[PATCH] wifi: ath11k: pci: Fix msi_irq crash on driver unload with QCN9074 PCIe WiFi 6 modules

[Vasanthakumar Thiagarajan]

On 4/30/2025 7:34 PM, Balsam Chihi wrote:
> Hello again,
> sorry for the delay.
> i did further debugging and the i found out that in
> "drivers/net/wireless/ath/ath11k/pci.c",
> "static void ath11k_pci_clear_dbg_registers(struct ath11k_base *ab)",
> the "val = ath11k_pcic_read32" calls are causing the kernel crash.
> by removing them the driver is working properly.
> so to answer your questions :
> @Baochen Qiang:
>> is this instruction causing the synchronous external abort? which src line is it pointing to?
> the src line is  "val = ath11k_pcic_read32(ab, PCIE_Q6_COOKIE_ADDR);"
> and any "val = ath11k_pcic_read32" in "static void
> ath11k_pci_clear_dbg_registers(struct ath11k_base *ab)".

Thanks for the information.

Just to confirm, so you are saying it worked when you skipped
val = ath11k_pcic_read32(ab, PCIE_Q6_COOKIE_ADDR)  alone but
still keeping the rest of the register accesses in ath11k_pci_clear_dbg_registers()?
Or you removed some more register reads in that function? May be you can share the
code diff that worked for you.


> @Vasanthakumar Thiagarajan:
>> i) Was this working before for you and broken recently or it has been
>>     like this since you tried first?
> the crash exists since began working on the project with the same hardware.
> also earlier versions of openwrt has the same bug.
> samething for Yocto kirkstone and scarthgap.
> FYI, i use two different QCN9074 chps one from Compex and the other
> from 8devices.
> both chips behave the same device_id=0xFF and they both crash because
> of this bug.
> so had to load different board-2.bin (provided by comprex and
> 8devices) for each WLAN device to have 2.4GHz and 5GHz bands.
> default board-2.bin (provided by linux-firmware) did not resolve the
> kernel crash.
>> ii) Running into same issue during rmmod of ath12k driver instead of
>>      reboot?
> both rmmod and reboot of the ath11k_pci driver cause the crash
>> iii) Is it possible to attach the console log during reboot also?
> rmmod and reboot are the same logs i already shared above

ath12k logs during reboot showing register read values
in ath11k_pci_clear_dbg_registers() will be helpful which
was not there in your log.

Vasanth