#include <linux/oprofile.h>
#include <linux/vmalloc.h>
#include <linux/errno.h>
-
+
#include "event_buffer.h"
#include "cpu_buffer.h"
#include "buffer_sync.h"
#include "oprof.h"
+#define OP_BUFFER_FLAGS 0
+
+/*
+ * Read and write access is using spin locking. Thus, writing to the
+ * buffer by 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.
+ */
+static struct ring_buffer *op_ring_buffer_read;
+static struct ring_buffer *op_ring_buffer_write;
DEFINE_PER_CPU(struct oprofile_cpu_buffer, cpu_buffer);
static void wq_sync_buffer(struct work_struct *work);
#define DEFAULT_TIMER_EXPIRE (HZ / 10)
static int work_enabled;
+unsigned long oprofile_get_cpu_buffer_size(void)
+{
+ return oprofile_cpu_buffer_size;
+}
+
+void oprofile_cpu_buffer_inc_smpl_lost(void)
+{
+ struct oprofile_cpu_buffer *cpu_buf
+ = &__get_cpu_var(cpu_buffer);
+
+ cpu_buf->sample_lost_overflow++;
+}
+
void free_cpu_buffers(void)
{
- int i;
-
- for_each_online_cpu(i) {
- vfree(per_cpu(cpu_buffer, i).buffer);
- per_cpu(cpu_buffer, i).buffer = NULL;
- }
+ if (op_ring_buffer_read)
+ ring_buffer_free(op_ring_buffer_read);
+ op_ring_buffer_read = NULL;
+ if (op_ring_buffer_write)
+ ring_buffer_free(op_ring_buffer_write);
+ op_ring_buffer_write = NULL;
}
int alloc_cpu_buffers(void)
{
int i;
-
- unsigned long buffer_size = fs_cpu_buffer_size;
-
- for_each_online_cpu(i) {
+
+ unsigned long buffer_size = oprofile_cpu_buffer_size;
+
+ op_ring_buffer_read = ring_buffer_alloc(buffer_size, OP_BUFFER_FLAGS);
+ if (!op_ring_buffer_read)
+ goto fail;
+ op_ring_buffer_write = ring_buffer_alloc(buffer_size, OP_BUFFER_FLAGS);
+ if (!op_ring_buffer_write)
+ goto fail;
+
+ for_each_possible_cpu(i) {
struct oprofile_cpu_buffer *b = &per_cpu(cpu_buffer, i);
-
- b->buffer = vmalloc_node(sizeof(struct op_sample) * buffer_size,
- cpu_to_node(i));
- if (!b->buffer)
- goto fail;
-
+
b->last_task = NULL;
b->last_is_kernel = -1;
b->tracing = 0;
b->buffer_size = buffer_size;
- b->tail_pos = 0;
- b->head_pos = 0;
b->sample_received = 0;
b->sample_lost_overflow = 0;
b->backtrace_aborted = 0;
flush_scheduled_work();
}
-/* Resets the cpu buffer to a sane state. */
-void cpu_buffer_reset(struct oprofile_cpu_buffer * cpu_buf)
+int op_cpu_buffer_write_entry(struct op_entry *entry)
{
- /* reset these to invalid values; the next sample
- * collected will populate the buffer with proper
- * values to initialize the buffer
- */
- cpu_buf->last_is_kernel = -1;
- cpu_buf->last_task = NULL;
+ entry->event = ring_buffer_lock_reserve(op_ring_buffer_write,
+ sizeof(struct op_sample),
+ &entry->irq_flags);
+ if (entry->event)
+ entry->sample = ring_buffer_event_data(entry->event);
+ else
+ entry->sample = NULL;
+
+ if (!entry->sample)
+ return -ENOMEM;
+
+ return 0;
}
-/* compute number of available slots in cpu_buffer queue */
-static unsigned long nr_available_slots(struct oprofile_cpu_buffer const * b)
+int op_cpu_buffer_write_commit(struct op_entry *entry)
{
- unsigned long head = b->head_pos;
- unsigned long tail = b->tail_pos;
+ return ring_buffer_unlock_commit(op_ring_buffer_write, entry->event,
+ entry->irq_flags);
+}
- if (tail > head)
- return (tail - head) - 1;
+struct op_sample *op_cpu_buffer_read_entry(int cpu)
+{
+ struct ring_buffer_event *e;
+ e = ring_buffer_consume(op_ring_buffer_read, cpu, NULL);
+ if (e)
+ return ring_buffer_event_data(e);
+ if (ring_buffer_swap_cpu(op_ring_buffer_read,
+ op_ring_buffer_write,
+ cpu))
+ return NULL;
+ e = ring_buffer_consume(op_ring_buffer_read, cpu, NULL);
+ if (e)
+ return ring_buffer_event_data(e);
+ return NULL;
+}
- return tail + (b->buffer_size - head) - 1;
+unsigned long op_cpu_buffer_entries(int cpu)
+{
+ return ring_buffer_entries_cpu(op_ring_buffer_read, cpu)
+ + ring_buffer_entries_cpu(op_ring_buffer_write, cpu);
}
-static void increment_head(struct oprofile_cpu_buffer * b)
+static inline int
+add_sample(struct oprofile_cpu_buffer *cpu_buf,
+ unsigned long pc, unsigned long event)
{
- unsigned long new_head = b->head_pos + 1;
+ struct op_entry entry;
+ int ret;
- /* Ensure anything written to the slot before we
- * increment is visible */
- wmb();
+ ret = op_cpu_buffer_write_entry(&entry);
+ if (ret)
+ return ret;
- if (new_head < b->buffer_size)
- b->head_pos = new_head;
- else
- b->head_pos = 0;
-}
+ entry.sample->eip = pc;
+ entry.sample->event = event;
-static inline void
-add_sample(struct oprofile_cpu_buffer * cpu_buf,
- unsigned long pc, unsigned long event)
-{
- struct op_sample * entry = &cpu_buf->buffer[cpu_buf->head_pos];
- entry->eip = pc;
- entry->event = event;
- increment_head(cpu_buf);
+ return op_cpu_buffer_write_commit(&entry);
}
-static inline void
-add_code(struct oprofile_cpu_buffer * buffer, unsigned long value)
+static inline int
+add_code(struct oprofile_cpu_buffer *buffer, unsigned long value)
{
- add_sample(buffer, ESCAPE_CODE, value);
+ return add_sample(buffer, ESCAPE_CODE, value);
}
/* This must be safe from any context. It's safe writing here
* pc. We tag this in the buffer by generating kernel enter/exit
* events whenever is_kernel changes
*/
-static int log_sample(struct oprofile_cpu_buffer * cpu_buf, unsigned long pc,
+static int log_sample(struct oprofile_cpu_buffer *cpu_buf, unsigned long pc,
int is_kernel, unsigned long event)
{
- struct task_struct * task;
+ struct task_struct *task;
cpu_buf->sample_received++;
return 0;
}
- if (nr_available_slots(cpu_buf) < 3) {
- cpu_buf->sample_lost_overflow++;
- return 0;
- }
-
is_kernel = !!is_kernel;
task = current;
/* notice a switch from user->kernel or vice versa */
if (cpu_buf->last_is_kernel != is_kernel) {
cpu_buf->last_is_kernel = is_kernel;
- add_code(cpu_buf, is_kernel);
+ if (add_code(cpu_buf, is_kernel))
+ goto fail;
}
/* notice a task switch */
if (cpu_buf->last_task != task) {
cpu_buf->last_task = task;
- add_code(cpu_buf, (unsigned long)task);
+ if (add_code(cpu_buf, (unsigned long)task))
+ goto fail;
}
-
- add_sample(cpu_buf, pc, event);
+
+ if (add_sample(cpu_buf, pc, event))
+ goto fail;
+
return 1;
+
+fail:
+ cpu_buf->sample_lost_overflow++;
+ return 0;
}
-static int oprofile_begin_trace(struct oprofile_cpu_buffer *cpu_buf)
+static inline void oprofile_begin_trace(struct oprofile_cpu_buffer *cpu_buf)
{
- if (nr_available_slots(cpu_buf) < 4) {
- cpu_buf->sample_lost_overflow++;
- return 0;
- }
-
add_code(cpu_buf, CPU_TRACE_BEGIN);
cpu_buf->tracing = 1;
- return 1;
}
-static void oprofile_end_trace(struct oprofile_cpu_buffer * cpu_buf)
+static inline void oprofile_end_trace(struct oprofile_cpu_buffer *cpu_buf)
{
cpu_buf->tracing = 0;
}
-void oprofile_add_ext_sample(unsigned long pc, struct pt_regs * const regs,
- unsigned long event, int is_kernel)
+static inline void
+__oprofile_add_ext_sample(unsigned long pc, struct pt_regs * const regs,
+ unsigned long event, int is_kernel)
{
struct oprofile_cpu_buffer *cpu_buf = &__get_cpu_var(cpu_buffer);
- if (!backtrace_depth) {
+ if (!oprofile_backtrace_depth) {
log_sample(cpu_buf, pc, is_kernel, event);
return;
}
- if (!oprofile_begin_trace(cpu_buf))
- return;
+ oprofile_begin_trace(cpu_buf);
- /* if log_sample() fail we can't backtrace since we lost the source
- * of this event */
+ /*
+ * if log_sample() fail we can't backtrace since we lost the
+ * source of this event
+ */
if (log_sample(cpu_buf, pc, is_kernel, event))
- oprofile_ops.backtrace(regs, backtrace_depth);
+ oprofile_ops.backtrace(regs, oprofile_backtrace_depth);
+
oprofile_end_trace(cpu_buf);
}
+void oprofile_add_ext_sample(unsigned long pc, struct pt_regs * const regs,
+ unsigned long event, int is_kernel)
+{
+ __oprofile_add_ext_sample(pc, regs, event, is_kernel);
+}
+
void oprofile_add_sample(struct pt_regs * const regs, unsigned long event)
{
int is_kernel = !user_mode(regs);
unsigned long pc = profile_pc(regs);
- oprofile_add_ext_sample(pc, regs, event, is_kernel);
+ __oprofile_add_ext_sample(pc, regs, event, is_kernel);
}
#ifdef CONFIG_OPROFILE_IBS
-#define MAX_IBS_SAMPLE_SIZE 14
-
-void oprofile_add_ibs_sample(struct pt_regs *const regs,
+void oprofile_add_ibs_sample(struct pt_regs * const regs,
unsigned int * const ibs_sample, int ibs_code)
{
int is_kernel = !user_mode(regs);
struct oprofile_cpu_buffer *cpu_buf = &__get_cpu_var(cpu_buffer);
struct task_struct *task;
+ int fail = 0;
cpu_buf->sample_received++;
- if (nr_available_slots(cpu_buf) < MAX_IBS_SAMPLE_SIZE) {
- /* we can't backtrace since we lost the source of this event */
- cpu_buf->sample_lost_overflow++;
- return;
- }
-
/* notice a switch from user->kernel or vice versa */
if (cpu_buf->last_is_kernel != is_kernel) {
+ if (add_code(cpu_buf, is_kernel))
+ goto fail;
cpu_buf->last_is_kernel = is_kernel;
- add_code(cpu_buf, is_kernel);
}
/* notice a task switch */
if (!is_kernel) {
task = current;
if (cpu_buf->last_task != task) {
+ if (add_code(cpu_buf, (unsigned long)task))
+ goto fail;
cpu_buf->last_task = task;
- add_code(cpu_buf, (unsigned long)task);
}
}
- add_code(cpu_buf, ibs_code);
- add_sample(cpu_buf, ibs_sample[0], ibs_sample[1]);
- add_sample(cpu_buf, ibs_sample[2], ibs_sample[3]);
- add_sample(cpu_buf, ibs_sample[4], ibs_sample[5]);
+ fail = fail || add_code(cpu_buf, ibs_code);
+ fail = fail || add_sample(cpu_buf, ibs_sample[0], ibs_sample[1]);
+ fail = fail || add_sample(cpu_buf, ibs_sample[2], ibs_sample[3]);
+ fail = fail || add_sample(cpu_buf, ibs_sample[4], ibs_sample[5]);
if (ibs_code == IBS_OP_BEGIN) {
- add_sample(cpu_buf, ibs_sample[6], ibs_sample[7]);
- add_sample(cpu_buf, ibs_sample[8], ibs_sample[9]);
- add_sample(cpu_buf, ibs_sample[10], ibs_sample[11]);
+ fail = fail || add_sample(cpu_buf, ibs_sample[6], ibs_sample[7]);
+ fail = fail || add_sample(cpu_buf, ibs_sample[8], ibs_sample[9]);
+ fail = fail || add_sample(cpu_buf, ibs_sample[10], ibs_sample[11]);
}
- if (backtrace_depth)
- oprofile_ops.backtrace(regs, backtrace_depth);
+ if (!fail)
+ return;
+
+fail:
+ cpu_buf->sample_lost_overflow++;
}
#endif
if (!cpu_buf->tracing)
return;
- if (nr_available_slots(cpu_buf) < 1) {
- cpu_buf->tracing = 0;
- cpu_buf->sample_lost_overflow++;
- return;
- }
+ /*
+ * broken frame can give an eip with the same value as an
+ * escape code, abort the trace if we get it
+ */
+ if (pc == ESCAPE_CODE)
+ goto fail;
- /* broken frame can give an eip with the same value as an escape code,
- * abort the trace if we get it */
- if (pc == ESCAPE_CODE) {
- cpu_buf->tracing = 0;
- cpu_buf->backtrace_aborted++;
- return;
- }
+ if (add_sample(cpu_buf, pc, 0))
+ goto fail;
- add_sample(cpu_buf, pc, 0);
+ return;
+fail:
+ cpu_buf->tracing = 0;
+ cpu_buf->backtrace_aborted++;
+ return;
}
/*
*/
static void wq_sync_buffer(struct work_struct *work)
{
- struct oprofile_cpu_buffer * b =
+ struct oprofile_cpu_buffer *b =
container_of(work, struct oprofile_cpu_buffer, work.work);
if (b->cpu != smp_processor_id()) {
printk(KERN_DEBUG "WQ on CPU%d, prefer CPU%d\n",
smp_processor_id(), b->cpu);
+
+ if (!cpu_online(b->cpu)) {
+ cancel_delayed_work(&b->work);
+ return;
+ }
}
sync_buffer(b->cpu);
git://ftp.safe.ca / safe / jmp / linux-2.6 / blobdiff