Received-SPF: pass (google.com: domain of linux-nfs-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Subject: [PATCH RFC net-next 3/3] mm: make zone->free_area[order] access
 faster
From:   Jesper Dangaard Brouer <brouer@redhat.com>
To:     Mel Gorman <mgorman@techsingularity.net>, linux-mm@kvack.org
Cc:     Jesper Dangaard Brouer <brouer@redhat.com>, chuck.lever@oracle.com,
        netdev@vger.kernel.org, linux-nfs@vger.kernel.org,
        linux-kernel@vger.kernel.org
Date:   Wed, 24 Feb 2021 19:56:51 +0100
Message-ID: <161419301128.2718959.4838557038019199822.stgit@firesoul>
In-Reply-To: <161419296941.2718959.12575257358107256094.stgit@firesoul>
References: <161419296941.2718959.12575257358107256094.stgit@firesoul>
User-Agent: StGit/0.19
MIME-Version: 1.0
Content-Type: text/plain; charset="utf-8"
Content-Transfer-Encoding: 8bit
Precedence: bulk

Avoid multiplication (imul) operations when accessing:
 zone->free_area[order].nr_free

This was really tricky to find. I was puzzled why perf reported that
rmqueue_bulk was using 44% of the time in an imul operation:

       │     del_page_from_free_list():
 44,54 │ e2:   imul   $0x58,%rax,%rax

This operation was generated (by compiler) because the struct free_area have
size 88 bytes or 0x58 hex. The compiler cannot find a shift operation to use
and instead choose to use a more expensive imul, to find the offset into the
array free_area[].

The patch align struct free_area to a cache-line, which cause the
compiler avoid the imul operation. The imul operation is very fast on
modern Intel CPUs. To help fast-path that decrement 'nr_free' move the
member 'nr_free' to be first element, which saves one 'add' operation.

Looking up instruction latency this exchange a 3-cycle imul with a
1-cycle shl, saving 2-cycles. It does trade some space to do this.

Used: gcc (GCC) 9.3.1 20200408 (Red Hat 9.3.1-2)

Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
---
 include/linux/mmzone.h |    6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
index b593316bff3d..4d83201717e1 100644
--- a/include/linux/mmzone.h
+++ b/include/linux/mmzone.h
@@ -93,10 +93,12 @@ extern int page_group_by_mobility_disabled;
 #define get_pageblock_migratetype(page)					\
 	get_pfnblock_flags_mask(page, page_to_pfn(page), MIGRATETYPE_MASK)
 
+/* Aligned struct to make zone->free_area[order] access faster */
 struct free_area {
-	struct list_head	free_list[MIGRATE_TYPES];
 	unsigned long		nr_free;
-};
+	unsigned long		__pad_to_align_free_list;
+	struct list_head	free_list[MIGRATE_TYPES];
+}  ____cacheline_aligned_in_smp;
 
 static inline struct page *get_page_from_free_area(struct free_area *area,
 					    int migratetype)