On Sat, Sep 22, 2018 at 08:33:00PM -0700, Nicolin Chen wrote:

On Sat, Sep 22, 2018 at 07:07:02PM -0700, Guenter Roeck wrote:

quoted

On 09/22/2018 05:38 PM, Nicolin Chen wrote:

quoted

On Sat, Sep 22, 2018 at 04:59:55PM -0700, Guenter Roeck wrote:

quoted

quoted

quoted

quoted

+	/* Disable channels if their inputs are disconnected */
+	for (i = 0, mask = 0; i < INA3221_NUM_CHANNELS; i++) {
+		if (ina->inputs[i].disconnected)
+			mask |= INA3221_CONFIG_CHx_EN(i);
+	}

Consequently, you should also _enable_ channels which are not explicitly disabled.

The register has enabled all channels by default. So I felt it'd
be neat to have disabling code only. My v1 actually had enabling
part as well, but I can add it back if you think it'd be better.

quoted

This can be tricky since you'll have to distinguish non-DT and DT configuration
and retain the original configuration if no channel configuration data is available
from devicetree.

quoted

quoted

For non-DT configurations, input->disconnected is always false by
default unless someone adds config for it (through platform_data).
If regmap_update_bits only does disabling like this version does,
non-DT configurations will not get affected since mask = 0. Or if
we change it to do both enabling and disabling, regmap_update_bits
will still ignore since there's no register value changed, though
it won't really hurt even if regmap writes correct configurations
to the register.

For DT configurations (without channel input source defined), it's
like the same as non-DT configurations. As we have platforms only
enabled ina3221 via DT while they don't have this new DT binding,
the driver has to be backward compatible, so my change only sets
input->disconnected=true when a status="disabled" is present, i.e.
those platforms are treated as all channels getting enabled until
they update their DTs.

quoted

I think your assumption may be that the chip is always in its reset state
when Linux is loaded. This is not necessarily the case; it may be
preconfigured by BIOS or ROMMON, or even by someone using i2cset before
loading the driver. If you add enable/disable functionality, you can
not make an assumption about the original state of the chip at probe time;
you have to read it from the chip itself.

I see. That made a point. In that case, I think the simplest way
is probably to do software reset before having configurations.

No. If the chip was configured by the BIOS/ROMMON, it is supposed
to be that way. We can not just override that.

For this driver, it does soft reset in the probe() so we're
sure that all channels are enabled at the moment of calling
this regmap_update_bits. So there's no assumption anymore.

But the case that you mentioned is a good one. It does give
me some insight about the use case and the things that will
need to be careful when adding in[123]_enable. Just it'd be
also possible that BIOS could disable a channel that is not
explicitly disabled in the DT -- then the driver should not
enable it.

Therefore, for both cases, it seems that disabling only the
disconnected channels as this version is a safe solution.

And the driver actually won't update input->disconnected as
this is a physical hardware status that won't be changed. I

[...]

probably could define the input->disconnected in const type.

[...]

Please ignore this line. It'd be read-only then. I got myself
confused as it's not a pointer like label.

For in[123]_enable, it'd be safer to read/write the register
directly.

Thanks
Nicolin

quoted

quoted

The "disconnected" here is to describe the physical connection,
not exact the switch control status because a channel could be
connected but disabled. However, a channel would not be enabled
if it's disconnected. So I think it'd be safe to just disable
the disconnected channels here as this version does, meanwhile,
I will add a soft reset to make sure all channels are enabled.

Thanks
Nicolin