Re: [patch 3/8] mutex subsystem, add atomic_*_call_if_*() to i386

From: Ingo Molnar
Date: Wed Dec 21 2005 - 15:00:40 EST



* Linus Torvalds <torvalds@xxxxxxxx> wrote:

> > Umm. This asm is broken. It doesn't mark %eax as changed, so this is only
> > reliable if the function you call is
> >
> > - a "fastcall" one
> > - always returns as its return value the pointer to the atomic count
> >
> > which is not true (you verify that it's a fastcall, but it's of type
> > "void").
>
> Actually (and re-reading the email I sent that wasn't obvious at all),
> my _preferred_ fix is to literally force the use of the above kind of
> function: not save/restore %eax at all, but just say that any function
> that is called by the magic "atomic_*_call_if()" needs to always
> return the argument it gets as its return value too.
>
> That allows the caller to not even have to care. And the callee
> obviously already _has_ that value, so it might as well return it (and
> in the best case it's not going to add any cost at all, either to the
> caller or the callee).
>
> So you might opt to keep the asm the same, just change the calling
> conventions.

ok, i've added this fix, thanks. Right now we dont do anything after
those functions (that's probably how the bug never showed up), but at
least one interim stage i tried to use the call_if functions at other
places too, so the potential is there.

Ingo
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/