[REGRESSION] 7.3: mt7925e hangs on hibernation restore; fix c182bfe944eb only queued for 7.4

[Илья Шидловский]

Hi Felix, Johannes,

commit c4edf9e3e0e5 ("wifi: mt76: mt7925: replace shared rx irq masks
with per-chip definitions"), merged for v7.3-rc1, breaks resume from
hibernation on MT7925 (PCIe). After the image is restored, the chip
reset path ends up on the MCU_WA NAPI, which is never set up on
MT7925, and the machine hangs hard (no console output, only a power-off
helps). Bert Karwatzki reported the same problem in August.

The fix is already applied:

c182bfe944eb ("wifi: mt76: mt7925: fix reset panic caused by the WM2
rx irq mask")

but only in wireless-next. As things stand, v7.3 will be released with
this regression and the fix will only arrive with 7.4-rc1. It is a small
fix for a regression introduced in this cycle and already carries a
Fixes: tag. Could it go to Linus through the wireless tree for v7.3? If
that is too late, please make sure it is picked up for 7.3.y stable.

Hardware:
Gigabyte Z890M AORUS ELITE WIFI7 ICE, BIOS F21
MT7925 [14c3:7925], subsystem [105b:e112], ASIC revision 79250000
Ubuntu 26.10 kernel 7.3.0-9-generic (based on v7.3-rc6)

Without the fix, every hibernation restore hangs. I narrowed it down by
unbinding groups of drivers before hibernating: only mt7925e triggers
it, and the 7.0 kernel is fine on the same machine. With c182bfe944eb
applied on top of 7.3.0-9 (built as an out-of-tree module), hibernation
and resume work again, including resume after booting another OS in
between.

Tested-by: Ilya Shidlovskiy <shidlovskiyilya at gmail.com>

#regzbot introduced: c4edf9e3e0e5
#regzbot fix: c182bfe944eb

Thanks,
Ilya