Re: [PATCH net-next v7 8/8] mfd: zl3073x: Register DPLL sub-device during init

On Thu, 22 May 2025, Ivan Vecera wrote:

On 22. 05. 25 12:45 odp., Lee Jones wrote:

quoted

On Thu, 22 May 2025, Ivan Vecera wrote:

quoted

On 22. 05. 25 9:39 dop., Lee Jones wrote:

quoted

On Tue, 13 May 2025, Ivan Vecera wrote:

quoted

On 13. 05. 25 11:41 dop., Lee Jones wrote:

quoted

On Mon, 12 May 2025, Ivan Vecera wrote:

quoted

On 07. 05. 25 5:26 odp., Lee Jones wrote:

quoted

On Wed, 07 May 2025, Andy Shevchenko wrote:

quoted

On Wed, May 07, 2025 at 03:56:37PM +0200, Ivan Vecera wrote:

quoted

On 07. 05. 25 3:41 odp., Andy Shevchenko wrote:

quoted

On Wed, May 7, 2025 at 3:45 PM Ivan Vecera [off-list ref] wrote:

...

quoted

quoted

quoted

+static const struct zl3073x_pdata zl3073x_pdata[ZL3073X_MAX_CHANNELS] = {
+       { .channel = 0, },
+       { .channel = 1, },
+       { .channel = 2, },
+       { .channel = 3, },
+       { .channel = 4, },
+};

quoted

+static const struct mfd_cell zl3073x_devs[] = {
+       ZL3073X_CELL("zl3073x-dpll", 0),
+       ZL3073X_CELL("zl3073x-dpll", 1),
+       ZL3073X_CELL("zl3073x-dpll", 2),
+       ZL3073X_CELL("zl3073x-dpll", 3),
+       ZL3073X_CELL("zl3073x-dpll", 4),
+};

quoted

+#define ZL3073X_MAX_CHANNELS   5

Btw, wouldn't be better to keep the above lists synchronised like

1. Make ZL3073X_CELL() to use indexed variant

[idx] = ...

2. Define the channel numbers

and use them in both data structures.

...

WDYM?

quoted

OTOH, I'm not sure why we even need this. If this is going to be
sequential, can't we make a core to decide which cell will be given
which id?

Just a note that after introduction of PHC sub-driver the array will look
like:
static const struct mfd_cell zl3073x_devs[] = {
           ZL3073X_CELL("zl3073x-dpll", 0),  // DPLL sub-dev for chan 0
           ZL3073X_CELL("zl3073x-phc", 0),   // PHC sub-dev for chan 0
           ZL3073X_CELL("zl3073x-dpll", 1),  // ...
           ZL3073X_CELL("zl3073x-phc", 1),
           ZL3073X_CELL("zl3073x-dpll", 2),
           ZL3073X_CELL("zl3073x-phc", 2),
           ZL3073X_CELL("zl3073x-dpll", 3),
           ZL3073X_CELL("zl3073x-phc", 3),
           ZL3073X_CELL("zl3073x-dpll", 4),
           ZL3073X_CELL("zl3073x-phc", 4),   // PHC sub-dev for chan 4
};

Ah, this is very important piece. Then I mean only this kind of change

enum {
	// this or whatever meaningful names
	..._CH_0	0
	..._CH_1	1
	...
};

static const struct zl3073x_pdata zl3073x_pdata[ZL3073X_MAX_CHANNELS] = {
           { .channel = ..._CH_0, },
           ...
};

static const struct mfd_cell zl3073x_devs[] = {
           ZL3073X_CELL("zl3073x-dpll", ..._CH_0),
           ZL3073X_CELL("zl3073x-phc", ..._CH_0),
           ...
};

This is getting hectic.  All for a sequential enumeration.  Seeing as
there are no other differentiations, why not use IDA in the child
instead?

For that, there have to be two IDAs, one for DPLLs and one for PHCs...

Sorry, can you explain a bit more.  Why is this a problem?

The IDA API is very simple.

Much better than building your own bespoke MACROs.

I will try to explain this in more detail... This MFD driver handles
chip family ZL3073x where the x == number of DPLL channels and can
be from <1, 5>.

The driver creates 'x' DPLL sub-devices during probe and has to pass
channel number that should this sub-device use. Here can be used IDA
in DPLL sub-driver:
e.g. ida_alloc_max(zldev->channels, zldev->max_channels, GFP_KERNEL);

This way the DPLL sub-device get its own unique channel ID to use.

The situation is getting more complicated with PHC sub-devices because
the chip can provide UP TO 'x' PHC sub-devices depending on HW
configuration. To handle this the MFD driver has to check this HW config
for particular channel if it is capable to provide PHC functionality.

E.g. ZL30735 chip has 5 channels, in this case the MFD driver should
create 5 DPLL sub-devices. And then lets say channel 0, 2 and 4 are
PHC capable. Then the MFD driver should create 3 PHC sub-devices and
pass 0, 2 resp. 4 for them.

Where is the code that determines which channels are PHC capable?

It is not included in this series and will be added once the PTP driver
will be added. But the code looks like:

for (i = 0; i < ZL3073X_MAX_CHANNELS; i++) {
	if (channel_is_in_nco_mode(..., i)) {
		struct mfd_cell phc_dev = ZL3073X_CELL("zl3073x-phc", i);
		rc = devm_mfd_add_devices(zldev->dev,
					  PLATFORM_DEVID_AUTO, &phc_dev,
					  1, NULL, 0, NULL);
		...
	}
}

It's the channel_is_in_nco_mode() code I wanted to see.

The function is like this:

static bool zl3073x_chan_in_nco_mode(struct zl3073x_dev *zldev, u8 ch)
{
	u8 mode, mode_refsel;
	int rc;

	rc = zl3073x_read_u8(zlptp->mfd,
			     ZL_REG_DPLL_MODE_REFSEL(ch), &mode_refsel);
	if (rc)
		return false;

	mode = FIELD_GET(ZL_DPLL_MODE_REFSEL_MODE, mode_refsel);

	return (mode == ZL_DPLL_MODE_REFSEL_MODE_NCO);
}

quoted

What if you register all PHC devices, then bomb out if
!channel_is_in_nco_mode()?  Presumably this can / should also live in
the PHC driver as well?

Yes, we can register PHC sub-dev for all channels disregard to channel
mode. The PHC driver checks for the mode and return -ENODEV when it is
different from NCO. But in this case the user will see PHC platform
devices under /sys/bus/platform/device and some of them won't have
driver bound (they will look like some kind of phantom devices).
I'm not sure if this is OK and not confusing.

There will be plenty of devices which do not successfully probe for one
reason or another.  This is all that these 'phantom devices' will
indicate.

-- 
Lee Jones [李琼斯]