[openib-general] [PATCH] mthca: fix PRM compliance problem in atomic-send completions

[Jack Morgenstein]

For Atomic-Send completions, according to the Tavor and Arbel
PRMs, the number of bytes transferred is not provided in the byte_cnt
field of the cqe. For atomic operations, the number of bytes transferred
is always 8 (when the status is "success"), and this value should be
inserted by the driver in the ib_wc entry returned to the poller.

Signed-off-by: Jack Morgenstein <jackm at dev.mellanox.co.il>

diff --git a/drivers/infiniband/hw/mthca/mthca_cq.c b/drivers/infiniband/hw/mthca/mthca_cq.c
index 8293c15..8fd0e9e 100644
--- a/drivers/infiniband/hw/mthca/mthca_cq.c
+++ b/drivers/infiniband/hw/mthca/mthca_cq.c
@@ -54,6 +54,10 @@ enum {
 	MTHCA_CQ_ENTRY_SIZE = 0x20
 };
 
+enum {
+	MTHCA_ATOMIC_BYTE_LEN = 8
+};
+
 /*
  * Must be packed because start is 64 bits but only aligned to 32 bits.
  */
@@ -598,11 +602,11 @@ static inline int mthca_poll_one(struct mthca_dev *dev,
 			break;
 		case MTHCA_OPCODE_ATOMIC_CS:
 			entry->opcode    = IB_WC_COMP_SWAP;
-			entry->byte_len  = be32_to_cpu(cqe->byte_cnt);
+			entry->byte_len  = MTHCA_ATOMIC_BYTE_LEN;
 			break;
 		case MTHCA_OPCODE_ATOMIC_FA:
 			entry->opcode    = IB_WC_FETCH_ADD;
-			entry->byte_len  = be32_to_cpu(cqe->byte_cnt);
+			entry->byte_len  = MTHCA_ATOMIC_BYTE_LEN;
 			break;
 		case MTHCA_OPCODE_BIND_MW:
 			entry->opcode    = IB_WC_BIND_MW;