In message <28270.1097278109@splode.eterna.com.au>matthew green writes

>   [...]    Any objections to splitting sys/dev/pci/if_re.c
>   into two (chipset, and pci-attachment), so I can try kluding up a
>   cardbus attachment?  (has anyone tried that already?)
>
>
>it needs splitting up, just do it.

OK, I have a split driver. The Netgear GA-511 (cardbus rtk8169S)
works, at least in 10/100 mode.  I don't have a PCI 8169 to test
I'm currently using the following file names:

sys/dev/ic/rtl8169plus.c	-- chipset driver
sys/dev/ic/rtl8169plusvar.h	-- functions exported to front-end attachments
sys/dev/pci/if_re_pci.c		-- PCI front-end
sys/dev/cardbus/if_re_cardbus.c	-- Cardbus front-end
sys/mii/rtlgphy.c		-- RTK 8169S/8110S internal 1000baseT phy

There's currently two glitches.:

#1: if the cardbus attachment puts the card to sleep during attach,
the card never gets woken up properly. For now, for debugging, I just
don't put the card to sleep.

#2: with cardbus (at least), the card interrupts in re_attach()
immediatley after writing the CPlus command register. There may we
some way to avoid this by reordering chip-register initialization.
But for now, I modified re_intr() to handle and dismiss interrupts
even if the interface is not IFF_UP: exactly like the rtk(4) backend
driver.

I can email diffs to anyone who wants, but I propose to commit the
split driver as-is. Fixing the above items will probably require some
reverse-engineering based on realtek-supplied Linux/FreeBSD drivers.


PS if anyone has better suggestions for better filenames, this is the
time to make them.  

PPS: anyone know why "rtl..." instead of "rtk..."?