On Fri, Dec 30, 2022 at 4:27 PM Andy Shevchenko <andy@xxxxxxxxxx> wrote:
>
> On Fri, Dec 30, 2022 at 09:46:25AM +0800, Binbin Zhou wrote:
> > On Thu, Dec 29, 2022 at 5:23 PM Andy Shevchenko <andy@xxxxxxxxxx> wrote:
> > >
> > > On Thu, Dec 29, 2022 at 11:20:01AM +0200, Andy Shevchenko wrote:
> > > > On Thu, Dec 29, 2022 at 03:31:47PM +0800, Binbin Zhou wrote:
> > > > > On Wed, Dec 28, 2022 at 5:57 AM Andy Shevchenko <andy@xxxxxxxxxx> wrote:
> > > > > > On Fri, Dec 23, 2022 at 05:00:51PM +0800, Binbin Zhou wrote:
>
> ...
>
> > > > > > > +static int ls2x_i2c_xfer_one(struct ls2x_i2c_priv *priv,
> > > > > > > + struct i2c_msg *msg, bool stop)
> > > > > > > +{
> > > > > > > + int ret;
> > > > > > > + bool is_read = msg->flags & I2C_M_RD;
> > > > > > > +
> > > > > > > + /* Contains steps to send start condition and address */
> > > > > > > + ret = ls2x_i2c_start(priv, msg);
> > > > > > > + if (!ret) {
> > > > > > > + if (is_read)
> > > > > > > + ret = ls2x_i2c_rx(priv, msg->buf, msg->len);
> > > > > > > + else
> > > > > > > + ret = ls2x_i2c_tx(priv, msg->buf, msg->len);
> > > > > > > +
> > > > > > > + if (!ret && stop)
> > > > > > > + ret = ls2x_i2c_stop(priv);
> > > > > > > + }
> > > > > > > +
> > > > > > > + if (ret == -ENXIO)
> > > > > > > + ls2x_i2c_stop(priv);
> > > > > > > + else if (ret < 0)
> > > > > > > + ls2x_i2c_init(priv);
> > > > > > > +
> > > > > > > + return ret;
> > > > > > > +}
> > > > > >
> > > > > > Still this code is odd from reader's perspective. It's in particular not clear
> > > > > > if the stop can be called twice in a row. I recommend to split it to two
> > > > >
> > > > > Sorry,
> > > > > Actually, I don't quite understand why you keep thinking that the stop
> > > > > can be called twice in a row.
> > > >
> > > > Because nothing in the code suggests otherwise. You need deeply understand
> > > > the flow to ensure that it won't. This means that the code is fragile and
> > > > needs refactoring (even comment, which you can do a least won't help, because
> > > > changing code in the other parts may break all this and you won't notice it).
> > > >
> > > > > As I said in my last email, the logic here should be:
> > > > > In the first case, stop is called when the last msg is transmitted successfully;
> > > > > In the second case, stop is called when there is a NOACK during the
> > > > > transmission;
> > > > > In the third case, init is called when other errors occur during the
> > > > > transmission, such as TIMEOUT.
> > > > >
> > > > > The key pointer is the stop function will only return a TIMEOUT error
> > > > > or 0 for success, so if the stop function above is failed, the stop
> > > > > function below will never be called twice.
> > > > >
> > > > > Anyway, I also admit that this part of the code may not be concise and
> > > > > clear enough, and I have tried the following changes:
> > > > >
> > > > > 1. put the start function into the rx/tx function respectively. As followers:
> > > > >
> > > > > @@ -177,10 +177,16 @@ static int ls2x_i2c_start(struct ls2x_i2c_priv
> > > > > *priv, struct i2c_msg *msgs)
> > > > > return ls2x_i2c_send_byte(priv, LS2X_CR_START | LS2X_CR_WRITE);
> > > > > }
> > > > >
> > > > > -static int ls2x_i2c_rx(struct ls2x_i2c_priv *priv, u8 *buf, u16 len)
> > > > > +static int ls2x_i2c_rx(struct ls2x_i2c_priv *priv, struct i2c_msg *msg)
> > > > > {
> > > > > int ret;
> > > > > - u8 rxdata;
> > > > > + u8 rxdata, *buf = msg->buf;
> > > > > + u16 len = msg->len;
> > > > > +
> > > > > + /* Contains steps to send start condition and address */
> > > > > + ret = ls2x_i2c_start(priv, msg);
> > > > > + if (ret)
> > > > > + return ret;
> > > > >
> > > > > while (len--) {
> > > > > ret = ls2x_i2c_xfer_byte(priv,
> > > > > @@ -195,9 +201,16 @@ static int ls2x_i2c_rx(struct ls2x_i2c_priv
> > > > > *priv, u8 *buf, u16 len)
> > > > > return 0;
> > > > > }
> > > > >
> > > > > -static int ls2x_i2c_tx(struct ls2x_i2c_priv *priv, u8 *buf, u16 len)
> > > > > +static int ls2x_i2c_tx(struct ls2x_i2c_priv *priv, struct i2c_msg *msg)
> > > > > {
> > > > > int ret;
> > > > > + u8 *buf = msg->buf;
> > > > > + u16 len = msg->len;
> > > > > +
> > > > > + /* Contains steps to send start condition and address */
> > > > > + ret = ls2x_i2c_start(priv, msg);
> > > > > + if (ret)
> > > > > + return ret;
> > > > >
> > > > > while (len--) {
> > > > > writeb(*buf++, priv->base + I2C_LS2X_TXR);
> > > > >
> > > > > 2. define the variable 'reinit' in the xfer_one function to mark the
> > > > > cases where reinit is needed. As follows:
> > > > >
> > > > > static int ls2x_i2c_xfer_one(struct ls2x_i2c_priv *priv,
> > > > > struct i2c_msg *msg, bool stop)
> > > > > {
> > > > > int ret, ret2;
> > > > > bool reinit = false;
> > > > > bool is_read = msg->flags & I2C_M_RD;
> > > > >
> > > > > if (is_read)
> > > > > ret = ls2x_i2c_rx(priv, msg);
> > > > > else
> > > > > ret = ls2x_i2c_tx(priv, msg);
> > > > >
> > > > > if (ret == -EAGAIN) /* could not acquire bus. bail out without STOP */
> > > > > return ret;
> > > > >
> > > > > if (ret == -ETIMEDOUT) {
> > > > > /* Fatal error. Needs reinit. */
> > > > > stop = false;
> > > > > reinit = true;
> > >
> > > Why do you need to initialize stop here?
> > > Why not to call reinit here and bailout?
> > >
> > > > > }
> > > > >
> > > > > if (stop) {
> > > > > ret2 = ls2x_i2c_stop(priv);
> > > > >
> > > > > if (ret2) {
> > > > > /* Failed to issue STOP. Needs reinit. */
> > > > > reinit = true;
> > > > > ret = ret ?: ret2;
> > >
> > > All the same, try to be less verbose with unneeded variables.
> >
> > Ok, the reinit and ret2 variables seem to be a bit redundant, I will
> > remove them.
> >
> > I will divide the whole thing into two parts:
> > The first part is to handle errors: if ret < 0, return ret directly.
> > One of the special handling is the fatal error timeout, which requires
> > reinit.
> >
> > if (ret < 0) {
> > if (ret == -ETIMEDOUT) /* Fatel error. Needs reinit. */
> > ls2x_i2c_init(priv);
> > return ret;
> > }
> >
> > The second part is to handle the final stop command: it should be
> > noted that if the stop command fails, reinit is also required.
> >
> > if (stop) {
> > /* Failed to issue STOP. Needs reinit. */
> > ret = ls2x_i2c_stop(priv);
> > if (ret)
> > ls2x_i2c_init(priv);
> > }
> >
> > The complete code is as follows:
>
> This looks much better!
> See a couple of nit-picks below.
>
> > static int ls2x_i2c_xfer_one(struct ls2x_i2c_priv *priv,
> > struct i2c_msg *msg, bool stop)
> > {
> > int ret;
>
> > bool is_read = msg->flags & I2C_M_RD;
> >
> > if (is_read)
>
> With this you don't need to have is_read variable.
Well, is_read is a bit redundant, I would use 'msg->flags & I2C_M_RD'
directly in the if judgment condition.
>
> > ret = ls2x_i2c_rx(priv, msg);
> > else
> > ret = ls2x_i2c_tx(priv, msg);
> >
> > if (ret < 0) {
> > if (ret == -ETIMEDOUT) /* Fatel error. Needs reinit. */
>
> Split comment and code, so the comment is followed by the code.
Got it.
In the meantime I'll double check the rest!
Thanks.
Binbin
>
> > ls2x_i2c_init(priv);
> > return ret;
> > }
> >
> > if (stop) {
> > /* Failed to issue STOP. Needs reinit. */
> > ret = ls2x_i2c_stop(priv);
> > if (ret)
> > ls2x_i2c_init(priv);
> > }
> >
> > return ret;
> > }
> >
> > Do you think this is better?
>
> > > > > }
> > > > > }
> > > > >
> > > > > if (reinit)
> > > > > ls2x_i2c_init(priv);
> > > > >
> > > > > return ret;
> > > > > }
> > > > >
> > > > > Do you think this is better?
> > > >
> > > > Slightly, but still twisted at the end with the play of error codes. Try to
> > > > make it even more clear.
> > > >
> > > > > > functions and then do something like
> > > > > >
> > > > > > _read_one()
> > > > > > {
> > > > > > ret = start();
> > > > > > if (ret)
> > > > > > goto _stop; // Do we really need this?
> > > > > >
> > > > > > ret = rx();
> > > > > > if (ret)
> > > > > > goto _stop; // Do we need this?
> > > > > >
> > > > > > /* By setting this call the stop */
> > > > > > if (stop)
> > > > > > ret = -ENXIO;
> > > > > >
> > > > > > out_send_stop:
> > > > > > if (ret == ...)
> > > > > > return _stop();
> > > > > > // I don't like above, so this error checking/setting parts
> > > > > > // also can be rethought and refactored accordingly
> > > > > >
> > > > > > return ret;
> > > > > > }
> > > > > >
> > > > > >
> > > > > > if (is_read)
> > > > > > ret = _read_one();
> > > > > > else
> > > > > > ret = _write_one();
> > > > > >
> > > > > > if (ret)
> > > > > > _init();
> > > > > >
> > > > > > return ret;
>
> --
> With Best Regards,
> Andy Shevchenko
>
>
