V4L/DVB (9801): cx18: Allow more than 63 capture buffers in rotation per stream

cx18: Allow more than 63 capture buffers in rotation per stream.  Implement
q_busy to hold buffers the firmware has for use.  q_free holds truly unused
buffers in a pool.  New buffers are given to the firmware as soon as the
firmware returns one, if there are any to give to the firmware.

Signed-off-by: Andy Walls <awalls@radix.net>
Signed-off-by: Mauro Carvalho Chehab <mchehab@redhat.com>
diff --git a/drivers/media/video/cx18/cx18-driver.h b/drivers/media/video/cx18/cx18-driver.h
index f88d823..4d56d07 100644
--- a/drivers/media/video/cx18/cx18-driver.h
+++ b/drivers/media/video/cx18/cx18-driver.h
@@ -289,7 +289,8 @@
 
 	/* Buffer Queues */
 	struct cx18_queue q_free;	/* free buffers */
-	struct cx18_queue q_full;	/* full buffers */
+	struct cx18_queue q_busy;	/* busy buffers - in use by firmware */
+	struct cx18_queue q_full;	/* full buffers - data for user apps */
 
 	/* DVB / Digital Transport */
 	struct cx18_dvb dvb;
diff --git a/drivers/media/video/cx18/cx18-fileops.c b/drivers/media/video/cx18/cx18-fileops.c
index 504a1f9..61192e6 100644
--- a/drivers/media/video/cx18/cx18-fileops.c
+++ b/drivers/media/video/cx18/cx18-fileops.c
@@ -187,7 +187,7 @@
 				while ((buf = cx18_dequeue(s_vbi, &s_vbi->q_full))) {
 					/* byteswap and process VBI data */
 /*					cx18_process_vbi_data(cx, buf, s_vbi->dma_pts, s_vbi->type); */
-					cx18_enqueue(s_vbi, buf, &s_vbi->q_free);
+					cx18_stream_put_buf_fw(s_vbi, buf);
 				}
 			}
 			buf = &cx->vbi.sliced_mpeg_buf;
@@ -361,15 +361,9 @@
 				tot_count - tot_written);
 
 		if (buf != &cx->vbi.sliced_mpeg_buf) {
-			if (buf->readpos == buf->bytesused) {
-				cx18_buf_sync_for_device(s, buf);
-				cx18_enqueue(s, buf, &s->q_free);
-				cx18_vapi(cx, CX18_CPU_DE_SET_MDL, 5,
-					s->handle,
-					(void __iomem *)&cx->scb->cpu_mdl[buf->id] -
-					  cx->enc_mem,
-					1, buf->id, s->buf_size);
-			} else
+			if (buf->readpos == buf->bytesused)
+				cx18_stream_put_buf_fw(s, buf);
+			else
 				cx18_push(s, buf, &s->q_full);
 		} else if (buf->readpos == buf->bytesused) {
 			int idx = cx->vbi.inserted_frame % CX18_VBI_FRAMES;
diff --git a/drivers/media/video/cx18/cx18-ioctl.c b/drivers/media/video/cx18/cx18-ioctl.c
index 8b2ec31b..ece799e 100644
--- a/drivers/media/video/cx18/cx18-ioctl.c
+++ b/drivers/media/video/cx18/cx18-ioctl.c
@@ -746,8 +746,7 @@
 			continue;
 		CX18_INFO("Stream %s: status 0x%04lx, %d%% of %d KiB (%d buffers) in use\n",
 			  s->name, s->s_flags,
-			  (s->buffers - atomic_read(&s->q_free.buffers))
-				* 100 / s->buffers,
+			  atomic_read(&s->q_full.buffers) * 100 / s->buffers,
 			  (s->buffers * s->buf_size) / 1024, s->buffers);
 	}
 	CX18_INFO("Read MPEG/VBI: %lld/%lld bytes\n",
diff --git a/drivers/media/video/cx18/cx18-mailbox.c b/drivers/media/video/cx18/cx18-mailbox.c
index b013e81..9d82205 100644
--- a/drivers/media/video/cx18/cx18-mailbox.c
+++ b/drivers/media/video/cx18/cx18-mailbox.c
@@ -189,16 +189,7 @@
 			dvb_dmx_swfilter(&s->dvb.demux, buf->buf,
 					 buf->bytesused);
 
-			cx18_buf_sync_for_device(s, buf);
-			cx18_enqueue(s, buf, &s->q_free);
-
-			if (s->handle != CX18_INVALID_TASK_HANDLE &&
-			    test_bit(CX18_F_S_STREAMING, &s->s_flags))
-				cx18_vapi(cx,
-				       CX18_CPU_DE_SET_MDL, 5, s->handle,
-				       (void __iomem *)
-				       &cx->scb->cpu_mdl[buf->id] - cx->enc_mem,
-				       1, buf->id, s->buf_size);
+			cx18_stream_put_buf_fw(s, buf);
 		} else
 			set_bit(CX18_F_B_NEED_BUF_SWAP, &buf->b_flags);
 	}
diff --git a/drivers/media/video/cx18/cx18-queue.c b/drivers/media/video/cx18/cx18-queue.c
index fdfc83e..40379d8 100644
--- a/drivers/media/video/cx18/cx18-queue.c
+++ b/drivers/media/video/cx18/cx18-queue.c
@@ -42,24 +42,32 @@
 	q->bytesused = 0;
 }
 
-void _cx18_enqueue(struct cx18_stream *s, struct cx18_buffer *buf,
-		   struct cx18_queue *q, int to_front)
+struct cx18_queue *_cx18_enqueue(struct cx18_stream *s, struct cx18_buffer *buf,
+				 struct cx18_queue *q, int to_front)
 {
-	/* clear the buffer if it is going to be enqueued to the free queue */
-	if (q == &s->q_free) {
+	/* clear the buffer if it is not to be enqueued to the full queue */
+	if (q != &s->q_full) {
 		buf->bytesused = 0;
 		buf->readpos = 0;
 		buf->b_flags = 0;
 		buf->skipped = 0;
 	}
+
 	mutex_lock(&s->qlock);
+
+	/* q_busy is restricted to 63 buffers to stay within firmware limits */
+	if (q == &s->q_busy && atomic_read(&q->buffers) >= 63)
+		q = &s->q_free;
+
 	if (to_front)
 		list_add(&buf->list, &q->list); /* LIFO */
 	else
 		list_add_tail(&buf->list, &q->list); /* FIFO */
-	atomic_inc(&q->buffers);
 	q->bytesused += buf->bytesused - buf->readpos;
+	atomic_inc(&q->buffers);
+
 	mutex_unlock(&s->qlock);
+	return q;
 }
 
 struct cx18_buffer *cx18_dequeue(struct cx18_stream *s, struct cx18_queue *q)
@@ -70,9 +78,9 @@
 	if (!list_empty(&q->list)) {
 		buf = list_entry(q->list.next, struct cx18_buffer, list);
 		list_del_init(q->list.next);
-		atomic_dec(&q->buffers);
 		q->bytesused -= buf->bytesused - buf->readpos;
 		buf->skipped = 0;
+		atomic_dec(&q->buffers);
 	}
 	mutex_unlock(&s->qlock);
 	return buf;
@@ -85,28 +93,30 @@
 	struct cx18_buffer *buf;
 	struct cx18_buffer *ret = NULL;
 	struct list_head *p, *t;
-	LIST_HEAD(r);
 
 	mutex_lock(&s->qlock);
-	list_for_each_safe(p, t, &s->q_free.list) {
+	list_for_each_safe(p, t, &s->q_busy.list) {
 		buf = list_entry(p, struct cx18_buffer, list);
 
 		if (buf->id != id) {
 			buf->skipped++;
-			if (buf->skipped >= atomic_read(&s->q_free.buffers)-1) {
+			if (buf->skipped >= atomic_read(&s->q_busy.buffers)-1) {
 				/* buffer must have fallen out of rotation */
-				atomic_dec(&s->q_free.buffers);
-				list_move_tail(&buf->list, &r);
 				CX18_WARN("Skipped %s, buffer %d, %d "
 					  "times - it must have dropped out of "
 					  "rotation\n", s->name, buf->id,
 					  buf->skipped);
+				/* move it to q_free */
+				list_move_tail(&buf->list, &s->q_free.list);
+				buf->bytesused = buf->readpos = buf->b_flags =
+					buf->skipped = 0;
+				atomic_dec(&s->q_busy.buffers);
+				atomic_inc(&s->q_free.buffers);
 			}
 			continue;
 		}
 
 		buf->bytesused = bytesused;
-		atomic_dec(&s->q_free.buffers);
 		if (s->type == CX18_ENC_STREAM_TYPE_TS) {
 			/*
 			 * TS doesn't use q_full, but for sweeping up lost
@@ -116,28 +126,19 @@
 			 */
 			list_del_init(&buf->list);
 		} else {
-			atomic_inc(&s->q_full.buffers);
-			s->q_full.bytesused += buf->bytesused;
 			list_move_tail(&buf->list, &s->q_full.list);
+			s->q_full.bytesused += buf->bytesused;
+			atomic_inc(&s->q_full.buffers);
 		}
+		atomic_dec(&s->q_busy.buffers);
 
 		ret = buf;
 		break;
 	}
-	mutex_unlock(&s->qlock);
 
-	/* Put lost buffers back into firmware transfer rotation */
-	while (!list_empty(&r)) {
-		buf = list_entry(r.next, struct cx18_buffer, list);
-		list_del_init(r.next);
-		cx18_enqueue(s, buf, &s->q_free);
-		cx18_vapi(cx, CX18_CPU_DE_SET_MDL, 5, s->handle,
-		       (void __iomem *)&cx->scb->cpu_mdl[buf->id] - cx->enc_mem,
-		       1, buf->id, s->buf_size);
-		CX18_INFO("Returning %s, buffer %d back to transfer rotation\n",
-			  s->name, buf->id);
-		/* and there was much rejoicing... */
-	}
+	/* Put more buffers into the transfer rotation from q_free, if we can */
+	cx18_stream_load_fw_queue_nolock(s);
+	mutex_unlock(&s->qlock);
 	return ret;
 }
 
@@ -162,6 +163,7 @@
 
 void cx18_flush_queues(struct cx18_stream *s)
 {
+	cx18_queue_flush(s, &s->q_busy);
 	cx18_queue_flush(s, &s->q_full);
 }
 
diff --git a/drivers/media/video/cx18/cx18-queue.h b/drivers/media/video/cx18/cx18-queue.h
index d184e55..456cec3 100644
--- a/drivers/media/video/cx18/cx18-queue.h
+++ b/drivers/media/video/cx18/cx18-queue.h
@@ -43,21 +43,21 @@
 void cx18_buf_swap(struct cx18_buffer *buf);
 
 /* cx18_queue utility functions */
-void _cx18_enqueue(struct cx18_stream *s, struct cx18_buffer *buf,
-		   struct cx18_queue *q, int to_front);
+struct cx18_queue *_cx18_enqueue(struct cx18_stream *s, struct cx18_buffer *buf,
+				 struct cx18_queue *q, int to_front);
 
 static inline
-void cx18_enqueue(struct cx18_stream *s, struct cx18_buffer *buf,
-		  struct cx18_queue *q)
+struct cx18_queue *cx18_enqueue(struct cx18_stream *s, struct cx18_buffer *buf,
+				struct cx18_queue *q)
 {
-	_cx18_enqueue(s, buf, q, 0); /* FIFO */
+	return _cx18_enqueue(s, buf, q, 0); /* FIFO */
 }
 
 static inline
-void cx18_push(struct cx18_stream *s, struct cx18_buffer *buf,
-	       struct cx18_queue *q)
+struct cx18_queue *cx18_push(struct cx18_stream *s, struct cx18_buffer *buf,
+			     struct cx18_queue *q)
 {
-	_cx18_enqueue(s, buf, q, 1); /* LIFO */
+	return _cx18_enqueue(s, buf, q, 1); /* LIFO */
 }
 
 void cx18_queue_init(struct cx18_queue *q);
diff --git a/drivers/media/video/cx18/cx18-streams.c b/drivers/media/video/cx18/cx18-streams.c
index a1cf434..6b0b7f75 100644
--- a/drivers/media/video/cx18/cx18-streams.c
+++ b/drivers/media/video/cx18/cx18-streams.c
@@ -127,16 +127,11 @@
 	s->buf_size = cx->stream_buf_size[type];
 	if (s->buf_size)
 		s->buffers = max_size / s->buf_size;
-	if (s->buffers > 63) {
-		/* Each stream has a maximum of 63 buffers,
-		   ensure we do not exceed that. */
-		s->buffers = 63;
-		s->buf_size = (max_size / s->buffers) & ~0xfff;
-	}
 	mutex_init(&s->qlock);
 	init_waitqueue_head(&s->waitq);
 	s->id = -1;
 	cx18_queue_init(&s->q_free);
+	cx18_queue_init(&s->q_busy);
 	cx18_queue_init(&s->q_full);
 }
 
@@ -401,11 +396,61 @@
 		cx18_api(cx, CX18_CPU_SET_RAW_VBI_PARAM, 6, data);
 }
 
+struct cx18_queue *cx18_stream_put_buf_fw(struct cx18_stream *s,
+					  struct cx18_buffer *buf)
+{
+	struct cx18 *cx = s->cx;
+	struct cx18_queue *q;
+
+	/* Don't give it to the firmware, if we're not running a capture */
+	if (s->handle == CX18_INVALID_TASK_HANDLE ||
+	    !test_bit(CX18_F_S_STREAMING, &s->s_flags))
+		return cx18_enqueue(s, buf, &s->q_free);
+
+	q = cx18_enqueue(s, buf, &s->q_busy);
+	if (q != &s->q_busy)
+		return q; /* The firmware has the max buffers it can handle */
+
+	cx18_buf_sync_for_device(s, buf);
+	cx18_vapi(cx, CX18_CPU_DE_SET_MDL, 5, s->handle,
+		  (void __iomem *) &cx->scb->cpu_mdl[buf->id] - cx->enc_mem,
+		  1, buf->id, s->buf_size);
+	return q;
+}
+
+/* Must hold s->qlock when calling */
+void cx18_stream_load_fw_queue_nolock(struct cx18_stream *s)
+{
+	struct cx18_buffer *buf;
+	struct cx18 *cx = s->cx;
+
+	/* Move from q_free to q_busy notifying the firmware: 63 buf limit */
+	while (s->handle != CX18_INVALID_TASK_HANDLE &&
+	       test_bit(CX18_F_S_STREAMING, &s->s_flags) &&
+	       atomic_read(&s->q_busy.buffers) < 63 &&
+	       !list_empty(&s->q_free.list)) {
+
+		/* Move from q_free to q_busy */
+		buf = list_entry(s->q_free.list.next, struct cx18_buffer, list);
+		list_move_tail(&buf->list, &s->q_busy.list);
+		buf->bytesused = buf->readpos = buf->b_flags = buf->skipped = 0;
+		atomic_dec(&s->q_free.buffers);
+		atomic_inc(&s->q_busy.buffers);
+
+		/* Notify firmware */
+		cx18_buf_sync_for_device(s, buf);
+		cx18_vapi(cx, CX18_CPU_DE_SET_MDL, 5, s->handle,
+		  (void __iomem *) &cx->scb->cpu_mdl[buf->id] - cx->enc_mem,
+		  1, buf->id, s->buf_size);
+	}
+}
+
 int cx18_start_v4l2_encode_stream(struct cx18_stream *s)
 {
 	u32 data[MAX_MB_ARGUMENTS];
 	struct cx18 *cx = s->cx;
 	struct list_head *p;
+	struct cx18_buffer *buf;
 	int ts = 0;
 	int captype = 0;
 
@@ -488,16 +533,18 @@
 		(void __iomem *)&cx->scb->cpu_mdl_ack[s->type][0] - cx->enc_mem,
 		(void __iomem *)&cx->scb->cpu_mdl_ack[s->type][1] - cx->enc_mem);
 
+	/* Init all the cpu_mdls for this stream */
+	cx18_flush_queues(s);
+	mutex_lock(&s->qlock);
 	list_for_each(p, &s->q_free.list) {
-		struct cx18_buffer *buf = list_entry(p, struct cx18_buffer, list);
-
+		buf = list_entry(p, struct cx18_buffer, list);
 		cx18_writel(cx, buf->dma_handle,
 					&cx->scb->cpu_mdl[buf->id].paddr);
 		cx18_writel(cx, s->buf_size, &cx->scb->cpu_mdl[buf->id].length);
-		cx18_vapi(cx, CX18_CPU_DE_SET_MDL, 5, s->handle,
-			(void __iomem *)&cx->scb->cpu_mdl[buf->id] - cx->enc_mem,
-			1, buf->id, s->buf_size);
 	}
+	cx18_stream_load_fw_queue_nolock(s);
+	mutex_unlock(&s->qlock);
+
 	/* begin_capture */
 	if (cx18_vapi(cx, CX18_CPU_CAPTURE_START, 1, s->handle)) {
 		CX18_DEBUG_WARN("Error starting capture!\n");
@@ -506,9 +553,15 @@
 			cx18_vapi(cx, CX18_CPU_CAPTURE_STOP, 2, s->handle, 1);
 		else
 			cx18_vapi(cx, CX18_CPU_CAPTURE_STOP, 1, s->handle);
+		clear_bit(CX18_F_S_STREAMING, &s->s_flags);
+		/* FIXME - CX18_F_S_STREAMOFF as well? */
 		cx18_vapi(cx, CX18_CPU_DE_RELEASE_MDL, 1, s->handle);
 		cx18_vapi(cx, CX18_DESTROY_TASK, 1, s->handle);
-		/* FIXME - clean-up DSP0_INT mask, i_flags, s_flags, etc. */
+		s->handle = CX18_INVALID_TASK_HANDLE;
+		if (atomic_read(&cx->tot_capturing) == 0) {
+			set_bit(CX18_F_I_EOS, &cx->i_flags);
+			cx18_write_reg(cx, 5, CX18_DSP0_INTERRUPT_MASK);
+		}
 		return -EINVAL;
 	}
 
diff --git a/drivers/media/video/cx18/cx18-streams.h b/drivers/media/video/cx18/cx18-streams.h
index 7218b15..635d34b 100644
--- a/drivers/media/video/cx18/cx18-streams.h
+++ b/drivers/media/video/cx18/cx18-streams.h
@@ -29,6 +29,9 @@
 void cx18_streams_cleanup(struct cx18 *cx, int unregister);
 
 /* Capture related */
+void cx18_stream_load_fw_queue_nolock(struct cx18_stream *s);
+struct cx18_queue *cx18_stream_put_buf_fw(struct cx18_stream *s,
+					  struct cx18_buffer *buf);
 int cx18_start_v4l2_encode_stream(struct cx18_stream *s);
 int cx18_stop_v4l2_encode_stream(struct cx18_stream *s, int gop_end);