Qualcomm WCN7850 WiFi 7 - Linux Firmware Fix

Hardware: Gigabyte X870E AORUS XTREME AI TOP (QCNCM865 module) WiFi: Qualcomm WCN7850 hw2.0 (FastConnect 7800) Status: FIXED - WiFi sees 40+ networks, Bluetooth working

---

Problem

The WCN7850 WiFi 7 card reports board_id 0xff (generic fallback) because the subsystem ID 105b:e0fb (Gigabyte X870E) is missing from the linux-firmware board-2.bin. This limits TX power calibration, though WiFi still achieves ~910 Mbps in practice.

ath12k_pci 0000:0d:00.0: failed to get ACPI BDF EXT: 0
ath12k_pci 0000:0d:00.0: board_id 0xff

Root Cause

Incorrect: Originally thought 105b:e0fb was missing from linux-firmware.

Actual Problem: The e0fb entry EXISTS in upstream board-2.bin, but in MULTIPLE groups:

1. Generic e0ee group (wrong calibration) ← Driver finds this FIRST
2. Generic e0dc group (also generic)
3. e0dc,variant=QC_5mm group (correct calibration)

When ACPI BDF lookup fails (as it does for us), the driver searches board-2.bin WITHOUT variant, finds the generic e0ee entry first, and uses conservative TX power calibration.

Result: Low TX power (~1 dBm), can only see own mesh network, Bluetooth crashes on scan.

Solution

Modify board-2.bin to:

1. Remove e0fb from all generic (non-variant) groups
2. Add standalone e0fb entry at position [0] pointing to QC_5mm.bin calibration data

This ensures the driver finds the correct entry first during search.

Current State

Item	Status
WiFi Networks	40+ detected (was: only own mesh)
Bluetooth	Working, stable (was: crash on scan)
board_id	Still 0xff (hardware-reported, expected)
Custom board-2.bin	Installed
Pacman hook	Active (survives updates)

---

Quick Start

The fix is installed. To test:

sudo reboot

After reboot, verify:

# Check if board_id changed from 0xff
sudo dmesg | grep -i "board_id\|ACPI BDF"

# Check subsystem ID detection
sudo dmesg | grep -i "105b:e0fb\|e0fb"

---

Documentation

File	Description
STATUS.md	Current state and test results
INSTALL.md	Installation guide
CHANGELOG.md	Version history
EMERGENCY-ROLLBACK.md	Rollback procedures

---

Technical Details

Root Cause Analysis

1. Driver loads board-2.bin from /usr/lib/firmware/ath12k/WCN7850/hw2.0/
2. Searches for entry matching PCI subsystem ID 105b:e0fb
3. Entry doesn't exist -> falls back to board_id 0xff (generic)
4. Generic board uses conservative TX power (1 dBm reported)

Fix Approach

1. Extract original board-2.bin using ath12k-bdencoder
2. Add entry for subsystem ID 105b:e0fb (shares calibration with e0dc)
3. Rebuild and install custom board-2.bin
4. Create pacman hook to restore after linux-firmware updates

Key Files

Project Files:
├── board-2.bin.zst                    # Custom firmware (source of truth)
├── board-2-fixed.json                 # Fixed JSON with standalone e0fb entry
├── board-2-fixed.bin                  # Rebuilt board-2.bin
├── fix-e0fb-entry.py                  # Script to fix JSON
├── board-2.json                       # Original extracted board data
├── qca-swiss-army-knife/              # Qualcomm tools (ath12k-bdencoder)

System Files:
├── /usr/lib/firmware/ath12k/WCN7850/hw2.0/board-2.bin.zst  # Installed firmware
├── /etc/pacman.d/hooks/99-wcn7850-board-fix.hook           # Pacman hook
└── /usr/local/bin/wcn7850-board-fix.sh                     # Restore script

Pacman Hook

Automatically restores custom board-2.bin after linux-firmware or linux-firmware-ath updates:

[Trigger]
Operation = Install
Operation = Upgrade
Type = Package
Target = linux-firmware
Target = linux-firmware-ath

[Action]
Description = Restoring WCN7850 board-2.bin with e0fb fix...
When = PostTransaction
Exec = /usr/local/bin/wcn7850-board-fix.sh

---

System Info

• Kernel: 6.17.8-arch1-1
• Bootloader: Limine (UKI)
• Firmware: linux-firmware 20251111-1
• PCI: 0000:0d:00.0 (17cb:1107, subsystem 105b:e0fb)

---

Previous Approaches (Historical)

SSDT Approach (v1-v3)

Attempted to provide board calibration hints via ACPI SSDT table. The SSDT loads at boot but the ath12k driver's ACPI lookup doesn't read _DSM from the device.

Status: SSDT loads, but driver ignores it. May require kernel patch.

Files in acpi/ directory are from this approach and remain for reference.

---

If It Doesn't Work

If board_id still shows 0xff after reboot, the entry format may need adjustment or upstream driver behavior may need investigation.

Current WiFi performance (~910 Mbps) is acceptable as fallback.

Upstream Submission

Once verified working, this fix should be submitted to linux-firmware upstream to add the 105b:e0fb subsystem ID entry for Gigabyte X870E motherboards.