Troubles with ath10k and QCA9882 Chipset

[Michal Kazior]

On 4 April 2016 at 14:53, Stefan Feirer <e0926020 at student.tuwien.ac.at> wrote:
> Hello,
>
> I'm trying to setup the PCIe Wifi Card Compex WLE600VX with QCA9882 Chipset
> and ath10k_pci kernel modul on an evaluation board (evaluation board
> var-som-mx6 from variscite).
> Therefore I compiled a linux embedded version with yocto and kernel version
> 3.14. The ath10k modules are backported.
> When I load the kernel modul ath10k_pci I get the following dmesg output:
[...]
> The output of dmesg with ath10k_pci irq_mode=1 and ath10k_core
> debug_mask=0x00000432 shows:
>
[...]
> [ 1626.991366] ath10k_pci 0000:01:00.0: bmi done
> [ 1626.991390] ath10k_pci 0000:01:00.0: htt tx max num pending tx 1424
> [ 1626.991420] ath10k_pci 0000:01:00.0: htt rx ring size 2048 fill_level
> 1023
> [ 1626.991428] ath10k_pci 0000:01:00.0: boot hif start
> [ 1627.993782] ath10k_pci 0000:01:00.0: failed to receive control response
> completion, polling..
> [ 1628.001136] ath10k_pci 0000:01:00.0: boot htc service 'Control' ul pipe 0
> dl pipe 1 eid 0 ready
> [ 1628.001157] ath10k_pci 0000:01:00.0: boot htc ep 0 ul polled 0 dl polled
> 0
> [ 1628.001175] ath10k_pci 0000:01:00.0: boot htc service 'Control' eid 0 TX
> flow control disabled
> [ 1628.001194] ath10k_pci 0000:01:00.0: boot htc service HTT Data does not
> allocate target credits
> [ 1628.993808] ath10k_pci 0000:01:00.0: Service connect timeout: 0
> [ 1628.998453] ath10k_pci 0000:01:00.0: failed to connect htt (-110)

Interesting. The success (htc control endpoint 0 being ready) after
failure to receive a completion.. Apparently interrupts aren't
delivered properly as a later attempt to wait for a completion times
out again.

Originally the polling was added to workaround weird virtual machine
pci-passthrough problem I was seeing (probably some race outside of
ath10k) that happened only for the initial device kick-off.

As for the actual reason why interrupts aren't delivered on your
system - no idea. Maybe the platform isn't brought up properly (i.e.
kernel/core issue doesn't properly unmask interrupts?) or ath10k lacks
some small bit in the pci probing that isn't necessary on other
platforms to work hence was missed.


Michał