Re: [PATCH] mmc: host: meson-gx-mmc: fix possible deadlock condition for preempt_rt

[Sebastian Andrzej Siewior]

On 2020-09-25 16:14:09 [+0200], Jerome Brunet wrote:
> Looks like we need to do manually what IRQF_ONESHOT was doing for us :(

IRQF_ONESHOT disables the IRQ at the irqchip level. You must ensure that
the device keeps quite. Usually you mast the interrupt source at the
device lee.

> This brings a few questions:
> 
> * The consideration you described is not mentioned near the description
>   of IRQF_ONESHOT. Maybe it should so other drivers with same intent
>   don't end up in the same pitfall ?

>From request_threaded_irq() ->
|  *      If you want to set up a threaded irq handler for your device
|  *      then you need to supply @handler and @thread_fn. @handler is
|  *      still called in hard interrupt context and has to check
|  *      whether the interrupt originates from the device. If yes it
|  *      needs to disable the interrupt on the device and return
|  *      IRQ_WAKE_THREAD which will wake up the handler thread and run
|  *      @thread_fn. This split handler design is necessary to support
|  *      shared interrupts.

Just the line that saying what needs to be done before returning
IRQ_WAKE_THREAD.

> * Why doesn't RT move the IRQ with this flag ? Seems completly unrelated
>   to RT (maybe it is the same documentation problem) 

It is unrelated to RT. Mostly. You end up with the same problem booting
with `threadirqs'. RT has the additional restrictions that you may not
acquire any sleeping locks in hardirq context. This you can see with
addinional lockdep magic.

> * Can't we have flag doing the irq disable in the same way while still
>   allowing to RT to do its magic ? seems better than open coding it in
>   the driver ?

Puh. That should be forwarded the IRQ department.
So we have IRQF_NO_THREAD to avoid force threading. This is documented
as such. Then we have IRQF_TIMER and IRQF_PERCPU which are also not
force threaded and it is not documented as such. However it is used for
the timer-IRQ, IPI, perf and such - things you obviously don't want to
thread and need to run in hard-IRQ context.

What you have ist a primary and secondary and IRQF_ONESHOT and don't
want the primary handler to be force-threaded. I can't answer why we
don't.
However, drivers usually disable the source themself if they providing
both handler.

Sebastian