linux-next: stall warnings and deadlock on Arm64 (was: [PATCH] kfence: Avoid stalling...)

[Marco Elver]

On Fri, Nov 20, 2020 at 10:26AM -0500, Steven Rostedt wrote:
> On Fri, 20 Nov 2020 15:19:28 +0100
> Marco Elver <elver at google.com> wrote:
> 
> > None of those triggered either.
> > 
> > I found that disabling ftrace for some of kernel/rcu (see below) solved
> > the stalls (and any mention of deadlocks as a side-effect I assume),
> > resulting in successful boot.
> > 
> > Does that provide any additional clues? I tried to narrow it down to 1-2
> > files, but that doesn't seem to work.
> > 
> > Thanks,
> > -- Marco
> > 
> > ------ >8 ------  
> > 
> > diff --git a/kernel/rcu/Makefile b/kernel/rcu/Makefile
> > index 0cfb009a99b9..678b4b094f94 100644
> > --- a/kernel/rcu/Makefile
> > +++ b/kernel/rcu/Makefile
> > @@ -3,6 +3,13 @@
> >  # and is generally not a function of system call inputs.
> >  KCOV_INSTRUMENT := n
> >  
> > +ifdef CONFIG_FUNCTION_TRACER
> > +CFLAGS_REMOVE_update.o = $(CC_FLAGS_FTRACE)
> > +CFLAGS_REMOVE_sync.o = $(CC_FLAGS_FTRACE)
> > +CFLAGS_REMOVE_srcutree.o = $(CC_FLAGS_FTRACE)
> > +CFLAGS_REMOVE_tree.o = $(CC_FLAGS_FTRACE)
> > +endif
> > +
> 
> Can you narrow it down further? That is, do you really need all of the
> above to stop the stalls?

I tried to reduce it to 1 or combinations of 2 files only, but that
didn't work.

> Also, since you are using linux-next, you have ftrace recursion debugging.
> Please enable:
> 
> CONFIG_FTRACE_RECORD_RECURSION=y
> CONFIG_RING_BUFFER_RECORD_RECURSION=y
> 
> when enabling any of the above. If you can get to a successful boot, you
> can then:
> 
>  # cat /sys/kernel/tracing/recursed_functions
> 
> Which would let me know if there's an recursion issue in RCU somewhere.

To get the system to boot in the first place (as mentioned in other
emails) I again needed to revert
  "rcu: Don't invoke try_invoke_on_locked_down_task() with irqs disabled",
as otherwise would run into the deadlock. That used to still result in
stall warnings, except when ftrace's recursion detection is on it seems.

With that, this is what I get:

| # cat /sys/kernel/tracing/recursed_functions
| trace_selftest_test_recursion_func+0x34/0x48:   trace_selftest_dynamic_test_func+0x4/0x28
| el1_irq+0xc0/0x180:     gic_handle_irq+0x4/0x108
| gic_handle_irq+0x70/0x108:      __handle_domain_irq+0x4/0x130
| __handle_domain_irq+0x7c/0x130: irq_enter+0x4/0x28
| trace_rcu_dyntick+0x168/0x190:  rcu_read_lock_sched_held+0x4/0x98
| rcu_read_lock_sched_held+0x30/0x98:     rcu_read_lock_held_common+0x4/0x88
| rcu_read_lock_held_common+0x50/0x88:    rcu_lockdep_current_cpu_online+0x4/0xd0
| irq_enter+0x1c/0x28:    irq_enter_rcu+0x4/0xa8
| irq_enter_rcu+0x3c/0xa8:        irqtime_account_irq+0x4/0x198
| irq_enter_rcu+0x44/0xa8:        preempt_count_add+0x4/0x1a0
| trace_hardirqs_off+0x254/0x2d8: __srcu_read_lock+0x4/0xa0
| trace_hardirqs_off+0x25c/0x2d8: rcu_irq_enter_irqson+0x4/0x78
| trace_rcu_dyntick+0xd8/0x190:   __traceiter_rcu_dyntick+0x4/0x80
| trace_hardirqs_off+0x294/0x2d8: rcu_irq_exit_irqson+0x4/0x78
| trace_hardirqs_off+0x2a0/0x2d8: __srcu_read_unlock+0x4/0x88

Thanks,
-- Marco