mmc: sdio: support suspend/resume while runtime suspended

Bring SDIO devices back to full power before their suspend
handler is invoked.

Doing so ensures that SDIO suspend/resume semantics are
maintained (drivers still get to decide whether their
card should be removed or kept during system suspend,
and at what power state), and that SDIO suspend/resume
execution paths are unchanged.

This is achieved by resuming a runtime-suspended SDIO device
in its ->prepare() PM callback (similary to the PCI subsystem).

Since the PM core always increments the run-time usage
counter before calling the ->prepare() callback and decrements
it after calling the ->complete() callback, it is guaranteed
that when the system will come out of suspend, our device's
power state will reflect its runtime PM usage counter.

Signed-off-by: Ohad Ben-Cohen <ohad@wizery.com>
Tested-by: Luciano Coelho <luciano.coelho@nokia.com>
Signed-off-by: Chris Ball <cjb@laptop.org>

1 files changed, 29 insertions, 0 deletions

diff --git a/drivers/mmc/core/sdio_bus.c b/drivers/mmc/core/sdio_bus.c
index 3637483..2716c7a 100644
--- a/drivers/mmc/core/sdio_bus.c
+++ b/drivers/mmc/core/sdio_bus.c
@@ -189,12 +189,41 @@ out:
 
 #ifdef CONFIG_PM_RUNTIME
 
+static int sdio_bus_pm_prepare(struct device *dev)
+{
+	/*
+	 * Resume an SDIO device which was suspended at run time at this
+	 * point, in order to allow standard SDIO suspend/resume paths
+	 * to keep working as usual.
+	 *
+	 * Ultimately, the SDIO driver itself will decide (in its
+	 * suspend handler, or lack thereof) whether the card should be
+	 * removed or kept, and if kept, at what power state.
+	 *
+	 * At this point, PM core have increased our use count, so it's
+	 * safe to directly resume the device. After system is resumed
+	 * again, PM core will drop back its runtime PM use count, and if
+	 * needed device will be suspended again.
+	 *
+	 * The end result is guaranteed to be a power state that is
+	 * coherent with the device's runtime PM use count.
+	 *
+	 * The return value of pm_runtime_resume is deliberately unchecked
+	 * since there is little point in failing system suspend if a
+	 * device can't be resumed.
+	 */
+	pm_runtime_resume(dev);
+
+	return 0;
+}
+
 static const struct dev_pm_ops sdio_bus_pm_ops = {
 	SET_RUNTIME_PM_OPS(
 		pm_generic_runtime_suspend,
 		pm_generic_runtime_resume,
 		pm_generic_runtime_idle
 	)
+	.prepare = sdio_bus_pm_prepare,
 };
 
 #define SDIO_PM_OPS_PTR	(&sdio_bus_pm_ops)