Received-SPF: pass (google.com: domain of linux-kernel+bounces-24219-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45e3:2400::1 as permitted sender) client-ip=2604:1380:45e3:2400::1;
Message-ID: <4a3031ff-49e5-4e95-914c-aa034ebfd8a7@joelfernandes.org>
Date: Thu, 11 Jan 2024 20:51:36 -0500
Precedence: bulk
MIME-Version: 1.0
User-Agent: Mozilla Thunderbird
Subject: Re: [PATCH] rcu/nocb: Fix WARN_ON_ONCE() in the
 rcu_nocb_bypass_lock()
Content-Language: en-US
To: paulmck@kernel.org, Z qiang <qiang.zhang1211@gmail.com>
Cc: frederic@kernel.org, quic_neeraju@quicinc.com, rcu@vger.kernel.org,
 linux-kernel@vger.kernel.org
References: <20240110081128.18683-1-qiang.zhang1211@gmail.com>
 <CALm+0cXQF0nS6HD5iZ1RhbWo0rg8adwvcHrsS8W04K1MxY51bA@mail.gmail.com>
 <06bb8028-a1be-431c-9bff-265777d64e27@paulmck-laptop>
From: Joel Fernandes <joel@joelfernandes.org>
In-Reply-To: <06bb8028-a1be-431c-9bff-265777d64e27@paulmck-laptop>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 7bit


On 1/11/2024 6:54 AM, Paul E. McKenney wrote:
> On Wed, Jan 10, 2024 at 04:36:46PM +0800, Z qiang wrote:
>>>
>>> For the kernels built with CONFIG_RCU_NOCB_CPU_DEFAULT_ALL=y and
>>> CONFIG_RCU_LAZY=y, here are the following scenarios that will trigger
>>> WARN_ON_ONCE() in the rcu_nocb_bypass_lock() and rcu_nocb_wait_contended().
>>>
>>>         CPU2                                               CPU11
>>> kthread
>>> rcu_nocb_cb_kthread                                       ksys_write
>>> rcu_do_batch                                              vfs_write
>>> rcu_torture_timer_cb                                      proc_sys_write
>>> __kmem_cache_free                                         proc_sys_call_handler
>>> kmemleak_free                                             drop_caches_sysctl_handler
>>> delete_object_full                                        drop_slab
>>> __delete_object                                           shrink_slab
>>> put_object                                                lazy_rcu_shrink_scan
>>> call_rcu                                                  rcu_nocb_flush_bypass
>>> __call_rcu_commn                                            rcu_nocb_bypass_lock
>>>                                                             raw_spin_trylock(&rdp->nocb_bypass_lock) fail
>>>                                                             atomic_inc(&rdp->nocb_lock_contended);
>>> rcu_nocb_wait_contended                                     WARN_ON_ONCE(smp_processor_id() != rdp->cpu);
>>>  WARN_ON_ONCE(atomic_read(&rdp->nocb_lock_contended))                                          |
>>>                             |_ _ _ _ _ _ _ _ _ _same rdp and rdp->cpu != 11_ _ _ _ _ _ _ _ _ __|
>>>
>>> This commit therefore use the rcu_nocb_try_flush_bypass() instead of
>>> rcu_nocb_flush_bypass() in lazy_rcu_shrink_scan(), if the nocb_bypass
>>> queue is being flushed, the rcu_nocb_try_flush_bypass will return directly.
>>>
>>> Signed-off-by: Zqiang <qiang.zhang1211@gmail.com>
> 
> Just to make sure I understand, the "echo" command called out below
> will trigger the two-CPU scenario called out above in kernels built with
> CONFIG_RCU_NOCB_CPU_DEFAULT_ALL=y and CONFIG_RCU_LAZY=y, correct?
> 
> If so, good catch!
> 
> Any objections to this patch?  Or to put it another way, is there a
> better fix via adjusting lazy RCU?

I think it is a good find and no objections to this patch. One thing we could
also do is have rcu_nocb_try_flush_bypass() return false if the
 trylock fails, and then retry till lock is available. That would give us
roughly the same behavior as rcu_nocb_flush_bypass() but I am not sure if it is
worth it, because the shrinker will just try again if memory pressure is not
relieved anyway.

Reviewed-by: Joel Fernandes (Google) <joel@joelfernandes.org>

thanks,

 - Joel


> 
> 							Thanx, Paul
> 
>>> ---
>>
>> During rcutorture testing, use echo 3 > /proc/sys/vm/drop_caches will trigger:
>>
>> [ 52.674359] WARNING: CPU: 11 PID: 505 at kernel/rcu/tree_nocb.h:104
>> rcu_nocb_bypass_lock+0xc7/0xd0
>> [ 52.674388] Modules linked in:
>> [ 52.674406] CPU: 11 PID: 505 Comm: sh Not tainted 6.6.0-rt14zqiang-dirty #103
>> [ 52.674422] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS
>> rel-1.16.2-0-gea1b7a073390-prebuilt.qemu.org 04/01/2014
>> [ 52.674436] RIP: 0010:rcu_nocb_bypass_lock+0xc7/0xd0
>> [ 52.674454] Code: 4c 89 e7 e8 4b 3d 6a 01 be 04 00 00 00 4c 89 ef e8
>> ce 92 31 00 f0 ff 8b 68 02 00 00 5b 41 5c 41 5d 41 5e 5d c3 cc cc cc
>> cc 90 <0f> 0b 90 eb d1 0f 1f 40 00 90 90 0
>> [ 52.674467] RSP: 0018:ffff88800af6fa68 EFLAGS: 00010093
>> [ 52.674487] RAX: 0000000000000000 RBX: ffff888069e0f540 RCX: ffffffffb5c12d44
>> [ 52.674497] RDX: 0000000000000003 RSI: dffffc0000000000 RDI: ffff888069e0fb10
>> [ 52.674508] RBP: ffff88800af6fa88 R08: ffffed100d3c1ef6 R09: ffffed100d3c1ef6
>> [ 52.674518] R10: ffffed100d3c1ef5 R11: ffff888069e0f7ab R12: ffff888069e0f8c0
>> [ 52.674529] R13: ffff888069e0f7a8 R14: 000000000000000b R15: ffff88800af6fb90
>> [ 52.674540] FS: 00007ff543132740(0000) GS:ffff88806c000000(0000)
>> knlGS:0000000000000000
>> [ 52.674555] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>> [ 52.674565] CR2: 000055df13465004 CR3: 00000000027f2000 CR4: 00000000001506e0
>> [ 52.674576] Call Trace:
>> [ 52.674583] <TASK>
>> [ 52.674598] ? show_regs+0x66/0x70
>> [ 52.674627] ? __warn+0xae/0x220
>> [ 52.674657] ? rcu_nocb_bypass_lock+0xc7/0xd0
>> [ 52.674693] ? report_bug+0x14a/0x240
>> [ 52.674756] ------------[ cut here ]------------
>> [ 52.674766] WARNING: CPU: 2 PID: 118 at kernel/rcu/tree_nocb.h:124
>> __call_rcu_common+0xd3f/0xd80
>> [ 52.674785] Modules linked in:
>> [ 52.674785] ? handle_bug+0x44/0x80
>> [ 52.674795] CPU: 2 PID: 118 Comm: rcuop/10 Not tainted
>> 6.6.0-rt14zqiang-dirty #103
>> [ 52.674806] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS
>> rel-1.16.2-0-gea1b7a073390-prebuilt.qemu.org 04/01/2014
>> [ 52.674812] RIP: 0010:__call_rcu_common+0xd3f/0xd80
>> [ 52.674816] ? exc_invalid_op+0x1c/0x50
>> [ 52.674822] Code: 9e 02 4c 89 e2 e8 a1 46 ff ff e9 d0 fa ff ff 48 8d
>> 7b 18 e8 43 5f 30 00 48 8b 7b 18 48 89 de e8 67 9a ff ff e9 69 f6 ff
>> ff 90 <0f> 0b 90 f3 90 e9 17 fa ff ff 90 b
>> [ 52.674832] RSP: 0018:ffff888002277af8 EFLAGS: 00010002
>> [ 52.674847] RAX: 0000000000000001 RBX: ffff888069e0f540 RCX: ffffffffb5c24655
>> [ 52.674848] ? asm_exc_invalid_op+0x1f/0x30
>> [ 52.674857] RDX: 0000000000000003 RSI: dffffc0000000000 RDI: ffff888069e0f7a8
>> [ 52.674867] RBP: ffff888002277bf0 R08: ffffed100d3c1ef6 R09: ffffed100d3c1ef6
>> [ 52.674877] R10: ffffed100d3c1ef5 R11: ffff888069e0f7ab R12: ffff88800bab62f8
>> [ 52.674887] R13: ffff888069e0f7a8 R14: 0000000000000000 R15: ffff888069e0f918
>> [ 52.674897] FS: 0000000000000000(0000) GS:ffff888069c00000(0000)
>> knlGS:0000000000000000
>> [ 52.674910] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
>> [ 52.674920] CR2: 000055df13419911 CR3: 00000000027f2000 CR4: 00000000001506e0
>> [ 52.674930] Call Trace:
>> [ 52.674936] <TASK>
>> [ 52.674934] ? rcu_nocb_bypass_lock+0x94/0xd0
>> [ 52.674950] ? show_regs+0x66/0x70
>> [ 52.674972] ? rcu_nocb_bypass_lock+0xc7/0xd0
>> [ 52.674974] ? __warn+0xae/0x220
>> [ 52.675002] ? __call_rcu_common+0xd3f/0xd80
>> [ 52.675010] ? rcu_nocb_bypass_lock+0x94/0xd0
>> [ 52.675037] ? report_bug+0x14a/0x240
>> [ 52.675054] rcu_nocb_flush_bypass+0x3a/0x60
>> [ 52.675094] ? handle_bug+0x44/0x80
>> [ 52.675095] lazy_rcu_shrink_scan+0x12e/0x220
>> [ 52.675114] ? exc_invalid_op+0x1c/0x50
>> [ 52.675139] ? asm_exc_invalid_op+0x1f/0x30
>> [ 52.675156] shrink_slab.constprop.116+0x2cd/0x6e0
>> [ 52.675206] ? __call_rcu_common+0x775/0xd80
>> [ 52.675239] ? __call_rcu_common+0xd3f/0xd80
>> [ 52.675259] ? __pfx_shrink_slab.constprop.116+0x10/0x10
>> [ 52.675272] ? __call_rcu_common+0x775/0xd80
>> [ 52.675335] ? __pfx_drop_pagecache_sb+0x10/0x10
>> [ 52.675365] ? __pfx___call_rcu_common+0x10/0x10
>> [ 52.675386] ? preempt_schedule+0x7f/0xa0
>> [ 52.675396] drop_slab+0x64/0x90
>> [ 52.675413] ? preempt_schedule_thunk+0x1a/0x30
>> [ 52.675426] drop_caches_sysctl_handler+0x82/0xe0
>> [ 52.675472] call_rcu+0x17/0x20
>> [ 52.675489] put_object+0x53/0x70
>> [ 52.675513] __delete_object+0x73/0x90
>> [ 52.675545] delete_object_full+0x1f/0x30
>> [ 52.675563] kmemleak_free+0x41/0x70
>> [ 52.675586] __kmem_cache_free+0x1bd/0x230
>> [ 52.675598] ? rcu_torture_timer_cb+0x12/0x20
>> [ 52.675623] ? rcu_do_batch+0x466/0xf50
>> [ 52.675649] kfree+0x90/0x110
>> [ 52.675661] ? __pfx_rcu_torture_timer_cb+0x10/0x10
>> [ 52.675678] rcu_torture_timer_cb+0x12/0x20
>> [ 52.675697] rcu_do_batch+0x46b/0xf50
>> [ 52.675464] proc_sys_call_handler+0x247/0x310
>> [ 52.675772] ? __pfx_rcu_do_batch+0x10/0x10
>> [ 52.675785] ? migrate_disable+0x2a/0xf0
>> [ 52.675815] ? lockdep_softirqs_off+0x13d/0x200
>> [ 52.675854] ? rcu_nocb_cb_kthread+0x29c/0x880
>> [ 52.675889] rcu_nocb_cb_kthread+0x2b1/0x880
>> [ 52.675910] ? __pfx_proc_sys_call_handler+0x10/0x10
>> [ 52.675945] ? vfs_write+0x3ea/0x7c0
>> [ 52.675958] ? vfs_write+0x3ea/0x7c0
>> [ 52.675959] ? __pfx_rcu_nocb_cb_kthread+0x10/0x10
>> [ 52.675974] ? trace_preempt_on+0x54/0xe0
>> [ 52.675990] ? __kthread_parkme+0x80/0x110
>> [ 52.676015] ? preempt_count_sub+0x50/0x80
>> [ 52.676031] proc_sys_write+0x17/0x20
>> [ 52.676050] vfs_write+0x58b/0x7c0
>> [ 52.676063] ? __kthread_parkme+0xf2/0x110
>> [ 52.676111] ? __pfx_rcu_nocb_cb_kthread+0x10/0x10
>> [ 52.676112] ? __pfx_vfs_write+0x10/0x10
>> [ 52.676139] kthread+0x1a8/0x1f0
>> [ 52.676161] ? kthread+0x107/0x1f0
>> [ 52.676163] ? __might_fault+0x84/0xd0
>> [ 52.676183] ? __pfx_kthread+0x10/0x10
>> [ 52.676197] ? __might_fault+0xbe/0xd0
>> [ 52.676213] ? __might_fault+0x84/0xd0
>> [ 52.676223] ret_from_fork+0x40/0x60
>> [ 52.676238] ? __pfx_kthread+0x10/0x10
>> [ 52.676272] ? __fget_light+0xb8/0x120
>> [ 52.676273] ret_from_fork_asm+0x1b/0x30
>> [ 52.676355] ksys_write+0xd0/0x170
>> [ 52.676386] ? __pfx_ksys_write+0x10/0x10
>> [ 52.676418] </TASK>
>> [ 52.676425] irq event stamp: 591689
>> [ 52.676433] hardirqs last enabled at (591688): [<ffffffffb72b7193>]
>> _raw_spin_unlock_irqrestore+0x63/0x80
>> [ 52.676452] hardirqs last disabled at (591689): [<ffffffffb5c242f3>]
>> __call_rcu_common+0x413/0xd80
>> [ 52.676467] softirqs last enabled at (591668): [<ffffffffb5af3489>]
>> __local_bh_enable_ip+0x109/0x160
>> [ 52.676486] softirqs last disabled at (591672): [<ffffffffb5c1ed2d>]
>> rcu_do_batch+0x5ad/0xf50
>> [ 52.676488] __x64_sys_write+0x47/0x60
>> [ 52.676500] ---[ end trace 0000000000000000 ]---
>> root@qemux86-64:~# [ 52.676526] do_syscall_64+0x47/0x90
>> [ 52.676552] entry_SYSCALL_64_after_hwframe+0x6f/0xd9
>> [ 52.676566] RIP: 0033:0x7ff5432260c4
>> [ 52.676583] Code: 15 59 7d 0d 00 f7 d8 64 89 02 48 c7 c0 ff ff ff ff
>> eb b7 0f 1f 00 f3 0f 1e fa 80 3d 1d 0d 0e 00 00 74 13 b8 01 00 00 00
>> 0f 05 <48> 3d 00 f0 ff ff 77 54 c3 0f 1f 8
>> [ 52.676596] RSP: 002b:00007ffcc9614998 EFLAGS: 00000202 ORIG_RAX:
>> 0000000000000001
>> [ 52.676613] RAX: ffffffffffffffda RBX: 0000000000000002 RCX: 00007ff5432260c4
>> [ 52.676624] RDX: 0000000000000002 RSI: 000055df1345df40 RDI: 0000000000000001
>> [ 52.676634] RBP: 000055df1345df40 R08: 000055df1345b700 R09: 0000000000000000
>> [ 52.676644] R10: 00000000000001b6 R11: 0000000000000202 R12: 0000000000000001
>> [ 52.676653] R13: 00007ff5431326c8 R14: 00007ffcc9614a00 R15: 000055df1345b6f8
>> [ 52.676789] </TASK>
>> [ 52.676796] irq event stamp: 502888
>> [ 52.676804] hardirqs last enabled at (502887): [<ffffffffb72b7193>]
>> _raw_spin_unlock_irqrestore+0x63/0x80
>> [ 52.676823] hardirqs last disabled at (502888): [<ffffffffb5c19753>]
>> lazy_rcu_shrink_scan+0x1d3/0x220
>> [ 52.676838] softirqs last enabled at (54708): [<ffffffffb5af3489>]
>> __local_bh_enable_ip+0x109/0x160
>> [ 52.676854] softirqs last disabled at (54698): [<ffffffffb6fd861a>]
>> unix_release_sock+0x26a/0x7c0
>> [ 52.676888] ---[ end trace 0000000000000000 ]---
>>
>> Thanks
>> Zqiang
>>
>>
>>>  kernel/rcu/tree_nocb.h | 2 +-
>>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>>
>>> diff --git a/kernel/rcu/tree_nocb.h b/kernel/rcu/tree_nocb.h
>>> index d82f96a66600..9b618842c324 100644
>>> --- a/kernel/rcu/tree_nocb.h
>>> +++ b/kernel/rcu/tree_nocb.h
>>> @@ -1381,7 +1381,7 @@ lazy_rcu_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
>>>                         rcu_nocb_unlock_irqrestore(rdp, flags);
>>>                         continue;
>>>                 }
>>> -               WARN_ON_ONCE(!rcu_nocb_flush_bypass(rdp, NULL, jiffies, false));
>>> +               rcu_nocb_try_flush_bypass(rdp, jiffies);
>>>                 rcu_nocb_unlock_irqrestore(rdp, flags);
>>>                 wake_nocb_gp(rdp, false);
>>>                 sc->nr_to_scan -= _count;
>>> --
>>> 2.17.1
>>>
>