On Mon, Apr 19, 2021 at 2:36 PM Tomas Melin <tomas.melin@xxxxxxxxxxx> wrote:
> On 4/19/21 2:14 PM, Andy Shevchenko wrote:
> > On Mon, Apr 19, 2021 at 1:29 PM Tomas Melin <tomas.melin@xxxxxxxxxxx> wrote:
> >> On 4/17/21 3:39 PM, Andy Shevchenko wrote:
> >>> On Fri, Apr 16, 2021 at 5:21 PM Tomas Melin <tomas.melin@xxxxxxxxxxx> wrote:
> >>>> Add initial support for Murata SCA3300 3-axis industrial
> >>>> accelerometer with digital SPI interface. This device also
> >>>> provides a temperature measurement.
> > ...
> >
> >>>> + ret = spi_sync_transfer(sca_data->spi, xfers, ARRAY_SIZE(xfers));
> >>>> + if (ret < 0) {
> >>>> + dev_err(&sca_data->spi->dev,
> >>>> + "transfer error, error: %d\n", ret);
> >>>> + return -EIO;
> >>> Why shadowing error code?
> >> Returning EIO here to have full control over the return value from this
> >> function. As return value of this is used for testing
> > Care to show what kind of testing requires this?
> > Also why can't it be refactored to accept all error codes?
>
> I was referring to this:
>
> +static int sca3300_read_reg(struct sca3300_data *sca_data, u8 reg, int *val)
> +{
> + int ret;
> +
> + mutex_lock(&sca_data->lock);
> + sca_data->txbuf[0] = 0x0 | (reg << 2);
> + ret = sca3300_transfer(sca_data, val);
> + mutex_unlock(&sca_data->lock);
> + if (ret == -EINVAL)
> + ret = sca3300_error_handler(sca_data);
> +
> + return ret;
> +}
> +
> +static int sca3300_write_reg(struct sca3300_data *sca_data, u8 reg, int val)
> +{
> + int reg_val = 0;
> + int ret;
> +
> + mutex_lock(&sca_data->lock);
> + sca_data->txbuf[0] = BIT(7) | (reg << 2);
> + put_unaligned_be16(val, &sca_data->txbuf[1]);
> + ret = sca3300_transfer(sca_data, ®_val);
> + mutex_unlock(&sca_data->lock);
> + if (ret == -EINVAL)
> + ret = sca3300_error_handler(sca_data);
> +
> + return ret;
> +}
>
> So this goes into error handling only when transfer indicates EINVAL
> (which happens when
>
> transfer otherwise is good, but device return status has error flags set
> i message).
In such cases I would recommend introducing your own error space (with
positive numbers) or playing around with the number of transfers (but
this usually works only if you anticipate several of them in a row).
Something like
#define SCA3300_ERROR_FLAGS 1
...
if (ret > 0)
return error_handler(..., ret); // ret in case if you want to
convert the code to something in Linux error code space.
> >> for possible status error (EINVAL), feels more confident to have it like
> >> this to avoid any confusion. And atleast spi_sync_transfer() return value
> >>
> >> would be visible in error message.
> >>>> + }
--
With Best Regards,
Andy Shevchenko
