Re: Get rid of IRQF_DISABLED - (was [PATCH] genirq: warn about IRQF_SHARED|IRQF_DISABLED)

[David Brownell <david-b@pacbell.net>]

On Monday 30 November 2009, Russell King - ARM Linux wrote:
> On Mon, Nov 30, 2009 at 02:47:02PM +0000, Alan Cox wrote:
> > SHARED|DISABLED ought to WARN_ON() and if that doesn't motivate people
> > then return -EINVAL.
> 
> That is an impossibility.  There is hardware out there (AT91) where
> the timer interrupt is shared with other peripherals, and you end
> up with a mixture of irqs-disabled and irqs-enabled handlers sharing
> the same interrupt.

For the record:  AT91 isn't restricted to the system timers hooked
up on irq 0 ... there's also drivers/clocksource/tcb_clksrc.c (not
at the same hardware priority).

But to concur, this is indeed messy.  Both the system timer and
the serial console generally share the same IRQ; both are very
timing-sensitive.  I've seen console character dropouts after
tweaking timer IRQ handling.  And I've never convinced myself
that Linux handles the hardware IRQ priority on those chips as
well as it could.  


> My point is that if we outlaw irqs-disabled shared interrupts, it puts
> Atmel AT91 support into immediate difficulties.

ISTR that those TCB modules don't share IRQs with other peripherals.

Also, that Linux doesn't use them for much else.  I've yet to see a
three-phase motor driver using the TCB's PWM capabilities, for example.

- Dave