Re: [RFC PATCH v2 1/5] rtc: migrate driver data to RTC device

["Danilo Krummrich" <dakr@kernel.org>]

On Thu Jan 8, 2026 at 1:24 AM CET, Ke Sun wrote:
> Correction: Should use devm_request_irq(&ldata->rtc->dev, ...) instead
>
> of devm_request_irq(&adev->dev, ...).
>
>
> While a lifecycle chain exists (ldata -> rtc -> amba_device), the IRQ 
> should
>
> be bound to the device that uses it (the RTC device), matching where driver
>
> data is stored, to avoid UAF.

No, that is wrong.

Your RTC device is a class device, not a bus device. I.e. it is not a
representation of physical device on the bus.

However, the requested IRQ is a physical device resource that belongs to a bus
device, which in your case is an AMBA device. (Please also see my reply [1] from
the previous version again.)

Therefore device resources are only valid to access and hold on to for a driver
as long as it is bound to the corresponding bus device.

This is the fundamental reason why it is desirable to guard the lifetime of the
class device and its callbacks with devres, to ensure that it is always valid to
access device resources from class device callbacks.

In C drivers, you just have to do the right thing, otherwise you indeed end up
with UAFs. In Rust, we can leverage the type system to make it impossible for
drivers to mess this up.

[1] https://lore.kernel.org/all/DFHJM2HAG7Q3.1HGZ3P7H55FD2@kernel.org/