Re: procfs problems

Dan Hollis (
Fri, 18 Apr 1997 17:55:22 -0700 (PDT)

On Fri, 18 Apr 1997, Christopher Horn wrote:
> Dan Hollis wrote:
> > Whether the information is in an external program or the kernel, it is
> > going to take roughly the same amount of space. I for one don't want these
> > tables (which will continue to grow!) statically in the kernel in
> > non-swappable space.
> >
> > There is currently about 5 kilobytes of static text in pci.c, and IMHO it
> > has absolutely no business being there.
> Now that initialization code and data can be jettisoned it might make
> sense to restructure the PCI code so that it just hangs onto the
> information for installed devices.

If you look at the current pci.c, it is going to be exceedingly difficult
to jettison part of the tables, and only keep others.

> There may be other areas of the kernel where some restructuring might
> also help. Question, of course, is would the gain be worth the effort?

I think rewriting current stuff that uses procfs to use the sysctl()
interface would end up being a big win overall.

> For the PCI stuff maybe, as it will keep growing.

I still maintain that the pci parser belongs in user space.

/proc/pci could be reduced to something like:

cat /proc/sys/hardware/pci
0 1 9 1 1 255 0 32 f0000000

Of course this would also be available through sysctl()

This is my ATI mach64 PCI card at irq 255, address f0000000.