On 02/17/2016 09:52 PM, Peter Ujfalusi wrote:
> On 02/17/2016 02:07 PM, Mark Brown wrote:
>> On Wed, Feb 17, 2016 at 10:13:35AM +0200, Peter Ujfalusi wrote:
>>
>>> With this change we don't need to write custom machine drivers for setup not
>>> using sysclk_id == 0.
>>> I do think this is reasonable change by itself.
>>
>>> However I do think that the current simple-card is flawed regarding to clock
>>> selection and the change Jyri and me are proposing is reasonable.
>>
>> But you define a new ABI to specify it in the process, I'd rather fix
>> the flaws by using the common clock ABI than extend any device stuff.
>> If it didn't define a new ABI I'd probably not worry about it but one of
>> the issues we have with DT is that we do end up making ABIs every time
>> we put something in DT.
>
> I understand. This should have been supported by simple-card since the
> beginning. When we tried to move all boards to use simple-card, we hit the
> wall by not being able to select and configure other sysclk_id than 0.
> I don't want to create yet another simple card which handles only sysclk_id 1 ;)
> On the other hand this ABI is backwards compatible since if it is missing it
> will default to the configuration we right now have regarding to sysclk_dir
> and sysclk_id.
>
> I will look at the CCF implementation for McASP first then for aic3x.
The first issue with converting the McASP to use CCF internally for clock
selection, muxing and rate configuration is that the daVinci platform does not
use CCF at all. Given that the davinci-mcasp driver is used by daVinci, we
need to have non CCF way supported in ASoC...
--
P?ter
On Mon, Apr 18, 2016 at 06:50:52PM +0300, Peter Ujfalusi wrote:
> On 02/17/2016 09:52 PM, Peter Ujfalusi wrote:
> > On the other hand this ABI is backwards compatible since if it is missing it
> > will default to the configuration we right now have regarding to sysclk_dir
> > and sysclk_id.
> > I will look at the CCF implementation for McASP first then for aic3x.
> The first issue with converting the McASP to use CCF internally for clock
> selection, muxing and rate configuration is that the daVinci platform does not
> use CCF at all. Given that the davinci-mcasp driver is used by daVinci, we
> need to have non CCF way supported in ASoC...
Well, at least long term we do need daVinci converting to CCF - this is
going to continue to cause problems, devices not part of the SoC can and
do contain clocks and are going to end up being supported via the clock
API.
On 04/18, Mark Brown wrote:
> On Mon, Apr 18, 2016 at 06:50:52PM +0300, Peter Ujfalusi wrote:
> > On 02/17/2016 09:52 PM, Peter Ujfalusi wrote:
>
> > > On the other hand this ABI is backwards compatible since if it is missing it
> > > will default to the configuration we right now have regarding to sysclk_dir
> > > and sysclk_id.
>
> > > I will look at the CCF implementation for McASP first then for aic3x.
>
> > The first issue with converting the McASP to use CCF internally for clock
> > selection, muxing and rate configuration is that the daVinci platform does not
> > use CCF at all. Given that the davinci-mcasp driver is used by daVinci, we
> > need to have non CCF way supported in ASoC...
>
> Well, at least long term we do need daVinci converting to CCF - this is
> going to continue to cause problems, devices not part of the SoC can and
> do contain clocks and are going to end up being supported via the clock
> API.
Does anyone here know what's involved in converting daVinci to
CCF? It doesn't look too far off from what is in the CCF today,
so I'm not sure what's blocking the transition.
--
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project
On 04/22/16 01:29, Stephen Boyd wrote:
>>> The first issue with converting the McASP to use CCF internally for clock
>>> selection, muxing and rate configuration is that the daVinci platform does not
>>> use CCF at all. Given that the davinci-mcasp driver is used by daVinci, we
>>> need to have non CCF way supported in ASoC...
>>
>> Well, at least long term we do need daVinci converting to CCF - this is
>> going to continue to cause problems, devices not part of the SoC can and
>> do contain clocks and are going to end up being supported via the clock
>> API.
>
> Does anyone here know what's involved in converting daVinci to
> CCF? It doesn't look too far off from what is in the CCF today,
> so I'm not sure what's blocking the transition.
Not entirely sure, but most likely new clk driver(s) for daVinci under
drivers/clk/ti/ new set of structures to describe the clocks if the ti_clk* is
not applicable I guess for starter. Support for DT, non DT boots as most of
daVinci is not booting with DT and most likely never will.
It might help to have different daVinci boards for testing the transition. I
only have OMAP-L138-evm. I don't think it is enough for testing an entire
architecture for this big change...
Tero might have better estimates on what is involved when switching an
architecture to CCF from custom, but at least synchronized API - so we don't
need to convert drivers at least.
--
P?ter
On 22/04/16 14:52, Peter Ujfalusi wrote:
> On 04/22/16 01:29, Stephen Boyd wrote:
>>>> The first issue with converting the McASP to use CCF internally for clock
>>>> selection, muxing and rate configuration is that the daVinci platform does not
>>>> use CCF at all. Given that the davinci-mcasp driver is used by daVinci, we
>>>> need to have non CCF way supported in ASoC...
>>>
>>> Well, at least long term we do need daVinci converting to CCF - this is
>>> going to continue to cause problems, devices not part of the SoC can and
>>> do contain clocks and are going to end up being supported via the clock
>>> API.
>>
>> Does anyone here know what's involved in converting daVinci to
>> CCF? It doesn't look too far off from what is in the CCF today,
>> so I'm not sure what's blocking the transition.
>
> Not entirely sure, but most likely new clk driver(s) for daVinci under
> drivers/clk/ti/ new set of structures to describe the clocks if the ti_clk* is
> not applicable I guess for starter. Support for DT, non DT boots as most of
> daVinci is not booting with DT and most likely never will.
> It might help to have different daVinci boards for testing the transition. I
> only have OMAP-L138-evm. I don't think it is enough for testing an entire
> architecture for this big change...
>
> Tero might have better estimates on what is involved when switching an
> architecture to CCF from custom, but at least synchronized API - so we don't
> need to convert drivers at least.
>
Davinci is currently a mutant architecture, it is overriding the common
clk APIs and using its own. Converting these to CCF may open a can of
worms in many ways.
All the clock data should be converted to support CCF, (from
arch/arm/mach-davinci/), along with whatever Peter said.
This also in a situation where many/most upstream people don't even have
davinci devices... Personally I have a grand total of zero davinci
boards on my desk so at least I am unable to work on this right now.
-Tero