linux-kernel - Re: [PATCH] rcu: Remove redundant rcu_read_lock/unlock() in spin

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <10adb9b7-ea19-407b-818e-79061a067e13@redhat.com>
Date: Sat, 13 Sep 2025 13:23:03 -0400
From: Waiman Long <llong@...hat.com>
To: Hillf Danton <hdanton@...a.com>, Waiman Long <llong@...hat.com>
Cc: Sebastian Andrzej Siewior <bigeasy@...utronix.de>,
 pengdonglin <dolinux.peng@...il.com>, linux-kernel@...r.kernel.org,
 "Paul E . McKenney" <paulmck@...nel.org>
Subject: Re: [PATCH] rcu: Remove redundant rcu_read_lock/unlock() in spin_lock
 critical sections

On 9/13/25 4:00 AM, Hillf Danton wrote:
> On Fri, 12 Sep 2025 20:33:31 -0400 Waiman Long wrote:
>> On 9/12/25 5:35 PM, Sebastian Andrzej Siewior wrote:
>>> On 2025-09-12 17:13:09 [-0400], Waiman Long wrote:
>>>> On 9/12/25 2:50 AM, pengdonglin wrote:
>>>>> From: pengdonglin <pengdonglin@...omi.com>
>>>>>
>>>>> When CONFIG_PREEMPT_RT is disabled, spin_lock*() operations 
>>>>> implicitly
>>>>> disable preemption, which provides RCU read-side protection. When
>>>>> CONFIG_PREEMPT_RT is enabled, spin_lock*() implementations internally
>>>>> manage RCU read-side critical sections.
>>>> I have some doubt about your claim that disabling preemption 
>>>> provides RCU
>>>> read-side protection. It is true for some flavors but probably not 
>>>> all. I do
>>>> know that disabling interrupt will provide RCU read-side 
>>>> protection. So for
>>>> spin_lock_irq*() calls, that is valid. I am not sure about 
>>>> spin_lock_bh(),
>>>> maybe it applies there too. we need some RCU people to confirm.
>>> The claim is valid since Paul merged the three flavours we had. Before
>>> that preempt_disable() (and disabling irqs) would match
>>> rcu_read_lock_sched(). rcu_read_lock() and rcu_read_lock_bh() were
>>> different in terms of grace period and clean up.
>>> So _now_ we could remove it if it makes things easier.
>>
>> Thanks for the clarification.
>>
>> In this case, I think the patch description should mention the commit 
>> that unify the 3 RCU flavors to make sure that this patch won't be 
>> accidentally backport'ed to an older kernel without the necessary 
>> prerequisite commit(s).
>
> This change also affects the dereference helpers.
>
> #define rcu_dereference_check(p, c) \
>     __rcu_dereference_check((p), __UNIQUE_ID(rcu), \
>                 (c) || rcu_read_lock_held(), __rcu)
>
Right, this macro will need to be updated as well to avoid false 
positive if we decide that preempt_disabled region is a valid 
rcu_read_lock critical section.

Cheers,
Longman