On Mon, 25 Mar 2019 at 15:58, Mel Gorman <mgorman@xxxxxxxxxxxxxxxxxxx> wrote:
>
> Ok, it's somewhat of a pity that we don't know what PFN that page
> corresponds to. Specifically it would be interesting to know if the PFN
> corresponds to a memory hole as DMA32 on your machine has a number of
> gaps. What I'm wondering is if the reinit fails to find good starting
> points that it picks a PFN that corresponds to an uninitialised page and
> trips up later.
>
> Can you try again with this patch please? It replaces the failed patch
> entirely.
>
> Thanks.
>
> diff --git a/mm/compaction.c b/mm/compaction.c
> index f171a83707ce..caac4b07eb33 100644
> --- a/mm/compaction.c
> +++ b/mm/compaction.c
> @@ -242,6 +242,7 @@ __reset_isolation_pfn(struct zone *zone, unsigned long pfn, bool check_source,
> bool check_target)
> {
> struct page *page = pfn_to_online_page(pfn);
> + struct page *block_page;
> struct page *end_page;
> unsigned long block_pfn;
>
> @@ -267,20 +268,26 @@ __reset_isolation_pfn(struct zone *zone, unsigned long pfn, bool check_source,
> get_pageblock_migratetype(page) != MIGRATE_MOVABLE)
> return false;
>
> + /* Ensure the start of the pageblock or zone is online and valid */
> + block_pfn = pageblock_start_pfn(pfn);
> + block_page = pfn_to_online_page(max(block_pfn, zone->zone_start_pfn));
> + if (block_page) {
> + page = block_page;
> + pfn = block_pfn;
> + }
> +
> + /* Ensure the end of the pageblock or zone is online and valid */
> + block_pfn += pageblock_nr_pages;
> + block_pfn = min(block_pfn, zone_end_pfn(zone));
> + end_page = pfn_to_online_page(block_pfn);
> + if (!end_page)
> + return false;
> +
> /*
> * Only clear the hint if a sample indicates there is either a
> * free page or an LRU page in the block. One or other condition
> * is necessary for the block to be a migration source/target.
> */
> - block_pfn = pageblock_start_pfn(pfn);
> - pfn = max(block_pfn, zone->zone_start_pfn);
> - page = pfn_to_page(pfn);
> - if (zone != page_zone(page))
> - return false;
> - pfn = block_pfn + pageblock_nr_pages;
> - pfn = min(pfn, zone_end_pfn(zone));
> - end_page = pfn_to_page(pfn);
> -
> do {
> if (pfn_valid_within(pfn)) {
> if (check_source && PageLRU(page)) {
> @@ -320,6 +327,16 @@ static void __reset_isolation_suitable(struct zone *zone)
>
> zone->compact_blockskip_flush = false;
>
> +
> + /*
> + * Re-init the scanners and attempt to find a better starting
> + * position below. This may result in redundant scanning if
> + * a better position is not found but it avoids the corner
> + * case whereby the cached PFNs are left in a memory hole with
> + * no proper struct page backing it.
> + */
> + reset_cached_positions(zone);
> +
> /*
> * Walk the zone and update pageblock skip information. Source looks
> * for PageLRU while target looks for PageBuddy. When the scanner
> @@ -349,13 +366,6 @@ static void __reset_isolation_suitable(struct zone *zone)
> zone->compact_cached_free_pfn = reset_free;
> }
> }
> -
> - /* Leave no distance if no suitable block was reset */
> - if (reset_migrate >= reset_free) {
> - zone->compact_cached_migrate_pfn[0] = migrate_pfn;
> - zone->compact_cached_migrate_pfn[1] = migrate_pfn;
> - zone->compact_cached_free_pfn = free_pfn;
> - }
> }
>
> void reset_isolation_suitable(pg_data_t *pgdat)
>
> --
> Mel Gorman
> SUSE Labs
Kernel panic are still occurs.
--
Best Regards,
Mike Gavrilov.
Attachment:
system-log12.tar.xz
Description: application/xz
