| Message ID | 535FE55D.40500@semaphore.gr (mailing list archive) |
|---|---|
| State | Not Applicable, archived |
| Headers | show |
On Tuesday, April 29, 2014 08:46:05 PM Stratos Karafotis wrote: > According to Intel 64 and IA-32 Architectures SDM, Volume 3, > Chapter 14.2, "Software needs to exercise care to avoid delays > between the two RDMSRs (for example interrupts)". > > So, disable interrupts during reading MSRs IA32_APERF and IA32_MPERF. > Since the TSC is also takes place in the calculation include it in > the protected section. This should increase the accuracy of the > calculations. > > Tested on Intel i7-3770 CPU @ 3.40GHz. > Benchmarks (Phoronix Linux compilation) show a slightly increase > in performance by ~0.1% and a decrease in power consumption by > ~0.13%. > > Signed-off-by: Stratos Karafotis <stratosk@semaphore.gr> This looks good to me, but I need Dirk Brandewie to review and comment all intel_pstate patches before I apply any of them. Thanks! > --- > > I hope it's not too inappropriate to post test results in > conjunction with my previous patch about the change of calculation > method of next p-state (https://lkml.org/lkml/2014/4/27/100). > I used Phoronix benchmark - Linux Kernel Compilation 3.1. > Below the test results using turbostat (5 iterations): > > Without 2 patches: > > Ph. avg Time Total time PkgWatt Total Energy > 79.63 266.416 57.74 15382.85984 > 79.63 265.609 57.87 15370.79283 > 79.57 266.994 57.54 15362.83476 > 79.53 265.304 57.83 15342.53032 > 79.71 265.977 57.76 15362.83152 > avg 79.61 266.06 57.74 15364.36985 > > With change calculation patch: > > Ph. avg Time Total time PkgWatt Total Energy > 78.23 258.826 59.14 15306.96964 > 78.41 259.110 59.15 15326.35650 > 78.40 258.530 59.26 15320.48780 > 78.46 258.673 59.20 15313.44160 > 78.19 259.075 59.16 15326.87700 > avg 78.34 258.842 59.18 15318.82650 > > With change calculation patch + this one > > Ph. avg Time Total time PkgWatt Total Energy > 78.33 259.320 59.06 15315.43920 > 78.28 258.245 59.20 15288.10400 > 78.08 257.523 59.32 15276.26436 > 78.38 259.084 59.13 15319.63692 > 78.31 258.232 59.22 15292.49904 > avg 78.27 258.480 59.18 15298.38870 > > > drivers/cpufreq/intel_pstate.c | 3 +++ > 1 file changed, 3 insertions(+) > > diff --git a/drivers/cpufreq/intel_pstate.c b/drivers/cpufreq/intel_pstate.c > index 8e309db..95b3958 100644 > --- a/drivers/cpufreq/intel_pstate.c > +++ b/drivers/cpufreq/intel_pstate.c > @@ -576,10 +576,13 @@ static inline void intel_pstate_sample(struct cpudata *cpu) > { > u64 aperf, mperf; > unsigned long long tsc; > + unsigned long flags; > > + local_irq_save(flags); > rdmsrl(MSR_IA32_APERF, aperf); > rdmsrl(MSR_IA32_MPERF, mperf); > tsc = native_read_tsc(); > + local_irq_restore(flags); > > aperf = aperf >> FRAC_BITS; > mperf = mperf >> FRAC_BITS; >
diff --git a/drivers/cpufreq/intel_pstate.c b/drivers/cpufreq/intel_pstate.c index 8e309db..95b3958 100644 --- a/drivers/cpufreq/intel_pstate.c +++ b/drivers/cpufreq/intel_pstate.c @@ -576,10 +576,13 @@ static inline void intel_pstate_sample(struct cpudata *cpu) { u64 aperf, mperf; unsigned long long tsc; + unsigned long flags; + local_irq_save(flags); rdmsrl(MSR_IA32_APERF, aperf); rdmsrl(MSR_IA32_MPERF, mperf); tsc = native_read_tsc(); + local_irq_restore(flags); aperf = aperf >> FRAC_BITS; mperf = mperf >> FRAC_BITS;
According to Intel 64 and IA-32 Architectures SDM, Volume 3, Chapter 14.2, "Software needs to exercise care to avoid delays between the two RDMSRs (for example interrupts)". So, disable interrupts during reading MSRs IA32_APERF and IA32_MPERF. Since the TSC is also takes place in the calculation include it in the protected section. This should increase the accuracy of the calculations. Tested on Intel i7-3770 CPU @ 3.40GHz. Benchmarks (Phoronix Linux compilation) show a slightly increase in performance by ~0.1% and a decrease in power consumption by ~0.13%. Signed-off-by: Stratos Karafotis <stratosk@semaphore.gr> --- I hope it's not too inappropriate to post test results in conjunction with my previous patch about the change of calculation method of next p-state (https://lkml.org/lkml/2014/4/27/100). I used Phoronix benchmark - Linux Kernel Compilation 3.1. Below the test results using turbostat (5 iterations): Without 2 patches: Ph. avg Time Total time PkgWatt Total Energy 79.63 266.416 57.74 15382.85984 79.63 265.609 57.87 15370.79283 79.57 266.994 57.54 15362.83476 79.53 265.304 57.83 15342.53032 79.71 265.977 57.76 15362.83152 avg 79.61 266.06 57.74 15364.36985 With change calculation patch: Ph. avg Time Total time PkgWatt Total Energy 78.23 258.826 59.14 15306.96964 78.41 259.110 59.15 15326.35650 78.40 258.530 59.26 15320.48780 78.46 258.673 59.20 15313.44160 78.19 259.075 59.16 15326.87700 avg 78.34 258.842 59.18 15318.82650 With change calculation patch + this one Ph. avg Time Total time PkgWatt Total Energy 78.33 259.320 59.06 15315.43920 78.28 258.245 59.20 15288.10400 78.08 257.523 59.32 15276.26436 78.38 259.084 59.13 15319.63692 78.31 258.232 59.22 15292.49904 avg 78.27 258.480 59.18 15298.38870 drivers/cpufreq/intel_pstate.c | 3 +++ 1 file changed, 3 insertions(+)