Re: [PATCH v2] ARM: imx: allow to disable board specific PHY fixups

[Florian Fainelli]

On 3/29/2020 10:26 PM, Oleksij Rempel wrote:
> Hi Andrew,
> 
> On Sun, Mar 29, 2020 at 05:08:54PM +0200, Andrew Lunn wrote:
>> On Sun, Mar 29, 2020 at 01:04:57PM +0200, Oleksij Rempel wrote:
>>
>> Hi Oleksij
>>
>>> +config DEPRECATED_PHY_FIXUPS
>>> +	bool "Enable deprecated PHY fixups"
>>> +	default y
>>> +	---help---
>>> +	  In the early days it was common practice to configure PHYs by adding a
>>> +	  phy_register_fixup*() in the machine code. This practice turned out to
>>> +	  be potentially dangerous, because:
>>> +	  - it affects all PHYs in the system
>>> +	  - these register changes are usually not preserved during PHY reset
>>> +	    or suspend/resume cycle.
>>> +	  - it complicates debugging, since these configuration changes were not
>>> +	    done by the actual PHY driver.
>>> +	  This option allows to disable all fixups which are identified as
>>> +	  potentially harmful and give the developers a chance to implement the
>>> +	  proper configuration via the device tree (e.g.: phy-mode) and/or the
>>> +	  related PHY drivers.
>>
>> This appears to be an IMX only problem. Everybody else seems to of got
>> this right. There is no need to bother everybody with this new
>> option. Please put this in arch/arm/mach-mxs/Kconfig and have IMX in
>> the name.
> 
> Actually, all fixups seems to do wring thing:
> arch/arm/mach-davinci/board-dm644x-evm.c:915:		phy_register_fixup_for_uid(LXT971_PHY_ID, LXT971_PHY_MASK,
> 
> Increased MII drive strength. Should be probably enabled by the PHY
> driver.
> 
> arch/arm/mach-imx/mach-imx6q.c:167:		phy_register_fixup_for_uid(PHY_ID_KSZ9021, MICREL_PHY_ID_MASK,
> arch/arm/mach-imx/mach-imx6q.c:169:		phy_register_fixup_for_uid(PHY_ID_KSZ9031, MICREL_PHY_ID_MASK,
> arch/arm/mach-imx/mach-imx6q.c:171:		phy_register_fixup_for_uid(PHY_ID_AR8031, 0xffffffef,
> arch/arm/mach-imx/mach-imx6q.c:173:		phy_register_fixup_for_uid(PHY_ID_AR8035, 0xffffffef,
> arch/arm/mach-imx/mach-imx6sx.c:40:		phy_register_fixup_for_uid(PHY_ID_AR8031, 0xffffffff,
> arch/arm/mach-imx/mach-imx6ul.c:47:		phy_register_fixup_for_uid(PHY_ID_KSZ8081, MICREL_PHY_ID_MASK,
> arch/arm/mach-imx/mach-imx7d.c:54:		phy_register_fixup_for_uid(PHY_ID_AR8031, 0xffffffff,
> arch/arm/mach-imx/mach-imx7d.c:56:		phy_register_fixup_for_uid(PHY_ID_BCM54220, 0xffffffff,
> arch/arm/mach-mxs/mach-mxs.c:262:		phy_register_fixup_for_uid(PHY_ID_KSZ8051, MICREL_PHY_ID_MASK,
> 
> Fix in some random manner PHY specific errata, enable clock output and
> configure the clock skew.
> 
> arch/arm/mach-orion5x/dns323-setup.c:645:		phy_register_fixup_for_uid(MARVELL_PHY_ID_88E1118,
> 
> Enable LED. Should be done in DT if supported.
> 
> arch/powerpc/platforms/85xx/mpc85xx_mds.c:305:		phy_register_fixup_for_id(phy_id, mpc8568_fixup_125_clock);
> arch/powerpc/platforms/85xx/mpc85xx_mds.c:306:		phy_register_fixup_for_id(phy_id, mpc8568_mds_phy_fixups);
> arch/powerpc/platforms/85xx/mpc85xx_mds.c:311:		phy_register_fixup_for_id(phy_id, mpc8568_mds_phy_fixups);
> 
> Fix in some random manner PHY specific errata, enable clock output and
> configure the clock skew.
> 
> drivers/net/ethernet/dnet.c:818:	err = phy_register_fixup_for_uid(0x01410cc0, 0xfffffff0,
> 
> Enable LED. Should be done in DT if supported.
> 
> drivers/net/usb/lan78xx.c:2071:		ret = phy_register_fixup_for_uid(PHY_KSZ9031RNX, 0xfffffff0,
> drivers/net/usb/lan78xx.c:2078:		ret = phy_register_fixup_for_uid(PHY_LAN8835, 0xfffffff0,
> 
> enable clock output and configure the clock skew.
> 
> As we can see, all of used fixups seem to be wrong. For example micrel
> PHY errata should be fixed in one place for all devices. Not only for
> some iMX6 SoC. I used this option for iMX, because i can test it.

"wrong" is a bit general without really trying to understand the history
of where this came from. Historically, those platforms were not DT
enabled for a while (except PPC) and there was no way to pass platform
specific to the PHY driver so the only way to key off specific
board/platform desired behavior was to register a PHY fixup.

There are also various configuration which are really policies (as in
policy vs. mechanism separation) for the board such as configuring LEDs
in a certain way etc. Very quickly you start putting more of that policy
into DT which is just frowned upon, unless there is a good abstraction
model whereby an Ethernet Device Tree node can also be a LED provider.

> 
>> There is no need to bother everybody with this new
>> option. Please put this in arch/arm/mach-mxs/Kconfig and have IMX in
>> the name.
> 
> A lot of work is needed to fix all of them. I just do not have enough
> time to do it.
> 
>> Having said that, i'm not sure this is the best solution. You cannot
>> build one kernel which runs on all machines.  Did you consider some
>> sort of DT property to disable these fixup? What other ideas did you
>> have before deciding on this solution?
> 
> As soon as all PHY driver support all needed bits used in this fixups,
> we can use drivers on top of fixups. Since changes made by fixups will
> be overwritten by PHY drivers any way. The Kconfig option is needed only for
> developers who has enough resource to investigate this issues and
> mainline needed changes.

We all know this is not going to happen, if people cared so much about
fixing such a problem, it would have been solved by now. If you do care
about IMX though, then please work on removing the fixups, but do not
introduce yet another config that you are guaranteed is going to be
turned on by default, thus creating another test matrix with no real value.
-- 
Florian