[PATCH] Implement Symbol-Based Video Driver Autoloading

David Nusinow dnusinow at speakeasy.net
Fri Nov 16 18:28:12 PST 2007


On Fri, Nov 16, 2007 at 04:45:07AM +0100, Stephane Marchesin wrote:
>    On 11/16/07, David Nusinow <[1]dnusinow at speakeasy.net> wrote:
> 
>      Ok, this method is the other one I had in mind. Basically the server
>      would
>      load each driver, call its probe function, then unload it. Once it's
>      collected a list of drivers that say they have something they can
>      support, load them. Is this more or less what you're thinking?
> 
>    Well to be honest we're mostly thinking about lowering our own maintenance
>    burden here. I also don't know how other driver devs feel, some might need
>    pciids for different reasons anyway.
> 
> 
>      This seems like it'd be fairly heavyweight to me. I don't want to slow
>      down
>      server startup too much if I can help it. Do you think that having a
>      dozen
>      drivers run this kind of probe won't slow down the startup?
> 
>    I frankly don't know, as far as I know we're the first driver to do that
>    (all other drivers have at least a pciid list + sometimes future device
>    detection)

Ok, I'm all for reducing the workload on driver maintainers. I've spent
more than enough time maintaining pci id lists to empathize. I'll try and
answer my own questions about the performance by implementing this and
benchmarking it and seeing how it goes. I think with a simple cache setup
it'll probably be pretty quick most of the time. I'd love to hear from
other driver maintainers if they'd prefer this sort of mechanism over
straight pci id's too.

 - David Nusinow



More information about the xorg mailing list