linux/drivers/infiniband/hw/qib/qib_mr.c

537 lines
12 KiB
C
Raw Normal View History

/*
* Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
* Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
*
* This software is available to you under a choice of one of two
* licenses. You may choose to be licensed under the terms of the GNU
* General Public License (GPL) Version 2, available from the file
* COPYING in the main directory of this source tree, or the
* OpenIB.org BSD license below:
*
* Redistribution and use in source and binary forms, with or
* without modification, are permitted provided that the following
* conditions are met:
*
* - Redistributions of source code must retain the above
* copyright notice, this list of conditions and the following
* disclaimer.
*
* - Redistributions in binary form must reproduce the above
* copyright notice, this list of conditions and the following
* disclaimer in the documentation and/or other materials
* provided with the distribution.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
* NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
* BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
* ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
#include <rdma/ib_umem.h>
#include <rdma/ib_smi.h>
#include "qib.h"
/* Fast memory region */
struct qib_fmr {
struct ib_fmr ibfmr;
struct qib_mregion mr; /* must be last */
};
static inline struct qib_fmr *to_ifmr(struct ib_fmr *ibfmr)
{
return container_of(ibfmr, struct qib_fmr, ibfmr);
}
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
static int init_qib_mregion(struct qib_mregion *mr, struct ib_pd *pd,
int count)
{
int m, i = 0;
int rval = 0;
m = (count + QIB_SEGSZ - 1) / QIB_SEGSZ;
for (; i < m; i++) {
mr->map[i] = kzalloc(sizeof *mr->map[0], GFP_KERNEL);
if (!mr->map[i])
goto bail;
}
mr->mapsz = m;
init_completion(&mr->comp);
/* count returning the ptr to user */
atomic_set(&mr->refcount, 1);
mr->pd = pd;
mr->max_segs = count;
out:
return rval;
bail:
while (i)
kfree(mr->map[--i]);
rval = -ENOMEM;
goto out;
}
static void deinit_qib_mregion(struct qib_mregion *mr)
{
int i = mr->mapsz;
mr->mapsz = 0;
while (i)
kfree(mr->map[--i]);
}
/**
* qib_get_dma_mr - get a DMA memory region
* @pd: protection domain for this memory region
* @acc: access flags
*
* Returns the memory region on success, otherwise returns an errno.
* Note that all DMA addresses should be created via the
* struct ib_dma_mapping_ops functions (see qib_dma.c).
*/
struct ib_mr *qib_get_dma_mr(struct ib_pd *pd, int acc)
{
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
struct qib_mr *mr = NULL;
struct ib_mr *ret;
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
int rval;
if (to_ipd(pd)->user) {
ret = ERR_PTR(-EPERM);
goto bail;
}
mr = kzalloc(sizeof *mr, GFP_KERNEL);
if (!mr) {
ret = ERR_PTR(-ENOMEM);
goto bail;
}
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
rval = init_qib_mregion(&mr->mr, pd, 0);
if (rval) {
ret = ERR_PTR(rval);
goto bail;
}
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
rval = qib_alloc_lkey(&mr->mr, 1);
if (rval) {
ret = ERR_PTR(rval);
goto bail_mregion;
}
mr->mr.access_flags = acc;
ret = &mr->ibmr;
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
done:
return ret;
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
bail_mregion:
deinit_qib_mregion(&mr->mr);
bail:
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
kfree(mr);
goto done;
}
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
static struct qib_mr *alloc_mr(int count, struct ib_pd *pd)
{
struct qib_mr *mr;
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
int rval = -ENOMEM;
int m;
/* Allocate struct plus pointers to first level page tables. */
m = (count + QIB_SEGSZ - 1) / QIB_SEGSZ;
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
mr = kzalloc(sizeof *mr + m * sizeof mr->mr.map[0], GFP_KERNEL);
if (!mr)
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
goto bail;
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
rval = init_qib_mregion(&mr->mr, pd, count);
if (rval)
goto bail;
/*
* ib_reg_phys_mr() will initialize mr->ibmr except for
* lkey and rkey.
*/
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
rval = qib_alloc_lkey(&mr->mr, 0);
if (rval)
goto bail_mregion;
mr->ibmr.lkey = mr->mr.lkey;
mr->ibmr.rkey = mr->mr.lkey;
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
done:
return mr;
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
bail_mregion:
deinit_qib_mregion(&mr->mr);
bail:
kfree(mr);
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
mr = ERR_PTR(rval);
goto done;
}
/**
* qib_reg_phys_mr - register a physical memory region
* @pd: protection domain for this memory region
* @buffer_list: pointer to the list of physical buffers to register
* @num_phys_buf: the number of physical buffers to register
* @iova_start: the starting address passed over IB which maps to this MR
*
* Returns the memory region on success, otherwise returns an errno.
*/
struct ib_mr *qib_reg_phys_mr(struct ib_pd *pd,
struct ib_phys_buf *buffer_list,
int num_phys_buf, int acc, u64 *iova_start)
{
struct qib_mr *mr;
int n, m, i;
struct ib_mr *ret;
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
mr = alloc_mr(num_phys_buf, pd);
if (IS_ERR(mr)) {
ret = (struct ib_mr *)mr;
goto bail;
}
mr->mr.user_base = *iova_start;
mr->mr.iova = *iova_start;
mr->mr.access_flags = acc;
m = 0;
n = 0;
for (i = 0; i < num_phys_buf; i++) {
mr->mr.map[m]->segs[n].vaddr = (void *) buffer_list[i].addr;
mr->mr.map[m]->segs[n].length = buffer_list[i].size;
mr->mr.length += buffer_list[i].size;
n++;
if (n == QIB_SEGSZ) {
m++;
n = 0;
}
}
ret = &mr->ibmr;
bail:
return ret;
}
/**
* qib_reg_user_mr - register a userspace memory region
* @pd: protection domain for this memory region
* @start: starting userspace address
* @length: length of region to register
* @mr_access_flags: access flags for this memory region
* @udata: unused by the QLogic_IB driver
*
* Returns the memory region on success, otherwise returns an errno.
*/
struct ib_mr *qib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
u64 virt_addr, int mr_access_flags,
struct ib_udata *udata)
{
struct qib_mr *mr;
struct ib_umem *umem;
struct ib_umem_chunk *chunk;
int n, m, i;
struct ib_mr *ret;
if (length == 0) {
ret = ERR_PTR(-EINVAL);
goto bail;
}
umem = ib_umem_get(pd->uobject->context, start, length,
mr_access_flags, 0);
if (IS_ERR(umem))
return (void *) umem;
n = 0;
list_for_each_entry(chunk, &umem->chunk_list, list)
n += chunk->nents;
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
mr = alloc_mr(n, pd);
if (IS_ERR(mr)) {
ret = (struct ib_mr *)mr;
ib_umem_release(umem);
goto bail;
}
mr->mr.user_base = start;
mr->mr.iova = virt_addr;
mr->mr.length = length;
mr->mr.offset = umem->offset;
mr->mr.access_flags = mr_access_flags;
mr->umem = umem;
if (is_power_of_2(umem->page_size))
mr->mr.page_shift = ilog2(umem->page_size);
m = 0;
n = 0;
list_for_each_entry(chunk, &umem->chunk_list, list) {
for (i = 0; i < chunk->nents; i++) {
void *vaddr;
vaddr = page_address(sg_page(&chunk->page_list[i]));
if (!vaddr) {
ret = ERR_PTR(-EINVAL);
goto bail;
}
mr->mr.map[m]->segs[n].vaddr = vaddr;
mr->mr.map[m]->segs[n].length = umem->page_size;
n++;
if (n == QIB_SEGSZ) {
m++;
n = 0;
}
}
}
ret = &mr->ibmr;
bail:
return ret;
}
/**
* qib_dereg_mr - unregister and free a memory region
* @ibmr: the memory region to free
*
* Returns 0 on success.
*
* Note that this is called to free MRs created by qib_get_dma_mr()
* or qib_reg_user_mr().
*/
int qib_dereg_mr(struct ib_mr *ibmr)
{
struct qib_mr *mr = to_imr(ibmr);
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
int ret = 0;
unsigned long timeout;
qib_free_lkey(&mr->mr);
qib_put_mr(&mr->mr); /* will set completion if last */
timeout = wait_for_completion_timeout(&mr->mr.comp,
5 * HZ);
if (!timeout) {
qib_get_mr(&mr->mr);
ret = -EBUSY;
goto out;
}
deinit_qib_mregion(&mr->mr);
if (mr->umem)
ib_umem_release(mr->umem);
kfree(mr);
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
out:
return ret;
}
/*
* Allocate a memory region usable with the
* IB_WR_FAST_REG_MR send work request.
*
* Return the memory region on success, otherwise return an errno.
*/
struct ib_mr *qib_alloc_fast_reg_mr(struct ib_pd *pd, int max_page_list_len)
{
struct qib_mr *mr;
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
mr = alloc_mr(max_page_list_len, pd);
if (IS_ERR(mr))
return (struct ib_mr *)mr;
return &mr->ibmr;
}
struct ib_fast_reg_page_list *
qib_alloc_fast_reg_page_list(struct ib_device *ibdev, int page_list_len)
{
unsigned size = page_list_len * sizeof(u64);
struct ib_fast_reg_page_list *pl;
if (size > PAGE_SIZE)
return ERR_PTR(-EINVAL);
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
pl = kzalloc(sizeof *pl, GFP_KERNEL);
if (!pl)
return ERR_PTR(-ENOMEM);
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
pl->page_list = kzalloc(size, GFP_KERNEL);
if (!pl->page_list)
goto err_free;
return pl;
err_free:
kfree(pl);
return ERR_PTR(-ENOMEM);
}
void qib_free_fast_reg_page_list(struct ib_fast_reg_page_list *pl)
{
kfree(pl->page_list);
kfree(pl);
}
/**
* qib_alloc_fmr - allocate a fast memory region
* @pd: the protection domain for this memory region
* @mr_access_flags: access flags for this memory region
* @fmr_attr: fast memory region attributes
*
* Returns the memory region on success, otherwise returns an errno.
*/
struct ib_fmr *qib_alloc_fmr(struct ib_pd *pd, int mr_access_flags,
struct ib_fmr_attr *fmr_attr)
{
struct qib_fmr *fmr;
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
int m;
struct ib_fmr *ret;
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
int rval = -ENOMEM;
/* Allocate struct plus pointers to first level page tables. */
m = (fmr_attr->max_pages + QIB_SEGSZ - 1) / QIB_SEGSZ;
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
fmr = kzalloc(sizeof *fmr + m * sizeof fmr->mr.map[0], GFP_KERNEL);
if (!fmr)
goto bail;
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
rval = init_qib_mregion(&fmr->mr, pd, fmr_attr->max_pages);
if (rval)
goto bail;
/*
* ib_alloc_fmr() will initialize fmr->ibfmr except for lkey &
* rkey.
*/
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
rval = qib_alloc_lkey(&fmr->mr, 0);
if (rval)
goto bail_mregion;
fmr->ibfmr.rkey = fmr->mr.lkey;
fmr->ibfmr.lkey = fmr->mr.lkey;
/*
* Resources are allocated but no valid mapping (RKEY can't be
* used).
*/
fmr->mr.access_flags = mr_access_flags;
fmr->mr.max_segs = fmr_attr->max_pages;
fmr->mr.page_shift = fmr_attr->page_shift;
ret = &fmr->ibfmr;
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
done:
return ret;
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
bail_mregion:
deinit_qib_mregion(&fmr->mr);
bail:
kfree(fmr);
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
ret = ERR_PTR(rval);
goto done;
}
/**
* qib_map_phys_fmr - set up a fast memory region
* @ibmfr: the fast memory region to set up
* @page_list: the list of pages to associate with the fast memory region
* @list_len: the number of pages to associate with the fast memory region
* @iova: the virtual address of the start of the fast memory region
*
* This may be called from interrupt context.
*/
int qib_map_phys_fmr(struct ib_fmr *ibfmr, u64 *page_list,
int list_len, u64 iova)
{
struct qib_fmr *fmr = to_ifmr(ibfmr);
struct qib_lkey_table *rkt;
unsigned long flags;
int m, n, i;
u32 ps;
int ret;
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
i = atomic_read(&fmr->mr.refcount);
if (i > 2)
return -EBUSY;
if (list_len > fmr->mr.max_segs) {
ret = -EINVAL;
goto bail;
}
rkt = &to_idev(ibfmr->device)->lk_table;
spin_lock_irqsave(&rkt->lock, flags);
fmr->mr.user_base = iova;
fmr->mr.iova = iova;
ps = 1 << fmr->mr.page_shift;
fmr->mr.length = list_len * ps;
m = 0;
n = 0;
for (i = 0; i < list_len; i++) {
fmr->mr.map[m]->segs[n].vaddr = (void *) page_list[i];
fmr->mr.map[m]->segs[n].length = ps;
if (++n == QIB_SEGSZ) {
m++;
n = 0;
}
}
spin_unlock_irqrestore(&rkt->lock, flags);
ret = 0;
bail:
return ret;
}
/**
* qib_unmap_fmr - unmap fast memory regions
* @fmr_list: the list of fast memory regions to unmap
*
* Returns 0 on success.
*/
int qib_unmap_fmr(struct list_head *fmr_list)
{
struct qib_fmr *fmr;
struct qib_lkey_table *rkt;
unsigned long flags;
list_for_each_entry(fmr, fmr_list, ibfmr.list) {
rkt = &to_idev(fmr->ibfmr.device)->lk_table;
spin_lock_irqsave(&rkt->lock, flags);
fmr->mr.user_base = 0;
fmr->mr.iova = 0;
fmr->mr.length = 0;
spin_unlock_irqrestore(&rkt->lock, flags);
}
return 0;
}
/**
* qib_dealloc_fmr - deallocate a fast memory region
* @ibfmr: the fast memory region to deallocate
*
* Returns 0 on success.
*/
int qib_dealloc_fmr(struct ib_fmr *ibfmr)
{
struct qib_fmr *fmr = to_ifmr(ibfmr);
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
int ret = 0;
unsigned long timeout;
qib_free_lkey(&fmr->mr);
qib_put_mr(&fmr->mr); /* will set completion if last */
timeout = wait_for_completion_timeout(&fmr->mr.comp,
5 * HZ);
if (!timeout) {
qib_get_mr(&fmr->mr);
ret = -EBUSY;
goto out;
}
deinit_qib_mregion(&fmr->mr);
kfree(fmr);
IB/qib: Avoid returning EBUSY from MR deregister A timing issue can occur where qib_mr_dereg can return -EBUSY if the MR use count is not zero. This can occur if the MR is de-registered while RDMA read response packets are being progressed from the SDMA ring. The suspicion is that the peer sent an RDMA read request, which has already been copied across to the peer. The peer sees the completion of his request and then communicates to the responder that the MR is not needed any longer. The responder tries to de-register the MR, catching some responses remaining in the SDMA ring holding the MR use count. The code now uses a get/put paradigm to track MR use counts and coordinates with the MR de-registration process using a completion when the count has reached zero. A timeout on the delay is in place to catch other EBUSY issues. The reference count protocol is as follows: - The return to the user counts as 1 - A reference from the lk_table or the qib_ibdev counts as 1. - Transient I/O operations increase/decrease as necessary A lot of code duplication has been folded into the new routines init_qib_mregion() and deinit_qib_mregion(). Additionally, explicit initialization of fields to zero is now handled by kzalloc(). Also, duplicated code 'while.*num_sge' that decrements reference counts have been consolidated in qib_put_ss(). Reviewed-by: Ramkrishna Vepa <ramkrishna.vepa@intel.com> Signed-off-by: Mike Marciniszyn <mike.marciniszyn@intel.com> Signed-off-by: Roland Dreier <roland@purestorage.com>
2012-06-28 06:33:12 +08:00
out:
return ret;
}
void mr_rcu_callback(struct rcu_head *list)
{
struct qib_mregion *mr = container_of(list, struct qib_mregion, list);
complete(&mr->comp);
}