Re: [PATCH RFC 1/2] mmc: sdhci: Allow platform controlled voltage switching

[Adrian Hunter]

On 17/07/18 08:14, Vijay Viswanath wrote:
> 
> 
> On 7/10/2018 4:37 PM, Adrian Hunter wrote:
>> On 21/06/18 15:23, Vijay Viswanath wrote:
>>> Some controllers can have internal mechanism to inform the SW that it
>>> is ready for voltage switching. For such controllers, changing voltage
>>> before the HW is ready can result in various issues.
>>>
>>> Add a quirk, which can be used by drivers of such controllers.
>>>
>>> Signed-off-by: Vijay Viswanath <vviswana@xxxxxxxxxxxxxx>
>>> ---
>>> Â drivers/mmc/host/sdhci.c | 20 +++++++++++++++-----
>>> Â drivers/mmc/host/sdhci.h |Â 2 ++
>>> Â 2 files changed, 17 insertions(+), 5 deletions(-)
>>>
>>> diff --git a/drivers/mmc/host/sdhci.c b/drivers/mmc/host/sdhci.c
>>> index 1c828e0..f0346d4 100644
>>> --- a/drivers/mmc/host/sdhci.c
>>> +++ b/drivers/mmc/host/sdhci.c
>>> @@ -1615,7 +1615,8 @@ void sdhci_set_power_noreg(struct sdhci_host *host,
>>> unsigned char mode,
>>> Â void sdhci_set_power(struct sdhci_host *host, unsigned char mode,
>>> ÂÂÂÂÂÂÂÂÂÂÂÂÂÂ unsigned short vdd)
>>> Â {
>>> -ÂÂÂ if (IS_ERR(host->mmc->supply.vmmc))
>>> +ÂÂÂ if (IS_ERR(host->mmc->supply.vmmc) ||
>>> +ÂÂÂÂÂÂÂÂÂÂÂ (host->quirks2 & SDHCI_QUIRK2_INTERNAL_PWR_CTL))
>>
>> I think you should provide your own ->set_power() instead of this
>>
> 
> will do
> 
>>> ÂÂÂÂÂÂÂÂÂ sdhci_set_power_noreg(host, mode, vdd);
>>> ÂÂÂÂÂ else
>>> ÂÂÂÂÂÂÂÂÂ sdhci_set_power_reg(host, mode, vdd);
>>> @@ -2009,7 +2010,9 @@ int sdhci_start_signal_voltage_switch(struct
>>> mmc_host *mmc,
>>> ÂÂÂÂÂÂÂÂÂ ctrl &= ~SDHCI_CTRL_VDD_180;
>>> ÂÂÂÂÂÂÂÂÂ sdhci_writew(host, ctrl, SDHCI_HOST_CONTROL2);
>>> Â -ÂÂÂÂÂÂÂ if (!IS_ERR(mmc->supply.vqmmc)) {
>>> +ÂÂÂÂÂÂÂ if (!IS_ERR(mmc->supply.vqmmc) &&
>>> +ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ !(host->quirks2 &
>>> +ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ SDHCI_QUIRK2_INTERNAL_PWR_CTL)) {
>>
>> And your own ->start_signal_voltage_switch()
>>
> 
> sdhci_msm_start_signal_voltage_switch() would be an exact copy of
> sdhci_start_signal_voltage_switch()..... will incorporate this if not using
> quirk.
> 
>>> ÂÂÂÂÂÂÂÂÂÂÂÂÂ ret = mmc_regulator_set_vqmmc(mmc, ios);
>>> ÂÂÂÂÂÂÂÂÂÂÂÂÂ if (ret) {
>>> ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ pr_warn("%s: Switching to 3.3V signalling voltage
>>> failed\n",
>>> @@ -2032,7 +2035,8 @@ int sdhci_start_signal_voltage_switch(struct
>>> mmc_host *mmc,
>>> ÂÂÂÂÂ case MMC_SIGNAL_VOLTAGE_180:
>>> ÂÂÂÂÂÂÂÂÂ if (!(host->flags & SDHCI_SIGNALING_180))
>>> ÂÂÂÂÂÂÂÂÂÂÂÂÂ return -EINVAL;
>>> -ÂÂÂÂÂÂÂ if (!IS_ERR(mmc->supply.vqmmc)) {
>>> +ÂÂÂÂÂÂÂ if (!IS_ERR(mmc->supply.vqmmc) &&
>>> +ÂÂÂÂÂÂÂÂÂÂÂ !(host->quirks2 & SDHCI_QUIRK2_INTERNAL_PWR_CTL)) {
>>> ÂÂÂÂÂÂÂÂÂÂÂÂÂ ret = mmc_regulator_set_vqmmc(mmc, ios);
>>> ÂÂÂÂÂÂÂÂÂÂÂÂÂ if (ret) {
>>> ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ pr_warn("%s: Switching to 1.8V signalling voltage
>>> failed\n",
>>> @@ -3485,7 +3489,10 @@ int sdhci_setup_host(struct sdhci_host *host)
>>> ÂÂÂÂÂÂ * the host can take the appropriate action if regulators are not
>>> ÂÂÂÂÂÂ * available.
>>> ÂÂÂÂÂÂ */
>>> -ÂÂÂ ret = mmc_regulator_get_supply(mmc);
>>> +ÂÂÂ if (!(host->quirks2 & SDHCI_QUIRK2_INTERNAL_PWR_CTL))
>>
>> Since we expect mmc_regulator_get_supply() to have been called, this could
>> be:
>>
>> ÂÂÂÂif (!mmc->supply.vmmc) {
>> ÂÂÂÂÂÂÂ ret = mmc_regulator_get_supply(mmc);
>> ÂÂÂÂÂÂÂ enable_vqmmc = true;
>> ÂÂÂÂ} else {
>> ÂÂÂÂÂÂÂ ret = 0;
>> ÂÂÂÂ}
>> >> +ÂÂÂÂÂÂÂ ret = mmc_regulator_get_supply(mmc);
>>> +ÂÂÂ else
>>> +ÂÂÂÂÂÂÂ ret = 0;
>>> ÂÂÂÂÂ if (ret)
>>> ÂÂÂÂÂÂÂÂÂ return ret;
>>> Â @@ -3736,7 +3743,10 @@ int sdhci_setup_host(struct sdhci_host *host)
>>> Â ÂÂÂÂÂ /* If vqmmc regulator and no 1.8V signalling, then there's no UHS */
>>> ÂÂÂÂÂ if (!IS_ERR(mmc->supply.vqmmc)) {
>>> -ÂÂÂÂÂÂÂ ret = regulator_enable(mmc->supply.vqmmc);
>>> +ÂÂÂÂÂÂÂ if (!(host->quirks2 & SDHCI_QUIRK2_INTERNAL_PWR_CTL))
>>
>> And this could be:
>>
>> ÂÂÂÂÂÂÂ if (enable_vqmmc)
>> ÂÂÂÂÂÂÂÂÂÂÂ ret = regulator_enable(mmc->supply.vqmmc);
>> ÂÂÂÂÂÂÂ else
>> ÂÂÂÂÂÂÂÂÂÂÂ ret = 0;
>> Â> However, you still need to ensure regulator_disable(mmc->supply.vqmmc) is
>> only called if regulator_enable() was called.
> I missed this. Will cover it.
> 
> Also I missed one more place where we are doing regulator_disable. During
> sdhci-msm unbinding, we would end up doing an extra regulator disable
> (thanks Evan for pointing it out) in sdhci_remove_host.
> 
> To avoid the quirk( or having any flag), it would require copying the code
> of sdhci_start_signal_voltage_switch() and sdhci_remove_host() and creating

You do not need to duplicate sdhci_remove_host(), just change it so that it
only  disables what was enabled i.e.

	if (host->vqmmc_enabled)
		regulator_disable(mmc->supply.vqmmc);

> 2 new functions in sdhci_msm layer which would do the exact same as above,
> with just the regulator parts removed.
> 
> This looks messy (considering any future changes to the 2 sdhci API will
> need to be copied to their duplicate sdhci_msm API) and a bit overkill to
> avoid quirk. At the same time, I don't know how useful such a quirk would be
> to other platform drivers.
> 
> Please let me know your view/suggestions.

Let's try without the quirk.

>>
>>> +ÂÂÂÂÂÂÂÂÂÂÂ ret = regulator_enable(mmc->supply.vqmmc);
>>> +ÂÂÂÂÂÂÂ else
>>> +ÂÂÂÂÂÂÂÂÂÂÂ ret = 0;
>>> ÂÂÂÂÂÂÂÂÂ if (!regulator_is_supported_voltage(mmc->supply.vqmmc, 1700000,
>>> ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ 1950000))
>>> ÂÂÂÂÂÂÂÂÂÂÂÂÂ host->caps1 &= ~(SDHCI_SUPPORT_SDR104 |
>>> diff --git a/drivers/mmc/host/sdhci.h b/drivers/mmc/host/sdhci.h
>>> index 23966f8..3b0c97a 100644
>>> --- a/drivers/mmc/host/sdhci.h
>>> +++ b/drivers/mmc/host/sdhci.h
>>> @@ -450,6 +450,8 @@ struct sdhci_host {
>>> ÂÂ * obtainable timeout.
>>> ÂÂ */
>>> Â #define SDHCI_QUIRK2_DISABLE_HW_TIMEOUTÂÂÂÂÂÂÂÂÂÂÂ (1<<17)
>>> +/* Regulator voltage changes are being done from platform layer */
>>> +#define SDHCI_QUIRK2_INTERNAL_PWR_CTLÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ (1<<18)
>>
>> So maybe the quirk is not needed.
>>
>>> Â ÂÂÂÂÂ int irq;ÂÂÂÂÂÂÂ /* Device IRQ */
>>> ÂÂÂÂÂ void __iomem *ioaddr;ÂÂÂ /* Mapped address */
>>>
>>
> 
> Thanks for the review & suggestions!
> Vijay
>