IB/qib: Add new qib driver for QLogic PCIe InfiniBand adapters

Add a low-level IB driver for QLogic PCIe adapters.

Signed-off-by: Ralph Campbell <ralph.campbell@qlogic.com>
Signed-off-by: Roland Dreier <rolandd@cisco.com>

drivers/infiniband/Kconfig

@@ -43,6 +43,7 @@ config INFINIBAND_ADDR_TRANS
 
 source "drivers/infiniband/hw/mthca/Kconfig"
 source "drivers/infiniband/hw/ipath/Kconfig"
+source "drivers/infiniband/hw/qib/Kconfig"
 source "drivers/infiniband/hw/ehca/Kconfig"
 source "drivers/infiniband/hw/amso1100/Kconfig"
 source "drivers/infiniband/hw/cxgb3/Kconfig"

drivers/infiniband/Makefile

@@ -1,6 +1,7 @@
 obj-$(CONFIG_INFINIBAND)		+= core/
 obj-$(CONFIG_INFINIBAND_MTHCA)		+= hw/mthca/
 obj-$(CONFIG_INFINIBAND_IPATH)		+= hw/ipath/
+obj-$(CONFIG_INFINIBAND_QIB)		+= hw/qib/
 obj-$(CONFIG_INFINIBAND_EHCA)		+= hw/ehca/
 obj-$(CONFIG_INFINIBAND_AMSO1100)	+= hw/amso1100/
 obj-$(CONFIG_INFINIBAND_CXGB3)		+= hw/cxgb3/

drivers/infiniband/hw/qib/Kconfig

@@ -0,0 +1,7 @@
+config INFINIBAND_QIB
+	tristate "QLogic PCIe HCA support"
+	depends on 64BIT && NET
+	---help---
+	This is a low-level driver for QLogic PCIe QLE InfiniBand host
+	channel adapters.  This driver does not support the QLogic
+	HyperTransport card (model QHT7140).

drivers/infiniband/hw/qib/Makefile

@@ -0,0 +1,15 @@
+obj-$(CONFIG_INFINIBAND_QIB) += ib_qib.o
+
+ib_qib-y := qib_cq.o qib_diag.o qib_dma.o qib_driver.o qib_eeprom.o \
+	qib_file_ops.o qib_fs.o qib_init.o qib_intr.o qib_keys.o \
+	qib_mad.o qib_mmap.o qib_mr.o qib_pcie.o qib_pio_copy.o \
+	qib_qp.o qib_qsfp.o qib_rc.o qib_ruc.o qib_sdma.o qib_srq.o \
+	qib_sysfs.o qib_twsi.o qib_tx.o qib_uc.o qib_ud.o \
+	qib_user_pages.o qib_user_sdma.o qib_verbs_mcast.o qib_iba7220.o \
+	qib_sd7220.o qib_sd7220_img.o qib_iba7322.o qib_verbs.o
+
+# 6120 has no fallback if no MSI interrupts, others can do INTx
+ib_qib-$(CONFIG_PCI_MSI) += qib_iba6120.o
+
+ib_qib-$(CONFIG_X86_64) += qib_wc_x86_64.o
+ib_qib-$(CONFIG_PPC64) += qib_wc_ppc64.o

drivers/infiniband/hw/qib/qib_6120_regs.h

@@ -0,0 +1,977 @@
+/*
+ * Copyright (c) 2008, 2009, 2010 QLogic Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+/* This file is mechanically generated from RTL. Any hand-edits will be lost! */
+
+#define QIB_6120_Revision_OFFS 0x0
+#define QIB_6120_Revision_R_Simulator_LSB 0x3F
+#define QIB_6120_Revision_R_Simulator_RMASK 0x1
+#define QIB_6120_Revision_Reserved_LSB 0x28
+#define QIB_6120_Revision_Reserved_RMASK 0x7FFFFF
+#define QIB_6120_Revision_BoardID_LSB 0x20
+#define QIB_6120_Revision_BoardID_RMASK 0xFF
+#define QIB_6120_Revision_R_SW_LSB 0x18
+#define QIB_6120_Revision_R_SW_RMASK 0xFF
+#define QIB_6120_Revision_R_Arch_LSB 0x10
+#define QIB_6120_Revision_R_Arch_RMASK 0xFF
+#define QIB_6120_Revision_R_ChipRevMajor_LSB 0x8
+#define QIB_6120_Revision_R_ChipRevMajor_RMASK 0xFF
+#define QIB_6120_Revision_R_ChipRevMinor_LSB 0x0
+#define QIB_6120_Revision_R_ChipRevMinor_RMASK 0xFF
+
+#define QIB_6120_Control_OFFS 0x8
+#define QIB_6120_Control_TxLatency_LSB 0x4
+#define QIB_6120_Control_TxLatency_RMASK 0x1
+#define QIB_6120_Control_PCIERetryBufDiagEn_LSB 0x3
+#define QIB_6120_Control_PCIERetryBufDiagEn_RMASK 0x1
+#define QIB_6120_Control_LinkEn_LSB 0x2
+#define QIB_6120_Control_LinkEn_RMASK 0x1
+#define QIB_6120_Control_FreezeMode_LSB 0x1
+#define QIB_6120_Control_FreezeMode_RMASK 0x1
+#define QIB_6120_Control_SyncReset_LSB 0x0
+#define QIB_6120_Control_SyncReset_RMASK 0x1
+
+#define QIB_6120_PageAlign_OFFS 0x10
+
+#define QIB_6120_PortCnt_OFFS 0x18
+
+#define QIB_6120_SendRegBase_OFFS 0x30
+
+#define QIB_6120_UserRegBase_OFFS 0x38
+
+#define QIB_6120_CntrRegBase_OFFS 0x40
+
+#define QIB_6120_Scratch_OFFS 0x48
+#define QIB_6120_Scratch_TopHalf_LSB 0x20
+#define QIB_6120_Scratch_TopHalf_RMASK 0xFFFFFFFF
+#define QIB_6120_Scratch_BottomHalf_LSB 0x0
+#define QIB_6120_Scratch_BottomHalf_RMASK 0xFFFFFFFF
+
+#define QIB_6120_IntBlocked_OFFS 0x60
+#define QIB_6120_IntBlocked_ErrorIntBlocked_LSB 0x1F
+#define QIB_6120_IntBlocked_ErrorIntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_PioSetIntBlocked_LSB 0x1E
+#define QIB_6120_IntBlocked_PioSetIntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_PioBufAvailIntBlocked_LSB 0x1D
+#define QIB_6120_IntBlocked_PioBufAvailIntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_assertGPIOIntBlocked_LSB 0x1C
+#define QIB_6120_IntBlocked_assertGPIOIntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_Reserved_LSB 0xF
+#define QIB_6120_IntBlocked_Reserved_RMASK 0x1FFF
+#define QIB_6120_IntBlocked_RcvAvail4IntBlocked_LSB 0x10
+#define QIB_6120_IntBlocked_RcvAvail4IntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_RcvAvail3IntBlocked_LSB 0xF
+#define QIB_6120_IntBlocked_RcvAvail3IntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_RcvAvail2IntBlocked_LSB 0xE
+#define QIB_6120_IntBlocked_RcvAvail2IntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_RcvAvail1IntBlocked_LSB 0xD
+#define QIB_6120_IntBlocked_RcvAvail1IntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_RcvAvail0IntBlocked_LSB 0xC
+#define QIB_6120_IntBlocked_RcvAvail0IntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_Reserved1_LSB 0x5
+#define QIB_6120_IntBlocked_Reserved1_RMASK 0x7F
+#define QIB_6120_IntBlocked_RcvUrg4IntBlocked_LSB 0x4
+#define QIB_6120_IntBlocked_RcvUrg4IntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_RcvUrg3IntBlocked_LSB 0x3
+#define QIB_6120_IntBlocked_RcvUrg3IntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_RcvUrg2IntBlocked_LSB 0x2
+#define QIB_6120_IntBlocked_RcvUrg2IntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_RcvUrg1IntBlocked_LSB 0x1
+#define QIB_6120_IntBlocked_RcvUrg1IntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_RcvUrg0IntBlocked_LSB 0x0
+#define QIB_6120_IntBlocked_RcvUrg0IntBlocked_RMASK 0x1
+
+#define QIB_6120_IntMask_OFFS 0x68
+#define QIB_6120_IntMask_ErrorIntMask_LSB 0x1F
+#define QIB_6120_IntMask_ErrorIntMask_RMASK 0x1
+#define QIB_6120_IntMask_PioSetIntMask_LSB 0x1E
+#define QIB_6120_IntMask_PioSetIntMask_RMASK 0x1
+#define QIB_6120_IntMask_PioBufAvailIntMask_LSB 0x1D
+#define QIB_6120_IntMask_PioBufAvailIntMask_RMASK 0x1
+#define QIB_6120_IntMask_assertGPIOIntMask_LSB 0x1C
+#define QIB_6120_IntMask_assertGPIOIntMask_RMASK 0x1
+#define QIB_6120_IntMask_Reserved_LSB 0x11
+#define QIB_6120_IntMask_Reserved_RMASK 0x7FF
+#define QIB_6120_IntMask_RcvAvail4IntMask_LSB 0x10
+#define QIB_6120_IntMask_RcvAvail4IntMask_RMASK 0x1
+#define QIB_6120_IntMask_RcvAvail3IntMask_LSB 0xF
+#define QIB_6120_IntMask_RcvAvail3IntMask_RMASK 0x1
+#define QIB_6120_IntMask_RcvAvail2IntMask_LSB 0xE
+#define QIB_6120_IntMask_RcvAvail2IntMask_RMASK 0x1
+#define QIB_6120_IntMask_RcvAvail1IntMask_LSB 0xD
+#define QIB_6120_IntMask_RcvAvail1IntMask_RMASK 0x1
+#define QIB_6120_IntMask_RcvAvail0IntMask_LSB 0xC
+#define QIB_6120_IntMask_RcvAvail0IntMask_RMASK 0x1
+#define QIB_6120_IntMask_Reserved1_LSB 0x5
+#define QIB_6120_IntMask_Reserved1_RMASK 0x7F
+#define QIB_6120_IntMask_RcvUrg4IntMask_LSB 0x4
+#define QIB_6120_IntMask_RcvUrg4IntMask_RMASK 0x1
+#define QIB_6120_IntMask_RcvUrg3IntMask_LSB 0x3
+#define QIB_6120_IntMask_RcvUrg3IntMask_RMASK 0x1
+#define QIB_6120_IntMask_RcvUrg2IntMask_LSB 0x2
+#define QIB_6120_IntMask_RcvUrg2IntMask_RMASK 0x1
+#define QIB_6120_IntMask_RcvUrg1IntMask_LSB 0x1
+#define QIB_6120_IntMask_RcvUrg1IntMask_RMASK 0x1
+#define QIB_6120_IntMask_RcvUrg0IntMask_LSB 0x0
+#define QIB_6120_IntMask_RcvUrg0IntMask_RMASK 0x1
+
+#define QIB_6120_IntStatus_OFFS 0x70
+#define QIB_6120_IntStatus_Error_LSB 0x1F
+#define QIB_6120_IntStatus_Error_RMASK 0x1
+#define QIB_6120_IntStatus_PioSent_LSB 0x1E
+#define QIB_6120_IntStatus_PioSent_RMASK 0x1
+#define QIB_6120_IntStatus_PioBufAvail_LSB 0x1D
+#define QIB_6120_IntStatus_PioBufAvail_RMASK 0x1
+#define QIB_6120_IntStatus_assertGPIO_LSB 0x1C
+#define QIB_6120_IntStatus_assertGPIO_RMASK 0x1
+#define QIB_6120_IntStatus_Reserved_LSB 0xF
+#define QIB_6120_IntStatus_Reserved_RMASK 0x1FFF
+#define QIB_6120_IntStatus_RcvAvail4_LSB 0x10
+#define QIB_6120_IntStatus_RcvAvail4_RMASK 0x1
+#define QIB_6120_IntStatus_RcvAvail3_LSB 0xF
+#define QIB_6120_IntStatus_RcvAvail3_RMASK 0x1
+#define QIB_6120_IntStatus_RcvAvail2_LSB 0xE
+#define QIB_6120_IntStatus_RcvAvail2_RMASK 0x1
+#define QIB_6120_IntStatus_RcvAvail1_LSB 0xD
+#define QIB_6120_IntStatus_RcvAvail1_RMASK 0x1
+#define QIB_6120_IntStatus_RcvAvail0_LSB 0xC
+#define QIB_6120_IntStatus_RcvAvail0_RMASK 0x1
+#define QIB_6120_IntStatus_Reserved1_LSB 0x5
+#define QIB_6120_IntStatus_Reserved1_RMASK 0x7F
+#define QIB_6120_IntStatus_RcvUrg4_LSB 0x4
+#define QIB_6120_IntStatus_RcvUrg4_RMASK 0x1
+#define QIB_6120_IntStatus_RcvUrg3_LSB 0x3
+#define QIB_6120_IntStatus_RcvUrg3_RMASK 0x1
+#define QIB_6120_IntStatus_RcvUrg2_LSB 0x2
+#define QIB_6120_IntStatus_RcvUrg2_RMASK 0x1
+#define QIB_6120_IntStatus_RcvUrg1_LSB 0x1
+#define QIB_6120_IntStatus_RcvUrg1_RMASK 0x1
+#define QIB_6120_IntStatus_RcvUrg0_LSB 0x0
+#define QIB_6120_IntStatus_RcvUrg0_RMASK 0x1
+
+#define QIB_6120_IntClear_OFFS 0x78
+#define QIB_6120_IntClear_ErrorIntClear_LSB 0x1F
+#define QIB_6120_IntClear_ErrorIntClear_RMASK 0x1
+#define QIB_6120_IntClear_PioSetIntClear_LSB 0x1E
+#define QIB_6120_IntClear_PioSetIntClear_RMASK 0x1
+#define QIB_6120_IntClear_PioBufAvailIntClear_LSB 0x1D
+#define QIB_6120_IntClear_PioBufAvailIntClear_RMASK 0x1
+#define QIB_6120_IntClear_assertGPIOIntClear_LSB 0x1C
+#define QIB_6120_IntClear_assertGPIOIntClear_RMASK 0x1
+#define QIB_6120_IntClear_Reserved_LSB 0xF
+#define QIB_6120_IntClear_Reserved_RMASK 0x1FFF
+#define QIB_6120_IntClear_RcvAvail4IntClear_LSB 0x10
+#define QIB_6120_IntClear_RcvAvail4IntClear_RMASK 0x1
+#define QIB_6120_IntClear_RcvAvail3IntClear_LSB 0xF
+#define QIB_6120_IntClear_RcvAvail3IntClear_RMASK 0x1
+#define QIB_6120_IntClear_RcvAvail2IntClear_LSB 0xE
+#define QIB_6120_IntClear_RcvAvail2IntClear_RMASK 0x1
+#define QIB_6120_IntClear_RcvAvail1IntClear_LSB 0xD
+#define QIB_6120_IntClear_RcvAvail1IntClear_RMASK 0x1
+#define QIB_6120_IntClear_RcvAvail0IntClear_LSB 0xC
+#define QIB_6120_IntClear_RcvAvail0IntClear_RMASK 0x1
+#define QIB_6120_IntClear_Reserved1_LSB 0x5
+#define QIB_6120_IntClear_Reserved1_RMASK 0x7F
+#define QIB_6120_IntClear_RcvUrg4IntClear_LSB 0x4
+#define QIB_6120_IntClear_RcvUrg4IntClear_RMASK 0x1
+#define QIB_6120_IntClear_RcvUrg3IntClear_LSB 0x3
+#define QIB_6120_IntClear_RcvUrg3IntClear_RMASK 0x1
+#define QIB_6120_IntClear_RcvUrg2IntClear_LSB 0x2
+#define QIB_6120_IntClear_RcvUrg2IntClear_RMASK 0x1
+#define QIB_6120_IntClear_RcvUrg1IntClear_LSB 0x1
+#define QIB_6120_IntClear_RcvUrg1IntClear_RMASK 0x1
+#define QIB_6120_IntClear_RcvUrg0IntClear_LSB 0x0
+#define QIB_6120_IntClear_RcvUrg0IntClear_RMASK 0x1
+
+#define QIB_6120_ErrMask_OFFS 0x80
+#define QIB_6120_ErrMask_Reserved_LSB 0x34
+#define QIB_6120_ErrMask_Reserved_RMASK 0xFFF
+#define QIB_6120_ErrMask_HardwareErrMask_LSB 0x33
+#define QIB_6120_ErrMask_HardwareErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_ResetNegatedMask_LSB 0x32
+#define QIB_6120_ErrMask_ResetNegatedMask_RMASK 0x1
+#define QIB_6120_ErrMask_InvalidAddrErrMask_LSB 0x31
+#define QIB_6120_ErrMask_InvalidAddrErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_IBStatusChangedMask_LSB 0x30
+#define QIB_6120_ErrMask_IBStatusChangedMask_RMASK 0x1
+#define QIB_6120_ErrMask_Reserved1_LSB 0x26
+#define QIB_6120_ErrMask_Reserved1_RMASK 0x3FF
+#define QIB_6120_ErrMask_SendUnsupportedVLErrMask_LSB 0x25
+#define QIB_6120_ErrMask_SendUnsupportedVLErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_SendUnexpectedPktNumErrMask_LSB 0x24
+#define QIB_6120_ErrMask_SendUnexpectedPktNumErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_SendPioArmLaunchErrMask_LSB 0x23
+#define QIB_6120_ErrMask_SendPioArmLaunchErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_SendDroppedDataPktErrMask_LSB 0x22
+#define QIB_6120_ErrMask_SendDroppedDataPktErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_SendDroppedSmpPktErrMask_LSB 0x21
+#define QIB_6120_ErrMask_SendDroppedSmpPktErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_SendPktLenErrMask_LSB 0x20
+#define QIB_6120_ErrMask_SendPktLenErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_SendUnderRunErrMask_LSB 0x1F
+#define QIB_6120_ErrMask_SendUnderRunErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_SendMaxPktLenErrMask_LSB 0x1E
+#define QIB_6120_ErrMask_SendMaxPktLenErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_SendMinPktLenErrMask_LSB 0x1D
+#define QIB_6120_ErrMask_SendMinPktLenErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_Reserved2_LSB 0x12
+#define QIB_6120_ErrMask_Reserved2_RMASK 0x7FF
+#define QIB_6120_ErrMask_RcvIBLostLinkErrMask_LSB 0x11
+#define QIB_6120_ErrMask_RcvIBLostLinkErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvHdrErrMask_LSB 0x10
+#define QIB_6120_ErrMask_RcvHdrErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvHdrLenErrMask_LSB 0xF
+#define QIB_6120_ErrMask_RcvHdrLenErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvBadTidErrMask_LSB 0xE
+#define QIB_6120_ErrMask_RcvBadTidErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvHdrFullErrMask_LSB 0xD
+#define QIB_6120_ErrMask_RcvHdrFullErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvEgrFullErrMask_LSB 0xC
+#define QIB_6120_ErrMask_RcvEgrFullErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvBadVersionErrMask_LSB 0xB
+#define QIB_6120_ErrMask_RcvBadVersionErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvIBFlowErrMask_LSB 0xA
+#define QIB_6120_ErrMask_RcvIBFlowErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvEBPErrMask_LSB 0x9
+#define QIB_6120_ErrMask_RcvEBPErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvUnsupportedVLErrMask_LSB 0x8
+#define QIB_6120_ErrMask_RcvUnsupportedVLErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvUnexpectedCharErrMask_LSB 0x7
+#define QIB_6120_ErrMask_RcvUnexpectedCharErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvShortPktLenErrMask_LSB 0x6
+#define QIB_6120_ErrMask_RcvShortPktLenErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvLongPktLenErrMask_LSB 0x5
+#define QIB_6120_ErrMask_RcvLongPktLenErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvMaxPktLenErrMask_LSB 0x4
+#define QIB_6120_ErrMask_RcvMaxPktLenErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvMinPktLenErrMask_LSB 0x3
+#define QIB_6120_ErrMask_RcvMinPktLenErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvICRCErrMask_LSB 0x2
+#define QIB_6120_ErrMask_RcvICRCErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvVCRCErrMask_LSB 0x1
+#define QIB_6120_ErrMask_RcvVCRCErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvFormatErrMask_LSB 0x0
+#define QIB_6120_ErrMask_RcvFormatErrMask_RMASK 0x1
+
+#define QIB_6120_ErrStatus_OFFS 0x88
+#define QIB_6120_ErrStatus_Reserved_LSB 0x34
+#define QIB_6120_ErrStatus_Reserved_RMASK 0xFFF
+#define QIB_6120_ErrStatus_HardwareErr_LSB 0x33
+#define QIB_6120_ErrStatus_HardwareErr_RMASK 0x1
+#define QIB_6120_ErrStatus_ResetNegated_LSB 0x32
+#define QIB_6120_ErrStatus_ResetNegated_RMASK 0x1
+#define QIB_6120_ErrStatus_InvalidAddrErr_LSB 0x31
+#define QIB_6120_ErrStatus_InvalidAddrErr_RMASK 0x1
+#define QIB_6120_ErrStatus_IBStatusChanged_LSB 0x30
+#define QIB_6120_ErrStatus_IBStatusChanged_RMASK 0x1
+#define QIB_6120_ErrStatus_Reserved1_LSB 0x26
+#define QIB_6120_ErrStatus_Reserved1_RMASK 0x3FF
+#define QIB_6120_ErrStatus_SendUnsupportedVLErr_LSB 0x25
+#define QIB_6120_ErrStatus_SendUnsupportedVLErr_RMASK 0x1
+#define QIB_6120_ErrStatus_SendUnexpectedPktNumErr_LSB 0x24
+#define QIB_6120_ErrStatus_SendUnexpectedPktNumErr_RMASK 0x1
+#define QIB_6120_ErrStatus_SendPioArmLaunchErr_LSB 0x23
+#define QIB_6120_ErrStatus_SendPioArmLaunchErr_RMASK 0x1
+#define QIB_6120_ErrStatus_SendDroppedDataPktErr_LSB 0x22
+#define QIB_6120_ErrStatus_SendDroppedDataPktErr_RMASK 0x1
+#define QIB_6120_ErrStatus_SendDroppedSmpPktErr_LSB 0x21
+#define QIB_6120_ErrStatus_SendDroppedSmpPktErr_RMASK 0x1
+#define QIB_6120_ErrStatus_SendPktLenErr_LSB 0x20
+#define QIB_6120_ErrStatus_SendPktLenErr_RMASK 0x1
+#define QIB_6120_ErrStatus_SendUnderRunErr_LSB 0x1F
+#define QIB_6120_ErrStatus_SendUnderRunErr_RMASK 0x1
+#define QIB_6120_ErrStatus_SendMaxPktLenErr_LSB 0x1E
+#define QIB_6120_ErrStatus_SendMaxPktLenErr_RMASK 0x1
+#define QIB_6120_ErrStatus_SendMinPktLenErr_LSB 0x1D
+#define QIB_6120_ErrStatus_SendMinPktLenErr_RMASK 0x1
+#define QIB_6120_ErrStatus_Reserved2_LSB 0x12
+#define QIB_6120_ErrStatus_Reserved2_RMASK 0x7FF
+#define QIB_6120_ErrStatus_RcvIBLostLinkErr_LSB 0x11
+#define QIB_6120_ErrStatus_RcvIBLostLinkErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvHdrErr_LSB 0x10
+#define QIB_6120_ErrStatus_RcvHdrErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvHdrLenErr_LSB 0xF
+#define QIB_6120_ErrStatus_RcvHdrLenErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvBadTidErr_LSB 0xE
+#define QIB_6120_ErrStatus_RcvBadTidErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvHdrFullErr_LSB 0xD
+#define QIB_6120_ErrStatus_RcvHdrFullErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvEgrFullErr_LSB 0xC
+#define QIB_6120_ErrStatus_RcvEgrFullErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvBadVersionErr_LSB 0xB
+#define QIB_6120_ErrStatus_RcvBadVersionErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvIBFlowErr_LSB 0xA
+#define QIB_6120_ErrStatus_RcvIBFlowErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvEBPErr_LSB 0x9
+#define QIB_6120_ErrStatus_RcvEBPErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvUnsupportedVLErr_LSB 0x8
+#define QIB_6120_ErrStatus_RcvUnsupportedVLErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvUnexpectedCharErr_LSB 0x7
+#define QIB_6120_ErrStatus_RcvUnexpectedCharErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvShortPktLenErr_LSB 0x6
+#define QIB_6120_ErrStatus_RcvShortPktLenErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvLongPktLenErr_LSB 0x5
+#define QIB_6120_ErrStatus_RcvLongPktLenErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvMaxPktLenErr_LSB 0x4
+#define QIB_6120_ErrStatus_RcvMaxPktLenErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvMinPktLenErr_LSB 0x3
+#define QIB_6120_ErrStatus_RcvMinPktLenErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvICRCErr_LSB 0x2
+#define QIB_6120_ErrStatus_RcvICRCErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvVCRCErr_LSB 0x1
+#define QIB_6120_ErrStatus_RcvVCRCErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvFormatErr_LSB 0x0
+#define QIB_6120_ErrStatus_RcvFormatErr_RMASK 0x1
+
+#define QIB_6120_ErrClear_OFFS 0x90
+#define QIB_6120_ErrClear_Reserved_LSB 0x34
+#define QIB_6120_ErrClear_Reserved_RMASK 0xFFF
+#define QIB_6120_ErrClear_HardwareErrClear_LSB 0x33
+#define QIB_6120_ErrClear_HardwareErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_ResetNegatedClear_LSB 0x32
+#define QIB_6120_ErrClear_ResetNegatedClear_RMASK 0x1
+#define QIB_6120_ErrClear_InvalidAddrErrClear_LSB 0x31
+#define QIB_6120_ErrClear_InvalidAddrErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_IBStatusChangedClear_LSB 0x30
+#define QIB_6120_ErrClear_IBStatusChangedClear_RMASK 0x1
+#define QIB_6120_ErrClear_Reserved1_LSB 0x26
+#define QIB_6120_ErrClear_Reserved1_RMASK 0x3FF
+#define QIB_6120_ErrClear_SendUnsupportedVLErrClear_LSB 0x25
+#define QIB_6120_ErrClear_SendUnsupportedVLErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_SendUnexpectedPktNumErrClear_LSB 0x24
+#define QIB_6120_ErrClear_SendUnexpectedPktNumErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_SendPioArmLaunchErrClear_LSB 0x23
+#define QIB_6120_ErrClear_SendPioArmLaunchErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_SendDroppedDataPktErrClear_LSB 0x22
+#define QIB_6120_ErrClear_SendDroppedDataPktErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_SendDroppedSmpPktErrClear_LSB 0x21
+#define QIB_6120_ErrClear_SendDroppedSmpPktErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_SendPktLenErrClear_LSB 0x20
+#define QIB_6120_ErrClear_SendPktLenErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_SendUnderRunErrClear_LSB 0x1F
+#define QIB_6120_ErrClear_SendUnderRunErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_SendMaxPktLenErrClear_LSB 0x1E
+#define QIB_6120_ErrClear_SendMaxPktLenErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_SendMinPktLenErrClear_LSB 0x1D
+#define QIB_6120_ErrClear_SendMinPktLenErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_Reserved2_LSB 0x12
+#define QIB_6120_ErrClear_Reserved2_RMASK 0x7FF
+#define QIB_6120_ErrClear_RcvIBLostLinkErrClear_LSB 0x11
+#define QIB_6120_ErrClear_RcvIBLostLinkErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvHdrErrClear_LSB 0x10
+#define QIB_6120_ErrClear_RcvHdrErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvHdrLenErrClear_LSB 0xF
+#define QIB_6120_ErrClear_RcvHdrLenErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvBadTidErrClear_LSB 0xE
+#define QIB_6120_ErrClear_RcvBadTidErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvHdrFullErrClear_LSB 0xD
+#define QIB_6120_ErrClear_RcvHdrFullErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvEgrFullErrClear_LSB 0xC
+#define QIB_6120_ErrClear_RcvEgrFullErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvBadVersionErrClear_LSB 0xB
+#define QIB_6120_ErrClear_RcvBadVersionErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvIBFlowErrClear_LSB 0xA
+#define QIB_6120_ErrClear_RcvIBFlowErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvEBPErrClear_LSB 0x9
+#define QIB_6120_ErrClear_RcvEBPErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvUnsupportedVLErrClear_LSB 0x8
+#define QIB_6120_ErrClear_RcvUnsupportedVLErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvUnexpectedCharErrClear_LSB 0x7
+#define QIB_6120_ErrClear_RcvUnexpectedCharErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvShortPktLenErrClear_LSB 0x6
+#define QIB_6120_ErrClear_RcvShortPktLenErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvLongPktLenErrClear_LSB 0x5
+#define QIB_6120_ErrClear_RcvLongPktLenErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvMaxPktLenErrClear_LSB 0x4
+#define QIB_6120_ErrClear_RcvMaxPktLenErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvMinPktLenErrClear_LSB 0x3
+#define QIB_6120_ErrClear_RcvMinPktLenErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvICRCErrClear_LSB 0x2
+#define QIB_6120_ErrClear_RcvICRCErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvVCRCErrClear_LSB 0x1
+#define QIB_6120_ErrClear_RcvVCRCErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvFormatErrClear_LSB 0x0
+#define QIB_6120_ErrClear_RcvFormatErrClear_RMASK 0x1
+
+#define QIB_6120_HwErrMask_OFFS 0x98
+#define QIB_6120_HwErrMask_IBCBusFromSPCParityErrMask_LSB 0x3F
+#define QIB_6120_HwErrMask_IBCBusFromSPCParityErrMask_RMASK 0x1
+#define QIB_6120_HwErrMask_IBCBusToSPCParityErrMask_LSB 0x3E
+#define QIB_6120_HwErrMask_IBCBusToSPCParityErrMask_RMASK 0x1
+#define QIB_6120_HwErrMask_Reserved_LSB 0x3D
+#define QIB_6120_HwErrMask_Reserved_RMASK 0x1
+#define QIB_6120_HwErrMask_IBSerdesPClkNotDetectMask_LSB 0x3C
+#define QIB_6120_HwErrMask_IBSerdesPClkNotDetectMask_RMASK 0x1
+#define QIB_6120_HwErrMask_PCIESerdesQ0PClkNotDetectMask_LSB 0x3B
+#define QIB_6120_HwErrMask_PCIESerdesQ0PClkNotDetectMask_RMASK 0x1
+#define QIB_6120_HwErrMask_PCIESerdesQ1PClkNotDetectMask_LSB 0x3A
+#define QIB_6120_HwErrMask_PCIESerdesQ1PClkNotDetectMask_RMASK 0x1
+#define QIB_6120_HwErrMask_Reserved1_LSB 0x39
+#define QIB_6120_HwErrMask_Reserved1_RMASK 0x1
+#define QIB_6120_HwErrMask_IBPLLrfSlipMask_LSB 0x38
+#define QIB_6120_HwErrMask_IBPLLrfSlipMask_RMASK 0x1
+#define QIB_6120_HwErrMask_IBPLLfbSlipMask_LSB 0x37
+#define QIB_6120_HwErrMask_IBPLLfbSlipMask_RMASK 0x1
+#define QIB_6120_HwErrMask_PowerOnBISTFailedMask_LSB 0x36
+#define QIB_6120_HwErrMask_PowerOnBISTFailedMask_RMASK 0x1
+#define QIB_6120_HwErrMask_Reserved2_LSB 0x33
+#define QIB_6120_HwErrMask_Reserved2_RMASK 0x7
+#define QIB_6120_HwErrMask_RXEMemParityErrMask_LSB 0x2C
+#define QIB_6120_HwErrMask_RXEMemParityErrMask_RMASK 0x7F
+#define QIB_6120_HwErrMask_TXEMemParityErrMask_LSB 0x28
+#define QIB_6120_HwErrMask_TXEMemParityErrMask_RMASK 0xF
+#define QIB_6120_HwErrMask_Reserved3_LSB 0x22
+#define QIB_6120_HwErrMask_Reserved3_RMASK 0x3F
+#define QIB_6120_HwErrMask_PCIeBusParityErrMask_LSB 0x1F
+#define QIB_6120_HwErrMask_PCIeBusParityErrMask_RMASK 0x7
+#define QIB_6120_HwErrMask_PcieCplTimeoutMask_LSB 0x1E
+#define QIB_6120_HwErrMask_PcieCplTimeoutMask_RMASK 0x1
+#define QIB_6120_HwErrMask_PoisonedTLPMask_LSB 0x1D
+#define QIB_6120_HwErrMask_PoisonedTLPMask_RMASK 0x1
+#define QIB_6120_HwErrMask_Reserved4_LSB 0x6
+#define QIB_6120_HwErrMask_Reserved4_RMASK 0x7FFFFF
+#define QIB_6120_HwErrMask_PCIeMemParityErrMask_LSB 0x0
+#define QIB_6120_HwErrMask_PCIeMemParityErrMask_RMASK 0x3F
+
+#define QIB_6120_HwErrStatus_OFFS 0xA0
+#define QIB_6120_HwErrStatus_IBCBusFromSPCParityErr_LSB 0x3F
+#define QIB_6120_HwErrStatus_IBCBusFromSPCParityErr_RMASK 0x1
+#define QIB_6120_HwErrStatus_IBCBusToSPCParityErr_LSB 0x3E
+#define QIB_6120_HwErrStatus_IBCBusToSPCParityErr_RMASK 0x1
+#define QIB_6120_HwErrStatus_Reserved_LSB 0x3D
+#define QIB_6120_HwErrStatus_Reserved_RMASK 0x1
+#define QIB_6120_HwErrStatus_IBSerdesPClkNotDetect_LSB 0x3C
+#define QIB_6120_HwErrStatus_IBSerdesPClkNotDetect_RMASK 0x1
+#define QIB_6120_HwErrStatus_PCIESerdesQ0PClkNotDetect_LSB 0x3B
+#define QIB_6120_HwErrStatus_PCIESerdesQ0PClkNotDetect_RMASK 0x1
+#define QIB_6120_HwErrStatus_PCIESerdesQ1PClkNotDetect_LSB 0x3A
+#define QIB_6120_HwErrStatus_PCIESerdesQ1PClkNotDetect_RMASK 0x1
+#define QIB_6120_HwErrStatus_Reserved1_LSB 0x39
+#define QIB_6120_HwErrStatus_Reserved1_RMASK 0x1
+#define QIB_6120_HwErrStatus_IBPLLrfSlip_LSB 0x38
+#define QIB_6120_HwErrStatus_IBPLLrfSlip_RMASK 0x1
+#define QIB_6120_HwErrStatus_IBPLLfbSlip_LSB 0x37
+#define QIB_6120_HwErrStatus_IBPLLfbSlip_RMASK 0x1
+#define QIB_6120_HwErrStatus_PowerOnBISTFailed_LSB 0x36
+#define QIB_6120_HwErrStatus_PowerOnBISTFailed_RMASK 0x1
+#define QIB_6120_HwErrStatus_Reserved2_LSB 0x33
+#define QIB_6120_HwErrStatus_Reserved2_RMASK 0x7
+#define QIB_6120_HwErrStatus_RXEMemParity_LSB 0x2C
+#define QIB_6120_HwErrStatus_RXEMemParity_RMASK 0x7F
+#define QIB_6120_HwErrStatus_TXEMemParity_LSB 0x28
+#define QIB_6120_HwErrStatus_TXEMemParity_RMASK 0xF
+#define QIB_6120_HwErrStatus_Reserved3_LSB 0x22
+#define QIB_6120_HwErrStatus_Reserved3_RMASK 0x3F
+#define QIB_6120_HwErrStatus_PCIeBusParity_LSB 0x1F
+#define QIB_6120_HwErrStatus_PCIeBusParity_RMASK 0x7
+#define QIB_6120_HwErrStatus_PcieCplTimeout_LSB 0x1E
+#define QIB_6120_HwErrStatus_PcieCplTimeout_RMASK 0x1
+#define QIB_6120_HwErrStatus_PoisenedTLP_LSB 0x1D
+#define QIB_6120_HwErrStatus_PoisenedTLP_RMASK 0x1
+#define QIB_6120_HwErrStatus_Reserved4_LSB 0x6
+#define QIB_6120_HwErrStatus_Reserved4_RMASK 0x7FFFFF
+#define QIB_6120_HwErrStatus_PCIeMemParity_LSB 0x0
+#define QIB_6120_HwErrStatus_PCIeMemParity_RMASK 0x3F
+
+#define QIB_6120_HwErrClear_OFFS 0xA8
+#define QIB_6120_HwErrClear_IBCBusFromSPCParityErrClear_LSB 0x3F
+#define QIB_6120_HwErrClear_IBCBusFromSPCParityErrClear_RMASK 0x1
+#define QIB_6120_HwErrClear_IBCBusToSPCparityErrClear_LSB 0x3E
+#define QIB_6120_HwErrClear_IBCBusToSPCparityErrClear_RMASK 0x1
+#define QIB_6120_HwErrClear_Reserved_LSB 0x3D
+#define QIB_6120_HwErrClear_Reserved_RMASK 0x1
+#define QIB_6120_HwErrClear_IBSerdesPClkNotDetectClear_LSB 0x3C
+#define QIB_6120_HwErrClear_IBSerdesPClkNotDetectClear_RMASK 0x1
+#define QIB_6120_HwErrClear_PCIESerdesQ0PClkNotDetectClear_LSB 0x3B
+#define QIB_6120_HwErrClear_PCIESerdesQ0PClkNotDetectClear_RMASK 0x1
+#define QIB_6120_HwErrClear_PCIESerdesQ1PClkNotDetectClear_LSB 0x3A
+#define QIB_6120_HwErrClear_PCIESerdesQ1PClkNotDetectClear_RMASK 0x1
+#define QIB_6120_HwErrClear_Reserved1_LSB 0x39
+#define QIB_6120_HwErrClear_Reserved1_RMASK 0x1
+#define QIB_6120_HwErrClear_IBPLLrfSlipClear_LSB 0x38
+#define QIB_6120_HwErrClear_IBPLLrfSlipClear_RMASK 0x1
+#define QIB_6120_HwErrClear_IBPLLfbSlipClear_LSB 0x37
+#define QIB_6120_HwErrClear_IBPLLfbSlipClear_RMASK 0x1
+#define QIB_6120_HwErrClear_PowerOnBISTFailedClear_LSB 0x36
+#define QIB_6120_HwErrClear_PowerOnBISTFailedClear_RMASK 0x1
+#define QIB_6120_HwErrClear_Reserved2_LSB 0x33
+#define QIB_6120_HwErrClear_Reserved2_RMASK 0x7
+#define QIB_6120_HwErrClear_RXEMemParityClear_LSB 0x2C
+#define QIB_6120_HwErrClear_RXEMemParityClear_RMASK 0x7F
+#define QIB_6120_HwErrClear_TXEMemParityClear_LSB 0x28
+#define QIB_6120_HwErrClear_TXEMemParityClear_RMASK 0xF
+#define QIB_6120_HwErrClear_Reserved3_LSB 0x22
+#define QIB_6120_HwErrClear_Reserved3_RMASK 0x3F
+#define QIB_6120_HwErrClear_PCIeBusParityClr_LSB 0x1F
+#define QIB_6120_HwErrClear_PCIeBusParityClr_RMASK 0x7
+#define QIB_6120_HwErrClear_PcieCplTimeoutClear_LSB 0x1E
+#define QIB_6120_HwErrClear_PcieCplTimeoutClear_RMASK 0x1
+#define QIB_6120_HwErrClear_PoisonedTLPClear_LSB 0x1D
+#define QIB_6120_HwErrClear_PoisonedTLPClear_RMASK 0x1
+#define QIB_6120_HwErrClear_Reserved4_LSB 0x6
+#define QIB_6120_HwErrClear_Reserved4_RMASK 0x7FFFFF
+#define QIB_6120_HwErrClear_PCIeMemParityClr_LSB 0x0
+#define QIB_6120_HwErrClear_PCIeMemParityClr_RMASK 0x3F
+
+#define QIB_6120_HwDiagCtrl_OFFS 0xB0
+#define QIB_6120_HwDiagCtrl_ForceIBCBusFromSPCParityErr_LSB 0x3F
+#define QIB_6120_HwDiagCtrl_ForceIBCBusFromSPCParityErr_RMASK 0x1
+#define QIB_6120_HwDiagCtrl_ForceIBCBusToSPCParityErr_LSB 0x3E
+#define QIB_6120_HwDiagCtrl_ForceIBCBusToSPCParityErr_RMASK 0x1
+#define QIB_6120_HwDiagCtrl_CounterWrEnable_LSB 0x3D
+#define QIB_6120_HwDiagCtrl_CounterWrEnable_RMASK 0x1
+#define QIB_6120_HwDiagCtrl_CounterDisable_LSB 0x3C
+#define QIB_6120_HwDiagCtrl_CounterDisable_RMASK 0x1
+#define QIB_6120_HwDiagCtrl_Reserved_LSB 0x33
+#define QIB_6120_HwDiagCtrl_Reserved_RMASK 0x1FF
+#define QIB_6120_HwDiagCtrl_ForceRxMemParityErr_LSB 0x2C
+#define QIB_6120_HwDiagCtrl_ForceRxMemParityErr_RMASK 0x7F
+#define QIB_6120_HwDiagCtrl_ForceTxMemparityErr_LSB 0x28
+#define QIB_6120_HwDiagCtrl_ForceTxMemparityErr_RMASK 0xF
+#define QIB_6120_HwDiagCtrl_Reserved1_LSB 0x23
+#define QIB_6120_HwDiagCtrl_Reserved1_RMASK 0x1F
+#define QIB_6120_HwDiagCtrl_forcePCIeBusParity_LSB 0x1F
+#define QIB_6120_HwDiagCtrl_forcePCIeBusParity_RMASK 0xF
+#define QIB_6120_HwDiagCtrl_Reserved2_LSB 0x6
+#define QIB_6120_HwDiagCtrl_Reserved2_RMASK 0x1FFFFFF
+#define QIB_6120_HwDiagCtrl_forcePCIeMemParity_LSB 0x0
+#define QIB_6120_HwDiagCtrl_forcePCIeMemParity_RMASK 0x3F
+
+#define QIB_6120_IBCStatus_OFFS 0xC0
+#define QIB_6120_IBCStatus_TxCreditOk_LSB 0x1F
+#define QIB_6120_IBCStatus_TxCreditOk_RMASK 0x1
+#define QIB_6120_IBCStatus_TxReady_LSB 0x1E
+#define QIB_6120_IBCStatus_TxReady_RMASK 0x1
+#define QIB_6120_IBCStatus_Reserved_LSB 0x7
+#define QIB_6120_IBCStatus_Reserved_RMASK 0x7FFFFF
+#define QIB_6120_IBCStatus_LinkState_LSB 0x4
+#define QIB_6120_IBCStatus_LinkState_RMASK 0x7
+#define QIB_6120_IBCStatus_LinkTrainingState_LSB 0x0
+#define QIB_6120_IBCStatus_LinkTrainingState_RMASK 0xF
+
+#define QIB_6120_IBCCtrl_OFFS 0xC8
+#define QIB_6120_IBCCtrl_Loopback_LSB 0x3F
+#define QIB_6120_IBCCtrl_Loopback_RMASK 0x1
+#define QIB_6120_IBCCtrl_LinkDownDefaultState_LSB 0x3E
+#define QIB_6120_IBCCtrl_LinkDownDefaultState_RMASK 0x1
+#define QIB_6120_IBCCtrl_Reserved_LSB 0x2B
+#define QIB_6120_IBCCtrl_Reserved_RMASK 0x7FFFF
+#define QIB_6120_IBCCtrl_CreditScale_LSB 0x28
+#define QIB_6120_IBCCtrl_CreditScale_RMASK 0x7
+#define QIB_6120_IBCCtrl_OverrunThreshold_LSB 0x24
+#define QIB_6120_IBCCtrl_OverrunThreshold_RMASK 0xF
+#define QIB_6120_IBCCtrl_PhyerrThreshold_LSB 0x20
+#define QIB_6120_IBCCtrl_PhyerrThreshold_RMASK 0xF
+#define QIB_6120_IBCCtrl_Reserved1_LSB 0x1F
+#define QIB_6120_IBCCtrl_Reserved1_RMASK 0x1
+#define QIB_6120_IBCCtrl_MaxPktLen_LSB 0x14
+#define QIB_6120_IBCCtrl_MaxPktLen_RMASK 0x7FF
+#define QIB_6120_IBCCtrl_LinkCmd_LSB 0x12
+#define QIB_6120_IBCCtrl_LinkCmd_RMASK 0x3
+#define QIB_6120_IBCCtrl_LinkInitCmd_LSB 0x10
+#define QIB_6120_IBCCtrl_LinkInitCmd_RMASK 0x3
+#define QIB_6120_IBCCtrl_FlowCtrlWaterMark_LSB 0x8
+#define QIB_6120_IBCCtrl_FlowCtrlWaterMark_RMASK 0xFF
+#define QIB_6120_IBCCtrl_FlowCtrlPeriod_LSB 0x0
+#define QIB_6120_IBCCtrl_FlowCtrlPeriod_RMASK 0xFF
+
+#define QIB_6120_EXTStatus_OFFS 0xD0
+#define QIB_6120_EXTStatus_GPIOIn_LSB 0x30
+#define QIB_6120_EXTStatus_GPIOIn_RMASK 0xFFFF
+#define QIB_6120_EXTStatus_Reserved_LSB 0x20
+#define QIB_6120_EXTStatus_Reserved_RMASK 0xFFFF
+#define QIB_6120_EXTStatus_Reserved1_LSB 0x10
+#define QIB_6120_EXTStatus_Reserved1_RMASK 0xFFFF
+#define QIB_6120_EXTStatus_MemBISTFoundErr_LSB 0xF
+#define QIB_6120_EXTStatus_MemBISTFoundErr_RMASK 0x1
+#define QIB_6120_EXTStatus_MemBISTEndTest_LSB 0xE
+#define QIB_6120_EXTStatus_MemBISTEndTest_RMASK 0x1
+#define QIB_6120_EXTStatus_Reserved2_LSB 0x0
+#define QIB_6120_EXTStatus_Reserved2_RMASK 0x3FFF
+
+#define QIB_6120_EXTCtrl_OFFS 0xD8
+#define QIB_6120_EXTCtrl_GPIOOe_LSB 0x30
+#define QIB_6120_EXTCtrl_GPIOOe_RMASK 0xFFFF
+#define QIB_6120_EXTCtrl_GPIOInvert_LSB 0x20
+#define QIB_6120_EXTCtrl_GPIOInvert_RMASK 0xFFFF
+#define QIB_6120_EXTCtrl_Reserved_LSB 0x4
+#define QIB_6120_EXTCtrl_Reserved_RMASK 0xFFFFFFF
+#define QIB_6120_EXTCtrl_LEDPriPortGreenOn_LSB 0x3
+#define QIB_6120_EXTCtrl_LEDPriPortGreenOn_RMASK 0x1
+#define QIB_6120_EXTCtrl_LEDPriPortYellowOn_LSB 0x2
+#define QIB_6120_EXTCtrl_LEDPriPortYellowOn_RMASK 0x1
+#define QIB_6120_EXTCtrl_LEDGblOkGreenOn_LSB 0x1
+#define QIB_6120_EXTCtrl_LEDGblOkGreenOn_RMASK 0x1
+#define QIB_6120_EXTCtrl_LEDGblErrRedOff_LSB 0x0
+#define QIB_6120_EXTCtrl_LEDGblErrRedOff_RMASK 0x1
+
+#define QIB_6120_GPIOOut_OFFS 0xE0
+
+#define QIB_6120_GPIOMask_OFFS 0xE8
+
+#define QIB_6120_GPIOStatus_OFFS 0xF0
+
+#define QIB_6120_GPIOClear_OFFS 0xF8
+
+#define QIB_6120_RcvCtrl_OFFS 0x100
+#define QIB_6120_RcvCtrl_TailUpd_LSB 0x1F
+#define QIB_6120_RcvCtrl_TailUpd_RMASK 0x1
+#define QIB_6120_RcvCtrl_RcvPartitionKeyDisable_LSB 0x1E
+#define QIB_6120_RcvCtrl_RcvPartitionKeyDisable_RMASK 0x1
+#define QIB_6120_RcvCtrl_Reserved_LSB 0x15
+#define QIB_6120_RcvCtrl_Reserved_RMASK 0x1FF
+#define QIB_6120_RcvCtrl_IntrAvail_LSB 0x10
+#define QIB_6120_RcvCtrl_IntrAvail_RMASK 0x1F
+#define QIB_6120_RcvCtrl_Reserved1_LSB 0x9
+#define QIB_6120_RcvCtrl_Reserved1_RMASK 0x7F
+#define QIB_6120_RcvCtrl_Reserved2_LSB 0x5
+#define QIB_6120_RcvCtrl_Reserved2_RMASK 0xF
+#define QIB_6120_RcvCtrl_PortEnable_LSB 0x0
+#define QIB_6120_RcvCtrl_PortEnable_RMASK 0x1F
+
+#define QIB_6120_RcvBTHQP_OFFS 0x108
+#define QIB_6120_RcvBTHQP_BTHQP_Mask_LSB 0x1E
+#define QIB_6120_RcvBTHQP_BTHQP_Mask_RMASK 0x3
+#define QIB_6120_RcvBTHQP_Reserved_LSB 0x18
+#define QIB_6120_RcvBTHQP_Reserved_RMASK 0x3F
+#define QIB_6120_RcvBTHQP_RcvBTHQP_LSB 0x0
+#define QIB_6120_RcvBTHQP_RcvBTHQP_RMASK 0xFFFFFF
+
+#define QIB_6120_RcvHdrSize_OFFS 0x110
+
+#define QIB_6120_RcvHdrCnt_OFFS 0x118
+
+#define QIB_6120_RcvHdrEntSize_OFFS 0x120
+
+#define QIB_6120_RcvTIDBase_OFFS 0x128
+
+#define QIB_6120_RcvTIDCnt_OFFS 0x130
+
+#define QIB_6120_RcvEgrBase_OFFS 0x138
+
+#define QIB_6120_RcvEgrCnt_OFFS 0x140
+
+#define QIB_6120_RcvBufBase_OFFS 0x148
+
+#define QIB_6120_RcvBufSize_OFFS 0x150
+
+#define QIB_6120_RxIntMemBase_OFFS 0x158
+
+#define QIB_6120_RxIntMemSize_OFFS 0x160
+
+#define QIB_6120_RcvPartitionKey_OFFS 0x168
+
+#define QIB_6120_RcvPktLEDCnt_OFFS 0x178
+#define QIB_6120_RcvPktLEDCnt_ONperiod_LSB 0x20
+#define QIB_6120_RcvPktLEDCnt_ONperiod_RMASK 0xFFFFFFFF
+#define QIB_6120_RcvPktLEDCnt_OFFperiod_LSB 0x0
+#define QIB_6120_RcvPktLEDCnt_OFFperiod_RMASK 0xFFFFFFFF
+
+#define QIB_6120_SendCtrl_OFFS 0x1C0
+#define QIB_6120_SendCtrl_Disarm_LSB 0x1F
+#define QIB_6120_SendCtrl_Disarm_RMASK 0x1
+#define QIB_6120_SendCtrl_Reserved_LSB 0x17
+#define QIB_6120_SendCtrl_Reserved_RMASK 0xFF
+#define QIB_6120_SendCtrl_DisarmPIOBuf_LSB 0x10
+#define QIB_6120_SendCtrl_DisarmPIOBuf_RMASK 0x7F
+#define QIB_6120_SendCtrl_Reserved1_LSB 0x4
+#define QIB_6120_SendCtrl_Reserved1_RMASK 0xFFF
+#define QIB_6120_SendCtrl_PIOEnable_LSB 0x3
+#define QIB_6120_SendCtrl_PIOEnable_RMASK 0x1
+#define QIB_6120_SendCtrl_PIOBufAvailUpd_LSB 0x2
+#define QIB_6120_SendCtrl_PIOBufAvailUpd_RMASK 0x1
+#define QIB_6120_SendCtrl_PIOIntBufAvail_LSB 0x1
+#define QIB_6120_SendCtrl_PIOIntBufAvail_RMASK 0x1
+#define QIB_6120_SendCtrl_Abort_LSB 0x0
+#define QIB_6120_SendCtrl_Abort_RMASK 0x1
+
+#define QIB_6120_SendPIOBufBase_OFFS 0x1C8
+#define QIB_6120_SendPIOBufBase_Reserved_LSB 0x35
+#define QIB_6120_SendPIOBufBase_Reserved_RMASK 0x7FF
+#define QIB_6120_SendPIOBufBase_BaseAddr_LargePIO_LSB 0x20
+#define QIB_6120_SendPIOBufBase_BaseAddr_LargePIO_RMASK 0x1FFFFF
+#define QIB_6120_SendPIOBufBase_Reserved1_LSB 0x15
+#define QIB_6120_SendPIOBufBase_Reserved1_RMASK 0x7FF
+#define QIB_6120_SendPIOBufBase_BaseAddr_SmallPIO_LSB 0x0
+#define QIB_6120_SendPIOBufBase_BaseAddr_SmallPIO_RMASK 0x1FFFFF
+
+#define QIB_6120_SendPIOSize_OFFS 0x1D0
+#define QIB_6120_SendPIOSize_Reserved_LSB 0x2D
+#define QIB_6120_SendPIOSize_Reserved_RMASK 0xFFFFF
+#define QIB_6120_SendPIOSize_Size_LargePIO_LSB 0x20
+#define QIB_6120_SendPIOSize_Size_LargePIO_RMASK 0x1FFF
+#define QIB_6120_SendPIOSize_Reserved1_LSB 0xC
+#define QIB_6120_SendPIOSize_Reserved1_RMASK 0xFFFFF
+#define QIB_6120_SendPIOSize_Size_SmallPIO_LSB 0x0
+#define QIB_6120_SendPIOSize_Size_SmallPIO_RMASK 0xFFF
+
+#define QIB_6120_SendPIOBufCnt_OFFS 0x1D8
+#define QIB_6120_SendPIOBufCnt_Reserved_LSB 0x24
+#define QIB_6120_SendPIOBufCnt_Reserved_RMASK 0xFFFFFFF
+#define QIB_6120_SendPIOBufCnt_Num_LargePIO_LSB 0x20
+#define QIB_6120_SendPIOBufCnt_Num_LargePIO_RMASK 0xF
+#define QIB_6120_SendPIOBufCnt_Reserved1_LSB 0x9
+#define QIB_6120_SendPIOBufCnt_Reserved1_RMASK 0x7FFFFF
+#define QIB_6120_SendPIOBufCnt_Num_SmallPIO_LSB 0x0
+#define QIB_6120_SendPIOBufCnt_Num_SmallPIO_RMASK 0x1FF
+
+#define QIB_6120_SendPIOAvailAddr_OFFS 0x1E0
+#define QIB_6120_SendPIOAvailAddr_SendPIOAvailAddr_LSB 0x6
+#define QIB_6120_SendPIOAvailAddr_SendPIOAvailAddr_RMASK 0x3FFFFFFFF
+#define QIB_6120_SendPIOAvailAddr_Reserved_LSB 0x0
+#define QIB_6120_SendPIOAvailAddr_Reserved_RMASK 0x3F
+
+#define QIB_6120_SendBufErr0_OFFS 0x240
+#define QIB_6120_SendBufErr0_SendBufErrPIO_63_0_LSB 0x0
+#define QIB_6120_SendBufErr0_SendBufErrPIO_63_0_RMASK 0x0
+
+#define QIB_6120_RcvHdrAddr0_OFFS 0x280
+#define QIB_6120_RcvHdrAddr0_RcvHdrAddr0_LSB 0x2
+#define QIB_6120_RcvHdrAddr0_RcvHdrAddr0_RMASK 0x3FFFFFFFFF
+#define QIB_6120_RcvHdrAddr0_Reserved_LSB 0x0
+#define QIB_6120_RcvHdrAddr0_Reserved_RMASK 0x3
+
+#define QIB_6120_RcvHdrTailAddr0_OFFS 0x300
+#define QIB_6120_RcvHdrTailAddr0_RcvHdrTailAddr0_LSB 0x2
+#define QIB_6120_RcvHdrTailAddr0_RcvHdrTailAddr0_RMASK 0x3FFFFFFFFF
+#define QIB_6120_RcvHdrTailAddr0_Reserved_LSB 0x0
+#define QIB_6120_RcvHdrTailAddr0_Reserved_RMASK 0x3
+
+#define QIB_6120_SerdesCfg0_OFFS 0x3C0
+#define QIB_6120_SerdesCfg0_DisableIBTxIdleDetect_LSB 0x3F
+#define QIB_6120_SerdesCfg0_DisableIBTxIdleDetect_RMASK 0x1
+#define QIB_6120_SerdesCfg0_Reserved_LSB 0x38
+#define QIB_6120_SerdesCfg0_Reserved_RMASK 0x7F
+#define QIB_6120_SerdesCfg0_RxEqCtl_LSB 0x36
+#define QIB_6120_SerdesCfg0_RxEqCtl_RMASK 0x3
+#define QIB_6120_SerdesCfg0_TxTermAdj_LSB 0x34
+#define QIB_6120_SerdesCfg0_TxTermAdj_RMASK 0x3
+#define QIB_6120_SerdesCfg0_RxTermAdj_LSB 0x32
+#define QIB_6120_SerdesCfg0_RxTermAdj_RMASK 0x3
+#define QIB_6120_SerdesCfg0_TermAdj1_LSB 0x31
+#define QIB_6120_SerdesCfg0_TermAdj1_RMASK 0x1
+#define QIB_6120_SerdesCfg0_TermAdj0_LSB 0x30
+#define QIB_6120_SerdesCfg0_TermAdj0_RMASK 0x1
+#define QIB_6120_SerdesCfg0_LPBKA_LSB 0x2F
+#define QIB_6120_SerdesCfg0_LPBKA_RMASK 0x1
+#define QIB_6120_SerdesCfg0_LPBKB_LSB 0x2E
+#define QIB_6120_SerdesCfg0_LPBKB_RMASK 0x1
+#define QIB_6120_SerdesCfg0_LPBKC_LSB 0x2D
+#define QIB_6120_SerdesCfg0_LPBKC_RMASK 0x1
+#define QIB_6120_SerdesCfg0_LPBKD_LSB 0x2C
+#define QIB_6120_SerdesCfg0_LPBKD_RMASK 0x1
+#define QIB_6120_SerdesCfg0_PW_LSB 0x2B
+#define QIB_6120_SerdesCfg0_PW_RMASK 0x1
+#define QIB_6120_SerdesCfg0_RefSel_LSB 0x29
+#define QIB_6120_SerdesCfg0_RefSel_RMASK 0x3
+#define QIB_6120_SerdesCfg0_ParReset_LSB 0x28
+#define QIB_6120_SerdesCfg0_ParReset_RMASK 0x1
+#define QIB_6120_SerdesCfg0_ParLPBK_LSB 0x27
+#define QIB_6120_SerdesCfg0_ParLPBK_RMASK 0x1
+#define QIB_6120_SerdesCfg0_OffsetEn_LSB 0x26
+#define QIB_6120_SerdesCfg0_OffsetEn_RMASK 0x1
+#define QIB_6120_SerdesCfg0_Offset_LSB 0x1E
+#define QIB_6120_SerdesCfg0_Offset_RMASK 0xFF
+#define QIB_6120_SerdesCfg0_L2PwrDn_LSB 0x1D
+#define QIB_6120_SerdesCfg0_L2PwrDn_RMASK 0x1
+#define QIB_6120_SerdesCfg0_ResetPLL_LSB 0x1C
+#define QIB_6120_SerdesCfg0_ResetPLL_RMASK 0x1
+#define QIB_6120_SerdesCfg0_RxTermEnX_LSB 0x18
+#define QIB_6120_SerdesCfg0_RxTermEnX_RMASK 0xF
+#define QIB_6120_SerdesCfg0_BeaconTxEnX_LSB 0x14
+#define QIB_6120_SerdesCfg0_BeaconTxEnX_RMASK 0xF
+#define QIB_6120_SerdesCfg0_RxDetEnX_LSB 0x10
+#define QIB_6120_SerdesCfg0_RxDetEnX_RMASK 0xF
+#define QIB_6120_SerdesCfg0_TxIdeEnX_LSB 0xC
+#define QIB_6120_SerdesCfg0_TxIdeEnX_RMASK 0xF
+#define QIB_6120_SerdesCfg0_RxIdleEnX_LSB 0x8
+#define QIB_6120_SerdesCfg0_RxIdleEnX_RMASK 0xF
+#define QIB_6120_SerdesCfg0_L1PwrDnA_LSB 0x7
+#define QIB_6120_SerdesCfg0_L1PwrDnA_RMASK 0x1
+#define QIB_6120_SerdesCfg0_L1PwrDnB_LSB 0x6
+#define QIB_6120_SerdesCfg0_L1PwrDnB_RMASK 0x1
+#define QIB_6120_SerdesCfg0_L1PwrDnC_LSB 0x5
+#define QIB_6120_SerdesCfg0_L1PwrDnC_RMASK 0x1
+#define QIB_6120_SerdesCfg0_L1PwrDnD_LSB 0x4
+#define QIB_6120_SerdesCfg0_L1PwrDnD_RMASK 0x1
+#define QIB_6120_SerdesCfg0_ResetA_LSB 0x3
+#define QIB_6120_SerdesCfg0_ResetA_RMASK 0x1
+#define QIB_6120_SerdesCfg0_ResetB_LSB 0x2
+#define QIB_6120_SerdesCfg0_ResetB_RMASK 0x1
+#define QIB_6120_SerdesCfg0_ResetC_LSB 0x1
+#define QIB_6120_SerdesCfg0_ResetC_RMASK 0x1
+#define QIB_6120_SerdesCfg0_ResetD_LSB 0x0
+#define QIB_6120_SerdesCfg0_ResetD_RMASK 0x1
+
+#define QIB_6120_SerdesStat_OFFS 0x3D0
+#define QIB_6120_SerdesStat_Reserved_LSB 0xC
+#define QIB_6120_SerdesStat_Reserved_RMASK 0xFFFFFFFFFFFFF
+#define QIB_6120_SerdesStat_BeaconDetA_LSB 0xB
+#define QIB_6120_SerdesStat_BeaconDetA_RMASK 0x1
+#define QIB_6120_SerdesStat_BeaconDetB_LSB 0xA
+#define QIB_6120_SerdesStat_BeaconDetB_RMASK 0x1
+#define QIB_6120_SerdesStat_BeaconDetC_LSB 0x9
+#define QIB_6120_SerdesStat_BeaconDetC_RMASK 0x1
+#define QIB_6120_SerdesStat_BeaconDetD_LSB 0x8
+#define QIB_6120_SerdesStat_BeaconDetD_RMASK 0x1
+#define QIB_6120_SerdesStat_RxDetA_LSB 0x7
+#define QIB_6120_SerdesStat_RxDetA_RMASK 0x1
+#define QIB_6120_SerdesStat_RxDetB_LSB 0x6
+#define QIB_6120_SerdesStat_RxDetB_RMASK 0x1
+#define QIB_6120_SerdesStat_RxDetC_LSB 0x5
+#define QIB_6120_SerdesStat_RxDetC_RMASK 0x1
+#define QIB_6120_SerdesStat_RxDetD_LSB 0x4
+#define QIB_6120_SerdesStat_RxDetD_RMASK 0x1
+#define QIB_6120_SerdesStat_TxIdleDetA_LSB 0x3
+#define QIB_6120_SerdesStat_TxIdleDetA_RMASK 0x1
+#define QIB_6120_SerdesStat_TxIdleDetB_LSB 0x2
+#define QIB_6120_SerdesStat_TxIdleDetB_RMASK 0x1
+#define QIB_6120_SerdesStat_TxIdleDetC_LSB 0x1
+#define QIB_6120_SerdesStat_TxIdleDetC_RMASK 0x1
+#define QIB_6120_SerdesStat_TxIdleDetD_LSB 0x0
+#define QIB_6120_SerdesStat_TxIdleDetD_RMASK 0x1
+
+#define QIB_6120_XGXSCfg_OFFS 0x3D8
+#define QIB_6120_XGXSCfg_ArmLaunchErrorDisable_LSB 0x3F
+#define QIB_6120_XGXSCfg_ArmLaunchErrorDisable_RMASK 0x1
+#define QIB_6120_XGXSCfg_Reserved_LSB 0x17
+#define QIB_6120_XGXSCfg_Reserved_RMASK 0xFFFFFFFFFF
+#define QIB_6120_XGXSCfg_polarity_inv_LSB 0x13
+#define QIB_6120_XGXSCfg_polarity_inv_RMASK 0xF
+#define QIB_6120_XGXSCfg_link_sync_mask_LSB 0x9
+#define QIB_6120_XGXSCfg_link_sync_mask_RMASK 0x3FF
+#define QIB_6120_XGXSCfg_port_addr_LSB 0x4
+#define QIB_6120_XGXSCfg_port_addr_RMASK 0x1F
+#define QIB_6120_XGXSCfg_mdd_30_LSB 0x3
+#define QIB_6120_XGXSCfg_mdd_30_RMASK 0x1
+#define QIB_6120_XGXSCfg_xcv_resetn_LSB 0x2
+#define QIB_6120_XGXSCfg_xcv_resetn_RMASK 0x1
+#define QIB_6120_XGXSCfg_Reserved1_LSB 0x1
+#define QIB_6120_XGXSCfg_Reserved1_RMASK 0x1
+#define QIB_6120_XGXSCfg_tx_rx_resetn_LSB 0x0
+#define QIB_6120_XGXSCfg_tx_rx_resetn_RMASK 0x1
+
+#define QIB_6120_LBIntCnt_OFFS 0x12000
+
+#define QIB_6120_LBFlowStallCnt_OFFS 0x12008
+
+#define QIB_6120_TxUnsupVLErrCnt_OFFS 0x12018
+
+#define QIB_6120_TxDataPktCnt_OFFS 0x12020
+
+#define QIB_6120_TxFlowPktCnt_OFFS 0x12028
+
+#define QIB_6120_TxDwordCnt_OFFS 0x12030
+
+#define QIB_6120_TxLenErrCnt_OFFS 0x12038
+
+#define QIB_6120_TxMaxMinLenErrCnt_OFFS 0x12040
+
+#define QIB_6120_TxUnderrunCnt_OFFS 0x12048
+
+#define QIB_6120_TxFlowStallCnt_OFFS 0x12050
+
+#define QIB_6120_TxDroppedPktCnt_OFFS 0x12058
+
+#define QIB_6120_RxDroppedPktCnt_OFFS 0x12060
+
+#define QIB_6120_RxDataPktCnt_OFFS 0x12068
+
+#define QIB_6120_RxFlowPktCnt_OFFS 0x12070
+
+#define QIB_6120_RxDwordCnt_OFFS 0x12078
+
+#define QIB_6120_RxLenErrCnt_OFFS 0x12080
+
+#define QIB_6120_RxMaxMinLenErrCnt_OFFS 0x12088
+
+#define QIB_6120_RxICRCErrCnt_OFFS 0x12090
+
+#define QIB_6120_RxVCRCErrCnt_OFFS 0x12098
+
+#define QIB_6120_RxFlowCtrlErrCnt_OFFS 0x120A0
+
+#define QIB_6120_RxBadFormatCnt_OFFS 0x120A8
+
+#define QIB_6120_RxLinkProblemCnt_OFFS 0x120B0
+
+#define QIB_6120_RxEBPCnt_OFFS 0x120B8
+
+#define QIB_6120_RxLPCRCErrCnt_OFFS 0x120C0
+
+#define QIB_6120_RxBufOvflCnt_OFFS 0x120C8
+
+#define QIB_6120_RxTIDFullErrCnt_OFFS 0x120D0
+
+#define QIB_6120_RxTIDValidErrCnt_OFFS 0x120D8
+
+#define QIB_6120_RxPKeyMismatchCnt_OFFS 0x120E0
+
+#define QIB_6120_RxP0HdrEgrOvflCnt_OFFS 0x120E8
+
+#define QIB_6120_IBStatusChangeCnt_OFFS 0x12140
+
+#define QIB_6120_IBLinkErrRecoveryCnt_OFFS 0x12148
+
+#define QIB_6120_IBLinkDownedCnt_OFFS 0x12150
+
+#define QIB_6120_IBSymbolErrCnt_OFFS 0x12158
+
+#define QIB_6120_PcieRetryBufDiagQwordCnt_OFFS 0x12170
+
+#define QIB_6120_RcvEgrArray0_OFFS 0x14000
+
+#define QIB_6120_RcvTIDArray0_OFFS 0x54000
+
+#define QIB_6120_PIOLaunchFIFO_OFFS 0x64000
+
+#define QIB_6120_SendPIOpbcCache_OFFS 0x64800
+
+#define QIB_6120_RcvBuf1_OFFS 0x72000
+
+#define QIB_6120_RcvBuf2_OFFS 0x75000
+
+#define QIB_6120_RcvFlags_OFFS 0x77000
+
+#define QIB_6120_RcvLookupBuf1_OFFS 0x79000
+
+#define QIB_6120_RcvDMABuf_OFFS 0x7B000
+
+#define QIB_6120_MiscRXEIntMem_OFFS 0x7C000
+
+#define QIB_6120_PCIERcvBuf_OFFS 0x80000
+
+#define QIB_6120_PCIERetryBuf_OFFS 0x82000
+
+#define QIB_6120_PCIERcvBufRdToWrAddr_OFFS 0x84000
+
+#define QIB_6120_PIOBuf0_MA_OFFS 0x100000

drivers/infiniband/hw/qib/qib_7220.h

@@ -0,0 +1,156 @@
+#ifndef _QIB_7220_H
+#define _QIB_7220_H
+/*
+ * Copyright (c) 2007, 2009, 2010 QLogic Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+/* grab register-defs auto-generated by HW */
+#include "qib_7220_regs.h"
+
+/* The number of eager receive TIDs for context zero. */
+#define IBA7220_KRCVEGRCNT      2048U
+
+#define IB_7220_LT_STATE_CFGRCVFCFG      0x09
+#define IB_7220_LT_STATE_CFGWAITRMT      0x0a
+#define IB_7220_LT_STATE_TXREVLANES      0x0d
+#define IB_7220_LT_STATE_CFGENH          0x10
+
+struct qib_chip_specific {
+	u64 __iomem *cregbase;
+	u64 *cntrs;
+	u64 *portcntrs;
+	spinlock_t sdepb_lock; /* serdes EPB bus */
+	spinlock_t rcvmod_lock; /* protect rcvctrl shadow changes */
+	spinlock_t gpio_lock; /* RMW of shadows/regs for ExtCtrl and GPIO */
+	u64 hwerrmask;
+	u64 errormask;
+	u64 gpio_out; /* shadow of kr_gpio_out, for rmw ops */
+	u64 gpio_mask; /* shadow the gpio mask register */
+	u64 extctrl; /* shadow the gpio output enable, etc... */
+	u32 ncntrs;
+	u32 nportcntrs;
+	u32 cntrnamelen;
+	u32 portcntrnamelen;
+	u32 numctxts;
+	u32 rcvegrcnt;
+	u32 autoneg_tries;
+	u32 serdes_first_init_done;
+	u32 sdmabufcnt;
+	u32 lastbuf_for_pio;
+	u32 updthresh; /* current AvailUpdThld */
+	u32 updthresh_dflt; /* default AvailUpdThld */
+	int irq;
+	u8 presets_needed;
+	u8 relock_timer_active;
+	char emsgbuf[128];
+	char sdmamsgbuf[192];
+	char bitsmsgbuf[64];
+	struct timer_list relock_timer;
+	unsigned int relock_interval; /* in jiffies */
+};
+
+struct qib_chippport_specific {
+	struct qib_pportdata pportdata;
+	wait_queue_head_t autoneg_wait;
+	struct delayed_work autoneg_work;
+	struct timer_list chase_timer;
+	/*
+	 * these 5 fields are used to establish deltas for IB symbol
+	 * errors and linkrecovery errors.  They can be reported on
+	 * some chips during link negotiation prior to INIT, and with
+	 * DDR when faking DDR negotiations with non-IBTA switches.
+	 * The chip counters are adjusted at driver unload if there is
+	 * a non-zero delta.
+	 */
+	u64 ibdeltainprog;
+	u64 ibsymdelta;
+	u64 ibsymsnap;
+	u64 iblnkerrdelta;
+	u64 iblnkerrsnap;
+	u64 ibcctrl; /* kr_ibcctrl shadow */
+	u64 ibcddrctrl; /* kr_ibcddrctrl shadow */
+	u64 chase_end;
+	u32 last_delay_mult;
+};
+
+/*
+ * This header file provides the declarations and common definitions
+ * for (mostly) manipulation of the SerDes blocks within the IBA7220.
+ * the functions declared should only be called from within other
+ * 7220-related files such as qib_iba7220.c or qib_sd7220.c.
+ */
+int qib_sd7220_presets(struct qib_devdata *dd);
+int qib_sd7220_init(struct qib_devdata *dd);
+int qib_sd7220_prog_ld(struct qib_devdata *dd, int sdnum, u8 *img,
+		       int len, int offset);
+int qib_sd7220_prog_vfy(struct qib_devdata *dd, int sdnum, const u8 *img,
+			int len, int offset);
+void qib_sd7220_clr_ibpar(struct qib_devdata *);
+/*
+ * Below used for sdnum parameter, selecting one of the two sections
+ * used for PCIe, or the single SerDes used for IB, which is the
+ * only one currently used
+ */
+#define IB_7220_SERDES 2
+
+int qib_sd7220_ib_load(struct qib_devdata *dd);
+int qib_sd7220_ib_vfy(struct qib_devdata *dd);
+
+static inline u32 qib_read_kreg32(const struct qib_devdata *dd,
+				  const u16 regno)
+{
+	if (!dd->kregbase || !(dd->flags & QIB_PRESENT))
+		return -1;
+	return readl((u32 __iomem *)&dd->kregbase[regno]);
+}
+
+static inline u64 qib_read_kreg64(const struct qib_devdata *dd,
+				  const u16 regno)
+{
+	if (!dd->kregbase || !(dd->flags & QIB_PRESENT))
+		return -1;
+
+	return readq(&dd->kregbase[regno]);
+}
+
+static inline void qib_write_kreg(const struct qib_devdata *dd,
+				  const u16 regno, u64 value)
+{
+	if (dd->kregbase)
+		writeq(value, &dd->kregbase[regno]);
+}
+
+void set_7220_relock_poll(struct qib_devdata *, int);
+void shutdown_7220_relock_poll(struct qib_devdata *);
+void toggle_7220_rclkrls(struct qib_devdata *);
+
+
+#endif /* _QIB_7220_H */

drivers/infiniband/hw/qib/qib_common.h

@@ -0,0 +1,758 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009, 2010 QLogic Corporation.
+ * All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#ifndef _QIB_COMMON_H
+#define _QIB_COMMON_H
+
+/*
+ * This file contains defines, structures, etc. that are used
+ * to communicate between kernel and user code.
+ */
+
+/* This is the IEEE-assigned OUI for QLogic Inc. QLogic_IB */
+#define QIB_SRC_OUI_1 0x00
+#define QIB_SRC_OUI_2 0x11
+#define QIB_SRC_OUI_3 0x75
+
+/* version of protocol header (known to chip also). In the long run,
+ * we should be able to generate and accept a range of version numbers;
+ * for now we only accept one, and it's compiled in.
+ */
+#define IPS_PROTO_VERSION 2
+
+/*
+ * These are compile time constants that you may want to enable or disable
+ * if you are trying to debug problems with code or performance.
+ * QIB_VERBOSE_TRACING define as 1 if you want additional tracing in
+ * fastpath code
+ * QIB_TRACE_REGWRITES define as 1 if you want register writes to be
+ * traced in faspath code
+ * _QIB_TRACING define as 0 if you want to remove all tracing in a
+ * compilation unit
+ */
+
+/*
+ * The value in the BTH QP field that QLogic_IB uses to differentiate
+ * an qlogic_ib protocol IB packet vs standard IB transport
+ * This it needs to be even (0x656b78), because the LSB is sometimes
+ * used for the MSB of context. The change may cause a problem
+ * interoperating with older software.
+ */
+#define QIB_KD_QP 0x656b78
+
+/*
+ * These are the status bits readable (in ascii form, 64bit value)
+ * from the "status" sysfs file.  For binary compatibility, values
+ * must remain as is; removed states can be reused for different
+ * purposes.
+ */
+#define QIB_STATUS_INITTED       0x1    /* basic initialization done */
+/* Chip has been found and initted */
+#define QIB_STATUS_CHIP_PRESENT 0x20
+/* IB link is at ACTIVE, usable for data traffic */
+#define QIB_STATUS_IB_READY     0x40
+/* link is configured, LID, MTU, etc. have been set */
+#define QIB_STATUS_IB_CONF      0x80
+/* A Fatal hardware error has occurred. */
+#define QIB_STATUS_HWERROR     0x200
+
+/*
+ * The list of usermode accessible registers.  Also see Reg_* later in file.
+ */
+enum qib_ureg {
+	/* (RO)  DMA RcvHdr to be used next. */
+	ur_rcvhdrtail = 0,
+	/* (RW)  RcvHdr entry to be processed next by host. */
+	ur_rcvhdrhead = 1,
+	/* (RO)  Index of next Eager index to use. */
+	ur_rcvegrindextail = 2,
+	/* (RW)  Eager TID to be processed next */
+	ur_rcvegrindexhead = 3,
+	/* For internal use only; max register number. */
+	_QIB_UregMax
+};
+
+/* bit values for spi_runtime_flags */
+#define QIB_RUNTIME_PCIE                0x0002
+#define QIB_RUNTIME_FORCE_WC_ORDER      0x0004
+#define QIB_RUNTIME_RCVHDR_COPY         0x0008
+#define QIB_RUNTIME_MASTER              0x0010
+#define QIB_RUNTIME_RCHK                0x0020
+#define QIB_RUNTIME_NODMA_RTAIL         0x0080
+#define QIB_RUNTIME_SPECIAL_TRIGGER     0x0100
+#define QIB_RUNTIME_SDMA                0x0200
+#define QIB_RUNTIME_FORCE_PIOAVAIL      0x0400
+#define QIB_RUNTIME_PIO_REGSWAPPED      0x0800
+#define QIB_RUNTIME_CTXT_MSB_IN_QP      0x1000
+#define QIB_RUNTIME_CTXT_REDIRECT       0x2000
+#define QIB_RUNTIME_HDRSUPP             0x4000
+
+/*
+ * This structure is returned by qib_userinit() immediately after
+ * open to get implementation-specific info, and info specific to this
+ * instance.
+ *
+ * This struct must have explict pad fields where type sizes
+ * may result in different alignments between 32 and 64 bit
+ * programs, since the 64 bit * bit kernel requires the user code
+ * to have matching offsets
+ */
+struct qib_base_info {
+	/* version of hardware, for feature checking. */
+	__u32 spi_hw_version;
+	/* version of software, for feature checking. */
+	__u32 spi_sw_version;
+	/* QLogic_IB context assigned, goes into sent packets */
+	__u16 spi_ctxt;
+	__u16 spi_subctxt;
+	/*
+	 * IB MTU, packets IB data must be less than this.
+	 * The MTU is in bytes, and will be a multiple of 4 bytes.
+	 */
+	__u32 spi_mtu;
+	/*
+	 * Size of a PIO buffer.  Any given packet's total size must be less
+	 * than this (in words).  Included is the starting control word, so
+	 * if 513 is returned, then total pkt size is 512 words or less.
+	 */
+	__u32 spi_piosize;
+	/* size of the TID cache in qlogic_ib, in entries */
+	__u32 spi_tidcnt;
+	/* size of the TID Eager list in qlogic_ib, in entries */
+	__u32 spi_tidegrcnt;
+	/* size of a single receive header queue entry in words. */
+	__u32 spi_rcvhdrent_size;
+	/*
+	 * Count of receive header queue entries allocated.
+	 * This may be less than the spu_rcvhdrcnt passed in!.
+	 */
+	__u32 spi_rcvhdr_cnt;
+
+	/* per-chip and other runtime features bitmap (QIB_RUNTIME_*) */
+	__u32 spi_runtime_flags;
+
+	/* address where hardware receive header queue is mapped */
+	__u64 spi_rcvhdr_base;
+
+	/* user program. */
+
+	/* base address of eager TID receive buffers used by hardware. */
+	__u64 spi_rcv_egrbufs;
+
+	/* Allocated by initialization code, not by protocol. */
+
+	/*
+	 * Size of each TID buffer in host memory, starting at
+	 * spi_rcv_egrbufs.  The buffers are virtually contiguous.
+	 */
+	__u32 spi_rcv_egrbufsize;
+	/*
+	 * The special QP (queue pair) value that identifies an qlogic_ib
+	 * protocol packet from standard IB packets.  More, probably much
+	 * more, to be added.
+	 */
+	__u32 spi_qpair;
+
+	/*
+	 * User register base for init code, not to be used directly by
+	 * protocol or applications.  Always points to chip registers,
+	 * for normal or shared context.
+	 */
+	__u64 spi_uregbase;
+	/*
+	 * Maximum buffer size in bytes that can be used in a single TID
+	 * entry (assuming the buffer is aligned to this boundary).  This is
+	 * the minimum of what the hardware and software support Guaranteed
+	 * to be a power of 2.
+	 */
+	__u32 spi_tid_maxsize;
+	/*
+	 * alignment of each pio send buffer (byte count
+	 * to add to spi_piobufbase to get to second buffer)
+	 */
+	__u32 spi_pioalign;
+	/*
+	 * The index of the first pio buffer available to this process;
+	 * needed to do lookup in spi_pioavailaddr; not added to
+	 * spi_piobufbase.
+	 */
+	__u32 spi_pioindex;
+	 /* number of buffers mapped for this process */
+	__u32 spi_piocnt;
+
+	/*
+	 * Base address of writeonly pio buffers for this process.
+	 * Each buffer has spi_piosize words, and is aligned on spi_pioalign
+	 * boundaries.  spi_piocnt buffers are mapped from this address
+	 */
+	__u64 spi_piobufbase;
+
+	/*
+	 * Base address of readonly memory copy of the pioavail registers.
+	 * There are 2 bits for each buffer.
+	 */
+	__u64 spi_pioavailaddr;
+
+	/*
+	 * Address where driver updates a copy of the interface and driver
+	 * status (QIB_STATUS_*) as a 64 bit value.  It's followed by a
+	 * link status qword (formerly combined with driver status), then a
+	 * string indicating hardware error, if there was one.
+	 */
+	__u64 spi_status;
+
+	/* number of chip ctxts available to user processes */
+	__u32 spi_nctxts;
+	__u16 spi_unit; /* unit number of chip we are using */
+	__u16 spi_port; /* IB port number we are using */
+	/* num bufs in each contiguous set */
+	__u32 spi_rcv_egrperchunk;
+	/* size in bytes of each contiguous set */
+	__u32 spi_rcv_egrchunksize;
+	/* total size of mmap to cover full rcvegrbuffers */
+	__u32 spi_rcv_egrbuftotlen;
+	__u32 spi_rhf_offset; /* dword offset in hdrqent for rcvhdr flags */
+	/* address of readonly memory copy of the rcvhdrq tail register. */
+	__u64 spi_rcvhdr_tailaddr;
+
+	/*
+	 * shared memory pages for subctxts if ctxt is shared; these cover
+	 * all the processes in the group sharing a single context.
+	 * all have enough space for the num_subcontexts value on this job.
+	 */
+	__u64 spi_subctxt_uregbase;
+	__u64 spi_subctxt_rcvegrbuf;
+	__u64 spi_subctxt_rcvhdr_base;
+
+	/* shared memory page for send buffer disarm status */
+	__u64 spi_sendbuf_status;
+} __attribute__ ((aligned(8)));
+
+/*
+ * This version number is given to the driver by the user code during
+ * initialization in the spu_userversion field of qib_user_info, so
+ * the driver can check for compatibility with user code.
+ *
+ * The major version changes when data structures
+ * change in an incompatible way.  The driver must be the same or higher
+ * for initialization to succeed.  In some cases, a higher version
+ * driver will not interoperate with older software, and initialization
+ * will return an error.
+ */
+#define QIB_USER_SWMAJOR 1
+
+/*
+ * Minor version differences are always compatible
+ * a within a major version, however if user software is larger
+ * than driver software, some new features and/or structure fields
+ * may not be implemented; the user code must deal with this if it
+ * cares, or it must abort after initialization reports the difference.
+ */
+#define QIB_USER_SWMINOR 10
+
+#define QIB_USER_SWVERSION ((QIB_USER_SWMAJOR << 16) | QIB_USER_SWMINOR)
+
+#ifndef QIB_KERN_TYPE
+#define QIB_KERN_TYPE 0
+#define QIB_IDSTR "QLogic kernel.org driver"
+#endif
+
+/*
+ * Similarly, this is the kernel version going back to the user.  It's
+ * slightly different, in that we want to tell if the driver was built as
+ * part of a QLogic release, or from the driver from openfabrics.org,
+ * kernel.org, or a standard distribution, for support reasons.
+ * The high bit is 0 for non-QLogic and 1 for QLogic-built/supplied.
+ *
+ * It's returned by the driver to the user code during initialization in the
+ * spi_sw_version field of qib_base_info, so the user code can in turn
+ * check for compatibility with the kernel.
+*/
+#define QIB_KERN_SWVERSION ((QIB_KERN_TYPE << 31) | QIB_USER_SWVERSION)
+
+/*
+ * This structure is passed to qib_userinit() to tell the driver where
+ * user code buffers are, sizes, etc.   The offsets and sizes of the
+ * fields must remain unchanged, for binary compatibility.  It can
+ * be extended, if userversion is changed so user code can tell, if needed
+ */
+struct qib_user_info {
+	/*
+	 * version of user software, to detect compatibility issues.
+	 * Should be set to QIB_USER_SWVERSION.
+	 */
+	__u32 spu_userversion;
+
+	__u32 _spu_unused2;
+
+	/* size of struct base_info to write to */
+	__u32 spu_base_info_size;
+
+	__u32 _spu_unused3;
+
+	/*
+	 * If two or more processes wish to share a context, each process
+	 * must set the spu_subctxt_cnt and spu_subctxt_id to the same
+	 * values.  The only restriction on the spu_subctxt_id is that
+	 * it be unique for a given node.
+	 */
+	__u16 spu_subctxt_cnt;
+	__u16 spu_subctxt_id;
+
+	__u32 spu_port; /* IB port requested by user if > 0 */
+
+	/*
+	 * address of struct base_info to write to
+	 */
+	__u64 spu_base_info;
+
+} __attribute__ ((aligned(8)));
+
+/* User commands. */
+
+/* 16 available, was: old set up userspace (for old user code) */
+#define QIB_CMD_CTXT_INFO       17      /* find out what resources we got */
+#define QIB_CMD_RECV_CTRL       18      /* control receipt of packets */
+#define QIB_CMD_TID_UPDATE      19      /* update expected TID entries */
+#define QIB_CMD_TID_FREE        20      /* free expected TID entries */
+#define QIB_CMD_SET_PART_KEY    21      /* add partition key */
+/* 22 available, was: return info on slave processes (for old user code) */
+#define QIB_CMD_ASSIGN_CTXT     23      /* allocate HCA and ctxt */
+#define QIB_CMD_USER_INIT       24      /* set up userspace */
+#define QIB_CMD_UNUSED_1        25
+#define QIB_CMD_UNUSED_2        26
+#define QIB_CMD_PIOAVAILUPD     27      /* force an update of PIOAvail reg */
+#define QIB_CMD_POLL_TYPE       28      /* set the kind of polling we want */
+#define QIB_CMD_ARMLAUNCH_CTRL  29      /* armlaunch detection control */
+/* 30 is unused */
+#define QIB_CMD_SDMA_INFLIGHT   31      /* sdma inflight counter request */
+#define QIB_CMD_SDMA_COMPLETE   32      /* sdma completion counter request */
+/* 33 available, was a testing feature  */
+#define QIB_CMD_DISARM_BUFS     34      /* disarm send buffers w/ errors */
+#define QIB_CMD_ACK_EVENT       35      /* ack & clear bits */
+#define QIB_CMD_CPUS_LIST       36      /* list of cpus allocated, for pinned
+					 * processes: qib_cpus_list */
+
+/*
+ * QIB_CMD_ACK_EVENT obsoletes QIB_CMD_DISARM_BUFS, but we keep it for
+ * compatibility with libraries from previous release.   The ACK_EVENT
+ * will take appropriate driver action (if any, just DISARM for now),
+ * then clear the bits passed in as part of the mask.  These bits are
+ * in the first 64bit word at spi_sendbuf_status, and are passed to
+ * the driver in the event_mask union as well.
+ */
+#define _QIB_EVENT_DISARM_BUFS_BIT	0
+#define _QIB_EVENT_LINKDOWN_BIT		1
+#define _QIB_EVENT_LID_CHANGE_BIT	2
+#define _QIB_EVENT_LMC_CHANGE_BIT	3
+#define _QIB_EVENT_SL2VL_CHANGE_BIT	4
+#define _QIB_MAX_EVENT_BIT _QIB_EVENT_SL2VL_CHANGE_BIT
+
+#define QIB_EVENT_DISARM_BUFS_BIT	(1UL << _QIB_EVENT_DISARM_BUFS_BIT)
+#define QIB_EVENT_LINKDOWN_BIT		(1UL << _QIB_EVENT_LINKDOWN_BIT)
+#define QIB_EVENT_LID_CHANGE_BIT	(1UL << _QIB_EVENT_LID_CHANGE_BIT)
+#define QIB_EVENT_LMC_CHANGE_BIT	(1UL << _QIB_EVENT_LMC_CHANGE_BIT)
+#define QIB_EVENT_SL2VL_CHANGE_BIT	(1UL << _QIB_EVENT_SL2VL_CHANGE_BIT)
+
+
+/*
+ * Poll types
+ */
+#define QIB_POLL_TYPE_ANYRCV     0x0
+#define QIB_POLL_TYPE_URGENT     0x1
+
+struct qib_ctxt_info {
+	__u16 num_active;       /* number of active units */
+	__u16 unit;             /* unit (chip) assigned to caller */
+	__u16 port;             /* IB port assigned to caller (1-based) */
+	__u16 ctxt;             /* ctxt on unit assigned to caller */
+	__u16 subctxt;          /* subctxt on unit assigned to caller */
+	__u16 num_ctxts;        /* number of ctxts available on unit */
+	__u16 num_subctxts;     /* number of subctxts opened on ctxt */
+	__u16 rec_cpu;          /* cpu # for affinity (ffff if none) */
+};
+
+struct qib_tid_info {
+	__u32 tidcnt;
+	/* make structure same size in 32 and 64 bit */
+	__u32 tid__unused;
+	/* virtual address of first page in transfer */
+	__u64 tidvaddr;
+	/* pointer (same size 32/64 bit) to __u16 tid array */
+	__u64 tidlist;
+
+	/*
+	 * pointer (same size 32/64 bit) to bitmap of TIDs used
+	 * for this call; checked for being large enough at open
+	 */
+	__u64 tidmap;
+};
+
+struct qib_cmd {
+	__u32 type;                     /* command type */
+	union {
+		struct qib_tid_info tid_info;
+		struct qib_user_info user_info;
+
+		/*
+		 * address in userspace where we should put the sdma
+		 * inflight counter
+		 */
+		__u64 sdma_inflight;
+		/*
+		 * address in userspace where we should put the sdma
+		 * completion counter
+		 */
+		__u64 sdma_complete;
+		/* address in userspace of struct qib_ctxt_info to
+		   write result to */
+		__u64 ctxt_info;
+		/* enable/disable receipt of packets */
+		__u32 recv_ctrl;
+		/* enable/disable armlaunch errors (non-zero to enable) */
+		__u32 armlaunch_ctrl;
+		/* partition key to set */
+		__u16 part_key;
+		/* user address of __u32 bitmask of active slaves */
+		__u64 slave_mask_addr;
+		/* type of polling we want */
+		__u16 poll_type;
+		/* back pressure enable bit for one particular context */
+		__u8 ctxt_bp;
+		/* qib_user_event_ack(), IPATH_EVENT_* bits */
+		__u64 event_mask;
+	} cmd;
+};
+
+struct qib_iovec {
+	/* Pointer to data, but same size 32 and 64 bit */
+	__u64 iov_base;
+
+	/*
+	 * Length of data; don't need 64 bits, but want
+	 * qib_sendpkt to remain same size as before 32 bit changes, so...
+	 */
+	__u64 iov_len;
+};
+
+/*
+ * Describes a single packet for send.  Each packet can have one or more
+ * buffers, but the total length (exclusive of IB headers) must be less
+ * than the MTU, and if using the PIO method, entire packet length,
+ * including IB headers, must be less than the qib_piosize value (words).
+ * Use of this necessitates including sys/uio.h
+ */
+struct __qib_sendpkt {
+	__u32 sps_flags;        /* flags for packet (TBD) */
+	__u32 sps_cnt;          /* number of entries to use in sps_iov */
+	/* array of iov's describing packet. TEMPORARY */
+	struct qib_iovec sps_iov[4];
+};
+
+/*
+ * Diagnostics can send a packet by "writing" the following
+ * structs to the diag data special file.
+ * This allows a custom
+ * pbc (+ static rate) qword, so that special modes and deliberate
+ * changes to CRCs can be used. The elements were also re-ordered
+ * for better alignment and to avoid padding issues.
+ */
+#define _DIAG_XPKT_VERS 3
+struct qib_diag_xpkt {
+	__u16 version;
+	__u16 unit;
+	__u16 port;
+	__u16 len;
+	__u64 data;
+	__u64 pbc_wd;
+};
+
+/*
+ * Data layout in I2C flash (for GUID, etc.)
+ * All fields are little-endian binary unless otherwise stated
+ */
+#define QIB_FLASH_VERSION 2
+struct qib_flash {
+	/* flash layout version (QIB_FLASH_VERSION) */
+	__u8 if_fversion;
+	/* checksum protecting if_length bytes */
+	__u8 if_csum;
+	/*
+	 * valid length (in use, protected by if_csum), including
+	 * if_fversion and if_csum themselves)
+	 */
+	__u8 if_length;
+	/* the GUID, in network order */
+	__u8 if_guid[8];
+	/* number of GUIDs to use, starting from if_guid */
+	__u8 if_numguid;
+	/* the (last 10 characters of) board serial number, in ASCII */
+	char if_serial[12];
+	/* board mfg date (YYYYMMDD ASCII) */
+	char if_mfgdate[8];
+	/* last board rework/test date (YYYYMMDD ASCII) */
+	char if_testdate[8];
+	/* logging of error counts, TBD */
+	__u8 if_errcntp[4];
+	/* powered on hours, updated at driver unload */
+	__u8 if_powerhour[2];
+	/* ASCII free-form comment field */
+	char if_comment[32];
+	/* Backwards compatible prefix for longer QLogic Serial Numbers */
+	char if_sprefix[4];
+	/* 82 bytes used, min flash size is 128 bytes */
+	__u8 if_future[46];
+};
+
+/*
+ * These are the counters implemented in the chip, and are listed in order.
+ * The InterCaps naming is taken straight from the chip spec.
+ */
+struct qlogic_ib_counters {
+	__u64 LBIntCnt;
+	__u64 LBFlowStallCnt;
+	__u64 TxSDmaDescCnt;    /* was Reserved1 */
+	__u64 TxUnsupVLErrCnt;
+	__u64 TxDataPktCnt;
+	__u64 TxFlowPktCnt;
+	__u64 TxDwordCnt;
+	__u64 TxLenErrCnt;
+	__u64 TxMaxMinLenErrCnt;
+	__u64 TxUnderrunCnt;
+	__u64 TxFlowStallCnt;
+	__u64 TxDroppedPktCnt;
+	__u64 RxDroppedPktCnt;
+	__u64 RxDataPktCnt;
+	__u64 RxFlowPktCnt;
+	__u64 RxDwordCnt;
+	__u64 RxLenErrCnt;
+	__u64 RxMaxMinLenErrCnt;
+	__u64 RxICRCErrCnt;
+	__u64 RxVCRCErrCnt;
+	__u64 RxFlowCtrlErrCnt;
+	__u64 RxBadFormatCnt;
+	__u64 RxLinkProblemCnt;
+	__u64 RxEBPCnt;
+	__u64 RxLPCRCErrCnt;
+	__u64 RxBufOvflCnt;
+	__u64 RxTIDFullErrCnt;
+	__u64 RxTIDValidErrCnt;
+	__u64 RxPKeyMismatchCnt;
+	__u64 RxP0HdrEgrOvflCnt;
+	__u64 RxP1HdrEgrOvflCnt;
+	__u64 RxP2HdrEgrOvflCnt;
+	__u64 RxP3HdrEgrOvflCnt;
+	__u64 RxP4HdrEgrOvflCnt;
+	__u64 RxP5HdrEgrOvflCnt;
+	__u64 RxP6HdrEgrOvflCnt;
+	__u64 RxP7HdrEgrOvflCnt;
+	__u64 RxP8HdrEgrOvflCnt;
+	__u64 RxP9HdrEgrOvflCnt;
+	__u64 RxP10HdrEgrOvflCnt;
+	__u64 RxP11HdrEgrOvflCnt;
+	__u64 RxP12HdrEgrOvflCnt;
+	__u64 RxP13HdrEgrOvflCnt;
+	__u64 RxP14HdrEgrOvflCnt;
+	__u64 RxP15HdrEgrOvflCnt;
+	__u64 RxP16HdrEgrOvflCnt;
+	__u64 IBStatusChangeCnt;
+	__u64 IBLinkErrRecoveryCnt;
+	__u64 IBLinkDownedCnt;
+	__u64 IBSymbolErrCnt;
+	__u64 RxVL15DroppedPktCnt;
+	__u64 RxOtherLocalPhyErrCnt;
+	__u64 PcieRetryBufDiagQwordCnt;
+	__u64 ExcessBufferOvflCnt;
+	__u64 LocalLinkIntegrityErrCnt;
+	__u64 RxVlErrCnt;
+	__u64 RxDlidFltrCnt;
+};
+
+/*
+ * The next set of defines are for packet headers, and chip register
+ * and memory bits that are visible to and/or used by user-mode software.
+ */
+
+/* RcvHdrFlags bits */
+#define QLOGIC_IB_RHF_LENGTH_MASK 0x7FF
+#define QLOGIC_IB_RHF_LENGTH_SHIFT 0
+#define QLOGIC_IB_RHF_RCVTYPE_MASK 0x7
+#define QLOGIC_IB_RHF_RCVTYPE_SHIFT 11
+#define QLOGIC_IB_RHF_EGRINDEX_MASK 0xFFF
+#define QLOGIC_IB_RHF_EGRINDEX_SHIFT 16
+#define QLOGIC_IB_RHF_SEQ_MASK 0xF
+#define QLOGIC_IB_RHF_SEQ_SHIFT 0
+#define QLOGIC_IB_RHF_HDRQ_OFFSET_MASK 0x7FF
+#define QLOGIC_IB_RHF_HDRQ_OFFSET_SHIFT 4
+#define QLOGIC_IB_RHF_H_ICRCERR   0x80000000
+#define QLOGIC_IB_RHF_H_VCRCERR   0x40000000
+#define QLOGIC_IB_RHF_H_PARITYERR 0x20000000
+#define QLOGIC_IB_RHF_H_LENERR    0x10000000
+#define QLOGIC_IB_RHF_H_MTUERR    0x08000000
+#define QLOGIC_IB_RHF_H_IHDRERR   0x04000000
+#define QLOGIC_IB_RHF_H_TIDERR    0x02000000
+#define QLOGIC_IB_RHF_H_MKERR     0x01000000
+#define QLOGIC_IB_RHF_H_IBERR     0x00800000
+#define QLOGIC_IB_RHF_H_ERR_MASK  0xFF800000
+#define QLOGIC_IB_RHF_L_USE_EGR   0x80000000
+#define QLOGIC_IB_RHF_L_SWA       0x00008000
+#define QLOGIC_IB_RHF_L_SWB       0x00004000
+
+/* qlogic_ib header fields */
+#define QLOGIC_IB_I_VERS_MASK 0xF
+#define QLOGIC_IB_I_VERS_SHIFT 28
+#define QLOGIC_IB_I_CTXT_MASK 0xF
+#define QLOGIC_IB_I_CTXT_SHIFT 24
+#define QLOGIC_IB_I_TID_MASK 0x7FF
+#define QLOGIC_IB_I_TID_SHIFT 13
+#define QLOGIC_IB_I_OFFSET_MASK 0x1FFF
+#define QLOGIC_IB_I_OFFSET_SHIFT 0
+
+/* K_PktFlags bits */
+#define QLOGIC_IB_KPF_INTR 0x1
+#define QLOGIC_IB_KPF_SUBCTXT_MASK 0x3
+#define QLOGIC_IB_KPF_SUBCTXT_SHIFT 1
+
+#define QLOGIC_IB_MAX_SUBCTXT   4
+
+/* SendPIO per-buffer control */
+#define QLOGIC_IB_SP_TEST    0x40
+#define QLOGIC_IB_SP_TESTEBP 0x20
+#define QLOGIC_IB_SP_TRIGGER_SHIFT  15
+
+/* SendPIOAvail bits */
+#define QLOGIC_IB_SENDPIOAVAIL_BUSY_SHIFT 1
+#define QLOGIC_IB_SENDPIOAVAIL_CHECK_SHIFT 0
+
+/* qlogic_ib header format */
+struct qib_header {
+	/*
+	 * Version - 4 bits, Context - 4 bits, TID - 10 bits and Offset -
+	 * 14 bits before ECO change ~28 Dec 03.  After that, Vers 4,
+	 * Context 4, TID 11, offset 13.
+	 */
+	__le32 ver_ctxt_tid_offset;
+	__le16 chksum;
+	__le16 pkt_flags;
+};
+
+/*
+ * qlogic_ib user message header format.
+ * This structure contains the first 4 fields common to all protocols
+ * that employ qlogic_ib.
+ */
+struct qib_message_header {
+	__be16 lrh[4];
+	__be32 bth[3];
+	/* fields below this point are in host byte order */
+	struct qib_header iph;
+	__u8 sub_opcode;
+};
+
+/* IB - LRH header consts */
+#define QIB_LRH_GRH 0x0003      /* 1. word of IB LRH - next header: GRH */
+#define QIB_LRH_BTH 0x0002      /* 1. word of IB LRH - next header: BTH */
+
+/* misc. */
+#define SIZE_OF_CRC 1
+
+#define QIB_DEFAULT_P_KEY 0xFFFF
+#define QIB_PERMISSIVE_LID 0xFFFF
+#define QIB_AETH_CREDIT_SHIFT 24
+#define QIB_AETH_CREDIT_MASK 0x1F
+#define QIB_AETH_CREDIT_INVAL 0x1F
+#define QIB_PSN_MASK 0xFFFFFF
+#define QIB_MSN_MASK 0xFFFFFF
+#define QIB_QPN_MASK 0xFFFFFF
+#define QIB_MULTICAST_LID_BASE 0xC000
+#define QIB_EAGER_TID_ID QLOGIC_IB_I_TID_MASK
+#define QIB_MULTICAST_QPN 0xFFFFFF
+
+/* Receive Header Queue: receive type (from qlogic_ib) */
+#define RCVHQ_RCV_TYPE_EXPECTED  0
+#define RCVHQ_RCV_TYPE_EAGER     1
+#define RCVHQ_RCV_TYPE_NON_KD    2
+#define RCVHQ_RCV_TYPE_ERROR     3
+
+#define QIB_HEADER_QUEUE_WORDS 9
+
+/* functions for extracting fields from rcvhdrq entries for the driver.
+ */
+static inline __u32 qib_hdrget_err_flags(const __le32 *rbuf)
+{
+	return __le32_to_cpu(rbuf[1]) & QLOGIC_IB_RHF_H_ERR_MASK;
+}
+
+static inline __u32 qib_hdrget_rcv_type(const __le32 *rbuf)
+{
+	return (__le32_to_cpu(rbuf[0]) >> QLOGIC_IB_RHF_RCVTYPE_SHIFT) &
+		QLOGIC_IB_RHF_RCVTYPE_MASK;
+}
+
+static inline __u32 qib_hdrget_length_in_bytes(const __le32 *rbuf)
+{
+	return ((__le32_to_cpu(rbuf[0]) >> QLOGIC_IB_RHF_LENGTH_SHIFT) &
+		QLOGIC_IB_RHF_LENGTH_MASK) << 2;
+}
+
+static inline __u32 qib_hdrget_index(const __le32 *rbuf)
+{
+	return (__le32_to_cpu(rbuf[0]) >> QLOGIC_IB_RHF_EGRINDEX_SHIFT) &
+		QLOGIC_IB_RHF_EGRINDEX_MASK;
+}
+
+static inline __u32 qib_hdrget_seq(const __le32 *rbuf)
+{
+	return (__le32_to_cpu(rbuf[1]) >> QLOGIC_IB_RHF_SEQ_SHIFT) &
+		QLOGIC_IB_RHF_SEQ_MASK;
+}
+
+static inline __u32 qib_hdrget_offset(const __le32 *rbuf)
+{
+	return (__le32_to_cpu(rbuf[1]) >> QLOGIC_IB_RHF_HDRQ_OFFSET_SHIFT) &
+		QLOGIC_IB_RHF_HDRQ_OFFSET_MASK;
+}
+
+static inline __u32 qib_hdrget_use_egr_buf(const __le32 *rbuf)
+{
+	return __le32_to_cpu(rbuf[0]) & QLOGIC_IB_RHF_L_USE_EGR;
+}
+
+static inline __u32 qib_hdrget_qib_ver(__le32 hdrword)
+{
+	return (__le32_to_cpu(hdrword) >> QLOGIC_IB_I_VERS_SHIFT) &
+		QLOGIC_IB_I_VERS_MASK;
+}
+
+#endif                          /* _QIB_COMMON_H */

drivers/infiniband/hw/qib/qib_cq.c

@@ -0,0 +1,484 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2010 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/err.h>
+#include <linux/slab.h>
+#include <linux/vmalloc.h>
+
+#include "qib_verbs.h"
+
+/**
+ * qib_cq_enter - add a new entry to the completion queue
+ * @cq: completion queue
+ * @entry: work completion entry to add
+ * @sig: true if @entry is a solicitated entry
+ *
+ * This may be called with qp->s_lock held.
+ */
+void qib_cq_enter(struct qib_cq *cq, struct ib_wc *entry, int solicited)
+{
+	struct qib_cq_wc *wc;
+	unsigned long flags;
+	u32 head;
+	u32 next;
+
+	spin_lock_irqsave(&cq->lock, flags);
+
+	/*
+	 * Note that the head pointer might be writable by user processes.
+	 * Take care to verify it is a sane value.
+	 */
+	wc = cq->queue;
+	head = wc->head;
+	if (head >= (unsigned) cq->ibcq.cqe) {
+		head = cq->ibcq.cqe;
+		next = 0;
+	} else
+		next = head + 1;
+	if (unlikely(next == wc->tail)) {
+		spin_unlock_irqrestore(&cq->lock, flags);
+		if (cq->ibcq.event_handler) {
+			struct ib_event ev;
+
+			ev.device = cq->ibcq.device;
+			ev.element.cq = &cq->ibcq;
+			ev.event = IB_EVENT_CQ_ERR;
+			cq->ibcq.event_handler(&ev, cq->ibcq.cq_context);
+		}
+		return;
+	}
+	if (cq->ip) {
+		wc->uqueue[head].wr_id = entry->wr_id;
+		wc->uqueue[head].status = entry->status;
+		wc->uqueue[head].opcode = entry->opcode;
+		wc->uqueue[head].vendor_err = entry->vendor_err;
+		wc->uqueue[head].byte_len = entry->byte_len;
+		wc->uqueue[head].ex.imm_data =
+			(__u32 __force)entry->ex.imm_data;
+		wc->uqueue[head].qp_num = entry->qp->qp_num;
+		wc->uqueue[head].src_qp = entry->src_qp;
+		wc->uqueue[head].wc_flags = entry->wc_flags;
+		wc->uqueue[head].pkey_index = entry->pkey_index;
+		wc->uqueue[head].slid = entry->slid;
+		wc->uqueue[head].sl = entry->sl;
+		wc->uqueue[head].dlid_path_bits = entry->dlid_path_bits;
+		wc->uqueue[head].port_num = entry->port_num;
+		/* Make sure entry is written before the head index. */
+		smp_wmb();
+	} else
+		wc->kqueue[head] = *entry;
+	wc->head = next;
+
+	if (cq->notify == IB_CQ_NEXT_COMP ||
+	    (cq->notify == IB_CQ_SOLICITED && solicited)) {
+		cq->notify = IB_CQ_NONE;
+		cq->triggered++;
+		/*
+		 * This will cause send_complete() to be called in
+		 * another thread.
+		 */
+		queue_work(qib_cq_wq, &cq->comptask);
+	}
+
+	spin_unlock_irqrestore(&cq->lock, flags);
+}
+
+/**
+ * qib_poll_cq - poll for work completion entries
+ * @ibcq: the completion queue to poll
+ * @num_entries: the maximum number of entries to return
+ * @entry: pointer to array where work completions are placed
+ *
+ * Returns the number of completion entries polled.
+ *
+ * This may be called from interrupt context.  Also called by ib_poll_cq()
+ * in the generic verbs code.
+ */
+int qib_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *entry)
+{
+	struct qib_cq *cq = to_icq(ibcq);
+	struct qib_cq_wc *wc;
+	unsigned long flags;
+	int npolled;
+	u32 tail;
+
+	/* The kernel can only poll a kernel completion queue */
+	if (cq->ip) {
+		npolled = -EINVAL;
+		goto bail;
+	}
+
+	spin_lock_irqsave(&cq->lock, flags);
+
+	wc = cq->queue;
+	tail = wc->tail;
+	if (tail > (u32) cq->ibcq.cqe)
+		tail = (u32) cq->ibcq.cqe;
+	for (npolled = 0; npolled < num_entries; ++npolled, ++entry) {
+		if (tail == wc->head)
+			break;
+		/* The kernel doesn't need a RMB since it has the lock. */
+		*entry = wc->kqueue[tail];
+		if (tail >= cq->ibcq.cqe)
+			tail = 0;
+		else
+			tail++;
+	}
+	wc->tail = tail;
+
+	spin_unlock_irqrestore(&cq->lock, flags);
+
+bail:
+	return npolled;
+}
+
+static void send_complete(struct work_struct *work)
+{
+	struct qib_cq *cq = container_of(work, struct qib_cq, comptask);
+
+	/*
+	 * The completion handler will most likely rearm the notification
+	 * and poll for all pending entries.  If a new completion entry
+	 * is added while we are in this routine, queue_work()
+	 * won't call us again until we return so we check triggered to
+	 * see if we need to call the handler again.
+	 */
+	for (;;) {
+		u8 triggered = cq->triggered;
+
+		/*
+		 * IPoIB connected mode assumes the callback is from a
+		 * soft IRQ. We simulate this by blocking "bottom halves".
+		 * See the implementation for ipoib_cm_handle_tx_wc(),
+		 * netif_tx_lock_bh() and netif_tx_lock().
+		 */
+		local_bh_disable();
+		cq->ibcq.comp_handler(&cq->ibcq, cq->ibcq.cq_context);
+		local_bh_enable();
+
+		if (cq->triggered == triggered)
+			return;
+	}
+}
+
+/**
+ * qib_create_cq - create a completion queue
+ * @ibdev: the device this completion queue is attached to
+ * @entries: the minimum size of the completion queue
+ * @context: unused by the QLogic_IB driver
+ * @udata: user data for libibverbs.so
+ *
+ * Returns a pointer to the completion queue or negative errno values
+ * for failure.
+ *
+ * Called by ib_create_cq() in the generic verbs code.
+ */
+struct ib_cq *qib_create_cq(struct ib_device *ibdev, int entries,
+			    int comp_vector, struct ib_ucontext *context,
+			    struct ib_udata *udata)
+{
+	struct qib_ibdev *dev = to_idev(ibdev);
+	struct qib_cq *cq;
+	struct qib_cq_wc *wc;
+	struct ib_cq *ret;
+	u32 sz;
+
+	if (entries < 1 || entries > ib_qib_max_cqes) {
+		ret = ERR_PTR(-EINVAL);
+		goto done;
+	}
+
+	/* Allocate the completion queue structure. */
+	cq = kmalloc(sizeof(*cq), GFP_KERNEL);
+	if (!cq) {
+		ret = ERR_PTR(-ENOMEM);
+		goto done;
+	}
+
+	/*
+	 * Allocate the completion queue entries and head/tail pointers.
+	 * This is allocated separately so that it can be resized and
+	 * also mapped into user space.
+	 * We need to use vmalloc() in order to support mmap and large
+	 * numbers of entries.
+	 */
+	sz = sizeof(*wc);
+	if (udata && udata->outlen >= sizeof(__u64))
+		sz += sizeof(struct ib_uverbs_wc) * (entries + 1);
+	else
+		sz += sizeof(struct ib_wc) * (entries + 1);
+	wc = vmalloc_user(sz);
+	if (!wc) {
+		ret = ERR_PTR(-ENOMEM);
+		goto bail_cq;
+	}
+
+	/*
+	 * Return the address of the WC as the offset to mmap.
+	 * See qib_mmap() for details.
+	 */
+	if (udata && udata->outlen >= sizeof(__u64)) {
+		int err;
+
+		cq->ip = qib_create_mmap_info(dev, sz, context, wc);
+		if (!cq->ip) {
+			ret = ERR_PTR(-ENOMEM);
+			goto bail_wc;
+		}
+
+		err = ib_copy_to_udata(udata, &cq->ip->offset,
+				       sizeof(cq->ip->offset));
+		if (err) {
+			ret = ERR_PTR(err);
+			goto bail_ip;
+		}
+	} else
+		cq->ip = NULL;
+
+	spin_lock(&dev->n_cqs_lock);
+	if (dev->n_cqs_allocated == ib_qib_max_cqs) {
+		spin_unlock(&dev->n_cqs_lock);
+		ret = ERR_PTR(-ENOMEM);
+		goto bail_ip;
+	}
+
+	dev->n_cqs_allocated++;
+	spin_unlock(&dev->n_cqs_lock);
+
+	if (cq->ip) {
+		spin_lock_irq(&dev->pending_lock);
+		list_add(&cq->ip->pending_mmaps, &dev->pending_mmaps);
+		spin_unlock_irq(&dev->pending_lock);
+	}
+
+	/*
+	 * ib_create_cq() will initialize cq->ibcq except for cq->ibcq.cqe.
+	 * The number of entries should be >= the number requested or return
+	 * an error.
+	 */
+	cq->ibcq.cqe = entries;
+	cq->notify = IB_CQ_NONE;
+	cq->triggered = 0;
+	spin_lock_init(&cq->lock);
+	INIT_WORK(&cq->comptask, send_complete);
+	wc->head = 0;
+	wc->tail = 0;
+	cq->queue = wc;
+
+	ret = &cq->ibcq;
+
+	goto done;
+
+bail_ip:
+	kfree(cq->ip);
+bail_wc:
+	vfree(wc);
+bail_cq:
+	kfree(cq);
+done:
+	return ret;
+}
+
+/**
+ * qib_destroy_cq - destroy a completion queue
+ * @ibcq: the completion queue to destroy.
+ *
+ * Returns 0 for success.
+ *
+ * Called by ib_destroy_cq() in the generic verbs code.
+ */
+int qib_destroy_cq(struct ib_cq *ibcq)
+{
+	struct qib_ibdev *dev = to_idev(ibcq->device);
+	struct qib_cq *cq = to_icq(ibcq);
+
+	flush_work(&cq->comptask);
+	spin_lock(&dev->n_cqs_lock);
+	dev->n_cqs_allocated--;
+	spin_unlock(&dev->n_cqs_lock);
+	if (cq->ip)
+		kref_put(&cq->ip->ref, qib_release_mmap_info);
+	else
+		vfree(cq->queue);
+	kfree(cq);
+
+	return 0;
+}
+
+/**
+ * qib_req_notify_cq - change the notification type for a completion queue
+ * @ibcq: the completion queue
+ * @notify_flags: the type of notification to request
+ *
+ * Returns 0 for success.
+ *
+ * This may be called from interrupt context.  Also called by
+ * ib_req_notify_cq() in the generic verbs code.
+ */
+int qib_req_notify_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags notify_flags)
+{
+	struct qib_cq *cq = to_icq(ibcq);
+	unsigned long flags;
+	int ret = 0;
+
+	spin_lock_irqsave(&cq->lock, flags);
+	/*
+	 * Don't change IB_CQ_NEXT_COMP to IB_CQ_SOLICITED but allow
+	 * any other transitions (see C11-31 and C11-32 in ch. 11.4.2.2).
+	 */
+	if (cq->notify != IB_CQ_NEXT_COMP)
+		cq->notify = notify_flags & IB_CQ_SOLICITED_MASK;
+
+	if ((notify_flags & IB_CQ_REPORT_MISSED_EVENTS) &&
+	    cq->queue->head != cq->queue->tail)
+		ret = 1;
+
+	spin_unlock_irqrestore(&cq->lock, flags);
+
+	return ret;
+}
+
+/**
+ * qib_resize_cq - change the size of the CQ
+ * @ibcq: the completion queue
+ *
+ * Returns 0 for success.
+ */
+int qib_resize_cq(struct ib_cq *ibcq, int cqe, struct ib_udata *udata)
+{
+	struct qib_cq *cq = to_icq(ibcq);
+	struct qib_cq_wc *old_wc;
+	struct qib_cq_wc *wc;
+	u32 head, tail, n;
+	int ret;
+	u32 sz;
+
+	if (cqe < 1 || cqe > ib_qib_max_cqes) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	/*
+	 * Need to use vmalloc() if we want to support large #s of entries.
+	 */
+	sz = sizeof(*wc);
+	if (udata && udata->outlen >= sizeof(__u64))
+		sz += sizeof(struct ib_uverbs_wc) * (cqe + 1);
+	else
+		sz += sizeof(struct ib_wc) * (cqe + 1);
+	wc = vmalloc_user(sz);
+	if (!wc) {
+		ret = -ENOMEM;
+		goto bail;
+	}
+
+	/* Check that we can write the offset to mmap. */
+	if (udata && udata->outlen >= sizeof(__u64)) {
+		__u64 offset = 0;
+
+		ret = ib_copy_to_udata(udata, &offset, sizeof(offset));
+		if (ret)
+			goto bail_free;
+	}
+
+	spin_lock_irq(&cq->lock);
+	/*
+	 * Make sure head and tail are sane since they
+	 * might be user writable.
+	 */
+	old_wc = cq->queue;
+	head = old_wc->head;
+	if (head > (u32) cq->ibcq.cqe)
+		head = (u32) cq->ibcq.cqe;
+	tail = old_wc->tail;
+	if (tail > (u32) cq->ibcq.cqe)
+		tail = (u32) cq->ibcq.cqe;
+	if (head < tail)
+		n = cq->ibcq.cqe + 1 + head - tail;
+	else
+		n = head - tail;
+	if (unlikely((u32)cqe < n)) {
+		ret = -EINVAL;
+		goto bail_unlock;
+	}
+	for (n = 0; tail != head; n++) {
+		if (cq->ip)
+			wc->uqueue[n] = old_wc->uqueue[tail];
+		else
+			wc->kqueue[n] = old_wc->kqueue[tail];
+		if (tail == (u32) cq->ibcq.cqe)
+			tail = 0;
+		else
+			tail++;
+	}
+	cq->ibcq.cqe = cqe;
+	wc->head = n;
+	wc->tail = 0;
+	cq->queue = wc;
+	spin_unlock_irq(&cq->lock);
+
+	vfree(old_wc);
+
+	if (cq->ip) {
+		struct qib_ibdev *dev = to_idev(ibcq->device);
+		struct qib_mmap_info *ip = cq->ip;
+
+		qib_update_mmap_info(dev, ip, sz, wc);
+
+		/*
+		 * Return the offset to mmap.
+		 * See qib_mmap() for details.
+		 */
+		if (udata && udata->outlen >= sizeof(__u64)) {
+			ret = ib_copy_to_udata(udata, &ip->offset,
+					       sizeof(ip->offset));
+			if (ret)
+				goto bail;
+		}
+
+		spin_lock_irq(&dev->pending_lock);
+		if (list_empty(&ip->pending_mmaps))
+			list_add(&ip->pending_mmaps, &dev->pending_mmaps);
+		spin_unlock_irq(&dev->pending_lock);
+	}
+
+	ret = 0;
+	goto bail;
+
+bail_unlock:
+	spin_unlock_irq(&cq->lock);
+bail_free:
+	vfree(wc);
+bail:
+	return ret;
+}

drivers/infiniband/hw/qib/qib_diag.c

@@ -0,0 +1,894 @@
+/*
+ * Copyright (c) 2010 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+/*
+ * This file contains support for diagnostic functions.  It is accessed by
+ * opening the qib_diag device, normally minor number 129.  Diagnostic use
+ * of the QLogic_IB chip may render the chip or board unusable until the
+ * driver is unloaded, or in some cases, until the system is rebooted.
+ *
+ * Accesses to the chip through this interface are not similar to going
+ * through the /sys/bus/pci resource mmap interface.
+ */
+
+#include <linux/io.h>
+#include <linux/pci.h>
+#include <linux/poll.h>
+#include <linux/vmalloc.h>
+#include <linux/fs.h>
+#include <linux/uaccess.h>
+
+#include "qib.h"
+#include "qib_common.h"
+
+/*
+ * Each client that opens the diag device must read then write
+ * offset 0, to prevent lossage from random cat or od. diag_state
+ * sequences this "handshake".
+ */
+enum diag_state { UNUSED = 0, OPENED, INIT, READY };
+
+/* State for an individual client. PID so children cannot abuse handshake */
+static struct qib_diag_client {
+	struct qib_diag_client *next;
+	struct qib_devdata *dd;
+	pid_t pid;
+	enum diag_state state;
+} *client_pool;
+
+/*
+ * Get a client struct. Recycled if possible, else kmalloc.
+ * Must be called with qib_mutex held
+ */
+static struct qib_diag_client *get_client(struct qib_devdata *dd)
+{
+	struct qib_diag_client *dc;
+
+	dc = client_pool;
+	if (dc)
+		/* got from pool remove it and use */
+		client_pool = dc->next;
+	else
+		/* None in pool, alloc and init */
+		dc = kmalloc(sizeof *dc, GFP_KERNEL);
+
+	if (dc) {
+		dc->next = NULL;
+		dc->dd = dd;
+		dc->pid = current->pid;
+		dc->state = OPENED;
+	}
+	return dc;
+}
+
+/*
+ * Return to pool. Must be called with qib_mutex held
+ */
+static void return_client(struct qib_diag_client *dc)
+{
+	struct qib_devdata *dd = dc->dd;
+	struct qib_diag_client *tdc, *rdc;
+
+	rdc = NULL;
+	if (dc == dd->diag_client) {
+		dd->diag_client = dc->next;
+		rdc = dc;
+	} else {
+		tdc = dc->dd->diag_client;
+		while (tdc) {
+			if (dc == tdc->next) {
+				tdc->next = dc->next;
+				rdc = dc;
+				break;
+			}
+			tdc = tdc->next;
+		}
+	}
+	if (rdc) {
+		rdc->state = UNUSED;
+		rdc->dd = NULL;
+		rdc->pid = 0;
+		rdc->next = client_pool;
+		client_pool = rdc;
+	}
+}
+
+static int qib_diag_open(struct inode *in, struct file *fp);
+static int qib_diag_release(struct inode *in, struct file *fp);
+static ssize_t qib_diag_read(struct file *fp, char __user *data,
+			     size_t count, loff_t *off);
+static ssize_t qib_diag_write(struct file *fp, const char __user *data,
+			      size_t count, loff_t *off);
+
+static const struct file_operations diag_file_ops = {
+	.owner = THIS_MODULE,
+	.write = qib_diag_write,
+	.read = qib_diag_read,
+	.open = qib_diag_open,
+	.release = qib_diag_release
+};
+
+static atomic_t diagpkt_count = ATOMIC_INIT(0);
+static struct cdev *diagpkt_cdev;
+static struct device *diagpkt_device;
+
+static ssize_t qib_diagpkt_write(struct file *fp, const char __user *data,
+				 size_t count, loff_t *off);
+
+static const struct file_operations diagpkt_file_ops = {
+	.owner = THIS_MODULE,
+	.write = qib_diagpkt_write,
+};
+
+int qib_diag_add(struct qib_devdata *dd)
+{
+	char name[16];
+	int ret = 0;
+
+	if (atomic_inc_return(&diagpkt_count) == 1) {
+		ret = qib_cdev_init(QIB_DIAGPKT_MINOR, "ipath_diagpkt",
+				    &diagpkt_file_ops, &diagpkt_cdev,
+				    &diagpkt_device);
+		if (ret)
+			goto done;
+	}
+
+	snprintf(name, sizeof(name), "ipath_diag%d", dd->unit);
+	ret = qib_cdev_init(QIB_DIAG_MINOR_BASE + dd->unit, name,
+			    &diag_file_ops, &dd->diag_cdev,
+			    &dd->diag_device);
+done:
+	return ret;
+}
+
+static void qib_unregister_observers(struct qib_devdata *dd);
+
+void qib_diag_remove(struct qib_devdata *dd)
+{
+	struct qib_diag_client *dc;
+
+	if (atomic_dec_and_test(&diagpkt_count))
+		qib_cdev_cleanup(&diagpkt_cdev, &diagpkt_device);
+
+	qib_cdev_cleanup(&dd->diag_cdev, &dd->diag_device);
+
+	/*
+	 * Return all diag_clients of this device. There should be none,
+	 * as we are "guaranteed" that no clients are still open
+	 */
+	while (dd->diag_client)
+		return_client(dd->diag_client);
+
+	/* Now clean up all unused client structs */
+	while (client_pool) {
+		dc = client_pool;
+		client_pool = dc->next;
+		kfree(dc);
+	}
+	/* Clean up observer list */
+	qib_unregister_observers(dd);
+}
+
+/* qib_remap_ioaddr32 - remap an offset into chip address space to __iomem *
+ *
+ * @dd: the qlogic_ib device
+ * @offs: the offset in chip-space
+ * @cntp: Pointer to max (byte) count for transfer starting at offset
+ * This returns a u32 __iomem * so it can be used for both 64 and 32-bit
+ * mapping. It is needed because with the use of PAT for control of
+ * write-combining, the logically contiguous address-space of the chip
+ * may be split into virtually non-contiguous spaces, with different
+ * attributes, which are them mapped to contiguous physical space
+ * based from the first BAR.
+ *
+ * The code below makes the same assumptions as were made in
+ * init_chip_wc_pat() (qib_init.c), copied here:
+ * Assumes chip address space looks like:
+ *		- kregs + sregs + cregs + uregs (in any order)
+ *		- piobufs (2K and 4K bufs in either order)
+ *	or:
+ *		- kregs + sregs + cregs (in any order)
+ *		- piobufs (2K and 4K bufs in either order)
+ *		- uregs
+ *
+ * If cntp is non-NULL, returns how many bytes from offset can be accessed
+ * Returns 0 if the offset is not mapped.
+ */
+static u32 __iomem *qib_remap_ioaddr32(struct qib_devdata *dd, u32 offset,
+				       u32 *cntp)
+{
+	u32 kreglen;
+	u32 snd_bottom, snd_lim = 0;
+	u32 __iomem *krb32 = (u32 __iomem *)dd->kregbase;
+	u32 __iomem *map = NULL;
+	u32 cnt = 0;
+
+	/* First, simplest case, offset is within the first map. */
+	kreglen = (dd->kregend - dd->kregbase) * sizeof(u64);
+	if (offset < kreglen) {
+		map = krb32 + (offset / sizeof(u32));
+		cnt = kreglen - offset;
+		goto mapped;
+	}
+
+	/*
+	 * Next check for user regs, the next most common case,
+	 * and a cheap check because if they are not in the first map
+	 * they are last in chip.
+	 */
+	if (dd->userbase) {
+		/* If user regs mapped, they are after send, so set limit. */
+		u32 ulim = (dd->cfgctxts * dd->ureg_align) + dd->uregbase;
+		snd_lim = dd->uregbase;
+		krb32 = (u32 __iomem *)dd->userbase;
+		if (offset >= dd->uregbase && offset < ulim) {
+			map = krb32 + (offset - dd->uregbase) / sizeof(u32);
+			cnt = ulim - offset;
+			goto mapped;
+		}
+	}
+
+	/*
+	 * Lastly, check for offset within Send Buffers.
+	 * This is gnarly because struct devdata is deliberately vague
+	 * about things like 7322 VL15 buffers, and we are not in
+	 * chip-specific code here, so should not make many assumptions.
+	 * The one we _do_ make is that the only chip that has more sndbufs
+	 * than we admit is the 7322, and it has userregs above that, so
+	 * we know the snd_lim.
+	 */
+	/* Assume 2K buffers are first. */
+	snd_bottom = dd->pio2k_bufbase;
+	if (snd_lim == 0) {
+		u32 tot2k = dd->piobcnt2k * ALIGN(dd->piosize2k, dd->palign);
+		snd_lim = snd_bottom + tot2k;
+	}
+	/* If 4k buffers exist, account for them by bumping
+	 * appropriate limit.
+	 */
+	if (dd->piobcnt4k) {
+		u32 tot4k = dd->piobcnt4k * dd->align4k;
+		u32 offs4k = dd->piobufbase >> 32;
+		if (snd_bottom > offs4k)
+			snd_bottom = offs4k;
+		else {
+			/* 4k above 2k. Bump snd_lim, if needed*/
+			if (!dd->userbase)
+				snd_lim = offs4k + tot4k;
+		}
+	}
+	/*
+	 * Judgement call: can we ignore the space between SendBuffs and
+	 * UserRegs, where we would like to see vl15 buffs, but not more?
+	 */
+	if (offset >= snd_bottom && offset < snd_lim) {
+		offset -= snd_bottom;
+		map = (u32 __iomem *)dd->piobase + (offset / sizeof(u32));
+		cnt = snd_lim - offset;
+	}
+
+mapped:
+	if (cntp)
+		*cntp = cnt;
+	return map;
+}
+
+/*
+ * qib_read_umem64 - read a 64-bit quantity from the chip into user space
+ * @dd: the qlogic_ib device
+ * @uaddr: the location to store the data in user memory
+ * @regoffs: the offset from BAR0 (_NOT_ full pointer, anymore)
+ * @count: number of bytes to copy (multiple of 32 bits)
+ *
+ * This function also localizes all chip memory accesses.
+ * The copy should be written such that we read full cacheline packets
+ * from the chip.  This is usually used for a single qword
+ *
+ * NOTE:  This assumes the chip address is 64-bit aligned.
+ */
+static int qib_read_umem64(struct qib_devdata *dd, void __user *uaddr,
+			   u32 regoffs, size_t count)
+{
+	const u64 __iomem *reg_addr;
+	const u64 __iomem *reg_end;
+	u32 limit;
+	int ret;
+
+	reg_addr = (const u64 __iomem *)qib_remap_ioaddr32(dd, regoffs, &limit);
+	if (reg_addr == NULL || limit == 0 || !(dd->flags & QIB_PRESENT)) {
+		ret = -EINVAL;
+		goto bail;
+	}
+	if (count >= limit)
+		count = limit;
+	reg_end = reg_addr + (count / sizeof(u64));
+
+	/* not very efficient, but it works for now */
+	while (reg_addr < reg_end) {
+		u64 data = readq(reg_addr);
+
+		if (copy_to_user(uaddr, &data, sizeof(u64))) {
+			ret = -EFAULT;
+			goto bail;
+		}
+		reg_addr++;
+		uaddr += sizeof(u64);
+	}
+	ret = 0;
+bail:
+	return ret;
+}
+
+/*
+ * qib_write_umem64 - write a 64-bit quantity to the chip from user space
+ * @dd: the qlogic_ib device
+ * @regoffs: the offset from BAR0 (_NOT_ full pointer, anymore)
+ * @uaddr: the source of the data in user memory
+ * @count: the number of bytes to copy (multiple of 32 bits)
+ *
+ * This is usually used for a single qword
+ * NOTE:  This assumes the chip address is 64-bit aligned.
+ */
+
+static int qib_write_umem64(struct qib_devdata *dd, u32 regoffs,
+			    const void __user *uaddr, size_t count)
+{
+	u64 __iomem *reg_addr;
+	const u64 __iomem *reg_end;
+	u32 limit;
+	int ret;
+
+	reg_addr = (u64 __iomem *)qib_remap_ioaddr32(dd, regoffs, &limit);
+	if (reg_addr == NULL || limit == 0 || !(dd->flags & QIB_PRESENT)) {
+		ret = -EINVAL;
+		goto bail;
+	}
+	if (count >= limit)
+		count = limit;
+	reg_end = reg_addr + (count / sizeof(u64));
+
+	/* not very efficient, but it works for now */
+	while (reg_addr < reg_end) {
+		u64 data;
+		if (copy_from_user(&data, uaddr, sizeof(data))) {
+			ret = -EFAULT;
+			goto bail;
+		}
+		writeq(data, reg_addr);
+
+		reg_addr++;
+		uaddr += sizeof(u64);
+	}
+	ret = 0;
+bail:
+	return ret;
+}
+
+/*
+ * qib_read_umem32 - read a 32-bit quantity from the chip into user space
+ * @dd: the qlogic_ib device
+ * @uaddr: the location to store the data in user memory
+ * @regoffs: the offset from BAR0 (_NOT_ full pointer, anymore)
+ * @count: number of bytes to copy
+ *
+ * read 32 bit values, not 64 bit; for memories that only
+ * support 32 bit reads; usually a single dword.
+ */
+static int qib_read_umem32(struct qib_devdata *dd, void __user *uaddr,
+			   u32 regoffs, size_t count)
+{
+	const u32 __iomem *reg_addr;
+	const u32 __iomem *reg_end;
+	u32 limit;
+	int ret;
+
+	reg_addr = qib_remap_ioaddr32(dd, regoffs, &limit);
+	if (reg_addr == NULL || limit == 0 || !(dd->flags & QIB_PRESENT)) {
+		ret = -EINVAL;
+		goto bail;
+	}
+	if (count >= limit)
+		count = limit;
+	reg_end = reg_addr + (count / sizeof(u32));
+
+	/* not very efficient, but it works for now */
+	while (reg_addr < reg_end) {
+		u32 data = readl(reg_addr);
+
+		if (copy_to_user(uaddr, &data, sizeof(data))) {
+			ret = -EFAULT;
+			goto bail;
+		}
+
+		reg_addr++;
+		uaddr += sizeof(u32);
+
+	}
+	ret = 0;
+bail:
+	return ret;
+}
+
+/*
+ * qib_write_umem32 - write a 32-bit quantity to the chip from user space
+ * @dd: the qlogic_ib device
+ * @regoffs: the offset from BAR0 (_NOT_ full pointer, anymore)
+ * @uaddr: the source of the data in user memory
+ * @count: number of bytes to copy
+ *
+ * write 32 bit values, not 64 bit; for memories that only
+ * support 32 bit write; usually a single dword.
+ */
+
+static int qib_write_umem32(struct qib_devdata *dd, u32 regoffs,
+			    const void __user *uaddr, size_t count)
+{
+	u32 __iomem *reg_addr;
+	const u32 __iomem *reg_end;
+	u32 limit;
+	int ret;
+
+	reg_addr = qib_remap_ioaddr32(dd, regoffs, &limit);
+	if (reg_addr == NULL || limit == 0 || !(dd->flags & QIB_PRESENT)) {
+		ret = -EINVAL;
+		goto bail;
+	}
+	if (count >= limit)
+		count = limit;
+	reg_end = reg_addr + (count / sizeof(u32));
+
+	while (reg_addr < reg_end) {
+		u32 data;
+
+		if (copy_from_user(&data, uaddr, sizeof(data))) {
+			ret = -EFAULT;
+			goto bail;
+		}
+		writel(data, reg_addr);
+
+		reg_addr++;
+		uaddr += sizeof(u32);
+	}
+	ret = 0;
+bail:
+	return ret;
+}
+
+static int qib_diag_open(struct inode *in, struct file *fp)
+{
+	int unit = iminor(in) - QIB_DIAG_MINOR_BASE;
+	struct qib_devdata *dd;
+	struct qib_diag_client *dc;
+	int ret;
+
+	mutex_lock(&qib_mutex);
+
+	dd = qib_lookup(unit);
+
+	if (dd == NULL || !(dd->flags & QIB_PRESENT) ||
+	    !dd->kregbase) {
+		ret = -ENODEV;
+		goto bail;
+	}
+
+	dc = get_client(dd);
+	if (!dc) {
+		ret = -ENOMEM;
+		goto bail;
+	}
+	dc->next = dd->diag_client;
+	dd->diag_client = dc;
+	fp->private_data = dc;
+	ret = 0;
+bail:
+	mutex_unlock(&qib_mutex);
+
+	return ret;
+}
+
+/**
+ * qib_diagpkt_write - write an IB packet
+ * @fp: the diag data device file pointer
+ * @data: qib_diag_pkt structure saying where to get the packet
+ * @count: size of data to write
+ * @off: unused by this code
+ */
+static ssize_t qib_diagpkt_write(struct file *fp,
+				 const char __user *data,
+				 size_t count, loff_t *off)
+{
+	u32 __iomem *piobuf;
+	u32 plen, clen, pbufn;
+	struct qib_diag_xpkt dp;
+	u32 *tmpbuf = NULL;
+	struct qib_devdata *dd;
+	struct qib_pportdata *ppd;
+	ssize_t ret = 0;
+
+	if (count != sizeof(dp)) {
+		ret = -EINVAL;
+		goto bail;
+	}
+	if (copy_from_user(&dp, data, sizeof(dp))) {
+		ret = -EFAULT;
+		goto bail;
+	}
+
+	dd = qib_lookup(dp.unit);
+	if (!dd || !(dd->flags & QIB_PRESENT) || !dd->kregbase) {
+		ret = -ENODEV;
+		goto bail;
+	}
+	if (!(dd->flags & QIB_INITTED)) {
+		/* no hardware, freeze, etc. */
+		ret = -ENODEV;
+		goto bail;
+	}
+
+	if (dp.version != _DIAG_XPKT_VERS) {
+		qib_dev_err(dd, "Invalid version %u for diagpkt_write\n",
+			    dp.version);
+		ret = -EINVAL;
+		goto bail;
+	}
+	/* send count must be an exact number of dwords */
+	if (dp.len & 3) {
+		ret = -EINVAL;
+		goto bail;
+	}
+	if (!dp.port || dp.port > dd->num_pports) {
+		ret = -EINVAL;
+		goto bail;
+	}
+	ppd = &dd->pport[dp.port - 1];
+
+	/* need total length before first word written */
+	/* +1 word is for the qword padding */
+	plen = sizeof(u32) + dp.len;
+	clen = dp.len >> 2;
+
+	if ((plen + 4) > ppd->ibmaxlen) {
+		ret = -EINVAL;
+		goto bail;      /* before writing pbc */
+	}
+	tmpbuf = vmalloc(plen);
+	if (!tmpbuf) {
+		qib_devinfo(dd->pcidev, "Unable to allocate tmp buffer, "
+			 "failing\n");
+		ret = -ENOMEM;
+		goto bail;
+	}
+
+	if (copy_from_user(tmpbuf,
+			   (const void __user *) (unsigned long) dp.data,
+			   dp.len)) {
+		ret = -EFAULT;
+		goto bail;
+	}
+
+	plen >>= 2;             /* in dwords */
+
+	if (dp.pbc_wd == 0)
+		dp.pbc_wd = plen;
+
+	piobuf = dd->f_getsendbuf(ppd, dp.pbc_wd, &pbufn);
+	if (!piobuf) {
+		ret = -EBUSY;
+		goto bail;
+	}
+	/* disarm it just to be extra sure */
+	dd->f_sendctrl(dd->pport, QIB_SENDCTRL_DISARM_BUF(pbufn));
+
+	/* disable header check on pbufn for this packet */
+	dd->f_txchk_change(dd, pbufn, 1, TXCHK_CHG_TYPE_DIS1, NULL);
+
+	writeq(dp.pbc_wd, piobuf);
+	/*
+	 * Copy all but the trigger word, then flush, so it's written
+	 * to chip before trigger word, then write trigger word, then
+	 * flush again, so packet is sent.
+	 */
+	if (dd->flags & QIB_PIO_FLUSH_WC) {
+		qib_flush_wc();
+		qib_pio_copy(piobuf + 2, tmpbuf, clen - 1);
+		qib_flush_wc();
+		__raw_writel(tmpbuf[clen - 1], piobuf + clen + 1);
+	} else
+		qib_pio_copy(piobuf + 2, tmpbuf, clen);
+
+	if (dd->flags & QIB_USE_SPCL_TRIG) {
+		u32 spcl_off = (pbufn >= dd->piobcnt2k) ? 2047 : 1023;
+
+		qib_flush_wc();
+		__raw_writel(0xaebecede, piobuf + spcl_off);
+	}
+
+	/*
+	 * Ensure buffer is written to the chip, then re-enable
+	 * header checks (if supported by chip).  The txchk
+	 * code will ensure seen by chip before returning.
+	 */
+	qib_flush_wc();
+	qib_sendbuf_done(dd, pbufn);
+	dd->f_txchk_change(dd, pbufn, 1, TXCHK_CHG_TYPE_ENAB1, NULL);
+
+	ret = sizeof(dp);
+
+bail:
+	vfree(tmpbuf);
+	return ret;
+}
+
+static int qib_diag_release(struct inode *in, struct file *fp)
+{
+	mutex_lock(&qib_mutex);
+	return_client(fp->private_data);
+	fp->private_data = NULL;
+	mutex_unlock(&qib_mutex);
+	return 0;
+}
+
+/*
+ * Chip-specific code calls to register its interest in
+ * a specific range.
+ */
+struct diag_observer_list_elt {
+	struct diag_observer_list_elt *next;
+	const struct diag_observer *op;
+};
+
+int qib_register_observer(struct qib_devdata *dd,
+			  const struct diag_observer *op)
+{
+	struct diag_observer_list_elt *olp;
+	int ret = -EINVAL;
+
+	if (!dd || !op)
+		goto bail;
+	ret = -ENOMEM;
+	olp = vmalloc(sizeof *olp);
+	if (!olp) {
+		printk(KERN_ERR QIB_DRV_NAME ": vmalloc for observer failed\n");
+		goto bail;
+	}
+	if (olp) {
+		unsigned long flags;
+
+		spin_lock_irqsave(&dd->qib_diag_trans_lock, flags);
+		olp->op = op;
+		olp->next = dd->diag_observer_list;
+		dd->diag_observer_list = olp;
+		spin_unlock_irqrestore(&dd->qib_diag_trans_lock, flags);
+		ret = 0;
+	}
+bail:
+	return ret;
+}
+
+/* Remove all registered observers when device is closed */
+static void qib_unregister_observers(struct qib_devdata *dd)
+{
+	struct diag_observer_list_elt *olp;
+	unsigned long flags;
+
+	spin_lock_irqsave(&dd->qib_diag_trans_lock, flags);
+	olp = dd->diag_observer_list;
+	while (olp) {
+		/* Pop one observer, let go of lock */
+		dd->diag_observer_list = olp->next;
+		spin_unlock_irqrestore(&dd->qib_diag_trans_lock, flags);
+		vfree(olp);
+		/* try again. */
+		spin_lock_irqsave(&dd->qib_diag_trans_lock, flags);
+		olp = dd->diag_observer_list;
+	}
+	spin_unlock_irqrestore(&dd->qib_diag_trans_lock, flags);
+}
+
+/*
+ * Find the observer, if any, for the specified address. Initial implementation
+ * is simple stack of observers. This must be called with diag transaction
+ * lock held.
+ */
+static const struct diag_observer *diag_get_observer(struct qib_devdata *dd,
+						     u32 addr)
+{
+	struct diag_observer_list_elt *olp;
+	const struct diag_observer *op = NULL;
+
+	olp = dd->diag_observer_list;
+	while (olp) {
+		op = olp->op;
+		if (addr >= op->bottom && addr <= op->top)
+			break;
+		olp = olp->next;
+	}
+	if (!olp)
+		op = NULL;
+
+	return op;
+}
+
+static ssize_t qib_diag_read(struct file *fp, char __user *data,
+			     size_t count, loff_t *off)
+{
+	struct qib_diag_client *dc = fp->private_data;
+	struct qib_devdata *dd = dc->dd;
+	void __iomem *kreg_base;
+	ssize_t ret;
+
+	if (dc->pid != current->pid) {
+		ret = -EPERM;
+		goto bail;
+	}
+
+	kreg_base = dd->kregbase;
+
+	if (count == 0)
+		ret = 0;
+	else if ((count % 4) || (*off % 4))
+		/* address or length is not 32-bit aligned, hence invalid */
+		ret = -EINVAL;
+	else if (dc->state < READY && (*off || count != 8))
+		ret = -EINVAL;  /* prevent cat /dev/qib_diag* */
+	else {
+		unsigned long flags;
+		u64 data64 = 0;
+		int use_32;
+		const struct diag_observer *op;
+
+		use_32 = (count % 8) || (*off % 8);
+		ret = -1;
+		spin_lock_irqsave(&dd->qib_diag_trans_lock, flags);
+		/*
+		 * Check for observer on this address range.
+		 * we only support a single 32 or 64-bit read
+		 * via observer, currently.
+		 */
+		op = diag_get_observer(dd, *off);
+		if (op) {
+			u32 offset = *off;
+			ret = op->hook(dd, op, offset, &data64, 0, use_32);
+		}
+		/*
+		 * We need to release lock before any copy_to_user(),
+		 * whether implicit in qib_read_umem* or explicit below.
+		 */
+		spin_unlock_irqrestore(&dd->qib_diag_trans_lock, flags);
+		if (!op) {
+			if (use_32)
+				/*
+				 * Address or length is not 64-bit aligned;
+				 * do 32-bit rd
+				 */
+				ret = qib_read_umem32(dd, data, (u32) *off,
+						      count);
+			else
+				ret = qib_read_umem64(dd, data, (u32) *off,
+						      count);
+		} else if (ret == count) {
+			/* Below finishes case where observer existed */
+			ret = copy_to_user(data, &data64, use_32 ?
+					   sizeof(u32) : sizeof(u64));
+			if (ret)
+				ret = -EFAULT;
+		}
+	}
+
+	if (ret >= 0) {
+		*off += count;
+		ret = count;
+		if (dc->state == OPENED)
+			dc->state = INIT;
+	}
+bail:
+	return ret;
+}
+
+static ssize_t qib_diag_write(struct file *fp, const char __user *data,
+			      size_t count, loff_t *off)
+{
+	struct qib_diag_client *dc = fp->private_data;
+	struct qib_devdata *dd = dc->dd;
+	void __iomem *kreg_base;
+	ssize_t ret;
+
+	if (dc->pid != current->pid) {
+		ret = -EPERM;
+		goto bail;
+	}
+
+	kreg_base = dd->kregbase;
+
+	if (count == 0)
+		ret = 0;
+	else if ((count % 4) || (*off % 4))
+		/* address or length is not 32-bit aligned, hence invalid */
+		ret = -EINVAL;
+	else if (dc->state < READY &&
+		((*off || count != 8) || dc->state != INIT))
+		/* No writes except second-step of init seq */
+		ret = -EINVAL;  /* before any other write allowed */
+	else {
+		unsigned long flags;
+		const struct diag_observer *op = NULL;
+		int use_32 =  (count % 8) || (*off % 8);
+
+		/*
+		 * Check for observer on this address range.
+		 * We only support a single 32 or 64-bit write
+		 * via observer, currently. This helps, because
+		 * we would otherwise have to jump through hoops
+		 * to make "diag transaction" meaningful when we
+		 * cannot do a copy_from_user while holding the lock.
+		 */
+		if (count == 4 || count == 8) {
+			u64 data64;
+			u32 offset = *off;
+			ret = copy_from_user(&data64, data, count);
+			if (ret) {
+				ret = -EFAULT;
+				goto bail;
+			}
+			spin_lock_irqsave(&dd->qib_diag_trans_lock, flags);
+			op = diag_get_observer(dd, *off);
+			if (op)
+				ret = op->hook(dd, op, offset, &data64, ~0Ull,
+					       use_32);
+			spin_unlock_irqrestore(&dd->qib_diag_trans_lock, flags);
+		}
+
+		if (!op) {
+			if (use_32)
+				/*
+				 * Address or length is not 64-bit aligned;
+				 * do 32-bit write
+				 */
+				ret = qib_write_umem32(dd, (u32) *off, data,
+						       count);
+			else
+				ret = qib_write_umem64(dd, (u32) *off, data,
+						       count);
+		}
+	}
+
+	if (ret >= 0) {
+		*off += count;
+		ret = count;
+		if (dc->state == INIT)
+			dc->state = READY; /* all read/write OK now */
+	}
+bail:
+	return ret;
+}

drivers/infiniband/hw/qib/qib_dma.c

@@ -0,0 +1,182 @@
+/*
+ * Copyright (c) 2006, 2009, 2010 QLogic, Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include <linux/types.h>
+#include <linux/scatterlist.h>
+
+#include "qib_verbs.h"
+
+#define BAD_DMA_ADDRESS ((u64) 0)
+
+/*
+ * The following functions implement driver specific replacements
+ * for the ib_dma_*() functions.
+ *
+ * These functions return kernel virtual addresses instead of
+ * device bus addresses since the driver uses the CPU to copy
+ * data instead of using hardware DMA.
+ */
+
+static int qib_mapping_error(struct ib_device *dev, u64 dma_addr)
+{
+	return dma_addr == BAD_DMA_ADDRESS;
+}
+
+static u64 qib_dma_map_single(struct ib_device *dev, void *cpu_addr,
+			      size_t size, enum dma_data_direction direction)
+{
+	BUG_ON(!valid_dma_direction(direction));
+	return (u64) cpu_addr;
+}
+
+static void qib_dma_unmap_single(struct ib_device *dev, u64 addr, size_t size,
+				 enum dma_data_direction direction)
+{
+	BUG_ON(!valid_dma_direction(direction));
+}
+
+static u64 qib_dma_map_page(struct ib_device *dev, struct page *page,
+			    unsigned long offset, size_t size,
+			    enum dma_data_direction direction)
+{
+	u64 addr;
+
+	BUG_ON(!valid_dma_direction(direction));
+
+	if (offset + size > PAGE_SIZE) {
+		addr = BAD_DMA_ADDRESS;
+		goto done;
+	}
+
+	addr = (u64) page_address(page);
+	if (addr)
+		addr += offset;
+	/* TODO: handle highmem pages */
+
+done:
+	return addr;
+}
+
+static void qib_dma_unmap_page(struct ib_device *dev, u64 addr, size_t size,
+			       enum dma_data_direction direction)
+{
+	BUG_ON(!valid_dma_direction(direction));
+}
+
+static int qib_map_sg(struct ib_device *dev, struct scatterlist *sgl,
+		      int nents, enum dma_data_direction direction)
+{
+	struct scatterlist *sg;
+	u64 addr;
+	int i;
+	int ret = nents;
+
+	BUG_ON(!valid_dma_direction(direction));
+
+	for_each_sg(sgl, sg, nents, i) {
+		addr = (u64) page_address(sg_page(sg));
+		/* TODO: handle highmem pages */
+		if (!addr) {
+			ret = 0;
+			break;
+		}
+	}
+	return ret;
+}
+
+static void qib_unmap_sg(struct ib_device *dev,
+			 struct scatterlist *sg, int nents,
+			 enum dma_data_direction direction)
+{
+	BUG_ON(!valid_dma_direction(direction));
+}
+
+static u64 qib_sg_dma_address(struct ib_device *dev, struct scatterlist *sg)
+{
+	u64 addr = (u64) page_address(sg_page(sg));
+
+	if (addr)
+		addr += sg->offset;
+	return addr;
+}
+
+static unsigned int qib_sg_dma_len(struct ib_device *dev,
+				   struct scatterlist *sg)
+{
+	return sg->length;
+}
+
+static void qib_sync_single_for_cpu(struct ib_device *dev, u64 addr,
+				    size_t size, enum dma_data_direction dir)
+{
+}
+
+static void qib_sync_single_for_device(struct ib_device *dev, u64 addr,
+				       size_t size,
+				       enum dma_data_direction dir)
+{
+}
+
+static void *qib_dma_alloc_coherent(struct ib_device *dev, size_t size,
+				    u64 *dma_handle, gfp_t flag)
+{
+	struct page *p;
+	void *addr = NULL;
+
+	p = alloc_pages(flag, get_order(size));
+	if (p)
+		addr = page_address(p);
+	if (dma_handle)
+		*dma_handle = (u64) addr;
+	return addr;
+}
+
+static void qib_dma_free_coherent(struct ib_device *dev, size_t size,
+				  void *cpu_addr, u64 dma_handle)
+{
+	free_pages((unsigned long) cpu_addr, get_order(size));
+}
+
+struct ib_dma_mapping_ops qib_dma_mapping_ops = {
+	.mapping_error = qib_mapping_error,
+	.map_single = qib_dma_map_single,
+	.unmap_single = qib_dma_unmap_single,
+	.map_page = qib_dma_map_page,
+	.unmap_page = qib_dma_unmap_page,
+	.map_sg = qib_map_sg,
+	.unmap_sg = qib_unmap_sg,
+	.dma_address = qib_sg_dma_address,
+	.dma_len = qib_sg_dma_len,
+	.sync_single_for_cpu = qib_sync_single_for_cpu,
+	.sync_single_for_device = qib_sync_single_for_device,
+	.alloc_coherent = qib_dma_alloc_coherent,
+	.free_coherent = qib_dma_free_coherent
+};

drivers/infiniband/hw/qib/qib_driver.c

@@ -0,0 +1,665 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/spinlock.h>
+#include <linux/pci.h>
+#include <linux/io.h>
+#include <linux/delay.h>
+#include <linux/netdevice.h>
+#include <linux/vmalloc.h>
+
+#include "qib.h"
+
+/*
+ * The size has to be longer than this string, so we can append
+ * board/chip information to it in the init code.
+ */
+const char ib_qib_version[] = QIB_IDSTR "\n";
+
+DEFINE_SPINLOCK(qib_devs_lock);
+LIST_HEAD(qib_dev_list);
+DEFINE_MUTEX(qib_mutex);	/* general driver use */
+
+unsigned qib_ibmtu;
+module_param_named(ibmtu, qib_ibmtu, uint, S_IRUGO);
+MODULE_PARM_DESC(ibmtu, "Set max IB MTU (0=2KB, 1=256, 2=512, ... 5=4096");
+
+unsigned qib_compat_ddr_negotiate = 1;
+module_param_named(compat_ddr_negotiate, qib_compat_ddr_negotiate, uint,
+		   S_IWUSR | S_IRUGO);
+MODULE_PARM_DESC(compat_ddr_negotiate,
+		 "Attempt pre-IBTA 1.2 DDR speed negotiation");
+
+MODULE_LICENSE("Dual BSD/GPL");
+MODULE_AUTHOR("QLogic <support@qlogic.com>");
+MODULE_DESCRIPTION("QLogic IB driver");
+
+/*
+ * QIB_PIO_MAXIBHDR is the max IB header size allowed for in our
+ * PIO send buffers.  This is well beyond anything currently
+ * defined in the InfiniBand spec.
+ */
+#define QIB_PIO_MAXIBHDR 128
+
+struct qlogic_ib_stats qib_stats;
+
+const char *qib_get_unit_name(int unit)
+{
+	static char iname[16];
+
+	snprintf(iname, sizeof iname, "infinipath%u", unit);
+	return iname;
+}
+
+/*
+ * Return count of units with at least one port ACTIVE.
+ */
+int qib_count_active_units(void)
+{
+	struct qib_devdata *dd;
+	struct qib_pportdata *ppd;
+	unsigned long flags;
+	int pidx, nunits_active = 0;
+
+	spin_lock_irqsave(&qib_devs_lock, flags);
+	list_for_each_entry(dd, &qib_dev_list, list) {
+		if (!(dd->flags & QIB_PRESENT) || !dd->kregbase)
+			continue;
+		for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+			ppd = dd->pport + pidx;
+			if (ppd->lid && (ppd->lflags & (QIBL_LINKINIT |
+					 QIBL_LINKARMED | QIBL_LINKACTIVE))) {
+				nunits_active++;
+				break;
+			}
+		}
+	}
+	spin_unlock_irqrestore(&qib_devs_lock, flags);
+	return nunits_active;
+}
+
+/*
+ * Return count of all units, optionally return in arguments
+ * the number of usable (present) units, and the number of
+ * ports that are up.
+ */
+int qib_count_units(int *npresentp, int *nupp)
+{
+	int nunits = 0, npresent = 0, nup = 0;
+	struct qib_devdata *dd;
+	unsigned long flags;
+	int pidx;
+	struct qib_pportdata *ppd;
+
+	spin_lock_irqsave(&qib_devs_lock, flags);
+
+	list_for_each_entry(dd, &qib_dev_list, list) {
+		nunits++;
+		if ((dd->flags & QIB_PRESENT) && dd->kregbase)
+			npresent++;
+		for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+			ppd = dd->pport + pidx;
+			if (ppd->lid && (ppd->lflags & (QIBL_LINKINIT |
+					 QIBL_LINKARMED | QIBL_LINKACTIVE)))
+				nup++;
+		}
+	}
+
+	spin_unlock_irqrestore(&qib_devs_lock, flags);
+
+	if (npresentp)
+		*npresentp = npresent;
+	if (nupp)
+		*nupp = nup;
+
+	return nunits;
+}
+
+/**
+ * qib_wait_linkstate - wait for an IB link state change to occur
+ * @dd: the qlogic_ib device
+ * @state: the state to wait for
+ * @msecs: the number of milliseconds to wait
+ *
+ * wait up to msecs milliseconds for IB link state change to occur for
+ * now, take the easy polling route.  Currently used only by
+ * qib_set_linkstate.  Returns 0 if state reached, otherwise
+ * -ETIMEDOUT state can have multiple states set, for any of several
+ * transitions.
+ */
+int qib_wait_linkstate(struct qib_pportdata *ppd, u32 state, int msecs)
+{
+	int ret;
+	unsigned long flags;
+
+	spin_lock_irqsave(&ppd->lflags_lock, flags);
+	if (ppd->state_wanted) {
+		spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+		ret = -EBUSY;
+		goto bail;
+	}
+	ppd->state_wanted = state;
+	spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+	wait_event_interruptible_timeout(ppd->state_wait,
+					 (ppd->lflags & state),
+					 msecs_to_jiffies(msecs));
+	spin_lock_irqsave(&ppd->lflags_lock, flags);
+	ppd->state_wanted = 0;
+	spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+
+	if (!(ppd->lflags & state))
+		ret = -ETIMEDOUT;
+	else
+		ret = 0;
+bail:
+	return ret;
+}
+
+int qib_set_linkstate(struct qib_pportdata *ppd, u8 newstate)
+{
+	u32 lstate;
+	int ret;
+	struct qib_devdata *dd = ppd->dd;
+	unsigned long flags;
+
+	switch (newstate) {
+	case QIB_IB_LINKDOWN_ONLY:
+		dd->f_set_ib_cfg(ppd, QIB_IB_CFG_LSTATE,
+				 IB_LINKCMD_DOWN | IB_LINKINITCMD_NOP);
+		/* don't wait */
+		ret = 0;
+		goto bail;
+
+	case QIB_IB_LINKDOWN:
+		dd->f_set_ib_cfg(ppd, QIB_IB_CFG_LSTATE,
+				 IB_LINKCMD_DOWN | IB_LINKINITCMD_POLL);
+		/* don't wait */
+		ret = 0;
+		goto bail;
+
+	case QIB_IB_LINKDOWN_SLEEP:
+		dd->f_set_ib_cfg(ppd, QIB_IB_CFG_LSTATE,
+				 IB_LINKCMD_DOWN | IB_LINKINITCMD_SLEEP);
+		/* don't wait */
+		ret = 0;
+		goto bail;
+
+	case QIB_IB_LINKDOWN_DISABLE:
+		dd->f_set_ib_cfg(ppd, QIB_IB_CFG_LSTATE,
+				 IB_LINKCMD_DOWN | IB_LINKINITCMD_DISABLE);
+		/* don't wait */
+		ret = 0;
+		goto bail;
+
+	case QIB_IB_LINKARM:
+		if (ppd->lflags & QIBL_LINKARMED) {
+			ret = 0;
+			goto bail;
+		}
+		if (!(ppd->lflags & (QIBL_LINKINIT | QIBL_LINKACTIVE))) {
+			ret = -EINVAL;
+			goto bail;
+		}
+		/*
+		 * Since the port can be ACTIVE when we ask for ARMED,
+		 * clear QIBL_LINKV so we can wait for a transition.
+		 * If the link isn't ARMED, then something else happened
+		 * and there is no point waiting for ARMED.
+		 */
+		spin_lock_irqsave(&ppd->lflags_lock, flags);
+		ppd->lflags &= ~QIBL_LINKV;
+		spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+		dd->f_set_ib_cfg(ppd, QIB_IB_CFG_LSTATE,
+				 IB_LINKCMD_ARMED | IB_LINKINITCMD_NOP);
+		lstate = QIBL_LINKV;
+		break;
+
+	case QIB_IB_LINKACTIVE:
+		if (ppd->lflags & QIBL_LINKACTIVE) {
+			ret = 0;
+			goto bail;
+		}
+		if (!(ppd->lflags & QIBL_LINKARMED)) {
+			ret = -EINVAL;
+			goto bail;
+		}
+		dd->f_set_ib_cfg(ppd, QIB_IB_CFG_LSTATE,
+				 IB_LINKCMD_ACTIVE | IB_LINKINITCMD_NOP);
+		lstate = QIBL_LINKACTIVE;
+		break;
+
+	default:
+		ret = -EINVAL;
+		goto bail;
+	}
+	ret = qib_wait_linkstate(ppd, lstate, 10);
+
+bail:
+	return ret;
+}
+
+/*
+ * Get address of eager buffer from it's index (allocated in chunks, not
+ * contiguous).
+ */
+static inline void *qib_get_egrbuf(const struct qib_ctxtdata *rcd, u32 etail)
+{
+	const u32 chunk = etail / rcd->rcvegrbufs_perchunk;
+	const u32 idx =  etail % rcd->rcvegrbufs_perchunk;
+
+	return rcd->rcvegrbuf[chunk] + idx * rcd->dd->rcvegrbufsize;
+}
+
+/*
+ * Returns 1 if error was a CRC, else 0.
+ * Needed for some chip's synthesized error counters.
+ */
+static u32 qib_rcv_hdrerr(struct qib_pportdata *ppd, u32 ctxt,
+			  u32 eflags, u32 l, u32 etail, __le32 *rhf_addr,
+			  struct qib_message_header *hdr)
+{
+	u32 ret = 0;
+
+	if (eflags & (QLOGIC_IB_RHF_H_ICRCERR | QLOGIC_IB_RHF_H_VCRCERR))
+		ret = 1;
+	return ret;
+}
+
+/*
+ * qib_kreceive - receive a packet
+ * @rcd: the qlogic_ib context
+ * @llic: gets count of good packets needed to clear lli,
+ *          (used with chips that need need to track crcs for lli)
+ *
+ * called from interrupt handler for errors or receive interrupt
+ * Returns number of CRC error packets, needed by some chips for
+ * local link integrity tracking.   crcs are adjusted down by following
+ * good packets, if any, and count of good packets is also tracked.
+ */
+u32 qib_kreceive(struct qib_ctxtdata *rcd, u32 *llic, u32 *npkts)
+{
+	struct qib_devdata *dd = rcd->dd;
+	struct qib_pportdata *ppd = rcd->ppd;
+	__le32 *rhf_addr;
+	void *ebuf;
+	const u32 rsize = dd->rcvhdrentsize;        /* words */
+	const u32 maxcnt = dd->rcvhdrcnt * rsize;   /* words */
+	u32 etail = -1, l, hdrqtail;
+	struct qib_message_header *hdr;
+	u32 eflags, etype, tlen, i = 0, updegr = 0, crcs = 0;
+	int last;
+	u64 lval;
+	struct qib_qp *qp, *nqp;
+
+	l = rcd->head;
+	rhf_addr = (__le32 *) rcd->rcvhdrq + l + dd->rhf_offset;
+	if (dd->flags & QIB_NODMA_RTAIL) {
+		u32 seq = qib_hdrget_seq(rhf_addr);
+		if (seq != rcd->seq_cnt)
+			goto bail;
+		hdrqtail = 0;
+	} else {
+		hdrqtail = qib_get_rcvhdrtail(rcd);
+		if (l == hdrqtail)
+			goto bail;
+		smp_rmb();  /* prevent speculative reads of dma'ed hdrq */
+	}
+
+	for (last = 0, i = 1; !last; i += !last) {
+		hdr = dd->f_get_msgheader(dd, rhf_addr);
+		eflags = qib_hdrget_err_flags(rhf_addr);
+		etype = qib_hdrget_rcv_type(rhf_addr);
+		/* total length */
+		tlen = qib_hdrget_length_in_bytes(rhf_addr);
+		ebuf = NULL;
+		if ((dd->flags & QIB_NODMA_RTAIL) ?
+		    qib_hdrget_use_egr_buf(rhf_addr) :
+		    (etype != RCVHQ_RCV_TYPE_EXPECTED)) {
+			etail = qib_hdrget_index(rhf_addr);
+			updegr = 1;
+			if (tlen > sizeof(*hdr) ||
+			    etype >= RCVHQ_RCV_TYPE_NON_KD)
+				ebuf = qib_get_egrbuf(rcd, etail);
+		}
+		if (!eflags) {
+			u16 lrh_len = be16_to_cpu(hdr->lrh[2]) << 2;
+
+			if (lrh_len != tlen) {
+				qib_stats.sps_lenerrs++;
+				goto move_along;
+			}
+		}
+		if (etype == RCVHQ_RCV_TYPE_NON_KD && !eflags &&
+		    ebuf == NULL &&
+		    tlen > (dd->rcvhdrentsize - 2 + 1 -
+				qib_hdrget_offset(rhf_addr)) << 2) {
+			goto move_along;
+		}
+
+		/*
+		 * Both tiderr and qibhdrerr are set for all plain IB
+		 * packets; only qibhdrerr should be set.
+		 */
+		if (unlikely(eflags))
+			crcs += qib_rcv_hdrerr(ppd, rcd->ctxt, eflags, l,
+					       etail, rhf_addr, hdr);
+		else if (etype == RCVHQ_RCV_TYPE_NON_KD) {
+			qib_ib_rcv(rcd, hdr, ebuf, tlen);
+			if (crcs)
+				crcs--;
+			else if (llic && *llic)
+				--*llic;
+		}
+move_along:
+		l += rsize;
+		if (l >= maxcnt)
+			l = 0;
+		rhf_addr = (__le32 *) rcd->rcvhdrq + l + dd->rhf_offset;
+		if (dd->flags & QIB_NODMA_RTAIL) {
+			u32 seq = qib_hdrget_seq(rhf_addr);
+
+			if (++rcd->seq_cnt > 13)
+				rcd->seq_cnt = 1;
+			if (seq != rcd->seq_cnt)
+				last = 1;
+		} else if (l == hdrqtail)
+			last = 1;
+		/*
+		 * Update head regs etc., every 16 packets, if not last pkt,
+		 * to help prevent rcvhdrq overflows, when many packets
+		 * are processed and queue is nearly full.
+		 * Don't request an interrupt for intermediate updates.
+		 */
+		lval = l;
+		if (!last && !(i & 0xf)) {
+			dd->f_update_usrhead(rcd, lval, updegr, etail);
+			updegr = 0;
+		}
+	}
+
+	rcd->head = l;
+	rcd->pkt_count += i;
+
+	/*
+	 * Iterate over all QPs waiting to respond.
+	 * The list won't change since the IRQ is only run on one CPU.
+	 */
+	list_for_each_entry_safe(qp, nqp, &rcd->qp_wait_list, rspwait) {
+		list_del_init(&qp->rspwait);
+		if (qp->r_flags & QIB_R_RSP_NAK) {
+			qp->r_flags &= ~QIB_R_RSP_NAK;
+			qib_send_rc_ack(qp);
+		}
+		if (qp->r_flags & QIB_R_RSP_SEND) {
+			unsigned long flags;
+
+			qp->r_flags &= ~QIB_R_RSP_SEND;
+			spin_lock_irqsave(&qp->s_lock, flags);
+			if (ib_qib_state_ops[qp->state] &
+					QIB_PROCESS_OR_FLUSH_SEND)
+				qib_schedule_send(qp);
+			spin_unlock_irqrestore(&qp->s_lock, flags);
+		}
+		if (atomic_dec_and_test(&qp->refcount))
+			wake_up(&qp->wait);
+	}
+
+bail:
+	/* Report number of packets consumed */
+	if (npkts)
+		*npkts = i;
+
+	/*
+	 * Always write head at end, and setup rcv interrupt, even
+	 * if no packets were processed.
+	 */
+	lval = (u64)rcd->head | dd->rhdrhead_intr_off;
+	dd->f_update_usrhead(rcd, lval, updegr, etail);
+	return crcs;
+}
+
+/**
+ * qib_set_mtu - set the MTU
+ * @ppd: the perport data
+ * @arg: the new MTU
+ *
+ * We can handle "any" incoming size, the issue here is whether we
+ * need to restrict our outgoing size.   For now, we don't do any
+ * sanity checking on this, and we don't deal with what happens to
+ * programs that are already running when the size changes.
+ * NOTE: changing the MTU will usually cause the IBC to go back to
+ * link INIT state...
+ */
+int qib_set_mtu(struct qib_pportdata *ppd, u16 arg)
+{
+	u32 piosize;
+	int ret, chk;
+
+	if (arg != 256 && arg != 512 && arg != 1024 && arg != 2048 &&
+	    arg != 4096) {
+		ret = -EINVAL;
+		goto bail;
+	}
+	chk = ib_mtu_enum_to_int(qib_ibmtu);
+	if (chk > 0 && arg > chk) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	piosize = ppd->ibmaxlen;
+	ppd->ibmtu = arg;
+
+	if (arg >= (piosize - QIB_PIO_MAXIBHDR)) {
+		/* Only if it's not the initial value (or reset to it) */
+		if (piosize != ppd->init_ibmaxlen) {
+			if (arg > piosize && arg <= ppd->init_ibmaxlen)
+				piosize = ppd->init_ibmaxlen - 2 * sizeof(u32);
+			ppd->ibmaxlen = piosize;
+		}
+	} else if ((arg + QIB_PIO_MAXIBHDR) != ppd->ibmaxlen) {
+		piosize = arg + QIB_PIO_MAXIBHDR - 2 * sizeof(u32);
+		ppd->ibmaxlen = piosize;
+	}
+
+	ppd->dd->f_set_ib_cfg(ppd, QIB_IB_CFG_MTU, 0);
+
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+int qib_set_lid(struct qib_pportdata *ppd, u32 lid, u8 lmc)
+{
+	struct qib_devdata *dd = ppd->dd;
+	ppd->lid = lid;
+	ppd->lmc = lmc;
+
+	dd->f_set_ib_cfg(ppd, QIB_IB_CFG_LIDLMC,
+			 lid | (~((1U << lmc) - 1)) << 16);
+
+	qib_devinfo(dd->pcidev, "IB%u:%u got a lid: 0x%x\n",
+		    dd->unit, ppd->port, lid);
+
+	return 0;
+}
+
+/*
+ * Following deal with the "obviously simple" task of overriding the state
+ * of the LEDS, which normally indicate link physical and logical status.
+ * The complications arise in dealing with different hardware mappings
+ * and the board-dependent routine being called from interrupts.
+ * and then there's the requirement to _flash_ them.
+ */
+#define LED_OVER_FREQ_SHIFT 8
+#define LED_OVER_FREQ_MASK (0xFF<<LED_OVER_FREQ_SHIFT)
+/* Below is "non-zero" to force override, but both actual LEDs are off */
+#define LED_OVER_BOTH_OFF (8)
+
+static void qib_run_led_override(unsigned long opaque)
+{
+	struct qib_pportdata *ppd = (struct qib_pportdata *)opaque;
+	struct qib_devdata *dd = ppd->dd;
+	int timeoff;
+	int ph_idx;
+
+	if (!(dd->flags & QIB_INITTED))
+		return;
+
+	ph_idx = ppd->led_override_phase++ & 1;
+	ppd->led_override = ppd->led_override_vals[ph_idx];
+	timeoff = ppd->led_override_timeoff;
+
+	dd->f_setextled(ppd, 1);
+	/*
+	 * don't re-fire the timer if user asked for it to be off; we let
+	 * it fire one more time after they turn it off to simplify
+	 */
+	if (ppd->led_override_vals[0] || ppd->led_override_vals[1])
+		mod_timer(&ppd->led_override_timer, jiffies + timeoff);
+}
+
+void qib_set_led_override(struct qib_pportdata *ppd, unsigned int val)
+{
+	struct qib_devdata *dd = ppd->dd;
+	int timeoff, freq;
+
+	if (!(dd->flags & QIB_INITTED))
+		return;
+
+	/* First check if we are blinking. If not, use 1HZ polling */
+	timeoff = HZ;
+	freq = (val & LED_OVER_FREQ_MASK) >> LED_OVER_FREQ_SHIFT;
+
+	if (freq) {
+		/* For blink, set each phase from one nybble of val */
+		ppd->led_override_vals[0] = val & 0xF;
+		ppd->led_override_vals[1] = (val >> 4) & 0xF;
+		timeoff = (HZ << 4)/freq;
+	} else {
+		/* Non-blink set both phases the same. */
+		ppd->led_override_vals[0] = val & 0xF;
+		ppd->led_override_vals[1] = val & 0xF;
+	}
+	ppd->led_override_timeoff = timeoff;
+
+	/*
+	 * If the timer has not already been started, do so. Use a "quick"
+	 * timeout so the function will be called soon, to look at our request.
+	 */
+	if (atomic_inc_return(&ppd->led_override_timer_active) == 1) {
+		/* Need to start timer */
+		init_timer(&ppd->led_override_timer);
+		ppd->led_override_timer.function = qib_run_led_override;
+		ppd->led_override_timer.data = (unsigned long) ppd;
+		ppd->led_override_timer.expires = jiffies + 1;
+		add_timer(&ppd->led_override_timer);
+	} else {
+		if (ppd->led_override_vals[0] || ppd->led_override_vals[1])
+			mod_timer(&ppd->led_override_timer, jiffies + 1);
+		atomic_dec(&ppd->led_override_timer_active);
+	}
+}
+
+/**
+ * qib_reset_device - reset the chip if possible
+ * @unit: the device to reset
+ *
+ * Whether or not reset is successful, we attempt to re-initialize the chip
+ * (that is, much like a driver unload/reload).  We clear the INITTED flag
+ * so that the various entry points will fail until we reinitialize.  For
+ * now, we only allow this if no user contexts are open that use chip resources
+ */
+int qib_reset_device(int unit)
+{
+	int ret, i;
+	struct qib_devdata *dd = qib_lookup(unit);
+	struct qib_pportdata *ppd;
+	unsigned long flags;
+	int pidx;
+
+	if (!dd) {
+		ret = -ENODEV;
+		goto bail;
+	}
+
+	qib_devinfo(dd->pcidev, "Reset on unit %u requested\n", unit);
+
+	if (!dd->kregbase || !(dd->flags & QIB_PRESENT)) {
+		qib_devinfo(dd->pcidev, "Invalid unit number %u or "
+			    "not initialized or not present\n", unit);
+		ret = -ENXIO;
+		goto bail;
+	}
+
+	spin_lock_irqsave(&dd->uctxt_lock, flags);
+	if (dd->rcd)
+		for (i = dd->first_user_ctxt; i < dd->cfgctxts; i++) {
+			if (!dd->rcd[i] || !dd->rcd[i]->cnt)
+				continue;
+			spin_unlock_irqrestore(&dd->uctxt_lock, flags);
+			ret = -EBUSY;
+			goto bail;
+		}
+	spin_unlock_irqrestore(&dd->uctxt_lock, flags);
+
+	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+		ppd = dd->pport + pidx;
+		if (atomic_read(&ppd->led_override_timer_active)) {
+			/* Need to stop LED timer, _then_ shut off LEDs */
+			del_timer_sync(&ppd->led_override_timer);
+			atomic_set(&ppd->led_override_timer_active, 0);
+		}
+
+		/* Shut off LEDs after we are sure timer is not running */
+		ppd->led_override = LED_OVER_BOTH_OFF;
+		dd->f_setextled(ppd, 0);
+		if (dd->flags & QIB_HAS_SEND_DMA)
+			qib_teardown_sdma(ppd);
+	}
+
+	ret = dd->f_reset(dd);
+	if (ret == 1)
+		ret = qib_init(dd, 1);
+	else
+		ret = -EAGAIN;
+	if (ret)
+		qib_dev_err(dd, "Reinitialize unit %u after "
+			    "reset failed with %d\n", unit, ret);
+	else
+		qib_devinfo(dd->pcidev, "Reinitialized unit %u after "
+			    "resetting\n", unit);
+
+bail:
+	return ret;
+}

drivers/infiniband/hw/qib/qib_eeprom.c

@@ -0,0 +1,451 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/delay.h>
+#include <linux/pci.h>
+#include <linux/vmalloc.h>
+
+#include "qib.h"
+
+/*
+ * Functions specific to the serial EEPROM on cards handled by ib_qib.
+ * The actual serail interface code is in qib_twsi.c. This file is a client
+ */
+
+/**
+ * qib_eeprom_read - receives bytes from the eeprom via I2C
+ * @dd: the qlogic_ib device
+ * @eeprom_offset: address to read from
+ * @buffer: where to store result
+ * @len: number of bytes to receive
+ */
+int qib_eeprom_read(struct qib_devdata *dd, u8 eeprom_offset,
+		    void *buff, int len)
+{
+	int ret;
+
+	ret = mutex_lock_interruptible(&dd->eep_lock);
+	if (!ret) {
+		ret = qib_twsi_reset(dd);
+		if (ret)
+			qib_dev_err(dd, "EEPROM Reset for read failed\n");
+		else
+			ret = qib_twsi_blk_rd(dd, dd->twsi_eeprom_dev,
+					      eeprom_offset, buff, len);
+		mutex_unlock(&dd->eep_lock);
+	}
+
+	return ret;
+}
+
+/*
+ * Actually update the eeprom, first doing write enable if
+ * needed, then restoring write enable state.
+ * Must be called with eep_lock held
+ */
+static int eeprom_write_with_enable(struct qib_devdata *dd, u8 offset,
+		     const void *buf, int len)
+{
+	int ret, pwen;
+
+	pwen = dd->f_eeprom_wen(dd, 1);
+	ret = qib_twsi_reset(dd);
+	if (ret)
+		qib_dev_err(dd, "EEPROM Reset for write failed\n");
+	else
+		ret = qib_twsi_blk_wr(dd, dd->twsi_eeprom_dev,
+				      offset, buf, len);
+	dd->f_eeprom_wen(dd, pwen);
+	return ret;
+}
+
+/**
+ * qib_eeprom_write - writes data to the eeprom via I2C
+ * @dd: the qlogic_ib device
+ * @eeprom_offset: where to place data
+ * @buffer: data to write
+ * @len: number of bytes to write
+ */
+int qib_eeprom_write(struct qib_devdata *dd, u8 eeprom_offset,
+		     const void *buff, int len)
+{
+	int ret;
+
+	ret = mutex_lock_interruptible(&dd->eep_lock);
+	if (!ret) {
+		ret = eeprom_write_with_enable(dd, eeprom_offset, buff, len);
+		mutex_unlock(&dd->eep_lock);
+	}
+
+	return ret;
+}
+
+static u8 flash_csum(struct qib_flash *ifp, int adjust)
+{
+	u8 *ip = (u8 *) ifp;
+	u8 csum = 0, len;
+
+	/*
+	 * Limit length checksummed to max length of actual data.
+	 * Checksum of erased eeprom will still be bad, but we avoid
+	 * reading past the end of the buffer we were passed.
+	 */
+	len = ifp->if_length;
+	if (len > sizeof(struct qib_flash))
+		len = sizeof(struct qib_flash);
+	while (len--)
+		csum += *ip++;
+	csum -= ifp->if_csum;
+	csum = ~csum;
+	if (adjust)
+		ifp->if_csum = csum;
+
+	return csum;
+}
+
+/**
+ * qib_get_eeprom_info- get the GUID et al. from the TSWI EEPROM device
+ * @dd: the qlogic_ib device
+ *
+ * We have the capability to use the nguid field, and get
+ * the guid from the first chip's flash, to use for all of them.
+ */
+void qib_get_eeprom_info(struct qib_devdata *dd)
+{
+	void *buf;
+	struct qib_flash *ifp;
+	__be64 guid;
+	int len, eep_stat;
+	u8 csum, *bguid;
+	int t = dd->unit;
+	struct qib_devdata *dd0 = qib_lookup(0);
+
+	if (t && dd0->nguid > 1 && t <= dd0->nguid) {
+		u8 oguid;
+		dd->base_guid = dd0->base_guid;
+		bguid = (u8 *) &dd->base_guid;
+
+		oguid = bguid[7];
+		bguid[7] += t;
+		if (oguid > bguid[7]) {
+			if (bguid[6] == 0xff) {
+				if (bguid[5] == 0xff) {
+					qib_dev_err(dd, "Can't set %s GUID"
+						    " from base, wraps to"
+						    " OUI!\n",
+						    qib_get_unit_name(t));
+					dd->base_guid = 0;
+					goto bail;
+				}
+				bguid[5]++;
+			}
+			bguid[6]++;
+		}
+		dd->nguid = 1;
+		goto bail;
+	}
+
+	/*
+	 * Read full flash, not just currently used part, since it may have
+	 * been written with a newer definition.
+	 * */
+	len = sizeof(struct qib_flash);
+	buf = vmalloc(len);
+	if (!buf) {
+		qib_dev_err(dd, "Couldn't allocate memory to read %u "
+			    "bytes from eeprom for GUID\n", len);
+		goto bail;
+	}
+
+	/*
+	 * Use "public" eeprom read function, which does locking and
+	 * figures out device. This will migrate to chip-specific.
+	 */
+	eep_stat = qib_eeprom_read(dd, 0, buf, len);
+
+	if (eep_stat) {
+		qib_dev_err(dd, "Failed reading GUID from eeprom\n");
+		goto done;
+	}
+	ifp = (struct qib_flash *)buf;
+
+	csum = flash_csum(ifp, 0);
+	if (csum != ifp->if_csum) {
+		qib_devinfo(dd->pcidev, "Bad I2C flash checksum: "
+			 "0x%x, not 0x%x\n", csum, ifp->if_csum);
+		goto done;
+	}
+	if (*(__be64 *) ifp->if_guid == cpu_to_be64(0) ||
+	    *(__be64 *) ifp->if_guid == ~cpu_to_be64(0)) {
+		qib_dev_err(dd, "Invalid GUID %llx from flash; ignoring\n",
+			    *(unsigned long long *) ifp->if_guid);
+		/* don't allow GUID if all 0 or all 1's */
+		goto done;
+	}
+
+	/* complain, but allow it */
+	if (*(u64 *) ifp->if_guid == 0x100007511000000ULL)
+		qib_devinfo(dd->pcidev, "Warning, GUID %llx is "
+			 "default, probably not correct!\n",
+			 *(unsigned long long *) ifp->if_guid);
+
+	bguid = ifp->if_guid;
+	if (!bguid[0] && !bguid[1] && !bguid[2]) {
+		/*
+		 * Original incorrect GUID format in flash; fix in
+		 * core copy, by shifting up 2 octets; don't need to
+		 * change top octet, since both it and shifted are 0.
+		 */
+		bguid[1] = bguid[3];
+		bguid[2] = bguid[4];
+		bguid[3] = 0;
+		bguid[4] = 0;
+		guid = *(__be64 *) ifp->if_guid;
+	} else
+		guid = *(__be64 *) ifp->if_guid;
+	dd->base_guid = guid;
+	dd->nguid = ifp->if_numguid;
+	/*
+	 * Things are slightly complicated by the desire to transparently
+	 * support both the Pathscale 10-digit serial number and the QLogic
+	 * 13-character version.
+	 */
+	if ((ifp->if_fversion > 1) && ifp->if_sprefix[0] &&
+	    ((u8 *) ifp->if_sprefix)[0] != 0xFF) {
+		char *snp = dd->serial;
+
+		/*
+		 * This board has a Serial-prefix, which is stored
+		 * elsewhere for backward-compatibility.
+		 */
+		memcpy(snp, ifp->if_sprefix, sizeof ifp->if_sprefix);
+		snp[sizeof ifp->if_sprefix] = '\0';
+		len = strlen(snp);
+		snp += len;
+		len = (sizeof dd->serial) - len;
+		if (len > sizeof ifp->if_serial)
+			len = sizeof ifp->if_serial;
+		memcpy(snp, ifp->if_serial, len);
+	} else
+		memcpy(dd->serial, ifp->if_serial,
+		       sizeof ifp->if_serial);
+	if (!strstr(ifp->if_comment, "Tested successfully"))
+		qib_dev_err(dd, "Board SN %s did not pass functional "
+			    "test: %s\n", dd->serial, ifp->if_comment);
+
+	memcpy(&dd->eep_st_errs, &ifp->if_errcntp, QIB_EEP_LOG_CNT);
+	/*
+	 * Power-on (actually "active") hours are kept as little-endian value
+	 * in EEPROM, but as seconds in a (possibly as small as 24-bit)
+	 * atomic_t while running.
+	 */
+	atomic_set(&dd->active_time, 0);
+	dd->eep_hrs = ifp->if_powerhour[0] | (ifp->if_powerhour[1] << 8);
+
+done:
+	vfree(buf);
+
+bail:;
+}
+
+/**
+ * qib_update_eeprom_log - copy active-time and error counters to eeprom
+ * @dd: the qlogic_ib device
+ *
+ * Although the time is kept as seconds in the qib_devdata struct, it is
+ * rounded to hours for re-write, as we have only 16 bits in EEPROM.
+ * First-cut code reads whole (expected) struct qib_flash, modifies,
+ * re-writes. Future direction: read/write only what we need, assuming
+ * that the EEPROM had to have been "good enough" for driver init, and
+ * if not, we aren't making it worse.
+ *
+ */
+int qib_update_eeprom_log(struct qib_devdata *dd)
+{
+	void *buf;
+	struct qib_flash *ifp;
+	int len, hi_water;
+	uint32_t new_time, new_hrs;
+	u8 csum;
+	int ret, idx;
+	unsigned long flags;
+
+	/* first, check if we actually need to do anything. */
+	ret = 0;
+	for (idx = 0; idx < QIB_EEP_LOG_CNT; ++idx) {
+		if (dd->eep_st_new_errs[idx]) {
+			ret = 1;
+			break;
+		}
+	}
+	new_time = atomic_read(&dd->active_time);
+
+	if (ret == 0 && new_time < 3600)
+		goto bail;
+
+	/*
+	 * The quick-check above determined that there is something worthy
+	 * of logging, so get current contents and do a more detailed idea.
+	 * read full flash, not just currently used part, since it may have
+	 * been written with a newer definition
+	 */
+	len = sizeof(struct qib_flash);
+	buf = vmalloc(len);
+	ret = 1;
+	if (!buf) {
+		qib_dev_err(dd, "Couldn't allocate memory to read %u "
+			    "bytes from eeprom for logging\n", len);
+		goto bail;
+	}
+
+	/* Grab semaphore and read current EEPROM. If we get an
+	 * error, let go, but if not, keep it until we finish write.
+	 */
+	ret = mutex_lock_interruptible(&dd->eep_lock);
+	if (ret) {
+		qib_dev_err(dd, "Unable to acquire EEPROM for logging\n");
+		goto free_bail;
+	}
+	ret = qib_twsi_blk_rd(dd, dd->twsi_eeprom_dev, 0, buf, len);
+	if (ret) {
+		mutex_unlock(&dd->eep_lock);
+		qib_dev_err(dd, "Unable read EEPROM for logging\n");
+		goto free_bail;
+	}
+	ifp = (struct qib_flash *)buf;
+
+	csum = flash_csum(ifp, 0);
+	if (csum != ifp->if_csum) {
+		mutex_unlock(&dd->eep_lock);
+		qib_dev_err(dd, "EEPROM cks err (0x%02X, S/B 0x%02X)\n",
+			    csum, ifp->if_csum);
+		ret = 1;
+		goto free_bail;
+	}
+	hi_water = 0;
+	spin_lock_irqsave(&dd->eep_st_lock, flags);
+	for (idx = 0; idx < QIB_EEP_LOG_CNT; ++idx) {
+		int new_val = dd->eep_st_new_errs[idx];
+		if (new_val) {
+			/*
+			 * If we have seen any errors, add to EEPROM values
+			 * We need to saturate at 0xFF (255) and we also
+			 * would need to adjust the checksum if we were
+			 * trying to minimize EEPROM traffic
+			 * Note that we add to actual current count in EEPROM,
+			 * in case it was altered while we were running.
+			 */
+			new_val += ifp->if_errcntp[idx];
+			if (new_val > 0xFF)
+				new_val = 0xFF;
+			if (ifp->if_errcntp[idx] != new_val) {
+				ifp->if_errcntp[idx] = new_val;
+				hi_water = offsetof(struct qib_flash,
+						    if_errcntp) + idx;
+			}
+			/*
+			 * update our shadow (used to minimize EEPROM
+			 * traffic), to match what we are about to write.
+			 */
+			dd->eep_st_errs[idx] = new_val;
+			dd->eep_st_new_errs[idx] = 0;
+		}
+	}
+	/*
+	 * Now update active-time. We would like to round to the nearest hour
+	 * but unless atomic_t are sure to be proper signed ints we cannot,
+	 * because we need to account for what we "transfer" to EEPROM and
+	 * if we log an hour at 31 minutes, then we would need to set
+	 * active_time to -29 to accurately count the _next_ hour.
+	 */
+	if (new_time >= 3600) {
+		new_hrs = new_time / 3600;
+		atomic_sub((new_hrs * 3600), &dd->active_time);
+		new_hrs += dd->eep_hrs;
+		if (new_hrs > 0xFFFF)
+			new_hrs = 0xFFFF;
+		dd->eep_hrs = new_hrs;
+		if ((new_hrs & 0xFF) != ifp->if_powerhour[0]) {
+			ifp->if_powerhour[0] = new_hrs & 0xFF;
+			hi_water = offsetof(struct qib_flash, if_powerhour);
+		}
+		if ((new_hrs >> 8) != ifp->if_powerhour[1]) {
+			ifp->if_powerhour[1] = new_hrs >> 8;
+			hi_water = offsetof(struct qib_flash, if_powerhour) + 1;
+		}
+	}
+	/*
+	 * There is a tiny possibility that we could somehow fail to write
+	 * the EEPROM after updating our shadows, but problems from holding
+	 * the spinlock too long are a much bigger issue.
+	 */
+	spin_unlock_irqrestore(&dd->eep_st_lock, flags);
+	if (hi_water) {
+		/* we made some change to the data, uopdate cksum and write */
+		csum = flash_csum(ifp, 1);
+		ret = eeprom_write_with_enable(dd, 0, buf, hi_water + 1);
+	}
+	mutex_unlock(&dd->eep_lock);
+	if (ret)
+		qib_dev_err(dd, "Failed updating EEPROM\n");
+
+free_bail:
+	vfree(buf);
+bail:
+	return ret;
+}
+
+/**
+ * qib_inc_eeprom_err - increment one of the four error counters
+ * that are logged to EEPROM.
+ * @dd: the qlogic_ib device
+ * @eidx: 0..3, the counter to increment
+ * @incr: how much to add
+ *
+ * Each counter is 8-bits, and saturates at 255 (0xFF). They
+ * are copied to the EEPROM (aka flash) whenever qib_update_eeprom_log()
+ * is called, but it can only be called in a context that allows sleep.
+ * This function can be called even at interrupt level.
+ */
+void qib_inc_eeprom_err(struct qib_devdata *dd, u32 eidx, u32 incr)
+{
+	uint new_val;
+	unsigned long flags;
+
+	spin_lock_irqsave(&dd->eep_st_lock, flags);
+	new_val = dd->eep_st_new_errs[eidx] + incr;
+	if (new_val > 255)
+		new_val = 255;
+	dd->eep_st_new_errs[eidx] = new_val;
+	spin_unlock_irqrestore(&dd->eep_st_lock, flags);
+}

drivers/infiniband/hw/qib/qib_fs.c

@@ -0,0 +1,613 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/module.h>
+#include <linux/fs.h>
+#include <linux/mount.h>
+#include <linux/pagemap.h>
+#include <linux/init.h>
+#include <linux/namei.h>
+
+#include "qib.h"
+
+#define QIBFS_MAGIC 0x726a77
+
+static struct super_block *qib_super;
+
+#define private2dd(file) ((file)->f_dentry->d_inode->i_private)
+
+static int qibfs_mknod(struct inode *dir, struct dentry *dentry,
+		       int mode, const struct file_operations *fops,
+		       void *data)
+{
+	int error;
+	struct inode *inode = new_inode(dir->i_sb);
+
+	if (!inode) {
+		error = -EPERM;
+		goto bail;
+	}
+
+	inode->i_mode = mode;
+	inode->i_uid = 0;
+	inode->i_gid = 0;
+	inode->i_blocks = 0;
+	inode->i_atime = CURRENT_TIME;
+	inode->i_mtime = inode->i_atime;
+	inode->i_ctime = inode->i_atime;
+	inode->i_private = data;
+	if ((mode & S_IFMT) == S_IFDIR) {
+		inode->i_op = &simple_dir_inode_operations;
+		inc_nlink(inode);
+		inc_nlink(dir);
+	}
+
+	inode->i_fop = fops;
+
+	d_instantiate(dentry, inode);
+	error = 0;
+
+bail:
+	return error;
+}
+
+static int create_file(const char *name, mode_t mode,
+		       struct dentry *parent, struct dentry **dentry,
+		       const struct file_operations *fops, void *data)
+{
+	int error;
+
+	*dentry = NULL;
+	mutex_lock(&parent->d_inode->i_mutex);
+	*dentry = lookup_one_len(name, parent, strlen(name));
+	if (!IS_ERR(*dentry))
+		error = qibfs_mknod(parent->d_inode, *dentry,
+				    mode, fops, data);
+	else
+		error = PTR_ERR(*dentry);
+	mutex_unlock(&parent->d_inode->i_mutex);
+
+	return error;
+}
+
+static ssize_t driver_stats_read(struct file *file, char __user *buf,
+				 size_t count, loff_t *ppos)
+{
+	return simple_read_from_buffer(buf, count, ppos, &qib_stats,
+				       sizeof qib_stats);
+}
+
+/*
+ * driver stats field names, one line per stat, single string.  Used by
+ * programs like ipathstats to print the stats in a way which works for
+ * different versions of drivers, without changing program source.
+ * if qlogic_ib_stats changes, this needs to change.  Names need to be
+ * 12 chars or less (w/o newline), for proper display by ipathstats utility.
+ */
+static const char qib_statnames[] =
+	"KernIntr\n"
+	"ErrorIntr\n"
+	"Tx_Errs\n"
+	"Rcv_Errs\n"
+	"H/W_Errs\n"
+	"NoPIOBufs\n"
+	"CtxtsOpen\n"
+	"RcvLen_Errs\n"
+	"EgrBufFull\n"
+	"EgrHdrFull\n"
+	;
+
+static ssize_t driver_names_read(struct file *file, char __user *buf,
+				 size_t count, loff_t *ppos)
+{
+	return simple_read_from_buffer(buf, count, ppos, qib_statnames,
+		sizeof qib_statnames - 1); /* no null */
+}
+
+static const struct file_operations driver_ops[] = {
+	{ .read = driver_stats_read, },
+	{ .read = driver_names_read, },
+};
+
+/* read the per-device counters */
+static ssize_t dev_counters_read(struct file *file, char __user *buf,
+				 size_t count, loff_t *ppos)
+{
+	u64 *counters;
+	struct qib_devdata *dd = private2dd(file);
+
+	return simple_read_from_buffer(buf, count, ppos, counters,
+		dd->f_read_cntrs(dd, *ppos, NULL, &counters));
+}
+
+/* read the per-device counters */
+static ssize_t dev_names_read(struct file *file, char __user *buf,
+			      size_t count, loff_t *ppos)
+{
+	char *names;
+	struct qib_devdata *dd = private2dd(file);
+
+	return simple_read_from_buffer(buf, count, ppos, names,
+		dd->f_read_cntrs(dd, *ppos, &names, NULL));
+}
+
+static const struct file_operations cntr_ops[] = {
+	{ .read = dev_counters_read, },
+	{ .read = dev_names_read, },
+};
+
+/*
+ * Could use file->f_dentry->d_inode->i_ino to figure out which file,
+ * instead of separate routine for each, but for now, this works...
+ */
+
+/* read the per-port names (same for each port) */
+static ssize_t portnames_read(struct file *file, char __user *buf,
+			      size_t count, loff_t *ppos)
+{
+	char *names;
+	struct qib_devdata *dd = private2dd(file);
+
+	return simple_read_from_buffer(buf, count, ppos, names,
+		dd->f_read_portcntrs(dd, *ppos, 0, &names, NULL));
+}
+
+/* read the per-port counters for port 1 (pidx 0) */
+static ssize_t portcntrs_1_read(struct file *file, char __user *buf,
+				size_t count, loff_t *ppos)
+{
+	u64 *counters;
+	struct qib_devdata *dd = private2dd(file);
+
+	return simple_read_from_buffer(buf, count, ppos, counters,
+		dd->f_read_portcntrs(dd, *ppos, 0, NULL, &counters));
+}
+
+/* read the per-port counters for port 2 (pidx 1) */
+static ssize_t portcntrs_2_read(struct file *file, char __user *buf,
+				size_t count, loff_t *ppos)
+{
+	u64 *counters;
+	struct qib_devdata *dd = private2dd(file);
+
+	return simple_read_from_buffer(buf, count, ppos, counters,
+		dd->f_read_portcntrs(dd, *ppos, 1, NULL, &counters));
+}
+
+static const struct file_operations portcntr_ops[] = {
+	{ .read = portnames_read, },
+	{ .read = portcntrs_1_read, },
+	{ .read = portcntrs_2_read, },
+};
+
+/*
+ * read the per-port QSFP data for port 1 (pidx 0)
+ */
+static ssize_t qsfp_1_read(struct file *file, char __user *buf,
+			   size_t count, loff_t *ppos)
+{
+	struct qib_devdata *dd = private2dd(file);
+	char *tmp;
+	int ret;
+
+	tmp = kmalloc(PAGE_SIZE, GFP_KERNEL);
+	if (!tmp)
+		return -ENOMEM;
+
+	ret = qib_qsfp_dump(dd->pport, tmp, PAGE_SIZE);
+	if (ret > 0)
+		ret = simple_read_from_buffer(buf, count, ppos, tmp, ret);
+	kfree(tmp);
+	return ret;
+}
+
+/*
+ * read the per-port QSFP data for port 2 (pidx 1)
+ */
+static ssize_t qsfp_2_read(struct file *file, char __user *buf,
+			   size_t count, loff_t *ppos)
+{
+	struct qib_devdata *dd = private2dd(file);
+	char *tmp;
+	int ret;
+
+	if (dd->num_pports < 2)
+		return -ENODEV;
+
+	tmp = kmalloc(PAGE_SIZE, GFP_KERNEL);
+	if (!tmp)
+		return -ENOMEM;
+
+	ret = qib_qsfp_dump(dd->pport + 1, tmp, PAGE_SIZE);
+	if (ret > 0)
+		ret = simple_read_from_buffer(buf, count, ppos, tmp, ret);
+	kfree(tmp);
+	return ret;
+}
+
+static const struct file_operations qsfp_ops[] = {
+	{ .read = qsfp_1_read, },
+	{ .read = qsfp_2_read, },
+};
+
+static ssize_t flash_read(struct file *file, char __user *buf,
+			  size_t count, loff_t *ppos)
+{
+	struct qib_devdata *dd;
+	ssize_t ret;
+	loff_t pos;
+	char *tmp;
+
+	pos = *ppos;
+
+	if (pos < 0) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	if (pos >= sizeof(struct qib_flash)) {
+		ret = 0;
+		goto bail;
+	}
+
+	if (count > sizeof(struct qib_flash) - pos)
+		count = sizeof(struct qib_flash) - pos;
+
+	tmp = kmalloc(count, GFP_KERNEL);
+	if (!tmp) {
+		ret = -ENOMEM;
+		goto bail;
+	}
+
+	dd = private2dd(file);
+	if (qib_eeprom_read(dd, pos, tmp, count)) {
+		qib_dev_err(dd, "failed to read from flash\n");
+		ret = -ENXIO;
+		goto bail_tmp;
+	}
+
+	if (copy_to_user(buf, tmp, count)) {
+		ret = -EFAULT;
+		goto bail_tmp;
+	}
+
+	*ppos = pos + count;
+	ret = count;
+
+bail_tmp:
+	kfree(tmp);
+
+bail:
+	return ret;
+}
+
+static ssize_t flash_write(struct file *file, const char __user *buf,
+			   size_t count, loff_t *ppos)
+{
+	struct qib_devdata *dd;
+	ssize_t ret;
+	loff_t pos;
+	char *tmp;
+
+	pos = *ppos;
+
+	if (pos != 0) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	if (count != sizeof(struct qib_flash)) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	tmp = kmalloc(count, GFP_KERNEL);
+	if (!tmp) {
+		ret = -ENOMEM;
+		goto bail;
+	}
+
+	if (copy_from_user(tmp, buf, count)) {
+		ret = -EFAULT;
+		goto bail_tmp;
+	}
+
+	dd = private2dd(file);
+	if (qib_eeprom_write(dd, pos, tmp, count)) {
+		ret = -ENXIO;
+		qib_dev_err(dd, "failed to write to flash\n");
+		goto bail_tmp;
+	}
+
+	*ppos = pos + count;
+	ret = count;
+
+bail_tmp:
+	kfree(tmp);
+
+bail:
+	return ret;
+}
+
+static const struct file_operations flash_ops = {
+	.read = flash_read,
+	.write = flash_write,
+};
+
+static int add_cntr_files(struct super_block *sb, struct qib_devdata *dd)
+{
+	struct dentry *dir, *tmp;
+	char unit[10];
+	int ret, i;
+
+	/* create the per-unit directory */
+	snprintf(unit, sizeof unit, "%u", dd->unit);
+	ret = create_file(unit, S_IFDIR|S_IRUGO|S_IXUGO, sb->s_root, &dir,
+			  &simple_dir_operations, dd);
+	if (ret) {
+		printk(KERN_ERR "create_file(%s) failed: %d\n", unit, ret);
+		goto bail;
+	}
+
+	/* create the files in the new directory */
+	ret = create_file("counters", S_IFREG|S_IRUGO, dir, &tmp,
+			  &cntr_ops[0], dd);
+	if (ret) {
+		printk(KERN_ERR "create_file(%s/counters) failed: %d\n",
+		       unit, ret);
+		goto bail;
+	}
+	ret = create_file("counter_names", S_IFREG|S_IRUGO, dir, &tmp,
+			  &cntr_ops[1], dd);
+	if (ret) {
+		printk(KERN_ERR "create_file(%s/counter_names) failed: %d\n",
+		       unit, ret);
+		goto bail;
+	}
+	ret = create_file("portcounter_names", S_IFREG|S_IRUGO, dir, &tmp,
+			  &portcntr_ops[0], dd);
+	if (ret) {
+		printk(KERN_ERR "create_file(%s/%s) failed: %d\n",
+		       unit, "portcounter_names", ret);
+		goto bail;
+	}
+	for (i = 1; i <= dd->num_pports; i++) {
+		char fname[24];
+
+		sprintf(fname, "port%dcounters", i);
+		/* create the files in the new directory */
+		ret = create_file(fname, S_IFREG|S_IRUGO, dir, &tmp,
+				  &portcntr_ops[i], dd);
+		if (ret) {
+			printk(KERN_ERR "create_file(%s/%s) failed: %d\n",
+				unit, fname, ret);
+			goto bail;
+		}
+		if (!(dd->flags & QIB_HAS_QSFP))
+			continue;
+		sprintf(fname, "qsfp%d", i);
+		ret = create_file(fname, S_IFREG|S_IRUGO, dir, &tmp,
+				  &qsfp_ops[i - 1], dd);
+		if (ret) {
+			printk(KERN_ERR "create_file(%s/%s) failed: %d\n",
+				unit, fname, ret);
+			goto bail;
+		}
+	}
+
+	ret = create_file("flash", S_IFREG|S_IWUSR|S_IRUGO, dir, &tmp,
+			  &flash_ops, dd);
+	if (ret)
+		printk(KERN_ERR "create_file(%s/flash) failed: %d\n",
+			unit, ret);
+bail:
+	return ret;
+}
+
+static int remove_file(struct dentry *parent, char *name)
+{
+	struct dentry *tmp;
+	int ret;
+
+	tmp = lookup_one_len(name, parent, strlen(name));
+
+	if (IS_ERR(tmp)) {
+		ret = PTR_ERR(tmp);
+		goto bail;
+	}
+
+	spin_lock(&dcache_lock);
+	spin_lock(&tmp->d_lock);
+	if (!(d_unhashed(tmp) && tmp->d_inode)) {
+		dget_locked(tmp);
+		__d_drop(tmp);
+		spin_unlock(&tmp->d_lock);
+		spin_unlock(&dcache_lock);
+		simple_unlink(parent->d_inode, tmp);
+	} else {
+		spin_unlock(&tmp->d_lock);
+		spin_unlock(&dcache_lock);
+	}
+
+	ret = 0;
+bail:
+	/*
+	 * We don't expect clients to care about the return value, but
+	 * it's there if they need it.
+	 */
+	return ret;
+}
+
+static int remove_device_files(struct super_block *sb,
+			       struct qib_devdata *dd)
+{
+	struct dentry *dir, *root;
+	char unit[10];
+	int ret, i;
+
+	root = dget(sb->s_root);
+	mutex_lock(&root->d_inode->i_mutex);
+	snprintf(unit, sizeof unit, "%u", dd->unit);
+	dir = lookup_one_len(unit, root, strlen(unit));
+
+	if (IS_ERR(dir)) {
+		ret = PTR_ERR(dir);
+		printk(KERN_ERR "Lookup of %s failed\n", unit);
+		goto bail;
+	}
+
+	remove_file(dir, "counters");
+	remove_file(dir, "counter_names");
+	remove_file(dir, "portcounter_names");
+	for (i = 0; i < dd->num_pports; i++) {
+		char fname[24];
+
+		sprintf(fname, "port%dcounters", i + 1);
+		remove_file(dir, fname);
+		if (dd->flags & QIB_HAS_QSFP) {
+			sprintf(fname, "qsfp%d", i + 1);
+			remove_file(dir, fname);
+		}
+	}
+	remove_file(dir, "flash");
+	d_delete(dir);
+	ret = simple_rmdir(root->d_inode, dir);
+
+bail:
+	mutex_unlock(&root->d_inode->i_mutex);
+	dput(root);
+	return ret;
+}
+
+/*
+ * This fills everything in when the fs is mounted, to handle umount/mount
+ * after device init.  The direct add_cntr_files() call handles adding
+ * them from the init code, when the fs is already mounted.
+ */
+static int qibfs_fill_super(struct super_block *sb, void *data, int silent)
+{
+	struct qib_devdata *dd, *tmp;
+	unsigned long flags;
+	int ret;
+
+	static struct tree_descr files[] = {
+		[2] = {"driver_stats", &driver_ops[0], S_IRUGO},
+		[3] = {"driver_stats_names", &driver_ops[1], S_IRUGO},
+		{""},
+	};
+
+	ret = simple_fill_super(sb, QIBFS_MAGIC, files);
+	if (ret) {
+		printk(KERN_ERR "simple_fill_super failed: %d\n", ret);
+		goto bail;
+	}
+
+	spin_lock_irqsave(&qib_devs_lock, flags);
+
+	list_for_each_entry_safe(dd, tmp, &qib_dev_list, list) {
+		spin_unlock_irqrestore(&qib_devs_lock, flags);
+		ret = add_cntr_files(sb, dd);
+		if (ret) {
+			deactivate_super(sb);
+			goto bail;
+		}
+		spin_lock_irqsave(&qib_devs_lock, flags);
+	}
+
+	spin_unlock_irqrestore(&qib_devs_lock, flags);
+
+bail:
+	return ret;
+}
+
+static int qibfs_get_sb(struct file_system_type *fs_type, int flags,
+			const char *dev_name, void *data, struct vfsmount *mnt)
+{
+	int ret = get_sb_single(fs_type, flags, data,
+				qibfs_fill_super, mnt);
+	if (ret >= 0)
+		qib_super = mnt->mnt_sb;
+	return ret;
+}
+
+static void qibfs_kill_super(struct super_block *s)
+{
+	kill_litter_super(s);
+	qib_super = NULL;
+}
+
+int qibfs_add(struct qib_devdata *dd)
+{
+	int ret;
+
+	/*
+	 * On first unit initialized, qib_super will not yet exist
+	 * because nobody has yet tried to mount the filesystem, so
+	 * we can't consider that to be an error; if an error occurs
+	 * during the mount, that will get a complaint, so this is OK.
+	 * add_cntr_files() for all units is done at mount from
+	 * qibfs_fill_super(), so one way or another, everything works.
+	 */
+	if (qib_super == NULL)
+		ret = 0;
+	else
+		ret = add_cntr_files(qib_super, dd);
+	return ret;
+}
+
+int qibfs_remove(struct qib_devdata *dd)
+{
+	int ret = 0;
+
+	if (qib_super)
+		ret = remove_device_files(qib_super, dd);
+
+	return ret;
+}
+
+static struct file_system_type qibfs_fs_type = {
+	.owner =        THIS_MODULE,
+	.name =         "ipathfs",
+	.get_sb =       qibfs_get_sb,
+	.kill_sb =      qibfs_kill_super,
+};
+
+int __init qib_init_qibfs(void)
+{
+	return register_filesystem(&qibfs_fs_type);
+}
+
+int __exit qib_exit_qibfs(void)
+{
+	return unregister_filesystem(&qibfs_fs_type);
+}

drivers/infiniband/hw/qib/qib_intr.c

@@ -0,0 +1,236 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009, 2010 QLogic Corporation.
+ * All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/pci.h>
+#include <linux/delay.h>
+
+#include "qib.h"
+#include "qib_common.h"
+
+/**
+ * qib_format_hwmsg - format a single hwerror message
+ * @msg message buffer
+ * @msgl length of message buffer
+ * @hwmsg message to add to message buffer
+ */
+static void qib_format_hwmsg(char *msg, size_t msgl, const char *hwmsg)
+{
+	strlcat(msg, "[", msgl);
+	strlcat(msg, hwmsg, msgl);
+	strlcat(msg, "]", msgl);
+}
+
+/**
+ * qib_format_hwerrors - format hardware error messages for display
+ * @hwerrs hardware errors bit vector
+ * @hwerrmsgs hardware error descriptions
+ * @nhwerrmsgs number of hwerrmsgs
+ * @msg message buffer
+ * @msgl message buffer length
+ */
+void qib_format_hwerrors(u64 hwerrs, const struct qib_hwerror_msgs *hwerrmsgs,
+			 size_t nhwerrmsgs, char *msg, size_t msgl)
+{
+	int i;
+
+	for (i = 0; i < nhwerrmsgs; i++)
+		if (hwerrs & hwerrmsgs[i].mask)
+			qib_format_hwmsg(msg, msgl, hwerrmsgs[i].msg);
+}
+
+static void signal_ib_event(struct qib_pportdata *ppd, enum ib_event_type ev)
+{
+	struct ib_event event;
+	struct qib_devdata *dd = ppd->dd;
+
+	event.device = &dd->verbs_dev.ibdev;
+	event.element.port_num = ppd->port;
+	event.event = ev;
+	ib_dispatch_event(&event);
+}
+
+void qib_handle_e_ibstatuschanged(struct qib_pportdata *ppd, u64 ibcs)
+{
+	struct qib_devdata *dd = ppd->dd;
+	unsigned long flags;
+	u32 lstate;
+	u8 ltstate;
+	enum ib_event_type ev = 0;
+
+	lstate = dd->f_iblink_state(ibcs); /* linkstate */
+	ltstate = dd->f_ibphys_portstate(ibcs);
+
+	/*
+	 * If linkstate transitions into INIT from any of the various down
+	 * states, or if it transitions from any of the up (INIT or better)
+	 * states into any of the down states (except link recovery), then
+	 * call the chip-specific code to take appropriate actions.
+	 */
+	if (lstate >= IB_PORT_INIT && (ppd->lflags & QIBL_LINKDOWN) &&
+	    ltstate == IB_PHYSPORTSTATE_LINKUP) {
+		/* transitioned to UP */
+		if (dd->f_ib_updown(ppd, 1, ibcs))
+			goto skip_ibchange; /* chip-code handled */
+	} else if (ppd->lflags & (QIBL_LINKINIT | QIBL_LINKARMED |
+		   QIBL_LINKACTIVE | QIBL_IB_FORCE_NOTIFY)) {
+		if (ltstate != IB_PHYSPORTSTATE_LINKUP &&
+		    ltstate <= IB_PHYSPORTSTATE_CFG_TRAIN &&
+		    dd->f_ib_updown(ppd, 0, ibcs))
+			goto skip_ibchange; /* chip-code handled */
+		qib_set_uevent_bits(ppd, _QIB_EVENT_LINKDOWN_BIT);
+	}
+
+	if (lstate != IB_PORT_DOWN) {
+		/* lstate is INIT, ARMED, or ACTIVE */
+		if (lstate != IB_PORT_ACTIVE) {
+			*ppd->statusp &= ~QIB_STATUS_IB_READY;
+			if (ppd->lflags & QIBL_LINKACTIVE)
+				ev = IB_EVENT_PORT_ERR;
+			spin_lock_irqsave(&ppd->lflags_lock, flags);
+			if (lstate == IB_PORT_ARMED) {
+				ppd->lflags |= QIBL_LINKARMED | QIBL_LINKV;
+				ppd->lflags &= ~(QIBL_LINKINIT |
+					QIBL_LINKDOWN | QIBL_LINKACTIVE);
+			} else {
+				ppd->lflags |= QIBL_LINKINIT | QIBL_LINKV;
+				ppd->lflags &= ~(QIBL_LINKARMED |
+					QIBL_LINKDOWN | QIBL_LINKACTIVE);
+			}
+			spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+			/* start a 75msec timer to clear symbol errors */
+			mod_timer(&ppd->symerr_clear_timer,
+				  msecs_to_jiffies(75));
+		} else if (ltstate == IB_PHYSPORTSTATE_LINKUP) {
+			/* active, but not active defered */
+			qib_hol_up(ppd); /* useful only for 6120 now */
+			*ppd->statusp |=
+				QIB_STATUS_IB_READY | QIB_STATUS_IB_CONF;
+			qib_clear_symerror_on_linkup((unsigned long)ppd);
+			spin_lock_irqsave(&ppd->lflags_lock, flags);
+			ppd->lflags |= QIBL_LINKACTIVE | QIBL_LINKV;
+			ppd->lflags &= ~(QIBL_LINKINIT |
+				QIBL_LINKDOWN | QIBL_LINKARMED);
+			spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+			if (dd->flags & QIB_HAS_SEND_DMA)
+				qib_sdma_process_event(ppd,
+					qib_sdma_event_e30_go_running);
+			ev = IB_EVENT_PORT_ACTIVE;
+			dd->f_setextled(ppd, 1);
+		}
+	} else { /* down */
+		if (ppd->lflags & QIBL_LINKACTIVE)
+			ev = IB_EVENT_PORT_ERR;
+		spin_lock_irqsave(&ppd->lflags_lock, flags);
+		ppd->lflags |= QIBL_LINKDOWN | QIBL_LINKV;
+		ppd->lflags &= ~(QIBL_LINKINIT |
+				 QIBL_LINKACTIVE | QIBL_LINKARMED);
+		spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+		*ppd->statusp &= ~QIB_STATUS_IB_READY;
+	}
+
+skip_ibchange:
+	ppd->lastibcstat = ibcs;
+	if (ev)
+		signal_ib_event(ppd, ev);
+	return;
+}
+
+void qib_clear_symerror_on_linkup(unsigned long opaque)
+{
+	struct qib_pportdata *ppd = (struct qib_pportdata *)opaque;
+
+	if (ppd->lflags & QIBL_LINKACTIVE)
+		return;
+
+	ppd->ibport_data.z_symbol_error_counter =
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_IBSYMBOLERR);
+}
+
+/*
+ * Handle receive interrupts for user ctxts; this means a user
+ * process was waiting for a packet to arrive, and didn't want
+ * to poll.
+ */
+void qib_handle_urcv(struct qib_devdata *dd, u64 ctxtr)
+{
+	struct qib_ctxtdata *rcd;
+	unsigned long flags;
+	int i;
+
+	spin_lock_irqsave(&dd->uctxt_lock, flags);
+	for (i = dd->first_user_ctxt; dd->rcd && i < dd->cfgctxts; i++) {
+		if (!(ctxtr & (1ULL << i)))
+			continue;
+		rcd = dd->rcd[i];
+		if (!rcd || !rcd->cnt)
+			continue;
+
+		if (test_and_clear_bit(QIB_CTXT_WAITING_RCV, &rcd->flag)) {
+			wake_up_interruptible(&rcd->wait);
+			dd->f_rcvctrl(rcd->ppd, QIB_RCVCTRL_INTRAVAIL_DIS,
+				      rcd->ctxt);
+		} else if (test_and_clear_bit(QIB_CTXT_WAITING_URG,
+					      &rcd->flag)) {
+			rcd->urgent++;
+			wake_up_interruptible(&rcd->wait);
+		}
+	}
+	spin_unlock_irqrestore(&dd->uctxt_lock, flags);
+}
+
+void qib_bad_intrstatus(struct qib_devdata *dd)
+{
+	static int allbits;
+
+	/* separate routine, for better optimization of qib_intr() */
+
+	/*
+	 * We print the message and disable interrupts, in hope of
+	 * having a better chance of debugging the problem.
+	 */
+	qib_dev_err(dd, "Read of chip interrupt status failed"
+		    " disabling interrupts\n");
+	if (allbits++) {
+		/* disable interrupt delivery, something is very wrong */
+		if (allbits == 2)
+			dd->f_set_intr_state(dd, 0);
+		if (allbits == 3) {
+			qib_dev_err(dd, "2nd bad interrupt status, "
+				    "unregistering interrupts\n");
+			dd->flags |= QIB_BADINTR;
+			dd->flags &= ~QIB_INITTED;
+			dd->f_free_irq(dd);
+		}
+	}
+}

drivers/infiniband/hw/qib/qib_keys.c

@@ -0,0 +1,328 @@
+/*
+ * Copyright (c) 2006, 2007, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include "qib.h"
+
+/**
+ * qib_alloc_lkey - allocate an lkey
+ * @rkt: lkey table in which to allocate the lkey
+ * @mr: memory region that this lkey protects
+ *
+ * Returns 1 if successful, otherwise returns 0.
+ */
+
+int qib_alloc_lkey(struct qib_lkey_table *rkt, struct qib_mregion *mr)
+{
+	unsigned long flags;
+	u32 r;
+	u32 n;
+	int ret;
+
+	spin_lock_irqsave(&rkt->lock, flags);
+
+	/* Find the next available LKEY */
+	r = rkt->next;
+	n = r;
+	for (;;) {
+		if (rkt->table[r] == NULL)
+			break;
+		r = (r + 1) & (rkt->max - 1);
+		if (r == n) {
+			spin_unlock_irqrestore(&rkt->lock, flags);
+			ret = 0;
+			goto bail;
+		}
+	}
+	rkt->next = (r + 1) & (rkt->max - 1);
+	/*
+	 * Make sure lkey is never zero which is reserved to indicate an
+	 * unrestricted LKEY.
+	 */
+	rkt->gen++;
+	mr->lkey = (r << (32 - ib_qib_lkey_table_size)) |
+		((((1 << (24 - ib_qib_lkey_table_size)) - 1) & rkt->gen)
+		 << 8);
+	if (mr->lkey == 0) {
+		mr->lkey |= 1 << 8;
+		rkt->gen++;
+	}
+	rkt->table[r] = mr;
+	spin_unlock_irqrestore(&rkt->lock, flags);
+
+	ret = 1;
+
+bail:
+	return ret;
+}
+
+/**
+ * qib_free_lkey - free an lkey
+ * @rkt: table from which to free the lkey
+ * @lkey: lkey id to free
+ */
+int qib_free_lkey(struct qib_ibdev *dev, struct qib_mregion *mr)
+{
+	unsigned long flags;
+	u32 lkey = mr->lkey;
+	u32 r;
+	int ret;
+
+	spin_lock_irqsave(&dev->lk_table.lock, flags);
+	if (lkey == 0) {
+		if (dev->dma_mr && dev->dma_mr == mr) {
+			ret = atomic_read(&dev->dma_mr->refcount);
+			if (!ret)
+				dev->dma_mr = NULL;
+		} else
+			ret = 0;
+	} else {
+		r = lkey >> (32 - ib_qib_lkey_table_size);
+		ret = atomic_read(&dev->lk_table.table[r]->refcount);
+		if (!ret)
+			dev->lk_table.table[r] = NULL;
+	}
+	spin_unlock_irqrestore(&dev->lk_table.lock, flags);
+
+	if (ret)
+		ret = -EBUSY;
+	return ret;
+}
+
+/**
+ * qib_lkey_ok - check IB SGE for validity and initialize
+ * @rkt: table containing lkey to check SGE against
+ * @isge: outgoing internal SGE
+ * @sge: SGE to check
+ * @acc: access flags
+ *
+ * Return 1 if valid and successful, otherwise returns 0.
+ *
+ * Check the IB SGE for validity and initialize our internal version
+ * of it.
+ */
+int qib_lkey_ok(struct qib_lkey_table *rkt, struct qib_pd *pd,
+		struct qib_sge *isge, struct ib_sge *sge, int acc)
+{
+	struct qib_mregion *mr;
+	unsigned n, m;
+	size_t off;
+	int ret = 0;
+	unsigned long flags;
+
+	/*
+	 * We use LKEY == zero for kernel virtual addresses
+	 * (see qib_get_dma_mr and qib_dma.c).
+	 */
+	spin_lock_irqsave(&rkt->lock, flags);
+	if (sge->lkey == 0) {
+		struct qib_ibdev *dev = to_idev(pd->ibpd.device);
+
+		if (pd->user)
+			goto bail;
+		if (!dev->dma_mr)
+			goto bail;
+		atomic_inc(&dev->dma_mr->refcount);
+		isge->mr = dev->dma_mr;
+		isge->vaddr = (void *) sge->addr;
+		isge->length = sge->length;
+		isge->sge_length = sge->length;
+		isge->m = 0;
+		isge->n = 0;
+		goto ok;
+	}
+	mr = rkt->table[(sge->lkey >> (32 - ib_qib_lkey_table_size))];
+	if (unlikely(mr == NULL || mr->lkey != sge->lkey ||
+		     mr->pd != &pd->ibpd))
+		goto bail;
+
+	off = sge->addr - mr->user_base;
+	if (unlikely(sge->addr < mr->user_base ||
+		     off + sge->length > mr->length ||
+		     (mr->access_flags & acc) != acc))
+		goto bail;
+
+	off += mr->offset;
+	m = 0;
+	n = 0;
+	while (off >= mr->map[m]->segs[n].length) {
+		off -= mr->map[m]->segs[n].length;
+		n++;
+		if (n >= QIB_SEGSZ) {
+			m++;
+			n = 0;
+		}
+	}
+	atomic_inc(&mr->refcount);
+	isge->mr = mr;
+	isge->vaddr = mr->map[m]->segs[n].vaddr + off;
+	isge->length = mr->map[m]->segs[n].length - off;
+	isge->sge_length = sge->length;
+	isge->m = m;
+	isge->n = n;
+ok:
+	ret = 1;
+bail:
+	spin_unlock_irqrestore(&rkt->lock, flags);
+	return ret;
+}
+
+/**
+ * qib_rkey_ok - check the IB virtual address, length, and RKEY
+ * @dev: infiniband device
+ * @ss: SGE state
+ * @len: length of data
+ * @vaddr: virtual address to place data
+ * @rkey: rkey to check
+ * @acc: access flags
+ *
+ * Return 1 if successful, otherwise 0.
+ */
+int qib_rkey_ok(struct qib_qp *qp, struct qib_sge *sge,
+		u32 len, u64 vaddr, u32 rkey, int acc)
+{
+	struct qib_lkey_table *rkt = &to_idev(qp->ibqp.device)->lk_table;
+	struct qib_mregion *mr;
+	unsigned n, m;
+	size_t off;
+	int ret = 0;
+	unsigned long flags;
+
+	/*
+	 * We use RKEY == zero for kernel virtual addresses
+	 * (see qib_get_dma_mr and qib_dma.c).
+	 */
+	spin_lock_irqsave(&rkt->lock, flags);
+	if (rkey == 0) {
+		struct qib_pd *pd = to_ipd(qp->ibqp.pd);
+		struct qib_ibdev *dev = to_idev(pd->ibpd.device);
+
+		if (pd->user)
+			goto bail;
+		if (!dev->dma_mr)
+			goto bail;
+		atomic_inc(&dev->dma_mr->refcount);
+		sge->mr = dev->dma_mr;
+		sge->vaddr = (void *) vaddr;
+		sge->length = len;
+		sge->sge_length = len;
+		sge->m = 0;
+		sge->n = 0;
+		goto ok;
+	}
+
+	mr = rkt->table[(rkey >> (32 - ib_qib_lkey_table_size))];
+	if (unlikely(mr == NULL || mr->lkey != rkey || qp->ibqp.pd != mr->pd))
+		goto bail;
+
+	off = vaddr - mr->iova;
+	if (unlikely(vaddr < mr->iova || off + len > mr->length ||
+		     (mr->access_flags & acc) == 0))
+		goto bail;
+
+	off += mr->offset;
+	m = 0;
+	n = 0;
+	while (off >= mr->map[m]->segs[n].length) {
+		off -= mr->map[m]->segs[n].length;
+		n++;
+		if (n >= QIB_SEGSZ) {
+			m++;
+			n = 0;
+		}
+	}
+	atomic_inc(&mr->refcount);
+	sge->mr = mr;
+	sge->vaddr = mr->map[m]->segs[n].vaddr + off;
+	sge->length = mr->map[m]->segs[n].length - off;
+	sge->sge_length = len;
+	sge->m = m;
+	sge->n = n;
+ok:
+	ret = 1;
+bail:
+	spin_unlock_irqrestore(&rkt->lock, flags);
+	return ret;
+}
+
+/*
+ * Initialize the memory region specified by the work reqeust.
+ */
+int qib_fast_reg_mr(struct qib_qp *qp, struct ib_send_wr *wr)
+{
+	struct qib_lkey_table *rkt = &to_idev(qp->ibqp.device)->lk_table;
+	struct qib_pd *pd = to_ipd(qp->ibqp.pd);
+	struct qib_mregion *mr;
+	u32 rkey = wr->wr.fast_reg.rkey;
+	unsigned i, n, m;
+	int ret = -EINVAL;
+	unsigned long flags;
+	u64 *page_list;
+	size_t ps;
+
+	spin_lock_irqsave(&rkt->lock, flags);
+	if (pd->user || rkey == 0)
+		goto bail;
+
+	mr = rkt->table[(rkey >> (32 - ib_qib_lkey_table_size))];
+	if (unlikely(mr == NULL || qp->ibqp.pd != mr->pd))
+		goto bail;
+
+	if (wr->wr.fast_reg.page_list_len > mr->max_segs)
+		goto bail;
+
+	ps = 1UL << wr->wr.fast_reg.page_shift;
+	if (wr->wr.fast_reg.length > ps * wr->wr.fast_reg.page_list_len)
+		goto bail;
+
+	mr->user_base = wr->wr.fast_reg.iova_start;
+	mr->iova = wr->wr.fast_reg.iova_start;
+	mr->lkey = rkey;
+	mr->length = wr->wr.fast_reg.length;
+	mr->access_flags = wr->wr.fast_reg.access_flags;
+	page_list = wr->wr.fast_reg.page_list->page_list;
+	m = 0;
+	n = 0;
+	for (i = 0; i < wr->wr.fast_reg.page_list_len; i++) {
+		mr->map[m]->segs[n].vaddr = (void *) page_list[i];
+		mr->map[m]->segs[n].length = ps;
+		if (++n == QIB_SEGSZ) {
+			m++;
+			n = 0;
+		}
+	}
+
+	ret = 0;
+bail:
+	spin_unlock_irqrestore(&rkt->lock, flags);
+	return ret;
+}

drivers/infiniband/hw/qib/qib_mad.h

@@ -0,0 +1,373 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009, 2010 QLogic Corporation.
+ * All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#define IB_SMP_UNSUP_VERSION    cpu_to_be16(0x0004)
+#define IB_SMP_UNSUP_METHOD     cpu_to_be16(0x0008)
+#define IB_SMP_UNSUP_METH_ATTR  cpu_to_be16(0x000C)
+#define IB_SMP_INVALID_FIELD    cpu_to_be16(0x001C)
+
+struct ib_node_info {
+	u8 base_version;
+	u8 class_version;
+	u8 node_type;
+	u8 num_ports;
+	__be64 sys_guid;
+	__be64 node_guid;
+	__be64 port_guid;
+	__be16 partition_cap;
+	__be16 device_id;
+	__be32 revision;
+	u8 local_port_num;
+	u8 vendor_id[3];
+} __attribute__ ((packed));
+
+struct ib_mad_notice_attr {
+	u8 generic_type;
+	u8 prod_type_msb;
+	__be16 prod_type_lsb;
+	__be16 trap_num;
+	__be16 issuer_lid;
+	__be16 toggle_count;
+
+	union {
+		struct {
+			u8	details[54];
+		} raw_data;
+
+		struct {
+			__be16	reserved;
+			__be16	lid;		/* where violation happened */
+			u8	port_num;	/* where violation happened */
+		} __attribute__ ((packed)) ntc_129_131;
+
+		struct {
+			__be16	reserved;
+			__be16	lid;		/* LID where change occured */
+			u8	reserved2;
+			u8	local_changes;	/* low bit - local changes */
+			__be32	new_cap_mask;	/* new capability mask */
+			u8	reserved3;
+			u8	change_flags;	/* low 3 bits only */
+		} __attribute__ ((packed)) ntc_144;
+
+		struct {
+			__be16	reserved;
+			__be16	lid;		/* lid where sys guid changed */
+			__be16	reserved2;
+			__be64	new_sys_guid;
+		} __attribute__ ((packed)) ntc_145;
+
+		struct {
+			__be16	reserved;
+			__be16	lid;
+			__be16	dr_slid;
+			u8	method;
+			u8	reserved2;
+			__be16	attr_id;
+			__be32	attr_mod;
+			__be64	mkey;
+			u8	reserved3;
+			u8	dr_trunc_hop;
+			u8	dr_rtn_path[30];
+		} __attribute__ ((packed)) ntc_256;
+
+		struct {
+			__be16		reserved;
+			__be16		lid1;
+			__be16		lid2;
+			__be32		key;
+			__be32		sl_qp1;	/* SL: high 4 bits */
+			__be32		qp2;	/* high 8 bits reserved */
+			union ib_gid	gid1;
+			union ib_gid	gid2;
+		} __attribute__ ((packed)) ntc_257_258;
+
+	} details;
+};
+
+/*
+ * Generic trap/notice types
+ */
+#define IB_NOTICE_TYPE_FATAL	0x80
+#define IB_NOTICE_TYPE_URGENT	0x81
+#define IB_NOTICE_TYPE_SECURITY	0x82
+#define IB_NOTICE_TYPE_SM	0x83
+#define IB_NOTICE_TYPE_INFO	0x84
+
+/*
+ * Generic trap/notice producers
+ */
+#define IB_NOTICE_PROD_CA		cpu_to_be16(1)
+#define IB_NOTICE_PROD_SWITCH		cpu_to_be16(2)
+#define IB_NOTICE_PROD_ROUTER		cpu_to_be16(3)
+#define IB_NOTICE_PROD_CLASS_MGR	cpu_to_be16(4)
+
+/*
+ * Generic trap/notice numbers
+ */
+#define IB_NOTICE_TRAP_LLI_THRESH	cpu_to_be16(129)
+#define IB_NOTICE_TRAP_EBO_THRESH	cpu_to_be16(130)
+#define IB_NOTICE_TRAP_FLOW_UPDATE	cpu_to_be16(131)
+#define IB_NOTICE_TRAP_CAP_MASK_CHG	cpu_to_be16(144)
+#define IB_NOTICE_TRAP_SYS_GUID_CHG	cpu_to_be16(145)
+#define IB_NOTICE_TRAP_BAD_MKEY		cpu_to_be16(256)
+#define IB_NOTICE_TRAP_BAD_PKEY		cpu_to_be16(257)
+#define IB_NOTICE_TRAP_BAD_QKEY		cpu_to_be16(258)
+
+/*
+ * Repress trap/notice flags
+ */
+#define IB_NOTICE_REPRESS_LLI_THRESH	(1 << 0)
+#define IB_NOTICE_REPRESS_EBO_THRESH	(1 << 1)
+#define IB_NOTICE_REPRESS_FLOW_UPDATE	(1 << 2)
+#define IB_NOTICE_REPRESS_CAP_MASK_CHG	(1 << 3)
+#define IB_NOTICE_REPRESS_SYS_GUID_CHG	(1 << 4)
+#define IB_NOTICE_REPRESS_BAD_MKEY	(1 << 5)
+#define IB_NOTICE_REPRESS_BAD_PKEY	(1 << 6)
+#define IB_NOTICE_REPRESS_BAD_QKEY	(1 << 7)
+
+/*
+ * Generic trap/notice other local changes flags (trap 144).
+ */
+#define IB_NOTICE_TRAP_LSE_CHG		0x04	/* Link Speed Enable changed */
+#define IB_NOTICE_TRAP_LWE_CHG		0x02	/* Link Width Enable changed */
+#define IB_NOTICE_TRAP_NODE_DESC_CHG	0x01
+
+/*
+ * Generic trap/notice M_Key volation flags in dr_trunc_hop (trap 256).
+ */
+#define IB_NOTICE_TRAP_DR_NOTICE	0x80
+#define IB_NOTICE_TRAP_DR_TRUNC		0x40
+
+struct ib_vl_weight_elem {
+	u8      vl;     /* Only low 4 bits, upper 4 bits reserved */
+	u8      weight;
+};
+
+#define IB_VLARB_LOWPRI_0_31    1
+#define IB_VLARB_LOWPRI_32_63   2
+#define IB_VLARB_HIGHPRI_0_31   3
+#define IB_VLARB_HIGHPRI_32_63  4
+
+/*
+ * PMA class portinfo capability mask bits
+ */
+#define IB_PMA_CLASS_CAP_ALLPORTSELECT  cpu_to_be16(1 << 8)
+#define IB_PMA_CLASS_CAP_EXT_WIDTH      cpu_to_be16(1 << 9)
+#define IB_PMA_CLASS_CAP_XMIT_WAIT      cpu_to_be16(1 << 12)
+
+#define IB_PMA_CLASS_PORT_INFO          cpu_to_be16(0x0001)
+#define IB_PMA_PORT_SAMPLES_CONTROL     cpu_to_be16(0x0010)
+#define IB_PMA_PORT_SAMPLES_RESULT      cpu_to_be16(0x0011)
+#define IB_PMA_PORT_COUNTERS            cpu_to_be16(0x0012)
+#define IB_PMA_PORT_COUNTERS_EXT        cpu_to_be16(0x001D)
+#define IB_PMA_PORT_SAMPLES_RESULT_EXT  cpu_to_be16(0x001E)
+#define IB_PMA_PORT_COUNTERS_CONG       cpu_to_be16(0xFF00)
+
+struct ib_perf {
+	u8 base_version;
+	u8 mgmt_class;
+	u8 class_version;
+	u8 method;
+	__be16 status;
+	__be16 unused;
+	__be64 tid;
+	__be16 attr_id;
+	__be16 resv;
+	__be32 attr_mod;
+	u8 reserved[40];
+	u8 data[192];
+} __attribute__ ((packed));
+
+struct ib_pma_classportinfo {
+	u8 base_version;
+	u8 class_version;
+	__be16 cap_mask;
+	u8 reserved[3];
+	u8 resp_time_value;     /* only lower 5 bits */
+	union ib_gid redirect_gid;
+	__be32 redirect_tc_sl_fl;       /* 8, 4, 20 bits respectively */
+	__be16 redirect_lid;
+	__be16 redirect_pkey;
+	__be32 redirect_qp;     /* only lower 24 bits */
+	__be32 redirect_qkey;
+	union ib_gid trap_gid;
+	__be32 trap_tc_sl_fl;   /* 8, 4, 20 bits respectively */
+	__be16 trap_lid;
+	__be16 trap_pkey;
+	__be32 trap_hl_qp;      /* 8, 24 bits respectively */
+	__be32 trap_qkey;
+} __attribute__ ((packed));
+
+struct ib_pma_portsamplescontrol {
+	u8 opcode;
+	u8 port_select;
+	u8 tick;
+	u8 counter_width;       /* only lower 3 bits */
+	__be32 counter_mask0_9; /* 2, 10 * 3, bits */
+	__be16 counter_mask10_14;       /* 1, 5 * 3, bits */
+	u8 sample_mechanisms;
+	u8 sample_status;       /* only lower 2 bits */
+	__be64 option_mask;
+	__be64 vendor_mask;
+	__be32 sample_start;
+	__be32 sample_interval;
+	__be16 tag;
+	__be16 counter_select[15];
+} __attribute__ ((packed));
+
+struct ib_pma_portsamplesresult {
+	__be16 tag;
+	__be16 sample_status;   /* only lower 2 bits */
+	__be32 counter[15];
+} __attribute__ ((packed));
+
+struct ib_pma_portsamplesresult_ext {
+	__be16 tag;
+	__be16 sample_status;   /* only lower 2 bits */
+	__be32 extended_width;  /* only upper 2 bits */
+	__be64 counter[15];
+} __attribute__ ((packed));
+
+struct ib_pma_portcounters {
+	u8 reserved;
+	u8 port_select;
+	__be16 counter_select;
+	__be16 symbol_error_counter;
+	u8 link_error_recovery_counter;
+	u8 link_downed_counter;
+	__be16 port_rcv_errors;
+	__be16 port_rcv_remphys_errors;
+	__be16 port_rcv_switch_relay_errors;
+	__be16 port_xmit_discards;
+	u8 port_xmit_constraint_errors;
+	u8 port_rcv_constraint_errors;
+	u8 reserved1;
+	u8 lli_ebor_errors;     /* 4, 4, bits */
+	__be16 reserved2;
+	__be16 vl15_dropped;
+	__be32 port_xmit_data;
+	__be32 port_rcv_data;
+	__be32 port_xmit_packets;
+	__be32 port_rcv_packets;
+} __attribute__ ((packed));
+
+struct ib_pma_portcounters_cong {
+	u8 reserved;
+	u8 reserved1;
+	__be16 port_check_rate;
+	__be16 symbol_error_counter;
+	u8 link_error_recovery_counter;
+	u8 link_downed_counter;
+	__be16 port_rcv_errors;
+	__be16 port_rcv_remphys_errors;
+	__be16 port_rcv_switch_relay_errors;
+	__be16 port_xmit_discards;
+	u8 port_xmit_constraint_errors;
+	u8 port_rcv_constraint_errors;
+	u8 reserved2;
+	u8 lli_ebor_errors;    /* 4, 4, bits */
+	__be16 reserved3;
+	__be16 vl15_dropped;
+	__be64 port_xmit_data;
+	__be64 port_rcv_data;
+	__be64 port_xmit_packets;
+	__be64 port_rcv_packets;
+	__be64 port_xmit_wait;
+	__be64 port_adr_events;
+} __attribute__ ((packed));
+
+#define IB_PMA_CONG_HW_CONTROL_TIMER            0x00
+#define IB_PMA_CONG_HW_CONTROL_SAMPLE           0x01
+
+#define QIB_XMIT_RATE_UNSUPPORTED               0x0
+#define QIB_XMIT_RATE_PICO                      0x7
+/* number of 4nsec cycles equaling 2secs */
+#define QIB_CONG_TIMER_PSINTERVAL               0x1DCD64EC
+
+#define IB_PMA_SEL_SYMBOL_ERROR                 cpu_to_be16(0x0001)
+#define IB_PMA_SEL_LINK_ERROR_RECOVERY          cpu_to_be16(0x0002)
+#define IB_PMA_SEL_LINK_DOWNED                  cpu_to_be16(0x0004)
+#define IB_PMA_SEL_PORT_RCV_ERRORS              cpu_to_be16(0x0008)
+#define IB_PMA_SEL_PORT_RCV_REMPHYS_ERRORS      cpu_to_be16(0x0010)
+#define IB_PMA_SEL_PORT_XMIT_DISCARDS           cpu_to_be16(0x0040)
+#define IB_PMA_SEL_LOCAL_LINK_INTEGRITY_ERRORS  cpu_to_be16(0x0200)
+#define IB_PMA_SEL_EXCESSIVE_BUFFER_OVERRUNS    cpu_to_be16(0x0400)
+#define IB_PMA_SEL_PORT_VL15_DROPPED            cpu_to_be16(0x0800)
+#define IB_PMA_SEL_PORT_XMIT_DATA               cpu_to_be16(0x1000)
+#define IB_PMA_SEL_PORT_RCV_DATA                cpu_to_be16(0x2000)
+#define IB_PMA_SEL_PORT_XMIT_PACKETS            cpu_to_be16(0x4000)
+#define IB_PMA_SEL_PORT_RCV_PACKETS             cpu_to_be16(0x8000)
+
+#define IB_PMA_SEL_CONG_ALL                     0x01
+#define IB_PMA_SEL_CONG_PORT_DATA               0x02
+#define IB_PMA_SEL_CONG_XMIT                    0x04
+#define IB_PMA_SEL_CONG_ROUTING                 0x08
+
+struct ib_pma_portcounters_ext {
+	u8 reserved;
+	u8 port_select;
+	__be16 counter_select;
+	__be32 reserved1;
+	__be64 port_xmit_data;
+	__be64 port_rcv_data;
+	__be64 port_xmit_packets;
+	__be64 port_rcv_packets;
+	__be64 port_unicast_xmit_packets;
+	__be64 port_unicast_rcv_packets;
+	__be64 port_multicast_xmit_packets;
+	__be64 port_multicast_rcv_packets;
+} __attribute__ ((packed));
+
+#define IB_PMA_SELX_PORT_XMIT_DATA              cpu_to_be16(0x0001)
+#define IB_PMA_SELX_PORT_RCV_DATA               cpu_to_be16(0x0002)
+#define IB_PMA_SELX_PORT_XMIT_PACKETS           cpu_to_be16(0x0004)
+#define IB_PMA_SELX_PORT_RCV_PACKETS            cpu_to_be16(0x0008)
+#define IB_PMA_SELX_PORT_UNI_XMIT_PACKETS       cpu_to_be16(0x0010)
+#define IB_PMA_SELX_PORT_UNI_RCV_PACKETS        cpu_to_be16(0x0020)
+#define IB_PMA_SELX_PORT_MULTI_XMIT_PACKETS     cpu_to_be16(0x0040)
+#define IB_PMA_SELX_PORT_MULTI_RCV_PACKETS      cpu_to_be16(0x0080)
+
+/*
+ * The PortSamplesControl.CounterMasks field is an array of 3 bit fields
+ * which specify the N'th counter's capabilities. See ch. 16.1.3.2.
+ * We support 5 counters which only count the mandatory quantities.
+ */
+#define COUNTER_MASK(q, n) (q << ((9 - n) * 3))
+#define COUNTER_MASK0_9 \
+	cpu_to_be32(COUNTER_MASK(1, 0) | \
+		    COUNTER_MASK(1, 1) | \
+		    COUNTER_MASK(1, 2) | \
+		    COUNTER_MASK(1, 3) | \
+		    COUNTER_MASK(1, 4))

drivers/infiniband/hw/qib/qib_mmap.c

@@ -0,0 +1,174 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/module.h>
+#include <linux/slab.h>
+#include <linux/vmalloc.h>
+#include <linux/mm.h>
+#include <linux/errno.h>
+#include <asm/pgtable.h>
+
+#include "qib_verbs.h"
+
+/**
+ * qib_release_mmap_info - free mmap info structure
+ * @ref: a pointer to the kref within struct qib_mmap_info
+ */
+void qib_release_mmap_info(struct kref *ref)
+{
+	struct qib_mmap_info *ip =
+		container_of(ref, struct qib_mmap_info, ref);
+	struct qib_ibdev *dev = to_idev(ip->context->device);
+
+	spin_lock_irq(&dev->pending_lock);
+	list_del(&ip->pending_mmaps);
+	spin_unlock_irq(&dev->pending_lock);
+
+	vfree(ip->obj);
+	kfree(ip);
+}
+
+/*
+ * open and close keep track of how many times the CQ is mapped,
+ * to avoid releasing it.
+ */
+static void qib_vma_open(struct vm_area_struct *vma)
+{
+	struct qib_mmap_info *ip = vma->vm_private_data;
+
+	kref_get(&ip->ref);
+}
+
+static void qib_vma_close(struct vm_area_struct *vma)
+{
+	struct qib_mmap_info *ip = vma->vm_private_data;
+
+	kref_put(&ip->ref, qib_release_mmap_info);
+}
+
+static struct vm_operations_struct qib_vm_ops = {
+	.open =     qib_vma_open,
+	.close =    qib_vma_close,
+};
+
+/**
+ * qib_mmap - create a new mmap region
+ * @context: the IB user context of the process making the mmap() call
+ * @vma: the VMA to be initialized
+ * Return zero if the mmap is OK. Otherwise, return an errno.
+ */
+int qib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
+{
+	struct qib_ibdev *dev = to_idev(context->device);
+	unsigned long offset = vma->vm_pgoff << PAGE_SHIFT;
+	unsigned long size = vma->vm_end - vma->vm_start;
+	struct qib_mmap_info *ip, *pp;
+	int ret = -EINVAL;
+
+	/*
+	 * Search the device's list of objects waiting for a mmap call.
+	 * Normally, this list is very short since a call to create a
+	 * CQ, QP, or SRQ is soon followed by a call to mmap().
+	 */
+	spin_lock_irq(&dev->pending_lock);
+	list_for_each_entry_safe(ip, pp, &dev->pending_mmaps,
+				 pending_mmaps) {
+		/* Only the creator is allowed to mmap the object */
+		if (context != ip->context || (__u64) offset != ip->offset)
+			continue;
+		/* Don't allow a mmap larger than the object. */
+		if (size > ip->size)
+			break;
+
+		list_del_init(&ip->pending_mmaps);
+		spin_unlock_irq(&dev->pending_lock);
+
+		ret = remap_vmalloc_range(vma, ip->obj, 0);
+		if (ret)
+			goto done;
+		vma->vm_ops = &qib_vm_ops;
+		vma->vm_private_data = ip;
+		qib_vma_open(vma);
+		goto done;
+	}
+	spin_unlock_irq(&dev->pending_lock);
+done:
+	return ret;
+}
+
+/*
+ * Allocate information for qib_mmap
+ */
+struct qib_mmap_info *qib_create_mmap_info(struct qib_ibdev *dev,
+					   u32 size,
+					   struct ib_ucontext *context,
+					   void *obj) {
+	struct qib_mmap_info *ip;
+
+	ip = kmalloc(sizeof *ip, GFP_KERNEL);
+	if (!ip)
+		goto bail;
+
+	size = PAGE_ALIGN(size);
+
+	spin_lock_irq(&dev->mmap_offset_lock);
+	if (dev->mmap_offset == 0)
+		dev->mmap_offset = PAGE_SIZE;
+	ip->offset = dev->mmap_offset;
+	dev->mmap_offset += size;
+	spin_unlock_irq(&dev->mmap_offset_lock);
+
+	INIT_LIST_HEAD(&ip->pending_mmaps);
+	ip->size = size;
+	ip->context = context;
+	ip->obj = obj;
+	kref_init(&ip->ref);
+
+bail:
+	return ip;
+}
+
+void qib_update_mmap_info(struct qib_ibdev *dev, struct qib_mmap_info *ip,
+			  u32 size, void *obj)
+{
+	size = PAGE_ALIGN(size);
+
+	spin_lock_irq(&dev->mmap_offset_lock);
+	if (dev->mmap_offset == 0)
+		dev->mmap_offset = PAGE_SIZE;
+	ip->offset = dev->mmap_offset;
+	dev->mmap_offset += size;
+	spin_unlock_irq(&dev->mmap_offset_lock);
+
+	ip->size = size;
+	ip->obj = obj;
+}

drivers/infiniband/hw/qib/qib_mr.c

@@ -0,0 +1,503 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <rdma/ib_umem.h>
+#include <rdma/ib_smi.h>
+
+#include "qib.h"
+
+/* Fast memory region */
+struct qib_fmr {
+	struct ib_fmr ibfmr;
+	u8 page_shift;
+	struct qib_mregion mr;        /* must be last */
+};
+
+static inline struct qib_fmr *to_ifmr(struct ib_fmr *ibfmr)
+{
+	return container_of(ibfmr, struct qib_fmr, ibfmr);
+}
+
+/**
+ * qib_get_dma_mr - get a DMA memory region
+ * @pd: protection domain for this memory region
+ * @acc: access flags
+ *
+ * Returns the memory region on success, otherwise returns an errno.
+ * Note that all DMA addresses should be created via the
+ * struct ib_dma_mapping_ops functions (see qib_dma.c).
+ */
+struct ib_mr *qib_get_dma_mr(struct ib_pd *pd, int acc)
+{
+	struct qib_ibdev *dev = to_idev(pd->device);
+	struct qib_mr *mr;
+	struct ib_mr *ret;
+	unsigned long flags;
+
+	if (to_ipd(pd)->user) {
+		ret = ERR_PTR(-EPERM);
+		goto bail;
+	}
+
+	mr = kzalloc(sizeof *mr, GFP_KERNEL);
+	if (!mr) {
+		ret = ERR_PTR(-ENOMEM);
+		goto bail;
+	}
+
+	mr->mr.access_flags = acc;
+	atomic_set(&mr->mr.refcount, 0);
+
+	spin_lock_irqsave(&dev->lk_table.lock, flags);
+	if (!dev->dma_mr)
+		dev->dma_mr = &mr->mr;
+	spin_unlock_irqrestore(&dev->lk_table.lock, flags);
+
+	ret = &mr->ibmr;
+
+bail:
+	return ret;
+}
+
+static struct qib_mr *alloc_mr(int count, struct qib_lkey_table *lk_table)
+{
+	struct qib_mr *mr;
+	int m, i = 0;
+
+	/* Allocate struct plus pointers to first level page tables. */
+	m = (count + QIB_SEGSZ - 1) / QIB_SEGSZ;
+	mr = kmalloc(sizeof *mr + m * sizeof mr->mr.map[0], GFP_KERNEL);
+	if (!mr)
+		goto done;
+
+	/* Allocate first level page tables. */
+	for (; i < m; i++) {
+		mr->mr.map[i] = kmalloc(sizeof *mr->mr.map[0], GFP_KERNEL);
+		if (!mr->mr.map[i])
+			goto bail;
+	}
+	mr->mr.mapsz = m;
+	mr->mr.max_segs = count;
+
+	/*
+	 * ib_reg_phys_mr() will initialize mr->ibmr except for
+	 * lkey and rkey.
+	 */
+	if (!qib_alloc_lkey(lk_table, &mr->mr))
+		goto bail;
+	mr->ibmr.lkey = mr->mr.lkey;
+	mr->ibmr.rkey = mr->mr.lkey;
+
+	atomic_set(&mr->mr.refcount, 0);
+	goto done;
+
+bail:
+	while (i)
+		kfree(mr->mr.map[--i]);
+	kfree(mr);
+	mr = NULL;
+
+done:
+	return mr;
+}
+
+/**
+ * qib_reg_phys_mr - register a physical memory region
+ * @pd: protection domain for this memory region
+ * @buffer_list: pointer to the list of physical buffers to register
+ * @num_phys_buf: the number of physical buffers to register
+ * @iova_start: the starting address passed over IB which maps to this MR
+ *
+ * Returns the memory region on success, otherwise returns an errno.
+ */
+struct ib_mr *qib_reg_phys_mr(struct ib_pd *pd,
+			      struct ib_phys_buf *buffer_list,
+			      int num_phys_buf, int acc, u64 *iova_start)
+{
+	struct qib_mr *mr;
+	int n, m, i;
+	struct ib_mr *ret;
+
+	mr = alloc_mr(num_phys_buf, &to_idev(pd->device)->lk_table);
+	if (mr == NULL) {
+		ret = ERR_PTR(-ENOMEM);
+		goto bail;
+	}
+
+	mr->mr.pd = pd;
+	mr->mr.user_base = *iova_start;
+	mr->mr.iova = *iova_start;
+	mr->mr.length = 0;
+	mr->mr.offset = 0;
+	mr->mr.access_flags = acc;
+	mr->umem = NULL;
+
+	m = 0;
+	n = 0;
+	for (i = 0; i < num_phys_buf; i++) {
+		mr->mr.map[m]->segs[n].vaddr = (void *) buffer_list[i].addr;
+		mr->mr.map[m]->segs[n].length = buffer_list[i].size;
+		mr->mr.length += buffer_list[i].size;
+		n++;
+		if (n == QIB_SEGSZ) {
+			m++;
+			n = 0;
+		}
+	}
+
+	ret = &mr->ibmr;
+
+bail:
+	return ret;
+}
+
+/**
+ * qib_reg_user_mr - register a userspace memory region
+ * @pd: protection domain for this memory region
+ * @start: starting userspace address
+ * @length: length of region to register
+ * @virt_addr: virtual address to use (from HCA's point of view)
+ * @mr_access_flags: access flags for this memory region
+ * @udata: unused by the QLogic_IB driver
+ *
+ * Returns the memory region on success, otherwise returns an errno.
+ */
+struct ib_mr *qib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
+			      u64 virt_addr, int mr_access_flags,
+			      struct ib_udata *udata)
+{
+	struct qib_mr *mr;
+	struct ib_umem *umem;
+	struct ib_umem_chunk *chunk;
+	int n, m, i;
+	struct ib_mr *ret;
+
+	if (length == 0) {
+		ret = ERR_PTR(-EINVAL);
+		goto bail;
+	}
+
+	umem = ib_umem_get(pd->uobject->context, start, length,
+			   mr_access_flags, 0);
+	if (IS_ERR(umem))
+		return (void *) umem;
+
+	n = 0;
+	list_for_each_entry(chunk, &umem->chunk_list, list)
+		n += chunk->nents;
+
+	mr = alloc_mr(n, &to_idev(pd->device)->lk_table);
+	if (!mr) {
+		ret = ERR_PTR(-ENOMEM);
+		ib_umem_release(umem);
+		goto bail;
+	}
+
+	mr->mr.pd = pd;
+	mr->mr.user_base = start;
+	mr->mr.iova = virt_addr;
+	mr->mr.length = length;
+	mr->mr.offset = umem->offset;
+	mr->mr.access_flags = mr_access_flags;
+	mr->umem = umem;
+
+	m = 0;
+	n = 0;
+	list_for_each_entry(chunk, &umem->chunk_list, list) {
+		for (i = 0; i < chunk->nents; i++) {
+			void *vaddr;
+
+			vaddr = page_address(sg_page(&chunk->page_list[i]));
+			if (!vaddr) {
+				ret = ERR_PTR(-EINVAL);
+				goto bail;
+			}
+			mr->mr.map[m]->segs[n].vaddr = vaddr;
+			mr->mr.map[m]->segs[n].length = umem->page_size;
+			n++;
+			if (n == QIB_SEGSZ) {
+				m++;
+				n = 0;
+			}
+		}
+	}
+	ret = &mr->ibmr;
+
+bail:
+	return ret;
+}
+
+/**
+ * qib_dereg_mr - unregister and free a memory region
+ * @ibmr: the memory region to free
+ *
+ * Returns 0 on success.
+ *
+ * Note that this is called to free MRs created by qib_get_dma_mr()
+ * or qib_reg_user_mr().
+ */
+int qib_dereg_mr(struct ib_mr *ibmr)
+{
+	struct qib_mr *mr = to_imr(ibmr);
+	struct qib_ibdev *dev = to_idev(ibmr->device);
+	int ret;
+	int i;
+
+	ret = qib_free_lkey(dev, &mr->mr);
+	if (ret)
+		return ret;
+
+	i = mr->mr.mapsz;
+	while (i)
+		kfree(mr->mr.map[--i]);
+	if (mr->umem)
+		ib_umem_release(mr->umem);
+	kfree(mr);
+	return 0;
+}
+
+/*
+ * Allocate a memory region usable with the
+ * IB_WR_FAST_REG_MR send work request.
+ *
+ * Return the memory region on success, otherwise return an errno.
+ */
+struct ib_mr *qib_alloc_fast_reg_mr(struct ib_pd *pd, int max_page_list_len)
+{
+	struct qib_mr *mr;
+
+	mr = alloc_mr(max_page_list_len, &to_idev(pd->device)->lk_table);
+	if (mr == NULL)
+		return ERR_PTR(-ENOMEM);
+
+	mr->mr.pd = pd;
+	mr->mr.user_base = 0;
+	mr->mr.iova = 0;
+	mr->mr.length = 0;
+	mr->mr.offset = 0;
+	mr->mr.access_flags = 0;
+	mr->umem = NULL;
+
+	return &mr->ibmr;
+}
+
+struct ib_fast_reg_page_list *
+qib_alloc_fast_reg_page_list(struct ib_device *ibdev, int page_list_len)
+{
+	unsigned size = page_list_len * sizeof(u64);
+	struct ib_fast_reg_page_list *pl;
+
+	if (size > PAGE_SIZE)
+		return ERR_PTR(-EINVAL);
+
+	pl = kmalloc(sizeof *pl, GFP_KERNEL);
+	if (!pl)
+		return ERR_PTR(-ENOMEM);
+
+	pl->page_list = kmalloc(size, GFP_KERNEL);
+	if (!pl->page_list)
+		goto err_free;
+
+	return pl;
+
+err_free:
+	kfree(pl);
+	return ERR_PTR(-ENOMEM);
+}
+
+void qib_free_fast_reg_page_list(struct ib_fast_reg_page_list *pl)
+{
+	kfree(pl->page_list);
+	kfree(pl);
+}
+
+/**
+ * qib_alloc_fmr - allocate a fast memory region
+ * @pd: the protection domain for this memory region
+ * @mr_access_flags: access flags for this memory region
+ * @fmr_attr: fast memory region attributes
+ *
+ * Returns the memory region on success, otherwise returns an errno.
+ */
+struct ib_fmr *qib_alloc_fmr(struct ib_pd *pd, int mr_access_flags,
+			     struct ib_fmr_attr *fmr_attr)
+{
+	struct qib_fmr *fmr;
+	int m, i = 0;
+	struct ib_fmr *ret;
+
+	/* Allocate struct plus pointers to first level page tables. */
+	m = (fmr_attr->max_pages + QIB_SEGSZ - 1) / QIB_SEGSZ;
+	fmr = kmalloc(sizeof *fmr + m * sizeof fmr->mr.map[0], GFP_KERNEL);
+	if (!fmr)
+		goto bail;
+
+	/* Allocate first level page tables. */
+	for (; i < m; i++) {
+		fmr->mr.map[i] = kmalloc(sizeof *fmr->mr.map[0],
+					 GFP_KERNEL);
+		if (!fmr->mr.map[i])
+			goto bail;
+	}
+	fmr->mr.mapsz = m;
+
+	/*
+	 * ib_alloc_fmr() will initialize fmr->ibfmr except for lkey &
+	 * rkey.
+	 */
+	if (!qib_alloc_lkey(&to_idev(pd->device)->lk_table, &fmr->mr))
+		goto bail;
+	fmr->ibfmr.rkey = fmr->mr.lkey;
+	fmr->ibfmr.lkey = fmr->mr.lkey;
+	/*
+	 * Resources are allocated but no valid mapping (RKEY can't be
+	 * used).
+	 */
+	fmr->mr.pd = pd;
+	fmr->mr.user_base = 0;
+	fmr->mr.iova = 0;
+	fmr->mr.length = 0;
+	fmr->mr.offset = 0;
+	fmr->mr.access_flags = mr_access_flags;
+	fmr->mr.max_segs = fmr_attr->max_pages;
+	fmr->page_shift = fmr_attr->page_shift;
+
+	atomic_set(&fmr->mr.refcount, 0);
+	ret = &fmr->ibfmr;
+	goto done;
+
+bail:
+	while (i)
+		kfree(fmr->mr.map[--i]);
+	kfree(fmr);
+	ret = ERR_PTR(-ENOMEM);
+
+done:
+	return ret;
+}
+
+/**
+ * qib_map_phys_fmr - set up a fast memory region
+ * @ibmfr: the fast memory region to set up
+ * @page_list: the list of pages to associate with the fast memory region
+ * @list_len: the number of pages to associate with the fast memory region
+ * @iova: the virtual address of the start of the fast memory region
+ *
+ * This may be called from interrupt context.
+ */
+
+int qib_map_phys_fmr(struct ib_fmr *ibfmr, u64 *page_list,
+		     int list_len, u64 iova)
+{
+	struct qib_fmr *fmr = to_ifmr(ibfmr);
+	struct qib_lkey_table *rkt;
+	unsigned long flags;
+	int m, n, i;
+	u32 ps;
+	int ret;
+
+	if (atomic_read(&fmr->mr.refcount))
+		return -EBUSY;
+
+	if (list_len > fmr->mr.max_segs) {
+		ret = -EINVAL;
+		goto bail;
+	}
+	rkt = &to_idev(ibfmr->device)->lk_table;
+	spin_lock_irqsave(&rkt->lock, flags);
+	fmr->mr.user_base = iova;
+	fmr->mr.iova = iova;
+	ps = 1 << fmr->page_shift;
+	fmr->mr.length = list_len * ps;
+	m = 0;
+	n = 0;
+	for (i = 0; i < list_len; i++) {
+		fmr->mr.map[m]->segs[n].vaddr = (void *) page_list[i];
+		fmr->mr.map[m]->segs[n].length = ps;
+		if (++n == QIB_SEGSZ) {
+			m++;
+			n = 0;
+		}
+	}
+	spin_unlock_irqrestore(&rkt->lock, flags);
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+/**
+ * qib_unmap_fmr - unmap fast memory regions
+ * @fmr_list: the list of fast memory regions to unmap
+ *
+ * Returns 0 on success.
+ */
+int qib_unmap_fmr(struct list_head *fmr_list)
+{
+	struct qib_fmr *fmr;
+	struct qib_lkey_table *rkt;
+	unsigned long flags;
+
+	list_for_each_entry(fmr, fmr_list, ibfmr.list) {
+		rkt = &to_idev(fmr->ibfmr.device)->lk_table;
+		spin_lock_irqsave(&rkt->lock, flags);
+		fmr->mr.user_base = 0;
+		fmr->mr.iova = 0;
+		fmr->mr.length = 0;
+		spin_unlock_irqrestore(&rkt->lock, flags);
+	}
+	return 0;
+}
+
+/**
+ * qib_dealloc_fmr - deallocate a fast memory region
+ * @ibfmr: the fast memory region to deallocate
+ *
+ * Returns 0 on success.
+ */
+int qib_dealloc_fmr(struct ib_fmr *ibfmr)
+{
+	struct qib_fmr *fmr = to_ifmr(ibfmr);
+	int ret;
+	int i;
+
+	ret = qib_free_lkey(to_idev(ibfmr->device), &fmr->mr);
+	if (ret)
+		return ret;
+
+	i = fmr->mr.mapsz;
+	while (i)
+		kfree(fmr->mr.map[--i]);
+	kfree(fmr);
+	return 0;
+}

drivers/infiniband/hw/qib/qib_pcie.c

@@ -0,0 +1,738 @@
+/*
+ * Copyright (c) 2008, 2009 QLogic Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/pci.h>
+#include <linux/io.h>
+#include <linux/delay.h>
+#include <linux/vmalloc.h>
+#include <linux/aer.h>
+
+#include "qib.h"
+
+/*
+ * This file contains PCIe utility routines that are common to the
+ * various QLogic InfiniPath adapters
+ */
+
+/*
+ * Code to adjust PCIe capabilities.
+ * To minimize the change footprint, we call it
+ * from qib_pcie_params, which every chip-specific
+ * file calls, even though this violates some
+ * expectations of harmlessness.
+ */
+static int qib_tune_pcie_caps(struct qib_devdata *);
+static int qib_tune_pcie_coalesce(struct qib_devdata *);
+
+/*
+ * Do all the common PCIe setup and initialization.
+ * devdata is not yet allocated, and is not allocated until after this
+ * routine returns success.  Therefore qib_dev_err() can't be used for error
+ * printing.
+ */
+int qib_pcie_init(struct pci_dev *pdev, const struct pci_device_id *ent)
+{
+	int ret;
+
+	ret = pci_enable_device(pdev);
+	if (ret) {
+		/*
+		 * This can happen (in theory) iff:
+		 * We did a chip reset, and then failed to reprogram the
+		 * BAR, or the chip reset due to an internal error.  We then
+		 * unloaded the driver and reloaded it.
+		 *
+		 * Both reset cases set the BAR back to initial state.  For
+		 * the latter case, the AER sticky error bit at offset 0x718
+		 * should be set, but the Linux kernel doesn't yet know
+		 * about that, it appears.  If the original BAR was retained
+		 * in the kernel data structures, this may be OK.
+		 */
+		qib_early_err(&pdev->dev, "pci enable failed: error %d\n",
+			      -ret);
+		goto done;
+	}
+
+	ret = pci_request_regions(pdev, QIB_DRV_NAME);
+	if (ret) {
+		qib_devinfo(pdev, "pci_request_regions fails: err %d\n", -ret);
+		goto bail;
+	}
+
+	ret = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
+	if (ret) {
+		/*
+		 * If the 64 bit setup fails, try 32 bit.  Some systems
+		 * do not setup 64 bit maps on systems with 2GB or less
+		 * memory installed.
+		 */
+		ret = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
+		if (ret) {
+			qib_devinfo(pdev, "Unable to set DMA mask: %d\n", ret);
+			goto bail;
+		}
+		ret = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
+	} else
+		ret = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
+	if (ret)
+		qib_early_err(&pdev->dev,
+			      "Unable to set DMA consistent mask: %d\n", ret);
+
+	pci_set_master(pdev);
+	ret = pci_enable_pcie_error_reporting(pdev);
+	if (ret)
+		qib_early_err(&pdev->dev,
+			      "Unable to enable pcie error reporting: %d\n",
+			      ret);
+	goto done;
+
+bail:
+	pci_disable_device(pdev);
+	pci_release_regions(pdev);
+done:
+	return ret;
+}
+
+/*
+ * Do remaining PCIe setup, once dd is allocated, and save away
+ * fields required to re-initialize after a chip reset, or for
+ * various other purposes
+ */
+int qib_pcie_ddinit(struct qib_devdata *dd, struct pci_dev *pdev,
+		    const struct pci_device_id *ent)
+{
+	unsigned long len;
+	resource_size_t addr;
+
+	dd->pcidev = pdev;
+	pci_set_drvdata(pdev, dd);
+
+	addr = pci_resource_start(pdev, 0);
+	len = pci_resource_len(pdev, 0);
+
+#if defined(__powerpc__)
+	/* There isn't a generic way to specify writethrough mappings */
+	dd->kregbase = __ioremap(addr, len, _PAGE_NO_CACHE | _PAGE_WRITETHRU);
+#else
+	dd->kregbase = ioremap_nocache(addr, len);
+#endif
+
+	if (!dd->kregbase)
+		return -ENOMEM;
+
+	dd->kregend = (u64 __iomem *)((void __iomem *) dd->kregbase + len);
+	dd->physaddr = addr;        /* used for io_remap, etc. */
+
+	/*
+	 * Save BARs to rewrite after device reset.  Save all 64 bits of
+	 * BAR, just in case.
+	 */
+	dd->pcibar0 = addr;
+	dd->pcibar1 = addr >> 32;
+	dd->deviceid = ent->device; /* save for later use */
+	dd->vendorid = ent->vendor;
+
+	return 0;
+}
+
+/*
+ * Do PCIe cleanup, after chip-specific cleanup, etc.  Just prior
+ * to releasing the dd memory.
+ * void because none of the core pcie cleanup returns are void
+ */
+void qib_pcie_ddcleanup(struct qib_devdata *dd)
+{
+	u64 __iomem *base = (void __iomem *) dd->kregbase;
+
+	dd->kregbase = NULL;
+	iounmap(base);
+	if (dd->piobase)
+		iounmap(dd->piobase);
+	if (dd->userbase)
+		iounmap(dd->userbase);
+
+	pci_disable_device(dd->pcidev);
+	pci_release_regions(dd->pcidev);
+
+	pci_set_drvdata(dd->pcidev, NULL);
+}
+
+static void qib_msix_setup(struct qib_devdata *dd, int pos, u32 *msixcnt,
+			   struct msix_entry *msix_entry)
+{
+	int ret;
+	u32 tabsize = 0;
+	u16 msix_flags;
+
+	pci_read_config_word(dd->pcidev, pos + PCI_MSIX_FLAGS, &msix_flags);
+	tabsize = 1 + (msix_flags & PCI_MSIX_FLAGS_QSIZE);
+	if (tabsize > *msixcnt)
+		tabsize = *msixcnt;
+	ret = pci_enable_msix(dd->pcidev, msix_entry, tabsize);
+	if (ret > 0) {
+		tabsize = ret;
+		ret = pci_enable_msix(dd->pcidev, msix_entry, tabsize);
+	}
+	if (ret) {
+		qib_dev_err(dd, "pci_enable_msix %d vectors failed: %d, "
+			    "falling back to INTx\n", tabsize, ret);
+		tabsize = 0;
+	}
+	*msixcnt = tabsize;
+
+	if (ret)
+		qib_enable_intx(dd->pcidev);
+
+}
+
+/**
+ * We save the msi lo and hi values, so we can restore them after
+ * chip reset (the kernel PCI infrastructure doesn't yet handle that
+ * correctly.
+ */
+static int qib_msi_setup(struct qib_devdata *dd, int pos)
+{
+	struct pci_dev *pdev = dd->pcidev;
+	u16 control;
+	int ret;
+
+	ret = pci_enable_msi(pdev);
+	if (ret)
+		qib_dev_err(dd, "pci_enable_msi failed: %d, "
+			    "interrupts may not work\n", ret);
+	/* continue even if it fails, we may still be OK... */
+
+	pci_read_config_dword(pdev, pos + PCI_MSI_ADDRESS_LO,
+			      &dd->msi_lo);
+	pci_read_config_dword(pdev, pos + PCI_MSI_ADDRESS_HI,
+			      &dd->msi_hi);
+	pci_read_config_word(pdev, pos + PCI_MSI_FLAGS, &control);
+	/* now save the data (vector) info */
+	pci_read_config_word(pdev, pos + ((control & PCI_MSI_FLAGS_64BIT)
+				    ? 12 : 8),
+			     &dd->msi_data);
+	return ret;
+}
+
+int qib_pcie_params(struct qib_devdata *dd, u32 minw, u32 *nent,
+		    struct msix_entry *entry)
+{
+	u16 linkstat, speed;
+	int pos = 0, pose, ret = 1;
+
+	pose = pci_find_capability(dd->pcidev, PCI_CAP_ID_EXP);
+	if (!pose) {
+		qib_dev_err(dd, "Can't find PCI Express capability!\n");
+		/* set up something... */
+		dd->lbus_width = 1;
+		dd->lbus_speed = 2500; /* Gen1, 2.5GHz */
+		goto bail;
+	}
+
+	pos = pci_find_capability(dd->pcidev, PCI_CAP_ID_MSIX);
+	if (nent && *nent && pos) {
+		qib_msix_setup(dd, pos, nent, entry);
+		ret = 0; /* did it, either MSIx or INTx */
+	} else {
+		pos = pci_find_capability(dd->pcidev, PCI_CAP_ID_MSI);
+		if (pos)
+			ret = qib_msi_setup(dd, pos);
+		else
+			qib_dev_err(dd, "No PCI MSI or MSIx capability!\n");
+	}
+	if (!pos)
+		qib_enable_intx(dd->pcidev);
+
+	pci_read_config_word(dd->pcidev, pose + PCI_EXP_LNKSTA, &linkstat);
+	/*
+	 * speed is bits 0-3, linkwidth is bits 4-8
+	 * no defines for them in headers
+	 */
+	speed = linkstat & 0xf;
+	linkstat >>= 4;
+	linkstat &= 0x1f;
+	dd->lbus_width = linkstat;
+
+	switch (speed) {
+	case 1:
+		dd->lbus_speed = 2500; /* Gen1, 2.5GHz */
+		break;
+	case 2:
+		dd->lbus_speed = 5000; /* Gen1, 5GHz */
+		break;
+	default: /* not defined, assume gen1 */
+		dd->lbus_speed = 2500;
+		break;
+	}
+
+	/*
+	 * Check against expected pcie width and complain if "wrong"
+	 * on first initialization, not afterwards (i.e., reset).
+	 */
+	if (minw && linkstat < minw)
+		qib_dev_err(dd,
+			    "PCIe width %u (x%u HCA), performance reduced\n",
+			    linkstat, minw);
+
+	qib_tune_pcie_caps(dd);
+
+	qib_tune_pcie_coalesce(dd);
+
+bail:
+	/* fill in string, even on errors */
+	snprintf(dd->lbus_info, sizeof(dd->lbus_info),
+		 "PCIe,%uMHz,x%u\n", dd->lbus_speed, dd->lbus_width);
+	return ret;
+}
+
+/*
+ * Setup pcie interrupt stuff again after a reset.  I'd like to just call
+ * pci_enable_msi() again for msi, but when I do that,
+ * the MSI enable bit doesn't get set in the command word, and
+ * we switch to to a different interrupt vector, which is confusing,
+ * so I instead just do it all inline.  Perhaps somehow can tie this
+ * into the PCIe hotplug support at some point
+ */
+int qib_reinit_intr(struct qib_devdata *dd)
+{
+	int pos;
+	u16 control;
+	int ret = 0;
+
+	/* If we aren't using MSI, don't restore it */
+	if (!dd->msi_lo)
+		goto bail;
+
+	pos = pci_find_capability(dd->pcidev, PCI_CAP_ID_MSI);
+	if (!pos) {
+		qib_dev_err(dd, "Can't find MSI capability, "
+			    "can't restore MSI settings\n");
+		ret = 0;
+		/* nothing special for MSIx, just MSI */
+		goto bail;
+	}
+	pci_write_config_dword(dd->pcidev, pos + PCI_MSI_ADDRESS_LO,
+			       dd->msi_lo);
+	pci_write_config_dword(dd->pcidev, pos + PCI_MSI_ADDRESS_HI,
+			       dd->msi_hi);
+	pci_read_config_word(dd->pcidev, pos + PCI_MSI_FLAGS, &control);
+	if (!(control & PCI_MSI_FLAGS_ENABLE)) {
+		control |= PCI_MSI_FLAGS_ENABLE;
+		pci_write_config_word(dd->pcidev, pos + PCI_MSI_FLAGS,
+				      control);
+	}
+	/* now rewrite the data (vector) info */
+	pci_write_config_word(dd->pcidev, pos +
+			      ((control & PCI_MSI_FLAGS_64BIT) ? 12 : 8),
+			      dd->msi_data);
+	ret = 1;
+bail:
+	if (!ret && (dd->flags & QIB_HAS_INTX)) {
+		qib_enable_intx(dd->pcidev);
+		ret = 1;
+	}
+
+	/* and now set the pci master bit again */
+	pci_set_master(dd->pcidev);
+
+	return ret;
+}
+
+/*
+ * Disable msi interrupt if enabled, and clear msi_lo.
+ * This is used primarily for the fallback to INTx, but
+ * is also used in reinit after reset, and during cleanup.
+ */
+void qib_nomsi(struct qib_devdata *dd)
+{
+	dd->msi_lo = 0;
+	pci_disable_msi(dd->pcidev);
+}
+
+/*
+ * Same as qib_nosmi, but for MSIx.
+ */
+void qib_nomsix(struct qib_devdata *dd)
+{
+	pci_disable_msix(dd->pcidev);
+}
+
+/*
+ * Similar to pci_intx(pdev, 1), except that we make sure
+ * msi(x) is off.
+ */
+void qib_enable_intx(struct pci_dev *pdev)
+{
+	u16 cw, new;
+	int pos;
+
+	/* first, turn on INTx */
+	pci_read_config_word(pdev, PCI_COMMAND, &cw);
+	new = cw & ~PCI_COMMAND_INTX_DISABLE;
+	if (new != cw)
+		pci_write_config_word(pdev, PCI_COMMAND, new);
+
+	pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
+	if (pos) {
+		/* then turn off MSI */
+		pci_read_config_word(pdev, pos + PCI_MSI_FLAGS, &cw);
+		new = cw & ~PCI_MSI_FLAGS_ENABLE;
+		if (new != cw)
+			pci_write_config_word(pdev, pos + PCI_MSI_FLAGS, new);
+	}
+	pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
+	if (pos) {
+		/* then turn off MSIx */
+		pci_read_config_word(pdev, pos + PCI_MSIX_FLAGS, &cw);
+		new = cw & ~PCI_MSIX_FLAGS_ENABLE;
+		if (new != cw)
+			pci_write_config_word(pdev, pos + PCI_MSIX_FLAGS, new);
+	}
+}
+
+/*
+ * These two routines are helper routines for the device reset code
+ * to move all the pcie code out of the chip-specific driver code.
+ */
+void qib_pcie_getcmd(struct qib_devdata *dd, u16 *cmd, u8 *iline, u8 *cline)
+{
+	pci_read_config_word(dd->pcidev, PCI_COMMAND, cmd);
+	pci_read_config_byte(dd->pcidev, PCI_INTERRUPT_LINE, iline);
+	pci_read_config_byte(dd->pcidev, PCI_CACHE_LINE_SIZE, cline);
+}
+
+void qib_pcie_reenable(struct qib_devdata *dd, u16 cmd, u8 iline, u8 cline)
+{
+	int r;
+	r = pci_write_config_dword(dd->pcidev, PCI_BASE_ADDRESS_0,
+				   dd->pcibar0);
+	if (r)
+		qib_dev_err(dd, "rewrite of BAR0 failed: %d\n", r);
+	r = pci_write_config_dword(dd->pcidev, PCI_BASE_ADDRESS_1,
+				   dd->pcibar1);
+	if (r)
+		qib_dev_err(dd, "rewrite of BAR1 failed: %d\n", r);
+	/* now re-enable memory access, and restore cosmetic settings */
+	pci_write_config_word(dd->pcidev, PCI_COMMAND, cmd);
+	pci_write_config_byte(dd->pcidev, PCI_INTERRUPT_LINE, iline);
+	pci_write_config_byte(dd->pcidev, PCI_CACHE_LINE_SIZE, cline);
+	r = pci_enable_device(dd->pcidev);
+	if (r)
+		qib_dev_err(dd, "pci_enable_device failed after "
+			    "reset: %d\n", r);
+}
+
+/* code to adjust PCIe capabilities. */
+
+static int fld2val(int wd, int mask)
+{
+	int lsbmask;
+
+	if (!mask)
+		return 0;
+	wd &= mask;
+	lsbmask = mask ^ (mask & (mask - 1));
+	wd /= lsbmask;
+	return wd;
+}
+
+static int val2fld(int wd, int mask)
+{
+	int lsbmask;
+
+	if (!mask)
+		return 0;
+	lsbmask = mask ^ (mask & (mask - 1));
+	wd *= lsbmask;
+	return wd;
+}
+
+static int qib_pcie_coalesce;
+module_param_named(pcie_coalesce, qib_pcie_coalesce, int, S_IRUGO);
+MODULE_PARM_DESC(pcie_coalesce, "tune PCIe colescing on some Intel chipsets");
+
+/*
+ * Enable PCIe completion and data coalescing, on Intel 5x00 and 7300
+ * chipsets.   This is known to be unsafe for some revisions of some
+ * of these chipsets, with some BIOS settings, and enabling it on those
+ * systems may result in the system crashing, and/or data corruption.
+ */
+static int qib_tune_pcie_coalesce(struct qib_devdata *dd)
+{
+	int r;
+	struct pci_dev *parent;
+	int ppos;
+	u16 devid;
+	u32 mask, bits, val;
+
+	if (!qib_pcie_coalesce)
+		return 0;
+
+	/* Find out supported and configured values for parent (root) */
+	parent = dd->pcidev->bus->self;
+	if (parent->bus->parent) {
+		qib_devinfo(dd->pcidev, "Parent not root\n");
+		return 1;
+	}
+	ppos = pci_find_capability(parent, PCI_CAP_ID_EXP);
+	if (!ppos)
+		return 1;
+	if (parent->vendor != 0x8086)
+		return 1;
+
+	/*
+	 *  - bit 12: Max_rdcmp_Imt_EN: need to set to 1
+	 *  - bit 11: COALESCE_FORCE: need to set to 0
+	 *  - bit 10: COALESCE_EN: need to set to 1
+	 *  (but limitations on some on some chipsets)
+	 *
+	 *  On the Intel 5000, 5100, and 7300 chipsets, there is
+	 *  also: - bit 25:24: COALESCE_MODE, need to set to 0
+	 */
+	devid = parent->device;
+	if (devid >= 0x25e2 && devid <= 0x25fa) {
+		u8 rev;
+
+		/* 5000 P/V/X/Z */
+		pci_read_config_byte(parent, PCI_REVISION_ID, &rev);
+		if (rev <= 0xb2)
+			bits = 1U << 10;
+		else
+			bits = 7U << 10;
+		mask = (3U << 24) | (7U << 10);
+	} else if (devid >= 0x65e2 && devid <= 0x65fa) {
+		/* 5100 */
+		bits = 1U << 10;
+		mask = (3U << 24) | (7U << 10);
+	} else if (devid >= 0x4021 && devid <= 0x402e) {
+		/* 5400 */
+		bits = 7U << 10;
+		mask = 7U << 10;
+	} else if (devid >= 0x3604 && devid <= 0x360a) {
+		/* 7300 */
+		bits = 7U << 10;
+		mask = (3U << 24) | (7U << 10);
+	} else {
+		/* not one of the chipsets that we know about */
+		return 1;
+	}
+	pci_read_config_dword(parent, 0x48, &val);
+	val &= ~mask;
+	val |= bits;
+	r = pci_write_config_dword(parent, 0x48, val);
+	return 0;
+}
+
+/*
+ * BIOS may not set PCIe bus-utilization parameters for best performance.
+ * Check and optionally adjust them to maximize our throughput.
+ */
+static int qib_pcie_caps;
+module_param_named(pcie_caps, qib_pcie_caps, int, S_IRUGO);
+MODULE_PARM_DESC(pcie_caps, "Max PCIe tuning: Payload (4lsb), ReadReq (D4..7)");
+
+static int qib_tune_pcie_caps(struct qib_devdata *dd)
+{
+	int ret = 1; /* Assume the worst */
+	struct pci_dev *parent;
+	int ppos, epos;
+	u16 pcaps, pctl, ecaps, ectl;
+	int rc_sup, ep_sup;
+	int rc_cur, ep_cur;
+
+	/* Find out supported and configured values for parent (root) */
+	parent = dd->pcidev->bus->self;
+	if (parent->bus->parent) {
+		qib_devinfo(dd->pcidev, "Parent not root\n");
+		goto bail;
+	}
+	ppos = pci_find_capability(parent, PCI_CAP_ID_EXP);
+	if (ppos) {
+		pci_read_config_word(parent, ppos + PCI_EXP_DEVCAP, &pcaps);
+		pci_read_config_word(parent, ppos + PCI_EXP_DEVCTL, &pctl);
+	} else
+		goto bail;
+	/* Find out supported and configured values for endpoint (us) */
+	epos = pci_find_capability(dd->pcidev, PCI_CAP_ID_EXP);
+	if (epos) {
+		pci_read_config_word(dd->pcidev, epos + PCI_EXP_DEVCAP, &ecaps);
+		pci_read_config_word(dd->pcidev, epos + PCI_EXP_DEVCTL, &ectl);
+	} else
+		goto bail;
+	ret = 0;
+	/* Find max payload supported by root, endpoint */
+	rc_sup = fld2val(pcaps, PCI_EXP_DEVCAP_PAYLOAD);
+	ep_sup = fld2val(ecaps, PCI_EXP_DEVCAP_PAYLOAD);
+	if (rc_sup > ep_sup)
+		rc_sup = ep_sup;
+
+	rc_cur = fld2val(pctl, PCI_EXP_DEVCTL_PAYLOAD);
+	ep_cur = fld2val(ectl, PCI_EXP_DEVCTL_PAYLOAD);
+
+	/* If Supported greater than limit in module param, limit it */
+	if (rc_sup > (qib_pcie_caps & 7))
+		rc_sup = qib_pcie_caps & 7;
+	/* If less than (allowed, supported), bump root payload */
+	if (rc_sup > rc_cur) {
+		rc_cur = rc_sup;
+		pctl = (pctl & ~PCI_EXP_DEVCTL_PAYLOAD) |
+			val2fld(rc_cur, PCI_EXP_DEVCTL_PAYLOAD);
+		pci_write_config_word(parent, ppos + PCI_EXP_DEVCTL, pctl);
+	}
+	/* If less than (allowed, supported), bump endpoint payload */
+	if (rc_sup > ep_cur) {
+		ep_cur = rc_sup;
+		ectl = (ectl & ~PCI_EXP_DEVCTL_PAYLOAD) |
+			val2fld(ep_cur, PCI_EXP_DEVCTL_PAYLOAD);
+		pci_write_config_word(dd->pcidev, epos + PCI_EXP_DEVCTL, ectl);
+	}
+
+	/*
+	 * Now the Read Request size.
+	 * No field for max supported, but PCIe spec limits it to 4096,
+	 * which is code '5' (log2(4096) - 7)
+	 */
+	rc_sup = 5;
+	if (rc_sup > ((qib_pcie_caps >> 4) & 7))
+		rc_sup = (qib_pcie_caps >> 4) & 7;
+	rc_cur = fld2val(pctl, PCI_EXP_DEVCTL_READRQ);
+	ep_cur = fld2val(ectl, PCI_EXP_DEVCTL_READRQ);
+
+	if (rc_sup > rc_cur) {
+		rc_cur = rc_sup;
+		pctl = (pctl & ~PCI_EXP_DEVCTL_READRQ) |
+			val2fld(rc_cur, PCI_EXP_DEVCTL_READRQ);
+		pci_write_config_word(parent, ppos + PCI_EXP_DEVCTL, pctl);
+	}
+	if (rc_sup > ep_cur) {
+		ep_cur = rc_sup;
+		ectl = (ectl & ~PCI_EXP_DEVCTL_READRQ) |
+			val2fld(ep_cur, PCI_EXP_DEVCTL_READRQ);
+		pci_write_config_word(dd->pcidev, epos + PCI_EXP_DEVCTL, ectl);
+	}
+bail:
+	return ret;
+}
+/* End of PCIe capability tuning */
+
+/*
+ * From here through qib_pci_err_handler definition is invoked via
+ * PCI error infrastructure, registered via pci
+ */
+static pci_ers_result_t
+qib_pci_error_detected(struct pci_dev *pdev, pci_channel_state_t state)
+{
+	struct qib_devdata *dd = pci_get_drvdata(pdev);
+	pci_ers_result_t ret = PCI_ERS_RESULT_RECOVERED;
+
+	switch (state) {
+	case pci_channel_io_normal:
+		qib_devinfo(pdev, "State Normal, ignoring\n");
+		break;
+
+	case pci_channel_io_frozen:
+		qib_devinfo(pdev, "State Frozen, requesting reset\n");
+		pci_disable_device(pdev);
+		ret = PCI_ERS_RESULT_NEED_RESET;
+		break;
+
+	case pci_channel_io_perm_failure:
+		qib_devinfo(pdev, "State Permanent Failure, disabling\n");
+		if (dd) {
+			/* no more register accesses! */
+			dd->flags &= ~QIB_PRESENT;
+			qib_disable_after_error(dd);
+		}
+		 /* else early, or other problem */
+		ret =  PCI_ERS_RESULT_DISCONNECT;
+		break;
+
+	default: /* shouldn't happen */
+		qib_devinfo(pdev, "QIB PCI errors detected (state %d)\n",
+			state);
+		break;
+	}
+	return ret;
+}
+
+static pci_ers_result_t
+qib_pci_mmio_enabled(struct pci_dev *pdev)
+{
+	u64 words = 0U;
+	struct qib_devdata *dd = pci_get_drvdata(pdev);
+	pci_ers_result_t ret = PCI_ERS_RESULT_RECOVERED;
+
+	if (dd && dd->pport) {
+		words = dd->f_portcntr(dd->pport, QIBPORTCNTR_WORDRCV);
+		if (words == ~0ULL)
+			ret = PCI_ERS_RESULT_NEED_RESET;
+	}
+	qib_devinfo(pdev, "QIB mmio_enabled function called, "
+		 "read wordscntr %Lx, returning %d\n", words, ret);
+	return  ret;
+}
+
+static pci_ers_result_t
+qib_pci_slot_reset(struct pci_dev *pdev)
+{
+	qib_devinfo(pdev, "QIB link_reset function called, ignored\n");
+	return PCI_ERS_RESULT_CAN_RECOVER;
+}
+
+static pci_ers_result_t
+qib_pci_link_reset(struct pci_dev *pdev)
+{
+	qib_devinfo(pdev, "QIB link_reset function called, ignored\n");
+	return PCI_ERS_RESULT_CAN_RECOVER;
+}
+
+static void
+qib_pci_resume(struct pci_dev *pdev)
+{
+	struct qib_devdata *dd = pci_get_drvdata(pdev);
+	qib_devinfo(pdev, "QIB resume function called\n");
+	pci_cleanup_aer_uncorrect_error_status(pdev);
+	/*
+	 * Running jobs will fail, since it's asynchronous
+	 * unlike sysfs-requested reset.   Better than
+	 * doing nothing.
+	 */
+	qib_init(dd, 1); /* same as re-init after reset */
+}
+
+struct pci_error_handlers qib_pci_err_handler = {
+	.error_detected = qib_pci_error_detected,
+	.mmio_enabled = qib_pci_mmio_enabled,
+	.link_reset = qib_pci_link_reset,
+	.slot_reset = qib_pci_slot_reset,
+	.resume = qib_pci_resume,
+};

drivers/infiniband/hw/qib/qib_pio_copy.c

@@ -0,0 +1,64 @@
+/*
+ * Copyright (c) 2009 QLogic Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include "qib.h"
+
+/**
+ * qib_pio_copy - copy data to MMIO space, in multiples of 32-bits
+ * @to: destination, in MMIO space (must be 64-bit aligned)
+ * @from: source (must be 64-bit aligned)
+ * @count: number of 32-bit quantities to copy
+ *
+ * Copy data from kernel space to MMIO space, in multiples of 32 bits at a
+ * time.  Order of access is not guaranteed, nor is a memory barrier
+ * performed afterwards.
+ */
+void qib_pio_copy(void __iomem *to, const void *from, size_t count)
+{
+#ifdef CONFIG_64BIT
+	u64 __iomem *dst = to;
+	const u64 *src = from;
+	const u64 *end = src + (count >> 1);
+
+	while (src < end)
+		__raw_writeq(*src++, dst++);
+	if (count & 1)
+		__raw_writel(*(const u32 *)src, dst);
+#else
+	u32 __iomem *dst = to;
+	const u32 *src = from;
+	const u32 *end = src + count;
+
+	while (src < end)
+		__raw_writel(*src++, dst++);
+#endif
+}

drivers/infiniband/hw/qib/qib_qsfp.c

@@ -0,0 +1,564 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/delay.h>
+#include <linux/pci.h>
+#include <linux/vmalloc.h>
+
+#include "qib.h"
+#include "qib_qsfp.h"
+
+/*
+ * QSFP support for ib_qib driver, using "Two Wire Serial Interface" driver
+ * in qib_twsi.c
+ */
+#define QSFP_MAX_RETRY 4
+
+static int qsfp_read(struct qib_pportdata *ppd, int addr, void *bp, int len)
+{
+	struct qib_devdata *dd = ppd->dd;
+	u32 out, mask;
+	int ret, cnt, pass = 0;
+	int stuck = 0;
+	u8 *buff = bp;
+
+	ret = mutex_lock_interruptible(&dd->eep_lock);
+	if (ret)
+		goto no_unlock;
+
+	if (dd->twsi_eeprom_dev == QIB_TWSI_NO_DEV) {
+		ret = -ENXIO;
+		goto bail;
+	}
+
+	/*
+	 * We presume, if we are called at all, that this board has
+	 * QSFP. This is on the same i2c chain as the legacy parts,
+	 * but only responds if the module is selected via GPIO pins.
+	 * Further, there are very long setup and hold requirements
+	 * on MODSEL.
+	 */
+	mask = QSFP_GPIO_MOD_SEL_N | QSFP_GPIO_MOD_RST_N | QSFP_GPIO_LP_MODE;
+	out = QSFP_GPIO_MOD_RST_N | QSFP_GPIO_LP_MODE;
+	if (ppd->hw_pidx) {
+		mask <<= QSFP_GPIO_PORT2_SHIFT;
+		out <<= QSFP_GPIO_PORT2_SHIFT;
+	}
+
+	dd->f_gpio_mod(dd, out, mask, mask);
+
+	/*
+	 * Module could take up to 2 Msec to respond to MOD_SEL, and there
+	 * is no way to tell if it is ready, so we must wait.
+	 */
+	msleep(2);
+
+	/* Make sure TWSI bus is in sane state. */
+	ret = qib_twsi_reset(dd);
+	if (ret) {
+		qib_dev_porterr(dd, ppd->port,
+				"QSFP interface Reset for read failed\n");
+		ret = -EIO;
+		stuck = 1;
+		goto deselect;
+	}
+
+	/* All QSFP modules are at A0 */
+
+	cnt = 0;
+	while (cnt < len) {
+		unsigned in_page;
+		int wlen = len - cnt;
+		in_page = addr % QSFP_PAGESIZE;
+		if ((in_page + wlen) > QSFP_PAGESIZE)
+			wlen = QSFP_PAGESIZE - in_page;
+		ret = qib_twsi_blk_rd(dd, QSFP_DEV, addr, buff + cnt, wlen);
+		/* Some QSFP's fail first try. Retry as experiment */
+		if (ret && cnt == 0 && ++pass < QSFP_MAX_RETRY)
+			continue;
+		if (ret) {
+			/* qib_twsi_blk_rd() 1 for error, else 0 */
+			ret = -EIO;
+			goto deselect;
+		}
+		addr += wlen;
+		cnt += wlen;
+	}
+	ret = cnt;
+
+deselect:
+	/*
+	 * Module could take up to 10 uSec after transfer before
+	 * ready to respond to MOD_SEL negation, and there is no way
+	 * to tell if it is ready, so we must wait.
+	 */
+	udelay(10);
+	/* set QSFP MODSEL, RST. LP all high */
+	dd->f_gpio_mod(dd, mask, mask, mask);
+
+	/*
+	 * Module could take up to 2 Msec to respond to MOD_SEL
+	 * going away, and there is no way to tell if it is ready.
+	 * so we must wait.
+	 */
+	if (stuck)
+		qib_dev_err(dd, "QSFP interface bus stuck non-idle\n");
+
+	if (pass >= QSFP_MAX_RETRY && ret)
+		qib_dev_porterr(dd, ppd->port, "QSFP failed even retrying\n");
+	else if (pass)
+		qib_dev_porterr(dd, ppd->port, "QSFP retries: %d\n", pass);
+
+	msleep(2);
+
+bail:
+	mutex_unlock(&dd->eep_lock);
+
+no_unlock:
+	return ret;
+}
+
+/*
+ * qsfp_write
+ * We do not ordinarily write the QSFP, but this is needed to select
+ * the page on non-flat QSFPs, and possibly later unusual cases
+ */
+static int qib_qsfp_write(struct qib_pportdata *ppd, int addr, void *bp,
+			  int len)
+{
+	struct qib_devdata *dd = ppd->dd;
+	u32 out, mask;
+	int ret, cnt;
+	u8 *buff = bp;
+
+	ret = mutex_lock_interruptible(&dd->eep_lock);
+	if (ret)
+		goto no_unlock;
+
+	if (dd->twsi_eeprom_dev == QIB_TWSI_NO_DEV) {
+		ret = -ENXIO;
+		goto bail;
+	}
+
+	/*
+	 * We presume, if we are called at all, that this board has
+	 * QSFP. This is on the same i2c chain as the legacy parts,
+	 * but only responds if the module is selected via GPIO pins.
+	 * Further, there are very long setup and hold requirements
+	 * on MODSEL.
+	 */
+	mask = QSFP_GPIO_MOD_SEL_N | QSFP_GPIO_MOD_RST_N | QSFP_GPIO_LP_MODE;
+	out = QSFP_GPIO_MOD_RST_N | QSFP_GPIO_LP_MODE;
+	if (ppd->hw_pidx) {
+		mask <<= QSFP_GPIO_PORT2_SHIFT;
+		out <<= QSFP_GPIO_PORT2_SHIFT;
+	}
+	dd->f_gpio_mod(dd, out, mask, mask);
+
+	/*
+	 * Module could take up to 2 Msec to respond to MOD_SEL,
+	 * and there is no way to tell if it is ready, so we must wait.
+	 */
+	msleep(2);
+
+	/* Make sure TWSI bus is in sane state. */
+	ret = qib_twsi_reset(dd);
+	if (ret) {
+		qib_dev_porterr(dd, ppd->port,
+				"QSFP interface Reset for write failed\n");
+		ret = -EIO;
+		goto deselect;
+	}
+
+	/* All QSFP modules are at A0 */
+
+	cnt = 0;
+	while (cnt < len) {
+		unsigned in_page;
+		int wlen = len - cnt;
+		in_page = addr % QSFP_PAGESIZE;
+		if ((in_page + wlen) > QSFP_PAGESIZE)
+			wlen = QSFP_PAGESIZE - in_page;
+		ret = qib_twsi_blk_wr(dd, QSFP_DEV, addr, buff + cnt, wlen);
+		if (ret) {
+			/* qib_twsi_blk_wr() 1 for error, else 0 */
+			ret = -EIO;
+			goto deselect;
+		}
+		addr += wlen;
+		cnt += wlen;
+	}
+	ret = cnt;
+
+deselect:
+	/*
+	 * Module could take up to 10 uSec after transfer before
+	 * ready to respond to MOD_SEL negation, and there is no way
+	 * to tell if it is ready, so we must wait.
+	 */
+	udelay(10);
+	/* set QSFP MODSEL, RST, LP high */
+	dd->f_gpio_mod(dd, mask, mask, mask);
+	/*
+	 * Module could take up to 2 Msec to respond to MOD_SEL
+	 * going away, and there is no way to tell if it is ready.
+	 * so we must wait.
+	 */
+	msleep(2);
+
+bail:
+	mutex_unlock(&dd->eep_lock);
+
+no_unlock:
+	return ret;
+}
+
+/*
+ * For validation, we want to check the checksums, even of the
+ * fields we do not otherwise use. This function reads the bytes from
+ * <first> to <next-1> and returns the 8lsbs of the sum, or <0 for errors
+ */
+static int qsfp_cks(struct qib_pportdata *ppd, int first, int next)
+{
+	int ret;
+	u16 cks;
+	u8 bval;
+
+	cks = 0;
+	while (first < next) {
+		ret = qsfp_read(ppd, first, &bval, 1);
+		if (ret < 0)
+			goto bail;
+		cks += bval;
+		++first;
+	}
+	ret = cks & 0xFF;
+bail:
+	return ret;
+
+}
+
+int qib_refresh_qsfp_cache(struct qib_pportdata *ppd, struct qib_qsfp_cache *cp)
+{
+	int ret;
+	int idx;
+	u16 cks;
+	u32 mask;
+	u8 peek[4];
+
+	/* ensure sane contents on invalid reads, for cable swaps */
+	memset(cp, 0, sizeof(*cp));
+
+	mask = QSFP_GPIO_MOD_PRS_N;
+	if (ppd->hw_pidx)
+		mask <<= QSFP_GPIO_PORT2_SHIFT;
+
+	ret = ppd->dd->f_gpio_mod(ppd->dd, 0, 0, 0);
+	if (ret & mask) {
+		ret = -ENODEV;
+		goto bail;
+	}
+
+	ret = qsfp_read(ppd, 0, peek, 3);
+	if (ret < 0)
+		goto bail;
+	if ((peek[0] & 0xFE) != 0x0C)
+		qib_dev_porterr(ppd->dd, ppd->port,
+				"QSFP byte0 is 0x%02X, S/B 0x0C/D\n", peek[0]);
+
+	if ((peek[2] & 2) == 0) {
+		/*
+		 * If cable is paged, rather than "flat memory", we need to
+		 * set the page to zero, Even if it already appears to be zero.
+		 */
+		u8 poke = 0;
+		ret = qib_qsfp_write(ppd, 127, &poke, 1);
+		udelay(50);
+		if (ret != 1) {
+			qib_dev_porterr(ppd->dd, ppd->port,
+					"Failed QSFP Page set\n");
+			goto bail;
+		}
+	}
+
+	ret = qsfp_read(ppd, QSFP_MOD_ID_OFFS, &cp->id, 1);
+	if (ret < 0)
+		goto bail;
+	if ((cp->id & 0xFE) != 0x0C)
+		qib_dev_porterr(ppd->dd, ppd->port,
+				"QSFP ID byte is 0x%02X, S/B 0x0C/D\n", cp->id);
+	cks = cp->id;
+
+	ret = qsfp_read(ppd, QSFP_MOD_PWR_OFFS, &cp->pwr, 1);
+	if (ret < 0)
+		goto bail;
+	cks += cp->pwr;
+
+	ret = qsfp_cks(ppd, QSFP_MOD_PWR_OFFS + 1, QSFP_MOD_LEN_OFFS);
+	if (ret < 0)
+		goto bail;
+	cks += ret;
+
+	ret = qsfp_read(ppd, QSFP_MOD_LEN_OFFS, &cp->len, 1);
+	if (ret < 0)
+		goto bail;
+	cks += cp->len;
+
+	ret = qsfp_read(ppd, QSFP_MOD_TECH_OFFS, &cp->tech, 1);
+	if (ret < 0)
+		goto bail;
+	cks += cp->tech;
+
+	ret = qsfp_read(ppd, QSFP_VEND_OFFS, &cp->vendor, QSFP_VEND_LEN);
+	if (ret < 0)
+		goto bail;
+	for (idx = 0; idx < QSFP_VEND_LEN; ++idx)
+		cks += cp->vendor[idx];
+
+	ret = qsfp_read(ppd, QSFP_IBXCV_OFFS, &cp->xt_xcv, 1);
+	if (ret < 0)
+		goto bail;
+	cks += cp->xt_xcv;
+
+	ret = qsfp_read(ppd, QSFP_VOUI_OFFS, &cp->oui, QSFP_VOUI_LEN);
+	if (ret < 0)
+		goto bail;
+	for (idx = 0; idx < QSFP_VOUI_LEN; ++idx)
+		cks += cp->oui[idx];
+
+	ret = qsfp_read(ppd, QSFP_PN_OFFS, &cp->partnum, QSFP_PN_LEN);
+	if (ret < 0)
+		goto bail;
+	for (idx = 0; idx < QSFP_PN_LEN; ++idx)
+		cks += cp->partnum[idx];
+
+	ret = qsfp_read(ppd, QSFP_REV_OFFS, &cp->rev, QSFP_REV_LEN);
+	if (ret < 0)
+		goto bail;
+	for (idx = 0; idx < QSFP_REV_LEN; ++idx)
+		cks += cp->rev[idx];
+
+	ret = qsfp_read(ppd, QSFP_ATTEN_OFFS, &cp->atten, QSFP_ATTEN_LEN);
+	if (ret < 0)
+		goto bail;
+	for (idx = 0; idx < QSFP_ATTEN_LEN; ++idx)
+		cks += cp->atten[idx];
+
+	ret = qsfp_cks(ppd, QSFP_ATTEN_OFFS + QSFP_ATTEN_LEN, QSFP_CC_OFFS);
+	if (ret < 0)
+		goto bail;
+	cks += ret;
+
+	cks &= 0xFF;
+	ret = qsfp_read(ppd, QSFP_CC_OFFS, &cp->cks1, 1);
+	if (ret < 0)
+		goto bail;
+	if (cks != cp->cks1)
+		qib_dev_porterr(ppd->dd, ppd->port,
+				"QSFP cks1 is %02X, computed %02X\n", cp->cks1,
+				cks);
+
+	/* Second checksum covers 192 to (serial, date, lot) */
+	ret = qsfp_cks(ppd, QSFP_CC_OFFS + 1, QSFP_SN_OFFS);
+	if (ret < 0)
+		goto bail;
+	cks = ret;
+
+	ret = qsfp_read(ppd, QSFP_SN_OFFS, &cp->serial, QSFP_SN_LEN);
+	if (ret < 0)
+		goto bail;
+	for (idx = 0; idx < QSFP_SN_LEN; ++idx)
+		cks += cp->serial[idx];
+
+	ret = qsfp_read(ppd, QSFP_DATE_OFFS, &cp->date, QSFP_DATE_LEN);
+	if (ret < 0)
+		goto bail;
+	for (idx = 0; idx < QSFP_DATE_LEN; ++idx)
+		cks += cp->date[idx];
+
+	ret = qsfp_read(ppd, QSFP_LOT_OFFS, &cp->lot, QSFP_LOT_LEN);
+	if (ret < 0)
+		goto bail;
+	for (idx = 0; idx < QSFP_LOT_LEN; ++idx)
+		cks += cp->lot[idx];
+
+	ret = qsfp_cks(ppd, QSFP_LOT_OFFS + QSFP_LOT_LEN, QSFP_CC_EXT_OFFS);
+	if (ret < 0)
+		goto bail;
+	cks += ret;
+
+	ret = qsfp_read(ppd, QSFP_CC_EXT_OFFS, &cp->cks2, 1);
+	if (ret < 0)
+		goto bail;
+	cks &= 0xFF;
+	if (cks != cp->cks2)
+		qib_dev_porterr(ppd->dd, ppd->port,
+				"QSFP cks2 is %02X, computed %02X\n", cp->cks2,
+				cks);
+	return 0;
+
+bail:
+	cp->id = 0;
+	return ret;
+}
+
+const char * const qib_qsfp_devtech[16] = {
+	"850nm VCSEL", "1310nm VCSEL", "1550nm VCSEL", "1310nm FP",
+	"1310nm DFB", "1550nm DFB", "1310nm EML", "1550nm EML",
+	"Cu Misc", "1490nm DFB", "Cu NoEq", "Cu Eq",
+	"Undef", "Cu Active BothEq", "Cu FarEq", "Cu NearEq"
+};
+
+#define QSFP_DUMP_CHUNK 16 /* Holds longest string */
+#define QSFP_DEFAULT_HDR_CNT 224
+
+static const char *pwr_codes = "1.5W2.0W2.5W3.5W";
+
+/*
+ * Initialize structures that control access to QSFP. Called once per port
+ * on cards that support QSFP.
+ */
+void qib_qsfp_init(struct qib_qsfp_data *qd,
+		   void (*fevent)(struct work_struct *))
+{
+	u32 mask, highs;
+	int pins;
+
+	struct qib_devdata *dd = qd->ppd->dd;
+
+	/* Initialize work struct for later QSFP events */
+	INIT_WORK(&qd->work, fevent);
+
+	/*
+	 * Later, we may want more validation. For now, just set up pins and
+	 * blip reset. If module is present, call qib_refresh_qsfp_cache(),
+	 * to do further init.
+	 */
+	mask = QSFP_GPIO_MOD_SEL_N | QSFP_GPIO_MOD_RST_N | QSFP_GPIO_LP_MODE;
+	highs = mask - QSFP_GPIO_MOD_RST_N;
+	if (qd->ppd->hw_pidx) {
+		mask <<= QSFP_GPIO_PORT2_SHIFT;
+		highs <<= QSFP_GPIO_PORT2_SHIFT;
+	}
+	dd->f_gpio_mod(dd, highs, mask, mask);
+	udelay(20); /* Generous RST dwell */
+
+	dd->f_gpio_mod(dd, mask, mask, mask);
+	/* Spec says module can take up to two seconds! */
+	mask = QSFP_GPIO_MOD_PRS_N;
+	if (qd->ppd->hw_pidx)
+		mask <<= QSFP_GPIO_PORT2_SHIFT;
+
+	/* Do not try to wait here. Better to let event handle it */
+	pins = dd->f_gpio_mod(dd, 0, 0, 0);
+	if (pins & mask)
+		goto bail;
+	/* We see a module, but it may be unwise to look yet. Just schedule */
+	qd->t_insert = get_jiffies_64();
+	schedule_work(&qd->work);
+bail:
+	return;
+}
+
+void qib_qsfp_deinit(struct qib_qsfp_data *qd)
+{
+	/*
+	 * There is nothing to do here for now.  our
+	 * work is scheduled with schedule_work(), and
+	 * flush_scheduled_work() from remove_one will
+	 * block until all work ssetup with schedule_work()
+	 * completes.
+	 */
+}
+
+int qib_qsfp_dump(struct qib_pportdata *ppd, char *buf, int len)
+{
+	struct qib_qsfp_cache cd;
+	u8 bin_buff[QSFP_DUMP_CHUNK];
+	char lenstr[6];
+	int sofar, ret;
+	int bidx = 0;
+
+	sofar = 0;
+	ret = qib_refresh_qsfp_cache(ppd, &cd);
+	if (ret < 0)
+		goto bail;
+
+	lenstr[0] = ' ';
+	lenstr[1] = '\0';
+	if (QSFP_IS_CU(cd.tech))
+		sprintf(lenstr, "%dM ", cd.len);
+
+	sofar += scnprintf(buf + sofar, len - sofar, "PWR:%.3sW\n", pwr_codes +
+			   (QSFP_PWR(cd.pwr) * 4));
+
+	sofar += scnprintf(buf + sofar, len - sofar, "TECH:%s%s\n", lenstr,
+			   qib_qsfp_devtech[cd.tech >> 4]);
+
+	sofar += scnprintf(buf + sofar, len - sofar, "Vendor:%.*s\n",
+			   QSFP_VEND_LEN, cd.vendor);
+
+	sofar += scnprintf(buf + sofar, len - sofar, "OUI:%06X\n",
+			   QSFP_OUI(cd.oui));
+
+	sofar += scnprintf(buf + sofar, len - sofar, "Part#:%.*s\n",
+			   QSFP_PN_LEN, cd.partnum);
+	sofar += scnprintf(buf + sofar, len - sofar, "Rev:%.*s\n",
+			   QSFP_REV_LEN, cd.rev);
+	if (QSFP_IS_CU(cd.tech))
+		sofar += scnprintf(buf + sofar, len - sofar, "Atten:%d, %d\n",
+				   QSFP_ATTEN_SDR(cd.atten),
+				   QSFP_ATTEN_DDR(cd.atten));
+	sofar += scnprintf(buf + sofar, len - sofar, "Serial:%.*s\n",
+			   QSFP_SN_LEN, cd.serial);
+	sofar += scnprintf(buf + sofar, len - sofar, "Date:%.*s\n",
+			   QSFP_DATE_LEN, cd.date);
+	sofar += scnprintf(buf + sofar, len - sofar, "Lot:%.*s\n",
+			   QSFP_LOT_LEN, cd.date);
+
+	while (bidx < QSFP_DEFAULT_HDR_CNT) {
+		int iidx;
+		ret = qsfp_read(ppd, bidx, bin_buff, QSFP_DUMP_CHUNK);
+		if (ret < 0)
+			goto bail;
+		for (iidx = 0; iidx < ret; ++iidx) {
+			sofar += scnprintf(buf + sofar, len-sofar, " %02X",
+				bin_buff[iidx]);
+		}
+		sofar += scnprintf(buf + sofar, len - sofar, "\n");
+		bidx += QSFP_DUMP_CHUNK;
+	}
+	ret = sofar;
+bail:
+	return ret;
+}

drivers/infiniband/hw/qib/qib_qsfp.h

@@ -0,0 +1,184 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+/* QSFP support common definitions, for ib_qib driver */
+
+#define QSFP_DEV 0xA0
+#define QSFP_PWR_LAG_MSEC 2000
+
+/*
+ * Below are masks for various QSFP signals, for Port 1.
+ * Port2 equivalents are shifted by QSFP_GPIO_PORT2_SHIFT.
+ * _N means asserted low
+ */
+#define QSFP_GPIO_MOD_SEL_N (4)
+#define QSFP_GPIO_MOD_PRS_N (8)
+#define QSFP_GPIO_INT_N (0x10)
+#define QSFP_GPIO_MOD_RST_N (0x20)
+#define QSFP_GPIO_LP_MODE (0x40)
+#define QSFP_GPIO_PORT2_SHIFT 5
+
+#define QSFP_PAGESIZE 128
+/* Defined fields that QLogic requires of qualified cables */
+/* Byte 0 is Identifier, not checked */
+/* Byte 1 is reserved "status MSB" */
+/* Byte 2 is "status LSB" We only care that D2 "Flat Mem" is set. */
+/*
+ * Rest of first 128 not used, although 127 is reserved for page select
+ * if module is not "Flat memory".
+ */
+/* Byte 128 is Identifier: must be 0x0c for QSFP, or 0x0d for QSFP+ */
+#define QSFP_MOD_ID_OFFS 128
+/*
+ * Byte 129 is "Extended Identifier". We only care about D7,D6: Power class
+ *  0:1.5W, 1:2.0W, 2:2.5W, 3:3.5W
+ */
+#define QSFP_MOD_PWR_OFFS 129
+/* Byte 130 is Connector type. Not QLogic req'd */
+/* Bytes 131..138 are Transceiver types, bit maps for various tech, none IB */
+/* Byte 139 is encoding. code 0x01 is 8b10b. Not QLogic req'd */
+/* byte 140 is nominal bit-rate, in units of 100Mbits/sec Not QLogic req'd */
+/* Byte 141 is Extended Rate Select. Not QLogic req'd */
+/* Bytes 142..145 are lengths for various fiber types. Not QLogic req'd */
+/* Byte 146 is length for Copper. Units of 1 meter */
+#define QSFP_MOD_LEN_OFFS 146
+/*
+ * Byte 147 is Device technology. D0..3 not Qlogc req'd
+ * D4..7 select from 15 choices, translated by table:
+ */
+#define QSFP_MOD_TECH_OFFS 147
+extern const char *const qib_qsfp_devtech[16];
+/* Active Equalization includes fiber, copper full EQ, and copper near Eq */
+#define QSFP_IS_ACTIVE(tech) ((0xA2FF >> ((tech) >> 4)) & 1)
+/* Attenuation should be valid for copper other than full/near Eq */
+#define QSFP_HAS_ATTEN(tech) ((0x4D00 >> ((tech) >> 4)) & 1)
+/* Length is only valid if technology is "copper" */
+#define QSFP_IS_CU(tech) ((0xED00 >> ((tech) >> 4)) & 1)
+#define QSFP_TECH_1490 9
+
+#define QSFP_OUI(oui) (((unsigned)oui[0] << 16) | ((unsigned)oui[1] << 8) | \
+			oui[2])
+#define QSFP_OUI_AMPHENOL 0x415048
+#define QSFP_OUI_FINISAR  0x009065
+#define QSFP_OUI_GORE     0x002177
+
+/* Bytes 148..163 are Vendor Name, Left-justified Blank-filled */
+#define QSFP_VEND_OFFS 148
+#define QSFP_VEND_LEN 16
+/* Byte 164 is IB Extended tranceiver codes Bits D0..3 are SDR,DDR,QDR,EDR */
+#define QSFP_IBXCV_OFFS 164
+/* Bytes 165..167 are Vendor OUI number */
+#define QSFP_VOUI_OFFS 165
+#define QSFP_VOUI_LEN 3
+/* Bytes 168..183 are Vendor Part Number, string */
+#define QSFP_PN_OFFS 168
+#define QSFP_PN_LEN 16
+/* Bytes 184,185 are Vendor Rev. Left Justified, Blank-filled */
+#define QSFP_REV_OFFS 184
+#define QSFP_REV_LEN 2
+/*
+ * Bytes 186,187 are Wavelength, if Optical. Not Qlogic req'd
+ *  If copper, they are attenuation in dB:
+ * Byte 186 is at 2.5Gb/sec (SDR), Byte 187 at 5.0Gb/sec (DDR)
+ */
+#define QSFP_ATTEN_OFFS 186
+#define QSFP_ATTEN_LEN 2
+/* Bytes 188,189 are Wavelength tolerance, not QLogic req'd */
+/* Byte 190 is Max Case Temp. Not QLogic req'd */
+/* Byte 191 is LSB of sum of bytes 128..190. Not QLogic req'd */
+#define QSFP_CC_OFFS 191
+/* Bytes 192..195 are Options implemented in qsfp. Not Qlogic req'd */
+/* Bytes 196..211 are Serial Number, String */
+#define QSFP_SN_OFFS 196
+#define QSFP_SN_LEN 16
+/* Bytes 212..219 are date-code YYMMDD (MM==1 for Jan) */
+#define QSFP_DATE_OFFS 212
+#define QSFP_DATE_LEN 6
+/* Bytes 218,219 are optional lot-code, string */
+#define QSFP_LOT_OFFS 218
+#define QSFP_LOT_LEN 2
+/* Bytes 220, 221 indicate monitoring options, Not QLogic req'd */
+/* Byte 223 is LSB of sum of bytes 192..222 */
+#define QSFP_CC_EXT_OFFS 223
+
+/*
+ * struct qib_qsfp_data encapsulates state of QSFP device for one port.
+ * it will be part of port-chip-specific data if a board supports QSFP.
+ *
+ * Since multiple board-types use QSFP, and their pport_data structs
+ * differ (in the chip-specific section), we need a pointer to its head.
+ *
+ * Avoiding premature optimization, we will have one work_struct per port,
+ * and let the (increasingly inaccurately named) eep_lock arbitrate
+ * access to common resources.
+ *
+ */
+
+/*
+ * Hold the parts of the onboard EEPROM that we care about, so we aren't
+ * coonstantly bit-boffing
+ */
+struct qib_qsfp_cache {
+	u8 id;	/* must be 0x0C or 0x0D; 0 indicates invalid EEPROM read */
+	u8 pwr; /* in D6,7 */
+	u8 len;	/* in meters, Cu only */
+	u8 tech;
+	char vendor[QSFP_VEND_LEN];
+	u8 xt_xcv; /* Ext. tranceiver codes, 4 lsbs are IB speed supported */
+	u8 oui[QSFP_VOUI_LEN];
+	u8 partnum[QSFP_PN_LEN];
+	u8 rev[QSFP_REV_LEN];
+	u8 atten[QSFP_ATTEN_LEN];
+	u8 cks1;	/* Checksum of bytes 128..190 */
+	u8 serial[QSFP_SN_LEN];
+	u8 date[QSFP_DATE_LEN];
+	u8 lot[QSFP_LOT_LEN];
+	u8 cks2;	/* Checsum of bytes 192..222 */
+};
+
+#define QSFP_PWR(pbyte) (((pbyte) >> 6) & 3)
+#define QSFP_ATTEN_SDR(attenarray) (attenarray[0])
+#define QSFP_ATTEN_DDR(attenarray) (attenarray[1])
+
+struct qib_qsfp_data {
+	/* Helps to find our way */
+	struct qib_pportdata *ppd;
+	struct work_struct work;
+	struct qib_qsfp_cache cache;
+	u64 t_insert;
+};
+
+extern int qib_refresh_qsfp_cache(struct qib_pportdata *ppd,
+				  struct qib_qsfp_cache *cp);
+extern void qib_qsfp_init(struct qib_qsfp_data *qd,
+			  void (*fevent)(struct work_struct *));
+extern void qib_qsfp_deinit(struct qib_qsfp_data *qd);

drivers/infiniband/hw/qib/qib_ruc.c

@@ -0,0 +1,817 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/spinlock.h>
+
+#include "qib.h"
+#include "qib_mad.h"
+
+/*
+ * Convert the AETH RNR timeout code into the number of microseconds.
+ */
+const u32 ib_qib_rnr_table[32] = {
+	655360,	/* 00: 655.36 */
+	10,	/* 01:    .01 */
+	20,	/* 02     .02 */
+	30,	/* 03:    .03 */
+	40,	/* 04:    .04 */
+	60,	/* 05:    .06 */
+	80,	/* 06:    .08 */
+	120,	/* 07:    .12 */
+	160,	/* 08:    .16 */
+	240,	/* 09:    .24 */
+	320,	/* 0A:    .32 */
+	480,	/* 0B:    .48 */
+	640,	/* 0C:    .64 */
+	960,	/* 0D:    .96 */
+	1280,	/* 0E:   1.28 */
+	1920,	/* 0F:   1.92 */
+	2560,	/* 10:   2.56 */
+	3840,	/* 11:   3.84 */
+	5120,	/* 12:   5.12 */
+	7680,	/* 13:   7.68 */
+	10240,	/* 14:  10.24 */
+	15360,	/* 15:  15.36 */
+	20480,	/* 16:  20.48 */
+	30720,	/* 17:  30.72 */
+	40960,	/* 18:  40.96 */
+	61440,	/* 19:  61.44 */
+	81920,	/* 1A:  81.92 */
+	122880,	/* 1B: 122.88 */
+	163840,	/* 1C: 163.84 */
+	245760,	/* 1D: 245.76 */
+	327680,	/* 1E: 327.68 */
+	491520	/* 1F: 491.52 */
+};
+
+/*
+ * Validate a RWQE and fill in the SGE state.
+ * Return 1 if OK.
+ */
+static int qib_init_sge(struct qib_qp *qp, struct qib_rwqe *wqe)
+{
+	int i, j, ret;
+	struct ib_wc wc;
+	struct qib_lkey_table *rkt;
+	struct qib_pd *pd;
+	struct qib_sge_state *ss;
+
+	rkt = &to_idev(qp->ibqp.device)->lk_table;
+	pd = to_ipd(qp->ibqp.srq ? qp->ibqp.srq->pd : qp->ibqp.pd);
+	ss = &qp->r_sge;
+	ss->sg_list = qp->r_sg_list;
+	qp->r_len = 0;
+	for (i = j = 0; i < wqe->num_sge; i++) {
+		if (wqe->sg_list[i].length == 0)
+			continue;
+		/* Check LKEY */
+		if (!qib_lkey_ok(rkt, pd, j ? &ss->sg_list[j - 1] : &ss->sge,
+				 &wqe->sg_list[i], IB_ACCESS_LOCAL_WRITE))
+			goto bad_lkey;
+		qp->r_len += wqe->sg_list[i].length;
+		j++;
+	}
+	ss->num_sge = j;
+	ss->total_len = qp->r_len;
+	ret = 1;
+	goto bail;
+
+bad_lkey:
+	while (j) {
+		struct qib_sge *sge = --j ? &ss->sg_list[j - 1] : &ss->sge;
+
+		atomic_dec(&sge->mr->refcount);
+	}
+	ss->num_sge = 0;
+	memset(&wc, 0, sizeof(wc));
+	wc.wr_id = wqe->wr_id;
+	wc.status = IB_WC_LOC_PROT_ERR;
+	wc.opcode = IB_WC_RECV;
+	wc.qp = &qp->ibqp;
+	/* Signal solicited completion event. */
+	qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc, 1);
+	ret = 0;
+bail:
+	return ret;
+}
+
+/**
+ * qib_get_rwqe - copy the next RWQE into the QP's RWQE
+ * @qp: the QP
+ * @wr_id_only: update qp->r_wr_id only, not qp->r_sge
+ *
+ * Return -1 if there is a local error, 0 if no RWQE is available,
+ * otherwise return 1.
+ *
+ * Can be called from interrupt level.
+ */
+int qib_get_rwqe(struct qib_qp *qp, int wr_id_only)
+{
+	unsigned long flags;
+	struct qib_rq *rq;
+	struct qib_rwq *wq;
+	struct qib_srq *srq;
+	struct qib_rwqe *wqe;
+	void (*handler)(struct ib_event *, void *);
+	u32 tail;
+	int ret;
+
+	if (qp->ibqp.srq) {
+		srq = to_isrq(qp->ibqp.srq);
+		handler = srq->ibsrq.event_handler;
+		rq = &srq->rq;
+	} else {
+		srq = NULL;
+		handler = NULL;
+		rq = &qp->r_rq;
+	}
+
+	spin_lock_irqsave(&rq->lock, flags);
+	if (!(ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK)) {
+		ret = 0;
+		goto unlock;
+	}
+
+	wq = rq->wq;
+	tail = wq->tail;
+	/* Validate tail before using it since it is user writable. */
+	if (tail >= rq->size)
+		tail = 0;
+	if (unlikely(tail == wq->head)) {
+		ret = 0;
+		goto unlock;
+	}
+	/* Make sure entry is read after head index is read. */
+	smp_rmb();
+	wqe = get_rwqe_ptr(rq, tail);
+	/*
+	 * Even though we update the tail index in memory, the verbs
+	 * consumer is not supposed to post more entries until a
+	 * completion is generated.
+	 */
+	if (++tail >= rq->size)
+		tail = 0;
+	wq->tail = tail;
+	if (!wr_id_only && !qib_init_sge(qp, wqe)) {
+		ret = -1;
+		goto unlock;
+	}
+	qp->r_wr_id = wqe->wr_id;
+
+	ret = 1;
+	set_bit(QIB_R_WRID_VALID, &qp->r_aflags);
+	if (handler) {
+		u32 n;
+
+		/*
+		 * Validate head pointer value and compute
+		 * the number of remaining WQEs.
+		 */
+		n = wq->head;
+		if (n >= rq->size)
+			n = 0;
+		if (n < tail)
+			n += rq->size - tail;
+		else
+			n -= tail;
+		if (n < srq->limit) {
+			struct ib_event ev;
+
+			srq->limit = 0;
+			spin_unlock_irqrestore(&rq->lock, flags);
+			ev.device = qp->ibqp.device;
+			ev.element.srq = qp->ibqp.srq;
+			ev.event = IB_EVENT_SRQ_LIMIT_REACHED;
+			handler(&ev, srq->ibsrq.srq_context);
+			goto bail;
+		}
+	}
+unlock:
+	spin_unlock_irqrestore(&rq->lock, flags);
+bail:
+	return ret;
+}
+
+/*
+ * Switch to alternate path.
+ * The QP s_lock should be held and interrupts disabled.
+ */
+void qib_migrate_qp(struct qib_qp *qp)
+{
+	struct ib_event ev;
+
+	qp->s_mig_state = IB_MIG_MIGRATED;
+	qp->remote_ah_attr = qp->alt_ah_attr;
+	qp->port_num = qp->alt_ah_attr.port_num;
+	qp->s_pkey_index = qp->s_alt_pkey_index;
+
+	ev.device = qp->ibqp.device;
+	ev.element.qp = &qp->ibqp;
+	ev.event = IB_EVENT_PATH_MIG;
+	qp->ibqp.event_handler(&ev, qp->ibqp.qp_context);
+}
+
+static __be64 get_sguid(struct qib_ibport *ibp, unsigned index)
+{
+	if (!index) {
+		struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+
+		return ppd->guid;
+	} else
+		return ibp->guids[index - 1];
+}
+
+static int gid_ok(union ib_gid *gid, __be64 gid_prefix, __be64 id)
+{
+	return (gid->global.interface_id == id &&
+		(gid->global.subnet_prefix == gid_prefix ||
+		 gid->global.subnet_prefix == IB_DEFAULT_GID_PREFIX));
+}
+
+/*
+ *
+ * This should be called with the QP s_lock held.
+ */
+int qib_ruc_check_hdr(struct qib_ibport *ibp, struct qib_ib_header *hdr,
+		      int has_grh, struct qib_qp *qp, u32 bth0)
+{
+	__be64 guid;
+
+	if (qp->s_mig_state == IB_MIG_ARMED && (bth0 & IB_BTH_MIG_REQ)) {
+		if (!has_grh) {
+			if (qp->alt_ah_attr.ah_flags & IB_AH_GRH)
+				goto err;
+		} else {
+			if (!(qp->alt_ah_attr.ah_flags & IB_AH_GRH))
+				goto err;
+			guid = get_sguid(ibp, qp->alt_ah_attr.grh.sgid_index);
+			if (!gid_ok(&hdr->u.l.grh.dgid, ibp->gid_prefix, guid))
+				goto err;
+			if (!gid_ok(&hdr->u.l.grh.sgid,
+			    qp->alt_ah_attr.grh.dgid.global.subnet_prefix,
+			    qp->alt_ah_attr.grh.dgid.global.interface_id))
+				goto err;
+		}
+		if (!qib_pkey_ok((u16)bth0,
+				 qib_get_pkey(ibp, qp->s_alt_pkey_index))) {
+			qib_bad_pqkey(ibp, IB_NOTICE_TRAP_BAD_PKEY,
+				      (u16)bth0,
+				      (be16_to_cpu(hdr->lrh[0]) >> 4) & 0xF,
+				      0, qp->ibqp.qp_num,
+				      hdr->lrh[3], hdr->lrh[1]);
+			goto err;
+		}
+		/* Validate the SLID. See Ch. 9.6.1.5 and 17.2.8 */
+		if (be16_to_cpu(hdr->lrh[3]) != qp->alt_ah_attr.dlid ||
+		    ppd_from_ibp(ibp)->port != qp->alt_ah_attr.port_num)
+			goto err;
+		qib_migrate_qp(qp);
+	} else {
+		if (!has_grh) {
+			if (qp->remote_ah_attr.ah_flags & IB_AH_GRH)
+				goto err;
+		} else {
+			if (!(qp->remote_ah_attr.ah_flags & IB_AH_GRH))
+				goto err;
+			guid = get_sguid(ibp,
+					 qp->remote_ah_attr.grh.sgid_index);
+			if (!gid_ok(&hdr->u.l.grh.dgid, ibp->gid_prefix, guid))
+				goto err;
+			if (!gid_ok(&hdr->u.l.grh.sgid,
+			    qp->remote_ah_attr.grh.dgid.global.subnet_prefix,
+			    qp->remote_ah_attr.grh.dgid.global.interface_id))
+				goto err;
+		}
+		if (!qib_pkey_ok((u16)bth0,
+				 qib_get_pkey(ibp, qp->s_pkey_index))) {
+			qib_bad_pqkey(ibp, IB_NOTICE_TRAP_BAD_PKEY,
+				      (u16)bth0,
+				      (be16_to_cpu(hdr->lrh[0]) >> 4) & 0xF,
+				      0, qp->ibqp.qp_num,
+				      hdr->lrh[3], hdr->lrh[1]);
+			goto err;
+		}
+		/* Validate the SLID. See Ch. 9.6.1.5 */
+		if (be16_to_cpu(hdr->lrh[3]) != qp->remote_ah_attr.dlid ||
+		    ppd_from_ibp(ibp)->port != qp->port_num)
+			goto err;
+		if (qp->s_mig_state == IB_MIG_REARM &&
+		    !(bth0 & IB_BTH_MIG_REQ))
+			qp->s_mig_state = IB_MIG_ARMED;
+	}
+
+	return 0;
+
+err:
+	return 1;
+}
+
+/**
+ * qib_ruc_loopback - handle UC and RC lookback requests
+ * @sqp: the sending QP
+ *
+ * This is called from qib_do_send() to
+ * forward a WQE addressed to the same HCA.
+ * Note that although we are single threaded due to the tasklet, we still
+ * have to protect against post_send().  We don't have to worry about
+ * receive interrupts since this is a connected protocol and all packets
+ * will pass through here.
+ */
+static void qib_ruc_loopback(struct qib_qp *sqp)
+{
+	struct qib_ibport *ibp = to_iport(sqp->ibqp.device, sqp->port_num);
+	struct qib_qp *qp;
+	struct qib_swqe *wqe;
+	struct qib_sge *sge;
+	unsigned long flags;
+	struct ib_wc wc;
+	u64 sdata;
+	atomic64_t *maddr;
+	enum ib_wc_status send_status;
+	int release;
+	int ret;
+
+	/*
+	 * Note that we check the responder QP state after
+	 * checking the requester's state.
+	 */
+	qp = qib_lookup_qpn(ibp, sqp->remote_qpn);
+
+	spin_lock_irqsave(&sqp->s_lock, flags);
+
+	/* Return if we are already busy processing a work request. */
+	if ((sqp->s_flags & (QIB_S_BUSY | QIB_S_ANY_WAIT)) ||
+	    !(ib_qib_state_ops[sqp->state] & QIB_PROCESS_OR_FLUSH_SEND))
+		goto unlock;
+
+	sqp->s_flags |= QIB_S_BUSY;
+
+again:
+	if (sqp->s_last == sqp->s_head)
+		goto clr_busy;
+	wqe = get_swqe_ptr(sqp, sqp->s_last);
+
+	/* Return if it is not OK to start a new work reqeust. */
+	if (!(ib_qib_state_ops[sqp->state] & QIB_PROCESS_NEXT_SEND_OK)) {
+		if (!(ib_qib_state_ops[sqp->state] & QIB_FLUSH_SEND))
+			goto clr_busy;
+		/* We are in the error state, flush the work request. */
+		send_status = IB_WC_WR_FLUSH_ERR;
+		goto flush_send;
+	}
+
+	/*
+	 * We can rely on the entry not changing without the s_lock
+	 * being held until we update s_last.
+	 * We increment s_cur to indicate s_last is in progress.
+	 */
+	if (sqp->s_last == sqp->s_cur) {
+		if (++sqp->s_cur >= sqp->s_size)
+			sqp->s_cur = 0;
+	}
+	spin_unlock_irqrestore(&sqp->s_lock, flags);
+
+	if (!qp || !(ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK) ||
+	    qp->ibqp.qp_type != sqp->ibqp.qp_type) {
+		ibp->n_pkt_drops++;
+		/*
+		 * For RC, the requester would timeout and retry so
+		 * shortcut the timeouts and just signal too many retries.
+		 */
+		if (sqp->ibqp.qp_type == IB_QPT_RC)
+			send_status = IB_WC_RETRY_EXC_ERR;
+		else
+			send_status = IB_WC_SUCCESS;
+		goto serr;
+	}
+
+	memset(&wc, 0, sizeof wc);
+	send_status = IB_WC_SUCCESS;
+
+	release = 1;
+	sqp->s_sge.sge = wqe->sg_list[0];
+	sqp->s_sge.sg_list = wqe->sg_list + 1;
+	sqp->s_sge.num_sge = wqe->wr.num_sge;
+	sqp->s_len = wqe->length;
+	switch (wqe->wr.opcode) {
+	case IB_WR_SEND_WITH_IMM:
+		wc.wc_flags = IB_WC_WITH_IMM;
+		wc.ex.imm_data = wqe->wr.ex.imm_data;
+		/* FALLTHROUGH */
+	case IB_WR_SEND:
+		ret = qib_get_rwqe(qp, 0);
+		if (ret < 0)
+			goto op_err;
+		if (!ret)
+			goto rnr_nak;
+		break;
+
+	case IB_WR_RDMA_WRITE_WITH_IMM:
+		if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_WRITE)))
+			goto inv_err;
+		wc.wc_flags = IB_WC_WITH_IMM;
+		wc.ex.imm_data = wqe->wr.ex.imm_data;
+		ret = qib_get_rwqe(qp, 1);
+		if (ret < 0)
+			goto op_err;
+		if (!ret)
+			goto rnr_nak;
+		/* FALLTHROUGH */
+	case IB_WR_RDMA_WRITE:
+		if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_WRITE)))
+			goto inv_err;
+		if (wqe->length == 0)
+			break;
+		if (unlikely(!qib_rkey_ok(qp, &qp->r_sge.sge, wqe->length,
+					  wqe->wr.wr.rdma.remote_addr,
+					  wqe->wr.wr.rdma.rkey,
+					  IB_ACCESS_REMOTE_WRITE)))
+			goto acc_err;
+		qp->r_sge.sg_list = NULL;
+		qp->r_sge.num_sge = 1;
+		qp->r_sge.total_len = wqe->length;
+		break;
+
+	case IB_WR_RDMA_READ:
+		if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_READ)))
+			goto inv_err;
+		if (unlikely(!qib_rkey_ok(qp, &sqp->s_sge.sge, wqe->length,
+					  wqe->wr.wr.rdma.remote_addr,
+					  wqe->wr.wr.rdma.rkey,
+					  IB_ACCESS_REMOTE_READ)))
+			goto acc_err;
+		release = 0;
+		sqp->s_sge.sg_list = NULL;
+		sqp->s_sge.num_sge = 1;
+		qp->r_sge.sge = wqe->sg_list[0];
+		qp->r_sge.sg_list = wqe->sg_list + 1;
+		qp->r_sge.num_sge = wqe->wr.num_sge;
+		qp->r_sge.total_len = wqe->length;
+		break;
+
+	case IB_WR_ATOMIC_CMP_AND_SWP:
+	case IB_WR_ATOMIC_FETCH_AND_ADD:
+		if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_ATOMIC)))
+			goto inv_err;
+		if (unlikely(!qib_rkey_ok(qp, &qp->r_sge.sge, sizeof(u64),
+					  wqe->wr.wr.atomic.remote_addr,
+					  wqe->wr.wr.atomic.rkey,
+					  IB_ACCESS_REMOTE_ATOMIC)))
+			goto acc_err;
+		/* Perform atomic OP and save result. */
+		maddr = (atomic64_t *) qp->r_sge.sge.vaddr;
+		sdata = wqe->wr.wr.atomic.compare_add;
+		*(u64 *) sqp->s_sge.sge.vaddr =
+			(wqe->wr.opcode == IB_WR_ATOMIC_FETCH_AND_ADD) ?
+			(u64) atomic64_add_return(sdata, maddr) - sdata :
+			(u64) cmpxchg((u64 *) qp->r_sge.sge.vaddr,
+				      sdata, wqe->wr.wr.atomic.swap);
+		atomic_dec(&qp->r_sge.sge.mr->refcount);
+		qp->r_sge.num_sge = 0;
+		goto send_comp;
+
+	default:
+		send_status = IB_WC_LOC_QP_OP_ERR;
+		goto serr;
+	}
+
+	sge = &sqp->s_sge.sge;
+	while (sqp->s_len) {
+		u32 len = sqp->s_len;
+
+		if (len > sge->length)
+			len = sge->length;
+		if (len > sge->sge_length)
+			len = sge->sge_length;
+		BUG_ON(len == 0);
+		qib_copy_sge(&qp->r_sge, sge->vaddr, len, release);
+		sge->vaddr += len;
+		sge->length -= len;
+		sge->sge_length -= len;
+		if (sge->sge_length == 0) {
+			if (!release)
+				atomic_dec(&sge->mr->refcount);
+			if (--sqp->s_sge.num_sge)
+				*sge = *sqp->s_sge.sg_list++;
+		} else if (sge->length == 0 && sge->mr->lkey) {
+			if (++sge->n >= QIB_SEGSZ) {
+				if (++sge->m >= sge->mr->mapsz)
+					break;
+				sge->n = 0;
+			}
+			sge->vaddr =
+				sge->mr->map[sge->m]->segs[sge->n].vaddr;
+			sge->length =
+				sge->mr->map[sge->m]->segs[sge->n].length;
+		}
+		sqp->s_len -= len;
+	}
+	if (release)
+		while (qp->r_sge.num_sge) {
+			atomic_dec(&qp->r_sge.sge.mr->refcount);
+			if (--qp->r_sge.num_sge)
+				qp->r_sge.sge = *qp->r_sge.sg_list++;
+		}
+
+	if (!test_and_clear_bit(QIB_R_WRID_VALID, &qp->r_aflags))
+		goto send_comp;
+
+	if (wqe->wr.opcode == IB_WR_RDMA_WRITE_WITH_IMM)
+		wc.opcode = IB_WC_RECV_RDMA_WITH_IMM;
+	else
+		wc.opcode = IB_WC_RECV;
+	wc.wr_id = qp->r_wr_id;
+	wc.status = IB_WC_SUCCESS;
+	wc.byte_len = wqe->length;
+	wc.qp = &qp->ibqp;
+	wc.src_qp = qp->remote_qpn;
+	wc.slid = qp->remote_ah_attr.dlid;
+	wc.sl = qp->remote_ah_attr.sl;
+	wc.port_num = 1;
+	/* Signal completion event if the solicited bit is set. */
+	qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
+		       wqe->wr.send_flags & IB_SEND_SOLICITED);
+
+send_comp:
+	spin_lock_irqsave(&sqp->s_lock, flags);
+	ibp->n_loop_pkts++;
+flush_send:
+	sqp->s_rnr_retry = sqp->s_rnr_retry_cnt;
+	qib_send_complete(sqp, wqe, send_status);
+	goto again;
+
+rnr_nak:
+	/* Handle RNR NAK */
+	if (qp->ibqp.qp_type == IB_QPT_UC)
+		goto send_comp;
+	ibp->n_rnr_naks++;
+	/*
+	 * Note: we don't need the s_lock held since the BUSY flag
+	 * makes this single threaded.
+	 */
+	if (sqp->s_rnr_retry == 0) {
+		send_status = IB_WC_RNR_RETRY_EXC_ERR;
+		goto serr;
+	}
+	if (sqp->s_rnr_retry_cnt < 7)
+		sqp->s_rnr_retry--;
+	spin_lock_irqsave(&sqp->s_lock, flags);
+	if (!(ib_qib_state_ops[sqp->state] & QIB_PROCESS_RECV_OK))
+		goto clr_busy;
+	sqp->s_flags |= QIB_S_WAIT_RNR;
+	sqp->s_timer.function = qib_rc_rnr_retry;
+	sqp->s_timer.expires = jiffies +
+		usecs_to_jiffies(ib_qib_rnr_table[qp->r_min_rnr_timer]);
+	add_timer(&sqp->s_timer);
+	goto clr_busy;
+
+op_err:
+	send_status = IB_WC_REM_OP_ERR;
+	wc.status = IB_WC_LOC_QP_OP_ERR;
+	goto err;
+
+inv_err:
+	send_status = IB_WC_REM_INV_REQ_ERR;
+	wc.status = IB_WC_LOC_QP_OP_ERR;
+	goto err;
+
+acc_err:
+	send_status = IB_WC_REM_ACCESS_ERR;
+	wc.status = IB_WC_LOC_PROT_ERR;
+err:
+	/* responder goes to error state */
+	qib_rc_error(qp, wc.status);
+
+serr:
+	spin_lock_irqsave(&sqp->s_lock, flags);
+	qib_send_complete(sqp, wqe, send_status);
+	if (sqp->ibqp.qp_type == IB_QPT_RC) {
+		int lastwqe = qib_error_qp(sqp, IB_WC_WR_FLUSH_ERR);
+
+		sqp->s_flags &= ~QIB_S_BUSY;
+		spin_unlock_irqrestore(&sqp->s_lock, flags);
+		if (lastwqe) {
+			struct ib_event ev;
+
+			ev.device = sqp->ibqp.device;
+			ev.element.qp = &sqp->ibqp;
+			ev.event = IB_EVENT_QP_LAST_WQE_REACHED;
+			sqp->ibqp.event_handler(&ev, sqp->ibqp.qp_context);
+		}
+		goto done;
+	}
+clr_busy:
+	sqp->s_flags &= ~QIB_S_BUSY;
+unlock:
+	spin_unlock_irqrestore(&sqp->s_lock, flags);
+done:
+	if (qp && atomic_dec_and_test(&qp->refcount))
+		wake_up(&qp->wait);
+}
+
+/**
+ * qib_make_grh - construct a GRH header
+ * @ibp: a pointer to the IB port
+ * @hdr: a pointer to the GRH header being constructed
+ * @grh: the global route address to send to
+ * @hwords: the number of 32 bit words of header being sent
+ * @nwords: the number of 32 bit words of data being sent
+ *
+ * Return the size of the header in 32 bit words.
+ */
+u32 qib_make_grh(struct qib_ibport *ibp, struct ib_grh *hdr,
+		 struct ib_global_route *grh, u32 hwords, u32 nwords)
+{
+	hdr->version_tclass_flow =
+		cpu_to_be32((IB_GRH_VERSION << IB_GRH_VERSION_SHIFT) |
+			    (grh->traffic_class << IB_GRH_TCLASS_SHIFT) |
+			    (grh->flow_label << IB_GRH_FLOW_SHIFT));
+	hdr->paylen = cpu_to_be16((hwords - 2 + nwords + SIZE_OF_CRC) << 2);
+	/* next_hdr is defined by C8-7 in ch. 8.4.1 */
+	hdr->next_hdr = IB_GRH_NEXT_HDR;
+	hdr->hop_limit = grh->hop_limit;
+	/* The SGID is 32-bit aligned. */
+	hdr->sgid.global.subnet_prefix = ibp->gid_prefix;
+	hdr->sgid.global.interface_id = grh->sgid_index ?
+		ibp->guids[grh->sgid_index - 1] : ppd_from_ibp(ibp)->guid;
+	hdr->dgid = grh->dgid;
+
+	/* GRH header size in 32-bit words. */
+	return sizeof(struct ib_grh) / sizeof(u32);
+}
+
+void qib_make_ruc_header(struct qib_qp *qp, struct qib_other_headers *ohdr,
+			 u32 bth0, u32 bth2)
+{
+	struct qib_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
+	u16 lrh0;
+	u32 nwords;
+	u32 extra_bytes;
+
+	/* Construct the header. */
+	extra_bytes = -qp->s_cur_size & 3;
+	nwords = (qp->s_cur_size + extra_bytes) >> 2;
+	lrh0 = QIB_LRH_BTH;
+	if (unlikely(qp->remote_ah_attr.ah_flags & IB_AH_GRH)) {
+		qp->s_hdrwords += qib_make_grh(ibp, &qp->s_hdr.u.l.grh,
+					       &qp->remote_ah_attr.grh,
+					       qp->s_hdrwords, nwords);
+		lrh0 = QIB_LRH_GRH;
+	}
+	lrh0 |= ibp->sl_to_vl[qp->remote_ah_attr.sl] << 12 |
+		qp->remote_ah_attr.sl << 4;
+	qp->s_hdr.lrh[0] = cpu_to_be16(lrh0);
+	qp->s_hdr.lrh[1] = cpu_to_be16(qp->remote_ah_attr.dlid);
+	qp->s_hdr.lrh[2] = cpu_to_be16(qp->s_hdrwords + nwords + SIZE_OF_CRC);
+	qp->s_hdr.lrh[3] = cpu_to_be16(ppd_from_ibp(ibp)->lid |
+				       qp->remote_ah_attr.src_path_bits);
+	bth0 |= qib_get_pkey(ibp, qp->s_pkey_index);
+	bth0 |= extra_bytes << 20;
+	if (qp->s_mig_state == IB_MIG_MIGRATED)
+		bth0 |= IB_BTH_MIG_REQ;
+	ohdr->bth[0] = cpu_to_be32(bth0);
+	ohdr->bth[1] = cpu_to_be32(qp->remote_qpn);
+	ohdr->bth[2] = cpu_to_be32(bth2);
+}
+
+/**
+ * qib_do_send - perform a send on a QP
+ * @work: contains a pointer to the QP
+ *
+ * Process entries in the send work queue until credit or queue is
+ * exhausted.  Only allow one CPU to send a packet per QP (tasklet).
+ * Otherwise, two threads could send packets out of order.
+ */
+void qib_do_send(struct work_struct *work)
+{
+	struct qib_qp *qp = container_of(work, struct qib_qp, s_work);
+	struct qib_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	int (*make_req)(struct qib_qp *qp);
+	unsigned long flags;
+
+	if ((qp->ibqp.qp_type == IB_QPT_RC ||
+	     qp->ibqp.qp_type == IB_QPT_UC) &&
+	    (qp->remote_ah_attr.dlid & ~((1 << ppd->lmc) - 1)) == ppd->lid) {
+		qib_ruc_loopback(qp);
+		return;
+	}
+
+	if (qp->ibqp.qp_type == IB_QPT_RC)
+		make_req = qib_make_rc_req;
+	else if (qp->ibqp.qp_type == IB_QPT_UC)
+		make_req = qib_make_uc_req;
+	else
+		make_req = qib_make_ud_req;
+
+	spin_lock_irqsave(&qp->s_lock, flags);
+
+	/* Return if we are already busy processing a work request. */
+	if (!qib_send_ok(qp)) {
+		spin_unlock_irqrestore(&qp->s_lock, flags);
+		return;
+	}
+
+	qp->s_flags |= QIB_S_BUSY;
+
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+
+	do {
+		/* Check for a constructed packet to be sent. */
+		if (qp->s_hdrwords != 0) {
+			/*
+			 * If the packet cannot be sent now, return and
+			 * the send tasklet will be woken up later.
+			 */
+			if (qib_verbs_send(qp, &qp->s_hdr, qp->s_hdrwords,
+					   qp->s_cur_sge, qp->s_cur_size))
+				break;
+			/* Record that s_hdr is empty. */
+			qp->s_hdrwords = 0;
+		}
+	} while (make_req(qp));
+}
+
+/*
+ * This should be called with s_lock held.
+ */
+void qib_send_complete(struct qib_qp *qp, struct qib_swqe *wqe,
+		       enum ib_wc_status status)
+{
+	u32 old_last, last;
+	unsigned i;
+
+	if (!(ib_qib_state_ops[qp->state] & QIB_PROCESS_OR_FLUSH_SEND))
+		return;
+
+	for (i = 0; i < wqe->wr.num_sge; i++) {
+		struct qib_sge *sge = &wqe->sg_list[i];
+
+		atomic_dec(&sge->mr->refcount);
+	}
+	if (qp->ibqp.qp_type == IB_QPT_UD ||
+	    qp->ibqp.qp_type == IB_QPT_SMI ||
+	    qp->ibqp.qp_type == IB_QPT_GSI)
+		atomic_dec(&to_iah(wqe->wr.wr.ud.ah)->refcount);
+
+	/* See ch. 11.2.4.1 and 10.7.3.1 */
+	if (!(qp->s_flags & QIB_S_SIGNAL_REQ_WR) ||
+	    (wqe->wr.send_flags & IB_SEND_SIGNALED) ||
+	    status != IB_WC_SUCCESS) {
+		struct ib_wc wc;
+
+		memset(&wc, 0, sizeof wc);
+		wc.wr_id = wqe->wr.wr_id;
+		wc.status = status;
+		wc.opcode = ib_qib_wc_opcode[wqe->wr.opcode];
+		wc.qp = &qp->ibqp;
+		if (status == IB_WC_SUCCESS)
+			wc.byte_len = wqe->length;
+		qib_cq_enter(to_icq(qp->ibqp.send_cq), &wc,
+			     status != IB_WC_SUCCESS);
+	}
+
+	last = qp->s_last;
+	old_last = last;
+	if (++last >= qp->s_size)
+		last = 0;
+	qp->s_last = last;
+	if (qp->s_acked == old_last)
+		qp->s_acked = last;
+	if (qp->s_cur == old_last)
+		qp->s_cur = last;
+	if (qp->s_tail == old_last)
+		qp->s_tail = last;
+	if (qp->state == IB_QPS_SQD && last == qp->s_cur)
+		qp->s_draining = 0;
+}

drivers/infiniband/hw/qib/qib_sdma.c

@@ -0,0 +1,973 @@
+/*
+ * Copyright (c) 2007, 2008, 2009, 2010 QLogic Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/spinlock.h>
+#include <linux/netdevice.h>
+
+#include "qib.h"
+#include "qib_common.h"
+
+/* default pio off, sdma on */
+static ushort sdma_descq_cnt = 256;
+module_param_named(sdma_descq_cnt, sdma_descq_cnt, ushort, S_IRUGO);
+MODULE_PARM_DESC(sdma_descq_cnt, "Number of SDMA descq entries");
+
+/*
+ * Bits defined in the send DMA descriptor.
+ */
+#define SDMA_DESC_LAST          (1ULL << 11)
+#define SDMA_DESC_FIRST         (1ULL << 12)
+#define SDMA_DESC_DMA_HEAD      (1ULL << 13)
+#define SDMA_DESC_USE_LARGE_BUF (1ULL << 14)
+#define SDMA_DESC_INTR          (1ULL << 15)
+#define SDMA_DESC_COUNT_LSB     16
+#define SDMA_DESC_GEN_LSB       30
+
+char *qib_sdma_state_names[] = {
+	[qib_sdma_state_s00_hw_down]          = "s00_HwDown",
+	[qib_sdma_state_s10_hw_start_up_wait] = "s10_HwStartUpWait",
+	[qib_sdma_state_s20_idle]             = "s20_Idle",
+	[qib_sdma_state_s30_sw_clean_up_wait] = "s30_SwCleanUpWait",
+	[qib_sdma_state_s40_hw_clean_up_wait] = "s40_HwCleanUpWait",
+	[qib_sdma_state_s50_hw_halt_wait]     = "s50_HwHaltWait",
+	[qib_sdma_state_s99_running]          = "s99_Running",
+};
+
+char *qib_sdma_event_names[] = {
+	[qib_sdma_event_e00_go_hw_down]   = "e00_GoHwDown",
+	[qib_sdma_event_e10_go_hw_start]  = "e10_GoHwStart",
+	[qib_sdma_event_e20_hw_started]   = "e20_HwStarted",
+	[qib_sdma_event_e30_go_running]   = "e30_GoRunning",
+	[qib_sdma_event_e40_sw_cleaned]   = "e40_SwCleaned",
+	[qib_sdma_event_e50_hw_cleaned]   = "e50_HwCleaned",
+	[qib_sdma_event_e60_hw_halted]    = "e60_HwHalted",
+	[qib_sdma_event_e70_go_idle]      = "e70_GoIdle",
+	[qib_sdma_event_e7220_err_halted] = "e7220_ErrHalted",
+	[qib_sdma_event_e7322_err_halted] = "e7322_ErrHalted",
+	[qib_sdma_event_e90_timer_tick]   = "e90_TimerTick",
+};
+
+/* declare all statics here rather than keep sorting */
+static int alloc_sdma(struct qib_pportdata *);
+static void sdma_complete(struct kref *);
+static void sdma_finalput(struct qib_sdma_state *);
+static void sdma_get(struct qib_sdma_state *);
+static void sdma_put(struct qib_sdma_state *);
+static void sdma_set_state(struct qib_pportdata *, enum qib_sdma_states);
+static void sdma_start_sw_clean_up(struct qib_pportdata *);
+static void sdma_sw_clean_up_task(unsigned long);
+static void unmap_desc(struct qib_pportdata *, unsigned);
+
+static void sdma_get(struct qib_sdma_state *ss)
+{
+	kref_get(&ss->kref);
+}
+
+static void sdma_complete(struct kref *kref)
+{
+	struct qib_sdma_state *ss =
+		container_of(kref, struct qib_sdma_state, kref);
+
+	complete(&ss->comp);
+}
+
+static void sdma_put(struct qib_sdma_state *ss)
+{
+	kref_put(&ss->kref, sdma_complete);
+}
+
+static void sdma_finalput(struct qib_sdma_state *ss)
+{
+	sdma_put(ss);
+	wait_for_completion(&ss->comp);
+}
+
+/*
+ * Complete all the sdma requests on the active list, in the correct
+ * order, and with appropriate processing.   Called when cleaning up
+ * after sdma shutdown, and when new sdma requests are submitted for
+ * a link that is down.   This matches what is done for requests
+ * that complete normally, it's just the full list.
+ *
+ * Must be called with sdma_lock held
+ */
+static void clear_sdma_activelist(struct qib_pportdata *ppd)
+{
+	struct qib_sdma_txreq *txp, *txp_next;
+
+	list_for_each_entry_safe(txp, txp_next, &ppd->sdma_activelist, list) {
+		list_del_init(&txp->list);
+		if (txp->flags & QIB_SDMA_TXREQ_F_FREEDESC) {
+			unsigned idx;
+
+			idx = txp->start_idx;
+			while (idx != txp->next_descq_idx) {
+				unmap_desc(ppd, idx);
+				if (++idx == ppd->sdma_descq_cnt)
+					idx = 0;
+			}
+		}
+		if (txp->callback)
+			(*txp->callback)(txp, QIB_SDMA_TXREQ_S_ABORTED);
+	}
+}
+
+static void sdma_sw_clean_up_task(unsigned long opaque)
+{
+	struct qib_pportdata *ppd = (struct qib_pportdata *) opaque;
+	unsigned long flags;
+
+	spin_lock_irqsave(&ppd->sdma_lock, flags);
+
+	/*
+	 * At this point, the following should always be true:
+	 * - We are halted, so no more descriptors are getting retired.
+	 * - We are not running, so no one is submitting new work.
+	 * - Only we can send the e40_sw_cleaned, so we can't start
+	 *   running again until we say so.  So, the active list and
+	 *   descq are ours to play with.
+	 */
+
+	/* Process all retired requests. */
+	qib_sdma_make_progress(ppd);
+
+	clear_sdma_activelist(ppd);
+
+	/*
+	 * Resync count of added and removed.  It is VERY important that
+	 * sdma_descq_removed NEVER decrement - user_sdma depends on it.
+	 */
+	ppd->sdma_descq_removed = ppd->sdma_descq_added;
+
+	/*
+	 * Reset our notion of head and tail.
+	 * Note that the HW registers will be reset when switching states
+	 * due to calling __qib_sdma_process_event() below.
+	 */
+	ppd->sdma_descq_tail = 0;
+	ppd->sdma_descq_head = 0;
+	ppd->sdma_head_dma[0] = 0;
+	ppd->sdma_generation = 0;
+
+	__qib_sdma_process_event(ppd, qib_sdma_event_e40_sw_cleaned);
+
+	spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+}
+
+/*
+ * This is called when changing to state qib_sdma_state_s10_hw_start_up_wait
+ * as a result of send buffer errors or send DMA descriptor errors.
+ * We want to disarm the buffers in these cases.
+ */
+static void sdma_hw_start_up(struct qib_pportdata *ppd)
+{
+	struct qib_sdma_state *ss = &ppd->sdma_state;
+	unsigned bufno;
+
+	for (bufno = ss->first_sendbuf; bufno < ss->last_sendbuf; ++bufno)
+		ppd->dd->f_sendctrl(ppd, QIB_SENDCTRL_DISARM_BUF(bufno));
+
+	ppd->dd->f_sdma_hw_start_up(ppd);
+}
+
+static void sdma_sw_tear_down(struct qib_pportdata *ppd)
+{
+	struct qib_sdma_state *ss = &ppd->sdma_state;
+
+	/* Releasing this reference means the state machine has stopped. */
+	sdma_put(ss);
+}
+
+static void sdma_start_sw_clean_up(struct qib_pportdata *ppd)
+{
+	tasklet_hi_schedule(&ppd->sdma_sw_clean_up_task);
+}
+
+static void sdma_set_state(struct qib_pportdata *ppd,
+	enum qib_sdma_states next_state)
+{
+	struct qib_sdma_state *ss = &ppd->sdma_state;
+	struct sdma_set_state_action *action = ss->set_state_action;
+	unsigned op = 0;
+
+	/* debugging bookkeeping */
+	ss->previous_state = ss->current_state;
+	ss->previous_op = ss->current_op;
+
+	ss->current_state = next_state;
+
+	if (action[next_state].op_enable)
+		op |= QIB_SDMA_SENDCTRL_OP_ENABLE;
+
+	if (action[next_state].op_intenable)
+		op |= QIB_SDMA_SENDCTRL_OP_INTENABLE;
+
+	if (action[next_state].op_halt)
+		op |= QIB_SDMA_SENDCTRL_OP_HALT;
+
+	if (action[next_state].op_drain)
+		op |= QIB_SDMA_SENDCTRL_OP_DRAIN;
+
+	if (action[next_state].go_s99_running_tofalse)
+		ss->go_s99_running = 0;
+
+	if (action[next_state].go_s99_running_totrue)
+		ss->go_s99_running = 1;
+
+	ss->current_op = op;
+
+	ppd->dd->f_sdma_sendctrl(ppd, ss->current_op);
+}
+
+static void unmap_desc(struct qib_pportdata *ppd, unsigned head)
+{
+	__le64 *descqp = &ppd->sdma_descq[head].qw[0];
+	u64 desc[2];
+	dma_addr_t addr;
+	size_t len;
+
+	desc[0] = le64_to_cpu(descqp[0]);
+	desc[1] = le64_to_cpu(descqp[1]);
+
+	addr = (desc[1] << 32) | (desc[0] >> 32);
+	len = (desc[0] >> 14) & (0x7ffULL << 2);
+	dma_unmap_single(&ppd->dd->pcidev->dev, addr, len, DMA_TO_DEVICE);
+}
+
+static int alloc_sdma(struct qib_pportdata *ppd)
+{
+	ppd->sdma_descq_cnt = sdma_descq_cnt;
+	if (!ppd->sdma_descq_cnt)
+		ppd->sdma_descq_cnt = 256;
+
+	/* Allocate memory for SendDMA descriptor FIFO */
+	ppd->sdma_descq = dma_alloc_coherent(&ppd->dd->pcidev->dev,
+		ppd->sdma_descq_cnt * sizeof(u64[2]), &ppd->sdma_descq_phys,
+		GFP_KERNEL);
+
+	if (!ppd->sdma_descq) {
+		qib_dev_err(ppd->dd, "failed to allocate SendDMA descriptor "
+			    "FIFO memory\n");
+		goto bail;
+	}
+
+	/* Allocate memory for DMA of head register to memory */
+	ppd->sdma_head_dma = dma_alloc_coherent(&ppd->dd->pcidev->dev,
+		PAGE_SIZE, &ppd->sdma_head_phys, GFP_KERNEL);
+	if (!ppd->sdma_head_dma) {
+		qib_dev_err(ppd->dd, "failed to allocate SendDMA "
+			    "head memory\n");
+		goto cleanup_descq;
+	}
+	ppd->sdma_head_dma[0] = 0;
+	return 0;
+
+cleanup_descq:
+	dma_free_coherent(&ppd->dd->pcidev->dev,
+		ppd->sdma_descq_cnt * sizeof(u64[2]), (void *)ppd->sdma_descq,
+		ppd->sdma_descq_phys);
+	ppd->sdma_descq = NULL;
+	ppd->sdma_descq_phys = 0;
+bail:
+	ppd->sdma_descq_cnt = 0;
+	return -ENOMEM;
+}
+
+static void free_sdma(struct qib_pportdata *ppd)
+{
+	struct qib_devdata *dd = ppd->dd;
+
+	if (ppd->sdma_head_dma) {
+		dma_free_coherent(&dd->pcidev->dev, PAGE_SIZE,
+				  (void *)ppd->sdma_head_dma,
+				  ppd->sdma_head_phys);
+		ppd->sdma_head_dma = NULL;
+		ppd->sdma_head_phys = 0;
+	}
+
+	if (ppd->sdma_descq) {
+		dma_free_coherent(&dd->pcidev->dev,
+				  ppd->sdma_descq_cnt * sizeof(u64[2]),
+				  ppd->sdma_descq, ppd->sdma_descq_phys);
+		ppd->sdma_descq = NULL;
+		ppd->sdma_descq_phys = 0;
+	}
+}
+
+static inline void make_sdma_desc(struct qib_pportdata *ppd,
+				  u64 *sdmadesc, u64 addr, u64 dwlen,
+				  u64 dwoffset)
+{
+
+	WARN_ON(addr & 3);
+	/* SDmaPhyAddr[47:32] */
+	sdmadesc[1] = addr >> 32;
+	/* SDmaPhyAddr[31:0] */
+	sdmadesc[0] = (addr & 0xfffffffcULL) << 32;
+	/* SDmaGeneration[1:0] */
+	sdmadesc[0] |= (ppd->sdma_generation & 3ULL) <<
+		SDMA_DESC_GEN_LSB;
+	/* SDmaDwordCount[10:0] */
+	sdmadesc[0] |= (dwlen & 0x7ffULL) << SDMA_DESC_COUNT_LSB;
+	/* SDmaBufOffset[12:2] */
+	sdmadesc[0] |= dwoffset & 0x7ffULL;
+}
+
+/* sdma_lock must be held */
+int qib_sdma_make_progress(struct qib_pportdata *ppd)
+{
+	struct list_head *lp = NULL;
+	struct qib_sdma_txreq *txp = NULL;
+	struct qib_devdata *dd = ppd->dd;
+	int progress = 0;
+	u16 hwhead;
+	u16 idx = 0;
+
+	hwhead = dd->f_sdma_gethead(ppd);
+
+	/* The reason for some of the complexity of this code is that
+	 * not all descriptors have corresponding txps.  So, we have to
+	 * be able to skip over descs until we wander into the range of
+	 * the next txp on the list.
+	 */
+
+	if (!list_empty(&ppd->sdma_activelist)) {
+		lp = ppd->sdma_activelist.next;
+		txp = list_entry(lp, struct qib_sdma_txreq, list);
+		idx = txp->start_idx;
+	}
+
+	while (ppd->sdma_descq_head != hwhead) {
+		/* if desc is part of this txp, unmap if needed */
+		if (txp && (txp->flags & QIB_SDMA_TXREQ_F_FREEDESC) &&
+		    (idx == ppd->sdma_descq_head)) {
+			unmap_desc(ppd, ppd->sdma_descq_head);
+			if (++idx == ppd->sdma_descq_cnt)
+				idx = 0;
+		}
+
+		/* increment dequed desc count */
+		ppd->sdma_descq_removed++;
+
+		/* advance head, wrap if needed */
+		if (++ppd->sdma_descq_head == ppd->sdma_descq_cnt)
+			ppd->sdma_descq_head = 0;
+
+		/* if now past this txp's descs, do the callback */
+		if (txp && txp->next_descq_idx == ppd->sdma_descq_head) {
+			/* remove from active list */
+			list_del_init(&txp->list);
+			if (txp->callback)
+				(*txp->callback)(txp, QIB_SDMA_TXREQ_S_OK);
+			/* see if there is another txp */
+			if (list_empty(&ppd->sdma_activelist))
+				txp = NULL;
+			else {
+				lp = ppd->sdma_activelist.next;
+				txp = list_entry(lp, struct qib_sdma_txreq,
+					list);
+				idx = txp->start_idx;
+			}
+		}
+		progress = 1;
+	}
+	if (progress)
+		qib_verbs_sdma_desc_avail(ppd, qib_sdma_descq_freecnt(ppd));
+	return progress;
+}
+
+/*
+ * This is called from interrupt context.
+ */
+void qib_sdma_intr(struct qib_pportdata *ppd)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&ppd->sdma_lock, flags);
+
+	__qib_sdma_intr(ppd);
+
+	spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+}
+
+void __qib_sdma_intr(struct qib_pportdata *ppd)
+{
+	if (__qib_sdma_running(ppd))
+		qib_sdma_make_progress(ppd);
+}
+
+int qib_setup_sdma(struct qib_pportdata *ppd)
+{
+	struct qib_devdata *dd = ppd->dd;
+	unsigned long flags;
+	int ret = 0;
+
+	ret = alloc_sdma(ppd);
+	if (ret)
+		goto bail;
+
+	/* set consistent sdma state */
+	ppd->dd->f_sdma_init_early(ppd);
+	spin_lock_irqsave(&ppd->sdma_lock, flags);
+	sdma_set_state(ppd, qib_sdma_state_s00_hw_down);
+	spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+
+	/* set up reference counting */
+	kref_init(&ppd->sdma_state.kref);
+	init_completion(&ppd->sdma_state.comp);
+
+	ppd->sdma_generation = 0;
+	ppd->sdma_descq_head = 0;
+	ppd->sdma_descq_removed = 0;
+	ppd->sdma_descq_added = 0;
+
+	INIT_LIST_HEAD(&ppd->sdma_activelist);
+
+	tasklet_init(&ppd->sdma_sw_clean_up_task, sdma_sw_clean_up_task,
+		(unsigned long)ppd);
+
+	ret = dd->f_init_sdma_regs(ppd);
+	if (ret)
+		goto bail_alloc;
+
+	qib_sdma_process_event(ppd, qib_sdma_event_e10_go_hw_start);
+
+	return 0;
+
+bail_alloc:
+	qib_teardown_sdma(ppd);
+bail:
+	return ret;
+}
+
+void qib_teardown_sdma(struct qib_pportdata *ppd)
+{
+	qib_sdma_process_event(ppd, qib_sdma_event_e00_go_hw_down);
+
+	/*
+	 * This waits for the state machine to exit so it is not
+	 * necessary to kill the sdma_sw_clean_up_task to make sure
+	 * it is not running.
+	 */
+	sdma_finalput(&ppd->sdma_state);
+
+	free_sdma(ppd);
+}
+
+int qib_sdma_running(struct qib_pportdata *ppd)
+{
+	unsigned long flags;
+	int ret;
+
+	spin_lock_irqsave(&ppd->sdma_lock, flags);
+	ret = __qib_sdma_running(ppd);
+	spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+
+	return ret;
+}
+
+/*
+ * Complete a request when sdma not running; likely only request
+ * but to simplify the code, always queue it, then process the full
+ * activelist.  We process the entire list to ensure that this particular
+ * request does get it's callback, but in the correct order.
+ * Must be called with sdma_lock held
+ */
+static void complete_sdma_err_req(struct qib_pportdata *ppd,
+				  struct qib_verbs_txreq *tx)
+{
+	atomic_inc(&tx->qp->s_dma_busy);
+	/* no sdma descriptors, so no unmap_desc */
+	tx->txreq.start_idx = 0;
+	tx->txreq.next_descq_idx = 0;
+	list_add_tail(&tx->txreq.list, &ppd->sdma_activelist);
+	clear_sdma_activelist(ppd);
+}
+
+/*
+ * This function queues one IB packet onto the send DMA queue per call.
+ * The caller is responsible for checking:
+ * 1) The number of send DMA descriptor entries is less than the size of
+ *    the descriptor queue.
+ * 2) The IB SGE addresses and lengths are 32-bit aligned
+ *    (except possibly the last SGE's length)
+ * 3) The SGE addresses are suitable for passing to dma_map_single().
+ */
+int qib_sdma_verbs_send(struct qib_pportdata *ppd,
+			struct qib_sge_state *ss, u32 dwords,
+			struct qib_verbs_txreq *tx)
+{
+	unsigned long flags;
+	struct qib_sge *sge;
+	struct qib_qp *qp;
+	int ret = 0;
+	u16 tail;
+	__le64 *descqp;
+	u64 sdmadesc[2];
+	u32 dwoffset;
+	dma_addr_t addr;
+
+	spin_lock_irqsave(&ppd->sdma_lock, flags);
+
+retry:
+	if (unlikely(!__qib_sdma_running(ppd))) {
+		complete_sdma_err_req(ppd, tx);
+		goto unlock;
+	}
+
+	if (tx->txreq.sg_count > qib_sdma_descq_freecnt(ppd)) {
+		if (qib_sdma_make_progress(ppd))
+			goto retry;
+		if (ppd->dd->flags & QIB_HAS_SDMA_TIMEOUT)
+			ppd->dd->f_sdma_set_desc_cnt(ppd,
+					ppd->sdma_descq_cnt / 2);
+		goto busy;
+	}
+
+	dwoffset = tx->hdr_dwords;
+	make_sdma_desc(ppd, sdmadesc, (u64) tx->txreq.addr, dwoffset, 0);
+
+	sdmadesc[0] |= SDMA_DESC_FIRST;
+	if (tx->txreq.flags & QIB_SDMA_TXREQ_F_USELARGEBUF)
+		sdmadesc[0] |= SDMA_DESC_USE_LARGE_BUF;
+
+	/* write to the descq */
+	tail = ppd->sdma_descq_tail;
+	descqp = &ppd->sdma_descq[tail].qw[0];
+	*descqp++ = cpu_to_le64(sdmadesc[0]);
+	*descqp++ = cpu_to_le64(sdmadesc[1]);
+
+	/* increment the tail */
+	if (++tail == ppd->sdma_descq_cnt) {
+		tail = 0;
+		descqp = &ppd->sdma_descq[0].qw[0];
+		++ppd->sdma_generation;
+	}
+
+	tx->txreq.start_idx = tail;
+
+	sge = &ss->sge;
+	while (dwords) {
+		u32 dw;
+		u32 len;
+
+		len = dwords << 2;
+		if (len > sge->length)
+			len = sge->length;
+		if (len > sge->sge_length)
+			len = sge->sge_length;
+		BUG_ON(len == 0);
+		dw = (len + 3) >> 2;
+		addr = dma_map_single(&ppd->dd->pcidev->dev, sge->vaddr,
+				      dw << 2, DMA_TO_DEVICE);
+		if (dma_mapping_error(&ppd->dd->pcidev->dev, addr))
+			goto unmap;
+		sdmadesc[0] = 0;
+		make_sdma_desc(ppd, sdmadesc, (u64) addr, dw, dwoffset);
+		/* SDmaUseLargeBuf has to be set in every descriptor */
+		if (tx->txreq.flags & QIB_SDMA_TXREQ_F_USELARGEBUF)
+			sdmadesc[0] |= SDMA_DESC_USE_LARGE_BUF;
+		/* write to the descq */
+		*descqp++ = cpu_to_le64(sdmadesc[0]);
+		*descqp++ = cpu_to_le64(sdmadesc[1]);
+
+		/* increment the tail */
+		if (++tail == ppd->sdma_descq_cnt) {
+			tail = 0;
+			descqp = &ppd->sdma_descq[0].qw[0];
+			++ppd->sdma_generation;
+		}
+		sge->vaddr += len;
+		sge->length -= len;
+		sge->sge_length -= len;
+		if (sge->sge_length == 0) {
+			if (--ss->num_sge)
+				*sge = *ss->sg_list++;
+		} else if (sge->length == 0 && sge->mr->lkey) {
+			if (++sge->n >= QIB_SEGSZ) {
+				if (++sge->m >= sge->mr->mapsz)
+					break;
+				sge->n = 0;
+			}
+			sge->vaddr =
+				sge->mr->map[sge->m]->segs[sge->n].vaddr;
+			sge->length =
+				sge->mr->map[sge->m]->segs[sge->n].length;
+		}
+
+		dwoffset += dw;
+		dwords -= dw;
+	}
+
+	if (!tail)
+		descqp = &ppd->sdma_descq[ppd->sdma_descq_cnt].qw[0];
+	descqp -= 2;
+	descqp[0] |= cpu_to_le64(SDMA_DESC_LAST);
+	if (tx->txreq.flags & QIB_SDMA_TXREQ_F_HEADTOHOST)
+		descqp[0] |= cpu_to_le64(SDMA_DESC_DMA_HEAD);
+	if (tx->txreq.flags & QIB_SDMA_TXREQ_F_INTREQ)
+		descqp[0] |= cpu_to_le64(SDMA_DESC_INTR);
+
+	atomic_inc(&tx->qp->s_dma_busy);
+	tx->txreq.next_descq_idx = tail;
+	ppd->dd->f_sdma_update_tail(ppd, tail);
+	ppd->sdma_descq_added += tx->txreq.sg_count;
+	list_add_tail(&tx->txreq.list, &ppd->sdma_activelist);
+	goto unlock;
+
+unmap:
+	for (;;) {
+		if (!tail)
+			tail = ppd->sdma_descq_cnt - 1;
+		else
+			tail--;
+		if (tail == ppd->sdma_descq_tail)
+			break;
+		unmap_desc(ppd, tail);
+	}
+	qp = tx->qp;
+	qib_put_txreq(tx);
+	spin_lock(&qp->s_lock);
+	if (qp->ibqp.qp_type == IB_QPT_RC) {
+		/* XXX what about error sending RDMA read responses? */
+		if (ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK)
+			qib_error_qp(qp, IB_WC_GENERAL_ERR);
+	} else if (qp->s_wqe)
+		qib_send_complete(qp, qp->s_wqe, IB_WC_GENERAL_ERR);
+	spin_unlock(&qp->s_lock);
+	/* return zero to process the next send work request */
+	goto unlock;
+
+busy:
+	qp = tx->qp;
+	spin_lock(&qp->s_lock);
+	if (ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK) {
+		struct qib_ibdev *dev;
+
+		/*
+		 * If we couldn't queue the DMA request, save the info
+		 * and try again later rather than destroying the
+		 * buffer and undoing the side effects of the copy.
+		 */
+		tx->ss = ss;
+		tx->dwords = dwords;
+		qp->s_tx = tx;
+		dev = &ppd->dd->verbs_dev;
+		spin_lock(&dev->pending_lock);
+		if (list_empty(&qp->iowait)) {
+			struct qib_ibport *ibp;
+
+			ibp = &ppd->ibport_data;
+			ibp->n_dmawait++;
+			qp->s_flags |= QIB_S_WAIT_DMA_DESC;
+			list_add_tail(&qp->iowait, &dev->dmawait);
+		}
+		spin_unlock(&dev->pending_lock);
+		qp->s_flags &= ~QIB_S_BUSY;
+		spin_unlock(&qp->s_lock);
+		ret = -EBUSY;
+	} else {
+		spin_unlock(&qp->s_lock);
+		qib_put_txreq(tx);
+	}
+unlock:
+	spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+	return ret;
+}
+
+void qib_sdma_process_event(struct qib_pportdata *ppd,
+	enum qib_sdma_events event)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&ppd->sdma_lock, flags);
+
+	__qib_sdma_process_event(ppd, event);
+
+	if (ppd->sdma_state.current_state == qib_sdma_state_s99_running)
+		qib_verbs_sdma_desc_avail(ppd, qib_sdma_descq_freecnt(ppd));
+
+	spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+}
+
+void __qib_sdma_process_event(struct qib_pportdata *ppd,
+	enum qib_sdma_events event)
+{
+	struct qib_sdma_state *ss = &ppd->sdma_state;
+
+	switch (ss->current_state) {
+	case qib_sdma_state_s00_hw_down:
+		switch (event) {
+		case qib_sdma_event_e00_go_hw_down:
+			break;
+		case qib_sdma_event_e30_go_running:
+			/*
+			 * If down, but running requested (usually result
+			 * of link up, then we need to start up.
+			 * This can happen when hw down is requested while
+			 * bringing the link up with traffic active on
+			 * 7220, e.g. */
+			ss->go_s99_running = 1;
+			/* fall through and start dma engine */
+		case qib_sdma_event_e10_go_hw_start:
+			/* This reference means the state machine is started */
+			sdma_get(&ppd->sdma_state);
+			sdma_set_state(ppd,
+				       qib_sdma_state_s10_hw_start_up_wait);
+			break;
+		case qib_sdma_event_e20_hw_started:
+			break;
+		case qib_sdma_event_e40_sw_cleaned:
+			sdma_sw_tear_down(ppd);
+			break;
+		case qib_sdma_event_e50_hw_cleaned:
+			break;
+		case qib_sdma_event_e60_hw_halted:
+			break;
+		case qib_sdma_event_e70_go_idle:
+			break;
+		case qib_sdma_event_e7220_err_halted:
+			break;
+		case qib_sdma_event_e7322_err_halted:
+			break;
+		case qib_sdma_event_e90_timer_tick:
+			break;
+		}
+		break;
+
+	case qib_sdma_state_s10_hw_start_up_wait:
+		switch (event) {
+		case qib_sdma_event_e00_go_hw_down:
+			sdma_set_state(ppd, qib_sdma_state_s00_hw_down);
+			sdma_sw_tear_down(ppd);
+			break;
+		case qib_sdma_event_e10_go_hw_start:
+			break;
+		case qib_sdma_event_e20_hw_started:
+			sdma_set_state(ppd, ss->go_s99_running ?
+				       qib_sdma_state_s99_running :
+				       qib_sdma_state_s20_idle);
+			break;
+		case qib_sdma_event_e30_go_running:
+			ss->go_s99_running = 1;
+			break;
+		case qib_sdma_event_e40_sw_cleaned:
+			break;
+		case qib_sdma_event_e50_hw_cleaned:
+			break;
+		case qib_sdma_event_e60_hw_halted:
+			break;
+		case qib_sdma_event_e70_go_idle:
+			ss->go_s99_running = 0;
+			break;
+		case qib_sdma_event_e7220_err_halted:
+			break;
+		case qib_sdma_event_e7322_err_halted:
+			break;
+		case qib_sdma_event_e90_timer_tick:
+			break;
+		}
+		break;
+
+	case qib_sdma_state_s20_idle:
+		switch (event) {
+		case qib_sdma_event_e00_go_hw_down:
+			sdma_set_state(ppd, qib_sdma_state_s00_hw_down);
+			sdma_sw_tear_down(ppd);
+			break;
+		case qib_sdma_event_e10_go_hw_start:
+			break;
+		case qib_sdma_event_e20_hw_started:
+			break;
+		case qib_sdma_event_e30_go_running:
+			sdma_set_state(ppd, qib_sdma_state_s99_running);
+			ss->go_s99_running = 1;
+			break;
+		case qib_sdma_event_e40_sw_cleaned:
+			break;
+		case qib_sdma_event_e50_hw_cleaned:
+			break;
+		case qib_sdma_event_e60_hw_halted:
+			break;
+		case qib_sdma_event_e70_go_idle:
+			break;
+		case qib_sdma_event_e7220_err_halted:
+			break;
+		case qib_sdma_event_e7322_err_halted:
+			break;
+		case qib_sdma_event_e90_timer_tick:
+			break;
+		}
+		break;
+
+	case qib_sdma_state_s30_sw_clean_up_wait:
+		switch (event) {
+		case qib_sdma_event_e00_go_hw_down:
+			sdma_set_state(ppd, qib_sdma_state_s00_hw_down);
+			break;
+		case qib_sdma_event_e10_go_hw_start:
+			break;
+		case qib_sdma_event_e20_hw_started:
+			break;
+		case qib_sdma_event_e30_go_running:
+			ss->go_s99_running = 1;
+			break;
+		case qib_sdma_event_e40_sw_cleaned:
+			sdma_set_state(ppd,
+				       qib_sdma_state_s10_hw_start_up_wait);
+			sdma_hw_start_up(ppd);
+			break;
+		case qib_sdma_event_e50_hw_cleaned:
+			break;
+		case qib_sdma_event_e60_hw_halted:
+			break;
+		case qib_sdma_event_e70_go_idle:
+			ss->go_s99_running = 0;
+			break;
+		case qib_sdma_event_e7220_err_halted:
+			break;
+		case qib_sdma_event_e7322_err_halted:
+			break;
+		case qib_sdma_event_e90_timer_tick:
+			break;
+		}
+		break;
+
+	case qib_sdma_state_s40_hw_clean_up_wait:
+		switch (event) {
+		case qib_sdma_event_e00_go_hw_down:
+			sdma_set_state(ppd, qib_sdma_state_s00_hw_down);
+			sdma_start_sw_clean_up(ppd);
+			break;
+		case qib_sdma_event_e10_go_hw_start:
+			break;
+		case qib_sdma_event_e20_hw_started:
+			break;
+		case qib_sdma_event_e30_go_running:
+			ss->go_s99_running = 1;
+			break;
+		case qib_sdma_event_e40_sw_cleaned:
+			break;
+		case qib_sdma_event_e50_hw_cleaned:
+			sdma_set_state(ppd,
+				       qib_sdma_state_s30_sw_clean_up_wait);
+			sdma_start_sw_clean_up(ppd);
+			break;
+		case qib_sdma_event_e60_hw_halted:
+			break;
+		case qib_sdma_event_e70_go_idle:
+			ss->go_s99_running = 0;
+			break;
+		case qib_sdma_event_e7220_err_halted:
+			break;
+		case qib_sdma_event_e7322_err_halted:
+			break;
+		case qib_sdma_event_e90_timer_tick:
+			break;
+		}
+		break;
+
+	case qib_sdma_state_s50_hw_halt_wait:
+		switch (event) {
+		case qib_sdma_event_e00_go_hw_down:
+			sdma_set_state(ppd, qib_sdma_state_s00_hw_down);
+			sdma_start_sw_clean_up(ppd);
+			break;
+		case qib_sdma_event_e10_go_hw_start:
+			break;
+		case qib_sdma_event_e20_hw_started:
+			break;
+		case qib_sdma_event_e30_go_running:
+			ss->go_s99_running = 1;
+			break;
+		case qib_sdma_event_e40_sw_cleaned:
+			break;
+		case qib_sdma_event_e50_hw_cleaned:
+			break;
+		case qib_sdma_event_e60_hw_halted:
+			sdma_set_state(ppd,
+				       qib_sdma_state_s40_hw_clean_up_wait);
+			ppd->dd->f_sdma_hw_clean_up(ppd);
+			break;
+		case qib_sdma_event_e70_go_idle:
+			ss->go_s99_running = 0;
+			break;
+		case qib_sdma_event_e7220_err_halted:
+			break;
+		case qib_sdma_event_e7322_err_halted:
+			break;
+		case qib_sdma_event_e90_timer_tick:
+			break;
+		}
+		break;
+
+	case qib_sdma_state_s99_running:
+		switch (event) {
+		case qib_sdma_event_e00_go_hw_down:
+			sdma_set_state(ppd, qib_sdma_state_s00_hw_down);
+			sdma_start_sw_clean_up(ppd);
+			break;
+		case qib_sdma_event_e10_go_hw_start:
+			break;
+		case qib_sdma_event_e20_hw_started:
+			break;
+		case qib_sdma_event_e30_go_running:
+			break;
+		case qib_sdma_event_e40_sw_cleaned:
+			break;
+		case qib_sdma_event_e50_hw_cleaned:
+			break;
+		case qib_sdma_event_e60_hw_halted:
+			sdma_set_state(ppd,
+				       qib_sdma_state_s30_sw_clean_up_wait);
+			sdma_start_sw_clean_up(ppd);
+			break;
+		case qib_sdma_event_e70_go_idle:
+			sdma_set_state(ppd, qib_sdma_state_s50_hw_halt_wait);
+			ss->go_s99_running = 0;
+			break;
+		case qib_sdma_event_e7220_err_halted:
+			sdma_set_state(ppd,
+				       qib_sdma_state_s30_sw_clean_up_wait);
+			sdma_start_sw_clean_up(ppd);
+			break;
+		case qib_sdma_event_e7322_err_halted:
+			sdma_set_state(ppd, qib_sdma_state_s50_hw_halt_wait);
+			break;
+		case qib_sdma_event_e90_timer_tick:
+			break;
+		}
+		break;
+	}
+
+	ss->last_event = event;
+}

drivers/infiniband/hw/qib/qib_srq.c

@@ -0,0 +1,375 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/err.h>
+#include <linux/slab.h>
+#include <linux/vmalloc.h>
+
+#include "qib_verbs.h"
+
+/**
+ * qib_post_srq_receive - post a receive on a shared receive queue
+ * @ibsrq: the SRQ to post the receive on
+ * @wr: the list of work requests to post
+ * @bad_wr: A pointer to the first WR to cause a problem is put here
+ *
+ * This may be called from interrupt context.
+ */
+int qib_post_srq_receive(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
+			 struct ib_recv_wr **bad_wr)
+{
+	struct qib_srq *srq = to_isrq(ibsrq);
+	struct qib_rwq *wq;
+	unsigned long flags;
+	int ret;
+
+	for (; wr; wr = wr->next) {
+		struct qib_rwqe *wqe;
+		u32 next;
+		int i;
+
+		if ((unsigned) wr->num_sge > srq->rq.max_sge) {
+			*bad_wr = wr;
+			ret = -EINVAL;
+			goto bail;
+		}
+
+		spin_lock_irqsave(&srq->rq.lock, flags);
+		wq = srq->rq.wq;
+		next = wq->head + 1;
+		if (next >= srq->rq.size)
+			next = 0;
+		if (next == wq->tail) {
+			spin_unlock_irqrestore(&srq->rq.lock, flags);
+			*bad_wr = wr;
+			ret = -ENOMEM;
+			goto bail;
+		}
+
+		wqe = get_rwqe_ptr(&srq->rq, wq->head);
+		wqe->wr_id = wr->wr_id;
+		wqe->num_sge = wr->num_sge;
+		for (i = 0; i < wr->num_sge; i++)
+			wqe->sg_list[i] = wr->sg_list[i];
+		/* Make sure queue entry is written before the head index. */
+		smp_wmb();
+		wq->head = next;
+		spin_unlock_irqrestore(&srq->rq.lock, flags);
+	}
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+/**
+ * qib_create_srq - create a shared receive queue
+ * @ibpd: the protection domain of the SRQ to create
+ * @srq_init_attr: the attributes of the SRQ
+ * @udata: data from libibverbs when creating a user SRQ
+ */
+struct ib_srq *qib_create_srq(struct ib_pd *ibpd,
+			      struct ib_srq_init_attr *srq_init_attr,
+			      struct ib_udata *udata)
+{
+	struct qib_ibdev *dev = to_idev(ibpd->device);
+	struct qib_srq *srq;
+	u32 sz;
+	struct ib_srq *ret;
+
+	if (srq_init_attr->attr.max_sge == 0 ||
+	    srq_init_attr->attr.max_sge > ib_qib_max_srq_sges ||
+	    srq_init_attr->attr.max_wr == 0 ||
+	    srq_init_attr->attr.max_wr > ib_qib_max_srq_wrs) {
+		ret = ERR_PTR(-EINVAL);
+		goto done;
+	}
+
+	srq = kmalloc(sizeof(*srq), GFP_KERNEL);
+	if (!srq) {
+		ret = ERR_PTR(-ENOMEM);
+		goto done;
+	}
+
+	/*
+	 * Need to use vmalloc() if we want to support large #s of entries.
+	 */
+	srq->rq.size = srq_init_attr->attr.max_wr + 1;
+	srq->rq.max_sge = srq_init_attr->attr.max_sge;
+	sz = sizeof(struct ib_sge) * srq->rq.max_sge +
+		sizeof(struct qib_rwqe);
+	srq->rq.wq = vmalloc_user(sizeof(struct qib_rwq) + srq->rq.size * sz);
+	if (!srq->rq.wq) {
+		ret = ERR_PTR(-ENOMEM);
+		goto bail_srq;
+	}
+
+	/*
+	 * Return the address of the RWQ as the offset to mmap.
+	 * See qib_mmap() for details.
+	 */
+	if (udata && udata->outlen >= sizeof(__u64)) {
+		int err;
+		u32 s = sizeof(struct qib_rwq) + srq->rq.size * sz;
+
+		srq->ip =
+		    qib_create_mmap_info(dev, s, ibpd->uobject->context,
+					 srq->rq.wq);
+		if (!srq->ip) {
+			ret = ERR_PTR(-ENOMEM);
+			goto bail_wq;
+		}
+
+		err = ib_copy_to_udata(udata, &srq->ip->offset,
+				       sizeof(srq->ip->offset));
+		if (err) {
+			ret = ERR_PTR(err);
+			goto bail_ip;
+		}
+	} else
+		srq->ip = NULL;
+
+	/*
+	 * ib_create_srq() will initialize srq->ibsrq.
+	 */
+	spin_lock_init(&srq->rq.lock);
+	srq->rq.wq->head = 0;
+	srq->rq.wq->tail = 0;
+	srq->limit = srq_init_attr->attr.srq_limit;
+
+	spin_lock(&dev->n_srqs_lock);
+	if (dev->n_srqs_allocated == ib_qib_max_srqs) {
+		spin_unlock(&dev->n_srqs_lock);
+		ret = ERR_PTR(-ENOMEM);
+		goto bail_ip;
+	}
+
+	dev->n_srqs_allocated++;
+	spin_unlock(&dev->n_srqs_lock);
+
+	if (srq->ip) {
+		spin_lock_irq(&dev->pending_lock);
+		list_add(&srq->ip->pending_mmaps, &dev->pending_mmaps);
+		spin_unlock_irq(&dev->pending_lock);
+	}
+
+	ret = &srq->ibsrq;
+	goto done;
+
+bail_ip:
+	kfree(srq->ip);
+bail_wq:
+	vfree(srq->rq.wq);
+bail_srq:
+	kfree(srq);
+done:
+	return ret;
+}
+
+/**
+ * qib_modify_srq - modify a shared receive queue
+ * @ibsrq: the SRQ to modify
+ * @attr: the new attributes of the SRQ
+ * @attr_mask: indicates which attributes to modify
+ * @udata: user data for libibverbs.so
+ */
+int qib_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
+		   enum ib_srq_attr_mask attr_mask,
+		   struct ib_udata *udata)
+{
+	struct qib_srq *srq = to_isrq(ibsrq);
+	struct qib_rwq *wq;
+	int ret = 0;
+
+	if (attr_mask & IB_SRQ_MAX_WR) {
+		struct qib_rwq *owq;
+		struct qib_rwqe *p;
+		u32 sz, size, n, head, tail;
+
+		/* Check that the requested sizes are below the limits. */
+		if ((attr->max_wr > ib_qib_max_srq_wrs) ||
+		    ((attr_mask & IB_SRQ_LIMIT) ?
+		     attr->srq_limit : srq->limit) > attr->max_wr) {
+			ret = -EINVAL;
+			goto bail;
+		}
+
+		sz = sizeof(struct qib_rwqe) +
+			srq->rq.max_sge * sizeof(struct ib_sge);
+		size = attr->max_wr + 1;
+		wq = vmalloc_user(sizeof(struct qib_rwq) + size * sz);
+		if (!wq) {
+			ret = -ENOMEM;
+			goto bail;
+		}
+
+		/* Check that we can write the offset to mmap. */
+		if (udata && udata->inlen >= sizeof(__u64)) {
+			__u64 offset_addr;
+			__u64 offset = 0;
+
+			ret = ib_copy_from_udata(&offset_addr, udata,
+						 sizeof(offset_addr));
+			if (ret)
+				goto bail_free;
+			udata->outbuf =
+				(void __user *) (unsigned long) offset_addr;
+			ret = ib_copy_to_udata(udata, &offset,
+					       sizeof(offset));
+			if (ret)
+				goto bail_free;
+		}
+
+		spin_lock_irq(&srq->rq.lock);
+		/*
+		 * validate head and tail pointer values and compute
+		 * the number of remaining WQEs.
+		 */
+		owq = srq->rq.wq;
+		head = owq->head;
+		tail = owq->tail;
+		if (head >= srq->rq.size || tail >= srq->rq.size) {
+			ret = -EINVAL;
+			goto bail_unlock;
+		}
+		n = head;
+		if (n < tail)
+			n += srq->rq.size - tail;
+		else
+			n -= tail;
+		if (size <= n) {
+			ret = -EINVAL;
+			goto bail_unlock;
+		}
+		n = 0;
+		p = wq->wq;
+		while (tail != head) {
+			struct qib_rwqe *wqe;
+			int i;
+
+			wqe = get_rwqe_ptr(&srq->rq, tail);
+			p->wr_id = wqe->wr_id;
+			p->num_sge = wqe->num_sge;
+			for (i = 0; i < wqe->num_sge; i++)
+				p->sg_list[i] = wqe->sg_list[i];
+			n++;
+			p = (struct qib_rwqe *)((char *) p + sz);
+			if (++tail >= srq->rq.size)
+				tail = 0;
+		}
+		srq->rq.wq = wq;
+		srq->rq.size = size;
+		wq->head = n;
+		wq->tail = 0;
+		if (attr_mask & IB_SRQ_LIMIT)
+			srq->limit = attr->srq_limit;
+		spin_unlock_irq(&srq->rq.lock);
+
+		vfree(owq);
+
+		if (srq->ip) {
+			struct qib_mmap_info *ip = srq->ip;
+			struct qib_ibdev *dev = to_idev(srq->ibsrq.device);
+			u32 s = sizeof(struct qib_rwq) + size * sz;
+
+			qib_update_mmap_info(dev, ip, s, wq);
+
+			/*
+			 * Return the offset to mmap.
+			 * See qib_mmap() for details.
+			 */
+			if (udata && udata->inlen >= sizeof(__u64)) {
+				ret = ib_copy_to_udata(udata, &ip->offset,
+						       sizeof(ip->offset));
+				if (ret)
+					goto bail;
+			}
+
+			/*
+			 * Put user mapping info onto the pending list
+			 * unless it already is on the list.
+			 */
+			spin_lock_irq(&dev->pending_lock);
+			if (list_empty(&ip->pending_mmaps))
+				list_add(&ip->pending_mmaps,
+					 &dev->pending_mmaps);
+			spin_unlock_irq(&dev->pending_lock);
+		}
+	} else if (attr_mask & IB_SRQ_LIMIT) {
+		spin_lock_irq(&srq->rq.lock);
+		if (attr->srq_limit >= srq->rq.size)
+			ret = -EINVAL;
+		else
+			srq->limit = attr->srq_limit;
+		spin_unlock_irq(&srq->rq.lock);
+	}
+	goto bail;
+
+bail_unlock:
+	spin_unlock_irq(&srq->rq.lock);
+bail_free:
+	vfree(wq);
+bail:
+	return ret;
+}
+
+int qib_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr)
+{
+	struct qib_srq *srq = to_isrq(ibsrq);
+
+	attr->max_wr = srq->rq.size - 1;
+	attr->max_sge = srq->rq.max_sge;
+	attr->srq_limit = srq->limit;
+	return 0;
+}
+
+/**
+ * qib_destroy_srq - destroy a shared receive queue
+ * @ibsrq: the SRQ to destroy
+ */
+int qib_destroy_srq(struct ib_srq *ibsrq)
+{
+	struct qib_srq *srq = to_isrq(ibsrq);
+	struct qib_ibdev *dev = to_idev(ibsrq->device);
+
+	spin_lock(&dev->n_srqs_lock);
+	dev->n_srqs_allocated--;
+	spin_unlock(&dev->n_srqs_lock);
+	if (srq->ip)
+		kref_put(&srq->ip->ref, qib_release_mmap_info);
+	else
+		vfree(srq->rq.wq);
+	kfree(srq);
+
+	return 0;
+}

drivers/infiniband/hw/qib/qib_sysfs.c

@@ -0,0 +1,691 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include <linux/ctype.h>
+
+#include "qib.h"
+
+/**
+ * qib_parse_ushort - parse an unsigned short value in an arbitrary base
+ * @str: the string containing the number
+ * @valp: where to put the result
+ *
+ * Returns the number of bytes consumed, or negative value on error.
+ */
+static int qib_parse_ushort(const char *str, unsigned short *valp)
+{
+	unsigned long val;
+	char *end;
+	int ret;
+
+	if (!isdigit(str[0])) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	val = simple_strtoul(str, &end, 0);
+
+	if (val > 0xffff) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	*valp = val;
+
+	ret = end + 1 - str;
+	if (ret == 0)
+		ret = -EINVAL;
+
+bail:
+	return ret;
+}
+
+/* start of per-port functions */
+/*
+ * Get/Set heartbeat enable. OR of 1=enabled, 2=auto
+ */
+static ssize_t show_hrtbt_enb(struct qib_pportdata *ppd, char *buf)
+{
+	struct qib_devdata *dd = ppd->dd;
+	int ret;
+
+	ret = dd->f_get_ib_cfg(ppd, QIB_IB_CFG_HRTBT);
+	ret = scnprintf(buf, PAGE_SIZE, "%d\n", ret);
+	return ret;
+}
+
+static ssize_t store_hrtbt_enb(struct qib_pportdata *ppd, const char *buf,
+			       size_t count)
+{
+	struct qib_devdata *dd = ppd->dd;
+	int ret;
+	u16 val;
+
+	ret = qib_parse_ushort(buf, &val);
+
+	/*
+	 * Set the "intentional" heartbeat enable per either of
+	 * "Enable" and "Auto", as these are normally set together.
+	 * This bit is consulted when leaving loopback mode,
+	 * because entering loopback mode overrides it and automatically
+	 * disables heartbeat.
+	 */
+	if (ret >= 0)
+		ret = dd->f_set_ib_cfg(ppd, QIB_IB_CFG_HRTBT, val);
+	if (ret < 0)
+		qib_dev_err(dd, "attempt to set invalid Heartbeat enable\n");
+	return ret < 0 ? ret : count;
+}
+
+static ssize_t store_loopback(struct qib_pportdata *ppd, const char *buf,
+			      size_t count)
+{
+	struct qib_devdata *dd = ppd->dd;
+	int ret = count, r;
+
+	r = dd->f_set_ib_loopback(ppd, buf);
+	if (r < 0)
+		ret = r;
+
+	return ret;
+}
+
+static ssize_t store_led_override(struct qib_pportdata *ppd, const char *buf,
+				  size_t count)
+{
+	struct qib_devdata *dd = ppd->dd;
+	int ret;
+	u16 val;
+
+	ret = qib_parse_ushort(buf, &val);
+	if (ret > 0)
+		qib_set_led_override(ppd, val);
+	else
+		qib_dev_err(dd, "attempt to set invalid LED override\n");
+	return ret < 0 ? ret : count;
+}
+
+static ssize_t show_status(struct qib_pportdata *ppd, char *buf)
+{
+	ssize_t ret;
+
+	if (!ppd->statusp)
+		ret = -EINVAL;
+	else
+		ret = scnprintf(buf, PAGE_SIZE, "0x%llx\n",
+				(unsigned long long) *(ppd->statusp));
+	return ret;
+}
+
+/*
+ * For userland compatibility, these offsets must remain fixed.
+ * They are strings for QIB_STATUS_*
+ */
+static const char *qib_status_str[] = {
+	"Initted",
+	"",
+	"",
+	"",
+	"",
+	"Present",
+	"IB_link_up",
+	"IB_configured",
+	"",
+	"Fatal_Hardware_Error",
+	NULL,
+};
+
+static ssize_t show_status_str(struct qib_pportdata *ppd, char *buf)
+{
+	int i, any;
+	u64 s;
+	ssize_t ret;
+
+	if (!ppd->statusp) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	s = *(ppd->statusp);
+	*buf = '\0';
+	for (any = i = 0; s && qib_status_str[i]; i++) {
+		if (s & 1) {
+			/* if overflow */
+			if (any && strlcat(buf, " ", PAGE_SIZE) >= PAGE_SIZE)
+				break;
+			if (strlcat(buf, qib_status_str[i], PAGE_SIZE) >=
+					PAGE_SIZE)
+				break;
+			any = 1;
+		}
+		s >>= 1;
+	}
+	if (any)
+		strlcat(buf, "\n", PAGE_SIZE);
+
+	ret = strlen(buf);
+
+bail:
+	return ret;
+}
+
+/* end of per-port functions */
+
+/*
+ * Start of per-port file structures and support code
+ * Because we are fitting into other infrastructure, we have to supply the
+ * full set of kobject/sysfs_ops structures and routines.
+ */
+#define QIB_PORT_ATTR(name, mode, show, store) \
+	static struct qib_port_attr qib_port_attr_##name = \
+		__ATTR(name, mode, show, store)
+
+struct qib_port_attr {
+	struct attribute attr;
+	ssize_t (*show)(struct qib_pportdata *, char *);
+	ssize_t (*store)(struct qib_pportdata *, const char *, size_t);
+};
+
+QIB_PORT_ATTR(loopback, S_IWUSR, NULL, store_loopback);
+QIB_PORT_ATTR(led_override, S_IWUSR, NULL, store_led_override);
+QIB_PORT_ATTR(hrtbt_enable, S_IWUSR | S_IRUGO, show_hrtbt_enb,
+	      store_hrtbt_enb);
+QIB_PORT_ATTR(status, S_IRUGO, show_status, NULL);
+QIB_PORT_ATTR(status_str, S_IRUGO, show_status_str, NULL);
+
+static struct attribute *port_default_attributes[] = {
+	&qib_port_attr_loopback.attr,
+	&qib_port_attr_led_override.attr,
+	&qib_port_attr_hrtbt_enable.attr,
+	&qib_port_attr_status.attr,
+	&qib_port_attr_status_str.attr,
+	NULL
+};
+
+static ssize_t qib_portattr_show(struct kobject *kobj,
+	struct attribute *attr, char *buf)
+{
+	struct qib_port_attr *pattr =
+		container_of(attr, struct qib_port_attr, attr);
+	struct qib_pportdata *ppd =
+		container_of(kobj, struct qib_pportdata, pport_kobj);
+
+	return pattr->show(ppd, buf);
+}
+
+static ssize_t qib_portattr_store(struct kobject *kobj,
+	struct attribute *attr, const char *buf, size_t len)
+{
+	struct qib_port_attr *pattr =
+		container_of(attr, struct qib_port_attr, attr);
+	struct qib_pportdata *ppd =
+		container_of(kobj, struct qib_pportdata, pport_kobj);
+
+	return pattr->store(ppd, buf, len);
+}
+
+static void qib_port_release(struct kobject *kobj)
+{
+	/* nothing to do since memory is freed by qib_free_devdata() */
+}
+
+static const struct sysfs_ops qib_port_ops = {
+	.show = qib_portattr_show,
+	.store = qib_portattr_store,
+};
+
+static struct kobj_type qib_port_ktype = {
+	.release = qib_port_release,
+	.sysfs_ops = &qib_port_ops,
+	.default_attrs = port_default_attributes
+};
+
+/* Start sl2vl */
+
+#define QIB_SL2VL_ATTR(N) \
+	static struct qib_sl2vl_attr qib_sl2vl_attr_##N = { \
+		.attr = { .name = __stringify(N), .mode = 0444 }, \
+		.sl = N \
+	}
+
+struct qib_sl2vl_attr {
+	struct attribute attr;
+	int sl;
+};
+
+QIB_SL2VL_ATTR(0);
+QIB_SL2VL_ATTR(1);
+QIB_SL2VL_ATTR(2);
+QIB_SL2VL_ATTR(3);
+QIB_SL2VL_ATTR(4);
+QIB_SL2VL_ATTR(5);
+QIB_SL2VL_ATTR(6);
+QIB_SL2VL_ATTR(7);
+QIB_SL2VL_ATTR(8);
+QIB_SL2VL_ATTR(9);
+QIB_SL2VL_ATTR(10);
+QIB_SL2VL_ATTR(11);
+QIB_SL2VL_ATTR(12);
+QIB_SL2VL_ATTR(13);
+QIB_SL2VL_ATTR(14);
+QIB_SL2VL_ATTR(15);
+
+static struct attribute *sl2vl_default_attributes[] = {
+	&qib_sl2vl_attr_0.attr,
+	&qib_sl2vl_attr_1.attr,
+	&qib_sl2vl_attr_2.attr,
+	&qib_sl2vl_attr_3.attr,
+	&qib_sl2vl_attr_4.attr,
+	&qib_sl2vl_attr_5.attr,
+	&qib_sl2vl_attr_6.attr,
+	&qib_sl2vl_attr_7.attr,
+	&qib_sl2vl_attr_8.attr,
+	&qib_sl2vl_attr_9.attr,
+	&qib_sl2vl_attr_10.attr,
+	&qib_sl2vl_attr_11.attr,
+	&qib_sl2vl_attr_12.attr,
+	&qib_sl2vl_attr_13.attr,
+	&qib_sl2vl_attr_14.attr,
+	&qib_sl2vl_attr_15.attr,
+	NULL
+};
+
+static ssize_t sl2vl_attr_show(struct kobject *kobj, struct attribute *attr,
+			       char *buf)
+{
+	struct qib_sl2vl_attr *sattr =
+		container_of(attr, struct qib_sl2vl_attr, attr);
+	struct qib_pportdata *ppd =
+		container_of(kobj, struct qib_pportdata, sl2vl_kobj);
+	struct qib_ibport *qibp = &ppd->ibport_data;
+
+	return sprintf(buf, "%u\n", qibp->sl_to_vl[sattr->sl]);
+}
+
+static const struct sysfs_ops qib_sl2vl_ops = {
+	.show = sl2vl_attr_show,
+};
+
+static struct kobj_type qib_sl2vl_ktype = {
+	.release = qib_port_release,
+	.sysfs_ops = &qib_sl2vl_ops,
+	.default_attrs = sl2vl_default_attributes
+};
+
+/* End sl2vl */
+
+/* Start diag_counters */
+
+#define QIB_DIAGC_ATTR(N) \
+	static struct qib_diagc_attr qib_diagc_attr_##N = { \
+		.attr = { .name = __stringify(N), .mode = 0444 }, \
+		.counter = offsetof(struct qib_ibport, n_##N) \
+	}
+
+struct qib_diagc_attr {
+	struct attribute attr;
+	size_t counter;
+};
+
+QIB_DIAGC_ATTR(rc_resends);
+QIB_DIAGC_ATTR(rc_acks);
+QIB_DIAGC_ATTR(rc_qacks);
+QIB_DIAGC_ATTR(rc_delayed_comp);
+QIB_DIAGC_ATTR(seq_naks);
+QIB_DIAGC_ATTR(rdma_seq);
+QIB_DIAGC_ATTR(rnr_naks);
+QIB_DIAGC_ATTR(other_naks);
+QIB_DIAGC_ATTR(rc_timeouts);
+QIB_DIAGC_ATTR(loop_pkts);
+QIB_DIAGC_ATTR(pkt_drops);
+QIB_DIAGC_ATTR(dmawait);
+QIB_DIAGC_ATTR(unaligned);
+QIB_DIAGC_ATTR(rc_dupreq);
+QIB_DIAGC_ATTR(rc_seqnak);
+
+static struct attribute *diagc_default_attributes[] = {
+	&qib_diagc_attr_rc_resends.attr,
+	&qib_diagc_attr_rc_acks.attr,
+	&qib_diagc_attr_rc_qacks.attr,
+	&qib_diagc_attr_rc_delayed_comp.attr,
+	&qib_diagc_attr_seq_naks.attr,
+	&qib_diagc_attr_rdma_seq.attr,
+	&qib_diagc_attr_rnr_naks.attr,
+	&qib_diagc_attr_other_naks.attr,
+	&qib_diagc_attr_rc_timeouts.attr,
+	&qib_diagc_attr_loop_pkts.attr,
+	&qib_diagc_attr_pkt_drops.attr,
+	&qib_diagc_attr_dmawait.attr,
+	&qib_diagc_attr_unaligned.attr,
+	&qib_diagc_attr_rc_dupreq.attr,
+	&qib_diagc_attr_rc_seqnak.attr,
+	NULL
+};
+
+static ssize_t diagc_attr_show(struct kobject *kobj, struct attribute *attr,
+			       char *buf)
+{
+	struct qib_diagc_attr *dattr =
+		container_of(attr, struct qib_diagc_attr, attr);
+	struct qib_pportdata *ppd =
+		container_of(kobj, struct qib_pportdata, diagc_kobj);
+	struct qib_ibport *qibp = &ppd->ibport_data;
+
+	return sprintf(buf, "%u\n", *(u32 *)((char *)qibp + dattr->counter));
+}
+
+static const struct sysfs_ops qib_diagc_ops = {
+	.show = diagc_attr_show,
+};
+
+static struct kobj_type qib_diagc_ktype = {
+	.release = qib_port_release,
+	.sysfs_ops = &qib_diagc_ops,
+	.default_attrs = diagc_default_attributes
+};
+
+/* End diag_counters */
+
+/* end of per-port file structures and support code */
+
+/*
+ * Start of per-unit (or driver, in some cases, but replicated
+ * per unit) functions (these get a device *)
+ */
+static ssize_t show_rev(struct device *device, struct device_attribute *attr,
+			char *buf)
+{
+	struct qib_ibdev *dev =
+		container_of(device, struct qib_ibdev, ibdev.dev);
+
+	return sprintf(buf, "%x\n", dd_from_dev(dev)->minrev);
+}
+
+static ssize_t show_hca(struct device *device, struct device_attribute *attr,
+			char *buf)
+{
+	struct qib_ibdev *dev =
+		container_of(device, struct qib_ibdev, ibdev.dev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+	int ret;
+
+	if (!dd->boardname)
+		ret = -EINVAL;
+	else
+		ret = scnprintf(buf, PAGE_SIZE, "%s\n", dd->boardname);
+	return ret;
+}
+
+static ssize_t show_version(struct device *device,
+			    struct device_attribute *attr, char *buf)
+{
+	/* The string printed here is already newline-terminated. */
+	return scnprintf(buf, PAGE_SIZE, "%s", (char *)ib_qib_version);
+}
+
+static ssize_t show_boardversion(struct device *device,
+				 struct device_attribute *attr, char *buf)
+{
+	struct qib_ibdev *dev =
+		container_of(device, struct qib_ibdev, ibdev.dev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+
+	/* The string printed here is already newline-terminated. */
+	return scnprintf(buf, PAGE_SIZE, "%s", dd->boardversion);
+}
+
+
+static ssize_t show_localbus_info(struct device *device,
+				  struct device_attribute *attr, char *buf)
+{
+	struct qib_ibdev *dev =
+		container_of(device, struct qib_ibdev, ibdev.dev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+
+	/* The string printed here is already newline-terminated. */
+	return scnprintf(buf, PAGE_SIZE, "%s", dd->lbus_info);
+}
+
+
+static ssize_t show_nctxts(struct device *device,
+			   struct device_attribute *attr, char *buf)
+{
+	struct qib_ibdev *dev =
+		container_of(device, struct qib_ibdev, ibdev.dev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+
+	/* Return the number of user ports (contexts) available. */
+	return scnprintf(buf, PAGE_SIZE, "%u\n", dd->cfgctxts -
+		dd->first_user_ctxt);
+}
+
+static ssize_t show_serial(struct device *device,
+			   struct device_attribute *attr, char *buf)
+{
+	struct qib_ibdev *dev =
+		container_of(device, struct qib_ibdev, ibdev.dev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+
+	buf[sizeof dd->serial] = '\0';
+	memcpy(buf, dd->serial, sizeof dd->serial);
+	strcat(buf, "\n");
+	return strlen(buf);
+}
+
+static ssize_t store_chip_reset(struct device *device,
+				struct device_attribute *attr, const char *buf,
+				size_t count)
+{
+	struct qib_ibdev *dev =
+		container_of(device, struct qib_ibdev, ibdev.dev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+	int ret;
+
+	if (count < 5 || memcmp(buf, "reset", 5) || !dd->diag_client) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	ret = qib_reset_device(dd->unit);
+bail:
+	return ret < 0 ? ret : count;
+}
+
+static ssize_t show_logged_errs(struct device *device,
+				struct device_attribute *attr, char *buf)
+{
+	struct qib_ibdev *dev =
+		container_of(device, struct qib_ibdev, ibdev.dev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+	int idx, count;
+
+	/* force consistency with actual EEPROM */
+	if (qib_update_eeprom_log(dd) != 0)
+		return -ENXIO;
+
+	count = 0;
+	for (idx = 0; idx < QIB_EEP_LOG_CNT; ++idx) {
+		count += scnprintf(buf + count, PAGE_SIZE - count, "%d%c",
+				   dd->eep_st_errs[idx],
+				   idx == (QIB_EEP_LOG_CNT - 1) ? '\n' : ' ');
+	}
+
+	return count;
+}
+
+/*
+ * Dump tempsense regs. in decimal, to ease shell-scripts.
+ */
+static ssize_t show_tempsense(struct device *device,
+			      struct device_attribute *attr, char *buf)
+{
+	struct qib_ibdev *dev =
+		container_of(device, struct qib_ibdev, ibdev.dev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+	int ret;
+	int idx;
+	u8 regvals[8];
+
+	ret = -ENXIO;
+	for (idx = 0; idx < 8; ++idx) {
+		if (idx == 6)
+			continue;
+		ret = dd->f_tempsense_rd(dd, idx);
+		if (ret < 0)
+			break;
+		regvals[idx] = ret;
+	}
+	if (idx == 8)
+		ret = scnprintf(buf, PAGE_SIZE, "%d %d %02X %02X %d %d\n",
+				*(signed char *)(regvals),
+				*(signed char *)(regvals + 1),
+				regvals[2], regvals[3],
+				*(signed char *)(regvals + 5),
+				*(signed char *)(regvals + 7));
+	return ret;
+}
+
+/*
+ * end of per-unit (or driver, in some cases, but replicated
+ * per unit) functions
+ */
+
+/* start of per-unit file structures and support code */
+static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
+static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
+static DEVICE_ATTR(board_id, S_IRUGO, show_hca, NULL);
+static DEVICE_ATTR(version, S_IRUGO, show_version, NULL);
+static DEVICE_ATTR(nctxts, S_IRUGO, show_nctxts, NULL);
+static DEVICE_ATTR(serial, S_IRUGO, show_serial, NULL);
+static DEVICE_ATTR(boardversion, S_IRUGO, show_boardversion, NULL);
+static DEVICE_ATTR(logged_errors, S_IRUGO, show_logged_errs, NULL);
+static DEVICE_ATTR(tempsense, S_IRUGO, show_tempsense, NULL);
+static DEVICE_ATTR(localbus_info, S_IRUGO, show_localbus_info, NULL);
+static DEVICE_ATTR(chip_reset, S_IWUSR, NULL, store_chip_reset);
+
+static struct device_attribute *qib_attributes[] = {
+	&dev_attr_hw_rev,
+	&dev_attr_hca_type,
+	&dev_attr_board_id,
+	&dev_attr_version,
+	&dev_attr_nctxts,
+	&dev_attr_serial,
+	&dev_attr_boardversion,
+	&dev_attr_logged_errors,
+	&dev_attr_tempsense,
+	&dev_attr_localbus_info,
+	&dev_attr_chip_reset,
+};
+
+int qib_create_port_files(struct ib_device *ibdev, u8 port_num,
+			  struct kobject *kobj)
+{
+	struct qib_pportdata *ppd;
+	struct qib_devdata *dd = dd_from_ibdev(ibdev);
+	int ret;
+
+	if (!port_num || port_num > dd->num_pports) {
+		qib_dev_err(dd, "Skipping infiniband class with "
+			    "invalid port %u\n", port_num);
+		ret = -ENODEV;
+		goto bail;
+	}
+	ppd = &dd->pport[port_num - 1];
+
+	ret = kobject_init_and_add(&ppd->pport_kobj, &qib_port_ktype, kobj,
+				   "linkcontrol");
+	if (ret) {
+		qib_dev_err(dd, "Skipping linkcontrol sysfs info, "
+			    "(err %d) port %u\n", ret, port_num);
+		goto bail;
+	}
+	kobject_uevent(&ppd->pport_kobj, KOBJ_ADD);
+
+	ret = kobject_init_and_add(&ppd->sl2vl_kobj, &qib_sl2vl_ktype, kobj,
+				   "sl2vl");
+	if (ret) {
+		qib_dev_err(dd, "Skipping sl2vl sysfs info, "
+			    "(err %d) port %u\n", ret, port_num);
+		goto bail_sl;
+	}
+	kobject_uevent(&ppd->sl2vl_kobj, KOBJ_ADD);
+
+	ret = kobject_init_and_add(&ppd->diagc_kobj, &qib_diagc_ktype, kobj,
+				   "diag_counters");
+	if (ret) {
+		qib_dev_err(dd, "Skipping diag_counters sysfs info, "
+			    "(err %d) port %u\n", ret, port_num);
+		goto bail_diagc;
+	}
+	kobject_uevent(&ppd->diagc_kobj, KOBJ_ADD);
+
+	return 0;
+
+bail_diagc:
+	kobject_put(&ppd->sl2vl_kobj);
+bail_sl:
+	kobject_put(&ppd->pport_kobj);
+bail:
+	return ret;
+}
+
+/*
+ * Register and create our files in /sys/class/infiniband.
+ */
+int qib_verbs_register_sysfs(struct qib_devdata *dd)
+{
+	struct ib_device *dev = &dd->verbs_dev.ibdev;
+	int i, ret;
+
+	for (i = 0; i < ARRAY_SIZE(qib_attributes); ++i) {
+		ret = device_create_file(&dev->dev, qib_attributes[i]);
+		if (ret)
+			return ret;
+	}
+
+	return 0;
+}
+
+/*
+ * Unregister and remove our files in /sys/class/infiniband.
+ */
+void qib_verbs_unregister_sysfs(struct qib_devdata *dd)
+{
+	struct qib_pportdata *ppd;
+	int i;
+
+	for (i = 0; i < dd->num_pports; i++) {
+		ppd = &dd->pport[i];
+		kobject_put(&ppd->pport_kobj);
+		kobject_put(&ppd->sl2vl_kobj);
+	}
+}

drivers/infiniband/hw/qib/qib_twsi.c

@@ -0,0 +1,498 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/delay.h>
+#include <linux/pci.h>
+#include <linux/vmalloc.h>
+
+#include "qib.h"
+
+/*
+ * QLogic_IB "Two Wire Serial Interface" driver.
+ * Originally written for a not-quite-i2c serial eeprom, which is
+ * still used on some supported boards. Later boards have added a
+ * variety of other uses, most board-specific, so teh bit-boffing
+ * part has been split off to this file, while the other parts
+ * have been moved to chip-specific files.
+ *
+ * We have also dropped all pretense of fully generic (e.g. pretend
+ * we don't know whether '1' is the higher voltage) interface, as
+ * the restrictions of the generic i2c interface (e.g. no access from
+ * driver itself) make it unsuitable for this use.
+ */
+
+#define READ_CMD 1
+#define WRITE_CMD 0
+
+/**
+ * i2c_wait_for_writes - wait for a write
+ * @dd: the qlogic_ib device
+ *
+ * We use this instead of udelay directly, so we can make sure
+ * that previous register writes have been flushed all the way
+ * to the chip.  Since we are delaying anyway, the cost doesn't
+ * hurt, and makes the bit twiddling more regular
+ */
+static void i2c_wait_for_writes(struct qib_devdata *dd)
+{
+	/*
+	 * implicit read of EXTStatus is as good as explicit
+	 * read of scratch, if all we want to do is flush
+	 * writes.
+	 */
+	dd->f_gpio_mod(dd, 0, 0, 0);
+	rmb(); /* inlined, so prevent compiler reordering */
+}
+
+/*
+ * QSFP modules are allowed to hold SCL low for 500uSec. Allow twice that
+ * for "almost compliant" modules
+ */
+#define SCL_WAIT_USEC 1000
+
+/* BUF_WAIT is time bus must be free between STOP or ACK and to next START.
+ * Should be 20, but some chips need more.
+ */
+#define TWSI_BUF_WAIT_USEC 60
+
+static void scl_out(struct qib_devdata *dd, u8 bit)
+{
+	u32 mask;
+
+	udelay(1);
+
+	mask = 1UL << dd->gpio_scl_num;
+
+	/* SCL is meant to be bare-drain, so never set "OUT", just DIR */
+	dd->f_gpio_mod(dd, 0, bit ? 0 : mask, mask);
+
+	/*
+	 * Allow for slow slaves by simple
+	 * delay for falling edge, sampling on rise.
+	 */
+	if (!bit)
+		udelay(2);
+	else {
+		int rise_usec;
+		for (rise_usec = SCL_WAIT_USEC; rise_usec > 0; rise_usec -= 2) {
+			if (mask & dd->f_gpio_mod(dd, 0, 0, 0))
+				break;
+			udelay(2);
+		}
+		if (rise_usec <= 0)
+			qib_dev_err(dd, "SCL interface stuck low > %d uSec\n",
+				    SCL_WAIT_USEC);
+	}
+	i2c_wait_for_writes(dd);
+}
+
+static void sda_out(struct qib_devdata *dd, u8 bit)
+{
+	u32 mask;
+
+	mask = 1UL << dd->gpio_sda_num;
+
+	/* SDA is meant to be bare-drain, so never set "OUT", just DIR */
+	dd->f_gpio_mod(dd, 0, bit ? 0 : mask, mask);
+
+	i2c_wait_for_writes(dd);
+	udelay(2);
+}
+
+static u8 sda_in(struct qib_devdata *dd, int wait)
+{
+	int bnum;
+	u32 read_val, mask;
+
+	bnum = dd->gpio_sda_num;
+	mask = (1UL << bnum);
+	/* SDA is meant to be bare-drain, so never set "OUT", just DIR */
+	dd->f_gpio_mod(dd, 0, 0, mask);
+	read_val = dd->f_gpio_mod(dd, 0, 0, 0);
+	if (wait)
+		i2c_wait_for_writes(dd);
+	return (read_val & mask) >> bnum;
+}
+
+/**
+ * i2c_ackrcv - see if ack following write is true
+ * @dd: the qlogic_ib device
+ */
+static int i2c_ackrcv(struct qib_devdata *dd)
+{
+	u8 ack_received;
+
+	/* AT ENTRY SCL = LOW */
+	/* change direction, ignore data */
+	ack_received = sda_in(dd, 1);
+	scl_out(dd, 1);
+	ack_received = sda_in(dd, 1) == 0;
+	scl_out(dd, 0);
+	return ack_received;
+}
+
+static void stop_cmd(struct qib_devdata *dd);
+
+/**
+ * rd_byte - read a byte, sending STOP on last, else ACK
+ * @dd: the qlogic_ib device
+ *
+ * Returns byte shifted out of device
+ */
+static int rd_byte(struct qib_devdata *dd, int last)
+{
+	int bit_cntr, data;
+
+	data = 0;
+
+	for (bit_cntr = 7; bit_cntr >= 0; --bit_cntr) {
+		data <<= 1;
+		scl_out(dd, 1);
+		data |= sda_in(dd, 0);
+		scl_out(dd, 0);
+	}
+	if (last) {
+		scl_out(dd, 1);
+		stop_cmd(dd);
+	} else {
+		sda_out(dd, 0);
+		scl_out(dd, 1);
+		scl_out(dd, 0);
+		sda_out(dd, 1);
+	}
+	return data;
+}
+
+/**
+ * wr_byte - write a byte, one bit at a time
+ * @dd: the qlogic_ib device
+ * @data: the byte to write
+ *
+ * Returns 0 if we got the following ack, otherwise 1
+ */
+static int wr_byte(struct qib_devdata *dd, u8 data)
+{
+	int bit_cntr;
+	u8 bit;
+
+	for (bit_cntr = 7; bit_cntr >= 0; bit_cntr--) {
+		bit = (data >> bit_cntr) & 1;
+		sda_out(dd, bit);
+		scl_out(dd, 1);
+		scl_out(dd, 0);
+	}
+	return (!i2c_ackrcv(dd)) ? 1 : 0;
+}
+
+/*
+ * issue TWSI start sequence:
+ * (both clock/data high, clock high, data low while clock is high)
+ */
+static void start_seq(struct qib_devdata *dd)
+{
+	sda_out(dd, 1);
+	scl_out(dd, 1);
+	sda_out(dd, 0);
+	udelay(1);
+	scl_out(dd, 0);
+}
+
+/**
+ * stop_seq - transmit the stop sequence
+ * @dd: the qlogic_ib device
+ *
+ * (both clock/data low, clock high, data high while clock is high)
+ */
+static void stop_seq(struct qib_devdata *dd)
+{
+	scl_out(dd, 0);
+	sda_out(dd, 0);
+	scl_out(dd, 1);
+	sda_out(dd, 1);
+}
+
+/**
+ * stop_cmd - transmit the stop condition
+ * @dd: the qlogic_ib device
+ *
+ * (both clock/data low, clock high, data high while clock is high)
+ */
+static void stop_cmd(struct qib_devdata *dd)
+{
+	stop_seq(dd);
+	udelay(TWSI_BUF_WAIT_USEC);
+}
+
+/**
+ * qib_twsi_reset - reset I2C communication
+ * @dd: the qlogic_ib device
+ */
+
+int qib_twsi_reset(struct qib_devdata *dd)
+{
+	int clock_cycles_left = 9;
+	int was_high = 0;
+	u32 pins, mask;
+
+	/* Both SCL and SDA should be high. If not, there
+	 * is something wrong.
+	 */
+	mask = (1UL << dd->gpio_scl_num) | (1UL << dd->gpio_sda_num);
+
+	/*
+	 * Force pins to desired innocuous state.
+	 * This is the default power-on state with out=0 and dir=0,
+	 * So tri-stated and should be floating high (barring HW problems)
+	 */
+	dd->f_gpio_mod(dd, 0, 0, mask);
+
+	/*
+	 * Clock nine times to get all listeners into a sane state.
+	 * If SDA does not go high at any point, we are wedged.
+	 * One vendor recommends then issuing START followed by STOP.
+	 * we cannot use our "normal" functions to do that, because
+	 * if SCL drops between them, another vendor's part will
+	 * wedge, dropping SDA and keeping it low forever, at the end of
+	 * the next transaction (even if it was not the device addressed).
+	 * So our START and STOP take place with SCL held high.
+	 */
+	while (clock_cycles_left--) {
+		scl_out(dd, 0);
+		scl_out(dd, 1);
+		/* Note if SDA is high, but keep clocking to sync slave */
+		was_high |= sda_in(dd, 0);
+	}
+
+	if (was_high) {
+		/*
+		 * We saw a high, which we hope means the slave is sync'd.
+		 * Issue START, STOP, pause for T_BUF.
+		 */
+
+		pins = dd->f_gpio_mod(dd, 0, 0, 0);
+		if ((pins & mask) != mask)
+			qib_dev_err(dd, "GPIO pins not at rest: %d\n",
+				    pins & mask);
+		/* Drop SDA to issue START */
+		udelay(1); /* Guarantee .6 uSec setup */
+		sda_out(dd, 0);
+		udelay(1); /* Guarantee .6 uSec hold */
+		/* At this point, SCL is high, SDA low. Raise SDA for STOP */
+		sda_out(dd, 1);
+		udelay(TWSI_BUF_WAIT_USEC);
+	}
+
+	return !was_high;
+}
+
+#define QIB_TWSI_START 0x100
+#define QIB_TWSI_STOP 0x200
+
+/* Write byte to TWSI, optionally prefixed with START or suffixed with
+ * STOP.
+ * returns 0 if OK (ACK received), else != 0
+ */
+static int qib_twsi_wr(struct qib_devdata *dd, int data, int flags)
+{
+	int ret = 1;
+	if (flags & QIB_TWSI_START)
+		start_seq(dd);
+
+	ret = wr_byte(dd, data); /* Leaves SCL low (from i2c_ackrcv()) */
+
+	if (flags & QIB_TWSI_STOP)
+		stop_cmd(dd);
+	return ret;
+}
+
+/* Added functionality for IBA7220-based cards */
+#define QIB_TEMP_DEV 0x98
+
+/*
+ * qib_twsi_blk_rd
+ * Formerly called qib_eeprom_internal_read, and only used for eeprom,
+ * but now the general interface for data transfer from twsi devices.
+ * One vestige of its former role is that it recognizes a device
+ * QIB_TWSI_NO_DEV and does the correct operation for the legacy part,
+ * which responded to all TWSI device codes, interpreting them as
+ * address within device. On all other devices found on board handled by
+ * this driver, the device is followed by a one-byte "address" which selects
+ * the "register" or "offset" within the device from which data should
+ * be read.
+ */
+int qib_twsi_blk_rd(struct qib_devdata *dd, int dev, int addr,
+		    void *buffer, int len)
+{
+	int ret;
+	u8 *bp = buffer;
+
+	ret = 1;
+
+	if (dev == QIB_TWSI_NO_DEV) {
+		/* legacy not-really-I2C */
+		addr = (addr << 1) | READ_CMD;
+		ret = qib_twsi_wr(dd, addr, QIB_TWSI_START);
+	} else {
+		/* Actual I2C */
+		ret = qib_twsi_wr(dd, dev | WRITE_CMD, QIB_TWSI_START);
+		if (ret) {
+			stop_cmd(dd);
+			ret = 1;
+			goto bail;
+		}
+		/*
+		 * SFF spec claims we do _not_ stop after the addr
+		 * but simply issue a start with the "read" dev-addr.
+		 * Since we are implicitely waiting for ACK here,
+		 * we need t_buf (nominally 20uSec) before that start,
+		 * and cannot rely on the delay built in to the STOP
+		 */
+		ret = qib_twsi_wr(dd, addr, 0);
+		udelay(TWSI_BUF_WAIT_USEC);
+
+		if (ret) {
+			qib_dev_err(dd,
+				"Failed to write interface read addr %02X\n",
+				addr);
+			ret = 1;
+			goto bail;
+		}
+		ret = qib_twsi_wr(dd, dev | READ_CMD, QIB_TWSI_START);
+	}
+	if (ret) {
+		stop_cmd(dd);
+		ret = 1;
+		goto bail;
+	}
+
+	/*
+	 * block devices keeps clocking data out as long as we ack,
+	 * automatically incrementing the address. Some have "pages"
+	 * whose boundaries will not be crossed, but the handling
+	 * of these is left to the caller, who is in a better
+	 * position to know.
+	 */
+	while (len-- > 0) {
+		/*
+		 * Get and store data, sending ACK if length remaining,
+		 * else STOP
+		 */
+		*bp++ = rd_byte(dd, !len);
+	}
+
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+/*
+ * qib_twsi_blk_wr
+ * Formerly called qib_eeprom_internal_write, and only used for eeprom,
+ * but now the general interface for data transfer to twsi devices.
+ * One vestige of its former role is that it recognizes a device
+ * QIB_TWSI_NO_DEV and does the correct operation for the legacy part,
+ * which responded to all TWSI device codes, interpreting them as
+ * address within device. On all other devices found on board handled by
+ * this driver, the device is followed by a one-byte "address" which selects
+ * the "register" or "offset" within the device to which data should
+ * be written.
+ */
+int qib_twsi_blk_wr(struct qib_devdata *dd, int dev, int addr,
+		    const void *buffer, int len)
+{
+	int sub_len;
+	const u8 *bp = buffer;
+	int max_wait_time, i;
+	int ret;
+	ret = 1;
+
+	while (len > 0) {
+		if (dev == QIB_TWSI_NO_DEV) {
+			if (qib_twsi_wr(dd, (addr << 1) | WRITE_CMD,
+					QIB_TWSI_START)) {
+				goto failed_write;
+			}
+		} else {
+			/* Real I2C */
+			if (qib_twsi_wr(dd, dev | WRITE_CMD, QIB_TWSI_START))
+				goto failed_write;
+			ret = qib_twsi_wr(dd, addr, 0);
+			if (ret) {
+				qib_dev_err(dd, "Failed to write interface"
+					    " write addr %02X\n", addr);
+				goto failed_write;
+			}
+		}
+
+		sub_len = min(len, 4);
+		addr += sub_len;
+		len -= sub_len;
+
+		for (i = 0; i < sub_len; i++)
+			if (qib_twsi_wr(dd, *bp++, 0))
+				goto failed_write;
+
+		stop_cmd(dd);
+
+		/*
+		 * Wait for write complete by waiting for a successful
+		 * read (the chip replies with a zero after the write
+		 * cmd completes, and before it writes to the eeprom.
+		 * The startcmd for the read will fail the ack until
+		 * the writes have completed.   We do this inline to avoid
+		 * the debug prints that are in the real read routine
+		 * if the startcmd fails.
+		 * We also use the proper device address, so it doesn't matter
+		 * whether we have real eeprom_dev. Legacy likes any address.
+		 */
+		max_wait_time = 100;
+		while (qib_twsi_wr(dd, dev | READ_CMD, QIB_TWSI_START)) {
+			stop_cmd(dd);
+			if (!--max_wait_time)
+				goto failed_write;
+		}
+		/* now read (and ignore) the resulting byte */
+		rd_byte(dd, 1);
+	}
+
+	ret = 0;
+	goto bail;
+
+failed_write:
+	stop_cmd(dd);
+	ret = 1;
+
+bail:
+	return ret;
+}

drivers/infiniband/hw/qib/qib_tx.c

@@ -0,0 +1,557 @@
+/*
+ * Copyright (c) 2008, 2009, 2010 QLogic Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/spinlock.h>
+#include <linux/pci.h>
+#include <linux/io.h>
+#include <linux/delay.h>
+#include <linux/netdevice.h>
+#include <linux/vmalloc.h>
+
+#include "qib.h"
+
+static unsigned qib_hol_timeout_ms = 3000;
+module_param_named(hol_timeout_ms, qib_hol_timeout_ms, uint, S_IRUGO);
+MODULE_PARM_DESC(hol_timeout_ms,
+		 "duration of user app suspension after link failure");
+
+unsigned qib_sdma_fetch_arb = 1;
+module_param_named(fetch_arb, qib_sdma_fetch_arb, uint, S_IRUGO);
+MODULE_PARM_DESC(fetch_arb, "IBA7220: change SDMA descriptor arbitration");
+
+/**
+ * qib_disarm_piobufs - cancel a range of PIO buffers
+ * @dd: the qlogic_ib device
+ * @first: the first PIO buffer to cancel
+ * @cnt: the number of PIO buffers to cancel
+ *
+ * Cancel a range of PIO buffers. Used at user process close,
+ * in case it died while writing to a PIO buffer.
+ */
+void qib_disarm_piobufs(struct qib_devdata *dd, unsigned first, unsigned cnt)
+{
+	unsigned long flags;
+	unsigned i;
+	unsigned last;
+
+	last = first + cnt;
+	spin_lock_irqsave(&dd->pioavail_lock, flags);
+	for (i = first; i < last; i++) {
+		__clear_bit(i, dd->pio_need_disarm);
+		dd->f_sendctrl(dd->pport, QIB_SENDCTRL_DISARM_BUF(i));
+	}
+	spin_unlock_irqrestore(&dd->pioavail_lock, flags);
+}
+
+/*
+ * This is called by a user process when it sees the DISARM_BUFS event
+ * bit is set.
+ */
+int qib_disarm_piobufs_ifneeded(struct qib_ctxtdata *rcd)
+{
+	struct qib_devdata *dd = rcd->dd;
+	unsigned i;
+	unsigned last;
+	unsigned n = 0;
+
+	last = rcd->pio_base + rcd->piocnt;
+	/*
+	 * Don't need uctxt_lock here, since user has called in to us.
+	 * Clear at start in case more interrupts set bits while we
+	 * are disarming
+	 */
+	if (rcd->user_event_mask) {
+		/*
+		 * subctxt_cnt is 0 if not shared, so do base
+		 * separately, first, then remaining subctxt, if any
+		 */
+		clear_bit(_QIB_EVENT_DISARM_BUFS_BIT, &rcd->user_event_mask[0]);
+		for (i = 1; i < rcd->subctxt_cnt; i++)
+			clear_bit(_QIB_EVENT_DISARM_BUFS_BIT,
+				  &rcd->user_event_mask[i]);
+	}
+	spin_lock_irq(&dd->pioavail_lock);
+	for (i = rcd->pio_base; i < last; i++) {
+		if (__test_and_clear_bit(i, dd->pio_need_disarm)) {
+			n++;
+			dd->f_sendctrl(rcd->ppd, QIB_SENDCTRL_DISARM_BUF(i));
+		}
+	}
+	spin_unlock_irq(&dd->pioavail_lock);
+	return 0;
+}
+
+static struct qib_pportdata *is_sdma_buf(struct qib_devdata *dd, unsigned i)
+{
+	struct qib_pportdata *ppd;
+	unsigned pidx;
+
+	for (pidx = 0; pidx < dd->num_pports; pidx++) {
+		ppd = dd->pport + pidx;
+		if (i >= ppd->sdma_state.first_sendbuf &&
+		    i < ppd->sdma_state.last_sendbuf)
+			return ppd;
+	}
+	return NULL;
+}
+
+/*
+ * Return true if send buffer is being used by a user context.
+ * Sets  _QIB_EVENT_DISARM_BUFS_BIT in user_event_mask as a side effect
+ */
+static int find_ctxt(struct qib_devdata *dd, unsigned bufn)
+{
+	struct qib_ctxtdata *rcd;
+	unsigned ctxt;
+	int ret = 0;
+
+	spin_lock(&dd->uctxt_lock);
+	for (ctxt = dd->first_user_ctxt; ctxt < dd->cfgctxts; ctxt++) {
+		rcd = dd->rcd[ctxt];
+		if (!rcd || bufn < rcd->pio_base ||
+		    bufn >= rcd->pio_base + rcd->piocnt)
+			continue;
+		if (rcd->user_event_mask) {
+			int i;
+			/*
+			 * subctxt_cnt is 0 if not shared, so do base
+			 * separately, first, then remaining subctxt, if any
+			 */
+			set_bit(_QIB_EVENT_DISARM_BUFS_BIT,
+				&rcd->user_event_mask[0]);
+			for (i = 1; i < rcd->subctxt_cnt; i++)
+				set_bit(_QIB_EVENT_DISARM_BUFS_BIT,
+					&rcd->user_event_mask[i]);
+		}
+		ret = 1;
+		break;
+	}
+	spin_unlock(&dd->uctxt_lock);
+
+	return ret;
+}
+
+/*
+ * Disarm a set of send buffers.  If the buffer might be actively being
+ * written to, mark the buffer to be disarmed later when it is not being
+ * written to.
+ *
+ * This should only be called from the IRQ error handler.
+ */
+void qib_disarm_piobufs_set(struct qib_devdata *dd, unsigned long *mask,
+			    unsigned cnt)
+{
+	struct qib_pportdata *ppd, *pppd[dd->num_pports];
+	unsigned i;
+	unsigned long flags;
+
+	for (i = 0; i < dd->num_pports; i++)
+		pppd[i] = NULL;
+
+	for (i = 0; i < cnt; i++) {
+		int which;
+		if (!test_bit(i, mask))
+			continue;
+		/*
+		 * If the buffer is owned by the DMA hardware,
+		 * reset the DMA engine.
+		 */
+		ppd = is_sdma_buf(dd, i);
+		if (ppd) {
+			pppd[ppd->port] = ppd;
+			continue;
+		}
+		/*
+		 * If the kernel is writing the buffer or the buffer is
+		 * owned by a user process, we can't clear it yet.
+		 */
+		spin_lock_irqsave(&dd->pioavail_lock, flags);
+		if (test_bit(i, dd->pio_writing) ||
+		    (!test_bit(i << 1, dd->pioavailkernel) &&
+		     find_ctxt(dd, i))) {
+			__set_bit(i, dd->pio_need_disarm);
+			which = 0;
+		} else {
+			which = 1;
+			dd->f_sendctrl(dd->pport, QIB_SENDCTRL_DISARM_BUF(i));
+		}
+		spin_unlock_irqrestore(&dd->pioavail_lock, flags);
+	}
+
+	/* do cancel_sends once per port that had sdma piobufs in error */
+	for (i = 0; i < dd->num_pports; i++)
+		if (pppd[i])
+			qib_cancel_sends(pppd[i]);
+}
+
+/**
+ * update_send_bufs - update shadow copy of the PIO availability map
+ * @dd: the qlogic_ib device
+ *
+ * called whenever our local copy indicates we have run out of send buffers
+ */
+static void update_send_bufs(struct qib_devdata *dd)
+{
+	unsigned long flags;
+	unsigned i;
+	const unsigned piobregs = dd->pioavregs;
+
+	/*
+	 * If the generation (check) bits have changed, then we update the
+	 * busy bit for the corresponding PIO buffer.  This algorithm will
+	 * modify positions to the value they already have in some cases
+	 * (i.e., no change), but it's faster than changing only the bits
+	 * that have changed.
+	 *
+	 * We would like to do this atomicly, to avoid spinlocks in the
+	 * critical send path, but that's not really possible, given the
+	 * type of changes, and that this routine could be called on
+	 * multiple cpu's simultaneously, so we lock in this routine only,
+	 * to avoid conflicting updates; all we change is the shadow, and
+	 * it's a single 64 bit memory location, so by definition the update
+	 * is atomic in terms of what other cpu's can see in testing the
+	 * bits.  The spin_lock overhead isn't too bad, since it only
+	 * happens when all buffers are in use, so only cpu overhead, not
+	 * latency or bandwidth is affected.
+	 */
+	if (!dd->pioavailregs_dma)
+		return;
+	spin_lock_irqsave(&dd->pioavail_lock, flags);
+	for (i = 0; i < piobregs; i++) {
+		u64 pchbusy, pchg, piov, pnew;
+
+		piov = le64_to_cpu(dd->pioavailregs_dma[i]);
+		pchg = dd->pioavailkernel[i] &
+			~(dd->pioavailshadow[i] ^ piov);
+		pchbusy = pchg << QLOGIC_IB_SENDPIOAVAIL_BUSY_SHIFT;
+		if (pchg && (pchbusy & dd->pioavailshadow[i])) {
+			pnew = dd->pioavailshadow[i] & ~pchbusy;
+			pnew |= piov & pchbusy;
+			dd->pioavailshadow[i] = pnew;
+		}
+	}
+	spin_unlock_irqrestore(&dd->pioavail_lock, flags);
+}
+
+/*
+ * Debugging code and stats updates if no pio buffers available.
+ */
+static noinline void no_send_bufs(struct qib_devdata *dd)
+{
+	dd->upd_pio_shadow = 1;
+
+	/* not atomic, but if we lose a stat count in a while, that's OK */
+	qib_stats.sps_nopiobufs++;
+}
+
+/*
+ * Common code for normal driver send buffer allocation, and reserved
+ * allocation.
+ *
+ * Do appropriate marking as busy, etc.
+ * Returns buffer pointer if one is found, otherwise NULL.
+ */
+u32 __iomem *qib_getsendbuf_range(struct qib_devdata *dd, u32 *pbufnum,
+				  u32 first, u32 last)
+{
+	unsigned i, j, updated = 0;
+	unsigned nbufs;
+	unsigned long flags;
+	unsigned long *shadow = dd->pioavailshadow;
+	u32 __iomem *buf;
+
+	if (!(dd->flags & QIB_PRESENT))
+		return NULL;
+
+	nbufs = last - first + 1; /* number in range to check */
+	if (dd->upd_pio_shadow) {
+		/*
+		 * Minor optimization.  If we had no buffers on last call,
+		 * start out by doing the update; continue and do scan even
+		 * if no buffers were updated, to be paranoid.
+		 */
+		update_send_bufs(dd);
+		updated++;
+	}
+	i = first;
+rescan:
+	/*
+	 * While test_and_set_bit() is atomic, we do that and then the
+	 * change_bit(), and the pair is not.  See if this is the cause
+	 * of the remaining armlaunch errors.
+	 */
+	spin_lock_irqsave(&dd->pioavail_lock, flags);
+	for (j = 0; j < nbufs; j++, i++) {
+		if (i > last)
+			i = first;
+		if (__test_and_set_bit((2 * i) + 1, shadow))
+			continue;
+		/* flip generation bit */
+		__change_bit(2 * i, shadow);
+		/* remember that the buffer can be written to now */
+		__set_bit(i, dd->pio_writing);
+		break;
+	}
+	spin_unlock_irqrestore(&dd->pioavail_lock, flags);
+
+	if (j == nbufs) {
+		if (!updated) {
+			/*
+			 * First time through; shadow exhausted, but may be
+			 * buffers available, try an update and then rescan.
+			 */
+			update_send_bufs(dd);
+			updated++;
+			i = first;
+			goto rescan;
+		}
+		no_send_bufs(dd);
+		buf = NULL;
+	} else {
+		if (i < dd->piobcnt2k)
+			buf = (u32 __iomem *)(dd->pio2kbase +
+				i * dd->palign);
+		else
+			buf = (u32 __iomem *)(dd->pio4kbase +
+				(i - dd->piobcnt2k) * dd->align4k);
+		if (pbufnum)
+			*pbufnum = i;
+		dd->upd_pio_shadow = 0;
+	}
+
+	return buf;
+}
+
+/*
+ * Record that the caller is finished writing to the buffer so we don't
+ * disarm it while it is being written and disarm it now if needed.
+ */
+void qib_sendbuf_done(struct qib_devdata *dd, unsigned n)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&dd->pioavail_lock, flags);
+	__clear_bit(n, dd->pio_writing);
+	if (__test_and_clear_bit(n, dd->pio_need_disarm))
+		dd->f_sendctrl(dd->pport, QIB_SENDCTRL_DISARM_BUF(n));
+	spin_unlock_irqrestore(&dd->pioavail_lock, flags);
+}
+
+/**
+ * qib_chg_pioavailkernel - change which send buffers are available for kernel
+ * @dd: the qlogic_ib device
+ * @start: the starting send buffer number
+ * @len: the number of send buffers
+ * @avail: true if the buffers are available for kernel use, false otherwise
+ */
+void qib_chg_pioavailkernel(struct qib_devdata *dd, unsigned start,
+	unsigned len, u32 avail, struct qib_ctxtdata *rcd)
+{
+	unsigned long flags;
+	unsigned end;
+	unsigned ostart = start;
+
+	/* There are two bits per send buffer (busy and generation) */
+	start *= 2;
+	end = start + len * 2;
+
+	spin_lock_irqsave(&dd->pioavail_lock, flags);
+	/* Set or clear the busy bit in the shadow. */
+	while (start < end) {
+		if (avail) {
+			unsigned long dma;
+			int i;
+
+			/*
+			 * The BUSY bit will never be set, because we disarm
+			 * the user buffers before we hand them back to the
+			 * kernel.  We do have to make sure the generation
+			 * bit is set correctly in shadow, since it could
+			 * have changed many times while allocated to user.
+			 * We can't use the bitmap functions on the full
+			 * dma array because it is always little-endian, so
+			 * we have to flip to host-order first.
+			 * BITS_PER_LONG is slightly wrong, since it's
+			 * always 64 bits per register in chip...
+			 * We only work on 64 bit kernels, so that's OK.
+			 */
+			i = start / BITS_PER_LONG;
+			__clear_bit(QLOGIC_IB_SENDPIOAVAIL_BUSY_SHIFT + start,
+				    dd->pioavailshadow);
+			dma = (unsigned long)
+				le64_to_cpu(dd->pioavailregs_dma[i]);
+			if (test_bit((QLOGIC_IB_SENDPIOAVAIL_CHECK_SHIFT +
+				      start) % BITS_PER_LONG, &dma))
+				__set_bit(QLOGIC_IB_SENDPIOAVAIL_CHECK_SHIFT +
+					  start, dd->pioavailshadow);
+			else
+				__clear_bit(QLOGIC_IB_SENDPIOAVAIL_CHECK_SHIFT
+					    + start, dd->pioavailshadow);
+			__set_bit(start, dd->pioavailkernel);
+		} else {
+			__set_bit(start + QLOGIC_IB_SENDPIOAVAIL_BUSY_SHIFT,
+				  dd->pioavailshadow);
+			__clear_bit(start, dd->pioavailkernel);
+		}
+		start += 2;
+	}
+
+	spin_unlock_irqrestore(&dd->pioavail_lock, flags);
+
+	dd->f_txchk_change(dd, ostart, len, avail, rcd);
+}
+
+/*
+ * Flush all sends that might be in the ready to send state, as well as any
+ * that are in the process of being sent.  Used whenever we need to be
+ * sure the send side is idle.  Cleans up all buffer state by canceling
+ * all pio buffers, and issuing an abort, which cleans up anything in the
+ * launch fifo.  The cancel is superfluous on some chip versions, but
+ * it's safer to always do it.
+ * PIOAvail bits are updated by the chip as if a normal send had happened.
+ */
+void qib_cancel_sends(struct qib_pportdata *ppd)
+{
+	struct qib_devdata *dd = ppd->dd;
+	struct qib_ctxtdata *rcd;
+	unsigned long flags;
+	unsigned ctxt;
+	unsigned i;
+	unsigned last;
+
+	/*
+	 * Tell PSM to disarm buffers again before trying to reuse them.
+	 * We need to be sure the rcd doesn't change out from under us
+	 * while we do so.  We hold the two locks sequentially.  We might
+	 * needlessly set some need_disarm bits as a result, if the
+	 * context is closed after we release the uctxt_lock, but that's
+	 * fairly benign, and safer than nesting the locks.
+	 */
+	for (ctxt = dd->first_user_ctxt; ctxt < dd->cfgctxts; ctxt++) {
+		spin_lock_irqsave(&dd->uctxt_lock, flags);
+		rcd = dd->rcd[ctxt];
+		if (rcd && rcd->ppd == ppd) {
+			last = rcd->pio_base + rcd->piocnt;
+			if (rcd->user_event_mask) {
+				/*
+				 * subctxt_cnt is 0 if not shared, so do base
+				 * separately, first, then remaining subctxt,
+				 * if any
+				 */
+				set_bit(_QIB_EVENT_DISARM_BUFS_BIT,
+					&rcd->user_event_mask[0]);
+				for (i = 1; i < rcd->subctxt_cnt; i++)
+					set_bit(_QIB_EVENT_DISARM_BUFS_BIT,
+						&rcd->user_event_mask[i]);
+			}
+			i = rcd->pio_base;
+			spin_unlock_irqrestore(&dd->uctxt_lock, flags);
+			spin_lock_irqsave(&dd->pioavail_lock, flags);
+			for (; i < last; i++)
+				__set_bit(i, dd->pio_need_disarm);
+			spin_unlock_irqrestore(&dd->pioavail_lock, flags);
+		} else
+			spin_unlock_irqrestore(&dd->uctxt_lock, flags);
+	}
+
+	if (!(dd->flags & QIB_HAS_SEND_DMA))
+		dd->f_sendctrl(ppd, QIB_SENDCTRL_DISARM_ALL |
+				    QIB_SENDCTRL_FLUSH);
+}
+
+/*
+ * Force an update of in-memory copy of the pioavail registers, when
+ * needed for any of a variety of reasons.
+ * If already off, this routine is a nop, on the assumption that the
+ * caller (or set of callers) will "do the right thing".
+ * This is a per-device operation, so just the first port.
+ */
+void qib_force_pio_avail_update(struct qib_devdata *dd)
+{
+	dd->f_sendctrl(dd->pport, QIB_SENDCTRL_AVAIL_BLIP);
+}
+
+void qib_hol_down(struct qib_pportdata *ppd)
+{
+	/*
+	 * Cancel sends when the link goes DOWN so that we aren't doing it
+	 * at INIT when we might be trying to send SMI packets.
+	 */
+	if (!(ppd->lflags & QIBL_IB_AUTONEG_INPROG))
+		qib_cancel_sends(ppd);
+}
+
+/*
+ * Link is at INIT.
+ * We start the HoL timer so we can detect stuck packets blocking SMP replies.
+ * Timer may already be running, so use mod_timer, not add_timer.
+ */
+void qib_hol_init(struct qib_pportdata *ppd)
+{
+	if (ppd->hol_state != QIB_HOL_INIT) {
+		ppd->hol_state = QIB_HOL_INIT;
+		mod_timer(&ppd->hol_timer,
+			  jiffies + msecs_to_jiffies(qib_hol_timeout_ms));
+	}
+}
+
+/*
+ * Link is up, continue any user processes, and ensure timer
+ * is a nop, if running.  Let timer keep running, if set; it
+ * will nop when it sees the link is up.
+ */
+void qib_hol_up(struct qib_pportdata *ppd)
+{
+	ppd->hol_state = QIB_HOL_UP;
+}
+
+/*
+ * This is only called via the timer.
+ */
+void qib_hol_event(unsigned long opaque)
+{
+	struct qib_pportdata *ppd = (struct qib_pportdata *)opaque;
+
+	/* If hardware error, etc, skip. */
+	if (!(ppd->dd->flags & QIB_INITTED))
+		return;
+
+	if (ppd->hol_state != QIB_HOL_UP) {
+		/*
+		 * Try to flush sends in case a stuck packet is blocking
+		 * SMP replies.
+		 */
+		qib_hol_down(ppd);
+		mod_timer(&ppd->hol_timer,
+			  jiffies + msecs_to_jiffies(qib_hol_timeout_ms));
+	}
+}

drivers/infiniband/hw/qib/qib_uc.c

@@ -0,0 +1,555 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009, 2010 QLogic Corporation.
+ * All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include "qib.h"
+
+/* cut down ridiculously long IB macro names */
+#define OP(x) IB_OPCODE_UC_##x
+
+/**
+ * qib_make_uc_req - construct a request packet (SEND, RDMA write)
+ * @qp: a pointer to the QP
+ *
+ * Return 1 if constructed; otherwise, return 0.
+ */
+int qib_make_uc_req(struct qib_qp *qp)
+{
+	struct qib_other_headers *ohdr;
+	struct qib_swqe *wqe;
+	unsigned long flags;
+	u32 hwords;
+	u32 bth0;
+	u32 len;
+	u32 pmtu = ib_mtu_enum_to_int(qp->path_mtu);
+	int ret = 0;
+
+	spin_lock_irqsave(&qp->s_lock, flags);
+
+	if (!(ib_qib_state_ops[qp->state] & QIB_PROCESS_SEND_OK)) {
+		if (!(ib_qib_state_ops[qp->state] & QIB_FLUSH_SEND))
+			goto bail;
+		/* We are in the error state, flush the work request. */
+		if (qp->s_last == qp->s_head)
+			goto bail;
+		/* If DMAs are in progress, we can't flush immediately. */
+		if (atomic_read(&qp->s_dma_busy)) {
+			qp->s_flags |= QIB_S_WAIT_DMA;
+			goto bail;
+		}
+		wqe = get_swqe_ptr(qp, qp->s_last);
+		qib_send_complete(qp, wqe, IB_WC_WR_FLUSH_ERR);
+		goto done;
+	}
+
+	ohdr = &qp->s_hdr.u.oth;
+	if (qp->remote_ah_attr.ah_flags & IB_AH_GRH)
+		ohdr = &qp->s_hdr.u.l.oth;
+
+	/* header size in 32-bit words LRH+BTH = (8+12)/4. */
+	hwords = 5;
+	bth0 = 0;
+
+	/* Get the next send request. */
+	wqe = get_swqe_ptr(qp, qp->s_cur);
+	qp->s_wqe = NULL;
+	switch (qp->s_state) {
+	default:
+		if (!(ib_qib_state_ops[qp->state] &
+		    QIB_PROCESS_NEXT_SEND_OK))
+			goto bail;
+		/* Check if send work queue is empty. */
+		if (qp->s_cur == qp->s_head)
+			goto bail;
+		/*
+		 * Start a new request.
+		 */
+		wqe->psn = qp->s_next_psn;
+		qp->s_psn = qp->s_next_psn;
+		qp->s_sge.sge = wqe->sg_list[0];
+		qp->s_sge.sg_list = wqe->sg_list + 1;
+		qp->s_sge.num_sge = wqe->wr.num_sge;
+		qp->s_sge.total_len = wqe->length;
+		len = wqe->length;
+		qp->s_len = len;
+		switch (wqe->wr.opcode) {
+		case IB_WR_SEND:
+		case IB_WR_SEND_WITH_IMM:
+			if (len > pmtu) {
+				qp->s_state = OP(SEND_FIRST);
+				len = pmtu;
+				break;
+			}
+			if (wqe->wr.opcode == IB_WR_SEND)
+				qp->s_state = OP(SEND_ONLY);
+			else {
+				qp->s_state =
+					OP(SEND_ONLY_WITH_IMMEDIATE);
+				/* Immediate data comes after the BTH */
+				ohdr->u.imm_data = wqe->wr.ex.imm_data;
+				hwords += 1;
+			}
+			if (wqe->wr.send_flags & IB_SEND_SOLICITED)
+				bth0 |= IB_BTH_SOLICITED;
+			qp->s_wqe = wqe;
+			if (++qp->s_cur >= qp->s_size)
+				qp->s_cur = 0;
+			break;
+
+		case IB_WR_RDMA_WRITE:
+		case IB_WR_RDMA_WRITE_WITH_IMM:
+			ohdr->u.rc.reth.vaddr =
+				cpu_to_be64(wqe->wr.wr.rdma.remote_addr);
+			ohdr->u.rc.reth.rkey =
+				cpu_to_be32(wqe->wr.wr.rdma.rkey);
+			ohdr->u.rc.reth.length = cpu_to_be32(len);
+			hwords += sizeof(struct ib_reth) / 4;
+			if (len > pmtu) {
+				qp->s_state = OP(RDMA_WRITE_FIRST);
+				len = pmtu;
+				break;
+			}
+			if (wqe->wr.opcode == IB_WR_RDMA_WRITE)
+				qp->s_state = OP(RDMA_WRITE_ONLY);
+			else {
+				qp->s_state =
+					OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE);
+				/* Immediate data comes after the RETH */
+				ohdr->u.rc.imm_data = wqe->wr.ex.imm_data;
+				hwords += 1;
+				if (wqe->wr.send_flags & IB_SEND_SOLICITED)
+					bth0 |= IB_BTH_SOLICITED;
+			}
+			qp->s_wqe = wqe;
+			if (++qp->s_cur >= qp->s_size)
+				qp->s_cur = 0;
+			break;
+
+		default:
+			goto bail;
+		}
+		break;
+
+	case OP(SEND_FIRST):
+		qp->s_state = OP(SEND_MIDDLE);
+		/* FALLTHROUGH */
+	case OP(SEND_MIDDLE):
+		len = qp->s_len;
+		if (len > pmtu) {
+			len = pmtu;
+			break;
+		}
+		if (wqe->wr.opcode == IB_WR_SEND)
+			qp->s_state = OP(SEND_LAST);
+		else {
+			qp->s_state = OP(SEND_LAST_WITH_IMMEDIATE);
+			/* Immediate data comes after the BTH */
+			ohdr->u.imm_data = wqe->wr.ex.imm_data;
+			hwords += 1;
+		}
+		if (wqe->wr.send_flags & IB_SEND_SOLICITED)
+			bth0 |= IB_BTH_SOLICITED;
+		qp->s_wqe = wqe;
+		if (++qp->s_cur >= qp->s_size)
+			qp->s_cur = 0;
+		break;
+
+	case OP(RDMA_WRITE_FIRST):
+		qp->s_state = OP(RDMA_WRITE_MIDDLE);
+		/* FALLTHROUGH */
+	case OP(RDMA_WRITE_MIDDLE):
+		len = qp->s_len;
+		if (len > pmtu) {
+			len = pmtu;
+			break;
+		}
+		if (wqe->wr.opcode == IB_WR_RDMA_WRITE)
+			qp->s_state = OP(RDMA_WRITE_LAST);
+		else {
+			qp->s_state =
+				OP(RDMA_WRITE_LAST_WITH_IMMEDIATE);
+			/* Immediate data comes after the BTH */
+			ohdr->u.imm_data = wqe->wr.ex.imm_data;
+			hwords += 1;
+			if (wqe->wr.send_flags & IB_SEND_SOLICITED)
+				bth0 |= IB_BTH_SOLICITED;
+		}
+		qp->s_wqe = wqe;
+		if (++qp->s_cur >= qp->s_size)
+			qp->s_cur = 0;
+		break;
+	}
+	qp->s_len -= len;
+	qp->s_hdrwords = hwords;
+	qp->s_cur_sge = &qp->s_sge;
+	qp->s_cur_size = len;
+	qib_make_ruc_header(qp, ohdr, bth0 | (qp->s_state << 24),
+			    qp->s_next_psn++ & QIB_PSN_MASK);
+done:
+	ret = 1;
+	goto unlock;
+
+bail:
+	qp->s_flags &= ~QIB_S_BUSY;
+unlock:
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+	return ret;
+}
+
+/**
+ * qib_uc_rcv - handle an incoming UC packet
+ * @ibp: the port the packet came in on
+ * @hdr: the header of the packet
+ * @has_grh: true if the packet has a GRH
+ * @data: the packet data
+ * @tlen: the length of the packet
+ * @qp: the QP for this packet.
+ *
+ * This is called from qib_qp_rcv() to process an incoming UC packet
+ * for the given QP.
+ * Called at interrupt level.
+ */
+void qib_uc_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr,
+		int has_grh, void *data, u32 tlen, struct qib_qp *qp)
+{
+	struct qib_other_headers *ohdr;
+	unsigned long flags;
+	u32 opcode;
+	u32 hdrsize;
+	u32 psn;
+	u32 pad;
+	struct ib_wc wc;
+	u32 pmtu = ib_mtu_enum_to_int(qp->path_mtu);
+	struct ib_reth *reth;
+	int ret;
+
+	/* Check for GRH */
+	if (!has_grh) {
+		ohdr = &hdr->u.oth;
+		hdrsize = 8 + 12;       /* LRH + BTH */
+	} else {
+		ohdr = &hdr->u.l.oth;
+		hdrsize = 8 + 40 + 12;  /* LRH + GRH + BTH */
+	}
+
+	opcode = be32_to_cpu(ohdr->bth[0]);
+	spin_lock_irqsave(&qp->s_lock, flags);
+	if (qib_ruc_check_hdr(ibp, hdr, has_grh, qp, opcode))
+		goto sunlock;
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+
+	psn = be32_to_cpu(ohdr->bth[2]);
+	opcode >>= 24;
+	memset(&wc, 0, sizeof wc);
+
+	/* Prevent simultaneous processing after APM on different CPUs */
+	spin_lock(&qp->r_lock);
+
+	/* Compare the PSN verses the expected PSN. */
+	if (unlikely(qib_cmp24(psn, qp->r_psn) != 0)) {
+		/*
+		 * Handle a sequence error.
+		 * Silently drop any current message.
+		 */
+		qp->r_psn = psn;
+inv:
+		if (qp->r_state == OP(SEND_FIRST) ||
+		    qp->r_state == OP(SEND_MIDDLE)) {
+			set_bit(QIB_R_REWIND_SGE, &qp->r_aflags);
+			qp->r_sge.num_sge = 0;
+		} else
+			while (qp->r_sge.num_sge) {
+				atomic_dec(&qp->r_sge.sge.mr->refcount);
+				if (--qp->r_sge.num_sge)
+					qp->r_sge.sge = *qp->r_sge.sg_list++;
+			}
+		qp->r_state = OP(SEND_LAST);
+		switch (opcode) {
+		case OP(SEND_FIRST):
+		case OP(SEND_ONLY):
+		case OP(SEND_ONLY_WITH_IMMEDIATE):
+			goto send_first;
+
+		case OP(RDMA_WRITE_FIRST):
+		case OP(RDMA_WRITE_ONLY):
+		case OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE):
+			goto rdma_first;
+
+		default:
+			goto drop;
+		}
+	}
+
+	/* Check for opcode sequence errors. */
+	switch (qp->r_state) {
+	case OP(SEND_FIRST):
+	case OP(SEND_MIDDLE):
+		if (opcode == OP(SEND_MIDDLE) ||
+		    opcode == OP(SEND_LAST) ||
+		    opcode == OP(SEND_LAST_WITH_IMMEDIATE))
+			break;
+		goto inv;
+
+	case OP(RDMA_WRITE_FIRST):
+	case OP(RDMA_WRITE_MIDDLE):
+		if (opcode == OP(RDMA_WRITE_MIDDLE) ||
+		    opcode == OP(RDMA_WRITE_LAST) ||
+		    opcode == OP(RDMA_WRITE_LAST_WITH_IMMEDIATE))
+			break;
+		goto inv;
+
+	default:
+		if (opcode == OP(SEND_FIRST) ||
+		    opcode == OP(SEND_ONLY) ||
+		    opcode == OP(SEND_ONLY_WITH_IMMEDIATE) ||
+		    opcode == OP(RDMA_WRITE_FIRST) ||
+		    opcode == OP(RDMA_WRITE_ONLY) ||
+		    opcode == OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE))
+			break;
+		goto inv;
+	}
+
+	if (qp->state == IB_QPS_RTR && !(qp->r_flags & QIB_R_COMM_EST)) {
+		qp->r_flags |= QIB_R_COMM_EST;
+		if (qp->ibqp.event_handler) {
+			struct ib_event ev;
+
+			ev.device = qp->ibqp.device;
+			ev.element.qp = &qp->ibqp;
+			ev.event = IB_EVENT_COMM_EST;
+			qp->ibqp.event_handler(&ev, qp->ibqp.qp_context);
+		}
+	}
+
+	/* OK, process the packet. */
+	switch (opcode) {
+	case OP(SEND_FIRST):
+	case OP(SEND_ONLY):
+	case OP(SEND_ONLY_WITH_IMMEDIATE):
+send_first:
+		if (test_and_clear_bit(QIB_R_REWIND_SGE, &qp->r_aflags))
+			qp->r_sge = qp->s_rdma_read_sge;
+		else {
+			ret = qib_get_rwqe(qp, 0);
+			if (ret < 0)
+				goto op_err;
+			if (!ret)
+				goto drop;
+			/*
+			 * qp->s_rdma_read_sge will be the owner
+			 * of the mr references.
+			 */
+			qp->s_rdma_read_sge = qp->r_sge;
+		}
+		qp->r_rcv_len = 0;
+		if (opcode == OP(SEND_ONLY))
+			goto send_last;
+		else if (opcode == OP(SEND_ONLY_WITH_IMMEDIATE))
+			goto send_last_imm;
+		/* FALLTHROUGH */
+	case OP(SEND_MIDDLE):
+		/* Check for invalid length PMTU or posted rwqe len. */
+		if (unlikely(tlen != (hdrsize + pmtu + 4)))
+			goto rewind;
+		qp->r_rcv_len += pmtu;
+		if (unlikely(qp->r_rcv_len > qp->r_len))
+			goto rewind;
+		qib_copy_sge(&qp->r_sge, data, pmtu, 0);
+		break;
+
+	case OP(SEND_LAST_WITH_IMMEDIATE):
+send_last_imm:
+		wc.ex.imm_data = ohdr->u.imm_data;
+		hdrsize += 4;
+		wc.wc_flags = IB_WC_WITH_IMM;
+		/* FALLTHROUGH */
+	case OP(SEND_LAST):
+send_last:
+		/* Get the number of bytes the message was padded by. */
+		pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
+		/* Check for invalid length. */
+		/* XXX LAST len should be >= 1 */
+		if (unlikely(tlen < (hdrsize + pad + 4)))
+			goto rewind;
+		/* Don't count the CRC. */
+		tlen -= (hdrsize + pad + 4);
+		wc.byte_len = tlen + qp->r_rcv_len;
+		if (unlikely(wc.byte_len > qp->r_len))
+			goto rewind;
+		wc.opcode = IB_WC_RECV;
+last_imm:
+		qib_copy_sge(&qp->r_sge, data, tlen, 0);
+		while (qp->s_rdma_read_sge.num_sge) {
+			atomic_dec(&qp->s_rdma_read_sge.sge.mr->refcount);
+			if (--qp->s_rdma_read_sge.num_sge)
+				qp->s_rdma_read_sge.sge =
+					*qp->s_rdma_read_sge.sg_list++;
+		}
+		wc.wr_id = qp->r_wr_id;
+		wc.status = IB_WC_SUCCESS;
+		wc.qp = &qp->ibqp;
+		wc.src_qp = qp->remote_qpn;
+		wc.slid = qp->remote_ah_attr.dlid;
+		wc.sl = qp->remote_ah_attr.sl;
+		/* Signal completion event if the solicited bit is set. */
+		qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
+			     (ohdr->bth[0] &
+				cpu_to_be32(IB_BTH_SOLICITED)) != 0);
+		break;
+
+	case OP(RDMA_WRITE_FIRST):
+	case OP(RDMA_WRITE_ONLY):
+	case OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE): /* consume RWQE */
+rdma_first:
+		if (unlikely(!(qp->qp_access_flags &
+			       IB_ACCESS_REMOTE_WRITE))) {
+			goto drop;
+		}
+		reth = &ohdr->u.rc.reth;
+		hdrsize += sizeof(*reth);
+		qp->r_len = be32_to_cpu(reth->length);
+		qp->r_rcv_len = 0;
+		qp->r_sge.sg_list = NULL;
+		if (qp->r_len != 0) {
+			u32 rkey = be32_to_cpu(reth->rkey);
+			u64 vaddr = be64_to_cpu(reth->vaddr);
+			int ok;
+
+			/* Check rkey */
+			ok = qib_rkey_ok(qp, &qp->r_sge.sge, qp->r_len,
+					 vaddr, rkey, IB_ACCESS_REMOTE_WRITE);
+			if (unlikely(!ok))
+				goto drop;
+			qp->r_sge.num_sge = 1;
+		} else {
+			qp->r_sge.num_sge = 0;
+			qp->r_sge.sge.mr = NULL;
+			qp->r_sge.sge.vaddr = NULL;
+			qp->r_sge.sge.length = 0;
+			qp->r_sge.sge.sge_length = 0;
+		}
+		if (opcode == OP(RDMA_WRITE_ONLY))
+			goto rdma_last;
+		else if (opcode == OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE))
+			goto rdma_last_imm;
+		/* FALLTHROUGH */
+	case OP(RDMA_WRITE_MIDDLE):
+		/* Check for invalid length PMTU or posted rwqe len. */
+		if (unlikely(tlen != (hdrsize + pmtu + 4)))
+			goto drop;
+		qp->r_rcv_len += pmtu;
+		if (unlikely(qp->r_rcv_len > qp->r_len))
+			goto drop;
+		qib_copy_sge(&qp->r_sge, data, pmtu, 1);
+		break;
+
+	case OP(RDMA_WRITE_LAST_WITH_IMMEDIATE):
+rdma_last_imm:
+		wc.ex.imm_data = ohdr->u.imm_data;
+		hdrsize += 4;
+		wc.wc_flags = IB_WC_WITH_IMM;
+
+		/* Get the number of bytes the message was padded by. */
+		pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
+		/* Check for invalid length. */
+		/* XXX LAST len should be >= 1 */
+		if (unlikely(tlen < (hdrsize + pad + 4)))
+			goto drop;
+		/* Don't count the CRC. */
+		tlen -= (hdrsize + pad + 4);
+		if (unlikely(tlen + qp->r_rcv_len != qp->r_len))
+			goto drop;
+		if (test_and_clear_bit(QIB_R_REWIND_SGE, &qp->r_aflags))
+			while (qp->s_rdma_read_sge.num_sge) {
+				atomic_dec(&qp->s_rdma_read_sge.sge.mr->
+					   refcount);
+				if (--qp->s_rdma_read_sge.num_sge)
+					qp->s_rdma_read_sge.sge =
+						*qp->s_rdma_read_sge.sg_list++;
+			}
+		else {
+			ret = qib_get_rwqe(qp, 1);
+			if (ret < 0)
+				goto op_err;
+			if (!ret)
+				goto drop;
+		}
+		wc.byte_len = qp->r_len;
+		wc.opcode = IB_WC_RECV_RDMA_WITH_IMM;
+		goto last_imm;
+
+	case OP(RDMA_WRITE_LAST):
+rdma_last:
+		/* Get the number of bytes the message was padded by. */
+		pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
+		/* Check for invalid length. */
+		/* XXX LAST len should be >= 1 */
+		if (unlikely(tlen < (hdrsize + pad + 4)))
+			goto drop;
+		/* Don't count the CRC. */
+		tlen -= (hdrsize + pad + 4);
+		if (unlikely(tlen + qp->r_rcv_len != qp->r_len))
+			goto drop;
+		qib_copy_sge(&qp->r_sge, data, tlen, 1);
+		while (qp->r_sge.num_sge) {
+			atomic_dec(&qp->r_sge.sge.mr->refcount);
+			if (--qp->r_sge.num_sge)
+				qp->r_sge.sge = *qp->r_sge.sg_list++;
+		}
+		break;
+
+	default:
+		/* Drop packet for unknown opcodes. */
+		goto drop;
+	}
+	qp->r_psn++;
+	qp->r_state = opcode;
+	spin_unlock(&qp->r_lock);
+	return;
+
+rewind:
+	set_bit(QIB_R_REWIND_SGE, &qp->r_aflags);
+	qp->r_sge.num_sge = 0;
+drop:
+	ibp->n_pkt_drops++;
+	spin_unlock(&qp->r_lock);
+	return;
+
+op_err:
+	qib_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
+	spin_unlock(&qp->r_lock);
+	return;
+
+sunlock:
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+}

drivers/infiniband/hw/qib/qib_ud.c

@@ -0,0 +1,607 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <rdma/ib_smi.h>
+
+#include "qib.h"
+#include "qib_mad.h"
+
+/**
+ * qib_ud_loopback - handle send on loopback QPs
+ * @sqp: the sending QP
+ * @swqe: the send work request
+ *
+ * This is called from qib_make_ud_req() to forward a WQE addressed
+ * to the same HCA.
+ * Note that the receive interrupt handler may be calling qib_ud_rcv()
+ * while this is being called.
+ */
+static void qib_ud_loopback(struct qib_qp *sqp, struct qib_swqe *swqe)
+{
+	struct qib_ibport *ibp = to_iport(sqp->ibqp.device, sqp->port_num);
+	struct qib_pportdata *ppd;
+	struct qib_qp *qp;
+	struct ib_ah_attr *ah_attr;
+	unsigned long flags;
+	struct qib_sge_state ssge;
+	struct qib_sge *sge;
+	struct ib_wc wc;
+	u32 length;
+
+	qp = qib_lookup_qpn(ibp, swqe->wr.wr.ud.remote_qpn);
+	if (!qp) {
+		ibp->n_pkt_drops++;
+		return;
+	}
+	if (qp->ibqp.qp_type != sqp->ibqp.qp_type ||
+	    !(ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK)) {
+		ibp->n_pkt_drops++;
+		goto drop;
+	}
+
+	ah_attr = &to_iah(swqe->wr.wr.ud.ah)->attr;
+	ppd = ppd_from_ibp(ibp);
+
+	if (qp->ibqp.qp_num > 1) {
+		u16 pkey1;
+		u16 pkey2;
+		u16 lid;
+
+		pkey1 = qib_get_pkey(ibp, sqp->s_pkey_index);
+		pkey2 = qib_get_pkey(ibp, qp->s_pkey_index);
+		if (unlikely(!qib_pkey_ok(pkey1, pkey2))) {
+			lid = ppd->lid | (ah_attr->src_path_bits &
+					  ((1 << ppd->lmc) - 1));
+			qib_bad_pqkey(ibp, IB_NOTICE_TRAP_BAD_PKEY, pkey1,
+				      ah_attr->sl,
+				      sqp->ibqp.qp_num, qp->ibqp.qp_num,
+				      cpu_to_be16(lid),
+				      cpu_to_be16(ah_attr->dlid));
+			goto drop;
+		}
+	}
+
+	/*
+	 * Check that the qkey matches (except for QP0, see 9.6.1.4.1).
+	 * Qkeys with the high order bit set mean use the
+	 * qkey from the QP context instead of the WR (see 10.2.5).
+	 */
+	if (qp->ibqp.qp_num) {
+		u32 qkey;
+
+		qkey = (int)swqe->wr.wr.ud.remote_qkey < 0 ?
+			sqp->qkey : swqe->wr.wr.ud.remote_qkey;
+		if (unlikely(qkey != qp->qkey)) {
+			u16 lid;
+
+			lid = ppd->lid | (ah_attr->src_path_bits &
+					  ((1 << ppd->lmc) - 1));
+			qib_bad_pqkey(ibp, IB_NOTICE_TRAP_BAD_QKEY, qkey,
+				      ah_attr->sl,
+				      sqp->ibqp.qp_num, qp->ibqp.qp_num,
+				      cpu_to_be16(lid),
+				      cpu_to_be16(ah_attr->dlid));
+			goto drop;
+		}
+	}
+
+	/*
+	 * A GRH is expected to preceed the data even if not
+	 * present on the wire.
+	 */
+	length = swqe->length;
+	memset(&wc, 0, sizeof wc);
+	wc.byte_len = length + sizeof(struct ib_grh);
+
+	if (swqe->wr.opcode == IB_WR_SEND_WITH_IMM) {
+		wc.wc_flags = IB_WC_WITH_IMM;
+		wc.ex.imm_data = swqe->wr.ex.imm_data;
+	}
+
+	spin_lock_irqsave(&qp->r_lock, flags);
+
+	/*
+	 * Get the next work request entry to find where to put the data.
+	 */
+	if (qp->r_flags & QIB_R_REUSE_SGE)
+		qp->r_flags &= ~QIB_R_REUSE_SGE;
+	else {
+		int ret;
+
+		ret = qib_get_rwqe(qp, 0);
+		if (ret < 0) {
+			qib_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
+			goto bail_unlock;
+		}
+		if (!ret) {
+			if (qp->ibqp.qp_num == 0)
+				ibp->n_vl15_dropped++;
+			goto bail_unlock;
+		}
+	}
+	/* Silently drop packets which are too big. */
+	if (unlikely(wc.byte_len > qp->r_len)) {
+		qp->r_flags |= QIB_R_REUSE_SGE;
+		ibp->n_pkt_drops++;
+		goto bail_unlock;
+	}
+
+	if (ah_attr->ah_flags & IB_AH_GRH) {
+		qib_copy_sge(&qp->r_sge, &ah_attr->grh,
+			     sizeof(struct ib_grh), 1);
+		wc.wc_flags |= IB_WC_GRH;
+	} else
+		qib_skip_sge(&qp->r_sge, sizeof(struct ib_grh), 1);
+	ssge.sg_list = swqe->sg_list + 1;
+	ssge.sge = *swqe->sg_list;
+	ssge.num_sge = swqe->wr.num_sge;
+	sge = &ssge.sge;
+	while (length) {
+		u32 len = sge->length;
+
+		if (len > length)
+			len = length;
+		if (len > sge->sge_length)
+			len = sge->sge_length;
+		BUG_ON(len == 0);
+		qib_copy_sge(&qp->r_sge, sge->vaddr, len, 1);
+		sge->vaddr += len;
+		sge->length -= len;
+		sge->sge_length -= len;
+		if (sge->sge_length == 0) {
+			if (--ssge.num_sge)
+				*sge = *ssge.sg_list++;
+		} else if (sge->length == 0 && sge->mr->lkey) {
+			if (++sge->n >= QIB_SEGSZ) {
+				if (++sge->m >= sge->mr->mapsz)
+					break;
+				sge->n = 0;
+			}
+			sge->vaddr =
+				sge->mr->map[sge->m]->segs[sge->n].vaddr;
+			sge->length =
+				sge->mr->map[sge->m]->segs[sge->n].length;
+		}
+		length -= len;
+	}
+	while (qp->r_sge.num_sge) {
+		atomic_dec(&qp->r_sge.sge.mr->refcount);
+		if (--qp->r_sge.num_sge)
+			qp->r_sge.sge = *qp->r_sge.sg_list++;
+	}
+	if (!test_and_clear_bit(QIB_R_WRID_VALID, &qp->r_aflags))
+		goto bail_unlock;
+	wc.wr_id = qp->r_wr_id;
+	wc.status = IB_WC_SUCCESS;
+	wc.opcode = IB_WC_RECV;
+	wc.qp = &qp->ibqp;
+	wc.src_qp = sqp->ibqp.qp_num;
+	wc.pkey_index = qp->ibqp.qp_type == IB_QPT_GSI ?
+		swqe->wr.wr.ud.pkey_index : 0;
+	wc.slid = ppd->lid | (ah_attr->src_path_bits & ((1 << ppd->lmc) - 1));
+	wc.sl = ah_attr->sl;
+	wc.dlid_path_bits = ah_attr->dlid & ((1 << ppd->lmc) - 1);
+	wc.port_num = qp->port_num;
+	/* Signal completion event if the solicited bit is set. */
+	qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
+		     swqe->wr.send_flags & IB_SEND_SOLICITED);
+	ibp->n_loop_pkts++;
+bail_unlock:
+	spin_unlock_irqrestore(&qp->r_lock, flags);
+drop:
+	if (atomic_dec_and_test(&qp->refcount))
+		wake_up(&qp->wait);
+}
+
+/**
+ * qib_make_ud_req - construct a UD request packet
+ * @qp: the QP
+ *
+ * Return 1 if constructed; otherwise, return 0.
+ */
+int qib_make_ud_req(struct qib_qp *qp)
+{
+	struct qib_other_headers *ohdr;
+	struct ib_ah_attr *ah_attr;
+	struct qib_pportdata *ppd;
+	struct qib_ibport *ibp;
+	struct qib_swqe *wqe;
+	unsigned long flags;
+	u32 nwords;
+	u32 extra_bytes;
+	u32 bth0;
+	u16 lrh0;
+	u16 lid;
+	int ret = 0;
+	int next_cur;
+
+	spin_lock_irqsave(&qp->s_lock, flags);
+
+	if (!(ib_qib_state_ops[qp->state] & QIB_PROCESS_NEXT_SEND_OK)) {
+		if (!(ib_qib_state_ops[qp->state] & QIB_FLUSH_SEND))
+			goto bail;
+		/* We are in the error state, flush the work request. */
+		if (qp->s_last == qp->s_head)
+			goto bail;
+		/* If DMAs are in progress, we can't flush immediately. */
+		if (atomic_read(&qp->s_dma_busy)) {
+			qp->s_flags |= QIB_S_WAIT_DMA;
+			goto bail;
+		}
+		wqe = get_swqe_ptr(qp, qp->s_last);
+		qib_send_complete(qp, wqe, IB_WC_WR_FLUSH_ERR);
+		goto done;
+	}
+
+	if (qp->s_cur == qp->s_head)
+		goto bail;
+
+	wqe = get_swqe_ptr(qp, qp->s_cur);
+	next_cur = qp->s_cur + 1;
+	if (next_cur >= qp->s_size)
+		next_cur = 0;
+
+	/* Construct the header. */
+	ibp = to_iport(qp->ibqp.device, qp->port_num);
+	ppd = ppd_from_ibp(ibp);
+	ah_attr = &to_iah(wqe->wr.wr.ud.ah)->attr;
+	if (ah_attr->dlid >= QIB_MULTICAST_LID_BASE) {
+		if (ah_attr->dlid != QIB_PERMISSIVE_LID)
+			ibp->n_multicast_xmit++;
+		else
+			ibp->n_unicast_xmit++;
+	} else {
+		ibp->n_unicast_xmit++;
+		lid = ah_attr->dlid & ~((1 << ppd->lmc) - 1);
+		if (unlikely(lid == ppd->lid)) {
+			/*
+			 * If DMAs are in progress, we can't generate
+			 * a completion for the loopback packet since
+			 * it would be out of order.
+			 * XXX Instead of waiting, we could queue a
+			 * zero length descriptor so we get a callback.
+			 */
+			if (atomic_read(&qp->s_dma_busy)) {
+				qp->s_flags |= QIB_S_WAIT_DMA;
+				goto bail;
+			}
+			qp->s_cur = next_cur;
+			spin_unlock_irqrestore(&qp->s_lock, flags);
+			qib_ud_loopback(qp, wqe);
+			spin_lock_irqsave(&qp->s_lock, flags);
+			qib_send_complete(qp, wqe, IB_WC_SUCCESS);
+			goto done;
+		}
+	}
+
+	qp->s_cur = next_cur;
+	extra_bytes = -wqe->length & 3;
+	nwords = (wqe->length + extra_bytes) >> 2;
+
+	/* header size in 32-bit words LRH+BTH+DETH = (8+12+8)/4. */
+	qp->s_hdrwords = 7;
+	qp->s_cur_size = wqe->length;
+	qp->s_cur_sge = &qp->s_sge;
+	qp->s_srate = ah_attr->static_rate;
+	qp->s_wqe = wqe;
+	qp->s_sge.sge = wqe->sg_list[0];
+	qp->s_sge.sg_list = wqe->sg_list + 1;
+	qp->s_sge.num_sge = wqe->wr.num_sge;
+	qp->s_sge.total_len = wqe->length;
+
+	if (ah_attr->ah_flags & IB_AH_GRH) {
+		/* Header size in 32-bit words. */
+		qp->s_hdrwords += qib_make_grh(ibp, &qp->s_hdr.u.l.grh,
+					       &ah_attr->grh,
+					       qp->s_hdrwords, nwords);
+		lrh0 = QIB_LRH_GRH;
+		ohdr = &qp->s_hdr.u.l.oth;
+		/*
+		 * Don't worry about sending to locally attached multicast
+		 * QPs.  It is unspecified by the spec. what happens.
+		 */
+	} else {
+		/* Header size in 32-bit words. */
+		lrh0 = QIB_LRH_BTH;
+		ohdr = &qp->s_hdr.u.oth;
+	}
+	if (wqe->wr.opcode == IB_WR_SEND_WITH_IMM) {
+		qp->s_hdrwords++;
+		ohdr->u.ud.imm_data = wqe->wr.ex.imm_data;
+		bth0 = IB_OPCODE_UD_SEND_ONLY_WITH_IMMEDIATE << 24;
+	} else
+		bth0 = IB_OPCODE_UD_SEND_ONLY << 24;
+	lrh0 |= ah_attr->sl << 4;
+	if (qp->ibqp.qp_type == IB_QPT_SMI)
+		lrh0 |= 0xF000; /* Set VL (see ch. 13.5.3.1) */
+	else
+		lrh0 |= ibp->sl_to_vl[ah_attr->sl] << 12;
+	qp->s_hdr.lrh[0] = cpu_to_be16(lrh0);
+	qp->s_hdr.lrh[1] = cpu_to_be16(ah_attr->dlid);  /* DEST LID */
+	qp->s_hdr.lrh[2] = cpu_to_be16(qp->s_hdrwords + nwords + SIZE_OF_CRC);
+	lid = ppd->lid;
+	if (lid) {
+		lid |= ah_attr->src_path_bits & ((1 << ppd->lmc) - 1);
+		qp->s_hdr.lrh[3] = cpu_to_be16(lid);
+	} else
+		qp->s_hdr.lrh[3] = IB_LID_PERMISSIVE;
+	if (wqe->wr.send_flags & IB_SEND_SOLICITED)
+		bth0 |= IB_BTH_SOLICITED;
+	bth0 |= extra_bytes << 20;
+	bth0 |= qp->ibqp.qp_type == IB_QPT_SMI ? QIB_DEFAULT_P_KEY :
+		qib_get_pkey(ibp, qp->ibqp.qp_type == IB_QPT_GSI ?
+			     wqe->wr.wr.ud.pkey_index : qp->s_pkey_index);
+	ohdr->bth[0] = cpu_to_be32(bth0);
+	/*
+	 * Use the multicast QP if the destination LID is a multicast LID.
+	 */
+	ohdr->bth[1] = ah_attr->dlid >= QIB_MULTICAST_LID_BASE &&
+		ah_attr->dlid != QIB_PERMISSIVE_LID ?
+		cpu_to_be32(QIB_MULTICAST_QPN) :
+		cpu_to_be32(wqe->wr.wr.ud.remote_qpn);
+	ohdr->bth[2] = cpu_to_be32(qp->s_next_psn++ & QIB_PSN_MASK);
+	/*
+	 * Qkeys with the high order bit set mean use the
+	 * qkey from the QP context instead of the WR (see 10.2.5).
+	 */
+	ohdr->u.ud.deth[0] = cpu_to_be32((int)wqe->wr.wr.ud.remote_qkey < 0 ?
+					 qp->qkey : wqe->wr.wr.ud.remote_qkey);
+	ohdr->u.ud.deth[1] = cpu_to_be32(qp->ibqp.qp_num);
+
+done:
+	ret = 1;
+	goto unlock;
+
+bail:
+	qp->s_flags &= ~QIB_S_BUSY;
+unlock:
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+	return ret;
+}
+
+static unsigned qib_lookup_pkey(struct qib_ibport *ibp, u16 pkey)
+{
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	struct qib_devdata *dd = ppd->dd;
+	unsigned ctxt = ppd->hw_pidx;
+	unsigned i;
+
+	pkey &= 0x7fff;	/* remove limited/full membership bit */
+
+	for (i = 0; i < ARRAY_SIZE(dd->rcd[ctxt]->pkeys); ++i)
+		if ((dd->rcd[ctxt]->pkeys[i] & 0x7fff) == pkey)
+			return i;
+
+	/*
+	 * Should not get here, this means hardware failed to validate pkeys.
+	 * Punt and return index 0.
+	 */
+	return 0;
+}
+
+/**
+ * qib_ud_rcv - receive an incoming UD packet
+ * @ibp: the port the packet came in on
+ * @hdr: the packet header
+ * @has_grh: true if the packet has a GRH
+ * @data: the packet data
+ * @tlen: the packet length
+ * @qp: the QP the packet came on
+ *
+ * This is called from qib_qp_rcv() to process an incoming UD packet
+ * for the given QP.
+ * Called at interrupt level.
+ */
+void qib_ud_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr,
+		int has_grh, void *data, u32 tlen, struct qib_qp *qp)
+{
+	struct qib_other_headers *ohdr;
+	int opcode;
+	u32 hdrsize;
+	u32 pad;
+	struct ib_wc wc;
+	u32 qkey;
+	u32 src_qp;
+	u16 dlid;
+
+	/* Check for GRH */
+	if (!has_grh) {
+		ohdr = &hdr->u.oth;
+		hdrsize = 8 + 12 + 8;   /* LRH + BTH + DETH */
+	} else {
+		ohdr = &hdr->u.l.oth;
+		hdrsize = 8 + 40 + 12 + 8; /* LRH + GRH + BTH + DETH */
+	}
+	qkey = be32_to_cpu(ohdr->u.ud.deth[0]);
+	src_qp = be32_to_cpu(ohdr->u.ud.deth[1]) & QIB_QPN_MASK;
+
+	/* Get the number of bytes the message was padded by. */
+	pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
+	if (unlikely(tlen < (hdrsize + pad + 4))) {
+		/* Drop incomplete packets. */
+		ibp->n_pkt_drops++;
+		goto bail;
+	}
+	tlen -= hdrsize + pad + 4;
+
+	/*
+	 * Check that the permissive LID is only used on QP0
+	 * and the QKEY matches (see 9.6.1.4.1 and 9.6.1.5.1).
+	 */
+	if (qp->ibqp.qp_num) {
+		if (unlikely(hdr->lrh[1] == IB_LID_PERMISSIVE ||
+			     hdr->lrh[3] == IB_LID_PERMISSIVE)) {
+			ibp->n_pkt_drops++;
+			goto bail;
+		}
+		if (qp->ibqp.qp_num > 1) {
+			u16 pkey1, pkey2;
+
+			pkey1 = be32_to_cpu(ohdr->bth[0]);
+			pkey2 = qib_get_pkey(ibp, qp->s_pkey_index);
+			if (unlikely(!qib_pkey_ok(pkey1, pkey2))) {
+				qib_bad_pqkey(ibp, IB_NOTICE_TRAP_BAD_PKEY,
+					      pkey1,
+					      (be16_to_cpu(hdr->lrh[0]) >> 4) &
+						0xF,
+					      src_qp, qp->ibqp.qp_num,
+					      hdr->lrh[3], hdr->lrh[1]);
+				goto bail;
+			}
+		}
+		if (unlikely(qkey != qp->qkey)) {
+			qib_bad_pqkey(ibp, IB_NOTICE_TRAP_BAD_QKEY, qkey,
+				      (be16_to_cpu(hdr->lrh[0]) >> 4) & 0xF,
+				      src_qp, qp->ibqp.qp_num,
+				      hdr->lrh[3], hdr->lrh[1]);
+			goto bail;
+		}
+		/* Drop invalid MAD packets (see 13.5.3.1). */
+		if (unlikely(qp->ibqp.qp_num == 1 &&
+			     (tlen != 256 ||
+			      (be16_to_cpu(hdr->lrh[0]) >> 12) == 15))) {
+			ibp->n_pkt_drops++;
+			goto bail;
+		}
+	} else {
+		struct ib_smp *smp;
+
+		/* Drop invalid MAD packets (see 13.5.3.1). */
+		if (tlen != 256 || (be16_to_cpu(hdr->lrh[0]) >> 12) != 15) {
+			ibp->n_pkt_drops++;
+			goto bail;
+		}
+		smp = (struct ib_smp *) data;
+		if ((hdr->lrh[1] == IB_LID_PERMISSIVE ||
+		     hdr->lrh[3] == IB_LID_PERMISSIVE) &&
+		    smp->mgmt_class != IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) {
+			ibp->n_pkt_drops++;
+			goto bail;
+		}
+	}
+
+	/*
+	 * The opcode is in the low byte when its in network order
+	 * (top byte when in host order).
+	 */
+	opcode = be32_to_cpu(ohdr->bth[0]) >> 24;
+	if (qp->ibqp.qp_num > 1 &&
+	    opcode == IB_OPCODE_UD_SEND_ONLY_WITH_IMMEDIATE) {
+		wc.ex.imm_data = ohdr->u.ud.imm_data;
+		wc.wc_flags = IB_WC_WITH_IMM;
+		hdrsize += sizeof(u32);
+	} else if (opcode == IB_OPCODE_UD_SEND_ONLY) {
+		wc.ex.imm_data = 0;
+		wc.wc_flags = 0;
+	} else {
+		ibp->n_pkt_drops++;
+		goto bail;
+	}
+
+	/*
+	 * A GRH is expected to preceed the data even if not
+	 * present on the wire.
+	 */
+	wc.byte_len = tlen + sizeof(struct ib_grh);
+
+	/*
+	 * We need to serialize getting a receive work queue entry and
+	 * generating a completion for it against QPs sending to this QP
+	 * locally.
+	 */
+	spin_lock(&qp->r_lock);
+
+	/*
+	 * Get the next work request entry to find where to put the data.
+	 */
+	if (qp->r_flags & QIB_R_REUSE_SGE)
+		qp->r_flags &= ~QIB_R_REUSE_SGE;
+	else {
+		int ret;
+
+		ret = qib_get_rwqe(qp, 0);
+		if (ret < 0) {
+			qib_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
+			goto bail_unlock;
+		}
+		if (!ret) {
+			if (qp->ibqp.qp_num == 0)
+				ibp->n_vl15_dropped++;
+			goto bail_unlock;
+		}
+	}
+	/* Silently drop packets which are too big. */
+	if (unlikely(wc.byte_len > qp->r_len)) {
+		qp->r_flags |= QIB_R_REUSE_SGE;
+		ibp->n_pkt_drops++;
+		goto bail_unlock;
+	}
+	if (has_grh) {
+		qib_copy_sge(&qp->r_sge, &hdr->u.l.grh,
+			     sizeof(struct ib_grh), 1);
+		wc.wc_flags |= IB_WC_GRH;
+	} else
+		qib_skip_sge(&qp->r_sge, sizeof(struct ib_grh), 1);
+	qib_copy_sge(&qp->r_sge, data, wc.byte_len - sizeof(struct ib_grh), 1);
+	while (qp->r_sge.num_sge) {
+		atomic_dec(&qp->r_sge.sge.mr->refcount);
+		if (--qp->r_sge.num_sge)
+			qp->r_sge.sge = *qp->r_sge.sg_list++;
+	}
+	if (!test_and_clear_bit(QIB_R_WRID_VALID, &qp->r_aflags))
+		goto bail_unlock;
+	wc.wr_id = qp->r_wr_id;
+	wc.status = IB_WC_SUCCESS;
+	wc.opcode = IB_WC_RECV;
+	wc.vendor_err = 0;
+	wc.qp = &qp->ibqp;
+	wc.src_qp = src_qp;
+	wc.pkey_index = qp->ibqp.qp_type == IB_QPT_GSI ?
+		qib_lookup_pkey(ibp, be32_to_cpu(ohdr->bth[0])) : 0;
+	wc.slid = be16_to_cpu(hdr->lrh[3]);
+	wc.sl = (be16_to_cpu(hdr->lrh[0]) >> 4) & 0xF;
+	dlid = be16_to_cpu(hdr->lrh[1]);
+	/*
+	 * Save the LMC lower bits if the destination LID is a unicast LID.
+	 */
+	wc.dlid_path_bits = dlid >= QIB_MULTICAST_LID_BASE ? 0 :
+		dlid & ((1 << ppd_from_ibp(ibp)->lmc) - 1);
+	wc.port_num = qp->port_num;
+	/* Signal completion event if the solicited bit is set. */
+	qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
+		     (ohdr->bth[0] &
+			cpu_to_be32(IB_BTH_SOLICITED)) != 0);
+bail_unlock:
+	spin_unlock(&qp->r_lock);
+bail:;
+}

drivers/infiniband/hw/qib/qib_user_pages.c

@@ -0,0 +1,157 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/mm.h>
+#include <linux/device.h>
+
+#include "qib.h"
+
+static void __qib_release_user_pages(struct page **p, size_t num_pages,
+				     int dirty)
+{
+	size_t i;
+
+	for (i = 0; i < num_pages; i++) {
+		if (dirty)
+			set_page_dirty_lock(p[i]);
+		put_page(p[i]);
+	}
+}
+
+/*
+ * Call with current->mm->mmap_sem held.
+ */
+static int __get_user_pages(unsigned long start_page, size_t num_pages,
+			    struct page **p, struct vm_area_struct **vma)
+{
+	unsigned long lock_limit;
+	size_t got;
+	int ret;
+
+	lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
+
+	if (num_pages > lock_limit && !capable(CAP_IPC_LOCK)) {
+		ret = -ENOMEM;
+		goto bail;
+	}
+
+	for (got = 0; got < num_pages; got += ret) {
+		ret = get_user_pages(current, current->mm,
+				     start_page + got * PAGE_SIZE,
+				     num_pages - got, 1, 1,
+				     p + got, vma);
+		if (ret < 0)
+			goto bail_release;
+	}
+
+	current->mm->locked_vm += num_pages;
+
+	ret = 0;
+	goto bail;
+
+bail_release:
+	__qib_release_user_pages(p, got, 0);
+bail:
+	return ret;
+}
+
+/**
+ * qib_map_page - a safety wrapper around pci_map_page()
+ *
+ * A dma_addr of all 0's is interpreted by the chip as "disabled".
+ * Unfortunately, it can also be a valid dma_addr returned on some
+ * architectures.
+ *
+ * The powerpc iommu assigns dma_addrs in ascending order, so we don't
+ * have to bother with retries or mapping a dummy page to insure we
+ * don't just get the same mapping again.
+ *
+ * I'm sure we won't be so lucky with other iommu's, so FIXME.
+ */
+dma_addr_t qib_map_page(struct pci_dev *hwdev, struct page *page,
+			unsigned long offset, size_t size, int direction)
+{
+	dma_addr_t phys;
+
+	phys = pci_map_page(hwdev, page, offset, size, direction);
+
+	if (phys == 0) {
+		pci_unmap_page(hwdev, phys, size, direction);
+		phys = pci_map_page(hwdev, page, offset, size, direction);
+		/*
+		 * FIXME: If we get 0 again, we should keep this page,
+		 * map another, then free the 0 page.
+		 */
+	}
+
+	return phys;
+}
+
+/**
+ * qib_get_user_pages - lock user pages into memory
+ * @start_page: the start page
+ * @num_pages: the number of pages
+ * @p: the output page structures
+ *
+ * This function takes a given start page (page aligned user virtual
+ * address) and pins it and the following specified number of pages.  For
+ * now, num_pages is always 1, but that will probably change at some point
+ * (because caller is doing expected sends on a single virtually contiguous
+ * buffer, so we can do all pages at once).
+ */
+int qib_get_user_pages(unsigned long start_page, size_t num_pages,
+		       struct page **p)
+{
+	int ret;
+
+	down_write(&current->mm->mmap_sem);
+
+	ret = __get_user_pages(start_page, num_pages, p, NULL);
+
+	up_write(&current->mm->mmap_sem);
+
+	return ret;
+}
+
+void qib_release_user_pages(struct page **p, size_t num_pages)
+{
+	if (current->mm) /* during close after signal, mm can be NULL */
+		down_write(&current->mm->mmap_sem);
+
+	__qib_release_user_pages(p, num_pages, 1);
+
+	if (current->mm) {
+		current->mm->locked_vm -= num_pages;
+		up_write(&current->mm->mmap_sem);
+	}
+}

drivers/infiniband/hw/qib/qib_user_sdma.c

@@ -0,0 +1,897 @@
+/*
+ * Copyright (c) 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include <linux/mm.h>
+#include <linux/types.h>
+#include <linux/device.h>
+#include <linux/dmapool.h>
+#include <linux/slab.h>
+#include <linux/list.h>
+#include <linux/highmem.h>
+#include <linux/io.h>
+#include <linux/uio.h>
+#include <linux/rbtree.h>
+#include <linux/spinlock.h>
+#include <linux/delay.h>
+
+#include "qib.h"
+#include "qib_user_sdma.h"
+
+/* minimum size of header */
+#define QIB_USER_SDMA_MIN_HEADER_LENGTH 64
+/* expected size of headers (for dma_pool) */
+#define QIB_USER_SDMA_EXP_HEADER_LENGTH 64
+/* attempt to drain the queue for 5secs */
+#define QIB_USER_SDMA_DRAIN_TIMEOUT 500
+
+struct qib_user_sdma_pkt {
+	u8 naddr;               /* dimension of addr (1..3) ... */
+	u32 counter;            /* sdma pkts queued counter for this entry */
+	u64 added;              /* global descq number of entries */
+
+	struct {
+		u32 offset;                     /* offset for kvaddr, addr */
+		u32 length;                     /* length in page */
+		u8  put_page;                   /* should we put_page? */
+		u8  dma_mapped;                 /* is page dma_mapped? */
+		struct page *page;              /* may be NULL (coherent mem) */
+		void *kvaddr;                   /* FIXME: only for pio hack */
+		dma_addr_t addr;
+	} addr[4];   /* max pages, any more and we coalesce */
+	struct list_head list;  /* list element */
+};
+
+struct qib_user_sdma_queue {
+	/*
+	 * pkts sent to dma engine are queued on this
+	 * list head.  the type of the elements of this
+	 * list are struct qib_user_sdma_pkt...
+	 */
+	struct list_head sent;
+
+	/* headers with expected length are allocated from here... */
+	char header_cache_name[64];
+	struct dma_pool *header_cache;
+
+	/* packets are allocated from the slab cache... */
+	char pkt_slab_name[64];
+	struct kmem_cache *pkt_slab;
+
+	/* as packets go on the queued queue, they are counted... */
+	u32 counter;
+	u32 sent_counter;
+
+	/* dma page table */
+	struct rb_root dma_pages_root;
+
+	/* protect everything above... */
+	struct mutex lock;
+};
+
+struct qib_user_sdma_queue *
+qib_user_sdma_queue_create(struct device *dev, int unit, int ctxt, int sctxt)
+{
+	struct qib_user_sdma_queue *pq =
+		kmalloc(sizeof(struct qib_user_sdma_queue), GFP_KERNEL);
+
+	if (!pq)
+		goto done;
+
+	pq->counter = 0;
+	pq->sent_counter = 0;
+	INIT_LIST_HEAD(&pq->sent);
+
+	mutex_init(&pq->lock);
+
+	snprintf(pq->pkt_slab_name, sizeof(pq->pkt_slab_name),
+		 "qib-user-sdma-pkts-%u-%02u.%02u", unit, ctxt, sctxt);
+	pq->pkt_slab = kmem_cache_create(pq->pkt_slab_name,
+					 sizeof(struct qib_user_sdma_pkt),
+					 0, 0, NULL);
+
+	if (!pq->pkt_slab)
+		goto err_kfree;
+
+	snprintf(pq->header_cache_name, sizeof(pq->header_cache_name),
+		 "qib-user-sdma-headers-%u-%02u.%02u", unit, ctxt, sctxt);
+	pq->header_cache = dma_pool_create(pq->header_cache_name,
+					   dev,
+					   QIB_USER_SDMA_EXP_HEADER_LENGTH,
+					   4, 0);
+	if (!pq->header_cache)
+		goto err_slab;
+
+	pq->dma_pages_root = RB_ROOT;
+
+	goto done;
+
+err_slab:
+	kmem_cache_destroy(pq->pkt_slab);
+err_kfree:
+	kfree(pq);
+	pq = NULL;
+
+done:
+	return pq;
+}
+
+static void qib_user_sdma_init_frag(struct qib_user_sdma_pkt *pkt,
+				    int i, size_t offset, size_t len,
+				    int put_page, int dma_mapped,
+				    struct page *page,
+				    void *kvaddr, dma_addr_t dma_addr)
+{
+	pkt->addr[i].offset = offset;
+	pkt->addr[i].length = len;
+	pkt->addr[i].put_page = put_page;
+	pkt->addr[i].dma_mapped = dma_mapped;
+	pkt->addr[i].page = page;
+	pkt->addr[i].kvaddr = kvaddr;
+	pkt->addr[i].addr = dma_addr;
+}
+
+static void qib_user_sdma_init_header(struct qib_user_sdma_pkt *pkt,
+				      u32 counter, size_t offset,
+				      size_t len, int dma_mapped,
+				      struct page *page,
+				      void *kvaddr, dma_addr_t dma_addr)
+{
+	pkt->naddr = 1;
+	pkt->counter = counter;
+	qib_user_sdma_init_frag(pkt, 0, offset, len, 0, dma_mapped, page,
+				kvaddr, dma_addr);
+}
+
+/* we've too many pages in the iovec, coalesce to a single page */
+static int qib_user_sdma_coalesce(const struct qib_devdata *dd,
+				  struct qib_user_sdma_pkt *pkt,
+				  const struct iovec *iov,
+				  unsigned long niov)
+{
+	int ret = 0;
+	struct page *page = alloc_page(GFP_KERNEL);
+	void *mpage_save;
+	char *mpage;
+	int i;
+	int len = 0;
+	dma_addr_t dma_addr;
+
+	if (!page) {
+		ret = -ENOMEM;
+		goto done;
+	}
+
+	mpage = kmap(page);
+	mpage_save = mpage;
+	for (i = 0; i < niov; i++) {
+		int cfur;
+
+		cfur = copy_from_user(mpage,
+				      iov[i].iov_base, iov[i].iov_len);
+		if (cfur) {
+			ret = -EFAULT;
+			goto free_unmap;
+		}
+
+		mpage += iov[i].iov_len;
+		len += iov[i].iov_len;
+	}
+
+	dma_addr = dma_map_page(&dd->pcidev->dev, page, 0, len,
+				DMA_TO_DEVICE);
+	if (dma_mapping_error(&dd->pcidev->dev, dma_addr)) {
+		ret = -ENOMEM;
+		goto free_unmap;
+	}
+
+	qib_user_sdma_init_frag(pkt, 1, 0, len, 0, 1, page, mpage_save,
+				dma_addr);
+	pkt->naddr = 2;
+
+	goto done;
+
+free_unmap:
+	kunmap(page);
+	__free_page(page);
+done:
+	return ret;
+}
+
+/*
+ * How many pages in this iovec element?
+ */
+static int qib_user_sdma_num_pages(const struct iovec *iov)
+{
+	const unsigned long addr  = (unsigned long) iov->iov_base;
+	const unsigned long  len  = iov->iov_len;
+	const unsigned long spage = addr & PAGE_MASK;
+	const unsigned long epage = (addr + len - 1) & PAGE_MASK;
+
+	return 1 + ((epage - spage) >> PAGE_SHIFT);
+}
+
+/*
+ * Truncate length to page boundry.
+ */
+static int qib_user_sdma_page_length(unsigned long addr, unsigned long len)
+{
+	const unsigned long offset = addr & ~PAGE_MASK;
+
+	return ((offset + len) > PAGE_SIZE) ? (PAGE_SIZE - offset) : len;
+}
+
+static void qib_user_sdma_free_pkt_frag(struct device *dev,
+					struct qib_user_sdma_queue *pq,
+					struct qib_user_sdma_pkt *pkt,
+					int frag)
+{
+	const int i = frag;
+
+	if (pkt->addr[i].page) {
+		if (pkt->addr[i].dma_mapped)
+			dma_unmap_page(dev,
+				       pkt->addr[i].addr,
+				       pkt->addr[i].length,
+				       DMA_TO_DEVICE);
+
+		if (pkt->addr[i].kvaddr)
+			kunmap(pkt->addr[i].page);
+
+		if (pkt->addr[i].put_page)
+			put_page(pkt->addr[i].page);
+		else
+			__free_page(pkt->addr[i].page);
+	} else if (pkt->addr[i].kvaddr)
+		/* free coherent mem from cache... */
+		dma_pool_free(pq->header_cache,
+			      pkt->addr[i].kvaddr, pkt->addr[i].addr);
+}
+
+/* return number of pages pinned... */
+static int qib_user_sdma_pin_pages(const struct qib_devdata *dd,
+				   struct qib_user_sdma_pkt *pkt,
+				   unsigned long addr, int tlen, int npages)
+{
+	struct page *pages[2];
+	int j;
+	int ret;
+
+	ret = get_user_pages(current, current->mm, addr,
+			     npages, 0, 1, pages, NULL);
+
+	if (ret != npages) {
+		int i;
+
+		for (i = 0; i < ret; i++)
+			put_page(pages[i]);
+
+		ret = -ENOMEM;
+		goto done;
+	}
+
+	for (j = 0; j < npages; j++) {
+		/* map the pages... */
+		const int flen = qib_user_sdma_page_length(addr, tlen);
+		dma_addr_t dma_addr =
+			dma_map_page(&dd->pcidev->dev,
+				     pages[j], 0, flen, DMA_TO_DEVICE);
+		unsigned long fofs = addr & ~PAGE_MASK;
+
+		if (dma_mapping_error(&dd->pcidev->dev, dma_addr)) {
+			ret = -ENOMEM;
+			goto done;
+		}
+
+		qib_user_sdma_init_frag(pkt, pkt->naddr, fofs, flen, 1, 1,
+					pages[j], kmap(pages[j]), dma_addr);
+
+		pkt->naddr++;
+		addr += flen;
+		tlen -= flen;
+	}
+
+done:
+	return ret;
+}
+
+static int qib_user_sdma_pin_pkt(const struct qib_devdata *dd,
+				 struct qib_user_sdma_queue *pq,
+				 struct qib_user_sdma_pkt *pkt,
+				 const struct iovec *iov,
+				 unsigned long niov)
+{
+	int ret = 0;
+	unsigned long idx;
+
+	for (idx = 0; idx < niov; idx++) {
+		const int npages = qib_user_sdma_num_pages(iov + idx);
+		const unsigned long addr = (unsigned long) iov[idx].iov_base;
+
+		ret = qib_user_sdma_pin_pages(dd, pkt, addr,
+					      iov[idx].iov_len, npages);
+		if (ret < 0)
+			goto free_pkt;
+	}
+
+	goto done;
+
+free_pkt:
+	for (idx = 0; idx < pkt->naddr; idx++)
+		qib_user_sdma_free_pkt_frag(&dd->pcidev->dev, pq, pkt, idx);
+
+done:
+	return ret;
+}
+
+static int qib_user_sdma_init_payload(const struct qib_devdata *dd,
+				      struct qib_user_sdma_queue *pq,
+				      struct qib_user_sdma_pkt *pkt,
+				      const struct iovec *iov,
+				      unsigned long niov, int npages)
+{
+	int ret = 0;
+
+	if (npages >= ARRAY_SIZE(pkt->addr))
+		ret = qib_user_sdma_coalesce(dd, pkt, iov, niov);
+	else
+		ret = qib_user_sdma_pin_pkt(dd, pq, pkt, iov, niov);
+
+	return ret;
+}
+
+/* free a packet list -- return counter value of last packet */
+static void qib_user_sdma_free_pkt_list(struct device *dev,
+					struct qib_user_sdma_queue *pq,
+					struct list_head *list)
+{
+	struct qib_user_sdma_pkt *pkt, *pkt_next;
+
+	list_for_each_entry_safe(pkt, pkt_next, list, list) {
+		int i;
+
+		for (i = 0; i < pkt->naddr; i++)
+			qib_user_sdma_free_pkt_frag(dev, pq, pkt, i);
+
+		kmem_cache_free(pq->pkt_slab, pkt);
+	}
+}
+
+/*
+ * copy headers, coalesce etc -- pq->lock must be held
+ *
+ * we queue all the packets to list, returning the
+ * number of bytes total.  list must be empty initially,
+ * as, if there is an error we clean it...
+ */
+static int qib_user_sdma_queue_pkts(const struct qib_devdata *dd,
+				    struct qib_user_sdma_queue *pq,
+				    struct list_head *list,
+				    const struct iovec *iov,
+				    unsigned long niov,
+				    int maxpkts)
+{
+	unsigned long idx = 0;
+	int ret = 0;
+	int npkts = 0;
+	struct page *page = NULL;
+	__le32 *pbc;
+	dma_addr_t dma_addr;
+	struct qib_user_sdma_pkt *pkt = NULL;
+	size_t len;
+	size_t nw;
+	u32 counter = pq->counter;
+	int dma_mapped = 0;
+
+	while (idx < niov && npkts < maxpkts) {
+		const unsigned long addr = (unsigned long) iov[idx].iov_base;
+		const unsigned long idx_save = idx;
+		unsigned pktnw;
+		unsigned pktnwc;
+		int nfrags = 0;
+		int npages = 0;
+		int cfur;
+
+		dma_mapped = 0;
+		len = iov[idx].iov_len;
+		nw = len >> 2;
+		page = NULL;
+
+		pkt = kmem_cache_alloc(pq->pkt_slab, GFP_KERNEL);
+		if (!pkt) {
+			ret = -ENOMEM;
+			goto free_list;
+		}
+
+		if (len < QIB_USER_SDMA_MIN_HEADER_LENGTH ||
+		    len > PAGE_SIZE || len & 3 || addr & 3) {
+			ret = -EINVAL;
+			goto free_pkt;
+		}
+
+		if (len == QIB_USER_SDMA_EXP_HEADER_LENGTH)
+			pbc = dma_pool_alloc(pq->header_cache, GFP_KERNEL,
+					     &dma_addr);
+		else
+			pbc = NULL;
+
+		if (!pbc) {
+			page = alloc_page(GFP_KERNEL);
+			if (!page) {
+				ret = -ENOMEM;
+				goto free_pkt;
+			}
+			pbc = kmap(page);
+		}
+
+		cfur = copy_from_user(pbc, iov[idx].iov_base, len);
+		if (cfur) {
+			ret = -EFAULT;
+			goto free_pbc;
+		}
+
+		/*
+		 * This assignment is a bit strange.  it's because the
+		 * the pbc counts the number of 32 bit words in the full
+		 * packet _except_ the first word of the pbc itself...
+		 */
+		pktnwc = nw - 1;
+
+		/*
+		 * pktnw computation yields the number of 32 bit words
+		 * that the caller has indicated in the PBC.  note that
+		 * this is one less than the total number of words that
+		 * goes to the send DMA engine as the first 32 bit word
+		 * of the PBC itself is not counted.  Armed with this count,
+		 * we can verify that the packet is consistent with the
+		 * iovec lengths.
+		 */
+		pktnw = le32_to_cpu(*pbc) & QIB_PBC_LENGTH_MASK;
+		if (pktnw < pktnwc || pktnw > pktnwc + (PAGE_SIZE >> 2)) {
+			ret = -EINVAL;
+			goto free_pbc;
+		}
+
+		idx++;
+		while (pktnwc < pktnw && idx < niov) {
+			const size_t slen = iov[idx].iov_len;
+			const unsigned long faddr =
+				(unsigned long) iov[idx].iov_base;
+
+			if (slen & 3 || faddr & 3 || !slen ||
+			    slen > PAGE_SIZE) {
+				ret = -EINVAL;
+				goto free_pbc;
+			}
+
+			npages++;
+			if ((faddr & PAGE_MASK) !=
+			    ((faddr + slen - 1) & PAGE_MASK))
+				npages++;
+
+			pktnwc += slen >> 2;
+			idx++;
+			nfrags++;
+		}
+
+		if (pktnwc != pktnw) {
+			ret = -EINVAL;
+			goto free_pbc;
+		}
+
+		if (page) {
+			dma_addr = dma_map_page(&dd->pcidev->dev,
+						page, 0, len, DMA_TO_DEVICE);
+			if (dma_mapping_error(&dd->pcidev->dev, dma_addr)) {
+				ret = -ENOMEM;
+				goto free_pbc;
+			}
+
+			dma_mapped = 1;
+		}
+
+		qib_user_sdma_init_header(pkt, counter, 0, len, dma_mapped,
+					  page, pbc, dma_addr);
+
+		if (nfrags) {
+			ret = qib_user_sdma_init_payload(dd, pq, pkt,
+							 iov + idx_save + 1,
+							 nfrags, npages);
+			if (ret < 0)
+				goto free_pbc_dma;
+		}
+
+		counter++;
+		npkts++;
+
+		list_add_tail(&pkt->list, list);
+	}
+
+	ret = idx;
+	goto done;
+
+free_pbc_dma:
+	if (dma_mapped)
+		dma_unmap_page(&dd->pcidev->dev, dma_addr, len, DMA_TO_DEVICE);
+free_pbc:
+	if (page) {
+		kunmap(page);
+		__free_page(page);
+	} else
+		dma_pool_free(pq->header_cache, pbc, dma_addr);
+free_pkt:
+	kmem_cache_free(pq->pkt_slab, pkt);
+free_list:
+	qib_user_sdma_free_pkt_list(&dd->pcidev->dev, pq, list);
+done:
+	return ret;
+}
+
+static void qib_user_sdma_set_complete_counter(struct qib_user_sdma_queue *pq,
+					       u32 c)
+{
+	pq->sent_counter = c;
+}
+
+/* try to clean out queue -- needs pq->lock */
+static int qib_user_sdma_queue_clean(struct qib_pportdata *ppd,
+				     struct qib_user_sdma_queue *pq)
+{
+	struct qib_devdata *dd = ppd->dd;
+	struct list_head free_list;
+	struct qib_user_sdma_pkt *pkt;
+	struct qib_user_sdma_pkt *pkt_prev;
+	int ret = 0;
+
+	INIT_LIST_HEAD(&free_list);
+
+	list_for_each_entry_safe(pkt, pkt_prev, &pq->sent, list) {
+		s64 descd = ppd->sdma_descq_removed - pkt->added;
+
+		if (descd < 0)
+			break;
+
+		list_move_tail(&pkt->list, &free_list);
+
+		/* one more packet cleaned */
+		ret++;
+	}
+
+	if (!list_empty(&free_list)) {
+		u32 counter;
+
+		pkt = list_entry(free_list.prev,
+				 struct qib_user_sdma_pkt, list);
+		counter = pkt->counter;
+
+		qib_user_sdma_free_pkt_list(&dd->pcidev->dev, pq, &free_list);
+		qib_user_sdma_set_complete_counter(pq, counter);
+	}
+
+	return ret;
+}
+
+void qib_user_sdma_queue_destroy(struct qib_user_sdma_queue *pq)
+{
+	if (!pq)
+		return;
+
+	kmem_cache_destroy(pq->pkt_slab);
+	dma_pool_destroy(pq->header_cache);
+	kfree(pq);
+}
+
+/* clean descriptor queue, returns > 0 if some elements cleaned */
+static int qib_user_sdma_hwqueue_clean(struct qib_pportdata *ppd)
+{
+	int ret;
+	unsigned long flags;
+
+	spin_lock_irqsave(&ppd->sdma_lock, flags);
+	ret = qib_sdma_make_progress(ppd);
+	spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+
+	return ret;
+}
+
+/* we're in close, drain packets so that we can cleanup successfully... */
+void qib_user_sdma_queue_drain(struct qib_pportdata *ppd,
+			       struct qib_user_sdma_queue *pq)
+{
+	struct qib_devdata *dd = ppd->dd;
+	int i;
+
+	if (!pq)
+		return;
+
+	for (i = 0; i < QIB_USER_SDMA_DRAIN_TIMEOUT; i++) {
+		mutex_lock(&pq->lock);
+		if (list_empty(&pq->sent)) {
+			mutex_unlock(&pq->lock);
+			break;
+		}
+		qib_user_sdma_hwqueue_clean(ppd);
+		qib_user_sdma_queue_clean(ppd, pq);
+		mutex_unlock(&pq->lock);
+		msleep(10);
+	}
+
+	if (!list_empty(&pq->sent)) {
+		struct list_head free_list;
+
+		qib_dev_err(dd, "user sdma lists not empty: forcing!\n");
+		INIT_LIST_HEAD(&free_list);
+		mutex_lock(&pq->lock);
+		list_splice_init(&pq->sent, &free_list);
+		qib_user_sdma_free_pkt_list(&dd->pcidev->dev, pq, &free_list);
+		mutex_unlock(&pq->lock);
+	}
+}
+
+static inline __le64 qib_sdma_make_desc0(struct qib_pportdata *ppd,
+					 u64 addr, u64 dwlen, u64 dwoffset)
+{
+	u8 tmpgen;
+
+	tmpgen = ppd->sdma_generation;
+
+	return cpu_to_le64(/* SDmaPhyAddr[31:0] */
+			   ((addr & 0xfffffffcULL) << 32) |
+			   /* SDmaGeneration[1:0] */
+			   ((tmpgen & 3ULL) << 30) |
+			   /* SDmaDwordCount[10:0] */
+			   ((dwlen & 0x7ffULL) << 16) |
+			   /* SDmaBufOffset[12:2] */
+			   (dwoffset & 0x7ffULL));
+}
+
+static inline __le64 qib_sdma_make_first_desc0(__le64 descq)
+{
+	return descq | cpu_to_le64(1ULL << 12);
+}
+
+static inline __le64 qib_sdma_make_last_desc0(__le64 descq)
+{
+					      /* last */  /* dma head */
+	return descq | cpu_to_le64(1ULL << 11 | 1ULL << 13);
+}
+
+static inline __le64 qib_sdma_make_desc1(u64 addr)
+{
+	/* SDmaPhyAddr[47:32] */
+	return cpu_to_le64(addr >> 32);
+}
+
+static void qib_user_sdma_send_frag(struct qib_pportdata *ppd,
+				    struct qib_user_sdma_pkt *pkt, int idx,
+				    unsigned ofs, u16 tail)
+{
+	const u64 addr = (u64) pkt->addr[idx].addr +
+		(u64) pkt->addr[idx].offset;
+	const u64 dwlen = (u64) pkt->addr[idx].length / 4;
+	__le64 *descqp;
+	__le64 descq0;
+
+	descqp = &ppd->sdma_descq[tail].qw[0];
+
+	descq0 = qib_sdma_make_desc0(ppd, addr, dwlen, ofs);
+	if (idx == 0)
+		descq0 = qib_sdma_make_first_desc0(descq0);
+	if (idx == pkt->naddr - 1)
+		descq0 = qib_sdma_make_last_desc0(descq0);
+
+	descqp[0] = descq0;
+	descqp[1] = qib_sdma_make_desc1(addr);
+}
+
+/* pq->lock must be held, get packets on the wire... */
+static int qib_user_sdma_push_pkts(struct qib_pportdata *ppd,
+				   struct qib_user_sdma_queue *pq,
+				   struct list_head *pktlist)
+{
+	struct qib_devdata *dd = ppd->dd;
+	int ret = 0;
+	unsigned long flags;
+	u16 tail;
+	u8 generation;
+	u64 descq_added;
+
+	if (list_empty(pktlist))
+		return 0;
+
+	if (unlikely(!(ppd->lflags & QIBL_LINKACTIVE)))
+		return -ECOMM;
+
+	spin_lock_irqsave(&ppd->sdma_lock, flags);
+
+	/* keep a copy for restoring purposes in case of problems */
+	generation = ppd->sdma_generation;
+	descq_added = ppd->sdma_descq_added;
+
+	if (unlikely(!__qib_sdma_running(ppd))) {
+		ret = -ECOMM;
+		goto unlock;
+	}
+
+	tail = ppd->sdma_descq_tail;
+	while (!list_empty(pktlist)) {
+		struct qib_user_sdma_pkt *pkt =
+			list_entry(pktlist->next, struct qib_user_sdma_pkt,
+				   list);
+		int i;
+		unsigned ofs = 0;
+		u16 dtail = tail;
+
+		if (pkt->naddr > qib_sdma_descq_freecnt(ppd))
+			goto unlock_check_tail;
+
+		for (i = 0; i < pkt->naddr; i++) {
+			qib_user_sdma_send_frag(ppd, pkt, i, ofs, tail);
+			ofs += pkt->addr[i].length >> 2;
+
+			if (++tail == ppd->sdma_descq_cnt) {
+				tail = 0;
+				++ppd->sdma_generation;
+			}
+		}
+
+		if ((ofs << 2) > ppd->ibmaxlen) {
+			ret = -EMSGSIZE;
+			goto unlock;
+		}
+
+		/*
+		 * If the packet is >= 2KB mtu equivalent, we have to use
+		 * the large buffers, and have to mark each descriptor as
+		 * part of a large buffer packet.
+		 */
+		if (ofs > dd->piosize2kmax_dwords) {
+			for (i = 0; i < pkt->naddr; i++) {
+				ppd->sdma_descq[dtail].qw[0] |=
+					cpu_to_le64(1ULL << 14);
+				if (++dtail == ppd->sdma_descq_cnt)
+					dtail = 0;
+			}
+		}
+
+		ppd->sdma_descq_added += pkt->naddr;
+		pkt->added = ppd->sdma_descq_added;
+		list_move_tail(&pkt->list, &pq->sent);
+		ret++;
+	}
+
+unlock_check_tail:
+	/* advance the tail on the chip if necessary */
+	if (ppd->sdma_descq_tail != tail)
+		dd->f_sdma_update_tail(ppd, tail);
+
+unlock:
+	if (unlikely(ret < 0)) {
+		ppd->sdma_generation = generation;
+		ppd->sdma_descq_added = descq_added;
+	}
+	spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+
+	return ret;
+}
+
+int qib_user_sdma_writev(struct qib_ctxtdata *rcd,
+			 struct qib_user_sdma_queue *pq,
+			 const struct iovec *iov,
+			 unsigned long dim)
+{
+	struct qib_devdata *dd = rcd->dd;
+	struct qib_pportdata *ppd = rcd->ppd;
+	int ret = 0;
+	struct list_head list;
+	int npkts = 0;
+
+	INIT_LIST_HEAD(&list);
+
+	mutex_lock(&pq->lock);
+
+	/* why not -ECOMM like qib_user_sdma_push_pkts() below? */
+	if (!qib_sdma_running(ppd))
+		goto done_unlock;
+
+	if (ppd->sdma_descq_added != ppd->sdma_descq_removed) {
+		qib_user_sdma_hwqueue_clean(ppd);
+		qib_user_sdma_queue_clean(ppd, pq);
+	}
+
+	while (dim) {
+		const int mxp = 8;
+
+		down_write(&current->mm->mmap_sem);
+		ret = qib_user_sdma_queue_pkts(dd, pq, &list, iov, dim, mxp);
+		up_write(&current->mm->mmap_sem);
+
+		if (ret <= 0)
+			goto done_unlock;
+		else {
+			dim -= ret;
+			iov += ret;
+		}
+
+		/* force packets onto the sdma hw queue... */
+		if (!list_empty(&list)) {
+			/*
+			 * Lazily clean hw queue.  the 4 is a guess of about
+			 * how many sdma descriptors a packet will take (it
+			 * doesn't have to be perfect).
+			 */
+			if (qib_sdma_descq_freecnt(ppd) < ret * 4) {
+				qib_user_sdma_hwqueue_clean(ppd);
+				qib_user_sdma_queue_clean(ppd, pq);
+			}
+
+			ret = qib_user_sdma_push_pkts(ppd, pq, &list);
+			if (ret < 0)
+				goto done_unlock;
+			else {
+				npkts += ret;
+				pq->counter += ret;
+
+				if (!list_empty(&list))
+					goto done_unlock;
+			}
+		}
+	}
+
+done_unlock:
+	if (!list_empty(&list))
+		qib_user_sdma_free_pkt_list(&dd->pcidev->dev, pq, &list);
+	mutex_unlock(&pq->lock);
+
+	return (ret < 0) ? ret : npkts;
+}
+
+int qib_user_sdma_make_progress(struct qib_pportdata *ppd,
+				struct qib_user_sdma_queue *pq)
+{
+	int ret = 0;
+
+	mutex_lock(&pq->lock);
+	qib_user_sdma_hwqueue_clean(ppd);
+	ret = qib_user_sdma_queue_clean(ppd, pq);
+	mutex_unlock(&pq->lock);
+
+	return ret;
+}
+
+u32 qib_user_sdma_complete_counter(const struct qib_user_sdma_queue *pq)
+{
+	return pq ? pq->sent_counter : 0;
+}
+
+u32 qib_user_sdma_inflight_counter(struct qib_user_sdma_queue *pq)
+{
+	return pq ? pq->counter : 0;
+}

drivers/infiniband/hw/qib/qib_user_sdma.h

@@ -0,0 +1,52 @@
+/*
+ * Copyright (c) 2007, 2008 QLogic Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include <linux/device.h>
+
+struct qib_user_sdma_queue;
+
+struct qib_user_sdma_queue *
+qib_user_sdma_queue_create(struct device *dev, int unit, int port, int sport);
+void qib_user_sdma_queue_destroy(struct qib_user_sdma_queue *pq);
+
+int qib_user_sdma_writev(struct qib_ctxtdata *pd,
+			 struct qib_user_sdma_queue *pq,
+			 const struct iovec *iov,
+			 unsigned long dim);
+
+int qib_user_sdma_make_progress(struct qib_pportdata *ppd,
+				struct qib_user_sdma_queue *pq);
+
+void qib_user_sdma_queue_drain(struct qib_pportdata *ppd,
+			       struct qib_user_sdma_queue *pq);
+
+u32 qib_user_sdma_complete_counter(const struct qib_user_sdma_queue *pq);
+u32 qib_user_sdma_inflight_counter(struct qib_user_sdma_queue *pq);

drivers/infiniband/hw/qib/qib_verbs_mcast.c

@@ -0,0 +1,368 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/rculist.h>
+
+#include "qib.h"
+
+/**
+ * qib_mcast_qp_alloc - alloc a struct to link a QP to mcast GID struct
+ * @qp: the QP to link
+ */
+static struct qib_mcast_qp *qib_mcast_qp_alloc(struct qib_qp *qp)
+{
+	struct qib_mcast_qp *mqp;
+
+	mqp = kmalloc(sizeof *mqp, GFP_KERNEL);
+	if (!mqp)
+		goto bail;
+
+	mqp->qp = qp;
+	atomic_inc(&qp->refcount);
+
+bail:
+	return mqp;
+}
+
+static void qib_mcast_qp_free(struct qib_mcast_qp *mqp)
+{
+	struct qib_qp *qp = mqp->qp;
+
+	/* Notify qib_destroy_qp() if it is waiting. */
+	if (atomic_dec_and_test(&qp->refcount))
+		wake_up(&qp->wait);
+
+	kfree(mqp);
+}
+
+/**
+ * qib_mcast_alloc - allocate the multicast GID structure
+ * @mgid: the multicast GID
+ *
+ * A list of QPs will be attached to this structure.
+ */
+static struct qib_mcast *qib_mcast_alloc(union ib_gid *mgid)
+{
+	struct qib_mcast *mcast;
+
+	mcast = kmalloc(sizeof *mcast, GFP_KERNEL);
+	if (!mcast)
+		goto bail;
+
+	mcast->mgid = *mgid;
+	INIT_LIST_HEAD(&mcast->qp_list);
+	init_waitqueue_head(&mcast->wait);
+	atomic_set(&mcast->refcount, 0);
+	mcast->n_attached = 0;
+
+bail:
+	return mcast;
+}
+
+static void qib_mcast_free(struct qib_mcast *mcast)
+{
+	struct qib_mcast_qp *p, *tmp;
+
+	list_for_each_entry_safe(p, tmp, &mcast->qp_list, list)
+		qib_mcast_qp_free(p);
+
+	kfree(mcast);
+}
+
+/**
+ * qib_mcast_find - search the global table for the given multicast GID
+ * @ibp: the IB port structure
+ * @mgid: the multicast GID to search for
+ *
+ * Returns NULL if not found.
+ *
+ * The caller is responsible for decrementing the reference count if found.
+ */
+struct qib_mcast *qib_mcast_find(struct qib_ibport *ibp, union ib_gid *mgid)
+{
+	struct rb_node *n;
+	unsigned long flags;
+	struct qib_mcast *mcast;
+
+	spin_lock_irqsave(&ibp->lock, flags);
+	n = ibp->mcast_tree.rb_node;
+	while (n) {
+		int ret;
+
+		mcast = rb_entry(n, struct qib_mcast, rb_node);
+
+		ret = memcmp(mgid->raw, mcast->mgid.raw,
+			     sizeof(union ib_gid));
+		if (ret < 0)
+			n = n->rb_left;
+		else if (ret > 0)
+			n = n->rb_right;
+		else {
+			atomic_inc(&mcast->refcount);
+			spin_unlock_irqrestore(&ibp->lock, flags);
+			goto bail;
+		}
+	}
+	spin_unlock_irqrestore(&ibp->lock, flags);
+
+	mcast = NULL;
+
+bail:
+	return mcast;
+}
+
+/**
+ * qib_mcast_add - insert mcast GID into table and attach QP struct
+ * @mcast: the mcast GID table
+ * @mqp: the QP to attach
+ *
+ * Return zero if both were added.  Return EEXIST if the GID was already in
+ * the table but the QP was added.  Return ESRCH if the QP was already
+ * attached and neither structure was added.
+ */
+static int qib_mcast_add(struct qib_ibdev *dev, struct qib_ibport *ibp,
+			 struct qib_mcast *mcast, struct qib_mcast_qp *mqp)
+{
+	struct rb_node **n = &ibp->mcast_tree.rb_node;
+	struct rb_node *pn = NULL;
+	int ret;
+
+	spin_lock_irq(&ibp->lock);
+
+	while (*n) {
+		struct qib_mcast *tmcast;
+		struct qib_mcast_qp *p;
+
+		pn = *n;
+		tmcast = rb_entry(pn, struct qib_mcast, rb_node);
+
+		ret = memcmp(mcast->mgid.raw, tmcast->mgid.raw,
+			     sizeof(union ib_gid));
+		if (ret < 0) {
+			n = &pn->rb_left;
+			continue;
+		}
+		if (ret > 0) {
+			n = &pn->rb_right;
+			continue;
+		}
+
+		/* Search the QP list to see if this is already there. */
+		list_for_each_entry_rcu(p, &tmcast->qp_list, list) {
+			if (p->qp == mqp->qp) {
+				ret = ESRCH;
+				goto bail;
+			}
+		}
+		if (tmcast->n_attached == ib_qib_max_mcast_qp_attached) {
+			ret = ENOMEM;
+			goto bail;
+		}
+
+		tmcast->n_attached++;
+
+		list_add_tail_rcu(&mqp->list, &tmcast->qp_list);
+		ret = EEXIST;
+		goto bail;
+	}
+
+	spin_lock(&dev->n_mcast_grps_lock);
+	if (dev->n_mcast_grps_allocated == ib_qib_max_mcast_grps) {
+		spin_unlock(&dev->n_mcast_grps_lock);
+		ret = ENOMEM;
+		goto bail;
+	}
+
+	dev->n_mcast_grps_allocated++;
+	spin_unlock(&dev->n_mcast_grps_lock);
+
+	mcast->n_attached++;
+
+	list_add_tail_rcu(&mqp->list, &mcast->qp_list);
+
+	atomic_inc(&mcast->refcount);
+	rb_link_node(&mcast->rb_node, pn, n);
+	rb_insert_color(&mcast->rb_node, &ibp->mcast_tree);
+
+	ret = 0;
+
+bail:
+	spin_unlock_irq(&ibp->lock);
+
+	return ret;
+}
+
+int qib_multicast_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
+{
+	struct qib_qp *qp = to_iqp(ibqp);
+	struct qib_ibdev *dev = to_idev(ibqp->device);
+	struct qib_ibport *ibp;
+	struct qib_mcast *mcast;
+	struct qib_mcast_qp *mqp;
+	int ret;
+
+	if (ibqp->qp_num <= 1 || qp->state == IB_QPS_RESET) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	/*
+	 * Allocate data structures since its better to do this outside of
+	 * spin locks and it will most likely be needed.
+	 */
+	mcast = qib_mcast_alloc(gid);
+	if (mcast == NULL) {
+		ret = -ENOMEM;
+		goto bail;
+	}
+	mqp = qib_mcast_qp_alloc(qp);
+	if (mqp == NULL) {
+		qib_mcast_free(mcast);
+		ret = -ENOMEM;
+		goto bail;
+	}
+	ibp = to_iport(ibqp->device, qp->port_num);
+	switch (qib_mcast_add(dev, ibp, mcast, mqp)) {
+	case ESRCH:
+		/* Neither was used: OK to attach the same QP twice. */
+		qib_mcast_qp_free(mqp);
+		qib_mcast_free(mcast);
+		break;
+
+	case EEXIST:            /* The mcast wasn't used */
+		qib_mcast_free(mcast);
+		break;
+
+	case ENOMEM:
+		/* Exceeded the maximum number of mcast groups. */
+		qib_mcast_qp_free(mqp);
+		qib_mcast_free(mcast);
+		ret = -ENOMEM;
+		goto bail;
+
+	default:
+		break;
+	}
+
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+int qib_multicast_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
+{
+	struct qib_qp *qp = to_iqp(ibqp);
+	struct qib_ibdev *dev = to_idev(ibqp->device);
+	struct qib_ibport *ibp = to_iport(ibqp->device, qp->port_num);
+	struct qib_mcast *mcast = NULL;
+	struct qib_mcast_qp *p, *tmp;
+	struct rb_node *n;
+	int last = 0;
+	int ret;
+
+	if (ibqp->qp_num <= 1 || qp->state == IB_QPS_RESET) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	spin_lock_irq(&ibp->lock);
+
+	/* Find the GID in the mcast table. */
+	n = ibp->mcast_tree.rb_node;
+	while (1) {
+		if (n == NULL) {
+			spin_unlock_irq(&ibp->lock);
+			ret = -EINVAL;
+			goto bail;
+		}
+
+		mcast = rb_entry(n, struct qib_mcast, rb_node);
+		ret = memcmp(gid->raw, mcast->mgid.raw,
+			     sizeof(union ib_gid));
+		if (ret < 0)
+			n = n->rb_left;
+		else if (ret > 0)
+			n = n->rb_right;
+		else
+			break;
+	}
+
+	/* Search the QP list. */
+	list_for_each_entry_safe(p, tmp, &mcast->qp_list, list) {
+		if (p->qp != qp)
+			continue;
+		/*
+		 * We found it, so remove it, but don't poison the forward
+		 * link until we are sure there are no list walkers.
+		 */
+		list_del_rcu(&p->list);
+		mcast->n_attached--;
+
+		/* If this was the last attached QP, remove the GID too. */
+		if (list_empty(&mcast->qp_list)) {
+			rb_erase(&mcast->rb_node, &ibp->mcast_tree);
+			last = 1;
+		}
+		break;
+	}
+
+	spin_unlock_irq(&ibp->lock);
+
+	if (p) {
+		/*
+		 * Wait for any list walkers to finish before freeing the
+		 * list element.
+		 */
+		wait_event(mcast->wait, atomic_read(&mcast->refcount) <= 1);
+		qib_mcast_qp_free(p);
+	}
+	if (last) {
+		atomic_dec(&mcast->refcount);
+		wait_event(mcast->wait, !atomic_read(&mcast->refcount));
+		qib_mcast_free(mcast);
+		spin_lock_irq(&dev->n_mcast_grps_lock);
+		dev->n_mcast_grps_allocated--;
+		spin_unlock_irq(&dev->n_mcast_grps_lock);
+	}
+
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+int qib_mcast_tree_empty(struct qib_ibport *ibp)
+{
+	return ibp->mcast_tree.rb_node == NULL;
+}

drivers/infiniband/hw/qib/qib_wc_ppc64.c

@@ -0,0 +1,62 @@
+/*
+ * Copyright (c) 2006, 2007, 2008 QLogic Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+/*
+ * This file is conditionally built on PowerPC only.  Otherwise weak symbol
+ * versions of the functions exported from here are used.
+ */
+
+#include "qib.h"
+
+/**
+ * qib_enable_wc - enable write combining for MMIO writes to the device
+ * @dd: qlogic_ib device
+ *
+ * Nothing to do on PowerPC, so just return without error.
+ */
+int qib_enable_wc(struct qib_devdata *dd)
+{
+	return 0;
+}
+
+/**
+ * qib_unordered_wc - indicate whether write combining is unordered
+ *
+ * Because our performance depends on our ability to do write
+ * combining mmio writes in the most efficient way, we need to
+ * know if we are on a processor that may reorder stores when
+ * write combining.
+ */
+int qib_unordered_wc(void)
+{
+	return 1;
+}

drivers/infiniband/hw/qib/qib_wc_x86_64.c

@@ -0,0 +1,171 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+/*
+ * This file is conditionally built on x86_64 only.  Otherwise weak symbol
+ * versions of the functions exported from here are used.
+ */
+
+#include <linux/pci.h>
+#include <asm/mtrr.h>
+#include <asm/processor.h>
+
+#include "qib.h"
+
+/**
+ * qib_enable_wc - enable write combining for MMIO writes to the device
+ * @dd: qlogic_ib device
+ *
+ * This routine is x86_64-specific; it twiddles the CPU's MTRRs to enable
+ * write combining.
+ */
+int qib_enable_wc(struct qib_devdata *dd)
+{
+	int ret = 0;
+	u64 pioaddr, piolen;
+	unsigned bits;
+	const unsigned long addr = pci_resource_start(dd->pcidev, 0);
+	const size_t len = pci_resource_len(dd->pcidev, 0);
+
+	/*
+	 * Set the PIO buffers to be WCCOMB, so we get HT bursts to the
+	 * chip.  Linux (possibly the hardware) requires it to be on a power
+	 * of 2 address matching the length (which has to be a power of 2).
+	 * For rev1, that means the base address, for rev2, it will be just
+	 * the PIO buffers themselves.
+	 * For chips with two sets of buffers, the calculations are
+	 * somewhat more complicated; we need to sum, and the piobufbase
+	 * register has both offsets, 2K in low 32 bits, 4K in high 32 bits.
+	 * The buffers are still packed, so a single range covers both.
+	 */
+	if (dd->piobcnt2k && dd->piobcnt4k) {
+		/* 2 sizes for chip */
+		unsigned long pio2kbase, pio4kbase;
+		pio2kbase = dd->piobufbase & 0xffffffffUL;
+		pio4kbase = (dd->piobufbase >> 32) & 0xffffffffUL;
+		if (pio2kbase < pio4kbase) {
+			/* all current chips */
+			pioaddr = addr + pio2kbase;
+			piolen = pio4kbase - pio2kbase +
+				dd->piobcnt4k * dd->align4k;
+		} else {
+			pioaddr = addr + pio4kbase;
+			piolen = pio2kbase - pio4kbase +
+				dd->piobcnt2k * dd->palign;
+		}
+	} else {  /* single buffer size (2K, currently) */
+		pioaddr = addr + dd->piobufbase;
+		piolen = dd->piobcnt2k * dd->palign +
+			dd->piobcnt4k * dd->align4k;
+	}
+
+	for (bits = 0; !(piolen & (1ULL << bits)); bits++)
+		/* do nothing */ ;
+
+	if (piolen != (1ULL << bits)) {
+		piolen >>= bits;
+		while (piolen >>= 1)
+			bits++;
+		piolen = 1ULL << (bits + 1);
+	}
+	if (pioaddr & (piolen - 1)) {
+		u64 atmp;
+		atmp = pioaddr & ~(piolen - 1);
+		if (atmp < addr || (atmp + piolen) > (addr + len)) {
+			qib_dev_err(dd, "No way to align address/size "
+				    "(%llx/%llx), no WC mtrr\n",
+				    (unsigned long long) atmp,
+				    (unsigned long long) piolen << 1);
+			ret = -ENODEV;
+		} else {
+			pioaddr = atmp;
+			piolen <<= 1;
+		}
+	}
+
+	if (!ret) {
+		int cookie;
+
+		cookie = mtrr_add(pioaddr, piolen, MTRR_TYPE_WRCOMB, 0);
+		if (cookie < 0) {
+			{
+				qib_devinfo(dd->pcidev,
+					 "mtrr_add()  WC for PIO bufs "
+					 "failed (%d)\n",
+					 cookie);
+				ret = -EINVAL;
+			}
+		} else {
+			dd->wc_cookie = cookie;
+			dd->wc_base = (unsigned long) pioaddr;
+			dd->wc_len = (unsigned long) piolen;
+		}
+	}
+
+	return ret;
+}
+
+/**
+ * qib_disable_wc - disable write combining for MMIO writes to the device
+ * @dd: qlogic_ib device
+ */
+void qib_disable_wc(struct qib_devdata *dd)
+{
+	if (dd->wc_cookie) {
+		int r;
+
+		r = mtrr_del(dd->wc_cookie, dd->wc_base,
+			     dd->wc_len);
+		if (r < 0)
+			qib_devinfo(dd->pcidev,
+				 "mtrr_del(%lx, %lx, %lx) failed: %d\n",
+				 dd->wc_cookie, dd->wc_base,
+				 dd->wc_len, r);
+		dd->wc_cookie = 0; /* even on failure */
+	}
+}
+
+/**
+ * qib_unordered_wc - indicate whether write combining is ordered
+ *
+ * Because our performance depends on our ability to do write combining mmio
+ * writes in the most efficient way, we need to know if we are on an Intel
+ * or AMD x86_64 processor.  AMD x86_64 processors flush WC buffers out in
+ * the order completed, and so no special flushing is required to get
+ * correct ordering.  Intel processors, however, will flush write buffers
+ * out in "random" orders, and so explicit ordering is needed at times.
+ */
+int qib_unordered_wc(void)
+{
+	return boot_cpu_data.x86_vendor != X86_VENDOR_AMD;
+}