All of lore.kernel.org
 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 v2 05/21] xprtrdma: rpcrdma_regbuf alignment
Date: Tue, 16 Apr 2019 09:37:17 -0400	[thread overview]
Message-ID: <20190416133717.23113.38965.stgit@manet.1015granger.net> (raw)
In-Reply-To: <20190416133156.23113.91846.stgit@manet.1015granger.net>

Allocate the struct rpcrdma_regbuf separately from the I/O buffer
to better guarantee the alignment of the I/O buffer and eliminate
the wasted space between the rpcrdma_regbuf metadata and the buffer
itself.

Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
---
 net/sunrpc/xprtrdma/backchannel.c |    6 +++---
 net/sunrpc/xprtrdma/rpc_rdma.c    |    4 ++--
 net/sunrpc/xprtrdma/transport.c   |    8 ++++----
 net/sunrpc/xprtrdma/verbs.c       |   27 ++++++++++++++++-----------
 net/sunrpc/xprtrdma/xprt_rdma.h   |   19 ++++++++++---------
 5 files changed, 35 insertions(+), 29 deletions(-)

diff --git a/net/sunrpc/xprtrdma/backchannel.c b/net/sunrpc/xprtrdma/backchannel.c
index 713961a..6170ec7 100644
--- a/net/sunrpc/xprtrdma/backchannel.c
+++ b/net/sunrpc/xprtrdma/backchannel.c
@@ -45,10 +45,10 @@ static int rpcrdma_bc_setup_reqs(struct rpcrdma_xprt *r_xprt,
 
 		size = r_xprt->rx_data.inline_rsize;
 		rb = rpcrdma_alloc_regbuf(size, DMA_TO_DEVICE, GFP_KERNEL);
-		if (IS_ERR(rb))
+		if (!rb)
 			goto out_fail;
 		req->rl_sendbuf = rb;
-		xdr_buf_init(&rqst->rq_snd_buf, rb->rg_base,
+		xdr_buf_init(&rqst->rq_snd_buf, rdmab_data(rb),
 			     min_t(size_t, size, PAGE_SIZE));
 	}
 	return 0;
@@ -123,7 +123,7 @@ static int rpcrdma_bc_marshal_reply(struct rpc_rqst *rqst)
 
 	rpcrdma_set_xdrlen(&req->rl_hdrbuf, 0);
 	xdr_init_encode(&req->rl_stream, &req->rl_hdrbuf,
-			req->rl_rdmabuf->rg_base, rqst);
+			rdmab_data(req->rl_rdmabuf), rqst);
 
 	p = xdr_reserve_space(&req->rl_stream, 28);
 	if (unlikely(!p))
diff --git a/net/sunrpc/xprtrdma/rpc_rdma.c b/net/sunrpc/xprtrdma/rpc_rdma.c
index b759b16..cf99c55 100644
--- a/net/sunrpc/xprtrdma/rpc_rdma.c
+++ b/net/sunrpc/xprtrdma/rpc_rdma.c
@@ -747,8 +747,8 @@ static bool rpcrdma_results_inline(struct rpcrdma_xprt *r_xprt,
 	int ret;
 
 	rpcrdma_set_xdrlen(&req->rl_hdrbuf, 0);
-	xdr_init_encode(xdr, &req->rl_hdrbuf,
-			req->rl_rdmabuf->rg_base, rqst);
+	xdr_init_encode(xdr, &req->rl_hdrbuf, rdmab_data(req->rl_rdmabuf),
+			rqst);
 
 	/* Fixed header fields */
 	ret = -EMSGSIZE;
diff --git a/net/sunrpc/xprtrdma/transport.c b/net/sunrpc/xprtrdma/transport.c
index 5d26135..e3b5b91 100644
--- a/net/sunrpc/xprtrdma/transport.c
+++ b/net/sunrpc/xprtrdma/transport.c
@@ -595,7 +595,7 @@ void xprt_rdma_close(struct rpc_xprt *xprt)
 		return true;
 
 	rb = rpcrdma_alloc_regbuf(size, DMA_TO_DEVICE, flags);
-	if (IS_ERR(rb))
+	if (!rb)
 		return false;
 
 	rpcrdma_free_regbuf(req->rl_sendbuf);
@@ -625,7 +625,7 @@ void xprt_rdma_close(struct rpc_xprt *xprt)
 		return true;
 
 	rb = rpcrdma_alloc_regbuf(size, DMA_NONE, flags);
-	if (IS_ERR(rb))
+	if (!rb)
 		return false;
 
 	rpcrdma_free_regbuf(req->rl_recvbuf);
@@ -660,8 +660,8 @@ void xprt_rdma_close(struct rpc_xprt *xprt)
 	if (!rpcrdma_get_recvbuf(r_xprt, req, rqst->rq_rcvsize, flags))
 		goto out_fail;
 
-	rqst->rq_buffer = req->rl_sendbuf->rg_base;
-	rqst->rq_rbuffer = req->rl_recvbuf->rg_base;
+	rqst->rq_buffer = rdmab_data(req->rl_sendbuf);
+	rqst->rq_rbuffer = rdmab_data(req->rl_recvbuf);
 	trace_xprtrdma_op_allocate(task, req);
 	return 0;
 
diff --git a/net/sunrpc/xprtrdma/verbs.c b/net/sunrpc/xprtrdma/verbs.c
index caa6a5d..f88fd39 100644
--- a/net/sunrpc/xprtrdma/verbs.c
+++ b/net/sunrpc/xprtrdma/verbs.c
@@ -1013,12 +1013,12 @@ struct rpcrdma_req *rpcrdma_req_create(struct rpcrdma_xprt *r_xprt, gfp_t flags)
 		return NULL;
 
 	rb = rpcrdma_alloc_regbuf(RPCRDMA_HDRBUF_SIZE, DMA_TO_DEVICE, flags);
-	if (IS_ERR(rb)) {
+	if (!rb) {
 		kfree(req);
 		return NULL;
 	}
 	req->rl_rdmabuf = rb;
-	xdr_buf_init(&req->rl_hdrbuf, rb->rg_base, rdmab_length(rb));
+	xdr_buf_init(&req->rl_hdrbuf, rdmab_data(rb), rdmab_length(rb));
 	req->rl_buffer = buffer;
 	INIT_LIST_HEAD(&req->rl_registered);
 
@@ -1040,9 +1040,9 @@ static bool rpcrdma_rep_create(struct rpcrdma_xprt *r_xprt, bool temp)
 
 	rep->rr_rdmabuf = rpcrdma_alloc_regbuf(cdata->inline_rsize,
 					       DMA_FROM_DEVICE, GFP_KERNEL);
-	if (IS_ERR(rep->rr_rdmabuf))
+	if (!rep->rr_rdmabuf)
 		goto out_free;
-	xdr_buf_init(&rep->rr_hdrbuf, rep->rr_rdmabuf->rg_base,
+	xdr_buf_init(&rep->rr_hdrbuf, rdmab_data(rep->rr_rdmabuf),
 		     rdmab_length(rep->rr_rdmabuf));
 
 	rep->rr_cqe.done = rpcrdma_wc_receive;
@@ -1354,8 +1354,7 @@ struct rpcrdma_req *
  * @direction: direction of data movement
  * @flags: GFP flags
  *
- * Returns an ERR_PTR, or a pointer to a regbuf, a buffer that
- * can be persistently DMA-mapped for I/O.
+ * Returns a pointer to a rpcrdma_regbuf object, or NULL.
  *
  * xprtrdma uses a regbuf for posting an outgoing RDMA SEND, or for
  * receiving the payload of RDMA RECV operations. During Long Calls
@@ -1367,14 +1366,18 @@ struct rpcrdma_regbuf *
 {
 	struct rpcrdma_regbuf *rb;
 
-	rb = kmalloc(sizeof(*rb) + size, flags);
-	if (rb == NULL)
-		return ERR_PTR(-ENOMEM);
+	rb = kmalloc(sizeof(*rb), flags);
+	if (!rb)
+		return NULL;
+	rb->rg_data = kmalloc(size, flags);
+	if (!rb->rg_data) {
+		kfree(rb);
+		return NULL;
+	}
 
 	rb->rg_device = NULL;
 	rb->rg_direction = direction;
 	rb->rg_iov.length = size;
-
 	return rb;
 }
 
@@ -1392,7 +1395,7 @@ struct rpcrdma_regbuf *
 		return false;
 
 	rb->rg_iov.addr = ib_dma_map_single(device,
-					    (void *)rb->rg_base,
+					    rdmab_data(rb),
 					    rdmab_length(rb),
 					    rb->rg_direction);
 	if (ib_dma_mapping_error(device, rdmab_addr(rb))) {
@@ -1427,6 +1430,8 @@ struct rpcrdma_regbuf *
 rpcrdma_free_regbuf(struct rpcrdma_regbuf *rb)
 {
 	rpcrdma_dma_unmap_regbuf(rb);
+	if (rb)
+		kfree(rb->rg_data);
 	kfree(rb);
 }
 
diff --git a/net/sunrpc/xprtrdma/xprt_rdma.h b/net/sunrpc/xprtrdma/xprt_rdma.h
index 539558f..1af9674 100644
--- a/net/sunrpc/xprtrdma/xprt_rdma.h
+++ b/net/sunrpc/xprtrdma/xprt_rdma.h
@@ -121,33 +121,34 @@ struct rpcrdma_regbuf {
 	struct ib_sge		rg_iov;
 	struct ib_device	*rg_device;
 	enum dma_data_direction	rg_direction;
-	__be32			rg_base[0] __attribute__ ((aligned(256)));
+	void			*rg_data;
 };
 
-static inline u64
-rdmab_addr(struct rpcrdma_regbuf *rb)
+static inline u64 rdmab_addr(struct rpcrdma_regbuf *rb)
 {
 	return rb->rg_iov.addr;
 }
 
-static inline u32
-rdmab_length(struct rpcrdma_regbuf *rb)
+static inline u32 rdmab_length(struct rpcrdma_regbuf *rb)
 {
 	return rb->rg_iov.length;
 }
 
-static inline u32
-rdmab_lkey(struct rpcrdma_regbuf *rb)
+static inline u32 rdmab_lkey(struct rpcrdma_regbuf *rb)
 {
 	return rb->rg_iov.lkey;
 }
 
-static inline struct ib_device *
-rdmab_device(struct rpcrdma_regbuf *rb)
+static inline struct ib_device *rdmab_device(struct rpcrdma_regbuf *rb)
 {
 	return rb->rg_device;
 }
 
+static inline void *rdmab_data(const struct rpcrdma_regbuf *rb)
+{
+	return rb->rg_data;
+}
+
 #define RPCRDMA_DEF_GFP		(GFP_NOIO | __GFP_NOWARN)
 
 /* To ensure a transport can always make forward progress,


  parent reply	other threads:[~2019-04-16 13:37 UTC|newest]

Thread overview: 22+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2019-04-16 13:36 [PATCH v2 00/21] Proposed NFS/RDMA patches for v5.2 Chuck Lever
2019-04-16 13:36 ` [PATCH v2 01/21] SUNRPC: Avoid digging into the ATOMIC pool Chuck Lever
2019-04-16 13:37 ` [PATCH v2 02/21] xprtrdma: Fix an frwr_map recovery nit Chuck Lever
2019-04-16 13:37 ` [PATCH v2 03/21] xprtrdma: Clean up rpcrdma_create_req() Chuck Lever
2019-04-16 13:37 ` [PATCH v2 04/21] xprtrdma: Clean up rpcrdma_create_rep() and rpcrdma_destroy_rep() Chuck Lever
2019-04-16 13:37 ` Chuck Lever [this message]
2019-04-16 13:37 ` [PATCH v2 06/21] xprtrdma: Allocate req's regbufs at xprt create time Chuck Lever
2019-04-16 13:37 ` [PATCH v2 07/21] xprtrdma: De-duplicate "allocate new, free old regbuf" Chuck Lever
2019-04-16 13:37 ` [PATCH v2 08/21] xprtrdma: Clean up regbuf helpers Chuck Lever
2019-04-16 13:37 ` [PATCH v2 09/21] xprtrdma: Backchannel can use GFP_KERNEL allocations Chuck Lever
2019-04-16 13:37 ` [PATCH v2 10/21] xprtrdma: Increase maximum number of backchannel request Chuck Lever
2019-04-16 13:37 ` [PATCH v2 11/21] xprtrdma: Trace marshaling failures Chuck Lever
2019-04-16 13:37 ` [PATCH v2 12/21] xprtrdma: Clean up sendctx functions Chuck Lever
2019-04-16 13:38 ` [PATCH v2 13/21] xprtrdma: More Send completion batching Chuck Lever
2019-04-16 13:38 ` [PATCH v2 14/21] xprtrdma: Eliminate rpcrdma_ia::ri_device Chuck Lever
2019-04-16 13:38 ` [PATCH v2 15/21] SUNRPC: Update comments based on recent changes Chuck Lever
2019-04-16 13:38 ` [PATCH v2 16/21] xprtrdma: Remove rpcrdma_create_data_internal::rsize and wsize Chuck Lever
2019-04-16 13:38 ` [PATCH v2 17/21] xprtrdma: Aggregate the inline settings in struct rpcrdma_ep Chuck Lever
2019-04-16 13:38 ` [PATCH v2 18/21] xprtrdma: Eliminate struct rpcrdma_create_data_internal Chuck Lever
2019-04-16 13:38 ` [PATCH v2 19/21] xprtrdma: Remove pr_err() call sites from completion handlers Chuck Lever
2019-04-16 13:38 ` [PATCH v2 20/21] xprtrdma: Update comments that reference ib_drain_qp Chuck Lever
2019-04-16 13:38 ` [PATCH v2 21/21] xprtrdma: Remove stale comment 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=20190416133717.23113.38965.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 an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.