Merge branch 'master' into for-next

Sync with Linus' tree to be able to apply fixes on top of newer things
in tree (efi-stub).

Signed-off-by: Jiri Kosina <jkosina@suse.cz>

5 files changed, 1023 insertions, 280 deletions

diff --git a/drivers/ntb/Kconfig b/drivers/ntb/Kconfig
index 37ee6495acc1..f69df793dbe2 100644
--- a/drivers/ntb/Kconfig
+++ b/drivers/ntb/Kconfig
@@ -1,7 +1,7 @@
 config NTB
        tristate "Intel Non-Transparent Bridge support"
        depends on PCI
-       depends on X86_64
+       depends on X86
        help
         The PCI-E Non-transparent bridge hardware is a point-to-point PCI-E bus
         connecting 2 systems.  When configured, writes to the device's PCI
diff --git a/drivers/ntb/ntb_hw.c b/drivers/ntb/ntb_hw.c
index 2dacd19e1b8a..170e8e60cdb7 100644
--- a/drivers/ntb/ntb_hw.c
+++ b/drivers/ntb/ntb_hw.c
@@ -46,24 +46,30 @@
  * Jon Mason <jon.mason@intel.com>
  */
 #include <linux/debugfs.h>
+#include <linux/delay.h>
 #include <linux/init.h>
 #include <linux/interrupt.h>
 #include <linux/module.h>
 #include <linux/pci.h>
+#include <linux/random.h>
 #include <linux/slab.h>
 #include "ntb_hw.h"
 #include "ntb_regs.h"
 
 #define NTB_NAME	"Intel(R) PCI-E Non-Transparent Bridge Driver"
-#define NTB_VER		"0.25"
+#define NTB_VER		"1.0"
 
 MODULE_DESCRIPTION(NTB_NAME);
 MODULE_VERSION(NTB_VER);
 MODULE_LICENSE("Dual BSD/GPL");
 MODULE_AUTHOR("Intel Corporation");
 
+static bool xeon_errata_workaround = true;
+module_param(xeon_errata_workaround, bool, 0644);
+MODULE_PARM_DESC(xeon_errata_workaround, "Workaround for the Xeon Errata");
+
 enum {
-	NTB_CONN_CLASSIC = 0,
+	NTB_CONN_TRANSPARENT = 0,
 	NTB_CONN_B2B,
 	NTB_CONN_RP,
 };
@@ -78,17 +84,27 @@ enum {
 	BWD_HW,
 };
 
+static struct dentry *debugfs_dir;
+
+#define BWD_LINK_RECOVERY_TIME	500
+
 /* Translate memory window 0,1 to BAR 2,4 */
-#define MW_TO_BAR(mw)	(mw * 2 + 2)
+#define MW_TO_BAR(mw)	(mw * NTB_MAX_NUM_MW + 2)
 
 static DEFINE_PCI_DEVICE_TABLE(ntb_pci_tbl) = {
 	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_B2B_BWD)},
 	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_B2B_JSF)},
-	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_CLASSIC_JSF)},
-	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_RP_JSF)},
-	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_RP_SNB)},
 	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_B2B_SNB)},
-	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_CLASSIC_SNB)},
+	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_B2B_IVT)},
+	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_B2B_HSX)},
+	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_PS_JSF)},
+	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_PS_SNB)},
+	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_PS_IVT)},
+	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_PS_HSX)},
+	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_SS_JSF)},
+	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_SS_SNB)},
+	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_SS_IVT)},
+	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_SS_HSX)},
 	{0}
 };
 MODULE_DEVICE_TABLE(pci, ntb_pci_tbl);
@@ -125,10 +141,29 @@ void ntb_unregister_event_callback(struct ntb_device *ndev)
 	ndev->event_cb = NULL;
 }
 
+static void ntb_irq_work(unsigned long data)
+{
+	struct ntb_db_cb *db_cb = (struct ntb_db_cb *)data;
+	int rc;
+
+	rc = db_cb->callback(db_cb->data, db_cb->db_num);
+	if (rc)
+		tasklet_schedule(&db_cb->irq_work);
+	else {
+		struct ntb_device *ndev = db_cb->ndev;
+		unsigned long mask;
+
+		mask = readw(ndev->reg_ofs.ldb_mask);
+		clear_bit(db_cb->db_num * ndev->bits_per_vector, &mask);
+		writew(mask, ndev->reg_ofs.ldb_mask);
+	}
+}
+
 /**
  * ntb_register_db_callback() - register a callback for doorbell interrupt
  * @ndev: pointer to ntb_device instance
  * @idx: doorbell index to register callback, zero based
+ * @data: pointer to be returned to caller with every callback
  * @func: callback function to register
  *
  * This function registers a callback function for the doorbell interrupt
@@ -138,7 +173,7 @@ void ntb_unregister_event_callback(struct ntb_device *ndev)
  * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
  */
 int ntb_register_db_callback(struct ntb_device *ndev, unsigned int idx,
-			     void *data, void (*func)(void *data, int db_num))
+			     void *data, int (*func)(void *data, int db_num))
 {
 	unsigned long mask;
 
@@ -149,11 +184,15 @@ int ntb_register_db_callback(struct ntb_device *ndev, unsigned int idx,
 
 	ndev->db_cb[idx].callback = func;
 	ndev->db_cb[idx].data = data;
+	ndev->db_cb[idx].ndev = ndev;
+
+	tasklet_init(&ndev->db_cb[idx].irq_work, ntb_irq_work,
+		     (unsigned long) &ndev->db_cb[idx]);
 
 	/* unmask interrupt */
-	mask = readw(ndev->reg_ofs.pdb_mask);
+	mask = readw(ndev->reg_ofs.ldb_mask);
 	clear_bit(idx * ndev->bits_per_vector, &mask);
-	writew(mask, ndev->reg_ofs.pdb_mask);
+	writew(mask, ndev->reg_ofs.ldb_mask);
 
 	return 0;
 }
@@ -173,9 +212,11 @@ void ntb_unregister_db_callback(struct ntb_device *ndev, unsigned int idx)
 	if (idx >= ndev->max_cbs || !ndev->db_cb[idx].callback)
 		return;
 
-	mask = readw(ndev->reg_ofs.pdb_mask);
+	mask = readw(ndev->reg_ofs.ldb_mask);
 	set_bit(idx * ndev->bits_per_vector, &mask);
-	writew(mask, ndev->reg_ofs.pdb_mask);
+	writew(mask, ndev->reg_ofs.ldb_mask);
+
+	tasklet_disable(&ndev->db_cb[idx].irq_work);
 
 	ndev->db_cb[idx].callback = NULL;
 }
@@ -334,6 +375,23 @@ int ntb_read_remote_spad(struct ntb_device *ndev, unsigned int idx, u32 *val)
 }
 
 /**
+ * ntb_get_mw_base() - get addr for the NTB memory window
+ * @ndev: pointer to ntb_device instance
+ * @mw: memory window number
+ *
+ * This function provides the base address of the memory window specified.
+ *
+ * RETURNS: address, or NULL on error.
+ */
+resource_size_t ntb_get_mw_base(struct ntb_device *ndev, unsigned int mw)
+{
+	if (mw >= ntb_max_mw(ndev))
+		return 0;
+
+	return pci_resource_start(ndev->pdev, MW_TO_BAR(mw));
+}
+
+/**
  * ntb_get_mw_vbase() - get virtual addr for the NTB memory window
  * @ndev: pointer to ntb_device instance
  * @mw: memory window number
@@ -345,7 +403,7 @@ int ntb_read_remote_spad(struct ntb_device *ndev, unsigned int idx, u32 *val)
  */
 void __iomem *ntb_get_mw_vbase(struct ntb_device *ndev, unsigned int mw)
 {
-	if (mw >= NTB_NUM_MW)
+	if (mw >= ntb_max_mw(ndev))
 		return NULL;
 
 	return ndev->mw[mw].vbase;
@@ -360,9 +418,9 @@ void __iomem *ntb_get_mw_vbase(struct ntb_device *ndev, unsigned int mw)
  *
  * RETURNS: the size of the memory window or zero on error
  */
-resource_size_t ntb_get_mw_size(struct ntb_device *ndev, unsigned int mw)
+u64 ntb_get_mw_size(struct ntb_device *ndev, unsigned int mw)
 {
-	if (mw >= NTB_NUM_MW)
+	if (mw >= ntb_max_mw(ndev))
 		return 0;
 
 	return ndev->mw[mw].bar_sz;
@@ -380,7 +438,7 @@ resource_size_t ntb_get_mw_size(struct ntb_device *ndev, unsigned int mw)
  */
 void ntb_set_mw_addr(struct ntb_device *ndev, unsigned int mw, u64 addr)
 {
-	if (mw >= NTB_NUM_MW)
+	if (mw >= ntb_max_mw(ndev))
 		return;
 
 	dev_dbg(&ndev->pdev->dev, "Writing addr %Lx to BAR %d\n", addr,
@@ -390,16 +448,16 @@ void ntb_set_mw_addr(struct ntb_device *ndev, unsigned int mw, u64 addr)
 
 	switch (MW_TO_BAR(mw)) {
 	case NTB_BAR_23:
-		writeq(addr, ndev->reg_ofs.sbar2_xlat);
+		writeq(addr, ndev->reg_ofs.bar2_xlat);
 		break;
 	case NTB_BAR_45:
-		writeq(addr, ndev->reg_ofs.sbar4_xlat);
+		writeq(addr, ndev->reg_ofs.bar4_xlat);
 		break;
 	}
 }
 
 /**
- * ntb_ring_sdb() - Set the doorbell on the secondary/external side
+ * ntb_ring_doorbell() - Set the doorbell on the secondary/external side
  * @ndev: pointer to ntb_device instance
  * @db: doorbell to ring
  *
@@ -408,15 +466,58 @@ void ntb_set_mw_addr(struct ntb_device *ndev, unsigned int mw, u64 addr)
  *
  * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
  */
-void ntb_ring_sdb(struct ntb_device *ndev, unsigned int db)
+void ntb_ring_doorbell(struct ntb_device *ndev, unsigned int db)
 {
 	dev_dbg(&ndev->pdev->dev, "%s: ringing doorbell %d\n", __func__, db);
 
 	if (ndev->hw_type == BWD_HW)
-		writeq((u64) 1 << db, ndev->reg_ofs.sdb);
+		writeq((u64) 1 << db, ndev->reg_ofs.rdb);
 	else
 		writew(((1 << ndev->bits_per_vector) - 1) <<
-		       (db * ndev->bits_per_vector), ndev->reg_ofs.sdb);
+		       (db * ndev->bits_per_vector), ndev->reg_ofs.rdb);
+}
+
+static void bwd_recover_link(struct ntb_device *ndev)
+{
+	u32 status;
+
+	/* Driver resets the NTB ModPhy lanes - magic! */
+	writeb(0xe0, ndev->reg_base + BWD_MODPHY_PCSREG6);
+	writeb(0x40, ndev->reg_base + BWD_MODPHY_PCSREG4);
+	writeb(0x60, ndev->reg_base + BWD_MODPHY_PCSREG4);
+	writeb(0x60, ndev->reg_base + BWD_MODPHY_PCSREG6);
+
+	/* Driver waits 100ms to allow the NTB ModPhy to settle */
+	msleep(100);
+
+	/* Clear AER Errors, write to clear */
+	status = readl(ndev->reg_base + BWD_ERRCORSTS_OFFSET);
+	dev_dbg(&ndev->pdev->dev, "ERRCORSTS = %x\n", status);
+	status &= PCI_ERR_COR_REP_ROLL;
+	writel(status, ndev->reg_base + BWD_ERRCORSTS_OFFSET);
+
+	/* Clear unexpected electrical idle event in LTSSM, write to clear */
+	status = readl(ndev->reg_base + BWD_LTSSMERRSTS0_OFFSET);
+	dev_dbg(&ndev->pdev->dev, "LTSSMERRSTS0 = %x\n", status);
+	status |= BWD_LTSSMERRSTS0_UNEXPECTEDEI;
+	writel(status, ndev->reg_base + BWD_LTSSMERRSTS0_OFFSET);
+
+	/* Clear DeSkew Buffer error, write to clear */
+	status = readl(ndev->reg_base + BWD_DESKEWSTS_OFFSET);
+	dev_dbg(&ndev->pdev->dev, "DESKEWSTS = %x\n", status);
+	status |= BWD_DESKEWSTS_DBERR;
+	writel(status, ndev->reg_base + BWD_DESKEWSTS_OFFSET);
+
+	status = readl(ndev->reg_base + BWD_IBSTERRRCRVSTS0_OFFSET);
+	dev_dbg(&ndev->pdev->dev, "IBSTERRRCRVSTS0 = %x\n", status);
+	status &= BWD_IBIST_ERR_OFLOW;
+	writel(status, ndev->reg_base + BWD_IBSTERRRCRVSTS0_OFFSET);
+
+	/* Releases the NTB state machine to allow the link to retrain */
+	status = readl(ndev->reg_base + BWD_LTSSMSTATEJMP_OFFSET);
+	dev_dbg(&ndev->pdev->dev, "LTSSMSTATEJMP = %x\n", status);
+	status &= ~BWD_LTSSMSTATEJMP_FORCEDETECT;
+	writel(status, ndev->reg_base + BWD_LTSSMSTATEJMP_OFFSET);
 }
 
 static void ntb_link_event(struct ntb_device *ndev, int link_state)
@@ -433,7 +534,8 @@ static void ntb_link_event(struct ntb_device *ndev, int link_state)
 		ndev->link_status = NTB_LINK_UP;
 		event = NTB_EVENT_HW_LINK_UP;
 
-		if (ndev->hw_type == BWD_HW)
+		if (ndev->hw_type == BWD_HW ||
+		    ndev->conn_type == NTB_CONN_TRANSPARENT)
 			status = readw(ndev->reg_ofs.lnk_stat);
 		else {
 			int rc = pci_read_config_word(ndev->pdev,
@@ -442,13 +544,16 @@ static void ntb_link_event(struct ntb_device *ndev, int link_state)
 			if (rc)
 				return;
 		}
+
+		ndev->link_width = (status & NTB_LINK_WIDTH_MASK) >> 4;
+		ndev->link_speed = (status & NTB_LINK_SPEED_MASK);
 		dev_info(&ndev->pdev->dev, "Link Width %d, Link Speed %d\n",
-			 (status & NTB_LINK_WIDTH_MASK) >> 4,
-			 (status & NTB_LINK_SPEED_MASK));
+			 ndev->link_width, ndev->link_speed);
 	} else {
 		dev_info(&ndev->pdev->dev, "Link Down\n");
 		ndev->link_status = NTB_LINK_DOWN;
 		event = NTB_EVENT_HW_LINK_DOWN;
+		/* Don't modify link width/speed, we need it in link recovery */
 	}
 
 	/* notify the upper layer if we have an event change */
@@ -488,6 +593,47 @@ static int ntb_link_status(struct ntb_device *ndev)
 	return 0;
 }
 
+static void bwd_link_recovery(struct work_struct *work)
+{
+	struct ntb_device *ndev = container_of(work, struct ntb_device,
+					       lr_timer.work);
+	u32 status32;
+
+	bwd_recover_link(ndev);
+	/* There is a potential race between the 2 NTB devices recovering at the
+	 * same time.  If the times are the same, the link will not recover and
+	 * the driver will be stuck in this loop forever.  Add a random interval
+	 * to the recovery time to prevent this race.
+	 */
+	msleep(BWD_LINK_RECOVERY_TIME + prandom_u32() % BWD_LINK_RECOVERY_TIME);
+
+	status32 = readl(ndev->reg_base + BWD_LTSSMSTATEJMP_OFFSET);
+	if (status32 & BWD_LTSSMSTATEJMP_FORCEDETECT)
+		goto retry;
+
+	status32 = readl(ndev->reg_base + BWD_IBSTERRRCRVSTS0_OFFSET);
+	if (status32 & BWD_IBIST_ERR_OFLOW)
+		goto retry;
+
+	status32 = readl(ndev->reg_ofs.lnk_cntl);
+	if (!(status32 & BWD_CNTL_LINK_DOWN)) {
+		unsigned char speed, width;
+		u16 status16;
+
+		status16 = readw(ndev->reg_ofs.lnk_stat);
+		width = (status16 & NTB_LINK_WIDTH_MASK) >> 4;
+		speed = (status16 & NTB_LINK_SPEED_MASK);
+		if (ndev->link_width != width || ndev->link_speed != speed)
+			goto retry;
+	}
+
+	schedule_delayed_work(&ndev->hb_timer, NTB_HB_TIMEOUT);
+	return;
+
+retry:
+	schedule_delayed_work(&ndev->lr_timer, NTB_HB_TIMEOUT);
+}
+
 /* BWD doesn't have link status interrupt, poll on that platform */
 static void bwd_link_poll(struct work_struct *work)
 {
@@ -503,6 +649,16 @@ static void bwd_link_poll(struct work_struct *work)
 		if (rc)
 			dev_err(&ndev->pdev->dev,
 				"Error determining link status\n");
+
+		/* Check to see if a link error is the cause of the link down */
+		if (ndev->link_status == NTB_LINK_DOWN) {
+			u32 status32 = readl(ndev->reg_base +
+					     BWD_LTSSMSTATEJMP_OFFSET);
+			if (status32 & BWD_LTSSMSTATEJMP_FORCEDETECT) {
+				schedule_delayed_work(&ndev->lr_timer, 0);
+				return;
+			}
+		}
 	}
 
 	schedule_delayed_work(&ndev->hb_timer, NTB_HB_TIMEOUT);
@@ -519,42 +675,196 @@ static int ntb_xeon_setup(struct ntb_device *ndev)
 	if (rc)
 		return rc;
 
+	if (val & SNB_PPD_DEV_TYPE)
+		ndev->dev_type = NTB_DEV_USD;
+	else
+		ndev->dev_type = NTB_DEV_DSD;
+
 	switch (val & SNB_PPD_CONN_TYPE) {
 	case NTB_CONN_B2B:
+		dev_info(&ndev->pdev->dev, "Conn Type = B2B\n");
 		ndev->conn_type = NTB_CONN_B2B;
+		ndev->reg_ofs.ldb = ndev->reg_base + SNB_PDOORBELL_OFFSET;
+		ndev->reg_ofs.ldb_mask = ndev->reg_base + SNB_PDBMSK_OFFSET;
+		ndev->reg_ofs.spad_read = ndev->reg_base + SNB_SPAD_OFFSET;
+		ndev->reg_ofs.bar2_xlat = ndev->reg_base + SNB_SBAR2XLAT_OFFSET;
+		ndev->reg_ofs.bar4_xlat = ndev->reg_base + SNB_SBAR4XLAT_OFFSET;
+		ndev->limits.max_spads = SNB_MAX_B2B_SPADS;
+
+		/* There is a Xeon hardware errata related to writes to
+		 * SDOORBELL or B2BDOORBELL in conjunction with inbound access
+		 * to NTB MMIO Space, which may hang the system.  To workaround
+		 * this use the second memory window to access the interrupt and
+		 * scratch pad registers on the remote system.
+		 */
+		if (xeon_errata_workaround) {
+			if (!ndev->mw[1].bar_sz)
+				return -EINVAL;
+
+			ndev->limits.max_mw = SNB_ERRATA_MAX_MW;
+			ndev->limits.max_db_bits = SNB_MAX_DB_BITS;
+			ndev->reg_ofs.spad_write = ndev->mw[1].vbase +
+						   SNB_SPAD_OFFSET;
+			ndev->reg_ofs.rdb = ndev->mw[1].vbase +
+					    SNB_PDOORBELL_OFFSET;
+
+			/* Set the Limit register to 4k, the minimum size, to
+			 * prevent an illegal access
+			 */
+			writeq(ndev->mw[1].bar_sz + 0x1000, ndev->reg_base +
+			       SNB_PBAR4LMT_OFFSET);
+			/* HW errata on the Limit registers.  They can only be
+			 * written when the base register is 4GB aligned and
+			 * < 32bit.  This should already be the case based on the
+			 * driver defaults, but write the Limit registers first
+			 * just in case.
+			 */
+		} else {
+			ndev->limits.max_mw = SNB_MAX_MW;
+
+			/* HW Errata on bit 14 of b2bdoorbell register.  Writes
+			 * will not be mirrored to the remote system.  Shrink
+			 * the number of bits by one, since bit 14 is the last
+			 * bit.
+			 */
+			ndev->limits.max_db_bits = SNB_MAX_DB_BITS - 1;
+			ndev->reg_ofs.spad_write = ndev->reg_base +
+						   SNB_B2B_SPAD_OFFSET;
+			ndev->reg_ofs.rdb = ndev->reg_base +
+					    SNB_B2B_DOORBELL_OFFSET;
+
+			/* Disable the Limit register, just incase it is set to
+			 * something silly
+			 */
+			writeq(0, ndev->reg_base + SNB_PBAR4LMT_OFFSET);
+			/* HW errata on the Limit registers.  They can only be
+			 * written when the base register is 4GB aligned and
+			 * < 32bit.  This should already be the case based on the
+			 * driver defaults, but write the Limit registers first
+			 * just in case.
+			 */
+		}
+
+		/* The Xeon errata workaround requires setting SBAR Base
+		 * addresses to known values, so that the PBAR XLAT can be
+		 * pointed at SBAR0 of the remote system.
+		 */
+		if (ndev->dev_type == NTB_DEV_USD) {
+			writeq(SNB_MBAR23_DSD_ADDR, ndev->reg_base +
+			       SNB_PBAR2XLAT_OFFSET);
+			if (xeon_errata_workaround)
+				writeq(SNB_MBAR01_DSD_ADDR, ndev->reg_base +
+				       SNB_PBAR4XLAT_OFFSET);
+			else {
+				writeq(SNB_MBAR45_DSD_ADDR, ndev->reg_base +
+				       SNB_PBAR4XLAT_OFFSET);
+				/* B2B_XLAT_OFFSET is a 64bit register, but can
+				 * only take 32bit writes
+				 */
+				writel(SNB_MBAR01_DSD_ADDR & 0xffffffff,
+				       ndev->reg_base + SNB_B2B_XLAT_OFFSETL);
+				writel(SNB_MBAR01_DSD_ADDR >> 32,
+				       ndev->reg_base + SNB_B2B_XLAT_OFFSETU);
+			}
+
+			writeq(SNB_MBAR01_USD_ADDR, ndev->reg_base +
+			       SNB_SBAR0BASE_OFFSET);
+			writeq(SNB_MBAR23_USD_ADDR, ndev->reg_base +
+			       SNB_SBAR2BASE_OFFSET);
+			writeq(SNB_MBAR45_USD_ADDR, ndev->reg_base +
+			       SNB_SBAR4BASE_OFFSET);
+		} else {
+			writeq(SNB_MBAR23_USD_ADDR, ndev->reg_base +
+			       SNB_PBAR2XLAT_OFFSET);
+			if (xeon_errata_workaround)
+				writeq(SNB_MBAR01_USD_ADDR, ndev->reg_base +
+				       SNB_PBAR4XLAT_OFFSET);
+			else {
+				writeq(SNB_MBAR45_USD_ADDR, ndev->reg_base +
+				       SNB_PBAR4XLAT_OFFSET);
+				/* B2B_XLAT_OFFSET is a 64bit register, but can
+				 * only take 32bit writes
+				 */
+				writel(SNB_MBAR01_DSD_ADDR & 0xffffffff,
+				       ndev->reg_base + SNB_B2B_XLAT_OFFSETL);
+				writel(SNB_MBAR01_USD_ADDR >> 32,
+				       ndev->reg_base + SNB_B2B_XLAT_OFFSETU);
+			}
+			writeq(SNB_MBAR01_DSD_ADDR, ndev->reg_base +
+			       SNB_SBAR0BASE_OFFSET);
+			writeq(SNB_MBAR23_DSD_ADDR, ndev->reg_base +
+			       SNB_SBAR2BASE_OFFSET);
+			writeq(SNB_MBAR45_DSD_ADDR, ndev->reg_base +
+			       SNB_SBAR4BASE_OFFSET);
+		}
 		break;
-	case NTB_CONN_CLASSIC:
 	case NTB_CONN_RP:
+		dev_info(&ndev->pdev->dev, "Conn Type = RP\n");
+		ndev->conn_type = NTB_CONN_RP;
+
+		if (xeon_errata_workaround) {
+			dev_err(&ndev->pdev->dev, 
+				"NTB-RP disabled due to hardware errata.  To disregard this warning and potentially lock-up the system, add the parameter 'xeon_errata_workaround=0'.\n");
+			return -EINVAL;
+		}
+
+		/* Scratch pads need to have exclusive access from the primary
+		 * or secondary side.  Halve the num spads so that each side can
+		 * have an equal amount.
+		 */
+		ndev->limits.max_spads = SNB_MAX_COMPAT_SPADS / 2;
+		ndev->limits.max_db_bits = SNB_MAX_DB_BITS;
+		/* Note: The SDOORBELL is the cause of the errata.  You REALLY
+		 * don't want to touch it.
+		 */
+		ndev->reg_ofs.rdb = ndev->reg_base + SNB_SDOORBELL_OFFSET;
+		ndev->reg_ofs.ldb = ndev->reg_base + SNB_PDOORBELL_OFFSET;
+		ndev->reg_ofs.ldb_mask = ndev->reg_base + SNB_PDBMSK_OFFSET;
+		/* Offset the start of the spads to correspond to whether it is
+		 * primary or secondary
+		 */
+		ndev->reg_ofs.spad_write = ndev->reg_base + SNB_SPAD_OFFSET +
+					   ndev->limits.max_spads * 4;
+		ndev->reg_ofs.spad_read = ndev->reg_base + SNB_SPAD_OFFSET;
+		ndev->reg_ofs.bar2_xlat = ndev->reg_base + SNB_SBAR2XLAT_OFFSET;
+		ndev->reg_ofs.bar4_xlat = ndev->reg_base + SNB_SBAR4XLAT_OFFSET;
+		ndev->limits.max_mw = SNB_MAX_MW;
+		break;
+	case NTB_CONN_TRANSPARENT:
+		dev_info(&ndev->pdev->dev, "Conn Type = TRANSPARENT\n");
+		ndev->conn_type = NTB_CONN_TRANSPARENT;
+		/* Scratch pads need to have exclusive access from the primary
+		 * or secondary side.  Halve the num spads so that each side can
+		 * have an equal amount.
+		 */
+		ndev->limits.max_spads = SNB_MAX_COMPAT_SPADS / 2;
+		ndev->limits.max_db_bits = SNB_MAX_DB_BITS;
+		ndev->reg_ofs.rdb = ndev->reg_base + SNB_PDOORBELL_OFFSET;
+		ndev->reg_ofs.ldb = ndev->reg_base + SNB_SDOORBELL_OFFSET;
+		ndev->reg_ofs.ldb_mask = ndev->reg_base + SNB_SDBMSK_OFFSET;
+		ndev->reg_ofs.spad_write = ndev->reg_base + SNB_SPAD_OFFSET;
+		/* Offset the start of the spads to correspond to whether it is
+		 * primary or secondary
+		 */
+		ndev->reg_ofs.spad_read = ndev->reg_base + SNB_SPAD_OFFSET +
+					  ndev->limits.max_spads * 4;
+		ndev->reg_ofs.bar2_xlat = ndev->reg_base + SNB_PBAR2XLAT_OFFSET;
+		ndev->reg_ofs.bar4_xlat = ndev->reg_base + SNB_PBAR4XLAT_OFFSET;
+
+		ndev->limits.max_mw = SNB_MAX_MW;
+		break;
 	default:
-		dev_err(&ndev->pdev->dev, "Only B2B supported at this time\n");
+		/* Most likely caused by the remote NTB-RP device not being
+		 * configured
+		 */
+		dev_err(&ndev->pdev->dev, "Unknown PPD %x\n", val);
 		return -EINVAL;
 	}
 
-	if (val & SNB_PPD_DEV_TYPE)
-		ndev->dev_type = NTB_DEV_DSD;
-	else
-		ndev->dev_type = NTB_DEV_USD;
-
-	ndev->reg_ofs.pdb = ndev->reg_base + SNB_PDOORBELL_OFFSET;
-	ndev->reg_ofs.pdb_mask = ndev->reg_base + SNB_PDBMSK_OFFSET;
-	ndev->reg_ofs.sbar2_xlat = ndev->reg_base + SNB_SBAR2XLAT_OFFSET;
-	ndev->reg_ofs.sbar4_xlat = ndev->reg_base + SNB_SBAR4XLAT_OFFSET;
 	ndev->reg_ofs.lnk_cntl = ndev->reg_base + SNB_NTBCNTL_OFFSET;
-	ndev->reg_ofs.lnk_stat = ndev->reg_base + SNB_LINK_STATUS_OFFSET;
-	ndev->reg_ofs.spad_read = ndev->reg_base + SNB_SPAD_OFFSET;
+	ndev->reg_ofs.lnk_stat = ndev->reg_base + SNB_SLINK_STATUS_OFFSET;
 	ndev->reg_ofs.spci_cmd = ndev->reg_base + SNB_PCICMD_OFFSET;
 
-	if (ndev->conn_type == NTB_CONN_B2B) {
-		ndev->reg_ofs.sdb = ndev->reg_base + SNB_B2B_DOORBELL_OFFSET;
-		ndev->reg_ofs.spad_write = ndev->reg_base + SNB_B2B_SPAD_OFFSET;
-		ndev->limits.max_spads = SNB_MAX_SPADS;
-	} else {
-		ndev->reg_ofs.sdb = ndev->reg_base + SNB_SDOORBELL_OFFSET;
-		ndev->reg_ofs.spad_write = ndev->reg_base + SNB_SPAD_OFFSET;
-		ndev->limits.max_spads = SNB_MAX_COMPAT_SPADS;
-	}
-
-	ndev->limits.max_db_bits = SNB_MAX_DB_BITS;
 	ndev->limits.msix_cnt = SNB_MSIX_CNT;
 	ndev->bits_per_vector = SNB_DB_BITS_PER_VEC;
 
@@ -578,7 +888,7 @@ static int ntb_bwd_setup(struct ntb_device *ndev)
 		break;
 	case NTB_CONN_RP:
 	default:
-		dev_err(&ndev->pdev->dev, "Only B2B supported at this time\n");
+		dev_err(&ndev->pdev->dev, "Unsupported NTB configuration\n");
 		return -EINVAL;
 	}
 
@@ -593,31 +903,25 @@ static int ntb_bwd_setup(struct ntb_device *ndev)
 	if (rc)
 		return rc;
 
-	ndev->reg_ofs.pdb = ndev->reg_base + BWD_PDOORBELL_OFFSET;
-	ndev->reg_ofs.pdb_mask = ndev->reg_base + BWD_PDBMSK_OFFSET;
-	ndev->reg_ofs.sbar2_xlat = ndev->reg_base + BWD_SBAR2XLAT_OFFSET;
-	ndev->reg_ofs.sbar4_xlat = ndev->reg_base + BWD_SBAR4XLAT_OFFSET;
+	ndev->reg_ofs.ldb = ndev->reg_base + BWD_PDOORBELL_OFFSET;
+	ndev->reg_ofs.ldb_mask = ndev->reg_base + BWD_PDBMSK_OFFSET;
+	ndev->reg_ofs.rdb = ndev->reg_base + BWD_B2B_DOORBELL_OFFSET;
+	ndev->reg_ofs.bar2_xlat = ndev->reg_base + BWD_SBAR2XLAT_OFFSET;
+	ndev->reg_ofs.bar4_xlat = ndev->reg_base + BWD_SBAR4XLAT_OFFSET;
 	ndev->reg_ofs.lnk_cntl = ndev->reg_base + BWD_NTBCNTL_OFFSET;
 	ndev->reg_ofs.lnk_stat = ndev->reg_base + BWD_LINK_STATUS_OFFSET;
 	ndev->reg_ofs.spad_read = ndev->reg_base + BWD_SPAD_OFFSET;
+	ndev->reg_ofs.spad_write = ndev->reg_base + BWD_B2B_SPAD_OFFSET;
 	ndev->reg_ofs.spci_cmd = ndev->reg_base + BWD_PCICMD_OFFSET;
-
-	if (ndev->conn_type == NTB_CONN_B2B) {
-		ndev->reg_ofs.sdb = ndev->reg_base + BWD_B2B_DOORBELL_OFFSET;
-		ndev->reg_ofs.spad_write = ndev->reg_base + BWD_B2B_SPAD_OFFSET;
-		ndev->limits.max_spads = BWD_MAX_SPADS;
-	} else {
-		ndev->reg_ofs.sdb = ndev->reg_base + BWD_PDOORBELL_OFFSET;
-		ndev->reg_ofs.spad_write = ndev->reg_base + BWD_SPAD_OFFSET;
-		ndev->limits.max_spads = BWD_MAX_COMPAT_SPADS;
-	}
-
+	ndev->limits.max_mw = BWD_MAX_MW;
+	ndev->limits.max_spads = BWD_MAX_SPADS;
 	ndev->limits.max_db_bits = BWD_MAX_DB_BITS;
 	ndev->limits.msix_cnt = BWD_MSIX_CNT;
 	ndev->bits_per_vector = BWD_DB_BITS_PER_VEC;
 
 	/* Since bwd doesn't have a link interrupt, setup a poll timer */
 	INIT_DELAYED_WORK(&ndev->hb_timer, bwd_link_poll);
+	INIT_DELAYED_WORK(&ndev->lr_timer, bwd_link_recovery);
 	schedule_delayed_work(&ndev->hb_timer, NTB_HB_TIMEOUT);
 
 	return 0;
@@ -628,13 +932,18 @@ static int ntb_device_setup(struct ntb_device *ndev)
 	int rc;
 
 	switch (ndev->pdev->device) {
-	case PCI_DEVICE_ID_INTEL_NTB_2ND_SNB:
-	case PCI_DEVICE_ID_INTEL_NTB_RP_JSF:
-	case PCI_DEVICE_ID_INTEL_NTB_RP_SNB:
-	case PCI_DEVICE_ID_INTEL_NTB_CLASSIC_JSF:
-	case PCI_DEVICE_ID_INTEL_NTB_CLASSIC_SNB:
+	case PCI_DEVICE_ID_INTEL_NTB_SS_JSF:
+	case PCI_DEVICE_ID_INTEL_NTB_SS_SNB:
+	case PCI_DEVICE_ID_INTEL_NTB_SS_IVT:
+	case PCI_DEVICE_ID_INTEL_NTB_SS_HSX:
+	case PCI_DEVICE_ID_INTEL_NTB_PS_JSF:
+	case PCI_DEVICE_ID_INTEL_NTB_PS_SNB:
+	case PCI_DEVICE_ID_INTEL_NTB_PS_IVT:
+	case PCI_DEVICE_ID_INTEL_NTB_PS_HSX:
 	case PCI_DEVICE_ID_INTEL_NTB_B2B_JSF:
 	case PCI_DEVICE_ID_INTEL_NTB_B2B_SNB:
+	case PCI_DEVICE_ID_INTEL_NTB_B2B_IVT:
+	case PCI_DEVICE_ID_INTEL_NTB_B2B_HSX:
 		rc = ntb_xeon_setup(ndev);
 		break;
 	case PCI_DEVICE_ID_INTEL_NTB_B2B_BWD:
@@ -644,35 +953,49 @@ static int ntb_device_setup(struct ntb_device *ndev)
 		rc = -ENODEV;
 	}
 
-	/* Enable Bus Master and Memory Space on the secondary side */
-	writew(PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER, ndev->reg_ofs.spci_cmd);
+	if (rc)
+		return rc;
 
-	return rc;
+	dev_info(&ndev->pdev->dev, "Device Type = %s\n",
+		 ndev->dev_type == NTB_DEV_USD ? "USD/DSP" : "DSD/USP");
+
+	if (ndev->conn_type == NTB_CONN_B2B)
+		/* Enable Bus Master and Memory Space on the secondary side */
+		writew(PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER,
+		       ndev->reg_ofs.spci_cmd);
+
+	return 0;
 }
 
 static void ntb_device_free(struct ntb_device *ndev)
 {
-	if (ndev->hw_type == BWD_HW)
+	if (ndev->hw_type == BWD_HW) {
 		cancel_delayed_work_sync(&ndev->hb_timer);
+		cancel_delayed_work_sync(&ndev->lr_timer);
+	}
 }
 
 static irqreturn_t bwd_callback_msix_irq(int irq, void *data)
 {
 	struct ntb_db_cb *db_cb = data;
 	struct ntb_device *ndev = db_cb->ndev;
+	unsigned long mask;
 
 	dev_dbg(&ndev->pdev->dev, "MSI-X irq %d received for DB %d\n", irq,
 		db_cb->db_num);
 
-	if (db_cb->callback)
-		db_cb->callback(db_cb->data, db_cb->db_num);
+	mask = readw(ndev->reg_ofs.ldb_mask);
+	set_bit(db_cb->db_num * ndev->bits_per_vector, &mask);
+	writew(mask, ndev->reg_ofs.ldb_mask);
+
+	tasklet_schedule(&db_cb->irq_work);
 
 	/* No need to check for the specific HB irq, any interrupt means
 	 * we're connected.
 	 */
 	ndev->last_ts = jiffies;
 
-	writeq((u64) 1 << db_cb->db_num, ndev->reg_ofs.pdb);
+	writeq((u64) 1 << db_cb->db_num, ndev->reg_ofs.ldb);
 
 	return IRQ_HANDLED;
 }
@@ -681,12 +1004,16 @@ static irqreturn_t xeon_callback_msix_irq(int irq, void *data)
 {
 	struct ntb_db_cb *db_cb = data;
 	struct ntb_device *ndev = db_cb->ndev;
+	unsigned long mask;
 
 	dev_dbg(&ndev->pdev->dev, "MSI-X irq %d received for DB %d\n", irq,
 		db_cb->db_num);
 
-	if (db_cb->callback)
-		db_cb->callback(db_cb->data, db_cb->db_num);
+	mask = readw(ndev->reg_ofs.ldb_mask);
+	set_bit(db_cb->db_num * ndev->bits_per_vector, &mask);
+	writew(mask, ndev->reg_ofs.ldb_mask);
+
+	tasklet_schedule(&db_cb->irq_work);
 
 	/* On Sandybridge, there are 16 bits in the interrupt register
 	 * but only 4 vectors.  So, 5 bits are assigned to the first 3
@@ -694,7 +1021,7 @@ static irqreturn_t xeon_callback_msix_irq(int irq, void *data)
 	 * interrupts.
 	 */
 	writew(((1 << ndev->bits_per_vector) - 1) <<
-	       (db_cb->db_num * ndev->bits_per_vector), ndev->reg_ofs.pdb);
+	       (db_cb->db_num * ndev->bits_per_vector), ndev->reg_ofs.ldb);
 
 	return IRQ_HANDLED;
 }
@@ -712,7 +1039,7 @@ static irqreturn_t xeon_event_msix_irq(int irq, void *dev)
 		dev_err(&ndev->pdev->dev, "Error determining link status\n");
 
 	/* bit 15 is always the link bit */
-	writew(1 << ndev->limits.max_db_bits, ndev->reg_ofs.pdb);
+	writew(1 << SNB_LINK_DB, ndev->reg_ofs.ldb);
 
 	return IRQ_HANDLED;
 }
@@ -723,29 +1050,28 @@ static irqreturn_t ntb_interrupt(int irq, void *dev)
 	unsigned int i = 0;
 
 	if (ndev->hw_type == BWD_HW) {
-		u64 pdb = readq(ndev->reg_ofs.pdb);
+		u64 ldb = readq(ndev->reg_ofs.ldb);
 
-		dev_dbg(&ndev->pdev->dev, "irq %d - pdb = %Lx\n", irq, pdb);
+		dev_dbg(&ndev->pdev->dev, "irq %d - ldb = %Lx\n", irq, ldb);
 
-		while (pdb) {
-			i = __ffs(pdb);
-			pdb &= pdb - 1;
+		while (ldb) {
+			i = __ffs(ldb);
+			ldb &= ldb - 1;
 			bwd_callback_msix_irq(irq, &ndev->db_cb[i]);
 		}
 	} else {
-		u16 pdb = readw(ndev->reg_ofs.pdb);
+		u16 ldb = readw(ndev->reg_ofs.ldb);
 
-		dev_dbg(&ndev->pdev->dev, "irq %d - pdb = %x sdb %x\n", irq,
-			pdb, readw(ndev->reg_ofs.sdb));
+		dev_dbg(&ndev->pdev->dev, "irq %d - ldb = %x\n", irq, ldb);
 
-		if (pdb & SNB_DB_HW_LINK) {
+		if (ldb & SNB_DB_HW_LINK) {
 			xeon_event_msix_irq(irq, dev);
-			pdb &= ~SNB_DB_HW_LINK;
+			ldb &= ~SNB_DB_HW_LINK;
 		}
 
-		while (pdb) {
-			i = __ffs(pdb);
-			pdb &= pdb - 1;
+		while (ldb) {
+			i = __ffs(ldb);
+			ldb &= ldb - 1;
 			xeon_callback_msix_irq(irq, &ndev->db_cb[i]);
 		}
 	}
@@ -758,16 +1084,15 @@ static int ntb_setup_msix(struct ntb_device *ndev)
 	struct pci_dev *pdev = ndev->pdev;
 	struct msix_entry *msix;
 	int msix_entries;
-	int rc, i, pos;
+	int rc, i;
 	u16 val;
 
-	pos = pci_find_capability(pdev, PCI_CAP_ID_MSIX);
-	if (!pos) {
+	if (!pdev->msix_cap) {
 		rc = -EIO;
 		goto err;
 	}
 
-	rc = pci_read_config_word(pdev, pos + PCI_MSIX_FLAGS, &val);
+	rc = pci_read_config_word(pdev, pdev->msix_cap + PCI_MSIX_FLAGS, &val);
 	if (rc)
 		goto err;
 
@@ -803,6 +1128,10 @@ static int ntb_setup_msix(struct ntb_device *ndev)
 			 "Only %d MSI-X vectors.  Limiting the number of queues to that number.\n",
 			 rc);
 		msix_entries = rc;
+
+		rc = pci_enable_msix(pdev, ndev->msix_entries, msix_entries);
+		if (rc)
+			goto err1;
 	}
 
 	for (i = 0; i < msix_entries; i++) {
@@ -903,10 +1232,11 @@ static int ntb_setup_interrupts(struct ntb_device *ndev)
 	 * Interrupt.  The rest will be unmasked as callbacks are registered.
 	 */
 	if (ndev->hw_type == BWD_HW)
-		writeq(~0, ndev->reg_ofs.pdb_mask);
-	else
-		writew(~(1 << ndev->limits.max_db_bits),
-		       ndev->reg_ofs.pdb_mask);
+		writeq(~0, ndev->reg_ofs.ldb_mask);
+	else {
+		u16 var = 1 << SNB_LINK_DB;
+		writew(~var, ndev->reg_ofs.ldb_mask);
+	}
 
 	rc = ntb_setup_msix(ndev);
 	if (!rc)
@@ -935,9 +1265,9 @@ static void ntb_free_interrupts(struct ntb_device *ndev)
 
 	/* mask interrupts */
 	if (ndev->hw_type == BWD_HW)
-		writeq(~0, ndev->reg_ofs.pdb_mask);
+		writeq(~0, ndev->reg_ofs.ldb_mask);
 	else
-		writew(~0, ndev->reg_ofs.pdb_mask);
+		writew(~0, ndev->reg_ofs.ldb_mask);
 
 	if (ndev->num_msix) {
 		struct msix_entry *msix;
@@ -963,9 +1293,9 @@ static int ntb_create_callbacks(struct ntb_device *ndev)
 {
 	int i;
 
-	/* Checken-egg issue.  We won't know how many callbacks are necessary
+	/* Chicken-egg issue.  We won't know how many callbacks are necessary
 	 * until we see how many MSI-X vectors we get, but these pointers need
-	 * to be passed into the MSI-X register fucntion.  So, we allocate the
+	 * to be passed into the MSI-X register function.  So, we allocate the
 	 * max, knowing that they might not all be used, to work around this.
 	 */
 	ndev->db_cb = kcalloc(ndev->limits.max_db_bits,
@@ -992,6 +1322,61 @@ static void ntb_free_callbacks(struct ntb_device *ndev)
 	kfree(ndev->db_cb);
 }
 
+static void ntb_setup_debugfs(struct ntb_device *ndev)
+{
+	if (!debugfs_initialized())
+		return;
+
+	if (!debugfs_dir)
+		debugfs_dir = debugfs_create_dir(KBUILD_MODNAME, NULL);
+
+	ndev->debugfs_dir = debugfs_create_dir(pci_name(ndev->pdev),
+					       debugfs_dir);
+}
+
+static void ntb_free_debugfs(struct ntb_device *ndev)
+{
+	debugfs_remove_recursive(ndev->debugfs_dir);
+
+	if (debugfs_dir && simple_empty(debugfs_dir)) {
+		debugfs_remove_recursive(debugfs_dir);
+		debugfs_dir = NULL;
+	}
+}
+
+static void ntb_hw_link_up(struct ntb_device *ndev)
+{
+	if (ndev->conn_type == NTB_CONN_TRANSPARENT)
+		ntb_link_event(ndev, NTB_LINK_UP);
+	else {
+		u32 ntb_cntl;
+
+		/* Let's bring the NTB link up */
+		ntb_cntl = readl(ndev->reg_ofs.lnk_cntl);
+		ntb_cntl &= ~(NTB_CNTL_LINK_DISABLE | NTB_CNTL_CFG_LOCK);
+		ntb_cntl |= NTB_CNTL_P2S_BAR23_SNOOP | NTB_CNTL_S2P_BAR23_SNOOP;
+		ntb_cntl |= NTB_CNTL_P2S_BAR45_SNOOP | NTB_CNTL_S2P_BAR45_SNOOP;
+		writel(ntb_cntl, ndev->reg_ofs.lnk_cntl);
+	}
+}
+
+static void ntb_hw_link_down(struct ntb_device *ndev)
+{
+	u32 ntb_cntl;
+
+	if (ndev->conn_type == NTB_CONN_TRANSPARENT) {
+		ntb_link_event(ndev, NTB_LINK_DOWN);
+		return;
+	}
+
+	/* Bring NTB link down */
+	ntb_cntl = readl(ndev->reg_ofs.lnk_cntl);
+	ntb_cntl &= ~(NTB_CNTL_P2S_BAR23_SNOOP | NTB_CNTL_S2P_BAR23_SNOOP);
+	ntb_cntl &= ~(NTB_CNTL_P2S_BAR45_SNOOP | NTB_CNTL_S2P_BAR45_SNOOP);
+	ntb_cntl |= NTB_CNTL_LINK_DISABLE | NTB_CNTL_CFG_LOCK;
+	writel(ntb_cntl, ndev->reg_ofs.lnk_cntl);
+}
+
 static int ntb_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 {
 	struct ntb_device *ndev;
@@ -1004,6 +1389,7 @@ static int ntb_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 	ndev->pdev = pdev;
 	ndev->link_status = NTB_LINK_DOWN;
 	pci_set_drvdata(pdev, ndev);
+	ntb_setup_debugfs(ndev);
 
 	rc = pci_enable_device(pdev);
 	if (rc)
@@ -1022,13 +1408,13 @@ static int ntb_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 		goto err2;
 	}
 
-	for (i = 0; i < NTB_NUM_MW; i++) {
+	for (i = 0; i < NTB_MAX_NUM_MW; i++) {
 		ndev->mw[i].bar_sz = pci_resource_len(pdev, MW_TO_BAR(i));
 		ndev->mw[i].vbase =
 		    ioremap_wc(pci_resource_start(pdev, MW_TO_BAR(i)),
 			       ndev->mw[i].bar_sz);
 		dev_info(&pdev->dev, "MW %d size %llu\n", i,
-			 pci_resource_len(pdev, MW_TO_BAR(i)));
+			 (unsigned long long) ndev->mw[i].bar_sz);
 		if (!ndev->mw[i].vbase) {
 			dev_warn(&pdev->dev, "Cannot remap BAR %d\n",
 				 MW_TO_BAR(i));
@@ -1079,9 +1465,7 @@ static int ntb_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
 	if (rc)
 		goto err6;
 
-	/* Let's bring the NTB link up */
-	writel(NTB_CNTL_BAR23_SNOOP | NTB_CNTL_BAR45_SNOOP,
-	       ndev->reg_ofs.lnk_cntl);
+	ntb_hw_link_up(ndev);
 
 	return 0;
 
@@ -1100,6 +1484,7 @@ err2:
 err1:
 	pci_disable_device(pdev);
 err:
+	ntb_free_debugfs(ndev);
 	kfree(ndev);
 
 	dev_err(&pdev->dev, "Error loading %s module\n", KBUILD_MODNAME);
@@ -1110,12 +1495,8 @@ static void ntb_pci_remove(struct pci_dev *pdev)
 {
 	struct ntb_device *ndev = pci_get_drvdata(pdev);
 	int i;
-	u32 ntb_cntl;
 
-	/* Bring NTB link down */
-	ntb_cntl = readl(ndev->reg_ofs.lnk_cntl);
-	ntb_cntl |= NTB_LINK_DISABLE;
-	writel(ntb_cntl, ndev->reg_ofs.lnk_cntl);
+	ntb_hw_link_down(ndev);
 
 	ntb_transport_free(ndev->ntb_transport);
 
@@ -1123,12 +1504,13 @@ static void ntb_pci_remove(struct pci_dev *pdev)
 	ntb_free_callbacks(ndev);
 	ntb_device_free(ndev);
 
-	for (i = 0; i < NTB_NUM_MW; i++)
+	for (i = 0; i < NTB_MAX_NUM_MW; i++)
 		iounmap(ndev->mw[i].vbase);
 
 	iounmap(ndev->reg_base);
 	pci_release_selected_regions(pdev, NTB_BAR_MASK);
 	pci_disable_device(pdev);
+	ntb_free_debugfs(ndev);
 	kfree(ndev);
 }
 
diff --git a/drivers/ntb/ntb_hw.h b/drivers/ntb/ntb_hw.h
index 3a3038ca83e6..bbdb7edca10c 100644
--- a/drivers/ntb/ntb_hw.h
+++ b/drivers/ntb/ntb_hw.h
@@ -47,16 +47,36 @@
  */
 
 #define PCI_DEVICE_ID_INTEL_NTB_B2B_JSF		0x3725
-#define PCI_DEVICE_ID_INTEL_NTB_CLASSIC_JSF	0x3726
-#define PCI_DEVICE_ID_INTEL_NTB_RP_JSF		0x3727
-#define PCI_DEVICE_ID_INTEL_NTB_RP_SNB		0x3C08
+#define PCI_DEVICE_ID_INTEL_NTB_PS_JSF		0x3726
+#define PCI_DEVICE_ID_INTEL_NTB_SS_JSF		0x3727
 #define PCI_DEVICE_ID_INTEL_NTB_B2B_SNB		0x3C0D
-#define PCI_DEVICE_ID_INTEL_NTB_CLASSIC_SNB	0x3C0E
-#define PCI_DEVICE_ID_INTEL_NTB_2ND_SNB		0x3C0F
+#define PCI_DEVICE_ID_INTEL_NTB_PS_SNB		0x3C0E
+#define PCI_DEVICE_ID_INTEL_NTB_SS_SNB		0x3C0F
+#define PCI_DEVICE_ID_INTEL_NTB_B2B_IVT		0x0E0D
+#define PCI_DEVICE_ID_INTEL_NTB_PS_IVT		0x0E0E
+#define PCI_DEVICE_ID_INTEL_NTB_SS_IVT		0x0E0F
+#define PCI_DEVICE_ID_INTEL_NTB_B2B_HSX		0x2F0D
+#define PCI_DEVICE_ID_INTEL_NTB_PS_HSX		0x2F0E
+#define PCI_DEVICE_ID_INTEL_NTB_SS_HSX		0x2F0F
 #define PCI_DEVICE_ID_INTEL_NTB_B2B_BWD		0x0C4E
 
 #define msix_table_size(control)	((control & PCI_MSIX_FLAGS_QSIZE)+1)
 
+#ifndef readq
+static inline u64 readq(void __iomem *addr)
+{
+	return readl(addr) | (((u64) readl(addr + 4)) << 32LL);
+}
+#endif
+
+#ifndef writeq
+static inline void writeq(u64 val, void __iomem *addr)
+{
+	writel(val & 0xffffffff, addr);
+	writel(val >> 32, addr + 4);
+}
+#endif
+
 #define NTB_BAR_MMIO		0
 #define NTB_BAR_23		2
 #define NTB_BAR_45		4
@@ -68,7 +88,7 @@
 
 #define NTB_HB_TIMEOUT		msecs_to_jiffies(1000)
 
-#define NTB_NUM_MW		2
+#define NTB_MAX_NUM_MW		2
 
 enum ntb_hw_event {
 	NTB_EVENT_SW_EVENT0 = 0,
@@ -86,28 +106,30 @@ struct ntb_mw {
 };
 
 struct ntb_db_cb {
-	void (*callback) (void *data, int db_num);
+	int (*callback)(void *data, int db_num);
 	unsigned int db_num;
 	void *data;
 	struct ntb_device *ndev;
+	struct tasklet_struct irq_work;
 };
 
 struct ntb_device {
 	struct pci_dev *pdev;
 	struct msix_entry *msix_entries;
 	void __iomem *reg_base;
-	struct ntb_mw mw[NTB_NUM_MW];
+	struct ntb_mw mw[NTB_MAX_NUM_MW];
 	struct {
-		unsigned int max_spads;
-		unsigned int max_db_bits;
-		unsigned int msix_cnt;
+		unsigned char max_mw;
+		unsigned char max_spads;
+		unsigned char max_db_bits;
+		unsigned char msix_cnt;
 	} limits;
 	struct {
-		void __iomem *pdb;
-		void __iomem *pdb_mask;
-		void __iomem *sdb;
-		void __iomem *sbar2_xlat;
-		void __iomem *sbar4_xlat;
+		void __iomem *ldb;
+		void __iomem *ldb_mask;
+		void __iomem *rdb;
+		void __iomem *bar2_xlat;
+		void __iomem *bar4_xlat;
 		void __iomem *spad_write;
 		void __iomem *spad_read;
 		void __iomem *lnk_cntl;
@@ -124,12 +146,45 @@ struct ntb_device {
 	unsigned char num_msix;
 	unsigned char bits_per_vector;
 	unsigned char max_cbs;
+	unsigned char link_width;
+	unsigned char link_speed;
 	unsigned char link_status;
+
 	struct delayed_work hb_timer;
 	unsigned long last_ts;
+
+	struct delayed_work lr_timer;
+
+	struct dentry *debugfs_dir;
 };
 
 /**
+ * ntb_max_cbs() - return the max callbacks
+ * @ndev: pointer to ntb_device instance
+ *
+ * Given the ntb pointer, return the maximum number of callbacks
+ *
+ * RETURNS: the maximum number of callbacks
+ */
+static inline unsigned char ntb_max_cbs(struct ntb_device *ndev)
+{
+	return ndev->max_cbs;
+}
+
+/**
+ * ntb_max_mw() - return the max number of memory windows
+ * @ndev: pointer to ntb_device instance
+ *
+ * Given the ntb pointer, return the maximum number of memory windows
+ *
+ * RETURNS: the maximum number of memory windows
+ */
+static inline unsigned char ntb_max_mw(struct ntb_device *ndev)
+{
+	return ndev->limits.max_mw;
+}
+
+/**
  * ntb_hw_link_status() - return the hardware link status
  * @ndev: pointer to ntb_device instance
  *
@@ -146,7 +201,7 @@ static inline bool ntb_hw_link_status(struct ntb_device *ndev)
  * ntb_query_pdev() - return the pci_dev pointer
  * @ndev: pointer to ntb_device instance
  *
- * Given the ntb pointer return the pci_dev pointerfor the NTB hardware device
+ * Given the ntb pointer, return the pci_dev pointer for the NTB hardware device
  *
  * RETURNS: a pointer to the ntb pci_dev
  */
@@ -155,13 +210,27 @@ static inline struct pci_dev *ntb_query_pdev(struct ntb_device *ndev)
 	return ndev->pdev;
 }
 
+/**
+ * ntb_query_debugfs() - return the debugfs pointer
+ * @ndev: pointer to ntb_device instance
+ *
+ * Given the ntb pointer, return the debugfs directory pointer for the NTB
+ * hardware device
+ *
+ * RETURNS: a pointer to the debugfs directory
+ */
+static inline struct dentry *ntb_query_debugfs(struct ntb_device *ndev)
+{
+	return ndev->debugfs_dir;
+}
+
 struct ntb_device *ntb_register_transport(struct pci_dev *pdev,
 					  void *transport);
 void ntb_unregister_transport(struct ntb_device *ndev);
 void ntb_set_mw_addr(struct ntb_device *ndev, unsigned int mw, u64 addr);
 int ntb_register_db_callback(struct ntb_device *ndev, unsigned int idx,
-			     void *data, void (*db_cb_func) (void *data,
-							     int db_num));
+			     void *data, int (*db_cb_func)(void *data,
+							   int db_num));
 void ntb_unregister_db_callback(struct ntb_device *ndev, unsigned int idx);
 int ntb_register_event_callback(struct ntb_device *ndev,
 				void (*event_cb_func) (void *handle,
@@ -172,9 +241,10 @@ int ntb_write_local_spad(struct ntb_device *ndev, unsigned int idx, u32 val);
 int ntb_read_local_spad(struct ntb_device *ndev, unsigned int idx, u32 *val);
 int ntb_write_remote_spad(struct ntb_device *ndev, unsigned int idx, u32 val);
 int ntb_read_remote_spad(struct ntb_device *ndev, unsigned int idx, u32 *val);
+resource_size_t ntb_get_mw_base(struct ntb_device *ndev, unsigned int mw);
 void __iomem *ntb_get_mw_vbase(struct ntb_device *ndev, unsigned int mw);
-resource_size_t ntb_get_mw_size(struct ntb_device *ndev, unsigned int mw);
-void ntb_ring_sdb(struct ntb_device *ndev, unsigned int idx);
+u64 ntb_get_mw_size(struct ntb_device *ndev, unsigned int mw);
+void ntb_ring_doorbell(struct ntb_device *ndev, unsigned int idx);
 void *ntb_find_transport(struct pci_dev *pdev);
 
 int ntb_transport_init(struct pci_dev *pdev);
diff --git a/drivers/ntb/ntb_regs.h b/drivers/ntb/ntb_regs.h
index 5bfa8c06c059..9774506419d7 100644
--- a/drivers/ntb/ntb_regs.h
+++ b/drivers/ntb/ntb_regs.h
@@ -46,23 +46,25 @@
  * Jon Mason <jon.mason@intel.com>
  */
 
-#define NTB_LINK_ENABLE		0x0000
-#define NTB_LINK_DISABLE	0x0002
 #define NTB_LINK_STATUS_ACTIVE	0x2000
 #define NTB_LINK_SPEED_MASK	0x000f
 #define NTB_LINK_WIDTH_MASK	0x03f0
 
 #define SNB_MSIX_CNT		4
-#define SNB_MAX_SPADS		16
-#define SNB_MAX_COMPAT_SPADS	8
+#define SNB_MAX_B2B_SPADS	16
+#define SNB_MAX_COMPAT_SPADS	16
 /* Reserve the uppermost bit for link interrupt */
 #define SNB_MAX_DB_BITS		15
+#define SNB_LINK_DB		15
 #define SNB_DB_BITS_PER_VEC	5
+#define SNB_MAX_MW		2
+#define SNB_ERRATA_MAX_MW	1
 
 #define SNB_DB_HW_LINK		0x8000
 
 #define SNB_PCICMD_OFFSET	0x0504
 #define SNB_DEVCTRL_OFFSET	0x0598
+#define SNB_SLINK_STATUS_OFFSET	0x05A2
 #define SNB_LINK_STATUS_OFFSET	0x01A2
 
 #define SNB_PBAR2LMT_OFFSET	0x0000
@@ -88,19 +90,28 @@
 #define SNB_WCCNTRL_OFFSET	0x00e0
 #define SNB_B2B_SPAD_OFFSET	0x0100
 #define SNB_B2B_DOORBELL_OFFSET	0x0140
-#define SNB_B2B_XLAT_OFFSET	0x0144
+#define SNB_B2B_XLAT_OFFSETL	0x0144
+#define SNB_B2B_XLAT_OFFSETU	0x0148
+
+#define SNB_MBAR01_USD_ADDR	0x000000210000000CULL
+#define SNB_MBAR23_USD_ADDR	0x000000410000000CULL
+#define SNB_MBAR45_USD_ADDR	0x000000810000000CULL
+#define SNB_MBAR01_DSD_ADDR	0x000000200000000CULL
+#define SNB_MBAR23_DSD_ADDR	0x000000400000000CULL
+#define SNB_MBAR45_DSD_ADDR	0x000000800000000CULL
 
 #define BWD_MSIX_CNT		34
 #define BWD_MAX_SPADS		16
-#define BWD_MAX_COMPAT_SPADS	16
 #define BWD_MAX_DB_BITS		34
 #define BWD_DB_BITS_PER_VEC	1
+#define BWD_MAX_MW		2
 
 #define BWD_PCICMD_OFFSET	0xb004
 #define BWD_MBAR23_OFFSET	0xb018
 #define BWD_MBAR45_OFFSET	0xb020
 #define BWD_DEVCTRL_OFFSET	0xb048
 #define BWD_LINK_STATUS_OFFSET	0xb052
+#define BWD_ERRCORSTS_OFFSET	0xb110
 
 #define BWD_SBAR2XLAT_OFFSET	0x0008
 #define BWD_SBAR4XLAT_OFFSET	0x0010
@@ -118,9 +129,27 @@
 #define BWD_B2B_SPADSEMA_OFFSET	0x80c0
 #define BWD_B2B_STKYSPAD_OFFSET	0x80c4
 
-#define NTB_CNTL_BAR23_SNOOP	(1 << 2)
-#define NTB_CNTL_BAR45_SNOOP	(1 << 6)
-#define BWD_CNTL_LINK_DOWN	(1 << 16)
+#define BWD_MODPHY_PCSREG4	0x1c004
+#define BWD_MODPHY_PCSREG6	0x1c006
+
+#define BWD_IP_BASE		0xC000
+#define BWD_DESKEWSTS_OFFSET	(BWD_IP_BASE + 0x3024)
+#define BWD_LTSSMERRSTS0_OFFSET (BWD_IP_BASE + 0x3180)
+#define BWD_LTSSMSTATEJMP_OFFSET	(BWD_IP_BASE + 0x3040)
+#define BWD_IBSTERRRCRVSTS0_OFFSET	(BWD_IP_BASE + 0x3324)
+
+#define BWD_DESKEWSTS_DBERR	(1 << 15)
+#define BWD_LTSSMERRSTS0_UNEXPECTEDEI	(1 << 20)
+#define BWD_LTSSMSTATEJMP_FORCEDETECT	(1 << 2)
+#define BWD_IBIST_ERR_OFLOW	0x7FFF7FFF
+
+#define NTB_CNTL_CFG_LOCK		(1 << 0)
+#define NTB_CNTL_LINK_DISABLE		(1 << 1)
+#define NTB_CNTL_S2P_BAR23_SNOOP	(1 << 2)
+#define NTB_CNTL_P2S_BAR23_SNOOP	(1 << 4)
+#define NTB_CNTL_S2P_BAR45_SNOOP	(1 << 6)
+#define NTB_CNTL_P2S_BAR45_SNOOP	(1 << 8)
+#define BWD_CNTL_LINK_DOWN		(1 << 16)
 
 #define NTB_PPD_OFFSET		0x00D4
 #define SNB_PPD_CONN_TYPE	0x0003
@@ -128,12 +157,3 @@
 #define BWD_PPD_INIT_LINK	0x0008
 #define BWD_PPD_CONN_TYPE	0x0300
 #define BWD_PPD_DEV_TYPE	0x1000
-
-#define BWD_PBAR2XLAT_USD_ADDR	0x0000004000000000
-#define BWD_PBAR4XLAT_USD_ADDR	0x0000008000000000
-#define BWD_MBAR23_USD_ADDR	0x000000410000000C
-#define BWD_MBAR45_USD_ADDR	0x000000810000000C
-#define BWD_PBAR2XLAT_DSD_ADDR	0x0000004100000000
-#define BWD_PBAR4XLAT_DSD_ADDR	0x0000008100000000
-#define BWD_MBAR23_DSD_ADDR	0x000000400000000C
-#define BWD_MBAR45_DSD_ADDR	0x000000800000000C
diff --git a/drivers/ntb/ntb_transport.c b/drivers/ntb/ntb_transport.c
index f8d7081ee301..3217f394d45b 100644
--- a/drivers/ntb/ntb_transport.c
+++ b/drivers/ntb/ntb_transport.c
@@ -47,6 +47,7 @@
  */
 #include <linux/debugfs.h>
 #include <linux/delay.h>
+#include <linux/dmaengine.h>
 #include <linux/dma-mapping.h>
 #include <linux/errno.h>
 #include <linux/export.h>
@@ -64,10 +65,14 @@ static unsigned int transport_mtu = 0x401E;
 module_param(transport_mtu, uint, 0644);
 MODULE_PARM_DESC(transport_mtu, "Maximum size of NTB transport packets");
 
-static unsigned char max_num_clients = 2;
+static unsigned char max_num_clients;
 module_param(max_num_clients, byte, 0644);
 MODULE_PARM_DESC(max_num_clients, "Maximum number of NTB transport clients");
 
+static unsigned int copy_bytes = 1024;
+module_param(copy_bytes, uint, 0644);
+MODULE_PARM_DESC(copy_bytes, "Threshold under which NTB will use the CPU to copy instead of DMA");
+
 struct ntb_queue_entry {
 	/* ntb_queue list reference */
 	struct list_head entry;
@@ -76,6 +81,13 @@ struct ntb_queue_entry {
 	void *buf;
 	unsigned int len;
 	unsigned int flags;
+
+	struct ntb_transport_qp *qp;
+	union {
+		struct ntb_payload_header __iomem *tx_hdr;
+		struct ntb_payload_header *rx_hdr;
+	};
+	unsigned int index;
 };
 
 struct ntb_rx_info {
@@ -86,6 +98,7 @@ struct ntb_transport_qp {
 	struct ntb_transport *transport;
 	struct ntb_device *ndev;
 	void *cb_data;
+	struct dma_chan *dma_chan;
 
 	bool client_ready;
 	bool qp_link;
@@ -99,13 +112,13 @@ struct ntb_transport_qp {
 	struct list_head tx_free_q;
 	spinlock_t ntb_tx_free_q_lock;
 	void __iomem *tx_mw;
+	dma_addr_t tx_mw_phys;
 	unsigned int tx_index;
 	unsigned int tx_max_entry;
 	unsigned int tx_max_frame;
 
 	void (*rx_handler) (struct ntb_transport_qp *qp, void *qp_data,
 			    void *data, int len);
-	struct tasklet_struct rx_work;
 	struct list_head rx_pend_q;
 	struct list_head rx_free_q;
 	spinlock_t ntb_rx_pend_q_lock;
@@ -114,6 +127,7 @@ struct ntb_transport_qp {
 	unsigned int rx_index;
 	unsigned int rx_max_entry;
 	unsigned int rx_max_frame;
+	dma_cookie_t last_cookie;
 
 	void (*event_handler) (void *data, int status);
 	struct delayed_work link_work;
@@ -129,9 +143,14 @@ struct ntb_transport_qp {
 	u64 rx_err_no_buf;
 	u64 rx_err_oflow;
 	u64 rx_err_ver;
+	u64 rx_memcpy;
+	u64 rx_async;
 	u64 tx_bytes;
 	u64 tx_pkts;
 	u64 tx_ring_full;
+	u64 tx_err_no_buf;
+	u64 tx_memcpy;
+	u64 tx_async;
 };
 
 struct ntb_transport_mw {
@@ -150,14 +169,13 @@ struct ntb_transport {
 	struct list_head client_devs;
 
 	struct ntb_device *ndev;
-	struct ntb_transport_mw mw[NTB_NUM_MW];
+	struct ntb_transport_mw *mw;
 	struct ntb_transport_qp *qps;
 	unsigned int max_qps;
 	unsigned long qp_bitmap;
 	bool transport_link;
 	struct delayed_work link_work;
 	struct work_struct link_cleanup;
-	struct dentry *debugfs_dir;
 };
 
 enum {
@@ -183,7 +201,7 @@ enum {
 	MAX_SPAD,
 };
 
-#define QP_TO_MW(qp)		((qp) % NTB_NUM_MW)
+#define QP_TO_MW(ndev, qp)	((qp) % ntb_max_mw(ndev))
 #define NTB_QP_DEF_NUM_ENTRIES	100
 #define NTB_LINK_DOWN_TIMEOUT	10
 
@@ -382,7 +400,7 @@ static ssize_t debugfs_read(struct file *filp, char __user *ubuf, size_t count,
 	char *buf;
 	ssize_t ret, out_offset, out_count;
 
-	out_count = 600;
+	out_count = 1000;
 
 	buf = kmalloc(out_count, GFP_KERNEL);
 	if (!buf)
@@ -397,6 +415,10 @@ static ssize_t debugfs_read(struct file *filp, char __user *ubuf, size_t count,
 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
 			       "rx_pkts - \t%llu\n", qp->rx_pkts);
 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
+			       "rx_memcpy - \t%llu\n", qp->rx_memcpy);
+	out_offset += snprintf(buf + out_offset, out_count - out_offset,
+			       "rx_async - \t%llu\n", qp->rx_async);
+	out_offset += snprintf(buf + out_offset, out_count - out_offset,
 			       "rx_ring_empty - %llu\n", qp->rx_ring_empty);
 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
 			       "rx_err_no_buf - %llu\n", qp->rx_err_no_buf);
@@ -416,8 +438,14 @@ static ssize_t debugfs_read(struct file *filp, char __user *ubuf, size_t count,
 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
 			       "tx_pkts - \t%llu\n", qp->tx_pkts);
 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
+			       "tx_memcpy - \t%llu\n", qp->tx_memcpy);
+	out_offset += snprintf(buf + out_offset, out_count - out_offset,
+			       "tx_async - \t%llu\n", qp->tx_async);
+	out_offset += snprintf(buf + out_offset, out_count - out_offset,
 			       "tx_ring_full - \t%llu\n", qp->tx_ring_full);
 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
+			       "tx_err_no_buf - %llu\n", qp->tx_err_no_buf);
+	out_offset += snprintf(buf + out_offset, out_count - out_offset,
 			       "tx_mw - \t%p\n", qp->tx_mw);
 	out_offset += snprintf(buf + out_offset, out_count - out_offset,
 			       "tx_index - \t%u\n", qp->tx_index);
@@ -475,22 +503,25 @@ static void ntb_transport_setup_qp_mw(struct ntb_transport *nt,
 {
 	struct ntb_transport_qp *qp = &nt->qps[qp_num];
 	unsigned int rx_size, num_qps_mw;
-	u8 mw_num = QP_TO_MW(qp_num);
+	u8 mw_num, mw_max;
 	unsigned int i;
 
+	mw_max = ntb_max_mw(nt->ndev);
+	mw_num = QP_TO_MW(nt->ndev, qp_num);
+
 	WARN_ON(nt->mw[mw_num].virt_addr == NULL);
 
-	if (nt->max_qps % NTB_NUM_MW && mw_num < nt->max_qps % NTB_NUM_MW)
-		num_qps_mw = nt->max_qps / NTB_NUM_MW + 1;
+	if (nt->max_qps % mw_max && mw_num < nt->max_qps % mw_max)
+		num_qps_mw = nt->max_qps / mw_max + 1;
 	else
-		num_qps_mw = nt->max_qps / NTB_NUM_MW;
+		num_qps_mw = nt->max_qps / mw_max;
 
 	rx_size = (unsigned int) nt->mw[mw_num].size / num_qps_mw;
-	qp->remote_rx_info = nt->mw[mw_num].virt_addr +
-			     (qp_num / NTB_NUM_MW * rx_size);
+	qp->rx_buff = nt->mw[mw_num].virt_addr + qp_num / mw_max * rx_size;
 	rx_size -= sizeof(struct ntb_rx_info);
 
-	qp->rx_buff = qp->remote_rx_info + 1;
+	qp->remote_rx_info = qp->rx_buff + rx_size;
+
 	/* Due to housekeeping, there must be atleast 2 buffs */
 	qp->rx_max_frame = min(transport_mtu, rx_size / 2);
 	qp->rx_max_entry = rx_size / qp->rx_max_frame;
@@ -552,11 +583,8 @@ static int ntb_set_mw(struct ntb_transport *nt, int num_mw, unsigned int size)
 	return 0;
 }
 
-static void ntb_qp_link_cleanup(struct work_struct *work)
+static void ntb_qp_link_cleanup(struct ntb_transport_qp *qp)
 {
-	struct ntb_transport_qp *qp = container_of(work,
-						   struct ntb_transport_qp,
-						   link_cleanup);
 	struct ntb_transport *nt = qp->transport;
 	struct pci_dev *pdev = ntb_query_pdev(nt->ndev);
 
@@ -570,6 +598,16 @@ static void ntb_qp_link_cleanup(struct work_struct *work)
 
 	dev_info(&pdev->dev, "qp %d: Link Down\n", qp->qp_num);
 	qp->qp_link = NTB_LINK_DOWN;
+}
+
+static void ntb_qp_link_cleanup_work(struct work_struct *work)
+{
+	struct ntb_transport_qp *qp = container_of(work,
+						   struct ntb_transport_qp,
+						   link_cleanup);
+	struct ntb_transport *nt = qp->transport;
+
+	ntb_qp_link_cleanup(qp);
 
 	if (nt->transport_link == NTB_LINK_UP)
 		schedule_delayed_work(&qp->link_work,
@@ -581,22 +619,20 @@ static void ntb_qp_link_down(struct ntb_transport_qp *qp)
 	schedule_work(&qp->link_cleanup);
 }
 
-static void ntb_transport_link_cleanup(struct work_struct *work)
+static void ntb_transport_link_cleanup(struct ntb_transport *nt)
 {
-	struct ntb_transport *nt = container_of(work, struct ntb_transport,
-						link_cleanup);
 	int i;
 
+	/* Pass along the info to any clients */
+	for (i = 0; i < nt->max_qps; i++)
+		if (!test_bit(i, &nt->qp_bitmap))
+			ntb_qp_link_cleanup(&nt->qps[i]);
+
 	if (nt->transport_link == NTB_LINK_DOWN)
 		cancel_delayed_work_sync(&nt->link_work);
 	else
 		nt->transport_link = NTB_LINK_DOWN;
 
-	/* Pass along the info to any clients */
-	for (i = 0; i < nt->max_qps; i++)
-		if (!test_bit(i, &nt->qp_bitmap))
-			ntb_qp_link_down(&nt->qps[i]);
-
 	/* The scratchpad registers keep the values if the remote side
 	 * goes down, blast them now to give them a sane value the next
 	 * time they are accessed
@@ -605,6 +641,14 @@ static void ntb_transport_link_cleanup(struct work_struct *work)
 		ntb_write_local_spad(nt->ndev, i, 0);
 }
 
+static void ntb_transport_link_cleanup_work(struct work_struct *work)
+{
+	struct ntb_transport *nt = container_of(work, struct ntb_transport,
+						link_cleanup);
+
+	ntb_transport_link_cleanup(nt);
+}
+
 static void ntb_transport_event_callback(void *data, enum ntb_hw_event event)
 {
 	struct ntb_transport *nt = data;
@@ -631,7 +675,7 @@ static void ntb_transport_link_work(struct work_struct *work)
 	int rc, i;
 
 	/* send the local info, in the opposite order of the way we read it */
-	for (i = 0; i < NTB_NUM_MW; i++) {
+	for (i = 0; i < ntb_max_mw(ndev); i++) {
 		rc = ntb_write_remote_spad(ndev, MW0_SZ_HIGH + (i * 2),
 					   ntb_get_mw_size(ndev, i) >> 32);
 		if (rc) {
@@ -651,10 +695,10 @@ static void ntb_transport_link_work(struct work_struct *work)
 		}
 	}
 
-	rc = ntb_write_remote_spad(ndev, NUM_MWS, NTB_NUM_MW);
+	rc = ntb_write_remote_spad(ndev, NUM_MWS, ntb_max_mw(ndev));
 	if (rc) {
 		dev_err(&pdev->dev, "Error writing %x to remote spad %d\n",
-			NTB_NUM_MW, NUM_MWS);
+			ntb_max_mw(ndev), NUM_MWS);
 		goto out;
 	}
 
@@ -699,11 +743,11 @@ static void ntb_transport_link_work(struct work_struct *work)
 		goto out;
 	}
 
-	if (val != NTB_NUM_MW)
+	if (val != ntb_max_mw(ndev))
 		goto out;
 	dev_dbg(&pdev->dev, "Remote number of mws = %d\n", val);
 
-	for (i = 0; i < NTB_NUM_MW; i++) {
+	for (i = 0; i < ntb_max_mw(ndev); i++) {
 		u64 val64;
 
 		rc = ntb_read_remote_spad(ndev, MW0_SZ_HIGH + (i * 2), &val);
@@ -745,7 +789,7 @@ static void ntb_transport_link_work(struct work_struct *work)
 	return;
 
 out1:
-	for (i = 0; i < NTB_NUM_MW; i++)
+	for (i = 0; i < ntb_max_mw(ndev); i++)
 		ntb_free_mw(nt, i);
 out:
 	if (ntb_hw_link_status(ndev))
@@ -794,12 +838,16 @@ static void ntb_qp_link_work(struct work_struct *work)
 				      msecs_to_jiffies(NTB_LINK_DOWN_TIMEOUT));
 }
 
-static void ntb_transport_init_queue(struct ntb_transport *nt,
+static int ntb_transport_init_queue(struct ntb_transport *nt,
 				     unsigned int qp_num)
 {
 	struct ntb_transport_qp *qp;
 	unsigned int num_qps_mw, tx_size;
-	u8 mw_num = QP_TO_MW(qp_num);
+	u8 mw_num, mw_max;
+	u64 qp_offset;
+
+	mw_max = ntb_max_mw(nt->ndev);
+	mw_num = QP_TO_MW(nt->ndev, qp_num);
 
 	qp = &nt->qps[qp_num];
 	qp->qp_num = qp_num;
@@ -809,27 +857,34 @@ static void ntb_transport_init_queue(struct ntb_transport *nt,
 	qp->client_ready = NTB_LINK_DOWN;
 	qp->event_handler = NULL;
 
-	if (nt->max_qps % NTB_NUM_MW && mw_num < nt->max_qps % NTB_NUM_MW)
-		num_qps_mw = nt->max_qps / NTB_NUM_MW + 1;
+	if (nt->max_qps % mw_max && mw_num < nt->max_qps % mw_max)
+		num_qps_mw = nt->max_qps / mw_max + 1;
 	else
-		num_qps_mw = nt->max_qps / NTB_NUM_MW;
+		num_qps_mw = nt->max_qps / mw_max;
 
 	tx_size = (unsigned int) ntb_get_mw_size(qp->ndev, mw_num) / num_qps_mw;
-	qp->rx_info = ntb_get_mw_vbase(nt->ndev, mw_num) +
-		      (qp_num / NTB_NUM_MW * tx_size);
+	qp_offset = qp_num / mw_max * tx_size;
+	qp->tx_mw = ntb_get_mw_vbase(nt->ndev, mw_num) + qp_offset;
+	if (!qp->tx_mw)
+		return -EINVAL;
+
+	qp->tx_mw_phys = ntb_get_mw_base(qp->ndev, mw_num) + qp_offset;
+	if (!qp->tx_mw_phys)
+		return -EINVAL;
+
 	tx_size -= sizeof(struct ntb_rx_info);
+	qp->rx_info = qp->tx_mw + tx_size;
 
-	qp->tx_mw = qp->rx_info + 1;
 	/* Due to housekeeping, there must be atleast 2 buffs */
 	qp->tx_max_frame = min(transport_mtu, tx_size / 2);
 	qp->tx_max_entry = tx_size / qp->tx_max_frame;
 
-	if (nt->debugfs_dir) {
+	if (ntb_query_debugfs(nt->ndev)) {
 		char debugfs_name[4];
 
 		snprintf(debugfs_name, 4, "qp%d", qp_num);
 		qp->debugfs_dir = debugfs_create_dir(debugfs_name,
-						     nt->debugfs_dir);
+						 ntb_query_debugfs(nt->ndev));
 
 		qp->debugfs_stats = debugfs_create_file("stats", S_IRUSR,
 							qp->debugfs_dir, qp,
@@ -837,7 +892,7 @@ static void ntb_transport_init_queue(struct ntb_transport *nt,
 	}
 
 	INIT_DELAYED_WORK(&qp->link_work, ntb_qp_link_work);
-	INIT_WORK(&qp->link_cleanup, ntb_qp_link_cleanup);
+	INIT_WORK(&qp->link_cleanup, ntb_qp_link_cleanup_work);
 
 	spin_lock_init(&qp->ntb_rx_pend_q_lock);
 	spin_lock_init(&qp->ntb_rx_free_q_lock);
@@ -846,6 +901,8 @@ static void ntb_transport_init_queue(struct ntb_transport *nt,
 	INIT_LIST_HEAD(&qp->rx_pend_q);
 	INIT_LIST_HEAD(&qp->rx_free_q);
 	INIT_LIST_HEAD(&qp->tx_free_q);
+
+	return 0;
 }
 
 int ntb_transport_init(struct pci_dev *pdev)
@@ -857,57 +914,66 @@ int ntb_transport_init(struct pci_dev *pdev)
 	if (!nt)
 		return -ENOMEM;
 
-	if (debugfs_initialized())
-		nt->debugfs_dir = debugfs_create_dir(KBUILD_MODNAME, NULL);
-	else
-		nt->debugfs_dir = NULL;
-
 	nt->ndev = ntb_register_transport(pdev, nt);
 	if (!nt->ndev) {
 		rc = -EIO;
 		goto err;
 	}
 
-	nt->max_qps = min(nt->ndev->max_cbs, max_num_clients);
+	nt->mw = kcalloc(ntb_max_mw(nt->ndev), sizeof(struct ntb_transport_mw),
+			 GFP_KERNEL);
+	if (!nt->mw) {
+		rc = -ENOMEM;
+		goto err1;
+	}
+
+	if (max_num_clients)
+		nt->max_qps = min(ntb_max_cbs(nt->ndev), max_num_clients);
+	else
+		nt->max_qps = min(ntb_max_cbs(nt->ndev), ntb_max_mw(nt->ndev));
 
 	nt->qps = kcalloc(nt->max_qps, sizeof(struct ntb_transport_qp),
 			  GFP_KERNEL);
 	if (!nt->qps) {
 		rc = -ENOMEM;
-		goto err1;
+		goto err2;
 	}
 
 	nt->qp_bitmap = ((u64) 1 << nt->max_qps) - 1;
 
-	for (i = 0; i < nt->max_qps; i++)
-		ntb_transport_init_queue(nt, i);
+	for (i = 0; i < nt->max_qps; i++) {
+		rc = ntb_transport_init_queue(nt, i);
+		if (rc)
+			goto err3;
+	}
 
 	INIT_DELAYED_WORK(&nt->link_work, ntb_transport_link_work);
-	INIT_WORK(&nt->link_cleanup, ntb_transport_link_cleanup);
+	INIT_WORK(&nt->link_cleanup, ntb_transport_link_cleanup_work);
 
 	rc = ntb_register_event_callback(nt->ndev,
 					 ntb_transport_event_callback);
 	if (rc)
-		goto err2;
+		goto err3;
 
 	INIT_LIST_HEAD(&nt->client_devs);
 	rc = ntb_bus_init(nt);
 	if (rc)
-		goto err3;
+		goto err4;
 
 	if (ntb_hw_link_status(nt->ndev))
 		schedule_delayed_work(&nt->link_work, 0);
 
 	return 0;
 
-err3:
+err4:
 	ntb_unregister_event_callback(nt->ndev);
-err2:
+err3:
 	kfree(nt->qps);
+err2:
+	kfree(nt->mw);
 err1:
 	ntb_unregister_transport(nt->ndev);
 err:
-	debugfs_remove_recursive(nt->debugfs_dir);
 	kfree(nt);
 	return rc;
 }
@@ -915,41 +981,46 @@ err:
 void ntb_transport_free(void *transport)
 {
 	struct ntb_transport *nt = transport;
-	struct pci_dev *pdev;
+	struct ntb_device *ndev = nt->ndev;
 	int i;
 
-	nt->transport_link = NTB_LINK_DOWN;
+	ntb_transport_link_cleanup(nt);
 
 	/* verify that all the qp's are freed */
-	for (i = 0; i < nt->max_qps; i++)
+	for (i = 0; i < nt->max_qps; i++) {
 		if (!test_bit(i, &nt->qp_bitmap))
 			ntb_transport_free_queue(&nt->qps[i]);
+		debugfs_remove_recursive(nt->qps[i].debugfs_dir);
+	}
 
 	ntb_bus_remove(nt);
 
 	cancel_delayed_work_sync(&nt->link_work);
 
-	debugfs_remove_recursive(nt->debugfs_dir);
-
-	ntb_unregister_event_callback(nt->ndev);
-
-	pdev = ntb_query_pdev(nt->ndev);
+	ntb_unregister_event_callback(ndev);
 
-	for (i = 0; i < NTB_NUM_MW; i++)
+	for (i = 0; i < ntb_max_mw(ndev); i++)
 		ntb_free_mw(nt, i);
 
 	kfree(nt->qps);
-	ntb_unregister_transport(nt->ndev);
+	kfree(nt->mw);
+	ntb_unregister_transport(ndev);
 	kfree(nt);
 }
 
-static void ntb_rx_copy_task(struct ntb_transport_qp *qp,
-			     struct ntb_queue_entry *entry, void *offset)
+static void ntb_rx_copy_callback(void *data)
 {
+	struct ntb_queue_entry *entry = data;
+	struct ntb_transport_qp *qp = entry->qp;
 	void *cb_data = entry->cb_data;
 	unsigned int len = entry->len;
+	struct ntb_payload_header *hdr = entry->rx_hdr;
+
+	/* Ensure that the data is fully copied out before clearing the flag */
+	wmb();
+	hdr->flags = 0;
 
-	memcpy(entry->buf, offset, entry->len);
+	iowrite32(entry->index, &qp->rx_info->entry);
 
 	ntb_list_add(&qp->ntb_rx_free_q_lock, &entry->entry, &qp->rx_free_q);
 
@@ -957,6 +1028,99 @@ static void ntb_rx_copy_task(struct ntb_transport_qp *qp,
 		qp->rx_handler(qp, qp->cb_data, cb_data, len);
 }
 
+static void ntb_memcpy_rx(struct ntb_queue_entry *entry, void *offset)
+{
+	void *buf = entry->buf;
+	size_t len = entry->len;
+
+	memcpy(buf, offset, len);
+
+	ntb_rx_copy_callback(entry);
+}
+
+static void ntb_async_rx(struct ntb_queue_entry *entry, void *offset,
+			 size_t len)
+{
+	struct dma_async_tx_descriptor *txd;
+	struct ntb_transport_qp *qp = entry->qp;
+	struct dma_chan *chan = qp->dma_chan;
+	struct dma_device *device;
+	size_t pay_off, buff_off;
+	struct dmaengine_unmap_data *unmap;
+	dma_cookie_t cookie;
+	void *buf = entry->buf;
+
+	entry->len = len;
+
+	if (!chan)
+		goto err;
+
+	if (len < copy_bytes) 
+		goto err_wait;
+
+	device = chan->device;
+	pay_off = (size_t) offset & ~PAGE_MASK;
+	buff_off = (size_t) buf & ~PAGE_MASK;
+
+	if (!is_dma_copy_aligned(device, pay_off, buff_off, len))
+		goto err_wait;
+
+	unmap = dmaengine_get_unmap_data(device->dev, 2, GFP_NOWAIT);
+	if (!unmap)
+		goto err_wait;
+
+	unmap->len = len;
+	unmap->addr[0] = dma_map_page(device->dev, virt_to_page(offset),
+				      pay_off, len, DMA_TO_DEVICE);
+	if (dma_mapping_error(device->dev, unmap->addr[0]))
+		goto err_get_unmap;
+
+	unmap->to_cnt = 1;
+
+	unmap->addr[1] = dma_map_page(device->dev, virt_to_page(buf),
+				      buff_off, len, DMA_FROM_DEVICE);
+	if (dma_mapping_error(device->dev, unmap->addr[1]))
+		goto err_get_unmap;
+
+	unmap->from_cnt = 1;
+
+	txd = device->device_prep_dma_memcpy(chan, unmap->addr[1],
+					     unmap->addr[0], len,
+					     DMA_PREP_INTERRUPT);
+	if (!txd)
+		goto err_get_unmap;
+
+	txd->callback = ntb_rx_copy_callback;
+	txd->callback_param = entry;
+	dma_set_unmap(txd, unmap);
+
+	cookie = dmaengine_submit(txd);
+	if (dma_submit_error(cookie))
+		goto err_set_unmap;
+
+	dmaengine_unmap_put(unmap);
+
+	qp->last_cookie = cookie;
+
+	qp->rx_async++;
+
+	return;
+
+err_set_unmap:
+	dmaengine_unmap_put(unmap);
+err_get_unmap:
+	dmaengine_unmap_put(unmap);
+err_wait:
+	/* If the callbacks come out of order, the writing of the index to the
+	 * last completed will be out of order.  This may result in the
+	 * receive stalling forever.
+	 */
+	dma_sync_wait(chan, qp->last_cookie);
+err:
+	ntb_memcpy_rx(entry, offset);
+	qp->rx_memcpy++;
+}
+
 static int ntb_process_rxc(struct ntb_transport_qp *qp)
 {
 	struct ntb_payload_header *hdr;
@@ -995,48 +1159,55 @@ static int ntb_process_rxc(struct ntb_transport_qp *qp)
 	if (hdr->flags & LINK_DOWN_FLAG) {
 		ntb_qp_link_down(qp);
 
-		ntb_list_add(&qp->ntb_rx_pend_q_lock, &entry->entry,
-			     &qp->rx_pend_q);
-		goto out;
+		goto err;
 	}
 
 	dev_dbg(&ntb_query_pdev(qp->ndev)->dev,
 		"rx offset %u, ver %u - %d payload received, buf size %d\n",
 		qp->rx_index, hdr->ver, hdr->len, entry->len);
 
-	if (hdr->len <= entry->len) {
-		entry->len = hdr->len;
-		ntb_rx_copy_task(qp, entry, offset);
-	} else {
-		ntb_list_add(&qp->ntb_rx_pend_q_lock, &entry->entry,
-			     &qp->rx_pend_q);
+	qp->rx_bytes += hdr->len;
+	qp->rx_pkts++;
 
+	if (hdr->len > entry->len) {
 		qp->rx_err_oflow++;
 		dev_dbg(&ntb_query_pdev(qp->ndev)->dev,
 			"RX overflow! Wanted %d got %d\n",
 			hdr->len, entry->len);
+
+		goto err;
 	}
 
-	qp->rx_bytes += hdr->len;
-	qp->rx_pkts++;
+	entry->index = qp->rx_index;
+	entry->rx_hdr = hdr;
+
+	ntb_async_rx(entry, offset, hdr->len);
 
 out:
+	qp->rx_index++;
+	qp->rx_index %= qp->rx_max_entry;
+
+	return 0;
+
+err:
+	ntb_list_add(&qp->ntb_rx_pend_q_lock, &entry->entry,
+		     &qp->rx_pend_q);
 	/* Ensure that the data is fully copied out before clearing the flag */
 	wmb();
 	hdr->flags = 0;
 	iowrite32(qp->rx_index, &qp->rx_info->entry);
 
-	qp->rx_index++;
-	qp->rx_index %= qp->rx_max_entry;
-
-	return 0;
+	goto out;
 }
 
-static void ntb_transport_rx(unsigned long data)
+static int ntb_transport_rxc_db(void *data, int db_num)
 {
-	struct ntb_transport_qp *qp = (struct ntb_transport_qp *)data;
+	struct ntb_transport_qp *qp = data;
 	int rc, i;
 
+	dev_dbg(&ntb_query_pdev(qp->ndev)->dev, "%s: doorbell %d received\n",
+		__func__, db_num);
+
 	/* Limit the number of packets processed in a single interrupt to
 	 * provide fairness to others
 	 */
@@ -1045,35 +1216,24 @@ static void ntb_transport_rx(unsigned long data)
 		if (rc)
 			break;
 	}
-}
 
-static void ntb_transport_rxc_db(void *data, int db_num)
-{
-	struct ntb_transport_qp *qp = data;
-
-	dev_dbg(&ntb_query_pdev(qp->ndev)->dev, "%s: doorbell %d received\n",
-		__func__, db_num);
+	if (qp->dma_chan)
+		dma_async_issue_pending(qp->dma_chan);
 
-	tasklet_schedule(&qp->rx_work);
+	return i;
 }
 
-static void ntb_tx_copy_task(struct ntb_transport_qp *qp,
-			     struct ntb_queue_entry *entry,
-			     void __iomem *offset)
+static void ntb_tx_copy_callback(void *data)
 {
-	struct ntb_payload_header __iomem *hdr;
-
-	memcpy_toio(offset, entry->buf, entry->len);
-
-	hdr = offset + qp->tx_max_frame - sizeof(struct ntb_payload_header);
-	iowrite32(entry->len, &hdr->len);
-	iowrite32((u32) qp->tx_pkts, &hdr->ver);
+	struct ntb_queue_entry *entry = data;
+	struct ntb_transport_qp *qp = entry->qp;
+	struct ntb_payload_header __iomem *hdr = entry->tx_hdr;
 
-	/* Ensure that the data is fully copied out before setting the flag */
+	/* Ensure that the data is fully copied out before setting the flags */
 	wmb();
 	iowrite32(entry->flags | DESC_DONE_FLAG, &hdr->flags);
 
-	ntb_ring_sdb(qp->ndev, qp->qp_num);
+	ntb_ring_doorbell(qp->ndev, qp->qp_num);
 
 	/* The entry length can only be zero if the packet is intended to be a
 	 * "link down" or similar.  Since no payload is being sent in these
@@ -1090,15 +1250,94 @@ static void ntb_tx_copy_task(struct ntb_transport_qp *qp,
 	ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry, &qp->tx_free_q);
 }
 
-static int ntb_process_tx(struct ntb_transport_qp *qp,
-			  struct ntb_queue_entry *entry)
+static void ntb_memcpy_tx(struct ntb_queue_entry *entry, void __iomem *offset)
 {
+	memcpy_toio(offset, entry->buf, entry->len);
+
+	ntb_tx_copy_callback(entry);
+}
+
+static void ntb_async_tx(struct ntb_transport_qp *qp,
+			 struct ntb_queue_entry *entry)
+{
+	struct ntb_payload_header __iomem *hdr;
+	struct dma_async_tx_descriptor *txd;
+	struct dma_chan *chan = qp->dma_chan;
+	struct dma_device *device;
+	size_t dest_off, buff_off;
+	struct dmaengine_unmap_data *unmap;
+	dma_addr_t dest;
+	dma_cookie_t cookie;
 	void __iomem *offset;
+	size_t len = entry->len;
+	void *buf = entry->buf;
 
 	offset = qp->tx_mw + qp->tx_max_frame * qp->tx_index;
+	hdr = offset + qp->tx_max_frame - sizeof(struct ntb_payload_header);
+	entry->tx_hdr = hdr;
+
+	iowrite32(entry->len, &hdr->len);
+	iowrite32((u32) qp->tx_pkts, &hdr->ver);
+
+	if (!chan)
+		goto err;
+
+	if (len < copy_bytes)
+		goto err;
+
+	device = chan->device;
+	dest = qp->tx_mw_phys + qp->tx_max_frame * qp->tx_index;
+	buff_off = (size_t) buf & ~PAGE_MASK;
+	dest_off = (size_t) dest & ~PAGE_MASK;
+
+	if (!is_dma_copy_aligned(device, buff_off, dest_off, len))
+		goto err;
+
+	unmap = dmaengine_get_unmap_data(device->dev, 1, GFP_NOWAIT);
+	if (!unmap)
+		goto err;
+
+	unmap->len = len;
+	unmap->addr[0] = dma_map_page(device->dev, virt_to_page(buf),
+				      buff_off, len, DMA_TO_DEVICE);
+	if (dma_mapping_error(device->dev, unmap->addr[0]))
+		goto err_get_unmap;
+
+	unmap->to_cnt = 1;
+
+	txd = device->device_prep_dma_memcpy(chan, dest, unmap->addr[0], len,
+					     DMA_PREP_INTERRUPT);
+	if (!txd)
+		goto err_get_unmap;
+
+	txd->callback = ntb_tx_copy_callback;
+	txd->callback_param = entry;
+	dma_set_unmap(txd, unmap);
+
+	cookie = dmaengine_submit(txd);
+	if (dma_submit_error(cookie))
+		goto err_set_unmap;
 
-	dev_dbg(&ntb_query_pdev(qp->ndev)->dev, "%lld - offset %p, tx %u, entry len %d flags %x buff %p\n",
-		qp->tx_pkts, offset, qp->tx_index, entry->len, entry->flags,
+	dmaengine_unmap_put(unmap);
+
+	dma_async_issue_pending(chan);
+	qp->tx_async++;
+
+	return;
+err_set_unmap:
+	dmaengine_unmap_put(unmap);
+err_get_unmap:
+	dmaengine_unmap_put(unmap);
+err:
+	ntb_memcpy_tx(entry, offset);
+	qp->tx_memcpy++;
+}
+
+static int ntb_process_tx(struct ntb_transport_qp *qp,
+			  struct ntb_queue_entry *entry)
+{
+	dev_dbg(&ntb_query_pdev(qp->ndev)->dev, "%lld - tx %u, entry len %d flags %x buff %p\n",
+		qp->tx_pkts, qp->tx_index, entry->len, entry->flags,
 		entry->buf);
 	if (qp->tx_index == qp->remote_rx_info->entry) {
 		qp->tx_ring_full++;
@@ -1114,7 +1353,7 @@ static int ntb_process_tx(struct ntb_transport_qp *qp,
 		return 0;
 	}
 
-	ntb_tx_copy_task(qp, entry, offset);
+	ntb_async_tx(qp, entry);
 
 	qp->tx_index++;
 	qp->tx_index %= qp->tx_max_entry;
@@ -1200,11 +1439,19 @@ ntb_transport_create_queue(void *data, struct pci_dev *pdev,
 	qp->tx_handler = handlers->tx_handler;
 	qp->event_handler = handlers->event_handler;
 
+	dmaengine_get();
+	qp->dma_chan = dma_find_channel(DMA_MEMCPY);
+	if (!qp->dma_chan) {
+		dmaengine_put();
+		dev_info(&pdev->dev, "Unable to allocate DMA channel, using CPU instead\n");
+	}
+
 	for (i = 0; i < NTB_QP_DEF_NUM_ENTRIES; i++) {
 		entry = kzalloc(sizeof(struct ntb_queue_entry), GFP_ATOMIC);
 		if (!entry)
 			goto err1;
 
+		entry->qp = qp;
 		ntb_list_add(&qp->ntb_rx_free_q_lock, &entry->entry,
 			     &qp->rx_free_q);
 	}
@@ -1214,29 +1461,28 @@ ntb_transport_create_queue(void *data, struct pci_dev *pdev,
 		if (!entry)
 			goto err2;
 
+		entry->qp = qp;
 		ntb_list_add(&qp->ntb_tx_free_q_lock, &entry->entry,
 			     &qp->tx_free_q);
 	}
 
-	tasklet_init(&qp->rx_work, ntb_transport_rx, (unsigned long) qp);
-
 	rc = ntb_register_db_callback(qp->ndev, free_queue, qp,
 				      ntb_transport_rxc_db);
 	if (rc)
-		goto err3;
+		goto err2;
 
 	dev_info(&pdev->dev, "NTB Transport QP %d created\n", qp->qp_num);
 
 	return qp;
 
-err3:
-	tasklet_disable(&qp->rx_work);
 err2:
 	while ((entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q)))
 		kfree(entry);
 err1:
 	while ((entry = ntb_list_rm(&qp->ntb_rx_free_q_lock, &qp->rx_free_q)))
 		kfree(entry);
+	if (qp->dma_chan)
+		dmaengine_put();
 	set_bit(free_queue, &nt->qp_bitmap);
 err:
 	return NULL;
@@ -1259,10 +1505,24 @@ void ntb_transport_free_queue(struct ntb_transport_qp *qp)
 
 	pdev = ntb_query_pdev(qp->ndev);
 
-	cancel_delayed_work_sync(&qp->link_work);
+	if (qp->dma_chan) {
+		struct dma_chan *chan = qp->dma_chan;
+		/* Putting the dma_chan to NULL will force any new traffic to be
+		 * processed by the CPU instead of the DAM engine
+		 */
+		qp->dma_chan = NULL;
+
+		/* Try to be nice and wait for any queued DMA engine
+		 * transactions to process before smashing it with a rock
+		 */
+		dma_sync_wait(chan, qp->last_cookie);
+		dmaengine_terminate_all(chan);
+		dmaengine_put();
+	}
 
 	ntb_unregister_db_callback(qp->ndev, qp->qp_num);
-	tasklet_disable(&qp->rx_work);
+
+	cancel_delayed_work_sync(&qp->link_work);
 
 	while ((entry = ntb_list_rm(&qp->ntb_rx_free_q_lock, &qp->rx_free_q)))
 		kfree(entry);
@@ -1354,7 +1614,7 @@ EXPORT_SYMBOL_GPL(ntb_transport_rx_enqueue);
  * @len: length of the data buffer
  *
  * Enqueue a new transmit buffer onto the transport queue from which a NTB
- * payload will be transmitted.  This assumes that a lock is behing held to
+ * payload will be transmitted.  This assumes that a lock is being held to
  * serialize access to the qp.
  *
  * RETURNS: An appropriate -ERRNO error value on error, or zero for success.
@@ -1369,8 +1629,10 @@ int ntb_transport_tx_enqueue(struct ntb_transport_qp *qp, void *cb, void *data,
 		return -EINVAL;
 
 	entry = ntb_list_rm(&qp->ntb_tx_free_q_lock, &qp->tx_free_q);
-	if (!entry)
+	if (!entry) {
+		qp->tx_err_no_buf++;
 		return -ENOMEM;
+	}
 
 	entry->cb_data = cb;
 	entry->buf = data;
@@ -1410,7 +1672,7 @@ EXPORT_SYMBOL_GPL(ntb_transport_link_up);
  *
  * Notify NTB transport layer of client's desire to no longer receive data on
  * transport queue specified.  It is the client's responsibility to ensure all
- * entries on queue are purged or otherwise handled appropraitely.
+ * entries on queue are purged or otherwise handled appropriately.
  */
 void ntb_transport_link_down(struct ntb_transport_qp *qp)
 {
@@ -1486,9 +1748,18 @@ EXPORT_SYMBOL_GPL(ntb_transport_qp_num);
  */
 unsigned int ntb_transport_max_size(struct ntb_transport_qp *qp)
 {
+	unsigned int max;
+
 	if (!qp)
 		return 0;
 
-	return qp->tx_max_frame - sizeof(struct ntb_payload_header);
+	if (!qp->dma_chan)
+		return qp->tx_max_frame - sizeof(struct ntb_payload_header);
+
+	/* If DMA engine usage is possible, try to find the max size for that */
+	max = qp->tx_max_frame - sizeof(struct ntb_payload_header);
+	max -= max % (1 << qp->dma_chan->device->copy_align);
+
+	return max;
 }
 EXPORT_SYMBOL_GPL(ntb_transport_max_size);