summaryrefslogtreecommitdiff
path: root/net/rds/ib_recv.c
diff options
context:
space:
mode:
Diffstat (limited to 'net/rds/ib_recv.c')
-rw-r--r--net/rds/ib_recv.c47
1 files changed, 32 insertions, 15 deletions
diff --git a/net/rds/ib_recv.c b/net/rds/ib_recv.c
index 70559854837e..694d411dc72f 100644
--- a/net/rds/ib_recv.c
+++ b/net/rds/ib_recv.c
@@ -1,5 +1,5 @@
/*
- * Copyright (c) 2006, 2017 Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2006, 2019 Oracle and/or its affiliates. All rights reserved.
*
* This software is available to you under a choice of one of two
* licenses. You may choose to be licensed under the terms of the GNU
@@ -61,7 +61,7 @@ void rds_ib_recv_init_ring(struct rds_ib_connection *ic)
recv->r_wr.num_sge = RDS_IB_RECV_SGE;
sge = &recv->r_sge[0];
- sge->addr = ic->i_recv_hdrs_dma + (i * sizeof(struct rds_header));
+ sge->addr = ic->i_recv_hdrs_dma[i];
sge->length = sizeof(struct rds_header);
sge->lkey = ic->i_pd->local_dma_lkey;
@@ -168,6 +168,7 @@ void rds_ib_recv_free_caches(struct rds_ib_connection *ic)
list_del(&inc->ii_cache_entry);
WARN_ON(!list_empty(&inc->ii_frags));
kmem_cache_free(rds_ib_incoming_slab, inc);
+ atomic_dec(&rds_ib_allocation);
}
rds_ib_cache_xfer_to_ready(&ic->i_cache_frags);
@@ -342,7 +343,7 @@ static int rds_ib_recv_refill_one(struct rds_connection *conn,
WARN_ON(ret != 1);
sge = &recv->r_sge[0];
- sge->addr = ic->i_recv_hdrs_dma + (recv - ic->i_recvs) * sizeof(struct rds_header);
+ sge->addr = ic->i_recv_hdrs_dma[recv - ic->i_recvs];
sge->length = sizeof(struct rds_header);
sge = &recv->r_sge[1];
@@ -384,6 +385,7 @@ void rds_ib_recv_refill(struct rds_connection *conn, int prefill, gfp_t gfp)
unsigned int posted = 0;
int ret = 0;
bool can_wait = !!(gfp & __GFP_DIRECT_RECLAIM);
+ bool must_wake = false;
u32 pos;
/* the goal here is to just make sure that someone, somewhere
@@ -404,6 +406,7 @@ void rds_ib_recv_refill(struct rds_connection *conn, int prefill, gfp_t gfp)
recv = &ic->i_recvs[pos];
ret = rds_ib_recv_refill_one(conn, recv, gfp);
if (ret) {
+ must_wake = true;
break;
}
@@ -422,6 +425,11 @@ void rds_ib_recv_refill(struct rds_connection *conn, int prefill, gfp_t gfp)
}
posted++;
+
+ if ((posted > 128 && need_resched()) || posted > 8192) {
+ must_wake = true;
+ break;
+ }
}
/* We're doing flow control - update the window. */
@@ -444,10 +452,13 @@ void rds_ib_recv_refill(struct rds_connection *conn, int prefill, gfp_t gfp)
* if we should requeue.
*/
if (rds_conn_up(conn) &&
- ((can_wait && rds_ib_ring_low(&ic->i_recv_ring)) ||
+ (must_wake ||
+ (can_wait && rds_ib_ring_low(&ic->i_recv_ring)) ||
rds_ib_ring_empty(&ic->i_recv_ring))) {
queue_delayed_work(rds_wq, &conn->c_recv_w, 1);
}
+ if (can_wait)
+ cond_resched();
}
/*
@@ -772,7 +783,7 @@ static void rds_ib_cong_recv(struct rds_connection *conn,
unsigned long frag_off;
unsigned long to_copy;
unsigned long copied;
- uint64_t uncongested = 0;
+ __le64 uncongested = 0;
void *addr;
/* catch completely corrupt packets */
@@ -789,7 +800,7 @@ static void rds_ib_cong_recv(struct rds_connection *conn,
copied = 0;
while (copied < RDS_CONG_MAP_BYTES) {
- uint64_t *src, *dst;
+ __le64 *src, *dst;
unsigned int k;
to_copy = min(RDS_FRAG_SIZE - frag_off, PAGE_SIZE - map_off);
@@ -824,9 +835,7 @@ static void rds_ib_cong_recv(struct rds_connection *conn,
}
/* the congestion map is in little endian order */
- uncongested = le64_to_cpu(uncongested);
-
- rds_cong_map_updated(map, uncongested);
+ rds_cong_map_updated(map, le64_to_cpu(uncongested));
}
static void rds_ib_process_recv(struct rds_connection *conn,
@@ -852,7 +861,7 @@ static void rds_ib_process_recv(struct rds_connection *conn,
}
data_len -= sizeof(struct rds_header);
- ihdr = &ic->i_recv_hdrs[recv - ic->i_recvs];
+ ihdr = ic->i_recv_hdrs[recv - ic->i_recvs];
/* Validate the checksum. */
if (!rds_message_verify_checksum(ihdr)) {
@@ -984,10 +993,11 @@ void rds_ib_recv_cqe_handler(struct rds_ib_connection *ic,
} else {
/* We expect errors as the qp is drained during shutdown */
if (rds_conn_up(conn) || rds_conn_connecting(conn))
- rds_ib_conn_error(conn, "recv completion on <%pI6c,%pI6c, %d> had status %u (%s), disconnecting and reconnecting\n",
+ rds_ib_conn_error(conn, "recv completion on <%pI6c,%pI6c, %d> had status %u (%s), vendor err 0x%x, disconnecting and reconnecting\n",
&conn->c_laddr, &conn->c_faddr,
conn->c_tos, wc->status,
- ib_wc_status_msg(wc->status));
+ ib_wc_status_msg(wc->status),
+ wc->vendor_err);
}
/* rds_ib_process_recv() doesn't always consume the frag, and
@@ -1039,9 +1049,14 @@ int rds_ib_recv_init(void)
si_meminfo(&si);
rds_ib_sysctl_max_recv_allocation = si.totalram / 3 * PAGE_SIZE / RDS_FRAG_SIZE;
- rds_ib_incoming_slab = kmem_cache_create("rds_ib_incoming",
- sizeof(struct rds_ib_incoming),
- 0, SLAB_HWCACHE_ALIGN, NULL);
+ rds_ib_incoming_slab =
+ kmem_cache_create_usercopy("rds_ib_incoming",
+ sizeof(struct rds_ib_incoming),
+ 0, SLAB_HWCACHE_ALIGN,
+ offsetof(struct rds_ib_incoming,
+ ii_inc.i_usercopy),
+ sizeof(struct rds_inc_usercopy),
+ NULL);
if (!rds_ib_incoming_slab)
goto out;
@@ -1059,6 +1074,8 @@ out:
void rds_ib_recv_exit(void)
{
+ WARN_ON(atomic_read(&rds_ib_allocation));
+
kmem_cache_destroy(rds_ib_incoming_slab);
kmem_cache_destroy(rds_ib_frag_slab);
}