2009-03-19 15:26:15 -04:00
|
|
|
/*
|
2010-03-04 23:35:37 -05:00
|
|
|
* trace event based perf event profiling/tracing
|
2009-03-19 15:26:15 -04:00
|
|
|
*
|
|
|
|
* Copyright (C) 2009 Red Hat Inc, Peter Zijlstra <pzijlstr@redhat.com>
|
2010-03-03 01:16:16 -05:00
|
|
|
* Copyright (C) 2009-2010 Frederic Weisbecker <fweisbec@gmail.com>
|
2009-03-19 15:26:15 -04:00
|
|
|
*/
|
|
|
|
|
2009-08-24 00:19:47 -04:00
|
|
|
#include <linux/module.h>
|
2010-01-27 20:32:29 -05:00
|
|
|
#include <linux/kprobes.h>
|
2009-03-19 15:26:15 -04:00
|
|
|
#include "trace.h"
|
|
|
|
|
2010-08-10 23:47:59 -04:00
|
|
|
static char __percpu *perf_trace_buf[PERF_NR_CONTEXTS];
|
2009-09-18 00:10:28 -04:00
|
|
|
|
2010-03-22 19:08:59 -04:00
|
|
|
/*
|
|
|
|
* Force it to be aligned to unsigned long to avoid misaligned accesses
|
|
|
|
* suprises
|
|
|
|
*/
|
|
|
|
typedef typeof(unsigned long [PERF_MAX_TRACE_SIZE / sizeof(unsigned long)])
|
|
|
|
perf_trace_t;
|
2009-11-21 23:26:55 -05:00
|
|
|
|
2009-09-18 00:10:28 -04:00
|
|
|
/* Count the events in use (per event id, not per instance) */
|
2010-03-04 23:35:37 -05:00
|
|
|
static int total_ref_count;
|
2009-09-18 00:10:28 -04:00
|
|
|
|
2010-05-19 08:02:22 -04:00
|
|
|
static int perf_trace_event_init(struct ftrace_event_call *tp_event,
|
|
|
|
struct perf_event *p_event)
|
2009-09-17 18:54:43 -04:00
|
|
|
{
|
2010-08-10 23:47:59 -04:00
|
|
|
struct hlist_head __percpu *list;
|
2009-09-18 00:10:28 -04:00
|
|
|
int ret = -ENOMEM;
|
2010-05-19 08:02:22 -04:00
|
|
|
int cpu;
|
2009-09-18 00:10:28 -04:00
|
|
|
|
2010-05-19 08:02:22 -04:00
|
|
|
p_event->tp_event = tp_event;
|
|
|
|
if (tp_event->perf_refcount++ > 0)
|
2009-09-17 18:54:43 -04:00
|
|
|
return 0;
|
|
|
|
|
2010-05-19 08:02:22 -04:00
|
|
|
list = alloc_percpu(struct hlist_head);
|
|
|
|
if (!list)
|
|
|
|
goto fail;
|
|
|
|
|
|
|
|
for_each_possible_cpu(cpu)
|
|
|
|
INIT_HLIST_HEAD(per_cpu_ptr(list, cpu));
|
2009-09-18 00:10:28 -04:00
|
|
|
|
2010-05-19 08:02:22 -04:00
|
|
|
tp_event->perf_events = list;
|
2009-09-17 18:54:43 -04:00
|
|
|
|
2010-03-04 23:35:37 -05:00
|
|
|
if (!total_ref_count) {
|
2010-08-10 23:47:59 -04:00
|
|
|
char __percpu *buf;
|
2010-05-19 04:52:27 -04:00
|
|
|
int i;
|
2009-09-18 00:10:28 -04:00
|
|
|
|
2010-08-14 14:45:13 -04:00
|
|
|
for (i = 0; i < PERF_NR_CONTEXTS; i++) {
|
2010-08-10 23:47:59 -04:00
|
|
|
buf = (char __percpu *)alloc_percpu(perf_trace_t);
|
2010-05-19 04:52:27 -04:00
|
|
|
if (!buf)
|
2010-05-19 08:02:22 -04:00
|
|
|
goto fail;
|
2009-09-18 00:10:28 -04:00
|
|
|
|
2010-05-19 08:02:22 -04:00
|
|
|
perf_trace_buf[i] = buf;
|
2010-05-19 04:52:27 -04:00
|
|
|
}
|
2009-09-18 00:10:28 -04:00
|
|
|
}
|
|
|
|
|
2010-06-08 11:22:06 -04:00
|
|
|
ret = tp_event->class->reg(tp_event, TRACE_REG_PERF_REGISTER);
|
2010-05-19 08:02:22 -04:00
|
|
|
if (ret)
|
|
|
|
goto fail;
|
2009-09-18 00:10:28 -04:00
|
|
|
|
2010-05-19 08:02:22 -04:00
|
|
|
total_ref_count++;
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
fail:
|
2010-03-04 23:35:37 -05:00
|
|
|
if (!total_ref_count) {
|
2010-05-19 04:52:27 -04:00
|
|
|
int i;
|
|
|
|
|
2010-08-14 14:45:13 -04:00
|
|
|
for (i = 0; i < PERF_NR_CONTEXTS; i++) {
|
2010-05-19 04:52:27 -04:00
|
|
|
free_percpu(perf_trace_buf[i]);
|
|
|
|
perf_trace_buf[i] = NULL;
|
|
|
|
}
|
2009-10-03 08:55:18 -04:00
|
|
|
}
|
2010-05-19 08:02:22 -04:00
|
|
|
|
|
|
|
if (!--tp_event->perf_refcount) {
|
|
|
|
free_percpu(tp_event->perf_events);
|
|
|
|
tp_event->perf_events = NULL;
|
2009-10-03 08:55:18 -04:00
|
|
|
}
|
2009-09-18 00:10:28 -04:00
|
|
|
|
|
|
|
return ret;
|
2009-09-17 18:54:43 -04:00
|
|
|
}
|
|
|
|
|
2010-05-19 08:02:22 -04:00
|
|
|
int perf_trace_init(struct perf_event *p_event)
|
2009-03-19 15:26:15 -04:00
|
|
|
{
|
2010-05-19 08:02:22 -04:00
|
|
|
struct ftrace_event_call *tp_event;
|
|
|
|
int event_id = p_event->attr.config;
|
2009-05-05 22:33:45 -04:00
|
|
|
int ret = -EINVAL;
|
2009-03-19 15:26:15 -04:00
|
|
|
|
2009-05-05 22:33:45 -04:00
|
|
|
mutex_lock(&event_mutex);
|
2010-05-19 08:02:22 -04:00
|
|
|
list_for_each_entry(tp_event, &ftrace_events, list) {
|
2010-05-21 11:49:57 -04:00
|
|
|
if (tp_event->event.type == event_id &&
|
2010-06-08 11:22:06 -04:00
|
|
|
tp_event->class && tp_event->class->reg &&
|
2010-05-19 08:02:22 -04:00
|
|
|
try_module_get(tp_event->mod)) {
|
|
|
|
ret = perf_trace_event_init(tp_event, p_event);
|
2010-09-01 06:58:43 -04:00
|
|
|
if (ret)
|
|
|
|
module_put(tp_event->mod);
|
2009-05-05 22:33:45 -04:00
|
|
|
break;
|
|
|
|
}
|
2009-03-19 15:26:15 -04:00
|
|
|
}
|
2009-05-05 22:33:45 -04:00
|
|
|
mutex_unlock(&event_mutex);
|
2009-03-19 15:26:15 -04:00
|
|
|
|
2009-05-05 22:33:45 -04:00
|
|
|
return ret;
|
2009-03-19 15:26:15 -04:00
|
|
|
}
|
|
|
|
|
perf: Rework the PMU methods
Replace pmu::{enable,disable,start,stop,unthrottle} with
pmu::{add,del,start,stop}, all of which take a flags argument.
The new interface extends the capability to stop a counter while
keeping it scheduled on the PMU. We replace the throttled state with
the generic stopped state.
This also allows us to efficiently stop/start counters over certain
code paths (like IRQ handlers).
It also allows scheduling a counter without it starting, allowing for
a generic frozen state (useful for rotating stopped counters).
The stopped state is implemented in two different ways, depending on
how the architecture implemented the throttled state:
1) We disable the counter:
a) the pmu has per-counter enable bits, we flip that
b) we program a NOP event, preserving the counter state
2) We store the counter state and ignore all read/overflow events
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: paulus <paulus@samba.org>
Cc: stephane eranian <eranian@googlemail.com>
Cc: Robert Richter <robert.richter@amd.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Cyrill Gorcunov <gorcunov@gmail.com>
Cc: Lin Ming <ming.m.lin@intel.com>
Cc: Yanmin <yanmin_zhang@linux.intel.com>
Cc: Deng-Cheng Zhu <dengcheng.zhu@gmail.com>
Cc: David Miller <davem@davemloft.net>
Cc: Michael Cree <mcree@orcon.net.nz>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-06-16 08:37:10 -04:00
|
|
|
int perf_trace_add(struct perf_event *p_event, int flags)
|
2009-09-17 18:54:43 -04:00
|
|
|
{
|
2010-05-19 08:02:22 -04:00
|
|
|
struct ftrace_event_call *tp_event = p_event->tp_event;
|
2010-08-10 23:47:59 -04:00
|
|
|
struct hlist_head __percpu *pcpu_list;
|
2010-05-19 08:02:22 -04:00
|
|
|
struct hlist_head *list;
|
2009-09-18 00:10:28 -04:00
|
|
|
|
2010-08-10 23:47:59 -04:00
|
|
|
pcpu_list = tp_event->perf_events;
|
|
|
|
if (WARN_ON_ONCE(!pcpu_list))
|
2010-05-19 08:02:22 -04:00
|
|
|
return -EINVAL;
|
2009-09-18 00:10:28 -04:00
|
|
|
|
perf: Rework the PMU methods
Replace pmu::{enable,disable,start,stop,unthrottle} with
pmu::{add,del,start,stop}, all of which take a flags argument.
The new interface extends the capability to stop a counter while
keeping it scheduled on the PMU. We replace the throttled state with
the generic stopped state.
This also allows us to efficiently stop/start counters over certain
code paths (like IRQ handlers).
It also allows scheduling a counter without it starting, allowing for
a generic frozen state (useful for rotating stopped counters).
The stopped state is implemented in two different ways, depending on
how the architecture implemented the throttled state:
1) We disable the counter:
a) the pmu has per-counter enable bits, we flip that
b) we program a NOP event, preserving the counter state
2) We store the counter state and ignore all read/overflow events
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: paulus <paulus@samba.org>
Cc: stephane eranian <eranian@googlemail.com>
Cc: Robert Richter <robert.richter@amd.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Cyrill Gorcunov <gorcunov@gmail.com>
Cc: Lin Ming <ming.m.lin@intel.com>
Cc: Yanmin <yanmin_zhang@linux.intel.com>
Cc: Deng-Cheng Zhu <dengcheng.zhu@gmail.com>
Cc: David Miller <davem@davemloft.net>
Cc: Michael Cree <mcree@orcon.net.nz>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-06-16 08:37:10 -04:00
|
|
|
if (!(flags & PERF_EF_START))
|
|
|
|
p_event->hw.state = PERF_HES_STOPPED;
|
|
|
|
|
2010-08-10 23:47:59 -04:00
|
|
|
list = this_cpu_ptr(pcpu_list);
|
2010-05-19 08:02:22 -04:00
|
|
|
hlist_add_head_rcu(&p_event->hlist_entry, list);
|
2009-09-18 00:10:28 -04:00
|
|
|
|
2010-05-19 08:02:22 -04:00
|
|
|
return 0;
|
|
|
|
}
|
2009-09-18 00:10:28 -04:00
|
|
|
|
perf: Rework the PMU methods
Replace pmu::{enable,disable,start,stop,unthrottle} with
pmu::{add,del,start,stop}, all of which take a flags argument.
The new interface extends the capability to stop a counter while
keeping it scheduled on the PMU. We replace the throttled state with
the generic stopped state.
This also allows us to efficiently stop/start counters over certain
code paths (like IRQ handlers).
It also allows scheduling a counter without it starting, allowing for
a generic frozen state (useful for rotating stopped counters).
The stopped state is implemented in two different ways, depending on
how the architecture implemented the throttled state:
1) We disable the counter:
a) the pmu has per-counter enable bits, we flip that
b) we program a NOP event, preserving the counter state
2) We store the counter state and ignore all read/overflow events
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: paulus <paulus@samba.org>
Cc: stephane eranian <eranian@googlemail.com>
Cc: Robert Richter <robert.richter@amd.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Paul Mundt <lethal@linux-sh.org>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Cyrill Gorcunov <gorcunov@gmail.com>
Cc: Lin Ming <ming.m.lin@intel.com>
Cc: Yanmin <yanmin_zhang@linux.intel.com>
Cc: Deng-Cheng Zhu <dengcheng.zhu@gmail.com>
Cc: David Miller <davem@davemloft.net>
Cc: Michael Cree <mcree@orcon.net.nz>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
2010-06-16 08:37:10 -04:00
|
|
|
void perf_trace_del(struct perf_event *p_event, int flags)
|
2010-05-19 08:02:22 -04:00
|
|
|
{
|
|
|
|
hlist_del_rcu(&p_event->hlist_entry);
|
2009-09-17 18:54:43 -04:00
|
|
|
}
|
|
|
|
|
2010-05-19 08:02:22 -04:00
|
|
|
void perf_trace_destroy(struct perf_event *p_event)
|
2009-03-19 15:26:15 -04:00
|
|
|
{
|
2010-05-19 08:02:22 -04:00
|
|
|
struct ftrace_event_call *tp_event = p_event->tp_event;
|
|
|
|
int i;
|
2009-03-19 15:26:15 -04:00
|
|
|
|
2010-05-21 10:22:33 -04:00
|
|
|
mutex_lock(&event_mutex);
|
2010-05-19 08:02:22 -04:00
|
|
|
if (--tp_event->perf_refcount > 0)
|
2010-05-21 10:22:33 -04:00
|
|
|
goto out;
|
2010-05-19 08:02:22 -04:00
|
|
|
|
2010-06-08 11:22:06 -04:00
|
|
|
tp_event->class->reg(tp_event, TRACE_REG_PERF_UNREGISTER);
|
2010-05-19 08:02:22 -04:00
|
|
|
|
2010-05-21 06:31:09 -04:00
|
|
|
/*
|
2010-07-20 11:29:54 -04:00
|
|
|
* Ensure our callback won't be called anymore. The buffers
|
|
|
|
* will be freed after that.
|
2010-05-21 06:31:09 -04:00
|
|
|
*/
|
2010-07-20 11:29:54 -04:00
|
|
|
tracepoint_synchronize_unregister();
|
2010-05-21 06:31:09 -04:00
|
|
|
|
2010-05-19 08:02:22 -04:00
|
|
|
free_percpu(tp_event->perf_events);
|
|
|
|
tp_event->perf_events = NULL;
|
|
|
|
|
|
|
|
if (!--total_ref_count) {
|
2010-08-14 14:45:13 -04:00
|
|
|
for (i = 0; i < PERF_NR_CONTEXTS; i++) {
|
2010-05-19 08:02:22 -04:00
|
|
|
free_percpu(perf_trace_buf[i]);
|
|
|
|
perf_trace_buf[i] = NULL;
|
2009-05-05 22:33:45 -04:00
|
|
|
}
|
2009-03-19 15:26:15 -04:00
|
|
|
}
|
2010-05-21 10:22:33 -04:00
|
|
|
out:
|
2010-09-01 06:58:43 -04:00
|
|
|
module_put(tp_event->mod);
|
2010-05-21 10:22:33 -04:00
|
|
|
mutex_unlock(&event_mutex);
|
2009-03-19 15:26:15 -04:00
|
|
|
}
|
2010-01-27 20:32:29 -05:00
|
|
|
|
2010-03-04 23:35:37 -05:00
|
|
|
__kprobes void *perf_trace_buf_prepare(int size, unsigned short type,
|
2010-05-19 04:52:27 -04:00
|
|
|
struct pt_regs *regs, int *rctxp)
|
2010-01-27 20:32:29 -05:00
|
|
|
{
|
|
|
|
struct trace_entry *entry;
|
2010-05-25 05:02:55 -04:00
|
|
|
unsigned long flags;
|
2010-05-19 08:02:22 -04:00
|
|
|
char *raw_data;
|
2010-05-19 04:52:27 -04:00
|
|
|
int pc;
|
2010-01-27 20:32:29 -05:00
|
|
|
|
2010-03-22 19:08:59 -04:00
|
|
|
BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(unsigned long));
|
|
|
|
|
2010-01-27 20:32:29 -05:00
|
|
|
pc = preempt_count();
|
|
|
|
|
|
|
|
*rctxp = perf_swevent_get_recursion_context();
|
|
|
|
if (*rctxp < 0)
|
2010-05-19 08:02:22 -04:00
|
|
|
return NULL;
|
2010-01-27 20:32:29 -05:00
|
|
|
|
2010-05-21 06:31:09 -04:00
|
|
|
raw_data = this_cpu_ptr(perf_trace_buf[*rctxp]);
|
2010-01-27 20:32:29 -05:00
|
|
|
|
|
|
|
/* zero the dead bytes from align to not leak stack to user */
|
2010-03-22 19:08:59 -04:00
|
|
|
memset(&raw_data[size - sizeof(u64)], 0, sizeof(u64));
|
2010-01-27 20:32:29 -05:00
|
|
|
|
|
|
|
entry = (struct trace_entry *)raw_data;
|
2010-05-25 05:02:55 -04:00
|
|
|
local_save_flags(flags);
|
|
|
|
tracing_generic_entry_update(entry, flags, pc);
|
2010-01-27 20:32:29 -05:00
|
|
|
entry->type = type;
|
|
|
|
|
|
|
|
return raw_data;
|
|
|
|
}
|
2010-03-04 23:35:37 -05:00
|
|
|
EXPORT_SYMBOL_GPL(perf_trace_buf_prepare);
|