[ Upstream commit ec1c7ad476 ]
CPUfreq governors request CPU frequencies using information
on current CPU usage. The CPPC driver converts them to
performance requests. Frequency targets are computed as:
target_freq = (util / cpu_capacity) * max_freq
target_freq is then clamped between [policy->min, policy->max].
The CPPC driver converts performance values to frequencies
(and vice-versa) using cppc_cpufreq_perf_to_khz() and
cppc_cpufreq_khz_to_perf(). These functions both use two different
factors depending on the range of the input value. For
cppc_cpufreq_khz_to_perf():
- (NOMINAL_PERF / NOMINAL_FREQ) or
- (LOWEST_PERF / LOWEST_FREQ)
and for cppc_cpufreq_perf_to_khz():
- (NOMINAL_FREQ / NOMINAL_PERF) or
- ((NOMINAL_PERF - LOWEST_FREQ) / (NOMINAL_PERF - LOWEST_PERF))
This means:
1- the functions are not inverse for some values:
(perf_to_khz(khz_to_perf(x)) != x)
2- cppc_cpufreq_perf_to_khz(LOWEST_PERF) can sometimes give
a different value from LOWEST_FREQ due to integer approximation
3- it is implied that performance and frequency are proportional
(NOMINAL_FREQ / NOMINAL_PERF) == (LOWEST_PERF / LOWEST_FREQ)
This patch changes the conversion functions to an affine function.
This fixes the 3 points above.
Suggested-by: Lukasz Luba <lukasz.luba@arm.com>
Suggested-by: Morten Rasmussen <morten.rasmussen@arm.com>
Signed-off-by: Pierre Gondois <Pierre.Gondois@arm.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
[ Upstream commit 4a8a77abf0 ]
The fuse consists of 64 bits, with this statement we're supposed to get
the upper 32 bits but it actually read out of bounds and got 0 instead
of the desired value which lead to the "PVS bin not set." codepath being
run resetting our pvs value.
Fixes: a8811ec764 ("cpufreq: qcom: Add support for krait based socs")
Signed-off-by: Luca Weiss <luca@z3ntu.xyz>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Sasha Levin <sashal@kernel.org>
In the event that the SoC is under thermal pressure while booting it's
possible for the dcvs notification to happen inbetween the cpufreq
framework calling init and it actually updating the policy's
related_cpus cpumask.
Prior to the introduction of the thermal pressure update helper an empty
cpumask would simply result in the thermal pressure of no cpus being
updated, but the new code will attempt to dereference an invalid per_cpu
variable.
Avoid this problem by using the newly reintroduced "ready" callback, to
postpone enabling the IRQ until the related_cpus cpumask is filled in.
Fixes: 0258cb19c7 ("cpufreq: qcom-cpufreq-hw: Use new thermal pressure update function")
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
This effectively revert '4bf8e582119e ("cpufreq: Remove ready()
callback")', in order to reintroduce the ready callback.
This is needed in order to be able to leave the thermal pressure
interrupts in the Qualcomm CPUfreq driver disabled during
initialization, so that it doesn't fire while related_cpus are still 0.
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
[ Viresh: Added the Chinese translation as well and updated commit msg ]
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
The AMD P-State driver is based on ACPI CPPC function, so ACPI should be
dependence of this driver in the kernel config.
In file included from ../drivers/cpufreq/amd-pstate.c:40:0:
../include/acpi/processor.h:226:2: error: unknown type name ‘phys_cpuid_t’
phys_cpuid_t phys_id; /* CPU hardware ID such as APIC ID for x86 */
^~~~~~~~~~~~
../include/acpi/processor.h:355:1: error: unknown type name ‘phys_cpuid_t’; did you mean ‘phys_addr_t’?
phys_cpuid_t acpi_get_phys_id(acpi_handle, int type, u32 acpi_id);
^~~~~~~~~~~~
phys_addr_t
CC drivers/rtc/rtc-rv3029c2.o
../include/acpi/processor.h:356:1: error: unknown type name ‘phys_cpuid_t’; did you mean ‘phys_addr_t’?
phys_cpuid_t acpi_map_madt_entry(u32 acpi_id);
^~~~~~~~~~~~
phys_addr_t
../include/acpi/processor.h:357:20: error: unknown type name ‘phys_cpuid_t’; did you mean ‘phys_addr_t’?
int acpi_map_cpuid(phys_cpuid_t phys_id, u32 acpi_id);
^~~~~~~~~~~~
phys_addr_t
See https://lore.kernel.org/lkml/20e286d4-25d7-fb6e-31a1-4349c805aae3@infradead.org/.
Reported-by: Randy Dunlap <rdunlap@infradead.org>
Reported-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Huang Rui <ray.huang@amd.com>
[ rjw: Subject edits ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Add the description of @req and @boost_supported in struct amd_cpudata
kernel-doc comment to remove warnings found by running scripts/kernel-doc,
which is caused by using 'make W=1'.
drivers/cpufreq/amd-pstate.c:104: warning: Function parameter or member
'req' not described in 'amd_cpudata'
drivers/cpufreq/amd-pstate.c:104: warning: Function parameter or member
'boost_supported' not described in 'amd_cpudata'
Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Signed-off-by: Yang Li <yang.lee@linux.alibaba.com>
Acked-by: Huang Rui <ray.huang@amd.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Introduce sysfs attributes to get the different level AMD P-State
performances.
Signed-off-by: Huang Rui <ray.huang@amd.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Introduce sysfs attributes to get the different level processor
frequencies.
Signed-off-by: Huang Rui <ray.huang@amd.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
If the sbios supports the boost mode of AMD P-State, let's switch to
boost enabled by default.
Signed-off-by: Huang Rui <ray.huang@amd.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Add trace event to monitor the performance value changes which is
controlled by cpu governors.
Signed-off-by: Huang Rui <ray.huang@amd.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
In some of Zen2 and Zen3 based processors, they are using the shared
memory that exposed from ACPI SBIOS. In this kind of the processors,
there is no MSR support, so we add acpi cppc function as the backend for
them.
It is using a module param (shared_mem) to enable related processors
manually. We will enable this by default once we address performance
issue on this solution.
Signed-off-by: Jinzhou Su <Jinzhou.Su@amd.com>
Signed-off-by: Huang Rui <ray.huang@amd.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Introduce the fast switch function for AMD P-State on the AMD processors
which support the full MSR register control. It's able to decrease the
latency on interrupt context.
Signed-off-by: Huang Rui <ray.huang@amd.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
AMD P-State is the AMD CPU performance scaling driver that introduces a
new CPU frequency control mechanism on AMD Zen based CPU series in Linux
kernel. The new mechanism is based on Collaborative processor
performance control (CPPC) which is finer grain frequency management
than legacy ACPI hardware P-States. Current AMD CPU platforms are using
the ACPI P-states driver to manage CPU frequency and clocks with
switching only in 3 P-states. AMD P-State is to replace the ACPI
P-states controls, allows a flexible, low-latency interface for the
Linux kernel to directly communicate the performance hints to hardware.
AMD P-State leverages the Linux kernel governors such as *schedutil*,
*ondemand*, etc. to manage the performance hints which are provided by CPPC
hardware functionality. The first version for AMD P-State is to support one
of the Zen3 processors, and we will support more in future after we verify
the hardware and SBIOS functionalities.
There are two types of hardware implementations for AMD P-State: one is full
MSR support and another is shared memory support. It can use
X86_FEATURE_CPPC feature flag to distinguish the different types.
Using the new AMD P-State method + kernel governors (*schedutil*,
*ondemand*, ...) to manage the frequency update is the most appropriate
bridge between AMD Zen based hardware processor and Linux kernel, the
processor is able to adjust to the most efficiency frequency according to
the kernel scheduler loading.
Please check the detailed CPU feature and MSR register description in
Processor Programming Reference (PPR) for AMD Family 19h Model 51h,
Revision A1 Processors:
https://www.amd.com/system/files/TechDocs/56569-A1-PUB.zip
Signed-off-by: Huang Rui <ray.huang@amd.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Pull ARM cpufreq updates for 5.17-rc1 from Viresh Kumar:
"- Qcom cpufreq driver updates improve irq support (Ard Biesheuvel, Stephen Boyd,
and Vladimir Zapolskiy).
- Fixes double devm_remap for mediatek driver (Hector Yuan).
- Introduces thermal pressure helpers (Lukasz Luba)."
* 'cpufreq/arm/linux-next' of git://git.kernel.org/pub/scm/linux/kernel/git/vireshk/pm:
cpufreq: mediatek-hw: Fix double devm_remap in hotplug case
cpufreq: qcom-hw: Use optional irq API
cpufreq: qcom-hw: Set CPU affinity of dcvsh interrupts
cpufreq: qcom-hw: Fix probable nested interrupt handling
cpufreq: qcom-cpufreq-hw: Avoid stack buffer for IRQ name
arch_topology: Remove unused topology_set_thermal_pressure() and related
cpufreq: qcom-cpufreq-hw: Use new thermal pressure update function
cpufreq: qcom-cpufreq-hw: Update offline CPUs per-cpu thermal pressure
thermal: cpufreq_cooling: Use new thermal pressure update function
arch_topology: Introduce thermal pressure update function
There are currently 2 ways to create a set of sysfs files for a
kobj_type, through the default_attrs field, and the default_groups
field. Move the cpufreq code to use default_groups field which has been
the preferred way since aa30f47cf6 ("kobject: Add support for default
attribute groups to kobj_type") so that we can soon get rid of the
obsolete default_attrs field.
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
When hotpluging policy cpu, cpu policy init will be called multiple times.
Unplug CPU7 -> CPU6 -> CPU5 -> CPU4, then plug CPU4 again.
In this case, devm_remap will double remap and resource allocate fail.
So replace devm_remap to ioremap and release resources in cpu policy exit.
Signed-off-by: Hector.Yuan <hector.yuan@mediatek.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
With HWP enabled, when the turbo range of performance levels is
disabled by the platform firmware, the CPU capacity is given by
the "guaranteed performance" field in MSR_HWP_CAPABILITIES which
is generally dynamic. When it changes, the kernel receives an HWP
notification interrupt handled by notify_hwp_interrupt().
When the "guaranteed performance" value changes in the above
configuration, the CPU performance scaling needs to be adjusted so
as to use the new CPU capacity in computations, which means that
the cpuinfo.max_freq value needs to be updated for that CPU.
Accordingly, modify intel_pstate_notify_work() to read
MSR_HWP_CAPABILITIES and update cpuinfo.max_freq to reflect the
new configuration (this update can be carried out even if the
configuration doesn't actually change, because it simply doesn't
matter then and it takes less time to update it than to do extra
checks to decide whether or not a change has really occurred).
Reported-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Tested-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The min and max frequency QoS requests in the cpufreq core are
initialized to whatever the current min and max frequency values are
at the init time, but if any of these values change later (for
example, cpuinfo.max_freq is updated by the driver), these initial
request values will be limiting the CPU frequency unnecessarily
unless they are changed by user space via sysfs.
To address this, initialize min_freq_req and max_freq_req to
FREQ_QOS_MIN_DEFAULT_VALUE and FREQ_QOS_MAX_DEFAULT_VALUE,
respectively, so they don't really limit anything until user
space updates them.
Reported-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Tested-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
There is an expectation from users that they can get frequency specified
by cpufreq/cpuinfo_max_freq when conditions permit. But with AlderLake
mobile it may not be possible. This is possible that frequency is clipped
based on the system power-up EPP value. In this case users can update
cpufreq/energy_performance_preference to some performance oriented EPP to
limit clipping of frequencies.
To get out of box behavior as the prior generations of CPUs, update EPP
for AlderLake mobile CPUs on boot. On prior generations of CPUs EPP = 128
was enough to get maximum frequency, but with AlderLake mobile the
equivalent EPP is 102. Since EPP is model specific, this is possible that
they have different meaning on each generation of CPU.
The current EPP string "balance_performance" corresponds to EPP = 128.
Change the EPP corresponding to "balance_performance" to 102 for only
AlderLake mobile CPUs and update this on each CPU during boot.
To implement reuse epp_values[] array and update the modified EPP at the
index for BALANCE_PERFORMANCE. Add a dummy EPP_INDEX_DEFAULT to
epp_values[] to match indexes in the energy_perf_strings[].
After HWP PM is enabled also update EPP when "balance_performance" is
redefined for the very first time after the boot on each CPU. On
subsequent suspend/resume or offline/online the old EPP is restored,
so no specific action is needed.
Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
It is not necessary to call intel_pstate_get_hwp_cap() from
intel_pstate_update_perf_limits(), because it gets called from
intel_pstate_verify_cpu_policy() which is either invoked directly
right before intel_pstate_update_perf_limits(), in
intel_cpufreq_verify_policy() in the passive mode, or called
from driver callbacks in a sequence that causes it to be followed
by an immediate intel_pstate_update_perf_limits().
Namely, in the active mode intel_cpufreq_verify_policy() is called
by intel_pstate_verify_policy() which is the ->verify() callback
routine of intel_pstate and gets called by the cpufreq core right
before intel_pstate_set_policy(), which is the driver's ->setoplicy()
callback routine, where intel_pstate_update_perf_limits() is called.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Use platform_get_irq_optional() to avoid a noisy error message when the
irq isn't specified. The irq is definitely optional given that we only
care about errors that are -EPROBE_DEFER here.
Cc: Thara Gopinath <thara.gopinath@linaro.org>
Signed-off-by: Stephen Boyd <swboyd@chromium.org>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Make the comment in blocking_notifier_call_chain() easier to
understand.
Signed-off-by: Tang Yizhou <tangyizhou@huawei.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
When I hot added a CPU, I found 'cpufreq' directory was not created
below /sys/devices/system/cpu/cpuX/.
It is because get_cpu_device() failed in add_cpu_dev_symlink().
cpufreq_add_dev() is the .add_dev callback of a CPU subsys interface.
It will be called when the CPU device registered into the system.
The call chain is as follows:
register_cpu()
->device_register()
->device_add()
->bus_probe_device()
->cpufreq_add_dev()
But only after the CPU device has been registered, we can get the
CPU device by get_cpu_device(), otherwise it will return NULL.
Since we already have the CPU device in cpufreq_add_dev(), pass
it to add_cpu_dev_symlink().
I noticed that the 'kobj' of the CPU device has been added into
the system before cpufreq_add_dev().
Fixes: 2f0ba790df ("cpufreq: Fix creation of symbolic links to policy directories")
Signed-off-by: Xiongfeng Wang <wangxiongfeng2@huawei.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Cc: All applicable <stable@vger.kernel.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
In runtime CPU cluster specific dcvsh interrupts may be handled on
unrelated CPU cores, it leads to an issue of too excessive number of
received and handled interrupts, but this is not observed, if CPU
affinity of the interrupt handler is set in accordance to CPU clusters.
The change reduces a number of received interrupts in about 10-100 times.
Signed-off-by: Vladimir Zapolskiy <vladimir.zapolskiy@linaro.org>
Reviewed-by: Matthias Kaehlcke <mka@chromium.org>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Re-enabling an interrupt from its own interrupt handler may cause
an interrupt storm, if there is a pending interrupt and because its
handling is disabled due to already done entrance into the handler
above in the stack.
Also, apparently it is improper to lock a mutex in an interrupt contex.
Fixes: 275157b367 ("cpufreq: qcom-cpufreq-hw: Add dcvs interrupt support")
Signed-off-by: Vladimir Zapolskiy <vladimir.zapolskiy@linaro.org>
Reviewed-by: Matthias Kaehlcke <mka@chromium.org>
Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Registering an IRQ requires the string buffer containing the name to
remain allocated, as the name is not copied into another buffer.
So let's add a irq_name field to the data struct instead, which is
guaranteed to have the appropriate lifetime.
Cc: Thara Gopinath <thara.gopinath@linaro.org>
Cc: Bjorn Andersson <bjorn.andersson@linaro.org>
Cc: Andy Gross <agross@kernel.org>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Reviewed-by: Thara Gopinath <thara.gopinath@linaro.org>
Tested-by: Steev Klimaszewski <steev@kali.org>
Signed-off-by: Vladimir Zapolskiy <vladimir.zapolskiy@linaro.org>
Reviewed-by: Matthias Kaehlcke <mka@chromium.org>
Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
On systems with overclocking enabled, CPPC Highest Performance can be
hard coded to 0xff. In this case even if we have cores with different
highest performance, ITMT can't be enabled as the current implementation
depends on CPPC Highest Performance.
On such systems we can use MSR_HWP_CAPABILITIES maximum performance field
when CPPC.Highest Performance is 0xff.
Due to legacy reasons, we can't solely depend on MSR_HWP_CAPABILITIES as
in some older systems CPPC Highest Performance is the only way to identify
different performing cores.
Reported-by: Michael Larabel <Michael@MichaelLarabel.com>
Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Tested-by: Michael Larabel <Michael@MichaelLarabel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
After commit 4adcf2e582 ("cpufreq: intel_pstate: Add ->offline and
->online callbacks") the EPP value set by the "performance" scaling
algorithm in the active mode is not restored after an offline/online
cycle which replaces it with the saved EPP value coming from user
space.
Address this issue by forcing intel_pstate_hwp_set() to set a new
EPP value when it runs first time after online.
Fixes: 4adcf2e582 ("cpufreq: intel_pstate: Add ->offline and ->online callbacks")
Link: https://lore.kernel.org/linux-pm/adc7132c8655bd4d1c8b6129578e931a14fe1db2.camel@linux.intel.com/
Reported-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Cc: 5.9+ <stable@vger.kernel.org> # 5.9+
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Commit fbdc21e9b0 ("cpufreq: intel_pstate: Add Icelake servers
support in no-HWP mode") enabled the use of Intel P-State driver
for Ice Lake servers.
But it doesn't cover the case when OS can't control P-States.
Therefore, for Ice Lake server, if MSR_MISC_PWR_MGMT bits 8 or 18
are enabled, then the Intel P-State driver should exit as OS can't
control P-States.
Fixes: fbdc21e9b0 ("cpufreq: intel_pstate: Add Icelake servers support in no-HWP mode")
Signed-off-by: Adamos Ttofari <attofari@amazon.de>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Thermal pressure provides a new API, which allows to use CPU frequency
as an argument. That removes the need of local conversion to capacity.
Use this new API and remove old local conversion code.
The new arch_update_thermal_pressure() also accepts boost frequencies,
which solves issue in the driver code with wrong reduced capacity
calculation. The reduced capacity was calculated wrongly due to
'policy->cpuinfo.max_freq' used as a divider. The value present there was
actually the boost frequency. Thus, even a normal maximum frequency value
which corresponds to max CPU capacity (arch_scale_cpu_capacity(cpu_id))
is not able to remove the capping.
The second side effect which is solved is that the reduced frequency wasn't
properly translated into the right reduced capacity,
e.g.
boost frequency = 3000MHz (stored in policy->cpuinfo.max_freq)
max normal frequency = 2500MHz (which is 1024 capacity)
2nd highest frequency = 2000MHz (which translates to 819 capacity)
Then in a scenario when the 'throttled_freq' max allowed frequency was
2000MHz the driver translated it into 682 capacity:
capacity = 1024 * 2000 / 3000 = 682
Then set the pressure value bigger than actually applied by the HW:
max_capacity - capacity => 1024 - 682 = 342 (<- thermal pressure)
Which was causing higher throttling and misleading task scheduler
about available CPU capacity.
A proper calculation in such case should be:
capacity = 1024 * 2000 / 2500 = 819
1024 - 819 = 205 (<- thermal pressure)
This patch relies on the new arch_update_thermal_pressure() handling
correctly such use case (with boost frequencies).
Signed-off-by: Lukasz Luba <lukasz.luba@arm.com>
Reviewed-by: Thara Gopinath <thara.gopinath@linaro.org>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
The thermal pressure signal gives information to the scheduler about
reduced CPU capacity due to thermal. It is based on a value stored in
a per-cpu 'thermal_pressure' variable. The online CPUs will get the
new value there, while the offline won't. Unfortunately, when the CPU
is back online, the value read from per-cpu variable might be wrong
(stale data). This might affect the scheduler decisions, since it
sees the CPU capacity differently than what is actually available.
Fix it by making sure that all online+offline CPUs would get the
proper value in their per-cpu variable when there is throttling
or throttling is removed.
Fixes: 275157b367 ("cpufreq: qcom-cpufreq-hw: Add dcvs interrupt support")
Reviewed-by: Thara Gopinath <thara.gopinath@linaro.org>
Signed-off-by: Lukasz Luba <lukasz.luba@arm.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
It is possible that some performance excursions happened before OS boot
or enable HWP interrupts. So clear MSR_HWP_STATUS bits when we enable
HWP interrupt. In this way a next excursion will results in a HWP
interrupt.
The status bits of MSR_HWP_STATUS must be cleared (0) by software so
that a new status condition change will cause the hardware to set the
bit again and issue the notification.
Fixes: 57577c996d ("cpufreq: intel_pstate: Process HWP Guaranteed change notification")
Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
It is possible that on some platforms HWP interrupts are disabled. In
that case accessing MSR 0x773 will result in warning.
So check X86_FEATURE_HWP_NOTIFY feature to access MSR 0x773. The other
places in code where this MSR is accessed, already checks this feature
except during disable path called during cpufreq offline and suspend
callbacks.
Fixes: 57577c996d ("cpufreq: intel_pstate: Process HWP Guaranteed change notification")
Reported-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Tested-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Commit a365ab6b9d ("cpufreq: intel_pstate: Implement the
->adjust_perf() callback") caused intel_pstate to use nonzero HWP
desired values in certain usage scenarios, but it did not prevent
them from being leaked into the confugirations in which HWP desired
is expected to be 0.
The failing scenarios are switching the driver from the passive
mode to the active mode and starting a new kernel via kexec() while
intel_pstate is running in the passive mode.
To address this issue, ensure that HWP desired will be cleared on
offline and suspend/shutdown.
Fixes: a365ab6b9d ("cpufreq: intel_pstate: Implement the ->adjust_perf() callback")
Reported-by: Julia Lawall <julia.lawall@inria.fr>
Tested-by: Julia Lawall <julia.lawall@inria.fr>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Merge Energy Model and power capping updates for 5.16-rc1:
- Add support for inefficient operating performance points to the
Energy Model and modify cpufreq to use them properly (Vincent
Donnefort).
- Rearrange the DTPM framework code to simplify it and make it easier
to follow (Daniel Lezcano).
- Fix power intialization in DTPM (Daniel Lezcano).
- Add CPU load consideration when estimating the instaneous power
consumption in DTPM (Daniel Lezcano).
* pm-em:
cpufreq: mediatek-hw: Fix cpufreq_table_find_index_dl() call
PM: EM: Mark inefficiencies in CPUFreq
cpufreq: Use CPUFREQ_RELATION_E in DVFS governors
cpufreq: Introducing CPUFREQ_RELATION_E
cpufreq: Add an interface to mark inefficient frequencies
cpufreq: Make policy min/max hard requirements
PM: EM: Allow skipping inefficient states
PM: EM: Extend em_perf_domain with a flag field
PM: EM: Mark inefficient states
PM: EM: Fix inefficient states detection
* powercap:
powercap/drivers/dtpm: Fix power limit initialization
powercap/drivers/dtpm: Scale the power with the load
powercap/drivers/dtpm: Use container_of instead of a private data field
powercap/drivers/dtpm: Simplify the dtpm table
powercap/drivers/dtpm: Encapsulate even more the code
Pull ARM cpufreq updates for 5.16-rc1 from Viresh Kumar:
"- Fix tegra driver to handle BPMP errors properly (Mikko Perttunen).
- Fix the parameter usage of the newly added perf-domain API (Hector
Yuan).
- Minor cleanups to cppc, vexpress and s3c244x drivers (Han Wang,
Guenter Roeck, and Arnd Bergmann)."
* 'cpufreq/arm/linux-next' of git://git.kernel.org/pub/scm/linux/kernel/git/vireshk/pm:
cpufreq: Fix parameter in parse_perf_domain()
cpufreq: tegra186/tegra194: Handle errors in BPMP response
cpufreq: remove useless INIT_LIST_HEAD()
cpufreq: s3c244x: add fallthrough comments for switch
cpufreq: vexpress: Drop unused variable
Fix a problem in active mode that cpu->pstate.turbo_freq is initialized
only if HWP-to-frequency scaling factor is refined.
In passive mode, this problem is not exposed, because
cpu->pstate.turbo_freq is set again, later in
intel_cpufreq_cpu_init()->intel_pstate_get_hwp_cap().
Fixes: eb3693f052 ("cpufreq: intel_pstate: hybrid: CPU-specific scaling factor")
Signed-off-by: Zhang Rui <rui.zhang@intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The new cpufreq table flag RELATION_E introduced a new "efficient"
parameter for the cpufreq_table_find*() functions.
Fixes: 1f39fa0dcc (cpufreq: Introducing CPUFREQ_RELATION_E)
Signed-off-by: Vincent Donnefort <vincent.donnefort@arm.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Let the governors schedutil, conservative and ondemand to work, if possible
on efficient frequencies only.
Signed-off-by: Vincent Donnefort <vincent.donnefort@arm.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
This newly introduced flag can be applied by a governor to a CPUFreq
relation, when looking for a frequency within the policy table. The
resolution would then only walk through efficient frequencies.
Even with the flag set, the policy max limit will still be honoured. If no
efficient frequencies can be found within the limits of the policy, an
inefficient one would be returned.
Signed-off-by: Vincent Donnefort <vincent.donnefort@arm.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
When applying the policy min/max limits, the requested frequency is
simply clamped to not be out of range. It means, however, if one of the
boundaries isn't an available frequency, the frequency resolution can
return a value out of those limits, depending on the relation used.
e.g. freq{0,1,2} being available frequencies.
freq0 policy->min freq1 policy->max freq2
| | | | |
17kHz 18kHz 19kHz 20kHz 21kHz
__resolve_freq(21kHz, CPUFREQ_RELATION_L) -> 21kHz (out of bounds)
__resolve_freq(17kHz, CPUFREQ_RELATION_H) -> 17kHz (out of bounds)
If, during the policy init, we resolve the requested min/max to existing
frequencies, we ensure that any CPUFREQ_RELATION_* would resolve to a
frequency which is inside the policy min/max range.
Making the policy limits rigid helps to introduce the inefficient
frequencies support. Resolving an inefficient frequency to an efficient
one should not transgress policy->max (which can be set for thermal
reason) and having a value we can trust simplify this comparison.
Signed-off-by: Vincent Donnefort <vincent.donnefort@arm.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
It is possible that HWP guaranteed ratio is changed in response to
change in power and thermal limits. For example when Intel Speed Select
performance profile is changed or there is change in TDP, hardware can
send notifications. It is possible that the guaranteed ratio is
increased. This creates an issue when turbo is disabled, as the old
limits set in MSR_HWP_REQUEST are still lower and hardware will clip
to older limits.
This change enables HWP interrupt and process HWP interrupts. When
guaranteed is changed, calls cpufreq_update_policy() so that driver
callbacks are called to update to new HWP limits. This callback
is called from a delayed workqueue of 10ms to avoid frequent updates.
Although the scope of IA32_HWP_INTERRUPT is per logical cpu, on some
plaforms interrupt is generated on all CPUs. This is particularly a
problem during initialization, when the driver didn't allocated
data for other CPUs. So this change uses a cpumask of enabled CPUs and
process interrupts on those CPUs only.
When the cpufreq offline() or suspend() callback is called, HWP interrupt
is disabled on those CPUs and also cancels any pending work item.
Spin lock is used to protect data and processing shared with interrupt
handler. Here READ_ONCE(), WRITE_ONCE() macros are used to designate
shared data, even though spin lock act as an optimization barrier here.
Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Tested-by: pablomh@gmail.com
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The return value from tegra_bpmp_transfer indicates the success or
failure of the IPC transaction with BPMP. If the transaction
succeeded, we also need to check the actual command's result code.
Add code to do this.
While at it, explicitly handle missing CPU clusters, which can
occur on floorswept chips. This worked before as well, but
possibly only by accident.
Signed-off-by: Mikko Perttunen <mperttunen@nvidia.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
list cpu_data_list has been inited staticly through LIST_HEAD,
so there's no need to call another INIT_LIST_HEAD. Simply remove
it from cppc_cpufreq_init.
Signed-off-by: Han Wang <zjuwanghan@outlook.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Apparently nobody has so far caught this warning, I hit it in randconfig
build testing:
drivers/cpufreq/s3c2440-cpufreq.c: In function 's3c2440_cpufreq_setdivs':
drivers/cpufreq/s3c2440-cpufreq.c:175:10: error: this statement may fall through [-Werror=implicit-fallthrough=]
camdiv |= S3C2440_CAMDIVN_HCLK3_HALF;
^
drivers/cpufreq/s3c2440-cpufreq.c:176:2: note: here
case 3:
^~~~
drivers/cpufreq/s3c2440-cpufreq.c:181:10: error: this statement may fall through [-Werror=implicit-fallthrough=]
camdiv |= S3C2440_CAMDIVN_HCLK4_HALF;
^
drivers/cpufreq/s3c2440-cpufreq.c:182:2: note: here
case 4:
^~~~
Both look like the fallthrough is intentional, so add the new
"fallthrough;" keyword.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: Krzysztof Kozlowski <krzysztof.kozlowski@canonical.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
- Prevent intel_pstate from avoiding to use HWP, even if instructed
to do so via the kernel command line, when HWP has been enabled
already by the platform firmware (Doug Smythies).
- Prevent use-after-free from occurring in the schedutil cpufreq
governor on exit by fixing a core helper function that attempts
to access memory associated with a kobject after calling
kobject_put() on it (James Morse).
-----BEGIN PGP SIGNATURE-----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=ytGW
-----END PGP SIGNATURE-----
Merge tag 'pm-5.15-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm
Pull power management fixes from Rafael Wysocki:
"These fix two cpufreq issues, one in the intel_pstate driver and one
in the core.
Specifics:
- Prevent intel_pstate from avoiding to use HWP, even if instructed
to do so via the kernel command line, when HWP has been enabled
already by the platform firmware (Doug Smythies).
- Prevent use-after-free from occurring in the schedutil cpufreq
governor on exit by fixing a core helper function that attempts to
access memory associated with a kobject after calling kobject_put()
on it (James Morse)"
* tag 'pm-5.15-rc2' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
cpufreq: schedutil: Destroy mutex before kobject_put() frees the memory
cpufreq: intel_pstate: Override parameters if HWP forced by BIOS
Since commit e5c6b312ce ("cpufreq: schedutil: Use kobject release()
method to free sugov_tunables") kobject_put() has kfree()d the
attr_set before gov_attr_set_put() returns.
kobject_put() isn't the last user of attr_set in gov_attr_set_put(),
the subsequent mutex_destroy() triggers a use-after-free:
| BUG: KASAN: use-after-free in mutex_is_locked+0x20/0x60
| Read of size 8 at addr ffff000800ca4250 by task cpuhp/2/20
|
| CPU: 2 PID: 20 Comm: cpuhp/2 Not tainted 5.15.0-rc1 #12369
| Hardware name: ARM LTD ARM Juno Development Platform/ARM Juno Development
| Platform, BIOS EDK II Jul 30 2018
| Call trace:
| dump_backtrace+0x0/0x380
| show_stack+0x1c/0x30
| dump_stack_lvl+0x8c/0xb8
| print_address_description.constprop.0+0x74/0x2b8
| kasan_report+0x1f4/0x210
| kasan_check_range+0xfc/0x1a4
| __kasan_check_read+0x38/0x60
| mutex_is_locked+0x20/0x60
| mutex_destroy+0x80/0x100
| gov_attr_set_put+0xfc/0x150
| sugov_exit+0x78/0x190
| cpufreq_offline.isra.0+0x2c0/0x660
| cpuhp_cpufreq_offline+0x14/0x24
| cpuhp_invoke_callback+0x430/0x6d0
| cpuhp_thread_fun+0x1b0/0x624
| smpboot_thread_fn+0x5e0/0xa6c
| kthread+0x3a0/0x450
| ret_from_fork+0x10/0x20
Swap the order of the calls.
Fixes: e5c6b312ce ("cpufreq: schedutil: Use kobject release() method to free sugov_tunables")
Cc: 4.7+ <stable@vger.kernel.org> # 4.7+
Signed-off-by: James Morse <james.morse@arm.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
If HWP has been already been enabled by BIOS, it may be
necessary to override some kernel command line parameters.
Once it has been enabled it requires a reset to be disabled.
Suggested-by: Rafael J. Wysocki <rafael@kernel.org>
Signed-off-by: Doug Smythies <dsmythies@telus.net>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Pierre Gondois