Re: [PATCH 2/9] mm/page_alloc: Add a bulk page allocator

On Mon, Apr 12, 2021 at 12:21:42PM +0200, Vlastimil Babka wrote:

quoted

diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 8a3e13277e22..eb547470a7e4 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c

@@ -4965,6 +4965,124 @@ static inline bool prepare_alloc_pages(gfp_t gfp_mask, unsigned int order,
 	return true;
 }
 
+/*
+ * __alloc_pages_bulk - Allocate a number of order-0 pages to a list
+ * @gfp: GFP flags for the allocation
+ * @preferred_nid: The preferred NUMA node ID to allocate from
+ * @nodemask: Set of nodes to allocate from, may be NULL
+ * @nr_pages: The number of pages desired on the list
+ * @page_list: List to store the allocated pages
+ *
+ * This is a batched version of the page allocator that attempts to
+ * allocate nr_pages quickly and add them to a list.
+ *
+ * Returns the number of pages on the list.
+ */
+int __alloc_pages_bulk(gfp_t gfp, int preferred_nid,
+			nodemask_t *nodemask, int nr_pages,
+			struct list_head *page_list)
+{
+	struct page *page;
+	unsigned long flags;
+	struct zone *zone;
+	struct zoneref *z;
+	struct per_cpu_pages *pcp;
+	struct list_head *pcp_list;
+	struct alloc_context ac;
+	gfp_t alloc_gfp;
+	unsigned int alloc_flags;

Was going to complain that this is not set to ALLOC_WMARK_LOW. Must be faster
next time...

Good that you caught it anyway!

quoted

+	int allocated = 0;
+
+	if (WARN_ON_ONCE(nr_pages <= 0))
+		return 0;
+
+	/* Use the single page allocator for one page. */
+	if (nr_pages == 1)
+		goto failed;
+
+	/* May set ALLOC_NOFRAGMENT, fragmentation will return 1 page. */

I don't understand this comment. Only alloc_flags_nofragment() sets this flag
and we don't use it here?

It's there as a reminder that there are non-obvious consequences
to ALLOC_NOFRAGMENT that may affect the bulk allocation success
rate. __rmqueue_fallback will only select pageblock_order pages and if that
fails, we fall into the slow path that allocates a single page. I didn't
deal with it because it was not obvious that it's even relevant but I bet
in 6 months time, I'll forget that ALLOC_NOFRAGMENT may affect success
rates without the comment. I'm waiting for a bug that can trivially trigger
a case with a meaningful workload where the success rate is poor enough to
affect latency before adding complexity. Ideally by then, the allocation
paths would be unified a bit better.

quoted

+	gfp &= gfp_allowed_mask;
+	alloc_gfp = gfp;
+	if (!prepare_alloc_pages(gfp, 0, preferred_nid, nodemask, &ac, &alloc_gfp, &alloc_flags))
+		return 0;
+	gfp = alloc_gfp;
+
+	/* Find an allowed local zone that meets the high watermark. */

Should it say "low watermark"?

Yeah, that's leftover from an earlier prototype :(

-- 
Mel Gorman
SUSE Labs