On Sat, Dec 3, 2011 at 00:22, Alan Cox [off-list ref] wrote:

quoted

This is now broken on ARM where, for good or bad, NO_IRQ currently is
used and is -1.

Good.

ARM developers have been told to change this for several years. The nice
approach hasn't worked, the patient approach hasn't worked so now finally
ARM is going to be dragged kicking and screaming into doing the work
everyone else did several years ago.

I have so little sympathy over this that you'll need a quantum physicist
to measure it.

quoted

Half-removing NO_IRQ is going to be problematic, though...
I really don't care whether the "no irq" value is 0 or -1, but it is
abundantly clear that choosing different values to mean the same thing
on opposite sides of an interface does not work.

You've had years to fix it. If I were you I'd delete NO_IRQ from your
tree, type make and get it done. It's not even a big job to clean it out.

At that point various other drivers will also start working properly on
ARM because they use 0 for polled mode.

Not just ARM:

arch/arm/include/asm/irq.h:#ifndef
NO_IRQarch/arm/include/asm/irq.h:#define NO_IRQ       ((unsigned
int)(-1))arch/microblaze/include/asm/irq.h:#define NO_IRQ
(-1)arch/mn10300/include/asm/irq.h:#define NO_IRQ
INT_MAXarch/openrisc/include/asm/irq.h:#define NO_IRQ
(-1)arch/parisc/include/asm/irq.h:#define NO_IRQ
(-1)arch/powerpc/include/asm/irq.h:#define NO_IRQ
(0)arch/powerpc/include/asm/machdep.h:     /* Return an irq, or NO_IRQ
to indicate arch/powerpc/include/asm/parport.h:             if (virq
== NO_IRQ)arch/powerpc/include/asm/qe_ic.h:       if (cascade_irq !=
NO_IRQ)arch/powerpc/include/asm/qe_ic.h:       if (cascade_irq !=
NO_IRQ)arch/powerpc/include/asm/qe_ic.h:       if (cascade_irq !=
NO_IRQ)arch/powerpc/include/asm/qe_ic.h:       if (cascade_irq !=
NO_IRQ)arch/powerpc/include/asm/qe_ic.h:       if (cascade_irq ==
NO_IRQ)arch/powerpc/include/asm/qe_ic.h:       if (cascade_irq !=
NO_IRQ)arch/sparc/include/asm/irq_32.h:#define NO_IRQ
0xffffffffarch/sparc/include/asm/irq_64.h:#define NO_IRQ
0xffffffff

And it's not just definitions of NO_IRQ. These are easy to find.
On some archs (notably ARM) zero still seems to be a valid IRQ number,
e.g. IRQ_LOCOMO_KEY and IRQ_DMA0C0.

Also, UML has TIMER_IRQ being zero.

A quick grep found many more IRQ definitions being zero, but
surprisingly the few
I looked into were definitions without users (e.g. SE7343_FPGA_IRQ_MRSHPC0,
ROUTE_VIA_IRQ0 aka IRQ_MB93493_VDC_ROUTE).

Perhaps request_irq() should just reject zero to find all of them?

Gr{oetje,eeting}s,

                        Geert

--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@linux-m68k.org

In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
                                -- Linus Torvalds