Re: [COUNTERPATCH] mm: avoid overflowing preempt_count() in mmu_take_all_locks()

[Andrea Arcangeli <aarcange@redhat.com>]

On Thu, Apr 01, 2010 at 06:12:34PM +0200, Peter Zijlstra wrote:
> One thing we can do there is to mutex_trylock() if we get the lock, see
> if we've got the right object, if the trylock fails we can do the
> refcount thing and sleep. That would allow the fast-path to remain a
> single atomic.

But then how do you know which anon_vma_unlink has to decrease the
refcount and which not? That info should need to be stored in the
kernel stack, can't be stored in the vma. I guess it's feasible but
passing that info around sounds more tricky than the trylock itself
(adding params to those functions with int &refcount).

> The only thing is growing that anon_vma struct, but KSM seems to already
> do that for the most common .configs.

Ok, from my initial review of memory compaction I see that it already
adds its own recount and unifies it too with the ksm_refcount. So it's
worth stop calling it ksm_refcount and to remove the #ifdef (which
memory compaction didn't remove but added an defined(CONFIG_KSM) ||
defined(CONFIG_MIGRATION)).

If you do this change we can have your change applied first, and then
Mel can adapt memory compaction to it sharing the same unconditional
compiled-in refcount.