On 8/17/2021 1:53 AM, Stephen Boyd wrote:
Quoting Deepak Kumar Singh (2021-08-16 03:05:35)
Remote susbsystems notify fatal crash throught smp2p interrupt.
When modem/wifi crashes it can cause soc to come out of low power state
and may not allow again to enter in low power state until crash is handled.
Mark smp2p interrupt wakeup capable so that interrupt handler is executed
and remote susbsystem crash can be handled in system resume path.
Signed-off-by: Deepak Kumar Singh <deesin@xxxxxxxxxxxxxx>
---
drivers/soc/qcom/smp2p.c | 12 ++++++++++++
1 file changed, 12 insertions(+)
diff --git a/drivers/soc/qcom/smp2p.c b/drivers/soc/qcom/smp2p.c
index 2df4883..646848b 100644
--- a/drivers/soc/qcom/smp2p.c
+++ b/drivers/soc/qcom/smp2p.c
@@ -18,6 +18,7 @@
#include <linux/soc/qcom/smem.h>
#include <linux/soc/qcom/smem_state.h>
#include <linux/spinlock.h>
+#include <linux/pm_wakeirq.h>
Please sort alphabetically by include name, 'p' before 's'.
/*
* The Shared Memory Point to Point (SMP2P) protocol facilitates communication
@@ -538,9 +539,20 @@ static int qcom_smp2p_probe(struct platform_device *pdev)
goto unwind_interfaces;
}
+ /* Setup smp2p interrupt as wakeup source */
This comment is bad. Please don't reiterate what the code is doing.
Instead, write something like
/*
* Treat remoteproc crashes as wakeups by default so we handle
* them sooner rather than along with the next wakeup (e.g.
* power button). This avoids leaving the system in a shallower
* suspend power state if a remoteproc crashes during suspend,
* but requires userspace to actively suspend the device after
* handling the crash, or CONFIG_PM_AUTOSLEEP to be true.
*/
+ ret = device_init_wakeup(&pdev->dev, true);
I still wonder if it's better to leave this off by default and only
enable it if the kernel is using autosuspend (PM_AUTOSLEEP). Then
userspace is responsible to decide if it can handle the wakeup with the
screen off, reload the remoteproc, and go back to suspend if it isn't
using autosuspend.
Seems like not all targets use PM_AUTOSLEEP feature, even those targets
may require wakeup to handle
modem crash so that important modem events are not missed. I think we
can keep wake up as default behavior
and let the user space disable it through sysfs if it doesn't want it as
wake up source.
+ if (ret)
+ goto unwind_interfaces;
+
