ath11k: QCN9074 public firmware stalls at QMI cap request on C-360
Christopher Griffin
christophergriffin05 at gmail.com
Mon Nov 9 22:54:16 PST 2026
Baochen, Rameshkumar,
A direct question on this one, since a month of testing has ruled out
everything on the host side that I can reach.
The board is an Arista C-360: an IPQ8074 host with two PCIe cards,
both 17cb:1104 and both reporting board_id 0xff. One is a 4x4 QCN9074
(6 GHz); the other is 2x2, probably a QCN9072 (2.4/5/6 GHz scan
radio). ath11k from backports 7.2 on Linux 6.18.
No QCN9074 hw1.0 build that Qualcomm has posted boots on it. That is
24 builds now: the 15 in ath11k-firmware (2.5.0.1-01100 through
2.9.0.1-02146; 2.9.0.1-02175 has 02146's amss.bin), on both cards,
and this week the nine under testing/ in quic/upstream-wifi-fw, on
the 2x2 card (2.6.0.1.r6-00010, 2.6.0.1.r7-00002, 2.7.0.1.r1-00003,
2.7.0.1.r2-00005, 2.6.0.1-00828, 2.7.0.1-01701, 2.9.0.1-01385,
-01713, -01832). Two builds that are not in linux-firmware do boot:
the AP vendor's WLAN.HK.2.5.r4-00683, in 2 s, and the
2.15.0.1.r2-00004 from my earlier mail. amss.bin alone decides it:
the vendor's amss.bin boots with any public m3.bin, and a public
amss.bin fails with the vendor's m3.bin.
The failure is always the same, on either card, with the other card
bound or unbound:
MHI SBL -> mission mode, WLFW service connects, indication register,
host capability, memory request (type 1 36700160, type 4 3784704,
type 3 1048576, type 9 8388608, type 10 8192), the host answers in
2 MiB segments, FW_MEM_READY arrives, the host sends the target
capability request, and the firmware never answers it:
"failed to wait qmi cap request: -110". No SYS_ERR, no RDDM, MHI
stays in mission mode.
The same public 2.9.0.1-02146 boots on a QCN9072 on an IPQ6018 host
(an Arista W-318) with a QMI trace that is identical up to that
point, including the segmented memory answer; there the capability
reply comes at once.
So, the questions:
1. Between FW_MEM_READY and the capability reply, what does the
firmware do that can hang on these cards behind an IPQ8074 host? Is
there a known dependency that the 2.5.r4 and 2.15 builds handle and
the posted 0.1 builds do not?
2. What can I capture to show where it sits? I have forced RDDM dumps
of both cards from the stalled state and can send them; tell me if
another log or MHI channel would show more.
3. If the fix exists in a release branch, can a current QCN9074 hw1.0
build from it go into linux-firmware?
I can run any test build on this hardware the day I get it.
Christopher
The tests and this note were prepared with Claude, an AI coding
assistant.
More information about the ath11k
mailing list