[vbox-dev] Kernel API changes for kernel 4.10

[Larry Finger]

On 01/15/2017 11:49 AM, Larry Finger wrote:
> On 01/06/2017 11:06 AM, Larry Finger wrote:
>> On 01/06/2017 08:06 AM, Frank Mehnert wrote:
>>> Hi Larry,
>>>
>>> On Dienstag, 3. Januar 2017 21:48:06 CET Larry Finger wrote:
>>>> On 12/27/2016 11:16 AM, Larry Finger wrote:
>>>>> On 12/27/2016 10:56 AM, Larry Finger wrote:
>>>>>> In kernel 4.10-rc1, the kernel routines register_cpu_notifier() and
>>>>>> unregister_cpu_notifier() have been removed and their function replaced by
>>>>>> callbacks from the hotplug state machine. After looking at the changes
>>>>>> required
>>>>>> in some other drivers, I have seen that the required changes will not be
>>>>>> trivial. Before I start doing the work, I want to know if the changes have
>>>>>> been
>>>>>> started and/or done by Oracle.
>>>>>
>>>>> One thing I failed to mention. Commit 7fd8329ba502 in 4.10-rc1 also messes up
>>>>> the ability to define true/false in an out-of-kernel driver. I have
>>>>> submitted a
>>>>> kernel patch to fix this
>>>>> (http://lkml.iu.edu/hypermail/linux/kernel/1612.3/00086.html), but it has not
>>>>> yet been merged or received any comments.
>>>>
>>>> My patch was merged today. It has not yet hit mainline, but I expect it to be
>>>> there as of 4.10-rc3. I still have no fix for the missing
>>>> register_cpu_notifier(), etc.
>>>
>>> thanks for the information. I assume we will do it like the Linux kernel and use
>>> cpuhp_setup_state_nocalls() to register CPU hotplug callbacks. But before we can
>>> provide a fix we need to do a few tests.
>>
>> I think that is the correct approach, but I have not had time to actually do it
>> yet. Unlike most API changes, this one is a bit more intrusive. Once you get a
>> patch, I would be pleased to help test it. Having to boot an old kernel to run a
>> VM is getting old. In addition, the openSUSE kernel HEAD project is running
>> 4.10-rcX kernels, and the VB build fails every time they trigger a new build.
>

Version 5.1.14 has a proper fix, which is better than the hack I posted. In 
addition, the fix for the true/false problem was merged into kernel mainline 
yesterday. It will definitely be in kernel 4.10-rc5.

Larry