On Mon, Apr 11, 2022 at 04:48:15PM +0100, Paul Cercueil wrote:
> Hi Aidan,
>
> Le lun., avril 11 2022 at 16:42:41 +0100, Aidan MacDonald
> <aidanmacdonald.0x0@xxxxxxxxx> a écrit :
> > The TCU clock gate on X1000 wasn't requested by the driver and could
> > be gated automatically later on in boot, which prevents timers from
> > running and breaks PWM.
> >
> > Add a workaround to support old device trees that don't specify the
> > "tcu" clock gate. In this case the kernel will print a warning and
> > attempt to continue without the clock, which is wrong, but it could
> > work if "clk_ignore_unused" is in the kernel arguments.
> >
> > Signed-off-by: Aidan MacDonald <aidanmacdonald.0x0@xxxxxxxxx>
> > ---
> > drivers/clk/ingenic/tcu.c | 38 ++++++++++++++++++++++++++------------
> > 1 file changed, 26 insertions(+), 12 deletions(-)
> >
> > diff --git a/drivers/clk/ingenic/tcu.c b/drivers/clk/ingenic/tcu.c
> > index 77acfbeb4830..ce8c768db997 100644
> > --- a/drivers/clk/ingenic/tcu.c
> > +++ b/drivers/clk/ingenic/tcu.c
> > @@ -31,6 +31,7 @@ struct ingenic_soc_info {
> > unsigned int num_channels;
> > bool has_ost;
> > bool has_tcu_clk;
> > + bool allow_missing_tcu_clk;
> > };
> >
> > struct ingenic_tcu_clk_info {
> > @@ -320,7 +321,8 @@ static const struct ingenic_soc_info
> > jz4770_soc_info = {
> > static const struct ingenic_soc_info x1000_soc_info = {
> > .num_channels = 8,
> > .has_ost = false, /* X1000 has OST, but it not belong TCU */
> > - .has_tcu_clk = false,
> > + .has_tcu_clk = true,
> > + .allow_missing_tcu_clk = true,
> > };
> >
> > static const struct of_device_id __maybe_unused
> > ingenic_tcu_of_match[] __initconst = {
> > @@ -354,15 +356,27 @@ static int __init ingenic_tcu_probe(struct
> > device_node *np)
> > if (tcu->soc_info->has_tcu_clk) {
> > tcu->clk = of_clk_get_by_name(np, "tcu");
> > if (IS_ERR(tcu->clk)) {
> > - ret = PTR_ERR(tcu->clk);
> > - pr_crit("Cannot get TCU clock\n");
> > - goto err_free_tcu;
> > - }
> > -
> > - ret = clk_prepare_enable(tcu->clk);
> > - if (ret) {
> > - pr_crit("Unable to enable TCU clock\n");
> > - goto err_put_clk;
> > + /*
> > + * Old device trees for some SoCs did not include the
> > + * TCU clock because this driver (incorrectly) didn't
> > + * use it. In this case we complain loudly and attempt
> > + * to continue without the clock, which might work if
> > + * booting with workarounds like "clk_ignore_unused".
> > + */
>
> Why not unconditionally enable it instead? Then it would boot without
> clk_ignore_unused.
>
> Cheers,
> -Paul
I could, but why add essentially dead code to the kernel? Maintaining the
old behavior has the "advantage" that it remains broken in the same way as
before, so any workarounds anyone was using will continue to work the same.
And if they were not using workarounds and got a broken kernel, this patch
will not make anything *more* broken, in fact it will not cause any change
in behavior in that case (aside from the warning message).
But if you think it's best to just enable the clock anyway, let me know
and I'll send a new patch.
Regards,
Aidan
>
> > + if (tcu->soc_info->allow_missing_tcu_clk &&
> > + PTR_ERR(tcu->clk) == -EINVAL) {
> > + pr_warn("TCU clock missing from device tree, please update your
> > device tree\n");
> > + tcu->clk = NULL;
> > + } else {
> > + pr_crit("Cannot get TCU clock from device tree\n");
> > + goto err_free_tcu;
> > + }
> > + } else {
> > + ret = clk_prepare_enable(tcu->clk);
> > + if (ret) {
> > + pr_crit("Unable to enable TCU clock\n");
> > + goto err_put_clk;
> > + }
> > }
> > }
> >
> > @@ -432,10 +446,10 @@ static int __init ingenic_tcu_probe(struct
> > device_node *np)
> > clk_hw_unregister(tcu->clocks->hws[i]);
> > kfree(tcu->clocks);
> > err_clk_disable:
> > - if (tcu->soc_info->has_tcu_clk)
> > + if (tcu->clk)
> > clk_disable_unprepare(tcu->clk);
> > err_put_clk:
> > - if (tcu->soc_info->has_tcu_clk)
> > + if (tcu->clk)
> > clk_put(tcu->clk);
> > err_free_tcu:
> > kfree(tcu);
> > --
> > 2.35.1
> >
>
>
