Merge branch 'bpf-Add-BPF-support-to-all-perf_event'

Merge branch 'bpf-Add-BPF-support-to-all-perf_event'

Alexei Starovoitov says:

====================
bpf: Add BPF support to all perf_event

v3->v4: one more tweak to reject unsupported events at map
update time as Peter suggested

v2->v3: more refactoring to address Peter's feedback.
Now all perf_events are attachable and readable

v1->v2: address Peter's feedback. Refactor patch 1 to allow attaching
bpf programs to all event types and reading counters from all of them as well
patch 2 - more tests
patch 3 - address Dave's feedback and document bpf_perf_event_read()
and bpf_perf_event_output() properly
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

include/linux/perf_event.h

@@ -896,7 +896,7 @@ perf_event_create_kernel_counter(struct perf_event_attr *attr,
 				void *context);
 extern void perf_pmu_migrate_context(struct pmu *pmu,
 				int src_cpu, int dst_cpu);
-extern u64 perf_event_read_local(struct perf_event *event);
+int perf_event_read_local(struct perf_event *event, u64 *value);
 extern u64 perf_event_read_value(struct perf_event *event,
 				 u64 *enabled, u64 *running);
 
@@ -1301,7 +1301,10 @@ static inline const struct perf_event_attr *perf_event_attrs(struct perf_event *
 {
 	return ERR_PTR(-EINVAL);
 }
-static inline u64 perf_event_read_local(struct perf_event *event)	{ return -EINVAL; }
+static inline int perf_event_read_local(struct perf_event *event, u64 *value)
+{
+	return -EINVAL;
+}
 static inline void perf_event_print_debug(void)				{ }
 static inline int perf_event_task_disable(void)				{ return -EINVAL; }
 static inline int perf_event_task_enable(void)				{ return -EINVAL; }

include/uapi/linux/bpf.h

@@ -313,8 +313,11 @@ union bpf_attr {
  *     @flags: room for future extensions
  *     Return: 0 on success or negative error
  *
- * u64 bpf_perf_event_read(&map, index)
- *     Return: Number events read or error code
+ * u64 bpf_perf_event_read(map, flags)
+ *     read perf event counter value
+ *     @map: pointer to perf_event_array map
+ *     @flags: index of event in the map or bitmask flags
+ *     Return: value of perf event counter read or error code
  *
  * int bpf_redirect(ifindex, flags)
  *     redirect to another netdev
@@ -328,11 +331,11 @@ union bpf_attr {
  *     @skb: pointer to skb
  *     Return: realm if != 0
  *
- * int bpf_perf_event_output(ctx, map, index, data, size)
+ * int bpf_perf_event_output(ctx, map, flags, data, size)
  *     output perf raw sample
  *     @ctx: struct pt_regs*
  *     @map: pointer to perf_event_array map
- *     @index: index of event in the map
+ *     @flags: index of event in the map or bitmask flags
  *     @data: data on stack to be output as raw data
  *     @size: size of data
  *     Return: 0 on success or negative error

kernel/bpf/arraymap.c

@@ -452,38 +452,24 @@ static void bpf_event_entry_free_rcu(struct bpf_event_entry *ee)
 static void *perf_event_fd_array_get_ptr(struct bpf_map *map,
 					 struct file *map_file, int fd)
 {
-	const struct perf_event_attr *attr;
 	struct bpf_event_entry *ee;
 	struct perf_event *event;
 	struct file *perf_file;
+	u64 value;
 
 	perf_file = perf_event_get(fd);
 	if (IS_ERR(perf_file))
 		return perf_file;
 
+	ee = ERR_PTR(-EOPNOTSUPP);
 	event = perf_file->private_data;
-	ee = ERR_PTR(-EINVAL);
-
-	attr = perf_event_attrs(event);
-	if (IS_ERR(attr) || attr->inherit)
+	if (perf_event_read_local(event, &value) == -EOPNOTSUPP)
 		goto err_out;
 
-	switch (attr->type) {
-	case PERF_TYPE_SOFTWARE:
-		if (attr->config != PERF_COUNT_SW_BPF_OUTPUT)
-			goto err_out;
-		/* fall-through */
-	case PERF_TYPE_RAW:
-	case PERF_TYPE_HARDWARE:
-		ee = bpf_event_entry_gen(perf_file, map_file);
-		if (ee)
-			return ee;
-		ee = ERR_PTR(-ENOMEM);
-		/* fall-through */
-	default:
-		break;
-	}
-
+	ee = bpf_event_entry_gen(perf_file, map_file);
+	if (ee)
+		return ee;
+	ee = ERR_PTR(-ENOMEM);
 err_out:
 	fput(perf_file);
 	return ee;

kernel/events/core.c

@@ -3636,36 +3636,48 @@ static inline u64 perf_event_count(struct perf_event *event)
  *     will not be local and we cannot read them atomically
  *   - must not have a pmu::count method
  */
-u64 perf_event_read_local(struct perf_event *event)
+int perf_event_read_local(struct perf_event *event, u64 *value)
 {
 	unsigned long flags;
-	u64 val;
+	int ret = 0;
 
 	/*
 	 * Disabling interrupts avoids all counter scheduling (context
 	 * switches, timer based rotation and IPIs).
 	 */
 	local_irq_save(flags);
 
-	/* If this is a per-task event, it must be for current */
-	WARN_ON_ONCE((event->attach_state & PERF_ATTACH_TASK) &&
-		     event->hw.target != current);
-
-	/* If this is a per-CPU event, it must be for this CPU */
-	WARN_ON_ONCE(!(event->attach_state & PERF_ATTACH_TASK) &&
-		     event->cpu != smp_processor_id());
-
 	/*
 	 * It must not be an event with inherit set, we cannot read
 	 * all child counters from atomic context.
 	 */
-	WARN_ON_ONCE(event->attr.inherit);
+	if (event->attr.inherit) {
+		ret = -EOPNOTSUPP;
+		goto out;
+	}
 
 	/*
 	 * It must not have a pmu::count method, those are not
 	 * NMI safe.
 	 */
-	WARN_ON_ONCE(event->pmu->count);
+	if (event->pmu->count) {
+		ret = -EOPNOTSUPP;
+		goto out;
+	}
+
+	/* If this is a per-task event, it must be for current */
+	if ((event->attach_state & PERF_ATTACH_TASK) &&
+	    event->hw.target != current) {
+		ret = -EINVAL;
+		goto out;
+	}
+
+	/* If this is a per-CPU event, it must be for this CPU */
+	if (!(event->attach_state & PERF_ATTACH_TASK) &&
+	    event->cpu != smp_processor_id()) {
+		ret = -EINVAL;
+		goto out;
+	}
 
 	/*
 	 * If the event is currently on this CPU, its either a per-task event,
@@ -3675,10 +3687,11 @@ u64 perf_event_read_local(struct perf_event *event)
 	if (event->oncpu == smp_processor_id())
 		event->pmu->read(event);
 
-	val = local64_read(&event->count);
+	*value = local64_read(&event->count);
+out:
 	local_irq_restore(flags);
 
-	return val;
+	return ret;
 }
 
 static int perf_event_read(struct perf_event *event, bool group)
@@ -8037,12 +8050,8 @@ static int perf_event_set_bpf_prog(struct perf_event *event, u32 prog_fd)
 	bool is_kprobe, is_tracepoint;
 	struct bpf_prog *prog;
 
-	if (event->attr.type == PERF_TYPE_HARDWARE ||
-	    event->attr.type == PERF_TYPE_SOFTWARE)
-		return perf_event_set_bpf_handler(event, prog_fd);
-
 	if (event->attr.type != PERF_TYPE_TRACEPOINT)
-		return -EINVAL;
+		return perf_event_set_bpf_handler(event, prog_fd);
 
 	if (event->tp_event->prog)
 		return -EEXIST;

kernel/trace/bpf_trace.c

@@ -234,7 +234,8 @@ BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
 	unsigned int cpu = smp_processor_id();
 	u64 index = flags & BPF_F_INDEX_MASK;
 	struct bpf_event_entry *ee;
-	struct perf_event *event;
+	u64 value = 0;
+	int err;
 
 	if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
 		return -EINVAL;
@@ -247,21 +248,14 @@ BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
 	if (!ee)
 		return -ENOENT;
 
-	event = ee->event;
-	if (unlikely(event->attr.type != PERF_TYPE_HARDWARE &&
-		     event->attr.type != PERF_TYPE_RAW))
-		return -EINVAL;
-
-	/* make sure event is local and doesn't have pmu::count */
-	if (unlikely(event->oncpu != cpu || event->pmu->count))
-		return -EINVAL;
-
+	err = perf_event_read_local(ee->event, &value);
 	/*
-	 * we don't know if the function is run successfully by the
-	 * return value. It can be judged in other places, such as
-	 * eBPF programs.
+	 * this api is ugly since we miss [-22..-2] range of valid
+	 * counter values, but that's uapi
 	 */
-	return perf_event_read_local(event);
+	if (err)
+		return err;
+	return value;
 }
 
 static const struct bpf_func_proto bpf_perf_event_read_proto = {

samples/bpf/bpf_helpers.h

@@ -31,7 +31,8 @@ static unsigned long long (*bpf_get_current_uid_gid)(void) =
 	(void *) BPF_FUNC_get_current_uid_gid;
 static int (*bpf_get_current_comm)(void *buf, int buf_size) =
 	(void *) BPF_FUNC_get_current_comm;
-static int (*bpf_perf_event_read)(void *map, int index) =
+static unsigned long long (*bpf_perf_event_read)(void *map,
+						 unsigned long long flags) =
 	(void *) BPF_FUNC_perf_event_read;
 static int (*bpf_clone_redirect)(void *ctx, int ifindex, int flags) =
 	(void *) BPF_FUNC_clone_redirect;

samples/bpf/trace_event_user.c

@@ -75,7 +75,10 @@ static void print_stack(struct key_t *key, __u64 count)
 		for (i = PERF_MAX_STACK_DEPTH - 1; i >= 0; i--)
 			print_addr(ip[i]);
 	}
-	printf("\n");
+	if (count < 6)
+		printf("\r");
+	else
+		printf("\n");
 
 	if (key->kernstack == -EEXIST && !warned) {
 		printf("stackmap collisions seen. Consider increasing size\n");
@@ -105,7 +108,7 @@ static void print_stacks(void)
 		bpf_map_delete_elem(fd, &next_key);
 		key = next_key;
 	}
-
+	printf("\n");
 	if (!sys_read_seen || !sys_write_seen) {
 		printf("BUG kernel stack doesn't contain sys_read() and sys_write()\n");
 		int_exit(0);
@@ -122,24 +125,29 @@ static void test_perf_event_all_cpu(struct perf_event_attr *attr)
 {
 	int nr_cpus = sysconf(_SC_NPROCESSORS_CONF);
 	int *pmu_fd = malloc(nr_cpus * sizeof(int));
-	int i;
+	int i, error = 0;
 
 	/* open perf_event on all cpus */
 	for (i = 0; i < nr_cpus; i++) {
 		pmu_fd[i] = sys_perf_event_open(attr, -1, i, -1, 0);
 		if (pmu_fd[i] < 0) {
 			printf("sys_perf_event_open failed\n");
+			error = 1;
 			goto all_cpu_err;
 		}
 		assert(ioctl(pmu_fd[i], PERF_EVENT_IOC_SET_BPF, prog_fd[0]) == 0);
-		assert(ioctl(pmu_fd[i], PERF_EVENT_IOC_ENABLE, 0) == 0);
+		assert(ioctl(pmu_fd[i], PERF_EVENT_IOC_ENABLE) == 0);
 	}
-	system("dd if=/dev/zero of=/dev/null count=5000k");
+	system("dd if=/dev/zero of=/dev/null count=5000k status=none");
 	print_stacks();
 all_cpu_err:
-	for (i--; i >= 0; i--)
+	for (i--; i >= 0; i--) {
+		ioctl(pmu_fd[i], PERF_EVENT_IOC_DISABLE);
 		close(pmu_fd[i]);
+	}
 	free(pmu_fd);
+	if (error)
+		int_exit(0);
 }
 
 static void test_perf_event_task(struct perf_event_attr *attr)
@@ -150,12 +158,13 @@ static void test_perf_event_task(struct perf_event_attr *attr)
 	pmu_fd = sys_perf_event_open(attr, 0, -1, -1, 0);
 	if (pmu_fd < 0) {
 		printf("sys_perf_event_open failed\n");
-		return;
+		int_exit(0);
 	}
 	assert(ioctl(pmu_fd, PERF_EVENT_IOC_SET_BPF, prog_fd[0]) == 0);
-	assert(ioctl(pmu_fd, PERF_EVENT_IOC_ENABLE, 0) == 0);
-	system("dd if=/dev/zero of=/dev/null count=5000k");
+	assert(ioctl(pmu_fd, PERF_EVENT_IOC_ENABLE) == 0);
+	system("dd if=/dev/zero of=/dev/null count=5000k status=none");
 	print_stacks();
+	ioctl(pmu_fd, PERF_EVENT_IOC_DISABLE);
 	close(pmu_fd);
 }
 
@@ -175,11 +184,56 @@ static void test_bpf_perf_event(void)
 		.config = PERF_COUNT_SW_CPU_CLOCK,
 		.inherit = 1,
 	};
+	struct perf_event_attr attr_hw_cache_l1d = {
+		.sample_freq = SAMPLE_FREQ,
+		.freq = 1,
+		.type = PERF_TYPE_HW_CACHE,
+		.config =
+			PERF_COUNT_HW_CACHE_L1D |
+			(PERF_COUNT_HW_CACHE_OP_READ << 8) |
+			(PERF_COUNT_HW_CACHE_RESULT_ACCESS << 16),
+		.inherit = 1,
+	};
+	struct perf_event_attr attr_hw_cache_branch_miss = {
+		.sample_freq = SAMPLE_FREQ,
+		.freq = 1,
+		.type = PERF_TYPE_HW_CACHE,
+		.config =
+			PERF_COUNT_HW_CACHE_BPU |
+			(PERF_COUNT_HW_CACHE_OP_READ << 8) |
+			(PERF_COUNT_HW_CACHE_RESULT_MISS << 16),
+		.inherit = 1,
+	};
+	struct perf_event_attr attr_type_raw = {
+		.sample_freq = SAMPLE_FREQ,
+		.freq = 1,
+		.type = PERF_TYPE_RAW,
+		/* Intel Instruction Retired */
+		.config = 0xc0,
+		.inherit = 1,
+	};
 
+	printf("Test HW_CPU_CYCLES\n");
 	test_perf_event_all_cpu(&attr_type_hw);
 	test_perf_event_task(&attr_type_hw);
+
+	printf("Test SW_CPU_CLOCK\n");
 	test_perf_event_all_cpu(&attr_type_sw);
 	test_perf_event_task(&attr_type_sw);
+
+	printf("Test HW_CACHE_L1D\n");
+	test_perf_event_all_cpu(&attr_hw_cache_l1d);
+	test_perf_event_task(&attr_hw_cache_l1d);
+
+	printf("Test HW_CACHE_BPU\n");
+	test_perf_event_all_cpu(&attr_hw_cache_branch_miss);
+	test_perf_event_task(&attr_hw_cache_branch_miss);
+
+	printf("Test Instruction Retired\n");
+	test_perf_event_all_cpu(&attr_type_raw);
+	test_perf_event_task(&attr_type_raw);
+
+	printf("*** PASS ***\n");
 }
 
 
@@ -209,7 +263,6 @@ int main(int argc, char **argv)
 		return 0;
 	}
 	test_bpf_perf_event();
-
 	int_exit(0);
 	return 0;
 }