| Message ID | 20210112052127.4557-1-yu.c.chen@intel.com (mailing list archive) |
|---|---|
| State | Mainlined, archived |
| Headers | show |
| Series | [v3] cpufreq: intel_pstate: Get percpu max freq via HWP MSR register if available | expand |
On Tue, Jan 12, 2021 at 6:19 AM Chen Yu <yu.c.chen@intel.com> wrote: > > Currently when turbo is disabled(either by BIOS or by the user), the > intel_pstate driver reads the max non-turbo frequency from the package-wide > MSR_PLATFORM_INFO(0xce) register. However on asymmetric platforms it is > possible in theory that small and big core with HWP enabled might have > different max non-turbo cpu frequency, because the MSR_HWP_CAPABILITIES > is percpu scope according to Intel Software Developer Manual. > > The turbo max freq is already percpu basis in current code, thus make > similar change to the max non-turbo frequency as well. > > Reported-by: Wendy Wang <wendy.wang@intel.com> > Signed-off-by: Chen Yu <yu.c.chen@intel.com> > --- > v2: Per Srinivas' suggestion, avoid duplicated assignment of max_pstate. > v3: Per Rafael's suggestion, do not add new argument in intel_pstate_get_hwp_max() > to avoid redundant local vars. > Per Srinivas' suggestion, refined the commit log to reflect the 'non-turbo' > max frequency. Looks good now, thanks! Is it needed in -stable and if so, which -stable series should it go into? > -- > drivers/cpufreq/intel_pstate.c | 5 +++-- > 1 file changed, 3 insertions(+), 2 deletions(-) > > diff --git a/drivers/cpufreq/intel_pstate.c b/drivers/cpufreq/intel_pstate.c > index eaf32ef7a030..99e180f644c3 100644 > --- a/drivers/cpufreq/intel_pstate.c > +++ b/drivers/cpufreq/intel_pstate.c > @@ -1724,11 +1724,9 @@ static void intel_pstate_max_within_limits(struct cpudata *cpu) > static void intel_pstate_get_cpu_pstates(struct cpudata *cpu) > { > cpu->pstate.min_pstate = pstate_funcs.get_min(); > - cpu->pstate.max_pstate = pstate_funcs.get_max(); > cpu->pstate.max_pstate_physical = pstate_funcs.get_max_physical(); > cpu->pstate.turbo_pstate = pstate_funcs.get_turbo(); > cpu->pstate.scaling = pstate_funcs.get_scaling(); > - cpu->pstate.max_freq = cpu->pstate.max_pstate * cpu->pstate.scaling; > > if (hwp_active && !hwp_mode_bdw) { > unsigned int phy_max, current_max; > @@ -1736,9 +1734,12 @@ static void intel_pstate_get_cpu_pstates(struct cpudata *cpu) > intel_pstate_get_hwp_max(cpu, &phy_max, ¤t_max); > cpu->pstate.turbo_freq = phy_max * cpu->pstate.scaling; > cpu->pstate.turbo_pstate = phy_max; > + cpu->pstate.max_pstate = HWP_GUARANTEED_PERF(READ_ONCE(cpu->hwp_cap_cached)); > } else { > cpu->pstate.turbo_freq = cpu->pstate.turbo_pstate * cpu->pstate.scaling; > + cpu->pstate.max_pstate = pstate_funcs.get_max(); > } > + cpu->pstate.max_freq = cpu->pstate.max_pstate * cpu->pstate.scaling; > > if (pstate_funcs.get_aperf_mperf_shift) > cpu->aperf_mperf_shift = pstate_funcs.get_aperf_mperf_shift(); > -- > 2.17.1 >
On Tue, Jan 12, 2021 at 02:52:50PM +0100, Rafael J. Wysocki wrote: > On Tue, Jan 12, 2021 at 6:19 AM Chen Yu <yu.c.chen@intel.com> wrote: > > > > Currently when turbo is disabled(either by BIOS or by the user), the > > intel_pstate driver reads the max non-turbo frequency from the package-wide > > MSR_PLATFORM_INFO(0xce) register. However on asymmetric platforms it is > > possible in theory that small and big core with HWP enabled might have > > different max non-turbo cpu frequency, because the MSR_HWP_CAPABILITIES > > is percpu scope according to Intel Software Developer Manual. > > > > The turbo max freq is already percpu basis in current code, thus make > > similar change to the max non-turbo frequency as well. > > > > Reported-by: Wendy Wang <wendy.wang@intel.com> > > Signed-off-by: Chen Yu <yu.c.chen@intel.com> > > --- > > v2: Per Srinivas' suggestion, avoid duplicated assignment of max_pstate. > > v3: Per Rafael's suggestion, do not add new argument in intel_pstate_get_hwp_max() > > to avoid redundant local vars. > > Per Srinivas' suggestion, refined the commit log to reflect the 'non-turbo' > > max frequency. > > Looks good now, thanks! > > Is it needed in -stable and if so, which -stable series should it go into? > Yes, I think so, it should be Cc: stable@vger.kernel.org # 4.18+ as the HWP reading turbo frequency was firstly introduced in v4.18-rc2 and it was easier to be applied. BTW, this patch is on top of your previous patch set on intel_pstate clean up: https://patchwork.kernel.org/project/linux-pm/list/?series=410797 thanks, Chenyu
On Tue, Jan 12, 2021 at 4:00 PM Chen Yu <yu.c.chen@intel.com> wrote: > > On Tue, Jan 12, 2021 at 02:52:50PM +0100, Rafael J. Wysocki wrote: > > On Tue, Jan 12, 2021 at 6:19 AM Chen Yu <yu.c.chen@intel.com> wrote: > > > > > > Currently when turbo is disabled(either by BIOS or by the user), the > > > intel_pstate driver reads the max non-turbo frequency from the package-wide > > > MSR_PLATFORM_INFO(0xce) register. However on asymmetric platforms it is > > > possible in theory that small and big core with HWP enabled might have > > > different max non-turbo cpu frequency, because the MSR_HWP_CAPABILITIES > > > is percpu scope according to Intel Software Developer Manual. > > > > > > The turbo max freq is already percpu basis in current code, thus make > > > similar change to the max non-turbo frequency as well. > > > > > > Reported-by: Wendy Wang <wendy.wang@intel.com> > > > Signed-off-by: Chen Yu <yu.c.chen@intel.com> > > > --- > > > v2: Per Srinivas' suggestion, avoid duplicated assignment of max_pstate. > > > v3: Per Rafael's suggestion, do not add new argument in intel_pstate_get_hwp_max() > > > to avoid redundant local vars. > > > Per Srinivas' suggestion, refined the commit log to reflect the 'non-turbo' > > > max frequency. > > > > Looks good now, thanks! > > > > Is it needed in -stable and if so, which -stable series should it go into? > > > Yes, I think so, it should be > Cc: stable@vger.kernel.org # 4.18+ > as the HWP reading turbo frequency was firstly introduced in v4.18-rc2 and it > was easier to be applied. > BTW, this patch is on top of your previous patch set on intel_pstate clean up: > https://patchwork.kernel.org/project/linux-pm/list/?series=410797 OK, applied (with a few minor edits in the subject and changelog) as 5.12 material along with those patches , thanks!
diff --git a/drivers/cpufreq/intel_pstate.c b/drivers/cpufreq/intel_pstate.c index eaf32ef7a030..99e180f644c3 100644 --- a/drivers/cpufreq/intel_pstate.c +++ b/drivers/cpufreq/intel_pstate.c @@ -1724,11 +1724,9 @@ static void intel_pstate_max_within_limits(struct cpudata *cpu) static void intel_pstate_get_cpu_pstates(struct cpudata *cpu) { cpu->pstate.min_pstate = pstate_funcs.get_min(); - cpu->pstate.max_pstate = pstate_funcs.get_max(); cpu->pstate.max_pstate_physical = pstate_funcs.get_max_physical(); cpu->pstate.turbo_pstate = pstate_funcs.get_turbo(); cpu->pstate.scaling = pstate_funcs.get_scaling(); - cpu->pstate.max_freq = cpu->pstate.max_pstate * cpu->pstate.scaling; if (hwp_active && !hwp_mode_bdw) { unsigned int phy_max, current_max; @@ -1736,9 +1734,12 @@ static void intel_pstate_get_cpu_pstates(struct cpudata *cpu) intel_pstate_get_hwp_max(cpu, &phy_max, ¤t_max); cpu->pstate.turbo_freq = phy_max * cpu->pstate.scaling; cpu->pstate.turbo_pstate = phy_max; + cpu->pstate.max_pstate = HWP_GUARANTEED_PERF(READ_ONCE(cpu->hwp_cap_cached)); } else { cpu->pstate.turbo_freq = cpu->pstate.turbo_pstate * cpu->pstate.scaling; + cpu->pstate.max_pstate = pstate_funcs.get_max(); } + cpu->pstate.max_freq = cpu->pstate.max_pstate * cpu->pstate.scaling; if (pstate_funcs.get_aperf_mperf_shift) cpu->aperf_mperf_shift = pstate_funcs.get_aperf_mperf_shift();
Currently when turbo is disabled(either by BIOS or by the user), the intel_pstate driver reads the max non-turbo frequency from the package-wide MSR_PLATFORM_INFO(0xce) register. However on asymmetric platforms it is possible in theory that small and big core with HWP enabled might have different max non-turbo cpu frequency, because the MSR_HWP_CAPABILITIES is percpu scope according to Intel Software Developer Manual. The turbo max freq is already percpu basis in current code, thus make similar change to the max non-turbo frequency as well. Reported-by: Wendy Wang <wendy.wang@intel.com> Signed-off-by: Chen Yu <yu.c.chen@intel.com> --- v2: Per Srinivas' suggestion, avoid duplicated assignment of max_pstate. v3: Per Rafael's suggestion, do not add new argument in intel_pstate_get_hwp_max() to avoid redundant local vars. Per Srinivas' suggestion, refined the commit log to reflect the 'non-turbo' max frequency. -- drivers/cpufreq/intel_pstate.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)