Merge tag 'pm-4.10-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm

Pull power management fixes from Rafael Wysocki:
 "These add a quirk to intel_pstate to work around a firmware setting
  that leads to frequency scaling issues (discovered recently) on some
  Intel Kaby Lake processors, fix up the recently added brcmstb-avs
  cpufreq driver and avoid false-positive warnings from the runtime PM
  framework triggered by recent changes in i915.

  Specifics:

   - Add an intel_pstate driver quirk to work around a firmware setting
     that leads to frequency scaling issues on desktop Intel Kaby Lake
     processors in some configurations if the hardware-managed P-states
     (HWP) feature is in use (Srinivas Pandruvada)

   - Fix up the recently added brcmstb-avs cpufreq driver: fix a bug
     related to system suspend and change the sysfs interface to match
     the user space expectations (Markus Mayer)

   - Modify the runtime PM framework to avoid false-positive warnings
     from the might_sleep_if() assertions in it (Rafael Wysocki)"

* tag 'pm-4.10-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  PM / runtime: Avoid false-positive warnings from might_sleep_if()
  cpufreq: intel_pstate: Disable energy efficiency optimization
  cpufreq: brcmstb-avs-cpufreq: properly retrieve P-state upon suspend
  cpufreq: brcmstb-avs-cpufreq: extend sysfs entry brcm_avs_pmap

Author: torvalds

drivers/base/power/runtime.c

@@ -966,13 +966,13 @@ int __pm_runtime_idle(struct device *dev, int rpmflags)
 	unsigned long flags;
 	int retval;
 
-	might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe);
-
 	if (rpmflags & RPM_GET_PUT) {
 		if (!atomic_dec_and_test(&dev->power.usage_count))
 			return 0;
 	}
 
+	might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe);
+
 	spin_lock_irqsave(&dev->power.lock, flags);
 	retval = rpm_idle(dev, rpmflags);
 	spin_unlock_irqrestore(&dev->power.lock, flags);
@@ -998,13 +998,13 @@ int __pm_runtime_suspend(struct device *dev, int rpmflags)
 	unsigned long flags;
 	int retval;
 
-	might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe);
-
 	if (rpmflags & RPM_GET_PUT) {
 		if (!atomic_dec_and_test(&dev->power.usage_count))
 			return 0;
 	}
 
+	might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe);
+
 	spin_lock_irqsave(&dev->power.lock, flags);
 	retval = rpm_suspend(dev, rpmflags);
 	spin_unlock_irqrestore(&dev->power.lock, flags);
@@ -1029,7 +1029,8 @@ int __pm_runtime_resume(struct device *dev, int rpmflags)
 	unsigned long flags;
 	int retval;
 
-	might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe);
+	might_sleep_if(!(rpmflags & RPM_ASYNC) && !dev->power.irq_safe &&
+			dev->power.runtime_status != RPM_ACTIVE);
 
 	if (rpmflags & RPM_GET_PUT)
 		atomic_inc(&dev->power.usage_count);

drivers/cpufreq/brcmstb-avs-cpufreq.c

@@ -784,8 +784,19 @@ static int brcm_avs_target_index(struct cpufreq_policy *policy,
 static int brcm_avs_suspend(struct cpufreq_policy *policy)
 {
 	struct private_data *priv = policy->driver_data;
+	int ret;
+
+	ret = brcm_avs_get_pmap(priv, &priv->pmap);
+	if (ret)
+		return ret;
 
-	return brcm_avs_get_pmap(priv, &priv->pmap);
+	/*
+	 * We can't use the P-state returned by brcm_avs_get_pmap(), since
+	 * that's the initial P-state from when the P-map was downloaded to the
+	 * AVS co-processor, not necessarily the P-state we are running at now.
+	 * So, we get the current P-state explicitly.
+	 */
+	return brcm_avs_get_pstate(priv, &priv->pmap.state);
 }
 
 static int brcm_avs_resume(struct cpufreq_policy *policy)
@@ -954,9 +965,9 @@ static ssize_t show_brcm_avs_pmap(struct cpufreq_policy *policy, char *buf)
 	brcm_avs_parse_p1(pmap.p1, &mdiv_p0, &pdiv, &ndiv);
 	brcm_avs_parse_p2(pmap.p2, &mdiv_p1, &mdiv_p2, &mdiv_p3, &mdiv_p4);
 
-	return sprintf(buf, "0x%08x 0x%08x %u %u %u %u %u %u %u\n",
+	return sprintf(buf, "0x%08x 0x%08x %u %u %u %u %u %u %u %u %u\n",
 		pmap.p1, pmap.p2, ndiv, pdiv, mdiv_p0, mdiv_p1, mdiv_p2,
-		mdiv_p3, mdiv_p4);
+		mdiv_p3, mdiv_p4, pmap.mode, pmap.state);
 }
 
 static ssize_t show_brcm_avs_voltage(struct cpufreq_policy *policy, char *buf)

drivers/cpufreq/intel_pstate.c

@@ -1235,6 +1235,25 @@ static void intel_pstate_hwp_enable(struct cpudata *cpudata)
 		cpudata->epp_default = intel_pstate_get_epp(cpudata, 0);
 }
 
+#define MSR_IA32_POWER_CTL_BIT_EE	19
+
+/* Disable energy efficiency optimization */
+static void intel_pstate_disable_ee(int cpu)
+{
+	u64 power_ctl;
+	int ret;
+
+	ret = rdmsrl_on_cpu(cpu, MSR_IA32_POWER_CTL, &power_ctl);
+	if (ret)
+		return;
+
+	if (!(power_ctl & BIT(MSR_IA32_POWER_CTL_BIT_EE))) {
+		pr_info("Disabling energy efficiency optimization\n");
+		power_ctl |= BIT(MSR_IA32_POWER_CTL_BIT_EE);
+		wrmsrl_on_cpu(cpu, MSR_IA32_POWER_CTL, power_ctl);
+	}
+}
+
 static int atom_get_min_pstate(void)
 {
 	u64 value;
@@ -1845,6 +1864,11 @@ static const struct x86_cpu_id intel_pstate_cpu_oob_ids[] __initconst = {
 	{}
 };
 
+static const struct x86_cpu_id intel_pstate_cpu_ee_disable_ids[] = {
+	ICPU(INTEL_FAM6_KABYLAKE_DESKTOP, core_params),
+	{}
+};
+
 static int intel_pstate_init_cpu(unsigned int cpunum)
 {
 	struct cpudata *cpu;
@@ -1875,6 +1899,12 @@ static int intel_pstate_init_cpu(unsigned int cpunum)
 	cpu->cpu = cpunum;
 
 	if (hwp_active) {
+		const struct x86_cpu_id *id;
+
+		id = x86_match_cpu(intel_pstate_cpu_ee_disable_ids);
+		if (id)
+			intel_pstate_disable_ee(cpunum);
+
 		intel_pstate_hwp_enable(cpu);
 		pid_params.sample_rate_ms = 50;
 		pid_params.sample_rate_ns = 50 * NSEC_PER_MSEC;