On Tue, Dec 16, 2025 at 09:30:08PM +0300, Dan Carpenter wrote:

On Tue, Dec 16, 2025 at 09:42:06AM -0500, Frank Li wrote:

quoted

quoted

quoted

Why not implement standard phy interface,
phy_set_mode_ext(PHY_MODE_ETHERNET, RGMII);

For example:  drivers/pci/controller/dwc/pci-imx6.c

In legency platform, it use syscon to set some registers. It becomes mess
when more platform added.  And it becomes hard to convert because avoid
break compatibltiy now.

It doesn't become worse since new platforms switched to use standard
inteface, (phy, reset ...).

This happens below that layer, this is just saying where the registers
are found.  The GMAC_0_CTRL_STS is just one register in the GPR region,
most of the others are unrelated to PHY.

The other register should work as other function's providor with mfd.

Syscons are a really standard way to do register accesses.

It is quite like back door. Many clock/reset also use phandle to node to
controller by raw register read/write.

The
pci-imx6.c driver you mentioned earlier does it that way...

It is not preferred when we tried to add new one. Give me some time to look
for original threads.

The only
thing which my code does differently is I put the offset into the
phandle, but that's not so unusual and it's arguably a cleaner way
because now both the base address and offset are in the same file.

It is not big deal about offset. The key is if use phande to direct access
other module's register.

Frank

regards,
dan carpenter