Re: [PATCH] pkt_sched: QFQ Plus: fair-queueing service at DRR cost

On Tue, 20 Nov 2012 18:45:02 +0100
Paolo Valente [off-list ref] wrote:

Il 20/11/2012 00:48, David Miller ha scritto:

quoted

From: Paolo Valente <redacted>
Date: Mon, 12 Nov 2012 17:48:33 +0100

quoted

[This patch received positive feedback from Stephen Hemminger ("put in
net-next"), but no further feedback or decision. So I am (re)sending
an updated version of it. The only differences with respect to the
previous version are the support for TSO/GSO (taken from QFQ), and a
hopefully improved description.]

Can you rearrange the logic so that the compiler doesn't emit this
warning?

In file included from net/sched/sch_qfq.c:18:0:
net/sched/sch_qfq.c: In function ‘qfq_dequeue’:
include/net/sch_generic.h:480:15: warning: ‘skb’ may be used uninitialized in this function [-Wmaybe-uninitialized]
net/sched/sch_qfq.c:1007:18: note: ‘skb’ was declared here

You and I both know that SKB will be initialized at this point, but
the compiler can't see it clearly enough.

Unfortunately I could not reproduce the warning (with
gcc-4.7 -Wmaybe-uninitialized). I am however about to send a new version 
with skb initialized to NULL. I hope that this fix properly addresses 
this issue.

There are actually lots of bogus warnings than seem to only occur
because gcc 4.4 does a bad job of checking. Later versions are fixed
and don't generate warnings.

My preference is to not add the unnecessary initialization because
if you get in the habit of doing it. The whole purpose of the uninitialized
check is lost.