On Thu, Sep 10, 2020 at 02:14:32PM +1000, Stephen Rothwell wrote:
> Hi all,
>
> On Mon, 7 Sep 2020 10:55:47 +0200 Anders Roxell <anders.roxell@xxxxxxxxxx> wrote:
> >
> > On Thu, 3 Sep 2020 at 18:14, Paul E. McKenney <paulmck@xxxxxxxxxx> wrote:
> > >
> > > On Thu, Sep 03, 2020 at 10:39:10AM +0200, Anders Roxell wrote:
> > > > Hi Paul,
> > > >
> > > > On Sat, 29 Aug 2020 at 00:59, Paul E. McKenney <paulmck@xxxxxxxxxx> wrote:
> > > > >
> > > > > On Fri, Aug 28, 2020 at 09:24:19PM +0200, Anders Roxell wrote:
> > > > > > On Fri, 28 Aug 2020 at 15:29, Paul E. McKenney <paulmck@xxxxxxxxxx> wrote:
> > > > > > >
> > > > > > > On Fri, Aug 28, 2020 at 09:37:17AM +0200, Anders Roxell wrote:
> > > > > > > > On Wed, 26 Aug 2020 at 21:39, Paul E. McKenney <paulmck@xxxxxxxxxx> wrote:
> > > > > > > > >
> > > > > > > > > On Wed, Aug 26, 2020 at 08:19:01PM +0200, Anders Roxell wrote:
> > > > > > > > > > On Wed, 26 Aug 2020 at 08:33, Stephen Rothwell <sfr@xxxxxxxxxxxxxxxx> wrote:
> > > > > > > > >
> > > > > > > > > [ . . . ]
> > > > > > > > >
> > > > > > > > > > I've built and run an arm64 allmodconfig kernel where I use the
> > > > > > > > > > defconfig as the base, I do this for testing purposes.
> > > > > > > > > > I can see the following call trace [1]:
> > > > > > > > > >
> > > > > > > > > > [ 2595.811453][ T1] Running tests on all trace events:
> > > > > > > > > > [ 2595.860933][ T1] Testing all events:
> > > > > > > > > > [ 4316.066072][ T8] kworker/dying (8) used greatest stack depth:
> > > > > > > > > > 27056 bytes left
> > > > > > > > > > [ 8561.924871][ C0] watchdog: BUG: soft lockup - CPU#0 stuck for
> > > > > > > > > > 22s! [migration/0:14]
> > > > > > > > > > [ 8561.934498][ C0] Modules linked in:
> > > > > > > > > > [ 8561.942303][ C0] irq event stamp: 4044
> > > > > > > > > > [ 8561.949044][ C0] hardirqs last enabled at (4043):
> > > > > > > > > > [<ffffa000126b530c>] _raw_spin_unlock_irqrestore+0xac/0x138
> > > > > > > > > > [ 8561.960848][ C0] hardirqs last disabled at (4044):
> > > > > > > > > > [<ffffa000126a89f8>] __schedule+0xf8/0x7e0
> > > > > > > > > > [ 8561.971418][ C0] softirqs last enabled at (3698):
> > > > > > > > > > [<ffffa00010001b04>] __do_softirq+0x524/0x5f8
> > > > > > > > > > [ 8561.982191][ C0] softirqs last disabled at (3689):
> > > > > > > > > > [<ffffa000101216c8>] __irq_exit_rcu+0x128/0x1a0
> > > > > > > > > > [ 8561.993068][ C0] CPU: 0 PID: 14 Comm: migration/0 Tainted: G
> > > > > > > > > > W 5.9.0-rc2-next-20200826-00005-g24628bb4c0bf #1
> > > > > > > > > > [ 8562.005684][ C0] Hardware name: linux,dummy-virt (DT)
> > > > > > > > > > [ 8562.013247][ C0] pstate: 80400005 (Nzcv daif +PAN -UAO BTYPE=--)
> > > > > > > > > > [ 8562.021657][ C0] pc : arch_local_irq_enable+0x58/0x80
> > > > > > > > > > [ 8562.029323][ C0] lr : _raw_spin_unlock_irq+0x84/0xc0
> > > > > > > > > > [ 8562.036739][ C0] sp : ffff0000698efaa0
> > > > > > > > > > [ 8562.042984][ C0] x29: ffff0000698efaa0 x28: ffff00006ad0f270
> > > > > > > > > > [ 8562.053814][ C0] x27: ffff00006ad0f248 x26: ffff0000698d4718
> > > > > > > > > > [ 8562.064687][ C0] x25: ffff00006ad0e798 x24: ffffa000139e3a40
> > > > > > > > > > [ 8562.075506][ C0] x23: 0000000000000001 x22: ffffa000154f5000
> > > > > > > > > > [ 8562.086425][ C0] x21: ffff00006ad0e798 x20: ffff00006ad0e780
> > > > > > > > > > [ 8562.097255][ C0] x19: ffffa000126a905c x18: 00000000000014c0
> > > > > > > > > > [ 8562.108071][ C0] x17: 0000000000001500 x16: 0000000000001440
> > > > > > > > > > [ 8562.118918][ C0] x15: 00000000f1f1f1f1 x14: 003d090000000000
> > > > > > > > > > [ 8562.129739][ C0] x13: 00003d0900000000 x12: ffff80000d31df41
> > > > > > > > > > [ 8562.140544][ C0] x11: 1fffe0000d31df40 x10: ffff80000d31df40
> > > > > > > > > > [ 8562.151366][ C0] x9 : dfffa00000000000 x8 : ffff0000698efa07
> > > > > > > > > > [ 8562.162247][ C0] x7 : 0000000000000001 x6 : 00007ffff2ce20c0
> > > > > > > > > > [ 8562.173072][ C0] x5 : ffff0000698d4040 x4 : dfffa00000000000
> > > > > > > > > > [ 8562.183954][ C0] x3 : ffffa0001040f904 x2 : 0000000000000007
> > > > > > > > > > [ 8562.194811][ C0] x1 : ffffa00014080000 x0 : 00000000000000e0
> > > > > > > > > > [ 8562.205858][ C0] Call trace:
> > > > > > > > > > [ 8562.211739][ C0] arch_local_irq_enable+0x58/0x80
> > > > > > > > > > [ 8562.219076][ C0] _raw_spin_unlock_irq+0x84/0xc0
> > > > > > > > > > [ 8562.226394][ C0] __schedule+0x75c/0x7e0
> > > > > > > > > > [ 8562.233074][ C0] preempt_schedule_notrace+0x64/0xc0
> > > > > > > > > > [ 8562.268210][ C0] ftrace_ops_list_func+0x494/0x4e0
> > > > > > > > > > [ 8562.275735][ C0] ftrace_graph_call+0x0/0x4
> > > > > > > > > > [ 8562.282647][ C0] preempt_count_add+0xc/0x240
> > > > > > > > > > [ 8562.289686][ C0] schedule+0xe4/0x160
> > > > > > > > > > [ 8562.296187][ C0] smpboot_thread_fn+0x47c/0x540
> > > > > > > > > > [ 8562.303377][ C0] kthread+0x23c/0x260
> > > > > > > > > > [ 8562.309906][ C0] ret_from_fork+0x10/0x18
> > > > > > > > > > [ 8562.316604][ C0] Kernel panic - not syncing: softlockup: hung tasks
> > > > > > > > > > [ 8562.325230][ C0] CPU: 0 PID: 14 Comm: migration/0 Tainted: G
> > > > > > > > > > W L 5.9.0-rc2-next-20200826-00005-g24628bb4c0bf #1
> > > > > > > > > > [ 8562.337861][ C0] Hardware name: linux,dummy-virt (DT)
> > > > > > > > > > [ 8562.345374][ C0] Call trace:
> > > > > > > > > > [ 8562.351228][ C0] dump_backtrace+0x0/0x320
> > > > > > > > > > [ 8562.358070][ C0] show_stack+0x38/0x60
> > > > > > > > > > [ 8562.364728][ C0] dump_stack+0x1c0/0x280
> > > > > > > > > > [ 8562.371447][ C0] panic+0x32c/0x614
> > > > > > > > > > [ 8562.377868][ C0] watchdog_timer_fn+0x49c/0x560
> > > > > > > > > > [ 8562.385076][ C0] __run_hrtimer+0x1cc/0x360
> > > > > > > > > > [ 8562.392021][ C0] __hrtimer_run_queues+0x1a0/0x220
> > > > > > > > > > [ 8562.399500][ C0] hrtimer_interrupt+0x1f8/0x440
> > > > > > > > > > [ 8562.406807][ C0] arch_timer_handler_virt+0x68/0xa0
> > > > > > > > > > [ 8562.414338][ C0] handle_percpu_devid_irq+0x118/0x2a0
> > > > > > > > > > [ 8562.421992][ C0] __handle_domain_irq+0x150/0x1c0
> > > > > > > > > > [ 8562.429315][ C0] gic_handle_irq+0x98/0x120
> > > > > > > > > > [ 8562.436297][ C0] el1_irq+0xd4/0x1c0
> > > > > > > > >
> > > > > > > > > We appear to have taken an interrupt here, just after releasing
> > > > > > > > > an irq-disabled lock and enabling interrupts.
> > > > > > > > >
> > > > > > > > > > [ 8562.442748][ C0] arch_local_irq_enable+0x58/0x80
> > > > > > > > > > [ 8562.450116][ C0] _raw_spin_unlock_irq+0x84/0xc0
> > > > > > > > > > [ 8562.457360][ C0] __schedule+0x75c/0x7e0
> > > > > > > > > > [ 8562.464142][ C0] preempt_schedule_notrace+0x64/0xc0
> > > > > > > > > > [ 8562.471745][ C0] ftrace_ops_list_func+0x494/0x4e0
> > > > > > > > > > [ 8562.479195][ C0] ftrace_graph_call+0x0/0x4
> > > > > > > > > > [ 8562.486159][ C0] preempt_count_add+0xc/0x240
> > > > > > > > > > [ 8562.493210][ C0] schedule+0xe4/0x160
> > > > > > > > >
> > > > > > > > > We are trying to sleep, so we took a pass through the scheduler and
> > > > > > > > > did some tracing.
> > > > > > > > >
> > > > > > > > > > [ 8562.499737][ C0] smpboot_thread_fn+0x47c/0x540
> > > > > > > > >
> > > > > > > > > Here we might be bringing up a CPU? Except that according to your
> > > > > > > > > dmesg, there is only one CPU. ("RCU restricting CPUs from NR_CPUS=256
> > > > > > > > > to nr_cpu_ids=1"). So this seems unlikely.
> > > > > > > > >
> > > > > > > > > Huh. The first dmesg in output-next-20200826.log is instead a
> > > > > > > > > DEBUG_LOCKS_WARN_ON() from lockdep. Might this be related? Or do you
> > > > > > > > > also see this warning when boot happens quickly?
> > > > > > > > >
> > > > > > > > > (I do see the soft lockup later on.)
> > > > > > > > >
> > > > > > > > > > [ 8562.506960][ C0] kthread+0x23c/0x260
> > > > > > > > > > [ 8562.513496][ C0] ret_from_fork+0x10/0x18
> > > > > > > > > > [ 8562.521052][ C0] Kernel Offset: disabled
> > > > > > > > > > [ 8562.527725][ C0] CPU features: 0x0240002,20002004
> > > > > > > > > > [ 8562.534950][ C0] Memory Limit: none
> > > > > > > > > > [ 8562.543830][ C0] ---[ end Kernel panic - not syncing:
> > > > > > > > > > softlockup: hung tasks ]---
> > > > > > > > > >
> > > > > > > > > > When I tested to checkout and build e6df9766894d ("Merge remote-tracking branch
> > > > > > > > > > 'irqchip/irq/irqchip-next' into master") that was merged into today's
> > > > > > > > > > tag I was able to
> > > > > > > > > > boot [2]. When I checked out 25e6e115dd1d ("Merge remote-tracking
> > > > > > > > > > branch 'rcu/rcu/next' into master") and built and tested that I was
> > > > > > > > > > able to boot [3] after a
> > > > > > > > > > looong time. To clarify, the rcu merge comes after the irqchip merge.
> > > > > > > > > >
> > > > > > > > > > I also tried to only revert the rcu tree from todays next tag like this:
> > > > > > > > > > "git diff e6df9766894d..25e6e115dd1d | patch -Rp1". When I built that
> > > > > > > > > > I was able to
> > > > > > > > > > boot [4] too.
> > > > > > > > > >
> > > > > > > > > > Any idea what's going on here?
> > > > > > > > >
> > > > > > > > > Could you please try bisecting the RCU commits?
> > > > > > > >
> > > > > > > > I did a bisect and found:
> > > > > > > > 3c8e1e3eca23 ("rcu: Always set .need_qs from __rcu_read_lock() for strict GPs")
> > > > > > > >
> > > > > > > > When I reverted that on next-20200826 and on next-20200827 the
> > > > > > > > allmodconfig kernel boots [1], there's other issues that I can see in
> > > > > > > > the log but not this one.
> > > > > > >
> > > > > > > Thank you for bisecting!
> > > > > > >
> > > > > > > I take it that you have CONFIG_RCU_STRICT_GRACE_PERIOD=y in your .config?
> > > > > >
> > > > > > Correct.
> > > > > >
> > > > > > > Ah, you do, according to this line in your dmesg:
> > > > > > >
> > > > > > > rcu: RCU strict (and thus non-scalable) grace periods enabled.
> > > > > >
> > > > > > Yup.
> > > > > >
> > > > > > >
> > > > > > > And much else besides, serious debugging enabled! ;-)
> > > > > >
> > > > > > as you can see from the time it takes to boot it is alot =)
> > > > > >
> > > > > > >
> > > > > > > Does this reproduce with CONFIG_RCU_STRICT_GRACE_PERIOD=n?
> > > > > >
> > > > > > I trying that now.
> > > > > >
> > > > > > > My guess,
> > > > > > > given the commit that your bisection converged on, is that it will not
> > > > > > > reproduce in that case.
> > > > > >
> > > > > > Your guess was correct, it worked =)
> > > > >
> > > > > I cannot reproduce this. I would suggest enabling KASAN, but you have
> > > > > already enabled it. At this point, I suggest clearing up the other
> > > > > issues. If this issue remains, please let me know and I will see what
> > > > > additional diagnostics I can provide.
> > > >
> > > > On tag next-20200902 I see this workqueue lockup [1], does that give
> > > > you any ideas?
> > > > when I disable CONFIG_RCU_STRICT_GRACE_PERIOD it works.
> > >
> > > Does exactly the same commit as you bisected to last time determine
> > > whether this happens or not?
> >
> > Yes it's the exact same commit.
>
> Has this been fixed?
Here is the situation...
Historically, asynchronous RCU grace periods (call_rcu() rather
than the synchronous synchronize_rcu()) have taken some milliseconds
to complete. They still do. I was asked to produce a debugging option
that makes the grace periods much shorter, as in down to 40 microseconds.
This apparently can break things that assume longer call_rcu()-induced
grace periods. (Things like rcupdate.rcu_expedited and single-CPU
operation have trained the code to expect the occasional very short
grace period from synchronize_rcu().)
If you don't want to help debug these problems, then you should always
always always build with CONFIG_RCU_STRICT_GRACE_PERIOD=n. I have not
yet set aside time to go look into this particular issue.
Fair enough?
Thanx, Paul
