2005-04-16 22:20:36 +00:00
|
|
|
/**
|
|
|
|
* @file buffer_sync.c
|
|
|
|
*
|
|
|
|
* @remark Copyright 2002 OProfile authors
|
|
|
|
* @remark Read the file COPYING
|
|
|
|
*
|
|
|
|
* @author John Levon <levon@movementarian.org>
|
2008-07-22 19:08:54 +00:00
|
|
|
* @author Barry Kasindorf
|
2005-04-16 22:20:36 +00:00
|
|
|
*
|
|
|
|
* This is the core of the buffer management. Each
|
|
|
|
* CPU buffer is processed and entered into the
|
|
|
|
* global event buffer. Such processing is necessary
|
|
|
|
* in several circumstances, mentioned below.
|
|
|
|
*
|
|
|
|
* The processing does the job of converting the
|
|
|
|
* transitory EIP value into a persistent dentry/offset
|
|
|
|
* value that the profiler can record at its leisure.
|
|
|
|
*
|
|
|
|
* See fs/dcookies.c for a description of the dentry/offset
|
|
|
|
* objects.
|
|
|
|
*/
|
|
|
|
|
|
|
|
#include <linux/mm.h>
|
|
|
|
#include <linux/workqueue.h>
|
|
|
|
#include <linux/notifier.h>
|
|
|
|
#include <linux/dcookies.h>
|
|
|
|
#include <linux/profile.h>
|
|
|
|
#include <linux/module.h>
|
|
|
|
#include <linux/fs.h>
|
2007-07-20 19:39:53 +00:00
|
|
|
#include <linux/oprofile.h>
|
Detach sched.h from mm.h
First thing mm.h does is including sched.h solely for can_do_mlock() inline
function which has "current" dereference inside. By dealing with can_do_mlock()
mm.h can be detached from sched.h which is good. See below, why.
This patch
a) removes unconditional inclusion of sched.h from mm.h
b) makes can_do_mlock() normal function in mm/mlock.c
c) exports can_do_mlock() to not break compilation
d) adds sched.h inclusions back to files that were getting it indirectly.
e) adds less bloated headers to some files (asm/signal.h, jiffies.h) that were
getting them indirectly
Net result is:
a) mm.h users would get less code to open, read, preprocess, parse, ... if
they don't need sched.h
b) sched.h stops being dependency for significant number of files:
on x86_64 allmodconfig touching sched.h results in recompile of 4083 files,
after patch it's only 3744 (-8.3%).
Cross-compile tested on
all arm defconfigs, all mips defconfigs, all powerpc defconfigs,
alpha alpha-up
arm
i386 i386-up i386-defconfig i386-allnoconfig
ia64 ia64-up
m68k
mips
parisc parisc-up
powerpc powerpc-up
s390 s390-up
sparc sparc-up
sparc64 sparc64-up
um-x86_64
x86_64 x86_64-up x86_64-defconfig x86_64-allnoconfig
as well as my two usual configs.
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-05-20 21:22:52 +00:00
|
|
|
#include <linux/sched.h>
|
2007-07-20 19:39:53 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
#include "oprofile_stats.h"
|
|
|
|
#include "event_buffer.h"
|
|
|
|
#include "cpu_buffer.h"
|
|
|
|
#include "buffer_sync.h"
|
2008-07-22 19:08:51 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
static LIST_HEAD(dying_tasks);
|
|
|
|
static LIST_HEAD(dead_tasks);
|
|
|
|
static cpumask_t marked_cpus = CPU_MASK_NONE;
|
|
|
|
static DEFINE_SPINLOCK(task_mortuary);
|
|
|
|
static void process_task_mortuary(void);
|
|
|
|
|
|
|
|
/* Take ownership of the task struct and place it on the
|
|
|
|
* list for processing. Only after two full buffer syncs
|
|
|
|
* does the task eventually get freed, because by then
|
|
|
|
* we are sure we will not reference it again.
|
2006-01-08 09:01:35 +00:00
|
|
|
* Can be invoked from softirq via RCU callback due to
|
|
|
|
* call_rcu() of the task struct, hence the _irqsave.
|
2005-04-16 22:20:36 +00:00
|
|
|
*/
|
2008-07-22 19:08:51 +00:00
|
|
|
static int
|
|
|
|
task_free_notify(struct notifier_block *self, unsigned long val, void *data)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2006-01-08 09:01:35 +00:00
|
|
|
unsigned long flags;
|
2008-07-22 19:08:51 +00:00
|
|
|
struct task_struct *task = data;
|
2006-01-08 09:01:35 +00:00
|
|
|
spin_lock_irqsave(&task_mortuary, flags);
|
2005-04-16 22:20:36 +00:00
|
|
|
list_add(&task->tasks, &dying_tasks);
|
2006-01-08 09:01:35 +00:00
|
|
|
spin_unlock_irqrestore(&task_mortuary, flags);
|
2005-04-16 22:20:36 +00:00
|
|
|
return NOTIFY_OK;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/* The task is on its way out. A sync of the buffer means we can catch
|
|
|
|
* any remaining samples for this task.
|
|
|
|
*/
|
2008-07-22 19:08:51 +00:00
|
|
|
static int
|
|
|
|
task_exit_notify(struct notifier_block *self, unsigned long val, void *data)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
/* To avoid latency problems, we only process the current CPU,
|
|
|
|
* hoping that most samples for the task are on this CPU
|
|
|
|
*/
|
2005-06-22 00:14:34 +00:00
|
|
|
sync_buffer(raw_smp_processor_id());
|
2008-07-22 19:08:51 +00:00
|
|
|
return 0;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/* The task is about to try a do_munmap(). We peek at what it's going to
|
|
|
|
* do, and if it's an executable region, process the samples first, so
|
|
|
|
* we don't lose any. This does not have to be exact, it's a QoI issue
|
|
|
|
* only.
|
|
|
|
*/
|
2008-07-22 19:08:51 +00:00
|
|
|
static int
|
|
|
|
munmap_notify(struct notifier_block *self, unsigned long val, void *data)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
unsigned long addr = (unsigned long)data;
|
2008-07-22 19:08:51 +00:00
|
|
|
struct mm_struct *mm = current->mm;
|
|
|
|
struct vm_area_struct *mpnt;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
down_read(&mm->mmap_sem);
|
|
|
|
|
|
|
|
mpnt = find_vma(mm, addr);
|
|
|
|
if (mpnt && mpnt->vm_file && (mpnt->vm_flags & VM_EXEC)) {
|
|
|
|
up_read(&mm->mmap_sem);
|
|
|
|
/* To avoid latency problems, we only process the current CPU,
|
|
|
|
* hoping that most samples for the task are on this CPU
|
|
|
|
*/
|
2005-06-22 00:14:34 +00:00
|
|
|
sync_buffer(raw_smp_processor_id());
|
2005-04-16 22:20:36 +00:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
up_read(&mm->mmap_sem);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2008-07-22 19:08:51 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/* We need to be told about new modules so we don't attribute to a previously
|
|
|
|
* loaded module, or drop the samples on the floor.
|
|
|
|
*/
|
2008-07-22 19:08:51 +00:00
|
|
|
static int
|
|
|
|
module_load_notify(struct notifier_block *self, unsigned long val, void *data)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
#ifdef CONFIG_MODULES
|
|
|
|
if (val != MODULE_STATE_COMING)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
/* FIXME: should we process all CPU buffers ? */
|
2006-06-25 12:47:33 +00:00
|
|
|
mutex_lock(&buffer_mutex);
|
2005-04-16 22:20:36 +00:00
|
|
|
add_event_entry(ESCAPE_CODE);
|
|
|
|
add_event_entry(MODULE_LOADED_CODE);
|
2006-06-25 12:47:33 +00:00
|
|
|
mutex_unlock(&buffer_mutex);
|
2005-04-16 22:20:36 +00:00
|
|
|
#endif
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2008-07-22 19:08:51 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
static struct notifier_block task_free_nb = {
|
|
|
|
.notifier_call = task_free_notify,
|
|
|
|
};
|
|
|
|
|
|
|
|
static struct notifier_block task_exit_nb = {
|
|
|
|
.notifier_call = task_exit_notify,
|
|
|
|
};
|
|
|
|
|
|
|
|
static struct notifier_block munmap_nb = {
|
|
|
|
.notifier_call = munmap_notify,
|
|
|
|
};
|
|
|
|
|
|
|
|
static struct notifier_block module_load_nb = {
|
|
|
|
.notifier_call = module_load_notify,
|
|
|
|
};
|
|
|
|
|
2008-07-22 19:08:51 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
static void end_sync(void)
|
|
|
|
{
|
|
|
|
end_cpu_work();
|
|
|
|
/* make sure we don't leak task structs */
|
|
|
|
process_task_mortuary();
|
|
|
|
process_task_mortuary();
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
int sync_start(void)
|
|
|
|
{
|
|
|
|
int err;
|
|
|
|
|
|
|
|
start_cpu_work();
|
|
|
|
|
|
|
|
err = task_handoff_register(&task_free_nb);
|
|
|
|
if (err)
|
|
|
|
goto out1;
|
|
|
|
err = profile_event_register(PROFILE_TASK_EXIT, &task_exit_nb);
|
|
|
|
if (err)
|
|
|
|
goto out2;
|
|
|
|
err = profile_event_register(PROFILE_MUNMAP, &munmap_nb);
|
|
|
|
if (err)
|
|
|
|
goto out3;
|
|
|
|
err = register_module_notifier(&module_load_nb);
|
|
|
|
if (err)
|
|
|
|
goto out4;
|
|
|
|
|
|
|
|
out:
|
|
|
|
return err;
|
|
|
|
out4:
|
|
|
|
profile_event_unregister(PROFILE_MUNMAP, &munmap_nb);
|
|
|
|
out3:
|
|
|
|
profile_event_unregister(PROFILE_TASK_EXIT, &task_exit_nb);
|
|
|
|
out2:
|
|
|
|
task_handoff_unregister(&task_free_nb);
|
|
|
|
out1:
|
|
|
|
end_sync();
|
|
|
|
goto out;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
void sync_stop(void)
|
|
|
|
{
|
|
|
|
unregister_module_notifier(&module_load_nb);
|
|
|
|
profile_event_unregister(PROFILE_MUNMAP, &munmap_nb);
|
|
|
|
profile_event_unregister(PROFILE_TASK_EXIT, &task_exit_nb);
|
|
|
|
task_handoff_unregister(&task_free_nb);
|
|
|
|
end_sync();
|
|
|
|
}
|
|
|
|
|
2008-02-15 03:38:36 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/* Optimisation. We can manage without taking the dcookie sem
|
|
|
|
* because we cannot reach this code without at least one
|
|
|
|
* dcookie user still being registered (namely, the reader
|
|
|
|
* of the event buffer). */
|
2008-02-15 03:38:36 +00:00
|
|
|
static inline unsigned long fast_get_dcookie(struct path *path)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
unsigned long cookie;
|
2008-02-15 03:38:36 +00:00
|
|
|
|
2008-12-01 08:33:43 +00:00
|
|
|
if (path->dentry->d_flags & DCACHE_COOKIE)
|
2008-02-15 03:38:36 +00:00
|
|
|
return (unsigned long)path->dentry;
|
|
|
|
get_dcookie(path, &cookie);
|
2005-04-16 22:20:36 +00:00
|
|
|
return cookie;
|
|
|
|
}
|
|
|
|
|
2008-02-15 03:38:36 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/* Look up the dcookie for the task's first VM_EXECUTABLE mapping,
|
|
|
|
* which corresponds loosely to "application name". This is
|
|
|
|
* not strictly necessary but allows oprofile to associate
|
|
|
|
* shared-library samples with particular applications
|
|
|
|
*/
|
2008-07-22 19:08:51 +00:00
|
|
|
static unsigned long get_exec_dcookie(struct mm_struct *mm)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2005-06-24 05:02:47 +00:00
|
|
|
unsigned long cookie = NO_COOKIE;
|
2008-07-22 19:08:51 +00:00
|
|
|
struct vm_area_struct *vma;
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
if (!mm)
|
|
|
|
goto out;
|
2008-07-22 19:08:51 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
for (vma = mm->mmap; vma; vma = vma->vm_next) {
|
|
|
|
if (!vma->vm_file)
|
|
|
|
continue;
|
|
|
|
if (!(vma->vm_flags & VM_EXECUTABLE))
|
|
|
|
continue;
|
2008-02-15 03:38:36 +00:00
|
|
|
cookie = fast_get_dcookie(&vma->vm_file->f_path);
|
2005-04-16 22:20:36 +00:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
out:
|
|
|
|
return cookie;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/* Convert the EIP value of a sample into a persistent dentry/offset
|
|
|
|
* pair that can then be added to the global event buffer. We make
|
|
|
|
* sure to do this lookup before a mm->mmap modification happens so
|
|
|
|
* we don't lose track.
|
|
|
|
*/
|
2008-07-22 19:08:51 +00:00
|
|
|
static unsigned long
|
|
|
|
lookup_dcookie(struct mm_struct *mm, unsigned long addr, off_t *offset)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2005-06-24 05:02:47 +00:00
|
|
|
unsigned long cookie = NO_COOKIE;
|
2008-07-22 19:08:51 +00:00
|
|
|
struct vm_area_struct *vma;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
for (vma = find_vma(mm, addr); vma; vma = vma->vm_next) {
|
2008-07-22 19:08:51 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
if (addr < vma->vm_start || addr >= vma->vm_end)
|
|
|
|
continue;
|
|
|
|
|
2005-06-24 05:02:47 +00:00
|
|
|
if (vma->vm_file) {
|
2008-02-15 03:38:36 +00:00
|
|
|
cookie = fast_get_dcookie(&vma->vm_file->f_path);
|
2005-06-24 05:02:47 +00:00
|
|
|
*offset = (vma->vm_pgoff << PAGE_SHIFT) + addr -
|
|
|
|
vma->vm_start;
|
|
|
|
} else {
|
|
|
|
/* must be an anonymous map */
|
|
|
|
*offset = addr;
|
|
|
|
}
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
2005-06-24 05:02:47 +00:00
|
|
|
if (!vma)
|
|
|
|
cookie = INVALID_COOKIE;
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
return cookie;
|
|
|
|
}
|
|
|
|
|
2005-06-24 05:02:47 +00:00
|
|
|
static unsigned long last_cookie = INVALID_COOKIE;
|
2008-07-22 19:08:51 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
static void add_cpu_switch(int i)
|
|
|
|
{
|
|
|
|
add_event_entry(ESCAPE_CODE);
|
|
|
|
add_event_entry(CPU_SWITCH_CODE);
|
|
|
|
add_event_entry(i);
|
2005-06-24 05:02:47 +00:00
|
|
|
last_cookie = INVALID_COOKIE;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
static void add_kernel_ctx_switch(unsigned int in_kernel)
|
|
|
|
{
|
|
|
|
add_event_entry(ESCAPE_CODE);
|
|
|
|
if (in_kernel)
|
2008-07-22 19:08:51 +00:00
|
|
|
add_event_entry(KERNEL_ENTER_SWITCH_CODE);
|
2005-04-16 22:20:36 +00:00
|
|
|
else
|
2008-07-22 19:08:51 +00:00
|
|
|
add_event_entry(KERNEL_EXIT_SWITCH_CODE);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2008-07-22 19:08:51 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
static void
|
2008-07-22 19:08:51 +00:00
|
|
|
add_user_ctx_switch(struct task_struct const *task, unsigned long cookie)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
add_event_entry(ESCAPE_CODE);
|
2008-07-22 19:08:51 +00:00
|
|
|
add_event_entry(CTX_SWITCH_CODE);
|
2005-04-16 22:20:36 +00:00
|
|
|
add_event_entry(task->pid);
|
|
|
|
add_event_entry(cookie);
|
|
|
|
/* Another code for daemon back-compat */
|
|
|
|
add_event_entry(ESCAPE_CODE);
|
|
|
|
add_event_entry(CTX_TGID_CODE);
|
|
|
|
add_event_entry(task->tgid);
|
|
|
|
}
|
|
|
|
|
2008-07-22 19:08:51 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
static void add_cookie_switch(unsigned long cookie)
|
|
|
|
{
|
|
|
|
add_event_entry(ESCAPE_CODE);
|
|
|
|
add_event_entry(COOKIE_SWITCH_CODE);
|
|
|
|
add_event_entry(cookie);
|
|
|
|
}
|
|
|
|
|
2008-07-22 19:08:51 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
static void add_trace_begin(void)
|
|
|
|
{
|
|
|
|
add_event_entry(ESCAPE_CODE);
|
|
|
|
add_event_entry(TRACE_BEGIN_CODE);
|
|
|
|
}
|
|
|
|
|
2008-07-22 19:09:06 +00:00
|
|
|
#ifdef CONFIG_OPROFILE_IBS
|
|
|
|
|
2008-07-22 19:08:54 +00:00
|
|
|
#define IBS_FETCH_CODE_SIZE 2
|
|
|
|
#define IBS_OP_CODE_SIZE 5
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Add IBS fetch and op entries to event buffer
|
|
|
|
*/
|
oprofile: port to the new ring_buffer
This patch replaces the current oprofile cpu buffer implementation
with the ring buffer provided by the tracing framework. The motivation
here is to leave the pain of implementing ring buffers to others. Oh,
no, there are more advantages. Main reason is the support of different
sample sizes that could be stored in the buffer. Use cases for this
are IBS and Cell spu profiling. Using the new ring buffer ensures
valid and complete samples and allows copying the cpu buffer stateless
without knowing its content. Second it will use generic kernel API and
also reduce code size. And hopefully, there are less bugs.
Since the new tracing ring buffer implementation uses spin locks to
protect the buffer during read/write access, it is difficult to use
the buffer in an NMI handler. In this case, writing to the buffer by
the NMI handler (x86) could occur also during critical sections when
reading the buffer. To avoid this, there are 2 buffers for independent
read and write access. Read access is in process context only, write
access only in the NMI handler. If the read buffer runs empty, both
buffers are swapped atomically. There is potentially a small window
during swapping where the buffers are disabled and samples could be
lost.
Using 2 buffers is a little bit overhead, but the solution is clear
and does not require changes in the ring buffer implementation. It can
be changed to a single buffer solution when the ring buffer access is
implemented as non-locking atomic code.
The new buffer requires more size to store the same amount of samples
because each sample includes an u32 header. Also, there is more code
to execute for buffer access. Nonetheless, the buffer implementation
is proven in the ftrace environment and worth to use also in oprofile.
Patches that changes the internal IBS buffer usage will follow.
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Robert Richter <robert.richter@amd.com>
2008-12-09 00:21:32 +00:00
|
|
|
static void add_ibs_begin(int cpu, int code, struct mm_struct *mm)
|
2008-07-22 19:08:54 +00:00
|
|
|
{
|
|
|
|
unsigned long rip;
|
|
|
|
int i, count;
|
|
|
|
unsigned long ibs_cookie = 0;
|
|
|
|
off_t offset;
|
oprofile: port to the new ring_buffer
This patch replaces the current oprofile cpu buffer implementation
with the ring buffer provided by the tracing framework. The motivation
here is to leave the pain of implementing ring buffers to others. Oh,
no, there are more advantages. Main reason is the support of different
sample sizes that could be stored in the buffer. Use cases for this
are IBS and Cell spu profiling. Using the new ring buffer ensures
valid and complete samples and allows copying the cpu buffer stateless
without knowing its content. Second it will use generic kernel API and
also reduce code size. And hopefully, there are less bugs.
Since the new tracing ring buffer implementation uses spin locks to
protect the buffer during read/write access, it is difficult to use
the buffer in an NMI handler. In this case, writing to the buffer by
the NMI handler (x86) could occur also during critical sections when
reading the buffer. To avoid this, there are 2 buffers for independent
read and write access. Read access is in process context only, write
access only in the NMI handler. If the read buffer runs empty, both
buffers are swapped atomically. There is potentially a small window
during swapping where the buffers are disabled and samples could be
lost.
Using 2 buffers is a little bit overhead, but the solution is clear
and does not require changes in the ring buffer implementation. It can
be changed to a single buffer solution when the ring buffer access is
implemented as non-locking atomic code.
The new buffer requires more size to store the same amount of samples
because each sample includes an u32 header. Also, there is more code
to execute for buffer access. Nonetheless, the buffer implementation
is proven in the ftrace environment and worth to use also in oprofile.
Patches that changes the internal IBS buffer usage will follow.
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Robert Richter <robert.richter@amd.com>
2008-12-09 00:21:32 +00:00
|
|
|
struct op_sample *sample;
|
2008-07-22 19:08:54 +00:00
|
|
|
|
oprofile: port to the new ring_buffer
This patch replaces the current oprofile cpu buffer implementation
with the ring buffer provided by the tracing framework. The motivation
here is to leave the pain of implementing ring buffers to others. Oh,
no, there are more advantages. Main reason is the support of different
sample sizes that could be stored in the buffer. Use cases for this
are IBS and Cell spu profiling. Using the new ring buffer ensures
valid and complete samples and allows copying the cpu buffer stateless
without knowing its content. Second it will use generic kernel API and
also reduce code size. And hopefully, there are less bugs.
Since the new tracing ring buffer implementation uses spin locks to
protect the buffer during read/write access, it is difficult to use
the buffer in an NMI handler. In this case, writing to the buffer by
the NMI handler (x86) could occur also during critical sections when
reading the buffer. To avoid this, there are 2 buffers for independent
read and write access. Read access is in process context only, write
access only in the NMI handler. If the read buffer runs empty, both
buffers are swapped atomically. There is potentially a small window
during swapping where the buffers are disabled and samples could be
lost.
Using 2 buffers is a little bit overhead, but the solution is clear
and does not require changes in the ring buffer implementation. It can
be changed to a single buffer solution when the ring buffer access is
implemented as non-locking atomic code.
The new buffer requires more size to store the same amount of samples
because each sample includes an u32 header. Also, there is more code
to execute for buffer access. Nonetheless, the buffer implementation
is proven in the ftrace environment and worth to use also in oprofile.
Patches that changes the internal IBS buffer usage will follow.
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Robert Richter <robert.richter@amd.com>
2008-12-09 00:21:32 +00:00
|
|
|
sample = cpu_buffer_read_entry(cpu);
|
|
|
|
if (!sample)
|
|
|
|
goto Error;
|
|
|
|
rip = sample->eip;
|
2008-07-22 19:08:54 +00:00
|
|
|
|
|
|
|
#ifdef __LP64__
|
oprofile: port to the new ring_buffer
This patch replaces the current oprofile cpu buffer implementation
with the ring buffer provided by the tracing framework. The motivation
here is to leave the pain of implementing ring buffers to others. Oh,
no, there are more advantages. Main reason is the support of different
sample sizes that could be stored in the buffer. Use cases for this
are IBS and Cell spu profiling. Using the new ring buffer ensures
valid and complete samples and allows copying the cpu buffer stateless
without knowing its content. Second it will use generic kernel API and
also reduce code size. And hopefully, there are less bugs.
Since the new tracing ring buffer implementation uses spin locks to
protect the buffer during read/write access, it is difficult to use
the buffer in an NMI handler. In this case, writing to the buffer by
the NMI handler (x86) could occur also during critical sections when
reading the buffer. To avoid this, there are 2 buffers for independent
read and write access. Read access is in process context only, write
access only in the NMI handler. If the read buffer runs empty, both
buffers are swapped atomically. There is potentially a small window
during swapping where the buffers are disabled and samples could be
lost.
Using 2 buffers is a little bit overhead, but the solution is clear
and does not require changes in the ring buffer implementation. It can
be changed to a single buffer solution when the ring buffer access is
implemented as non-locking atomic code.
The new buffer requires more size to store the same amount of samples
because each sample includes an u32 header. Also, there is more code
to execute for buffer access. Nonetheless, the buffer implementation
is proven in the ftrace environment and worth to use also in oprofile.
Patches that changes the internal IBS buffer usage will follow.
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Robert Richter <robert.richter@amd.com>
2008-12-09 00:21:32 +00:00
|
|
|
rip += sample->event << 32;
|
2008-07-22 19:08:54 +00:00
|
|
|
#endif
|
|
|
|
|
|
|
|
if (mm) {
|
|
|
|
ibs_cookie = lookup_dcookie(mm, rip, &offset);
|
|
|
|
|
|
|
|
if (ibs_cookie == NO_COOKIE)
|
|
|
|
offset = rip;
|
|
|
|
if (ibs_cookie == INVALID_COOKIE) {
|
|
|
|
atomic_inc(&oprofile_stats.sample_lost_no_mapping);
|
|
|
|
offset = rip;
|
|
|
|
}
|
|
|
|
if (ibs_cookie != last_cookie) {
|
|
|
|
add_cookie_switch(ibs_cookie);
|
|
|
|
last_cookie = ibs_cookie;
|
|
|
|
}
|
|
|
|
} else
|
|
|
|
offset = rip;
|
|
|
|
|
|
|
|
add_event_entry(ESCAPE_CODE);
|
|
|
|
add_event_entry(code);
|
|
|
|
add_event_entry(offset); /* Offset from Dcookie */
|
|
|
|
|
|
|
|
/* we send the Dcookie offset, but send the raw Linear Add also*/
|
oprofile: port to the new ring_buffer
This patch replaces the current oprofile cpu buffer implementation
with the ring buffer provided by the tracing framework. The motivation
here is to leave the pain of implementing ring buffers to others. Oh,
no, there are more advantages. Main reason is the support of different
sample sizes that could be stored in the buffer. Use cases for this
are IBS and Cell spu profiling. Using the new ring buffer ensures
valid and complete samples and allows copying the cpu buffer stateless
without knowing its content. Second it will use generic kernel API and
also reduce code size. And hopefully, there are less bugs.
Since the new tracing ring buffer implementation uses spin locks to
protect the buffer during read/write access, it is difficult to use
the buffer in an NMI handler. In this case, writing to the buffer by
the NMI handler (x86) could occur also during critical sections when
reading the buffer. To avoid this, there are 2 buffers for independent
read and write access. Read access is in process context only, write
access only in the NMI handler. If the read buffer runs empty, both
buffers are swapped atomically. There is potentially a small window
during swapping where the buffers are disabled and samples could be
lost.
Using 2 buffers is a little bit overhead, but the solution is clear
and does not require changes in the ring buffer implementation. It can
be changed to a single buffer solution when the ring buffer access is
implemented as non-locking atomic code.
The new buffer requires more size to store the same amount of samples
because each sample includes an u32 header. Also, there is more code
to execute for buffer access. Nonetheless, the buffer implementation
is proven in the ftrace environment and worth to use also in oprofile.
Patches that changes the internal IBS buffer usage will follow.
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Robert Richter <robert.richter@amd.com>
2008-12-09 00:21:32 +00:00
|
|
|
add_event_entry(sample->eip);
|
|
|
|
add_event_entry(sample->event);
|
2008-07-22 19:08:54 +00:00
|
|
|
|
|
|
|
if (code == IBS_FETCH_CODE)
|
|
|
|
count = IBS_FETCH_CODE_SIZE; /*IBS FETCH is 2 int64s*/
|
|
|
|
else
|
|
|
|
count = IBS_OP_CODE_SIZE; /*IBS OP is 5 int64s*/
|
|
|
|
|
|
|
|
for (i = 0; i < count; i++) {
|
oprofile: port to the new ring_buffer
This patch replaces the current oprofile cpu buffer implementation
with the ring buffer provided by the tracing framework. The motivation
here is to leave the pain of implementing ring buffers to others. Oh,
no, there are more advantages. Main reason is the support of different
sample sizes that could be stored in the buffer. Use cases for this
are IBS and Cell spu profiling. Using the new ring buffer ensures
valid and complete samples and allows copying the cpu buffer stateless
without knowing its content. Second it will use generic kernel API and
also reduce code size. And hopefully, there are less bugs.
Since the new tracing ring buffer implementation uses spin locks to
protect the buffer during read/write access, it is difficult to use
the buffer in an NMI handler. In this case, writing to the buffer by
the NMI handler (x86) could occur also during critical sections when
reading the buffer. To avoid this, there are 2 buffers for independent
read and write access. Read access is in process context only, write
access only in the NMI handler. If the read buffer runs empty, both
buffers are swapped atomically. There is potentially a small window
during swapping where the buffers are disabled and samples could be
lost.
Using 2 buffers is a little bit overhead, but the solution is clear
and does not require changes in the ring buffer implementation. It can
be changed to a single buffer solution when the ring buffer access is
implemented as non-locking atomic code.
The new buffer requires more size to store the same amount of samples
because each sample includes an u32 header. Also, there is more code
to execute for buffer access. Nonetheless, the buffer implementation
is proven in the ftrace environment and worth to use also in oprofile.
Patches that changes the internal IBS buffer usage will follow.
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Robert Richter <robert.richter@amd.com>
2008-12-09 00:21:32 +00:00
|
|
|
sample = cpu_buffer_read_entry(cpu);
|
|
|
|
if (!sample)
|
|
|
|
goto Error;
|
|
|
|
add_event_entry(sample->eip);
|
|
|
|
add_event_entry(sample->event);
|
2008-07-22 19:08:54 +00:00
|
|
|
}
|
oprofile: port to the new ring_buffer
This patch replaces the current oprofile cpu buffer implementation
with the ring buffer provided by the tracing framework. The motivation
here is to leave the pain of implementing ring buffers to others. Oh,
no, there are more advantages. Main reason is the support of different
sample sizes that could be stored in the buffer. Use cases for this
are IBS and Cell spu profiling. Using the new ring buffer ensures
valid and complete samples and allows copying the cpu buffer stateless
without knowing its content. Second it will use generic kernel API and
also reduce code size. And hopefully, there are less bugs.
Since the new tracing ring buffer implementation uses spin locks to
protect the buffer during read/write access, it is difficult to use
the buffer in an NMI handler. In this case, writing to the buffer by
the NMI handler (x86) could occur also during critical sections when
reading the buffer. To avoid this, there are 2 buffers for independent
read and write access. Read access is in process context only, write
access only in the NMI handler. If the read buffer runs empty, both
buffers are swapped atomically. There is potentially a small window
during swapping where the buffers are disabled and samples could be
lost.
Using 2 buffers is a little bit overhead, but the solution is clear
and does not require changes in the ring buffer implementation. It can
be changed to a single buffer solution when the ring buffer access is
implemented as non-locking atomic code.
The new buffer requires more size to store the same amount of samples
because each sample includes an u32 header. Also, there is more code
to execute for buffer access. Nonetheless, the buffer implementation
is proven in the ftrace environment and worth to use also in oprofile.
Patches that changes the internal IBS buffer usage will follow.
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Robert Richter <robert.richter@amd.com>
2008-12-09 00:21:32 +00:00
|
|
|
|
|
|
|
return;
|
|
|
|
|
|
|
|
Error:
|
|
|
|
return;
|
2008-07-22 19:08:54 +00:00
|
|
|
}
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2008-07-22 19:09:06 +00:00
|
|
|
#endif
|
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
static void add_sample_entry(unsigned long offset, unsigned long event)
|
|
|
|
{
|
|
|
|
add_event_entry(offset);
|
|
|
|
add_event_entry(event);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2008-07-22 19:08:51 +00:00
|
|
|
static int add_us_sample(struct mm_struct *mm, struct op_sample *s)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
unsigned long cookie;
|
|
|
|
off_t offset;
|
2008-07-22 19:08:51 +00:00
|
|
|
|
|
|
|
cookie = lookup_dcookie(mm, s->eip, &offset);
|
|
|
|
|
2005-06-24 05:02:47 +00:00
|
|
|
if (cookie == INVALID_COOKIE) {
|
2005-04-16 22:20:36 +00:00
|
|
|
atomic_inc(&oprofile_stats.sample_lost_no_mapping);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (cookie != last_cookie) {
|
|
|
|
add_cookie_switch(cookie);
|
|
|
|
last_cookie = cookie;
|
|
|
|
}
|
|
|
|
|
|
|
|
add_sample_entry(offset, s->event);
|
|
|
|
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
2008-07-22 19:08:51 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
/* Add a sample to the global event buffer. If possible the
|
|
|
|
* sample is converted into a persistent dentry/offset pair
|
|
|
|
* for later lookup from userspace.
|
|
|
|
*/
|
|
|
|
static int
|
2008-07-22 19:08:51 +00:00
|
|
|
add_sample(struct mm_struct *mm, struct op_sample *s, int in_kernel)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
if (in_kernel) {
|
|
|
|
add_sample_entry(s->eip, s->event);
|
|
|
|
return 1;
|
|
|
|
} else if (mm) {
|
|
|
|
return add_us_sample(mm, s);
|
|
|
|
} else {
|
|
|
|
atomic_inc(&oprofile_stats.sample_lost_no_mm);
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2008-07-22 19:08:51 +00:00
|
|
|
|
|
|
|
static void release_mm(struct mm_struct *mm)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
|
|
|
if (!mm)
|
|
|
|
return;
|
|
|
|
up_read(&mm->mmap_sem);
|
|
|
|
mmput(mm);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
2008-07-22 19:08:51 +00:00
|
|
|
static struct mm_struct *take_tasks_mm(struct task_struct *task)
|
2005-04-16 22:20:36 +00:00
|
|
|
{
|
2008-07-22 19:08:51 +00:00
|
|
|
struct mm_struct *mm = get_task_mm(task);
|
2005-04-16 22:20:36 +00:00
|
|
|
if (mm)
|
|
|
|
down_read(&mm->mmap_sem);
|
|
|
|
return mm;
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static inline int is_code(unsigned long val)
|
|
|
|
{
|
|
|
|
return val == ESCAPE_CODE;
|
|
|
|
}
|
2008-07-22 19:08:51 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
/* Move tasks along towards death. Any tasks on dead_tasks
|
|
|
|
* will definitely have no remaining references in any
|
|
|
|
* CPU buffers at this point, because we use two lists,
|
|
|
|
* and to have reached the list, it must have gone through
|
|
|
|
* one full sync already.
|
|
|
|
*/
|
|
|
|
static void process_task_mortuary(void)
|
|
|
|
{
|
2006-01-08 09:01:35 +00:00
|
|
|
unsigned long flags;
|
|
|
|
LIST_HEAD(local_dead_tasks);
|
2008-07-22 19:08:51 +00:00
|
|
|
struct task_struct *task;
|
|
|
|
struct task_struct *ttask;
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2006-01-08 09:01:35 +00:00
|
|
|
spin_lock_irqsave(&task_mortuary, flags);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2006-01-08 09:01:35 +00:00
|
|
|
list_splice_init(&dead_tasks, &local_dead_tasks);
|
|
|
|
list_splice_init(&dying_tasks, &dead_tasks);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
2006-01-08 09:01:35 +00:00
|
|
|
spin_unlock_irqrestore(&task_mortuary, flags);
|
|
|
|
|
|
|
|
list_for_each_entry_safe(task, ttask, &local_dead_tasks, tasks) {
|
2005-04-16 22:20:36 +00:00
|
|
|
list_del(&task->tasks);
|
2006-01-08 09:01:35 +00:00
|
|
|
free_task(task);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
static void mark_done(int cpu)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
|
|
|
|
cpu_set(cpu, marked_cpus);
|
|
|
|
|
|
|
|
for_each_online_cpu(i) {
|
|
|
|
if (!cpu_isset(i, marked_cpus))
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* All CPUs have been processed at least once,
|
|
|
|
* we can process the mortuary once
|
|
|
|
*/
|
|
|
|
process_task_mortuary();
|
|
|
|
|
|
|
|
cpus_clear(marked_cpus);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/* FIXME: this is not sufficient if we implement syscall barrier backtrace
|
|
|
|
* traversal, the code switch to sb_sample_start at first kernel enter/exit
|
|
|
|
* switch so we need a fifth state and some special handling in sync_buffer()
|
|
|
|
*/
|
|
|
|
typedef enum {
|
|
|
|
sb_bt_ignore = -2,
|
|
|
|
sb_buffer_start,
|
|
|
|
sb_bt_start,
|
|
|
|
sb_sample_start,
|
|
|
|
} sync_buffer_state;
|
|
|
|
|
|
|
|
/* Sync one of the CPU's buffers into the global event buffer.
|
|
|
|
* Here we need to go through each batch of samples punctuated
|
|
|
|
* by context switch notes, taking the task's mmap_sem and doing
|
|
|
|
* lookup in task->mm->mmap to convert EIP into dcookie/offset
|
|
|
|
* value.
|
|
|
|
*/
|
|
|
|
void sync_buffer(int cpu)
|
|
|
|
{
|
|
|
|
struct mm_struct *mm = NULL;
|
2008-09-26 21:50:31 +00:00
|
|
|
struct mm_struct *oldmm;
|
2008-07-22 19:08:51 +00:00
|
|
|
struct task_struct *new;
|
2005-04-16 22:20:36 +00:00
|
|
|
unsigned long cookie = 0;
|
|
|
|
int in_kernel = 1;
|
|
|
|
sync_buffer_state state = sb_buffer_start;
|
2008-07-14 22:10:36 +00:00
|
|
|
unsigned int i;
|
2005-04-16 22:20:36 +00:00
|
|
|
unsigned long available;
|
|
|
|
|
2006-06-25 12:47:33 +00:00
|
|
|
mutex_lock(&buffer_mutex);
|
2008-07-22 19:08:51 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
add_cpu_switch(cpu);
|
|
|
|
|
2008-12-04 15:27:00 +00:00
|
|
|
cpu_buffer_reset(cpu);
|
oprofile: port to the new ring_buffer
This patch replaces the current oprofile cpu buffer implementation
with the ring buffer provided by the tracing framework. The motivation
here is to leave the pain of implementing ring buffers to others. Oh,
no, there are more advantages. Main reason is the support of different
sample sizes that could be stored in the buffer. Use cases for this
are IBS and Cell spu profiling. Using the new ring buffer ensures
valid and complete samples and allows copying the cpu buffer stateless
without knowing its content. Second it will use generic kernel API and
also reduce code size. And hopefully, there are less bugs.
Since the new tracing ring buffer implementation uses spin locks to
protect the buffer during read/write access, it is difficult to use
the buffer in an NMI handler. In this case, writing to the buffer by
the NMI handler (x86) could occur also during critical sections when
reading the buffer. To avoid this, there are 2 buffers for independent
read and write access. Read access is in process context only, write
access only in the NMI handler. If the read buffer runs empty, both
buffers are swapped atomically. There is potentially a small window
during swapping where the buffers are disabled and samples could be
lost.
Using 2 buffers is a little bit overhead, but the solution is clear
and does not require changes in the ring buffer implementation. It can
be changed to a single buffer solution when the ring buffer access is
implemented as non-locking atomic code.
The new buffer requires more size to store the same amount of samples
because each sample includes an u32 header. Also, there is more code
to execute for buffer access. Nonetheless, the buffer implementation
is proven in the ftrace environment and worth to use also in oprofile.
Patches that changes the internal IBS buffer usage will follow.
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Robert Richter <robert.richter@amd.com>
2008-12-09 00:21:32 +00:00
|
|
|
available = cpu_buffer_entries(cpu);
|
2005-04-16 22:20:36 +00:00
|
|
|
|
|
|
|
for (i = 0; i < available; ++i) {
|
oprofile: port to the new ring_buffer
This patch replaces the current oprofile cpu buffer implementation
with the ring buffer provided by the tracing framework. The motivation
here is to leave the pain of implementing ring buffers to others. Oh,
no, there are more advantages. Main reason is the support of different
sample sizes that could be stored in the buffer. Use cases for this
are IBS and Cell spu profiling. Using the new ring buffer ensures
valid and complete samples and allows copying the cpu buffer stateless
without knowing its content. Second it will use generic kernel API and
also reduce code size. And hopefully, there are less bugs.
Since the new tracing ring buffer implementation uses spin locks to
protect the buffer during read/write access, it is difficult to use
the buffer in an NMI handler. In this case, writing to the buffer by
the NMI handler (x86) could occur also during critical sections when
reading the buffer. To avoid this, there are 2 buffers for independent
read and write access. Read access is in process context only, write
access only in the NMI handler. If the read buffer runs empty, both
buffers are swapped atomically. There is potentially a small window
during swapping where the buffers are disabled and samples could be
lost.
Using 2 buffers is a little bit overhead, but the solution is clear
and does not require changes in the ring buffer implementation. It can
be changed to a single buffer solution when the ring buffer access is
implemented as non-locking atomic code.
The new buffer requires more size to store the same amount of samples
because each sample includes an u32 header. Also, there is more code
to execute for buffer access. Nonetheless, the buffer implementation
is proven in the ftrace environment and worth to use also in oprofile.
Patches that changes the internal IBS buffer usage will follow.
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Robert Richter <robert.richter@amd.com>
2008-12-09 00:21:32 +00:00
|
|
|
struct op_sample *s = cpu_buffer_read_entry(cpu);
|
|
|
|
if (!s)
|
|
|
|
break;
|
2008-07-22 19:08:51 +00:00
|
|
|
|
2005-04-16 22:20:36 +00:00
|
|
|
if (is_code(s->eip)) {
|
2008-09-26 21:50:31 +00:00
|
|
|
switch (s->event) {
|
|
|
|
case 0:
|
|
|
|
case CPU_IS_KERNEL:
|
2005-04-16 22:20:36 +00:00
|
|
|
/* kernel/userspace switch */
|
|
|
|
in_kernel = s->event;
|
|
|
|
if (state == sb_buffer_start)
|
|
|
|
state = sb_sample_start;
|
|
|
|
add_kernel_ctx_switch(s->event);
|
2008-09-26 21:50:31 +00:00
|
|
|
break;
|
|
|
|
case CPU_TRACE_BEGIN:
|
2005-04-16 22:20:36 +00:00
|
|
|
state = sb_bt_start;
|
|
|
|
add_trace_begin();
|
2008-09-26 21:50:31 +00:00
|
|
|
break;
|
2008-07-22 19:09:06 +00:00
|
|
|
#ifdef CONFIG_OPROFILE_IBS
|
2008-09-26 21:50:31 +00:00
|
|
|
case IBS_FETCH_BEGIN:
|
2008-07-22 19:08:54 +00:00
|
|
|
state = sb_bt_start;
|
oprofile: port to the new ring_buffer
This patch replaces the current oprofile cpu buffer implementation
with the ring buffer provided by the tracing framework. The motivation
here is to leave the pain of implementing ring buffers to others. Oh,
no, there are more advantages. Main reason is the support of different
sample sizes that could be stored in the buffer. Use cases for this
are IBS and Cell spu profiling. Using the new ring buffer ensures
valid and complete samples and allows copying the cpu buffer stateless
without knowing its content. Second it will use generic kernel API and
also reduce code size. And hopefully, there are less bugs.
Since the new tracing ring buffer implementation uses spin locks to
protect the buffer during read/write access, it is difficult to use
the buffer in an NMI handler. In this case, writing to the buffer by
the NMI handler (x86) could occur also during critical sections when
reading the buffer. To avoid this, there are 2 buffers for independent
read and write access. Read access is in process context only, write
access only in the NMI handler. If the read buffer runs empty, both
buffers are swapped atomically. There is potentially a small window
during swapping where the buffers are disabled and samples could be
lost.
Using 2 buffers is a little bit overhead, but the solution is clear
and does not require changes in the ring buffer implementation. It can
be changed to a single buffer solution when the ring buffer access is
implemented as non-locking atomic code.
The new buffer requires more size to store the same amount of samples
because each sample includes an u32 header. Also, there is more code
to execute for buffer access. Nonetheless, the buffer implementation
is proven in the ftrace environment and worth to use also in oprofile.
Patches that changes the internal IBS buffer usage will follow.
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Robert Richter <robert.richter@amd.com>
2008-12-09 00:21:32 +00:00
|
|
|
add_ibs_begin(cpu, IBS_FETCH_CODE, mm);
|
2008-09-26 21:50:31 +00:00
|
|
|
break;
|
|
|
|
case IBS_OP_BEGIN:
|
2008-07-22 19:08:54 +00:00
|
|
|
state = sb_bt_start;
|
oprofile: port to the new ring_buffer
This patch replaces the current oprofile cpu buffer implementation
with the ring buffer provided by the tracing framework. The motivation
here is to leave the pain of implementing ring buffers to others. Oh,
no, there are more advantages. Main reason is the support of different
sample sizes that could be stored in the buffer. Use cases for this
are IBS and Cell spu profiling. Using the new ring buffer ensures
valid and complete samples and allows copying the cpu buffer stateless
without knowing its content. Second it will use generic kernel API and
also reduce code size. And hopefully, there are less bugs.
Since the new tracing ring buffer implementation uses spin locks to
protect the buffer during read/write access, it is difficult to use
the buffer in an NMI handler. In this case, writing to the buffer by
the NMI handler (x86) could occur also during critical sections when
reading the buffer. To avoid this, there are 2 buffers for independent
read and write access. Read access is in process context only, write
access only in the NMI handler. If the read buffer runs empty, both
buffers are swapped atomically. There is potentially a small window
during swapping where the buffers are disabled and samples could be
lost.
Using 2 buffers is a little bit overhead, but the solution is clear
and does not require changes in the ring buffer implementation. It can
be changed to a single buffer solution when the ring buffer access is
implemented as non-locking atomic code.
The new buffer requires more size to store the same amount of samples
because each sample includes an u32 header. Also, there is more code
to execute for buffer access. Nonetheless, the buffer implementation
is proven in the ftrace environment and worth to use also in oprofile.
Patches that changes the internal IBS buffer usage will follow.
Cc: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Robert Richter <robert.richter@amd.com>
2008-12-09 00:21:32 +00:00
|
|
|
add_ibs_begin(cpu, IBS_OP_CODE, mm);
|
2008-09-26 21:50:31 +00:00
|
|
|
break;
|
2008-07-22 19:09:06 +00:00
|
|
|
#endif
|
2008-09-26 21:50:31 +00:00
|
|
|
default:
|
2005-04-16 22:20:36 +00:00
|
|
|
/* userspace context switch */
|
2008-09-26 21:50:31 +00:00
|
|
|
oldmm = mm;
|
2005-04-16 22:20:36 +00:00
|
|
|
new = (struct task_struct *)s->event;
|
|
|
|
release_mm(oldmm);
|
|
|
|
mm = take_tasks_mm(new);
|
|
|
|
if (mm != oldmm)
|
|
|
|
cookie = get_exec_dcookie(mm);
|
|
|
|
add_user_ctx_switch(new, cookie);
|
2008-09-26 21:50:31 +00:00
|
|
|
break;
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2008-07-22 19:08:51 +00:00
|
|
|
} else if (state >= sb_bt_start &&
|
|
|
|
!add_sample(mm, s, in_kernel)) {
|
|
|
|
if (state == sb_bt_start) {
|
|
|
|
state = sb_bt_ignore;
|
|
|
|
atomic_inc(&oprofile_stats.bt_lost_no_mapping);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
release_mm(mm);
|
|
|
|
|
|
|
|
mark_done(cpu);
|
|
|
|
|
2006-06-25 12:47:33 +00:00
|
|
|
mutex_unlock(&buffer_mutex);
|
2005-04-16 22:20:36 +00:00
|
|
|
}
|
2008-10-14 23:37:01 +00:00
|
|
|
|
|
|
|
/* The function can be used to add a buffer worth of data directly to
|
|
|
|
* the kernel buffer. The buffer is assumed to be a circular buffer.
|
|
|
|
* Take the entries from index start and end at index end, wrapping
|
|
|
|
* at max_entries.
|
|
|
|
*/
|
|
|
|
void oprofile_put_buff(unsigned long *buf, unsigned int start,
|
|
|
|
unsigned int stop, unsigned int max)
|
|
|
|
{
|
|
|
|
int i;
|
|
|
|
|
|
|
|
i = start;
|
|
|
|
|
|
|
|
mutex_lock(&buffer_mutex);
|
|
|
|
while (i != stop) {
|
|
|
|
add_event_entry(buf[i++]);
|
|
|
|
|
|
|
|
if (i >= max)
|
|
|
|
i = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
mutex_unlock(&buffer_mutex);
|
|
|
|
}
|
|
|
|
|