On Fri, Dec 20, 2013 at 5:40 AM, Ingo Molnar <mingo@xxxxxxxxxx> wrote:
>
> The first mb() looks superfluous, because see
> current_set_polling_and_test():
>
> static inline bool __must_check current_set_polling_and_test(void)
> {
> __current_set_polling();
>
> /*
> * Polling state must be visible before we test NEED_RESCHED,
> * paired by resched_task()
> */
> smp_mb();
>
> return unlikely(tif_need_resched());
> }
>
> So it already has a (MFENCE) barrier after ->flags is modified.
So what?
The mb() isn't necessarily against the *write*, it is also against the *read*.
If the cflush is needed before the monitor, it's likely because the
monitor instruction has some bug with already-existing cachelines in
shared state or whatever.
Which means that we want to make sure that the cflush is ordered wrt
*reads* from that cacheline too.
cflush has nothing specifically to do with writes.
Linus
--
To unsubscribe from this list: send the line "unsubscribe linux-tip-commits" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
|