Re: [syzbot] [kernel?] WARNING: proc registration bug in unregister_irq_proc (2)

[Thomas Gleixner]

On Mon, Jul 27 2026 at 16:33, Greg Kroah-Hartman wrote:
> On Mon, Jul 27, 2026 at 04:22:58PM +0200, Thomas Gleixner wrote:
>> I rather go and harden the interrupt core against such bogosities.
>
> Sure, but really, anyone doing a manual "unbind" gets to keep the pieces
> if the kernel can't handle it. The comedi driver really aren't all that
> ready for stuff like this as they are in systems that are fixed to the
> shop floor, driving big stone cutters.
>
> So no real rush/need that I can see here.

I understand, but it bothers me that the leakage is detected way too
late and only unearthed by removing a proc entry. There might be actual
references dangling. The interrupt core code knows all that at the point
where the teardown starts and that's where it can cleanup the mess and
emit a warning message.

Thanks,

        tglx