Re: [PATCH v2 1/2] x86/delay: Fix the wrong asm constraint in `delay_loop()`

[Ammar Faizi]

Hi Greg,

On 3/1/22 3:26 PM, Greg KH wrote:
> On Tue, Mar 01, 2022 at 02:32:22PM +0700, Ammar Faizi wrote:
> > The asm constraint does not reflect that the asm statement can modify
> > the value of @loops. But the asm statement in delay_loop() does change
> > the @loops.
> >
> > If by any chance the compiler inlines this function, it may clobber
> > random stuff (e.g. local variable, important temporary value in reg,
> > etc.).
> >
> > Fortunately, delay_loop() is only called indirectly (so it can't
> > inline), and then the register it clobbers is %rax (which is by the
> > nature of the calling convention, it's a caller saved register), so it
> > didn't yield any bug.
> >
> > ^ That shouldn't be an excuse for using the wrong constraint anyway.
> >
> > This changes "a" (as an input) to "+a" (as an input and output).
> >
> > Cc: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
> > Cc: Ingo Molnar <mingo@xxxxxxxxxx>
> > Cc: Borislav Petkov <bp@xxxxxxxxx>
> > Cc: Dave Hansen <dave.hansen@xxxxxxxxxxxxxxx>
> > Cc: "H. Peter Anvin" <hpa@xxxxxxxxx>
> > Fixes: e01b70ef3eb3080fecc35e15f68cd274c0a48163 ("x86: fix bug in arch/i386/lib/delay.c file, delay_loop function")
>
> You only need 12 characters here :)

Ah well, that's too verbose. Will fix it in the v4.

> > Signed-off-by: Ammar Faizi <ammarfaizi2@xxxxxxxxxxx>
> > ---
>
> Why is this one not tagged for stable?

As far as I can tell, the compiler will never inline that function,
because despite the function is static, it's assigned to a global
variable and it's called indirectly via a function pointer variable,
so it can't be inline. Therefore, it will always be a function call.

Note that %eax is a call clobbered register w.r.t. System V ABI. As
such, *by luck*, this wrong constraint doesn't yield any bug.

The compiler will not assume the %eax value is the same as before the
function call is done. So the compiler isn't aware that constraint
violation. Not sure if it's worth it for backport.

x86 maintainers, any comment on this?

--
Ammar Faizi