V4L/DVB (6047): ivtv: Fix scatter/gather DMA timeouts

It turns out that the cx23415/6 DMA engine cannot do scatter/gather DMA
reliably. Every so often depending on the phase of the moon and your
hardware configuration the cx2341x DMA engine simply chokes on it and
you have to reboot to get it working again.

This change replaced the scatter/gather DMA by single transfers at a time,
where the driver is now responsible for DMA-ing each buffer.

UDMA is still done using scatter/gather DMA, that will be fixed soon.

Many thanks to Mark Bryars <mark.bryars@etvinteractive.com> for discovering
the link between scatter/gather and the DMA timeouts.

Signed-off-by: Hans Verkuil <hverkuil@xs4all.nl>
Signed-off-by: Mauro Carvalho Chehab <mchehab@infradead.org>

5 files changed, 196 insertions, 133 deletions

diff --git a/drivers/media/video/ivtv/ivtv-driver.h b/drivers/media/video/ivtv/ivtv-driver.h
index f5de2fd01b1a..e80f9f65a905 100644
--- a/drivers/media/video/ivtv/ivtv-driver.h
+++ b/drivers/media/video/ivtv/ivtv-driver.h
@@ -382,7 +382,6 @@ struct ivtv_mailbox_data {
 #define IVTV_F_I_RADIO_USER	   5 	/* The radio tuner is selected */
 #define IVTV_F_I_DIG_RST	   6 	/* Reset digitizer */
 #define IVTV_F_I_DEC_YUV	   7 	/* YUV instead of MPG is being decoded */
-#define IVTV_F_I_ENC_VBI	   8 	/* VBI DMA */
 #define IVTV_F_I_UPDATE_CC	   9  	/* CC should be updated */
 #define IVTV_F_I_UPDATE_WSS	   10 	/* WSS should be updated */
 #define IVTV_F_I_UPDATE_VPS	   11 	/* VPS should be updated */
@@ -405,7 +404,7 @@ struct ivtv_mailbox_data {
 #define IVTV_F_I_EV_VSYNC_ENABLED  31 	/* VSYNC event enabled */
 
 /* Scatter-Gather array element, used in DMA transfers */
-struct ivtv_SG_element {
+struct ivtv_sg_element {
 	u32 src;
 	u32 dst;
 	u32 size;
@@ -417,7 +416,7 @@ struct ivtv_user_dma {
 	struct page *map[IVTV_DMA_SG_OSD_ENT];
 
 	/* Base Dev SG Array for cx23415/6 */
-	struct ivtv_SG_element SGarray[IVTV_DMA_SG_OSD_ENT];
+	struct ivtv_sg_element SGarray[IVTV_DMA_SG_OSD_ENT];
 	dma_addr_t SG_handle;
 	int SG_length;
 
@@ -468,6 +467,10 @@ struct ivtv_stream {
 	int dma;		/* can be PCI_DMA_TODEVICE,
 				   PCI_DMA_FROMDEVICE or
 				   PCI_DMA_NONE */
+	u32 pending_offset;
+	u32 pending_backup;
+	u64 pending_pts;
+
 	u32 dma_offset;
 	u32 dma_backup;
 	u64 dma_pts;
@@ -493,10 +496,13 @@ struct ivtv_stream {
 	u16 dma_xfer_cnt;
 
 	/* Base Dev SG Array for cx23415/6 */
-	struct ivtv_SG_element *SGarray;
-	struct ivtv_SG_element *PIOarray;
-	dma_addr_t SG_handle;
-	int SG_length;
+	struct ivtv_sg_element *sg_pending;
+	struct ivtv_sg_element *sg_processing;
+	struct ivtv_sg_element *sg_dma;
+	dma_addr_t sg_handle;
+	int sg_pending_size;
+	int sg_processing_size;
+	int sg_processed;
 
 	/* SG List of Buffers */
 	struct scatterlist *SGlist;
@@ -637,7 +643,6 @@ struct vbi_info {
 	u32 enc_start, enc_size;
 	int fpi;
 	u32 frame;
-	u32 dma_offset;
 	u8 cc_data_odd[256];
 	u8 cc_data_even[256];
 	int cc_pos;
@@ -724,6 +729,7 @@ struct ivtv {
 	int cur_pio_stream;	/* index of stream doing PIO */
 	u32 dma_data_req_offset;
 	u32 dma_data_req_size;
+	int dma_retries;
 	int output_mode;        /* NONE, MPG, YUV, UDMA YUV, passthrough */
 	spinlock_t lock;        /* lock access to this struct */
 	int search_pack_header;
diff --git a/drivers/media/video/ivtv/ivtv-irq.c b/drivers/media/video/ivtv/ivtv-irq.c
index 8644f3dda31e..9695e5356163 100644
--- a/drivers/media/video/ivtv/ivtv-irq.c
+++ b/drivers/media/video/ivtv/ivtv-irq.c
@@ -60,18 +60,18 @@ static void ivtv_pio_work_handler(struct ivtv *itv)
 	buf = list_entry(s->q_dma.list.next, struct ivtv_buffer, list);
 	list_for_each(p, &s->q_dma.list) {
 		struct ivtv_buffer *buf = list_entry(p, struct ivtv_buffer, list);
-		u32 size = s->PIOarray[i].size & 0x3ffff;
+		u32 size = s->sg_processing[i].size & 0x3ffff;
 
 		/* Copy the data from the card to the buffer */
 		if (s->type == IVTV_DEC_STREAM_TYPE_VBI) {
-			memcpy_fromio(buf->buf, itv->dec_mem + s->PIOarray[i].src - IVTV_DECODER_OFFSET, size);
+			memcpy_fromio(buf->buf, itv->dec_mem + s->sg_processing[i].src - IVTV_DECODER_OFFSET, size);
 		}
 		else {
-			memcpy_fromio(buf->buf, itv->enc_mem + s->PIOarray[i].src, size);
+			memcpy_fromio(buf->buf, itv->enc_mem + s->sg_processing[i].src, size);
 		}
-		if (s->PIOarray[i].size & 0x80000000)
-			break;
 		i++;
+		if (i == s->sg_processing_size)
+			break;
 	}
 	write_reg(IVTV_IRQ_ENC_PIO_COMPLETE, 0x44);
 }
@@ -105,7 +105,7 @@ static int stream_enc_dma_append(struct ivtv_stream *s, u32 data[CX2341X_MBOX_MA
 	u32 offset, size;
 	u32 UVoffset = 0, UVsize = 0;
 	int skip_bufs = s->q_predma.buffers;
-	int idx = s->SG_length;
+	int idx = s->sg_pending_size;
 	int rc;
 
 	/* sanity checks */
@@ -123,7 +123,7 @@ static int stream_enc_dma_append(struct ivtv_stream *s, u32 data[CX2341X_MBOX_MA
 		case IVTV_ENC_STREAM_TYPE_MPG:
 			offset = data[1];
 			size = data[2];
-			s->dma_pts = 0;
+			s->pending_pts = 0;
 			break;
 
 		case IVTV_ENC_STREAM_TYPE_YUV:
@@ -131,13 +131,13 @@ static int stream_enc_dma_append(struct ivtv_stream *s, u32 data[CX2341X_MBOX_MA
 			size = data[2];
 			UVoffset = data[3];
 			UVsize = data[4];
-			s->dma_pts = ((u64) data[5] << 32) | data[6];
+			s->pending_pts = ((u64) data[5] << 32) | data[6];
 			break;
 
 		case IVTV_ENC_STREAM_TYPE_PCM:
 			offset = data[1] + 12;
 			size = data[2] - 12;
-			s->dma_pts = read_dec(offset - 8) |
+			s->pending_pts = read_dec(offset - 8) |
 				((u64)(read_dec(offset - 12)) << 32);
 			if (itv->has_cx23415)
 				offset += IVTV_DECODER_OFFSET;
@@ -150,13 +150,13 @@ static int stream_enc_dma_append(struct ivtv_stream *s, u32 data[CX2341X_MBOX_MA
 				IVTV_DEBUG_INFO("VBI offset == 0\n");
 				return -1;
 			}
-			s->dma_pts = read_enc(offset - 4) | ((u64)read_enc(offset - 8) << 32);
+			s->pending_pts = read_enc(offset - 4) | ((u64)read_enc(offset - 8) << 32);
 			break;
 
 		case IVTV_DEC_STREAM_TYPE_VBI:
 			size = read_dec(itv->vbi.dec_start + 4) + 8;
 			offset = read_dec(itv->vbi.dec_start) + itv->vbi.dec_start;
-			s->dma_pts = 0;
+			s->pending_pts = 0;
 			offset += IVTV_DECODER_OFFSET;
 			break;
 		default:
@@ -165,17 +165,17 @@ static int stream_enc_dma_append(struct ivtv_stream *s, u32 data[CX2341X_MBOX_MA
 	}
 
 	/* if this is the start of the DMA then fill in the magic cookie */
-	if (s->SG_length == 0) {
+	if (s->sg_pending_size == 0) {
 		if (itv->has_cx23415 && (s->type == IVTV_ENC_STREAM_TYPE_PCM ||
 		    s->type == IVTV_DEC_STREAM_TYPE_VBI)) {
-			s->dma_backup = read_dec(offset - IVTV_DECODER_OFFSET);
+			s->pending_backup = read_dec(offset - IVTV_DECODER_OFFSET);
 			write_dec_sync(cpu_to_le32(DMA_MAGIC_COOKIE), offset - IVTV_DECODER_OFFSET);
 		}
 		else {
-			s->dma_backup = read_enc(offset);
+			s->pending_backup = read_enc(offset);
 			write_enc_sync(cpu_to_le32(DMA_MAGIC_COOKIE), offset);
 		}
-		s->dma_offset = offset;
+		s->pending_offset = offset;
 	}
 
 	bytes_needed = size;
@@ -202,7 +202,7 @@ static int stream_enc_dma_append(struct ivtv_stream *s, u32 data[CX2341X_MBOX_MA
 	}
 	s->buffers_stolen = rc;
 
-	/* got the buffers, now fill in SGarray (DMA) */
+	/* got the buffers, now fill in sg_pending */
 	buf = list_entry(s->q_predma.list.next, struct ivtv_buffer, list);
 	memset(buf->buf, 0, 128);
 	list_for_each(p, &s->q_predma.list) {
@@ -210,9 +210,9 @@ static int stream_enc_dma_append(struct ivtv_stream *s, u32 data[CX2341X_MBOX_MA
 
 		if (skip_bufs-- > 0)
 			continue;
-		s->SGarray[idx].dst = cpu_to_le32(buf->dma_handle);
-		s->SGarray[idx].src = cpu_to_le32(offset);
-		s->SGarray[idx].size = cpu_to_le32(s->buf_size);
+		s->sg_pending[idx].dst = buf->dma_handle;
+		s->sg_pending[idx].src = offset;
+		s->sg_pending[idx].size = s->buf_size;
 		buf->bytesused = (size < s->buf_size) ? size : s->buf_size;
 		buf->dma_xfer_cnt = s->dma_xfer_cnt;
 
@@ -230,7 +230,7 @@ static int stream_enc_dma_append(struct ivtv_stream *s, u32 data[CX2341X_MBOX_MA
 		}
 		idx++;
 	}
-	s->SG_length = idx;
+	s->sg_pending_size = idx;
 	return 0;
 }
 
@@ -332,9 +332,9 @@ void ivtv_dma_stream_dec_prepare(struct ivtv_stream *s, u32 offset, int lock)
 			offset = uv_offset;
 			y_done = 1;
 		}
-		s->SGarray[idx].src = cpu_to_le32(buf->dma_handle);
-		s->SGarray[idx].dst = cpu_to_le32(offset);
-		s->SGarray[idx].size = cpu_to_le32(buf->bytesused);
+		s->sg_pending[idx].src = buf->dma_handle;
+		s->sg_pending[idx].dst = offset;
+		s->sg_pending[idx].size = buf->bytesused;
 
 		offset += buf->bytesused;
 		bytes_written += buf->bytesused;
@@ -343,10 +343,7 @@ void ivtv_dma_stream_dec_prepare(struct ivtv_stream *s, u32 offset, int lock)
 		ivtv_buf_sync_for_device(s, buf);
 		idx++;
 	}
-	s->SG_length = idx;
-
-	/* Mark last buffer size for Interrupt flag */
-	s->SGarray[s->SG_length - 1].size |= cpu_to_le32(0x80000000);
+	s->sg_pending_size = idx;
 
 	/* Sync Hardware SG List of buffers */
 	ivtv_stream_sync_for_device(s);
@@ -362,6 +359,34 @@ void ivtv_dma_stream_dec_prepare(struct ivtv_stream *s, u32 offset, int lock)
 		spin_unlock_irqrestore(&itv->dma_reg_lock, flags);
 }
 
+static void ivtv_dma_enc_start_xfer(struct ivtv_stream *s)
+{
+	struct ivtv *itv = s->itv;
+
+	s->sg_dma->src = cpu_to_le32(s->sg_processing[s->sg_processed].src);
+	s->sg_dma->dst = cpu_to_le32(s->sg_processing[s->sg_processed].dst);
+	s->sg_dma->size = cpu_to_le32(s->sg_processing[s->sg_processed].size | 0x80000000);
+	s->sg_processed++;
+	/* Sync Hardware SG List of buffers */
+	ivtv_stream_sync_for_device(s);
+	write_reg(s->sg_handle, IVTV_REG_ENCDMAADDR);
+	write_reg_sync(read_reg(IVTV_REG_DMAXFER) | 0x02, IVTV_REG_DMAXFER);
+}
+
+static void ivtv_dma_dec_start_xfer(struct ivtv_stream *s)
+{
+	struct ivtv *itv = s->itv;
+
+	s->sg_dma->src = cpu_to_le32(s->sg_processing[s->sg_processed].src);
+	s->sg_dma->dst = cpu_to_le32(s->sg_processing[s->sg_processed].dst);
+	s->sg_dma->size = cpu_to_le32(s->sg_processing[s->sg_processed].size | 0x80000000);
+	s->sg_processed++;
+	/* Sync Hardware SG List of buffers */
+	ivtv_stream_sync_for_device(s);
+	write_reg(s->sg_handle, IVTV_REG_DECDMAADDR);
+	write_reg_sync(read_reg(IVTV_REG_DMAXFER) | 0x01, IVTV_REG_DMAXFER);
+}
+
 /* start the encoder DMA */
 static void ivtv_dma_enc_start(struct ivtv_stream *s)
 {
@@ -375,8 +400,7 @@ static void ivtv_dma_enc_start(struct ivtv_stream *s)
 		ivtv_queue_move(s, &s->q_predma, NULL, &s->q_dma, s->q_predma.bytesused);
 
 	if (ivtv_use_dma(s))
-		s->SGarray[s->SG_length - 1].size =
-			cpu_to_le32(le32_to_cpu(s->SGarray[s->SG_length - 1].size) + 256);
+		s->sg_pending[s->sg_pending_size - 1].size += 256;
 
 	/* If this is an MPEG stream, and VBI data is also pending, then append the
 	   VBI DMA to the MPEG DMA and transfer both sets of data at once.
@@ -387,45 +411,39 @@ static void ivtv_dma_enc_start(struct ivtv_stream *s)
 	   sure we only use the MPEG DMA to transfer the VBI DMA if both are in
 	   use. This way no conflicts occur. */
 	clear_bit(IVTV_F_S_DMA_HAS_VBI, &s->s_flags);
-	if (s->type == IVTV_ENC_STREAM_TYPE_MPG && s_vbi->SG_length &&
-			s->SG_length + s_vbi->SG_length <= s->buffers) {
+	if (s->type == IVTV_ENC_STREAM_TYPE_MPG && s_vbi->sg_pending_size &&
+			s->sg_pending_size + s_vbi->sg_pending_size <= s->buffers) {
 		ivtv_queue_move(s_vbi, &s_vbi->q_predma, NULL, &s_vbi->q_dma, s_vbi->q_predma.bytesused);
 		if (ivtv_use_dma(s_vbi))
-			s_vbi->SGarray[s_vbi->SG_length - 1].size = cpu_to_le32(le32_to_cpu(s_vbi->SGarray[s->SG_length - 1].size) + 256);
-		for (i = 0; i < s_vbi->SG_length; i++) {
-			s->SGarray[s->SG_length++] = s_vbi->SGarray[i];
+			s_vbi->sg_pending[s_vbi->sg_pending_size - 1].size += 256;
+		for (i = 0; i < s_vbi->sg_pending_size; i++) {
+			s->sg_pending[s->sg_pending_size++] = s_vbi->sg_pending[i];
 		}
-		itv->vbi.dma_offset = s_vbi->dma_offset;
-		s_vbi->SG_length = 0;
+		s_vbi->dma_offset = s_vbi->pending_offset;
+		s_vbi->sg_pending_size = 0;
 		s_vbi->dma_xfer_cnt++;
 		set_bit(IVTV_F_S_DMA_HAS_VBI, &s->s_flags);
 		IVTV_DEBUG_HI_DMA("include DMA for %s\n", s->name);
 	}
 
-	/* Mark last buffer size for Interrupt flag */
-	s->SGarray[s->SG_length - 1].size |= cpu_to_le32(0x80000000);
 	s->dma_xfer_cnt++;
-
-	if (s->type == IVTV_ENC_STREAM_TYPE_VBI)
-		set_bit(IVTV_F_I_ENC_VBI, &itv->i_flags);
-	else
-		clear_bit(IVTV_F_I_ENC_VBI, &itv->i_flags);
+	memcpy(s->sg_processing, s->sg_pending, sizeof(struct ivtv_sg_element) * s->sg_pending_size);
+	s->sg_processing_size = s->sg_pending_size;
+	s->sg_pending_size = 0;
+	s->sg_processed = 0;
+	s->dma_offset = s->pending_offset;
+	s->dma_backup = s->pending_backup;
+	s->dma_pts = s->pending_pts;
 
 	if (ivtv_use_pio(s)) {
-		for (i = 0; i < s->SG_length; i++) {
-			s->PIOarray[i].src = le32_to_cpu(s->SGarray[i].src);
-			s->PIOarray[i].size = le32_to_cpu(s->SGarray[i].size);
-		}
 		set_bit(IVTV_F_I_WORK_HANDLER_PIO, &itv->i_flags);
 		set_bit(IVTV_F_I_HAVE_WORK, &itv->i_flags);
 		set_bit(IVTV_F_I_PIO, &itv->i_flags);
 		itv->cur_pio_stream = s->type;
 	}
 	else {
-		/* Sync Hardware SG List of buffers */
-		ivtv_stream_sync_for_device(s);
-		write_reg(s->SG_handle, IVTV_REG_ENCDMAADDR);
-		write_reg_sync(read_reg(IVTV_REG_DMAXFER) | 0x02, IVTV_REG_DMAXFER);
+		itv->dma_retries = 0;
+		ivtv_dma_enc_start_xfer(s);
 		set_bit(IVTV_F_I_DMA, &itv->i_flags);
 		itv->cur_dma_stream = s->type;
 		itv->dma_timer.expires = jiffies + msecs_to_jiffies(100);
@@ -439,10 +457,15 @@ static void ivtv_dma_dec_start(struct ivtv_stream *s)
 
 	if (s->q_predma.bytesused)
 		ivtv_queue_move(s, &s->q_predma, NULL, &s->q_dma, s->q_predma.bytesused);
+	s->dma_xfer_cnt++;
+	memcpy(s->sg_processing, s->sg_pending, sizeof(struct ivtv_sg_element) * s->sg_pending_size);
+	s->sg_processing_size = s->sg_pending_size;
+	s->sg_pending_size = 0;
+	s->sg_processed = 0;
+
 	IVTV_DEBUG_HI_DMA("start DMA for %s\n", s->name);
-	/* put SG Handle into register 0x0c */
-	write_reg(s->SG_handle, IVTV_REG_DECDMAADDR);
-	write_reg_sync(read_reg(IVTV_REG_DMAXFER) | 0x01, IVTV_REG_DMAXFER);
+	itv->dma_retries = 0;
+	ivtv_dma_dec_start_xfer(s);
 	set_bit(IVTV_F_I_DMA, &itv->i_flags);
 	itv->cur_dma_stream = s->type;
 	itv->dma_timer.expires = jiffies + msecs_to_jiffies(100);
@@ -453,27 +476,42 @@ static void ivtv_irq_dma_read(struct ivtv *itv)
 {
 	struct ivtv_stream *s = NULL;
 	struct ivtv_buffer *buf;
-	int hw_stream_type;
+	int hw_stream_type = 0;
 
 	IVTV_DEBUG_HI_IRQ("DEC DMA READ\n");
-	del_timer(&itv->dma_timer);
-	if (read_reg(IVTV_REG_DMASTATUS) & 0x14) {
-		IVTV_DEBUG_WARN("DEC DMA ERROR %x\n", read_reg(IVTV_REG_DMASTATUS));
-		write_reg(read_reg(IVTV_REG_DMASTATUS) & 3, IVTV_REG_DMASTATUS);
+	if (!test_bit(IVTV_F_I_UDMA, &itv->i_flags) && itv->cur_dma_stream < 0) {
+		del_timer(&itv->dma_timer);
+		return;
 	}
+
 	if (!test_bit(IVTV_F_I_UDMA, &itv->i_flags)) {
-		if (test_bit(IVTV_F_I_DEC_YUV, &itv->i_flags)) {
-			s = &itv->streams[IVTV_DEC_STREAM_TYPE_YUV];
-			hw_stream_type = 2;
+		s = &itv->streams[itv->cur_dma_stream];
+		ivtv_stream_sync_for_cpu(s);
+
+		if (read_reg(IVTV_REG_DMASTATUS) & 0x14) {
+			IVTV_DEBUG_WARN("DEC DMA ERROR %x (xfer %d of %d, retry %d)\n",
+					read_reg(IVTV_REG_DMASTATUS),
+					s->sg_processed, s->sg_processing_size, itv->dma_retries);
+			write_reg(read_reg(IVTV_REG_DMASTATUS) & 3, IVTV_REG_DMASTATUS);
+			if (itv->dma_retries == 3) {
+				itv->dma_retries = 0;
+			}
+			else {
+				/* Retry, starting with the first xfer segment.
+				   Just retrying the current segment is not sufficient. */
+				s->sg_processed = 0;
+				itv->dma_retries++;
+			}
 		}
-		else {
-			s = &itv->streams[IVTV_DEC_STREAM_TYPE_MPG];
-			hw_stream_type = 0;
+		if (s->sg_processed < s->sg_processing_size) {
+			/* DMA next buffer */
+			ivtv_dma_dec_start_xfer(s);
+			return;
 		}
+		if (s->type == IVTV_DEC_STREAM_TYPE_YUV)
+			hw_stream_type = 2;
 		IVTV_DEBUG_HI_DMA("DEC DATA READ %s: %d\n", s->name, s->q_dma.bytesused);
 
-		ivtv_stream_sync_for_cpu(s);
-
 		/* For some reason must kick the firmware, like PIO mode,
 		   I think this tells the firmware we are done and the size
 		   of the xfer so it can calculate what we need next.
@@ -490,6 +528,7 @@ static void ivtv_irq_dma_read(struct ivtv *itv)
 		}
 		wake_up(&s->waitq);
 	}
+	del_timer(&itv->dma_timer);
 	clear_bit(IVTV_F_I_UDMA, &itv->i_flags);
 	clear_bit(IVTV_F_I_DMA, &itv->i_flags);
 	itv->cur_dma_stream = -1;
@@ -501,33 +540,44 @@ static void ivtv_irq_enc_dma_complete(struct ivtv *itv)
 	u32 data[CX2341X_MBOX_MAX_DATA];
 	struct ivtv_stream *s;
 
-	del_timer(&itv->dma_timer);
 	ivtv_api_get_data(&itv->enc_mbox, IVTV_MBOX_DMA_END, data);
-	IVTV_DEBUG_HI_IRQ("ENC DMA COMPLETE %x %d\n", data[0], data[1]);
-	if (test_and_clear_bit(IVTV_F_I_ENC_VBI, &itv->i_flags))
-		data[1] = 3;
-	else if (data[1] > 2)
+	IVTV_DEBUG_HI_IRQ("ENC DMA COMPLETE %x %d (%d)\n", data[0], data[1], itv->cur_dma_stream);
+	if (itv->cur_dma_stream < 0) {
+		del_timer(&itv->dma_timer);
 		return;
-	s = &itv->streams[ivtv_stream_map[data[1]]];
+	}
+	s = &itv->streams[itv->cur_dma_stream];
+	ivtv_stream_sync_for_cpu(s);
+
 	if (data[0] & 0x18) {
-		IVTV_DEBUG_WARN("ENC DMA ERROR %x\n", data[0]);
+		IVTV_DEBUG_WARN("ENC DMA ERROR %x (offset %08x, xfer %d of %d, retry %d)\n", data[0],
+			s->dma_offset, s->sg_processed, s->sg_processing_size, itv->dma_retries);
 		write_reg(read_reg(IVTV_REG_DMASTATUS) & 3, IVTV_REG_DMASTATUS);
-		ivtv_vapi(itv, CX2341X_ENC_SCHED_DMA_TO_HOST, 3, 0, 0, data[1]);
+		if (itv->dma_retries == 3) {
+			itv->dma_retries = 0;
+		}
+		else {
+			/* Retry, starting with the first xfer segment.
+			   Just retrying the current segment is not sufficient. */
+			s->sg_processed = 0;
+			itv->dma_retries++;
+		}
 	}
-	s->SG_length = 0;
+	if (s->sg_processed < s->sg_processing_size) {
+		/* DMA next buffer */
+		ivtv_dma_enc_start_xfer(s);
+		return;
+	}
+	del_timer(&itv->dma_timer);
 	clear_bit(IVTV_F_I_DMA, &itv->i_flags);
 	itv->cur_dma_stream = -1;
 	dma_post(s);
-	ivtv_stream_sync_for_cpu(s);
 	if (test_and_clear_bit(IVTV_F_S_DMA_HAS_VBI, &s->s_flags)) {
-		u32 tmp;
-
 		s = &itv->streams[IVTV_ENC_STREAM_TYPE_VBI];
-		tmp = s->dma_offset;
-		s->dma_offset = itv->vbi.dma_offset;
 		dma_post(s);
-		s->dma_offset = tmp;
 	}
+	s->sg_processing_size = 0;
+	s->sg_processed = 0;
 	wake_up(&itv->dma_waitq);
 }
 
@@ -541,8 +591,7 @@ static void ivtv_irq_enc_pio_complete(struct ivtv *itv)
 	}
 	s = &itv->streams[itv->cur_pio_stream];
 	IVTV_DEBUG_HI_IRQ("ENC PIO COMPLETE %s\n", s->name);
-	s->SG_length = 0;
-	clear_bit(IVTV_F_I_ENC_VBI, &itv->i_flags);
+	s->sg_pending_size = 0;
 	clear_bit(IVTV_F_I_PIO, &itv->i_flags);
 	itv->cur_pio_stream = -1;
 	dma_post(s);
@@ -554,13 +603,8 @@ static void ivtv_irq_enc_pio_complete(struct ivtv *itv)
 		ivtv_vapi(itv, CX2341X_ENC_SCHED_DMA_TO_HOST, 3, 0, 0, 2);
 	clear_bit(IVTV_F_I_PIO, &itv->i_flags);
 	if (test_and_clear_bit(IVTV_F_S_DMA_HAS_VBI, &s->s_flags)) {
-		u32 tmp;
-
 		s = &itv->streams[IVTV_ENC_STREAM_TYPE_VBI];
-		tmp = s->dma_offset;
-		s->dma_offset = itv->vbi.dma_offset;
 		dma_post(s);
-		s->dma_offset = tmp;
 	}
 	wake_up(&itv->dma_waitq);
 }
@@ -572,19 +616,23 @@ static void ivtv_irq_dma_err(struct ivtv *itv)
 	del_timer(&itv->dma_timer);
 	ivtv_api_get_data(&itv->enc_mbox, IVTV_MBOX_DMA_END, data);
 	IVTV_DEBUG_WARN("DMA ERROR %08x %08x %08x %d\n", data[0], data[1],
-					read_reg(IVTV_REG_DMASTATUS), itv->cur_dma_stream);
+				read_reg(IVTV_REG_DMASTATUS), itv->cur_dma_stream);
+	write_reg(read_reg(IVTV_REG_DMASTATUS) & 3, IVTV_REG_DMASTATUS);
 	if (!test_bit(IVTV_F_I_UDMA, &itv->i_flags) &&
 	    itv->cur_dma_stream >= 0 && itv->cur_dma_stream < IVTV_MAX_STREAMS) {
 		struct ivtv_stream *s = &itv->streams[itv->cur_dma_stream];
 
 		/* retry */
-		write_reg(read_reg(IVTV_REG_DMASTATUS) & 3, IVTV_REG_DMASTATUS);
 		if (s->type >= IVTV_DEC_STREAM_TYPE_MPG)
 			ivtv_dma_dec_start(s);
 		else
 			ivtv_dma_enc_start(s);
 		return;
 	}
+	if (test_bit(IVTV_F_I_UDMA, &itv->i_flags)) {
+		ivtv_udma_start(itv);
+		return;
+	}
 	clear_bit(IVTV_F_I_UDMA, &itv->i_flags);
 	clear_bit(IVTV_F_I_DMA, &itv->i_flags);
 	itv->cur_dma_stream = -1;
@@ -628,14 +676,14 @@ static void ivtv_irq_enc_vbi_cap(struct ivtv *itv)
 	   DMA the data. Since at most four VBI DMA buffers are available,
 	   we just drop the old requests when there are already three
 	   requests queued. */
-	if (s->SG_length > 2) {
+	if (s->sg_pending_size > 2) {
 		struct list_head *p;
 		list_for_each(p, &s->q_predma.list) {
 			struct ivtv_buffer *buf = list_entry(p, struct ivtv_buffer, list);
 			ivtv_buf_sync_for_cpu(s, buf);
 		}
 		ivtv_queue_move(s, &s->q_predma, NULL, &s->q_free, 0);
-		s->SG_length = 0;
+		s->sg_pending_size = 0;
 	}
 	/* if we can append the data, and the MPEG stream isn't capturing,
 	   then start a DMA request for just the VBI data. */
diff --git a/drivers/media/video/ivtv/ivtv-queue.c b/drivers/media/video/ivtv/ivtv-queue.c
index bff75aeee0a0..d9a1478ca1f7 100644
--- a/drivers/media/video/ivtv/ivtv-queue.c
+++ b/drivers/media/video/ivtv/ivtv-queue.c
@@ -195,7 +195,7 @@ void ivtv_flush_queues(struct ivtv_stream *s)
 int ivtv_stream_alloc(struct ivtv_stream *s)
 {
 	struct ivtv *itv = s->itv;
-	int SGsize = sizeof(struct ivtv_SG_element) * s->buffers;
+	int SGsize = sizeof(struct ivtv_sg_element) * s->buffers;
 	int i;
 
 	if (s->buffers == 0)
@@ -205,27 +205,33 @@ int ivtv_stream_alloc(struct ivtv_stream *s)
 		s->dma != PCI_DMA_NONE ? "DMA " : "",
 		s->name, s->buffers, s->buf_size, s->buffers * s->buf_size / 1024);
 
-	if (ivtv_might_use_pio(s)) {
-		s->PIOarray = (struct ivtv_SG_element *)kzalloc(SGsize, GFP_KERNEL);
-		if (s->PIOarray == NULL) {
-			IVTV_ERR("Could not allocate PIOarray for %s stream\n", s->name);
-			return -ENOMEM;
-		}
+	s->sg_pending = (struct ivtv_sg_element *)kzalloc(SGsize, GFP_KERNEL);
+	if (s->sg_pending == NULL) {
+		IVTV_ERR("Could not allocate sg_pending for %s stream\n", s->name);
+		return -ENOMEM;
 	}
+	s->sg_pending_size = 0;
 
-	/* Allocate DMA SG Arrays */
-	s->SGarray = (struct ivtv_SG_element *)kzalloc(SGsize, GFP_KERNEL);
-	if (s->SGarray == NULL) {
-		IVTV_ERR("Could not allocate SGarray for %s stream\n", s->name);
-		if (ivtv_might_use_pio(s)) {
-			kfree(s->PIOarray);
-			s->PIOarray = NULL;
-		}
+	s->sg_processing = (struct ivtv_sg_element *)kzalloc(SGsize, GFP_KERNEL);
+	if (s->sg_processing == NULL) {
+		IVTV_ERR("Could not allocate sg_processing for %s stream\n", s->name);
+		kfree(s->sg_pending);
+		s->sg_pending = NULL;
+		return -ENOMEM;
+	}
+	s->sg_processing_size = 0;
+
+	s->sg_dma = (struct ivtv_sg_element *)kzalloc(sizeof(struct ivtv_sg_element), GFP_KERNEL);
+	if (s->sg_dma == NULL) {
+		IVTV_ERR("Could not allocate sg_dma for %s stream\n", s->name);
+		kfree(s->sg_pending);
+		s->sg_pending = NULL;
+		kfree(s->sg_processing);
+		s->sg_processing = NULL;
 		return -ENOMEM;
 	}
-	s->SG_length = 0;
 	if (ivtv_might_use_dma(s)) {
-		s->SG_handle = pci_map_single(itv->dev, s->SGarray, SGsize, s->dma);
+		s->sg_handle = pci_map_single(itv->dev, s->sg_dma, sizeof(struct ivtv_sg_element), s->dma);
 		ivtv_stream_sync_for_cpu(s);
 	}
 
@@ -272,16 +278,19 @@ void ivtv_stream_free(struct ivtv_stream *s)
 	}
 
 	/* Free SG Array/Lists */
-	if (s->SGarray != NULL) {
-		if (s->SG_handle != IVTV_DMA_UNMAPPED) {
-			pci_unmap_single(s->itv->dev, s->SG_handle,
-				 sizeof(struct ivtv_SG_element) * s->buffers, PCI_DMA_TODEVICE);
-			s->SG_handle = IVTV_DMA_UNMAPPED;
+	if (s->sg_dma != NULL) {
+		if (s->sg_handle != IVTV_DMA_UNMAPPED) {
+			pci_unmap_single(s->itv->dev, s->sg_handle,
+				 sizeof(struct ivtv_sg_element), PCI_DMA_TODEVICE);
+			s->sg_handle = IVTV_DMA_UNMAPPED;
 		}
-		kfree(s->SGarray);
-		kfree(s->PIOarray);
-		s->PIOarray = NULL;
-		s->SGarray = NULL;
-		s->SG_length = 0;
+		kfree(s->sg_pending);
+		kfree(s->sg_processing);
+		kfree(s->sg_dma);
+		s->sg_pending = NULL;
+		s->sg_processing = NULL;
+		s->sg_dma = NULL;
+		s->sg_pending_size = 0;
+		s->sg_processing_size = 0;
 	}
 }
diff --git a/drivers/media/video/ivtv/ivtv-queue.h b/drivers/media/video/ivtv/ivtv-queue.h
index 2ed8d548255d..14a9f7fe50aa 100644
--- a/drivers/media/video/ivtv/ivtv-queue.h
+++ b/drivers/media/video/ivtv/ivtv-queue.h
@@ -79,13 +79,13 @@ void ivtv_stream_free(struct ivtv_stream *s);
 static inline void ivtv_stream_sync_for_cpu(struct ivtv_stream *s)
 {
 	if (ivtv_use_dma(s))
-		pci_dma_sync_single_for_cpu(s->itv->dev, s->SG_handle,
-			sizeof(struct ivtv_SG_element) * s->buffers, PCI_DMA_TODEVICE);
+		pci_dma_sync_single_for_cpu(s->itv->dev, s->sg_handle,
+			sizeof(struct ivtv_sg_element), PCI_DMA_TODEVICE);
 }
 
 static inline void ivtv_stream_sync_for_device(struct ivtv_stream *s)
 {
 	if (ivtv_use_dma(s))
-		pci_dma_sync_single_for_device(s->itv->dev, s->SG_handle,
-			sizeof(struct ivtv_SG_element) * s->buffers, PCI_DMA_TODEVICE);
+		pci_dma_sync_single_for_device(s->itv->dev, s->sg_handle,
+			sizeof(struct ivtv_sg_element), PCI_DMA_TODEVICE);
 }
diff --git a/drivers/media/video/ivtv/ivtv-streams.c b/drivers/media/video/ivtv/ivtv-streams.c
index 0582b9d57c55..d1cc366c2a35 100644
--- a/drivers/media/video/ivtv/ivtv-streams.c
+++ b/drivers/media/video/ivtv/ivtv-streams.c
@@ -154,7 +154,7 @@ static void ivtv_stream_init(struct ivtv *itv, int type)
 	spin_lock_init(&s->qlock);
 	init_waitqueue_head(&s->waitq);
 	s->id = -1;
-	s->SG_handle = IVTV_DMA_UNMAPPED;
+	s->sg_handle = IVTV_DMA_UNMAPPED;
 	ivtv_queue_init(&s->q_free);
 	ivtv_queue_init(&s->q_full);
 	ivtv_queue_init(&s->q_dma);