Re: [PATCH v3 2/4] mm: failfast mode with __GFP_NORETRY in alloc_contig_range

[David Hildenbrand <david@redhat.com>]

On 13.01.21 02:21, Minchan Kim wrote:
> Contiguous memory allocation can be stalled due to waiting
> on page writeback and/or page lock which causes unpredictable
> delay. It's a unavoidable cost for the requestor to get *big*
> contiguous memory but it's expensive for *small* contiguous
> memory(e.g., order-4) because caller could retry the request
> in diffrent range where would have easy migratable pages
> without stalling.

s/diffrent/different/

> 
> This patch introduce __GFP_NORETRY as compaction gfp_mask in
> alloc_contig_range so it will fail fast without blocking
> when it encounters pages needed waitting.

s/waitting/waiting/

> 
> Signed-off-by: Minchan Kim <minchan@kernel.org>
> ---
>  mm/page_alloc.c | 8 ++++++--
>  1 file changed, 6 insertions(+), 2 deletions(-)
> 
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index 5b3923db9158..ff41ceb4db51 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -8489,12 +8489,16 @@ static int __alloc_contig_migrate_range(struct compact_control *cc,
>  	unsigned int nr_reclaimed;
>  	unsigned long pfn = start;
>  	unsigned int tries = 0;
> +	unsigned int max_tries = 5;
>  	int ret = 0;
>  	struct migration_target_control mtc = {
>  		.nid = zone_to_nid(cc->zone),
>  		.gfp_mask = GFP_USER | __GFP_MOVABLE | __GFP_RETRY_MAYFAIL,
>  	};
>  
> +	if (cc->alloc_contig && cc->mode == MIGRATE_ASYNC)
> +		max_tries = 1;
> +
>  	migrate_prep();
>  
>  	while (pfn < end || !list_empty(&cc->migratepages)) {
> @@ -8511,7 +8515,7 @@ static int __alloc_contig_migrate_range(struct compact_control *cc,
>  				break;
>  			}
>  			tries = 0;
> -		} else if (++tries == 5) {
> +		} else if (++tries == max_tries) {
>  			ret = ret < 0 ? ret : -EBUSY;
>  			break;
>  		}
> @@ -8562,7 +8566,7 @@ int alloc_contig_range(unsigned long start, unsigned long end,
>  		.nr_migratepages = 0,
>  		.order = -1,
>  		.zone = page_zone(pfn_to_page(start)),
> -		.mode = MIGRATE_SYNC,
> +		.mode = gfp_mask & __GFP_NORETRY ? MIGRATE_ASYNC : MIGRATE_SYNC,
>  		.ignore_skip_hint = true,
>  		.no_set_skip_hint = true,
>  		.gfp_mask = current_gfp_context(gfp_mask),
> 

I'm fine with using gfp flags (e.g., __GFP_NORETRY) as long as they
don't enable other implicit behavior (e.g., move draining X to the
caller) that's hard to get from the flag name.

IMHO, if we ever want to move draining to the caller, or change the
behavior of alloc_contig_range() in different ways (e.g., disable PCP),
we won't get around introducing a separate set of flags for
alloc_contig_range().

Let's see what Michal thinks. Thanks!

-- 
Thanks,

David / dhildenb