[Arm-dev] [arm-dev] Gigabyte MP30-AR0

Sun Apr 24 20:00:21 UTC 2016
Gordan Bobic <gordan at redsleeve.org>

On 2016-04-24 17:20, Marcin Juszkiewicz wrote:
> W dniu 24.04.2016 o 17:46, Ronald Maas pisze:
>> On Sun, Apr 24, 2016 at 2:09 AM, Gordan Bobic <gordan at redsleeve.org>
>>  wrote:
> 
>>> Given that we are seeing kernel panics with different SATA, SAS,
>>> AMD and Nvidia cards, the only thing I can conclude is that there
>>> is a major bug somewhere in the PCIe driver. The main difference
>>> between my kernel 4.4.x kernels and the CentOS 4.2.0 is that the
>>> latter uses 64KB memory pages, so that is probably a good place to
>>> start looking.
> 
>> You would expect 64 KB pages to give more issues compared to 4 KB as
>> the latter has been the norm for decades. But I see if I can buy some
>> other more PCIe graphics cards on Craigslist for cheap and test again
>> with the various distros.
> 
>> Otherwise hope these issues will be ironed out over time when
>> 96Boards EE boards finally become widely available.
> 
> I have used Sil SATA controller, Matrox g550, few NVidia cards and two
> Radeon ones (HD5450 and R240) with APM Mustang (also X-Gene1 like
> Gigabyte). All with 64KB kernels.

Have you tried any of the above with kernels that have 4KB pages?
64KB memory pages aren't merely inefficient, they are a complete
show-stopper for me.

> Radeon sits in this machine since my "AArch64 desktop week" which was
> September 2015. Used that card since 3.19 kernel and went through all
> versions used in Fedora/rawhide.

Yeah, I wish Nvidia provided a driver for aarch64. I could
really do with NVENC on my system.

> I rather suspect Gigabyte firmware having some issues. Try booting
> with "acpi=off" and proper DTB to make sure that it is not ACPI
> related?

Can you elaborate on the firmware and DTB issues you are suspecting
and where to get better firmware and DTBs?

Gordan