[PATCH] ARM: OMAP2+: omap_device: drop broken RPM status update from suspend_noirq

Tue Jul 25 10:48:40 PDT 2017

Hi Johan,

On 07/25/2017 03:24 AM, Johan Hovold wrote:
> On Mon, Jul 24, 2017 at 05:16:02PM -0500, Grygorii Strashko wrote:
>> On 07/24/2017 04:52 AM, Johan Hovold wrote:
>>> Since commit a8636c89648a ("PM / Runtime: Don't allow to suspend a
>>> device with an active child"), which went into 4.10, it is no longer
>>> permitted to set RPM_SUSPENDED state for a device with active children
>>> (unless power.ignore_children is set).
>>>
>>> This specifically means that the attempts to do just that from the omap
>>> pm-domain suspend_noirq callback have since been failing whenever a
>>> child is active, for example:
>>>
>>>     am335x-usb-childs 47400000.usb: runtime PM trying to suspend
>>>       device but active child
>>>
>>> Silence this warning by dropping the broken pm_runtime_set_suspended()
>>> call from the omap suspend_noirq callback along with the redundant
>>> pm_runtime_set_active() in resume_noirq.
>>>
>>> This effectively reverts commit 3522bf7bfa24 ("ARM: OMAP2+: omap_device:
>>> maintain sane runtime pm status around suspend/resume"), which started
>>> updating the RPM state after the runtime_suspend callback (!) for active
>>> omap devices had been called during system suspend. The rationale was
>>> that a later pm_runtime_get_sync() would then fail (even after runtime
>>> pm had been disabled) and that this in turn would avoid any external
>>> aborts when accessing registers with clocks disabled. (See also commit
>>> 6f3c77b040fc ("PM / Runtime: let rpm_resume() succeed if RPM_ACTIVE,
>>> even when disabled, v2").
>>>
>>> But during the suspend_noirq phase all children would already have been
>>> suspended and their drivers would specifically not attempt any further
>>> register accesses. And if this was all just a workaround for random
>>> device drivers doing cross-tree calls during system suspend, those
>>> drivers should be fixed and updated to explicitly model such
>>> dependencies using device-links instead (and either way, any such calls
>>> have been causing crashes since 4.10).
>>
>> I'd like to provide some background and comments here.
>>
>> 1) OMAP platform is PM runtime centric - in other words all device's
>> PM operations (power on/off) are done through PM runtime calls and
>> devices drivers do not really change device's Power states during
>> suspend/resume - just do preparation for suspend. The final disabling
>> of devices is done form _od_suspend_noirq() or if device driver calls
>> pm_runtime_force_suspend() directly during suspend.
>>
>> 2) Initially all this suspend_noirq code was introduced due to commit
>>
>> commit 1e2ef05bb8cf851a694d38e9170c89e7ff052741
>> Author: Rafael J. Wysocki <rjw at sisk.pl>
>> Date:   Wed Jul 6 10:51:58 2011 +0200
>>
>>      PM: Limit race conditions between runtime PM and system sleep (v2)
>>
>> which blocks PM runtime during suspend resume. And, at that time,
>> there were no pm_runtime_force_suspend/resume() neither proper support
>> for power domains.
>>
>> As result, below two calls really switch OFF device and its state
>> should be RPM_SUSPENDED as it will reflect real PM state of HW:
>> 	m_generic_runtime_suspend(dev)
>> 	omap_device_idle(pdev);
> 
> That's the point to be discussed; does the runtime status really need to
> reflect the hardware state *during* suspend?

At least there are no restriction, as I know.

> 
> As you've noticed, not all subsystems enforce that, even if omap seems
> to have been expecting it.

True.

> 
> And as I mentioned in the commit message, at least the point about
> wanting to have late pm_runtime_get_sync() fail during suspend appears
> to be moot.

Correct. It was introduced long time ago (before device-links).

> 
>> In general, now the code in _od_resume/resume_noirq() callback is
>> equivalent to pm_runtime_force_suspend/resume().
>>
>> 3) It's expected to ignore -EBUSY return code from
>> pm_generic_runtime_suspend(dev) in _od_suspend_noirq(), as such code
>> (-EBUSY) tells OMAP device framework that device should be kept active
>> during suspend (for example serial device in case of
>> "no_console_suspend").
>>
>> commit 4b7ec5accecdb136c7afaf8739a06d5335cc05aa
>> Author: Sourav Poddar <sourav.poddar at ti.com>
>> Date:   Sat Apr 27 01:55:34 2013 +0530
>>
>>      arm: omap2+: omap_device: remove no_idle_on_suspend
>>
>> As result, when "runtime PM trying to suspend device but active child"
>> happens the OMAP device framework will ignore it and gracefully
>> continue to suspend where target device will finally powered off
>> (depending on suspend mode and device).  On resume, target device will
>> not be powered on, because its state was not marked as
>> OMAP_DEVICE_SUSPENDED, so further attempts to power on device with
>> pm_runtime_get_sync() will be a NOP (because device state is
>> RPM_ACTIVE) and any access to the device will fail (system crash).
> 
> I believe you're mistaken here; _od_suspend_noirq() will continue to
> power-off *and* set OMAP_DEVICE_SUSPENDED, so that device is again
> powered during resume_noirq also after this patch has been applied.
> 
> Specifically, note that the return value of pm_runtime_set_suspended()
> was never checked, so the only difference here would be that the error
> message is suppressed.

Correct. Sorry. My bad - it was Monday. 

> 
>> 4) Unfortunately, commit a8636c89648a ("PM / Runtime: Don't allow to
>> suspend a device with an active child") "breaks" OMAP suspend
>> functionality (or makes visible more fundamental issues of PM runtime
>> vs Sys suspend), because not all Kernel subsystem (especially USB)
>> maintain their devices PM runtime state properly during suspend, which
>> cause pm_runtime_set_suspended() (or pm_runtime_force_suspend()) to
>> return -EBUSY and, as result, misbehave of OMAP device framework.
> 
> Again, the return value of pm_runtime_set_suspended() was never checked
> so the only thing commit a8636c89648a ("PM / Runtime: Don't allow to
> suspend a device with an active child") did was to trigger those error
> messages and leave the runtime PM state at RPM_ACTIVE. The latter does
> not seems to have any other side effects, and specifically, the device
> would still be powered off.

Again sorry for the noise.

> 
>> My personal thought here is that removing of pm_runtime_set_active()
>> will not fix root cause of the problem, but rather hide it :( and,
>> probably, real fix will be to update USB framework to ensure that all
>> suspend devices are also PM runtime suspend (not sure how) or add few
>> more pm_suspend_ignore_children() calls (for example as I've tried to
>> do in [2], but this was unfinished).
>>
>> I've found very simple steps to reproduce suspend failure on
>> am335x-evm (should also > work on BBB) -  do below sequence with USB
>> device plugged:
>>
>>     echo platform > /sys/power/pm_test
>>     echo 1 > /sys/power/pm_print_times
>>     [ echo 0 > /sys/module/printk/parameters/console_suspend ]
>>     echo mem > /sys/power/state
>>
>> [   95.499685] calling  47400000.usb+ @ 733, parent: ocp
>> [   95.504818] am335x-usb-childs 47400000.usb: runtime PM trying to suspend device but active child
>> [   95.513750] am335x-usb-childs 47400000.usb: omap device suspend failure 0
>>
>> Below I've attached possible patch which converts OMAP device to
>> use pm_runtime_force_suspend/resume().
> 
> What code are you running above? The OMAP PM code should not be failing
> due to that error message at least (as mentioned twice above).
> 
> I have musb suspend working on BBB with the patch I posted yesterday
> which keeps the controller at RPM_ACTIVE during suspend:
> 
> 	https://marc.info/?i=20170724094939.21477-1-johan%40kernel.org

I re-checked it on clean master (Linux 4.13-rc2) with and without your patches
and see no issues, just "runtime PM trying to suspend device but active child"
message is gone. Above test sequence still can be used without any additional patches.

So, thank you for your patches and sorry for the noise.

Tested-by: Grygorii Strashko <grygorii.strashko at ti.com>

-- 
regards,
-grygorii