lib: start polling for fw boot complete only when sof card is not pre…

[fredoh9]

…sent

SOF soundcard is already listed it is ready to start the test. If not, set the KERNEL_CHECKPOINT and start polling for FW boot complete. This is a bug fix for fceeee0.

[fredoh9]

without --since=@"$KERNEL_CHECKPOINT" is the biggest the problem from fceeee0

[marc-hb]

Summary of an off-line review with @fredoh9 :

• There's a tiny TOCTOU race window between is_sof_used() and setup_kernel_check_point. But @fredoh9 found that there's already a -1 in setup_kernel_check_point, so we're good. Just need a comment.
• By fixing a bug, this PR makes it impossible to design any test where the modules are not already loaded. Fred will add an escape. there is already one.

[fredoh9]

There is no TOCTOU race condition problem, because setup_kernel_check_point() has -3 or -1 second

[marc-hb]

There is no TOCTOU race condition problem, because setup_kernel_check_point() has -3 or -1 second

That's a bit of hack :-) It's never going to happen but in computer science theory you could have things like:

• the lib.sh process being suspended by the OS for more than 1 second, or
• the firmware being UNloaded just after the -3 seconds and before the is_sof_used() grep.

Current code is good enough in practice but please add a comment, something like "TOCTOU race avoided thanks to -1 second inside setup_kernel_check_point()

[fredoh9]

Note, NO_POLL_FW_LOADING variable is already available to skip this polling

[fredoh9]

Looks good for device test results also. Merging now

[marc-hb]

Here's an interesting side-effect.

thesofproject/sof#7698 broke the MTL build. Tests were started anyway without firmware (tracked in internal issue 139). Each test took 70s to timeout in https://sof-ci.01.org/sofpr/PR7698/build9948/devicetest

The whole PR test plan uselessly hogged 3 MTL devices for more than 200 minutes.