Merge tag 'ntb-4.16' of git://github.com/jonmason/ntb

Pull NTB updates from Jon Mason:
 "Bug fixes galore, removal of the ntb atom driver, and updates to the
  ntb tools and tests to support the multi-port interface"

* tag 'ntb-4.16' of git://github.com/jonmason/ntb: (37 commits)
  NTB: ntb_perf: fix cast to restricted __le32
  ntb_perf: Fix an error code in perf_copy_chunk()
  ntb_hw_switchtec: Make function switchtec_ntb_remove() static
  NTB: ntb_tool: fix memory leak on 'buf' on error exit path
  NTB: ntb_perf: fix printing of resource_size_t
  NTB: ntb_hw_idt: Set NTB_TOPO_SWITCH topology
  NTB: ntb_test: Update ntb_perf tests
  NTB: ntb_test: Update ntb_tool MW tests
  NTB: ntb_test: Add ntb_tool Message tests
  NTB: ntb_test: Update ntb_tool Scratchpad tests
  NTB: ntb_test: Update ntb_tool DB tests
  NTB: ntb_test: Update ntb_tool link tests
  NTB: ntb_test: Add ntb_tool port tests
  NTB: ntb_test: Safely use paths with whitespace
  NTB: ntb_perf: Add full multi-port NTB API support
  NTB: ntb_tool: Add full multi-port NTB API support
  NTB: ntb_pp: Add full multi-port NTB API support
  NTB: Fix UB/bug in ntb_mw_get_align()
  NTB: Set dma mask and dma coherent mask to NTB devices
  NTB: Rename NTB messaging API methods
  ...

MAINTAINERS

@@ -9801,7 +9801,7 @@ F:	drivers/ntb/hw/amd/
 NTB DRIVER CORE
 M:	Jon Mason <jdmason@kudzu.us>
 M:	Dave Jiang <dave.jiang@intel.com>
-M:	Allen Hubbe <Allen.Hubbe@emc.com>
+M:	Allen Hubbe <allenbh@gmail.com>
 L:	linux-ntb@googlegroups.com
 S:	Supported
 W:	https://github.com/jonmason/ntb/wiki

drivers/ntb/hw/amd/ntb_hw_amd.c

@@ -1020,6 +1020,10 @@ static int amd_ntb_init_pci(struct amd_ntb_dev *ndev,
 			goto err_dma_mask;
 		dev_warn(&pdev->dev, "Cannot DMA consistent highmem\n");
 	}
+	rc = dma_coerce_mask_and_coherent(&ndev->ntb.dev,
+					  dma_get_mask(&pdev->dev));
+	if (rc)
+		goto err_dma_mask;
 
 	ndev->self_mmio = pci_iomap(pdev, 0, 0);
 	if (!ndev->self_mmio) {

drivers/ntb/hw/idt/ntb_hw_idt.c

@@ -1744,20 +1744,19 @@ static int idt_ntb_msg_clear_mask(struct ntb_dev *ntb, u64 mask_bits)
  * idt_ntb_msg_read() - read message register with specified index
  *			(NTB API callback)
  * @ntb:	NTB device context.
- * @midx:	Message register index
  * @pidx:	OUT - Port index of peer device a message retrieved from
- * @msg:	OUT - Data
+ * @midx:	Message register index
  *
  * Read data from the specified message register and source register.
  *
- * Return: zero on success, negative error if invalid argument passed.
+ * Return: inbound message register value.
  */
-static int idt_ntb_msg_read(struct ntb_dev *ntb, int midx, int *pidx, u32 *msg)
+static u32 idt_ntb_msg_read(struct ntb_dev *ntb, int *pidx, int midx)
 {
 	struct idt_ntb_dev *ndev = to_ndev_ntb(ntb);
 
 	if (midx < 0 || IDT_MSG_CNT <= midx)
-		return -EINVAL;
+		return ~(u32)0;
 
 	/* Retrieve source port index of the message */
 	if (pidx != NULL) {
@@ -1772,18 +1771,15 @@ static int idt_ntb_msg_read(struct ntb_dev *ntb, int midx, int *pidx, u32 *msg)
 	}
 
 	/* Retrieve data of the corresponding message register */
-	if (msg != NULL)
-		*msg = idt_nt_read(ndev, ntdata_tbl.msgs[midx].in);
-
-	return 0;
+	return idt_nt_read(ndev, ntdata_tbl.msgs[midx].in);
 }
 
 /*
- * idt_ntb_msg_write() - write data to the specified message register
- *			 (NTB API callback)
+ * idt_ntb_peer_msg_write() - write data to the specified message register
+ *			      (NTB API callback)
  * @ntb:	NTB device context.
- * @midx:	Message register index
  * @pidx:	Port index of peer device a message being sent to
+ * @midx:	Message register index
  * @msg:	Data to send
  *
  * Just try to send data to a peer. Message status register should be
@@ -1791,7 +1787,8 @@ static int idt_ntb_msg_read(struct ntb_dev *ntb, int midx, int *pidx, u32 *msg)
  *
  * Return: zero on success, negative error if invalid argument passed.
  */
-static int idt_ntb_msg_write(struct ntb_dev *ntb, int midx, int pidx, u32 msg)
+static int idt_ntb_peer_msg_write(struct ntb_dev *ntb, int pidx, int midx,
+				  u32 msg)
 {
 	struct idt_ntb_dev *ndev = to_ndev_ntb(ntb);
 	unsigned long irqflags;
@@ -2058,7 +2055,7 @@ static const struct ntb_dev_ops idt_ntb_ops = {
 	.msg_set_mask		= idt_ntb_msg_set_mask,
 	.msg_clear_mask		= idt_ntb_msg_clear_mask,
 	.msg_read		= idt_ntb_msg_read,
-	.msg_write		= idt_ntb_msg_write
+	.peer_msg_write		= idt_ntb_peer_msg_write
 };
 
 /*
@@ -2073,7 +2070,7 @@ static int idt_register_device(struct idt_ntb_dev *ndev)
 
 	/* Initialize the rest of NTB device structure and register it */
 	ndev->ntb.ops = &idt_ntb_ops;
-	ndev->ntb.topo = NTB_TOPO_PRI;
+	ndev->ntb.topo = NTB_TOPO_SWITCH;
 
 	ret = ntb_register_device(&ndev->ntb);
 	if (ret != 0) {
@@ -2269,7 +2266,7 @@ static ssize_t idt_dbgfs_info_read(struct file *filp, char __user *ubuf,
 		 "Message data:\n");
 	for (idx = 0; idx < IDT_MSG_CNT; idx++) {
 		int src;
-		(void)idt_ntb_msg_read(&ndev->ntb, idx, &src, &data);
+		data = idt_ntb_msg_read(&ndev->ntb, &src, idx);
 		off += scnprintf(strbuf + off, size - off,
 			"\t%hhu. 0x%08x from peer %hhu (Port %hhu)\n",
 			idx, data, src, ndev->peers[src].port);
@@ -2429,7 +2426,7 @@ static int idt_init_pci(struct idt_ntb_dev *ndev)
 	struct pci_dev *pdev = ndev->ntb.pdev;
 	int ret;
 
-	/* Initialize the bit mask of DMA */
+	/* Initialize the bit mask of PCI/NTB DMA */
 	ret = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
 	if (ret != 0) {
 		ret = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
@@ -2450,6 +2447,12 @@ static int idt_init_pci(struct idt_ntb_dev *ndev)
 		dev_warn(&pdev->dev,
 			"Cannot set consistent DMA highmem bit mask\n");
 	}
+	ret = dma_coerce_mask_and_coherent(&ndev->ntb.dev,
+					   dma_get_mask(&pdev->dev));
+	if (ret != 0) {
+		dev_err(&pdev->dev, "Failed to set NTB device DMA bit mask\n");
+		return ret;
+	}
 
 	/*
 	 * Enable the device advanced error reporting. It's not critical to

drivers/ntb/hw/intel/ntb_hw_intel.c

@@ -74,12 +74,6 @@ MODULE_AUTHOR("Intel Corporation");
 #define bar0_off(base, bar) ((base) + ((bar) << 2))
 #define bar2_off(base, bar) bar0_off(base, (bar) - 2)
 
-static const struct intel_ntb_reg atom_reg;
-static const struct intel_ntb_alt_reg atom_pri_reg;
-static const struct intel_ntb_alt_reg atom_sec_reg;
-static const struct intel_ntb_alt_reg atom_b2b_reg;
-static const struct intel_ntb_xlat_reg atom_pri_xlat;
-static const struct intel_ntb_xlat_reg atom_sec_xlat;
 static const struct intel_ntb_reg xeon_reg;
 static const struct intel_ntb_alt_reg xeon_pri_reg;
 static const struct intel_ntb_alt_reg xeon_sec_reg;
@@ -184,15 +178,6 @@ static inline void _iowrite64(u64 val, void __iomem *mmio)
 #endif
 #endif
 
-static inline int pdev_is_atom(struct pci_dev *pdev)
-{
-	switch (pdev->device) {
-	case PCI_DEVICE_ID_INTEL_NTB_B2B_BWD:
-		return 1;
-	}
-	return 0;
-}
-
 static inline int pdev_is_xeon(struct pci_dev *pdev)
 {
 	switch (pdev->device) {
@@ -1006,8 +991,7 @@ static ssize_t ndev_debugfs_read(struct file *filp, char __user *ubuf,
 {
 	struct intel_ntb_dev *ndev = filp->private_data;
 
-	if (pdev_is_xeon(ndev->ntb.pdev) ||
-	    pdev_is_atom(ndev->ntb.pdev))
+	if (pdev_is_xeon(ndev->ntb.pdev))
 		return ndev_ntb_debugfs_read(filp, ubuf, count, offp);
 	else if (pdev_is_skx_xeon(ndev->ntb.pdev))
 		return ndev_ntb3_debugfs_read(filp, ubuf, count, offp);
@@ -1439,242 +1423,6 @@ static int intel_ntb_peer_spad_write(struct ntb_dev *ntb, int pidx,
 			       ndev->peer_reg->spad);
 }
 
-/* ATOM */
-
-static u64 atom_db_ioread(void __iomem *mmio)
-{
-	return ioread64(mmio);
-}
-
-static void atom_db_iowrite(u64 bits, void __iomem *mmio)
-{
-	iowrite64(bits, mmio);
-}
-
-static int atom_poll_link(struct intel_ntb_dev *ndev)
-{
-	u32 ntb_ctl;
-
-	ntb_ctl = ioread32(ndev->self_mmio + ATOM_NTBCNTL_OFFSET);
-
-	if (ntb_ctl == ndev->ntb_ctl)
-		return 0;
-
-	ndev->ntb_ctl = ntb_ctl;
-
-	ndev->lnk_sta = ioread32(ndev->self_mmio + ATOM_LINK_STATUS_OFFSET);
-
-	return 1;
-}
-
-static int atom_link_is_up(struct intel_ntb_dev *ndev)
-{
-	return ATOM_NTB_CTL_ACTIVE(ndev->ntb_ctl);
-}
-
-static int atom_link_is_err(struct intel_ntb_dev *ndev)
-{
-	if (ioread32(ndev->self_mmio + ATOM_LTSSMSTATEJMP_OFFSET)
-	    & ATOM_LTSSMSTATEJMP_FORCEDETECT)
-		return 1;
-
-	if (ioread32(ndev->self_mmio + ATOM_IBSTERRRCRVSTS0_OFFSET)
-	    & ATOM_IBIST_ERR_OFLOW)
-		return 1;
-
-	return 0;
-}
-
-static inline enum ntb_topo atom_ppd_topo(struct intel_ntb_dev *ndev, u32 ppd)
-{
-	struct device *dev = &ndev->ntb.pdev->dev;
-
-	switch (ppd & ATOM_PPD_TOPO_MASK) {
-	case ATOM_PPD_TOPO_B2B_USD:
-		dev_dbg(dev, "PPD %d B2B USD\n", ppd);
-		return NTB_TOPO_B2B_USD;
-
-	case ATOM_PPD_TOPO_B2B_DSD:
-		dev_dbg(dev, "PPD %d B2B DSD\n", ppd);
-		return NTB_TOPO_B2B_DSD;
-
-	case ATOM_PPD_TOPO_PRI_USD:
-	case ATOM_PPD_TOPO_PRI_DSD: /* accept bogus PRI_DSD */
-	case ATOM_PPD_TOPO_SEC_USD:
-	case ATOM_PPD_TOPO_SEC_DSD: /* accept bogus SEC_DSD */
-		dev_dbg(dev, "PPD %d non B2B disabled\n", ppd);
-		return NTB_TOPO_NONE;
-	}
-
-	dev_dbg(dev, "PPD %d invalid\n", ppd);
-	return NTB_TOPO_NONE;
-}
-
-static void atom_link_hb(struct work_struct *work)
-{
-	struct intel_ntb_dev *ndev = hb_ndev(work);
-	struct device *dev = &ndev->ntb.pdev->dev;
-	unsigned long poll_ts;
-	void __iomem *mmio;
-	u32 status32;
-
-	poll_ts = ndev->last_ts + ATOM_LINK_HB_TIMEOUT;
-
-	/* Delay polling the link status if an interrupt was received,
-	 * unless the cached link status says the link is down.
-	 */
-	if (time_after(poll_ts, jiffies) && atom_link_is_up(ndev)) {
-		schedule_delayed_work(&ndev->hb_timer, poll_ts - jiffies);
-		return;
-	}
-
-	if (atom_poll_link(ndev))
-		ntb_link_event(&ndev->ntb);
-
-	if (atom_link_is_up(ndev) || !atom_link_is_err(ndev)) {
-		schedule_delayed_work(&ndev->hb_timer, ATOM_LINK_HB_TIMEOUT);
-		return;
-	}
-
-	/* Link is down with error: recover the link! */
-
-	mmio = ndev->self_mmio;
-
-	/* Driver resets the NTB ModPhy lanes - magic! */
-	iowrite8(0xe0, mmio + ATOM_MODPHY_PCSREG6);
-	iowrite8(0x40, mmio + ATOM_MODPHY_PCSREG4);
-	iowrite8(0x60, mmio + ATOM_MODPHY_PCSREG4);
-	iowrite8(0x60, mmio + ATOM_MODPHY_PCSREG6);
-
-	/* Driver waits 100ms to allow the NTB ModPhy to settle */
-	msleep(100);
-
-	/* Clear AER Errors, write to clear */
-	status32 = ioread32(mmio + ATOM_ERRCORSTS_OFFSET);
-	dev_dbg(dev, "ERRCORSTS = %x\n", status32);
-	status32 &= PCI_ERR_COR_REP_ROLL;
-	iowrite32(status32, mmio + ATOM_ERRCORSTS_OFFSET);
-
-	/* Clear unexpected electrical idle event in LTSSM, write to clear */
-	status32 = ioread32(mmio + ATOM_LTSSMERRSTS0_OFFSET);
-	dev_dbg(dev, "LTSSMERRSTS0 = %x\n", status32);
-	status32 |= ATOM_LTSSMERRSTS0_UNEXPECTEDEI;
-	iowrite32(status32, mmio + ATOM_LTSSMERRSTS0_OFFSET);
-
-	/* Clear DeSkew Buffer error, write to clear */
-	status32 = ioread32(mmio + ATOM_DESKEWSTS_OFFSET);
-	dev_dbg(dev, "DESKEWSTS = %x\n", status32);
-	status32 |= ATOM_DESKEWSTS_DBERR;
-	iowrite32(status32, mmio + ATOM_DESKEWSTS_OFFSET);
-
-	status32 = ioread32(mmio + ATOM_IBSTERRRCRVSTS0_OFFSET);
-	dev_dbg(dev, "IBSTERRRCRVSTS0 = %x\n", status32);
-	status32 &= ATOM_IBIST_ERR_OFLOW;
-	iowrite32(status32, mmio + ATOM_IBSTERRRCRVSTS0_OFFSET);
-
-	/* Releases the NTB state machine to allow the link to retrain */
-	status32 = ioread32(mmio + ATOM_LTSSMSTATEJMP_OFFSET);
-	dev_dbg(dev, "LTSSMSTATEJMP = %x\n", status32);
-	status32 &= ~ATOM_LTSSMSTATEJMP_FORCEDETECT;
-	iowrite32(status32, mmio + ATOM_LTSSMSTATEJMP_OFFSET);
-
-	/* 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.
-	 */
-	schedule_delayed_work(&ndev->hb_timer, ATOM_LINK_RECOVERY_TIME
-			      + prandom_u32() % ATOM_LINK_RECOVERY_TIME);
-}
-
-static int atom_init_isr(struct intel_ntb_dev *ndev)
-{
-	int rc;
-
-	rc = ndev_init_isr(ndev, 1, ATOM_DB_MSIX_VECTOR_COUNT,
-			   ATOM_DB_MSIX_VECTOR_SHIFT, ATOM_DB_TOTAL_SHIFT);
-	if (rc)
-		return rc;
-
-	/* ATOM doesn't have link status interrupt, poll on that platform */
-	ndev->last_ts = jiffies;
-	INIT_DELAYED_WORK(&ndev->hb_timer, atom_link_hb);
-	schedule_delayed_work(&ndev->hb_timer, ATOM_LINK_HB_TIMEOUT);
-
-	return 0;
-}
-
-static void atom_deinit_isr(struct intel_ntb_dev *ndev)
-{
-	cancel_delayed_work_sync(&ndev->hb_timer);
-	ndev_deinit_isr(ndev);
-}
-
-static int atom_init_ntb(struct intel_ntb_dev *ndev)
-{
-	ndev->mw_count = ATOM_MW_COUNT;
-	ndev->spad_count = ATOM_SPAD_COUNT;
-	ndev->db_count = ATOM_DB_COUNT;
-
-	switch (ndev->ntb.topo) {
-	case NTB_TOPO_B2B_USD:
-	case NTB_TOPO_B2B_DSD:
-		ndev->self_reg = &atom_pri_reg;
-		ndev->peer_reg = &atom_b2b_reg;
-		ndev->xlat_reg = &atom_sec_xlat;
-
-		/* Enable Bus Master and Memory Space on the secondary side */
-		iowrite16(PCI_COMMAND_MEMORY | PCI_COMMAND_MASTER,
-			  ndev->self_mmio + ATOM_SPCICMD_OFFSET);
-
-		break;
-
-	default:
-		return -EINVAL;
-	}
-
-	ndev->db_valid_mask = BIT_ULL(ndev->db_count) - 1;
-
-	return 0;
-}
-
-static int atom_init_dev(struct intel_ntb_dev *ndev)
-{
-	u32 ppd;
-	int rc;
-
-	rc = pci_read_config_dword(ndev->ntb.pdev, ATOM_PPD_OFFSET, &ppd);
-	if (rc)
-		return -EIO;
-
-	ndev->ntb.topo = atom_ppd_topo(ndev, ppd);
-	if (ndev->ntb.topo == NTB_TOPO_NONE)
-		return -EINVAL;
-
-	rc = atom_init_ntb(ndev);
-	if (rc)
-		return rc;
-
-	rc = atom_init_isr(ndev);
-	if (rc)
-		return rc;
-
-	if (ndev->ntb.topo != NTB_TOPO_SEC) {
-		/* Initiate PCI-E link training */
-		rc = pci_write_config_dword(ndev->ntb.pdev, ATOM_PPD_OFFSET,
-					    ppd | ATOM_PPD_INIT_LINK);
-		if (rc)
-			return rc;
-	}
-
-	return 0;
-}
-
-static void atom_deinit_dev(struct intel_ntb_dev *ndev)
-{
-	atom_deinit_isr(ndev);
-}
-
 /* Skylake Xeon NTB */
 
 static int skx_poll_link(struct intel_ntb_dev *ndev)
@@ -2586,6 +2334,10 @@ static int intel_ntb_init_pci(struct intel_ntb_dev *ndev, struct pci_dev *pdev)
 			goto err_dma_mask;
 		dev_warn(&pdev->dev, "Cannot DMA consistent highmem\n");
 	}
+	rc = dma_coerce_mask_and_coherent(&ndev->ntb.dev,
+					  dma_get_mask(&pdev->dev));
+	if (rc)
+		goto err_dma_mask;
 
 	ndev->self_mmio = pci_iomap(pdev, 0, 0);
 	if (!ndev->self_mmio) {
@@ -2658,24 +2410,7 @@ static int intel_ntb_pci_probe(struct pci_dev *pdev,
 
 	node = dev_to_node(&pdev->dev);
 
-	if (pdev_is_atom(pdev)) {
-		ndev = kzalloc_node(sizeof(*ndev), GFP_KERNEL, node);
-		if (!ndev) {
-			rc = -ENOMEM;
-			goto err_ndev;
-		}
-
-		ndev_init_struct(ndev, pdev);
-
-		rc = intel_ntb_init_pci(ndev, pdev);
-		if (rc)
-			goto err_init_pci;
-
-		rc = atom_init_dev(ndev);
-		if (rc)
-			goto err_init_dev;
-
-	} else if (pdev_is_xeon(pdev)) {
+	if (pdev_is_xeon(pdev)) {
 		ndev = kzalloc_node(sizeof(*ndev), GFP_KERNEL, node);
 		if (!ndev) {
 			rc = -ENOMEM;
@@ -2731,9 +2466,7 @@ static int intel_ntb_pci_probe(struct pci_dev *pdev,
 
 err_register:
 	ndev_deinit_debugfs(ndev);
-	if (pdev_is_atom(pdev))
-		atom_deinit_dev(ndev);
-	else if (pdev_is_xeon(pdev) || pdev_is_skx_xeon(pdev))
+	if (pdev_is_xeon(pdev) || pdev_is_skx_xeon(pdev))
 		xeon_deinit_dev(ndev);
 err_init_dev:
 	intel_ntb_deinit_pci(ndev);
@@ -2749,41 +2482,12 @@ static void intel_ntb_pci_remove(struct pci_dev *pdev)
 
 	ntb_unregister_device(&ndev->ntb);
 	ndev_deinit_debugfs(ndev);
-	if (pdev_is_atom(pdev))
-		atom_deinit_dev(ndev);
-	else if (pdev_is_xeon(pdev) || pdev_is_skx_xeon(pdev))
+	if (pdev_is_xeon(pdev) || pdev_is_skx_xeon(pdev))
 		xeon_deinit_dev(ndev);
 	intel_ntb_deinit_pci(ndev);
 	kfree(ndev);
 }
 
-static const struct intel_ntb_reg atom_reg = {
-	.poll_link		= atom_poll_link,
-	.link_is_up		= atom_link_is_up,
-	.db_ioread		= atom_db_ioread,
-	.db_iowrite		= atom_db_iowrite,
-	.db_size		= sizeof(u64),
-	.ntb_ctl		= ATOM_NTBCNTL_OFFSET,
-	.mw_bar			= {2, 4},
-};
-
-static const struct intel_ntb_alt_reg atom_pri_reg = {
-	.db_bell		= ATOM_PDOORBELL_OFFSET,
-	.db_mask		= ATOM_PDBMSK_OFFSET,
-	.spad			= ATOM_SPAD_OFFSET,
-};
-
-static const struct intel_ntb_alt_reg atom_b2b_reg = {
-	.db_bell		= ATOM_B2B_DOORBELL_OFFSET,
-	.spad			= ATOM_B2B_SPAD_OFFSET,
-};
-
-static const struct intel_ntb_xlat_reg atom_sec_xlat = {
-	/* FIXME : .bar0_base	= ATOM_SBAR0BASE_OFFSET, */
-	/* FIXME : .bar2_limit	= ATOM_SBAR2LMT_OFFSET, */
-	.bar2_xlat		= ATOM_SBAR2XLAT_OFFSET,
-};
-
 static const struct intel_ntb_reg xeon_reg = {
 	.poll_link		= xeon_poll_link,
 	.link_is_up		= xeon_link_is_up,
@@ -2940,7 +2644,6 @@ static const struct file_operations intel_ntb_debugfs_info = {
 };
 
 static const struct pci_device_id intel_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_B2B_SNB)},
 	{PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_NTB_B2B_IVT)},

drivers/ntb/hw/intel/ntb_hw_intel.h

@@ -66,7 +66,6 @@
 #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 PCI_DEVICE_ID_INTEL_NTB_B2B_BDX	0x6F0D
 #define PCI_DEVICE_ID_INTEL_NTB_PS_BDX	0x6F0E
 #define PCI_DEVICE_ID_INTEL_NTB_SS_BDX	0x6F0F
@@ -196,63 +195,6 @@
 #define SKX_DB_TOTAL_SHIFT		33
 #define SKX_SPAD_COUNT			16
 
-/* Intel Atom hardware */
-
-#define ATOM_SBAR2XLAT_OFFSET		0x0008
-#define ATOM_PDOORBELL_OFFSET		0x0020
-#define ATOM_PDBMSK_OFFSET		0x0028
-#define ATOM_NTBCNTL_OFFSET		0x0060
-#define ATOM_SPAD_OFFSET			0x0080
-#define ATOM_PPD_OFFSET			0x00d4
-#define ATOM_PBAR2XLAT_OFFSET		0x8008
-#define ATOM_B2B_DOORBELL_OFFSET		0x8020
-#define ATOM_B2B_SPAD_OFFSET		0x8080
-#define ATOM_SPCICMD_OFFSET		0xb004
-#define ATOM_LINK_STATUS_OFFSET		0xb052
-#define ATOM_ERRCORSTS_OFFSET		0xb110
-#define ATOM_IP_BASE			0xc000
-#define ATOM_DESKEWSTS_OFFSET		(ATOM_IP_BASE + 0x3024)
-#define ATOM_LTSSMERRSTS0_OFFSET		(ATOM_IP_BASE + 0x3180)
-#define ATOM_LTSSMSTATEJMP_OFFSET	(ATOM_IP_BASE + 0x3040)
-#define ATOM_IBSTERRRCRVSTS0_OFFSET	(ATOM_IP_BASE + 0x3324)
-#define ATOM_MODPHY_PCSREG4		0x1c004
-#define ATOM_MODPHY_PCSREG6		0x1c006
-
-#define ATOM_PPD_INIT_LINK		0x0008
-#define ATOM_PPD_CONN_MASK		0x0300
-#define ATOM_PPD_CONN_TRANSPARENT	0x0000
-#define ATOM_PPD_CONN_B2B		0x0100
-#define ATOM_PPD_CONN_RP			0x0200
-#define ATOM_PPD_DEV_MASK		0x1000
-#define ATOM_PPD_DEV_USD			0x0000
-#define ATOM_PPD_DEV_DSD			0x1000
-#define ATOM_PPD_TOPO_MASK	(ATOM_PPD_CONN_MASK | ATOM_PPD_DEV_MASK)
-#define ATOM_PPD_TOPO_PRI_USD	(ATOM_PPD_CONN_TRANSPARENT | ATOM_PPD_DEV_USD)
-#define ATOM_PPD_TOPO_PRI_DSD	(ATOM_PPD_CONN_TRANSPARENT | ATOM_PPD_DEV_DSD)
-#define ATOM_PPD_TOPO_SEC_USD	(ATOM_PPD_CONN_RP | ATOM_PPD_DEV_USD)
-#define ATOM_PPD_TOPO_SEC_DSD	(ATOM_PPD_CONN_RP | ATOM_PPD_DEV_DSD)
-#define ATOM_PPD_TOPO_B2B_USD	(ATOM_PPD_CONN_B2B | ATOM_PPD_DEV_USD)
-#define ATOM_PPD_TOPO_B2B_DSD	(ATOM_PPD_CONN_B2B | ATOM_PPD_DEV_DSD)
-
-#define ATOM_MW_COUNT			2
-#define ATOM_DB_COUNT			34
-#define ATOM_DB_VALID_MASK		(BIT_ULL(ATOM_DB_COUNT) - 1)
-#define ATOM_DB_MSIX_VECTOR_COUNT	34
-#define ATOM_DB_MSIX_VECTOR_SHIFT	1
-#define ATOM_DB_TOTAL_SHIFT		34
-#define ATOM_SPAD_COUNT			16
-
-#define ATOM_NTB_CTL_DOWN_BIT		BIT(16)
-#define ATOM_NTB_CTL_ACTIVE(x)		!(x & ATOM_NTB_CTL_DOWN_BIT)
-
-#define ATOM_DESKEWSTS_DBERR		BIT(15)
-#define ATOM_LTSSMERRSTS0_UNEXPECTEDEI	BIT(20)
-#define ATOM_LTSSMSTATEJMP_FORCEDETECT	BIT(2)
-#define ATOM_IBIST_ERR_OFLOW		0x7FFF7FFF
-
-#define ATOM_LINK_HB_TIMEOUT		msecs_to_jiffies(1000)
-#define ATOM_LINK_RECOVERY_TIME		msecs_to_jiffies(500)
-
 /* Ntb control and link status */
 
 #define NTB_CTL_CFG_LOCK		BIT(0)

drivers/ntb/hw/mscc/ntb_hw_switchtec.c

@@ -94,6 +94,9 @@ struct switchtec_ntb {
 	struct ntb_ctrl_regs __iomem *mmio_self_ctrl;
 	struct ntb_ctrl_regs __iomem *mmio_peer_ctrl;
 	struct ntb_dbmsg_regs __iomem *mmio_self_dbmsg;
+	struct ntb_dbmsg_regs __iomem *mmio_peer_dbmsg;
+
+	void __iomem *mmio_xlink_win;
 
 	struct shared_mw *self_shared;
 	struct shared_mw __iomem *peer_shared;
@@ -109,6 +112,7 @@ struct switchtec_ntb {
 
 	int nr_direct_mw;
 	int nr_lut_mw;
+	int nr_rsvd_luts;
 	int direct_mw_to_bar[MAX_DIRECT_MW];
 
 	int peer_nr_direct_mw;
@@ -118,6 +122,7 @@ struct switchtec_ntb {
 	bool link_is_up;
 	enum ntb_speed link_speed;
 	enum ntb_width link_width;
+	struct work_struct link_reinit_work;
 };
 
 static struct switchtec_ntb *ntb_sndev(struct ntb_dev *ntb)
@@ -172,7 +177,7 @@ static int switchtec_ntb_part_op(struct switchtec_ntb *sndev,
 
 	if (ps == status) {
 		dev_err(&sndev->stdev->dev,
-			"Timed out while peforming %s (%d). (%08x)",
+			"Timed out while performing %s (%d). (%08x)\n",
 			op_text[op], op,
 			ioread32(&ctl->partition_status));
 
@@ -185,10 +190,10 @@ static int switchtec_ntb_part_op(struct switchtec_ntb *sndev,
 static int switchtec_ntb_send_msg(struct switchtec_ntb *sndev, int idx,
 				  u32 val)
 {
-	if (idx < 0 || idx >= ARRAY_SIZE(sndev->mmio_self_dbmsg->omsg))
+	if (idx < 0 || idx >= ARRAY_SIZE(sndev->mmio_peer_dbmsg->omsg))
 		return -EINVAL;
 
-	iowrite32(val, &sndev->mmio_self_dbmsg->omsg[idx].msg);
+	iowrite32(val, &sndev->mmio_peer_dbmsg->omsg[idx].msg);
 
 	return 0;
 }
@@ -197,7 +202,7 @@ static int switchtec_ntb_mw_count(struct ntb_dev *ntb, int pidx)
 {
 	struct switchtec_ntb *sndev = ntb_sndev(ntb);
 	int nr_direct_mw = sndev->peer_nr_direct_mw;
-	int nr_lut_mw = sndev->peer_nr_lut_mw - 1;
+	int nr_lut_mw = sndev->peer_nr_lut_mw - sndev->nr_rsvd_luts;
 
 	if (pidx != NTB_DEF_PEER_IDX)
 		return -EINVAL;
@@ -210,12 +215,12 @@ static int switchtec_ntb_mw_count(struct ntb_dev *ntb, int pidx)
 
 static int lut_index(struct switchtec_ntb *sndev, int mw_idx)
 {
-	return mw_idx - sndev->nr_direct_mw + 1;
+	return mw_idx - sndev->nr_direct_mw + sndev->nr_rsvd_luts;
 }
 
 static int peer_lut_index(struct switchtec_ntb *sndev, int mw_idx)
 {
-	return mw_idx - sndev->peer_nr_direct_mw + 1;
+	return mw_idx - sndev->peer_nr_direct_mw + sndev->nr_rsvd_luts;
 }
 
 static int switchtec_ntb_mw_get_align(struct ntb_dev *ntb, int pidx,
@@ -306,7 +311,7 @@ static int switchtec_ntb_mw_set_trans(struct ntb_dev *ntb, int pidx, int widx,
 	if (pidx != NTB_DEF_PEER_IDX)
 		return -EINVAL;
 
-	dev_dbg(&sndev->stdev->dev, "MW %d: part %d addr %pad size %pap",
+	dev_dbg(&sndev->stdev->dev, "MW %d: part %d addr %pad size %pap\n",
 		widx, pidx, &addr, &size);
 
 	if (widx >= switchtec_ntb_mw_count(ntb, pidx))
@@ -315,6 +320,19 @@ static int switchtec_ntb_mw_set_trans(struct ntb_dev *ntb, int pidx, int widx,
 	if (xlate_pos < 12)
 		return -EINVAL;
 
+	if (!IS_ALIGNED(addr, BIT_ULL(xlate_pos))) {
+		/*
+		 * In certain circumstances we can get a buffer that is
+		 * not aligned to its size. (Most of the time
+		 * dma_alloc_coherent ensures this). This can happen when
+		 * using large buffers allocated by the CMA
+		 * (see CMA_CONFIG_ALIGNMENT)
+		 */
+		dev_err(&sndev->stdev->dev,
+			"ERROR: Memory window address is not aligned to it's size!\n");
+		return -EINVAL;
+	}
+
 	rc = switchtec_ntb_part_op(sndev, ctl, NTB_CTRL_PART_OP_LOCK,
 				   NTB_CTRL_PART_STATUS_LOCKED);
 	if (rc)
@@ -337,7 +355,7 @@ static int switchtec_ntb_mw_set_trans(struct ntb_dev *ntb, int pidx, int widx,
 
 	if (rc == -EIO) {
 		dev_err(&sndev->stdev->dev,
-			"Hardware reported an error configuring mw %d: %08x",
+			"Hardware reported an error configuring mw %d: %08x\n",
 			widx, ioread32(&ctl->bar_error));
 
 		if (widx < nr_direct_mw)
@@ -355,8 +373,9 @@ static int switchtec_ntb_mw_set_trans(struct ntb_dev *ntb, int pidx, int widx,
 static int switchtec_ntb_peer_mw_count(struct ntb_dev *ntb)
 {
 	struct switchtec_ntb *sndev = ntb_sndev(ntb);
+	int nr_lut_mw = sndev->nr_lut_mw - sndev->nr_rsvd_luts;
 
-	return sndev->nr_direct_mw + (use_lut_mws ? sndev->nr_lut_mw - 1 : 0);
+	return sndev->nr_direct_mw + (use_lut_mws ? nr_lut_mw : 0);
 }
 
 static int switchtec_ntb_direct_get_addr(struct switchtec_ntb *sndev,
@@ -463,18 +482,69 @@ static void switchtec_ntb_set_link_speed(struct switchtec_ntb *sndev)
 	sndev->link_width = min(self_width, peer_width);
 }
 
-enum {
+static int crosslink_is_enabled(struct switchtec_ntb *sndev)
+{
+	struct ntb_info_regs __iomem *inf = sndev->mmio_ntb;
+
+	return ioread8(&inf->ntp_info[sndev->peer_partition].xlink_enabled);
+}
+
+static void crosslink_init_dbmsgs(struct switchtec_ntb *sndev)
+{
+	int i;
+	u32 msg_map = 0;
+
+	if (!crosslink_is_enabled(sndev))
+		return;
+
+	for (i = 0; i < ARRAY_SIZE(sndev->mmio_peer_dbmsg->imsg); i++) {
+		int m = i | sndev->self_partition << 2;
+
+		msg_map |= m << i * 8;
+	}
+
+	iowrite32(msg_map, &sndev->mmio_peer_dbmsg->msg_map);
+	iowrite64(sndev->db_valid_mask << sndev->db_peer_shift,
+		  &sndev->mmio_peer_dbmsg->odb_mask);
+}
+
+enum switchtec_msg {
 	LINK_MESSAGE = 0,
 	MSG_LINK_UP = 1,
 	MSG_LINK_DOWN = 2,
 	MSG_CHECK_LINK = 3,
+	MSG_LINK_FORCE_DOWN = 4,
 };
 
-static void switchtec_ntb_check_link(struct switchtec_ntb *sndev)
+static int switchtec_ntb_reinit_peer(struct switchtec_ntb *sndev);
+
+static void link_reinit_work(struct work_struct *work)
+{
+	struct switchtec_ntb *sndev;
+
+	sndev = container_of(work, struct switchtec_ntb, link_reinit_work);
+
+	switchtec_ntb_reinit_peer(sndev);
+}
+
+static void switchtec_ntb_check_link(struct switchtec_ntb *sndev,
+				     enum switchtec_msg msg)
 {
 	int link_sta;
 	int old = sndev->link_is_up;
 
+	if (msg == MSG_LINK_FORCE_DOWN) {
+		schedule_work(&sndev->link_reinit_work);
+
+		if (sndev->link_is_up) {
+			sndev->link_is_up = 0;
+			ntb_link_event(&sndev->ntb);
+			dev_info(&sndev->stdev->dev, "ntb link forced down\n");
+		}
+
+		return;
+	}
+
 	link_sta = sndev->self_shared->link_sta;
 	if (link_sta) {
 		u64 peer = ioread64(&sndev->peer_shared->magic);
@@ -491,8 +561,11 @@ static void switchtec_ntb_check_link(struct switchtec_ntb *sndev)
 	if (link_sta != old) {
 		switchtec_ntb_send_msg(sndev, LINK_MESSAGE, MSG_CHECK_LINK);
 		ntb_link_event(&sndev->ntb);
-		dev_info(&sndev->stdev->dev, "ntb link %s",
+		dev_info(&sndev->stdev->dev, "ntb link %s\n",
 			 link_sta ? "up" : "down");
+
+		if (link_sta)
+			crosslink_init_dbmsgs(sndev);
 	}
 }
 
@@ -500,7 +573,7 @@ static void switchtec_ntb_link_notification(struct switchtec_dev *stdev)
 {
 	struct switchtec_ntb *sndev = stdev->sndev;
 
-	switchtec_ntb_check_link(sndev);
+	switchtec_ntb_check_link(sndev, MSG_CHECK_LINK);
 }
 
 static u64 switchtec_ntb_link_is_up(struct ntb_dev *ntb,
@@ -523,12 +596,12 @@ static int switchtec_ntb_link_enable(struct ntb_dev *ntb,
 {
 	struct switchtec_ntb *sndev = ntb_sndev(ntb);
 
-	dev_dbg(&sndev->stdev->dev, "enabling link");
+	dev_dbg(&sndev->stdev->dev, "enabling link\n");
 
 	sndev->self_shared->link_sta = 1;
 	switchtec_ntb_send_msg(sndev, LINK_MESSAGE, MSG_LINK_UP);
 
-	switchtec_ntb_check_link(sndev);
+	switchtec_ntb_check_link(sndev, MSG_CHECK_LINK);
 
 	return 0;
 }
@@ -537,12 +610,12 @@ static int switchtec_ntb_link_disable(struct ntb_dev *ntb)
 {
 	struct switchtec_ntb *sndev = ntb_sndev(ntb);
 
-	dev_dbg(&sndev->stdev->dev, "disabling link");
+	dev_dbg(&sndev->stdev->dev, "disabling link\n");
 
 	sndev->self_shared->link_sta = 0;
-	switchtec_ntb_send_msg(sndev, LINK_MESSAGE, MSG_LINK_UP);
+	switchtec_ntb_send_msg(sndev, LINK_MESSAGE, MSG_LINK_DOWN);
 
-	switchtec_ntb_check_link(sndev);
+	switchtec_ntb_check_link(sndev, MSG_CHECK_LINK);
 
 	return 0;
 }
@@ -638,7 +711,7 @@ static int switchtec_ntb_peer_db_addr(struct ntb_dev *ntb,
 	struct switchtec_ntb *sndev = ntb_sndev(ntb);
 	unsigned long offset;
 
-	offset = (unsigned long)sndev->mmio_self_dbmsg->odb -
+	offset = (unsigned long)sndev->mmio_peer_dbmsg->odb -
 		(unsigned long)sndev->stdev->mmio;
 
 	offset += sndev->db_shift / 8;
@@ -656,7 +729,7 @@ static int switchtec_ntb_peer_db_set(struct ntb_dev *ntb, u64 db_bits)
 	struct switchtec_ntb *sndev = ntb_sndev(ntb);
 
 	iowrite64(db_bits << sndev->db_peer_shift,
-		  &sndev->mmio_self_dbmsg->odb);
+		  &sndev->mmio_peer_dbmsg->odb);
 
 	return 0;
 }
@@ -777,24 +850,63 @@ static const struct ntb_dev_ops switchtec_ntb_ops = {
 	.peer_spad_addr		= switchtec_ntb_peer_spad_addr,
 };
 
-static void switchtec_ntb_init_sndev(struct switchtec_ntb *sndev)
+static int switchtec_ntb_init_sndev(struct switchtec_ntb *sndev)
 {
+	u64 tpart_vec;
+	int self;
 	u64 part_map;
+	int bit;
 
 	sndev->ntb.pdev = sndev->stdev->pdev;
 	sndev->ntb.topo = NTB_TOPO_SWITCH;
 	sndev->ntb.ops = &switchtec_ntb_ops;
 
+	INIT_WORK(&sndev->link_reinit_work, link_reinit_work);
+
 	sndev->self_partition = sndev->stdev->partition;
 
 	sndev->mmio_ntb = sndev->stdev->mmio_ntb;
+
+	self = sndev->self_partition;
+	tpart_vec = ioread32(&sndev->mmio_ntb->ntp_info[self].target_part_high);
+	tpart_vec <<= 32;
+	tpart_vec |= ioread32(&sndev->mmio_ntb->ntp_info[self].target_part_low);
+
 	part_map = ioread64(&sndev->mmio_ntb->ep_map);
 	part_map &= ~(1 << sndev->self_partition);
-	sndev->peer_partition = ffs(part_map) - 1;
 
-	dev_dbg(&sndev->stdev->dev, "Partition ID %d of %d (%llx)",
-		sndev->self_partition, sndev->stdev->partition_count,
-		part_map);
+	if (!ffs(tpart_vec)) {
+		if (sndev->stdev->partition_count != 2) {
+			dev_err(&sndev->stdev->dev,
+				"ntb target partition not defined\n");
+			return -ENODEV;
+		}
+
+		bit = ffs(part_map);
+		if (!bit) {
+			dev_err(&sndev->stdev->dev,
+				"peer partition is not NT partition\n");
+			return -ENODEV;
+		}
+
+		sndev->peer_partition = bit - 1;
+	} else {
+		if (ffs(tpart_vec) != fls(tpart_vec)) {
+			dev_err(&sndev->stdev->dev,
+				"ntb driver only supports 1 pair of 1-1 ntb mapping\n");
+			return -ENODEV;
+		}
+
+		sndev->peer_partition = ffs(tpart_vec) - 1;
+		if (!(part_map & (1 << sndev->peer_partition))) {
+			dev_err(&sndev->stdev->dev,
+				"ntb target partition is not NT partition\n");
+			return -ENODEV;
+		}
+	}
+
+	dev_dbg(&sndev->stdev->dev, "Partition ID %d of %d\n",
+		sndev->self_partition, sndev->stdev->partition_count);
 
 	sndev->mmio_ctrl = (void * __iomem)sndev->mmio_ntb +
 		SWITCHTEC_NTB_REG_CTRL_OFFSET;
@@ -804,6 +916,283 @@ static void switchtec_ntb_init_sndev(struct switchtec_ntb *sndev)
 	sndev->mmio_self_ctrl = &sndev->mmio_ctrl[sndev->self_partition];
 	sndev->mmio_peer_ctrl = &sndev->mmio_ctrl[sndev->peer_partition];
 	sndev->mmio_self_dbmsg = &sndev->mmio_dbmsg[sndev->self_partition];
+	sndev->mmio_peer_dbmsg = sndev->mmio_self_dbmsg;
+
+	return 0;
+}
+
+static int config_rsvd_lut_win(struct switchtec_ntb *sndev,
+			       struct ntb_ctrl_regs __iomem *ctl,
+			       int lut_idx, int partition, u64 addr)
+{
+	int peer_bar = sndev->peer_direct_mw_to_bar[0];
+	u32 ctl_val;
+	int rc;
+
+	rc = switchtec_ntb_part_op(sndev, ctl, NTB_CTRL_PART_OP_LOCK,
+				   NTB_CTRL_PART_STATUS_LOCKED);
+	if (rc)
+		return rc;
+
+	ctl_val = ioread32(&ctl->bar_entry[peer_bar].ctl);
+	ctl_val &= 0xFF;
+	ctl_val |= NTB_CTRL_BAR_LUT_WIN_EN;
+	ctl_val |= ilog2(LUT_SIZE) << 8;
+	ctl_val |= (sndev->nr_lut_mw - 1) << 14;
+	iowrite32(ctl_val, &ctl->bar_entry[peer_bar].ctl);
+
+	iowrite64((NTB_CTRL_LUT_EN | (partition << 1) | addr),
+		  &ctl->lut_entry[lut_idx]);
+
+	rc = switchtec_ntb_part_op(sndev, ctl, NTB_CTRL_PART_OP_CFG,
+				   NTB_CTRL_PART_STATUS_NORMAL);
+	if (rc) {
+		u32 bar_error, lut_error;
+
+		bar_error = ioread32(&ctl->bar_error);
+		lut_error = ioread32(&ctl->lut_error);
+		dev_err(&sndev->stdev->dev,
+			"Error setting up reserved lut window: %08x / %08x\n",
+			bar_error, lut_error);
+		return rc;
+	}
+
+	return 0;
+}
+
+static int config_req_id_table(struct switchtec_ntb *sndev,
+			       struct ntb_ctrl_regs __iomem *mmio_ctrl,
+			       int *req_ids, int count)
+{
+	int i, rc = 0;
+	u32 error;
+	u32 proxy_id;
+
+	if (ioread32(&mmio_ctrl->req_id_table_size) < count) {
+		dev_err(&sndev->stdev->dev,
+			"Not enough requester IDs available.\n");
+		return -EFAULT;
+	}
+
+	rc = switchtec_ntb_part_op(sndev, mmio_ctrl,
+				   NTB_CTRL_PART_OP_LOCK,
+				   NTB_CTRL_PART_STATUS_LOCKED);
+	if (rc)
+		return rc;
+
+	iowrite32(NTB_PART_CTRL_ID_PROT_DIS,
+		  &mmio_ctrl->partition_ctrl);
+
+	for (i = 0; i < count; i++) {
+		iowrite32(req_ids[i] << 16 | NTB_CTRL_REQ_ID_EN,
+			  &mmio_ctrl->req_id_table[i]);
+
+		proxy_id = ioread32(&mmio_ctrl->req_id_table[i]);
+		dev_dbg(&sndev->stdev->dev,
+			"Requester ID %02X:%02X.%X -> BB:%02X.%X\n",
+			req_ids[i] >> 8, (req_ids[i] >> 3) & 0x1F,
+			req_ids[i] & 0x7, (proxy_id >> 4) & 0x1F,
+			(proxy_id >> 1) & 0x7);
+	}
+
+	rc = switchtec_ntb_part_op(sndev, mmio_ctrl,
+				   NTB_CTRL_PART_OP_CFG,
+				   NTB_CTRL_PART_STATUS_NORMAL);
+
+	if (rc == -EIO) {
+		error = ioread32(&mmio_ctrl->req_id_error);
+		dev_err(&sndev->stdev->dev,
+			"Error setting up the requester ID table: %08x\n",
+			error);
+	}
+
+	return 0;
+}
+
+static int crosslink_setup_mws(struct switchtec_ntb *sndev, int ntb_lut_idx,
+			       u64 *mw_addrs, int mw_count)
+{
+	int rc, i;
+	struct ntb_ctrl_regs __iomem *ctl = sndev->mmio_self_ctrl;
+	u64 addr;
+	size_t size, offset;
+	int bar;
+	int xlate_pos;
+	u32 ctl_val;
+
+	rc = switchtec_ntb_part_op(sndev, ctl, NTB_CTRL_PART_OP_LOCK,
+				   NTB_CTRL_PART_STATUS_LOCKED);
+	if (rc)
+		return rc;
+
+	for (i = 0; i < sndev->nr_lut_mw; i++) {
+		if (i == ntb_lut_idx)
+			continue;
+
+		addr = mw_addrs[0] + LUT_SIZE * i;
+
+		iowrite64((NTB_CTRL_LUT_EN | (sndev->peer_partition << 1) |
+			   addr),
+			  &ctl->lut_entry[i]);
+	}
+
+	sndev->nr_direct_mw = min_t(int, sndev->nr_direct_mw, mw_count);
+
+	for (i = 0; i < sndev->nr_direct_mw; i++) {
+		bar = sndev->direct_mw_to_bar[i];
+		offset = (i == 0) ? LUT_SIZE * sndev->nr_lut_mw : 0;
+		addr = mw_addrs[i] + offset;
+		size = pci_resource_len(sndev->ntb.pdev, bar) - offset;
+		xlate_pos = ilog2(size);
+
+		if (offset && size > offset)
+			size = offset;
+
+		ctl_val = ioread32(&ctl->bar_entry[bar].ctl);
+		ctl_val |= NTB_CTRL_BAR_DIR_WIN_EN;
+
+		iowrite32(ctl_val, &ctl->bar_entry[bar].ctl);
+		iowrite32(xlate_pos | size, &ctl->bar_entry[bar].win_size);
+		iowrite64(sndev->peer_partition | addr,
+			  &ctl->bar_entry[bar].xlate_addr);
+	}
+
+	rc = switchtec_ntb_part_op(sndev, ctl, NTB_CTRL_PART_OP_CFG,
+				   NTB_CTRL_PART_STATUS_NORMAL);
+	if (rc) {
+		u32 bar_error, lut_error;
+
+		bar_error = ioread32(&ctl->bar_error);
+		lut_error = ioread32(&ctl->lut_error);
+		dev_err(&sndev->stdev->dev,
+			"Error setting up cross link windows: %08x / %08x\n",
+			bar_error, lut_error);
+		return rc;
+	}
+
+	return 0;
+}
+
+static int crosslink_setup_req_ids(struct switchtec_ntb *sndev,
+	struct ntb_ctrl_regs __iomem *mmio_ctrl)
+{
+	int req_ids[16];
+	int i;
+	u32 proxy_id;
+
+	for (i = 0; i < ARRAY_SIZE(req_ids); i++) {
+		proxy_id = ioread32(&sndev->mmio_self_ctrl->req_id_table[i]);
+
+		if (!(proxy_id & NTB_CTRL_REQ_ID_EN))
+			break;
+
+		req_ids[i] = ((proxy_id >> 1) & 0xFF);
+	}
+
+	return config_req_id_table(sndev, mmio_ctrl, req_ids, i);
+}
+
+/*
+ * In crosslink configuration there is a virtual partition in the
+ * middle of the two switches. The BARs in this partition have to be
+ * enumerated and assigned addresses.
+ */
+static int crosslink_enum_partition(struct switchtec_ntb *sndev,
+				    u64 *bar_addrs)
+{
+	struct part_cfg_regs __iomem *part_cfg =
+		&sndev->stdev->mmio_part_cfg_all[sndev->peer_partition];
+	u32 pff = ioread32(&part_cfg->vep_pff_inst_id);
+	struct pff_csr_regs __iomem *mmio_pff =
+		&sndev->stdev->mmio_pff_csr[pff];
+	const u64 bar_space = 0x1000000000LL;
+	u64 bar_addr;
+	int bar_cnt = 0;
+	int i;
+
+	iowrite16(0x6, &mmio_pff->pcicmd);
+
+	for (i = 0; i < ARRAY_SIZE(mmio_pff->pci_bar64); i++) {
+		iowrite64(bar_space * i, &mmio_pff->pci_bar64[i]);
+		bar_addr = ioread64(&mmio_pff->pci_bar64[i]);
+		bar_addr &= ~0xf;
+
+		dev_dbg(&sndev->stdev->dev,
+			"Crosslink BAR%d addr: %llx\n",
+			i, bar_addr);
+
+		if (bar_addr != bar_space * i)
+			continue;
+
+		bar_addrs[bar_cnt++] = bar_addr;
+	}
+
+	return bar_cnt;
+}
+
+static int switchtec_ntb_init_crosslink(struct switchtec_ntb *sndev)
+{
+	int rc;
+	int bar = sndev->direct_mw_to_bar[0];
+	const int ntb_lut_idx = 1;
+	u64 bar_addrs[6];
+	u64 addr;
+	int offset;
+	int bar_cnt;
+
+	if (!crosslink_is_enabled(sndev))
+		return 0;
+
+	dev_info(&sndev->stdev->dev, "Using crosslink configuration\n");
+	sndev->ntb.topo = NTB_TOPO_CROSSLINK;
+
+	bar_cnt = crosslink_enum_partition(sndev, bar_addrs);
+	if (bar_cnt < sndev->nr_direct_mw + 1) {
+		dev_err(&sndev->stdev->dev,
+			"Error enumerating crosslink partition\n");
+		return -EINVAL;
+	}
+
+	addr = (bar_addrs[0] + SWITCHTEC_GAS_NTB_OFFSET +
+		SWITCHTEC_NTB_REG_DBMSG_OFFSET +
+		sizeof(struct ntb_dbmsg_regs) * sndev->peer_partition);
+
+	offset = addr & (LUT_SIZE - 1);
+	addr -= offset;
+
+	rc = config_rsvd_lut_win(sndev, sndev->mmio_self_ctrl, ntb_lut_idx,
+				 sndev->peer_partition, addr);
+	if (rc)
+		return rc;
+
+	rc = crosslink_setup_mws(sndev, ntb_lut_idx, &bar_addrs[1],
+				 bar_cnt - 1);
+	if (rc)
+		return rc;
+
+	rc = crosslink_setup_req_ids(sndev, sndev->mmio_peer_ctrl);
+	if (rc)
+		return rc;
+
+	sndev->mmio_xlink_win = pci_iomap_range(sndev->stdev->pdev, bar,
+						LUT_SIZE, LUT_SIZE);
+	if (!sndev->mmio_xlink_win) {
+		rc = -ENOMEM;
+		return rc;
+	}
+
+	sndev->mmio_peer_dbmsg = sndev->mmio_xlink_win + offset;
+	sndev->nr_rsvd_luts++;
+
+	crosslink_init_dbmsgs(sndev);
+
+	return 0;
+}
+
+static void switchtec_ntb_deinit_crosslink(struct switchtec_ntb *sndev)
+{
+	if (sndev->mmio_xlink_win)
+		pci_iounmap(sndev->stdev->pdev, sndev->mmio_xlink_win);
 }
 
 static int map_bars(int *map, struct ntb_ctrl_regs __iomem *ctrl)
@@ -829,7 +1218,7 @@ static void switchtec_ntb_init_mw(struct switchtec_ntb *sndev)
 	sndev->nr_lut_mw = ioread16(&sndev->mmio_self_ctrl->lut_table_entries);
 	sndev->nr_lut_mw = rounddown_pow_of_two(sndev->nr_lut_mw);
 
-	dev_dbg(&sndev->stdev->dev, "MWs: %d direct, %d lut",
+	dev_dbg(&sndev->stdev->dev, "MWs: %d direct, %d lut\n",
 		sndev->nr_direct_mw, sndev->nr_lut_mw);
 
 	sndev->peer_nr_direct_mw = map_bars(sndev->peer_direct_mw_to_bar,
@@ -839,7 +1228,7 @@ static void switchtec_ntb_init_mw(struct switchtec_ntb *sndev)
 		ioread16(&sndev->mmio_peer_ctrl->lut_table_entries);
 	sndev->peer_nr_lut_mw = rounddown_pow_of_two(sndev->peer_nr_lut_mw);
 
-	dev_dbg(&sndev->stdev->dev, "Peer MWs: %d direct, %d lut",
+	dev_dbg(&sndev->stdev->dev, "Peer MWs: %d direct, %d lut\n",
 		sndev->peer_nr_direct_mw, sndev->peer_nr_lut_mw);
 
 }
@@ -849,24 +1238,35 @@ static void switchtec_ntb_init_mw(struct switchtec_ntb *sndev)
  * shared among all partitions. So we must split them in half
  * (32 for each partition). However, the message interrupts are
  * also shared with the top 4 doorbells so we just limit this to
- * 28 doorbells per partition
+ * 28 doorbells per partition.
+ *
+ * In crosslink mode, each side has it's own dbmsg register so
+ * they can each use all 60 of the available doorbells.
  */
 static void switchtec_ntb_init_db(struct switchtec_ntb *sndev)
 {
-	sndev->db_valid_mask = 0x0FFFFFFF;
+	sndev->db_mask = 0x0FFFFFFFFFFFFFFFULL;
 
-	if (sndev->self_partition < sndev->peer_partition) {
+	if (sndev->mmio_peer_dbmsg != sndev->mmio_self_dbmsg) {
+		sndev->db_shift = 0;
+		sndev->db_peer_shift = 0;
+		sndev->db_valid_mask = sndev->db_mask;
+	} else if (sndev->self_partition < sndev->peer_partition) {
 		sndev->db_shift = 0;
 		sndev->db_peer_shift = 32;
+		sndev->db_valid_mask = 0x0FFFFFFF;
 	} else {
 		sndev->db_shift = 32;
 		sndev->db_peer_shift = 0;
+		sndev->db_valid_mask = 0x0FFFFFFF;
 	}
 
-	sndev->db_mask = 0x0FFFFFFFFFFFFFFFULL;
 	iowrite64(~sndev->db_mask, &sndev->mmio_self_dbmsg->idb_mask);
 	iowrite64(sndev->db_valid_mask << sndev->db_peer_shift,
-		  &sndev->mmio_self_dbmsg->odb_mask);
+		  &sndev->mmio_peer_dbmsg->odb_mask);
+
+	dev_dbg(&sndev->stdev->dev, "dbs: shift %d/%d, mask %016llx\n",
+		sndev->db_shift, sndev->db_peer_shift, sndev->db_valid_mask);
 }
 
 static void switchtec_ntb_init_msgs(struct switchtec_ntb *sndev)
@@ -887,52 +1287,23 @@ static void switchtec_ntb_init_msgs(struct switchtec_ntb *sndev)
 			  &sndev->mmio_self_dbmsg->imsg[i]);
 }
 
-static int switchtec_ntb_init_req_id_table(struct switchtec_ntb *sndev)
+static int
+switchtec_ntb_init_req_id_table(struct switchtec_ntb *sndev)
 {
-	int rc = 0;
-	u16 req_id;
-	u32 error;
-
-	req_id = ioread16(&sndev->mmio_ntb->requester_id);
-
-	if (ioread32(&sndev->mmio_self_ctrl->req_id_table_size) < 2) {
-		dev_err(&sndev->stdev->dev,
-			"Not enough requester IDs available.");
-		return -EFAULT;
-	}
-
-	rc = switchtec_ntb_part_op(sndev, sndev->mmio_self_ctrl,
-				   NTB_CTRL_PART_OP_LOCK,
-				   NTB_CTRL_PART_STATUS_LOCKED);
-	if (rc)
-		return rc;
-
-	iowrite32(NTB_PART_CTRL_ID_PROT_DIS,
-		  &sndev->mmio_self_ctrl->partition_ctrl);
+	int req_ids[2];
 
 	/*
 	 * Root Complex Requester ID (which is 0:00.0)
 	 */
-	iowrite32(0 << 16 | NTB_CTRL_REQ_ID_EN,
-		  &sndev->mmio_self_ctrl->req_id_table[0]);
+	req_ids[0] = 0;
 
 	/*
 	 * Host Bridge Requester ID (as read from the mmap address)
 	 */
-	iowrite32(req_id << 16 | NTB_CTRL_REQ_ID_EN,
-		  &sndev->mmio_self_ctrl->req_id_table[1]);
-
-	rc = switchtec_ntb_part_op(sndev, sndev->mmio_self_ctrl,
-				   NTB_CTRL_PART_OP_CFG,
-				   NTB_CTRL_PART_STATUS_NORMAL);
-	if (rc == -EIO) {
-		error = ioread32(&sndev->mmio_self_ctrl->req_id_error);
-		dev_err(&sndev->stdev->dev,
-			"Error setting up the requester ID table: %08x",
-			error);
-	}
+	req_ids[1] = ioread16(&sndev->mmio_ntb->requester_id);
 
-	return rc;
+	return config_req_id_table(sndev, sndev->mmio_self_ctrl, req_ids,
+				   ARRAY_SIZE(req_ids));
 }
 
 static void switchtec_ntb_init_shared(struct switchtec_ntb *sndev)
@@ -963,59 +1334,35 @@ static void switchtec_ntb_init_shared(struct switchtec_ntb *sndev)
 
 static int switchtec_ntb_init_shared_mw(struct switchtec_ntb *sndev)
 {
-	struct ntb_ctrl_regs __iomem *ctl = sndev->mmio_peer_ctrl;
-	int bar = sndev->direct_mw_to_bar[0];
-	u32 ctl_val;
+	int self_bar = sndev->direct_mw_to_bar[0];
 	int rc;
 
+	sndev->nr_rsvd_luts++;
 	sndev->self_shared = dma_zalloc_coherent(&sndev->stdev->pdev->dev,
 						 LUT_SIZE,
 						 &sndev->self_shared_dma,
 						 GFP_KERNEL);
 	if (!sndev->self_shared) {
 		dev_err(&sndev->stdev->dev,
-			"unable to allocate memory for shared mw");
+			"unable to allocate memory for shared mw\n");
 		return -ENOMEM;
 	}
 
 	switchtec_ntb_init_shared(sndev);
 
-	rc = switchtec_ntb_part_op(sndev, ctl, NTB_CTRL_PART_OP_LOCK,
-				   NTB_CTRL_PART_STATUS_LOCKED);
+	rc = config_rsvd_lut_win(sndev, sndev->mmio_peer_ctrl, 0,
+				 sndev->self_partition,
+				 sndev->self_shared_dma);
 	if (rc)
 		goto unalloc_and_exit;
 
-	ctl_val = ioread32(&ctl->bar_entry[bar].ctl);
-	ctl_val &= 0xFF;
-	ctl_val |= NTB_CTRL_BAR_LUT_WIN_EN;
-	ctl_val |= ilog2(LUT_SIZE) << 8;
-	ctl_val |= (sndev->nr_lut_mw - 1) << 14;
-	iowrite32(ctl_val, &ctl->bar_entry[bar].ctl);
-
-	iowrite64((NTB_CTRL_LUT_EN | (sndev->self_partition << 1) |
-		   sndev->self_shared_dma),
-		  &ctl->lut_entry[0]);
-
-	rc = switchtec_ntb_part_op(sndev, ctl, NTB_CTRL_PART_OP_CFG,
-				   NTB_CTRL_PART_STATUS_NORMAL);
-	if (rc) {
-		u32 bar_error, lut_error;
-
-		bar_error = ioread32(&ctl->bar_error);
-		lut_error = ioread32(&ctl->lut_error);
-		dev_err(&sndev->stdev->dev,
-			"Error setting up shared MW: %08x / %08x",
-			bar_error, lut_error);
-		goto unalloc_and_exit;
-	}
-
-	sndev->peer_shared = pci_iomap(sndev->stdev->pdev, bar, LUT_SIZE);
+	sndev->peer_shared = pci_iomap(sndev->stdev->pdev, self_bar, LUT_SIZE);
 	if (!sndev->peer_shared) {
 		rc = -ENOMEM;
 		goto unalloc_and_exit;
 	}
 
-	dev_dbg(&sndev->stdev->dev, "Shared MW Ready");
+	dev_dbg(&sndev->stdev->dev, "Shared MW Ready\n");
 	return 0;
 
 unalloc_and_exit:
@@ -1034,6 +1381,7 @@ static void switchtec_ntb_deinit_shared_mw(struct switchtec_ntb *sndev)
 		dma_free_coherent(&sndev->stdev->pdev->dev, LUT_SIZE,
 				  sndev->self_shared,
 				  sndev->self_shared_dma);
+	sndev->nr_rsvd_luts--;
 }
 
 static irqreturn_t switchtec_ntb_doorbell_isr(int irq, void *dev)
@@ -1056,12 +1404,12 @@ static irqreturn_t switchtec_ntb_message_isr(int irq, void *dev)
 		u64 msg = ioread64(&sndev->mmio_self_dbmsg->imsg[i]);
 
 		if (msg & NTB_DBMSG_IMSG_STATUS) {
-			dev_dbg(&sndev->stdev->dev, "message: %d %08x\n", i,
-				(u32)msg);
+			dev_dbg(&sndev->stdev->dev, "message: %d %08x\n",
+				i, (u32)msg);
 			iowrite8(1, &sndev->mmio_self_dbmsg->imsg[i].status);
 
 			if (i == LINK_MESSAGE)
-				switchtec_ntb_check_link(sndev);
+				switchtec_ntb_check_link(sndev, msg);
 		}
 	}
 
@@ -1085,7 +1433,7 @@ static int switchtec_ntb_init_db_msg_irq(struct switchtec_ntb *sndev)
 	       message_irq == event_irq)
 		message_irq++;
 
-	dev_dbg(&sndev->stdev->dev, "irqs - event: %d, db: %d, msgs: %d",
+	dev_dbg(&sndev->stdev->dev, "irqs - event: %d, db: %d, msgs: %d\n",
 		event_irq, doorbell_irq, message_irq);
 
 	for (i = 0; i < idb_vecs - 4; i++)
@@ -1122,6 +1470,14 @@ static void switchtec_ntb_deinit_db_msg_irq(struct switchtec_ntb *sndev)
 	free_irq(sndev->message_irq, sndev);
 }
 
+static int switchtec_ntb_reinit_peer(struct switchtec_ntb *sndev)
+{
+	dev_info(&sndev->stdev->dev, "peer reinitialized\n");
+	switchtec_ntb_deinit_shared_mw(sndev);
+	switchtec_ntb_init_mw(sndev);
+	return switchtec_ntb_init_shared_mw(sndev);
+}
+
 static int switchtec_ntb_add(struct device *dev,
 			     struct class_interface *class_intf)
 {
@@ -1134,38 +1490,50 @@ static int switchtec_ntb_add(struct device *dev,
 	if (stdev->pdev->class != MICROSEMI_NTB_CLASSCODE)
 		return -ENODEV;
 
-	if (stdev->partition_count != 2)
-		dev_warn(dev, "ntb driver only supports 2 partitions");
-
 	sndev = kzalloc_node(sizeof(*sndev), GFP_KERNEL, dev_to_node(dev));
 	if (!sndev)
 		return -ENOMEM;
 
 	sndev->stdev = stdev;
-	switchtec_ntb_init_sndev(sndev);
+	rc = switchtec_ntb_init_sndev(sndev);
+	if (rc)
+		goto free_and_exit;
+
 	switchtec_ntb_init_mw(sndev);
-	switchtec_ntb_init_db(sndev);
-	switchtec_ntb_init_msgs(sndev);
 
 	rc = switchtec_ntb_init_req_id_table(sndev);
 	if (rc)
 		goto free_and_exit;
 
-	rc = switchtec_ntb_init_shared_mw(sndev);
+	rc = switchtec_ntb_init_crosslink(sndev);
 	if (rc)
 		goto free_and_exit;
 
+	switchtec_ntb_init_db(sndev);
+	switchtec_ntb_init_msgs(sndev);
+
+	rc = switchtec_ntb_init_shared_mw(sndev);
+	if (rc)
+		goto deinit_crosslink;
+
 	rc = switchtec_ntb_init_db_msg_irq(sndev);
 	if (rc)
 		goto deinit_shared_and_exit;
 
+	/*
+	 * If this host crashed, the other host may think the link is
+	 * still up. Tell them to force it down (it will go back up
+	 * once we register the ntb device).
+	 */
+	switchtec_ntb_send_msg(sndev, LINK_MESSAGE, MSG_LINK_FORCE_DOWN);
+
 	rc = ntb_register_device(&sndev->ntb);
 	if (rc)
 		goto deinit_and_exit;
 
 	stdev->sndev = sndev;
 	stdev->link_notifier = switchtec_ntb_link_notification;
-	dev_info(dev, "NTB device registered");
+	dev_info(dev, "NTB device registered\n");
 
 	return 0;
 
@@ -1173,14 +1541,16 @@ deinit_and_exit:
 	switchtec_ntb_deinit_db_msg_irq(sndev);
 deinit_shared_and_exit:
 	switchtec_ntb_deinit_shared_mw(sndev);
+deinit_crosslink:
+	switchtec_ntb_deinit_crosslink(sndev);
 free_and_exit:
 	kfree(sndev);
-	dev_err(dev, "failed to register ntb device: %d", rc);
+	dev_err(dev, "failed to register ntb device: %d\n", rc);
 	return rc;
 }
 
-void switchtec_ntb_remove(struct device *dev,
-			  struct class_interface *class_intf)
+static void switchtec_ntb_remove(struct device *dev,
+				 struct class_interface *class_intf)
 {
 	struct switchtec_dev *stdev = to_stdev(dev);
 	struct switchtec_ntb *sndev = stdev->sndev;
@@ -1193,8 +1563,9 @@ void switchtec_ntb_remove(struct device *dev,
 	ntb_unregister_device(&sndev->ntb);
 	switchtec_ntb_deinit_db_msg_irq(sndev);
 	switchtec_ntb_deinit_shared_mw(sndev);
+	switchtec_ntb_deinit_crosslink(sndev);
 	kfree(sndev);
-	dev_info(dev, "ntb device unregistered");
+	dev_info(dev, "ntb device unregistered\n");
 }
 
 static struct class_interface switchtec_interface  = {

drivers/ntb/ntb.c

@@ -63,12 +63,11 @@
 #define DRIVER_NAME			"ntb"
 #define DRIVER_DESCRIPTION		"PCIe NTB Driver Framework"
 
-#define DRIVER_LICENSE			"Dual BSD/GPL"
 #define DRIVER_VERSION			"1.0"
 #define DRIVER_RELDATE			"24 March 2015"
 #define DRIVER_AUTHOR			"Allen Hubbe <Allen.Hubbe@emc.com>"
 
-MODULE_LICENSE(DRIVER_LICENSE);
+MODULE_LICENSE("Dual BSD/GPL");
 MODULE_VERSION(DRIVER_VERSION);
 MODULE_AUTHOR(DRIVER_AUTHOR);
 MODULE_DESCRIPTION(DRIVER_DESCRIPTION);
@@ -112,7 +111,6 @@ int ntb_register_device(struct ntb_dev *ntb)
 
 	init_completion(&ntb->released);
 
-	memset(&ntb->dev, 0, sizeof(ntb->dev));
 	ntb->dev.bus = &ntb_bus;
 	ntb->dev.parent = &ntb->pdev->dev;
 	ntb->dev.release = ntb_dev_release;

drivers/ntb/ntb_transport.c

@@ -1003,6 +1003,9 @@ static int ntb_transport_init_queue(struct ntb_transport_ctx *nt,
 	mw_base = nt->mw_vec[mw_num].phys_addr;
 	mw_size = nt->mw_vec[mw_num].phys_size;
 
+	if (max_mw_size && mw_size > max_mw_size)
+		mw_size = max_mw_size;
+
 	tx_size = (unsigned int)mw_size / num_qps_mw;
 	qp_offset = tx_size * (qp_num / mw_count);
 

drivers/ntb/test/ntb_perf.c

@@ -5,6 +5,7 @@
  *   GPL LICENSE SUMMARY
  *
  *   Copyright(c) 2015 Intel Corporation. All rights reserved.
+ *   Copyright(c) 2017 T-Platforms. All Rights Reserved.
  *
  *   This program is free software; you can redistribute it and/or modify
  *   it under the terms of version 2 of the GNU General Public License as
@@ -13,6 +14,7 @@
  *   BSD LICENSE
  *
  *   Copyright(c) 2015 Intel Corporation. All rights reserved.
+ *   Copyright(c) 2017 T-Platforms. All Rights Reserved.
  *
  *   Redistribution and use in source and binary forms, with or without
  *   modification, are permitted provided that the following conditions
@@ -40,860 +42,1474 @@
  *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  *
- *   PCIe NTB Perf Linux driver
+ * PCIe NTB Perf Linux driver
+ */
+
+/*
+ * How to use this tool, by example.
+ *
+ * Assuming $DBG_DIR is something like:
+ * '/sys/kernel/debug/ntb_perf/0000:00:03.0'
+ * Suppose aside from local device there is at least one remote device
+ * connected to NTB with index 0.
+ *-----------------------------------------------------------------------------
+ * Eg: install driver with specified chunk/total orders and dma-enabled flag
+ *
+ * root@self# insmod ntb_perf.ko chunk_order=19 total_order=28 use_dma
+ *-----------------------------------------------------------------------------
+ * Eg: check NTB ports (index) and MW mapping information
+ *
+ * root@self# cat $DBG_DIR/info
+ *-----------------------------------------------------------------------------
+ * Eg: start performance test with peer (index 0) and get the test metrics
+ *
+ * root@self# echo 0 > $DBG_DIR/run
+ * root@self# cat $DBG_DIR/run
  */
 
 #include <linux/init.h>
 #include <linux/kernel.h>
 #include <linux/module.h>
-#include <linux/kthread.h>
-#include <linux/time.h>
-#include <linux/timer.h>
+#include <linux/sched.h>
+#include <linux/wait.h>
 #include <linux/dma-mapping.h>
+#include <linux/dmaengine.h>
 #include <linux/pci.h>
+#include <linux/ktime.h>
 #include <linux/slab.h>
-#include <linux/spinlock.h>
-#include <linux/debugfs.h>
-#include <linux/dmaengine.h>
 #include <linux/delay.h>
 #include <linux/sizes.h>
+#include <linux/workqueue.h>
+#include <linux/debugfs.h>
+#include <linux/random.h>
 #include <linux/ntb.h>
-#include <linux/mutex.h>
 
 #define DRIVER_NAME		"ntb_perf"
-#define DRIVER_DESCRIPTION	"PCIe NTB Performance Measurement Tool"
-
-#define DRIVER_LICENSE		"Dual BSD/GPL"
-#define DRIVER_VERSION		"1.0"
-#define DRIVER_AUTHOR		"Dave Jiang <dave.jiang@intel.com>"
-
-#define PERF_LINK_DOWN_TIMEOUT	10
-#define PERF_VERSION		0xffff0001
-#define MAX_THREADS		32
-#define MAX_TEST_SIZE		SZ_1M
-#define MAX_SRCS		32
-#define DMA_OUT_RESOURCE_TO	msecs_to_jiffies(50)
-#define DMA_RETRIES		20
-#define SZ_4G			(1ULL << 32)
-#define MAX_SEG_ORDER		20 /* no larger than 1M for kmalloc buffer */
-#define PIDX			NTB_DEF_PEER_IDX
-
-MODULE_LICENSE(DRIVER_LICENSE);
+#define DRIVER_VERSION		"2.0"
+
+MODULE_LICENSE("Dual BSD/GPL");
 MODULE_VERSION(DRIVER_VERSION);
-MODULE_AUTHOR(DRIVER_AUTHOR);
-MODULE_DESCRIPTION(DRIVER_DESCRIPTION);
+MODULE_AUTHOR("Dave Jiang <dave.jiang@intel.com>");
+MODULE_DESCRIPTION("PCIe NTB Performance Measurement Tool");
+
+#define MAX_THREADS_CNT		32
+#define DEF_THREADS_CNT		1
+#define MAX_CHUNK_SIZE		SZ_1M
+#define MAX_CHUNK_ORDER		20 /* no larger than 1M */
 
-static struct dentry *perf_debugfs_dir;
+#define DMA_TRIES		100
+#define DMA_MDELAY		10
+
+#define MSG_TRIES		500
+#define MSG_UDELAY_LOW		1000
+#define MSG_UDELAY_HIGH		2000
+
+#define PERF_BUF_LEN 1024
 
 static unsigned long max_mw_size;
 module_param(max_mw_size, ulong, 0644);
-MODULE_PARM_DESC(max_mw_size, "Limit size of large memory windows");
+MODULE_PARM_DESC(max_mw_size, "Upper limit of memory window size");
 
-static unsigned int seg_order = 19; /* 512K */
-module_param(seg_order, uint, 0644);
-MODULE_PARM_DESC(seg_order, "size order [2^n] of buffer segment for testing");
+static unsigned char chunk_order = 19; /* 512K */
+module_param(chunk_order, byte, 0644);
+MODULE_PARM_DESC(chunk_order, "Data chunk order [2^n] to transfer");
 
-static unsigned int run_order = 32; /* 4G */
-module_param(run_order, uint, 0644);
-MODULE_PARM_DESC(run_order, "size order [2^n] of total data to transfer");
+static unsigned char total_order = 30; /* 1G */
+module_param(total_order, byte, 0644);
+MODULE_PARM_DESC(total_order, "Total data order [2^n] to transfer");
 
 static bool use_dma; /* default to 0 */
 module_param(use_dma, bool, 0644);
-MODULE_PARM_DESC(use_dma, "Using DMA engine to measure performance");
-
-static bool on_node = true; /* default to 1 */
-module_param(on_node, bool, 0644);
-MODULE_PARM_DESC(on_node, "Run threads only on NTB device node (default: true)");
-
-struct perf_mw {
-	phys_addr_t	phys_addr;
-	resource_size_t	phys_size;
-	void __iomem	*vbase;
-	size_t		xlat_size;
-	size_t		buf_size;
-	void		*virt_addr;
-	dma_addr_t	dma_addr;
+MODULE_PARM_DESC(use_dma, "Use DMA engine to measure performance");
+
+/*==============================================================================
+ *                         Perf driver data definition
+ *==============================================================================
+ */
+
+enum perf_cmd {
+	PERF_CMD_INVAL = -1,/* invalid spad command */
+	PERF_CMD_SSIZE = 0, /* send out buffer size */
+	PERF_CMD_RSIZE = 1, /* recv in  buffer size */
+	PERF_CMD_SXLAT = 2, /* send in  buffer xlat */
+	PERF_CMD_RXLAT = 3, /* recv out buffer xlat */
+	PERF_CMD_CLEAR = 4, /* clear allocated memory */
+	PERF_STS_DONE  = 5, /* init is done */
+	PERF_STS_LNKUP = 6, /* link up state flag */
 };
 
 struct perf_ctx;
 
-struct pthr_ctx {
-	struct task_struct	*thread;
-	struct perf_ctx		*perf;
-	atomic_t		dma_sync;
-	struct dma_chan		*dma_chan;
-	int			dma_prep_err;
-	int			src_idx;
-	void			*srcs[MAX_SRCS];
-	wait_queue_head_t       *wq;
-	int			status;
-	u64			copied;
-	u64			diff_us;
+struct perf_peer {
+	struct perf_ctx	*perf;
+	int pidx;
+	int gidx;
+
+	/* Outbound MW params */
+	u64 outbuf_xlat;
+	resource_size_t outbuf_size;
+	void __iomem *outbuf;
+
+	/* Inbound MW params */
+	dma_addr_t inbuf_xlat;
+	resource_size_t inbuf_size;
+	void		*inbuf;
+
+	/* NTB connection setup service */
+	struct work_struct	service;
+	unsigned long		sts;
 };
+#define to_peer_service(__work) \
+	container_of(__work, struct perf_peer, service)
 
-struct perf_ctx {
-	struct ntb_dev		*ntb;
-	spinlock_t		db_lock;
-	struct perf_mw		mw;
-	bool			link_is_up;
-	struct delayed_work	link_work;
-	wait_queue_head_t	link_wq;
-	u8			perf_threads;
-	/* mutex ensures only one set of threads run at once */
-	struct mutex		run_mutex;
-	struct pthr_ctx		pthr_ctx[MAX_THREADS];
-	atomic_t		tsync;
-	atomic_t                tdone;
+struct perf_thread {
+	struct perf_ctx *perf;
+	int tidx;
+
+	/* DMA-based test sync parameters */
+	atomic_t dma_sync;
+	wait_queue_head_t dma_wait;
+	struct dma_chan *dma_chan;
+
+	/* Data source and measured statistics */
+	void *src;
+	u64 copied;
+	ktime_t duration;
+	int status;
+	struct work_struct work;
 };
+#define to_thread_work(__work) \
+	container_of(__work, struct perf_thread, work)
 
-enum {
-	VERSION = 0,
-	MW_SZ_HIGH,
-	MW_SZ_LOW,
-	MAX_SPAD
+struct perf_ctx {
+	struct ntb_dev *ntb;
+
+	/* Global device index and peers descriptors */
+	int gidx;
+	int pcnt;
+	struct perf_peer *peers;
+
+	/* Performance measuring work-threads interface */
+	unsigned long busy_flag;
+	wait_queue_head_t twait;
+	atomic_t tsync;
+	u8 tcnt;
+	struct perf_peer *test_peer;
+	struct perf_thread threads[MAX_THREADS_CNT];
+
+	/* Scratchpad/Message IO operations */
+	int (*cmd_send)(struct perf_peer *peer, enum perf_cmd cmd, u64 data);
+	int (*cmd_recv)(struct perf_ctx *perf, int *pidx, enum perf_cmd *cmd,
+			u64 *data);
+
+	struct dentry *dbgfs_dir;
 };
 
+/*
+ * Scratchpads-base commands interface
+ */
+#define PERF_SPAD_CNT(_pcnt) \
+	(3*((_pcnt) + 1))
+#define PERF_SPAD_CMD(_gidx) \
+	(3*(_gidx))
+#define PERF_SPAD_LDATA(_gidx) \
+	(3*(_gidx) + 1)
+#define PERF_SPAD_HDATA(_gidx) \
+	(3*(_gidx) + 2)
+#define PERF_SPAD_NOTIFY(_gidx) \
+	(BIT_ULL(_gidx))
+
+/*
+ * Messages-base commands interface
+ */
+#define PERF_MSG_CNT		3
+#define PERF_MSG_CMD		0
+#define PERF_MSG_LDATA		1
+#define PERF_MSG_HDATA		2
+
+/*==============================================================================
+ *                           Static data declarations
+ *==============================================================================
+ */
+
+static struct dentry *perf_dbgfs_topdir;
+
+static struct workqueue_struct *perf_wq __read_mostly;
+
+/*==============================================================================
+ *                  NTB cross-link commands execution service
+ *==============================================================================
+ */
+
+static void perf_terminate_test(struct perf_ctx *perf);
+
+static inline bool perf_link_is_up(struct perf_peer *peer)
+{
+	u64 link;
+
+	link = ntb_link_is_up(peer->perf->ntb, NULL, NULL);
+	return !!(link & BIT_ULL_MASK(peer->pidx));
+}
+
+static int perf_spad_cmd_send(struct perf_peer *peer, enum perf_cmd cmd,
+			      u64 data)
+{
+	struct perf_ctx *perf = peer->perf;
+	int try;
+	u32 sts;
+
+	dev_dbg(&perf->ntb->dev, "CMD send: %d 0x%llx\n", cmd, data);
+
+	/*
+	 * Perform predefined number of attempts before give up.
+	 * We are sending the data to the port specific scratchpad, so
+	 * to prevent a multi-port access race-condition. Additionally
+	 * there is no need in local locking since only thread-safe
+	 * service work is using this method.
+	 */
+	for (try = 0; try < MSG_TRIES; try++) {
+		if (!perf_link_is_up(peer))
+			return -ENOLINK;
+
+		sts = ntb_peer_spad_read(perf->ntb, peer->pidx,
+					 PERF_SPAD_CMD(perf->gidx));
+		if (sts != PERF_CMD_INVAL) {
+			usleep_range(MSG_UDELAY_LOW, MSG_UDELAY_HIGH);
+			continue;
+		}
+
+		ntb_peer_spad_write(perf->ntb, peer->pidx,
+				    PERF_SPAD_LDATA(perf->gidx),
+				    lower_32_bits(data));
+		ntb_peer_spad_write(perf->ntb, peer->pidx,
+				    PERF_SPAD_HDATA(perf->gidx),
+				    upper_32_bits(data));
+		mmiowb();
+		ntb_peer_spad_write(perf->ntb, peer->pidx,
+				    PERF_SPAD_CMD(perf->gidx),
+				    cmd);
+		mmiowb();
+		ntb_peer_db_set(perf->ntb, PERF_SPAD_NOTIFY(peer->gidx));
+
+		dev_dbg(&perf->ntb->dev, "DB ring peer %#llx\n",
+			PERF_SPAD_NOTIFY(peer->gidx));
+
+		break;
+	}
+
+	return try < MSG_TRIES ? 0 : -EAGAIN;
+}
+
+static int perf_spad_cmd_recv(struct perf_ctx *perf, int *pidx,
+			      enum perf_cmd *cmd, u64 *data)
+{
+	struct perf_peer *peer;
+	u32 val;
+
+	ntb_db_clear(perf->ntb, PERF_SPAD_NOTIFY(perf->gidx));
+
+	/*
+	 * We start scanning all over, since cleared DB may have been set
+	 * by any peer. Yes, it makes peer with smaller index being
+	 * serviced with greater priority, but it's convenient for spad
+	 * and message code unification and simplicity.
+	 */
+	for (*pidx = 0; *pidx < perf->pcnt; (*pidx)++) {
+		peer = &perf->peers[*pidx];
+
+		if (!perf_link_is_up(peer))
+			continue;
+
+		val = ntb_spad_read(perf->ntb, PERF_SPAD_CMD(peer->gidx));
+		if (val == PERF_CMD_INVAL)
+			continue;
+
+		*cmd = val;
+
+		val = ntb_spad_read(perf->ntb, PERF_SPAD_LDATA(peer->gidx));
+		*data = val;
+
+		val = ntb_spad_read(perf->ntb, PERF_SPAD_HDATA(peer->gidx));
+		*data |= (u64)val << 32;
+
+		/* Next command can be retrieved from now */
+		ntb_spad_write(perf->ntb, PERF_SPAD_CMD(peer->gidx),
+			       PERF_CMD_INVAL);
+
+		dev_dbg(&perf->ntb->dev, "CMD recv: %d 0x%llx\n", *cmd, *data);
+
+		return 0;
+	}
+
+	return -ENODATA;
+}
+
+static int perf_msg_cmd_send(struct perf_peer *peer, enum perf_cmd cmd,
+			     u64 data)
+{
+	struct perf_ctx *perf = peer->perf;
+	int try, ret;
+	u64 outbits;
+
+	dev_dbg(&perf->ntb->dev, "CMD send: %d 0x%llx\n", cmd, data);
+
+	/*
+	 * Perform predefined number of attempts before give up. Message
+	 * registers are free of race-condition problem when accessed
+	 * from different ports, so we don't need splitting registers
+	 * by global device index. We also won't have local locking,
+	 * since the method is used from service work only.
+	 */
+	outbits = ntb_msg_outbits(perf->ntb);
+	for (try = 0; try < MSG_TRIES; try++) {
+		if (!perf_link_is_up(peer))
+			return -ENOLINK;
+
+		ret = ntb_msg_clear_sts(perf->ntb, outbits);
+		if (ret)
+			return ret;
+
+		ntb_peer_msg_write(perf->ntb, peer->pidx, PERF_MSG_LDATA,
+				   lower_32_bits(data));
+
+		if (ntb_msg_read_sts(perf->ntb) & outbits) {
+			usleep_range(MSG_UDELAY_LOW, MSG_UDELAY_HIGH);
+			continue;
+		}
+
+		ntb_peer_msg_write(perf->ntb, peer->pidx, PERF_MSG_HDATA,
+				   upper_32_bits(data));
+		mmiowb();
+
+		/* This call shall trigger peer message event */
+		ntb_peer_msg_write(perf->ntb, peer->pidx, PERF_MSG_CMD, cmd);
+
+		break;
+	}
+
+	return try < MSG_TRIES ? 0 : -EAGAIN;
+}
+
+static int perf_msg_cmd_recv(struct perf_ctx *perf, int *pidx,
+			     enum perf_cmd *cmd, u64 *data)
+{
+	u64 inbits;
+	u32 val;
+
+	inbits = ntb_msg_inbits(perf->ntb);
+
+	if (hweight64(ntb_msg_read_sts(perf->ntb) & inbits) < 3)
+		return -ENODATA;
+
+	val = ntb_msg_read(perf->ntb, pidx, PERF_MSG_CMD);
+	*cmd = val;
+
+	val = ntb_msg_read(perf->ntb, pidx, PERF_MSG_LDATA);
+	*data = val;
+
+	val = ntb_msg_read(perf->ntb, pidx, PERF_MSG_HDATA);
+	*data |= (u64)val << 32;
+
+	/* Next command can be retrieved from now */
+	ntb_msg_clear_sts(perf->ntb, inbits);
+
+	dev_dbg(&perf->ntb->dev, "CMD recv: %d 0x%llx\n", *cmd, *data);
+
+	return 0;
+}
+
+static int perf_cmd_send(struct perf_peer *peer, enum perf_cmd cmd, u64 data)
+{
+	struct perf_ctx *perf = peer->perf;
+
+	if (cmd == PERF_CMD_SSIZE || cmd == PERF_CMD_SXLAT)
+		return perf->cmd_send(peer, cmd, data);
+
+	dev_err(&perf->ntb->dev, "Send invalid command\n");
+	return -EINVAL;
+}
+
+static int perf_cmd_exec(struct perf_peer *peer, enum perf_cmd cmd)
+{
+	switch (cmd) {
+	case PERF_CMD_SSIZE:
+	case PERF_CMD_RSIZE:
+	case PERF_CMD_SXLAT:
+	case PERF_CMD_RXLAT:
+	case PERF_CMD_CLEAR:
+		break;
+	default:
+		dev_err(&peer->perf->ntb->dev, "Exec invalid command\n");
+		return -EINVAL;
+	}
+
+	/* No need of memory barrier, since bit ops have invernal lock */
+	set_bit(cmd, &peer->sts);
+
+	dev_dbg(&peer->perf->ntb->dev, "CMD exec: %d\n", cmd);
+
+	(void)queue_work(system_highpri_wq, &peer->service);
+
+	return 0;
+}
+
+static int perf_cmd_recv(struct perf_ctx *perf)
+{
+	struct perf_peer *peer;
+	int ret, pidx, cmd;
+	u64 data;
+
+	while (!(ret = perf->cmd_recv(perf, &pidx, &cmd, &data))) {
+		peer = &perf->peers[pidx];
+
+		switch (cmd) {
+		case PERF_CMD_SSIZE:
+			peer->inbuf_size = data;
+			return perf_cmd_exec(peer, PERF_CMD_RSIZE);
+		case PERF_CMD_SXLAT:
+			peer->outbuf_xlat = data;
+			return perf_cmd_exec(peer, PERF_CMD_RXLAT);
+		default:
+			dev_err(&perf->ntb->dev, "Recv invalid command\n");
+			return -EINVAL;
+		}
+	}
+
+	/* Return 0 if no data left to process, otherwise an error */
+	return ret == -ENODATA ? 0 : ret;
+}
+
 static void perf_link_event(void *ctx)
 {
 	struct perf_ctx *perf = ctx;
+	struct perf_peer *peer;
+	bool lnk_up;
+	int pidx;
 
-	if (ntb_link_is_up(perf->ntb, NULL, NULL) == 1) {
-		schedule_delayed_work(&perf->link_work, 2*HZ);
-	} else {
-		dev_dbg(&perf->ntb->pdev->dev, "link down\n");
+	for (pidx = 0; pidx < perf->pcnt; pidx++) {
+		peer = &perf->peers[pidx];
 
-		if (!perf->link_is_up)
-			cancel_delayed_work_sync(&perf->link_work);
+		lnk_up = perf_link_is_up(peer);
 
-		perf->link_is_up = false;
+		if (lnk_up &&
+		    !test_and_set_bit(PERF_STS_LNKUP, &peer->sts)) {
+			perf_cmd_exec(peer, PERF_CMD_SSIZE);
+		} else if (!lnk_up &&
+			   test_and_clear_bit(PERF_STS_LNKUP, &peer->sts)) {
+			perf_cmd_exec(peer, PERF_CMD_CLEAR);
+		}
 	}
 }
 
 static void perf_db_event(void *ctx, int vec)
 {
 	struct perf_ctx *perf = ctx;
-	u64 db_bits, db_mask;
 
-	db_mask = ntb_db_vector_mask(perf->ntb, vec);
-	db_bits = ntb_db_read(perf->ntb);
+	dev_dbg(&perf->ntb->dev, "DB vec %d mask %#llx bits %#llx\n", vec,
+		ntb_db_vector_mask(perf->ntb, vec), ntb_db_read(perf->ntb));
+
+	/* Just receive all available commands */
+	(void)perf_cmd_recv(perf);
+}
+
+static void perf_msg_event(void *ctx)
+{
+	struct perf_ctx *perf = ctx;
+
+	dev_dbg(&perf->ntb->dev, "Msg status bits %#llx\n",
+		ntb_msg_read_sts(perf->ntb));
 
-	dev_dbg(&perf->ntb->dev, "doorbell vec %d mask %#llx bits %#llx\n",
-		vec, db_mask, db_bits);
+	/* Messages are only sent one-by-one */
+	(void)perf_cmd_recv(perf);
 }
 
 static const struct ntb_ctx_ops perf_ops = {
 	.link_event = perf_link_event,
 	.db_event = perf_db_event,
+	.msg_event = perf_msg_event
 };
 
-static void perf_copy_callback(void *data)
+static void perf_free_outbuf(struct perf_peer *peer)
+{
+	(void)ntb_peer_mw_clear_trans(peer->perf->ntb, peer->pidx, peer->gidx);
+}
+
+static int perf_setup_outbuf(struct perf_peer *peer)
 {
-	struct pthr_ctx *pctx = data;
+	struct perf_ctx *perf = peer->perf;
+	int ret;
+
+	/* Outbuf size can be unaligned due to custom max_mw_size */
+	ret = ntb_peer_mw_set_trans(perf->ntb, peer->pidx, peer->gidx,
+				    peer->outbuf_xlat, peer->outbuf_size);
+	if (ret) {
+		dev_err(&perf->ntb->dev, "Failed to set outbuf translation\n");
+		return ret;
+	}
 
-	atomic_dec(&pctx->dma_sync);
+	/* Initialization is finally done */
+	set_bit(PERF_STS_DONE, &peer->sts);
+
+	return 0;
 }
 
-static ssize_t perf_copy(struct pthr_ctx *pctx, char __iomem *dst,
-			 char *src, size_t size)
+static void perf_free_inbuf(struct perf_peer *peer)
 {
-	struct perf_ctx *perf = pctx->perf;
-	struct dma_async_tx_descriptor *txd;
-	struct dma_chan *chan = pctx->dma_chan;
-	struct dma_device *device;
-	struct dmaengine_unmap_data *unmap;
-	dma_cookie_t cookie;
-	size_t src_off, dst_off;
-	struct perf_mw *mw = &perf->mw;
-	void __iomem *vbase;
-	void __iomem *dst_vaddr;
-	dma_addr_t dst_phys;
-	int retries = 0;
+	if (!peer->inbuf)
+		return;
 
-	if (!use_dma) {
-		memcpy_toio(dst, src, size);
-		return size;
+	(void)ntb_mw_clear_trans(peer->perf->ntb, peer->pidx, peer->gidx);
+	dma_free_coherent(&peer->perf->ntb->dev, peer->inbuf_size,
+			  peer->inbuf, peer->inbuf_xlat);
+	peer->inbuf = NULL;
+}
+
+static int perf_setup_inbuf(struct perf_peer *peer)
+{
+	resource_size_t xlat_align, size_align, size_max;
+	struct perf_ctx *perf = peer->perf;
+	int ret;
+
+	/* Get inbound MW parameters */
+	ret = ntb_mw_get_align(perf->ntb, peer->pidx, perf->gidx,
+			       &xlat_align, &size_align, &size_max);
+	if (ret) {
+		dev_err(&perf->ntb->dev, "Couldn't get inbuf restrictions\n");
+		return ret;
 	}
 
-	if (!chan) {
-		dev_err(&perf->ntb->dev, "DMA engine does not exist\n");
+	if (peer->inbuf_size > size_max) {
+		dev_err(&perf->ntb->dev, "Too big inbuf size %pa > %pa\n",
+			&peer->inbuf_size, &size_max);
 		return -EINVAL;
 	}
 
-	device = chan->device;
-	src_off = (uintptr_t)src & ~PAGE_MASK;
-	dst_off = (uintptr_t __force)dst & ~PAGE_MASK;
-
-	if (!is_dma_copy_aligned(device, src_off, dst_off, size))
-		return -ENODEV;
+	peer->inbuf_size = round_up(peer->inbuf_size, size_align);
 
-	vbase = mw->vbase;
-	dst_vaddr = dst;
-	dst_phys = mw->phys_addr + (dst_vaddr - vbase);
+	perf_free_inbuf(peer);
 
-	unmap = dmaengine_get_unmap_data(device->dev, 1, GFP_NOWAIT);
-	if (!unmap)
+	peer->inbuf = dma_alloc_coherent(&perf->ntb->dev, peer->inbuf_size,
+					 &peer->inbuf_xlat, GFP_KERNEL);
+	if (!peer->inbuf) {
+		dev_err(&perf->ntb->dev, "Failed to alloc inbuf of %pa\n",
+			&peer->inbuf_size);
 		return -ENOMEM;
+	}
+	if (!IS_ALIGNED(peer->inbuf_xlat, xlat_align)) {
+		dev_err(&perf->ntb->dev, "Unaligned inbuf allocated\n");
+		goto err_free_inbuf;
+	}
 
-	unmap->len = size;
-	unmap->addr[0] = dma_map_page(device->dev, virt_to_page(src),
-				      src_off, size, DMA_TO_DEVICE);
-	if (dma_mapping_error(device->dev, unmap->addr[0]))
-		goto err_get_unmap;
+	ret = ntb_mw_set_trans(perf->ntb, peer->pidx, peer->gidx,
+			       peer->inbuf_xlat, peer->inbuf_size);
+	if (ret) {
+		dev_err(&perf->ntb->dev, "Failed to set inbuf translation\n");
+		goto err_free_inbuf;
+	}
 
-	unmap->to_cnt = 1;
+	/*
+	 * We submit inbuf xlat transmission cmd for execution here to follow
+	 * the code architecture, even though this method is called from service
+	 * work itself so the command will be executed right after it returns.
+	 */
+	(void)perf_cmd_exec(peer, PERF_CMD_SXLAT);
 
-	do {
-		txd = device->device_prep_dma_memcpy(chan, dst_phys,
-						     unmap->addr[0],
-						     size, DMA_PREP_INTERRUPT);
-		if (!txd) {
-			set_current_state(TASK_INTERRUPTIBLE);
-			schedule_timeout(DMA_OUT_RESOURCE_TO);
-		}
-	} while (!txd && (++retries < DMA_RETRIES));
+	return 0;
 
-	if (!txd) {
-		pctx->dma_prep_err++;
-		goto err_get_unmap;
-	}
+err_free_inbuf:
+	perf_free_inbuf(peer);
 
-	txd->callback = perf_copy_callback;
-	txd->callback_param = pctx;
-	dma_set_unmap(txd, unmap);
+	return ret;
+}
 
-	cookie = dmaengine_submit(txd);
-	if (dma_submit_error(cookie))
-		goto err_set_unmap;
+static void perf_service_work(struct work_struct *work)
+{
+	struct perf_peer *peer = to_peer_service(work);
 
-	dmaengine_unmap_put(unmap);
+	if (test_and_clear_bit(PERF_CMD_SSIZE, &peer->sts))
+		perf_cmd_send(peer, PERF_CMD_SSIZE, peer->outbuf_size);
 
-	atomic_inc(&pctx->dma_sync);
-	dma_async_issue_pending(chan);
+	if (test_and_clear_bit(PERF_CMD_RSIZE, &peer->sts))
+		perf_setup_inbuf(peer);
 
-	return size;
+	if (test_and_clear_bit(PERF_CMD_SXLAT, &peer->sts))
+		perf_cmd_send(peer, PERF_CMD_SXLAT, peer->inbuf_xlat);
 
-err_set_unmap:
-	dmaengine_unmap_put(unmap);
-err_get_unmap:
-	dmaengine_unmap_put(unmap);
-	return 0;
-}
+	if (test_and_clear_bit(PERF_CMD_RXLAT, &peer->sts))
+		perf_setup_outbuf(peer);
 
-static int perf_move_data(struct pthr_ctx *pctx, char __iomem *dst, char *src,
-			  u64 buf_size, u64 win_size, u64 total)
-{
-	int chunks, total_chunks, i;
-	int copied_chunks = 0;
-	u64 copied = 0, result;
-	char __iomem *tmp = dst;
-	u64 perf, diff_us;
-	ktime_t kstart, kstop, kdiff;
-	unsigned long last_sleep = jiffies;
-
-	chunks = div64_u64(win_size, buf_size);
-	total_chunks = div64_u64(total, buf_size);
-	kstart = ktime_get();
-
-	for (i = 0; i < total_chunks; i++) {
-		result = perf_copy(pctx, tmp, src, buf_size);
-		copied += result;
-		copied_chunks++;
-		if (copied_chunks == chunks) {
-			tmp = dst;
-			copied_chunks = 0;
-		} else
-			tmp += buf_size;
-
-		/* Probably should schedule every 5s to prevent soft hang. */
-		if (unlikely((jiffies - last_sleep) > 5 * HZ)) {
-			last_sleep = jiffies;
-			set_current_state(TASK_INTERRUPTIBLE);
-			schedule_timeout(1);
+	if (test_and_clear_bit(PERF_CMD_CLEAR, &peer->sts)) {
+		clear_bit(PERF_STS_DONE, &peer->sts);
+		if (test_bit(0, &peer->perf->busy_flag) &&
+		    peer == peer->perf->test_peer) {
+			dev_warn(&peer->perf->ntb->dev,
+				"Freeing while test on-fly\n");
+			perf_terminate_test(peer->perf);
 		}
+		perf_free_outbuf(peer);
+		perf_free_inbuf(peer);
+	}
+}
+
+static int perf_init_service(struct perf_ctx *perf)
+{
+	u64 mask;
 
-		if (unlikely(kthread_should_stop()))
-			break;
+	if (ntb_peer_mw_count(perf->ntb) < perf->pcnt + 1) {
+		dev_err(&perf->ntb->dev, "Not enough memory windows\n");
+		return -EINVAL;
 	}
 
-	if (use_dma) {
-		pr_debug("%s: All DMA descriptors submitted\n", current->comm);
-		while (atomic_read(&pctx->dma_sync) != 0) {
-			if (kthread_should_stop())
-				break;
-			msleep(20);
-		}
+	if (ntb_msg_count(perf->ntb) >= PERF_MSG_CNT) {
+		perf->cmd_send = perf_msg_cmd_send;
+		perf->cmd_recv = perf_msg_cmd_recv;
+
+		dev_dbg(&perf->ntb->dev, "Message service initialized\n");
+
+		return 0;
 	}
 
-	kstop = ktime_get();
-	kdiff = ktime_sub(kstop, kstart);
-	diff_us = ktime_to_us(kdiff);
+	dev_dbg(&perf->ntb->dev, "Message service unsupported\n");
 
-	pr_debug("%s: copied %llu bytes\n", current->comm, copied);
+	mask = GENMASK_ULL(perf->pcnt, 0);
+	if (ntb_spad_count(perf->ntb) >= PERF_SPAD_CNT(perf->pcnt) &&
+	    (ntb_db_valid_mask(perf->ntb) & mask) == mask) {
+		perf->cmd_send = perf_spad_cmd_send;
+		perf->cmd_recv = perf_spad_cmd_recv;
 
-	pr_debug("%s: lasted %llu usecs\n", current->comm, diff_us);
+		dev_dbg(&perf->ntb->dev, "Scratchpad service initialized\n");
 
-	perf = div64_u64(copied, diff_us);
+		return 0;
+	}
 
-	pr_debug("%s: MBytes/s: %llu\n", current->comm, perf);
+	dev_dbg(&perf->ntb->dev, "Scratchpad service unsupported\n");
 
-	pctx->copied = copied;
-	pctx->diff_us = diff_us;
+	dev_err(&perf->ntb->dev, "Command services unsupported\n");
 
-	return 0;
+	return -EINVAL;
 }
 
-static bool perf_dma_filter_fn(struct dma_chan *chan, void *node)
+static int perf_enable_service(struct perf_ctx *perf)
 {
-	/* Is the channel required to be on the same node as the device? */
-	if (!on_node)
-		return true;
+	u64 mask, incmd_bit;
+	int ret, sidx, scnt;
 
-	return dev_to_node(&chan->dev->device) == (int)(unsigned long)node;
-}
+	mask = ntb_db_valid_mask(perf->ntb);
+	(void)ntb_db_set_mask(perf->ntb, mask);
 
-static int ntb_perf_thread(void *data)
-{
-	struct pthr_ctx *pctx = data;
-	struct perf_ctx *perf = pctx->perf;
-	struct pci_dev *pdev = perf->ntb->pdev;
-	struct perf_mw *mw = &perf->mw;
-	char __iomem *dst;
-	u64 win_size, buf_size, total;
-	void *src;
-	int rc, node, i;
-	struct dma_chan *dma_chan = NULL;
+	ret = ntb_set_ctx(perf->ntb, perf, &perf_ops);
+	if (ret)
+		return ret;
 
-	pr_debug("kthread %s starting...\n", current->comm);
+	if (perf->cmd_send == perf_msg_cmd_send) {
+		u64 inbits, outbits;
 
-	node = on_node ? dev_to_node(&pdev->dev) : NUMA_NO_NODE;
+		inbits = ntb_msg_inbits(perf->ntb);
+		outbits = ntb_msg_outbits(perf->ntb);
+		(void)ntb_msg_set_mask(perf->ntb, inbits | outbits);
 
-	if (use_dma && !pctx->dma_chan) {
-		dma_cap_mask_t dma_mask;
+		incmd_bit = BIT_ULL(__ffs64(inbits));
+		ret = ntb_msg_clear_mask(perf->ntb, incmd_bit);
 
-		dma_cap_zero(dma_mask);
-		dma_cap_set(DMA_MEMCPY, dma_mask);
-		dma_chan = dma_request_channel(dma_mask, perf_dma_filter_fn,
-					       (void *)(unsigned long)node);
-		if (!dma_chan) {
-			pr_warn("%s: cannot acquire DMA channel, quitting\n",
-				current->comm);
-			return -ENODEV;
-		}
-		pctx->dma_chan = dma_chan;
+		dev_dbg(&perf->ntb->dev, "MSG sts unmasked %#llx\n", incmd_bit);
+	} else {
+		scnt = ntb_spad_count(perf->ntb);
+		for (sidx = 0; sidx < scnt; sidx++)
+			ntb_spad_write(perf->ntb, sidx, PERF_CMD_INVAL);
+		incmd_bit = PERF_SPAD_NOTIFY(perf->gidx);
+		ret = ntb_db_clear_mask(perf->ntb, incmd_bit);
+
+		dev_dbg(&perf->ntb->dev, "DB bits unmasked %#llx\n", incmd_bit);
+	}
+	if (ret) {
+		ntb_clear_ctx(perf->ntb);
+		return ret;
 	}
 
-	for (i = 0; i < MAX_SRCS; i++) {
-		pctx->srcs[i] = kmalloc_node(MAX_TEST_SIZE, GFP_KERNEL, node);
-		if (!pctx->srcs[i]) {
-			rc = -ENOMEM;
-			goto err;
-		}
+	ntb_link_enable(perf->ntb, NTB_SPEED_AUTO, NTB_WIDTH_AUTO);
+	/* Might be not necessary */
+	ntb_link_event(perf->ntb);
+
+	return 0;
+}
+
+static void perf_disable_service(struct perf_ctx *perf)
+{
+	int pidx;
+
+	ntb_link_disable(perf->ntb);
+
+	if (perf->cmd_send == perf_msg_cmd_send) {
+		u64 inbits;
+
+		inbits = ntb_msg_inbits(perf->ntb);
+		(void)ntb_msg_set_mask(perf->ntb, inbits);
+	} else {
+		(void)ntb_db_set_mask(perf->ntb, PERF_SPAD_NOTIFY(perf->gidx));
 	}
 
-	win_size = mw->phys_size;
-	buf_size = 1ULL << seg_order;
-	total = 1ULL << run_order;
+	ntb_clear_ctx(perf->ntb);
 
-	if (buf_size > MAX_TEST_SIZE)
-		buf_size = MAX_TEST_SIZE;
+	for (pidx = 0; pidx < perf->pcnt; pidx++)
+		perf_cmd_exec(&perf->peers[pidx], PERF_CMD_CLEAR);
 
-	dst = (char __iomem *)mw->vbase;
+	for (pidx = 0; pidx < perf->pcnt; pidx++)
+		flush_work(&perf->peers[pidx].service);
+}
 
-	atomic_inc(&perf->tsync);
-	while (atomic_read(&perf->tsync) != perf->perf_threads)
-		schedule();
+/*==============================================================================
+ *                      Performance measuring work-thread
+ *==============================================================================
+ */
 
-	src = pctx->srcs[pctx->src_idx];
-	pctx->src_idx = (pctx->src_idx + 1) & (MAX_SRCS - 1);
+static void perf_dma_copy_callback(void *data)
+{
+	struct perf_thread *pthr = data;
 
-	rc = perf_move_data(pctx, dst, src, buf_size, win_size, total);
+	atomic_dec(&pthr->dma_sync);
+	wake_up(&pthr->dma_wait);
+}
 
-	atomic_dec(&perf->tsync);
+static int perf_copy_chunk(struct perf_thread *pthr,
+			   void __iomem *dst, void *src, size_t len)
+{
+	struct dma_async_tx_descriptor *tx;
+	struct dmaengine_unmap_data *unmap;
+	struct device *dma_dev;
+	int try = 0, ret = 0;
 
-	if (rc < 0) {
-		pr_err("%s: failed\n", current->comm);
-		rc = -ENXIO;
-		goto err;
+	if (!use_dma) {
+		memcpy_toio(dst, src, len);
+		goto ret_check_tsync;
 	}
 
-	for (i = 0; i < MAX_SRCS; i++) {
-		kfree(pctx->srcs[i]);
-		pctx->srcs[i] = NULL;
+	dma_dev = pthr->dma_chan->device->dev;
+
+	if (!is_dma_copy_aligned(pthr->dma_chan->device, offset_in_page(src),
+				 offset_in_page(dst), len))
+		return -EIO;
+
+	unmap = dmaengine_get_unmap_data(dma_dev, 2, GFP_NOWAIT);
+	if (!unmap)
+		return -ENOMEM;
+
+	unmap->len = len;
+	unmap->addr[0] = dma_map_page(dma_dev, virt_to_page(src),
+		offset_in_page(src), len, DMA_TO_DEVICE);
+	if (dma_mapping_error(dma_dev, unmap->addr[0])) {
+		ret = -EIO;
+		goto err_free_resource;
 	}
+	unmap->to_cnt = 1;
 
-	atomic_inc(&perf->tdone);
-	wake_up(pctx->wq);
-	rc = 0;
-	goto done;
+	unmap->addr[1] = dma_map_page(dma_dev, virt_to_page(dst),
+		offset_in_page(dst), len, DMA_FROM_DEVICE);
+	if (dma_mapping_error(dma_dev, unmap->addr[1])) {
+		ret = -EIO;
+		goto err_free_resource;
+	}
+	unmap->from_cnt = 1;
 
-err:
-	for (i = 0; i < MAX_SRCS; i++) {
-		kfree(pctx->srcs[i]);
-		pctx->srcs[i] = NULL;
+	do {
+		tx = dmaengine_prep_dma_memcpy(pthr->dma_chan, unmap->addr[1],
+			unmap->addr[0], len, DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
+		if (!tx)
+			msleep(DMA_MDELAY);
+	} while (!tx && (try++ < DMA_TRIES));
+
+	if (!tx) {
+		ret = -EIO;
+		goto err_free_resource;
 	}
 
-	if (dma_chan) {
-		dma_release_channel(dma_chan);
-		pctx->dma_chan = NULL;
+	tx->callback = perf_dma_copy_callback;
+	tx->callback_param = pthr;
+	dma_set_unmap(tx, unmap);
+
+	ret = dma_submit_error(dmaengine_submit(tx));
+	if (ret) {
+		dmaengine_unmap_put(unmap);
+		goto err_free_resource;
 	}
 
-done:
-	/* Wait until we are told to stop */
-	for (;;) {
-		set_current_state(TASK_INTERRUPTIBLE);
-		if (kthread_should_stop())
-			break;
-		schedule();
+	dmaengine_unmap_put(unmap);
+
+	atomic_inc(&pthr->dma_sync);
+	dma_async_issue_pending(pthr->dma_chan);
+
+ret_check_tsync:
+	return likely(atomic_read(&pthr->perf->tsync) > 0) ? 0 : -EINTR;
+
+err_free_resource:
+	dmaengine_unmap_put(unmap);
+
+	return ret;
+}
+
+static bool perf_dma_filter(struct dma_chan *chan, void *data)
+{
+	struct perf_ctx *perf = data;
+	int node;
+
+	node = dev_to_node(&perf->ntb->dev);
+
+	return node == NUMA_NO_NODE || node == dev_to_node(chan->device->dev);
+}
+
+static int perf_init_test(struct perf_thread *pthr)
+{
+	struct perf_ctx *perf = pthr->perf;
+	dma_cap_mask_t dma_mask;
+
+	pthr->src = kmalloc_node(perf->test_peer->outbuf_size, GFP_KERNEL,
+				 dev_to_node(&perf->ntb->dev));
+	if (!pthr->src)
+		return -ENOMEM;
+
+	get_random_bytes(pthr->src, perf->test_peer->outbuf_size);
+
+	if (!use_dma)
+		return 0;
+
+	dma_cap_zero(dma_mask);
+	dma_cap_set(DMA_MEMCPY, dma_mask);
+	pthr->dma_chan = dma_request_channel(dma_mask, perf_dma_filter, perf);
+	if (!pthr->dma_chan) {
+		dev_err(&perf->ntb->dev, "%d: Failed to get DMA channel\n",
+			pthr->tidx);
+		atomic_dec(&perf->tsync);
+		wake_up(&perf->twait);
+		kfree(pthr->src);
+		return -ENODEV;
 	}
-	__set_current_state(TASK_RUNNING);
 
-	return rc;
+	atomic_set(&pthr->dma_sync, 0);
+
+	return 0;
 }
 
-static void perf_free_mw(struct perf_ctx *perf)
+static int perf_run_test(struct perf_thread *pthr)
 {
-	struct perf_mw *mw = &perf->mw;
-	struct pci_dev *pdev = perf->ntb->pdev;
+	struct perf_peer *peer = pthr->perf->test_peer;
+	struct perf_ctx *perf = pthr->perf;
+	void __iomem *flt_dst, *bnd_dst;
+	u64 total_size, chunk_size;
+	void *flt_src;
+	int ret = 0;
+
+	total_size = 1ULL << total_order;
+	chunk_size = 1ULL << chunk_order;
+	chunk_size = min_t(u64, peer->outbuf_size, chunk_size);
+
+	flt_src = pthr->src;
+	bnd_dst = peer->outbuf + peer->outbuf_size;
+	flt_dst = peer->outbuf;
+
+	pthr->duration = ktime_get();
+
+	/* Copied field is cleared on test launch stage */
+	while (pthr->copied < total_size) {
+		ret = perf_copy_chunk(pthr, flt_dst, flt_src, chunk_size);
+		if (ret) {
+			dev_err(&perf->ntb->dev, "%d: Got error %d on test\n",
+				pthr->tidx, ret);
+			return ret;
+		}
 
-	if (!mw->virt_addr)
-		return;
+		pthr->copied += chunk_size;
+
+		flt_dst += chunk_size;
+		flt_src += chunk_size;
+		if (flt_dst >= bnd_dst || flt_dst < peer->outbuf) {
+			flt_dst = peer->outbuf;
+			flt_src = pthr->src;
+		}
 
-	ntb_mw_clear_trans(perf->ntb, PIDX, 0);
-	dma_free_coherent(&pdev->dev, mw->buf_size,
-			  mw->virt_addr, mw->dma_addr);
-	mw->xlat_size = 0;
-	mw->buf_size = 0;
-	mw->virt_addr = NULL;
+		/* Give up CPU to give a chance for other threads to use it */
+		schedule();
+	}
+
+	return 0;
 }
 
-static int perf_set_mw(struct perf_ctx *perf, resource_size_t size)
+static int perf_sync_test(struct perf_thread *pthr)
 {
-	struct perf_mw *mw = &perf->mw;
-	size_t xlat_size, buf_size;
-	resource_size_t	xlat_align;
-	resource_size_t	xlat_align_size;
-	int rc;
+	struct perf_ctx *perf = pthr->perf;
 
-	if (!size)
-		return -EINVAL;
+	if (!use_dma)
+		goto no_dma_ret;
 
-	rc = ntb_mw_get_align(perf->ntb, PIDX, 0, &xlat_align,
-			      &xlat_align_size, NULL);
-	if (rc)
-		return rc;
+	wait_event(pthr->dma_wait,
+		   (atomic_read(&pthr->dma_sync) == 0 ||
+		    atomic_read(&perf->tsync) < 0));
 
-	xlat_size = round_up(size, xlat_align_size);
-	buf_size = round_up(size, xlat_align);
+	if (atomic_read(&perf->tsync) < 0)
+		return -EINTR;
 
-	if (mw->xlat_size == xlat_size)
-		return 0;
+no_dma_ret:
+	pthr->duration = ktime_sub(ktime_get(), pthr->duration);
 
-	if (mw->buf_size)
-		perf_free_mw(perf);
+	dev_dbg(&perf->ntb->dev, "%d: copied %llu bytes\n",
+		pthr->tidx, pthr->copied);
 
-	mw->xlat_size = xlat_size;
-	mw->buf_size = buf_size;
+	dev_dbg(&perf->ntb->dev, "%d: lasted %llu usecs\n",
+		pthr->tidx, ktime_to_us(pthr->duration));
+
+	dev_dbg(&perf->ntb->dev, "%d: %llu MBytes/s\n", pthr->tidx,
+		div64_u64(pthr->copied, ktime_to_us(pthr->duration)));
+
+	return 0;
+}
+
+static void perf_clear_test(struct perf_thread *pthr)
+{
+	struct perf_ctx *perf = pthr->perf;
+
+	if (!use_dma)
+		goto no_dma_notify;
+
+	/*
+	 * If test finished without errors, termination isn't needed.
+	 * We call it anyway just to be sure of the transfers completion.
+	 */
+	(void)dmaengine_terminate_sync(pthr->dma_chan);
+
+	dma_release_channel(pthr->dma_chan);
+
+no_dma_notify:
+	atomic_dec(&perf->tsync);
+	wake_up(&perf->twait);
+	kfree(pthr->src);
+}
 
-	mw->virt_addr = dma_alloc_coherent(&perf->ntb->pdev->dev, buf_size,
-					   &mw->dma_addr, GFP_KERNEL);
-	if (!mw->virt_addr) {
-		mw->xlat_size = 0;
-		mw->buf_size = 0;
+static void perf_thread_work(struct work_struct *work)
+{
+	struct perf_thread *pthr = to_thread_work(work);
+	int ret;
+
+	/*
+	 * Perform stages in compliance with use_dma flag value.
+	 * Test status is changed only if error happened, otherwise
+	 * status -ENODATA is kept while test is on-fly. Results
+	 * synchronization is performed only if test fininshed
+	 * without an error or interruption.
+	 */
+	ret = perf_init_test(pthr);
+	if (ret) {
+		pthr->status = ret;
+		return;
 	}
 
-	rc = ntb_mw_set_trans(perf->ntb, PIDX, 0, mw->dma_addr, mw->xlat_size);
-	if (rc) {
-		dev_err(&perf->ntb->dev, "Unable to set mw0 translation\n");
-		perf_free_mw(perf);
-		return -EIO;
+	ret = perf_run_test(pthr);
+	if (ret) {
+		pthr->status = ret;
+		goto err_clear_test;
 	}
 
-	return 0;
+	pthr->status = perf_sync_test(pthr);
+
+err_clear_test:
+	perf_clear_test(pthr);
 }
 
-static void perf_link_work(struct work_struct *work)
+static int perf_set_tcnt(struct perf_ctx *perf, u8 tcnt)
 {
-	struct perf_ctx *perf =
-		container_of(work, struct perf_ctx, link_work.work);
-	struct ntb_dev *ndev = perf->ntb;
-	struct pci_dev *pdev = ndev->pdev;
-	u32 val;
-	u64 size;
-	int rc;
+	if (tcnt == 0 || tcnt > MAX_THREADS_CNT)
+		return -EINVAL;
 
-	dev_dbg(&perf->ntb->pdev->dev, "%s called\n", __func__);
+	if (test_and_set_bit_lock(0, &perf->busy_flag))
+		return -EBUSY;
+
+	perf->tcnt = tcnt;
+
+	clear_bit_unlock(0, &perf->busy_flag);
 
-	size = perf->mw.phys_size;
+	return 0;
+}
 
-	if (max_mw_size && size > max_mw_size)
-		size = max_mw_size;
+static void perf_terminate_test(struct perf_ctx *perf)
+{
+	int tidx;
 
-	ntb_peer_spad_write(ndev, PIDX, MW_SZ_HIGH, upper_32_bits(size));
-	ntb_peer_spad_write(ndev, PIDX, MW_SZ_LOW, lower_32_bits(size));
-	ntb_peer_spad_write(ndev, PIDX, VERSION, PERF_VERSION);
+	atomic_set(&perf->tsync, -1);
+	wake_up(&perf->twait);
 
-	/* now read what peer wrote */
-	val = ntb_spad_read(ndev, VERSION);
-	if (val != PERF_VERSION) {
-		dev_dbg(&pdev->dev, "Remote version = %#x\n", val);
-		goto out;
+	for (tidx = 0; tidx < MAX_THREADS_CNT; tidx++) {
+		wake_up(&perf->threads[tidx].dma_wait);
+		cancel_work_sync(&perf->threads[tidx].work);
 	}
+}
+
+static int perf_submit_test(struct perf_peer *peer)
+{
+	struct perf_ctx *perf = peer->perf;
+	struct perf_thread *pthr;
+	int tidx, ret;
 
-	val = ntb_spad_read(ndev, MW_SZ_HIGH);
-	size = (u64)val << 32;
+	if (!test_bit(PERF_STS_DONE, &peer->sts))
+		return -ENOLINK;
 
-	val = ntb_spad_read(ndev, MW_SZ_LOW);
-	size |= val;
+	if (test_and_set_bit_lock(0, &perf->busy_flag))
+		return -EBUSY;
 
-	dev_dbg(&pdev->dev, "Remote MW size = %#llx\n", size);
+	perf->test_peer = peer;
+	atomic_set(&perf->tsync, perf->tcnt);
 
-	rc = perf_set_mw(perf, size);
-	if (rc)
-		goto out1;
+	for (tidx = 0; tidx < MAX_THREADS_CNT; tidx++) {
+		pthr = &perf->threads[tidx];
 
-	perf->link_is_up = true;
-	wake_up(&perf->link_wq);
+		pthr->status = -ENODATA;
+		pthr->copied = 0;
+		pthr->duration = ktime_set(0, 0);
+		if (tidx < perf->tcnt)
+			(void)queue_work(perf_wq, &pthr->work);
+	}
 
-	return;
+	ret = wait_event_interruptible(perf->twait,
+				       atomic_read(&perf->tsync) <= 0);
+	if (ret == -ERESTARTSYS) {
+		perf_terminate_test(perf);
+		ret = -EINTR;
+	}
 
-out1:
-	perf_free_mw(perf);
+	clear_bit_unlock(0, &perf->busy_flag);
 
-out:
-	if (ntb_link_is_up(ndev, NULL, NULL) == 1)
-		schedule_delayed_work(&perf->link_work,
-				      msecs_to_jiffies(PERF_LINK_DOWN_TIMEOUT));
+	return ret;
 }
 
-static int perf_setup_mw(struct ntb_dev *ntb, struct perf_ctx *perf)
+static int perf_read_stats(struct perf_ctx *perf, char *buf,
+			   size_t size, ssize_t *pos)
 {
-	struct perf_mw *mw;
-	int rc;
+	struct perf_thread *pthr;
+	int tidx;
+
+	if (test_and_set_bit_lock(0, &perf->busy_flag))
+		return -EBUSY;
 
-	mw = &perf->mw;
+	(*pos) += scnprintf(buf + *pos, size - *pos,
+		"    Peer %d test statistics:\n", perf->test_peer->pidx);
 
-	rc = ntb_peer_mw_get_addr(ntb, 0, &mw->phys_addr, &mw->phys_size);
-	if (rc)
-		return rc;
+	for (tidx = 0; tidx < MAX_THREADS_CNT; tidx++) {
+		pthr = &perf->threads[tidx];
 
-	perf->mw.vbase = ioremap_wc(mw->phys_addr, mw->phys_size);
-	if (!mw->vbase)
-		return -ENOMEM;
+		if (pthr->status == -ENODATA)
+			continue;
+
+		if (pthr->status) {
+			(*pos) += scnprintf(buf + *pos, size - *pos,
+				"%d: error status %d\n", tidx, pthr->status);
+			continue;
+		}
+
+		(*pos) += scnprintf(buf + *pos, size - *pos,
+			"%d: copied %llu bytes in %llu usecs, %llu MBytes/s\n",
+			tidx, pthr->copied, ktime_to_us(pthr->duration),
+			div64_u64(pthr->copied, ktime_to_us(pthr->duration)));
+	}
+
+	clear_bit_unlock(0, &perf->busy_flag);
 
 	return 0;
 }
 
-static ssize_t debugfs_run_read(struct file *filp, char __user *ubuf,
-				size_t count, loff_t *offp)
+static void perf_init_threads(struct perf_ctx *perf)
 {
-	struct perf_ctx *perf = filp->private_data;
+	struct perf_thread *pthr;
+	int tidx;
+
+	perf->tcnt = DEF_THREADS_CNT;
+	perf->test_peer = &perf->peers[0];
+	init_waitqueue_head(&perf->twait);
+
+	for (tidx = 0; tidx < MAX_THREADS_CNT; tidx++) {
+		pthr = &perf->threads[tidx];
+
+		pthr->perf = perf;
+		pthr->tidx = tidx;
+		pthr->status = -ENODATA;
+		init_waitqueue_head(&pthr->dma_wait);
+		INIT_WORK(&pthr->work, perf_thread_work);
+	}
+}
+
+static void perf_clear_threads(struct perf_ctx *perf)
+{
+	perf_terminate_test(perf);
+}
+
+/*==============================================================================
+ *                               DebugFS nodes
+ *==============================================================================
+ */
+
+static ssize_t perf_dbgfs_read_info(struct file *filep, char __user *ubuf,
+				    size_t size, loff_t *offp)
+{
+	struct perf_ctx *perf = filep->private_data;
+	struct perf_peer *peer;
+	size_t buf_size;
+	ssize_t pos = 0;
+	int ret, pidx;
 	char *buf;
-	ssize_t ret, out_off = 0;
-	struct pthr_ctx *pctx;
-	int i;
-	u64 rate;
 
-	if (!perf)
-		return 0;
+	buf_size = min_t(size_t, size, 0x1000U);
 
-	buf = kmalloc(1024, GFP_KERNEL);
+	buf = kmalloc(buf_size, GFP_KERNEL);
 	if (!buf)
 		return -ENOMEM;
 
-	if (mutex_is_locked(&perf->run_mutex)) {
-		out_off = scnprintf(buf, 64, "running\n");
-		goto read_from_buf;
+	pos += scnprintf(buf + pos, buf_size - pos,
+		"    Performance measuring tool info:\n\n");
+
+	pos += scnprintf(buf + pos, buf_size - pos,
+		"Local port %d, Global index %d\n", ntb_port_number(perf->ntb),
+		perf->gidx);
+	pos += scnprintf(buf + pos, buf_size - pos, "Test status: ");
+	if (test_bit(0, &perf->busy_flag)) {
+		pos += scnprintf(buf + pos, buf_size - pos,
+			"on-fly with port %d (%d)\n",
+			ntb_peer_port_number(perf->ntb, perf->test_peer->pidx),
+			perf->test_peer->pidx);
+	} else {
+		pos += scnprintf(buf + pos, buf_size - pos, "idle\n");
 	}
 
-	for (i = 0; i < MAX_THREADS; i++) {
-		pctx = &perf->pthr_ctx[i];
+	for (pidx = 0; pidx < perf->pcnt; pidx++) {
+		peer = &perf->peers[pidx];
+
+		pos += scnprintf(buf + pos, buf_size - pos,
+			"Port %d (%d), Global index %d:\n",
+			ntb_peer_port_number(perf->ntb, peer->pidx), peer->pidx,
+			peer->gidx);
+
+		pos += scnprintf(buf + pos, buf_size - pos,
+			"\tLink status: %s\n",
+			test_bit(PERF_STS_LNKUP, &peer->sts) ? "up" : "down");
+
+		pos += scnprintf(buf + pos, buf_size - pos,
+			"\tOut buffer addr 0x%pK\n", peer->outbuf);
 
-		if (pctx->status == -ENODATA)
-			break;
+		pos += scnprintf(buf + pos, buf_size - pos,
+			"\tOut buffer size %pa\n", &peer->outbuf_size);
 
-		if (pctx->status) {
-			out_off += scnprintf(buf + out_off, 1024 - out_off,
-					    "%d: error %d\n", i,
-					    pctx->status);
+		pos += scnprintf(buf + pos, buf_size - pos,
+			"\tOut buffer xlat 0x%016llx[p]\n", peer->outbuf_xlat);
+
+		if (!peer->inbuf) {
+			pos += scnprintf(buf + pos, buf_size - pos,
+				"\tIn buffer addr: unallocated\n");
 			continue;
 		}
 
-		rate = div64_u64(pctx->copied, pctx->diff_us);
-		out_off += scnprintf(buf + out_off, 1024 - out_off,
-			"%d: copied %llu bytes in %llu usecs, %llu MBytes/s\n",
-			i, pctx->copied, pctx->diff_us, rate);
+		pos += scnprintf(buf + pos, buf_size - pos,
+			"\tIn buffer addr 0x%pK\n", peer->inbuf);
+
+		pos += scnprintf(buf + pos, buf_size - pos,
+			"\tIn buffer size %pa\n", &peer->inbuf_size);
+
+		pos += scnprintf(buf + pos, buf_size - pos,
+			"\tIn buffer xlat %pad[p]\n", &peer->inbuf_xlat);
 	}
 
-read_from_buf:
-	ret = simple_read_from_buffer(ubuf, count, offp, buf, out_off);
+	ret = simple_read_from_buffer(ubuf, size, offp, buf, pos);
 	kfree(buf);
 
 	return ret;
 }
 
-static void threads_cleanup(struct perf_ctx *perf)
+static const struct file_operations perf_dbgfs_info = {
+	.open = simple_open,
+	.read = perf_dbgfs_read_info
+};
+
+static ssize_t perf_dbgfs_read_run(struct file *filep, char __user *ubuf,
+				   size_t size, loff_t *offp)
 {
-	struct pthr_ctx *pctx;
-	int i;
+	struct perf_ctx *perf = filep->private_data;
+	ssize_t ret, pos = 0;
+	char *buf;
 
-	for (i = 0; i < MAX_THREADS; i++) {
-		pctx = &perf->pthr_ctx[i];
-		if (pctx->thread) {
-			pctx->status = kthread_stop(pctx->thread);
-			pctx->thread = NULL;
-		}
-	}
-}
+	buf = kmalloc(PERF_BUF_LEN, GFP_KERNEL);
+	if (!buf)
+		return -ENOMEM;
 
-static void perf_clear_thread_status(struct perf_ctx *perf)
-{
-	int i;
+	ret = perf_read_stats(perf, buf, PERF_BUF_LEN, &pos);
+	if (ret)
+		goto err_free;
+
+	ret = simple_read_from_buffer(ubuf, size, offp, buf, pos);
+err_free:
+	kfree(buf);
 
-	for (i = 0; i < MAX_THREADS; i++)
-		perf->pthr_ctx[i].status = -ENODATA;
+	return ret;
 }
 
-static ssize_t debugfs_run_write(struct file *filp, const char __user *ubuf,
-				 size_t count, loff_t *offp)
+static ssize_t perf_dbgfs_write_run(struct file *filep, const char __user *ubuf,
+				    size_t size, loff_t *offp)
 {
-	struct perf_ctx *perf = filp->private_data;
-	int node, i;
-	DECLARE_WAIT_QUEUE_HEAD(wq);
+	struct perf_ctx *perf = filep->private_data;
+	struct perf_peer *peer;
+	int pidx, ret;
 
-	if (wait_event_interruptible(perf->link_wq, perf->link_is_up))
-		return -ENOLINK;
+	ret = kstrtoint_from_user(ubuf, size, 0, &pidx);
+	if (ret)
+		return ret;
 
-	if (perf->perf_threads == 0)
+	if (pidx < 0 || pidx >= perf->pcnt)
 		return -EINVAL;
 
-	if (!mutex_trylock(&perf->run_mutex))
-		return -EBUSY;
+	peer = &perf->peers[pidx];
 
-	perf_clear_thread_status(perf);
+	ret = perf_submit_test(peer);
+	if (ret)
+		return ret;
 
-	if (perf->perf_threads > MAX_THREADS) {
-		perf->perf_threads = MAX_THREADS;
-		pr_info("Reset total threads to: %u\n", MAX_THREADS);
-	}
+	return size;
+}
 
-	/* no greater than 1M */
-	if (seg_order > MAX_SEG_ORDER) {
-		seg_order = MAX_SEG_ORDER;
-		pr_info("Fix seg_order to %u\n", seg_order);
-	}
+static const struct file_operations perf_dbgfs_run = {
+	.open = simple_open,
+	.read = perf_dbgfs_read_run,
+	.write = perf_dbgfs_write_run
+};
 
-	if (run_order < seg_order) {
-		run_order = seg_order;
-		pr_info("Fix run_order to %u\n", run_order);
-	}
+static ssize_t perf_dbgfs_read_tcnt(struct file *filep, char __user *ubuf,
+				    size_t size, loff_t *offp)
+{
+	struct perf_ctx *perf = filep->private_data;
+	char buf[8];
+	ssize_t pos;
 
-	node = on_node ? dev_to_node(&perf->ntb->pdev->dev)
-		       : NUMA_NO_NODE;
-	atomic_set(&perf->tdone, 0);
+	pos = scnprintf(buf, sizeof(buf), "%hhu\n", perf->tcnt);
 
-	/* launch kernel thread */
-	for (i = 0; i < perf->perf_threads; i++) {
-		struct pthr_ctx *pctx;
+	return simple_read_from_buffer(ubuf, size, offp, buf, pos);
+}
 
-		pctx = &perf->pthr_ctx[i];
-		atomic_set(&pctx->dma_sync, 0);
-		pctx->perf = perf;
-		pctx->wq = &wq;
-		pctx->thread =
-			kthread_create_on_node(ntb_perf_thread,
-					       (void *)pctx,
-					       node, "ntb_perf %d", i);
-		if (IS_ERR(pctx->thread)) {
-			pctx->thread = NULL;
-			goto err;
-		} else {
-			wake_up_process(pctx->thread);
-		}
-	}
+static ssize_t perf_dbgfs_write_tcnt(struct file *filep,
+				     const char __user *ubuf,
+				     size_t size, loff_t *offp)
+{
+	struct perf_ctx *perf = filep->private_data;
+	int ret;
+	u8 val;
 
-	wait_event_interruptible(wq,
-		atomic_read(&perf->tdone) == perf->perf_threads);
+	ret = kstrtou8_from_user(ubuf, size, 0, &val);
+	if (ret)
+		return ret;
 
-	threads_cleanup(perf);
-	mutex_unlock(&perf->run_mutex);
-	return count;
+	ret = perf_set_tcnt(perf, val);
+	if (ret)
+		return ret;
 
-err:
-	threads_cleanup(perf);
-	mutex_unlock(&perf->run_mutex);
-	return -ENXIO;
+	return size;
 }
 
-static const struct file_operations ntb_perf_debugfs_run = {
-	.owner = THIS_MODULE,
+static const struct file_operations perf_dbgfs_tcnt = {
 	.open = simple_open,
-	.read = debugfs_run_read,
-	.write = debugfs_run_write,
+	.read = perf_dbgfs_read_tcnt,
+	.write = perf_dbgfs_write_tcnt
 };
 
-static int perf_debugfs_setup(struct perf_ctx *perf)
+static void perf_setup_dbgfs(struct perf_ctx *perf)
 {
 	struct pci_dev *pdev = perf->ntb->pdev;
-	struct dentry *debugfs_node_dir;
-	struct dentry *debugfs_run;
-	struct dentry *debugfs_threads;
-	struct dentry *debugfs_seg_order;
-	struct dentry *debugfs_run_order;
-	struct dentry *debugfs_use_dma;
-	struct dentry *debugfs_on_node;
-
-	if (!debugfs_initialized())
-		return -ENODEV;
 
-	/* Assumpion: only one NTB device in the system */
-	if (!perf_debugfs_dir) {
-		perf_debugfs_dir = debugfs_create_dir(KBUILD_MODNAME, NULL);
-		if (!perf_debugfs_dir)
-			return -ENODEV;
-	}
-
-	debugfs_node_dir = debugfs_create_dir(pci_name(pdev),
-					      perf_debugfs_dir);
-	if (!debugfs_node_dir)
-		goto err;
-
-	debugfs_run = debugfs_create_file("run", S_IRUSR | S_IWUSR,
-					  debugfs_node_dir, perf,
-					  &ntb_perf_debugfs_run);
-	if (!debugfs_run)
-		goto err;
-
-	debugfs_threads = debugfs_create_u8("threads", S_IRUSR | S_IWUSR,
-					    debugfs_node_dir,
-					    &perf->perf_threads);
-	if (!debugfs_threads)
-		goto err;
-
-	debugfs_seg_order = debugfs_create_u32("seg_order", 0600,
-					       debugfs_node_dir,
-					       &seg_order);
-	if (!debugfs_seg_order)
-		goto err;
-
-	debugfs_run_order = debugfs_create_u32("run_order", 0600,
-					       debugfs_node_dir,
-					       &run_order);
-	if (!debugfs_run_order)
-		goto err;
-
-	debugfs_use_dma = debugfs_create_bool("use_dma", 0600,
-					       debugfs_node_dir,
-					       &use_dma);
-	if (!debugfs_use_dma)
-		goto err;
-
-	debugfs_on_node = debugfs_create_bool("on_node", 0600,
-					      debugfs_node_dir,
-					      &on_node);
-	if (!debugfs_on_node)
-		goto err;
+	perf->dbgfs_dir = debugfs_create_dir(pci_name(pdev), perf_dbgfs_topdir);
+	if (!perf->dbgfs_dir) {
+		dev_warn(&perf->ntb->dev, "DebugFS unsupported\n");
+		return;
+	}
+
+	debugfs_create_file("info", 0600, perf->dbgfs_dir, perf,
+			    &perf_dbgfs_info);
 
-	return 0;
+	debugfs_create_file("run", 0600, perf->dbgfs_dir, perf,
+			    &perf_dbgfs_run);
 
-err:
-	debugfs_remove_recursive(perf_debugfs_dir);
-	perf_debugfs_dir = NULL;
-	return -ENODEV;
+	debugfs_create_file("threads_count", 0600, perf->dbgfs_dir, perf,
+			    &perf_dbgfs_tcnt);
+
+	/* They are made read-only for test exec safety and integrity */
+	debugfs_create_u8("chunk_order", 0500, perf->dbgfs_dir, &chunk_order);
+
+	debugfs_create_u8("total_order", 0500, perf->dbgfs_dir, &total_order);
+
+	debugfs_create_bool("use_dma", 0500, perf->dbgfs_dir, &use_dma);
 }
 
-static int perf_probe(struct ntb_client *client, struct ntb_dev *ntb)
+static void perf_clear_dbgfs(struct perf_ctx *perf)
+{
+	debugfs_remove_recursive(perf->dbgfs_dir);
+}
+
+/*==============================================================================
+ *                        Basic driver initialization
+ *==============================================================================
+ */
+
+static struct perf_ctx *perf_create_data(struct ntb_dev *ntb)
 {
-	struct pci_dev *pdev = ntb->pdev;
 	struct perf_ctx *perf;
-	int node;
-	int rc = 0;
 
-	if (ntb_spad_count(ntb) < MAX_SPAD) {
-		dev_err(&ntb->dev, "Not enough scratch pad registers for %s",
-			DRIVER_NAME);
-		return -EIO;
-	}
+	perf = devm_kzalloc(&ntb->dev, sizeof(*perf), GFP_KERNEL);
+	if (!perf)
+		return ERR_PTR(-ENOMEM);
 
-	if (!ntb->ops->mw_set_trans) {
-		dev_err(&ntb->dev, "Need inbound MW based NTB API\n");
-		return -EINVAL;
+	perf->pcnt = ntb_peer_port_count(ntb);
+	perf->peers = devm_kcalloc(&ntb->dev, perf->pcnt, sizeof(*perf->peers),
+				  GFP_KERNEL);
+	if (!perf->peers)
+		return ERR_PTR(-ENOMEM);
+
+	perf->ntb = ntb;
+
+	return perf;
+}
+
+static int perf_setup_peer_mw(struct perf_peer *peer)
+{
+	struct perf_ctx *perf = peer->perf;
+	phys_addr_t phys_addr;
+	int ret;
+
+	/* Get outbound MW parameters and map it */
+	ret = ntb_peer_mw_get_addr(perf->ntb, peer->gidx, &phys_addr,
+				   &peer->outbuf_size);
+	if (ret)
+		return ret;
+
+	peer->outbuf = devm_ioremap_wc(&perf->ntb->dev, phys_addr,
+					peer->outbuf_size);
+	if (!peer->outbuf)
+		return -ENOMEM;
+
+	if (max_mw_size && peer->outbuf_size > max_mw_size) {
+		peer->outbuf_size = max_mw_size;
+		dev_warn(&peer->perf->ntb->dev,
+			"Peer %d outbuf reduced to %pa\n", peer->pidx,
+			&peer->outbuf_size);
 	}
 
-	if (ntb_peer_port_count(ntb) != NTB_DEF_PEER_CNT)
-		dev_warn(&ntb->dev, "Multi-port NTB devices unsupported\n");
+	return 0;
+}
 
-	node = on_node ? dev_to_node(&pdev->dev) : NUMA_NO_NODE;
-	perf = kzalloc_node(sizeof(*perf), GFP_KERNEL, node);
-	if (!perf) {
-		rc = -ENOMEM;
-		goto err_perf;
+static int perf_init_peers(struct perf_ctx *perf)
+{
+	struct perf_peer *peer;
+	int pidx, lport, ret;
+
+	lport = ntb_port_number(perf->ntb);
+	perf->gidx = -1;
+	for (pidx = 0; pidx < perf->pcnt; pidx++) {
+		peer = &perf->peers[pidx];
+
+		peer->perf = perf;
+		peer->pidx = pidx;
+		if (lport < ntb_peer_port_number(perf->ntb, pidx)) {
+			if (perf->gidx == -1)
+				perf->gidx = pidx;
+			peer->gidx = pidx + 1;
+		} else {
+			peer->gidx = pidx;
+		}
+		INIT_WORK(&peer->service, perf_service_work);
 	}
+	if (perf->gidx == -1)
+		perf->gidx = pidx;
 
-	perf->ntb = ntb;
-	perf->perf_threads = 1;
-	atomic_set(&perf->tsync, 0);
-	mutex_init(&perf->run_mutex);
-	spin_lock_init(&perf->db_lock);
-	perf_setup_mw(ntb, perf);
-	init_waitqueue_head(&perf->link_wq);
-	INIT_DELAYED_WORK(&perf->link_work, perf_link_work);
+	for (pidx = 0; pidx < perf->pcnt; pidx++) {
+		ret = perf_setup_peer_mw(&perf->peers[pidx]);
+		if (ret)
+			return ret;
+	}
+
+	dev_dbg(&perf->ntb->dev, "Global port index %d\n", perf->gidx);
+
+	return 0;
+}
 
-	rc = ntb_set_ctx(ntb, perf, &perf_ops);
-	if (rc)
-		goto err_ctx;
+static int perf_probe(struct ntb_client *client, struct ntb_dev *ntb)
+{
+	struct perf_ctx *perf;
+	int ret;
 
-	perf->link_is_up = false;
-	ntb_link_enable(ntb, NTB_SPEED_AUTO, NTB_WIDTH_AUTO);
-	ntb_link_event(ntb);
+	perf = perf_create_data(ntb);
+	if (IS_ERR(perf))
+		return PTR_ERR(perf);
 
-	rc = perf_debugfs_setup(perf);
-	if (rc)
-		goto err_ctx;
+	ret = perf_init_peers(perf);
+	if (ret)
+		return ret;
 
-	perf_clear_thread_status(perf);
+	perf_init_threads(perf);
 
-	return 0;
+	ret = perf_init_service(perf);
+	if (ret)
+		return ret;
 
-err_ctx:
-	cancel_delayed_work_sync(&perf->link_work);
-	kfree(perf);
-err_perf:
-	return rc;
+	ret = perf_enable_service(perf);
+	if (ret)
+		return ret;
+
+	perf_setup_dbgfs(perf);
+
+	return 0;
 }
 
 static void perf_remove(struct ntb_client *client, struct ntb_dev *ntb)
 {
 	struct perf_ctx *perf = ntb->ctx;
-	int i;
 
-	dev_dbg(&perf->ntb->dev, "%s called\n", __func__);
+	perf_clear_dbgfs(perf);
 
-	mutex_lock(&perf->run_mutex);
+	perf_disable_service(perf);
 
-	cancel_delayed_work_sync(&perf->link_work);
+	perf_clear_threads(perf);
+}
 
-	ntb_clear_ctx(ntb);
-	ntb_link_disable(ntb);
+static struct ntb_client perf_client = {
+	.ops = {
+		.probe = perf_probe,
+		.remove = perf_remove
+	}
+};
 
-	debugfs_remove_recursive(perf_debugfs_dir);
-	perf_debugfs_dir = NULL;
+static int __init perf_init(void)
+{
+	int ret;
 
-	if (use_dma) {
-		for (i = 0; i < MAX_THREADS; i++) {
-			struct pthr_ctx *pctx = &perf->pthr_ctx[i];
+	if (chunk_order > MAX_CHUNK_ORDER) {
+		chunk_order = MAX_CHUNK_ORDER;
+		pr_info("Chunk order reduced to %hhu\n", chunk_order);
+	}
 
-			if (pctx->dma_chan)
-				dma_release_channel(pctx->dma_chan);
-		}
+	if (total_order < chunk_order) {
+		total_order = chunk_order;
+		pr_info("Total data order reduced to %hhu\n", total_order);
 	}
 
-	kfree(perf);
+	perf_wq = alloc_workqueue("perf_wq", WQ_UNBOUND | WQ_SYSFS, 0);
+	if (!perf_wq)
+		return -ENOMEM;
+
+	if (debugfs_initialized())
+		perf_dbgfs_topdir = debugfs_create_dir(KBUILD_MODNAME, NULL);
+
+	ret = ntb_register_client(&perf_client);
+	if (ret) {
+		debugfs_remove_recursive(perf_dbgfs_topdir);
+		destroy_workqueue(perf_wq);
+	}
+
+	return ret;
 }
+module_init(perf_init);
+
+static void __exit perf_exit(void)
+{
+	ntb_unregister_client(&perf_client);
+	debugfs_remove_recursive(perf_dbgfs_topdir);
+	destroy_workqueue(perf_wq);
+}
+module_exit(perf_exit);
 
-static struct ntb_client perf_client = {
-	.ops = {
-		.probe = perf_probe,
-		.remove = perf_remove,
-	},
-};
-module_ntb_client(perf_client);

drivers/ntb/test/ntb_pingpong.c

@@ -1,10 +1,11 @@
 /*
- * This file is provided under a dual BSD/GPLv2 license.  When using or
+ *   This file is provided under a dual BSD/GPLv2 license.  When using or
  *   redistributing this file, you may do so under either license.
  *
  *   GPL LICENSE SUMMARY
  *
  *   Copyright (C) 2015 EMC Corporation. All Rights Reserved.
+ *   Copyright (C) 2017 T-Platforms. All Rights Reserved.
  *
  *   This program is free software; you can redistribute it and/or modify
  *   it under the terms of version 2 of the GNU General Public License as
@@ -18,6 +19,7 @@
  *   BSD LICENSE
  *
  *   Copyright (C) 2015 EMC Corporation. All Rights Reserved.
+ *   Copyright (C) 2017 T-Platforms. All Rights Reserved.
  *
  *   Redistribution and use in source and binary forms, with or without
  *   modification, are permitted provided that the following conditions
@@ -46,37 +48,45 @@
  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  *
  * PCIe NTB Pingpong Linux driver
- *
- * Contact Information:
- * Allen Hubbe <Allen.Hubbe@emc.com>
  */
 
-/* Note: load this module with option 'dyndbg=+p' */
+/*
+ * How to use this tool, by example.
+ *
+ * Assuming $DBG_DIR is something like:
+ * '/sys/kernel/debug/ntb_perf/0000:00:03.0'
+ * Suppose aside from local device there is at least one remote device
+ * connected to NTB with index 0.
+ *-----------------------------------------------------------------------------
+ * Eg: install driver with specified delay between doorbell event and response
+ *
+ * root@self# insmod ntb_pingpong.ko delay_ms=1000
+ *-----------------------------------------------------------------------------
+ * Eg: get number of ping-pong cycles performed
+ *
+ * root@self# cat $DBG_DIR/count
+ */
 
 #include <linux/init.h>
 #include <linux/kernel.h>
 #include <linux/module.h>
+#include <linux/device.h>
+#include <linux/bitops.h>
 
-#include <linux/dma-mapping.h>
 #include <linux/pci.h>
 #include <linux/slab.h>
-#include <linux/spinlock.h>
+#include <linux/hrtimer.h>
 #include <linux/debugfs.h>
 
 #include <linux/ntb.h>
 
-#define DRIVER_NAME			"ntb_pingpong"
-#define DRIVER_DESCRIPTION		"PCIe NTB Simple Pingpong Client"
-
-#define DRIVER_LICENSE			"Dual BSD/GPL"
-#define DRIVER_VERSION			"1.0"
-#define DRIVER_RELDATE			"24 March 2015"
-#define DRIVER_AUTHOR			"Allen Hubbe <Allen.Hubbe@emc.com>"
+#define DRIVER_NAME		"ntb_pingpong"
+#define DRIVER_VERSION		"2.0"
 
-MODULE_LICENSE(DRIVER_LICENSE);
+MODULE_LICENSE("Dual BSD/GPL");
 MODULE_VERSION(DRIVER_VERSION);
-MODULE_AUTHOR(DRIVER_AUTHOR);
-MODULE_DESCRIPTION(DRIVER_DESCRIPTION);
+MODULE_AUTHOR("Allen Hubbe <Allen.Hubbe@emc.com>");
+MODULE_DESCRIPTION("PCIe NTB Simple Pingpong Client");
 
 static unsigned int unsafe;
 module_param(unsafe, uint, 0644);
@@ -86,237 +96,343 @@ static unsigned int delay_ms = 1000;
 module_param(delay_ms, uint, 0644);
 MODULE_PARM_DESC(delay_ms, "Milliseconds to delay the response to peer");
 
-static unsigned long db_init = 0x7;
-module_param(db_init, ulong, 0644);
-MODULE_PARM_DESC(db_init, "Initial doorbell bits to ring on the peer");
-
-/* Only two-ports NTB devices are supported */
-#define PIDX		NTB_DEF_PEER_IDX
-
 struct pp_ctx {
-	struct ntb_dev			*ntb;
-	u64				db_bits;
-	/* synchronize access to db_bits by ping and pong */
-	spinlock_t			db_lock;
-	struct timer_list		db_timer;
-	unsigned long			db_delay;
-	struct dentry			*debugfs_node_dir;
-	struct dentry			*debugfs_count;
-	atomic_t			count;
+	struct ntb_dev *ntb;
+	struct hrtimer timer;
+	u64 in_db;
+	u64 out_db;
+	int out_pidx;
+	u64 nmask;
+	u64 pmask;
+	atomic_t count;
+	spinlock_t lock;
+	struct dentry *dbgfs_dir;
 };
+#define to_pp_timer(__timer) \
+	container_of(__timer, struct pp_ctx, timer)
 
-static struct dentry *pp_debugfs_dir;
+static struct dentry *pp_dbgfs_topdir;
 
-static void pp_ping(struct timer_list *t)
+static int pp_find_next_peer(struct pp_ctx *pp)
 {
-	struct pp_ctx *pp = from_timer(pp, t, db_timer);
-	unsigned long irqflags;
-	u64 db_bits, db_mask;
-	u32 spad_rd, spad_wr;
+	u64 link, out_db;
+	int pidx;
+
+	link = ntb_link_is_up(pp->ntb, NULL, NULL);
+
+	/* Find next available peer */
+	if (link & pp->nmask) {
+		pidx = __ffs64(link & pp->nmask);
+		out_db = BIT_ULL(pidx + 1);
+	} else if (link & pp->pmask) {
+		pidx = __ffs64(link & pp->pmask);
+		out_db = BIT_ULL(pidx);
+	} else {
+		return -ENODEV;
+	}
 
-	spin_lock_irqsave(&pp->db_lock, irqflags);
-	{
-		db_mask = ntb_db_valid_mask(pp->ntb);
-		db_bits = ntb_db_read(pp->ntb);
+	spin_lock(&pp->lock);
+	pp->out_pidx = pidx;
+	pp->out_db = out_db;
+	spin_unlock(&pp->lock);
 
-		if (db_bits) {
-			dev_dbg(&pp->ntb->dev,
-				"Masked pongs %#llx\n",
-				db_bits);
-			ntb_db_clear(pp->ntb, db_bits);
-		}
+	return 0;
+}
 
-		db_bits = ((pp->db_bits | db_bits) << 1) & db_mask;
+static void pp_setup(struct pp_ctx *pp)
+{
+	int ret;
 
-		if (!db_bits)
-			db_bits = db_init;
+	ntb_db_set_mask(pp->ntb, pp->in_db);
 
-		spad_rd = ntb_spad_read(pp->ntb, 0);
-		spad_wr = spad_rd + 1;
+	hrtimer_cancel(&pp->timer);
 
-		dev_dbg(&pp->ntb->dev,
-			"Ping bits %#llx read %#x write %#x\n",
-			db_bits, spad_rd, spad_wr);
+	ret = pp_find_next_peer(pp);
+	if (ret == -ENODEV) {
+		dev_dbg(&pp->ntb->dev, "Got no peers, so cancel\n");
+		return;
+	}
 
-		ntb_peer_spad_write(pp->ntb, PIDX, 0, spad_wr);
-		ntb_peer_db_set(pp->ntb, db_bits);
-		ntb_db_clear_mask(pp->ntb, db_mask);
+	dev_dbg(&pp->ntb->dev, "Ping-pong started with port %d, db %#llx\n",
+		ntb_peer_port_number(pp->ntb, pp->out_pidx), pp->out_db);
 
-		pp->db_bits = 0;
-	}
-	spin_unlock_irqrestore(&pp->db_lock, irqflags);
+	hrtimer_start(&pp->timer, ms_to_ktime(delay_ms), HRTIMER_MODE_REL);
 }
 
-static void pp_link_event(void *ctx)
+static void pp_clear(struct pp_ctx *pp)
 {
-	struct pp_ctx *pp = ctx;
+	hrtimer_cancel(&pp->timer);
 
-	if (ntb_link_is_up(pp->ntb, NULL, NULL) == 1) {
-		dev_dbg(&pp->ntb->dev, "link is up\n");
-		pp_ping(&pp->db_timer);
-	} else {
-		dev_dbg(&pp->ntb->dev, "link is down\n");
-		del_timer(&pp->db_timer);
-	}
+	ntb_db_set_mask(pp->ntb, pp->in_db);
+
+	dev_dbg(&pp->ntb->dev, "Ping-pong cancelled\n");
 }
 
-static void pp_db_event(void *ctx, int vec)
+static void pp_ping(struct pp_ctx *pp)
 {
-	struct pp_ctx *pp = ctx;
-	u64 db_bits, db_mask;
-	unsigned long irqflags;
+	u32 count;
 
-	spin_lock_irqsave(&pp->db_lock, irqflags);
-	{
-		db_mask = ntb_db_vector_mask(pp->ntb, vec);
-		db_bits = db_mask & ntb_db_read(pp->ntb);
-		ntb_db_set_mask(pp->ntb, db_mask);
-		ntb_db_clear(pp->ntb, db_bits);
+	count = atomic_read(&pp->count);
 
-		pp->db_bits |= db_bits;
+	spin_lock(&pp->lock);
+	ntb_peer_spad_write(pp->ntb, pp->out_pidx, 0, count);
+	ntb_peer_msg_write(pp->ntb, pp->out_pidx, 0, count);
 
-		mod_timer(&pp->db_timer, jiffies + pp->db_delay);
+	dev_dbg(&pp->ntb->dev, "Ping port %d spad %#x, msg %#x\n",
+		ntb_peer_port_number(pp->ntb, pp->out_pidx), count, count);
 
-		dev_dbg(&pp->ntb->dev,
-			"Pong vec %d bits %#llx\n",
-			vec, db_bits);
-		atomic_inc(&pp->count);
-	}
-	spin_unlock_irqrestore(&pp->db_lock, irqflags);
+	ntb_peer_db_set(pp->ntb, pp->out_db);
+	ntb_db_clear_mask(pp->ntb, pp->in_db);
+	spin_unlock(&pp->lock);
 }
 
-static int pp_debugfs_setup(struct pp_ctx *pp)
+static void pp_pong(struct pp_ctx *pp)
 {
-	struct pci_dev *pdev = pp->ntb->pdev;
+	u32 msg_data = -1, spad_data = -1;
+	int pidx = 0;
 
-	if (!pp_debugfs_dir)
-		return -ENODEV;
+	/* Read pong data */
+	spad_data = ntb_spad_read(pp->ntb, 0);
+	msg_data = ntb_msg_read(pp->ntb, &pidx, 0);
+	ntb_msg_clear_sts(pp->ntb, -1);
 
-	pp->debugfs_node_dir = debugfs_create_dir(pci_name(pdev),
-						  pp_debugfs_dir);
-	if (!pp->debugfs_node_dir)
-		return -ENODEV;
+	/*
+	 * Scratchpad and message data may differ, since message register can't
+	 * be rewritten unless status is cleared. Additionally either of them
+	 * might be unsupported
+	 */
+	dev_dbg(&pp->ntb->dev, "Pong spad %#x, msg %#x (port %d)\n",
+		spad_data, msg_data, ntb_peer_port_number(pp->ntb, pidx));
 
-	pp->debugfs_count = debugfs_create_atomic_t("count", S_IRUSR | S_IWUSR,
-						    pp->debugfs_node_dir,
-						    &pp->count);
-	if (!pp->debugfs_count)
-		return -ENODEV;
+	atomic_inc(&pp->count);
 
-	return 0;
+	ntb_db_set_mask(pp->ntb, pp->in_db);
+	ntb_db_clear(pp->ntb, pp->in_db);
+
+	hrtimer_start(&pp->timer, ms_to_ktime(delay_ms), HRTIMER_MODE_REL);
+}
+
+static enum hrtimer_restart pp_timer_func(struct hrtimer *t)
+{
+	struct pp_ctx *pp = to_pp_timer(t);
+
+	pp_ping(pp);
+
+	return HRTIMER_NORESTART;
+}
+
+static void pp_link_event(void *ctx)
+{
+	struct pp_ctx *pp = ctx;
+
+	pp_setup(pp);
+}
+
+static void pp_db_event(void *ctx, int vec)
+{
+	struct pp_ctx *pp = ctx;
+
+	pp_pong(pp);
 }
 
 static const struct ntb_ctx_ops pp_ops = {
 	.link_event = pp_link_event,
-	.db_event = pp_db_event,
+	.db_event = pp_db_event
 };
 
-static int pp_probe(struct ntb_client *client,
-		    struct ntb_dev *ntb)
+static int pp_check_ntb(struct ntb_dev *ntb)
 {
-	struct pp_ctx *pp;
-	int rc;
+	u64 pmask;
 
 	if (ntb_db_is_unsafe(ntb)) {
-		dev_dbg(&ntb->dev, "doorbell is unsafe\n");
-		if (!unsafe) {
-			rc = -EINVAL;
-			goto err_pp;
-		}
-	}
-
-	if (ntb_spad_count(ntb) < 1) {
-		dev_dbg(&ntb->dev, "no enough scratchpads\n");
-		rc = -EINVAL;
-		goto err_pp;
+		dev_dbg(&ntb->dev, "Doorbell is unsafe\n");
+		if (!unsafe)
+			return -EINVAL;
 	}
 
 	if (ntb_spad_is_unsafe(ntb)) {
-		dev_dbg(&ntb->dev, "scratchpad is unsafe\n");
-		if (!unsafe) {
-			rc = -EINVAL;
-			goto err_pp;
-		}
+		dev_dbg(&ntb->dev, "Scratchpad is unsafe\n");
+		if (!unsafe)
+			return -EINVAL;
 	}
 
-	if (ntb_peer_port_count(ntb) != NTB_DEF_PEER_CNT)
-		dev_warn(&ntb->dev, "multi-port NTB is unsupported\n");
+	pmask = GENMASK_ULL(ntb_peer_port_count(ntb), 0);
+	if ((ntb_db_valid_mask(ntb) & pmask) != pmask) {
+		dev_err(&ntb->dev, "Unsupported DB configuration\n");
+		return -EINVAL;
+	}
 
-	pp = kmalloc(sizeof(*pp), GFP_KERNEL);
-	if (!pp) {
-		rc = -ENOMEM;
-		goto err_pp;
+	if (ntb_spad_count(ntb) < 1 && ntb_msg_count(ntb) < 1) {
+		dev_err(&ntb->dev, "Scratchpads and messages unsupported\n");
+		return -EINVAL;
+	} else if (ntb_spad_count(ntb) < 1) {
+		dev_dbg(&ntb->dev, "Scratchpads unsupported\n");
+	} else if (ntb_msg_count(ntb) < 1) {
+		dev_dbg(&ntb->dev, "Messages unsupported\n");
 	}
 
+	return 0;
+}
+
+static struct pp_ctx *pp_create_data(struct ntb_dev *ntb)
+{
+	struct pp_ctx *pp;
+
+	pp = devm_kzalloc(&ntb->dev, sizeof(*pp), GFP_KERNEL);
+	if (!pp)
+		return ERR_PTR(-ENOMEM);
+
 	pp->ntb = ntb;
-	pp->db_bits = 0;
 	atomic_set(&pp->count, 0);
-	spin_lock_init(&pp->db_lock);
-	timer_setup(&pp->db_timer, pp_ping, 0);
-	pp->db_delay = msecs_to_jiffies(delay_ms);
+	spin_lock_init(&pp->lock);
+	hrtimer_init(&pp->timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
+	pp->timer.function = pp_timer_func;
+
+	return pp;
+}
+
+static void pp_init_flds(struct pp_ctx *pp)
+{
+	int pidx, lport, pcnt;
+
+	/* Find global port index */
+	lport = ntb_port_number(pp->ntb);
+	pcnt = ntb_peer_port_count(pp->ntb);
+	for (pidx = 0; pidx < pcnt; pidx++) {
+		if (lport < ntb_peer_port_number(pp->ntb, pidx))
+			break;
+	}
 
-	rc = ntb_set_ctx(ntb, pp, &pp_ops);
-	if (rc)
-		goto err_ctx;
+	pp->in_db = BIT_ULL(pidx);
+	pp->pmask = GENMASK_ULL(pidx, 0) >> 1;
+	pp->nmask = GENMASK_ULL(pcnt - 1, pidx);
 
-	rc = pp_debugfs_setup(pp);
-	if (rc)
-		goto err_ctx;
+	dev_dbg(&pp->ntb->dev, "Inbound db %#llx, prev %#llx, next %#llx\n",
+		pp->in_db, pp->pmask, pp->nmask);
+}
+
+static int pp_mask_events(struct pp_ctx *pp)
+{
+	u64 db_mask, msg_mask;
+	int ret;
+
+	db_mask = ntb_db_valid_mask(pp->ntb);
+	ret = ntb_db_set_mask(pp->ntb, db_mask);
+	if (ret)
+		return ret;
 
-	ntb_link_enable(ntb, NTB_SPEED_AUTO, NTB_WIDTH_AUTO);
-	ntb_link_event(ntb);
+	/* Skip message events masking if unsupported */
+	if (ntb_msg_count(pp->ntb) < 1)
+		return 0;
+
+	msg_mask = ntb_msg_outbits(pp->ntb) | ntb_msg_inbits(pp->ntb);
+	return ntb_msg_set_mask(pp->ntb, msg_mask);
+}
+
+static int pp_setup_ctx(struct pp_ctx *pp)
+{
+	int ret;
+
+	ret = ntb_set_ctx(pp->ntb, pp, &pp_ops);
+	if (ret)
+		return ret;
+
+	ntb_link_enable(pp->ntb, NTB_SPEED_AUTO, NTB_WIDTH_AUTO);
+	/* Might be not necessary */
+	ntb_link_event(pp->ntb);
 
 	return 0;
+}
+
+static void pp_clear_ctx(struct pp_ctx *pp)
+{
+	ntb_link_disable(pp->ntb);
 
-err_ctx:
-	kfree(pp);
-err_pp:
-	return rc;
+	ntb_clear_ctx(pp->ntb);
 }
 
-static void pp_remove(struct ntb_client *client,
-		      struct ntb_dev *ntb)
+static void pp_setup_dbgfs(struct pp_ctx *pp)
+{
+	struct pci_dev *pdev = pp->ntb->pdev;
+	void *ret;
+
+	pp->dbgfs_dir = debugfs_create_dir(pci_name(pdev), pp_dbgfs_topdir);
+
+	ret = debugfs_create_atomic_t("count", 0600, pp->dbgfs_dir, &pp->count);
+	if (!ret)
+		dev_warn(&pp->ntb->dev, "DebugFS unsupported\n");
+}
+
+static void pp_clear_dbgfs(struct pp_ctx *pp)
+{
+	debugfs_remove_recursive(pp->dbgfs_dir);
+}
+
+static int pp_probe(struct ntb_client *client, struct ntb_dev *ntb)
+{
+	struct pp_ctx *pp;
+	int ret;
+
+	ret = pp_check_ntb(ntb);
+	if (ret)
+		return ret;
+
+	pp = pp_create_data(ntb);
+	if (IS_ERR(pp))
+		return PTR_ERR(pp);
+
+	pp_init_flds(pp);
+
+	ret = pp_mask_events(pp);
+	if (ret)
+		return ret;
+
+	ret = pp_setup_ctx(pp);
+	if (ret)
+		return ret;
+
+	pp_setup_dbgfs(pp);
+
+	return 0;
+}
+
+static void pp_remove(struct ntb_client *client, struct ntb_dev *ntb)
 {
 	struct pp_ctx *pp = ntb->ctx;
 
-	debugfs_remove_recursive(pp->debugfs_node_dir);
+	pp_clear_dbgfs(pp);
 
-	ntb_clear_ctx(ntb);
-	del_timer_sync(&pp->db_timer);
-	ntb_link_disable(ntb);
+	pp_clear_ctx(pp);
 
-	kfree(pp);
+	pp_clear(pp);
 }
 
 static struct ntb_client pp_client = {
 	.ops = {
 		.probe = pp_probe,
-		.remove = pp_remove,
-	},
+		.remove = pp_remove
+	}
 };
 
 static int __init pp_init(void)
 {
-	int rc;
+	int ret;
 
 	if (debugfs_initialized())
-		pp_debugfs_dir = debugfs_create_dir(KBUILD_MODNAME, NULL);
+		pp_dbgfs_topdir = debugfs_create_dir(KBUILD_MODNAME, NULL);
 
-	rc = ntb_register_client(&pp_client);
-	if (rc)
-		goto err_client;
+	ret = ntb_register_client(&pp_client);
+	if (ret)
+		debugfs_remove_recursive(pp_dbgfs_topdir);
 
-	return 0;
-
-err_client:
-	debugfs_remove_recursive(pp_debugfs_dir);
-	return rc;
+	return ret;
 }
 module_init(pp_init);
 
 static void __exit pp_exit(void)
 {
 	ntb_unregister_client(&pp_client);
-	debugfs_remove_recursive(pp_debugfs_dir);
+	debugfs_remove_recursive(pp_dbgfs_topdir);
 }
 module_exit(pp_exit);
+

drivers/ntb/test/ntb_tool.c

@@ -5,6 +5,7 @@
  *   GPL LICENSE SUMMARY
  *
  *   Copyright (C) 2015 EMC Corporation. All Rights Reserved.
+ *   Copyright (C) 2017 T-Platforms All Rights Reserved.
  *
  *   This program is free software; you can redistribute it and/or modify
  *   it under the terms of version 2 of the GNU General Public License as
@@ -18,6 +19,7 @@
  *   BSD LICENSE
  *
  *   Copyright (C) 2015 EMC Corporation. All Rights Reserved.
+ *   Copyright (C) 2017 T-Platforms All Rights Reserved.
  *
  *   Redistribution and use in source and binary forms, with or without
  *   modification, are permitted provided that the following conditions
@@ -46,9 +48,6 @@
  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  *
  * PCIe NTB Debugging Tool Linux driver
- *
- * Contact Information:
- * Allen Hubbe <Allen.Hubbe@emc.com>
  */
 
 /*
@@ -56,42 +55,125 @@
  *
  * Assuming $DBG_DIR is something like:
  * '/sys/kernel/debug/ntb_tool/0000:00:03.0'
+ * Suppose aside from local device there is at least one remote device
+ * connected to NTB with index 0.
+ *-----------------------------------------------------------------------------
+ * Eg: check local/peer device information.
+ *
+ * # Get local device port number
+ * root@self# cat $DBG_DIR/port
+ *
+ * # Check local device functionality
+ * root@self# ls $DBG_DIR
+ * db            msg1          msg_sts     peer4/        port
+ * db_event      msg2          peer0/      peer5/        spad0
+ * db_mask       msg3          peer1/      peer_db       spad1
+ * link          msg_event     peer2/      peer_db_mask  spad2
+ * msg0          msg_mask      peer3/      peer_spad     spad3
+ * # As one can see it supports:
+ * # 1) four inbound message registers
+ * # 2) four inbound scratchpads
+ * # 3) up to six peer devices
+ *
+ * # Check peer device port number
+ * root@self# cat $DBG_DIR/peer0/port
  *
- * Eg: check if clearing the doorbell mask generates an interrupt.
+ * # Check peer device(s) functionality to be used
+ * root@self# ls $DBG_DIR/peer0
+ * link             mw_trans0       mw_trans6        port
+ * link_event       mw_trans1       mw_trans7        spad0
+ * msg0             mw_trans2       peer_mw_trans0   spad1
+ * msg1             mw_trans3       peer_mw_trans1   spad2
+ * msg2             mw_trans4       peer_mw_trans2   spad3
+ * msg3             mw_trans5       peer_mw_trans3
+ * # As one can see we got:
+ * # 1) four outbound message registers
+ * # 2) four outbound scratchpads
+ * # 3) eight inbound memory windows
+ * # 4) four outbound memory windows
+ *-----------------------------------------------------------------------------
+ * Eg: NTB link tests
  *
- * # Check the link status
- * root@self# cat $DBG_DIR/link
+ * # Set local link up/down
+ * root@self# echo Y > $DBG_DIR/link
+ * root@self# echo N > $DBG_DIR/link
  *
- * # Block until the link is up
- * root@self# echo Y > $DBG_DIR/link_event
+ * # Check if link with peer device is up/down:
+ * root@self# cat $DBG_DIR/peer0/link
  *
- * # Set the doorbell mask
- * root@self# echo 's 1' > $DBG_DIR/mask
+ * # Block until the link is up/down
+ * root@self# echo Y > $DBG_DIR/peer0/link_event
+ * root@self# echo N > $DBG_DIR/peer0/link_event
+ *-----------------------------------------------------------------------------
+ * Eg: Doorbell registers tests (some functionality might be absent)
  *
- * # Ring the doorbell from the peer
+ * # Set/clear/get local doorbell
+ * root@self# echo 's 1' > $DBG_DIR/db
+ * root@self# echo 'c 1' > $DBG_DIR/db
+ * root@self# cat  $DBG_DIR/db
+ *
+ * # Set/clear/get local doorbell mask
+ * root@self# echo 's 1' > $DBG_DIR/db_mask
+ * root@self# echo 'c 1' > $DBG_DIR/db_mask
+ * root@self# cat $DBG_DIR/db_mask
+ *
+ * # Ring/clear/get peer doorbell
  * root@peer# echo 's 1' > $DBG_DIR/peer_db
+ * root@peer# echo 'c 1' > $DBG_DIR/peer_db
+ * root@peer# cat $DBG_DIR/peer_db
+ *
+ * # Set/clear/get peer doorbell mask
+ * root@self# echo 's 1' > $DBG_DIR/peer_db_mask
+ * root@self# echo 'c 1' > $DBG_DIR/peer_db_mask
+ * root@self# cat $DBG_DIR/peer_db_mask
+ *
+ * # Block until local doorbell is set with specified value
+ * root@self# echo 1 > $DBG_DIR/db_event
+ *-----------------------------------------------------------------------------
+ * Eg: Message registers tests (functionality might be absent)
  *
- * # Clear the doorbell mask
- * root@self# echo 'c 1' > $DBG_DIR/mask
+ * # Set/clear/get in/out message registers status
+ * root@self# echo 's 1' > $DBG_DIR/msg_sts
+ * root@self# echo 'c 1' > $DBG_DIR/msg_sts
+ * root@self# cat $DBG_DIR/msg_sts
  *
- * Observe debugging output in dmesg or your console.  You should see a
- * doorbell event triggered by clearing the mask.  If not, this may indicate an
- * issue with the hardware that needs to be worked around in the driver.
+ * # Set/clear in/out message registers mask
+ * root@self# echo 's 1' > $DBG_DIR/msg_mask
+ * root@self# echo 'c 1' > $DBG_DIR/msg_mask
  *
- * Eg: read and write scratchpad registers
+ * # Get inbound message register #0 value and source of port index
+ * root@self# cat  $DBG_DIR/msg0
  *
- * root@peer# echo '0 0x01010101 1 0x7f7f7f7f' > $DBG_DIR/peer_spad
+ * # Send some data to peer over outbound message register #0
+ * root@self# echo 0x01020304 > $DBG_DIR/peer0/msg0
+ *-----------------------------------------------------------------------------
+ * Eg: Scratchpad registers tests (functionality might be absent)
  *
- * root@self# cat $DBG_DIR/spad
+ * # Write/read to/from local scratchpad register #0
+ * root@peer# echo 0x01020304 > $DBG_DIR/spad0
+ * root@peer# cat $DBG_DIR/spad0
  *
- * Observe that spad 0 and 1 have the values set by the peer.
+ * # Write/read to/from peer scratchpad register #0
+ * root@peer# echo 0x01020304 > $DBG_DIR/peer0/spad0
+ * root@peer# cat $DBG_DIR/peer0/spad0
+ *-----------------------------------------------------------------------------
+ * Eg: Memory windows tests
  *
- * # Check the memory window translation info
- * cat $DBG_DIR/peer_trans0
+ * # Create inbound memory window buffer of specified size/get its base address
+ * root@peer# echo 16384 > $DBG_DIR/peer0/mw_trans0
+ * root@peer# cat $DBG_DIR/peer0/mw_trans0
  *
- * # Setup a 16k memory window buffer
- * echo 16384 > $DBG_DIR/peer_trans0
+ * # Write/read data to/from inbound memory window
+ * root@peer# echo Hello > $DBG_DIR/peer0/mw0
+ * root@peer# head -c 7 $DBG_DIR/peer0/mw0
  *
+ * # Map outbound memory window/check it settings (on peer device)
+ * root@peer# echo 0xADD0BA5E:16384 > $DBG_DIR/peer0/peer_mw_trans0
+ * root@peer# cat $DBG_DIR/peer0/peer_mw_trans0
+ *
+ * # Write/read data to/from outbound memory window (on peer device)
+ * root@peer# echo olleH > $DBG_DIR/peer0/peer_mw0
+ * root@peer# head -c 7 $DBG_DIR/peer0/peer_mw0
  */
 
 #include <linux/init.h>
@@ -106,49 +188,87 @@
 
 #include <linux/ntb.h>
 
-#define DRIVER_NAME			"ntb_tool"
-#define DRIVER_DESCRIPTION		"PCIe NTB Debugging Tool"
-
-#define DRIVER_LICENSE			"Dual BSD/GPL"
-#define DRIVER_VERSION			"1.0"
-#define DRIVER_RELDATE			"22 April 2015"
-#define DRIVER_AUTHOR			"Allen Hubbe <Allen.Hubbe@emc.com>"
+#define DRIVER_NAME		"ntb_tool"
+#define DRIVER_VERSION		"2.0"
 
-MODULE_LICENSE(DRIVER_LICENSE);
+MODULE_LICENSE("Dual BSD/GPL");
 MODULE_VERSION(DRIVER_VERSION);
-MODULE_AUTHOR(DRIVER_AUTHOR);
-MODULE_DESCRIPTION(DRIVER_DESCRIPTION);
-
-/* It is rare to have hadrware with greater than six MWs */
-#define MAX_MWS	6
-/* Only two-ports devices are supported */
-#define PIDX	NTB_DEF_PEER_IDX
-
-static struct dentry *tool_dbgfs;
+MODULE_AUTHOR("Allen Hubbe <Allen.Hubbe@emc.com>");
+MODULE_DESCRIPTION("PCIe NTB Debugging Tool");
 
+/*
+ * Inbound and outbound memory windows descriptor. Union members selection
+ * depends on the MW type the structure describes. mm_base/dma_base are the
+ * virtual and DMA address of an inbound MW. io_base/tr_base are the MMIO
+ * mapped virtual and xlat addresses of an outbound MW respectively.
+ */
 struct tool_mw {
-	int idx;
+	int widx;
+	int pidx;
 	struct tool_ctx *tc;
-	resource_size_t win_size;
+	union {
+		u8 *mm_base;
+		u8 __iomem *io_base;
+	};
+	union {
+		dma_addr_t dma_base;
+		u64 tr_base;
+	};
 	resource_size_t size;
-	u8 __iomem *local;
-	u8 *peer;
-	dma_addr_t peer_dma;
-	struct dentry *peer_dbg_file;
+	struct dentry *dbgfs_file;
+};
+
+/*
+ * Wrapper structure is used to distinguish the outbound MW peers reference
+ * within the corresponding DebugFS directory IO operation.
+ */
+struct tool_mw_wrap {
+	int pidx;
+	struct tool_mw *mw;
+};
+
+struct tool_msg {
+	int midx;
+	int pidx;
+	struct tool_ctx *tc;
+};
+
+struct tool_spad {
+	int sidx;
+	int pidx;
+	struct tool_ctx *tc;
+};
+
+struct tool_peer {
+	int pidx;
+	struct tool_ctx *tc;
+	int inmw_cnt;
+	struct tool_mw *inmws;
+	int outmw_cnt;
+	struct tool_mw_wrap *outmws;
+	int outmsg_cnt;
+	struct tool_msg *outmsgs;
+	int outspad_cnt;
+	struct tool_spad *outspads;
+	struct dentry *dbgfs_dir;
 };
 
 struct tool_ctx {
 	struct ntb_dev *ntb;
-	struct dentry *dbgfs;
 	wait_queue_head_t link_wq;
-	int mw_count;
-	struct tool_mw mws[MAX_MWS];
+	wait_queue_head_t db_wq;
+	wait_queue_head_t msg_wq;
+	int outmw_cnt;
+	struct tool_mw *outmws;
+	int peer_cnt;
+	struct tool_peer *peers;
+	int inmsg_cnt;
+	struct tool_msg *inmsgs;
+	int inspad_cnt;
+	struct tool_spad *inspads;
+	struct dentry *dbgfs_dir;
 };
 
-#define SPAD_FNAME_SIZE 0x10
-#define INT_PTR(x) ((void *)(unsigned long)x)
-#define PTR_INT(x) ((int)(unsigned long)x)
-
 #define TOOL_FOPS_RDWR(__name, __read, __write) \
 	const struct file_operations __name = {	\
 		.owner = THIS_MODULE,		\
@@ -157,6 +277,15 @@ struct tool_ctx {
 		.write = __write,		\
 	}
 
+#define TOOL_BUF_LEN 32
+
+static struct dentry *tool_dbgfs_topdir;
+
+/*==============================================================================
+ *                               NTB events handlers
+ *==============================================================================
+ */
+
 static void tool_link_event(void *ctx)
 {
 	struct tool_ctx *tc = ctx;
@@ -182,580 +311,578 @@ static void tool_db_event(void *ctx, int vec)
 
 	dev_dbg(&tc->ntb->dev, "doorbell vec %d mask %#llx bits %#llx\n",
 		vec, db_mask, db_bits);
+
+	wake_up(&tc->db_wq);
+}
+
+static void tool_msg_event(void *ctx)
+{
+	struct tool_ctx *tc = ctx;
+	u64 msg_sts;
+
+	msg_sts = ntb_msg_read_sts(tc->ntb);
+
+	dev_dbg(&tc->ntb->dev, "message bits %#llx\n", msg_sts);
+
+	wake_up(&tc->msg_wq);
 }
 
 static const struct ntb_ctx_ops tool_ops = {
 	.link_event = tool_link_event,
 	.db_event = tool_db_event,
+	.msg_event = tool_msg_event
 };
 
-static ssize_t tool_dbfn_read(struct tool_ctx *tc, char __user *ubuf,
-			      size_t size, loff_t *offp,
-			      u64 (*db_read_fn)(struct ntb_dev *))
+/*==============================================================================
+ *                        Common read/write methods
+ *==============================================================================
+ */
+
+static ssize_t tool_fn_read(struct tool_ctx *tc, char __user *ubuf,
+			    size_t size, loff_t *offp,
+			    u64 (*fn_read)(struct ntb_dev *))
 {
 	size_t buf_size;
-	char *buf;
-	ssize_t pos, rc;
+	char buf[TOOL_BUF_LEN];
+	ssize_t pos;
 
-	if (!db_read_fn)
+	if (!fn_read)
 		return -EINVAL;
 
-	buf_size = min_t(size_t, size, 0x20);
-
-	buf = kmalloc(buf_size, GFP_KERNEL);
-	if (!buf)
-		return -ENOMEM;
-
-	pos = scnprintf(buf, buf_size, "%#llx\n",
-			db_read_fn(tc->ntb));
+	buf_size = min(size, sizeof(buf));
 
-	rc = simple_read_from_buffer(ubuf, size, offp, buf, pos);
+	pos = scnprintf(buf, buf_size, "%#llx\n", fn_read(tc->ntb));
 
-	kfree(buf);
-
-	return rc;
+	return simple_read_from_buffer(ubuf, size, offp, buf, pos);
 }
 
-static ssize_t tool_dbfn_write(struct tool_ctx *tc,
-			       const char __user *ubuf,
-			       size_t size, loff_t *offp,
-			       int (*db_set_fn)(struct ntb_dev *, u64),
-			       int (*db_clear_fn)(struct ntb_dev *, u64))
+static ssize_t tool_fn_write(struct tool_ctx *tc,
+			     const char __user *ubuf,
+			     size_t size, loff_t *offp,
+			     int (*fn_set)(struct ntb_dev *, u64),
+			     int (*fn_clear)(struct ntb_dev *, u64))
 {
-	u64 db_bits;
 	char *buf, cmd;
-	ssize_t rc;
+	ssize_t ret;
+	u64 bits;
 	int n;
 
 	buf = kmalloc(size + 1, GFP_KERNEL);
 	if (!buf)
 		return -ENOMEM;
 
-	rc = simple_write_to_buffer(buf, size, offp, ubuf, size);
-	if (rc < 0) {
+	ret = simple_write_to_buffer(buf, size, offp, ubuf, size);
+	if (ret < 0) {
 		kfree(buf);
-		return rc;
+		return ret;
 	}
 
 	buf[size] = 0;
 
-	n = sscanf(buf, "%c %lli", &cmd, &db_bits);
+	n = sscanf(buf, "%c %lli", &cmd, &bits);
 
 	kfree(buf);
 
 	if (n != 2) {
-		rc = -EINVAL;
+		ret = -EINVAL;
 	} else if (cmd == 's') {
-		if (!db_set_fn)
-			rc = -EINVAL;
+		if (!fn_set)
+			ret = -EINVAL;
 		else
-			rc = db_set_fn(tc->ntb, db_bits);
+			ret = fn_set(tc->ntb, bits);
 	} else if (cmd == 'c') {
-		if (!db_clear_fn)
-			rc = -EINVAL;
+		if (!fn_clear)
+			ret = -EINVAL;
 		else
-			rc = db_clear_fn(tc->ntb, db_bits);
+			ret = fn_clear(tc->ntb, bits);
 	} else {
-		rc = -EINVAL;
+		ret = -EINVAL;
 	}
 
-	return rc ? : size;
+	return ret ? : size;
 }
 
-static ssize_t tool_spadfn_read(struct tool_ctx *tc, char __user *ubuf,
-				size_t size, loff_t *offp,
-				u32 (*spad_read_fn)(struct ntb_dev *, int))
-{
-	size_t buf_size;
-	char *buf;
-	ssize_t pos, rc;
-	int i, spad_count;
-
-	if (!spad_read_fn)
-		return -EINVAL;
-
-	spad_count = ntb_spad_count(tc->ntb);
+/*==============================================================================
+ *                            Port read/write methods
+ *==============================================================================
+ */
 
-	/*
-	 * We multiply the number of spads by 15 to get the buffer size
-	 * this is from 3 for the %d, 10 for the largest hex value
-	 * (0x00000000) and 2 for the tab and line feed.
-	 */
-	buf_size = min_t(size_t, size, spad_count * 15);
+static ssize_t tool_port_read(struct file *filep, char __user *ubuf,
+			      size_t size, loff_t *offp)
+{
+	struct tool_ctx *tc = filep->private_data;
+	char buf[TOOL_BUF_LEN];
+	int pos;
 
-	buf = kmalloc(buf_size, GFP_KERNEL);
-	if (!buf)
-		return -ENOMEM;
+	pos = scnprintf(buf, sizeof(buf), "%d\n", ntb_port_number(tc->ntb));
 
-	pos = 0;
+	return simple_read_from_buffer(ubuf, size, offp, buf, pos);
+}
 
-	for (i = 0; i < spad_count; ++i) {
-		pos += scnprintf(buf + pos, buf_size - pos, "%d\t%#x\n",
-				 i, spad_read_fn(tc->ntb, i));
-	}
+static TOOL_FOPS_RDWR(tool_port_fops,
+		      tool_port_read,
+		      NULL);
 
-	rc = simple_read_from_buffer(ubuf, size, offp, buf, pos);
+static ssize_t tool_peer_port_read(struct file *filep, char __user *ubuf,
+				   size_t size, loff_t *offp)
+{
+	struct tool_peer *peer = filep->private_data;
+	struct tool_ctx *tc = peer->tc;
+	char buf[TOOL_BUF_LEN];
+	int pos;
 
-	kfree(buf);
+	pos = scnprintf(buf, sizeof(buf), "%d\n",
+		ntb_peer_port_number(tc->ntb, peer->pidx));
 
-	return rc;
+	return simple_read_from_buffer(ubuf, size, offp, buf, pos);
 }
 
-static ssize_t tool_spadfn_write(struct tool_ctx *tc,
-				 const char __user *ubuf,
-				 size_t size, loff_t *offp,
-				 int (*spad_write_fn)(struct ntb_dev *,
-						      int, u32))
+static TOOL_FOPS_RDWR(tool_peer_port_fops,
+		      tool_peer_port_read,
+		      NULL);
+
+static int tool_init_peers(struct tool_ctx *tc)
 {
-	int spad_idx;
-	u32 spad_val;
-	char *buf, *buf_ptr;
-	int pos, n;
-	ssize_t rc;
-
-	if (!spad_write_fn) {
-		dev_dbg(&tc->ntb->dev, "no spad write fn\n");
-		return -EINVAL;
-	}
+	int pidx;
 
-	buf = kmalloc(size + 1, GFP_KERNEL);
-	if (!buf)
+	tc->peer_cnt = ntb_peer_port_count(tc->ntb);
+	tc->peers = devm_kcalloc(&tc->ntb->dev, tc->peer_cnt,
+				 sizeof(*tc->peers), GFP_KERNEL);
+	if (tc->peers == NULL)
 		return -ENOMEM;
 
-	rc = simple_write_to_buffer(buf, size, offp, ubuf, size);
-	if (rc < 0) {
-		kfree(buf);
-		return rc;
-	}
-
-	buf[size] = 0;
-	buf_ptr = buf;
-	n = sscanf(buf_ptr, "%d %i%n", &spad_idx, &spad_val, &pos);
-	while (n == 2) {
-		buf_ptr += pos;
-		rc = spad_write_fn(tc->ntb, spad_idx, spad_val);
-		if (rc)
-			break;
-
-		n = sscanf(buf_ptr, "%d %i%n", &spad_idx, &spad_val, &pos);
+	for (pidx = 0; pidx < tc->peer_cnt; pidx++) {
+		tc->peers[pidx].pidx = pidx;
+		tc->peers[pidx].tc = tc;
 	}
 
-	if (n < 0)
-		rc = n;
-
-	kfree(buf);
-
-	return rc ? : size;
+	return 0;
 }
 
-static ssize_t tool_db_read(struct file *filep, char __user *ubuf,
-			    size_t size, loff_t *offp)
-{
-	struct tool_ctx *tc = filep->private_data;
-
-	return tool_dbfn_read(tc, ubuf, size, offp,
-			      tc->ntb->ops->db_read);
-}
+/*==============================================================================
+ *                       Link state read/write methods
+ *==============================================================================
+ */
 
-static ssize_t tool_db_write(struct file *filep, const char __user *ubuf,
-			     size_t size, loff_t *offp)
+static ssize_t tool_link_write(struct file *filep, const char __user *ubuf,
+			       size_t size, loff_t *offp)
 {
 	struct tool_ctx *tc = filep->private_data;
+	bool val;
+	int ret;
 
-	return tool_dbfn_write(tc, ubuf, size, offp,
-			       tc->ntb->ops->db_set,
-			       tc->ntb->ops->db_clear);
-}
+	ret = kstrtobool_from_user(ubuf, size, &val);
+	if (ret)
+		return ret;
 
-static TOOL_FOPS_RDWR(tool_db_fops,
-		      tool_db_read,
-		      tool_db_write);
+	if (val)
+		ret = ntb_link_enable(tc->ntb, NTB_SPEED_AUTO, NTB_WIDTH_AUTO);
+	else
+		ret = ntb_link_disable(tc->ntb);
 
-static ssize_t tool_mask_read(struct file *filep, char __user *ubuf,
-			      size_t size, loff_t *offp)
-{
-	struct tool_ctx *tc = filep->private_data;
+	if (ret)
+		return ret;
 
-	return tool_dbfn_read(tc, ubuf, size, offp,
-			      tc->ntb->ops->db_read_mask);
+	return size;
 }
 
-static ssize_t tool_mask_write(struct file *filep, const char __user *ubuf,
-			       size_t size, loff_t *offp)
+static TOOL_FOPS_RDWR(tool_link_fops,
+		      NULL,
+		      tool_link_write);
+
+static ssize_t tool_peer_link_read(struct file *filep, char __user *ubuf,
+				   size_t size, loff_t *offp)
 {
-	struct tool_ctx *tc = filep->private_data;
+	struct tool_peer *peer = filep->private_data;
+	struct tool_ctx *tc = peer->tc;
+	char buf[3];
 
-	return tool_dbfn_write(tc, ubuf, size, offp,
-			       tc->ntb->ops->db_set_mask,
-			       tc->ntb->ops->db_clear_mask);
+	if (ntb_link_is_up(tc->ntb, NULL, NULL) & BIT(peer->pidx))
+		buf[0] = 'Y';
+	else
+		buf[0] = 'N';
+	buf[1] = '\n';
+	buf[2] = '\0';
+
+	return simple_read_from_buffer(ubuf, size, offp, buf, 3);
 }
 
-static TOOL_FOPS_RDWR(tool_mask_fops,
-		      tool_mask_read,
-		      tool_mask_write);
+static TOOL_FOPS_RDWR(tool_peer_link_fops,
+		      tool_peer_link_read,
+		      NULL);
 
-static ssize_t tool_peer_db_read(struct file *filep, char __user *ubuf,
-				 size_t size, loff_t *offp)
+static ssize_t tool_peer_link_event_write(struct file *filep,
+					  const char __user *ubuf,
+					  size_t size, loff_t *offp)
 {
-	struct tool_ctx *tc = filep->private_data;
+	struct tool_peer *peer = filep->private_data;
+	struct tool_ctx *tc = peer->tc;
+	u64 link_msk;
+	bool val;
+	int ret;
 
-	return tool_dbfn_read(tc, ubuf, size, offp,
-			      tc->ntb->ops->peer_db_read);
-}
+	ret = kstrtobool_from_user(ubuf, size, &val);
+	if (ret)
+		return ret;
 
-static ssize_t tool_peer_db_write(struct file *filep, const char __user *ubuf,
-				  size_t size, loff_t *offp)
-{
-	struct tool_ctx *tc = filep->private_data;
+	link_msk = BIT_ULL_MASK(peer->pidx);
+
+	if (wait_event_interruptible(tc->link_wq,
+		!!(ntb_link_is_up(tc->ntb, NULL, NULL) & link_msk) == val))
+		return -ERESTART;
 
-	return tool_dbfn_write(tc, ubuf, size, offp,
-			       tc->ntb->ops->peer_db_set,
-			       tc->ntb->ops->peer_db_clear);
+	return size;
 }
 
-static TOOL_FOPS_RDWR(tool_peer_db_fops,
-		      tool_peer_db_read,
-		      tool_peer_db_write);
+static TOOL_FOPS_RDWR(tool_peer_link_event_fops,
+		      NULL,
+		      tool_peer_link_event_write);
 
-static ssize_t tool_peer_mask_read(struct file *filep, char __user *ubuf,
-				   size_t size, loff_t *offp)
+/*==============================================================================
+ *                  Memory windows read/write/setting methods
+ *==============================================================================
+ */
+
+static ssize_t tool_mw_read(struct file *filep, char __user *ubuf,
+			    size_t size, loff_t *offp)
 {
-	struct tool_ctx *tc = filep->private_data;
+	struct tool_mw *inmw = filep->private_data;
+
+	if (inmw->mm_base == NULL)
+		return -ENXIO;
 
-	return tool_dbfn_read(tc, ubuf, size, offp,
-			      tc->ntb->ops->peer_db_read_mask);
+	return simple_read_from_buffer(ubuf, size, offp,
+				       inmw->mm_base, inmw->size);
 }
 
-static ssize_t tool_peer_mask_write(struct file *filep, const char __user *ubuf,
-				    size_t size, loff_t *offp)
+static ssize_t tool_mw_write(struct file *filep, const char __user *ubuf,
+			     size_t size, loff_t *offp)
 {
-	struct tool_ctx *tc = filep->private_data;
+	struct tool_mw *inmw = filep->private_data;
+
+	if (inmw->mm_base == NULL)
+		return -ENXIO;
 
-	return tool_dbfn_write(tc, ubuf, size, offp,
-			       tc->ntb->ops->peer_db_set_mask,
-			       tc->ntb->ops->peer_db_clear_mask);
+	return simple_write_to_buffer(inmw->mm_base, inmw->size, offp,
+				      ubuf, size);
 }
 
-static TOOL_FOPS_RDWR(tool_peer_mask_fops,
-		      tool_peer_mask_read,
-		      tool_peer_mask_write);
+static TOOL_FOPS_RDWR(tool_mw_fops,
+		      tool_mw_read,
+		      tool_mw_write);
 
-static ssize_t tool_spad_read(struct file *filep, char __user *ubuf,
-			      size_t size, loff_t *offp)
+static int tool_setup_mw(struct tool_ctx *tc, int pidx, int widx,
+			 size_t req_size)
 {
-	struct tool_ctx *tc = filep->private_data;
+	resource_size_t size, addr_align, size_align;
+	struct tool_mw *inmw = &tc->peers[pidx].inmws[widx];
+	char buf[TOOL_BUF_LEN];
+	int ret;
 
-	return tool_spadfn_read(tc, ubuf, size, offp,
-				tc->ntb->ops->spad_read);
-}
+	if (inmw->mm_base != NULL)
+		return 0;
 
-static ssize_t tool_spad_write(struct file *filep, const char __user *ubuf,
-			       size_t size, loff_t *offp)
-{
-	struct tool_ctx *tc = filep->private_data;
+	ret = ntb_mw_get_align(tc->ntb, pidx, widx, &addr_align,
+				&size_align, &size);
+	if (ret)
+		return ret;
+
+	inmw->size = min_t(resource_size_t, req_size, size);
+	inmw->size = round_up(inmw->size, addr_align);
+	inmw->size = round_up(inmw->size, size_align);
+	inmw->mm_base = dma_alloc_coherent(&tc->ntb->dev, inmw->size,
+					   &inmw->dma_base, GFP_KERNEL);
+	if (!inmw->mm_base)
+		return -ENOMEM;
 
-	return tool_spadfn_write(tc, ubuf, size, offp,
-				 tc->ntb->ops->spad_write);
-}
+	if (!IS_ALIGNED(inmw->dma_base, addr_align)) {
+		ret = -ENOMEM;
+		goto err_free_dma;
+	}
 
-static TOOL_FOPS_RDWR(tool_spad_fops,
-		      tool_spad_read,
-		      tool_spad_write);
+	ret = ntb_mw_set_trans(tc->ntb, pidx, widx, inmw->dma_base, inmw->size);
+	if (ret)
+		goto err_free_dma;
 
-static u32 ntb_tool_peer_spad_read(struct ntb_dev *ntb, int sidx)
-{
-	return ntb_peer_spad_read(ntb, PIDX, sidx);
-}
+	snprintf(buf, sizeof(buf), "mw%d", widx);
+	inmw->dbgfs_file = debugfs_create_file(buf, 0600,
+					       tc->peers[pidx].dbgfs_dir, inmw,
+					       &tool_mw_fops);
 
-static ssize_t tool_peer_spad_read(struct file *filep, char __user *ubuf,
-				   size_t size, loff_t *offp)
-{
-	struct tool_ctx *tc = filep->private_data;
+	return 0;
 
-	return tool_spadfn_read(tc, ubuf, size, offp, ntb_tool_peer_spad_read);
-}
+err_free_dma:
+	dma_free_coherent(&tc->ntb->dev, inmw->size, inmw->mm_base,
+			  inmw->dma_base);
+	inmw->mm_base = NULL;
+	inmw->dma_base = 0;
+	inmw->size = 0;
 
-static int ntb_tool_peer_spad_write(struct ntb_dev *ntb, int sidx, u32 val)
-{
-	return ntb_peer_spad_write(ntb, PIDX, sidx, val);
+	return ret;
 }
 
-static ssize_t tool_peer_spad_write(struct file *filep, const char __user *ubuf,
-				    size_t size, loff_t *offp)
+static void tool_free_mw(struct tool_ctx *tc, int pidx, int widx)
 {
-	struct tool_ctx *tc = filep->private_data;
-
-	return tool_spadfn_write(tc, ubuf, size, offp,
-				 ntb_tool_peer_spad_write);
-}
-
-static TOOL_FOPS_RDWR(tool_peer_spad_fops,
-		      tool_peer_spad_read,
-		      tool_peer_spad_write);
+	struct tool_mw *inmw = &tc->peers[pidx].inmws[widx];
 
-static ssize_t tool_link_read(struct file *filep, char __user *ubuf,
-			      size_t size, loff_t *offp)
-{
-	struct tool_ctx *tc = filep->private_data;
-	char buf[3];
+	debugfs_remove(inmw->dbgfs_file);
 
-	buf[0] = ntb_link_is_up(tc->ntb, NULL, NULL) ? 'Y' : 'N';
-	buf[1] = '\n';
-	buf[2] = '\0';
+	if (inmw->mm_base != NULL) {
+		ntb_mw_clear_trans(tc->ntb, pidx, widx);
+		dma_free_coherent(&tc->ntb->dev, inmw->size,
+				  inmw->mm_base, inmw->dma_base);
+	}
 
-	return simple_read_from_buffer(ubuf, size, offp, buf, 2);
+	inmw->mm_base = NULL;
+	inmw->dma_base = 0;
+	inmw->size = 0;
+	inmw->dbgfs_file = NULL;
 }
 
-static ssize_t tool_link_write(struct file *filep, const char __user *ubuf,
-			       size_t size, loff_t *offp)
+static ssize_t tool_mw_trans_read(struct file *filep, char __user *ubuf,
+				  size_t size, loff_t *offp)
 {
-	struct tool_ctx *tc = filep->private_data;
-	char buf[32];
+	struct tool_mw *inmw = filep->private_data;
+	resource_size_t addr_align;
+	resource_size_t size_align;
+	resource_size_t size_max;
+	ssize_t ret, off = 0;
 	size_t buf_size;
-	bool val;
-	int rc;
+	char *buf;
 
-	buf_size = min(size, (sizeof(buf) - 1));
-	if (copy_from_user(buf, ubuf, buf_size))
-		return -EFAULT;
+	buf_size = min_t(size_t, size, 512);
 
-	buf[buf_size] = '\0';
+	buf = kmalloc(buf_size, GFP_KERNEL);
+	if (!buf)
+		return -ENOMEM;
 
-	rc = strtobool(buf, &val);
-	if (rc)
-		return rc;
+	ret = ntb_mw_get_align(inmw->tc->ntb, inmw->pidx, inmw->widx,
+			       &addr_align, &size_align, &size_max);
+	if (ret)
+		goto err;
 
-	if (val)
-		rc = ntb_link_enable(tc->ntb, NTB_SPEED_AUTO, NTB_WIDTH_AUTO);
-	else
-		rc = ntb_link_disable(tc->ntb);
+	off += scnprintf(buf + off, buf_size - off,
+			 "Inbound MW     \t%d\n",
+			 inmw->widx);
 
-	if (rc)
-		return rc;
+	off += scnprintf(buf + off, buf_size - off,
+			 "Port           \t%d (%d)\n",
+			 ntb_peer_port_number(inmw->tc->ntb, inmw->pidx),
+			 inmw->pidx);
 
-	return size;
-}
+	off += scnprintf(buf + off, buf_size - off,
+			 "Window Address \t0x%pK\n", inmw->mm_base);
 
-static TOOL_FOPS_RDWR(tool_link_fops,
-		      tool_link_read,
-		      tool_link_write);
+	off += scnprintf(buf + off, buf_size - off,
+			 "DMA Address    \t%pad\n",
+			 &inmw->dma_base);
 
-static ssize_t tool_link_event_write(struct file *filep,
-				     const char __user *ubuf,
-				     size_t size, loff_t *offp)
-{
-	struct tool_ctx *tc = filep->private_data;
-	char buf[32];
-	size_t buf_size;
-	bool val;
-	int rc;
+	off += scnprintf(buf + off, buf_size - off,
+			 "Window Size    \t%pa[p]\n",
+			 &inmw->size);
 
-	buf_size = min(size, (sizeof(buf) - 1));
-	if (copy_from_user(buf, ubuf, buf_size))
-		return -EFAULT;
+	off += scnprintf(buf + off, buf_size - off,
+			 "Alignment      \t%pa[p]\n",
+			 &addr_align);
 
-	buf[buf_size] = '\0';
+	off += scnprintf(buf + off, buf_size - off,
+			 "Size Alignment \t%pa[p]\n",
+			 &size_align);
+
+	off += scnprintf(buf + off, buf_size - off,
+			 "Size Max       \t%pa[p]\n",
+			 &size_max);
 
-	rc = strtobool(buf, &val);
-	if (rc)
-		return rc;
+	ret = simple_read_from_buffer(ubuf, size, offp, buf, off);
 
-	if (wait_event_interruptible(tc->link_wq,
-		ntb_link_is_up(tc->ntb, NULL, NULL) == val))
-		return -ERESTART;
+err:
+	kfree(buf);
+
+	return ret;
+}
+
+static ssize_t tool_mw_trans_write(struct file *filep, const char __user *ubuf,
+				   size_t size, loff_t *offp)
+{
+	struct tool_mw *inmw = filep->private_data;
+	unsigned int val;
+	int ret;
+
+	ret = kstrtouint_from_user(ubuf, size, 0, &val);
+	if (ret)
+		return ret;
+
+	tool_free_mw(inmw->tc, inmw->pidx, inmw->widx);
+	if (val) {
+		ret = tool_setup_mw(inmw->tc, inmw->pidx, inmw->widx, val);
+		if (ret)
+			return ret;
+	}
 
 	return size;
 }
 
-static TOOL_FOPS_RDWR(tool_link_event_fops,
-		      NULL,
-		      tool_link_event_write);
+static TOOL_FOPS_RDWR(tool_mw_trans_fops,
+		      tool_mw_trans_read,
+		      tool_mw_trans_write);
 
-static ssize_t tool_mw_read(struct file *filep, char __user *ubuf,
-			    size_t size, loff_t *offp)
+static ssize_t tool_peer_mw_read(struct file *filep, char __user *ubuf,
+				 size_t size, loff_t *offp)
 {
-	struct tool_mw *mw = filep->private_data;
-	ssize_t rc;
+	struct tool_mw *outmw = filep->private_data;
 	loff_t pos = *offp;
+	ssize_t ret;
 	void *buf;
 
-	if (mw->local == NULL)
+	if (outmw->io_base == NULL)
 		return -EIO;
-	if (pos < 0)
-		return -EINVAL;
-	if (pos >= mw->win_size || !size)
+
+	if (pos >= outmw->size || !size)
 		return 0;
-	if (size > mw->win_size - pos)
-		size = mw->win_size - pos;
+
+	if (size > outmw->size - pos)
+		size = outmw->size - pos;
 
 	buf = kmalloc(size, GFP_KERNEL);
 	if (!buf)
 		return -ENOMEM;
 
-	memcpy_fromio(buf, mw->local + pos, size);
-	rc = copy_to_user(ubuf, buf, size);
-	if (rc == size) {
-		rc = -EFAULT;
+	memcpy_fromio(buf, outmw->io_base + pos, size);
+	ret = copy_to_user(ubuf, buf, size);
+	if (ret == size) {
+		ret = -EFAULT;
 		goto err_free;
 	}
 
-	size -= rc;
+	size -= ret;
 	*offp = pos + size;
-	rc = size;
+	ret = size;
 
 err_free:
 	kfree(buf);
 
-	return rc;
+	return ret;
 }
 
-static ssize_t tool_mw_write(struct file *filep, const char __user *ubuf,
-			     size_t size, loff_t *offp)
-{
-	struct tool_mw *mw = filep->private_data;
-	ssize_t rc;
+static ssize_t tool_peer_mw_write(struct file *filep, const char __user *ubuf,
+				  size_t size, loff_t *offp)
+{
+	struct tool_mw *outmw = filep->private_data;
+	ssize_t ret;
 	loff_t pos = *offp;
 	void *buf;
 
-	if (pos < 0)
-		return -EINVAL;
-	if (pos >= mw->win_size || !size)
+	if (outmw->io_base == NULL)
+		return -EIO;
+
+	if (pos >= outmw->size || !size)
 		return 0;
-	if (size > mw->win_size - pos)
-		size = mw->win_size - pos;
+	if (size > outmw->size - pos)
+		size = outmw->size - pos;
 
 	buf = kmalloc(size, GFP_KERNEL);
 	if (!buf)
 		return -ENOMEM;
 
-	rc = copy_from_user(buf, ubuf, size);
-	if (rc == size) {
-		rc = -EFAULT;
+	ret = copy_from_user(buf, ubuf, size);
+	if (ret == size) {
+		ret = -EFAULT;
 		goto err_free;
 	}
 
-	size -= rc;
+	size -= ret;
 	*offp = pos + size;
-	rc = size;
+	ret = size;
 
-	memcpy_toio(mw->local + pos, buf, size);
+	memcpy_toio(outmw->io_base + pos, buf, size);
 
 err_free:
 	kfree(buf);
 
-	return rc;
-}
-
-static TOOL_FOPS_RDWR(tool_mw_fops,
-		      tool_mw_read,
-		      tool_mw_write);
-
-static ssize_t tool_peer_mw_read(struct file *filep, char __user *ubuf,
-				 size_t size, loff_t *offp)
-{
-	struct tool_mw *mw = filep->private_data;
-
-	if (!mw->peer)
-		return -ENXIO;
-
-	return simple_read_from_buffer(ubuf, size, offp, mw->peer, mw->size);
-}
-
-static ssize_t tool_peer_mw_write(struct file *filep, const char __user *ubuf,
-				  size_t size, loff_t *offp)
-{
-	struct tool_mw *mw = filep->private_data;
-
-	if (!mw->peer)
-		return -ENXIO;
-
-	return simple_write_to_buffer(mw->peer, mw->size, offp, ubuf, size);
+	return ret;
 }
 
 static TOOL_FOPS_RDWR(tool_peer_mw_fops,
 		      tool_peer_mw_read,
 		      tool_peer_mw_write);
 
-static int tool_setup_mw(struct tool_ctx *tc, int idx, size_t req_size)
+static int tool_setup_peer_mw(struct tool_ctx *tc, int pidx, int widx,
+			      u64 req_addr, size_t req_size)
 {
-	int rc;
-	struct tool_mw *mw = &tc->mws[idx];
-	resource_size_t size, align_addr, align_size;
-	char buf[16];
+	struct tool_mw *outmw = &tc->outmws[widx];
+	resource_size_t map_size;
+	phys_addr_t map_base;
+	char buf[TOOL_BUF_LEN];
+	int ret;
 
-	if (mw->peer)
+	if (outmw->io_base != NULL)
 		return 0;
 
-	rc = ntb_mw_get_align(tc->ntb, PIDX, idx, &align_addr,
-				&align_size, &size);
-	if (rc)
-		return rc;
+	ret = ntb_peer_mw_get_addr(tc->ntb, widx, &map_base, &map_size);
+	if (ret)
+		return ret;
 
-	mw->size = min_t(resource_size_t, req_size, size);
-	mw->size = round_up(mw->size, align_addr);
-	mw->size = round_up(mw->size, align_size);
-	mw->peer = dma_alloc_coherent(&tc->ntb->pdev->dev, mw->size,
-				      &mw->peer_dma, GFP_KERNEL);
+	ret = ntb_peer_mw_set_trans(tc->ntb, pidx, widx, req_addr, req_size);
+	if (ret)
+		return ret;
 
-	if (!mw->peer || !IS_ALIGNED(mw->peer_dma, align_addr))
-		return -ENOMEM;
+	outmw->io_base = ioremap_wc(map_base, map_size);
+	if (outmw->io_base == NULL) {
+		ret = -EFAULT;
+		goto err_clear_trans;
+	}
 
-	rc = ntb_mw_set_trans(tc->ntb, PIDX, idx, mw->peer_dma, mw->size);
-	if (rc)
-		goto err_free_dma;
+	outmw->tr_base = req_addr;
+	outmw->size = req_size;
+	outmw->pidx = pidx;
 
-	snprintf(buf, sizeof(buf), "peer_mw%d", idx);
-	mw->peer_dbg_file = debugfs_create_file(buf, S_IRUSR | S_IWUSR,
-						mw->tc->dbgfs, mw,
-						&tool_peer_mw_fops);
+	snprintf(buf, sizeof(buf), "peer_mw%d", widx);
+	outmw->dbgfs_file = debugfs_create_file(buf, 0600,
+					       tc->peers[pidx].dbgfs_dir, outmw,
+					       &tool_peer_mw_fops);
 
 	return 0;
 
-err_free_dma:
-	dma_free_coherent(&tc->ntb->pdev->dev, mw->size,
-			  mw->peer,
-			  mw->peer_dma);
-	mw->peer = NULL;
-	mw->peer_dma = 0;
-	mw->size = 0;
-
-	return rc;
+err_clear_trans:
+	ntb_peer_mw_clear_trans(tc->ntb, pidx, widx);
+
+	return ret;
 }
 
-static void tool_free_mw(struct tool_ctx *tc, int idx)
+static void tool_free_peer_mw(struct tool_ctx *tc, int widx)
 {
-	struct tool_mw *mw = &tc->mws[idx];
+	struct tool_mw *outmw = &tc->outmws[widx];
 
-	if (mw->peer) {
-		ntb_mw_clear_trans(tc->ntb, PIDX, idx);
-		dma_free_coherent(&tc->ntb->pdev->dev, mw->size,
-				  mw->peer,
-				  mw->peer_dma);
-	}
-
-	mw->peer = NULL;
-	mw->peer_dma = 0;
+	debugfs_remove(outmw->dbgfs_file);
 
-	debugfs_remove(mw->peer_dbg_file);
+	if (outmw->io_base != NULL) {
+		iounmap(tc->outmws[widx].io_base);
+		ntb_peer_mw_clear_trans(tc->ntb, outmw->pidx, widx);
+	}
 
-	mw->peer_dbg_file = NULL;
+	outmw->io_base = NULL;
+	outmw->tr_base = 0;
+	outmw->size = 0;
+	outmw->pidx = -1;
+	outmw->dbgfs_file = NULL;
 }
 
-static ssize_t tool_peer_mw_trans_read(struct file *filep,
-				       char __user *ubuf,
-				       size_t size, loff_t *offp)
+static ssize_t tool_peer_mw_trans_read(struct file *filep, char __user *ubuf,
+					size_t size, loff_t *offp)
 {
-	struct tool_mw *mw = filep->private_data;
-
-	char *buf;
+	struct tool_mw_wrap *outmw_wrap = filep->private_data;
+	struct tool_mw *outmw = outmw_wrap->mw;
+	resource_size_t map_size;
+	phys_addr_t map_base;
+	ssize_t off = 0;
 	size_t buf_size;
-	ssize_t ret, off = 0;
+	char *buf;
+	int ret;
 
-	phys_addr_t base;
-	resource_size_t mw_size;
-	resource_size_t align_addr = 0;
-	resource_size_t align_size = 0;
-	resource_size_t max_size = 0;
+	ret = ntb_peer_mw_get_addr(outmw->tc->ntb, outmw->widx,
+				  &map_base, &map_size);
+	if (ret)
+		return ret;
 
 	buf_size = min_t(size_t, size, 512);
 
@@ -763,43 +890,37 @@ static ssize_t tool_peer_mw_trans_read(struct file *filep,
 	if (!buf)
 		return -ENOMEM;
 
-	ntb_mw_get_align(mw->tc->ntb, PIDX, mw->idx,
-			 &align_addr, &align_size, &max_size);
-	ntb_peer_mw_get_addr(mw->tc->ntb, mw->idx, &base, &mw_size);
-
 	off += scnprintf(buf + off, buf_size - off,
-			 "Peer MW %d Information:\n", mw->idx);
+			 "Outbound MW:        \t%d\n", outmw->widx);
 
-	off += scnprintf(buf + off, buf_size - off,
-			 "Physical Address      \t%pa[p]\n",
-			 &base);
-
-	off += scnprintf(buf + off, buf_size - off,
-			 "Window Size           \t%lld\n",
-			 (unsigned long long)mw_size);
+	if (outmw->io_base != NULL) {
+		off += scnprintf(buf + off, buf_size - off,
+			"Port attached       \t%d (%d)\n",
+			ntb_peer_port_number(outmw->tc->ntb, outmw->pidx),
+			outmw->pidx);
+	} else {
+		off += scnprintf(buf + off, buf_size - off,
+				 "Port attached       \t-1 (-1)\n");
+	}
 
 	off += scnprintf(buf + off, buf_size - off,
-			 "Alignment             \t%lld\n",
-			 (unsigned long long)align_addr);
+			 "Virtual address     \t0x%pK\n", outmw->io_base);
 
 	off += scnprintf(buf + off, buf_size - off,
-			 "Size Alignment        \t%lld\n",
-			 (unsigned long long)align_size);
+			 "Phys Address        \t%pa[p]\n", &map_base);
 
 	off += scnprintf(buf + off, buf_size - off,
-			 "Size Max              \t%lld\n",
-			 (unsigned long long)max_size);
+			 "Mapping Size        \t%pa[p]\n", &map_size);
 
 	off += scnprintf(buf + off, buf_size - off,
-			 "Ready                 \t%c\n",
-			 (mw->peer) ? 'Y' : 'N');
+			 "Translation Address \t0x%016llx\n", outmw->tr_base);
 
 	off += scnprintf(buf + off, buf_size - off,
-			 "Allocated Size       \t%zd\n",
-			 (mw->peer) ? (size_t)mw->size : 0);
+			 "Window Size         \t%pa[p]\n", &outmw->size);
 
 	ret = simple_read_from_buffer(ubuf, size, offp, buf, off);
 	kfree(buf);
+
 	return ret;
 }
 
@@ -807,12 +928,12 @@ static ssize_t tool_peer_mw_trans_write(struct file *filep,
 					const char __user *ubuf,
 					size_t size, loff_t *offp)
 {
-	struct tool_mw *mw = filep->private_data;
-
-	char buf[32];
-	size_t buf_size;
-	unsigned long long val;
-	int rc;
+	struct tool_mw_wrap *outmw_wrap = filep->private_data;
+	struct tool_mw *outmw = outmw_wrap->mw;
+	size_t buf_size, wsize;
+	char buf[TOOL_BUF_LEN];
+	int ret, n;
+	u64 addr;
 
 	buf_size = min(size, (sizeof(buf) - 1));
 	if (copy_from_user(buf, ubuf, buf_size))
@@ -820,16 +941,17 @@ static ssize_t tool_peer_mw_trans_write(struct file *filep,
 
 	buf[buf_size] = '\0';
 
-	rc = kstrtoull(buf, 0, &val);
-	if (rc)
-		return rc;
-
-	tool_free_mw(mw->tc, mw->idx);
-	if (val)
-		rc = tool_setup_mw(mw->tc, mw->idx, val);
+	n = sscanf(buf, "%lli:%zi", &addr, &wsize);
+	if (n != 2)
+		return -EINVAL;
 
-	if (rc)
-		return rc;
+	tool_free_peer_mw(outmw->tc, outmw->widx);
+	if (wsize) {
+		ret = tool_setup_peer_mw(outmw->tc, outmw_wrap->pidx,
+					 outmw->widx, addr, wsize);
+		if (ret)
+			return ret;
+	}
 
 	return size;
 }
@@ -838,195 +960,734 @@ static TOOL_FOPS_RDWR(tool_peer_mw_trans_fops,
 		      tool_peer_mw_trans_read,
 		      tool_peer_mw_trans_write);
 
-static int tool_init_mw(struct tool_ctx *tc, int idx)
+static int tool_init_mws(struct tool_ctx *tc)
 {
-	struct tool_mw *mw = &tc->mws[idx];
-	phys_addr_t base;
-	int rc;
-
-	rc = ntb_peer_mw_get_addr(tc->ntb, idx, &base, &mw->win_size);
-	if (rc)
-		return rc;
-
-	mw->tc = tc;
-	mw->idx = idx;
-	mw->local = ioremap_wc(base, mw->win_size);
-	if (!mw->local)
-		return -EFAULT;
+	int widx, pidx;
+
+	/* Initialize outbound memory windows */
+	tc->outmw_cnt = ntb_peer_mw_count(tc->ntb);
+	tc->outmws = devm_kcalloc(&tc->ntb->dev, tc->outmw_cnt,
+				  sizeof(*tc->outmws), GFP_KERNEL);
+	if (tc->outmws == NULL)
+		return -ENOMEM;
+
+	for (widx = 0; widx < tc->outmw_cnt; widx++) {
+		tc->outmws[widx].widx = widx;
+		tc->outmws[widx].pidx = -1;
+		tc->outmws[widx].tc = tc;
+	}
+
+	/* Initialize inbound memory windows and outbound MWs wrapper */
+	for (pidx = 0; pidx < tc->peer_cnt; pidx++) {
+		tc->peers[pidx].inmw_cnt = ntb_mw_count(tc->ntb, pidx);
+		tc->peers[pidx].inmws =
+			devm_kcalloc(&tc->ntb->dev, tc->peers[pidx].inmw_cnt,
+				    sizeof(*tc->peers[pidx].inmws), GFP_KERNEL);
+		if (tc->peers[pidx].inmws == NULL)
+			return -ENOMEM;
+
+		for (widx = 0; widx < tc->peers[pidx].inmw_cnt; widx++) {
+			tc->peers[pidx].inmws[widx].widx = widx;
+			tc->peers[pidx].inmws[widx].pidx = pidx;
+			tc->peers[pidx].inmws[widx].tc = tc;
+		}
+
+		tc->peers[pidx].outmw_cnt = ntb_peer_mw_count(tc->ntb);
+		tc->peers[pidx].outmws =
+			devm_kcalloc(&tc->ntb->dev, tc->peers[pidx].outmw_cnt,
+				   sizeof(*tc->peers[pidx].outmws), GFP_KERNEL);
+
+		for (widx = 0; widx < tc->peers[pidx].outmw_cnt; widx++) {
+			tc->peers[pidx].outmws[widx].pidx = pidx;
+			tc->peers[pidx].outmws[widx].mw = &tc->outmws[widx];
+		}
+	}
 
 	return 0;
 }
 
-static void tool_free_mws(struct tool_ctx *tc)
+static void tool_clear_mws(struct tool_ctx *tc)
 {
-	int i;
+	int widx, pidx;
 
-	for (i = 0; i < tc->mw_count; i++) {
-		tool_free_mw(tc, i);
+	/* Free outbound memory windows */
+	for (widx = 0; widx < tc->outmw_cnt; widx++)
+		tool_free_peer_mw(tc, widx);
 
-		if (tc->mws[i].local)
-			iounmap(tc->mws[i].local);
+	/* Free outbound memory windows */
+	for (pidx = 0; pidx < tc->peer_cnt; pidx++)
+		for (widx = 0; widx < tc->peers[pidx].inmw_cnt; widx++)
+			tool_free_mw(tc, pidx, widx);
+}
 
-		tc->mws[i].local = NULL;
-	}
+/*==============================================================================
+ *                       Doorbell read/write methods
+ *==============================================================================
+ */
+
+static ssize_t tool_db_read(struct file *filep, char __user *ubuf,
+			    size_t size, loff_t *offp)
+{
+	struct tool_ctx *tc = filep->private_data;
+
+	return tool_fn_read(tc, ubuf, size, offp, tc->ntb->ops->db_read);
 }
 
-static void tool_setup_dbgfs(struct tool_ctx *tc)
+static ssize_t tool_db_write(struct file *filep, const char __user *ubuf,
+			     size_t size, loff_t *offp)
 {
-	int i;
+	struct tool_ctx *tc = filep->private_data;
 
-	/* This modules is useless without dbgfs... */
-	if (!tool_dbgfs) {
-		tc->dbgfs = NULL;
-		return;
+	return tool_fn_write(tc, ubuf, size, offp, tc->ntb->ops->db_set,
+			     tc->ntb->ops->db_clear);
+}
+
+static TOOL_FOPS_RDWR(tool_db_fops,
+		      tool_db_read,
+		      tool_db_write);
+
+static ssize_t tool_db_valid_mask_read(struct file *filep, char __user *ubuf,
+				       size_t size, loff_t *offp)
+{
+	struct tool_ctx *tc = filep->private_data;
+
+	return tool_fn_read(tc, ubuf, size, offp, tc->ntb->ops->db_valid_mask);
+}
+
+static TOOL_FOPS_RDWR(tool_db_valid_mask_fops,
+		      tool_db_valid_mask_read,
+		      NULL);
+
+static ssize_t tool_db_mask_read(struct file *filep, char __user *ubuf,
+				 size_t size, loff_t *offp)
+{
+	struct tool_ctx *tc = filep->private_data;
+
+	return tool_fn_read(tc, ubuf, size, offp, tc->ntb->ops->db_read_mask);
+}
+
+static ssize_t tool_db_mask_write(struct file *filep, const char __user *ubuf,
+			       size_t size, loff_t *offp)
+{
+	struct tool_ctx *tc = filep->private_data;
+
+	return tool_fn_write(tc, ubuf, size, offp, tc->ntb->ops->db_set_mask,
+			     tc->ntb->ops->db_clear_mask);
+}
+
+static TOOL_FOPS_RDWR(tool_db_mask_fops,
+		      tool_db_mask_read,
+		      tool_db_mask_write);
+
+static ssize_t tool_peer_db_read(struct file *filep, char __user *ubuf,
+				 size_t size, loff_t *offp)
+{
+	struct tool_ctx *tc = filep->private_data;
+
+	return tool_fn_read(tc, ubuf, size, offp, tc->ntb->ops->peer_db_read);
+}
+
+static ssize_t tool_peer_db_write(struct file *filep, const char __user *ubuf,
+				  size_t size, loff_t *offp)
+{
+	struct tool_ctx *tc = filep->private_data;
+
+	return tool_fn_write(tc, ubuf, size, offp, tc->ntb->ops->peer_db_set,
+			     tc->ntb->ops->peer_db_clear);
+}
+
+static TOOL_FOPS_RDWR(tool_peer_db_fops,
+		      tool_peer_db_read,
+		      tool_peer_db_write);
+
+static ssize_t tool_peer_db_mask_read(struct file *filep, char __user *ubuf,
+				   size_t size, loff_t *offp)
+{
+	struct tool_ctx *tc = filep->private_data;
+
+	return tool_fn_read(tc, ubuf, size, offp,
+			    tc->ntb->ops->peer_db_read_mask);
+}
+
+static ssize_t tool_peer_db_mask_write(struct file *filep,
+				       const char __user *ubuf,
+				       size_t size, loff_t *offp)
+{
+	struct tool_ctx *tc = filep->private_data;
+
+	return tool_fn_write(tc, ubuf, size, offp,
+			     tc->ntb->ops->peer_db_set_mask,
+			     tc->ntb->ops->peer_db_clear_mask);
+}
+
+static TOOL_FOPS_RDWR(tool_peer_db_mask_fops,
+		      tool_peer_db_mask_read,
+		      tool_peer_db_mask_write);
+
+static ssize_t tool_db_event_write(struct file *filep,
+				   const char __user *ubuf,
+				   size_t size, loff_t *offp)
+{
+	struct tool_ctx *tc = filep->private_data;
+	u64 val;
+	int ret;
+
+	ret = kstrtou64_from_user(ubuf, size, 0, &val);
+	if (ret)
+		return ret;
+
+	if (wait_event_interruptible(tc->db_wq, ntb_db_read(tc->ntb) == val))
+		return -ERESTART;
+
+	return size;
+}
+
+static TOOL_FOPS_RDWR(tool_db_event_fops,
+		      NULL,
+		      tool_db_event_write);
+
+/*==============================================================================
+ *                       Scratchpads read/write methods
+ *==============================================================================
+ */
+
+static ssize_t tool_spad_read(struct file *filep, char __user *ubuf,
+			      size_t size, loff_t *offp)
+{
+	struct tool_spad *spad = filep->private_data;
+	char buf[TOOL_BUF_LEN];
+	ssize_t pos;
+
+	if (!spad->tc->ntb->ops->spad_read)
+		return -EINVAL;
+
+	pos = scnprintf(buf, sizeof(buf), "%#x\n",
+		ntb_spad_read(spad->tc->ntb, spad->sidx));
+
+	return simple_read_from_buffer(ubuf, size, offp, buf, pos);
+}
+
+static ssize_t tool_spad_write(struct file *filep, const char __user *ubuf,
+			       size_t size, loff_t *offp)
+{
+	struct tool_spad *spad = filep->private_data;
+	u32 val;
+	int ret;
+
+	if (!spad->tc->ntb->ops->spad_write) {
+		dev_dbg(&spad->tc->ntb->dev, "no spad write fn\n");
+		return -EINVAL;
 	}
 
-	tc->dbgfs = debugfs_create_dir(dev_name(&tc->ntb->dev),
-				       tool_dbgfs);
-	if (!tc->dbgfs)
-		return;
+	ret = kstrtou32_from_user(ubuf, size, 0, &val);
+	if (ret)
+		return ret;
 
-	debugfs_create_file("db", S_IRUSR | S_IWUSR, tc->dbgfs,
-			    tc, &tool_db_fops);
+	ret = ntb_spad_write(spad->tc->ntb, spad->sidx, val);
 
-	debugfs_create_file("mask", S_IRUSR | S_IWUSR, tc->dbgfs,
-			    tc, &tool_mask_fops);
+	return ret ?: size;
+}
 
-	debugfs_create_file("peer_db", S_IRUSR | S_IWUSR, tc->dbgfs,
-			    tc, &tool_peer_db_fops);
+static TOOL_FOPS_RDWR(tool_spad_fops,
+		      tool_spad_read,
+		      tool_spad_write);
 
-	debugfs_create_file("peer_mask", S_IRUSR | S_IWUSR, tc->dbgfs,
-			    tc, &tool_peer_mask_fops);
+static ssize_t tool_peer_spad_read(struct file *filep, char __user *ubuf,
+				   size_t size, loff_t *offp)
+{
+	struct tool_spad *spad = filep->private_data;
+	char buf[TOOL_BUF_LEN];
+	ssize_t pos;
 
-	debugfs_create_file("spad", S_IRUSR | S_IWUSR, tc->dbgfs,
-			    tc, &tool_spad_fops);
+	if (!spad->tc->ntb->ops->peer_spad_read)
+		return -EINVAL;
 
-	debugfs_create_file("peer_spad", S_IRUSR | S_IWUSR, tc->dbgfs,
-			    tc, &tool_peer_spad_fops);
+	pos = scnprintf(buf, sizeof(buf), "%#x\n",
+		ntb_peer_spad_read(spad->tc->ntb, spad->pidx, spad->sidx));
 
-	debugfs_create_file("link", S_IRUSR | S_IWUSR, tc->dbgfs,
-			    tc, &tool_link_fops);
+	return simple_read_from_buffer(ubuf, size, offp, buf, pos);
+}
+
+static ssize_t tool_peer_spad_write(struct file *filep, const char __user *ubuf,
+				    size_t size, loff_t *offp)
+{
+	struct tool_spad *spad = filep->private_data;
+	u32 val;
+	int ret;
+
+	if (!spad->tc->ntb->ops->peer_spad_write) {
+		dev_dbg(&spad->tc->ntb->dev, "no spad write fn\n");
+		return -EINVAL;
+	}
 
-	debugfs_create_file("link_event", S_IWUSR, tc->dbgfs,
-			    tc, &tool_link_event_fops);
+	ret = kstrtou32_from_user(ubuf, size, 0, &val);
+	if (ret)
+		return ret;
 
-	for (i = 0; i < tc->mw_count; i++) {
-		char buf[30];
+	ret = ntb_peer_spad_write(spad->tc->ntb, spad->pidx, spad->sidx, val);
 
-		snprintf(buf, sizeof(buf), "mw%d", i);
-		debugfs_create_file(buf, S_IRUSR | S_IWUSR, tc->dbgfs,
-				    &tc->mws[i], &tool_mw_fops);
+	return ret ?: size;
+}
+
+static TOOL_FOPS_RDWR(tool_peer_spad_fops,
+		      tool_peer_spad_read,
+		      tool_peer_spad_write);
+
+static int tool_init_spads(struct tool_ctx *tc)
+{
+	int sidx, pidx;
+
+	/* Initialize inbound scratchpad structures */
+	tc->inspad_cnt = ntb_spad_count(tc->ntb);
+	tc->inspads = devm_kcalloc(&tc->ntb->dev, tc->inspad_cnt,
+				   sizeof(*tc->inspads), GFP_KERNEL);
+	if (tc->inspads == NULL)
+		return -ENOMEM;
+
+	for (sidx = 0; sidx < tc->inspad_cnt; sidx++) {
+		tc->inspads[sidx].sidx = sidx;
+		tc->inspads[sidx].pidx = -1;
+		tc->inspads[sidx].tc = tc;
+	}
 
-		snprintf(buf, sizeof(buf), "peer_trans%d", i);
-		debugfs_create_file(buf, S_IRUSR | S_IWUSR, tc->dbgfs,
-				    &tc->mws[i], &tool_peer_mw_trans_fops);
+	/* Initialize outbound scratchpad structures */
+	for (pidx = 0; pidx < tc->peer_cnt; pidx++) {
+		tc->peers[pidx].outspad_cnt = ntb_spad_count(tc->ntb);
+		tc->peers[pidx].outspads =
+			devm_kcalloc(&tc->ntb->dev, tc->peers[pidx].outspad_cnt,
+				sizeof(*tc->peers[pidx].outspads), GFP_KERNEL);
+		if (tc->peers[pidx].outspads == NULL)
+			return -ENOMEM;
+
+		for (sidx = 0; sidx < tc->peers[pidx].outspad_cnt; sidx++) {
+			tc->peers[pidx].outspads[sidx].sidx = sidx;
+			tc->peers[pidx].outspads[sidx].pidx = pidx;
+			tc->peers[pidx].outspads[sidx].tc = tc;
+		}
 	}
+
+	return 0;
 }
 
-static int tool_probe(struct ntb_client *self, struct ntb_dev *ntb)
+/*==============================================================================
+ *                       Messages read/write methods
+ *==============================================================================
+ */
+
+static ssize_t tool_inmsg_read(struct file *filep, char __user *ubuf,
+			       size_t size, loff_t *offp)
 {
-	struct tool_ctx *tc;
-	int rc;
-	int i;
+	struct tool_msg *msg = filep->private_data;
+	char buf[TOOL_BUF_LEN];
+	ssize_t pos;
+	u32 data;
+	int pidx;
+
+	data = ntb_msg_read(msg->tc->ntb, &pidx, msg->midx);
+
+	pos = scnprintf(buf, sizeof(buf), "0x%08x<-%d\n", data, pidx);
+
+	return simple_read_from_buffer(ubuf, size, offp, buf, pos);
+}
+
+static TOOL_FOPS_RDWR(tool_inmsg_fops,
+		      tool_inmsg_read,
+		      NULL);
+
+static ssize_t tool_outmsg_write(struct file *filep,
+				 const char __user *ubuf,
+				 size_t size, loff_t *offp)
+{
+	struct tool_msg *msg = filep->private_data;
+	u32 val;
+	int ret;
+
+	ret = kstrtou32_from_user(ubuf, size, 0, &val);
+	if (ret)
+		return ret;
+
+	ret = ntb_peer_msg_write(msg->tc->ntb, msg->pidx, msg->midx, val);
+
+	return ret ? : size;
+}
+
+static TOOL_FOPS_RDWR(tool_outmsg_fops,
+		      NULL,
+		      tool_outmsg_write);
+
+static ssize_t tool_msg_sts_read(struct file *filep, char __user *ubuf,
+				 size_t size, loff_t *offp)
+{
+	struct tool_ctx *tc = filep->private_data;
+
+	return tool_fn_read(tc, ubuf, size, offp, tc->ntb->ops->msg_read_sts);
+}
+
+static ssize_t tool_msg_sts_write(struct file *filep, const char __user *ubuf,
+				  size_t size, loff_t *offp)
+{
+	struct tool_ctx *tc = filep->private_data;
+
+	return tool_fn_write(tc, ubuf, size, offp, NULL,
+			     tc->ntb->ops->msg_clear_sts);
+}
+
+static TOOL_FOPS_RDWR(tool_msg_sts_fops,
+		      tool_msg_sts_read,
+		      tool_msg_sts_write);
+
+static ssize_t tool_msg_inbits_read(struct file *filep, char __user *ubuf,
+				    size_t size, loff_t *offp)
+{
+	struct tool_ctx *tc = filep->private_data;
+
+	return tool_fn_read(tc, ubuf, size, offp, tc->ntb->ops->msg_inbits);
+}
+
+static TOOL_FOPS_RDWR(tool_msg_inbits_fops,
+		      tool_msg_inbits_read,
+		      NULL);
+
+static ssize_t tool_msg_outbits_read(struct file *filep, char __user *ubuf,
+				     size_t size, loff_t *offp)
+{
+	struct tool_ctx *tc = filep->private_data;
+
+	return tool_fn_read(tc, ubuf, size, offp, tc->ntb->ops->msg_outbits);
+}
+
+static TOOL_FOPS_RDWR(tool_msg_outbits_fops,
+		      tool_msg_outbits_read,
+		      NULL);
+
+static ssize_t tool_msg_mask_write(struct file *filep, const char __user *ubuf,
+				   size_t size, loff_t *offp)
+{
+	struct tool_ctx *tc = filep->private_data;
+
+	return tool_fn_write(tc, ubuf, size, offp,
+			     tc->ntb->ops->msg_set_mask,
+			     tc->ntb->ops->msg_clear_mask);
+}
+
+static TOOL_FOPS_RDWR(tool_msg_mask_fops,
+		      NULL,
+		      tool_msg_mask_write);
+
+static ssize_t tool_msg_event_write(struct file *filep,
+				    const char __user *ubuf,
+				    size_t size, loff_t *offp)
+{
+	struct tool_ctx *tc = filep->private_data;
+	u64 val;
+	int ret;
+
+	ret = kstrtou64_from_user(ubuf, size, 0, &val);
+	if (ret)
+		return ret;
+
+	if (wait_event_interruptible(tc->msg_wq,
+		ntb_msg_read_sts(tc->ntb) == val))
+		return -ERESTART;
+
+	return size;
+}
+
+static TOOL_FOPS_RDWR(tool_msg_event_fops,
+		      NULL,
+		      tool_msg_event_write);
 
-	if (!ntb->ops->mw_set_trans) {
-		dev_dbg(&ntb->dev, "need inbound MW based NTB API\n");
-		rc = -EINVAL;
-		goto err_tc;
+static int tool_init_msgs(struct tool_ctx *tc)
+{
+	int midx, pidx;
+
+	/* Initialize inbound message structures */
+	tc->inmsg_cnt = ntb_msg_count(tc->ntb);
+	tc->inmsgs = devm_kcalloc(&tc->ntb->dev, tc->inmsg_cnt,
+				   sizeof(*tc->inmsgs), GFP_KERNEL);
+	if (tc->inmsgs == NULL)
+		return -ENOMEM;
+
+	for (midx = 0; midx < tc->inmsg_cnt; midx++) {
+		tc->inmsgs[midx].midx = midx;
+		tc->inmsgs[midx].pidx = -1;
+		tc->inmsgs[midx].tc = tc;
 	}
 
-	if (ntb_spad_count(ntb) < 1) {
-		dev_dbg(&ntb->dev, "no enough scratchpads\n");
-		rc = -EINVAL;
-		goto err_tc;
+	/* Initialize outbound message structures */
+	for (pidx = 0; pidx < tc->peer_cnt; pidx++) {
+		tc->peers[pidx].outmsg_cnt = ntb_msg_count(tc->ntb);
+		tc->peers[pidx].outmsgs =
+			devm_kcalloc(&tc->ntb->dev, tc->peers[pidx].outmsg_cnt,
+				sizeof(*tc->peers[pidx].outmsgs), GFP_KERNEL);
+		if (tc->peers[pidx].outmsgs == NULL)
+			return -ENOMEM;
+
+		for (midx = 0; midx < tc->peers[pidx].outmsg_cnt; midx++) {
+			tc->peers[pidx].outmsgs[midx].midx = midx;
+			tc->peers[pidx].outmsgs[midx].pidx = pidx;
+			tc->peers[pidx].outmsgs[midx].tc = tc;
+		}
 	}
 
+	return 0;
+}
+
+/*==============================================================================
+ *                          Initialization methods
+ *==============================================================================
+ */
+
+static struct tool_ctx *tool_create_data(struct ntb_dev *ntb)
+{
+	struct tool_ctx *tc;
+
+	tc = devm_kzalloc(&ntb->dev, sizeof(*tc), GFP_KERNEL);
+	if (tc == NULL)
+		return ERR_PTR(-ENOMEM);
+
+	tc->ntb = ntb;
+	init_waitqueue_head(&tc->link_wq);
+	init_waitqueue_head(&tc->db_wq);
+	init_waitqueue_head(&tc->msg_wq);
+
 	if (ntb_db_is_unsafe(ntb))
 		dev_dbg(&ntb->dev, "doorbell is unsafe\n");
 
 	if (ntb_spad_is_unsafe(ntb))
 		dev_dbg(&ntb->dev, "scratchpad is unsafe\n");
 
-	if (ntb_peer_port_count(ntb) != NTB_DEF_PEER_CNT)
-		dev_warn(&ntb->dev, "multi-port NTB is unsupported\n");
+	return tc;
+}
+
+static void tool_clear_data(struct tool_ctx *tc)
+{
+	wake_up(&tc->link_wq);
+	wake_up(&tc->db_wq);
+	wake_up(&tc->msg_wq);
+}
+
+static int tool_init_ntb(struct tool_ctx *tc)
+{
+	return ntb_set_ctx(tc->ntb, tc, &tool_ops);
+}
+
+static void tool_clear_ntb(struct tool_ctx *tc)
+{
+	ntb_clear_ctx(tc->ntb);
+	ntb_link_disable(tc->ntb);
+}
 
-	tc = kzalloc(sizeof(*tc), GFP_KERNEL);
-	if (!tc) {
-		rc = -ENOMEM;
-		goto err_tc;
+static void tool_setup_dbgfs(struct tool_ctx *tc)
+{
+	int pidx, widx, sidx, midx;
+	char buf[TOOL_BUF_LEN];
+
+	/* This modules is useless without dbgfs... */
+	if (!tool_dbgfs_topdir) {
+		tc->dbgfs_dir = NULL;
+		return;
 	}
 
-	tc->ntb = ntb;
-	init_waitqueue_head(&tc->link_wq);
+	tc->dbgfs_dir = debugfs_create_dir(dev_name(&tc->ntb->dev),
+					   tool_dbgfs_topdir);
+	if (!tc->dbgfs_dir)
+		return;
 
-	tc->mw_count = min(ntb_peer_mw_count(tc->ntb), MAX_MWS);
-	for (i = 0; i < tc->mw_count; i++) {
-		rc = tool_init_mw(tc, i);
-		if (rc)
-			goto err_ctx;
+	debugfs_create_file("port", 0600, tc->dbgfs_dir,
+			    tc, &tool_port_fops);
+
+	debugfs_create_file("link", 0600, tc->dbgfs_dir,
+			    tc, &tool_link_fops);
+
+	debugfs_create_file("db", 0600, tc->dbgfs_dir,
+			    tc, &tool_db_fops);
+
+	debugfs_create_file("db_valid_mask", 0600, tc->dbgfs_dir,
+			    tc, &tool_db_valid_mask_fops);
+
+	debugfs_create_file("db_mask", 0600, tc->dbgfs_dir,
+			    tc, &tool_db_mask_fops);
+
+	debugfs_create_file("db_event", 0600, tc->dbgfs_dir,
+			    tc, &tool_db_event_fops);
+
+	debugfs_create_file("peer_db", 0600, tc->dbgfs_dir,
+			    tc, &tool_peer_db_fops);
+
+	debugfs_create_file("peer_db_mask", 0600, tc->dbgfs_dir,
+			    tc, &tool_peer_db_mask_fops);
+
+	if (tc->inspad_cnt != 0) {
+		for (sidx = 0; sidx < tc->inspad_cnt; sidx++) {
+			snprintf(buf, sizeof(buf), "spad%d", sidx);
+
+			debugfs_create_file(buf, 0600, tc->dbgfs_dir,
+					   &tc->inspads[sidx], &tool_spad_fops);
+		}
 	}
 
-	tool_setup_dbgfs(tc);
+	if (tc->inmsg_cnt != 0) {
+		for (midx = 0; midx < tc->inmsg_cnt; midx++) {
+			snprintf(buf, sizeof(buf), "msg%d", midx);
+			debugfs_create_file(buf, 0600, tc->dbgfs_dir,
+					   &tc->inmsgs[midx], &tool_inmsg_fops);
+		}
+
+		debugfs_create_file("msg_sts", 0600, tc->dbgfs_dir,
+				    tc, &tool_msg_sts_fops);
+
+		debugfs_create_file("msg_inbits", 0600, tc->dbgfs_dir,
+				    tc, &tool_msg_inbits_fops);
+
+		debugfs_create_file("msg_outbits", 0600, tc->dbgfs_dir,
+				    tc, &tool_msg_outbits_fops);
+
+		debugfs_create_file("msg_mask", 0600, tc->dbgfs_dir,
+				    tc, &tool_msg_mask_fops);
+
+		debugfs_create_file("msg_event", 0600, tc->dbgfs_dir,
+				    tc, &tool_msg_event_fops);
+	}
+
+	for (pidx = 0; pidx < tc->peer_cnt; pidx++) {
+		snprintf(buf, sizeof(buf), "peer%d", pidx);
+		tc->peers[pidx].dbgfs_dir =
+			debugfs_create_dir(buf, tc->dbgfs_dir);
+
+		debugfs_create_file("port", 0600,
+				    tc->peers[pidx].dbgfs_dir,
+				    &tc->peers[pidx], &tool_peer_port_fops);
+
+		debugfs_create_file("link", 0200,
+				    tc->peers[pidx].dbgfs_dir,
+				    &tc->peers[pidx], &tool_peer_link_fops);
+
+		debugfs_create_file("link_event", 0200,
+				  tc->peers[pidx].dbgfs_dir,
+				  &tc->peers[pidx], &tool_peer_link_event_fops);
+
+		for (widx = 0; widx < tc->peers[pidx].inmw_cnt; widx++) {
+			snprintf(buf, sizeof(buf), "mw_trans%d", widx);
+			debugfs_create_file(buf, 0600,
+					    tc->peers[pidx].dbgfs_dir,
+					    &tc->peers[pidx].inmws[widx],
+					    &tool_mw_trans_fops);
+		}
+
+		for (widx = 0; widx < tc->peers[pidx].outmw_cnt; widx++) {
+			snprintf(buf, sizeof(buf), "peer_mw_trans%d", widx);
+			debugfs_create_file(buf, 0600,
+					    tc->peers[pidx].dbgfs_dir,
+					    &tc->peers[pidx].outmws[widx],
+					    &tool_peer_mw_trans_fops);
+		}
+
+		for (sidx = 0; sidx < tc->peers[pidx].outspad_cnt; sidx++) {
+			snprintf(buf, sizeof(buf), "spad%d", sidx);
+
+			debugfs_create_file(buf, 0600,
+					    tc->peers[pidx].dbgfs_dir,
+					    &tc->peers[pidx].outspads[sidx],
+					    &tool_peer_spad_fops);
+		}
+
+		for (midx = 0; midx < tc->peers[pidx].outmsg_cnt; midx++) {
+			snprintf(buf, sizeof(buf), "msg%d", midx);
+			debugfs_create_file(buf, 0600,
+					    tc->peers[pidx].dbgfs_dir,
+					    &tc->peers[pidx].outmsgs[midx],
+					    &tool_outmsg_fops);
+		}
+	}
+}
+
+static void tool_clear_dbgfs(struct tool_ctx *tc)
+{
+	debugfs_remove_recursive(tc->dbgfs_dir);
+}
+
+static int tool_probe(struct ntb_client *self, struct ntb_dev *ntb)
+{
+	struct tool_ctx *tc;
+	int ret;
 
-	rc = ntb_set_ctx(ntb, tc, &tool_ops);
-	if (rc)
-		goto err_ctx;
+	tc = tool_create_data(ntb);
+	if (IS_ERR(tc))
+		return PTR_ERR(tc);
 
-	ntb_link_enable(ntb, NTB_SPEED_AUTO, NTB_WIDTH_AUTO);
-	ntb_link_event(ntb);
+	ret = tool_init_peers(tc);
+	if (ret != 0)
+		goto err_clear_data;
+
+	ret = tool_init_mws(tc);
+	if (ret != 0)
+		goto err_clear_data;
+
+	ret = tool_init_spads(tc);
+	if (ret != 0)
+		goto err_clear_mws;
+
+	ret = tool_init_msgs(tc);
+	if (ret != 0)
+		goto err_clear_mws;
+
+	ret = tool_init_ntb(tc);
+	if (ret != 0)
+		goto err_clear_mws;
+
+	tool_setup_dbgfs(tc);
 
 	return 0;
 
-err_ctx:
-	tool_free_mws(tc);
-	debugfs_remove_recursive(tc->dbgfs);
-	kfree(tc);
-err_tc:
-	return rc;
+err_clear_mws:
+	tool_clear_mws(tc);
+
+err_clear_data:
+	tool_clear_data(tc);
+
+	return ret;
 }
 
 static void tool_remove(struct ntb_client *self, struct ntb_dev *ntb)
 {
 	struct tool_ctx *tc = ntb->ctx;
 
-	tool_free_mws(tc);
+	tool_clear_dbgfs(tc);
+
+	tool_clear_ntb(tc);
 
-	ntb_clear_ctx(ntb);
-	ntb_link_disable(ntb);
+	tool_clear_mws(tc);
 
-	debugfs_remove_recursive(tc->dbgfs);
-	kfree(tc);
+	tool_clear_data(tc);
 }
 
 static struct ntb_client tool_client = {
 	.ops = {
 		.probe = tool_probe,
 		.remove = tool_remove,
-	},
+	}
 };
 
 static int __init tool_init(void)
 {
-	int rc;
+	int ret;
 
 	if (debugfs_initialized())
-		tool_dbgfs = debugfs_create_dir(KBUILD_MODNAME, NULL);
-
-	rc = ntb_register_client(&tool_client);
-	if (rc)
-		goto err_client;
+		tool_dbgfs_topdir = debugfs_create_dir(KBUILD_MODNAME, NULL);
 
-	return 0;
+	ret = ntb_register_client(&tool_client);
+	if (ret)
+		debugfs_remove_recursive(tool_dbgfs_topdir);
 
-err_client:
-	debugfs_remove_recursive(tool_dbgfs);
-	return rc;
+	return ret;
 }
 module_init(tool_init);
 
 static void __exit tool_exit(void)
 {
 	ntb_unregister_client(&tool_client);
-	debugfs_remove_recursive(tool_dbgfs);
+	debugfs_remove_recursive(tool_dbgfs_topdir);
 }
 module_exit(tool_exit);
+

include/linux/ntb.h

@@ -71,6 +71,7 @@ struct pci_dev;
  * @NTB_TOPO_B2B_USD:	On primary side of local ntb upstream of remote ntb.
  * @NTB_TOPO_B2B_DSD:	On primary side of local ntb downstream of remote ntb.
  * @NTB_TOPO_SWITCH:	Connected via a switch which supports ntb.
+ * @NTB_TOPO_CROSSLINK: Connected via two symmetric switchecs
  */
 enum ntb_topo {
 	NTB_TOPO_NONE = -1,
@@ -79,6 +80,7 @@ enum ntb_topo {
 	NTB_TOPO_B2B_USD,
 	NTB_TOPO_B2B_DSD,
 	NTB_TOPO_SWITCH,
+	NTB_TOPO_CROSSLINK,
 };
 
 static inline int ntb_topo_is_b2b(enum ntb_topo topo)
@@ -94,12 +96,13 @@ static inline int ntb_topo_is_b2b(enum ntb_topo topo)
 static inline char *ntb_topo_string(enum ntb_topo topo)
 {
 	switch (topo) {
-	case NTB_TOPO_NONE:	return "NTB_TOPO_NONE";
-	case NTB_TOPO_PRI:	return "NTB_TOPO_PRI";
-	case NTB_TOPO_SEC:	return "NTB_TOPO_SEC";
-	case NTB_TOPO_B2B_USD:	return "NTB_TOPO_B2B_USD";
-	case NTB_TOPO_B2B_DSD:	return "NTB_TOPO_B2B_DSD";
-	case NTB_TOPO_SWITCH:	return "NTB_TOPO_SWITCH";
+	case NTB_TOPO_NONE:		return "NTB_TOPO_NONE";
+	case NTB_TOPO_PRI:		return "NTB_TOPO_PRI";
+	case NTB_TOPO_SEC:		return "NTB_TOPO_SEC";
+	case NTB_TOPO_B2B_USD:		return "NTB_TOPO_B2B_USD";
+	case NTB_TOPO_B2B_DSD:		return "NTB_TOPO_B2B_DSD";
+	case NTB_TOPO_SWITCH:		return "NTB_TOPO_SWITCH";
+	case NTB_TOPO_CROSSLINK:	return "NTB_TOPO_CROSSLINK";
 	}
 	return "NTB_TOPO_INVALID";
 }
@@ -250,7 +253,7 @@ static inline int ntb_ctx_ops_is_valid(const struct ntb_ctx_ops *ops)
  * @msg_set_mask:	See ntb_msg_set_mask().
  * @msg_clear_mask:	See ntb_msg_clear_mask().
  * @msg_read:		See ntb_msg_read().
- * @msg_write:		See ntb_msg_write().
+ * @peer_msg_write:	See ntb_peer_msg_write().
  */
 struct ntb_dev_ops {
 	int (*port_number)(struct ntb_dev *ntb);
@@ -321,8 +324,8 @@ struct ntb_dev_ops {
 	int (*msg_clear_sts)(struct ntb_dev *ntb, u64 sts_bits);
 	int (*msg_set_mask)(struct ntb_dev *ntb, u64 mask_bits);
 	int (*msg_clear_mask)(struct ntb_dev *ntb, u64 mask_bits);
-	int (*msg_read)(struct ntb_dev *ntb, int midx, int *pidx, u32 *msg);
-	int (*msg_write)(struct ntb_dev *ntb, int midx, int pidx, u32 msg);
+	u32 (*msg_read)(struct ntb_dev *ntb, int *pidx, int midx);
+	int (*peer_msg_write)(struct ntb_dev *ntb, int pidx, int midx, u32 msg);
 };
 
 static inline int ntb_dev_ops_is_valid(const struct ntb_dev_ops *ops)
@@ -384,7 +387,7 @@ static inline int ntb_dev_ops_is_valid(const struct ntb_dev_ops *ops)
 		/* !ops->msg_set_mask == !ops->msg_count	&& */
 		/* !ops->msg_clear_mask == !ops->msg_count	&& */
 		!ops->msg_read == !ops->msg_count		&&
-		!ops->msg_write == !ops->msg_count		&&
+		!ops->peer_msg_write == !ops->msg_count		&&
 		1;
 }
 
@@ -764,7 +767,7 @@ static inline int ntb_mw_get_align(struct ntb_dev *ntb, int pidx, int widx,
 				   resource_size_t *size_align,
 				   resource_size_t *size_max)
 {
-	if (!(ntb_link_is_up(ntb, NULL, NULL) & (1 << pidx)))
+	if (!(ntb_link_is_up(ntb, NULL, NULL) & BIT_ULL(pidx)))
 		return -ENOTCONN;
 
 	return ntb->ops->mw_get_align(ntb, pidx, widx, addr_align, size_align,
@@ -1459,31 +1462,29 @@ static inline int ntb_msg_clear_mask(struct ntb_dev *ntb, u64 mask_bits)
 }
 
 /**
- * ntb_msg_read() - read message register with specified index
+ * ntb_msg_read() - read inbound message register with specified index
  * @ntb:	NTB device context.
- * @midx:	Message register index
  * @pidx:	OUT - Port index of peer device a message retrieved from
- * @msg:	OUT - Data
+ * @midx:	Message register index
  *
  * Read data from the specified message register. Source port index of a
  * message is retrieved as well.
  *
- * Return: Zero on success, otherwise a negative error number.
+ * Return: The value of the inbound message register.
  */
-static inline int ntb_msg_read(struct ntb_dev *ntb, int midx, int *pidx,
-			       u32 *msg)
+static inline u32 ntb_msg_read(struct ntb_dev *ntb, int *pidx, int midx)
 {
 	if (!ntb->ops->msg_read)
-		return -EINVAL;
+		return ~(u32)0;
 
-	return ntb->ops->msg_read(ntb, midx, pidx, msg);
+	return ntb->ops->msg_read(ntb, pidx, midx);
 }
 
 /**
- * ntb_msg_write() - write data to the specified message register
+ * ntb_peer_msg_write() - write data to the specified peer message register
  * @ntb:	NTB device context.
- * @midx:	Message register index
  * @pidx:	Port index of peer device a message being sent to
+ * @midx:	Message register index
  * @msg:	Data to send
  *
  * Send data to a specified peer device using the defined message register.
@@ -1492,13 +1493,13 @@ static inline int ntb_msg_read(struct ntb_dev *ntb, int midx, int *pidx,
  *
  * Return: Zero on success, otherwise a negative error number.
  */
-static inline int ntb_msg_write(struct ntb_dev *ntb, int midx, int pidx,
-				u32 msg)
+static inline int ntb_peer_msg_write(struct ntb_dev *ntb, int pidx, int midx,
+				     u32 msg)
 {
-	if (!ntb->ops->msg_write)
+	if (!ntb->ops->peer_msg_write)
 		return -EINVAL;
 
-	return ntb->ops->msg_write(ntb, midx, pidx, msg);
+	return ntb->ops->peer_msg_write(ntb, pidx, midx, msg);
 }
 
 #endif

include/linux/switchtec.h

@@ -168,6 +168,14 @@ struct ntb_info_regs {
 	u16 reserved1;
 	u64 ep_map;
 	u16 requester_id;
+	u16 reserved2;
+	u32 reserved3[4];
+	struct nt_partition_info {
+		u32 xlink_enabled;
+		u32 target_part_low;
+		u32 target_part_high;
+		u32 reserved;
+	} ntp_info[48];
 } __packed;
 
 struct part_cfg_regs {
@@ -284,7 +292,20 @@ enum {
 struct pff_csr_regs {
 	u16 vendor_id;
 	u16 device_id;
-	u32 pci_cfg_header[15];
+	u16 pcicmd;
+	u16 pcists;
+	u32 pci_class;
+	u32 pci_opts;
+	union {
+		u32 pci_bar[6];
+		u64 pci_bar64[3];
+	};
+	u32 pci_cardbus;
+	u32 pci_subsystem_id;
+	u32 pci_expansion_rom;
+	u32 pci_cap_ptr;
+	u32 reserved1;
+	u32 pci_irq;
 	u32 pci_cap_region[48];
 	u32 pcie_cap_region[448];
 	u32 indirect_gas_window[128];

tools/testing/selftests/ntb/ntb_test.sh

@@ -18,7 +18,6 @@ LIST_DEVS=FALSE
 
 DEBUGFS=${DEBUGFS-/sys/kernel/debug}
 
-DB_BITMASK=0x7FFF
 PERF_RUN_ORDER=32
 MAX_MW_SIZE=0
 RUN_DMA_TESTS=
@@ -39,15 +38,17 @@ function show_help()
 	echo "be highly recommended."
 	echo
 	echo "Options:"
-	echo "  -b BITMASK      doorbell clear bitmask for ntb_tool"
 	echo "  -C              don't cleanup ntb modules on exit"
-	echo "  -d              run dma tests"
 	echo "  -h              show this help message"
 	echo "  -l              list available local and remote PCI ids"
 	echo "  -r REMOTE_HOST  specify the remote's hostname to connect"
-        echo "                  to for the test (using ssh)"
-	echo "  -p NUM          ntb_perf run order (default: $PERF_RUN_ORDER)"
-	echo "  -w max_mw_size  maxmium memory window size"
+	echo "                  to for the test (using ssh)"
+	echo "  -m MW_SIZE      memory window size for ntb_tool"
+	echo "                  (default: $MW_SIZE)"
+	echo "  -d              run dma tests for ntb_perf"
+	echo "  -p ORDER        total data order for ntb_perf"
+	echo "                  (default: $PERF_RUN_ORDER)"
+	echo "  -w MAX_MW_SIZE  maxmium memory window size for ntb_perf"
 	echo
 }
 
@@ -56,7 +57,6 @@ function parse_args()
 	OPTIND=0
 	while getopts "b:Cdhlm:r:p:w:" opt; do
 		case "$opt" in
-		b)  DB_BITMASK=${OPTARG} ;;
 		C)  DONT_CLEANUP=1 ;;
 		d)  RUN_DMA_TESTS=1 ;;
 		h)  show_help; exit 0 ;;
@@ -87,7 +87,7 @@ set -e
 
 function _modprobe()
 {
-        modprobe "$@"
+	modprobe "$@"
 
 	if [[ "$REMOTE_HOST" != "" ]]; then
 		ssh "$REMOTE_HOST" modprobe "$@"
@@ -127,15 +127,70 @@ function write_file()
 	fi
 }
 
+function check_file()
+{
+	split_remote $1
+
+	if [[ "$REMOTE" != "" ]]; then
+		ssh "$REMOTE" "[[ -e ${VPATH} ]]"
+	else
+		[[ -e ${VPATH} ]]
+	fi
+}
+
+function subdirname()
+{
+	echo $(basename $(dirname $1)) 2> /dev/null
+}
+
+function find_pidx()
+{
+	PORT=$1
+	PPATH=$2
+
+	for ((i = 0; i < 64; i++)); do
+		PEER_DIR="$PPATH/peer$i"
+
+		check_file ${PEER_DIR} || break
+
+		PEER_PORT=$(read_file "${PEER_DIR}/port")
+		if [[ ${PORT} -eq $PEER_PORT ]]; then
+			echo $i
+			return 0
+		fi
+	done
+
+	return 1
+}
+
+function port_test()
+{
+	LOC=$1
+	REM=$2
+
+	echo "Running port tests on: $(basename $LOC) / $(basename $REM)"
+
+	LOCAL_PORT=$(read_file "$LOC/port")
+	REMOTE_PORT=$(read_file "$REM/port")
+
+	LOCAL_PIDX=$(find_pidx ${REMOTE_PORT} "$LOC")
+	REMOTE_PIDX=$(find_pidx ${LOCAL_PORT} "$REM")
+
+	echo "Local port ${LOCAL_PORT} with index ${REMOTE_PIDX} on remote host"
+	echo "Peer port ${REMOTE_PORT} with index ${LOCAL_PIDX} on local host"
+
+	echo "  Passed"
+}
+
 function link_test()
 {
 	LOC=$1
 	REM=$2
 	EXP=0
 
-	echo "Running link tests on: $(basename $LOC) / $(basename $REM)"
+	echo "Running link tests on: $(subdirname $LOC) / $(subdirname $REM)"
 
-	if ! write_file "N" "$LOC/link" 2> /dev/null; then
+	if ! write_file "N" "$LOC/../link" 2> /dev/null; then
 		echo "  Unsupported"
 		return
 	fi
@@ -143,12 +198,11 @@ function link_test()
 	write_file "N" "$LOC/link_event"
 
 	if [[ $(read_file "$REM/link") != "N" ]]; then
-		echo "Expected remote link to be down in $REM/link" >&2
+		echo "Expected link to be down in $REM/link" >&2
 		exit -1
 	fi
 
-	write_file "Y" "$LOC/link"
-	write_file "Y" "$LOC/link_event"
+	write_file "Y" "$LOC/../link"
 
 	echo "  Passed"
 }
@@ -161,58 +215,136 @@ function doorbell_test()
 
 	echo "Running db tests on: $(basename $LOC) / $(basename $REM)"
 
-	write_file "c $DB_BITMASK" "$REM/db"
+	DB_VALID_MASK=$(read_file "$LOC/db_valid_mask")
+
+	write_file "c $DB_VALID_MASK" "$REM/db"
 
-	for ((i=1; i <= 8; i++)); do
-		let DB=$(read_file "$REM/db") || true
-		if [[ "$DB" != "$EXP" ]]; then
+	for ((i = 0; i < 64; i++)); do
+		DB=$(read_file "$REM/db")
+		if [[ "$DB" -ne "$EXP" ]]; then
 			echo "Doorbell doesn't match expected value $EXP " \
 			     "in $REM/db" >&2
 			exit -1
 		fi
 
-		let "MASK=1 << ($i-1)" || true
-		let "EXP=$EXP | $MASK" || true
+		let "MASK = (1 << $i) & $DB_VALID_MASK" || true
+		let "EXP = $EXP | $MASK" || true
+
 		write_file "s $MASK" "$LOC/peer_db"
 	done
 
+	write_file "c $DB_VALID_MASK" "$REM/db_mask"
+	write_file $DB_VALID_MASK "$REM/db_event"
+	write_file "s $DB_VALID_MASK" "$REM/db_mask"
+
+	write_file "c $DB_VALID_MASK" "$REM/db"
+
 	echo "  Passed"
 }
 
-function read_spad()
+function get_files_count()
 {
-       VPATH=$1
-       IDX=$2
+	NAME=$1
+	LOC=$2
+
+	split_remote $LOC
 
-       ROW=($(read_file "$VPATH" | grep -e "^$IDX"))
-       let VAL=${ROW[1]} || true
-       echo $VAL
+	if [[ "$REMOTE" == "" ]]; then
+		echo $(ls -1 "$LOC"/${NAME}* 2>/dev/null | wc -l)
+	else
+		echo $(ssh "$REMOTE" "ls -1 \"$VPATH\"/${NAME}* | \
+		       wc -l" 2> /dev/null)
+	fi
 }
 
 function scratchpad_test()
 {
 	LOC=$1
 	REM=$2
-	CNT=$(read_file "$LOC/spad" | wc -l)
 
-	echo "Running spad tests on: $(basename $LOC) / $(basename $REM)"
+	echo "Running spad tests on: $(subdirname $LOC) / $(subdirname $REM)"
+
+	CNT=$(get_files_count "spad" "$LOC")
+
+	if [[ $CNT -eq 0 ]]; then
+		echo "  Unsupported"
+		return
+	fi
 
 	for ((i = 0; i < $CNT; i++)); do
 		VAL=$RANDOM
-		write_file "$i $VAL" "$LOC/peer_spad"
-		RVAL=$(read_spad "$REM/spad" $i)
+		write_file "$VAL" "$LOC/spad$i"
+		RVAL=$(read_file "$REM/../spad$i")
 
-		if [[ "$VAL" != "$RVAL" ]]; then
-			echo "Scratchpad doesn't match expected value $VAL " \
-			     "in $REM/spad, got $RVAL" >&2
+		if [[ "$VAL" -ne "$RVAL" ]]; then
+			echo "Scratchpad $i value $RVAL doesn't match $VAL" >&2
 			exit -1
 		fi
+	done
+
+	echo "  Passed"
+}
+
+function message_test()
+{
+	LOC=$1
+	REM=$2
+
+	echo "Running msg tests on: $(subdirname $LOC) / $(subdirname $REM)"
+
+	CNT=$(get_files_count "msg" "$LOC")
 
+	if [[ $CNT -eq 0 ]]; then
+		echo "  Unsupported"
+		return
+	fi
+
+	MSG_OUTBITS_MASK=$(read_file "$LOC/../msg_inbits")
+	MSG_INBITS_MASK=$(read_file "$REM/../msg_inbits")
+
+	write_file "c $MSG_OUTBITS_MASK" "$LOC/../msg_sts"
+	write_file "c $MSG_INBITS_MASK" "$REM/../msg_sts"
+
+	for ((i = 0; i < $CNT; i++)); do
+		VAL=$RANDOM
+		write_file "$VAL" "$LOC/msg$i"
+		RVAL=$(read_file "$REM/../msg$i")
+
+		if [[ "$VAL" -ne "${RVAL%%<-*}" ]]; then
+			echo "Message $i value $RVAL doesn't match $VAL" >&2
+			exit -1
+		fi
 	done
 
 	echo "  Passed"
 }
 
+function get_number()
+{
+	KEY=$1
+
+	sed -n "s/^\(${KEY}\)[ \t]*\(0x[0-9a-fA-F]*\)\(\[p\]\)\?$/\2/p"
+}
+
+function mw_alloc()
+{
+	IDX=$1
+	LOC=$2
+	REM=$3
+
+	write_file $MW_SIZE "$LOC/mw_trans$IDX"
+
+	INB_MW=$(read_file "$LOC/mw_trans$IDX")
+	MW_ALIGNED_SIZE=$(echo "$INB_MW" | get_number "Window Size")
+	MW_DMA_ADDR=$(echo "$INB_MW" | get_number "DMA Address")
+
+	write_file "$MW_DMA_ADDR:$(($MW_ALIGNED_SIZE))" "$REM/peer_mw_trans$IDX"
+
+	if [[ $MW_SIZE -ne $MW_ALIGNED_SIZE ]]; then
+		echo "MW $IDX size aligned to $MW_ALIGNED_SIZE"
+	fi
+}
+
 function write_mw()
 {
 	split_remote $2
@@ -225,17 +357,15 @@ function write_mw()
 	fi
 }
 
-function mw_test()
+function mw_check()
 {
 	IDX=$1
 	LOC=$2
 	REM=$3
 
-	echo "Running $IDX tests on: $(basename $LOC) / $(basename $REM)"
+	write_mw "$LOC/mw$IDX"
 
-	write_mw "$LOC/$IDX"
-
-	split_remote "$LOC/$IDX"
+	split_remote "$LOC/mw$IDX"
 	if [[ "$REMOTE" == "" ]]; then
 		A=$VPATH
 	else
@@ -243,7 +373,7 @@ function mw_test()
 		ssh "$REMOTE" cat "$VPATH" > "$A"
 	fi
 
-	split_remote "$REM/peer_$IDX"
+	split_remote "$REM/peer_mw$IDX"
 	if [[ "$REMOTE" == "" ]]; then
 		B=$VPATH
 	else
@@ -251,7 +381,7 @@ function mw_test()
 		ssh "$REMOTE" cat "$VPATH" > "$B"
 	fi
 
-	cmp -n $MW_SIZE "$A" "$B"
+	cmp -n $MW_ALIGNED_SIZE "$A" "$B"
 	if [[ $? != 0 ]]; then
 		echo "Memory window $MW did not match!" >&2
 	fi
@@ -263,8 +393,39 @@ function mw_test()
 	if [[ "$B" == "/tmp/*" ]]; then
 		rm "$B"
 	fi
+}
+
+function mw_free()
+{
+	IDX=$1
+	LOC=$2
+	REM=$3
+
+	write_file "$MW_DMA_ADDR:0" "$REM/peer_mw_trans$IDX"
+
+	write_file 0 "$LOC/mw_trans$IDX"
+}
+
+function mw_test()
+{
+	LOC=$1
+	REM=$2
+
+	CNT=$(get_files_count "mw_trans" "$LOC")
+
+	for ((i = 0; i < $CNT; i++)); do
+		echo "Running mw$i tests on: $(subdirname $LOC) / " \
+		     "$(subdirname $REM)"
+
+		mw_alloc $i $LOC $REM
+
+		mw_check $i $LOC $REM
+
+		mw_free $i $LOC  $REM
+
+		echo "  Passed"
+	done
 
-	echo "  Passed"
 }
 
 function pingpong_test()
@@ -274,13 +435,13 @@ function pingpong_test()
 
 	echo "Running ping pong tests on: $(basename $LOC) / $(basename $REM)"
 
-	LOC_START=$(read_file $LOC/count)
-	REM_START=$(read_file $REM/count)
+	LOC_START=$(read_file "$LOC/count")
+	REM_START=$(read_file "$REM/count")
 
 	sleep 7
 
-	LOC_END=$(read_file $LOC/count)
-	REM_END=$(read_file $REM/count)
+	LOC_END=$(read_file "$LOC/count")
+	REM_END=$(read_file "$REM/count")
 
 	if [[ $LOC_START == $LOC_END ]] || [[ $REM_START == $REM_END ]]; then
 		echo "Ping pong counter not incrementing!" >&2
@@ -300,19 +461,19 @@ function perf_test()
 		WITH="without"
 	fi
 
-	_modprobe ntb_perf run_order=$PERF_RUN_ORDER \
+	_modprobe ntb_perf total_order=$PERF_RUN_ORDER \
 		max_mw_size=$MAX_MW_SIZE use_dma=$USE_DMA
 
 	echo "Running local perf test $WITH DMA"
-	write_file "" $LOCAL_PERF/run
+	write_file "$LOCAL_PIDX" "$LOCAL_PERF/run"
 	echo -n "  "
-	read_file $LOCAL_PERF/run
+	read_file "$LOCAL_PERF/run"
 	echo "  Passed"
 
 	echo "Running remote perf test $WITH DMA"
-	write_file "" $REMOTE_PERF/run
+	write_file "$REMOTE_PIDX" "$REMOTE_PERF/run"
 	echo -n "  "
-	read_file $REMOTE_PERF/run
+	read_file "$REMOTE_PERF/run"
 	echo "  Passed"
 
 	_modprobe -r ntb_perf
@@ -320,48 +481,44 @@ function perf_test()
 
 function ntb_tool_tests()
 {
-	LOCAL_TOOL=$DEBUGFS/ntb_tool/$LOCAL_DEV
-	REMOTE_TOOL=$REMOTE_HOST:$DEBUGFS/ntb_tool/$REMOTE_DEV
+	LOCAL_TOOL="$DEBUGFS/ntb_tool/$LOCAL_DEV"
+	REMOTE_TOOL="$REMOTE_HOST:$DEBUGFS/ntb_tool/$REMOTE_DEV"
 
 	echo "Starting ntb_tool tests..."
 
 	_modprobe ntb_tool
 
-	write_file Y $LOCAL_TOOL/link_event
-	write_file Y $REMOTE_TOOL/link_event
+	port_test "$LOCAL_TOOL" "$REMOTE_TOOL"
 
-	link_test $LOCAL_TOOL $REMOTE_TOOL
-	link_test $REMOTE_TOOL $LOCAL_TOOL
+	LOCAL_PEER_TOOL="$LOCAL_TOOL/peer$LOCAL_PIDX"
+	REMOTE_PEER_TOOL="$REMOTE_TOOL/peer$REMOTE_PIDX"
+
+	link_test "$LOCAL_PEER_TOOL" "$REMOTE_PEER_TOOL"
+	link_test "$REMOTE_PEER_TOOL" "$LOCAL_PEER_TOOL"
 
 	#Ensure the link is up on both sides before continuing
-	write_file Y $LOCAL_TOOL/link_event
-	write_file Y $REMOTE_TOOL/link_event
+	write_file "Y" "$LOCAL_PEER_TOOL/link_event"
+	write_file "Y" "$REMOTE_PEER_TOOL/link_event"
 
-	for PEER_TRANS in $(ls $LOCAL_TOOL/peer_trans*); do
-		PT=$(basename $PEER_TRANS)
-		write_file $MW_SIZE $LOCAL_TOOL/$PT
-		write_file $MW_SIZE $REMOTE_TOOL/$PT
-	done
+	doorbell_test "$LOCAL_TOOL" "$REMOTE_TOOL"
+	doorbell_test "$REMOTE_TOOL" "$LOCAL_TOOL"
 
-	doorbell_test $LOCAL_TOOL $REMOTE_TOOL
-	doorbell_test $REMOTE_TOOL $LOCAL_TOOL
-	scratchpad_test $LOCAL_TOOL $REMOTE_TOOL
-	scratchpad_test $REMOTE_TOOL $LOCAL_TOOL
+	scratchpad_test "$LOCAL_PEER_TOOL" "$REMOTE_PEER_TOOL"
+	scratchpad_test "$REMOTE_PEER_TOOL" "$LOCAL_PEER_TOOL"
 
-	for MW in $(ls $LOCAL_TOOL/mw*); do
-		MW=$(basename $MW)
+	message_test "$LOCAL_PEER_TOOL" "$REMOTE_PEER_TOOL"
+	message_test "$REMOTE_PEER_TOOL" "$LOCAL_PEER_TOOL"
 
-		mw_test $MW $LOCAL_TOOL $REMOTE_TOOL
-		mw_test $MW $REMOTE_TOOL $LOCAL_TOOL
-	done
+	mw_test "$LOCAL_PEER_TOOL" "$REMOTE_PEER_TOOL"
+	mw_test "$REMOTE_PEER_TOOL" "$LOCAL_PEER_TOOL"
 
 	_modprobe -r ntb_tool
 }
 
 function ntb_pingpong_tests()
 {
-	LOCAL_PP=$DEBUGFS/ntb_pingpong/$LOCAL_DEV
-	REMOTE_PP=$REMOTE_HOST:$DEBUGFS/ntb_pingpong/$REMOTE_DEV
+	LOCAL_PP="$DEBUGFS/ntb_pingpong/$LOCAL_DEV"
+	REMOTE_PP="$REMOTE_HOST:$DEBUGFS/ntb_pingpong/$REMOTE_DEV"
 
 	echo "Starting ntb_pingpong tests..."
 
@@ -374,8 +531,8 @@ function ntb_pingpong_tests()
 
 function ntb_perf_tests()
 {
-	LOCAL_PERF=$DEBUGFS/ntb_perf/$LOCAL_DEV
-	REMOTE_PERF=$REMOTE_HOST:$DEBUGFS/ntb_perf/$REMOTE_DEV
+	LOCAL_PERF="$DEBUGFS/ntb_perf/$LOCAL_DEV"
+	REMOTE_PERF="$REMOTE_HOST:$DEBUGFS/ntb_perf/$REMOTE_DEV"
 
 	echo "Starting ntb_perf tests..."