Skip to contentSkip to conversation

Fix generation of IBV_WC_GRH

PATCH/libmlx4 Jason Gunthorpe 3 participants 6 messages Mar 5, 2011 – Mar 9, 2011 3 messages not in the archive InfiniBand

lore
Share & tools
lore
#0 lore raw

Byte swap was erroneously using short not long, so the test always returns false as 0x80000000 is greater than 16 bits.

---
 src/cq.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

Found and tested on a MT25418

Looks like this is a cut and paste error from libmthca which has a similar line.

Also confirmed that the kernel has a correct version.

src/cq.c +1 -1
index 2d04883..bc53f9b 100644
@@ -317,7 +317,7 @@ static int mlx4_poll_one(struct mlx4_cq *cq,
 		wc->src_qp	   = ntohl(cqe->g_mlpath_rqpn) & 0xffffff;
 		wc->dlid_path_bits = (ntohl(cqe->g_mlpath_rqpn) >> 24) & 0x7f;
 		wc->pkey_index     = ntohl(cqe->immed_rss_invalid) >> 16;
-		wc->wc_flags      |= ntohs(cqe->g_mlpath_rqpn) & 0x80000000 ?
+		wc->wc_flags      |= ntohl(cqe->g_mlpath_rqpn) & 0x80000000 ?
 			IBV_WC_GRH : 0;
 	}
 
Signature · 6 lines
1.7.1

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
#2 lore raw
In reply to a message not in the archive
Quoted 2 lines · Jason Gunthorpe

Byte swap was erroneously using short not long, so the test always returns false as 0x80000000 is greater than 16 bits.

This was fixed in libmlx4 1.0.1 a while ago. I guess I really need to get that version into Debian/Ubuntu.

- R.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

#4 lore raw
In reply to a message not in the archive
Quoted 7 lines · Roland Dreier
Quoted 2 lines · Jason Gunthorpe

Byte swap was erroneously using short not long, so the test always returns false as 0x80000000 is greater than 16 bits.

This was fixed in libmlx4 1.0.1 a while ago. I guess I really need to get that version into Debian/Ubuntu.

Hurm. OFED needs to pick up 1.0.1.. I've been using your 1.0 with an upstream kernel because that is what OFED has labeled their 1.5.1/1.5.2 versions of libmlx4. Looks like it is actually 1.0.1 anyhow. Sigh.

Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

#5 lore raw
In reply to a message not in the archive
Quoted 2 lines · Roland Dreier

This was fixed in libmlx4 1.0.1 a while ago. I guess I really need to get that version into Debian/Ubuntu.

Roland,

I see now that the below commit of yours dropped the ntohl around the GRH flag setting which was added by jack in commit b2b26f8a73108f971ade988fac3eb9d0446bbbb6

Or.

Quoted 30 lines

commit e6a3ccdffb36f94ee72b2bc97944a50777ff72e0

Author: Roland Dreier <[email protected]>
Date:   Thu Jan 3 19:59:05 2008 -0800

Micro-optimize mlx4_poll_one()

Rather than byte-swapping cqe->g_mlpath_rqpn each time we extract a field from it, byte-swap it once into a temporary variable. This results in smaller, better code.

src/cq.c +1 -2
index bc53f9b..06ae9e2 100644
@@ -196,6 +196,7 @@ static int mlx4_poll_one(struct mlx4_cq *cq,
        struct mlx4_cqe *cqe;
        struct mlx4_srq *srq;
        uint32_t qpn;
+       uint32_t g_mlpath_rqpn;
        uint16_t wqe_index;
        int is_error;
        int is_send;
@@ -314,11 +315,11 @@ static int mlx4_poll_one(struct mlx4_cq *cq,
 
                wc->slid           = ntohs(cqe->rlid);
                wc->sl             = cqe->sl >> 4;
-               wc->src_qp         = ntohl(cqe->g_mlpath_rqpn) & 0xffffff;
-               wc->dlid_path_bits = (ntohl(cqe->g_mlpath_rqpn) >> 24) & 0x7f;

src/cq.c static int mlx4_poll_one(struct mlx4_cq *cq, · line 325

Quoted 10 lines
src/cq.c +4 -2
+               g_mlpath_rqpn      = ntohl(cqe->g_mlpath_rqpn);
+               wc->src_qp         = g_mlpath_rqpn & 0xffffff;
+               wc->dlid_path_bits = (g_mlpath_rqpn >> 24) & 0x7f;
+               wc->wc_flags      |= g_mlpath_rqpn & 0x80000000 ? IBV_WC_GRH : 0;
                wc->pkey_index     = ntohl(cqe->immed_rss_invalid) >> 16;
-               wc->wc_flags      |= ntohl(cqe->g_mlpath_rqpn) & 0x80000000 ?
-                       IBV_WC_GRH : 0;
        }
 
        return CQ_OK;

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html

#8 lore raw
In reply to a message not in the archive
Quoted 3 lines · Or Gerlitz

I see now that the below commit of yours dropped the ntohl around the GRH flag setting which was added by jack in commit b2b26f8a73108f971ade988fac3eb9d0446bbbb6

I think the changelog, which you quoted, explains what is going on with the byte-swapping perfectly:

Quoted 9 lines
Quoted 9 lines

commit e6a3ccdffb36f94ee72b2bc97944a50777ff72e0
Author: Roland Dreier <[email protected]>
Date:   Thu Jan 3 19:59:05 2008 -0800

   Micro-optimize mlx4_poll_one()

   Rather than byte-swapping cqe->g_mlpath_rqpn each time we extract a field from it, byte-swap it once into a temporary variable.  This results in smaller, better code.

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html