Re: [PATCH v3 13/14] reset: convert reset core to using firmware nodes

[Bartosz Golaszewski]

On Tue, Mar 10, 2026 at 3:49 PM Philipp Zabel <p.zabel@xxxxxxxxxxxxxx> wrote:
>
> On Di, 2026-03-10 at 14:16 +0000, Mark Brown wrote:
> > On Fri, Mar 06, 2026 at 06:22:57PM +0100, Bartosz Golaszewski wrote:
> > > With everything else now in place, we can convert the remaining parts of
> > > the reset subsystem to becoming fwnode-agnostic - meaning it will work
> > > with all kinds of firmware nodes, not only devicetree.
> >
> > This patch is causing issues for me in yesterday's -next on a Toradax
> > Mallow platform with a TI K3 AM625.  Things start to go wrong with a
> > WARN_ON():
> >
> > [   11.544610] WARNING: drivers/reset/core.c:1195 at __fwnode_reset_control_get+0x488/0x580, CPU#0: (udev-worker)/129
>
> Thank you, that's reset_controller_register() setting
> fwnode_reset_n_cells to 2 since rcdev->of_node is set and rcdev-
> >of_reset_n_cells == 2:
>
>           if (rcdev->of_node) {
>                   rcdev->fwnode = of_fwnode_handle(rcdev->of_node);
>                   rcdev->fwnode_reset_n_cells = rcdev->of_reset_n_cells;
>           }
>
> and then overwriting it with 1, because rcdev->fwnode is now set as
> well, rcdev->fwnode_xlate is NULL, and rcdev->of_xlate is ignored:
>
>           if (rcdev->fwnode && !rcdev->fwnode_xlate) {
>                   rcdev->fwnode_reset_n_cells = 1;
>                   rcdev->fwnode_xlate = fwnode_reset_simple_xlate;
>           }
>
> That then fails because #reset-cells is <2> for k3_reset.
> We also have to check !rcdev->of_xlate here. Could you check if this
> fixes the issue?
>
> ----------8<----------
> diff --git a/drivers/reset/core.c b/drivers/reset/core.c
> index e625cf59cfb0..d52595db5f87 100644
> --- a/drivers/reset/core.c
> +++ b/drivers/reset/core.c
> @@ -143,7 +143,7 @@ int reset_controller_register(struct reset_controller_dev *rcdev)
>                 rcdev->fwnode_reset_n_cells = rcdev->of_reset_n_cells;
>         }
>
> -       if (rcdev->fwnode && !rcdev->fwnode_xlate) {
> +       if (rcdev->fwnode && !rcdev->fwnode_xlate && !rcdev->of_xlate) {
>                 rcdev->fwnode_reset_n_cells = 1;
>                 rcdev->fwnode_xlate = fwnode_reset_simple_xlate;
>         }
> ---------->8----------
>

Yeah, I just noticed the same. I'm fixing it and will send a patch shortly.

Bart