xprtrdma: Cap req_cqinit

Recent work made FRMR registration and invalidation completions
unsignaled. This greatly reduces the adapter interrupt rate.

Every so often, however, a posted send Work Request is allowed to
signal. Otherwise, the provider's Work Queue will wrap and the
workload will hang.

The number of Work Requests that are allowed to remain unsignaled is
determined by the value of req_cqinit. Currently, this is set to the
size of the send Work Queue divided by two, minus 1.

For FRMR, the send Work Queue is the maximum number of concurrent
RPCs (currently 32) times the maximum number of Work Requests an
RPC might use (currently 7, though some adapters may need more).

For mlx4, this is 224 entries. This leaves completion signaling
disabled for 111 send Work Requests.

Some providers hold back dispatching Work Requests until a CQE is
generated.  If completions are disabled, then no CQEs are generated
for quite some time, and that can stall the Work Queue.

I've seen this occur running xfstests generic/113 over NFSv4, where
eventually, posting a FAST_REG_MR Work Request fails with -ENOMEM
because the Work Queue has overflowed. The connection is dropped
and re-established.

Cap the rep_cqinit setting so completions are not left turned off
for too long.

BugLink: https://bugzilla.linux-nfs.org/show_bug.cgi?id=269
Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Anna Schumaker <Anna.Schumaker@Netapp.com>

diff --git a/net/sunrpc/xprtrdma/verbs.c b/net/sunrpc/xprtrdma/verbs.c
index 6ea29420f0acb1ce204851496efdd897ec17a98a..af45cf390126089ddbab36ecd51b293600a5fef1 100644 (file)
--- a/net/sunrpc/xprtrdma/verbs.c
+++ b/net/sunrpc/xprtrdma/verbs.c
@@ -733,7 +733,9 @@ rpcrdma_ep_create(struct rpcrdma_ep *ep, struct rpcrdma_ia *ia,
 
        /* set trigger for requesting send completion */
        ep->rep_cqinit = ep->rep_attr.cap.max_send_wr/2 - 1;
-       if (ep->rep_cqinit <= 2)
+       if (ep->rep_cqinit > RPCRDMA_MAX_UNSIGNALED_SENDS)
+               ep->rep_cqinit = RPCRDMA_MAX_UNSIGNALED_SENDS;
+       else if (ep->rep_cqinit <= 2)
                ep->rep_cqinit = 0;
        INIT_CQCOUNT(ep);
        ep->rep_ia = ia;

diff --git a/net/sunrpc/xprtrdma/xprt_rdma.h b/net/sunrpc/xprtrdma/xprt_rdma.h
index ac7fc9a3134276bba8f67ad2cf9b84714ee3b3f4..b799041b75bf9efd01dc0c3a5bb77cc832bdf020 100644 (file)
--- a/net/sunrpc/xprtrdma/xprt_rdma.h
+++ b/net/sunrpc/xprtrdma/xprt_rdma.h
@@ -97,6 +97,12 @@ struct rpcrdma_ep {
        struct ib_wc            rep_recv_wcs[RPCRDMA_POLLSIZE];
 };
 
+/*
+ * Force a signaled SEND Work Request every so often,
+ * in case the provider needs to do some housekeeping.
+ */
+#define RPCRDMA_MAX_UNSIGNALED_SENDS   (32)
+
 #define INIT_CQCOUNT(ep) atomic_set(&(ep)->rep_cqcount, (ep)->rep_cqinit)
 #define DECR_CQCOUNT(ep) atomic_sub_return(1, &(ep)->rep_cqcount)