Issue with Master_vehicle crash

[Tony3dr]

Today we tested Master v1.9.0 (beta) (7f3b170) with the Pixracer, Pixhawk 4 mini, Pixhawk 4. Created mission. The vehicle armed good, took off. At 30 meters the vehicle started spinning and came down to the ground. Resulted in a crash. Same behavior for all 3 platforms.

Steps to reproduce the behavior:

1. Uploaded mission
2. Drone switched on
3. Vehicle armed
4. Took off
5. Elevated to 30 meters
6. vehicle started spinning out of control
7. The vehicle came down and crashed

Pixracer:
https://review.px4.io/plot_app?log=4f0c2692-5f27-4d83-bfbd-765bf48a989e

Pixhawk 4 Mini:
https://review.px4.io/plot_app?log=53fa724b-906d-4e93-87fe-5a4f4c2a50b8

Pixhawk 4:
https://review.px4.io/plot_app?log=69f42b00-6259-4a97-96c1-bd8f06b4d29d

@dagar

[mhkabir]

Potentially #11376

[dagar]

Wow, ouch. Looking.

[dagar]

In both cases (fmu-v4 and fmu-v5) the pc points to math::max (mathlib/math/Limits.hpp:72) called from navigator MissionBlock::get_time_inside().

https://github.com/PX4/Firmware/blob/c3d6a042e6332792fe85a1121c08fefa7dbfcb43/src/modules/navigator/mission_block.cpp#L480-L491

I think the problem is passing the reference of a packed struct (mission_item) member to the template function. Taking a closer look...

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=77388

[dagar]

This is easy enough to fix once off, but I'm looking for a most exhaustive solution (compiler warning perhaps) otherwise we'll need to be careful around anything packed (mission items, mavlink in general).

https://stackoverflow.com/questions/8568432/is-gccs-attribute-packed-pragma-pack-unsafe/8568441#8568441

[dagar]

Newer gcc has -Waddress-of-packed-member.

[dagar]

WIP fixes in #11411

[dagar]

We need to do a quick pass of packed structure usage looking for anything that might take the address of a member, create a reference, or pass it into a function that does.

This includes navigator (mission_item), mavlink (anything from the mavlink spec), and a few drivers.