linux/drivers/cpufreq, branch v5.11

cpufreq: ACPI: Update arch scale-invariance max perf ratio if CPPC is not there

2021-02-08T12:45:51Z

If the maximum performance level taken for computing the arch_max_freq_ratio value used in the x86 scale-invariance code is higher than the one corresponding to the cpuinfo.max_freq value coming from the acpi_cpufreq driver, the scale-invariant utilization falls below 100% even if the CPU runs at cpuinfo.max_freq or slightly faster, which causes the schedutil governor to select a frequency below cpuinfo.max_freq. That frequency corresponds to a frequency table entry below the maximum performance level necessary to get to the "boost" range of CPU frequencies which prevents "boost" frequencies from being used in some workloads. While this issue is related to scale-invariance, it may be amplified by commit db865272d9c4 ("cpufreq: Avoid configuring old governors as default with intel_pstate") from the 5.10 development cycle which made it extremely easy to default to schedutil even if the preferred driver is acpi_cpufreq as long as intel_pstate is built too, because the mere presence of the latter effectively removes the ondemand governor from the defaults. Distro kernels are likely to include both intel_pstate and acpi_cpufreq on x86, so their users who cannot use intel_pstate or choose to use acpi_cpufreq may easily be affectecd by this issue. If CPPC is available, it can be used to address this issue by extending the frequency tables created by acpi_cpufreq to cover the entire available frequency range (including "boost" frequencies) for each CPU, but if CPPC is not there, acpi_cpufreq has no idea what the maximum "boost" frequency is and the frequency tables created by it cannot be extended in a meaningful way, so in that case make it ask the arch scale-invariance code to to use the "nominal" performance level for CPU utilization scaling in order to avoid the issue at hand. Fixes: db865272d9c4 ("cpufreq: Avoid configuring old governors as default with intel_pstate") Signed-off-by: Rafael J. Wysocki Reviewed-by: Giovanni Gherdovich Acked-by: Peter Zijlstra (Intel)

cpufreq: ACPI: Extend frequency tables to cover boost frequencies

2021-02-08T12:45:51Z

A severe performance regression on AMD EPYC processors when using the schedutil scaling governor was discovered by Phoronix.com and attributed to the following commits: 41ea667227ba ("x86, sched: Calculate frequency invariance for AMD systems") 976df7e5730e ("x86, sched: Use midpoint of max_boost and max_P for frequency invariance on AMD EPYC") The source of the problem is that the maximum performance level taken for computing the arch_max_freq_ratio value used in the x86 scale- invariance code is higher than the one corresponding to the cpuinfo.max_freq value coming from the acpi_cpufreq driver. This effectively causes the scale-invariant utilization to fall below 100% even if the CPU runs at cpuinfo.max_freq or slightly faster, so the schedutil governor selects a frequency below cpuinfo.max_freq then. That frequency corresponds to a frequency table entry below the maximum performance level necessary to get to the "boost" range of CPU frequencies. However, if the cpuinfo.max_freq value coming from acpi_cpufreq was higher, the schedutil governor would select higher frequencies which in turn would allow acpi_cpufreq to set more adequate performance levels and to get to the "boost" range of CPU frequencies more often. This issue affects any systems where acpi_cpufreq is used and the "boost" (or "turbo") frequencies are enabled, not just AMD EPYC. Moreover, commit db865272d9c4 ("cpufreq: Avoid configuring old governors as default with intel_pstate") from the 5.10 development cycle made it extremely easy to default to schedutil even if the preferred driver is acpi_cpufreq as long as intel_pstate is built too, because the mere presence of the latter effectively removes the ondemand governor from the defaults. Distro kernels are likely to include both intel_pstate and acpi_cpufreq on x86, so their users who cannot use intel_pstate or choose to use acpi_cpufreq may easily be affectecd by this issue. To address this issue, extend the frequency table constructed by acpi_cpufreq for each CPU to cover the entire range of available frequencies (including the "boost" ones) if CPPC is available and indicates that "boost" (or "turbo") frequencies are enabled. That causes cpuinfo.max_freq to become the maximum "boost" frequency of the given CPU (instead of the maximum frequency returned by the ACPI _PSS object that corresponds to the "nominal" performance level). Fixes: 41ea667227ba ("x86, sched: Calculate frequency invariance for AMD systems") Fixes: 976df7e5730e ("x86, sched: Use midpoint of max_boost and max_P for frequency invariance on AMD EPYC") Fixes: db865272d9c4 ("cpufreq: Avoid configuring old governors as default with intel_pstate") Link: https://www.phoronix.com/scan.php?page=article&item=linux511-amd-schedutil&num=1 Link: https://lore.kernel.org/linux-pm/20210203135321.12253-2-ggherdovich@suse.cz/ Reported-by: Michael Larabel Diagnosed-by: Giovanni Gherdovich Signed-off-by: Rafael J. Wysocki Tested-by: Giovanni Gherdovich Reviewed-by: Giovanni Gherdovich Tested-by: Michael Larabel

cpufreq: intel_pstate: remove obsolete functions

2021-01-07T17:22:46Z

percent_fp() was used in intel_pstate_pid_reset(), which was removed in commit 9d0ef7af1f2d ("cpufreq: intel_pstate: Do not use PID-based P-state selection") and hence, percent_fp() is unused since then. percent_ext_fp() was last used in intel_pstate_update_perf_limits(), which was refactored in commit 1a4fe38add8b ("cpufreq: intel_pstate: Remove max/min fractions to limit performance"), and hence, percent_ext_fp() is unused since then. make CC=clang W=1 points us those unused functions: drivers/cpufreq/intel_pstate.c:79:23: warning: unused function 'percent_fp' [-Wunused-function] static inline int32_t percent_fp(int percent) ^ drivers/cpufreq/intel_pstate.c:94:23: warning: unused function 'percent_ext_fp' [-Wunused-function] static inline int32_t percent_ext_fp(int percent) ^ Remove those obsolete functions. Signed-off-by: Lukas Bulwahn Reviewed-by: Nathan Chancellor Signed-off-by: Rafael J. Wysocki

cpufreq: powernow-k8: pass policy rather than use cpufreq_cpu_get()

2021-01-07T16:37:33Z

Currently there is an unlikely case where cpufreq_cpu_get() returns a NULL policy and this will cause a NULL pointer dereference later on. Fix this by passing the policy to transition_frequency_fidvid() from the caller and hence eliminating the need for the cpufreq_cpu_get() and cpufreq_cpu_put(). Thanks to Viresh Kumar for suggesting the fix. Addresses-Coverity: ("Dereference null return") Fixes: b43a7ffbf33b ("cpufreq: Notify all policy->cpus in cpufreq_notify_transition()") Suggested-by: Viresh Kumar Signed-off-by: Colin Ian King Acked-by: Viresh Kumar Signed-off-by: Rafael J. Wysocki

cpufreq: intel_pstate: Use HWP capabilities in intel_cpufreq_adjust_perf()

2021-01-07T16:34:32Z

If turbo P-states cannot be used, either due to the configuration of the processor, or because intel_pstate is not allowed to used them, the maximum available P-state with HWP enabled corresponds to the HWP_CAP.GUARANTEED value which is not static. It can be adjusted by an out-of-band agent or during an Intel Speed Select performance level change, so long as it remains less than or equal to HWP_CAP.MAX. However, if turbo P-states cannot be used, intel_cpufreq_adjust_perf() always uses pstate.max_pstate (set during the initialization of the driver only) as the maximum available P-state, so it may miss a change of the HWP_CAP.GUARANTEED value. Prevent that from happening by modifyig intel_cpufreq_adjust_perf() to always read the "guaranteed" and "maximum turbo" performance levels from the cached HWP_CAP value. Fixes: a365ab6b9dfb ("cpufreq: intel_pstate: Implement the ->adjust_perf() callback") Signed-off-by: Rafael J. Wysocki Acked-by: Srinivas Pandruvada

cpufreq: intel_pstate: Fix fast-switch fallback path

2020-12-30T17:22:17Z

When sugov_update_single_perf() falls back to the "frequency" path due to the missing scale-invariance, it will call cpufreq_driver_fast_switch() via sugov_fast_switch() and the driver's ->fast_switch() callback will be invoked, so it must not be NULL. However, after commit a365ab6b9dfb ("cpufreq: intel_pstate: Implement the ->adjust_perf() callback") intel_pstate sets ->fast_switch() to NULL when it is going to use intel_cpufreq_adjust_perf(), which is a mistake, because on x86 the scale-invariance may be turned off dynamically, so modify it to retain the original ->adjust_perf() callback pointer. Fixes: a365ab6b9dfb ("cpufreq: intel_pstate: Implement the ->adjust_perf() callback") Reported-by: Kenneth R. Crudup Tested-by: Kenneth R. Crudup Signed-off-by: Rafael J. Wysocki

Merge branch 'pm-cpufreq'

2020-12-22T16:59:11Z

* pm-cpufreq: cpufreq: intel_pstate: Use most recent guaranteed performance values cpufreq: intel_pstate: Implement the ->adjust_perf() callback cpufreq: Add special-purpose fast-switching callback for drivers cpufreq: schedutil: Add util to struct sg_cpu cppc_cpufreq: replace per-cpu data array with a list cppc_cpufreq: expose information on frequency domains cppc_cpufreq: clarify support for coordination types cppc_cpufreq: use policy->cpu as driver of frequency setting ACPI: processor: fix NONE coordination for domain mapping failure ACPI: processor: Drop duplicate setting of shared_cpu_map

cpufreq: intel_pstate: Use most recent guaranteed performance values

2020-12-21T09:51:00Z

When turbo has been disabled by the BIOS, but HWP_CAP.GUARANTEED is changed later, user space may want to take advantage of this increased guaranteed performance. HWP_CAP.GUARANTEED is not a static value. It can be adjusted by an out-of-band agent or during an Intel Speed Select performance level change. The HWP_CAP.MAX is still the maximum achievable performance with turbo disabled by the BIOS, so HWP_CAP.GUARANTEED can still change as long as it remains less than or equal to HWP_CAP.MAX. When HWP_CAP.GUARANTEED is changed, the sysfs base_frequency attribute shows the most recent guaranteed frequency value. This attribute can be used by user space software to update the scaling min/max limits of the CPU. Currently, the ->setpolicy() callback already uses the latest HWP_CAP values when setting HWP_REQ, but the ->verify() callback will restrict the user settings to the to old guaranteed performance value which prevents user space from making use of the extra CPU capacity theoretically available to it after increasing HWP_CAP.GUARANTEED. To address this, read HWP_CAP in intel_pstate_verify_cpu_policy() to obtain the maximum P-state that can be used and use that to confine the policy max limit instead of using the cached and possibly stale pstate.max_freq value for this purpose. For consistency, update intel_pstate_update_perf_limits() to use the maximum available P-state returned by intel_pstate_get_hwp_max() to compute the maximum frequency instead of using the return value of intel_pstate_get_max_freq() which, again, may be stale. This issue is a side-effect of fixing the scaling frequency limits in commit eacc9c5a927e ("cpufreq: intel_pstate: Fix intel_pstate_get_hwp_max() for turbo disabled") which corrected the setting of the reduced scaling frequency values, but caused stale HWP_CAP.GUARANTEED to be used in the case at hand. Fixes: eacc9c5a927e ("cpufreq: intel_pstate: Fix intel_pstate_get_hwp_max() for turbo disabled") Reported-by: Srinivas Pandruvada Tested-by: Srinivas Pandruvada Cc: 5.8+ # 5.8+ Signed-off-by: Rafael J. Wysocki

cpufreq: intel_pstate: Implement the ->adjust_perf() callback

2020-12-15T18:24:19Z

Make intel_pstate expose the ->adjust_perf() callback when it operates in the passive mode with HWP enabled which causes the schedutil governor to use that callback instead of ->fast_switch(). The minimum and target performance-level values passed by the governor to ->adjust_perf() are converted to HWP.REQ.MIN and HWP.REQ.DESIRED, respectively, which allows the processor to adjust its configuration to maximize energy-efficiency while providing sufficient capacity. Signed-off-by: Rafael J. Wysocki Acked-by: Srinivas Pandruvada Acked-by: Viresh Kumar

cpufreq: Add special-purpose fast-switching callback for drivers

2020-12-15T18:24:18Z

First off, some cpufreq drivers (eg. intel_pstate) can pass hints beyond the current target frequency to the hardware and there are no provisions for doing that in the cpufreq framework. In particular, today the driver has to assume that it should not allow the frequency to fall below the one requested by the governor (or the required capacity may not be provided) which may not be the case and which may lead to excessive energy usage in some scenarios. Second, the hints passed by these drivers to the hardware need not be in terms of the frequency, so representing the utilization numbers coming from the scheduler as frequency before passing them to those drivers is not really useful. Address the two points above by adding a special-purpose replacement for the ->fast_switch callback, called ->adjust_perf, allowing the governor to pass abstract performance level (rather than frequency) values for the minimum (required) and target (desired) performance along with the CPU capacity to compare them to. Also update the schedutil governor to use the new callback instead of ->fast_switch if present and if the utilization mertics are frequency-invariant (that is requisite for the direct mapping between the utilization and the CPU performance levels to be a reasonable approximation). Signed-off-by: Rafael J. Wysocki Acked-by: Viresh Kumar