Re: [PATCH net-next v15 03/13] dt-bindings: dpll: add SiTime SiT95316 clock generator

From: Ali Rouhi

Date: Thu Nov 19 2026 - 16:20:42 EST


On Mon, 16 Nov 2026 18:18:06 -0800, Jakub Kicinski wrote:

> > + maxItems: 1
> > + description:
> > + The rate wired to the input, in Hz, as a single entry. An
> > + input runs at the rate the board feeds it and has no
> > + frequency to select, so that entry is what the pin reports
> > + as its frequency.
>
> [Severity: Low]
> Does maxItems: 1 actually limit this to a single rate?
>
> dpll-pin.yaml gives supported-frequencies-hz no explicit type, so the type
> comes from the dtschema -hz unit-suffix rule. If that rule makes it a uint32
> matrix, a top-level maxItems only counts bracket groups. A list written the
> way microchip,zl30731.yaml writes one would then count as a single group and
> pass:
>
> supported-frequencies-hz = <10000000 25000000>;
>
> This depends on how dtschema types -hz properties, so it may not apply.
>
> If it does, the driver later in the series takes freqs[0] as the input's rate.
> It still advertises every listed value as a supported frequency, even though
> sit9531x_dpll_input_pin_ops has no .frequency_set.
>
> Would limiting the inner dimension (for example items: [ maxItems: 1 ]) or
> giving an explicit shape guarantee a single value?

dtschema applies maxItems of an integer array to the cells, not only to
the bracket groups: an input pin with `<10000000 25000000>` fails
dt_binding_check. No change.

> > + output-pins:
> > + patternProperties:
> > + "^pin@[0-9a-f]+$":
> > + properties:
> > + reg:
> > + enum: [0, 3, 4, 5, 7, 8, 9, 11]
>
> [Severity: Low]
> On the SiT95317, don't the two numberings in this node collide?
>
> Output pin reg is the die output index. The vddoN-supply names follow the
> SiT95317 datasheet numbering, VDDO0 to VDDO7.
>
> So output-pins/pin@3 is datasheet OUT1. But vddo3-supply powers datasheet
> OUT3, which is die output 5. The same thing happens for 4/7, 5/8 and 7/11.
>
> One supply name also maps to different outputs on the two parts. vddo3 feeds
> reg 3 on the SiT95316 but reg 5 on the SiT95317.
>
> This enum accepts 0, 3, 4, 5 and 7, which are also valid datasheet output
> numbers. A board DT that numbers SiT95317 output pins by datasheet OUTn would
> pass validation for five of the eight outputs. Its label, frequency and
> enable state would then describe a different physical output. Only datasheet
> OUT1, OUT2 and OUT6 would be rejected.
>
> The driver does not enable the supplies today. A later change that looks up
> per-output supplies by pin reg would work on the SiT95316 but switch the wrong
> output driver on the SiT95317.
>
> Could one numbering be used for both reg and the supply names on the
> SiT95317?

Both numberings are intended and the binding text gives the mapping in
both places: reg is the die output, which is what the registers and the
shared SiT95316 numbering use, and a supply is named after the package
pin (VDDOn) as supplies are. A board author reads the pin names from the
datasheet, so renaming VDDO3 to match reg 5 would be the surprising
choice.

> > + /* IN0P and IN0N: pair 0 runs single-ended */
> > + pin@0 {
> > + reg = <0>;
> > + label = "clkin0";
> > + connection-type = "ext";
> > + supported-frequencies-hz = <10000000>;
>
> [Severity: Medium]
> Should the cell width of supported-frequencies-hz be fixed here before this
> becomes ABI?
>
> This example uses 32-bit cells. The only other in-tree DPLL binding that
> uses the same dpll-pin.yaml property, microchip,zl30731.yaml, uses 64-bit
> cells:
>
> supported-frequencies-hz = /bits/ 64 <10000 10000000>;
>
> Its consumer, zl3073x_pin_props_get() in drivers/dpll/zl3073x/prop.c, reads
> the property with fwnode_property_count_u64(). dpll-pin.yaml gives the
> property no type, and this binding does not say which width a SiT9531x node
> has to use.
>
> The SiT9531x driver added later in this series reads it as u32:
>
> drivers/dpll/sit9531x/prop.c:
> /* A -hz property is 32-bit; no rate of this part needs more. */
> num_freqs = fwnode_property_count_u32(props->fwnode,
> "supported-frequencies-hz");
>
> Suppose a SiT9531x node is written the way zl30731.yaml does it:
>
> supported-frequencies-hz = /bits/ 64 <10000000>;
>
> The u32 count is then 2, and fwnode_property_read_u32_array() returns
> {0, 10000000}, because the high word comes first.
>
> With that input, curr_freq = freqs[0] is 0 and the reference frequency is
> never seeded. sit9531x_dpll_input_pin_frequency_get() then reports 0 Hz, and
> the pin also advertises a 0 Hz supported frequency next to the real one.
>
> Output pins have no item limit on this property, so that encoding would pass
> dt_binding_check there too.

Correct that the driver does not detect it. An input written in 64-bit
cells fails dt_binding_check, since the single item is then two 32-bit
cells. An output passes and advertises an extra 0 Hz entry beside each
rate; the rates themselves still work and a request for 0 Hz is refused,
so a check in the driver is not worth the code. No change.