On 9/14/19 9:08 PM, Corey Minyard wrote:
>>
>>>
>>> {disable,enable}_si_irq() themselves are racy:
>>>
>>> static inline bool disable_si_irq(struct smi_info *smi_info)
>>> {
>>> if ((smi_info->io.irq) && (!smi_info->interrupt_disabled)) {
>>> smi_info->interrupt_disabled = true;
>>>
>>> Basically interrupt_disabled need to be atomic here to have any value,
>>> unless you ensure to have a spin lock around every access to it.
>>
>> It needs to be atomic, yes, but I think just adding the spinlock like
>> I suggested will work. You are right, the check for timer_running is
>> not necessary here, and I'm fine with removing it, but there are other
>> issues with interrupt_disabled (as you said) and with memory ordering
>> in the timer case. So even if you remove the timer running check, the
>> lock is still required here.
>
> It turns out you were right, all that really needs to be done is the
> del_timer_sync(). I've added your patch to my queue.
>
> Sorry for the trouble.
Awesome!
Sorry I was going to get back and look at it again, but Linux Plumbers
and playing sardine i a tin can got in the way.
Cheers,
Jes
