10/stable virtualbox-ose crashes

[Russell L. Carter]

Hi,

On 05/29/15 10:52, Kimmo Paasiala wrote:
> On Fri, May 29, 2015 at 8:45 PM, Adam McDougall <mcdouga9 at egr.msu.edu> wrote:
>> On 05/29/2015 13:38, Kevin Oberman wrote:
>>> On Fri, May 29, 2015 at 10:10 AM, Russell L. Carter <rcarter at pinyon.org>
>>> wrote:
>>>
>>>> Hi,
>>>> kldload vboxsrv crashes recent 10/stables.  Last known working
>>>> kernel/module pair is from May 5th.
>>>>
>>>> Not sure what is the optimal next step, suggestions welcome.
>>>>
>>>
>>> Sounds like you have done this, but several reports have been made of this
>>> recently.  All were "fixed" by rebuilding virtualbox-ose-kmod. Always
>>> rebuild all kernel modules that are in ports when rebuilding the kernel,
>>> preferably by adding the appropriate PORTS_MODULES to /etc/src.conf.
>>> --
>>> Kevin Oberman, Network Engineer, Retired
>>> E-mail: rkoberman at gmail.com
>>
>> Won't this will be a problem if people use official packages (built on
>> 10.x-RELEASE) on a system running -stable?  Wouldn't it mean an ABI was
>> violated?
>
>
> The KBI (kernel binary interface that the kernel modules use) is not
> part of the promised stability, only the userland ABI is guaranteed to
> stay compatible.

According to:

http://blog.shatow.net/posts/2015-04-27-Poudriere-FreeBSD-Journal/

    "Packages are built for the oldest release of each branch. These
    packages are supposed to be ABI/KBI compatible with all future
    releases on those branches as well as the STABLE branch for that
    release. This means that packages built for 8.3 will work on 8.4
    but are not guaranteed to work on 9.x."

After sleeping on the problem, I think that the simplest approach to
managing the evolution of the KBI in stable is to mark packages that
depend on the KBI, port kernel modules for instance, as dependent on
poudriere's world source tree.  Then when the source tree is updated,
on the first subsequent bulk package build port kernel modules would
be correctly updated.  This would eliminate the maintenance complexity
I outlined in previous messages that arises from maintaining port
kernel modules outside of poudriere.

I just verified that updating poudriere's source tree does not
currently cause any port kernel modules to be rebuilt.

Comments?

Thanks,
Russell