Merge branch 'ioat' into dmaengine

17 files changed, 170 insertions, 209 deletions

diff --git a/drivers/dma/at_hdmac.c b/drivers/dma/at_hdmac.c
index ee805a4..bd5250e 100644
--- a/drivers/dma/at_hdmac.c
+++ b/drivers/dma/at_hdmac.c
@@ -22,6 +22,7 @@
 #include <linux/interrupt.h>
 #include <linux/module.h>
 #include <linux/platform_device.h>
+#include <linux/slab.h>
 
 #include "at_hdmac_regs.h"
 
diff --git a/drivers/dma/coh901318_lli.c b/drivers/dma/coh901318_lli.c
index 71d58c1..9f7e0e6a 100644
--- a/drivers/dma/coh901318_lli.c
+++ b/drivers/dma/coh901318_lli.c
@@ -11,6 +11,7 @@
 #include <linux/spinlock.h>
 #include <linux/dmapool.h>
 #include <linux/memory.h>
+#include <linux/gfp.h>
 #include <mach/coh901318.h>
 
 #include "coh901318_lli.h"
diff --git a/drivers/dma/dmaengine.c b/drivers/dma/dmaengine.c
index 21eb896..9d31d5e 100644
--- a/drivers/dma/dmaengine.c
+++ b/drivers/dma/dmaengine.c
@@ -58,6 +58,7 @@
 #include <linux/jiffies.h>
 #include <linux/rculist.h>
 #include <linux/idr.h>
+#include <linux/slab.h>
 
 static DEFINE_MUTEX(dma_list_mutex);
 static LIST_HEAD(dma_device_list);
@@ -975,7 +976,9 @@ void dma_async_tx_descriptor_init(struct dma_async_tx_descriptor *tx,
 	struct dma_chan *chan)
 {
 	tx->chan = chan;
+	#ifndef CONFIG_ASYNC_TX_DISABLE_CHANNEL_SWITCH
 	spin_lock_init(&tx->lock);
+	#endif
 }
 EXPORT_SYMBOL(dma_async_tx_descriptor_init);
 
@@ -1008,7 +1011,7 @@ EXPORT_SYMBOL_GPL(dma_wait_for_async_tx);
  */
 void dma_run_dependencies(struct dma_async_tx_descriptor *tx)
 {
-	struct dma_async_tx_descriptor *dep = tx->next;
+	struct dma_async_tx_descriptor *dep = txd_next(tx);
 	struct dma_async_tx_descriptor *dep_next;
 	struct dma_chan *chan;
 
@@ -1016,7 +1019,7 @@ void dma_run_dependencies(struct dma_async_tx_descriptor *tx)
 		return;
 
 	/* we'll submit tx->next now, so clear the link */
-	tx->next = NULL;
+	txd_clear_next(tx);
 	chan = dep->chan;
 
 	/* keep submitting up until a channel switch is detected
@@ -1024,14 +1027,14 @@ void dma_run_dependencies(struct dma_async_tx_descriptor *tx)
 	 * processing the interrupt from async_tx_channel_switch
 	 */
 	for (; dep; dep = dep_next) {
-		spin_lock_bh(&dep->lock);
-		dep->parent = NULL;
-		dep_next = dep->next;
+		txd_lock(dep);
+		txd_clear_parent(dep);
+		dep_next = txd_next(dep);
 		if (dep_next && dep_next->chan == chan)
-			dep->next = NULL; /* ->next will be submitted */
+			txd_clear_next(dep); /* ->next will be submitted */
 		else
 			dep_next = NULL; /* submit current dep and terminate */
-		spin_unlock_bh(&dep->lock);
+		txd_unlock(dep);
 
 		dep->tx_submit(dep);
 	}
diff --git a/drivers/dma/dmatest.c b/drivers/dma/dmatest.c
index 6fa55fe..68d58c4 100644
--- a/drivers/dma/dmatest.c
+++ b/drivers/dma/dmatest.c
@@ -14,6 +14,7 @@
 #include <linux/module.h>
 #include <linux/moduleparam.h>
 #include <linux/random.h>
+#include <linux/slab.h>
 #include <linux/wait.h>
 
 static unsigned int test_buf_size = 16384;
diff --git a/drivers/dma/fsldma.c b/drivers/dma/fsldma.c
index 005329d..1fdf180 100644
--- a/drivers/dma/fsldma.c
+++ b/drivers/dma/fsldma.c
@@ -27,6 +27,7 @@
 #include <linux/init.h>
 #include <linux/module.h>
 #include <linux/pci.h>
+#include <linux/slab.h>
 #include <linux/interrupt.h>
 #include <linux/dmaengine.h>
 #include <linux/delay.h>
diff --git a/drivers/dma/ioat/dma.c b/drivers/dma/ioat/dma.c
index 59cebbf..c9213ea 100644
--- a/drivers/dma/ioat/dma.c
+++ b/drivers/dma/ioat/dma.c
@@ -27,6 +27,7 @@
 
 #include <linux/init.h>
 #include <linux/module.h>
+#include <linux/slab.h>
 #include <linux/pci.h>
 #include <linux/interrupt.h>
 #include <linux/dmaengine.h>
diff --git a/drivers/dma/ioat/dma.h b/drivers/dma/ioat/dma.h
index 26f48ef..6d3a73b 100644
--- a/drivers/dma/ioat/dma.h
+++ b/drivers/dma/ioat/dma.h
@@ -96,6 +96,7 @@ struct ioat_chan_common {
 	#define IOAT_COMPLETION_ACK 1
 	#define IOAT_RESET_PENDING 2
 	#define IOAT_KOBJ_INIT_FAIL 3
+	#define IOAT_RESHAPE_PENDING 4
 	struct timer_list timer;
 	#define COMPLETION_TIMEOUT msecs_to_jiffies(100)
 	#define IDLE_TIMEOUT msecs_to_jiffies(2000)
diff --git a/drivers/dma/ioat/dma_v2.c b/drivers/dma/ioat/dma_v2.c
index f540e0b..3c8b32a 100644
--- a/drivers/dma/ioat/dma_v2.c
+++ b/drivers/dma/ioat/dma_v2.c
@@ -27,6 +27,7 @@
 
 #include <linux/init.h>
 #include <linux/module.h>
+#include <linux/slab.h>
 #include <linux/pci.h>
 #include <linux/interrupt.h>
 #include <linux/dmaengine.h>
@@ -55,8 +56,6 @@ void __ioat2_issue_pending(struct ioat2_dma_chan *ioat)
 
 	ioat->dmacount += ioat2_ring_pending(ioat);
 	ioat->issued = ioat->head;
-	/* make descriptor updates globally visible before notifying channel */
-	wmb();
 	writew(ioat->dmacount, chan->reg_base + IOAT_CHAN_DMACOUNT_OFFSET);
 	dev_dbg(to_dev(chan),
 		"%s: head: %#x tail: %#x issued: %#x count: %#x\n",
@@ -68,9 +67,9 @@ void ioat2_issue_pending(struct dma_chan *c)
 	struct ioat2_dma_chan *ioat = to_ioat2_chan(c);
 
 	if (ioat2_ring_pending(ioat)) {
-		spin_lock_bh(&ioat->ring_lock);
+		spin_lock_bh(&ioat->prep_lock);
 		__ioat2_issue_pending(ioat);
-		spin_unlock_bh(&ioat->ring_lock);
+		spin_unlock_bh(&ioat->prep_lock);
 	}
 }
 
@@ -79,7 +78,7 @@ void ioat2_issue_pending(struct dma_chan *c)
  * @ioat: ioat2+ channel
  *
  * Check if the number of unsubmitted descriptors has exceeded the
- * watermark.  Called with ring_lock held
+ * watermark.  Called with prep_lock held
  */
 static void ioat2_update_pending(struct ioat2_dma_chan *ioat)
 {
@@ -91,7 +90,6 @@ static void __ioat2_start_null_desc(struct ioat2_dma_chan *ioat)
 {
 	struct ioat_ring_ent *desc;
 	struct ioat_dma_descriptor *hw;
-	int idx;
 
 	if (ioat2_ring_space(ioat) < 1) {
 		dev_err(to_dev(&ioat->base),
@@ -101,8 +99,7 @@ static void __ioat2_start_null_desc(struct ioat2_dma_chan *ioat)
 
 	dev_dbg(to_dev(&ioat->base), "%s: head: %#x tail: %#x issued: %#x\n",
 		__func__, ioat->head, ioat->tail, ioat->issued);
-	idx = ioat2_desc_alloc(ioat, 1);
-	desc = ioat2_get_ring_ent(ioat, idx);
+	desc = ioat2_get_ring_ent(ioat, ioat->head);
 
 	hw = desc->hw;
 	hw->ctl = 0;
@@ -116,14 +113,16 @@ static void __ioat2_start_null_desc(struct ioat2_dma_chan *ioat)
 	async_tx_ack(&desc->txd);
 	ioat2_set_chainaddr(ioat, desc->txd.phys);
 	dump_desc_dbg(ioat, desc);
+	wmb();
+	ioat->head += 1;
 	__ioat2_issue_pending(ioat);
 }
 
 static void ioat2_start_null_desc(struct ioat2_dma_chan *ioat)
 {
-	spin_lock_bh(&ioat->ring_lock);
+	spin_lock_bh(&ioat->prep_lock);
 	__ioat2_start_null_desc(ioat);
-	spin_unlock_bh(&ioat->ring_lock);
+	spin_unlock_bh(&ioat->prep_lock);
 }
 
 static void __cleanup(struct ioat2_dma_chan *ioat, unsigned long phys_complete)
@@ -133,15 +132,16 @@ static void __cleanup(struct ioat2_dma_chan *ioat, unsigned long phys_complete)
 	struct ioat_ring_ent *desc;
 	bool seen_current = false;
 	u16 active;
-	int i;
+	int idx = ioat->tail, i;
 
 	dev_dbg(to_dev(chan), "%s: head: %#x tail: %#x issued: %#x\n",
 		__func__, ioat->head, ioat->tail, ioat->issued);
 
 	active = ioat2_ring_active(ioat);
 	for (i = 0; i < active && !seen_current; i++) {
-		prefetch(ioat2_get_ring_ent(ioat, ioat->tail + i + 1));
-		desc = ioat2_get_ring_ent(ioat, ioat->tail + i);
+		smp_read_barrier_depends();
+		prefetch(ioat2_get_ring_ent(ioat, idx + i + 1));
+		desc = ioat2_get_ring_ent(ioat, idx + i);
 		tx = &desc->txd;
 		dump_desc_dbg(ioat, desc);
 		if (tx->cookie) {
@@ -157,11 +157,12 @@ static void __cleanup(struct ioat2_dma_chan *ioat, unsigned long phys_complete)
 		if (tx->phys == phys_complete)
 			seen_current = true;
 	}
-	ioat->tail += i;
+	smp_mb(); /* finish all descriptor reads before incrementing tail */
+	ioat->tail = idx + i;
 	BUG_ON(active && !seen_current); /* no active descs have written a completion? */
 
 	chan->last_completion = phys_complete;
-	if (ioat->head == ioat->tail) {
+	if (active - i == 0) {
 		dev_dbg(to_dev(chan), "%s: cancel completion timeout\n",
 			__func__);
 		clear_bit(IOAT_COMPLETION_PENDING, &chan->state);
@@ -178,24 +179,9 @@ static void ioat2_cleanup(struct ioat2_dma_chan *ioat)
 	struct ioat_chan_common *chan = &ioat->base;
 	unsigned long phys_complete;
 
-	prefetch(chan->completion);
-
-	if (!spin_trylock_bh(&chan->cleanup_lock))
-		return;
-
-	if (!ioat_cleanup_preamble(chan, &phys_complete)) {
-		spin_unlock_bh(&chan->cleanup_lock);
-		return;
-	}
-
-	if (!spin_trylock_bh(&ioat->ring_lock)) {
-		spin_unlock_bh(&chan->cleanup_lock);
-		return;
-	}
-
-	__cleanup(ioat, phys_complete);
-
-	spin_unlock_bh(&ioat->ring_lock);
+	spin_lock_bh(&chan->cleanup_lock);
+	if (ioat_cleanup_preamble(chan, &phys_complete))
+		__cleanup(ioat, phys_complete);
 	spin_unlock_bh(&chan->cleanup_lock);
 }
 
@@ -286,12 +272,10 @@ void ioat2_timer_event(unsigned long data)
 	struct ioat2_dma_chan *ioat = to_ioat2_chan((void *) data);
 	struct ioat_chan_common *chan = &ioat->base;
 
-	spin_lock_bh(&chan->cleanup_lock);
 	if (test_bit(IOAT_COMPLETION_PENDING, &chan->state)) {
 		unsigned long phys_complete;
 		u64 status;
 
-		spin_lock_bh(&ioat->ring_lock);
 		status = ioat_chansts(chan);
 
 		/* when halted due to errors check for channel
@@ -310,26 +294,31 @@ void ioat2_timer_event(unsigned long data)
 		 * acknowledged a pending completion once, then be more
 		 * forceful with a restart
 		 */
-		if (ioat_cleanup_preamble(chan, &phys_complete))
+		spin_lock_bh(&chan->cleanup_lock);
+		if (ioat_cleanup_preamble(chan, &phys_complete)) {
 			__cleanup(ioat, phys_complete);
-		else if (test_bit(IOAT_COMPLETION_ACK, &chan->state))
+		} else if (test_bit(IOAT_COMPLETION_ACK, &chan->state)) {
+			spin_lock_bh(&ioat->prep_lock);
 			ioat2_restart_channel(ioat);
-		else {
+			spin_unlock_bh(&ioat->prep_lock);
+		} else {
 			set_bit(IOAT_COMPLETION_ACK, &chan->state);
 			mod_timer(&chan->timer, jiffies + COMPLETION_TIMEOUT);
 		}
-		spin_unlock_bh(&ioat->ring_lock);
+		spin_unlock_bh(&chan->cleanup_lock);
 	} else {
 		u16 active;
 
 		/* if the ring is idle, empty, and oversized try to step
 		 * down the size
 		 */
-		spin_lock_bh(&ioat->ring_lock);
+		spin_lock_bh(&chan->cleanup_lock);
+		spin_lock_bh(&ioat->prep_lock);
 		active = ioat2_ring_active(ioat);
 		if (active == 0 && ioat->alloc_order > ioat_get_alloc_order())
 			reshape_ring(ioat, ioat->alloc_order-1);
-		spin_unlock_bh(&ioat->ring_lock);
+		spin_unlock_bh(&ioat->prep_lock);
+		spin_unlock_bh(&chan->cleanup_lock);
 
 		/* keep shrinking until we get back to our minimum
 		 * default size
@@ -337,7 +326,6 @@ void ioat2_timer_event(unsigned long data)
 		if (ioat->alloc_order > ioat_get_alloc_order())
 			mod_timer(&chan->timer, jiffies + IDLE_TIMEOUT);
 	}
-	spin_unlock_bh(&chan->cleanup_lock);
 }
 
 static int ioat2_reset_hw(struct ioat_chan_common *chan)
@@ -391,7 +379,7 @@ int ioat2_enumerate_channels(struct ioatdma_device *device)
 
 		ioat_init_channel(device, &ioat->base, i);
 		ioat->xfercap_log = xfercap_log;
-		spin_lock_init(&ioat->ring_lock);
+		spin_lock_init(&ioat->prep_lock);
 		if (device->reset_hw(&ioat->base)) {
 			i = 0;
 			break;
@@ -417,8 +405,17 @@ static dma_cookie_t ioat2_tx_submit_unlock(struct dma_async_tx_descriptor *tx)
 
 	if (!test_and_set_bit(IOAT_COMPLETION_PENDING, &chan->state))
 		mod_timer(&chan->timer, jiffies + COMPLETION_TIMEOUT);
+
+	/* make descriptor updates visible before advancing ioat->head,
+	 * this is purposefully not smp_wmb() since we are also
+	 * publishing the descriptor updates to a dma device
+	 */
+	wmb();
+
+	ioat->head += ioat->produce;
+
 	ioat2_update_pending(ioat);
-	spin_unlock_bh(&ioat->ring_lock);
+	spin_unlock_bh(&ioat->prep_lock);
 
 	return cookie;
 }
@@ -530,13 +527,15 @@ int ioat2_alloc_chan_resources(struct dma_chan *c)
 	if (!ring)
 		return -ENOMEM;
 
-	spin_lock_bh(&ioat->ring_lock);
+	spin_lock_bh(&chan->cleanup_lock);
+	spin_lock_bh(&ioat->prep_lock);
 	ioat->ring = ring;
 	ioat->head = 0;
 	ioat->issued = 0;
 	ioat->tail = 0;
 	ioat->alloc_order = order;
-	spin_unlock_bh(&ioat->ring_lock);
+	spin_unlock_bh(&ioat->prep_lock);
+	spin_unlock_bh(&chan->cleanup_lock);
 
 	tasklet_enable(&chan->cleanup_task);
 	ioat2_start_null_desc(ioat);
@@ -552,7 +551,7 @@ bool reshape_ring(struct ioat2_dma_chan *ioat, int order)
 	 */
 	struct ioat_chan_common *chan = &ioat->base;
 	struct dma_chan *c = &chan->common;
-	const u16 curr_size = ioat2_ring_mask(ioat) + 1;
+	const u16 curr_size = ioat2_ring_size(ioat);
 	const u16 active = ioat2_ring_active(ioat);
 	const u16 new_size = 1 << order;
 	struct ioat_ring_ent **ring;
@@ -652,54 +651,61 @@ bool reshape_ring(struct ioat2_dma_chan *ioat, int order)
 }
 
 /**
- * ioat2_alloc_and_lock - common descriptor alloc boilerplate for ioat2,3 ops
- * @idx: gets starting descriptor index on successful allocation
+ * ioat2_check_space_lock - verify space and grab ring producer lock
  * @ioat: ioat2,3 channel (ring) to operate on
  * @num_descs: allocation length
  */
-int ioat2_alloc_and_lock(u16 *idx, struct ioat2_dma_chan *ioat, int num_descs)
+int ioat2_check_space_lock(struct ioat2_dma_chan *ioat, int num_descs)
 {
 	struct ioat_chan_common *chan = &ioat->base;
+	bool retry;
 
-	spin_lock_bh(&ioat->ring_lock);
+ retry:
+	spin_lock_bh(&ioat->prep_lock);
 	/* never allow the last descriptor to be consumed, we need at
 	 * least one free at all times to allow for on-the-fly ring
 	 * resizing.
 	 */
-	while (unlikely(ioat2_ring_space(ioat) <= num_descs)) {
-		if (reshape_ring(ioat, ioat->alloc_order + 1) &&
-		    ioat2_ring_space(ioat) > num_descs)
-				break;
-
-		if (printk_ratelimit())
-			dev_dbg(to_dev(chan),
-				"%s: ring full! num_descs: %d (%x:%x:%x)\n",
-				__func__, num_descs, ioat->head, ioat->tail,
-				ioat->issued);
-		spin_unlock_bh(&ioat->ring_lock);
-
-		/* progress reclaim in the allocation failure case we
-		 * may be called under bh_disabled so we need to trigger
-		 * the timer event directly
-		 */
-		spin_lock_bh(&chan->cleanup_lock);
-		if (jiffies > chan->timer.expires &&
-		    timer_pending(&chan->timer)) {
-			struct ioatdma_device *device = chan->device;
-
-			mod_timer(&chan->timer, jiffies + COMPLETION_TIMEOUT);
-			spin_unlock_bh(&chan->cleanup_lock);
-			device->timer_fn((unsigned long) &chan->common);
-		} else
-			spin_unlock_bh(&chan->cleanup_lock);
-		return -ENOMEM;
+	if (likely(ioat2_ring_space(ioat) > num_descs)) {
+		dev_dbg(to_dev(chan), "%s: num_descs: %d (%x:%x:%x)\n",
+			__func__, num_descs, ioat->head, ioat->tail, ioat->issued);
+		ioat->produce = num_descs;
+		return 0;  /* with ioat->prep_lock held */
 	}
+	retry = test_and_set_bit(IOAT_RESHAPE_PENDING, &chan->state);
+	spin_unlock_bh(&ioat->prep_lock);
 
-	dev_dbg(to_dev(chan), "%s: num_descs: %d (%x:%x:%x)\n",
-		__func__, num_descs, ioat->head, ioat->tail, ioat->issued);
+	/* is another cpu already trying to expand the ring? */
+	if (retry)
+		goto retry;
 
-	*idx = ioat2_desc_alloc(ioat, num_descs);
-	return 0;  /* with ioat->ring_lock held */
+	spin_lock_bh(&chan->cleanup_lock);
+	spin_lock_bh(&ioat->prep_lock);
+	retry = reshape_ring(ioat, ioat->alloc_order + 1);
+	clear_bit(IOAT_RESHAPE_PENDING, &chan->state);
+	spin_unlock_bh(&ioat->prep_lock);
+	spin_unlock_bh(&chan->cleanup_lock);
+
+	/* if we were able to expand the ring retry the allocation */
+	if (retry)
+		goto retry;
+
+	if (printk_ratelimit())
+		dev_dbg(to_dev(chan), "%s: ring full! num_descs: %d (%x:%x:%x)\n",
+			__func__, num_descs, ioat->head, ioat->tail, ioat->issued);
+
+	/* progress reclaim in the allocation failure case we may be
+	 * called under bh_disabled so we need to trigger the timer
+	 * event directly
+	 */
+	if (jiffies > chan->timer.expires && timer_pending(&chan->timer)) {
+		struct ioatdma_device *device = chan->device;
+
+		mod_timer(&chan->timer, jiffies + COMPLETION_TIMEOUT);
+		device->timer_fn((unsigned long) &chan->common);
+	}
+
+	return -ENOMEM;
 }
 
 struct dma_async_tx_descriptor *
@@ -712,14 +718,11 @@ ioat2_dma_prep_memcpy_lock(struct dma_chan *c, dma_addr_t dma_dest,
 	dma_addr_t dst = dma_dest;
 	dma_addr_t src = dma_src;
 	size_t total_len = len;
-	int num_descs;
-	u16 idx;
-	int i;
+	int num_descs, idx, i;
 
 	num_descs = ioat2_xferlen_to_descs(ioat, len);
-	if (likely(num_descs) &&
-	    ioat2_alloc_and_lock(&idx, ioat, num_descs) == 0)
-		/* pass */;
+	if (likely(num_descs) && ioat2_check_space_lock(ioat, num_descs) == 0)
+		idx = ioat->head;
 	else
 		return NULL;
 	i = 0;
@@ -776,7 +779,8 @@ void ioat2_free_chan_resources(struct dma_chan *c)
 	device->cleanup_fn((unsigned long) c);
 	device->reset_hw(chan);
 
-	spin_lock_bh(&ioat->ring_lock);
+	spin_lock_bh(&chan->cleanup_lock);
+	spin_lock_bh(&ioat->prep_lock);
 	descs = ioat2_ring_space(ioat);
 	dev_dbg(to_dev(chan), "freeing %d idle descriptors\n", descs);
 	for (i = 0; i < descs; i++) {
@@ -799,7 +803,8 @@ void ioat2_free_chan_resources(struct dma_chan *c)
 	ioat->alloc_order = 0;
 	pci_pool_free(device->completion_pool, chan->completion,
 		      chan->completion_dma);
-	spin_unlock_bh(&ioat->ring_lock);
+	spin_unlock_bh(&ioat->prep_lock);
+	spin_unlock_bh(&chan->cleanup_lock);
 
 	chan->last_completion = 0;
 	chan->completion_dma = 0;
diff --git a/drivers/dma/ioat/dma_v2.h b/drivers/dma/ioat/dma_v2.h
index ef2871f..a2c413b 100644
--- a/drivers/dma/ioat/dma_v2.h
+++ b/drivers/dma/ioat/dma_v2.h
@@ -22,6 +22,7 @@
 #define IOATDMA_V2_H
 
 #include <linux/dmaengine.h>
+#include <linux/circ_buf.h>
 #include "dma.h"
 #include "hw.h"
 
@@ -49,8 +50,9 @@ extern int ioat_ring_alloc_order;
  * @tail: cleanup index
  * @dmacount: identical to 'head' except for occasionally resetting to zero
  * @alloc_order: log2 of the number of allocated descriptors
+ * @produce: number of descriptors to produce at submit time
  * @ring: software ring buffer implementation of hardware ring
- * @ring_lock: protects ring attributes
+ * @prep_lock: serializes descriptor preparation (producers)
  */
 struct ioat2_dma_chan {
 	struct ioat_chan_common base;
@@ -60,8 +62,9 @@ struct ioat2_dma_chan {
 	u16 tail;
 	u16 dmacount;
 	u16 alloc_order;
+	u16 produce;
 	struct ioat_ring_ent **ring;
-	spinlock_t ring_lock;
+	spinlock_t prep_lock;
 };
 
 static inline struct ioat2_dma_chan *to_ioat2_chan(struct dma_chan *c)
@@ -71,38 +74,26 @@ static inline struct ioat2_dma_chan *to_ioat2_chan(struct dma_chan *c)
 	return container_of(chan, struct ioat2_dma_chan, base);
 }
 
-static inline u16 ioat2_ring_mask(struct ioat2_dma_chan *ioat)
+static inline u16 ioat2_ring_size(struct ioat2_dma_chan *ioat)
 {
-	return (1 << ioat->alloc_order) - 1;
+	return 1 << ioat->alloc_order;
 }
 
 /* count of descriptors in flight with the engine */
 static inline u16 ioat2_ring_active(struct ioat2_dma_chan *ioat)
 {
-	return (ioat->head - ioat->tail) & ioat2_ring_mask(ioat);
+	return CIRC_CNT(ioat->head, ioat->tail, ioat2_ring_size(ioat));
 }
 
 /* count of descriptors pending submission to hardware */
 static inline u16 ioat2_ring_pending(struct ioat2_dma_chan *ioat)
 {
-	return (ioat->head - ioat->issued) & ioat2_ring_mask(ioat);
+	return CIRC_CNT(ioat->head, ioat->issued, ioat2_ring_size(ioat));
 }
 
 static inline u16 ioat2_ring_space(struct ioat2_dma_chan *ioat)
 {
-	u16 num_descs = ioat2_ring_mask(ioat) + 1;
-	u16 active = ioat2_ring_active(ioat);
-
-	BUG_ON(active > num_descs);
-
-	return num_descs - active;
-}
-
-/* assumes caller already checked space */
-static inline u16 ioat2_desc_alloc(struct ioat2_dma_chan *ioat, u16 len)
-{
-	ioat->head += len;
-	return ioat->head - len;
+	return ioat2_ring_size(ioat) - ioat2_ring_active(ioat);
 }
 
 static inline u16 ioat2_xferlen_to_descs(struct ioat2_dma_chan *ioat, size_t len)
@@ -151,7 +142,7 @@ struct ioat_ring_ent {
 static inline struct ioat_ring_ent *
 ioat2_get_ring_ent(struct ioat2_dma_chan *ioat, u16 idx)
 {
-	return ioat->ring[idx & ioat2_ring_mask(ioat)];
+	return ioat->ring[idx & (ioat2_ring_size(ioat) - 1)];
 }
 
 static inline void ioat2_set_chainaddr(struct ioat2_dma_chan *ioat, u64 addr)
@@ -168,7 +159,7 @@ int __devinit ioat2_dma_probe(struct ioatdma_device *dev, int dca);
 int __devinit ioat3_dma_probe(struct ioatdma_device *dev, int dca);
 struct dca_provider * __devinit ioat2_dca_init(struct pci_dev *pdev, void __iomem *iobase);
 struct dca_provider * __devinit ioat3_dca_init(struct pci_dev *pdev, void __iomem *iobase);
-int ioat2_alloc_and_lock(u16 *idx, struct ioat2_dma_chan *ioat, int num_descs);
+int ioat2_check_space_lock(struct ioat2_dma_chan *ioat, int num_descs);
 int ioat2_enumerate_channels(struct ioatdma_device *device);
 struct dma_async_tx_descriptor *
 ioat2_dma_prep_memcpy_lock(struct dma_chan *c, dma_addr_t dma_dest,
diff --git a/drivers/dma/ioat/dma_v3.c b/drivers/dma/ioat/dma_v3.c
index d1adbf3..1cdd22e 100644
--- a/drivers/dma/ioat/dma_v3.c
+++ b/drivers/dma/ioat/dma_v3.c
@@ -57,6 +57,7 @@
  */
 
 #include <linux/pci.h>
+#include <linux/gfp.h>
 #include <linux/dmaengine.h>
 #include <linux/dma-mapping.h>
 #include "registers.h"
@@ -259,8 +260,8 @@ static void __cleanup(struct ioat2_dma_chan *ioat, unsigned long phys_complete)
 	struct ioat_chan_common *chan = &ioat->base;
 	struct ioat_ring_ent *desc;
 	bool seen_current = false;
+	int idx = ioat->tail, i;
 	u16 active;
-	int i;
 
 	dev_dbg(to_dev(chan), "%s: head: %#x tail: %#x issued: %#x\n",
 		__func__, ioat->head, ioat->tail, ioat->issued);
@@ -269,13 +270,14 @@ static void __cleanup(struct ioat2_dma_chan *ioat, unsigned long phys_complete)
 	for (i = 0; i < active && !seen_current; i++) {
 		struct dma_async_tx_descriptor *tx;
 
-		prefetch(ioat2_get_ring_ent(ioat, ioat->tail + i + 1));
-		desc = ioat2_get_ring_ent(ioat, ioat->tail + i);
+		smp_read_barrier_depends();
+		prefetch(ioat2_get_ring_ent(ioat, idx + i + 1));
+		desc = ioat2_get_ring_ent(ioat, idx + i);
 		dump_desc_dbg(ioat, desc);
 		tx = &desc->txd;
 		if (tx->cookie) {
 			chan->completed_cookie = tx->cookie;
-			ioat3_dma_unmap(ioat, desc, ioat->tail + i);
+			ioat3_dma_unmap(ioat, desc, idx + i);
 			tx->cookie = 0;
 			if (tx->callback) {
 				tx->callback(tx->callback_param);
@@ -292,69 +294,30 @@ static void __cleanup(struct ioat2_dma_chan *ioat, unsigned long phys_complete)
 			i++;
 		}
 	}
-	ioat->tail += i;
+	smp_mb(); /* finish all descriptor reads before incrementing tail */
+	ioat->tail = idx + i;
 	BUG_ON(active && !seen_current); /* no active descs have written a completion? */
 	chan->last_completion = phys_complete;
 
-	active = ioat2_ring_active(ioat);
-	if (active == 0) {
+	if (active - i == 0) {
 		dev_dbg(to_dev(chan), "%s: cancel completion timeout\n",
 			__func__);
 		clear_bit(IOAT_COMPLETION_PENDING, &chan->state);
 		mod_timer(&chan->timer, jiffies + IDLE_TIMEOUT);
 	}
 	/* 5 microsecond delay per pending descriptor */
-	writew(min((5 * active), IOAT_INTRDELAY_MASK),
+	writew(min((5 * (active - i)), IOAT_INTRDELAY_MASK),
 	       chan->device->reg_base + IOAT_INTRDELAY_OFFSET);
 }
 
-/* try to cleanup, but yield (via spin_trylock) to incoming submissions
- * with the expectation that we will immediately poll again shortly
- */
-static void ioat3_cleanup_poll(struct ioat2_dma_chan *ioat)
+static void ioat3_cleanup(struct ioat2_dma_chan *ioat)
 {
 	struct ioat_chan_common *chan = &ioat->base;
 	unsigned long phys_complete;
 
-	prefetch(chan->completion);
-
-	if (!spin_trylock_bh(&chan->cleanup_lock))
-		return;
-
-	if (!ioat_cleanup_preamble(chan, &phys_complete)) {
-		spin_unlock_bh(&chan->cleanup_lock);
-		return;
-	}
-
-	if (!spin_trylock_bh(&ioat->ring_lock)) {
-		spin_unlock_bh(&chan->cleanup_lock);
-		return;
-	}
-
-	__cleanup(ioat, phys_complete);
-
-	spin_unlock_bh(&ioat->ring_lock);
-	spin_unlock_bh(&chan->cleanup_lock);
-}
-
-/* run cleanup now because we already delayed the interrupt via INTRDELAY */
-static void ioat3_cleanup_sync(struct ioat2_dma_chan *ioat)
-{
-	struct ioat_chan_common *chan = &ioat->base;
-	unsigned long phys_complete;
-
-	prefetch(chan->completion);
-
 	spin_lock_bh(&chan->cleanup_lock);
-	if (!ioat_cleanup_preamble(chan, &phys_complete)) {
-		spin_unlock_bh(&chan->cleanup_lock);
-		return;
-	}
-	spin_lock_bh(&ioat->ring_lock);
-
-	__cleanup(ioat, phys_complete);
-
-	spin_unlock_bh(&ioat->ring_lock);
+	if (ioat_cleanup_preamble(chan, &phys_complete))
+		__cleanup(ioat, phys_complete);
 	spin_unlock_bh(&chan->cleanup_lock);
 }
 
@@ -362,7 +325,7 @@ static void ioat3_cleanup_event(unsigned long data)
 {
 	struct ioat2_dma_chan *ioat = to_ioat2_chan((void *) data);
 
-	ioat3_cleanup_sync(ioat);
+	ioat3_cleanup(ioat);
 	writew(IOAT_CHANCTRL_RUN, ioat->base.reg_base + IOAT_CHANCTRL_OFFSET);
 }
 
@@ -383,12 +346,10 @@ static void ioat3_timer_event(unsigned long data)
 	struct ioat2_dma_chan *ioat = to_ioat2_chan((void *) data);
 	struct ioat_chan_common *chan = &ioat->base;
 
-	spin_lock_bh(&chan->cleanup_lock);
 	if (test_bit(IOAT_COMPLETION_PENDING, &chan->state)) {
 		unsigned long phys_complete;
 		u64 status;
 
-		spin_lock_bh(&ioat->ring_lock);
 		status = ioat_chansts(chan);
 
 		/* when halted due to errors check for channel
@@ -407,26 +368,31 @@ static void ioat3_timer_event(unsigned long data)
 		 * acknowledged a pending completion once, then be more
 		 * forceful with a restart
 		 */
+		spin_lock_bh(&chan->cleanup_lock);
 		if (ioat_cleanup_preamble(chan, &phys_complete))
 			__cleanup(ioat, phys_complete);
-		else if (test_bit(IOAT_COMPLETION_ACK, &chan->state))
+		else if (test_bit(IOAT_COMPLETION_ACK, &chan->state)) {
+			spin_lock_bh(&ioat->prep_lock);
 			ioat3_restart_channel(ioat);
-		else {
+			spin_unlock_bh(&ioat->prep_lock);
+		} else {
 			set_bit(IOAT_COMPLETION_ACK, &chan->state);
 			mod_timer(&chan->timer, jiffies + COMPLETION_TIMEOUT);
 		}
-		spin_unlock_bh(&ioat->ring_lock);
+		spin_unlock_bh(&chan->cleanup_lock);
 	} else {
 		u16 active;
 
 		/* if the ring is idle, empty, and oversized try to step
 		 * down the size
 		 */
-		spin_lock_bh(&ioat->ring_lock);
+		spin_lock_bh(&chan->cleanup_lock);
+		spin_lock_bh(&ioat->prep_lock);
 		active = ioat2_ring_active(ioat);
 		if (active == 0 && ioat->alloc_order > ioat_get_alloc_order())
 			reshape_ring(ioat, ioat->alloc_order-1);
-		spin_unlock_bh(&ioat->ring_lock);
+		spin_unlock_bh(&ioat->prep_lock);
+		spin_unlock_bh(&chan->cleanup_lock);
 
 		/* keep shrinking until we get back to our minimum
 		 * default size
@@ -434,7 +400,6 @@ static void ioat3_timer_event(unsigned long data)
 		if (ioat->alloc_order > ioat_get_alloc_order())
 			mod_timer(&chan->timer, jiffies + IDLE_TIMEOUT);
 	}
-	spin_unlock_bh(&chan->cleanup_lock);
 }
 
 static enum dma_status
@@ -446,7 +411,7 @@ ioat3_tx_status(struct dma_chan *c, dma_cookie_t cookie,
 	if (ioat_tx_status(c, cookie, txstate) == DMA_SUCCESS)
 		return DMA_SUCCESS;
 
-	ioat3_cleanup_poll(ioat);
+	ioat3_cleanup(ioat);
 
 	return ioat_tx_status(c, cookie, txstate);
 }
@@ -459,15 +424,12 @@ ioat3_prep_memset_lock(struct dma_chan *c, dma_addr_t dest, int value,
 	struct ioat_ring_ent *desc;
 	size_t total_len = len;
 	struct ioat_fill_descriptor *fill;
-	int num_descs;
 	u64 src_data = (0x0101010101010101ULL) * (value & 0xff);
-	u16 idx;
-	int i;
+	int num_descs, idx, i;
 
 	num_descs = ioat2_xferlen_to_descs(ioat, len);
-	if (likely(num_descs) &&
-	    ioat2_alloc_and_lock(&idx, ioat, num_descs) == 0)
-		/* pass */;
+	if (likely(num_descs) && ioat2_check_space_lock(ioat, num_descs) == 0)
+		idx = ioat->head;
 	else
 		return NULL;
 	i = 0;
@@ -512,11 +474,8 @@ __ioat3_prep_xor_lock(struct dma_chan *c, enum sum_check_flags *result,
 	struct ioat_xor_descriptor *xor;
 	struct ioat_xor_ext_descriptor *xor_ex = NULL;
 	struct ioat_dma_descriptor *hw;
+	int num_descs, with_ext, idx, i;
 	u32 offset = 0;
-	int num_descs;
-	int with_ext;
-	int i;
-	u16 idx;
 	u8 op = result ? IOAT_OP_XOR_VAL : IOAT_OP_XOR;
 
 	BUG_ON(src_cnt < 2);
@@ -536,9 +495,8 @@ __ioat3_prep_xor_lock(struct dma_chan *c, enum sum_check_flags *result,
 	 * (legacy) descriptor to ensure all completion writes arrive in
 	 * order.
 	 */
-	if (likely(num_descs) &&
-	    ioat2_alloc_and_lock(&idx, ioat, num_descs+1) == 0)
-		/* pass */;
+	if (likely(num_descs) && ioat2_check_space_lock(ioat, num_descs+1) == 0)
+		idx = ioat->head;
 	else
 		return NULL;
 	i = 0;
@@ -656,11 +614,8 @@ __ioat3_prep_pq_lock(struct dma_chan *c, enum sum_check_flags *result,
 	struct ioat_pq_ext_descriptor *pq_ex = NULL;
 	struct ioat_dma_descriptor *hw;
 	u32 offset = 0;
-	int num_descs;
-	int with_ext;
-	int i, s;
-	u16 idx;
 	u8 op = result ? IOAT_OP_PQ_VAL : IOAT_OP_PQ;
+	int i, s, idx, with_ext, num_descs;
 
 	dev_dbg(to_dev(chan), "%s\n", __func__);
 	/* the engine requires at least two sources (we provide
@@ -686,8 +641,8 @@ __ioat3_prep_pq_lock(struct dma_chan *c, enum sum_check_flags *result,
 	 * order.
 	 */
 	if (likely(num_descs) &&
-	    ioat2_alloc_and_lock(&idx, ioat, num_descs+1) == 0)
-		/* pass */;
+	    ioat2_check_space_lock(ioat, num_descs+1) == 0)
+		idx = ioat->head;
 	else
 		return NULL;
 	i = 0;
@@ -850,10 +805,9 @@ ioat3_prep_interrupt_lock(struct dma_chan *c, unsigned long flags)
 	struct ioat2_dma_chan *ioat = to_ioat2_chan(c);
 	struct ioat_ring_ent *desc;
 	struct ioat_dma_descriptor *hw;
-	u16 idx;
 
-	if (ioat2_alloc_and_lock(&idx, ioat, 1) == 0)
-		desc = ioat2_get_ring_ent(ioat, idx);
+	if (ioat2_check_space_lock(ioat, 1) == 0)
+		desc = ioat2_get_ring_ent(ioat, ioat->head);
 	else
 		return NULL;
 
@@ -1221,7 +1175,7 @@ int __devinit ioat3_dma_probe(struct ioatdma_device *device, int dca)
 	if (cap & IOAT_CAP_XOR) {
 		is_raid_device = true;
 		dma->max_xor = 8;
-		dma->xor_align = 2;
+		dma->xor_align = 6;
 
 		dma_cap_set(DMA_XOR, dma->cap_mask);
 		dma->device_prep_dma_xor = ioat3_prep_xor;
@@ -1232,7 +1186,7 @@ int __devinit ioat3_dma_probe(struct ioatdma_device *device, int dca)
 	if (cap & IOAT_CAP_PQ) {
 		is_raid_device = true;
 		dma_set_maxpq(dma, 8, 0);
-		dma->pq_align = 2;
+		dma->pq_align = 6;
 
 		dma_cap_set(DMA_PQ, dma->cap_mask);
 		dma->device_prep_dma_pq = ioat3_prep_pq;
@@ -1242,7 +1196,7 @@ int __devinit ioat3_dma_probe(struct ioatdma_device *device, int dca)
 
 		if (!(cap & IOAT_CAP_XOR)) {
 			dma->max_xor = 8;
-			dma->xor_align = 2;
+			dma->xor_align = 6;
 
 			dma_cap_set(DMA_XOR, dma->cap_mask);
 			dma->device_prep_dma_xor = ioat3_prep_pqxor;
diff --git a/drivers/dma/ioat/pci.c b/drivers/dma/ioat/pci.c
index d545fae3..fab37d1 100644
--- a/drivers/dma/ioat/pci.c
+++ b/drivers/dma/ioat/pci.c
@@ -30,6 +30,7 @@
 #include <linux/pci.h>
 #include <linux/interrupt.h>
 #include <linux/dca.h>
+#include <linux/slab.h>
 #include "dma.h"
 #include "dma_v2.h"
 #include "registers.h"
@@ -137,15 +138,10 @@ static int __devinit ioat_pci_probe(struct pci_dev *pdev, const struct pci_devic
 	if (err)
 		return err;
 
-	device = devm_kzalloc(dev, sizeof(*device), GFP_KERNEL);
-	if (!device)
-		return -ENOMEM;
-
-	pci_set_master(pdev);
-
 	device = alloc_ioatdma(pdev, iomap[IOAT_MMIO_BAR]);
 	if (!device)
 		return -ENOMEM;
+	pci_set_master(pdev);
 	pci_set_drvdata(pdev, device);
 
 	device->version = readb(device->reg_base + IOAT_VER_OFFSET);
diff --git a/drivers/dma/iop-adma.c b/drivers/dma/iop-adma.c
index e5d4b97..161c452 100644
--- a/drivers/dma/iop-adma.c
+++ b/drivers/dma/iop-adma.c
@@ -32,6 +32,7 @@
 #include <linux/memory.h>
 #include <linux/ioport.h>
 #include <linux/raid/pq.h>
+#include <linux/slab.h>
 
 #include <mach/adma.h>
 
diff --git a/drivers/dma/iovlock.c b/drivers/dma/iovlock.c
index c0a272c7..bb48a57 100644
--- a/drivers/dma/iovlock.c
+++ b/drivers/dma/iovlock.c
@@ -27,6 +27,7 @@
 
 #include <linux/dmaengine.h>
 #include <linux/pagemap.h>
+#include <linux/slab.h>
 #include <net/tcp.h> /* for memcpy_toiovec */
 #include <asm/io.h>
 #include <asm/uaccess.h>
diff --git a/drivers/dma/mpc512x_dma.c b/drivers/dma/mpc512x_dma.c
index e41fb8b..201e6e1 100644
--- a/drivers/dma/mpc512x_dma.c
+++ b/drivers/dma/mpc512x_dma.c
@@ -37,6 +37,7 @@
 #include <linux/dma-mapping.h>
 #include <linux/interrupt.h>
 #include <linux/io.h>
+#include <linux/slab.h>
 #include <linux/of_device.h>
 #include <linux/of_platform.h>
 
diff --git a/drivers/dma/mv_xor.c b/drivers/dma/mv_xor.c
index 4b8c1fc..86c5ae9 100644
--- a/drivers/dma/mv_xor.c
+++ b/drivers/dma/mv_xor.c
@@ -18,6 +18,7 @@
 
 #include <linux/init.h>
 #include <linux/module.h>
+#include <linux/slab.h>
 #include <linux/delay.h>
 #include <linux/dma-mapping.h>
 #include <linux/spinlock.h>
diff --git a/drivers/dma/ppc4xx/adma.c b/drivers/dma/ppc4xx/adma.c
index 5558419..c6079fc 100644
--- a/drivers/dma/ppc4xx/adma.c
+++ b/drivers/dma/ppc4xx/adma.c
@@ -38,6 +38,7 @@
 #include <linux/dma-mapping.h>
 #include <linux/spinlock.h>
 #include <linux/interrupt.h>
+#include <linux/slab.h>
 #include <linux/uaccess.h>
 #include <linux/proc_fs.h>
 #include <linux/of.h>
diff --git a/drivers/dma/shdma.c b/drivers/dma/shdma.c
index e9de1d3..544340d 100644
--- a/drivers/dma/shdma.c
+++ b/drivers/dma/shdma.c
@@ -19,6 +19,7 @@
 
 #include <linux/init.h>
 #include <linux/module.h>
+#include <linux/slab.h>
 #include <linux/interrupt.h>
 #include <linux/dmaengine.h>
 #include <linux/delay.h>