[svlug] Expectations for Windows XP and Linux
Rick Moen
rick at linuxmafia.com
Thu Jul 27 21:04:43 PDT 2006
Quoting Jim Thompson (jim at netgate.com):
[I'm of necessity skipping a lot, for lack of time.]
> 1) I got trained (with a switch) by a couple lawyers who didn't like
> to use the word "partnership" where such a legal entity wasn't intended.
The term "business partner" outside of a legal context -- which this was
-- means something decidedly different. And (the larger point) all of
that was (yet another) time-consuming distraction.
I really do appreciate talking to you, Jim, but sometimes the weight of
digressions, rhetorical timewasting, and incredibly long-winded mistaken
assumptions is pretty brutal.
> 2) I really like *not* being an employee. :-)
I hear ya.
> My only reason(s) for taking it private were:
>
> 1) I didn't really want to subscribe to svlug (I'm not in Silicon(e)
> Valley, and have no intent of leaving Hawaii to move there. I'm
> involved with what passes for the local LUG here, though HOSEF has
> different goals than most LUGs.
> 2) The message to svlug would bounce until such a time as I subscribed.
> 3) I wanted to give you a chance to understand the 'why' without
> seeming to attack you in public.
Thank you. This does you credit, and I appreciate it.
However, in the long term, these sorts of things really do need to be
discussed in public. (Even better would be some lasting documentation
of relevant analysis and disclosure _from_ companies like Atheros that
are involved.) It's a sort of spinal impulse I tend to have: I see
technical subjects rehashed repeatedly, and think "This needs to be
researched and FAQed, for clarity's sake." I'm typically not out to
convince anyone of anything; I'm pursuing a clear understanding of the
issues.
(Again, I have no stake in the hardware in question. My interest is
mostly in documenting a controversy I keep seeing coming up.)
[right to redistribute is not present for, e.g., Conexant/Intersil
and Intel "firmware images".]
> But is for the madwifi HAL.
Correct, and noted.
> >To repeat the earlier comparison, I wouldn't particularly have
> >minded if my Lucent Orinoco and Cisco Aironet 802.11b cards had used
> >"binary image" files instead of ROMs, as long as -- like ROMs -- they
> >would Just Work in perpetuity until the device burned out, and be
> >OS-neutral.
>
> You might have minded if you wanted your XXX card to do something new
> and different (say, WPA?) and couldn't make it do that.
>
> You might have minded if you wanted to experiment with alternative
> MACs, and couldn't.
Yes. These suggest that _minimal_ functionality should be locked up
in {microcode|ROMs|blobs}, right? Sorry if my assessment is vague;
I don't try to play EDA designer on the Internet. ("I'm a sysadmin,
Cap'n, not a chip designer.")
Trying to look at this in a broader perspective, it's _nice_ if you can
make hardware do new tricks that it couldn't do at the time you bought
it, but you don't feel cheated if you can't, right? You either buy a
replacement, or you hack a software implementation, or you do without.
(I'm unclear on what would interfere with switching MACs. That ought
to be controllable within the network stack. To my knowledge, it always
is -- and in theory is even the case on OSes that lack features like the
"hw ether [MAC]" parameters to ifconfig(8).
> I'm OK betting you a beer (or three, though I recently stopped
> drinking for no good reason), but you're asking Atheros to
> (literally) "bet the farm" by requiring them to "open source" the HAL.
Er, you've not been heeding what I said. I didn't say that.
At the time of the rapidly-pyramiding private e-mail conversation with
which you (pardon _my_ being surly for a minute) bushwhacked me, I
hadn't had a chance to think things through very thoroughly. Wifi is
not a professional concern of _mine_, and I have a whole lot of other
things to worry about, so my off-the-cuff reaction to Madwifi after
a friend on the CABAL mailing list advocated it is "This thing that's
heavily proprietary to the point of not even being portable to changed
architectures. That's wacked, and is a problem."
Other wifi cards have managed to avoid that excessive _degree_ of
proprietisation; what's wrong with Atheros?
I've asked that question in such a number of ways that I'm not sure for
which value of "n" that's the nth time.
In that private mail conversation that I found myself thrust into
unexpectedly, you were rather vehemently saying (summing here, and
sacrificing fine accuracy for brevity) that open source in this context
was not possible, and reverse-engineering could not be permitted. I
rejoined to the contrary, saying _that_ situation was not permissible,
and we of the open source community insist on being able to be in charge
of our own software.
You seem to have translated that as "Rick's position is that everything
in the Atheros HAL must be open sourced". But we never got to that
point in the conversation -- and that is _not_ my view.
> >By your logic, Atheros is already in violation, since it has
> >reportedly taken _no_ steps as required on "software radios" per
> >section 2.932, the section immediately following the above:
> >
> > (e) Manufacturers must take steps to ensure that only software
> >that has been approved by the FCC or a TCB can be loaded into a
> >transmitter. The software must not allow the user to operate the
> >transmitter with frequencies, output power, modulation types or other
> >parameters outside of those that were approved. Manufacturers may use
> >authentication codes or any other means to meet these requirements,
> >and must describe the methods in their application for equipment
> >authorization.
>
> Atheros took such steps by limiting access to the HAL code. You
> seem to object to such steps.
You seem to have utterly and completely ignored my point. 2.932(e)
requires technical measures to ensure that no _alternative_ software
capable of changing frequences, power output, etc. be capable of being
loaded.
> a) If you can view the HAL source and this "third-party HAL", you can
> quite clearly see that one is a derivative work of the other. [...]
As you and I discussed, obviously nobody who's not an Atheros licensee
can verify this assertion -- and I didn't deny it; I just cannot (and
did not) comment on its merits either way.
Surely that was clear the first several times?
> Prism & Hermes (and even the Broadcom and Intel 'Centrino') MACs all
> have computational resources on board. They *run* the firmware. You
> seem to be under the (false) impression that they simply push some
> values into the correct registers and get out of the way.
I neither thought nor expressed that view.
> The Atheros (and rtl818x, as well as others) MACs are all hardware
> state machines. They have no CPU resources to exploit. (They have
> AES/TKIP/WEP cores which can operate on packets in external memory,
> and they are capable of generating some control frames that have
> tight timing requirements (CTS, ACKs), but they are a fundamentally
> different architecture from the chipsets you endorse.
>
> Are you taking the position that the hardware is broken (or unusable)
> if you don't have source code with which to operate it?
What an utterly wacky, bizarre, and non-sequitur question! Why would
you think I might hold that view? Did your eyeglasses break before you
read what I said to the contrary?
> Moreover, the actual "HAL" (not the OS-specific wrapper) is OS-
> agnostic, though not CPU agnostic.
Well, there you go (and I thank you sincerely for the technical
rundown). That's the respect in which I think something's wrong:
One should be able to take a peripheral device and move it to the same
bus on a new architecture and it should _work_ at a hardware level
(possibly after lobbing binary code into it, in some documented fashion)
-- without needing some new type of blessed binary thing. The
programming interface specs should be available, so that people can
write drivers for it on arbitrary present and future operating systems
-- even if the manufacturer goes broke and thus unable to help.
When those things are not true, I consider the hardware in question
undesirable as not being general-purpose computing gear. It's not
future-proofed; it's inherently crippled.
Now, please note: I am not _married_ to the above view. It's my best
current effort at a reasonable strategy in dealing with a world where
hardware functions are popping up embodied in software, and deciding
what proprietary aspects of the latter are of concern, and what are not.
If I later revise my opinions because I learned something new, kindly
don't accuse me of being "inconsistent", unless you don't mind being
called a pinhead in return. ;->
> >>From time to time I debate making someone who is a) trustworthy and
> >>b) trusted in the community subject to a "netgate" NDA and showing
> >>them the HAL source code, likely acting as a contractor to the
> >>company, just so yet another person could say, "Yep, there isn't
> >>anything in there to fear", but then, you don't want to believe me
> >>when I say that, and I have no real reason to toe the Atheros party
> >>line (other than it happens to be the truth), so why would adding
> >>one more person help quell the discussion?
> >
> >I'd be glad to answer your (rather bizarre, aside from being
> >pointlessly aggressive) question if you can calm down, and then tell
> >me what in Sam Hill it has to do with the preceding discussion. I
> >can't for the life of me imagine why you've asked it.
>
> Other smart, honest people have looked at this and honestly see no
> way around it.
You're missing the point: You're apparently attributing to me some view
that I don't hold, since otherwise posing the question to _me_ makes no
sense whatsoever.
> >I have no objection to binary-only code in my option ROMs, or in
> >ROM-equivalent "firmware images". I live with it every day, as does
> >everyone else on this mailing list: The server I'm posting this from
> >has binary-only motherboard firmware, binary-only SCSI firmware, and
> >probably binary-only ethernet firmware, not to mention binary-only
> >microcode in various places.
>
> Few of these would survive if you moved them to a PowerPC machine, or
> even a 64-bit x86 machine. (I'm assuming you're on something 32-bit
> x86 based.)
We consider (e.g.) PCI v. 3.0 NICs and SCSI cards defective if they didn't
work equally well in x86_64 machines as they did in the IA32 machines
for which they were designed. And, if memory serves, I _think_ I
remember Adaptec HBAs for "PCs" working just fine if you moved them to
PPC Macs with the right kind of slot.
And all of those would have been perfectly acceptable (if a bit
unwieldy) if the same OS-neutral, CPU-neutral, architecture-neutral code
were embodied in a "blob" instead of firmware. Which is the distinction
I'm trying to draw between, on the one hand, the likes of Intersil Prism
GT (which has such environment-neutral code embodied in software) and
the old Lucent Orinoco (which has that in a ROM), and on the other, the
likes of Atheros/Madwifi, which doesn't.
> If by "hardware platforms" you mean CPU architectures, then yes, its
> a straight-forward recompile after building a cross-compiler or
> hosting a compiler on
...which can be done, if I understand correctly, only by the
proprietary-code maintainers. That is the sort of thing that is not
OK, for reasons above among others. Vital interests are not compromised
by having a software blackbox exactly analogous to the hardware
blackboxes used for that same function, but they _are_ by not being able
to compile a functional system without needing to get a blessed binary
for any new environment from some proprietary code maintainer who may
not even be around any more.
> The actual "madwifi" project doesn't have access to the HAL source to
> do this. Atheros does, Sam Leffler does, I do, and around 20 other
> people that I could name do. Any one of these parties could run the
> compiler for you, should you develop or encounter a new CPU
> architecture that needed wireless.
>
> The situation is not ideal, but its very likely as good as it can get.
And yet, you got better a situation in that respect with Intersil Prism
GT/Duette cards, Intel Wireless Pro 2100 et alii, and others. Thus my
point.
More information about the svlug
mailing list