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 6CF7AC77B7C for ; Fri, 18 Aug 2023 16:24:13 +0000 (UTC) Received: by kanga.kvack.org (Postfix) id BA86A940065; Fri, 18 Aug 2023 12:24:12 -0400 (EDT) Received: by kanga.kvack.org (Postfix, from userid 40) id B57BA940012; Fri, 18 Aug 2023 12:24:12 -0400 (EDT) X-Delivered-To: int-list-linux-mm@kvack.org Received: by kanga.kvack.org (Postfix, from userid 63042) id 9F8B7940065; Fri, 18 Aug 2023 12:24:12 -0400 (EDT) X-Delivered-To: linux-mm@kvack.org Received: from relay.hostedemail.com (smtprelay0013.hostedemail.com [216.40.44.13]) by kanga.kvack.org (Postfix) with ESMTP id 89702940012 for ; Fri, 18 Aug 2023 12:24:12 -0400 (EDT) Received: from smtpin16.hostedemail.com (a10.router.float.18 [10.200.18.1]) by unirelay09.hostedemail.com (Postfix) with ESMTP id 396C18048E for ; Fri, 18 Aug 2023 16:24:12 +0000 (UTC) X-FDA: 81137747544.16.CB687EA Received: from mail-il1-f173.google.com (mail-il1-f173.google.com [209.85.166.173]) by imf06.hostedemail.com (Postfix) with ESMTP id 4862A180022 for ; Fri, 18 Aug 2023 16:24:10 +0000 (UTC) Authentication-Results: imf06.hostedemail.com; dkim=pass header.d=gmail.com header.s=20221208 header.b=PXxE2sic; spf=pass (imf06.hostedemail.com: domain of nphamcs@gmail.com designates 209.85.166.173 as permitted sender) smtp.mailfrom=nphamcs@gmail.com; dmarc=pass (policy=none) header.from=gmail.com ARC-Seal: i=1; s=arc-20220608; d=hostedemail.com; t=1692375850; a=rsa-sha256; cv=none; b=HrxsVHidxWJOpQc6V+KM43CQAZlQhXmV4IcgiCZal1WGTe9UiWytVC/MlEfTZ734TERkou nbCGWSKnJiVLHjF7HwenL87MNj6v6WJFu0cw1j9bM3nRpYfFsengWOga1/nQly7IcH+NEp cI2/Jk1CcuO9hvburJSBvdF60vcYm7o= ARC-Authentication-Results: i=1; imf06.hostedemail.com; dkim=pass header.d=gmail.com header.s=20221208 header.b=PXxE2sic; spf=pass (imf06.hostedemail.com: domain of nphamcs@gmail.com designates 209.85.166.173 as permitted sender) smtp.mailfrom=nphamcs@gmail.com; dmarc=pass (policy=none) header.from=gmail.com ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=hostedemail.com; s=arc-20220608; t=1692375850; 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=xDWQjXaNJb8KAIE3XBNR1OmBmtzTUxAeMkedUxNPbYY=; b=ipXbaDpwLd0cTTpu3ut6+emEU+HTe+xROx7fSPc8WETzgJeVck0iBa7KUaVW69PrfIhrbR hLeao5t/tEhkGJglKMugjKM7j0ziJcUGAhPCv0F/f4odbEJD4eVeNlqxOZI481d0XotzeB GKnpSs4iDiB0KleleY4roV2IWig7h+o= Received: by mail-il1-f173.google.com with SMTP id e9e14a558f8ab-34aa1572740so3746445ab.0 for ; Fri, 18 Aug 2023 09:24:10 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20221208; t=1692375849; x=1692980649; 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=xDWQjXaNJb8KAIE3XBNR1OmBmtzTUxAeMkedUxNPbYY=; b=PXxE2sicxghfbCd5UJh2GDYifav0XHaVv9I03ECFqbLj3EWz17KtvhPynEOYuc3n1t vSt+e1/8GmgwaSxkbKL2W4keEvlXvM3L3TMKinoMOOmTmjZjcu8mz8s/KAlApWsU9hbW jOFC0CDqGWOnHIAnZWvo0j7CIGl3aT3UjyHGtPqd4VtFOv+9YzP9glPtuMq4kZ6QPktn w3Rrcn3bXSahX+bWGvOpc4CozDxGfq8t8HxZlpOAmqbsg2rXXCTyBjT5+/UhE2jGYsRZ hS/0ytPd4UnZyVm+BGjsw7S68jOXP983HL9En0pZqexhCpICCiJn40Lq/WaBbI7k4eaj txrA== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20221208; t=1692375849; x=1692980649; 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=xDWQjXaNJb8KAIE3XBNR1OmBmtzTUxAeMkedUxNPbYY=; b=h6nT6PZoXTJEUvO1kID0uTdH1Rkcy/Hy+qs+AQgDa3loSeNTDaZsW7fwLpBBUhmDVt v66EgAAcoxQyPPqaG0ObPK25wn4u+oIHyjiiva0tEJsQt86tt8KfcD+50ZNd8PX+8Rwk KYQ5tD+Gzv/hyT76fz9VYQ6TeMWcJ/VtfW1awmaltTHjm8r14ZVT7moDVD+hnGuKqbgg a7RniAmOaYE8a/A9ryatAl+drpf0M/iCFqYBmgWuA7UrG/yY76O17TS00TtjpgdberZf aR0lF2/WyDoUUYVtThTLTaCR81t/gHY9tsamct+qeaFJdH1ASePI4tw0JtTIQO/CeJpy mEmA== X-Gm-Message-State: AOJu0YxryQZEAELRAJRlne4vxr/EVigCEZxy+gsw/78LryEQuZDMVK9G fVnyF8mLZRkggN+g5lXId7xpfQkbyqGcZhg149Q= X-Google-Smtp-Source: AGHT+IFzprIN6kBhXE5UmGSmPLWghIBiKuc7cvG8f1WjG8+H6bu4QjHHM/cq2m+Os1yDQbEeO4r1HJrtYeEVm0nfK0o= X-Received: by 2002:a05:6e02:104c:b0:348:cb93:56b8 with SMTP id p12-20020a056e02104c00b00348cb9356b8mr3371842ilj.11.1692375849152; Fri, 18 Aug 2023 09:24:09 -0700 (PDT) MIME-Version: 1.0 References: <20230817164733.2475092-1-nphamcs@gmail.com> <20230817190126.3155299-1-nphamcs@gmail.com> <20230818134906.GA138967@cmpxchg.org> In-Reply-To: From: Nhat Pham Date: Fri, 18 Aug 2023 09:23:57 -0700 Message-ID: Subject: Re: [PATCH v2] workingset: ensure memcg is valid for recency check To: Yosry Ahmed Cc: Johannes Weiner , Yu Zhao , akpm@linux-foundation.org, kernel-team@meta.com, linux-kernel@vger.kernel.org, linux-mm@kvack.org, stable@vger.kernel.org Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: quoted-printable X-Rspamd-Server: rspam08 X-Rspamd-Queue-Id: 4862A180022 X-Stat-Signature: h5mppwubt4b3gebm6nxw19x9knnk7e6b X-Rspam-User: X-HE-Tag: 1692375850-18643 X-HE-Meta: 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 U+zTPg9K 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 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, Aug 18, 2023 at 7:57 AM Yosry Ahmed wrote: > > On Fri, Aug 18, 2023 at 6:49=E2=80=AFAM Johannes Weiner wrote: > > > > On Thu, Aug 17, 2023 at 05:12:17PM -0600, Yu Zhao wrote: > > > On Thu, Aug 17, 2023 at 4:50=E2=80=AFPM Yosry Ahmed wrote: > > > > > > > > On Thu, Aug 17, 2023 at 3:43=E2=80=AFPM Nhat Pham wrote: > > > > > > > > > > On Thu, Aug 17, 2023 at 1:50 PM Yosry Ahmed wrote: > > > > > > > > > > > > On Thu, Aug 17, 2023 at 12:01=E2=80=AFPM Nhat Pham wrote: > > > > > > > > > > > > > > In eviction recency check, we are currently not holding a loc= al > > > > > > > reference to the memcg that the refaulted folio belonged to w= hen it was > > > > > > > evicted. This could cause serious memcg lifetime issues, for = e.g in the > > > > > > > memcg hierarchy traversal done in mem_cgroup_get_nr_swap_page= s(). This > > > > > > > has occurred in production: > > > > > > > > > > > > > > [ 155757.793456] BUG: kernel NULL pointer dereference, addres= s: 00000000000000c0 > > > > > > > [ 155757.807568] #PF: supervisor read access in kernel mode > > > > > > > [ 155757.818024] #PF: error_code(0x0000) - not-present page > > > > > > > [ 155757.828482] PGD 401f77067 P4D 401f77067 PUD 401f76067 PM= D 0 > > > > > > > [ 155757.839985] Oops: 0000 [#1] SMP > > > > > > > [ 155757.846444] CPU: 7 PID: 1380944 Comm: ThriftSrv-pri3- Kd= ump: loaded Tainted: G S 6.4.3-0_fbk1_rc0_594_g8d0cbcaa67ba= #1 > > > > > > > [ 155757.870808] Hardware name: Wiwynn Twin Lakes MP/Twin Lak= es Passive MP, BIOS YMM16 05/24/2021 > > > > > > > [ 155757.887870] RIP: 0010:mem_cgroup_get_nr_swap_pages+0x3d/= 0xb0 > > > > > > > [ 155757.899377] Code: 29 19 4a 02 48 39 f9 74 63 48 8b 97 c0= 00 00 00 48 8b b7 58 02 00 00 48 2b b7 c0 01 00 00 48 39 f0 48 0f 4d c6 48= 39 d1 74 42 <48> 8b b2 c0 00 00 00 48 8b ba 58 02 00 00 48 2b ba c0 01 00 = 00 48 > > > > > > > [ 155757.937125] RSP: 0018:ffffc9002ecdfbc8 EFLAGS: 00010286 > > > > > > > [ 155757.947755] RAX: 00000000003a3b1c RBX: 000007ffffffffff = RCX: ffff888280183000 > > > > > > > [ 155757.962202] RDX: 0000000000000000 RSI: 0007ffffffffffff = RDI: ffff888bbc2d1000 > > > > > > > [ 155757.976648] RBP: 0000000000000001 R08: 000000000000000b = R09: ffff888ad9cedba0 > > > > > > > [ 155757.991094] R10: ffffea0039c07900 R11: 0000000000000010 = R12: ffff888b23a7b000 > > > > > > > [ 155758.005540] R13: 0000000000000000 R14: ffff888bbc2d1000 = R15: 000007ffffc71354 > > > > > > > [ 155758.019991] FS: 00007f6234c68640(0000) GS:ffff88903f9c0= 000(0000) knlGS:0000000000000000 > > > > > > > [ 155758.036356] CS: 0010 DS: 0000 ES: 0000 CR0: 00000000800= 50033 > > > > > > > [ 155758.048023] CR2: 00000000000000c0 CR3: 0000000a83eb8004 = CR4: 00000000007706e0 > > > > > > > [ 155758.062473] DR0: 0000000000000000 DR1: 0000000000000000 = DR2: 0000000000000000 > > > > > > > [ 155758.076924] DR3: 0000000000000000 DR6: 00000000fffe0ff0 = DR7: 0000000000000400 > > > > > > > [ 155758.091376] PKRU: 55555554 > > > > > > > [ 155758.096957] Call Trace: > > > > > > > [ 155758.102016] > > > > > > > [ 155758.106502] ? __die+0x78/0xc0 > > > > > > > [ 155758.112793] ? page_fault_oops+0x286/0x380 > > > > > > > [ 155758.121175] ? exc_page_fault+0x5d/0x110 > > > > > > > [ 155758.129209] ? asm_exc_page_fault+0x22/0x30 > > > > > > > [ 155758.137763] ? mem_cgroup_get_nr_swap_pages+0x3d/0xb0 > > > > > > > [ 155758.148060] workingset_test_recent+0xda/0x1b0 > > > > > > > [ 155758.157133] workingset_refault+0xca/0x1e0 > > > > > > > [ 155758.165508] filemap_add_folio+0x4d/0x70 > > > > > > > [ 155758.173538] page_cache_ra_unbounded+0xed/0x190 > > > > > > > [ 155758.182919] page_cache_sync_ra+0xd6/0x1e0 > > > > > > > [ 155758.191738] filemap_read+0x68d/0xdf0 > > > > > > > [ 155758.199495] ? mlx5e_napi_poll+0x123/0x940 > > > > > > > [ 155758.207981] ? __napi_schedule+0x55/0x90 > > > > > > > [ 155758.216095] __x64_sys_pread64+0x1d6/0x2c0 > > > > > > > [ 155758.224601] do_syscall_64+0x3d/0x80 > > > > > > > [ 155758.232058] entry_SYSCALL_64_after_hwframe+0x46/0xb0 > > > > > > > [ 155758.242473] RIP: 0033:0x7f62c29153b5 > > > > > > > [ 155758.249938] Code: e8 48 89 75 f0 89 7d f8 48 89 4d e0 e8= b4 e6 f7 ff 41 89 c0 4c 8b 55 e0 48 8b 55 e8 48 8b 75 f0 8b 7d f8 b8 11 00= 00 00 0f 05 <48> 3d 00 f0 ff ff 77 33 44 89 c7 48 89 45 f8 e8 e7 e6 f7 ff = 48 8b > > > > > > > [ 155758.288005] RSP: 002b:00007f6234c5ffd0 EFLAGS: 00000293 = ORIG_RAX: 0000000000000011 > > > > > > > [ 155758.303474] RAX: ffffffffffffffda RBX: 00007f628c4e70c0 = RCX: 00007f62c29153b5 > > > > > > > [ 155758.318075] RDX: 000000000003c041 RSI: 00007f61d2986000 = RDI: 0000000000000076 > > > > > > > [ 155758.332678] RBP: 00007f6234c5fff0 R08: 0000000000000000 = R09: 0000000064d5230c > > > > > > > [ 155758.347452] R10: 000000000027d450 R11: 0000000000000293 = R12: 000000000003c041 > > > > > > > [ 155758.362044] R13: 00007f61d2986000 R14: 00007f629e11b060 = R15: 000000000027d450 > > > > > > > [ 155758.376661] > > > > > > > > > > > > > > This patch fixes the issue by getting a local reference insid= e > > > > > > > unpack_shadow(). > > > > > > > > > > > > > > Fixes: f78dfc7b77d5 ("workingset: fix confusion around evicti= on vs refault container") > > > > > > > > > > > > Beyond mem_cgroup_get_nr_swap_pages(), we still use the evictio= n_memcg > > > > > > without grabbing a ref to it first in workingset_test_recent() = (and in > > > > > > workingset_refault() before that) as well as lru_gen_test_recen= t(). > > > > > > > > > > > > Wouldn't the fix go back even further? or am I misinterpreting = the problem? > > > > > Hmm I don't see eviction_memcg being used outside of *_test_recen= t > > > > > (the rest just uses memcg =3D folio_memcg(folio), which if I'm no= t mistaken is > > > > > the memcg that is refaulting the folio into memory). > > > > > > > > > > Inside workingset_test_recent(), the only other place where evict= ion_memcg > > > > > is used is for mem_cgroup_lruvec. This function call won't crash = whether > > > > > eviction_memcg is valid or not. > > > > > > > > If eviction_memcg is invalid because the memory was already freed, = we > > > > are basically dereferencing garbage in mem_cgroup_lruvec() aren't w= e? > > > > > > > > > The crash only happens during > > > > > mem_cgroup_get_nr_swap_pages, which has an upward traversal from > > > > > eviction_memcg to root. > > > > > > > > > > Let me know if this does not make sense and/or is insufficient to= ensure > > > > > safe upward traversal from eviction_memcg to root! > > > > > > > > > > > > > > > > > > > > > > > > > Signed-off-by: Nhat Pham > > > > > > > Cc: stable@vger.kernel.org > > > > > > > --- > > > > > > > mm/workingset.c | 65 ++++++++++++++++++++++++++++++++-------= ---------- > > > > > > > 1 file changed, 43 insertions(+), 22 deletions(-) > > > > > > > > > > > > > > diff --git a/mm/workingset.c b/mm/workingset.c > > > > > > > index da58a26d0d4d..03cadad4e484 100644 > > > > > > > --- a/mm/workingset.c > > > > > > > +++ b/mm/workingset.c > > > > > > > @@ -206,10 +206,19 @@ static void *pack_shadow(int memcgid, p= g_data_t *pgdat, unsigned long eviction, > > > > > > > return xa_mk_value(eviction); > > > > > > > } > > > > > > > > > > > > > > -static void unpack_shadow(void *shadow, int *memcgidp, pg_da= ta_t **pgdat, > > > > > > > - unsigned long *evictionp, bool *wor= kingsetp) > > > > > > > +/* > > > > > > > + * Unpacks the stored fields of a shadow entry into the give= n pointers. > > > > > > > + * > > > > > > > + * The memcg pointer is only populated if the memcg recorded= in the shadow > > > > > > > + * entry is valid. In this case, a reference to the memcg wi= ll be acquired, > > > > > > > + * and a corresponding mem_cgroup_put() will be needed when = we no longer > > > > > > > + * need the memcg. > > > > > > > + */ > > > > > > > +static void unpack_shadow(void *shadow, struct mem_cgroup **= memcgp, > > > > > > > + pg_data_t **pgdat, unsigned long *evi= ctionp, bool *workingsetp) > > > > > > > { > > > > > > > unsigned long entry =3D xa_to_value(shadow); > > > > > > > + struct mem_cgroup *memcg; > > > > > > > int memcgid, nid; > > > > > > > bool workingset; > > > > > > > > > > > > > > @@ -220,7 +229,24 @@ static void unpack_shadow(void *shadow, = int *memcgidp, pg_data_t **pgdat, > > > > > > > memcgid =3D entry & ((1UL << MEM_CGROUP_ID_SHIFT) - 1= ); > > > > > > > entry >>=3D MEM_CGROUP_ID_SHIFT; > > > > > > > > > > > > > > - *memcgidp =3D memcgid; > > > > > > > + /* > > > > > > > + * Look up the memcg associated with the stored ID. I= t might > > > > > > > + * have been deleted since the folio's eviction. > > > > > > > + * > > > > > > > + * Note that in rare events the ID could have been re= cycled > > > > > > > + * for a new cgroup that refaults a shared folio. Thi= s is > > > > > > > + * impossible to tell from the available data. Howeve= r, this > > > > > > > + * should be a rare and limited disturbance, and acti= vations > > > > > > > + * are always speculative anyway. Ultimately, it's th= e aging > > > > > > > + * algorithm's job to shake out the minimum access fr= equency > > > > > > > + * for the active cache. > > > > > > > + */ > > > > > > > + memcg =3D mem_cgroup_from_id(memcgid); > > > > > > > + if (memcg && css_tryget(&memcg->css)) > > > > > > > + *memcgp =3D memcg; > > > > > > > + else > > > > > > > + *memcgp =3D NULL; > > > > > > > + > > > > > > > *pgdat =3D NODE_DATA(nid); > > > > > > > *evictionp =3D entry; > > > > > > > *workingsetp =3D workingset; > > > > > > > @@ -262,15 +288,16 @@ static void *lru_gen_eviction(struct fo= lio *folio) > > > > > > > static bool lru_gen_test_recent(void *shadow, bool file, str= uct lruvec **lruvec, > > > > > > > unsigned long *token, bool *w= orkingset) > > > > > > > { > > > > > > > - int memcg_id; > > > > > > > unsigned long min_seq; > > > > > > > struct mem_cgroup *memcg; > > > > > > > struct pglist_data *pgdat; > > > > > > > > > > > > > > - unpack_shadow(shadow, &memcg_id, &pgdat, token, worki= ngset); > > > > > > > + unpack_shadow(shadow, &memcg, &pgdat, token, workings= et); > > > > > > > + if (!mem_cgroup_disabled() && !memcg) > > > > > > > + return false; > > > > > > > > > > > > +Yu Zhao > > > > > > > > > > > > There is a change of behavior here, right? > > > > > > > > > > > > The existing code will continue if !mem_cgroup_disabled() && !m= emcg is > > > > > > true, and mem_cgroup_lruvec() will return the lruvec of the roo= t > > > > > > memcg. Now we are just returning false. > > > > > > > > > > > > Is this intentional? > > > > > Oh right, there is. Should have cc-ed Yu Zhao as well, my bad. > > > > > get_maintainers.pl isn't always sufficient I guess :) > > > > > > > > > > But yeah, this behavioral change is intentional. > > > > > > > > > > Correct me if I'm wrong of course, but it seems like MGLRU should > > > > > follow the same pattern here. That is, once we return from unpack= _shadow, > > > > > the possible scenarios are the same as prescribed in workingset_t= est_recent: > > > > > > > > > > 1. If mem_cgroup is disabled, we can ignore this check. > > > > > 2. If mem_cgroup is enabled, then the only reason why we get NULL > > > > > memcg from unpack_shadow is if the eviction_memcg is no longer > > > > > valid. We should not try to get its lruvec, or substitute it wit= h the > > > > > root memcg, but return false right away (i.e not recent). > > > > > > > > > > > > > > I will leave this for Yu :) > > > > > > Thanks, Yosry. > > > > > > Hi Nhat, it seems unnecessary to me to introduce a get/put into > > > lru_gen_test_recent() because it doesn't suffer from the bug this > > > patch tries to fix. In theory, the extra get/put can impact > > > performance, though admittedly the impact is unlikely to be > > > measurable. Regardless, the general practice is to fix the bug > > > locally, i.e., when the mem_cgroup_get_nr_swap_pages() path is taken, > > > rather than change the unrelated path. Thank you. > > > > Hey guys, > > > > I had suggested to have it in unpack_shadow() to keep things simple, > > and not further complicate the lifetime rules in this code. The > > tryget() is against a per-cpu counter, so it's not expensive. > > > > The NULL deref is evidence that while *some* cgroup members are still > > accessible once it's dead, not all of it is. There is no explicit > > guarantee from the cgroup code that anything BUT the tryget() is still > > valid against group that is under rcu freeing. > > > > Since it isn't expensive, let's keep it simple and robust, and prevent > > future bugs of the same class, by always ensuring the cgroup is alive > > before accessing random members. Especially in non-cgroup code. > > I looked at this again today with fresh eyes, and I want to go back to > what I initially said. Isn't RCU protection in this case enough to > keep the memcg "valid" (i.e accessible, not garbage)? The tryget is > not a lot of complexity or performance tax, but I want to really > understand what's happening here. > > Looking at the code again, this seems to be the sequence of events on > the cgroup side: > - css_put() puts the last reference invoking a call to css_release() > - css_release() queues css_release_work_fn() > - css_release() does some bookkeeping, makes some callbacks, and > queues css_free_rwork_fn() to run *after* an RCU grace period. > - css_free_rwork_fn() makes callbacks to free the memory, ultimately > freeing the memcg. > > On the memcg idr side, the removal sequence of events seem to be: > - mem_cgroup_id_put() will decrement the id ref and check if falls to 0 > - If the id ref falls to 0, we call mem_cgroup_id_remove() *then* css_put= () > > On the workingset_refault() side, the sequence of events seems to be: > - rcu_read_lock() > - memcg =3D mem_cgroup_from_id() > - ... // use memcg > - rcu_read_unlock() > > So technically, after holding the rcu read lock, if we find the memcg > in the idr, it must be valid, and it must not be freed until after the > rcu read section is completed. It's not just the cgroup internal > implementation, it's the contract between cgroup core and controllers > such as memcg. > > The memory controller expects a sequence of callbacks during freeing: > css_offline() -> css_released() -> css_free(). So memcg code is within > its right to access any fields of struct mem_cgroup that are not freed > by the css_offline() or css_released() until css_free() is called, > right? > > > Here is a guess / question, because I am not really familiar with > memory barriers and such, but is it at all possible that the actual > problem is reordering of instructions in mem_cgroup_id_put_many(), > such that we actually execute css_put() *before* > mem_cgroup_id_remove()? > > If this happens it seems possible for this to happen: > > cpu #1 cpu#2 > css_put() > /* css_free_rwork_fn is queu= ed */ > rcu_read_lock() > mem_cgroup_from_id() > mem_cgroup_id_remove() > /* access memcg */ > > If I understand correctly, if css_free_rwork_fn() is queued before the > rcu_read_lock in workingset_refault() begins, then it can be executed > during the rcu read section, and the memcg can be freed at any point > from under us. Perhaps what we need is memory barriers to ensure > correct ordering in mem_cgroup_id_put_many()? I am not sure if > rcu_read_lock() implies a barrier on the other side. > > Sorry if this is all off, I am just trying to understand what's going on. Ah that is wild. That does sound plausible. In this case, maybe something like this? mem_cgroup_id_remove(memcg); /* * Preventing css_put from happening before id removal due to * instruction reordering. * * This guarantees that if a non-null memcg is acquired from ID within * an RCU read section, its css won't be freed for the * duration of this section. */ smp_mb(); /* Memcg ID pins CSS */ css_put(&memcg->css);