[RFC 0/8] trace: improve support

[Simon Glass]

Hi Heinrich,

On Sat, 20 Dec 2025 at 18:59, Heinrich Schuchardt
<heinrich.schuchardt at canonical.com> wrote:
>
> In the current state the trace functionality is not usable.
> With CONFIG_TRACE=y not even QEMU can boot.
>
> With this series CONFIG_TRACE becomes usable on
> qemu-riscv64_smode_defconfig.

I would be great to enable this feature on QEMU!

>
> It all starts with correcting the handling of recursive calls
> to the measuring functions at the entry and exit points.
>
> When a function is entered we record the time and reading the
> timer leads to a recursion which should not switch of recording

off

> traces but just be ignored.

See my comment on the patch.

>
> The meminfo patches are in the series because the output
> of the meminfo command depends on CONFIG_TRACE.
>
> Also the Pythontrace test does not consider that output depends on
> configuration.
>
> There are still issues with qemu-x86_64_defconfig when trying to
> trace. But I would like to collect early feedback.

We have sandbox tests for tracing, so eventually I suppose this could
be added to CI.

>
> Thanks to Patrick Rudolph for pointing to his issues with tracing.
>
> Heinrich Schuchardt (8):
>   trace: correct handling of recursive call
>   trace: initr_trace must succeed initr_dm
>   test: test_trace: remove invalid assert
>   trace: increase TRACE_CALL_DEPTH_LIMIT
>   cmd/meminfo: display of addresses above 4 GiB
>   test: cmd: consider configuration in meminfo test
>   test: initf_malloc is only traced with EARLY_TRACE
>   test: clean up test_trace.py code
>
>  cmd/meminfo.c               |  8 ++---
>  common/board_r.c            |  2 +-
>  lib/Kconfig                 |  2 +-
>  lib/trace.c                 | 59 +++++++++++++++++--------------------
>  test/cmd/meminfo.c          | 20 +++++++++----
>  test/py/tests/test_trace.py | 52 +++++++++++++-------------------
>  6 files changed, 68 insertions(+), 75 deletions(-)
>
> --
> 2.51.0
>

Regards,
Simon