From: Srinivas Pandruvada <[email protected]>
Each TPMI power domain includes a group of CPUs. Several power
management settings in this case applicable to a group of CPUs.
There can be several power domains in a CPU package. So, provide
interfaces for:
- Get power domain id for a Linux CPU
- Get mask of Linux CPUs in a power domain
Hardware Punit uses different CPU numbering, which is not based on
APIC (Advanced Programmable Interrupt Controller) CPU numbering.
The Linux CPU numbering is based on APIC CPU numbering. Some PM features
like Intel Speed Select, the CPU core mask provided by the hardware is
based on the Punit CPU numbering. To use the core mask, this mask
needs to be converted to a Linux CPUs mask. So, provide interfaces for:
- Convert to a Linux CPU number from a Punit CPU number
- Convert to a Punit CPU number from a Linux CPU number
On each CPU online, MSR 0x54 is used to read the mapping and stores in
a per cpu array. Create a hash for faster searching of a Linux CPU number
from a Punit CPU number.
Signed-off-by: Srinivas Pandruvada <[email protected]>
[tero.kristo: minor updates]
Reviewed-by: Andy Shevchenko <[email protected]>
Reviewed-by: Ilpo Järvinen <[email protected]>
Signed-off-by: Tero Kristo <[email protected]>
---
drivers/platform/x86/intel/Kconfig | 4 +
drivers/platform/x86/intel/Makefile | 3 +
.../platform/x86/intel/tpmi_power_domains.c | 236 ++++++++++++++++++
.../platform/x86/intel/tpmi_power_domains.h | 19 ++
4 files changed, 262 insertions(+)
create mode 100644 drivers/platform/x86/intel/tpmi_power_domains.c
create mode 100644 drivers/platform/x86/intel/tpmi_power_domains.h
diff --git a/drivers/platform/x86/intel/Kconfig b/drivers/platform/x86/intel/Kconfig
index e9dc0c021029..e97a97355d5a 100644
--- a/drivers/platform/x86/intel/Kconfig
+++ b/drivers/platform/x86/intel/Kconfig
@@ -192,10 +192,14 @@ config INTEL_SMARTCONNECT
This driver checks to determine whether the device has Intel Smart
Connect enabled, and if so disables it.
+config INTEL_TPMI_POWER_DOMAINS
+ tristate
+
config INTEL_TPMI
tristate "Intel Topology Aware Register and PM Capsule Interface (TPMI)"
depends on INTEL_VSEC
depends on X86_64
+ select INTEL_TPMI_POWER_DOMAINS
help
The Intel Topology Aware Register and PM Capsule Interface (TPMI),
provides enumerable MMIO interface for power management features.
diff --git a/drivers/platform/x86/intel/Makefile b/drivers/platform/x86/intel/Makefile
index c1d5fe05e3f3..10437e56027d 100644
--- a/drivers/platform/x86/intel/Makefile
+++ b/drivers/platform/x86/intel/Makefile
@@ -53,6 +53,9 @@ obj-$(CONFIG_INTEL_PUNIT_IPC) += intel_punit_ipc.o
intel_vsec_tpmi-y := tpmi.o
obj-$(CONFIG_INTEL_TPMI) += intel_vsec_tpmi.o
+intel_tpmi_power_domains-y := tpmi_power_domains.o
+obj-$(CONFIG_INTEL_TPMI_POWER_DOMAINS) += intel_tpmi_power_domains.o
+
# Intel Uncore drivers
intel-rst-y := rst.o
obj-$(CONFIG_INTEL_RST) += intel-rst.o
diff --git a/drivers/platform/x86/intel/tpmi_power_domains.c b/drivers/platform/x86/intel/tpmi_power_domains.c
new file mode 100644
index 000000000000..40f994814248
--- /dev/null
+++ b/drivers/platform/x86/intel/tpmi_power_domains.c
@@ -0,0 +1,236 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Mapping of TPMI power domains CPU mapping
+ *
+ * Copyright (c) 2024, Intel Corporation.
+ * All Rights Reserved.
+ */
+
+#include <linux/bitfield.h>
+#include <linux/cleanup.h>
+#include <linux/cpuhotplug.h>
+#include <linux/cpumask.h>
+#include <linux/errno.h>
+#include <linux/export.h>
+#include <linux/hashtable.h>
+#include <linux/module.h>
+#include <linux/mutex.h>
+#include <linux/overflow.h>
+#include <linux/slab.h>
+#include <linux/topology.h>
+#include <linux/types.h>
+
+#include <asm/cpu_device_id.h>
+#include <asm/intel-family.h>
+#include <asm/msr.h>
+
+#include "tpmi_power_domains.h"
+
+#define MSR_PM_LOGICAL_ID 0x54
+
+/*
+ * Struct of MSR 0x54
+ * [15:11] PM_DOMAIN_ID
+ * [10:3] MODULE_ID (aka IDI_AGENT_ID)
+ * [2:0] LP_ID
+ * For Atom:
+ * [2] Always 0
+ * [1:0] core ID within module
+ * For Core
+ * [2:1] Always 0
+ * [0] thread ID
+ */
+
+#define LP_ID_MASK GENMASK_ULL(2, 0)
+#define MODULE_ID_MASK GENMASK_ULL(10, 3)
+#define PM_DOMAIN_ID_MASK GENMASK_ULL(15, 11)
+
+/**
+ * struct tpmi_cpu_info - Mapping information for a CPU
+ * @hnode: Used to add mapping information to hash list
+ * @linux_cpu: Linux CPU number
+ * @pkg_id: Package ID of this CPU
+ * @punit_thread_id: Punit thread id of this CPU
+ * @punit_core_id: Punit core id
+ * @punit_domain_id: Power domain id from Punit
+ *
+ * Structure to store mapping information for a Linux CPU
+ * to a Punit core, thread and power domain.
+ */
+struct tpmi_cpu_info {
+ struct hlist_node hnode;
+ int linux_cpu;
+ u8 pkg_id;
+ u8 punit_thread_id;
+ u8 punit_core_id;
+ u8 punit_domain_id;
+};
+
+static DEFINE_PER_CPU(struct tpmi_cpu_info, tpmi_cpu_info);
+
+/* The dynamically assigned cpu hotplug state to free later */
+static enum cpuhp_state tpmi_hp_state __read_mostly;
+
+#define MAX_POWER_DOMAINS 8
+
+static cpumask_t *tpmi_power_domain_mask;
+
+/* Lock to protect tpmi_power_domain_mask and tpmi_cpu_hash */
+static DEFINE_MUTEX(tpmi_lock);
+
+static const struct x86_cpu_id tpmi_cpu_ids[] = {
+ X86_MATCH_INTEL_FAM6_MODEL(GRANITERAPIDS_X, NULL),
+ X86_MATCH_INTEL_FAM6_MODEL(ATOM_CRESTMONT_X, NULL),
+ X86_MATCH_INTEL_FAM6_MODEL(ATOM_CRESTMONT, NULL),
+ X86_MATCH_INTEL_FAM6_MODEL(GRANITERAPIDS_D, NULL),
+ {}
+};
+MODULE_DEVICE_TABLE(x86cpu, tpmi_cpu_ids);
+
+static DECLARE_HASHTABLE(tpmi_cpu_hash, 8);
+
+static bool tpmi_domain_is_valid(struct tpmi_cpu_info *info)
+{
+ return info->pkg_id < topology_max_packages() &&
+ info->punit_domain_id < MAX_POWER_DOMAINS;
+}
+
+int tpmi_get_linux_cpu_number(int package_id, int domain_id, int punit_core_id)
+{
+ struct tpmi_cpu_info *info;
+ int ret = -EINVAL;
+
+ guard(mutex)(&tpmi_lock);
+ hash_for_each_possible(tpmi_cpu_hash, info, hnode, punit_core_id) {
+ if (info->punit_domain_id == domain_id && info->pkg_id == package_id) {
+ ret = info->linux_cpu;
+ break;
+ }
+ }
+
+ return ret;
+}
+EXPORT_SYMBOL_NS_GPL(tpmi_get_linux_cpu_number, INTEL_TPMI_POWER_DOMAIN);
+
+int tpmi_get_punit_core_number(int cpu_no)
+{
+ if (cpu_no >= num_possible_cpus())
+ return -EINVAL;
+
+ return per_cpu(tpmi_cpu_info, cpu_no).punit_core_id;
+}
+EXPORT_SYMBOL_NS_GPL(tpmi_get_punit_core_number, INTEL_TPMI_POWER_DOMAIN);
+
+int tpmi_get_power_domain_id(int cpu_no)
+{
+ if (cpu_no >= num_possible_cpus())
+ return -EINVAL;
+
+ return per_cpu(tpmi_cpu_info, cpu_no).punit_domain_id;
+}
+EXPORT_SYMBOL_NS_GPL(tpmi_get_power_domain_id, INTEL_TPMI_POWER_DOMAIN);
+
+cpumask_t *tpmi_get_power_domain_mask(int cpu_no)
+{
+ struct tpmi_cpu_info *info;
+ cpumask_t *mask;
+ int index;
+
+ if (cpu_no >= num_possible_cpus())
+ return NULL;
+
+ info = &per_cpu(tpmi_cpu_info, cpu_no);
+ if (!tpmi_domain_is_valid(info))
+ return NULL;
+
+ index = info->pkg_id * MAX_POWER_DOMAINS + info->punit_domain_id;
+ guard(mutex)(&tpmi_lock);
+ mask = &tpmi_power_domain_mask[index];
+
+ return mask;
+}
+EXPORT_SYMBOL_NS_GPL(tpmi_get_power_domain_mask, INTEL_TPMI_POWER_DOMAIN);
+
+static int tpmi_get_logical_id(unsigned int cpu, struct tpmi_cpu_info *info)
+{
+ u64 data;
+ int ret;
+
+ ret = rdmsrl_safe(MSR_PM_LOGICAL_ID, &data);
+ if (ret)
+ return ret;
+
+ info->punit_domain_id = FIELD_GET(PM_DOMAIN_ID_MASK, data);
+ if (info->punit_domain_id >= MAX_POWER_DOMAINS)
+ return -EINVAL;
+
+ info->punit_thread_id = FIELD_GET(LP_ID_MASK, data);
+ info->punit_core_id = FIELD_GET(MODULE_ID_MASK, data);
+ info->pkg_id = topology_physical_package_id(cpu);
+ info->linux_cpu = cpu;
+
+ return 0;
+}
+
+static int tpmi_cpu_online(unsigned int cpu)
+{
+ struct tpmi_cpu_info *info = &per_cpu(tpmi_cpu_info, cpu);
+ int ret, index;
+
+ /* Don't fail CPU online for some bad mapping of CPUs */
+ ret = tpmi_get_logical_id(cpu, info);
+ if (ret)
+ return 0;
+
+ index = info->pkg_id * MAX_POWER_DOMAINS + info->punit_domain_id;
+
+ guard(mutex)(&tpmi_lock);
+ cpumask_set_cpu(cpu, &tpmi_power_domain_mask[index]);
+ hash_add(tpmi_cpu_hash, &info->hnode, info->punit_core_id);
+
+ return 0;
+}
+
+static int __init tpmi_init(void)
+{
+ const struct x86_cpu_id *id;
+ u64 data;
+ int ret;
+
+ id = x86_match_cpu(tpmi_cpu_ids);
+ if (!id)
+ return -ENODEV;
+
+ /* Check for MSR 0x54 presence */
+ ret = rdmsrl_safe(MSR_PM_LOGICAL_ID, &data);
+ if (ret)
+ return ret;
+
+ tpmi_power_domain_mask = kcalloc(size_mul(topology_max_packages(), MAX_POWER_DOMAINS),
+ sizeof(*tpmi_power_domain_mask), GFP_KERNEL);
+ if (!tpmi_power_domain_mask)
+ return -ENOMEM;
+
+ 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;
+ }
+
+ tpmi_hp_state = ret;
+
+ return 0;
+}
+module_init(tpmi_init)
+
+static void __exit tpmi_exit(void)
+{
+ cpuhp_remove_state(tpmi_hp_state);
+ kfree(tpmi_power_domain_mask);
+}
+module_exit(tpmi_exit)
+
+MODULE_DESCRIPTION("TPMI Power Domains Mapping");
+MODULE_LICENSE("GPL");
diff --git a/drivers/platform/x86/intel/tpmi_power_domains.h b/drivers/platform/x86/intel/tpmi_power_domains.h
new file mode 100644
index 000000000000..0c2154bd941f
--- /dev/null
+++ b/drivers/platform/x86/intel/tpmi_power_domains.h
@@ -0,0 +1,19 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Mapping of TPMI power domain and CPUs
+ *
+ * Copyright (c) 2024, Intel Corporation.
+ * All rights reserved.
+ */
+
+#ifndef _TPMI_POWER_DOMAINS_H_
+#define _TPMI_POWER_DOMAINS_H_
+
+#include <linux/cpumask.h>
+
+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);
+
+#endif
--
2.43.1
On Mon, 2024-05-27 at 16:29 +0300, Tero Kristo wrote:
> From: Srinivas Pandruvada <[email protected]>
>
> Each TPMI power domain includes a group of CPUs. Several power
> management settings in this case applicable to a group of CPUs.
> There can be several power domains in a CPU package. So, provide
> interfaces for:
> - Get power domain id for a Linux CPU
> - Get mask of Linux CPUs in a power domain
>
> Hardware Punit uses different CPU numbering, which is not based on
> APIC (Advanced Programmable Interrupt Controller) CPU numbering.
> The Linux CPU numbering is based on APIC CPU numbering. Some PM
> features
> like Intel Speed Select, the CPU core mask provided by the hardware
> is
> based on the Punit CPU numbering. To use the core mask, this mask
> needs to be converted to a Linux CPUs mask. So, provide interfaces
> for:
> - Convert to a Linux CPU number from a Punit CPU number
> - Convert to a Punit CPU number from a Linux CPU number
>
> On each CPU online, MSR 0x54 is used to read the mapping and stores
> in
> a per cpu array. Create a hash for faster searching of a Linux CPU
> number
> from a Punit CPU number.
>
> Signed-off-by: Srinivas Pandruvada
> <[email protected]>
> [tero.kristo: minor updates]
> Reviewed-by: Andy Shevchenko <[email protected]>
> Reviewed-by: Ilpo Järvinen <[email protected]>
> Signed-off-by: Tero Kristo <[email protected]>
> ---
> drivers/platform/x86/intel/Kconfig | 4 +
> drivers/platform/x86/intel/Makefile | 3 +
> .../platform/x86/intel/tpmi_power_domains.c | 236
> ++++++++++++++++++
> .../platform/x86/intel/tpmi_power_domains.h | 19 ++
> 4 files changed, 262 insertions(+)
> create mode 100644 drivers/platform/x86/intel/tpmi_power_domains.c
> create mode 100644 drivers/platform/x86/intel/tpmi_power_domains.h
>
> diff --git a/drivers/platform/x86/intel/Kconfig
> b/drivers/platform/x86/intel/Kconfig
> index e9dc0c021029..e97a97355d5a 100644
> --- a/drivers/platform/x86/intel/Kconfig
> +++ b/drivers/platform/x86/intel/Kconfig
> @@ -192,10 +192,14 @@ config INTEL_SMARTCONNECT
> This driver checks to determine whether the device has
> Intel Smart
> Connect enabled, and if so disables it.
>
> +config INTEL_TPMI_POWER_DOMAINS
> + tristate
> +
> config INTEL_TPMI
> tristate "Intel Topology Aware Register and PM Capsule
> Interface (TPMI)"
> depends on INTEL_VSEC
> depends on X86_64
> + select INTEL_TPMI_POWER_DOMAINS
> help
> The Intel Topology Aware Register and PM Capsule Interface
> (TPMI),
> provides enumerable MMIO interface for power management
> features.
> diff --git a/drivers/platform/x86/intel/Makefile
> b/drivers/platform/x86/intel/Makefile
> index c1d5fe05e3f3..10437e56027d 100644
> --- a/drivers/platform/x86/intel/Makefile
> +++ b/drivers/platform/x86/intel/Makefile
> @@ -53,6 +53,9 @@ obj-$(CONFIG_INTEL_PUNIT_IPC) +=
> intel_punit_ipc.o
> intel_vsec_tpmi-y := tpmi.o
> obj-$(CONFIG_INTEL_TPMI) += intel_vsec_tpmi.o
>
> +intel_tpmi_power_domains-y := tpmi_power_domains.o
> +obj-$(CONFIG_INTEL_TPMI_POWER_DOMAINS) += intel_tpmi_power_domains.o
> +
> # Intel Uncore drivers
> intel-rst-y := rst.o
> obj-$(CONFIG_INTEL_RST) += intel-rst.o
> diff --git a/drivers/platform/x86/intel/tpmi_power_domains.c
> b/drivers/platform/x86/intel/tpmi_power_domains.c
> new file mode 100644
> index 000000000000..40f994814248
> --- /dev/null
> +++ b/drivers/platform/x86/intel/tpmi_power_domains.c
> @@ -0,0 +1,236 @@
> +// SPDX-License-Identifier: GPL-2.0-only
> +/*
> + * Mapping of TPMI power domains CPU mapping
> + *
> + * Copyright (c) 2024, Intel Corporation.
> + * All Rights Reserved.
> + */
> +
> +#include <linux/bitfield.h>
> +#include <linux/cleanup.h>
> +#include <linux/cpuhotplug.h>
> +#include <linux/cpumask.h>
> +#include <linux/errno.h>
> +#include <linux/export.h>
> +#include <linux/hashtable.h>
> +#include <linux/module.h>
> +#include <linux/mutex.h>
> +#include <linux/overflow.h>
> +#include <linux/slab.h>
> +#include <linux/topology.h>
> +#include <linux/types.h>
> +
> +#include <asm/cpu_device_id.h>
> +#include <asm/intel-family.h>
> +#include <asm/msr.h>
> +
> +#include "tpmi_power_domains.h"
> +
> +#define MSR_PM_LOGICAL_ID 0x54
> +
> +/*
> + * Struct of MSR 0x54
> + * [15:11] PM_DOMAIN_ID
> + * [10:3] MODULE_ID (aka IDI_AGENT_ID)
> + * [2:0] LP_ID
> + * For Atom:
> + * [2] Always 0
> + * [1:0] core ID within module
> + * For Core
> + * [2:1] Always 0
> + * [0] thread ID
> + */
> +
> +#define LP_ID_MASK GENMASK_ULL(2, 0)
> +#define MODULE_ID_MASK GENMASK_ULL(10, 3)
> +#define PM_DOMAIN_ID_MASK GENMASK_ULL(15, 11)
> +
> +/**
> + * struct tpmi_cpu_info - Mapping information for a CPU
> + * @hnode: Used to add mapping information to hash list
> + * @linux_cpu: Linux CPU number
> + * @pkg_id: Package ID of this CPU
> + * @punit_thread_id: Punit thread id of this CPU
> + * @punit_core_id: Punit core id
> + * @punit_domain_id: Power domain id from Punit
> + *
> + * Structure to store mapping information for a Linux CPU
> + * to a Punit core, thread and power domain.
> + */
> +struct tpmi_cpu_info {
> + struct hlist_node hnode;
> + int linux_cpu;
> + u8 pkg_id;
> + u8 punit_thread_id;
> + u8 punit_core_id;
> + u8 punit_domain_id;
> +};
> +
> +static DEFINE_PER_CPU(struct tpmi_cpu_info, tpmi_cpu_info);
> +
> +/* The dynamically assigned cpu hotplug state to free later */
> +static enum cpuhp_state tpmi_hp_state __read_mostly;
> +
> +#define MAX_POWER_DOMAINS 8
> +
> +static cpumask_t *tpmi_power_domain_mask;
> +
> +/* Lock to protect tpmi_power_domain_mask and tpmi_cpu_hash */
> +static DEFINE_MUTEX(tpmi_lock);
> +
> +static const struct x86_cpu_id tpmi_cpu_ids[] = {
> + X86_MATCH_INTEL_FAM6_MODEL(GRANITERAPIDS_X, NULL),
> + X86_MATCH_INTEL_FAM6_MODEL(ATOM_CRESTMONT_X, NULL),
> + X86_MATCH_INTEL_FAM6_MODEL(ATOM_CRESTMONT, NULL),
> + X86_MATCH_INTEL_FAM6_MODEL(GRANITERAPIDS_D, NULL),
> + {}
> +};
From 6.10 kernel we are supposed to use
X86_MATCH_VFM() macro.
It is better to either change here or apply a patch on top.
Thanks,
Srinivas
> +MODULE_DEVICE_TABLE(x86cpu, tpmi_cpu_ids);
> +
> +static DECLARE_HASHTABLE(tpmi_cpu_hash, 8);
> +
> +static bool tpmi_domain_is_valid(struct tpmi_cpu_info *info)
> +{
> + return info->pkg_id < topology_max_packages() &&
> + info->punit_domain_id < MAX_POWER_DOMAINS;
> +}
> +
> +int tpmi_get_linux_cpu_number(int package_id, int domain_id, int
> punit_core_id)
> +{
> + struct tpmi_cpu_info *info;
> + int ret = -EINVAL;
> +
> + guard(mutex)(&tpmi_lock);
> + hash_for_each_possible(tpmi_cpu_hash, info, hnode,
> punit_core_id) {
> + if (info->punit_domain_id == domain_id && info-
> >pkg_id == package_id) {
> + ret = info->linux_cpu;
> + break;
> + }
> + }
> +
> + return ret;
> +}
> +EXPORT_SYMBOL_NS_GPL(tpmi_get_linux_cpu_number,
> INTEL_TPMI_POWER_DOMAIN);
> +
> +int tpmi_get_punit_core_number(int cpu_no)
> +{
> + if (cpu_no >= num_possible_cpus())
> + return -EINVAL;
> +
> + return per_cpu(tpmi_cpu_info, cpu_no).punit_core_id;
> +}
> +EXPORT_SYMBOL_NS_GPL(tpmi_get_punit_core_number,
> INTEL_TPMI_POWER_DOMAIN);
> +
> +int tpmi_get_power_domain_id(int cpu_no)
> +{
> + if (cpu_no >= num_possible_cpus())
> + return -EINVAL;
> +
> + return per_cpu(tpmi_cpu_info, cpu_no).punit_domain_id;
> +}
> +EXPORT_SYMBOL_NS_GPL(tpmi_get_power_domain_id,
> INTEL_TPMI_POWER_DOMAIN);
> +
> +cpumask_t *tpmi_get_power_domain_mask(int cpu_no)
> +{
> + struct tpmi_cpu_info *info;
> + cpumask_t *mask;
> + int index;
> +
> + if (cpu_no >= num_possible_cpus())
> + return NULL;
> +
> + info = &per_cpu(tpmi_cpu_info, cpu_no);
> + if (!tpmi_domain_is_valid(info))
> + return NULL;
> +
> + index = info->pkg_id * MAX_POWER_DOMAINS + info-
> >punit_domain_id;
> + guard(mutex)(&tpmi_lock);
> + mask = &tpmi_power_domain_mask[index];
> +
> + return mask;
> +}
> +EXPORT_SYMBOL_NS_GPL(tpmi_get_power_domain_mask,
> INTEL_TPMI_POWER_DOMAIN);
> +
> +static int tpmi_get_logical_id(unsigned int cpu, struct
> tpmi_cpu_info *info)
> +{
> + u64 data;
> + int ret;
> +
> + ret = rdmsrl_safe(MSR_PM_LOGICAL_ID, &data);
> + if (ret)
> + return ret;
> +
> + info->punit_domain_id = FIELD_GET(PM_DOMAIN_ID_MASK, data);
> + if (info->punit_domain_id >= MAX_POWER_DOMAINS)
> + return -EINVAL;
> +
> + info->punit_thread_id = FIELD_GET(LP_ID_MASK, data);
> + info->punit_core_id = FIELD_GET(MODULE_ID_MASK, data);
> + info->pkg_id = topology_physical_package_id(cpu);
> + info->linux_cpu = cpu;
> +
> + return 0;
> +}
> +
> +static int tpmi_cpu_online(unsigned int cpu)
> +{
> + struct tpmi_cpu_info *info = &per_cpu(tpmi_cpu_info, cpu);
> + int ret, index;
> +
> + /* Don't fail CPU online for some bad mapping of CPUs */
> + ret = tpmi_get_logical_id(cpu, info);
> + if (ret)
> + return 0;
> +
> + index = info->pkg_id * MAX_POWER_DOMAINS + info-
> >punit_domain_id;
> +
> + guard(mutex)(&tpmi_lock);
> + cpumask_set_cpu(cpu, &tpmi_power_domain_mask[index]);
> + hash_add(tpmi_cpu_hash, &info->hnode, info->punit_core_id);
> +
> + return 0;
> +}
> +
> +static int __init tpmi_init(void)
> +{
> + const struct x86_cpu_id *id;
> + u64 data;
> + int ret;
> +
> + id = x86_match_cpu(tpmi_cpu_ids);
> + if (!id)
> + return -ENODEV;
> +
> + /* Check for MSR 0x54 presence */
> + ret = rdmsrl_safe(MSR_PM_LOGICAL_ID, &data);
> + if (ret)
> + return ret;
> +
> + tpmi_power_domain_mask =
> kcalloc(size_mul(topology_max_packages(), MAX_POWER_DOMAINS),
> +
> sizeof(*tpmi_power_domain_mask), GFP_KERNEL);
> + if (!tpmi_power_domain_mask)
> + return -ENOMEM;
> +
> + ret = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN,
> + "platform/x86/tpmi_power_domains:onli
> ne",
> + tpmi_cpu_online, NULL);
> + if (ret < 0) {
> + kfree(tpmi_power_domain_mask);
> + return ret;
> + }
> +
> + tpmi_hp_state = ret;
> +
> + return 0;
> +}
> +module_init(tpmi_init)
> +
> +static void __exit tpmi_exit(void)
> +{
> + cpuhp_remove_state(tpmi_hp_state);
> + kfree(tpmi_power_domain_mask);
> +}
> +module_exit(tpmi_exit)
> +
> +MODULE_DESCRIPTION("TPMI Power Domains Mapping");
> +MODULE_LICENSE("GPL");
> diff --git a/drivers/platform/x86/intel/tpmi_power_domains.h
> b/drivers/platform/x86/intel/tpmi_power_domains.h
> new file mode 100644
> index 000000000000..0c2154bd941f
> --- /dev/null
> +++ b/drivers/platform/x86/intel/tpmi_power_domains.h
> @@ -0,0 +1,19 @@
> +/* SPDX-License-Identifier: GPL-2.0-only */
> +/*
> + * Mapping of TPMI power domain and CPUs
> + *
> + * Copyright (c) 2024, Intel Corporation.
> + * All rights reserved.
> + */
> +
> +#ifndef _TPMI_POWER_DOMAINS_H_
> +#define _TPMI_POWER_DOMAINS_H_
> +
> +#include <linux/cpumask.h>
> +
> +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);
> +
> +#endif
From: Srinivas Pandruvada <[email protected]>
Each TPMI power domain includes a group of CPUs. Several power
management settings in this case applicable to a group of CPUs.
There can be several power domains in a CPU package. So, provide
interfaces for:
- Get power domain id for a Linux CPU
- Get mask of Linux CPUs in a power domain
Hardware Punit uses different CPU numbering, which is not based on
APIC (Advanced Programmable Interrupt Controller) CPU numbering.
The Linux CPU numbering is based on APIC CPU numbering. Some PM features
like Intel Speed Select, the CPU core mask provided by the hardware is
based on the Punit CPU numbering. To use the core mask, this mask
needs to be converted to a Linux CPUs mask. So, provide interfaces for:
- Convert to a Linux CPU number from a Punit CPU number
- Convert to a Punit CPU number from a Linux CPU number
On each CPU online, MSR 0x54 is used to read the mapping and stores in
a per cpu array. Create a hash for faster searching of a Linux CPU number
from a Punit CPU number.
Signed-off-by: Srinivas Pandruvada <[email protected]>
[tero.kristo: minor updates]
Reviewed-by: Andy Shevchenko <[email protected]>
Reviewed-by: Ilpo Järvinen <[email protected]>
Signed-off-by: Tero Kristo <[email protected]>
---
v2:
* changed to use X86_MATCH_VFM() instead of X86_MATCH_INTEL_FAM6_MODEL()
drivers/platform/x86/intel/Kconfig | 4 +
drivers/platform/x86/intel/Makefile | 3 +
.../platform/x86/intel/tpmi_power_domains.c | 236 ++++++++++++++++++
.../platform/x86/intel/tpmi_power_domains.h | 19 ++
4 files changed, 262 insertions(+)
create mode 100644 drivers/platform/x86/intel/tpmi_power_domains.c
create mode 100644 drivers/platform/x86/intel/tpmi_power_domains.h
diff --git a/drivers/platform/x86/intel/Kconfig b/drivers/platform/x86/intel/Kconfig
index e9dc0c021029..e97a97355d5a 100644
--- a/drivers/platform/x86/intel/Kconfig
+++ b/drivers/platform/x86/intel/Kconfig
@@ -192,10 +192,14 @@ config INTEL_SMARTCONNECT
This driver checks to determine whether the device has Intel Smart
Connect enabled, and if so disables it.
+config INTEL_TPMI_POWER_DOMAINS
+ tristate
+
config INTEL_TPMI
tristate "Intel Topology Aware Register and PM Capsule Interface (TPMI)"
depends on INTEL_VSEC
depends on X86_64
+ select INTEL_TPMI_POWER_DOMAINS
help
The Intel Topology Aware Register and PM Capsule Interface (TPMI),
provides enumerable MMIO interface for power management features.
diff --git a/drivers/platform/x86/intel/Makefile b/drivers/platform/x86/intel/Makefile
index c1d5fe05e3f3..10437e56027d 100644
--- a/drivers/platform/x86/intel/Makefile
+++ b/drivers/platform/x86/intel/Makefile
@@ -53,6 +53,9 @@ obj-$(CONFIG_INTEL_PUNIT_IPC) += intel_punit_ipc.o
intel_vsec_tpmi-y := tpmi.o
obj-$(CONFIG_INTEL_TPMI) += intel_vsec_tpmi.o
+intel_tpmi_power_domains-y := tpmi_power_domains.o
+obj-$(CONFIG_INTEL_TPMI_POWER_DOMAINS) += intel_tpmi_power_domains.o
+
# Intel Uncore drivers
intel-rst-y := rst.o
obj-$(CONFIG_INTEL_RST) += intel-rst.o
diff --git a/drivers/platform/x86/intel/tpmi_power_domains.c b/drivers/platform/x86/intel/tpmi_power_domains.c
new file mode 100644
index 000000000000..40f994814248
--- /dev/null
+++ b/drivers/platform/x86/intel/tpmi_power_domains.c
@@ -0,0 +1,236 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * Mapping of TPMI power domains CPU mapping
+ *
+ * Copyright (c) 2024, Intel Corporation.
+ * All Rights Reserved.
+ */
+
+#include <linux/bitfield.h>
+#include <linux/cleanup.h>
+#include <linux/cpuhotplug.h>
+#include <linux/cpumask.h>
+#include <linux/errno.h>
+#include <linux/export.h>
+#include <linux/hashtable.h>
+#include <linux/module.h>
+#include <linux/mutex.h>
+#include <linux/overflow.h>
+#include <linux/slab.h>
+#include <linux/topology.h>
+#include <linux/types.h>
+
+#include <asm/cpu_device_id.h>
+#include <asm/intel-family.h>
+#include <asm/msr.h>
+
+#include "tpmi_power_domains.h"
+
+#define MSR_PM_LOGICAL_ID 0x54
+
+/*
+ * Struct of MSR 0x54
+ * [15:11] PM_DOMAIN_ID
+ * [10:3] MODULE_ID (aka IDI_AGENT_ID)
+ * [2:0] LP_ID
+ * For Atom:
+ * [2] Always 0
+ * [1:0] core ID within module
+ * For Core
+ * [2:1] Always 0
+ * [0] thread ID
+ */
+
+#define LP_ID_MASK GENMASK_ULL(2, 0)
+#define MODULE_ID_MASK GENMASK_ULL(10, 3)
+#define PM_DOMAIN_ID_MASK GENMASK_ULL(15, 11)
+
+/**
+ * struct tpmi_cpu_info - Mapping information for a CPU
+ * @hnode: Used to add mapping information to hash list
+ * @linux_cpu: Linux CPU number
+ * @pkg_id: Package ID of this CPU
+ * @punit_thread_id: Punit thread id of this CPU
+ * @punit_core_id: Punit core id
+ * @punit_domain_id: Power domain id from Punit
+ *
+ * Structure to store mapping information for a Linux CPU
+ * to a Punit core, thread and power domain.
+ */
+struct tpmi_cpu_info {
+ struct hlist_node hnode;
+ int linux_cpu;
+ u8 pkg_id;
+ u8 punit_thread_id;
+ u8 punit_core_id;
+ u8 punit_domain_id;
+};
+
+static DEFINE_PER_CPU(struct tpmi_cpu_info, tpmi_cpu_info);
+
+/* The dynamically assigned cpu hotplug state to free later */
+static enum cpuhp_state tpmi_hp_state __read_mostly;
+
+#define MAX_POWER_DOMAINS 8
+
+static cpumask_t *tpmi_power_domain_mask;
+
+/* Lock to protect tpmi_power_domain_mask and tpmi_cpu_hash */
+static DEFINE_MUTEX(tpmi_lock);
+
+static const struct x86_cpu_id tpmi_cpu_ids[] = {
+ X86_MATCH_VFM(INTEL_GRANITERAPIDS_X, NULL),
+ X86_MATCH_VFM(INTEL_ATOM_CRESTMONT_X, NULL),
+ X86_MATCH_VFM(INTEL_ATOM_CRESTMONT, NULL),
+ X86_MATCH_VFM(INTEL_GRANITERAPIDS_D, NULL),
+ {}
+};
+MODULE_DEVICE_TABLE(x86cpu, tpmi_cpu_ids);
+
+static DECLARE_HASHTABLE(tpmi_cpu_hash, 8);
+
+static bool tpmi_domain_is_valid(struct tpmi_cpu_info *info)
+{
+ return info->pkg_id < topology_max_packages() &&
+ info->punit_domain_id < MAX_POWER_DOMAINS;
+}
+
+int tpmi_get_linux_cpu_number(int package_id, int domain_id, int punit_core_id)
+{
+ struct tpmi_cpu_info *info;
+ int ret = -EINVAL;
+
+ guard(mutex)(&tpmi_lock);
+ hash_for_each_possible(tpmi_cpu_hash, info, hnode, punit_core_id) {
+ if (info->punit_domain_id == domain_id && info->pkg_id == package_id) {
+ ret = info->linux_cpu;
+ break;
+ }
+ }
+
+ return ret;
+}
+EXPORT_SYMBOL_NS_GPL(tpmi_get_linux_cpu_number, INTEL_TPMI_POWER_DOMAIN);
+
+int tpmi_get_punit_core_number(int cpu_no)
+{
+ if (cpu_no >= num_possible_cpus())
+ return -EINVAL;
+
+ return per_cpu(tpmi_cpu_info, cpu_no).punit_core_id;
+}
+EXPORT_SYMBOL_NS_GPL(tpmi_get_punit_core_number, INTEL_TPMI_POWER_DOMAIN);
+
+int tpmi_get_power_domain_id(int cpu_no)
+{
+ if (cpu_no >= num_possible_cpus())
+ return -EINVAL;
+
+ return per_cpu(tpmi_cpu_info, cpu_no).punit_domain_id;
+}
+EXPORT_SYMBOL_NS_GPL(tpmi_get_power_domain_id, INTEL_TPMI_POWER_DOMAIN);
+
+cpumask_t *tpmi_get_power_domain_mask(int cpu_no)
+{
+ struct tpmi_cpu_info *info;
+ cpumask_t *mask;
+ int index;
+
+ if (cpu_no >= num_possible_cpus())
+ return NULL;
+
+ info = &per_cpu(tpmi_cpu_info, cpu_no);
+ if (!tpmi_domain_is_valid(info))
+ return NULL;
+
+ index = info->pkg_id * MAX_POWER_DOMAINS + info->punit_domain_id;
+ guard(mutex)(&tpmi_lock);
+ mask = &tpmi_power_domain_mask[index];
+
+ return mask;
+}
+EXPORT_SYMBOL_NS_GPL(tpmi_get_power_domain_mask, INTEL_TPMI_POWER_DOMAIN);
+
+static int tpmi_get_logical_id(unsigned int cpu, struct tpmi_cpu_info *info)
+{
+ u64 data;
+ int ret;
+
+ ret = rdmsrl_safe(MSR_PM_LOGICAL_ID, &data);
+ if (ret)
+ return ret;
+
+ info->punit_domain_id = FIELD_GET(PM_DOMAIN_ID_MASK, data);
+ if (info->punit_domain_id >= MAX_POWER_DOMAINS)
+ return -EINVAL;
+
+ info->punit_thread_id = FIELD_GET(LP_ID_MASK, data);
+ info->punit_core_id = FIELD_GET(MODULE_ID_MASK, data);
+ info->pkg_id = topology_physical_package_id(cpu);
+ info->linux_cpu = cpu;
+
+ return 0;
+}
+
+static int tpmi_cpu_online(unsigned int cpu)
+{
+ struct tpmi_cpu_info *info = &per_cpu(tpmi_cpu_info, cpu);
+ int ret, index;
+
+ /* Don't fail CPU online for some bad mapping of CPUs */
+ ret = tpmi_get_logical_id(cpu, info);
+ if (ret)
+ return 0;
+
+ index = info->pkg_id * MAX_POWER_DOMAINS + info->punit_domain_id;
+
+ guard(mutex)(&tpmi_lock);
+ cpumask_set_cpu(cpu, &tpmi_power_domain_mask[index]);
+ hash_add(tpmi_cpu_hash, &info->hnode, info->punit_core_id);
+
+ return 0;
+}
+
+static int __init tpmi_init(void)
+{
+ const struct x86_cpu_id *id;
+ u64 data;
+ int ret;
+
+ id = x86_match_cpu(tpmi_cpu_ids);
+ if (!id)
+ return -ENODEV;
+
+ /* Check for MSR 0x54 presence */
+ ret = rdmsrl_safe(MSR_PM_LOGICAL_ID, &data);
+ if (ret)
+ return ret;
+
+ tpmi_power_domain_mask = kcalloc(size_mul(topology_max_packages(), MAX_POWER_DOMAINS),
+ sizeof(*tpmi_power_domain_mask), GFP_KERNEL);
+ if (!tpmi_power_domain_mask)
+ return -ENOMEM;
+
+ 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;
+ }
+
+ tpmi_hp_state = ret;
+
+ return 0;
+}
+module_init(tpmi_init)
+
+static void __exit tpmi_exit(void)
+{
+ cpuhp_remove_state(tpmi_hp_state);
+ kfree(tpmi_power_domain_mask);
+}
+module_exit(tpmi_exit)
+
+MODULE_DESCRIPTION("TPMI Power Domains Mapping");
+MODULE_LICENSE("GPL");
diff --git a/drivers/platform/x86/intel/tpmi_power_domains.h b/drivers/platform/x86/intel/tpmi_power_domains.h
new file mode 100644
index 000000000000..0c2154bd941f
--- /dev/null
+++ b/drivers/platform/x86/intel/tpmi_power_domains.h
@@ -0,0 +1,19 @@
+/* SPDX-License-Identifier: GPL-2.0-only */
+/*
+ * Mapping of TPMI power domain and CPUs
+ *
+ * Copyright (c) 2024, Intel Corporation.
+ * All rights reserved.
+ */
+
+#ifndef _TPMI_POWER_DOMAINS_H_
+#define _TPMI_POWER_DOMAINS_H_
+
+#include <linux/cpumask.h>
+
+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);
+
+#endif
--
2.43.1
On Tue, 28 May 2024, Tero Kristo wrote:
> From: Srinivas Pandruvada <[email protected]>
>
> Each TPMI power domain includes a group of CPUs. Several power
> management settings in this case applicable to a group of CPUs.
> There can be several power domains in a CPU package. So, provide
> interfaces for:
> - Get power domain id for a Linux CPU
> - Get mask of Linux CPUs in a power domain
>
> Hardware Punit uses different CPU numbering, which is not based on
> APIC (Advanced Programmable Interrupt Controller) CPU numbering.
> The Linux CPU numbering is based on APIC CPU numbering. Some PM features
> like Intel Speed Select, the CPU core mask provided by the hardware is
> based on the Punit CPU numbering. To use the core mask, this mask
> needs to be converted to a Linux CPUs mask. So, provide interfaces for:
> - Convert to a Linux CPU number from a Punit CPU number
> - Convert to a Punit CPU number from a Linux CPU number
>
> On each CPU online, MSR 0x54 is used to read the mapping and stores in
> a per cpu array. Create a hash for faster searching of a Linux CPU number
> from a Punit CPU number.
>
> Signed-off-by: Srinivas Pandruvada <[email protected]>
> [tero.kristo: minor updates]
> Reviewed-by: Andy Shevchenko <[email protected]>
> Reviewed-by: Ilpo Järvinen <[email protected]>
> Signed-off-by: Tero Kristo <[email protected]>
> ---
> v2:
> * changed to use X86_MATCH_VFM() instead of X86_MATCH_INTEL_FAM6_MODEL()
I've applied this v2 + the other patches from v1 series now to review-ilpo
branch.
For the record, I removed "All rights reserved." lines from the patches
while applying. I asking first (privately) a permission from Tero whether
it's okay with him I remove those lines.
--
i.
On Fri, 2024-05-31 at 16:45 +0300, Ilpo Järvinen wrote:
> On Tue, 28 May 2024, Tero Kristo wrote:
>
> > From: Srinivas Pandruvada <[email protected]>
> >
> > Each TPMI power domain includes a group of CPUs. Several power
> > management settings in this case applicable to a group of CPUs.
> > There can be several power domains in a CPU package. So, provide
> > interfaces for:
> > - Get power domain id for a Linux CPU
> > - Get mask of Linux CPUs in a power domain
> >
> > Hardware Punit uses different CPU numbering, which is not based on
> > APIC (Advanced Programmable Interrupt Controller) CPU numbering.
> > The Linux CPU numbering is based on APIC CPU numbering. Some PM
> > features
> > like Intel Speed Select, the CPU core mask provided by the hardware
> > is
> > based on the Punit CPU numbering. To use the core mask, this mask
> > needs to be converted to a Linux CPUs mask. So, provide interfaces
> > for:
> > - Convert to a Linux CPU number from a Punit CPU number
> > - Convert to a Punit CPU number from a Linux CPU number
> >
> > On each CPU online, MSR 0x54 is used to read the mapping and stores
> > in
> > a per cpu array. Create a hash for faster searching of a Linux CPU
> > number
> > from a Punit CPU number.
> >
> > Signed-off-by: Srinivas Pandruvada
> > <[email protected]>
> > [tero.kristo: minor updates]
> > Reviewed-by: Andy Shevchenko <[email protected]>
> > Reviewed-by: Ilpo Järvinen <[email protected]>
> > Signed-off-by: Tero Kristo <[email protected]>
> > ---
> > v2:
> > * changed to use X86_MATCH_VFM() instead of
> > X86_MATCH_INTEL_FAM6_MODEL()
>
> I've applied this v2 + the other patches from v1 series now to
> review-ilpo
> branch.
>
> For the record, I removed "All rights reserved." lines from the
> patches
> while applying. I asking first (privately) a permission from Tero
> whether
> it's okay with him I remove those lines.
>
Thanks Ilpo!
-Tero