On Tue, Jan 13, 2004 at 06:09:26PM -0800, Andrew Morton wrote:
> Why is kyro_modedb[] and the code which touches it under #ifdef MODULE?
> 
No particular reason, really. I hadn't been following some of the recent fbdev
changes, so I wasn't sure if this was necessary or not (I was using tdfxfb as
a reference when tracking API changes, and noticed that it did the #ifdef
MODULE stuff which it never did earlier). I'll happily drop the #ifdef MODULE
bits if this isn't necessary. James?

> kyrofb_pci_tbl[] cannot be marked __devinitdata - it is linked into
> system-wide PCI tables and will be walked at various times in the kernel. 
> If it is in a discarded section the kernel oopses.
> 
Okay, I'll remove that.

> Shouldn't the module_exit(kyrofb_exit); be inside #ifdef MODULE too?
> 
Yes, I suppose.

> 
> In general, what are the fbdev drivers doing here btw?  It's very strange
> that some core file (fbmem.c) has to "know" about individual card drivers. 
> It should be sufficient for the driver to just register itself with the core
> at the drivers's module_init() time.
> 
This is done for device ordering as far as I recall. Devices listed first on
the command line are numbered first, with the rest coming up in the order that
they're listed in fb_drivers[].

I had some patches against this awhile ago that redid some of this, getting
rid of fb_drivers[], putting ordered devices in a list and logically assigning
IDs based off of that, etc. but still letting the drivers come up on their own
through module_init(). I was planning on working on this for 2.7, as I didn't
have enough time under 2.5.

> Sigh.  Too late to fix that now I guess.

I'll send you another patch for this as soon as I hear back from James or
someone else about the modedb / #ifdef MODULE stuff, I don't see any reason
why it would be required, though.