Message-ID: <c02e856e-6ec5-49d0-8527-2647695a0174@huawei.com>
Date: Wed, 15 Jan 2025 19:33:51 +0800
From: Yunsheng Lin <linyunsheng@...wei.com>
To: Jesper Dangaard Brouer <hawk@...nel.org>, <davem@...emloft.net>,
	<kuba@...nel.org>, <pabeni@...hat.com>
CC: <zhangkun09@...wei.com>, <liuyonglong@...wei.com>,
	<fanghaiqing@...wei.com>, Alexander Lobakin <aleksander.lobakin@...el.com>,
	Robin Murphy <robin.murphy@....com>, Alexander Duyck
	<alexander.duyck@...il.com>, Andrew Morton <akpm@...ux-foundation.org>, IOMMU
	<iommu@...ts.linux.dev>, MM <linux-mm@...ck.org>, Alexei Starovoitov
	<ast@...nel.org>, Daniel Borkmann <daniel@...earbox.net>, John Fastabend
	<john.fastabend@...il.com>, Matthias Brugger <matthias.bgg@...il.com>,
	AngeloGioacchino Del Regno <angelogioacchino.delregno@...labora.com>,
	<netdev@...r.kernel.org>, <intel-wired-lan@...ts.osuosl.org>,
	<bpf@...r.kernel.org>, <linux-kernel@...r.kernel.org>,
	<linux-arm-kernel@...ts.infradead.org>, <linux-mediatek@...ts.infradead.org>
Subject: Re: [PATCH net-next v7 0/8] fix two bugs related to page_pool

On 2025/1/14 22:31, Jesper Dangaard Brouer wrote:
> 
> 
> On 10/01/2025 14.06, Yunsheng Lin wrote:
>> This patchset fix a possible time window problem for page_pool and
>> the dma API misuse problem as mentioned in [1], and try to avoid the
>> overhead of the fixing using some optimization.
>>
>>  From the below performance data, the overhead is not so obvious
>> due to performance variations for time_bench_page_pool01_fast_path()
>> and time_bench_page_pool02_ptr_ring, and there is about 20ns overhead
>> for time_bench_page_pool03_slow() for fixing the bug.
>>
> 
> My benchmarking on x86_64 CPUs looks significantly different.
>  - CPU: Intel(R) Xeon(R) CPU E5-1650 v4 @ 3.60GHz
> 
> Benchmark (bench_page_pool_simple) results from before and after patchset:
> 
> | Test name  | Cycles |       |    |Nanosec |        |       |      % |
> | (tasklet_*)| Before | After |diff| Before |  After |  diff | change |
> |------------+--------+-------+----+--------+--------+-------+--------|
> | fast_path  |     19 |    24 |   5|  5.399 |  6.928 | 1.529 |   28.3 |
> | ptr_ring   |     54 |    79 |  25| 15.090 | 21.976 | 6.886 |   45.6 |
> | slow       |    238 |   299 |  61| 66.134 | 83.298 |17.164 |   26.0 |
> #+TBLFM: $4=$3-$2::$7=$6-$5::$8=(($7/$5)*100);%.1f
> 
> My above testing show a clear performance regressions across three
> different page_pool operating modes.

I retested it on arm64 server patch by patch as the raw performance
data in the attachment, it seems the result seemed similar as before.

Before this patchset:
            fast_path              ptr_ring            slow
1.         31.171 ns               60.980 ns          164.917 ns
2.         28.824 ns               60.891 ns          170.241 ns
3.         14.236 ns               60.583 ns          164.355 ns

With patch 1-4:
4.         31.443 ns               53.242 ns          210.148 ns
5.         31.406 ns               53.270 ns          210.189 ns

With patch 1-5:
6.         26.163 ns               53.781 ns          189.450 ns
7.         26.189 ns               53.798 ns          189.466 ns

With patch 1-8:
8.         28.108 ns               68.199 ns          202.516 ns
9.         16.128 ns               55.904 ns          202.711 ns

I am not able to get hold of a x86 server yet, I might be able
to get one during weekend.

Theoretically, patch 1-4 or 1-5 should not have much performance
impact for fast_path and ptr_ring except for the rcu_lock mentioned
in page_pool_napi_local(), so it would be good if patch 1-5 is also
tested in your testlab with the rcu_lock removing in
page_pool_napi_local().

> 
> 
> Data also available in:
>  - https://github.com/xdp-project/xdp-project/blob/main/areas/mem/page_pool07_bench_DMA_fix.org
> 
> Raw data below
> 
> Before this patchset:
> 
> [  157.186644] bench_page_pool_simple: Loaded
> [  157.475084] time_bench: Type:for_loop Per elem: 1 cycles(tsc) 0.284 ns (step:0) - (measurement period time:0.284327440 sec time_interval:284327440) - (invoke count:1000000000 tsc_interval:1023590451)
> [  162.262752] time_bench: Type:atomic_inc Per elem: 17 cycles(tsc) 4.769 ns (step:0) - (measurement period time:4.769757001 sec time_interval:4769757001) - (invoke count:1000000000 tsc_interval:17171776113)
> [  163.324091] time_bench: Type:lock Per elem: 37 cycles(tsc) 10.431 ns (step:0) - (measurement period time:1.043182161 sec time_interval:1043182161) - (invoke count:100000000 tsc_interval:3755514465)
> [  163.341702] bench_page_pool_simple: time_bench_page_pool01_fast_path(): Cannot use page_pool fast-path
> [  163.922466] time_bench: Type:no-softirq-page_pool01 Per elem: 20 cycles(tsc) 5.713 ns (step:0) - (measurement period time:0.571357387 sec time_interval:571357387) - (invoke count:100000000 tsc_interval:2056911063)
> [  163.941429] bench_page_pool_simple: time_bench_page_pool02_ptr_ring(): Cannot use page_pool fast-path
> [  165.506796] time_bench: Type:no-softirq-page_pool02 Per elem: 56 cycles(tsc) 15.560 ns (step:0) - (measurement period time:1.556080558 sec time_interval:1556080558) - (invoke count:100000000 tsc_interval:5601960921)
> [  165.525978] bench_page_pool_simple: time_bench_page_pool03_slow(): Cannot use page_pool fast-path
> [  171.811289] time_bench: Type:no-softirq-page_pool03 Per elem: 225 cycles(tsc) 62.763 ns (step:0) - (measurement period time:6.276301531 sec time_interval:6276301531) - (invoke count:100000000 tsc_interval:22594974468)
> [  171.830646] bench_page_pool_simple: pp_tasklet_handler(): in_serving_softirq fast-path
> [  171.838561] bench_page_pool_simple: time_bench_page_pool01_fast_path(): in_serving_softirq fast-path
> [  172.387597] time_bench: Type:tasklet_page_pool01_fast_path Per elem: 19 cycles(tsc) 5.399 ns (step:0) - (measurement period time:0.539904228 sec time_interval:539904228) - (invoke count:100000000 tsc_interval:1943679246)
> [  172.407130] bench_page_pool_simple: time_bench_page_pool02_ptr_ring(): in_serving_softirq fast-path
> [  173.925266] time_bench: Type:tasklet_page_pool02_ptr_ring Per elem: 54 cycles(tsc) 15.090 ns (step:0) - (measurement period time:1.509075496 sec time_interval:1509075496) - (invoke count:100000000 tsc_interval:5432740575)
> [  173.944878] bench_page_pool_simple: time_bench_page_pool03_slow(): in_serving_softirq fast-path
> [  180.567094] time_bench: Type:tasklet_page_pool03_slow Per elem: 238 cycles(tsc) 66.134 ns (step:0) - (measurement period time:6.613430605 sec time_interval:6613430605) - (invoke count:100000000 tsc_interval:23808654870)
> 
> 
> 
> After this patchset:
> [  860.519918] bench_page_pool_simple: Loaded
> [  860.781605] time_bench: Type:for_loop Per elem: 0 cycles(tsc) 0.257 ns (step:0) - (measurement period time:0.257573336 sec time_interval:257573336) - (invoke count:1000000000 tsc_interval:927275355)
> [  865.613893] time_bench: Type:atomic_inc Per elem: 17 cycles(tsc) 4.814 ns (step:0) - (measurement period time:4.814593429 sec time_interval:4814593429) - (invoke count:1000000000 tsc_interval:17332768494)
> [  866.708420] time_bench: Type:lock Per elem: 38 cycles(tsc) 10.763 ns (step:0) - (measurement period time:1.076362960 sec time_interval:1076362960) - (invoke count:100000000 tsc_interval:3874955595)
> [  866.726118] bench_page_pool_simple: time_bench_page_pool01_fast_path(): Cannot use page_pool fast-path
> [  867.423572] time_bench: Type:no-softirq-page_pool01 Per elem: 24 cycles(tsc) 6.880 ns (step:0) - (measurement period time:0.688069107 sec time_interval:688069107) - (invoke count:100000000 tsc_interval:2477080260)
> [  867.442517] bench_page_pool_simple: time_bench_page_pool02_ptr_ring(): Cannot use page_pool fast-path
> [  869.436286] time_bench: Type:no-softirq-page_pool02 Per elem: 71 cycles(tsc) 19.844 ns (step:0) - (measurement period time:1.984451929 sec time_interval:1984451929) - (invoke count:100000000 tsc_interval:7144120329)
> [  869.455492] bench_page_pool_simple: time_bench_page_pool03_slow(): Cannot use page_pool fast-path
> [  877.071437] time_bench: Type:no-softirq-page_pool03 Per elem: 273 cycles(tsc) 76.069 ns (step:0) - (measurement period time:7.606911291 sec time_interval:7606911291) - (invoke count:100000000 tsc_interval:27385252251)
> [  877.090762] bench_page_pool_simple: pp_tasklet_handler(): in_serving_softirq fast-path
> [  877.098683] bench_page_pool_simple: time_bench_page_pool01_fast_path(): in_serving_softirq fast-path
> [  877.800696] time_bench: Type:tasklet_page_pool01_fast_path Per elem: 24 cycles(tsc) 6.928 ns (step:0) - (measurement period time:0.692852876 sec time_interval:692852876) - (invoke count:100000000 tsc_interval:2494303293)
> [  877.820224] bench_page_pool_simple: time_bench_page_pool02_ptr_ring(): in_serving_softirq fast-path
> [  880.026911] time_bench: Type:tasklet_page_pool02_ptr_ring Per elem: 79 cycles(tsc) 21.976 ns (step:0) - (measurement period time:2.197615122 sec time_interval:2197615122) - (invoke count:100000000 tsc_interval:7911521190)
> [  880.046528] bench_page_pool_simple: time_bench_page_pool03_slow(): in_serving_softirq fast-path
> [  888.385235] time_bench: Type:tasklet_page_pool03_slow Per elem: 299 cycles(tsc) 83.298 ns (step:0) - (measurement period time:8.329893717 sec time_interval:8329893717) - (invoke count:100000000 tsc_interval:29988024696)

As mentioned by Toke, we may be able to reduce the performance difference
between tasklet and non-tasklet testcases by removing the rcu_lock in
page_pool_napi_local() for patch 1 as in_softirq() checking in
page_pool_napi_local() should ensure RCU-bh read-side critical section.
View attachment "pp_inflight_fix_v7_perf_data.txt" of type "text/plain" (69039 bytes)

Powered by blists - more mailing lists