* Thanks to Peter Zijlstra for suggesting this idea.
*/
struct buffer_page {
- union {
- struct {
- unsigned long flags; /* mandatory */
- atomic_t _count; /* mandatory */
- u64 time_stamp; /* page time stamp */
- unsigned size; /* size of page data */
- struct list_head list; /* list of free pages */
- };
- struct page page;
- };
+ u64 time_stamp; /* page time stamp */
+ local_t write; /* index for next write */
+ local_t commit; /* write commited index */
+ unsigned read; /* index for next read */
+ struct list_head list; /* list of free pages */
+ void *page; /* Actual data page */
};
+/*
+ * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing
+ * this issue out.
+ */
+static inline void free_buffer_page(struct buffer_page *bpage)
+{
+ if (bpage->page)
+ free_page((unsigned long)bpage->page);
+ kfree(bpage);
+}
+
/*
* We need to fit the time_stamp delta into 27 bits.
*/
spinlock_t lock;
struct lock_class_key lock_key;
struct list_head pages;
- unsigned long head; /* read from head */
- unsigned long tail; /* write to tail */
- struct buffer_page *head_page;
- struct buffer_page *tail_page;
+ struct buffer_page *head_page; /* read from head */
+ struct buffer_page *tail_page; /* write to tail */
+ struct buffer_page *commit_page; /* commited pages */
+ struct buffer_page *reader_page;
unsigned long overrun;
unsigned long entries;
u64 write_stamp;
u64 read_stamp;
};
-#define RB_WARN_ON(buffer, cond) \
- if (unlikely(cond)) { \
- atomic_inc(&buffer->record_disabled); \
- WARN_ON(1); \
- return -1; \
- }
+#define RB_WARN_ON(buffer, cond) \
+ do { \
+ if (unlikely(cond)) { \
+ atomic_inc(&buffer->record_disabled); \
+ WARN_ON(1); \
+ } \
+ } while (0)
+
+#define RB_WARN_ON_RET(buffer, cond) \
+ do { \
+ if (unlikely(cond)) { \
+ atomic_inc(&buffer->record_disabled); \
+ WARN_ON(1); \
+ return -1; \
+ } \
+ } while (0)
+
+#define RB_WARN_ON_ONCE(buffer, cond) \
+ do { \
+ static int once; \
+ if (unlikely(cond) && !once) { \
+ once++; \
+ atomic_inc(&buffer->record_disabled); \
+ WARN_ON(1); \
+ } \
+ } while (0)
/**
* check_pages - integrity check of buffer pages
struct list_head *head = &cpu_buffer->pages;
struct buffer_page *page, *tmp;
- RB_WARN_ON(cpu_buffer, head->next->prev != head);
- RB_WARN_ON(cpu_buffer, head->prev->next != head);
+ RB_WARN_ON_RET(cpu_buffer, head->next->prev != head);
+ RB_WARN_ON_RET(cpu_buffer, head->prev->next != head);
list_for_each_entry_safe(page, tmp, head, list) {
- RB_WARN_ON(cpu_buffer, page->list.next->prev != &page->list);
- RB_WARN_ON(cpu_buffer, page->list.prev->next != &page->list);
+ RB_WARN_ON_RET(cpu_buffer,
+ page->list.next->prev != &page->list);
+ RB_WARN_ON_RET(cpu_buffer,
+ page->list.prev->next != &page->list);
}
return 0;
}
-static unsigned rb_head_size(struct ring_buffer_per_cpu *cpu_buffer)
-{
- return cpu_buffer->head_page->size;
-}
-
static int rb_allocate_pages(struct ring_buffer_per_cpu *cpu_buffer,
unsigned nr_pages)
{
unsigned i;
for (i = 0; i < nr_pages; i++) {
+ page = kzalloc_node(ALIGN(sizeof(*page), cache_line_size()),
+ GFP_KERNEL, cpu_to_node(cpu_buffer->cpu));
+ if (!page)
+ goto free_pages;
+ list_add(&page->list, &pages);
+
addr = __get_free_page(GFP_KERNEL);
if (!addr)
goto free_pages;
- page = (struct buffer_page *)virt_to_page(addr);
- list_add(&page->list, &pages);
+ page->page = (void *)addr;
}
list_splice(&pages, head);
free_pages:
list_for_each_entry_safe(page, tmp, &pages, list) {
list_del_init(&page->list);
- __free_page(&page->page);
+ free_buffer_page(page);
}
return -ENOMEM;
}
rb_allocate_cpu_buffer(struct ring_buffer *buffer, int cpu)
{
struct ring_buffer_per_cpu *cpu_buffer;
+ struct buffer_page *page;
+ unsigned long addr;
int ret;
cpu_buffer = kzalloc_node(ALIGN(sizeof(*cpu_buffer), cache_line_size()),
spin_lock_init(&cpu_buffer->lock);
INIT_LIST_HEAD(&cpu_buffer->pages);
+ page = kzalloc_node(ALIGN(sizeof(*page), cache_line_size()),
+ GFP_KERNEL, cpu_to_node(cpu));
+ if (!page)
+ goto fail_free_buffer;
+
+ cpu_buffer->reader_page = page;
+ addr = __get_free_page(GFP_KERNEL);
+ if (!addr)
+ goto fail_free_reader;
+ page->page = (void *)addr;
+
+ INIT_LIST_HEAD(&cpu_buffer->reader_page->list);
+
ret = rb_allocate_pages(cpu_buffer, buffer->pages);
if (ret < 0)
- goto fail_free_buffer;
+ goto fail_free_reader;
cpu_buffer->head_page
= list_entry(cpu_buffer->pages.next, struct buffer_page, list);
- cpu_buffer->tail_page
- = list_entry(cpu_buffer->pages.next, struct buffer_page, list);
+ cpu_buffer->tail_page = cpu_buffer->commit_page = cpu_buffer->head_page;
return cpu_buffer;
+ fail_free_reader:
+ free_buffer_page(cpu_buffer->reader_page);
+
fail_free_buffer:
kfree(cpu_buffer);
return NULL;
struct list_head *head = &cpu_buffer->pages;
struct buffer_page *page, *tmp;
+ list_del_init(&cpu_buffer->reader_page->list);
+ free_buffer_page(cpu_buffer->reader_page);
+
list_for_each_entry_safe(page, tmp, head, list) {
list_del_init(&page->list);
- __free_page(&page->page);
+ free_buffer_page(page);
}
kfree(cpu_buffer);
}
+/*
+ * Causes compile errors if the struct buffer_page gets bigger
+ * than the struct page.
+ */
+extern int ring_buffer_page_too_big(void);
+
/**
* ring_buffer_alloc - allocate a new ring_buffer
* @size: the size in bytes that is needed.
int bsize;
int cpu;
+ /* Paranoid! Optimizes out when all is well */
+ if (sizeof(struct buffer_page) > sizeof(struct page))
+ ring_buffer_page_too_big();
+
+
/* keep it in its own cache line */
buffer = kzalloc(ALIGN(sizeof(*buffer), cache_line_size()),
GFP_KERNEL);
p = cpu_buffer->pages.next;
page = list_entry(p, struct buffer_page, list);
list_del_init(&page->list);
- __free_page(&page->page);
+ free_buffer_page(page);
}
BUG_ON(list_empty(&cpu_buffer->pages));
for_each_buffer_cpu(buffer, cpu) {
for (i = 0; i < new_pages; i++) {
+ page = kzalloc_node(ALIGN(sizeof(*page),
+ cache_line_size()),
+ GFP_KERNEL, cpu_to_node(cpu));
+ if (!page)
+ goto free_pages;
+ list_add(&page->list, &pages);
addr = __get_free_page(GFP_KERNEL);
if (!addr)
goto free_pages;
- page = (struct buffer_page *)virt_to_page(addr);
- list_add(&page->list, &pages);
+ page->page = (void *)addr;
}
}
free_pages:
list_for_each_entry_safe(page, tmp, &pages, list) {
list_del_init(&page->list);
- __free_page(&page->page);
+ free_buffer_page(page);
}
return -ENOMEM;
}
-static inline int rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer)
-{
- return cpu_buffer->head_page == cpu_buffer->tail_page &&
- cpu_buffer->head == cpu_buffer->tail;
-}
-
static inline int rb_null_event(struct ring_buffer_event *event)
{
return event->type == RINGBUF_TYPE_PADDING;
}
-static inline void *rb_page_index(struct buffer_page *page, unsigned index)
+static inline void *__rb_page_index(struct buffer_page *page, unsigned index)
{
- void *addr = page_address(&page->page);
+ return page->page + index;
+}
- return addr + index;
+static inline struct ring_buffer_event *
+rb_reader_event(struct ring_buffer_per_cpu *cpu_buffer)
+{
+ return __rb_page_index(cpu_buffer->reader_page,
+ cpu_buffer->reader_page->read);
}
static inline struct ring_buffer_event *
rb_head_event(struct ring_buffer_per_cpu *cpu_buffer)
{
- return rb_page_index(cpu_buffer->head_page,
- cpu_buffer->head);
+ return __rb_page_index(cpu_buffer->head_page,
+ cpu_buffer->head_page->read);
}
static inline struct ring_buffer_event *
rb_iter_head_event(struct ring_buffer_iter *iter)
{
- return rb_page_index(iter->head_page,
- iter->head);
+ return __rb_page_index(iter->head_page, iter->head);
+}
+
+static inline unsigned rb_page_write(struct buffer_page *bpage)
+{
+ return local_read(&bpage->write);
+}
+
+static inline unsigned rb_page_commit(struct buffer_page *bpage)
+{
+ return local_read(&bpage->commit);
+}
+
+/* Size is determined by what has been commited */
+static inline unsigned rb_page_size(struct buffer_page *bpage)
+{
+ return rb_page_commit(bpage);
+}
+
+static inline unsigned
+rb_commit_index(struct ring_buffer_per_cpu *cpu_buffer)
+{
+ return rb_page_commit(cpu_buffer->commit_page);
+}
+
+static inline unsigned rb_head_size(struct ring_buffer_per_cpu *cpu_buffer)
+{
+ return rb_page_commit(cpu_buffer->head_page);
}
/*
for (head = 0; head < rb_head_size(cpu_buffer);
head += rb_event_length(event)) {
- event = rb_page_index(cpu_buffer->head_page, head);
+ event = __rb_page_index(cpu_buffer->head_page, head);
BUG_ON(rb_null_event(event));
/* Only count data entries */
if (event->type != RINGBUF_TYPE_DATA)
*page = list_entry(p, struct buffer_page, list);
}
+static inline unsigned
+rb_event_index(struct ring_buffer_event *event)
+{
+ unsigned long addr = (unsigned long)event;
+
+ return (addr & ~PAGE_MASK) - (PAGE_SIZE - BUF_PAGE_SIZE);
+}
+
+static inline int
+rb_is_commit(struct ring_buffer_per_cpu *cpu_buffer,
+ struct ring_buffer_event *event)
+{
+ unsigned long addr = (unsigned long)event;
+ unsigned long index;
+
+ index = rb_event_index(event);
+ addr &= PAGE_MASK;
+
+ return cpu_buffer->commit_page->page == (void *)addr &&
+ rb_commit_index(cpu_buffer) == index;
+}
+
+static inline void
+rb_set_commit_event(struct ring_buffer_per_cpu *cpu_buffer,
+ struct ring_buffer_event *event)
+{
+ unsigned long addr = (unsigned long)event;
+ unsigned long index;
+
+ index = rb_event_index(event);
+ addr &= PAGE_MASK;
+
+ while (cpu_buffer->commit_page->page != (void *)addr) {
+ RB_WARN_ON(cpu_buffer,
+ cpu_buffer->commit_page == cpu_buffer->tail_page);
+ cpu_buffer->commit_page->commit =
+ cpu_buffer->commit_page->write;
+ rb_inc_page(cpu_buffer, &cpu_buffer->commit_page);
+ cpu_buffer->write_stamp = cpu_buffer->commit_page->time_stamp;
+ }
+
+ /* Now set the commit to the event's index */
+ local_set(&cpu_buffer->commit_page->commit, index);
+}
+
static inline void
-rb_add_stamp(struct ring_buffer_per_cpu *cpu_buffer, u64 *ts)
+rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer)
{
- cpu_buffer->tail_page->time_stamp = *ts;
- cpu_buffer->write_stamp = *ts;
+ /*
+ * We only race with interrupts and NMIs on this CPU.
+ * If we own the commit event, then we can commit
+ * all others that interrupted us, since the interruptions
+ * are in stack format (they finish before they come
+ * back to us). This allows us to do a simple loop to
+ * assign the commit to the tail.
+ */
+ while (cpu_buffer->commit_page != cpu_buffer->tail_page) {
+ cpu_buffer->commit_page->commit =
+ cpu_buffer->commit_page->write;
+ rb_inc_page(cpu_buffer, &cpu_buffer->commit_page);
+ cpu_buffer->write_stamp = cpu_buffer->commit_page->time_stamp;
+ /* add barrier to keep gcc from optimizing too much */
+ barrier();
+ }
+ while (rb_commit_index(cpu_buffer) !=
+ rb_page_write(cpu_buffer->commit_page)) {
+ cpu_buffer->commit_page->commit =
+ cpu_buffer->commit_page->write;
+ barrier();
+ }
}
-static void rb_reset_read_page(struct ring_buffer_per_cpu *cpu_buffer)
+static void rb_reset_reader_page(struct ring_buffer_per_cpu *cpu_buffer)
{
- cpu_buffer->read_stamp = cpu_buffer->head_page->time_stamp;
- cpu_buffer->head = 0;
+ cpu_buffer->read_stamp = cpu_buffer->reader_page->time_stamp;
+ cpu_buffer->reader_page->read = 0;
}
-static void
-rb_reset_iter_read_page(struct ring_buffer_iter *iter)
+static inline void rb_inc_iter(struct ring_buffer_iter *iter)
{
+ struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
+
+ /*
+ * The iterator could be on the reader page (it starts there).
+ * But the head could have moved, since the reader was
+ * found. Check for this case and assign the iterator
+ * to the head page instead of next.
+ */
+ if (iter->head_page == cpu_buffer->reader_page)
+ iter->head_page = cpu_buffer->head_page;
+ else
+ rb_inc_page(cpu_buffer, &iter->head_page);
+
iter->read_stamp = iter->head_page->time_stamp;
iter->head = 0;
}
__rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer,
unsigned type, unsigned long length, u64 *ts)
{
- struct buffer_page *head_page, *tail_page;
- unsigned long tail;
+ struct buffer_page *tail_page, *head_page, *reader_page;
+ unsigned long tail, write;
struct ring_buffer *buffer = cpu_buffer->buffer;
struct ring_buffer_event *event;
+ unsigned long flags;
tail_page = cpu_buffer->tail_page;
- head_page = cpu_buffer->head_page;
- tail = cpu_buffer->tail;
+ write = local_add_return(length, &tail_page->write);
+ tail = write - length;
- if (tail + length > BUF_PAGE_SIZE) {
+ /* See if we shot pass the end of this buffer page */
+ if (write > BUF_PAGE_SIZE) {
struct buffer_page *next_page = tail_page;
+ spin_lock_irqsave(&cpu_buffer->lock, flags);
+
rb_inc_page(cpu_buffer, &next_page);
- if (next_page == head_page) {
- if (!(buffer->flags & RB_FL_OVERWRITE))
- return NULL;
+ head_page = cpu_buffer->head_page;
+ reader_page = cpu_buffer->reader_page;
- /* count overflows */
- rb_update_overflow(cpu_buffer);
+ /* we grabbed the lock before incrementing */
+ RB_WARN_ON(cpu_buffer, next_page == reader_page);
- rb_inc_page(cpu_buffer, &head_page);
- cpu_buffer->head_page = head_page;
- rb_reset_read_page(cpu_buffer);
+ /*
+ * If for some reason, we had an interrupt storm that made
+ * it all the way around the buffer, bail, and warn
+ * about it.
+ */
+ if (unlikely(next_page == cpu_buffer->commit_page)) {
+ WARN_ON_ONCE(1);
+ goto out_unlock;
}
- if (tail != BUF_PAGE_SIZE) {
- event = rb_page_index(tail_page, tail);
- /* page padding */
+ if (next_page == head_page) {
+ if (!(buffer->flags & RB_FL_OVERWRITE)) {
+ /* reset write */
+ if (tail <= BUF_PAGE_SIZE)
+ local_set(&tail_page->write, tail);
+ goto out_unlock;
+ }
+
+ /* tail_page has not moved yet? */
+ if (tail_page == cpu_buffer->tail_page) {
+ /* count overflows */
+ rb_update_overflow(cpu_buffer);
+
+ rb_inc_page(cpu_buffer, &head_page);
+ cpu_buffer->head_page = head_page;
+ cpu_buffer->head_page->read = 0;
+ }
+ }
+
+ /*
+ * If the tail page is still the same as what we think
+ * it is, then it is up to us to update the tail
+ * pointer.
+ */
+ if (tail_page == cpu_buffer->tail_page) {
+ local_set(&next_page->write, 0);
+ local_set(&next_page->commit, 0);
+ cpu_buffer->tail_page = next_page;
+
+ /* reread the time stamp */
+ *ts = ring_buffer_time_stamp(cpu_buffer->cpu);
+ cpu_buffer->tail_page->time_stamp = *ts;
+ }
+
+ /*
+ * The actual tail page has moved forward.
+ */
+ if (tail < BUF_PAGE_SIZE) {
+ /* Mark the rest of the page with padding */
+ event = __rb_page_index(tail_page, tail);
event->type = RINGBUF_TYPE_PADDING;
}
- tail_page->size = tail;
- tail_page = next_page;
- tail_page->size = 0;
- tail = 0;
- cpu_buffer->tail_page = tail_page;
- cpu_buffer->tail = tail;
- rb_add_stamp(cpu_buffer, ts);
+ if (tail <= BUF_PAGE_SIZE)
+ /* Set the write back to the previous setting */
+ local_set(&tail_page->write, tail);
+
+ /*
+ * If this was a commit entry that failed,
+ * increment that too
+ */
+ if (tail_page == cpu_buffer->commit_page &&
+ tail == rb_commit_index(cpu_buffer)) {
+ rb_set_commit_to_write(cpu_buffer);
+ }
+
+ spin_unlock_irqrestore(&cpu_buffer->lock, flags);
+
+ /* fail and let the caller try again */
+ return ERR_PTR(-EAGAIN);
}
- BUG_ON(tail + length > BUF_PAGE_SIZE);
+ /* We reserved something on the buffer */
- event = rb_page_index(tail_page, tail);
+ BUG_ON(write > BUF_PAGE_SIZE);
+
+ event = __rb_page_index(tail_page, tail);
rb_update_event(event, type, length);
+ /*
+ * If this is a commit and the tail is zero, then update
+ * this page's time stamp.
+ */
+ if (!tail && rb_is_commit(cpu_buffer, event))
+ cpu_buffer->commit_page->time_stamp = *ts;
+
return event;
+
+ out_unlock:
+ spin_unlock_irqrestore(&cpu_buffer->lock, flags);
+ return NULL;
}
static int
{
struct ring_buffer_event *event;
static int once;
+ int ret;
if (unlikely(*delta > (1ULL << 59) && !once++)) {
printk(KERN_WARNING "Delta way too big! %llu"
" ts=%llu write stamp = %llu\n",
- *delta, *ts, cpu_buffer->write_stamp);
+ (unsigned long long)*delta,
+ (unsigned long long)*ts,
+ (unsigned long long)cpu_buffer->write_stamp);
WARN_ON(1);
}
RB_LEN_TIME_EXTEND,
ts);
if (!event)
- return -1;
+ return -EBUSY;
- /* check to see if we went to the next page */
- if (cpu_buffer->tail) {
- /* Still on same page, update timestamp */
- event->time_delta = *delta & TS_MASK;
- event->array[0] = *delta >> TS_SHIFT;
- /* commit the time event */
- cpu_buffer->tail +=
- rb_event_length(event);
+ if (PTR_ERR(event) == -EAGAIN)
+ return -EAGAIN;
+
+ /* Only a commited time event can update the write stamp */
+ if (rb_is_commit(cpu_buffer, event)) {
+ /*
+ * If this is the first on the page, then we need to
+ * update the page itself, and just put in a zero.
+ */
+ if (rb_event_index(event)) {
+ event->time_delta = *delta & TS_MASK;
+ event->array[0] = *delta >> TS_SHIFT;
+ } else {
+ cpu_buffer->commit_page->time_stamp = *ts;
+ event->time_delta = 0;
+ event->array[0] = 0;
+ }
cpu_buffer->write_stamp = *ts;
- *delta = 0;
+ /* let the caller know this was the commit */
+ ret = 1;
+ } else {
+ /* Darn, this is just wasted space */
+ event->time_delta = 0;
+ event->array[0] = 0;
+ ret = 0;
}
- return 0;
+ *delta = 0;
+
+ return ret;
}
static struct ring_buffer_event *
{
struct ring_buffer_event *event;
u64 ts, delta;
+ int commit = 0;
+ int nr_loops = 0;
+
+ again:
+ /*
+ * We allow for interrupts to reenter here and do a trace.
+ * If one does, it will cause this original code to loop
+ * back here. Even with heavy interrupts happening, this
+ * should only happen a few times in a row. If this happens
+ * 1000 times in a row, there must be either an interrupt
+ * storm or we have something buggy.
+ * Bail!
+ */
+ if (unlikely(++nr_loops > 1000)) {
+ RB_WARN_ON(cpu_buffer, 1);
+ return NULL;
+ }
ts = ring_buffer_time_stamp(cpu_buffer->cpu);
- if (cpu_buffer->tail) {
+ /*
+ * Only the first commit can update the timestamp.
+ * Yes there is a race here. If an interrupt comes in
+ * just after the conditional and it traces too, then it
+ * will also check the deltas. More than one timestamp may
+ * also be made. But only the entry that did the actual
+ * commit will be something other than zero.
+ */
+ if (cpu_buffer->tail_page == cpu_buffer->commit_page &&
+ rb_page_write(cpu_buffer->tail_page) ==
+ rb_commit_index(cpu_buffer)) {
+
delta = ts - cpu_buffer->write_stamp;
+ /* make sure this delta is calculated here */
+ barrier();
+
+ /* Did the write stamp get updated already? */
+ if (unlikely(ts < cpu_buffer->write_stamp))
+ goto again;
+
if (test_time_stamp(delta)) {
- int ret;
- ret = rb_add_time_stamp(cpu_buffer, &ts, &delta);
- if (ret < 0)
+ commit = rb_add_time_stamp(cpu_buffer, &ts, &delta);
+
+ if (commit == -EBUSY)
return NULL;
+
+ if (commit == -EAGAIN)
+ goto again;
+
+ RB_WARN_ON(cpu_buffer, commit < 0);
}
- } else {
- rb_add_stamp(cpu_buffer, &ts);
+ } else
+ /* Non commits have zero deltas */
delta = 0;
- }
event = __rb_reserve_next(cpu_buffer, type, length, &ts);
- if (!event)
+ if (PTR_ERR(event) == -EAGAIN)
+ goto again;
+
+ if (!event) {
+ if (unlikely(commit))
+ /*
+ * Ouch! We needed a timestamp and it was commited. But
+ * we didn't get our event reserved.
+ */
+ rb_set_commit_to_write(cpu_buffer);
return NULL;
+ }
- /* If the reserve went to the next page, our delta is zero */
- if (!cpu_buffer->tail)
+ /*
+ * If the timestamp was commited, make the commit our entry
+ * now so that we will update it when needed.
+ */
+ if (commit)
+ rb_set_commit_event(cpu_buffer, event);
+ else if (!rb_is_commit(cpu_buffer, event))
delta = 0;
event->time_delta = delta;
return event;
}
+static DEFINE_PER_CPU(int, rb_need_resched);
+
/**
* ring_buffer_lock_reserve - reserve a part of the buffer
* @buffer: the ring buffer to reserve from
{
struct ring_buffer_per_cpu *cpu_buffer;
struct ring_buffer_event *event;
- int cpu;
+ int cpu, resched;
if (atomic_read(&buffer->record_disabled))
return NULL;
- raw_local_irq_save(*flags);
+ /* If we are tracing schedule, we don't want to recurse */
+ resched = need_resched();
+ preempt_disable_notrace();
+
cpu = raw_smp_processor_id();
if (!cpu_isset(cpu, buffer->cpumask))
- goto out_irq;
+ goto out;
cpu_buffer = buffer->buffers[cpu];
- spin_lock(&cpu_buffer->lock);
if (atomic_read(&cpu_buffer->record_disabled))
- goto no_record;
+ goto out;
length = rb_calculate_event_length(length);
if (length > BUF_PAGE_SIZE)
- return NULL;
+ goto out;
event = rb_reserve_next_event(cpu_buffer, RINGBUF_TYPE_DATA, length);
if (!event)
- goto no_record;
+ goto out;
+
+ /*
+ * Need to store resched state on this cpu.
+ * Only the first needs to.
+ */
+
+ if (preempt_count() == 1)
+ per_cpu(rb_need_resched, cpu) = resched;
return event;
- no_record:
- spin_unlock(&cpu_buffer->lock);
- out_irq:
- local_irq_restore(*flags);
+ out:
+ if (resched)
+ preempt_enable_notrace();
+ else
+ preempt_enable_notrace();
return NULL;
}
static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer,
struct ring_buffer_event *event)
{
- cpu_buffer->tail += rb_event_length(event);
- cpu_buffer->tail_page->size = cpu_buffer->tail;
- cpu_buffer->write_stamp += event->time_delta;
cpu_buffer->entries++;
+
+ /* Only process further if we own the commit */
+ if (!rb_is_commit(cpu_buffer, event))
+ return;
+
+ cpu_buffer->write_stamp += event->time_delta;
+
+ rb_set_commit_to_write(cpu_buffer);
}
/**
cpu_buffer = buffer->buffers[cpu];
- assert_spin_locked(&cpu_buffer->lock);
-
rb_commit(cpu_buffer, event);
- spin_unlock(&cpu_buffer->lock);
- raw_local_irq_restore(flags);
+ /*
+ * Only the last preempt count needs to restore preemption.
+ */
+ if (preempt_count() == 1) {
+ if (per_cpu(rb_need_resched, cpu))
+ preempt_enable_no_resched_notrace();
+ else
+ preempt_enable_notrace();
+ } else
+ preempt_enable_no_resched_notrace();
return 0;
}
{
struct ring_buffer_per_cpu *cpu_buffer;
struct ring_buffer_event *event;
- unsigned long event_length, flags;
+ unsigned long event_length;
void *body;
int ret = -EBUSY;
- int cpu;
+ int cpu, resched;
if (atomic_read(&buffer->record_disabled))
return -EBUSY;
- local_irq_save(flags);
+ resched = need_resched();
+ preempt_disable_notrace();
+
cpu = raw_smp_processor_id();
if (!cpu_isset(cpu, buffer->cpumask))
- goto out_irq;
+ goto out;
cpu_buffer = buffer->buffers[cpu];
- spin_lock(&cpu_buffer->lock);
if (atomic_read(&cpu_buffer->record_disabled))
goto out;
ret = 0;
out:
- spin_unlock(&cpu_buffer->lock);
- out_irq:
- local_irq_restore(flags);
+ if (resched)
+ preempt_enable_no_resched_notrace();
+ else
+ preempt_enable_notrace();
return ret;
}
-/**
- * ring_buffer_lock - lock the ring buffer
- * @buffer: The ring buffer to lock
- * @flags: The place to store the interrupt flags
- *
- * This locks all the per CPU buffers.
- *
- * Must be unlocked by ring_buffer_unlock.
- */
-void ring_buffer_lock(struct ring_buffer *buffer, unsigned long *flags)
-{
- struct ring_buffer_per_cpu *cpu_buffer;
- int cpu;
-
- local_irq_save(*flags);
-
- for_each_buffer_cpu(buffer, cpu) {
- cpu_buffer = buffer->buffers[cpu];
- spin_lock(&cpu_buffer->lock);
- }
-}
-
-/**
- * ring_buffer_unlock - unlock a locked buffer
- * @buffer: The locked buffer to unlock
- * @flags: The interrupt flags received by ring_buffer_lock
- */
-void ring_buffer_unlock(struct ring_buffer *buffer, unsigned long flags)
+static inline int rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer)
{
- struct ring_buffer_per_cpu *cpu_buffer;
- int cpu;
-
- for (cpu = buffer->cpus - 1; cpu >= 0; cpu--) {
- if (!cpu_isset(cpu, buffer->cpumask))
- continue;
- cpu_buffer = buffer->buffers[cpu];
- spin_unlock(&cpu_buffer->lock);
- }
+ struct buffer_page *reader = cpu_buffer->reader_page;
+ struct buffer_page *head = cpu_buffer->head_page;
+ struct buffer_page *commit = cpu_buffer->commit_page;
- local_irq_restore(flags);
+ return reader->read == rb_page_commit(reader) &&
+ (commit == reader ||
+ (commit == head &&
+ head->read == rb_page_commit(commit)));
}
/**
{
struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
- iter->head_page = cpu_buffer->head_page;
- iter->head = cpu_buffer->head;
- rb_reset_iter_read_page(iter);
+ /* Iterator usage is expected to have record disabled */
+ if (list_empty(&cpu_buffer->reader_page->list)) {
+ iter->head_page = cpu_buffer->head_page;
+ iter->head = cpu_buffer->head_page->read;
+ } else {
+ iter->head_page = cpu_buffer->reader_page;
+ iter->head = cpu_buffer->reader_page->read;
+ }
+ if (iter->head)
+ iter->read_stamp = cpu_buffer->read_stamp;
+ else
+ iter->read_stamp = iter->head_page->time_stamp;
}
/**
cpu_buffer = iter->cpu_buffer;
- return iter->head_page == cpu_buffer->tail_page &&
- iter->head == cpu_buffer->tail;
+ return iter->head_page == cpu_buffer->commit_page &&
+ iter->head == rb_commit_index(cpu_buffer);
}
static void
return;
}
-static void rb_advance_head(struct ring_buffer_per_cpu *cpu_buffer)
+static struct buffer_page *
+rb_get_reader_page(struct ring_buffer_per_cpu *cpu_buffer)
{
- struct ring_buffer_event *event;
- unsigned length;
+ struct buffer_page *reader = NULL;
+ unsigned long flags;
+ int nr_loops = 0;
+ spin_lock_irqsave(&cpu_buffer->lock, flags);
+
+ again:
/*
- * Check if we are at the end of the buffer.
+ * This should normally only loop twice. But because the
+ * start of the reader inserts an empty page, it causes
+ * a case where we will loop three times. There should be no
+ * reason to loop four times (that I know of).
*/
- if (cpu_buffer->head >= cpu_buffer->head_page->size) {
- BUG_ON(cpu_buffer->head_page == cpu_buffer->tail_page);
- rb_inc_page(cpu_buffer, &cpu_buffer->head_page);
- rb_reset_read_page(cpu_buffer);
- return;
+ if (unlikely(++nr_loops > 3)) {
+ RB_WARN_ON(cpu_buffer, 1);
+ reader = NULL;
+ goto out;
}
- event = rb_head_event(cpu_buffer);
+ reader = cpu_buffer->reader_page;
- if (event->type == RINGBUF_TYPE_DATA)
- cpu_buffer->entries--;
+ /* If there's more to read, return this page */
+ if (cpu_buffer->reader_page->read < rb_page_size(reader))
+ goto out;
- length = rb_event_length(event);
+ /* Never should we have an index greater than the size */
+ RB_WARN_ON(cpu_buffer,
+ cpu_buffer->reader_page->read > rb_page_size(reader));
+
+ /* check if we caught up to the tail */
+ reader = NULL;
+ if (cpu_buffer->commit_page == cpu_buffer->reader_page)
+ goto out;
/*
- * This should not be called to advance the header if we are
- * at the tail of the buffer.
+ * Splice the empty reader page into the list around the head.
+ * Reset the reader page to size zero.
*/
- BUG_ON((cpu_buffer->head_page == cpu_buffer->tail_page) &&
- (cpu_buffer->head + length > cpu_buffer->tail));
- rb_update_read_stamp(cpu_buffer, event);
+ reader = cpu_buffer->head_page;
+ cpu_buffer->reader_page->list.next = reader->list.next;
+ cpu_buffer->reader_page->list.prev = reader->list.prev;
- cpu_buffer->head += length;
+ local_set(&cpu_buffer->reader_page->write, 0);
+ local_set(&cpu_buffer->reader_page->commit, 0);
+
+ /* Make the reader page now replace the head */
+ reader->list.prev->next = &cpu_buffer->reader_page->list;
+ reader->list.next->prev = &cpu_buffer->reader_page->list;
+
+ /*
+ * If the tail is on the reader, then we must set the head
+ * to the inserted page, otherwise we set it one before.
+ */
+ cpu_buffer->head_page = cpu_buffer->reader_page;
- /* check for end of page */
- if ((cpu_buffer->head >= cpu_buffer->head_page->size) &&
- (cpu_buffer->head_page != cpu_buffer->tail_page))
- rb_advance_head(cpu_buffer);
+ if (cpu_buffer->commit_page != reader)
+ rb_inc_page(cpu_buffer, &cpu_buffer->head_page);
+
+ /* Finally update the reader page to the new head */
+ cpu_buffer->reader_page = reader;
+ rb_reset_reader_page(cpu_buffer);
+
+ goto again;
+
+ out:
+ spin_unlock_irqrestore(&cpu_buffer->lock, flags);
+
+ return reader;
+}
+
+static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer)
+{
+ struct ring_buffer_event *event;
+ struct buffer_page *reader;
+ unsigned length;
+
+ reader = rb_get_reader_page(cpu_buffer);
+
+ /* This function should not be called when buffer is empty */
+ BUG_ON(!reader);
+
+ event = rb_reader_event(cpu_buffer);
+
+ if (event->type == RINGBUF_TYPE_DATA)
+ cpu_buffer->entries--;
+
+ rb_update_read_stamp(cpu_buffer, event);
+
+ length = rb_event_length(event);
+ cpu_buffer->reader_page->read += length;
}
static void rb_advance_iter(struct ring_buffer_iter *iter)
/*
* Check if we are at the end of the buffer.
*/
- if (iter->head >= iter->head_page->size) {
- BUG_ON(iter->head_page == cpu_buffer->tail_page);
- rb_inc_page(cpu_buffer, &iter->head_page);
- rb_reset_iter_read_page(iter);
+ if (iter->head >= rb_page_size(iter->head_page)) {
+ BUG_ON(iter->head_page == cpu_buffer->commit_page);
+ rb_inc_iter(iter);
return;
}
* This should not be called to advance the header if we are
* at the tail of the buffer.
*/
- BUG_ON((iter->head_page == cpu_buffer->tail_page) &&
- (iter->head + length > cpu_buffer->tail));
+ BUG_ON((iter->head_page == cpu_buffer->commit_page) &&
+ (iter->head + length > rb_commit_index(cpu_buffer)));
rb_update_iter_read_stamp(iter, event);
iter->head += length;
/* check for end of page padding */
- if ((iter->head >= iter->head_page->size) &&
- (iter->head_page != cpu_buffer->tail_page))
+ if ((iter->head >= rb_page_size(iter->head_page)) &&
+ (iter->head_page != cpu_buffer->commit_page))
rb_advance_iter(iter);
}
{
struct ring_buffer_per_cpu *cpu_buffer;
struct ring_buffer_event *event;
+ struct buffer_page *reader;
+ int nr_loops = 0;
if (!cpu_isset(cpu, buffer->cpumask))
return NULL;
cpu_buffer = buffer->buffers[cpu];
again:
- if (rb_per_cpu_empty(cpu_buffer))
+ /*
+ * We repeat when a timestamp is encountered. It is possible
+ * to get multiple timestamps from an interrupt entering just
+ * as one timestamp is about to be written. The max times
+ * that this can happen is the number of nested interrupts we
+ * can have. Nesting 10 deep of interrupts is clearly
+ * an anomaly.
+ */
+ if (unlikely(++nr_loops > 10)) {
+ RB_WARN_ON(cpu_buffer, 1);
return NULL;
+ }
- event = rb_head_event(cpu_buffer);
+ reader = rb_get_reader_page(cpu_buffer);
+ if (!reader)
+ return NULL;
+
+ event = rb_reader_event(cpu_buffer);
switch (event->type) {
case RINGBUF_TYPE_PADDING:
- rb_inc_page(cpu_buffer, &cpu_buffer->head_page);
- rb_reset_read_page(cpu_buffer);
- goto again;
+ RB_WARN_ON(cpu_buffer, 1);
+ rb_advance_reader(cpu_buffer);
+ return NULL;
case RINGBUF_TYPE_TIME_EXTEND:
/* Internal data, OK to advance */
- rb_advance_head(cpu_buffer);
+ rb_advance_reader(cpu_buffer);
goto again;
case RINGBUF_TYPE_TIME_STAMP:
/* FIXME: not implemented */
- rb_advance_head(cpu_buffer);
+ rb_advance_reader(cpu_buffer);
goto again;
case RINGBUF_TYPE_DATA:
struct ring_buffer *buffer;
struct ring_buffer_per_cpu *cpu_buffer;
struct ring_buffer_event *event;
+ int nr_loops = 0;
if (ring_buffer_iter_empty(iter))
return NULL;
buffer = cpu_buffer->buffer;
again:
+ /*
+ * We repeat when a timestamp is encountered. It is possible
+ * to get multiple timestamps from an interrupt entering just
+ * as one timestamp is about to be written. The max times
+ * that this can happen is the number of nested interrupts we
+ * can have. Nesting 10 deep of interrupts is clearly
+ * an anomaly.
+ */
+ if (unlikely(++nr_loops > 10)) {
+ RB_WARN_ON(cpu_buffer, 1);
+ return NULL;
+ }
+
if (rb_per_cpu_empty(cpu_buffer))
return NULL;
switch (event->type) {
case RINGBUF_TYPE_PADDING:
- rb_inc_page(cpu_buffer, &iter->head_page);
- rb_reset_iter_read_page(iter);
+ rb_inc_iter(iter);
goto again;
case RINGBUF_TYPE_TIME_EXTEND:
return NULL;
cpu_buffer = buffer->buffers[cpu];
- rb_advance_head(cpu_buffer);
+ rb_advance_reader(cpu_buffer);
return event;
}
{
struct ring_buffer_per_cpu *cpu_buffer;
struct ring_buffer_iter *iter;
+ unsigned long flags;
if (!cpu_isset(cpu, buffer->cpumask))
return NULL;
atomic_inc(&cpu_buffer->record_disabled);
synchronize_sched();
- spin_lock(&cpu_buffer->lock);
- iter->head = cpu_buffer->head;
- iter->head_page = cpu_buffer->head_page;
- rb_reset_iter_read_page(iter);
- spin_unlock(&cpu_buffer->lock);
+ spin_lock_irqsave(&cpu_buffer->lock, flags);
+ ring_buffer_iter_reset(iter);
+ spin_unlock_irqrestore(&cpu_buffer->lock, flags);
return iter;
}
{
cpu_buffer->head_page
= list_entry(cpu_buffer->pages.next, struct buffer_page, list);
- cpu_buffer->tail_page
- = list_entry(cpu_buffer->pages.next, struct buffer_page, list);
+ local_set(&cpu_buffer->head_page->write, 0);
+ local_set(&cpu_buffer->head_page->commit, 0);
+
+ cpu_buffer->head_page->read = 0;
+
+ cpu_buffer->tail_page = cpu_buffer->head_page;
+ cpu_buffer->commit_page = cpu_buffer->head_page;
+
+ INIT_LIST_HEAD(&cpu_buffer->reader_page->list);
+ local_set(&cpu_buffer->reader_page->write, 0);
+ local_set(&cpu_buffer->reader_page->commit, 0);
+ cpu_buffer->reader_page->read = 0;
- cpu_buffer->head = cpu_buffer->tail = 0;
cpu_buffer->overrun = 0;
cpu_buffer->entries = 0;
}
if (!cpu_isset(cpu, buffer->cpumask))
return;
- raw_local_irq_save(flags);
- spin_lock(&cpu_buffer->lock);
+ spin_lock_irqsave(&cpu_buffer->lock, flags);
rb_reset_cpu(cpu_buffer);
- spin_unlock(&cpu_buffer->lock);
- raw_local_irq_restore(flags);
+ spin_unlock_irqrestore(&cpu_buffer->lock, flags);
}
/**
*/
void ring_buffer_reset(struct ring_buffer *buffer)
{
- unsigned long flags;
int cpu;
- ring_buffer_lock(buffer, &flags);
-
for_each_buffer_cpu(buffer, cpu)
- rb_reset_cpu(buffer->buffers[cpu]);
-
- ring_buffer_unlock(buffer, flags);
+ ring_buffer_reset_cpu(buffer, cpu);
}
/**