amd-pstate preferred core

amd-pstate driver utilizes the functions and data structures provided by the ITMT architecture to enable the scheduler to favor scheduling on cores which can be get a higher frequency with lower voltage. We call it amd-pstate preferrred core.

Here sched_set_itmt_core_prio() is called to set priorities and sched_set_itmt_support() is called to enable ITMT feature. amd-pstate driver uses the highest performance value to indicate the priority of CPU. The higher value has a higher priority.

The initial core rankings are set up by amd-pstate when the system boots.

Add device attribute for hardware preferred core. It will check if the processor and power firmware support preferred core feature.

Add device attribute for preferred core. Only when hardware supports preferred core and user set enabled in early parameter, it can be set to enabled.

Add one new early parameter disable to allow user to disable the preferred core.

Reviewed-by Mario Limonciello <[email protected]>

Co-developed-by Perry Yuan <[email protected]>

Signed-off-by Perry Yuan <[email protected]>

Signed-off-by Meng Li <[email protected]>

---
 drivers/cpufreq/amd-pstate.c | 163 +++++++++++++++++++++++++++++++----
 1 file changed, 147 insertions(+), 16 deletions(-)

drivers/cpufreq/amd-pstate.c +73 -16

index 9a1e194d5cf8..050e23594057 100644
@@ -37,6 +37,7 @@
 #include <linux/uaccess.h>
 #include <linux/static_call.h>
 #include <linux/amd-pstate.h>
+#include <linux/topology.h>
 
 #include <acpi/processor.h>
 #include <acpi/cppc_acpi.h>

@@ -49,6 +50,8 @@
 
 #define AMD_PSTATE_TRANSITION_LATENCY	20000
 #define AMD_PSTATE_TRANSITION_DELAY	1000
+#define AMD_PSTATE_PREFCORE_THRESHOLD	166
+#define AMD_PSTATE_MAX_CPPC_PERF	255
 
 /*
  * TODO: We need more time to fine tune processors with shared memory solution

@@ -65,6 +68,12 @@ static struct cpufreq_driver amd_pstate_epp_driver;
 static int cppc_state = AMD_PSTATE_UNDEFINED;
 static bool cppc_enabled;
 
+/*HW preferred Core featue is supported*/
+static bool hw_prefcore = true;
+
+/*Preferred Core featue is supported*/
+static bool prefcore = true;
+
 /*
  * AMD Energy Preference Performance (EPP)
  * The EPP is used in the CCLK DPM controller to drive

@@ -290,23 +299,21 @@ static inline int amd_pstate_enable(bool enable)
 static int pstate_init_perf(struct amd_cpudata *cpudata)
 {
 	u64 cap1;
-	u32 highest_perf;
 
 	int ret = rdmsrl_safe_on_cpu(cpudata->cpu, MSR_AMD_CPPC_CAP1,
 				     &cap1);
 	if (ret)
 		return ret;
 
-	/*
-	 * TODO: Introduce AMD specific power feature.
-	 *
-	 * CPPC entry doesn't indicate the highest performance in some ASICs.
+	/* For platforms that do not support the preferred core feature, the
+	 * highest_pef may be configured with 166 or 255, to avoid max frequency
+	 * calculated wrongly. we take the AMD_CPPC_HIGHEST_PERF(cap1) value as
+	 * the default max perf.
 	 */
-	highest_perf = amd_get_highest_perf();
-	if (highest_perf > AMD_CPPC_HIGHEST_PERF(cap1))
-		highest_perf = AMD_CPPC_HIGHEST_PERF(cap1);
-
-	WRITE_ONCE(cpudata->highest_perf, highest_perf);
+	if (hw_prefcore)
+		WRITE_ONCE(cpudata->highest_perf, AMD_PSTATE_PREFCORE_THRESHOLD);
+	else
+		WRITE_ONCE(cpudata->highest_perf, AMD_CPPC_HIGHEST_PERF(cap1));
 
 	WRITE_ONCE(cpudata->nominal_perf, AMD_CPPC_NOMINAL_PERF(cap1));
 	WRITE_ONCE(cpudata->lowest_nonlinear_perf, AMD_CPPC_LOWNONLIN_PERF(cap1));

@@ -318,17 +325,15 @@ static int pstate_init_perf(struct amd_cpudata *cpudata)
 static int cppc_init_perf(struct amd_cpudata *cpudata)
 {
 	struct cppc_perf_caps cppc_perf;
-	u32 highest_perf;
 
 	int ret = cppc_get_perf_caps(cpudata->cpu, &cppc_perf);
 	if (ret)
 		return ret;
 
-	highest_perf = amd_get_highest_perf();
-	if (highest_perf > cppc_perf.highest_perf)
-		highest_perf = cppc_perf.highest_perf;
-
-	WRITE_ONCE(cpudata->highest_perf, highest_perf);
+	if (hw_prefcore)
+		WRITE_ONCE(cpudata->highest_perf, AMD_PSTATE_PREFCORE_THRESHOLD);
+	else
+		WRITE_ONCE(cpudata->highest_perf, cppc_perf.highest_perf);
 
 	WRITE_ONCE(cpudata->nominal_perf, cppc_perf.nominal_perf);
 	WRITE_ONCE(cpudata->lowest_nonlinear_perf,

@@ -676,6 +681,90 @@ static void amd_perf_ctl_reset(unsigned int cpu)
 	wrmsrl_on_cpu(cpu, MSR_AMD_PERF_CTL, 0);
 }
 
+/*
+ * Set amd-pstate preferred core enable can't be done directly from cpufreq callbacks
+ * due to locking, so queue the work for later.
+ */
+static void amd_pstste_sched_prefcore_workfn(struct work_struct *work)
+{
+	sched_set_itmt_support();
+}
+static DECLARE_WORK(sched_prefcore_work, amd_pstste_sched_prefcore_workfn);
+
+/*
+ * Get the highest performance register value.
+ * @cpu: CPU from which to get highest performance.
+ * @highest_perf: Return address.
+ *
+ * Return: 0 for success, -EIO otherwise.
+ */
+static int amd_pstate_get_highest_perf(int cpu, u32 *highest_perf)
+{
+	int ret;
+	u64 cppc_highest_perf;
+
+	if (boot_cpu_has(X86_FEATURE_CPPC)) {
+		u64 cap1;
+
+		ret = rdmsrl_safe_on_cpu(cpu, MSR_AMD_CPPC_CAP1, &cap1);
+		if (ret)
+			return ret;
+		WRITE_ONCE(*highest_perf, AMD_CPPC_HIGHEST_PERF(cap1));
+	} else {
+		ret = cppc_get_highest_perf(cpu, &cppc_highest_perf);
+		*highest_perf = (u32)(cppc_highest_perf & 0xFFFF);
+	}
+
+	return (ret);
+}
+
+static void amd_pstate_init_prefcore(unsigned int cpu)
+{
+	int ret;
+	u32 highest_perf;
+	static u32 max_highest_perf = 0, min_highest_perf = U32_MAX;
+
+	if (!prefcore)
+		return;
+
+	ret = amd_pstate_get_highest_perf(cpu, &highest_perf);
+	if (ret)
+		return;
+
+	/*
+	 * The priorities can be set regardless of whether or not

drivers/cpufreq/amd-pstate.c +64 -0

+		return;
+	}
+
+	if (max_highest_perf <= min_highest_perf) {
+		if (highest_perf > max_highest_perf)
+			max_highest_perf = highest_perf;
+
+		if (highest_perf < min_highest_perf)
+			min_highest_perf = highest_perf;
+
+		if (max_highest_perf > min_highest_perf) {
+			/*
+			 * This code can be run during CPU online under the
+			 * CPU hotplug locks, so sched_set_itmt_support()
+			 * cannot be called from here.  Queue up a work item
+			 * to invoke it.
+			 */
+			schedule_work(&sched_prefcore_work);
+		}
+	}
+}
+
 static int amd_pstate_cpu_init(struct cpufreq_policy *policy)
 {
 	int min_freq, max_freq, nominal_freq, lowest_nonlinear_freq, ret;

@@ -697,6 +786,8 @@ static int amd_pstate_cpu_init(struct cpufreq_policy *policy)
 
 	cpudata->cpu = policy->cpu;
 
+	amd_pstate_init_prefcore(policy->cpu);
+
 	ret = amd_pstate_init_perf(cpudata);
 	if (ret)
 		goto free_cpudata1;

@@ -763,6 +854,22 @@ static int amd_pstate_cpu_init(struct cpufreq_policy *policy)
 	return ret;
 }
 
+static int amd_pstate_cpu_online(struct cpufreq_policy *policy)
+{
+	struct amd_cpudata *cpudata = policy->driver_data;
+
+	pr_debug("CPU %d going online\n", cpudata->cpu);
+
+	amd_pstate_init_prefcore(cpudata->cpu);
+
+	return 0;
+}
+
+static int amd_pstate_cpu_offline(struct cpufreq_policy *policy)
+{
+	return 0;
+}
+
 static int amd_pstate_cpu_exit(struct cpufreq_policy *policy)
 {
 	struct amd_cpudata *cpudata = policy->driver_data;

@@ -1037,6 +1144,12 @@ static ssize_t status_store(struct device *a, struct device_attribute *b,
 	return ret < 0 ? ret : count;
 }
 
+static ssize_t prefcore_show(struct device *dev,
+			     struct device_attribute *attr, char *buf)
+{
+	return sysfs_emit(buf, "%s\n", prefcore ? "enabled" : "disabled");
+}
+
 cpufreq_freq_attr_ro(amd_pstate_max_freq);
 cpufreq_freq_attr_ro(amd_pstate_lowest_nonlinear_freq);
 

@@ -1044,6 +1157,7 @@ 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_RO(prefcore);
 
 static struct freq_attr *amd_pstate_attr[] = {
 	&amd_pstate_max_freq,

@@ -1063,6 +1177,7 @@ static struct freq_attr *amd_pstate_epp_attr[] = {
 
 static struct attribute *pstate_global_attributes[] = {
 	&dev_attr_status.attr,
+	&dev_attr_prefcore.attr,
 	NULL
 };
 

@@ -1114,6 +1229,8 @@ static int amd_pstate_epp_cpu_init(struct cpufreq_policy *policy)
 	cpudata->cpu = policy->cpu;
 	cpudata->epp_policy = 0;
 
+	amd_pstate_init_prefcore(policy->cpu);
+
 	ret = amd_pstate_init_perf(cpudata);
 	if (ret)
 		goto free_cpudata1;

@@ -1285,6 +1402,8 @@ static int amd_pstate_epp_cpu_online(struct cpufreq_policy *policy)
 
 	pr_debug("AMD CPU Core %d going online\n", cpudata->cpu);
 
+	amd_pstate_init_prefcore(cpudata->cpu);
+
 	if (cppc_state == AMD_PSTATE_ACTIVE) {
 		amd_pstate_epp_reenable(cpudata);
 		cpudata->suspended = false;

@@ -1389,6 +1508,8 @@ static struct cpufreq_driver amd_pstate_driver = {
 	.fast_switch    = amd_pstate_fast_switch,
 	.init		= amd_pstate_cpu_init,
 	.exit		= amd_pstate_cpu_exit,
+	.offline	= amd_pstate_cpu_offline,
+	.online		= amd_pstate_cpu_online,
 	.suspend	= amd_pstate_cpu_suspend,
 	.resume		= amd_pstate_cpu_resume,
 	.set_boost	= amd_pstate_set_boost,

@@ -1527,7 +1648,17 @@ static int __init amd_pstate_param(char *str)
 
 	return amd_pstate_set_driver(mode_idx);
 }
+
+static int __init amd_prefcore_param(char *str)
+{
+	if (!strcmp(str, "disable"))
+		prefcore = false;
+
+	return 0;
+}
+
 early_param("amd_pstate", amd_pstate_param);
+early_param("amd_prefcore", amd_prefcore_param);
 
 MODULE_AUTHOR("Huang Rui <[email protected]>");
 MODULE_DESCRIPTION("AMD Processor P-state Frequency Driver");

Signature · 1 line

2.34.1

Hi all:

The core frequency is subjected to the process variation in semiconductors. Not all cores are able to reach the maximum frequency respecting the infrastructure limits. Consequently, AMD has redefined the concept of maximum frequency of a part. This means that a fraction of cores can reach maximum frequency. To find the best process scheduling policy for a given scenario, OS needs to know the core ordering informed by the platform through highest performance capability register of the CPPC interface.

Earlier implementations of amd-pstate preferred core only support a static core ranking and targeted performance. Now it has the ability to dynamically change the preferred core based on the workload and platform conditions and accounting for thermals and aging.

Amd-pstate driver utilizes the functions and data structures provided by the ITMT architecture to enable the scheduler to favor scheduling on cores which can be get a higher frequency with lower voltage.
We call it amd-pstate preferred core.

Here sched_set_itmt_core_prio() is called to set priorities and sched_set_itmt_support() is called to enable ITMT feature. Amd-pstate driver uses the highest performance value to indicate the priority of CPU. The higher value has a higher priority.

Amd-pstate driver will provide an initial core ordering at boot time. It relies on the CPPC interface to communicate the core ranking to the operating system and scheduler to make sure that OS is choosing the cores with highest performance firstly for scheduling the process. When amd-pstate driver receives a message with the highest performance change, it will update the core ranking.

Changes form V6->V7:
- x86:
- - Modify kconfig about X86_AMD_PSTATE.
- cpufreq: amd-pstate:
- - modify incorrect comments about scheduler_work().
- - convert highest_perf data type.
- - modify preferred core init when cpu init and online.
- acpi: cppc:
- - modify link of CPPC highest performance.
- cpufreq:
- - modify link of CPPC highest performance changed.

Changes form V5->V6:
- cpufreq: amd-pstate:
- - modify the wrong tag order.
- - modify warning about hw_prefcore sysfs attribute.
- - delete duplicate comments.
- - modify the variable name cppc_highest_perf to prefcore_ranking.
- - modify judgment conditions for setting highest_perf.
- - modify sysfs attribute for CPPC highest perf to pr_debug message.
- Documentation: amd-pstate:
- - modify warning: title underline too short.

Changes form V4->V5:
- cpufreq: amd-pstate:
- - modify sysfs attribute for CPPC highest perf.
- - modify warning about comments
- - rebase linux-next
- cpufreq:
- - Moidfy warning about function declarations.
- Documentation: amd-pstate:
- - align with ``amd-pstat``

Changes form V3->V4:
- Documentation: amd-pstate:
- - Modify inappropriate descriptions.

Changes form V2->V3:
- x86:
- - Modify kconfig and description.
- cpufreq: amd-pstate:
- - Add Co-developed-by tag in commit message.
- cpufreq:
- - Modify commit message.
- Documentation: amd-pstate:
- - Modify inappropriate descriptions.

Changes form V1->V2:
- acpi: cppc:
- - Add reference link.
- cpufreq:
- - Moidfy link error.
- cpufreq: amd-pstate:
- - Init the priorities of all online CPUs
- - Use a single variable to represent the status of preferred core.
- Documentation:
- - Default enabled preferred core.
- Documentation: amd-pstate:
- - Modify inappropriate descriptions.
- - Default enabled preferred core.
- - Use a single variable to represent the status of preferred core.

Meng Li (7):
x86: Drop CPU_SUP_INTEL from SCHED_MC_PRIO for the expansion.
acpi: cppc: Add get the highest performance cppc control
cpufreq: amd-pstate: Enable amd-pstate preferred core supporting.
cpufreq: Add a notification message that the highest perf has changed
cpufreq: amd-pstate: Update amd-pstate preferred core ranking
dynamically
Documentation: amd-pstate: introduce amd-pstate preferred core
Documentation: introduce amd-pstate preferrd core mode kernel command
line options

  .../admin-guide/kernel-parameters.txt         |   5 +
  Documentation/admin-guide/pm/amd-pstate.rst   |  58 +++++-
  arch/x86/Kconfig                              |   5 +-
  drivers/acpi/cppc_acpi.c                      |  13 ++
  drivers/acpi/processor_driver.c               |   6 +
  drivers/cpufreq/amd-pstate.c                  | 197 ++++++++++++++++--
  drivers/cpufreq/cpufreq.c                     |  13 ++
  include/acpi/cppc_acpi.h                      |   5 +
  include/linux/amd-pstate.h                    |   6 +
  include/linux/cpufreq.h                       |   5 +
  10 files changed, 291 insertions(+), 22 deletions(-)

-----Original Message-----
From: Limonciello, Mario <[email protected]>
Sent: Tuesday, September 19, 2023 1:41 AM
To: Meng, Li (Jassmine) <[email protected]>; Rafael J . Wysocki <[email protected]>; Huang, Ray <[email protected]>
Cc: [email protected]; [email protected]; [email protected]; [email protected]; Shuah Khan <[email protected]>; [email protected]; Fontenot, Nathan <[email protected]>; Sharma, Deepak <[email protected]>; Deucher, Alexander <[email protected]>; Huang, Shimmer <[email protected]>; Yuan, Perry <[email protected]>; Du, Xiaojian <[email protected]>; Viresh Kumar <[email protected]>; Borislav Petkov <[email protected]>
Subject: Re: [PATCH V7 0/7] amd-pstate preferred core

Quoted 9 lines · Meng Li

Hi all:

The core frequency is subjected to the process variation in semiconductors. Not all cores are able to reach the maximum frequency respecting the infrastructure limits. Consequently, AMD has redefined the concept of maximum frequency of a part. This means that a fraction of cores can reach maximum frequency. To find the best process scheduling policy for a given scenario, OS needs to know the core ordering informed by the platform through highest performance capability register of the CPPC

interface.

Quoted 16 lines

Earlier implementations of amd-pstate preferred core only support a static core ranking and targeted performance. Now it has the ability to dynamically change the preferred core based on the workload and platform conditions and accounting for thermals and aging.

Amd-pstate driver utilizes the functions and data structures provided by the ITMT architecture to enable the scheduler to favor scheduling on cores which can be get a higher frequency with lower voltage. We call it amd-pstate preferred core.

Here sched_set_itmt_core_prio() is called to set priorities and sched_set_itmt_support() is called to enable ITMT feature. Amd-pstate driver uses the highest performance value to indicate the priority of CPU. The higher value has a higher priority.

Quoted 7 lines

Amd-pstate driver will provide an initial core ordering at boot time. It relies on the CPPC interface to communicate the core ranking to the operating system and scheduler to make sure that OS is choosing the cores with highest performance firstly for scheduling the process. When amd-pstate driver receives a message with the highest performance change, it will update the core ranking.

For the remaining patches missing my tag:

Reviewed-by Mario Limonciello <[email protected]>

Quoted 62 lines

Changes form V6->V7:
- x86:
- - Modify kconfig about X86_AMD_PSTATE.
- cpufreq: amd-pstate:
- - modify incorrect comments about scheduler_work().
- - convert highest_perf data type.
- - modify preferred core init when cpu init and online.
- acpi: cppc:
- - modify link of CPPC highest performance.
- cpufreq:
- - modify link of CPPC highest performance changed.

Changes form V5->V6:
- cpufreq: amd-pstate:
- - modify the wrong tag order.
- - modify warning about hw_prefcore sysfs attribute.
- - delete duplicate comments.
- - modify the variable name cppc_highest_perf to prefcore_ranking.
- - modify judgment conditions for setting highest_perf.
- - modify sysfs attribute for CPPC highest perf to pr_debug message.
- Documentation: amd-pstate:
- - modify warning: title underline too short.

Changes form V4->V5:
- cpufreq: amd-pstate:
- - modify sysfs attribute for CPPC highest perf.
- - modify warning about comments
- - rebase linux-next
- cpufreq:
- - Moidfy warning about function declarations.
- Documentation: amd-pstate:
- - align with ``amd-pstat``

Changes form V3->V4:
- Documentation: amd-pstate:
- - Modify inappropriate descriptions.

Changes form V2->V3:
- x86:
- - Modify kconfig and description.
- cpufreq: amd-pstate:
- - Add Co-developed-by tag in commit message.
- cpufreq:
- - Modify commit message.
- Documentation: amd-pstate:
- - Modify inappropriate descriptions.

Changes form V1->V2:
- acpi: cppc:
- - Add reference link.
- cpufreq:
- - Moidfy link error.
- cpufreq: amd-pstate:
- - Init the priorities of all online CPUs
- - Use a single variable to represent the status of preferred core.
- Documentation:
- - Default enabled preferred core.
- Documentation: amd-pstate:
- - Modify inappropriate descriptions.
- - Default enabled preferred core.
- - Use a single variable to represent the status of preferred core.

Quoted 9 lines

Meng Li (7):
x86: Drop CPU_SUP_INTEL from SCHED_MC_PRIO for the expansion.
acpi: cppc: Add get the highest performance cppc control
cpufreq: amd-pstate: Enable amd-pstate preferred core supporting.
cpufreq: Add a notification message that the highest perf has changed
cpufreq: amd-pstate: Update amd-pstate preferred core ranking
dynamically
Documentation: amd-pstate: introduce amd-pstate preferred core
Documentation: introduce amd-pstate preferrd core mode kernel

command

Quoted 14 lines

line options

  .../admin-guide/kernel-parameters.txt         |   5 +
  Documentation/admin-guide/pm/amd-pstate.rst   |  58 +++++-
  arch/x86/Kconfig                              |   5 +-
  drivers/acpi/cppc_acpi.c                      |  13 ++
  drivers/acpi/processor_driver.c               |   6 +
  drivers/cpufreq/amd-pstate.c                  | 197 ++++++++++++++++--
  drivers/cpufreq/cpufreq.c                     |  13 ++
  include/acpi/cppc_acpi.h                      |   5 +
  include/linux/amd-pstate.h                    |   6 +
  include/linux/cpufreq.h                       |   5 +
  10 files changed, 291 insertions(+), 22 deletions(-)

Hi all:

The core frequency is subjected to the process variation in semiconductors. Not all cores are able to reach the maximum frequency respecting the infrastructure limits. Consequently, AMD has redefined the concept of maximum frequency of a part. This means that a fraction of cores can reach maximum frequency. To find the best process scheduling policy for a given scenario, OS needs to know the core ordering informed by the platform through highest performance capability register of the CPPC interface.

Earlier implementations of amd-pstate preferred core only support a static core ranking and targeted performance. Now it has the ability to dynamically change the preferred core based on the workload and platform conditions and accounting for thermals and aging.

Amd-pstate driver utilizes the functions and data structures provided by the ITMT architecture to enable the scheduler to favor scheduling on cores which can be get a higher frequency with lower voltage.
We call it amd-pstate preferred core.

Here sched_set_itmt_core_prio() is called to set priorities and sched_set_itmt_support() is called to enable ITMT feature. Amd-pstate driver uses the highest performance value to indicate the priority of CPU. The higher value has a higher priority.

Amd-pstate driver will provide an initial core ordering at boot time. It relies on the CPPC interface to communicate the core ranking to the operating system and scheduler to make sure that OS is choosing the cores with highest performance firstly for scheduling the process. When amd-pstate driver receives a message with the highest performance change, it will update the core ranking.

Changes form V6->V7:
- x86:
- - Modify kconfig about X86_AMD_PSTATE.
- cpufreq: amd-pstate:
- - modify incorrect comments about scheduler_work().
- - convert highest_perf data type.
- - modify preferred core init when cpu init and online.
- acpi: cppc:
- - modify link of CPPC highest performance.
- cpufreq:
- - modify link of CPPC highest performance changed.

Quoted 32 lines · Meng Li

Hi all:

The core frequency is subjected to the process variation in semiconductors. Not all cores are able to reach the maximum frequency respecting the infrastructure limits. Consequently, AMD has redefined the concept of maximum frequency of a part. This means that a fraction of cores can reach maximum frequency. To find the best process scheduling policy for a given scenario, OS needs to know the core ordering informed by the platform through highest performance capability register of the CPPC interface.

Earlier implementations of amd-pstate preferred core only support a static core ranking and targeted performance. Now it has the ability to dynamically change the preferred core based on the workload and platform conditions and accounting for thermals and aging.

Amd-pstate driver utilizes the functions and data structures provided by the ITMT architecture to enable the scheduler to favor scheduling on cores which can be get a higher frequency with lower voltage.
We call it amd-pstate preferred core.

Here sched_set_itmt_core_prio() is called to set priorities and sched_set_itmt_support() is called to enable ITMT feature. Amd-pstate driver uses the highest performance value to indicate the priority of CPU. The higher value has a higher priority.

Amd-pstate driver will provide an initial core ordering at boot time. It relies on the CPPC interface to communicate the core ranking to the operating system and scheduler to make sure that OS is choosing the cores with highest performance firstly for scheduling the process. When amd-pstate driver receives a message with the highest performance change, it will update the core ranking.

Quoted 12 lines

Changes form V6->V7:
- x86:
- - Modify kconfig about X86_AMD_PSTATE.
- cpufreq: amd-pstate:
- - modify incorrect comments about scheduler_work().
- - convert highest_perf data type.
- - modify preferred core init when cpu init and online.
- acpi: cppc:
- - modify link of CPPC highest performance.
- cpufreq:
- - modify link of CPPC highest performance changed.

This series in now in linux-kselftest next branch for Linux 6.7-rc1.

If there are any changes and/or fixes, please send patches on top of linux-kselftest next branch.

Hi all:

The core frequency is subjected to the process variation in semiconductors. Not all cores are able to reach the maximum frequency respecting the infrastructure limits. Consequently, AMD has redefined the concept of maximum frequency of a part. This means that a fraction of cores can reach maximum frequency. To find the best process scheduling policy for a given scenario, OS needs to know the core ordering informed by the platform through highest performance capability register of the CPPC interface.

Earlier implementations of amd-pstate preferred core only support a static core ranking and targeted performance. Now it has the ability to dynamically change the preferred core based on the workload and platform conditions and accounting for thermals and aging.

Amd-pstate driver utilizes the functions and data structures provided by the ITMT architecture to enable the scheduler to favor scheduling on cores which can be get a higher frequency with lower voltage.
We call it amd-pstate preferred core.

Here sched_set_itmt_core_prio() is called to set priorities and sched_set_itmt_support() is called to enable ITMT feature. Amd-pstate driver uses the highest performance value to indicate the priority of CPU. The higher value has a higher priority.

Amd-pstate driver will provide an initial core ordering at boot time. It relies on the CPPC interface to communicate the core ranking to the operating system and scheduler to make sure that OS is choosing the cores with highest performance firstly for scheduling the process. When amd-pstate driver receives a message with the highest performance change, it will update the core ranking.

Changes form V6->V7:
- x86:
- - Modify kconfig about X86_AMD_PSTATE.
- cpufreq: amd-pstate:
- - modify incorrect comments about scheduler_work().
- - convert highest_perf data type.
- - modify preferred core init when cpu init and online.
- acpi: cppc:
- - modify link of CPPC highest performance.
- cpufreq:
- - modify link of CPPC highest performance changed.

Changes form V5->V6:
- cpufreq: amd-pstate:
- - modify the wrong tag order.
- - modify warning about hw_prefcore sysfs attribute.
- - delete duplicate comments.
- - modify the variable name cppc_highest_perf to prefcore_ranking.
- - modify judgment conditions for setting highest_perf.
- - modify sysfs attribute for CPPC highest perf to pr_debug message.
- Documentation: amd-pstate:
- - modify warning: title underline too short.

Changes form V4->V5:
- cpufreq: amd-pstate:
- - modify sysfs attribute for CPPC highest perf.
- - modify warning about comments
- - rebase linux-next
- cpufreq:
- - Moidfy warning about function declarations.
- Documentation: amd-pstate:
- - align with ``amd-pstat``

Changes form V3->V4:
- Documentation: amd-pstate:
- - Modify inappropriate descriptions.

Changes form V2->V3:
- x86:
- - Modify kconfig and description.
- cpufreq: amd-pstate:
- - Add Co-developed-by tag in commit message.
- cpufreq:
- - Modify commit message.
- Documentation: amd-pstate:
- - Modify inappropriate descriptions.

Changes form V1->V2:
- acpi: cppc:
- - Add reference link.
- cpufreq:
- - Moidfy link error.
- cpufreq: amd-pstate:
- - Init the priorities of all online CPUs
- - Use a single variable to represent the status of preferred core.
- Documentation:
- - Default enabled preferred core.
- Documentation: amd-pstate:
- - Modify inappropriate descriptions.
- - Default enabled preferred core.
- - Use a single variable to represent the status of preferred core.

Meng Li (7):
x86: Drop CPU_SUP_INTEL from SCHED_MC_PRIO for the expansion.
acpi: cppc: Add get the highest performance cppc control
cpufreq: amd-pstate: Enable amd-pstate preferred core supporting.
cpufreq: Add a notification message that the highest perf has changed
cpufreq: amd-pstate: Update amd-pstate preferred core ranking
dynamically
Documentation: amd-pstate: introduce amd-pstate preferred core
Documentation: introduce amd-pstate preferrd core mode kernel command
line options

 .../admin-guide/kernel-parameters.txt         |   5 +
 Documentation/admin-guide/pm/amd-pstate.rst   |  58 +++++-
 arch/x86/Kconfig                              |   5 +-
 drivers/acpi/cppc_acpi.c                      |  13 ++
 drivers/acpi/processor_driver.c               |   6 +
 drivers/cpufreq/amd-pstate.c                  | 197 ++++++++++++++++--
 drivers/cpufreq/cpufreq.c                     |  13 ++
 include/acpi/cppc_acpi.h                      |   5 +
 include/linux/amd-pstate.h                    |   6 +
 include/linux/cpufreq.h                       |   5 +
 10 files changed, 291 insertions(+), 22 deletions(-)

Quoted 11 lines

Meng Li (7):
x86: Drop CPU_SUP_INTEL from SCHED_MC_PRIO for the expansion.
acpi: cppc: Add get the highest performance cppc control
cpufreq: amd-pstate: Enable amd-pstate preferred core supporting.
cpufreq: Add a notification message that the highest perf has changed
cpufreq: amd-pstate: Update amd-pstate preferred core ranking
dynamically
Documentation: amd-pstate: introduce amd-pstate preferred core
Documentation: introduce amd-pstate preferrd core mode kernel command
line options

Quoted 11 lines

  .../admin-guide/kernel-parameters.txt         |   5 +
  Documentation/admin-guide/pm/amd-pstate.rst   |  58 +++++-
  arch/x86/Kconfig                              |   5 +-
  drivers/acpi/cppc_acpi.c                      |  13 ++
  drivers/acpi/processor_driver.c               |   6 +
  drivers/cpufreq/amd-pstate.c                  | 197 ++++++++++++++++--
  drivers/cpufreq/cpufreq.c                     |  13 ++
  include/acpi/cppc_acpi.h                      |   5 +
  include/linux/amd-pstate.h                    |   6 +
  include/linux/cpufreq.h                       |   5 +
  10 files changed, 291 insertions(+), 22 deletions(-)

When applied on top of v6.5.3 this breaks turbo on my 5950X after suspend/resume cycle. Please see the scenario description below.

If I boot v6.5.3 + this patchset, then turbostat reports ~4.9 GHz on core 0 where taskset -c 0 dd if=/dev/zero of=/dev/null is being run.

After I suspend the machine and then resume it, and run dd again, turbostat reports the core to be capped to a stock frequency of ~3.4 GHz. Rebooting the machine fixes this, and the CPU can boost again.

If this patchset is reverted, then the CPU can turbo after suspend/resume cycle just fine.

I'm using amd_pstate=guided.

Is this behaviour expected?

Quoted 8 lines

When applied on top of v6.5.3 this breaks turbo on my 5950X after suspend/resume cycle. Please see the scenario description below.

If I boot v6.5.3 + this patchset, then turbostat reports ~4.9 GHz on core 0 where taskset -c 0 dd if=/dev/zero of=/dev/null is being run.

After I suspend the machine and then resume it, and run dd again, turbostat reports the core to be capped to a stock frequency of ~3.4 GHz. Rebooting the machine fixes this, and the CPU can boost again.

If this patchset is reverted, then the CPU can turbo after suspend/resume cycle just fine.

Quoted 3 lines

I'm using amd_pstate=guided.

Is this behaviour expected?

To help confirm where the issue is, can I ask you to do three experiments with the patch series applied:

1) 'amd_pstate=active' on your kernel command line.

2) 'amd_pstate=active amd_prefcore=disable' on your kernel command line.

3) 'amd_pstate=guided amd_prefcore=disable' on your kernel command line.

Looking through the code, I anticipate from your report that it reproduces on "1" but not "2" and "3".

Meng,

Can you try to repro?

I think that it's probably a call to amd_pstate_init_prefcore() missing from amd_pstate_cpu_resume() and also amd_pstate_epp_resume().

Hello.

Quoted 8 lines · On středa 20. září 2023 18:56:09 CEST Mario Limonciello wrote:

Quoted 8 lines

When applied on top of v6.5.3 this breaks turbo on my 5950X after suspend/resume cycle. Please see the scenario description below.

If I boot v6.5.3 + this patchset, then turbostat reports ~4.9 GHz on core 0 where taskset -c 0 dd if=/dev/zero of=/dev/null is being run.

After I suspend the machine and then resume it, and run dd again, turbostat reports the core to be capped to a stock frequency of ~3.4 GHz. Rebooting the machine fixes this, and the CPU can boost again.

If this patchset is reverted, then the CPU can turbo after suspend/resume cycle just fine.

Quoted 8 lines

Quoted 3 lines

I'm using amd_pstate=guided.

Is this behaviour expected?

To help confirm where the issue is, can I ask you to do three experiments with the patch series applied:

1) 'amd_pstate=active' on your kernel command line.

The issue is reproducible. If I toggle the governor in cpupower to powersave and back to performance, boost is restored.

Quoted 1 line

2) 'amd_pstate=active amd_prefcore=disable' on your kernel command line.

The issue is not reproducible.

Quoted 1 line

3) 'amd_pstate=guided amd_prefcore=disable' on your kernel command line.

The issue is not reproducible.

I should also mention that in my initial configuration I use amd_pstate=guided and schedutil. If I switch to performance after suspend-resume cycle, the boost is restored. However, if I switch back to schedutil, the freq is capped.

Does this info help?

Quoted 9 lines

Looking through the code, I anticipate from your report that it reproduces on "1" but not "2" and "3".

Meng,

Can you try to repro?

I think that it's probably a call to amd_pstate_init_prefcore() missing from amd_pstate_cpu_resume() and also amd_pstate_epp_resume().

-----Original Message-----
From: Limonciello, Mario <[email protected]>
Sent: Thursday, September 21, 2023 4:12 AM
To: Oleksandr Natalenko <[email protected]>; Huang, Ray <[email protected]>; Meng, Li (Jassmine) <[email protected]>
Cc: [email protected]; [email protected]; [email protected]; [email protected]; Shuah Khan <[email protected]>; [email protected]; Fontenot, Nathan <[email protected]>; Sharma, Deepak <[email protected]>; Deucher, Alexander <[email protected]>; Huang, Shimmer <[email protected]>; Yuan, Perry <[email protected]>; Du, Xiaojian <[email protected]>; Viresh Kumar <[email protected]>; Borislav Petkov <[email protected]>; Rafael J . Wysocki <[email protected]>
Subject: Re: [PATCH V7 0/7] amd-pstate preferred core

Quoted 4 lines · Oleksandr Natalenko

Hello.

Quoted 1 line · On středa 20. září 2023 18:56:09 CEST Mario Limonciello wrote:

Quoted 1 line

When applied on top of v6.5.3 this breaks turbo on my 5950X after

suspend/resume cycle. Please see the scenario description below.

Quoted 2 lines

Quoted 2 lines

Quoted 2 lines

If I boot v6.5.3 + this patchset, then turbostat reports ~4.9 GHz on core

0 where taskset -c 0 dd if=/dev/zero of=/dev/null is being run.

Quoted 2 lines

Quoted 2 lines

Quoted 2 lines

After I suspend the machine and then resume it, and run dd again,

turbostat reports the core to be capped to a stock frequency of ~3.4 GHz. Rebooting the machine fixes this, and the CPU can boost again.

Quoted 2 lines

Quoted 2 lines

Quoted 2 lines

If this patchset is reverted, then the CPU can turbo after

suspend/resume cycle just fine.

Quoted 11 lines

Quoted 9 lines

Quoted 4 lines

I'm using amd_pstate=guided.

Is this behaviour expected?

To help confirm where the issue is, can I ask you to do three experiments with the patch series applied:

1) 'amd_pstate=active' on your kernel command line.

The issue is reproducible. If I toggle the governor in cpupower to

powersave and back to performance, boost is restored.

Quoted 2 lines

Quoted 2 lines

2) 'amd_pstate=active amd_prefcore=disable' on your kernel command

line.

Quoted 4 lines

The issue is not reproducible.

Quoted 1 line

3) 'amd_pstate=guided amd_prefcore=disable' on your kernel command

line.

Quoted 4 lines

The issue is not reproducible.

I should also mention that in my initial configuration I use

amd_pstate=guided and schedutil. If I switch to performance after suspend-resume cycle, the boost is restored. However, if I switch back to schedutil, the freq is capped.

Quoted 1 line

Quoted 2 lines

Does this info help?

Yeah, it matches my expectations for this issue you reported. Thanks!

Jassmine can dig into a fix for another spin of this series.

Quoted 10 lines

Quoted 10 lines

Looking through the code, I anticipate from your report that it reproduces on "1" but not "2" and "3".

Meng,

Can you try to repro?

I think that it's probably a call to amd_pstate_init_prefcore() missing from amd_pstate_cpu_resume() and also

amd_pstate_epp_resume().

Quoted 1 line

Hi all:

The core frequency is subjected to the process variation in semiconductors. Not all cores are able to reach the maximum frequency respecting the infrastructure limits. Consequently, AMD has redefined the concept of maximum frequency of a part. This means that a fraction of cores can reach maximum frequency. To find the best process scheduling policy for a given scenario, OS needs to know the core ordering informed by the platform through highest performance capability register of the CPPC interface.

Earlier implementations of amd-pstate preferred core only support a static core ranking and targeted performance. Now it has the ability to dynamically change the preferred core based on the workload and platform conditions and accounting for thermals and aging.

Amd-pstate driver utilizes the functions and data structures provided by the ITMT architecture to enable the scheduler to favor scheduling on cores which can be get a higher frequency with lower voltage.
We call it amd-pstate preferred core.

Here sched_set_itmt_core_prio() is called to set priorities and sched_set_itmt_support() is called to enable ITMT feature. Amd-pstate driver uses the highest performance value to indicate the priority of CPU. The higher value has a higher priority.

Amd-pstate driver will provide an initial core ordering at boot time. It relies on the CPPC interface to communicate the core ranking to the operating system and scheduler to make sure that OS is choosing the cores with highest performance firstly for scheduling the process. When amd-pstate driver receives a message with the highest performance change, it will update the core ranking.

Changes form V6->V7:
- x86:
- - Modify kconfig about X86_AMD_PSTATE.
- cpufreq: amd-pstate:
- - modify incorrect comments about scheduler_work().
- - convert highest_perf data type.
- - modify preferred core init when cpu init and online.
- acpi: cppc:
- - modify link of CPPC highest performance.
- cpufreq:
- - modify link of CPPC highest performance changed.

Changes form V5->V6:
- cpufreq: amd-pstate:
- - modify the wrong tag order.
- - modify warning about hw_prefcore sysfs attribute.
- - delete duplicate comments.
- - modify the variable name cppc_highest_perf to prefcore_ranking.
- - modify judgment conditions for setting highest_perf.
- - modify sysfs attribute for CPPC highest perf to pr_debug message.
- Documentation: amd-pstate:
- - modify warning: title underline too short.

Changes form V4->V5:
- cpufreq: amd-pstate:
- - modify sysfs attribute for CPPC highest perf.
- - modify warning about comments
- - rebase linux-next
- cpufreq:
- - Moidfy warning about function declarations.
- Documentation: amd-pstate:
- - align with ``amd-pstat``

Changes form V3->V4:
- Documentation: amd-pstate:
- - Modify inappropriate descriptions.

Changes form V2->V3:
- x86:
- - Modify kconfig and description.
- cpufreq: amd-pstate:
- - Add Co-developed-by tag in commit message.
- cpufreq:
- - Modify commit message.
- Documentation: amd-pstate:
- - Modify inappropriate descriptions.

Changes form V1->V2:
- acpi: cppc:
- - Add reference link.
- cpufreq:
- - Moidfy link error.
- cpufreq: amd-pstate:
- - Init the priorities of all online CPUs
- - Use a single variable to represent the status of preferred core.
- Documentation:
- - Default enabled preferred core.
- Documentation: amd-pstate:
- - Modify inappropriate descriptions.
- - Default enabled preferred core.
- - Use a single variable to represent the status of preferred core.

Meng Li (7):
x86: Drop CPU_SUP_INTEL from SCHED_MC_PRIO for the expansion.
acpi: cppc: Add get the highest performance cppc control
cpufreq: amd-pstate: Enable amd-pstate preferred core supporting.
cpufreq: Add a notification message that the highest perf has changed
cpufreq: amd-pstate: Update amd-pstate preferred core ranking
dynamically
Documentation: amd-pstate: introduce amd-pstate preferred core
Documentation: introduce amd-pstate preferrd core mode kernel command
line options

 .../admin-guide/kernel-parameters.txt         |   5 +
 Documentation/admin-guide/pm/amd-pstate.rst   |  58 +++++-
 arch/x86/Kconfig                              |   5 +-
 drivers/acpi/cppc_acpi.c                      |  13 ++
 drivers/acpi/processor_driver.c               |   6 +
 drivers/cpufreq/amd-pstate.c                  | 197 ++++++++++++++++--
 drivers/cpufreq/cpufreq.c                     |  13 ++
 include/acpi/cppc_acpi.h                      |   5 +
 include/linux/amd-pstate.h                    |   6 +
 include/linux/cpufreq.h                       |   5 +
 10 files changed, 291 insertions(+), 22 deletions(-)

Signature · 1 line

2.34.1