On Sun, Feb 17, 2019 at 04:27:32PM +0000, Russell King wrote:
> ======================================================
> 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]
> -> #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
>
> 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);
Hi Russell
I failed to reproduce it on a Freescale system. Which made me take a
closer look at the above. This is a false positive.
In #1 we are requesting the GPIO interrupt. In #2 we are requesting
the chained interrupt from the mv88e6xxx global 1 interrupt handler.
So these are different desc->request_mutex. The Freescale VF610 GPIO
driver uses gpiochip_irqchip_add(), which creates a lock class for the
GPIO. The marvell gpio-mvebu driver does not create a lock class. So
when i test on Freescale, lockdep can tell they are different mutex,
but on clearfog it cannot.
So i think the real fix is probably two fold, although just doing one
is sufficient:
1) Add lock classes to gpio-mvebu, by call irq_set_lockdep_class()
2) Add lock classes to chip.c global 1, by calling irq_set_lockdep_class()
There is probably more value in 1) since the mvebu gpio driver is much
more widely used than the mv88e6xxx driver.
Andrew
