Re: [Qemu-devel] [PATCH] monitor: Fix warning from clang

[Markus Armbruster]

Luiz Capitulino <address@hidden> writes:

> On Fri, 17 Aug 2012 16:10:12 +0200
> Markus Armbruster <address@hidden> wrote:
>
>> Stefan Weil <address@hidden> writes:
>> 
>> > ccc-analyzer reports these warnings:
>> >
>> > monitor.c:3532:21: warning: Division by zero
>> >                 val %= val2;
>> >                     ^
>> > monitor.c:3530:21: warning: Division by zero
>> >                 val /= val2;
>> >                     ^
>> >
>> > Rewriting the code fixes this (and also a style issue).
>> 
>> I'm afraid this doesn't actually fix anything, because...
>> 
>> > Signed-off-by: Stefan Weil <address@hidden>
>> > ---
>> >  monitor.c |    7 ++++---
>> >  1 file changed, 4 insertions(+), 3 deletions(-)
>> >
>> > diff --git a/monitor.c b/monitor.c
>> > index 0c34934..0ea2c14 100644
>> > --- a/monitor.c
>> > +++ b/monitor.c
>> > @@ -3524,12 +3524,13 @@ static int64_t expr_prod(Monitor *mon)
>> >              break;
>> >          case '/':
>> >          case '%':
>> > -            if (val2 == 0)
>> > +            if (val2 == 0) {
>> >                  expr_error(mon, "division by zero");
>> > -            if (op == '/')
>> > +            } else if (op == '/') {
>> >                  val /= val2;
>> > -            else
>> > +            } else {
>> >                  val %= val2;
>> > +            }
>> >              break;
>> >          }
>> >      }
>> 
>> ... expr_error() longjmp()s out.  The expression evaluator commonly
>> exploits that.
>
> And that's correct. As far far I understood it's fixing clang, not qemu.
>
>> If expr_error() returned, the code would be just as wrong after your
>> patch as before.
>
> Hmm, how? It checks for val2 == 0 first.

It would evaluate A % 0 into A, which is wrong.

>> Perhaps the checker can be shut up by making expr_error() QEMU_NORETURN.
>
> That's indeed a better solution.

Stefan, could you try that for us?