[media] tw686x: Add support for DMA scatter-gather mode

Now that the driver has the infrastructure to support more
DMA modes, let's add the DMA scatter-gather mode.

In this mode, the device delivers sequential top-bottom
frames. The scatter-gather logic is based on staging's
tw686x-kh driver (by Krzysztof Ha?asa).
Tested-by: NTim Harvey <tharvey@gateworks.com>
Signed-off-by: NEzequiel Garcia <ezequiel@vanguardiasur.com.ar>
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@s-opensource.com>

drivers/media/pci/tw686x/Kconfig

@@ -4,6 +4,7 @@ config VIDEO_TW686X
 	depends on HAS_DMA
 	select VIDEOBUF2_VMALLOC
 	select VIDEOBUF2_DMA_CONTIG
+	select VIDEOBUF2_DMA_SG
 	select SND_PCM
 	help
 	  Support for Intersil/Techwell TW686x-based frame grabber cards.

drivers/media/pci/tw686x/tw686x-core.c

@@ -65,6 +65,8 @@ static const char *dma_mode_name(unsigned int mode)
 		return "memcpy";
 	case TW686X_DMA_MODE_CONTIG:
 		return "contig";
+	case TW686X_DMA_MODE_SG:
+		return "sg";
 	default:
 		return "unknown";
 	}
@@ -81,6 +83,8 @@ static int tw686x_dma_mode_set(const char *val, struct kernel_param *kp)
 		dma_mode = TW686X_DMA_MODE_MEMCPY;
 	else if (!strcasecmp(val, dma_mode_name(TW686X_DMA_MODE_CONTIG)))
 		dma_mode = TW686X_DMA_MODE_CONTIG;
+	else if (!strcasecmp(val, dma_mode_name(TW686X_DMA_MODE_SG)))
+		dma_mode = TW686X_DMA_MODE_SG;
 	else
 		return -EINVAL;
 	return 0;

drivers/media/pci/tw686x/tw686x-video.c

@@ -20,6 +20,7 @@
 #include <media/v4l2-common.h>
 #include <media/v4l2-event.h>
 #include <media/videobuf2-dma-contig.h>
+#include <media/videobuf2-dma-sg.h>
 #include <media/videobuf2-vmalloc.h>
 #include "tw686x.h"
 #include "tw686x-regs.h"
@@ -28,6 +29,10 @@
 #define TW686X_VIDEO_WIDTH		720
 #define TW686X_VIDEO_HEIGHT(id)		((id & V4L2_STD_525_60) ? 480 : 576)
 
+#define TW686X_MAX_SG_ENTRY_SIZE	4096
+#define TW686X_MAX_SG_DESC_COUNT	256 /* PAL 720x576 needs 203 4-KB pages */
+#define TW686X_SG_TABLE_SIZE		(TW686X_MAX_SG_DESC_COUNT * sizeof(struct tw686x_sg_desc))
+
 static const struct tw686x_format formats[] = {
 	{
 		.fourcc = V4L2_PIX_FMT_UYVY,
@@ -196,6 +201,174 @@ const struct tw686x_dma_ops contig_dma_ops = {
 	.field		= V4L2_FIELD_INTERLACED,
 };
 
+static int tw686x_sg_desc_fill(struct tw686x_sg_desc *descs,
+			       struct tw686x_v4l2_buf *buf,
+			       unsigned int buf_len)
+{
+	struct sg_table *vbuf = vb2_dma_sg_plane_desc(&buf->vb.vb2_buf, 0);
+	unsigned int len, entry_len;
+	struct scatterlist *sg;
+	int i, count;
+
+	/* Clear the scatter-gather table */
+	memset(descs, 0, TW686X_SG_TABLE_SIZE);
+
+	count = 0;
+	for_each_sg(vbuf->sgl, sg, vbuf->nents, i) {
+		dma_addr_t phys = sg_dma_address(sg);
+		len = sg_dma_len(sg);
+
+		while (len && buf_len) {
+
+			if (count == TW686X_MAX_SG_DESC_COUNT)
+				return -ENOMEM;
+
+			entry_len = min_t(unsigned int, len,
+					  TW686X_MAX_SG_ENTRY_SIZE);
+			entry_len = min_t(unsigned int, entry_len, buf_len);
+			descs[count].phys = cpu_to_le32(phys);
+			descs[count++].flags_length =
+					cpu_to_le32(BIT(30) | entry_len);
+			phys += entry_len;
+			len -= entry_len;
+			buf_len -= entry_len;
+		}
+
+		if (!buf_len)
+			return 0;
+	}
+
+	return -ENOMEM;
+}
+
+static void tw686x_sg_buf_refill(struct tw686x_video_channel *vc,
+				 unsigned int pb)
+{
+	struct tw686x_dev *dev = vc->dev;
+	struct tw686x_v4l2_buf *buf;
+
+	while (!list_empty(&vc->vidq_queued)) {
+		unsigned int buf_len;
+
+		buf = list_first_entry(&vc->vidq_queued,
+			struct tw686x_v4l2_buf, list);
+		list_del(&buf->list);
+
+		buf_len = (vc->width * vc->height * vc->format->depth) >> 3;
+		if (tw686x_sg_desc_fill(vc->sg_descs[pb], buf, buf_len)) {
+			v4l2_err(&dev->v4l2_dev,
+				 "dma%d: unable to fill %s-buffer\n",
+				 vc->ch, pb ? "B" : "P");
+			vb2_buffer_done(&buf->vb.vb2_buf, VB2_BUF_STATE_ERROR);
+			continue;
+		}
+
+		buf->vb.vb2_buf.state = VB2_BUF_STATE_ACTIVE;
+		vc->curr_bufs[pb] = buf;
+		return;
+	}
+
+	vc->curr_bufs[pb] = NULL;
+}
+
+static void tw686x_sg_dma_free(struct tw686x_video_channel *vc,
+			       unsigned int pb)
+{
+	struct tw686x_dma_desc *desc = &vc->dma_descs[pb];
+	struct tw686x_dev *dev = vc->dev;
+
+	if (desc->size) {
+		pci_free_consistent(dev->pci_dev, desc->size,
+				    desc->virt, desc->phys);
+		desc->virt = NULL;
+	}
+
+	vc->sg_descs[pb] = NULL;
+}
+
+static int tw686x_sg_dma_alloc(struct tw686x_video_channel *vc,
+			       unsigned int pb)
+{
+	struct tw686x_dma_desc *desc = &vc->dma_descs[pb];
+	struct tw686x_dev *dev = vc->dev;
+	u32 reg = pb ? DMA_PAGE_TABLE1_ADDR[vc->ch] :
+		       DMA_PAGE_TABLE0_ADDR[vc->ch];
+	void *virt;
+
+	if (desc->size) {
+
+		virt = pci_alloc_consistent(dev->pci_dev, desc->size,
+					    &desc->phys);
+		if (!virt) {
+			v4l2_err(&dev->v4l2_dev,
+				 "dma%d: unable to allocate %s-buffer\n",
+				 vc->ch, pb ? "B" : "P");
+			return -ENOMEM;
+		}
+		desc->virt = virt;
+		reg_write(dev, reg, desc->phys);
+	} else {
+		virt = dev->video_channels[0].dma_descs[pb].virt +
+		       vc->ch * TW686X_SG_TABLE_SIZE;
+	}
+
+	vc->sg_descs[pb] = virt;
+	return 0;
+}
+
+static void tw686x_sg_cleanup(struct tw686x_dev *dev)
+{
+	vb2_dma_sg_cleanup_ctx(dev->alloc_ctx);
+}
+
+static int tw686x_sg_setup(struct tw686x_dev *dev)
+{
+	unsigned int sg_table_size, pb, ch, channels;
+
+	dev->alloc_ctx = vb2_dma_sg_init_ctx(&dev->pci_dev->dev);
+	if (IS_ERR(dev->alloc_ctx)) {
+		dev_err(&dev->pci_dev->dev, "unable to init DMA context\n");
+		return PTR_ERR(dev->alloc_ctx);
+	}
+
+	if (is_second_gen(dev)) {
+		/*
+		 * TW6865/TW6869: each channel needs a pair of
+		 * P-B descriptor tables.
+		 */
+		channels = max_channels(dev);
+		sg_table_size = TW686X_SG_TABLE_SIZE;
+	} else {
+		/*
+		 * TW6864/TW6868: we need to allocate a pair of
+		 * P-B descriptor tables, common for all channels.
+		 * Each table will be bigger than 4 KB.
+		 */
+		channels = 1;
+		sg_table_size = max_channels(dev) * TW686X_SG_TABLE_SIZE;
+	}
+
+	for (ch = 0; ch < channels; ch++) {
+		struct tw686x_video_channel *vc = &dev->video_channels[ch];
+
+		for (pb = 0; pb < 2; pb++)
+			vc->dma_descs[pb].size = sg_table_size;
+	}
+
+	return 0;
+}
+
+const struct tw686x_dma_ops sg_dma_ops = {
+	.setup		= tw686x_sg_setup,
+	.cleanup	= tw686x_sg_cleanup,
+	.alloc		= tw686x_sg_dma_alloc,
+	.free		= tw686x_sg_dma_free,
+	.buf_refill	= tw686x_sg_buf_refill,
+	.mem_ops	= &vb2_dma_sg_memops,
+	.hw_dma_mode	= TW686X_SG_MODE,
+	.field		= V4L2_FIELD_SEQ_TB,
+};
+
 static unsigned int tw686x_fields_map(v4l2_std_id std, unsigned int fps)
 {
 	static const unsigned int map[15] = {
@@ -554,6 +727,19 @@ static int tw686x_s_fmt_vid_cap(struct file *file, void *priv,
 	else
 		val &= ~BIT(24);
 
+	val &= ~0x7ffff;
+
+	/* Program the DMA scatter-gather */
+	if (dev->dma_mode == TW686X_DMA_MODE_SG) {
+		u32 start_idx, end_idx;
+
+		start_idx = is_second_gen(dev) ?
+				0 : vc->ch * TW686X_MAX_SG_DESC_COUNT;
+		end_idx = start_idx + TW686X_MAX_SG_DESC_COUNT - 1;
+
+		val |= (end_idx << 10) | start_idx;
+	}
+
 	val &= ~(0x7 << 20);
 	val |= vc->format->mode << 20;
 	reg_write(vc->dev, VDMA_CHANNEL_CONFIG[vc->ch], val);
@@ -891,6 +1077,8 @@ int tw686x_video_init(struct tw686x_dev *dev)
 		dev->dma_ops = &memcpy_dma_ops;
 	else if (dev->dma_mode == TW686X_DMA_MODE_CONTIG)
 		dev->dma_ops = &contig_dma_ops;
+	else if (dev->dma_mode == TW686X_DMA_MODE_SG)
+		dev->dma_ops = &sg_dma_ops;
 	else
 		return -EINVAL;
 

drivers/media/pci/tw686x/tw686x.h

@@ -34,6 +34,7 @@
 
 #define TW686X_DMA_MODE_MEMCPY		0
 #define TW686X_DMA_MODE_CONTIG		1
+#define TW686X_DMA_MODE_SG		2
 
 struct tw686x_format {
 	char *name;
@@ -48,6 +49,12 @@ struct tw686x_dma_desc {
 	unsigned int size;
 };
 
+struct tw686x_sg_desc {
+	/* 3 MSBits for flags, 13 LSBits for length */
+	__le32 flags_length;
+	__le32 phys;
+};
+
 struct tw686x_audio_buf {
 	dma_addr_t dma;
 	void *virt;
@@ -80,6 +87,7 @@ struct tw686x_video_channel {
 	struct video_device *device;
 	struct tw686x_v4l2_buf *curr_bufs[2];
 	struct tw686x_dma_desc dma_descs[2];
+	struct tw686x_sg_desc *sg_descs[2];
 
 	struct v4l2_ctrl_handler ctrl_handler;
 	const struct tw686x_format *format;
@@ -154,6 +162,12 @@ static inline unsigned int max_channels(struct tw686x_dev *dev)
 	return dev->type & TYPE_MAX_CHANNELS; /* 4 or 8 channels */
 }
 
+static inline unsigned is_second_gen(struct tw686x_dev *dev)
+{
+	/* each channel has its own DMA SG table */
+	return dev->type & TYPE_SECOND_GEN;
+}
+
 void tw686x_enable_channel(struct tw686x_dev *dev, unsigned int channel);
 void tw686x_disable_channel(struct tw686x_dev *dev, unsigned int channel);