[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-ID: <20251110234157.808080361@kernel.org>
Date: Mon, 10 Nov 2025 18:41:57 -0500
From: Steven Rostedt <rostedt@...nel.org>
To: linux-kernel@...r.kernel.org,
linux-trace-kernel@...r.kernel.org
Cc: Masami Hiramatsu <mhiramat@...nel.org>,
Mark Rutland <mark.rutland@....com>,
Mathieu Desnoyers <mathieu.desnoyers@...icios.com>,
Andrew Morton <akpm@...ux-foundation.org>
Subject: [PATCH v3 0/3] tracing: Fix tracer options per instance
The tracers (like function and function_graph) have options that modify
how they behave or how they are shown in the trace file. As these options
were created before instances (multiple buffers), the flags variable that
represent these options were global for the tracer.
Now that a tracer may exist in multiple instances, having a global flags
that kinda affect every instance of that tracer does not make sense.
The reason it "kinda" affects them, is that the code that updates an option
for an instance, only does the change for the tracer running in that instance.
If a tracer is already running in another instance, it will not be affected.
But because the option is saved in a global variable, if the tracer is stopped
and restarted in the instance, then the option will take affect.
Since each instance has its own interface to modify the options, the effect
of an option should only affect the tracer for the given instance. This means
that the state of the options for each tracer must be saved for each instance
separately.
Add a new struct tracers field in the trace_array (instance descriptor) that
holds a descriptor that points to a tracer and contains its flags values. This
will be used to see what tracers are available for an instance.
Now when a tracer option is modified for an instance, it only affects the
tracer in that instance and not semi-effects the tracers in other instances.
Changes since v2: https://lore.kernel.org/linux-trace-kernel/20251106171557.011333928@kernel.org/
- Fix check of tracer not found in instance. The loop is a
list_for_each_entry() and testing the value of the entry for NULL is not
going to work.
git://git.kernel.org/pub/scm/linux/kernel/git/trace/linux-trace.git
trace/core
Head SHA1: 23857cfa3f9e6c505bf6b7e7cc4344710801209c
Steven Rostedt (3):
tracing: Have tracer option be instance specific
tracing: Have function tracer define options per instance
tracing: Have function graph tracer define options per instance
----
kernel/trace/trace.c | 221 ++++++++++++++++++++++++-----------
kernel/trace/trace.h | 3 +
kernel/trace/trace_functions.c | 10 +-
kernel/trace/trace_functions_graph.c | 18 +--
4 files changed, 169 insertions(+), 83 deletions(-)
Powered by blists - more mailing lists