Hi Russell,

On Sat, Sep 27, 2014 at 10:25:36AM +0100, Russell King - ARM Linux wrote:

On Sat, Sep 27, 2014 at 10:54:42AM +0200, Maxime Ripard wrote:

quoted

dma_slave_caps is very important to the generic layers that might interact with
dmaengine, such as ASoC. Unfortunately, it has been added as yet another
dma_device callback, and most of the existing drivers haven't implemented it,
reducing its reliability.

Many haven't implemented it probably because either (a) they don't get used
with ASoC, or (b) they aren't aware of the new interface, or (c) can't be
bothered with the churn.

For a), I really see this as a chicken-egg issue. ASoC is the only
user of it because it's the only framework that has a generic layer on
top, and it's the only framework that has a generic layer because most
drivers don't implement it.

Now, there seems to be a trend to actually use a generic DMA layer in
other frameworks. SPI gained one recently, I think I saw something
about some discussions for IIO and I2C too. And in order for this to
work, we have to make it reliable, and as such, implemented on most
drivers.

However, trying to return something introduces a bug:

quoted

 static inline int dma_get_slave_caps(struct dma_chan *chan, struct dma_slave_caps *caps)
 {
+	struct dma_device *device;
+
 	if (!chan || !caps)
 		return -EINVAL;
 
+	device = chan->device;
+
 	/* check if the channel supports slave transactions */
-	if (!test_bit(DMA_SLAVE, chan->device->cap_mask.bits))
+	if (!test_bit(DMA_SLAVE, device->cap_mask.bits))
 		return -ENXIO;
 
-	if (chan->device->device_slave_caps)
-		return chan->device->device_slave_caps(chan, caps);
+	caps->cmd_pause = !!device->device_pause;
+	caps->cmd_terminate = !!device->device_terminate_all;
+
+	if (device->device_slave_caps)
+		return device->device_slave_caps(chan, caps);
 
-	return -ENXIO;
+	return 0;

So this now returns success if the driver doesn't implement device_slave_caps(),
but with most of the structure zero.

Now, consider what effect this has with:


        ret = dma_get_slave_caps(chan, &dma_caps);
        if (ret == 0) {
                if (dma_caps.cmd_pause)
                        hw.info |= SNDRV_PCM_INFO_PAUSE | SNDRV_PCM_INFO_RESUME;
                if (dma_caps.residue_granularity <= DMA_RESIDUE_GRANULARITY_SEGMENT)
                        hw.info |= SNDRV_PCM_INFO_BATCH;

                if (substream->stream == SNDRV_PCM_STREAM_PLAYBACK)
                        addr_widths = dma_caps.dstn_addr_widths;
                else
                        addr_widths = dma_caps.src_addr_widths;
        }

addr_widths becomes zero, and we also get SNDRV_PCM_INFO_BATCH turned
on for _all_ DMA engine drivers.  The first renders ASoC useless with
DMA engine.

It may be a good way to get people to implement it, but this will cause
regressions.

Hmmm, nasty indeed. Maybe we could add a test to see if any of the
field we're going to use are filled, and if not, return an error?

Thanks!
Maxime

-- 
Maxime Ripard, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: Digital signature
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20141001/262b71c3/attachment.sig>