Re: [PATCH net-next] svcrdma: Fix an uninitialized variable false warning

[Bart Van Assche <bvanassche@acm.org>]

On Fri, 2018-12-28 at 10:36 -0500, Chuck Lever wrote:
> > On Dec 20, 2018, at 4:49 AM, YueHaibing <yuehaibing@huawei.com> wrote:
> > 
> > smatch warning this:
> > net/sunrpc/xprtrdma/svc_rdma_rw.c:351 svc_rdma_post_chunk_ctxt() error: uninitialized symbol 'bad_wr'
> > net/sunrpc/xprtrdma/verbs.c:1569 rpcrdma_post_recvs() error: uninitialized symbol 'bad_wr'
> > 
> > 'bad_wr' is initialized in ib_post_send. But smatch
> > doesn't know that and warns this.
> > 
> > Signed-off-by: YueHaibing <yuehaibing@huawei.com>
> > ---
> > net/sunrpc/xprtrdma/svc_rdma_rw.c | 2 +-
> > net/sunrpc/xprtrdma/verbs.c       | 2 +-
> > 2 files changed, 2 insertions(+), 2 deletions(-)
> > 
> > diff --git a/net/sunrpc/xprtrdma/svc_rdma_rw.c b/net/sunrpc/xprtrdma/svc_rdma_rw.c
> > index dc19517..0954b25 100644
> > --- a/net/sunrpc/xprtrdma/svc_rdma_rw.c
> > +++ b/net/sunrpc/xprtrdma/svc_rdma_rw.c
> > @@ -308,7 +308,7 @@ static int svc_rdma_post_chunk_ctxt(struct svc_rdma_chunk_ctxt *cc)
> > 	struct svcxprt_rdma *rdma = cc->cc_rdma;
> > 	struct svc_xprt *xprt = &rdma->sc_xprt;
> > 	struct ib_send_wr *first_wr;
> > -	const struct ib_send_wr *bad_wr;
> > +	const struct ib_send_wr *bad_wr = NULL;
> > 	struct list_head *tmp;
> > 	struct ib_cqe *cqe;
> > 	int ret;
> > diff --git a/net/sunrpc/xprtrdma/verbs.c b/net/sunrpc/xprtrdma/verbs.c
> > index 3ddba94..37be70f 100644
> > --- a/net/sunrpc/xprtrdma/verbs.c
> > +++ b/net/sunrpc/xprtrdma/verbs.c
> > @@ -1518,7 +1518,7 @@ void
> > rpcrdma_post_recvs(struct rpcrdma_xprt *r_xprt, bool temp)
> > {
> > 	struct rpcrdma_buffer *buf = &r_xprt->rx_buf;
> > -	struct ib_recv_wr *wr, *bad_wr;
> > +	struct ib_recv_wr *wr, *bad_wr = NULL;
> > 	int needed, count, rc;
> > 
> > 	rc = 0;
> > -- 
> > 2.7.0
> 
> Does this need
> 
> Fixes: d34ac5cd3a73 ("RDMA, core and ULPs: Declare ib_post_send() and ib_post_recv() arguments const")  ???
> 
> Bart, any comments?

Hi Chuck,

My understanding is that the "Fixes:" tag should only be used for patches
that fix bugs. Since this patch addresses a false positive warning reported
by smatch I think that it would be misleading to use the "Fixes:" tag.
Before proceeding with this patch, I think that smatch should be improved.
The following patch namely was not sufficient to suppress the xprtrdma
warning reported by smatch:

diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h
index a3ceed3a040a..498eaa245d1a 100644
--- a/include/rdma/ib_verbs.h
+++ b/include/rdma/ib_verbs.h
@@ -3327,9 +3327,16 @@ static inline int ib_post_srq_recv(struct ib_srq *srq,
 				   const struct ib_recv_wr **bad_recv_wr)
 {
 	const struct ib_recv_wr *dummy;
-
-	return srq->device->ops.post_srq_recv(srq, recv_wr,
-					      bad_recv_wr ? : &dummy);
+	int ret;
+
+	ret = srq->device->ops.post_srq_recv(srq, recv_wr,
+					     bad_recv_wr ? : &dummy);
+#ifdef __CHECKER__
+	/* Tell static analyzers that *bad_send_wr is initialized if ret != 0 */
+	if (ret && bad_send_wr)
+		*bad_send_wr = *bad_send_wr;
+#endif
+	return ret;
 }
 
 /**
@@ -3431,8 +3438,15 @@ static inline int ib_post_send(struct ib_qp *qp,
 			       const struct ib_send_wr **bad_send_wr)
 {
 	const struct ib_send_wr *dummy;
+	int ret;
 
-	return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
+	ret = qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy);
+#ifdef __CHECKER__
+	/* Tell static analyzers that *bad_send_wr is initialized if ret != 0 */
+	if (ret && bad_send_wr)
+		*bad_send_wr = *bad_send_wr;
+#endif
+	return ret;
 }
 
 /**
@@ -3448,8 +3462,15 @@ static inline int ib_post_recv(struct ib_qp *qp,
 			       const struct ib_recv_wr **bad_recv_wr)
 {
 	const struct ib_recv_wr *dummy;
+	int ret;
 
-	return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
+	ret = qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy);
+#ifdef __CHECKER__
+	/* Tell static analyzers that *bad_send_wr is initialized if ret != 0 */
+	if (ret && bad_send_wr)
+		*bad_send_wr = *bad_send_wr;
+#endif
+	return ret;
 }
 
 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private,

Bart.