Hello. I recently acquired a NetWinder...

[Daniel Gimpelevich]

On Mar 6, 2006, at 3:24 PM, Ralph Siemsen wrote:

> Daniel Gimpelevich wrote:
>
>> I was referring to the website of NetWinder, Inc. The site currently  
>> at  that address appears to be nothing more than a server running  
>> blogging  software that had a few random quotes from the previous  
>> website pasted  in. WHOIS shows something in the South Pacific. I can  
>> only surmise that  at some point, NetWinder, Inc started hemorrhaging  
>> assets, and that's  who picked up the website.
>
> You know as much as I do at this point!

Regarding the website, perhaps, but you would know more about who ended  
up with the licenses and copyrights, which is what I originally asked.

>> I was under the impression that the firmware included a miniature  
>> Linux  kernel. Wouldn't that make it a derived work?
>
> Indeed, but the situation is complicated, please see old mailing list  
> achives if you really want to go there.  The short of it is we don't  
> have permission from the copyright holder, even though they should  
> release it, since there is GPL code inside.  Going after them isn't  
> likely to help, especially since they seem to not be interested in  
> netwinder.  It leaves me in an uneasy situation, I don't want to deal  
> with legal issues on either side, so the only choice was to pull the  
> binaries.

Evidently, any such discussion took place during an era of the mailing  
list archives that I have not yet found. This synopsis seems slightly  
paradoxical, since a lack of interest in that property is hardly  
consistent with actively turning down requests for permission to use it  
rather than taking advantage of opportunities to get rid of it.

>> Mark Lord's nw-9 live image is what I installed for the time being.  
>> It  had some brokenness that needed tweaking (e.g. X was not setuid  
>> root),  but it seems to be, in a way, the worst of both worlds:  
>> missing some of  the NetWinder-specific stuff characteristic of the  
>> DM images, yet  missing the currentness of Debian (I have not yet  
>> looked into Gentoo  for ARM, but based on my past Gentoo experiences,  
>> I don't expect to be  choosing that option.). BTW, the "stuff  
>> currently missing from the  netwinder.org website" also includes a  
>> period of mailing list archives  that may or may not prove somewhat  
>> enlightening to me.
>
> When netwinder.org moved to OSU, only the web and FTP sites moved, and  
> this means we have lost access to things like the pipermail list  
> archives.  I figured with services like gmane and marc this wouldn't  
> be a problem; but perhaps I am mistaken.  What era of stuff are you  
> trying to find?

The list archives in the old FTP site stop in early 2000, but the  
current pipermail begins in late 2004. I have not yet seen the  
pipermail archives for the intervening time period.

>> Easier said than done, since the board appeared to be multilayered.  
>> If  I knew why they actually included the connectors instead of just  
>> pads  like for the chips/magnetics, I might have a better idea.
>
> I believe it was an FCC regulation that you are not allowed empty  
> holes on the box, so the connector was populated for this reason.

Interesting. Makes me want to see a schematic even more...

> Helpful photos:
> http://netwinder.osuosl.org/devteam/ralphs/public_html/images/nw- 
> top2.jpg
> and also
> http://netwinder.osuosl.org/users/t/tinymoth/public_html/ 
> daughtercard.htm

So the board was fully populated at one point? Any idea how that  
interfaced with software? Also, I don't see a Winbond video codec chip  
anywhere in these pictures. Where would it have been? And what kind of  
signals would be accessible through that little card-edge connector  
under the fan?

>> Only 1.4-3 is available on debian.org, and most recent word on the   
>> matter appears to be this:
>> http://lists.debian.org/debian-arm/2005/11/msg00026.html
>> Version 1.8-1 has been "pending upload" for two weeks.
>
> There was a lot of off-list chatter to get all the little bugs worked  
> out; 1.8 was the result.  I can send you a copy privately if you wish.
> Actually I should just get my mirror master in order...
>
> > BTW, your
>> directory includes some of the NeTTrom binaries you said should not  
>> be  anywhere.
>
> Hmm, indeed, they are showing up elsewhere too.  We had chmod'ed all  
> these files to make them inaccessible (we were hoping to resolve the  
> issue with NetWinder Inc.), but it seems that the permissions have  
> returned.  I'll be correcting this ASAP.  Thanks for pointing it out.

Ouch! You even deleted the changelogs! I guess now I'd better hope the  
firmware in the machine never gets corrupted...

>> In the last kernel RPM available in the updates directory, dated   
>> 01/2004, FastFPE is compiled as a module while NWFPE is compiled into  
>>  the kernel. Boot messages show that NWFPE is in effect with double   
>> precision. How are they irrelevant to binaries that may also be used  
>> on  non-NetWinder ARM hardware with hard-float support? I think using  
>>  soft-float on those machines is rather dumb, unless libfloat can  
>> detect  hard-float support at run time and use it.
>
> Granted, if you want to share binaries across machines, there are  
> issues, however, as there are very few ARM systems with FPE's (the  
> StrongARM doesn't support one at all), I think you'll find that  
> soft-float will become the dominant way of doing things on ARM.

Seems slightly wasteful if libfloat can't use hard-float support even  
in the few instances when it's there...

>> http://www.gossamer-threads.com/lists/linux/kernel/610127
>
> So we're in unsupported land.  Wonderful!  Most of my netwinders run  
> 2.4.25 kernel so I haven't noticed what's missing in 2.6.x...

The actual post that shows up in Google is this one:

http://lkml.org/lkml/2006/1/20/427

It's not all that discouraging, yet...

>> That would explain why I couldn't get VGA mode to work with the  
>> 01/2004  kernel RPM. I would expect VGA mode to be faster. Should I  
>> expect it to  ever be fixed?
>
> Probably not.  Framebuffer is necessary for running X, as there is no  
> cyberpro support in X.org to my knowledge.  The DM disk images had a  
> hacked-up X server, but this work never got merged back, AFAIK.

That's the most disappointing thing I've heard so far. I think  
framebuffer reliance is a rather steep price for a software upgrade.

>> I was under the impression that those apps used the Philips and/or   
>> CyberPro chips, ignoring the Winbond. Am I wrong? I would expect the   
>> Winbond to improve performance over the Philips/CyberPro alone.
>
> Not sure to be honest.  The chip may be depopulated on later boards.

I would think its presence would be coupled with the presence of the  
Philips chip. Where on the board was this Winbond chip supposed to be?

>> Since NW9 and beyond (e.g. Debian) are the established future of the   
>> NetWinder, I believe that it would be worthwhile to preserve its past  
>>  in that manner.
>
> Agreed, and the recent work on nwutils is a step in that direction.

Maybe the first step, but everything Corel-specific and Rebel-specific  
in the DM images should ideally be available.

>> I may obtain an IDE flash drive, which I was considering for the   
>> NetWinder, but I can't figure out a way to have it physically inside   
>> the case along with a hard disk. Any ideas?
>
> The modern IDE drives are much thinner than the original ones; 9mm  
> versus 15mm if memory serves.  If you have an original drive, its  
> probably past its service life anyways.

It is the original drive. Since it hadn't even had power applied to it  
for a period of at least 5 years (as mentioned in the thread to which I  
linked at the top of my first post), I would assume it still has some  
service life left. The drive I was originally planning to replace it  
with is a DeskStar, which it says on your site that I can't use for  
that. The alternative drive I had in mind was the original drive pulled  
from a laptop that has been heavily used since it was new in 2000. It's  
not much thinner. The flash drive is a bit dated as well. Even two 9mm  
drives could not be easily sandwiched in that space AFAICT.

> You can also do some clever cable-folding to make use of "free" space  
> in the case.  Using a two-headed IDE cable, run it on top of the HDD,  
> then do a 90 degree diagonal bend, and you'll have room on the side.   
> The only issue is supporting it mechanically.

Room on the side? You mean put the second drive in sideways? I don't  
quite see how that's possible...

As for the power supply, I have two with the Corel label, and their  
plugs barely fit the power connector, becoming entirely dislodged at  
the slightest touch. One of them I have not tried with the unit because  
I was told that its operational status was unknown, and indeed, its  
voltage without load is slightly lower than that of the one that  
currently works.

>> Even today, 64-bit 256MB SODIMMs are not all that cheap. I figure  
>> very  few NetWinders ever had 256MB in them, even though they could  
>> really  use the boost. I think some type of banana-board would  
>> certainly be no  less feasible than some of the other RAM upgrade  
>> options that have been  proposed. I'd bet that some units are still  
>> stuck with only 32MB. Some  type of light at the end of the NetWinder  
>> RAM non-availability tunnel  would greatly expand the possibilities  
>> of NetWinder use. While the  prospect of every remaining unit getting  
>> upped to 256MB may only serve  to increase software bloat, their  
>> potential as internet appliances may  offset that if concurrent  
>> processes to run are carefully chosen.
>
> If you're adventurous and want to try it, sure :)  I have lots of 32MB  
> machines that would love more memory, then again, going from 64 to 128  
> didn't really give me the kind of boost I was expecting; I think  
> applications just assume you've got a gig nowadays.

An app's memory consumption expectations are really  
architecture-dependent. Going from 64 to 128 wouldn't sound as  
attention-grabbing to me as 32 to 256 or even 64 to 256. Anyway, at  
this point I have no clue what info would be needed to bring a  
banana-board solution any closer to reality.