>I've cleaned up after this change (which involved bumping the kernel
>version, updating libpci and bumping its major, and updating assorted
>documentation).

Cool - so it compiles now?  My sparc build was dying on:

pci_subr.c(656): warning: c89 C does not support 'long long' [265]

>+               cp += snprintf(cp, ep - cp, "%svendor 0x%04x product 0x%04x",

As you say, this is dangerous and all too common.  

It is an appealing approach though, so its probably worth doing a 
safe_snprinf() that will return 0 in the not enough space 
case to avoid the overflow issues.

If the above is in the kernel and DIAGNOSTIC is defined, a panic might
be in order for the out of room case?

--sjg