Re: [PATCH] vmalloc: Use atomic_long_add_return_relaxed()

[Mateusz Guzik <mjguzik@gmail.com>]

On Tue, Apr 15, 2025 at 01:26:46PM +0200, Uladzislau Rezki (Sony) wrote:
> Switch from the atomic_long_add_return() to its relaxed version.
> 
> We do not need a full memory barrier or any memory ordering during
> increasing the "vmap_lazy_nr" variable. What we only need is to do it
> atomically. This is what atomic_long_add_return_relaxed() guarantees.
> 
> AARCH64:
> 
> <snip>
> Default:
>     40ec:       d34cfe94        lsr     x20, x20, #12
>     40f0:       14000044        b       4200 <free_vmap_area_noflush+0x19c>
>     40f4:       94000000        bl      0 <__sanitizer_cov_trace_pc>
>     40f8:       90000000        adrp    x0, 0 <__traceiter_alloc_vmap_area>
>     40fc:       91000000        add     x0, x0, #0x0
>     4100:       f8f40016        ldaddal x20, x22, [x0]
>     4104:       8b160296        add     x22, x20, x22
> 
> Relaxed:
>     40ec:       d34cfe94        lsr     x20, x20, #12
>     40f0:       14000044        b       4200 <free_vmap_area_noflush+0x19c>
>     40f4:       94000000        bl      0 <__sanitizer_cov_trace_pc>
>     40f8:       90000000        adrp    x0, 0 <__traceiter_alloc_vmap_area>
>     40fc:       91000000        add     x0, x0, #0x0
>     4100:       f8340016        ldadd   x20, x22, [x0]
>     4104:       8b160296        add     x22, x20, x22
> <snip>
> 
> Signed-off-by: Uladzislau Rezki (Sony) <urezki@gmail.com>
> ---
>  mm/vmalloc.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/mm/vmalloc.c b/mm/vmalloc.c
> index 7bb32f498d39..9d4027041a3f 100644
> --- a/mm/vmalloc.c
> +++ b/mm/vmalloc.c
> @@ -2370,7 +2370,7 @@ static void free_vmap_area_noflush(struct vmap_area *va)
>  	if (WARN_ON_ONCE(!list_empty(&va->list)))
>  		return;
>  
> -	nr_lazy = atomic_long_add_return(va_size(va) >> PAGE_SHIFT,
> +	nr_lazy = atomic_long_add_return_relaxed(va_size(va) >> PAGE_SHIFT,
>  					 &vmap_lazy_nr);
>  
>  	/*

If touching this, maybe a step further -- I see false sharing with:
nm vmlinux | sort -k 1

[snip]
ffffffff845cb240 b vmap_lazy_nr
ffffffff845cb248 b nr_vmalloc_pages
ffffffff845cb260 b single
[/snip]

If this is used enough to warrant messing with eliding a fence, then it
probably also wants to elide some false sharing.

So at least the first two should be annotated with __cacheline_aligned_in_smp?

I am not going to push though, feel free to ignore.