Re: [PATCH 0/5] mm: tracking dirty pages -v11

[Hugh Dickins <hugh@veritas.com>]

On Sat, 24 Jun 2006, Peter Zijlstra wrote:
> 
> I hope to have addressed all Hugh's latest comments in this version.
> Its against 2.6.17-mm1, however I wasted most of the day trying to 
> test it on that kernel. But due to various circumstances that failed.

Looks good - I'm happy that we leave the do_wp_page test reordering
(to fix up that third order ptrace poke issue) to a subsequent patch,
it's better separated.

> So I've tested something like this against something 2.6.17'ish and 
> respun against the -mm lineup.

Your next (final?) spin should be against Linus' current git tree,
http://ftp.kernel.org/pub/linux/kernel/v2.6/snapshots/patch-2.6.17-git10.bz2
is the latest snapshot patch if you're not using git itself.  That will
suit Andrew better too: he prefers patches against Linus' current tree,
except when the changes are to work that's only in -mm.

You ought to respin, because the vma_wants_writenotify mods in mprotect.c
affect later patches in your series, giving rejects at present.  It does
look _much_ better with Linus' vma_wants_writenotify.  I did think of
asking you for that, but it seemed unfair because I knew you'd want
to use it in mprotect, and then get in trouble with backing-dev.h:
which you've solved by #including that now in mm.h - a pity,
but an unavoidable decision.

Given the reordering you had to make in mprotect_fixup to get its tests
working right (a little naughty!), I'd now do away with the "mask"
variable, and just work directly on "newflags" itself; but up to you.

> I've taken Hugh's msync changes too, looks a lot better and does indeed
> fix some boundary cases.

Thanks for reviewing: please add my
Signed-off-by: Hugh Dickins <hugh@veritas.com>
to that msync one.

In the respin of 1/5 you enquired:
> Bah Bah Bah, why didn't the page_mkwrite() patch re-protect clean pages?
> And is it a Bad-Thing (tm) that that can happen now?

You'll need a reply from David for the definitive answer, but I think
page_mkwrite is only wanting to know about the _first_ write to the
page e.g. so that it can allocate space on disk for that page.  And
many (most) calls to page_mkwrite won't be for that first write at
all, the filesystem already has to work out the irrelevant calls:
so it's no great problem that you'll be making some more such calls.

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/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>