[elrepo] EL7 kernel abi changes?

[Phil Perry]

On 03/03/17 20:17, Phil Perry wrote:
> On 03/03/17 16:16, Orion Poplawski wrote:
>> On 03/03/2017 09:03 AM, Orion Poplawski wrote:
>>> Looks like perhaps some ABI changes with 3.10.0-514.10.2.el7?
>>>
>>> depmod: WARNING:
>>> /lib/modules/3.10.0-514.10.2.el7.x86_64/weak-updates/nvidia-340xx/nvidia-uvm.ko
>>>
>>> needs unknown symbol nvUvmInterfaceChannelDestroy
>>> ...
>>>
>>> and so on.
>>>
>>> I tried to see what symbols changed, but none of them are explicitly
>>> requires
>>> by the kmod-nvidia package.  But maybe some are missing.
>>> 1c1
>>> < kernel = 3.10.0-514.10.2.el7
>>> ---
>>>> kernel = 3.10.0-514.6.2.el7
>>> 749c749
>>> < kernel(acpi_bus_get_device) = 0x03d75282
>>> ---
>>>> kernel(acpi_bus_get_device) = 0x459997bb
>>> 755c755
>>> < kernel(acpi_bus_register_driver) = 0x611af3b7
>>> ---
>>>> kernel(acpi_bus_register_driver) = 0x0fd3fb0d
>>> 758,759c758,759
>>> < kernel(acpi_bus_trim) = 0x64cf5a8d
>>> < kernel(acpi_bus_unregister_driver) = 0xa0b23612
>>> ---
>>>> kernel(acpi_bus_trim) = 0x597d42de
>>>> kernel(acpi_bus_unregister_driver) = 0xf628513f
>>
>> These seem to be the culprits:
>>
>> Module nvidia.ko from kernel 3.10.0-514.el7.x86_64 is not compatible
>> with
>> kernel 3.10.0-514.10.2.el7.x86_64 in symbols: acpi_bus_register_driver
>> acpi_bus_get_device acpi_bus_unregister_driver
>>
>
> Yes, there were a series of ACPI patches in the latest kernel and I
> think you are right, the ABI looks to have changed and that is causing
> the breakage in nvidia.ko
>
> The only way to fix this is to rebuild the nvidia driver against the
> latest kernel (3.10.0-514.10.2.el7.x86_64), but that will break backward
> compatibility with earlier kernels.
>
> This doesn't occur often, but when it does this is the only way to fix.
>
> I'll rebuild the affected nvidia packages this weekend, and add a
> Requires for the latest kernel.
>
> In the meantime, those needing nvidia support will need to revert to
> booting the previous kernel until updated packages are released.
>
> Apologies for the inconvenience, and thank you for your analysis.
>
> Phil
>

I've rebuilt the kmod-nvidia packages (legacy packages too) against the 
latest RHEL7 kernel and pushed these updates to the testing repository 
for testing.

The following packages are syncing to the mirrors and should be 
available shortly:

kmod-nvidia-304xx-304.135-2.el7.elrepo.x86_64.rpm
kmod-nvidia-340xx-340.102-2.el7.elrepo.x86_64.rpm
kmod-nvidia-375.39-2.el7.elrepo.x86_64.rpm

To test, please run the following:

yum --enablerepo=elrepo-testing update kmod-nvidia\*

and reboot to the 3.10.0-514.10.2.el7.x86_64 kernel

These packages will not be compatible with earlier kernel releases.

Please feel free to provide feedback here.

Thanks

Phil