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 9DB30C4332F for ; Fri, 18 Nov 2022 19:08:32 +0000 (UTC) Received: by kanga.kvack.org (Postfix) id 03E768E0001; Fri, 18 Nov 2022 14:08:32 -0500 (EST) Received: by kanga.kvack.org (Postfix, from userid 40) id F309D6B0072; Fri, 18 Nov 2022 14:08:31 -0500 (EST) X-Delivered-To: int-list-linux-mm@kvack.org Received: by kanga.kvack.org (Postfix, from userid 63042) id DAA478E0001; Fri, 18 Nov 2022 14:08:31 -0500 (EST) X-Delivered-To: linux-mm@kvack.org Received: from relay.hostedemail.com (smtprelay0012.hostedemail.com [216.40.44.12]) by kanga.kvack.org (Postfix) with ESMTP id C79DB6B0071 for ; Fri, 18 Nov 2022 14:08:31 -0500 (EST) Received: from smtpin09.hostedemail.com (a10.router.float.18 [10.200.18.1]) by unirelay06.hostedemail.com (Postfix) with ESMTP id 95CEDABC75 for ; Fri, 18 Nov 2022 19:08:31 +0000 (UTC) X-FDA: 80147499222.09.D855A81 Received: from mail-pj1-f43.google.com (mail-pj1-f43.google.com [209.85.216.43]) by imf22.hostedemail.com (Postfix) with ESMTP id 2A1F3C000C for ; Fri, 18 Nov 2022 19:08:30 +0000 (UTC) Received: by mail-pj1-f43.google.com with SMTP id v3-20020a17090ac90300b00218441ac0f6so8700579pjt.0 for ; Fri, 18 Nov 2022 11:08:30 -0800 (PST) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; h=in-reply-to:content-disposition:mime-version:references:message-id :subject:cc:to:from:date:x-gm-message-state:from:to:cc:subject:date :message-id:reply-to; bh=I/8r8Y1eiMtDrea4Bredv83OHEKoOW04VLm26sFfGLI=; b=2AaouBIHWgnwOVoHUtjK+GvwehtN/x8W+G811NIeIoQT+XjTkmF1N2Go59PaLO4FUC 8XjbGh09vf2/Cgg0F3qLjlrCt7eI8ETnk4Ni7wXocB3QEfliflk5CE1WddaccZy1UszM v6xnP8ciMuKiiQVz488BqUKL7s0YnzZpuer9w0KqvXVSp+GI01YNNxXQ6dBP5WG1gMBA pJZVB1IzCr6KjdDZJGawstlNFlHOC+CwmAhvSL/pHz7iZDj2MlgaTSSEmY8Boet331Pq eOzvtII15VwMejxLH8K7zIuBh0NrBrjj1YmKlsgYv/2rtcVM6OkmMt0nP1r6TsMMQrXB oBdw== X-Gm-Message-State: ANoB5pm7h0YKR+Mq8n3B/VznVi9UcUEv3MvqN/bprYH05SSlElMFQte+ Ou/zqmEYjlEGskJ8Cuaqh3w= X-Google-Smtp-Source: AA0mqf4xgOvIuy0PrG+CJHcSFjIJF5Y6ybarRduM5XmaJwvCAHUMIKgE1Gqn5+T8S8OeGVqMd3sdxg== X-Received: by 2002:a17:903:228c:b0:186:8d12:f51f with SMTP id b12-20020a170903228c00b001868d12f51fmr810014plh.129.1668798509794; Fri, 18 Nov 2022 11:08:29 -0800 (PST) Received: from fedora ([2607:f598:b99a:be1:7423:52fa:7f88:fa93]) by smtp.gmail.com with ESMTPSA id s10-20020a170902a50a00b0017f57787a4asm4093531plq.229.2022.11.18.11.08.28 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Fri, 18 Nov 2022 11:08:29 -0800 (PST) Date: Fri, 18 Nov 2022 11:08:26 -0800 From: Dennis Zhou To: Vlastimil Babka Cc: Baoquan He , kernel test robot , oe-kbuild-all@lists.linux.dev, Linux Memory Management List , 42.hyeyoo@gmail.com Subject: Re: [linux-next:master 5002/7443] include/linux/compiler_types.h:357:45: error: call to '__compiletime_assert_474' declared with attribute error: BUILD_BUG_ON failed: PERCPU_DYNAMIC_EARLY_SIZE < NR_KMALLOC_TYPES * KMALLOC_SHIFT_HIGH * sizeof(struct kmem_cache_cpu) Message-ID: References: <202211120436.HzD1i2yQ-lkp@intel.com> <97b6c8e7-36f3-1181-7ffb-d94e8a8d293e@suse.cz> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <97b6c8e7-36f3-1181-7ffb-d94e8a8d293e@suse.cz> ARC-Seal: i=1; s=arc-20220608; d=hostedemail.com; t=1668798511; a=rsa-sha256; cv=none; b=tNOKMFzpOnoz5nD0C0G3j9fkS3ZZo3zNkzGRzpui129iakUkBY0XdJ4N/saWiWGIYVTaoI BdWkp60mdEFyrPqlS6KqJDFUlM1qIGyM2zKv/3YGrdLfrHySpOLa1rDrejOpvD0AlOf4as bJPDVfoz5TMJFzhKGIa61TQ0hwyP+5M= ARC-Authentication-Results: i=1; imf22.hostedemail.com; dkim=none; dmarc=fail reason="SPF not aligned (relaxed), No valid DKIM" header.from=kernel.org (policy=none); spf=pass (imf22.hostedemail.com: domain of dennisszhou@gmail.com designates 209.85.216.43 as permitted sender) smtp.mailfrom=dennisszhou@gmail.com ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=hostedemail.com; s=arc-20220608; t=1668798511; 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: in-reply-to:in-reply-to:references:references; bh=I/8r8Y1eiMtDrea4Bredv83OHEKoOW04VLm26sFfGLI=; b=71OKCE51WCjBJ2aQoZdH7QPoN3G4P149GCMYjc+1Bmaq7eY5Cx96ocqPgs+67t5e6b9WXL 4f3fZWOu331mLh0BPytl31IVyz6/NHsYSbWHSwuJhuAFRhiB7/GQAu1JcXZE5sefqNc1o1 Vved1d5fFxP/uxUaNC+t9+osnXBfQ7Q= X-Stat-Signature: 6f6i7sh8c6w16k8gjharwzpqcip8ef3w X-Rspamd-Queue-Id: 2A1F3C000C Authentication-Results: imf22.hostedemail.com; dkim=none; dmarc=fail reason="SPF not aligned (relaxed), No valid DKIM" header.from=kernel.org (policy=none); spf=pass (imf22.hostedemail.com: domain of dennisszhou@gmail.com designates 209.85.216.43 as permitted sender) smtp.mailfrom=dennisszhou@gmail.com X-Rspam-User: X-Rspamd-Server: rspam10 X-HE-Tag: 1668798510-110121 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: On Fri, Nov 18, 2022 at 10:49:43AM +0100, Vlastimil Babka wrote: > On 11/17/22 20:23, Dennis Zhou wrote: > > On Wed, Nov 16, 2022 at 07:32:03PM +0800, Baoquan He wrote: > >> On 11/15/22 at 12:00pm, Dennis Zhou wrote: > >> > On Tue, Nov 15, 2022 at 05:08:52PM +0800, Baoquan He wrote: > >> > > Hi Dennis, > >> > > > >> > > On 11/14/22 at 08:13pm, Dennis Zhou wrote: > >> > > > Hi Vlastimil & Baoquan, > >> > > > > >> > > > On Mon, Nov 14, 2022 at 06:58:13PM +0100, Vlastimil Babka wrote: > >> > > > > On 11/14/22 08:44, Baoquan He wrote: > >> > > > > > Hi, > >> > > > > > > >> > > > > > I reproduced the build failure according to lkp report and made a patch > >> > > > > > as below to fix it. > >> > > > > > > >> > > > > > From dae7dd9705015ce36db757e88c78802584f949b1 Mon Sep 17 00:00:00 2001 > >> > > > > > From: Baoquan He > >> > > > > > Date: Sun, 13 Nov 2022 18:08:27 +0800 > >> > > > > > Subject: [PATCH] percpu: adjust the value of PERCPU_DYNAMIC_EARLY_SIZE > >> > > > > > Content-type: text/plain > >> > > > > > > >> > > > > > LKP reported a build failure as below on the patch "mm/slub, percpu: > >> > > > > > correct the calculation of early percpu allocation size" > >> > > > > > >> > > > > Since I have that patch in slab.git exposed to -next, should I take this fix > >> > > > > too, to make things simpler? Dennis? > >> > > > > > >> > > > > >> > > > I don't have any problems with you running a fix, but I'm not quite sure > >> > > > this is the right fix. Though this might cause a trivial merge conflict > >> > > > with: d667c94962c1 ("mm/percpu: remove unused PERCPU_DYNAMIC_EARLY_SLOTS") > >> > > > in my percpu#for-6.2 branch. > >> > > > > >> > > > If I'm understanding this correctly, slub requires additional percpu > >> > > > memory due to the use of 64k pages. By increasing > >> > > > PERCPU_DYNAMIC_EARLY_SIZE, we solve the problem for 64k page users, but > >> > > > require a few unnecessary pages that can bloat the size of subsequent > >> > > > percpu chunks. Though, I'm not sure if that's an issue today for > >> > > > embedded devices. > >> > > > >> > > Thanks for looking into this. > >> > > > >> > > I guess you are talking about PERCPU_DYNAMIC_EARLY_SIZE will impact the > >> > > first dynamic chunk size of page first chunk, because the embed first > >> > > chunk will take PERCPU_DYNAMIC_RESERVE. And the impact is done in below > >> > > max() invacation. > >> > > > >> > > static struct pcpu_alloc_info * __init __flatten pcpu_build_alloc_info( > >> > > size_t reserved_size, size_t dyn_size, > >> > > size_t atom_size, > >> > > pcpu_fc_cpu_distance_fn_t cpu_distance_fn) > >> > > { > >> > > ...... > >> > > /* calculate size_sum and ensure dyn_size is enough for early alloc */ > >> > > size_sum = PFN_ALIGN(static_size + reserved_size + > >> > > max_t(size_t, dyn_size, PERCPU_DYNAMIC_EARLY_SIZE)); > >> > > ...... > >> > > } > >> > > > >> > > > > >> > > > I think adding parity to PERCPU_DYNAMIC_EARLY_SIZE with > >> > > > PERCPU_DYNAMIC_RESERVE is defined by BITS_PER_LONG is a safer option > >> > > > here. A small TODO item would be to make PERCPU_DYNAMIC_RESERVE be a + > >> > > > value instead of a max() with PERCPU_DYNAMIC_EARLY_SIZE. > >> > > > >> > > Hmm, the below change may not take power arch into account. Please > >> > > check arch/powerpc/include/asm/page.h, seems the 32bit ppc could have > >> > > 256K pages too. Adding PERCPU_DYNAMIC_EARLY_SIZE to 20K may cost extra > >> > > memory during boot. But th left space of 1st dynamic chunk will join > >> > > the later percpu dynamic allocation, it's not wasted, right? > >> > > > >> > > Not sure if I got your point. > >> > > > >> > > > >> > > >> > Ah, I'm not familiar with all the PAGE_SIZE and word length > >> > combinations. > >> > > >> > The first chunk is smaller in the embedded case with the assumption that > >> > static percpu variables are highly accessed along with the limited > >> > initial allocations. While adding an additional 8KB is not the biggest > >> > deal to the first chunk, this can cause the unit_size for subsequent > >> > chunks to be larger. For example, x86 unit size jumps in powers of 2 due > >> > to alignment and packing against an allocation size of 2MB. So if we're > >> > at say 60KB for the first chunk, subsequent chunks could be 64KB. But > >> > adding 8KB, we'd go from 60KB -> 68KB and a chunk size of 64KB -> 128KB. > >> > >> I could have misunderstanding about the first chunk usage and percpu > >> code. Below is my personal uderstanding about the 1st chunk size and > >> how PERCPU_DYNAMIC_EARLY_SIZE could impact it, please help point out > >> if I am wrong. > >> > >> ~~~ > >> Abstract the definition of them here for reference. > >> /* > >> * Percpu allocator can serve percpu allocations before slab is > >> * initialized which allows slab to depend on the percpu allocator. > >> * The following parameter decide how much resource to preallocate > >> * for this. Keep PERCPU_DYNAMIC_RESERVE equal to or larger than > >> ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ > >> * PERCPU_DYNAMIC_EARLY_SIZE. > >> ~~~~~~~~~~~~~~~~~~~~~ > >> */ > >> #define PERCPU_DYNAMIC_EARLY_SIZE (12 << 10) > >> ...... > >> #if BITS_PER_LONG > 32 > >> #define PERCPU_DYNAMIC_RESERVE (28 << 10) > >> #else > >> #define PERCPU_DYNAMIC_RESERVE (20 << 10) > >> #endif > >> > >> From above definition, we can see that no matter how big > >> PERCPU_DYNAMIC_RESERVE is , it's >= PERCPU_DYNAMIC_EARLY_SIZE as the > >> code comment says. So the max() in pcpu_build_alloc_info() won't impact > >> the embeded 1st chunk at all. > >> > >> So, PERCPU_DYNAMIC_EARLY_SIZE can only impact the page 1st chunk case, > >> namely when calling pcpu_page_first_chunk() to do that. In > >> pcpu_page_first_chunk(), we don't provide dyn_size, so with the help of > >> max(), it will get final dyn_size as PERCPU_DYNAMIC_EARLY_SIZE. This is > >> the only place where PERCPU_DYNAMIC_EARLY_SIZE takes effect on percpu. > >> However, the atom size of page 1st chunk is PAGE_SIZE, it doesn't have > >> the issue of possible bloating unit_size by the atom size, e.g 2M on > >> x86_64. Since pcpu_page_first_chunk() is the fallback of > >> pcpu_embed_first_chunk(), if we decide to provide PERCPU_DYNAMIC_RESERVE > >> as the current value, why we grudge setting it as the smaller value, > >> 20K, whether it's 32bit or 64bit. > >> > > > > I think I might be overindexing on the out of tree modifications here. > > Currently, I think it's clear how modifying PERCPU_DYNAMIC_RESERVE > > affects the system with the lower bound being dictated by > > PERCPU_DYNAMIC_EARLY_SIZE. If we bump PERCPU_DYNAMIC_EARLY_SIZE, it's > > not inherently obvious you can drop that value lower depending on your > > system config. > > > > Ultimately, it is only a few pages, so is saving it that big of a deal > > today? Likely not, just a bit wasteful to potentially orphan a few extra > > pages unnecessarily. > > > > Let's just fix this now and I can massage this in the future if anything > > comes up. I appreciate you taking the time to have this discussion with > > me. > > > > Vlastimil, can you please pick up this fix. > > Sorry, got a bit lost, so do you mean the original uncoditional bump, or the > modification with BITS_PER_LONG > 32 (or PAGE_SHIFT > 12)? > No I've made this more complicated than necessary. Please pick up the original unconditional bump. There's a small chance you'll see a merge conflict in my percpu#for-6.2 tree: d667c94962c1 ("mm/percpu: remove unused PERCPU_DYNAMIC_EARLY_SLOTS") Thanks, Dennis > > Acked-by: Dennis Zhou > > > > Thanks, > > Dennis > > > >> > >> > > >> > If not `BITS_PER_LONG >32`, we could do `PAGE_SHIFT > 12`. > >> > > >> > Thanks, > >> > Dennis > >> > > >> > > > > >> > > > --- > >> > > > diff --git a/include/linux/percpu.h b/include/linux/percpu.h > >> > > > index f1ec5ad1351c..22ce3271eed2 100644 > >> > > > --- a/include/linux/percpu.h > >> > > > +++ b/include/linux/percpu.h > >> > > > @@ -42,7 +42,11 @@ > >> > > > * larger than PERCPU_DYNAMIC_EARLY_SIZE. > >> > > > */ > >> > > > #define PERCPU_DYNAMIC_EARLY_SLOTS 128 > >> > > > +#if BITS_PER_LONG > 32 > >> > > > +#define PERCPU_DYNAMIC_EARLY_SIZE (20 << 10) > >> > > > +#else > >> > > > #define PERCPU_DYNAMIC_EARLY_SIZE (12 << 10) > >> > > > +#endif > >> > > > > >> > > > /* > >> > > > * PERCPU_DYNAMIC_RESERVE indicates the amount of free area to piggy > >> > > > > >> > > > >> > > > >> > > >> > >> > >