Re: [PATCH] genirq: Deduplicate WARN_ON_ONCE() in generic_handle_domain_irq()

[Sebastian Andrzej Siewior <bigeasy@xxxxxxxxxxxxx>]

On 2022-05-10 09:56:05 [+0200], Lukas Wunner wrote:
> An example for irqchips where the warning is false positive are
> USB-attached GPIO controllers such as drivers/gpio/gpio-dln2.c:

They are not false positives because…

> USB gadgets are incapable of directly signaling an interrupt because
> they cannot initiate a bus transaction by themselves.  All communication
> on the bus is initiated by the host controller, which polls a gadget's
> Interrupt Endpoint in regular intervals.  If an interrupt is pending,
> that information is passed up the stack in softirq context, from which
> a hardirq is synthesized via generic_handle_domain_irq().

they tell you that the context is wrong. From looking at gpio-dln2
this is called from USB URB's callback which is softirq. In the end
dln2_gpio_event() is invoked while dln2_dev::event_cb_lock is acquired.
That lock is acquired by disabling interrupts which is what gets the
locking right for generic_handle_domain_irq(). If that lock lifted to
spin_lock_bh() (because it is always in urb's calback context and all
HCDs complete in one context unlike now) then this breaks. And
PREEMPT_RT is broken already. Therefore, last week, I've been promoting
generic_handle_domain_irq_safe()
   https://lkml.kernel.org/r/YnkfWFzvusFFktSt@xxxxxxxxxxxxx

and sadly I missed dln2. Please let me know if you have more users
similar to dln2. I will add those to my list once upstream buys that
interface.

Sebastian