amdgpu support for Polaris11, Radeon Pro 400 series graphics

[StDenis, Tom]

if PP is being sketchy can you force the performance to low (e.g. manual) instead of auto and see if keeping at lower clocks helps keep it stable?


Just an idea to try.


Tom


________________________________
From: amd-gfx <amd-gfx-bounces at lists.freedesktop.org> on behalf of Bernhard Froemel <bfroemel at gmail.com>
Sent: Friday, November 25, 2016 11:15
To: amd-gfx at lists.freedesktop.org
Subject: amdgpu support for Polaris11, Radeon Pro 400 series graphics

Hi everyone,

I'd very much appreciate any help to get the open source amdgpu driver working
for the Radeon Pro 460 [1] found in the recently released
Apple MacBook Pro, Late 2016, 15" (model: MacBookPro13,3).
Note that the Radeon Pro is part of a hybrid graphics setup, paired with an
Intel HD Graphics 530 and an Apple GMUX device. By default/after a system reboot
the GMUX connect the AMD GPU to the available displays. Consequently, the
hybrid setup should not cause any additional issues. For example, the Intel GPU
seems to be not powered, because it is not even enumerated.

lspci tells me:
> 01:00.0 VGA compatible controller: Advanced Micro Devices,
> Inc. [AMD/ATI] Baffin [Radeon RX 460] (rev c0) (prog-if 00 [VGA controller])
> Subsystem: Apple Inc. Baffin [Radeon RX 460]

The amdgpu driver in linux-4.9-rc6 (+ polaris11 firmware from
the firmware-amd-graphics Debian package) detects the card, but the system
crashes in the later stages of probing/initializing the driver (see attached
dmesg_gpuamd2.txt).

Indeed it seems that there is a divison by zero encountered in:
drivers/gpu/drm/amd/powerplay/smumgr/polaris10_smc.c:1918

> t_diff1 = hwmgr->thermal_controller.advanceFanControlParameters.usTMed -
> hwmgr->thermal_controller.advanceFanControlParameters.usTMin;
> t_diff2 = hwmgr->thermal_controller.advanceFanControlParameters.usTHigh -
> hwmgr->thermal_controller.advanceFanControlParameters.usTMed;
.
.
> slope1 = (uint16_t)((50 + ((16 * duty100 * pwm_diff1) / t_diff1)) / 100);
> slope2 = (uint16_t)((50 + ((16 * duty100 * pwm_diff2) / t_diff2)) / 100);

t_diff1 and t_diff2 are zero; probably because active cooling by fans
is realized
quite differently in this embedded/laptop version of the graphics card.

When commenting this simply out, the driver loads and I can run an X
server .. for a while; even glxgears as well as glxinfo seem to run fine.
However after some time or possibly after certain actions (e.g., on one occasion
when I set a terminal window to full-screen) the screen blanks out and the
system becomes sometimes completely unresponsive.

On one of  the more partial system crashes I could still read dmesg
(see attached
dmesg_gpuamd3.txt also detailing the log messages before):
> [  273.877708]
>                 failed to send message 5e ret is 0
> [  286.343630]
>                 failed to send message 5e ret is 0
> [  906.960345] amdgpu 0000:01:00.0: GPU fault detected: 146 0x04121014
> [  906.960348] amdgpu 0000:01:00.0:
> VM_CONTEXT1_PROTECTION_FAULT_ADDR   0x0010CE82
> [  906.960349] amdgpu 0000:01:00.0:
> VM_CONTEXT1_PROTECTION_FAULT_STATUS 0x0D010014
> [  906.960350] VM fault (0x14, vmid 6) at page 1101442, write from
>  'CB3' (0x43423300) (16)

Again, any help/advice/pointers highly appreciated!

Greetings,
  Bernhard

[1] http://www.amd.com/en-us/press-releases/Pages/radeon-pro-400-2016oct27.aspx
[http://www.amd.com/Style%20Library/Images/logo.png]<http://www.amd.com/en-us/press-releases/Pages/radeon-pro-400-2016oct27.aspx>

Radeon Pro 400 Series Graphics: AMD's Most Powerful ...<http://www.amd.com/en-us/press-releases/Pages/radeon-pro-400-2016oct27.aspx>
www.amd.com
Today AMD (NASDAQ: AMD) unveiled a new family of power-efficient graphics processors, Radeon(tm) Pro 400 Series Graphics. Available first in the all-new 15-inch ...



-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/amd-gfx/attachments/20161126/b47bf78b/attachment.html>