Skip to contentSkip to conversation

Repeated firmware crash when admin down.

DISCUSSION Ben Greear 2 participants 3 messages Mar 20, 2015 – Mar 24, 2015 Qualcomm Atheros ATH10K wireless driver

lore
Share & tools
lore

Ben Greear

author
#0 lore raw

Someone reported this bug to me. It is repeatable on various firmware, including my own and official QCA firmware.

I am not certain of the hardware platform, but I think it might be a Ventana imx6 board.

Kernel is based on 3.19-mumble.

I cannot reproduce on x86 (but testing on 3.17 kernel and different flavour of Linux, so who knows.)

They simply admin-down wlanX, and then the firmware crashes over and over. They cannot get any register dump.
When they admin up the interface, the crashes stop and system resumes normally.

I had them try it with my firmware and a driver patch
which will attempt to do ping-pong register dump
over pci register reads in case the firmware's normal PCI messaging transport is down, but the register value always reads as '0xdeadbeef'.

Driver patch is linked below, 'val' is 0xdeadbeef after the million read attempts.

(Need a few additional patches to define CT firmware flag and such, but you can
see the logic easily enough.)
http://dmz2.candelatech.com/?p=linux-3.17.dev.y/.git;a=commit;h=52333b889f0c818d2766d0d514ac31e5c53ce0ab

Any ideas what might be the problem?

Thanks,
Ben

Signature · 8 lines
Ben Greear <[email protected]>
Candela Technologies Inc  http://www.candelatech.com


_______________________________________________
ath10k mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/ath10k
#1 lore raw
Quoted 15 lines · Ben Greear

Someone reported this bug to me. It is repeatable on various firmware, including my own and official QCA firmware.

I am not certain of the hardware platform, but I think it might be a Ventana imx6 board.

Kernel is based on 3.19-mumble.

I cannot reproduce on x86 (but testing on 3.17 kernel and different flavour of Linux, so who knows.)

They simply admin-down wlanX, and then the firmware crashes over and over. They cannot get any register dump.
When they admin up the interface, the crashes stop and system resumes normally.

Quoted 6 lines

I had them try it with my firmware and a driver patch
which will attempt to do ping-pong register dump
over pci register reads in case the firmware's normal PCI messaging transport is down, but the register value always reads as '0xdeadbeef'.

Driver patch is linked below, 'val' is 0xdeadbeef after the million read attempts.

If you read MMIO registers while the device sleeps you get 0xdeadbeef. I've recently posted a patch which should fix some of these[1].

I'm guessing the imx6 board could be using shared irq and the line is shared with some other (active) device(s). This would lead to checking registers in ath10k after ath10k_pci_sleep(). In that case my patch should be sufficient.

[1]: https://github.com/kvalo/ath/commit/320e14b8db51a2d635897d521db4e5c79c3a8390

Michał

_______________________________________________
ath10k mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/ath10k

author
#2 lore raw

The user reports this does indeed fix the problem.

Thanks,
Ben

Quoted 36 lines · Michal Kazior
Quoted 15 lines · Ben Greear

Someone reported this bug to me. It is repeatable on various firmware, including my own and official QCA firmware.

I am not certain of the hardware platform, but I think it might be a Ventana imx6 board.

Kernel is based on 3.19-mumble.

I cannot reproduce on x86 (but testing on 3.17 kernel and different flavour of Linux, so who knows.)

They simply admin-down wlanX, and then the firmware crashes over and over. They cannot get any register dump.
When they admin up the interface, the crashes stop and system resumes normally.

Quoted 6 lines

I had them try it with my firmware and a driver patch
which will attempt to do ping-pong register dump
over pci register reads in case the firmware's normal PCI messaging transport is down, but the register value always reads as '0xdeadbeef'.

Driver patch is linked below, 'val' is 0xdeadbeef after the million read attempts.

If you read MMIO registers while the device sleeps you get 0xdeadbeef. I've recently posted a patch which should fix some of these[1].

I'm guessing the imx6 board could be using shared irq and the line is shared with some other (active) device(s). This would lead to checking registers in ath10k after ath10k_pci_sleep(). In that case my patch should be sufficient.

[1]: https://github.com/kvalo/ath/commit/320e14b8db51a2d635897d521db4e5c79c3a8390

Michał

Signature · 7 lines
Ben Greear <[email protected]>
Candela Technologies Inc  http://www.candelatech.com

_______________________________________________
ath10k mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/ath10k