Re: [PATCH V12 10/23] mmc: sdhci-uhs2: add reset function and uhs2_mode function

From: Ulf Hansson
Date: Wed Oct 04 2023 - 04:36:24 EST


On Tue, 3 Oct 2023 at 17:03, Adrian Hunter <adrian.hunter@xxxxxxxxx> wrote:
>
> On 3/10/23 15:22, Ulf Hansson wrote:
> > On Tue, 3 Oct 2023 at 13:37, Adrian Hunter <adrian.hunter@xxxxxxxxx> wrote:
> >>
> >> On 3/10/23 13:30, Ulf Hansson wrote:
> >>> On Fri, 15 Sept 2023 at 11:44, Victor Shih <victorshihgli@xxxxxxxxx> wrote:
> >>>>
> >>>> From: Victor Shih <victor.shih@xxxxxxxxxxxxxxxxxxx>
> >>>>
> >>>> Sdhci_uhs2_reset() does a UHS-II specific reset operation.
> >>>>
> >>>> Signed-off-by: Ben Chuang <ben.chuang@xxxxxxxxxxxxxxxxxxx>
> >>>> Signed-off-by: AKASHI Takahiro <takahiro.akashi@xxxxxxxxxx>
> >>>> Signed-off-by: Victor Shih <victor.shih@xxxxxxxxxxxxxxxxxxx>
> >>>> Acked-by: Adrian Hunter <adrian.hunter@xxxxxxxxx>
> >>>> ---
> >>>>
> >>>> Updates in V8:
> >>>> - Adjust the position of matching brackets.
> >>>>
> >>>> Updates in V6:
> >>>> - Remove unnecessary functions and simplify code.
> >>>>
> >>>> ---
> >>>>
> >>>> drivers/mmc/host/sdhci-uhs2.c | 45 +++++++++++++++++++++++++++++++++++
> >>>> drivers/mmc/host/sdhci-uhs2.h | 2 ++
> >>>> 2 files changed, 47 insertions(+)
> >>>>
> >>>> diff --git a/drivers/mmc/host/sdhci-uhs2.c b/drivers/mmc/host/sdhci-uhs2.c
> >>>> index e339821d3504..dfc80a7f1bad 100644
> >>>> --- a/drivers/mmc/host/sdhci-uhs2.c
> >>>> +++ b/drivers/mmc/host/sdhci-uhs2.c
> >>>> @@ -10,7 +10,9 @@
> >>>> * Author: AKASHI Takahiro <takahiro.akashi@xxxxxxxxxx>
> >>>> */
> >>>>
> >>>> +#include <linux/delay.h>
> >>>> #include <linux/module.h>
> >>>> +#include <linux/iopoll.h>
> >>>>
> >>>> #include "sdhci.h"
> >>>> #include "sdhci-uhs2.h"
> >>>> @@ -49,6 +51,49 @@ void sdhci_uhs2_dump_regs(struct sdhci_host *host)
> >>>> }
> >>>> EXPORT_SYMBOL_GPL(sdhci_uhs2_dump_regs);
> >>>>
> >>>> +/*****************************************************************************\
> >>>> + * *
> >>>> + * Low level functions *
> >>>> + * *
> >>>> +\*****************************************************************************/
> >>>> +
> >>>> +bool sdhci_uhs2_mode(struct sdhci_host *host)
> >>>> +{
> >>>> + return host->mmc->flags & MMC_UHS2_SUPPORT;
> >>>
> >>> The MMC_UHS2_SUPPORT bit looks redundant to me. Instead, I think we
> >>> should be using mmc->ios.timings, which already indicates whether we
> >>> are using UHS2 (MMC_TIMING_UHS2_SPEED_*). See patch2 where we added
> >>> this.
> >>>
> >>> That said, I think we should drop the sdhci_uhs2_mode() function
> >>> altogether and instead use mmc_card_uhs2(), which means we should move
> >>> it to include/linux/mmc/host.h, so it becomes available for host
> >>> drivers.
> >>>
> >>
> >> UHS2 mode starts at UHS2 initialization and ends either when UHS2
> >> initialization fails, or the card is removed.
> >>
> >> So it includes re-initialization and reset when the transfer mode
> >> currently transitions through MMC_TIMING_LEGACY.
> >>
> >> So mmc_card_uhs2() won't work correctly for the host callbacks
> >> unless something is done about that.
> >
> > Right, thanks for clarifying!
> >
> > In that case I wonder if we couldn't change the way we update the
> > ->ios.timing for UHS2. It seems silly to have two (similar) ways to
> > indicate that we have moved to UHS2.
>
> Perhaps something like below:
>
> diff --git a/drivers/mmc/core/sd_uhs2.c b/drivers/mmc/core/sd_uhs2.c
> index aacefdd6bc9e..e39d63d46041 100644
> --- a/drivers/mmc/core/sd_uhs2.c
> +++ b/drivers/mmc/core/sd_uhs2.c
> @@ -70,7 +70,8 @@ static int sd_uhs2_power_off(struct mmc_host *host)
>
> host->ios.vdd = 0;
> host->ios.clock = 0;
> - host->ios.timing = MMC_TIMING_LEGACY;
> + /* Must set UHS2 timing to identify UHS2 mode */
> + host->ios.timing = MMC_TIMING_UHS2_SPEED_A;
> host->ios.power_mode = MMC_POWER_OFF;
> if (host->flags & MMC_UHS2_SD_TRAN)
> host->flags &= ~MMC_UHS2_SD_TRAN;
> @@ -1095,7 +1096,8 @@ static void sd_uhs2_detect(struct mmc_host *host)
> mmc_claim_host(host);
> mmc_detach_bus(host);
> sd_uhs2_power_off(host);
> - host->flags &= ~MMC_UHS2_SUPPORT;
> + /* Remove UHS2 timing to indicate the end of UHS2 mode */
> + host->ios.timing = MMC_TIMING_LEGACY;
> mmc_release_host(host);
> }
> }
> @@ -1338,7 +1340,8 @@ static int sd_uhs2_attach(struct mmc_host *host)
> err:
> mmc_detach_bus(host);
> sd_uhs2_power_off(host);
> - host->flags &= ~MMC_UHS2_SUPPORT;
> + /* Remove UHS2 timing to indicate the end of UHS2 mode */
> + host->ios.timing = MMC_TIMING_LEGACY;
> return err;
> }

I wouldn't mind changing to the above. But, maybe an even better
option is to use the ->timing variable in the struct sdhci_host, as
it's there already to keep track of the current/previous timing state.
Would that work too?

>
> diff --git a/drivers/mmc/host/sdhci-uhs2.c b/drivers/mmc/host/sdhci-uhs2.c
> index 517c497112f4..d1f3318b7d3a 100644
> --- a/drivers/mmc/host/sdhci-uhs2.c
> +++ b/drivers/mmc/host/sdhci-uhs2.c
> @@ -267,10 +267,11 @@ static void __sdhci_uhs2_set_ios(struct mmc_host *mmc, struct mmc_ios *ios)
>
> /* UHS2 timing. Note, UHS2 timing is disabled when powering off */
> ctrl_2 = sdhci_readw(host, SDHCI_HOST_CONTROL2);
> - if (ios->timing == MMC_TIMING_UHS2_SPEED_A ||
> - ios->timing == MMC_TIMING_UHS2_SPEED_A_HD ||
> - ios->timing == MMC_TIMING_UHS2_SPEED_B ||
> - ios->timing == MMC_TIMING_UHS2_SPEED_B_HD)
> + if (ios->power_mode != MMC_POWER_OFF &&
> + (ios->timing == MMC_TIMING_UHS2_SPEED_A ||
> + ios->timing == MMC_TIMING_UHS2_SPEED_A_HD ||
> + ios->timing == MMC_TIMING_UHS2_SPEED_B ||
> + ios->timing == MMC_TIMING_UHS2_SPEED_B_HD))
> ctrl_2 |= SDHCI_CTRL_UHS2 | SDHCI_CTRL_UHS2_ENABLE;
> else
> ctrl_2 &= ~(SDHCI_CTRL_UHS2 | SDHCI_CTRL_UHS2_ENABLE);
>
>

Kind regards
Uffe