net: phy: avoid NPE if read_page/write_page callbacks are not available
Subsystems (3)
Heiner Kallweit
maintainerPATCH 1/1 net: phy: avoid NPE if read_page/write_page callbacks are not available
Currently there's a bug in the module subsystem [0] preventing load of the PHY driver module on certain systems (as one symptom). This results in a NPE on such systems for the following reason:
Instead of the correct PHY driver the genphy driver is loaded that doesn't implement the read_page/write_page callbacks. Every call to phy_read_paged() et al will result in a NPE therefore.
In parallel to fixing the root cause we should make sure that this one and maybe similar issues in other subsystems don't result in a NPE in phylib. So let's check for the callbacks before using them and warn once if they are not available.
[0] https://marc.info/?t=157072642100001&r=1&w=2
--- drivers/net/phy/phy-core.c | 6 ++++++ 1 file changed, 6 insertions(+)
drivers/net/phy/phy-core.c +6 -0
@@ -689,11 +689,17 @@ EXPORT_SYMBOL_GPL(phy_modify_mmd);static int __phy_read_page (struct phy_device *phydev) { +if (WARN_ONCE (!phydev->drv->read_page,"read_page callback not available, PHY driver not loaded?\n" )) +return -EOPNOTSUPP ; +return phydev->drv->read_page (phydev); }static int __phy_write_page (struct phy_device *phydev,int page) { +if (WARN_ONCE (!phydev->drv->write_page,"write_page callback not available, PHY driver not loaded?\n" )) +return -EOPNOTSUPP ; +return phydev->drv->write_page (phydev, page); }
Signature · 1 line
2.23.0
Quoted 11 lines · Heiner Kallweit
Currently there's a bug in the module subsystem [0] preventing load of the PHY driver module on certain systems (as one symptom). This results in a NPE on such systems for the following reason:
Instead of the correct PHY driver the genphy driver is loaded that doesn't implement the read_page/write_page callbacks. Every call tophy_read_paged()et al will result in a NPE therefore.In parallel to fixing the root cause we should make sure that this one and maybe similar issues in other subsystems don't result in a NPE in phylib. So let's check for the callbacks before using them and warn once if they are not available.
Everywhere else in the PHY library we tend to do:
if (!phydev->drv)
return -EIO;
maybe not the best choice for an error code, but we should be consistent.
Is the issue really that we do have a driver we are bound to, but somehow we cannot resolve the read_page/write_page callbacks to a valid function pointer?
Quoted 31 lines
[0] https://marc.info/?t=157072642100001&r=1&w=2
Signed-off-by Heiner Kallweit <[email protected]>--- drivers/net/phy/phy-core.c | 6 ++++++ 1 file changed, 6 insertions(+)drivers/net/phy/phy-core.c +6
-0@@ -689,11 +689,17 @@ EXPORT_SYMBOL_GPL(phy_modify_mmd);static int __phy_read_page (struct phy_device *phydev) { +if (WARN_ONCE (!phydev->drv->read_page,"read_page callback not available, PHY driver not loaded?\n" )) +return -EOPNOTSUPP ; +return phydev->drv->read_page (phydev); }static int __phy_write_page (struct phy_device *phydev,int page) { +if (WARN_ONCE (!phydev->drv->write_page,"write_page callback not available, PHY driver not loaded?\n" )) +return -EOPNOTSUPP ; +return phydev->drv->write_page (phydev, page); }
Signature · 1 line
Florian
Heiner Kallweit
maintainerQuoted 13 lines · Florian Fainelli
Quoted 11 lines · Heiner Kallweit
Currently there's a bug in the module subsystem [0] preventing load of the PHY driver module on certain systems (as one symptom). This results in a NPE on such systems for the following reason:
Instead of the correct PHY driver the genphy driver is loaded that doesn't implement the read_page/write_page callbacks. Every call tophy_read_paged()et al will result in a NPE therefore.In parallel to fixing the root cause we should make sure that this one and maybe similar issues in other subsystems don't result in a NPE in phylib. So let's check for the callbacks before using them and warn once if they are not available.
Quoted 5 lines
Everywhere else in the PHY library we tend to do:
if (!phydev->drv)
return -EIO;
A driver is bound, but genphy instead of the dedicated one. I think this check won't catch the error.
Quoted 6 lines
maybe not the best choice for an error code, but we should be consistent.
Is the issue really that we do have a driver we are bound to, but somehow we cannot resolve the read_page/write_page callbacks to a valid function pointer?
The issue is that loading the dedicated PHY driver module fails, therefore the (built-in) genphy driver is bound. Code in the r8169 driver expects that the dedicated driver for the internal PHY is bound and uses phy_read_page() et al what fails miserably with the genphy driver.
Quoted 33 lines
Quoted 31 lines
[0] https://marc.info/?t=157072642100001&r=1&w=2
Signed-off-by Heiner Kallweit <[email protected]>--- drivers/net/phy/phy-core.c | 6 ++++++ 1 file changed, 6 insertions(+)drivers/net/phy/phy-core.c +6
-0@@ -689,11 +689,17 @@ EXPORT_SYMBOL_GPL(phy_modify_mmd);static int __phy_read_page (struct phy_device *phydev) { +if (WARN_ONCE (!phydev->drv->read_page,"read_page callback not available, PHY driver not loaded?\n" )) +return -EOPNOTSUPP ; +return phydev->drv->read_page (phydev); }static int __phy_write_page (struct phy_device *phydev,int page) { +if (WARN_ONCE (!phydev->drv->write_page,"write_page callback not available, PHY driver not loaded?\n" )) +return -EOPNOTSUPP ; +return phydev->drv->write_page (phydev, page); }
David Miller
maintainerFrom: Heiner Kallweit <[email protected]>
Date: Wed, 16 Oct 2019 21:53:31 +0200
Quoted 7 lines
Currently there's a bug in the module subsystem [0] preventing load of the PHY driver module on certain systems (as one symptom). This results in a NPE on such systems for the following reason:
Instead of the correct PHY driver the genphy driver is loaded that doesn't implement the read_page/write_page callbacks. Every call tophy_read_paged()et al will result in a NPE therefore.
Quoted 8 lines
In parallel to fixing the root cause we should make sure that this one and maybe similar issues in other subsystems don't result in a NPE in phylib. So let's check for the callbacks before using them and warn once if they are not available.
[0] https://marc.info/?t=157072642100001&r=1&w=2
Signed-off-by Heiner Kallweit <[email protected]>
I read the discussion over a few times and this looks good to me for now, so applied.
Thanks.