thunderbolt: Add initial cactus ridge NHI support

Thunderbolt hotplug is supposed to be handled by the firmware. But Apple
decided to implement thunderbolt at the operating system level. The
firmare only initializes thunderbolt devices that are present at boot
time. This driver enables hotplug of thunderbolt of non-chained
thunderbolt devices on Apple systems with a cactus ridge controller.

This first patch adds the Kconfig file as well the parts of the driver
which talk directly to the hardware (that is pci device setup, interrupt
handling and RX/TX ring management).

Signed-off-by: Andreas Noever <andreas.noever@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

7 files changed, 863 insertions, 0 deletions

diff --git a/drivers/Kconfig b/drivers/Kconfig
index 0e87a34..9b2dcc2 100644
--- a/drivers/Kconfig
+++ b/drivers/Kconfig
@@ -176,4 +176,6 @@ source "drivers/powercap/Kconfig"
 
 source "drivers/mcb/Kconfig"
 
+source "drivers/thunderbolt/Kconfig"
+
 endmenu
diff --git a/drivers/Makefile b/drivers/Makefile
index f98b50d..37b9ed4 100644
--- a/drivers/Makefile
+++ b/drivers/Makefile
@@ -158,3 +158,4 @@ obj-$(CONFIG_NTB)		+= ntb/
 obj-$(CONFIG_FMC)		+= fmc/
 obj-$(CONFIG_POWERCAP)		+= powercap/
 obj-$(CONFIG_MCB)		+= mcb/
+obj-$(CONFIG_THUNDERBOLT)	+= thunderbolt/
diff --git a/drivers/thunderbolt/Kconfig b/drivers/thunderbolt/Kconfig
new file mode 100644
index 0000000..3a25529
--- /dev/null
+++ b/drivers/thunderbolt/Kconfig
@@ -0,0 +1,12 @@
+menuconfig THUNDERBOLT
+	tristate "Thunderbolt support for Apple devices"
+	default no
+	help
+	  Cactus Ridge Thunderbolt Controller driver
+	  This driver is required if you want to hotplug Thunderbolt devices on
+	  Apple hardware.
+
+	  Device chaining is currently not supported.
+
+	  To compile this driver a module, choose M here. The module will be
+	  called thunderbolt.
diff --git a/drivers/thunderbolt/Makefile b/drivers/thunderbolt/Makefile
new file mode 100644
index 0000000..d473ab9
--- /dev/null
+++ b/drivers/thunderbolt/Makefile
@@ -0,0 +1,3 @@
+obj-${CONFIG_THUNDERBOLT} := thunderbolt.o
+thunderbolt-objs := nhi.o
+
diff --git a/drivers/thunderbolt/nhi.c b/drivers/thunderbolt/nhi.c
new file mode 100644
index 0000000..11070ff
--- /dev/null
+++ b/drivers/thunderbolt/nhi.c
@@ -0,0 +1,630 @@
+/*
+ * Thunderbolt Cactus Ridge driver - NHI driver
+ *
+ * The NHI (native host interface) is the pci device that allows us to send and
+ * receive frames from the thunderbolt bus.
+ *
+ * Copyright (c) 2014 Andreas Noever <andreas.noever@gmail.com>
+ */
+
+#include <linux/slab.h>
+#include <linux/errno.h>
+#include <linux/pci.h>
+#include <linux/interrupt.h>
+#include <linux/module.h>
+#include <linux/dmi.h>
+
+#include "nhi.h"
+#include "nhi_regs.h"
+
+#define RING_TYPE(ring) ((ring)->is_tx ? "TX ring" : "RX ring")
+
+
+static int ring_interrupt_index(struct tb_ring *ring)
+{
+	int bit = ring->hop;
+	if (!ring->is_tx)
+		bit += ring->nhi->hop_count;
+	return bit;
+}
+
+/**
+ * ring_interrupt_active() - activate/deactivate interrupts for a single ring
+ *
+ * ring->nhi->lock must be held.
+ */
+static void ring_interrupt_active(struct tb_ring *ring, bool active)
+{
+	int reg = REG_RING_INTERRUPT_BASE + ring_interrupt_index(ring) / 32;
+	int bit = ring_interrupt_index(ring) & 31;
+	int mask = 1 << bit;
+	u32 old, new;
+	old = ioread32(ring->nhi->iobase + reg);
+	if (active)
+		new = old | mask;
+	else
+		new = old & ~mask;
+
+	dev_info(&ring->nhi->pdev->dev,
+		 "%s interrupt at register %#x bit %d (%#x -> %#x)\n",
+		 active ? "enabling" : "disabling", reg, bit, old, new);
+
+	if (new == old)
+		dev_WARN(&ring->nhi->pdev->dev,
+					 "interrupt for %s %d is already %s\n",
+					 RING_TYPE(ring), ring->hop,
+					 active ? "enabled" : "disabled");
+	iowrite32(new, ring->nhi->iobase + reg);
+}
+
+/**
+ * nhi_disable_interrupts() - disable interrupts for all rings
+ *
+ * Use only during init and shutdown.
+ */
+static void nhi_disable_interrupts(struct tb_nhi *nhi)
+{
+	int i = 0;
+	/* disable interrupts */
+	for (i = 0; i < RING_INTERRUPT_REG_COUNT(nhi); i++)
+		iowrite32(0, nhi->iobase + REG_RING_INTERRUPT_BASE + 4 * i);
+
+	/* clear interrupt status bits */
+	for (i = 0; i < RING_NOTIFY_REG_COUNT(nhi); i++)
+		ioread32(nhi->iobase + REG_RING_NOTIFY_BASE + 4 * i);
+}
+
+/* ring helper methods */
+
+static void __iomem *ring_desc_base(struct tb_ring *ring)
+{
+	void __iomem *io = ring->nhi->iobase;
+	io += ring->is_tx ? REG_TX_RING_BASE : REG_RX_RING_BASE;
+	io += ring->hop * 16;
+	return io;
+}
+
+static void __iomem *ring_options_base(struct tb_ring *ring)
+{
+	void __iomem *io = ring->nhi->iobase;
+	io += ring->is_tx ? REG_TX_OPTIONS_BASE : REG_RX_OPTIONS_BASE;
+	io += ring->hop * 32;
+	return io;
+}
+
+static void ring_iowrite16desc(struct tb_ring *ring, u32 value, u32 offset)
+{
+	iowrite16(value, ring_desc_base(ring) + offset);
+}
+
+static void ring_iowrite32desc(struct tb_ring *ring, u32 value, u32 offset)
+{
+	iowrite32(value, ring_desc_base(ring) + offset);
+}
+
+static void ring_iowrite64desc(struct tb_ring *ring, u64 value, u32 offset)
+{
+	iowrite32(value, ring_desc_base(ring) + offset);
+	iowrite32(value >> 32, ring_desc_base(ring) + offset + 4);
+}
+
+static void ring_iowrite32options(struct tb_ring *ring, u32 value, u32 offset)
+{
+	iowrite32(value, ring_options_base(ring) + offset);
+}
+
+static bool ring_full(struct tb_ring *ring)
+{
+	return ((ring->head + 1) % ring->size) == ring->tail;
+}
+
+static bool ring_empty(struct tb_ring *ring)
+{
+	return ring->head == ring->tail;
+}
+
+/**
+ * ring_write_descriptors() - post frames from ring->queue to the controller
+ *
+ * ring->lock is held.
+ */
+static void ring_write_descriptors(struct tb_ring *ring)
+{
+	struct ring_frame *frame, *n;
+	struct ring_desc *descriptor;
+	list_for_each_entry_safe(frame, n, &ring->queue, list) {
+		if (ring_full(ring))
+			break;
+		list_move_tail(&frame->list, &ring->in_flight);
+		descriptor = &ring->descriptors[ring->head];
+		descriptor->phys = frame->buffer_phy;
+		descriptor->time = 0;
+		descriptor->flags = RING_DESC_POSTED | RING_DESC_INTERRUPT;
+		if (ring->is_tx) {
+			descriptor->length = frame->size;
+			descriptor->eof = frame->eof;
+			descriptor->sof = frame->sof;
+		}
+		ring->head = (ring->head + 1) % ring->size;
+		ring_iowrite16desc(ring, ring->head, ring->is_tx ? 10 : 8);
+	}
+}
+
+/**
+ * ring_work() - progress completed frames
+ *
+ * If the ring is shutting down then all frames are marked as canceled and
+ * their callbacks are invoked.
+ *
+ * Otherwise we collect all completed frame from the ring buffer, write new
+ * frame to the ring buffer and invoke the callbacks for the completed frames.
+ */
+static void ring_work(struct work_struct *work)
+{
+	struct tb_ring *ring = container_of(work, typeof(*ring), work);
+	struct ring_frame *frame;
+	bool canceled = false;
+	LIST_HEAD(done);
+	mutex_lock(&ring->lock);
+
+	if (!ring->running) {
+		/*  Move all frames to done and mark them as canceled. */
+		list_splice_tail_init(&ring->in_flight, &done);
+		list_splice_tail_init(&ring->queue, &done);
+		canceled = true;
+		goto invoke_callback;
+	}
+
+	while (!ring_empty(ring)) {
+		if (!(ring->descriptors[ring->tail].flags
+				& RING_DESC_COMPLETED))
+			break;
+		frame = list_first_entry(&ring->in_flight, typeof(*frame),
+					 list);
+		list_move_tail(&frame->list, &done);
+		if (!ring->is_tx) {
+			frame->size = ring->descriptors[ring->tail].length;
+			frame->eof = ring->descriptors[ring->tail].eof;
+			frame->sof = ring->descriptors[ring->tail].sof;
+			frame->flags = ring->descriptors[ring->tail].flags;
+			if (frame->sof != 0)
+				dev_WARN(&ring->nhi->pdev->dev,
+					 "%s %d got unexpected SOF: %#x\n",
+					 RING_TYPE(ring), ring->hop,
+					 frame->sof);
+			/*
+			 * known flags:
+			 * raw not enabled, interupt not set: 0x2=0010
+			 * raw enabled: 0xa=1010
+			 * raw not enabled: 0xb=1011
+			 * partial frame (>MAX_FRAME_SIZE): 0xe=1110
+			 */
+			if (frame->flags != 0xa)
+				dev_WARN(&ring->nhi->pdev->dev,
+					 "%s %d got unexpected flags: %#x\n",
+					 RING_TYPE(ring), ring->hop,
+					 frame->flags);
+		}
+		ring->tail = (ring->tail + 1) % ring->size;
+	}
+	ring_write_descriptors(ring);
+
+invoke_callback:
+	mutex_unlock(&ring->lock); /* allow callbacks to schedule new work */
+	while (!list_empty(&done)) {
+		frame = list_first_entry(&done, typeof(*frame), list);
+		/*
+		 * The callback may reenqueue or delete frame.
+		 * Do not hold on to it.
+		 */
+		list_del_init(&frame->list);
+		frame->callback(ring, frame, canceled);
+	}
+}
+
+int __ring_enqueue(struct tb_ring *ring, struct ring_frame *frame)
+{
+	int ret = 0;
+	mutex_lock(&ring->lock);
+	if (ring->running) {
+		list_add_tail(&frame->list, &ring->queue);
+		ring_write_descriptors(ring);
+	} else {
+		ret = -ESHUTDOWN;
+	}
+	mutex_unlock(&ring->lock);
+	return ret;
+}
+
+static struct tb_ring *ring_alloc(struct tb_nhi *nhi, u32 hop, int size,
+				  bool transmit)
+{
+	struct tb_ring *ring = NULL;
+	dev_info(&nhi->pdev->dev, "allocating %s ring %d of size %d\n",
+		 transmit ? "TX" : "RX", hop, size);
+
+	mutex_lock(&nhi->lock);
+	if (hop >= nhi->hop_count) {
+		dev_WARN(&nhi->pdev->dev, "invalid hop: %d\n", hop);
+		goto err;
+	}
+	if (transmit && nhi->tx_rings[hop]) {
+		dev_WARN(&nhi->pdev->dev, "TX hop %d already allocated\n", hop);
+		goto err;
+	} else if (!transmit && nhi->rx_rings[hop]) {
+		dev_WARN(&nhi->pdev->dev, "RX hop %d already allocated\n", hop);
+		goto err;
+	}
+	ring = kzalloc(sizeof(*ring), GFP_KERNEL);
+	if (!ring)
+		goto err;
+
+	mutex_init(&ring->lock);
+	INIT_LIST_HEAD(&ring->queue);
+	INIT_LIST_HEAD(&ring->in_flight);
+	INIT_WORK(&ring->work, ring_work);
+
+	ring->nhi = nhi;
+	ring->hop = hop;
+	ring->is_tx = transmit;
+	ring->size = size;
+	ring->head = 0;
+	ring->tail = 0;
+	ring->running = false;
+	ring->descriptors = dma_alloc_coherent(&ring->nhi->pdev->dev,
+			size * sizeof(*ring->descriptors),
+			&ring->descriptors_dma, GFP_KERNEL | __GFP_ZERO);
+	if (!ring->descriptors)
+		goto err;
+
+	if (transmit)
+		nhi->tx_rings[hop] = ring;
+	else
+		nhi->rx_rings[hop] = ring;
+	mutex_unlock(&nhi->lock);
+	return ring;
+
+err:
+	if (ring)
+		mutex_destroy(&ring->lock);
+	kfree(ring);
+	mutex_unlock(&nhi->lock);
+	return NULL;
+}
+
+struct tb_ring *ring_alloc_tx(struct tb_nhi *nhi, int hop, int size)
+{
+	return ring_alloc(nhi, hop, size, true);
+}
+
+struct tb_ring *ring_alloc_rx(struct tb_nhi *nhi, int hop, int size)
+{
+	return ring_alloc(nhi, hop, size, false);
+}
+
+/**
+ * ring_start() - enable a ring
+ *
+ * Must not be invoked in parallel with ring_stop().
+ */
+void ring_start(struct tb_ring *ring)
+{
+	mutex_lock(&ring->nhi->lock);
+	mutex_lock(&ring->lock);
+	if (ring->running) {
+		dev_WARN(&ring->nhi->pdev->dev, "ring already started\n");
+		goto err;
+	}
+	dev_info(&ring->nhi->pdev->dev, "starting %s %d\n",
+		 RING_TYPE(ring), ring->hop);
+
+	ring_iowrite64desc(ring, ring->descriptors_dma, 0);
+	if (ring->is_tx) {
+		ring_iowrite32desc(ring, ring->size, 12);
+		ring_iowrite32options(ring, 0, 4); /* time releated ? */
+		ring_iowrite32options(ring,
+				      RING_FLAG_ENABLE | RING_FLAG_RAW, 0);
+	} else {
+		ring_iowrite32desc(ring,
+				   (TB_FRAME_SIZE << 16) | ring->size, 12);
+		ring_iowrite32options(ring, 0xffffffff, 4); /* SOF EOF mask */
+		ring_iowrite32options(ring,
+				      RING_FLAG_ENABLE | RING_FLAG_RAW, 0);
+	}
+	ring_interrupt_active(ring, true);
+	ring->running = true;
+err:
+	mutex_unlock(&ring->lock);
+	mutex_unlock(&ring->nhi->lock);
+}
+
+
+/**
+ * ring_stop() - shutdown a ring
+ *
+ * Must not be invoked from a callback.
+ *
+ * This method will disable the ring. Further calls to ring_tx/ring_rx will
+ * return -ESHUTDOWN until ring_stop has been called.
+ *
+ * All enqueued frames will be canceled and their callbacks will be executed
+ * with frame->canceled set to true (on the callback thread). This method
+ * returns only after all callback invocations have finished.
+ */
+void ring_stop(struct tb_ring *ring)
+{
+	mutex_lock(&ring->nhi->lock);
+	mutex_lock(&ring->lock);
+	dev_info(&ring->nhi->pdev->dev, "stopping %s %d\n",
+		 RING_TYPE(ring), ring->hop);
+	if (!ring->running) {
+		dev_WARN(&ring->nhi->pdev->dev, "%s %d already stopped\n",
+			 RING_TYPE(ring), ring->hop);
+		goto err;
+	}
+	ring_interrupt_active(ring, false);
+
+	ring_iowrite32options(ring, 0, 0);
+	ring_iowrite64desc(ring, 0, 0);
+	ring_iowrite16desc(ring, 0, ring->is_tx ? 10 : 8);
+	ring_iowrite32desc(ring, 0, 12);
+	ring->head = 0;
+	ring->tail = 0;
+	ring->running = false;
+
+err:
+	mutex_unlock(&ring->lock);
+	mutex_unlock(&ring->nhi->lock);
+
+	/*
+	 * schedule ring->work to invoke callbacks on all remaining frames.
+	 */
+	schedule_work(&ring->work);
+	flush_work(&ring->work);
+}
+
+/*
+ * ring_free() - free ring
+ *
+ * When this method returns all invocations of ring->callback will have
+ * finished.
+ *
+ * Ring must be stopped.
+ *
+ * Must NOT be called from ring_frame->callback!
+ */
+void ring_free(struct tb_ring *ring)
+{
+	mutex_lock(&ring->nhi->lock);
+	/*
+	 * Dissociate the ring from the NHI. This also ensures that
+	 * nhi_interrupt_work cannot reschedule ring->work.
+	 */
+	if (ring->is_tx)
+		ring->nhi->tx_rings[ring->hop] = NULL;
+	else
+		ring->nhi->rx_rings[ring->hop] = NULL;
+
+	if (ring->running) {
+		dev_WARN(&ring->nhi->pdev->dev, "%s %d still running\n",
+			 RING_TYPE(ring), ring->hop);
+	}
+
+	dma_free_coherent(&ring->nhi->pdev->dev,
+			  ring->size * sizeof(*ring->descriptors),
+			  ring->descriptors, ring->descriptors_dma);
+
+	ring->descriptors = 0;
+	ring->descriptors_dma = 0;
+
+
+	dev_info(&ring->nhi->pdev->dev,
+		 "freeing %s %d\n",
+		 RING_TYPE(ring),
+		 ring->hop);
+
+	mutex_unlock(&ring->nhi->lock);
+	/**
+	 * ring->work can no longer be scheduled (it is scheduled only by
+	 * nhi_interrupt_work and ring_stop). Wait for it to finish before
+	 * freeing the ring.
+	 */
+	flush_work(&ring->work);
+	mutex_destroy(&ring->lock);
+	kfree(ring);
+}
+
+static void nhi_interrupt_work(struct work_struct *work)
+{
+	struct tb_nhi *nhi = container_of(work, typeof(*nhi), interrupt_work);
+	int value = 0; /* Suppress uninitialized usage warning. */
+	int bit;
+	int hop = -1;
+	int type = 0; /* current interrupt type 0: TX, 1: RX, 2: RX overflow */
+	struct tb_ring *ring;
+
+	mutex_lock(&nhi->lock);
+
+	/*
+	 * Starting at REG_RING_NOTIFY_BASE there are three status bitfields
+	 * (TX, RX, RX overflow). We iterate over the bits and read a new
+	 * dwords as required. The registers are cleared on read.
+	 */
+	for (bit = 0; bit < 3 * nhi->hop_count; bit++) {
+		if (bit % 32 == 0)
+			value = ioread32(nhi->iobase
+					 + REG_RING_NOTIFY_BASE
+					 + 4 * (bit / 32));
+		if (++hop == nhi->hop_count) {
+			hop = 0;
+			type++;
+		}
+		if ((value & (1 << (bit % 32))) == 0)
+			continue;
+		if (type == 2) {
+			dev_warn(&nhi->pdev->dev,
+				 "RX overflow for ring %d\n",
+				 hop);
+			continue;
+		}
+		if (type == 0)
+			ring = nhi->tx_rings[hop];
+		else
+			ring = nhi->rx_rings[hop];
+		if (ring == NULL) {
+			dev_warn(&nhi->pdev->dev,
+				 "got interrupt for inactive %s ring %d\n",
+				 type ? "RX" : "TX",
+				 hop);
+			continue;
+		}
+		/* we do not check ring->running, this is done in ring->work */
+		schedule_work(&ring->work);
+	}
+	mutex_unlock(&nhi->lock);
+}
+
+static irqreturn_t nhi_msi(int irq, void *data)
+{
+	struct tb_nhi *nhi = data;
+	schedule_work(&nhi->interrupt_work);
+	return IRQ_HANDLED;
+}
+
+static void nhi_shutdown(struct tb_nhi *nhi)
+{
+	int i;
+	dev_info(&nhi->pdev->dev, "shutdown\n");
+
+	for (i = 0; i < nhi->hop_count; i++) {
+		if (nhi->tx_rings[i])
+			dev_WARN(&nhi->pdev->dev,
+				 "TX ring %d is still active\n", i);
+		if (nhi->rx_rings[i])
+			dev_WARN(&nhi->pdev->dev,
+				 "RX ring %d is still active\n", i);
+	}
+	nhi_disable_interrupts(nhi);
+	/*
+	 * We have to release the irq before calling flush_work. Otherwise an
+	 * already executing IRQ handler could call schedule_work again.
+	 */
+	devm_free_irq(&nhi->pdev->dev, nhi->pdev->irq, nhi);
+	flush_work(&nhi->interrupt_work);
+	mutex_destroy(&nhi->lock);
+}
+
+static int nhi_probe(struct pci_dev *pdev, const struct pci_device_id *id)
+{
+	struct tb_nhi *nhi;
+	int res;
+
+	res = pcim_enable_device(pdev);
+	if (res) {
+		dev_err(&pdev->dev, "cannot enable PCI device, aborting\n");
+		return res;
+	}
+
+	res = pci_enable_msi(pdev);
+	if (res) {
+		dev_err(&pdev->dev, "cannot enable MSI, aborting\n");
+		return res;
+	}
+
+	res = pcim_iomap_regions(pdev, 1 << 0, "thunderbolt");
+	if (res) {
+		dev_err(&pdev->dev, "cannot obtain PCI resources, aborting\n");
+		return res;
+	}
+
+	nhi = devm_kzalloc(&pdev->dev, sizeof(*nhi), GFP_KERNEL);
+	if (!nhi)
+		return -ENOMEM;
+
+	nhi->pdev = pdev;
+	/* cannot fail - table is allocated bin pcim_iomap_regions */
+	nhi->iobase = pcim_iomap_table(pdev)[0];
+	nhi->hop_count = ioread32(nhi->iobase + REG_HOP_COUNT) & 0x3ff;
+	if (nhi->hop_count != 12)
+		dev_warn(&pdev->dev, "unexpected hop count: %d\n",
+			 nhi->hop_count);
+	INIT_WORK(&nhi->interrupt_work, nhi_interrupt_work);
+
+	nhi->tx_rings = devm_kzalloc(&pdev->dev,
+				     nhi->hop_count * sizeof(struct tb_ring),
+				     GFP_KERNEL);
+	nhi->rx_rings = devm_kzalloc(&pdev->dev,
+				     nhi->hop_count * sizeof(struct tb_ring),
+				     GFP_KERNEL);
+	if (!nhi->tx_rings || !nhi->rx_rings)
+		return -ENOMEM;
+
+	nhi_disable_interrupts(nhi); /* In case someone left them on. */
+	res = devm_request_irq(&pdev->dev, pdev->irq, nhi_msi,
+			       IRQF_NO_SUSPEND, /* must work during _noirq */
+			       "thunderbolt", nhi);
+	if (res) {
+		dev_err(&pdev->dev, "request_irq failed, aborting\n");
+		return res;
+	}
+
+	mutex_init(&nhi->lock);
+
+	pci_set_master(pdev);
+
+	/* magic value - clock related? */
+	iowrite32(3906250 / 10000, nhi->iobase + 0x38c00);
+
+	pci_set_drvdata(pdev, nhi);
+
+	return 0;
+}
+
+static void nhi_remove(struct pci_dev *pdev)
+{
+	struct tb_nhi *nhi = pci_get_drvdata(pdev);
+	nhi_shutdown(nhi);
+}
+
+struct pci_device_id nhi_ids[] = {
+	/*
+	 * We have to specify class, the TB bridges use the same device and
+	 * vendor (sub)id.
+	 */
+	{
+		.class = PCI_CLASS_SYSTEM_OTHER << 8, .class_mask = ~0,
+		.vendor = PCI_VENDOR_ID_INTEL, .device = 0x1547,
+		.subvendor = 0x2222, .subdevice = 0x1111,
+	},
+	{
+		.class = PCI_CLASS_SYSTEM_OTHER << 8, .class_mask = ~0,
+		.vendor = PCI_VENDOR_ID_INTEL, .device = 0x156c,
+		.subvendor = 0x2222, .subdevice = 0x1111,
+	},
+	{ 0,}
+};
+
+MODULE_DEVICE_TABLE(pci, nhi_ids);
+MODULE_LICENSE("GPL");
+
+static struct pci_driver nhi_driver = {
+	.name = "thunderbolt",
+	.id_table = nhi_ids,
+	.probe = nhi_probe,
+	.remove = nhi_remove,
+};
+
+static int __init nhi_init(void)
+{
+	if (!dmi_match(DMI_BOARD_VENDOR, "Apple Inc."))
+		return -ENOSYS;
+	return pci_register_driver(&nhi_driver);
+}
+
+static void __exit nhi_unload(void)
+{
+	pci_unregister_driver(&nhi_driver);
+}
+
+module_init(nhi_init);
+module_exit(nhi_unload);
diff --git a/drivers/thunderbolt/nhi.h b/drivers/thunderbolt/nhi.h
new file mode 100644
index 0000000..3172429
--- /dev/null
+++ b/drivers/thunderbolt/nhi.h
@@ -0,0 +1,114 @@
+/*
+ * Thunderbolt Cactus Ridge driver - NHI driver
+ *
+ * Copyright (c) 2014 Andreas Noever <andreas.noever@gmail.com>
+ */
+
+#ifndef DSL3510_H_
+#define DSL3510_H_
+
+#include <linux/mutex.h>
+#include <linux/workqueue.h>
+
+/**
+ * struct tb_nhi - thunderbolt native host interface
+ */
+struct tb_nhi {
+	struct mutex lock; /*
+			    * Must be held during ring creation/destruction.
+			    * Is acquired by interrupt_work when dispatching
+			    * interrupts to individual rings.
+			    **/
+	struct pci_dev *pdev;
+	void __iomem *iobase;
+	struct tb_ring **tx_rings;
+	struct tb_ring **rx_rings;
+	struct work_struct interrupt_work;
+	u32 hop_count; /* Number of rings (end point hops) supported by NHI. */
+};
+
+/**
+ * struct tb_ring - thunderbolt TX or RX ring associated with a NHI
+ */
+struct tb_ring {
+	struct mutex lock; /* must be acquired after nhi->lock */
+	struct tb_nhi *nhi;
+	int size;
+	int hop;
+	int head; /* write next descriptor here */
+	int tail; /* complete next descriptor here */
+	struct ring_desc *descriptors;
+	dma_addr_t descriptors_dma;
+	struct list_head queue;
+	struct list_head in_flight;
+	struct work_struct work;
+	bool is_tx:1; /* rx otherwise */
+	bool running:1;
+};
+
+struct ring_frame;
+typedef void (*ring_cb)(struct tb_ring*, struct ring_frame*, bool canceled);
+
+/**
+ * struct ring_frame - for use with ring_rx/ring_tx
+ */
+struct ring_frame {
+	dma_addr_t buffer_phy;
+	ring_cb callback;
+	struct list_head list;
+	u32 size:12; /* TX: in, RX: out*/
+	u32 flags:12; /* RX: out */
+	u32 eof:4; /* TX:in, RX: out */
+	u32 sof:4; /* TX:in, RX: out */
+};
+
+#define TB_FRAME_SIZE 0x100    /* minimum size for ring_rx */
+
+struct tb_ring *ring_alloc_tx(struct tb_nhi *nhi, int hop, int size);
+struct tb_ring *ring_alloc_rx(struct tb_nhi *nhi, int hop, int size);
+void ring_start(struct tb_ring *ring);
+void ring_stop(struct tb_ring *ring);
+void ring_free(struct tb_ring *ring);
+
+int __ring_enqueue(struct tb_ring *ring, struct ring_frame *frame);
+
+/**
+ * ring_rx() - enqueue a frame on an RX ring
+ *
+ * frame->buffer, frame->buffer_phy and frame->callback have to be set. The
+ * buffer must contain at least TB_FRAME_SIZE bytes.
+ *
+ * frame->callback will be invoked with frame->size, frame->flags, frame->eof,
+ * frame->sof set once the frame has been received.
+ *
+ * If ring_stop is called after the packet has been enqueued frame->callback
+ * will be called with canceled set to true.
+ *
+ * Return: Returns ESHUTDOWN if ring_stop has been called. Zero otherwise.
+ */
+static inline int ring_rx(struct tb_ring *ring, struct ring_frame *frame)
+{
+	WARN_ON(ring->is_tx);
+	return __ring_enqueue(ring, frame);
+}
+
+/**
+ * ring_tx() - enqueue a frame on an TX ring
+ *
+ * frame->buffer, frame->buffer_phy, frame->callback, frame->size, frame->eof
+ * and frame->sof have to be set.
+ *
+ * frame->callback will be invoked with once the frame has been transmitted.
+ *
+ * If ring_stop is called after the packet has been enqueued frame->callback
+ * will be called with canceled set to true.
+ *
+ * Return: Returns ESHUTDOWN if ring_stop has been called. Zero otherwise.
+ */
+static inline int ring_tx(struct tb_ring *ring, struct ring_frame *frame)
+{
+	WARN_ON(!ring->is_tx);
+	return __ring_enqueue(ring, frame);
+}
+
+#endif
diff --git a/drivers/thunderbolt/nhi_regs.h b/drivers/thunderbolt/nhi_regs.h
new file mode 100644
index 0000000..86b996c
--- /dev/null
+++ b/drivers/thunderbolt/nhi_regs.h
@@ -0,0 +1,101 @@
+/*
+ * Thunderbolt Cactus Ridge driver - NHI registers
+ *
+ * Copyright (c) 2014 Andreas Noever <andreas.noever@gmail.com>
+ */
+
+#ifndef DSL3510_REGS_H_
+#define DSL3510_REGS_H_
+
+#include <linux/types.h>
+
+enum ring_flags {
+	RING_FLAG_ISOCH_ENABLE = 1 << 27, /* TX only? */
+	RING_FLAG_E2E_FLOW_CONTROL = 1 << 28,
+	RING_FLAG_PCI_NO_SNOOP = 1 << 29,
+	RING_FLAG_RAW = 1 << 30, /* ignore EOF/SOF mask, include checksum */
+	RING_FLAG_ENABLE = 1 << 31,
+};
+
+enum ring_desc_flags {
+	RING_DESC_ISOCH = 0x1, /* TX only? */
+	RING_DESC_COMPLETED = 0x2, /* set by NHI */
+	RING_DESC_POSTED = 0x4, /* always set this */
+	RING_DESC_INTERRUPT = 0x8, /* request an interrupt on completion */
+};
+
+/**
+ * struct ring_desc - TX/RX ring entry
+ *
+ * For TX set length/eof/sof.
+ * For RX length/eof/sof are set by the NHI.
+ */
+struct ring_desc {
+	u64 phys;
+	u32 length:12;
+	u32 eof:4;
+	u32 sof:4;
+	enum ring_desc_flags flags:12;
+	u32 time; /* write zero */
+} __packed;
+
+/* NHI registers in bar 0 */
+
+/*
+ * 16 bytes per entry, one entry for every hop (REG_HOP_COUNT)
+ * 00: physical pointer to an array of struct ring_desc
+ * 08: ring tail (set by NHI)
+ * 10: ring head (index of first non posted descriptor)
+ * 12: descriptor count
+ */
+#define REG_TX_RING_BASE	0x00000
+
+/*
+ * 16 bytes per entry, one entry for every hop (REG_HOP_COUNT)
+ * 00: physical pointer to an array of struct ring_desc
+ * 08: ring head (index of first not posted descriptor)
+ * 10: ring tail (set by NHI)
+ * 12: descriptor count
+ * 14: max frame sizes (anything larger than 0x100 has no effect)
+ */
+#define REG_RX_RING_BASE	0x08000
+
+/*
+ * 32 bytes per entry, one entry for every hop (REG_HOP_COUNT)
+ * 00: enum_ring_flags
+ * 04: isoch time stamp ?? (write 0)
+ * ..: unknown
+ */
+#define REG_TX_OPTIONS_BASE	0x19800
+
+/*
+ * 32 bytes per entry, one entry for every hop (REG_HOP_COUNT)
+ * 00: enum ring_flags
+ *     If RING_FLAG_E2E_FLOW_CONTROL is set then bits 13-23 must be set to
+ *     the corresponding TX hop id.
+ * 04: EOF/SOF mask (ignored for RING_FLAG_RAW rings)
+ * ..: unknown
+ */
+#define REG_RX_OPTIONS_BASE	0x29800
+
+/*
+ * three bitfields: tx, rx, rx overflow
+ * Every bitfield contains one bit for every hop (REG_HOP_COUNT). Registers are
+ * cleared on read. New interrupts are fired only after ALL registers have been
+ * read (even those containing only disabled rings).
+ */
+#define REG_RING_NOTIFY_BASE	0x37800
+#define RING_NOTIFY_REG_COUNT(nhi) ((31 + 3 * nhi->hop_count) / 32)
+
+/*
+ * two bitfields: rx, tx
+ * Both bitfields contains one bit for every hop (REG_HOP_COUNT). To
+ * enable/disable interrupts set/clear the corresponding bits.
+ */
+#define REG_RING_INTERRUPT_BASE	0x38200
+#define RING_INTERRUPT_REG_COUNT(nhi) ((31 + 2 * nhi->hop_count) / 32)
+
+/* The last 11 bits contain the number of hops supported by the NHI port. */
+#define REG_HOP_COUNT		0x39640
+
+#endif