Re: [RFC Patch 4/5] PERF-HW_HBKPT: Enable perf-events to use hw-breakpoints

[Frederic Weisbecker <fweisbec@gmail.com>]

On Sun, Nov 22, 2009 at 03:52:00AM +0100, Frederic Weisbecker wrote:
> On Fri, Oct 30, 2009 at 03:52:23AM +0530, K.Prasad wrote:
> > Enable perf-events to collect memory access statistics on kernel-space
> > data in the context of a running process.
> > 
> > Signed-off-by: K.Prasad <prasad@linux.vnet.ibm.com>
> 
> 
> 
> I'm going to take the userspace part of this patch to integrate breakpoints
> in perf tools.
> 
> But I'm not sure passing the name of the symbol to the kernel and store
> it in the perf attr is a good idea.
> We should probably resolve the name from userspace and pass the address
> to the kernel. Mostly because if we want to support breakpoints in
> userspace apps, we don't want conflicts between kernel and userspace
> symbols.
> 
> Thanks.
> 


Hmm, actually instead of memory-write/memory-readwrite
I would rather see:

mem:symbol:[r][w][x]

like mem:pid_max:rw

We could default it to rw for now.

This will be more flexible I guess.
And later for ranges:

mem:symbol[[start]:[len]:[rwx]

This could be:
mem:array_name[0:5]:rwx

So, I'll need an ad-hoc parsing function for this one. This
would be hard by integrating your patch.
I'll try something else. But I can probably take the 5th patch
of your series.

Thanks.