Re: [PATCH v3 07/11] dt-bindings: clock: Add StarFive JH7110 system clock and reset generator

[Hal Feng]

On Thu, 16 Feb 2023 18:20:34 +0000, Conor Dooley wrote:
> Hey Hal!
> 
> On Thu, Feb 16, 2023 at 10:42:20PM +0800, Hal Feng wrote:
>> On Tue, 27 Dec 2022 20:15:20 +0000, Conor Dooley wrote:
>> > On Mon, Dec 26, 2022 at 12:26:32AM +0800, Hal Feng wrote:
>> >> On Tue, 20 Dec 2022 23:14:39 +0000, Conor Dooley wrote:
>> >> > On Tue, Dec 20, 2022 at 08:50:50AM +0800, Hal Feng wrote:
>> >> > > From: Emil Renner Berthing <kernel@xxxxxxxx>
>> >> > > 
>> >> > > Add bindings for the system clock and reset generator (SYSCRG) on the
>> >> > > JH7110 RISC-V SoC by StarFive Ltd.
>> >> > > 
>> >> > > Signed-off-by: Emil Renner Berthing <kernel@xxxxxxxx>
>> >> > > Signed-off-by: Hal Feng <hal.feng@xxxxxxxxxxxxxxxx>
>> > 
>> >> > > +  clocks:
>> >> > > +    items:
>> >> > > +      - description: Main Oscillator (24 MHz)
>> >> > > +      - description: GMAC1 RMII reference
>> >> > > +      - description: GMAC1 RGMII RX
>> >> > > +      - description: External I2S TX bit clock
>> >> > > +      - description: External I2S TX left/right channel clock
>> >> > > +      - description: External I2S RX bit clock
>> >> > > +      - description: External I2S RX left/right channel clock
>> >> > > +      - description: External TDM clock
>> >> > > +      - description: External audio master clock
>> >> > 
>> >> > So, from peeking at the clock driver & the dt - it looks like a bunch of
>> >> > these are not actually required?
>> >> 
>> >> These clocks are used as root clocks or optional parent clocks in clock tree.
>> >> Some of them are optional, but they are required if we want to describe the
>> >> complete clock tree of JH7110 SoC.
>> > 
>> > Perhaps I have a misunderstand of what required means. To me, required
>> > means "you must provide this clock for the SoC to operate in all
>> > configurations".
>> > Optional therefore would be for things that are needed only for some
>> > configurations and may be omitted if not required.
>> > 
>> > From your comment below, boards with a JH7110 may choose not to populate
>> > both external clock inputs to a mux. In that case, "dummy" clocks should
>> > not have to be provided in the DT of such boards to satisfy this binding
>> > which seems wrong to me..
>> 
>> Please see the picture of these external clocks in clock tree.
>> 
>> # mount -t debugfs none /mnt
>> # cat /mnt/clk/clk_summary
>>                                  enable  prepare  protect                                duty  hardware
>>    clock                          count    count    count        rate   accuracy phase  cycle    enable
>> -------------------------------------------------------------------------------------------------------
>>  *mclk_ext*                             0        0        0    12288000          0     0  50000         Y
>>  *tdm_ext*                              0        0        0    49152000          0     0  50000         Y
>>  *i2srx_lrck_ext*                       0        0        0      192000          0     0  50000         Y
>>  *i2srx_bclk_ext*                       0        0        0    12288000          0     0  50000         Y
>>  *i2stx_lrck_ext*                       0        0        0      192000          0     0  50000         Y
>>  *i2stx_bclk_ext*                       0        0        0    12288000          0     0  50000         Y
>>  *gmac1_rgmii_rxin*                     0        0        0   125000000          0     0  50000         Y
>>     gmac1_rx                          0        0        0   125000000          0     0  50000         Y
>>        gmac1_rx_inv                   0        0        0   125000000          0   180  50000         Y
>>  *gmac1_rmii_refin*                     0        0        0    50000000          0     0  50000         Y
>>     gmac1_rmii_rtx                    0        0        0    50000000          0     0  50000         Y
>>        gmac1_tx                       0        0        0    50000000          0     0  50000         N
>>           gmac1_tx_inv                0        0        0    50000000          0   180  50000         Y
>>  *osc*                                  4        4        0    24000000          0     0  50000         Y
>>     apb_func                          0        0        0    24000000          0     0  50000         Y
>>  ...
>> 
>> The clock "gmac1_rgmii_rxin" and the clock "gmac1_rmii_refin" are
>> actually used as the parent of other clocks.
> 
>> The "dummy" clocks
>> you said are all internal clocks.
> 
> No, what I meant by "dummy" clocks is that if you make clocks "required"
> in the binding that are not needed by the hardware for operation a
> customer of yours might have to add "dummy" clocks to their devicetree
> to pass dtbs_check.
> 
>> For the audio related clocks (mclk_ext/tdm_ext/i2srx_lrck_ext/
>> i2srx_bclk_ext/i2stx_lrck_ext/i2stx_bclk_ext), they will be used
>> as the parent clocks in audio related drivers. Note that some
>> clocks need to select different clocks as parent according to
>> requirement.
>> So all these external clocks are required.
>> 
>> > 
>> > It would seem to me that you need to set minItems < maxItems here to
>> > account for that & you do in fact need clock-names.
>> > 
>> >> 
>> >> > I'd have ploughed through this, but having read Krzysztof's comments on
>> >> > the DTS I'm not sure that this binding is correct.
>> >> > https://lore.kernel.org/linux-riscv/20221220011247.35560-1-hal.feng@xxxxxxxxxxxxxxxx/T/#mdf67621a2344dce801aa8015d4963593a2c28bcc
>> >> > 
>> >> > I *think* the DT is correct - the fixed clocks are all inputs from clock
>> >> > sources on the board and as such they are empty in soc.dtsi and are
>> >> > populated in board.dts?
>> >> 
>> >> Yes, the fixed clocks are all clock sources on the board and input to the SoC.
>> >> 
>> >> > 
>> >> > However, are they all actually required? In the driver I see:
>> >> > 	JH71X0__MUX(JH7110_SYSCLK_GMAC1_RX, "gmac1_rx", 2,
>> >> > 		    JH7110_SYSCLK_GMAC1_RGMII_RXIN,
>> >> > 		    JH7110_SYSCLK_GMAC1_RMII_RTX),
>> >> > That macro is:
>> >> > #define JH71X0__MUX(_idx, _name, _nparents, ...) [_idx] = {			\
>> >> > 	.name = _name,								\
>> >> > 	.flags = 0,								\
>> >> > 	.max = ((_nparents) - 1) << JH71X0_CLK_MUX_SHIFT,			\
>> >> > 	.parents = { __VA_ARGS__ },						\
>> >> > }
> 
>> >> > AFAICT, RMII reference feeds RMII_RTX & RGMII RX *is* RGMII_RXIN?
>> >> > Does that mean you need to populate only one of GMAC1 RMII reference
>> >> > and GMAC1 RMGII RX and the other is optional?
> 
>> >> Yes, actually only one of them is chosen as the root clock
>> >> source of the clock "gmac1_rx".
> |  *gmac1_rgmii_rxin*   
> |     gmac1_rx          
> |        gmac1_rx_inv   
> |  *gmac1_rmii_refin*   
> |     gmac1_rmii_rtx    
> |        gmac1_tx       
> |           gmac1_tx_inv
> |
> | description: GMAC1 RMII reference
> | description: GMAC1 RGMII RX
> 
> 
> So you're telling me that you can either:
> - Provide GMAC1 RMII reference and GMAC1 RGMII RX & then use different
>   clocks for gmac1_rx and gmac1_tx
> - Provide only GMAC1 RMII reference & use it for both gmac1_tx *and*
>   gmac1_rx
> 
> Is that correct?

Yes, it is.

Best regards,
Hal

> 
>> >> > 
>> >> > > +
>> >> > > +  clock-names:
>> >> > > +    items:
>> >> > > +      - const: osc
>> >> > > +      - const: gmac1_rmii_refin
>> >> > > +      - const: gmac1_rgmii_rxin
>> >> > > +      - const: i2stx_bclk_ext
>> >> > > +      - const: i2stx_lrck_ext
>> >> > > +      - const: i2srx_bclk_ext
>> >> > > +      - const: i2srx_lrck_ext
>> >> > > +      - const: tdm_ext
>> >> > > +      - const: mclk_ext
>> >> > 
>> >> > If all clocks are in fact required though, isn't this kinda pointless to
>> >> > have since we already know that the order is fixed from the "clocks"
>> >> > property?
>> >> > Krzk/Rob?
>> >> 
>> >> The clock-names are used to easily identify these clocks in the clock driver.
>> > 
>> > *IF* all clocks were in fact required, which they aren't, you could rely
>> > on the order alone in the driver as it is enforced by the binding.
>> 
>> OK, I'll remove "clock-names" property in the bindings and device tree.
>> Instead, will use index to get these clocks in drivers.
> 
> Hang on until you answer my question above before deleting this from the
> dt-binding & driver ;)
>