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.
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include <asm/page.h>
|
|
|
|
#include <linux/io.h>
|
|
|
|
#include <linux/wait.h>
|
|
|
|
#include <rdma/ib_addr.h>
|
|
|
|
#include <rdma/ib_smi.h>
|
|
|
|
#include <rdma/ib_user_verbs.h>
|
|
|
|
|
|
|
|
#include "pvrdma.h"
|
|
|
|
|
|
|
|
static inline void get_cqs(struct pvrdma_qp *qp, struct pvrdma_cq **send_cq,
|
|
|
|
struct pvrdma_cq **recv_cq)
|
|
|
|
{
|
|
|
|
*send_cq = to_vcq(qp->ibqp.send_cq);
|
|
|
|
*recv_cq = to_vcq(qp->ibqp.recv_cq);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void pvrdma_lock_cqs(struct pvrdma_cq *scq, struct pvrdma_cq *rcq,
|
|
|
|
unsigned long *scq_flags,
|
|
|
|
unsigned long *rcq_flags)
|
|
|
|
__acquires(scq->cq_lock) __acquires(rcq->cq_lock)
|
|
|
|
{
|
|
|
|
if (scq == rcq) {
|
|
|
|
spin_lock_irqsave(&scq->cq_lock, *scq_flags);
|
|
|
|
__acquire(rcq->cq_lock);
|
|
|
|
} else if (scq->cq_handle < rcq->cq_handle) {
|
|
|
|
spin_lock_irqsave(&scq->cq_lock, *scq_flags);
|
|
|
|
spin_lock_irqsave_nested(&rcq->cq_lock, *rcq_flags,
|
|
|
|
SINGLE_DEPTH_NESTING);
|
|
|
|
} else {
|
|
|
|
spin_lock_irqsave(&rcq->cq_lock, *rcq_flags);
|
|
|
|
spin_lock_irqsave_nested(&scq->cq_lock, *scq_flags,
|
|
|
|
SINGLE_DEPTH_NESTING);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void pvrdma_unlock_cqs(struct pvrdma_cq *scq, struct pvrdma_cq *rcq,
|
|
|
|
unsigned long *scq_flags,
|
|
|
|
unsigned long *rcq_flags)
|
|
|
|
__releases(scq->cq_lock) __releases(rcq->cq_lock)
|
|
|
|
{
|
|
|
|
if (scq == rcq) {
|
|
|
|
__release(rcq->cq_lock);
|
|
|
|
spin_unlock_irqrestore(&scq->cq_lock, *scq_flags);
|
|
|
|
} else if (scq->cq_handle < rcq->cq_handle) {
|
|
|
|
spin_unlock_irqrestore(&rcq->cq_lock, *rcq_flags);
|
|
|
|
spin_unlock_irqrestore(&scq->cq_lock, *scq_flags);
|
|
|
|
} else {
|
|
|
|
spin_unlock_irqrestore(&scq->cq_lock, *scq_flags);
|
|
|
|
spin_unlock_irqrestore(&rcq->cq_lock, *rcq_flags);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void pvrdma_reset_qp(struct pvrdma_qp *qp)
|
|
|
|
{
|
|
|
|
struct pvrdma_cq *scq, *rcq;
|
|
|
|
unsigned long scq_flags, rcq_flags;
|
|
|
|
|
|
|
|
/* Clean up cqes */
|
|
|
|
get_cqs(qp, &scq, &rcq);
|
|
|
|
pvrdma_lock_cqs(scq, rcq, &scq_flags, &rcq_flags);
|
|
|
|
|
|
|
|
_pvrdma_flush_cqe(qp, scq);
|
|
|
|
if (scq != rcq)
|
|
|
|
_pvrdma_flush_cqe(qp, rcq);
|
|
|
|
|
|
|
|
pvrdma_unlock_cqs(scq, rcq, &scq_flags, &rcq_flags);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Reset queuepair. The checks are because usermode queuepairs won't
|
|
|
|
* have kernel ringstates.
|
|
|
|
*/
|
|
|
|
if (qp->rq.ring) {
|
|
|
|
atomic_set(&qp->rq.ring->cons_head, 0);
|
|
|
|
atomic_set(&qp->rq.ring->prod_tail, 0);
|
|
|
|
}
|
|
|
|
if (qp->sq.ring) {
|
|
|
|
atomic_set(&qp->sq.ring->cons_head, 0);
|
|
|
|
atomic_set(&qp->sq.ring->prod_tail, 0);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static int pvrdma_set_rq_size(struct pvrdma_dev *dev,
|
|
|
|
struct ib_qp_cap *req_cap,
|
|
|
|
struct pvrdma_qp *qp)
|
|
|
|
{
|
|
|
|
if (req_cap->max_recv_wr > dev->dsr->caps.max_qp_wr ||
|
|
|
|
req_cap->max_recv_sge > dev->dsr->caps.max_sge) {
|
|
|
|
dev_warn(&dev->pdev->dev, "recv queue size invalid\n");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
qp->rq.wqe_cnt = roundup_pow_of_two(max(1U, req_cap->max_recv_wr));
|
|
|
|
qp->rq.max_sg = roundup_pow_of_two(max(1U, req_cap->max_recv_sge));
|
|
|
|
|
|
|
|
/* Write back */
|
|
|
|
req_cap->max_recv_wr = qp->rq.wqe_cnt;
|
|
|
|
req_cap->max_recv_sge = qp->rq.max_sg;
|
|
|
|
|
|
|
|
qp->rq.wqe_size = roundup_pow_of_two(sizeof(struct pvrdma_rq_wqe_hdr) +
|
|
|
|
sizeof(struct pvrdma_sge) *
|
|
|
|
qp->rq.max_sg);
|
|
|
|
qp->npages_recv = (qp->rq.wqe_cnt * qp->rq.wqe_size + PAGE_SIZE - 1) /
|
|
|
|
PAGE_SIZE;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int pvrdma_set_sq_size(struct pvrdma_dev *dev, struct ib_qp_cap *req_cap,
|
2017-01-17 08:34:00 +00:00
|
|
|
struct pvrdma_qp *qp)
|
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
|
|
|
{
|
|
|
|
if (req_cap->max_send_wr > dev->dsr->caps.max_qp_wr ||
|
|
|
|
req_cap->max_send_sge > dev->dsr->caps.max_sge) {
|
|
|
|
dev_warn(&dev->pdev->dev, "send queue size invalid\n");
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
qp->sq.wqe_cnt = roundup_pow_of_two(max(1U, req_cap->max_send_wr));
|
|
|
|
qp->sq.max_sg = roundup_pow_of_two(max(1U, req_cap->max_send_sge));
|
|
|
|
|
|
|
|
/* Write back */
|
|
|
|
req_cap->max_send_wr = qp->sq.wqe_cnt;
|
|
|
|
req_cap->max_send_sge = qp->sq.max_sg;
|
|
|
|
|
|
|
|
qp->sq.wqe_size = roundup_pow_of_two(sizeof(struct pvrdma_sq_wqe_hdr) +
|
|
|
|
sizeof(struct pvrdma_sge) *
|
|
|
|
qp->sq.max_sg);
|
|
|
|
/* Note: one extra page for the header. */
|
2017-02-23 01:22:57 +00:00
|
|
|
qp->npages_send = PVRDMA_QP_NUM_HEADER_PAGES +
|
|
|
|
(qp->sq.wqe_cnt * qp->sq.wqe_size + PAGE_SIZE - 1) /
|
|
|
|
PAGE_SIZE;
|
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
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* pvrdma_create_qp - create queue pair
|
|
|
|
* @pd: protection domain
|
|
|
|
* @init_attr: queue pair attributes
|
|
|
|
* @udata: user data
|
|
|
|
*
|
|
|
|
* @return: the ib_qp pointer on success, otherwise returns an errno.
|
|
|
|
*/
|
|
|
|
struct ib_qp *pvrdma_create_qp(struct ib_pd *pd,
|
|
|
|
struct ib_qp_init_attr *init_attr,
|
|
|
|
struct ib_udata *udata)
|
|
|
|
{
|
|
|
|
struct pvrdma_qp *qp = NULL;
|
|
|
|
struct pvrdma_dev *dev = to_vdev(pd->device);
|
|
|
|
union pvrdma_cmd_req req;
|
|
|
|
union pvrdma_cmd_resp rsp;
|
|
|
|
struct pvrdma_cmd_create_qp *cmd = &req.create_qp;
|
|
|
|
struct pvrdma_cmd_create_qp_resp *resp = &rsp.create_qp_resp;
|
|
|
|
struct pvrdma_create_qp ucmd;
|
|
|
|
unsigned long flags;
|
|
|
|
int ret;
|
2017-11-06 19:48:53 +00:00
|
|
|
bool is_srq = !!init_attr->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
|
|
|
|
|
|
|
if (init_attr->create_flags) {
|
|
|
|
dev_warn(&dev->pdev->dev,
|
|
|
|
"invalid create queuepair flags %#x\n",
|
|
|
|
init_attr->create_flags);
|
|
|
|
return ERR_PTR(-EINVAL);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (init_attr->qp_type != IB_QPT_RC &&
|
|
|
|
init_attr->qp_type != IB_QPT_UD &&
|
|
|
|
init_attr->qp_type != IB_QPT_GSI) {
|
|
|
|
dev_warn(&dev->pdev->dev, "queuepair type %d not supported\n",
|
|
|
|
init_attr->qp_type);
|
|
|
|
return ERR_PTR(-EINVAL);
|
|
|
|
}
|
|
|
|
|
2017-11-06 19:48:53 +00:00
|
|
|
if (is_srq && !dev->dsr->caps.max_srq) {
|
|
|
|
dev_warn(&dev->pdev->dev,
|
|
|
|
"SRQs not supported by device\n");
|
|
|
|
return ERR_PTR(-EINVAL);
|
|
|
|
}
|
|
|
|
|
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
|
|
|
if (!atomic_add_unless(&dev->num_qps, 1, dev->dsr->caps.max_qp))
|
|
|
|
return ERR_PTR(-ENOMEM);
|
|
|
|
|
|
|
|
switch (init_attr->qp_type) {
|
|
|
|
case IB_QPT_GSI:
|
|
|
|
if (init_attr->port_num == 0 ||
|
|
|
|
init_attr->port_num > pd->device->phys_port_cnt ||
|
|
|
|
udata) {
|
|
|
|
dev_warn(&dev->pdev->dev, "invalid queuepair attrs\n");
|
|
|
|
ret = -EINVAL;
|
|
|
|
goto err_qp;
|
|
|
|
}
|
|
|
|
/* fall through */
|
|
|
|
case IB_QPT_RC:
|
|
|
|
case IB_QPT_UD:
|
|
|
|
qp = kzalloc(sizeof(*qp), GFP_KERNEL);
|
|
|
|
if (!qp) {
|
|
|
|
ret = -ENOMEM;
|
|
|
|
goto err_qp;
|
|
|
|
}
|
|
|
|
|
|
|
|
spin_lock_init(&qp->sq.lock);
|
|
|
|
spin_lock_init(&qp->rq.lock);
|
|
|
|
mutex_init(&qp->mutex);
|
2017-12-20 19:27:00 +00:00
|
|
|
refcount_set(&qp->refcnt, 1);
|
2017-12-20 17:51:40 +00:00
|
|
|
init_completion(&qp->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
|
|
|
|
|
|
|
qp->state = IB_QPS_RESET;
|
2018-12-17 15:15:18 +00:00
|
|
|
qp->is_kernel = !udata;
|
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
|
|
|
|
2017-12-20 19:24:33 +00:00
|
|
|
if (!qp->is_kernel) {
|
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
|
|
|
dev_dbg(&dev->pdev->dev,
|
|
|
|
"create queuepair from user space\n");
|
|
|
|
|
|
|
|
if (ib_copy_from_udata(&ucmd, udata, sizeof(ucmd))) {
|
|
|
|
ret = -EFAULT;
|
|
|
|
goto err_qp;
|
|
|
|
}
|
|
|
|
|
2017-11-06 19:48:53 +00:00
|
|
|
if (!is_srq) {
|
|
|
|
/* set qp->sq.wqe_cnt, shift, buf_size.. */
|
2019-01-09 09:15:16 +00:00
|
|
|
qp->rumem = ib_umem_get(udata, ucmd.rbuf_addr,
|
2017-11-06 19:48:53 +00:00
|
|
|
ucmd.rbuf_size, 0, 0);
|
|
|
|
if (IS_ERR(qp->rumem)) {
|
|
|
|
ret = PTR_ERR(qp->rumem);
|
|
|
|
goto err_qp;
|
|
|
|
}
|
|
|
|
qp->srq = NULL;
|
|
|
|
} else {
|
|
|
|
qp->rumem = NULL;
|
|
|
|
qp->srq = to_vsrq(init_attr->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
|
|
|
}
|
|
|
|
|
2019-01-09 09:15:16 +00:00
|
|
|
qp->sumem = ib_umem_get(udata, ucmd.sbuf_addr,
|
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
|
|
|
ucmd.sbuf_size, 0, 0);
|
|
|
|
if (IS_ERR(qp->sumem)) {
|
2017-11-06 19:48:53 +00:00
|
|
|
if (!is_srq)
|
|
|
|
ib_umem_release(qp->rumem);
|
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
|
|
|
ret = PTR_ERR(qp->sumem);
|
|
|
|
goto err_qp;
|
|
|
|
}
|
|
|
|
|
|
|
|
qp->npages_send = ib_umem_page_count(qp->sumem);
|
2017-11-06 19:48:53 +00:00
|
|
|
if (!is_srq)
|
|
|
|
qp->npages_recv = ib_umem_page_count(qp->rumem);
|
|
|
|
else
|
|
|
|
qp->npages_recv = 0;
|
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
|
|
|
qp->npages = qp->npages_send + qp->npages_recv;
|
|
|
|
} else {
|
|
|
|
ret = pvrdma_set_sq_size(to_vdev(pd->device),
|
2017-01-17 08:34:00 +00:00
|
|
|
&init_attr->cap, qp);
|
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
|
|
|
if (ret)
|
|
|
|
goto err_qp;
|
|
|
|
|
|
|
|
ret = pvrdma_set_rq_size(to_vdev(pd->device),
|
|
|
|
&init_attr->cap, qp);
|
|
|
|
if (ret)
|
|
|
|
goto err_qp;
|
|
|
|
|
|
|
|
qp->npages = qp->npages_send + qp->npages_recv;
|
|
|
|
|
|
|
|
/* Skip header page. */
|
2017-02-23 01:22:57 +00:00
|
|
|
qp->sq.offset = PVRDMA_QP_NUM_HEADER_PAGES * PAGE_SIZE;
|
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
|
|
|
|
|
|
|
/* Recv queue pages are after send pages. */
|
|
|
|
qp->rq.offset = qp->npages_send * PAGE_SIZE;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (qp->npages < 0 || qp->npages > PVRDMA_PAGE_DIR_MAX_PAGES) {
|
|
|
|
dev_warn(&dev->pdev->dev,
|
|
|
|
"overflow pages in queuepair\n");
|
|
|
|
ret = -EINVAL;
|
|
|
|
goto err_umem;
|
|
|
|
}
|
|
|
|
|
|
|
|
ret = pvrdma_page_dir_init(dev, &qp->pdir, qp->npages,
|
|
|
|
qp->is_kernel);
|
|
|
|
if (ret) {
|
|
|
|
dev_warn(&dev->pdev->dev,
|
|
|
|
"could not allocate page directory\n");
|
|
|
|
goto err_umem;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!qp->is_kernel) {
|
|
|
|
pvrdma_page_dir_insert_umem(&qp->pdir, qp->sumem, 0);
|
2017-11-06 19:48:53 +00:00
|
|
|
if (!is_srq)
|
|
|
|
pvrdma_page_dir_insert_umem(&qp->pdir,
|
|
|
|
qp->rumem,
|
|
|
|
qp->npages_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
|
|
|
} else {
|
|
|
|
/* Ring state is always the first page. */
|
|
|
|
qp->sq.ring = qp->pdir.pages[0];
|
2017-11-06 19:48:53 +00:00
|
|
|
qp->rq.ring = is_srq ? NULL : &qp->sq.ring[1];
|
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
|
|
|
}
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
ret = -EINVAL;
|
|
|
|
goto err_qp;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Not supported */
|
|
|
|
init_attr->cap.max_inline_data = 0;
|
|
|
|
|
|
|
|
memset(cmd, 0, sizeof(*cmd));
|
|
|
|
cmd->hdr.cmd = PVRDMA_CMD_CREATE_QP;
|
|
|
|
cmd->pd_handle = to_vpd(pd)->pd_handle;
|
|
|
|
cmd->send_cq_handle = to_vcq(init_attr->send_cq)->cq_handle;
|
|
|
|
cmd->recv_cq_handle = to_vcq(init_attr->recv_cq)->cq_handle;
|
2017-11-06 19:48:53 +00:00
|
|
|
if (is_srq)
|
|
|
|
cmd->srq_handle = to_vsrq(init_attr->srq)->srq_handle;
|
|
|
|
else
|
|
|
|
cmd->srq_handle = 0;
|
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
|
|
|
cmd->max_send_wr = init_attr->cap.max_send_wr;
|
|
|
|
cmd->max_recv_wr = init_attr->cap.max_recv_wr;
|
|
|
|
cmd->max_send_sge = init_attr->cap.max_send_sge;
|
|
|
|
cmd->max_recv_sge = init_attr->cap.max_recv_sge;
|
|
|
|
cmd->max_inline_data = init_attr->cap.max_inline_data;
|
|
|
|
cmd->sq_sig_all = (init_attr->sq_sig_type == IB_SIGNAL_ALL_WR) ? 1 : 0;
|
|
|
|
cmd->qp_type = ib_qp_type_to_pvrdma(init_attr->qp_type);
|
2017-11-06 19:48:53 +00:00
|
|
|
cmd->is_srq = is_srq;
|
|
|
|
cmd->lkey = 0;
|
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
|
|
|
cmd->access_flags = IB_ACCESS_LOCAL_WRITE;
|
|
|
|
cmd->total_chunks = qp->npages;
|
2017-02-23 01:22:57 +00:00
|
|
|
cmd->send_chunks = qp->npages_send - PVRDMA_QP_NUM_HEADER_PAGES;
|
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
|
|
|
cmd->pdir_dma = qp->pdir.dir_dma;
|
|
|
|
|
|
|
|
dev_dbg(&dev->pdev->dev, "create queuepair with %d, %d, %d, %d\n",
|
|
|
|
cmd->max_send_wr, cmd->max_recv_wr, cmd->max_send_sge,
|
|
|
|
cmd->max_recv_sge);
|
|
|
|
|
|
|
|
ret = pvrdma_cmd_post(dev, &req, &rsp, PVRDMA_CMD_CREATE_QP_RESP);
|
|
|
|
if (ret < 0) {
|
|
|
|
dev_warn(&dev->pdev->dev,
|
|
|
|
"could not create queuepair, error: %d\n", ret);
|
|
|
|
goto err_pdir;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* max_send_wr/_recv_wr/_send_sge/_recv_sge/_inline_data */
|
|
|
|
qp->qp_handle = resp->qpn;
|
|
|
|
qp->port = init_attr->port_num;
|
|
|
|
qp->ibqp.qp_num = resp->qpn;
|
|
|
|
spin_lock_irqsave(&dev->qp_tbl_lock, flags);
|
|
|
|
dev->qp_tbl[qp->qp_handle % dev->dsr->caps.max_qp] = qp;
|
|
|
|
spin_unlock_irqrestore(&dev->qp_tbl_lock, flags);
|
|
|
|
|
|
|
|
return &qp->ibqp;
|
|
|
|
|
|
|
|
err_pdir:
|
|
|
|
pvrdma_page_dir_cleanup(dev, &qp->pdir);
|
|
|
|
err_umem:
|
2017-12-20 19:24:33 +00:00
|
|
|
if (!qp->is_kernel) {
|
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
|
|
|
if (qp->rumem)
|
|
|
|
ib_umem_release(qp->rumem);
|
|
|
|
if (qp->sumem)
|
|
|
|
ib_umem_release(qp->sumem);
|
|
|
|
}
|
|
|
|
err_qp:
|
|
|
|
kfree(qp);
|
|
|
|
atomic_dec(&dev->num_qps);
|
|
|
|
|
|
|
|
return ERR_PTR(ret);
|
|
|
|
}
|
|
|
|
|
|
|
|
static void pvrdma_free_qp(struct pvrdma_qp *qp)
|
|
|
|
{
|
|
|
|
struct pvrdma_dev *dev = to_vdev(qp->ibqp.device);
|
|
|
|
struct pvrdma_cq *scq;
|
|
|
|
struct pvrdma_cq *rcq;
|
|
|
|
unsigned long flags, scq_flags, rcq_flags;
|
|
|
|
|
|
|
|
/* In case cq is polling */
|
|
|
|
get_cqs(qp, &scq, &rcq);
|
|
|
|
pvrdma_lock_cqs(scq, rcq, &scq_flags, &rcq_flags);
|
|
|
|
|
|
|
|
_pvrdma_flush_cqe(qp, scq);
|
|
|
|
if (scq != rcq)
|
|
|
|
_pvrdma_flush_cqe(qp, rcq);
|
|
|
|
|
|
|
|
spin_lock_irqsave(&dev->qp_tbl_lock, flags);
|
|
|
|
dev->qp_tbl[qp->qp_handle] = NULL;
|
|
|
|
spin_unlock_irqrestore(&dev->qp_tbl_lock, flags);
|
|
|
|
|
|
|
|
pvrdma_unlock_cqs(scq, rcq, &scq_flags, &rcq_flags);
|
|
|
|
|
2017-12-20 19:27:00 +00:00
|
|
|
if (refcount_dec_and_test(&qp->refcnt))
|
2017-12-20 17:51:40 +00:00
|
|
|
complete(&qp->free);
|
|
|
|
wait_for_completion(&qp->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
|
|
|
|
2017-12-20 17:49:03 +00:00
|
|
|
if (!qp->is_kernel) {
|
|
|
|
if (qp->rumem)
|
|
|
|
ib_umem_release(qp->rumem);
|
|
|
|
if (qp->sumem)
|
|
|
|
ib_umem_release(qp->sumem);
|
|
|
|
}
|
|
|
|
|
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
|
|
|
pvrdma_page_dir_cleanup(dev, &qp->pdir);
|
|
|
|
|
|
|
|
kfree(qp);
|
|
|
|
|
|
|
|
atomic_dec(&dev->num_qps);
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* pvrdma_destroy_qp - destroy a queue pair
|
|
|
|
* @qp: the queue pair to destroy
|
|
|
|
*
|
|
|
|
* @return: 0 on success.
|
|
|
|
*/
|
|
|
|
int pvrdma_destroy_qp(struct ib_qp *qp)
|
|
|
|
{
|
|
|
|
struct pvrdma_qp *vqp = to_vqp(qp);
|
|
|
|
union pvrdma_cmd_req req;
|
|
|
|
struct pvrdma_cmd_destroy_qp *cmd = &req.destroy_qp;
|
|
|
|
int ret;
|
|
|
|
|
|
|
|
memset(cmd, 0, sizeof(*cmd));
|
|
|
|
cmd->hdr.cmd = PVRDMA_CMD_DESTROY_QP;
|
|
|
|
cmd->qp_handle = vqp->qp_handle;
|
|
|
|
|
|
|
|
ret = pvrdma_cmd_post(to_vdev(qp->device), &req, NULL, 0);
|
|
|
|
if (ret < 0)
|
|
|
|
dev_warn(&to_vdev(qp->device)->pdev->dev,
|
|
|
|
"destroy queuepair failed, error: %d\n", ret);
|
|
|
|
|
|
|
|
pvrdma_free_qp(vqp);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* pvrdma_modify_qp - modify queue pair attributes
|
|
|
|
* @ibqp: the queue pair
|
|
|
|
* @attr: the new queue pair's attributes
|
|
|
|
* @attr_mask: attributes mask
|
|
|
|
* @udata: user data
|
|
|
|
*
|
|
|
|
* @returns 0 on success, otherwise returns an errno.
|
|
|
|
*/
|
|
|
|
int pvrdma_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
|
|
|
|
int attr_mask, struct ib_udata *udata)
|
|
|
|
{
|
|
|
|
struct pvrdma_dev *dev = to_vdev(ibqp->device);
|
|
|
|
struct pvrdma_qp *qp = to_vqp(ibqp);
|
|
|
|
union pvrdma_cmd_req req;
|
|
|
|
union pvrdma_cmd_resp rsp;
|
|
|
|
struct pvrdma_cmd_modify_qp *cmd = &req.modify_qp;
|
2018-03-11 11:51:34 +00:00
|
|
|
enum ib_qp_state cur_state, next_state;
|
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 ret;
|
|
|
|
|
|
|
|
/* Sanity checking. Should need lock here */
|
|
|
|
mutex_lock(&qp->mutex);
|
|
|
|
cur_state = (attr_mask & IB_QP_CUR_STATE) ? attr->cur_qp_state :
|
|
|
|
qp->state;
|
|
|
|
next_state = (attr_mask & IB_QP_STATE) ? attr->qp_state : cur_state;
|
|
|
|
|
|
|
|
if (!ib_modify_qp_is_ok(cur_state, next_state, ibqp->qp_type,
|
2018-10-02 13:11:21 +00:00
|
|
|
attr_mask)) {
|
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
|
|
|
ret = -EINVAL;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (attr_mask & IB_QP_PORT) {
|
|
|
|
if (attr->port_num == 0 ||
|
|
|
|
attr->port_num > ibqp->device->phys_port_cnt) {
|
|
|
|
ret = -EINVAL;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (attr_mask & IB_QP_MIN_RNR_TIMER) {
|
|
|
|
if (attr->min_rnr_timer > 31) {
|
|
|
|
ret = -EINVAL;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (attr_mask & IB_QP_PKEY_INDEX) {
|
|
|
|
if (attr->pkey_index >= dev->dsr->caps.max_pkeys) {
|
|
|
|
ret = -EINVAL;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (attr_mask & IB_QP_QKEY)
|
|
|
|
qp->qkey = attr->qkey;
|
|
|
|
|
|
|
|
if (cur_state == next_state && cur_state == IB_QPS_RESET) {
|
|
|
|
ret = 0;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
qp->state = next_state;
|
|
|
|
memset(cmd, 0, sizeof(*cmd));
|
|
|
|
cmd->hdr.cmd = PVRDMA_CMD_MODIFY_QP;
|
|
|
|
cmd->qp_handle = qp->qp_handle;
|
|
|
|
cmd->attr_mask = ib_qp_attr_mask_to_pvrdma(attr_mask);
|
|
|
|
cmd->attrs.qp_state = ib_qp_state_to_pvrdma(attr->qp_state);
|
|
|
|
cmd->attrs.cur_qp_state =
|
|
|
|
ib_qp_state_to_pvrdma(attr->cur_qp_state);
|
|
|
|
cmd->attrs.path_mtu = ib_mtu_to_pvrdma(attr->path_mtu);
|
|
|
|
cmd->attrs.path_mig_state =
|
|
|
|
ib_mig_state_to_pvrdma(attr->path_mig_state);
|
|
|
|
cmd->attrs.qkey = attr->qkey;
|
|
|
|
cmd->attrs.rq_psn = attr->rq_psn;
|
|
|
|
cmd->attrs.sq_psn = attr->sq_psn;
|
|
|
|
cmd->attrs.dest_qp_num = attr->dest_qp_num;
|
|
|
|
cmd->attrs.qp_access_flags =
|
|
|
|
ib_access_flags_to_pvrdma(attr->qp_access_flags);
|
|
|
|
cmd->attrs.pkey_index = attr->pkey_index;
|
|
|
|
cmd->attrs.alt_pkey_index = attr->alt_pkey_index;
|
|
|
|
cmd->attrs.en_sqd_async_notify = attr->en_sqd_async_notify;
|
|
|
|
cmd->attrs.sq_draining = attr->sq_draining;
|
|
|
|
cmd->attrs.max_rd_atomic = attr->max_rd_atomic;
|
|
|
|
cmd->attrs.max_dest_rd_atomic = attr->max_dest_rd_atomic;
|
|
|
|
cmd->attrs.min_rnr_timer = attr->min_rnr_timer;
|
|
|
|
cmd->attrs.port_num = attr->port_num;
|
|
|
|
cmd->attrs.timeout = attr->timeout;
|
|
|
|
cmd->attrs.retry_cnt = attr->retry_cnt;
|
|
|
|
cmd->attrs.rnr_retry = attr->rnr_retry;
|
|
|
|
cmd->attrs.alt_port_num = attr->alt_port_num;
|
|
|
|
cmd->attrs.alt_timeout = attr->alt_timeout;
|
|
|
|
ib_qp_cap_to_pvrdma(&cmd->attrs.cap, &attr->cap);
|
2017-04-29 18:41:26 +00:00
|
|
|
rdma_ah_attr_to_pvrdma(&cmd->attrs.ah_attr, &attr->ah_attr);
|
|
|
|
rdma_ah_attr_to_pvrdma(&cmd->attrs.alt_ah_attr, &attr->alt_ah_attr);
|
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
|
|
|
|
|
|
|
ret = pvrdma_cmd_post(dev, &req, &rsp, PVRDMA_CMD_MODIFY_QP_RESP);
|
|
|
|
if (ret < 0) {
|
|
|
|
dev_warn(&dev->pdev->dev,
|
|
|
|
"could not modify queuepair, error: %d\n", ret);
|
|
|
|
} else if (rsp.hdr.err > 0) {
|
|
|
|
dev_warn(&dev->pdev->dev,
|
|
|
|
"cannot modify queuepair, error: %d\n", rsp.hdr.err);
|
|
|
|
ret = -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (ret == 0 && next_state == IB_QPS_RESET)
|
|
|
|
pvrdma_reset_qp(qp);
|
|
|
|
|
|
|
|
out:
|
|
|
|
mutex_unlock(&qp->mutex);
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
2017-02-23 01:22:56 +00:00
|
|
|
static inline void *get_sq_wqe(struct pvrdma_qp *qp, unsigned int n)
|
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
|
|
|
{
|
|
|
|
return pvrdma_page_dir_get_ptr(&qp->pdir,
|
|
|
|
qp->sq.offset + n * qp->sq.wqe_size);
|
|
|
|
}
|
|
|
|
|
2017-02-23 01:22:56 +00:00
|
|
|
static inline void *get_rq_wqe(struct pvrdma_qp *qp, unsigned int n)
|
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
|
|
|
{
|
|
|
|
return pvrdma_page_dir_get_ptr(&qp->pdir,
|
|
|
|
qp->rq.offset + n * qp->rq.wqe_size);
|
|
|
|
}
|
|
|
|
|
2018-07-18 16:25:14 +00:00
|
|
|
static int set_reg_seg(struct pvrdma_sq_wqe_hdr *wqe_hdr,
|
|
|
|
const struct ib_reg_wr *wr)
|
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_user_mr *mr = to_vmr(wr->mr);
|
|
|
|
|
|
|
|
wqe_hdr->wr.fast_reg.iova_start = mr->ibmr.iova;
|
|
|
|
wqe_hdr->wr.fast_reg.pl_pdir_dma = mr->pdir.dir_dma;
|
|
|
|
wqe_hdr->wr.fast_reg.page_shift = mr->page_shift;
|
|
|
|
wqe_hdr->wr.fast_reg.page_list_len = mr->npages;
|
|
|
|
wqe_hdr->wr.fast_reg.length = mr->ibmr.length;
|
|
|
|
wqe_hdr->wr.fast_reg.access_flags = wr->access;
|
|
|
|
wqe_hdr->wr.fast_reg.rkey = wr->key;
|
|
|
|
|
|
|
|
return pvrdma_page_dir_insert_page_list(&mr->pdir, mr->pages,
|
|
|
|
mr->npages);
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* pvrdma_post_send - post send work request entries on a QP
|
|
|
|
* @ibqp: the QP
|
|
|
|
* @wr: work request list to post
|
|
|
|
* @bad_wr: the first bad WR returned
|
|
|
|
*
|
|
|
|
* @return: 0 on success, otherwise errno returned.
|
|
|
|
*/
|
2018-07-18 16:25:32 +00:00
|
|
|
int pvrdma_post_send(struct ib_qp *ibqp, const struct ib_send_wr *wr,
|
|
|
|
const struct ib_send_wr **bad_wr)
|
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 = to_vqp(ibqp);
|
|
|
|
struct pvrdma_dev *dev = to_vdev(ibqp->device);
|
|
|
|
unsigned long flags;
|
|
|
|
struct pvrdma_sq_wqe_hdr *wqe_hdr;
|
|
|
|
struct pvrdma_sge *sge;
|
2017-02-23 01:22:56 +00:00
|
|
|
int i, ret;
|
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
|
|
|
|
|
|
|
/*
|
|
|
|
* In states lower than RTS, we can fail immediately. In other states,
|
|
|
|
* just post and let the device figure it out.
|
|
|
|
*/
|
|
|
|
if (qp->state < IB_QPS_RTS) {
|
|
|
|
*bad_wr = wr;
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
|
|
|
spin_lock_irqsave(&qp->sq.lock, flags);
|
|
|
|
|
2017-02-23 01:22:56 +00:00
|
|
|
while (wr) {
|
|
|
|
unsigned int tail = 0;
|
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
|
|
|
|
|
|
|
if (unlikely(!pvrdma_idx_ring_has_space(
|
|
|
|
qp->sq.ring, qp->sq.wqe_cnt, &tail))) {
|
|
|
|
dev_warn_ratelimited(&dev->pdev->dev,
|
|
|
|
"send queue is full\n");
|
|
|
|
*bad_wr = wr;
|
|
|
|
ret = -ENOMEM;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (unlikely(wr->num_sge > qp->sq.max_sg || wr->num_sge < 0)) {
|
|
|
|
dev_warn_ratelimited(&dev->pdev->dev,
|
|
|
|
"send SGE overflow\n");
|
|
|
|
*bad_wr = wr;
|
|
|
|
ret = -EINVAL;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (unlikely(wr->opcode < 0)) {
|
|
|
|
dev_warn_ratelimited(&dev->pdev->dev,
|
|
|
|
"invalid send opcode\n");
|
|
|
|
*bad_wr = wr;
|
|
|
|
ret = -EINVAL;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Only support UD, RC.
|
|
|
|
* Need to check opcode table for thorough checking.
|
|
|
|
* opcode _UD _UC _RC
|
|
|
|
* _SEND x x x
|
|
|
|
* _SEND_WITH_IMM x x x
|
|
|
|
* _RDMA_WRITE x x
|
|
|
|
* _RDMA_WRITE_WITH_IMM x x
|
|
|
|
* _LOCAL_INV x x
|
|
|
|
* _SEND_WITH_INV x x
|
|
|
|
* _RDMA_READ x
|
|
|
|
* _ATOMIC_CMP_AND_SWP x
|
|
|
|
* _ATOMIC_FETCH_AND_ADD x
|
|
|
|
* _MASK_ATOMIC_CMP_AND_SWP x
|
|
|
|
* _MASK_ATOMIC_FETCH_AND_ADD x
|
|
|
|
* _REG_MR x
|
|
|
|
*
|
|
|
|
*/
|
|
|
|
if (qp->ibqp.qp_type != IB_QPT_UD &&
|
|
|
|
qp->ibqp.qp_type != IB_QPT_RC &&
|
|
|
|
wr->opcode != IB_WR_SEND) {
|
|
|
|
dev_warn_ratelimited(&dev->pdev->dev,
|
|
|
|
"unsupported queuepair type\n");
|
|
|
|
*bad_wr = wr;
|
|
|
|
ret = -EINVAL;
|
|
|
|
goto out;
|
|
|
|
} else if (qp->ibqp.qp_type == IB_QPT_UD ||
|
|
|
|
qp->ibqp.qp_type == IB_QPT_GSI) {
|
|
|
|
if (wr->opcode != IB_WR_SEND &&
|
|
|
|
wr->opcode != IB_WR_SEND_WITH_IMM) {
|
|
|
|
dev_warn_ratelimited(&dev->pdev->dev,
|
|
|
|
"invalid send opcode\n");
|
|
|
|
*bad_wr = wr;
|
|
|
|
ret = -EINVAL;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2017-02-23 01:22:56 +00:00
|
|
|
wqe_hdr = (struct pvrdma_sq_wqe_hdr *)get_sq_wqe(qp, tail);
|
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
|
|
|
memset(wqe_hdr, 0, sizeof(*wqe_hdr));
|
|
|
|
wqe_hdr->wr_id = wr->wr_id;
|
|
|
|
wqe_hdr->num_sge = wr->num_sge;
|
|
|
|
wqe_hdr->opcode = ib_wr_opcode_to_pvrdma(wr->opcode);
|
|
|
|
wqe_hdr->send_flags = ib_send_flags_to_pvrdma(wr->send_flags);
|
|
|
|
if (wr->opcode == IB_WR_SEND_WITH_IMM ||
|
|
|
|
wr->opcode == IB_WR_RDMA_WRITE_WITH_IMM)
|
|
|
|
wqe_hdr->ex.imm_data = wr->ex.imm_data;
|
|
|
|
|
|
|
|
switch (qp->ibqp.qp_type) {
|
|
|
|
case IB_QPT_GSI:
|
|
|
|
case IB_QPT_UD:
|
|
|
|
if (unlikely(!ud_wr(wr)->ah)) {
|
|
|
|
dev_warn_ratelimited(&dev->pdev->dev,
|
|
|
|
"invalid address handle\n");
|
|
|
|
*bad_wr = wr;
|
|
|
|
ret = -EINVAL;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Use qkey from qp context if high order bit set,
|
|
|
|
* otherwise from work request.
|
|
|
|
*/
|
|
|
|
wqe_hdr->wr.ud.remote_qpn = ud_wr(wr)->remote_qpn;
|
|
|
|
wqe_hdr->wr.ud.remote_qkey =
|
|
|
|
ud_wr(wr)->remote_qkey & 0x80000000 ?
|
|
|
|
qp->qkey : ud_wr(wr)->remote_qkey;
|
|
|
|
wqe_hdr->wr.ud.av = to_vah(ud_wr(wr)->ah)->av;
|
|
|
|
|
|
|
|
break;
|
|
|
|
case IB_QPT_RC:
|
|
|
|
switch (wr->opcode) {
|
|
|
|
case IB_WR_RDMA_READ:
|
|
|
|
case IB_WR_RDMA_WRITE:
|
|
|
|
case IB_WR_RDMA_WRITE_WITH_IMM:
|
|
|
|
wqe_hdr->wr.rdma.remote_addr =
|
|
|
|
rdma_wr(wr)->remote_addr;
|
|
|
|
wqe_hdr->wr.rdma.rkey = rdma_wr(wr)->rkey;
|
|
|
|
break;
|
|
|
|
case IB_WR_LOCAL_INV:
|
|
|
|
case IB_WR_SEND_WITH_INV:
|
|
|
|
wqe_hdr->ex.invalidate_rkey =
|
|
|
|
wr->ex.invalidate_rkey;
|
|
|
|
break;
|
|
|
|
case IB_WR_ATOMIC_CMP_AND_SWP:
|
|
|
|
case IB_WR_ATOMIC_FETCH_AND_ADD:
|
|
|
|
wqe_hdr->wr.atomic.remote_addr =
|
|
|
|
atomic_wr(wr)->remote_addr;
|
|
|
|
wqe_hdr->wr.atomic.rkey = atomic_wr(wr)->rkey;
|
|
|
|
wqe_hdr->wr.atomic.compare_add =
|
|
|
|
atomic_wr(wr)->compare_add;
|
|
|
|
if (wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP)
|
|
|
|
wqe_hdr->wr.atomic.swap =
|
|
|
|
atomic_wr(wr)->swap;
|
|
|
|
break;
|
|
|
|
case IB_WR_REG_MR:
|
|
|
|
ret = set_reg_seg(wqe_hdr, reg_wr(wr));
|
|
|
|
if (ret < 0) {
|
|
|
|
dev_warn_ratelimited(&dev->pdev->dev,
|
|
|
|
"Failed to set fast register work request\n");
|
|
|
|
*bad_wr = wr;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
dev_warn_ratelimited(&dev->pdev->dev,
|
|
|
|
"invalid queuepair type\n");
|
|
|
|
ret = -EINVAL;
|
|
|
|
*bad_wr = wr;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
sge = (struct pvrdma_sge *)(wqe_hdr + 1);
|
|
|
|
for (i = 0; i < wr->num_sge; i++) {
|
|
|
|
/* Need to check wqe_size 0 or max size */
|
|
|
|
sge->addr = wr->sg_list[i].addr;
|
|
|
|
sge->length = wr->sg_list[i].length;
|
|
|
|
sge->lkey = wr->sg_list[i].lkey;
|
|
|
|
sge++;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Make sure wqe is written before index update */
|
|
|
|
smp_wmb();
|
|
|
|
|
|
|
|
/* Update shared sq ring */
|
|
|
|
pvrdma_idx_ring_inc(&qp->sq.ring->prod_tail,
|
|
|
|
qp->sq.wqe_cnt);
|
2017-02-23 01:22:56 +00:00
|
|
|
|
|
|
|
wr = wr->next;
|
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
|
|
|
}
|
|
|
|
|
|
|
|
ret = 0;
|
|
|
|
|
|
|
|
out:
|
|
|
|
spin_unlock_irqrestore(&qp->sq.lock, flags);
|
|
|
|
|
|
|
|
if (!ret)
|
|
|
|
pvrdma_write_uar_qp(dev, PVRDMA_UAR_QP_SEND | qp->qp_handle);
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* pvrdma_post_receive - post receive work request entries on a QP
|
|
|
|
* @ibqp: the QP
|
|
|
|
* @wr: the work request list to post
|
|
|
|
* @bad_wr: the first bad WR returned
|
|
|
|
*
|
|
|
|
* @return: 0 on success, otherwise errno returned.
|
|
|
|
*/
|
2018-07-18 16:25:32 +00:00
|
|
|
int pvrdma_post_recv(struct ib_qp *ibqp, const struct ib_recv_wr *wr,
|
|
|
|
const struct ib_recv_wr **bad_wr)
|
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 *dev = to_vdev(ibqp->device);
|
|
|
|
unsigned long flags;
|
|
|
|
struct pvrdma_qp *qp = to_vqp(ibqp);
|
|
|
|
struct pvrdma_rq_wqe_hdr *wqe_hdr;
|
|
|
|
struct pvrdma_sge *sge;
|
|
|
|
int ret = 0;
|
|
|
|
int i;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* In the RESET state, we can fail immediately. For other states,
|
|
|
|
* just post and let the device figure it out.
|
|
|
|
*/
|
|
|
|
if (qp->state == IB_QPS_RESET) {
|
|
|
|
*bad_wr = wr;
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
2017-11-06 19:48:53 +00:00
|
|
|
if (qp->srq) {
|
|
|
|
dev_warn(&dev->pdev->dev, "QP associated with SRQ\n");
|
|
|
|
*bad_wr = wr;
|
|
|
|
return -EINVAL;
|
|
|
|
}
|
|
|
|
|
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
|
|
|
spin_lock_irqsave(&qp->rq.lock, flags);
|
|
|
|
|
2017-02-23 01:22:56 +00:00
|
|
|
while (wr) {
|
|
|
|
unsigned int tail = 0;
|
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
|
|
|
|
|
|
|
if (unlikely(wr->num_sge > qp->rq.max_sg ||
|
|
|
|
wr->num_sge < 0)) {
|
|
|
|
ret = -EINVAL;
|
|
|
|
*bad_wr = wr;
|
|
|
|
dev_warn_ratelimited(&dev->pdev->dev,
|
|
|
|
"recv SGE overflow\n");
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (unlikely(!pvrdma_idx_ring_has_space(
|
|
|
|
qp->rq.ring, qp->rq.wqe_cnt, &tail))) {
|
|
|
|
ret = -ENOMEM;
|
|
|
|
*bad_wr = wr;
|
|
|
|
dev_warn_ratelimited(&dev->pdev->dev,
|
|
|
|
"recv queue full\n");
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
2017-02-23 01:22:56 +00:00
|
|
|
wqe_hdr = (struct pvrdma_rq_wqe_hdr *)get_rq_wqe(qp, tail);
|
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
|
|
|
wqe_hdr->wr_id = wr->wr_id;
|
|
|
|
wqe_hdr->num_sge = wr->num_sge;
|
|
|
|
wqe_hdr->total_len = 0;
|
|
|
|
|
|
|
|
sge = (struct pvrdma_sge *)(wqe_hdr + 1);
|
|
|
|
for (i = 0; i < wr->num_sge; i++) {
|
|
|
|
sge->addr = wr->sg_list[i].addr;
|
|
|
|
sge->length = wr->sg_list[i].length;
|
|
|
|
sge->lkey = wr->sg_list[i].lkey;
|
|
|
|
sge++;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Make sure wqe is written before index update */
|
|
|
|
smp_wmb();
|
|
|
|
|
|
|
|
/* Update shared rq ring */
|
|
|
|
pvrdma_idx_ring_inc(&qp->rq.ring->prod_tail,
|
|
|
|
qp->rq.wqe_cnt);
|
2017-02-23 01:22:56 +00:00
|
|
|
|
|
|
|
wr = wr->next;
|
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
|
|
|
}
|
|
|
|
|
|
|
|
spin_unlock_irqrestore(&qp->rq.lock, flags);
|
|
|
|
|
|
|
|
pvrdma_write_uar_qp(dev, PVRDMA_UAR_QP_RECV | qp->qp_handle);
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
|
|
|
|
out:
|
|
|
|
spin_unlock_irqrestore(&qp->rq.lock, flags);
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* pvrdma_query_qp - query a queue pair's attributes
|
|
|
|
* @ibqp: the queue pair to query
|
|
|
|
* @attr: the queue pair's attributes
|
|
|
|
* @attr_mask: attributes mask
|
|
|
|
* @init_attr: initial queue pair attributes
|
|
|
|
*
|
|
|
|
* @returns 0 on success, otherwise returns an errno.
|
|
|
|
*/
|
|
|
|
int pvrdma_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
|
|
|
|
int attr_mask, struct ib_qp_init_attr *init_attr)
|
|
|
|
{
|
|
|
|
struct pvrdma_dev *dev = to_vdev(ibqp->device);
|
|
|
|
struct pvrdma_qp *qp = to_vqp(ibqp);
|
|
|
|
union pvrdma_cmd_req req;
|
|
|
|
union pvrdma_cmd_resp rsp;
|
|
|
|
struct pvrdma_cmd_query_qp *cmd = &req.query_qp;
|
|
|
|
struct pvrdma_cmd_query_qp_resp *resp = &rsp.query_qp_resp;
|
|
|
|
int ret = 0;
|
|
|
|
|
|
|
|
mutex_lock(&qp->mutex);
|
|
|
|
|
|
|
|
if (qp->state == IB_QPS_RESET) {
|
|
|
|
attr->qp_state = IB_QPS_RESET;
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
memset(cmd, 0, sizeof(*cmd));
|
|
|
|
cmd->hdr.cmd = PVRDMA_CMD_QUERY_QP;
|
|
|
|
cmd->qp_handle = qp->qp_handle;
|
|
|
|
cmd->attr_mask = ib_qp_attr_mask_to_pvrdma(attr_mask);
|
|
|
|
|
|
|
|
ret = pvrdma_cmd_post(dev, &req, &rsp, PVRDMA_CMD_QUERY_QP_RESP);
|
|
|
|
if (ret < 0) {
|
|
|
|
dev_warn(&dev->pdev->dev,
|
|
|
|
"could not query queuepair, error: %d\n", ret);
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
attr->qp_state = pvrdma_qp_state_to_ib(resp->attrs.qp_state);
|
|
|
|
attr->cur_qp_state =
|
|
|
|
pvrdma_qp_state_to_ib(resp->attrs.cur_qp_state);
|
|
|
|
attr->path_mtu = pvrdma_mtu_to_ib(resp->attrs.path_mtu);
|
|
|
|
attr->path_mig_state =
|
|
|
|
pvrdma_mig_state_to_ib(resp->attrs.path_mig_state);
|
|
|
|
attr->qkey = resp->attrs.qkey;
|
|
|
|
attr->rq_psn = resp->attrs.rq_psn;
|
|
|
|
attr->sq_psn = resp->attrs.sq_psn;
|
|
|
|
attr->dest_qp_num = resp->attrs.dest_qp_num;
|
|
|
|
attr->qp_access_flags =
|
|
|
|
pvrdma_access_flags_to_ib(resp->attrs.qp_access_flags);
|
|
|
|
attr->pkey_index = resp->attrs.pkey_index;
|
|
|
|
attr->alt_pkey_index = resp->attrs.alt_pkey_index;
|
|
|
|
attr->en_sqd_async_notify = resp->attrs.en_sqd_async_notify;
|
|
|
|
attr->sq_draining = resp->attrs.sq_draining;
|
|
|
|
attr->max_rd_atomic = resp->attrs.max_rd_atomic;
|
|
|
|
attr->max_dest_rd_atomic = resp->attrs.max_dest_rd_atomic;
|
|
|
|
attr->min_rnr_timer = resp->attrs.min_rnr_timer;
|
|
|
|
attr->port_num = resp->attrs.port_num;
|
|
|
|
attr->timeout = resp->attrs.timeout;
|
|
|
|
attr->retry_cnt = resp->attrs.retry_cnt;
|
|
|
|
attr->rnr_retry = resp->attrs.rnr_retry;
|
|
|
|
attr->alt_port_num = resp->attrs.alt_port_num;
|
|
|
|
attr->alt_timeout = resp->attrs.alt_timeout;
|
|
|
|
pvrdma_qp_cap_to_ib(&attr->cap, &resp->attrs.cap);
|
2017-04-29 18:41:26 +00:00
|
|
|
pvrdma_ah_attr_to_rdma(&attr->ah_attr, &resp->attrs.ah_attr);
|
|
|
|
pvrdma_ah_attr_to_rdma(&attr->alt_ah_attr, &resp->attrs.alt_ah_attr);
|
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
|
|
|
|
|
|
|
qp->state = attr->qp_state;
|
|
|
|
|
|
|
|
ret = 0;
|
|
|
|
|
|
|
|
out:
|
|
|
|
attr->cur_qp_state = attr->qp_state;
|
|
|
|
|
|
|
|
init_attr->event_handler = qp->ibqp.event_handler;
|
|
|
|
init_attr->qp_context = qp->ibqp.qp_context;
|
|
|
|
init_attr->send_cq = qp->ibqp.send_cq;
|
|
|
|
init_attr->recv_cq = qp->ibqp.recv_cq;
|
|
|
|
init_attr->srq = qp->ibqp.srq;
|
|
|
|
init_attr->xrcd = NULL;
|
|
|
|
init_attr->cap = attr->cap;
|
|
|
|
init_attr->sq_sig_type = 0;
|
|
|
|
init_attr->qp_type = qp->ibqp.qp_type;
|
|
|
|
init_attr->create_flags = 0;
|
|
|
|
init_attr->port_num = qp->port;
|
|
|
|
|
|
|
|
mutex_unlock(&qp->mutex);
|
|
|
|
return ret;
|
|
|
|
}
|