pkg/report: new Linux rcu stall reports are mis-parsed

[dvyukov]

All recent rcu stall reports are treated as corrupted. See #4966 for examples. Need to restore stall parsing.

[novitoll]

Hi, I'd like to understand more about this issue as I can't seem to reproduce it.

I've debugged with 718, 719, 720 testdata via report.TestParse test and all of them have Corrupted true field, so I'm confused here. If report says CORRUPTED: Y, should I expect report.Corrupted to be true?

718, 720 might be parsed via a new rule exc_nmi per NMI call trace. 719 matches current ?:apic_timer_interrupt rule in IRQ context.

[dvyukov]

If report says CORRUPTED: Y, should I expect report.Corrupted to be true?

Correct. Such reports will be thrown away and not reported as bugs. But these reports are not corrupted, and should not be marked as corrupted.

718, 720 might be parsed via a new rule exc_nmi per NMI call trace. 719 matches current ?:apic_timer_interrupt rule in IRQ context.

Yes, but as far as I remember lines that contain " ? " are removed before matching (stack unwinder says it's some garbage, not a real frame).

I am actually not sure what's the current state here b/c I see some recent "rcu stall" reports that are parsed correctly, e.g.: https://syzkaller.appspot.com/bug?extid=894cac40ce97c83618bb

But I am not sure if we are also throwing away more rcu stall reports that were parser as corrupted. This bucket contains some recent reports that were parsed as corrutped: https://syzkaller.appspot.com/bug?extid=56275b6e83c2fb060948 I see some "rcu stalls" there, but they just look truncated rather than mis-parsed.