MT7925 (0e8d:0717): HFP microphone unusable — firmware delivers zero-filled (e)SCO frames on sentinel handle 0x0E00
Hi,
Independent reproduction of this report on different hardware, a different headset and newer firmware.
Hardware: Framework Desktop (AMD Ryzen AI Max 300), BIOS 03.06
MediaTek MT7925 (RZ717), BT on USB 0e8d:0717
Firmware: BT_RAM_CODE_MT7925_1_1_hdr.bin build 20260813113236
(linux-firmware 20261019, newer than the builds tested above)
Kernel: 7.3.0-5-cachyos
Userspace: BlueZ 5.87, PipeWire 1.6.11, WirePlumber 0.5.19
Headset: Sony WH-1000XM4 (HFP, native backend)Symptoms match: A2DP is fine; in HFP the microphone stops delivering audio while the SCO link stays up. The kernel logs the same signatures:
Bluetooth: hci0: corrupted SCO packet
Bluetooth: hci0: SCO packet for unknown connection handle 3584
(32 "corrupted SCO packet" and 18 "handle 3584" lines in the current boot.)
Observations that may help:
- Codec-independent here. It fails with mSBC (Transparent) and also after pinning CVSD (headset-head-unit-cvsd): the CVSD session failed about two minutes after a fresh reconnect, with "corrupted SCO packet" x3
followed by
"handle 3584" at the moment the capture went silent. So removing
mSBC, the
workaround that helps MT7921 users in bluez#2545, does not avoid it
on this
MT7925.
- Two failure shapes seen at capture level (pw-record on the BT source):
1. Pure zeros (max amplitude 0.000000) for roughly 8-22 s after a "corrupted SCO packet", then audio recovered without any reconnect.
2. A persistent state where speech arrives about 35 dB too low (peaks around -45 dBFS vs about -18 dBFS when healthy). No userspace errors are logged; a profile switch (mSBC -> CVSD) does not clear it, only a full ACL disconnect/connect does.
- Not the headset or userspace: a USB headset on the same machine and the same PipeWire/Steam setup works, and the WH-1000XM4 mic is fine on other hosts.
I can run a btmon capture of a failing session, test patches against bluetooth-next or mainline, or try other firmware builds if that helps.
Is anyone from MediaTek looking at the 0x0E00 SCO frames?
Thanks,
Mo