Re: [RFC PATCH v2 1/3] tracing: Show last module text symbols in the stacktrace

[Google]

On Thu, 6 Mar 2025 19:46:27 -0500
Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:

> On Tue, 11 Feb 2025 00:43:57 +0900
> "Masami Hiramatsu (Google)" <mhiramat@xxxxxxxxxx> wrote:
> 
> > From: Masami Hiramatsu (Google) <mhiramat@xxxxxxxxxx>
> > 
> > Since the previous boot trace buffer can include module text address in
> > the stacktrace. As same as the kernel text address, convert the module
> > text address using the module address information.
> > 
> > Signed-off-by: Masami Hiramatsu (Google) <mhiramat@xxxxxxxxxx>
> > ---
> >  Changes in v2:
> >   - Add LRU based removed module information override.
> 
> I don't think we should bother supporting removal of modules yet.
> 
> It also makes this patch way too complex for what it should do.
> 
> >   - Fix to move module_delta initialization after tr->scratch check.
> 
> >   - Fix to make mod_delta on all loaded modules correctly.
> 
> Should be a separate patch.
> 
> >   - (Note, I don't add uname check yet)
> >   - Fix mod_addr_comp() overflow issue.
> >   - Add pr_info() when failed to allocate module_delta.
> 
> So basically this patch should just add the code to handle finding the
> module for stack traces. Not to mention, I have code that depends on that,
> but this patch currently does way to much.
> 
> Can you just send a patch that adds the trace_adjust_address() (although, I
> would shorten it to "trace_addr_adjust()").
> 
> And add what is needed to update the trace_stack_print().
> 
> You can base it off of the ring-buffer/for-next branch.

OK, BTW, if we always need tscratch, we can move module_delta in it.
Let's reimplement it.

Thanks,

> 
> Thanks,
> 
> -- Steve
> 
> 
> > ---
> >  kernel/trace/trace.c        |  173 ++++++++++++++++++++++++++++++++++++++++---
> >  kernel/trace/trace.h        |    4 +
> >  kernel/trace/trace_output.c |    3 -
> >  3 files changed, 167 insertions(+), 13 deletions(-)
> > 


-- 
Masami Hiramatsu (Google) <mhiramat@xxxxxxxxxx>