Slashdot Mirror

Since (almost) all wifi chipsets are closed, information and NDA-wise, it takes a huge amount of reverse engineering to find out the details of the devices, so that not only the internals of the hows and wheres in the chipset can be understood, but actually its firmware be improved/hacked/changed/redone in a way it does something much more complex than intended in the first place.

Having a look at the impressive effort of the Prism54 FreeMAC people gives an idea of the complexity of such work, and it also hints to another issue: To my experience, wifi chipsets vanish from the market within a year or less after they have been introduced, so once you might have successfully '0wned' them, craigslist and flea markets will be the only places to find your wifi dongle of choice. The fact that devices which have been made for "SoftMACs", i.e. those for moving most of the 802.11 complexity into the driver and thus away from the device hardware to the host computer, and that is the wide majority of devices today, have pretty scarce ressources for implementing anything not so trivial. For example, said Prism54 chipsets use an ARM946E core with a few dozens of kilobytes.

So, I think $100 isn't really enough to bring people into this, but rather the fascination of those devices and the vision of maybe finally being able to easily and very cheaply deploy wifi mesh networks which actually go below the one-watt-power consumption level per device - whether it's done for fun, sharing internet with your neighbors, providing networking in rural areas or doing IP-based sensor networks.

So, let's have a dream: Maybe Marvell, or the cozybit people that implemented meshing into the OLPC wifi chipset on Marvell's behalf, would be willing to release software, which would be a nice thing towards a less PITA realization of mesh micro devices, although OLPC's step to use a proprietary mesh protocol instead of open standards like OLSR, AODV, or the rather new but promising hands-on B.A.T.M.A.N. protocol wasn't highly acclaimed in the community. It would be even nicer if Marvell or any other wifi chipset vendor releases a chipset (or even a device) which comes with an extra application CPU-on-chip, so that wifi-specific and application-specific code could be decoupled, like we see it in most two-chip chipsets for mobile phones. Provided there is a defined and feature-complete interface, the vendor could still hide wifi-specific firmware from the eyes of the oh-so evil hackers, and yet have an embedded scale device that is capable of virtually everything Dan discusses - and more. And hey, throwing another ARM9 core and a few 100 kB flash and ram onto such a die can't be that expensive, can it? And who would not want such a chipset (or device), given the possibilites? Yes, you vendors can make it 50% more expensive than regular chipsets, I will still want them, and you will probably spearhead the market of embedded wifi capabilities within no time, cpmpanies will buy from you like crazy, and your ROI will skyrocket, so you actually can produce them for cheap... ;)

BTW, the reason for the closedness of the drivers and especially firmware (which becomes partially indistinguishable in "SoftMAC" setups) is, apart from the pure evilness of companies, the required FCC certification: since wifi adapters nowadays are pretty much software-defined radios, you could do a whole lot radio stuff with the devices, once you change their firmware, and that is something the FCC (and similar agencies around the world) wouldn't be so fond of. At lest this is what I heard Eben Moglin elaborate on during some conference, when he talked about the FSF's in

Your FLOSS ipw2200 driver relies on proprietary, non-auditable, possibly-evil, binary-blob firmware.

There are chipsets with FLOSS firmware - Prism FreeMAC for one: http://prism54.org/freemac.html

Oops... I forgot to mention the Prism54 project... I really like this driver a lot:

http://prism54.org/

Unless you want to do some snooping or personal intrusion detection with kismet.
Last I knew linuxant and ndiswrapper didn't support the hooks that were needed by kismet.

I was a happy Linuxant customer before needing this functionality.
Now I'm a happy customer of both madwifi on my laptop and prism54 on my desktop.

Mod redundant if you want but I have to second NdisWrapper. I remember downloading it back during the 0.x days so that I could get a NETGEAR PCI card working... now I've got that, about two tons of USB adapters, and who knows what else.

And there's also Prism54, although I couldn't really find the downloads or whatever because I have to admit they have a pretty miserably-organized site. And there are a few others as well that I can't remember of the top of my head – the RA-whatever ones come to mind.

So wireless support really isn't that much of a problem.

There's also this project:
http://prism54.org/

But be aware that it doesn't support WEP/WPA yet. The guy who's running the site says WEP/WPA is easy to add, so I'm puzzled as to why it's not in yet since it's really important for security :-)

I have tried to run both my Netgear cards WG511v1 and WG511v2 on Suse 9.3. Version one works pretty well as Suse has in built support for the prism chipset. Earlier, It wasnt working properly as i was using some other driver. I have used prism54s driver(http://prism54.org/ but failed. However, for the version 2 card having Marvel chipset i used ndiswrapper to get it working. Not to mention I tried using every other possibilites as well. Using Ndiswprapper was easy and no special configurations were required. All in all I would like to say for a user who is not well versed with linux. its difficult for him/her to follow. I would really like to divert the attention of companies like Netgear to provide driver support for Linux as well.

Newer Prism54 chipsets are not supported. Intersil, who made the Prism chipsets, were bought by Conexant, and they have always been among the worst to get information from. The Prism54 driver project are working on reverse engineering, and make much progress, but don't hold your breath. And don't buy a card with Prism54 chipset just now, because they all use one of the newer unsupported revisions.

Ralink, as you say, have good support. Thanks to them, I have working wireless in Linux on an old Mac. Drivers are at rt2x00.serialmonkey.com.

I'm not quite happy with configuration yet. My setup works fine, but it's static. I'd like a nice GUI for choosing APs, setting passwords,etc., when on the move. There is a utility for the old v.1 RT2500 driver, and it seems to work, but I didn't quite get internet access when trying to use it (I did get access to the LAN, though). For now, I use an ordinary network configuration (/etc/networking/interfaces in Debian). It works, and I'm happy. I don't think everyone else would be.

• Important Stuff

• Important Stuff

> nvidia implements a standard driver [which X can talk to and works in OpenGL].
> They follow standards. The fact that the source is closed shouldn't bother people.

Speaking only for myself, the main difference is that the nVidia driver has to be linked with the kernel.
This may seem like an unbelievably eggheaded reason to dislike it, but I'll try to explain. When nvidia's
giant binary blob is linked with my kernel, all bets are off as to debugging any problems that might arise. If
the driver crashes, it takes the kernel down with it. The nvidia module lives in the same place as my hard drive and
filesystem drivers, so it could presumably cause data loss. All manner of undesirable consequences.

I work a lot with wireless cards; here's another example. Atheros has generously assisted with the creation
of a very high-quality mostly-open driver for their cards in Linux. I say mostly because of a sizable
binary HAL that is linked in, to prevent free reprogramming of the radio (FCC regs, blah blah).

On the other hand, you have the Intersil prism 54g. It has a fully open-source driver, also high-quality
and functional. They comply with non-reprogrammability requirements by having a binary-only firmware
that is loaded at module init time. Now, there's still closed, untouchable code in there... but it runs on the card's chipset,
not on my CPU where my kernel and other things live. And, since the driver is fully open, it's included with
the mainstream kernel, so much more convenient and more likely to work with any of the weird kernels I cobble together.
And a bug in that card's firmware is much less likely to cause my whole system to come crashing down, although
it's certainly possible.

(All bitching aside, much thanks to nVidia, ATI, Atheros, Intersil, and all the other hardware companies out there
who throw Linux a bone every now and then.)

Prism I think is the best you're going to get. The firmware isn't open source, but the drivers are. Best PCMCIA card with the chip (IMHO) is the Netgear WG511 (notWG511T).

c.

Here, I would like to call everyone's attention so people get rid of the cloth in front of the eyes and see the real status: some do NOT provide info, like Conexant for their new generation prism54 (Intersil did for the first gen), Intel for their 2100 or 2200 chips or TI for the acx100. Others provide binary only drivers, like Atheros (dig in the OpenBSD source, they reverse engineered the atheros hw abstraction layer). If they do not want to help at all, fine, but do not say they are nice for PR reasons.

For me "Open Source Network Sniffer" covers the full kernel and the apps used for the sniffing. Please, think twice before affirming something is open source (binary drivers are not, even if the stub code is open source) and also that the company is open source friendly (provide help to Linux, *BSD and similar coders, maybe even the driver, is). At this moment, the only few I know that still are pro open source drivers, or even provide support (if my investigation isn't wrong), are Realtek and Ralink (and Intersil, but it doesn't exist anymore).

And for those that think binary is better than nothing... then why *BSD or Linux at all? If it starts with "but is wifi card" or "well, it is only the video card", I don't see why not apply that logic to the OS anyway, or all the apps too.

Thanks for your attention.

If you're using a Prism-type chipset, you can either use hostap (for 802.11b) or the Prism54 drivers (for 802.11g) to run your card in "Master" (AP) mode. Instant linux-based AP.

I play with this off and on with my laptop and the high-powered SMC-2532W-B card, which can take an external antenna. Crank up Apache, set up BIND to return the laptop's wifi IP address to every query, and away I go (hey, wifi isn't JUST for The Internet(tm) after all). Or if I feel the need I can bridge to a wired connection, but that's boring - ANYONE can do that...

Anybody know if any of the 802.11g Prism-chipset-based PCMCIA/Cardbus cards exist that can take and external antenna?

> Is the MM-20 less laggy than the MM-10? The CPU throttling & slow HDD sometimes makes me crazy.

It's quite a bit faster in terms of raw CPU torque, yes. HDD is not appreciably faster. But, it's got the best currently available chipset for 54mbps wireless under Linux. =)

If you want a huge leap in speed, without a great deal more mass, get a Raven. That's what I use. =)

Bullshit!

There are Linux drivers for both Prism and Atheros chipsets, which are used in Linksys and Netgear cards.

Unfortunately, your WG121 is not among those, but the Linux Prism GT driver project at least mentions it (although with a pretty disturbing "unknown status" and a "success rate" close to 90%, which seems kinda oxymoronous), so maybe it's worth a try. Atheros chipsets are supported by the MadWifi project, too...

I would suggest SMC's 2802W (PCI) and 2835W (PCMCIA). They're reasonably priced, and work flawlessly with the Prism54 driver, which is part of the most recent 2.6 kernels.

It's the fault of the Broadcom chip they started putting in them. Which is Linksys's fault for switching to a chip that has drivers closed tighter then a virgin's legs. The wrapper provided by Linuxant can use the Windows drivers. Also the Prism GT people are working on other chipsets at a better price.

Drivers for the prism54 chipset have finally been merged, which means that the vanilla kernel now has support for 802.11g (54 mbit) wireless lan. The prism54 chipset is used in whole bunch of 802.11g cards.

Drivers for the prism54 chipset have finally been merged, which means that the vanilla kernel now has support for 802.11g (54 mbit) wireless lan. The prism54 chipset is used in whole bunch of 802.11g cards.

If you have the Prism Chipset you should head to http://prism54.org
There's a working linux driver and lots of support. However, it's still VERY experimental and can be tough to get working... but it does work.

Just a followup to this part of my own post - evidently, according to the January 8th 2004 "News" post at www.prism54.org (Prism GT driver for Linux) they DO have "Master" mode working and you CAN use supported cards as an access point. Nifty, now I've got an excuse to upgrade my "main" laptop's wireless card to an 802.11g and use the 802.11b that I'm currently on for my little project...

802.11g is also accessible through Free Software/Open Source. Hell, Intersil wrote drivers for their Prism GT/Indigo/Duette chipsets on their own and then GPLed them. (See prism54.org for some of the work being done with them.)

These card are relatively inexpensive. There's no particular reason to pick a Centrino laptop because of the built-in WLAN support.

IMO there is nothing wrong with the current driver architecture in 2.6 - what I believe is wrong are the old attitudes among hardware manufacturers with regards to releasing programming information, which forces them to write binary drivers. And not to sound dumb, but the Slashdot crowd is definitely aware of all of the problems surrounding binary drivers in Linux.

The hardware folks need to learn that releasing programming specs and allowing the community to do their own drivers (example: the Prism54 folks) is not the Kiss of Death. If they have trade secrets that they want to hide, then they can hide them in firmware or in a BIOS.

The community is willing to do the work of maintaing drivers. All the hardware folks have to do is learn a new attitude with regards to the propagation of open-sources OSes.

For 11b, see this page - most of those drivers are open-source.

For 11g, prism54 is the only open source driver I know of (or it will be once a license issue is resolved - there are still two non-GPL header files). There's a non-free firmware file you have to load onto the card, but IMHO that doesn't make the driver non-free - most modern devices contain firmware anyway, whether you upload it or not, and none of that binary code runs on your CPU or taints your kernel. The driver is still portable to other architectures, etc.

The Atheros madwifi drivers are partially open-source, but require you to load some binary code into your kernel. Personally I'd avoid this non-free driver, but many people are happy with it.

Although the DriverLoader apparently supports these cards, please support these companies in either helping develop Linux driver support or releasing specifications (both of which Intel and Broadcom adamantly refuse to do) by

a) purchasing their products when you have a choice (e.g. buy Pentium-M instead of Centrino and add on a third-party wireless card, and don't buy 802.11g products from Linksys or Dell which use Broadcom), and

b) Use the open-source drivers rather than emulating windows drivers, let the chip (Atheros and Globespan/Virata nee Intersil) and the card companies know that you appreciate their linux support. Report bugs and feedback to the open source projects, too.

It's nice to have something like this around as a stopgap way to load drivers for hardware made by manufacturers with poor linux support, and even as a way for manufacturers to ship initial drivers for linux inexpensively for them (and claim "linux support out of the box"), but it is no substitute for published specs and real drivers (which, with published specs, the companies don't even have to develop themselves).