From mboxrd@z Thu Jan 1 00:00:00 1970 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 790F5D29FB8 for ; Wed, 6 Nov 2024 08:43:15 +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-Transfer-Encoding: Content-Type:In-Reply-To:MIME-Version:Date:Message-ID:From:References:To: Subject:CC:Reply-To:Content-ID:Content-Description:Resent-Date:Resent-From: Resent-Sender:Resent-To:Resent-Cc:Resent-Message-ID:List-Owner; bh=qcjnpJDcS9pr/1jLW7TlSymlCjKqx1U7IbD8OpRJMN4=; b=vMvBLYFFh5ePMkau++LmelWZ9y MWJGhfE+xBBBvrS1OpPFEXNz9IIUFqYcmetWQ/KAPSGQQxgphp2oZJkJSdD3FR+PBZnRIHJ2iXzrE 2VIqfzGZRr24IqzeGZeTFX+A1L86awAz2dpX357fYuUyOxBlXj5AP/4BEhX7uguuhwniCs87Bio4L tDYonba2nIDHayQNU5vOy8Tvb0ZkmCINQ429hf2cKhSXg1a37vG3K6MXg2ekQM9+O3uc/OE4VQ4JC m9PNzBPnvIb1riQesdUUa2WJR89IqLJdQt6415P8qnRRsR29PFSllAWy7Bw5QS3jqdUYpyrchYkOU MODgaLfg==; Received: from localhost ([::1] helo=bombadil.infradead.org) by bombadil.infradead.org with esmtp (Exim 4.98 #2 (Red Hat Linux)) id 1t8bd0-00000002Kp3-2dtG; Wed, 06 Nov 2024 08:43:02 +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 1t8bTq-00000002ItO-0B0A for linux-arm-kernel@lists.infradead.org; Wed, 06 Nov 2024 08:33:36 +0000 Received: from mail.maildlp.com (unknown [172.19.163.252]) by szxga01-in.huawei.com (SkyGuard) with ESMTP id 4Xjz2D3WZXz10V72; Wed, 6 Nov 2024 16:31:32 +0800 (CST) Received: from kwepemd200014.china.huawei.com (unknown [7.221.188.8]) by mail.maildlp.com (Postfix) with ESMTPS id 172571800A5; Wed, 6 Nov 2024 16:33:19 +0800 (CST) Received: from [10.67.121.177] (10.67.121.177) by kwepemd200014.china.huawei.com (7.221.188.8) with Microsoft SMTP Server (version=TLS1_2, cipher=TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384) id 15.2.1258.34; Wed, 6 Nov 2024 16:33:18 +0800 CC: , , , , , , , Subject: Re: [PATCH v3 2/8] drivers/perf: hisi: Improve the detection of associated CPUs To: Will Deacon References: <20241026072424.29887-1-yangyicong@huawei.com> <20241026072424.29887-3-yangyicong@huawei.com> <20241029132826.GB4416@willie-the-truck> <5895fd31-04f8-f1ea-0639-12dd1f19069c@huawei.com> From: Yicong Yang Message-ID: <9a64ed90-6c9f-a32a-85cf-9d0c41e2fd81@huawei.com> Date: Wed, 6 Nov 2024 16:33:18 +0800 User-Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:78.0) Gecko/20100101 Thunderbird/78.5.1 MIME-Version: 1.0 In-Reply-To: <5895fd31-04f8-f1ea-0639-12dd1f19069c@huawei.com> Content-Type: text/plain; charset="utf-8" Content-Transfer-Encoding: 7bit X-Originating-IP: [10.67.121.177] X-ClientProxiedBy: dggems705-chm.china.huawei.com (10.3.19.182) To kwepemd200014.china.huawei.com (7.221.188.8) X-CRM114-Version: 20100106-BlameMichelson ( TRE 0.8.0 (BSD) ) MR-646709E3 X-CRM114-CacheID: sfid-20241106_003334_491185_32F9F455 X-CRM114-Status: GOOD ( 36.11 ) 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 Hi Will, Further comment on this patch? On 2024/10/30 22:04, Yicong Yang wrote: > On 2024/10/29 21:28, Will Deacon wrote: >> On Sat, Oct 26, 2024 at 03:24:18PM +0800, Yicong Yang wrote: >>> From: Yicong Yang >>> >>> Currently the associated CPUs are detected in the cpuhp online >>> callback. If the CPU's sccl_id or the ccl_id matches the PMU's, >>> they're associated. There's an exception that some PMUs locate >>> on the SICL and will match no CPUs. The events of these PMUs >>> can be opened on any online CPUs. To handle this we just check >>> whether the PMU's sccl_id is -1, if so we know it locates on >>> SICL and make any CPU associated to it. >>> >>> This can be tweaked so in this patch just do the below changes: >>> - If the PMU doesn't match any CPU then associated it to online CPUs >>> - Choose the target CPU according to the NUMA affinity for opening >>> events >>> >>> The function is implemented by hisi_pmu_init_associated_cpus() and >>> invoked in hisi_pmu_init(). >>> >>> Also the associated_cpus are maintained with all the online CPUs. This >>> is redundant since we'll always schedule the events on the online CPUs. >>> Get rid of this and make associated_cpus contain offline CPUs as well. >> >> I don't really understand the rationale for this change. Why is the new >> behaviour better than the old one? >> > > Thanks for taking a look. > > As mentioned in the commit, we have 2 types of PMU here: > 1) PMUs locate on SCCL (Super CPU Cluster *), associated with certain CCL(CPU cluster *)(e.g. L3C PMU) > or not (e.g. DDRC PMU) > 2) PMUs locate on the SICL (Super IO Cluster *), which has no association with certain CPU > topology (e.g. CPA PMU) > > Currently we find associated CPUs in the cpuhp callbacks by comparing the CPU's > MPIDR with the PMU's SCCL ID and CCL ID. This will be fine for type 1) PMUs but > not for type 2) PMUs since no CPU will match a SICL ID. We do a trick here, > make type 2) PMUs's SCCL to -1 and match all the CPUs. So in fact type 2) PMUs > are already associated with online CPUs but in an implicit way. With this patch > the behaviour will be more clear. > > Another thing is about NUMA locality. Usually for type 2) PMUs they maybe on > different NUMA node which is not considered in current approach, all SICL PMUs > are stacked on CPU0. With the patch the NUMA node information will be considered > if provided by the firmware. > > [*] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/Documentation/admin-guide/perf/hisi-pmu.rst?h=v6.12-rc1 > Another thing is that if the associated CPUs are all offline, previous approach will show -1 by cpumask while with this patch user will always see a usable CPU (this is ok since the uncore events are is not associated with certain CPU context and can be open on any online CPU): // In the previous approach [root@localhost devices]# cat hisi_sccl3_l3c1/cpumask 8 // offline all the CPUs sharing sccl3_l3c1 [root@localhost devices]# cat hisi_sccl3_l3c1/cpumask -1 This should make no difference for the function since perf tool will find an online CPU for opening the event even if it gets -1 from the cpumask. But -1 maybe a bit confusing. Thanks. >>> >>> Reviewed-by: Jonathan Cameron >>> Signed-off-by: Yicong Yang >>> --- >>> drivers/perf/hisilicon/hisi_uncore_pmu.c | 56 +++++++++++++++++++----- >>> drivers/perf/hisilicon/hisi_uncore_pmu.h | 5 +++ >>> 2 files changed, 49 insertions(+), 12 deletions(-) >>> >>> diff --git a/drivers/perf/hisilicon/hisi_uncore_pmu.c b/drivers/perf/hisilicon/hisi_uncore_pmu.c >>> index 416f72a813fc..c3549e16e0c3 100644 >>> --- a/drivers/perf/hisilicon/hisi_uncore_pmu.c >>> +++ b/drivers/perf/hisilicon/hisi_uncore_pmu.c >>> @@ -399,6 +399,27 @@ void hisi_uncore_pmu_disable(struct pmu *pmu) >>> } >>> EXPORT_SYMBOL_NS_GPL(hisi_uncore_pmu_disable, HISI_PMU); >>> >>> +static void hisi_pmu_init_associated_cpus(struct hisi_pmu *hisi_pmu) >>> +{ >>> + /* >>> + * If the associated_cpus has already been initialized, for example >>> + * determined by comparing the sccl_id and ccl_id with the CPU's >>> + * mpidr_el1, then do nothing here. Otherwise the PMU has no affinity >>> + * and could be opened on any online CPU. >>> + */ >>> + if (!cpumask_empty(&hisi_pmu->associated_cpus)) >>> + return; >>> + >>> + cpumask_copy(&hisi_pmu->associated_cpus, cpu_online_mask); >> >> Is it always safe to access 'cpu_online_mask' here? >> > > It should be safe. This function will be called in hisi_pmu_init() in the module > init stage so the cpu_online_mask() should be populated, at least the current > running CPU is contained. The hisi_pmu->associated_cpus will be used along > with the online CPU checking in cpuhp callbacks so we're unlikely to use an > offline CPU mistakenly. Checked cpumask_local_spread() and it also use the > cpu_online_mask directly without synchronization against cpuhp process, so > it also should be ok here. Please correct my if any cases I missed. > > Thanks, > Yicong > > > > . >