Re: [PATCH v2] mmc: mmci: manage MMC_PM_KEEP_POWER per variant config

[Yann Gautier <yann.gautier@foss.st.com>]

On 3/15/22 10:07, Ulf Hansson wrote:
> On Mon, 14 Mar 2022 at 17:52, Yann Gautier <yann.gautier@foss.st.com> wrote:
>>
>> On 3/14/22 17:18, Ulf Hansson wrote:
>>> On Mon, 14 Mar 2022 at 15:34, Yann Gautier <yann.gautier@foss.st.com> wrote:
>>>>
>>>> On 3/14/22 14:03, Ulf Hansson wrote:
>>>>> On Mon, 14 Mar 2022 at 13:56, Yann Gautier <yann.gautier@foss.st.com> wrote:
>>>>>>
>>>>>> Add a disable_keep_power field in variant_data struct. The
>>>>>> MMC_PM_KEEP_POWER flag will be enabled if disable_keep_power is not set.
>>>>>> It is only set to true for stm32_sdmmc variants.
>>>>>>
>>>>>> The issue was seen on STM32MP157C-DK2 board, which embeds a wifi chip.
>>>>>> It doesn't correctly support low power on this board. The Wifi chip
>>>>>> awaits an always-on regulator, but it was connected to v3v3 which is off
>>>>>> in low-power sequence. MMC_PM_KEEP_POWER should then be disabled.
>>>>>
>>>>> Just to make sure I get this correct.
>>>>>
>>>>> Why can't the regulator stay on during system suspend? The point is,
>>>>> we don't need an always on regulator to cope with this.
>>>>>
>>>>> Kind regards
>>>>> Uffe
>>>>
>>>> Hi Ulf,
>>>>
>>>> This v3v3 regulator powers most of the devices on this board. So we need
>>>> to switch it off to gain power in suspend mode.
>>>
>>> I see. Thanks for sharing that information.
>>>
>>> The MMC_PM_KEEP_POWER flag is there to describe what is supported by
>>> the platform/host. It doesn't mean that the card *must* stay powered
>>> on during system suspend. Instead that depends on whether system
>>> wakeup for the SDIO/WiFi is supported too - and if that is enabled by
>>> userspace. If not, the regulator will be turned off for the SDIO card
>>> during system suspend.
>>>
>>> Assuming the regulator is implemented as a proper regulator and can
>>> remain on during system suspend, the right thing would be to keep the
>>> MMC_PM_KEEP_POWER flag around.
>>>
>>> Kind regards
>>> Uffe
>>>
>>
>> OK, but in the wifi driver we use on this platform (brcmfmac), the
>> suspend/resume functions (brcmf_ops_sdio_suspend/brcmf_ops_sdio_resume)
>> use the flag to check regu was off, and then call probe function during
>> resume, to re-init Wifi chip and reload its firmware.
> 
> I had a closer look at the brcmfmac driver, thanks for the pointers.
> 
> In my opinion, I think we should change the brcmfmac driver, so it
> decides to power off the SDIO card, unless the WiFi chip is configured
> to serve us with system wakeups.
> 
> I can send a patch for brcmfmac that we can try, unless you want to
> send it yourself?
> 

Hi Ulf,
If you already have an idea of the patch in the brcmfmac driver, can you 
propose something?
We'll be able to test it at our side.

Thanks,
Yann

>>
>>
>> Best regards,
>> Yann
> 
> Kind regards
> Uffe
> >>
>>>>
>>>>
>>>> Yann
>>>>
>>>>>
>>>>>>
>>>>>> The flag can still be enabled through DT property:
>>>>>> keep-power-in-suspend.
>>>>>>
>>>>>> Signed-off-by: Yann Gautier <yann.gautier@foss.st.com>
>>>>>> ---
>>>>>> Update in v2:
>>>>>> Reword commit message to better explain the issue.
>>>>>>
>>>>>> Resend the patch alone. It was previoulsy in a series [1] for which the
>>>>>> other patches will be reworked.
>>>>>>
>>>>>> [1] https://lore.kernel.org/lkml/20220304135134.47827-1-yann.gautier@foss.st.com/
>>>>>>
>>>>>>     drivers/mmc/host/mmci.c | 5 ++++-
>>>>>>     drivers/mmc/host/mmci.h | 1 +
>>>>>>     2 files changed, 5 insertions(+), 1 deletion(-)
>>>>>>
>>>>>> diff --git a/drivers/mmc/host/mmci.c b/drivers/mmc/host/mmci.c
>>>>>> index 45b8608c935c..0e2f2f5d6a52 100644
>>>>>> --- a/drivers/mmc/host/mmci.c
>>>>>> +++ b/drivers/mmc/host/mmci.c
>>>>>> @@ -274,6 +274,7 @@ static struct variant_data variant_stm32_sdmmc = {
>>>>>>            .busy_detect            = true,
>>>>>>            .busy_detect_flag       = MCI_STM32_BUSYD0,
>>>>>>            .busy_detect_mask       = MCI_STM32_BUSYD0ENDMASK,
>>>>>> +       .disable_keep_power     = true,
>>>>>>            .init                   = sdmmc_variant_init,
>>>>>>     };
>>>>>>
>>>>>> @@ -301,6 +302,7 @@ static struct variant_data variant_stm32_sdmmcv2 = {
>>>>>>            .busy_detect            = true,
>>>>>>            .busy_detect_flag       = MCI_STM32_BUSYD0,
>>>>>>            .busy_detect_mask       = MCI_STM32_BUSYD0ENDMASK,
>>>>>> +       .disable_keep_power     = true,
>>>>>>            .init                   = sdmmc_variant_init,
>>>>>>     };
>>>>>>
>>>>>> @@ -2172,7 +2174,8 @@ static int mmci_probe(struct amba_device *dev,
>>>>>>            host->stop_abort.flags = MMC_RSP_R1B | MMC_CMD_AC;
>>>>>>
>>>>>>            /* We support these PM capabilities. */
>>>>>> -       mmc->pm_caps |= MMC_PM_KEEP_POWER;
>>>>>> +       if (!variant->disable_keep_power)
>>>>>> +               mmc->pm_caps |= MMC_PM_KEEP_POWER;
>>>>>>
>>>>>>            /*
>>>>>>             * We can do SGIO
>>>>>> diff --git a/drivers/mmc/host/mmci.h b/drivers/mmc/host/mmci.h
>>>>>> index e1a9b96a3396..2cad1ef9766a 100644
>>>>>> --- a/drivers/mmc/host/mmci.h
>>>>>> +++ b/drivers/mmc/host/mmci.h
>>>>>> @@ -361,6 +361,7 @@ struct variant_data {
>>>>>>            u32                     opendrain;
>>>>>>            u8                      dma_lli:1;
>>>>>>            u32                     stm32_idmabsize_mask;
>>>>>> +       u8                      disable_keep_power:1;
>>>>>>            void (*init)(struct mmci_host *host);
>>>>>>     };
>>>>>>
>>>>>> --
>>>>>> 2.25.1
>>>>>>
>>>>
>>