Re: [RFC PATCH] hw-breakpoints, kgdb, x86: add a flag topassDIE_DEBUG notification

[Frederic Weisbecker <fweisbec@gmail.com>]

On Mon, Jul 26, 2010 at 07:13:23PM +0800, DDD wrote:
> Frederic Weisbecker wrote:
>> Why? It seems to me a kernel debugger should have the highest priority
>> over anything.
>
> In my option, the reason of kgdb set the lowest-prio for
> notifier is:
>
> For letting kgdb to keep simple, there is no codes to check the
> breakpoint event was generated by kgdb or not, thus it have to set kgdb
> as lowest priority to notifier.
>
> If the breakpoint event is not generated by kgdb, the source of the
> breakpoint event will consume that event before passing to kgdb's
> routine, so that the breakpoint event of kgdb getting must be generated  
> by kgdb itself.



Ok, but that makes it hard to differentiate from a spurious breakpoint
event.




>>
>>
>>
>>>> - Always returning NOTIFY_DONE from the breakpoint path.
>>>>
>>>>   
>>> Without some further investigation, I am not sure what this will do.
>>
>>
>>
>> Nothing, this NOTIFY_STOP is only an optimization. But now I think that
>> won't solve the problem. We still clear a dr6 trap bit for a debug
>> exception due to lazy dr7 switches we have to handle.
>>
>> This is why kgdb should have the highest priority, or use the overflow
>> callback.
>
>
> OK, I will try to use the overflow callback to let kgdb works with
> hw_breakpoints. :-)
>
> Thanks,
> Dongdong


Thanks.

kgdb sets breakpoints through arch_install_hw_breakpoint()
So when it triggers, it will be handled by perf through
perf_bp_event(), so it's quite natural it is considered as
handled and then it's bit removed from dr6. The only way
for kgdb to handle it properly is to set an overflow_handler.

Tell me if you encounter any problem.