Re: [patch] reduce IPI noise due to /dev/cdrom open/close

[Jes Sorensen <jes@sgi.com>]

Andrew Morton wrote:
> On Tue, 04 Jul 2006 15:32:19 +1000
> I expect raw_smp_processor_id() is used here as a a microoptimisation -
> avoid a might_sleep() which obviously will never trigger.
> 
> But I think it'd be better to do just a single raw_smp_processor_id() for
> this entire function:

I agree, should have done that instead.

>> Racy?  Start with an empty lru_in_use.
>>
>> Cpu A                         Cpu B
>> invalidate_bh_lrus()
>> mask = lru_in_use;
>> preempted
>>                               block I/O
>> 			      bh_lru_install()
>> 			      cpu_set(cpu, lru_in_use);
>> resume
>> cpus_clear(lru_in_use);
>> schedule_on_each_cpu_mask() - does not send IPI to cpu B
> 
> Yup.  I think we can fix that by doing a single cpu_clear() on each CPU
> just prior to that CPU clearing out its array, in invalidate_bh_lru().

I deliberately tried to avoid that to avoid the bitmask turning into a
bouncing cacheline.

> There's a possibility of course that new bh's will get installed somewhere,
> but higher-level code must ensure that those bh's do not belong to the
> device which we're trying to clean up.

Cheers,
Jes