Hi guys,
I've been thinking about turning my f-trace.c USB function [1] into an
implementation of the USB Debug Class [2]. In order to support all
requests, I would need means to control parts of the tracing subsystem,
for example enabling and disabling tracing, flushing the trace buffer,
changing trace buffer size, choose which trace events to enable, etc.
Some of those functions (e.g. print_trace_line()) are available for
things like trace_kdb.c [3], but they're not available for modules;
i.e. they're not exported symbols.
This will involve a rather large change to tracing subsystem however we
could rely on such infrastructure for remote debugging of e.g. IVIs,
phones, etc.
The USB Debug Class already predicts the presence of JTAG access and
Hardware tracing modules to be exported over USB. It also allows for
GDB-type access. Adding our linux kernel tracing would make it into a
complete tracing solution.
In order to fully support this we would need, not only to expose some
internal trace functions for modules, but I suppose the trace format
would have to be slightly better defined so that it be decoded remotely
give access to vmlinux.
What do you guys think about this? Would this be something that folks
would be interested in? Currently we can only export ftrace data, not
trace events or anything else. Ideally we would export raw trace buffer
entries to be decoded at a later moment.
cheers
[1] https://marc.info/[email protected]
[2] https://www.usb.org/sites/default/files/documents/usb_debug_class_rev_1_0_final_0.pdf
[3] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/kernel/trace/trace_kdb.c#n20
--
balbi
Hi,
Felipe Balbi <[email protected]> writes:
> Hi guys,
>
> I've been thinking about turning my f-trace.c USB function [1] into an
> implementation of the USB Debug Class [2]. In order to support all
> requests, I would need means to control parts of the tracing subsystem,
> for example enabling and disabling tracing, flushing the trace buffer,
> changing trace buffer size, choose which trace events to enable, etc.
>
> Some of those functions (e.g. print_trace_line()) are available for
> things like trace_kdb.c [3], but they're not available for modules;
> i.e. they're not exported symbols.
>
> This will involve a rather large change to tracing subsystem however we
> could rely on such infrastructure for remote debugging of e.g. IVIs,
> phones, etc.
>
> The USB Debug Class already predicts the presence of JTAG access and
> Hardware tracing modules to be exported over USB. It also allows for
> GDB-type access. Adding our linux kernel tracing would make it into a
> complete tracing solution.
>
> In order to fully support this we would need, not only to expose some
> internal trace functions for modules, but I suppose the trace format
> would have to be slightly better defined so that it be decoded remotely
> give access to vmlinux.
>
> What do you guys think about this? Would this be something that folks
> would be interested in? Currently we can only export ftrace data, not
> trace events or anything else. Ideally we would export raw trace buffer
> entries to be decoded at a later moment.
>
> cheers
>
> [1] https://marc.info/[email protected]
> [2] https://www.usb.org/sites/default/files/documents/usb_debug_class_rev_1_0_final_0.pdf
> [3] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/kernel/trace/trace_kdb.c#n20
A gentle reminder here.
--
balbi
On Tue, 02 Apr 2019 08:44:31 +0300
Felipe Balbi <[email protected]> wrote:
> Hi,
>
> Felipe Balbi <[email protected]> writes:
Hi Felipe,
I somehow missed both your original email and this gentle reminder.
> > Hi guys,
> >
> > I've been thinking about turning my f-trace.c USB function [1] into an
> > implementation of the USB Debug Class [2]. In order to support all
> > requests, I would need means to control parts of the tracing subsystem,
> > for example enabling and disabling tracing, flushing the trace buffer,
> > changing trace buffer size, choose which trace events to enable, etc.
I believe a lot of these are already available.
Also note, that Divya made it possible to have your own tracing
instance for a module.
> >
> > Some of those functions (e.g. print_trace_line()) are available for
> > things like trace_kdb.c [3], but they're not available for modules;
> > i.e. they're not exported symbols.
> >
> > This will involve a rather large change to tracing subsystem however we
> > could rely on such infrastructure for remote debugging of e.g. IVIs,
> > phones, etc.
> >
> > The USB Debug Class already predicts the presence of JTAG access and
> > Hardware tracing modules to be exported over USB. It also allows for
> > GDB-type access. Adding our linux kernel tracing would make it into a
> > complete tracing solution.
> >
> > In order to fully support this we would need, not only to expose some
> > internal trace functions for modules, but I suppose the trace format
> > would have to be slightly better defined so that it be decoded remotely
> > give access to vmlinux.
> >
> > What do you guys think about this? Would this be something that folks
> > would be interested in? Currently we can only export ftrace data, not
> > trace events or anything else. Ideally we would export raw trace buffer
> > entries to be decoded at a later moment.
> >
Let me know what changes are needed and we can work something out.
Sorry for the delay, it appears I missed quite a bit of emails in this period :-/
-- Steve
> > cheers
> >
> > [1] https://marc.info/[email protected]
> > [2] https://www.usb.org/sites/default/files/documents/usb_debug_class_rev_1_0_final_0.pdf
> > [3] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/kernel/trace/trace_kdb.c#n20
>
> A gentle reminder here.
>