clone of my dotfiles.ssp.sh
add check
+48
-5
+48
-5
hypr/.config/hypr/sspaeti/gpu-health-check
+48
-5
hypr/.config/hypr/sspaeti/gpu-health-check
···
6
6
YELLOW='\033[0;33m'
7
7
NC='\033[0m'
8
8
9
+
warn=0
10
+
9
11
echo "GPU Health Check"
10
12
echo "================"
11
13
···
18
20
echo -e "${RED}[WARN]${NC} GPU errors detected:"
19
21
echo "$errors" | head -5
20
22
echo -e "${YELLOW}Recommendation: Reboot before your meeting${NC}"
23
+
warn=1
21
24
fi
22
25
23
-
# Check GPU is responding
24
-
if busy=$(cat /sys/class/drm/card*/device/gpu_busy_percent 2>/dev/null | head -1); then
26
+
# Check MES errors specifically (early warning before full crash)
27
+
mes_count=$(journalctl -b -k --no-pager 2>/dev/null | grep -c "MES")
28
+
if [ "$mes_count" -gt 0 ]; then
29
+
echo -e "${RED}[WARN]${NC} MES errors this boot: ${mes_count} (crash imminent, reboot NOW)"
30
+
warn=1
31
+
else
32
+
echo -e "${GREEN}[OK]${NC} MES scheduler healthy"
33
+
fi
34
+
35
+
# Check GPU is responding (timeout = early freeze detection)
36
+
if busy=$(timeout 2 cat /sys/class/drm/card*/device/gpu_busy_percent 2>/dev/null | head -1); then
25
37
echo -e "${GREEN}[OK]${NC} GPU responding (${busy}% busy)"
26
38
else
27
-
echo -e "${RED}[FAIL]${NC} Cannot read GPU status"
39
+
echo -e "${RED}[FAIL]${NC} Cannot read GPU status (GPU may be hung)"
40
+
warn=1
28
41
fi
29
42
30
43
# Check VRAM
···
35
48
echo -e "${YELLOW}[SKIP]${NC} VRAM info not available"
36
49
fi
37
50
51
+
# Check CWSR status (known MES crash trigger on gfx1150)
52
+
cwsr=$(cat /sys/module/amdgpu/parameters/cwsr_enable 2>/dev/null)
53
+
if [ "$cwsr" = "1" ]; then
54
+
echo -e "${YELLOW}[INFO]${NC} CWSR enabled (known MES crash risk on gfx1150)"
55
+
echo -e " To disable: add amdgpu.cwsr_enable=0 to kernel params"
56
+
elif [ "$cwsr" = "0" ]; then
57
+
echo -e "${GREEN}[OK]${NC} CWSR disabled (MES crash mitigation active)"
58
+
fi
59
+
60
+
# Check if last suspend/resume left GPU degraded
61
+
resume_errors=$(journalctl -b -k --no-pager 2>/dev/null | grep -E "(SMU is resumed|ring .* uses VM)" | tail -1)
62
+
if [ -n "$resume_errors" ]; then
63
+
post_resume_errors=$(journalctl -b -k --no-pager 2>/dev/null | sed -n "/SMU is resumed/,\$p" | grep -cE "(fail|error|timeout)" 2>/dev/null)
64
+
if [ "$post_resume_errors" -gt 0 ]; then
65
+
echo -e "${RED}[WARN]${NC} GPU errors after last resume: ${post_resume_errors} (reboot recommended)"
66
+
warn=1
67
+
else
68
+
echo -e "${GREEN}[OK]${NC} Last suspend/resume was clean"
69
+
fi
70
+
fi
71
+
72
+
# Check kernel params are set
73
+
cmdline=$(cat /proc/cmdline 2>/dev/null)
74
+
missing=""
75
+
echo "$cmdline" | grep -q "gpu_recovery=1" || missing="${missing} gpu_recovery"
76
+
echo "$cmdline" | grep -q "ip_block_mask" || missing="${missing} ip_block_mask(VPE)"
77
+
if [ -n "$missing" ]; then
78
+
echo -e "${YELLOW}[INFO]${NC} Missing kernel params:${missing}"
79
+
fi
80
+
38
81
# Summary
39
82
echo ""
40
-
if [ -z "$errors" ]; then
41
-
echo -e "${GREEN}GPU is healthy - safe to start meeting${NC}"
83
+
if [ "$warn" -eq 0 ]; then
84
+
echo -e "${GREEN}GPU is healthy - safe to proceed${NC}"
42
85
else
43
86
echo -e "${RED}GPU may be unstable - consider rebooting${NC}"
44
87
fi