ring-buffer: fix timestamp in partial ring_buffer_page_read
authorSteven Rostedt <srostedt@redhat.com>
Wed, 4 Mar 2009 04:52:42 +0000 (23:52 -0500)
committerSteven Rostedt <srostedt@redhat.com>
Thu, 5 Mar 2009 00:01:45 +0000 (19:01 -0500)
If a partial ring_buffer_page_read happens, then some of the
incremental timestamps may be lost. This patch writes the
recent timestamp into the page that is passed back to the caller.

A partial ring_buffer_page_read is where the full page would not
be written back to the user, and instead, just part of the page
is copied to the user. A full page would be a page swap with the
ring buffer and the timestamps would be correct.

Signed-off-by: Steven Rostedt <srostedt@redhat.com>
kernel/trace/ring_buffer.c

index f2a163d..f747364 100644 (file)
@@ -2461,6 +2461,7 @@ int ring_buffer_read_page(struct ring_buffer *buffer,
        unsigned long flags;
        unsigned int commit;
        unsigned int read;
+       u64 save_timestamp;
        int ret = -1;
 
        /*
@@ -2515,6 +2516,9 @@ int ring_buffer_read_page(struct ring_buffer *buffer,
                if (len < size)
                        goto out;
 
+               /* save the current timestamp, since the user will need it */
+               save_timestamp = cpu_buffer->read_stamp;
+
                /* Need to copy one event at a time */
                do {
                        memcpy(bpage->data + pos, rpage->data + rpos, size);
@@ -2531,7 +2535,7 @@ int ring_buffer_read_page(struct ring_buffer *buffer,
 
                /* update bpage */
                local_set(&bpage->commit, pos);
-               bpage->time_stamp = rpage->time_stamp;
+               bpage->time_stamp = save_timestamp;
 
                /* we copied everything to the beginning */
                read = 0;