On Thu, 19 May 2011, KAMEZAWA Hiroyuki wrote:

On Wed, 18 May 2011 11:25:48 -0700 (PDT)
Hugh Dickins [off-list ref] wrote:

quoted

On Wed, 18 May 2011, Daisuke Nishimura wrote:

quoted

On Tue, 17 May 2011 11:24:40 -0700 (PDT)
Hugh Dickins [off-list ref] wrote:

quoted

mem_cgroup_count_vm_event() should update the PGMAJFAULT count for the
target mm, not for current mm (but of course they're usually the same).

hmm, why ?
In shmem_getpage(), we charge the page to the memcg where current mm belongs to,

(In the case when it's this fault which is creating the page.
Just as when filemap_fault() reads in the page, add_to_page_cache
will charge it to the current->mm's memcg, yes.  Arguably correct.)

quoted

so I think counting vm events of the memcg is right.

It should be consistent with which task gets the maj_flt++, and
it should be consistent with filemap_fault(), and it should be a
subset of what's counted by mem_cgroup_count_vm_event(mm, PGFAULT).

In each case, those work on target mm rather than current->mm.

Hmm, I have no strong opinion on this but yes, it makes sense to account
PGMAJFLT to the process whose mm->maj_flt++.

mm->maj_flt++ would be yet another story!  But it's tsk->maj_flt++.

BTW,  do you think memcg should
account shmem into vma->vm_mm rather than current->mm ? When vma->vm_mm
is different from current ? At get_user_pages() + MAJFLT ?

If what we have at present works well enough, I don't think we should
risk breaking it with a funny change like that.

Is there a reason to treat shmem differently from filemap there?
You can argue that if it's shm then yes, but if it's tmpfs then no.

I suppose shmem.c does usually know the difference (by VM_NORESERVE),
but does not export it; and I'd prefer to keep it that way.

Unless we've got a real bug to fix here, let's not mess with it.

Hugh

--
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>