Commit a11227dc authored by David S. Miller's avatar David S. Miller
Browse files

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: default avatarDavid S. Miller <davem@davemloft.net>
parents 5071034e b7d3ed5b

include/linux/perf_event.h

+5 −2
Original line number Diff line number Diff line
@@ -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

+7 −4
Original line number Diff line number Diff line
@@ -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

+7 −21
Original line number Diff line number Diff line
@@ -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;

	}



err_out:

	fput(perf_file);

	return ee;

kernel/events/core.c

+28 −19
Original line number Diff line number Diff line
@@ -3636,10 +3636,10 @@ 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
@@ -3647,25 +3647,37 @@ u64 perf_event_read_local(struct perf_event *event) 
	 */

	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

+8 −14
Original line number Diff line number Diff line
@@ -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 = {

CRA Git | Maintained and supported by SUSTech CRA and CCSE