Commit 3bb7dc3b authored Jul 02, 2020 by junj Committed by Charan Teja Reddy Jul 03, 2020

Revert "mm, page_alloc: fix core hung in free_pcppages_bulk()"



This reverts commit c1d75807 ("mm, page_alloc: fix core hung in
free_pcppages_bulk()) as it is causing the below crash.

(new_slab) from [<c02d728c>]
(__kmem_cache_create) from [<c1a22468>] (create_boot_cache+0x6c/0x94)
(create_boot_cache) from [<c1a24e38>] (kmem_cache_init+0x74/0x16c)
(kmem_cache_init) from [<c1a009d0>] (start_kernel+0x340/0x680)
(start_kernel) from [<00000000>] (  (null))

Reason being that pages are not being initialized properly as they first
goto pcplists rather than buddy explained in the below call flow:
Start_kernel() -->
	build_all_zonelists()-> pcp's ->batch = 1, ->high = 0,
	mm_init()->mem_init()->free_all_bootmem()--> add order-0 pages to
				pcp lists rather then to buddy.

	alloc_pages() -> This will cause crash as some of the page
			struct are not reliable.

Change-Id: I693fca5c7ef7f1d034355fc761fa623d4e78227d
Signed-off-by: junj <junj@codeaurora.org>
[charante@codeaurora.org]: updated the details in commit message
Signed-off-by: Charan Teja Reddy <charante@codeaurora.org>

parent 67938f3f

mm/page_alloc.c

+4 −4

Original line number	Diff line number	Diff line
		@@ -3036,7 +3036,7 @@ static void free_unref_page_commit(struct page *page, unsigned long pfn)
		{
		struct zone *zone = page_zone(page);
		struct per_cpu_pages *pcp;
		int migratetype, high, batch;
		int migratetype;

		migratetype = get_pcppage_migratetype(page);
		__count_vm_event(PGFREE);
		@@ -3059,11 +3059,11 @@ static void free_unref_page_commit(struct page *page, unsigned long pfn)
		pcp = &this_cpu_ptr(zone->pageset)->pcp;
		list_add(&page->lru, &pcp->lists[migratetype]);
		pcp->count++;
		high = READ_ONCE(pcp->high);
		batch = READ_ONCE(pcp->batch);
		if ((high > batch) && pcp->count >= high)
		if (pcp->count >= pcp->high) {
		unsigned long batch = READ_ONCE(pcp->batch);
		free_pcppages_bulk(zone, batch, pcp);
		}
		}

		/*
		* Free a 0-order page