Re: [PATCH] rcu/tree: Add a trace event for RCU stall warnings

[Neeraj Upadhyay <neeraju@codeaurora.org>]

Hi Sangmoon,


On 2/22/2021 1:55 PM, Sangmoon Kim wrote:
> 
> 
>> -----Original Message-----
>> From: Paul E. McKenney <paulmck@kernel.org>
>> Sent: Friday, February 19, 2021 3:59 AM
>>
>> On Thu, Feb 18, 2021 at 01:24:16PM -0500, Steven Rostedt wrote:
>>> On Thu, 18 Feb 2021 09:00:12 -0800
>>> "Paul E. McKenney" <paulmck@kernel.org> wrote:
>>>
>>>> OK, I will bite.  ;-)
>>>
>>> Do I turn into a wearwolf?
>>
>> Maybe even a wash-n-wear wolf.  ;-)
>>
>>>> Can the saving of console into trace buffers be easily enabled and
>>>> disabled within the code?
>>>
>>> Well, we do have this for ftrace:
>>>
>>>   echo 'enable_event:printk:console' > events/rcu/rcu_stall_warning/trigger
>>>
>>> Would enable the printk console to be traced to the ring buffer once the
>>> stall is triggered, and not before hand.
>>>
>>>>   If so, does enabling and disabling nest?
>>>
>>> It has a disable as well, but unfortunately its global and does not nest.
>>> We could always implement a percpu trigger that can nest though.
>>
>> So Sangmoon could have a module react to an RCU CPU stall warning
>> tracepoint by turning on tracing of console messages.  If there was a
>> start-stop pair of RCU CPU stall warning tracepoints, he could turn it
>> off as well.
>>
>> Not sure if that is what Sangmoon is looking for, but it is a possibility.
>>
>> 							Thanx, Paul
> 
> Thank you. I'm learning the new things through this review.
> 
> Copying console messages is not really necessary to me. I wanted to say
> this tracepoint can also be used in such way for debugging purpose.
> 

In case we want to use this trace event, to capture RCU stall 
information, which comes on console, into trace buffers; why not have 
these events at the start of the functions like print_other_cpu_stall(), 
print_cpu_stall() rather than  at the end ?


Thanks
Neeraj

> What I need is that a module records the stall happening and report the
> stall event when kernel panic occurred. Of course, there is no detailed
> information there, but it helps developers to recognize that the stall
> has occurred in the system.
> 
> The usage scenario is as follows.
> There is a kernel module for debugging.
> The module registers a probe function using register_trace_rcu_stall_warning().
> When a stall warning event occurs, the module records the event into its own buffer.
> When a kernel panic occurs, the developers can see the report generated
> by the module and can also see the buffer of the module directly through Ramdump.
> 
> Thanks,
> Sangmoon
> 

-- 
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a 
member of the Code Aurora Forum, hosted by The Linux Foundation