Freescale i.mx25 SDIO/Libertas issue

[Daniel Mack]

On Mon, May 30, 2011 at 3:52 AM, Andre Renaud <andre at bluewatersys.com> wrote:
> Hello,
> I am trying to use a Wi2Wi 802.11b/g (W2CBW009S) module attached to an
> i.MX25 cpu, specifically the i.MX25 PDK from freescale, although I've
> also tried it on a DeviceSolutions Topaz i.MX25. However it is unable to
> fully load the driver, leaving me with the following in the logs:
> [   30.149097] libertas_sdio mmc0:0001:1: (unregistered net_device):
> command 0x00a9 timed out
> [   30.153759] libertas_sdio mmc0:0001:1: (unregistered net_device):
> Timeout submitting command 0x00a9
> [   30.158843] libertas_sdio mmc0:0001:1: (unregistered net_device):
> PREP_CMD: command 0x00a9 failed: -110
> [   30.164126] libertas_sdio mmc0:0001:1: (unregistered net_device):
> CMD_FUNC_INIT cmd failed
> This wifi chip is based around the Marvell 8688 core.
>
> I have also tried using an 8686 device, and in that case it is unable to
> even complete loading the firmware, failing with the following:
> [   14.753132] libertas_sdio: Firmware req_size: 17
> [   14.755713] libertas_sdio: failed to load firmware: -5
> [   14.759272] libertas_sdio: probe of mmc0:0001:1 failed with error -5
> Note: I added the 'req_size: ' debugging message in

I assume this error happens after the driver uploaded the helper
firmware and switched to multi-block transfers?

Which kernel version are you working on? I'm asking because I fixed a
similar bug some months ago which I happend to see an a MX31 CPU with
a 8686 attach to the SDIO bus. The MX3 MMC controller has bug severe
bug in its CRC calculation code which causes multi-block transfers on
a 4-bit bus to fail, and hence the only solution was to switch to a
1-bit bus. But I don't known whether this also affects MX25 cores.

Anyway - this workaround was merged more than a year ago. See if you
have commit 1eb46d ("libertas/sdio: 8686: set ECSI bit for 1-bit
transfers") in your tree.


Daniel