Dear all,
I recently came across a very annoying question regarding Linux compatibility.
It rises a fundamental question which should be discussed, IMHO.
Facts are:
I bought a card from some vendor, claiming "support for Linux". I tried to make
it work in a configuration with a standard 2.4.20 kernel from kernel.org. The
drivers (kernel modules) are binary-only. They did not load because of a
version mismatch. Asking for versions loadable with standard kernels, I got the
response that they only support kernels from Red Hat and SuSE, but no standard
kernels.
This leads to my simple question: how can one claim his product supports linux,
if it does not work with a kernel.org kernel? Is there any paper or open
statement from big L (hello btw ;-) available what you have to do to call
yourself "supporting linux"?
I know that the technical background is ridiculous, because it should very well
be possible to recompile their drivers under stock 2.4.20, but it looks like
they don't want to, simply.
I am in fact a bit worried about this behaviour, because I take it as a first
step to a general market split up already known to *nix.
My general conclusion would be that something not working with a standard
kernel cannot be called "supporting linux", no matter what distros ever are
supported. You may call me purist...
Any ideas?
Regards,
Stephan
> This leads to my simple question: how can one claim his product supports
> linux, if it does not work with a kernel.org kernel? Is there any paper or
> open statement from big L (hello btw ;-) available what you have to do to
> call yourself "supporting linux"?
What about the following: a vendor provides linux drivers for one of its
products (thanks!). These drivers simply do not work with some of its
other products (all variants of the same basic product). There are no
linux drivers for these other products. All the products claim to be
supported under linux. This "linux support" is explicitly stated on
each product's web-page whether it is really supported or not.
Duncan.
On Maw, 2003-05-13 at 14:16, Stephan von Krawczynski wrote:
> I bought a card from some vendor, claiming "support for Linux". I tried to make
> it work in a configuration with a standard 2.4.20 kernel from kernel.org. The
> drivers (kernel modules) are binary-only. They did not load because of a
> version mismatch. Asking for versions loadable with standard kernels, I got the
> response that they only support kernels from Red Hat and SuSE, but no standard
> kernels.
So send it back and consult local trading standards rules.
> My general conclusion would be that something not working with a standard
> kernel cannot be called "supporting linux", no matter what distros ever are
> supported. You may call me purist...
> Any ideas?
Linus owns the trademark for the USA certainly.
On Tue, 13 May 2003, Alan Cox wrote:
> On Maw, 2003-05-13 at 14:16, Stephan von Krawczynski wrote:
> > I bought a card from some vendor, claiming "support for Linux". I tried
> to make
> > it work in a configuration with a standard 2.4.20 kernel from kernel.org.
> The
> > drivers (kernel modules) are binary-only. They did not load because of a
> > version mismatch. Asking for versions loadable with standard kernels,
> I got the
> > response that they only support kernels from Red Hat and SuSE, but
> no standard
> > kernels.
>
If you really want it to work, try `insmod -f modulename.o`. See of it
works. RedHat supplies kernels with "intermediate" version numbers
like linux-2.4.18-24. A perfectly-good module from linux-2.4.18
may fail to load without the '-f' option, even though it is probably
compatible. Try it, it may work fine. You can modify /etc/rc.d/rc.local
to insert the module during startup so you don't have to muck with
/etc/modules.conf (and having other startup-code change it when it
"finds" new hardware.
Cheers,
Dick Johnson
Penguin : Linux version 2.4.20 on an i686 machine (797.90 BogoMips).
Why is the government concerned about the lunatic fringe? Think about it.
Richard B. Johnson wrote:
>If you really want it to work, try `insmod -f modulename.o`.[...]
>
Of course, don't do this unless you can live with the following :
1/ nobody will support you in this configuration.
2/ you have zero guarantee that your hardware will be supported in the
future (what will happen when you upgrade to a new kernel with different
interfaces and the vendor marketing folks decide that Linux support
isn't a good thing anymore ?).
Stephan von Krawczynski wrote:
> My general conclusion would be that something not working with a standard
> kernel cannot be called "supporting linux", no matter what distros ever are
> supported. You may call me purist...
If their drivers don't come with full source code then their claims
of supporting Linux are just a bad joke AFAIC.
On Tue, May 13, 2003 at 01:53:37PM +0100, Alan Cox wrote:
> On Maw, 2003-05-13 at 14:16, Stephan von Krawczynski wrote:
[snip]
>
> > My general conclusion would be that something not working with a standard
> > kernel cannot be called "supporting linux", no matter what distros ever are
> > supported. You may call me purist...
> > Any ideas?
>
> Linus owns the trademark for the USA certainly.
Any ideas who (if anyone) owns it in the UK? Is there anywhere to go to
find this out, perhaps for other countries too?
jc
Chuck Ebbert <[email protected]> writes:
> > My general conclusion would be that something not working with a standard
> > kernel cannot be called "supporting linux", no matter what distros ever are
> > supported. You may call me purist...
>
> If their drivers don't come with full source code then their claims
> of supporting Linux are just a bad joke AFAIC.
Even when they do, it's often far from what I would call "Linux
support". I've seen vendor drivers that made such assumptions about
the machine that they would only work on IA-32 machines. I'm talking
about things like assuming that sizof(int) == sizeof(void *) == 4, or
that physical memory addresses are the same seen from the CPU and from
the PCI bus.
--
M?ns Rullg?rd
[email protected]
On Tue, May 13, 2003 at 10:36:01PM +0200, M?ns Rullg?rd wrote:
> Chuck Ebbert <[email protected]> writes:
>
> > > My general conclusion would be that something not working with a standard
> > > kernel cannot be called "supporting linux", no matter what distros ever are
> > > supported. You may call me purist...
> >
> > If their drivers don't come with full source code then their claims
> > of supporting Linux are just a bad joke AFAIC.
>
> Even when they do, it's often far from what I would call "Linux
> support". I've seen vendor drivers that made such assumptions about
> the machine that they would only work on IA-32 machines. I'm talking
> about things like assuming that sizof(int) == sizeof(void *) == 4, or
> that physical memory addresses are the same seen from the CPU and from
> the PCI bus.
This is really a trademark related labelling issue. The
trademark allows Linus or his assignee to specify in what
way Linux(tm) may be used in labelling and advertising.
Linux is just like other products with third-party parts and
supplies. If Linus's assignee (Linux international?) where
to specify explicit guidelines then people would know what
to expect. Something like:
Linux certified:
Mainline kernel has driver and it has been certified
as functioning with this hardware by OSDL or some
other officially sanctioned lab.
Linux supported:
Mainline kernel has driver.
Linux compatible:
Source code driver available as a patch.
Runs on Linux:
Binary only driver available that can be used with
mainline kernel.
Supports Linux:
Portion of the purchase price will be donated to
Linux International.
You will notice this all relates to mainline kernels (Linus
and Marcello). If the product requires a vendor kernel they
need to negotiate with the vendor to say so.
These are just suggestions. Many other products (including
MS windows) have similar labelling restrictions, often with
logos. Use of the term "Linux" in packaging or advertising
or products inconsistent with the official designations would
be trademark infringement. Different rules would apply to
products that exist strictly in user-space.
--
________________________________________________________________
J.W. Schultz Pegasystems Technologies
email address: [email protected]
Remember Cernan and Schmitt
Hi.
> This is really a trademark related labelling issue. The trademark
> allows Linus or his assignee to specify in what way Linux (tm) may
> be used in labelling and advertising. Linux is just like other
> products with third-party parts and supplies. If Linus's assignee
> (Linux international?) where to specify explicit guidelines then
> people would know what to expect. Something like:
>
> Linux certified:
> The mainline kernel has a driver and it has been certified
> as functioning with this hardware by OSDL or some other
> officially sanctioned lab.
>
> Linux supported:
> The mainline kernel has a driver.
Fine so far.
> Linux compatible:
> Source code driver is available as a patch.
In other words, if a patch is available for the 1.0.0 kernel, they
can claim "Linux compatible" ??? That's meaningless...replace with
something like...
Linux 2.2.2 compatible:
Source code driver is available as a patch for the stated
mainline kernel.
...with the specific version to be made explicit. As a minimum, it
needs to state the actual kernel series the patch is for.
> Runs on Linux:
> A binary only driver is available that can be used with
> mainline kernel.
Similar comments apply. Again, require that the kernel that driver
works with is made explicit.
> Supports Linux:
> A portion of the purchase price will be donated to
> Linux International.
So a company provides a product for $5,000.00 and donates $0.01 of
the purchase price to the specified organisation, thus entitling
themselves to say "Supports Linux" by this rule. Can I suggest
this alternative definition:
Supports Linux:
At least 1% of the purchase price will be donated to
Linux International.
> You will notice this all relates to mainline kernels (Linus and
> Marcello). If the product requires a vendor kernel they need to
> negotiate with the vendor to say so.
Agreed.
> These are just suggestions. Many other products (including MS
> windows) have similar labelling restrictions, often with logos.
> Use of the term "Linux" in packaging or advertising or products
> inconsistent with the official designations would be trademark
> infringement.
Agreed.
> Different rules would apply to products that exist strictly in
> user-space.
That's not necessary - just define:
Runs under Linux:
This product work on a system based on a mainline Linux
kernel without making any modification to the kernel
itself or loading any kernel modules.
That is all that is required there.
Best wishes from Riley.
---
* Nothing as pretty as a smile, nothing as ugly as a frown.
---
Outgoing mail is certified Virus Free.
Checked by AVG anti-virus system (http://www.grisoft.com).
Version: 6.0.481 / Virus Database: 277 - Release Date: 13-May-2003
[email protected] (=?iso-8859-1?q?M=E5ns_Rullg=E5rd?=) writes:
>Even when they do, it's often far from what I would call "Linux
>support". I've seen vendor drivers that made such assumptions about
>the machine that they would only work on IA-32 machines. I'm talking
>about things like assuming that sizof(int) == sizeof(void *) == 4, or
>that physical memory addresses are the same seen from the CPU and from
>the PCI bus.
This is why "the other OS" has the WHQL, signs drivers and generally
does many things (including getting some cash from driver vendors) to
ensure (and enforce!) that the "designed for Windows xxx" logo really
helps customers.
And might be one of the reasons why they're releasing a new 'kernel'
version only once every two years and go through many pains to ensure,
that old drivers still run most of the times. This is an area where
Linux (and many of the Linux advocates) could really learn.
>From a user land perspective, only major Linux vendors or
organizations could enforce such a logo program, it would cost wads of
cash and it will really suck if you currently run the certification
process for Linux 2.5.102 for your driver and right before you're
done, 2.5.103 is released and you have to start all over again.
That's why most of the companies that _do_ provide drivers, provide
them for _one_ kernel release of some Linux distributions.
Heck, I was working with an IDS (won't tell you which one), which
shipped its security relevant kernel module _only_ for a truly well
known distribution with the stock kernel release which had remotely
exploitable holes.
While nVidia bashing is very popular on this list, this is one of the
few companies that, while distributing binary only drivers, try to
make an effort to keep their drivers reasonably up to date _and_ work
with most new kernel releases.
Regards
Henning
--
Dipl.-Inf. (Univ.) Henning P. Schmiedehausen INTERMETA GmbH
[email protected] +49 9131 50 654 0 http://www.intermeta.de/
Java, perl, Solaris, Linux, xSP Consulting, Web Services
freelance consultant -- Jakarta Turbine Development -- hero for hire
jw schultz <[email protected]> writes:
>This is really a trademark related labelling issue. The
>trademark allows Linus or his assignee to specify in what
>way Linux(tm) may be used in labelling and advertising.
>Linux is just like other products with third-party parts and
>supplies. If Linus's assignee (Linux international?) where
>to specify explicit guidelines then people would know what
>to expect. Something like:
>Linux certified:
> Mainline kernel has driver and it has been certified
> as functioning with this hardware by OSDL or some
> other officially sanctioned lab.
>Linux supported:
> Mainline kernel has driver.
>Linux compatible:
> Source code driver available as a patch.
>Runs on Linux:
> Binary only driver available that can be used with
> mainline kernel.
>Supports Linux:
> Portion of the purchase price will be donated to
> Linux International.
As there is no real body to enforce misuse of these labels, they're
moot.
And you forgot:
Linus certified:
Used by Linus Torvalds to develop new versions of Linux. :-)
Regards
Henning
--
Dipl.-Inf. (Univ.) Henning P. Schmiedehausen INTERMETA GmbH
[email protected] +49 9131 50 654 0 http://www.intermeta.de/
Java, perl, Solaris, Linux, xSP Consulting, Web Services
freelance consultant -- Jakarta Turbine Development -- hero for hire
On Wed, May 14, 2003 at 02:09:33PM +0000, Henning P. Schmiedehausen wrote:
> >From a user land perspective, only major Linux vendors or
> organizations could enforce such a logo program, it would cost wads of
> cash and it will really suck if you currently run the certification
> process for Linux 2.5.102 for your driver and right before you're
> done, 2.5.103 is released and you have to start all over again.
Certifying anything against a development series kernel is completely
pointless. Breakage outside the driver itself could have adverse
affects. Example: For the last dozen or so kernels, the i845 AGP driver
crashed on exiting X. Turned out to be a VM bug.
Dave
On Wed, 14 May 2003, Henning P. Schmiedehausen wrote:
> And you forgot:
>
> Linus certified:
> Used by Linus Torvalds to develop new versions of Linux. :-)
Penguin Holy Pee:
Come rain, sleet, snow, or stampeding penguins, this driver
_WORKS_.
Mike
Using a "2.5.xxx" number was a bad example. Consider it replaced by
"2.4.17" and "2.4.18".
Regards
Henning
On Wed, 2003-05-14 at 16:44, Dave Jones wrote:
> On Wed, May 14, 2003 at 02:09:33PM +0000, Henning P. Schmiedehausen wrote:
>
> > >From a user land perspective, only major Linux vendors or
> > organizations could enforce such a logo program, it would cost wads of
> > cash and it will really suck if you currently run the certification
> > process for Linux 2.5.102 for your driver and right before you're
> > done, 2.5.103 is released and you have to start all over again.
>
> Certifying anything against a development series kernel is completely
> pointless. Breakage outside the driver itself could have adverse
> affects. Example: For the last dozen or so kernels, the i845 AGP driver
> crashed on exiting X. Turned out to be a VM bug.
>
>
> Dave
--
Dipl.-Inf. (Univ.) Henning P. Schmiedehausen INTERMETA GmbH
[email protected] +49 9131 50 654 0 http://www.intermeta.de/
Java, perl, Solaris, Linux, xSP Consulting, Web Services
freelance consultant -- Jakarta Turbine Development -- hero for hire
On Wed, May 14, 2003 at 05:58:03PM +0200, Henning Schmiedehausen wrote:
> Using a "2.5.xxx" number was a bad example. Consider it replaced by
> "2.4.17" and "2.4.18".
There's typically several months between stable releases.
That's more than adequate for driver certification.
Dave
"Riley Williams" <[email protected]> writes:
> > This is really a trademark related labelling issue. The trademark
> > allows Linus or his assignee to specify in what way Linux (tm) may
> > be used in labelling and advertising. Linux is just like other
> > products with third-party parts and supplies. If Linus's assignee
> > (Linux international?) where to specify explicit guidelines then
> > people would know what to expect. Something like:
> >
> > Linux certified:
> > The mainline kernel has a driver and it has been certified
> > as functioning with this hardware by OSDL or some other
> > officially sanctioned lab.
> >
> > Linux supported:
> > The mainline kernel has a driver.
>
> Fine so far.
>
> > Linux compatible:
> > Source code driver is available as a patch.
>
> In other words, if a patch is available for the 1.0.0 kernel, they
> can claim "Linux compatible" ??? That's meaningless...replace with
> something like...
>
> Linux 2.2.2 compatible:
> Source code driver is available as a patch for the stated
> mainline kernel.
>
> ...with the specific version to be made explicit. As a minimum, it
> needs to state the actual kernel series the patch is for.
It should also be stated which architectures it works on. Something
like
Compatible with Linux 2.4.20 on foo hardware
might be getting closer.
--
M?ns Rullg?rd
[email protected]
> On Wed, May 14, 2003 at 02:09:33PM +0000, Henning P. Schmiedehausen wrote:
> > >From a user land perspective, only major Linux vendors or
> > organizations could enforce such a logo program, it would cost wads of
> > cash and it will really suck if you currently run the certification
> > process for Linux 2.5.102 for your driver and right before you're
> > done, 2.5.103 is released and you have to start all over again.
> Certifying anything against a development series kernel is completely
> pointless. Breakage outside the driver itself could have adverse
> affects. Example: For the last dozen or so kernels, the i845 AGP driver
> crashed on exiting X. Turned out to be a VM bug.
This is why I think it only makes sense to certify a product that either
provides a source code driver or sufficient documentation to allow someone
to write one. Even if the driver is bugfree, you still have to be able to
debug around it.
DS
On Wed, May 14, 2003 at 02:11:10PM +0000, Henning P. Schmiedehausen wrote:
> jw schultz <[email protected]> writes:
>
> >This is really a trademark related labelling issue. The
> >trademark allows Linus or his assignee to specify in what
> >way Linux(tm) may be used in labelling and advertising.
> >Linux is just like other products with third-party parts and
> >supplies. If Linus's assignee (Linux international?) where
> >to specify explicit guidelines then people would know what
> >to expect. Something like:
>
> >Linux certified:
> > Mainline kernel has driver and it has been certified
> > as functioning with this hardware by OSDL or some
> > other officially sanctioned lab.
>
> >Linux supported:
> > Mainline kernel has driver.
>
> >Linux compatible:
> > Source code driver available as a patch.
>
> >Runs on Linux:
> > Binary only driver available that can be used with
> > mainline kernel.
>
> >Supports Linux:
> > Portion of the purchase price will be donated to
> > Linux International.
>
I left out versioning because i don't see that as belonging
on the front of the box. That falls under "system
requirements" with a URL for downloading up-to-date drivers.
Hardware that has a driver in, or for, the mainline
development tree but not the stable tree should probably
not qualify or should have another another category or a
category modifier.
> As there is no real body to enforce misuse of these labels, they're
> moot.
I'd suggest Linux International. I think Linus has assigned
trademark enforcement to them.
With decent guidelines most manufacturers would be only to
glad to comply. Some might even be willing to pony up $$
for certification or the right to use some copyrighted
logo.
Those who play fast-and-loose will generally fall in line if
asked to do so. Witness the effectiveness of the EFF in
defending the GPL. The resistant would face public
criticism for false labelling (possible criminal fraud
charges?) and the potential of a trademark infringement
suit. The categories i suggest don't really leave anyone
out. Even if you make a piece of crap hardware with a
binary only driver that only supports one kernel version as
long as you hide which version in the "system requirements"
(where we all look first) you can claim "runs on Linux".
--
________________________________________________________________
J.W. Schultz Pegasystems Technologies
email address: [email protected]
Remember Cernan and Schmitt
Hi!
> > > Linux compatible:
> > > Source code driver is available as a patch.
> >
> > In other words, if a patch is available for the 1.0.0 kernel, they
> > can claim "Linux compatible" ??? That's meaningless...replace with
> > something like...
Yep, I guess patch for 1.0.0 kernel would count as Linux compatible. I
even believe that patch for FreeBSD would count as Linux
compatible. As long as you can get source of the driver, it is easy.
Maybe better sticker would be
"includes driver sources".
[Long time ago I got double-speed cdrom, and on its driver disk, there
was C source of the driver. Porting it to linux was matter of two
days, IIRC. And it had korean comments ;-)].
Pavel
--
When do you have a heart between your knees?
[Johanka's followup: and *two* hearts?]
On Sat, May 17, 2003 at 06:05:09PM +0200, Pavel Machek wrote:
> Hi!
>
> > > > Linux compatible:
> > > > Source code driver is available as a patch.
> > >
> > > In other words, if a patch is available for the 1.0.0 kernel, they
> > > can claim "Linux compatible" ??? That's meaningless...replace with
> > > something like...
>
> Yep, I guess patch for 1.0.0 kernel would count as Linux compatible. I
If the box has been on the shelf for 8 years it would. If
the driver isn't in mainline i'm not very interested in what
drivers are in the box. I want a URL where i can go and
look to see the versions for download.
> even believe that patch for FreeBSD would count as Linux
> compatible. As long as you can get source of the driver, it is easy.
There is a big difference between "some assembly required"
and "some machining required". Driver source that has to be
modified doesn't count. This isn't labelling for kernel
hackers. Some people need it to just work, others are OK
following a script to patch and rebuild or setup an insmod.
What we are talking about is enough info that someone should
feel comfortable buying the box with the expectation of
support at the level they require.
> Maybe better sticker would be
>
> "includes driver sources".
>
> [Long time ago I got double-speed cdrom, and on its driver disk, there
> was C source of the driver. Porting it to linux was matter of two
> days, IIRC. And it had korean comments ;-)].
Perhaps you'd be willing to help Aunt Tillie do this. Your
definition of "includes driver sources" will appeal to such a
small market segment it isn't worth the ink on the box.
--
________________________________________________________________
J.W. Schultz Pegasystems Technologies
email address: [email protected]
Remember Cernan and Schmitt
jw schultz wrote:
> Perhaps you'd be willing to help Aunt Tillie do this. Your
> definition of "includes driver sources" will appeal to such a
> small market segment it isn't worth the ink on the box.
I think Pavel's idea is brilliant. "Includes driver sources"
captures exactly the essence of dependable support for Linux.
Aunt Tillie will eventually learn that this strange sentence means
that, even if the thing doesn't work when she attaches it to her
PC, her nephew will be able to make it work for her.
And in a few years, when she buys a new PC with a new distribution,
there will probably already be a driver included.
What else could she ask for ?
- Werner
--
_________________________________________________________________________
/ Werner Almesberger, Buenos Aires, Argentina [email protected] /
/_http://www.almesberger.net/____________________________________________/
Hi!
> > Maybe better sticker would be
> >
> > "includes driver sources".
> >
> > [Long time ago I got double-speed cdrom, and on its driver disk, there
> > was C source of the driver. Porting it to linux was matter of two
> > days, IIRC. And it had korean comments ;-)].
>
> Perhaps you'd be willing to help Aunt Tillie do this. Your
> definition of "includes driver sources" will appeal to such a
> small market segment it isn't worth the ink on the box.
Sure, Aunt Tillie can pay SuSE.cz to have that driver updated to any
kernel she wants ;-).
[Really, if the card is common enough, and driver sources are
available, they are going to be in kernel.org kernel within few weeks.]
Pavel
--
When do you have a heart between your knees?
[Johanka's followup: and *two* hearts?]
On Sun, 18 May 2003, Pavel Machek wrote:
> > > [Long time ago I got double-speed cdrom, and on its driver disk, there
> > > was C source of the driver. Porting it to linux was matter of two
> > > days, IIRC. And it had korean comments ;-)].
> >
> > Perhaps you'd be willing to help Aunt Tillie do this. Your
> > definition of "includes driver sources" will appeal to such a
> > small market segment it isn't worth the ink on the box.
>
> Sure, Aunt Tillie can pay SuSE.cz to have that driver updated to any
> kernel she wants ;-).
>
> [Really, if the card is common enough, and driver sources are
> available, they are going to be in kernel.org kernel within few weeks.]
Iff:
(1) the sources etc constitute "sufficient" documentation (e.g. poke magic
value into magic register might not help enough :-( ).
(2) the sources are not only "included" but are "freely" modifiable and
redistributable.
Regards,
Neale.
Neale Banks wrote:
> On Sun, 18 May 2003, Pavel Machek wrote:
>>[Really, if the card is common enough, and driver sources are
>>available, they are going to be in kernel.org kernel within few weeks.]
>
> Iff:
> (1) the sources etc constitute "sufficient" documentation (e.g. poke magic
> value into magic register might not help enough :-( ).
Sure.
> (2) the sources are not only "included" but are "freely" modifiable and
> redistributable.
No need. As long as there is no NDA, which there isn't when
you get a cd bundled with the thing. Restricted source then
mean that you can't edit it so it fits your kernel and
redistribute. But you can read it and see how the thing works,
if (1) is satisfied. Then you use your knowledge and write
a driver from scratch. More work, but legal.
Of course a completely free driver is even better. :-)
Helge Hafting