Re: pci.h and pci.c

Ulrich Windl (ulrich.windl@rrzc1.rz.uni-regensburg.de)
Thu, 11 Sep 1997 11:12:08 +0200

> References: <199709031929.MAA21535@icarus.icarus.com> <Pine.LNX.3.96.970903214027.3799A-100000@filesrv1.baby-dragons.com> <5una0q$pkg$1@palladium.transmeta.com>
> On 4 Sep 1997, H. Peter Anvin wrote:
> > Followup to: <Pine.LNX.3.96.970903214027.3799A-100000@filesrv1.baby-dragons.com>
> > By author: "Mr. James W. Laferriere Network Engineer" <babydr@nwrain.net>
> > In newsgroup: linux.dev.kernel
> > > Hello Stephen, If nothing else just for the 'pretty'
> > > messages from the kernel at boot time that actually
> > > tell people what they have on board... ;-)
> > This is pretty broken, really. It would be better to have /proc/pci
> > give the PCI device ID's and then have a user-space utility map those
> > to human-legible strings. This is kernel bloat if anything...
>
> I suggested this a while back but some zealots shot it down claiming they
> needed full text output when booting "recovery disks" from floppy, and
> didn't want external utilities to process /proc/pci output.
>
> I did some checking and found that about 8k (or was it 14k?) could be
> shaved from the kernel by ripping out all the verbose text in pci.c

I also thought about that. The easiest thing would be a configuration
option (like for SCSI messages) to suppress verbose textual output.

Maybe one can make a module and some function pointers: With that
text-module you would get full messages, without just numbers...

I personally also thought of a simple text-compression algorithm to
store all these strings more compactly...

Ulrich
P.S. Just my food for thought