[linux kernel 5.16, HS38x2] uio driver waiting queue can't wakeup by calling function wake_up_interruptible().

[joshualin-petaio]

Hi expert,

I try to use uio driver to handle memory access and one IRQ signal. I use those functions to reach this goal: open, write, read. After open uo device and before waiting interrupt by read function. I use write function to renew the IRQ. But sometimes I will stuck in read function.
Then I use "cat /proc/interrupts" to check interrupt trigger time. That uio has one trigger. And I observe the function uio_event_notify(), it has execute the function wake_up_interruptible(). But the uio_read() still stuck in while loop.

Do you know why or there has more method can debug this issue ??

Thanks, Joshua

[EvgeniiDidin]

Hi @joshualin-petaio,

If I understood you correctly, uio_read runs in a while loope because this check:
https://elixir.bootlin.com/linux/latest/source/drivers/uio/uio.c#L591
is false.

Could you please check the contents of idev->event and listener->event_count?
Also, please check what is happening here:
https://elixir.bootlin.com/linux/latest/source/drivers/uio/uio.c#L433
and here:
https://elixir.bootlin.com/linux/latest/source/drivers/uio/uio.c#L489
?
Just wondering if open_uio is correctly setting this value and notify updates the value correctly.

[joshualin-petaio]

Hi Evgenii,

Long time no see!!! Thanks for your quick reply.

Yes, you fully understand my question.

As your suggestion, I try to print some message when uio waiting IRQ and got IRQ. It seems idev->event had update correctly(log file line 3). But read function can't wakeup and run the while loop when notify function do wake_up_interruptible(). So, I didn't get the event count check message that in read while loop. Does there has other thing need confirm??

I had attach the uio.c and log file as uio_code_and_log_0.zip
uio_code_and_log_0.zip
. You can check it also.

Thanks, Joshua

[joshualin-petaio]

Hi Evgenii,

When I stuck in schedule() after execute wake_up_interruptible(). I got following message period.

rcu: INFO: rcu_preempt self-detected stall on CPU
rcu: 0-....: (1 GPs behind) idle=2e3/1/0x40000004 softirq=371/372 fqs=955
(t=2100 jiffies g=-11 q=24)
Task dump for CPU 0:
task:ipcTest state:R running task stack: 0 pid: 70 ppid: 1 flags:0x00000008

Stack Trace:
arc_unwind_core+0xe8/0x118
rcu_dump_cpu_stacks+0xc4/0x100
rcu_sched_clock_irq+0x818/0xa90
update_process_times+0x96/0xc0
tick_sched_timer+0x72/0xec
__hrtimer_run_queues+0x132/0x1dc
hrtimer_interrupt+0xde/0x25c
timer_irq_handler+0x28/0x30
__handle_irq_event_percpu+0x4e/0x154
handle_irq_event_percpu+0x14/0x4c
handle_percpu_irq+0x46/0x64
generic_handle_domain_irq+0x74/0xb8
arch_do_IRQ+0x3a/0x58
ret_from_exception+0x0/0x8

Thanks, Joshua

[EvgeniiDidin]

Hi Joshua,

Sorry for delayed reply!

Indeed, for some reason wakeup is not working properly.
It is hard to tell what might be wrong there, add_wait_queue() and schedule() functions are common and well tested, as well as uio driver. Probably, idev->wait value can be checked in both uio_read() and uio_event_notify() function calls.
Also, could you please share .dts you are using if possible?

Thanks,
Evgeniy

[joshualin-petaio]

Hi Evgeniy,

Thanks for your reply.
I'll try to observe the idev->wait. And attachment is my .dts.

Thanks,
Joshua
fpga_hs_idu.txt

[joshualin-petaio]

Hi Evgenii,

I found some phenomena about this issue.
Success case: wakeup thread run on CPU1.
Fail case: wakeup thread run on CPU0. <= will reschedule after wakeup directly in schedule().

Do you have any recommend for this phenomena ?

Thanks, Joshua