HID based IPTS and ITHC support, 6.1 edition

[StollD]

iptsd feels pretty much ready and will work better than what is currently in the repo, so lets start doing this ...

Things that need to be done once this is merged:

• Update CONFIG_MISC_IPTS=m to CONFIG_HID_IPTS=m in our kernel configs
• Add CONFIG_HID_ITHC=m to the kernel configs
• Bump iptsd version to 0.6 and tag a release

Changes since #132:

• ITHC: quo/ithc-linux@67ebf09...e1c3978
• IPTS: linux-surface/intel-precise-touch@f4c781a...19e8512

[qzed]

You could replace those with set_bit() and test_bit() (https://www.kernel.org/doc/html/latest/core-api/kernel-api.html?highlight=set_bit#bit-operations) as full mutexes are a bit overkill for that task. Alternatives are atomics (which is probably better for explicit acquire/release semantics without needing rmb/wmb) or I guess you could even use READ_ONCE() and WRITE_ONCE() with memory barriers if you want to keep the bool.

So with bitops you'd do something like:

int ipts_thread_stop(struct ipts_thread *thread)
{
	int ret = 0;

	if (!thread)
		return -EFAULT;

	if (!thread->thread)
		return 0;

	set_bit(IPTS_THREAD_STOP_BIT, &thread->flags);
	wmb();  // ensure bit is set before we start waiting

	wait_for_completion(&thread->done);
	ret = kthread_stop(thread->thread);

	thread->thread = NULL;
	thread->data = NULL;
	thread->threadfn = NULL;

	return ret;
}

and

bool ipts_thread_should_stop(struct ipts_thread *thread)
{
	bool ret = false;

	if (!thread)
		return false;

	// optional: add rmb(); here to have "acquire" semantics, but that's not really necessary
	return test_bit(IPTS_THREAD_STOP_BIT, thread->flags);
}

[StollD]

Would smp_load_acquire / smp_store_release work too? That could be the easiest solution, because it already seems to combine READ_ONCE and WRITE_ONCE with a barrier.

[qzed]

Oh, I forgot about those. Yeah I think those should work too.

[qzed]

Although... After checking, those have the barriers exactly the opposite way (which is what you more often want), but I think we only need the wmb() after setting the flag. If at all. I'm not entirely sure there how the semantics are for the wait_for_completion() call, but the point is that we want to make sure that the bit is set before waiting.

[StollD]

Done

[qzed]

Looks good to me (I haven't extensively checked it though). I've added one comment for a potential optimization, but that's something that we can work on later.