Re: [PATCH 0/5] [GIT PULL] updates for tip/tracing/ftrace

[Frederic Weisbecker <fweisbec@gmail.com>]

On Sat, Mar 21, 2009 at 01:44:07PM -0400, Steven Rostedt wrote:
> 
> On Sat, 21 Mar 2009, Frederic Weisbecker wrote:
> > > Testing tracer sched_switch: PASSED
> > > initcall init_sched_switch_trace+0x0/0x12 returned 0 after 99609 usecs
> > > calling  init_stack_trace+0x0/0x12 @ 1
> > > Testing tracer sysprof: .. no entries found ..FAILED!
> > > initcall init_stack_trace+0x0/0x12 returned -1 after 101562 usecs
> > > initcall init_stack_trace+0x0/0x12 returned with error code -1 
> > > calling  init_function_trace+0x0/0x12 @ 1
> > > Testing tracer function: PASSED
> > > initcall init_function_trace+0x0/0x12 returned 0 after 104492 usecs
> > > calling  init_irqsoff_tracer+0x0/0x2c @ 1
> > > Testing tracer irqsoff: .. no entries found ..FAILED!
> > > Testing tracer preemptoff: .. no entries found ..FAILED!
> > > Testing tracer preemptirqsoff: .. no entries found ..FAILED!
> > 
> > 
> > It's strange that the {*}_off tracers have failed.
> 
> Does this have your changes in it?  The ones that solved this before.


You mean Lai's patch for RCU?
I haven't seen such tracers failures since it has been merged.
I don't think it's related.

 
> > 
> > 
> > > initcall init_irqsoff_tracer+0x0/0x2c returned 0 after 8789 usecs
> > > calling  init_wakeup_tracer+0x0/0x58 @ 1
> > > Testing tracer wakeup: .. no entries found ..FAILED!
> > 
> > 
> > This one too. (sysprof doesn't count, it fails for some weeks, I think
> > it's not a hard deal to fix).
> > 
> > 
> > > initcall init_wakeup_tracer+0x0/0x58 returned -1 after 298828 usecs
> > > initcall init_wakeup_tracer+0x0/0x58 returned with error code -1 
> > > calling  stack_trace_init+0x0/0xc7 @ 1
> > > initcall stack_trace_init+0x0/0xc7 returned 0 after 0 usecs
> > > calling  init_mmio_trace+0x0/0x12 @ 1
> > > initcall init_mmio_trace+0x0/0x12 returned 0 after 0 usecs
> > > calling  init_graph_trace+0x0/0x12 @ 1
> > > Testing tracer function_graph: <3>INFO: RCU detected CPU 0 stall (t=4294678940/10000 jiffies)
> > > Pid: 1, comm: swapper Not tainted 2.6.29-rc8-tip-02752-g47b1aea-dirty #3264
> > > Call Trace:
> > >  <IRQ>  [<ffffffff8020c79d>] return_to_handler+0x0/0x73
> > >  [<ffffffff80211150>] print_context_stack+0xa0/0xd3
> > >  [<ffffffff8020c79d>] return_to_handler+0x0/0x73
> > >  [<ffffffff8020fb26>] dump_trace+0x22d/0x2cc
> > >  [<ffffffff8020c79d>] return_to_handler+0x0/0x73
> > >  [<ffffffff80211008>] show_trace_log_lvl+0x51/0x5d
> > >  [<ffffffff8020c79d>] return_to_handler+0x0/0x73
> > >  [<ffffffff80211029>] show_trace+0x15/0x17
> > >  [<ffffffff8020c79d>] return_to_handler+0x0/0x73
> > >  [<ffffffff802111fa>] dump_stack+0x77/0x81
> > >  [<ffffffff8020c79d>] return_to_handler+0x0/0x73
> > >  [<ffffffff8029e6dd>] print_cpu_stall+0x40/0xa4
> > >  [<ffffffff8020c79d>] return_to_handler+0x0/0x73
> > >  [<ffffffff8029e8be>] check_cpu_stall+0x49/0x76
> > >  [<ffffffff8020c79d>] return_to_handler+0x0/0x73
> > >  [<ffffffff8029e902>] __rcu_pending+0x17/0xfc
> > >  [<ffffffff8020c79d>] return_to_handler+0x0/0x73
> > >  [<ffffffff8029ea13>] rcu_pending+0x2c/0x5e
> > >  [<ffffffff8020c79d>] return_to_handler+0x0/0x73
> > >  [<ffffffff8026abef>] update_process_times+0x3c/0x77
> > >  [<ffffffff8020c79d>] return_to_handler+0x0/0x73
> > >  [<ffffffff802875dd>] tick_periodic+0x6e/0x70
> > 
> > 
> > Still hanging in the timer interrupt.
> > I guess it makes the timer interrupt servicing too slow and then
> > once it is serviced, another one is raised.
> > 
> > But the cause is perhaps more complex
> > 
> > I think you have had too much hanging of this type.
> > I'm preparing a fix that checks periodically if the function graph
> > tracer is spending too much time in an interrupt.
> > 
> > I guess I could count the number of function executed between the irq entry
> > and its exit.
> > 
> > That's the best: if we are hanging in an interrupt, it could be whatever
> > interrupt and the jiffies could not be progressing so I can't rely
> > on time but only on number of functions executed.
> > 
> > May be 10000 calls is a good threshold before killing the function graph
> > inside an interrupt?
> > 
> > Let's try, I will also provide a way to dump the function graph traces from
> > the ring-buffer on the screen, it could help to debug it in this case.
> 
> I was thinking the same thing. All you need to do is add a "ftrace_dump()" 
> in the print_cpu_stall() function in kernel/rcuclassic.c.


Perhaps not relying on rcu cpu ctall detector, because it could perhaps
hang without it.
I think I should directly call ftrace_dump() from the tracer and not
rely on CONFIG that might not be enabled.


> You would need to add "#include <linux/ftrace.h>" too.
> 
> /me wonders if we should add ftrace_dump() to kernel.h to remove that 
> requirement?
> 
> -- Steve
>