Re: [PATCH] mm/percpu.c: change GFP_KERNEL to GFP_ATOMIC

[Baoquan He]

On 07/18/23 at 10:17am, Andrew Morton wrote:
> On Tue, 18 Jul 2023 14:35:11 +0800 sunran001@xxxxxxxxxx wrote:
> 
> > ERROR: function pcpu_balance_populated called on line 2238 inside lock
> > on line 2234 but uses GFP_KERNEL
> > 
> > Generated by: scripts/coccinelle/locks/call_kern.cocci
> > 
> > ...
> >
> > --- a/mm/percpu.c
> > +++ b/mm/percpu.c
> > @@ -2033,7 +2033,7 @@ static void pcpu_balance_free(bool empty_only)
> >   static void pcpu_balance_populated(void)
> >   {
> >       /* gfp flags passed to underlying allocators */
> > -    const gfp_t gfp = GFP_KERNEL | __GFP_NORETRY | __GFP_NOWARN;
> > +    const gfp_t gfp = GFP_ATOMIC | __GFP_NORETRY | __GFP_NOWARN;
> >       struct pcpu_chunk *chunk;
> >       int slot, nr_to_pop, ret;
> 
> I don't believe this warning is correct.  

Yeah, the warning is false positive. It releases the lock when
requesting memory allocation, then take the lock again.

> 
> 			spin_unlock_irq(&pcpu_lock);
> 			ret = pcpu_populate_chunk(chunk, rs, rs + nr, gfp);
> 			cond_resched();
> 			spin_lock_irq(&pcpu_lock);
>