Memory space was fine.
I have big mem usage as i allocating 95% memory into huge pages 1G
so the rest is 32-64GB of memory for system. Anyway i had plenty at that time when this happened
It was on kernel Kernel: 6.18.8-pbk
Claude was able to remind me that issue:
walk_pgd_range BUG on PVE12 — Crash Details
Server: PVE12 (10.10.42.12)
Date: February 4, 2026, 21:42:16 UTC
Kernel: 6.18.8-pbk (custom build, compiled January 30, 2026) - this is from https://prebuiltkernels.com/
Error:
BUG: unable to handle page fault for address: ff164aee00000000
#PF: supervisor read access in kernel mode
#PF: error_code(0x0000) - not-present page
PGD 4602067 P4D 0
Oops: 0000 [#1] SMP NOPTI
CPU: 125 UID: 0 PID: 783442 Comm: qm Not tainted 6.18.8-pbk #1 PREEMPT(voluntary)
Hardware: TURIN2D24G-2L+/500W (AMD Turin platform)
RIP: 0010:walk_pgd_range+0x6ff/0xbb0
Call Trace:
walk_pgd_range+0x6ff/0xbb0
→ __walk_page_range+0x8e/0x220
→ walk_page_vma+0x92/0xe0
→ smap_gather_stats.part.0+0x8c/0xd0
→ show_smaps_rollup+0x258/0x420
→ seq_read_iter → seq_read → vfs_read → ksys_read
→ __x64_sys_read → do_syscall_64 → entry_SYSCALL_64
Root Cause Analysis:
The qm process (Proxmox QEMU VM manager, PID 783442) triggered a kernel page fault while reading /proc/<pid>/smaps_rollup. The fault occurred in walk_pgd_range() when the kernel attempted to dereference address ff164aee00000000 (stored in R12), which was a not-present page. The corrupted
pointer pattern suggests a stale or corrupted page table entry.
Likely Causes (ranked):
1. Race condition between KVM/QEMU VM memory operations and /proc/smaps_rollup reads
2. Bug in vanilla kernel 6.18.x page table handling with VFIO passthrough (mlx5, vfio-pci modules were loaded)
3. Possible instability introduced by custom -pbk kernel patches
4. Memory corruption from VFIO PCI passthrough affecting page table structures
Key Registers at Crash:
- R12 (faulting pointer): ff164aee00000000
- CR2 (fault address): ff164aee00000000
- RSI: 000071afffffffff (end of VMA range)
- R14: 000071afc0000000 (start of VMA range)
Free Memory: Not checked during that session — no memory statistics were collected.
Previous Kernel (boot -1): Same version 6.18.8-pbk — the system was rebooted after the crash and came back on the same kernel.
Server: PVE12 (10.10.42.12)
Date: February 4, 2026, 21:42:16 UTC
Kernel: 6.18.8-pbk (custom build, compiled January 30, 2026) - this is from https://prebuiltkernels.com/
Error:
BUG: unable to handle page fault for address: ff164aee00000000
#PF: supervisor read access in kernel mode
#PF: error_code(0x0000) - not-present page
PGD 4602067 P4D 0
Oops: 0000 [#1] SMP NOPTI
CPU: 125 UID: 0 PID: 783442 Comm: qm Not tainted 6.18.8-pbk #1 PREEMPT(voluntary)
Hardware: TURIN2D24G-2L+/500W (AMD Turin platform)
RIP: 0010:walk_pgd_range+0x6ff/0xbb0
Call Trace:
walk_pgd_range+0x6ff/0xbb0
→ __walk_page_range+0x8e/0x220
→ walk_page_vma+0x92/0xe0
→ smap_gather_stats.part.0+0x8c/0xd0
→ show_smaps_rollup+0x258/0x420
→ seq_read_iter → seq_read → vfs_read → ksys_read
→ __x64_sys_read → do_syscall_64 → entry_SYSCALL_64
Root Cause Analysis:
The qm process (Proxmox QEMU VM manager, PID 783442) triggered a kernel page fault while reading /proc/<pid>/smaps_rollup. The fault occurred in walk_pgd_range() when the kernel attempted to dereference address ff164aee00000000 (stored in R12), which was a not-present page. The corrupted
pointer pattern suggests a stale or corrupted page table entry.
Likely Causes (ranked):
1. Race condition between KVM/QEMU VM memory operations and /proc/smaps_rollup reads
2. Bug in vanilla kernel 6.18.x page table handling with VFIO passthrough (mlx5, vfio-pci modules were loaded)
3. Possible instability introduced by custom -pbk kernel patches
4. Memory corruption from VFIO PCI passthrough affecting page table structures
Key Registers at Crash:
- R12 (faulting pointer): ff164aee00000000
- CR2 (fault address): ff164aee00000000
- RSI: 000071afffffffff (end of VMA range)
- R14: 000071afc0000000 (start of VMA range)
Free Memory: Not checked during that session — no memory statistics were collected.
Previous Kernel (boot -1): Same version 6.18.8-pbk — the system was rebooted after the crash and came back on the same kernel.
--
tel. 790 202 300
Tytus Rogalewski
Dolina Krzemowa 6A
83-010 Jagatowo
NIP: 9570976234