On Fri, 6 Jul 2007, M. Warner Losh wrote:

> Personally, I think USBVERBOSE was a bad idea and should be retired.
> The place for large tables of device names is in userland, not stuck
> in wired memory in the kernel.  Without USBVERBOSE, the amount of
> information is adequate to determine a number of things about the
> ugen device that just attached, for example.  Since there are dozens
> of entries in this file already that NetBSD doesn't support, adding
> more only adds size to the USBVERBOSE kernels.
>
> However, I know people like to be able to more completely identify usb
> devices.  I've started working on installing src/sys/dev/usb/usbdevs
> into the same place that FreeBSD installs its pci device database
> file.  I've also started modifications to usbdevs(1) to read this file
> and optionally print out a longer name from this file.
>
> If I were to do that, would it be acceptible to the larger NetBSD
> community for me to commit the merged file and fix the drivers that
> reference the names that will change?  I'm happy to do the leg work
> here, and I do have access to the NetBSD repository to do so.

 	Another idea, which may be more work than needed, but would
 	potentially cover more bases:

 	Have {USB,PCI,MII}VERBOSE add the strings to a separate
 	ELF section of the kernel. If present this section is loaded
 	and referenced for autoconfig, then freed. If dynamic
 	autoconfig occurs after userland has started you could
 	optionally attempt to load it in again, but diminishing
 	returns really kick in then.

 	Whether or nor any of that occurs, I'd be happy to see
 	usbdevs(1) referencing a file with the verbose data, though
 	my opinion may be worth the paper on which its written :)

-- 
 		David/absolute       -- www.NetBSD.org: No hype required --