Public bug reported:

Description:    Ubuntu 24.04.5 LTS
Release:        24.04
On gb10 based Asus GX10

After upgrading from linux-image-7.0.0-1019-nvidia
(7.0.0-1019.19~24.04.2) to linux-image-7.0.0-1024-nvidia
(7.0.0-1024.25~24.04.1) on an ASUS Ascent GX10 (NVIDIA GB10, DGX OS 7.x
on Ubuntu 24.04, arm64), the machine crashes and hard-resets after 2 to
40 minutes of uptime under light desktop use (GNOME, Chromium, a
terminal). Seven crash-reboots within five hours on the first day. On
7.0.0-1019 the same machine ran for up to four days without incident.

Most crashes leave nothing in the journal, consistent with an instant
reset. Two captured via pstore show the same kernel oops in the MediaTek
MT7925 Wi-Fi driver (mt7925e / mt792x_lib / mt76), triggered when the
card wakes from runtime power-save and the driver falls into its WPDMA
reset path:

Unable to handle kernel paging request at virtual address fbfff9ffbf6189b0
Workqueue: mt76 mt792x_pm_wake_work [mt792x_lib]
pc : iommu_put_pages_list
Call trace:
 iommu_put_pages_list
 dma_unmap_page_attrs
 page_pool_return_netmem
 mt76_dma_rx_cleanup+0x104/0x210 [mt76]
 mt792x_wpdma_reset+0xf0/0x340 [mt792x_lib]
 mt792x_wpdma_reinit_cond+0xb4/0x138 [mt792x_lib]
 mt792xe_mcu_drv_pmctrl+0x34/0xc0 [mt792x_lib]
 mt792x_mcu_drv_pmctrl+0x50/0xd0 [mt792x_lib]
 mt792x_pm_wake_work+0x4c/0x258 [mt792x_lib]

The fault address is not a valid arm64 kernel pointer, suggesting the rx
page_pool list being freed is already stale or corrupted. In the one
crash with a complete trace, the IOMMU damage then propagated: about 38
seconds later the GPU (nvidia open module 580.178.04) reported Xid 13,
followed by RCU stalls, a GSP crash and an Xid 45 storm until the
machine reset. The GPU failure appears to be a consequence of the IOMMU
state, not an independent fault.

Workaround: booting 7.0.0-1019 (still installed) with Wi-Fi power-save
disabled (wifi.powersave = 2 in NetworkManager) has been stable. Power-
save alone was not the only factor; 1019 ran for days with power-save
on. The regression is between 1019 and 1024.

Hardware: ASUS Ascent GX10, MediaTek MT7925 (mt7925e), machine is on Wi-
Fi only with no ethernet connected.

7.0.0-1024-nvidia, and the first reboot into it was at 15:42. Since then there
  have been 7 boots, most lasting 2–40 minutes. The previous kernel,
  7.0.0-1019, ran for up to 4 days at a time without crashing.

  The one crash that left a full trace is in the Wi-Fi driver (boot at 16:52,
  crashed at 17:00):

  Unable to handle kernel paging request ...
  Workqueue: mt76 mt792x_pm_wake_work [mt792x_lib]
  pc : iommu_put_pages_list
    ... dma_unmap_page_attrs → page_pool_return_netmem
    → mt76_dma_rx_cleanup → mt792x_wpdma_reset → mt792x_pm_wake_work

  Your Wi-Fi card is a MediaTek MT7925 (mt7925e driver). It is running with
  power-save on. When it woke from power-save, the driver crashed while
  releasing DMA memory and left the IOMMU's state corrupted. The GPU depends on
  that IOMMU, and 38 seconds later it failed: Xid 13 (FECS UCODE fatal error),
  then CPU stalls, then a GSP crash, then an Xid 45 storm until the machine
  died. The GPU failure was a consequence, not the cause.

  The other crashed boots stopped mid-log with no error at all, which fits an
  instant hard reset. I can't prove they're the same bug. Two things point that
  way: they all happened on the new kernel, and the 15:43 boot died while the
  Wi-Fi was stuck in an authentication loop. The 16:30 restart was a normal
  reboot, not a crash. The ethernet port enP7s7 shows no cable, so all your
  traffic goes over this Wi-Fi card.


Quickest fix: go back to the previous kernel, which is still installed. At
the GRUB menu, choose Advanced options → 7.0.0-1019-nvidia. To make it the
default:
! sudo sed -i 's/^GRUB_DEFAULT=.*/GRUB_DEFAULT="Advanced options for
Ubuntu>Ubuntu, with Linux 7.0.0-1019-nvidia"/' /etc/default/grub && sudo
update-grub
(Check the exact menu entry names with grep menuentry /boot/grub/grub.cfg
first.)

Fix
Moving back to 7.0.0-1019-nvidia and disabling power-save on wifi fixed it.

ProblemType: Bug
DistroRelease: Ubuntu 24.04
Package: linux-nvidia (not installed)
ProcVersionSignature: Ubuntu 7.0.0-1019.19~24.04.2-nvidia 7.0.14
Uname: Linux 7.0.0-1019-nvidia aarch64
ApportVersion: 2.28.3-0ubuntu0.1
Architecture: arm64
AudioDevicesInUse:
 USER        PID ACCESS COMMAND
 /dev/snd/controlC1:  maiken     4725 F.... wireplumber
 /dev/snd/controlC0:  maiken     4725 F.... wireplumber
 /dev/snd/seq:        maiken     4722 F.... pipewire
CasperMD5json:
 {
   "result": "skip"
 }
CurrentDesktop: ubuntu:GNOME
Date: Sat Oct 31 19:31:34 2026
InstallationDate: Installed on 2026-02-09 (264 days ago)
InstallationMedia: Ubuntu-Server 24.04.3 LTS "Noble Numbat" - Release arm64 
(20250806.1)
MachineType: ASUSTeK COMPUTER INC. GX10
ProcEnviron:
 LANG=en_US.utf8
 PATH=(custom, no user)
 SHELL=/bin/bash
 TERM=xterm-256color
 XDG_RUNTIME_DIR=<set>
ProcFB: 0 simpledrmdrmfb
ProcKernelCmdLine: BOOT_IMAGE=/boot/vmlinuz-7.0.0-1019-nvidia 
root=UUID=bb2e58a9-bd50-43b5-b2b9-71a79a3c98d1 ro init_on_alloc=0 
iommu.passthrough=0 console=tty0 plymouth.ignore-serial-consoles 
plymouth.use-simpledrm earlycon=uart,mmio32,0x16A00000 crashkernel=1G-:0M quiet 
splash initcall_blacklist=tegra234_cbb_init kho=off pci=pcie_bus_safe 
vt.handoff=7
RelatedPackageVersions:
 linux-restricted-modules-7.0.0-1019-nvidia N/A
 linux-backports-modules-7.0.0-1019-nvidia  N/A
 linux-firmware                             20240318.git3b128b60.0ubuntu3.1
SourcePackage: linux-nvidia
UpgradeStatus: No upgrade log present (probably fresh install)
dmi.bios.date: 07/07/2026
dmi.bios.release: 5.36
dmi.bios.vendor: American Megatrends International, LLC.
dmi.bios.version: GX10DGX.0106.2026.0707.0731
dmi.board.asset.tag: Default string
dmi.board.name: GX10
dmi.board.vendor: ASUSTeK COMPUTER INC.
dmi.board.version: Default string
dmi.chassis.asset.tag: Default string
dmi.chassis.type: 17
dmi.chassis.vendor: ASUSTeK COMPUTER INC.
dmi.chassis.version: Default string
dmi.modalias: 
dmi:bvnAmericanMegatrendsInternational,LLC.:bvrGX10DGX.0106.2026.0707.0731:bd07/07/2026:br5.36:svnASUSTeKCOMPUTERINC.:pnGX10:pvr5.36_GX10DGX:rvnASUSTeKCOMPUTERINC.:rnGX10:rvrDefaultstring:cvnASUSTeKCOMPUTERINC.:ct17:cvrDefaultstring:skuDefaultstring:pfaDGXSpark:
dmi.product.family: DGX Spark
dmi.product.name: GX10
dmi.product.sku: Default string
dmi.product.version: 5.36_GX10DGX
dmi.sys.vendor: ASUSTeK COMPUTER INC.

** Affects: linux-nvidia (Ubuntu)
     Importance: Undecided
         Status: New


** Tags: apport-bug arm64 noble

-- 
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2172676

Title:
  linux-nvidia 7.0.0-1024: kernel paging fault in mt792x_pm_wake_work
  (MT7925 Wi-Fi) corrupts IOMMU, machine hard-resets; regression from
  7.0.0-1019

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux-nvidia/+bug/2172676/+subscriptions


-- 
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs

Reply via email to