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>,
Sebastian Sanchez
<sebastian.sanchez-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Subject: [PATCH 3/6] IB/hfi1: Optimize lkey validation structures
Date: Mon, 10 Oct 2016 06:14:39 -0700 [thread overview]
Message-ID: <20161010131439.17586.29496.stgit@scvm10.sc.intel.com> (raw)
In-Reply-To: <20161010131249.17586.23789.stgit-9QXIwq+3FY+1XWohqUldA0EOCMrvLtNR@public.gmane.org>
From: Mike Marciniszyn <mike.marciniszyn-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
Profiling shows that the key validation is susceptible
to cache line trading when accessing the lkey table.
Fix by separating out the read mostly fields from the write
fields. In addition the shift amount, which is function
of the lkey table size, is precomputed and stored with the
table pointer. Since both the shift and table pointer
are in the same read mostly cacheline, this saves a cache
line in this hot path.
Reviewed-by: Sebastian Sanchez <sebastian.sanchez-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 | 10 +++++-----
include/rdma/rdmavt_mr.h | 10 +++++++---
2 files changed, 12 insertions(+), 8 deletions(-)
diff --git a/drivers/infiniband/sw/rdmavt/mr.c b/drivers/infiniband/sw/rdmavt/mr.c
index 46b6497..4acf179 100644
--- a/drivers/infiniband/sw/rdmavt/mr.c
+++ b/drivers/infiniband/sw/rdmavt/mr.c
@@ -84,6 +84,7 @@ int rvt_driver_mr_init(struct rvt_dev_info *rdi)
lkey_table_size = rdi->dparms.lkey_table_size;
}
rdi->lkey_table.max = 1 << lkey_table_size;
+ rdi->lkey_table.shift = 32 - lkey_table_size;
lk_tab_size = rdi->lkey_table.max * sizeof(*rdi->lkey_table.table);
rdi->lkey_table.table = (struct rvt_mregion __rcu **)
vmalloc_node(lk_tab_size, rdi->dparms.node);
@@ -774,7 +775,6 @@ int rvt_lkey_ok(struct rvt_lkey_table *rkt, struct rvt_pd *pd,
struct rvt_mregion *mr;
unsigned n, m;
size_t off;
- struct rvt_dev_info *dev = ib_to_rvt(pd->ibpd.device);
/*
* We use LKEY == zero for kernel virtual addresses
@@ -782,6 +782,8 @@ int rvt_lkey_ok(struct rvt_lkey_table *rkt, struct rvt_pd *pd,
*/
rcu_read_lock();
if (sge->lkey == 0) {
+ struct rvt_dev_info *dev = ib_to_rvt(pd->ibpd.device);
+
if (pd->user)
goto bail;
mr = rcu_dereference(dev->dma_mr);
@@ -798,8 +800,7 @@ int rvt_lkey_ok(struct rvt_lkey_table *rkt, struct rvt_pd *pd,
isge->n = 0;
goto ok;
}
- mr = rcu_dereference(
- rkt->table[(sge->lkey >> (32 - dev->dparms.lkey_table_size))]);
+ 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))
goto bail;
@@ -899,8 +900,7 @@ int rvt_rkey_ok(struct rvt_qp *qp, struct rvt_sge *sge,
goto ok;
}
- mr = rcu_dereference(
- rkt->table[(rkey >> (32 - dev->dparms.lkey_table_size))]);
+ mr = rcu_dereference(rkt->table[rkey >> rkt->shift]);
if (unlikely(!mr || atomic_read(&mr->lkey_invalid) ||
mr->lkey != rkey || qp->ibqp.pd != mr->pd))
goto bail;
diff --git a/include/rdma/rdmavt_mr.h b/include/rdma/rdmavt_mr.h
index 6b3c6c8..de59de2 100644
--- a/include/rdma/rdmavt_mr.h
+++ b/include/rdma/rdmavt_mr.h
@@ -90,11 +90,15 @@ struct rvt_mregion {
#define RVT_MAX_LKEY_TABLE_BITS 23
struct rvt_lkey_table {
- spinlock_t lock; /* protect changes in this struct */
- u32 next; /* next unused index (speeds search) */
- u32 gen; /* generation count */
+ /* read mostly fields */
u32 max; /* size of the table */
+ u32 shift; /* lkey/rkey shift */
struct rvt_mregion __rcu **table;
+ /* writeable fields */
+ /* protect changes in this struct */
+ spinlock_t lock ____cacheline_aligned_in_smp;
+ u32 next; /* next unused index (speeds search) */
+ u32 gen; /* generation count */
};
/*
--
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:[~2016-10-10 13:14 UTC|newest]
Thread overview: 7+ messages / expand[flat|nested] mbox.gz Atom feed top
2016-10-10 13:14 [PATCH 0/6] IB/hfi1,rdmavt: Few more for 4.9 incl stable fix Dennis Dalessandro
[not found] ` <20161010131249.17586.23789.stgit-9QXIwq+3FY+1XWohqUldA0EOCMrvLtNR@public.gmane.org>
2016-10-10 13:14 ` [PATCH 1/6] IB/hfi1: Add unique txwait_lock for txreq events Dennis Dalessandro
2016-10-10 13:14 ` [PATCH 2/6] IB/hfi1: Inline sdma_txclean() for verbs pio Dennis Dalessandro
2016-10-10 13:14 ` Dennis Dalessandro [this message]
2016-10-10 13:14 ` [PATCH 4/6] IB/rdmavt: rdmavt can handle non aligned page maps Dennis Dalessandro
2016-10-10 13:14 ` [PATCH 5/6] IB/hfi1: Remove redundant sysfs irq affinity entry Dennis Dalessandro
2016-10-10 13:14 ` [PATCH 6/6] IB/hfi1: Fix integrity check flags default values 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=20161010131439.17586.29496.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 \
--cc=sebastian.sanchez-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).