[PATCH v5 03/21] gpu: host1x: Show number of pending waiters in debugfs

[Mikko Perttunen]

On 4/8/21 7:25 AM, Michał Mirosław wrote:
> On Thu, Apr 08, 2021 at 06:13:44AM +0200, Michał Mirosław wrote:
>> On Fri, Apr 02, 2021 at 07:02:32PM +0300, Dmitry Osipenko wrote:
>>> 02.04.2021 00:19, Michał Mirosław пишет:
>>>> On Fri, Mar 26, 2021 at 04:34:13PM +0200, Mikko Perttunen wrote:
>>>>> On 3/23/21 12:16 PM, Thierry Reding wrote:
>>>>>> On Mon, Jan 11, 2021 at 03:00:01PM +0200, Mikko Perttunen wrote:
>>>>>>> Show the number of pending waiters in the debugfs status file.
>>>>>>> This is useful for testing to verify that waiters do not leak
>>>>>>> or accumulate incorrectly.
>>>>>>>
>>>>>>> Signed-off-by: Mikko Perttunen <mperttunen at nvidia.com>
>>>>>>> ---
>>>>>>>    drivers/gpu/host1x/debug.c | 14 +++++++++++---
>>>>>>>    1 file changed, 11 insertions(+), 3 deletions(-)
>>>>>>>
>>>>>>> diff --git a/drivers/gpu/host1x/debug.c b/drivers/gpu/host1x/debug.c
>>>>>>> index 1b4997bda1c7..8a14880c61bb 100644
>>>>>>> --- a/drivers/gpu/host1x/debug.c
>>>>>>> +++ b/drivers/gpu/host1x/debug.c
>>>>>>> @@ -69,6 +69,7 @@ static int show_channel(struct host1x_channel *ch, void *data, bool show_fifo)
>>>>>>>    static void show_syncpts(struct host1x *m, struct output *o)
>>>>>>>    {
>>>>>>> +	struct list_head *pos;
>>>>>>>    	unsigned int i;
>>>>>>>    	host1x_debug_output(o, "---- syncpts ----\n");
>>>>>>> @@ -76,12 +77,19 @@ static void show_syncpts(struct host1x *m, struct output *o)
>>>>>>>    	for (i = 0; i < host1x_syncpt_nb_pts(m); i++) {
>>>>>>>    		u32 max = host1x_syncpt_read_max(m->syncpt + i);
>>>>>>>    		u32 min = host1x_syncpt_load(m->syncpt + i);
>>>>>>> +		unsigned int waiters = 0;
>>>>>>> -		if (!min && !max)
>>>>>>> +		spin_lock(&m->syncpt[i].intr.lock);
>>>>>>> +		list_for_each(pos, &m->syncpt[i].intr.wait_head)
>>>>>>> +			waiters++;
>>>>>>> +		spin_unlock(&m->syncpt[i].intr.lock);
>>>>>>
>>>>>> Would it make sense to keep a running count so that we don't have to
>>>>>> compute it here?
>>>>>
>>>>> Considering this is just a debug facility, I think I prefer not adding a new
>>>>> field just for it.
>>>>
>>>> This looks like IRQ-disabled region, so unless only root can trigger
>>>> this code, maybe the additional field could save a potential headache?
>>>> How many waiters can there be in the worst case?
>>>
>>> The host1x's IRQ handler runs in a workqueue, so it should be okay.
>>
>> Why, then, this uses a spinlock (and it has 'intr' in its name)?
> 
> The critical sections are already O(n) in number of waiters, so this
> patch doesn't make things worse as I previously thought. The questions
> remain: What is the expected number and upper bound of workers?
> Shouldn't this be a mutex instead?

Everything is primarily for historical reasons. The name 'intr' is 
because this is in the part of the host1x driver that handles syncpoint 
threshold interrupts - just some of it is in interrupt context and some not.

In any case, this code is scheduled for a complete redesign once we get 
the UAPI changes done. I'll take this into account at that point.

Cheers,
Mikko

> 
> Best Regards
> Michał Mirosław
>