lkml.org 
[lkml]   [2021]   [Oct]   [13]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [RFC PATCH] tracing: BTF testing for kprobe-events
On Mon, 11 Oct 2021 18:23:34 -0400
Steven Rostedt <rostedt@goodmis.org> wrote:

> On Thu, 23 Sep 2021 21:39:43 +0900
> Masami Hiramatsu <mhiramat@kernel.org> wrote:
>
> > Hi Steve,
> >
>
> Hi Masami,
>
> Sorry for the late reply, but Plumbers followed by OSS put me way behind,
> and I just got to this email :-/
>
> > Here I share my testing patch of the BTF for kprobe events.
> > Currently this only allow user to specify '$$args' for
> > tracing all arguments of the function. This is only
> > avaialbe if
> > - the probe point is on the function entry
> > - the kernel is compiled with BTF (CONFIG_DEBUG_INFO_BTF)
> > - the kernel is enables BPF (CONFIG_BPF_SYSCALL)
> >
> > And Special thanks to Sven! Most of BTF handling part of
> > this patch comes from his patch [1]
> >
> > [1] https://stackframe.org/0001-ftrace-arg-hack.patch
>
> Which is newer than this patch because he sent a v2, and that's a couple
> patches down in my queue. I'll be looking at that one shortly as well.

Did he send his BTF hack patch to you ?
I didn't notice that.


> > What I thought while coding this were;
> > - kernel/bpf/btf.c can be moved under lib/ so that
> > the other subsystems can reuse it, independent
> > from BPF. (Also, this should depends on CONFIG_DEBUG_INFO_BTF)
>
> Makes sense.
>
> > - some more utility functions can be exposed.
> > e.g. I copied btf_type_int() from btf.c
>
> Agreed.
>
> > - If there are more comments for the BTF APIs, it will
> > be more useful...
> > - Overall, the BTF is easy to understand for who
> > already understand DWARF. Great work!
>
> Great to hear.
>
> > - I think I need 'ptr' and 'bool' types for fetcharg types.
> >
> > Anyway, this is just for testing. I have to add some
> > more cleanup, features and documentations, etc.
>
> This is awesome, and something to look at for a generic ftrace args point
> of view too.
>
> One issue is how do we handle multiple register values? Like a u64 type on
> 32 bit? As $arg1 is just a register that is in $arg1, for a u64 parameter
> on 32 bit, that is usually handled with two registers.
>
> Have thoughts on that?

Oh, that's a good point! The probe event supports such case, since I expected
the user will use 2 arguments to record it. But indeed, using BTF means we need
such extension.
OK, let me consider how to extend fetchargs to support it.

Thank you!

>
> I'll play with your patch today.
>
> -- Steve


--
Masami Hiramatsu <mhiramat@kernel.org>

\
 
 \ /
  Last update: 2021-10-13 14:52    [W:0.078 / U:0.284 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site