[PATCH v7 2/3] cpufreq: CPPC: Add ospm_nominal_freq attribute

[Sumit Gupta <sumitg@nvidia.com>]

OSPM Nominal Performance (ACPI 6.6, Section 8.4.6.1.2.6) lets the OS
request a nominal performance level below the platform-reported one. The
platform treats performance above that level as boosted and below it as
throttled for its power and thermal decisions. A lower value moves that
boundary down, so sustained work runs at a lower point while the range
above it remains available as boost.

Expose it as a per-policy cpufreq attribute in kHz, matching the unit
convention of the other frequency attributes:

  /sys/devices/system/cpu/cpuX/cpufreq/ospm_nominal_freq

The attribute is write-only as the register cannot be read back. Writes
are converted with cppc_khz_to_perf() and rejected unless they fall in
[Lowest Performance, Nominal Performance].

Also track the register in the OSPM-set save/restore table, so a
requested value survives CPU hotplug and suspend/resume. The store
handler records the request rather than init() capturing a firmware
value, and driver unload reverts the register to the platform-reported
Nominal Performance.

Signed-off-by: Sumit Gupta <sumitg@nvidia.com>
---
 .../ABI/testing/sysfs-devices-system-cpu      | 24 +++++++
 drivers/cpufreq/cppc_cpufreq.c                | 70 +++++++++++++++++--
 2 files changed, 90 insertions(+), 4 deletions(-)

diff --git a/Documentation/ABI/testing/sysfs-devices-system-cpu b/Documentation/ABI/testing/sysfs-devices-system-cpu
index 82d10d556cc8..59aafcb2af97 100644
--- a/Documentation/ABI/testing/sysfs-devices-system-cpu
+++ b/Documentation/ABI/testing/sysfs-devices-system-cpu
@@ -346,6 +346,30 @@ Description:	Performance Limited
 
 		This file is only present if the cppc-cpufreq driver is in use.
 
+What:		/sys/devices/system/cpu/cpuX/cpufreq/ospm_nominal_freq
+Date:		August 2026
+Contact:	linux-pm@vger.kernel.org
+Description:	OSPM Nominal Performance (kHz), write-only
+
+		OSPM uses this attribute to request a nominal performance level
+		lower than the platform-reported nominal. The platform treats
+		performance above this level as boost and below as throttle for
+		power and thermal decisions.
+
+		Write a value in kHz, between the frequencies corresponding to
+		Lowest Performance and Nominal Performance. The register cannot
+		be read back, so this attribute has no read side.
+
+		Note that tasks may be migrated from one CPU to another by the
+		scheduler's load-balancing algorithm, and if different OSPM
+		Nominal Performance values are set for those CPUs (through
+		different cpufreq policies), that may lead to undesirable
+		outcomes. To avoid such issues it is better to set the same
+		value across all policies, or to pin every task potentially
+		sensitive to it to a specific CPU.
+
+		This file is only present if the cppc-cpufreq driver is in use.
+
 What:		/sys/devices/system/cpu/cpu*/cache/index3/cache_disable_{0,1}
 Date:		August 2008
 KernelVersion:	2.6.27
diff --git a/drivers/cpufreq/cppc_cpufreq.c b/drivers/cpufreq/cppc_cpufreq.c
index 32f38b0c492b..fe714e71826a 100644
--- a/drivers/cpufreq/cppc_cpufreq.c
+++ b/drivers/cpufreq/cppc_cpufreq.c
@@ -33,11 +33,13 @@ static struct cpufreq_driver cppc_cpufreq_driver;
  * reapplied from online() across CPU hotplug, and the firmware value is
  * restored from offline().
  *
- * Autonomous Selection (auto_sel) is kept first, as writes to the registers
- * listed after it only have meaning while autonomous selection is enabled.
+ * Autonomous Selection (auto_sel) splits the list: the registers before it are
+ * independent of it, and those after it have meaning only while autonomous
+ * selection is enabled. Place a new register on the matching side.
  */
 enum cppc_saved_reg_id {
-	CPPC_SAVED_AUTO_SEL,
+	CPPC_SAVED_OSPM_NOMINAL_PERF,
+	CPPC_SAVED_AUTO_SEL,	/* Entries below need auto_sel enabled. */
 	CPPC_SAVED_EPP,
 	CPPC_SAVED_AUTO_ACT_WINDOW,
 	CPPC_NR_SAVED_REGS,
@@ -50,6 +52,11 @@ struct cppc_saved_reg {
 };
 
 static const struct cppc_saved_reg cppc_saved_regs[CPPC_NR_SAVED_REGS] = {
+	/* Write-only: the requested value is tracked in software. */
+	[CPPC_SAVED_OSPM_NOMINAL_PERF] = {
+		.name = "ospm_nominal_perf",
+		.set = cppc_set_ospm_nominal_perf,
+	},
 	[CPPC_SAVED_AUTO_SEL] = {
 		.name = "auto_sel",
 		.get = cppc_get_auto_sel,
@@ -79,6 +86,9 @@ enum cppc_saved_type {
  *                   could not be read
  *   requested_val - value in effect when the policy last went offline,
  *                   reapplied at online(). U64_MAX if none
+ *
+ * A write-only register cannot be read back, so its store handler sets
+ * firmware_val and requested_val instead of init() and offline().
  */
 struct cppc_saved_vals {
 	u64 firmware_val;
@@ -135,7 +145,17 @@ static void cppc_cpufreq_save_regs(struct cpufreq_policy *policy,
 		st->suspend_regs_handled = false;
 
 	for (i = 0; i < CPPC_NR_SAVED_REGS; i++) {
-		if (cppc_saved_regs[i].get(cpu, &val))
+		const struct cppc_saved_reg *reg = &cppc_saved_regs[i];
+
+		/*
+		 * A write-only register cannot be read back. It has no
+		 * firmware value to capture, and its requested value comes
+		 * from the store handler, so do not overwrite it here.
+		 */
+		if (!reg->get && saved_type == CPPC_SAVED_REQUESTED)
+			continue;
+
+		if (!reg->get || reg->get(cpu, &val))
 			val = U64_MAX;
 
 		if (saved_type == CPPC_SAVED_FIRMWARE) {
@@ -197,6 +217,12 @@ static void cppc_cpufreq_apply_saved_regs(struct cpufreq_policy *policy,
 	u64 auto_sel, val;
 	int i;
 
+	/* Registers before auto_sel do not depend on it. */
+	for (i = 0; i < CPPC_SAVED_AUTO_SEL; i++) {
+		val = cppc_cpufreq_saved_reg_value(st, i, saved_type);
+		cppc_cpufreq_write_saved_reg(cpu, i, val, saved_type);
+	}
+
 	auto_sel = cppc_cpufreq_saved_reg_value(st, CPPC_SAVED_AUTO_SEL,
 						saved_type);
 
@@ -1385,11 +1411,46 @@ static int cppc_get_perf_limited_filtered(int cpu, u64 *perf_limited)
 CPPC_CPUFREQ_ATTR_RW_U64(perf_limited, cppc_get_perf_limited_filtered,
 			 cppc_set_perf_limited)
 
+static ssize_t store_ospm_nominal_freq(struct cpufreq_policy *policy,
+				       const char *buf, size_t count)
+{
+	struct cppc_cpudata *cpu_data = policy->driver_data;
+	struct cppc_saved_vals *st;
+	unsigned int freq_khz;
+	u32 perf;
+	int ret;
+
+	ret = kstrtouint(buf, 0, &freq_khz);
+	if (ret)
+		return ret;
+
+	perf = cppc_khz_to_perf(&cpu_data->perf_caps, freq_khz);
+	if (perf < cpu_data->perf_caps.lowest_perf ||
+	    perf > cpu_data->perf_caps.nominal_perf)
+		return -EINVAL;
+
+	ret = cppc_set_ospm_nominal_perf(policy->cpu, perf);
+	if (ret)
+		return ret;
+
+	/*
+	 * Track the request in software: requested_val is reapplied across
+	 * hotplug, and firmware_val makes the register revert to the platform
+	 * Nominal on driver unload, since the value cannot be read back.
+	 */
+	st = &cppc_cpufreq_policy_state(policy)->regs[CPPC_SAVED_OSPM_NOMINAL_PERF];
+	st->requested_val = perf;
+	st->firmware_val = cpu_data->perf_caps.nominal_perf;
+
+	return count;
+}
+
 cpufreq_freq_attr_ro(freqdomain_cpus);
 cpufreq_freq_attr_rw(auto_select);
 cpufreq_freq_attr_rw(auto_act_window);
 cpufreq_freq_attr_rw(energy_performance_preference_val);
 cpufreq_freq_attr_rw(perf_limited);
+cpufreq_freq_attr_wo(ospm_nominal_freq);
 
 static struct freq_attr *cppc_cpufreq_attr[] = {
 	&freqdomain_cpus,
@@ -1397,6 +1458,7 @@ static struct freq_attr *cppc_cpufreq_attr[] = {
 	&auto_act_window,
 	&energy_performance_preference_val,
 	&perf_limited,
+	&ospm_nominal_freq,
 	NULL,
 };
 
-- 
2.34.1