summaryrefslogtreecommitdiff
path: root/drivers/infiniband
diff options
context:
space:
mode:
authorSteve Wise <swise@opengridcomputing.com>2010-05-20 16:57:43 -0500
committerRoland Dreier <rolandd@cisco.com>2010-05-24 21:08:01 -0700
commit84172dee05cbce6ae791eac481ef4d8590cda791 (patch)
treed631f0991e8be72d1d4d1cd2e1ce8d32eb5116e5 /drivers/infiniband
parent895cf5f3d6cb72825919a39ae48a41476a821e65 (diff)
downloadlinux-3.10-84172dee05cbce6ae791eac481ef4d8590cda791.tar.gz
linux-3.10-84172dee05cbce6ae791eac481ef4d8590cda791.tar.bz2
linux-3.10-84172dee05cbce6ae791eac481ef4d8590cda791.zip
RDMA/cxgb4: Optimize CQ overflow detection
1) save the timestamp flit in the cq when we consume a CQE. 2) always compare the saved flit with the previous entry flit when reading the next CQE entry. If the flits don't compare, then we have overflowed. Signed-off-by: Steve Wise <swise@opengridcomputing.com> Signed-off-by: Roland Dreier <rolandd@cisco.com>
Diffstat (limited to 'drivers/infiniband')
-rw-r--r--drivers/infiniband/hw/cxgb4/cq.c1
-rw-r--r--drivers/infiniband/hw/cxgb4/t4.h28
2 files changed, 17 insertions, 12 deletions
diff --git a/drivers/infiniband/hw/cxgb4/cq.c b/drivers/infiniband/hw/cxgb4/cq.c
index 46ac00f728f..2447f529548 100644
--- a/drivers/infiniband/hw/cxgb4/cq.c
+++ b/drivers/infiniband/hw/cxgb4/cq.c
@@ -373,6 +373,7 @@ static void create_read_req_cqe(struct t4_wq *wq, struct t4_cqe *hw_cqe,
V_CQE_SWCQE(SW_CQE(hw_cqe)) |
V_CQE_OPCODE(FW_RI_READ_REQ) |
V_CQE_TYPE(1));
+ read_cqe->bits_type_ts = hw_cqe->bits_type_ts;
}
/*
diff --git a/drivers/infiniband/hw/cxgb4/t4.h b/drivers/infiniband/hw/cxgb4/t4.h
index d0e8af35240..712bc5620d3 100644
--- a/drivers/infiniband/hw/cxgb4/t4.h
+++ b/drivers/infiniband/hw/cxgb4/t4.h
@@ -434,7 +434,7 @@ struct t4_cq {
struct c4iw_rdev *rdev;
u64 ugts;
size_t memsize;
- u64 timestamp;
+ __be64 bits_type_ts;
u32 cqid;
u16 size; /* including status page */
u16 cidx;
@@ -487,6 +487,7 @@ static inline void t4_swcq_consume(struct t4_cq *cq)
static inline void t4_hwcq_consume(struct t4_cq *cq)
{
+ cq->bits_type_ts = cq->queue[cq->cidx].bits_type_ts;
cq->cidx_inc++;
if (++cq->cidx == cq->size) {
cq->cidx = 0;
@@ -501,20 +502,23 @@ static inline int t4_valid_cqe(struct t4_cq *cq, struct t4_cqe *cqe)
static inline int t4_next_hw_cqe(struct t4_cq *cq, struct t4_cqe **cqe)
{
- int ret = 0;
- u64 bits_type_ts = be64_to_cpu(cq->queue[cq->cidx].bits_type_ts);
+ int ret;
+ u16 prev_cidx;
- if (G_CQE_GENBIT(bits_type_ts) == cq->gen) {
- *cqe = &cq->queue[cq->cidx];
- cq->timestamp = G_CQE_TS(bits_type_ts);
- } else if (G_CQE_TS(bits_type_ts) > cq->timestamp)
- ret = -EOVERFLOW;
+ if (cq->cidx == 0)
+ prev_cidx = cq->size - 1;
else
- ret = -ENODATA;
- if (ret == -EOVERFLOW) {
- printk(KERN_ERR MOD "cq overflow cqid %u\n", cq->cqid);
+ prev_cidx = cq->cidx - 1;
+
+ if (cq->queue[prev_cidx].bits_type_ts != cq->bits_type_ts) {
+ ret = -EOVERFLOW;
cq->error = 1;
- }
+ printk(KERN_ERR MOD "cq overflow cqid %u\n", cq->cqid);
+ } else if (t4_valid_cqe(cq, &cq->queue[cq->cidx])) {
+ *cqe = &cq->queue[cq->cidx];
+ ret = 0;
+ } else
+ ret = -ENODATA;
return ret;
}