2022-07-15 20:28:37

by Petko Manolov

[permalink] [raw]
Subject: GCC fails to spot uninitialized variable

Guys,

Today i was bitten by a stupid bug that i introduced myself while writing some
v4l2 code. Looking at it a bit more carefully i was surprised that GCC didn't
catch this one, as it was something that should definitely emit a warning.

When included into the driver, this particular code:

int blah(int a, int *b)
{
int ret;

switch (a) {
case 0:
ret = a;
break;
case 1:
ret = *b;
break;
case 2:
*b = a;
break;
default:
ret = 0;
}

return ret;
}

somehow managed to defeat GCC checks. Compiling it as a standalone .c file
with:

gcc -Wall -O2 -c t.c

gives me nice:

t.c:19:16: warning: 'ret' may be used uninitialized in this function [-Wmaybe-uninitialized]
19 | return ret;
| ^~~

Any idea what might have gone wrong?


cheers,
Petko


2022-07-15 22:37:39

by Nathan Chancellor

[permalink] [raw]
Subject: Re: GCC fails to spot uninitialized variable

On Fri, Jul 15, 2022 at 11:09:17PM +0300, Petko Manolov wrote:
> Guys,
>
> Today i was bitten by a stupid bug that i introduced myself while writing some
> v4l2 code. Looking at it a bit more carefully i was surprised that GCC didn't
> catch this one, as it was something that should definitely emit a warning.
>
> When included into the driver, this particular code:
>
> int blah(int a, int *b)
> {
> int ret;
>
> switch (a) {
> case 0:
> ret = a;
> break;
> case 1:
> ret = *b;
> break;
> case 2:
> *b = a;
> break;
> default:
> ret = 0;
> }
>
> return ret;
> }
>
> somehow managed to defeat GCC checks. Compiling it as a standalone .c file
> with:
>
> gcc -Wall -O2 -c t.c
>
> gives me nice:
>
> t.c:19:16: warning: 'ret' may be used uninitialized in this function [-Wmaybe-uninitialized]
> 19 | return ret;
> | ^~~
>
> Any idea what might have gone wrong?

See commit 78a5255ffb6a ("Stop the ad-hoc games with
-Wno-maybe-initialized") in 5.7, which disabled that warning for a
default kernel build. You have to either include 'W=2' (which will
introduce other warnings which might be noisy) or
'KCFLAGS=-Wmaybe-uninitialized' (which will just add that warning) in
your make command to see those warnings.

As an aside, your mailer adds a "Mail-Followup-To:" header that was set
to LKML, meaning that you would not have seen this reply unless you were
subscribed to LKML. Might be something worth looking into.

Cheers,
Nathan

2022-07-16 06:39:34

by Petko Manolov

[permalink] [raw]
Subject: Re: GCC fails to spot uninitialized variable

On 22-07-15 15:03:37, Nathan Chancellor wrote:
> On Fri, Jul 15, 2022 at 11:09:17PM +0300, Petko Manolov wrote:
> > Guys,
> >
> > Today i was bitten by a stupid bug that i introduced myself while writing
> > some v4l2 code. Looking at it a bit more carefully i was surprised that GCC
> > didn't catch this one, as it was something that should definitely emit a
> > warning.
> >
> > When included into the driver, this particular code:
> >
> > int blah(int a, int *b)
> > {
> > int ret;
> >
> > switch (a) {
> > case 0:
> > ret = a;
> > break;
> > case 1:
> > ret = *b;
> > break;
> > case 2:
> > *b = a;
> > break;
> > default:
> > ret = 0;
> > }
> >
> > return ret;
> > }
> >
> > somehow managed to defeat GCC checks. Compiling it as a standalone .c file
> > with:
> >
> > gcc -Wall -O2 -c t.c
> >
> > gives me nice:
> >
> > t.c:19:16: warning: 'ret' may be used uninitialized in this function [-Wmaybe-uninitialized]
> > 19 | return ret;
> > | ^~~
> >
> > Any idea what might have gone wrong?
>
> See commit 78a5255ffb6a ("Stop the ad-hoc games with -Wno-maybe-initialized")
> in 5.7, which disabled that warning for a default kernel build. You have to
> either include 'W=2' (which will introduce other warnings which might be
> noisy) or 'KCFLAGS=-Wmaybe-uninitialized' (which will just add that warning)
> in your make command to see those warnings.

I see. I guess i'll end up enabling W=2 only for this particular driver and
only while in development.

> As an aside, your mailer adds a "Mail-Followup-To:" header that was set to
> LKML, meaning that you would not have seen this reply unless you were
> subscribed to LKML. Might be something worth looking into.

That would be "set followup_to=no" in mutt speak. Thanks for looking into this.
I am subscribed to all list i'm replying to but, if i understand this properly,
with the old setup non-subscribers may not get my messages.


cheers,
Petko