Kernel Bug? Use of IRQF_SHARED + IRQF_DISABLED

["Rajat Jain" <rajat.noida.india@gmail.com>]

Hi,

Based on suggestion from Thomas Petazzoni, I'm moving this to LKML.

This is regarding the following code in kernel/irq/handle.c. Consider
the case of a shared IRQ line, where two handlers are registered such
that first handler does not specify IRQF_DISABLED, but the second one
does. But it seems both the handlers will be called with IRQs ENABLED
(which is certainly not what the second handler expects).

I also checked but could not find anything that stops me from
registering two shared ISRs - one with IRQF_DISABLED & another without
this flag. Am I missing something here?

irqreturn_t handle_IRQ_event(unsigned int irq, struct irqaction *action)
{
   irqreturn_t ret, retval = IRQ_NONE;
   unsigned int status = 0;

   handle_dynamic_tick(action);

   if (!(action->flags & IRQF_DISABLED))
       local_irq_enable_in_hardirq();

   do {
       ret = action->handler(irq, action->dev_id);
       if (ret == IRQ_HANDLED)
           status |= action->flags;
       retval |= ret;
       action = action->next;
   } while (action);

   if (status & IRQF_SAMPLE_RANDOM)
       add_interrupt_randomness(irq);
   local_irq_disable();

   return retval;
}

I'd like to submit a patch but was wondering which of the following is
the correct startegy to deal with above situation (I personally think
(1) below is more appropriate):

1) IN the above code while calling shared ISRs, check for each ISR
whether it specified IRQF_DISABLED or not. Enable IRQs only for ISR
that did not specify IRQF_DISABLED.

2) While installing ISR, check that all the ISRs for that IRQ should
have consistent use of IRQF_DISABLED. Don't allow insonsistent use of
IRQF_DISABLED on a shared IRQ.

Thanks,

Rajat