GLDv3 isn't ready for "stabilization", I think. We need to get the
following projects integrated, and I believe this should happen *before*
we publicize the GLDv3 interfaces:
* Brussels (changes properties/ndd, and possibly also the MAC_STAT
interface which is absolutely horrid right now for any hardware that
doesn't maintain the stats in hardware in super-cheaply read registers.)
* Clearview. This is going to change for VLANs, and probably a
bunch of other smaller stuff, including possibly stats.
* Crossbow. This is going to change the mac_resources API
undoubtedly. Possibly other bits as well.
Given this, I'd prefer to see 3rd parties pursue one of two courses of
action (or both courses!) until we have a public GLDv3:
1) Integration into Nevada. This is an awesome way to use GLDv3,
and also gets a wider code base/experience for GLDv3 engineering. Plus,
after integration, it is up to project teams to make sure that they
don't break stuff.
2) GLDv2. This is the only way to produce a driver for S10 and
earlier (short of doing the NSN thing and implementing the whole DLPI
layer yourself), and that isn't likely to change anytime soon, if ever.
To third party IHVs that want to integrate into Nevada with a GLDv3
hardware, let me express the following open invitation.
If you will:
a) do the engineering to make the driver work under OpenSolaris and
GLDv3
b) address any serious bugs (P3 or higher, at my judgment) found
during testing
c) release your code under a CDDL license using the SCA
d) make hardware (2 NICs please!) available to me (Sun) for testing
Then, I promise to:
a) sponsor them for putback (or will find someone else to do so)
into Nevada
b) making sure that they are passing the rigorous NICDRV test suites
c) perform other important tests, including suspend/resume and link
status notification
d) help them through any of the various alphabet-soup committees
(PSARC, ONSC, etc.)
e) provide engineering guidance to improve the driver and make best
of Solaris features
f) assuming the above are done successfully, ultimately commit their
code into OpenSolaris
I am happy to report that Tehuti has already responded to this
invitation, as I have a pair of Tehuti 10GbE CX4 NICs in my office now.
I hope to start testing their stuff next week.
-- Garrett
Sebastien Roy wrote:
> Andrew Gallatin wrote:
>
>> Sebastien Roy writes:
>> > Cyril Plisko wrote:
>> > > I noticed that uts/common/sys/mac.h file is not delivered
>> > > by any package into final snv image. Does that mean
>> > > that 3rd party has no means to build Nemo-based
>> > > ethernet driver outside of the ON consolidation ?
>> > > If so is it on purpose ?
>> >
>> > This is the case, and it is on purpose. The Nemo driver interfaces are
>> > consolidation-private for now, and will remain so until the interfaces
>> > stabilize and are formally documented.
>>
>> It would be nice if you could at least ship the header files which
>> match the running kernel so that 3rd parties can distribute source
>> based drivers.
>>
>
> You know this already, but hey, this is a mailing list and there's more
> than one person reading: One of the defining attributes of a
> Consolidation-Private interface is that changes to the interface can be
> managed within that software consolidation without affecting untold
> numbers of unknown consumers outside of the consolidation. Use of these
> interface by 3rd parties in this way would have a high likelihood of
> resulting in broken software as experienced by end-users.
>
> -Seb
> _______________________________________________
> networking-discuss mailing list
> [email protected]
>
_______________________________________________
networking-discuss mailing list
[email protected]
