Re: hardware detection using libpci
Joey Hess wrote:
> Glenn McGrath wrote:
> > I guess the card names are handy for the user if all doesnt go to plan,
> > i just left them out because the program didnt _need_ them, and they
> > take the most space.
> That's true, it is nice to be able to look at /proc/pci and figure out
> what module to load by hand. But that 80k is pretty nice too.
If we have the pci id's we can always get the hardware description later
from a seperate list, but then i guess the information is of less use.
> > From the three modules i looked at there appears to be a few pci ids
> > that arent in libdetects list, so maybe a kernel module based list would
> > be more complete.
> Yes, if at all possible we really need to grub this list of of the
> kernel so you don't have to keep maintaining it.
I started to setup a sourceforge page for pcidetect, but now am having
second thoughts, im not sure that pcidetect is the best aproach purely
because it is only pci.
> > I noticed that libdetect uses /proc to get the pci ids, it cant do
> > hardware probes to get the ids. Whilst Joey has said that we wouldnt
> > ever ship a kernel without /proc support, i think that having the option
> > to do without /proc is a good thing, if we want to make it easy for
> > users to build a disk with there own kernel then its best is we depend
> > on as few kernel features as possible.
> I think we're surely going to depend on some features, probably even
> features that a user wouldn't think to compile into their kernel. I'm
> afraid doing without /proc is going to waste more space than it saves by
> making code try to do without it. Thus, asking users to compile in
> /proc, which _is_ the default, is not undue hardship.
> Remember making a kernel for the boot-floppies requires it be built with
> all kinds of unlikely stuff like msdos fs support. That is well
> documented though, and I guess some people have managed to do it.
Yea, i now conceed that the linux kernel demands /proc support, there is
no reasonable way of determining which devices the kernel has detected
Im starting to feel that the linux kernel is inflexible and limiting, we
could do better hardware detection but it would need to be a kernel
module, this would be a bad idea as i think we want as much kernel
independence as possible.