Thanks Steve, this is Chelsio's Unified Wire package. That additional line fixed the leak.
- Dan
From: Steve Wise <swise@xxxxxxxxxxxxxxxxxxxxx>
Sent: Monday, June 26, 2017 3:45:44 PM
To: Dan Coombs; linux-rdma@xxxxxxxxxxxxxxx
Cc: 'Michael Chen'
Subject: RE: IWARP Valgrind Issue
Hey Dan:
>
> Hello,
>
> I'm getting a valgrind error when using the cxgb4/iw_cxgb4 drivers for a
Chelsio
> IWARP card.
>
> System info:
> Distribution: Ubuntu 16.04.2 LTS
> Kernel: 4.4.0-79-generic
> Hardware: Chelsio T420-BT
> Drivers:
> Chelsio T4/T5/T6 Offload Network Driver - version 3.2.0.0
> Chelsio T420-BT rev 2
> iw_cxgb4:Chelsio T4/T5/T6 RDMA Driver - version 3.2.0.0
>
This is Chelsio's Unified Wire package, yes?
>
> Bug:
>
> Running with sudo valgrind --leak-check=full --undef-value-errors=no
>
> 12,720 bytes in 1 blocks are definitely lost in loss record 146 of 150
>
>
> Call stack:
>
> ibv_create_cq (verbs.c:328)
> c4iw_create_cq (verbs.c:245)
> calloc (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
>
> Code simply opens the device, creates a completion channel and a completion
> queue with ibv_create_*, then destroys the completion queue and completion
> channel with ibv_destroy_*.
>
That looks like a bug in Chelsio's Unified Wire libcxgb4 core (ie not in the
upstream rdma-core libcxgb4 code). I See the leak. Something like this should
probably fix it.
diff -r fbf715bf1eb7 linux_libs/libcxgb4/src/verbs.c
--- a/linux_libs/libcxgb4/src/verbs.c Fri Jun 23 13:44:07 2017 -0700
+++ b/linux_libs/libcxgb4/src/verbs.c Mon Jun 26 13:43:52 2017 -0700
@@ -304,6 +304,7 @@ int c4iw_destroy_cq(struct ibv_cq *ibcq)
dev->cqid2ptr[chp->cq.cqid] = NULL;
pthread_spin_unlock(&dev->lock);
+ free(chp->cq.swiq_queue);
free(chp->cq.sw_queue);
free(chp);
return 0;
Steve.
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
