[PATCH v5 00/10] NFS/RDMA server patches for v4.5

[PATCH v5 00/10] NFS/RDMA server patches for v4.5

[Chuck Lever]

Hi Doug-

Here are patches to support server-side bi-directional RPC/RDMA
operation (to enable NFSv4.1 on RPC/RDMA transports). Thanks to
all who reviewed earlier versions.

Doug, please take these for 4.5. These are Acked-by Bruce and ready
to go.

Also available in the "nfsd-rdma-for-4.5" topic branch of this git
repo:

git://git.linux-nfs.org/projects/cel/cel-2.6.git

Or for browsing:

http://git.linux-nfs.org/?p=cel/cel-2.6.git;a=log;h=refs/heads/nfsd-rdma-for-4.5


Changes since v4:
- Rebased on Doug's k.o/for-4.5 branch
- Dropped patch to prevent writing XDR pads
- Squashed in Dan Carpenter's fixes


Changes since v3:
- Rebased on Christoph's ib_device_attr branch
- Backchannel patches have been squashed together
- Memory allocation overhaul to prevent blocking allocation
   when sending backchannel calls


Changes since v2:
- Rebased on v4.4-rc4
- Backchannel code in new source file to address dprintk issues
- svc_rdma_get_context() now uses a pre-allocated cache
- Dropped svc_rdma_send clean up


Changes since v1:

- Rebased on v4.4-rc3
- Removed the use of CONFIG_SUNRPC_BACKCHANNEL
- Fixed computation of forward and backward max_requests
- Updated some comments and patch descriptions
- pr_err and pr_info converted to dprintk
- Simplified svc_rdma_get_context()
- Dropped patch removing access_flags field
- NFSv4.1 callbacks tested with for-4.5 client

---

Chuck Lever (10):
      svcrdma: Clean up rdma_create_xprt()
      svcrdma: Clean up process_context()
      svcrdma: Improve allocation of struct svc_rdma_op_ctxt
      svcrdma: Improve allocation of struct svc_rdma_req_map
      svcrdma: Remove unused req_map and ctxt kmem_caches
      svcrdma: Add gfp flags to svc_rdma_post_recv()
      svcrdma: Remove last two __GFP_NOFAIL call sites
      svcrdma: Make map_xdr non-static
      svcrdma: Define maximum number of backchannel requests
      svcrdma: Add class for RDMA backwards direction transport


 include/linux/sunrpc/svc_rdma.h            |   37 ++-
 net/sunrpc/xprt.c                          |    1 
 net/sunrpc/xprtrdma/Makefile               |    2 
 net/sunrpc/xprtrdma/svc_rdma.c             |   41 ---
 net/sunrpc/xprtrdma/svc_rdma_backchannel.c |  371 ++++++++++++++++++++++++++++
 net/sunrpc/xprtrdma/svc_rdma_recvfrom.c    |   52 ++++
 net/sunrpc/xprtrdma/svc_rdma_sendto.c      |   27 +-
 net/sunrpc/xprtrdma/svc_rdma_transport.c   |  284 ++++++++++++++++-----
 net/sunrpc/xprtrdma/transport.c            |   30 +-
 net/sunrpc/xprtrdma/xprt_rdma.h            |   20 +-
 10 files changed, 723 insertions(+), 142 deletions(-)
 create mode 100644 net/sunrpc/xprtrdma/svc_rdma_backchannel.c

--
Chuck Lever

[PATCH v5 01/10] svcrdma: Clean up rdma_create_xprt()

[Chuck Lever]

kzalloc is used here, so setting the atomic fields to zero is
unnecessary. sc_ord is set again in handle_connect_req. The other
fields are re-initialized in svc_rdma_accept().

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Acked-by: Bruce Fields <bfields@fieldses.org>
---
 net/sunrpc/xprtrdma/svc_rdma_transport.c |    9 +--------
 1 file changed, 1 insertion(+), 8 deletions(-)

diff --git a/net/sunrpc/xprtrdma/svc_rdma_transport.c b/net/sunrpc/xprtrdma/svc_rdma_transport.c
index 4d695af..45d1824 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_transport.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_transport.c
@@ -529,14 +529,6 @@ static struct svcxprt_rdma *rdma_create_xprt(struct svc_serv *serv,
 	spin_lock_init(&cma_xprt->sc_rq_dto_lock);
 	spin_lock_init(&cma_xprt->sc_frmr_q_lock);
 
-	cma_xprt->sc_ord = svcrdma_ord;
-
-	cma_xprt->sc_max_req_size = svcrdma_max_req_size;
-	cma_xprt->sc_max_requests = svcrdma_max_requests;
-	cma_xprt->sc_sq_depth = svcrdma_max_requests * RPCRDMA_SQ_DEPTH_MULT;
-	atomic_set(&cma_xprt->sc_sq_count, 0);
-	atomic_set(&cma_xprt->sc_ctxt_used, 0);
-
 	if (listener)
 		set_bit(XPT_LISTENER, &cma_xprt->sc_xprt.xpt_flags);
 
@@ -918,6 +910,7 @@ static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt)
 				  (size_t)RPCSVC_MAXPAGES);
 	newxprt->sc_max_sge_rd = min_t(size_t, dev->attrs.max_sge_rd,
 				       RPCSVC_MAXPAGES);
+	newxprt->sc_max_req_size = svcrdma_max_req_size;
 	newxprt->sc_max_requests = min((size_t)dev->attrs.max_qp_wr,
 				   (size_t)svcrdma_max_requests);
 	newxprt->sc_sq_depth = RPCRDMA_SQ_DEPTH_MULT * newxprt->sc_max_requests;

[PATCH v5 02/10] svcrdma: Clean up process_context()

[Chuck Lever]

Be sure the completed ctxt is put in every path.

The xprt enqueue can take a while, so put the completed ctxt back
in circulation _before_ enqueuing the xprt.

Remove/disable debugging.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Acked-by: Bruce Fields <bfields@fieldses.org>
---
 net/sunrpc/xprtrdma/svc_rdma_transport.c |   44 ++++++++++++++----------------
 1 file changed, 21 insertions(+), 23 deletions(-)

diff --git a/net/sunrpc/xprtrdma/svc_rdma_transport.c b/net/sunrpc/xprtrdma/svc_rdma_transport.c
index 45d1824..a100d56 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_transport.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_transport.c
@@ -386,46 +386,44 @@ static void rq_cq_reap(struct svcxprt_rdma *xprt)
 static void process_context(struct svcxprt_rdma *xprt,
 			    struct svc_rdma_op_ctxt *ctxt)
 {
+	struct svc_rdma_op_ctxt *read_hdr;
+	int free_pages = 0;
+
 	svc_rdma_unmap_dma(ctxt);
 
 	switch (ctxt->wr_op) {
 	case IB_WR_SEND:
-		if (ctxt->frmr)
-			pr_err("svcrdma: SEND: ctxt->frmr != NULL\n");
-		svc_rdma_put_context(ctxt, 1);
+		free_pages = 1;
 		break;
 
 	case IB_WR_RDMA_WRITE:
-		if (ctxt->frmr)
-			pr_err("svcrdma: WRITE: ctxt->frmr != NULL\n");
-		svc_rdma_put_context(ctxt, 0);
 		break;
 
 	case IB_WR_RDMA_READ:
 	case IB_WR_RDMA_READ_WITH_INV:
 		svc_rdma_put_frmr(xprt, ctxt->frmr);
-		if (test_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags)) {
-			struct svc_rdma_op_ctxt *read_hdr = ctxt->read_hdr;
-			if (read_hdr) {
-				spin_lock_bh(&xprt->sc_rq_dto_lock);
-				set_bit(XPT_DATA, &xprt->sc_xprt.xpt_flags);
-				list_add_tail(&read_hdr->dto_q,
-					      &xprt->sc_read_complete_q);
-				spin_unlock_bh(&xprt->sc_rq_dto_lock);
-			} else {
-				pr_err("svcrdma: ctxt->read_hdr == NULL\n");
-			}
-			svc_xprt_enqueue(&xprt->sc_xprt);
-		}
+
+		if (!test_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags))
+			break;
+
+		read_hdr = ctxt->read_hdr;
 		svc_rdma_put_context(ctxt, 0);
-		break;
+
+		spin_lock_bh(&xprt->sc_rq_dto_lock);
+		set_bit(XPT_DATA, &xprt->sc_xprt.xpt_flags);
+		list_add_tail(&read_hdr->dto_q,
+			      &xprt->sc_read_complete_q);
+		spin_unlock_bh(&xprt->sc_rq_dto_lock);
+		svc_xprt_enqueue(&xprt->sc_xprt);
+		return;
 
 	default:
-		printk(KERN_ERR "svcrdma: unexpected completion type, "
-		       "opcode=%d\n",
-		       ctxt->wr_op);
+		dprintk("svcrdma: unexpected completion opcode=%d\n",
+			ctxt->wr_op);
 		break;
 	}
+
+	svc_rdma_put_context(ctxt, free_pages);
 }
 
 /*

[PATCH v5 03/10] svcrdma: Improve allocation of struct svc_rdma_op_ctxt

[Chuck Lever]

When the maximum payload size of NFS READ and WRITE was increased
by commit cc9a903d915c ("svcrdma: Change maximum server payload back
to RPCSVC_MAXPAYLOAD"), the size of struct svc_rdma_op_ctxt
increased to over 6KB (on x86_64). That makes allocating one of
these from a kmem_cache more likely to fail in situations when
system memory is exhausted.

Since I'm about to add a caller where this allocation must always
work _and_ it cannot sleep, pre-allocate ctxts for each connection.

Another motivation for this change is that NFSv4.x servers are
required by specification not to drop NFS requests. Pre-allocating
memory resources reduces the likelihood of a drop.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Acked-by: Bruce Fields <bfields@fieldses.org>
---
 include/linux/sunrpc/svc_rdma.h          |    6 +-
 net/sunrpc/xprtrdma/svc_rdma_transport.c |  102 ++++++++++++++++++++++++++----
 2 files changed, 94 insertions(+), 14 deletions(-)

diff --git a/include/linux/sunrpc/svc_rdma.h b/include/linux/sunrpc/svc_rdma.h
index f869807..be2804b 100644
--- a/include/linux/sunrpc/svc_rdma.h
+++ b/include/linux/sunrpc/svc_rdma.h
@@ -69,6 +69,7 @@ extern atomic_t rdma_stat_sq_prod;
  * completes.
  */
 struct svc_rdma_op_ctxt {
+	struct list_head free;
 	struct svc_rdma_op_ctxt *read_hdr;
 	struct svc_rdma_fastreg_mr *frmr;
 	int hdr_count;
@@ -141,7 +142,10 @@ struct svcxprt_rdma {
 	struct ib_pd         *sc_pd;
 
 	atomic_t	     sc_dma_used;
-	atomic_t	     sc_ctxt_used;
+	spinlock_t	     sc_ctxt_lock;
+	struct list_head     sc_ctxts;
+	int		     sc_ctxt_used;
+
 	struct list_head     sc_rq_dto_q;
 	spinlock_t	     sc_rq_dto_lock;
 	struct ib_qp         *sc_qp;
diff --git a/net/sunrpc/xprtrdma/svc_rdma_transport.c b/net/sunrpc/xprtrdma/svc_rdma_transport.c
index a100d56..9801115 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_transport.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_transport.c
@@ -153,18 +153,76 @@ static void svc_rdma_bc_free(struct svc_xprt *xprt)
 }
 #endif	/* CONFIG_SUNRPC_BACKCHANNEL */
 
-struct svc_rdma_op_ctxt *svc_rdma_get_context(struct svcxprt_rdma *xprt)
+static struct svc_rdma_op_ctxt *alloc_ctxt(struct svcxprt_rdma *xprt,
+					   gfp_t flags)
 {
 	struct svc_rdma_op_ctxt *ctxt;
 
-	ctxt = kmem_cache_alloc(svc_rdma_ctxt_cachep,
-				GFP_KERNEL | __GFP_NOFAIL);
-	ctxt->xprt = xprt;
-	INIT_LIST_HEAD(&ctxt->dto_q);
+	ctxt = kmalloc(sizeof(*ctxt), flags);
+	if (ctxt) {
+		ctxt->xprt = xprt;
+		INIT_LIST_HEAD(&ctxt->free);
+		INIT_LIST_HEAD(&ctxt->dto_q);
+	}
+	return ctxt;
+}
+
+static bool svc_rdma_prealloc_ctxts(struct svcxprt_rdma *xprt)
+{
+	int i;
+
+	/* Each RPC/RDMA credit can consume a number of send
+	 * and receive WQEs. One ctxt is allocated for each.
+	 */
+	i = xprt->sc_sq_depth + xprt->sc_max_requests;
+
+	while (i--) {
+		struct svc_rdma_op_ctxt *ctxt;
+
+		ctxt = alloc_ctxt(xprt, GFP_KERNEL);
+		if (!ctxt) {
+			dprintk("svcrdma: No memory for RDMA ctxt\n");
+			return false;
+		}
+		list_add(&ctxt->free, &xprt->sc_ctxts);
+	}
+	return true;
+}
+
+struct svc_rdma_op_ctxt *svc_rdma_get_context(struct svcxprt_rdma *xprt)
+{
+	struct svc_rdma_op_ctxt *ctxt = NULL;
+
+	spin_lock_bh(&xprt->sc_ctxt_lock);
+	xprt->sc_ctxt_used++;
+	if (list_empty(&xprt->sc_ctxts))
+		goto out_empty;
+
+	ctxt = list_first_entry(&xprt->sc_ctxts,
+				struct svc_rdma_op_ctxt, free);
+	list_del_init(&ctxt->free);
+	spin_unlock_bh(&xprt->sc_ctxt_lock);
+
+out:
 	ctxt->count = 0;
 	ctxt->frmr = NULL;
-	atomic_inc(&xprt->sc_ctxt_used);
 	return ctxt;
+
+out_empty:
+	/* Either pre-allocation missed the mark, or send
+	 * queue accounting is broken.
+	 */
+	spin_unlock_bh(&xprt->sc_ctxt_lock);
+
+	ctxt = alloc_ctxt(xprt, GFP_NOIO);
+	if (ctxt)
+		goto out;
+
+	spin_lock_bh(&xprt->sc_ctxt_lock);
+	xprt->sc_ctxt_used--;
+	spin_unlock_bh(&xprt->sc_ctxt_lock);
+	WARN_ONCE(1, "svcrdma: empty RDMA ctxt list?\n");
+	return NULL;
 }
 
 void svc_rdma_unmap_dma(struct svc_rdma_op_ctxt *ctxt)
@@ -190,16 +248,29 @@ void svc_rdma_unmap_dma(struct svc_rdma_op_ctxt *ctxt)
 
 void svc_rdma_put_context(struct svc_rdma_op_ctxt *ctxt, int free_pages)
 {
-	struct svcxprt_rdma *xprt;
+	struct svcxprt_rdma *xprt = ctxt->xprt;
 	int i;
 
-	xprt = ctxt->xprt;
 	if (free_pages)
 		for (i = 0; i < ctxt->count; i++)
 			put_page(ctxt->pages[i]);
 
-	kmem_cache_free(svc_rdma_ctxt_cachep, ctxt);
-	atomic_dec(&xprt->sc_ctxt_used);
+	spin_lock_bh(&xprt->sc_ctxt_lock);
+	xprt->sc_ctxt_used--;
+	list_add(&ctxt->free, &xprt->sc_ctxts);
+	spin_unlock_bh(&xprt->sc_ctxt_lock);
+}
+
+static void svc_rdma_destroy_ctxts(struct svcxprt_rdma *xprt)
+{
+	while (!list_empty(&xprt->sc_ctxts)) {
+		struct svc_rdma_op_ctxt *ctxt;
+
+		ctxt = list_first_entry(&xprt->sc_ctxts,
+					struct svc_rdma_op_ctxt, free);
+		list_del(&ctxt->free);
+		kfree(ctxt);
+	}
 }
 
 /*
@@ -521,11 +592,13 @@ static struct svcxprt_rdma *rdma_create_xprt(struct svc_serv *serv,
 	INIT_LIST_HEAD(&cma_xprt->sc_rq_dto_q);
 	INIT_LIST_HEAD(&cma_xprt->sc_read_complete_q);
 	INIT_LIST_HEAD(&cma_xprt->sc_frmr_q);
+	INIT_LIST_HEAD(&cma_xprt->sc_ctxts);
 	init_waitqueue_head(&cma_xprt->sc_send_wait);
 
 	spin_lock_init(&cma_xprt->sc_lock);
 	spin_lock_init(&cma_xprt->sc_rq_dto_lock);
 	spin_lock_init(&cma_xprt->sc_frmr_q_lock);
+	spin_lock_init(&cma_xprt->sc_ctxt_lock);
 
 	if (listener)
 		set_bit(XPT_LISTENER, &cma_xprt->sc_xprt.xpt_flags);
@@ -913,6 +986,9 @@ static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt)
 				   (size_t)svcrdma_max_requests);
 	newxprt->sc_sq_depth = RPCRDMA_SQ_DEPTH_MULT * newxprt->sc_max_requests;
 
+	if (!svc_rdma_prealloc_ctxts(newxprt))
+		goto errout;
+
 	/*
 	 * Limit ORD based on client limit, local device limit, and
 	 * configured svcrdma limit.
@@ -1174,15 +1250,15 @@ static void __svc_rdma_free(struct work_struct *work)
 	}
 
 	/* Warn if we leaked a resource or under-referenced */
-	if (atomic_read(&rdma->sc_ctxt_used) != 0)
+	if (rdma->sc_ctxt_used != 0)
 		pr_err("svcrdma: ctxt still in use? (%d)\n",
-		       atomic_read(&rdma->sc_ctxt_used));
+		       rdma->sc_ctxt_used);
 	if (atomic_read(&rdma->sc_dma_used) != 0)
 		pr_err("svcrdma: dma still in use? (%d)\n",
 		       atomic_read(&rdma->sc_dma_used));
 
-	/* De-allocate fastreg mr */
 	rdma_dealloc_frmr_q(rdma);
+	svc_rdma_destroy_ctxts(rdma);
 
 	/* Destroy the QP if present (not a listener) */
 	if (rdma->sc_qp && !IS_ERR(rdma->sc_qp))

[PATCH v5 04/10] svcrdma: Improve allocation of struct svc_rdma_req_map

[Chuck Lever]

To ensure this allocation cannot fail and will not sleep,
pre-allocate the req_map structures per-connection.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Acked-by: Bruce Fields <bfields@fieldses.org>
---
 include/linux/sunrpc/svc_rdma.h          |    8 ++-
 net/sunrpc/xprtrdma/svc_rdma_sendto.c    |    6 +-
 net/sunrpc/xprtrdma/svc_rdma_transport.c |   85 ++++++++++++++++++++++++++----
 3 files changed, 84 insertions(+), 15 deletions(-)

diff --git a/include/linux/sunrpc/svc_rdma.h b/include/linux/sunrpc/svc_rdma.h
index be2804b..05bf4fe 100644
--- a/include/linux/sunrpc/svc_rdma.h
+++ b/include/linux/sunrpc/svc_rdma.h
@@ -113,6 +113,7 @@ struct svc_rdma_fastreg_mr {
 	struct list_head frmr_list;
 };
 struct svc_rdma_req_map {
+	struct list_head free;
 	unsigned long count;
 	union {
 		struct kvec sge[RPCSVC_MAXPAGES];
@@ -145,6 +146,8 @@ struct svcxprt_rdma {
 	spinlock_t	     sc_ctxt_lock;
 	struct list_head     sc_ctxts;
 	int		     sc_ctxt_used;
+	spinlock_t	     sc_map_lock;
+	struct list_head     sc_maps;
 
 	struct list_head     sc_rq_dto_q;
 	spinlock_t	     sc_rq_dto_lock;
@@ -223,8 +226,9 @@ extern int svc_rdma_create_listen(struct svc_serv *, int, struct sockaddr *);
 extern struct svc_rdma_op_ctxt *svc_rdma_get_context(struct svcxprt_rdma *);
 extern void svc_rdma_put_context(struct svc_rdma_op_ctxt *, int);
 extern void svc_rdma_unmap_dma(struct svc_rdma_op_ctxt *ctxt);
-extern struct svc_rdma_req_map *svc_rdma_get_req_map(void);
-extern void svc_rdma_put_req_map(struct svc_rdma_req_map *);
+extern struct svc_rdma_req_map *svc_rdma_get_req_map(struct svcxprt_rdma *);
+extern void svc_rdma_put_req_map(struct svcxprt_rdma *,
+				 struct svc_rdma_req_map *);
 extern struct svc_rdma_fastreg_mr *svc_rdma_get_frmr(struct svcxprt_rdma *);
 extern void svc_rdma_put_frmr(struct svcxprt_rdma *,
 			      struct svc_rdma_fastreg_mr *);
diff --git a/net/sunrpc/xprtrdma/svc_rdma_sendto.c b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
index 969a1ab..9a097f9 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_sendto.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
@@ -591,7 +591,7 @@ int svc_rdma_sendto(struct svc_rqst *rqstp)
 	/* Build an req vec for the XDR */
 	ctxt = svc_rdma_get_context(rdma);
 	ctxt->direction = DMA_TO_DEVICE;
-	vec = svc_rdma_get_req_map();
+	vec = svc_rdma_get_req_map(rdma);
 	ret = map_xdr(rdma, &rqstp->rq_res, vec);
 	if (ret)
 		goto err0;
@@ -630,14 +630,14 @@ int svc_rdma_sendto(struct svc_rqst *rqstp)
 
 	ret = send_reply(rdma, rqstp, res_page, rdma_resp, ctxt, vec,
 			 inline_bytes);
-	svc_rdma_put_req_map(vec);
+	svc_rdma_put_req_map(rdma, vec);
 	dprintk("svcrdma: send_reply returns %d\n", ret);
 	return ret;
 
  err1:
 	put_page(res_page);
  err0:
-	svc_rdma_put_req_map(vec);
+	svc_rdma_put_req_map(rdma, vec);
 	svc_rdma_put_context(ctxt, 0);
 	return ret;
 }
diff --git a/net/sunrpc/xprtrdma/svc_rdma_transport.c b/net/sunrpc/xprtrdma/svc_rdma_transport.c
index 9801115..0b9e17e 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_transport.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_transport.c
@@ -273,23 +273,83 @@ static void svc_rdma_destroy_ctxts(struct svcxprt_rdma *xprt)
 	}
 }
 
-/*
- * Temporary NFS req mappings are shared across all transport
- * instances. These are short lived and should be bounded by the number
- * of concurrent server threads * depth of the SQ.
- */
-struct svc_rdma_req_map *svc_rdma_get_req_map(void)
+static struct svc_rdma_req_map *alloc_req_map(gfp_t flags)
 {
 	struct svc_rdma_req_map *map;
-	map = kmem_cache_alloc(svc_rdma_map_cachep,
-			       GFP_KERNEL | __GFP_NOFAIL);
+
+	map = kmalloc(sizeof(*map), flags);
+	if (map)
+		INIT_LIST_HEAD(&map->free);
+	return map;
+}
+
+static bool svc_rdma_prealloc_maps(struct svcxprt_rdma *xprt)
+{
+	int i;
+
+	/* One for each receive buffer on this connection. */
+	i = xprt->sc_max_requests;
+
+	while (i--) {
+		struct svc_rdma_req_map *map;
+
+		map = alloc_req_map(GFP_KERNEL);
+		if (!map) {
+			dprintk("svcrdma: No memory for request map\n");
+			return false;
+		}
+		list_add(&map->free, &xprt->sc_maps);
+	}
+	return true;
+}
+
+struct svc_rdma_req_map *svc_rdma_get_req_map(struct svcxprt_rdma *xprt)
+{
+	struct svc_rdma_req_map *map = NULL;
+
+	spin_lock(&xprt->sc_map_lock);
+	if (list_empty(&xprt->sc_maps))
+		goto out_empty;
+
+	map = list_first_entry(&xprt->sc_maps,
+			       struct svc_rdma_req_map, free);
+	list_del_init(&map->free);
+	spin_unlock(&xprt->sc_map_lock);
+
+out:
 	map->count = 0;
 	return map;
+
+out_empty:
+	spin_unlock(&xprt->sc_map_lock);
+
+	/* Pre-allocation amount was incorrect */
+	map = alloc_req_map(GFP_NOIO);
+	if (map)
+		goto out;
+
+	WARN_ONCE(1, "svcrdma: empty request map list?\n");
+	return NULL;
 }
 
-void svc_rdma_put_req_map(struct svc_rdma_req_map *map)
+void svc_rdma_put_req_map(struct svcxprt_rdma *xprt,
+			  struct svc_rdma_req_map *map)
 {
-	kmem_cache_free(svc_rdma_map_cachep, map);
+	spin_lock(&xprt->sc_map_lock);
+	list_add(&map->free, &xprt->sc_maps);
+	spin_unlock(&xprt->sc_map_lock);
+}
+
+static void svc_rdma_destroy_maps(struct svcxprt_rdma *xprt)
+{
+	while (!list_empty(&xprt->sc_maps)) {
+		struct svc_rdma_req_map *map;
+
+		map = list_first_entry(&xprt->sc_maps,
+				       struct svc_rdma_req_map, free);
+		list_del(&map->free);
+		kfree(map);
+	}
 }
 
 /* ib_cq event handler */
@@ -593,12 +653,14 @@ static struct svcxprt_rdma *rdma_create_xprt(struct svc_serv *serv,
 	INIT_LIST_HEAD(&cma_xprt->sc_read_complete_q);
 	INIT_LIST_HEAD(&cma_xprt->sc_frmr_q);
 	INIT_LIST_HEAD(&cma_xprt->sc_ctxts);
+	INIT_LIST_HEAD(&cma_xprt->sc_maps);
 	init_waitqueue_head(&cma_xprt->sc_send_wait);
 
 	spin_lock_init(&cma_xprt->sc_lock);
 	spin_lock_init(&cma_xprt->sc_rq_dto_lock);
 	spin_lock_init(&cma_xprt->sc_frmr_q_lock);
 	spin_lock_init(&cma_xprt->sc_ctxt_lock);
+	spin_lock_init(&cma_xprt->sc_map_lock);
 
 	if (listener)
 		set_bit(XPT_LISTENER, &cma_xprt->sc_xprt.xpt_flags);
@@ -988,6 +1050,8 @@ static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt)
 
 	if (!svc_rdma_prealloc_ctxts(newxprt))
 		goto errout;
+	if (!svc_rdma_prealloc_maps(newxprt))
+		goto errout;
 
 	/*
 	 * Limit ORD based on client limit, local device limit, and
@@ -1259,6 +1323,7 @@ static void __svc_rdma_free(struct work_struct *work)
 
 	rdma_dealloc_frmr_q(rdma);
 	svc_rdma_destroy_ctxts(rdma);
+	svc_rdma_destroy_maps(rdma);
 
 	/* Destroy the QP if present (not a listener) */
 	if (rdma->sc_qp && !IS_ERR(rdma->sc_qp))

[PATCH v5 05/10] svcrdma: Remove unused req_map and ctxt kmem_caches

[Chuck Lever]

Clean up.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Acked-by: Bruce Fields <bfields@fieldses.org>
---
 include/linux/sunrpc/svc_rdma.h |    1 +
 net/sunrpc/xprtrdma/svc_rdma.c  |   35 -----------------------------------
 net/sunrpc/xprtrdma/xprt_rdma.h |    7 -------
 3 files changed, 1 insertion(+), 42 deletions(-)

diff --git a/include/linux/sunrpc/svc_rdma.h b/include/linux/sunrpc/svc_rdma.h
index 05bf4fe..141edbb 100644
--- a/include/linux/sunrpc/svc_rdma.h
+++ b/include/linux/sunrpc/svc_rdma.h
@@ -242,6 +242,7 @@ extern struct svc_xprt_class svc_rdma_bc_class;
 #endif
 
 /* svc_rdma.c */
+extern struct workqueue_struct *svc_rdma_wq;
 extern int svc_rdma_init(void);
 extern void svc_rdma_cleanup(void);
 
diff --git a/net/sunrpc/xprtrdma/svc_rdma.c b/net/sunrpc/xprtrdma/svc_rdma.c
index 1b7051b..e894e06 100644
--- a/net/sunrpc/xprtrdma/svc_rdma.c
+++ b/net/sunrpc/xprtrdma/svc_rdma.c
@@ -71,10 +71,6 @@ atomic_t rdma_stat_rq_prod;
 atomic_t rdma_stat_sq_poll;
 atomic_t rdma_stat_sq_prod;
 
-/* Temporary NFS request map and context caches */
-struct kmem_cache *svc_rdma_map_cachep;
-struct kmem_cache *svc_rdma_ctxt_cachep;
-
 struct workqueue_struct *svc_rdma_wq;
 
 /*
@@ -243,8 +239,6 @@ void svc_rdma_cleanup(void)
 	svc_unreg_xprt_class(&svc_rdma_bc_class);
 #endif
 	svc_unreg_xprt_class(&svc_rdma_class);
-	kmem_cache_destroy(svc_rdma_map_cachep);
-	kmem_cache_destroy(svc_rdma_ctxt_cachep);
 }
 
 int svc_rdma_init(void)
@@ -264,39 +258,10 @@ int svc_rdma_init(void)
 		svcrdma_table_header =
 			register_sysctl_table(svcrdma_root_table);
 
-	/* Create the temporary map cache */
-	svc_rdma_map_cachep = kmem_cache_create("svc_rdma_map_cache",
-						sizeof(struct svc_rdma_req_map),
-						0,
-						SLAB_HWCACHE_ALIGN,
-						NULL);
-	if (!svc_rdma_map_cachep) {
-		printk(KERN_INFO "Could not allocate map cache.\n");
-		goto err0;
-	}
-
-	/* Create the temporary context cache */
-	svc_rdma_ctxt_cachep =
-		kmem_cache_create("svc_rdma_ctxt_cache",
-				  sizeof(struct svc_rdma_op_ctxt),
-				  0,
-				  SLAB_HWCACHE_ALIGN,
-				  NULL);
-	if (!svc_rdma_ctxt_cachep) {
-		printk(KERN_INFO "Could not allocate WR ctxt cache.\n");
-		goto err1;
-	}
-
 	/* Register RDMA with the SVC transport switch */
 	svc_reg_xprt_class(&svc_rdma_class);
 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
 	svc_reg_xprt_class(&svc_rdma_bc_class);
 #endif
 	return 0;
- err1:
-	kmem_cache_destroy(svc_rdma_map_cachep);
- err0:
-	unregister_sysctl_table(svcrdma_table_header);
-	destroy_workqueue(svc_rdma_wq);
-	return -ENOMEM;
 }
diff --git a/net/sunrpc/xprtrdma/xprt_rdma.h b/net/sunrpc/xprtrdma/xprt_rdma.h
index 4197191..72276c7 100644
--- a/net/sunrpc/xprtrdma/xprt_rdma.h
+++ b/net/sunrpc/xprtrdma/xprt_rdma.h
@@ -528,11 +528,4 @@ void xprt_rdma_bc_free_rqst(struct rpc_rqst *);
 void xprt_rdma_bc_destroy(struct rpc_xprt *, unsigned int);
 #endif	/* CONFIG_SUNRPC_BACKCHANNEL */
 
-/* Temporary NFS request map cache. Created in svc_rdma.c  */
-extern struct kmem_cache *svc_rdma_map_cachep;
-/* WR context cache. Created in svc_rdma.c  */
-extern struct kmem_cache *svc_rdma_ctxt_cachep;
-/* Workqueue created in svc_rdma.c */
-extern struct workqueue_struct *svc_rdma_wq;
-
 #endif				/* _LINUX_SUNRPC_XPRT_RDMA_H */

[PATCH v5 06/10] svcrdma: Add gfp flags to svc_rdma_post_recv()

[Chuck Lever]

svc_rdma_post_recv() allocates pages for receive buffers on-demand.
It uses GFP_KERNEL so the allocator tries hard, and may sleep. But
I'm about to add a call to svc_rdma_post_recv() from a function
that may not sleep.

Since all svc_rdma_post_recv() call sites can tolerate its failure,
allow it to fail if the page allocator returns nothing. Longer term,
receive buffers, being a finite resource per-connection, should be
pre-allocated and re-used.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Acked-by: Bruce Fields <bfields@fieldses.org>
---
 include/linux/sunrpc/svc_rdma.h          |    2 +-
 net/sunrpc/xprtrdma/svc_rdma_sendto.c    |    2 +-
 net/sunrpc/xprtrdma/svc_rdma_transport.c |    8 +++++---
 3 files changed, 7 insertions(+), 5 deletions(-)

diff --git a/include/linux/sunrpc/svc_rdma.h b/include/linux/sunrpc/svc_rdma.h
index 141edbb..729ff35 100644
--- a/include/linux/sunrpc/svc_rdma.h
+++ b/include/linux/sunrpc/svc_rdma.h
@@ -221,7 +221,7 @@ extern struct rpcrdma_read_chunk *
 extern int svc_rdma_send(struct svcxprt_rdma *, struct ib_send_wr *);
 extern void svc_rdma_send_error(struct svcxprt_rdma *, struct rpcrdma_msg *,
 				enum rpcrdma_errcode);
-extern int svc_rdma_post_recv(struct svcxprt_rdma *);
+extern int svc_rdma_post_recv(struct svcxprt_rdma *, gfp_t);
 extern int svc_rdma_create_listen(struct svc_serv *, int, struct sockaddr *);
 extern struct svc_rdma_op_ctxt *svc_rdma_get_context(struct svcxprt_rdma *);
 extern void svc_rdma_put_context(struct svc_rdma_op_ctxt *, int);
diff --git a/net/sunrpc/xprtrdma/svc_rdma_sendto.c b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
index 9a097f9..aeaec7a 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_sendto.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
@@ -465,7 +465,7 @@ static int send_reply(struct svcxprt_rdma *rdma,
 	int ret;
 
 	/* Post a recv buffer to handle another request. */
-	ret = svc_rdma_post_recv(rdma);
+	ret = svc_rdma_post_recv(rdma, GFP_KERNEL);
 	if (ret) {
 		printk(KERN_INFO
 		       "svcrdma: could not post a receive buffer, err=%d."
diff --git a/net/sunrpc/xprtrdma/svc_rdma_transport.c b/net/sunrpc/xprtrdma/svc_rdma_transport.c
index 0b9e17e..20fc095 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_transport.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_transport.c
@@ -668,7 +668,7 @@ static struct svcxprt_rdma *rdma_create_xprt(struct svc_serv *serv,
 	return cma_xprt;
 }
 
-int svc_rdma_post_recv(struct svcxprt_rdma *xprt)
+int svc_rdma_post_recv(struct svcxprt_rdma *xprt, gfp_t flags)
 {
 	struct ib_recv_wr recv_wr, *bad_recv_wr;
 	struct svc_rdma_op_ctxt *ctxt;
@@ -686,7 +686,9 @@ int svc_rdma_post_recv(struct svcxprt_rdma *xprt)
 			pr_err("svcrdma: Too many sges (%d)\n", sge_no);
 			goto err_put_ctxt;
 		}
-		page = alloc_page(GFP_KERNEL | __GFP_NOFAIL);
+		page = alloc_page(flags);
+		if (!page)
+			goto err_put_ctxt;
 		ctxt->pages[sge_no] = page;
 		pa = ib_dma_map_page(xprt->sc_cm_id->device,
 				     page, 0, PAGE_SIZE,
@@ -1182,7 +1184,7 @@ static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt)
 
 	/* Post receive buffers */
 	for (i = 0; i < newxprt->sc_max_requests; i++) {
-		ret = svc_rdma_post_recv(newxprt);
+		ret = svc_rdma_post_recv(newxprt, GFP_KERNEL);
 		if (ret) {
 			dprintk("svcrdma: failure posting receive buffers\n");
 			goto errout;

[PATCH v5 07/10] svcrdma: Remove last two __GFP_NOFAIL call sites

[Chuck Lever]

Clean up.

These functions can otherwise fail, so check for page allocation
failures too.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Acked-by: Bruce Fields <bfields@fieldses.org>
---
 net/sunrpc/xprtrdma/svc_rdma_sendto.c    |    5 ++++-
 net/sunrpc/xprtrdma/svc_rdma_transport.c |    4 +++-
 2 files changed, 7 insertions(+), 2 deletions(-)

diff --git a/net/sunrpc/xprtrdma/svc_rdma_sendto.c b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
index aeaec7a..de7df7b 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_sendto.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
@@ -598,7 +598,10 @@ int svc_rdma_sendto(struct svc_rqst *rqstp)
 	inline_bytes = rqstp->rq_res.len;
 
 	/* Create the RDMA response header */
-	res_page = alloc_page(GFP_KERNEL | __GFP_NOFAIL);
+	ret = -ENOMEM;
+	res_page = alloc_page(GFP_KERNEL);
+	if (!res_page)
+		goto err0;
 	rdma_resp = page_address(res_page);
 	reply_ary = svc_rdma_get_reply_array(rdma_argp);
 	if (reply_ary)
diff --git a/net/sunrpc/xprtrdma/svc_rdma_transport.c b/net/sunrpc/xprtrdma/svc_rdma_transport.c
index 20fc095..8b3ee04 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_transport.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_transport.c
@@ -1445,7 +1445,9 @@ void svc_rdma_send_error(struct svcxprt_rdma *xprt, struct rpcrdma_msg *rmsgp,
 	int length;
 	int ret;
 
-	p = alloc_page(GFP_KERNEL | __GFP_NOFAIL);
+	p = alloc_page(GFP_KERNEL);
+	if (!p)
+		return;
 	va = page_address(p);
 
 	/* XDR encode error */

[PATCH v5 08/10] svcrdma: Make map_xdr non-static

[Chuck Lever]

Pre-requisite to use map_xdr in the backchannel code.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Acked-by: Bruce Fields <bfields@fieldses.org>
---
 include/linux/sunrpc/svc_rdma.h       |    2 ++
 net/sunrpc/xprtrdma/svc_rdma_sendto.c |   14 +++++++-------
 2 files changed, 9 insertions(+), 7 deletions(-)

diff --git a/include/linux/sunrpc/svc_rdma.h b/include/linux/sunrpc/svc_rdma.h
index 729ff35..aeffa30 100644
--- a/include/linux/sunrpc/svc_rdma.h
+++ b/include/linux/sunrpc/svc_rdma.h
@@ -213,6 +213,8 @@ extern int rdma_read_chunk_frmr(struct svcxprt_rdma *, struct svc_rqst *,
 				u32, u32, u64, bool);
 
 /* svc_rdma_sendto.c */
+extern int svc_rdma_map_xdr(struct svcxprt_rdma *, struct xdr_buf *,
+			    struct svc_rdma_req_map *);
 extern int svc_rdma_sendto(struct svc_rqst *);
 extern struct rpcrdma_read_chunk *
 	svc_rdma_get_read_chunk(struct rpcrdma_msg *);
diff --git a/net/sunrpc/xprtrdma/svc_rdma_sendto.c b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
index de7df7b..3c25052 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_sendto.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
@@ -50,9 +50,9 @@
 
 #define RPCDBG_FACILITY	RPCDBG_SVCXPRT
 
-static int map_xdr(struct svcxprt_rdma *xprt,
-		   struct xdr_buf *xdr,
-		   struct svc_rdma_req_map *vec)
+int svc_rdma_map_xdr(struct svcxprt_rdma *xprt,
+		     struct xdr_buf *xdr,
+		     struct svc_rdma_req_map *vec)
 {
 	int sge_no;
 	u32 sge_bytes;
@@ -62,7 +62,7 @@ static int map_xdr(struct svcxprt_rdma *xprt,
 
 	if (xdr->len !=
 	    (xdr->head[0].iov_len + xdr->page_len + xdr->tail[0].iov_len)) {
-		pr_err("svcrdma: map_xdr: XDR buffer length error\n");
+		pr_err("svcrdma: %s: XDR buffer length error\n", __func__);
 		return -EIO;
 	}
 
@@ -97,9 +97,9 @@ static int map_xdr(struct svcxprt_rdma *xprt,
 		sge_no++;
 	}
 
-	dprintk("svcrdma: map_xdr: sge_no %d page_no %d "
+	dprintk("svcrdma: %s: sge_no %d page_no %d "
 		"page_base %u page_len %u head_len %zu tail_len %zu\n",
-		sge_no, page_no, xdr->page_base, xdr->page_len,
+		__func__, sge_no, page_no, xdr->page_base, xdr->page_len,
 		xdr->head[0].iov_len, xdr->tail[0].iov_len);
 
 	vec->count = sge_no;
@@ -592,7 +592,7 @@ int svc_rdma_sendto(struct svc_rqst *rqstp)
 	ctxt = svc_rdma_get_context(rdma);
 	ctxt->direction = DMA_TO_DEVICE;
 	vec = svc_rdma_get_req_map(rdma);
-	ret = map_xdr(rdma, &rqstp->rq_res, vec);
+	ret = svc_rdma_map_xdr(rdma, &rqstp->rq_res, vec);
 	if (ret)
 		goto err0;
 	inline_bytes = rqstp->rq_res.len;

[PATCH v5 09/10] svcrdma: Define maximum number of backchannel requests

[Chuck Lever]

Extra resources for handling backchannel requests have to be
pre-allocated when a transport instance is created. Set up
additional fields in svcxprt_rdma to track these resources.

The max_requests fields are elements of the RPC-over-RDMA
protocol, so they should be u32. To ensure that unsigned
arithmetic is used everywhere, some other fields in the
svcxprt_rdma struct are updated.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Acked-by: Bruce Fields <bfields@fieldses.org>
---
 include/linux/sunrpc/svc_rdma.h          |   13 ++++++++++---
 net/sunrpc/xprtrdma/svc_rdma.c           |    6 ++++--
 net/sunrpc/xprtrdma/svc_rdma_transport.c |   24 ++++++++++++++----------
 3 files changed, 28 insertions(+), 15 deletions(-)

diff --git a/include/linux/sunrpc/svc_rdma.h b/include/linux/sunrpc/svc_rdma.h
index aeffa30..9a2c418 100644
--- a/include/linux/sunrpc/svc_rdma.h
+++ b/include/linux/sunrpc/svc_rdma.h
@@ -51,6 +51,7 @@
 /* RPC/RDMA parameters and stats */
 extern unsigned int svcrdma_ord;
 extern unsigned int svcrdma_max_requests;
+extern unsigned int svcrdma_max_bc_requests;
 extern unsigned int svcrdma_max_req_size;
 
 extern atomic_t rdma_stat_recv;
@@ -134,10 +135,11 @@ struct svcxprt_rdma {
 	int                  sc_max_sge;
 	int                  sc_max_sge_rd;	/* max sge for read target */
 
-	int                  sc_sq_depth;	/* Depth of SQ */
 	atomic_t             sc_sq_count;	/* Number of SQ WR on queue */
-
-	int                  sc_max_requests;	/* Depth of RQ */
+	unsigned int	     sc_sq_depth;	/* Depth of SQ */
+	unsigned int	     sc_rq_depth;	/* Depth of RQ */
+	u32		     sc_max_requests;	/* Forward credits */
+	u32		     sc_max_bc_requests;/* Backward credits */
 	int                  sc_max_req_size;	/* Size of each RQ WR buf */
 
 	struct ib_pd         *sc_pd;
@@ -186,6 +188,11 @@ struct svcxprt_rdma {
 #define RPCRDMA_MAX_REQUESTS    32
 #define RPCRDMA_MAX_REQ_SIZE    4096
 
+/* Typical ULP usage of BC requests is NFSv4.1 backchannel. Our
+ * current NFSv4.1 implementation supports one backchannel slot.
+ */
+#define RPCRDMA_MAX_BC_REQUESTS	2
+
 #define RPCSVC_MAXPAYLOAD_RDMA	RPCSVC_MAXPAYLOAD
 
 /* svc_rdma_marshal.c */
diff --git a/net/sunrpc/xprtrdma/svc_rdma.c b/net/sunrpc/xprtrdma/svc_rdma.c
index e894e06..c846ca9 100644
--- a/net/sunrpc/xprtrdma/svc_rdma.c
+++ b/net/sunrpc/xprtrdma/svc_rdma.c
@@ -55,6 +55,7 @@ unsigned int svcrdma_ord = RPCRDMA_ORD;
 static unsigned int min_ord = 1;
 static unsigned int max_ord = 4096;
 unsigned int svcrdma_max_requests = RPCRDMA_MAX_REQUESTS;
+unsigned int svcrdma_max_bc_requests = RPCRDMA_MAX_BC_REQUESTS;
 static unsigned int min_max_requests = 4;
 static unsigned int max_max_requests = 16384;
 unsigned int svcrdma_max_req_size = RPCRDMA_MAX_REQ_SIZE;
@@ -245,9 +246,10 @@ int svc_rdma_init(void)
 {
 	dprintk("SVCRDMA Module Init, register RPC RDMA transport\n");
 	dprintk("\tsvcrdma_ord      : %d\n", svcrdma_ord);
-	dprintk("\tmax_requests     : %d\n", svcrdma_max_requests);
-	dprintk("\tsq_depth         : %d\n",
+	dprintk("\tmax_requests     : %u\n", svcrdma_max_requests);
+	dprintk("\tsq_depth         : %u\n",
 		svcrdma_max_requests * RPCRDMA_SQ_DEPTH_MULT);
+	dprintk("\tmax_bc_requests  : %u\n", svcrdma_max_bc_requests);
 	dprintk("\tmax_inline       : %d\n", svcrdma_max_req_size);
 
 	svc_rdma_wq = alloc_workqueue("svc_rdma", 0, 0);
diff --git a/net/sunrpc/xprtrdma/svc_rdma_transport.c b/net/sunrpc/xprtrdma/svc_rdma_transport.c
index 8b3ee04..af86dfe 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_transport.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_transport.c
@@ -169,12 +169,12 @@ static struct svc_rdma_op_ctxt *alloc_ctxt(struct svcxprt_rdma *xprt,
 
 static bool svc_rdma_prealloc_ctxts(struct svcxprt_rdma *xprt)
 {
-	int i;
+	unsigned int i;
 
 	/* Each RPC/RDMA credit can consume a number of send
 	 * and receive WQEs. One ctxt is allocated for each.
 	 */
-	i = xprt->sc_sq_depth + xprt->sc_max_requests;
+	i = xprt->sc_sq_depth + xprt->sc_rq_depth;
 
 	while (i--) {
 		struct svc_rdma_op_ctxt *ctxt;
@@ -285,7 +285,7 @@ static struct svc_rdma_req_map *alloc_req_map(gfp_t flags)
 
 static bool svc_rdma_prealloc_maps(struct svcxprt_rdma *xprt)
 {
-	int i;
+	unsigned int i;
 
 	/* One for each receive buffer on this connection. */
 	i = xprt->sc_max_requests;
@@ -1016,8 +1016,8 @@ static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt)
 	struct ib_device *dev;
 	int uninitialized_var(dma_mr_acc);
 	int need_dma_mr = 0;
+	unsigned int i;
 	int ret = 0;
-	int i;
 
 	listen_rdma = container_of(xprt, struct svcxprt_rdma, sc_xprt);
 	clear_bit(XPT_CONN, &xprt->xpt_flags);
@@ -1046,9 +1046,13 @@ static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt)
 	newxprt->sc_max_sge_rd = min_t(size_t, dev->attrs.max_sge_rd,
 				       RPCSVC_MAXPAGES);
 	newxprt->sc_max_req_size = svcrdma_max_req_size;
-	newxprt->sc_max_requests = min((size_t)dev->attrs.max_qp_wr,
-				   (size_t)svcrdma_max_requests);
-	newxprt->sc_sq_depth = RPCRDMA_SQ_DEPTH_MULT * newxprt->sc_max_requests;
+	newxprt->sc_max_requests = min_t(u32, dev->attrs.max_qp_wr,
+					 svcrdma_max_requests);
+	newxprt->sc_max_bc_requests = min_t(u32, dev->attrs.max_qp_wr,
+					    svcrdma_max_bc_requests);
+	newxprt->sc_rq_depth = newxprt->sc_max_requests +
+			       newxprt->sc_max_bc_requests;
+	newxprt->sc_sq_depth = RPCRDMA_SQ_DEPTH_MULT * newxprt->sc_rq_depth;
 
 	if (!svc_rdma_prealloc_ctxts(newxprt))
 		goto errout;
@@ -1077,7 +1081,7 @@ static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt)
 		dprintk("svcrdma: error creating SQ CQ for connect request\n");
 		goto errout;
 	}
-	cq_attr.cqe = newxprt->sc_max_requests;
+	cq_attr.cqe = newxprt->sc_rq_depth;
 	newxprt->sc_rq_cq = ib_create_cq(dev,
 					 rq_comp_handler,
 					 cq_event_handler,
@@ -1092,7 +1096,7 @@ static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt)
 	qp_attr.event_handler = qp_event_handler;
 	qp_attr.qp_context = &newxprt->sc_xprt;
 	qp_attr.cap.max_send_wr = newxprt->sc_sq_depth;
-	qp_attr.cap.max_recv_wr = newxprt->sc_max_requests;
+	qp_attr.cap.max_recv_wr = newxprt->sc_rq_depth;
 	qp_attr.cap.max_send_sge = newxprt->sc_max_sge;
 	qp_attr.cap.max_recv_sge = newxprt->sc_max_sge;
 	qp_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
@@ -1183,7 +1187,7 @@ static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt)
 		newxprt->sc_dma_lkey = dev->local_dma_lkey;
 
 	/* Post receive buffers */
-	for (i = 0; i < newxprt->sc_max_requests; i++) {
+	for (i = 0; i < newxprt->sc_rq_depth; i++) {
 		ret = svc_rdma_post_recv(newxprt, GFP_KERNEL);
 		if (ret) {
 			dprintk("svcrdma: failure posting receive buffers\n");

[PATCH v5 10/10] svcrdma: Add class for RDMA backwards direction transport

[Chuck Lever]

To support the server-side of an NFSv4.1 backchannel on RDMA
connections, add a transport class that enables backward
direction messages on an existing forward channel connection.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Acked-by: Bruce Fields <bfields@fieldses.org>
---
 include/linux/sunrpc/svc_rdma.h            |    5 
 net/sunrpc/xprt.c                          |    1 
 net/sunrpc/xprtrdma/Makefile               |    2 
 net/sunrpc/xprtrdma/svc_rdma_backchannel.c |  371 ++++++++++++++++++++++++++++
 net/sunrpc/xprtrdma/svc_rdma_recvfrom.c    |   52 ++++
 net/sunrpc/xprtrdma/svc_rdma_transport.c   |   14 +
 net/sunrpc/xprtrdma/transport.c            |   30 +-
 net/sunrpc/xprtrdma/xprt_rdma.h            |   15 +
 8 files changed, 475 insertions(+), 15 deletions(-)
 create mode 100644 net/sunrpc/xprtrdma/svc_rdma_backchannel.c

diff --git a/include/linux/sunrpc/svc_rdma.h b/include/linux/sunrpc/svc_rdma.h
index 9a2c418..b13513a 100644
--- a/include/linux/sunrpc/svc_rdma.h
+++ b/include/linux/sunrpc/svc_rdma.h
@@ -195,6 +195,11 @@ struct svcxprt_rdma {
 
 #define RPCSVC_MAXPAYLOAD_RDMA	RPCSVC_MAXPAYLOAD
 
+/* svc_rdma_backchannel.c */
+extern int svc_rdma_handle_bc_reply(struct rpc_xprt *xprt,
+				    struct rpcrdma_msg *rmsgp,
+				    struct xdr_buf *rcvbuf);
+
 /* svc_rdma_marshal.c */
 extern int svc_rdma_xdr_decode_req(struct rpcrdma_msg **, struct svc_rqst *);
 extern int svc_rdma_xdr_encode_error(struct svcxprt_rdma *,
diff --git a/net/sunrpc/xprt.c b/net/sunrpc/xprt.c
index 2e98f4a..37edea6 100644
--- a/net/sunrpc/xprt.c
+++ b/net/sunrpc/xprt.c
@@ -1425,3 +1425,4 @@ void xprt_put(struct rpc_xprt *xprt)
 	if (atomic_dec_and_test(&xprt->count))
 		xprt_destroy(xprt);
 }
+EXPORT_SYMBOL_GPL(xprt_put);
diff --git a/net/sunrpc/xprtrdma/Makefile b/net/sunrpc/xprtrdma/Makefile
index 33f99d3..dc9f3b5 100644
--- a/net/sunrpc/xprtrdma/Makefile
+++ b/net/sunrpc/xprtrdma/Makefile
@@ -2,7 +2,7 @@ obj-$(CONFIG_SUNRPC_XPRT_RDMA) += rpcrdma.o
 
 rpcrdma-y := transport.o rpc_rdma.o verbs.o \
 	fmr_ops.o frwr_ops.o physical_ops.o \
-	svc_rdma.o svc_rdma_transport.o \
+	svc_rdma.o svc_rdma_backchannel.o svc_rdma_transport.o \
 	svc_rdma_marshal.o svc_rdma_sendto.o svc_rdma_recvfrom.o \
 	module.o
 rpcrdma-$(CONFIG_SUNRPC_BACKCHANNEL) += backchannel.o
diff --git a/net/sunrpc/xprtrdma/svc_rdma_backchannel.c b/net/sunrpc/xprtrdma/svc_rdma_backchannel.c
new file mode 100644
index 0000000..deff06a
--- /dev/null
+++ b/net/sunrpc/xprtrdma/svc_rdma_backchannel.c
@@ -0,0 +1,371 @@
+/*
+ * Copyright (c) 2015 Oracle.  All rights reserved.
+ *
+ * Support for backward direction RPCs on RPC/RDMA (server-side).
+ */
+
+#include <linux/sunrpc/svc_rdma.h>
+#include "xprt_rdma.h"
+
+#define RPCDBG_FACILITY	RPCDBG_SVCXPRT
+
+#undef SVCRDMA_BACKCHANNEL_DEBUG
+
+int svc_rdma_handle_bc_reply(struct rpc_xprt *xprt, struct rpcrdma_msg *rmsgp,
+			     struct xdr_buf *rcvbuf)
+{
+	struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
+	struct kvec *dst, *src = &rcvbuf->head[0];
+	struct rpc_rqst *req;
+	unsigned long cwnd;
+	u32 credits;
+	size_t len;
+	__be32 xid;
+	__be32 *p;
+	int ret;
+
+	p = (__be32 *)src->iov_base;
+	len = src->iov_len;
+	xid = rmsgp->rm_xid;
+
+#ifdef SVCRDMA_BACKCHANNEL_DEBUG
+	pr_info("%s: xid=%08x, length=%zu\n",
+		__func__, be32_to_cpu(xid), len);
+	pr_info("%s: RPC/RDMA: %*ph\n",
+		__func__, (int)RPCRDMA_HDRLEN_MIN, rmsgp);
+	pr_info("%s:      RPC: %*ph\n",
+		__func__, (int)len, p);
+#endif
+
+	ret = -EAGAIN;
+	if (src->iov_len < 24)
+		goto out_shortreply;
+
+	spin_lock_bh(&xprt->transport_lock);
+	req = xprt_lookup_rqst(xprt, xid);
+	if (!req)
+		goto out_notfound;
+
+	dst = &req->rq_private_buf.head[0];
+	memcpy(&req->rq_private_buf, &req->rq_rcv_buf, sizeof(struct xdr_buf));
+	if (dst->iov_len < len)
+		goto out_unlock;
+	memcpy(dst->iov_base, p, len);
+
+	credits = be32_to_cpu(rmsgp->rm_credit);
+	if (credits == 0)
+		credits = 1;	/* don't deadlock */
+	else if (credits > r_xprt->rx_buf.rb_bc_max_requests)
+		credits = r_xprt->rx_buf.rb_bc_max_requests;
+
+	cwnd = xprt->cwnd;
+	xprt->cwnd = credits << RPC_CWNDSHIFT;
+	if (xprt->cwnd > cwnd)
+		xprt_release_rqst_cong(req->rq_task);
+
+	ret = 0;
+	xprt_complete_rqst(req->rq_task, rcvbuf->len);
+	rcvbuf->len = 0;
+
+out_unlock:
+	spin_unlock_bh(&xprt->transport_lock);
+out:
+	return ret;
+
+out_shortreply:
+	dprintk("svcrdma: short bc reply: xprt=%p, len=%zu\n",
+		xprt, src->iov_len);
+	goto out;
+
+out_notfound:
+	dprintk("svcrdma: unrecognized bc reply: xprt=%p, xid=%08x\n",
+		xprt, be32_to_cpu(xid));
+
+	goto out_unlock;
+}
+
+/* Send a backwards direction RPC call.
+ *
+ * Caller holds the connection's mutex and has already marshaled
+ * the RPC/RDMA request.
+ *
+ * This is similar to svc_rdma_reply, but takes an rpc_rqst
+ * instead, does not support chunks, and avoids blocking memory
+ * allocation.
+ *
+ * XXX: There is still an opportunity to block in svc_rdma_send()
+ * if there are no SQ entries to post the Send. This may occur if
+ * the adapter has a small maximum SQ depth.
+ */
+static int svc_rdma_bc_sendto(struct svcxprt_rdma *rdma,
+			      struct rpc_rqst *rqst)
+{
+	struct xdr_buf *sndbuf = &rqst->rq_snd_buf;
+	struct svc_rdma_op_ctxt *ctxt;
+	struct svc_rdma_req_map *vec;
+	struct ib_send_wr send_wr;
+	int ret;
+
+	vec = svc_rdma_get_req_map(rdma);
+	ret = svc_rdma_map_xdr(rdma, sndbuf, vec);
+	if (ret)
+		goto out_err;
+
+	/* Post a recv buffer to handle the reply for this request. */
+	ret = svc_rdma_post_recv(rdma, GFP_NOIO);
+	if (ret) {
+		pr_err("svcrdma: Failed to post bc receive buffer, err=%d.\n",
+		       ret);
+		pr_err("svcrdma: closing transport %p.\n", rdma);
+		set_bit(XPT_CLOSE, &rdma->sc_xprt.xpt_flags);
+		ret = -ENOTCONN;
+		goto out_err;
+	}
+
+	ctxt = svc_rdma_get_context(rdma);
+	ctxt->pages[0] = virt_to_page(rqst->rq_buffer);
+	ctxt->count = 1;
+
+	ctxt->wr_op = IB_WR_SEND;
+	ctxt->direction = DMA_TO_DEVICE;
+	ctxt->sge[0].lkey = rdma->sc_dma_lkey;
+	ctxt->sge[0].length = sndbuf->len;
+	ctxt->sge[0].addr =
+	    ib_dma_map_page(rdma->sc_cm_id->device, ctxt->pages[0], 0,
+			    sndbuf->len, DMA_TO_DEVICE);
+	if (ib_dma_mapping_error(rdma->sc_cm_id->device, ctxt->sge[0].addr)) {
+		ret = -EIO;
+		goto out_unmap;
+	}
+	atomic_inc(&rdma->sc_dma_used);
+
+	memset(&send_wr, 0, sizeof(send_wr));
+	send_wr.wr_id = (unsigned long)ctxt;
+	send_wr.sg_list = ctxt->sge;
+	send_wr.num_sge = 1;
+	send_wr.opcode = IB_WR_SEND;
+	send_wr.send_flags = IB_SEND_SIGNALED;
+
+	ret = svc_rdma_send(rdma, &send_wr);
+	if (ret) {
+		ret = -EIO;
+		goto out_unmap;
+	}
+
+out_err:
+	svc_rdma_put_req_map(rdma, vec);
+	dprintk("svcrdma: %s returns %d\n", __func__, ret);
+	return ret;
+
+out_unmap:
+	svc_rdma_unmap_dma(ctxt);
+	svc_rdma_put_context(ctxt, 1);
+	goto out_err;
+}
+
+/* Server-side transport endpoint wants a whole page for its send
+ * buffer. The client RPC code constructs the RPC header in this
+ * buffer before it invokes ->send_request.
+ *
+ * Returns NULL if there was a temporary allocation failure.
+ */
+static void *
+xprt_rdma_bc_allocate(struct rpc_task *task, size_t size)
+{
+	struct rpc_rqst *rqst = task->tk_rqstp;
+	struct svc_xprt *sxprt = rqst->rq_xprt->bc_xprt;
+	struct svcxprt_rdma *rdma;
+	struct page *page;
+
+	rdma = container_of(sxprt, struct svcxprt_rdma, sc_xprt);
+
+	/* Prevent an infinite loop: try to make this case work */
+	if (size > PAGE_SIZE)
+		WARN_ONCE(1, "svcrdma: large bc buffer request (size %zu)\n",
+			  size);
+
+	page = alloc_page(RPCRDMA_DEF_GFP);
+	if (!page)
+		return NULL;
+
+	return page_address(page);
+}
+
+static void
+xprt_rdma_bc_free(void *buffer)
+{
+	/* No-op: ctxt and page have already been freed. */
+}
+
+static int
+rpcrdma_bc_send_request(struct svcxprt_rdma *rdma, struct rpc_rqst *rqst)
+{
+	struct rpc_xprt *xprt = rqst->rq_xprt;
+	struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
+	struct rpcrdma_msg *headerp = (struct rpcrdma_msg *)rqst->rq_buffer;
+	int rc;
+
+	/* Space in the send buffer for an RPC/RDMA header is reserved
+	 * via xprt->tsh_size.
+	 */
+	headerp->rm_xid = rqst->rq_xid;
+	headerp->rm_vers = rpcrdma_version;
+	headerp->rm_credit = cpu_to_be32(r_xprt->rx_buf.rb_bc_max_requests);
+	headerp->rm_type = rdma_msg;
+	headerp->rm_body.rm_chunks[0] = xdr_zero;
+	headerp->rm_body.rm_chunks[1] = xdr_zero;
+	headerp->rm_body.rm_chunks[2] = xdr_zero;
+
+#ifdef SVCRDMA_BACKCHANNEL_DEBUG
+	pr_info("%s: %*ph\n", __func__, 64, rqst->rq_buffer);
+#endif
+
+	rc = svc_rdma_bc_sendto(rdma, rqst);
+	if (rc)
+		goto drop_connection;
+	return rc;
+
+drop_connection:
+	dprintk("svcrdma: failed to send bc call\n");
+	xprt_disconnect_done(xprt);
+	return -ENOTCONN;
+}
+
+/* Send an RPC call on the passive end of a transport
+ * connection.
+ */
+static int
+xprt_rdma_bc_send_request(struct rpc_task *task)
+{
+	struct rpc_rqst *rqst = task->tk_rqstp;
+	struct svc_xprt *sxprt = rqst->rq_xprt->bc_xprt;
+	struct svcxprt_rdma *rdma;
+	int ret;
+
+	dprintk("svcrdma: sending bc call with xid: %08x\n",
+		be32_to_cpu(rqst->rq_xid));
+
+	if (!mutex_trylock(&sxprt->xpt_mutex)) {
+		rpc_sleep_on(&sxprt->xpt_bc_pending, task, NULL);
+		if (!mutex_trylock(&sxprt->xpt_mutex))
+			return -EAGAIN;
+		rpc_wake_up_queued_task(&sxprt->xpt_bc_pending, task);
+	}
+
+	ret = -ENOTCONN;
+	rdma = container_of(sxprt, struct svcxprt_rdma, sc_xprt);
+	if (!test_bit(XPT_DEAD, &sxprt->xpt_flags))
+		ret = rpcrdma_bc_send_request(rdma, rqst);
+
+	mutex_unlock(&sxprt->xpt_mutex);
+
+	if (ret < 0)
+		return ret;
+	return 0;
+}
+
+static void
+xprt_rdma_bc_close(struct rpc_xprt *xprt)
+{
+	dprintk("svcrdma: %s: xprt %p\n", __func__, xprt);
+}
+
+static void
+xprt_rdma_bc_put(struct rpc_xprt *xprt)
+{
+	dprintk("svcrdma: %s: xprt %p\n", __func__, xprt);
+
+	xprt_free(xprt);
+	module_put(THIS_MODULE);
+}
+
+static struct rpc_xprt_ops xprt_rdma_bc_procs = {
+	.reserve_xprt		= xprt_reserve_xprt_cong,
+	.release_xprt		= xprt_release_xprt_cong,
+	.alloc_slot		= xprt_alloc_slot,
+	.release_request	= xprt_release_rqst_cong,
+	.buf_alloc		= xprt_rdma_bc_allocate,
+	.buf_free		= xprt_rdma_bc_free,
+	.send_request		= xprt_rdma_bc_send_request,
+	.set_retrans_timeout	= xprt_set_retrans_timeout_def,
+	.close			= xprt_rdma_bc_close,
+	.destroy		= xprt_rdma_bc_put,
+	.print_stats		= xprt_rdma_print_stats
+};
+
+static const struct rpc_timeout xprt_rdma_bc_timeout = {
+	.to_initval = 60 * HZ,
+	.to_maxval = 60 * HZ,
+};
+
+/* It shouldn't matter if the number of backchannel session slots
+ * doesn't match the number of RPC/RDMA credits. That just means
+ * one or the other will have extra slots that aren't used.
+ */
+static struct rpc_xprt *
+xprt_setup_rdma_bc(struct xprt_create *args)
+{
+	struct rpc_xprt *xprt;
+	struct rpcrdma_xprt *new_xprt;
+
+	if (args->addrlen > sizeof(xprt->addr)) {
+		dprintk("RPC:       %s: address too large\n", __func__);
+		return ERR_PTR(-EBADF);
+	}
+
+	xprt = xprt_alloc(args->net, sizeof(*new_xprt),
+			  RPCRDMA_MAX_BC_REQUESTS,
+			  RPCRDMA_MAX_BC_REQUESTS);
+	if (!xprt) {
+		dprintk("RPC:       %s: couldn't allocate rpc_xprt\n",
+			__func__);
+		return ERR_PTR(-ENOMEM);
+	}
+
+	xprt->timeout = &xprt_rdma_bc_timeout;
+	xprt_set_bound(xprt);
+	xprt_set_connected(xprt);
+	xprt->bind_timeout = RPCRDMA_BIND_TO;
+	xprt->reestablish_timeout = RPCRDMA_INIT_REEST_TO;
+	xprt->idle_timeout = RPCRDMA_IDLE_DISC_TO;
+
+	xprt->prot = XPRT_TRANSPORT_BC_RDMA;
+	xprt->tsh_size = RPCRDMA_HDRLEN_MIN / sizeof(__be32);
+	xprt->ops = &xprt_rdma_bc_procs;
+
+	memcpy(&xprt->addr, args->dstaddr, args->addrlen);
+	xprt->addrlen = args->addrlen;
+	xprt_rdma_format_addresses(xprt, (struct sockaddr *)&xprt->addr);
+	xprt->resvport = 0;
+
+	xprt->max_payload = xprt_rdma_max_inline_read;
+
+	new_xprt = rpcx_to_rdmax(xprt);
+	new_xprt->rx_buf.rb_bc_max_requests = xprt->max_reqs;
+
+	xprt_get(xprt);
+	args->bc_xprt->xpt_bc_xprt = xprt;
+	xprt->bc_xprt = args->bc_xprt;
+
+	if (!try_module_get(THIS_MODULE))
+		goto out_fail;
+
+	/* Final put for backchannel xprt is in __svc_rdma_free */
+	xprt_get(xprt);
+	return xprt;
+
+out_fail:
+	xprt_rdma_free_addresses(xprt);
+	args->bc_xprt->xpt_bc_xprt = NULL;
+	xprt_put(xprt);
+	xprt_free(xprt);
+	return ERR_PTR(-EINVAL);
+}
+
+struct xprt_class xprt_rdma_bc = {
+	.list			= LIST_HEAD_INIT(xprt_rdma_bc.list),
+	.name			= "rdma backchannel",
+	.owner			= THIS_MODULE,
+	.ident			= XPRT_TRANSPORT_BC_RDMA,
+	.setup			= xprt_setup_rdma_bc,
+};
diff --git a/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c b/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c
index ff4f01e..3dfe464 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c
@@ -567,6 +567,38 @@ static int rdma_read_complete(struct svc_rqst *rqstp,
 	return ret;
 }
 
+/* By convention, backchannel calls arrive via rdma_msg type
+ * messages, and never populate the chunk lists. This makes
+ * the RPC/RDMA header small and fixed in size, so it is
+ * straightforward to check the RPC header's direction field.
+ */
+static bool
+svc_rdma_is_backchannel_reply(struct svc_xprt *xprt, struct rpcrdma_msg *rmsgp)
+{
+	__be32 *p = (__be32 *)rmsgp;
+
+	if (!xprt->xpt_bc_xprt)
+		return false;
+
+	if (rmsgp->rm_type != rdma_msg)
+		return false;
+	if (rmsgp->rm_body.rm_chunks[0] != xdr_zero)
+		return false;
+	if (rmsgp->rm_body.rm_chunks[1] != xdr_zero)
+		return false;
+	if (rmsgp->rm_body.rm_chunks[2] != xdr_zero)
+		return false;
+
+	/* sanity */
+	if (p[7] != rmsgp->rm_xid)
+		return false;
+	/* call direction */
+	if (p[8] == cpu_to_be32(RPC_CALL))
+		return false;
+
+	return true;
+}
+
 /*
  * Set up the rqstp thread context to point to the RQ buffer. If
  * necessary, pull additional data from the client with an RDMA_READ
@@ -632,6 +664,15 @@ int svc_rdma_recvfrom(struct svc_rqst *rqstp)
 		goto close_out;
 	}
 
+	if (svc_rdma_is_backchannel_reply(xprt, rmsgp)) {
+		ret = svc_rdma_handle_bc_reply(xprt->xpt_bc_xprt, rmsgp,
+					       &rqstp->rq_arg);
+		svc_rdma_put_context(ctxt, 0);
+		if (ret)
+			goto repost;
+		return ret;
+	}
+
 	/* Read read-list data. */
 	ret = rdma_read_chunks(rdma_xprt, rmsgp, rqstp, ctxt);
 	if (ret > 0) {
@@ -668,4 +709,15 @@ int svc_rdma_recvfrom(struct svc_rqst *rqstp)
 	set_bit(XPT_CLOSE, &xprt->xpt_flags);
 defer:
 	return 0;
+
+repost:
+	ret = svc_rdma_post_recv(rdma_xprt, GFP_KERNEL);
+	if (ret) {
+		pr_err("svcrdma: could not post a receive buffer, err=%d.\n",
+		       ret);
+		pr_err("svcrdma: closing transport %p.\n", rdma_xprt);
+		set_bit(XPT_CLOSE, &rdma_xprt->sc_xprt.xpt_flags);
+		ret = -ENOTCONN;
+	}
+	return ret;
 }
diff --git a/net/sunrpc/xprtrdma/svc_rdma_transport.c b/net/sunrpc/xprtrdma/svc_rdma_transport.c
index af86dfe..7fd2395 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_transport.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_transport.c
@@ -1287,12 +1287,14 @@ static void __svc_rdma_free(struct work_struct *work)
 {
 	struct svcxprt_rdma *rdma =
 		container_of(work, struct svcxprt_rdma, sc_work);
-	dprintk("svcrdma: svc_rdma_free(%p)\n", rdma);
+	struct svc_xprt *xprt = &rdma->sc_xprt;
+
+	dprintk("svcrdma: %s(%p)\n", __func__, rdma);
 
 	/* We should only be called from kref_put */
-	if (atomic_read(&rdma->sc_xprt.xpt_ref.refcount) != 0)
+	if (atomic_read(&xprt->xpt_ref.refcount) != 0)
 		pr_err("svcrdma: sc_xprt still in use? (%d)\n",
-		       atomic_read(&rdma->sc_xprt.xpt_ref.refcount));
+		       atomic_read(&xprt->xpt_ref.refcount));
 
 	/*
 	 * Destroy queued, but not processed read completions. Note
@@ -1327,6 +1329,12 @@ static void __svc_rdma_free(struct work_struct *work)
 		pr_err("svcrdma: dma still in use? (%d)\n",
 		       atomic_read(&rdma->sc_dma_used));
 
+	/* Final put of backchannel client transport */
+	if (xprt->xpt_bc_xprt) {
+		xprt_put(xprt->xpt_bc_xprt);
+		xprt->xpt_bc_xprt = NULL;
+	}
+
 	rdma_dealloc_frmr_q(rdma);
 	svc_rdma_destroy_ctxts(rdma);
 	svc_rdma_destroy_maps(rdma);
diff --git a/net/sunrpc/xprtrdma/transport.c b/net/sunrpc/xprtrdma/transport.c
index 8c545f7..5c7d235 100644
--- a/net/sunrpc/xprtrdma/transport.c
+++ b/net/sunrpc/xprtrdma/transport.c
@@ -63,7 +63,7 @@
  */
 
 static unsigned int xprt_rdma_slot_table_entries = RPCRDMA_DEF_SLOT_TABLE;
-static unsigned int xprt_rdma_max_inline_read = RPCRDMA_DEF_INLINE;
+unsigned int xprt_rdma_max_inline_read = RPCRDMA_DEF_INLINE;
 static unsigned int xprt_rdma_max_inline_write = RPCRDMA_DEF_INLINE;
 static unsigned int xprt_rdma_inline_write_padding;
 static unsigned int xprt_rdma_memreg_strategy = RPCRDMA_FRMR;
@@ -143,12 +143,7 @@ static struct ctl_table sunrpc_table[] = {
 
 #endif
 
-#define RPCRDMA_BIND_TO		(60U * HZ)
-#define RPCRDMA_INIT_REEST_TO	(5U * HZ)
-#define RPCRDMA_MAX_REEST_TO	(30U * HZ)
-#define RPCRDMA_IDLE_DISC_TO	(5U * 60 * HZ)
-
-static struct rpc_xprt_ops xprt_rdma_procs;	/* forward reference */
+static struct rpc_xprt_ops xprt_rdma_procs;	/*forward reference */
 
 static void
 xprt_rdma_format_addresses4(struct rpc_xprt *xprt, struct sockaddr *sap)
@@ -174,7 +169,7 @@ xprt_rdma_format_addresses6(struct rpc_xprt *xprt, struct sockaddr *sap)
 	xprt->address_strings[RPC_DISPLAY_NETID] = RPCBIND_NETID_RDMA6;
 }
 
-static void
+void
 xprt_rdma_format_addresses(struct rpc_xprt *xprt, struct sockaddr *sap)
 {
 	char buf[128];
@@ -203,7 +198,7 @@ xprt_rdma_format_addresses(struct rpc_xprt *xprt, struct sockaddr *sap)
 	xprt->address_strings[RPC_DISPLAY_PROTO] = "rdma";
 }
 
-static void
+void
 xprt_rdma_free_addresses(struct rpc_xprt *xprt)
 {
 	unsigned int i;
@@ -499,7 +494,7 @@ xprt_rdma_allocate(struct rpc_task *task, size_t size)
 	if (req == NULL)
 		return NULL;
 
-	flags = GFP_NOIO | __GFP_NOWARN;
+	flags = RPCRDMA_DEF_GFP;
 	if (RPC_IS_SWAPPER(task))
 		flags = __GFP_MEMALLOC | GFP_NOWAIT | __GFP_NOWARN;
 
@@ -639,7 +634,7 @@ drop_connection:
 	return -ENOTCONN;	/* implies disconnect */
 }
 
-static void xprt_rdma_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
+void xprt_rdma_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
 {
 	struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
 	long idle_time = 0;
@@ -740,6 +735,11 @@ void xprt_rdma_cleanup(void)
 
 	rpcrdma_destroy_wq();
 	frwr_destroy_recovery_wq();
+
+	rc = xprt_unregister_transport(&xprt_rdma_bc);
+	if (rc)
+		dprintk("RPC:       %s: xprt_unregister(bc) returned %i\n",
+			__func__, rc);
 }
 
 int xprt_rdma_init(void)
@@ -763,6 +763,14 @@ int xprt_rdma_init(void)
 		return rc;
 	}
 
+	rc = xprt_register_transport(&xprt_rdma_bc);
+	if (rc) {
+		xprt_unregister_transport(&xprt_rdma);
+		rpcrdma_destroy_wq();
+		frwr_destroy_recovery_wq();
+		return rc;
+	}
+
 	dprintk("RPCRDMA Module Init, register RPC RDMA transport\n");
 
 	dprintk("Defaults:\n");
diff --git a/net/sunrpc/xprtrdma/xprt_rdma.h b/net/sunrpc/xprtrdma/xprt_rdma.h
index 72276c7..5a38236 100644
--- a/net/sunrpc/xprtrdma/xprt_rdma.h
+++ b/net/sunrpc/xprtrdma/xprt_rdma.h
@@ -55,6 +55,11 @@
 #define RDMA_RESOLVE_TIMEOUT	(5000)	/* 5 seconds */
 #define RDMA_CONNECT_RETRY_MAX	(2)	/* retries if no listener backlog */
 
+#define RPCRDMA_BIND_TO		(60U * HZ)
+#define RPCRDMA_INIT_REEST_TO	(5U * HZ)
+#define RPCRDMA_MAX_REEST_TO	(30U * HZ)
+#define RPCRDMA_IDLE_DISC_TO	(5U * 60 * HZ)
+
 /*
  * Interface Adapter -- one per transport instance
  */
@@ -147,6 +152,8 @@ rdmab_to_msg(struct rpcrdma_regbuf *rb)
 	return (struct rpcrdma_msg *)rb->rg_base;
 }
 
+#define RPCRDMA_DEF_GFP		(GFP_NOIO | __GFP_NOWARN)
+
 /*
  * struct rpcrdma_rep -- this structure encapsulates state required to recv
  * and complete a reply, asychronously. It needs several pieces of
@@ -308,6 +315,8 @@ struct rpcrdma_buffer {
 	u32			rb_bc_srv_max_requests;
 	spinlock_t		rb_reqslock;	/* protect rb_allreqs */
 	struct list_head	rb_allreqs;
+
+	u32			rb_bc_max_requests;
 };
 #define rdmab_to_ia(b) (&container_of((b), struct rpcrdma_xprt, rx_buf)->rx_ia)
 
@@ -513,6 +522,10 @@ int rpcrdma_marshal_req(struct rpc_rqst *);
 
 /* RPC/RDMA module init - xprtrdma/transport.c
  */
+extern unsigned int xprt_rdma_max_inline_read;
+void xprt_rdma_format_addresses(struct rpc_xprt *xprt, struct sockaddr *sap);
+void xprt_rdma_free_addresses(struct rpc_xprt *xprt);
+void xprt_rdma_print_stats(struct rpc_xprt *xprt, struct seq_file *seq);
 int xprt_rdma_init(void);
 void xprt_rdma_cleanup(void);
 
@@ -528,4 +541,6 @@ void xprt_rdma_bc_free_rqst(struct rpc_rqst *);
 void xprt_rdma_bc_destroy(struct rpc_xprt *, unsigned int);
 #endif	/* CONFIG_SUNRPC_BACKCHANNEL */
 
+extern struct xprt_class xprt_rdma_bc;
+
 #endif				/* _LINUX_SUNRPC_XPRT_RDMA_H */

[PATCH v5 11/10] svc_rdma: use local_dma_lkey

[Christoph Hellwig]

We now alwasy have a per-PD local_dma_lkey available.  Make use of that
fact in svc_rdma and stop registering our own MR.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagig@mellanox.com>
Reviewed-by: Jason Gunthorpe <jgunthorpe@obsidianresearch.com>
Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
Reviewed-by: Steve Wise <swise@opengridcomputing.com>
---
 include/linux/sunrpc/svc_rdma.h            |  2 --
 net/sunrpc/xprtrdma/svc_rdma_backchannel.c |  2 +-
 net/sunrpc/xprtrdma/svc_rdma_recvfrom.c    |  4 ++--
 net/sunrpc/xprtrdma/svc_rdma_sendto.c      |  6 ++---
 net/sunrpc/xprtrdma/svc_rdma_transport.c   | 36 ++++--------------------------
 5 files changed, 10 insertions(+), 40 deletions(-)

diff --git a/include/linux/sunrpc/svc_rdma.h b/include/linux/sunrpc/svc_rdma.h
index b13513a..5322fea 100644
--- a/include/linux/sunrpc/svc_rdma.h
+++ b/include/linux/sunrpc/svc_rdma.h
@@ -156,13 +156,11 @@ struct svcxprt_rdma {
 	struct ib_qp         *sc_qp;
 	struct ib_cq         *sc_rq_cq;
 	struct ib_cq         *sc_sq_cq;
-	struct ib_mr         *sc_phys_mr;	/* MR for server memory */
 	int		     (*sc_reader)(struct svcxprt_rdma *,
 					  struct svc_rqst *,
 					  struct svc_rdma_op_ctxt *,
 					  int *, u32 *, u32, u32, u64, bool);
 	u32		     sc_dev_caps;	/* distilled device caps */
-	u32		     sc_dma_lkey;	/* local dma key */
 	unsigned int	     sc_frmr_pg_list_len;
 	struct list_head     sc_frmr_q;
 	spinlock_t	     sc_frmr_q_lock;
diff --git a/net/sunrpc/xprtrdma/svc_rdma_backchannel.c b/net/sunrpc/xprtrdma/svc_rdma_backchannel.c
index deff06a..65a7c23 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_backchannel.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_backchannel.c
@@ -128,7 +128,7 @@ static int svc_rdma_bc_sendto(struct svcxprt_rdma *rdma,
 
 	ctxt->wr_op = IB_WR_SEND;
 	ctxt->direction = DMA_TO_DEVICE;
-	ctxt->sge[0].lkey = rdma->sc_dma_lkey;
+	ctxt->sge[0].lkey = rdma->sc_pd->local_dma_lkey;
 	ctxt->sge[0].length = sndbuf->len;
 	ctxt->sge[0].addr =
 	    ib_dma_map_page(rdma->sc_cm_id->device, ctxt->pages[0], 0,
diff --git a/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c b/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c
index 3dfe464..c8b8a8b 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c
@@ -144,6 +144,7 @@ int rdma_read_chunk_lcl(struct svcxprt_rdma *xprt,
 
 		head->arg.pages[pg_no] = rqstp->rq_arg.pages[pg_no];
 		head->arg.page_len += len;
+
 		head->arg.len += len;
 		if (!pg_off)
 			head->count++;
@@ -160,8 +161,7 @@ int rdma_read_chunk_lcl(struct svcxprt_rdma *xprt,
 			goto err;
 		atomic_inc(&xprt->sc_dma_used);
 
-		/* The lkey here is either a local dma lkey or a dma_mr lkey */
-		ctxt->sge[pno].lkey = xprt->sc_dma_lkey;
+		ctxt->sge[pno].lkey = xprt->sc_pd->local_dma_lkey;
 		ctxt->sge[pno].length = len;
 		ctxt->count++;
 
diff --git a/net/sunrpc/xprtrdma/svc_rdma_sendto.c b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
index 3c25052..df57f3c 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_sendto.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
@@ -265,7 +265,7 @@ static int send_write(struct svcxprt_rdma *xprt, struct svc_rqst *rqstp,
 					 sge[sge_no].addr))
 			goto err;
 		atomic_inc(&xprt->sc_dma_used);
-		sge[sge_no].lkey = xprt->sc_dma_lkey;
+		sge[sge_no].lkey = xprt->sc_pd->local_dma_lkey;
 		ctxt->count++;
 		sge_off = 0;
 		sge_no++;
@@ -480,7 +480,7 @@ static int send_reply(struct svcxprt_rdma *rdma,
 	ctxt->count = 1;
 
 	/* Prepare the SGE for the RPCRDMA Header */
-	ctxt->sge[0].lkey = rdma->sc_dma_lkey;
+	ctxt->sge[0].lkey = rdma->sc_pd->local_dma_lkey;
 	ctxt->sge[0].length = svc_rdma_xdr_get_reply_hdr_len(rdma_resp);
 	ctxt->sge[0].addr =
 	    ib_dma_map_page(rdma->sc_cm_id->device, page, 0,
@@ -504,7 +504,7 @@ static int send_reply(struct svcxprt_rdma *rdma,
 					 ctxt->sge[sge_no].addr))
 			goto err;
 		atomic_inc(&rdma->sc_dma_used);
-		ctxt->sge[sge_no].lkey = rdma->sc_dma_lkey;
+		ctxt->sge[sge_no].lkey = rdma->sc_pd->local_dma_lkey;
 		ctxt->sge[sge_no].length = sge_bytes;
 	}
 	if (byte_count != 0) {
diff --git a/net/sunrpc/xprtrdma/svc_rdma_transport.c b/net/sunrpc/xprtrdma/svc_rdma_transport.c
index 7fd2395..5763825 100644
--- a/net/sunrpc/xprtrdma/svc_rdma_transport.c
+++ b/net/sunrpc/xprtrdma/svc_rdma_transport.c
@@ -232,11 +232,11 @@ void svc_rdma_unmap_dma(struct svc_rdma_op_ctxt *ctxt)
 	for (i = 0; i < ctxt->count && ctxt->sge[i].length; i++) {
 		/*
 		 * Unmap the DMA addr in the SGE if the lkey matches
-		 * the sc_dma_lkey, otherwise, ignore it since it is
+		 * the local_dma_lkey, otherwise, ignore it since it is
 		 * an FRMR lkey and will be unmapped later when the
 		 * last WR that uses it completes.
 		 */
-		if (ctxt->sge[i].lkey == xprt->sc_dma_lkey) {
+		if (ctxt->sge[i].lkey == xprt->sc_pd->local_dma_lkey) {
 			atomic_dec(&xprt->sc_dma_used);
 			ib_dma_unmap_page(xprt->sc_cm_id->device,
 					    ctxt->sge[i].addr,
@@ -698,7 +698,7 @@ int svc_rdma_post_recv(struct svcxprt_rdma *xprt, gfp_t flags)
 		atomic_inc(&xprt->sc_dma_used);
 		ctxt->sge[sge_no].addr = pa;
 		ctxt->sge[sge_no].length = PAGE_SIZE;
-		ctxt->sge[sge_no].lkey = xprt->sc_dma_lkey;
+		ctxt->sge[sge_no].lkey = xprt->sc_pd->local_dma_lkey;
 		ctxt->count = sge_no + 1;
 		buflen += PAGE_SIZE;
 	}
@@ -1014,8 +1014,6 @@ static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt)
 	struct ib_cq_init_attr cq_attr = {};
 	struct ib_qp_init_attr qp_attr;
 	struct ib_device *dev;
-	int uninitialized_var(dma_mr_acc);
-	int need_dma_mr = 0;
 	unsigned int i;
 	int ret = 0;
 
@@ -1160,32 +1158,9 @@ static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt)
 	    !rdma_ib_or_roce(dev, newxprt->sc_cm_id->port_num))
 		goto errout;
 
-	if (!(newxprt->sc_dev_caps & SVCRDMA_DEVCAP_FAST_REG) ||
-	    !(dev->attrs.device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY)) {
-		need_dma_mr = 1;
-		dma_mr_acc = IB_ACCESS_LOCAL_WRITE;
-		if (rdma_protocol_iwarp(dev, newxprt->sc_cm_id->port_num) &&
-		    !(newxprt->sc_dev_caps & SVCRDMA_DEVCAP_FAST_REG))
-			dma_mr_acc |= IB_ACCESS_REMOTE_WRITE;
-	}
-
 	if (rdma_protocol_iwarp(dev, newxprt->sc_cm_id->port_num))
 		newxprt->sc_dev_caps |= SVCRDMA_DEVCAP_READ_W_INV;
 
-	/* Create the DMA MR if needed, otherwise, use the DMA LKEY */
-	if (need_dma_mr) {
-		/* Register all of physical memory */
-		newxprt->sc_phys_mr =
-			ib_get_dma_mr(newxprt->sc_pd, dma_mr_acc);
-		if (IS_ERR(newxprt->sc_phys_mr)) {
-			dprintk("svcrdma: Failed to create DMA MR ret=%d\n",
-				ret);
-			goto errout;
-		}
-		newxprt->sc_dma_lkey = newxprt->sc_phys_mr->lkey;
-	} else
-		newxprt->sc_dma_lkey = dev->local_dma_lkey;
-
 	/* Post receive buffers */
 	for (i = 0; i < newxprt->sc_rq_depth; i++) {
 		ret = svc_rdma_post_recv(newxprt, GFP_KERNEL);
@@ -1349,9 +1324,6 @@ static void __svc_rdma_free(struct work_struct *work)
 	if (rdma->sc_rq_cq && !IS_ERR(rdma->sc_rq_cq))
 		ib_destroy_cq(rdma->sc_rq_cq);
 
-	if (rdma->sc_phys_mr && !IS_ERR(rdma->sc_phys_mr))
-		ib_dereg_mr(rdma->sc_phys_mr);
-
 	if (rdma->sc_pd && !IS_ERR(rdma->sc_pd))
 		ib_dealloc_pd(rdma->sc_pd);
 
@@ -1479,7 +1451,7 @@ void svc_rdma_send_error(struct svcxprt_rdma *xprt, struct rpcrdma_msg *rmsgp,
 		return;
 	}
 	atomic_inc(&xprt->sc_dma_used);
-	ctxt->sge[0].lkey = xprt->sc_dma_lkey;
+	ctxt->sge[0].lkey = xprt->sc_pd->local_dma_lkey;
 	ctxt->sge[0].length = length;
 
 	/* Prepare SEND WR */
-- 
1.9.1

Re: [PATCH v5 11/10] svc_rdma: use local_dma_lkey

[J. Bruce Fields]

On Thu, Jan 07, 2016 at 11:53:41PM -0800, Christoph Hellwig wrote:
> We now alwasy have a per-PD local_dma_lkey available.  Make use of that
> fact in svc_rdma and stop registering our own MR.

Acked-by: J. Bruce Fields <bfields@redhat.com>

--b.

> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> Reviewed-by: Sagi Grimberg <sagig@mellanox.com>
> Reviewed-by: Jason Gunthorpe <jgunthorpe@obsidianresearch.com>
> Reviewed-by: Chuck Lever <chuck.lever@oracle.com>
> Reviewed-by: Steve Wise <swise@opengridcomputing.com>
> ---
>  include/linux/sunrpc/svc_rdma.h            |  2 --
>  net/sunrpc/xprtrdma/svc_rdma_backchannel.c |  2 +-
>  net/sunrpc/xprtrdma/svc_rdma_recvfrom.c    |  4 ++--
>  net/sunrpc/xprtrdma/svc_rdma_sendto.c      |  6 ++---
>  net/sunrpc/xprtrdma/svc_rdma_transport.c   | 36 ++++--------------------------
>  5 files changed, 10 insertions(+), 40 deletions(-)
> 
> diff --git a/include/linux/sunrpc/svc_rdma.h b/include/linux/sunrpc/svc_rdma.h
> index b13513a..5322fea 100644
> --- a/include/linux/sunrpc/svc_rdma.h
> +++ b/include/linux/sunrpc/svc_rdma.h
> @@ -156,13 +156,11 @@ struct svcxprt_rdma {
>  	struct ib_qp         *sc_qp;
>  	struct ib_cq         *sc_rq_cq;
>  	struct ib_cq         *sc_sq_cq;
> -	struct ib_mr         *sc_phys_mr;	/* MR for server memory */
>  	int		     (*sc_reader)(struct svcxprt_rdma *,
>  					  struct svc_rqst *,
>  					  struct svc_rdma_op_ctxt *,
>  					  int *, u32 *, u32, u32, u64, bool);
>  	u32		     sc_dev_caps;	/* distilled device caps */
> -	u32		     sc_dma_lkey;	/* local dma key */
>  	unsigned int	     sc_frmr_pg_list_len;
>  	struct list_head     sc_frmr_q;
>  	spinlock_t	     sc_frmr_q_lock;
> diff --git a/net/sunrpc/xprtrdma/svc_rdma_backchannel.c b/net/sunrpc/xprtrdma/svc_rdma_backchannel.c
> index deff06a..65a7c23 100644
> --- a/net/sunrpc/xprtrdma/svc_rdma_backchannel.c
> +++ b/net/sunrpc/xprtrdma/svc_rdma_backchannel.c
> @@ -128,7 +128,7 @@ static int svc_rdma_bc_sendto(struct svcxprt_rdma *rdma,
>  
>  	ctxt->wr_op = IB_WR_SEND;
>  	ctxt->direction = DMA_TO_DEVICE;
> -	ctxt->sge[0].lkey = rdma->sc_dma_lkey;
> +	ctxt->sge[0].lkey = rdma->sc_pd->local_dma_lkey;
>  	ctxt->sge[0].length = sndbuf->len;
>  	ctxt->sge[0].addr =
>  	    ib_dma_map_page(rdma->sc_cm_id->device, ctxt->pages[0], 0,
> diff --git a/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c b/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c
> index 3dfe464..c8b8a8b 100644
> --- a/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c
> +++ b/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c
> @@ -144,6 +144,7 @@ int rdma_read_chunk_lcl(struct svcxprt_rdma *xprt,
>  
>  		head->arg.pages[pg_no] = rqstp->rq_arg.pages[pg_no];
>  		head->arg.page_len += len;
> +
>  		head->arg.len += len;
>  		if (!pg_off)
>  			head->count++;
> @@ -160,8 +161,7 @@ int rdma_read_chunk_lcl(struct svcxprt_rdma *xprt,
>  			goto err;
>  		atomic_inc(&xprt->sc_dma_used);
>  
> -		/* The lkey here is either a local dma lkey or a dma_mr lkey */
> -		ctxt->sge[pno].lkey = xprt->sc_dma_lkey;
> +		ctxt->sge[pno].lkey = xprt->sc_pd->local_dma_lkey;
>  		ctxt->sge[pno].length = len;
>  		ctxt->count++;
>  
> diff --git a/net/sunrpc/xprtrdma/svc_rdma_sendto.c b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
> index 3c25052..df57f3c 100644
> --- a/net/sunrpc/xprtrdma/svc_rdma_sendto.c
> +++ b/net/sunrpc/xprtrdma/svc_rdma_sendto.c
> @@ -265,7 +265,7 @@ static int send_write(struct svcxprt_rdma *xprt, struct svc_rqst *rqstp,
>  					 sge[sge_no].addr))
>  			goto err;
>  		atomic_inc(&xprt->sc_dma_used);
> -		sge[sge_no].lkey = xprt->sc_dma_lkey;
> +		sge[sge_no].lkey = xprt->sc_pd->local_dma_lkey;
>  		ctxt->count++;
>  		sge_off = 0;
>  		sge_no++;
> @@ -480,7 +480,7 @@ static int send_reply(struct svcxprt_rdma *rdma,
>  	ctxt->count = 1;
>  
>  	/* Prepare the SGE for the RPCRDMA Header */
> -	ctxt->sge[0].lkey = rdma->sc_dma_lkey;
> +	ctxt->sge[0].lkey = rdma->sc_pd->local_dma_lkey;
>  	ctxt->sge[0].length = svc_rdma_xdr_get_reply_hdr_len(rdma_resp);
>  	ctxt->sge[0].addr =
>  	    ib_dma_map_page(rdma->sc_cm_id->device, page, 0,
> @@ -504,7 +504,7 @@ static int send_reply(struct svcxprt_rdma *rdma,
>  					 ctxt->sge[sge_no].addr))
>  			goto err;
>  		atomic_inc(&rdma->sc_dma_used);
> -		ctxt->sge[sge_no].lkey = rdma->sc_dma_lkey;
> +		ctxt->sge[sge_no].lkey = rdma->sc_pd->local_dma_lkey;
>  		ctxt->sge[sge_no].length = sge_bytes;
>  	}
>  	if (byte_count != 0) {
> diff --git a/net/sunrpc/xprtrdma/svc_rdma_transport.c b/net/sunrpc/xprtrdma/svc_rdma_transport.c
> index 7fd2395..5763825 100644
> --- a/net/sunrpc/xprtrdma/svc_rdma_transport.c
> +++ b/net/sunrpc/xprtrdma/svc_rdma_transport.c
> @@ -232,11 +232,11 @@ void svc_rdma_unmap_dma(struct svc_rdma_op_ctxt *ctxt)
>  	for (i = 0; i < ctxt->count && ctxt->sge[i].length; i++) {
>  		/*
>  		 * Unmap the DMA addr in the SGE if the lkey matches
> -		 * the sc_dma_lkey, otherwise, ignore it since it is
> +		 * the local_dma_lkey, otherwise, ignore it since it is
>  		 * an FRMR lkey and will be unmapped later when the
>  		 * last WR that uses it completes.
>  		 */
> -		if (ctxt->sge[i].lkey == xprt->sc_dma_lkey) {
> +		if (ctxt->sge[i].lkey == xprt->sc_pd->local_dma_lkey) {
>  			atomic_dec(&xprt->sc_dma_used);
>  			ib_dma_unmap_page(xprt->sc_cm_id->device,
>  					    ctxt->sge[i].addr,
> @@ -698,7 +698,7 @@ int svc_rdma_post_recv(struct svcxprt_rdma *xprt, gfp_t flags)
>  		atomic_inc(&xprt->sc_dma_used);
>  		ctxt->sge[sge_no].addr = pa;
>  		ctxt->sge[sge_no].length = PAGE_SIZE;
> -		ctxt->sge[sge_no].lkey = xprt->sc_dma_lkey;
> +		ctxt->sge[sge_no].lkey = xprt->sc_pd->local_dma_lkey;
>  		ctxt->count = sge_no + 1;
>  		buflen += PAGE_SIZE;
>  	}
> @@ -1014,8 +1014,6 @@ static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt)
>  	struct ib_cq_init_attr cq_attr = {};
>  	struct ib_qp_init_attr qp_attr;
>  	struct ib_device *dev;
> -	int uninitialized_var(dma_mr_acc);
> -	int need_dma_mr = 0;
>  	unsigned int i;
>  	int ret = 0;
>  
> @@ -1160,32 +1158,9 @@ static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt)
>  	    !rdma_ib_or_roce(dev, newxprt->sc_cm_id->port_num))
>  		goto errout;
>  
> -	if (!(newxprt->sc_dev_caps & SVCRDMA_DEVCAP_FAST_REG) ||
> -	    !(dev->attrs.device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY)) {
> -		need_dma_mr = 1;
> -		dma_mr_acc = IB_ACCESS_LOCAL_WRITE;
> -		if (rdma_protocol_iwarp(dev, newxprt->sc_cm_id->port_num) &&
> -		    !(newxprt->sc_dev_caps & SVCRDMA_DEVCAP_FAST_REG))
> -			dma_mr_acc |= IB_ACCESS_REMOTE_WRITE;
> -	}
> -
>  	if (rdma_protocol_iwarp(dev, newxprt->sc_cm_id->port_num))
>  		newxprt->sc_dev_caps |= SVCRDMA_DEVCAP_READ_W_INV;
>  
> -	/* Create the DMA MR if needed, otherwise, use the DMA LKEY */
> -	if (need_dma_mr) {
> -		/* Register all of physical memory */
> -		newxprt->sc_phys_mr =
> -			ib_get_dma_mr(newxprt->sc_pd, dma_mr_acc);
> -		if (IS_ERR(newxprt->sc_phys_mr)) {
> -			dprintk("svcrdma: Failed to create DMA MR ret=%d\n",
> -				ret);
> -			goto errout;
> -		}
> -		newxprt->sc_dma_lkey = newxprt->sc_phys_mr->lkey;
> -	} else
> -		newxprt->sc_dma_lkey = dev->local_dma_lkey;
> -
>  	/* Post receive buffers */
>  	for (i = 0; i < newxprt->sc_rq_depth; i++) {
>  		ret = svc_rdma_post_recv(newxprt, GFP_KERNEL);
> @@ -1349,9 +1324,6 @@ static void __svc_rdma_free(struct work_struct *work)
>  	if (rdma->sc_rq_cq && !IS_ERR(rdma->sc_rq_cq))
>  		ib_destroy_cq(rdma->sc_rq_cq);
>  
> -	if (rdma->sc_phys_mr && !IS_ERR(rdma->sc_phys_mr))
> -		ib_dereg_mr(rdma->sc_phys_mr);
> -
>  	if (rdma->sc_pd && !IS_ERR(rdma->sc_pd))
>  		ib_dealloc_pd(rdma->sc_pd);
>  
> @@ -1479,7 +1451,7 @@ void svc_rdma_send_error(struct svcxprt_rdma *xprt, struct rpcrdma_msg *rmsgp,
>  		return;
>  	}
>  	atomic_inc(&xprt->sc_dma_used);
> -	ctxt->sge[0].lkey = xprt->sc_dma_lkey;
> +	ctxt->sge[0].lkey = xprt->sc_pd->local_dma_lkey;
>  	ctxt->sge[0].length = length;
>  
>  	/* Prepare SEND WR */
> -- 
> 1.9.1

Re: [PATCH v5 11/10] svc_rdma: use local_dma_lkey

[Doug Ledford]

[-- Attachment #1: Type: text/plain, Size: 491 bytes --]

On 01/08/2016 09:02 AM, J. Bruce Fields wrote:
> On Thu, Jan 07, 2016 at 11:53:41PM -0800, Christoph Hellwig wrote:
>> We now alwasy have a per-PD local_dma_lkey available.  Make use of that
>> fact in svc_rdma and stop registering our own MR.
> 
> Acked-by: J. Bruce Fields <bfields@redhat.com>
> 
> --b.

Per off-list communication with Bruce, this series has been applied to
my tree.  Thanks.


-- 
Doug Ledford <dledford@redhat.com>
              GPG KeyID: 0E572FDD



[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 884 bytes --]