| Message ID | 1391529538-21685-6-git-send-email-ulf.hansson@linaro.org |
|---|---|
| State | New |
| Headers | show |
On Tue, Feb 4, 2014 at 8:22 PM, Kevin Hilman <khilman@linaro.org> wrote: > Ulf Hansson <ulf.hansson@linaro.org> writes: > >> Due to the available runtime PM callbacks, we are now able to put our >> device into low power state at system suspend. (...) > I'm trying to thing of a good reason to not make PM_SLEEP depend on > PM_RUNTIME for platforms like this. Isn't the typical Android platform using PM_SLEEP without using PM_RUNTIME? Yours, Linus Walleij -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
On 4 February 2014 20:22, Kevin Hilman <khilman@linaro.org> wrote: > Ulf Hansson <ulf.hansson@linaro.org> writes: > >> Due to the available runtime PM callbacks, we are now able to put our >> device into low power state at system suspend. >> >> Earlier we could not accomplish this without trusting a power domain >> for the device to take care of it. Now we are able to cope with >> scenarios both with and without a power domain. >> >> Cc: Russell King <linux@arm.linux.org.uk> >> Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org> >> --- >> drivers/mmc/host/mmci.c | 45 +++++++++++++++++++++++++-------------------- >> 1 file changed, 25 insertions(+), 20 deletions(-) >> >> diff --git a/drivers/mmc/host/mmci.c b/drivers/mmc/host/mmci.c >> index c88da1c..074e0cb 100644 >> --- a/drivers/mmc/host/mmci.c >> +++ b/drivers/mmc/host/mmci.c >> @@ -1723,33 +1723,38 @@ static int mmci_remove(struct amba_device *dev) >> return 0; >> } >> >> -#ifdef CONFIG_SUSPEND >> -static int mmci_suspend(struct device *dev) >> +#ifdef CONFIG_PM_SLEEP >> +static int mmci_suspend_late(struct device *dev) >> { >> - struct amba_device *adev = to_amba_device(dev); >> - struct mmc_host *mmc = amba_get_drvdata(adev); >> + int ret = 0; >> >> - if (mmc) { >> - struct mmci_host *host = mmc_priv(mmc); >> - pm_runtime_get_sync(dev); >> - writel(0, host->base + MMCIMASK0); >> - } >> + if (pm_runtime_status_suspended(dev)) >> + return 0; >> >> - return 0; >> + if (dev->pm_domain && dev->pm_domain->ops.runtime_suspend) >> + ret = dev->pm_domain->ops.runtime_suspend(dev); >> + else >> + ret = dev->bus->pm->runtime_suspend(dev); >> + >> + if (!ret) >> + pm_runtime_set_suspended(dev); > > Isn't this basically open-coding pm_runtime_suspend()... It is similar, but with once big difference. Since the PM core prevents pm_runtime_suspend() from invoking our ->runtime_suspend callback during system suspend (it does so by invoking pm_runtime_get_sync() before starting the suspend sequence), we then need to make the driver handle that by itself. > >> + return ret; >> } >> >> -static int mmci_resume(struct device *dev) >> +static int mmci_resume_early(struct device *dev) >> { >> - struct amba_device *adev = to_amba_device(dev); >> - struct mmc_host *mmc = amba_get_drvdata(adev); >> + int ret = 0; >> >> - if (mmc) { >> - struct mmci_host *host = mmc_priv(mmc); >> - writel(MCI_IRQENABLE, host->base + MMCIMASK0); >> - pm_runtime_put(dev); >> - } >> + if (pm_runtime_status_suspended(dev)) >> + return 0; >> >> - return 0; >> + if (dev->pm_domain && dev->pm_domain->ops.runtime_resume) >> + ret = dev->pm_domain->ops.runtime_resume(dev); >> + else >> + ret = dev->bus->pm->runtime_resume(dev); >> + >> + return ret; > > ...and this is pm_runtime_resume()? (though both terribly simplified.) Correct, but again with a big difference. See comment above. > > This is starting to show that building with PM_SLEEP but not PM_RUNTIME > is going to force open-coding a lot of stuff that the runtime PM > framework already provides. So either we need some helper functions so > we're not sprinkling manual calls to bus/pm_domain callbacks all over I have send a patch a while ago for the PM core, that tried to implement something similar like this, I wasn't accepted. I will follow up on that asap. Still, do you think we could go ahead with this patch? If/when we can get an acceptance for a PM runtime helper function in the PM core, we can easily convert to use it later on. > the place, or maybe where we need to go is have a way for platforms that > really are "runtime PM centric" to declare that even PM_SLEEP depends on > PM_RUNTIME. > > I'm trying to thing of a good reason to not make PM_SLEEP depend on > PM_RUNTIME for platforms like this. This wont help. The PM core will still prevent the runtime_suspend callback from being invoked during system suspend. Kind regards Ulf Hansson > > Kevin > -- 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
diff --git a/drivers/mmc/host/mmci.c b/drivers/mmc/host/mmci.c index c88da1c..074e0cb 100644 --- a/drivers/mmc/host/mmci.c +++ b/drivers/mmc/host/mmci.c @@ -1723,33 +1723,38 @@ static int mmci_remove(struct amba_device *dev) return 0; } -#ifdef CONFIG_SUSPEND -static int mmci_suspend(struct device *dev) +#ifdef CONFIG_PM_SLEEP +static int mmci_suspend_late(struct device *dev) { - struct amba_device *adev = to_amba_device(dev); - struct mmc_host *mmc = amba_get_drvdata(adev); + int ret = 0; - if (mmc) { - struct mmci_host *host = mmc_priv(mmc); - pm_runtime_get_sync(dev); - writel(0, host->base + MMCIMASK0); - } + if (pm_runtime_status_suspended(dev)) + return 0; - return 0; + if (dev->pm_domain && dev->pm_domain->ops.runtime_suspend) + ret = dev->pm_domain->ops.runtime_suspend(dev); + else + ret = dev->bus->pm->runtime_suspend(dev); + + if (!ret) + pm_runtime_set_suspended(dev); + + return ret; } -static int mmci_resume(struct device *dev) +static int mmci_resume_early(struct device *dev) { - struct amba_device *adev = to_amba_device(dev); - struct mmc_host *mmc = amba_get_drvdata(adev); + int ret = 0; - if (mmc) { - struct mmci_host *host = mmc_priv(mmc); - writel(MCI_IRQENABLE, host->base + MMCIMASK0); - pm_runtime_put(dev); - } + if (pm_runtime_status_suspended(dev)) + return 0; - return 0; + if (dev->pm_domain && dev->pm_domain->ops.runtime_resume) + ret = dev->pm_domain->ops.runtime_resume(dev); + else + ret = dev->bus->pm->runtime_resume(dev); + + return ret; } #endif @@ -1820,7 +1825,7 @@ static int mmci_runtime_resume(struct device *dev) #endif static const struct dev_pm_ops mmci_dev_pm_ops = { - SET_SYSTEM_SLEEP_PM_OPS(mmci_suspend, mmci_resume) + SET_LATE_SYSTEM_SLEEP_PM_OPS(mmci_suspend_late, mmci_resume_early) SET_PM_RUNTIME_PM_OPS(mmci_runtime_suspend, mmci_runtime_resume, NULL) };
Due to the available runtime PM callbacks, we are now able to put our device into low power state at system suspend. Earlier we could not accomplish this without trusting a power domain for the device to take care of it. Now we are able to cope with scenarios both with and without a power domain. Cc: Russell King <linux@arm.linux.org.uk> Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org> --- drivers/mmc/host/mmci.c | 45 +++++++++++++++++++++++++-------------------- 1 file changed, 25 insertions(+), 20 deletions(-)