Re: Linux drivers management

[Jesper Juhl <jesper.juhl@gmail.com>]

On 2/6/06, David Chow <davidchow@shaolinmicro.com> wrote:
>
> > Please read Documentation/stable_api_nonsense.txt in your copy of
> > Linux source.
> I've read the document, I strongly disagree, because it is not relavant
> to my question or to my original purpose of this question.
>
It has a lot of relevance.
If you split up drivers and the core kernel into two sepperate
projects I can easily imagine the two getting out of sync whenever an
API change needs to be made.
Currently whenever someone changes a kernel API that person is
required to also update all users of that API. Since all users are
in-tree that's resonably easy to do. If the drivers are out-of-tree
it's a *lot* harder for several reasons;

a) to fix users of the API I can no longer just 'cd drivers/' I now
have to go download & extract a sepperate source tree.

b) if an API change I make gets accepted by the 'core kernel' team but
the 'driver-tree' maintainers refuse my updates we get out-of-sync -
just as if the 'core kernel' tree rejects the API change but the
'driver-tree' maintainers already merged the update to the drivers to
use the new API.

c) presumably the two projects 'driver-tree' and 'core-kernel' would
use sepperate mailing lists, making it harder to communicate about API
changes.

Maintaining the drivers out of tree would make API changes a lot more
painful and thus we'd need to lock down the API a lot harder than we
do today. This in turn would mean that some types of bugs will be
harder to solve (since a proper fix would involve a painful API
change) and advancement of kernel caabilities/features etc will be
slower.


> Putting the driver source code in the kernel source tree has nothing to
> do with talking about a stable kernel API . Even you put the driver
> sources into the main kernel tree, it will still need a lot of work to
> port all drivers if the API changes.

You can't dodge the work that needs to be done when API's change, but
you /can/ try to make the work as easy to do as possible by keeping
the core kernel code and the drivers close together and maintained as
a single unit.


> Driver sources can still host in a
> different project (e.g. projects in sf.net) and maintain open-source and
> om by the community, no difference than before
>
> For different compile time options that affect data structures, this is
> well known a bad idea . These types of techniques no longer allowed in
> Java and other OO languages .

The kernel is written in C. What other languages do/allow/recommend
etc is irrelevant.


>Because I can simply say the code is not
> portable. If really need a recompile and optimize, the distro vendor
> should bare this, but according to the document, "As Linux supports a
> larger number of different devices "out of the box" than any other
> operating system" , do you think Linux should one day or some day grow
> to 1TB source tree to include all possible drivers for all hw come from
> the world?

First of all, drivers are regularly dropped from the tree; either
because they become unmaintained and bitrot or because the hardware
becomes extremely obsolete and rare.
Secondly, I predict that available storage sizes and bandwith
available to users of the kernel will grow faster than the size of the
source tree (and cost of storage & bandwith will likely continue to
drop as well).
Thirdly, the day Linux supports "all hw come from the world" I'll be
dancing with joy.


> I don't see there is reason why a kernel or OS need to
> include all the drivers for all the hardware.

This is *extremely* convenient for our users.


> I don't think there is any
> OS vendors on the market to capable to distribute all drivers integrity,
> then the choice is to make a disabled Linux OS because of an OSV who has
> only limited supporting resources to suppport and certify limited
> hardware devices.
>
The user always has the option of building any driver they need
themselves. This is easy since the drivers are all in the main kernel
source tree, the user doesn't have to go hunt for them online
(assuming the user even has an internet connection).


> Please see my other email responded to Jes about the learning curve and
> documentation issues of a Linux driver developer to pick up Linux skills.
>
The fact that all documentation (well, at least a lot of it) is kept
in a central place (Documentation/) is a very nice thing for someone
trying to learn their way around the kernel - I know I've personally
bennefitted from that.
Also the fact that core kernel code, drivers and supporting scripts
are all kept in a single source tree is very convenient for new
developers since there's only one thing to download and then you have
a complete tree with the full picture to learn from - again I say this
based on personal experience.


But in any case, if you want to maintain one or more drivers
out-of-tree, then go ahead, noone's stopping you. But your maintenance
work will be a lot lower if you instead submit your drivers for
inclusion in mainline since then other people will keep your driver
up-to-date and in sync with API changes when they happen. You might
even get a few bugs fixed for free and get the code cleaned up for
free.


--
Jesper Juhl <jesper.juhl@gmail.com>
Don't top-post  http://www.catb.org/~esr/jargon/html/T/top-post.html
Plain text mails only, please      http://www.expita.com/nomime.html