Re: [PATCH 0/3 v5] Introduce a bulk order-0 page allocator

On Tue, Mar 23, 2021 at 04:08:14PM +0100, Jesper Dangaard Brouer wrote:

On Tue, 23 Mar 2021 10:44:21 +0000
Mel Gorman [off-list ref] wrote:

quoted

On Mon, Mar 22, 2021 at 09:18:42AM +0000, Mel Gorman wrote:

quoted

This series is based on top of Matthew Wilcox's series "Rationalise
__alloc_pages wrapper" and does not apply to 5.12-rc2. If you want to
test and are not using Andrew's tree as a baseline, I suggest using the
following git tree

git://git.kernel.org/pub/scm/linux/kernel/git/mel/linux.git mm-bulk-rebase-v5r9
  

Jesper and Chuck, would you mind rebasing on top of the following branch
please? 

git://git.kernel.org/pub/scm/linux/kernel/git/mel/linux.git mm-bulk-rebase-v6r2

The interface is the same so the rebase should be trivial.

Jesper, I'm hoping you see no differences in performance but it's best
to check.

I will rebase and check again.

The current performance tests that I'm running, I observe that the
compiler layout the code in unfortunate ways, which cause I-cache
performance issues.  I wonder if you could integrate below patch with
your patchset? (just squash it)

Yes but I'll keep it as a separate patch that is modified slightly.
Otherwise it might get "fixed" as likely/unlikely has been used
inappropriately in the past. If there is pushback, I'll squash them
together.

Ok, I can make this change. It was a defensive check for the new callers
in case insane values were being passed in. 

-	if (WARN_ON_ONCE(page_list && !list_empty(page_list)))
+	if (unlikely(page_list && !list_empty(page_list)))
 		return 0;
 
 	/* Skip populated array elements. */

FWIW, this check is now gone. The list only had to be empty if
prep_new_page was deferred until IRQs were enabled to avoid accidentally
calling prep_new_page() on a page that was already on the list when
alloc_pages_bulk was called.

I'm dropping this because nr_pages == 1 is common for the sunrpc user.

Ok.

Hmmm, ok. It depends on memory pressure but I agree !page is unlikely.

Current version applied is

--8<--
mm/page_alloc: optimize code layout for __alloc_pages_bulk

From: Jesper Dangaard Brouer <redacted>

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: Minor changes and rebasing]
Signed-off-by: Jesper Dangaard Brouer <redacted>
Signed-off-by: Mel Gorman <redacted>

diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index be1e33a4df39..1ec18121268b 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c

@@ -5001,7 +5001,7 @@ int __alloc_pages_bulk(gfp_t gfp, int preferred_nid,
 	unsigned int alloc_flags;
 	int nr_populated = 0;
 
-	if (WARN_ON_ONCE(nr_pages <= 0))
+	if (unlikely(nr_pages <= 0))
 		return 0;
 
 	/*

@@ -5048,7 +5048,7 @@ int __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 */

@@ -5066,7 +5066,7 @@ int __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;