A better way to sequence driver initialization?

[Bill Gatliff]

Paul Mundt wrote:
> In some cases that might be valid, but there are many cases where drivers
> can reconfigure their capability sets based on which GPIOs are and aren't
> available. Just because a pin isn't available doesn't make it a
> show-stopper for the probe path..
>   

Understood.

I just tweaked my kernel to run all probe()s  in their own kthreads. 
The results were a mixed bag, as expected.

On the one hand, it's pretty cool to see everything running in parallel!

On the other hand, I can see now yet another big reason why a probe
free-for-all won't work.  Busses are also devices, so there are plenty
of places where a device for a particular bus type won't be able to
probe because the target bus simply doesn't exist yet.  That's yet
another dependency that I hadn't thought about.

Were I to rewrite Linux :), I would make probing parallel from the
beginning.  But I think I'm going to have to settle for now with
kthreading my probes that might want to sleep, and adding a wait queue
to gpio_request() and a few others.  It ain't perfect, but it is
achieveable.  *sigh*


b.g.

-- 
Bill Gatliff
bgat at billgatliff.com