Re: [PATCH] mmc: tegra: Write xfer_mode, CMD regs in together

[Ulf Hansson]

On 9 February 2015 at 07:22, Alexandre Courbot <gnurou@xxxxxxxxx> wrote:
> On Thu, Jan 29, 2015 at 1:30 AM, Rhyland Klein <rklein@xxxxxxxxxx> wrote:
>> On 1/28/2015 1:06 AM, Alexandre Courbot wrote:
>>> On Wed, Jan 28, 2015 at 2:23 AM, Rhyland Klein <rklein@xxxxxxxxxx> wrote:
>>>> From: Pavan Kunapuli <pkunapuli@xxxxxxxxxx>
>>>>
>>>> If there is a gap between xfer mode and command register writes,
>>>> tegra SDMMC controller can sometimes issue a spurious command before
>>>> the CMD register is written. To avoid this, these two registers need
>>>> to be written together in a single write operation.
>>>>
>>>> Signed-off-by: Pavan Kunapuli <pkunapuli@xxxxxxxxxx>
>>>> Signed-off-by: Rhyland Klein <rklein@xxxxxxxxxx>
>>>> ---
>>>>  drivers/mmc/host/sdhci-tegra.c |   31 ++++++++++++++++++++++++++++++-
>>>>  1 file changed, 30 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/drivers/mmc/host/sdhci-tegra.c b/drivers/mmc/host/sdhci-tegra.c
>>>> index 59797106af93..3d34de47e57e 100644
>>>> --- a/drivers/mmc/host/sdhci-tegra.c
>>>> +++ b/drivers/mmc/host/sdhci-tegra.c
>>>> @@ -41,6 +41,7 @@
>>>>  #define NVQUIRK_DISABLE_SDR50          BIT(3)
>>>>  #define NVQUIRK_DISABLE_SDR104         BIT(4)
>>>>  #define NVQUIRK_DISABLE_DDR50          BIT(5)
>>>> +#define NVQUIRK_SHADOW_XFER_MODE_REG   BIT(6)
>>>>
>>>>  struct sdhci_tegra_soc_data {
>>>>         const struct sdhci_pltfm_data *pdata;
>>>> @@ -67,6 +68,32 @@ static u16 tegra_sdhci_readw(struct sdhci_host *host, int reg)
>>>>         return readw(host->ioaddr + reg);
>>>>  }
>>>>
>>>> +static void tegra_sdhci_writew(struct sdhci_host *host, u16 val, int reg)
>>>> +{
>>>> +       struct sdhci_pltfm_host *pltfm_host = sdhci_priv(host);
>>>> +       struct sdhci_tegra *tegra_host = pltfm_host->priv;
>>>> +       const struct sdhci_tegra_soc_data *soc_data = tegra_host->soc_data;
>>>> +
>>>> +       if (soc_data->nvquirks * NVQUIRK_SHADOW_XFER_MODE_REG) {
>>>
>>> Isn't the '*' supposed to be a '&' here?
>>
>> Yah .. not sure how that happened, but it should be '&' good catch.
>>
>>>
>>>> +               switch (reg) {
>>>> +               case SDHCI_TRANSFER_MODE:
>>>> +                       /*
>>>> +                        * Postpone this write, we must do it together with a
>>>> +                        * command write that is down below.
>>>> +                        */
>>>> +                       pltfm_host->xfer_mode_shadow = val;
>>>> +                       return;
>>>> +               case SDHCI_COMMAND:
>>>> +                       writel((val << 16) | pltfm_host->xfer_mode_shadow,
>>>> +                               host->ioaddr + SDHCI_TRANSFER_MODE);
>>>> +                       pltfm_host->xfer_mode_shadow = 0;
>>>
>>> That last line is probably not needed and could actually be harmful -
>>> if we try to write SDHCI_COMMAND twice in a raw without a write to
>>> SDHCI_TRANSFER_MODE in between, the zero will overwrite the previous
>>> value of SDHCI_TRANSFER_MODE.
>>
>> True, will remove it.
>>
>>>
>>>> +                       return;
>>>> +               }
>>>> +       }
>>>> +
>>>> +       writew(val, host->ioaddr + reg);
>>>> +}
>>>> +
>>>>  static void tegra_sdhci_writel(struct sdhci_host *host, u32 val, int reg)
>>>>  {
>>>>         struct sdhci_pltfm_host *pltfm_host = sdhci_priv(host);
>>>> @@ -147,6 +174,7 @@ static void tegra_sdhci_set_bus_width(struct sdhci_host *host, int bus_width)
>>>>  static const struct sdhci_ops tegra_sdhci_ops = {
>>>>         .get_ro     = tegra_sdhci_get_ro,
>>>>         .read_w     = tegra_sdhci_readw,
>>>> +       .write_w    = tegra_sdhci_writew,
>>>>         .write_l    = tegra_sdhci_writel,
>>>>         .set_clock  = sdhci_set_clock,
>>>>         .set_bus_width = tegra_sdhci_set_bus_width,
>>>> @@ -201,7 +229,8 @@ static struct sdhci_tegra_soc_data soc_data_tegra114 = {
>>>>         .pdata = &sdhci_tegra114_pdata,
>>>>         .nvquirks = NVQUIRK_DISABLE_SDR50 |
>>>>                     NVQUIRK_DISABLE_DDR50 |
>>>> -                   NVQUIRK_DISABLE_SDR104,
>>>> +                   NVQUIRK_DISABLE_SDR104 |
>>>> +                   NVQUIRK_SHADOW_XFER_MODE_REG,
>>>>  };
>>>
>>> Since this only applies to Tegra114 (?), I wonder whether it would not
>>> be better to have a dedicated tegra114_sdhci_ops that implements
>>> tegra_sdhci_writew, and use it only in tegra_sdhci_writew. That way
>>> you could get rid of the NVQUIRK_SHADOW_XFER_MODE_REG and the test for
>>> it in tegra_sdhci_writew(), and chips prior to Tegra114 will not have
>>> to needlessly check for it every time they write a register.
>>
>> The reason I did it this way, is that this doesn't explicitly just apply
>> to T114. It actually applies to T114, T124 and T132. In that case, I
>> think it makes sense to keep the QUIRK and I can update the commit
>> description to reflect that.
>
> All the same, I don't see what advantage we have checking for that
> condition for every single write while we could simply set the right
> function to use at probe time?

The above seems like a reasonable optimization to do!

Now, since $subject patch has been applied, I suggest you send a new
patch addressing this issue.

Kind regards
Uffe
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/