On 2019-09-29 12:36, Biwen Li wrote:
> On some Layerscape boards like LS2085ARDB and LS2088ARDB,
> input pull-up resistors on PCA954x MUX device are missing on board,
> So, if MUX are disabled after powered-on, input lines will float
> leading to incorrect functionality.
Hi!
Are you saying that the parent bus of the mux is relying on some
pull-ups inside the mux?
> Hence, PCA954x MUX device should never be turned-off after
> power-on.
>
> Add property to skip disabling PCA954x MUX device
> if device tree contains "i2c-mux-never-disable"
> for PCA954x device node.
>
> Errata ID: E-00013 on board LS2085ARDB and LS2088ARDB
> (Board revision found on Rev.B, Rev.C and Rev.D)
I think you should follow the example of the i2c-mux-gpio driver
and implement the idle-state property instead.
That is a lot more consistent, assuming it solves the problem
at hand?
>
> Signed-off-by: Biwen Li <biwen.li@xxxxxxx>
> ---
> drivers/i2c/muxes/i2c-mux-pca954x.c | 33 +++++++++++++++++++++++++----
> 1 file changed, 29 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/i2c/muxes/i2c-mux-pca954x.c b/drivers/i2c/muxes/i2c-mux-pca954x.c
> index 923aa3a5a3dc..ea8aca54d572 100644
> --- a/drivers/i2c/muxes/i2c-mux-pca954x.c
> +++ b/drivers/i2c/muxes/i2c-mux-pca954x.c
> @@ -93,6 +93,7 @@ struct pca954x {
> struct irq_domain *irq;
> unsigned int irq_mask;
> raw_spinlock_t lock;
> + u8 disable_mux; /* do not disable mux if val not 0 */
Awful number of negations there. The name is also backwards given that
a non-zero value means that the mux should *not* be disabled. I would
have reused the name from the binding.
bool never_disable;
A bit less confusing...
> };
>
> /* Provide specs for the PCA954x types we know about */
> @@ -258,6 +259,11 @@ static int pca954x_deselect_mux(struct i2c_mux_core *muxc, u32 chan)
> struct i2c_client *client = data->client;
> s8 idle_state;
>
> + if (data->disable_mux != 0) {
Please drop " != 0" and use the variable as a truth value. More
instances below...
> + data->last_chan = data->chip->nchans;
> + return pca954x_reg_write(muxc->parent, client, data->disable_mux);
> + }
> +
> idle_state = READ_ONCE(data->idle_state);
> if (idle_state >= 0)
> /* Set the mux back to a predetermined channel */
> @@ -462,16 +468,32 @@ static int pca954x_probe(struct i2c_client *client,
> }
> }
>
> + /* Errata ID E-00013 on board LS2088ARDB and LS2088ARDB:
> + * The point here is that you must not disable a mux if there
> + * are no pullups on the input or you mess up the I2C. This
> + * needs to be put into the DTS really as the kernel cannot
> + * know this otherwise.
> + */
> +
> + data->disable_mux = np &&
> + of_property_read_bool(np, "i2c-mux-never-disable") &&
i2c-mux-never-disable needs to be documented. However see above for my
point that you should do it like the i2c-mux-gpio driver. Any usage
of idle-state still needs to be documented for pca954x binding.
> + data->chip->muxtype == pca954x_ismux ?
> + data->chip->enable : 0;
Why do you not allow never-disable for switches?
Cheers,
Peter
> +
> /* Write the mux register at addr to verify
> * that the mux is in fact present. This also
> * initializes the mux to disconnected state.
> */
> - if (i2c_smbus_write_byte(client, 0) < 0) {
> + if (i2c_smbus_write_byte(client, data->disable_mux) < 0) {
> dev_warn(dev, "probe failed\n");
> return -ENODEV;
> }
>
> - data->last_chan = 0; /* force the first selection */
> + if (data->disable_mux != 0)
> + data->last_chan = data->chip->nchans;
> + else
> + data->last_chan = 0; /* force the first selection */
> +
> data->idle_state = MUX_IDLE_AS_IS;
>
> idle_disconnect_dt = np &&
> @@ -531,8 +553,11 @@ static int pca954x_resume(struct device *dev)
> struct i2c_mux_core *muxc = i2c_get_clientdata(client);
> struct pca954x *data = i2c_mux_priv(muxc);
>
> - data->last_chan = 0;
> - return i2c_smbus_write_byte(client, 0);
> + if (data->disable_mux != 0)
> + data->last_chan = data->chip->nchans;
> + else
> + data->last_chan = 0;
> + return i2c_smbus_write_byte(client, data->disable_mux);
> }
> #endif
>
>
