sdhci: runtime suspend/resume on card insert/removal

[Vaibhav Hiremath]

On Monday 14 September 2015 01:48 PM, Jisheng Zhang wrote:
> On Mon, 14 Sep 2015 13:43:32 +0530
> Vaibhav Hiremath <vaibhav.hiremath at linaro.org> wrote:
>
>>
>>
>> On Monday 14 September 2015 11:58 AM, Jisheng Zhang wrote:
>>> On Mon, 14 Sep 2015 11:55:56 +0530
>>> Vaibhav Hiremath <vaibhav.hiremath at linaro.org> wrote:
>>>
>>>>
>>>>
>>>> On Thursday 10 September 2015 01:34 PM, Jisheng Zhang wrote:
>>>>> Hi Russell,
>>>>>
>>>>> On Thu, 10 Sep 2015 09:02:33 +0100
>>>>> Russell King - ARM Linux <linux at arm.linux.org.uk> wrote:
>>>>>
>>>>>> On Thu, Sep 10, 2015 at 03:31:29PM +0800, Jisheng Zhang wrote:
>>>>>>> Hi Vaibhav,
>>>>>>>
>>>>>>> On Thu, 10 Sep 2015 12:40:48 +0530
>>>>>>> Vaibhav Hiremath <vaibhav.hiremath at linaro.org> wrote:
>>>>>>>
>>>>>>>> Hi,
>>>>>>>>
>>>>>>>> During my testing of SDHCI-PXAV3 driver on Marvell's pxa1928
>>>>>>>> based platform, I observed that runtime PM suspend/resume is having
>>>>>>>> issues with card insertion and removal.
>>>>>>>>
>>>>>>>> Let me try to explain it using execution sequence -
>>>>>>>>
>>>>>>>> During boot:
>>>>>>>>
>>>>>>>> MMC SD card gets detected as expected.
>>>>>>>>
>>>>>>>> [    2.431012] mmc1: new high speed SDHC card at address 1234
>>>>>>>> [    2.437235] mmcblk1: mmc1:1234 SA04G 3.63 GiB
>>>>>>>> [    2.444841]  mmcblk1: p1
>>>>>>>>
>>>>>>>>
>>>>>>>> Now after coming to the linux prompt, if card removal event occurs
>>>>>>>> then the call sequence is -
>>>>>>>>
>>>>>>>>      sdhci_irq() -->
>>>>>>>>       -> sdhci_thread_irq(): host->thread_isr - 0x80
>>>>>>>>          -> sdhci_card_event()
>>>>>>>>          -> mmc_detect_change()
>>>>>>>>             --> _mmc_detect_change()
>>>>>>>>                 --->  mmc_sd_detect()
>>>>>>>>                       mmc_sd_remove()
>>>>>>>>                       mmc_remove_card()
>>>>>>>>                       mmc_bus_remove()
>>>>>>>>                       mmc_power_off()
>>>>>>>>                       mmc_set_initial_state()
>>>>>>>>                       sdhci_set_ios()
>>>>>>>>                       ...
>>>>>>>>            sdhci_pxav3_runtime_suspend()
>>>>>>>>            sdhci_runtime_suspend_host()
>>>>>>>>
>>>>>>>>
>>>>>>>> Till here everything looks perfect :) (if I got it right)
>>>>>>>>
>>>>>>>> Now on card insertion again, the expectation is, runtime resume should
>>>>>>>> get called as part of interrupt trigger from the SDHCI controller on
>>>>>>>> card insertion.
>>>>>>>
>>>>>>> AFAIK, card insertion => wakeup irq, this irq doesn't come from SDHCI
>>>>>>
>>>>>> Wakeup IRQs are what happens when the _system_ is in suspend, not
>>>>>> when the device is runtime suspended.
>>>>>
>>>>> Oh, yes. Sorry for misleading, I didn't express myself clearly. What
>>>>> I really means is that the card insertion/remove irq which could finally
>>>>> cause sdhci host resumed.
>>>>>
>>>>
>>>> So Jisheng,
>>>>
>>>> You are ok with above change, right?
>>>
>>> Nope. the above is just to clarify my mixing "wakeup IRQs" and the card
>>> insertion/remove irq that brings sdhci host resumed.
>>>
>>> IMHO, your patch is still not necessary and perhaps wrong.
>>>
>>
>> Don't be in hurry to conclude,
>> Let's understand the technical aspect fist.
>>
>>
>> Let me clarify once again here,
>>
>>
>> The card detect interrupt is coming from SDHCI controller itself,
>
> But how the SDHCI controller generate CD interrupt when it's in runtime suspended
> state, i.e clock gated or power gated?
>

Due to lack of detailed Documentation its difficult to conclude here


>> I have confirmed that the runtime PM is working perfectly fine,
>> clock is gated/disabled on card removal.
>
> But seems we still didn't know the sdhci host is clock gated or power gated.
>

Register above is
SDIO Host 1 Clock/Reset Control Register

And bit 3 & 4 says

BIT 3:

SD Host 1 Peripheral Clock Enable
   0x1: Peripheral clock enabled
   0x0: Peripheral clock disabled

BIT4:

SD Host 1 AXI Clock Enable
   0x1: AXI clock enabled
   0x0: AXI clock disabled


>>
>>
>> Clock Gating:
>> =============
>> #
>> # devmem 0xd4282854
>> 0x0000181B
>> # [ 1318.948460] mmc1: card 1234 removed
>>
>> #
>> # devmem 0xd4282854
>> 0x00001800
>> #
>
> Can you plz read the SDHCI host registers after this step? If the host is
> clock gated, we should get "asynchronous external abort" in theory.
>

Ok, that's another way. I can cross check on this.

Thanks,
Vaibhav