Re: [RFC] Add implicit barriers to irqsave/restore class of functions

[Vineet Gupta]

Hi Peter,

On 04/04/2013 09:43 PM, Peter Zijlstra wrote:
> -       : "cc");
> +       : "memory", "cc");
> That's not a memory barrier, that a memory clobber, aka a compiler
> barrier.

For the problem under consideration we indeed want a compiler barrier because the
error shows up due to a stale register which is live across a spinlock for
!PREEMPT_COUNT config.

However IMO doing this in irq save/restore macros is semantically incorrect since
those macros might be used elsewhere which don't need the compiler reload reg
semantics. Further per tglx' suggestion fixing preempt_* macros for !PREEMPT_COUNT
case would fix this independent of what arch is doing.

A patch to that effect was already posted to lists:
http://www.spinics.net/lists/kernel/msg1510885.html
Please let us know what you think.

Thx,
-Vineet


--
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/