[PATCH v2 1/2] mmc: dw_mmc: fix bug that cause 'Timeout sending command'

[Addy]

On 2015/02/10 23:22, Alim Akhtar wrote:
> Hi Addy,
>
> On Mon, Feb 9, 2015 at 12:55 PM, Addy Ke <addy.ke at rock-chips.com> wrote:
>> Because of some uncertain factors, such as worse card or worse hardware,
>> DAT[3:0](the data lines) may be pulled down by card, and mmc controller
>> will be in busy state. This should not happend when mmc controller
>> send command to update card clocks. If this happends, mci_send_cmd will
>> be failed and we will get 'Timeout sending command', and then system will
>> be blocked. To avoid this, we need reset mmc controller.
>>
>> Signed-off-by: Addy Ke <addy.ke at rock-chips.com>
>> ---
>>   drivers/mmc/host/dw_mmc.c | 28 ++++++++++++++++++++++++++++
>>   1 file changed, 28 insertions(+)
>>
>> diff --git a/drivers/mmc/host/dw_mmc.c b/drivers/mmc/host/dw_mmc.c
>> index 4d2e3c2..b0b57e3 100644
>> --- a/drivers/mmc/host/dw_mmc.c
>> +++ b/drivers/mmc/host/dw_mmc.c
>> @@ -100,6 +100,7 @@ struct idmac_desc {
>>   };
>>   #endif /* CONFIG_MMC_DW_IDMAC */
>>
>> +static int dw_mci_card_busy(struct mmc_host *mmc);
>>   static bool dw_mci_reset(struct dw_mci *host);
>>   static bool dw_mci_ctrl_reset(struct dw_mci *host, u32 reset);
>>
>> @@ -888,6 +889,31 @@ static void mci_send_cmd(struct dw_mci_slot *slot, u32 cmd, u32 arg)
>>                  cmd, arg, cmd_status);
>>   }
>>
>> +static void dw_mci_wait_busy(struct dw_mci_slot *slot)
>> +{
>> +       struct dw_mci *host = slot->host;
>> +       unsigned long timeout = jiffies + msecs_to_jiffies(500);
>> +
> Why 500 msec?
This timeout value is the same as  mci_send_cmd:
static void mci_send_cmd(struct dw_mci_slot *slot, u32 cmd, u32 arg)
{
         struct dw_mci *host = slot->host;
         unsigned long timeout = jiffies + msecs_to_jiffies(500);
         ....
}

I have not clear that which is suitable.
Do you have any suggestion on it?
>
>> +       do {
>> +               if (!dw_mci_card_busy(slot->mmc))
>> +                       return;
>> +               cpu_relax();
>> +       } while (time_before(jiffies, timeout));
>> +
>> +       dev_err(host->dev, "Data busy (status %#x)\n",
>> +               mci_readl(slot->host, STATUS));
>> +
>> +       /*
>> +        * Data busy, this should not happend when mmc controller send command
>> +        * to update card clocks in non-volt-switch state. If it happends, we
>> +        * should reset controller to avoid getting "Timeout sending command".
>> +        */
>> +       dw_mci_ctrl_reset(host, SDMMC_CTRL_ALL_RESET_FLAGS);
>> +
> Why you need to reset all blocks? may be CTRL_RESET is good enough here.
I have tested on rk3288, if only reset ctroller, data busy bit will not 
be cleaned,and we will still get

"Timeout sending command".

>
>> +       /* Fail to reset controller or still data busy, WARN_ON! */
>> +       WARN_ON(dw_mci_card_busy(slot->mmc));
>> +}
>> +
>>   static void dw_mci_setup_bus(struct dw_mci_slot *slot, bool force_clkinit)
>>   {
>>          struct dw_mci *host = slot->host;
>> @@ -899,6 +925,8 @@ static void dw_mci_setup_bus(struct dw_mci_slot *slot, bool force_clkinit)
>>          /* We must continue to set bit 28 in CMD until the change is complete */
>>          if (host->state == STATE_WAITING_CMD11_DONE)
>>                  sdmmc_cmd_bits |= SDMMC_CMD_VOLT_SWITCH;
>> +       else
>> +               dw_mci_wait_busy(slot);
>>
> hmm...I would suggest you to call dw_mci_wait_busy() from inside
> mci_send_cmd(), seems like dw_mmc hangs while sending update clock cmd
> in multiple cases.see [1]
>
> [1]: http://permalink.gmane.org/gmane.linux.kernel.mmc/31140
I think this patch is more reasonable.
So I will resend patches based on this patch.
thank you!
>
>>          if (!clock) {
>>                  mci_writel(host, CLKENA, 0);
>> --
>> 1.8.3.2
>>
>>
>>
>> _______________________________________________
>> linux-arm-kernel mailing list
>> linux-arm-kernel at lists.infradead.org
>> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
>
>