RE: [PATCH 2/2] clk: renesas: rzv2h: Deassert reset on assert timeout

[Biju Das]

Hi Philipp Zabel,

> -----Original Message-----
> From: Philipp Zabel <p.zabel@xxxxxxxxxxxxxx>
> Sent: 08 December 2025 11:22
> Subject: Re: [PATCH 2/2] clk: renesas: rzv2h: Deassert reset on assert timeout
> 
> On Mo, 2025-12-08 at 10:13 +0000, Biju wrote:
> > From: Biju Das <biju.das.jz@xxxxxxxxxxxxxx>
> >
> > If the assert() fails due to timeout error, set the reset register bit
> > back to deasserted state. This change is needed especially for
> > handling assert error in suspend() callback that expect the device to
> > be in operational state in case of failure.
> >
> > Signed-off-by: Biju Das <biju.das.jz@xxxxxxxxxxxxxx>
> > ---
> >  drivers/clk/renesas/rzv2h-cpg.c | 5 ++++-
> >  1 file changed, 4 insertions(+), 1 deletion(-)
> >
> > diff --git a/drivers/clk/renesas/rzv2h-cpg.c
> > b/drivers/clk/renesas/rzv2h-cpg.c index 3f6299b9fec0..c0ee2dcc858c
> > 100644
> > --- a/drivers/clk/renesas/rzv2h-cpg.c
> > +++ b/drivers/clk/renesas/rzv2h-cpg.c
> > @@ -1366,8 +1366,11 @@ static int __rzv2h_cpg_assert(struct
> > reset_controller_dev *rcdev,
> >
> >  	ret = readl_poll_timeout_atomic(priv->base + reg, value,
> >  					assert == !!(value & mask), 10, 200);
> > -	if (ret && !assert) {
> > +	if (ret) {
> >  		value = mask << 16;
> > +		if (assert)
> > +			value |= mask;
> > +
> >  		writel(value, priv->base +
> > GET_RST_OFFSET(priv->resets[id].reset_index));
> 
> This writel() could reuse
> 
> 	unsigned int reg = GET_RST_OFFSET(priv->resets[id].reset_index);

For that I need to declare rst_mon for handling GET_RST_MON_OFFSET(priv->resets[id].mon_index) and
reuse the reg variable here.

> 
> >  	}
> 
> How does the hardware behave when __rzv2h_cpg_assert() is called on an already asserted reset? Is it
> possible for the
> readl_poll_timeout_atomic() timeout to trigger, or can this only ever happen for asserted <->
> deasserted transitions? 

Basically, there are 2 separate registers

Reset Control Registers (CPG_RST_m):
0b: Reset on (reset assert)
1b: Reset off (reset_deassert)

Reset Monitor Registers (CPG_RSTMON_m):
0b: The target unit is released from the reset state.
1b: The target unit is in the reset state.

The monitor transition happens, only if there is change is CPG_RST_m register.

If the __rzv2h_cpg_assert() is called on an already asserted reset, readl_poll_timeout_atomic() will return
immediately.

The readl_poll_timeout_atomic() timeout to trigger can only happen during transitions.

> Having a failed
> reset_control_assert() deassert the reset if it was previously asserted would be surprising.

I agree.

Cheers,
Biju