[PATCH 08/12] platform/x86: ISST: Add SST-CP support via TPMI

From: Srinivas Pandruvada
Date: Sat Feb 11 2023 - 01:33:45 EST


Intel Speed Select Technology Core Power (SST-CP) is an interface that
allows users to define per core priority. This defines a mechanism to
distribute power among cores when there is a power constrained
scenario. This defines a class of service (CLOS) configuration.

Three new IOCTLs are added:
ISST_IF_CORE_POWER_STATE : Enable/Disable SST-CP
ISST_IF_CLOS_PARAM : Configure CLOS parameters
ISST_IF_CLOS_ASSOC : Associate CPUs to a CLOS

To associate CPUs to CLOS, either Linux CPU numbering or PUNIT numbering
scheme can be used, using parameter punit_cpu_map (1: for PUNIT numbering
0 for Linux CPU number).

There is no change to IOCTL to get PUNIT CPU number for a CPU.

Introduce get_instance() function, which is used by majority of IOCTLs
processing to convert a socket and power domain to
tpmi_per_power_domain_info * instance. This instance has all the MMIO
offsets stored to read a particular field.

Once an instance is identified, read or write from correct MMIO
offset for a given field as defined in the specification.

For details on SST CP operations using intel-speed-selet utility,
refer to:
Documentation/admin-guide/pm/intel-speed-select.rst
under the kernel documentation

Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@xxxxxxxxxxxxxxx>
---
.../intel/speed_select_if/isst_tpmi_core.c | 264 ++++++++++++++++++
include/uapi/linux/isst_if.h | 79 ++++++
2 files changed, 343 insertions(+)

diff --git a/drivers/platform/x86/intel/speed_select_if/isst_tpmi_core.c b/drivers/platform/x86/intel/speed_select_if/isst_tpmi_core.c
index 3453708c2dd0..bcdf51517c88 100644
--- a/drivers/platform/x86/intel/speed_select_if/isst_tpmi_core.c
+++ b/drivers/platform/x86/intel/speed_select_if/isst_tpmi_core.c
@@ -31,6 +31,18 @@
/* Supported SST hardware version by this driver */
#define ISST_HEADER_VERSION 1

+/*
+ * Used to indicate if value read from MMIO needs to get multiplied
+ * to get to a standard unit or not.
+ */
+#define SST_MUL_FACTOR_NONE 1
+
+/* Define 100 as a scaling factor frequency ratio to frequency conversion */
+#define SST_MUL_FACTOR_FREQ 100
+
+/* All SST regs are 64 bit size */
+#define SST_REG_SIZE 8
+
/**
* struct sst_header - SST main header
* @interface_version: Version number for this interface
@@ -359,6 +371,249 @@ static int sst_main(struct auxiliary_device *auxdev, struct tpmi_per_power_domai
return 0;
}

+/*
+ * Map a package and power_domain id to SST information structure unique for a power_domain.
+ * The caller should call under isst_tpmi_dev_lock.
+ */
+static struct tpmi_per_power_domain_info *get_instance(int pkg_id, int power_domain_id)
+{
+ struct tpmi_per_power_domain_info *power_domain_info;
+ struct tpmi_sst_struct *sst_inst;
+
+ if (pkg_id < 0 || pkg_id > isst_common.max_index ||
+ pkg_id >= topology_max_packages())
+ return NULL;
+
+ sst_inst = isst_common.sst_inst[pkg_id];
+ if (!sst_inst)
+ return NULL;
+
+ if (power_domain_id < 0 || power_domain_id >= sst_inst->number_of_power_domains)
+ return NULL;
+
+ power_domain_info = &sst_inst->power_domain_info[power_domain_id];
+
+ if (power_domain_info && !power_domain_info->sst_base)
+ return NULL;
+
+ return power_domain_info;
+}
+
+static bool disable_dynamic_sst_features(void)
+{
+ u64 value;
+
+ rdmsrl(MSR_PM_ENABLE, value);
+ return !(value & 0x1);
+}
+
+#define _read_cp_info(name_str, name, offset, start, width, mult_factor)\
+{\
+ u64 val, mask;\
+ \
+ val = readq(power_domain_info->sst_base + power_domain_info->sst_header.cp_offset +\
+ (offset));\
+ mask = GENMASK_ULL((start + width - 1), start);\
+ val &= mask; \
+ val >>= start;\
+ name = (val * mult_factor);\
+}
+
+#define _write_cp_info(name_str, name, offset, start, width, div_factor)\
+{\
+ u64 val, mask;\
+ \
+ val = readq(power_domain_info->sst_base +\
+ power_domain_info->sst_header.cp_offset + (offset));\
+ mask = GENMASK_ULL((start + width - 1), start);\
+ val &= ~mask;\
+ val |= (name / div_factor) << start;\
+ writeq(val, power_domain_info->sst_base + power_domain_info->sst_header.cp_offset +\
+ (offset));\
+}
+
+#define SST_CP_CONTROL_OFFSET 8
+#define SST_CP_STATUS_OFFSET 16
+
+#define SST_CP_ENABLE_START 0
+#define SST_CP_ENABLE_WIDTH 1
+
+#define SST_CP_PRIORITY_TYPE_START 1
+#define SST_CP_PRIORITY_TYPE_WIDTH 1
+
+static long isst_if_core_power_state(void __user *argp)
+{
+ struct tpmi_per_power_domain_info *power_domain_info;
+ struct isst_core_power core_power;
+
+ if (disable_dynamic_sst_features())
+ return -EFAULT;
+
+ if (copy_from_user(&core_power, argp, sizeof(core_power)))
+ return -EFAULT;
+
+ power_domain_info = get_instance(core_power.socket_id, core_power.power_domain_id);
+ if (!power_domain_info)
+ return -EINVAL;
+
+ if (core_power.get_set) {
+ _write_cp_info("cp_enable", core_power.enable, SST_CP_CONTROL_OFFSET,
+ SST_CP_ENABLE_START, SST_CP_ENABLE_WIDTH, SST_MUL_FACTOR_NONE)
+ _write_cp_info("cp_prio_type", core_power.priority_type, SST_CP_CONTROL_OFFSET,
+ SST_CP_PRIORITY_TYPE_START, SST_CP_PRIORITY_TYPE_WIDTH,
+ SST_MUL_FACTOR_NONE)
+ } else {
+ /* get */
+ _read_cp_info("cp_enable", core_power.enable, SST_CP_STATUS_OFFSET,
+ SST_CP_ENABLE_START, SST_CP_ENABLE_WIDTH, SST_MUL_FACTOR_NONE)
+ _read_cp_info("cp_prio_type", core_power.priority_type, SST_CP_STATUS_OFFSET,
+ SST_CP_PRIORITY_TYPE_START, SST_CP_PRIORITY_TYPE_WIDTH,
+ SST_MUL_FACTOR_NONE)
+ core_power.supported = !!(power_domain_info->sst_header.cap_mask & BIT(0));
+ if (copy_to_user(argp, &core_power, sizeof(core_power)))
+ return -EFAULT;
+ }
+
+ return 0;
+}
+
+#define SST_CLOS_CONFIG_0_OFFSET 24
+
+#define SST_CLOS_CONFIG_PRIO_START 4
+#define SST_CLOS_CONFIG_PRIO_WIDTH 4
+
+#define SST_CLOS_CONFIG_MIN_START 8
+#define SST_CLOS_CONFIG_MIN_WIDTH 8
+
+#define SST_CLOS_CONFIG_MAX_START 16
+#define SST_CLOS_CONFIG_MAX_WIDTH 8
+
+static long isst_if_clos_param(void __user *argp)
+{
+ struct tpmi_per_power_domain_info *power_domain_info;
+ struct isst_clos_param clos_param;
+
+ if (copy_from_user(&clos_param, argp, sizeof(clos_param)))
+ return -EFAULT;
+
+ power_domain_info = get_instance(clos_param.socket_id, clos_param.power_domain_id);
+ if (!power_domain_info)
+ return -EINVAL;
+
+ if (clos_param.get_set) {
+ _write_cp_info("clos.min_freq", clos_param.min_freq_mhz,
+ (SST_CLOS_CONFIG_0_OFFSET + clos_param.clos * SST_REG_SIZE),
+ SST_CLOS_CONFIG_MIN_START, SST_CLOS_CONFIG_MIN_WIDTH,
+ SST_MUL_FACTOR_FREQ);
+ _write_cp_info("clos.max_freq", clos_param.max_freq_mhz,
+ (SST_CLOS_CONFIG_0_OFFSET + clos_param.clos * SST_REG_SIZE),
+ SST_CLOS_CONFIG_MAX_START, SST_CLOS_CONFIG_MAX_WIDTH,
+ SST_MUL_FACTOR_FREQ);
+ _write_cp_info("clos.prio", clos_param.prop_prio,
+ (SST_CLOS_CONFIG_0_OFFSET + clos_param.clos * SST_REG_SIZE),
+ SST_CLOS_CONFIG_PRIO_START, SST_CLOS_CONFIG_PRIO_WIDTH,
+ SST_MUL_FACTOR_NONE);
+ } else {
+ /* get */
+ _read_cp_info("clos.min_freq", clos_param.min_freq_mhz,
+ (SST_CLOS_CONFIG_0_OFFSET + clos_param.clos * SST_REG_SIZE),
+ SST_CLOS_CONFIG_MIN_START, SST_CLOS_CONFIG_MIN_WIDTH,
+ SST_MUL_FACTOR_FREQ)
+ _read_cp_info("clos.max_freq", clos_param.max_freq_mhz,
+ (SST_CLOS_CONFIG_0_OFFSET + clos_param.clos * SST_REG_SIZE),
+ SST_CLOS_CONFIG_MAX_START, SST_CLOS_CONFIG_MAX_WIDTH,
+ SST_MUL_FACTOR_FREQ)
+ _read_cp_info("clos.prio", clos_param.prop_prio,
+ (SST_CLOS_CONFIG_0_OFFSET + clos_param.clos * SST_REG_SIZE),
+ SST_CLOS_CONFIG_PRIO_START, SST_CLOS_CONFIG_PRIO_WIDTH,
+ SST_MUL_FACTOR_NONE)
+
+ if (copy_to_user(argp, &clos_param, sizeof(clos_param)))
+ return -EFAULT;
+ }
+
+ return 0;
+}
+
+#define SST_CLOS_ASSOC_0_OFFSET 56
+#define SST_CLOS_ASSOC_CPUS_PER_REG 16
+#define SST_CLOS_ASSOC_BITS_PER_CPU 4
+
+static long isst_if_clos_assoc(void __user *argp)
+{
+ struct isst_if_clos_assoc_cmds assoc_cmds;
+ unsigned char __user *ptr;
+ int i;
+
+ /* Each multi command has u16 command count as the first field */
+ if (copy_from_user(&assoc_cmds, argp, sizeof(assoc_cmds)))
+ return -EFAULT;
+
+ if (!assoc_cmds.cmd_count || assoc_cmds.cmd_count > ISST_IF_CMD_LIMIT)
+ return -EINVAL;
+
+ ptr = argp + offsetof(struct isst_if_clos_assoc_cmds, assoc_info);
+ for (i = 0; i < assoc_cmds.cmd_count; ++i) {
+ struct tpmi_per_power_domain_info *power_domain_info;
+ struct isst_if_clos_assoc clos_assoc;
+ int punit_id, punit_cpu_no, pkg_id;
+ struct tpmi_sst_struct *sst_inst;
+ int offset, shift, cpu;
+ u64 val, mask, clos;
+
+ if (copy_from_user(&clos_assoc, ptr, sizeof(clos_assoc)))
+ return -EFAULT;
+
+ if (clos_assoc.socket_id > topology_max_packages())
+ return -EINVAL;
+
+ cpu = clos_assoc.logical_cpu;
+ clos = clos_assoc.clos;
+
+ if (assoc_cmds.punit_cpu_map)
+ punit_cpu_no = cpu;
+ else
+ return -EOPNOTSUPP;
+
+ if (punit_cpu_no < 0)
+ return -EINVAL;
+
+ punit_id = clos_assoc.power_domain_id;
+ pkg_id = clos_assoc.socket_id;
+
+ sst_inst = isst_common.sst_inst[pkg_id];
+
+ if (clos_assoc.power_domain_id > sst_inst->number_of_power_domains)
+ return -EINVAL;
+
+ power_domain_info = &sst_inst->power_domain_info[punit_id];
+
+ offset = SST_CLOS_ASSOC_0_OFFSET +
+ (punit_cpu_no / SST_CLOS_ASSOC_CPUS_PER_REG) * SST_REG_SIZE;
+ shift = punit_cpu_no % SST_CLOS_ASSOC_CPUS_PER_REG;
+ shift *= SST_CLOS_ASSOC_BITS_PER_CPU;
+
+ val = readq(power_domain_info->sst_base +
+ power_domain_info->sst_header.cp_offset + offset);
+ if (assoc_cmds.get_set) {
+ mask = GENMASK_ULL((shift + SST_CLOS_ASSOC_BITS_PER_CPU - 1), shift);
+ val &= ~mask;
+ val |= (clos << shift);
+ writeq(val, power_domain_info->sst_base +
+ power_domain_info->sst_header.cp_offset + offset);
+ } else {
+ val >>= shift;
+ clos_assoc.clos = val & GENMASK(SST_CLOS_ASSOC_BITS_PER_CPU - 1, 0);
+ if (copy_to_user(ptr, &clos_assoc, sizeof(clos_assoc)))
+ return -EFAULT;
+ }
+
+ ptr += sizeof(clos_assoc);
+ }
+
+ return 0;
+}
+
static int isst_if_get_tpmi_instance_count(void __user *argp)
{
struct isst_tpmi_instance_count tpmi_inst;
@@ -400,6 +655,15 @@ static long isst_if_def_ioctl(struct file *file, unsigned int cmd,
case ISST_IF_COUNT_TPMI_INSTANCES:
ret = isst_if_get_tpmi_instance_count(argp);
break;
+ case ISST_IF_CORE_POWER_STATE:
+ ret = isst_if_core_power_state(argp);
+ break;
+ case ISST_IF_CLOS_PARAM:
+ ret = isst_if_clos_param(argp);
+ break;
+ case ISST_IF_CLOS_ASSOC:
+ ret = isst_if_clos_assoc(argp);
+ break;
default:
break;
}
diff --git a/include/uapi/linux/isst_if.h b/include/uapi/linux/isst_if.h
index bf32d959f6e8..32687d8023ef 100644
--- a/include/uapi/linux/isst_if.h
+++ b/include/uapi/linux/isst_if.h
@@ -163,6 +163,82 @@ struct isst_if_msr_cmds {
struct isst_if_msr_cmd msr_cmd[1];
};

+/**
+ * struct isst_core_power - Structure to get/set core_power feature
+ * @get_set: 0: Get, 1: Set
+ * @socket_id: Socket/package id
+ * @power_domain: Power Domain id
+ * @enable: Feature enable status
+ * @priority_type: Priority type for the feature (ordered/proportional)
+ *
+ * Structure to get/set core_power feature state using IOCTL
+ * ISST_IF_CORE_POWER_STATE.
+ */
+struct isst_core_power {
+ __u8 get_set;
+ __u8 socket_id;
+ __u8 power_domain_id;
+ __u8 enable;
+ __u8 supported;
+ __u8 priority_type;
+};
+
+/**
+ * struct isst_clos_param - Structure to get/set clos praram
+ * @get_set: 0: Get, 1: Set
+ * @socket_id: Socket/package id
+ * @power_domain: Power Domain id
+ * clos: Clos ID for the parameters
+ * min_freq_mhz: Minimum frequency in MHz
+ * max_freq_mhz: Maximum frequency in MHz
+ * prop_prio: Proportional priority from 0-15
+ *
+ * Structure to get/set per clos property using IOCTL
+ * ISST_IF_CLOS_PARAM.
+ */
+struct isst_clos_param {
+ __u8 get_set;
+ __u8 socket_id;
+ __u8 power_domain_id;
+ __u8 clos;
+ __u16 min_freq_mhz;
+ __u16 max_freq_mhz;
+ __u8 prop_prio;
+};
+
+/**
+ * struct isst_if_clos_assoc - Structure to assign clos to a CPU
+ * @socket_id: Socket/package id
+ * @power_domain: Power Domain id
+ * @logical_cpu: CPU number
+ * @clos: Clos ID to assign to the logical CPU
+ *
+ * Structure to get/set core_power feature.
+ */
+struct isst_if_clos_assoc {
+ __u8 socket_id;
+ __u8 power_domain_id;
+ __u16 logical_cpu;
+ __u16 clos;
+};
+
+/**
+ * struct isst_if_clos_assoc_cmds - Structure to assign clos to CPUs
+ * @cmd_count: Number of cmds (cpus) in this request
+ * @get_set: Request is for get or set
+ * @punit_cpu_map: Set to 1 if the CPU number is punit numbering not
+ * Linux CPU number
+ *
+ * Structure used to get/set associate CPUs to clos using IOCTL
+ * ISST_IF_CLOS_ASSOC.
+ */
+struct isst_if_clos_assoc_cmds {
+ __u16 cmd_count;
+ __u16 get_set;
+ __u16 punit_cpu_map;
+ struct isst_if_clos_assoc assoc_info[1];
+};
+
/**
* struct isst_tpmi_instance_count - Get number of TPMI instances per socket
* @socket_id: Socket/package id
@@ -186,5 +262,8 @@ struct isst_tpmi_instance_count {
#define ISST_IF_MSR_COMMAND _IOWR(ISST_IF_MAGIC, 4, struct isst_if_msr_cmds *)

#define ISST_IF_COUNT_TPMI_INSTANCES _IOR(ISST_IF_MAGIC, 5, struct isst_tpmi_instance_count *)
+#define ISST_IF_CORE_POWER_STATE _IOWR(ISST_IF_MAGIC, 6, struct isst_core_power *)
+#define ISST_IF_CLOS_PARAM _IOWR(ISST_IF_MAGIC, 7, struct isst_clos_param *)
+#define ISST_IF_CLOS_ASSOC _IOWR(ISST_IF_MAGIC, 8, struct isst_if_clos_assoc_cmds *)

#endif
--
2.39.1