Re: [PATCH v6 8/8] dmaengine: sh: rz-dmac: Add device_{pause,resume}() callbacks

[Claudiu Beznea <claudiu.beznea@tuxon.dev>]

Hi, Biju,

On 12/23/25 16:43, Biju Das wrote:
> Hi Claudiu,
> 
>> -----Original Message-----
>> From: Claudiu <claudiu.beznea@tuxon.dev>
>> Sent: 23 December 2025 13:50
>> Subject: [PATCH v6 8/8] dmaengine: sh: rz-dmac: Add device_{pause,resume}() callbacks
>>
>> From: Claudiu Beznea <claudiu.beznea.uj@bp.renesas.com>
>>
>> Add support for device_{pause, resume}() callbacks. These are required by the RZ/G2L SCIFA driver.
>>
>> Signed-off-by: Claudiu Beznea <claudiu.beznea.uj@bp.renesas.com>
>> ---
>>
>> Changes in v6:
>> - set CHCTRL_SETSUS for pause and CHCTRL_CLRSUS for resume
>> - dropped read-modify-update approach for CHCTRL updates as the
>>    HW returns zero when reading CHCTRL
>> - moved the read_poll_timeout_atomic() under spin lock to
>>    ensure avoid any races b/w pause and resume functionalities
>>
>> Changes in v5:
>> - used suspend capability of the controller to pause/resume
>>    the transfers
>>
>>   drivers/dma/sh/rz-dmac.c | 36 ++++++++++++++++++++++++++++++++++++
>>   1 file changed, 36 insertions(+)
>>
>> diff --git a/drivers/dma/sh/rz-dmac.c b/drivers/dma/sh/rz-dmac.c index 44f0f72cbcf1..377bdd5c9425
>> 100644
>> --- a/drivers/dma/sh/rz-dmac.c
>> +++ b/drivers/dma/sh/rz-dmac.c
>> @@ -135,10 +135,12 @@ struct rz_dmac {
>>   #define CHANNEL_8_15_COMMON_BASE	0x0700
>>
>>   #define CHSTAT_ER			BIT(4)
>> +#define CHSTAT_SUS			BIT(3)
>>   #define CHSTAT_EN			BIT(0)
>>
>>   #define CHCTRL_CLRINTMSK		BIT(17)
>>   #define CHCTRL_CLRSUS			BIT(9)
>> +#define CHCTRL_SETSUS			BIT(8)
>>   #define CHCTRL_CLRTC			BIT(6)
>>   #define CHCTRL_CLREND			BIT(5)
>>   #define CHCTRL_CLRRQ			BIT(4)
>> @@ -827,6 +829,38 @@ static enum dma_status rz_dmac_tx_status(struct dma_chan *chan,
>>   	return status;
>>   }
>>
>> +static int rz_dmac_device_pause(struct dma_chan *chan) {
>> +	struct rz_dmac_chan *channel = to_rz_dmac_chan(chan);
>> +	u32 val;
>> +	int ret;
>> +
>> +	scoped_guard(spinlock_irqsave, &channel->vc.lock) {
> 
>> +		rz_dmac_ch_writel(channel, CHCTRL_SETSUS, CHCTRL, 1);
> 
> 
> Probably first you need to check CHSTAT_EN first before setting CHCTRL_SETSUS??
> 
> As per the hardware manual
> 
> "
> Suspends the current DMA transfer. Setting this bit to 1 when 1 is set in EN of the
> CHSTAT_n/nS register can suspend the current DMA transfer."

OK, I'll update it as follows:

static int rz_dmac_device_pause(struct dma_chan *chan)
{
	struct rz_dmac_chan *channel = to_rz_dmac_chan(chan);
	u32 val;

	guard(spinlock_irqsave)(&channel->vc.lock);

	val = rz_dmac_ch_readl(channel, CHSTAT, 1);
	if (!(val & CHSTAT_EN))
		return 0;

	rz_dmac_ch_writel(channel, CHCTRL_SETSUS, CHCTRL, 1);
	return read_poll_timeout_atomic(rz_dmac_ch_readl, val,
					(val & CHSTAT_SUS), 1, 1024,
					false, channel, CHSTAT, 1);
}

This avoids timeouts reported by read_poll_timeout_atomic() when pause 
is set for a disabled channel.

> 
> 
>> +		ret = read_poll_timeout_atomic(rz_dmac_ch_readl, val,
>> +					       (val & CHSTAT_SUS), 1, 1024,
>> +					       false, channel, CHSTAT, 1);
>> +	}
>> +
>> +	return ret;
>> +}
>> +
>> +static int rz_dmac_device_resume(struct dma_chan *chan) {
>> +	struct rz_dmac_chan *channel = to_rz_dmac_chan(chan);
>> +	u32 val;
>> +	int ret;
>> +
>> +	scoped_guard(spinlock_irqsave, &channel->vc.lock) {
> 
> 
>> +		rz_dmac_ch_writel(channel, CHCTRL_CLRSUS, CHCTRL, 1);
> 
> 
> Similarly, first you need to check  CHSTAT_SUS bit first and then clear suspend state.
> 
> 
> Clears the suspend status. Setting this bit to 1 when 1 is set in SUS of the
> CHSTAT_n/nS register can clear the suspend status.

I'll update this one as follows, to keep the code simple:

static int rz_dmac_device_resume(struct dma_chan *chan)
{
	struct rz_dmac_chan *channel = to_rz_dmac_chan(chan);
	u32 val;

	guard(spinlock_irqsave)(&channel->vc.lock);

         /* Do not check CHSTAT_SUS but rely on HW capabilities. */
	rz_dmac_ch_writel(channel, CHCTRL_CLRSUS, CHCTRL, 1);
	return read_poll_timeout_atomic(rz_dmac_ch_readl, val,
					!(val & CHSTAT_SUS), 1, 1024,
					false, channel, CHSTAT, 1);
}

With this:

1/ in case the channel is not suspended and the CHCTRL_CLRSUS is set, 
the read_poll_timeout_atomic() will not timeout, as the default value of 
the CHSTAT_SUS is zero.

2/ in case the channel is suspended and the CLRSUS is set, it is 
behaving as expected but without an extra check of the CHSTAT_SUS bit 
before setting CHCTRL_CLRSUS.

Thank you,
Claudiu