AMDGPU Page Fault on Boot
System information
12System: 12Kernel 5.10.12-200.fc33.x86_64 x86_64 12bits 64 12compiler gcc 12v 2.35-18.fc33)
12Desktop GNOME 3.38.3 12tk GTK 3.24.24 12wm gnome-shell 12dm GDM
12Distro Fedora release 33 (Thirty Three)
12CPU: 12Info 8-Core 12model AMD Ryzen 7 3700X 12bits 64 12type MT MCP 12arch Zen 2 12L2 cache 4096 KiB
12flags avx avx2 lm nx pae sse sse2 sse3 sse4_1 sse4_2 sse4a ssse3 svm 12bogomips 114987
12Speed 2170 MHz 12min/max 2200/3600 MHz 12boost enabled 12Core speeds (MHz) 121 2155 122 4317
123 2158 124 2158 125 2156 126 4316 127 4317 128 2158 129 2158 1210 2158 1211 2158 1212 4317 1213 4312
1214 2158 1215 2158 1216 4317
12Graphics: 12Device-1 AMD Navi 10 [Radeon RX 5600 OEM/5600 XT / 5700/5700 XT] 12vendor Tul
12driver amdgpu 12v kernel 12bus ID 0b:00.0 12chip ID 1002:731f
12Display wayland 12server Fedora Project X.org 1.20.10 12compositor gnome-shell
12driver amdgpu,ati 12unloaded fbdev,modesetting,radeon,vesa 12resolution 2560x1440~144Hz
12s-dpi 96
12OpenGL
12renderer AMD Radeon RX 5700 XT (NAVI10 DRM 3.40.0 5.10.12-200.fc33.x86_64 LLVM 11.0.0)
12v 4.6 Mesa 20.3.4 12direct render Yes
Describe the issue
On boot, amdgpu crashes before Plymouth/Gnome can load, and hangs the system until I perform a hard reset. Frequency seems related to kernel versions - 5.10.10 generally does not crash, 5.10.11 crashes every time, and 5.10.12 crashes about 50% of the time. Occasionally, the system will get far enough that Fedora's built-in error reporting catches it - this is the automated report filed from my system: https://bugzilla.redhat.com/show_bug.cgi?id=1924384
I am running an unmodified kernel, and I have not moved away from the default Mesa included with Fedora. I typically run with amdgpu.ppfeaturemask=0xffffffff
enabled, but this crash appears with similar frequency if I remove that kernel option.
Regression
This issue first appeared when Fedora offered kernel 5.10.11 as an update. This update coincided with Mesa 20.3.4 being offered as an update (1 Feb 2021). Before this update, I did not experience this problem.