* Peter Zijlstra wrote:
> I never can seem to figure out how to use the scripts mess :/
> acme, is there anything we could do to make that stuff usable?
> There's a ton of crap under scripts/ but I don't even know how
> to get that stuff to run.
>
> What's more, all that nonsense is in
* Peter Zijlstra a.p.zijls...@chello.nl wrote:
I never can seem to figure out how to use the scripts mess :/
acme, is there anything we could do to make that stuff usable?
There's a ton of crap under scripts/ but I don't even know how
to get that stuff to run.
What's more, all that
... NIH snipped ...
> If you do this inside perf, you have access to more infrastructure
> code, e.g., dwarf.
> For instance, I am interested in getting a value profiling mode. That
> means sampling
> the values of function arguments. That needs some dwarf support and the PEBS
> machine state.
On Tue, Oct 23, 2012 at 3:45 PM, Andi Kleen wrote:
>> I believe we can use a similar approach to PERF_SAMPLE_REGS to expose
>> the PEBS machine state. We need to add PERF_SAMPLE_IREGS and then
>> return interrupt machine registers with regular sampling and the PEBS machine
>> registers in precise
On Tue, Oct 23, 2012 at 03:50:50PM +0200, Peter Zijlstra wrote:
> PERF_SAMPLE_RAW has a u32 size header and starts the data after that.
> This means you PERF_SAMPLE_RAW output ends up on a u32 aligned end
> address -- assuming the data is a u64 multiple, this is not good.
>
> > It appears to work
On Tue, 2012-10-23 at 15:30 +0200, Andi Kleen wrote:
> > Also, there's an alignment issue there, the raw.data is 32bit offset,
> > the record is u64 aligned, leaving the output stream offset, wrecking
> > things.
>
> Can you explain more? Not sure I understand.
PERF_SAMPLE_RAW has a u32 size
> I believe we can use a similar approach to PERF_SAMPLE_REGS to expose
> the PEBS machine state. We need to add PERF_SAMPLE_IREGS and then
> return interrupt machine registers with regular sampling and the PEBS machine
> registers in precise mode. A while back I wrote a patch to do just this.
> Also, there's an alignment issue there, the raw.data is 32bit offset,
> the record is u64 aligned, leaving the output stream offset, wrecking
> things.
Can you explain more? Not sure I understand.
It appears to work at least.
>
> And as with any ABI extension, it should come with useful
On Tue, Oct 23, 2012 at 2:31 PM, Peter Zijlstra wrote:
> On Thu, 2012-10-18 at 16:19 -0700, Andi Kleen wrote:
>> + if (event->attr.sample_type & PERF_SAMPLE_RAW) {
>> + raw.size = x86_pmu.pebs_record_size;
>> + raw.data = __pebs;
>> + data.raw =
>>
On Thu, 2012-10-18 at 16:19 -0700, Andi Kleen wrote:
> + if (event->attr.sample_type & PERF_SAMPLE_RAW) {
> + raw.size = x86_pmu.pebs_record_size;
> + raw.data = __pebs;
> + data.raw =
> + }
The Changelog babbles about registers, yet you
On Thu, 2012-10-18 at 16:19 -0700, Andi Kleen wrote:
+ if (event-attr.sample_type PERF_SAMPLE_RAW) {
+ raw.size = x86_pmu.pebs_record_size;
+ raw.data = __pebs;
+ data.raw = raw;
+ }
The Changelog babbles about registers, yet you export
On Tue, Oct 23, 2012 at 2:31 PM, Peter Zijlstra a.p.zijls...@chello.nl wrote:
On Thu, 2012-10-18 at 16:19 -0700, Andi Kleen wrote:
+ if (event-attr.sample_type PERF_SAMPLE_RAW) {
+ raw.size = x86_pmu.pebs_record_size;
+ raw.data = __pebs;
+
Also, there's an alignment issue there, the raw.data is 32bit offset,
the record is u64 aligned, leaving the output stream offset, wrecking
things.
Can you explain more? Not sure I understand.
It appears to work at least.
And as with any ABI extension, it should come with useful userspace
I believe we can use a similar approach to PERF_SAMPLE_REGS to expose
the PEBS machine state. We need to add PERF_SAMPLE_IREGS and then
return interrupt machine registers with regular sampling and the PEBS machine
registers in precise mode. A while back I wrote a patch to do just this. Once
On Tue, 2012-10-23 at 15:30 +0200, Andi Kleen wrote:
Also, there's an alignment issue there, the raw.data is 32bit offset,
the record is u64 aligned, leaving the output stream offset, wrecking
things.
Can you explain more? Not sure I understand.
PERF_SAMPLE_RAW has a u32 size header and
On Tue, Oct 23, 2012 at 03:50:50PM +0200, Peter Zijlstra wrote:
PERF_SAMPLE_RAW has a u32 size header and starts the data after that.
This means you PERF_SAMPLE_RAW output ends up on a u32 aligned end
address -- assuming the data is a u64 multiple, this is not good.
It appears to work at
On Tue, Oct 23, 2012 at 3:45 PM, Andi Kleen a...@firstfloor.org wrote:
I believe we can use a similar approach to PERF_SAMPLE_REGS to expose
the PEBS machine state. We need to add PERF_SAMPLE_IREGS and then
return interrupt machine registers with regular sampling and the PEBS machine
registers
... NIH snipped ...
If you do this inside perf, you have access to more infrastructure
code, e.g., dwarf.
For instance, I am interested in getting a value profiling mode. That
means sampling
the values of function arguments. That needs some dwarf support and the PEBS
machine state. With
From: Andi Kleen
Add support for reporting PEBS records in a raw format that can
be then parsed by perf script.
We exposed most of the Haswell PEBS fields in a generic way
in this patchkit:
- Aborted cycles is in weight
- Memory latency is in weight
- DataLA is in address
- EventingRIP is used
From: Andi Kleen a...@linux.intel.com
Add support for reporting PEBS records in a raw format that can
be then parsed by perf script.
We exposed most of the Haswell PEBS fields in a generic way
in this patchkit:
- Aborted cycles is in weight
- Memory latency is in weight
- DataLA is in address
-
20 matches
Mail list logo
