Cannot Get Socket Wifi Card To Work Under Oz 3.5.3

[Mickeyl]

The 4137 (prism2.5 deriative) is _not_ compatible with the 4100 (prismII deriative) spectrum_cs driver. You _may_ have more luck with using hostap_cs + firmware upload. See the prism3-support package as well as the README contained inside.

[rjenkins]

I don't follow your reasoning..

If that's the case, how did russilwvong get exactly the same card as mine running with the same Orinoco driver, albeit a slightly older version?

I've been studying the Orinoco sources, I get the impression that it's a build problem or bug with the driver rather than a fundamental incompatibility.
(The Sharp 0.12 build works, the Opie 0.13e build does not.)

Note this line:
hermes_read_ltv(): rid (0xfd20) does not match type (0x0021)

The working example has Station identity 0021:0002:0002:0001
The 0021 is what identifies the card as a Symbol type to the driver.

The duff one is apparently reading the 0x0021, but when it's finished building the station identity block it's corrupt, probably offset by two words to give the 0002:0001:33fb:33cb

The driver then sees the initial 0002 in place of 0021 and goes haywire from that point, trying to handle it as an Agere device rather than Symbol.

More:
Working throught the hermes & orinoco code, it looks like the 0xfd20 is a command sent to the card to retrieve a specific data block.
This should return the block length, the original request code, then the data.
It's seeing the first word of the 'station identity' data in place of the block type code, then the remainder of the data is offset by two words.

It seems that somehow the buffering is getting out of step with the read data in the 'hermes' module.

This is during the basic, low-level, card init & ID stuff, before the driver tries to identify which specific card it's working with.

Could the be something to do with the structure alignment problem I've seen mentioned elsewhere?


Robert Jenkins.

[rjenkins]

And more:
It looks as if loading the card firmware is done during the PCMCIA 'card inserted' setup & this is working - the light on the card is flashing.

According to the Zaurus user group WiFi page, half the supported cards are Prism 2.5, including the Symbol & Socket cards.
http://www.zaurususergroup.org/modules.php...lessCardSupport

The driver they specify for these is 'the Spectrum/Socket CF driver by Pavel Roskin'. The page this is on is dated 2002, not exactly a new development.
http://www.cypherpunks.ca/zaurus/socket.html

(It's a different build of the package to the previous working one I'd seen from the Sharp site).

I don't want to go back to the Sharp rom, but it's starting to look like the only way of getting this Symbol card working in the near future
The reason I bought this in the first place was because of all the recommendations I'd seen for it... I assumed anything that worked with the rather old Sharp roms would have been covered in the OZ roms.

From some of the posts on the Orinoco mailing lists, it appears there could be a problem with builds for ARM against the more recent kernels, which does not seem to have been followed up?

Does anyone on here have the sources for the Orinoco drivers, as used on the 3.5.4Rc build & do you know a (working) link to the Sharp sources, or has anyone got a copy they have grabbed?
I'm aware of the Orinoco project & CVS, but I would like to compare the exact files used in the Z builds after any patching etc.

I'm setting up to do a full OE build from scratch so I can try & debug the drivers, but from what I have read so far that's probably going to take a very long time to get running sucessfully..

[lardman]

but from what I have read so far that's probably going to take a very long time to get running sucessfully..

No, it's quite easy actually (but I have been using it for a long while) - drop me a line if you need any help setting it up.


Si

[rjenkins]

Hi Si,
many thanks for the offer of assistance!

I've now got a machine set up with a clean install of Fedora core 4, plus all the required packages mentioned in the Open Embedded info & the ARM toolchain.

It looks like it should work, but I've not had time yet to try actually building anything.

One thing that bothers me with this setup, is how do you guarantee a consistent build? It appears to download and patch sources on the fly as it needs them, so each person building the 'same' (eg) OZ 3.5.4 package could actually end up with different binaries - is this correct??

Do the 'final' patched sources get saved, like into a local copy of the kernel source tree etc?


On the Symbol / Orinoco drivers again, I found this which I think is a definitive answer (from hostap mailing list archive):
-------------------
On Tue, 2005-05-17 at 13:51 -0500, Adam Roach wrote:
> I have a Symbol Spectrum24 High Rate LA4137 wireless lan card.  I was
> wondering if this is what I need to be using to get it working.

No, this card has Symbol chipset, whereas HostAP requires Prism chipset.
You should use Orinoco driver instead.

--
Regards,
Pavel Roskin
--------------------

I've also found more references to other ROM builds for the Zaurus series in which the Orinoco package works with the Symbol LA4137 Card.

Note that my Symbol card is 'new' but with a manufacturing date of 2003 & driver CD dated 2001; it's hardly the latest Symbol hardware version.
I get the impression that the very latest Symbol cards do use a different chipset, possibly Hermes2 from some list posts, but the 4137 is covered by the Orinoco spectrum driver.

I still think the problem is in the OZ build or the particular Orinoco source version used in it, or the libraries / tool chain used with OZ.
The dmesg errors are very similar to the ones Michael Lauer reported in the Orinoco mailing list, as if there is a problem with the 'hermes' module accessing the card hardware?


Robert Jenkins.