net: mdio: Add dedicated C45 API to MDIO bus drivers
Hello Andrew Lunn,
The patch 4e4aafcddbbf: "net: mdio: Add dedicated C45 API to MDIO bus drivers" from Jan 9, 2023, leads to the following Smatch static checker warning:
drivers/net/phy/mdio_bus.c:1003 mdiobus_c45_read() warn: sleeping in atomic context drivers/net/phy/mdio_bus.c:1104 mdiobus_c45_write() warn: sleeping in atomic context
drivers/net/phy/mdio_bus.c
993 * @regnum: register number to read
994 *
995 * NOTE: MUST NOT be called from interrupt context,
996 * because the bus read/write functions may wait for an interrupt
997 * to conclude the operation.
998 */
999 int mdiobus_c45_read(struct mii_bus *bus, int addr, int devad, u32 regnum)
1000 {
1001 int retval;
1002
--> 1003 mutex_lock(&bus->mdio_lock);
^^^^^^^^^^^^^^^^^^^^^^^^^^^This takes a mutex.
1004 retval = __mdiobus_c45_read(bus, addr, devad, regnum);
1005 mutex_unlock(&bus->mdio_lock);
1006
1007 return retval;
1008 }The problematic call tree is:
b44_ioctl() <- disables preempt
-> phy_mii_ioctl()
-> mdiobus_c45_read()
regards,
dan carpenter
Quoted 1 line · Dan Carpenter
Hello Andrew Lunn,
Adding Florian
Quoted 28 lines
The patch
4e4aafcddbbf: "net: mdio: Add dedicated C45 API to MDIO bus drivers" from Jan 9, 2023, leads to the following Smatch static checker warning:
drivers/net/phy/mdio_bus.c:1003mdiobus_c45_read()warn: sleeping in atomic contextdrivers/net/phy/mdio_bus.c:1104mdiobus_c45_write()warn: sleeping in atomic context
drivers/net/phy/mdio_bus.c993 * @regnum: register number to read 994 * 995 * NOTE: MUST NOT be called from interrupt context, 996 * because the bus read/write functions may wait for an interrupt 997 * to conclude the operation. 998 */ 999 int mdiobus_c45_read(struct mii_bus *bus, int addr, int devad, u32 regnum) 1000 { 1001 int retval; 1002 --> 1003 mutex_lock(&bus->mdio_lock); ^^^^^^^^^^^^^^^^^^^^^^^^^^^This takes a mutex.
1004 retval = __mdiobus_c45_read(bus, addr, devad, regnum); 1005 mutex_unlock(&bus->mdio_lock); 1006 1007 return retval; 1008 }
Quoted 5 lines
The problematic call tree is:
b44_ioctl()<- disables preempt
->phy_mii_ioctl()
->mdiobus_c45_read()
This seems like an old bug which my change has brought to light. I believe MDIO operations have always been blocking. So i would say b44_ioctl should not be taking a spinlock here.
Andrew