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 kanga.kvack.org (kanga.kvack.org [205.233.56.17]) by smtp.lore.kernel.org (Postfix) with ESMTP id 14E35C83F17 for ; Fri, 11 Jul 2025 00:42:20 +0000 (UTC) Received: by kanga.kvack.org (Postfix) id 64FBA6B009A; Thu, 10 Jul 2025 20:42:20 -0400 (EDT) Received: by kanga.kvack.org (Postfix, from userid 40) id 626C66B009B; Thu, 10 Jul 2025 20:42:20 -0400 (EDT) X-Delivered-To: int-list-linux-mm@kvack.org Received: by kanga.kvack.org (Postfix, from userid 63042) id 4C7236B009E; Thu, 10 Jul 2025 20:42:20 -0400 (EDT) X-Delivered-To: linux-mm@kvack.org Received: from relay.hostedemail.com (smtprelay0016.hostedemail.com [216.40.44.16]) by kanga.kvack.org (Postfix) with ESMTP id 3AA4E6B009B for ; Thu, 10 Jul 2025 20:42:20 -0400 (EDT) Received: from smtpin06.hostedemail.com (a10.router.float.18 [10.200.18.1]) by unirelay02.hostedemail.com (Postfix) with ESMTP id DED9612D7F1 for ; Fri, 11 Jul 2025 00:42:19 +0000 (UTC) X-FDA: 83650132398.06.C68118B Received: from mail-pg1-f175.google.com (mail-pg1-f175.google.com [209.85.215.175]) by imf11.hostedemail.com (Postfix) with ESMTP id BFBE240010 for ; Fri, 11 Jul 2025 00:42:17 +0000 (UTC) Authentication-Results: imf11.hostedemail.com; dkim=pass header.d=purestorage.com header.s=google2022 header.b=IaUSx5JU; spf=pass (imf11.hostedemail.com: domain of cachen@purestorage.com designates 209.85.215.175 as permitted sender) smtp.mailfrom=cachen@purestorage.com; dmarc=pass (policy=reject) header.from=purestorage.com ARC-Seal: i=1; s=arc-20220608; d=hostedemail.com; t=1752194538; a=rsa-sha256; cv=none; b=QWvhv3pFyqModU+gGzM9VTwUoS1aJHUVtAi7P7cL2fXSV+Zvkqkzh0vGlZWOciGlZclwnw LVgXkC3CU/dAguEAfpZluhPf85g+rej4bwrt0qgbIp4LCF4rAeEz2k/J89du/+bFSOWx30 gpUe6P9oS9hhHPuB3l65LECTSpNvgZg= ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=hostedemail.com; s=arc-20220608; t=1752194538; h=from:from:sender:reply-to:subject:subject:date:date: message-id:message-id:to:to:cc:cc:mime-version:mime-version: content-type:content-type: content-transfer-encoding:content-transfer-encoding: in-reply-to:in-reply-to:references:references:dkim-signature; bh=uM/h3yNoEs9CsIFWHK14qXfh4moxbnCvRDFEjWNVf0Y=; b=g/4xacTGRR3UgW0NqjSquH5+DLaH8LqylxqrrCQQcx1ZA/dSEMFbu7ImDOYlWU2ELqarfN ibme8J78MdENgMLXwmOOn1H5PRnMwAXLdQJdtAZwSy/7kUOspRVgkVfVz6+/SjfvPzNQI9 36EBi3bP8779Ybg0rO3Iskw/qKQSl3w= ARC-Authentication-Results: i=1; imf11.hostedemail.com; dkim=pass header.d=purestorage.com header.s=google2022 header.b=IaUSx5JU; spf=pass (imf11.hostedemail.com: domain of cachen@purestorage.com designates 209.85.215.175 as permitted sender) smtp.mailfrom=cachen@purestorage.com; dmarc=pass (policy=reject) header.from=purestorage.com Received: by mail-pg1-f175.google.com with SMTP id 41be03b00d2f7-b34b770868dso230515a12.2 for ; Thu, 10 Jul 2025 17:42:17 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=purestorage.com; s=google2022; t=1752194536; x=1752799336; darn=kvack.org; h=content-transfer-encoding:cc:to:subject:message-id:date:from :in-reply-to:references:mime-version:from:to:cc:subject:date :message-id:reply-to; bh=uM/h3yNoEs9CsIFWHK14qXfh4moxbnCvRDFEjWNVf0Y=; b=IaUSx5JUDLMpwli8KPSrag4Gk2zB3d94woO9RJwXU2hqzNs66bJ8glVa+RfL7mfYe5 JMFtu71DlzldO2Dat4zuXvfeOitGBHVa19DpapoRwi1dISSQpPWW97kUvD1FFyoW/pLx rJKAUHwmFNHKQs5cKu9OZpfwRGD5T5x8SmK0lme68x6VkUcerLw7yfH8IA9psDzhUCau YF/mHrfamSYtYoj4GVgrO2iUySFV2BytKd8u7UIvUFwh0pIk87jWPzZm2SHw1pbbW1g+ C6ds/GfIdjosjQlxbjuLZH/Y2WI32bCqANmpOtL+QzI4+ZLTq4zYKlPj2P3Cz89tGCkk 3ktg== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20230601; t=1752194536; x=1752799336; h=content-transfer-encoding:cc:to:subject:message-id:date:from :in-reply-to:references:mime-version:x-gm-message-state:from:to:cc :subject:date:message-id:reply-to; bh=uM/h3yNoEs9CsIFWHK14qXfh4moxbnCvRDFEjWNVf0Y=; b=CEP3lfS2G7zGjLmtfBjP28SuS/zopDUkWilshqnXKLPnf72WXfs6HkbJS/X+nM80m3 Z7FH+p7nhCIhEcOAC0EkJjD+1TDikPqUQGGDGQbqWnUZ6F2tzHDtYV2ikZuwqJWyK9KC hummOgS+HACruBozVKjjYkwIf4MyIjqKz/XxJsTLzKMeTjWws8JRXVn9c4Wru19sk3c2 VY12jRXMMpbuExMzxzDTYOkmLG8jX5qjnGu1gdRuzaY/Gz0vXdbaU9HBRFFkB1YpU+vE kZbisxWbQwv2IvG6C1wJtRgSTovU6hoF7gsqRAs/5n2O3UMgo/Wkg7AjJ0wmIfwDLEPw bxAg== X-Forwarded-Encrypted: i=1; AJvYcCUrBeLma8W5CUWEOxOHHr5DG9o0QOcq4+Ta4m3LSxFnMee2OoSGEOeoMg/9BzETsVen9akLf+hZmA==@kvack.org X-Gm-Message-State: AOJu0Yx3BBsNXTdbZw3WJjJhPVTBXPIxIMG1fnHrYj1Dt5gHkZvAZbON 2dmbrXtflTOnQrgQBRrv4Q65/A4vv6RiSRXPsZ7jHUiFGNzgQ12H3AwSWMoP2Z7MQmZex5I2wqS KIz0ySVZiz4uLmIOZN9CHNasb2tU92gwGsacrmkh0Ig== X-Gm-Gg: ASbGncsmr066Bj81dFdsoFuP0tKn4GjhUaaOOQgeJJW4RU/I5yA8OlPv5gRvLodUz3f U85tUVaHI1Ko+EZKRBS8V80sEpaxFaEKj+iuKRSBTWvBDqPOt64Ux/rJd/8YgiiDazrjqbT1ZI2 o1PbP5Pvv/95Dp2JBrGiZF7IDVDubC4oLIGLZNECaCcgQcaM6T8EyuE1Gi6vxga4txrZpWvluSF FJRBdc= X-Google-Smtp-Source: AGHT+IEIrbepiNMlFDkddxoGgoQTe3H2om5nGFBUlxSGtrjszr0HRcS+nnbPrUlVyaqEUszXmYUO6Gk95/+mWE/V8bc= X-Received: by 2002:a17:90b:3a46:b0:312:ec:411a with SMTP id 98e67ed59e1d1-31c4cce366fmr791614a91.3.1752194536081; Thu, 10 Jul 2025 17:42:16 -0700 (PDT) MIME-Version: 1.0 References: <20250711002322.1303421-1-cachen@purestorage.com> In-Reply-To: <20250711002322.1303421-1-cachen@purestorage.com> From: Casey Chen Date: Thu, 10 Jul 2025 17:42:05 -0700 X-Gm-Features: Ac12FXx3dZ34TbUeDk6x4XjpBwJPTiPYHqVoucQN-FWJIdlIxqkMKkhUfztfxMo Message-ID: Subject: Re: [PATCH v3] alloc_tag: add per-NUMA node stats To: akpm@linux-foundation.org, surenb@google.com Cc: kent.overstreet@linux.dev, corbet@lwn.net, dennis@kernel.org, tj@kernel.org, cl@gentwo.org, vbabka@suse.cz, mhocko@suse.com, jackmanb@google.com, hannes@cmpxchg.org, ziy@nvidia.com, rientjes@google.com, roman.gushchin@linux.dev, harry.yoo@oracle.com, linux-mm@kvack.org, linux-kernel@vger.kernel.org, linux-doc@vger.kernel.org, yzhong@purestorage.com, souravpanda@google.com, 00107082@163.com Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: quoted-printable X-Rspam-User: X-Rspamd-Server: rspam05 X-Rspamd-Queue-Id: BFBE240010 X-Stat-Signature: thty49zxar3y4w9mq6u1456t8rh6ny5g X-HE-Tag: 1752194537-924468 X-HE-Meta: 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 iQm8/NfZ yai0oiaqG1k1DMu2gzTzSNb9jU403sSsmo5hCJQFc3xgZ5jupwZC6CwlD9MSOBmqhbQhgv3FO9iz+/xPA0ajMtwmZX0W9I87Tilu/LEQuzYstWRVUlgZ2rb8IejSNIQg/Kv2EjHP/L4oF/qPlSQDjAhqKENOcQz9/58q9vhHPfibJ/O9YhdFUYLNVPhN/6Y+sFAl6t6nY6uEcoRBlp9oTQgnIRvt+iaf6nGWCZ2PHVGEMgkDIxBzECL2aakQdmkCJxfHu2NQw3t3G/6/ta4c6OvCRe7t2Yptst18tg6tLpzsqG4XDE91OLYw91VIauTMu9t5CL3yM6hXmd8nrx7gkXUfGN892pnbiQHU/XBoRiZo9ki1AhRvmOV0/SWZ4VHHMLBl2ewF5QQpHgoY= X-Bogosity: Ham, tests=bogofilter, spamicity=0.000000, version=1.2.4 Sender: owner-linux-mm@kvack.org Precedence: bulk X-Loop: owner-majordomo@kvack.org List-ID: List-Subscribe: List-Unsubscribe: Hi All, Thanks for reviewing my previous patches. I am replying some comments in our previous discussion https://lore.kernel.org/all/CAJuCfpHhSUhxer-6MP3503w6520YLfgBTGp7Q9Qm9kgN4T= Nsfw@mail.gmail.com/T/#u Most people care about the motivations and usages of this feature. Internally, we used to have systems having asymmetric memory to NUMA nodes. Node 0 uses a lot of memory but node 1 is pretty empty. Requests to allocate memory on node 0 always fail. With this patch, we can find the imbalance and optimize the memory usage. Also, David Rientjes and Sourav Panda provide their scenarios in which this patch would be very useful. It is easy to turn on an off so I think it is nice to have, enabling more scenarios in the future. Andrew / Kent, * I agree with Kent on using for_each_possible_cpu rather than for_each_online_cpu, considering CPU online/offline. * When failing to allocate counters for in-kernel alloc_tag, panic() is better than WARN(), eventually the kernel would panic at invalid memory access. * percpu stats would bloat data structures quite a bit. David Wang, I don't really understand what is 'granularity of calling sites'. If NUMA imbalance is found, the calling site could request memory allocation from different nodes. Other factors can affect NUMA balance, those information can be implemented in a different patch. Thanks, Casey On Thu, Jul 10, 2025 at 5:23=E2=80=AFPM Casey Chen = wrote: > > This patch adds per-NUMA node breakdown of memory allocation, > enabling more precise visibility into memory usage patterns across nodes. > It is particularly valuable in cloud environments, > where tracking asymmetric memory usage and identifying NUMA imbalances > down to the allocation caller helps optimize memory efficiency, avoid > CPU stranding, and improve system responsiveness under memory pressure. > > As implementation, it adds per-NUMA node statistics in /proc/allocinfo. > Previously, each alloc_tag had a single set of counters (bytes and > calls), aggregated across all CPUs. With this change, each CPU can > maintain separate counters for each NUMA node, allowing finer-grained > memory allocation profiling. > > This feature is controlled by the new > CONFIG_MEM_ALLOC_PROFILING_PER_NUMA_STATS option: > > * When enabled (=3Dy), the output includes per-node statistics following > the total bytes/calls: > > > ... > 315456 9858 mm/dmapool.c:338 func:pool_alloc_page > nid0 94912 2966 > nid1 220544 6892 > 7680 60 mm/dmapool.c:254 func:dma_pool_create > nid0 4224 33 > nid1 3456 27 > > * When disabled (=3Dn), the output remains unchanged: > > ... > 315456 9858 mm/dmapool.c:338 func:pool_alloc_page > 7680 60 mm/dmapool.c:254 func:dma_pool_create > > To minimize memory overhead, per-NUMA stats counters are dynamically > allocated using the percpu allocator. PERCPU_DYNAMIC_RESERVE has been > increased to ensure sufficient space for in-kernel alloc_tag counters. > > For in-kernel alloc_tag instances, pcpu_alloc_noprof() is used to > allocate counters. These allocations are excluded from the profiling > statistics themselves. > > Link: https://lore.kernel.org/all/20250610233053.973796-1-cachen@purestor= age.com > Link: https://lore.kernel.org/all/20250530003944.2929392-1-cachen@puresto= rage.com > Signed-off-by: Casey Chen > Reviewed-by: Yuanyuan Zhong > Cc: David Rientjes > Cc: Sourav Panda > --- > Documentation/mm/allocation-profiling.rst | 3 ++ > include/linux/alloc_tag.h | 52 ++++++++++++++++------ > include/linux/codetag.h | 4 ++ > include/linux/percpu.h | 2 +- > lib/Kconfig.debug | 7 +++ > lib/alloc_tag.c | 54 ++++++++++++++++++++--- > mm/page_alloc.c | 35 ++++++++------- > mm/percpu.c | 8 +++- > mm/show_mem.c | 25 ++++++++--- > mm/slub.c | 11 +++-- > 10 files changed, 150 insertions(+), 51 deletions(-) > > diff --git a/Documentation/mm/allocation-profiling.rst b/Documentation/mm= /allocation-profiling.rst > index 316311240e6a..13d1d0cb91bf 100644 > --- a/Documentation/mm/allocation-profiling.rst > +++ b/Documentation/mm/allocation-profiling.rst > @@ -17,6 +17,9 @@ kconfig options: > adds warnings for allocations that weren't accounted because of a > missing annotation > > +- CONFIG_MEM_ALLOC_PROFILING_PER_NUMA_STATS > + adds memory allocation profiling stats for each numa node, off by defa= ult. > + > Boot parameter: > sysctl.vm.mem_profiling=3D{0|1|never}[,compressed] > > diff --git a/include/linux/alloc_tag.h b/include/linux/alloc_tag.h > index 9ef2633e2c08..f714f1a436ec 100644 > --- a/include/linux/alloc_tag.h > +++ b/include/linux/alloc_tag.h > @@ -15,6 +15,12 @@ > #include > #include > > +#ifdef CONFIG_MEM_ALLOC_PROFILING_PER_NUMA_STATS > +#define ALLOC_TAG_NUM_NODES num_possible_nodes() > +#else > +#define ALLOC_TAG_NUM_NODES 1 > +#endif > + > struct alloc_tag_counters { > u64 bytes; > u64 calls; > @@ -134,16 +140,33 @@ static inline bool mem_alloc_profiling_enabled(void= ) > &mem_alloc_profiling_key); > } > > +static inline struct alloc_tag_counters alloc_tag_read_nid(struct alloc_= tag *tag, int nid) > +{ > + struct alloc_tag_counters v =3D { 0, 0 }; > + struct alloc_tag_counters *counters; > + int cpu; > + > + for_each_possible_cpu(cpu) { > + counters =3D per_cpu_ptr(tag->counters, cpu); > + v.bytes +=3D counters[nid].bytes; > + v.calls +=3D counters[nid].calls; > + } > + > + return v; > +} > + > static inline struct alloc_tag_counters alloc_tag_read(struct alloc_tag = *tag) > { > struct alloc_tag_counters v =3D { 0, 0 }; > - struct alloc_tag_counters *counter; > + struct alloc_tag_counters *counters; > int cpu; > > for_each_possible_cpu(cpu) { > - counter =3D per_cpu_ptr(tag->counters, cpu); > - v.bytes +=3D counter->bytes; > - v.calls +=3D counter->calls; > + counters =3D per_cpu_ptr(tag->counters, cpu); > + for (int nid =3D 0; nid < ALLOC_TAG_NUM_NODES; nid++) { > + v.bytes +=3D counters[nid].bytes; > + v.calls +=3D counters[nid].calls; > + } > } > > return v; > @@ -179,7 +202,7 @@ static inline bool __alloc_tag_ref_set(union codetag_= ref *ref, struct alloc_tag > return true; > } > > -static inline bool alloc_tag_ref_set(union codetag_ref *ref, struct allo= c_tag *tag) > +static inline bool alloc_tag_ref_set(union codetag_ref *ref, struct allo= c_tag *tag, int nid) > { > if (unlikely(!__alloc_tag_ref_set(ref, tag))) > return false; > @@ -190,17 +213,18 @@ static inline bool alloc_tag_ref_set(union codetag_= ref *ref, struct alloc_tag *t > * Each new reference for every sub-allocation needs to increment= call > * counter because when we free each part the counter will be dec= remented. > */ > - this_cpu_inc(tag->counters->calls); > + this_cpu_inc(tag->counters[nid].calls); > return true; > } > > -static inline void alloc_tag_add(union codetag_ref *ref, struct alloc_ta= g *tag, size_t bytes) > +static inline void alloc_tag_add(union codetag_ref *ref, struct alloc_ta= g *tag, > + int nid, size_t bytes) > { > - if (likely(alloc_tag_ref_set(ref, tag))) > - this_cpu_add(tag->counters->bytes, bytes); > + if (likely(alloc_tag_ref_set(ref, tag, nid))) > + this_cpu_add(tag->counters[nid].bytes, bytes); > } > > -static inline void alloc_tag_sub(union codetag_ref *ref, size_t bytes) > +static inline void alloc_tag_sub(union codetag_ref *ref, int nid, size_t= bytes) > { > struct alloc_tag *tag; > > @@ -215,8 +239,8 @@ static inline void alloc_tag_sub(union codetag_ref *r= ef, size_t bytes) > > tag =3D ct_to_alloc_tag(ref->ct); > > - this_cpu_sub(tag->counters->bytes, bytes); > - this_cpu_dec(tag->counters->calls); > + this_cpu_sub(tag->counters[nid].bytes, bytes); > + this_cpu_dec(tag->counters[nid].calls); > > ref->ct =3D NULL; > } > @@ -228,8 +252,8 @@ static inline void alloc_tag_sub(union codetag_ref *r= ef, size_t bytes) > #define DEFINE_ALLOC_TAG(_alloc_tag) > static inline bool mem_alloc_profiling_enabled(void) { return false; } > static inline void alloc_tag_add(union codetag_ref *ref, struct alloc_ta= g *tag, > - size_t bytes) {} > -static inline void alloc_tag_sub(union codetag_ref *ref, size_t bytes) {= } > + int nid, size_t bytes) {} > +static inline void alloc_tag_sub(union codetag_ref *ref, int nid, size_t= bytes) {} > #define alloc_tag_record(p) do {} while (0) > > #endif /* CONFIG_MEM_ALLOC_PROFILING */ > diff --git a/include/linux/codetag.h b/include/linux/codetag.h > index 457ed8fd3214..35b314b36633 100644 > --- a/include/linux/codetag.h > +++ b/include/linux/codetag.h > @@ -16,6 +16,10 @@ struct module; > #define CODETAG_SECTION_START_PREFIX "__start_" > #define CODETAG_SECTION_STOP_PREFIX "__stop_" > > +enum codetag_flags { > + CODETAG_PERCPU_ALLOC =3D (1 << 0), /* codetag tracking percpu = allocation */ > +}; > + > /* > * An instance of this structure is created in a special ELF section at = every > * code location being tagged. At runtime, the special section is treat= ed as > diff --git a/include/linux/percpu.h b/include/linux/percpu.h > index 85bf8dd9f087..d92c27fbcd0d 100644 > --- a/include/linux/percpu.h > +++ b/include/linux/percpu.h > @@ -43,7 +43,7 @@ > # define PERCPU_DYNAMIC_SIZE_SHIFT 12 > #endif /* LOCKDEP and PAGE_SIZE > 4KiB */ > #else > -#define PERCPU_DYNAMIC_SIZE_SHIFT 10 > +#define PERCPU_DYNAMIC_SIZE_SHIFT 13 > #endif > > /* > diff --git a/lib/Kconfig.debug b/lib/Kconfig.debug > index ebe33181b6e6..b2a35cc78635 100644 > --- a/lib/Kconfig.debug > +++ b/lib/Kconfig.debug > @@ -1037,6 +1037,13 @@ config MEM_ALLOC_PROFILING_DEBUG > Adds warnings with helpful error messages for memory allocation > profiling. > > +config MEM_ALLOC_PROFILING_PER_NUMA_STATS > + bool "Memory allocation profiling per-NUMA stats" > + default n > + depends on MEM_ALLOC_PROFILING > + help > + Display allocation stats on every NUMA node. > + > source "lib/Kconfig.kasan" > source "lib/Kconfig.kfence" > source "lib/Kconfig.kmsan" > diff --git a/lib/alloc_tag.c b/lib/alloc_tag.c > index e9b33848700a..3b170847f547 100644 > --- a/lib/alloc_tag.c > +++ b/lib/alloc_tag.c > @@ -40,6 +40,9 @@ struct alloc_tag_kernel_section kernel_tags =3D { NULL,= 0 }; > unsigned long alloc_tag_ref_mask; > int alloc_tag_ref_offs; > > +/* Total size of all alloc_tag_counters of each CPU */ > +static unsigned long pcpu_counters_size; > + > struct allocinfo_private { > struct codetag_iterator iter; > bool print_header; > @@ -81,7 +84,7 @@ static void print_allocinfo_header(struct seq_buf *buf) > { > /* Output format version, so we can change it. */ > seq_buf_printf(buf, "allocinfo - version: 1.0\n"); > - seq_buf_printf(buf, "# \n"); > + seq_buf_printf(buf, " \n"); > } > > static void alloc_tag_to_text(struct seq_buf *out, struct codetag *ct) > @@ -90,12 +93,32 @@ static void alloc_tag_to_text(struct seq_buf *out, st= ruct codetag *ct) > struct alloc_tag_counters counter =3D alloc_tag_read(tag); > s64 bytes =3D counter.bytes; > > - seq_buf_printf(out, "%12lli %8llu ", bytes, counter.calls); > + seq_buf_printf(out, "%-12lli %-8llu ", bytes, counter.calls); > codetag_to_text(out, ct); > seq_buf_putc(out, ' '); > seq_buf_putc(out, '\n'); > } > > +#ifdef CONFIG_MEM_ALLOC_PROFILING_PER_NUMA_STATS > +static void alloc_tag_to_text_all_nids(struct seq_buf *out, struct codet= ag *ct) > +{ > + struct alloc_tag *tag =3D ct_to_alloc_tag(ct); > + struct alloc_tag_counters counter; > + s64 bytes; > + > + for (int nid =3D 0; nid < ALLOC_TAG_NUM_NODES; nid++) { > + counter =3D alloc_tag_read_nid(tag, nid); > + bytes =3D counter.bytes; > + seq_buf_printf(out, " nid%-5u %-12lli %-8llu\n", > + nid, bytes, counter.calls); > + } > +} > +#else > +static void alloc_tag_to_text_all_nids(struct seq_buf *out, struct codet= ag *ct) > +{ > +} > +#endif > + > static int allocinfo_show(struct seq_file *m, void *arg) > { > struct allocinfo_private *priv =3D (struct allocinfo_private *)ar= g; > @@ -109,6 +132,7 @@ static int allocinfo_show(struct seq_file *m, void *a= rg) > priv->print_header =3D false; > } > alloc_tag_to_text(&buf, priv->iter.ct); > + alloc_tag_to_text_all_nids(&buf, priv->iter.ct); > seq_commit(m, seq_buf_used(&buf)); > return 0; > } > @@ -180,7 +204,7 @@ void pgalloc_tag_split(struct folio *folio, int old_o= rder, int new_order) > > if (get_page_tag_ref(folio_page(folio, i), &ref, &handle)= ) { > /* Set new reference to point to the original tag= */ > - alloc_tag_ref_set(&ref, tag); > + alloc_tag_ref_set(&ref, tag, folio_nid(folio)); > update_page_tag_ref(handle, &ref); > put_page_tag_ref(handle); > } > @@ -247,15 +271,29 @@ void __init alloc_tag_sec_init(void) > if (!mem_profiling_support) > return; > > - if (!static_key_enabled(&mem_profiling_compressed)) > - return; > - > kernel_tags.first_tag =3D (struct alloc_tag *)kallsyms_lookup_nam= e( > SECTION_START(ALLOC_TAG_SECTION_N= AME)); > last_codetag =3D (struct alloc_tag *)kallsyms_lookup_name( > SECTION_STOP(ALLOC_TAG_SECTION_NA= ME)); > kernel_tags.count =3D last_codetag - kernel_tags.first_tag; > > + pcpu_counters_size =3D ALLOC_TAG_NUM_NODES * sizeof(struct alloc_= tag_counters); > + for (int i =3D 0; i < kernel_tags.count; i++) { > + /* Each CPU has one alloc_tag_counters per numa node */ > + kernel_tags.first_tag[i].counters =3D > + pcpu_alloc_noprof(pcpu_counters_size, > + sizeof(struct alloc_tag_counter= s), > + false, GFP_KERNEL | __GFP_ZERO)= ; > + if (!kernel_tags.first_tag[i].counters) { > + while (--i >=3D 0) > + free_percpu(kernel_tags.first_tag[i].coun= ters); > + panic("Failed to allocate per-cpu alloc_tag count= ers\n"); > + } > + } > + > + if (!static_key_enabled(&mem_profiling_compressed)) > + return; > + > /* Check if kernel tags fit into page flags */ > if (kernel_tags.count > (1UL << NR_UNUSED_PAGEFLAG_BITS)) { > shutdown_mem_profiling(false); /* allocinfo file does not= exist yet */ > @@ -618,7 +656,9 @@ static int load_module(struct module *mod, struct cod= etag *start, struct codetag > stop_tag =3D ct_to_alloc_tag(stop); > for (tag =3D start_tag; tag < stop_tag; tag++) { > WARN_ON(tag->counters); > - tag->counters =3D alloc_percpu(struct alloc_tag_counters)= ; > + tag->counters =3D __alloc_percpu_gfp(pcpu_counters_size, > + sizeof(struct alloc_ta= g_counters), > + GFP_KERNEL | __GFP_ZER= O); > if (!tag->counters) { > while (--tag >=3D start_tag) { > free_percpu(tag->counters); > diff --git a/mm/page_alloc.c b/mm/page_alloc.c > index 78ddf1d43c6c..7c4d10f6873c 100644 > --- a/mm/page_alloc.c > +++ b/mm/page_alloc.c > @@ -1247,58 +1247,59 @@ void __clear_page_tag_ref(struct page *page) > /* Should be called only if mem_alloc_profiling_enabled() */ > static noinline > void __pgalloc_tag_add(struct page *page, struct task_struct *task, > - unsigned int nr) > + int nid, unsigned int nr) > { > union pgtag_ref_handle handle; > union codetag_ref ref; > > if (get_page_tag_ref(page, &ref, &handle)) { > - alloc_tag_add(&ref, task->alloc_tag, PAGE_SIZE * nr); > + alloc_tag_add(&ref, task->alloc_tag, nid, PAGE_SIZE * nr)= ; > update_page_tag_ref(handle, &ref); > put_page_tag_ref(handle); > } > } > > static inline void pgalloc_tag_add(struct page *page, struct task_struct= *task, > - unsigned int nr) > + int nid, unsigned int nr) > { > if (mem_alloc_profiling_enabled()) > - __pgalloc_tag_add(page, task, nr); > + __pgalloc_tag_add(page, task, nid, nr); > } > > /* Should be called only if mem_alloc_profiling_enabled() */ > static noinline > -void __pgalloc_tag_sub(struct page *page, unsigned int nr) > +void __pgalloc_tag_sub(struct page *page, int nid, unsigned int nr) > { > union pgtag_ref_handle handle; > union codetag_ref ref; > > if (get_page_tag_ref(page, &ref, &handle)) { > - alloc_tag_sub(&ref, PAGE_SIZE * nr); > + alloc_tag_sub(&ref, nid, PAGE_SIZE * nr); > update_page_tag_ref(handle, &ref); > put_page_tag_ref(handle); > } > } > > -static inline void pgalloc_tag_sub(struct page *page, unsigned int nr) > +static inline void pgalloc_tag_sub(struct page *page, int nid, unsigned = int nr) > { > if (mem_alloc_profiling_enabled()) > - __pgalloc_tag_sub(page, nr); > + __pgalloc_tag_sub(page, nid, nr); > } > > /* When tag is not NULL, assuming mem_alloc_profiling_enabled */ > -static inline void pgalloc_tag_sub_pages(struct alloc_tag *tag, unsigned= int nr) > +static inline void pgalloc_tag_sub_pages(struct alloc_tag *tag, > + int nid, unsigned int nr) > { > if (tag) > - this_cpu_sub(tag->counters->bytes, PAGE_SIZE * nr); > + this_cpu_sub(tag->counters[nid].bytes, PAGE_SIZE * nr); > } > > #else /* CONFIG_MEM_ALLOC_PROFILING */ > > static inline void pgalloc_tag_add(struct page *page, struct task_struct= *task, > - unsigned int nr) {} > -static inline void pgalloc_tag_sub(struct page *page, unsigned int nr) {= } > -static inline void pgalloc_tag_sub_pages(struct alloc_tag *tag, unsigned= int nr) {} > + int nid, unsigned int nr) {} > +static inline void pgalloc_tag_sub(struct page *page, int nid, unsigned = int nr) {} > +static inline void pgalloc_tag_sub_pages(struct alloc_tag *tag, int nid,= unsigned int nr) {} > > #endif /* CONFIG_MEM_ALLOC_PROFILING */ > > @@ -1337,7 +1338,7 @@ __always_inline bool free_pages_prepare(struct page= *page, > /* Do not let hwpoison pages hit pcplists/buddy */ > reset_page_owner(page, order); > page_table_check_free(page, order); > - pgalloc_tag_sub(page, 1 << order); > + pgalloc_tag_sub(page, page_to_nid(page), 1 << order); > > /* > * The page is isolated and accounted for. > @@ -1394,7 +1395,7 @@ __always_inline bool free_pages_prepare(struct page= *page, > page->flags &=3D ~PAGE_FLAGS_CHECK_AT_PREP; > reset_page_owner(page, order); > page_table_check_free(page, order); > - pgalloc_tag_sub(page, 1 << order); > + pgalloc_tag_sub(page, page_to_nid(page), 1 << order); > > if (!PageHighMem(page)) { > debug_check_no_locks_freed(page_address(page), > @@ -1850,7 +1851,7 @@ inline void post_alloc_hook(struct page *page, unsi= gned int order, > > set_page_owner(page, order, gfp_flags); > page_table_check_alloc(page, order); > - pgalloc_tag_add(page, current, 1 << order); > + pgalloc_tag_add(page, current, page_to_nid(page), 1 << order); > } > > static void prep_new_page(struct page *page, unsigned int order, gfp_t g= fp_flags, > @@ -5228,7 +5229,7 @@ static void ___free_pages(struct page *page, unsign= ed int order, > if (put_page_testzero(page)) > __free_frozen_pages(page, order, fpi_flags); > else if (!head) { > - pgalloc_tag_sub_pages(tag, (1 << order) - 1); > + pgalloc_tag_sub_pages(tag, page_to_nid(page), (1 << order= ) - 1); > while (order-- > 0) > __free_frozen_pages(page + (1 << order), order, > fpi_flags); > diff --git a/mm/percpu.c b/mm/percpu.c > index 782cc148b39c..4c5369a40323 100644 > --- a/mm/percpu.c > +++ b/mm/percpu.c > @@ -1691,15 +1691,19 @@ static void pcpu_alloc_tag_alloc_hook(struct pcpu= _chunk *chunk, int off, > size_t size) > { > if (mem_alloc_profiling_enabled() && likely(chunk->obj_exts)) { > + /* For percpu allocation, store all alloc_tag stats on nu= ma node 0 */ > alloc_tag_add(&chunk->obj_exts[off >> PCPU_MIN_ALLOC_SHIF= T].tag, > - current->alloc_tag, size); > + current->alloc_tag, 0, size); > + if (current->alloc_tag) > + current->alloc_tag->ct.flags |=3D CODETAG_PERCPU_= ALLOC; > } > } > > static void pcpu_alloc_tag_free_hook(struct pcpu_chunk *chunk, int off, = size_t size) > { > + /* percpu alloc_tag stats is stored on numa node 0 so subtract fr= om node 0 */ > if (mem_alloc_profiling_enabled() && likely(chunk->obj_exts)) > - alloc_tag_sub(&chunk->obj_exts[off >> PCPU_MIN_ALLOC_SHIF= T].tag, size); > + alloc_tag_sub(&chunk->obj_exts[off >> PCPU_MIN_ALLOC_SHIF= T].tag, 0, size); > } > #else > static void pcpu_alloc_tag_alloc_hook(struct pcpu_chunk *chunk, int off, > diff --git a/mm/show_mem.c b/mm/show_mem.c > index 41999e94a56d..3939c58e55c4 100644 > --- a/mm/show_mem.c > +++ b/mm/show_mem.c > @@ -5,6 +5,7 @@ > * Copyright (C) 2008 Johannes Weiner > */ > > +#include > #include > #include > #include > @@ -426,6 +427,7 @@ void __show_mem(unsigned int filter, nodemask_t *node= mask, int max_zone_idx) > nr =3D alloc_tag_top_users(tags, ARRAY_SIZE(tags), false)= ; > if (nr) { > pr_notice("Memory allocations:\n"); > + pr_notice(" \n"); > for (i =3D 0; i < nr; i++) { > struct codetag *ct =3D tags[i].ct; > struct alloc_tag *tag =3D ct_to_alloc_tag= (ct); > @@ -433,16 +435,25 @@ void __show_mem(unsigned int filter, nodemask_t *no= demask, int max_zone_idx) > char bytes[10]; > > string_get_size(counter.bytes, 1, STRING_= UNITS_2, bytes, sizeof(bytes)); > - > /* Same as alloc_tag_to_text() but w/o in= termediate buffer */ > if (ct->modname) > - pr_notice("%12s %8llu %s:%u [%s] = func:%s\n", > - bytes, counter.calls, c= t->filename, > - ct->lineno, ct->modname= , ct->function); > + pr_notice("%-12s %-8llu %s:%u [%s= ] func:%s\n", > + bytes, counter.calls, ct-= >filename, > + ct->lineno, ct->modname, = ct->function); > else > - pr_notice("%12s %8llu %s:%u func:= %s\n", > - bytes, counter.calls, c= t->filename, > - ct->lineno, ct->functio= n); > + pr_notice("%-12s %-8llu %s:%u fun= c:%s\n", > + bytes, counter.calls, > + ct->filename, ct->lineno,= ct->function); > + > +#ifdef CONFIG_MEM_ALLOC_PROFILING_PER_NUMA_STATS > + for (int nid =3D 0; nid < ALLOC_TAG_NUM_N= ODES; nid++) { > + counter =3D alloc_tag_read_nid(ta= g, nid); > + string_get_size(counter.bytes, 1,= STRING_UNITS_2, > + bytes, sizeof(byt= es)); > + pr_notice(" nid%-5u %-12s = %-8llu\n", > + nid, bytes, counter.cal= ls); > + } > +#endif > } > } > } > diff --git a/mm/slub.c b/mm/slub.c > index c4b64821e680..1c7b10befa7c 100644 > --- a/mm/slub.c > +++ b/mm/slub.c > @@ -2106,8 +2106,12 @@ __alloc_tagging_slab_alloc_hook(struct kmem_cache = *s, void *object, gfp_t flags) > * If other users appear then mem_alloc_profiling_enabled() > * check should be added before alloc_tag_add(). > */ > - if (likely(obj_exts)) > - alloc_tag_add(&obj_exts->ref, current->alloc_tag, s->size= ); > + if (likely(obj_exts)) { > + struct page *page =3D virt_to_page(object); > + > + alloc_tag_add(&obj_exts->ref, current->alloc_tag, > + page_to_nid(page), s->size); > + } > } > > static inline void > @@ -2135,8 +2139,9 @@ __alloc_tagging_slab_free_hook(struct kmem_cache *s= , struct slab *slab, void **p > > for (i =3D 0; i < objects; i++) { > unsigned int off =3D obj_to_index(s, slab, p[i]); > + struct page *page =3D virt_to_page(p[i]); > > - alloc_tag_sub(&obj_exts[off].ref, s->size); > + alloc_tag_sub(&obj_exts[off].ref, page_to_nid(page), s->s= ize); > } > } > > -- > 2.34.1 >