libertas sdio on bf548 w/ uclinux

[Nick Moszer]

Pierre Ossman wrote:
> On Tue, 22 Apr 2008 14:28:36 -0400
> Dan Williams <dcbw at redhat.com> wrote:
>
>   
>> On Tue, 2008-04-22 at 13:04 -0500, Nick Moszer wrote:
>>     
>>> As a follow up to this, here is the libertas debug info right before it 
>>> bombs.
>>>       
>> Probably need to get Pierre's input here...  I'm not sure how the SDIO
>> stack is supposed to interact with controllers that have certain
>> padding/alignment constraints.
>>
>> Pierre?
>>     
>
> What a delightful mess of top posting...
>
> As for controllers with certain padding/alignment constraints, the
> answer is simply tough luck, get some other hardware that isn't broken.
> The MMC/SD/SDIO specs have mandated varying transfer sizes since day
> one, so hardware that cannot support that doesn't deserve to be called
> a MMC/SD/SDIO controller.
>
> The controller shouldn't BUG though as such requests are perfectly
> valid. It should fail them with -EINVAL, indicating that the hardware
> is unable to satisfy the request.
>
> And yes, the behaviour of the libertas card is consistent with what
> I've been seeing. My guess is that it transfers one chunk that tells it
> how much data that follows.
>
> Rgds
> Pierre
>   
Wondering if I could ask one more question of you gentlemen.

This is VERY close to working now.  I added a workaround for this 
blackfin board where all the data passed it bumped up to the nearest
power of 2. I'm able to load the libertas and libertas_sdio modules just 
fine.  I get a ethX device and am able to associate with an AP.
I can throw iwconfig a hundred times and it works, so basic 
communication with the chip appears to be working.

Sometimes it will get an IP through DHCP, when it does once I try to 
send data through the wifi (ping, ssh) it will sometimes make it one or
two pings in, sometimes 5 or 6 and then die.

Output from libertas debug when it fails to get an IP:
libertas enter: handle_cmd_response():547
libertas leave: handle_cmd_response():712
libertas enter: cleanup_cmdnode():1585
libertas leave: cleanup_cmdnode():1599
libertas leave: libertas_process_rx_command():865, ret 0
libertas host: EXEC_NEXT_CMD: sending command 0x0028
libertas enter: DownloadcommandToStation():981
libertas host: DNLD_CMD: command 0x0028, size 12, jiffies 42721
libertas enter: if_sdio_host_to_card(type 1, bytes 12):726
libertas leave: if_sdio_host_to_card():800, ret 0
libertas cmd: DNLD_CMD: sent command 0x0028, jiffies 42721
libertas leave: DownloadcommandToStation():1037, ret 0
libertas enter: if_sdio_host_to_card_worker():379
libertas leave: if_sdio_host_to_card_worker():419
libertas enter: if_sdio_interrupt():846
libertas thread: interrupt: 0x2
libertas leave: if_sdio_interrupt():880, ret 0
libertas enter (INT): if_sdio_host_to_card(type 0, bytes 614):726
libertas leave (INT): if_sdio_host_to_card():800, ret 0
libertas enter: if_sdio_host_to_card_worker():379
libertas leave: if_sdio_host_to_card_worker():419
libertas enter: if_sdio_interrupt():846
libertas thread: interrupt: 0x1
libertas enter: if_sdio_card_to_host():273
libertas fw (INT): command_timer_fn fired, cmd 28
libertas fw (INT): re-sending same command because of timeout
libertas enter (INT): libertas_queue_cmd():924
libertas host (INT): QUEUE_CMD: inserted command 0x0028 into cmdpendingq
libertas leave (INT): libertas_queue_cmd():961
NETDEV WATCHDOG: eth1: transmit timed out
libertas: tx watch dog timeout
libertas enter (INT): if_sdio_host_to_card(type 0, bytes 614):726
libertas leave (INT): if_sdio_host_to_card():800, ret 0
libertas enter: if_sdio_host_to_card_worker():379
NETDEV WATCHDOG: eth1: transmit timed out
libertas: tx watch dog timeout
libertas enter (INT): if_sdio_host_to_card(type 0, bytes 614):726
libertas leave (INT): if_sdio_host_to_card():800, ret 0
NETDEV WATCHDOG: eth1: transmit timed out
libertas: tx watch dog timeout
libertas enter (INT): if_sdio_host_to_card(type 0, bytes 614):726
libertas leave (INT): if_sdio_host_to_card():800, ret 0
NETDEV WATCHDOG: eth1: transmit timed out
libertas: tx watch dog timeout
libertas enter (INT): if_sdio_host_to_card(type 0, bytes 614):726
libertas leave (INT): if_sdio_host_to_card():800, ret 0
NETDEV WATCHDOG: eth1: transmit timed out
libertas: tx watch dog timeout


In this instance it seems like it failed on a CMD_MAC_CONTROL but I've
also seen it fail on a CMD_MAC_MULTICAST_ADR and a CMD_802_11_GET_LOG so
I don't think it's that particular command.


Looking at scope output it APPEARS the SDIO_D1 pin is being held low by 
the Marvell chip, when I reset the SDH controller on the blackfin the
pin is still low. It won't go back high until the card is reset.  
Turning on mmc debugging it appears the controller is still trying to 
send commands.


Another thing of interest is in the status register of the blackfin SDH 
controller I get a "missing start bit" flag set when this fails.


So, my thought is that the firmware on the Marvell chip is going dumb 
and dies.  Which kind of makes this not easy to fix.  Does anyone else
have an opinion?  I'm hoping it's something I'm overlooking and not the 
firmware.

Thanks!