linux/drivers/oprofile/buffer_sync.c

621 lines
14 KiB
C
Raw Normal View History

/**
* @file buffer_sync.c
*
* @remark Copyright 2002 OProfile authors
* @remark Read the file COPYING
*
* @author John Levon <levon@movementarian.org>
* @author Barry Kasindorf
*
* 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>
#include <linux/oprofile.h>
#include <linux/sched.h>
#include "oprofile_stats.h"
#include "event_buffer.h"
#include "cpu_buffer.h"
#include "buffer_sync.h"
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.
* Can be invoked from softirq via RCU callback due to
* call_rcu() of the task struct, hence the _irqsave.
*/
static int
task_free_notify(struct notifier_block *self, unsigned long val, void *data)
{
unsigned long flags;
struct task_struct *task = data;
spin_lock_irqsave(&task_mortuary, flags);
list_add(&task->tasks, &dying_tasks);
spin_unlock_irqrestore(&task_mortuary, flags);
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.
*/
static int
task_exit_notify(struct notifier_block *self, unsigned long val, void *data)
{
/* To avoid latency problems, we only process the current CPU,
* hoping that most samples for the task are on this CPU
*/
sync_buffer(raw_smp_processor_id());
return 0;
}
/* 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.
*/
static int
munmap_notify(struct notifier_block *self, unsigned long val, void *data)
{
unsigned long addr = (unsigned long)data;
struct mm_struct *mm = current->mm;
struct vm_area_struct *mpnt;
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
*/
sync_buffer(raw_smp_processor_id());
return 0;
}
up_read(&mm->mmap_sem);
return 0;
}
/* 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.
*/
static int
module_load_notify(struct notifier_block *self, unsigned long val, void *data)
{
#ifdef CONFIG_MODULES
if (val != MODULE_STATE_COMING)
return 0;
/* FIXME: should we process all CPU buffers ? */
mutex_lock(&buffer_mutex);
add_event_entry(ESCAPE_CODE);
add_event_entry(MODULE_LOADED_CODE);
mutex_unlock(&buffer_mutex);
#endif
return 0;
}
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,
};
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();
}
/* 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). */
static inline unsigned long fast_get_dcookie(struct path *path)
{
unsigned long cookie;
shrink struct dentry struct dentry is one of the most critical structures in the kernel. So it's sad to see it going neglected. With CONFIG_PROFILING turned on (which is probably the common case at least for distros and kernel developers), sizeof(struct dcache) == 208 here (64-bit). This gives 19 objects per slab. I packed d_mounted into a hole, and took another 4 bytes off the inline name length to take the padding out from the end of the structure. This shinks it to 200 bytes. I could have gone the other way and increased the length to 40, but I'm aiming for a magic number, read on... I then got rid of the d_cookie pointer. This shrinks it to 192 bytes. Rant: why was this ever a good idea? The cookie system should increase its hash size or use a tree or something if lookups are a problem. Also the "fast dcookie lookups" in oprofile should be moved into the dcookie code -- how can oprofile possibly care about the dcookie_mutex? It gets dropped after get_dcookie() returns so it can't be providing any sort of protection. At 192 bytes, 21 objects fit into a 4K page, saving about 3MB on my system with ~140 000 entries allocated. 192 is also a multiple of 64, so we get nice cacheline alignment on 64 and 32 byte line systems -- any given dentry will now require 3 cachelines to touch all fields wheras previously it would require 4. I know the inline name size was chosen quite carefully, however with the reduction in cacheline footprint, it should actually be just about as fast to do a name lookup for a 36 character name as it was before the patch (and faster for other sizes). The memory footprint savings for names which are <= 32 or > 36 bytes long should more than make up for the memory cost for 33-36 byte names. Performance is a feature... Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
2008-12-01 08:33:43 +00:00
if (path->dentry->d_flags & DCACHE_COOKIE)
return (unsigned long)path->dentry;
get_dcookie(path, &cookie);
return cookie;
}
/* 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
*/
static unsigned long get_exec_dcookie(struct mm_struct *mm)
{
unsigned long cookie = NO_COOKIE;
struct vm_area_struct *vma;
if (!mm)
goto out;
for (vma = mm->mmap; vma; vma = vma->vm_next) {
if (!vma->vm_file)
continue;
if (!(vma->vm_flags & VM_EXECUTABLE))
continue;
cookie = fast_get_dcookie(&vma->vm_file->f_path);
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.
*/
static unsigned long
lookup_dcookie(struct mm_struct *mm, unsigned long addr, off_t *offset)
{
unsigned long cookie = NO_COOKIE;
struct vm_area_struct *vma;
for (vma = find_vma(mm, addr); vma; vma = vma->vm_next) {
if (addr < vma->vm_start || addr >= vma->vm_end)
continue;
if (vma->vm_file) {
cookie = fast_get_dcookie(&vma->vm_file->f_path);
*offset = (vma->vm_pgoff << PAGE_SHIFT) + addr -
vma->vm_start;
} else {
/* must be an anonymous map */
*offset = addr;
}
break;
}
if (!vma)
cookie = INVALID_COOKIE;
return cookie;
}
static unsigned long last_cookie = INVALID_COOKIE;
static void add_cpu_switch(int i)
{
add_event_entry(ESCAPE_CODE);
add_event_entry(CPU_SWITCH_CODE);
add_event_entry(i);
last_cookie = INVALID_COOKIE;
}
static void add_kernel_ctx_switch(unsigned int in_kernel)
{
add_event_entry(ESCAPE_CODE);
if (in_kernel)
add_event_entry(KERNEL_ENTER_SWITCH_CODE);
else
add_event_entry(KERNEL_EXIT_SWITCH_CODE);
}
static void
add_user_ctx_switch(struct task_struct const *task, unsigned long cookie)
{
add_event_entry(ESCAPE_CODE);
add_event_entry(CTX_SWITCH_CODE);
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);
}
static void add_cookie_switch(unsigned long cookie)
{
add_event_entry(ESCAPE_CODE);
add_event_entry(COOKIE_SWITCH_CODE);
add_event_entry(cookie);
}
static void add_trace_begin(void)
{
add_event_entry(ESCAPE_CODE);
add_event_entry(TRACE_BEGIN_CODE);
}
#ifdef CONFIG_OPROFILE_IBS
#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)
{
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;
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;
#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;
#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);
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);
}
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;
}
#endif
static void add_sample_entry(unsigned long offset, unsigned long event)
{
add_event_entry(offset);
add_event_entry(event);
}
static int add_us_sample(struct mm_struct *mm, struct op_sample *s)
{
unsigned long cookie;
off_t offset;
cookie = lookup_dcookie(mm, s->eip, &offset);
if (cookie == INVALID_COOKIE) {
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;
}
/* 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
add_sample(struct mm_struct *mm, struct op_sample *s, int in_kernel)
{
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;
}
static void release_mm(struct mm_struct *mm)
{
if (!mm)
return;
up_read(&mm->mmap_sem);
mmput(mm);
}
static struct mm_struct *take_tasks_mm(struct task_struct *task)
{
struct mm_struct *mm = get_task_mm(task);
if (mm)
down_read(&mm->mmap_sem);
return mm;
}
static inline int is_code(unsigned long val)
{
return val == ESCAPE_CODE;
}
/* 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)
{
unsigned long flags;
LIST_HEAD(local_dead_tasks);
struct task_struct *task;
struct task_struct *ttask;
spin_lock_irqsave(&task_mortuary, flags);
list_splice_init(&dead_tasks, &local_dead_tasks);
list_splice_init(&dying_tasks, &dead_tasks);
spin_unlock_irqrestore(&task_mortuary, flags);
list_for_each_entry_safe(task, ttask, &local_dead_tasks, tasks) {
list_del(&task->tasks);
free_task(task);
}
}
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;
struct mm_struct *oldmm;
struct task_struct *new;
unsigned long cookie = 0;
int in_kernel = 1;
sync_buffer_state state = sb_buffer_start;
unsigned int i;
unsigned long available;
mutex_lock(&buffer_mutex);
add_cpu_switch(cpu);
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);
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;
if (is_code(s->eip)) {
switch (s->event) {
case 0:
case CPU_IS_KERNEL:
/* kernel/userspace switch */
in_kernel = s->event;
if (state == sb_buffer_start)
state = sb_sample_start;
add_kernel_ctx_switch(s->event);
break;
case CPU_TRACE_BEGIN:
state = sb_bt_start;
add_trace_begin();
break;
#ifdef CONFIG_OPROFILE_IBS
case IBS_FETCH_BEGIN:
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);
break;
case IBS_OP_BEGIN:
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);
break;
#endif
default:
/* userspace context switch */
oldmm = mm;
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);
break;
}
} 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);
}
}
}
release_mm(mm);
mark_done(cpu);
mutex_unlock(&buffer_mutex);
}
/* 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);
}