tracing/events: Convert format output to seq_file
authorSteven Rostedt <srostedt@redhat.com>
Thu, 3 Jun 2010 19:21:34 +0000 (15:21 -0400)
committerSteven Rostedt <rostedt@goodmis.org>
Thu, 12 Aug 2010 20:59:29 +0000 (16:59 -0400)
Two new events were added that broke the current format output.

Both from the SCSI system: scsi_dispatch_cmd_done and scsi_dispatch_cmd_timeout

The reason is that their print_fmt exceeded a page size. Since the output
of the format used simple_read_from_buffer and trace_seq, it was limited
to a page size in output.

This patch converts the printing of the format of an event into seq_file,
which allows greater than a page size to be shown.

I diffed all event formats comparing the output with and without this
patch. All matched except for the above two, which showed just:

  FORMAT TOO BIG

without this patch, but now properly displays the output with this patch.

v2: Remove updating *pos in seq start function.
   [ Thanks to Li Zefan for pointing that out ]

Reviewed-by: Li Zefan <lizf@cn.fujitsu.com>
Cc: Martin K. Petersen <martin.petersen@oracle.com>
Cc: Kei Tokunaga <tokunaga.keiich@jp.fujitsu.com>
Cc: James Bottomley <James.Bottomley@suse.de>
Cc: Tomohiro Kusumi <kusumi.tomohiro@jp.fujitsu.com>
Cc: Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
Signed-off-by: Steven Rostedt <rostedt@goodmis.org>
kernel/trace/trace_events.c

index 53cffc0b08014db76d9e87da0d8bbc3c9768bd49..45a8968707aa53a62e1fe9283937323384ae34c1 100644 (file)
@@ -29,6 +29,8 @@ DEFINE_MUTEX(event_mutex);
 
 LIST_HEAD(ftrace_events);
 
+#define COMMON_FIELD_COUNT     5
+
 struct list_head *
 trace_get_fields(struct ftrace_event_call *event_call)
 {
@@ -544,85 +546,155 @@ out:
        return ret;
 }
 
-static ssize_t
-event_format_read(struct file *filp, char __user *ubuf, size_t cnt,
-                 loff_t *ppos)
+enum {
+       FORMAT_HEADER           = 1,
+       FORMAT_PRINTFMT         = 2,
+};
+
+static void *f_next(struct seq_file *m, void *v, loff_t *pos)
 {
-       struct ftrace_event_call *call = filp->private_data;
+       struct ftrace_event_call *call = m->private;
        struct ftrace_event_field *field;
        struct list_head *head;
-       struct trace_seq *s;
-       int common_field_count = 5;
-       char *buf;
-       int r = 0;
-
-       if (*ppos)
-               return 0;
+       loff_t index = *pos;
 
-       s = kmalloc(sizeof(*s), GFP_KERNEL);
-       if (!s)
-               return -ENOMEM;
+       (*pos)++;
 
-       trace_seq_init(s);
+       head = trace_get_fields(call);
 
-       trace_seq_printf(s, "name: %s\n", call->name);
-       trace_seq_printf(s, "ID: %d\n", call->event.type);
-       trace_seq_printf(s, "format:\n");
+       switch ((unsigned long)v) {
+       case FORMAT_HEADER:
 
-       head = trace_get_fields(call);
-       list_for_each_entry_reverse(field, head, link) {
-               /*
-                * Smartly shows the array type(except dynamic array).
-                * Normal:
-                *      field:TYPE VAR
-                * If TYPE := TYPE[LEN], it is shown:
-                *      field:TYPE VAR[LEN]
-                */
-               const char *array_descriptor = strchr(field->type, '[');
-
-               if (!strncmp(field->type, "__data_loc", 10))
-                       array_descriptor = NULL;
-
-               if (!array_descriptor) {
-                       r = trace_seq_printf(s, "\tfield:%s %s;\toffset:%u;"
-                                       "\tsize:%u;\tsigned:%d;\n",
-                                       field->type, field->name, field->offset,
-                                       field->size, !!field->is_signed);
-               } else {
-                       r = trace_seq_printf(s, "\tfield:%.*s %s%s;\toffset:%u;"
-                                       "\tsize:%u;\tsigned:%d;\n",
-                                       (int)(array_descriptor - field->type),
-                                       field->type, field->name,
-                                       array_descriptor, field->offset,
-                                       field->size, !!field->is_signed);
-               }
+               if (unlikely(list_empty(head)))
+                       return NULL;
 
-               if (--common_field_count == 0)
-                       r = trace_seq_printf(s, "\n");
+               field = list_entry(head->prev, struct ftrace_event_field, link);
+               return field;
 
-               if (!r)
-                       break;
+       case FORMAT_PRINTFMT:
+               /* all done */
+               return NULL;
        }
 
-       if (r)
-               r = trace_seq_printf(s, "\nprint fmt: %s\n",
-                               call->print_fmt);
+       /*
+        * To separate common fields from event fields, the
+        * LSB is set on the first event field. Clear it in case.
+        */
+       v = (void *)((unsigned long)v & ~1L);
 
-       if (!r) {
-               /*
-                * ug!  The format output is bigger than a PAGE!!
-                */
-               buf = "FORMAT TOO BIG\n";
-               r = simple_read_from_buffer(ubuf, cnt, ppos,
-                                             buf, strlen(buf));
-               goto out;
+       field = v;
+       if (field->link.prev == head)
+               return (void *)FORMAT_PRINTFMT;
+
+       field = list_entry(field->link.prev, struct ftrace_event_field, link);
+
+       /* Set the LSB to notify f_show to print an extra newline */
+       if (index == COMMON_FIELD_COUNT)
+               field = (struct ftrace_event_field *)
+                       ((unsigned long)field | 1);
+
+       return field;
+}
+
+static void *f_start(struct seq_file *m, loff_t *pos)
+{
+       loff_t l = 0;
+       void *p;
+
+       /* Start by showing the header */
+       if (!*pos)
+               return (void *)FORMAT_HEADER;
+
+       p = (void *)FORMAT_HEADER;
+       do {
+               p = f_next(m, p, &l);
+       } while (p && l < *pos);
+
+       return p;
+}
+
+static int f_show(struct seq_file *m, void *v)
+{
+       struct ftrace_event_call *call = m->private;
+       struct ftrace_event_field *field;
+       const char *array_descriptor;
+
+       switch ((unsigned long)v) {
+       case FORMAT_HEADER:
+               seq_printf(m, "name: %s\n", call->name);
+               seq_printf(m, "ID: %d\n", call->event.type);
+               seq_printf(m, "format:\n");
+               return 0;
+
+       case FORMAT_PRINTFMT:
+               seq_printf(m, "\nprint fmt: %s\n",
+                          call->print_fmt);
+               return 0;
        }
 
-       r = simple_read_from_buffer(ubuf, cnt, ppos,
-                                   s->buffer, s->len);
- out:
-       kfree(s);
-       return r;
+       /*
+        * To separate common fields from event fields, the
+        * LSB is set on the first event field. Clear it and
+        * print a newline if it is set.
+        */
+       if ((unsigned long)v & 1) {
+               seq_putc(m, '\n');
+               v = (void *)((unsigned long)v & ~1L);
+       }
+
+       field = v;
+
+       /*
+        * Smartly shows the array type(except dynamic array).
+        * Normal:
+        *      field:TYPE VAR
+        * If TYPE := TYPE[LEN], it is shown:
+        *      field:TYPE VAR[LEN]
+        */
+       array_descriptor = strchr(field->type, '[');
+
+       if (!strncmp(field->type, "__data_loc", 10))
+               array_descriptor = NULL;
+
+       if (!array_descriptor)
+               seq_printf(m, "\tfield:%s %s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
+                          field->type, field->name, field->offset,
+                          field->size, !!field->is_signed);
+       else
+               seq_printf(m, "\tfield:%.*s %s%s;\toffset:%u;\tsize:%u;\tsigned:%d;\n",
+                          (int)(array_descriptor - field->type),
+                          field->type, field->name,
+                          array_descriptor, field->offset,
+                          field->size, !!field->is_signed);
+
+       return 0;
+}
+
+static void f_stop(struct seq_file *m, void *p)
+{
+}
+
+static const struct seq_operations trace_format_seq_ops = {
+       .start          = f_start,
+       .next           = f_next,
+       .stop           = f_stop,
+       .show           = f_show,
+};
+
+static int trace_format_open(struct inode *inode, struct file *file)
+{
+       struct ftrace_event_call *call = inode->i_private;
+       struct seq_file *m;
+       int ret;
+
+       ret = seq_open(file, &trace_format_seq_ops);
+       if (ret < 0)
+               return ret;
+
+       m = file->private_data;
+       m->private = call;
+
+       return 0;
 }
 
 static ssize_t
@@ -820,8 +892,10 @@ static const struct file_operations ftrace_enable_fops = {
 };
 
 static const struct file_operations ftrace_event_format_fops = {
-       .open = tracing_open_generic,
-       .read = event_format_read,
+       .open = trace_format_open,
+       .read = seq_read,
+       .llseek = seq_lseek,
+       .release = seq_release,
 };
 
 static const struct file_operations ftrace_event_id_fops = {