From patchwork Wed Sep 18 11:18:23 2024 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Yunsheng Lin X-Patchwork-Id: 13806836 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from bombadil.infradead.org (bombadil.infradead.org [198.137.202.133]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.lore.kernel.org (Postfix) with ESMTPS id 4948FCCD1A0 for ; Wed, 18 Sep 2024 11:26:07 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; q=dns/txt; c=relaxed/relaxed; d=lists.infradead.org; s=bombadil.20210309; h=Sender:List-Subscribe:List-Help :List-Post:List-Archive:List-Unsubscribe:List-Id:Content-Type: Content-Transfer-Encoding:MIME-Version:Message-ID:Date:Subject:CC:To:From: Reply-To:Content-ID:Content-Description:Resent-Date:Resent-From:Resent-Sender :Resent-To:Resent-Cc:Resent-Message-ID:In-Reply-To:References:List-Owner; bh=kI9Y4R6SliCUq0ZrjxXhYRFYLywr93KKFQ+qad/cw2g=; b=auKQY+wWr6KofTZVEQ8JWCCaWC vu3vv5vHH3hXGa84F0e0mkrRVgOGEwS+s2AAttijgWm/5d7bpjTU77S6K33LQW/4De8zG2Okqt1Aj 3uv53Th5MjldCdxZ32Pbfw0K45GArlMeizLv6OSCV1faYutOKGC2WPU7jQBd/DjYgEK4MoAnKfnsT XNQfj1jcDaZqFsoebB+7q88Hi8VUBTX7ALkUjp9XnMK3BH/n1dMbEmPCNALV/4QVPPZ03I4l/BGjk A0QGpWSjPypvwkL8scUyOhiwA/lFrFuzenuDl9J1xSxPrrmQfwfmWrph018izUmCdgQAkkw2F5mCR RzHILl/Q==; Received: from localhost ([::1] helo=bombadil.infradead.org) by bombadil.infradead.org with esmtp (Exim 4.98 #2 (Red Hat Linux)) id 1sqsom-000000088Kl-204F; Wed, 18 Sep 2024 11:25:56 +0000 Received: from szxga01-in.huawei.com ([45.249.212.187]) by bombadil.infradead.org with esmtps (Exim 4.98 #2 (Red Hat Linux)) id 1sqsnV-000000087vd-0Npo; Wed, 18 Sep 2024 11:24:40 +0000 Received: from mail.maildlp.com (unknown [172.19.163.174]) by szxga01-in.huawei.com (SkyGuard) with ESMTP id 4X7x8r3ylrzyRhL; Wed, 18 Sep 2024 19:23:08 +0800 (CST) Received: from dggpemf200006.china.huawei.com (unknown [7.185.36.61]) by mail.maildlp.com (Postfix) with ESMTPS id A7D5B1400DD; Wed, 18 Sep 2024 19:24:24 +0800 (CST) Received: from localhost.localdomain (10.90.30.45) by dggpemf200006.china.huawei.com (7.185.36.61) with Microsoft SMTP Server (version=TLS1_2, cipher=TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384) id 15.2.1544.11; Wed, 18 Sep 2024 19:24:24 +0800 From: Yunsheng Lin To: , , CC: , , , Yunsheng Lin , Alexander Lobakin , Robin Murphy , Alexander Duyck , IOMMU , Alexei Starovoitov , Daniel Borkmann , Jesper Dangaard Brouer , John Fastabend , Matthias Brugger , AngeloGioacchino Del Regno , , , , , , Subject: [PATCH net 0/2] fix two bugs related to page_pool Date: Wed, 18 Sep 2024 19:18:23 +0800 Message-ID: <20240918111826.863596-1-linyunsheng@huawei.com> X-Mailer: git-send-email 2.30.0 MIME-Version: 1.0 X-Originating-IP: [10.90.30.45] X-ClientProxiedBy: dggems703-chm.china.huawei.com (10.3.19.180) To dggpemf200006.china.huawei.com (7.185.36.61) X-CRM114-Version: 20100106-BlameMichelson ( TRE 0.8.0 (BSD) ) MR-646709E3 X-CRM114-CacheID: sfid-20240918_042437_665373_549D0315 X-CRM114-Status: GOOD ( 19.61 ) X-BeenThere: linux-arm-kernel@lists.infradead.org X-Mailman-Version: 2.1.34 Precedence: list List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Sender: "linux-arm-kernel" Errors-To: linux-arm-kernel-bounces+linux-arm-kernel=archiver.kernel.org@lists.infradead.org Patch 1 fix a possible time window problem for page_pool. Patch 2 fix the kernel crash problem at iommu_get_dma_domain reported in [1]. When page_pool_put_unrefed_netmem() is called with allow_direct being true, there is added checking overhead introduced in patch 1, which seem to be no noticeable performance impact. When page_pool_put_unrefed_netmem() is called with allow_direct being false, there is a rcu read lock overhead introduced in patch 1, and the overhead is about 13ns using the below test code, but 'time_bench_page_pool02_ptr_ring' only show about 2ns overhead, which is about 2% degradation. +static int time_bench_rcu( + struct time_bench_record *rec, void *data) +{ + uint64_t loops_cnt = 0; + int i; + + time_bench_start(rec); + /** Loop to measure **/ + for (i = 0; i < rec->loops; i++) { + rcu_read_lock(); + loops_cnt++; + barrier(); /* avoid compiler to optimize this loop */ + rcu_read_unlock(); + } + time_bench_stop(rec, loops_cnt); + return loops_cnt; +} When page_pool need to be refilled from or flushed to the page allocator, the added overhead is the page_pool_item_add() and page_pool_item_del() calling overhead, using below patch to enable Jesper's testing running in arm64, the overhead is 0~20ns, which is quite variable Before this patchset: root@(none)$ taskset -c 1 insmod bench_page_pool_simple.ko [ 136.641453] bench_page_pool_simple: Loaded [ 136.722560] time_bench: Type:for_loop Per elem: 0 cycles(tsc) 0.769 ns (step:0) - (measurement period time:0.076968720 sec time_interval:76968720) - (invoke count:100000000 tsc_interval:7696855) [ 137.317006] time_bench: Type:atomic_inc Per elem: 0 cycles(tsc) 5.771 ns (step:0) - (measurement period time:0.577164350 sec time_interval:577164350) - (invoke count:100000000 tsc_interval:57716429) [ 137.480852] time_bench: Type:lock Per elem: 1 cycles(tsc) 14.621 ns (step:0) - (measurement period time:0.146218730 sec time_interval:146218730) - (invoke count:10000000 tsc_interval:14621868) [ 138.842377] time_bench: Type:rcu Per elem: 1 cycles(tsc) 13.444 ns (step:0) - (measurement period time:1.344419820 sec time_interval:1344419820) - (invoke count:100000000 tsc_interval:134441975) [ 138.859656] bench_page_pool_simple: time_bench_page_pool01_fast_path(): Cannot use page_pool fast-path [ 139.132102] time_bench: Type:no-softirq-page_pool01 Per elem: 2 cycles(tsc) 26.315 ns (step:0) - (measurement period time:0.263151430 sec time_interval:263151430) - (invoke count:10000000 tsc_interval:26315135) [ 139.150769] bench_page_pool_simple: time_bench_page_pool02_ptr_ring(): Cannot use page_pool fast-path [ 139.910642] time_bench: Type:no-softirq-page_pool02 Per elem: 7 cycles(tsc) 75.066 ns (step:0) - (measurement period time:0.750663200 sec time_interval:750663200) - (invoke count:10000000 tsc_interval:75066312) [ 139.929312] bench_page_pool_simple: time_bench_page_pool03_slow(): Cannot use page_pool fast-path [ 141.673951] time_bench: Type:no-softirq-page_pool03 Per elem: 17 cycles(tsc) 173.578 ns (step:0) - (measurement period time:1.735781610 sec time_interval:1735781610) - (invoke count:10000000 tsc_interval:173578155) [ 141.692970] bench_page_pool_simple: pp_tasklet_handler(): in_serving_softirq fast-path [ 141.700874] bench_page_pool_simple: time_bench_page_pool01_fast_path(): in_serving_softirq fast-path [ 141.973638] time_bench: Type:tasklet_page_pool01_fast_path Per elem: 2 cycles(tsc) 26.364 ns (step:0) - (measurement period time:0.263645150 sec time_interval:263645150) - (invoke count:10000000 tsc_interval:26364508) [ 141.992912] bench_page_pool_simple: time_bench_page_pool02_ptr_ring(): in_serving_softirq fast-path [ 142.531745] time_bench: Type:tasklet_page_pool02_ptr_ring Per elem: 5 cycles(tsc) 52.980 ns (step:0) - (measurement period time:0.529801250 sec time_interval:529801250) - (invoke count:10000000 tsc_interval:52980119) [ 142.550933] bench_page_pool_simple: time_bench_page_pool03_slow(): in_serving_softirq fast-path [ 144.297646] time_bench: Type:tasklet_page_pool03_slow Per elem: 17 cycles(tsc) 173.802 ns (step:0) - (measurement period time:1.738029000 sec time_interval:1738029000) - (invoke count:10000000 tsc_interval:173802894) After this patchset: root@(none)$ taskset -c 1 insmod bench_page_pool_simple.ko [ 149.865799] bench_page_pool_simple: Loaded [ 149.946907] time_bench: Type:for_loop Per elem: 0 cycles(tsc) 0.769 ns (step:0) - (measurement period time:0.076965620 sec time_interval:76965620) - (invoke count:100000000 tsc_interval:7696556) [ 150.722282] time_bench: Type:atomic_inc Per elem: 0 cycles(tsc) 7.580 ns (step:0) - (measurement period time:0.758094660 sec time_interval:758094660) - (invoke count:100000000 tsc_interval:75809459) [ 150.886335] time_bench: Type:lock Per elem: 1 cycles(tsc) 14.640 ns (step:0) - (measurement period time:0.146405830 sec time_interval:146405830) - (invoke count:10000000 tsc_interval:14640578) [ 152.249454] time_bench: Type:rcu Per elem: 1 cycles(tsc) 13.460 ns (step:0) - (measurement period time:1.346009570 sec time_interval:1346009570) - (invoke count:100000000 tsc_interval:134600951) [ 152.266734] bench_page_pool_simple: time_bench_page_pool01_fast_path(): Cannot use page_pool fast-path [ 152.537046] time_bench: Type:no-softirq-page_pool01 Per elem: 2 cycles(tsc) 26.100 ns (step:0) - (measurement period time:0.261007670 sec time_interval:261007670) - (invoke count:10000000 tsc_interval:26100761) [ 152.555714] bench_page_pool_simple: time_bench_page_pool02_ptr_ring(): Cannot use page_pool fast-path [ 153.342212] time_bench: Type:no-softirq-page_pool02 Per elem: 7 cycles(tsc) 77.729 ns (step:0) - (measurement period time:0.777293380 sec time_interval:777293380) - (invoke count:10000000 tsc_interval:77729331) [ 153.360881] bench_page_pool_simple: time_bench_page_pool03_slow(): Cannot use page_pool fast-path [ 155.287747] time_bench: Type:no-softirq-page_pool03 Per elem: 19 cycles(tsc) 191.800 ns (step:0) - (measurement period time:1.918007990 sec time_interval:1918007990) - (invoke count:10000000 tsc_interval:191800791) [ 155.306766] bench_page_pool_simple: pp_tasklet_handler(): in_serving_softirq fast-path [ 155.314670] bench_page_pool_simple: time_bench_page_pool01_fast_path(): in_serving_softirq fast-path [ 155.584313] time_bench: Type:tasklet_page_pool01_fast_path Per elem: 2 cycles(tsc) 26.052 ns (step:0) - (measurement period time:0.260524810 sec time_interval:260524810) - (invoke count:10000000 tsc_interval:26052476) [ 155.603588] bench_page_pool_simple: time_bench_page_pool02_ptr_ring(): in_serving_softirq fast-path [ 156.183214] time_bench: Type:tasklet_page_pool02_ptr_ring Per elem: 5 cycles(tsc) 57.059 ns (step:0) - (measurement period time:0.570594850 sec time_interval:570594850) - (invoke count:10000000 tsc_interval:57059478) [ 156.202402] bench_page_pool_simple: time_bench_page_pool03_slow(): in_serving_softirq fast-path [ 158.045594] time_bench: Type:tasklet_page_pool03_slow Per elem: 18 cycles(tsc) 183.450 ns (step:0) - (measurement period time:1.834507700 sec time_interval:1834507700) - (invoke count:10000000 tsc_interval:183450764) Patch for time_bench.h enable the out of tree testing on arm64 system: @@ -101,6 +101,7 @@ struct time_bench_cpu { * CPUID clears the high 32-bits of all (rax/rbx/rcx/rdx) */ static __always_inline uint64_t tsc_start_clock(void) { +#if defined(__i386__) || defined(__x86_64__) /* See: Intel Doc #324264 */ unsigned hi, lo; asm volatile ( @@ -111,9 +112,13 @@ static __always_inline uint64_t tsc_start_clock(void) { "%rax", "%rbx", "%rcx", "%rdx"); //FIXME: on 32bit use clobbered %eax + %edx return ((uint64_t)lo) | (((uint64_t)hi) << 32); +#else + return get_cycles(); +#endif } static __always_inline uint64_t tsc_stop_clock(void) { +#if defined(__i386__) || defined(__x86_64__) /* See: Intel Doc #324264 */ unsigned hi, lo; asm volatile( @@ -123,6 +128,9 @@ static __always_inline uint64_t tsc_stop_clock(void) { "CPUID\n\t": "=r" (hi), "=r" (lo):: "%rax", "%rbx", "%rcx", "%rdx"); return ((uint64_t)lo) | (((uint64_t)hi) << 32); +#else + return get_cycles(); +#endif } /* Notes for RDTSC and RDTSCP @@ -186,10 +194,14 @@ enum { static __always_inline unsigned long long p_rdpmc(unsigned in) { +#if defined(__i386__) || defined(__x86_64__) unsigned d, a; asm volatile("rdpmc" : "=d" (d), "=a" (a) : "c" (in) : "memory"); return ((unsigned long long)d << 32) | a; +#else + return 0; +#endif } /* These PMU counter needs to be enabled, but I don't have the @@ -216,7 +228,11 @@ static __always_inline unsigned long long pmc_clk(void) #define MSR_IA32_PCM2 0x400000C3 inline uint64_t msr_inst(unsigned long long *msr_result) { +#if defined(__i386__) || defined(__x86_64__) return rdmsrl_safe(MSR_IA32_PCM0, msr_result); +#else + return 0; +#endif } 1. https://lore.kernel.org/lkml/8067f204-1380-4d37-8ffd-007fc6f26738@kernel.org/T/ CC: Alexander Lobakin CC: Robin Murphy CC: Alexander Duyck CC: IOMMU Yunsheng Lin (2): page_pool: fix timing for checking and disabling napi_local page_pool: fix IOMMU crash when driver has already unbound drivers/net/ethernet/freescale/fec_main.c | 8 +- drivers/net/ethernet/intel/iavf/iavf_txrx.c | 6 +- drivers/net/ethernet/intel/idpf/idpf_txrx.c | 14 +- drivers/net/ethernet/intel/libeth/rx.c | 2 +- .../net/ethernet/mellanox/mlx5/core/en/xdp.c | 3 +- drivers/net/netdevsim/netdev.c | 6 +- drivers/net/wireless/mediatek/mt76/mt76.h | 2 +- include/linux/mm_types.h | 2 +- include/linux/skbuff.h | 1 + include/net/libeth/rx.h | 3 +- include/net/netmem.h | 10 +- include/net/page_pool/helpers.h | 11 + include/net/page_pool/types.h | 15 +- net/core/devmem.c | 4 +- net/core/netmem_priv.h | 5 +- net/core/page_pool.c | 188 +++++++++++++++--- net/core/page_pool_priv.h | 10 +- net/core/skbuff.c | 3 +- net/core/xdp.c | 3 +- 19 files changed, 238 insertions(+), 58 deletions(-)