On 2025-09-01 19:07, Alex Bennée wrote:
Pierrick Bouvier <[email protected]> writes:
On 2025-08-08 13:41, Pierrick Bouvier wrote:
This plugin generates a binary trace compatible with the excellent uftrace:
https://github.com/namhyung/uftrace
In short, it tracks all function calls performed during execution,
based on
frame pointer analysis. A big advantage over "uftrace record" is that it works
in system mode, allowing to trace a full system execution, which was the
original goal. It works as well in user mode, but uftrace itself already does
this. It's implemented for aarch64 and x86_64.
Let's start with concrete examples of the result.
First, in system mode, booting a stack using TF-A + U-boot + Linux:
- Two first stages of boot sequence in Arm Trusted Firmware (EL3 and S-EL1)
https://fileserver.linaro.org/s/kkxBS552W7nYESX/preview
- Stat and open syscalls in kernel
https://fileserver.linaro.org/s/dXe4MfraKg2F476/preview
- Poweroff sequence (from kernel back to firmware, NS-EL2 to EL3)
https://fileserver.linaro.org/s/oR2PtyGKJrqnfRf/preview
Full trace is available here:
https://fileserver.linaro.org/s/WsemLboPEzo24nw/download/aarch64_boot.json.gz
You can download and open it on https://ui.perfetto.dev/ to explore it.
Second, in user mode, tracing qemu-aarch64 (itself) running git
--help:
- Loading program and its interpreter
https://fileserver.linaro.org/s/fie8JgX76yyL5cq/preview
- TB creation
https://fileserver.linaro.org/s/GXY6NKMw5EeRCew/preview
Full trace is available here:
https://fileserver.linaro.org/s/N8X8fnZ5yGRZLsT/download/qemu_aarch64_git_help.json.gz
If you had curiosity and now you're ready to give some attention,
most of the
details you want to read are included in the documentation patch (final one).
Overhead is around x5-x15, and long traces can be directly filtered
with uftrace
if needed.
I hope this plugin can help people trying to understand what happens
out of the
user space, and get a better grasp of how firmwares, bootloader, and kernel
interact behind the curtain.
v6
--
- fix types and spelling mistakes
- use _MAX for privilege level enums
- uftrace info file: use empty strings
- trace flush: assert we wrote all data
- doc: added style for keyboard keys
- added URL to document uftrace entry format
- updated comment in vcpu_tb_trans to explain why we always instrument the first
instruction in tb
v5
--
- addressed Alex comments
- split plugin implementation in several commits
- removed instruction based timestamps (only use time based timestamps)
- removed sampling implementation
v4
--
- add support for x64
v3
--
- fix missing include unistd.h (build failed on MacOS only)
v2
--
- trace active stacks on exit
- do not erase map generated in system_emulation
- add documentation to generate restricted visual traces around specific events
of execution
Pierrick Bouvier (9):
contrib/plugins/uftrace: skeleton file
contrib/plugins/uftrace: define cpu operations and implement aarch64
contrib/plugins/uftrace: track callstack
contrib/plugins/uftrace: implement tracing
contrib/plugins/uftrace: implement privilege level tracing
contrib/plugins/uftrace: generate additional files for uftrace
contrib/plugins/uftrace: implement x64 support
contrib/plugins/uftrace_symbols.py
contrib/plugins/uftrace: add documentation
docs/about/emulation.rst | 199 +++++++
contrib/plugins/uftrace.c | 876 +++++++++++++++++++++++++++++
contrib/plugins/meson.build | 3 +-
contrib/plugins/uftrace_symbols.py | 152 +++++
4 files changed, 1229 insertions(+), 1 deletion(-)
create mode 100644 contrib/plugins/uftrace.c
create mode 100755 contrib/plugins/uftrace_symbols.py
Ping on this series.
Comments from reviewers have been addressed.
Alex, are there other changes you would like before pulling this?
I thought there were the some changes Philippe asked for? I can pull v7
once it is posted. There is no massive rush as the tree has just
re-opened.
It was a polite way to ask if you had other comments to address before
sending a v7, which is mostly about styling issues, so maybe not worth
being sent as it is.
I just posted v7, let me know there if you need other changes before
considering it as ready.
I don't mind if the series is pulled later, but I'm trying to address
all comments as soon as possible and move forward on other topics, thus
my previous ping.
Regards,
Pierrick
