Re: [GIT PULL] perf fixes

[Frederic Weisbecker <fweisbec@gmail.com>]

On Mon, Feb 22, 2010 at 09:01:39AM -0800, Linus Torvalds wrote:
> 
> 
> On Mon, 22 Feb 2010, Ingo Molnar wrote:
> > 
> > The reason i'm sending them post -rc8 is the fix for the Wine breakage from 
> > Frederic, which would be nice to see resolved in -final.
> 
> Umm. That commit seems to be crap.
> 
> Yes, it was wrong to not allow a NULL address breakpoint. But if you call 
> kallsyms_lookup_name(), you should check the "I couldn't find it" end 
> result for _that_ case, and 0 is the correct check there.
> 
> So the code _should_ do something like
> 
> 	if (info->name) {
> 		info->address = kallsyms_lookup_name(info->name);
> 		if (!info->address)
> 			return -EINVAL;
> 	}
> 
> and that would correctly handle both the NULL original address _and_ the 
> name lookup failure.



Ah you're right. That part in itself is buggy. Fortunately this won't ever
be reached because we don't use the info->name field anymore . It's a field I
forgot to remove when I re-adapted the new hw-breakpoint API on top of perf.
The name resolution (used by the ksym tracer and the hw-breakpoint sample
only) is done in an upper layer now, and never stored in the arch back end in any
case.

I'll remove this field and the associated dead-code for .34

Thanks.