On Mon, 31 Mar 2025 at 12:46, Wolfram Sang
<wsa+renesas@xxxxxxxxxxxxxxxxxxxx> wrote:
>
> Hi Ulf,
>
> > > > > +static bool mmc_may_poweroff_notify(const struct mmc_host *host,
> > > > > + bool is_suspend)
> > >
> > > Maybe add some comments about the difference between
> > > mmc_can_poweroff_notify() and mmc_may_poweroff_notify()? Like make it
> > > super-obvious, so I will easily remember next year again :)
> >
> > mmc_can_* functions are mostly about checking what the card is capable
> > of. So mmc_can_poweroff_notify() should be consistent with the other
> > similar functions.
> >
> > For eMMC power-off notifications in particular, it has become more
> > complicated as we need to check the power-off scenario along with the
> > host's capabilities, to understand what we should do.
> >
> > I am certainly open to another name than mmc_may_power_off_notify(),
> > if that is what you are suggesting. Do you have a proposal? :-)
>
> Initially, I didn't think of new names but some explanation in comments.
> But since you are mentioning a rename now, how about:
>
> mmc_card_can_poweroff_notify() and mmc_host_can_poweroff_notify()?
mmc_card_can_poweroff_notify() would not be consistent with all the
other mmc_can_* helpers, so I rather stay with
mmc_can_poweroff_notify(), for now. If you think a rename makes sense,
I suggest we do that as a follow up and rename all the helpers.
mmc_host_can_poweroff_notify() seems fine to me!
>
> Similar to the commit 32f18e596141 ("mmc: improve API to make clear
> hw_reset callback is for cards") where I renamed 'hw_reset' to
> 'card_hw_reset' for AFAICS similar reasons.
>
> > > > > if (mmc_can_poweroff_notify(host->card) &&
> > > > > - !(host->caps2 & MMC_CAP2_FULL_PWR_CYCLE))
> > > > > + !mmc_may_poweroff_notify(host, true))
> > > > I guess this deserve some extra documentation because:
> > > > If MMC_CAP2_FULL_PWR_CYCLE is not set but MMC_CAP2_FULL_PWR_CYCLE_IN_SUSPEND is set,
> > > > !mmc_may_poweroff_notify(host, true) will evaluate to false while !(host->caps2 & MMC_CAP2_FULL_PWR_CYCLE) will evaluate to true.
> >
> > Right. See more below.
> >
> > >
> > > I agree, I neither get this. Another way to express my confusion is: Why
> > > do we set the 'is_suspend' flag to true in the shutdown function?
> > >
> >
> > I understand your concern and I agree that this is rather messy.
> > Anyway, for shutdown, we set the is_suspend flag to false. The
> > reasoning behind this is that during shutdown we know that the card
> > will be fully powered-down (both vcc and vccq will be cut).
> >
> > In suspend/runtime_suspend, we don't really know as it depends on what
> > the platform/host is capable of. If we can't do a full power-off
> > (maybe just vcc can be cut), then we prefer the sleep command instead.
>
> I do understand that. I don't see why this needs a change in the
> existing logic as Alan pointed out above.
Aha. I get your point now. As stated in the commit message:
Due to an earlier suspend request the eMMC may already have been properly
powered-off, hence we are sometimes leaving the eMMC in its current state.
However, in one case when the host has MMC_CAP2_FULL_PWR_CYCLE_IN_SUSPEND
set we may unnecessarily restore the power to the eMMC, let's avoid this.
To further clarify, at a system suspend we issue a poweroff-notify for
the case above. At system resume we leave the card in powered-off
state until there is I/O (when we runtime resume it). If a shutdown
occurs before I/O, we would unnecessarily re-initialize the card as
it's already in the correct state.
Let me try to clarify the commit message a bit with this information.
>
> > I was hoping that patch3 should make this more clear (using an enum
>
> Sadly, it didn't. Using MMC_POWEROFF_SUSPEND first and then later
> MMC_POWEROFF_SHUTDOWN in mmc_shutdown() is still confusing. Do you want
> to return false in case none of the two PWR_CYCLE flags is set?
>
> > type), but I can try to add some comment(s) in the code to further
> > clarify the policy.
>
> Please do.
>
> All the best,
>
> Wolfram
>
Thanks!
Kind regards
Uffe
