[PATCH v5] perf/x86/amd/power: Add AMD accumulated power reporting mechanism

[Huang Rui]

Introduce an AMD accumlated power reporting mechanism for the Family
15h, Model 60h processor that can be used to calculate the average
power consumed by a processor during a measurement interval. The
feature support is indicated by CPUID Fn8000_0007_EDX[12].

This feature will be implemented both in hwmon and perf. The current
design provides one event to report per package/processor power
consumption by counting each compute unit power value.

Here the gory details of how the computation is done:

---------------------------------------------------------------------
* Tsample: compute unit power accumulator sample period
* Tref: the PTSC counter period (PTSC: performance timestamp counter)
* N: the ratio of compute unit power accumulator sample period to the
  PTSC period

* Jmax: max compute unit accumulated power which is indicated by
  MSR_C001007b[MaxCpuSwPwrAcc]

* Jx/Jy: compute unit accumulated power which is indicated by
  MSR_C001007a[CpuSwPwrAcc]

* Tx/Ty: the value of performance timestamp counter which is indicated
  by CU_PTSC MSR_C0010280[PTSC]
* PwrCPUave: CPU average power

i. Determine the ratio of Tsample to Tref by executing CPUID Fn8000_0007.
	N = value of CPUID Fn8000_0007_ECX[CpuPwrSampleTimeRatio[15:0]].

ii. Read the full range of the cumulative energy value from the new
    MSR MaxCpuSwPwrAcc.
	Jmax = value returned.

iii. At time x, software reads CpuSwPwrAcc and samples the PTSC.
	Jx = value read from CpuSwPwrAcc and Tx = value read from PTSC.

iv. At time y, software reads CpuSwPwrAcc and samples the PTSC.
	Jy = value read from CpuSwPwrAcc and Ty = value read from PTSC.

v. Calculate the average power consumption for a compute unit over
time period (y-x). Unit of result is uWatt:

	if (Jy < Jx) // Rollover has occurred
		Jdelta = (Jy + Jmax) - Jx
	else
		Jdelta = Jy - Jx
	PwrCPUave = N * Jdelta * 1000 / (Ty - Tx)
----------------------------------------------------------------------

Simple example:

  root@hr-zp:/home/ray/tip# ./tools/perf/perf stat -a -e 'power/power-pkg/' make -j4
    CHK     include/config/kernel.release
    CHK     include/generated/uapi/linux/version.h
    CHK     include/generated/utsrelease.h
    CHK     include/generated/timeconst.h
    CHK     include/generated/bounds.h
    CHK     include/generated/asm-offsets.h
    CALL    scripts/checksyscalls.sh
    CHK     include/generated/compile.h
    SKIPPED include/generated/compile.h
    Building modules, stage 2.
  Kernel: arch/x86/boot/bzImage is ready  (#40)
    MODPOST 4225 modules

   Performance counter stats for 'system wide':

              183.44 mWatts power/power-pkg/

       341.837270111 seconds time elapsed

  root@hr-zp:/home/ray/tip# ./tools/perf/perf stat -a -e 'power/power-pkg/' sleep 10

   Performance counter stats for 'system wide':

                0.18 mWatts power/power-pkg/

        10.012551815 seconds time elapsed

Suggested-by: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
Suggested-by: Ingo Molnar <mingo@xxxxxxxxxx>
Suggested-by: Borislav Petkov <bp@xxxxxxx>
Signed-off-by: Huang Rui <ray.huang@xxxxxxx>
Cc: Guenter Roeck <linux@xxxxxxxxxxxx>
---

Hi,

This series of patches introduces the perf implementation of
accumulated power reporting algorithm. It will calculate the average
power consumption for the processor. The CPU feature flag is
CPUID.8000_0007H:EDX[12].

The V5 is rebased on bp/tip-perf.

Changes from v1 -> v2:
- Add a patch to fix the build issue which is reported by kbuild test
  robot.

Changes from v2 -> v3:
- Use raw_spinlock_t instead of spinlock_t, because it need meet the
  -rt mode use case.
- Use topology_sibling_cpumask to make the cpumask operation easier.

Changes from v3 -> v4:
- Remove active_list, because it is not iterated.
- Capitalize sentences consistently and fix some typos.
- Fix some code style issues.
- Initialize structures in a vertically aligned manner.
- Remove unnecessary comment.
- Fix the runtime bug, and do some testing on CPU-hotplug scenario.

Changes from v4 -> v5:
- Remove "struct pmu" and lock from power_pmu, and rename it to
  power_pmu_masks
- As Peter's suggestion, add a new struct to hw_perf_event, and track
  these values from per-event.

Thanks,
Rui

---
 arch/x86/kernel/cpu/Makefile               |   1 +
 arch/x86/kernel/cpu/perf_event_amd_power.c | 445 +++++++++++++++++++++++++++++
 include/linux/perf_event.h                 |   4 +
 3 files changed, 450 insertions(+)
 create mode 100644 arch/x86/kernel/cpu/perf_event_amd_power.c

diff --git a/arch/x86/kernel/cpu/Makefile b/arch/x86/kernel/cpu/Makefile
index faa7b52..ffc9650 100644
--- a/arch/x86/kernel/cpu/Makefile
+++ b/arch/x86/kernel/cpu/Makefile
@@ -34,6 +34,7 @@ obj-$(CONFIG_PERF_EVENTS)		+= perf_event.o
 
 ifdef CONFIG_PERF_EVENTS
 obj-$(CONFIG_CPU_SUP_AMD)		+= perf_event_amd.o perf_event_amd_uncore.o
+obj-$(CONFIG_CPU_SUP_AMD)		+= perf_event_amd_power.o
 ifdef CONFIG_AMD_IOMMU
 obj-$(CONFIG_CPU_SUP_AMD)		+= perf_event_amd_iommu.o
 endif
diff --git a/arch/x86/kernel/cpu/perf_event_amd_power.c b/arch/x86/kernel/cpu/perf_event_amd_power.c
new file mode 100644
index 0000000..fe2e5e0
--- /dev/null
+++ b/arch/x86/kernel/cpu/perf_event_amd_power.c
@@ -0,0 +1,445 @@
+/*
+ * Performance events - AMD Processor Power Reporting Mechanism
+ *
+ * Copyright (C) 2016 Advanced Micro Devices, Inc.
+ *
+ * Author: Huang Rui <ray.huang@xxxxxxx>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include <linux/module.h>
+#include <linux/slab.h>
+#include <linux/perf_event.h>
+#include <asm/cpu_device_id.h>
+#include "perf_event.h"
+
+#define MSR_F15H_CU_PWR_ACCUMULATOR     0xc001007a
+#define MSR_F15H_CU_MAX_PWR_ACCUMULATOR 0xc001007b
+#define MSR_F15H_PTSC			0xc0010280
+
+/* Event code: LSB 8 bits, passed in attr->config any other bit is reserved. */
+#define AMD_POWER_EVENT_MASK	0xFFULL
+
+#define MAX_CUS	8
+
+/*
+ * Accumulated power status counters.
+ */
+#define AMD_POWER_EVENTSEL_PKG		1
+
+/*
+ * The ratio of compute unit power accumulator sample period to the
+ * PTSC period.
+ */
+static unsigned int cpu_pwr_sample_ratio;
+static unsigned int cores_per_cu;
+static unsigned int cu_num;
+
+/* Maximum accumulated power of a compute unit. */
+static u64 max_cu_acc_power;
+
+struct power_pmu_masks {
+	/*
+	 * These two cpumasks are used for avoiding the allocations on the
+	 * CPU_STARTING phase because power_cpu_prepare() will be called with
+	 * IRQs disabled.
+	 */
+	cpumask_var_t		mask;
+	cpumask_var_t		tmp_mask;
+};
+
+static struct pmu pmu_class;
+
+/*
+ * Accumulated power represents the sum of each compute unit's (CU) power
+ * consumption. On any core of each CU we read the total accumulated power from
+ * MSR_F15H_CU_PWR_ACCUMULATOR. cpu_mask represents CPU bit map of all cores
+ * which are picked to measure the power for the CUs they belong to.
+ */
+static cpumask_t cpu_mask;
+
+static DEFINE_PER_CPU(struct power_pmu_masks *, amd_power_pmu);
+
+static void event_update(struct perf_event *event)
+{
+	struct hw_perf_event *hwc = &event->hw;
+	u64 prev_pwr_acc, new_pwr_acc, prev_ptsc, new_ptsc;
+	u64 delta, tdelta;
+
+	prev_pwr_acc = hwc->pwr_acc;
+	prev_ptsc = hwc->ptsc;
+	rdmsrl(MSR_F15H_CU_PWR_ACCUMULATOR, new_pwr_acc);
+	rdmsrl(MSR_F15H_PTSC, new_ptsc);
+
+	/*
+	 * Calculate the CU power consumption over a time period, the unit of
+	 * final value (delta) is micro-Watts. Then add it to the event count.
+	 */
+	if (new_pwr_acc < prev_pwr_acc) {
+		delta = max_cu_acc_power + new_pwr_acc;
+		delta -= prev_pwr_acc;
+	} else
+		delta = new_pwr_acc - prev_pwr_acc;
+
+	delta *= cpu_pwr_sample_ratio * 1000;
+	tdelta = new_ptsc - prev_ptsc;
+
+	do_div(delta, tdelta);
+	local64_add(delta, &event->count);
+}
+
+static void __pmu_event_start(struct perf_event *event)
+{
+	if (WARN_ON_ONCE(!(event->hw.state & PERF_HES_STOPPED)))
+		return;
+
+	event->hw.state = 0;
+
+	rdmsrl(MSR_F15H_PTSC, event->hw.ptsc);
+	rdmsrl(MSR_F15H_CU_PWR_ACCUMULATOR, event->hw.pwr_acc);
+}
+
+static void pmu_event_start(struct perf_event *event, int mode)
+{
+	__pmu_event_start(event);
+}
+
+static void pmu_event_stop(struct perf_event *event, int mode)
+{
+	struct hw_perf_event *hwc = &event->hw;
+
+	/* Mark event as deactivated and stopped. */
+	if (!(hwc->state & PERF_HES_STOPPED))
+		hwc->state |= PERF_HES_STOPPED;
+
+	/* Check if software counter update is necessary. */
+	if ((mode & PERF_EF_UPDATE) && !(hwc->state & PERF_HES_UPTODATE)) {
+		/*
+		 * Drain the remaining delta count out of an event
+		 * that we are disabling:
+		 */
+		event_update(event);
+		hwc->state |= PERF_HES_UPTODATE;
+	}
+}
+
+static int pmu_event_add(struct perf_event *event, int mode)
+{
+	struct hw_perf_event *hwc = &event->hw;
+
+	hwc->state = PERF_HES_UPTODATE | PERF_HES_STOPPED;
+
+	if (mode & PERF_EF_START)
+		__pmu_event_start(event);
+
+	return 0;
+}
+
+static void pmu_event_del(struct perf_event *event, int flags)
+{
+	pmu_event_stop(event, PERF_EF_UPDATE);
+}
+
+static int pmu_event_init(struct perf_event *event)
+{
+	u64 cfg = event->attr.config & AMD_POWER_EVENT_MASK;
+
+	/* Only look at AMD power events. */
+	if (event->attr.type != pmu_class.type)
+		return -ENOENT;
+
+	/* Unsupported modes and filters. */
+	if (event->attr.exclude_user   ||
+	    event->attr.exclude_kernel ||
+	    event->attr.exclude_hv     ||
+	    event->attr.exclude_idle   ||
+	    event->attr.exclude_host   ||
+	    event->attr.exclude_guest  ||
+	    /* no sampling */
+	    event->attr.sample_period)
+		return -EINVAL;
+
+	if (cfg != AMD_POWER_EVENTSEL_PKG)
+		return -EINVAL;
+
+	return 0;
+}
+
+static void pmu_event_read(struct perf_event *event)
+{
+	event_update(event);
+}
+
+static ssize_t
+get_attr_cpumask(struct device *dev, struct device_attribute *attr, char *buf)
+{
+	return cpumap_print_to_pagebuf(true, buf, &cpu_mask);
+}
+
+static DEVICE_ATTR(cpumask, S_IRUGO, get_attr_cpumask, NULL);
+
+static struct attribute *pmu_attrs[] = {
+	&dev_attr_cpumask.attr,
+	NULL,
+};
+
+static struct attribute_group pmu_attr_group = {
+	.attrs = pmu_attrs,
+};
+
+/*
+ * Currently it only supports to report the power of each
+ * processor/package.
+ */
+EVENT_ATTR_STR(power-pkg, power_pkg, "event=0x01");
+
+EVENT_ATTR_STR(power-pkg.unit, power_pkg_unit, "mWatts");
+
+/* Convert the count from micro-Watts to milli-Watts. */
+EVENT_ATTR_STR(power-pkg.scale, power_pkg_scale, "1.000000e-3");
+
+static struct attribute *events_attr[] = {
+	EVENT_PTR(power_pkg),
+	EVENT_PTR(power_pkg_unit),
+	EVENT_PTR(power_pkg_scale),
+	NULL,
+};
+
+static struct attribute_group pmu_events_group = {
+	.name	= "events",
+	.attrs	= events_attr,
+};
+
+PMU_FORMAT_ATTR(event, "config:0-7");
+
+static struct attribute *formats_attr[] = {
+	&format_attr_event.attr,
+	NULL,
+};
+
+static struct attribute_group pmu_format_group = {
+	.name	= "format",
+	.attrs	= formats_attr,
+};
+
+static const struct attribute_group *attr_groups[] = {
+	&pmu_attr_group,
+	&pmu_format_group,
+	&pmu_events_group,
+	NULL,
+};
+
+static struct pmu pmu_class = {
+	.attr_groups	= attr_groups,
+	/* system-wide only */
+	.task_ctx_nr	= perf_invalid_context,
+	.event_init	= pmu_event_init,
+	.add		= pmu_event_add,
+	.del		= pmu_event_del,
+	.start		= pmu_event_start,
+	.stop		= pmu_event_stop,
+	.read		= pmu_event_read,
+};
+
+static int power_cpu_exit(int cpu)
+{
+	struct power_pmu_masks *pmu = per_cpu(amd_power_pmu, cpu);
+	int target = nr_cpumask_bits;
+	int ret = 0;
+
+	cpumask_copy(pmu->mask, topology_sibling_cpumask(cpu));
+
+	cpumask_clear_cpu(cpu, &cpu_mask);
+	cpumask_clear_cpu(cpu, pmu->mask);
+
+	if (!cpumask_and(pmu->tmp_mask, pmu->mask, cpu_online_mask))
+		goto out;
+
+	/*
+	 * Find a new CPU on the same compute unit, if was set in cpumask
+	 * and still some CPUs on compute unit. Then move on to the new CPU.
+	 */
+	target = cpumask_any(pmu->tmp_mask);
+	if (target < nr_cpumask_bits && target != cpu)
+		cpumask_set_cpu(target, &cpu_mask);
+
+	WARN_ON(cpumask_empty(&cpu_mask));
+
+out:
+	/*
+	 * Migrate event and context to new CPU.
+	 */
+	if (target < nr_cpumask_bits)
+		perf_pmu_migrate_context(&pmu_class, cpu, target);
+
+	return ret;
+
+}
+
+static int power_cpu_init(int cpu)
+{
+	struct power_pmu_masks *pmu = per_cpu(amd_power_pmu, cpu);
+
+	if (!pmu)
+		return 0;
+
+	if (!cpumask_and(pmu->mask, topology_sibling_cpumask(cpu), &cpu_mask))
+		cpumask_set_cpu(cpu, &cpu_mask);
+
+	return 0;
+}
+
+static int power_cpu_prepare(int cpu)
+{
+	struct power_pmu_masks *pmu = per_cpu(amd_power_pmu, cpu);
+	int phys_id = topology_physical_package_id(cpu);
+	int ret = 0;
+
+	if (pmu)
+		return 0;
+
+	if (phys_id < 0)
+		return -EINVAL;
+
+	pmu = kzalloc_node(sizeof(*pmu), GFP_KERNEL, cpu_to_node(cpu));
+	if (!pmu)
+		return -ENOMEM;
+
+	if (!zalloc_cpumask_var(&pmu->mask, GFP_KERNEL)) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	if (!zalloc_cpumask_var(&pmu->tmp_mask, GFP_KERNEL)) {
+		ret = -ENOMEM;
+		goto out1;
+	}
+
+	per_cpu(amd_power_pmu, cpu) = pmu;
+
+	return 0;
+
+out1:
+	free_cpumask_var(pmu->mask);
+out:
+	kfree(pmu);
+
+	return ret;
+}
+
+static void power_cpu_kfree(int cpu)
+{
+	struct power_pmu_masks *pmu = per_cpu(amd_power_pmu, cpu);
+
+	if (!pmu)
+		return;
+
+	free_cpumask_var(pmu->mask);
+	free_cpumask_var(pmu->tmp_mask);
+	kfree(pmu);
+
+	per_cpu(amd_power_pmu, cpu) = NULL;
+}
+
+static int
+power_cpu_notifier(struct notifier_block *self, unsigned long action, void *hcpu)
+{
+	unsigned int cpu = (long)hcpu;
+
+	switch (action & ~CPU_TASKS_FROZEN) {
+	case CPU_UP_PREPARE:
+		if (power_cpu_prepare(cpu))
+			return NOTIFY_BAD;
+		break;
+	case CPU_STARTING:
+		if (power_cpu_init(cpu))
+			return NOTIFY_BAD;
+		break;
+	case CPU_DEAD:
+		power_cpu_kfree(cpu);
+		break;
+	case CPU_DOWN_PREPARE:
+		if (power_cpu_exit(cpu))
+			return NOTIFY_BAD;
+		break;
+	default:
+		break;
+	}
+
+	return NOTIFY_OK;
+}
+
+static const struct x86_cpu_id cpu_match[] = {
+	{ .vendor = X86_VENDOR_AMD, .family = 0x15 },
+	{},
+};
+
+static int __init amd_power_pmu_init(void)
+{
+	int i, ret;
+	u64 tmp;
+
+	if (!x86_match_cpu(cpu_match))
+		return 0;
+
+	if (!boot_cpu_has(X86_FEATURE_ACC_POWER))
+		return -ENODEV;
+
+	cores_per_cu = amd_get_cores_per_cu();
+	cu_num = boot_cpu_data.x86_max_cores / cores_per_cu;
+
+	if (WARN_ON_ONCE(cu_num > MAX_CUS))
+		return -EINVAL;
+
+	cpu_pwr_sample_ratio = cpuid_ecx(0x80000007);
+
+	if (rdmsrl_safe(MSR_F15H_CU_MAX_PWR_ACCUMULATOR, &tmp)) {
+		pr_err("Failed to read max compute unit power accumulator MSR\n");
+		return -ENODEV;
+	}
+	max_cu_acc_power = tmp;
+
+	cpu_notifier_register_begin();
+
+	/* Choose one online core of each compute unit.  */
+	for (i = 0; i < boot_cpu_data.x86_max_cores; i += cores_per_cu) {
+		WARN_ON(cpumask_empty(topology_sibling_cpumask(i)));
+		cpumask_set_cpu(cpumask_any(topology_sibling_cpumask(i)), &cpu_mask);
+	}
+
+	for_each_present_cpu(i) {
+		ret = power_cpu_prepare(i);
+		if (ret) {
+			/* Unwind on [0 ... i-1] CPUs. */
+			while (i--)
+				power_cpu_kfree(i);
+			goto out;
+		}
+		ret = power_cpu_init(i);
+		if (ret) {
+			/* Unwind on [0 ... i] CPUs. */
+			while (i >= 0)
+				power_cpu_kfree(i--);
+			goto out;
+		}
+	}
+
+	__perf_cpu_notifier(power_cpu_notifier);
+
+	ret = perf_pmu_register(&pmu_class, "power", -1);
+	if (WARN_ON(ret)) {
+		pr_warn("AMD Power PMU registration failed\n");
+		goto out;
+	}
+
+	pr_info("AMD Power PMU detected, %d compute units\n", cu_num);
+
+out:
+	cpu_notifier_register_done();
+
+	return ret;
+}
+device_initcall(amd_power_pmu_init);
diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index f9828a4..01ea21c 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -128,6 +128,10 @@ struct hw_perf_event {
 		struct { /* itrace */
 			int			itrace_started;
 		};
+		struct { /* amd_power */
+			u64	pwr_acc;
+			u64	ptsc;
+		};
 #ifdef CONFIG_HAVE_HW_BREAKPOINT
 		struct { /* breakpoint */
 			/*
-- 
1.9.1