[v1] ARM i.MX6q: make sure PHY fixup for KSZ9031 is applied only on one board
diff mbox series

Message ID 20191209084430.11107-1-o.rempel@pengutronix.de
State Not Applicable
Delegated to: David Miller
Headers show
Series
  • [v1] ARM i.MX6q: make sure PHY fixup for KSZ9031 is applied only on one board
Related show

Commit Message

Oleksij Rempel Dec. 9, 2019, 8:44 a.m. UTC
Currently all board specific fixups defined in mach-imx*.c are not
working properly. They are applied on all boards with the same iMX SoC
variant and even if:
- PHY needs different settings because of different board design
- PHY needs different settings if it is not connected directly to the SoC.
  For example, the PHY is connected to a switch and the switch is connected
  to the SoC.

Since most PHY drivers don't know about changed default settings, these
settings will not be restored by the PHY driver after reset or
suspend/resume cycle.

This patch changes the MICREL KSZ9031 fixup, which was introduced for
the "Data Modul eDM-QMX6" board in following patch, to be only activated
for this specific board.

|commit dbf6719a4a080669acb5087893704586c791aa41
|Author: Sascha Hauer <s.hauer@pengutronix.de>
|Date:   Thu Jun 20 17:34:33 2013 +0200
|
| ARM: i.MX6: add ethernet phy fixup for KSZ9031
|
| The KSZ9031 is used on the i.MX6 based Data Modul eDM-QMX6
| board. It needs the same fixup to the rx/tx delays as other
| i.MX6 boards.

If some board was working by accident with this fixup, it will probably
break and should be fixed properly by setting related device tree
properties.

To fix this properly the "eDM-QMX6" devicetree:

    arch/arm/boot/dts/imx6q-dmo-edmqmx6.dts

should have board specific *-skew-ps properties. See:

    Documentation/devicetree/bindings/net/micrel-ksz90x1.txt

Cc: Sascha Hauer <s.hauer@pengutronix.de>
Signed-off-by: Oleksij Rempel <o.rempel@pengutronix.de>
---
 arch/arm/mach-imx/mach-imx6q.c | 12 ++++++++++--
 1 file changed, 10 insertions(+), 2 deletions(-)

Comments

Andrew Lunn Dec. 9, 2019, 5:15 p.m. UTC | #1
Hi Oleksij

> This patch changes the MICREL KSZ9031 fixup, which was introduced for
> the "Data Modul eDM-QMX6" board in following patch, to be only activated
> for this specific board.

...

>  static void __init imx6q_enet_phy_init(void)
>  {
> +	/* Warning: please do not extend this fixup list. This fixups are
> +	 * applied even on boards where related PHY is not directly connected
> +	 * to the ethernet controller. For example with switch in the middle.
> +	 */
>  	if (IS_BUILTIN(CONFIG_PHYLIB)) {
>  		phy_register_fixup_for_uid(PHY_ID_KSZ9021, MICREL_PHY_ID_MASK,
>  				ksz9021rn_phy_fixup);
> -		phy_register_fixup_for_uid(PHY_ID_KSZ9031, MICREL_PHY_ID_MASK,
> -				ksz9031rn_phy_fixup);
> +
> +		if (of_machine_is_compatible("dmo,imx6q-edmqmx6"))
> +			phy_register_fixup_for_uid(PHY_ID_KSZ9031,
> +						   MICREL_PHY_ID_MASK,
> +						   ksz9031rn_phy_fixup);
> +
>  		phy_register_fixup_for_uid(PHY_ID_AR8031, 0xffffffef,
>  				ar8031_phy_fixup);
>  		phy_register_fixup_for_uid(PHY_ID_AR8035, 0xffffffef,

What about the other 3 fixups? Are they not also equally broken,
applied for all boards, not specific boards?

	Andrew
Oleksij Rempel Dec. 9, 2019, 5:39 p.m. UTC | #2
Hi Andrew,

On Mon, Dec 09, 2019 at 06:15:08PM +0100, Andrew Lunn wrote:
> Hi Oleksij
> 
> > This patch changes the MICREL KSZ9031 fixup, which was introduced for
> > the "Data Modul eDM-QMX6" board in following patch, to be only activated
> > for this specific board.
> 
> ...
> 
> >  static void __init imx6q_enet_phy_init(void)
> >  {
> > +	/* Warning: please do not extend this fixup list. This fixups are
> > +	 * applied even on boards where related PHY is not directly connected
> > +	 * to the ethernet controller. For example with switch in the middle.
> > +	 */
> >  	if (IS_BUILTIN(CONFIG_PHYLIB)) {
> >  		phy_register_fixup_for_uid(PHY_ID_KSZ9021, MICREL_PHY_ID_MASK,
> >  				ksz9021rn_phy_fixup);
> > -		phy_register_fixup_for_uid(PHY_ID_KSZ9031, MICREL_PHY_ID_MASK,
> > -				ksz9031rn_phy_fixup);
> > +
> > +		if (of_machine_is_compatible("dmo,imx6q-edmqmx6"))
> > +			phy_register_fixup_for_uid(PHY_ID_KSZ9031,
> > +						   MICREL_PHY_ID_MASK,
> > +						   ksz9031rn_phy_fixup);
> > +
> >  		phy_register_fixup_for_uid(PHY_ID_AR8031, 0xffffffef,
> >  				ar8031_phy_fixup);
> >  		phy_register_fixup_for_uid(PHY_ID_AR8035, 0xffffffef,
> 
> What about the other 3 fixups? Are they not also equally broken,
> applied for all boards, not specific boards?

Yes. all of them are broken.
I just trying to not wake all wasp at one time. Most probably there are
board working by accident. So, it will be good to have at least separate
patches for each fixup.
Andrew Lunn Dec. 9, 2019, 5:51 p.m. UTC | #3
> Yes. all of them are broken.
> I just trying to not wake all wasp at one time. Most probably there are
> board working by accident. So, it will be good to have at least separate
> patches for each fixup.

I agree about a patch per fixup. Can you try to generate such patches?
See if there is enough history in git to determine which boards
actually need these fixups?

Thanks
	Andrew
Oleksij Rempel Dec. 10, 2019, 7:54 a.m. UTC | #4
On 09.12.19 18:51, Andrew Lunn wrote:
>> Yes. all of them are broken.
>> I just trying to not wake all wasp at one time. Most probably there are
>> board working by accident. So, it will be good to have at least separate
>> patches for each fixup.
> 
> I agree about a patch per fixup. Can you try to generate such patches?
> See if there is enough history in git to determine which boards
> actually need these fixups?

Ok,

then please ignore this patch. I'll prepare a series of patches related to imx phy fixups.

Kind regards,
Oleksij Rempel
Oleksij Rempel Jan. 20, 2020, 9:50 a.m. UTC | #5
On 09.12.19 18:51, Andrew Lunn wrote:
>> Yes. all of them are broken.
>> I just trying to not wake all wasp at one time. Most probably there are
>> board working by accident. So, it will be good to have at least separate
>> patches for each fixup.
> 
> I agree about a patch per fixup. Can you try to generate such patches?
> See if there is enough history in git to determine which boards
> actually need these fixups?

After some attempts to make it and more discussions, I tend to provide a Kconfig to 
disable/enable this fixups. This will provide an option for users with old devicetree
and possibility to continue development on clean setup.

What do you think about this way and what default state it should be? FIXUPS default on or 
off?

Kind regards,
Oleksij Rempel

Patch
diff mbox series

diff --git a/arch/arm/mach-imx/mach-imx6q.c b/arch/arm/mach-imx/mach-imx6q.c
index 6441281cf4f2..2370cb5d8501 100644
--- a/arch/arm/mach-imx/mach-imx6q.c
+++ b/arch/arm/mach-imx/mach-imx6q.c
@@ -162,11 +162,19 @@  static int ar8035_phy_fixup(struct phy_device *dev)
 
 static void __init imx6q_enet_phy_init(void)
 {
+	/* Warning: please do not extend this fixup list. This fixups are
+	 * applied even on boards where related PHY is not directly connected
+	 * to the ethernet controller. For example with switch in the middle.
+	 */
 	if (IS_BUILTIN(CONFIG_PHYLIB)) {
 		phy_register_fixup_for_uid(PHY_ID_KSZ9021, MICREL_PHY_ID_MASK,
 				ksz9021rn_phy_fixup);
-		phy_register_fixup_for_uid(PHY_ID_KSZ9031, MICREL_PHY_ID_MASK,
-				ksz9031rn_phy_fixup);
+
+		if (of_machine_is_compatible("dmo,imx6q-edmqmx6"))
+			phy_register_fixup_for_uid(PHY_ID_KSZ9031,
+						   MICREL_PHY_ID_MASK,
+						   ksz9031rn_phy_fixup);
+
 		phy_register_fixup_for_uid(PHY_ID_AR8031, 0xffffffef,
 				ar8031_phy_fixup);
 		phy_register_fixup_for_uid(PHY_ID_AR8035, 0xffffffef,