On Thu, Dec 23, 2021 at 12:07:54PM +0100, Vladimir Sementsov-Ogievskiy wrote:
Add possibility to generate trace points for each qmp command.
We should generate both trace points and trace-events file, for further
trace point code generation.
Signed-off-by: Vladimir Sementsov-Ogievskiy <vsementsov@virtuozzo.com>
---
scripts/qapi/commands.py | 84 ++++++++++++++++++++++++++++++++++------
1 file changed, 73 insertions(+), 11 deletions(-)
diff --git a/scripts/qapi/commands.py b/scripts/qapi/commands.py
index 21001bbd6b..9691c11f96 100644
--- a/scripts/qapi/commands.py
+++ b/scripts/qapi/commands.py
@@ -53,7 +53,8 @@ def gen_command_decl(name: str,
def gen_call(name: str,
arg_type: Optional[QAPISchemaObjectType],
boxed: bool,
- ret_type: Optional[QAPISchemaType]) -> str:
+ ret_type: Optional[QAPISchemaType],
+ add_trace_points: bool) -> str:
Please use the term "trace events" instead of "trace points". That's the
term that docs/devel/tracing.rst uses.
@@ -122,10 +167,14 @@ def gen_marshal_decl(name: str) -> str:
proto=build_marshal_proto(name))
+def gen_trace(name: str) -> str:
+ return f'qmp_{c_name(name)}(const char *tag, const char *json) "%s%s"\n'
This trace event is emitted in 3 different ways:
1. For arguments before calling a QMP command.
2. For the error message when the QMP command fails.
3. For the return value when a QMP command succeeds.
This makes parsing the trace akward because you get two events in
succession for a single call and they both have the same name.
Please generate 2 trace events:
1. qmp_enter_<name> <args>
2. qmp_exit_<name> <ret> <succeeded>
(That's similar to how the syscalls Linux kernel trace events work.)
Scripts processing the trace can easily differentiate between enter
(args) and exit (return value) events without parsing or keeping state
to count the second event.