Re: [RESEND][PATCH v2] i2c: designwear: Fix clk warning on suspend/resume

From: Jarkko Nikula
Date: Thu Mar 30 2017 - 08:39:48 EST


Hi

On 03/30/2017 06:26 AM, John Stultz wrote:
On my HiKey board, I'm seeing clk warnings on suspend/resume,
which seem to be caused by runtime pm suspending the device,
then the same suspend hook being called again on suspend time.

This patch fixes this by setting the SYSTEM_SLEEP_PM_OPS to
using pm_runtime_force_suspend and pm_runtime_force_resume.

Cc: Jarkko Nikula <jarkko.nikula@xxxxxxxxxxxxxxx>
Cc: Andy Shevchenko <andriy.shevchenko@xxxxxxxxxxxxxxx>
Cc: Mika Westerberg <mika.westerberg@xxxxxxxxxxxxxxx>
Cc: Wolfram Sang <wsa@xxxxxxxxxxxxx>
Cc: linux-i2c@xxxxxxxxxxxxxxx
Suggested-by: Jarkko Nikula <jarkko.nikula@xxxxxxxxxxxxxxx>
Signed-off-by: John Stultz <john.stultz@xxxxxxxxxx>
---
v2: Switch to suggested fix by Jarkko.
---
drivers/i2c/busses/i2c-designware-platdrv.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/i2c/busses/i2c-designware-platdrv.c b/drivers/i2c/busses/i2c-designware-platdrv.c
index 79c4b4e..286447c 100644
--- a/drivers/i2c/busses/i2c-designware-platdrv.c
+++ b/drivers/i2c/busses/i2c-designware-platdrv.c
@@ -382,7 +382,7 @@ static int dw_i2c_plat_resume(struct device *dev)
static const struct dev_pm_ops dw_i2c_dev_pm_ops = {
.prepare = dw_i2c_plat_prepare,
.complete = dw_i2c_plat_complete,
- SET_SYSTEM_SLEEP_PM_OPS(dw_i2c_plat_suspend, dw_i2c_plat_resume)
+ SET_SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend, pm_runtime_force_resume)
SET_RUNTIME_PM_OPS(dw_i2c_plat_suspend, dw_i2c_plat_resume, NULL)

We cannot do this since it will break our platforms that use the drivers/acpi/acpi_lpss.c power domain. Reason why it fails the pm_runtime_force_suspend() forces acpi_lpss to power down too early before suspend_late state where context is expected to be saved during system suspend.

acpi_lpss.c power domain
.suspend = acpi_subsys_suspend
-> pm_runtime_resume
-> acpi_lpss_runtime_resume
-> pm_generic_runtime_resume
-> dw_i2c_plat_resume
-> pm_generic_suspend
-> pm_runtime_force_suspend
-> goes through __rpm_get_callback which selects domain ops
-> acpi_lpss_runtime_suspend (context save + power down)
-> pm_generic_runtime_suspend
-> dw_i2c_plat_suspend
...
.suspend_late = acpi_lpss_suspend_late
-> acpi_lpss is already power gated here due acpi_lpss_runtime_suspend() done during suspend phase and context save will save just 0xFFs and resume restores wrong context.

I managed to found a use case where I'm able to trigger dw_i2c_plat_suspend() callback twice during suspend. Quite deep dive to PM core, lot of debugging and end result is a trivial patch that just checks is i2c-designware runtime suspended in its suspend/resume callbacks... This feels somewhat boilerplate and I figured also an alternative solution to PM core.

I just sent that patch out and I Cc'ed both you and Rafael so he can comment is this the right thing to do in driver or should it be fixed in the PM core.

--
Jarkko