Merge branch 'pm-x86'

* pm-x86:
  x86: tsc: Rework time_cpufreq_notifier()
  admin-guide: pm: intel_epb: Add SPDX license tag and copyright notice
  PM / arch: x86: MSR_IA32_ENERGY_PERF_BIAS sysfs interface
  PM / arch: x86: Rework the MSR_IA32_ENERGY_PERF_BIAS handling

6 files changed, 231 insertions, 68 deletions

diff --git a/arch/x86/kernel/cpu/Makefile b/arch/x86/kernel/cpu/Makefile
index cfd24f9f7614..1796d2bdcaaa 100644
--- a/arch/x86/kernel/cpu/Makefile
+++ b/arch/x86/kernel/cpu/Makefile
@@ -28,7 +28,7 @@ obj-y			+= cpuid-deps.o
 obj-$(CONFIG_PROC_FS)	+= proc.o
 obj-$(CONFIG_X86_FEATURE_NAMES) += capflags.o powerflags.o
 
-obj-$(CONFIG_CPU_SUP_INTEL)		+= intel.o intel_pconfig.o
+obj-$(CONFIG_CPU_SUP_INTEL)		+= intel.o intel_pconfig.o intel_epb.o
 obj-$(CONFIG_CPU_SUP_AMD)		+= amd.o
 obj-$(CONFIG_CPU_SUP_HYGON)		+= hygon.o
 obj-$(CONFIG_CPU_SUP_CYRIX_32)		+= cyrix.o
diff --git a/arch/x86/kernel/cpu/common.c b/arch/x86/kernel/cpu/common.c
index cb28e98a0659..5e37dfa4d9df 100644
--- a/arch/x86/kernel/cpu/common.c
+++ b/arch/x86/kernel/cpu/common.c
@@ -1864,23 +1864,6 @@ void cpu_init(void)
 }
 #endif
 
-static void bsp_resume(void)
-{
-	if (this_cpu->c_bsp_resume)
-		this_cpu->c_bsp_resume(&boot_cpu_data);
-}
-
-static struct syscore_ops cpu_syscore_ops = {
-	.resume		= bsp_resume,
-};
-
-static int __init init_cpu_syscore(void)
-{
-	register_syscore_ops(&cpu_syscore_ops);
-	return 0;
-}
-core_initcall(init_cpu_syscore);
-
 /*
  * The microcode loader calls this upon late microcode load to recheck features,
  * only when microcode has been updated. Caller holds microcode_mutex and CPU
diff --git a/arch/x86/kernel/cpu/cpu.h b/arch/x86/kernel/cpu/cpu.h
index 5eb946b9a9f3..c0e2407abdd6 100644
--- a/arch/x86/kernel/cpu/cpu.h
+++ b/arch/x86/kernel/cpu/cpu.h
@@ -14,7 +14,6 @@ struct cpu_dev {
 	void		(*c_init)(struct cpuinfo_x86 *);
 	void		(*c_identify)(struct cpuinfo_x86 *);
 	void		(*c_detect_tlb)(struct cpuinfo_x86 *);
-	void		(*c_bsp_resume)(struct cpuinfo_x86 *);
 	int		c_x86_vendor;
 #ifdef CONFIG_X86_32
 	/* Optional vendor specific routine to obtain the cache size. */
diff --git a/arch/x86/kernel/cpu/intel.c b/arch/x86/kernel/cpu/intel.c
index 3142fd7a9b32..f17c1a714779 100644
--- a/arch/x86/kernel/cpu/intel.c
+++ b/arch/x86/kernel/cpu/intel.c
@@ -596,36 +596,6 @@ detect_keyid_bits:
 	c->x86_phys_bits -= keyid_bits;
 }
 
-static void init_intel_energy_perf(struct cpuinfo_x86 *c)
-{
-	u64 epb;
-
-	/*
-	 * Initialize MSR_IA32_ENERGY_PERF_BIAS if not already initialized.
-	 * (x86_energy_perf_policy(8) is available to change it at run-time.)
-	 */
-	if (!cpu_has(c, X86_FEATURE_EPB))
-		return;
-
-	rdmsrl(MSR_IA32_ENERGY_PERF_BIAS, epb);
-	if ((epb & 0xF) != ENERGY_PERF_BIAS_PERFORMANCE)
-		return;
-
-	pr_info_once("ENERGY_PERF_BIAS: Set to 'normal', was 'performance'\n");
-	pr_info_once("ENERGY_PERF_BIAS: View and update with x86_energy_perf_policy(8)\n");
-	epb = (epb & ~0xF) | ENERGY_PERF_BIAS_NORMAL;
-	wrmsrl(MSR_IA32_ENERGY_PERF_BIAS, epb);
-}
-
-static void intel_bsp_resume(struct cpuinfo_x86 *c)
-{
-	/*
-	 * MSR_IA32_ENERGY_PERF_BIAS is lost across suspend/resume,
-	 * so reinitialize it properly like during bootup:
-	 */
-	init_intel_energy_perf(c);
-}
-
 static void init_cpuid_fault(struct cpuinfo_x86 *c)
 {
 	u64 msr;
@@ -763,8 +733,6 @@ static void init_intel(struct cpuinfo_x86 *c)
 	if (cpu_has(c, X86_FEATURE_TME))
 		detect_tme(c);
 
-	init_intel_energy_perf(c);
-
 	init_intel_misc_features(c);
 }
 
@@ -1023,9 +991,7 @@ static const struct cpu_dev intel_cpu_dev = {
 	.c_detect_tlb	= intel_detect_tlb,
 	.c_early_init   = early_init_intel,
 	.c_init		= init_intel,
-	.c_bsp_resume	= intel_bsp_resume,
 	.c_x86_vendor	= X86_VENDOR_INTEL,
 };
 
 cpu_dev_register(intel_cpu_dev);
-
diff --git a/arch/x86/kernel/cpu/intel_epb.c b/arch/x86/kernel/cpu/intel_epb.c
new file mode 100644
index 000000000000..f4dd73396f28
--- /dev/null
+++ b/arch/x86/kernel/cpu/intel_epb.c
@@ -0,0 +1,216 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Intel Performance and Energy Bias Hint support.
+ *
+ * Copyright (C) 2019 Intel Corporation
+ *
+ * Author:
+ *	Rafael J. Wysocki <rafael.j.wysocki@intel.com>
+ */
+
+#include <linux/cpuhotplug.h>
+#include <linux/cpu.h>
+#include <linux/device.h>
+#include <linux/kernel.h>
+#include <linux/string.h>
+#include <linux/syscore_ops.h>
+#include <linux/pm.h>
+
+#include <asm/cpufeature.h>
+#include <asm/msr.h>
+
+/**
+ * DOC: overview
+ *
+ * The Performance and Energy Bias Hint (EPB) allows software to specify its
+ * preference with respect to the power-performance tradeoffs present in the
+ * processor.  Generally, the EPB is expected to be set by user space (directly
+ * via sysfs or with the help of the x86_energy_perf_policy tool), but there are
+ * two reasons for the kernel to update it.
+ *
+ * First, there are systems where the platform firmware resets the EPB during
+ * system-wide transitions from sleep states back into the working state
+ * effectively causing the previous EPB updates by user space to be lost.
+ * Thus the kernel needs to save the current EPB values for all CPUs during
+ * system-wide transitions to sleep states and restore them on the way back to
+ * the working state.  That can be achieved by saving EPB for secondary CPUs
+ * when they are taken offline during transitions into system sleep states and
+ * for the boot CPU in a syscore suspend operation, so that it can be restored
+ * for the boot CPU in a syscore resume operation and for the other CPUs when
+ * they are brought back online.  However, CPUs that are already offline when
+ * a system-wide PM transition is started are not taken offline again, but their
+ * EPB values may still be reset by the platform firmware during the transition,
+ * so in fact it is necessary to save the EPB of any CPU taken offline and to
+ * restore it when the given CPU goes back online at all times.
+ *
+ * Second, on many systems the initial EPB value coming from the platform
+ * firmware is 0 ('performance') and at least on some of them that is because
+ * the platform firmware does not initialize EPB at all with the assumption that
+ * the OS will do that anyway.  That sometimes is problematic, as it may cause
+ * the system battery to drain too fast, for example, so it is better to adjust
+ * it on CPU bring-up and if the initial EPB value for a given CPU is 0, the
+ * kernel changes it to 6 ('normal').
+ */
+
+static DEFINE_PER_CPU(u8, saved_epb);
+
+#define EPB_MASK	0x0fULL
+#define EPB_SAVED	0x10ULL
+#define MAX_EPB		EPB_MASK
+
+static int intel_epb_save(void)
+{
+	u64 epb;
+
+	rdmsrl(MSR_IA32_ENERGY_PERF_BIAS, epb);
+	/*
+	 * Ensure that saved_epb will always be nonzero after this write even if
+	 * the EPB value read from the MSR is 0.
+	 */
+	this_cpu_write(saved_epb, (epb & EPB_MASK) | EPB_SAVED);
+
+	return 0;
+}
+
+static void intel_epb_restore(void)
+{
+	u64 val = this_cpu_read(saved_epb);
+	u64 epb;
+
+	rdmsrl(MSR_IA32_ENERGY_PERF_BIAS, epb);
+	if (val) {
+		val &= EPB_MASK;
+	} else {
+		/*
+		 * Because intel_epb_save() has not run for the current CPU yet,
+		 * it is going online for the first time, so if its EPB value is
+		 * 0 ('performance') at this point, assume that it has not been
+		 * initialized by the platform firmware and set it to 6
+		 * ('normal').
+		 */
+		val = epb & EPB_MASK;
+		if (val == ENERGY_PERF_BIAS_PERFORMANCE) {
+			val = ENERGY_PERF_BIAS_NORMAL;
+			pr_warn_once("ENERGY_PERF_BIAS: Set to 'normal', was 'performance'\n");
+		}
+	}
+	wrmsrl(MSR_IA32_ENERGY_PERF_BIAS, (epb & ~EPB_MASK) | val);
+}
+
+static struct syscore_ops intel_epb_syscore_ops = {
+	.suspend = intel_epb_save,
+	.resume = intel_epb_restore,
+};
+
+static const char * const energy_perf_strings[] = {
+	"performance",
+	"balance-performance",
+	"normal",
+	"balance-power",
+	"power"
+};
+static const u8 energ_perf_values[] = {
+	ENERGY_PERF_BIAS_PERFORMANCE,
+	ENERGY_PERF_BIAS_BALANCE_PERFORMANCE,
+	ENERGY_PERF_BIAS_NORMAL,
+	ENERGY_PERF_BIAS_BALANCE_POWERSAVE,
+	ENERGY_PERF_BIAS_POWERSAVE
+};
+
+static ssize_t energy_perf_bias_show(struct device *dev,
+				     struct device_attribute *attr,
+				     char *buf)
+{
+	unsigned int cpu = dev->id;
+	u64 epb;
+	int ret;
+
+	ret = rdmsrl_on_cpu(cpu, MSR_IA32_ENERGY_PERF_BIAS, &epb);
+	if (ret < 0)
+		return ret;
+
+	return sprintf(buf, "%llu\n", epb);
+}
+
+static ssize_t energy_perf_bias_store(struct device *dev,
+				      struct device_attribute *attr,
+				      const char *buf, size_t count)
+{
+	unsigned int cpu = dev->id;
+	u64 epb, val;
+	int ret;
+
+	ret = __sysfs_match_string(energy_perf_strings,
+				   ARRAY_SIZE(energy_perf_strings), buf);
+	if (ret >= 0)
+		val = energ_perf_values[ret];
+	else if (kstrtou64(buf, 0, &val) || val > MAX_EPB)
+		return -EINVAL;
+
+	ret = rdmsrl_on_cpu(cpu, MSR_IA32_ENERGY_PERF_BIAS, &epb);
+	if (ret < 0)
+		return ret;
+
+	ret = wrmsrl_on_cpu(cpu, MSR_IA32_ENERGY_PERF_BIAS,
+			    (epb & ~EPB_MASK) | val);
+	if (ret < 0)
+		return ret;
+
+	return count;
+}
+
+static DEVICE_ATTR_RW(energy_perf_bias);
+
+static struct attribute *intel_epb_attrs[] = {
+	&dev_attr_energy_perf_bias.attr,
+	NULL
+};
+
+static const struct attribute_group intel_epb_attr_group = {
+	.name = power_group_name,
+	.attrs =  intel_epb_attrs
+};
+
+static int intel_epb_online(unsigned int cpu)
+{
+	struct device *cpu_dev = get_cpu_device(cpu);
+
+	intel_epb_restore();
+	if (!cpuhp_tasks_frozen)
+		sysfs_merge_group(&cpu_dev->kobj, &intel_epb_attr_group);
+
+	return 0;
+}
+
+static int intel_epb_offline(unsigned int cpu)
+{
+	struct device *cpu_dev = get_cpu_device(cpu);
+
+	if (!cpuhp_tasks_frozen)
+		sysfs_unmerge_group(&cpu_dev->kobj, &intel_epb_attr_group);
+
+	intel_epb_save();
+	return 0;
+}
+
+static __init int intel_epb_init(void)
+{
+	int ret;
+
+	if (!boot_cpu_has(X86_FEATURE_EPB))
+		return -ENODEV;
+
+	ret = cpuhp_setup_state(CPUHP_AP_X86_INTEL_EPB_ONLINE,
+				"x86/intel/epb:online", intel_epb_online,
+				intel_epb_offline);
+	if (ret < 0)
+		goto err_out_online;
+
+	register_syscore_ops(&intel_epb_syscore_ops);
+	return 0;
+
+err_out_online:
+	cpuhp_remove_state(CPUHP_AP_X86_INTEL_EPB_ONLINE);
+	return ret;
+}
+subsys_initcall(intel_epb_init);
diff --git a/arch/x86/kernel/tsc.c b/arch/x86/kernel/tsc.c
index 3fae23834069..cc6df5c6d7b3 100644
--- a/arch/x86/kernel/tsc.c
+++ b/arch/x86/kernel/tsc.c
@@ -185,8 +185,7 @@ static void __init cyc2ns_init_boot_cpu(void)
 /*
  * Secondary CPUs do not run through tsc_init(), so set up
  * all the scale factors for all CPUs, assuming the same
- * speed as the bootup CPU. (cpufreq notifiers will fix this
- * up if their speed diverges)
+ * speed as the bootup CPU.
  */
 static void __init cyc2ns_init_secondary_cpus(void)
 {
@@ -937,12 +936,12 @@ void tsc_restore_sched_clock_state(void)
 }
 
 #ifdef CONFIG_CPU_FREQ
-/* Frequency scaling support. Adjust the TSC based timer when the cpu frequency
+/*
+ * Frequency scaling support. Adjust the TSC based timer when the CPU frequency
  * changes.
  *
- * RED-PEN: On SMP we assume all CPUs run with the same frequency.  It's
- * not that important because current Opteron setups do not support
- * scaling on SMP anyroads.
+ * NOTE: On SMP the situation is not fixable in general, so simply mark the TSC
+ * as unstable and give up in those cases.
  *
  * Should fix up last_tsc too. Currently gettimeofday in the
  * first tick after the change will be slightly wrong.
@@ -956,22 +955,22 @@ static int time_cpufreq_notifier(struct notifier_block *nb, unsigned long val,
 				void *data)
 {
 	struct cpufreq_freqs *freq = data;
-	unsigned long *lpj;
 
-	lpj = &boot_cpu_data.loops_per_jiffy;
-#ifdef CONFIG_SMP
-	if (!(freq->flags & CPUFREQ_CONST_LOOPS))
-		lpj = &cpu_data(freq->cpu).loops_per_jiffy;
-#endif
+	if (num_online_cpus() > 1) {
+		mark_tsc_unstable("cpufreq changes on SMP");
+		return 0;
+	}
 
 	if (!ref_freq) {
 		ref_freq = freq->old;
-		loops_per_jiffy_ref = *lpj;
+		loops_per_jiffy_ref = boot_cpu_data.loops_per_jiffy;
 		tsc_khz_ref = tsc_khz;
 	}
+
 	if ((val == CPUFREQ_PRECHANGE  && freq->old < freq->new) ||
-			(val == CPUFREQ_POSTCHANGE && freq->old > freq->new)) {
-		*lpj = cpufreq_scale(loops_per_jiffy_ref, ref_freq, freq->new);
+	    (val == CPUFREQ_POSTCHANGE && freq->old > freq->new)) {
+		boot_cpu_data.loops_per_jiffy =
+			cpufreq_scale(loops_per_jiffy_ref, ref_freq, freq->new);
 
 		tsc_khz = cpufreq_scale(tsc_khz_ref, ref_freq, freq->new);
 		if (!(freq->flags & CPUFREQ_CONST_LOOPS))