On 11/11/2025 6:01 pm, Ian Rogers wrote:
> On Mon, Nov 10, 2025 at 10:21 PM Namhyung Kim <namhyung@kernel.org> wrote:
>>
>> Hi Ian,
>>
>> On Sun, Nov 09, 2025 at 05:31:47PM -0800, Ian Rogers wrote:
>>> Switch the __get_cpuid feature for intel-pt to use the provided cpuid
>>> function in perf, this removes the need for NO_AUXTRACE when the
>>> feature detection fails. Remove the now unnecessary feature
>>> detection. Remove NO_AUXTRACE as it just builds a more broken version
>>
>> Can you please elaborate what the broken part is?
>
> Sure. I'll summarize what alters in patch 4. NO_AUXTRACE is
> controlling 3 main things:
> * set up of aux options for PMUs (code in the arch directory)
> * ARM: coresight and SPE
> * Intel: BTS and PT
> * PowerPC: VPA DTL
> * S390: cpumsf
> * support for decoding aux events (common code that can be
> cross-compiled assuming other library dependencies are available)
> * ARM: coresight
> * HiSi: PTT decoder
> * Intel: BTS and PT
> * PowerPC: VPA DTL
> * S390: cpumsf
> * Tool support for aux buffers (common shared builtin code):
> * perf record: aux options for events, snapshot, aux-sample
> * perf inject: aux events will fail the entire perf inject command
>
> So somebody with a NO_AUXTRACE build would generally experience a very
> sad perf command. Having the option made sense when there were feature
> tests that could fail, but possibly that should have just controlled
> not compiling intel-pt. Having the option is extra burden on
> developers/maintainers, as shown in my comment:
>
>> This was prompted by needing to make a v2 patch set of:
>> https://lore.kernel.org/lkml/20251107170712.2302714-1-irogers@google.com/
>> due to a broken NO_AUXTRACE configuration.
>
> Somebody may have been using NO_AUXTRACE as a proxy for not having
> some library, but I don't see that in the code. If this is the case we
> should add the appropriate feature test, ..
> Not having NO_AUXTRACE may have been a bug work around for someone, in
> which case we should work to fix the bug. Again, I don't know of this
> case and don't see it in the code.
>
> Thanks,
> Ian
>
>> Thanks,
>> Namhyung
Seems like a nice simplification even if nothing was badly broken.
Reviewed-by: James Clark <james.clark@linaro.org>