Re: [linus:master] [mm/vmalloc] a061578043: BUG:spinlock_trylock_failure_on_UP_on_CPU

On Thu, Dec 11, 2025 at 12:47:10PM -0800, Andrew Morton wrote:
> On Wed, 10 Dec 2025 15:30:51 -0800 "Vishal Moola (Oracle)" <[email protected]> wrote:
> 
> > > The kernel config and materials to reproduce are available at:
> > > https://download.01.org/0day-ci/archive/20251210/[email protected]
> > > 
> > 
> > Hmmm. This looks like a race condition tied to reclaim. I'm assuming
> > we fail to allocate a page and kick off kswapd. Then when we fall back
> > to the bulk allocator which tries to remove a pcp page at the same time as
> > kswapd tries to reclaim it. Maybe?
> > 
> > Does something like this fix it?
> 
> Concerned.  If pcp page removal is racing with kswapd reclaim then
> that's a problem.  Altering a caller so it doesn't tickle this bug
> doesn't fix the bug?

Agreed. I'll look into it. 

> > diff --git a/mm/vmalloc.c b/mm/vmalloc.c
> > index ecbac900c35f..0d1480723ddc 100644
> > --- a/mm/vmalloc.c
> > +++ b/mm/vmalloc.c
> > @@ -3634,7 +3634,7 @@ vm_area_alloc_pages(gfp_t gfp, int nid,
> >  	struct page *page;
> >  	int i;
> >  	unsigned int large_order = ilog2(nr_remaining);
> > -	gfp_t large_gfp = vmalloc_gfp_adjust(gfp, large_order) & ~__GFP_DIRECT_RECLAIM;
> > +	gfp_t large_gfp = vmalloc_gfp_adjust(gfp, large_order) & ~__GFP_RECLAIM;
> >  
> >  	large_order = min(max_attempt_order, large_order);