#define PERF_PMU_CAP_NO_EXCLUDE 0x80
#define PERF_PMU_CAP_AUX_OUTPUT 0x100
+struct perf_output_handle;
+
/**
* struct pmu - generic performance monitoring unit
*/
*/
void (*free_aux) (void *aux); /* optional */
+ /*
+ * Take a snapshot of the AUX buffer without touching the event
+ * state, so that preempting ->start()/->stop() callbacks does
+ * not interfere with their logic. Called in PMI context.
+ *
+ * Returns the size of AUX data copied to the output handle.
+ *
+ * Optional.
+ */
+ long (*snapshot_aux) (struct perf_event *event,
+ struct perf_output_handle *handle,
+ unsigned long size);
+
/*
* Validate address range filters: make sure the HW supports the
* requested configuration and number of filters; return 0 if the
u32 reserved;
} cpu_entry;
struct perf_callchain_entry *callchain;
+ u64 aux_size;
/*
* regs_user may point to task_pt_regs or to regs_user_copy, depending
const void *buf, unsigned int len);
extern unsigned int perf_output_skip(struct perf_output_handle *handle,
unsigned int len);
+extern long perf_output_copy_aux(struct perf_output_handle *aux_handle,
+ struct perf_output_handle *handle,
+ unsigned long from, unsigned long to);
extern int perf_swevent_get_recursion_context(void);
extern void perf_swevent_put_recursion_context(int rctx);
extern u64 perf_swevent_set_period(struct perf_event *event);
}
}
+static bool perf_need_aux_event(struct perf_event *event)
+{
+ return !!event->attr.aux_output || !!event->attr.aux_sample_size;
+}
+
static int perf_get_aux_event(struct perf_event *event,
struct perf_event *group_leader)
{
if (!group_leader)
return 0;
- if (!perf_aux_output_match(event, group_leader))
+ /*
+ * aux_output and aux_sample_size are mutually exclusive.
+ */
+ if (event->attr.aux_output && event->attr.aux_sample_size)
+ return 0;
+
+ if (event->attr.aux_output &&
+ !perf_aux_output_match(event, group_leader))
+ return 0;
+
+ if (event->attr.aux_sample_size && !group_leader->pmu->snapshot_aux)
return 0;
if (!atomic_long_inc_not_zero(&group_leader->refcount))
}
}
+static unsigned long perf_prepare_sample_aux(struct perf_event *event,
+ struct perf_sample_data *data,
+ size_t size)
+{
+ struct perf_event *sampler = event->aux_event;
+ struct ring_buffer *rb;
+
+ data->aux_size = 0;
+
+ if (!sampler)
+ goto out;
+
+ if (WARN_ON_ONCE(READ_ONCE(sampler->state) != PERF_EVENT_STATE_ACTIVE))
+ goto out;
+
+ if (WARN_ON_ONCE(READ_ONCE(sampler->oncpu) != smp_processor_id()))
+ goto out;
+
+ rb = ring_buffer_get(sampler->parent ? sampler->parent : sampler);
+ if (!rb)
+ goto out;
+
+ /*
+ * If this is an NMI hit inside sampling code, don't take
+ * the sample. See also perf_aux_sample_output().
+ */
+ if (READ_ONCE(rb->aux_in_sampling)) {
+ data->aux_size = 0;
+ } else {
+ size = min_t(size_t, size, perf_aux_size(rb));
+ data->aux_size = ALIGN(size, sizeof(u64));
+ }
+ ring_buffer_put(rb);
+
+out:
+ return data->aux_size;
+}
+
+long perf_pmu_snapshot_aux(struct ring_buffer *rb,
+ struct perf_event *event,
+ struct perf_output_handle *handle,
+ unsigned long size)
+{
+ unsigned long flags;
+ long ret;
+
+ /*
+ * Normal ->start()/->stop() callbacks run in IRQ mode in scheduler
+ * paths. If we start calling them in NMI context, they may race with
+ * the IRQ ones, that is, for example, re-starting an event that's just
+ * been stopped, which is why we're using a separate callback that
+ * doesn't change the event state.
+ *
+ * IRQs need to be disabled to prevent IPIs from racing with us.
+ */
+ local_irq_save(flags);
+ /*
+ * Guard against NMI hits inside the critical section;
+ * see also perf_prepare_sample_aux().
+ */
+ WRITE_ONCE(rb->aux_in_sampling, 1);
+ barrier();
+
+ ret = event->pmu->snapshot_aux(event, handle, size);
+
+ barrier();
+ WRITE_ONCE(rb->aux_in_sampling, 0);
+ local_irq_restore(flags);
+
+ return ret;
+}
+
+static void perf_aux_sample_output(struct perf_event *event,
+ struct perf_output_handle *handle,
+ struct perf_sample_data *data)
+{
+ struct perf_event *sampler = event->aux_event;
+ unsigned long pad;
+ struct ring_buffer *rb;
+ long size;
+
+ if (WARN_ON_ONCE(!sampler || !data->aux_size))
+ return;
+
+ rb = ring_buffer_get(sampler->parent ? sampler->parent : sampler);
+ if (!rb)
+ return;
+
+ size = perf_pmu_snapshot_aux(rb, sampler, handle, data->aux_size);
+
+ /*
+ * An error here means that perf_output_copy() failed (returned a
+ * non-zero surplus that it didn't copy), which in its current
+ * enlightened implementation is not possible. If that changes, we'd
+ * like to know.
+ */
+ if (WARN_ON_ONCE(size < 0))
+ goto out_put;
+
+ /*
+ * The pad comes from ALIGN()ing data->aux_size up to u64 in
+ * perf_prepare_sample_aux(), so should not be more than that.
+ */
+ pad = data->aux_size - size;
+ if (WARN_ON_ONCE(pad >= sizeof(u64)))
+ pad = 8;
+
+ if (pad) {
+ u64 zero = 0;
+ perf_output_copy(handle, &zero, pad);
+ }
+
+out_put:
+ ring_buffer_put(rb);
+}
+
static void __perf_event_header__init_id(struct perf_event_header *header,
struct perf_sample_data *data,
struct perf_event *event)
if (sample_type & PERF_SAMPLE_PHYS_ADDR)
perf_output_put(handle, data->phys_addr);
+ if (sample_type & PERF_SAMPLE_AUX) {
+ perf_output_put(handle, data->aux_size);
+
+ if (data->aux_size)
+ perf_aux_sample_output(event, handle, data);
+ }
+
if (!event->attr.watermark) {
int wakeup_events = event->attr.wakeup_events;
if (sample_type & PERF_SAMPLE_PHYS_ADDR)
data->phys_addr = perf_virt_to_phys(data->addr);
+
+ if (sample_type & PERF_SAMPLE_AUX) {
+ u64 size;
+
+ header->size += sizeof(u64); /* size */
+
+ /*
+ * Given the 16bit nature of header::size, an AUX sample can
+ * easily overflow it, what with all the preceding sample bits.
+ * Make sure this doesn't happen by using up to U16_MAX bytes
+ * per sample in total (rounded down to 8 byte boundary).
+ */
+ size = min_t(size_t, U16_MAX - header->size,
+ event->attr.aux_sample_size);
+ size = rounddown(size, 8);
+ size = perf_prepare_sample_aux(event, data, size);
+
+ WARN_ON_ONCE(size + header->size > U16_MAX);
+ header->size += size;
+ }
+ /*
+ * If you're adding more sample types here, you likely need to do
+ * something about the overflowing header::size, like repurpose the
+ * lowest 3 bits of size, which should be always zero at the moment.
+ * This raises a more important question, do we really need 512k sized
+ * samples and why, so good argumentation is in order for whatever you
+ * do here next.
+ */
+ WARN_ON_ONCE(header->size & 7);
}
static __always_inline int
attr->size = size;
- if (attr->__reserved_1 || attr->__reserved_2)
+ if (attr->__reserved_1 || attr->__reserved_2 || attr->__reserved_3)
return -EINVAL;
if (attr->sample_type & ~(PERF_SAMPLE_MAX-1))
}
}
- if (event->attr.aux_output && !perf_get_aux_event(event, group_leader))
+ if (perf_need_aux_event(event) && !perf_get_aux_event(event, group_leader))
goto err_locked;
/*
projects / users / hch / configfs.git / commitdiff