Hi Finn,
Am 08.03.2024 um 13:56 schrieb Finn Thain:
I realized (belatedly) that scheduler_tick() does not run the task
itself but just causes a reschedule if appropriate, so the probability
for this condition is quite low. The question is - does the VIA1 timer
interrupt ever get reentered?
Can you add a printk_once() warning when you see
arch/m68k/mac/via.c:via_timer_handler() reentered, Guenter?
It's not possible for an interrupt handler to be interrupted unless by an
hard interrupt at a higher priority level. Maybe QEMU could be so broken
Right - not sure why I remembered interrupt handlers lowering the IPL
there.
That leaves other users of the run queue lock. Queuing new tasks would
be one obvious such user for the init task, but that is run with IRQs
disabled (we'd see these warnings a lot more else). And comparing with
how early I see the 'cblist_init_generic: Setting adjustable number of
callback queues' message in my boot logs, this may all happen too early
for init to start queuing tasks.
(?!) but recall that I saw this BUG on a Powerbook 180 ten years ago.
I'll take a look at the scheduler code from that era (though probably
not much has changed since).
IMHO, Gueunter would do better to instrument the spinlock tests on the
assumption that the locking in arch/m68k really is correct.
I've come to agree - maybe logging any run queue locks taken by the init
task with IRQs enabled might help?
Cheers,
Michael
