Re: [PATCH] mm/memblock: skip kmemleak for kasan_init()

[Catalin Marinas]

On Wed, Nov 28, 2018 at 12:40:33PM -0500, Qian Cai wrote:
> Kmemleak does not play well with KASAN (tested on both HPE Apollo 70 and
> Huawei TaiShan 2280 aarch64 servers).
> 
> After calling start_kernel()->setup_arch()->kasan_init(), kmemleak early
> log buffer went from something like 280 to 260000 which caused kmemleak
> disabled and crash dump memory reservation failed. The multitude of
> kmemleak_alloc() calls is from,
> 
> for_each_memblock(memory, reg) x \
> while (pgdp++, addr = next, addr != end) x \
> while (pudp++, addr = next, addr != end && pud_none(READ_ONCE(*pudp))) \
> while (pmdp++, addr = next, addr != end && pmd_none(READ_ONCE(*pmdp))) \
> while (ptep++, addr = next, addr != end && pte_none(READ_ONCE(*ptep)))
> 
> Signed-off-by: Qian Cai <cai@xxxxxx>

Sorry, I didn't get the chance to investigate this further. Hopefully
early next week.

> diff --git a/mm/memblock.c b/mm/memblock.c
> index 9a2d5ae..fd78e39 100644
> --- a/mm/memblock.c
> +++ b/mm/memblock.c
> @@ -1412,6 +1412,8 @@ static void * __init memblock_alloc_internal(
>  done:
>  	ptr = phys_to_virt(alloc);
>  
> +/* Skip kmemleak for kasan_init() due to high volume. */
> +#ifndef CONFIG_KASAN
>  	/*
>  	 * The min_count is set to 0 so that bootmem allocated blocks
>  	 * are never reported as leaks. This is because many of these blocks
> @@ -1419,6 +1421,7 @@ static void * __init memblock_alloc_internal(
>  	 * looked up by kmemleak.
>  	 */
>  	kmemleak_alloc(ptr, size, 0, 0);
> +#endif

This may render kmemleak unusable since it is not aware of the memblock
allocations and it would trigger lots of false positives.

-- 
Catalin