Probably due to a lot of refactorization, the hardware was not brought
into a known state in probe. This may be a problem when a hypervisor
restarts Linux without resetting the hardware, leaving an old state
running. Make sure the hardware gets initialized, especially interrupts
should be cleared and disabled.

Reported-by: Dirk Behme <dirk.behme@de.bosch.com>
Reported-by: Geert Uytterhoeven <geert+renesas@glider.be>
Closes: https://lore.kernel.org/r/20240702045535.2000393-1-dirk.behme@de.bosch.com
Signed-off-by: Wolfram Sang <wsa+renesas@sang-engineering.com>
---

Here is my proposal to fix the issue reported by Dirk. Build tested.
I can do proper testing on HW only tomorrow. But so you know already...

It is strange to add another "_slave" function to the driver while I
work on removing such language from I2C somewhere else. "Consistency" is
the answer here. The driver will be converted as well. But then as a
whole.

 drivers/i2c/busses/i2c-rcar.c | 17 ++++++++++++-----
 1 file changed, 12 insertions(+), 5 deletions(-)

diff --git a/drivers/i2c/busses/i2c-rcar.c b/drivers/i2c/busses/i2c-rcar.c
index 828aa2ea0fe4..00a885205075 100644
--- a/drivers/i2c/busses/i2c-rcar.c
+++ b/drivers/i2c/busses/i2c-rcar.c
@@ -257,6 +257,14 @@ static void rcar_i2c_init(struct rcar_i2c_priv *priv)
 	}
 }
 
+static void rcar_i2c_init_slave(struct rcar_i2c_priv *priv)
+{
+	rcar_i2c_write(priv, ICSIER, 0);
+	rcar_i2c_write(priv, ICSSR, 0);
+	rcar_i2c_write(priv, ICSCR, SDBS);
+	rcar_i2c_write(priv, ICSAR, 0); /* Gen2: must be 0 if not using slave */
+}
+
 static int rcar_i2c_bus_barrier(struct rcar_i2c_priv *priv)
 {
 	int ret;
@@ -1033,11 +1041,8 @@ static int rcar_unreg_slave(struct i2c_client *slave)
 
 	/* ensure no irq is running before clearing ptr */
 	disable_irq(priv->irq);
-	rcar_i2c_write(priv, ICSIER, 0);
-	rcar_i2c_write(priv, ICSSR, 0);
+	rcar_i2c_init_slave(priv);
 	enable_irq(priv->irq);
-	rcar_i2c_write(priv, ICSCR, SDBS);
-	rcar_i2c_write(priv, ICSAR, 0); /* Gen2: must be 0 if not using slave */
 
 	priv->slave = NULL;
 
@@ -1152,7 +1157,9 @@ static int rcar_i2c_probe(struct platform_device *pdev)
 		goto out_pm_disable;
 	}
 
-	rcar_i2c_write(priv, ICSAR, 0); /* Gen2: must be 0 if not using slave */
+	/* Bring hardware to known state */
+	rcar_i2c_init(priv);
+	rcar_i2c_init_slave(priv);
 
 	if (priv->devtype < I2C_RCAR_GEN3) {
 		irqflags |= IRQF_NO_THREAD;
-- 
2.43.0

Hi Wolfram,

On Wed, Jul 03, 2024 at 09:12:03AM GMT, Wolfram Sang wrote:
> Probably due to a lot of refactorization, the hardware was not brought
> into a known state in probe. This may be a problem when a hypervisor
> restarts Linux without resetting the hardware, leaving an old state
> running. Make sure the hardware gets initialized, especially interrupts
> should be cleared and disabled.
> 
> Reported-by: Dirk Behme <dirk.behme@de.bosch.com>
> Reported-by: Geert Uytterhoeven <geert+renesas@glider.be>
> Closes: https://lore.kernel.org/r/20240702045535.2000393-1-dirk.behme@de.bosch.com
> Signed-off-by: Wolfram Sang <wsa+renesas@sang-engineering.com>

Do we need the Fixes tag here?

> ---
> 
> Here is my proposal to fix the issue reported by Dirk. Build tested.
> I can do proper testing on HW only tomorrow. But so you know already...

Looks reasonable. If testing is fine I can queue this up for this
week's pull request.

> It is strange to add another "_slave" function to the driver while I
> work on removing such language from I2C somewhere else. "Consistency" is
> the answer here. The driver will be converted as well. But then as a
> whole.

Ack!

Thanks,
Andi

> Do we need the Fixes tag here?

We can add the initial commit for this driver. That means the commit
message is wrong because refactorization is not the culprit. So, I will
resend with Fixes-tag and updated commit message.

> Looks reasonable. If testing is fine I can queue this up for this
> week's pull request.

First testing looks good. I will do some more.