Re: [cpufreq] 06c4d00466: will-it-scale.per_process_ops -53.4% regression

From: Thomas Gleixner
Date: Wed Mar 25 2020 - 06:32:56 EST


Rong Chen <rong.a.chen@xxxxxxxxx> writes:
> On 3/24/20 6:24 PM, Andy Shevchenko wrote:
>> On Tue, Mar 24, 2020 at 8:02 AM kernel test robot <rong.a.chen@xxxxxxxxx> wrote:
>>> Greeting,
>>>
>>> FYI, we noticed a -53.4% regression of will-it-scale.per_process_ops due to commit:
>>> commit: 06c4d00466eb374841bc84c39af19b3161ff6917 ("[patch 09/22] cpufreq: Convert to new X86 CPU match macros")
>>> url: https://github.com/0day-ci/linux/commits/Thomas-Gleixner/x86-devicetable-Move-x86-specific-macro-out-of-generic-code/20200321-031729
>>> base: https://git.kernel.org/cgit/linux/kernel/git/rafael/linux-pm.git linux-next
>>>
>>> in testcase: will-it-scale
>>> on test machine: 4 threads Intel(R) Core(TM) i3-3220 CPU @ 3.30GHz with 8G memory
>>> with following parameters:
>>
>> drivers/cpufreq/speedstep-centrino.c change missed the terminator,
>> perhaps it's a culprit, because I don't believe removing dups and
>> reordering lines may affect this.
>> Can you restore terminator there and re-test?
>>
>
> I have retested with the change, but it has no effect on the performance.

Bah. The binary equivalence testing detected this, but I obvioulsy
missed it. Delta fix below.

Thanks,

tglx

8<--------------
--- a/drivers/cpufreq/intel_pstate.c
+++ b/drivers/cpufreq/intel_pstate.c
@@ -2727,7 +2727,7 @@ static inline void intel_pstate_request_

#define X86_MATCH_HWP(model, hwp_mode) \
X86_MATCH_VENDOR_FAM_MODEL_FEATURE(INTEL, 6, INTEL_FAM6_##model, \
- X86_FEATURE_APERFMPERF, hwp_mode)
+ X86_FEATURE_HWP, hwp_mode)

static const struct x86_cpu_id hwp_support_ids[] __initconst = {
X86_MATCH_HWP(BROADWELL_X, INTEL_PSTATE_HWP_BROADWELL),