From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from mgamail.intel.com (mgamail.intel.com [198.175.65.10]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by smtp.subspace.kernel.org (Postfix) with ESMTPS id 57A17175A85; Thu, 15 Oct 2026 23:17:51 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; arc=none smtp.client-ip=198.175.65.10 ARC-Seal:i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1792106273; cv=none; b=AqeEhkjx5l9Gr3p4j9lznzLQHMHcw82LM9gS33hveSQauT8s/l1H94aT06Z5w16ylDRAjbqreNkrguLI7FwimxOZqFxVOOmLo6ssrYapKND89O88AnvOUpMmTuc6lBVGl8YJ44OEhalv0XWfU00NXIrQOxviiwjF9tMiE5pMd+Y= ARC-Message-Signature:i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1792106273; c=relaxed/simple; bh=Iich2VyA9Yl56RfAzD+pgFuU2K921AKcJLk7Up8k2OQ=; h=Message-ID:Subject:From:To:Date:In-Reply-To:References: Content-Type:MIME-Version; b=F6D+IYEbXvu4RYVjhFVfb7beYtlAjV7+q1fSfcAOc2TygknExQbG2hRPU2YgRQ4Hol0q9LVup8PQsYVzrzZuyVKXRn836LrMDHmMd0Mb8U3b/8Ma4HxMC/CPx83gfNG3VJZ9i6UhTrJqnIDDKPf7ThGnyjdH30UMGi/BR/1hNX0= ARC-Authentication-Results:i=1; smtp.subspace.kernel.org; dmarc=pass (p=none dis=none) header.from=linux.intel.com; spf=pass smtp.mailfrom=linux.intel.com; dkim=pass (2048-bit key) header.d=intel.com header.i=@intel.com header.b=JIhFkpTb; arc=none smtp.client-ip=198.175.65.10 Authentication-Results: smtp.subspace.kernel.org; dmarc=pass (p=none dis=none) header.from=linux.intel.com Authentication-Results: smtp.subspace.kernel.org; spf=pass smtp.mailfrom=linux.intel.com Authentication-Results: smtp.subspace.kernel.org; dkim=pass (2048-bit key) header.d=intel.com header.i=@intel.com header.b="JIhFkpTb" DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=intel.com; i=@intel.com; q=dns/txt; s=Intel; t=1792106272; x=1823642272; h=message-id:subject:from:to:date:in-reply-to:references: content-transfer-encoding:mime-version; bh=Iich2VyA9Yl56RfAzD+pgFuU2K921AKcJLk7Up8k2OQ=; b=JIhFkpTb0AdC8QP7jU5Fsu7GpYJ8UbvBX49SxVa/nlbpcPvXpG2DCCyK 9NM44/Y8ToReSFahy/dkhDfMrPaLDruOR4Umz4/LDhkrK/Te0aXZTbXbT z3/ITogwGlQHGeG9PYrRyPHvhBAeK+rOjkHSZidmBnEExSeD9YM1ojCRU 93BMaEiXi4+ZQvVwhrzHO16MfKlRo9JqcmyL88AMF11h1bR/kOQpQPLjh lLizr3Ryf8PWQjL+ZH/NR4VYHDE8X7PgqsPDVqeimpgwn4dljzBrcms3C rXRDrQ+NbRoAjn1D7GP4Z0OsN/k+YGDxb+eHOlI74rOeUpM+aXxwHnF7G g==; X-CSE-ConnectionGUID: +YjikdF0R4GbIcTZ2mDCrQ== X-CSE-MsgGUID: w7lGbnYgSmW1zcHh6CZheQ== X-IronPort-AV: E=McAfee;i="6800,10657,11936"; a="781041" X-IronPort-AV: E=Sophos;i="6.27,160,1787036400"; d="scan'208";a="781041" Received: from fmviesa002.fm.intel.com ([10.60.135.142]) by orvoesa102.jf.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 15 Oct 2026 16:17:51 -0700 X-CSE-ConnectionGUID: QPDXIi4JQEuXWbbkvfN5YQ== X-CSE-MsgGUID: KujZTx+hRAChXpirRYufLw== X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="6.27,160,1787036400"; d="scan'208";a="1904693" Received: from unknown (HELO [10.241.242.105]) ([10.241.242.105]) by smtpauth.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 15 Oct 2026 16:17:49 -0700 Message-ID: <1ec1aab8eabdad173ddef7483a940bfa3f6c3238.camel@linux.intel.com> Subject: Re: [PATCH v1 7/8] sched/topology: Avoid redundant percpu allocations and bitmap walks in build_sched_domains() From: Tim Chen To: Pasha Tatashin , kees@kernel.org, rppt@kernel.org, juri.lelli@redhat.com, nik.borisov@suse.com, tglx@kernel.org, vschneid@redhat.com, kprateek.nayak@amd.com, vincent.guittot@linaro.org, bp@alien8.de, thorsten.blum@linux.dev, cgroups@vger.kernel.org, tony.luck@intel.com, david@kernel.org, mingo@redhat.com, akpm@linux-foundation.org, linux-mm@kvack.org, seanjc@google.com, ljs@kernel.org, zhangguopeng@kylinos.cn, dietmar.eggemann@arm.com, mhocko@suse.com, vinicius.gomes@intel.com, tj@kernel.org, vbabka@kernel.org, linux-kernel@vger.kernel.org, peterz@infradead.org, surenb@google.com, hpa@zytor.com, x86@kernel.org, dmatlack@google.com, kexec@lists.infradead.org, pratyush@kernel.org, dave.hansen@linux.intel.com, rostedt@goodmis.org, gmonaco@redhat.com, jgross@suse.com, mgorman@suse.de, graf@amazon.com, linux-edac@vger.kernel.org, liam@infradead.org, ridong.chen@linux.dev, longman@redhat.com, bsegall@google.com Date: Thu, 15 Oct 2026 16:17:48 -0700 In-Reply-To: <20261015153331.1872641-8-pasha.tatashin@soleen.com> References: <20261015153331.1872641-1-pasha.tatashin@soleen.com> <20261015153331.1872641-8-pasha.tatashin@soleen.com> Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: quoted-printable User-Agent: Evolution 3.58.1 (3.58.1-1.fc43) Precedence: bulk X-Mailing-List: cgroups@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 On Thu, 2026-10-15 at 11:33 -0400, Pasha Tatashin wrote: > During CPU hotplug, build_sched_domains() is invoked on every CPU > activation and deactivation. Profiling build_sched_domains() on a > 480-CPU Intel Xeon 6985P-C and a 512-CPU AMD EPYC 9B45 shows several > sources of avoidable O(N^2) and O(N) work per step: >=20 > 1. build_sched_groups() and build_overlap_sched_groups() iterate over > every bit of sched_domain_span(sd) via for_each_cpu_wrap(i, span, cpu) > and check cpumask_test_cpu(i, covered) one CPU at a time, even after > entire child groups (e.g., 120-CPU NUMA nodes) have been added to > covered. Instead, track the remaining uncovered CPUs > (cpumask_copy(uncovered, span)), clear each built group's span with > cpumask_andnot(), and advance directly to the next uncovered CPU. >=20 > 2. __visit_domain_allocation_hell() and __free_domain_allocs() allocate > and free 3 * nr_levels + 2 temporary per-CPU pointer arrays via > alloc_percpu() / free_percpu() on every domain rebuild. Reuse the > per-CPU pointer arrays across domain rebuilds under > sched_domains_mutex, clearing entries in __free_domain_allocs() and > only freeing the topology-level arrays when sched_domain_topology is > replaced in sched_init_numa() / sched_reset_numa(). sched_domains_mutex does not protect the free: [snip] > #ifdef CONFIG_NUMA > +static void __sdt_free_percpu(void) > +{ > + struct sched_domain_topology_level *tl; > + > + for_each_sd_topology(tl) { > + struct sd_data *sdd =3D &tl->data; > + sched_init_numa() and sched_reset_numa() are called from sched_update_numa(), which runs from sched_cpu_activate() and sched_cpu_deactivate() without sched_domains_mutex. The free is safe because every domain builder holds at least cpus_read_lock(), and sched_update_numa() runs under the hotplug write lock.=C2=A0 Please fix the changelog and add lockdep_assert_cpus_held() in __sdt_free_percpu(), so a future caller cannot break this without a warn= ing. > + free_percpu(sdd->sd); > + sdd->sd =3D NULL; > + free_percpu(sdd->sg); > + sdd->sg =3D NULL; > + free_percpu(sdd->sgc); > + sdd->sgc =3D NULL; > + } > +} > + > static int cpu_numa_flags(void) > { > return SD_NUMA; > @@ -2449,6 +2483,7 @@ void sched_init_numa(int offline_node) > /* > * Copy the default topology bits.. > */ > + __sdt_free_percpu(); > for (i =3D 0; sched_domain_topology[i].mask; i++) > tl[i] =3D sched_domain_topology[i]; > =20 > @@ -2507,6 +2542,7 @@ static void sched_reset_numa(void) > kfree(masks); > } > if (sched_domain_topology_saved) { > + __sdt_free_percpu(); > kfree(sched_domain_topology); > sched_domain_topology =3D sched_domain_topology_saved; > sched_domain_topology_saved =3D NULL; > @@ -2708,17 +2744,23 @@ static int __sdt_alloc(const struct cpumask *cpu_= map) > for_each_sd_topology(tl) { > struct sd_data *sdd =3D &tl->data; > =20 > - sdd->sd =3D alloc_percpu(struct sched_domain *); > - if (!sdd->sd) > - return -ENOMEM; > + if (!sdd->sd) { > + sdd->sd =3D alloc_percpu(struct sched_domain *); > + if (!sdd->sd) > + return -ENOMEM; > + } > =20 > - sdd->sg =3D alloc_percpu(struct sched_group *); > - if (!sdd->sg) > - return -ENOMEM; > + if (!sdd->sg) { > + sdd->sg =3D alloc_percpu(struct sched_group *); > + if (!sdd->sg) > + return -ENOMEM; > + } > =20 > - sdd->sgc =3D alloc_percpu(struct sched_group_capacity *); > - if (!sdd->sgc) > - return -ENOMEM; > + if (!sdd->sgc) { > + sdd->sgc =3D alloc_percpu(struct sched_group_capacity *); > + if (!sdd->sgc) > + return -ENOMEM; > + } > =20 Keeping the arrays also means every slot must be NULL when a build=20 starts. Patch 5 depends on that: get_group(),=20 init_overlap_sched_group() and init_sched_domain_shared() treat a NULL=20 slot as "not allocated yet", and the overlap fast path treats=20 "sgc->ref > 0" as "set up in this build". An old pointer would pass these checks and be used without a warning. The rule holds today, but nothing says so. It also depends on builds never filling a slot outside cpu_map, because the clearing loops only visit cpu_map. Please write this down and check it, for example in __sdt_alloc(): > for_each_cpu(j, cpu_map) { > struct sched_domain *sd; Maybe add /* * The slot arrays are kept between builds. Every slot for a CPU * in cpu_map must be NULL here; lazy allocation treats NULL as * "not allocated in this build". claim_allocations() and * __sdt_free()/__sds_free() clear the slots, but only for * cpu_map, so a build must not fill slots outside cpu_map. */ for_each_cpu(j, cpu_map) { struct sched_domain *sd; WARN_ON_ONCE(*per_cpu_ptr(sdd->sd, j) || *per_cpu_ptr(sdd->sg, j) || *per_cpu_ptr(sdd->sgc, j)); > @@ -2748,30 +2790,36 @@ static void __sdt_free(const struct cpumask *cpu_= map) > =20 > if (sdd->sd) { > sd =3D *per_cpu_ptr(sdd->sd, j); > - if (sd && (sd->flags & SD_NUMA)) > - free_sched_groups(sd->groups, 0); > - kfree(*per_cpu_ptr(sdd->sd, j)); > + if (sd) { > + if (sd->flags & SD_NUMA) > + free_sched_groups(sd->groups, 0); > + kfree(sd); > + *per_cpu_ptr(sdd->sd, j) =3D NULL; > + } > } > =20 > - if (sdd->sg) > + if (sdd->sg) { > kfree(*per_cpu_ptr(sdd->sg, j)); > - if (sdd->sgc) > + *per_cpu_ptr(sdd->sg, j) =3D NULL; > + } > + if (sdd->sgc) { > kfree(*per_cpu_ptr(sdd->sgc, j)); > + *per_cpu_ptr(sdd->sgc, j) =3D NULL; > + } > } > - free_percpu(sdd->sd); > - sdd->sd =3D NULL; > - free_percpu(sdd->sg); > - sdd->sg =3D NULL; > - free_percpu(sdd->sgc); > - sdd->sgc =3D NULL; > } > } > =20 > +static struct sched_domain_shared * __percpu *sched_domains_sds; > + > static int __sds_alloc(struct s_data *d, const struct cpumask *cpu_map) > { and maybe similar check=20 for_each_cpu(j, cpu_map) WARN_ON_ONCE(*per_cpu_ptr(d->sds, j)); > - d->sds =3D alloc_percpu(struct sched_domain_shared *); > - if (!d->sds) > - return -ENOMEM; > + if (!sched_domains_sds) { > + sched_domains_sds =3D alloc_percpu(struct sched_domain_shared *); > + if (!sched_domains_sds) > + return -ENOMEM; > + } > + d->sds =3D sched_domains_sds; > =20 > return 0; > } Thanks. Tim