IB: Add vmw_pvrdma driver
This patch series adds a driver for a paravirtual RDMA device. The
device is developed for VMware's Virtual Machines and allows existing RDMA
applications to continue to use existing Verbs API when deployed in VMs
on ESXi. We recently did a presentation in the OFA Workshop [1] regarding
this device.
Description and RDMA Support
============================
The virtual device is exposed as a dual function PCIe device. One part
is a virtual network device (VMXNet3) which provides networking properties
like MAC, IP addresses to the RDMA part of the device. The networking
properties are used to register GIDs required by RDMA applications to
communicate.
These patches add support and the all required infrastructure for
letting applications use such a device. We support the mandatory Verbs API as
well as the base memory management extensions (Local Inv, Send with Inv and
Fast Register Work Requests). We currently support both Reliable Connected
and Unreliable Datagram QPs but do not support Shared Receive Queues
(SRQs).
Also, we support the following types of Work Requests:
o Send/Receive (with or without Immediate Data)
o RDMA Write (with or without Immediate Data)
o RDMA Read
o Local Invalidate
o Send with Invalidate
o Fast Register Work Requests
This version only adds support for version 1 of RoCE. We will add RoCEv2
support in a future patch. We do support registration of both MAC-based
and IP-based GIDs. I have also created a git tree for our user-level driver
[2].
Testing
=======
We have tested this internally for various types of Guest OS - Red Hat,
Centos, Ubuntu 12.04/14.04/16.04, Oracle Enterprise Linux, SLES 12
using backported versions of this driver. The tests included several
runs of the performance tests (included with OFED), Intel MPI PingPong
benchmark on OpenMPI, krping for FRWRs. Mellanox has been kind enough
to test the backported version of the driver internally on their hardware
using a VMware provided ESX build. I have also applied and tested this
with Doug's k.o/for-4.9 branch (commit 5603910b). Note, that this patch
series should be applied all together. I split out the commits so that
it may be easier to review.
PVRDMA Resources
================
[1] OFA Workshop Presentation -
https://openfabrics.org/images/eventpresos/2016presentations/102parardma.pdf
[2] Libpvrdma User-level library -
http://git.openfabrics.org/?p=~aditr/libpvrdma.git;a=summary
Reviewed-by: Jorgen Hansen <jhansen@vmware.com>
Reviewed-by: George Zhang <georgezhang@vmware.com>
Reviewed-by: Aditya Sarwade <asarwade@vmware.com>
Reviewed-by: Bryan Tan <bryantan@vmware.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Adit Ranadive <aditr@vmware.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
2016-10-03 02:10:22 +00:00
|
|
|
/*
|
|
|
|
* Copyright (c) 2012-2016 VMware, Inc. All rights reserved.
|
|
|
|
*
|
|
|
|
* This program is free software; you can redistribute it and/or
|
|
|
|
* modify it under the terms of EITHER the GNU General Public License
|
|
|
|
* version 2 as published by the Free Software Foundation or the BSD
|
|
|
|
* 2-Clause License. This program is distributed in the hope that it
|
|
|
|
* will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED
|
|
|
|
* WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
|
|
|
|
* See the GNU General Public License version 2 for more details at
|
|
|
|
* http://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html.
|
|
|
|
*
|
|
|
|
* You should have received a copy of the GNU General Public License
|
|
|
|
* along with this program available in the file COPYING in the main
|
|
|
|
* directory of this source tree.
|
|
|
|
*
|
|
|
|
* The BSD 2-Clause License
|
|
|
|
*
|
|
|
|
* 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.
|
|
|
|
*
|
|
|
|
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
|
|
|
|
* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
|
|
|
|
* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
|
|
|
|
* FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
|
|
|
|
* COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
|
|
|
|
* INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
|
|
|
|
* (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
|
|
|
|
* SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
|
|
|
|
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
|
|
|
|
* STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
|
|
|
|
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
|
|
|
|
* OF THE POSSIBILITY OF SUCH DAMAGE.
|
|
|
|
*/
|
|
|
|
|
|
|
|
#ifndef __PVRDMA_H__
|
|
|
|
#define __PVRDMA_H__
|
|
|
|
|
|
|
|
#include <linux/compiler.h>
|
|
|
|
#include <linux/interrupt.h>
|
|
|
|
#include <linux/list.h>
|
|
|
|
#include <linux/mutex.h>
|
|
|
|
#include <linux/pci.h>
|
|
|
|
#include <linux/semaphore.h>
|
|
|
|
#include <linux/workqueue.h>
|
|
|
|
#include <rdma/ib_umem.h>
|
|
|
|
#include <rdma/ib_verbs.h>
|
|
|
|
#include <rdma/vmw_pvrdma-abi.h>
|
|
|
|
|
|
|
|
#include "pvrdma_ring.h"
|
|
|
|
#include "pvrdma_dev_api.h"
|
|
|
|
#include "pvrdma_verbs.h"
|
|
|
|
|
|
|
|
/* NOT the same as BIT_MASK(). */
|
|
|
|
#define PVRDMA_MASK(n) ((n << 1) - 1)
|
|
|
|
|
|
|
|
/*
|
|
|
|
* VMware PVRDMA PCI device id.
|
|
|
|
*/
|
|
|
|
#define PCI_DEVICE_ID_VMWARE_PVRDMA 0x0820
|
|
|
|
|
2017-02-23 01:22:56 +00:00
|
|
|
#define PVRDMA_NUM_RING_PAGES 4
|
2017-02-23 01:22:57 +00:00
|
|
|
#define PVRDMA_QP_NUM_HEADER_PAGES 1
|
2017-02-23 01:22:56 +00:00
|
|
|
|
IB: Add vmw_pvrdma driver
This patch series adds a driver for a paravirtual RDMA device. The
device is developed for VMware's Virtual Machines and allows existing RDMA
applications to continue to use existing Verbs API when deployed in VMs
on ESXi. We recently did a presentation in the OFA Workshop [1] regarding
this device.
Description and RDMA Support
============================
The virtual device is exposed as a dual function PCIe device. One part
is a virtual network device (VMXNet3) which provides networking properties
like MAC, IP addresses to the RDMA part of the device. The networking
properties are used to register GIDs required by RDMA applications to
communicate.
These patches add support and the all required infrastructure for
letting applications use such a device. We support the mandatory Verbs API as
well as the base memory management extensions (Local Inv, Send with Inv and
Fast Register Work Requests). We currently support both Reliable Connected
and Unreliable Datagram QPs but do not support Shared Receive Queues
(SRQs).
Also, we support the following types of Work Requests:
o Send/Receive (with or without Immediate Data)
o RDMA Write (with or without Immediate Data)
o RDMA Read
o Local Invalidate
o Send with Invalidate
o Fast Register Work Requests
This version only adds support for version 1 of RoCE. We will add RoCEv2
support in a future patch. We do support registration of both MAC-based
and IP-based GIDs. I have also created a git tree for our user-level driver
[2].
Testing
=======
We have tested this internally for various types of Guest OS - Red Hat,
Centos, Ubuntu 12.04/14.04/16.04, Oracle Enterprise Linux, SLES 12
using backported versions of this driver. The tests included several
runs of the performance tests (included with OFED), Intel MPI PingPong
benchmark on OpenMPI, krping for FRWRs. Mellanox has been kind enough
to test the backported version of the driver internally on their hardware
using a VMware provided ESX build. I have also applied and tested this
with Doug's k.o/for-4.9 branch (commit 5603910b). Note, that this patch
series should be applied all together. I split out the commits so that
it may be easier to review.
PVRDMA Resources
================
[1] OFA Workshop Presentation -
https://openfabrics.org/images/eventpresos/2016presentations/102parardma.pdf
[2] Libpvrdma User-level library -
http://git.openfabrics.org/?p=~aditr/libpvrdma.git;a=summary
Reviewed-by: Jorgen Hansen <jhansen@vmware.com>
Reviewed-by: George Zhang <georgezhang@vmware.com>
Reviewed-by: Aditya Sarwade <asarwade@vmware.com>
Reviewed-by: Bryan Tan <bryantan@vmware.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Adit Ranadive <aditr@vmware.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
2016-10-03 02:10:22 +00:00
|
|
|
struct pvrdma_dev;
|
|
|
|
|
|
|
|
struct pvrdma_page_dir {
|
|
|
|
dma_addr_t dir_dma;
|
|
|
|
u64 *dir;
|
|
|
|
int ntables;
|
|
|
|
u64 **tables;
|
|
|
|
u64 npages;
|
|
|
|
void **pages;
|
|
|
|
};
|
|
|
|
|
|
|
|
struct pvrdma_cq {
|
|
|
|
struct ib_cq ibcq;
|
|
|
|
int offset;
|
|
|
|
spinlock_t cq_lock; /* Poll lock. */
|
|
|
|
struct pvrdma_uar_map *uar;
|
|
|
|
struct ib_umem *umem;
|
|
|
|
struct pvrdma_ring_state *ring_state;
|
|
|
|
struct pvrdma_page_dir pdir;
|
|
|
|
u32 cq_handle;
|
|
|
|
bool is_kernel;
|
2017-12-20 19:27:00 +00:00
|
|
|
refcount_t refcnt;
|
2017-12-20 17:51:40 +00:00
|
|
|
struct completion free;
|
IB: Add vmw_pvrdma driver
This patch series adds a driver for a paravirtual RDMA device. The
device is developed for VMware's Virtual Machines and allows existing RDMA
applications to continue to use existing Verbs API when deployed in VMs
on ESXi. We recently did a presentation in the OFA Workshop [1] regarding
this device.
Description and RDMA Support
============================
The virtual device is exposed as a dual function PCIe device. One part
is a virtual network device (VMXNet3) which provides networking properties
like MAC, IP addresses to the RDMA part of the device. The networking
properties are used to register GIDs required by RDMA applications to
communicate.
These patches add support and the all required infrastructure for
letting applications use such a device. We support the mandatory Verbs API as
well as the base memory management extensions (Local Inv, Send with Inv and
Fast Register Work Requests). We currently support both Reliable Connected
and Unreliable Datagram QPs but do not support Shared Receive Queues
(SRQs).
Also, we support the following types of Work Requests:
o Send/Receive (with or without Immediate Data)
o RDMA Write (with or without Immediate Data)
o RDMA Read
o Local Invalidate
o Send with Invalidate
o Fast Register Work Requests
This version only adds support for version 1 of RoCE. We will add RoCEv2
support in a future patch. We do support registration of both MAC-based
and IP-based GIDs. I have also created a git tree for our user-level driver
[2].
Testing
=======
We have tested this internally for various types of Guest OS - Red Hat,
Centos, Ubuntu 12.04/14.04/16.04, Oracle Enterprise Linux, SLES 12
using backported versions of this driver. The tests included several
runs of the performance tests (included with OFED), Intel MPI PingPong
benchmark on OpenMPI, krping for FRWRs. Mellanox has been kind enough
to test the backported version of the driver internally on their hardware
using a VMware provided ESX build. I have also applied and tested this
with Doug's k.o/for-4.9 branch (commit 5603910b). Note, that this patch
series should be applied all together. I split out the commits so that
it may be easier to review.
PVRDMA Resources
================
[1] OFA Workshop Presentation -
https://openfabrics.org/images/eventpresos/2016presentations/102parardma.pdf
[2] Libpvrdma User-level library -
http://git.openfabrics.org/?p=~aditr/libpvrdma.git;a=summary
Reviewed-by: Jorgen Hansen <jhansen@vmware.com>
Reviewed-by: George Zhang <georgezhang@vmware.com>
Reviewed-by: Aditya Sarwade <asarwade@vmware.com>
Reviewed-by: Bryan Tan <bryantan@vmware.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Adit Ranadive <aditr@vmware.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
2016-10-03 02:10:22 +00:00
|
|
|
};
|
|
|
|
|
|
|
|
struct pvrdma_id_table {
|
|
|
|
u32 last;
|
|
|
|
u32 top;
|
|
|
|
u32 max;
|
|
|
|
u32 mask;
|
|
|
|
spinlock_t lock; /* Table lock. */
|
|
|
|
unsigned long *table;
|
|
|
|
};
|
|
|
|
|
|
|
|
struct pvrdma_uar_map {
|
|
|
|
unsigned long pfn;
|
|
|
|
void __iomem *map;
|
|
|
|
int index;
|
|
|
|
};
|
|
|
|
|
|
|
|
struct pvrdma_uar_table {
|
|
|
|
struct pvrdma_id_table tbl;
|
|
|
|
int size;
|
|
|
|
};
|
|
|
|
|
|
|
|
struct pvrdma_ucontext {
|
|
|
|
struct ib_ucontext ibucontext;
|
|
|
|
struct pvrdma_dev *dev;
|
|
|
|
struct pvrdma_uar_map uar;
|
|
|
|
u64 ctx_handle;
|
|
|
|
};
|
|
|
|
|
|
|
|
struct pvrdma_pd {
|
|
|
|
struct ib_pd ibpd;
|
|
|
|
u32 pdn;
|
|
|
|
u32 pd_handle;
|
|
|
|
int privileged;
|
|
|
|
};
|
|
|
|
|
|
|
|
struct pvrdma_mr {
|
|
|
|
u32 mr_handle;
|
|
|
|
u64 iova;
|
|
|
|
u64 size;
|
|
|
|
};
|
|
|
|
|
|
|
|
struct pvrdma_user_mr {
|
|
|
|
struct ib_mr ibmr;
|
|
|
|
struct ib_umem *umem;
|
|
|
|
struct pvrdma_mr mmr;
|
|
|
|
struct pvrdma_page_dir pdir;
|
|
|
|
u64 *pages;
|
|
|
|
u32 npages;
|
|
|
|
u32 max_pages;
|
|
|
|
u32 page_shift;
|
|
|
|
};
|
|
|
|
|
|
|
|
struct pvrdma_wq {
|
|
|
|
struct pvrdma_ring *ring;
|
|
|
|
spinlock_t lock; /* Work queue lock. */
|
|
|
|
int wqe_cnt;
|
|
|
|
int wqe_size;
|
|
|
|
int max_sg;
|
|
|
|
int offset;
|
|
|
|
};
|
|
|
|
|
|
|
|
struct pvrdma_ah {
|
|
|
|
struct ib_ah ibah;
|
|
|
|
struct pvrdma_av av;
|
|
|
|
};
|
|
|
|
|
2017-11-06 19:48:53 +00:00
|
|
|
struct pvrdma_srq {
|
|
|
|
struct ib_srq ibsrq;
|
|
|
|
int offset;
|
|
|
|
spinlock_t lock; /* SRQ lock. */
|
|
|
|
int wqe_cnt;
|
|
|
|
int wqe_size;
|
|
|
|
int max_gs;
|
|
|
|
struct ib_umem *umem;
|
|
|
|
struct pvrdma_ring_state *ring;
|
|
|
|
struct pvrdma_page_dir pdir;
|
|
|
|
u32 srq_handle;
|
|
|
|
int npages;
|
|
|
|
refcount_t refcnt;
|
2017-12-20 17:51:40 +00:00
|
|
|
struct completion free;
|
2017-11-06 19:48:53 +00:00
|
|
|
};
|
|
|
|
|
IB: Add vmw_pvrdma driver
This patch series adds a driver for a paravirtual RDMA device. The
device is developed for VMware's Virtual Machines and allows existing RDMA
applications to continue to use existing Verbs API when deployed in VMs
on ESXi. We recently did a presentation in the OFA Workshop [1] regarding
this device.
Description and RDMA Support
============================
The virtual device is exposed as a dual function PCIe device. One part
is a virtual network device (VMXNet3) which provides networking properties
like MAC, IP addresses to the RDMA part of the device. The networking
properties are used to register GIDs required by RDMA applications to
communicate.
These patches add support and the all required infrastructure for
letting applications use such a device. We support the mandatory Verbs API as
well as the base memory management extensions (Local Inv, Send with Inv and
Fast Register Work Requests). We currently support both Reliable Connected
and Unreliable Datagram QPs but do not support Shared Receive Queues
(SRQs).
Also, we support the following types of Work Requests:
o Send/Receive (with or without Immediate Data)
o RDMA Write (with or without Immediate Data)
o RDMA Read
o Local Invalidate
o Send with Invalidate
o Fast Register Work Requests
This version only adds support for version 1 of RoCE. We will add RoCEv2
support in a future patch. We do support registration of both MAC-based
and IP-based GIDs. I have also created a git tree for our user-level driver
[2].
Testing
=======
We have tested this internally for various types of Guest OS - Red Hat,
Centos, Ubuntu 12.04/14.04/16.04, Oracle Enterprise Linux, SLES 12
using backported versions of this driver. The tests included several
runs of the performance tests (included with OFED), Intel MPI PingPong
benchmark on OpenMPI, krping for FRWRs. Mellanox has been kind enough
to test the backported version of the driver internally on their hardware
using a VMware provided ESX build. I have also applied and tested this
with Doug's k.o/for-4.9 branch (commit 5603910b). Note, that this patch
series should be applied all together. I split out the commits so that
it may be easier to review.
PVRDMA Resources
================
[1] OFA Workshop Presentation -
https://openfabrics.org/images/eventpresos/2016presentations/102parardma.pdf
[2] Libpvrdma User-level library -
http://git.openfabrics.org/?p=~aditr/libpvrdma.git;a=summary
Reviewed-by: Jorgen Hansen <jhansen@vmware.com>
Reviewed-by: George Zhang <georgezhang@vmware.com>
Reviewed-by: Aditya Sarwade <asarwade@vmware.com>
Reviewed-by: Bryan Tan <bryantan@vmware.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Adit Ranadive <aditr@vmware.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
2016-10-03 02:10:22 +00:00
|
|
|
struct pvrdma_qp {
|
|
|
|
struct ib_qp ibqp;
|
|
|
|
u32 qp_handle;
|
|
|
|
u32 qkey;
|
|
|
|
struct pvrdma_wq sq;
|
|
|
|
struct pvrdma_wq rq;
|
|
|
|
struct ib_umem *rumem;
|
|
|
|
struct ib_umem *sumem;
|
|
|
|
struct pvrdma_page_dir pdir;
|
2017-11-06 19:48:53 +00:00
|
|
|
struct pvrdma_srq *srq;
|
IB: Add vmw_pvrdma driver
This patch series adds a driver for a paravirtual RDMA device. The
device is developed for VMware's Virtual Machines and allows existing RDMA
applications to continue to use existing Verbs API when deployed in VMs
on ESXi. We recently did a presentation in the OFA Workshop [1] regarding
this device.
Description and RDMA Support
============================
The virtual device is exposed as a dual function PCIe device. One part
is a virtual network device (VMXNet3) which provides networking properties
like MAC, IP addresses to the RDMA part of the device. The networking
properties are used to register GIDs required by RDMA applications to
communicate.
These patches add support and the all required infrastructure for
letting applications use such a device. We support the mandatory Verbs API as
well as the base memory management extensions (Local Inv, Send with Inv and
Fast Register Work Requests). We currently support both Reliable Connected
and Unreliable Datagram QPs but do not support Shared Receive Queues
(SRQs).
Also, we support the following types of Work Requests:
o Send/Receive (with or without Immediate Data)
o RDMA Write (with or without Immediate Data)
o RDMA Read
o Local Invalidate
o Send with Invalidate
o Fast Register Work Requests
This version only adds support for version 1 of RoCE. We will add RoCEv2
support in a future patch. We do support registration of both MAC-based
and IP-based GIDs. I have also created a git tree for our user-level driver
[2].
Testing
=======
We have tested this internally for various types of Guest OS - Red Hat,
Centos, Ubuntu 12.04/14.04/16.04, Oracle Enterprise Linux, SLES 12
using backported versions of this driver. The tests included several
runs of the performance tests (included with OFED), Intel MPI PingPong
benchmark on OpenMPI, krping for FRWRs. Mellanox has been kind enough
to test the backported version of the driver internally on their hardware
using a VMware provided ESX build. I have also applied and tested this
with Doug's k.o/for-4.9 branch (commit 5603910b). Note, that this patch
series should be applied all together. I split out the commits so that
it may be easier to review.
PVRDMA Resources
================
[1] OFA Workshop Presentation -
https://openfabrics.org/images/eventpresos/2016presentations/102parardma.pdf
[2] Libpvrdma User-level library -
http://git.openfabrics.org/?p=~aditr/libpvrdma.git;a=summary
Reviewed-by: Jorgen Hansen <jhansen@vmware.com>
Reviewed-by: George Zhang <georgezhang@vmware.com>
Reviewed-by: Aditya Sarwade <asarwade@vmware.com>
Reviewed-by: Bryan Tan <bryantan@vmware.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Adit Ranadive <aditr@vmware.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
2016-10-03 02:10:22 +00:00
|
|
|
int npages;
|
|
|
|
int npages_send;
|
|
|
|
int npages_recv;
|
|
|
|
u32 flags;
|
|
|
|
u8 port;
|
|
|
|
u8 state;
|
|
|
|
bool is_kernel;
|
|
|
|
struct mutex mutex; /* QP state mutex. */
|
2017-12-20 19:27:00 +00:00
|
|
|
refcount_t refcnt;
|
2017-12-20 17:51:40 +00:00
|
|
|
struct completion free;
|
IB: Add vmw_pvrdma driver
This patch series adds a driver for a paravirtual RDMA device. The
device is developed for VMware's Virtual Machines and allows existing RDMA
applications to continue to use existing Verbs API when deployed in VMs
on ESXi. We recently did a presentation in the OFA Workshop [1] regarding
this device.
Description and RDMA Support
============================
The virtual device is exposed as a dual function PCIe device. One part
is a virtual network device (VMXNet3) which provides networking properties
like MAC, IP addresses to the RDMA part of the device. The networking
properties are used to register GIDs required by RDMA applications to
communicate.
These patches add support and the all required infrastructure for
letting applications use such a device. We support the mandatory Verbs API as
well as the base memory management extensions (Local Inv, Send with Inv and
Fast Register Work Requests). We currently support both Reliable Connected
and Unreliable Datagram QPs but do not support Shared Receive Queues
(SRQs).
Also, we support the following types of Work Requests:
o Send/Receive (with or without Immediate Data)
o RDMA Write (with or without Immediate Data)
o RDMA Read
o Local Invalidate
o Send with Invalidate
o Fast Register Work Requests
This version only adds support for version 1 of RoCE. We will add RoCEv2
support in a future patch. We do support registration of both MAC-based
and IP-based GIDs. I have also created a git tree for our user-level driver
[2].
Testing
=======
We have tested this internally for various types of Guest OS - Red Hat,
Centos, Ubuntu 12.04/14.04/16.04, Oracle Enterprise Linux, SLES 12
using backported versions of this driver. The tests included several
runs of the performance tests (included with OFED), Intel MPI PingPong
benchmark on OpenMPI, krping for FRWRs. Mellanox has been kind enough
to test the backported version of the driver internally on their hardware
using a VMware provided ESX build. I have also applied and tested this
with Doug's k.o/for-4.9 branch (commit 5603910b). Note, that this patch
series should be applied all together. I split out the commits so that
it may be easier to review.
PVRDMA Resources
================
[1] OFA Workshop Presentation -
https://openfabrics.org/images/eventpresos/2016presentations/102parardma.pdf
[2] Libpvrdma User-level library -
http://git.openfabrics.org/?p=~aditr/libpvrdma.git;a=summary
Reviewed-by: Jorgen Hansen <jhansen@vmware.com>
Reviewed-by: George Zhang <georgezhang@vmware.com>
Reviewed-by: Aditya Sarwade <asarwade@vmware.com>
Reviewed-by: Bryan Tan <bryantan@vmware.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Adit Ranadive <aditr@vmware.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
2016-10-03 02:10:22 +00:00
|
|
|
};
|
|
|
|
|
|
|
|
struct pvrdma_dev {
|
|
|
|
/* PCI device-related information. */
|
|
|
|
struct ib_device ib_dev;
|
|
|
|
struct pci_dev *pdev;
|
|
|
|
void __iomem *regs;
|
|
|
|
struct pvrdma_device_shared_region *dsr; /* Shared region pointer */
|
|
|
|
dma_addr_t dsrbase; /* Shared region base address */
|
|
|
|
void *cmd_slot;
|
|
|
|
void *resp_slot;
|
|
|
|
unsigned long flags;
|
|
|
|
struct list_head device_link;
|
2017-08-23 06:19:00 +00:00
|
|
|
unsigned int dsr_version;
|
IB: Add vmw_pvrdma driver
This patch series adds a driver for a paravirtual RDMA device. The
device is developed for VMware's Virtual Machines and allows existing RDMA
applications to continue to use existing Verbs API when deployed in VMs
on ESXi. We recently did a presentation in the OFA Workshop [1] regarding
this device.
Description and RDMA Support
============================
The virtual device is exposed as a dual function PCIe device. One part
is a virtual network device (VMXNet3) which provides networking properties
like MAC, IP addresses to the RDMA part of the device. The networking
properties are used to register GIDs required by RDMA applications to
communicate.
These patches add support and the all required infrastructure for
letting applications use such a device. We support the mandatory Verbs API as
well as the base memory management extensions (Local Inv, Send with Inv and
Fast Register Work Requests). We currently support both Reliable Connected
and Unreliable Datagram QPs but do not support Shared Receive Queues
(SRQs).
Also, we support the following types of Work Requests:
o Send/Receive (with or without Immediate Data)
o RDMA Write (with or without Immediate Data)
o RDMA Read
o Local Invalidate
o Send with Invalidate
o Fast Register Work Requests
This version only adds support for version 1 of RoCE. We will add RoCEv2
support in a future patch. We do support registration of both MAC-based
and IP-based GIDs. I have also created a git tree for our user-level driver
[2].
Testing
=======
We have tested this internally for various types of Guest OS - Red Hat,
Centos, Ubuntu 12.04/14.04/16.04, Oracle Enterprise Linux, SLES 12
using backported versions of this driver. The tests included several
runs of the performance tests (included with OFED), Intel MPI PingPong
benchmark on OpenMPI, krping for FRWRs. Mellanox has been kind enough
to test the backported version of the driver internally on their hardware
using a VMware provided ESX build. I have also applied and tested this
with Doug's k.o/for-4.9 branch (commit 5603910b). Note, that this patch
series should be applied all together. I split out the commits so that
it may be easier to review.
PVRDMA Resources
================
[1] OFA Workshop Presentation -
https://openfabrics.org/images/eventpresos/2016presentations/102parardma.pdf
[2] Libpvrdma User-level library -
http://git.openfabrics.org/?p=~aditr/libpvrdma.git;a=summary
Reviewed-by: Jorgen Hansen <jhansen@vmware.com>
Reviewed-by: George Zhang <georgezhang@vmware.com>
Reviewed-by: Aditya Sarwade <asarwade@vmware.com>
Reviewed-by: Bryan Tan <bryantan@vmware.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Adit Ranadive <aditr@vmware.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
2016-10-03 02:10:22 +00:00
|
|
|
|
|
|
|
/* Locking and interrupt information. */
|
|
|
|
spinlock_t cmd_lock; /* Command lock. */
|
|
|
|
struct semaphore cmd_sema;
|
|
|
|
struct completion cmd_done;
|
2017-02-15 07:47:04 +00:00
|
|
|
unsigned int nr_vectors;
|
IB: Add vmw_pvrdma driver
This patch series adds a driver for a paravirtual RDMA device. The
device is developed for VMware's Virtual Machines and allows existing RDMA
applications to continue to use existing Verbs API when deployed in VMs
on ESXi. We recently did a presentation in the OFA Workshop [1] regarding
this device.
Description and RDMA Support
============================
The virtual device is exposed as a dual function PCIe device. One part
is a virtual network device (VMXNet3) which provides networking properties
like MAC, IP addresses to the RDMA part of the device. The networking
properties are used to register GIDs required by RDMA applications to
communicate.
These patches add support and the all required infrastructure for
letting applications use such a device. We support the mandatory Verbs API as
well as the base memory management extensions (Local Inv, Send with Inv and
Fast Register Work Requests). We currently support both Reliable Connected
and Unreliable Datagram QPs but do not support Shared Receive Queues
(SRQs).
Also, we support the following types of Work Requests:
o Send/Receive (with or without Immediate Data)
o RDMA Write (with or without Immediate Data)
o RDMA Read
o Local Invalidate
o Send with Invalidate
o Fast Register Work Requests
This version only adds support for version 1 of RoCE. We will add RoCEv2
support in a future patch. We do support registration of both MAC-based
and IP-based GIDs. I have also created a git tree for our user-level driver
[2].
Testing
=======
We have tested this internally for various types of Guest OS - Red Hat,
Centos, Ubuntu 12.04/14.04/16.04, Oracle Enterprise Linux, SLES 12
using backported versions of this driver. The tests included several
runs of the performance tests (included with OFED), Intel MPI PingPong
benchmark on OpenMPI, krping for FRWRs. Mellanox has been kind enough
to test the backported version of the driver internally on their hardware
using a VMware provided ESX build. I have also applied and tested this
with Doug's k.o/for-4.9 branch (commit 5603910b). Note, that this patch
series should be applied all together. I split out the commits so that
it may be easier to review.
PVRDMA Resources
================
[1] OFA Workshop Presentation -
https://openfabrics.org/images/eventpresos/2016presentations/102parardma.pdf
[2] Libpvrdma User-level library -
http://git.openfabrics.org/?p=~aditr/libpvrdma.git;a=summary
Reviewed-by: Jorgen Hansen <jhansen@vmware.com>
Reviewed-by: George Zhang <georgezhang@vmware.com>
Reviewed-by: Aditya Sarwade <asarwade@vmware.com>
Reviewed-by: Bryan Tan <bryantan@vmware.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Adit Ranadive <aditr@vmware.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
2016-10-03 02:10:22 +00:00
|
|
|
|
|
|
|
/* RDMA-related device information. */
|
|
|
|
union ib_gid *sgid_tbl;
|
|
|
|
struct pvrdma_ring_state *async_ring_state;
|
|
|
|
struct pvrdma_page_dir async_pdir;
|
|
|
|
struct pvrdma_ring_state *cq_ring_state;
|
|
|
|
struct pvrdma_page_dir cq_pdir;
|
|
|
|
struct pvrdma_cq **cq_tbl;
|
|
|
|
spinlock_t cq_tbl_lock;
|
2017-11-06 19:48:53 +00:00
|
|
|
struct pvrdma_srq **srq_tbl;
|
|
|
|
spinlock_t srq_tbl_lock;
|
IB: Add vmw_pvrdma driver
This patch series adds a driver for a paravirtual RDMA device. The
device is developed for VMware's Virtual Machines and allows existing RDMA
applications to continue to use existing Verbs API when deployed in VMs
on ESXi. We recently did a presentation in the OFA Workshop [1] regarding
this device.
Description and RDMA Support
============================
The virtual device is exposed as a dual function PCIe device. One part
is a virtual network device (VMXNet3) which provides networking properties
like MAC, IP addresses to the RDMA part of the device. The networking
properties are used to register GIDs required by RDMA applications to
communicate.
These patches add support and the all required infrastructure for
letting applications use such a device. We support the mandatory Verbs API as
well as the base memory management extensions (Local Inv, Send with Inv and
Fast Register Work Requests). We currently support both Reliable Connected
and Unreliable Datagram QPs but do not support Shared Receive Queues
(SRQs).
Also, we support the following types of Work Requests:
o Send/Receive (with or without Immediate Data)
o RDMA Write (with or without Immediate Data)
o RDMA Read
o Local Invalidate
o Send with Invalidate
o Fast Register Work Requests
This version only adds support for version 1 of RoCE. We will add RoCEv2
support in a future patch. We do support registration of both MAC-based
and IP-based GIDs. I have also created a git tree for our user-level driver
[2].
Testing
=======
We have tested this internally for various types of Guest OS - Red Hat,
Centos, Ubuntu 12.04/14.04/16.04, Oracle Enterprise Linux, SLES 12
using backported versions of this driver. The tests included several
runs of the performance tests (included with OFED), Intel MPI PingPong
benchmark on OpenMPI, krping for FRWRs. Mellanox has been kind enough
to test the backported version of the driver internally on their hardware
using a VMware provided ESX build. I have also applied and tested this
with Doug's k.o/for-4.9 branch (commit 5603910b). Note, that this patch
series should be applied all together. I split out the commits so that
it may be easier to review.
PVRDMA Resources
================
[1] OFA Workshop Presentation -
https://openfabrics.org/images/eventpresos/2016presentations/102parardma.pdf
[2] Libpvrdma User-level library -
http://git.openfabrics.org/?p=~aditr/libpvrdma.git;a=summary
Reviewed-by: Jorgen Hansen <jhansen@vmware.com>
Reviewed-by: George Zhang <georgezhang@vmware.com>
Reviewed-by: Aditya Sarwade <asarwade@vmware.com>
Reviewed-by: Bryan Tan <bryantan@vmware.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Adit Ranadive <aditr@vmware.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
2016-10-03 02:10:22 +00:00
|
|
|
struct pvrdma_qp **qp_tbl;
|
|
|
|
spinlock_t qp_tbl_lock;
|
|
|
|
struct pvrdma_uar_table uar_table;
|
|
|
|
struct pvrdma_uar_map driver_uar;
|
|
|
|
__be64 sys_image_guid;
|
|
|
|
spinlock_t desc_lock; /* Device modification lock. */
|
|
|
|
u32 port_cap_mask;
|
|
|
|
struct mutex port_mutex; /* Port modification mutex. */
|
|
|
|
bool ib_active;
|
|
|
|
atomic_t num_qps;
|
|
|
|
atomic_t num_cqs;
|
2017-11-06 19:48:53 +00:00
|
|
|
atomic_t num_srqs;
|
IB: Add vmw_pvrdma driver
This patch series adds a driver for a paravirtual RDMA device. The
device is developed for VMware's Virtual Machines and allows existing RDMA
applications to continue to use existing Verbs API when deployed in VMs
on ESXi. We recently did a presentation in the OFA Workshop [1] regarding
this device.
Description and RDMA Support
============================
The virtual device is exposed as a dual function PCIe device. One part
is a virtual network device (VMXNet3) which provides networking properties
like MAC, IP addresses to the RDMA part of the device. The networking
properties are used to register GIDs required by RDMA applications to
communicate.
These patches add support and the all required infrastructure for
letting applications use such a device. We support the mandatory Verbs API as
well as the base memory management extensions (Local Inv, Send with Inv and
Fast Register Work Requests). We currently support both Reliable Connected
and Unreliable Datagram QPs but do not support Shared Receive Queues
(SRQs).
Also, we support the following types of Work Requests:
o Send/Receive (with or without Immediate Data)
o RDMA Write (with or without Immediate Data)
o RDMA Read
o Local Invalidate
o Send with Invalidate
o Fast Register Work Requests
This version only adds support for version 1 of RoCE. We will add RoCEv2
support in a future patch. We do support registration of both MAC-based
and IP-based GIDs. I have also created a git tree for our user-level driver
[2].
Testing
=======
We have tested this internally for various types of Guest OS - Red Hat,
Centos, Ubuntu 12.04/14.04/16.04, Oracle Enterprise Linux, SLES 12
using backported versions of this driver. The tests included several
runs of the performance tests (included with OFED), Intel MPI PingPong
benchmark on OpenMPI, krping for FRWRs. Mellanox has been kind enough
to test the backported version of the driver internally on their hardware
using a VMware provided ESX build. I have also applied and tested this
with Doug's k.o/for-4.9 branch (commit 5603910b). Note, that this patch
series should be applied all together. I split out the commits so that
it may be easier to review.
PVRDMA Resources
================
[1] OFA Workshop Presentation -
https://openfabrics.org/images/eventpresos/2016presentations/102parardma.pdf
[2] Libpvrdma User-level library -
http://git.openfabrics.org/?p=~aditr/libpvrdma.git;a=summary
Reviewed-by: Jorgen Hansen <jhansen@vmware.com>
Reviewed-by: George Zhang <georgezhang@vmware.com>
Reviewed-by: Aditya Sarwade <asarwade@vmware.com>
Reviewed-by: Bryan Tan <bryantan@vmware.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Adit Ranadive <aditr@vmware.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
2016-10-03 02:10:22 +00:00
|
|
|
atomic_t num_pds;
|
|
|
|
atomic_t num_ahs;
|
|
|
|
|
|
|
|
/* Network device information. */
|
|
|
|
struct net_device *netdev;
|
|
|
|
struct notifier_block nb_netdev;
|
|
|
|
};
|
|
|
|
|
|
|
|
struct pvrdma_netdevice_work {
|
|
|
|
struct work_struct work;
|
|
|
|
struct net_device *event_netdev;
|
|
|
|
unsigned long event;
|
|
|
|
};
|
|
|
|
|
|
|
|
static inline struct pvrdma_dev *to_vdev(struct ib_device *ibdev)
|
|
|
|
{
|
|
|
|
return container_of(ibdev, struct pvrdma_dev, ib_dev);
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline struct
|
|
|
|
pvrdma_ucontext *to_vucontext(struct ib_ucontext *ibucontext)
|
|
|
|
{
|
|
|
|
return container_of(ibucontext, struct pvrdma_ucontext, ibucontext);
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline struct pvrdma_pd *to_vpd(struct ib_pd *ibpd)
|
|
|
|
{
|
|
|
|
return container_of(ibpd, struct pvrdma_pd, ibpd);
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline struct pvrdma_cq *to_vcq(struct ib_cq *ibcq)
|
|
|
|
{
|
|
|
|
return container_of(ibcq, struct pvrdma_cq, ibcq);
|
|
|
|
}
|
|
|
|
|
2017-11-06 19:48:53 +00:00
|
|
|
static inline struct pvrdma_srq *to_vsrq(struct ib_srq *ibsrq)
|
|
|
|
{
|
|
|
|
return container_of(ibsrq, struct pvrdma_srq, ibsrq);
|
|
|
|
}
|
|
|
|
|
IB: Add vmw_pvrdma driver
This patch series adds a driver for a paravirtual RDMA device. The
device is developed for VMware's Virtual Machines and allows existing RDMA
applications to continue to use existing Verbs API when deployed in VMs
on ESXi. We recently did a presentation in the OFA Workshop [1] regarding
this device.
Description and RDMA Support
============================
The virtual device is exposed as a dual function PCIe device. One part
is a virtual network device (VMXNet3) which provides networking properties
like MAC, IP addresses to the RDMA part of the device. The networking
properties are used to register GIDs required by RDMA applications to
communicate.
These patches add support and the all required infrastructure for
letting applications use such a device. We support the mandatory Verbs API as
well as the base memory management extensions (Local Inv, Send with Inv and
Fast Register Work Requests). We currently support both Reliable Connected
and Unreliable Datagram QPs but do not support Shared Receive Queues
(SRQs).
Also, we support the following types of Work Requests:
o Send/Receive (with or without Immediate Data)
o RDMA Write (with or without Immediate Data)
o RDMA Read
o Local Invalidate
o Send with Invalidate
o Fast Register Work Requests
This version only adds support for version 1 of RoCE. We will add RoCEv2
support in a future patch. We do support registration of both MAC-based
and IP-based GIDs. I have also created a git tree for our user-level driver
[2].
Testing
=======
We have tested this internally for various types of Guest OS - Red Hat,
Centos, Ubuntu 12.04/14.04/16.04, Oracle Enterprise Linux, SLES 12
using backported versions of this driver. The tests included several
runs of the performance tests (included with OFED), Intel MPI PingPong
benchmark on OpenMPI, krping for FRWRs. Mellanox has been kind enough
to test the backported version of the driver internally on their hardware
using a VMware provided ESX build. I have also applied and tested this
with Doug's k.o/for-4.9 branch (commit 5603910b). Note, that this patch
series should be applied all together. I split out the commits so that
it may be easier to review.
PVRDMA Resources
================
[1] OFA Workshop Presentation -
https://openfabrics.org/images/eventpresos/2016presentations/102parardma.pdf
[2] Libpvrdma User-level library -
http://git.openfabrics.org/?p=~aditr/libpvrdma.git;a=summary
Reviewed-by: Jorgen Hansen <jhansen@vmware.com>
Reviewed-by: George Zhang <georgezhang@vmware.com>
Reviewed-by: Aditya Sarwade <asarwade@vmware.com>
Reviewed-by: Bryan Tan <bryantan@vmware.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Adit Ranadive <aditr@vmware.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
2016-10-03 02:10:22 +00:00
|
|
|
static inline struct pvrdma_user_mr *to_vmr(struct ib_mr *ibmr)
|
|
|
|
{
|
|
|
|
return container_of(ibmr, struct pvrdma_user_mr, ibmr);
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline struct pvrdma_qp *to_vqp(struct ib_qp *ibqp)
|
|
|
|
{
|
|
|
|
return container_of(ibqp, struct pvrdma_qp, ibqp);
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline struct pvrdma_ah *to_vah(struct ib_ah *ibah)
|
|
|
|
{
|
|
|
|
return container_of(ibah, struct pvrdma_ah, ibah);
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline void pvrdma_write_reg(struct pvrdma_dev *dev, u32 reg, u32 val)
|
|
|
|
{
|
|
|
|
writel(cpu_to_le32(val), dev->regs + reg);
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline u32 pvrdma_read_reg(struct pvrdma_dev *dev, u32 reg)
|
|
|
|
{
|
|
|
|
return le32_to_cpu(readl(dev->regs + reg));
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline void pvrdma_write_uar_cq(struct pvrdma_dev *dev, u32 val)
|
|
|
|
{
|
|
|
|
writel(cpu_to_le32(val), dev->driver_uar.map + PVRDMA_UAR_CQ_OFFSET);
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline void pvrdma_write_uar_qp(struct pvrdma_dev *dev, u32 val)
|
|
|
|
{
|
|
|
|
writel(cpu_to_le32(val), dev->driver_uar.map + PVRDMA_UAR_QP_OFFSET);
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline void *pvrdma_page_dir_get_ptr(struct pvrdma_page_dir *pdir,
|
|
|
|
u64 offset)
|
|
|
|
{
|
|
|
|
return pdir->pages[offset / PAGE_SIZE] + (offset % PAGE_SIZE);
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline enum pvrdma_mtu ib_mtu_to_pvrdma(enum ib_mtu mtu)
|
|
|
|
{
|
|
|
|
return (enum pvrdma_mtu)mtu;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline enum ib_mtu pvrdma_mtu_to_ib(enum pvrdma_mtu mtu)
|
|
|
|
{
|
|
|
|
return (enum ib_mtu)mtu;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline enum pvrdma_port_state ib_port_state_to_pvrdma(
|
|
|
|
enum ib_port_state state)
|
|
|
|
{
|
|
|
|
return (enum pvrdma_port_state)state;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline enum ib_port_state pvrdma_port_state_to_ib(
|
|
|
|
enum pvrdma_port_state state)
|
|
|
|
{
|
|
|
|
return (enum ib_port_state)state;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline int ib_port_cap_flags_to_pvrdma(int flags)
|
|
|
|
{
|
|
|
|
return flags & PVRDMA_MASK(PVRDMA_PORT_CAP_FLAGS_MAX);
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline int pvrdma_port_cap_flags_to_ib(int flags)
|
|
|
|
{
|
|
|
|
return flags;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline enum pvrdma_port_width ib_port_width_to_pvrdma(
|
|
|
|
enum ib_port_width width)
|
|
|
|
{
|
|
|
|
return (enum pvrdma_port_width)width;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline enum ib_port_width pvrdma_port_width_to_ib(
|
|
|
|
enum pvrdma_port_width width)
|
|
|
|
{
|
|
|
|
return (enum ib_port_width)width;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline enum pvrdma_port_speed ib_port_speed_to_pvrdma(
|
|
|
|
enum ib_port_speed speed)
|
|
|
|
{
|
|
|
|
return (enum pvrdma_port_speed)speed;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline enum ib_port_speed pvrdma_port_speed_to_ib(
|
|
|
|
enum pvrdma_port_speed speed)
|
|
|
|
{
|
|
|
|
return (enum ib_port_speed)speed;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline int ib_qp_attr_mask_to_pvrdma(int attr_mask)
|
|
|
|
{
|
|
|
|
return attr_mask & PVRDMA_MASK(PVRDMA_QP_ATTR_MASK_MAX);
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline enum pvrdma_mig_state ib_mig_state_to_pvrdma(
|
|
|
|
enum ib_mig_state state)
|
|
|
|
{
|
|
|
|
return (enum pvrdma_mig_state)state;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline enum ib_mig_state pvrdma_mig_state_to_ib(
|
|
|
|
enum pvrdma_mig_state state)
|
|
|
|
{
|
|
|
|
return (enum ib_mig_state)state;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline int ib_access_flags_to_pvrdma(int flags)
|
|
|
|
{
|
|
|
|
return flags;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline int pvrdma_access_flags_to_ib(int flags)
|
|
|
|
{
|
|
|
|
return flags & PVRDMA_MASK(PVRDMA_ACCESS_FLAGS_MAX);
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline enum pvrdma_qp_type ib_qp_type_to_pvrdma(enum ib_qp_type type)
|
|
|
|
{
|
|
|
|
return (enum pvrdma_qp_type)type;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline enum ib_qp_type pvrdma_qp_type_to_ib(enum pvrdma_qp_type type)
|
|
|
|
{
|
|
|
|
return (enum ib_qp_type)type;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline enum pvrdma_qp_state ib_qp_state_to_pvrdma(enum ib_qp_state state)
|
|
|
|
{
|
|
|
|
return (enum pvrdma_qp_state)state;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline enum ib_qp_state pvrdma_qp_state_to_ib(enum pvrdma_qp_state state)
|
|
|
|
{
|
|
|
|
return (enum ib_qp_state)state;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline enum pvrdma_wr_opcode ib_wr_opcode_to_pvrdma(enum ib_wr_opcode op)
|
|
|
|
{
|
2019-01-09 23:08:49 +00:00
|
|
|
switch (op) {
|
|
|
|
case IB_WR_RDMA_WRITE:
|
|
|
|
return PVRDMA_WR_RDMA_WRITE;
|
|
|
|
case IB_WR_RDMA_WRITE_WITH_IMM:
|
|
|
|
return PVRDMA_WR_RDMA_WRITE_WITH_IMM;
|
|
|
|
case IB_WR_SEND:
|
|
|
|
return PVRDMA_WR_SEND;
|
|
|
|
case IB_WR_SEND_WITH_IMM:
|
|
|
|
return PVRDMA_WR_SEND_WITH_IMM;
|
|
|
|
case IB_WR_RDMA_READ:
|
|
|
|
return PVRDMA_WR_RDMA_READ;
|
|
|
|
case IB_WR_ATOMIC_CMP_AND_SWP:
|
|
|
|
return PVRDMA_WR_ATOMIC_CMP_AND_SWP;
|
|
|
|
case IB_WR_ATOMIC_FETCH_AND_ADD:
|
|
|
|
return PVRDMA_WR_ATOMIC_FETCH_AND_ADD;
|
|
|
|
case IB_WR_LSO:
|
|
|
|
return PVRDMA_WR_LSO;
|
|
|
|
case IB_WR_SEND_WITH_INV:
|
|
|
|
return PVRDMA_WR_SEND_WITH_INV;
|
|
|
|
case IB_WR_RDMA_READ_WITH_INV:
|
|
|
|
return PVRDMA_WR_RDMA_READ_WITH_INV;
|
|
|
|
case IB_WR_LOCAL_INV:
|
|
|
|
return PVRDMA_WR_LOCAL_INV;
|
|
|
|
case IB_WR_REG_MR:
|
|
|
|
return PVRDMA_WR_FAST_REG_MR;
|
|
|
|
case IB_WR_MASKED_ATOMIC_CMP_AND_SWP:
|
|
|
|
return PVRDMA_WR_MASKED_ATOMIC_CMP_AND_SWP;
|
|
|
|
case IB_WR_MASKED_ATOMIC_FETCH_AND_ADD:
|
|
|
|
return PVRDMA_WR_MASKED_ATOMIC_FETCH_AND_ADD;
|
2019-06-11 15:52:54 +00:00
|
|
|
case IB_WR_REG_MR_INTEGRITY:
|
2019-01-09 23:08:49 +00:00
|
|
|
return PVRDMA_WR_REG_SIG_MR;
|
|
|
|
default:
|
|
|
|
return PVRDMA_WR_ERROR;
|
|
|
|
}
|
IB: Add vmw_pvrdma driver
This patch series adds a driver for a paravirtual RDMA device. The
device is developed for VMware's Virtual Machines and allows existing RDMA
applications to continue to use existing Verbs API when deployed in VMs
on ESXi. We recently did a presentation in the OFA Workshop [1] regarding
this device.
Description and RDMA Support
============================
The virtual device is exposed as a dual function PCIe device. One part
is a virtual network device (VMXNet3) which provides networking properties
like MAC, IP addresses to the RDMA part of the device. The networking
properties are used to register GIDs required by RDMA applications to
communicate.
These patches add support and the all required infrastructure for
letting applications use such a device. We support the mandatory Verbs API as
well as the base memory management extensions (Local Inv, Send with Inv and
Fast Register Work Requests). We currently support both Reliable Connected
and Unreliable Datagram QPs but do not support Shared Receive Queues
(SRQs).
Also, we support the following types of Work Requests:
o Send/Receive (with or without Immediate Data)
o RDMA Write (with or without Immediate Data)
o RDMA Read
o Local Invalidate
o Send with Invalidate
o Fast Register Work Requests
This version only adds support for version 1 of RoCE. We will add RoCEv2
support in a future patch. We do support registration of both MAC-based
and IP-based GIDs. I have also created a git tree for our user-level driver
[2].
Testing
=======
We have tested this internally for various types of Guest OS - Red Hat,
Centos, Ubuntu 12.04/14.04/16.04, Oracle Enterprise Linux, SLES 12
using backported versions of this driver. The tests included several
runs of the performance tests (included with OFED), Intel MPI PingPong
benchmark on OpenMPI, krping for FRWRs. Mellanox has been kind enough
to test the backported version of the driver internally on their hardware
using a VMware provided ESX build. I have also applied and tested this
with Doug's k.o/for-4.9 branch (commit 5603910b). Note, that this patch
series should be applied all together. I split out the commits so that
it may be easier to review.
PVRDMA Resources
================
[1] OFA Workshop Presentation -
https://openfabrics.org/images/eventpresos/2016presentations/102parardma.pdf
[2] Libpvrdma User-level library -
http://git.openfabrics.org/?p=~aditr/libpvrdma.git;a=summary
Reviewed-by: Jorgen Hansen <jhansen@vmware.com>
Reviewed-by: George Zhang <georgezhang@vmware.com>
Reviewed-by: Aditya Sarwade <asarwade@vmware.com>
Reviewed-by: Bryan Tan <bryantan@vmware.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Adit Ranadive <aditr@vmware.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
2016-10-03 02:10:22 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static inline enum ib_wc_status pvrdma_wc_status_to_ib(
|
|
|
|
enum pvrdma_wc_status status)
|
|
|
|
{
|
|
|
|
return (enum ib_wc_status)status;
|
|
|
|
}
|
|
|
|
|
2017-09-21 22:56:21 +00:00
|
|
|
static inline int pvrdma_wc_opcode_to_ib(unsigned int opcode)
|
|
|
|
{
|
|
|
|
switch (opcode) {
|
|
|
|
case PVRDMA_WC_SEND:
|
|
|
|
return IB_WC_SEND;
|
|
|
|
case PVRDMA_WC_RDMA_WRITE:
|
|
|
|
return IB_WC_RDMA_WRITE;
|
|
|
|
case PVRDMA_WC_RDMA_READ:
|
|
|
|
return IB_WC_RDMA_READ;
|
|
|
|
case PVRDMA_WC_COMP_SWAP:
|
|
|
|
return IB_WC_COMP_SWAP;
|
|
|
|
case PVRDMA_WC_FETCH_ADD:
|
|
|
|
return IB_WC_FETCH_ADD;
|
|
|
|
case PVRDMA_WC_LOCAL_INV:
|
|
|
|
return IB_WC_LOCAL_INV;
|
|
|
|
case PVRDMA_WC_FAST_REG_MR:
|
|
|
|
return IB_WC_REG_MR;
|
|
|
|
case PVRDMA_WC_MASKED_COMP_SWAP:
|
|
|
|
return IB_WC_MASKED_COMP_SWAP;
|
|
|
|
case PVRDMA_WC_MASKED_FETCH_ADD:
|
|
|
|
return IB_WC_MASKED_FETCH_ADD;
|
|
|
|
case PVRDMA_WC_RECV:
|
|
|
|
return IB_WC_RECV;
|
|
|
|
case PVRDMA_WC_RECV_RDMA_WITH_IMM:
|
|
|
|
return IB_WC_RECV_RDMA_WITH_IMM;
|
|
|
|
default:
|
|
|
|
return IB_WC_SEND;
|
|
|
|
}
|
IB: Add vmw_pvrdma driver
This patch series adds a driver for a paravirtual RDMA device. The
device is developed for VMware's Virtual Machines and allows existing RDMA
applications to continue to use existing Verbs API when deployed in VMs
on ESXi. We recently did a presentation in the OFA Workshop [1] regarding
this device.
Description and RDMA Support
============================
The virtual device is exposed as a dual function PCIe device. One part
is a virtual network device (VMXNet3) which provides networking properties
like MAC, IP addresses to the RDMA part of the device. The networking
properties are used to register GIDs required by RDMA applications to
communicate.
These patches add support and the all required infrastructure for
letting applications use such a device. We support the mandatory Verbs API as
well as the base memory management extensions (Local Inv, Send with Inv and
Fast Register Work Requests). We currently support both Reliable Connected
and Unreliable Datagram QPs but do not support Shared Receive Queues
(SRQs).
Also, we support the following types of Work Requests:
o Send/Receive (with or without Immediate Data)
o RDMA Write (with or without Immediate Data)
o RDMA Read
o Local Invalidate
o Send with Invalidate
o Fast Register Work Requests
This version only adds support for version 1 of RoCE. We will add RoCEv2
support in a future patch. We do support registration of both MAC-based
and IP-based GIDs. I have also created a git tree for our user-level driver
[2].
Testing
=======
We have tested this internally for various types of Guest OS - Red Hat,
Centos, Ubuntu 12.04/14.04/16.04, Oracle Enterprise Linux, SLES 12
using backported versions of this driver. The tests included several
runs of the performance tests (included with OFED), Intel MPI PingPong
benchmark on OpenMPI, krping for FRWRs. Mellanox has been kind enough
to test the backported version of the driver internally on their hardware
using a VMware provided ESX build. I have also applied and tested this
with Doug's k.o/for-4.9 branch (commit 5603910b). Note, that this patch
series should be applied all together. I split out the commits so that
it may be easier to review.
PVRDMA Resources
================
[1] OFA Workshop Presentation -
https://openfabrics.org/images/eventpresos/2016presentations/102parardma.pdf
[2] Libpvrdma User-level library -
http://git.openfabrics.org/?p=~aditr/libpvrdma.git;a=summary
Reviewed-by: Jorgen Hansen <jhansen@vmware.com>
Reviewed-by: George Zhang <georgezhang@vmware.com>
Reviewed-by: Aditya Sarwade <asarwade@vmware.com>
Reviewed-by: Bryan Tan <bryantan@vmware.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Adit Ranadive <aditr@vmware.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
2016-10-03 02:10:22 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static inline int pvrdma_wc_flags_to_ib(int flags)
|
|
|
|
{
|
|
|
|
return flags;
|
|
|
|
}
|
|
|
|
|
|
|
|
static inline int ib_send_flags_to_pvrdma(int flags)
|
|
|
|
{
|
|
|
|
return flags & PVRDMA_MASK(PVRDMA_SEND_FLAGS_MAX);
|
|
|
|
}
|
|
|
|
|
|
|
|
void pvrdma_qp_cap_to_ib(struct ib_qp_cap *dst,
|
|
|
|
const struct pvrdma_qp_cap *src);
|
|
|
|
void ib_qp_cap_to_pvrdma(struct pvrdma_qp_cap *dst,
|
|
|
|
const struct ib_qp_cap *src);
|
|
|
|
void pvrdma_gid_to_ib(union ib_gid *dst, const union pvrdma_gid *src);
|
|
|
|
void ib_gid_to_pvrdma(union pvrdma_gid *dst, const union ib_gid *src);
|
|
|
|
void pvrdma_global_route_to_ib(struct ib_global_route *dst,
|
|
|
|
const struct pvrdma_global_route *src);
|
|
|
|
void ib_global_route_to_pvrdma(struct pvrdma_global_route *dst,
|
|
|
|
const struct ib_global_route *src);
|
2017-04-29 18:41:26 +00:00
|
|
|
void pvrdma_ah_attr_to_rdma(struct rdma_ah_attr *dst,
|
|
|
|
const struct pvrdma_ah_attr *src);
|
|
|
|
void rdma_ah_attr_to_pvrdma(struct pvrdma_ah_attr *dst,
|
|
|
|
const struct rdma_ah_attr *src);
|
2017-08-23 06:19:00 +00:00
|
|
|
u8 ib_gid_type_to_pvrdma(enum ib_gid_type gid_type);
|
IB: Add vmw_pvrdma driver
This patch series adds a driver for a paravirtual RDMA device. The
device is developed for VMware's Virtual Machines and allows existing RDMA
applications to continue to use existing Verbs API when deployed in VMs
on ESXi. We recently did a presentation in the OFA Workshop [1] regarding
this device.
Description and RDMA Support
============================
The virtual device is exposed as a dual function PCIe device. One part
is a virtual network device (VMXNet3) which provides networking properties
like MAC, IP addresses to the RDMA part of the device. The networking
properties are used to register GIDs required by RDMA applications to
communicate.
These patches add support and the all required infrastructure for
letting applications use such a device. We support the mandatory Verbs API as
well as the base memory management extensions (Local Inv, Send with Inv and
Fast Register Work Requests). We currently support both Reliable Connected
and Unreliable Datagram QPs but do not support Shared Receive Queues
(SRQs).
Also, we support the following types of Work Requests:
o Send/Receive (with or without Immediate Data)
o RDMA Write (with or without Immediate Data)
o RDMA Read
o Local Invalidate
o Send with Invalidate
o Fast Register Work Requests
This version only adds support for version 1 of RoCE. We will add RoCEv2
support in a future patch. We do support registration of both MAC-based
and IP-based GIDs. I have also created a git tree for our user-level driver
[2].
Testing
=======
We have tested this internally for various types of Guest OS - Red Hat,
Centos, Ubuntu 12.04/14.04/16.04, Oracle Enterprise Linux, SLES 12
using backported versions of this driver. The tests included several
runs of the performance tests (included with OFED), Intel MPI PingPong
benchmark on OpenMPI, krping for FRWRs. Mellanox has been kind enough
to test the backported version of the driver internally on their hardware
using a VMware provided ESX build. I have also applied and tested this
with Doug's k.o/for-4.9 branch (commit 5603910b). Note, that this patch
series should be applied all together. I split out the commits so that
it may be easier to review.
PVRDMA Resources
================
[1] OFA Workshop Presentation -
https://openfabrics.org/images/eventpresos/2016presentations/102parardma.pdf
[2] Libpvrdma User-level library -
http://git.openfabrics.org/?p=~aditr/libpvrdma.git;a=summary
Reviewed-by: Jorgen Hansen <jhansen@vmware.com>
Reviewed-by: George Zhang <georgezhang@vmware.com>
Reviewed-by: Aditya Sarwade <asarwade@vmware.com>
Reviewed-by: Bryan Tan <bryantan@vmware.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: Adit Ranadive <aditr@vmware.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
2016-10-03 02:10:22 +00:00
|
|
|
|
|
|
|
int pvrdma_uar_table_init(struct pvrdma_dev *dev);
|
|
|
|
void pvrdma_uar_table_cleanup(struct pvrdma_dev *dev);
|
|
|
|
|
|
|
|
int pvrdma_uar_alloc(struct pvrdma_dev *dev, struct pvrdma_uar_map *uar);
|
|
|
|
void pvrdma_uar_free(struct pvrdma_dev *dev, struct pvrdma_uar_map *uar);
|
|
|
|
|
|
|
|
void _pvrdma_flush_cqe(struct pvrdma_qp *qp, struct pvrdma_cq *cq);
|
|
|
|
|
|
|
|
int pvrdma_page_dir_init(struct pvrdma_dev *dev, struct pvrdma_page_dir *pdir,
|
|
|
|
u64 npages, bool alloc_pages);
|
|
|
|
void pvrdma_page_dir_cleanup(struct pvrdma_dev *dev,
|
|
|
|
struct pvrdma_page_dir *pdir);
|
|
|
|
int pvrdma_page_dir_insert_dma(struct pvrdma_page_dir *pdir, u64 idx,
|
|
|
|
dma_addr_t daddr);
|
|
|
|
int pvrdma_page_dir_insert_umem(struct pvrdma_page_dir *pdir,
|
|
|
|
struct ib_umem *umem, u64 offset);
|
|
|
|
dma_addr_t pvrdma_page_dir_get_dma(struct pvrdma_page_dir *pdir, u64 idx);
|
|
|
|
int pvrdma_page_dir_insert_page_list(struct pvrdma_page_dir *pdir,
|
|
|
|
u64 *page_list, int num_pages);
|
|
|
|
|
|
|
|
int pvrdma_cmd_post(struct pvrdma_dev *dev, union pvrdma_cmd_req *req,
|
|
|
|
union pvrdma_cmd_resp *rsp, unsigned resp_code);
|
|
|
|
|
|
|
|
#endif /* __PVRDMA_H__ */
|