[PATCH v1 05/11] itrace: Add functionality to include traces in perf event samples
From: Alexander Shishkin
Date: Thu Feb 06 2014 - 05:51:33 EST
Trace data from itrace PMUs can be used to annotate other perf events
by including it in sample records when PERF_SAMPLE_ITRACE flag is set. In
this case, a PT kernel counter is created for each such event and trace data
is retrieved from it and stored in the perf data stream.
Signed-off-by: Alexander Shishkin <alexander.shishkin@xxxxxxxxxxxxxxx>
---
include/linux/itrace.h | 37 +++++++++
include/linux/perf_event.h | 15 ++++
include/uapi/linux/perf_event.h | 5 +-
kernel/events/core.c | 35 +++++++++
kernel/events/itrace.c | 169 ++++++++++++++++++++++++++++++++++++++--
5 files changed, 252 insertions(+), 9 deletions(-)
diff --git a/include/linux/itrace.h b/include/linux/itrace.h
index 735baaf4..6adbb32 100644
--- a/include/linux/itrace.h
+++ b/include/linux/itrace.h
@@ -54,12 +54,27 @@ struct itrace_pmu {
int (*event_init)(struct perf_event *event);
+ /*
+ * Calculate the size of a sample to be written out
+ */
+ unsigned long (*sample_trace)(struct perf_event *event,
+ struct perf_sample_data *data);
+
+ /*
+ * Write out a trace sample to the given output handle
+ */
+ void (*sample_output)(struct perf_event *event,
+ struct perf_output_handle *handle,
+ struct perf_sample_data *data);
char *name;
};
#define to_itrace_pmu(x) container_of((x), struct itrace_pmu, pmu)
#ifdef CONFIG_PERF_EVENTS
+
+extern int itrace_kernel_event(struct perf_event *event,
+ struct task_struct *task);
extern int itrace_inherit_event(struct perf_event *event,
struct task_struct *task);
extern void itrace_lost_data(struct perf_event *event, u64 offset);
@@ -72,7 +87,17 @@ extern void itrace_wake_up(struct perf_event *event);
extern bool is_itrace_event(struct perf_event *event);
+extern int itrace_sampler_init(struct perf_event *event,
+ struct task_struct *task);
+extern void itrace_sampler_fini(struct perf_event *event);
+extern unsigned long itrace_sampler_trace(struct perf_event *event,
+ struct perf_sample_data *data);
+extern void itrace_sampler_output(struct perf_event *event,
+ struct perf_output_handle *handle,
+ struct perf_sample_data *data);
#else
+static int itrace_kernel_event(struct perf_event *event,
+ struct task_struct *task) { return 0; }
static int itrace_inherit_event(struct perf_event *event,
struct task_struct *task) { return 0; }
static inline void
@@ -84,6 +109,18 @@ itrace_event_installable(struct perf_event *event,
struct perf_event_context *ctx) { return -EINVAL; }
static inline void itrace_wake_up(struct perf_event *event) {}
static inline bool is_itrace_event(struct perf_event *event) { return false; }
+
+static inline int itrace_sampler_init(struct perf_event *event,
+ struct task_struct *task) {}
+static inline void
+itrace_sampler_fini(struct perf_event *event) {}
+static inline unsigned long
+itrace_sampler_trace(struct perf_event *event,
+ struct perf_sample_data *data) { return 0; }
+static inline void
+itrace_sampler_output(struct perf_event *event,
+ struct perf_output_handle *handle,
+ struct perf_sample_data *data) {}
#endif
#endif /* _LINUX_PERF_EVENT_H */
diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index b0147e0..11eb133 100644
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -83,6 +83,12 @@ struct perf_regs_user {
struct pt_regs *regs;
};
+struct perf_trace_record {
+ u64 size;
+ unsigned long from;
+ unsigned long to;
+};
+
struct task_struct;
/*
@@ -97,6 +103,11 @@ struct hw_perf_event_extra {
struct event_constraint;
+enum perf_itrace_counter_type {
+ PERF_ITRACE_USER = BIT(1),
+ PERF_ITRACE_SAMPLING = BIT(2),
+};
+
/**
* struct hw_perf_event - performance event hardware details:
*/
@@ -129,6 +140,7 @@ struct hw_perf_event {
struct { /* itrace */
struct file *itrace_file;
struct task_struct *itrace_target;
+ unsigned int counter_type;
};
#ifdef CONFIG_HAVE_HW_BREAKPOINT
struct { /* breakpoint */
@@ -434,6 +446,7 @@ struct perf_event {
perf_overflow_handler_t overflow_handler;
void *overflow_handler_context;
+ struct perf_event *trace_event;
#ifdef CONFIG_EVENT_TRACING
struct ftrace_event_call *tp_event;
struct event_filter *filter;
@@ -591,6 +604,7 @@ struct perf_sample_data {
union perf_mem_data_src data_src;
struct perf_callchain_entry *callchain;
struct perf_raw_record *raw;
+ struct perf_trace_record trace;
struct perf_branch_stack *br_stack;
struct perf_regs_user regs_user;
u64 stack_user_size;
@@ -611,6 +625,7 @@ static inline void perf_sample_data_init(struct perf_sample_data *data,
data->period = period;
data->regs_user.abi = PERF_SAMPLE_REGS_ABI_NONE;
data->regs_user.regs = NULL;
+ data->trace.from = data->trace.to = data->trace.size = 0;
data->stack_user_size = 0;
data->weight = 0;
data->data_src.val = 0;
diff --git a/include/uapi/linux/perf_event.h b/include/uapi/linux/perf_event.h
index 2dd57db..a06cf4b 100644
--- a/include/uapi/linux/perf_event.h
+++ b/include/uapi/linux/perf_event.h
@@ -137,8 +137,9 @@ enum perf_event_sample_format {
PERF_SAMPLE_DATA_SRC = 1U << 15,
PERF_SAMPLE_IDENTIFIER = 1U << 16,
PERF_SAMPLE_TRANSACTION = 1U << 17,
+ PERF_SAMPLE_ITRACE = 1U << 18,
- PERF_SAMPLE_MAX = 1U << 18, /* non-ABI */
+ PERF_SAMPLE_MAX = 1U << 19, /* non-ABI */
};
/*
@@ -689,6 +690,8 @@ enum perf_event_type {
* { u64 weight; } && PERF_SAMPLE_WEIGHT
* { u64 data_src; } && PERF_SAMPLE_DATA_SRC
* { u64 transaction; } && PERF_SAMPLE_TRANSACTION
+ * { u64 size;
+ * char data[size]; } && PERF_SAMPLE_ITRACE
* };
*/
PERF_RECORD_SAMPLE = 9,
diff --git a/kernel/events/core.c b/kernel/events/core.c
index ff6e286..e1388a5 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -1576,6 +1576,9 @@ void perf_event_disable(struct perf_event *event)
struct perf_event_context *ctx = event->ctx;
struct task_struct *task = ctx->task;
+ if (event->trace_event)
+ perf_event_disable(event->trace_event);
+
if (!task) {
/*
* Disable the event on the cpu that it's on
@@ -2070,6 +2073,8 @@ void perf_event_enable(struct perf_event *event)
struct perf_event_context *ctx = event->ctx;
struct task_struct *task = ctx->task;
+ if (event->trace_event)
+ perf_event_enable(event->trace_event);
if (!task) {
/*
* Enable the event on the cpu that it's on
@@ -3209,6 +3214,8 @@ static void unaccount_event(struct perf_event *event)
static_key_slow_dec_deferred(&perf_sched_events);
if (has_branch_stack(event))
static_key_slow_dec_deferred(&perf_sched_events);
+ if ((event->attr.sample_type & PERF_SAMPLE_ITRACE) && event->trace_event)
+ itrace_sampler_fini(event);
unaccount_event_cpu(event, event->cpu);
}
@@ -4664,6 +4671,13 @@ void perf_output_sample(struct perf_output_handle *handle,
if (sample_type & PERF_SAMPLE_TRANSACTION)
perf_output_put(handle, data->txn);
+ if (sample_type & PERF_SAMPLE_ITRACE) {
+ perf_output_put(handle, data->trace.size);
+
+ if (data->trace.size)
+ itrace_sampler_output(event, handle, data);
+ }
+
if (!event->attr.watermark) {
int wakeup_events = event->attr.wakeup_events;
@@ -4771,6 +4785,14 @@ void perf_prepare_sample(struct perf_event_header *header,
data->stack_user_size = stack_size;
header->size += size;
}
+
+ if (sample_type & PERF_SAMPLE_ITRACE) {
+ u64 size = sizeof(u64);
+
+ size += itrace_sampler_trace(event, data);
+
+ header->size += size;
+ }
}
static void perf_event_output(struct perf_event *event,
@@ -6795,6 +6817,15 @@ perf_event_alloc(struct perf_event_attr *attr, int cpu,
if (err)
goto err_pmu;
}
+
+ if (event->attr.sample_type & PERF_SAMPLE_ITRACE) {
+ err = itrace_sampler_init(event, task);
+ if (err) {
+ /* XXX: either clean up callchain buffers too
+ or forbid them to go together */
+ goto err_pmu;
+ }
+ }
}
return event;
@@ -7369,6 +7400,10 @@ perf_event_create_kernel_counter(struct perf_event_attr *attr, int cpu,
account_event(event);
+ err = itrace_kernel_event(event, task);
+ if (err)
+ goto err_free;
+
ctx = find_get_context(event->pmu, task, cpu);
if (IS_ERR(ctx)) {
err = PTR_ERR(ctx);
diff --git a/kernel/events/itrace.c b/kernel/events/itrace.c
index ec26373..f003530 100644
--- a/kernel/events/itrace.c
+++ b/kernel/events/itrace.c
@@ -89,6 +89,22 @@ bool is_itrace_event(struct perf_event *event)
return !!itrace_pmu_find(event->attr.type);
}
+static void itrace_event_destroy(struct perf_event *event)
+{
+ struct ring_buffer *rb = event->rb[PERF_RB_ITRACE];
+
+ if (!rb)
+ return;
+
+ if (event->hw.counter_type != PERF_ITRACE_USER) {
+ atomic_dec(&rb->mmap_count);
+ atomic_dec(&event->mmap_count[PERF_RB_ITRACE]);
+ ring_buffer_detach(event, rb);
+ rcu_assign_pointer(event->rb[PERF_RB_ITRACE], NULL);
+ ring_buffer_put(rb); /* should be last */
+ }
+}
+
int itrace_event_installable(struct perf_event *event,
struct perf_event_context *ctx)
{
@@ -115,8 +131,16 @@ int itrace_event_installable(struct perf_event *event,
static int itrace_event_init(struct perf_event *event)
{
struct itrace_pmu *ipmu = to_itrace_pmu(event->pmu);
+ int ret;
- return ipmu->event_init(event);
+ ret = ipmu->event_init(event);
+ if (ret)
+ return ret;
+
+ event->destroy = itrace_event_destroy;
+ event->hw.counter_type = PERF_ITRACE_USER;
+
+ return 0;
}
static unsigned long itrace_rb_get_size(int nr_pages)
@@ -214,9 +238,16 @@ out:
mutex_unlock(&event->mmap_mutex);
}
+static size_t roundup_buffer_size(u64 size)
+{
+ return 1ul << (__get_order(size) + PAGE_SHIFT);
+}
+
int itrace_inherit_event(struct perf_event *event, struct task_struct *task)
{
+ size_t size = event->attr.itrace_sample_size;
struct perf_event *parent = event->parent;
+ struct ring_buffer *rb;
struct itrace_pmu *ipmu;
if (!is_itrace_event(event))
@@ -224,14 +255,59 @@ int itrace_inherit_event(struct perf_event *event, struct task_struct *task)
ipmu = to_itrace_pmu(event->pmu);
- /*
- * inherited user's counters should inherit buffers IF
- * they aren't cpu==-1
- */
- if (parent->cpu == -1)
- return -EINVAL;
+ if (parent->hw.counter_type == PERF_ITRACE_USER) {
+ /*
+ * inherited user's counters should inherit buffers IF
+ * they aren't cpu==-1
+ */
+ if (parent->cpu == -1)
+ return -EINVAL;
+
+ itrace_set_output(event, parent);
+ return 0;
+ }
+
+ event->hw.counter_type = parent->hw.counter_type;
+
+ size = roundup_buffer_size(size);
+ rb = rb_alloc(event, size >> PAGE_SHIFT, 0, event->cpu, 0,
+ &itrace_rb_ops);
+ if (!rb)
+ return -ENOMEM;
+
+ ring_buffer_attach(event, rb);
+ rcu_assign_pointer(event->rb[PERF_RB_ITRACE], rb);
+ atomic_set(&rb->mmap_count, 1);
+ atomic_set(&event->mmap_count[PERF_RB_ITRACE], 1);
+
+ return 0;
+}
+
+int itrace_kernel_event(struct perf_event *event, struct task_struct *task)
+{
+ struct itrace_pmu *ipmu;
+ struct ring_buffer *rb;
+ size_t size;
+
+ if (!is_itrace_event(event))
+ return 0;
- itrace_set_output(event, parent);
+ ipmu = to_itrace_pmu(event->pmu);
+
+ if (!event->attr.itrace_sample_size)
+ return 0;
+
+ size = roundup_buffer_size(event->attr.itrace_sample_size);
+
+ rb = rb_alloc(event, size >> PAGE_SHIFT, 0, event->cpu, 0,
+ &itrace_rb_ops);
+ if (!rb)
+ return -ENOMEM;
+
+ ring_buffer_attach(event, rb);
+ rcu_assign_pointer(event->rb[PERF_RB_ITRACE], rb);
+ atomic_set(&rb->mmap_count, 1);
+ atomic_set(&event->mmap_count[PERF_RB_ITRACE], 1);
return 0;
}
@@ -269,3 +345,80 @@ int itrace_pmu_register(struct itrace_pmu *ipmu)
return ret;
}
+
+/*
+ * Trace sample annotation
+ * For events that have attr.sample_type & PERF_SAMPLE_ITRACE, perf calls here
+ * to configure and obtain itrace samples.
+ */
+
+int itrace_sampler_init(struct perf_event *event, struct task_struct *task)
+{
+ struct perf_event_attr attr;
+ struct perf_event *tevt;
+ struct itrace_pmu *ipmu;
+
+ ipmu = itrace_pmu_find(event->attr.itrace_sample_type);
+ if (!ipmu || !ipmu->sample_trace || !ipmu->sample_output)
+ return -ENOTSUPP;
+
+ memset(&attr, 0, sizeof(attr));
+ attr.type = ipmu->pmu.type;
+ attr.config = 0;
+ attr.sample_type = 0;
+ attr.exclude_user = event->attr.exclude_user;
+ attr.exclude_kernel = event->attr.exclude_kernel;
+ attr.itrace_sample_size = event->attr.itrace_sample_size;
+ attr.itrace_config = event->attr.itrace_config;
+
+ tevt = perf_event_create_kernel_counter(&attr, event->cpu, task, NULL, NULL);
+ if (IS_ERR(tevt))
+ return PTR_ERR(tevt);
+
+ if (!itrace_priv(tevt)) {
+ perf_event_release_kernel(tevt);
+ return -EINVAL;
+ }
+
+ event->trace_event = tevt;
+ tevt->hw.counter_type = PERF_ITRACE_SAMPLING;
+ if (event->state != PERF_EVENT_STATE_OFF)
+ perf_event_enable(event->trace_event);
+
+ return 0;
+}
+
+void itrace_sampler_fini(struct perf_event *event)
+{
+ struct perf_event *tevt = event->trace_event;
+
+ perf_event_release_kernel(tevt);
+ event->trace_event = NULL;
+}
+
+unsigned long itrace_sampler_trace(struct perf_event *event,
+ struct perf_sample_data *data)
+{
+ struct perf_event *tevt = event->trace_event;
+ struct itrace_pmu *ipmu;
+
+ if (!tevt)
+ return 0;
+
+ ipmu = to_itrace_pmu(tevt->pmu);
+ return ipmu->sample_trace(tevt, data);
+}
+
+void itrace_sampler_output(struct perf_event *event,
+ struct perf_output_handle *handle,
+ struct perf_sample_data *data)
+{
+ struct perf_event *tevt = event->trace_event;
+ struct itrace_pmu *ipmu;
+
+ if (!tevt || !data->trace.size)
+ return;
+
+ ipmu = to_itrace_pmu(tevt->pmu);
+ ipmu->sample_output(tevt, handle, data);
+}
--
1.8.5.2
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/