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>
25 #define RING_TYPE(ring) ((ring)->is_tx ? "TX ring" : "RX ring")
27 #define RING_FIRST_USABLE_HOPID 1
30 * Minimal number of vectors when we use MSI-X. Two for control channel
31 * Rx/Tx and the rest four are for cross domain DMA paths.
33 #define MSIX_MIN_VECS 6
34 #define MSIX_MAX_VECS 16
36 #define NHI_MAILBOX_TIMEOUT 500 /* ms */
38 static int ring_interrupt_index(struct tb_ring *ring)
42 bit += ring->nhi->hop_count;
47 * ring_interrupt_active() - activate/deactivate interrupts for a single ring
49 * ring->nhi->lock must be held.
51 static void ring_interrupt_active(struct tb_ring *ring, bool active)
53 int reg = REG_RING_INTERRUPT_BASE +
54 ring_interrupt_index(ring) / 32 * 4;
55 int bit = ring_interrupt_index(ring) & 31;
60 u32 step, shift, ivr, misc;
61 void __iomem *ivr_base;
67 index = ring->hop + ring->nhi->hop_count;
70 * Ask the hardware to clear interrupt status bits automatically
71 * since we already know which interrupt was triggered.
73 misc = ioread32(ring->nhi->iobase + REG_DMA_MISC);
74 if (!(misc & REG_DMA_MISC_INT_AUTO_CLEAR)) {
75 misc |= REG_DMA_MISC_INT_AUTO_CLEAR;
76 iowrite32(misc, ring->nhi->iobase + REG_DMA_MISC);
79 ivr_base = ring->nhi->iobase + REG_INT_VEC_ALLOC_BASE;
80 step = index / REG_INT_VEC_ALLOC_REGS * REG_INT_VEC_ALLOC_BITS;
81 shift = index % REG_INT_VEC_ALLOC_REGS * REG_INT_VEC_ALLOC_BITS;
82 ivr = ioread32(ivr_base + step);
83 ivr &= ~(REG_INT_VEC_ALLOC_MASK << shift);
85 ivr |= ring->vector << shift;
86 iowrite32(ivr, ivr_base + step);
89 old = ioread32(ring->nhi->iobase + reg);
95 dev_dbg(&ring->nhi->pdev->dev,
96 "%s interrupt at register %#x bit %d (%#x -> %#x)\n",
97 active ? "enabling" : "disabling", reg, bit, old, new);
100 dev_WARN(&ring->nhi->pdev->dev,
101 "interrupt for %s %d is already %s\n",
102 RING_TYPE(ring), ring->hop,
103 active ? "enabled" : "disabled");
104 iowrite32(new, ring->nhi->iobase + reg);
108 * nhi_disable_interrupts() - disable interrupts for all rings
110 * Use only during init and shutdown.
112 static void nhi_disable_interrupts(struct tb_nhi *nhi)
115 /* disable interrupts */
116 for (i = 0; i < RING_INTERRUPT_REG_COUNT(nhi); i++)
117 iowrite32(0, nhi->iobase + REG_RING_INTERRUPT_BASE + 4 * i);
119 /* clear interrupt status bits */
120 for (i = 0; i < RING_NOTIFY_REG_COUNT(nhi); i++)
121 ioread32(nhi->iobase + REG_RING_NOTIFY_BASE + 4 * i);
124 /* ring helper methods */
126 static void __iomem *ring_desc_base(struct tb_ring *ring)
128 void __iomem *io = ring->nhi->iobase;
129 io += ring->is_tx ? REG_TX_RING_BASE : REG_RX_RING_BASE;
130 io += ring->hop * 16;
134 static void __iomem *ring_options_base(struct tb_ring *ring)
136 void __iomem *io = ring->nhi->iobase;
137 io += ring->is_tx ? REG_TX_OPTIONS_BASE : REG_RX_OPTIONS_BASE;
138 io += ring->hop * 32;
142 static void ring_iowrite_cons(struct tb_ring *ring, u16 cons)
145 * The other 16-bits in the register is read-only and writes to it
146 * are ignored by the hardware so we can save one ioread32() by
147 * filling the read-only bits with zeroes.
149 iowrite32(cons, ring_desc_base(ring) + 8);
152 static void ring_iowrite_prod(struct tb_ring *ring, u16 prod)
154 /* See ring_iowrite_cons() above for explanation */
155 iowrite32(prod << 16, ring_desc_base(ring) + 8);
158 static void ring_iowrite32desc(struct tb_ring *ring, u32 value, u32 offset)
160 iowrite32(value, ring_desc_base(ring) + offset);
163 static void ring_iowrite64desc(struct tb_ring *ring, u64 value, u32 offset)
165 iowrite32(value, ring_desc_base(ring) + offset);
166 iowrite32(value >> 32, ring_desc_base(ring) + offset + 4);
169 static void ring_iowrite32options(struct tb_ring *ring, u32 value, u32 offset)
171 iowrite32(value, ring_options_base(ring) + offset);
174 static bool ring_full(struct tb_ring *ring)
176 return ((ring->head + 1) % ring->size) == ring->tail;
179 static bool ring_empty(struct tb_ring *ring)
181 return ring->head == ring->tail;
185 * ring_write_descriptors() - post frames from ring->queue to the controller
187 * ring->lock is held.
189 static void ring_write_descriptors(struct tb_ring *ring)
191 struct ring_frame *frame, *n;
192 struct ring_desc *descriptor;
193 list_for_each_entry_safe(frame, n, &ring->queue, list) {
196 list_move_tail(&frame->list, &ring->in_flight);
197 descriptor = &ring->descriptors[ring->head];
198 descriptor->phys = frame->buffer_phy;
199 descriptor->time = 0;
200 descriptor->flags = RING_DESC_POSTED | RING_DESC_INTERRUPT;
202 descriptor->length = frame->size;
203 descriptor->eof = frame->eof;
204 descriptor->sof = frame->sof;
206 ring->head = (ring->head + 1) % ring->size;
208 ring_iowrite_prod(ring, ring->head);
210 ring_iowrite_cons(ring, ring->head);
215 * ring_work() - progress completed frames
217 * If the ring is shutting down then all frames are marked as canceled and
218 * their callbacks are invoked.
220 * Otherwise we collect all completed frame from the ring buffer, write new
221 * frame to the ring buffer and invoke the callbacks for the completed frames.
223 static void ring_work(struct work_struct *work)
225 struct tb_ring *ring = container_of(work, typeof(*ring), work);
226 struct ring_frame *frame;
227 bool canceled = false;
231 spin_lock_irqsave(&ring->lock, flags);
233 if (!ring->running) {
234 /* Move all frames to done and mark them as canceled. */
235 list_splice_tail_init(&ring->in_flight, &done);
236 list_splice_tail_init(&ring->queue, &done);
238 goto invoke_callback;
241 while (!ring_empty(ring)) {
242 if (!(ring->descriptors[ring->tail].flags
243 & RING_DESC_COMPLETED))
245 frame = list_first_entry(&ring->in_flight, typeof(*frame),
247 list_move_tail(&frame->list, &done);
249 frame->size = ring->descriptors[ring->tail].length;
250 frame->eof = ring->descriptors[ring->tail].eof;
251 frame->sof = ring->descriptors[ring->tail].sof;
252 frame->flags = ring->descriptors[ring->tail].flags;
254 ring->tail = (ring->tail + 1) % ring->size;
256 ring_write_descriptors(ring);
259 /* allow callbacks to schedule new work */
260 spin_unlock_irqrestore(&ring->lock, flags);
261 while (!list_empty(&done)) {
262 frame = list_first_entry(&done, typeof(*frame), list);
264 * The callback may reenqueue or delete frame.
265 * Do not hold on to it.
267 list_del_init(&frame->list);
269 frame->callback(ring, frame, canceled);
273 int __tb_ring_enqueue(struct tb_ring *ring, struct ring_frame *frame)
278 spin_lock_irqsave(&ring->lock, flags);
280 list_add_tail(&frame->list, &ring->queue);
281 ring_write_descriptors(ring);
285 spin_unlock_irqrestore(&ring->lock, flags);
288 EXPORT_SYMBOL_GPL(__tb_ring_enqueue);
291 * tb_ring_poll() - Poll one completed frame from the ring
292 * @ring: Ring to poll
294 * This function can be called when @start_poll callback of the @ring
295 * has been called. It will read one completed frame from the ring and
296 * return it to the caller. Returns %NULL if there is no more completed
299 struct ring_frame *tb_ring_poll(struct tb_ring *ring)
301 struct ring_frame *frame = NULL;
304 spin_lock_irqsave(&ring->lock, flags);
307 if (ring_empty(ring))
310 if (ring->descriptors[ring->tail].flags & RING_DESC_COMPLETED) {
311 frame = list_first_entry(&ring->in_flight, typeof(*frame),
313 list_del_init(&frame->list);
316 frame->size = ring->descriptors[ring->tail].length;
317 frame->eof = ring->descriptors[ring->tail].eof;
318 frame->sof = ring->descriptors[ring->tail].sof;
319 frame->flags = ring->descriptors[ring->tail].flags;
322 ring->tail = (ring->tail + 1) % ring->size;
326 spin_unlock_irqrestore(&ring->lock, flags);
329 EXPORT_SYMBOL_GPL(tb_ring_poll);
331 static void __ring_interrupt_mask(struct tb_ring *ring, bool mask)
333 int idx = ring_interrupt_index(ring);
334 int reg = REG_RING_INTERRUPT_BASE + idx / 32 * 4;
338 val = ioread32(ring->nhi->iobase + reg);
343 iowrite32(val, ring->nhi->iobase + reg);
346 /* Both @nhi->lock and @ring->lock should be held */
347 static void __ring_interrupt(struct tb_ring *ring)
352 if (ring->start_poll) {
353 __ring_interrupt_mask(ring, true);
354 ring->start_poll(ring->poll_data);
356 schedule_work(&ring->work);
361 * tb_ring_poll_complete() - Re-start interrupt for the ring
362 * @ring: Ring to re-start the interrupt
364 * This will re-start (unmask) the ring interrupt once the user is done
367 void tb_ring_poll_complete(struct tb_ring *ring)
371 spin_lock_irqsave(&ring->nhi->lock, flags);
372 spin_lock(&ring->lock);
373 if (ring->start_poll)
374 __ring_interrupt_mask(ring, false);
375 spin_unlock(&ring->lock);
376 spin_unlock_irqrestore(&ring->nhi->lock, flags);
378 EXPORT_SYMBOL_GPL(tb_ring_poll_complete);
380 static irqreturn_t ring_msix(int irq, void *data)
382 struct tb_ring *ring = data;
384 spin_lock(&ring->nhi->lock);
385 spin_lock(&ring->lock);
386 __ring_interrupt(ring);
387 spin_unlock(&ring->lock);
388 spin_unlock(&ring->nhi->lock);
393 static int ring_request_msix(struct tb_ring *ring, bool no_suspend)
395 struct tb_nhi *nhi = ring->nhi;
396 unsigned long irqflags;
399 if (!nhi->pdev->msix_enabled)
402 ret = ida_simple_get(&nhi->msix_ida, 0, MSIX_MAX_VECS, GFP_KERNEL);
408 ret = pci_irq_vector(ring->nhi->pdev, ring->vector);
414 irqflags = no_suspend ? IRQF_NO_SUSPEND : 0;
415 ret = request_irq(ring->irq, ring_msix, irqflags, "thunderbolt", ring);
422 ida_simple_remove(&nhi->msix_ida, ring->vector);
427 static void ring_release_msix(struct tb_ring *ring)
432 free_irq(ring->irq, ring);
433 ida_simple_remove(&ring->nhi->msix_ida, ring->vector);
438 static int nhi_alloc_hop(struct tb_nhi *nhi, struct tb_ring *ring)
442 spin_lock_irq(&nhi->lock);
448 * Automatically allocate HopID from the non-reserved
449 * range 1 .. hop_count - 1.
451 for (i = RING_FIRST_USABLE_HOPID; i < nhi->hop_count; i++) {
453 if (!nhi->tx_rings[i]) {
458 if (!nhi->rx_rings[i]) {
466 if (ring->hop < 0 || ring->hop >= nhi->hop_count) {
467 dev_warn(&nhi->pdev->dev, "invalid hop: %d\n", ring->hop);
471 if (ring->is_tx && nhi->tx_rings[ring->hop]) {
472 dev_warn(&nhi->pdev->dev, "TX hop %d already allocated\n",
476 } else if (!ring->is_tx && nhi->rx_rings[ring->hop]) {
477 dev_warn(&nhi->pdev->dev, "RX hop %d already allocated\n",
484 nhi->tx_rings[ring->hop] = ring;
486 nhi->rx_rings[ring->hop] = ring;
489 spin_unlock_irq(&nhi->lock);
494 static struct tb_ring *tb_ring_alloc(struct tb_nhi *nhi, u32 hop, int size,
495 bool transmit, unsigned int flags,
496 int e2e_tx_hop, u16 sof_mask, u16 eof_mask,
497 void (*start_poll)(void *),
500 struct tb_ring *ring = NULL;
502 dev_dbg(&nhi->pdev->dev, "allocating %s ring %d of size %d\n",
503 transmit ? "TX" : "RX", hop, size);
505 ring = kzalloc(sizeof(*ring), GFP_KERNEL);
509 spin_lock_init(&ring->lock);
510 INIT_LIST_HEAD(&ring->queue);
511 INIT_LIST_HEAD(&ring->in_flight);
512 INIT_WORK(&ring->work, ring_work);
516 ring->is_tx = transmit;
519 ring->e2e_tx_hop = e2e_tx_hop;
520 ring->sof_mask = sof_mask;
521 ring->eof_mask = eof_mask;
524 ring->running = false;
525 ring->start_poll = start_poll;
526 ring->poll_data = poll_data;
528 ring->descriptors = dma_alloc_coherent(&ring->nhi->pdev->dev,
529 size * sizeof(*ring->descriptors),
530 &ring->descriptors_dma, GFP_KERNEL | __GFP_ZERO);
531 if (!ring->descriptors)
534 if (ring_request_msix(ring, flags & RING_FLAG_NO_SUSPEND))
537 if (nhi_alloc_hop(nhi, ring))
538 goto err_release_msix;
543 ring_release_msix(ring);
545 dma_free_coherent(&ring->nhi->pdev->dev,
546 ring->size * sizeof(*ring->descriptors),
547 ring->descriptors, ring->descriptors_dma);
555 * tb_ring_alloc_tx() - Allocate DMA ring for transmit
556 * @nhi: Pointer to the NHI the ring is to be allocated
557 * @hop: HopID (ring) to allocate
558 * @size: Number of entries in the ring
559 * @flags: Flags for the ring
561 struct tb_ring *tb_ring_alloc_tx(struct tb_nhi *nhi, int hop, int size,
564 return tb_ring_alloc(nhi, hop, size, true, flags, 0, 0, 0, NULL, NULL);
566 EXPORT_SYMBOL_GPL(tb_ring_alloc_tx);
569 * tb_ring_alloc_rx() - Allocate DMA ring for receive
570 * @nhi: Pointer to the NHI the ring is to be allocated
571 * @hop: HopID (ring) to allocate. Pass %-1 for automatic allocation.
572 * @size: Number of entries in the ring
573 * @flags: Flags for the ring
574 * @e2e_tx_hop: Transmit HopID when E2E is enabled in @flags
575 * @sof_mask: Mask of PDF values that start a frame
576 * @eof_mask: Mask of PDF values that end a frame
577 * @start_poll: If not %NULL the ring will call this function when an
578 * interrupt is triggered and masked, instead of callback
580 * @poll_data: Optional data passed to @start_poll
582 struct tb_ring *tb_ring_alloc_rx(struct tb_nhi *nhi, int hop, int size,
583 unsigned int flags, int e2e_tx_hop,
584 u16 sof_mask, u16 eof_mask,
585 void (*start_poll)(void *), void *poll_data)
587 return tb_ring_alloc(nhi, hop, size, false, flags, e2e_tx_hop, sof_mask, eof_mask,
588 start_poll, poll_data);
590 EXPORT_SYMBOL_GPL(tb_ring_alloc_rx);
593 * tb_ring_start() - enable a ring
594 * @ring: Ring to start
596 * Must not be invoked in parallel with tb_ring_stop().
598 void tb_ring_start(struct tb_ring *ring)
603 spin_lock_irq(&ring->nhi->lock);
604 spin_lock(&ring->lock);
605 if (ring->nhi->going_away)
608 dev_WARN(&ring->nhi->pdev->dev, "ring already started\n");
611 dev_dbg(&ring->nhi->pdev->dev, "starting %s %d\n",
612 RING_TYPE(ring), ring->hop);
614 if (ring->flags & RING_FLAG_FRAME) {
617 flags = RING_FLAG_ENABLE;
619 frame_size = TB_FRAME_SIZE;
620 flags = RING_FLAG_ENABLE | RING_FLAG_RAW;
623 ring_iowrite64desc(ring, ring->descriptors_dma, 0);
625 ring_iowrite32desc(ring, ring->size, 12);
626 ring_iowrite32options(ring, 0, 4); /* time releated ? */
627 ring_iowrite32options(ring, flags, 0);
629 u32 sof_eof_mask = ring->sof_mask << 16 | ring->eof_mask;
631 ring_iowrite32desc(ring, (frame_size << 16) | ring->size, 12);
632 ring_iowrite32options(ring, sof_eof_mask, 4);
633 ring_iowrite32options(ring, flags, 0);
637 * Now that the ring valid bit is set we can configure E2E if
638 * enabled for the ring.
640 if (ring->flags & RING_FLAG_E2E) {
644 hop = ring->e2e_tx_hop << REG_RX_OPTIONS_E2E_HOP_SHIFT;
645 hop &= REG_RX_OPTIONS_E2E_HOP_MASK;
648 dev_dbg(&ring->nhi->pdev->dev,
649 "enabling E2E for %s %d with TX HopID %d\n",
650 RING_TYPE(ring), ring->hop, ring->e2e_tx_hop);
652 dev_dbg(&ring->nhi->pdev->dev, "enabling E2E for %s %d\n",
653 RING_TYPE(ring), ring->hop);
656 flags |= RING_FLAG_E2E_FLOW_CONTROL;
657 ring_iowrite32options(ring, flags, 0);
660 ring_interrupt_active(ring, true);
661 ring->running = true;
663 spin_unlock(&ring->lock);
664 spin_unlock_irq(&ring->nhi->lock);
666 EXPORT_SYMBOL_GPL(tb_ring_start);
669 * tb_ring_stop() - shutdown a ring
670 * @ring: Ring to stop
672 * Must not be invoked from a callback.
674 * This method will disable the ring. Further calls to
675 * tb_ring_tx/tb_ring_rx will return -ESHUTDOWN until ring_stop has been
678 * All enqueued frames will be canceled and their callbacks will be executed
679 * with frame->canceled set to true (on the callback thread). This method
680 * returns only after all callback invocations have finished.
682 void tb_ring_stop(struct tb_ring *ring)
684 spin_lock_irq(&ring->nhi->lock);
685 spin_lock(&ring->lock);
686 dev_dbg(&ring->nhi->pdev->dev, "stopping %s %d\n",
687 RING_TYPE(ring), ring->hop);
688 if (ring->nhi->going_away)
690 if (!ring->running) {
691 dev_WARN(&ring->nhi->pdev->dev, "%s %d already stopped\n",
692 RING_TYPE(ring), ring->hop);
695 ring_interrupt_active(ring, false);
697 ring_iowrite32options(ring, 0, 0);
698 ring_iowrite64desc(ring, 0, 0);
699 ring_iowrite32desc(ring, 0, 8);
700 ring_iowrite32desc(ring, 0, 12);
703 ring->running = false;
706 spin_unlock(&ring->lock);
707 spin_unlock_irq(&ring->nhi->lock);
710 * schedule ring->work to invoke callbacks on all remaining frames.
712 schedule_work(&ring->work);
713 flush_work(&ring->work);
715 EXPORT_SYMBOL_GPL(tb_ring_stop);
718 * tb_ring_free() - free ring
720 * When this method returns all invocations of ring->callback will have
723 * Ring must be stopped.
725 * Must NOT be called from ring_frame->callback!
727 void tb_ring_free(struct tb_ring *ring)
729 spin_lock_irq(&ring->nhi->lock);
731 * Dissociate the ring from the NHI. This also ensures that
732 * nhi_interrupt_work cannot reschedule ring->work.
735 ring->nhi->tx_rings[ring->hop] = NULL;
737 ring->nhi->rx_rings[ring->hop] = NULL;
740 dev_WARN(&ring->nhi->pdev->dev, "%s %d still running\n",
741 RING_TYPE(ring), ring->hop);
743 spin_unlock_irq(&ring->nhi->lock);
745 ring_release_msix(ring);
747 dma_free_coherent(&ring->nhi->pdev->dev,
748 ring->size * sizeof(*ring->descriptors),
749 ring->descriptors, ring->descriptors_dma);
751 ring->descriptors = NULL;
752 ring->descriptors_dma = 0;
755 dev_dbg(&ring->nhi->pdev->dev, "freeing %s %d\n", RING_TYPE(ring),
759 * ring->work can no longer be scheduled (it is scheduled only
760 * by nhi_interrupt_work, ring_stop and ring_msix). Wait for it
761 * to finish before freeing the ring.
763 flush_work(&ring->work);
766 EXPORT_SYMBOL_GPL(tb_ring_free);
769 * nhi_mailbox_cmd() - Send a command through NHI mailbox
770 * @nhi: Pointer to the NHI structure
771 * @cmd: Command to send
772 * @data: Data to be send with the command
774 * Sends mailbox command to the firmware running on NHI. Returns %0 in
775 * case of success and negative errno in case of failure.
777 int nhi_mailbox_cmd(struct tb_nhi *nhi, enum nhi_mailbox_cmd cmd, u32 data)
782 iowrite32(data, nhi->iobase + REG_INMAIL_DATA);
784 val = ioread32(nhi->iobase + REG_INMAIL_CMD);
785 val &= ~(REG_INMAIL_CMD_MASK | REG_INMAIL_ERROR);
786 val |= REG_INMAIL_OP_REQUEST | cmd;
787 iowrite32(val, nhi->iobase + REG_INMAIL_CMD);
789 timeout = ktime_add_ms(ktime_get(), NHI_MAILBOX_TIMEOUT);
791 val = ioread32(nhi->iobase + REG_INMAIL_CMD);
792 if (!(val & REG_INMAIL_OP_REQUEST))
794 usleep_range(10, 20);
795 } while (ktime_before(ktime_get(), timeout));
797 if (val & REG_INMAIL_OP_REQUEST)
799 if (val & REG_INMAIL_ERROR)
806 * nhi_mailbox_mode() - Return current firmware operation mode
807 * @nhi: Pointer to the NHI structure
809 * The function reads current firmware operation mode using NHI mailbox
810 * registers and returns it to the caller.
812 enum nhi_fw_mode nhi_mailbox_mode(struct tb_nhi *nhi)
816 val = ioread32(nhi->iobase + REG_OUTMAIL_CMD);
817 val &= REG_OUTMAIL_CMD_OPMODE_MASK;
818 val >>= REG_OUTMAIL_CMD_OPMODE_SHIFT;
820 return (enum nhi_fw_mode)val;
823 static void nhi_interrupt_work(struct work_struct *work)
825 struct tb_nhi *nhi = container_of(work, typeof(*nhi), interrupt_work);
826 int value = 0; /* Suppress uninitialized usage warning. */
829 int type = 0; /* current interrupt type 0: TX, 1: RX, 2: RX overflow */
830 struct tb_ring *ring;
832 spin_lock_irq(&nhi->lock);
835 * Starting at REG_RING_NOTIFY_BASE there are three status bitfields
836 * (TX, RX, RX overflow). We iterate over the bits and read a new
837 * dwords as required. The registers are cleared on read.
839 for (bit = 0; bit < 3 * nhi->hop_count; bit++) {
841 value = ioread32(nhi->iobase
842 + REG_RING_NOTIFY_BASE
844 if (++hop == nhi->hop_count) {
848 if ((value & (1 << (bit % 32))) == 0)
851 dev_warn(&nhi->pdev->dev,
852 "RX overflow for ring %d\n",
857 ring = nhi->tx_rings[hop];
859 ring = nhi->rx_rings[hop];
861 dev_warn(&nhi->pdev->dev,
862 "got interrupt for inactive %s ring %d\n",
868 spin_lock(&ring->lock);
869 __ring_interrupt(ring);
870 spin_unlock(&ring->lock);
872 spin_unlock_irq(&nhi->lock);
875 static irqreturn_t nhi_msi(int irq, void *data)
877 struct tb_nhi *nhi = data;
878 schedule_work(&nhi->interrupt_work);
882 static int __nhi_suspend_noirq(struct device *dev, bool wakeup)
884 struct pci_dev *pdev = to_pci_dev(dev);
885 struct tb *tb = pci_get_drvdata(pdev);
886 struct tb_nhi *nhi = tb->nhi;
889 ret = tb_domain_suspend_noirq(tb);
893 if (nhi->ops && nhi->ops->suspend_noirq) {
894 ret = nhi->ops->suspend_noirq(tb->nhi, wakeup);
902 static int nhi_suspend_noirq(struct device *dev)
904 return __nhi_suspend_noirq(dev, device_may_wakeup(dev));
907 static int nhi_freeze_noirq(struct device *dev)
909 struct pci_dev *pdev = to_pci_dev(dev);
910 struct tb *tb = pci_get_drvdata(pdev);
912 return tb_domain_freeze_noirq(tb);
915 static int nhi_thaw_noirq(struct device *dev)
917 struct pci_dev *pdev = to_pci_dev(dev);
918 struct tb *tb = pci_get_drvdata(pdev);
920 return tb_domain_thaw_noirq(tb);
923 static bool nhi_wake_supported(struct pci_dev *pdev)
928 * If power rails are sustainable for wakeup from S4 this
929 * property is set by the BIOS.
931 if (device_property_read_u8(&pdev->dev, "WAKE_SUPPORTED", &val))
937 static int nhi_poweroff_noirq(struct device *dev)
939 struct pci_dev *pdev = to_pci_dev(dev);
942 wakeup = device_may_wakeup(dev) && nhi_wake_supported(pdev);
943 return __nhi_suspend_noirq(dev, wakeup);
946 static void nhi_enable_int_throttling(struct tb_nhi *nhi)
948 /* Throttling is specified in 256ns increments */
949 u32 throttle = DIV_ROUND_UP(128 * NSEC_PER_USEC, 256);
953 * Configure interrupt throttling for all vectors even if we
956 for (i = 0; i < MSIX_MAX_VECS; i++) {
957 u32 reg = REG_INT_THROTTLING_RATE + i * 4;
958 iowrite32(throttle, nhi->iobase + reg);
962 static int nhi_resume_noirq(struct device *dev)
964 struct pci_dev *pdev = to_pci_dev(dev);
965 struct tb *tb = pci_get_drvdata(pdev);
966 struct tb_nhi *nhi = tb->nhi;
970 * Check that the device is still there. It may be that the user
971 * unplugged last device which causes the host controller to go
974 if (!pci_device_is_present(pdev)) {
975 nhi->going_away = true;
977 if (nhi->ops && nhi->ops->resume_noirq) {
978 ret = nhi->ops->resume_noirq(nhi);
982 nhi_enable_int_throttling(tb->nhi);
985 return tb_domain_resume_noirq(tb);
988 static int nhi_suspend(struct device *dev)
990 struct pci_dev *pdev = to_pci_dev(dev);
991 struct tb *tb = pci_get_drvdata(pdev);
993 return tb_domain_suspend(tb);
996 static void nhi_complete(struct device *dev)
998 struct pci_dev *pdev = to_pci_dev(dev);
999 struct tb *tb = pci_get_drvdata(pdev);
1002 * If we were runtime suspended when system suspend started,
1003 * schedule runtime resume now. It should bring the domain back
1004 * to functional state.
1006 if (pm_runtime_suspended(&pdev->dev))
1007 pm_runtime_resume(&pdev->dev);
1009 tb_domain_complete(tb);
1012 static int nhi_runtime_suspend(struct device *dev)
1014 struct pci_dev *pdev = to_pci_dev(dev);
1015 struct tb *tb = pci_get_drvdata(pdev);
1016 struct tb_nhi *nhi = tb->nhi;
1019 ret = tb_domain_runtime_suspend(tb);
1023 if (nhi->ops && nhi->ops->runtime_suspend) {
1024 ret = nhi->ops->runtime_suspend(tb->nhi);
1031 static int nhi_runtime_resume(struct device *dev)
1033 struct pci_dev *pdev = to_pci_dev(dev);
1034 struct tb *tb = pci_get_drvdata(pdev);
1035 struct tb_nhi *nhi = tb->nhi;
1038 if (nhi->ops && nhi->ops->runtime_resume) {
1039 ret = nhi->ops->runtime_resume(nhi);
1044 nhi_enable_int_throttling(nhi);
1045 return tb_domain_runtime_resume(tb);
1048 static void nhi_shutdown(struct tb_nhi *nhi)
1052 dev_dbg(&nhi->pdev->dev, "shutdown\n");
1054 for (i = 0; i < nhi->hop_count; i++) {
1055 if (nhi->tx_rings[i])
1056 dev_WARN(&nhi->pdev->dev,
1057 "TX ring %d is still active\n", i);
1058 if (nhi->rx_rings[i])
1059 dev_WARN(&nhi->pdev->dev,
1060 "RX ring %d is still active\n", i);
1062 nhi_disable_interrupts(nhi);
1064 * We have to release the irq before calling flush_work. Otherwise an
1065 * already executing IRQ handler could call schedule_work again.
1067 if (!nhi->pdev->msix_enabled) {
1068 devm_free_irq(&nhi->pdev->dev, nhi->pdev->irq, nhi);
1069 flush_work(&nhi->interrupt_work);
1071 ida_destroy(&nhi->msix_ida);
1073 if (nhi->ops && nhi->ops->shutdown)
1074 nhi->ops->shutdown(nhi);
1077 static int nhi_init_msi(struct tb_nhi *nhi)
1079 struct pci_dev *pdev = nhi->pdev;
1082 /* In case someone left them on. */
1083 nhi_disable_interrupts(nhi);
1085 nhi_enable_int_throttling(nhi);
1087 ida_init(&nhi->msix_ida);
1090 * The NHI has 16 MSI-X vectors or a single MSI. We first try to
1091 * get all MSI-X vectors and if we succeed, each ring will have
1092 * one MSI-X. If for some reason that does not work out, we
1093 * fallback to a single MSI.
1095 nvec = pci_alloc_irq_vectors(pdev, MSIX_MIN_VECS, MSIX_MAX_VECS,
1098 nvec = pci_alloc_irq_vectors(pdev, 1, 1, PCI_IRQ_MSI);
1102 INIT_WORK(&nhi->interrupt_work, nhi_interrupt_work);
1104 irq = pci_irq_vector(nhi->pdev, 0);
1108 res = devm_request_irq(&pdev->dev, irq, nhi_msi,
1109 IRQF_NO_SUSPEND, "thunderbolt", nhi);
1111 dev_err(&pdev->dev, "request_irq failed, aborting\n");
1119 static bool nhi_imr_valid(struct pci_dev *pdev)
1123 if (!device_property_read_u8(&pdev->dev, "IMR_VALID", &val))
1129 static struct tb *nhi_select_cm(struct tb_nhi *nhi)
1134 * USB4 case is simple. If we got control of any of the
1135 * capabilities, we use software CM.
1137 if (tb_acpi_is_native())
1138 return tb_probe(nhi);
1141 * Either firmware based CM is running (we did not get control
1142 * from the firmware) or this is pre-USB4 PC so try first
1143 * firmware CM and then fallback to software CM.
1145 tb = icm_probe(nhi);
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 = nhi_select_cm(nhi);
1219 dev_err(&nhi->pdev->dev,
1220 "failed to determine connection manager, aborting\n");
1224 dev_dbg(&nhi->pdev->dev, "NHI initialized, starting thunderbolt\n");
1226 res = tb_domain_add(tb);
1229 * At this point the RX/TX rings might already have been
1230 * activated. Do a proper shutdown.
1236 pci_set_drvdata(pdev, tb);
1238 device_wakeup_enable(&pdev->dev);
1240 pm_runtime_allow(&pdev->dev);
1241 pm_runtime_set_autosuspend_delay(&pdev->dev, TB_AUTOSUSPEND_DELAY);
1242 pm_runtime_use_autosuspend(&pdev->dev);
1243 pm_runtime_put_autosuspend(&pdev->dev);
1248 static void nhi_remove(struct pci_dev *pdev)
1250 struct tb *tb = pci_get_drvdata(pdev);
1251 struct tb_nhi *nhi = tb->nhi;
1253 pm_runtime_get_sync(&pdev->dev);
1254 pm_runtime_dont_use_autosuspend(&pdev->dev);
1255 pm_runtime_forbid(&pdev->dev);
1257 tb_domain_remove(tb);
1262 * The tunneled pci bridges are siblings of us. Use resume_noirq to reenable
1263 * the tunnels asap. A corresponding pci quirk blocks the downstream bridges
1264 * resume_noirq until we are done.
1266 static const struct dev_pm_ops nhi_pm_ops = {
1267 .suspend_noirq = nhi_suspend_noirq,
1268 .resume_noirq = nhi_resume_noirq,
1269 .freeze_noirq = nhi_freeze_noirq, /*
1270 * we just disable hotplug, the
1271 * pci-tunnels stay alive.
1273 .thaw_noirq = nhi_thaw_noirq,
1274 .restore_noirq = nhi_resume_noirq,
1275 .suspend = nhi_suspend,
1276 .poweroff_noirq = nhi_poweroff_noirq,
1277 .poweroff = nhi_suspend,
1278 .complete = nhi_complete,
1279 .runtime_suspend = nhi_runtime_suspend,
1280 .runtime_resume = nhi_runtime_resume,
1283 static struct pci_device_id nhi_ids[] = {
1285 * We have to specify class, the TB bridges use the same device and
1286 * vendor (sub)id on gen 1 and gen 2 controllers.
1289 .class = PCI_CLASS_SYSTEM_OTHER << 8, .class_mask = ~0,
1290 .vendor = PCI_VENDOR_ID_INTEL,
1291 .device = PCI_DEVICE_ID_INTEL_LIGHT_RIDGE,
1292 .subvendor = 0x2222, .subdevice = 0x1111,
1295 .class = PCI_CLASS_SYSTEM_OTHER << 8, .class_mask = ~0,
1296 .vendor = PCI_VENDOR_ID_INTEL,
1297 .device = PCI_DEVICE_ID_INTEL_CACTUS_RIDGE_4C,
1298 .subvendor = 0x2222, .subdevice = 0x1111,
1301 .class = PCI_CLASS_SYSTEM_OTHER << 8, .class_mask = ~0,
1302 .vendor = PCI_VENDOR_ID_INTEL,
1303 .device = PCI_DEVICE_ID_INTEL_FALCON_RIDGE_2C_NHI,
1304 .subvendor = PCI_ANY_ID, .subdevice = PCI_ANY_ID,
1307 .class = PCI_CLASS_SYSTEM_OTHER << 8, .class_mask = ~0,
1308 .vendor = PCI_VENDOR_ID_INTEL,
1309 .device = PCI_DEVICE_ID_INTEL_FALCON_RIDGE_4C_NHI,
1310 .subvendor = PCI_ANY_ID, .subdevice = PCI_ANY_ID,
1314 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_ALPINE_RIDGE_2C_NHI) },
1315 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_ALPINE_RIDGE_4C_NHI) },
1316 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_ALPINE_RIDGE_USBONLY_NHI) },
1317 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_ALPINE_RIDGE_LP_NHI) },
1318 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_ALPINE_RIDGE_LP_USBONLY_NHI) },
1319 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_ALPINE_RIDGE_C_2C_NHI) },
1320 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_ALPINE_RIDGE_C_4C_NHI) },
1321 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_ALPINE_RIDGE_C_USBONLY_NHI) },
1322 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_TITAN_RIDGE_2C_NHI) },
1323 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_TITAN_RIDGE_4C_NHI) },
1324 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_ICL_NHI0),
1325 .driver_data = (kernel_ulong_t)&icl_nhi_ops },
1326 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_ICL_NHI1),
1327 .driver_data = (kernel_ulong_t)&icl_nhi_ops },
1328 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_TGL_NHI0),
1329 .driver_data = (kernel_ulong_t)&icl_nhi_ops },
1330 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_TGL_NHI1),
1331 .driver_data = (kernel_ulong_t)&icl_nhi_ops },
1332 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_TGL_H_NHI0),
1333 .driver_data = (kernel_ulong_t)&icl_nhi_ops },
1334 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_TGL_H_NHI1),
1335 .driver_data = (kernel_ulong_t)&icl_nhi_ops },
1336 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_ADL_NHI0),
1337 .driver_data = (kernel_ulong_t)&icl_nhi_ops },
1338 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_ADL_NHI1),
1339 .driver_data = (kernel_ulong_t)&icl_nhi_ops },
1341 /* Any USB4 compliant host */
1342 { PCI_DEVICE_CLASS(PCI_CLASS_SERIAL_USB_USB4, ~0) },
1347 MODULE_DEVICE_TABLE(pci, nhi_ids);
1348 MODULE_LICENSE("GPL");
1350 static struct pci_driver nhi_driver = {
1351 .name = "thunderbolt",
1352 .id_table = nhi_ids,
1354 .remove = nhi_remove,
1355 .shutdown = nhi_remove,
1356 .driver.pm = &nhi_pm_ops,
1359 static int __init nhi_init(void)
1363 ret = tb_domain_init();
1366 ret = pci_register_driver(&nhi_driver);
1372 static void __exit nhi_unload(void)
1374 pci_unregister_driver(&nhi_driver);
1378 rootfs_initcall(nhi_init);
1379 module_exit(nhi_unload);