Messages in this thread | | | Date | Fri, 20 Dec 2013 09:47:10 +0100 | From | Ingo Molnar <> | Subject | Re: [tip:x86/idle] x86, idle: Use static_cpu_has() for CLFLUSH workaround, add barriers |
| |
* Peter Zijlstra <peterz@infradead.org> wrote:
> On Thu, Dec 19, 2013 at 09:40:51PM +0100, Ingo Molnar wrote: > > > + if (static_cpu_has(X86_FEATURE_CLFLUSH_MONITOR)) { > > > + mb(); > > > clflush((void *)¤t_thread_info()->flags); > > > + mb(); > > > + } > > > if (!current_set_polling_and_test()) { > > if (static_cpu_has(X86_FEATURE_CLFLUSH_MONITOR)) { > > - mb(); > > + /* > > + * There's no need for an extra barrier here: current_set_polling_and_test() > > + * already has an smp_mb() after ->flags gets modified, see sched.h. > > + */ > > clflush((void *)¤t_thread_info()->flags); > > mb(); > > } > > So almost, except that smp_mb() will disappear on an UP kernel whereas > this errata might just always need the MFENCE.
Yeah. Sigh, so I noticed that and even thought through the UP case and concluded: on UP the race cannot happen - which is bogus. Yesterday was so much fail, maybe today's a better day ;-)
Thanks,
ngo
| |