Re: [PATCH 2/5] mm: compaction: simplify should_compact_retry()

[Vlastimil Babka]

On 5/19/23 14:39, Johannes Weiner wrote:
> The different branches for retry are unnecessarily complicated. There
> are really only three outcomes: progress (retry n times), skipped
> (retry if reclaim can help), failed (retry with higher priority).
> 
> Rearrange the branches and the retry counter to make it simpler.
> 
> v2:
> - fix trace point build (Mel)
> - fix max_retries logic for costly allocs (Huang)
> 
> Signed-off-by: Johannes Weiner <hannes@xxxxxxxxxxx>
> ---
>  mm/page_alloc.c | 53 +++++++++++++++----------------------------------
>  1 file changed, 16 insertions(+), 37 deletions(-)
> 
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index 5a84a0bebc37..72660e924b95 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -3772,16 +3772,22 @@ should_compact_retry(struct alloc_context *ac, int order, int alloc_flags,
>  	 * Compaction managed to coalesce some page blocks, but the
>  	 * allocation failed presumably due to a race. Retry some.
>  	 */
> -	if (compact_result == COMPACT_SUCCESS)
> -		(*compaction_retries)++;
> +	if (compact_result == COMPACT_SUCCESS) {
> +		/*
> +		 * !costly requests are much more important than
> +		 * __GFP_RETRY_MAYFAIL costly ones because they are de
> +		 * facto nofail and invoke OOM killer to move on while
> +		 * costly can fail and users are ready to cope with
> +		 * that. 1/4 retries is rather arbitrary but we would
> +		 * need much more detailed feedback from compaction to
> +		 * make a better decision.
> +		 */
> +		if (order > PAGE_ALLOC_COSTLY_ORDER)
> +			max_retries /= 4;
>  
> -	/*
> -	 * All zones were scanned completely and still no result. It
> -	 * doesn't really make much sense to retry except when the
> -	 * failure could be caused by insufficient priority
> -	 */
> -	if (compact_result == COMPACT_COMPLETE)
> -		goto check_priority;
> +		ret = ++(*compaction_retries) <= max_retries;
> +		goto out;

I think you simplified this part too much, so now once it runs out of
retries, it will return false, while previously it would increase the priority.

> +	}
>  
>  	/*
>  	 * Compaction was skipped due to a lack of free order-0
> @@ -3793,35 +3799,8 @@ should_compact_retry(struct alloc_context *ac, int order, int alloc_flags,
>  	}
>  
>  	/*
> -	 * If compaction backed due to being deferred, due to
> -	 * contended locks in async mode, or due to scanners meeting
> -	 * after a partial scan, retry with increased priority.
> -	 */
> -	if (compact_result == COMPACT_DEFERRED ||
> -	    compact_result == COMPACT_CONTENDED ||
> -	    compact_result == COMPACT_PARTIAL_SKIPPED)
> -		goto check_priority;
> -
> -	/*
> -	 * !costly requests are much more important than __GFP_RETRY_MAYFAIL
> -	 * costly ones because they are de facto nofail and invoke OOM
> -	 * killer to move on while costly can fail and users are ready
> -	 * to cope with that. 1/4 retries is rather arbitrary but we
> -	 * would need much more detailed feedback from compaction to
> -	 * make a better decision.
> -	 */
> -	if (order > PAGE_ALLOC_COSTLY_ORDER)
> -		max_retries /= 4;
> -	if (*compaction_retries <= max_retries) {
> -		ret = true;
> -		goto out;
> -	}
> -
> -	/*
> -	 * Make sure there are attempts at the highest priority if we exhausted
> -	 * all retries or failed at the lower priorities.
> +	 * Compaction failed. Retry with increasing priority.
>  	 */
> -check_priority:
>  	min_priority = (order > PAGE_ALLOC_COSTLY_ORDER) ?
>  			MIN_COMPACT_COSTLY_PRIORITY : MIN_COMPACT_PRIORITY;
>