1 // SPDX-License-Identifier: GPL-2.0-only
3 * Thunderbolt driver - NHI driver
5 * The NHI (native host interface) is the pci device that allows us to send and
6 * receive frames from the thunderbolt bus.
8 * Copyright (c) 2014 Andreas Noever <andreas.noever@gmail.com>
9 * Copyright (C) 2018, Intel Corporation
12 #include <linux/pm_runtime.h>
13 #include <linux/slab.h>
14 #include <linux/errno.h>
15 #include <linux/pci.h>
16 #include <linux/interrupt.h>
17 #include <linux/module.h>
18 #include <linux/delay.h>
19 #include <linux/property.h>
20 #include <linux/platform_data/x86/apple.h>
26 #define RING_TYPE(ring) ((ring)->is_tx ? "TX ring" : "RX ring")
28 #define RING_FIRST_USABLE_HOPID 1
31 * Minimal number of vectors when we use MSI-X. Two for control channel
32 * Rx/Tx and the rest four are for cross domain DMA paths.
34 #define MSIX_MIN_VECS 6
35 #define MSIX_MAX_VECS 16
37 #define NHI_MAILBOX_TIMEOUT 500 /* ms */
39 static int ring_interrupt_index(struct tb_ring *ring)
43 bit += ring->nhi->hop_count;
48 * ring_interrupt_active() - activate/deactivate interrupts for a single ring
50 * ring->nhi->lock must be held.
52 static void ring_interrupt_active(struct tb_ring *ring, bool active)
54 int reg = REG_RING_INTERRUPT_BASE +
55 ring_interrupt_index(ring) / 32 * 4;
56 int bit = ring_interrupt_index(ring) & 31;
61 u32 step, shift, ivr, misc;
62 void __iomem *ivr_base;
68 index = ring->hop + ring->nhi->hop_count;
71 * Ask the hardware to clear interrupt status bits automatically
72 * since we already know which interrupt was triggered.
74 misc = ioread32(ring->nhi->iobase + REG_DMA_MISC);
75 if (!(misc & REG_DMA_MISC_INT_AUTO_CLEAR)) {
76 misc |= REG_DMA_MISC_INT_AUTO_CLEAR;
77 iowrite32(misc, ring->nhi->iobase + REG_DMA_MISC);
80 ivr_base = ring->nhi->iobase + REG_INT_VEC_ALLOC_BASE;
81 step = index / REG_INT_VEC_ALLOC_REGS * REG_INT_VEC_ALLOC_BITS;
82 shift = index % REG_INT_VEC_ALLOC_REGS * REG_INT_VEC_ALLOC_BITS;
83 ivr = ioread32(ivr_base + step);
84 ivr &= ~(REG_INT_VEC_ALLOC_MASK << shift);
86 ivr |= ring->vector << shift;
87 iowrite32(ivr, ivr_base + step);
90 old = ioread32(ring->nhi->iobase + reg);
96 dev_dbg(&ring->nhi->pdev->dev,
97 "%s interrupt at register %#x bit %d (%#x -> %#x)\n",
98 active ? "enabling" : "disabling", reg, bit, old, new);
101 dev_WARN(&ring->nhi->pdev->dev,
102 "interrupt for %s %d is already %s\n",
103 RING_TYPE(ring), ring->hop,
104 active ? "enabled" : "disabled");
105 iowrite32(new, ring->nhi->iobase + reg);
109 * nhi_disable_interrupts() - disable interrupts for all rings
111 * Use only during init and shutdown.
113 static void nhi_disable_interrupts(struct tb_nhi *nhi)
116 /* disable interrupts */
117 for (i = 0; i < RING_INTERRUPT_REG_COUNT(nhi); i++)
118 iowrite32(0, nhi->iobase + REG_RING_INTERRUPT_BASE + 4 * i);
120 /* clear interrupt status bits */
121 for (i = 0; i < RING_NOTIFY_REG_COUNT(nhi); i++)
122 ioread32(nhi->iobase + REG_RING_NOTIFY_BASE + 4 * i);
125 /* ring helper methods */
127 static void __iomem *ring_desc_base(struct tb_ring *ring)
129 void __iomem *io = ring->nhi->iobase;
130 io += ring->is_tx ? REG_TX_RING_BASE : REG_RX_RING_BASE;
131 io += ring->hop * 16;
135 static void __iomem *ring_options_base(struct tb_ring *ring)
137 void __iomem *io = ring->nhi->iobase;
138 io += ring->is_tx ? REG_TX_OPTIONS_BASE : REG_RX_OPTIONS_BASE;
139 io += ring->hop * 32;
143 static void ring_iowrite_cons(struct tb_ring *ring, u16 cons)
146 * The other 16-bits in the register is read-only and writes to it
147 * are ignored by the hardware so we can save one ioread32() by
148 * filling the read-only bits with zeroes.
150 iowrite32(cons, ring_desc_base(ring) + 8);
153 static void ring_iowrite_prod(struct tb_ring *ring, u16 prod)
155 /* See ring_iowrite_cons() above for explanation */
156 iowrite32(prod << 16, ring_desc_base(ring) + 8);
159 static void ring_iowrite32desc(struct tb_ring *ring, u32 value, u32 offset)
161 iowrite32(value, ring_desc_base(ring) + offset);
164 static void ring_iowrite64desc(struct tb_ring *ring, u64 value, u32 offset)
166 iowrite32(value, ring_desc_base(ring) + offset);
167 iowrite32(value >> 32, ring_desc_base(ring) + offset + 4);
170 static void ring_iowrite32options(struct tb_ring *ring, u32 value, u32 offset)
172 iowrite32(value, ring_options_base(ring) + offset);
175 static bool ring_full(struct tb_ring *ring)
177 return ((ring->head + 1) % ring->size) == ring->tail;
180 static bool ring_empty(struct tb_ring *ring)
182 return ring->head == ring->tail;
186 * ring_write_descriptors() - post frames from ring->queue to the controller
188 * ring->lock is held.
190 static void ring_write_descriptors(struct tb_ring *ring)
192 struct ring_frame *frame, *n;
193 struct ring_desc *descriptor;
194 list_for_each_entry_safe(frame, n, &ring->queue, list) {
197 list_move_tail(&frame->list, &ring->in_flight);
198 descriptor = &ring->descriptors[ring->head];
199 descriptor->phys = frame->buffer_phy;
200 descriptor->time = 0;
201 descriptor->flags = RING_DESC_POSTED | RING_DESC_INTERRUPT;
203 descriptor->length = frame->size;
204 descriptor->eof = frame->eof;
205 descriptor->sof = frame->sof;
207 ring->head = (ring->head + 1) % ring->size;
209 ring_iowrite_prod(ring, ring->head);
211 ring_iowrite_cons(ring, ring->head);
216 * ring_work() - progress completed frames
218 * If the ring is shutting down then all frames are marked as canceled and
219 * their callbacks are invoked.
221 * Otherwise we collect all completed frame from the ring buffer, write new
222 * frame to the ring buffer and invoke the callbacks for the completed frames.
224 static void ring_work(struct work_struct *work)
226 struct tb_ring *ring = container_of(work, typeof(*ring), work);
227 struct ring_frame *frame;
228 bool canceled = false;
232 spin_lock_irqsave(&ring->lock, flags);
234 if (!ring->running) {
235 /* Move all frames to done and mark them as canceled. */
236 list_splice_tail_init(&ring->in_flight, &done);
237 list_splice_tail_init(&ring->queue, &done);
239 goto invoke_callback;
242 while (!ring_empty(ring)) {
243 if (!(ring->descriptors[ring->tail].flags
244 & RING_DESC_COMPLETED))
246 frame = list_first_entry(&ring->in_flight, typeof(*frame),
248 list_move_tail(&frame->list, &done);
250 frame->size = ring->descriptors[ring->tail].length;
251 frame->eof = ring->descriptors[ring->tail].eof;
252 frame->sof = ring->descriptors[ring->tail].sof;
253 frame->flags = ring->descriptors[ring->tail].flags;
255 ring->tail = (ring->tail + 1) % ring->size;
257 ring_write_descriptors(ring);
260 /* allow callbacks to schedule new work */
261 spin_unlock_irqrestore(&ring->lock, flags);
262 while (!list_empty(&done)) {
263 frame = list_first_entry(&done, typeof(*frame), list);
265 * The callback may reenqueue or delete frame.
266 * Do not hold on to it.
268 list_del_init(&frame->list);
270 frame->callback(ring, frame, canceled);
274 int __tb_ring_enqueue(struct tb_ring *ring, struct ring_frame *frame)
279 spin_lock_irqsave(&ring->lock, flags);
281 list_add_tail(&frame->list, &ring->queue);
282 ring_write_descriptors(ring);
286 spin_unlock_irqrestore(&ring->lock, flags);
289 EXPORT_SYMBOL_GPL(__tb_ring_enqueue);
292 * tb_ring_poll() - Poll one completed frame from the ring
293 * @ring: Ring to poll
295 * This function can be called when @start_poll callback of the @ring
296 * has been called. It will read one completed frame from the ring and
297 * return it to the caller. Returns %NULL if there is no more completed
300 struct ring_frame *tb_ring_poll(struct tb_ring *ring)
302 struct ring_frame *frame = NULL;
305 spin_lock_irqsave(&ring->lock, flags);
308 if (ring_empty(ring))
311 if (ring->descriptors[ring->tail].flags & RING_DESC_COMPLETED) {
312 frame = list_first_entry(&ring->in_flight, typeof(*frame),
314 list_del_init(&frame->list);
317 frame->size = ring->descriptors[ring->tail].length;
318 frame->eof = ring->descriptors[ring->tail].eof;
319 frame->sof = ring->descriptors[ring->tail].sof;
320 frame->flags = ring->descriptors[ring->tail].flags;
323 ring->tail = (ring->tail + 1) % ring->size;
327 spin_unlock_irqrestore(&ring->lock, flags);
330 EXPORT_SYMBOL_GPL(tb_ring_poll);
332 static void __ring_interrupt_mask(struct tb_ring *ring, bool mask)
334 int idx = ring_interrupt_index(ring);
335 int reg = REG_RING_INTERRUPT_BASE + idx / 32 * 4;
339 val = ioread32(ring->nhi->iobase + reg);
344 iowrite32(val, ring->nhi->iobase + reg);
347 /* Both @nhi->lock and @ring->lock should be held */
348 static void __ring_interrupt(struct tb_ring *ring)
353 if (ring->start_poll) {
354 __ring_interrupt_mask(ring, true);
355 ring->start_poll(ring->poll_data);
357 schedule_work(&ring->work);
362 * tb_ring_poll_complete() - Re-start interrupt for the ring
363 * @ring: Ring to re-start the interrupt
365 * This will re-start (unmask) the ring interrupt once the user is done
368 void tb_ring_poll_complete(struct tb_ring *ring)
372 spin_lock_irqsave(&ring->nhi->lock, flags);
373 spin_lock(&ring->lock);
374 if (ring->start_poll)
375 __ring_interrupt_mask(ring, false);
376 spin_unlock(&ring->lock);
377 spin_unlock_irqrestore(&ring->nhi->lock, flags);
379 EXPORT_SYMBOL_GPL(tb_ring_poll_complete);
381 static irqreturn_t ring_msix(int irq, void *data)
383 struct tb_ring *ring = data;
385 spin_lock(&ring->nhi->lock);
386 spin_lock(&ring->lock);
387 __ring_interrupt(ring);
388 spin_unlock(&ring->lock);
389 spin_unlock(&ring->nhi->lock);
394 static int ring_request_msix(struct tb_ring *ring, bool no_suspend)
396 struct tb_nhi *nhi = ring->nhi;
397 unsigned long irqflags;
400 if (!nhi->pdev->msix_enabled)
403 ret = ida_simple_get(&nhi->msix_ida, 0, MSIX_MAX_VECS, GFP_KERNEL);
409 ring->irq = pci_irq_vector(ring->nhi->pdev, ring->vector);
413 irqflags = no_suspend ? IRQF_NO_SUSPEND : 0;
414 return request_irq(ring->irq, ring_msix, irqflags, "thunderbolt", ring);
417 static void ring_release_msix(struct tb_ring *ring)
422 free_irq(ring->irq, ring);
423 ida_simple_remove(&ring->nhi->msix_ida, ring->vector);
428 static int nhi_alloc_hop(struct tb_nhi *nhi, struct tb_ring *ring)
432 spin_lock_irq(&nhi->lock);
438 * Automatically allocate HopID from the non-reserved
439 * range 1 .. hop_count - 1.
441 for (i = RING_FIRST_USABLE_HOPID; i < nhi->hop_count; i++) {
443 if (!nhi->tx_rings[i]) {
448 if (!nhi->rx_rings[i]) {
456 if (ring->hop < 0 || ring->hop >= nhi->hop_count) {
457 dev_warn(&nhi->pdev->dev, "invalid hop: %d\n", ring->hop);
461 if (ring->is_tx && nhi->tx_rings[ring->hop]) {
462 dev_warn(&nhi->pdev->dev, "TX hop %d already allocated\n",
466 } else if (!ring->is_tx && nhi->rx_rings[ring->hop]) {
467 dev_warn(&nhi->pdev->dev, "RX hop %d already allocated\n",
474 nhi->tx_rings[ring->hop] = ring;
476 nhi->rx_rings[ring->hop] = ring;
479 spin_unlock_irq(&nhi->lock);
484 static struct tb_ring *tb_ring_alloc(struct tb_nhi *nhi, u32 hop, int size,
485 bool transmit, unsigned int flags,
486 u16 sof_mask, u16 eof_mask,
487 void (*start_poll)(void *),
490 struct tb_ring *ring = NULL;
492 dev_dbg(&nhi->pdev->dev, "allocating %s ring %d of size %d\n",
493 transmit ? "TX" : "RX", hop, size);
495 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
499 spin_lock_init(&ring->lock);
500 INIT_LIST_HEAD(&ring->queue);
501 INIT_LIST_HEAD(&ring->in_flight);
502 INIT_WORK(&ring->work, ring_work);
506 ring->is_tx = transmit;
509 ring->sof_mask = sof_mask;
510 ring->eof_mask = eof_mask;
513 ring->running = false;
514 ring->start_poll = start_poll;
515 ring->poll_data = poll_data;
517 ring->descriptors = dma_alloc_coherent(&ring->nhi->pdev->dev,
518 size * sizeof(*ring->descriptors),
519 &ring->descriptors_dma, GFP_KERNEL | __GFP_ZERO);
520 if (!ring->descriptors)
523 if (ring_request_msix(ring, flags & RING_FLAG_NO_SUSPEND))
526 if (nhi_alloc_hop(nhi, ring))
527 goto err_release_msix;
532 ring_release_msix(ring);
534 dma_free_coherent(&ring->nhi->pdev->dev,
535 ring->size * sizeof(*ring->descriptors),
536 ring->descriptors, ring->descriptors_dma);
544 * tb_ring_alloc_tx() - Allocate DMA ring for transmit
545 * @nhi: Pointer to the NHI the ring is to be allocated
546 * @hop: HopID (ring) to allocate
547 * @size: Number of entries in the ring
548 * @flags: Flags for the ring
550 struct tb_ring *tb_ring_alloc_tx(struct tb_nhi *nhi, int hop, int size,
553 return tb_ring_alloc(nhi, hop, size, true, flags, 0, 0, NULL, NULL);
555 EXPORT_SYMBOL_GPL(tb_ring_alloc_tx);
558 * tb_ring_alloc_rx() - Allocate DMA ring for receive
559 * @nhi: Pointer to the NHI the ring is to be allocated
560 * @hop: HopID (ring) to allocate. Pass %-1 for automatic allocation.
561 * @size: Number of entries in the ring
562 * @flags: Flags for the ring
563 * @sof_mask: Mask of PDF values that start a frame
564 * @eof_mask: Mask of PDF values that end a frame
565 * @start_poll: If not %NULL the ring will call this function when an
566 * interrupt is triggered and masked, instead of callback
568 * @poll_data: Optional data passed to @start_poll
570 struct tb_ring *tb_ring_alloc_rx(struct tb_nhi *nhi, int hop, int size,
571 unsigned int flags, u16 sof_mask, u16 eof_mask,
572 void (*start_poll)(void *), void *poll_data)
574 return tb_ring_alloc(nhi, hop, size, false, flags, sof_mask, eof_mask,
575 start_poll, poll_data);
577 EXPORT_SYMBOL_GPL(tb_ring_alloc_rx);
580 * tb_ring_start() - enable a ring
582 * Must not be invoked in parallel with tb_ring_stop().
584 void tb_ring_start(struct tb_ring *ring)
589 spin_lock_irq(&ring->nhi->lock);
590 spin_lock(&ring->lock);
591 if (ring->nhi->going_away)
594 dev_WARN(&ring->nhi->pdev->dev, "ring already started\n");
597 dev_dbg(&ring->nhi->pdev->dev, "starting %s %d\n",
598 RING_TYPE(ring), ring->hop);
600 if (ring->flags & RING_FLAG_FRAME) {
603 flags = RING_FLAG_ENABLE;
605 frame_size = TB_FRAME_SIZE;
606 flags = RING_FLAG_ENABLE | RING_FLAG_RAW;
609 ring_iowrite64desc(ring, ring->descriptors_dma, 0);
611 ring_iowrite32desc(ring, ring->size, 12);
612 ring_iowrite32options(ring, 0, 4); /* time releated ? */
613 ring_iowrite32options(ring, flags, 0);
615 u32 sof_eof_mask = ring->sof_mask << 16 | ring->eof_mask;
617 ring_iowrite32desc(ring, (frame_size << 16) | ring->size, 12);
618 ring_iowrite32options(ring, sof_eof_mask, 4);
619 ring_iowrite32options(ring, flags, 0);
621 ring_interrupt_active(ring, true);
622 ring->running = true;
624 spin_unlock(&ring->lock);
625 spin_unlock_irq(&ring->nhi->lock);
627 EXPORT_SYMBOL_GPL(tb_ring_start);
630 * tb_ring_stop() - shutdown a ring
632 * Must not be invoked from a callback.
634 * This method will disable the ring. Further calls to
635 * tb_ring_tx/tb_ring_rx will return -ESHUTDOWN until ring_stop has been
638 * All enqueued frames will be canceled and their callbacks will be executed
639 * with frame->canceled set to true (on the callback thread). This method
640 * returns only after all callback invocations have finished.
642 void tb_ring_stop(struct tb_ring *ring)
644 spin_lock_irq(&ring->nhi->lock);
645 spin_lock(&ring->lock);
646 dev_dbg(&ring->nhi->pdev->dev, "stopping %s %d\n",
647 RING_TYPE(ring), ring->hop);
648 if (ring->nhi->going_away)
650 if (!ring->running) {
651 dev_WARN(&ring->nhi->pdev->dev, "%s %d already stopped\n",
652 RING_TYPE(ring), ring->hop);
655 ring_interrupt_active(ring, false);
657 ring_iowrite32options(ring, 0, 0);
658 ring_iowrite64desc(ring, 0, 0);
659 ring_iowrite32desc(ring, 0, 8);
660 ring_iowrite32desc(ring, 0, 12);
663 ring->running = false;
666 spin_unlock(&ring->lock);
667 spin_unlock_irq(&ring->nhi->lock);
670 * schedule ring->work to invoke callbacks on all remaining frames.
672 schedule_work(&ring->work);
673 flush_work(&ring->work);
675 EXPORT_SYMBOL_GPL(tb_ring_stop);
678 * tb_ring_free() - free ring
680 * When this method returns all invocations of ring->callback will have
683 * Ring must be stopped.
685 * Must NOT be called from ring_frame->callback!
687 void tb_ring_free(struct tb_ring *ring)
689 spin_lock_irq(&ring->nhi->lock);
691 * Dissociate the ring from the NHI. This also ensures that
692 * nhi_interrupt_work cannot reschedule ring->work.
695 ring->nhi->tx_rings[ring->hop] = NULL;
697 ring->nhi->rx_rings[ring->hop] = NULL;
700 dev_WARN(&ring->nhi->pdev->dev, "%s %d still running\n",
701 RING_TYPE(ring), ring->hop);
703 spin_unlock_irq(&ring->nhi->lock);
705 ring_release_msix(ring);
707 dma_free_coherent(&ring->nhi->pdev->dev,
708 ring->size * sizeof(*ring->descriptors),
709 ring->descriptors, ring->descriptors_dma);
711 ring->descriptors = NULL;
712 ring->descriptors_dma = 0;
715 dev_dbg(&ring->nhi->pdev->dev, "freeing %s %d\n", RING_TYPE(ring),
719 * ring->work can no longer be scheduled (it is scheduled only
720 * by nhi_interrupt_work, ring_stop and ring_msix). Wait for it
721 * to finish before freeing the ring.
723 flush_work(&ring->work);
726 EXPORT_SYMBOL_GPL(tb_ring_free);
729 * nhi_mailbox_cmd() - Send a command through NHI mailbox
730 * @nhi: Pointer to the NHI structure
731 * @cmd: Command to send
732 * @data: Data to be send with the command
734 * Sends mailbox command to the firmware running on NHI. Returns %0 in
735 * case of success and negative errno in case of failure.
737 int nhi_mailbox_cmd(struct tb_nhi *nhi, enum nhi_mailbox_cmd cmd, u32 data)
742 iowrite32(data, nhi->iobase + REG_INMAIL_DATA);
744 val = ioread32(nhi->iobase + REG_INMAIL_CMD);
745 val &= ~(REG_INMAIL_CMD_MASK | REG_INMAIL_ERROR);
746 val |= REG_INMAIL_OP_REQUEST | cmd;
747 iowrite32(val, nhi->iobase + REG_INMAIL_CMD);
749 timeout = ktime_add_ms(ktime_get(), NHI_MAILBOX_TIMEOUT);
751 val = ioread32(nhi->iobase + REG_INMAIL_CMD);
752 if (!(val & REG_INMAIL_OP_REQUEST))
754 usleep_range(10, 20);
755 } while (ktime_before(ktime_get(), timeout));
757 if (val & REG_INMAIL_OP_REQUEST)
759 if (val & REG_INMAIL_ERROR)
766 * nhi_mailbox_mode() - Return current firmware operation mode
767 * @nhi: Pointer to the NHI structure
769 * The function reads current firmware operation mode using NHI mailbox
770 * registers and returns it to the caller.
772 enum nhi_fw_mode nhi_mailbox_mode(struct tb_nhi *nhi)
776 val = ioread32(nhi->iobase + REG_OUTMAIL_CMD);
777 val &= REG_OUTMAIL_CMD_OPMODE_MASK;
778 val >>= REG_OUTMAIL_CMD_OPMODE_SHIFT;
780 return (enum nhi_fw_mode)val;
783 static void nhi_interrupt_work(struct work_struct *work)
785 struct tb_nhi *nhi = container_of(work, typeof(*nhi), interrupt_work);
786 int value = 0; /* Suppress uninitialized usage warning. */
789 int type = 0; /* current interrupt type 0: TX, 1: RX, 2: RX overflow */
790 struct tb_ring *ring;
792 spin_lock_irq(&nhi->lock);
795 * Starting at REG_RING_NOTIFY_BASE there are three status bitfields
796 * (TX, RX, RX overflow). We iterate over the bits and read a new
797 * dwords as required. The registers are cleared on read.
799 for (bit = 0; bit < 3 * nhi->hop_count; bit++) {
801 value = ioread32(nhi->iobase
802 + REG_RING_NOTIFY_BASE
804 if (++hop == nhi->hop_count) {
808 if ((value & (1 << (bit % 32))) == 0)
811 dev_warn(&nhi->pdev->dev,
812 "RX overflow for ring %d\n",
817 ring = nhi->tx_rings[hop];
819 ring = nhi->rx_rings[hop];
821 dev_warn(&nhi->pdev->dev,
822 "got interrupt for inactive %s ring %d\n",
828 spin_lock(&ring->lock);
829 __ring_interrupt(ring);
830 spin_unlock(&ring->lock);
832 spin_unlock_irq(&nhi->lock);
835 static irqreturn_t nhi_msi(int irq, void *data)
837 struct tb_nhi *nhi = data;
838 schedule_work(&nhi->interrupt_work);
842 static int __nhi_suspend_noirq(struct device *dev, bool wakeup)
844 struct pci_dev *pdev = to_pci_dev(dev);
845 struct tb *tb = pci_get_drvdata(pdev);
846 struct tb_nhi *nhi = tb->nhi;
849 ret = tb_domain_suspend_noirq(tb);
853 if (nhi->ops && nhi->ops->suspend_noirq) {
854 ret = nhi->ops->suspend_noirq(tb->nhi, wakeup);
862 static int nhi_suspend_noirq(struct device *dev)
864 return __nhi_suspend_noirq(dev, device_may_wakeup(dev));
867 static int nhi_freeze_noirq(struct device *dev)
869 struct pci_dev *pdev = to_pci_dev(dev);
870 struct tb *tb = pci_get_drvdata(pdev);
872 return tb_domain_freeze_noirq(tb);
875 static int nhi_thaw_noirq(struct device *dev)
877 struct pci_dev *pdev = to_pci_dev(dev);
878 struct tb *tb = pci_get_drvdata(pdev);
880 return tb_domain_thaw_noirq(tb);
883 static bool nhi_wake_supported(struct pci_dev *pdev)
888 * If power rails are sustainable for wakeup from S4 this
889 * property is set by the BIOS.
891 if (device_property_read_u8(&pdev->dev, "WAKE_SUPPORTED", &val))
897 static int nhi_poweroff_noirq(struct device *dev)
899 struct pci_dev *pdev = to_pci_dev(dev);
902 wakeup = device_may_wakeup(dev) && nhi_wake_supported(pdev);
903 return __nhi_suspend_noirq(dev, wakeup);
906 static void nhi_enable_int_throttling(struct tb_nhi *nhi)
908 /* Throttling is specified in 256ns increments */
909 u32 throttle = DIV_ROUND_UP(128 * NSEC_PER_USEC, 256);
913 * Configure interrupt throttling for all vectors even if we
916 for (i = 0; i < MSIX_MAX_VECS; i++) {
917 u32 reg = REG_INT_THROTTLING_RATE + i * 4;
918 iowrite32(throttle, nhi->iobase + reg);
922 static int nhi_resume_noirq(struct device *dev)
924 struct pci_dev *pdev = to_pci_dev(dev);
925 struct tb *tb = pci_get_drvdata(pdev);
926 struct tb_nhi *nhi = tb->nhi;
930 * Check that the device is still there. It may be that the user
931 * unplugged last device which causes the host controller to go
934 if (!pci_device_is_present(pdev)) {
935 nhi->going_away = true;
937 if (nhi->ops && nhi->ops->resume_noirq) {
938 ret = nhi->ops->resume_noirq(nhi);
942 nhi_enable_int_throttling(tb->nhi);
945 return tb_domain_resume_noirq(tb);
948 static int nhi_suspend(struct device *dev)
950 struct pci_dev *pdev = to_pci_dev(dev);
951 struct tb *tb = pci_get_drvdata(pdev);
953 return tb_domain_suspend(tb);
956 static void nhi_complete(struct device *dev)
958 struct pci_dev *pdev = to_pci_dev(dev);
959 struct tb *tb = pci_get_drvdata(pdev);
962 * If we were runtime suspended when system suspend started,
963 * schedule runtime resume now. It should bring the domain back
964 * to functional state.
966 if (pm_runtime_suspended(&pdev->dev))
967 pm_runtime_resume(&pdev->dev);
969 tb_domain_complete(tb);
972 static int nhi_runtime_suspend(struct device *dev)
974 struct pci_dev *pdev = to_pci_dev(dev);
975 struct tb *tb = pci_get_drvdata(pdev);
976 struct tb_nhi *nhi = tb->nhi;
979 ret = tb_domain_runtime_suspend(tb);
983 if (nhi->ops && nhi->ops->runtime_suspend) {
984 ret = nhi->ops->runtime_suspend(tb->nhi);
991 static int nhi_runtime_resume(struct device *dev)
993 struct pci_dev *pdev = to_pci_dev(dev);
994 struct tb *tb = pci_get_drvdata(pdev);
995 struct tb_nhi *nhi = tb->nhi;
998 if (nhi->ops && nhi->ops->runtime_resume) {
999 ret = nhi->ops->runtime_resume(nhi);
1004 nhi_enable_int_throttling(nhi);
1005 return tb_domain_runtime_resume(tb);
1008 static void nhi_shutdown(struct tb_nhi *nhi)
1012 dev_dbg(&nhi->pdev->dev, "shutdown\n");
1014 for (i = 0; i < nhi->hop_count; i++) {
1015 if (nhi->tx_rings[i])
1016 dev_WARN(&nhi->pdev->dev,
1017 "TX ring %d is still active\n", i);
1018 if (nhi->rx_rings[i])
1019 dev_WARN(&nhi->pdev->dev,
1020 "RX ring %d is still active\n", i);
1022 nhi_disable_interrupts(nhi);
1024 * We have to release the irq before calling flush_work. Otherwise an
1025 * already executing IRQ handler could call schedule_work again.
1027 if (!nhi->pdev->msix_enabled) {
1028 devm_free_irq(&nhi->pdev->dev, nhi->pdev->irq, nhi);
1029 flush_work(&nhi->interrupt_work);
1031 ida_destroy(&nhi->msix_ida);
1033 if (nhi->ops && nhi->ops->shutdown)
1034 nhi->ops->shutdown(nhi);
1037 static int nhi_init_msi(struct tb_nhi *nhi)
1039 struct pci_dev *pdev = nhi->pdev;
1042 /* In case someone left them on. */
1043 nhi_disable_interrupts(nhi);
1045 nhi_enable_int_throttling(nhi);
1047 ida_init(&nhi->msix_ida);
1050 * The NHI has 16 MSI-X vectors or a single MSI. We first try to
1051 * get all MSI-X vectors and if we succeed, each ring will have
1052 * one MSI-X. If for some reason that does not work out, we
1053 * fallback to a single MSI.
1055 nvec = pci_alloc_irq_vectors(pdev, MSIX_MIN_VECS, MSIX_MAX_VECS,
1058 nvec = pci_alloc_irq_vectors(pdev, 1, 1, PCI_IRQ_MSI);
1062 INIT_WORK(&nhi->interrupt_work, nhi_interrupt_work);
1064 irq = pci_irq_vector(nhi->pdev, 0);
1068 res = devm_request_irq(&pdev->dev, irq, nhi_msi,
1069 IRQF_NO_SUSPEND, "thunderbolt", nhi);
1071 dev_err(&pdev->dev, "request_irq failed, aborting\n");
1079 static bool nhi_imr_valid(struct pci_dev *pdev)
1083 if (!device_property_read_u8(&pdev->dev, "IMR_VALID", &val))
1090 * During suspend the Thunderbolt controller is reset and all PCIe
1091 * tunnels are lost. The NHI driver will try to reestablish all tunnels
1092 * during resume. This adds device links between the tunneled PCIe
1093 * downstream ports and the NHI so that the device core will make sure
1094 * NHI is resumed first before the rest.
1096 static void tb_apple_add_links(struct tb_nhi *nhi)
1098 struct pci_dev *upstream, *pdev;
1100 if (!x86_apple_machine)
1103 switch (nhi->pdev->device) {
1104 case PCI_DEVICE_ID_INTEL_LIGHT_RIDGE:
1105 case PCI_DEVICE_ID_INTEL_CACTUS_RIDGE_4C:
1106 case PCI_DEVICE_ID_INTEL_FALCON_RIDGE_2C_NHI:
1107 case PCI_DEVICE_ID_INTEL_FALCON_RIDGE_4C_NHI:
1113 upstream = pci_upstream_bridge(nhi->pdev);
1115 if (!pci_is_pcie(upstream))
1117 if (pci_pcie_type(upstream) == PCI_EXP_TYPE_UPSTREAM)
1119 upstream = pci_upstream_bridge(upstream);
1126 * For each hotplug downstream port, create add device link
1127 * back to NHI so that PCIe tunnels can be re-established after
1130 for_each_pci_bridge(pdev, upstream->subordinate) {
1131 const struct device_link *link;
1133 if (!pci_is_pcie(pdev))
1135 if (pci_pcie_type(pdev) != PCI_EXP_TYPE_DOWNSTREAM ||
1136 !pdev->is_hotplug_bridge)
1139 link = device_link_add(&pdev->dev, &nhi->pdev->dev,
1140 DL_FLAG_AUTOREMOVE_SUPPLIER |
1141 DL_FLAG_PM_RUNTIME);
1143 dev_dbg(&nhi->pdev->dev, "created link from %s\n",
1144 dev_name(&pdev->dev));
1146 dev_warn(&nhi->pdev->dev, "device link creation from %s failed\n",
1147 dev_name(&pdev->dev));
1152 static int nhi_probe(struct pci_dev *pdev, const struct pci_device_id *id)
1158 if (!nhi_imr_valid(pdev)) {
1159 dev_warn(&pdev->dev, "firmware image not valid, aborting\n");
1163 res = pcim_enable_device(pdev);
1165 dev_err(&pdev->dev, "cannot enable PCI device, aborting\n");
1169 res = pcim_iomap_regions(pdev, 1 << 0, "thunderbolt");
1171 dev_err(&pdev->dev, "cannot obtain PCI resources, aborting\n");
1175 nhi = devm_kzalloc(&pdev->dev, sizeof(*nhi), GFP_KERNEL);
1180 nhi->ops = (const struct tb_nhi_ops *)id->driver_data;
1181 /* cannot fail - table is allocated bin pcim_iomap_regions */
1182 nhi->iobase = pcim_iomap_table(pdev)[0];
1183 nhi->hop_count = ioread32(nhi->iobase + REG_HOP_COUNT) & 0x3ff;
1184 dev_dbg(&pdev->dev, "total paths: %d\n", nhi->hop_count);
1186 nhi->tx_rings = devm_kcalloc(&pdev->dev, nhi->hop_count,
1187 sizeof(*nhi->tx_rings), GFP_KERNEL);
1188 nhi->rx_rings = devm_kcalloc(&pdev->dev, nhi->hop_count,
1189 sizeof(*nhi->rx_rings), GFP_KERNEL);
1190 if (!nhi->tx_rings || !nhi->rx_rings)
1193 res = nhi_init_msi(nhi);
1195 dev_err(&pdev->dev, "cannot enable MSI, aborting\n");
1199 spin_lock_init(&nhi->lock);
1201 res = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
1203 res = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32));
1205 dev_err(&pdev->dev, "failed to set DMA mask\n");
1209 pci_set_master(pdev);
1211 if (nhi->ops && nhi->ops->init) {
1212 res = nhi->ops->init(nhi);
1217 tb_apple_add_links(nhi);
1218 tb_acpi_add_links(nhi);
1220 tb = icm_probe(nhi);
1224 dev_err(&nhi->pdev->dev,
1225 "failed to determine connection manager, aborting\n");
1229 dev_dbg(&nhi->pdev->dev, "NHI initialized, starting thunderbolt\n");
1231 res = tb_domain_add(tb);
1234 * At this point the RX/TX rings might already have been
1235 * activated. Do a proper shutdown.
1241 pci_set_drvdata(pdev, tb);
1243 device_wakeup_enable(&pdev->dev);
1245 pm_runtime_allow(&pdev->dev);
1246 pm_runtime_set_autosuspend_delay(&pdev->dev, TB_AUTOSUSPEND_DELAY);
1247 pm_runtime_use_autosuspend(&pdev->dev);
1248 pm_runtime_put_autosuspend(&pdev->dev);
1253 static void nhi_remove(struct pci_dev *pdev)
1255 struct tb *tb = pci_get_drvdata(pdev);
1256 struct tb_nhi *nhi = tb->nhi;
1258 pm_runtime_get_sync(&pdev->dev);
1259 pm_runtime_dont_use_autosuspend(&pdev->dev);
1260 pm_runtime_forbid(&pdev->dev);
1262 tb_domain_remove(tb);
1267 * The tunneled pci bridges are siblings of us. Use resume_noirq to reenable
1268 * the tunnels asap. A corresponding pci quirk blocks the downstream bridges
1269 * resume_noirq until we are done.
1271 static const struct dev_pm_ops nhi_pm_ops = {
1272 .suspend_noirq = nhi_suspend_noirq,
1273 .resume_noirq = nhi_resume_noirq,
1274 .freeze_noirq = nhi_freeze_noirq, /*
1275 * we just disable hotplug, the
1276 * pci-tunnels stay alive.
1278 .thaw_noirq = nhi_thaw_noirq,
1279 .restore_noirq = nhi_resume_noirq,
1280 .suspend = nhi_suspend,
1281 .poweroff_noirq = nhi_poweroff_noirq,
1282 .poweroff = nhi_suspend,
1283 .complete = nhi_complete,
1284 .runtime_suspend = nhi_runtime_suspend,
1285 .runtime_resume = nhi_runtime_resume,
1288 static struct pci_device_id nhi_ids[] = {
1290 * We have to specify class, the TB bridges use the same device and
1291 * vendor (sub)id on gen 1 and gen 2 controllers.
1294 .class = PCI_CLASS_SYSTEM_OTHER << 8, .class_mask = ~0,
1295 .vendor = PCI_VENDOR_ID_INTEL,
1296 .device = PCI_DEVICE_ID_INTEL_LIGHT_RIDGE,
1297 .subvendor = 0x2222, .subdevice = 0x1111,
1300 .class = PCI_CLASS_SYSTEM_OTHER << 8, .class_mask = ~0,
1301 .vendor = PCI_VENDOR_ID_INTEL,
1302 .device = PCI_DEVICE_ID_INTEL_CACTUS_RIDGE_4C,
1303 .subvendor = 0x2222, .subdevice = 0x1111,
1306 .class = PCI_CLASS_SYSTEM_OTHER << 8, .class_mask = ~0,
1307 .vendor = PCI_VENDOR_ID_INTEL,
1308 .device = PCI_DEVICE_ID_INTEL_FALCON_RIDGE_2C_NHI,
1309 .subvendor = PCI_ANY_ID, .subdevice = PCI_ANY_ID,
1312 .class = PCI_CLASS_SYSTEM_OTHER << 8, .class_mask = ~0,
1313 .vendor = PCI_VENDOR_ID_INTEL,
1314 .device = PCI_DEVICE_ID_INTEL_FALCON_RIDGE_4C_NHI,
1315 .subvendor = PCI_ANY_ID, .subdevice = PCI_ANY_ID,
1319 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_ALPINE_RIDGE_2C_NHI) },
1320 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_ALPINE_RIDGE_4C_NHI) },
1321 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_ALPINE_RIDGE_USBONLY_NHI) },
1322 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_ALPINE_RIDGE_LP_NHI) },
1323 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_ALPINE_RIDGE_LP_USBONLY_NHI) },
1324 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_ALPINE_RIDGE_C_2C_NHI) },
1325 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_ALPINE_RIDGE_C_4C_NHI) },
1326 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_ALPINE_RIDGE_C_USBONLY_NHI) },
1327 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_TITAN_RIDGE_2C_NHI) },
1328 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_TITAN_RIDGE_4C_NHI) },
1329 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_ICL_NHI0),
1330 .driver_data = (kernel_ulong_t)&icl_nhi_ops },
1331 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_ICL_NHI1),
1332 .driver_data = (kernel_ulong_t)&icl_nhi_ops },
1333 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_TGL_NHI0),
1334 .driver_data = (kernel_ulong_t)&icl_nhi_ops },
1335 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_TGL_NHI1),
1336 .driver_data = (kernel_ulong_t)&icl_nhi_ops },
1338 /* Any USB4 compliant host */
1339 { PCI_DEVICE_CLASS(PCI_CLASS_SERIAL_USB_USB4, ~0) },
1344 MODULE_DEVICE_TABLE(pci, nhi_ids);
1345 MODULE_LICENSE("GPL");
1347 static struct pci_driver nhi_driver = {
1348 .name = "thunderbolt",
1349 .id_table = nhi_ids,
1351 .remove = nhi_remove,
1352 .shutdown = nhi_remove,
1353 .driver.pm = &nhi_pm_ops,
1356 static int __init nhi_init(void)
1360 ret = tb_domain_init();
1363 ret = pci_register_driver(&nhi_driver);
1369 static void __exit nhi_unload(void)
1371 pci_unregister_driver(&nhi_driver);
1375 rootfs_initcall(nhi_init);
1376 module_exit(nhi_unload);