On Thu 29 Jul 2010 11:29, Mike Frysinger pondered:
> On Thu, Jul 29, 2010 at 03:35, Zhang, Sonic <[email protected]> wrote:
> >From: Mike Frysinger [mailto:[email protected]]
> >>On Wed, Jul 28, 2010 at 06:53,  <[email protected]> wrote:
> >>> Modified: trunk/arch/blackfin/mach-bf561/atomic.S (9019 => 9020)
> >>> @@ -71,6 +71,8 @@
> >>>      SSYNC(r2);
> >>>      jump .Lretry_corelock_noflush
> >>>  .Ldone_corelock_noflush:
> >>> +    nop;
> >>> +    CSYNC(r2);
> >>
> >>why the NOP ?  the CSYNC() macro already takes care of padding
> >>things correctly.
> >
> > I also think NOP is no necessary. But, the SMP kgdb single step fails
> > without this nop.
> 
> add an inline comment then noting this ?  otherwise someone might
> scrub it in the future without retesting.

This seems fishy to me...

The CSYNC already does:

#define CSYNC(scratch) cli scratch; nop; nop; nop; CSYNC; sti scratch;

Do things work if it is SSYNC? 

-Robin
_______________________________________________
Linux-kernel-commits mailing list
[email protected]
https://blackfin.uclinux.org/mailman/listinfo/linux-kernel-commits

Reply via email to