Re: [PATCH v3] of: property: add i2c-parent to fw_devlink supplier bindings

["Abdurrahman Hussain" <abdurrahman@nexthop.ai>]

On Mon Aug 17, 2026 at 2:00 PM PDT, Rob Herring wrote:
> On Mon, Aug 17, 2026 at 07:36:46PM +0000, Roman Gushchin wrote:
>> Rob Herring <robh@kernel.org> writes:
>> 
>> > On Sun, Aug 16, 2026 at 10:17:08PM -0700, Saravana Kannan wrote:
>> >> On Sat, Aug 15, 2026 at 8:33 PM Abdurrahman Hussain
>> >> <abdurrahman@nexthop.ai> wrote:
>> >> >
>> >> > Platform-device i2c muxes (i2c-mux-reg, i2c-mux-gpio, ...) reference
>> >> > the bus they multiplex with an i2c-parent phandle and hold a reference
>> >> > on that adapter from probe until remove. Unlike muxes that are clients
>> >> > on the parent bus, they live outside the adapter's device hierarchy,
>> >> > so nothing orders their teardown against it: if the adapter's device
>> >> > is unbound first — e.g. while reverting a device-tree overlay whose
>> >> > changeset attached the mux nodes before the controller nodes —
>> >> > i2c_del_adapter() blocks forever in
>> >> > wait_for_completion(&adap->dev_released), waiting for a reference that
>> >> > is only dropped later in the same teardown sequence.
>> >> >
>> >> > Teach fw_devlink about i2c-parent so the core creates the
>> >> > corresponding device links: consumers are unbound before the parent
>> >> > adapter's device, and probe ordering no longer needs -EPROBE_DEFER.
>> >> >
>> >> > A plain DEFINE_SIMPLE_PROP() cannot be used because the property has
>> >> > two incompatible layouts: the i2c mux bindings hold a list of bare
>> >> > phandles (i2c-demux-pinctrl takes several), while toshiba,tc9563 holds
>> >> > a single phandle followed by an i2c slave-address cell. The two forms
>> >> > are indistinguishable in the flattened tree, and a 0-cell parse of the
>> >> > tc9563 form would read the slave address as a phandle, linking the
>> >> > consumer to whatever node happens to carry that phandle value. Use a
>> >> > custom parser that only takes entry 0 for toshiba,tc9563 nodes.
>> >> >
>> >> > Signed-off-by: Abdurrahman Hussain <abdurrahman@nexthop.ai>
>> >> > ---
>> >> > Changes in v3:
>> >> > - Fix the tc9563 compatible check: match "pci1179,0623" (the actual
>> >> >   binding compatible) instead of "toshiba,tc9563", which never matched
>> >> >   and let the bare-phandle parser misread the i2c slave-address cell as
>> >> >   a phandle (Sashiko AI review)
>> >> > - Link to v2: https://patch.msgid.link/20260815-b4-of-property-add-i2c-parent-v2-1-239c6da9e097@nexthop.ai
>> >> >
>> >> > Changes in v2:
>> >> > - Use a custom parser instead of DEFINE_SIMPLE_PROP: toshiba,tc9563 puts
>> >> >   an i2c slave-address cell after the phandle, which a 0-cell parse
>> >> >   would misread as a phandle (Sashiko AI review, Rob). Kept the full
>> >> 
>> >> A bit of a tangent, is this a bug in Sashiko bot?
>> >> It just dropped me from the reply in the v1 patch set. Any idea why?
>> 
>> Hm, can you, please, clarify, what do you mean?
>> https://sashiko.dev/#/patchset/20260813-b4-of-property-add-i2c-parent-v1-1-a2487e920ce1%40nexthop.ai
>> has your email in to according to the email policy.
>
> It has mine, but not Saravana's. The problem is Saravana only gets 
> drivers/of/ emails and not all the bindings.
>
> Saravana, If I enable your email, then you will get all the 
> sashiko binding replies. Is that okay?
>
> Rob

Hi Rob, Saravana,

Gentle ping on this one. The v3 thread turned into a discussion about
Sashiko's recipient list, but the patch itself has not had any review
comments since v2.

To recap the state: v2 switched from DEFINE_SIMPLE_PROP() to a custom
parser because toshiba,tc9563 puts an i2c slave-address cell after the
phandle, and v3 fixed the compatible check to match "pci1179,0623",
which is what the tc9563 binding actually uses. Every other i2c-parent
user is a bare phandle list, and i2c-demux-pinctrl relies on the full
list being parsed.

Saravana, since this touches the fw_devlink supplier table, I would
appreciate a look from you in particular. Is the per-compatible special
case acceptable, or would you prefer a different approach for handling
the two property layouts?

Happy to respin if anything is still open.

Thanks,
Abdurrahman