Re: [PATCH V2 1/4] perf: Add hardware breakpoint address mask

[Jacob Shin]

On Fri, Apr 26, 2013 at 05:20:44PM +0100, Will Deacon wrote:
> Hi Jacob,
> 
> On Fri, Apr 26, 2013 at 12:19:11AM +0100, Jacob Shin wrote:
> > On Thu, Apr 25, 2013 at 10:17:35AM -0700, H. Peter Anvin wrote:
> > > On 04/25/2013 10:06 AM, Oleg Nesterov wrote:
> > > >>
> > > >> The downside is that in userland perf tool we need differing documentation
> > > >> on what the mask syntax means for each architecture.
> > > > 
> > > > Personally I think this is acceptable.
> > > > 
> > > > But I am new to this code, so...
> > > > 
> > > 
> > > That would seem really, really awkward.  Yes, perf has a bunch of
> > > low-level stuff, but it would seem highly undesirable to force the user
> > > to deal with something like that.
> > > 
> > > It would be good to have a user-friendly syntax that covers most of what
> > > users may want to do and perhaps a longer form that can express
> > > everything including ARM's byte selects; if the system can't honor the
> > > request it should return an error.
> > 
> > Okay,
> > 
> > If arch specific masks are a no go, then I think I'm convinced that
> > Oleg's idea of using bp_len is the right thing to do. Right now perf
> > userland tool hard codes bp_len to 4, so I need to modify it to allow
> > user to override the length if desired.
> 
> So what value ends up in the bp_len field: a length or a mask? I just want
> to make sure it's flexible enough that, if we add another user interface for
> the byte-select stuff, we don't need to butcher the attr in another way.

It is length. Just as it is today, userland API does not change at all.

> 
> > Oleg, Frederic, et al.
> > 
> > Which syntax do you prefer?
> > 
> > If we want to set bp_len to 16:
> > 
> >   $ perf stat -e mem:0x1000:rw:16
> > 
> > Or
> > 
> >   $ perf stat -e mem:0x1000:16
> > 
> > Or
> > 
> >   $ perf stat -e mem:0x1000/16
> > 
> > If no bp_len value is specified, it will still default to 4 as it did
> > before.
> 
> I certainly like the ability to change the length of an execute breakpoint,
> as that helps for halfword instructions on ARM (not sure if your first
> syntax precludes that, but just checking...).

All three are equivalent, I just need to pick one and implement/document.

-Jacob

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/