Re: [PATCH] ASoC: Add support for multi register mux

[Lars-Peter Clausen <lars@metafoo.de>]

On 03/20/2014 08:05 PM, Lars-Peter Clausen wrote:
> On 03/20/2014 07:36 PM, Mark Brown wrote:
>> On Thu, Mar 20, 2014 at 12:20:17PM -0600, Stephen Warren wrote:
>>> On 03/20/2014 05:48 AM, Mark Brown wrote:
>>>> On Wed, Mar 19, 2014 at 04:44:00PM -0700, Arun Shamanna Lakshmi wrote:
>>
>>>>> If each bit of a 32 bit register maps to an input of a mux, then with
>>>>> the current 'soc_enum' structure we cannot have more than 64 inputs
>>>>> for the mux (because of reg and reg2 only).
>>
>>>> What makes you say that?  We currently have devices in mainline which
>>>> have well over 32 inputs to muxes.
>>
>>> I think their register layout is different.
>>
>>> I found a number of large muxes where the register stores a 'integer'
>>> indicating which mux input to select, e.g. Arizona, WM2200, etc. In this
>>> case, an N-bit register could support up to 2^N inputs.
>>
>>> However, the registers in the Tegra AHUB use 1 bit position per input,
>>> and require you to set one single bit at a time. Hence, an N bit
>>> register (or string of registers) can support up to N inputs. In more
>>> recent Tegra chips, we have at least >32 inputs and I think Arun was
>>> saying even >64 inputs. That requires 2 or 3 or more .reg fields in
>>> struct soc_enum.
>>
>> Right, that was my guess too (the mail wasn't terribly clear with the
>> formatting, references to unpublished documents and so on) but that's
>> not a straight mux, it's a value mux, and the limit with the current
>> code is much lower on 32 bit systems (like at least some of the K1s)
>> since muxes only use one of the current register fields.
>
> It might make sense to add special code for supported muxes with a one-hot
> encoding instead of using a value mux. Having an large array where each
> entry is just 1<<n is a bit ugly in my opinion, especially if the value
> needs to be able to be larger than 2**64. But anyway the patch that modifies
> the soc_enum struct should also add the code that makes use of the new
> struct layout.

On the other hand this can actually already be implemented without any core 
changes by using a virtual mux and connecting a supply widget to each input 
which sets the bit for that input. DAPM will take care that only one of the 
supply widgets is enabled at a time.

- Lars