reference implementation of runtime PM

reference implementation of runtime PM

[Daniel Drake]

Hi,

I'm working on getting runtime pm working on the XO-1.5 laptop
(sdhci-pci) in case it is related to some other power-related issues
we have been having.

Right now, when runtime is enabled, the card cannot be brought back
online again after it has been powered off.

I've dug down and I think that runtime PM is doing basically nothing.
i.e. when the system decides to runtime-suspend my SD card, nothing
actually happens to the SD card. It remains fully powered and
untouched. I'm basing this assumption on mmc_power_save_host() which
relies entirely on host->bus_ops->power_save() doing the actual
power-down work. On sdhci this method does not exist.

I figure that if I write sdhci power_save and power_restore functions,
the problem might be solved.

I looked at the other host controllers hoping to get a feel for the
kind of thing that should be done in such a handler but didn't find
any host controller that implements this.

Is there a host controller patch somewhere that implements
power_save/power_restore? Maybe even for sdhci?

cheers,
Daniel

Re: reference implementation of runtime PM

[Ben Dooks]

On Mon, Dec 06, 2010 at 05:57:22PM +0000, Daniel Drake wrote:
> Hi,
> 
> I'm working on getting runtime pm working on the XO-1.5 laptop
> (sdhci-pci) in case it is related to some other power-related issues
> we have been having.
> 
> Right now, when runtime is enabled, the card cannot be brought back
> online again after it has been powered off.
> 
> I've dug down and I think that runtime PM is doing basically nothing.
> i.e. when the system decides to runtime-suspend my SD card, nothing
> actually happens to the SD card. It remains fully powered and
> untouched. I'm basing this assumption on mmc_power_save_host() which
> relies entirely on host->bus_ops->power_save() doing the actual
> power-down work. On sdhci this method does not exist.

you might want to add the linux-pm list to this.

-- 
Ben

Q:      What's a light-year?
A:      One-third less calories than a regular year.

Re: reference implementation of runtime PM

[Daniel Drake]

On 6 December 2010 17:57, Daniel Drake <dsd@laptop.org> wrote:
> I've dug down and I think that runtime PM is doing basically nothing.
> i.e. when the system decides to runtime-suspend my SD card, nothing
> actually happens to the SD card. It remains fully powered and
> untouched. I'm basing this assumption on mmc_power_save_host() which
> relies entirely on host->bus_ops->power_save() doing the actual
> power-down work. On sdhci this method does not exist.

Sorry, I now realise there are a number of confusions in my earlier mail.
power_save is done by the bus, not the host, and the host does get its
opportunity to power down via the set_ios callback that comes in
telling it to turn the card off.

Still can't get our card to reset properly though. Tried sprinking
some sdhci_reset and sdhci_reinit calls around, no luck.

We do know that it can be powered down and up just fine at runtime via
mmc_stop_host then mmc_start_host. So there's something that happens
in that path which does not occur with runtime pm...

Daniel

Re: reference implementation of runtime PM

[Ohad Ben-Cohen]

On Mon, Dec 6, 2010 at 10:03 PM, Daniel Drake <dsd@laptop.org> wrote:
> power_save is done by the bus, not the host, and the host does get its
> opportunity to power down via the set_ios callback that comes in
> telling it to turn the card off.

yes.

In fact, the actual work is being handled by mmc_power_off, and not by
the ->power_save() handler (which is not being used at all today).

>
> Still can't get our card to reset properly though. Tried sprinking
> some sdhci_reset and sdhci_reinit calls around, no luck.
>
> We do know that it can be powered down and up just fine at runtime via
> mmc_stop_host then mmc_start_host.

Just wondering.. how did you verify this ? rmmod/insmod your host controller ?

> So there's something that happens
> in that path which does not occur with runtime pm...

Then you can probably compare the flow (ftrace ?) and find the difference.

>
> Daniel
> --
> To unsubscribe from this list: send the line "unsubscribe linux-mmc" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>