Public bug reported:

On kernel 7.0.0-42-generic (HWE, Ubuntu 24.04), every boot hits IOMMU (DMAR) 
faults from the AHCI SATA controller
about 70 seconds after boot. Writes to the root SSD fail, and ext4 remounts / 
read-only. On the same machine,
6.8.0-149-generic has run for days with no DMAR faults and no SATA errors.

Kernel log on 7.0 (monotonic timestamps; the DMAR fault comes ~15 ms
before the first AHCI error):

    [   68.031183] DMAR: [DMA Read NO_PASID] Request device [00:1f.2] fault 
addr 0xd0400000 [fault reason 0x0c] non-zero reserved fields in PTE
    [   68.046124] ata2.00: irq_stat 0x20000000, host bus error
    [   68.046130] ata2.00: cmd 61/00:b0:00:50:ca/20:00:26:00:00/40 tag 22 ncq 
dma 4194304 ou
                            res 40/00:01:00:00:00/00:00:00:00:00/00 Emask 0x60 
(host bus error)
    ata2: SError: { HostInt }
    ata2.00: failed command: WRITE FPDMA QUEUED
    ata2: hard resetting link
    I/O error, dev sdb, sector 651194368 op 0x1:(WRITE) flags 0x800 phys_seg 64 
prio class 2
    EXT4-fs warning (device sdb2): ext4_end_bio:369: I/O error 10 writing to 
inode 11567604 ...
    EXT4-fs (sdb2): failed to convert unwritten extents to written extents -- 
potential data loss!
    EXT4-fs (sdb2): Remounting filesystem read-only

(Excerpt shortened. The read-only remount line is from a photo of the
screen; the journal could no longer be written by then.)

Observations:
- Reproduced on 3 boots of 7.0.0-42 (2026-11-08 twice, 2026-11-11 once). It 
always starts ~68-71 s after boot, during
  normal startup activity. Two of the three boots ended with / remounted 
read-only. In the third, writes failed but the fs stayed rw.
- The faulting device is always 00:1f.2 (the SATA controller), always "DMA 
Read", always fault reason 0x0c.
- All fault addresses are 4 MiB-aligned: 0xcf800000, 0xcfc00000, 0xd0000000, 
0xd0400000, 0xd0c00000.
- The failing commands are WRITE FPDMA QUEUED of exactly 4 MiB (`ncq dma 
4194304`). In the 2026-11-11 boot, 52 of 52 listed
  failed commands were 4 MiB.
- /sys/block/sdb/queue/max_sectors_kb is 4096 on 7.0, but 1280 on 6.8 
(max_hw_sectors_kb = 32767 on both).
  So 6.8 never issues 4 MiB requests. My guess is that 7.0 maps these 4 MiB 
buffers with an IOMMU
  page-table entry (superpage?) that this Haswell-era VT-d rejects, but I 
haven't verified that.
- The IOMMU is in the same mode on both kernels: "iommu: Default domain type: 
Translated", lazy TLB invalidation,
  CONFIG_INTEL_IOMMU_DEFAULT_ON=y in both configs.
- 6.8.0-149: zero "DMAR: [" lines over several days and several boots; no ata 
errors.
- Not SATA link power management: the 2026-11-11 failure happened with 
ahci.mobile_lpm_policy=1 (lpm-pol 1 /
  max_performance on all ports).
- Not the drive: SMART PASSED, 0 reallocated/pending/uncorrectable sectors. The 
drive's CRC_Error_Count rises only during
  7.0 boots (by about 35000 in the ~90 s failed boot), and stays flat on 6.8.
- The second, identical SSD (sda, ata1) holds Windows and is mostly idle under 
Linux; it showed no errors.

Hardware / software:
- CPU: Intel Core i7-4790K (Haswell)
- Board: Gigabyte Z97X-UD5H-BK, BIOS F5 (05/30/2014)
- SATA: 00:1f.2 Intel 9 Series Chipset Family SATA Controller [AHCI Mode] 
[8086:8c82], subsystem Gigabyte [1458:b005]
- Disks: 2x INTEL SSDSC2BB480G4 (fw D2010370); root is /dev/sdb2 (ext4) on ata2
- Ubuntu 24.04.5 LTS
- Bad: 7.0.0-42-generic (Ubuntu 7.0.0-42.42~24.04.1-generic 7.0.14)
- Good: 6.8.0-149-generic (Ubuntu 6.8.0-149.149-generic 6.8.12)
- Kernel command line: `quiet splash` (plus `ahci.mobile_lpm_policy=1` in the 
2026-11-11 boots)

Not tested yet (I can test if useful): intel_iommu=sp_off, iommu=pt,
intel_iommu=off, or limiting max_sectors_kb to 1280 on 7.0.

Similar reports: the same "non-zero reserved fields in PTE" DMAR fault on 
Haswell/Broadwell Macs with failed writes
(https://github.com/omacom/omarchy/issues/14160), and on Intel audio with 
Debian's 6.16 kernel (Debian bug #1112627).

Attached: full kernel logs (journalctl -k -o short-monotonic) of two
failing 7.0 boots and one good 6.8 boot.

** Affects: linux-hwe-7.0 (Ubuntu)
     Importance: Undecided
         Status: New


** Tags: noble regression-update

** Attachment added: "kernel-log-7.0-failed-boot-2026-11-11.txt"
   
https://bugs.launchpad.net/bugs/2174419/+attachment/6017745/+files/kernel-log-7.0-failed-boot-2026-11-11.txt

-- 
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2174419

Title:
  7.0 HWE: DMAR fault "non-zero reserved fields in PTE" on Intel 9
  Series AHCI [8086:8c82] causes SATA host bus errors, failed writes and
  ext4 read-only remount (6.8 unaffected)

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux-hwe-7.0/+bug/2174419/+subscriptions


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

Reply via email to