trace-cmd: A couple of fixes
cover trace-cmd: A couple of fixes
Here's a couple of fixes to bugs that I found while playing with the p2p code.
Steven Rostedt (VMware) (2):
trace-cmd: Duplicate trace_clock in tracecmd_input handle tools lib traceevent: Add builtin handler for trace_marker_raw
----
lib/trace-cmd/trace-input.c | 7 ++++++ lib/traceevent/plugins/plugin_function.c | 41 ++++++++++++++++++++++++++++++++ 2 files changed, 48 insertions(+)
PATCH 1/2 trace-cmd: Duplicate trace_clock in tracecmd_input handle
From: "Steven Rostedt (VMware)" <[email protected]>
The following crashes:
# trace-cmd record -C local -e sched -B foo -e irq sleep 1 # trace-cmd report
The issue is that new instances are copied from the top instance descriptor and their values are set. This means that the trace_clock field is also copied which is a pointer to a string.
On freeing of the tracecmd_input handlers, the trace_clock is freed. This is an issue if the trace_clock was added as an option, because the instance just has a copy of the top instance, and when the instance descriptor is freed, it will free the same pointer that was already freed by the descruction of the top instance descriptor and we have a double free.
Have the creation of the instance tracecmd_input handler descriptor perform a strdup() and have its own copy of the trace_clock.
--- lib/trace-cmd/trace-input.c | 7 +++++++ 1 file changed, 7 insertions(+)
lib/trace-cmd/trace-input.c +7 -0
@@ -3398,6 +3398,13 @@ tracecmd_buffer_instance_handle(struct tracecmd_input *handle, int indx) new_handle->nr_buffers =0 ; new_handle->buffers =NULL ; new_handle->ref =1 ; +if (handle->trace_clock) { + new_handle->trace_clock =strdup (handle->trace_clock); +if (!new_handle->trace_clock) { +free (new_handle); +return NULL ; + } + } new_handle->parent = handle; new_handle->cpustats =NULL ; new_handle->hooks =NULL ;
Signature · 1 line
2.24.0
PATCH 2/2 tools lib traceevent: Add builtin handler for trace_marker_raw
In reply to Steven RostedtFrom: "Steven Rostedt (VMware)" <[email protected]>
When something is written into trace_marker_raw, it goes in as a binary. But the printk_fmt() of the event that is created (raw_data)'s format file only prints the first byte of data:
print fmt: "id:%04x %08x", REC->id, (int)REC->buf[0]
This is no very useful if we want to see the full data output.
Implement the processing of the raw_data event like it is in the kernel.
--- lib/traceevent/plugins/plugin_function.c | 41 ++++++++++++++++++++++++ 1 file changed, 41 insertions(+)
lib/traceevent/plugins/plugin_function.c +41 -0
@@ -207,6 +207,44 @@ trace_stack_handler(struct trace_seq *s, struct tep_record *record,return 0 ; } +static int +trace_raw_data_handler (struct trace_seq *s,struct tep_record *record, +struct tep_event *event,void *context) +{ +struct tep_format_field *field; +unsigned long long id; +int long_size; +void *data = record->data; + +if (tep_get_field_val (s, event,"id" , record, &id,1 )) +return trace_seq_putc (s,'!' ); + +trace_seq_printf (s,"# %llx" , id); + + field =tep_find_any_field (event,"buf" ); +if (!field) { +trace_seq_printf (s,"<CANT FIND FIELD %s>" ,"buf" ); +return 0 ; + } + + long_size =tep_get_long_size (event->tep); + +for (data += field->offset; data < record->data + record->size; + data += long_size) { +int size =sizeof (long ); +int left = (record->data + record->size) - data; +int i; + +if (size > left) + size = left; + +for (i =0 ; i < size; i++) +trace_seq_printf (s," %02x" , *(unsigned char *)(data + i)); + } + +return 0 ; +} +int TEP_PLUGIN_LOADER (struct tep_handle *tep) { tep_register_event_handler(tep, -1, "ftrace", "function",
@@ -215,6 +253,9 @@ int TEP_PLUGIN_LOADER(struct tep_handle *tep)tep_register_event_handler (tep,-1 ,"ftrace" ,"kernel_stack" , trace_stack_handler,NULL ); +tep_register_event_handler (tep,-1 ,"ftrace" ,"raw_data" , + trace_raw_data_handler,NULL ); +tep_plugin_add_options ("ftrace" , plugin_options);return 0 ;
Signature · 1 line
2.24.0