Deadlock in grgpio_irq_unmap()

Deadlock in grgpio_irq_unmap()

[Alexey Khoroshilov]

Dear colleagus,

grgpio_irq_unmap() code looks quite suspicious regarding usage of
priv->bgc.lock spinlock.

It locks the spinlock in line 310:
        spin_lock_irqsave(&priv->bgc.lock, flags);
and then it can call grgpio_set_imask() in line 317:
        grgpio_set_imask(priv, i, 0);

But grgpio_set_imask() unconditionally locks the spinlock by itself.

static void grgpio_set_imask(struct grgpio_priv *priv, unsigned int offset,
			     int val)
{
	struct bgpio_chip *bgc = &priv->bgc;
	unsigned long mask = bgc->pin2mask(bgc, offset);
	unsigned long flags;

	spin_lock_irqsave(&bgc->lock, flags);

	if (val)
		priv->imask |= mask;
	else
		priv->imask &= ~mask;
	bgc->write_reg(priv->regs + GRGPIO_IMASK, priv->imask);

	spin_unlock_irqrestore(&bgc->lock, flags);
}

So, it looks like unavoidable deadlock here.

Found by Linux Driver Verification project (linuxtesting.org).

--
Alexey Khoroshilov
Linux Verification Center, ISPRAS
web: http://linuxtesting.org

Re: Deadlock in grgpio_irq_unmap()

[Alexandre Courbot]

On Mon, Aug 3, 2015 at 1:15 AM, Alexey Khoroshilov
<khoroshilov@ispras.ru> wrote:
> Dear colleagus,
>
> grgpio_irq_unmap() code looks quite suspicious regarding usage of
> priv->bgc.lock spinlock.
>
> It locks the spinlock in line 310:
>         spin_lock_irqsave(&priv->bgc.lock, flags);
> and then it can call grgpio_set_imask() in line 317:
>         grgpio_set_imask(priv, i, 0);
>
> But grgpio_set_imask() unconditionally locks the spinlock by itself.

Indeed, this looks plain wrong. I have sent a fix that moves the
spinlock acquisition outside of grgpio_set_imask().