On Tue, Dec 17, 2024 at 01:45:08PM +0300, Dan Carpenter wrote:
> On Tue, Dec 17, 2024 at 05:12:08PM +0800, Xu Yang wrote:
> > With edge irq support, the ALERT event may be missed currently. The reason
> > is that ALERT_MASK register is written before devm_request_threaded_irq().
> > If ALERT event happens in this time gap, it will be missed and ALERT line
> > will not recover to high level. However, we don't meet this issue with
> > level irq. To avoid the issue, this will set ALERT_MASK register after
> > devm_request_threaded_irq() return.
> >
> > Fixes: 77e85107a771 ("usb: typec: tcpci: support edge irq")
>
> I agree with Francesco that this was introduced by patch 1 so the
> patches need to be folded together into one patch. The commit
> message would be something like:
>
> The ordering of setting up the IRQs is not correct here.
> We need to call tcpci_register_port() devm_request_threaded_irq().
> Otherwise if we recieve an IRQ before tcpci_register_port() has
> completed it leads to a NULL dereference in tcpci_irq() because
> tcpci->regmap and other pointers are NULL.
>
> However, moving tcpci_register_port() earlier creates a problem
> of its own because there is a potential that tcpci_init() will be
> called before devm_request_threaded_irq(). The tcpci_init()
> writes the ALERT_MASK to the hardware to tell it to start
> generating interrupts but we're not ready to deal with them yet.
>
> Move the ALERT_MASK stuff until after the call to
> devm_request_threaded_irq() has finished.
>
> Something like that.
>
Okay.
Thanks,
Xu Yang
