Re: [PATCH net-next v15 05/13] dpll: sit9531x: read DPLL types and pin properties from system firmware

From: Ali Rouhi

Date: Thu Nov 19 2026 - 16:21:25 EST


On Mon, 16 Nov 2026 18:18:11 -0800, Jakub Kicinski wrote:

> > Read the DPLL type and, for each pin, its label, direction, capabilities
> > and supported frequencies from the firmware node, with defaults for a
> > node that gives none.
>
> [Severity: Low]
> Does this description match what sit9531x_pin_props_get() does?
>
> The direction is not read from the firmware node. It is the dir argument
> passed in by the caller. It only selects which sub-node ("input-pins" or
> "output-pins") sit9531x_prop_pin_fwnode_get() searches.
>
> The capabilities are not read from firmware either. Each pin class gets a
> fixed value: DPLL_PIN_CAPABILITIES_STATE_CAN_CHANGE, or 0 for the XO.
>
> The message also leaves out two things the code does:
>
> - It parses the per-pin connection-type property into DPLL_PIN_TYPE_*.
>
> - The getter writes sitdev->ref[index].freq, using the first
> supported-frequencies-hz entry.
>
> Could the commit message be updated to match the code?

Agreed. The v15 message said the direction and the capabilities come
from the firmware node, and they don't; the message now says which
properties are read from the node and that the driver sets the other
two.

> > + goto err_alloc_ranges;
> > + }
> > +
> > + /* The first listed rate seeds the input's rate */
> > + if (num_freqs > 0 && dir == DPLL_PIN_DIRECTION_INPUT && !xo)
> > + curr_freq = freqs[0];
>
> [Severity: Medium]
> This reads supported-frequencies-hz as u32. Does that match the other
> driver that uses this generic property?
>
> The property is defined in the shared dpll-pin.yaml, which does not give a
> cell width. zl3073x_pin_props_get() reads the same property as u64:
>
> drivers/dpll/zl3073x/prop.c:zl3073x_pin_props_get() {
> ...
> num_freqs = fwnode_property_count_u64(props->fwnode,
> "supported-frequencies-hz");
> ...
> fwnode_property_read_u64_array(props->fwnode,
> "supported-frequencies-hz", freqs,
> num_freqs);
> ...
> }
>
> The microchip,zl30731.yaml examples also use /bits/ 64 for it.
>
> Suppose a SiT9531x board DT follows that convention:
>
> supported-frequencies-hz = /bits/ 64 <10000000>;
>
> That is one item, so it passes the sitime binding's maxItems: 1. But
> fwnode_property_count_u32() returns 2, and freqs[] becomes {0, 10000000}.
>
> curr_freq = freqs[0] is then 0, so sitdev->ref[index].freq never gets
> set. The pin also advertises a DPLL_PIN_FREQUENCY(0) entry next to the
> real rate.
>
> The reverse also breaks: zl3073x would read a 32-bit cell list as packed
> u64 values.
>
> Should the shared binding define one width that both drivers follow, or
> should the sitime binding state the 32-bit width explicitly?

dtschema already types a -hz property as 32-bit cells, which is why an
input written in 64-bit cells fails dt_binding_check. As for the binding
patch, an output written that way advertises an extra 0 Hz entry that
the driver refuses, and the rates still work. No change.