Re: [RFC] Hardware Breakpoint support for systemtap translator

[Prerna Saxena <prerna@linux.vnet.ibm.com>]

Hi Roland,

Roland McGrath wrote:
>> 1. probe kernel.data(ADDRESS).write     {....}   : To probe writes at 
>> the given address
>> 2. probe kernel.data(ADDRESS).rw         {....}   : To probe read & 
>> write access to the given address.
>> 3. probe kernel.data("SYMBOL").write   {....}
>> 4. probe kernel.data("SYMBOL").rw       {....}   : Similar to 1,2, but 
>> using a symbol name as argument.
>>     
>
> That is a nice start!  But what about dynamic uses?
> i.e.
>
> 	probe kernel.function("foobar")
> 	{
> 	  w1 = watch $foo->bar
> 	}
>
> 	probe kernel.function.return("foobar")
> 	{
> 	  unwatch w1
> 	}
>
> 	probe watch.w1
> 	{
> 	  val = $$watch.baz # i.e. $foo->bar.baz
> 	}
>
> or probably some entirely different syntax.  But I think you get the idea:
> enable/disable dynamically-discovered addresses to trigger watchpoint
> probes of some sort.  I think it would be really nice if the watch probe
> could somehow be embedded in the enabling probe's context so it can use
> $bar there and have those address translations taken at enable-time.
>
>   

Enabling systemtap to set hardware breakpoints at runtime would be a 
very useful feature to have, but it presents a problem. Kprobe handlers 
get executed in exception context, and it is not possible to register a 
hardware breakpoint from within exception context. I'm wondering how it 
could be implemented..

> Also, what about specifying access-size for the watchpoint, on machines
> where that can be done (i.e. x86)?
>   

Thats a nice suggestion, I'll add this !

>
> Thanks,
> Roland
>   
Thanks,

-- 
Prerna Saxena

Linux Technology Centre,
IBM Systems and Technology Lab,
Bangalore, India