[FEATURE] mechanism to ensure the consistent locking order

[yamt]

Is your feature request related to a problem? Please describe.

it would be nice to have some debug feature which can detect issues like #15314

Describe the solution you'd like

precedents:
https://man.freebsd.org/cgi/man.cgi?witness(4)
https://docs.kernel.org/locking/lockdep-design.html

Describe alternatives you've considered

although adhoc local assertions like yamt@ac3dda4
can work, it would be nicer to have a generic mechanism in the upstream.

Verification

• I have verified before submitting the report.

[xiaoxiang781216]

@yamt can you try https://github.com/apache/nuttx/blob/master/sched/misc/deadlock.c?

[yamt]

@yamt can you try https://github.com/apache/nuttx/blob/master/sched/misc/deadlock.c?

glancing at the code, it requires you to actually make a deadlock happen, and then somehow trigger an assertion failure.
is it right?

reproducing a deadlock is not always easy.
well, if you know how to reproduce a deadlock, i guess you likely already know the problem well. just fix it. :-)

what i want to see is something more aggressive.
eg. detect locking attempts with a wrong order. (in most cases it might not actually cause a deadlock. but it can cause a deadlock.)

[xiaoxiang781216]

@yamt can you try https://github.com/apache/nuttx/blob/master/sched/misc/deadlock.c?

glancing at the code, it requires you to actually make a deadlock happen, and then somehow trigger an assertion failure. is it right?

Yes, the code assume that the deadlock will trigger panic finally. but it's easy to expose a nsh command to trigger the check periodly.

reproducing a deadlock is not always easy. well, if you know how to reproduce a deadlock, i guess you likely already know the problem well. just fix it. :-)

in a complex system, it's hard to know which mutexs and threads participant the deadlock chain, this tool help people identify the problem mutex and thread after the deadlock happen.

what i want to see is something more aggressive. eg. detect locking attempts with a wrong order. (in most cases it might not actually cause a deadlock. but it can cause a deadlock.)

this feature requires to record the locking order.

[yamt]

what i want to see is something more aggressive. eg. detect locking attempts with a wrong order. (in most cases it might not actually cause a deadlock. but it can cause a deadlock.)

this feature requires to record the locking order.

yes.