Re: [PATCH] mm: hugetlb: flush dcache before returning zeroed hugepage to userspace

[Will Deacon]

On Tue, Jul 10, 2012 at 10:45:13AM +0100, Will Deacon wrote:
> On Tue, Jul 10, 2012 at 12:57:14AM +0100, Hugh Dickins wrote:
> > If I start to grep the architectures for non-empty flush_dcache_page(),
> > I soon find things in arch/arm such as v4_mc_copy_user_highpage() doing
> > if (!test_and_set_bit(PG_dcache_clean,)) __flush_dcache_page() - where
> > the naming suggests that I'm right, it's the architecture's responsibility
> > to arrange whatever flushing is needed in its copy and clear page functions.
> 
> On ARM the flushing is there to deal with dcache aliasing and highmem, so the
> clear/copy functions won't actually do explicit flushing on modern (ARMv7)
> cores. Instead we flush the page when writing the pte and noticing that
> PG_arch_1 (PG_dcache_clean) is clear...
> 
> ...so the real question is why this wasn't being triggered for huge pages.
> I'll go and take another look since I would expect PG_arch_1 to be cleared
> for pages coming back from alloc_huge_page.

Ok, so this is exactly the problem. The hugetlb allocator uses its own
pool of huge pages, so free_huge_page followed by a later alloc_huge_page
will give you something where the page flags of the compound head do not
guarantee that PG_arch_1 is clear.

I tried hacking arch_release_hugepage to clear the bit, but that's only
called when actually releasing the hugepages via __free_pages which is
precisely the case that works correctly anyway!

Will
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/