[RFC,3/3] stk1160: Use non-coherent buffers for USB transfers

[RFC,3/3] stk1160: Use non-coherent buffers for USB transfers

[Ezequiel Garcia]

Platforms without hardware coherency can benefit a lot
from using non-coherent buffers. Moreover, platforms
with hardware coherency aren't impacted by this change.

For instance, on AM335x, while it's still not possible
to capture full resolution frames, this patch enables
half-resolution frame streams to work.

Signed-off-by: Ezequiel Garcia <ezequiel@collabora.com>
---
 drivers/media/usb/stk1160/stk1160-video.c | 22 ++++++----------------
 1 file changed, 6 insertions(+), 16 deletions(-)

diff --git a/drivers/media/usb/stk1160/stk1160-video.c b/drivers/media/usb/stk1160/stk1160-video.c
index 2811f612820f..aeb4264d1998 100644
--- a/drivers/media/usb/stk1160/stk1160-video.c
+++ b/drivers/media/usb/stk1160/stk1160-video.c
@@ -240,6 +240,9 @@ static void stk1160_process_isoc(struct stk1160 *dev, struct urb *urb)
 		return;
 	}
 
+	dma_sync_single_for_cpu(&urb->dev->dev, urb->transfer_dma,
+		urb->transfer_buffer_length, DMA_FROM_DEVICE);
+
 	for (i = 0; i < urb->number_of_packets; i++) {
 		status = urb->iso_frame_desc[i].status;
 		if (status < 0) {
@@ -379,16 +382,11 @@ void stk1160_free_isoc(struct stk1160 *dev)
 		urb = dev->isoc_ctl.urb[i];
 		if (urb) {
 
-			if (dev->isoc_ctl.transfer_buffer[i]) {
-#ifndef CONFIG_DMA_NONCOHERENT
-				usb_free_coherent(dev->udev,
+			if (dev->isoc_ctl.transfer_buffer[i])
+				usb_free_noncoherent(dev->udev,
 					urb->transfer_buffer_length,
 					dev->isoc_ctl.transfer_buffer[i],
 					urb->transfer_dma);
-#else
-				kfree(dev->isoc_ctl.transfer_buffer[i]);
-#endif
-			}
 			usb_free_urb(urb);
 			dev->isoc_ctl.urb[i] = NULL;
 		}
@@ -461,12 +459,8 @@ int stk1160_alloc_isoc(struct stk1160 *dev)
 			goto free_i_bufs;
 		dev->isoc_ctl.urb[i] = urb;
 
-#ifndef CONFIG_DMA_NONCOHERENT
-		dev->isoc_ctl.transfer_buffer[i] = usb_alloc_coherent(dev->udev,
+		dev->isoc_ctl.transfer_buffer[i] = usb_alloc_noncoherent(dev->udev,
 			sb_size, GFP_KERNEL, &urb->transfer_dma);
-#else
-		dev->isoc_ctl.transfer_buffer[i] = kmalloc(sb_size, GFP_KERNEL);
-#endif
 		if (!dev->isoc_ctl.transfer_buffer[i]) {
 			stk1160_err("cannot alloc %d bytes for tx[%d] buffer\n",
 				sb_size, i);
@@ -490,11 +484,7 @@ int stk1160_alloc_isoc(struct stk1160 *dev)
 		urb->interval = 1;
 		urb->start_frame = 0;
 		urb->number_of_packets = max_packets;
-#ifndef CONFIG_DMA_NONCOHERENT
 		urb->transfer_flags = URB_ISO_ASAP | URB_NO_TRANSFER_DMA_MAP;
-#else
-		urb->transfer_flags = URB_ISO_ASAP;
-#endif
 
 		k = 0;
 		for (j = 0; j < max_packets; j++) {

[RFC,3/3] stk1160: Use non-coherent buffers for USB transfers

[Christoph Hellwig]

> +	dma_sync_single_for_cpu(&urb->dev->dev, urb->transfer_dma,
> +		urb->transfer_buffer_length, DMA_FROM_DEVICE);

You can't ue dma_sync_single_for_cpu on non-coherent dma buffers,
which is one of the major issues with them.

[RFC,3/3] stk1160: Use non-coherent buffers for USB transfers

[Tomasz Figa]

On Fri, Aug 31, 2018 at 2:59 AM Christoph Hellwig <hch@infradead.org> wrote:
>
> > +     dma_sync_single_for_cpu(&urb->dev->dev, urb->transfer_dma,
> > +             urb->transfer_buffer_length, DMA_FROM_DEVICE);
>
> You can't ue dma_sync_single_for_cpu on non-coherent dma buffers,
> which is one of the major issues with them.

It's not an issue of DMA API, but just an API mismatch. By design,
memory allocated for device (e.g. by DMA API) doesn't have to be
physically contiguous, while dma_*_single() API expects a _single_,
physically contiguous region of memory.

We need a way to allocate non-coherent memory using DMA API to handle
(on USB example, but applies to virtually any class of devices doing
DMA):
 - DMA address range limitations (e.g. dma_mask) - while a USB HCD
driver is normally aware of those, USB device driver should have no
idea,
 - memory mapping capability === whether contiguous memory or a set of
random pages can be allocated - this is a platform integration detail,
which even a USB HCD driver may not be aware of, if a SoC IOMMU is
just stuffed between the bus and HCD,
 - platform coherency specifics - there are practical scenarios when
on a coherent-by-default system it's more efficient to allocate
non-coherent memory and manage caches explicitly to avoid the costs of
cache snooping.

If DMA_ATTR_NON_CONSISTENT is not the right way to do it, there should
be definitely a new API introduced, coupled closely to DMA API
implementation on given platform, since it's the only place which can
solve all the constraints above.

Best regards,
Tomasz