Merge branch 'next' of git://git.kernel.org/pub/scm/linux/kernel/git/djbw/async_tx

* 'next' of git://git.kernel.org/pub/scm/linux/kernel/git/djbw/async_tx:
  dma: Add SoF and EoF debugging to ipu_idmac.c, minor cleanup
  dw_dmac: add cyclic API to DW DMA driver
  dmaengine: Add privatecnt to revert DMA_PRIVATE property
  dmatest: add dma interrupts and callbacks
  dmatest: add xor test
  dmaengine: allow dma support for async_tx to be toggled
  async_tx: provide __async_inline for HAS_DMA=n archs
  dmaengine: kill some unused headers
  dmaengine: initialize tx_list in dma_async_tx_descriptor_init
  dma: i.MX31 IPU DMA robustness improvements
  dma: improve section assignment in i.MX31 IPU DMA driver
  dma: ipu_idmac driver cosmetic clean-up
  dmaengine: fail device registration if channel registration fails

16 files changed, 925 insertions, 241 deletions

diff --git a/crypto/async_tx/async_tx.c b/crypto/async_tx/async_tx.c
index f21147f3626..06eb6cc09fe 100644
--- a/crypto/async_tx/async_tx.c
+++ b/crypto/async_tx/async_tx.c
@@ -30,7 +30,7 @@
 #ifdef CONFIG_DMA_ENGINE
 static int __init async_tx_init(void)
 {
-	dmaengine_get();
+	async_dmaengine_get();
 
 	printk(KERN_INFO "async_tx: api initialized (async)\n");
 
@@ -39,7 +39,7 @@ static int __init async_tx_init(void)
 
 static void __exit async_tx_exit(void)
 {
-	dmaengine_put();
+	async_dmaengine_put();
 }
 
 /**
@@ -56,7 +56,7 @@ __async_tx_find_channel(struct dma_async_tx_descriptor *depend_tx,
 	if (depend_tx &&
 	    dma_has_cap(tx_type, depend_tx->chan->device->cap_mask))
 		return depend_tx->chan;
-	return dma_find_channel(tx_type);
+	return async_dma_find_channel(tx_type);
 }
 EXPORT_SYMBOL_GPL(__async_tx_find_channel);
 #else
diff --git a/crypto/async_tx/async_xor.c b/crypto/async_tx/async_xor.c
index 595b78672b3..95fe2c8d6c5 100644
--- a/crypto/async_tx/async_xor.c
+++ b/crypto/async_tx/async_xor.c
@@ -30,11 +30,8 @@
 #include <linux/raid/xor.h>
 #include <linux/async_tx.h>
 
-/* do_async_xor - dma map the pages and perform the xor with an engine.
- * 	This routine is marked __always_inline so it can be compiled away
- * 	when CONFIG_DMA_ENGINE=n
- */
-static __always_inline struct dma_async_tx_descriptor *
+/* do_async_xor - dma map the pages and perform the xor with an engine */
+static __async_inline struct dma_async_tx_descriptor *
 do_async_xor(struct dma_chan *chan, struct page *dest, struct page **src_list,
 	     unsigned int offset, int src_cnt, size_t len,
 	     enum async_tx_flags flags,
diff --git a/drivers/dma/Kconfig b/drivers/dma/Kconfig
index 48ea59e7967..3b3c01b6f1e 100644
--- a/drivers/dma/Kconfig
+++ b/drivers/dma/Kconfig
@@ -98,6 +98,17 @@ config NET_DMA
 	  Say Y here if you enabled INTEL_IOATDMA or FSL_DMA, otherwise
 	  say N.
 
+config ASYNC_TX_DMA
+	bool "Async_tx: Offload support for the async_tx api"
+	depends on DMA_ENGINE
+	help
+	  This allows the async_tx api to take advantage of offload engines for
+	  memcpy, memset, xor, and raid6 p+q operations.  If your platform has
+	  a dma engine that can perform raid operations and you have enabled
+	  MD_RAID456 say Y.
+
+	  If unsure, say N.
+
 config DMATEST
 	tristate "DMA Test client"
 	depends on DMA_ENGINE
diff --git a/drivers/dma/dmaengine.c b/drivers/dma/dmaengine.c
index 280a9d263eb..92438e9dacc 100644
--- a/drivers/dma/dmaengine.c
+++ b/drivers/dma/dmaengine.c
@@ -507,6 +507,7 @@ struct dma_chan *__dma_request_channel(dma_cap_mask_t *mask, dma_filter_fn fn, v
 			 * published in the general-purpose allocator
 			 */
 			dma_cap_set(DMA_PRIVATE, device->cap_mask);
+			device->privatecnt++;
 			err = dma_chan_get(chan);
 
 			if (err == -ENODEV) {
@@ -518,6 +519,8 @@ struct dma_chan *__dma_request_channel(dma_cap_mask_t *mask, dma_filter_fn fn, v
 				       dma_chan_name(chan), err);
 			else
 				break;
+			if (--device->privatecnt == 0)
+				dma_cap_clear(DMA_PRIVATE, device->cap_mask);
 			chan->private = NULL;
 			chan = NULL;
 		}
@@ -537,6 +540,9 @@ void dma_release_channel(struct dma_chan *chan)
 	WARN_ONCE(chan->client_count != 1,
 		  "chan reference count %d != 1\n", chan->client_count);
 	dma_chan_put(chan);
+	/* drop PRIVATE cap enabled by __dma_request_channel() */
+	if (--chan->device->privatecnt == 0)
+		dma_cap_clear(DMA_PRIVATE, chan->device->cap_mask);
 	chan->private = NULL;
 	mutex_unlock(&dma_list_mutex);
 }
@@ -602,6 +608,24 @@ void dmaengine_put(void)
 }
 EXPORT_SYMBOL(dmaengine_put);
 
+static int get_dma_id(struct dma_device *device)
+{
+	int rc;
+
+ idr_retry:
+	if (!idr_pre_get(&dma_idr, GFP_KERNEL))
+		return -ENOMEM;
+	mutex_lock(&dma_list_mutex);
+	rc = idr_get_new(&dma_idr, NULL, &device->dev_id);
+	mutex_unlock(&dma_list_mutex);
+	if (rc == -EAGAIN)
+		goto idr_retry;
+	else if (rc != 0)
+		return rc;
+
+	return 0;
+}
+
 /**
  * dma_async_device_register - registers DMA devices found
  * @device: &dma_device
@@ -640,27 +664,25 @@ int dma_async_device_register(struct dma_device *device)
 	idr_ref = kmalloc(sizeof(*idr_ref), GFP_KERNEL);
 	if (!idr_ref)
 		return -ENOMEM;
-	atomic_set(idr_ref, 0);
- idr_retry:
-	if (!idr_pre_get(&dma_idr, GFP_KERNEL))
-		return -ENOMEM;
-	mutex_lock(&dma_list_mutex);
-	rc = idr_get_new(&dma_idr, NULL, &device->dev_id);
-	mutex_unlock(&dma_list_mutex);
-	if (rc == -EAGAIN)
-		goto idr_retry;
-	else if (rc != 0)
+	rc = get_dma_id(device);
+	if (rc != 0) {
+		kfree(idr_ref);
 		return rc;
+	}
+
+	atomic_set(idr_ref, 0);
 
 	/* represent channels in sysfs. Probably want devs too */
 	list_for_each_entry(chan, &device->channels, device_node) {
+		rc = -ENOMEM;
 		chan->local = alloc_percpu(typeof(*chan->local));
 		if (chan->local == NULL)
-			continue;
+			goto err_out;
 		chan->dev = kzalloc(sizeof(*chan->dev), GFP_KERNEL);
 		if (chan->dev == NULL) {
 			free_percpu(chan->local);
-			continue;
+			chan->local = NULL;
+			goto err_out;
 		}
 
 		chan->chan_id = chancnt++;
@@ -677,6 +699,8 @@ int dma_async_device_register(struct dma_device *device)
 		if (rc) {
 			free_percpu(chan->local);
 			chan->local = NULL;
+			kfree(chan->dev);
+			atomic_dec(idr_ref);
 			goto err_out;
 		}
 		chan->client_count = 0;
@@ -701,12 +725,23 @@ int dma_async_device_register(struct dma_device *device)
 			}
 		}
 	list_add_tail_rcu(&device->global_node, &dma_device_list);
+	if (dma_has_cap(DMA_PRIVATE, device->cap_mask))
+		device->privatecnt++;	/* Always private */
 	dma_channel_rebalance();
 	mutex_unlock(&dma_list_mutex);
 
 	return 0;
 
 err_out:
+	/* if we never registered a channel just release the idr */
+	if (atomic_read(idr_ref) == 0) {
+		mutex_lock(&dma_list_mutex);
+		idr_remove(&dma_idr, device->dev_id);
+		mutex_unlock(&dma_list_mutex);
+		kfree(idr_ref);
+		return rc;
+	}
+
 	list_for_each_entry(chan, &device->channels, device_node) {
 		if (chan->local == NULL)
 			continue;
@@ -893,6 +928,7 @@ void dma_async_tx_descriptor_init(struct dma_async_tx_descriptor *tx,
 {
 	tx->chan = chan;
 	spin_lock_init(&tx->lock);
+	INIT_LIST_HEAD(&tx->tx_list);
 }
 EXPORT_SYMBOL(dma_async_tx_descriptor_init);
 
diff --git a/drivers/dma/dmatest.c b/drivers/dma/dmatest.c
index e190d8b3070..a27c0fb1bc1 100644
--- a/drivers/dma/dmatest.c
+++ b/drivers/dma/dmatest.c
@@ -38,6 +38,11 @@ module_param(max_channels, uint, S_IRUGO);
 MODULE_PARM_DESC(max_channels,
 		"Maximum number of channels to use (default: all)");
 
+static unsigned int xor_sources = 3;
+module_param(xor_sources, uint, S_IRUGO);
+MODULE_PARM_DESC(xor_sources,
+		"Number of xor source buffers (default: 3)");
+
 /*
  * Initialization patterns. All bytes in the source buffer has bit 7
  * set, all bytes in the destination buffer has bit 7 cleared.
@@ -59,8 +64,9 @@ struct dmatest_thread {
 	struct list_head	node;
 	struct task_struct	*task;
 	struct dma_chan		*chan;
-	u8			*srcbuf;
-	u8			*dstbuf;
+	u8			**srcs;
+	u8			**dsts;
+	enum dma_transaction_type type;
 };
 
 struct dmatest_chan {
@@ -98,30 +104,37 @@ static unsigned long dmatest_random(void)
 	return buf;
 }
 
-static void dmatest_init_srcbuf(u8 *buf, unsigned int start, unsigned int len)
+static void dmatest_init_srcs(u8 **bufs, unsigned int start, unsigned int len)
 {
 	unsigned int i;
-
-	for (i = 0; i < start; i++)
-		buf[i] = PATTERN_SRC | (~i & PATTERN_COUNT_MASK);
-	for ( ; i < start + len; i++)
-		buf[i] = PATTERN_SRC | PATTERN_COPY
-			| (~i & PATTERN_COUNT_MASK);;
-	for ( ; i < test_buf_size; i++)
-		buf[i] = PATTERN_SRC | (~i & PATTERN_COUNT_MASK);
+	u8 *buf;
+
+	for (; (buf = *bufs); bufs++) {
+		for (i = 0; i < start; i++)
+			buf[i] = PATTERN_SRC | (~i & PATTERN_COUNT_MASK);
+		for ( ; i < start + len; i++)
+			buf[i] = PATTERN_SRC | PATTERN_COPY
+				| (~i & PATTERN_COUNT_MASK);;
+		for ( ; i < test_buf_size; i++)
+			buf[i] = PATTERN_SRC | (~i & PATTERN_COUNT_MASK);
+		buf++;
+	}
 }
 
-static void dmatest_init_dstbuf(u8 *buf, unsigned int start, unsigned int len)
+static void dmatest_init_dsts(u8 **bufs, unsigned int start, unsigned int len)
 {
 	unsigned int i;
-
-	for (i = 0; i < start; i++)
-		buf[i] = PATTERN_DST | (~i & PATTERN_COUNT_MASK);
-	for ( ; i < start + len; i++)
-		buf[i] = PATTERN_DST | PATTERN_OVERWRITE
-			| (~i & PATTERN_COUNT_MASK);
-	for ( ; i < test_buf_size; i++)
-		buf[i] = PATTERN_DST | (~i & PATTERN_COUNT_MASK);
+	u8 *buf;
+
+	for (; (buf = *bufs); bufs++) {
+		for (i = 0; i < start; i++)
+			buf[i] = PATTERN_DST | (~i & PATTERN_COUNT_MASK);
+		for ( ; i < start + len; i++)
+			buf[i] = PATTERN_DST | PATTERN_OVERWRITE
+				| (~i & PATTERN_COUNT_MASK);
+		for ( ; i < test_buf_size; i++)
+			buf[i] = PATTERN_DST | (~i & PATTERN_COUNT_MASK);
+	}
 }
 
 static void dmatest_mismatch(u8 actual, u8 pattern, unsigned int index,
@@ -150,23 +163,30 @@ static void dmatest_mismatch(u8 actual, u8 pattern, unsigned int index,
 				thread_name, index, expected, actual);
 }
 
-static unsigned int dmatest_verify(u8 *buf, unsigned int start,
+static unsigned int dmatest_verify(u8 **bufs, unsigned int start,
 		unsigned int end, unsigned int counter, u8 pattern,
 		bool is_srcbuf)
 {
 	unsigned int i;
 	unsigned int error_count = 0;
 	u8 actual;
-
-	for (i = start; i < end; i++) {
-		actual = buf[i];
-		if (actual != (pattern | (~counter & PATTERN_COUNT_MASK))) {
-			if (error_count < 32)
-				dmatest_mismatch(actual, pattern, i, counter,
-						is_srcbuf);
-			error_count++;
+	u8 expected;
+	u8 *buf;
+	unsigned int counter_orig = counter;
+
+	for (; (buf = *bufs); bufs++) {
+		counter = counter_orig;
+		for (i = start; i < end; i++) {
+			actual = buf[i];
+			expected = pattern | (~counter & PATTERN_COUNT_MASK);
+			if (actual != expected) {
+				if (error_count < 32)
+					dmatest_mismatch(actual, pattern, i,
+							 counter, is_srcbuf);
+				error_count++;
+			}
+			counter++;
 		}
-		counter++;
 	}
 
 	if (error_count > 32)
@@ -176,12 +196,17 @@ static unsigned int dmatest_verify(u8 *buf, unsigned int start,
 	return error_count;
 }
 
+static void dmatest_callback(void *completion)
+{
+	complete(completion);
+}
+
 /*
  * This function repeatedly tests DMA transfers of various lengths and
- * offsets until it is told to exit by kthread_stop(). There may be
- * multiple threads running this function in parallel for a single
- * channel, and there may be multiple channels being tested in
- * parallel.
+ * offsets for a given operation type until it is told to exit by
+ * kthread_stop(). There may be multiple threads running this function
+ * in parallel for a single channel, and there may be multiple channels
+ * being tested in parallel.
  *
  * Before each test, the source and destination buffer is initialized
  * with a known pattern. This pattern is different depending on
@@ -201,25 +226,57 @@ static int dmatest_func(void *data)
 	unsigned int		total_tests = 0;
 	dma_cookie_t		cookie;
 	enum dma_status		status;
+	enum dma_ctrl_flags 	flags;
 	int			ret;
+	int			src_cnt;
+	int			dst_cnt;
+	int			i;
 
 	thread_name = current->comm;
 
 	ret = -ENOMEM;
-	thread->srcbuf = kmalloc(test_buf_size, GFP_KERNEL);
-	if (!thread->srcbuf)
-		goto err_srcbuf;
-	thread->dstbuf = kmalloc(test_buf_size, GFP_KERNEL);
-	if (!thread->dstbuf)
-		goto err_dstbuf;
 
 	smp_rmb();
 	chan = thread->chan;
+	if (thread->type == DMA_MEMCPY)
+		src_cnt = dst_cnt = 1;
+	else if (thread->type == DMA_XOR) {
+		src_cnt = xor_sources | 1; /* force odd to ensure dst = src */
+		dst_cnt = 1;
+	} else
+		goto err_srcs;
+
+	thread->srcs = kcalloc(src_cnt+1, sizeof(u8 *), GFP_KERNEL);
+	if (!thread->srcs)
+		goto err_srcs;
+	for (i = 0; i < src_cnt; i++) {
+		thread->srcs[i] = kmalloc(test_buf_size, GFP_KERNEL);
+		if (!thread->srcs[i])
+			goto err_srcbuf;
+	}
+	thread->srcs[i] = NULL;
+
+	thread->dsts = kcalloc(dst_cnt+1, sizeof(u8 *), GFP_KERNEL);
+	if (!thread->dsts)
+		goto err_dsts;
+	for (i = 0; i < dst_cnt; i++) {
+		thread->dsts[i] = kmalloc(test_buf_size, GFP_KERNEL);
+		if (!thread->dsts[i])
+			goto err_dstbuf;
+	}
+	thread->dsts[i] = NULL;
+
+	set_user_nice(current, 10);
+
+	flags = DMA_CTRL_ACK | DMA_COMPL_SKIP_DEST_UNMAP | DMA_PREP_INTERRUPT;
 
 	while (!kthread_should_stop()) {
 		struct dma_device *dev = chan->device;
-		struct dma_async_tx_descriptor *tx;
-		dma_addr_t dma_src, dma_dest;
+		struct dma_async_tx_descriptor *tx = NULL;
+		dma_addr_t dma_srcs[src_cnt];
+		dma_addr_t dma_dsts[dst_cnt];
+		struct completion cmp;
+		unsigned long tmo = msecs_to_jiffies(3000);
 
 		total_tests++;
 
@@ -227,22 +284,41 @@ static int dmatest_func(void *data)
 		src_off = dmatest_random() % (test_buf_size - len + 1);
 		dst_off = dmatest_random() % (test_buf_size - len + 1);
 
-		dmatest_init_srcbuf(thread->srcbuf, src_off, len);
-		dmatest_init_dstbuf(thread->dstbuf, dst_off, len);
+		dmatest_init_srcs(thread->srcs, src_off, len);
+		dmatest_init_dsts(thread->dsts, dst_off, len);
 
-		dma_src = dma_map_single(dev->dev, thread->srcbuf + src_off,
-				len, DMA_TO_DEVICE);
+		for (i = 0; i < src_cnt; i++) {
+			u8 *buf = thread->srcs[i] + src_off;
+
+			dma_srcs[i] = dma_map_single(dev->dev, buf, len,
+						     DMA_TO_DEVICE);
+		}
 		/* map with DMA_BIDIRECTIONAL to force writeback/invalidate */
-		dma_dest = dma_map_single(dev->dev, thread->dstbuf,
-				test_buf_size, DMA_BIDIRECTIONAL);
+		for (i = 0; i < dst_cnt; i++) {
+			dma_dsts[i] = dma_map_single(dev->dev, thread->dsts[i],
+						     test_buf_size,
+						     DMA_BIDIRECTIONAL);
+		}
+
+		if (thread->type == DMA_MEMCPY)
+			tx = dev->device_prep_dma_memcpy(chan,
+							 dma_dsts[0] + dst_off,
+							 dma_srcs[0], len,
+							 flags);
+		else if (thread->type == DMA_XOR)
+			tx = dev->device_prep_dma_xor(chan,
+						      dma_dsts[0] + dst_off,
+						      dma_srcs, xor_sources,
+						      len, flags);
 
-		tx = dev->device_prep_dma_memcpy(chan, dma_dest + dst_off,
-				dma_src, len,
-				DMA_CTRL_ACK | DMA_COMPL_SKIP_DEST_UNMAP);
 		if (!tx) {
-			dma_unmap_single(dev->dev, dma_src, len, DMA_TO_DEVICE);
-			dma_unmap_single(dev->dev, dma_dest,
-					test_buf_size, DMA_BIDIRECTIONAL);
+			for (i = 0; i < src_cnt; i++)
+				dma_unmap_single(dev->dev, dma_srcs[i], len,
+						 DMA_TO_DEVICE);
+			for (i = 0; i < dst_cnt; i++)
+				dma_unmap_single(dev->dev, dma_dsts[i],
+						 test_buf_size,
+						 DMA_BIDIRECTIONAL);
 			pr_warning("%s: #%u: prep error with src_off=0x%x "
 					"dst_off=0x%x len=0x%x\n",
 					thread_name, total_tests - 1,
@@ -251,7 +327,10 @@ static int dmatest_func(void *data)
 			failed_tests++;
 			continue;
 		}
-		tx->callback = NULL;
+
+		init_completion(&cmp);
+		tx->callback = dmatest_callback;
+		tx->callback_param = &cmp;
 		cookie = tx->tx_submit(tx);
 
 		if (dma_submit_error(cookie)) {
@@ -263,44 +342,50 @@ static int dmatest_func(void *data)
 			failed_tests++;
 			continue;
 		}
-		dma_async_memcpy_issue_pending(chan);
+		dma_async_issue_pending(chan);
 
-		do {
-			msleep(1);
-			status = dma_async_memcpy_complete(
-					chan, cookie, NULL, NULL);
-		} while (status == DMA_IN_PROGRESS);
+		tmo = wait_for_completion_timeout(&cmp, tmo);
+		status = dma_async_is_tx_complete(chan, cookie, NULL, NULL);
 
-		if (status == DMA_ERROR) {
-			pr_warning("%s: #%u: error during copy\n",
-					thread_name, total_tests - 1);
+		if (tmo == 0) {
+			pr_warning("%s: #%u: test timed out\n",
+				   thread_name, total_tests - 1);
+			failed_tests++;
+			continue;
+		} else if (status != DMA_SUCCESS) {
+			pr_warning("%s: #%u: got completion callback,"
+				   " but status is \'%s\'\n",
+				   thread_name, total_tests - 1,
+				   status == DMA_ERROR ? "error" : "in progress");
 			failed_tests++;
 			continue;
 		}
+
 		/* Unmap by myself (see DMA_COMPL_SKIP_DEST_UNMAP above) */
-		dma_unmap_single(dev->dev, dma_dest,
-				test_buf_size, DMA_BIDIRECTIONAL);
+		for (i = 0; i < dst_cnt; i++)
+			dma_unmap_single(dev->dev, dma_dsts[i], test_buf_size,
+					 DMA_BIDIRECTIONAL);
 
 		error_count = 0;
 
 		pr_debug("%s: verifying source buffer...\n", thread_name);
-		error_count += dmatest_verify(thread->srcbuf, 0, src_off,
+		error_count += dmatest_verify(thread->srcs, 0, src_off,
 				0, PATTERN_SRC, true);
-		error_count += dmatest_verify(thread->srcbuf, src_off,
+		error_count += dmatest_verify(thread->srcs, src_off,
 				src_off + len, src_off,
 				PATTERN_SRC | PATTERN_COPY, true);
-		error_count += dmatest_verify(thread->srcbuf, src_off + len,
+		error_count += dmatest_verify(thread->srcs, src_off + len,
 				test_buf_size, src_off + len,
 				PATTERN_SRC, true);
 
 		pr_debug("%s: verifying dest buffer...\n",
 				thread->task->comm);
-		error_count += dmatest_verify(thread->dstbuf, 0, dst_off,
+		error_count += dmatest_verify(thread->dsts, 0, dst_off,
 				0, PATTERN_DST, false);
-		error_count += dmatest_verify(thread->dstbuf, dst_off,
+		error_count += dmatest_verify(thread->dsts, dst_off,
 				dst_off + len, src_off,
 				PATTERN_SRC | PATTERN_COPY, false);
-		error_count += dmatest_verify(thread->dstbuf, dst_off + len,
+		error_count += dmatest_verify(thread->dsts, dst_off + len,
 				test_buf_size, dst_off + len,
 				PATTERN_DST, false);
 
@@ -319,10 +404,16 @@ static int dmatest_func(void *data)
 	}
 
 	ret = 0;
-	kfree(thread->dstbuf);
+	for (i = 0; thread->dsts[i]; i++)
+		kfree(thread->dsts[i]);
 err_dstbuf:
-	kfree(thread->srcbuf);
+	kfree(thread->dsts);
+err_dsts:
+	for (i = 0; thread->srcs[i]; i++)
+		kfree(thread->srcs[i]);
 err_srcbuf:
+	kfree(thread->srcs);
+err_srcs:
 	pr_notice("%s: terminating after %u tests, %u failures (status %d)\n",
 			thread_name, total_tests, failed_tests, ret);
 	return ret;
@@ -344,35 +435,36 @@ static void dmatest_cleanup_channel(struct dmatest_chan *dtc)
 	kfree(dtc);
 }
 
-static int dmatest_add_channel(struct dma_chan *chan)
+static int dmatest_add_threads(struct dmatest_chan *dtc, enum dma_transaction_type type)
 {
-	struct dmatest_chan	*dtc;
-	struct dmatest_thread	*thread;
-	unsigned int		i;
-
-	dtc = kmalloc(sizeof(struct dmatest_chan), GFP_KERNEL);
-	if (!dtc) {
-		pr_warning("dmatest: No memory for %s\n", dma_chan_name(chan));
-		return -ENOMEM;
-	}
+	struct dmatest_thread *thread;
+	struct dma_chan *chan = dtc->chan;
+	char *op;
+	unsigned int i;
 
-	dtc->chan = chan;
-	INIT_LIST_HEAD(&dtc->threads);
+	if (type == DMA_MEMCPY)
+		op = "copy";
+	else if (type == DMA_XOR)
+		op = "xor";
+	else
+		return -EINVAL;
 
 	for (i = 0; i < threads_per_chan; i++) {
 		thread = kzalloc(sizeof(struct dmatest_thread), GFP_KERNEL);
 		if (!thread) {
-			pr_warning("dmatest: No memory for %s-test%u\n",
-				   dma_chan_name(chan), i);
+			pr_warning("dmatest: No memory for %s-%s%u\n",
+				   dma_chan_name(chan), op, i);
+
 			break;
 		}
 		thread->chan = dtc->chan;
+		thread->type = type;
 		smp_wmb();
-		thread->task = kthread_run(dmatest_func, thread, "%s-test%u",
-				dma_chan_name(chan), i);
+		thread->task = kthread_run(dmatest_func, thread, "%s-%s%u",
+				dma_chan_name(chan), op, i);
 		if (IS_ERR(thread->task)) {
-			pr_warning("dmatest: Failed to run thread %s-test%u\n",
-					dma_chan_name(chan), i);
+			pr_warning("dmatest: Failed to run thread %s-%s%u\n",
+					dma_chan_name(chan), op, i);
 			kfree(thread);
 			break;
 		}
@@ -382,7 +474,36 @@ static int dmatest_add_channel(struct dma_chan *chan)
 		list_add_tail(&thread->node, &dtc->threads);
 	}
 
-	pr_info("dmatest: Started %u threads using %s\n", i, dma_chan_name(chan));
+	return i;
+}
+
+static int dmatest_add_channel(struct dma_chan *chan)
+{
+	struct dmatest_chan	*dtc;
+	struct dma_device	*dma_dev = chan->device;
+	unsigned int		thread_count = 0;
+	unsigned int		cnt;
+
+	dtc = kmalloc(sizeof(struct dmatest_chan), GFP_KERNEL);
+	if (!dtc) {
+		pr_warning("dmatest: No memory for %s\n", dma_chan_name(chan));
+		return -ENOMEM;
+	}
+
+	dtc->chan = chan;
+	INIT_LIST_HEAD(&dtc->threads);
+
+	if (dma_has_cap(DMA_MEMCPY, dma_dev->cap_mask)) {
+		cnt = dmatest_add_threads(dtc, DMA_MEMCPY);
+		thread_count += cnt > 0 ?: 0;
+	}
+	if (dma_has_cap(DMA_XOR, dma_dev->cap_mask)) {
+		cnt = dmatest_add_threads(dtc, DMA_XOR);
+		thread_count += cnt > 0 ?: 0;
+	}
+
+	pr_info("dmatest: Started %u threads using %s\n",
+		thread_count, dma_chan_name(chan));
 
 	list_add_tail(&dtc->node, &dmatest_channels);
 	nr_channels++;
diff --git a/drivers/dma/dw_dmac.c b/drivers/dma/dw_dmac.c
index 20ad3d26bec..98c9a847bf5 100644
--- a/drivers/dma/dw_dmac.c
+++ b/drivers/dma/dw_dmac.c
@@ -363,6 +363,82 @@ static void dwc_handle_error(struct dw_dma *dw, struct dw_dma_chan *dwc)
 	dwc_descriptor_complete(dwc, bad_desc);
 }
 
+/* --------------------- Cyclic DMA API extensions -------------------- */
+
+inline dma_addr_t dw_dma_get_src_addr(struct dma_chan *chan)
+{
+	struct dw_dma_chan *dwc = to_dw_dma_chan(chan);
+	return channel_readl(dwc, SAR);
+}
+EXPORT_SYMBOL(dw_dma_get_src_addr);
+
+inline dma_addr_t dw_dma_get_dst_addr(struct dma_chan *chan)
+{
+	struct dw_dma_chan *dwc = to_dw_dma_chan(chan);
+	return channel_readl(dwc, DAR);
+}
+EXPORT_SYMBOL(dw_dma_get_dst_addr);
+
+/* called with dwc->lock held and all DMAC interrupts disabled */
+static void dwc_handle_cyclic(struct dw_dma *dw, struct dw_dma_chan *dwc,
+		u32 status_block, u32 status_err, u32 status_xfer)
+{
+	if (status_block & dwc->mask) {
+		void (*callback)(void *param);
+		void *callback_param;
+
+		dev_vdbg(chan2dev(&dwc->chan), "new cyclic period llp 0x%08x\n",
+				channel_readl(dwc, LLP));
+		dma_writel(dw, CLEAR.BLOCK, dwc->mask);
+
+		callback = dwc->cdesc->period_callback;
+		callback_param = dwc->cdesc->period_callback_param;
+		if (callback) {
+			spin_unlock(&dwc->lock);
+			callback(callback_param);
+			spin_lock(&dwc->lock);
+		}
+	}
+
+	/*
+	 * Error and transfer complete are highly unlikely, and will most
+	 * likely be due to a configuration error by the user.
+	 */
+	if (unlikely(status_err & dwc->mask) ||
+			unlikely(status_xfer & dwc->mask)) {
+		int i;
+
+		dev_err(chan2dev(&dwc->chan), "cyclic DMA unexpected %s "
+				"interrupt, stopping DMA transfer\n",
+				status_xfer ? "xfer" : "error");
+		dev_err(chan2dev(&dwc->chan),
+			"  SAR: 0x%x DAR: 0x%x LLP: 0x%x CTL: 0x%x:%08x\n",
+			channel_readl(dwc, SAR),
+			channel_readl(dwc, DAR),
+			channel_readl(dwc, LLP),
+			channel_readl(dwc, CTL_HI),
+			channel_readl(dwc, CTL_LO));
+
+		channel_clear_bit(dw, CH_EN, dwc->mask);
+		while (dma_readl(dw, CH_EN) & dwc->mask)
+			cpu_relax();
+
+		/* make sure DMA does not restart by loading a new list */
+		channel_writel(dwc, LLP, 0);
+		channel_writel(dwc, CTL_LO, 0);
+		channel_writel(dwc, CTL_HI, 0);
+
+		dma_writel(dw, CLEAR.BLOCK, dwc->mask);
+		dma_writel(dw, CLEAR.ERROR, dwc->mask);
+		dma_writel(dw, CLEAR.XFER, dwc->mask);
+
+		for (i = 0; i < dwc->cdesc->periods; i++)
+			dwc_dump_lli(dwc, &dwc->cdesc->desc[i]->lli);
+	}
+}
+
+/* ------------------------------------------------------------------------- */
+
 static void dw_dma_tasklet(unsigned long data)
 {
 	struct dw_dma *dw = (struct dw_dma *)data;
@@ -382,7 +458,10 @@ static void dw_dma_tasklet(unsigned long data)
 	for (i = 0; i < dw->dma.chancnt; i++) {
 		dwc = &dw->chan[i];
 		spin_lock(&dwc->lock);
-		if (status_err & (1 << i))
+		if (test_bit(DW_DMA_IS_CYCLIC, &dwc->flags))
+			dwc_handle_cyclic(dw, dwc, status_block, status_err,
+					status_xfer);
+		else if (status_err & (1 << i))
 			dwc_handle_error(dw, dwc);
 		else if ((status_block | status_xfer) & (1 << i))
 			dwc_scan_descriptors(dw, dwc);
@@ -826,7 +905,6 @@ static int dwc_alloc_chan_resources(struct dma_chan *chan)
 		dma_async_tx_descriptor_init(&desc->txd, chan);
 		desc->txd.tx_submit = dwc_tx_submit;
 		desc->txd.flags = DMA_CTRL_ACK;
-		INIT_LIST_HEAD(&desc->txd.tx_list);
 		desc->txd.phys = dma_map_single(chan2parent(chan), &desc->lli,
 				sizeof(desc->lli), DMA_TO_DEVICE);
 		dwc_desc_put(dwc, desc);
@@ -884,6 +962,257 @@ static void dwc_free_chan_resources(struct dma_chan *chan)
 	dev_vdbg(chan2dev(chan), "free_chan_resources done\n");
 }
 
+/* --------------------- Cyclic DMA API extensions -------------------- */
+
+/**
+ * dw_dma_cyclic_start - start the cyclic DMA transfer
+ * @chan: the DMA channel to start
+ *
+ * Must be called with soft interrupts disabled. Returns zero on success or
+ * -errno on failure.
+ */
+int dw_dma_cyclic_start(struct dma_chan *chan)
+{
+	struct dw_dma_chan	*dwc = to_dw_dma_chan(chan);
+	struct dw_dma		*dw = to_dw_dma(dwc->chan.device);
+
+	if (!test_bit(DW_DMA_IS_CYCLIC, &dwc->flags)) {
+		dev_err(chan2dev(&dwc->chan), "missing prep for cyclic DMA\n");
+		return -ENODEV;
+	}
+
+	spin_lock(&dwc->lock);
+
+	/* assert channel is idle */
+	if (dma_readl(dw, CH_EN) & dwc->mask) {
+		dev_err(chan2dev(&dwc->chan),
+			"BUG: Attempted to start non-idle channel\n");
+		dev_err(chan2dev(&dwc->chan),
+			"  SAR: 0x%x DAR: 0x%x LLP: 0x%x CTL: 0x%x:%08x\n",
+			channel_readl(dwc, SAR),
+			channel_readl(dwc, DAR),
+			channel_readl(dwc, LLP),
+			channel_readl(dwc, CTL_HI),
+			channel_readl(dwc, CTL_LO));
+		spin_unlock(&dwc->lock);
+		return -EBUSY;
+	}
+
+	dma_writel(dw, CLEAR.BLOCK, dwc->mask);
+	dma_writel(dw, CLEAR.ERROR, dwc->mask);
+	dma_writel(dw, CLEAR.XFER, dwc->mask);
+
+	/* setup DMAC channel registers */
+	channel_writel(dwc, LLP, dwc->cdesc->desc[0]->txd.phys);
+	channel_writel(dwc, CTL_LO, DWC_CTLL_LLP_D_EN | DWC_CTLL_LLP_S_EN);
+	channel_writel(dwc, CTL_HI, 0);
+
+	channel_set_bit(dw, CH_EN, dwc->mask);
+
+	spin_unlock(&dwc->lock);
+
+	return 0;
+}
+EXPORT_SYMBOL(dw_dma_cyclic_start);
+
+/**
+ * dw_dma_cyclic_stop - stop the cyclic DMA transfer
+ * @chan: the DMA channel to stop
+ *
+ * Must be called with soft interrupts disabled.
+ */
+void dw_dma_cyclic_stop(struct dma_chan *chan)
+{
+	struct dw_dma_chan	*dwc = to_dw_dma_chan(chan);
+	struct dw_dma		*dw = to_dw_dma(dwc->chan.device);
+
+	spin_lock(&dwc->lock);
+
+	channel_clear_bit(dw, CH_EN, dwc->mask);
+	while (dma_readl(dw, CH_EN) & dwc->mask)
+		cpu_relax();
+
+	spin_unlock(&dwc->lock);
+}
+EXPORT_SYMBOL(dw_dma_cyclic_stop);
+
+/**
+ * dw_dma_cyclic_prep - prepare the cyclic DMA transfer
+ * @chan: the DMA channel to prepare
+ * @buf_addr: physical DMA address where the buffer starts
+ * @buf_len: total number of bytes for the entire buffer
+ * @period_len: number of bytes for each period
+ * @direction: transfer direction, to or from device
+ *
+ * Must be called before trying to start the transfer. Returns a valid struct
+ * dw_cyclic_desc if successful or an ERR_PTR(-errno) if not successful.
+ */
+struct dw_cyclic_desc *dw_dma_cyclic_prep(struct dma_chan *chan,
+		dma_addr_t buf_addr, size_t buf_len, size_t period_len,
+		enum dma_data_direction direction)
+{
+	struct dw_dma_chan		*dwc = to_dw_dma_chan(chan);
+	struct dw_cyclic_desc		*cdesc;
+	struct dw_cyclic_desc		*retval = NULL;
+	struct dw_desc			*desc;
+	struct dw_desc			*last = NULL;
+	struct dw_dma_slave		*dws = chan->private;
+	unsigned long			was_cyclic;
+	unsigned int			reg_width;
+	unsigned int			periods;
+	unsigned int			i;
+
+	spin_lock_bh(&dwc->lock);
+	if (!list_empty(&dwc->queue) || !list_empty(&dwc->active_list)) {
+		spin_unlock_bh(&dwc->lock);
+		dev_dbg(chan2dev(&dwc->chan),
+				"queue and/or active list are not empty\n");
+		return ERR_PTR(-EBUSY);
+	}
+
+	was_cyclic = test_and_set_bit(DW_DMA_IS_CYCLIC, &dwc->flags);
+	spin_unlock_bh(&dwc->lock);
+	if (was_cyclic) {
+		dev_dbg(chan2dev(&dwc->chan),
+				"channel already prepared for cyclic DMA\n");
+		return ERR_PTR(-EBUSY);
+	}
+
+	retval = ERR_PTR(-EINVAL);
+	reg_width = dws->reg_width;
+	periods = buf_len / period_len;
+
+	/* Check for too big/unaligned periods and unaligned DMA buffer. */
+	if (period_len > (DWC_MAX_COUNT << reg_width))
+		goto out_err;
+	if (unlikely(period_len & ((1 << reg_width) - 1)))
+		goto out_err;
+	if (unlikely(buf_addr & ((1 << reg_width) - 1)))
+		goto out_err;
+	if (unlikely(!(direction & (DMA_TO_DEVICE | DMA_FROM_DEVICE))))
+		goto out_err;
+
+	retval = ERR_PTR(-ENOMEM);
+
+	if (periods > NR_DESCS_PER_CHANNEL)
+		goto out_err;
+
+	cdesc = kzalloc(sizeof(struct dw_cyclic_desc), GFP_KERNEL);
+	if (!cdesc)
+		goto out_err;
+
+	cdesc->desc = kzalloc(sizeof(struct dw_desc *) * periods, GFP_KERNEL);
+	if (!cdesc->desc)
+		goto out_err_alloc;
+
+	for (i = 0; i < periods; i++) {
+		desc = dwc_desc_get(dwc);
+		if (!desc)
+			goto out_err_desc_get;
+
+		switch (direction) {
+		case DMA_TO_DEVICE:
+			desc->lli.dar = dws->tx_reg;
+			desc->lli.sar = buf_addr + (period_len * i);
+			desc->lli.ctllo = (DWC_DEFAULT_CTLLO
+					| DWC_CTLL_DST_WIDTH(reg_width)
+					| DWC_CTLL_SRC_WIDTH(reg_width)
+					| DWC_CTLL_DST_FIX
+					| DWC_CTLL_SRC_INC
+					| DWC_CTLL_FC_M2P
+					| DWC_CTLL_INT_EN);
+			break;
+		case DMA_FROM_DEVICE:
+			desc->lli.dar = buf_addr + (period_len * i);
+			desc->lli.sar = dws->rx_reg;
+			desc->lli.ctllo = (DWC_DEFAULT_CTLLO
+					| DWC_CTLL_SRC_WIDTH(reg_width)
+					| DWC_CTLL_DST_WIDTH(reg_width)
+					| DWC_CTLL_DST_INC
+					| DWC_CTLL_SRC_FIX
+					| DWC_CTLL_FC_P2M
+					| DWC_CTLL_INT_EN);
+			break;
+		default:
+			break;
+		}
+
+		desc->lli.ctlhi = (period_len >> reg_width);
+		cdesc->desc[i] = desc;
+
+		if (last) {
+			last->lli.llp = desc->txd.phys;
+			dma_sync_single_for_device(chan2parent(chan),
+					last->txd.phys, sizeof(last->lli),
+					DMA_TO_DEVICE);
+		}
+
+		last = desc;
+	}
+
+	/* lets make a cyclic list */
+	last->lli.llp = cdesc->desc[0]->txd.phys;
+	dma_sync_single_for_device(chan2parent(chan), last->txd.phys,
+			sizeof(last->lli), DMA_TO_DEVICE);
+
+	dev_dbg(chan2dev(&dwc->chan), "cyclic prepared buf 0x%08x len %zu "
+			"period %zu periods %d\n", buf_addr, buf_len,
+			period_len, periods);
+
+	cdesc->periods = periods;
+	dwc->cdesc = cdesc;
+
+	return cdesc;
+
+out_err_desc_get:
+	while (i--)
+		dwc_desc_put(dwc, cdesc->desc[i]);
+out_err_alloc:
+	kfree(cdesc);
+out_err:
+	clear_bit(DW_DMA_IS_CYCLIC, &dwc->flags);
+	return (struct dw_cyclic_desc *)retval;
+}
+EXPORT_SYMBOL(dw_dma_cyclic_prep);
+
+/**
+ * dw_dma_cyclic_free - free a prepared cyclic DMA transfer
+ * @chan: the DMA channel to free
+ */
+void dw_dma_cyclic_free(struct dma_chan *chan)
+{
+	struct dw_dma_chan	*dwc = to_dw_dma_chan(chan);
+	struct dw_dma		*dw = to_dw_dma(dwc->chan.device);
+	struct dw_cyclic_desc	*cdesc = dwc->cdesc;
+	int			i;
+
+	dev_dbg(chan2dev(&dwc->chan), "cyclic free\n");
+
+	if (!cdesc)
+		return;
+
+	spin_lock_bh(&dwc->lock);
+
+	channel_clear_bit(dw, CH_EN, dwc->mask);
+	while (dma_readl(dw, CH_EN) & dwc->mask)
+		cpu_relax();
+
+	dma_writel(dw, CLEAR.BLOCK, dwc->mask);
+	dma_writel(dw, CLEAR.ERROR, dwc->mask);
+	dma_writel(dw, CLEAR.XFER, dwc->mask);
+
+	spin_unlock_bh(&dwc->lock);
+
+	for (i = 0; i < cdesc->periods; i++)
+		dwc_desc_put(dwc, cdesc->desc[i]);
+
+	kfree(cdesc->desc);
+	kfree(cdesc);
+
+	clear_bit(DW_DMA_IS_CYCLIC, &dwc->flags);
+}
+EXPORT_SYMBOL(dw_dma_cyclic_free);
+
 /*----------------------------------------------------------------------*/
 
 static void dw_dma_off(struct dw_dma *dw)
diff --git a/drivers/dma/dw_dmac_regs.h b/drivers/dma/dw_dmac_regs.h
index b252b202c5c..13a58076703 100644
--- a/drivers/dma/dw_dmac_regs.h
+++ b/drivers/dma/dw_dmac_regs.h
@@ -126,6 +126,10 @@ struct dw_dma_regs {
 
 #define DW_REGLEN		0x400
 
+enum dw_dmac_flags {
+	DW_DMA_IS_CYCLIC = 0,
+};
+
 struct dw_dma_chan {
 	struct dma_chan		chan;
 	void __iomem		*ch_regs;
@@ -134,10 +138,12 @@ struct dw_dma_chan {
 	spinlock_t		lock;
 
 	/* these other elements are all protected by lock */
+	unsigned long		flags;
 	dma_cookie_t		completed;
 	struct list_head	active_list;
 	struct list_head	queue;
 	struct list_head	free_list;
+	struct dw_cyclic_desc	*cdesc;
 
 	unsigned int		descs_allocated;
 };
@@ -158,7 +164,6 @@ static inline struct dw_dma_chan *to_dw_dma_chan(struct dma_chan *chan)
 	return container_of(chan, struct dw_dma_chan, chan);
 }
 
-
 struct dw_dma {
 	struct dma_device	dma;
 	void __iomem		*regs;
diff --git a/drivers/dma/fsldma.c b/drivers/dma/fsldma.c
index 86d6da47f55..da8a8ed9e41 100644
--- a/drivers/dma/fsldma.c
+++ b/drivers/dma/fsldma.c
@@ -354,7 +354,6 @@ static struct fsl_desc_sw *fsl_dma_alloc_descriptor(
 		dma_async_tx_descriptor_init(&desc_sw->async_tx,
 						&fsl_chan->common);
 		desc_sw->async_tx.tx_submit = fsl_dma_tx_submit;
-		INIT_LIST_HEAD(&desc_sw->async_tx.tx_list);
 		desc_sw->async_tx.phys = pdesc;
 	}
 
diff --git a/drivers/dma/ioat_dma.c b/drivers/dma/ioat_dma.c
index 5905cd36bcd..e4fc33c1c32 100644
--- a/drivers/dma/ioat_dma.c
+++ b/drivers/dma/ioat_dma.c
@@ -693,7 +693,6 @@ static struct ioat_desc_sw *ioat_dma_alloc_descriptor(
 		desc_sw->async_tx.tx_submit = ioat2_tx_submit;
 		break;
 	}
-	INIT_LIST_HEAD(&desc_sw->async_tx.tx_list);
 
 	desc_sw->hw = desc;
 	desc_sw->async_tx.phys = phys;
diff --git a/drivers/dma/iop-adma.c b/drivers/dma/iop-adma.c
index 16adbe61cfb..2f052265122 100644
--- a/drivers/dma/iop-adma.c
+++ b/drivers/dma/iop-adma.c
@@ -498,7 +498,6 @@ static int iop_adma_alloc_chan_resources(struct dma_chan *chan)
 		slot->async_tx.tx_submit = iop_adma_tx_submit;
 		INIT_LIST_HEAD(&slot->chain_node);
 		INIT_LIST_HEAD(&slot->slot_node);
-		INIT_LIST_HEAD(&slot->async_tx.tx_list);
 		hw_desc = (char *) iop_chan->device->dma_desc_pool;
 		slot->async_tx.phys =
 			(dma_addr_t) &hw_desc[idx * IOP_ADMA_SLOT_SIZE];
diff --git a/drivers/dma/ipu/ipu_idmac.c b/drivers/dma/ipu/ipu_idmac.c
index da781d10789..e202a6ce557 100644
--- a/drivers/dma/ipu/ipu_idmac.c
+++ b/drivers/dma/ipu/ipu_idmac.c
@@ -28,6 +28,9 @@
 #define FS_VF_IN_VALID	0x00000002
 #define FS_ENC_IN_VALID	0x00000001
 
+static int ipu_disable_channel(struct idmac *idmac, struct idmac_channel *ichan,
+			       bool wait_for_stop);
+
 /*
  * There can be only one, we could allocate it dynamically, but then we'd have
  * to add an extra parameter to some functions, and use something as ugly as
@@ -107,7 +110,7 @@ static uint32_t bytes_per_pixel(enum pixel_fmt fmt)
 	}
 }
 
-/* Enable / disable direct write to memory by the Camera Sensor Interface */
+/* Enable direct write to memory by the Camera Sensor Interface */
 static void ipu_ic_enable_task(struct ipu *ipu, enum ipu_channel channel)
 {
 	uint32_t ic_conf, mask;
@@ -126,6 +129,7 @@ static void ipu_ic_enable_task(struct ipu *ipu, enum ipu_channel channel)
 	idmac_write_icreg(ipu, ic_conf, IC_CONF);
 }
 
+/* Called under spin_lock_irqsave(&ipu_data.lock) */
 static void ipu_ic_disable_task(struct ipu *ipu, enum ipu_channel channel)
 {
 	uint32_t ic_conf, mask;
@@ -422,7 +426,7 @@ static void ipu_ch_param_set_size(union chan_param_mem *params,
 		break;
 	default:
 		dev_err(ipu_data.dev,
-			"mxc ipu: unimplemented pixel format %d\n", pixel_fmt);
+			"mx3 ipu: unimplemented pixel format %d\n", pixel_fmt);
 		break;
 	}
 
@@ -433,20 +437,20 @@ static void ipu_ch_param_set_burst_size(union chan_param_mem *params,
 					uint16_t burst_pixels)
 {
 	params->pp.npb = burst_pixels - 1;
-};
+}
 
 static void ipu_ch_param_set_buffer(union chan_param_mem *params,
 				    dma_addr_t buf0, dma_addr_t buf1)
 {
 	params->pp.eba0 = buf0;
 	params->pp.eba1 = buf1;
-};
+}
 
 static void ipu_ch_param_set_rotation(union chan_param_mem *params,
 				      enum ipu_rotate_mode rotate)
 {
 	params->pp.bam = rotate;
-};
+}
 
 static void ipu_write_param_mem(uint32_t addr, uint32_t *data,
 				uint32_t num_words)
@@ -571,7 +575,7 @@ static uint32_t dma_param_addr(uint32_t dma_ch)
 {
 	/* Channel Parameter Memory */
 	return 0x10000 | (dma_ch << 4);
-};
+}
 
 static void ipu_channel_set_priority(struct ipu *ipu, enum ipu_channel channel,
 				     bool prio)
@@ -611,7 +615,8 @@ static uint32_t ipu_channel_conf_mask(enum ipu_channel channel)
 
 /**
  * ipu_enable_channel() - enable an IPU channel.
- * @channel:	channel ID.
+ * @idmac:	IPU DMAC context.
+ * @ichan:	IDMAC channel.
  * @return:	0 on success or negative error code on failure.
  */
 static int ipu_enable_channel(struct idmac *idmac, struct idmac_channel *ichan)
@@ -649,7 +654,7 @@ static int ipu_enable_channel(struct idmac *idmac, struct idmac_channel *ichan)
 
 /**
  * ipu_init_channel_buffer() - initialize a buffer for logical IPU channel.
- * @channel:	channel ID.
+ * @ichan:	IDMAC channel.
  * @pixel_fmt:	pixel format of buffer. Pixel format is a FOURCC ASCII code.
  * @width:	width of buffer in pixels.
  * @height:	height of buffer in pixels.
@@ -687,7 +692,7 @@ static int ipu_init_channel_buffer(struct idmac_channel *ichan,
 	}
 
 	/* IC channel's stride must be a multiple of 8 pixels */
-	if ((channel <= 13) && (stride % 8)) {
+	if ((channel <= IDMAC_IC_13) && (stride % 8)) {
 		dev_err(ipu->dev, "Stride must be 8 pixel multiple\n");
 		return -EINVAL;
 	}
@@ -752,7 +757,7 @@ static void ipu_select_buffer(enum ipu_channel channel, int buffer_n)
 
 /**
  * ipu_update_channel_buffer() - update physical address of a channel buffer.
- * @channel:	channel ID.
+ * @ichan:	IDMAC channel.
  * @buffer_n:	buffer number to update.
  *		0 or 1 are the only valid values.
  * @phyaddr:	buffer physical address.
@@ -760,9 +765,10 @@ static void ipu_select_buffer(enum ipu_channel channel, int buffer_n)
  *              function will fail if the buffer is set to ready.
  */
 /* Called under spin_lock(_irqsave)(&ichan->lock) */
-static int ipu_update_channel_buffer(enum ipu_channel channel,
+static int ipu_update_channel_buffer(struct idmac_channel *ichan,
 				     int buffer_n, dma_addr_t phyaddr)
 {
+	enum ipu_channel channel = ichan->dma_chan.chan_id;
 	uint32_t reg;
 	unsigned long flags;
 
@@ -771,8 +777,8 @@ static int ipu_update_channel_buffer(enum ipu_channel channel,
 	if (buffer_n == 0) {
 		reg = idmac_read_ipureg(&ipu_data, IPU_CHA_BUF0_RDY);
 		if (reg & (1UL << channel)) {
-			spin_unlock_irqrestore(&ipu_data.lock, flags);
-			return -EACCES;
+			ipu_ic_disable_task(&ipu_data, channel);
+			ichan->status = IPU_CHANNEL_READY;
 		}
 
 		/* 44.3.3.1.9 - Row Number 1 (WORD1, offset 0) */
@@ -782,8 +788,8 @@ static int ipu_update_channel_buffer(enum ipu_channel channel,
 	} else {
 		reg = idmac_read_ipureg(&ipu_data, IPU_CHA_BUF1_RDY);
 		if (reg & (1UL << channel)) {
-			spin_unlock_irqrestore(&ipu_data.lock, flags);
-			return -EACCES;
+			ipu_ic_disable_task(&ipu_data, channel);
+			ichan->status = IPU_CHANNEL_READY;
 		}
 
 		/* Check if double-buffering is already enabled */
@@ -805,6 +811,39 @@ static int ipu_update_channel_buffer(enum ipu_channel channel,
 }
 
 /* Called under spin_lock_irqsave(&ichan->lock) */
+static int ipu_submit_buffer(struct idmac_channel *ichan,
+	struct idmac_tx_desc *desc, struct scatterlist *sg, int buf_idx)
+{
+	unsigned int chan_id = ichan->dma_chan.chan_id;
+	struct device *dev = &ichan->dma_chan.dev->device;
+	int ret;
+
+	if (async_tx_test_ack(&desc->txd))
+		return -EINTR;
+
+	/*
+	 * On first invocation this shouldn't be necessary, the call to
+	 * ipu_init_channel_buffer() above will set addresses for us, so we
+	 * could make it conditional on status >= IPU_CHANNEL_ENABLED, but
+	 * doing it again shouldn't hurt either.
+	 */
+	ret = ipu_update_channel_buffer(ichan, buf_idx,
+					sg_dma_address(sg));
+
+	if (ret < 0) {
+		dev_err(dev, "Updating sg %p on channel 0x%x buffer %d failed!\n",
+			sg, chan_id, buf_idx);
+		return ret;
+	}
+
+	ipu_select_buffer(chan_id, buf_idx);
+	dev_dbg(dev, "Updated sg %p on channel 0x%x buffer %d\n",
+		sg, chan_id, buf_idx);
+
+	return 0;
+}
+
+/* Called under spin_lock_irqsave(&ichan->lock) */
 static int ipu_submit_channel_buffers(struct idmac_channel *ichan,
 				      struct idmac_tx_desc *desc)
 {
@@ -815,20 +854,10 @@ static int ipu_submit_channel_buffers(struct idmac_channel *ichan,
 		if (!ichan->sg[i]) {
 			ichan->sg[i] = sg;
 
-			/*
-			 * On first invocation this shouldn't be necessary, the
-			 * call to ipu_init_channel_buffer() above will set
-			 * addresses for us, so we could make it conditional
-			 * on status >= IPU_CHANNEL_ENABLED, but doing it again
-			 * shouldn't hurt either.
-			 */
-			ret = ipu_update_channel_buffer(ichan->dma_chan.chan_id, i,
-							sg_dma_address(sg));
+			ret = ipu_submit_buffer(ichan, desc, sg, i);
 			if (ret < 0)
 				return ret;
 
-			ipu_select_buffer(ichan->dma_chan.chan_id, i);
-
 			sg = sg_next(sg);
 		}
 	}
@@ -842,19 +871,22 @@ static dma_cookie_t idmac_tx_submit(struct dma_async_tx_descriptor *tx)
 	struct idmac_channel *ichan = to_idmac_chan(tx->chan);
 	struct idmac *idmac = to_idmac(tx->chan->device);
 	struct ipu *ipu = to_ipu(idmac);
+	struct device *dev = &ichan->dma_chan.dev->device;
 	dma_cookie_t cookie;
 	unsigned long flags;
+	int ret;
 
 	/* Sanity check */
 	if (!list_empty(&desc->list)) {
 		/* The descriptor doesn't belong to client */
-		dev_err(&ichan->dma_chan.dev->device,
-			"Descriptor %p not prepared!\n", tx);
+		dev_err(dev, "Descriptor %p not prepared!\n", tx);
 		return -EBUSY;
 	}
 
 	mutex_lock(&ichan->chan_mutex);
 
+	async_tx_clear_ack(tx);
+
 	if (ichan->status < IPU_CHANNEL_READY) {
 		struct idmac_video_param *video = &ichan->params.video;
 		/*
@@ -878,16 +910,7 @@ static dma_cookie_t idmac_tx_submit(struct dma_async_tx_descriptor *tx)
 			goto out;
 	}
 
-	/* ipu->lock can be taken under ichan->lock, but not v.v. */
-	spin_lock_irqsave(&ichan->lock, flags);
-
-	/* submit_buffers() atomically verifies and fills empty sg slots */
-	cookie = ipu_submit_channel_buffers(ichan, desc);
-
-	spin_unlock_irqrestore(&ichan->lock, flags);
-
-	if (cookie < 0)
-		goto out;
+	dev_dbg(dev, "Submitting sg %p\n", &desc->sg[0]);
 
 	cookie = ichan->dma_chan.cookie;
 
@@ -897,24 +920,40 @@ static dma_cookie_t idmac_tx_submit(struct dma_async_tx_descriptor *tx)
 	/* from dmaengine.h: "last cookie value returned to client" */
 	ichan->dma_chan.cookie = cookie;
 	tx->cookie = cookie;
+
+	/* ipu->lock can be taken under ichan->lock, but not v.v. */
 	spin_lock_irqsave(&ichan->lock, flags);
+
 	list_add_tail(&desc->list, &ichan->queue);
+	/* submit_buffers() atomically verifies and fills empty sg slots */
+	ret = ipu_submit_channel_buffers(ichan, desc);
+
 	spin_unlock_irqrestore(&ichan->lock, flags);
 
+	if (ret < 0) {
+		cookie = ret;
+		goto dequeue;
+	}
+
 	if (ichan->status < IPU_CHANNEL_ENABLED) {
-		int ret = ipu_enable_channel(idmac, ichan);
+		ret = ipu_enable_channel(idmac, ichan);
 		if (ret < 0) {
 			cookie = ret;
-			spin_lock_irqsave(&ichan->lock, flags);
-			list_del_init(&desc->list);
-			spin_unlock_irqrestore(&ichan->lock, flags);
-			tx->cookie = cookie;
-			ichan->dma_chan.cookie = cookie;
+			goto dequeue;
 		}
 	}
 
 	dump_idmac_reg(ipu);
 
+dequeue:
+	if (cookie < 0) {
+		spin_lock_irqsave(&ichan->lock, flags);
+		list_del_init(&desc->list);
+		spin_unlock_irqrestore(&ichan->lock, flags);
+		tx->cookie = cookie;
+		ichan->dma_chan.cookie = cookie;
+	}
+
 out:
 	mutex_unlock(&ichan->chan_mutex);
 
@@ -944,8 +983,6 @@ static int idmac_desc_alloc(struct idmac_channel *ichan, int n)
 		memset(txd, 0, sizeof(*txd));
 		dma_async_tx_descriptor_init(txd, &ichan->dma_chan);
 		txd->tx_submit		= idmac_tx_submit;
-		txd->chan		= &ichan->dma_chan;
-		INIT_LIST_HEAD(&txd->tx_list);
 
 		list_add(&desc->list, &ichan->free_list);
 
@@ -1161,6 +1198,24 @@ static int ipu_disable_channel(struct idmac *idmac, struct idmac_channel *ichan,
 	return 0;
 }
 
+static struct scatterlist *idmac_sg_next(struct idmac_channel *ichan,
+	struct idmac_tx_desc **desc, struct scatterlist *sg)
+{
+	struct scatterlist *sgnew = sg ? sg_next(sg) : NULL;
+
+	if (sgnew)
+		/* next sg-element in this list */
+		return sgnew;
+
+	if ((*desc)->list.next == &ichan->queue)
+		/* No more descriptors on the queue */
+		return NULL;
+
+	/* Fetch next descriptor */
+	*desc = list_entry((*desc)->list.next, struct idmac_tx_desc, list);
+	return (*desc)->sg;
+}
+
 /*
  * We have several possibilities here:
  * current BUF		next BUF
@@ -1176,23 +1231,46 @@ static int ipu_disable_channel(struct idmac *idmac, struct idmac_channel *ichan,
 static irqreturn_t idmac_interrupt(int irq, void *dev_id)
 {
 	struct idmac_channel *ichan = dev_id;
+	struct device *dev = &ichan->dma_chan.dev->device;
 	unsigned int chan_id = ichan->dma_chan.chan_id;
 	struct scatterlist **sg, *sgnext, *sgnew = NULL;
 	/* Next transfer descriptor */
-	struct idmac_tx_desc *desc = NULL, *descnew;
+	struct idmac_tx_desc *desc, *descnew;
 	dma_async_tx_callback callback;
 	void *callback_param;
 	bool done = false;
-	u32	ready0 = idmac_read_ipureg(&ipu_data, IPU_CHA_BUF0_RDY),
-		ready1 = idmac_read_ipureg(&ipu_data, IPU_CHA_BUF1_RDY),
-		curbuf = idmac_read_ipureg(&ipu_data, IPU_CHA_CUR_BUF);
+	u32 ready0, ready1, curbuf, err;
+	unsigned long flags;
 
 	/* IDMAC has cleared the respective BUFx_RDY bit, we manage the buffer */
 
-	pr_debug("IDMAC irq %d\n", irq);
+	dev_dbg(dev, "IDMAC irq %d, buf %d\n", irq, ichan->active_buffer);
+
+	spin_lock_irqsave(&ipu_data.lock, flags);
+
+	ready0	= idmac_read_ipureg(&ipu_data, IPU_CHA_BUF0_RDY);
+	ready1	= idmac_read_ipureg(&ipu_data, IPU_CHA_BUF1_RDY);
+	curbuf	= idmac_read_ipureg(&ipu_data, IPU_CHA_CUR_BUF);
+	err	= idmac_read_ipureg(&ipu_data, IPU_INT_STAT_4);
+
+	if (err & (1 << chan_id)) {
+		idmac_write_ipureg(&ipu_data, 1 << chan_id, IPU_INT_STAT_4);
+		spin_unlock_irqrestore(&ipu_data.lock, flags);
+		/*
+		 * Doing this
+		 * ichan->sg[0] = ichan->sg[1] = NULL;
+		 * you can force channel re-enable on the next tx_submit(), but
+		 * this is dirty - think about descriptors with multiple
+		 * sg elements.
+		 */
+		dev_warn(dev, "NFB4EOF on channel %d, ready %x, %x, cur %x\n",
+			 chan_id, ready0, ready1, curbuf);
+		return IRQ_HANDLED;
+	}
+	spin_unlock_irqrestore(&ipu_data.lock, flags);
+
 	/* Other interrupts do not interfere with this channel */
 	spin_lock(&ichan->lock);
-
 	if (unlikely(chan_id != IDMAC_SDC_0 && chan_id != IDMAC_SDC_1 &&
 		     ((curbuf >> chan_id) & 1) == ichan->active_buffer)) {
 		int i = 100;
@@ -1207,19 +1285,23 @@ static irqreturn_t idmac_interrupt(int irq, void *dev_id)
 
 		if (!i) {
 			spin_unlock(&ichan->lock);
-			dev_dbg(ichan->dma_chan.device->dev,
+			dev_dbg(dev,
 				"IRQ on active buffer on channel %x, active "
 				"%d, ready %x, %x, current %x!\n", chan_id,
 				ichan->active_buffer, ready0, ready1, curbuf);
 			return IRQ_NONE;
-		}
+		} else
+			dev_dbg(dev,
+				"Buffer deactivated on channel %x, active "
+				"%d, ready %x, %x, current %x, rest %d!\n", chan_id,
+				ichan->active_buffer, ready0, ready1, curbuf, i);
 	}
 
 	if (unlikely((ichan->active_buffer && (ready1 >> chan_id) & 1) ||
 		     (!ichan->active_buffer && (ready0 >> chan_id) & 1)
 		     )) {
 		spin_unlock(&ichan->lock);
-		dev_dbg(ichan->dma_chan.device->dev,
+		dev_dbg(dev,
 			"IRQ with active buffer still ready on channel %x, "
 			"active %d, ready %x, %x!\n", chan_id,
 			ichan->active_buffer, ready0, ready1);
@@ -1227,8 +1309,9 @@ static irqreturn_t idmac_interrupt(int irq, void *dev_id)
 	}
 
 	if (unlikely(list_empty(&ichan->queue))) {
+		ichan->sg[ichan->active_buffer] = NULL;
 		spin_unlock(&ichan->lock);
-		dev_err(ichan->dma_chan.device->dev,
+		dev_err(dev,
 			"IRQ without queued buffers on channel %x, active %d, "
 			"ready %x, %x!\n", chan_id,
 			ichan->active_buffer, ready0, ready1);
@@ -1243,40 +1326,44 @@ static irqreturn_t idmac_interrupt(int irq, void *dev_id)
 	sg = &ichan->sg[ichan->active_buffer];
 	sgnext = ichan->sg[!ichan->active_buffer];
 
+	if (!*sg) {
+		spin_unlock(&ichan->lock);
+		return IRQ_HANDLED;
+	}
+
+	desc = list_entry(ichan->queue.next, struct idmac_tx_desc, list);
+	descnew = desc;
+
+	dev_dbg(dev, "IDMAC irq %d, dma 0x%08x, next dma 0x%08x, current %d, curbuf 0x%08x\n",
+		irq, sg_dma_address(*sg), sgnext ? sg_dma_address(sgnext) : 0, ichan->active_buffer, curbuf);
+
+	/* Find the descriptor of sgnext */
+	sgnew = idmac_sg_next(ichan, &descnew, *sg);
+	if (sgnext != sgnew)
+		dev_err(dev, "Submitted buffer %p, next buffer %p\n", sgnext, sgnew);
+
 	/*
 	 * if sgnext == NULL sg must be the last element in a scatterlist and
 	 * queue must be empty
 	 */
 	if (unlikely(!sgnext)) {
-		if (unlikely(sg_next(*sg))) {
-			dev_err(ichan->dma_chan.device->dev,
-				"Broken buffer-update locking on channel %x!\n",
-				chan_id);
-			/* We'll let the user catch up */
+		if (!WARN_ON(sg_next(*sg)))
+			dev_dbg(dev, "Underrun on channel %x\n", chan_id);
+		ichan->sg[!ichan->active_buffer] = sgnew;
+
+		if (unlikely(sgnew)) {
+			ipu_submit_buffer(ichan, descnew, sgnew, !ichan->active_buffer);
 		} else {
-			/* Underrun */
+			spin_lock_irqsave(&ipu_data.lock, flags);
 			ipu_ic_disable_task(&ipu_data, chan_id);
-			dev_dbg(ichan->dma_chan.device->dev,
-				"Underrun on channel %x\n", chan_id);
+			spin_unlock_irqrestore(&ipu_data.lock, flags);
 			ichan->status = IPU_CHANNEL_READY;
 			/* Continue to check for complete descriptor */
 		}
 	}
 
-	desc = list_entry(ichan->queue.next, struct idmac_tx_desc, list);
-
-	/* First calculate and submit the next sg element */
-	if (likely(sgnext))
-		sgnew = sg_next(sgnext);
-
-	if (unlikely(!sgnew)) {
-		/* Start a new scatterlist, if any queued */
-		if (likely(desc->list.next != &ichan->queue)) {
-			descnew = list_entry(desc->list.next,
-					     struct idmac_tx_desc, list);
-			sgnew = &descnew->sg[0];
-		}
-	}
+	/* Calculate and submit the next sg element */
+	sgnew = idmac_sg_next(ichan, &descnew, sgnew);
 
 	if (unlikely(!sg_next(*sg)) || !sgnext) {
 		/*
@@ -1289,17 +1376,13 @@ static irqreturn_t idmac_interrupt(int irq, void *dev_id)
 
 	*sg = sgnew;
 
-	if (likely(sgnew)) {
-		int ret;
-
-		ret = ipu_update_channel_buffer(chan_id, ichan->active_buffer,
-						sg_dma_address(*sg));
-		if (ret < 0)
-			dev_err(ichan->dma_chan.device->dev,
-				"Failed to update buffer on channel %x buffer %d!\n",
-				chan_id, ichan->active_buffer);
-		else
-			ipu_select_buffer(chan_id, ichan->active_buffer);
+	if (likely(sgnew) &&
+	    ipu_submit_buffer(ichan, descnew, sgnew, ichan->active_buffer) < 0) {
+		callback = desc->txd.callback;
+		callback_param = desc->txd.callback_param;
+		spin_unlock(&ichan->lock);
+		callback(callback_param);
+		spin_lock(&ichan->lock);
 	}
 
 	/* Flip the active buffer - even if update above failed */
@@ -1327,13 +1410,20 @@ static void ipu_gc_tasklet(unsigned long arg)
 		struct idmac_channel *ichan = ipu->channel + i;
 		struct idmac_tx_desc *desc;
 		unsigned long flags;
-		int j;
+		struct scatterlist *sg;
+		int j, k;
 
 		for (j = 0; j < ichan->n_tx_desc; j++) {
 			desc = ichan->desc + j;
 			spin_lock_irqsave(&ichan->lock, flags);
 			if (async_tx_test_ack(&desc->txd)) {
 				list_move(&desc->list, &ichan->free_list);
+				for_each_sg(desc->sg, sg, desc->sg_len, k) {
+					if (ichan->sg[0] == sg)
+						ichan->sg[0] = NULL;
+					else if (ichan->sg[1] == sg)
+						ichan->sg[1] = NULL;
+				}
 				async_tx_clear_ack(&desc->txd);
 			}
 			spin_unlock_irqrestore(&ichan->lock, flags);
@@ -1341,13 +1431,7 @@ static void ipu_gc_tasklet(unsigned long arg)
 	}
 }
 
-/*
- * At the time .device_alloc_chan_resources() method is called, we cannot know,
- * whether the client will accept the channel. Thus we must only check, if we
- * can satisfy client's request but the only real criterion to verify, whether
- * the client has accepted our offer is the client_count. That's why we have to
- * perform the rest of our allocation tasks on the first call to this function.
- */
+/* Allocate and initialise a transfer descriptor. */
 static struct dma_async_tx_descriptor *idmac_prep_slave_sg(struct dma_chan *chan,
 		struct scatterlist *sgl, unsigned int sg_len,
 		enum dma_data_direction direction, unsigned long tx_flags)
@@ -1358,8 +1442,8 @@ static struct dma_async_tx_descriptor *idmac_prep_slave_sg(struct dma_chan *chan
 	unsigned long flags;
 
 	/* We only can handle these three channels so far */
-	if (ichan->dma_chan.chan_id != IDMAC_SDC_0 && ichan->dma_chan.chan_id != IDMAC_SDC_1 &&
-	    ichan->dma_chan.chan_id != IDMAC_IC_7)
+	if (chan->chan_id != IDMAC_SDC_0 && chan->chan_id != IDMAC_SDC_1 &&
+	    chan->chan_id != IDMAC_IC_7)
 		return NULL;
 
 	if (direction != DMA_FROM_DEVICE && direction != DMA_TO_DEVICE) {
@@ -1400,7 +1484,7 @@ static void idmac_issue_pending(struct dma_chan *chan)
 
 	/* This is not always needed, but doesn't hurt either */
 	spin_lock_irqsave(&ipu->lock, flags);
-	ipu_select_buffer(ichan->dma_chan.chan_id, ichan->active_buffer);
+	ipu_select_buffer(chan->chan_id, ichan->active_buffer);
 	spin_unlock_irqrestore(&ipu->lock, flags);
 
 	/*
@@ -1432,8 +1516,7 @@ static void __idmac_terminate_all(struct dma_chan *chan)
 			struct idmac_tx_desc *desc = ichan->desc + i;
 			if (list_empty(&desc->list))
 				/* Descriptor was prepared, but not submitted */
-				list_add(&desc->list,
-					 &ichan->free_list);
+				list_add(&desc->list, &ichan->free_list);
 
 			async_tx_clear_ack(&desc->txd);
 		}
@@ -1458,6 +1541,28 @@ static void idmac_terminate_all(struct dma_chan *chan)
 	mutex_unlock(&ichan->chan_mutex);
 }
 
+#ifdef DEBUG
+static irqreturn_t ic_sof_irq(int irq, void *dev_id)
+{
+	struct idmac_channel *ichan = dev_id;
+	printk(KERN_DEBUG "Got SOF IRQ %d on Channel %d\n",
+	       irq, ichan->dma_chan.chan_id);
+	disable_irq(irq);
+	return IRQ_HANDLED;
+}
+
+static irqreturn_t ic_eof_irq(int irq, void *dev_id)
+{
+	struct idmac_channel *ichan = dev_id;
+	printk(KERN_DEBUG "Got EOF IRQ %d on Channel %d\n",
+	       irq, ichan->dma_chan.chan_id);
+	disable_irq(irq);
+	return IRQ_HANDLED;
+}
+
+static int ic_sof = -EINVAL, ic_eof = -EINVAL;
+#endif
+
 static int idmac_alloc_chan_resources(struct dma_chan *chan)
 {
 	struct idmac_channel *ichan = to_idmac_chan(chan);
@@ -1471,31 +1576,49 @@ static int idmac_alloc_chan_resources(struct dma_chan *chan)
 	chan->cookie		= 1;
 	ichan->completed	= -ENXIO;
 
-	ret = ipu_irq_map(ichan->dma_chan.chan_id);
+	ret = ipu_irq_map(chan->chan_id);
 	if (ret < 0)
 		goto eimap;
 
 	ichan->eof_irq = ret;
+
+	/*
+	 * Important to first disable the channel, because maybe someone
+	 * used it before us, e.g., the bootloader
+	 */
+	ipu_disable_channel(idmac, ichan, true);
+
+	ret = ipu_init_channel(idmac, ichan);
+	if (ret < 0)
+		goto eichan;
+
 	ret = request_irq(ichan->eof_irq, idmac_interrupt, 0,
 			  ichan->eof_name, ichan);
 	if (ret < 0)
 		goto erirq;
 
-	ret = ipu_init_channel(idmac, ichan);
-	if (ret < 0)
-		goto eichan;
+#ifdef DEBUG
+	if (chan->chan_id == IDMAC_IC_7) {
+		ic_sof = ipu_irq_map(69);
+		if (ic_sof > 0)
+			request_irq(ic_sof, ic_sof_irq, 0, "IC SOF", ichan);
+		ic_eof = ipu_irq_map(70);
+		if (ic_eof > 0)
+			request_irq(ic_eof, ic_eof_irq, 0, "IC EOF", ichan);
+	}
+#endif
 
 	ichan->status = IPU_CHANNEL_INITIALIZED;
 
-	dev_dbg(&ichan->dma_chan.dev->device, "Found channel 0x%x, irq %d\n",
-		ichan->dma_chan.chan_id, ichan->eof_irq);
+	dev_dbg(&chan->dev->device, "Found channel 0x%x, irq %d\n",
+		chan->chan_id, ichan->eof_irq);
 
 	return ret;
 
-eichan:
-	free_irq(ichan->eof_irq, ichan);
 erirq:
-	ipu_irq_unmap(ichan->dma_chan.chan_id);
+	ipu_uninit_channel(idmac, ichan);
+eichan:
+	ipu_irq_unmap(chan->chan_id);
 eimap:
 	return ret;
 }
@@ -1510,8 +1633,22 @@ static void idmac_free_chan_resources(struct dma_chan *chan)
 	__idmac_terminate_all(chan);
 
 	if (ichan->status > IPU_CHANNEL_FREE) {
+#ifdef DEBUG
+		if (chan->chan_id == IDMAC_IC_7) {
+			if (ic_sof > 0) {
+				free_irq(ic_sof, ichan);
+				ipu_irq_unmap(69);
+				ic_sof = -EINVAL;
+			}
+			if (ic_eof > 0) {
+				free_irq(ic_eof, ichan);
+				ipu_irq_unmap(70);
+				ic_eof = -EINVAL;
+			}
+		}
+#endif
 		free_irq(ichan->eof_irq, ichan);
-		ipu_irq_unmap(ichan->dma_chan.chan_id);
+		ipu_irq_unmap(chan->chan_id);
 	}
 
 	ichan->status = IPU_CHANNEL_FREE;
@@ -1573,7 +1710,7 @@ static int __init ipu_idmac_init(struct ipu *ipu)
 		dma_chan->device	= &idmac->dma;
 		dma_chan->cookie	= 1;
 		dma_chan->chan_id	= i;
-		list_add_tail(&ichan->dma_chan.device_node, &dma->channels);
+		list_add_tail(&dma_chan->device_node, &dma->channels);
 	}
 
 	idmac_write_icreg(ipu, 0x00000070, IDMAC_CONF);
@@ -1581,7 +1718,7 @@ static int __init ipu_idmac_init(struct ipu *ipu)
 	return dma_async_device_register(&idmac->dma);
 }
 
-static void ipu_idmac_exit(struct ipu *ipu)
+static void __exit ipu_idmac_exit(struct ipu *ipu)
 {
 	int i;
 	struct idmac *idmac = &ipu->idmac;
@@ -1600,7 +1737,7 @@ static void ipu_idmac_exit(struct ipu *ipu)
  * IPU common probe / remove
  */
 
-static int ipu_probe(struct platform_device *pdev)
+static int __init ipu_probe(struct platform_device *pdev)
 {
 	struct ipu_platform_data *pdata = pdev->dev.platform_data;
 	struct resource *mem_ipu, *mem_ic;
@@ -1700,7 +1837,7 @@ err_noirq:
 	return ret;
 }
 
-static int ipu_remove(struct platform_device *pdev)
+static int __exit ipu_remove(struct platform_device *pdev)
 {
 	struct ipu *ipu = platform_get_drvdata(pdev);
 
@@ -1725,7 +1862,7 @@ static struct platform_driver ipu_platform_driver = {
 		.name	= "ipu-core",
 		.owner	= THIS_MODULE,
 	},
-	.remove		= ipu_remove,
+	.remove		= __exit_p(ipu_remove),
 };
 
 static int __init ipu_init(void)
diff --git a/drivers/dma/ipu/ipu_irq.c b/drivers/dma/ipu/ipu_irq.c
index 83f532cc767..dd8ebc75b66 100644
--- a/drivers/dma/ipu/ipu_irq.c
+++ b/drivers/dma/ipu/ipu_irq.c
@@ -352,7 +352,7 @@ static struct irq_chip ipu_irq_chip = {
 };
 
 /* Install the IRQ handler */
-int ipu_irq_attach_irq(struct ipu *ipu, struct platform_device *dev)
+int __init ipu_irq_attach_irq(struct ipu *ipu, struct platform_device *dev)
 {
 	struct ipu_platform_data *pdata = dev->dev.platform_data;
 	unsigned int irq, irq_base, i;
diff --git a/drivers/dma/mv_xor.c b/drivers/dma/mv_xor.c
index cb7f26fb9f1..ddab94f5122 100644
--- a/drivers/dma/mv_xor.c
+++ b/drivers/dma/mv_xor.c
@@ -632,7 +632,6 @@ static int mv_xor_alloc_chan_resources(struct dma_chan *chan)
 		slot->async_tx.tx_submit = mv_xor_tx_submit;
 		INIT_LIST_HEAD(&slot->chain_node);
 		INIT_LIST_HEAD(&slot->slot_node);
-		INIT_LIST_HEAD(&slot->async_tx.tx_list);
 		hw_desc = (char *) mv_chan->device->dma_desc_pool;
 		slot->async_tx.phys =
 			(dma_addr_t) &hw_desc[idx * MV_XOR_SLOT_SIZE];
diff --git a/include/linux/async_tx.h b/include/linux/async_tx.h
index 45f6297821b..5fc2ef8d97f 100644
--- a/include/linux/async_tx.h
+++ b/include/linux/async_tx.h
@@ -21,6 +21,15 @@
 #include <linux/spinlock.h>
 #include <linux/interrupt.h>
 
+/* on architectures without dma-mapping capabilities we need to ensure
+ * that the asynchronous path compiles away
+ */
+#ifdef CONFIG_HAS_DMA
+#define __async_inline
+#else
+#define __async_inline __always_inline
+#endif
+
 /**
  * dma_chan_ref - object used to manage dma channels received from the
  *   dmaengine core.
diff --git a/include/linux/dmaengine.h b/include/linux/dmaengine.h
index 1956c8d46d3..2e2aa3df170 100644
--- a/include/linux/dmaengine.h
+++ b/include/linux/dmaengine.h
@@ -23,9 +23,6 @@
 
 #include <linux/device.h>
 #include <linux/uio.h>
-#include <linux/kref.h>
-#include <linux/completion.h>
-#include <linux/rcupdate.h>
 #include <linux/dma-mapping.h>
 
 /**
@@ -205,6 +202,7 @@ struct dma_async_tx_descriptor {
 /**
  * struct dma_device - info on the entity supplying DMA services
  * @chancnt: how many DMA channels are supported
+ * @privatecnt: how many DMA channels are requested by dma_request_channel
  * @channels: the list of struct dma_chan
  * @global_node: list_head for global dma_device_list
  * @cap_mask: one or more dma_capability flags
@@ -227,6 +225,7 @@ struct dma_async_tx_descriptor {
 struct dma_device {
 
 	unsigned int chancnt;
+	unsigned int privatecnt;
 	struct list_head channels;
 	struct list_head global_node;
 	dma_cap_mask_t  cap_mask;
@@ -291,6 +290,24 @@ static inline void net_dmaengine_put(void)
 }
 #endif
 
+#ifdef CONFIG_ASYNC_TX_DMA
+#define async_dmaengine_get()	dmaengine_get()
+#define async_dmaengine_put()	dmaengine_put()
+#define async_dma_find_channel(type) dma_find_channel(type)
+#else
+static inline void async_dmaengine_get(void)
+{
+}
+static inline void async_dmaengine_put(void)
+{
+}
+static inline struct dma_chan *
+async_dma_find_channel(enum dma_transaction_type type)
+{
+	return NULL;
+}
+#endif
+
 dma_cookie_t dma_async_memcpy_buf_to_buf(struct dma_chan *chan,
 	void *dest, void *src, size_t len);
 dma_cookie_t dma_async_memcpy_buf_to_pg(struct dma_chan *chan,
@@ -337,6 +354,13 @@ __dma_cap_set(enum dma_transaction_type tx_type, dma_cap_mask_t *dstp)
 	set_bit(tx_type, dstp->bits);
 }
 
+#define dma_cap_clear(tx, mask) __dma_cap_clear((tx), &(mask))
+static inline void
+__dma_cap_clear(enum dma_transaction_type tx_type, dma_cap_mask_t *dstp)
+{
+	clear_bit(tx_type, dstp->bits);
+}
+
 #define dma_cap_zero(mask) __dma_cap_zero(&(mask))
 static inline void __dma_cap_zero(dma_cap_mask_t *dstp)
 {
diff --git a/include/linux/dw_dmac.h b/include/linux/dw_dmac.h
index d797dde247f..c8aad713a04 100644
--- a/include/linux/dw_dmac.h
+++ b/include/linux/dw_dmac.h
@@ -74,4 +74,23 @@ struct dw_dma_slave {
 #define DWC_CFGL_HS_DST_POL	(1 << 18)	/* dst handshake active low */
 #define DWC_CFGL_HS_SRC_POL	(1 << 19)	/* src handshake active low */
 
+/* DMA API extensions */
+struct dw_cyclic_desc {
+	struct dw_desc	**desc;
+	unsigned long	periods;
+	void		(*period_callback)(void *param);
+	void		*period_callback_param;
+};
+
+struct dw_cyclic_desc *dw_dma_cyclic_prep(struct dma_chan *chan,
+		dma_addr_t buf_addr, size_t buf_len, size_t period_len,
+		enum dma_data_direction direction);
+void dw_dma_cyclic_free(struct dma_chan *chan);
+int dw_dma_cyclic_start(struct dma_chan *chan);
+void dw_dma_cyclic_stop(struct dma_chan *chan);
+
+dma_addr_t dw_dma_get_src_addr(struct dma_chan *chan);
+
+dma_addr_t dw_dma_get_dst_addr(struct dma_chan *chan);
+
 #endif /* DW_DMAC_H */