RDP(4) BSD/i386 Kernel Interfaces Manual RDP(4)NAMErdp — Ethernet driver for RealTek RTL 8002 pocket ethernet
SYNOPSIS
device rdp0 at isa? port 0x378 irq 7
device rdp0 at isa? port 0x378 irq 7 flags 0x2
DESCRIPTION
The rdp device driver supports RealTek RTL 8002-based pocket ethernet
adapters, connected to a standard parallel port.
These adapters seem to belong to the cheaper choices among pocket ether‐
net adapters. The RTL 8002 is the central part, containing an interface
to BNC and UTP (10 Mbit/s) media, as well as a host interface that is
designed to talk to standard parallel printer adapters. For the full
ethernet adapter to work, it is completed by an external RAM used as the
Tx and Rx packet buffer (16 K x 4 for the RTL 8002), and an EEPROM to
hold the assigned ethernet hardware address. For the RTL 8002, the EEP‐
ROM can be either a standard 93C46 serial EEPROM (which seems to be a
common choice), or a 74S288 parallel one. The latter variant needs the
device configuration flag 0x1 in order to work.
Since standard printer adapters seem to vary wildly among their timing
requirements, there are currently two possible choices for the way data
are being exchanged between the pocket ethernet adapter and the printer
interface. The default is the fastest mode the RTL 8002 supports. If
the printer adapter to use is particularly slow (which can be noticed by
watching the ethernet wire for crippled packets, or by not seeing cor‐
rectly received packets), the configuration flag 0x2 can be set in order
to throttle down the rdp driver. Note that in fast mode, the data rate
is asymmetric, sending is a little faster (up to two times) than receiv‐
ing. Rates like 150 KB/s for sending and 80 KB/s for receiving are com‐
mon. For slow mode, both rates are about the same, and in the range of
50 KB/s through 70 KB/s. As always, your mileage may vary.
In case the adapter isn't recognized at boot-time, setting the
bootverbose flag (‘-v’) might help in diagnosing the reason. Since the
RTL 8002 requires the availability of a working interrupt for the printer
adapter (unlike the ppc(4) driver), the rdp driver fails to attach if the
ethernet adapter cannot assert an interrupt at probe time.
The RTL 8002 doesn't support (hardware) multicast.
The rdp driver internally sets a flag so it gets probed very early. This
way, it is possible to configure both, an rdp driver as well as a ppc(4)
driver into the same kernel. If no RTL 8002 hardware is present, probing
will eventually detect the printer driver.
DIAGNOSTICS
rdp0: configured IRQ (7) cannot be asserted by device
The probe routine was unable to get the RTL 8002 asserting an interrupt
request through the printer adapter.
rdp0: failed to find a valid hardware address in EEPROM
Since there doesn't seem to be a standard place for storing the hardware
ethernet address within the EEPROM, the rdp driver walks the entire
(serial) EEPROM contents until it finds something that looks like a valid
ethernet hardware address, based on the IEEE's OUI assignments. This
diagnostic tells the driver was unable to find one. Note: it might as
well be the current adapter is one of the rare examples with a 74S288
EEPROM, so ‘flags 0x1’ should be tried.
rdp0: Device timeout
After initiating a packet transmission, the ethernet adapter didn't
return a notification of the (successful or failed) transmission. The
hardware is likely to be wedged, and is being reset.
SEE ALSOng_ether(4), ppc(4), ifconfig(8)AUTHORS
This driver was written by Jörg Wunsch, based on RealTek's packet driver
for the RTL 8002, as well as on some description of the successor chip,
RTL 8012, gracefully provided by RealTek.
BUGS
There are certainly many of them.
Since the rdp driver wants to probe its hardware at boot-time, the
adapter needs to be present then in order to be detected.
Only two out of the eight different speed modes RealTek's packet driver
could handle are implemented. Thus there might be hardware where even
the current slow mode is too fast.
There should be a DMA transfer test in the probe routine that figures out
the usable mode automatically.
Abusing a standard printer interface for data exchange is error-prone.
Occasional stuck hardware shouldn't surprise too much, hopefully the
timeout routine will catch these cases. Flood-pinging is a good example
of triggering this problem. Likewise, albeit BPF is of course supported,
it's certainly a bad idea attempting to watch a crowded ethernet wire
using promiscuous mode.
Since the RTL 8002 has only 4 KB of Rx buffer space (2 x 2 KB are used as
Tx buffers), the usual NFS deadlock with large packets arriving too
quickly could happen if a machine using the rdp driver NFS-mounts some
fast server with the standard NFS blocksize of 8 KB. (Since NFS can only
retransmit entire NFS packets, the same packet will be retransmitted over
and over again.)
The heuristic to find out the ethernet hardware address from the EEPROM
sucks, but seems to be the only sensible generic way that doesn't depend
on the actual location in EEPROM. RealTek's sample driver placed it
directly at address 0, other vendors picked something like 15, with other
junk in front of it that must not be confused with a valid ethernet
address.
The driver should support the successor chip RTL 8012, which seems to be
available and used these days. (The RTL 8002 is already somewhat aged,
around 1992/93.) The RTL 8012 offers support for advanced printer
adapter hardware, like bidirectional SPP, or EPP, which could speed up
the transfers substantially. The RTL 8012 also supports hardware multi‐
cast, and has the ability to address 64 K x 4 packet buffer RAM.
The driver should be layered upon the ppc driver, instead of working
standalone, and should be available as a loadable module, so the device
probing can be deferred until the pocket ethernet adapter has actually
been attached.
BSD December 21, 1998 BSD
