The communication mean of the _CPC desired performance can be
PCC, System Memory, System IO, or Functional Fixed Hardware (FFH).
PCC, SystemMemory and SystemIo address spaces are available from any
CPU. Thus, dvfs_possible_from_any_cpu should be enabled in such case.
For FFH, let the FFH implementation do smp_call_function_*() calls.
Signed-off-by: Pierre Gondois <pierre.gondois@arm.com>
Reviewed-by: Sudeep Holla <sudeep.holla@arm.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The communication mean of the _CPC desired performance can be
PCC, System Memory, System IO, or Functional Fixed Hardware.
commit b7898fda5b ("cpufreq: Support for fast frequency switching")
fast_switching is 'for switching CPU frequencies from interrupt
context'.
Writes to SystemMemory and SystemIo are fast and suitable this.
This is not the case for PCC and might not be the case for FFH.
Enable fast_switching for the cppc_cpufreq driver in above cases.
Add cppc_allow_fast_switch() to check the desired performance
register address space and set fast_switching accordingly.
Signed-off-by: Pierre Gondois <pierre.gondois@arm.com>
Reviewed-by: Sudeep Holla <sudeep.holla@arm.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
cpufreq_offline() calls offline() and exit() under the policy rwsem
But they are called outside the rwsem in cpufreq_online().
Make cpufreq_online() call offline() and exit() as well as online() and
init() under the policy rwsem to achieve a clear lock relationship.
All of the init() and online() implementations in the tree only
initialize the policy object without attempting to acquire the policy
rwsem and they won't call cpufreq APIs attempting to acquire it.
Signed-off-by: Schspa Shi <schspa@gmail.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
[ rjw: Changelog edits ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
If policy initialization fails after the sysfs files are created,
there is a possibility to end up running show()/store() callbacks
for half-initialized policies, which may have unpredictable
outcomes.
Abort show()/store() in such a case by making sure the policy is active.
Also dectivate the policy on such failures.
Signed-off-by: Schspa Shi <schspa@gmail.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
[ rjw: Subject and changelog edits ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Currently, cpufreq_remove_dev() invokes the ->exit() driver callback
without holding the policy rwsem which is inconsistent with what
happens if ->exit() is invoked directly from cpufreq_offline().
It also manipulates the real_cpus mask and removes the CPU device
symlink without holding the policy rwsem, but cpufreq_offline() holds
the rwsem around the modifications thereof.
For consistency, modify cpufreq_remove_dev() to hold the policy rwsem
until the ->exit() callback has been called (or it has been determined
that it is not necessary to call it).
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Split the "core" part running under the policy rwsem out of
cpufreq_offline() to allow the locking in cpufreq_remove_dev() to be
rearranged more easily.
As a side-effect this eliminates the unlock label that's not needed
any more.
No expected functional impact.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Notice that cpufreq_offline() only needs to check policy_is_inactive()
once and rearrange the code in there to make that happen.
No expected functional impact.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
add_cpu_dev_symlink() is responsible for setting the CPUs in the
real_cpus mask, the reverse of which should be done from
remove_cpu_dev_symlink() to make it look clean and avoid any breakage
later on.
Move the call to clear the mask to remove_cpu_dev_symlink().
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Prevent intel_pstate to load when OOB (Out Of Band) P-states mode is
enabled in Sapphire Rapids. The OOB identifying bits are same as the
prior generation CPUs like Ice Lake servers. So, also add Sapphire
Rapids to intel_pstate_cpu_oob_ids list.
Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
This reverts commit f346e96267.
The commit tried to fix a possible real bug but it made it even worse.
The fix was simply buggy as now an error out to out_offline_policy or
out_exit_policy will try to release a semaphore which was never taken in
the first place. This works fine only if we failed late, i.e. via
out_destroy_policy.
Fixes: f346e96267 ("cpufreq: Fix possible race in cpufreq online error path")
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Performance states and energy consumption values are not advertised
in ACPI. In the GicC structure of the MADT table, the "Processor
Power Efficiency Class field" (called efficiency class from now)
allows to describe the relative energy efficiency of CPUs.
To leverage the EM and EAS, the CPPC driver creates a set of
artificial performance states and registers them in the Energy Model
(EM), such as:
- Every 20 capacity unit, a performance state is created.
- The energy cost of each performance state gradually increases.
No power value is generated as only the cost is used in the EM.
During task placement, a task can raise the frequency of its whole
pd. This can make EAS place a task on a pd with CPUs that are
individually less energy efficient.
As cost values are artificial, and to place tasks on CPUs with the
lower efficiency class, a gap in cost values is generated for adjacent
efficiency classes.
E.g.:
- efficiency class = 0, capacity is in [0-1024], so cost values
are in [0: 51] (one performance state every 20 capacity unit)
- efficiency class = 1, capacity is in [0-1024], cost values
are in [1*gap+0: 1*gap+51].
The value of the cost gap is chosen to absorb a the energy of 4 CPUs
at their maximum capacity. This means that between:
1- a pd of 4 CPUs, each of them being used at almost their full
capacity. Their efficiency class is N.
2- a CPU using almost none of its capacity. Its efficiency class is
N+1
EAS will choose the first option.
This patch also populates the (struct cpufreq_driver).register_em
callback if the valid efficiency_class ACPI values are provided.
Signed-off-by: Pierre Gondois <Pierre.Gondois@arm.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
In ACPI, describing power efficiency of CPUs can be done through the
following arm specific field:
ACPI 6.4, s5.2.12.14 'GIC CPU Interface (GICC) Structure',
'Processor Power Efficiency Class field':
Describes the relative power efficiency of the associated pro-
cessor. Lower efficiency class numbers are more efficient than
higher ones (e.g. efficiency class 0 should be treated as more
efficient than efficiency class 1). However, absolute values
of this number have no meaning: 2 isn’t necessarily half as
efficient as 1.
The efficiency_class field is stored in the GicC structure of the
ACPI MADT table and it's currently supported in Linux for arm64 only.
Thus, this new functionality is introduced for arm64 only.
To allow the cppc_cpufreq driver to know and preprocess the
efficiency_class values of all the CPUs, add a per_cpu efficiency_class
variable to store them.
At least 2 different efficiency classes must be present,
otherwise there is no use in creating an Energy Model.
The efficiency_class values are squeezed in [0:#efficiency_class-1]
while conserving the order. For instance, efficiency classes of:
[111, 212, 250]
will be mapped to:
[0 (was 111), 1 (was 212), 2 (was 250)].
Each policy being independently registered in the driver, populating
the per_cpu efficiency_class is done only once at the driver
initialization. This prevents from having each policy re-searching the
efficiency_class values of other CPUs. The EM will be registered in a
following patch.
The patch also exports acpi_cpu_get_madt_gicc() to fetch the GicC
structure of the ACPI MADT table for each CPU.
Acked-by: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Pierre Gondois <Pierre.Gondois@arm.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
For some platforms, the frequency returned by hardware may be slightly
different from what is provided in the frequency table. For example,
hardware may return 499 MHz instead of 500 MHz. In such cases it is
better to avoid getting into unnecessary frequency updates, as we may
end up switching policy->cur between the two and sending unnecessary
pre/post update notifications, etc.
This patch has chosen allows the hardware frequency and table frequency
to deviate by 1 MHz for now, we may want to increase it a bit later on
if someone still complains.
Reported-by: Rex-BC Chen <rex-bc.chen@mediatek.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Tested-by: Jia-wei Chang <jia-wei.chang@mediatek.com>
Reviewed-by: Matthias Brugger <matthias.bgg@gmail.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
It's noted that dcvs interrupts are not self-clearing, thus an interrupt
handler runs constantly, which leads to a severe regression in runtime.
To fix the problem an explicit write to clear interrupt register is
required, note that on OSM platforms the register may not be present.
Fixes: 275157b367 ("cpufreq: qcom-cpufreq-hw: Add dcvs interrupt support")
Signed-off-by: Vladimir Zapolskiy <vladimir.zapolskiy@linaro.org>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
When cpufreq online fails, the policy->cpus mask is not cleared and
policy->rwsem is released too early, so the driver can be invoked
via the cpuinfo_cur_freq sysfs attribute while its ->offline() or
->exit() callbacks are being run.
Take policy->clk as an example:
static int cpufreq_online(unsigned int cpu)
{
...
// policy->cpus != 0 at this time
down_write(&policy->rwsem);
ret = cpufreq_add_dev_interface(policy);
up_write(&policy->rwsem);
return 0;
out_destroy_policy:
for_each_cpu(j, policy->real_cpus)
remove_cpu_dev_symlink(policy, get_cpu_device(j));
up_write(&policy->rwsem);
...
out_exit_policy:
if (cpufreq_driver->exit)
cpufreq_driver->exit(policy);
clk_put(policy->clk);
// policy->clk is a wild pointer
...
^
|
Another process access
__cpufreq_get
cpufreq_verify_current_freq
cpufreq_generic_get
// acces wild pointer of policy->clk;
|
|
out_offline_policy: |
cpufreq_policy_free(policy); |
// deleted here, and will wait for no body reference
cpufreq_policy_put_kobj(policy);
}
Address this by modifying cpufreq_online() to release policy->rwsem
in the error path after the driver callbacks have run and to clear
policy->cpus before releasing the semaphore.
Fixes: 7106e02bae ("cpufreq: release policy->rwsem on error")
Signed-off-by: Schspa Shi <schspa@gmail.com>
[ rjw: Subject and changelog edits ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Problem statement:
Once the user has disabled turbo frequency by
# echo 1 > /sys/devices/system/cpu/intel_pstate/no_turbo
the cfs_rq's util_avg becomes quite small when compared with
CPU capacity.
Step to reproduce:
# echo 1 > /sys/devices/system/cpu/intel_pstate/no_turbo
# ./x86_cpuload --count 1 --start 3 --timeout 100 --busy 99
would launch 1 thread and bind it to CPU3, lasting for 100 seconds,
with a CPU utilization of 99%. [1]
top result:
%Cpu3 : 98.4 us, 0.0 sy, 0.0 ni, 1.6 id, 0.0 wa, 0.0 hi, 0.0 si, 0.0 st
check util_avg:
cat /sys/kernel/debug/sched/debug | grep "cfs_rq\[3\]" -A 20 | grep util_avg
.util_avg : 611
So the util_avg/cpu capacity is 611/1024, which is much smaller than
98.4% shown in the top result.
This might impact some logic in the scheduler. For example,
group_is_overloaded() would compare the group_capacity and group_util
in the sched group, to check if this sched group is overloaded or not.
With this gap, even when there is a nearly 100% workload, the sched
group will not be regarded as overloaded. Besides group_is_overloaded(),
there are also other victims. There is a ongoing work that aims to
optimize the task wakeup in a LLC domain. The main idea is to stop
searching idle CPUs if the sched domain is overloaded[2]. This proposal
also relies on the util_avg/CPU capacity to decide whether the LLC
domain is overloaded.
Analysis:
CPU frequency invariance has caused this difference. In summary,
the util_sum of cfs rq would decay quite fast when the CPU is in
idle, when the CPU frequency invariance is enabled.
The detail is as followed:
As depicted in update_rq_clock_pelt(), when the frequency invariance
is enabled, there would be two clock variables on each rq, clock_task
and clock_pelt:
The clock_pelt scales the time to reflect the effective amount of
computation done during the running delta time but then syncs back to
clock_task when rq is idle.
absolute time | 1| 2| 3| 4| 5| 6| 7| 8| 9|10|11|12|13|14|15|16
@ max frequency ------******---------------******---------------
@ half frequency ------************---------************---------
clock pelt | 1| 2| 3| 4| 7| 8| 9| 10| 11|14|15|16
The fast decay of util_sum during idle is due to:
1. rq->clock_pelt is always behind rq->clock_task
2. rq->last_update is updated to rq->clock_pelt' after invoking
___update_load_sum()
3. Then the CPU becomes idle, the rq->clock_pelt' would be suddenly
increased a lot to rq->clock_task
4. Enters ___update_load_sum() again, the idle period is calculated by
rq->clock_task - rq->last_update, AKA, rq->clock_task - rq->clock_pelt'.
The lower the CPU frequency is, the larger the delta =
rq->clock_task - rq->clock_pelt' will be. Since the idle period will be
used to decay the util_sum only, the util_sum drops significantly during
idle period.
Proposal:
This symptom is not only caused by disabling turbo frequency, but it
would also appear if the user limits the max frequency at runtime.
Because, if the frequency is always lower than the max frequency,
CPU frequency invariance would decay the util_sum quite fast during
idle.
As some end users would disable turbo after boot up, this patch aims to
present this symptom and deals with turbo scenarios for now.
It might be ideal if CPU frequency invariance is aware of the max CPU
frequency (user specified) at runtime in the future.
Link: https://github.com/yu-chen-surf/x86_cpuload.git#1
Link: https://lore.kernel.org/lkml/20220310005228.11737-1-yu.c.chen@intel.com/#2
Signed-off-by: Chen Yu <yu.c.chen@intel.com>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Giovanni Gherdovich <ggherdovich@suse.cz>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
powerpc's asm/prom.h brings some headers that it doesn't
need itself.
In order to clean it up, first add missing headers in
users of asm/prom.h
Signed-off-by: Christophe Leroy <christophe.leroy@csgroup.eu>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The .active_power() callback passes the device pointer when it's called.
Aligned with a convetion present in other subsystems and pass the 'dev'
as a first argument. It looks more cleaner.
Adjust all affected drivers which implement that API callback.
Suggested-by: Ionela Voinescu <ionela.voinescu@arm.com>
Signed-off-by: Lukasz Luba <lukasz.luba@arm.com>
Reviewed-by: Ionela Voinescu <ionela.voinescu@arm.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The struct dbs_data embeds a struct gov_attr_set and
the struct gov_attr_set embeds a kobject. Since every kobject must have
a release() method and we can't use kfree() to free it directly,
so introduce cpufreq_dbs_data_release() to release the dbs_data via
the kobject::release() method. This fixes the calltrace like below:
ODEBUG: free active (active state 0) object type: timer_list hint: delayed_work_timer_fn+0x0/0x34
WARNING: CPU: 12 PID: 810 at lib/debugobjects.c:505 debug_print_object+0xb8/0x100
Modules linked in:
CPU: 12 PID: 810 Comm: sh Not tainted 5.16.0-next-20220120-yocto-standard+ #536
Hardware name: Marvell OcteonTX CN96XX board (DT)
pstate: 60400009 (nZCv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
pc : debug_print_object+0xb8/0x100
lr : debug_print_object+0xb8/0x100
sp : ffff80001dfcf9a0
x29: ffff80001dfcf9a0 x28: 0000000000000001 x27: ffff0001464f0000
x26: 0000000000000000 x25: ffff8000090e3f00 x24: ffff80000af60210
x23: ffff8000094dfb78 x22: ffff8000090e3f00 x21: ffff0001080b7118
x20: ffff80000aeb2430 x19: ffff800009e8f5e0 x18: 0000000000000000
x17: 0000000000000002 x16: 00004d62e58be040 x15: 013590470523aff8
x14: ffff8000090e1828 x13: 0000000001359047 x12: 00000000f5257d14
x11: 0000000000040591 x10: 0000000066c1ffea x9 : ffff8000080d15e0
x8 : ffff80000a1765a8 x7 : 0000000000000000 x6 : 0000000000000001
x5 : ffff800009e8c000 x4 : ffff800009e8c760 x3 : 0000000000000000
x2 : 0000000000000000 x1 : 0000000000000000 x0 : ffff0001474ed040
Call trace:
debug_print_object+0xb8/0x100
__debug_check_no_obj_freed+0x1d0/0x25c
debug_check_no_obj_freed+0x24/0xa0
kfree+0x11c/0x440
cpufreq_dbs_governor_exit+0xa8/0xac
cpufreq_exit_governor+0x44/0x90
cpufreq_set_policy+0x29c/0x570
store_scaling_governor+0x110/0x154
store+0xb0/0xe0
sysfs_kf_write+0x58/0x84
kernfs_fop_write_iter+0x12c/0x1c0
new_sync_write+0xf0/0x18c
vfs_write+0x1cc/0x220
ksys_write+0x74/0x100
__arm64_sys_write+0x28/0x3c
invoke_syscall.constprop.0+0x58/0xf0
do_el0_svc+0x70/0x170
el0_svc+0x54/0x190
el0t_64_sync_handler+0xa4/0x130
el0t_64_sync+0x1a0/0x1a4
irq event stamp: 189006
hardirqs last enabled at (189005): [<ffff8000080849d0>] finish_task_switch.isra.0+0xe0/0x2c0
hardirqs last disabled at (189006): [<ffff8000090667a4>] el1_dbg+0x24/0xa0
softirqs last enabled at (188966): [<ffff8000080106d0>] __do_softirq+0x4b0/0x6a0
softirqs last disabled at (188957): [<ffff80000804a618>] __irq_exit_rcu+0x108/0x1a4
[ rjw: Because can be freed by the gov_attr_set_put() in
cpufreq_dbs_governor_exit() now, it is also necessary to put the
invocation of the governor ->exit() callback into the new
cpufreq_dbs_data_release() function. ]
Fixes: c443563036 ("cpufreq: governor: New sysfs show/store callbacks for governor tunables")
Signed-off-by: Kevin Hao <haokexin@gmail.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
On QCOM platforms with EPSS flavour of cpufreq IP a throttled frequency is
obtained from another register REG_DOMAIN_STATE, thus the helper function
qcom_lmh_get_throttle_freq() should be modified accordingly, as for now
it returns gibberish since .reg_current_vote is unset for EPSS hardware.
To exclude a hardcoded magic number 19200 it is replaced by "xo" clock rate
in KHz.
Fixes: 275157b367 ("cpufreq: qcom-cpufreq-hw: Add dcvs interrupt support")
Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Signed-off-by: Vladimir Zapolskiy <vladimir.zapolskiy@linaro.org>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Provide lightweight online and offline operations. This saves us from
parsing and tearing down the OPP tables each time the CPU is put online
or offline.
Tested-by: Vladimir Zapolskiy <vladimir.zapolskiy@linaro.org>
Reviewed-by: Vladimir Zapolskiy <vladimir.zapolskiy@linaro.org>
Reviewed-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@linaro.org>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
Pull ARM cpufreq updates for 5.18-rc1 fron Viresh Kumar:
"- Add per core DVFS support for QCom SoC (Bjorn Andersson), convert to yaml
binding (Manivannan Sadhasivam) and various other fixes to the QCom drivers
(Luca Weiss).
- Add OPP table for imx7s SoC (Denys Drozdov) and minor fixes (Stefan Agner).
- Fix CPPC driver's freq/performance conversions (Pierre Gondois).
- Minor generic cleanups (Yury Norov)."
* 'cpufreq/arm/linux-next' of git://git.kernel.org/pub/scm/linux/kernel/git/vireshk/pm:
dt-bindings: cpufreq: cpufreq-qcom-hw: Convert to YAML bindings
dt-bindings: dvfs: Use MediaTek CPUFREQ HW as an example
cpufreq: blocklist Qualcomm sc8280xp and sa8540p in cpufreq-dt-platdev
cpufreq: qcom-hw: Add support for per-core-dcvs
cpufreq: CPPC: Fix performance/frequency conversion
cpufreq: Add i.MX7S to cpufreq-dt-platdev blocklist
ARM: dts: imx7s: Define operating points table for cpufreq
cpufreq: qcom-cpufreq-nvmem: fix reading of PVS Valid fuse
cpufreq: replace cpumask_weight with cpumask_empty where appropriate
Merge power management utilities changes for 5.18-rc1:
- Add tracer tool for the amd-pstate driver (Jinzhou Su).
- Fix PC6 displaying in turbostat on some systems (Artem Bityutskiy).
- Add AMD P-State support to the cpupower utility (Huang Rui).
* pm-tools:
Documentation: amd-pstate: add tracer tool introduction
tools/power/x86/amd_pstate_tracer: Add tracer tool for AMD P-state
tools/power/x86/intel_pstate_tracer: make tracer as a module
cpufreq: amd-pstate: Add more tracepoint for AMD P-State module
turbostat: fix PC6 displaying on some systems
cpupower: Add "perf" option to print AMD P-State information
cpupower: Add function to print AMD P-State performance capabilities
cpupower: Move print_speed function into misc helper
cpupower: Enable boost state support for AMD P-State module
cpupower: Add AMD P-State sysfs definition and access helper
cpupower: Introduce ACPI CPPC library
cpupower: Add the function to get the sysfs value from specific table
cpupower: Initial AMD P-State capability
cpupower: Add the function to check AMD P-State enabled
cpupower: Add AMD P-State capability flag
tools/power/cpupower/{ToDo => TODO}: Rename the todo file
tools: cpupower: fix typo in cpupower-idle-set(1) manpage
The powernow-k8 driver will do checks at startup that the current
active driver is acpi-cpufreq and show a warning when they're not
expected.
Because of this the following warning comes up on systems that
support amd-pstate and compiled in both drivers:
`WTF driver: amd-pstate`
The systems that support powernow-k8 will not support amd-pstate,
so re-order the checks to validate the CPU model number first to
avoid this warning being displayed on modern SOCs.
Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
Reviewed-by: Huang Rui <ray.huang@amd.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
For some specific platforms (E.g. AlderLake) the balance performance
EPP is updated from the hard coded value in the driver. This acts as
the default and balance_performance EPP. The purpose of this EPP
update is to reach maximum 1 core turbo frequency (when possible) out
of the box.
Although we can achieve the objective by using hard coded value in the
driver, there can be other EPP which can be better in terms of power.
But that will be very subjective based on platform and use cases.
This is not practical to have a per platform specific default hard coded
in the driver.
If a platform wants to specify default EPP, it can be set in the firmware.
If this EPP is not the chipset default of 0x80 (balance_perf_epp unless
driver changed it) and more performance oriented but not 0, the driver
can use this as the default and balanced_perf EPP. In this case no driver
update is required every time there is some new platform and default EPP.
If the firmware didn't update the EPP from the chipset default then
the hard coded value is used as per existing implementation.
Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Usually, sysfs attributes have .show and .store and their naming
convention is filename_show() and filename_store().
But in cpufreq the naming convention of these functions is
show_filename() and store_filename() which prevents __ATTR_RW() and
__ATTR_RO() from being used in there to simplify code.
Accordingly, change the naming convention of the sysfs .show and
.store methods in cpufreq to follow the one expected by __ATTR_RW()
and __ATTR_RO() and use these macros in that code.
Signed-off-by: Lianjie Zhang <zhanglianjie@uniontech.com>
[ rjw: Subject and changelog edits ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Add frequency, mperf, aperf and tsc in the trace. This can be used
to debug and tune the performance of AMD P-state driver.
Use the time difference between amd_pstate_update to calculate CPU
frequency. There could be sleep in arch_freq_get_on_cpu, so do not
use it here.
Signed-off-by: Jinzhou Su <Jinzhou.Su@amd.com>
Co-developed-by: Huang Rui <ray.huang@amd.com>
Signed-off-by: Huang Rui <ray.huang@amd.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The Qualcomm sc8280xp and sa8540p platforms also uses the
qcom-cpufreq-hw driver, so add them to the cpufreq-dt-platdev driver's
blocklist.
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
The OSM and EPSS hardware controls the frequency of each cluster in the
system based on requests from the OS and various limiting factors, such
as input from LMH.
In most systems the vote from the OS is done using a single register per
cluster, but some systems are configured to instead take one request per
core. In this configuration a set of consecutive registers are used for
the OS to request the frequency of each of the cores within the cluster.
The information is then aggregated in the hardware and the frequency for
the cluster is determined.
As the current implementation ends up only requesting a frequency for
the first core in each cluster and only the vote of non-idle cores are
considered it's often the case that the cluster will be clocked (much)
lower than expected.
It's possible that there are benefits of performing the per-core
requests from the OS, but more investigation of the outcome is needed
before introducing such support. As such this patch extends the request
for the cluster to be written to all the cores.
The weight of the policy's related_cpus is used to determine how many
cores, and hence consecutive registers, each cluster has.
The OS is not permitted to disable the per-core dcvs feature.
Signed-off-by: Bjorn Andersson <bjorn.andersson@linaro.org>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
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>
The i.MX 7Solo currently does not have multiple operating points,
however, in order for the i.MX Thermal driver to successfully probe
a cpufreq device is required. Add it to the cpufreq-dt-platdev
driver's blocklist to allow using imx-cpufreq-dt.
Signed-off-by: Stefan Agner <stefan.agner@toradex.com>
Cc: Stefan Agner <stefan@agner.ch>
Signed-off-by: Francesco Dolcini <francesco.dolcini@toradex.com>
Reviewed-by: Fabio Estevam <festevam@gmail.com>
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.org>
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>
drivers/cpufreq calls cpumask_weight() to check if any bit of a given
cpumask is set. We can do it more efficiently with cpumask_empty() because
cpumask_empty() stops traversing the cpumask as soon as it finds first set
bit, while cpumask_weight() counts all bits unconditionally.
Signed-off-by: Yury Norov <yury.norov@gmail.com>
Reviewed-by: Sudeep Holla <sudeep.holla@arm.com> (for SCMI cpufreq driver)
Signed-off-by: Viresh Kumar <viresh.kumar@linaro.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>
Replace acpi_bus_get_device() that is going to be dropped with
acpi_fetch_acpi_dev().
No intentional functional impact.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
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>
