lkml.org 
[lkml]   [2012]   [Sep]   [28]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
SubjectRe: [PATCH 01/31] perf, x86: Add PEBSv2 record support
From
On Fri, Sep 28, 2012 at 11:28 AM, Peter Zijlstra <a.p.zijlstra@chello.nl> wrote:
> On Fri, 2012-09-28 at 10:54 +0200, Stephane Eranian wrote:
>> On Fri, Sep 28, 2012 at 10:43 AM, Peter Zijlstra <a.p.zijlstra@chello.nl> wrote:
>> > On Thu, 2012-09-27 at 21:31 -0700, Andi Kleen wrote:
>> >> + if (event->attr.precise_ip > 1 && x86_pmu.intel_cap.pebs_format < 2) {
>> >
>> > Shouldn't that be: && x86_pmu.intel_cap.pebs_trap, like most other sites
>> > instead? Or didn't they flip the trap capability on Haswell?
>>
>> On Haswell, you get the event_ip which points to the sampled
>> instruction, i.e., the off-by-one
>> error can be avoided by using that value instead of pebs.rip. The nice
>> side effect is that you
>> free the LBR and minimize the overhead (no fixups). Therfore the LBR
>> filter can have any
>> setting when combined with PEBS, thus we do not need to check for
>> compatibility nor force
>> any setting for the LBR filter.
>
> Yes I got that, but what good is that trap capability flag if they don't
> use it? Them adding a second u64 to the format to report it seems to
> suggest their trap capability is pointless, but nowhere has this been
> explained.

I suspect this trap field was only meaningful for P4. Since then, the
implementation has changed and I don't think this flag is relevant anymore.
To be cautious, you could always or this (pebs_fmt > 1 || trap).


\
 
 \ /
  Last update: 2012-09-28 14:01    [W:0.300 / U:0.664 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site