linux-sg2042/drivers/oprofile/buffer_sync.c

598 lines
14 KiB
C
Raw Normal View History

/**
* @file buffer_sync.c
*
* @remark Copyright 2002-2009 OProfile authors
* @remark Read the file COPYING
*
* @author John Levon <levon@movementarian.org>
* @author Barry Kasindorf
* @author Robert Richter <robert.richter@amd.com>
*
* 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 cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h percpu.h is included by sched.h and module.h and thus ends up being included when building most .c files. percpu.h includes slab.h which in turn includes gfp.h making everything defined by the two files universally available and complicating inclusion dependencies. percpu.h -> slab.h dependency is about to be removed. Prepare for this change by updating users of gfp and slab facilities include those headers directly instead of assuming availability. As this conversion needs to touch large number of source files, the following script is used as the basis of conversion. http://userweb.kernel.org/~tj/misc/slabh-sweep.py The script does the followings. * Scan files for gfp and slab usages and update includes such that only the necessary includes are there. ie. if only gfp is used, gfp.h, if slab is used, slab.h. * When the script inserts a new include, it looks at the include blocks and try to put the new include such that its order conforms to its surrounding. It's put in the include block which contains core kernel includes, in the same order that the rest are ordered - alphabetical, Christmas tree, rev-Xmas-tree or at the end if there doesn't seem to be any matching order. * If the script can't find a place to put a new include (mostly because the file doesn't have fitting include block), it prints out an error message indicating which .h file needs to be added to the file. The conversion was done in the following steps. 1. The initial automatic conversion of all .c files updated slightly over 4000 files, deleting around 700 includes and adding ~480 gfp.h and ~3000 slab.h inclusions. The script emitted errors for ~400 files. 2. Each error was manually checked. Some didn't need the inclusion, some needed manual addition while adding it to implementation .h or embedding .c file was more appropriate for others. This step added inclusions to around 150 files. 3. The script was run again and the output was compared to the edits from #2 to make sure no file was left behind. 4. Several build tests were done and a couple of problems were fixed. e.g. lib/decompress_*.c used malloc/free() wrappers around slab APIs requiring slab.h to be added manually. 5. The script was run on all .h files but without automatically editing them as sprinkling gfp.h and slab.h inclusions around .h files could easily lead to inclusion dependency hell. Most gfp.h inclusion directives were ignored as stuff from gfp.h was usually wildly available and often used in preprocessor macros. Each slab.h inclusion directive was examined and added manually as necessary. 6. percpu.h was updated not to include slab.h. 7. Build test were done on the following configurations and failures were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my distributed build env didn't work with gcov compiles) and a few more options had to be turned off depending on archs to make things build (like ipr on powerpc/64 which failed due to missing writeq). * x86 and x86_64 UP and SMP allmodconfig and a custom test config. * powerpc and powerpc64 SMP allmodconfig * sparc and sparc64 SMP allmodconfig * ia64 SMP allmodconfig * s390 SMP allmodconfig * alpha SMP allmodconfig * um on x86_64 SMP allmodconfig 8. percpu.h modifications were reverted so that it could be applied as a separate patch and serve as bisection point. Given the fact that I had only a couple of failures from tests on step 6, I'm fairly confident about the coverage of this conversion patch. If there is a breakage, it's likely to be something in one of the arch headers which should be easily discoverable easily on most builds of the specific arch. Signed-off-by: Tejun Heo <tj@kernel.org> Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
2010-03-24 16:04:11 +08:00
#include <linux/gfp.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_var_t marked_cpus;
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,
};
int sync_start(void)
{
int err;
if (!zalloc_cpumask_var(&marked_cpus, GFP_KERNEL))
return -ENOMEM;
mutex_lock(&buffer_mutex);
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;
start_cpu_work();
out:
mutex_unlock(&buffer_mutex);
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:
free_cpumask_var(marked_cpus);
goto out;
}
void sync_stop(void)
{
/* flush buffers */
mutex_lock(&buffer_mutex);
end_cpu_work();
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);
mutex_unlock(&buffer_mutex);
flush_cpu_work();
/* make sure we don't leak task structs */
process_task_mortuary();
process_task_mortuary();
free_cpumask_var(marked_cpus);
}
/* 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 16:33:43 +08:00
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);
}
static void add_data(struct op_entry *entry, struct mm_struct *mm)
{
unsigned long code, pc, val;
unsigned long cookie;
off_t offset;
if (!op_cpu_buffer_get_data(entry, &code))
return;
if (!op_cpu_buffer_get_data(entry, &pc))
return;
if (!op_cpu_buffer_get_size(entry))
return;
if (mm) {
cookie = lookup_dcookie(mm, pc, &offset);
if (cookie == NO_COOKIE)
offset = pc;
if (cookie == INVALID_COOKIE) {
atomic_inc(&oprofile_stats.sample_lost_no_mapping);
offset = pc;
}
if (cookie != last_cookie) {
add_cookie_switch(cookie);
last_cookie = cookie;
}
} else
offset = pc;
add_event_entry(ESCAPE_CODE);
add_event_entry(code);
add_event_entry(offset); /* Offset from Dcookie */
while (op_cpu_buffer_get_data(entry, &val))
add_event_entry(val);
}
static inline void add_sample_entry(unsigned long offset, unsigned long event)
{
add_event_entry(offset);
add_event_entry(event);
}
/*
* 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. Return 0 on failure.
*/
static int
add_sample(struct mm_struct *mm, struct op_sample *s, int in_kernel)
{
unsigned long cookie;
off_t offset;
if (in_kernel) {
add_sample_entry(s->eip, s->event);
return 1;
}
/* add userspace sample */
if (!mm) {
atomic_inc(&oprofile_stats.sample_lost_no_mm);
return 0;
}
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;
}
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;
cpumask_set_cpu(cpu, marked_cpus);
for_each_online_cpu(i) {
if (!cpumask_test_cpu(i, marked_cpus))
return;
}
/* All CPUs have been processed at least once,
* we can process the mortuary once
*/
process_task_mortuary();
cpumask_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;
unsigned long val;
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;
unsigned long flags;
struct op_entry entry;
struct op_sample *sample;
mutex_lock(&buffer_mutex);
add_cpu_switch(cpu);
op_cpu_buffer_reset(cpu);
available = op_cpu_buffer_entries(cpu);
for (i = 0; i < available; ++i) {
sample = op_cpu_buffer_read_entry(&entry, cpu);
if (!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 08:21:32 +08:00
break;
if (is_code(sample->eip)) {
flags = sample->event;
if (flags & TRACE_BEGIN) {
state = sb_bt_start;
add_trace_begin();
}
if (flags & KERNEL_CTX_SWITCH) {
/* kernel/userspace switch */
in_kernel = flags & IS_KERNEL;
if (state == sb_buffer_start)
state = sb_sample_start;
add_kernel_ctx_switch(flags & IS_KERNEL);
}
if (flags & USER_CTX_SWITCH
&& op_cpu_buffer_get_data(&entry, &val)) {
/* userspace context switch */
new = (struct task_struct *)val;
oldmm = mm;
release_mm(oldmm);
mm = take_tasks_mm(new);
if (mm != oldmm)
cookie = get_exec_dcookie(mm);
add_user_ctx_switch(new, cookie);
}
if (op_cpu_buffer_get_size(&entry))
add_data(&entry, mm);
continue;
}
if (state < sb_bt_start)
/* ignore sample */
continue;
if (add_sample(mm, sample, in_kernel))
continue;
/* ignore backtraces if failed to add a sample */
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);
}