Re: [PATCH 01/15] iio: buffer-dma: Get rid of incoming/outgoing queues

[Paul Cercueil <paul@crapouillou.net>]

Le lun., nov. 22 2021 at 16:17:59 +0100, Lars-Peter Clausen 
<lars@metafoo.de> a écrit :
> On 11/22/21 4:16 PM, Paul Cercueil wrote:
>> Hi Lars,
>> 
>> Le lun., nov. 22 2021 at 16:08:51 +0100, Lars-Peter Clausen 
>> \x7f<lars@metafoo.de> a écrit :
>>> On 11/21/21 9:08 PM, Paul Cercueil wrote:
>>>> 
>>>> 
>>>> Le dim., nov. 21 2021 at 19:49:03 +0100, Lars-Peter Clausen 
>>>> \x7f\x7f\x7f\x7f<lars@metafoo.de> a écrit :
>>>>> On 11/21/21 6:52 PM, Paul Cercueil wrote:
>>>>>> Hi Lars,
>>>>>> 
>>>>>> Le dim., nov. 21 2021 at 17:23:35 +0100, Lars-Peter Clausen 
>>>>>> \x7f\x7f\x7f\x7f\x7f\x7f\x7f\x7f\x7f<lars@metafoo.de> a écrit :
>>>>>>> On 11/15/21 3:19 PM, Paul Cercueil wrote:
>>>>>>>> The buffer-dma code was using two queues, incoming and 
>>>>>>>> outgoing, to
>>>>>>>> manage the state of the blocks in use.
>>>>>>>> 
>>>>>>>> While this totally works, it adds some complexity to the code,
>>>>>>>> especially since the code only manages 2 blocks. It is much 
>>>>>>>> \x7f\x7f\x7f\x7f\x7f\x7f\x7feasier to
>>>>>>>> just check each block's state manually, and keep a counter for 
>>>>>>>> \x7f\x7f\x7f\x7f\x7f\x7f\x7fthe \x7f\x7f\x7f\x7f\x7fnext
>>>>>>>> block to dequeue.
>>>>>>>> 
>>>>>>>> Since the new DMABUF based API wouldn't use these incoming and 
>>>>>>>> \x7f\x7f\x7f\x7f\x7f\x7f\x7f\x7f\x7f\x7f\x7f\x7foutgoing
>>>>>>>> queues anyway, getting rid of them now makes the upcoming 
>>>>>>>> changes
>>>>>>>> simpler.
>>>>>>>> 
>>>>>>>> Signed-off-by: Paul Cercueil <paul@crapouillou.net>
>>>>>>> The outgoing queue is going to be replaced by fences, but I 
>>>>>>> think \x7f\x7f\x7f\x7f\x7f\x7f\x7f\x7f\x7f\x7fwe \x7f\x7fneed to keep the incoming queue.
>>>>>> 
>>>>>> Blocks are always accessed in sequential order, so we now have a 
>>>>>> \x7f\x7f\x7f\x7f\x7f\x7f\x7f\x7f\x7f"queue->next_dequeue" that cycles between the buffers 
>>>>>> \x7f\x7f\x7f\x7f\x7fallocated for \x7f\x7f\x7f\x7ffileio.
>>>>>> 
>>>>>>>> [...]
>>>>>>>> @@ -442,28 +435,33 @@ 
>>>>>>>> EXPORT_SYMBOL_GPL(iio_dma_buffer_disable);
>>>>>>>>   static void iio_dma_buffer_enqueue(struct 
>>>>>>>> iio_dma_buffer_queue \x7f\x7f\x7f\x7f\x7f\x7f\x7f\x7f\x7f\x7f\x7f\x7f\x7f\x7f\x7f*queue,
>>>>>>>>       struct iio_dma_buffer_block *block)
>>>>>>>>   {
>>>>>>>> -    if (block->state == IIO_BLOCK_STATE_DEAD) {
>>>>>>>> +    if (block->state == IIO_BLOCK_STATE_DEAD)
>>>>>>>>           iio_buffer_block_put(block);
>>>>>>>> -    } else if (queue->active) {
>>>>>>>> +    else if (queue->active)
>>>>>>>>           iio_dma_buffer_submit_block(queue, block);
>>>>>>>> -    } else {
>>>>>>>> +    else
>>>>>>>>           block->state = IIO_BLOCK_STATE_QUEUED;
>>>>>>>> -        list_add_tail(&block->head, &queue->incoming);
>>>>>>> If iio_dma_buffer_enqueue() is called with a dmabuf and the 
>>>>>>> \x7f\x7f\x7f\x7f\x7f\x7fbuffer \x7f\x7f\x7f\x7fis \x7f\x7fnot active, it will be marked as queued, 
>>>>>>> but we \x7f\x7f\x7f\x7f\x7f\x7fdon't actually \x7f\x7f\x7f\x7fkeep a \x7f\x7freference to it 
>>>>>>> anywhere. It will \x7f\x7f\x7f\x7f\x7f\x7fnever be submitted to \x7f\x7f\x7f\x7fthe DMA, and 
>>>>>>> \x7f\x7fit will never be \x7f\x7f\x7f\x7f\x7f\x7fsignaled as completed.
>>>>>> 
>>>>>> We do keep a reference to the buffers, in the 
>>>>>> queue->fileio.blocks \x7f\x7f\x7f\x7f\x7f\x7f\x7f\x7f\x7farray. When the buffer is enabled, 
>>>>>> all the blocks in that \x7f\x7f\x7f\x7f\x7farray \x7f\x7f\x7fthat \x7fare in the "queued" 
>>>>>> state will be submitted to the \x7f\x7f\x7f\x7f\x7fDMA.
>>>>>> 
>>>>> But not when used in combination with the DMA buf changes later 
>>>>> in \x7f\x7f\x7f\x7f\x7f\x7fthis series.
>>>>> 
>>>> 
>>>> That's still the case after the DMABUF changes of the series. Or 
>>>> can \x7f\x7f\x7f\x7fyou point me exactly what you think is broken?
>>>> 
>>> When you allocate a DMABUF with the allocate IOCTL and then submit 
>>> it \x7f\x7fwith the enqueue IOCTL before the buffer is enabled it will 
>>> end up \x7f\x7fmarked as queued, but not actually be queued anywhere.
>>> 
>> 
>> Ok, it works for me because I never enqueue blocks before enabling 
>> the \x7fbuffer. I can add a requirement that blocks must be enqueued 
>> only \x7fafter the buffer is enabled.
> 
> I don't think that is a good idea. This way you are going to 
> potentially drop data at the begining of your stream when the DMA 
> isn't ready yet.
> 

You wouldn't drop data, but it could cause an underrun, yes. Is it such 
a big deal, knowing that the buffer was just enabled? I don't think you 
can disable then enable the buffer without causing a discontinuity in 
the stream.

-Paul