mm/page_alloc: optimize code layout for __alloc_pages_bulk
authorJesper Dangaard Brouer <brouer@redhat.com>
Fri, 30 Apr 2021 06:01:51 +0000 (23:01 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 30 Apr 2021 18:20:43 +0000 (11:20 -0700)
Looking at perf-report and ASM-code for __alloc_pages_bulk() it is clear
that the code activated is suboptimal.  The compiler guesses wrong and
places unlikely code at the beginning.  Due to the use of WARN_ON_ONCE()
macro the UD2 asm instruction is added to the code, which confuse the
I-cache prefetcher in the CPU.

[mgorman@techsingularity.net: minor changes and rebasing]

Link: https://lkml.kernel.org/r/20210325114228.27719-5-mgorman@techsingularity.net
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Reviewed-by: Alexander Lobakin <alobakin@pm.me>
Acked-By: Vlastimil Babka <vbabka@suse.cz>
Cc: Alexander Duyck <alexander.duyck@gmail.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Chuck Lever <chuck.lever@oracle.com>
Cc: David Miller <davem@davemloft.net>
Cc: Ilias Apalodimas <ilias.apalodimas@linaro.org>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/page_alloc.c

index e240704..3f22916 100644 (file)
@@ -5042,7 +5042,7 @@ unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid,
        unsigned int alloc_flags = ALLOC_WMARK_LOW;
        int nr_populated = 0;
 
-       if (WARN_ON_ONCE(nr_pages <= 0))
+       if (unlikely(nr_pages <= 0))
                return 0;
 
        /*
@@ -5089,7 +5089,7 @@ unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid,
         * If there are no allowed local zones that meets the watermarks then
         * try to allocate a single page and reclaim if necessary.
         */
-       if (!zone)
+       if (unlikely(!zone))
                goto failed;
 
        /* Attempt the batch allocation */
@@ -5107,7 +5107,7 @@ unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid,
 
                page = __rmqueue_pcplist(zone, ac.migratetype, alloc_flags,
                                                                pcp, pcp_list);
-               if (!page) {
+               if (unlikely(!page)) {
                        /* Try and get at least one page */
                        if (!nr_populated)
                                goto failed_irq;