Discussion:
[Madwifi-devel] compat-drivers and build test server
Luis R. Rodriguez
2012-07-26 18:54:01 UTC
Permalink
The project you may know as "compat-wireless" [0] will soon be renamed
to "compat-drivers" given the growth in support for different
subsystems and general interest by a few Linux distributions to see
this work become more decentralized from the wireless subsystem. After
my presentation about the project at the last San Francisco Linux
Collaboration summit [1] I was asked if the project could be moved
onto some vendor / subsystem neutral place. I decided to work with the
Linux Foundation and see what I could do about it. As it turns out the
Linux Foundation already had a working group working on similar
project but such efforts received little traction and it seems the
efforts we have completed through the compat / compat-wireless project
actually ended up accomplishing some of the more difficult aspects of
the project envisioned by the LF backport working group.

MadWifi served as an example framework which provided a full 802.11
stack and one driver backported to a few kernel releases. The
compat-wireless driver started supporting all 802.11 Linux upstream
drivers backported in similar way but with the new 802.11 framework:
cfg80211, mac80211. The other subsystems contributed later were
support for Bluetooth and also Ethernet. Ozan is a Google Summer of
Code student and is working with great progress on adding Video
support. As it is now the package provides backport support drivers
across a range of 21 kernels. Each commit is test compiled prior to
integration. We have daily snapshots based on linux-next.git and then
stable releases based on linux-stable.git and Linus' RC releases /
final releases. Each test compile is currently taking me about 120
minutes -- that is after every patch committed (obviously sometimes I
do not test compile between a series of patches and then I'd bisect,
it depends). At times, I have to test compile 3 different releases in
a day, for a total time spent just on compiling to 6 hours. I have a
pretty decent system but its time to grow the project up and get a bit
more serious about compilation. I've reached out to a few
organizations for assistance but would like to ensure that if I get a
build server I'd like to ensure that other contributors to the project
get similar access to the system to be able to do their own test
compiles as well. I looked into distributed compiling with Alex
Mendoza but after a few promising results in the end it seems big iron
wins in terms of practicality and setup. At this point I am
coordinating with possibilities in working with the Linux Foundation
and John Hawley, Chief kernel.org Systems Administrator, on options.
After a few tests and reviews of technologies it seems using a big
iron box hosted and maintained exclusively by kernel.org folks would
be the best option. The estimated server and cost is:

HP DL165G7 (last generation though still on sale)
2 x AMD 6262HE ( 16-core cpu )
96G ram (though you can take the system to 384M, for another 8K)
3 x 300G 10K rpm sas drives
5yr warranty.
$7,807 before any discounts, taxes, etc.

There is good progress in this direction but as could be expected this
can take time. Pavel mentioned to e-mail this distribution list given
that the compat-driver backport project's mission statement parallels
the best interests of getting users access to Atheros 802.11 hardware
device drivers as easily / fast as possible. If the project can help
with this -- it'd be great. Even of the Linux Foundation can secure
somehow the build server alone for the project I do wonder if
MadwWifi's own funding can be further considered for supporting
similar upstream endeavors, perhaps now is a good time to review. My
preference would be that if its decided supporting general backporting
effort is useful -- for the money to be used specifically for this
project alone and for the server to be administered by kernel.org
folks. Please me know what you think.

[0] http://wireless.kernel.org/en/users/Download
[1] https://docs.google.com/presentation/d/1axVNEGwKZjnzG1ocdd289WMqPxzJ3qfMv70ghGcnUKc/edit

Luis
Pavel Roskin
2012-07-27 20:48:05 UTC
Permalink
On Thu, 26 Jul 2012 11:54:01 -0700
Post by Luis R. Rodriguez
There is good progress in this direction but as could be expected this
can take time. Pavel mentioned to e-mail this distribution list given
that the compat-driver backport project's mission statement parallels
the best interests of getting users access to Atheros 802.11 hardware
device drivers as easily / fast as possible. If the project can help
with this -- it'd be great. Even of the Linux Foundation can secure
somehow the build server alone for the project I do wonder if
MadwWifi's own funding can be further considered for supporting
similar upstream endeavors, perhaps now is a good time to review. My
preference would be that if its decided supporting general backporting
effort is useful -- for the money to be used specifically for this
project alone and for the server to be administered by kernel.org
folks. Please me know what you think.
Yes, it was my suggestion. I believe the MadWifi project should not
sit on the donations indefinitely. It would be unfair to the donors.
I think it would be better to give the remaining funds to the project
that essentially continues what MadWifi was aiming to do, namely an
Atheros wireless driver for a wide range of the kernels.

I believe the build server should also compile MadWifi against those
kernels. That should not take much CPU time.

Some part of the MadWifi funds could be spent on extending the SSL
certificates for the related sites (including ath5k and ath9k) if
having free certificates is not an option. The rest could be paid for
the build server.

I'm not associated with the build server or with the lab where it would
be installed.
--
Regards,
Pavel Roskin
Loading...