Dear Maintainer,

*** Summary ***
Severe SATA link degradation and host bus fatal errors (ICRC ABRT / -5 EIO)
introduced on Intel 9-Series Chipset (8086:8c82) when upgrading from Debian
12
(kernel 6.1) to Debian 13 (kernel 7.2.11+deb13-amd64).

*** Environment & Hardware ***
- Debian Release: Debian 13 (Trixie)
- Regression: Worked cleanly with 0 CRC errors under Debian 12 (Linux 6.1.x)
- SATA Controller: Intel Corporation 9 Series Chipset Family SATA
Controller [AHCI Mode] [8086:8c82] (subsystem [1458:b005])
- Driver: ahci
- Storage: 2x 8TB WD Red Plus (WD80EFPX) in Linux Software RAID 1 (mdadm) +
1x Samsung SATA SSD boot drive
- Filesystem: XFS on md0

*** Behavior Under Debian 12 (Kernel 6.1) ***
- Stable 6.0 Gbps / 3.0 Gbps link negotiation across all drives.
- Sustained concurrent write workloads without errors.
- SMART 199 UDMA_CRC_Error_Count remained static (0 increments).

*** Regressive Behavior Under Debian 13 (Kernel 7.x) ***
Upon upgrading to Debian 13, any heavy write operation (such as XFS mounting
and log replay) immediately triggers concurrent bus aborts across multiple
ATA ports (ata5 and ata6):

[  112.127842] ata5.00: irq_stat 0x28000000, host bus error, interface
fatal error
[  112.127857] ata5: SError: { UnrecovData HostInt CommWake Handshk }
[  112.127878] ata5.00: failed command: WRITE FPDMA QUEUED
[  112.127972] ata5.00: error: { ICRC ABRT }
[  112.507854] ata5: limiting SATA link speed to 1.5 Gbps
...
[  115.009694] I/O error, dev sdc, sector 8590198824 op 0x1:(WRITE)
[  116.993733] I/O error, dev sda, sector 8590198824 op 0x1:(WRITE)
[  117.256602] XFS (md0): log recovery write I/O error at daddr 0x0 len
4096 error -5

Under repeated retries, the root filesystem disk also encountered I/O
aborts
and remounted read-only. SMART attribute 199 (UDMA_CRC_Error_Count)
incremented
on every mount attempt.

*** Workarounds Attempted (Unsuccessful in 7.x) ***
- Tested with known-good SATA cables: No change.
- Isolated drive power across separate native PSU rails: No change.
- Forced ALPM policy to max_performance: No change.
- Kernel parameters 'libata.force=3.0G', and 'libata.force=1.5G':
  Failed to prevent link speed renegotiation loops and subsequent bus drops.

*** Resolution / Workaround ***
Moving the identical drives, cables, and power supply to an alternative
motherboard running the exact same Debian 13 installation resolved the
issue
completely (drives link cleanly at 6.0 Gbps, XFS mounted clean, scrub
completed).

The newer libahci/libata driver appears to handle PHY signal tolerances,
ASPM/ALPM, or FIFO DMA queue management on Intel Wildcat Point / 9-series
PCH silicon significantly less tolerantly than kernel 6.1 did, turning
minor bus jitter into catastrophic link degradation and unrecoverable EIO.
EOF