Interrupts & total mess

Interrupts & total mess

[Benjamin Herrenschmidt]

Ok so my simple project of adding NO_IRQ definitions all over the place
is turning into a nightmare for various reasons (the probe_irq_* stuff
beeing one of them, as it currently prevents using -1, so I'm leaning
toward defining NO_IRQ as beeing INT_MIN, nothing against that ?)

However, while trying to do that in a simple way, that is with a
 #ifndef NO_IRQ
 #define NO_IRQ		(INT_MIN)
 #endif

Somewhere in some generic piece of include after we has some asm/* stuff
included to let the arch a chance to override it, I figured that, first,
there are a number of places where "irq" is defined as beeing unsigned
long... So neither INT_MIN nor -1 are appropriate. Then I noticed while
looking for the right files to add this stuff that we have, at least:

include/linux/interrupts.h
include/linux/irq.h
include/linux/hardirq.h
include/asm-*/irq.h
include/asm-*/hw_irq.h
include/asm-*/hardirq.h

Which is seriously starting to make no sense, especially when you don't
really know who is including who, with also the strange rule of never
including linux/irq.h directly from a driver since the arch may not use
the definitions in there, etc... it's a complete can of worms...

So basically, linux/irq.h should be asm-generic/irq.h right ?

Then, all of the CONFIG_HARDIRQS_GENERIC stuff in linux/interrupts.h
should be moved there as well, since that's pretty much what the things
in linux/irq.h already define.

So our path should be:

toplevel include
	linux/interrupts.h (or rename it to linux/irq.h)
	  asm/irq.h (the arch implementation)
	    [asm-generic/irq.h] (optionally using the common defs)

but I'm not sure what to do with the various hardirq.h & hw_irq.h
thingies, at least one of the 2 arch ones should die.

I'm ready to start the (painful) work of cleaning that up, though
that will probably end up in a giga-patch touching hundreds of files
(just to change a #include directive most of the time) though I won't
fix all archs, I prefer not mucking aroudn with things I don't
understand.

But I'm not sure what we want to do here, so let's discuss it a bit.

Ben.

Re: Interrupts & total mess

[Ben Dooks]

On Thu, Oct 21, 2004 at 04:07:58PM +1000, Benjamin Herrenschmidt wrote:
> Ok so my simple project of adding NO_IRQ definitions all over the place
> is turning into a nightmare for various reasons (the probe_irq_* stuff
> beeing one of them, as it currently prevents using -1, so I'm leaning
> toward defining NO_IRQ as beeing INT_MIN, nothing against that ?)
> 
> However, while trying to do that in a simple way, that is with a
>  #ifndef NO_IRQ
>  #define NO_IRQ		(INT_MIN)
>  #endif
> 
> Somewhere in some generic piece of include after we has some asm/* stuff
> included to let the arch a chance to override it, I figured that, first,
> there are a number of places where "irq" is defined as beeing unsigned
> long... So neither INT_MIN nor -1 are appropriate. Then I noticed while
> looking for the right files to add this stuff that we have, at least:
> 
> include/linux/interrupts.h
> include/linux/irq.h
> include/linux/hardirq.h
> include/asm-*/irq.h
> include/asm-*/hw_irq.h
> include/asm-*/hardirq.h

also see the drivers/base/platform.c for the definition of
platform_get_irq() which also should be considered for this
(see earlier posting about the return code).

-- 
Ben (ben@fluff.org, http://www.fluff.org/)

  'a smiley only costs 4 bytes'