Re: [PATCH 6/9] lockdep: Introduce wait-type checks

[bigeasy@xxxxxxxxxxxxx]

On 2025-11-17 14:59:28 [+0000], Sverdlin, Alexander wrote:
> Hi Paul,
Hi Alexander,

…
> > > > > Call trace:
…
> > > > >   counter_push_event+0x68/0x430 [counter]
> > > > >   interrupt_cnt_isr+0x40/0x78 [interrupt_cnt]
> > > > >   __handle_irq_event_percpu+0xa4/0x398
…
> > Does that help?
> 
> Mmm. Not yet... Does this mean we are not allowed to use "spinlock_t" in
> IRQs any longer? Where has this been announced? This probably means that
> the majority of IRQ handlers were already converted tree-wide?
> 
> This also would mean that spin_lock_irqsave() shall not exist at all
> so that we are forced to use raw_spin_lock_irqsave()?

As I tried to explain, if you use IRQF_NO_THREAD then the handler will
not be threaded. Looking at drivers/counter/interrupt-cnt.c, it does

|         irq_set_status_flags(priv->irq, IRQ_NOAUTOEN);
|         ret = devm_request_irq(dev, priv->irq, interrupt_cnt_isr,
|                                IRQF_TRIGGER_RISING | IRQF_NO_THREAD,
|                                dev_name(dev), counter);
|         if (ret)

So the irq-core does not thread the handler as requested. So it must not
use spinlock_t in such a case.
And judging from the code, that IRQF_NO_THREAD should be removed. Not
sure why you are the only seeing that splat.

Sebastian