1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
|
AMD Strix Halo VPE/CWSR Freeze Fix Instructions
===============================================
Created: 2026-01-22
Machine: ratio (Framework Desktop, AMD Ryzen AI Max 300)
PROBLEM SUMMARY
---------------
Two AMD GPU bugs cause random freezes on Strix Halo:
1. VPE Power Gating Bug
- VPE (Video Processing Engine) tries to power gate after 1 second idle
- SMU hangs, system freezes
- Fix: amdgpu.pg_mask=0 (disables power gating)
2. CWSR Bug (Compute Wavefront Save/Restore)
- MES firmware hang under compute loads
- Causes GPU reset loops and crashes
- Fix: amdgpu.cwsr_enable=0
Current state on ratio:
- pg_mask = 4294967295 (power gating ENABLED - bad)
- cwsr_enable = 1 (CWSR ENABLED - bad)
- Neither workaround is applied
PART 1: GRUB CMDLINE FIX (Quick, can do now)
============================================
This adds the parameters to the kernel command line via GRUB.
Can be done on the running system, takes effect on next boot.
Step 1: Edit GRUB defaults
--------------------------
sudo nano /etc/default/grub
Find the line:
GRUB_CMDLINE_LINUX_DEFAULT="..."
Add these parameters (keep existing ones):
GRUB_CMDLINE_LINUX_DEFAULT="... amdgpu.pg_mask=0 amdgpu.cwsr_enable=0"
Example - if current line is:
GRUB_CMDLINE_LINUX_DEFAULT="loglevel=2 rd.systemd.show_status=auto rd.udev.log_level=2 nvme.noacpi=1 mem_sleep_default=deep nowatchdog random.trust_cpu=off quiet splash"
Change to:
GRUB_CMDLINE_LINUX_DEFAULT="loglevel=2 rd.systemd.show_status=auto rd.udev.log_level=2 nvme.noacpi=1 mem_sleep_default=deep nowatchdog random.trust_cpu=off quiet splash amdgpu.pg_mask=0 amdgpu.cwsr_enable=0"
Step 2: Regenerate GRUB config
------------------------------
sudo grub-mkconfig -o /boot/grub/grub.cfg
Step 3: Reboot and verify
-------------------------
sudo reboot
After reboot, verify:
cat /sys/module/amdgpu/parameters/pg_mask
# Should show: 0
cat /sys/module/amdgpu/parameters/cwsr_enable
# Should show: 0
cat /proc/cmdline | grep -oE "(pg_mask|cwsr_enable)=[^ ]+"
# Should show:
# pg_mask=0
# cwsr_enable=0
PART 2: MODPROBE.D FIX (Permanent, requires live ISO)
=====================================================
This embeds the parameters in the initramfs so they're always applied.
MUST be done from live ISO because mkinitcpio triggers the freeze.
Step 1: Boot archzfs live ISO
-----------------------------
- Boot from USB with archzfs ISO
- Get to root shell
Step 2: Import and mount ZFS
----------------------------
zpool import -f zroot
zfs mount zroot/ROOT/default
mount /dev/nvme1n1p1 /mnt/boot # Note: nvme1n1p1, not nvme0n1p1!
Verify:
ls /mnt/boot/vmlinuz*
# Should show kernel images
Step 3: Create modprobe config
------------------------------
cat > /mnt/etc/modprobe.d/amdgpu.conf << 'EOF'
# Workarounds for AMD Strix Halo GPU bugs
# Created: 2026-01-22
# Remove when kernel has proper fixes (check linux-lts >= 6.18 with fixes)
# Disable power gating to prevent VPE freeze
# VPE tries to power gate after 1s idle, causing SMU hang
options amdgpu pg_mask=0
# Disable Compute Wavefront Save/Restore to prevent MES hang
# CWSR causes MES firmware 0x80 hang under compute loads
options amdgpu cwsr_enable=0
EOF
Step 4: Chroot and rebuild initramfs
------------------------------------
# Mount system directories
mount --rbind /dev /mnt/dev
mount --rbind /sys /mnt/sys
mount --rbind /proc /mnt/proc
mount --rbind /run /mnt/run
# Chroot
arch-chroot /mnt
# Rebuild initramfs (this is safe from live ISO)
mkinitcpio -P
# Verify amdgpu.conf is in initramfs
lsinitcpio /boot/initramfs-linux.img | grep amdgpu
# Should show: etc/modprobe.d/amdgpu.conf
# Exit chroot
exit
Step 5: Clean up and reboot
---------------------------
# Unmount everything
umount -R /mnt/dev /mnt/sys /mnt/proc /mnt/run
zfs unmount -a
zpool export zroot
# Reboot
reboot
Step 6: Verify after reboot
---------------------------
cat /sys/module/amdgpu/parameters/pg_mask
# Should show: 0
cat /sys/module/amdgpu/parameters/cwsr_enable
# Should show: 0
lsinitcpio /boot/initramfs-linux.img | grep amdgpu.conf
# Should show: etc/modprobe.d/amdgpu.conf
VERIFICATION CHECKLIST
======================
After applying fixes, verify:
[ ] pg_mask shows 0 (not 4294967295)
[ ] cwsr_enable shows 0 (not 1)
[ ] Parameters visible in /proc/cmdline (if using GRUB method)
[ ] amdgpu.conf in initramfs (if using modprobe.d method)
[ ] System stable - no freezes during idle
[ ] mkinitcpio -P completes without freeze (test after fix applied)
IMPORTANT NOTES
===============
1. Boot partition UUID
ratio has mirrored NVMe drives. The boot partition is on nvme1n1p1:
- nvme0n1p1: 6A4B-47A4 (NOT the boot partition)
- nvme1n1p1: 6A4A-93B1 (THIS is /boot)
2. Kernel is pinned
/etc/pacman.conf has: IgnorePkg = linux
This prevents upgrading from 6.15.2 until manually unpinned.
DO NOT upgrade to 6.18.x - it has worse bugs for Strix Halo.
3. When to remove workarounds
Monitor Framework Community and AMD-gfx mailing list for proper fixes.
When linux-lts has confirmed VPE and CWSR fixes, can try removing.
Test by commenting out lines in amdgpu.conf, rebuild initramfs, test.
4. If system freezes during mkinitcpio
This means the fix isn't active yet. Must do from live ISO.
The modconf hook reads /etc/modprobe.d/ at build time, but the
running kernel still has the old parameters until reboot.
TROUBLESHOOTING
===============
System still freezes after GRUB fix:
- Check /proc/cmdline - are parameters there?
- Check /sys/module/amdgpu/parameters/* - are values correct?
- If cmdline has them but sysfs doesn't, driver may have loaded before
parsing. Need modprobe.d method instead.
Can't import zpool from live ISO:
- Try: zpool import -f zroot
- If "pool was previously in use": zpool import -f zroot
- Check hostid: cat /etc/hostid on installed system
mkinitcpio says "Preset not found":
- Check /etc/mkinitcpio.d/*.preset files exist
- For linux kernel: linux.preset
- For linux-lts: linux-lts.preset
After chroot, wrong mountpoints:
- Reset mountpoints after any chroot work:
zfs set mountpoint=/ zroot/ROOT/default
zfs set mountpoint=/home zroot/home
(etc. for all datasets)
REFERENCES
==========
VPE timeout patch (not merged):
https://www.mail-archive.com/amd-gfx@lists.freedesktop.org/msg127724.html
Framework Community - critical 6.18 bugs:
https://community.frame.work/t/attn-critical-bugs-in-amdgpu-driver-included-with-kernel-6-18-x-6-19-x/79221
CWSR workaround:
https://github.com/ROCm/ROCm/issues/5590
Session documentation:
- docs/2026-01-22-ratio-boot-fix-session.org
- docs/2026-01-22-mkinitcpio-config-boot-failure.org
- assets/2026-01-22-mkinitcpio-freeze-during-rebuild.org
|
