[ofa-general] [PATCH] ipoib: Bug fix in ipoib_poll - resend

[Eli Cohen]

On Tue, 2007-11-27 at 07:32 -0800, Roland Dreier wrote:
> > > > +      if (unlikely(n < 1))
>  > > > +         break;
>  > > >
>  > > >        for (i = 0; i < n; i++) {
>  > > >           struct ib_wc *wc = priv->ibwc + i;
>  > 
>  > > The 'for' loop (followed by the "if (n != t) break" check)
>  > > should take care of this, isn't it?
>  > 
>  > Oh you're saying that since this is signed arithmetic we should never
>  > enter the for loop. I saw a problem when I worked on some patch and that
>  > seemed to solve the problem...
> 
> What does signed arithmetic have to do with anything?  I don't see how
> your patch changes the behavior at all: if the condition (n < 1) ever
> tests as true then the for loop cannot be entered, and since t is
> always at least 1, the condition (n != t) will be true and we'll break
> out of the while loop anyway.
> 
I mean that I agree that the patch does not fix anything since even if
poll_cq returns a negative number, "i < n" in the for loop will cause
the block to not be executed (since "0 < negative number" is false).