Skip to content

Conversation

@a-nogikh
Copy link
Collaborator

@a-nogikh a-nogikh commented Aug 5, 2025

We used to do it after query completion, but that leaves some window open for a completely broken executor fork to mess with the data. At the same time, executor expects the area to be completely clean when it starts.

@a-nogikh a-nogikh requested a review from dvyukov August 5, 2025 13:10
We used to do it after query completion, but that leaves some window
open for a completely broken executor fork to mess with the data. At the
same time, executor expects the area to be completely clean when it starts.
@a-nogikh a-nogikh force-pushed the features/move-cleanup branch from 89eb419 to 3b07dad Compare August 5, 2025 13:22
@a-nogikh
Copy link
Collaborator Author

a-nogikh commented Aug 5, 2025

run_test.go:112: expect feature DevlinkPCI to be false, but it is true

So far I couldn't reproduce it locally. Devlink should indeed be unavailable on a test os:

#if SYZ_EXECUTOR || SYZ_DEVLINK_PCI
#if SYZ_EXECUTOR
if (flag_devlink_pci)
#endif
exitf("devlink_pci is not supported");
#endif

UPD: it reproduces, but very rarely.

@a-nogikh
Copy link
Collaborator Author

a-nogikh commented Aug 5, 2025

If we do not clear output data at the end of HandleCompletion (like in this PR), and we do Restart() because of the env change (like it's done during machine check), and if that check fails (like it should, for devlink + test os), then there's a chance we will report a stale result together with the failure.

It happens rarely because, for that, the proc that gets the devlink check request must happen to have already executed something.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants