Re: [RFC][PATCH] proc: export more page flags in /proc/kpageflags

[KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>]

> On Tue, Apr 14, 2009 at 03:54:40PM +0900, KOSAKI Motohiro wrote:
> > Hi
> 
> There are two use cases here:
> 
> First what is useful for the administrator as a general abstraction.
> And what is useful for the kernel hacker for debugging.
> 
> The kernel hacker wants everything even if it's subject to change,
> the administrator wants a higher level abstraction they can make
> sense of and that doesn't change too often.
> 
> I think there's a case for both usages, but perhaps they 
> should be separated (in a public and a internal interface perhaps?)
> 
> My comments below are about abstractions for the first case.
> 
> 
> > 
> > > On Tue, Apr 14, 2009 at 12:37:10PM +0800, KOSAKI Motohiro wrote:
> > > > > Export the following page flags in /proc/kpageflags,
> > > > > just in case they will be useful to someone:
> > > > > 
> > > > > - PG_swapcache
> > > > > - PG_swapbacked
> > > > > - PG_mappedtodisk
> > > > > - PG_reserved
> 
> PG_reserved should be exported as PG_KERNEL or somesuch.

OK.

rest problem is, how do we write document this.
PG_reserved have multiple meanings...


> > > > > - PG_private
> > > > > - PG_private_2
> > > > > - PG_owner_priv_1
> > > > > 
> > > > > - PG_head
> > > > > - PG_tail
> > > > > - PG_compound
> 
> I would combine these three into a pseudo "large page" flag.

Ah good idea.

> > > > > 
> > > > > - PG_unevictable
> > > > > - PG_mlocked
> > > > > 
> > > > > - PG_poison
> 
> PG_poison is also useful to export. But since it depends on my
> patchkit I will pull a patch for that into the HWPOISON series.

Yes, I agree.

> > > > > - PG_unevictable
> > > > > - PG_mlocked
> > 
> > this 9 flags shouldn't exported.
> > I can't imazine administrator use what purpose those flags.
> 
> I think an abstraced "PG_pinned" or somesuch flag that combines
> page lock, unevictable, mlocked would be useful for the administrator.

PG_unevictable and PG_mlocked have a bit delicate meaning.
it gurantee the page isn't evicted. but mlock(2) don't gurantee
turn page on PG_mlocked. some race prevent it.

I'm afraid administrator confuse it. but if someone can write good
document, my worriness will vanished.