The patchset series add core performance boost feature for AMD pstate
driver including passisve ,guide and active mode support.
User can change core frequency boost control with a new sysfs entry:
"/sys/devices/system/cpu/amd_pstate/cpb_boost"
The legancy boost interface has been removed due to the function
conflict with new cpb_boost which can support all modes.
1). enable core boost:
$ sudo bash -c "echo 0 > /sys/devices/system/cpu/amd_pstate/cpb_boost"
$ lscpu -ae
CPU NODE SOCKET CORE L1d:L1i:L2:L3 ONLINE MAXMHZ MINMHZ MHZ
0 0 0 0 0:0:0:0 yes 4201.0000 400.0000 2983.578
1 0 0 1 1:1:1:0 yes 4201.0000 400.0000 2983.578
2 0 0 2 2:2:2:0 yes 4201.0000 400.0000 2583.855
3 0 0 3 3:3:3:0 yes 4201.0000 400.0000 2983.578
4 0 0 4 4:4:4:0 yes 4201.0000 400.0000 2983.578
2). disabble core boost:
$ sudo bash -c "echo 1 > /sys/devices/system/cpu/amd_pstate/cpb_boost"
$ lscpu -ae
0 0 0 0 0:0:0:0 yes 5759.0000 400.0000 2983.578
1 0 0 1 1:1:1:0 yes 5759.0000 400.0000 2983.578
2 0 0 2 2:2:2:0 yes 5759.0000 400.0000 2983.578
3 0 0 3 3:3:3:0 yes 5759.0000 400.0000 2983.578
4 0 0 4 4:4:4:0 yes 5759.0000 400.0000 2983.578
The patches have been tested with the AMD 7950X processor and many users
would like to get core boost control enabled for power saving.
Changes from v1:
* drop suspend/resume fix patch 6/7 because of the fix should be in
another fix series instead of CPB feature
* move the set_boost remove patch to the last(Mario)
* Fix commit info with "Closes:" (Mario)
* simplified global.cpb_supported initialization(Mario)
* Add guide mode support for CPB control
* Fixed some Doc typos and add guide mode info to Doc as well.
v1: https://lore.kernel.org/all/[email protected]/
Perry Yuan (6):
cpufreq: amd-pstate: initialize new core precision boost state
cpufreq: amd-pstate: implement cpb_boost sysfs entry for boost control
cpufreq: amd-pstate: fix max_perf calculation for amd_get_max_freq()
cpufreq: amd-pstate: fix the MSR highest perf will be reset issue
while cpb boost off
Documentation: cpufreq: amd-pstate: introduce the new cpu boost
control method
cpufreq: amd-pstate: remove legacy set_boost callback for passive mode
Documentation/admin-guide/pm/amd-pstate.rst | 11 ++
drivers/cpufreq/amd-pstate.c | 168 ++++++++++++++++----
include/linux/amd-pstate.h | 1 -
3 files changed, 145 insertions(+), 35 deletions(-)
--
2.34.1
From: Perry Yuan <[email protected]>
Add gloal global_params to represent current CPU Performance Boost(cpb)
state for cpu frequency scaling, both active and passive modes all can
support CPU cores frequency boosting control which is based on the BIOS
setting, while BIOS turn on the "Core Performance Boost", it will
allow OS control each core highest perf limitation from OS side.
If core performance boost is disabled while a core is in a boosted P-state,
the core transitions to the highest performance non-boosted P-state,
that is the same as the nominal frequency limit.
Reported-by: Artem S. Tashkinov" <[email protected]>
Closes: https://bugzilla.kernel.org/show_bug.cgi?id=217931
Signed-off-by: Perry Yuan <[email protected]>
---
drivers/cpufreq/amd-pstate.c | 61 +++++++++++++++++-------------------
1 file changed, 28 insertions(+), 33 deletions(-)
diff --git a/drivers/cpufreq/amd-pstate.c b/drivers/cpufreq/amd-pstate.c
index 9a1e194d5cf8..921324ab9580 100644
--- a/drivers/cpufreq/amd-pstate.c
+++ b/drivers/cpufreq/amd-pstate.c
@@ -65,6 +65,19 @@ static struct cpufreq_driver amd_pstate_epp_driver;
static int cppc_state = AMD_PSTATE_UNDEFINED;
static bool cppc_enabled;
+/**
+ * struct global_params - Global parameters, mostly tunable via sysfs.
+ * @cpb_boost: Whether or not to use boost CPU P-states.
+ * @cpb_supported: Whether or not CPU boost P-states are available
+ * based on the MSR_K7_HWCR bit[25] state
+ */
+struct global_params {
+ bool cpb_boost;
+ bool cpb_supported;
+};
+
+static struct global_params global;
+
/*
* AMD Energy Preference Performance (EPP)
* The EPP is used in the CCLK DPM controller to drive
@@ -632,43 +645,21 @@ static int amd_get_lowest_nonlinear_freq(struct amd_cpudata *cpudata)
return lowest_nonlinear_freq * 1000;
}
-static int amd_pstate_set_boost(struct cpufreq_policy *policy, int state)
+static int amd_pstate_boost_init(struct amd_cpudata *cpudata)
{
- struct amd_cpudata *cpudata = policy->driver_data;
+ u64 boost_val;
int ret;
- if (!cpudata->boost_supported) {
- pr_err("Boost mode is not supported by this processor or SBIOS\n");
- return -EINVAL;
- }
-
- if (state)
- policy->cpuinfo.max_freq = cpudata->max_freq;
- else
- policy->cpuinfo.max_freq = cpudata->nominal_freq;
-
- policy->max = policy->cpuinfo.max_freq;
-
- ret = freq_qos_update_request(&cpudata->req[1],
- policy->cpuinfo.max_freq);
- if (ret < 0)
+ ret = rdmsrl_on_cpu(cpudata->cpu, MSR_K7_HWCR, &boost_val);
+ if (ret) {
+ pr_err_once("failed to read initial CPU boost state!\n");
return ret;
+ }
- return 0;
-}
-
-static void amd_pstate_boost_init(struct amd_cpudata *cpudata)
-{
- u32 highest_perf, nominal_perf;
-
- highest_perf = READ_ONCE(cpudata->highest_perf);
- nominal_perf = READ_ONCE(cpudata->nominal_perf);
-
- if (highest_perf <= nominal_perf)
- return;
+ global.cpb_supported = !((boost_val >> 25) & 0x1);
+ global.cpb_boost = global.cpb_supported;
- cpudata->boost_supported = true;
- current_pstate_driver->boost_enabled = true;
+ return ret;
}
static void amd_perf_ctl_reset(unsigned int cpu)
@@ -701,6 +692,9 @@ static int amd_pstate_cpu_init(struct cpufreq_policy *policy)
if (ret)
goto free_cpudata1;
+ /* initialize cpu cores boot state */
+ amd_pstate_boost_init(cpudata);
+
min_freq = amd_get_min_freq(cpudata);
max_freq = amd_get_max_freq(cpudata);
nominal_freq = amd_get_nominal_freq(cpudata);
@@ -750,7 +744,6 @@ static int amd_pstate_cpu_init(struct cpufreq_policy *policy)
policy->driver_data = cpudata;
- amd_pstate_boost_init(cpudata);
if (!current_pstate_driver->adjust_perf)
current_pstate_driver->adjust_perf = amd_pstate_adjust_perf;
@@ -1118,6 +1111,9 @@ static int amd_pstate_epp_cpu_init(struct cpufreq_policy *policy)
if (ret)
goto free_cpudata1;
+ /* initialize cpu cores boot state */
+ amd_pstate_boost_init(cpudata);
+
min_freq = amd_get_min_freq(cpudata);
max_freq = amd_get_max_freq(cpudata);
nominal_freq = amd_get_nominal_freq(cpudata);
@@ -1168,7 +1164,6 @@ static int amd_pstate_epp_cpu_init(struct cpufreq_policy *policy)
return ret;
WRITE_ONCE(cpudata->cppc_cap1_cached, value);
}
- amd_pstate_boost_init(cpudata);
return 0;
--
2.34.1
From: Perry Yuan <[email protected]>
With this new sysfs entry `cpb_boost`created, user can change CPU boost
state dynamically under `active`, `guided` and `passive` modes.
And the highest perf and frequency will also be updated as the boost
state changing.
0): check current boost state
cat /sys/devices/system/cpu/amd_pstate/cpb_boost
1): disable CPU boost
sudo bash -c "echo 0 > /sys/devices/system/cpu/amd_pstate/cpb_boost"
2): enable CPU boost
sudo bash -c "echo 1 > /sys/devices/system/cpu/amd_pstate/cpb_boost"
Link: https://bugzilla.kernel.org/show_bug.cgi?id=217931
Link: https://bugzilla.kernel.org/show_bug.cgi?id=217618
Signed-off-by: Perry Yuan <[email protected]>
---
drivers/cpufreq/amd-pstate.c | 97 ++++++++++++++++++++++++++++++++++++
1 file changed, 97 insertions(+)
diff --git a/drivers/cpufreq/amd-pstate.c b/drivers/cpufreq/amd-pstate.c
index 921324ab9580..018ea7cdda2b 100644
--- a/drivers/cpufreq/amd-pstate.c
+++ b/drivers/cpufreq/amd-pstate.c
@@ -1030,6 +1030,101 @@ static ssize_t status_store(struct device *a, struct device_attribute *b,
return ret < 0 ? ret : count;
}
+static int amd_cpu_boost_update(struct amd_cpudata *cpudata, u32 on)
+{
+ struct cpufreq_policy *policy = cpufreq_cpu_acquire(cpudata->cpu);
+ struct cppc_perf_ctrls perf_ctrls;
+ u32 highest_perf, nominal_perf;
+ int ret;
+
+ if (!policy)
+ return -ENODATA;
+
+ highest_perf = READ_ONCE(cpudata->highest_perf);
+ nominal_perf = READ_ONCE(cpudata->nominal_perf);
+
+ if (boot_cpu_has(X86_FEATURE_CPPC)) {
+ u64 value = READ_ONCE(cpudata->cppc_req_cached);
+
+ value &= ~GENMASK_ULL(7, 0);
+ value |= on ? highest_perf : nominal_perf;
+ WRITE_ONCE(cpudata->cppc_req_cached, value);
+
+ wrmsrl_on_cpu(cpudata->cpu, MSR_AMD_CPPC_REQ, value);
+
+ } else {
+ perf_ctrls.max_perf = on ? highest_perf : nominal_perf;
+ ret = cppc_set_epp_perf(cpudata->cpu, &perf_ctrls, 1);
+ if (ret) {
+ pr_debug("failed to set energy perf value (%d)\n", ret);
+ return ret;
+ }
+ }
+
+ if (on)
+ policy->cpuinfo.max_freq = cpudata->max_freq;
+ else
+ policy->cpuinfo.max_freq = cpudata->nominal_freq;
+
+ policy->max = policy->cpuinfo.max_freq;
+
+ if (cppc_state == AMD_PSTATE_PASSIVE) {
+ ret = freq_qos_update_request(&cpudata->req[1],
+ policy->cpuinfo.max_freq);
+ }
+
+ cpufreq_cpu_release(policy);
+
+ return ret;
+}
+
+static ssize_t cpb_boost_show(struct device *dev,
+ struct device_attribute *attr, char *buf)
+{
+ return sysfs_emit(buf, "%u\n", global.cpb_boost);
+}
+
+static ssize_t cpb_boost_store(struct device *dev, struct device_attribute *b,
+ const char *buf, size_t count)
+{
+ bool new_state;
+ ssize_t ret;
+ int cpu;
+
+ mutex_lock(&amd_pstate_driver_lock);
+ if (!global.cpb_supported) {
+ pr_err("Boost mode is not supported by this processor or SBIOS\n");
+ return -EINVAL;
+ }
+
+ ret = kstrtobool(buf, &new_state);
+ if (ret)
+ return -EINVAL;
+
+ global.cpb_boost = !!new_state;
+
+ for_each_possible_cpu(cpu) {
+
+ struct cpufreq_policy *policy = cpufreq_cpu_get(cpu);
+ struct amd_cpudata *cpudata = policy->driver_data;
+
+ if (!cpudata) {
+ pr_err("cpudata is NULL\n");
+ ret = -ENODATA;
+ cpufreq_cpu_put(policy);
+ goto err_exit;
+ }
+
+ amd_cpu_boost_update(cpudata, global.cpb_boost);
+ refresh_frequency_limits(policy);
+ cpufreq_cpu_put(policy);
+ }
+
+err_exit:
+ mutex_unlock(&amd_pstate_driver_lock);
+ return ret < 0 ? ret : count;
+}
+
cpufreq_freq_attr_ro(amd_pstate_max_freq);
cpufreq_freq_attr_ro(amd_pstate_lowest_nonlinear_freq);
@@ -1037,6 +1132,7 @@ cpufreq_freq_attr_ro(amd_pstate_highest_perf);
cpufreq_freq_attr_rw(energy_performance_preference);
cpufreq_freq_attr_ro(energy_performance_available_preferences);
static DEVICE_ATTR_RW(status);
+static DEVICE_ATTR_RW(cpb_boost);
static struct freq_attr *amd_pstate_attr[] = {
&amd_pstate_max_freq,
@@ -1056,6 +1152,7 @@ static struct freq_attr *amd_pstate_epp_attr[] = {
static struct attribute *pstate_global_attributes[] = {
&dev_attr_status.attr,
+ &dev_attr_cpb_boost.attr,
NULL
};
--
2.34.1
From: Perry Yuan <[email protected]>
When CPU core Precision Boost state changed, the max frequency will also
need to be updated according to the current boost state, if boost is
disabled now, the max perf will be limited to nominal perf values.
otherwise the max frequency will be showed wrongly.
Signed-off-by: Perry Yuan <[email protected]>
---
drivers/cpufreq/amd-pstate.c | 4 ++++
1 file changed, 4 insertions(+)
diff --git a/drivers/cpufreq/amd-pstate.c b/drivers/cpufreq/amd-pstate.c
index 018ea7cdda2b..107883cf0877 100644
--- a/drivers/cpufreq/amd-pstate.c
+++ b/drivers/cpufreq/amd-pstate.c
@@ -599,6 +599,10 @@ static int amd_get_max_freq(struct amd_cpudata *cpudata)
nominal_perf = READ_ONCE(cpudata->nominal_perf);
max_perf = READ_ONCE(cpudata->highest_perf);
+ /* when boost is off, the highest perf will be limited to nominal_perf */
+ if (!global.cpb_boost)
+ max_perf = nominal_perf;
+
boost_ratio = div_u64(max_perf << SCHED_CAPACITY_SHIFT,
nominal_perf);
--
2.34.1
From: Perry Yuan <[email protected]>
Select the min perf to fix the highest perf value while update pstate
CPPC request MSR register, here we need to limit the max perf value when
CPU boost is disabled in case of that highest perf value in the MSR will be
reset to original highest perf value which cause the BOOST control
failed.
Signed-off-by: Perry Yuan <[email protected]>
---
drivers/cpufreq/amd-pstate.c | 5 +++++
1 file changed, 5 insertions(+)
diff --git a/drivers/cpufreq/amd-pstate.c b/drivers/cpufreq/amd-pstate.c
index 107883cf0877..02a76b8fb298 100644
--- a/drivers/cpufreq/amd-pstate.c
+++ b/drivers/cpufreq/amd-pstate.c
@@ -443,6 +443,7 @@ static void amd_pstate_update(struct amd_cpudata *cpudata, u32 min_perf,
u32 des_perf, u32 max_perf, bool fast_switch, int gov_flags)
{
u64 prev = READ_ONCE(cpudata->cppc_req_cached);
+ u32 nominal_perf = READ_ONCE(cpudata->nominal_perf);
u64 value = prev;
des_perf = clamp_t(unsigned long, des_perf, min_perf, max_perf);
@@ -458,6 +459,10 @@ static void amd_pstate_update(struct amd_cpudata *cpudata, u32 min_perf,
value &= ~AMD_CPPC_DES_PERF(~0L);
value |= AMD_CPPC_DES_PERF(des_perf);
+ /* limit the max perf when core performance boost feature is disabled */
+ if (!global.cpb_boost)
+ max_perf = min_t(unsigned long, nominal_perf, max_perf);
+
value &= ~AMD_CPPC_MAX_PERF(~0L);
value |= AMD_CPPC_MAX_PERF(max_perf);
--
2.34.1
From: Perry Yuan <[email protected]>
Introduce AMD CPU frequency boosting control sysfs entry which used for
switching boost on and boost off.
If core performance boost is disabled while a core is in a boosted P-state,
the core automatically transitions to the highest performance non-boosted P-state
The highest perf and frequency will be limited by the setting value.
Signed-off-by: Perry Yuan <[email protected]>
---
Documentation/admin-guide/pm/amd-pstate.rst | 11 +++++++++++
1 file changed, 11 insertions(+)
diff --git a/Documentation/admin-guide/pm/amd-pstate.rst b/Documentation/admin-guide/pm/amd-pstate.rst
index 1cf40f69278c..abea29ba3d54 100644
--- a/Documentation/admin-guide/pm/amd-pstate.rst
+++ b/Documentation/admin-guide/pm/amd-pstate.rst
@@ -385,6 +385,17 @@ control its functionality at the system level. They are located in the
to the operation mode represented by that string - or to be
unregistered in the "disable" case.
+``cpb_boost``
+ Specifies whether core performance boost is requested to be enabled or disabled
+ If core performance boost is disabled while a core is in a boosted P-state, the
+ core automatically transitions to the highest performance non-boosted P-state.
+ AMD Core Performance Boost(CPB) is controlled by this new attribute file which
+ allow user to change all cores frequency boosting state. It supports both
+ ``active``, ``passive`` and ``guided`` mode control with below value write to it.
+
+ "0" Disable Core performance Boosting
+ "1" Enable Core performance Boosting
+
``cpupower`` tool support for ``amd-pstate``
===============================================
--
2.34.1
With new freqency boost interface supported, legacy boost control
doesn't make sense any more which only support passive mode.
so it can remove the legacy set_boost interface from amd-pstate driver
in case of there is conflict with new boost control logic.
Signed-off-by: Perry Yuan <[email protected]>
---
drivers/cpufreq/amd-pstate.c | 1 -
include/linux/amd-pstate.h | 1 -
2 files changed, 2 deletions(-)
diff --git a/drivers/cpufreq/amd-pstate.c b/drivers/cpufreq/amd-pstate.c
index 02a76b8fb298..4e377efdd4ed 100644
--- a/drivers/cpufreq/amd-pstate.c
+++ b/drivers/cpufreq/amd-pstate.c
@@ -1492,7 +1492,6 @@ static struct cpufreq_driver amd_pstate_driver = {
.exit = amd_pstate_cpu_exit,
.suspend = amd_pstate_cpu_suspend,
.resume = amd_pstate_cpu_resume,
- .set_boost = amd_pstate_set_boost,
.name = "amd-pstate",
.attr = amd_pstate_attr,
};
diff --git a/include/linux/amd-pstate.h b/include/linux/amd-pstate.h
index 446394f84606..66d939a344b1 100644
--- a/include/linux/amd-pstate.h
+++ b/include/linux/amd-pstate.h
@@ -80,7 +80,6 @@ struct amd_cpudata {
struct amd_aperf_mperf prev;
u64 freq;
- bool boost_supported;
/* EPP feature related attributes*/
s16 epp_policy;
--
2.34.1
Hello.
On neděle 4. února 2024 13:12:57 CET Perry Yuan wrote:
> With new freqency boost interface supported, legacy boost control
> doesn't make sense any more which only support passive mode.
> so it can remove the legacy set_boost interface from amd-pstate driver
> in case of there is conflict with new boost control logic.
>
> Signed-off-by: Perry Yuan <[email protected]>
> ---
> drivers/cpufreq/amd-pstate.c | 1 -
> include/linux/amd-pstate.h | 1 -
> 2 files changed, 2 deletions(-)
>
> diff --git a/drivers/cpufreq/amd-pstate.c b/drivers/cpufreq/amd-pstate.c
> index 02a76b8fb298..4e377efdd4ed 100644
> --- a/drivers/cpufreq/amd-pstate.c
> +++ b/drivers/cpufreq/amd-pstate.c
> @@ -1492,7 +1492,6 @@ static struct cpufreq_driver amd_pstate_driver = {
> .exit = amd_pstate_cpu_exit,
> .suspend = amd_pstate_cpu_suspend,
> .resume = amd_pstate_cpu_resume,
> - .set_boost = amd_pstate_set_boost,
> .name = "amd-pstate",
> .attr = amd_pstate_attr,
> };
> diff --git a/include/linux/amd-pstate.h b/include/linux/amd-pstate.h
> index 446394f84606..66d939a344b1 100644
> --- a/include/linux/amd-pstate.h
> +++ b/include/linux/amd-pstate.h
> @@ -80,7 +80,6 @@ struct amd_cpudata {
> struct amd_aperf_mperf prev;
>
> u64 freq;
> - bool boost_supported;
This leaves amd_pstate_ut_check_freq() in drivers/cpufreq/amd-pstate-ut.c broken. Likely, the whole `if (cpudata->boost_supported) {` hunk should be removed there too.
Also, in the header file, there's kernel-doc before `struct amd_cpudata`, where boost_supported is mentioned. It should be removed too then.
>
> /* EPP feature related attributes*/
> s16 epp_policy;
>
--
Oleksandr Natalenko (post-factum)
On 2/4/2024 08:57, Oleksandr Natalenko wrote:
> Hello.
>
> On neděle 4. února 2024 13:12:57 CET Perry Yuan wrote:
>> With new freqency boost interface supported, legacy boost control
>> doesn't make sense any more which only support passive mode.
>> so it can remove the legacy set_boost interface from amd-pstate driver
>> in case of there is conflict with new boost control logic.
>>
>> Signed-off-by: Perry Yuan <[email protected]>
>> ---
>> drivers/cpufreq/amd-pstate.c | 1 -
>> include/linux/amd-pstate.h | 1 -
>> 2 files changed, 2 deletions(-)
>>
>> diff --git a/drivers/cpufreq/amd-pstate.c b/drivers/cpufreq/amd-pstate.c
>> index 02a76b8fb298..4e377efdd4ed 100644
>> --- a/drivers/cpufreq/amd-pstate.c
>> +++ b/drivers/cpufreq/amd-pstate.c
>> @@ -1492,7 +1492,6 @@ static struct cpufreq_driver amd_pstate_driver = {
>> .exit = amd_pstate_cpu_exit,
>> .suspend = amd_pstate_cpu_suspend,
>> .resume = amd_pstate_cpu_resume,
>> - .set_boost = amd_pstate_set_boost,
>> .name = "amd-pstate",
>> .attr = amd_pstate_attr,
>> };
>> diff --git a/include/linux/amd-pstate.h b/include/linux/amd-pstate.h
>> index 446394f84606..66d939a344b1 100644
>> --- a/include/linux/amd-pstate.h
>> +++ b/include/linux/amd-pstate.h
>> @@ -80,7 +80,6 @@ struct amd_cpudata {
>> struct amd_aperf_mperf prev;
>>
>> u64 freq;
>> - bool boost_supported;
>
> This leaves amd_pstate_ut_check_freq() in drivers/cpufreq/amd-pstate-ut.c broken. Likely, the whole `if (cpudata->boost_supported) {` hunk should be removed there too.
>
> Also, in the header file, there's kernel-doc before `struct amd_cpudata`, where boost_supported is mentioned. It should be removed too then.
Yeah; I though the kernel robot caught this on v1 too. Make sure you
run the unit tests at *each* patch so that this code remains bisectable.
>
>>
>> /* EPP feature related attributes*/
>> s16 epp_policy;
>>
>
>