net: dsa: mv88e6xxx: Add lockdep classes to fix false positive splat
authorAndrew Lunn <andrew@lunn.ch>
Sat, 23 Feb 2019 16:43:56 +0000 (17:43 +0100)
committerDavid S. Miller <davem@davemloft.net>
Mon, 25 Feb 2019 06:21:23 +0000 (22:21 -0800)
The following false positive lockdep splat has been observed.

======================================================
WARNING: possible circular locking dependency detected
4.20.0+ #302 Not tainted
------------------------------------------------------
systemd-udevd/160 is trying to acquire lock:
edea6080 (&chip->reg_lock){+.+.}, at: __setup_irq+0x640/0x704

but task is already holding lock:
edff0340 (&desc->request_mutex){+.+.}, at: __setup_irq+0xa0/0x704

which lock already depends on the new lock.

the existing dependency chain (in reverse order) is:

-> #1 (&desc->request_mutex){+.+.}:
       mutex_lock_nested+0x1c/0x24
       __setup_irq+0xa0/0x704
       request_threaded_irq+0xd0/0x150
       mv88e6xxx_probe+0x41c/0x694 [mv88e6xxx]
       mdio_probe+0x2c/0x54
       really_probe+0x200/0x2c4
       driver_probe_device+0x5c/0x174
       __driver_attach+0xd8/0xdc
       bus_for_each_dev+0x58/0x7c
       bus_add_driver+0xe4/0x1f0
       driver_register+0x7c/0x110
       mdio_driver_register+0x24/0x58
       do_one_initcall+0x74/0x2e8
       do_init_module+0x60/0x1d0
       load_module+0x1968/0x1ff4
       sys_finit_module+0x8c/0x98
       ret_fast_syscall+0x0/0x28
       0xbedf2ae8

-> #0 (&chip->reg_lock){+.+.}:
       __mutex_lock+0x50/0x8b8
       mutex_lock_nested+0x1c/0x24
       __setup_irq+0x640/0x704
       request_threaded_irq+0xd0/0x150
       mv88e6xxx_g2_irq_setup+0xcc/0x1b4 [mv88e6xxx]
       mv88e6xxx_probe+0x44c/0x694 [mv88e6xxx]
       mdio_probe+0x2c/0x54
       really_probe+0x200/0x2c4
       driver_probe_device+0x5c/0x174
       __driver_attach+0xd8/0xdc
       bus_for_each_dev+0x58/0x7c
       bus_add_driver+0xe4/0x1f0
       driver_register+0x7c/0x110
       mdio_driver_register+0x24/0x58
       do_one_initcall+0x74/0x2e8
       do_init_module+0x60/0x1d0
       load_module+0x1968/0x1ff4
       sys_finit_module+0x8c/0x98
       ret_fast_syscall+0x0/0x28
       0xbedf2ae8

other info that might help us debug this:

 Possible unsafe locking scenario:

       CPU0                    CPU1
       ----                    ----
  lock(&desc->request_mutex);
                               lock(&chip->reg_lock);
                               lock(&desc->request_mutex);
  lock(&chip->reg_lock);

&desc->request_mutex refer to two different mutex. #1 is the GPIO for
the chip interrupt. #2 is the chained interrupt between global 1 and
global 2.

Add lockdep classes to the GPIO interrupt to avoid this.

Reported-by: Russell King <rmk+kernel@armlinux.org.uk>
Signed-off-by: Andrew Lunn <andrew@lunn.ch>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/dsa/mv88e6xxx/chip.c

index 3f9a3fc52a0e7179510d9bd7c07a82b2e55f592f..3546c5e2b924241e1f1a21f7d51d0f1174e197bc 100644 (file)
@@ -442,12 +442,20 @@ out_mapping:
 
 static int mv88e6xxx_g1_irq_setup(struct mv88e6xxx_chip *chip)
 {
+       static struct lock_class_key lock_key;
+       static struct lock_class_key request_key;
        int err;
 
        err = mv88e6xxx_g1_irq_setup_common(chip);
        if (err)
                return err;
 
+       /* These lock classes tells lockdep that global 1 irqs are in
+        * a different category than their parent GPIO, so it won't
+        * report false recursion.
+        */
+       irq_set_lockdep_class(chip->irq, &lock_key, &request_key);
+
        err = request_threaded_irq(chip->irq, NULL,
                                   mv88e6xxx_g1_irq_thread_fn,
                                   IRQF_ONESHOT | IRQF_SHARED,