[PATCH v2 0/5] intel-uncore-freq: Add agent_types and die_id attributes

[PATCH v2 0/5] intel-uncore-freq: Add agent_types and die_id attributes

[Srinivas Pandruvada]

Add two new attributes, so that orchestration software like Kubernetes can
target specific dies and agents for uncore frequency control.

v2:
In patch 5/5 fix grammar as reported by Alok Tiwari

Srinivas Pandruvada (5):
  platform/x86/intel-uncore-freq: Add attributes to show agent types
  Documentation: admin-guide: pm: Add documentation for agent_types
  platform/x86/intel: power-domains: Add interface to get Linux die ID
  platform/x86/intel-uncore-freq: Add attributes to show die_id
  Documentation: admin-guide: pm: Add documentation for die_id

 .../pm/intel_uncore_frequency_scaling.rst     | 10 ++++
 .../platform/x86/intel/tpmi_power_domains.c   | 34 +++++++++--
 .../platform/x86/intel/tpmi_power_domains.h   |  1 +
 .../uncore-frequency-common.c                 | 31 ++++++++++
 .../uncore-frequency-common.h                 | 19 ++++++-
 .../uncore-frequency/uncore-frequency-tpmi.c  | 56 +++++++++++++++++++
 6 files changed, 146 insertions(+), 5 deletions(-)

-- 
2.48.1

[PATCH v2 1/5] platform/x86/intel-uncore-freq: Add attributes to show agent types

[Srinivas Pandruvada]

Currently, users need detailed hardware information to understand the
scope of controls within each uncore domain. Uncore frequency controls
manage subsystems such as core, cache, memory, and I/O. The UFS TPMI
provides this information, which can be used to present the scope more
clearly.

Each uncore domain consists of one or more agent types, with each agent
type controlling one or more uncore hardware subsystems. For example, a
single agent might control both the core and cache.

Introduce a new attribute called "agent_types." This attribute displays
a list of agents, separated by space character.

The string representations for agent types are as follows:
	For core agent: core
	For cache agent: cache
	For memory agent: memory
	For I/O agent: io

These agent types are read during probe time for each cluster and stored
as part of the struct uncore_data.

Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
---
v2:
No change

 .../uncore-frequency-common.c                 | 24 ++++++++++++++++
 .../uncore-frequency-common.h                 | 17 ++++++++++-
 .../uncore-frequency/uncore-frequency-tpmi.c  | 28 +++++++++++++++++++
 3 files changed, 68 insertions(+), 1 deletion(-)

diff --git a/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-common.c b/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-common.c
index 4e2c6a2d7e6e..cfa3039a0e39 100644
--- a/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-common.c
+++ b/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-common.c
@@ -43,6 +43,28 @@ static ssize_t show_package_id(struct kobject *kobj, struct kobj_attribute *attr
 	return sprintf(buf, "%u\n", data->package_id);
 }
 
+static ssize_t show_agent_types(struct kobject *kobj, struct kobj_attribute *attr, char *buf)
+{
+	struct uncore_data *data = container_of(attr, struct uncore_data, agent_types_kobj_attr);
+	int length = 0;
+
+	if (data->agent_type_mask & AGENT_TYPE_CORE)
+		length += sysfs_emit_at(buf, length, "core ");
+
+	if (data->agent_type_mask & AGENT_TYPE_CACHE)
+		length += sysfs_emit_at(buf, length, "cache ");
+
+	if (data->agent_type_mask & AGENT_TYPE_MEMORY)
+		length += sysfs_emit_at(buf, length, "memory ");
+
+	if (data->agent_type_mask & AGENT_TYPE_IO)
+		length += sysfs_emit_at(buf, length, "io ");
+
+	length += sysfs_emit_at(buf, length, "\n");
+
+	return length;
+}
+
 static ssize_t show_attr(struct uncore_data *data, char *buf, enum uncore_index index)
 {
 	unsigned int value;
@@ -179,6 +201,8 @@ static int create_attr_group(struct uncore_data *data, char *name)
 		data->uncore_attrs[index++] = &data->fabric_cluster_id_kobj_attr.attr;
 		init_attribute_root_ro(package_id);
 		data->uncore_attrs[index++] = &data->package_id_kobj_attr.attr;
+		init_attribute_ro(agent_types);
+		data->uncore_attrs[index++] = &data->agent_types_kobj_attr.attr;
 	}
 
 	data->uncore_attrs[index++] = &data->max_freq_khz_kobj_attr.attr;
diff --git a/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-common.h b/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-common.h
index 26c854cd5d97..197ca2ad327f 100644
--- a/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-common.h
+++ b/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-common.h
@@ -11,6 +11,17 @@
 
 #include <linux/device.h>
 
+/*
+ * Define uncore agents, which are under uncore frequency control.
+ * It is possible that there are common uncore frequency control to more than
+ * one agents. So these defines are used as a bit mask.
+ */
+#define	AGENT_TYPE_NONE		0
+#define AGENT_TYPE_CORE		0x01
+#define	AGENT_TYPE_CACHE	0x02
+#define	AGENT_TYPE_MEMORY	0x04
+#define	AGENT_TYPE_IO		0x08
+
 /**
  * struct uncore_data - Encapsulate all uncore data
  * @stored_uncore_data: Last user changed MSR 620 value, which will be restored
@@ -25,6 +36,7 @@
  * @cluster_id:		cluster id in a domain
  * @instance_id:	Unique instance id to append to directory name
  * @name:		Sysfs entry name for this instance
+ * @agent_type_mask:	Bit mask of all hardware agents for this domain
  * @uncore_attr_group:	Attribute group storage
  * @max_freq_khz_kobj_attr: Storage for kobject attribute max_freq_khz
  * @mix_freq_khz_kobj_attr: Storage for kobject attribute min_freq_khz
@@ -41,6 +53,7 @@
  * @elc_high_threshold_enable_kobj_attr:
 		Storage for kobject attribute elc_high_threshold_enable
  * @elc_floor_freq_khz_kobj_attr: Storage for kobject attribute elc_floor_freq_khz
+ * @agent_types_kobj_attr: Storage for kobject attribute agent_type
  * @uncore_attrs:	Attribute storage for group creation
  *
  * This structure is used to encapsulate all data related to uncore sysfs
@@ -58,6 +71,7 @@ struct uncore_data {
 	int cluster_id;
 	int instance_id;
 	char name[32];
+	u16  agent_type_mask;
 
 	struct attribute_group uncore_attr_group;
 	struct kobj_attribute max_freq_khz_kobj_attr;
@@ -72,7 +86,8 @@ struct uncore_data {
 	struct kobj_attribute elc_high_threshold_percent_kobj_attr;
 	struct kobj_attribute elc_high_threshold_enable_kobj_attr;
 	struct kobj_attribute elc_floor_freq_khz_kobj_attr;
-	struct attribute *uncore_attrs[13];
+	struct kobj_attribute agent_types_kobj_attr;
+	struct attribute *uncore_attrs[14];
 };
 
 #define UNCORE_DOMAIN_ID_INVALID	-1
diff --git a/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-tpmi.c b/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-tpmi.c
index 4aa6c227ec82..70415d80773c 100644
--- a/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-tpmi.c
+++ b/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-tpmi.c
@@ -74,6 +74,12 @@ struct tpmi_uncore_struct {
 /* Bit definitions for STATUS register */
 #define UNCORE_CURRENT_RATIO_MASK			GENMASK_ULL(6, 0)
 
+/* Uncore agent type bits */
+#define UNCORE_CORE_PRESENT				BIT(23)
+#define UNCORE_CACHE_PRESENT				BIT(24)
+#define UNCORE_MEMORY_PRESENT				BIT(25)
+#define UNCORE_IO_PRESENT				BIT(26)
+
 /* Bit definitions for CONTROL register */
 #define UNCORE_MAX_RATIO_MASK				GENMASK_ULL(14, 8)
 #define UNCORE_MIN_RATIO_MASK				GENMASK_ULL(21, 15)
@@ -347,6 +353,26 @@ static int uncore_read_freq(struct uncore_data *data, unsigned int *freq)
 	return 0;
 }
 
+/* Helper function to read agent type over MMIO and set the agent type */
+static void uncore_set_agent_type(struct tpmi_uncore_cluster_info *cluster_info)
+{
+	u64 status;
+
+	status = readq((u8 __iomem *)cluster_info->cluster_base + UNCORE_STATUS_INDEX);
+
+	if (status & UNCORE_CORE_PRESENT)
+		cluster_info->uncore_data.agent_type_mask |= AGENT_TYPE_CORE;
+
+	if (status & UNCORE_CACHE_PRESENT)
+		cluster_info->uncore_data.agent_type_mask |= AGENT_TYPE_CACHE;
+
+	if (status & UNCORE_MEMORY_PRESENT)
+		cluster_info->uncore_data.agent_type_mask |= AGENT_TYPE_MEMORY;
+
+	if (status & UNCORE_IO_PRESENT)
+		cluster_info->uncore_data.agent_type_mask |= AGENT_TYPE_IO;
+}
+
 /* Callback for sysfs read for TPMI uncore values. Called under mutex locks. */
 static int uncore_read(struct uncore_data *data, unsigned int *value, enum uncore_index index)
 {
@@ -552,6 +578,8 @@ static int uncore_probe(struct auxiliary_device *auxdev, const struct auxiliary_
 
 			cluster_info->cluster_base = pd_info->uncore_base + mask;
 
+			uncore_set_agent_type(cluster_info);
+
 			cluster_info->uncore_data.package_id = pkg;
 			/* There are no dies like Cascade Lake */
 			cluster_info->uncore_data.die_id = 0;
-- 
2.48.1

Re: [PATCH v2 1/5] platform/x86/intel-uncore-freq: Add attributes to show agent types

[Ilpo Järvinen]

On Mon, 28 Apr 2025, Srinivas Pandruvada wrote:

> Currently, users need detailed hardware information to understand the
> scope of controls within each uncore domain. Uncore frequency controls
> manage subsystems such as core, cache, memory, and I/O. The UFS TPMI
> provides this information, which can be used to present the scope more
> clearly.
> 
> Each uncore domain consists of one or more agent types, with each agent
> type controlling one or more uncore hardware subsystems. For example, a
> single agent might control both the core and cache.
> 
> Introduce a new attribute called "agent_types." This attribute displays
> a list of agents, separated by space character.
> 
> The string representations for agent types are as follows:
> 	For core agent: core
> 	For cache agent: cache
> 	For memory agent: memory
> 	For I/O agent: io
> 
> These agent types are read during probe time for each cluster and stored
> as part of the struct uncore_data.
> 
> Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
> ---
> v2:
> No change
> 
>  .../uncore-frequency-common.c                 | 24 ++++++++++++++++
>  .../uncore-frequency-common.h                 | 17 ++++++++++-
>  .../uncore-frequency/uncore-frequency-tpmi.c  | 28 +++++++++++++++++++
>  3 files changed, 68 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-common.c b/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-common.c
> index 4e2c6a2d7e6e..cfa3039a0e39 100644
> --- a/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-common.c
> +++ b/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-common.c
> @@ -43,6 +43,28 @@ static ssize_t show_package_id(struct kobject *kobj, struct kobj_attribute *attr
>  	return sprintf(buf, "%u\n", data->package_id);
>  }
>  
> +static ssize_t show_agent_types(struct kobject *kobj, struct kobj_attribute *attr, char *buf)
> +{
> +	struct uncore_data *data = container_of(attr, struct uncore_data, agent_types_kobj_attr);
> +	int length = 0;
> +
> +	if (data->agent_type_mask & AGENT_TYPE_CORE)
> +		length += sysfs_emit_at(buf, length, "core ");
> +
> +	if (data->agent_type_mask & AGENT_TYPE_CACHE)
> +		length += sysfs_emit_at(buf, length, "cache ");
> +
> +	if (data->agent_type_mask & AGENT_TYPE_MEMORY)
> +		length += sysfs_emit_at(buf, length, "memory ");
> +
> +	if (data->agent_type_mask & AGENT_TYPE_IO)
> +		length += sysfs_emit_at(buf, length, "io ");

Is this set going to get expanded soon? It would feel more future proof to
do this mapping using a loop and array. You also chose the quick and dirty 
approach wrt. trailing spaces as getting rid of the extra space is a bit 
tedious when open coding the mapping like that ;-).

-- 
 i.

> +
> +	length += sysfs_emit_at(buf, length, "\n");
> +
> +	return length;
> +}
> +
>  static ssize_t show_attr(struct uncore_data *data, char *buf, enum uncore_index index)
>  {
>  	unsigned int value;
> @@ -179,6 +201,8 @@ static int create_attr_group(struct uncore_data *data, char *name)
>  		data->uncore_attrs[index++] = &data->fabric_cluster_id_kobj_attr.attr;
>  		init_attribute_root_ro(package_id);
>  		data->uncore_attrs[index++] = &data->package_id_kobj_attr.attr;
> +		init_attribute_ro(agent_types);
> +		data->uncore_attrs[index++] = &data->agent_types_kobj_attr.attr;
>  	}
>  
>  	data->uncore_attrs[index++] = &data->max_freq_khz_kobj_attr.attr;
> diff --git a/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-common.h b/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-common.h
> index 26c854cd5d97..197ca2ad327f 100644
> --- a/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-common.h
> +++ b/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-common.h
> @@ -11,6 +11,17 @@
>  
>  #include <linux/device.h>
>  
> +/*
> + * Define uncore agents, which are under uncore frequency control.
> + * It is possible that there are common uncore frequency control to more than
> + * one agents. So these defines are used as a bit mask.
> + */
> +#define	AGENT_TYPE_NONE		0
> +#define AGENT_TYPE_CORE		0x01
> +#define	AGENT_TYPE_CACHE	0x02
> +#define	AGENT_TYPE_MEMORY	0x04
> +#define	AGENT_TYPE_IO		0x08
> +
>  /**
>   * struct uncore_data - Encapsulate all uncore data
>   * @stored_uncore_data: Last user changed MSR 620 value, which will be restored
> @@ -25,6 +36,7 @@
>   * @cluster_id:		cluster id in a domain
>   * @instance_id:	Unique instance id to append to directory name
>   * @name:		Sysfs entry name for this instance
> + * @agent_type_mask:	Bit mask of all hardware agents for this domain
>   * @uncore_attr_group:	Attribute group storage
>   * @max_freq_khz_kobj_attr: Storage for kobject attribute max_freq_khz
>   * @mix_freq_khz_kobj_attr: Storage for kobject attribute min_freq_khz
> @@ -41,6 +53,7 @@
>   * @elc_high_threshold_enable_kobj_attr:
>  		Storage for kobject attribute elc_high_threshold_enable
>   * @elc_floor_freq_khz_kobj_attr: Storage for kobject attribute elc_floor_freq_khz
> + * @agent_types_kobj_attr: Storage for kobject attribute agent_type
>   * @uncore_attrs:	Attribute storage for group creation
>   *
>   * This structure is used to encapsulate all data related to uncore sysfs
> @@ -58,6 +71,7 @@ struct uncore_data {
>  	int cluster_id;
>  	int instance_id;
>  	char name[32];
> +	u16  agent_type_mask;
>  
>  	struct attribute_group uncore_attr_group;
>  	struct kobj_attribute max_freq_khz_kobj_attr;
> @@ -72,7 +86,8 @@ struct uncore_data {
>  	struct kobj_attribute elc_high_threshold_percent_kobj_attr;
>  	struct kobj_attribute elc_high_threshold_enable_kobj_attr;
>  	struct kobj_attribute elc_floor_freq_khz_kobj_attr;
> -	struct attribute *uncore_attrs[13];
> +	struct kobj_attribute agent_types_kobj_attr;
> +	struct attribute *uncore_attrs[14];
>  };
>  
>  #define UNCORE_DOMAIN_ID_INVALID	-1
> diff --git a/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-tpmi.c b/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-tpmi.c
> index 4aa6c227ec82..70415d80773c 100644
> --- a/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-tpmi.c
> +++ b/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-tpmi.c
> @@ -74,6 +74,12 @@ struct tpmi_uncore_struct {
>  /* Bit definitions for STATUS register */
>  #define UNCORE_CURRENT_RATIO_MASK			GENMASK_ULL(6, 0)
>  
> +/* Uncore agent type bits */
> +#define UNCORE_CORE_PRESENT				BIT(23)
> +#define UNCORE_CACHE_PRESENT				BIT(24)
> +#define UNCORE_MEMORY_PRESENT				BIT(25)
> +#define UNCORE_IO_PRESENT				BIT(26)
> +
>  /* Bit definitions for CONTROL register */
>  #define UNCORE_MAX_RATIO_MASK				GENMASK_ULL(14, 8)
>  #define UNCORE_MIN_RATIO_MASK				GENMASK_ULL(21, 15)
> @@ -347,6 +353,26 @@ static int uncore_read_freq(struct uncore_data *data, unsigned int *freq)
>  	return 0;
>  }
>  
> +/* Helper function to read agent type over MMIO and set the agent type */
> +static void uncore_set_agent_type(struct tpmi_uncore_cluster_info *cluster_info)
> +{
> +	u64 status;
> +
> +	status = readq((u8 __iomem *)cluster_info->cluster_base + UNCORE_STATUS_INDEX);
> +
> +	if (status & UNCORE_CORE_PRESENT)
> +		cluster_info->uncore_data.agent_type_mask |= AGENT_TYPE_CORE;
> +
> +	if (status & UNCORE_CACHE_PRESENT)
> +		cluster_info->uncore_data.agent_type_mask |= AGENT_TYPE_CACHE;
> +
> +	if (status & UNCORE_MEMORY_PRESENT)
> +		cluster_info->uncore_data.agent_type_mask |= AGENT_TYPE_MEMORY;
> +
> +	if (status & UNCORE_IO_PRESENT)
> +		cluster_info->uncore_data.agent_type_mask |= AGENT_TYPE_IO;
> +}
> +
>  /* Callback for sysfs read for TPMI uncore values. Called under mutex locks. */
>  static int uncore_read(struct uncore_data *data, unsigned int *value, enum uncore_index index)
>  {
> @@ -552,6 +578,8 @@ static int uncore_probe(struct auxiliary_device *auxdev, const struct auxiliary_
>  
>  			cluster_info->cluster_base = pd_info->uncore_base + mask;
>  
> +			uncore_set_agent_type(cluster_info);
> +
>  			cluster_info->uncore_data.package_id = pkg;
>  			/* There are no dies like Cascade Lake */
>  			cluster_info->uncore_data.die_id = 0;
> 

Re: [PATCH v2 1/5] platform/x86/intel-uncore-freq: Add attributes to show agent types

[srinivas pandruvada]

On Wed, 2025-05-07 at 16:02 +0300, Ilpo Järvinen wrote:
> On Mon, 28 Apr 2025, Srinivas Pandruvada wrote:
> 
> > Currently, users need detailed hardware information to understand
> > the
> > scope of controls within each uncore domain. Uncore frequency
> > controls
> > manage subsystems such as core, cache, memory, and I/O. The UFS
> > TPMI
> > provides this information, which can be used to present the scope
> > more
> > clearly.
> > 
> > Each uncore domain consists of one or more agent types, with each
> > agent
> > type controlling one or more uncore hardware subsystems. For
> > example, a
> > single agent might control both the core and cache.
> > 
> > Introduce a new attribute called "agent_types." This attribute
> > displays
> > a list of agents, separated by space character.
> > 
> > The string representations for agent types are as follows:
> > 	For core agent: core
> > 	For cache agent: cache
> > 	For memory agent: memory
> > 	For I/O agent: io
> > 
> > These agent types are read during probe time for each cluster and
> > stored
> > as part of the struct uncore_data.
> > 
> > Signed-off-by: Srinivas Pandruvada
> > <srinivas.pandruvada@linux.intel.com>
> > ---
> > v2:
> > No change
> > 
> >  .../uncore-frequency-common.c                 | 24
> > ++++++++++++++++
> >  .../uncore-frequency-common.h                 | 17 ++++++++++-
> >  .../uncore-frequency/uncore-frequency-tpmi.c  | 28
> > +++++++++++++++++++
> >  3 files changed, 68 insertions(+), 1 deletion(-)
> > 
> > diff --git a/drivers/platform/x86/intel/uncore-frequency/uncore-
> > frequency-common.c b/drivers/platform/x86/intel/uncore-
> > frequency/uncore-frequency-common.c
> > index 4e2c6a2d7e6e..cfa3039a0e39 100644
> > --- a/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-
> > common.c
> > +++ b/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-
> > common.c
> > @@ -43,6 +43,28 @@ static ssize_t show_package_id(struct kobject
> > *kobj, struct kobj_attribute *attr
> >  	return sprintf(buf, "%u\n", data->package_id);
> >  }
> >  
> > +static ssize_t show_agent_types(struct kobject *kobj, struct
> > kobj_attribute *attr, char *buf)
> > +{
> > +	struct uncore_data *data = container_of(attr, struct
> > uncore_data, agent_types_kobj_attr);
> > +	int length = 0;
> > +
> > +	if (data->agent_type_mask & AGENT_TYPE_CORE)
> > +		length += sysfs_emit_at(buf, length, "core ");
> > +
> > +	if (data->agent_type_mask & AGENT_TYPE_CACHE)
> > +		length += sysfs_emit_at(buf, length, "cache ");
> > +
> > +	if (data->agent_type_mask & AGENT_TYPE_MEMORY)
> > +		length += sysfs_emit_at(buf, length, "memory ");
> > +
> > +	if (data->agent_type_mask & AGENT_TYPE_IO)
> > +		length += sysfs_emit_at(buf, length, "io ");
> 
> Is this set going to get expanded soon?

Unlikely, as this list is adding every frequency scaled subsystem.

But I will try to change to loop. No issue.

Thanks,
Srinivas


>  It would feel more future proof to
> do this mapping using a loop and array. You also chose the quick and
> dirty 
> approach wrt. trailing spaces as getting rid of the extra space is a
> bit 
> tedious when open coding the mapping like that ;-).
> 

[PATCH v2 2/5] Documentation: admin-guide: pm: Add documentation for agent_types

[Srinivas Pandruvada]

Add documentation to describe agent_types attribute.

Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
---
v2:
No change

 .../admin-guide/pm/intel_uncore_frequency_scaling.rst        | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/Documentation/admin-guide/pm/intel_uncore_frequency_scaling.rst b/Documentation/admin-guide/pm/intel_uncore_frequency_scaling.rst
index 5151ec312dc0..84608dad84bd 100644
--- a/Documentation/admin-guide/pm/intel_uncore_frequency_scaling.rst
+++ b/Documentation/admin-guide/pm/intel_uncore_frequency_scaling.rst
@@ -97,6 +97,11 @@ Attributes in each directory:
 ``package_id``
 	This attribute is used to get the package id of this instance.
 
+``agent_types``
+	This attribute displays all the hardware agents present within the
+	domain. Each agent has the capability to control one or more hardware
+	subsystems, which include: core, cache, memory, and I/O.
+
 The other attributes are same as presented at package_*_die_* level.
 
 In most of current use cases, the "max_freq_khz" and "min_freq_khz"
-- 
2.48.1

[PATCH v2 3/5] platform/x86/intel: power-domains: Add interface to get Linux die ID

[Srinivas Pandruvada]

The die ID in the Linux topology sysfs is a logical identifier that
differs from the one presented in CPUID leaf 0x1F or via MSR 0x54.

Introduce an interface that returns the Linux CPU die ID based on a
given package ID and power domain ID. This mapping is stored during the
CPU online callback in an array.

Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
---
v2:
No change

 .../platform/x86/intel/tpmi_power_domains.c   | 34 ++++++++++++++++---
 .../platform/x86/intel/tpmi_power_domains.h   |  1 +
 2 files changed, 31 insertions(+), 4 deletions(-)

diff --git a/drivers/platform/x86/intel/tpmi_power_domains.c b/drivers/platform/x86/intel/tpmi_power_domains.c
index 2f01cd22a6ee..9aaebd74a2af 100644
--- a/drivers/platform/x86/intel/tpmi_power_domains.c
+++ b/drivers/platform/x86/intel/tpmi_power_domains.c
@@ -74,6 +74,8 @@ static enum cpuhp_state tpmi_hp_state __read_mostly;
 
 static cpumask_t *tpmi_power_domain_mask;
 
+static u16 *domain_die_map;
+
 /* Lock to protect tpmi_power_domain_mask and tpmi_cpu_hash */
 static DEFINE_MUTEX(tpmi_lock);
 
@@ -152,6 +154,15 @@ cpumask_t *tpmi_get_power_domain_mask(int cpu_no)
 }
 EXPORT_SYMBOL_NS_GPL(tpmi_get_power_domain_mask, "INTEL_TPMI_POWER_DOMAIN");
 
+int tpmi_get_linux_die_id(int pkg_id, int domain_id)
+{
+	if (pkg_id >= topology_max_packages() || domain_id >= MAX_POWER_DOMAINS)
+		return -EINVAL;
+
+	return domain_die_map[pkg_id * MAX_POWER_DOMAINS + domain_id];
+}
+EXPORT_SYMBOL_NS_GPL(tpmi_get_linux_die_id, "INTEL_TPMI_POWER_DOMAIN");
+
 static int tpmi_get_logical_id(unsigned int cpu, struct tpmi_cpu_info *info)
 {
 	u64 data;
@@ -189,6 +200,9 @@ static int tpmi_cpu_online(unsigned int cpu)
 	cpumask_set_cpu(cpu, &tpmi_power_domain_mask[index]);
 	hash_add(tpmi_cpu_hash, &info->hnode, info->punit_core_id);
 
+	domain_die_map[info->pkg_id * MAX_POWER_DOMAINS + info->punit_domain_id] =
+			topology_die_id(cpu);
+
 	return 0;
 }
 
@@ -212,17 +226,28 @@ static int __init tpmi_init(void)
 	if (!tpmi_power_domain_mask)
 		return -ENOMEM;
 
+	domain_die_map = kcalloc(size_mul(topology_max_packages(), MAX_POWER_DOMAINS),
+				 sizeof(*domain_die_map), GFP_KERNEL);
+	if (!domain_die_map)
+		goto free_domain_mask;
+
 	ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN,
 				"platform/x86/tpmi_power_domains:online",
 				tpmi_cpu_online, NULL);
-	if (ret < 0) {
-		kfree(tpmi_power_domain_mask);
-		return ret;
-	}
+	if (ret < 0)
+		goto free_domain_map;
 
 	tpmi_hp_state = ret;
 
 	return 0;
+
+free_domain_map:
+	kfree(domain_die_map);
+
+free_domain_mask:
+	kfree(tpmi_power_domain_mask);
+
+	return ret;
 }
 module_init(tpmi_init)
 
@@ -230,6 +255,7 @@ static void __exit tpmi_exit(void)
 {
 	cpuhp_remove_state(tpmi_hp_state);
 	kfree(tpmi_power_domain_mask);
+	kfree(domain_die_map);
 }
 module_exit(tpmi_exit)
 
diff --git a/drivers/platform/x86/intel/tpmi_power_domains.h b/drivers/platform/x86/intel/tpmi_power_domains.h
index e35750dd9273..2fd0dd7afbd2 100644
--- a/drivers/platform/x86/intel/tpmi_power_domains.h
+++ b/drivers/platform/x86/intel/tpmi_power_domains.h
@@ -14,5 +14,6 @@ int tpmi_get_linux_cpu_number(int package_id, int die_id, int punit_core_id);
 int tpmi_get_punit_core_number(int cpu_no);
 int tpmi_get_power_domain_id(int cpu_no);
 cpumask_t *tpmi_get_power_domain_mask(int cpu_no);
+int tpmi_get_linux_die_id(int pkg_id, int domain_id);
 
 #endif
-- 
2.48.1

[PATCH v2 4/5] platform/x86/intel-uncore-freq: Add attributes to show die_id

[Srinivas Pandruvada]

For domains with agents to control cores (compute dies) show matching
Linux CPU die ID. Linux CPU ID is a logical die ID, so this may not match
physical die ID or domain_id. So, a mapping is required to get Linux CPU
die ID. This attribute is only presented when CPUID enumerates die ids.

This attribute can be used by orchestration software like Kubernetes to
target specific dies for uncore frequency control.

Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
---
v2:
No change

 .../uncore-frequency-common.c                 |  7 +++++
 .../uncore-frequency-common.h                 |  4 ++-
 .../uncore-frequency/uncore-frequency-tpmi.c  | 28 +++++++++++++++++++
 3 files changed, 38 insertions(+), 1 deletion(-)

diff --git a/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-common.c b/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-common.c
index cfa3039a0e39..3411d33e5f13 100644
--- a/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-common.c
+++ b/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-common.c
@@ -142,6 +142,8 @@ show_uncore_attr(elc_high_threshold_enable,
 		 UNCORE_INDEX_EFF_LAT_CTRL_HIGH_THRESHOLD_ENABLE);
 show_uncore_attr(elc_floor_freq_khz, UNCORE_INDEX_EFF_LAT_CTRL_FREQ);
 
+show_uncore_attr(die_id, UNCORE_INDEX_DIE_ID);
+
 #define show_uncore_data(member_name)					\
 	static ssize_t show_##member_name(struct kobject *kobj,	\
 					   struct kobj_attribute *attr, char *buf)\
@@ -203,6 +205,11 @@ static int create_attr_group(struct uncore_data *data, char *name)
 		data->uncore_attrs[index++] = &data->package_id_kobj_attr.attr;
 		init_attribute_ro(agent_types);
 		data->uncore_attrs[index++] = &data->agent_types_kobj_attr.attr;
+		if (topology_max_dies_per_package() > 1 &&
+		    data->agent_type_mask & AGENT_TYPE_CORE) {
+			init_attribute_ro(die_id);
+			data->uncore_attrs[index++] = &data->die_id_kobj_attr.attr;
+		}
 	}
 
 	data->uncore_attrs[index++] = &data->max_freq_khz_kobj_attr.attr;
diff --git a/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-common.h b/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-common.h
index 197ca2ad327f..9e4d7f44a41f 100644
--- a/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-common.h
+++ b/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-common.h
@@ -87,7 +87,8 @@ struct uncore_data {
 	struct kobj_attribute elc_high_threshold_enable_kobj_attr;
 	struct kobj_attribute elc_floor_freq_khz_kobj_attr;
 	struct kobj_attribute agent_types_kobj_attr;
-	struct attribute *uncore_attrs[14];
+	struct kobj_attribute die_id_kobj_attr;
+	struct attribute *uncore_attrs[15];
 };
 
 #define UNCORE_DOMAIN_ID_INVALID	-1
@@ -100,6 +101,7 @@ enum uncore_index {
 	UNCORE_INDEX_EFF_LAT_CTRL_HIGH_THRESHOLD,
 	UNCORE_INDEX_EFF_LAT_CTRL_HIGH_THRESHOLD_ENABLE,
 	UNCORE_INDEX_EFF_LAT_CTRL_FREQ,
+	UNCORE_INDEX_DIE_ID,
 };
 
 int uncore_freq_common_init(int (*read)(struct uncore_data *data, unsigned int *value,
diff --git a/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-tpmi.c b/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-tpmi.c
index 70415d80773c..adcf992e4ff0 100644
--- a/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-tpmi.c
+++ b/drivers/platform/x86/intel/uncore-frequency/uncore-frequency-tpmi.c
@@ -26,6 +26,7 @@
 #include <linux/module.h>
 #include <linux/intel_tpmi.h>
 
+#include "../tpmi_power_domains.h"
 #include "uncore-frequency-common.h"
 
 #define	UNCORE_MAJOR_VERSION		0
@@ -49,6 +50,7 @@ struct tpmi_uncore_cluster_info {
 	bool root_domain;
 	bool elc_supported;
 	u8 __iomem *cluster_base;
+	u16 cdie_id;
 	struct uncore_data uncore_data;
 	struct tpmi_uncore_struct *uncore_root;
 };
@@ -376,6 +378,9 @@ static void uncore_set_agent_type(struct tpmi_uncore_cluster_info *cluster_info)
 /* Callback for sysfs read for TPMI uncore values. Called under mutex locks. */
 static int uncore_read(struct uncore_data *data, unsigned int *value, enum uncore_index index)
 {
+	struct tpmi_uncore_cluster_info *cluster_info;
+	int ret;
+
 	switch (index) {
 	case UNCORE_INDEX_MIN_FREQ:
 	case UNCORE_INDEX_MAX_FREQ:
@@ -390,6 +395,16 @@ static int uncore_read(struct uncore_data *data, unsigned int *value, enum uncor
 	case UNCORE_INDEX_EFF_LAT_CTRL_FREQ:
 		return read_eff_lat_ctrl(data, value, index);
 
+	case UNCORE_INDEX_DIE_ID:
+		cluster_info = container_of(data, struct tpmi_uncore_cluster_info, uncore_data);
+		ret = tpmi_get_linux_die_id(cluster_info->uncore_data.package_id,
+					    cluster_info->cdie_id);
+		if (ret < 0)
+			return ret;
+
+		*value = ret;
+		return 0;
+
 	default:
 		break;
 	}
@@ -439,6 +454,16 @@ static void remove_cluster_entries(struct tpmi_uncore_struct *tpmi_uncore)
 	}
 }
 
+static void set_cdie_id(int domain_id, struct tpmi_uncore_cluster_info *cluster_info,
+		       struct intel_tpmi_plat_info *plat_info)
+{
+
+	cluster_info->cdie_id = domain_id;
+
+	if (plat_info->cdie_mask && cluster_info->uncore_data.agent_type_mask & AGENT_TYPE_CORE)
+		cluster_info->cdie_id = domain_id + ffs(plat_info->cdie_mask) - 1;
+}
+
 #define UNCORE_VERSION_MASK			GENMASK_ULL(7, 0)
 #define UNCORE_LOCAL_FABRIC_CLUSTER_ID_MASK	GENMASK_ULL(15, 8)
 #define UNCORE_CLUSTER_OFF_MASK			GENMASK_ULL(7, 0)
@@ -586,6 +611,8 @@ static int uncore_probe(struct auxiliary_device *auxdev, const struct auxiliary_
 			cluster_info->uncore_data.domain_id = i;
 			cluster_info->uncore_data.cluster_id = j;
 
+			set_cdie_id(i, cluster_info, plat_info);
+
 			cluster_info->uncore_root = tpmi_uncore;
 
 			if (TPMI_MINOR_VERSION(pd_info->ufs_header_ver) >= UNCORE_ELC_SUPPORTED_VERSION)
@@ -659,5 +686,6 @@ module_auxiliary_driver(intel_uncore_aux_driver);
 
 MODULE_IMPORT_NS("INTEL_TPMI");
 MODULE_IMPORT_NS("INTEL_UNCORE_FREQUENCY");
+MODULE_IMPORT_NS("INTEL_TPMI_POWER_DOMAIN");
 MODULE_DESCRIPTION("Intel TPMI UFS Driver");
 MODULE_LICENSE("GPL");
-- 
2.48.1

[PATCH v2 5/5] Documentation: admin-guide: pm: Add documentation for die_id

[Srinivas Pandruvada]

Add documentation to describe die_id attribute.

Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
---
v2:
Change "attributes" to "attribute"

 .../admin-guide/pm/intel_uncore_frequency_scaling.rst        | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/Documentation/admin-guide/pm/intel_uncore_frequency_scaling.rst b/Documentation/admin-guide/pm/intel_uncore_frequency_scaling.rst
index 84608dad84bd..d7ffda6a8095 100644
--- a/Documentation/admin-guide/pm/intel_uncore_frequency_scaling.rst
+++ b/Documentation/admin-guide/pm/intel_uncore_frequency_scaling.rst
@@ -91,6 +91,11 @@ Attributes in each directory:
 ``domain_id``
 	This attribute is used to get the power domain id of this instance.
 
+``die_id``
+	This attribute is used to get the Linux die id of this instance.
+	This attribute is only present for domains with core agents and
+        when the CPUID leaf 0x1f presents die ID.
+
 ``fabric_cluster_id``
 	This attribute is used to get the fabric cluster id of this instance.
 
-- 
2.48.1

Re: [PATCH v2 5/5] Documentation: admin-guide: pm: Add documentation for die_id

[Ilpo Järvinen]

[-- Attachment #1: Type: text/plain, Size: 1534 bytes --]

On Mon, 28 Apr 2025, Srinivas Pandruvada wrote:

> Add documentation to describe die_id attribute.
> 
> Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
> ---
> v2:
> Change "attributes" to "attribute"
> 
>  .../admin-guide/pm/intel_uncore_frequency_scaling.rst        | 5 +++++
>  1 file changed, 5 insertions(+)
> 
> diff --git a/Documentation/admin-guide/pm/intel_uncore_frequency_scaling.rst b/Documentation/admin-guide/pm/intel_uncore_frequency_scaling.rst
> index 84608dad84bd..d7ffda6a8095 100644
> --- a/Documentation/admin-guide/pm/intel_uncore_frequency_scaling.rst
> +++ b/Documentation/admin-guide/pm/intel_uncore_frequency_scaling.rst
> @@ -91,6 +91,11 @@ Attributes in each directory:
>  ``domain_id``
>  	This attribute is used to get the power domain id of this instance.
>  
> +``die_id``
> +	This attribute is used to get the Linux die id of this instance.
> +	This attribute is only present for domains with core agents and
> +        when the CPUID leaf 0x1f presents die ID.
> +
>  ``fabric_cluster_id``
>  	This attribute is used to get the fabric cluster id of this instance.
>  
> 

For patches #2-#5,

Reviewed-by: Ilpo Järvinen <ilpo.jarvinen@linux.intel.com>

It would have been nice to use cleanup.h in #3 but there's no good 
alternative for no_free_ptr() that doesn't have __must_check (IIRC, 
somebody proposed a solution to this relatively recently but I don't see 
that in linux/cleanup.h currently so lets forget that for now).

-- 
 i.