Re: [PATCH v2 3/3] media: qcom: iris: Fix IRQ disable depth imbalance on watchdog interrupt

[Bryan O'Donoghue]

On 11/11/2026 09:05, Vishnu Reddy wrote:
> iris_hfi_isr() disables the IRQ for every interrupt and the threaded
> handler enables it again

This reads like a statement of a bug to me.

> except for a watchdog interrupt, which is left
> masked until the firmware is reset. 

Which then gets worse.

iris_vpu_power_off() decides whether
> to add its own disable from core->intr_status, but that is only updated
> by the threaded handler. If a watchdog interrupt is still pending when
> power-off checks it, power-off adds a disable and the handler then skips
> its enable, leaving the depth one higher than the single enable_irq() in
> iris_vpu_power_on() can undo. The IRQ then stays disabled after the
> reset and the firmware can no longer be reached.

Why don't we just do the _simple_ thing and enable the watchdog @ 
power-on and disable it @ power-off ? Let's at least investigate that.

If its infeasible or too reductive consider: passing IRQF_ONESHOT to 
devm_request_threaded_irq() then the kernel will automagically mask the 
ISR for you and will re-enable once your threaded handler has run.

We can then remove the watchdog interrupt as an exceptional (no pun) 
case from the IRQ.

Either way this flag mechanism smacks of spaghetti code and needs a rethink.

---
bod