| 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
| ||
|
Message-ID: <711fe3bf-0ac2-4ae9-9dda-97ba047eb64f@paulmck-laptop>
Date: Sun, 31 Aug 2025 18:06:03 -0700
From: "Paul E. McKenney" <paulmck@...nel.org>
To: Zqiang <qiang.zhang@...ux.dev>
Cc: kernel test robot <oliver.sang@...el.com>, oe-lkp@...ts.linux.dev,
lkp@...el.com, Andrii Nakryiko <andrii@...nel.org>,
Alexei Starovoitov <ast@...nel.org>,
Peter Zijlstra <peterz@...radead.org>, rcu@...r.kernel.org,
linux-kernel@...r.kernel.org
Subject: Re: [paulmckrcu:dev.2025.08.21a] [rcu] 8bd9383727:
WARNING:possible_circular_locking_dependency_detected
On Sun, Aug 31, 2025 at 11:52:40PM +0000, Zqiang wrote:
> >
> > On Sun, Aug 31, 2025 at 02:22:56AM +0000, Zqiang wrote:
> >
> > >
> > > On Sat, Aug 30, 2025 at 02:38:35AM +0000, Zqiang wrote:
> > >
> > > >
> > > > On Tue, Aug 26, 2025 at 04:47:22PM +0800, kernel test robot wrote:
> > > >
> > > > >
> > > > > hi, Paul,
> > > > >
> > > > > the similar issue still exists on this dev.2025.08.21a branch.
> > > > > again, if the issue is already fixed on later branches, please just ignore.
> > > > > thanks
> > > > >
> > > > >
> > > > > Hello,
> > > > >
> > > > > kernel test robot noticed "WARNING:possible_circular_locking_dependency_detected" on:
> > > > >
> > > > > commit: 8bd9383727068a5a18acfecefbdfa44a7d6bd838 ("rcu: Re-implement RCU Tasks Trace in terms of SRCU-fast")
> > > > > https://github.com/paulmckrcu/linux dev.2025.08.21a
> > > > >
> > > > > in testcase: rcutorture
> > > > > version:
> > > > > with following parameters:
> > > > >
> > > > > runtime: 300s
> > > > > test: default
> > > > > torture_type: tasks-tracing
> > > > >
> > > > >
> > > > >
> > > > > config: x86_64-randconfig-003-20250824
> > > > > compiler: clang-20
> > > > > test machine: qemu-system-x86_64 -enable-kvm -cpu SandyBridge -smp 2 -m 16G
> > > > >
> > > > > (please refer to attached dmesg/kmsg for entire log/backtrace)
> > > > >
> > > > Again, apologies for being slow, and thank you for your testing efforts.
> > > >
> > > > Idiot here forgot about Tiny SRCU, so please see the end of this email
> > > > for an alleged fix. Does it do the trick for you?
> > > >
> > > > Thanx, Paul
> > > >
> > > > >
> > > > > If you fix the issue in a separate patch/commit (i.e. not just a new version of
> > > > > the same patch/commit), kindly add following tags
> > > > > | Reported-by: kernel test robot <oliver.sang@...el.com>
> > > > > | Closes: https://lore.kernel.org/oe-lkp/202508261642.b15eefbb-lkp@intel.com
> > > > >
> > > > >
> > > > > [ 42.365933][ T393] WARNING: possible circular locking dependency detected
> > > > > [ 42.366428][ T393] 6.17.0-rc1-00035-g8bd938372706 #1 Tainted: G T
> > > > > [ 42.366985][ T393] ------------------------------------------------------
> > > > > [ 42.367490][ T393] rcu_torture_rea/393 is trying to acquire lock:
> > > > > [ 42.367952][ T393] ffffffffad41dc88 (rcu_tasks_trace_srcu_struct.srcu_wq.lock){....}-{2:2}, at: swake_up_one (kernel/sched/swait.c:52 (discriminator 1))
> > > > > [ 42.368775][ T393]
> > > > > [ 42.368775][ T393] but task is already holding lock:
> > > > > [ 42.369278][ T393] ffff88813d1ff2e8 (&p->pi_lock){-.-.}-{2:2}, at: rcutorture_one_extend (kernel/rcu/rcutorture.c:?) rcutorture
> > > > > [ 42.370043][ T393]
> > > > > [ 42.370043][ T393] which lock already depends on the new lock.
> > > > > [ 42.370043][ T393]
> > > > > [ 42.370755][ T393]
> > > > > [ 42.370755][ T393] the existing dependency chain (in reverse order) is:
> > > > > [ 42.371388][ T393]
> > > > > [ 42.371388][ T393] -> #1 (&p->pi_lock){-.-.}-{2:2}:
> > > > > [ 42.371903][ T393] _raw_spin_lock_irqsave (include/linux/spinlock_api_smp.h:110 kernel/locking/spinlock.c:162)
> > > > > [ 42.372309][ T393] try_to_wake_up (include/linux/spinlock.h:557 (discriminator 1) kernel/sched/core.c:4216 (discriminator 1))
> > > > > [ 42.372669][ T393] swake_up_locked (include/linux/list.h:111)
> > > > > [ 42.373029][ T393] swake_up_one (kernel/sched/swait.c:54 (discriminator 1))
> > > > > [ 42.373380][ T393] tasks_tracing_torture_read_unlock (include/linux/srcu.h:408 (discriminator 1) include/linux/rcupdate_trace.h:81 (discriminator 1) kernel/rcu/rcutorture.c:1112 (discriminator 1)) rcutorture
> > > > > [ 42.373952][ T393] rcutorture_one_extend (kernel/rcu/rcutorture.c:2141) rcutorture
> > > > > [ 42.374452][ T393] rcu_torture_one_read_end (kernel/rcu/rcutorture.c:2357) rcutorture
> > > > > [ 42.374976][ T393] rcu_torture_one_read (kernel/rcu/rcutorture.c:?) rcutorture
> > > > > [ 42.375460][ T393] rcu_torture_reader (kernel/rcu/rcutorture.c:2443) rcutorture
> > > > > [ 42.375920][ T393] kthread (kernel/kthread.c:465)
> > > > > [ 42.376241][ T393] ret_from_fork (arch/x86/kernel/process.c:154)
> > > > > [ 42.376603][ T393] ret_from_fork_asm (arch/x86/entry/entry_64.S:255)
> > > > > [ 42.376973][ T393]
> > > > > [ 42.376973][ T393] -> #0 (rcu_tasks_trace_srcu_struct.srcu_wq.lock){....}-{2:2}:
> > > > > [ 42.377657][ T393] __lock_acquire (kernel/locking/lockdep.c:3166)
> > > > > [ 42.378031][ T393] lock_acquire (kernel/locking/lockdep.c:5868)
> > > > > [ 42.378378][ T393] _raw_spin_lock_irqsave (include/linux/spinlock_api_smp.h:110 kernel/locking/spinlock.c:162)
> > > > > [ 42.378794][ T393] swake_up_one (kernel/sched/swait.c:52 (discriminator 1))
> > > > > [ 42.379152][ T393] tasks_tracing_torture_read_unlock (include/linux/srcu.h:408 (discriminator 1) include/linux/rcupdate_trace.h:81 (discriminator 1) kernel/rcu/rcutorture.c:1112 (discriminator 1)) rcutorture
> > > > > [ 42.379714][ T393] rcutorture_one_extend (kernel/rcu/rcutorture.c:2141) rcutorture
> > > > > [ 42.380217][ T393] rcu_torture_one_read_end (kernel/rcu/rcutorture.c:2357) rcutorture
> > > > > [ 42.380731][ T393] rcu_torture_one_read (kernel/rcu/rcutorture.c:?) rcutorture
> > > > > [ 42.381220][ T393] rcu_torture_reader (kernel/rcu/rcutorture.c:2443) rcutorture
> > > > > [ 42.381714][ T393] kthread (kernel/kthread.c:465)
> > > > > [ 42.382060][ T393] ret_from_fork (arch/x86/kernel/process.c:154)
> > > > > [ 42.382420][ T393] ret_from_fork_asm (arch/x86/entry/entry_64.S:255)
> > > > > [ 42.382796][ T393]
> > > > > [ 42.382796][ T393] other info that might help us debug this:
> > > > > [ 42.382796][ T393]
> > > > > [ 42.383515][ T393] Possible unsafe locking scenario:
> > > > > [ 42.383515][ T393]
> > > > > [ 42.384052][ T393] CPU0 CPU1
> > > > > [ 42.384428][ T393] ---- ----
> > > > > [ 42.384799][ T393] lock(&p->pi_lock);
> > > > > [ 42.385083][ T393] lock(rcu_tasks_trace_srcu_struct.srcu_wq.lock);
> > > > > [ 42.385707][ T393] lock(&p->pi_lock);
> > > > > [ 42.386180][ T393] lock(rcu_tasks_trace_srcu_struct.srcu_wq.lock);
> > > > > [ 42.386663][ T393]
> > > > > [ 42.386663][ T393] *** DEADLOCK ***
> > > > > [ 42.386663][ T393]
> > > > > [ 42.387236][ T393] 1 lock held by rcu_torture_rea/393:
> > > > > [ 42.387626][ T393] #0: ffff88813d1ff2e8 (&p->pi_lock){-.-.}-{2:2}, at: rcutorture_one_extend (kernel/rcu/rcutorture.c:?) rcutorture
> > > > > [ 42.388419][ T393]
> > > > > [ 42.388419][ T393] stack backtrace:
> > > > > [ 42.388852][ T393] CPU: 0 UID: 0 PID: 393 Comm: rcu_torture_rea Tainted: G T 6.17.0-rc1-00035-g8bd938372706 #1 PREEMPT(full)
> > > > > [ 42.389758][ T393] Tainted: [T]=RANDSTRUCT
> > > > > [ 42.390057][ T393] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.16.2-debian-1.16.2-1 04/01/2014
> > > > > [ 42.390786][ T393] Call Trace:
> > > > > [ 42.391020][ T393] <TASK>
> > > > > [ 42.391225][ T393] dump_stack_lvl (lib/dump_stack.c:123 (discriminator 2))
> > > > > [ 42.391544][ T393] print_circular_bug (kernel/locking/lockdep.c:2045)
> > > > > [ 42.391898][ T393] check_noncircular (kernel/locking/lockdep.c:?)
> > > > > [ 42.392242][ T393] __lock_acquire (kernel/locking/lockdep.c:3166)
> > > > > [ 42.392594][ T393] ? __schedule (kernel/sched/sched.h:1531 (discriminator 1) kernel/sched/core.c:6969 (discriminator 1))
> > > > > [ 42.392930][ T393] ? lock_release (kernel/locking/lockdep.c:470 (discriminator 3))
> > > > > [ 42.393272][ T393] ? swake_up_one (kernel/sched/swait.c:52 (discriminator 1))
> > > > > [ 42.393610][ T393] lock_acquire (kernel/locking/lockdep.c:5868)
> > > > > [ 42.393930][ T393] ? swake_up_one (kernel/sched/swait.c:52 (discriminator 1))
> > > > > [ 42.394264][ T393] _raw_spin_lock_irqsave (include/linux/spinlock_api_smp.h:110 kernel/locking/spinlock.c:162)
> > > > > [ 42.394640][ T393] ? swake_up_one (kernel/sched/swait.c:52 (discriminator 1))
> > > > > [ 42.394969][ T393] swake_up_one (kernel/sched/swait.c:52 (discriminator 1))
> > > > > [ 42.395281][ T393] tasks_tracing_torture_read_unlock (include/linux/srcu.h:408 (discriminator 1) include/linux/rcupdate_trace.h:81 (discriminator 1) kernel/rcu/rcutorture.c:1112 (discriminator 1)) rcutorture
> > > > > [ 42.395814][ T393] rcutorture_one_extend (kernel/rcu/rcutorture.c:2141) rcutorture
> > > > > [ 42.396276][ T393] rcu_torture_one_read_end (kernel/rcu/rcutorture.c:2357) rcutorture
> > > > > [ 42.396756][ T393] rcu_torture_one_read (kernel/rcu/rcutorture.c:?) rcutorture
> > > > > [ 42.397219][ T393] ? __cfi_rcu_torture_reader (kernel/rcu/rcutorture.c:2426) rcutorture
> > > > > [ 42.397690][ T393] rcu_torture_reader (kernel/rcu/rcutorture.c:2443) rcutorture
> > > > > [ 42.398126][ T393] ? __cfi_rcu_torture_timer (kernel/rcu/rcutorture.c:2405) rcutorture
> > > > > [ 42.398565][ T393] kthread (kernel/kthread.c:465)
> > > > > [ 42.398857][ T393] ? __cfi_kthread (kernel/kthread.c:412)
> > > > > [ 42.399169][ T393] ret_from_fork (arch/x86/kernel/process.c:154)
> > > > > [ 42.399491][ T393] ? __cfi_kthread (kernel/kthread.c:412)
> > > > > [ 42.399815][ T393] ret_from_fork_asm (arch/x86/entry/entry_64.S:255)
> > > > > [ 42.400151][ T393] </TASK>
> > > > >
> > > > >
> > > > > The kernel config and materials to reproduce are available at:
> > > > > https://download.01.org/0day-ci/archive/20250826/202508261642.b15eefbb-lkp@intel.com
> > > > >
> > > > >
> > > > >
> > > > > --
> > > > > 0-DAY CI Kernel Test Service
> > > > > https://github.com/intel/lkp-tests/wiki
> > > > >
> > > > ------------------------------------------------------------------------
> > > >
> > > > diff --git a/kernel/rcu/srcutiny.c b/kernel/rcu/srcutiny.c
> > > > index 6e9fe2ce1075d5..db63378f062051 100644
> > > > --- a/kernel/rcu/srcutiny.c
> > > > +++ b/kernel/rcu/srcutiny.c
> > > > @@ -106,7 +106,7 @@ void __srcu_read_unlock(struct srcu_struct *ssp, int idx)
> > > > newval = READ_ONCE(ssp->srcu_lock_nesting[idx]) - 1;
> > > > WRITE_ONCE(ssp->srcu_lock_nesting[idx], newval);
> > > > preempt_enable();
> > > > - if (!newval && READ_ONCE(ssp->srcu_gp_waiting) && in_task())
> > > > + if (!newval && READ_ONCE(ssp->srcu_gp_waiting) && in_task() && !irqs_disabled())
> > > >
> > > >
> > > > The fllowing case may exist:
> > > >
> > > >
> > > > CPU0
> > > >
> > > > task1:
> > > > __srcu_read_lock()
> > > >
> > > For mainline kernels, here we must have blocked, correct?
> > >
> > > In -rcu, there is of course:
> > >
> > > 740cda2fe1a9 ("EXP srcu: Enable Tiny SRCU On all CONFIG_SMP=n kernels")
> > >
> > > And this means that in -rcu kernels built with CONFIG_PREEMPT_NONE=y,
> > > we could be preempted.
> > >
> > > And maybe this is a reason to drop this commit. Or...
> > >
> > >
> > > For tiny srcu, even if the preempt schedule not happend in
> > > srcu read ctrical section, we can still do voluntary
> > > scheduling in srcu_read ctrical section, this case is
> > > also still happend.
> > >
> > > >
> > > > ....
> > > >
> > > >
> > > > task2 preempt run:
> > > >
> > > > srcu_drive_gp()
> > > > ->swait_event_exclusive()
> > > >
> > > >
> > > > ....
> > > > task1 continue run:
> > > > ....
> > > > raw_spin_lock_irqsave
> > > > __srcu_read_unlock()
> > > > ->find all previours condition are met
> > > > but the irqs_disable() return true,
> > > > not invoke swake_up_one().
> > > >
> > > > task2 maybe always hung.
> > > >
> > > The bug that kernel test robot reported existed for a long time.
> > > The offending commit simply introduced the use case that exercised
> > > this bug. So we do need a fix.
> > >
> > > One approach would be to impose a rule like we used to have for RCU,
> > > namely that if interrupts were disabled across srcu_read_unlock(),
> > > then they must have been disabled since the matching srcu_read_lock().
> > > Another would be to make the current swait_event_exclusive() in
> > > srcu_drive_gp() instead be a loop around wait_event_timeout_exclusive()
> > > that checks ssp->srcu_lock_nesting[].
> > >
> > > But is there a better way?
> > >
> > > I think the second approach is enough :)
> > >
> > Hmmm... OK, how about the incremental patch below?
> >
> > Thanx, Paul
> >
> > ------------------------------------------------------------------------
> >
> > commit a543d73eeaa491021040a02bdf0e8a9148b5c186
> > Author: Paul E. McKenney <paulmck@...nel.org>
> > Date: Sun Aug 31 09:38:44 2025 -0700
> >
> > squash! rcu: Re-implement RCU Tasks Trace in terms of SRCU-fast
> >
> > [ paulmck: Apply Zqiang feedback. ]
> >
> > Signed-off-by: Paul E. McKenney <paulmck@...nel.org>
>
> Should the previous fix that added irqs_disabled() also be
> added to this patch? or can we use preemptible() instead of
> in_tasks() && irqs_disabled()?
Yes, both this fix and the addition of the !irqw_disabled() check
are to be squashed into the original commit:
f56bf5dd7ffc ("rcu: Re-implement RCU Tasks Trace in terms of SRCU-fast")
But I do not immediately see how checking preemptible() would help,
especially in (say) CONFIG_PREEMPT_NONE=y kernels in which this function
always returns zero. What am I missing here?
Thanx, Paul
> Thanks
> Zqiang
>
>
> >
> > diff --git a/kernel/rcu/srcutiny.c b/kernel/rcu/srcutiny.c
> > index db63378f062051..b52ec45698e85b 100644
> > --- a/kernel/rcu/srcutiny.c
> > +++ b/kernel/rcu/srcutiny.c
> > @@ -113,8 +113,8 @@ EXPORT_SYMBOL_GPL(__srcu_read_unlock);
> >
> > /*
> > * Workqueue handler to drive one grace period and invoke any callbacks
> > - * that become ready as a result. Single-CPU and !PREEMPTION operation
> > - * means that we get away with murder on synchronization. ;-)
> > + * that become ready as a result. Single-CPU operation and preemption
> > + * disabling mean that we get away with murder on synchronization. ;-)
> > */
> > void srcu_drive_gp(struct work_struct *wp)
> > {
> > @@ -141,7 +141,12 @@ void srcu_drive_gp(struct work_struct *wp)
> > WRITE_ONCE(ssp->srcu_idx, ssp->srcu_idx + 1);
> > WRITE_ONCE(ssp->srcu_gp_waiting, true); /* srcu_read_unlock() wakes! */
> > preempt_enable();
> > - swait_event_exclusive(ssp->srcu_wq, !READ_ONCE(ssp->srcu_lock_nesting[idx]));
> > + do {
> > + // Deadlock issues prevent __srcu_read_unlock() from
> > + // doing an unconditional wakeup, so polling is required.
> > + swait_event_timeout_exclusive(ssp->srcu_wq,
> > + !READ_ONCE(ssp->srcu_lock_nesting[idx]), HZ / 10);
> > + } while (READ_ONCE(ssp->srcu_lock_nesting[idx]));
> > preempt_disable(); // Needed for PREEMPT_LAZY
> > WRITE_ONCE(ssp->srcu_gp_waiting, false); /* srcu_read_unlock() cheap. */
> > WRITE_ONCE(ssp->srcu_idx, ssp->srcu_idx + 1);
> >
Powered by blists - more mailing lists