Re: [RFC][PATCH 2/3] pagemap: export KPF_THP

[Dave Hansen <dave@linux.vnet.ibm.com>]

On 12/19/2011 12:45 PM, KOSAKI Motohiro wrote:
> (12/19/11 3:31 PM), Dave Hansen wrote:
>> Let's say you profiled a application and the data shows you're missing
>> the TLB a bunch, but you're also using THP.  This might give you a shot
>> at figuring out which parts of your application are *TRULY* THP-backed
>> instead of just the areas you *think* are backed.
>>
>> I'm not sure there's another way to figure it out at the moment.
> 
> A snapshot status of THP doesn't help your purpose. I think you need
> perf or similar profiling subsystem enhancement.
>
> Because of, if you've seen KPF_THP at once, It has no guarantee to keep
> hugepages until applications run. Opposite, If you only need rough
> statistics, the best way is to add some new stat to
> /sys/kernel/mm/transparent_hugepage.

But, every single one of the pagemap flags is really just a snapshot
KPF_DIRTY, KPF_LOCKED, etc...  The entire interface is inherently a racy
snapshot, and there's not a whole lot you can do about it.
sys_mincore() has the exact same issues.  But, that does not make them
useless, nor mean they shouldn't be in the kernel.

A tracepoint or something similar to watch for THP promotions or
demotions would be a great addition to this interface.  That way, you at
least have a concept if the data you got has become stale.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>