Commit c938ab4da0eb ("net: phy: Manual remove LEDs to ensure correct ordering") correctly fixed a problem with using devm_ but missed removing the LED entry from the LEDs list.
This cause kernel panic on specific scenario where the port for the PHY is torn down and up and the kmod for the PHY is removed.
On setting the port down the first time, the assosiacted LEDs are correctly unregistered. The associated kmod for the PHY is now removed. The kmod is now added again and the port is now put up, the associated LED are registered again. On putting the port down again for the second time after these step, the LED list now have 4 elements. With the first 2 already unregistered previously and the 2 new one registered again.
This cause a kernel panic as the first 2 element should have been removed.
Fix this by correctly removing the element when LED is unregistered.
Reported-by: Daniel Golle daniel@makrotopia.org Tested-by: Daniel Golle daniel@makrotopia.org Cc: stable@vger.kernel.org Fixes: c938ab4da0eb ("net: phy: Manual remove LEDs to ensure correct ordering") Signed-off-by: Christian Marangi ansuelsmth@gmail.com --- drivers/net/phy/phy_device.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
diff --git a/drivers/net/phy/phy_device.c b/drivers/net/phy/phy_device.c index 560e338b307a..499797646580 100644 --- a/drivers/net/phy/phy_device.c +++ b/drivers/net/phy/phy_device.c @@ -3326,10 +3326,11 @@ static __maybe_unused int phy_led_hw_is_supported(struct led_classdev *led_cdev,
static void phy_leds_unregister(struct phy_device *phydev) { - struct phy_led *phyled; + struct phy_led *phyled, *tmp;
- list_for_each_entry(phyled, &phydev->leds, list) { + list_for_each_entry_safe(phyled, tmp, &phydev->leds, list) { led_classdev_unregister(&phyled->led_cdev); + list_del(&phyled->list); } }
It might happen that a PHY driver fails to probe or is not present in the system as it's a kmod. In such case the Device Tree might have LED entry but the Generic PHY is probed instead.
In this scenario, PHY LEDs OF registration should be skipped as controlling the PHY LEDs is not possible.
Tested-by: Daniel Golle daniel@makrotopia.org Cc: stable@vger.kernel.org Fixes: 01e5b728e9e4 ("net: phy: Add a binding for PHY LEDs") Signed-off-by: Christian Marangi ansuelsmth@gmail.com --- drivers/net/phy/phy_device.c | 5 +++++ 1 file changed, 5 insertions(+)
diff --git a/drivers/net/phy/phy_device.c b/drivers/net/phy/phy_device.c index 499797646580..af088bf00bae 100644 --- a/drivers/net/phy/phy_device.c +++ b/drivers/net/phy/phy_device.c @@ -3411,6 +3411,11 @@ static int of_phy_leds(struct phy_device *phydev) struct device_node *leds; int err;
+ /* Skip LED registration if we are Generic PHY */ + if (phy_driver_is_genphy(phydev) || + phy_driver_is_genphy_10g(phydev)) + return 0; + if (!IS_ENABLED(CONFIG_OF_MDIO)) return 0;
On Fri, Oct 04, 2024 at 12:10:05AM +0200, Christian Marangi wrote:
It might happen that a PHY driver fails to probe or is not present in the system as it's a kmod. In such case the Device Tree might have LED entry but the Generic PHY is probed instead.
In this scenario, PHY LEDs OF registration should be skipped as controlling the PHY LEDs is not possible.
Tested-by: Daniel Golle daniel@makrotopia.org Cc: stable@vger.kernel.org Fixes: 01e5b728e9e4 ("net: phy: Add a binding for PHY LEDs") Signed-off-by: Christian Marangi ansuelsmth@gmail.com
drivers/net/phy/phy_device.c | 5 +++++ 1 file changed, 5 insertions(+)
diff --git a/drivers/net/phy/phy_device.c b/drivers/net/phy/phy_device.c index 499797646580..af088bf00bae 100644 --- a/drivers/net/phy/phy_device.c +++ b/drivers/net/phy/phy_device.c @@ -3411,6 +3411,11 @@ static int of_phy_leds(struct phy_device *phydev) struct device_node *leds; int err;
- /* Skip LED registration if we are Generic PHY */
- if (phy_driver_is_genphy(phydev) ||
phy_driver_is_genphy_10g(phydev))return 0;
Why fix it link this, when what you propose for net-next, that the drv ops must also exist, would fix it.
I don't see any need to special case genphy.
Andrew
On Fri, Oct 04, 2024 at 12:50:39AM +0200, Andrew Lunn wrote:
On Fri, Oct 04, 2024 at 12:10:05AM +0200, Christian Marangi wrote:
It might happen that a PHY driver fails to probe or is not present in the system as it's a kmod. In such case the Device Tree might have LED entry but the Generic PHY is probed instead.
In this scenario, PHY LEDs OF registration should be skipped as controlling the PHY LEDs is not possible.
Tested-by: Daniel Golle daniel@makrotopia.org Cc: stable@vger.kernel.org Fixes: 01e5b728e9e4 ("net: phy: Add a binding for PHY LEDs") Signed-off-by: Christian Marangi ansuelsmth@gmail.com
drivers/net/phy/phy_device.c | 5 +++++ 1 file changed, 5 insertions(+)
diff --git a/drivers/net/phy/phy_device.c b/drivers/net/phy/phy_device.c index 499797646580..af088bf00bae 100644 --- a/drivers/net/phy/phy_device.c +++ b/drivers/net/phy/phy_device.c @@ -3411,6 +3411,11 @@ static int of_phy_leds(struct phy_device *phydev) struct device_node *leds; int err;
- /* Skip LED registration if we are Generic PHY */
- if (phy_driver_is_genphy(phydev) ||
phy_driver_is_genphy_10g(phydev))return 0;Why fix it link this, when what you propose for net-next, that the drv ops must also exist, would fix it.
I don't see any need to special case genphy.
While the patch in net-next fix a broken condition (PHY driver exist but doesn't have LEDs OPs), this account a much possible scenario.
It's totally ok if the PHY driver is not loaded and we fallback to the Generic PHY and there are LEDs node.
This is the case with something like ip link set eth0 down rmmod air_en8811h ip link set eth0 up
On this up, the Generic PHY is loaded and LEDs will wrongly be registered. We should not add the LED to the phydev LEDs list.
Do you think this logic is wrong and we should print a warning also in this case? Or should we bite it and just return 0 with no warning at all? (again my concern is the additional LEDs entry in sysfs that won't be actually usable as everything will be rejected)
While the patch in net-next fix a broken condition (PHY driver exist but doesn't have LEDs OPs), this account a much possible scenario.
It's totally ok if the PHY driver is not loaded and we fallback to the Generic PHY and there are LEDs node.
This is the case with something like ip link set eth0 down rmmod air_en8811h ip link set eth0 up
On this up, the Generic PHY is loaded and LEDs will wrongly be registered. We should not add the LED to the phydev LEDs list.
Do you think this logic is wrong and we should print a warning also in this case? Or should we bite it and just return 0 with no warning at all? (again my concern is the additional LEDs entry in sysfs that won't be actually usable as everything will be rejected)
We should not add LEDs which we cannot drive. That much is clear to me.
I would also agree that LEDs in DT which we cannot drive is not fatal. So the return value should be 0.
The only really open point is phydev_err(), phydev_warn() or phydev_dbg(). Since it is not fatal, phydev_err() is wrong. I would probably go with phydev_dbg(), to aid somebody debugging why the LEDs don't appear in some conditions.
Andrew
On Fri, Oct 04, 2024 at 03:44:33PM +0200, Andrew Lunn wrote:
While the patch in net-next fix a broken condition (PHY driver exist but doesn't have LEDs OPs), this account a much possible scenario.
It's totally ok if the PHY driver is not loaded and we fallback to the Generic PHY and there are LEDs node.
This is the case with something like ip link set eth0 down rmmod air_en8811h ip link set eth0 up
On this up, the Generic PHY is loaded and LEDs will wrongly be registered. We should not add the LED to the phydev LEDs list.
Do you think this logic is wrong and we should print a warning also in this case? Or should we bite it and just return 0 with no warning at all? (again my concern is the additional LEDs entry in sysfs that won't be actually usable as everything will be rejected)
We should not add LEDs which we cannot drive. That much is clear to me.
I would also agree that LEDs in DT which we cannot drive is not fatal. So the return value should be 0.
The only really open point is phydev_err(), phydev_warn() or phydev_dbg(). Since it is not fatal, phydev_err() is wrong. I would probably go with phydev_dbg(), to aid somebody debugging why the LEDs don't appear in some conditions.
Ok I will squash this and the net-next patch and change to dbg.
Do you think it's still "net" content? I'm more tempted to post in net-next since I have to drop the Generic PHY condition.
Ok I will squash this and the net-next patch and change to dbg.
Do you think it's still "net" content? I'm more tempted to post in net-next since I have to drop the Generic PHY condition.
Does it cause real problems for users? That is the requirement for stable.
Andrew
On Fri, Oct 04, 2024 at 04:11:22PM +0200, Andrew Lunn wrote:
Ok I will squash this and the net-next patch and change to dbg.
Do you think it's still "net" content? I'm more tempted to post in net-next since I have to drop the Generic PHY condition.
Does it cause real problems for users? That is the requirement for stable.
Not strictly bugs or kernel panic, just annoyance, ok will post to net-next.
On Fri, Oct 04, 2024 at 12:10:04AM +0200, Christian Marangi wrote:
Commit c938ab4da0eb ("net: phy: Manual remove LEDs to ensure correct ordering") correctly fixed a problem with using devm_ but missed removing the LED entry from the LEDs list.
This cause kernel panic on specific scenario where the port for the PHY is torn down and up and the kmod for the PHY is removed.
On setting the port down the first time, the assosiacted LEDs are correctly unregistered. The associated kmod for the PHY is now removed. The kmod is now added again and the port is now put up, the associated LED are registered again. On putting the port down again for the second time after these step, the LED list now have 4 elements. With the first 2 already unregistered previously and the 2 new one registered again.
This cause a kernel panic as the first 2 element should have been removed.
Fix this by correctly removing the element when LED is unregistered.
Reported-by: Daniel Golle daniel@makrotopia.org Tested-by: Daniel Golle daniel@makrotopia.org Cc: stable@vger.kernel.org Fixes: c938ab4da0eb ("net: phy: Manual remove LEDs to ensure correct ordering") Signed-off-by: Christian Marangi ansuelsmth@gmail.com
Reviewed-by: Andrew Lunn andrew@lunn.ch
Andrew
linux-stable-mirror@lists.linaro.org
-
Andrew Lunn -
Christian Marangi