Re: [PATCH] mm/percpu: Fix gfp flag in pcpu_balance_populated

[Dennis Zhou <dennis@kernel.org>]

Hello,

On Fri, Jun 18, 2021 at 11:14:36PM +0800, Wan Jiabing wrote:
> Fix coccicheck warning:
> 
> ./mm/percpu.c:2045:19-29: ERROR: function pcpu_balance_populated
> called on line 2232 inside lock on line 2228 but uses GFP_KERNEL
> 
> When pcpu_balance_populated() is called in pcpu_balance_workfn(),
> it helds spin_lock but use GFP_KERNEL to alloc mem, which is unsafe.
> 
> Signed-off-by: Wan Jiabing <wanjiabing@vivo.com>
> ---
>  mm/percpu.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/mm/percpu.c b/mm/percpu.c
> index b4cebeca4c0c..4031f32e6975 100644
> --- a/mm/percpu.c
> +++ b/mm/percpu.c
> @@ -2042,7 +2042,7 @@ 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;
>  
> -- 
> 2.30.2
> 

In both places gfp flags are passed, the pcpu_lock is dropped. So I
think this is an issue with coccicheck. Regardless, the fix wouldn't be
to switch to GFP_ATOMIC but to make the locking correct.

Thanks,
Dennis