The topic of binary modules has been raised many times recently (I'm
quite new here :-) in respect of the ndiswrapper and 4k stacks,
discussion of 3D graphics drivers and most recently Arjan's "Doomsday
Scenario" thread. Whilst reading the Doomsday thread it became
apparent to me, and to several other people (please see my apology at
the end of this mail), that what is required is a Logo program
similar to the "Designed for Windows XP" logo.
The primary motivation for this is that it leverages the individual
power of each purchaser (of a system or individual piece of hardware)
be they a consumer, SME, system builder, tier 1 or 2 PC manufacturer,
government dept., or Linux distro company, into a single point of
pressure that can be applied to OEMs to ensure that they provide open
source drivers.
Unlike previous lists of equipment that works with Linux, this would
of course have to be centrally administered.
Simply, hardware manufacturers would be able to use the LHQL logo
(whatever it looks like or is called) once their kit had been
certified, the primary requirement for which would be either that a
fully featured open source driver was available or that all the
relevant documentation had been put in to the public domain so that
anybody could write one.
Why will this work? Here is how I see it developing in the server
market:
1. The logo will be launched along with a campaign to explain to
technical users that (a) open source drivers are better quality, (b)
that binary drivers threaten the future of the Linux ecosystem.
2. Any company which manufactures equipment with open source drivers
already available will have the logo granted for their current
equipment.
This means that right from the outset it will be possible to
put together a system which is wholly conformant, i.e. the entire PC
is eligible to carry the logo.
3. Certain small system builders will get on board.
4. Obviously, everyone on LKML will only buy/recommend hardware which
carries "the mark of the penguin". As will geeks the world over.
5. One of the distros will decide that their "Supported Hardware"
list will only feature items with the logo.
6. Medium sized system builders, maybe "partners" of the distro will
begin to ship systems that fully conform.
7. All the supporters above will be using the the new Open Source
Graphics hardware from http://www.opengraphics.org. (Important later)
8. Corporate or government buyers will by now become aware of the
value of buying certified equipment.
Up until this point the OEMs that we wish to influence have not taken
much notice.
However, once stage 8 is reached, having the logo will be easily
understood to be a competitive advantage.
The inflection point has then been reached. Each of the producer
groups i.e. OEMs, PC vendors and distros then increasing use the logo.
Who Pays?
1. _Not_ the OEM. Producing the open source driver is their
contribution.
2. PC vendors who build system from multiple conformant parts who
then wish to badge the entire system can pay an amount.
3. OSDL or a similar body would (or a wealthy benefactor) would be
required to provide initial funding.
Who Certifies?
1. This was discussed in the other threads, the feeling seemed to be
that it should be kept close to the LKML/maintainers and should not
be controlled by a body which could be subjected to financial pressure.
What is required for Certification? (I am not qualified to answer
this question)
1. The driver must be GPLed and fully support the features of the
hardware (or match the support given to Windows Server/Client)
2. Any other software quality / security tests that you think is
appropriate.
Desktop
======
So far I have been talking only of the server.
My assumptions when discussing the Desktop segment are as follows:
(1) Early comments in the "Doomsday Scenario" thread were indicating
that the state of drivers for consumer level machines was fairly
poor. But also that despite much hype the Linux Desktop really isn't
here yet. The 2 things are obviously somewhat connected, as discussed
at the recent OSDL desktop summit.
(2) Also, the major point of pain on the desktop seems to concern the
graphics chipsets.
(3) Cutting edge graphics cards in PCs are mainly required for
gamers. I read today that the PC games market is going to be dead in
5 years :-).
Since the market for consumer machines has not yet reached a level at
which OEMs feel they need to seriously address it, this obviously
provides some time for the Logo program to become established and
successful in the server market. During this period the level of
graphics support available from OEMs other than ATI & NVidia will
improve.
For example:
i) the OpenGraphics project will hopefully have flourished,
partially due to the Logo program.
ii) A number of smaller graphics OEMs will supply Open Source kit.
iii) Intel will provide fully conformant on-board graphics
solutions.
Therefore a consumer wishing to purchase a modern system with a
reasonably powerful graphics engine (i.e. more powerful than anything
available today) will have several options, even if the tier 1
graphics vendors are not yet providing Open Source drivers.
This will of course place them at a competitive disadvantage that
they will wish to avoid. So they apply for the logo. :-)
Thanks for listening.
regards,
Felix
Apologies
1. Apologies to anybody who thought of this months/years ago.
2. Apologies to everyone in the "Doomsday Scenario" thread who
suggested this. I am not crediting you by name simply because I do
not want to re-read all 100 entries to find your names.
3. Apologies to Benjamin LaHaise who has already started a thread
called "Runs with Linux (TM)" based on his prior independent
conclusion that this is the correct course of action.
4. Apologies to everyone I have CC'ed based on the particular input
they made on the "Doomsday Scenario" thread.
5. To anybody I upset, I am only posting this because I really do
think that it is The Answer (tm) and I want many people to consider it.
Felix Oxley wrote:
> Unlike previous lists of equipment that works with Linux, this would of
> course have to be centrally administered.
> Simply, hardware manufacturers would be able to use the LHQL logo
> (whatever it looks like or is called) once their kit had been certified,
> the primary requirement for which would be either that a fully featured
> open source driver was available or that all the relevant documentation
> had been put in to the public domain so that anybody could write one.
I think the current focus on the logo idea is pretty sad. The only real
"feature" a logo adds is it will have to be maintained by Someone Else.
Which is always nice. Except when the real info is in the kernel
maintainers heads, as is the real love of free software, being
maintained by Someone Else means it'll fail.
A good exhaustive online centralised hardware database, blessed and
maintained by kernel people, will have influence with or without a logo.
A pretty logo/HQL program without kernel people involvement will fail
quickly.
The problem with maintaining your own database is its a lot of work (as
Lee Revell pointed). However, if kernel people don't want the system to
be abused, distorted, etc they'll have to make this effort.
Kernel people complain users don't buy the right hardware, but they
don't bother pointing out which hardware is good. I'm pretty sure when
*they* buy new systems they don't do the long hours of googling and ML
analysis they'd want the users to perform (because right now that's what
needed to find out which hardware is linux-friendly)
Also, kernel people use the power of the internet all day round but put
their faith in a paper logo. Which is pretty ridiculous. Give users a
good database and they won't need the paper thingy (not to mention
drivers are completed after hardware ships, so all the already-packaged
hardware won't get a linux logo by magic)
Regards,
--
Nicolas Mailhot
On Thu, 8 Dec 2005, Nicolas Mailhot wrote:
> Felix Oxley wrote:
> A good exhaustive online centralised hardware database, blessed and
> maintained by kernel people, will have influence with or without a logo.
I don't think this is quite right.
The beauty of having a simple and easily-recognizable logo plastered on
a website or product box or what have you is exactly that -- simplicity.
That's a good thing, because it means that the person looking to
purchase some bit of hardware can merely look for that one sign, and
it'll be "guaranteed" to work.
Simplicity is a very good thing, in this case -- the easier it is for a
consumer to check/notice/comprehend something, the more likely they are
to use it and put value into that process. This leads to greater
mindshare for that logo. And besides, there are some cases where an
online database does not help you much: for example, if you find
yourself shopping in a brick-and-mortar store.
An online database is definitely a very useful thing to have, for those
who know to look for it, and who can look at it. But aside from that, a
simple logo is what the vast majority of people wishing to purchase
hardware would benefit from the most.
> Regards,
>
> --
> Nicolas Mailhot
> -
-Vadim Lobanov
Vadim Lobanov <vlobanov <at> speakeasy.net> writes:
>
> On Thu, 8 Dec 2005, Nicolas Mailhot wrote:
>
> > Felix Oxley wrote:
> > A good exhaustive online centralised hardware database, blessed and
> > maintained by kernel people, will have influence with or without a logo.
>
> I don't think this is quite right.
>
> The beauty of having a simple and easily-recognizable logo plastered on
> a website or product box or what have you is exactly that -- simplicity.
> That's a good thing, because it means that the person looking to
> purchase some bit of hardware can merely look for that one sign, and
> it'll be "guaranteed" to work.
>
> Simplicity is a very good thing, in this case -- the easier it is for a
> consumer to check/notice/comprehend something, the more likely they are
> to use it and put value into that process. This leads to greater
> mindshare for that logo. And besides, there are some cases where an
> online database does not help you much: for example, if you find
> yourself shopping in a brick-and-mortar store.
>
> An online database is definitely a very useful thing to have, for those
> who know to look for it, and who can look at it. But aside from that, a
> simple logo is what the vast majority of people wishing to purchase
> hardware would benefit from the most.
>
Sorry for posting this in the other thread, so heres my list again, (i'm pretty
much in favour of felix' scenario):
- get all open operating system folks to join
- have an approval organisation, that everyone is happy with
- lets call it "free driver" support
- create apropriate Logos for each operating system i.e. "Penguin-Logo"
with "free driver since kernel-a.b.c.d" or "BSD-Daemon-Logo" with "free driver
since a.b"
- the hardware version numer (usb-device id, pci-id)? will be held in a
database.
If in doubt people can look it up there. In case future kernels will
drop support, it can be marked there as well.
A yearly logo is too much
confusion, since hardware support for open drivers stays pretty long in
the kernel. (Imagine you license in december - you only got a month...)
Maybe educate people that support for kernel 2.6.x.y series means a
penguin logo on blue ground.
If dramatic changes to the kernel are introduced, which all
drivers affect call the kernel 2.8.x.y an educate people about a new linux
in town and create a penguin logo on yellow ground
- hardware vendors pay someone, or provide source code themselves
to be reviewed by apropriate kernel folks/bsd board...
- small licence fee and endless advertising possiblilities pay for
organisation folks and hosting.
Keeping up database records could help kernel developers:
There could be a log file attached to each device stating the
current affairs about support in varying OSs
Felix had a good senario for financing this, it should be an independend body
providing legal/licensing and marketing support. And the kernel folks should do
the technical review.
dirk
On Thursday 08 December 2005 18:23, Dirk Steuwer wrote:
> - hardware vendors pay someone, or provide source code themselves
> to be reviewed by apropriate kernel folks/bsd board...
The first part probably won't work too well. The moment money is involved,
financial pressure starts to come into play, and said person might be
inclined to "overlook" things so to speak. We want the second one in
reality. If it's supposed to work with linux, it better damn well be
reviewed by the people that made it.
Chris White
On 8 Dec 2005, at 07:43, Nicolas Mailhot wrote:
>
> I think the current focus on the logo idea is pretty sad. The only
> real
> "feature" a logo adds is it will have to be maintained by Someone
> Else.
> Which is always nice. Except when the real info is in the kernel
> maintainers heads, as is the real love of free software, being
> maintained by Someone Else means it'll fail.
>
> A good exhaustive online centralised hardware database, blessed and
> maintained by kernel people, will have influence with or without a
> logo.
>
The benefits of a 'paper' logo are as follows:
1. It builds 'brand' awareness
2. It means that when you take a piece of hardware of the shelf in
the store you don't need to check an online database for compatibility.
> their faith in a paper logo. Which is pretty ridiculous. Give users a
> good database and they won't need the paper thingy (not to mention
> drivers are completed after hardware ships, so all the already-
> packaged
> hardware won't get a linux logo by magic)
I presume that drivers will be developed alongside the hardware
because you can't sell the kit until the drivers are on the CD in the
nice box with the instruction manual.
Also you can't test the hardware properly unless you have drivers for
it.
regards,
Felix
> I presume that drivers will be developed alongside the hardware
> because you can't sell the kit until the drivers are on the CD in the
> nice box with the instruction manual.
It wil be real "plug and play". The driver is not in the hardwarebox, its in
the linux/bsd/whatever kernel.
Ideally, people with the right kernel stick the hardware in and it just works.
No driver installation involved. This is an important aspect, that currently
nobody out there links to linux. If you ask someone why they not use linux
its hardware problems - mostly on desktop/laptop. But any free operation system
with free drivers should be known for an "instant run" of supported hardware.
No more driver installation worries is a fantastic side effect of free drivers.
> Also you can't test the hardware properly unless you have drivers for
> it.
>
The hardware vendor has to provide
a) documentation of the interface
b) samples to maybe three kernel developers for review
if three developers certify that it works, the overlooking Org. can release
the logo to be included on the box.
If the vendor is not cooperating alongside hardware development, the inital box
with css drivers will go into market, and only later a linux sticker is added.
On Thu, Dec 08, 2005 at 10:26:53AM +0000, Felix Oxley wrote:
> I presume that drivers will be developed alongside the hardware
> because you can't sell the kit until the drivers are on the CD in the
> nice box with the instruction manual.
> Also you can't test the hardware properly unless you have drivers for
> it.
No. Any external reviewer will be immediately put in a nasty position
of standing in the way of vendor's schedule. That's _the_ recipe for
mess. And that's besides the joy of NDA, etc.
Chris White <chriswhite <at> gentoo.org> writes:
>
> On Thursday 08 December 2005 18:23, Dirk Steuwer wrote:
> > - hardware vendors pay someone, or provide source code themselves
> > to be reviewed by apropriate kernel folks/bsd board...
>
> The first part probably won't work too well. The moment money is involved,
> financial pressure starts to come into play, and said person might be
> inclined to "overlook" things so to speak. We want the second one in
> reality. If it's supposed to work with linux, it better damn well be
> reviewed by the people that made it.
>
> Chris White
>
Your are probably right.
1)Lets say there is a small licence fee.
50% to the developer that writes/intergrates the code into the kernel,
50% to the Org that provides the database/legal backup...
or to get things going
2)no fees, just providing a certain level of docs
such that a driver with full hardware support can be created.
This enables the company to add the "Linux" sticker to their boxes.
On 8 Dec 2005, at 10:46, Dirk Steuwer wrote:
>
>> I presume that drivers will be developed alongside the hardware
>> because you can't sell the kit until the drivers are on the CD in the
>> nice box with the instruction manual.
>
> It wil be real "plug and play". The driver is not in the
> hardwarebox, its in
> the linux/bsd/whatever kernel.
> Ideally, people with the right kernel stick the hardware in and it
> just works.
> No driver installation involved. This is an important aspect, that
> currently
> nobody out there links to linux. If you ask someone why they not
> use linux
> its hardware problems - mostly on desktop/laptop. But any free
> operation system
> with free drivers should be known for an "instant run" of supported
> hardware.
> No more driver installation worries is a fantastic side effect of
> free drivers.
>
>
>> Also you can't test the hardware properly unless you have drivers for
>> it.
>>
>
> The hardware vendor has to provide
> a) documentation of the interface
> b) samples to maybe three kernel developers for review
>
> if three developers certify that it works, the overlooking Org. can
> release
> the logo to be included on the box.
> If the vendor is not cooperating alongside hardware development,
> the inital box
> with css drivers will go into market, and only later a linux
> sticker is added.
>
The documentation aspect could solve this chicken and egg problem.
If the OEM has provided documentation and initail versions of the
driver under GPL (and we assume that they intend to complete the driver)
then we can give them the logo before the driver is 100% perfect in
order that they can print their product boxes etc.
However, an OEM with a certified device which has become broken over
time (or they didn't bother to finalise the driver after they had
been granted the logo) will not be able to get any further hardware
certified. They will have to fix the existing stuff first.
regards,
Felix
How about starting the whole thing with these steps:
short term:
A central place where kernel hackers document the hardware support for a given
device would be a good starting point.
Also they can start to document how well a company provided information to
create a free driver.
This would be completely independend from financial aspects.
It should be with some structure, as opposed to a pure wiki approach, which was
suggested somewhere in the original thread.
Kernel folks are allowed to enter devices/change Status/log ongoing development
steps
All others are allowed to comment as testers/reviewers
see where it goes.
long term:
establish and define the conditions under which a free driver logo is used
i.e. award companies/hardware for excellent linux support with
"penguin driver logo"
or grant use of logo "linux ready" for use on boxed retail.
Dirk
On 8 Dec 2005, at 15:39, Benjamin LaHaise wrote:
> No worries, I'm just happy to see people being interested in putting
> the program together. Where abouts are you located? We'll likely
> need to set up legal entities in the .us and elsewhere for the
> administration of the program. I'm happy to do the canadian side of
> things, and will be talking to a lawyer shortly on what is required.
I'm in the UK.
But my lawyer is busy getting rid of a wife for me so I am not going
to volunteer.
:-)
regards,
Felix
>
> 1)Lets say there is a small licence fee.
> 50% to the developer that writes/intergrates the code into the kernel,
> 50% to the Org that provides the database/legal backup...
>
> or to get things going
> 2)no fees, just providing a certain level of docs
> such that a driver with full hardware support can be created.
> This enables the company to add the "Linux" sticker to their boxes.
If the OEM wants to stick the logo on the box when the hardware ships
then _they_ will have to write the driver (they can of course
subcontract to a maintainer if they like).
If they choose to rely on the system where the driver is written by
somebody who wants to use the hardware themselves then the OEM still
has to provide the necessary documentation.
And of course they would not be able to use the logo until such time
as the driver was available.
Documentation must be available for on going maintenance as well as
initial coding. This would be a requirement for the logo.
Therefore by far the best solution is that the documentation is put
into the public domain.
If the OEM has a problem with releasing the documentation then they
would have to enter into some sort of ongoing agreement with the
certification body to give confidence that the driver was
maintainable (i.e. ongoing NDA agreement). Without this the hardware
could not be certified.
I don't know how many drivers are currently provided by the OEM
rather than written by A. Random Developer, but this process should
encourage the OEM to provide the driver themselves.
This should reduce the kernel exposure to patent attacks (??). (IANAL)
regards,
Felix