Re: [PATCH] Pass priority to shrink_slab

[Andrew Morton <akpm@linux-foundation.org>]

On Mon, 22 Nov 2010 18:09:33 -0800 Michel Lespinasse <walken@google.com> wrote:

> On Mon, Nov 22, 2010 at 3:06 PM, Andrew Morton
> <akpm@linux-foundation.org> wrote:
> > On Fri, 19 Nov 2010 19:23:22 -0800
> > Ying Han <yinghan@google.com> wrote:
> >> Yes, and it would be much easier later to add a small feature (like this
> >> one) w/o
> >> touching so many files of the shrinkers. I am thinking if we can extend the
> >> scan_control
> >> from page reclaim and pass it down to the shrinker ?
> >
> > Yes, that might work. __All callers of shrink_slab() already have a
> > scan_control on the stack, so passing all that extra info to the
> > shrinkers (along with some extra fields if needed) is pretty cheap, and
> > I don't see a great downside to exposing unneeded fields to the
> > shrinkers, given they're already on the stack somewhere.
> 
> The only downside I can see is that it makes struct scan_control
> public - it'll need to be declared in a public header file so that all
> shrinkers can access it.

We've done worse things ;) Put it in scan_control.h and it will only be
exposed to code which has a legitimate need for it.

> Maybe one way to mitigate this would be if we can make the shrinker
> api take a *const* struct scan_control pointer as an argument, so that
> it'll be clear that we expect the shrinkers to only read the
> information in that struct.

Well, we might want callees to update fields in there, say "number of
bytes I managed to reclaim" or such.  We do that with
writeback_control.pages_skipped and it is comfortable enough.

--
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 policy in Canada: sign http://dissolvethecrtc.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>