Re: [PATCH 9/9] perf/x86: add syfs entry to disable HT bug workaround
From: Stephane Eranian
Date: Thu Jun 05 2014 - 06:19:52 EST
On Thu, Jun 5, 2014 at 12:01 PM, Matt Fleming <matt@xxxxxxxxxxxxxxxxx> wrote:
> On 5 June 2014 10:29, Stephane Eranian <eranian@xxxxxxxxxx> wrote:
>>
>> If you know what you are doing (poweruser), then there are measurements
>> which works fine with the HT erratum. This is why we have the option.
>>
>> For instance if you only measure events 4x4 in system-wide mode
>> and you know which counters these event are going to use, you don't
>> need the workaround. For instance:
>>
>> # perf stat -a -e r81d0,r01d1,r08d0,r20d1 sleep 5
>>
>> Works well if you have a uniform workload across all CPUs.
>> All those events leak, but the leaks balance themselves and you
>> get the correct counts in the end. The advantage is that you don't
>> have to multiplex. With the workaround enable, this would multiplex
>> a lot.
>>
>> But as I said, this is for experts only.
>
> Is it not possible to detect this in the kernel and only enable the
> workaround for the case where the leaks don't balance? It may not be
> possible (or practical) but I do think it's worth having the
> discussion.
>
How would you know that you have a uniform workload from inside
the kernel?
>> Another reason is for systems with HT disabled. It turned out to be
>> very difficult to determine at kernel BOOT TIME if HT was enabled
>> or not. Note what I said: ENABLED and not SUPPORTED. The latter is
>> easy to detect. The former needs some model specific code which is
>> quite complicated. I wish the kernel had this capability abstracted
>> somehow. Consequently, the workaround is always enabled. When
>> HT is disabled, there won't be multiplexing because there will never
>> be conflict, but you pay a little price for accessing the extra data
>> state.
>
> Does cpu_sibling_map not give you some indication of whether HT is
> enabled? I think the topology_thread_cpumask() is the topology API for
> that. But I could most definitely be wrong. Hopefully someone on the
> Cc list will know.
>
Remember trying some of that, but when perf_event is initialized, those
masks are not yet setup properly.
>>An init script could well detect HT is off and thus disable the workaround altogether.
>
> This is exactly the kind of thing I think we should try to avoid. The
> ideal is that things just work out of the box and don't require these
> magic knobs to be tweaked.
>
>> Those are the two main reasons for this control in sysfs.
>
> Thanks for the info!
--
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/