[PATCH v6 0/2] perf/x86/power: Introduce AMD accumlated power reporting mechanism

[PATCH v6 0/2] perf/x86/power: Introduce AMD accumlated power reporting mechanism

[Huang Rui]

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 V6 is rebased on bp/tip-perf. And need two dependent patches at
tip because of modular perf driver:
http://git.kernel.org/tip/675965b00d734c985e4285f5bec7e524d15fc4e1
http://git.kernel.org/tip/3712bba1a260ad851f3aa8ddea9cb7326f6aa0b3


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.

Changes from v5 -> v6:
- Remove MAX_CUS check
- Remove DEFINE_PER_CPU(struct power_pmu_masks *, amd_power_pmu)
- Remove power_cpu_prepare power_cpu_free and refine power_cpu_init
  and power_cpu_exit
- Use smp_num_siblings instead of cores_per_cu variable.
- Make this driver as a module.
- Add a patch to export events_sysfs_show.

Thanks,
Rui

Huang Rui (2):
  perf/x86: Export events_sysfs_show()
  perf/x86/amd/power: Add AMD accumulated power reporting mechanism

 arch/x86/Kconfig                           |   9 +
 arch/x86/kernel/cpu/Makefile               |   1 +
 arch/x86/kernel/cpu/perf_event.c           |   1 +
 arch/x86/kernel/cpu/perf_event_amd_power.c | 360 +++++++++++++++++++++++++++++
 include/linux/perf_event.h                 |   4 +
 5 files changed, 375 insertions(+)
 create mode 100644 arch/x86/kernel/cpu/perf_event_amd_power.c

-- 
1.9.1

[PATCH v6 1/2] perf/x86: Export events_sysfs_show()

[Huang Rui]

This interface will be used in modular perf drivers.

Signed-off-by: Huang Rui <ray.huang@amd.com>
---
 arch/x86/kernel/cpu/perf_event.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/arch/x86/kernel/cpu/perf_event.c b/arch/x86/kernel/cpu/perf_event.c
index 1b443db..47f673c 100644
--- a/arch/x86/kernel/cpu/perf_event.c
+++ b/arch/x86/kernel/cpu/perf_event.c
@@ -1601,6 +1601,7 @@ ssize_t events_sysfs_show(struct device *dev, struct device_attribute *attr,
 
 	return x86_pmu.events_sysfs_show(page, config);
 }
+EXPORT_SYMBOL_GPL(events_sysfs_show);
 
 EVENT_ATTR(cpu-cycles,			CPU_CYCLES		);
 EVENT_ATTR(instructions,		INSTRUCTIONS		);
-- 
1.9.1

[PATCH v6 2/2] 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@infradead.org>
Suggested-by: Ingo Molnar <mingo@kernel.org>
Suggested-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Huang Rui <ray.huang@amd.com>
Cc: Guenter Roeck <linux@roeck-us.net>
---

Hi Thomas,

Thanks to suggest the updates for power_cpu_init(), but there are some
minor issues with below codes.


target = cpumask_any_and(&cpu_mask, topology_sibling_cpumask(cpu));
if (target < nr_cpu_ids)
	cpumask_set_cpu(cpu, &cpu_mask);


For example: cores number is 4, and compute unit number is 2 (2 cores
per compute units). 

When amd_power_pmu completes initialization, the cpumask will be set
as 1111. Actually, I just would like to choose one core per compute
unit. So the expected value should be 0101.

when core 2 and core 3 are both online (they are in same compute
unit) and cpu_mask is 1111,
set core 2 offline, then 1111 -> 1011,
set core 2 online again, then 1011 -> 1111.
Actually, we don't expect two cores both set at cpu_mask in the same
compute unit.

So as inspired by you, I do below change for power_cpu_init and tested
in my platform, please see my codes. :-)

Thanks,
Rui

---
 arch/x86/Kconfig                           |   9 +
 arch/x86/kernel/cpu/Makefile               |   1 +
 arch/x86/kernel/cpu/perf_event_amd_power.c | 360 +++++++++++++++++++++++++++++
 include/linux/perf_event.h                 |   4 +
 4 files changed, 374 insertions(+)
 create mode 100644 arch/x86/kernel/cpu/perf_event_amd_power.c

diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index 330e738..a0d3fb0 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -1202,6 +1202,15 @@ config MICROCODE_OLD_INTERFACE
 	def_bool y
 	depends on MICROCODE
 
+config PERF_EVENTS_AMD_POWER
+	depends on PERF_EVENTS && CPU_SUP_AMD
+	tristate "AMD Processor Power Reporting Mechanism"
+	---help---
+	  Provide power reporting mechanism support for AMD processors.
+	  Currently, it leverages X86_FEATURE_ACC_POWER
+	  (CPUID Fn8000_0007_EDX[12]) interface to calculate the
+	  average power consumption on Family 15h processors.
+
 config X86_MSR
 	tristate "/dev/cpu/*/msr - Model-specific register support"
 	---help---
diff --git a/arch/x86/kernel/cpu/Makefile b/arch/x86/kernel/cpu/Makefile
index faa7b52..95d8419 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_PERF_EVENTS_AMD_POWER)	+= 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..42d5072
--- /dev/null
+++ b/arch/x86/kernel/cpu/perf_event_amd_power.c
@@ -0,0 +1,360 @@
+/*
+ * Performance events - AMD Processor Power Reporting Mechanism
+ *
+ * Copyright (C) 2016 Advanced Micro Devices, Inc.
+ *
+ * Author: Huang Rui <ray.huang@amd.com>
+ *
+ * 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
+
+/*
+ * 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 cu_num;
+
+/* Maximum accumulated power of a compute unit. */
+static u64 max_cu_acc_power;
+
+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 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 void power_cpu_exit(int cpu)
+{
+	int target = nr_cpumask_bits;
+
+	if (!cpumask_test_and_clear_cpu(cpu, &cpu_mask))
+		return;
+
+	/*
+	 * Find a new CPU on the same compute unit, if was set in cpumask
+	 * and still some CPUs on compute unit. Then migrate event and
+	 * context to new CPU.
+	 */
+	target = cpumask_any_but(topology_sibling_cpumask(cpu), cpu);
+	if (target < nr_cpumask_bits) {
+		cpumask_set_cpu(target, &cpu_mask);
+		perf_pmu_migrate_context(&pmu_class, cpu, target);
+	}
+}
+
+static void power_cpu_init(int cpu)
+{
+	/*
+	 * 1) If any CPU is set at cpu_mask in the same compute unit, do
+	 * nothing.
+	 * 2) If no CPU is set at cpu_mask in the same compute unit,
+	 * set current STARTING CPU.
+	 *
+	 * If cpu_mask and topology_sibling_cpumask has intersected
+	 * bits, that means any CPU is set in the same compute unit.
+	 * But cpumask_weight(topology_sibling_cpumask(cpu)) == 1
+	 * means no CPU is set on cpu_mask in the same compute unit
+	 * before init current STARTING CPU.
+	 */
+	if (!cpumask_intersects(&cpu_mask, topology_sibling_cpumask(cpu)) &&
+	    cpumask_weight(topology_sibling_cpumask(cpu)) == 1)
+			cpumask_set_cpu(cpu, &cpu_mask);
+}
+
+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_STARTING:
+		power_cpu_init(cpu);
+		break;
+	case CPU_DOWN_PREPARE:
+		power_cpu_exit(cpu);
+		break;
+	default:
+		break;
+	}
+
+	return NOTIFY_OK;
+}
+
+static struct notifier_block power_cpu_notifier_nb = {
+	.notifier_call = power_cpu_notifier,
+	.priority = CPU_PRI_PERF,
+};
+
+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;
+
+	cu_num = boot_cpu_data.x86_max_cores / smp_num_siblings;
+
+	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 += smp_num_siblings) {
+		WARN_ON(cpumask_empty(topology_sibling_cpumask(i)));
+		cpumask_set_cpu(cpumask_any(topology_sibling_cpumask(i)), &cpu_mask);
+	}
+
+	for_each_online_cpu(i)
+		power_cpu_init(i);
+
+	__register_cpu_notifier(&power_cpu_notifier_nb);
+
+	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;
+}
+module_init(amd_power_pmu_init);
+
+static void __exit amd_power_pmu_exit(void)
+{
+	cpu_notifier_register_begin();
+	__unregister_cpu_notifier(&power_cpu_notifier_nb);
+	cpu_notifier_register_done();
+
+	perf_pmu_unregister(&pmu_class);
+}
+module_exit(amd_power_pmu_exit);
+
+MODULE_AUTHOR("Huang Rui <ray.huang@amd.com>");
+MODULE_DESCRIPTION("AMD Processor Power Reporting Mechanism");
+MODULE_LICENSE("GPL v2");
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

Re: [PATCH v6 2/2] perf/x86/amd/power: Add AMD accumulated power reporting mechanism

[Thomas Gleixner]

On Thu, 3 Mar 2016, Huang Rui wrote:
> +/*
> + * The ratio of compute unit power accumulator sample period to the
> + * PTSC period.
> + */
> +static unsigned int cpu_pwr_sample_ratio;
> +static unsigned int cu_num;

Why do you need static storage for that information when the only purpose is
to printk it in init?

> +static void power_cpu_exit(int cpu)
> +{
> +	int target = nr_cpumask_bits;

What's that initialization for?

> +
> +	if (!cpumask_test_and_clear_cpu(cpu, &cpu_mask))
> +		return;
> +
> +	/*
> +	 * Find a new CPU on the same compute unit, if was set in cpumask
> +	 * and still some CPUs on compute unit. Then migrate event and
> +	 * context to new CPU.
> +	 */
> +	target = cpumask_any_but(topology_sibling_cpumask(cpu), cpu);
> +	if (target < nr_cpumask_bits) {
> +		cpumask_set_cpu(target, &cpu_mask);
> +		perf_pmu_migrate_context(&pmu_class, cpu, target);
> +	}
> +}
> +
> +static void power_cpu_init(int cpu)
> +{
> +	/*
> +	 * 1) If any CPU is set at cpu_mask in the same compute unit, do
> +	 * nothing.
> +	 * 2) If no CPU is set at cpu_mask in the same compute unit,
> +	 * set current STARTING CPU.
> +	 *
> +	 * If cpu_mask and topology_sibling_cpumask has intersected
> +	 * bits, that means any CPU is set in the same compute unit.
> +	 * But cpumask_weight(topology_sibling_cpumask(cpu)) == 1
> +	 * means no CPU is set on cpu_mask in the same compute unit
> +	 * before init current STARTING CPU.
> +	 */
> +	if (!cpumask_intersects(&cpu_mask, topology_sibling_cpumask(cpu)) &&
> +	    cpumask_weight(topology_sibling_cpumask(cpu)) == 1)
> +			cpumask_set_cpu(cpu, &cpu_mask);

I don't think you need that complexity.

	target = cpumask_any_but(topology_sibling_cpumask(cpu), cpu);
	if (target >= nr_cpumask_bits)
		cpumask_set_cpu(cpu, &cpu_mask);
	   
Simply because if there is a cpu aside of the new one already in the sibling
mask, then it is also in cpu_mask. Hmm?

> +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_STARTING:
> +		power_cpu_init(cpu);
> +		break;
> +	case CPU_DOWN_PREPARE:
> +		power_cpu_exit(cpu);
> +		break;

And of course if CPU_DOWN_PREPARE fails and this is the last cpu in the
compute unit, nothing takes over the duty for this compute unit. So you need
to handle CPU_DOWN_FAILED ....

> +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;
> +
> +	cu_num = boot_cpu_data.x86_max_cores / smp_num_siblings;
> +
> +	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;

Why do you need an intermediate 'tmp' for this?

> +	cpu_notifier_register_begin();
> +
> +	/* Choose one online core of each compute unit. */
> +	for (i = 0; i < boot_cpu_data.x86_max_cores; i += smp_num_siblings) {
> +		WARN_ON(cpumask_empty(topology_sibling_cpumask(i)));

Err. What guarantees that in each compute unit is one sibling online? And what
value has that WARN_ON? We don't care about the stack trace here, because it's
known already.

> +		cpumask_set_cpu(cpumask_any(topology_sibling_cpumask(i)), &cpu_mask);

Of course you just continue in that case and end up with:

       	      cpumask_set_cpu(nr_cpu_ids, &cpu_mask);

i.e. you try to do that on an invalid bit, which will trigger a justified
warning in cpumask_set_cpu() if CONFIG_DEBUG_PER_CPU_MAPS is enabled.

Aside of that this only handles a single socket. And why do you do the above
if you handle the same thing in the loop below?

> +	}
> +
> +	for_each_online_cpu(i)
> +		power_cpu_init(i);
> +
> +	__register_cpu_notifier(&power_cpu_notifier_nb);
> +
> +	ret = perf_pmu_register(&pmu_class, "power", -1);
> +	if (WARN_ON(ret)) {
> +		pr_warn("AMD Power PMU registration failed\n");

This still leaks the cpu notifier. .....

> +		goto out;
> +	}
> +
> +	pr_info("AMD Power PMU detected, %d compute units\n", cu_num);

Why is the number of compute units interesting at all?

Thanks,

	tglx

Re: [PATCH v6 2/2] perf/x86/amd/power: Add AMD accumulated power reporting mechanism

[Huang Rui]

On Thu, Mar 03, 2016 at 09:50:11AM +0100, Thomas Gleixner wrote:
> On Thu, 3 Mar 2016, Huang Rui wrote:
> > +
> > +static void power_cpu_init(int cpu)
> > +{
> > +	/*
> > +	 * 1) If any CPU is set at cpu_mask in the same compute unit, do
> > +	 * nothing.
> > +	 * 2) If no CPU is set at cpu_mask in the same compute unit,
> > +	 * set current STARTING CPU.
> > +	 *
> > +	 * If cpu_mask and topology_sibling_cpumask has intersected
> > +	 * bits, that means any CPU is set in the same compute unit.
> > +	 * But cpumask_weight(topology_sibling_cpumask(cpu)) == 1
> > +	 * means no CPU is set on cpu_mask in the same compute unit
> > +	 * before init current STARTING CPU.
> > +	 */
> > +	if (!cpumask_intersects(&cpu_mask, topology_sibling_cpumask(cpu)) &&
> > +	    cpumask_weight(topology_sibling_cpumask(cpu)) == 1)
> > +			cpumask_set_cpu(cpu, &cpu_mask);
> 
> I don't think you need that complexity.
> 
> 	target = cpumask_any_but(topology_sibling_cpumask(cpu), cpu);
> 	if (target >= nr_cpumask_bits)
> 		cpumask_set_cpu(cpu, &cpu_mask);
> 	   
> Simply because if there is a cpu aside of the new one already in the sibling
> mask, then it is also in cpu_mask. Hmm?
> 

Make sense. Thanks. Will update.

> > +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_STARTING:
> > +		power_cpu_init(cpu);
> > +		break;
> > +	case CPU_DOWN_PREPARE:
> > +		power_cpu_exit(cpu);
> > +		break;
> 
> And of course if CPU_DOWN_PREPARE fails and this is the last cpu in the
> compute unit, nothing takes over the duty for this compute unit. So you need
> to handle CPU_DOWN_FAILED ....
> 

OK, so I need to do power_cpu_init when notified CPU_DOWN_FAILED, am I
right?

> > +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;
> > +
> > +	cu_num = boot_cpu_data.x86_max_cores / smp_num_siblings;
> > +
> > +	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;
> 
> Why do you need an intermediate 'tmp' for this?
> 

Will use max_cu_acc_power directly.

> > +	cpu_notifier_register_begin();
> > +
> > +	/* Choose one online core of each compute unit. */
> > +	for (i = 0; i < boot_cpu_data.x86_max_cores; i += smp_num_siblings) {
> > +		WARN_ON(cpumask_empty(topology_sibling_cpumask(i)));
> 
> Err. What guarantees that in each compute unit is one sibling online? And what
> value has that WARN_ON? We don't care about the stack trace here, because it's
> known already.
> 

When this driver is not as module before, I think there should be one
sibling online at least at initialization phase. But now, you're
right, we cannot guarantee it.

> > +		cpumask_set_cpu(cpumask_any(topology_sibling_cpumask(i)), &cpu_mask);
> 
> Of course you just continue in that case and end up with:
> 
>        	      cpumask_set_cpu(nr_cpu_ids, &cpu_mask);
> 
> i.e. you try to do that on an invalid bit, which will trigger a justified
> warning in cpumask_set_cpu() if CONFIG_DEBUG_PER_CPU_MAPS is enabled.
> 
> Aside of that this only handles a single socket. And why do you do the above
> if you handle the same thing in the loop below?
> 

Because the sibling online shouldn't be empty at initialization phase
if the driver is not module before. So...

Thanks to catch it.

How about below update:

for (i = 0; i < boot_cpu_data.x86_max_cores; i += smp_num_siblings) {
        if (!cpumask_empty(topology_sibling_cpumask(i)))
                cpumask_set_cpu(cpumask_any(topology_sibling_cpumask(i)), &cpu_mask);
}

> > +	}
> > +
> > +	for_each_online_cpu(i)
> > +		power_cpu_init(i);
> > +
> > +	__register_cpu_notifier(&power_cpu_notifier_nb);
> > +
> > +	ret = perf_pmu_register(&pmu_class, "power", -1);
> > +	if (WARN_ON(ret)) {
> > +		pr_warn("AMD Power PMU registration failed\n");
> 
> This still leaks the cpu notifier. .....
> 

OK, so I should do __unregister_cpu_notifier(&power_cpu_notifier_nb)
here.

> > +		goto out;
> > +	}
> > +
> > +	pr_info("AMD Power PMU detected, %d compute units\n", cu_num);
> 
> Why is the number of compute units interesting at all?
> 

Because the accumulated power bases on compute units.
We can see the mask from /sys/devices/power/cpumask and number of
compute units to know if all compute units are set at cpumask.
So I add a printk here, does it make sense?

Thanks,
Rui

Re: [PATCH v6 2/2] perf/x86/amd/power: Add AMD accumulated power reporting mechanism

[Thomas Gleixner]

On Thu, 3 Mar 2016, Huang Rui wrote:
> On Thu, Mar 03, 2016 at 09:50:11AM +0100, Thomas Gleixner wrote:
> > On Thu, 3 Mar 2016, Huang Rui wrote:
> > And of course if CPU_DOWN_PREPARE fails and this is the last cpu in the
> > compute unit, nothing takes over the duty for this compute unit. So you need
> > to handle CPU_DOWN_FAILED ....
> > 
> 
> OK, so I need to do power_cpu_init when notified CPU_DOWN_FAILED, am I
> right?

Yes.
 
> > > +	cpu_notifier_register_begin();
> > > +
> > > +	/* Choose one online core of each compute unit. */
> > > +	for (i = 0; i < boot_cpu_data.x86_max_cores; i += smp_num_siblings) {
> > > +		WARN_ON(cpumask_empty(topology_sibling_cpumask(i)));
> > 
> > Err. What guarantees that in each compute unit is one sibling online? And what
> > value has that WARN_ON? We don't care about the stack trace here, because it's
> > known already.
> > 
> 
> When this driver is not as module before, I think there should be one
> sibling online at least at initialization phase. But now, you're
> right, we cannot guarantee it.
> 
> > > +		cpumask_set_cpu(cpumask_any(topology_sibling_cpumask(i)), &cpu_mask);
> > 
> > Of course you just continue in that case and end up with:
> > 
> >        	      cpumask_set_cpu(nr_cpu_ids, &cpu_mask);
> > 
> > i.e. you try to do that on an invalid bit, which will trigger a justified
> > warning in cpumask_set_cpu() if CONFIG_DEBUG_PER_CPU_MAPS is enabled.
> > 
> > Aside of that this only handles a single socket. And why do you do the above
> > if you handle the same thing in the loop below?
> > 
> 
> Because the sibling online shouldn't be empty at initialization phase
> if the driver is not module before. So...
> 
> Thanks to catch it.
> 
> How about below update:
> 
> for (i = 0; i < boot_cpu_data.x86_max_cores; i += smp_num_siblings) {
>         if (!cpumask_empty(topology_sibling_cpumask(i)))
>                 cpumask_set_cpu(cpumask_any(topology_sibling_cpumask(i)), &cpu_mask);
> }

Why? You do a full for_each_online_cpu(i) loop after that, which does
exactly the same thing, right?
 
> > > +	}
> > > +
> > > +	for_each_online_cpu(i)
> > > +		power_cpu_init(i);
> > > +
> > > +	__register_cpu_notifier(&power_cpu_notifier_nb);
> > > +
> > > +	ret = perf_pmu_register(&pmu_class, "power", -1);
> > > +	if (WARN_ON(ret)) {
> > > +		pr_warn("AMD Power PMU registration failed\n");
> > 
> > This still leaks the cpu notifier. .....
> > 
> 
> OK, so I should do __unregister_cpu_notifier(&power_cpu_notifier_nb)
> here.

You can register the notifier after perf_pmu_register succeeded, right?
 
> > > +	pr_info("AMD Power PMU detected, %d compute units\n", cu_num);
> > 
> > Why is the number of compute units interesting at all?
> > 
> 
> Because the accumulated power bases on compute units.
> We can see the mask from /sys/devices/power/cpumask and number of
> compute units to know if all compute units are set at cpumask.
> So I add a printk here, does it make sense?

No, because it's completely non intuitive. How on earth am I supposed to get
the connection between /sys/devices/power/cpumask and number of compute units
without staring at the code? So this is only interesting for a developer who
can deduce that number from /proc/cpuinfo or dmesg as well.

Thanks,

	tglx

Re: [PATCH v6 2/2] perf/x86/amd/power: Add AMD accumulated power reporting mechanism

[Huang Rui]

On Thu, Mar 03, 2016 at 04:26:46PM +0100, Thomas Gleixner wrote:
> On Thu, 3 Mar 2016, Huang Rui wrote:
> > On Thu, Mar 03, 2016 at 09:50:11AM +0100, Thomas Gleixner wrote:
> > > On Thu, 3 Mar 2016, Huang Rui wrote:
> > > And of course if CPU_DOWN_PREPARE fails and this is the last cpu in the
> > > compute unit, nothing takes over the duty for this compute unit. So you need
> > > to handle CPU_DOWN_FAILED ....
> > > 
> > 
> > OK, so I need to do power_cpu_init when notified CPU_DOWN_FAILED, am I
> > right?
> 
> Yes.
>  
> > > > +	cpu_notifier_register_begin();
> > > > +
> > > > +	/* Choose one online core of each compute unit. */
> > > > +	for (i = 0; i < boot_cpu_data.x86_max_cores; i += smp_num_siblings) {
> > > > +		WARN_ON(cpumask_empty(topology_sibling_cpumask(i)));
> > > 
> > > Err. What guarantees that in each compute unit is one sibling online? And what
> > > value has that WARN_ON? We don't care about the stack trace here, because it's
> > > known already.
> > > 
> > 
> > When this driver is not as module before, I think there should be one
> > sibling online at least at initialization phase. But now, you're
> > right, we cannot guarantee it.
> > 
> > > > +		cpumask_set_cpu(cpumask_any(topology_sibling_cpumask(i)), &cpu_mask);
> > > 
> > > Of course you just continue in that case and end up with:
> > > 
> > >        	      cpumask_set_cpu(nr_cpu_ids, &cpu_mask);
> > > 
> > > i.e. you try to do that on an invalid bit, which will trigger a justified
> > > warning in cpumask_set_cpu() if CONFIG_DEBUG_PER_CPU_MAPS is enabled.
> > > 
> > > Aside of that this only handles a single socket. And why do you do the above
> > > if you handle the same thing in the loop below?
> > > 
> > 
> > Because the sibling online shouldn't be empty at initialization phase
> > if the driver is not module before. So...
> > 
> > Thanks to catch it.
> > 
> > How about below update:
> > 
> > for (i = 0; i < boot_cpu_data.x86_max_cores; i += smp_num_siblings) {
> >         if (!cpumask_empty(topology_sibling_cpumask(i)))
> >                 cpumask_set_cpu(cpumask_any(topology_sibling_cpumask(i)), &cpu_mask);
> > }
> 
> Why? You do a full for_each_online_cpu(i) loop after that, which does
> exactly the same thing, right?
>  

But looks like power_cpu_init cannot handle it if we don't take any
action here.

e. g. 
cpu_mask: 0000 and online mask: 1111 -> power_cpu_init(0) -> cpu_mask is still: 0000

topology_sibling_cpumask(0): 0011
target: 1 (i. e. we cannot do cpumask_set_cpu(0, &cpu_mask))

Maybe, we need to think out a stronger power_cpu_init if you want to
only do init thing at for_each_online_cpu.

> > > > +	}
> > > > +
> > > > +	for_each_online_cpu(i)
> > > > +		power_cpu_init(i);
> > > > +
> > > > +	__register_cpu_notifier(&power_cpu_notifier_nb);
> > > > +
> > > > +	ret = perf_pmu_register(&pmu_class, "power", -1);
> > > > +	if (WARN_ON(ret)) {
> > > > +		pr_warn("AMD Power PMU registration failed\n");
> > > 
> > > This still leaks the cpu notifier. .....
> > > 
> > 
> > OK, so I should do __unregister_cpu_notifier(&power_cpu_notifier_nb)
> > here.
> 
> You can register the notifier after perf_pmu_register succeeded, right?
>  

Yep.

> > > > +	pr_info("AMD Power PMU detected, %d compute units\n", cu_num);
> > > 
> > > Why is the number of compute units interesting at all?
> > > 
> > 
> > Because the accumulated power bases on compute units.
> > We can see the mask from /sys/devices/power/cpumask and number of
> > compute units to know if all compute units are set at cpumask.
> > So I add a printk here, does it make sense?
> 
> No, because it's completely non intuitive. How on earth am I supposed to get
> the connection between /sys/devices/power/cpumask and number of compute units
> without staring at the code? So this is only interesting for a developer who
> can deduce that number from /proc/cpuinfo or dmesg as well.
> 

OK, I will remove unnecessary cu_num next version. 

Thanks,
Rui

Re: [PATCH v6 2/2] perf/x86/amd/power: Add AMD accumulated power reporting mechanism

[Thomas Gleixner]

On Fri, 4 Mar 2016, Huang Rui wrote:
> On Thu, Mar 03, 2016 at 04:26:46PM +0100, Thomas Gleixner wrote:
> > Why? You do a full for_each_online_cpu(i) loop after that, which does
> > exactly the same thing, right?
> >  
> 
> But looks like power_cpu_init cannot handle it if we don't take any
> action here.
> 
> e. g. 
> cpu_mask: 0000 and online mask: 1111 -> power_cpu_init(0) -> cpu_mask is still: 0000
> 
> topology_sibling_cpumask(0): 0011
> target: 1 (i. e. we cannot do cpumask_set_cpu(0, &cpu_mask))

Fair enough, but then you don't need the power_cpu_init() call at all.

But your loop does not cover anything beyond the first socket. So you need a
separate init function which does:

   for_each_online_cpu(cpu) {
   	target = cpumask_first(topology_sibling_cpumask(cpu));
	if (!cpumask_test_cpu(target, cpumask))
	   	cpumask_set_cpu(target, cpumask);
   }	      
	   	
Thanks,

	tglx

   

Re: [PATCH v6 2/2] perf/x86/amd/power: Add AMD accumulated power reporting mechanism

[Huang Rui]

On Thu, Mar 03, 2016 at 06:57:53PM +0100, Thomas Gleixner wrote:
> On Fri, 4 Mar 2016, Huang Rui wrote:
> > On Thu, Mar 03, 2016 at 04:26:46PM +0100, Thomas Gleixner wrote:
> > > Why? You do a full for_each_online_cpu(i) loop after that, which does
> > > exactly the same thing, right?
> > >  
> > 
> > But looks like power_cpu_init cannot handle it if we don't take any
> > action here.
> > 
> > e. g. 
> > cpu_mask: 0000 and online mask: 1111 -> power_cpu_init(0) -> cpu_mask is still: 0000
> > 
> > topology_sibling_cpumask(0): 0011
> > target: 1 (i. e. we cannot do cpumask_set_cpu(0, &cpu_mask))
> 
> Fair enough, but then you don't need the power_cpu_init() call at all.
> 
> But your loop does not cover anything beyond the first socket. So you need a
> separate init function which does:
> 
>    for_each_online_cpu(cpu) {
>    	target = cpumask_first(topology_sibling_cpumask(cpu));
> 	if (!cpumask_test_cpu(target, cpumask))
> 	   	cpumask_set_cpu(target, cpumask);
>    }	      
> 	   	

Looks good. Thanks Thomas. Will update it.

Rui