8686 with SDIO on AT91SAM9263

[Hans-Jürgen Koch]

Am Thu, 24 Jan 2008 08:12:18 +0100
schrieb Marc Pignat <marc.pignat at hevs.ch>:

> Hi!

Hi Marc, lots of thanks for your help!

(added Pierre Ossman and Nicolas Ferre to CC)

> 
> On Thursday 24 January 2008, Dan Williams wrote:
> > On Wed, 2008-01-23 at 19:20 +0100, Hans J. Koch wrote:
> > > I made some progress in my attempts to make a Marvell 8686 work
> > > on an AT91SAM9263 board. I use latest netdev-2.6.25 plus a patch
> > > that solves the modulo-4-problem of the AT91 MCI driver. That
> > > patch was recently posted by Marc Pignat on linux-arm-kernel ML.
> > > 
> > > ATM, firmware is downloaded without problems. But after that, it
> > > hangs somewhere in libertas core code. The last thing I see is:
> > > 
> > > libertas: 00:19:88:05:1c:53, fw 8.73.7p3, cap 0x00000393
> > > libertas: Command 28 timed out
> > > libertas: requeueing command 28 due to timeout (#1)
> > > 
> > > This is somewhere in lbs_setup_firmware(), called from
> > > lbs_start_card() in libertas/main.c.
> > > 
> > > Could somebody tell me what the above messages mean?
> > > Are there any other users of similar hardware out there?
> > 
> > It means that the firmware did not respond to the command in the
> > time the driver expected it to.  Command 0x0028 is CMD_MAC_CONTROL,
> > which is probably coming from lbs_set_mac_packet_filter(), which is
> > called from libertas_setup_firmware().
> Can you try this patch?
> If it says something like "mmc0: Timeout waiting end of packet", then
> this is the mci controller state machine that has crashed, and not
> the 8686.

Yes, it seems this is what happens:

libertas: 00:19:88:05:1c:53, fw 8.73.7p3, cap 0x00000393
mmc0: Timeout waiting end of packet
libertas: Command 28 timed out
libertas: requeueing command 28 due to timeout (#1)

So, it seems to be a problem in at91_mci. Anyway, I'd like to consider
this a bug in libertas as well. From the code, I think the intention
was to reqeue a failed command, and abort after a maximum retry count.
This doesn't happen, ATM I have to kill modprobe manually...

Pierre, Nicolas could you shed some light on this?

Thanks,
Hans

(I'm leaving the patch below for easy reference)

> 
> Regards
> 
> Marc
> 
> patch-at91-mci-1-timeout:
> 
> Detect card or host timeout.
> 
> Signed-off-by: Marc Pignat <marc.pignat at hevs.ch>
> ---
> 
> 
> --- a/drivers/mmc/host/at91_mci.c	2008-01-23
> 15:47:58.000000000 +0100 +++ b/drivers/mmc/host/at91_mci.c
> 2008-01-23 16:35:10.000000000 +0100 @@ -124,8 +124,34 @@ struct
> at91mci_host 
>  	/* Latest in the scatterlist that has been enabled for
> transfer */ int transfer_index;
> +
> +	/* Timer for timeouts */
> +	struct timer_list timer;
>  };
>  
> +static void at91_timeout_timer(unsigned long data)
> +{
> +	struct at91mci_host *host;
> +
> +	host = (struct at91mci_host *)data;
> +
> +	if (host->request) {
> +		printk(KERN_ERR "%s: Timeout waiting end of
> packet\n",
> +			mmc_hostname(host->mmc));
> +
> +		if (host->cmd && host->cmd->data) {
> +			host->cmd->data->error = -ETIMEDOUT;
> +		} else {
> +			if (host->cmd)
> +				host->cmd->error = -ETIMEDOUT;
> +			else
> +				host->request->cmd->error =
> -ETIMEDOUT;
> +		}
> +
> +		mmc_request_done(host->mmc, host->request);
> +	}
> +}
> +
>  /*
>   * Copy from sg to a dma block - used for transfers
>   */
> @@ -551,9 +577,10 @@ static void at91_mci_process_next(struct
>  	else if ((!(host->flags & FL_SENT_STOP)) &&
> host->request->stop) { host->flags |= FL_SENT_STOP;
>  		at91_mci_send_command(host, host->request->stop);
> -	}
> -	else
> +	} else {
> +		del_timer(&host->timer);
>  		mmc_request_done(host->mmc, host->request);
> +	}
>  }
>  
>  /*
> @@ -612,6 +639,8 @@ static void at91_mci_request(struct mmc_
>  	host->request = mrq;
>  	host->flags = 0;
>  
> +	mod_timer(&host->timer, jiffies +  HZ);
> +
>  	at91_mci_process_next(host);
>  }
>  
> @@ -915,6 +944,8 @@ static int __init at91_mci_probe(struct 
>  
>  	mmc_add_host(mmc);
>  
> +	setup_timer(&host->timer, at91_timeout_timer, (unsigned
> long)host); +
>  	/*
>  	 * monitor card insertion/removal if we can
>  	 */