Skip to contentSkip to conversation

cpufreq: intel_pstate: A fix and some cleanups

PATCH In review Rafael J. Wysocki 2 participants 6 messages Jun 24 – Jun 30

Subsystems (2)
lore
Share & tools
lore

Rafael J. Wysocki

author maintainer
#0 lore raw

v3 cover cpufreq: intel_pstate: A fix and some cleanups

Hi,

This replaces

https://lore.kernel.org/linux-pm/[email protected]/

and drops the most significant change in it which was making intel_pstate set cpuinfo_min_freq to a lower value with HWP enabled. That change went too far and needs to be reconsidered.

The other changes in the series, and in particular the fix in the first patch, are useful and will facilitate further improvements, so here they go.

This series is independent of

https://lore.kernel.org/linux-pm/[email protected]/

and applies to Linux 7.1.

Thanks!

author maintainer
#1 lore raw

PATCH v3 1/4 cpufreq: intel_pstate: Fix setting minimum P-state at init time

From: "Rafael J. Wysocki" <[email protected]>

If HWP is enabled, writes to MSR_IA32_PERF_CTL have no effect, so intel_pstate_get_cpu_pstates() should not attempt to call intel_pstate_set_min_pstate() to set the minimum P-state for the given CPU in that case.

Accordingly, remove the intel_pstate_set_min_pstate() call from intel_pstate_get_cpu_pstates() and make both intel_pstate_cpu_init() and intel_cpufreq_cpu_init() call that function in their non-HWP code paths.

The HWP code path in intel_pstate_cpu_init() does not need to update the current P-state of the CPU directly at all because it is taken care of the processor automatically, but the HWP code path of intel_cpufreq_cpu_init() should update it in principle to initialize the DESIRED_PERF field in MSR_HWP_REQUEST. For this purpose, make it call intel_cpufreq_hwp_update() and pass the minimum P-state limit to it as the current target value along with the current minimum and maximum limits.

Fixes f6ebbcf08f37 ("cpufreq: intel_pstate: Implement passive mode with HWP enabled")
---
 drivers/cpufreq/intel_pstate.c | 19 +++++++++++--------
 1 file changed, 11 insertions(+), 8 deletions(-)
drivers/cpufreq/intel_pstate.c +11 -8
index 1f093e346430..95e41e975c45 100644
@@ -2354,8 +2354,6 @@ static void intel_pstate_get_cpu_pstates(struct cpudata *cpu)
 
 	if (pstate_funcs.get_vid)
 		pstate_funcs.get_vid(cpu);
-
-	intel_pstate_set_min_pstate(cpu);
 }
 
 /*
@@ -3062,6 +3060,7 @@ static int __intel_pstate_cpu_init(struct cpufreq_policy *policy)
 static int intel_pstate_cpu_init(struct cpufreq_policy *policy)
 {
 	int ret = __intel_pstate_cpu_init(policy);
+	struct cpudata *cpu;
 
 	if (ret)
 		return ret;
@@ -3072,11 +3071,11 @@ static int intel_pstate_cpu_init(struct cpufreq_policy *policy)
 	 */
 	policy->policy = CPUFREQ_POLICY_POWERSAVE;
 
-	if (hwp_active) {
-		struct cpudata *cpu = all_cpu_data[policy->cpu];
-
+	cpu = all_cpu_data[policy->cpu];
+	if (hwp_active)
 		cpu->epp_cached = intel_pstate_get_epp(cpu, 0);
-	}
+	else
+		intel_pstate_set_min_pstate(cpu);
 
 	return 0;
 }
@@ -3300,8 +3299,6 @@ static int intel_cpufreq_cpu_init(struct cpufreq_policy *policy)
 		return ret;
 
 	policy->cpuinfo.transition_latency = INTEL_CPUFREQ_TRANSITION_LATENCY;
-	/* This reflects the intel_pstate_get_cpu_pstates() setting. */
-	policy->cur = policy->cpuinfo.min_freq;
 
 	req = kzalloc_objs(*req, 2);
 	if (!req) {
@@ -3322,9 +3319,15 @@ static int intel_cpufreq_cpu_init(struct cpufreq_policy *policy)
 		WRITE_ONCE(cpu->hwp_req_cached, value);
 
 		cpu->epp_cached = intel_pstate_get_epp(cpu, value);
+
+		intel_cpufreq_hwp_update(cpu, cpu->pstate.min_pstate,
+					 cpu->pstate.max_pstate,
+					 cpu->pstate.min_pstate, false);
 	} else {
 		policy->transition_delay_us = INTEL_CPUFREQ_TRANSITION_DELAY;
+		intel_pstate_set_min_pstate(cpu);
 	}
+	policy->cur = policy->cpuinfo.min_freq;
 
 	freq = DIV_ROUND_UP(cpu->pstate.turbo_freq * global.min_perf_pct, 100);
 
Signature · 1 line
2.51.0
author maintainer
#2 lore raw

PATCH v3 2/4 cpufreq: intel_pstate: Introduce intel_pstate_update_freq_limits()

In reply to Rafael J. Wysocki

From: "Rafael J. Wysocki" <[email protected]>

Introduce a new helper function, intel_pstate_update_freq_limits(), for updating the max and turbo frequency values for the given CPU after updating the corresponding P-states.

Use it in intel_pstate_get_hwp_cap() and intel_pstate_get_cpu_pstates(), in the latter case instead of the direct updates of the max and turbo frequency values in intel_pstate_hybrid_hwp_adjust().

No intentional functional impact.

---
 drivers/cpufreq/intel_pstate.c | 36 +++++++++++++++++-----------------
 1 file changed, 18 insertions(+), 18 deletions(-)
drivers/cpufreq/intel_pstate.c +18 -18
index 95e41e975c45..e143c80a3215 100644
@@ -587,11 +587,6 @@ static void intel_pstate_hybrid_hwp_adjust(struct cpudata *cpu)
 
 	hwp_is_hybrid = true;
 
-	cpu->pstate.turbo_freq = rounddown(cpu->pstate.turbo_pstate * scaling,
-					   perf_ctl_scaling);
-	cpu->pstate.max_freq = rounddown(cpu->pstate.max_pstate * scaling,
-					 perf_ctl_scaling);
-
 	freq = perf_ctl_max_phys * perf_ctl_scaling;
 	cpu->pstate.max_pstate_physical = intel_pstate_freq_to_hwp(cpu, freq);
 
@@ -1183,6 +1178,22 @@ static bool hybrid_clear_max_perf_cpu(void)
 	return ret;
 }
 
+static void intel_pstate_update_freq_limits(struct cpudata *cpu)
+{
+	int scaling = cpu->pstate.scaling;
+	unsigned int turbo_freq = cpu->pstate.turbo_pstate * scaling;
+	unsigned int max_freq = cpu->pstate.max_pstate * scaling;
+	int perf_ctl_scaling = cpu->pstate.perf_ctl_scaling;
+
+	if (scaling != perf_ctl_scaling) {
+		turbo_freq = rounddown(turbo_freq, perf_ctl_scaling);
+		max_freq = rounddown(max_freq, perf_ctl_scaling);
+	}
+
+	cpu->pstate.turbo_freq = turbo_freq;
+	cpu->pstate.max_freq = max_freq;
+}
+
 static void __intel_pstate_get_hwp_cap(struct cpudata *cpu)
 {
 	u64 cap;
@@ -1195,20 +1206,8 @@ static void __intel_pstate_get_hwp_cap(struct cpudata *cpu)
 
 static void intel_pstate_get_hwp_cap(struct cpudata *cpu)
 {
-	int scaling = cpu->pstate.scaling;
-
 	__intel_pstate_get_hwp_cap(cpu);
-
-	cpu->pstate.max_freq = cpu->pstate.max_pstate * scaling;
-	cpu->pstate.turbo_freq = cpu->pstate.turbo_pstate * scaling;
-	if (scaling != cpu->pstate.perf_ctl_scaling) {
-		int perf_ctl_scaling = cpu->pstate.perf_ctl_scaling;
-
-		cpu->pstate.max_freq = rounddown(cpu->pstate.max_freq,
-						 perf_ctl_scaling);
-		cpu->pstate.turbo_freq = rounddown(cpu->pstate.turbo_freq,
-						   perf_ctl_scaling);
-	}
+	intel_pstate_update_freq_limits(cpu);
 }
 
 static void hybrid_update_capacity(struct cpudata *cpu)
@@ -2329,6 +2328,7 @@ static void intel_pstate_get_cpu_pstates(struct cpudata *cpu)
 		if (pstate_funcs.get_cpu_scaling) {
 			cpu->pstate.scaling = pstate_funcs.get_cpu_scaling(cpu->cpu);
 			intel_pstate_hybrid_hwp_adjust(cpu);
+			intel_pstate_update_freq_limits(cpu);
 		} else {
 			cpu->pstate.scaling = perf_ctl_scaling;
 		}
Signature · 1 line
2.51.0
author maintainer
#3 lore raw

PATCH v3 3/4 cpufreq: intel_pstate: Consolidate frequency values computation

In reply to Rafael J. Wysocki

From: "Rafael J. Wysocki" <[email protected]>

Update intel_pstate_get_cpu_pstates() to use intel_pstate_update_freq_limits() for computing the max and turbo frequency values in all cases, including non-hybrid HWP and HWP disabled.

No intentional functional impact.

---
 drivers/cpufreq/intel_pstate.c | 12 +++---------
 1 file changed, 3 insertions(+), 9 deletions(-)
drivers/cpufreq/intel_pstate.c +3 -9
index e143c80a3215..edc90ecb0b00 100644
@@ -590,13 +590,11 @@ static void intel_pstate_hybrid_hwp_adjust(struct cpudata *cpu)
 	freq = perf_ctl_max_phys * perf_ctl_scaling;
 	cpu->pstate.max_pstate_physical = intel_pstate_freq_to_hwp(cpu, freq);
 
-	freq = cpu->pstate.min_pstate * perf_ctl_scaling;
-	cpu->pstate.min_freq = freq;
 	/*
 	 * Cast the min P-state value retrieved via pstate_funcs.get_min() to
 	 * the effective range of HWP performance levels.
 	 */
-	cpu->pstate.min_pstate = intel_pstate_freq_to_hwp(cpu, freq);
+	cpu->pstate.min_pstate = intel_pstate_freq_to_hwp(cpu, cpu->pstate.min_freq);
 }
 
 static bool turbo_is_disabled(void)
@@ -2320,6 +2318,7 @@ static void intel_pstate_get_cpu_pstates(struct cpudata *cpu)
 
 	cpu->pstate.max_pstate_physical = pstate_funcs.get_max_physical(cpu->cpu);
 	cpu->pstate.min_pstate = pstate_funcs.get_min(cpu->cpu);
+	cpu->pstate.min_freq = cpu->pstate.min_pstate * perf_ctl_scaling;
 	cpu->pstate.perf_ctl_scaling = perf_ctl_scaling;
 
 	if (hwp_active && !hwp_mode_bdw) {
@@ -2328,7 +2327,6 @@ static void intel_pstate_get_cpu_pstates(struct cpudata *cpu)
 		if (pstate_funcs.get_cpu_scaling) {
 			cpu->pstate.scaling = pstate_funcs.get_cpu_scaling(cpu->cpu);
 			intel_pstate_hybrid_hwp_adjust(cpu);
-			intel_pstate_update_freq_limits(cpu);
 		} else {
 			cpu->pstate.scaling = perf_ctl_scaling;
 		}
@@ -2343,11 +2341,7 @@ static void intel_pstate_get_cpu_pstates(struct cpudata *cpu)
 		cpu->pstate.turbo_pstate = pstate_funcs.get_turbo(cpu->cpu);
 	}
 
-	if (cpu->pstate.scaling == perf_ctl_scaling) {
-		cpu->pstate.min_freq = cpu->pstate.min_pstate * perf_ctl_scaling;
-		cpu->pstate.max_freq = cpu->pstate.max_pstate * perf_ctl_scaling;
-		cpu->pstate.turbo_freq = cpu->pstate.turbo_pstate * perf_ctl_scaling;
-	}
+	intel_pstate_update_freq_limits(cpu);
 
 	if (pstate_funcs.get_aperf_mperf_shift)
 		cpu->aperf_mperf_shift = pstate_funcs.get_aperf_mperf_shift();
Signature · 1 line
2.51.0
author maintainer
#4 lore raw

PATCH v3 4/4 cpufreq: intel_pstate: Move two functions closer to callers

In reply to Rafael J. Wysocki

From: "Rafael J. Wysocki" <[email protected]>

Move intel_pstate_set_pstate() and intel_pstate_set_min_pstate() that are not used on systems with HWP enabled, closer to their first callers.

No intentional functional impact.

---
 drivers/cpufreq/intel_pstate.c | 36 +++++++++++++++++-----------------
 1 file changed, 18 insertions(+), 18 deletions(-)
drivers/cpufreq/intel_pstate.c +18 -18
index edc90ecb0b00..25ef2848eebf 100644
@@ -2294,24 +2294,6 @@ static int hwp_get_cpu_scaling(int cpu)
 	return intel_pstate_cppc_get_scaling(cpu);
 }
 
-static void intel_pstate_set_pstate(struct cpudata *cpu, int pstate)
-{
-	trace_cpu_frequency(pstate * cpu->pstate.scaling, cpu->cpu);
-	cpu->pstate.current_pstate = pstate;
-	/*
-	 * Generally, there is no guarantee that this code will always run on
-	 * the CPU being updated, so force the register update to run on the
-	 * right CPU.
-	 */
-	wrmsrq_on_cpu(cpu->cpu, MSR_IA32_PERF_CTL,
-		      pstate_funcs.get_val(cpu, pstate));
-}
-
-static void intel_pstate_set_min_pstate(struct cpudata *cpu)
-{
-	intel_pstate_set_pstate(cpu, cpu->pstate.min_pstate);
-}
-
 static void intel_pstate_get_cpu_pstates(struct cpudata *cpu)
 {
 	int perf_ctl_scaling = pstate_funcs.get_scaling();
@@ -2874,6 +2856,19 @@ static void intel_pstate_update_perf_limits(struct cpudata *cpu,
 		 cpu->min_perf_ratio);
 }
 
+static void intel_pstate_set_pstate(struct cpudata *cpu, int pstate)
+{
+	trace_cpu_frequency(pstate * cpu->pstate.scaling, cpu->cpu);
+	cpu->pstate.current_pstate = pstate;
+	/*
+	 * Generally, there is no guarantee that this code will always run on
+	 * the CPU being updated, so force the register update to run on the
+	 * right CPU.
+	 */
+	wrmsrq_on_cpu(cpu->cpu, MSR_IA32_PERF_CTL,
+		      pstate_funcs.get_val(cpu, pstate));
+}
+
 static int intel_pstate_set_policy(struct cpufreq_policy *policy)
 {
 	struct cpudata *cpu;
@@ -2961,6 +2956,11 @@ static int intel_pstate_verify_policy(struct cpufreq_policy_data *policy)
 	return 0;
 }
 
+static void intel_pstate_set_min_pstate(struct cpudata *cpu)
+{
+	intel_pstate_set_pstate(cpu, cpu->pstate.min_pstate);
+}
+
 static int intel_cpufreq_cpu_offline(struct cpufreq_policy *policy)
 {
 	struct cpudata *cpu = all_cpu_data[policy->cpu];
Signature · 1 line
2.51.0
#5 lore raw
In reply to Rafael J. Wysocki

Hi Rafael,

Quoted 6 lines · On 2026.06.24 10:32 Rafael wrote:

Hi,

This replaces

https://lore.kernel.org/linux-pm/[email protected]/

Quoted 7 lines

and drops the most significant change in it which was making intel_pstate set cpuinfo_min_freq to a lower value with HWP enabled. That change went too far and needs to be reconsidered.

The other changes in the series, and in particular the fix in the first patch, are useful and will facilitate further improvements, so here they go.

I tested this patch set a little. It seems okay.

... Doug

Quoted 8 lines

This series is independent of

https://lore.kernel.org/linux-pm/[email protected]/

and applies to Linux 7.1.

Thanks!