[PATCH v2] mmc: sdhci-esdhc-imx: fix abort due to missing runtime PM

[Ulf Hansson]

On 6 May 2015 at 22:23, Stefan Agner <stefan at agner.ch> wrote:
> On 2015-05-06 14:07, Ulf Hansson wrote:
>> On 30 March 2015 at 13:37, Stefan Agner <stefan at agner.ch> wrote:
>>> When entering suspend while the device is in runtime PM, the
>>> sdhci_(suspend|resume)_host function are called with disabled clocks.
>>> Since this functions access the SDHC host registers, this leads to an
>>> external abort on Vybrid SoC:
>>>
>>> [   37.772967] Unhandled fault: imprecise external abort (0x1c06) at 0x76f5f000
>>> [   37.780304] Internal error: : 1c06 [#1] ARM
>>> [   37.784670] Modules linked in:
>>> [   37.787908] CPU: 0 PID: 428 Comm: sh Not tainted 3.18.0-rc5-00119-geefd097-dirty #1540
>>> [   37.796142] task: 8e246c00 ti: 8ca6c000 task.ti: 8ca6c000
>>> [   37.801785] PC is at esdhc_writel_le+0x40/0xec
>>> [   37.806431] LR is at sdhci_set_card_detection+0xe0/0xe4
>>> [   37.811877] pc : [<803f0584>]    lr : [<803eaaa0>]    psr: 400f0013
>>> [   37.811877] sp : 8ca6dd28  ip : 00000001  fp : 8ca6dd3c
>>> [   37.823766] r10: 807a233c  r9 : 00000000  r8 : 8e8b7210
>>> [   37.829194] r7 : 802d8a08  r6 : 8082e928  r5 : 00000000  r4 : 00000002
>>> [   37.835974] r3 : 8ea34e90  r2 : 00000038  r1 : 00000000  r0 : 8ea32ac0
>>> ...
>>>
>>> Clocks need to be enabled to access the registers. Fix the issue by
>>> add driver specific implementation of suspend/resume functions which
>>> take care of clocks using runtime PM API.
>>>
>>> Signed-off-by: Stefan Agner <stefan at agner.ch>
>>
>> Hi Stefan,
>>
>> Sorry for the delay.
>>
>> Indeed sdhci seems to have some issues when combining runtime PM and system PM.
>>
>>> ---
>>> The first version of this patch was part of a patchset sent back in
>>> december. While the second patch of the patchset back then is invalid,
>>> this fix is required to avoid the abort when switching into suspend
>>> mode on Vybrid SoC.
>>>
>>> During review of this patch I realized that my proposed solution to
>>> fix this in the suspend/resume functions in sdhci-pltfm.c is not a
>>> good idea since a lot of drivers use this functions without using the
>>> runtime PM APIs.
>>>
>>> Changes since v1:
>>> - Create new sdhci_esdhc_(suspend|resume) functions in favor of adding a
>>>   hard requirement for runtime PM on SDHC platform implementations
>>>
>>>  drivers/mmc/host/sdhci-esdhc-imx.c | 28 +++++++++++++++++++++++++++-
>>>  1 file changed, 27 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/mmc/host/sdhci-esdhc-imx.c b/drivers/mmc/host/sdhci-esdhc-imx.c
>>> index 10ef824..84b3365 100644
>>> --- a/drivers/mmc/host/sdhci-esdhc-imx.c
>>> +++ b/drivers/mmc/host/sdhci-esdhc-imx.c
>>> @@ -1119,6 +1119,32 @@ static int sdhci_esdhc_imx_remove(struct platform_device *pdev)
>>>  }
>>>
>>>  #ifdef CONFIG_PM
>>> +static int sdhci_esdhc_suspend(struct device *dev)
>>> +{
>>> +       int ret;
>>> +       struct sdhci_host *host = dev_get_drvdata(dev);
>>> +
>>> +       pm_runtime_get_sync(dev);
>>
>> This is indeed needed, since else it isn't safe to invoke sdhci_suspend_host().
>>
>>> +       ret = sdhci_suspend_host(host);
>>> +       pm_runtime_mark_last_busy(dev);
>>> +       pm_runtime_put_autosuspend(dev);
>>
>> The problem is, that this wont do what you think I believe.
>>
>> The device will be kept runtime PM resumed, since the PM core has
>> increased a runtime PM reference count for your device, in the
>> ->prepare() phase (pm_runtime_get_noresume()).
>
> Which ->prepare() phase do you mean exactly? I don't see where
> pm_runtime_get_noresume gets from this driver.

It's not this driver that does it, it's the PM core.

drivers/base/power/main.c

dpm_prepare()
  -> device_prepare()
     -> pm_runtime_get_noresume()

>
>>
>> Potentially pm_runtime_force_suspend() could help you accomplish what you want.
>
> I copied the implementation from the PXA driver (sdhci-pxav3.c), is the
> driver suffering the same issue?

Yes!

Kind regards
Uffe