[PATCH] i2c: imx: increase retries on arbitration loss

[Francesco Dolcini]

On Fri, Dec 30, 2022 at 05:12:09PM +0100, Oleksij Rempel wrote:
> On Fri, Dec 30, 2022 at 03:40:58PM +0100, Francesco Dolcini wrote:
> > +Wolfram
> > 
> > On Wed, Dec 28, 2022 at 09:01:46AM +0100, Primoz Fiser wrote:
> > > On 16. 12. 22 13:51, Francesco Dolcini wrote:
> > > > On Fri, Dec 16, 2022 at 01:23:29PM +0100, Primoz Fiser wrote:
> > > > > The only solid point in the thread seems to be that in that case we are not
> > > > > covering up the potential i2c hardware issues?
> > > > 
> > > > I believe that in this case we should just have a warning in the kernel.
> > > > The retry potentially work-around a transient issue and we do not hide any hardware
> > > > issue at the same time. It seems an easy win-win solution.
> > > 
> > > I would agree about throwing a warning message in retry case.
> > > 
> > > Not sure how would it affect other i2c bus drivers using retries > 0.
> > > Retries might be pretty rare with i2c-imx but some other drivers set this to
> > > 5 for example. At least using _ratelimited printk is a must using this
> > > approach.
> > 
> > Wolfram, Uwe, Oleksij
> > 
> > Would it be acceptable to have a warning when we have I2C retries, and
> > with that in place enabling retries on the imx driver?
> > 
> > It exists hardware that requires this to work correctly,
> 
> Well, this is persistent confusion in this monolog. It will not make it
> correctly.
> 
> > and at a
> > minimum setting the retry count from user space is not going to solve
> > potential issues during initial driver probe.
> 
> I assume it is not clear from programmer point of view. Lets try other way:
> 
> - The I2C slave could not correctly interpret the data on SDA because the SDA
>   high or low-level voltages do not reach its appropriate input
>   thresholds.
> 
> This means:
> 
> You have this:
> 
>     /-\    /-\ ----- 2.5Vcc
> ___/   \__/   \___
> 
> Instead of this:
> 
>      /-\     /-\ ----- 3.3Vcc
>     /  \    /   \
> ___/    \__/     \___
> 
> This is bad, because master or slave will not be able to interpret the pick level
> correctly. It may see some times 0 instead of 1. This means, what ever we are
> writing we are to the slave or reading from the slave is potentially corrupt
> and only __sometimes__ the master was able to detect it. 
> 
> - The I2C slave missed an SCL cycle because the SCL high or low-level voltages
>   do not reach its appropriate input thresholds.
> 
> This means, the bus frequency is too high for current configured or physical PCB
> designed. So, you will have different kind of corruptions and some times they
> will be detected. 
> 
> - The I2C slave accidently interpreted a spike etc. as an SCL cycle.
> 
> This means the noise level is to high. The driver strange should be increased
> or PCB redesign should be made. May be there are more options. If not done,
> data corruption can be expected.
> 
> None of this issue can be "fixed" by retries or made more "robust".
> Doing more retries means: we do what ever we do until the system was not able to
> detect the error.

Hello Oleksij,
thanks for the detailed explanation, appreciated.

Given that is it correct that the i2c imx driver return EAGAIN in such a
case (arbitration error)? You made it crystal clear that there is no
such thing as try again for this error, I would be inclined to prepare a
patch to fix this.

diff --git a/drivers/i2c/busses/i2c-imx.c b/drivers/i2c/busses/i2c-imx.c
index cf5bacf3a488..a2a581c8ae07 100644
--- a/drivers/i2c/busses/i2c-imx.c
+++ b/drivers/i2c/busses/i2c-imx.c
@@ -492,7 +492,7 @@ static int i2c_imx_bus_busy(struct imx_i2c_struct *i2c_imx, int for_busy, bool a
                /* check for arbitration lost */
                if (temp & I2SR_IAL) {
                        i2c_imx_clear_irq(i2c_imx, I2SR_IAL);
-                       return -EAGAIN;
+                       return -EIO;
                }

                if (for_busy && (temp & I2SR_IBB)) {


In addition to that is there any valid use case of the i2c retry
mechanism? Is possible for an I2C controller to report anything that can
be recovered with a retry?

Francesco