[PATCH] net: fec: fix runtime PM when probing MII bus

In the case where there is no "mdio" bus specified in the devicetree a plain
mdiobus_register() will be called, which tries to probe the connected PHY
by doing mdio_read() on the bus.
Since 6c3e921b18ed (net: fec: Ensure clocks are enabled while using mdio bus)
this needs runtime PM to be available, but as RPM is only later set up in the
FEC probe function those calls will fail, which in turn prevents the FEC
driver to be registered successfully.

Fix this by moving the RPM setup calls before the MII bus probing.

Also move autosuspend init calls before runtime_pm_enable() so that the RPM
callbacks aren't invoked several times during the probe function.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
---
The offending commit got in with v4.2-rc3, so this should be applied as
a fix for 4.2.
---
 drivers/net/ethernet/freescale/fec_main.c | 9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

diff --git a/drivers/net/ethernet/freescale/fec_main.c b/drivers/net/ethernet/freescale/fec_main.c
index 42e20e5385ac..349365d85b92 100644
--- a/drivers/net/ethernet/freescale/fec_main.c
+++ b/drivers/net/ethernet/freescale/fec_main.c

@@ -3431,6 +3431,11 @@ fec_probe(struct platform_device *pdev)
 		fep->reg_phy = NULL;
 	}
 
+	pm_runtime_set_autosuspend_delay(&pdev->dev, FEC_MDIO_PM_TIMEOUT);
+	pm_runtime_use_autosuspend(&pdev->dev);
+	pm_runtime_set_active(&pdev->dev);
+	pm_runtime_enable(&pdev->dev);
+
 	fec_reset_phy(pdev);
 
 	if (fep->bufdesc_ex)

@@ -3465,8 +3470,6 @@ fec_probe(struct platform_device *pdev)
 	netif_carrier_off(ndev);
 	fec_enet_clk_enable(ndev, false);
 	pinctrl_pm_select_sleep_state(&pdev->dev);
-	pm_runtime_set_active(&pdev->dev);
-	pm_runtime_enable(&pdev->dev);
 
 	ret = register_netdev(ndev);
 	if (ret)

@@ -3481,8 +3484,6 @@ fec_probe(struct platform_device *pdev)
 	fep->rx_copybreak = COPYBREAK_DEFAULT;
 	INIT_WORK(&fep->tx_timeout_work, fec_enet_timeout_work);
 
-	pm_runtime_set_autosuspend_delay(&pdev->dev, FEC_MDIO_PM_TIMEOUT);
-	pm_runtime_use_autosuspend(&pdev->dev);
 	pm_runtime_mark_last_busy(&pdev->dev);
 	pm_runtime_put_autosuspend(&pdev->dev);
 

-- 
2.1.4

On Mon, Jul 20, 2015 at 03:48:14PM +0200, Lucas Stach wrote:

In the case where there is no "mdio" bus specified in the devicetree a plain
mdiobus_register() will be called, which tries to probe the connected PHY
by doing mdio_read() on the bus.
Since 6c3e921b18ed (net: fec: Ensure clocks are enabled while using mdio bus)
this needs runtime PM to be available, but as RPM is only later set up in the
FEC probe function those calls will fail, which in turn prevents the FEC
driver to be registered successfully.

Fix this by moving the RPM setup calls before the MII bus probing.

Also move autosuspend init calls before runtime_pm_enable() so that the RPM
callbacks aren't invoked several times during the probe function.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
---
The offending commit got in with v4.2-rc3, so this should be applied as
a fix for 4.2.

Hi Lucas

The patch adding runtime PM has been reverted by David.

Are you O.K. if i fold you fix into my patch and resubmit. I will add
your Signed-off-by.

Thanks
	Andrew

Am Montag, den 20.07.2015, 15:59 +0200 schrieb Andrew Lunn:

On Mon, Jul 20, 2015 at 03:48:14PM +0200, Lucas Stach wrote:

quoted

In the case where there is no "mdio" bus specified in the devicetree a plain
mdiobus_register() will be called, which tries to probe the connected PHY
by doing mdio_read() on the bus.
Since 6c3e921b18ed (net: fec: Ensure clocks are enabled while using mdio bus)
this needs runtime PM to be available, but as RPM is only later set up in the
FEC probe function those calls will fail, which in turn prevents the FEC
driver to be registered successfully.

Fix this by moving the RPM setup calls before the MII bus probing.

Also move autosuspend init calls before runtime_pm_enable() so that the RPM
callbacks aren't invoked several times during the probe function.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
---
The offending commit got in with v4.2-rc3, so this should be applied as
a fix for 4.2.

Hi Lucas

The patch adding runtime PM has been reverted by David.

Are you O.K. if i fold you fix into my patch and resubmit. I will add
your Signed-off-by.

Oh, I had missed that. Thanks for the hint.

I'm ok with folding this patch into yours. Please mention that this
fixes the regression seen on i.MX6 witch caused the revert of the
original patch.

Regards,
Lucas

-- 
Pengutronix e.K.             | Lucas Stach                 |
Industrial Linux Solutions   | http://www.pengutronix.de/  |

On Mon, Jul 20, 2015 at 03:48:14PM +0200, Lucas Stach wrote:

In the case where there is no "mdio" bus specified in the devicetree a plain
mdiobus_register() will be called, which tries to probe the connected PHY
by doing mdio_read() on the bus.
Since 6c3e921b18ed (net: fec: Ensure clocks are enabled while using mdio bus)
this needs runtime PM to be available, but as RPM is only later set up in the
FEC probe function those calls will fail, which in turn prevents the FEC
driver to be registered successfully.

Fix this by moving the RPM setup calls before the MII bus probing.

Also move autosuspend init calls before runtime_pm_enable() so that the RPM
callbacks aren't invoked several times during the probe function.

Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
---
The offending commit got in with v4.2-rc3, so this should be applied as
a fix for 4.2.

Hi David

You emailed saying you had reverted 6c3e921b18ed, but it is still in
net-next/master. If you are not going to revert it, please apply this
patch. I've tested it on my platform as well and it works for me.

If you still plan to revert 6c3e921b18ed, then i will squash this
patch into mine and resubmit.

Thanks
	Andrew

From: Andrew Lunn <andrew@lunn.ch>
Date: Tue, 21 Jul 2015 01:52:24 +0200

You emailed saying you had reverted 6c3e921b18ed, but it is still in
net-next/master.

It's reverted in the 'net' tree, and that's where bug fixes go.

'net' is periodically merged into 'net-next' so even if the revert
isn't there now, it should show up there eventually.