[PATCH] i2c: at91: support atomic write xfer

[PATCH] i2c: at91: support atomic write xfer

[Michał Mirosław]

Implement basic support for atomic write - enough to get a simple
write to PMIC on shutdown. Only for chips having ALT_CMD register,
eg. SAMA5D2.

Signed-off-by: Michał Mirosław <mirq-linux@rere.qmqm.pl>
---
 drivers/i2c/busses/i2c-at91-master.c | 73 +++++++++++++++++++++++++++-
 1 file changed, 71 insertions(+), 2 deletions(-)

diff --git a/drivers/i2c/busses/i2c-at91-master.c b/drivers/i2c/busses/i2c-at91-master.c
index ba6fbb9c7390..67091616987f 100644
--- a/drivers/i2c/busses/i2c-at91-master.c
+++ b/drivers/i2c/busses/i2c-at91-master.c
@@ -15,6 +15,7 @@
 
 #include <linux/clk.h>
 #include <linux/completion.h>
+#include <linux/delay.h>
 #include <linux/dma-mapping.h>
 #include <linux/dmaengine.h>
 #include <linux/err.h>
@@ -709,6 +710,73 @@ static int at91_twi_xfer(struct i2c_adapter *adap, struct i2c_msg *msg, int num)
 	return ret;
 }
 
+static int at91_twi_xfer_atomic(struct i2c_adapter *adap, struct i2c_msg *msg, int num)
+{
+	struct at91_twi_dev *dev = i2c_get_adapdata(adap);
+	unsigned long timeout;
+	__u32 stat;
+	int ret;
+
+	/* FIXME: only single write request supported to 7-bit addr */
+	if (num != 1)
+		return -EOPNOTSUPP;
+	if (msg->flags & I2C_M_RD)
+		return -EOPNOTSUPP;
+	if (msg->flags & I2C_M_TEN)
+		return -EOPNOTSUPP;
+	if (msg->len > dev->fifo_size && msg->len > 1)
+		return -EOPNOTSUPP;
+	if (!dev->pdata->has_alt_cmd)
+		return -EOPNOTSUPP;
+
+	ret = pm_runtime_get_sync(dev->dev);
+	if (ret < 0)
+		goto out;
+
+	/* Clear and disable pending interrupts, such as NACK. */
+	at91_twi_read(dev, AT91_TWI_SR);
+	at91_twi_write(dev, AT91_TWI_IDR, ~0);
+
+	at91_twi_write(dev, AT91_TWI_MMR, msg->addr << 16);
+
+	if (!msg->len) {
+		at91_twi_write(dev, AT91_TWI_CR,
+			       AT91_TWI_ACMDIS | AT91_TWI_QUICK);
+	} else {
+		size_t n = msg->len;
+		__u8 *p;
+
+		at91_twi_write(dev, AT91_TWI_CR,
+				    AT91_TWI_ACMEN |
+				    AT91_TWI_THRCLR | AT91_TWI_RHRCLR);
+		at91_twi_write(dev, AT91_TWI_ACR, AT91_TWI_ACR_DATAL(n));
+		for (p = msg->buf; n--; ++p)
+			writeb_relaxed(*p, dev->base + AT91_TWI_THR);
+	}
+
+	timeout = jiffies + (2 + msg->len) * HZ/1000;
+	for (;;) {
+		stat = at91_twi_read(dev, AT91_TWI_SR);
+		if (stat & AT91_TWI_TXCOMP)
+			break;
+		if (time_after(jiffies, timeout)) {
+			ret = -ETIMEDOUT;
+			goto out;
+		}
+		udelay(100);
+	}
+
+	if (stat & AT91_TWI_NACK)
+		ret = -EREMOTEIO;
+	else
+		ret = num;
+out:
+	pm_runtime_mark_last_busy(dev->dev);
+	pm_runtime_put_autosuspend(dev->dev);
+
+	return ret;
+}
+
 /*
  * The hardware can handle at most two messages concatenated by a
  * repeated start via it's internal address feature.
@@ -725,8 +793,9 @@ static u32 at91_twi_func(struct i2c_adapter *adapter)
 }
 
 static const struct i2c_algorithm at91_twi_algorithm = {
-	.master_xfer	= at91_twi_xfer,
-	.functionality	= at91_twi_func,
+	.master_xfer		= at91_twi_xfer,
+	.master_xfer_atomic	= at91_twi_xfer_atomic,
+	.functionality		= at91_twi_func,
 };
 
 static int at91_twi_configure_dma(struct at91_twi_dev *dev, u32 phy_addr)
-- 
2.20.1

Re: [PATCH] i2c: at91: support atomic write xfer

[Dmitry Osipenko]

15.03.2020 21:27, Michał Mirosław пишет:
> Implement basic support for atomic write - enough to get a simple
> write to PMIC on shutdown. Only for chips having ALT_CMD register,
> eg. SAMA5D2.
> 
> Signed-off-by: Michał Mirosław <mirq-linux@rere.qmqm.pl>
> ---

Hello Michał,

...
> +	ret = pm_runtime_get_sync(dev->dev);
> +	if (ret < 0)
> +		goto out;

Runtime PM can't be used while interrupts are disabled, unless
pm_runtime_irq_safe() is used and driver's RPM callback is IRQ-safe.

...
> +	timeout = jiffies + (2 + msg->len) * HZ/1000;
> +	for (;;) {
> +		stat = at91_twi_read(dev, AT91_TWI_SR);
> +		if (stat & AT91_TWI_TXCOMP)
> +			break;
> +		if (time_after(jiffies, timeout)) {
> +			ret = -ETIMEDOUT;
> +			goto out;
> +		}
> +		udelay(100);
> +	}

Jiffies can't be used with the disabled interrupts because jiffies are
updated by timer's interrupt.

Either ktime() API or iterator-based loop should be used.

Re: [PATCH] i2c: at91: support atomic write xfer

[Michał Mirosław]

On Sun, Mar 15, 2020 at 11:46:33PM +0300, Dmitry Osipenko wrote:
> 15.03.2020 21:27, Michał Mirosław пишет:
> > Implement basic support for atomic write - enough to get a simple
> > write to PMIC on shutdown. Only for chips having ALT_CMD register,
> > eg. SAMA5D2.
> > 
> > Signed-off-by: Michał Mirosław <mirq-linux@rere.qmqm.pl>
> > ---
> 
> Hello Michał,
> 
> ...
> > +	ret = pm_runtime_get_sync(dev->dev);
> > +	if (ret < 0)
> > +		goto out;
> 
> Runtime PM can't be used while interrupts are disabled, unless
> pm_runtime_irq_safe() is used and driver's RPM callback is IRQ-safe.

I didn't get any warnings from lockdep and friends, but I'll double
check if this is by luck.

> ...
> > +	timeout = jiffies + (2 + msg->len) * HZ/1000;
> > +	for (;;) {
> > +		stat = at91_twi_read(dev, AT91_TWI_SR);
> > +		if (stat & AT91_TWI_TXCOMP)
> > +			break;
> > +		if (time_after(jiffies, timeout)) {
> > +			ret = -ETIMEDOUT;
> > +			goto out;
> > +		}
> > +		udelay(100);
> > +	}
> 
> Jiffies can't be used with the disabled interrupts because jiffies are
> updated by timer's interrupt.
> 
> Either ktime() API or iterator-based loop should be used.

Thanks for the pointers. In my use-case power is cut from the CPU at this
point so it didn't matter that the loop was infinite.

Best Regards
Michał Mirosław

Re: [PATCH] i2c: at91: support atomic write xfer

[Stefan Lengfeld]

Hi Michał,

On Mon, Mar 16, 2020 at 03:42:21PM +0100, Michał Mirosław wrote:
> On Sun, Mar 15, 2020 at 11:46:33PM +0300, Dmitry Osipenko wrote:
> > 15.03.2020 21:27, Michał Mirosław пишет:
> > > Implement basic support for atomic write - enough to get a simple
> > > write to PMIC on shutdown. Only for chips having ALT_CMD register,
> > > eg. SAMA5D2.
> > > 
> > > Signed-off-by: Michał Mirosław <mirq-linux@rere.qmqm.pl>
> > > ---
> > 
> > Hello Michał,
> > 
> > ...
> > > +	ret = pm_runtime_get_sync(dev->dev);
> > > +	if (ret < 0)
> > > +		goto out;
> > 
> > Runtime PM can't be used while interrupts are disabled, unless
> > pm_runtime_irq_safe() is used and driver's RPM callback is IRQ-safe.
> 
> I didn't get any warnings from lockdep and friends, but I'll double
> check if this is by luck.

You can have a look at the I2C atomic patch for the imx-driver. See

   https://patchwork.ozlabs.org/patch/1225802/

In that patch Marco Felsch is using clk_enable() and clk_disable() calls.

> 
> > ...
> > > +	timeout = jiffies + (2 + msg->len) * HZ/1000;
> > > +	for (;;) {
> > > +		stat = at91_twi_read(dev, AT91_TWI_SR);
> > > +		if (stat & AT91_TWI_TXCOMP)
> > > +			break;
> > > +		if (time_after(jiffies, timeout)) {
> > > +			ret = -ETIMEDOUT;
> > > +			goto out;
> > > +		}
> > > +		udelay(100);
> > > +	}
> > 
> > Jiffies can't be used with the disabled interrupts because jiffies are
> > updated by timer's interrupt.
> > 
> > Either ktime() API or iterator-based loop should be used.
> 
> Thanks for the pointers. In my use-case power is cut from the CPU at this
> point so it didn't matter that the loop was infinite.

Here again you can have a look at Marco Felsch's patch. He used the
function readb_poll_timeout_atomic(). So the loop can potentially
replaced by a single line.

Kind regards,
Stefan

Re: [PATCH] i2c: at91: support atomic write xfer

[Marco Felsch]

On 20-03-16 18:20, Stefan Lengfeld wrote:
> Hi Michał,
> 
> On Mon, Mar 16, 2020 at 03:42:21PM +0100, Michał Mirosław wrote:
> > On Sun, Mar 15, 2020 at 11:46:33PM +0300, Dmitry Osipenko wrote:
> > > 15.03.2020 21:27, Michał Mirosław пишет:
> > > > Implement basic support for atomic write - enough to get a simple
> > > > write to PMIC on shutdown. Only for chips having ALT_CMD register,
> > > > eg. SAMA5D2.
> > > > 
> > > > Signed-off-by: Michał Mirosław <mirq-linux@rere.qmqm.pl>
> > > > ---
> > > 
> > > Hello Michał,
> > > 
> > > ...
> > > > +	ret = pm_runtime_get_sync(dev->dev);
> > > > +	if (ret < 0)
> > > > +		goto out;
> > > 
> > > Runtime PM can't be used while interrupts are disabled, unless
> > > pm_runtime_irq_safe() is used and driver's RPM callback is IRQ-safe.
> > 
> > I didn't get any warnings from lockdep and friends, but I'll double
> > check if this is by luck.
> 
> You can have a look at the I2C atomic patch for the imx-driver. See
> 
>    https://patchwork.ozlabs.org/patch/1225802/
> 
> In that patch Marco Felsch is using clk_enable() and clk_disable() calls.

Yep because we need to handle the runtime_pm stuff by our-self. So for
the imx case we need to handle the clk en-/disable stuff. Runtime pm is
using a workqueue which can't be used in that late case.

Regards,
  Marco

> > > ...
> > > > +	timeout = jiffies + (2 + msg->len) * HZ/1000;
> > > > +	for (;;) {
> > > > +		stat = at91_twi_read(dev, AT91_TWI_SR);
> > > > +		if (stat & AT91_TWI_TXCOMP)
> > > > +			break;
> > > > +		if (time_after(jiffies, timeout)) {
> > > > +			ret = -ETIMEDOUT;
> > > > +			goto out;
> > > > +		}
> > > > +		udelay(100);
> > > > +	}
> > > 
> > > Jiffies can't be used with the disabled interrupts because jiffies are
> > > updated by timer's interrupt.
> > > 
> > > Either ktime() API or iterator-based loop should be used.
> > 
> > Thanks for the pointers. In my use-case power is cut from the CPU at this
> > point so it didn't matter that the loop was infinite.
> 
> Here again you can have a look at Marco Felsch's patch. He used the
> function readb_poll_timeout_atomic(). So the loop can potentially
> replaced by a single line.
> 
> Kind regards,
> Stefan
> 

-- 
Pengutronix e.K.                           |                             |
Steuerwalder Str. 21                       | http://www.pengutronix.de/  |
31137 Hildesheim, Germany                  | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |