lkml.org 
[lkml]   [2008]   [Mar]   [10]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: quicklists confuse meminfo
Hugh Dickins wrote:
> On Mon, 10 Mar 2008, Andi Kleen wrote:
>
>> Christoph Lameter <clameter@sgi.com> writes:
>>
>>> Zeroed pages however will not address the issue of having initialized pgd
>>> (which seems to be what i386 needs).
>>>
>> pgd is tiny on i386 PAE (4 * 16 bytes). Are you sure reinitializing that
>> is a serious issue? ...
>>
>
> It used to be tiny (32 aligned bytes), then 2.6.22's quicklist enlarged
> that to a whole (lowmem) page. I think we were all too busy with other
> stuff to protest loudly enough about that bloat.
>

Yes, I was surprised about the pgd page allocation. It's currently
necessary for Xen (but I think I can fix that easily enough). But I was
especially surprised it was imposed for everyone.

> If the quicklists are going, it'd be good for PAE to go back to a
> kmem_cache of 32-byte entries as in 2.6.21 - I think Ingo's patch is
> still using a whole page there.
>

+1. We'd still need to maintain a list to link all the pgds together,
but I think we can just allocate that out of the cache too (either the
same object or a separate pgd list cache).

> Or have sl?b alignment changes, or virtualization issues (locking
> per underlying struct page?), made a kmem_cache awkward there now?
>
I think only Xen has a constraint. At the moment we rely on page-sized
pgds for two things:

1. Xen marks the whole page as being of pgd-type, and so it can't
have non-pgd contents (as the page would be RO, and any contents
would be validated as pgd entries). However I can fix that by
maintaining a separate per-cpu pgd page, and just copy the four
entries over when cr3 is reloaded. This would move the
Xen-specific requirements into the Xen code without affecting the
rest of the kernel.
2. We still need to maintain a list of pgds, as I discussed above.

So from my perspective there are no insoluble problems, and I'd fully
support the transition.

I've been working on unifying the pgalloc stuff, and the quicklist
(i386) vs non-quicklist (x86-64) use has been a bit of a thorn in my
side. Of course there'll still be the page-sized pgd vs non-page-sized
pgd, but we just have to live with that.

J


\
 
 \ /
  Last update: 2008-03-10 18:35    [W:0.113 / U:0.792 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site