linux-nfs.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Chuck Lever <chuck.lever@oracle.com>
To: linux-rdma@vger.kernel.org, linux-nfs@vger.kernel.org
Subject: [PATCH v1 09/20] xprtrdma: Limit the number of rpcrdma_mws
Date: Tue, 07 Jun 2016 15:47:32 -0400	[thread overview]
Message-ID: <20160607194732.18401.71941.stgit@manet.1015granger.net> (raw)
In-Reply-To: <20160607194001.18401.88592.stgit@manet.1015granger.net>

The fixed part of an R_key is 24 bits, but the variant part is only
8 bits, allowing 256 unique R_keys concurrenly in flight on one QP.

Thus an rpcrdma_xprt cannot use more than 256 rpcrdma_mws at a time.

Capping the number of rpcrdma_mws should prevent ro_map from
re-using an R_key, which could change the MR co-ordinates or access
settings while an RPC is still in progress. It also reduces the
number of pre-allocated FRMRs per transport, allowing more
transports per device.

To get more R_keys in flight at once, additional QPs will be
necessary.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
---
 net/sunrpc/xprtrdma/fmr_ops.c   |    6 +-----
 net/sunrpc/xprtrdma/frwr_ops.c  |    6 +-----
 net/sunrpc/xprtrdma/xprt_rdma.h |    8 ++++++++
 3 files changed, 10 insertions(+), 10 deletions(-)

diff --git a/net/sunrpc/xprtrdma/fmr_ops.c b/net/sunrpc/xprtrdma/fmr_ops.c
index 4632d25..33c4661 100644
--- a/net/sunrpc/xprtrdma/fmr_ops.c
+++ b/net/sunrpc/xprtrdma/fmr_ops.c
@@ -127,12 +127,8 @@ fmr_op_init(struct rpcrdma_xprt *r_xprt)
 	struct rpcrdma_mw *r;
 	int i, rc;
 
-	i = max_t(int, RPCRDMA_MAX_DATA_SEGS / RPCRDMA_MAX_FMR_SGES, 1);
-	i += 2;				/* head + tail */
-	i *= buf->rb_max_requests;	/* one set for each RPC slot */
-	dprintk("RPC:       %s: initalizing %d FMRs\n", __func__, i);
-
 	rc = -ENOMEM;
+	i = RPCRDMA_RKEYS_PER_QP;
 	while (i--) {
 		r = kzalloc(sizeof(*r), GFP_KERNEL);
 		if (!r)
diff --git a/net/sunrpc/xprtrdma/frwr_ops.c b/net/sunrpc/xprtrdma/frwr_ops.c
index e0f610c..04b8ab2e 100644
--- a/net/sunrpc/xprtrdma/frwr_ops.c
+++ b/net/sunrpc/xprtrdma/frwr_ops.c
@@ -317,11 +317,7 @@ frwr_op_init(struct rpcrdma_xprt *r_xprt)
 	struct ib_pd *pd = r_xprt->rx_ia.ri_pd;
 	int i;
 
-	i = max_t(int, RPCRDMA_MAX_DATA_SEGS / depth, 1);
-	i += 2;				/* head + tail */
-	i *= buf->rb_max_requests;	/* one set for each RPC slot */
-	dprintk("RPC:       %s: initalizing %d FRMRs\n", __func__, i);
-
+	i = RPCRDMA_RKEYS_PER_QP;
 	while (i--) {
 		struct rpcrdma_mw *r;
 		int rc;
diff --git a/net/sunrpc/xprtrdma/xprt_rdma.h b/net/sunrpc/xprtrdma/xprt_rdma.h
index ea8d4f8..8e53057 100644
--- a/net/sunrpc/xprtrdma/xprt_rdma.h
+++ b/net/sunrpc/xprtrdma/xprt_rdma.h
@@ -164,6 +164,14 @@ rdmab_to_msg(struct rpcrdma_regbuf *rb)
  */
 #define RPCRDMA_MAX_HDR_SEGS	(8)
 
+/* An R_key is 32 bits wide. 24 bits are fixed per QP, and
+ * 8 bits vary for each registered MR. Thus only 256 R_keys
+ * can be registered at one time per QP.
+ */
+enum {
+	RPCRDMA_RKEYS_PER_QP	= 256
+};
+
 /*
  * struct rpcrdma_rep -- this structure encapsulates state required to recv
  * and complete a reply, asychronously. It needs several pieces of


  parent reply	other threads:[~2016-06-07 19:47 UTC|newest]

Thread overview: 33+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2016-06-07 19:46 [PATCH v1 00/20] NFS/RDMA client patches proposed for v4.8 Chuck Lever
2016-06-07 19:46 ` [PATCH v1 01/20] xprtrdma: Remove ALLPHYSICAL memory registration mode Chuck Lever
2016-06-07 19:46 ` [PATCH v1 02/20] xprtrdma: Refactor ->ro_init Chuck Lever
2016-06-08 17:48   ` Anna Schumaker
2016-06-07 19:46 ` [PATCH v1 03/20] xprtrdma: Create common scatterlist fields in rpcrdma_mw Chuck Lever
2016-06-07 19:46 ` [PATCH v1 04/20] xprtrdma: Use scatterlist for DMA mapping and unmapping under FMR Chuck Lever
2016-06-07 19:46 ` [PATCH v1 05/20] xprtrdma: Remove rpcrdma_map_one() and friends Chuck Lever
2016-06-07 19:47 ` [PATCH v1 06/20] xprtrdma: Refactor MR recovery work queues Chuck Lever
2016-06-07 19:47 ` [PATCH v1 07/20] xprtrdma: Place registered MWs on a per-req list Chuck Lever
2016-06-07 19:47 ` [PATCH v1 08/20] xprtrdma: Reply buffer exhaustion can be catastrophic Chuck Lever
2016-06-07 19:47 ` Chuck Lever [this message]
2016-06-07 20:49   ` [PATCH v1 09/20] xprtrdma: Limit the number of rpcrdma_mws Jason Gunthorpe
2016-06-07 21:09     ` Chuck Lever
2016-06-07 21:28       ` Jason Gunthorpe
2016-06-07 21:51         ` Chuck Lever
2016-06-07 22:01           ` Jason Gunthorpe
2016-06-08 14:54             ` Tom Talpey
2016-06-08 15:06               ` Trond Myklebust
2016-06-08 17:40                 ` Jason Gunthorpe
2016-06-08 17:50                   ` Trond Myklebust
2016-06-08 17:53                   ` Chuck Lever
2016-06-08 18:45                     ` Tom Talpey
2016-06-07 19:47 ` [PATCH v1 10/20] xprtrdma: Chunk list encoders no longer share one rl_segments array Chuck Lever
2016-06-07 19:47 ` [PATCH v1 11/20] xprtrdma: rpcrdma_inline_fixup() overruns the receive page list Chuck Lever
2016-06-07 19:47 ` [PATCH v1 12/20] xprtrdma: Do not update {head, tail}.iov_len in rpcrdma_inline_fixup() Chuck Lever
2016-06-07 19:48 ` [PATCH v1 13/20] xprtrdma: Update only specific fields in private receive buffer Chuck Lever
2016-06-07 19:48 ` [PATCH v1 14/20] xprtrdma: Clean up fixup_copy_count accounting Chuck Lever
2016-06-07 19:48 ` [PATCH v1 15/20] xprtrdma: No direct data placement with krb5i and krb5p Chuck Lever
2016-06-07 19:48 ` [PATCH v1 16/20] svc: Avoid garbage replies when pc_func() returns rpc_drop_reply Chuck Lever
2016-06-07 19:48 ` [PATCH v1 17/20] NFS: Don't drop CB requests with invalid principals Chuck Lever
2016-06-07 19:48 ` [PATCH v1 18/20] xprtrdma: Eliminate rpcrdma_receive_worker() Chuck Lever
2016-06-07 19:48 ` [PATCH v1 19/20] xprtrdma: Eliminate INLINE_THRESHOLD macros Chuck Lever
2016-06-07 19:49 ` [PATCH v1 20/20] xprtrdma: Relocate connection helper functions Chuck Lever

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=20160607194732.18401.71941.stgit@manet.1015granger.net \
    --to=chuck.lever@oracle.com \
    --cc=linux-nfs@vger.kernel.org \
    --cc=linux-rdma@vger.kernel.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).