[PATCH v4 06/10] xprtrdma: Add ro_unmap_sync method for FRWR

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



FRWR's ro_unmap is asynchronous. The new ro_unmap_sync posts
LOCAL_INV Work Requests and waits for them to complete before
returning.

Note also, DMA unmapping is now done _after_ invalidation.

Signed-off-by: Chuck Lever <chuck.lever@xxxxxxxxxx>
Tested-by: Devesh Sharma <devesh.sharma@xxxxxxxxxxxxx>
---
 net/sunrpc/xprtrdma/frwr_ops.c  |  136 ++++++++++++++++++++++++++++++++++++++-
 net/sunrpc/xprtrdma/xprt_rdma.h |    2 +
 2 files changed, 134 insertions(+), 4 deletions(-)

diff --git a/net/sunrpc/xprtrdma/frwr_ops.c b/net/sunrpc/xprtrdma/frwr_ops.c
index 660d0b6..aa078a0 100644
--- a/net/sunrpc/xprtrdma/frwr_ops.c
+++ b/net/sunrpc/xprtrdma/frwr_ops.c
@@ -244,12 +244,14 @@ frwr_op_maxpages(struct rpcrdma_xprt *r_xprt)
 		     rpcrdma_max_segments(r_xprt) * ia->ri_max_frmr_depth);
 }
 
-/* If FAST_REG or LOCAL_INV failed, indicate the frmr needs to be reset. */
+/* If FAST_REG or LOCAL_INV failed, indicate the frmr needs
+ * to be reset.
+ *
+ * WARNING: Only wr_id and status are reliable at this point
+ */
 static void
-frwr_sendcompletion(struct ib_wc *wc)
+__frwr_sendcompletion_flush(struct ib_wc *wc, struct rpcrdma_mw *r)
 {
-	struct rpcrdma_mw *r;
-
 	if (likely(wc->status == IB_WC_SUCCESS))
 		return;
 
@@ -260,9 +262,23 @@ frwr_sendcompletion(struct ib_wc *wc)
 	else
 		pr_warn("RPC:       %s: frmr %p error, status %s (%d)\n",
 			__func__, r, ib_wc_status_msg(wc->status), wc->status);
+
 	r->r.frmr.fr_state = FRMR_IS_STALE;
 }
 
+static void
+frwr_sendcompletion(struct ib_wc *wc)
+{
+	struct rpcrdma_mw *r = (struct rpcrdma_mw *)(unsigned long)wc->wr_id;
+	struct rpcrdma_frmr *f = &r->r.frmr;
+
+	if (unlikely(wc->status != IB_WC_SUCCESS))
+		__frwr_sendcompletion_flush(wc, r);
+
+	if (f->fr_waiter)
+		complete(&f->fr_linv_done);
+}
+
 static int
 frwr_op_init(struct rpcrdma_xprt *r_xprt)
 {
@@ -334,6 +350,7 @@ frwr_op_map(struct rpcrdma_xprt *r_xprt, struct rpcrdma_mr_seg *seg,
 	} while (mw->r.frmr.fr_state != FRMR_IS_INVALID);
 	frmr = &mw->r.frmr;
 	frmr->fr_state = FRMR_IS_VALID;
+	frmr->fr_waiter = false;
 	mr = frmr->fr_mr;
 	reg_wr = &frmr->fr_regwr;
 
@@ -413,6 +430,116 @@ out_senderr:
 	return rc;
 }
 
+static struct ib_send_wr *
+__frwr_prepare_linv_wr(struct rpcrdma_mr_seg *seg)
+{
+	struct rpcrdma_mw *mw = seg->rl_mw;
+	struct rpcrdma_frmr *f = &mw->r.frmr;
+	struct ib_send_wr *invalidate_wr;
+
+	f->fr_waiter = false;
+	f->fr_state = FRMR_IS_INVALID;
+	invalidate_wr = &f->fr_invwr;
+
+	memset(invalidate_wr, 0, sizeof(*invalidate_wr));
+	invalidate_wr->wr_id = (unsigned long)(void *)mw;
+	invalidate_wr->opcode = IB_WR_LOCAL_INV;
+	invalidate_wr->ex.invalidate_rkey = f->fr_mr->rkey;
+
+	return invalidate_wr;
+}
+
+static void
+__frwr_dma_unmap(struct rpcrdma_xprt *r_xprt, struct rpcrdma_mr_seg *seg,
+		 int rc)
+{
+	struct ib_device *device = r_xprt->rx_ia.ri_device;
+	struct rpcrdma_mw *mw = seg->rl_mw;
+	struct rpcrdma_frmr *f = &mw->r.frmr;
+
+	seg->rl_mw = NULL;
+
+	ib_dma_unmap_sg(device, f->sg, f->sg_nents, seg->mr_dir);
+
+	if (!rc)
+		rpcrdma_put_mw(r_xprt, mw);
+	else
+		__frwr_queue_recovery(mw);
+}
+
+/* Invalidate all memory regions that were registered for "req".
+ *
+ * Sleeps until it is safe for the host CPU to access the
+ * previously mapped memory regions.
+ */
+static void
+frwr_op_unmap_sync(struct rpcrdma_xprt *r_xprt, struct rpcrdma_req *req)
+{
+	struct ib_send_wr *invalidate_wrs, *pos, *prev, *bad_wr;
+	struct rpcrdma_ia *ia = &r_xprt->rx_ia;
+	struct rpcrdma_mr_seg *seg;
+	unsigned int i, nchunks;
+	struct rpcrdma_frmr *f;
+	int rc;
+
+	dprintk("RPC:       %s: req %p\n", __func__, req);
+
+	/* ORDER: Invalidate all of the req's MRs first
+	 *
+	 * Chain the LOCAL_INV Work Requests and post them with
+	 * a single ib_post_send() call.
+	 */
+	invalidate_wrs = pos = prev = NULL;
+	seg = NULL;
+	for (i = 0, nchunks = req->rl_nchunks; nchunks; nchunks--) {
+		seg = &req->rl_segments[i];
+
+		pos = __frwr_prepare_linv_wr(seg);
+
+		if (!invalidate_wrs)
+			invalidate_wrs = pos;
+		else
+			prev->next = pos;
+		prev = pos;
+
+		i += seg->mr_nsegs;
+	}
+	f = &seg->rl_mw->r.frmr;
+
+	/* Strong send queue ordering guarantees that when the
+	 * last WR in the chain completes, all WRs in the chain
+	 * are complete.
+	 */
+	f->fr_invwr.send_flags = IB_SEND_SIGNALED;
+	f->fr_waiter = true;
+	init_completion(&f->fr_linv_done);
+	INIT_CQCOUNT(&r_xprt->rx_ep);
+
+	/* Transport disconnect drains the receive CQ before it
+	 * replaces the QP. The RPC reply handler won't call us
+	 * unless ri_id->qp is a valid pointer.
+	 */
+	rc = ib_post_send(ia->ri_id->qp, invalidate_wrs, &bad_wr);
+	if (rc)
+		pr_warn("%s: ib_post_send failed %i\n", __func__, rc);
+
+	wait_for_completion(&f->fr_linv_done);
+
+	/* ORDER: Now DMA unmap all of the req's MRs, and return
+	 * them to the free MW list.
+	 */
+	for (i = 0, nchunks = req->rl_nchunks; nchunks; nchunks--) {
+		seg = &req->rl_segments[i];
+
+		__frwr_dma_unmap(r_xprt, seg, rc);
+
+		i += seg->mr_nsegs;
+		seg->mr_nsegs = 0;
+	}
+
+	req->rl_nchunks = 0;
+}
+
 /* Post a LOCAL_INV Work Request to prevent further remote access
  * via RDMA READ or RDMA WRITE.
  */
@@ -472,6 +599,7 @@ frwr_op_destroy(struct rpcrdma_buffer *buf)
 
 const struct rpcrdma_memreg_ops rpcrdma_frwr_memreg_ops = {
 	.ro_map				= frwr_op_map,
+	.ro_unmap_sync			= frwr_op_unmap_sync,
 	.ro_unmap			= frwr_op_unmap,
 	.ro_open			= frwr_op_open,
 	.ro_maxpages			= frwr_op_maxpages,
diff --git a/net/sunrpc/xprtrdma/xprt_rdma.h b/net/sunrpc/xprtrdma/xprt_rdma.h
index 512184d..b8bac41 100644
--- a/net/sunrpc/xprtrdma/xprt_rdma.h
+++ b/net/sunrpc/xprtrdma/xprt_rdma.h
@@ -206,6 +206,8 @@ struct rpcrdma_frmr {
 	enum rpcrdma_frmr_state		fr_state;
 	struct work_struct		fr_work;
 	struct rpcrdma_xprt		*fr_xprt;
+	bool				fr_waiter;
+	struct completion		fr_linv_done;;
 	union {
 		struct ib_reg_wr	fr_regwr;
 		struct ib_send_wr	fr_invwr;

--
To unsubscribe from this list: send the line "unsubscribe linux-nfs" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html



[Index of Archives]     [Linux Filesystem Development]     [Linux USB Development]     [Linux Media Development]     [Video for Linux]     [Linux NILFS]     [Linux Audio Users]     [Yosemite Info]     [Linux SCSI]

  Powered by Linux