lkml.org 
[lkml]   [2020]   [Aug]   [30]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
SubjectRe: [PATCH v2 1/2] mm/pageblock: mitigation cmpxchg false sharing in pageblock flags
From
Date


在 2020/8/19 下午4:04, David Hildenbrand 写道:
> On 19.08.20 09:55, Anshuman Khandual wrote:
>>
>>
>> On 08/19/2020 11:17 AM, Alex Shi wrote:
>>> pageblock_flags is used as long, since every pageblock_flags is just 4
>>> bits, 'long' size will include 8(32bit machine) or 16 pageblocks' flags,
>>> that flag setting has to sync in cmpxchg with 7 or 15 other pageblock
>>> flags. It would cause long waiting for sync.
>>>
>>> If we could change the pageblock_flags variable as char, we could use
>>> char size cmpxchg, which just sync up with 2 pageblock flags. it could
>>> relief much false sharing in cmpxchg.
>>
>> Do you have numbers demonstrating claimed performance improvement
>> after this change ?
>>
>
> I asked for that in v1 and there are no performance numbers to justify
> the change. IMHO, that will be required to consider this for inclusion,
> otherwise it's just code churn resulting in an (although minimal)
> additional memory consumption.
>

Just got some time to run thpscale on my 4*HT cores machine, here is the data:
I run each of kernel for 3 times, pageblock kernel is the 5.9-rc2 with this 2
patches, the plp1 is the first patch on 5.9-rc2, and rc2 is 5.9-rc2 kernel.
We could found the system and total time is slight less than original kernel.

pageblock pageblock pageblock plp1 plp1 plp1 rc2 rc2 rc2 pageblock
16 16-2 16-3 1 2 3 a b c a
Duration User 14.81 15.24 14.55 15.28 14.66 14.63 14.76 14.97 14.38 15.07
Duration System 84.44 88.38 90.64 92.65 94.01 90.58 100.43 89.15 88.89 84.04
Duration Elapsed 98.83 99.06 99.81 99.65 100.26 99.90 100.30 99.24 99.14 98.87

And I also add tracing for patchset effect, which show the cmpxchg failure times
get clearly less.



Performance counter stats for './run-mmtests.sh -c configs/config-workload-thpscale rc2-b':

6,720 compaction:mm_compaction_isolate_migratepages
13,526 compaction:mm_compaction_isolate_freepages
4,052 compaction:mm_compaction_migratepages
34,199 compaction:mm_compaction_begin
34,199 compaction:mm_compaction_end
21,784 compaction:mm_compaction_try_to_compact_pages
71,606 compaction:mm_compaction_finished
106,545 compaction:mm_compaction_suitable
0 compaction:mm_compaction_deferred
0 compaction:mm_compaction_defer_compaction
2,977 compaction:mm_compaction_defer_reset
0 compaction:mm_compaction_kcompactd_sleep
0 compaction:mm_compaction_wakeup_kcompactd
0 compaction:mm_compaction_kcompactd_wake
1,046 pageblock:hit_cmpxchg

114.914303988 seconds time elapsed

15.754797000 seconds user
89.712251000 seconds sys





Performance counter stats for './run-mmtests.sh -c configs/config-workload-thpscale pageblock-a':

602 compaction:mm_compaction_isolate_migratepages
3,710 compaction:mm_compaction_isolate_freepages
402 compaction:mm_compaction_migratepages
43,116 compaction:mm_compaction_begin
43,116 compaction:mm_compaction_end
24,810 compaction:mm_compaction_try_to_compact_pages
86,527 compaction:mm_compaction_finished
125,819 compaction:mm_compaction_suitable
2 compaction:mm_compaction_deferred
0 compaction:mm_compaction_defer_compaction
271 compaction:mm_compaction_defer_reset
0 compaction:mm_compaction_kcompactd_sleep
0 compaction:mm_compaction_wakeup_kcompactd
0 compaction:mm_compaction_kcompactd_wake
369 pageblock:hit_cmpxchg

107.405499745 seconds time elapsed

15.830967000 seconds user
84.559767000 seconds sys


commit 36cea76895637c0c18ce8590c0f43a3e453fbf8f
Author: Alex Shi <alex.shi@linux.alibaba.com>
Date: Wed Aug 19 17:26:26 2020 +0800

add cmpxchg tracing

Signed-off-by: Alex Shi <alex.shi@linux.alibaba.com>

diff --git a/include/trace/events/pageblock.h b/include/trace/events/pageblock.h
new file mode 100644
index 000000000000..003c2d716f82
--- /dev/null
+++ b/include/trace/events/pageblock.h
@@ -0,0 +1,30 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM pageblock
+
+#if !defined(_TRACE_PAGEBLOCK_H) || defined(TRACE_HEADER_MULTI_READ)
+#define _TRACE_PAGEBLOCK_H
+
+#include <linux/tracepoint.h>
+
+TRACE_EVENT(hit_cmpxchg,
+
+ TP_PROTO(char byte),
+
+ TP_ARGS(byte),
+
+ TP_STRUCT__entry(
+ __field(char, byte)
+ ),
+
+ TP_fast_assign(
+ __entry->byte = byte;
+ ),
+
+ TP_printk("%d", __entry->byte)
+);
+
+#endif /* _TRACE_PAGE_ISOLATION_H */
+
+/* This part must be outside protection */
+#include <trace/define_trace.h>
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 60342e764090..2422dec00484 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -509,6 +509,9 @@ static __always_inline int get_pfnblock_migratetype(struct page *page, unsigned
* @pfn: The target page frame number
* @mask: mask of bits that the caller is interested in
*/
+#define CREATE_TRACE_POINTS
+#include <trace/events/pageblock.h>
+
void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
unsigned long pfn,
unsigned long mask)
@@ -536,6 +539,7 @@ void set_pfnblock_flags_mask(struct page *page, unsigned long flags,
if (byte == old_byte)
break;
byte = old_byte;
+ trace_hit_cmpxchg(byte);
}
}
\
 
 \ /
  Last update: 2020-08-30 12:10    [W:0.052 / U:6.240 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site