From: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
To: dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org
Cc: linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org,
Mike Marciniszyn
<mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Subject: [PATCH 18/20] IB/hfi1: Eliminate synchronize_rcu() in mr delete
Date: Wed, 01 Mar 2017 10:24:09 -0800 [thread overview]
Message-ID: <20170301182404.29989.63441.stgit@scvm10.sc.intel.com> (raw)
In-Reply-To: <20170301181719.29989.31238.stgit-9QXIwq+3FY+1XWohqUldA0EOCMrvLtNR@public.gmane.org>
From: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
The synchronize_rcu() call can be eliminated to improve memory deregistration
performance.
There are two key fields involved:
- The rcu pointer itself
- the lkey_published field
To close the window between the rcu read of the mregion pointer and the
reference count the code should:
1. To lkey/rkey validation (reader)
Read the rcu pointer. If the pointer is non-NULL, get a reference.
To the current validation tests use a READ_ONCE() on the lkey_published.
Upon any failure release the reference.
2. To the remove logic (delete)
Insure the published is zeroed prior to setting the pointer to NULL.
This requires using rcu_assign_pointer() to insure lkey_published
is written prior to the NULL.
3. To the insert logic (add)
Insure the published is set use an rcu_assign_pointer() to insure the
pointer is after all MR fields.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
---
drivers/infiniband/sw/rdmavt/mr.c | 49 +++++++++++++++++++++++++------------
1 files changed, 33 insertions(+), 16 deletions(-)
diff --git a/drivers/infiniband/sw/rdmavt/mr.c b/drivers/infiniband/sw/rdmavt/mr.c
index ae30b68..7c86955 100644
--- a/drivers/infiniband/sw/rdmavt/mr.c
+++ b/drivers/infiniband/sw/rdmavt/mr.c
@@ -191,8 +191,9 @@ static int rvt_alloc_lkey(struct rvt_mregion *mr, int dma_region)
tmr = rcu_access_pointer(dev->dma_mr);
if (!tmr) {
- rcu_assign_pointer(dev->dma_mr, mr);
mr->lkey_published = 1;
+ /* Insure published written first */
+ rcu_assign_pointer(dev->dma_mr, mr);
rvt_get_mr(mr);
}
goto success;
@@ -224,8 +225,9 @@ static int rvt_alloc_lkey(struct rvt_mregion *mr, int dma_region)
mr->lkey |= 1 << 8;
rkt->gen++;
}
- rcu_assign_pointer(rkt->table[r], mr);
mr->lkey_published = 1;
+ /* Insure published written first */
+ rcu_assign_pointer(rkt->table[r], mr);
success:
spin_unlock_irqrestore(&rkt->lock, flags);
out:
@@ -253,23 +255,24 @@ static void rvt_free_lkey(struct rvt_mregion *mr)
spin_lock_irqsave(&rkt->lock, flags);
if (!lkey) {
if (mr->lkey_published) {
- RCU_INIT_POINTER(dev->dma_mr, NULL);
+ mr->lkey_published = 0;
+ /* insure published is written before pointer */
+ rcu_assign_pointer(dev->dma_mr, NULL);
rvt_put_mr(mr);
}
} else {
if (!mr->lkey_published)
goto out;
r = lkey >> (32 - dev->dparms.lkey_table_size);
- RCU_INIT_POINTER(rkt->table[r], NULL);
+ mr->lkey_published = 0;
+ /* insure published is written before pointer */
+ rcu_assign_pointer(rkt->table[r], NULL);
}
- mr->lkey_published = 0;
freed++;
out:
spin_unlock_irqrestore(&rkt->lock, flags);
- if (freed) {
- synchronize_rcu();
+ if (freed)
percpu_ref_kill(&mr->refcount);
- }
}
static struct rvt_mr *__rvt_alloc_mr(int count, struct ib_pd *pd)
@@ -822,16 +825,21 @@ int rvt_lkey_ok(struct rvt_lkey_table *rkt, struct rvt_pd *pd,
goto ok;
}
mr = rcu_dereference(rkt->table[sge->lkey >> rkt->shift]);
- if (unlikely(!mr || atomic_read(&mr->lkey_invalid) ||
- mr->lkey != sge->lkey || mr->pd != &pd->ibpd))
+ if (!mr)
goto bail;
+ rvt_get_mr(mr);
+ if (!READ_ONCE(mr->lkey_published))
+ goto bail_unref;
+
+ if (unlikely(atomic_read(&mr->lkey_invalid) ||
+ mr->lkey != sge->lkey || mr->pd != &pd->ibpd))
+ goto bail_unref;
off = sge->addr - mr->user_base;
if (unlikely(sge->addr < mr->user_base ||
off + sge->length > mr->length ||
(mr->access_flags & acc) != acc))
- goto bail;
- rvt_get_mr(mr);
+ goto bail_unref;
rcu_read_unlock();
off += mr->offset;
@@ -867,6 +875,8 @@ int rvt_lkey_ok(struct rvt_lkey_table *rkt, struct rvt_pd *pd,
isge->n = n;
ok:
return 1;
+bail_unref:
+ rvt_put_mr(mr);
bail:
rcu_read_unlock();
return 0;
@@ -922,15 +932,20 @@ int rvt_rkey_ok(struct rvt_qp *qp, struct rvt_sge *sge,
}
mr = rcu_dereference(rkt->table[rkey >> rkt->shift]);
- if (unlikely(!mr || atomic_read(&mr->lkey_invalid) ||
- mr->lkey != rkey || qp->ibqp.pd != mr->pd))
+ if (!mr)
goto bail;
+ rvt_get_mr(mr);
+ /* insure mr read is before test */
+ if (!READ_ONCE(mr->lkey_published))
+ goto bail_unref;
+ if (unlikely(atomic_read(&mr->lkey_invalid) ||
+ mr->lkey != rkey || qp->ibqp.pd != mr->pd))
+ goto bail_unref;
off = vaddr - mr->iova;
if (unlikely(vaddr < mr->iova || off + len > mr->length ||
(mr->access_flags & acc) == 0))
- goto bail;
- rvt_get_mr(mr);
+ goto bail_unref;
rcu_read_unlock();
off += mr->offset;
@@ -966,6 +981,8 @@ int rvt_rkey_ok(struct rvt_qp *qp, struct rvt_sge *sge,
sge->n = n;
ok:
return 1;
+bail_unref:
+ rvt_put_mr(mr);
bail:
rcu_read_unlock();
return 0;
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
next prev parent reply other threads:[~2017-03-01 18:24 UTC|newest]
Thread overview: 29+ messages / expand[flat|nested] mbox.gz Atom feed top
2017-03-01 18:21 [PATCH 00/20] IB/hfi1, qib, rdmavt: Another round of patches for 4.11 Dennis Dalessandro
[not found] ` <20170301181719.29989.31238.stgit-9QXIwq+3FY+1XWohqUldA0EOCMrvLtNR@public.gmane.org>
2017-03-01 18:21 ` [PATCH 01/20] IB/hfi1: Force logical link down Dennis Dalessandro
2017-03-01 18:21 ` [PATCH 02/20] IB/hfi1: Race hazard avoidance in user SDMA driver Dennis Dalessandro
2017-03-01 18:21 ` [PATCH 03/20] IB/hfi1: Cache registers during state change Dennis Dalessandro
2017-03-01 18:21 ` [PATCH 04/20] IB/hfi1: NULL pointer dereference when freeing rhashtable Dennis Dalessandro
2017-03-01 18:22 ` [PATCH 05/20] IB/rdmavt, IB/hfi1, IB/qib: Make wc opcode translation driver dependent Dennis Dalessandro
2017-03-01 18:22 ` [PATCH 06/20] IB/rdmavt: Add additional fields to post send trace Dennis Dalessandro
2017-03-01 18:22 ` [PATCH 07/20] IB/rdmavt: Add tracing for cq entry and poll Dennis Dalessandro
2017-03-01 18:22 ` [PATCH 08/20] IB/rdmavt: Add swqe completion trace Dennis Dalessandro
2017-03-01 18:22 ` [PATCH 09/20] IB/hfi1: Check device id early during init Dennis Dalessandro
2017-03-01 18:22 ` [PATCH 10/20] IB/hfi1: Protect the global dev_cntr_names and port_cntr_names Dennis Dalessandro
2017-03-01 18:22 ` [PATCH 11/20] IB/hfi1: Check for QSFP presence before attempting reads Dennis Dalessandro
2017-03-01 18:23 ` [PATCH 12/20] IB/hfi1: Add a patch value to the firmware version string Dennis Dalessandro
2017-03-01 18:23 ` [PATCH 13/20] IB/rdmavt,IB/hfi1: Fix timer migration regressions Dennis Dalessandro
2017-03-01 18:23 ` [PATCH 14/20] IB/rdmavt: Avoid reseting wqe send_flags in unreserve Dennis Dalessandro
2017-03-01 18:23 ` [PATCH 15/20] IB/hfi1: Ensure VL index is within bounds Dennis Dalessandro
2017-03-01 18:23 ` [PATCH 16/20] IB/hfi1: Add receive fault injection feature Dennis Dalessandro
[not found] ` <20170301182344.29989.12032.stgit-9QXIwq+3FY+1XWohqUldA0EOCMrvLtNR@public.gmane.org>
2017-03-02 17:19 ` Dennis Dalessandro
2017-03-01 18:23 ` [PATCH 17/20] IB/hfi1: Add transmit " Dennis Dalessandro
2017-03-01 18:24 ` Dennis Dalessandro [this message]
2017-03-01 18:24 ` [PATCH 19/20] IB/rdmavt, IB/qib, IB/hfi1: Make percpu refcount optional for user MRs Dennis Dalessandro
2017-03-01 18:24 ` [PATCH 20/20] IB/core: If the MGID/MLID pair is not on the list return an error Dennis Dalessandro
[not found] ` <20170301182426.29989.77369.stgit-9QXIwq+3FY+1XWohqUldA0EOCMrvLtNR@public.gmane.org>
2017-03-02 14:00 ` Leon Romanovsky
[not found] ` <20170302140028.GE9525-U/DQcQFIOTAAJjI8aNfphQ@public.gmane.org>
2017-03-02 15:01 ` Dennis Dalessandro
[not found] ` <4d49f7ab-5bc5-029c-4321-fd349c9dc0f8-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
2017-03-02 15:13 ` Leon Romanovsky
2017-03-02 15:18 ` [PATCH v2 " Dennis Dalessandro
2017-03-02 18:26 ` [PATCH v2 16/20] IB/hfi1: Add receive fault injection feature Dennis Dalessandro
[not found] ` <20170302182610.28851.42687.stgit-9QXIwq+3FY+1XWohqUldA0EOCMrvLtNR@public.gmane.org>
2017-03-03 16:15 ` Leon Romanovsky
[not found] ` <20170303161508.GF14379-U/DQcQFIOTAAJjI8aNfphQ@public.gmane.org>
2017-03-03 17:02 ` Dennis Dalessandro
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20170301182404.29989.63441.stgit@scvm10.sc.intel.com \
--to=dennis.dalessandro-ral2jqcrhueavxtiumwx3w@public.gmane.org \
--cc=dledford-H+wXaHxf7aLQT0dZR+AlfA@public.gmane.org \
--cc=linux-rdma-u79uwXL29TY76Z2rM5mHXA@public.gmane.org \
--cc=mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).