[Prism54-users] 802.11g pci success story

[Eric "Shubes"]

Hats and Horns! (well, sort of)
I managed to get a pci wireless-g card working (finally), and want to
report my findings.

First and foremost, the most difficult task by far was finding and
acquiring a card that had any chance of working natively. It turns out,
you often can't tell what's what from the card's model number, as many
vendors use (very) different chipsets in different production runs, then
continue to use the same packaging with the same model number for
distribution. At best the vendor will use a sticker designating a
different part number, but that is of little value since the part number
is rarely used for identification purposes. For example, you can order
an SMC2802W (as I did), but you've no way of knowing whether you'll get
the V1 that works, or the V2 that doesn't (yet?), and in this case they
both even use the PrismGT chip! The only way to accurately tell what a
card actually has is to use the lspci command after it's installed.
Thank you very much (NOT) Mr. card vendor! Now, on with the story.

I already had a Linksys WMP54G working under W98SE (dual booting w/
FC2). The lspci command reported this card as having a Broadcom chipset
(Broadcom isn't Linux friendly, to say the least), so the only way to
get that working would be to use NDIS along with the windows drivers.
That 'solution' has several drawbacks, noted at
http://acx100.sourceforge.net/ndis_cludge.html, so I proceeded to find a
card that is supported on Linux.

The PRISM(R) chipset is reported to have REAL vendor Linux support (the
2.6.8 kernel even says so in the log!), so I proceeded in that
direction. The table of cards at http://prism54.org/supported_cards.php
was quite helpful, although somewhat outdated (Mar'04). I also referred
to the Wlan-ng table of cards (house of cards?) at
http://www.linux-wlan.org/docs/wlan_adapters.html.gz, a quite extensive
list, although it too is outdated (Feb'04).

I proceeded to shop stores in the Phoenix area, and was unable to find
ANY pci card containing the prism54 chipset, based on the data in these
tables. I was especially surprised that Fry's didn't have one, although
I must give them credit. They were willing to open every kind of pci
wireless-g card they had, looking for the (becoming mythical) prism54
chip. (Note, on some cards you can see the chip itself and tell who made
it, providing the vendor hasn't covered it up with a sheet metal cage.)

Delayed but not discouraged, I went online to shop. After doing some
price comparisons, I decided to purchase from newegg.com, as their
prices were competetively low, and I've had very good experience buying
from them in the past. My first choice was the Hawking HWP54G, as it was
reported to contain the Prism54 chipset, and I've had no problems with
Hawking products (nic, switch) in the past. Unfortunately, that card was
out of stock, so I opted to have them notify me when it came in. As a
second choice, I ordered the the SMC2802W, as it appeared to have a
large degree of success, and all models contained the PrismGT
(1260:3890) chipset. It shipped the same day, and I received it two days
later. In the meantime, the next day I received notice that the HWP54G
was in stock, so I ordered that one too (I could use it if need be).

The SMC2802W was recognized by kudzu (FC2), so that was encouraging. I
proceeded to follow the directions at
http://mysite.verizon.net/winterz/doc/prism54_fc2.txt while referring to
the wireless tools documentation at
http://www.hpl.hp.com/personal/Jean_Tourrilhes/Linux/Tools.html. First
there was one FC-specific bug (request_firmware() failed for 'isl3890')
that needed fixing (commented out 4 lines containing "hotplug" from
/etc/rc.d/init.d/network script, it was leftover from older hotplug
version). Then I was getting a "set bitrate failed" (Auto) error, so I
removed that configuration setting (letting it default). At that point,
I stumbled across the same bug as the one reported at
http://prism54.org/cgi-bin/bugzilla/show_bug.cgi?id=72. As it turned
out, I had acquired the V2 model, as verified by lspci (same
vendor:device, but different subsystem id). Since this bug's severity is
listed as "enhancement", I decided that this card needed to be returned. :(

In the meantime, the HWP54G had arrived, so I proceeded with its
installation (at the same time, removing the SMC2802W V2). This card was
not seen by kudzu the first time up, why I don't know (kudzu did
recognize it on a subsequent boot). I checked lspci to see what chipset
the card contained. Egads, it's a TI ACX111! Not to be dismayed, I went
to http://acx100.sourceforge.net/ to see what was up with that chipset.
"Craig's Acx100 Guide for Linux" at
http://www.houseofcraig.net/acx100_howto.php was extremely helpful in
getting me going. The driver is still extremely experimental (especially
for the ACX111), so I was a bit leery. It is only available as source,
but Craig's Guide is very detailed, so much so that even a newbie could
get it installed. I proceeded to follow the instructions. Everything
went very well, until I got to the point of installing the firmware.
This part of the guide is somewhat sketchy when it comes to the ACX111,
as it was written originally for the ACX100 (wireless-b, I presume). The
  driver files mentioned here weren't on the CD that came with the card.
I grep'd through the ACX100 "package" files for the names of the
firmware files I did have, and found it referenced in the README file,
along with  some c code. I read the README (should've done that
earlier), and it clarified which firmware file I could use, so I
installed it. At this point, Linux was apparently talking successfully
to the card (iwconfig looked ok), but it wasn't connecting to my access
point (Linksys WRT54G). I proceeded to open up the AP a bit (less
secure, but what the heck) to see if they'd connect. I turned WEP off,
disabled MAC filtering, and enabled both g,b rates. Low and behold, it
connected. Wheee!

I entered the card's mac address in the AP's mac filter list and turned
mac filtering back on, it still worked (no leechers). I bumped the rate
up to 54M on the card and it still worked, but only when AP is set to
"mixed" (b and g), not G-only (that's odd). Then I tried DHCP
configuration, no go. I think that's because the acx start_net script
doesn't follow FC conventions (device is wlanx instead of ethx). Also,
WEP isn't fully implemented for the ACX111 yet (ACX100 is), so I'll have
to wait a bit (or write the code myself?) before I can use encryption
again, but I can live with that for now.

BL, I have a bit of tweaking to do to get it working automagically on
startup, but at least it's working and is usable (enough for me). I'll
be reporting bugs and keeping in close contact with the acx developers
for a while!
-- 
-Eric 'shubes'
"There is no such thing as the People;
  it is a collectivist myth.
  There are only individual citizens
  with individual wills
  and individual purposes."
-William E. Simon (1927-2000),
     Secretary of the Treasury (1974-1977)
  "A Time For Truth" (1978), pg. 237



****************************************************
This message has been scanned using Contraxx
Technology Group mail server v2.3 and is virus free.
****************************************************