Merge branch 'perfcounters-fixes-for-linus' of... (3493e84d) · Commits · 戴 / test

arch/x86/Kconfig

+1 −1

Original line number	Original line	Diff line number	Diff line
	@@ -24,6 +24,7 @@ config X86
	select HAVE_UNSTABLE_SCHED_CLOCK		select HAVE_UNSTABLE_SCHED_CLOCK
	select HAVE_IDE		select HAVE_IDE
	select HAVE_OPROFILE		select HAVE_OPROFILE
			select HAVE_PERF_COUNTERS if (!M386 && !M486)
	select HAVE_IOREMAP_PROT		select HAVE_IOREMAP_PROT
	select HAVE_KPROBES		select HAVE_KPROBES
	select ARCH_WANT_OPTIONAL_GPIOLIB		select ARCH_WANT_OPTIONAL_GPIOLIB
	@@ -742,7 +743,6 @@ config X86_UP_IOAPIC
	config X86_LOCAL_APIC		config X86_LOCAL_APIC
	def_bool y		def_bool y
	depends on X86_64 \|\| SMP \|\| X86_32_NON_STANDARD \|\| X86_UP_APIC		depends on X86_64 \|\| SMP \|\| X86_32_NON_STANDARD \|\| X86_UP_APIC
	select HAVE_PERF_COUNTERS if (!M386 && !M486)

	config X86_IO_APIC		config X86_IO_APIC
	def_bool y		def_bool y

arch/x86/kernel/cpu/perf_counter.c

+33 −7

Original line number	Original line	Diff line number	Diff line
	@@ -55,6 +55,7 @@ struct x86_pmu {
	int num_counters_fixed;		int num_counters_fixed;
	int counter_bits;		int counter_bits;
	u64 counter_mask;		u64 counter_mask;
			int apic;
	u64 max_period;		u64 max_period;
	u64 intel_ctrl;		u64 intel_ctrl;
	};		};
	@@ -72,8 +73,8 @@ static const u64 p6_perfmon_event_map[] =
	{		{
	[PERF_COUNT_HW_CPU_CYCLES] = 0x0079,		[PERF_COUNT_HW_CPU_CYCLES] = 0x0079,
	[PERF_COUNT_HW_INSTRUCTIONS] = 0x00c0,		[PERF_COUNT_HW_INSTRUCTIONS] = 0x00c0,
	[PERF_COUNT_HW_CACHE_REFERENCES] = 0x0000,		[PERF_COUNT_HW_CACHE_REFERENCES] = 0x0f2e,
	[PERF_COUNT_HW_CACHE_MISSES] = 0x0000,		[PERF_COUNT_HW_CACHE_MISSES] = 0x012e,
	[PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = 0x00c4,		[PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = 0x00c4,
	[PERF_COUNT_HW_BRANCH_MISSES] = 0x00c5,		[PERF_COUNT_HW_BRANCH_MISSES] = 0x00c5,
	[PERF_COUNT_HW_BUS_CYCLES] = 0x0062,		[PERF_COUNT_HW_BUS_CYCLES] = 0x0062,
	@@ -613,6 +614,7 @@ static DEFINE_MUTEX(pmc_reserve_mutex);

	static bool reserve_pmc_hardware(void)		static bool reserve_pmc_hardware(void)
	{		{
			#ifdef CONFIG_X86_LOCAL_APIC
	int i;		int i;

	if (nmi_watchdog == NMI_LOCAL_APIC)		if (nmi_watchdog == NMI_LOCAL_APIC)
	@@ -627,9 +629,11 @@ static bool reserve_pmc_hardware(void)
	if (!reserve_evntsel_nmi(x86_pmu.eventsel + i))		if (!reserve_evntsel_nmi(x86_pmu.eventsel + i))
	goto eventsel_fail;		goto eventsel_fail;
	}		}
			#endif

	return true;		return true;

			#ifdef CONFIG_X86_LOCAL_APIC
	eventsel_fail:		eventsel_fail:
	for (i--; i >= 0; i--)		for (i--; i >= 0; i--)
	release_evntsel_nmi(x86_pmu.eventsel + i);		release_evntsel_nmi(x86_pmu.eventsel + i);
	@@ -644,10 +648,12 @@ perfctr_fail:
	enable_lapic_nmi_watchdog();		enable_lapic_nmi_watchdog();

	return false;		return false;
			#endif
	}		}

	static void release_pmc_hardware(void)		static void release_pmc_hardware(void)
	{		{
			#ifdef CONFIG_X86_LOCAL_APIC
	int i;		int i;

	for (i = 0; i < x86_pmu.num_counters; i++) {		for (i = 0; i < x86_pmu.num_counters; i++) {
	@@ -657,6 +663,7 @@ static void release_pmc_hardware(void)

	if (nmi_watchdog == NMI_LOCAL_APIC)		if (nmi_watchdog == NMI_LOCAL_APIC)
	enable_lapic_nmi_watchdog();		enable_lapic_nmi_watchdog();
			#endif
	}		}

	static void hw_perf_counter_destroy(struct perf_counter *counter)		static void hw_perf_counter_destroy(struct perf_counter *counter)
	@@ -748,6 +755,15 @@ static int __hw_perf_counter_init(struct perf_counter *counter)
	hwc->sample_period = x86_pmu.max_period;		hwc->sample_period = x86_pmu.max_period;
	hwc->last_period = hwc->sample_period;		hwc->last_period = hwc->sample_period;
	atomic64_set(&hwc->period_left, hwc->sample_period);		atomic64_set(&hwc->period_left, hwc->sample_period);
			} else {
			/*
			* If we have a PMU initialized but no APIC
			* interrupts, we cannot sample hardware
			* counters (user-space has to fall back and
			* sample via a hrtimer based software counter):
			*/
			if (!x86_pmu.apic)
			return -EOPNOTSUPP;
	}		}

	counter->destroy = hw_perf_counter_destroy;		counter->destroy = hw_perf_counter_destroy;
	@@ -1449,18 +1465,22 @@ void smp_perf_pending_interrupt(struct pt_regs *regs)

	void set_perf_counter_pending(void)		void set_perf_counter_pending(void)
	{		{
			#ifdef CONFIG_X86_LOCAL_APIC
	apic->send_IPI_self(LOCAL_PENDING_VECTOR);		apic->send_IPI_self(LOCAL_PENDING_VECTOR);
			#endif
	}		}

	void perf_counters_lapic_init(void)		void perf_counters_lapic_init(void)
	{		{
	if (!x86_pmu_initialized())		#ifdef CONFIG_X86_LOCAL_APIC
			if (!x86_pmu.apic \|\| !x86_pmu_initialized())
	return;		return;

	/*		/*
	* Always use NMI for PMU		* Always use NMI for PMU
	*/		*/
	apic_write(APIC_LVTPC, APIC_DM_NMI);		apic_write(APIC_LVTPC, APIC_DM_NMI);
			#endif
	}		}

	static int __kprobes		static int __kprobes
	@@ -1484,7 +1504,9 @@ perf_counter_nmi_handler(struct notifier_block *self,

	regs = args->regs;		regs = args->regs;

			#ifdef CONFIG_X86_LOCAL_APIC
	apic_write(APIC_LVTPC, APIC_DM_NMI);		apic_write(APIC_LVTPC, APIC_DM_NMI);
			#endif
	/*		/*
	* Can't rely on the handled return value to say it was our NMI, two		* Can't rely on the handled return value to say it was our NMI, two
	* counters could trigger 'simultaneously' raising two back-to-back NMIs.		* counters could trigger 'simultaneously' raising two back-to-back NMIs.
	@@ -1515,6 +1537,7 @@ static struct x86_pmu p6_pmu = {
	.event_map = p6_pmu_event_map,		.event_map = p6_pmu_event_map,
	.raw_event = p6_pmu_raw_event,		.raw_event = p6_pmu_raw_event,
	.max_events = ARRAY_SIZE(p6_perfmon_event_map),		.max_events = ARRAY_SIZE(p6_perfmon_event_map),
			.apic = 1,
	.max_period = (1ULL << 31) - 1,		.max_period = (1ULL << 31) - 1,
	.version = 0,		.version = 0,
	.num_counters = 2,		.num_counters = 2,
	@@ -1541,6 +1564,7 @@ static struct x86_pmu intel_pmu = {
	.event_map = intel_pmu_event_map,		.event_map = intel_pmu_event_map,
	.raw_event = intel_pmu_raw_event,		.raw_event = intel_pmu_raw_event,
	.max_events = ARRAY_SIZE(intel_perfmon_event_map),		.max_events = ARRAY_SIZE(intel_perfmon_event_map),
			.apic = 1,
	/*		/*
	* Intel PMCs cannot be accessed sanely above 32 bit width,		* Intel PMCs cannot be accessed sanely above 32 bit width,
	* so we install an artificial 1<<31 period regardless of		* so we install an artificial 1<<31 period regardless of
	@@ -1564,6 +1588,7 @@ static struct x86_pmu amd_pmu = {
	.num_counters = 4,		.num_counters = 4,
	.counter_bits = 48,		.counter_bits = 48,
	.counter_mask = (1ULL << 48) - 1,		.counter_mask = (1ULL << 48) - 1,
			.apic = 1,
	/* use highest bit to detect overflow */		/* use highest bit to detect overflow */
	.max_period = (1ULL << 47) - 1,		.max_period = (1ULL << 47) - 1,
	};		};
	@@ -1589,13 +1614,14 @@ static int p6_pmu_init(void)
	return -ENODEV;		return -ENODEV;
	}		}

			x86_pmu = p6_pmu;

	if (!cpu_has_apic) {		if (!cpu_has_apic) {
	pr_info("no Local APIC, try rebooting with lapic");		pr_info("no APIC, boot with the \"lapic\" boot parameter to force-enable it.\n");
	return -ENODEV;		pr_info("no hardware sampling interrupt available.\n");
			x86_pmu.apic = 0;
	}		}

	x86_pmu = p6_pmu;

	return 0;		return 0;
	}		}

include/linux/perf_counter.h

+38 −11

Original line number	Original line	Diff line number	Diff line
	@@ -115,7 +115,7 @@ enum perf_counter_sample_format {
	PERF_SAMPLE_TID = 1U << 1,		PERF_SAMPLE_TID = 1U << 1,
	PERF_SAMPLE_TIME = 1U << 2,		PERF_SAMPLE_TIME = 1U << 2,
	PERF_SAMPLE_ADDR = 1U << 3,		PERF_SAMPLE_ADDR = 1U << 3,
	PERF_SAMPLE_GROUP = 1U << 4,		PERF_SAMPLE_READ = 1U << 4,
	PERF_SAMPLE_CALLCHAIN = 1U << 5,		PERF_SAMPLE_CALLCHAIN = 1U << 5,
	PERF_SAMPLE_ID = 1U << 6,		PERF_SAMPLE_ID = 1U << 6,
	PERF_SAMPLE_CPU = 1U << 7,		PERF_SAMPLE_CPU = 1U << 7,
	@@ -127,16 +127,32 @@ enum perf_counter_sample_format {
	};		};

	/*		/*
	* Bits that can be set in attr.read_format to request that		* The format of the data returned by read() on a perf counter fd,
	* reads on the counter should return the indicated quantities,		* as specified by attr.read_format:
	* in increasing order of bit value, after the counter value.		*
			* struct read_format {
			* { u64 value;
			* { u64 time_enabled; } && PERF_FORMAT_ENABLED
			* { u64 time_running; } && PERF_FORMAT_RUNNING
			* { u64 id; } && PERF_FORMAT_ID
			* } && !PERF_FORMAT_GROUP
			*
			* { u64 nr;
			* { u64 time_enabled; } && PERF_FORMAT_ENABLED
			* { u64 time_running; } && PERF_FORMAT_RUNNING
			* { u64 value;
			* { u64 id; } && PERF_FORMAT_ID
			* } cntr[nr];
			* } && PERF_FORMAT_GROUP
			* };
	*/		*/
	enum perf_counter_read_format {		enum perf_counter_read_format {
	PERF_FORMAT_TOTAL_TIME_ENABLED = 1U << 0,		PERF_FORMAT_TOTAL_TIME_ENABLED = 1U << 0,
	PERF_FORMAT_TOTAL_TIME_RUNNING = 1U << 1,		PERF_FORMAT_TOTAL_TIME_RUNNING = 1U << 1,
	PERF_FORMAT_ID = 1U << 2,		PERF_FORMAT_ID = 1U << 2,
			PERF_FORMAT_GROUP = 1U << 3,

	PERF_FORMAT_MAX = 1U << 3, /* non-ABI */		PERF_FORMAT_MAX = 1U << 4, /* non-ABI */
	};		};

	#define PERF_ATTR_SIZE_VER0 64 /* sizeof first published struct */		#define PERF_ATTR_SIZE_VER0 64 /* sizeof first published struct */
	@@ -343,10 +359,8 @@ enum perf_event_type {
	* struct {		* struct {
	* struct perf_event_header header;		* struct perf_event_header header;
	* u32 pid, tid;		* u32 pid, tid;
	* u64 value;		*
	* { u64 time_enabled; } && PERF_FORMAT_ENABLED		* struct read_format values;
	* { u64 time_running; } && PERF_FORMAT_RUNNING
	* { u64 parent_id; } && PERF_FORMAT_ID
	* };		* };
	*/		*/
	PERF_EVENT_READ = 8,		PERF_EVENT_READ = 8,
	@@ -364,11 +378,22 @@ enum perf_event_type {
	* { u32 cpu, res; } && PERF_SAMPLE_CPU		* { u32 cpu, res; } && PERF_SAMPLE_CPU
	* { u64 period; } && PERF_SAMPLE_PERIOD		* { u64 period; } && PERF_SAMPLE_PERIOD
	*		*
	* { u64 nr;		* { struct read_format values; } && PERF_SAMPLE_READ
	* { u64 id, val; } cnt[nr]; } && PERF_SAMPLE_GROUP
	*		*
	* { u64 nr,		* { u64 nr,
	* u64 ips[nr]; } && PERF_SAMPLE_CALLCHAIN		* u64 ips[nr]; } && PERF_SAMPLE_CALLCHAIN
			*
			* #
			* # The RAW record below is opaque data wrt the ABI
			* #
			* # That is, the ABI doesn't make any promises wrt to
			* # the stability of its content, it may vary depending
			* # on event, hardware, kernel version and phase of
			* # the moon.
			* #
			* # In other words, PERF_SAMPLE_RAW contents are not an ABI.
			* #
			*
	* { u32 size;		* { u32 size;
	* char data[size];}&& PERF_SAMPLE_RAW		* char data[size];}&& PERF_SAMPLE_RAW
	* };		* };
	@@ -694,6 +719,8 @@ struct perf_sample_data {

	extern int perf_counter_overflow(struct perf_counter *counter, int nmi,		extern int perf_counter_overflow(struct perf_counter *counter, int nmi,
	struct perf_sample_data *data);		struct perf_sample_data *data);
			extern void perf_counter_output(struct perf_counter *counter, int nmi,
			struct perf_sample_data *data);

	/*		/*
	* Return 1 for a software counter, 0 for a hardware counter		* Return 1 for a software counter, 0 for a hardware counter

kernel/perf_counter.c

+236 −102

Original line number	Original line	Diff line number	Diff line
	@@ -88,6 +88,7 @@ void __weak hw_perf_disable(void) { barrier(); }
	void __weak hw_perf_enable(void) { barrier(); }		void __weak hw_perf_enable(void) { barrier(); }

	void __weak hw_perf_counter_setup(int cpu) { barrier(); }		void __weak hw_perf_counter_setup(int cpu) { barrier(); }
			void __weak hw_perf_counter_setup_online(int cpu) { barrier(); }

	int __weak		int __weak
	hw_perf_group_sched_in(struct perf_counter *group_leader,		hw_perf_group_sched_in(struct perf_counter *group_leader,
	@@ -306,6 +307,10 @@ counter_sched_out(struct perf_counter *counter,
	return;		return;

	counter->state = PERF_COUNTER_STATE_INACTIVE;		counter->state = PERF_COUNTER_STATE_INACTIVE;
			if (counter->pending_disable) {
			counter->pending_disable = 0;
			counter->state = PERF_COUNTER_STATE_OFF;
			}
	counter->tstamp_stopped = ctx->time;		counter->tstamp_stopped = ctx->time;
	counter->pmu->disable(counter);		counter->pmu->disable(counter);
	counter->oncpu = -1;		counter->oncpu = -1;
	@@ -1691,7 +1696,32 @@ static int perf_release(struct inode inode, struct file file)
	return 0;		return 0;
	}		}

	static u64 perf_counter_read_tree(struct perf_counter *counter)		static int perf_counter_read_size(struct perf_counter *counter)
			{
			int entry = sizeof(u64); /* value */
			int size = 0;
			int nr = 1;

			if (counter->attr.read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
			size += sizeof(u64);

			if (counter->attr.read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
			size += sizeof(u64);

			if (counter->attr.read_format & PERF_FORMAT_ID)
			entry += sizeof(u64);

			if (counter->attr.read_format & PERF_FORMAT_GROUP) {
			nr += counter->group_leader->nr_siblings;
			size += sizeof(u64);
			}

			size += entry * nr;

			return size;
			}

			static u64 perf_counter_read_value(struct perf_counter *counter)
	{		{
	struct perf_counter *child;		struct perf_counter *child;
	u64 total = 0;		u64 total = 0;
	@@ -1703,14 +1733,96 @@ static u64 perf_counter_read_tree(struct perf_counter *counter)
	return total;		return total;
	}		}

			static int perf_counter_read_entry(struct perf_counter *counter,
			u64 read_format, char __user *buf)
			{
			int n = 0, count = 0;
			u64 values[2];

			values[n++] = perf_counter_read_value(counter);
			if (read_format & PERF_FORMAT_ID)
			values[n++] = primary_counter_id(counter);

			count = n * sizeof(u64);

			if (copy_to_user(buf, values, count))
			return -EFAULT;

			return count;
			}

			static int perf_counter_read_group(struct perf_counter *counter,
			u64 read_format, char __user *buf)
			{
			struct perf_counter leader = counter->group_leader, sub;
			int n = 0, size = 0, err = -EFAULT;
			u64 values[3];

			values[n++] = 1 + leader->nr_siblings;
			if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) {
			values[n++] = leader->total_time_enabled +
			atomic64_read(&leader->child_total_time_enabled);
			}
			if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) {
			values[n++] = leader->total_time_running +
			atomic64_read(&leader->child_total_time_running);
			}

			size = n * sizeof(u64);

			if (copy_to_user(buf, values, size))
			return -EFAULT;

			err = perf_counter_read_entry(leader, read_format, buf + size);
			if (err < 0)
			return err;

			size += err;

			list_for_each_entry(sub, &leader->sibling_list, list_entry) {
			err = perf_counter_read_entry(counter, read_format,
			buf + size);
			if (err < 0)
			return err;

			size += err;
			}

			return size;
			}

			static int perf_counter_read_one(struct perf_counter *counter,
			u64 read_format, char __user *buf)
			{
			u64 values[4];
			int n = 0;

			values[n++] = perf_counter_read_value(counter);
			if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) {
			values[n++] = counter->total_time_enabled +
			atomic64_read(&counter->child_total_time_enabled);
			}
			if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) {
			values[n++] = counter->total_time_running +
			atomic64_read(&counter->child_total_time_running);
			}
			if (read_format & PERF_FORMAT_ID)
			values[n++] = primary_counter_id(counter);

			if (copy_to_user(buf, values, n * sizeof(u64)))
			return -EFAULT;

			return n * sizeof(u64);
			}

	/*		/*
	* Read the performance counter - simple non blocking version for now		* Read the performance counter - simple non blocking version for now
	*/		*/
	static ssize_t		static ssize_t
	perf_read_hw(struct perf_counter counter, char __user buf, size_t count)		perf_read_hw(struct perf_counter counter, char __user buf, size_t count)
	{		{
	u64 values[4];		u64 read_format = counter->attr.read_format;
	int n;		int ret;

	/*		/*
	* Return end-of-file for a read on a counter that is in		* Return end-of-file for a read on a counter that is in
	@@ -1720,28 +1832,18 @@ perf_read_hw(struct perf_counter counter, char __user buf, size_t count)
	if (counter->state == PERF_COUNTER_STATE_ERROR)		if (counter->state == PERF_COUNTER_STATE_ERROR)
	return 0;		return 0;

			if (count < perf_counter_read_size(counter))
			return -ENOSPC;

	WARN_ON_ONCE(counter->ctx->parent_ctx);		WARN_ON_ONCE(counter->ctx->parent_ctx);
	mutex_lock(&counter->child_mutex);		mutex_lock(&counter->child_mutex);
	values[0] = perf_counter_read_tree(counter);		if (read_format & PERF_FORMAT_GROUP)
	n = 1;		ret = perf_counter_read_group(counter, read_format, buf);
	if (counter->attr.read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)		else
	values[n++] = counter->total_time_enabled +		ret = perf_counter_read_one(counter, read_format, buf);
	atomic64_read(&counter->child_total_time_enabled);
	if (counter->attr.read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
	values[n++] = counter->total_time_running +
	atomic64_read(&counter->child_total_time_running);
	if (counter->attr.read_format & PERF_FORMAT_ID)
	values[n++] = primary_counter_id(counter);
	mutex_unlock(&counter->child_mutex);		mutex_unlock(&counter->child_mutex);

	if (count < n * sizeof(u64))		return ret;
	return -EINVAL;
	count = n * sizeof(u64);

	if (copy_to_user(buf, values, count))
	return -EFAULT;

	return count;
	}		}

	static ssize_t		static ssize_t
	@@ -2245,7 +2347,7 @@ static void perf_pending_counter(struct perf_pending_entry *entry)

	if (counter->pending_disable) {		if (counter->pending_disable) {
	counter->pending_disable = 0;		counter->pending_disable = 0;
	perf_counter_disable(counter);		__perf_counter_disable(counter);
	}		}

	if (counter->pending_wakeup) {		if (counter->pending_wakeup) {
	@@ -2630,7 +2732,80 @@ static u32 perf_counter_tid(struct perf_counter counter, struct task_struct p)
	return task_pid_nr_ns(p, counter->ns);		return task_pid_nr_ns(p, counter->ns);
	}		}

	static void perf_counter_output(struct perf_counter *counter, int nmi,		static void perf_output_read_one(struct perf_output_handle *handle,
			struct perf_counter *counter)
			{
			u64 read_format = counter->attr.read_format;
			u64 values[4];
			int n = 0;

			values[n++] = atomic64_read(&counter->count);
			if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) {
			values[n++] = counter->total_time_enabled +
			atomic64_read(&counter->child_total_time_enabled);
			}
			if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) {
			values[n++] = counter->total_time_running +
			atomic64_read(&counter->child_total_time_running);
			}
			if (read_format & PERF_FORMAT_ID)
			values[n++] = primary_counter_id(counter);

			perf_output_copy(handle, values, n * sizeof(u64));
			}

			/*
			* XXX PERF_FORMAT_GROUP vs inherited counters seems difficult.
			*/
			static void perf_output_read_group(struct perf_output_handle *handle,
			struct perf_counter *counter)
			{
			struct perf_counter leader = counter->group_leader, sub;
			u64 read_format = counter->attr.read_format;
			u64 values[5];
			int n = 0;

			values[n++] = 1 + leader->nr_siblings;

			if (read_format & PERF_FORMAT_TOTAL_TIME_ENABLED)
			values[n++] = leader->total_time_enabled;

			if (read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
			values[n++] = leader->total_time_running;

			if (leader != counter)
			leader->pmu->read(leader);

			values[n++] = atomic64_read(&leader->count);
			if (read_format & PERF_FORMAT_ID)
			values[n++] = primary_counter_id(leader);

			perf_output_copy(handle, values, n * sizeof(u64));

			list_for_each_entry(sub, &leader->sibling_list, list_entry) {
			n = 0;

			if (sub != counter)
			sub->pmu->read(sub);

			values[n++] = atomic64_read(&sub->count);
			if (read_format & PERF_FORMAT_ID)
			values[n++] = primary_counter_id(sub);

			perf_output_copy(handle, values, n * sizeof(u64));
			}
			}

			static void perf_output_read(struct perf_output_handle *handle,
			struct perf_counter *counter)
			{
			if (counter->attr.read_format & PERF_FORMAT_GROUP)
			perf_output_read_group(handle, counter);
			else
			perf_output_read_one(handle, counter);
			}

			void perf_counter_output(struct perf_counter *counter, int nmi,
	struct perf_sample_data *data)		struct perf_sample_data *data)
	{		{
	int ret;		int ret;
	@@ -2641,10 +2816,6 @@ static void perf_counter_output(struct perf_counter *counter, int nmi,
	struct {		struct {
	u32 pid, tid;		u32 pid, tid;
	} tid_entry;		} tid_entry;
	struct {
	u64 id;
	u64 counter;
	} group_entry;
	struct perf_callchain_entry *callchain = NULL;		struct perf_callchain_entry *callchain = NULL;
	int callchain_size = 0;		int callchain_size = 0;
	u64 time;		u64 time;
	@@ -2699,10 +2870,8 @@ static void perf_counter_output(struct perf_counter *counter, int nmi,
	if (sample_type & PERF_SAMPLE_PERIOD)		if (sample_type & PERF_SAMPLE_PERIOD)
	header.size += sizeof(u64);		header.size += sizeof(u64);

	if (sample_type & PERF_SAMPLE_GROUP) {		if (sample_type & PERF_SAMPLE_READ)
	header.size += sizeof(u64) +		header.size += perf_counter_read_size(counter);
	counter->nr_siblings * sizeof(group_entry);
	}

	if (sample_type & PERF_SAMPLE_CALLCHAIN) {		if (sample_type & PERF_SAMPLE_CALLCHAIN) {
	callchain = perf_callchain(data->regs);		callchain = perf_callchain(data->regs);
	@@ -2759,26 +2928,8 @@ static void perf_counter_output(struct perf_counter *counter, int nmi,
	if (sample_type & PERF_SAMPLE_PERIOD)		if (sample_type & PERF_SAMPLE_PERIOD)
	perf_output_put(&handle, data->period);		perf_output_put(&handle, data->period);

	/*		if (sample_type & PERF_SAMPLE_READ)
	* XXX PERF_SAMPLE_GROUP vs inherited counters seems difficult.		perf_output_read(&handle, counter);
	*/
	if (sample_type & PERF_SAMPLE_GROUP) {
	struct perf_counter leader, sub;
	u64 nr = counter->nr_siblings;

	perf_output_put(&handle, nr);

	leader = counter->group_leader;
	list_for_each_entry(sub, &leader->sibling_list, list_entry) {
	if (sub != counter)
	sub->pmu->read(sub);

	group_entry.id = primary_counter_id(sub);
	group_entry.counter = atomic64_read(&sub->count);

	perf_output_put(&handle, group_entry);
	}
	}

	if (sample_type & PERF_SAMPLE_CALLCHAIN) {		if (sample_type & PERF_SAMPLE_CALLCHAIN) {
	if (callchain)		if (callchain)
	@@ -2817,8 +2968,6 @@ struct perf_read_event {

	u32 pid;		u32 pid;
	u32 tid;		u32 tid;
	u64 value;
	u64 format[3];
	};		};

	static void		static void
	@@ -2830,34 +2979,20 @@ perf_counter_read_event(struct perf_counter *counter,
	.header = {		.header = {
	.type = PERF_EVENT_READ,		.type = PERF_EVENT_READ,
	.misc = 0,		.misc = 0,
	.size = sizeof(event) - sizeof(event.format),		.size = sizeof(event) + perf_counter_read_size(counter),
	},		},
	.pid = perf_counter_pid(counter, task),		.pid = perf_counter_pid(counter, task),
	.tid = perf_counter_tid(counter, task),		.tid = perf_counter_tid(counter, task),
	.value = atomic64_read(&counter->count),
	};		};
	int ret, i = 0;		int ret;

	if (counter->attr.read_format & PERF_FORMAT_TOTAL_TIME_ENABLED) {
	event.header.size += sizeof(u64);
	event.format[i++] = counter->total_time_enabled;
	}

	if (counter->attr.read_format & PERF_FORMAT_TOTAL_TIME_RUNNING) {
	event.header.size += sizeof(u64);
	event.format[i++] = counter->total_time_running;
	}

	if (counter->attr.read_format & PERF_FORMAT_ID) {
	event.header.size += sizeof(u64);
	event.format[i++] = primary_counter_id(counter);
	}

	ret = perf_output_begin(&handle, counter, event.header.size, 0, 0);		ret = perf_output_begin(&handle, counter, event.header.size, 0, 0);
	if (ret)		if (ret)
	return;		return;

	perf_output_copy(&handle, &event, event.header.size);		perf_output_put(&handle, event);
			perf_output_read(&handle, counter);

	perf_output_end(&handle);		perf_output_end(&handle);
	}		}

	@@ -2893,10 +3028,10 @@ static void perf_counter_task_output(struct perf_counter *counter,
	return;		return;

	task_event->event.pid = perf_counter_pid(counter, task);		task_event->event.pid = perf_counter_pid(counter, task);
	task_event->event.ppid = perf_counter_pid(counter, task->real_parent);		task_event->event.ppid = perf_counter_pid(counter, current);

	task_event->event.tid = perf_counter_tid(counter, task);		task_event->event.tid = perf_counter_tid(counter, task);
	task_event->event.ptid = perf_counter_tid(counter, task->real_parent);		task_event->event.ptid = perf_counter_tid(counter, current);

	perf_output_put(&handle, task_event->event);		perf_output_put(&handle, task_event->event);
	perf_output_end(&handle);		perf_output_end(&handle);
	@@ -3443,40 +3578,32 @@ static void perf_swcounter_add(struct perf_counter *counter, u64 nr,

	static int perf_swcounter_is_counting(struct perf_counter *counter)		static int perf_swcounter_is_counting(struct perf_counter *counter)
	{		{
	struct perf_counter_context *ctx;		/*
	unsigned long flags;		* The counter is active, we're good!
	int count;		*/

	if (counter->state == PERF_COUNTER_STATE_ACTIVE)		if (counter->state == PERF_COUNTER_STATE_ACTIVE)
	return 1;		return 1;

			/*
			* The counter is off/error, not counting.
			*/
	if (counter->state != PERF_COUNTER_STATE_INACTIVE)		if (counter->state != PERF_COUNTER_STATE_INACTIVE)
	return 0;		return 0;

	/*		/*
	* If the counter is inactive, it could be just because		* The counter is inactive, if the context is active
	* its task is scheduled out, or because it's in a group		* we're part of a group that didn't make it on the 'pmu',
	* which could not go on the PMU. We want to count in		* not counting.
	* the first case but not the second. If the context is
	* currently active then an inactive software counter must
	* be the second case. If it's not currently active then
	* we need to know whether the counter was active when the
	* context was last active, which we can determine by
	* comparing counter->tstamp_stopped with ctx->time.
	*
	* We are within an RCU read-side critical section,
	* which protects the existence of *ctx.
	*/		*/
	ctx = counter->ctx;		if (counter->ctx->is_active)
	spin_lock_irqsave(&ctx->lock, flags);		return 0;
	count = 1;
	/* Re-check state now we have the lock */		/*
	if (counter->state < PERF_COUNTER_STATE_INACTIVE \|\|		* We're inactive and the context is too, this means the
	counter->ctx->is_active \|\|		* task is scheduled out, we're counting events that happen
	counter->tstamp_stopped < ctx->time)		* to us, like migration events.
	count = 0;		*/
	spin_unlock_irqrestore(&ctx->lock, flags);		return 1;
	return count;
	}		}

	static int perf_swcounter_match(struct perf_counter *counter,		static int perf_swcounter_match(struct perf_counter *counter,
	@@ -3928,9 +4055,9 @@ perf_counter_alloc(struct perf_counter_attr *attr,
	atomic64_set(&hwc->period_left, hwc->sample_period);		atomic64_set(&hwc->period_left, hwc->sample_period);

	/*		/*
	* we currently do not support PERF_SAMPLE_GROUP on inherited counters		* we currently do not support PERF_FORMAT_GROUP on inherited counters
	*/		*/
	if (attr->inherit && (attr->sample_type & PERF_SAMPLE_GROUP))		if (attr->inherit && (attr->read_format & PERF_FORMAT_GROUP))
	goto done;		goto done;

	switch (attr->type) {		switch (attr->type) {
	@@ -4592,6 +4719,11 @@ perf_cpu_notify(struct notifier_block self, unsigned long action, void hcpu)
	perf_counter_init_cpu(cpu);		perf_counter_init_cpu(cpu);
	break;		break;

			case CPU_ONLINE:
			case CPU_ONLINE_FROZEN:
			hw_perf_counter_setup_online(cpu);
			break;

	case CPU_DOWN_PREPARE:		case CPU_DOWN_PREPARE:
	case CPU_DOWN_PREPARE_FROZEN:		case CPU_DOWN_PREPARE_FROZEN:
	perf_counter_exit_cpu(cpu);		perf_counter_exit_cpu(cpu);
	@@ -4616,6 +4748,8 @@ void __init perf_counter_init(void)
	{		{
	perf_cpu_notify(&perf_cpu_nb, (unsigned long)CPU_UP_PREPARE,		perf_cpu_notify(&perf_cpu_nb, (unsigned long)CPU_UP_PREPARE,
	(void *)(long)smp_processor_id());		(void *)(long)smp_processor_id());
			perf_cpu_notify(&perf_cpu_nb, (unsigned long)CPU_ONLINE,
			(void *)(long)smp_processor_id());
	register_cpu_notifier(&perf_cpu_nb);		register_cpu_notifier(&perf_cpu_nb);
	}		}

tools/perf/Makefile

+18 −11

Original line number	Original line	Diff line number	Diff line
	@@ -382,24 +382,31 @@ endif
	ifdef NO_DEMANGLE		ifdef NO_DEMANGLE
	BASIC_CFLAGS += -DNO_DEMANGLE		BASIC_CFLAGS += -DNO_DEMANGLE
	else		else

	has_bfd := $(shell sh -c "(echo '\#include <bfd.h>'; echo 'int main(void) { bfd_demangle(0, 0, 0); return 0; }') \| $(CC) -x c - $(ALL_CFLAGS) -o /dev/null $(ALL_LDFLAGS) -lbfd > /dev/null 2>&1 && echo y")		has_bfd := $(shell sh -c "(echo '\#include <bfd.h>'; echo 'int main(void) { bfd_demangle(0, 0, 0); return 0; }') \| $(CC) -x c - $(ALL_CFLAGS) -o /dev/null $(ALL_LDFLAGS) -lbfd > /dev/null 2>&1 && echo y")

	has_bfd_iberty := $(shell sh -c "(echo '\#include <bfd.h>'; echo 'int main(void) { bfd_demangle(0, 0, 0); return 0; }') \| $(CC) -x c - $(ALL_CFLAGS) -o /dev/null $(ALL_LDFLAGS) -lbfd -liberty > /dev/null 2>&1 && echo y")

	has_bfd_iberty_z := $(shell sh -c "(echo '\#include <bfd.h>'; echo 'int main(void) { bfd_demangle(0, 0, 0); return 0; }') \| $(CC) -x c - $(ALL_CFLAGS) -o /dev/null $(ALL_LDFLAGS) -lbfd -liberty -lz > /dev/null 2>&1 && echo y")

	ifeq ($(has_bfd),y)		ifeq ($(has_bfd),y)
	EXTLIBS += -lbfd		EXTLIBS += -lbfd
	else ifeq ($(has_bfd_iberty),y)		else
			has_bfd_iberty := $(shell sh -c "(echo '\#include <bfd.h>'; echo 'int main(void) { bfd_demangle(0, 0, 0); return 0; }') \| $(CC) -x c - $(ALL_CFLAGS) -o /dev/null $(ALL_LDFLAGS) -lbfd -liberty > /dev/null 2>&1 && echo y")
			ifeq ($(has_bfd_iberty),y)
	EXTLIBS += -lbfd -liberty		EXTLIBS += -lbfd -liberty
	else ifeq ($(has_bfd_iberty_z),y)		else
			has_bfd_iberty_z := $(shell sh -c "(echo '\#include <bfd.h>'; echo 'int main(void) { bfd_demangle(0, 0, 0); return 0; }') \| $(CC) -x c - $(ALL_CFLAGS) -o /dev/null $(ALL_LDFLAGS) -lbfd -liberty -lz > /dev/null 2>&1 && echo y")
			ifeq ($(has_bfd_iberty_z),y)
	EXTLIBS += -lbfd -liberty -lz		EXTLIBS += -lbfd -liberty -lz
			else
			has_cplus_demangle := $(shell sh -c "(echo 'extern char cplus_demangle(const char , int);'; echo 'int main(void) { cplus_demangle(0, 0); return 0; }') \| $(CC) -x c - $(ALL_CFLAGS) -o /dev/null $(ALL_LDFLAGS) -liberty > /dev/null 2>&1 && echo y")
			ifeq ($(has_cplus_demangle),y)
			EXTLIBS += -liberty
			BASIC_CFLAGS += -DHAVE_CPLUS_DEMANGLE
	else		else
	msg := $(warning No bfd.h/libbfd found, install binutils-dev[el] to gain symbol demangling)		msg := $(warning No bfd.h/libbfd found, install binutils-dev[el] to gain symbol demangling)
	BASIC_CFLAGS += -DNO_DEMANGLE		BASIC_CFLAGS += -DNO_DEMANGLE
	endif		endif
	endif		endif
			endif
			endif
			endif

	ifndef CC_LD_DYNPATH		ifndef CC_LD_DYNPATH
	ifdef NO_R_TO_GCC_LINKER		ifdef NO_R_TO_GCC_LINKER

Admin message