usb: musb: cppi41: restart hrtimer only if not yet done
authorThomas Gleixner <tglx@linutronix.de>
Thu, 2 Oct 2014 15:32:16 +0000 (17:32 +0200)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Fri, 14 Nov 2014 17:00:05 +0000 (09:00 -0800)
commitbe7cabe556695e136f983623e22c3c52de25dc1b
tree2601e8664b34df6a29c214332f7dc02dfa751c98
parentb8edc5dc0ae741cba4b094b220cf8721db5192ec
usb: musb: cppi41: restart hrtimer only if not yet done

commit d2e6d62c9cbbc2da4211f672dbeea08960e29a80 upstream.

commit c58d80f52 ("usb: musb: Ensure that cppi41 timer gets armed on
premature DMA TX irq") fixed hrtimer scheduling bug. There is one left
which does not trigger that often.
The following scenario is still possible:

    lock(&x->lock);
    hrtimer_start(&x->t);
    unlock(&x->lock);

expires:
    t->function();
                                lock(&x->lock);
    lock(&x->lock);             if (!hrtimer_queued(&x->t))
                                        hrtimer_start(&x->t);
                                unlock(&x->lock);

    if (!list_empty(x->early_tx_list))
           ret = HRTIMER_RESTART;
->         hrtimer_forward_now(...)
    } else
           ret = HRTIMER_NORESTART;

    unlock(&x->lock);

and the timer callback returns HRTIMER_RESTART for an armed timer. This
is wrong and we run into the BUG_ON() in __run_hrtimer().
This can happens on SMP or PREEMPT-RT.
The patch fixes the problem by only starting the timer if the timer is
not yet queued.

Reported-by: Torben Hohn <torbenh@linutronix.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
[bigeasy: collected information and created a patch + description based
          on it]
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Felipe Balbi <balbi@ti.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/usb/musb/musb_cppi41.c