From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from smtp.kernel.org (aws-us-west-2-korg-mail-1.web.codeaurora.org [10.30.226.201]) (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 26A303502B3 for ; Wed, 25 Mar 2026 07:36:52 +0000 (UTC) Authentication-Results: smtp.subspace.kernel.org; arc=none smtp.client-ip=10.30.226.201 ARC-Seal:i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1774424212; cv=none; b=hknsWaEadnKVsP2W8AQTRydwDe737ThzOQl7Po9jzlZE7r8FVbARO5cPyVZseeCX8jAoG+8RHGacM0DyMzX4kxmQSHRTNHHCM7ZRgid1INACqz6i0L/5yzKtTvu2DgE0pjtMx0UUgfvUdhsKOiVNcSu2YEBySzG3KPLAsUE3Sxg= ARC-Message-Signature:i=1; a=rsa-sha256; d=subspace.kernel.org; s=arc-20240116; t=1774424212; c=relaxed/simple; bh=opMqCnHb5NP/n4eovyOLUHlmdDCFdJVAToF0fcD3gVU=; h=Date:From:To:Cc:Subject:Message-ID:References:MIME-Version: Content-Type:Content-Disposition:In-Reply-To; b=jT+0Lt5YaMYcjrAO3OnWWT4ym/PKZy33SANR+0tdyXzoyCFFsniK86jXLiECZSkiNZlBguFhdYoR/noH99XP82LAg90eCD62nn7XuAuniGh9Rl3h27WI6V+oxy0hyHvLH3g6288MXTe/hfVwp1kR79nvaUhjYaHuNU8FV70usZ4= ARC-Authentication-Results:i=1; smtp.subspace.kernel.org; dkim=pass (2048-bit key) header.d=kernel.org header.i=@kernel.org header.b=VoZyamcp; arc=none smtp.client-ip=10.30.226.201 Authentication-Results: smtp.subspace.kernel.org; dkim=pass (2048-bit key) header.d=kernel.org header.i=@kernel.org header.b="VoZyamcp" Received: by smtp.kernel.org (Postfix) with ESMTPSA id 96FD5C4CEF7; Wed, 25 Mar 2026 07:36:51 +0000 (UTC) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=kernel.org; s=k20201202; t=1774424212; bh=opMqCnHb5NP/n4eovyOLUHlmdDCFdJVAToF0fcD3gVU=; h=Date:From:To:Cc:Subject:References:In-Reply-To:From; b=VoZyamcpvTqhe3Hs60wKIY2dMLK/Juv5gUS2KsvWJ26bDipnCR/28YSrriFXpdOnV 2Q6+eQjiMFsiPwQVGdVd3P1RrxUDm0d6brZygx10ZEmYjHszYinD7Kzqvhnl5Kb1Zj Cgz4ZC4yEc/3XrF1oC375hv3BbOcs7eKPpiUhFnahkQxOZO5SfpzEpaa7TpilK4Mj4 r8HsoT53C90bpXJCF+X18aLcr+xxsBj3YSVe+MNMPbwffDAlvZY3inGlOMi2z9Pjxa GTS3wRIgLjp2m1BVojuBBieD7ev+EwBHO5l3MlwLq8QFqyoeSHzQij4J5BG3qGr7Mz mJ2yCyP7W6nHA== Date: Wed, 25 Mar 2026 16:36:49 +0900 From: "Harry Yoo (Oracle)" To: Qi Zheng Cc: hannes@cmpxchg.org, hughd@google.com, mhocko@suse.com, roman.gushchin@linux.dev, shakeel.butt@linux.dev, muchun.song@linux.dev, david@kernel.org, lorenzo.stoakes@oracle.com, ziy@nvidia.com, yosry.ahmed@linux.dev, imran.f.khan@oracle.com, kamalesh.babulal@oracle.com, axelrasmussen@google.com, yuanchu@google.com, weixugc@google.com, chenridong@huaweicloud.com, mkoutny@suse.com, akpm@linux-foundation.org, hamzamahfooz@linux.microsoft.com, apais@linux.microsoft.com, lance.yang@linux.dev, bhe@redhat.com, usamaarif642@gmail.com, linux-mm@kvack.org, linux-kernel@vger.kernel.org, Qi Zheng Subject: Re: [PATCH 2/3] mm: memcontrol: correct the parameter type of __mod_memcg{_lruvec}_state() Message-ID: References: <90524ca3806e24105ab5f2d69435f57c2ae034cb.1774342371.git.zhengqi.arch@bytedance.com> <5a9d9152-8e23-4438-a322-ec524fd159a6@linux.dev> Precedence: bulk X-Mailing-List: linux-kernel@vger.kernel.org List-Id: List-Subscribe: List-Unsubscribe: MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: On Wed, Mar 25, 2026 at 03:26:46PM +0800, Qi Zheng wrote: > > > On 3/25/26 1:17 PM, Harry Yoo (Oracle) wrote: > > On Wed, Mar 25, 2026 at 11:25:06AM +0800, Qi Zheng wrote: > > > On 3/25/26 9:43 AM, Harry Yoo (Oracle) wrote: > > > > On Tue, Mar 24, 2026 at 07:31:28PM +0800, Qi Zheng wrote: > > > > > From: Qi Zheng > > > > > > > > > > The __mod_memcg_state() and __mod_memcg_lruvec_state() were used to > > > > > reparent non-hierarchical stats, the values passed to them might exceed > > > > > the upper limit of the type int, so correct the val parameter type of them > > > > > to long. > > > > > > > > > > Signed-off-by: Qi Zheng > > > > > --- > > > > > include/trace/events/memcg.h | 10 +++++----- > > > > > mm/memcontrol.c | 8 ++++---- > > > > > 2 files changed, 9 insertions(+), 9 deletions(-) > > > > > > > > > > diff --git a/mm/memcontrol.c b/mm/memcontrol.c > > > > > index 7fb9cbc10dfbb..4a78550f6174e 100644 > > > > > --- a/mm/memcontrol.c > > > > > +++ b/mm/memcontrol.c > > > > > @@ -527,7 +527,7 @@ unsigned long lruvec_page_state_local(struct lruvec *lruvec, > > > > > #ifdef CONFIG_MEMCG_V1 > > > > > static void __mod_memcg_lruvec_state(struct mem_cgroup_per_node *pn, > > > > > - enum node_stat_item idx, int val); > > > > > + enum node_stat_item idx, long val); > > > > > void reparent_memcg_lruvec_state_local(struct mem_cgroup *memcg, > > > > > struct mem_cgroup *parent, int idx) > > > > > @@ -784,7 +784,7 @@ static int memcg_page_state_unit(int item); > > > > > * Normalize the value passed into memcg_rstat_updated() to be in pages. Round > > > > > * up non-zero sub-page updates to 1 page as zero page updates are ignored. > > > > > */ > > > > > -static int memcg_state_val_in_pages(int idx, int val) > > > > > +static long memcg_state_val_in_pages(int idx, long val) > > > > > { > > > > > int unit = memcg_page_state_unit(idx); > > > > > > > > Sashiko AI made an interesting argument [1] that this could lead to > > > > incorrectly returning a very large positive number. Let me verify that. > > > > > > > > [1] https://sashiko.dev/#/patchset/cover.1774342371.git.zhengqi.arch%40bytedance.com > > > > > > > > Sashiko wrote: > > > > > Does this change inadvertently break the handling of negative byte-sized > > > > > updates? > > > > > Looking at the rest of the function: > > > > > if (!val || unit == PAGE_SIZE) > > > > > return val; > > > > > else > > > > > return max(val * unit / PAGE_SIZE, 1UL); > > > > > > > > > PAGE_SIZE is defined as an unsigned long. > > > > > > > > Right, it's defined as 1UL << PAGE_SHIFT. > > > > > > > > > When val is negative, such as during uncharging of byte-sized stats like > > > > > MEMCG_ZSWAP_B, the expression val * unit is a negative long. > > > > > > > > Right. > > > > > > > > > Dividing a signed long by an unsigned long causes the signed long to be > > > > > promoted to unsigned before division, > > > > > > > > Right. > > > > > > > > > resulting in a massive positive > > > > > number instead of a small negative one. > > > > > > > > Let's look at an example (assuming unit is 1). > > > > > > > > val = val * unit = -16384 (-16 KiB) > > > > val * unit / PAGE_SIZE = 0xFFFFFFFFFFFFC000 / PAGE_SIZE = 0x3FFFFFFFFFFFFF > > > > max(0x3FFFFFFFFFFFFF, 1UL) = 0x3FFFFFFFFFF > > > > > > > > Yeah, that's a massive positive number. > > > > > > > > Hmm but how did it work when it was int? > > > > > > > > val = val * unit = -16384 (-16KiB) > > > > val * unit / PAGE_SIZE = 0xFFFFFFFFFFFFC000 / PAGE_SIZE = 0x3FFFFFFFFFFFFF > > > > max(val * unit / PAGE_SIZE, 1UL) = 0x3FFFFFFFFFFFFF > > > > (int)0x3FFFFFFFFFFFFF = 0xFFFFFFFF = (-1) > > > > > > > > That's incorrect. It should have been -4? > > > > > > > > > Before this change, the function returned an int, which implicitly truncated > > > > > the massive unsigned 64-bit result to a 32-bit int, accidentally yielding the > > > > > correct negative arithmetic value. > > > > > > > > So... "accidentally yielding the correct negative arithemetic value" > > > > is wrong. > > > > > > > > Sounds like it's been subtly broken even before this patch and nobody > > > > noticed. > > > > > > Thank you for such a detailed analysis! And I think you are right. > > > > No problem ;) > > > > > The memcg_state_val_in_pages() is only to make @val to be in pages, so > > > perhaps we can avoid the above problem by taking the absolute value > > > first? > > For more clearly, here is the diff: > > diff --git a/mm/memcontrol.c b/mm/memcontrol.c > index 6491ca74b3398..2b34805b01476 100644 > --- a/mm/memcontrol.c > +++ b/mm/memcontrol.c > @@ -787,11 +787,17 @@ static int memcg_page_state_unit(int item); > static long memcg_state_val_in_pages(int idx, long val) > { > int unit = memcg_page_state_unit(idx); > + unsigned long uval; > + long res; > > if (!val || unit == PAGE_SIZE) > return val; > - else > - return max(val * unit / PAGE_SIZE, 1UL); > + > + uval = val < 0 ? -(unsigned long)val : (unsigned long)val; We could use abs() macro in linux/math.h here :) val should not be LONG_MIN, but it should be fine to assume it won't reach LONG_MIN. > + > + res = max(mult_frac(uval, unit, PAGE_SIZE), 1UL); > + > + return val < 0 ? -res : res; > } > > #ifdef CONFIG_MEMCG_V1 > > > That would work for memcg_rstat_updated(), > > but not for trace_mod_memcg_state()? > > Why? The trace_mod_memcg_state() seems to accept 'long' type, > am I missing something? Nevermind. Without the diff, I misunderstood what you meant. Thanks! -- Cheers, Harry / Hyeonggon