Re: [PATCH] x86/mm/pat: take cpa_lock around large-page collapse

[Mike Rapoport <rppt@kernel.org>]

On Fri, Jul 10, 2026 at 10:13:21AM -0700, Dave Hansen wrote:
> On 6/26/26 09:32, Denis V. Lunev wrote:
> > +	/*
> > +	 * debug_pagealloc bypasses cpa_lock, so __change_page_attr() walks
> > +	 * unserialized and freeing collapsed PTE-tables could race it; skip
> > +	 * the optional merge there.
> > +	 */
> > +	if (debug_pagealloc_enabled())
> > +		return;
> 
> Wow, that debug_pagealloc hack is ancient and came with the original
> introduction of cpa_lock:
> 
> > commit ad5ca55f6bdb47c957b681c7358bb3719ba4ee82
> > Author: Suresh Siddha <suresh.b.siddha@intel.com>
> > Date:   Tue Sep 23 14:00:42 2008 -0700
> > 
> >     x86, cpa: srlz cpa(), global flush tlb after splitting big page and before doing cpa
> 
> My only question is *why*!?!? Why add extra locking complexity and rules
> to optimize debug_pagealloc, which is already horrendously slow.

My guess would be that the logic was to lock only when a split is possible
and since debug_pagealloc does not split anything, skip the lock.
 
> I kinda think we should just _remove_ the locking which is conditional
> on debug_pagealloc_enabled().

I agree.

-- 
Sincerely yours,
Mike.