From: Vijay Viswanath <vviswana@xxxxxxxxxxxxxx>
mmc_host is a virtual device and it doesn't have any pm ops and so during
pm registration of device, no_pm_callback gets set as true. The
mmc_host device is not runtime enabled as it is a virtual device and
mmc_host is the parent device of mmc_card. As the mmc_host is runtime
disabled, mmc_card can runtime suspend/resume without depending on
state of mmc_host during normal operations. During system suspend, the
direct_complete flag of mmc_host device gets set as it has no pm_ops.
When mmc_card successfully suspends, it clears the direct_complete flag
of its parent (mmc_host).
But in certain cases during dpm_suspend, an async error can occur after
suspend work for mmc_card is scheduled and before it gets executed. In
that case, mmc_card suspend work will not clear the direct_complete flag
of mmc_host. When mmc_host suspend comes after that of mmc_card,
it too will skip all actions.
But by this time, the mmc_host device has been added to device_suspended
list. So during resume, mmc_host resume will do dpm resume of mmc_host.
In dpm_resume, all devices which has direct_complete flag set will be
runtime_enabled. This is because, in dpm_suspend, any device with
direct_complete flag will be runtime_disabled. Thus, mmc_host which has
direct_complete flag set, will get runtime enabled during dpm_resume.
This is a problem in pm framework with direct_complete flag
(runtime enabling a device in resume when it was not runtime disabled
in suspend path).
Now that mmc_host device is runtime enabled, to runtime resume the
mmc_card, the pm framework will try to runtime resume the mmc_host
device as well and will fail. This prevents mmc_card from runtime
resuming after a runtime_suspend.
Fix this by adding a dummy suspend_prepare() fn for mmc_host. This
prevents the direct_complete flag of mmc_host device from getting set.
Signed-off-by: Vijay Viswanath <vviswana@xxxxxxxxxxxxxx>
Signed-off-by: Ram Prakash Gupta <rampraka@xxxxxxxxxxxxxx>
Signed-off-by: Bao D. Nguyen <nguyenb@xxxxxxxxxxxxxx>
---
drivers/mmc/core/host.c | 19 +++++++++++++++++++
1 file changed, 19 insertions(+)
diff --git a/drivers/mmc/core/host.c b/drivers/mmc/core/host.c
index 105b7a7..242657b 100644
--- a/drivers/mmc/core/host.c
+++ b/drivers/mmc/core/host.c
@@ -40,9 +40,28 @@ static void mmc_host_classdev_release(struct device *dev)
kfree(host);
}
+static int mmc_host_prepare(struct device *dev)
+{
+ /*
+ * Since mmc_host is a virtual device, we don't have to do anything.
+ * If we return a positive value, the pm framework will consider that
+ * the runtime suspend and system suspend of this device is same and
+ * will set direct_complete flag as true. We don't want this as the
+ * mmc_host always has positive disable_depth and setting the flag
+ * will not speed up the suspend process.
+ * So return 0.
+ */
+ return 0;
+}
+
+static const struct dev_pm_ops mmc_pm_ops = {
+ .prepare = mmc_host_prepare,
+};
+
static struct class mmc_host_class = {
.name = "mmc_host",
.dev_release = mmc_host_classdev_release,
+ .pm = &mmc_pm_ops,
};
int mmc_register_host_class(void)
--
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum,
a Linux Foundation Collaborative Project
