Messages in this thread | | | Date | Mon, 22 Aug 2005 16:27:49 -0500 (CDT) | From | Brent Casavant <> | Subject | Re: [PATCH 0/2] external interrupts |
| |
On Sun, 21 Aug 2005, Pavel Machek wrote:
> > Here is a set of patches that implements an external interrupt capability > > in Linux, along with a device driver for a specific hardware device. I > > submitted the patches several weeks ago, and they drew no comments, which > > I take to be a good sign. Anyway, I'm > It was not good sign in this particular case. My reaction was "this is _so_ > overengineered tjat he's probably joking".
Laughter was not wholly unexpected, though I wasn't joking. I'm trying to be realistic about the lifetime of any given hardware, and IOC4 is several years old at this point. Couple that with a sincere desire to preserve application source compatability when (not if) new hardware appears, and an abstraction layer seemed to be a logical choice. I'm more than happy to discuss problems in the abstraction layer's interface and make appropriate changes -- I'm nothing if not obliging.
I do recognize that the code is a bit excessive given that there's only one known device (IOC4) that is currently supported. However, the abstraction isn't really all that complicated. The low-level driver simply provides to the abstraction layer a structure containing a handful of functions pointers: get/set output modes and timings, get/set interrupt sources, get timing roundoff information and device ID. The abstraction layer provides a function to call whenever an interrupt occurs. Then add in a bit of registration/deregistration glue so that kernel modules can load and unload. That's pretty much it -- nothing fancier than needed to decouple hardware implementation details from the programming interface.
Yes, I suppose I could implement completely seperate hardware drivers with a consistent interface, and do away with the abstraction layer. However, that leads to code duplication, which has its own downfalls. Personally, if I had to pick one poison or the other I'd take overengineering over maintaining divergent implementations, but the patch apparently made my predeliction obvious :).
Thanks, Brent Casavant
-- Brent Casavant If you had nothing to fear, bcasavan@sgi.com how then could you be brave? Silicon Graphics, Inc. -- Queen Dama, Source Wars - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |