Re: [PATCHv3] mm: skip CMA pages when they are not available

[Matthew Wilcox <willy@infradead.org>]

On Mon, May 22, 2023 at 11:08:02AM +0800, zhaoyang.huang wrote:
> +static bool skip_cma(struct page *page, struct scan_control *sc)
> +{
> +	if (!current_is_kswapd() && gfp_migratetype(sc->gfp_mask) != MIGRATE_MOVABLE
> +		&& get_pageblock_migratetype(page) == MIGRATE_CMA)
> +		return true;

Putting the 'return' at the same level of indentation as the second half
of the conditional is wrong.  It confuses the reader.  Also, the &&
needs to go at the end of the line not the beginning (read the
codingstyle documentation!)  Also there's no good reason to use such
long lines.  ie do this instead:

	if (!current_is_kswapd() &&
			gfp_migratetype(sc->gfp_mask) != MIGRATE_MOVABLE &&
			get_pageblock_migratetype(page) == MIGRATE_CMA)
		return true;

if you prefer, this style of indent is also acceptable:

	if (!current_is_kswapd() &&
	    gfp_migratetype(sc->gfp_mask) != MIGRATE_MOVABLE &&
	    get_pageblock_migratetype(page) == MIGRATE_CMA)
		return true;

> @@ -2225,10 +2242,12 @@ static unsigned long isolate_lru_folios(unsigned long nr_to_scan,
>  	unsigned long nr_skipped[MAX_NR_ZONES] = { 0, };
>  	unsigned long skipped = 0;
>  	unsigned long scan, total_scan, nr_pages;
> +	struct page *page;

No, don't do this.

>  	LIST_HEAD(folios_skipped);
>  
>  	total_scan = 0;
>  	scan = 0;
> +

Don't add this completely unrelated whitespace change either.

>  	while (scan < nr_to_scan && !list_empty(src)) {
>  		struct list_head *move_to = src;
>  		struct folio *folio;
> @@ -2239,12 +2258,14 @@ static unsigned long isolate_lru_folios(unsigned long nr_to_scan,
>  		nr_pages = folio_nr_pages(folio);
>  		total_scan += nr_pages;
>  
> -		if (folio_zonenum(folio) > sc->reclaim_idx) {
> +		page = &folio->page;
> +
> +		if (folio_zonenum(folio) > sc->reclaim_idx
> +			|| skip_cma(page, sc)) {

Again, this is not where the || goes.

And skip_cma() should take a folio, not a page.  It's unreasonable
to ask you to convert get_pageblock_migratetype(),
get_pfnblock_flags_mask(), __get_pfnblock_flags_mask(), etc to
use a folio (... although someone looking for a project could do that
...).  Instead, you should do the folio->page conversion inside
skip_cma().

>  			nr_skipped[folio_zonenum(folio)] += nr_pages;
>  			move_to = &folios_skipped;
>  			goto move;
>  		}
> -

No unnecessary whitespace changes.