Re: [PATCH] [RFC] dmaengine: add fifo_size member

From: Sameer Pujar
Date: Wed Jun 05 2019 - 23:53:37 EST

Sorry for late reply.
[Resending the reply since delivery failed for few recipients]

On 5/6/2019 9:20 PM, Vinod Koul wrote:
On 06-05-19, 18:34, Sameer Pujar wrote:
On 5/4/2019 3:53 PM, Vinod Koul wrote:
On 02-05-19, 18:59, Sameer Pujar wrote:
On 5/2/2019 5:55 PM, Vinod Koul wrote:
On 02-05-19, 16:23, Sameer Pujar wrote:
On 5/2/2019 11:34 AM, Vinod Koul wrote:
On 30-04-19, 17:00, Sameer Pujar wrote:
During the DMA transfers from memory to I/O, it was observed that transfers
were inconsistent and resulted in glitches for audio playback. It happened
because fifo size on DMA did not match with slave channel configuration.

currently 'dma_slave_config' structure does not have a field for fifo size.
Hence the platform pcm driver cannot pass the fifo size as a slave_config.
Note that 'snd_dmaengine_dai_dma_data' structure has fifo_size field which
cannot be used to pass the size info. This patch introduces fifo_size field
and the same can be populated on slave side. Users can set required size
for slave peripheral (multiple channels can be independently running with
different fifo sizes) and the corresponding sizes are programmed through
dma_slave_config on DMA side.
FIFO size is a hardware property not sure why you would want an
interface to program that?

On mismatch, I guess you need to take care of src/dst_maxburst..
Yes, FIFO size is a HW property. But it is SW configurable(atleast in my
case) on
slave side and can be set to different sizes. The src/dst_maxburst is
Are you sure, have you talked to HW folks on that? IIUC you are
programming the data to be used in FIFO not the FIFO length!
Yes, I mentioned about FIFO length.

1. MAX FIFO size is fixed in HW. But there is a way to limit the usage per
ÂÂ in multiples of 64 bytes.
2. Having a separate member would give independent control over MAX BURST
SIZE and
for specific values, I think this depends on few factors related to
needs of client, DMA needs of the system etc.,

In such cases how does DMA know the actual FIFO depth of slave peripheral?
Why should DMA know? Its job is to push/pull data as configured by
peripheral driver. The peripheral driver knows and configures DMA
I am not sure if there is any HW logic that mandates DMA to know the size
of configured FIFO depth on slave side. I will speak to HW folks and
would update here.
I still do not comprehend why dma would care about slave side
configuration. In the absence of patch which uses this I am not sure
what you are trying to do...
I am using DMA HW in cyclic mode for data transfers to Audio sub-system.
In such cases flow control on DMA transfers is essential, since I/O is
right and people use burst size for precisely that!

consuming/producing the data at slower rate. The DMA tranfer is enabled/
disabled during start/stop of audio playback/capture sessions through ALSA
callbacks and DMA runs in cyclic mode. Hence DMA is the one which is doing
flow control and it is necessary for it to know the peripheral FIFO depth
to avoid overruns/underruns.
not really, knowing that doesnt help anyway you have described! DMA
pushes/pulls data and that is controlled by burst configured by slave
(so it know what to expect and porgrams things accordingly)

you are really going other way around about the whole picture. FWIW that
is how *other* folks do audio with dmaengine!
I discussed this internally with HW folks and below is the reason why DMA needs
to know FIFO size.

- FIFOs reside in peripheral device(ADMAIF), which is the ADMA interface to the audio sub-system.
- ADMAIF has multiple channels and share FIFO buffer for individual operations. There is a provision
 to allocate specific fifo size for each individual ADMAIF channel from the shared buffer.
- Tegra Audio DMA(ADMA) architecture is different from the usual DMA engines, which you described earlier.
- The flow control logic is placed inside ADMA. Slave peripheral device(ADMAIF) signals ADMA whenever a
 read or write happens on the FIFO(per WORD basis). Please note that the signaling is per channel. There is
 no other signaling present from ADMAIF to ADMA.
- ADMA keeps a counter related to above signaling. Whenever a sufficient space is available, it initiates a transfer.
 But the question is, how does it know when to transfer. This is the reason, why ADMA has to be aware of FIFO
 depth of ADMAIF channel. Depending on the counters and FIFO depth, it knows exactly when a free space is available
 in the context of a specific channel. On ADMA, FIFO_SIZE is just a value which should match to actual FIFO_DEPTH/SIZE
 of ADMAIF channel.
- Now consider two cases based on above logic,
ÂÂÂ In this case, ADMA thinks that there is enough space available for transfer, when actually the FIFO data
ÂÂÂ on slave is not consumed yet. It would result in OVERRUN.
ÂÂÂ This is case where ADMA wonât transfer, even though sufficient space is available, resulting in UNDERRUN.
- The guideline is to program, DMA_FIFO_SIZE(on ADMA side) = SLAVE_FIFO_SIZE(on ADMAIF side) and hence we need a
 way to communicate fifo size info to ADMA.

Also please note that, peripheral device has multiple channels and share
a fixed MAX FIFO buffer. But SW can program different FIFO sizes for
individual channels.
yeah peripheral driver, yes. DMA driver nope!