Re: [PATCH/RFC] mmc: core: Issue power_off_notify for eMMC Suspend-to-RAM

[Ulf Hansson <ulf.hansson@linaro.org>]

On Mon, 8 Jun 2020 at 12:39, Yoshihiro Shimoda
<yoshihiro.shimoda.uh@renesas.com> wrote:
>
> > From: Ulf Hansson, Sent: Monday, June 8, 2020 5:14 PM
> > On Thu, 4 Jun 2020 at 14:17, Yoshihiro Shimoda
> > <yoshihiro.shimoda.uh@renesas.com> wrote:
> > >
> > > Hi again,
> > >
> > > > From: Yoshihiro Shimoda, Sent: Tuesday, May 19, 2020 8:33 PM
> > > >
> > > > The commit 432356793415 ("mmc: core: Enable power_off_notify for
> > > > eMMC shutdown sequence") enabled the power off notification
> > > > even if MMC_CAP2_POWEROFF_NOTIFY (MMC_CAP2_FULL_PWR_CYCLE now) is
> > > > not set. However, the mmc core lacks to issue the power off
> > > > notificaiton when Suspend-to-{RAM,Disk} happens on the system.
> > > >
> > > > So, add Suspend-to-RAM support at first because this is easy to
> > > > check by using pm_suspend_target_state condition on _mmc_suspend().
> > > >
> > > > Signed-off-by: Yoshihiro Shimoda <yoshihiro.shimoda.uh@renesas.com>
> > >
> > > I'd like to add more detail why this patch is needed.
> > > I think we should think some events (which are Shutdown, Suspend-to-idle,
> > > Suspend-to-RAM) for the Power off Notification control.
> > > I described these events like below.
> > >
> > > Assumption of the host : MMC_CAP2_FULL_PWR_CYCLE=false
> > > Assumption of the eMMC : in POWERED_ON
> > >
> > > 1) Event  : Shutdown
> > > - power   : going to VCC=OFF & VCCQ=OFF
> > > - ideal   : Either POWER_OFF_LONG or POWER_OFF_SHORT
> > > - current : POWER_OFF_LONG --> Perfect
> > > - Remarks : the commit 432356793415
> > >
> > > 2) Event  : Suspend-to-Idle
> > > - power   : Keep VCC=ON & VCCQ=ON
> > > - ideal   : issue MMC_SLEEP_AWAKE and keep the power (because the host could not change VCC=OFF)
> > > - current : issue MMC_SLEEP_AWAKE and keep the power --> Perfect
> > > - Remarks : IIUC, even if the eMMC is in POWERED_ON, a host can issue CMD5 (sleep).
> >
> > As a matter of fact, VCCQ *must* remain on in sleep state, while VCC
> > can be powered off.
>
> I got it.
>
> > >
> > > 3) Event  : Suspend-to-RAM
> > > - power   : going to VCC=OFF & VCCQ=OFF
> >
> > I don't understand why you think S2R should be treated differently
> > from S2I? At least from the MMC subsystem point of view, there is no
> > difference. No?
>
> On my environment, VCC & VCCQ condition differs like below.
>  S2I: VCC=ON & VCCQ=ON
>  S2R: VCC=OFF & VCCQ=OFF

Can you explain why it differs? Who is managing the regulators and who
decides to turn them off?

Perhaps this is a regulator-enable usage count problem?

>
> So, I think the MMC subsystem should not enter sleep mode
> on such environment. If this is board-specific, I'm thinking
> I should add a new flag to fix the issue which is entering
> sleep mode even if VCCQ=OFF.
>
> > > - ideal   : Either POWER_OFF_LONG or POWER_OFF_SHORT (because the same as the "Shutdown" event)
> > > - current : issue MMC_SLEEP_AWAKE --> NG
> > > - Remarks : So, I tried to fix this by the patch.
> > >
> > > In addition, we should also think about the event of unbind.
> >
> > Yes, very good point.
> >
> > >
> > > 4) Event  : Unbind
> > > - power   : Keep VCC=ON & VCCQ=ON
> > > - ideal   : NO_POWER_NOTIFICATION because user is possible to turn the power off
> > > - current : Keep POWERED_ON --> NG
> > > - Remarks : But, I didn't try to fix this yet.
> >
> > I don't quite understand why we should keep VCC and VCCQ on?
>
> Oops. I should have described a use case. I thought one of use cases was
> using mmc_test driver. So, I thought we should keep VCC and VCCQ on to
> use mmc_test driver.

Okay, let me think about this.

Actually, we can also unbind the mmc host. And if re-binding again,
that should still work.

>
> > In principle I think we should treat "unbind" in the similar way as we
> > treat S2R/S2I. Which means sending power-off-notification if the host
> > supports MMC_CAP2_FULL_PWR_CYCLE, otherwise we should send sleep.
>
> If we didn't take care of mmc_test driver after unbind, I think so.

Kind regards
Uffe