Re: [PATCH v2 0/6] ARM: STi reset controller support

[srinivas kandagatla <srinivas.kandagatla@st.com>]

On 24/02/14 15:16, Philipp Zabel wrote:
> Hi Srinivas,
> 
> Am Montag, den 24.02.2014, 14:03 +0000 schrieb srinivas kandagatla:
>> Thanks Philipp for your comments,
>>
>> On 24/02/14 10:33, Philipp Zabel wrote:
>>>>> Did Srini's explanations convinced you?
>>>>>
>>>>> If so, could you queue the series for v3.15?
>>> to be honest, I'm not comfortable with this explanation. If the
>>> "powerdown" bits only gate the clocks to those modules, calling it a
>>> reset control is clearly the wrong abstraction. If that is the case,
>>> couldn't you handle those bits via the clock framework?
>> I just had a re-look at the IPs specs for more information on where
>> these power-down signals are actually terminating on the IP side.
>>
>> For example: ST-Synopsis Ethernet GMAC IP has two pins
>> power_down_req[IN] and power_down_ack[OUT]. power_down_req is used by
>> the software to either put the IP in powerdown or bring it out of
>> powerdown state.
> 
> Now I'm a bit confused. There is no mention of GMAC in your patches,
> and for ETH[01] they contain only the SOFTRESET bits. I have no issue
> with the SOFTRESETs.
Yes, GMAC was a bad example indeed. However this same logic applies to
the USB IP as well.

GMAC power-down-reset can be added to the power-down-reset list for
consistency.

We did not define the power-down-reset for GMAC because the reset state
of GMAC will not be in power down. softreset should be enough to bring
the IP in to a usable state. So the software never drives the power
down-request but instead uses softreset in this particular case.

> 
>> The IP itself drives power_down_ack to indicate when the power down
>> request is successfully finished. For power_down/power_up request the IP
>> will change the internal state accordingly including powering up/down
>> its internal blocks and/or clock gating.
>>
>>> If on the other hand these powerdown bits also trigger reset machinery,
>>> such that asserting and deasserting that bit will change the module's
>>> internal state, I could be convinced to queue them like this.
>> This is true with ST IPs, these lines change the state of the IP as
>> described above. Reset framework seems to fits in very well with this
>> behavior rather than power-domains or clock framework.
> 
> If you put the IP in power down when it is idle, and then power it up
> again, will the IP registers have kept their previous state?
No, the context is lost, the IP needs re-initialization.

Thanks,
srini
> 
> regards
> Philipp
> 
> 
>