1// SPDX-License-Identifier: GPL-2.0
2/*
3 * xHCI host controller driver
4 *
5 * Copyright (C) 2008 Intel Corp.
6 *
7 * Author: Sarah Sharp
8 * Some code borrowed from the Linux EHCI driver.
9 */
10
11#include <linux/usb.h>
12#include <linux/overflow.h>
13#include <linux/pci.h>
14#include <linux/slab.h>
15#include <linux/dmapool.h>
16#include <linux/dma-mapping.h>
17
18#include "xhci.h"
19#include "xhci-trace.h"
20#include "xhci-debugfs.h"
21
22/*
23 * Allocates a generic ring segment from the ring pool, sets the dma address,
24 * initializes the segment to zero, and sets the private next pointer to NULL.
25 *
26 * Section 4.11.1.1:
27 * "All components of all Command and Transfer TRBs shall be initialized to '0'"
28 */
29static struct xhci_segment *xhci_segment_alloc(struct xhci_hcd *xhci,
30					       unsigned int cycle_state,
31					       unsigned int max_packet,
32					       unsigned int num,
33					       gfp_t flags)
34{
35	struct xhci_segment *seg;
36	dma_addr_t	dma;
37	int		i;
38	struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
39
40	seg = kzalloc_node(sizeof(*seg), flags, dev_to_node(dev));
41	if (!seg)
42		return NULL;
43
44	seg->trbs = dma_pool_zalloc(xhci->segment_pool, flags, &dma);
45	if (!seg->trbs) {
46		kfree(seg);
47		return NULL;
48	}
49
50	if (max_packet) {
51		seg->bounce_buf = kzalloc_node(max_packet, flags,
52					dev_to_node(dev));
53		if (!seg->bounce_buf) {
54			dma_pool_free(xhci->segment_pool, seg->trbs, dma);
55			kfree(seg);
56			return NULL;
57		}
58	}
59	/* If the cycle state is 0, set the cycle bit to 1 for all the TRBs */
60	if (cycle_state == 0) {
61		for (i = 0; i < TRBS_PER_SEGMENT; i++)
62			seg->trbs[i].link.control = cpu_to_le32(TRB_CYCLE);
63	}
64	seg->num = num;
65	seg->dma = dma;
66	seg->next = NULL;
67
68	return seg;
69}
70
71static void xhci_segment_free(struct xhci_hcd *xhci, struct xhci_segment *seg)
72{
73	if (seg->trbs) {
74		dma_pool_free(xhci->segment_pool, seg->trbs, seg->dma);
75		seg->trbs = NULL;
76	}
77	kfree(seg->bounce_buf);
78	kfree(seg);
79}
80
81static void xhci_free_segments_for_ring(struct xhci_hcd *xhci,
82				struct xhci_segment *first)
83{
84	struct xhci_segment *seg;
85
86	seg = first->next;
87	while (seg && seg != first) {
88		struct xhci_segment *next = seg->next;
89		xhci_segment_free(xhci, seg);
90		seg = next;
91	}
92	xhci_segment_free(xhci, first);
93}
94
95/*
96 * Make the prev segment point to the next segment.
97 *
98 * Change the last TRB in the prev segment to be a Link TRB which points to the
99 * DMA address of the next segment.  The caller needs to set any Link TRB
100 * related flags, such as End TRB, Toggle Cycle, and no snoop.
101 */
102static void xhci_link_segments(struct xhci_segment *prev,
103			       struct xhci_segment *next,
104			       enum xhci_ring_type type, bool chain_links)
105{
106	u32 val;
107
108	if (!prev || !next)
109		return;
110	prev->next = next;
111	if (type != TYPE_EVENT) {
112		prev->trbs[TRBS_PER_SEGMENT-1].link.segment_ptr =
113			cpu_to_le64(next->dma);
114
115		/* Set the last TRB in the segment to have a TRB type ID of Link TRB */
116		val = le32_to_cpu(prev->trbs[TRBS_PER_SEGMENT-1].link.control);
117		val &= ~TRB_TYPE_BITMASK;
118		val |= TRB_TYPE(TRB_LINK);
119		if (chain_links)
120			val |= TRB_CHAIN;
121		prev->trbs[TRBS_PER_SEGMENT-1].link.control = cpu_to_le32(val);
122	}
123}
124
125/*
126 * Link the ring to the new segments.
127 * Set Toggle Cycle for the new ring if needed.
128 */
129static void xhci_link_rings(struct xhci_hcd *xhci, struct xhci_ring *ring,
130		struct xhci_segment *first, struct xhci_segment *last,
131		unsigned int num_segs)
132{
133	struct xhci_segment *next, *seg;
134	bool chain_links;
135
136	if (!ring || !first || !last)
137		return;
138
139	/* Set chain bit for 0.95 hosts, and for isoc rings on AMD 0.96 host */
140	chain_links = !!(xhci_link_trb_quirk(xhci) ||
141			 (ring->type == TYPE_ISOC &&
142			  (xhci->quirks & XHCI_AMD_0x96_HOST)));
143
144	next = ring->enq_seg->next;
145	xhci_link_segments(ring->enq_seg, first, ring->type, chain_links);
146	xhci_link_segments(last, next, ring->type, chain_links);
147	ring->num_segs += num_segs;
148
149	if (ring->enq_seg == ring->last_seg) {
150		if (ring->type != TYPE_EVENT) {
151			ring->last_seg->trbs[TRBS_PER_SEGMENT-1].link.control
152				&= ~cpu_to_le32(LINK_TOGGLE);
153			last->trbs[TRBS_PER_SEGMENT-1].link.control
154				|= cpu_to_le32(LINK_TOGGLE);
155		}
156		ring->last_seg = last;
157	}
158
159	for (seg = last; seg != ring->last_seg; seg = seg->next)
160		seg->next->num = seg->num + 1;
161}
162
163/*
164 * We need a radix tree for mapping physical addresses of TRBs to which stream
165 * ID they belong to.  We need to do this because the host controller won't tell
166 * us which stream ring the TRB came from.  We could store the stream ID in an
167 * event data TRB, but that doesn't help us for the cancellation case, since the
168 * endpoint may stop before it reaches that event data TRB.
169 *
170 * The radix tree maps the upper portion of the TRB DMA address to a ring
171 * segment that has the same upper portion of DMA addresses.  For example, say I
172 * have segments of size 1KB, that are always 1KB aligned.  A segment may
173 * start at 0x10c91000 and end at 0x10c913f0.  If I use the upper 10 bits, the
174 * key to the stream ID is 0x43244.  I can use the DMA address of the TRB to
175 * pass the radix tree a key to get the right stream ID:
176 *
177 *	0x10c90fff >> 10 = 0x43243
178 *	0x10c912c0 >> 10 = 0x43244
179 *	0x10c91400 >> 10 = 0x43245
180 *
181 * Obviously, only those TRBs with DMA addresses that are within the segment
182 * will make the radix tree return the stream ID for that ring.
183 *
184 * Caveats for the radix tree:
185 *
186 * The radix tree uses an unsigned long as a key pair.  On 32-bit systems, an
187 * unsigned long will be 32-bits; on a 64-bit system an unsigned long will be
188 * 64-bits.  Since we only request 32-bit DMA addresses, we can use that as the
189 * key on 32-bit or 64-bit systems (it would also be fine if we asked for 64-bit
190 * PCI DMA addresses on a 64-bit system).  There might be a problem on 32-bit
191 * extended systems (where the DMA address can be bigger than 32-bits),
192 * if we allow the PCI dma mask to be bigger than 32-bits.  So don't do that.
193 */
194static int xhci_insert_segment_mapping(struct radix_tree_root *trb_address_map,
195		struct xhci_ring *ring,
196		struct xhci_segment *seg,
197		gfp_t mem_flags)
198{
199	unsigned long key;
200	int ret;
201
202	key = (unsigned long)(seg->dma >> TRB_SEGMENT_SHIFT);
203	/* Skip any segments that were already added. */
204	if (radix_tree_lookup(trb_address_map, key))
205		return 0;
206
207	ret = radix_tree_maybe_preload(mem_flags);
208	if (ret)
209		return ret;
210	ret = radix_tree_insert(trb_address_map,
211			key, ring);
212	radix_tree_preload_end();
213	return ret;
214}
215
216static void xhci_remove_segment_mapping(struct radix_tree_root *trb_address_map,
217		struct xhci_segment *seg)
218{
219	unsigned long key;
220
221	key = (unsigned long)(seg->dma >> TRB_SEGMENT_SHIFT);
222	if (radix_tree_lookup(trb_address_map, key))
223		radix_tree_delete(trb_address_map, key);
224}
225
226static int xhci_update_stream_segment_mapping(
227		struct radix_tree_root *trb_address_map,
228		struct xhci_ring *ring,
229		struct xhci_segment *first_seg,
230		struct xhci_segment *last_seg,
231		gfp_t mem_flags)
232{
233	struct xhci_segment *seg;
234	struct xhci_segment *failed_seg;
235	int ret;
236
237	if (WARN_ON_ONCE(trb_address_map == NULL))
238		return 0;
239
240	seg = first_seg;
241	do {
242		ret = xhci_insert_segment_mapping(trb_address_map,
243				ring, seg, mem_flags);
244		if (ret)
245			goto remove_streams;
246		if (seg == last_seg)
247			return 0;
248		seg = seg->next;
249	} while (seg != first_seg);
250
251	return 0;
252
253remove_streams:
254	failed_seg = seg;
255	seg = first_seg;
256	do {
257		xhci_remove_segment_mapping(trb_address_map, seg);
258		if (seg == failed_seg)
259			return ret;
260		seg = seg->next;
261	} while (seg != first_seg);
262
263	return ret;
264}
265
266static void xhci_remove_stream_mapping(struct xhci_ring *ring)
267{
268	struct xhci_segment *seg;
269
270	if (WARN_ON_ONCE(ring->trb_address_map == NULL))
271		return;
272
273	seg = ring->first_seg;
274	do {
275		xhci_remove_segment_mapping(ring->trb_address_map, seg);
276		seg = seg->next;
277	} while (seg != ring->first_seg);
278}
279
280static int xhci_update_stream_mapping(struct xhci_ring *ring, gfp_t mem_flags)
281{
282	return xhci_update_stream_segment_mapping(ring->trb_address_map, ring,
283			ring->first_seg, ring->last_seg, mem_flags);
284}
285
286/* XXX: Do we need the hcd structure in all these functions? */
287void xhci_ring_free(struct xhci_hcd *xhci, struct xhci_ring *ring)
288{
289	if (!ring)
290		return;
291
292	trace_xhci_ring_free(ring);
293
294	if (ring->first_seg) {
295		if (ring->type == TYPE_STREAM)
296			xhci_remove_stream_mapping(ring);
297		xhci_free_segments_for_ring(xhci, ring->first_seg);
298	}
299
300	kfree(ring);
301}
302
303void xhci_initialize_ring_info(struct xhci_ring *ring,
304			       unsigned int cycle_state)
305{
306	/* The ring is empty, so the enqueue pointer == dequeue pointer */
307	ring->enqueue = ring->first_seg->trbs;
308	ring->enq_seg = ring->first_seg;
309	ring->dequeue = ring->enqueue;
310	ring->deq_seg = ring->first_seg;
311	/* The ring is initialized to 0. The producer must write 1 to the cycle
312	 * bit to handover ownership of the TRB, so PCS = 1.  The consumer must
313	 * compare CCS to the cycle bit to check ownership, so CCS = 1.
314	 *
315	 * New rings are initialized with cycle state equal to 1; if we are
316	 * handling ring expansion, set the cycle state equal to the old ring.
317	 */
318	ring->cycle_state = cycle_state;
319
320	/*
321	 * Each segment has a link TRB, and leave an extra TRB for SW
322	 * accounting purpose
323	 */
324	ring->num_trbs_free = ring->num_segs * (TRBS_PER_SEGMENT - 1) - 1;
325}
326EXPORT_SYMBOL_GPL(xhci_initialize_ring_info);
327
328/* Allocate segments and link them for a ring */
329static int xhci_alloc_segments_for_ring(struct xhci_hcd *xhci,
330		struct xhci_segment **first, struct xhci_segment **last,
331		unsigned int num_segs, unsigned int num,
332		unsigned int cycle_state, enum xhci_ring_type type,
333		unsigned int max_packet, gfp_t flags)
334{
335	struct xhci_segment *prev;
336	bool chain_links;
337
338	/* Set chain bit for 0.95 hosts, and for isoc rings on AMD 0.96 host */
339	chain_links = !!(xhci_link_trb_quirk(xhci) ||
340			 (type == TYPE_ISOC &&
341			  (xhci->quirks & XHCI_AMD_0x96_HOST)));
342
343	prev = xhci_segment_alloc(xhci, cycle_state, max_packet, num, flags);
344	if (!prev)
345		return -ENOMEM;
346	num++;
347
348	*first = prev;
349	while (num < num_segs) {
350		struct xhci_segment	*next;
351
352		next = xhci_segment_alloc(xhci, cycle_state, max_packet, num,
353					  flags);
354		if (!next)
355			goto free_segments;
356
357		xhci_link_segments(prev, next, type, chain_links);
358		prev = next;
359		num++;
360	}
361	xhci_link_segments(prev, *first, type, chain_links);
362	*last = prev;
363
364	return 0;
365
366free_segments:
367	xhci_free_segments_for_ring(xhci, *first);
368	return -ENOMEM;
369}
370
371/*
372 * Create a new ring with zero or more segments.
373 *
374 * Link each segment together into a ring.
375 * Set the end flag and the cycle toggle bit on the last segment.
376 * See section 4.9.1 and figures 15 and 16.
377 */
378struct xhci_ring *xhci_ring_alloc(struct xhci_hcd *xhci,
379		unsigned int num_segs, unsigned int cycle_state,
380		enum xhci_ring_type type, unsigned int max_packet, gfp_t flags)
381{
382	struct xhci_ring	*ring;
383	int ret;
384	struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
385
386	ring = kzalloc_node(sizeof(*ring), flags, dev_to_node(dev));
387	if (!ring)
388		return NULL;
389
390	ring->num_segs = num_segs;
391	ring->bounce_buf_len = max_packet;
392	INIT_LIST_HEAD(&ring->td_list);
393	ring->type = type;
394	if (num_segs == 0)
395		return ring;
396
397	ret = xhci_alloc_segments_for_ring(xhci, &ring->first_seg,
398			&ring->last_seg, num_segs, 0, cycle_state, type,
399			max_packet, flags);
400	if (ret)
401		goto fail;
402
403	/* Only event ring does not use link TRB */
404	if (type != TYPE_EVENT) {
405		/* See section 4.9.2.1 and 6.4.4.1 */
406		ring->last_seg->trbs[TRBS_PER_SEGMENT - 1].link.control |=
407			cpu_to_le32(LINK_TOGGLE);
408	}
409	xhci_initialize_ring_info(ring, cycle_state);
410	trace_xhci_ring_alloc(ring);
411	return ring;
412
413fail:
414	kfree(ring);
415	return NULL;
416}
417
418void xhci_free_endpoint_ring(struct xhci_hcd *xhci,
419		struct xhci_virt_device *virt_dev,
420		unsigned int ep_index)
421{
422	xhci_ring_free(xhci, virt_dev->eps[ep_index].ring);
423	virt_dev->eps[ep_index].ring = NULL;
424}
425
426/*
427 * Expand an existing ring.
428 * Allocate a new ring which has same segment numbers and link the two rings.
429 */
430int xhci_ring_expansion(struct xhci_hcd *xhci, struct xhci_ring *ring,
431				unsigned int num_new_segs, gfp_t flags)
432{
433	struct xhci_segment	*first;
434	struct xhci_segment	*last;
435	int			ret;
436
437	ret = xhci_alloc_segments_for_ring(xhci, &first, &last,
438			num_new_segs, ring->enq_seg->num + 1,
439			ring->cycle_state, ring->type,
440			ring->bounce_buf_len, flags);
441	if (ret)
442		return -ENOMEM;
443
444	if (ring->type == TYPE_STREAM) {
445		ret = xhci_update_stream_segment_mapping(ring->trb_address_map,
446						ring, first, last, flags);
447		if (ret)
448			goto free_segments;
449	}
450
451	xhci_link_rings(xhci, ring, first, last, num_new_segs);
452	trace_xhci_ring_expansion(ring);
453	xhci_dbg_trace(xhci, trace_xhci_dbg_ring_expansion,
454			"ring expansion succeed, now has %d segments",
455			ring->num_segs);
456
457	return 0;
458
459free_segments:
460	xhci_free_segments_for_ring(xhci, first);
461	return ret;
462}
463
464struct xhci_container_ctx *xhci_alloc_container_ctx(struct xhci_hcd *xhci,
465						    int type, gfp_t flags)
466{
467	struct xhci_container_ctx *ctx;
468	struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
469
470	if ((type != XHCI_CTX_TYPE_DEVICE) && (type != XHCI_CTX_TYPE_INPUT))
471		return NULL;
472
473	ctx = kzalloc_node(sizeof(*ctx), flags, dev_to_node(dev));
474	if (!ctx)
475		return NULL;
476
477	ctx->type = type;
478	ctx->size = HCC_64BYTE_CONTEXT(xhci->hcc_params) ? 2048 : 1024;
479	if (type == XHCI_CTX_TYPE_INPUT)
480		ctx->size += CTX_SIZE(xhci->hcc_params);
481
482	ctx->bytes = dma_pool_zalloc(xhci->device_pool, flags, &ctx->dma);
483	if (!ctx->bytes) {
484		kfree(ctx);
485		return NULL;
486	}
487	return ctx;
488}
489
490void xhci_free_container_ctx(struct xhci_hcd *xhci,
491			     struct xhci_container_ctx *ctx)
492{
493	if (!ctx)
494		return;
495	dma_pool_free(xhci->device_pool, ctx->bytes, ctx->dma);
496	kfree(ctx);
497}
498
499struct xhci_input_control_ctx *xhci_get_input_control_ctx(
500					      struct xhci_container_ctx *ctx)
501{
502	if (ctx->type != XHCI_CTX_TYPE_INPUT)
503		return NULL;
504
505	return (struct xhci_input_control_ctx *)ctx->bytes;
506}
507
508struct xhci_slot_ctx *xhci_get_slot_ctx(struct xhci_hcd *xhci,
509					struct xhci_container_ctx *ctx)
510{
511	if (ctx->type == XHCI_CTX_TYPE_DEVICE)
512		return (struct xhci_slot_ctx *)ctx->bytes;
513
514	return (struct xhci_slot_ctx *)
515		(ctx->bytes + CTX_SIZE(xhci->hcc_params));
516}
517
518struct xhci_ep_ctx *xhci_get_ep_ctx(struct xhci_hcd *xhci,
519				    struct xhci_container_ctx *ctx,
520				    unsigned int ep_index)
521{
522	/* increment ep index by offset of start of ep ctx array */
523	ep_index++;
524	if (ctx->type == XHCI_CTX_TYPE_INPUT)
525		ep_index++;
526
527	return (struct xhci_ep_ctx *)
528		(ctx->bytes + (ep_index * CTX_SIZE(xhci->hcc_params)));
529}
530EXPORT_SYMBOL_GPL(xhci_get_ep_ctx);
531
532/***************** Streams structures manipulation *************************/
533
534static void xhci_free_stream_ctx(struct xhci_hcd *xhci,
535		unsigned int num_stream_ctxs,
536		struct xhci_stream_ctx *stream_ctx, dma_addr_t dma)
537{
538	struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
539	size_t size = sizeof(struct xhci_stream_ctx) * num_stream_ctxs;
540
541	if (size > MEDIUM_STREAM_ARRAY_SIZE)
542		dma_free_coherent(dev, size, stream_ctx, dma);
543	else if (size > SMALL_STREAM_ARRAY_SIZE)
544		dma_pool_free(xhci->medium_streams_pool, stream_ctx, dma);
545	else
546		dma_pool_free(xhci->small_streams_pool, stream_ctx, dma);
547}
548
549/*
550 * The stream context array for each endpoint with bulk streams enabled can
551 * vary in size, based on:
552 *  - how many streams the endpoint supports,
553 *  - the maximum primary stream array size the host controller supports,
554 *  - and how many streams the device driver asks for.
555 *
556 * The stream context array must be a power of 2, and can be as small as
557 * 64 bytes or as large as 1MB.
558 */
559static struct xhci_stream_ctx *xhci_alloc_stream_ctx(struct xhci_hcd *xhci,
560		unsigned int num_stream_ctxs, dma_addr_t *dma,
561		gfp_t mem_flags)
562{
563	struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
564	size_t size = size_mul(sizeof(struct xhci_stream_ctx), num_stream_ctxs);
565
566	if (size > MEDIUM_STREAM_ARRAY_SIZE)
567		return dma_alloc_coherent(dev, size, dma, mem_flags);
568	if (size > SMALL_STREAM_ARRAY_SIZE)
569		return dma_pool_zalloc(xhci->medium_streams_pool, mem_flags, dma);
570	else
571		return dma_pool_zalloc(xhci->small_streams_pool, mem_flags, dma);
572}
573
574struct xhci_ring *xhci_dma_to_transfer_ring(
575		struct xhci_virt_ep *ep,
576		u64 address)
577{
578	if (ep->ep_state & EP_HAS_STREAMS)
579		return radix_tree_lookup(&ep->stream_info->trb_address_map,
580				address >> TRB_SEGMENT_SHIFT);
581	return ep->ring;
582}
583
584/*
585 * Change an endpoint's internal structure so it supports stream IDs.  The
586 * number of requested streams includes stream 0, which cannot be used by device
587 * drivers.
588 *
589 * The number of stream contexts in the stream context array may be bigger than
590 * the number of streams the driver wants to use.  This is because the number of
591 * stream context array entries must be a power of two.
592 */
593struct xhci_stream_info *xhci_alloc_stream_info(struct xhci_hcd *xhci,
594		unsigned int num_stream_ctxs,
595		unsigned int num_streams,
596		unsigned int max_packet, gfp_t mem_flags)
597{
598	struct xhci_stream_info *stream_info;
599	u32 cur_stream;
600	struct xhci_ring *cur_ring;
601	u64 addr;
602	int ret;
603	struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
604
605	xhci_dbg(xhci, "Allocating %u streams and %u stream context array entries.\n",
606			num_streams, num_stream_ctxs);
607	if (xhci->cmd_ring_reserved_trbs == MAX_RSVD_CMD_TRBS) {
608		xhci_dbg(xhci, "Command ring has no reserved TRBs available\n");
609		return NULL;
610	}
611	xhci->cmd_ring_reserved_trbs++;
612
613	stream_info = kzalloc_node(sizeof(*stream_info), mem_flags,
614			dev_to_node(dev));
615	if (!stream_info)
616		goto cleanup_trbs;
617
618	stream_info->num_streams = num_streams;
619	stream_info->num_stream_ctxs = num_stream_ctxs;
620
621	/* Initialize the array of virtual pointers to stream rings. */
622	stream_info->stream_rings = kcalloc_node(
623			num_streams, sizeof(struct xhci_ring *), mem_flags,
624			dev_to_node(dev));
625	if (!stream_info->stream_rings)
626		goto cleanup_info;
627
628	/* Initialize the array of DMA addresses for stream rings for the HW. */
629	stream_info->stream_ctx_array = xhci_alloc_stream_ctx(xhci,
630			num_stream_ctxs, &stream_info->ctx_array_dma,
631			mem_flags);
632	if (!stream_info->stream_ctx_array)
633		goto cleanup_ring_array;
634
635	/* Allocate everything needed to free the stream rings later */
636	stream_info->free_streams_command =
637		xhci_alloc_command_with_ctx(xhci, true, mem_flags);
638	if (!stream_info->free_streams_command)
639		goto cleanup_ctx;
640
641	INIT_RADIX_TREE(&stream_info->trb_address_map, GFP_ATOMIC);
642
643	/* Allocate rings for all the streams that the driver will use,
644	 * and add their segment DMA addresses to the radix tree.
645	 * Stream 0 is reserved.
646	 */
647
648	for (cur_stream = 1; cur_stream < num_streams; cur_stream++) {
649		stream_info->stream_rings[cur_stream] =
650			xhci_ring_alloc(xhci, 2, 1, TYPE_STREAM, max_packet,
651					mem_flags);
652		cur_ring = stream_info->stream_rings[cur_stream];
653		if (!cur_ring)
654			goto cleanup_rings;
655		cur_ring->stream_id = cur_stream;
656		cur_ring->trb_address_map = &stream_info->trb_address_map;
657		/* Set deq ptr, cycle bit, and stream context type */
658		addr = cur_ring->first_seg->dma |
659			SCT_FOR_CTX(SCT_PRI_TR) |
660			cur_ring->cycle_state;
661		stream_info->stream_ctx_array[cur_stream].stream_ring =
662			cpu_to_le64(addr);
663		xhci_dbg(xhci, "Setting stream %d ring ptr to 0x%08llx\n", cur_stream, addr);
664
665		ret = xhci_update_stream_mapping(cur_ring, mem_flags);
666		if (ret) {
667			xhci_ring_free(xhci, cur_ring);
668			stream_info->stream_rings[cur_stream] = NULL;
669			goto cleanup_rings;
670		}
671	}
672	/* Leave the other unused stream ring pointers in the stream context
673	 * array initialized to zero.  This will cause the xHC to give us an
674	 * error if the device asks for a stream ID we don't have setup (if it
675	 * was any other way, the host controller would assume the ring is
676	 * "empty" and wait forever for data to be queued to that stream ID).
677	 */
678
679	return stream_info;
680
681cleanup_rings:
682	for (cur_stream = 1; cur_stream < num_streams; cur_stream++) {
683		cur_ring = stream_info->stream_rings[cur_stream];
684		if (cur_ring) {
685			xhci_ring_free(xhci, cur_ring);
686			stream_info->stream_rings[cur_stream] = NULL;
687		}
688	}
689	xhci_free_command(xhci, stream_info->free_streams_command);
690cleanup_ctx:
691	xhci_free_stream_ctx(xhci,
692		stream_info->num_stream_ctxs,
693		stream_info->stream_ctx_array,
694		stream_info->ctx_array_dma);
695cleanup_ring_array:
696	kfree(stream_info->stream_rings);
697cleanup_info:
698	kfree(stream_info);
699cleanup_trbs:
700	xhci->cmd_ring_reserved_trbs--;
701	return NULL;
702}
703/*
704 * Sets the MaxPStreams field and the Linear Stream Array field.
705 * Sets the dequeue pointer to the stream context array.
706 */
707void xhci_setup_streams_ep_input_ctx(struct xhci_hcd *xhci,
708		struct xhci_ep_ctx *ep_ctx,
709		struct xhci_stream_info *stream_info)
710{
711	u32 max_primary_streams;
712	/* MaxPStreams is the number of stream context array entries, not the
713	 * number we're actually using.  Must be in 2^(MaxPstreams + 1) format.
714	 * fls(0) = 0, fls(0x1) = 1, fls(0x10) = 2, fls(0x100) = 3, etc.
715	 */
716	max_primary_streams = fls(stream_info->num_stream_ctxs) - 2;
717	xhci_dbg_trace(xhci,  trace_xhci_dbg_context_change,
718			"Setting number of stream ctx array entries to %u",
719			1 << (max_primary_streams + 1));
720	ep_ctx->ep_info &= cpu_to_le32(~EP_MAXPSTREAMS_MASK);
721	ep_ctx->ep_info |= cpu_to_le32(EP_MAXPSTREAMS(max_primary_streams)
722				       | EP_HAS_LSA);
723	ep_ctx->deq  = cpu_to_le64(stream_info->ctx_array_dma);
724}
725
726/*
727 * Sets the MaxPStreams field and the Linear Stream Array field to 0.
728 * Reinstalls the "normal" endpoint ring (at its previous dequeue mark,
729 * not at the beginning of the ring).
730 */
731void xhci_setup_no_streams_ep_input_ctx(struct xhci_ep_ctx *ep_ctx,
732		struct xhci_virt_ep *ep)
733{
734	dma_addr_t addr;
735	ep_ctx->ep_info &= cpu_to_le32(~(EP_MAXPSTREAMS_MASK | EP_HAS_LSA));
736	addr = xhci_trb_virt_to_dma(ep->ring->deq_seg, ep->ring->dequeue);
737	ep_ctx->deq  = cpu_to_le64(addr | ep->ring->cycle_state);
738}
739
740/* Frees all stream contexts associated with the endpoint,
741 *
742 * Caller should fix the endpoint context streams fields.
743 */
744void xhci_free_stream_info(struct xhci_hcd *xhci,
745		struct xhci_stream_info *stream_info)
746{
747	int cur_stream;
748	struct xhci_ring *cur_ring;
749
750	if (!stream_info)
751		return;
752
753	for (cur_stream = 1; cur_stream < stream_info->num_streams;
754			cur_stream++) {
755		cur_ring = stream_info->stream_rings[cur_stream];
756		if (cur_ring) {
757			xhci_ring_free(xhci, cur_ring);
758			stream_info->stream_rings[cur_stream] = NULL;
759		}
760	}
761	xhci_free_command(xhci, stream_info->free_streams_command);
762	xhci->cmd_ring_reserved_trbs--;
763	if (stream_info->stream_ctx_array)
764		xhci_free_stream_ctx(xhci,
765				stream_info->num_stream_ctxs,
766				stream_info->stream_ctx_array,
767				stream_info->ctx_array_dma);
768
769	kfree(stream_info->stream_rings);
770	kfree(stream_info);
771}
772
773
774/***************** Device context manipulation *************************/
775
776static void xhci_free_tt_info(struct xhci_hcd *xhci,
777		struct xhci_virt_device *virt_dev,
778		int slot_id)
779{
780	struct list_head *tt_list_head;
781	struct xhci_tt_bw_info *tt_info, *next;
782	bool slot_found = false;
783
784	/* If the device never made it past the Set Address stage,
785	 * it may not have the root hub port pointer set correctly.
786	 */
787	if (!virt_dev->rhub_port) {
788		xhci_dbg(xhci, "Bad rhub port.\n");
789		return;
790	}
791
792	tt_list_head = &(xhci->rh_bw[virt_dev->rhub_port->hw_portnum].tts);
793	list_for_each_entry_safe(tt_info, next, tt_list_head, tt_list) {
794		/* Multi-TT hubs will have more than one entry */
795		if (tt_info->slot_id == slot_id) {
796			slot_found = true;
797			list_del(&tt_info->tt_list);
798			kfree(tt_info);
799		} else if (slot_found) {
800			break;
801		}
802	}
803}
804
805int xhci_alloc_tt_info(struct xhci_hcd *xhci,
806		struct xhci_virt_device *virt_dev,
807		struct usb_device *hdev,
808		struct usb_tt *tt, gfp_t mem_flags)
809{
810	struct xhci_tt_bw_info		*tt_info;
811	unsigned int			num_ports;
812	int				i, j;
813	struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
814
815	if (!tt->multi)
816		num_ports = 1;
817	else
818		num_ports = hdev->maxchild;
819
820	for (i = 0; i < num_ports; i++, tt_info++) {
821		struct xhci_interval_bw_table *bw_table;
822
823		tt_info = kzalloc_node(sizeof(*tt_info), mem_flags,
824				dev_to_node(dev));
825		if (!tt_info)
826			goto free_tts;
827		INIT_LIST_HEAD(&tt_info->tt_list);
828		list_add(&tt_info->tt_list,
829				&xhci->rh_bw[virt_dev->rhub_port->hw_portnum].tts);
830		tt_info->slot_id = virt_dev->udev->slot_id;
831		if (tt->multi)
832			tt_info->ttport = i+1;
833		bw_table = &tt_info->bw_table;
834		for (j = 0; j < XHCI_MAX_INTERVAL; j++)
835			INIT_LIST_HEAD(&bw_table->interval_bw[j].endpoints);
836	}
837	return 0;
838
839free_tts:
840	xhci_free_tt_info(xhci, virt_dev, virt_dev->udev->slot_id);
841	return -ENOMEM;
842}
843
844
845/* All the xhci_tds in the ring's TD list should be freed at this point.
846 * Should be called with xhci->lock held if there is any chance the TT lists
847 * will be manipulated by the configure endpoint, allocate device, or update
848 * hub functions while this function is removing the TT entries from the list.
849 */
850void xhci_free_virt_device(struct xhci_hcd *xhci, int slot_id)
851{
852	struct xhci_virt_device *dev;
853	int i;
854	int old_active_eps = 0;
855
856	/* Slot ID 0 is reserved */
857	if (slot_id == 0 || !xhci->devs[slot_id])
858		return;
859
860	dev = xhci->devs[slot_id];
861
862	xhci->dcbaa->dev_context_ptrs[slot_id] = 0;
863	if (!dev)
864		return;
865
866	trace_xhci_free_virt_device(dev);
867
868	if (dev->tt_info)
869		old_active_eps = dev->tt_info->active_eps;
870
871	for (i = 0; i < 31; i++) {
872		if (dev->eps[i].ring)
873			xhci_ring_free(xhci, dev->eps[i].ring);
874		if (dev->eps[i].stream_info)
875			xhci_free_stream_info(xhci,
876					dev->eps[i].stream_info);
877		/*
878		 * Endpoints are normally deleted from the bandwidth list when
879		 * endpoints are dropped, before device is freed.
880		 * If host is dying or being removed then endpoints aren't
881		 * dropped cleanly, so delete the endpoint from list here.
882		 * Only applicable for hosts with software bandwidth checking.
883		 */
884
885		if (!list_empty(&dev->eps[i].bw_endpoint_list)) {
886			list_del_init(&dev->eps[i].bw_endpoint_list);
887			xhci_dbg(xhci, "Slot %u endpoint %u not removed from BW list!\n",
888				 slot_id, i);
889		}
890	}
891	/* If this is a hub, free the TT(s) from the TT list */
892	xhci_free_tt_info(xhci, dev, slot_id);
893	/* If necessary, update the number of active TTs on this root port */
894	xhci_update_tt_active_eps(xhci, dev, old_active_eps);
895
896	if (dev->in_ctx)
897		xhci_free_container_ctx(xhci, dev->in_ctx);
898	if (dev->out_ctx)
899		xhci_free_container_ctx(xhci, dev->out_ctx);
900
901	if (dev->udev && dev->udev->slot_id)
902		dev->udev->slot_id = 0;
903	if (dev->rhub_port && dev->rhub_port->slot_id == slot_id)
904		dev->rhub_port->slot_id = 0;
905	kfree(xhci->devs[slot_id]);
906	xhci->devs[slot_id] = NULL;
907}
908
909/*
910 * Free a virt_device structure.
911 * If the virt_device added a tt_info (a hub) and has children pointing to
912 * that tt_info, then free the child first. Recursive.
913 * We can't rely on udev at this point to find child-parent relationships.
914 */
915static void xhci_free_virt_devices_depth_first(struct xhci_hcd *xhci, int slot_id)
916{
917	struct xhci_virt_device *vdev;
918	struct list_head *tt_list_head;
919	struct xhci_tt_bw_info *tt_info, *next;
920	int i;
921
922	vdev = xhci->devs[slot_id];
923	if (!vdev)
924		return;
925
926	if (!vdev->rhub_port) {
927		xhci_dbg(xhci, "Bad rhub port.\n");
928		goto out;
929	}
930
931	tt_list_head = &(xhci->rh_bw[vdev->rhub_port->hw_portnum].tts);
932	list_for_each_entry_safe(tt_info, next, tt_list_head, tt_list) {
933		/* is this a hub device that added a tt_info to the tts list */
934		if (tt_info->slot_id == slot_id) {
935			/* are any devices using this tt_info? */
936			for (i = 1; i < HCS_MAX_SLOTS(xhci->hcs_params1); i++) {
937				vdev = xhci->devs[i];
938				if (vdev && (vdev->tt_info == tt_info))
939					xhci_free_virt_devices_depth_first(
940						xhci, i);
941			}
942		}
943	}
944out:
945	/* we are now at a leaf device */
946	xhci_debugfs_remove_slot(xhci, slot_id);
947	xhci_free_virt_device(xhci, slot_id);
948}
949
950int xhci_alloc_virt_device(struct xhci_hcd *xhci, int slot_id,
951		struct usb_device *udev, gfp_t flags)
952{
953	struct xhci_virt_device *dev;
954	int i;
955
956	/* Slot ID 0 is reserved */
957	if (slot_id == 0 || xhci->devs[slot_id]) {
958		xhci_warn(xhci, "Bad Slot ID %d\n", slot_id);
959		return 0;
960	}
961
962	dev = kzalloc(sizeof(*dev), flags);
963	if (!dev)
964		return 0;
965
966	dev->slot_id = slot_id;
967
968	/* Allocate the (output) device context that will be used in the HC. */
969	dev->out_ctx = xhci_alloc_container_ctx(xhci, XHCI_CTX_TYPE_DEVICE, flags);
970	if (!dev->out_ctx)
971		goto fail;
972
973	xhci_dbg(xhci, "Slot %d output ctx = 0x%pad (dma)\n", slot_id, &dev->out_ctx->dma);
974
975	/* Allocate the (input) device context for address device command */
976	dev->in_ctx = xhci_alloc_container_ctx(xhci, XHCI_CTX_TYPE_INPUT, flags);
977	if (!dev->in_ctx)
978		goto fail;
979
980	xhci_dbg(xhci, "Slot %d input ctx = 0x%pad (dma)\n", slot_id, &dev->in_ctx->dma);
981
982	/* Initialize the cancellation and bandwidth list for each ep */
983	for (i = 0; i < 31; i++) {
984		dev->eps[i].ep_index = i;
985		dev->eps[i].vdev = dev;
986		dev->eps[i].xhci = xhci;
987		INIT_LIST_HEAD(&dev->eps[i].cancelled_td_list);
988		INIT_LIST_HEAD(&dev->eps[i].bw_endpoint_list);
989	}
990
991	/* Allocate endpoint 0 ring */
992	dev->eps[0].ring = xhci_ring_alloc(xhci, 2, 1, TYPE_CTRL, 0, flags);
993	if (!dev->eps[0].ring)
994		goto fail;
995
996	dev->udev = udev;
997
998	/* Point to output device context in dcbaa. */
999	xhci->dcbaa->dev_context_ptrs[slot_id] = cpu_to_le64(dev->out_ctx->dma);
1000	xhci_dbg(xhci, "Set slot id %d dcbaa entry %p to 0x%llx\n",
1001		 slot_id,
1002		 &xhci->dcbaa->dev_context_ptrs[slot_id],
1003		 le64_to_cpu(xhci->dcbaa->dev_context_ptrs[slot_id]));
1004
1005	trace_xhci_alloc_virt_device(dev);
1006
1007	xhci->devs[slot_id] = dev;
1008
1009	return 1;
1010fail:
1011
1012	if (dev->in_ctx)
1013		xhci_free_container_ctx(xhci, dev->in_ctx);
1014	if (dev->out_ctx)
1015		xhci_free_container_ctx(xhci, dev->out_ctx);
1016	kfree(dev);
1017
1018	return 0;
1019}
1020
1021void xhci_copy_ep0_dequeue_into_input_ctx(struct xhci_hcd *xhci,
1022		struct usb_device *udev)
1023{
1024	struct xhci_virt_device *virt_dev;
1025	struct xhci_ep_ctx	*ep0_ctx;
1026	struct xhci_ring	*ep_ring;
1027
1028	virt_dev = xhci->devs[udev->slot_id];
1029	ep0_ctx = xhci_get_ep_ctx(xhci, virt_dev->in_ctx, 0);
1030	ep_ring = virt_dev->eps[0].ring;
1031	/*
1032	 * FIXME we don't keep track of the dequeue pointer very well after a
1033	 * Set TR dequeue pointer, so we're setting the dequeue pointer of the
1034	 * host to our enqueue pointer.  This should only be called after a
1035	 * configured device has reset, so all control transfers should have
1036	 * been completed or cancelled before the reset.
1037	 */
1038	ep0_ctx->deq = cpu_to_le64(xhci_trb_virt_to_dma(ep_ring->enq_seg,
1039							ep_ring->enqueue)
1040				   | ep_ring->cycle_state);
1041}
1042
1043/*
1044 * The xHCI roothub may have ports of differing speeds in any order in the port
1045 * status registers.
1046 *
1047 * The xHCI hardware wants to know the roothub port that the USB device
1048 * is attached to (or the roothub port its ancestor hub is attached to).  All we
1049 * know is the index of that port under either the USB 2.0 or the USB 3.0
1050 * roothub, but that doesn't give us the real index into the HW port status
1051 * registers.
1052 */
1053static struct xhci_port *xhci_find_rhub_port(struct xhci_hcd *xhci, struct usb_device *udev)
1054{
1055	struct usb_device *top_dev;
1056	struct xhci_hub *rhub;
1057	struct usb_hcd *hcd;
1058
1059	if (udev->speed >= USB_SPEED_SUPER)
1060		hcd = xhci_get_usb3_hcd(xhci);
1061	else
1062		hcd = xhci->main_hcd;
1063
1064	for (top_dev = udev; top_dev->parent && top_dev->parent->parent;
1065			top_dev = top_dev->parent)
1066		/* Found device below root hub */;
1067
1068	rhub = xhci_get_rhub(hcd);
1069	return rhub->ports[top_dev->portnum - 1];
1070}
1071
1072/* Setup an xHCI virtual device for a Set Address command */
1073int xhci_setup_addressable_virt_dev(struct xhci_hcd *xhci, struct usb_device *udev)
1074{
1075	struct xhci_virt_device *dev;
1076	struct xhci_ep_ctx	*ep0_ctx;
1077	struct xhci_slot_ctx    *slot_ctx;
1078	u32			max_packets;
1079
1080	dev = xhci->devs[udev->slot_id];
1081	/* Slot ID 0 is reserved */
1082	if (udev->slot_id == 0 || !dev) {
1083		xhci_warn(xhci, "Slot ID %d is not assigned to this device\n",
1084				udev->slot_id);
1085		return -EINVAL;
1086	}
1087	ep0_ctx = xhci_get_ep_ctx(xhci, dev->in_ctx, 0);
1088	slot_ctx = xhci_get_slot_ctx(xhci, dev->in_ctx);
1089
1090	/* 3) Only the control endpoint is valid - one endpoint context */
1091	slot_ctx->dev_info |= cpu_to_le32(LAST_CTX(1) | udev->route);
1092	switch (udev->speed) {
1093	case USB_SPEED_SUPER_PLUS:
1094		slot_ctx->dev_info |= cpu_to_le32(SLOT_SPEED_SSP);
1095		max_packets = MAX_PACKET(512);
1096		break;
1097	case USB_SPEED_SUPER:
1098		slot_ctx->dev_info |= cpu_to_le32(SLOT_SPEED_SS);
1099		max_packets = MAX_PACKET(512);
1100		break;
1101	case USB_SPEED_HIGH:
1102		slot_ctx->dev_info |= cpu_to_le32(SLOT_SPEED_HS);
1103		max_packets = MAX_PACKET(64);
1104		break;
1105	/* USB core guesses at a 64-byte max packet first for FS devices */
1106	case USB_SPEED_FULL:
1107		slot_ctx->dev_info |= cpu_to_le32(SLOT_SPEED_FS);
1108		max_packets = MAX_PACKET(64);
1109		break;
1110	case USB_SPEED_LOW:
1111		slot_ctx->dev_info |= cpu_to_le32(SLOT_SPEED_LS);
1112		max_packets = MAX_PACKET(8);
1113		break;
1114	default:
1115		/* Speed was set earlier, this shouldn't happen. */
1116		return -EINVAL;
1117	}
1118	/* Find the root hub port this device is under */
1119	dev->rhub_port = xhci_find_rhub_port(xhci, udev);
1120	if (!dev->rhub_port)
1121		return -EINVAL;
1122	/* Slot ID is set to the device directly below the root hub */
1123	if (!udev->parent->parent)
1124		dev->rhub_port->slot_id = udev->slot_id;
1125	slot_ctx->dev_info2 |= cpu_to_le32(ROOT_HUB_PORT(dev->rhub_port->hw_portnum + 1));
1126	xhci_dbg(xhci, "Slot ID %d: HW portnum %d, hcd portnum %d\n",
1127		 udev->slot_id, dev->rhub_port->hw_portnum, dev->rhub_port->hcd_portnum);
1128
1129	/* Find the right bandwidth table that this device will be a part of.
1130	 * If this is a full speed device attached directly to a root port (or a
1131	 * decendent of one), it counts as a primary bandwidth domain, not a
1132	 * secondary bandwidth domain under a TT.  An xhci_tt_info structure
1133	 * will never be created for the HS root hub.
1134	 */
1135	if (!udev->tt || !udev->tt->hub->parent) {
1136		dev->bw_table = &xhci->rh_bw[dev->rhub_port->hw_portnum].bw_table;
1137	} else {
1138		struct xhci_root_port_bw_info *rh_bw;
1139		struct xhci_tt_bw_info *tt_bw;
1140
1141		rh_bw = &xhci->rh_bw[dev->rhub_port->hw_portnum];
1142		/* Find the right TT. */
1143		list_for_each_entry(tt_bw, &rh_bw->tts, tt_list) {
1144			if (tt_bw->slot_id != udev->tt->hub->slot_id)
1145				continue;
1146
1147			if (!dev->udev->tt->multi ||
1148					(udev->tt->multi &&
1149					 tt_bw->ttport == dev->udev->ttport)) {
1150				dev->bw_table = &tt_bw->bw_table;
1151				dev->tt_info = tt_bw;
1152				break;
1153			}
1154		}
1155		if (!dev->tt_info)
1156			xhci_warn(xhci, "WARN: Didn't find a matching TT\n");
1157	}
1158
1159	/* Is this a LS/FS device under an external HS hub? */
1160	if (udev->tt && udev->tt->hub->parent) {
1161		slot_ctx->tt_info = cpu_to_le32(udev->tt->hub->slot_id |
1162						(udev->ttport << 8));
1163		if (udev->tt->multi)
1164			slot_ctx->dev_info |= cpu_to_le32(DEV_MTT);
1165	}
1166	xhci_dbg(xhci, "udev->tt = %p\n", udev->tt);
1167	xhci_dbg(xhci, "udev->ttport = 0x%x\n", udev->ttport);
1168
1169	/* Step 4 - ring already allocated */
1170	/* Step 5 */
1171	ep0_ctx->ep_info2 = cpu_to_le32(EP_TYPE(CTRL_EP));
1172
1173	/* EP 0 can handle "burst" sizes of 1, so Max Burst Size field is 0 */
1174	ep0_ctx->ep_info2 |= cpu_to_le32(MAX_BURST(0) | ERROR_COUNT(3) |
1175					 max_packets);
1176
1177	ep0_ctx->deq = cpu_to_le64(dev->eps[0].ring->first_seg->dma |
1178				   dev->eps[0].ring->cycle_state);
1179
1180	trace_xhci_setup_addressable_virt_device(dev);
1181
1182	/* Steps 7 and 8 were done in xhci_alloc_virt_device() */
1183
1184	return 0;
1185}
1186
1187/*
1188 * Convert interval expressed as 2^(bInterval - 1) == interval into
1189 * straight exponent value 2^n == interval.
1190 *
1191 */
1192static unsigned int xhci_parse_exponent_interval(struct usb_device *udev,
1193		struct usb_host_endpoint *ep)
1194{
1195	unsigned int interval;
1196
1197	interval = clamp_val(ep->desc.bInterval, 1, 16) - 1;
1198	if (interval != ep->desc.bInterval - 1)
1199		dev_warn(&udev->dev,
1200			 "ep %#x - rounding interval to %d %sframes\n",
1201			 ep->desc.bEndpointAddress,
1202			 1 << interval,
1203			 udev->speed == USB_SPEED_FULL ? "" : "micro");
1204
1205	if (udev->speed == USB_SPEED_FULL) {
1206		/*
1207		 * Full speed isoc endpoints specify interval in frames,
1208		 * not microframes. We are using microframes everywhere,
1209		 * so adjust accordingly.
1210		 */
1211		interval += 3;	/* 1 frame = 2^3 uframes */
1212	}
1213
1214	return interval;
1215}
1216
1217/*
1218 * Convert bInterval expressed in microframes (in 1-255 range) to exponent of
1219 * microframes, rounded down to nearest power of 2.
1220 */
1221static unsigned int xhci_microframes_to_exponent(struct usb_device *udev,
1222		struct usb_host_endpoint *ep, unsigned int desc_interval,
1223		unsigned int min_exponent, unsigned int max_exponent)
1224{
1225	unsigned int interval;
1226
1227	interval = fls(desc_interval) - 1;
1228	interval = clamp_val(interval, min_exponent, max_exponent);
1229	if ((1 << interval) != desc_interval)
1230		dev_dbg(&udev->dev,
1231			 "ep %#x - rounding interval to %d microframes, ep desc says %d microframes\n",
1232			 ep->desc.bEndpointAddress,
1233			 1 << interval,
1234			 desc_interval);
1235
1236	return interval;
1237}
1238
1239static unsigned int xhci_parse_microframe_interval(struct usb_device *udev,
1240		struct usb_host_endpoint *ep)
1241{
1242	if (ep->desc.bInterval == 0)
1243		return 0;
1244	return xhci_microframes_to_exponent(udev, ep,
1245			ep->desc.bInterval, 0, 15);
1246}
1247
1248
1249static unsigned int xhci_parse_frame_interval(struct usb_device *udev,
1250		struct usb_host_endpoint *ep)
1251{
1252	return xhci_microframes_to_exponent(udev, ep,
1253			ep->desc.bInterval * 8, 3, 10);
1254}
1255
1256/* Return the polling or NAK interval.
1257 *
1258 * The polling interval is expressed in "microframes".  If xHCI's Interval field
1259 * is set to N, it will service the endpoint every 2^(Interval)*125us.
1260 *
1261 * The NAK interval is one NAK per 1 to 255 microframes, or no NAKs if interval
1262 * is set to 0.
1263 */
1264static unsigned int xhci_get_endpoint_interval(struct usb_device *udev,
1265		struct usb_host_endpoint *ep)
1266{
1267	unsigned int interval = 0;
1268
1269	switch (udev->speed) {
1270	case USB_SPEED_HIGH:
1271		/* Max NAK rate */
1272		if (usb_endpoint_xfer_control(&ep->desc) ||
1273		    usb_endpoint_xfer_bulk(&ep->desc)) {
1274			interval = xhci_parse_microframe_interval(udev, ep);
1275			break;
1276		}
1277		fallthrough;	/* SS and HS isoc/int have same decoding */
1278
1279	case USB_SPEED_SUPER_PLUS:
1280	case USB_SPEED_SUPER:
1281		if (usb_endpoint_xfer_int(&ep->desc) ||
1282		    usb_endpoint_xfer_isoc(&ep->desc)) {
1283			interval = xhci_parse_exponent_interval(udev, ep);
1284		}
1285		break;
1286
1287	case USB_SPEED_FULL:
1288		if (usb_endpoint_xfer_isoc(&ep->desc)) {
1289			interval = xhci_parse_exponent_interval(udev, ep);
1290			break;
1291		}
1292		/*
1293		 * Fall through for interrupt endpoint interval decoding
1294		 * since it uses the same rules as low speed interrupt
1295		 * endpoints.
1296		 */
1297		fallthrough;
1298
1299	case USB_SPEED_LOW:
1300		if (usb_endpoint_xfer_int(&ep->desc) ||
1301		    usb_endpoint_xfer_isoc(&ep->desc)) {
1302
1303			interval = xhci_parse_frame_interval(udev, ep);
1304		}
1305		break;
1306
1307	default:
1308		BUG();
1309	}
1310	return interval;
1311}
1312
1313/* The "Mult" field in the endpoint context is only set for SuperSpeed isoc eps.
1314 * High speed endpoint descriptors can define "the number of additional
1315 * transaction opportunities per microframe", but that goes in the Max Burst
1316 * endpoint context field.
1317 */
1318static u32 xhci_get_endpoint_mult(struct usb_device *udev,
1319		struct usb_host_endpoint *ep)
1320{
1321	if (udev->speed < USB_SPEED_SUPER ||
1322			!usb_endpoint_xfer_isoc(&ep->desc))
1323		return 0;
1324	return ep->ss_ep_comp.bmAttributes;
1325}
1326
1327static u32 xhci_get_endpoint_max_burst(struct usb_device *udev,
1328				       struct usb_host_endpoint *ep)
1329{
1330	/* Super speed and Plus have max burst in ep companion desc */
1331	if (udev->speed >= USB_SPEED_SUPER)
1332		return ep->ss_ep_comp.bMaxBurst;
1333
1334	if (udev->speed == USB_SPEED_HIGH &&
1335	    (usb_endpoint_xfer_isoc(&ep->desc) ||
1336	     usb_endpoint_xfer_int(&ep->desc)))
1337		return usb_endpoint_maxp_mult(&ep->desc) - 1;
1338
1339	return 0;
1340}
1341
1342static u32 xhci_get_endpoint_type(struct usb_host_endpoint *ep)
1343{
1344	int in;
1345
1346	in = usb_endpoint_dir_in(&ep->desc);
1347
1348	switch (usb_endpoint_type(&ep->desc)) {
1349	case USB_ENDPOINT_XFER_CONTROL:
1350		return CTRL_EP;
1351	case USB_ENDPOINT_XFER_BULK:
1352		return in ? BULK_IN_EP : BULK_OUT_EP;
1353	case USB_ENDPOINT_XFER_ISOC:
1354		return in ? ISOC_IN_EP : ISOC_OUT_EP;
1355	case USB_ENDPOINT_XFER_INT:
1356		return in ? INT_IN_EP : INT_OUT_EP;
1357	}
1358	return 0;
1359}
1360
1361/* Return the maximum endpoint service interval time (ESIT) payload.
1362 * Basically, this is the maxpacket size, multiplied by the burst size
1363 * and mult size.
1364 */
1365static u32 xhci_get_max_esit_payload(struct usb_device *udev,
1366		struct usb_host_endpoint *ep)
1367{
1368	int max_burst;
1369	int max_packet;
1370
1371	/* Only applies for interrupt or isochronous endpoints */
1372	if (usb_endpoint_xfer_control(&ep->desc) ||
1373			usb_endpoint_xfer_bulk(&ep->desc))
1374		return 0;
1375
1376	/* SuperSpeedPlus Isoc ep sending over 48k per esit */
1377	if ((udev->speed >= USB_SPEED_SUPER_PLUS) &&
1378	    USB_SS_SSP_ISOC_COMP(ep->ss_ep_comp.bmAttributes))
1379		return le32_to_cpu(ep->ssp_isoc_ep_comp.dwBytesPerInterval);
1380
1381	/* SuperSpeed or SuperSpeedPlus Isoc ep with less than 48k per esit */
1382	if (udev->speed >= USB_SPEED_SUPER)
1383		return le16_to_cpu(ep->ss_ep_comp.wBytesPerInterval);
1384
1385	max_packet = usb_endpoint_maxp(&ep->desc);
1386	max_burst = usb_endpoint_maxp_mult(&ep->desc);
1387	/* A 0 in max burst means 1 transfer per ESIT */
1388	return max_packet * max_burst;
1389}
1390
1391/* Set up an endpoint with one ring segment.  Do not allocate stream rings.
1392 * Drivers will have to call usb_alloc_streams() to do that.
1393 */
1394int xhci_endpoint_init(struct xhci_hcd *xhci,
1395		struct xhci_virt_device *virt_dev,
1396		struct usb_device *udev,
1397		struct usb_host_endpoint *ep,
1398		gfp_t mem_flags)
1399{
1400	unsigned int ep_index;
1401	struct xhci_ep_ctx *ep_ctx;
1402	struct xhci_ring *ep_ring;
1403	unsigned int max_packet;
1404	enum xhci_ring_type ring_type;
1405	u32 max_esit_payload;
1406	u32 endpoint_type;
1407	unsigned int max_burst;
1408	unsigned int interval;
1409	unsigned int mult;
1410	unsigned int avg_trb_len;
1411	unsigned int err_count = 0;
1412
1413	ep_index = xhci_get_endpoint_index(&ep->desc);
1414	ep_ctx = xhci_get_ep_ctx(xhci, virt_dev->in_ctx, ep_index);
1415
1416	endpoint_type = xhci_get_endpoint_type(ep);
1417	if (!endpoint_type)
1418		return -EINVAL;
1419
1420	ring_type = usb_endpoint_type(&ep->desc);
1421
1422	/*
1423	 * Get values to fill the endpoint context, mostly from ep descriptor.
1424	 * The average TRB buffer lengt for bulk endpoints is unclear as we
1425	 * have no clue on scatter gather list entry size. For Isoc and Int,
1426	 * set it to max available. See xHCI 1.1 spec 4.14.1.1 for details.
1427	 */
1428	max_esit_payload = xhci_get_max_esit_payload(udev, ep);
1429	interval = xhci_get_endpoint_interval(udev, ep);
1430
1431	/* Periodic endpoint bInterval limit quirk */
1432	if (usb_endpoint_xfer_int(&ep->desc) ||
1433	    usb_endpoint_xfer_isoc(&ep->desc)) {
1434		if ((xhci->quirks & XHCI_LIMIT_ENDPOINT_INTERVAL_7) &&
1435		    udev->speed >= USB_SPEED_HIGH &&
1436		    interval >= 7) {
1437			interval = 6;
1438		}
1439	}
1440
1441	mult = xhci_get_endpoint_mult(udev, ep);
1442	max_packet = usb_endpoint_maxp(&ep->desc);
1443	max_burst = xhci_get_endpoint_max_burst(udev, ep);
1444	avg_trb_len = max_esit_payload;
1445
1446	/* FIXME dig Mult and streams info out of ep companion desc */
1447
1448	/* Allow 3 retries for everything but isoc, set CErr = 3 */
1449	if (!usb_endpoint_xfer_isoc(&ep->desc))
1450		err_count = 3;
1451	/* HS bulk max packet should be 512, FS bulk supports 8, 16, 32 or 64 */
1452	if (usb_endpoint_xfer_bulk(&ep->desc)) {
1453		if (udev->speed == USB_SPEED_HIGH)
1454			max_packet = 512;
1455		if (udev->speed == USB_SPEED_FULL) {
1456			max_packet = rounddown_pow_of_two(max_packet);
1457			max_packet = clamp_val(max_packet, 8, 64);
1458		}
1459	}
1460	/* xHCI 1.0 and 1.1 indicates that ctrl ep avg TRB Length should be 8 */
1461	if (usb_endpoint_xfer_control(&ep->desc) && xhci->hci_version >= 0x100)
1462		avg_trb_len = 8;
1463	/* xhci 1.1 with LEC support doesn't use mult field, use RsvdZ */
1464	if ((xhci->hci_version > 0x100) && HCC2_LEC(xhci->hcc_params2))
1465		mult = 0;
1466
1467	/* Set up the endpoint ring */
1468	virt_dev->eps[ep_index].new_ring =
1469		xhci_ring_alloc(xhci, 2, 1, ring_type, max_packet, mem_flags);
1470	if (!virt_dev->eps[ep_index].new_ring)
1471		return -ENOMEM;
1472
1473	virt_dev->eps[ep_index].skip = false;
1474	ep_ring = virt_dev->eps[ep_index].new_ring;
1475
1476	/* Fill the endpoint context */
1477	ep_ctx->ep_info = cpu_to_le32(EP_MAX_ESIT_PAYLOAD_HI(max_esit_payload) |
1478				      EP_INTERVAL(interval) |
1479				      EP_MULT(mult));
1480	ep_ctx->ep_info2 = cpu_to_le32(EP_TYPE(endpoint_type) |
1481				       MAX_PACKET(max_packet) |
1482				       MAX_BURST(max_burst) |
1483				       ERROR_COUNT(err_count));
1484	ep_ctx->deq = cpu_to_le64(ep_ring->first_seg->dma |
1485				  ep_ring->cycle_state);
1486
1487	ep_ctx->tx_info = cpu_to_le32(EP_MAX_ESIT_PAYLOAD_LO(max_esit_payload) |
1488				      EP_AVG_TRB_LENGTH(avg_trb_len));
1489
1490	return 0;
1491}
1492
1493void xhci_endpoint_zero(struct xhci_hcd *xhci,
1494		struct xhci_virt_device *virt_dev,
1495		struct usb_host_endpoint *ep)
1496{
1497	unsigned int ep_index;
1498	struct xhci_ep_ctx *ep_ctx;
1499
1500	ep_index = xhci_get_endpoint_index(&ep->desc);
1501	ep_ctx = xhci_get_ep_ctx(xhci, virt_dev->in_ctx, ep_index);
1502
1503	ep_ctx->ep_info = 0;
1504	ep_ctx->ep_info2 = 0;
1505	ep_ctx->deq = 0;
1506	ep_ctx->tx_info = 0;
1507	/* Don't free the endpoint ring until the set interface or configuration
1508	 * request succeeds.
1509	 */
1510}
1511
1512void xhci_clear_endpoint_bw_info(struct xhci_bw_info *bw_info)
1513{
1514	bw_info->ep_interval = 0;
1515	bw_info->mult = 0;
1516	bw_info->num_packets = 0;
1517	bw_info->max_packet_size = 0;
1518	bw_info->type = 0;
1519	bw_info->max_esit_payload = 0;
1520}
1521
1522void xhci_update_bw_info(struct xhci_hcd *xhci,
1523		struct xhci_container_ctx *in_ctx,
1524		struct xhci_input_control_ctx *ctrl_ctx,
1525		struct xhci_virt_device *virt_dev)
1526{
1527	struct xhci_bw_info *bw_info;
1528	struct xhci_ep_ctx *ep_ctx;
1529	unsigned int ep_type;
1530	int i;
1531
1532	for (i = 1; i < 31; i++) {
1533		bw_info = &virt_dev->eps[i].bw_info;
1534
1535		/* We can't tell what endpoint type is being dropped, but
1536		 * unconditionally clearing the bandwidth info for non-periodic
1537		 * endpoints should be harmless because the info will never be
1538		 * set in the first place.
1539		 */
1540		if (!EP_IS_ADDED(ctrl_ctx, i) && EP_IS_DROPPED(ctrl_ctx, i)) {
1541			/* Dropped endpoint */
1542			xhci_clear_endpoint_bw_info(bw_info);
1543			continue;
1544		}
1545
1546		if (EP_IS_ADDED(ctrl_ctx, i)) {
1547			ep_ctx = xhci_get_ep_ctx(xhci, in_ctx, i);
1548			ep_type = CTX_TO_EP_TYPE(le32_to_cpu(ep_ctx->ep_info2));
1549
1550			/* Ignore non-periodic endpoints */
1551			if (ep_type != ISOC_OUT_EP && ep_type != INT_OUT_EP &&
1552					ep_type != ISOC_IN_EP &&
1553					ep_type != INT_IN_EP)
1554				continue;
1555
1556			/* Added or changed endpoint */
1557			bw_info->ep_interval = CTX_TO_EP_INTERVAL(
1558					le32_to_cpu(ep_ctx->ep_info));
1559			/* Number of packets and mult are zero-based in the
1560			 * input context, but we want one-based for the
1561			 * interval table.
1562			 */
1563			bw_info->mult = CTX_TO_EP_MULT(
1564					le32_to_cpu(ep_ctx->ep_info)) + 1;
1565			bw_info->num_packets = CTX_TO_MAX_BURST(
1566					le32_to_cpu(ep_ctx->ep_info2)) + 1;
1567			bw_info->max_packet_size = MAX_PACKET_DECODED(
1568					le32_to_cpu(ep_ctx->ep_info2));
1569			bw_info->type = ep_type;
1570			bw_info->max_esit_payload = CTX_TO_MAX_ESIT_PAYLOAD(
1571					le32_to_cpu(ep_ctx->tx_info));
1572		}
1573	}
1574}
1575
1576/* Copy output xhci_ep_ctx to the input xhci_ep_ctx copy.
1577 * Useful when you want to change one particular aspect of the endpoint and then
1578 * issue a configure endpoint command.
1579 */
1580void xhci_endpoint_copy(struct xhci_hcd *xhci,
1581		struct xhci_container_ctx *in_ctx,
1582		struct xhci_container_ctx *out_ctx,
1583		unsigned int ep_index)
1584{
1585	struct xhci_ep_ctx *out_ep_ctx;
1586	struct xhci_ep_ctx *in_ep_ctx;
1587
1588	out_ep_ctx = xhci_get_ep_ctx(xhci, out_ctx, ep_index);
1589	in_ep_ctx = xhci_get_ep_ctx(xhci, in_ctx, ep_index);
1590
1591	in_ep_ctx->ep_info = out_ep_ctx->ep_info;
1592	in_ep_ctx->ep_info2 = out_ep_ctx->ep_info2;
1593	in_ep_ctx->deq = out_ep_ctx->deq;
1594	in_ep_ctx->tx_info = out_ep_ctx->tx_info;
1595	if (xhci->quirks & XHCI_MTK_HOST) {
1596		in_ep_ctx->reserved[0] = out_ep_ctx->reserved[0];
1597		in_ep_ctx->reserved[1] = out_ep_ctx->reserved[1];
1598	}
1599}
1600
1601/* Copy output xhci_slot_ctx to the input xhci_slot_ctx.
1602 * Useful when you want to change one particular aspect of the endpoint and then
1603 * issue a configure endpoint command.  Only the context entries field matters,
1604 * but we'll copy the whole thing anyway.
1605 */
1606void xhci_slot_copy(struct xhci_hcd *xhci,
1607		struct xhci_container_ctx *in_ctx,
1608		struct xhci_container_ctx *out_ctx)
1609{
1610	struct xhci_slot_ctx *in_slot_ctx;
1611	struct xhci_slot_ctx *out_slot_ctx;
1612
1613	in_slot_ctx = xhci_get_slot_ctx(xhci, in_ctx);
1614	out_slot_ctx = xhci_get_slot_ctx(xhci, out_ctx);
1615
1616	in_slot_ctx->dev_info = out_slot_ctx->dev_info;
1617	in_slot_ctx->dev_info2 = out_slot_ctx->dev_info2;
1618	in_slot_ctx->tt_info = out_slot_ctx->tt_info;
1619	in_slot_ctx->dev_state = out_slot_ctx->dev_state;
1620}
1621
1622/* Set up the scratchpad buffer array and scratchpad buffers, if needed. */
1623static int scratchpad_alloc(struct xhci_hcd *xhci, gfp_t flags)
1624{
1625	int i;
1626	struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
1627	int num_sp = HCS_MAX_SCRATCHPAD(xhci->hcs_params2);
1628
1629	xhci_dbg_trace(xhci, trace_xhci_dbg_init,
1630			"Allocating %d scratchpad buffers", num_sp);
1631
1632	if (!num_sp)
1633		return 0;
1634
1635	xhci->scratchpad = kzalloc_node(sizeof(*xhci->scratchpad), flags,
1636				dev_to_node(dev));
1637	if (!xhci->scratchpad)
1638		goto fail_sp;
1639
1640	xhci->scratchpad->sp_array = dma_alloc_coherent(dev,
1641				     size_mul(sizeof(u64), num_sp),
1642				     &xhci->scratchpad->sp_dma, flags);
1643	if (!xhci->scratchpad->sp_array)
1644		goto fail_sp2;
1645
1646	xhci->scratchpad->sp_buffers = kcalloc_node(num_sp, sizeof(void *),
1647					flags, dev_to_node(dev));
1648	if (!xhci->scratchpad->sp_buffers)
1649		goto fail_sp3;
1650
1651	xhci->dcbaa->dev_context_ptrs[0] = cpu_to_le64(xhci->scratchpad->sp_dma);
1652	for (i = 0; i < num_sp; i++) {
1653		dma_addr_t dma;
1654		void *buf = dma_alloc_coherent(dev, xhci->page_size, &dma,
1655					       flags);
1656		if (!buf)
1657			goto fail_sp4;
1658
1659		xhci->scratchpad->sp_array[i] = dma;
1660		xhci->scratchpad->sp_buffers[i] = buf;
1661	}
1662
1663	return 0;
1664
1665 fail_sp4:
1666	while (i--)
1667		dma_free_coherent(dev, xhci->page_size,
1668				    xhci->scratchpad->sp_buffers[i],
1669				    xhci->scratchpad->sp_array[i]);
1670
1671	kfree(xhci->scratchpad->sp_buffers);
1672
1673 fail_sp3:
1674	dma_free_coherent(dev, num_sp * sizeof(u64),
1675			    xhci->scratchpad->sp_array,
1676			    xhci->scratchpad->sp_dma);
1677
1678 fail_sp2:
1679	kfree(xhci->scratchpad);
1680	xhci->scratchpad = NULL;
1681
1682 fail_sp:
1683	return -ENOMEM;
1684}
1685
1686static void scratchpad_free(struct xhci_hcd *xhci)
1687{
1688	int num_sp;
1689	int i;
1690	struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
1691
1692	if (!xhci->scratchpad)
1693		return;
1694
1695	num_sp = HCS_MAX_SCRATCHPAD(xhci->hcs_params2);
1696
1697	for (i = 0; i < num_sp; i++) {
1698		dma_free_coherent(dev, xhci->page_size,
1699				    xhci->scratchpad->sp_buffers[i],
1700				    xhci->scratchpad->sp_array[i]);
1701	}
1702	kfree(xhci->scratchpad->sp_buffers);
1703	dma_free_coherent(dev, num_sp * sizeof(u64),
1704			    xhci->scratchpad->sp_array,
1705			    xhci->scratchpad->sp_dma);
1706	kfree(xhci->scratchpad);
1707	xhci->scratchpad = NULL;
1708}
1709
1710struct xhci_command *xhci_alloc_command(struct xhci_hcd *xhci,
1711		bool allocate_completion, gfp_t mem_flags)
1712{
1713	struct xhci_command *command;
1714	struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
1715
1716	command = kzalloc_node(sizeof(*command), mem_flags, dev_to_node(dev));
1717	if (!command)
1718		return NULL;
1719
1720	if (allocate_completion) {
1721		command->completion =
1722			kzalloc_node(sizeof(struct completion), mem_flags,
1723				dev_to_node(dev));
1724		if (!command->completion) {
1725			kfree(command);
1726			return NULL;
1727		}
1728		init_completion(command->completion);
1729	}
1730
1731	command->status = 0;
1732	/* set default timeout to 5000 ms */
1733	command->timeout_ms = XHCI_CMD_DEFAULT_TIMEOUT;
1734	INIT_LIST_HEAD(&command->cmd_list);
1735	return command;
1736}
1737
1738struct xhci_command *xhci_alloc_command_with_ctx(struct xhci_hcd *xhci,
1739		bool allocate_completion, gfp_t mem_flags)
1740{
1741	struct xhci_command *command;
1742
1743	command = xhci_alloc_command(xhci, allocate_completion, mem_flags);
1744	if (!command)
1745		return NULL;
1746
1747	command->in_ctx = xhci_alloc_container_ctx(xhci, XHCI_CTX_TYPE_INPUT,
1748						   mem_flags);
1749	if (!command->in_ctx) {
1750		kfree(command->completion);
1751		kfree(command);
1752		return NULL;
1753	}
1754	return command;
1755}
1756
1757void xhci_urb_free_priv(struct urb_priv *urb_priv)
1758{
1759	kfree(urb_priv);
1760}
1761
1762void xhci_free_command(struct xhci_hcd *xhci,
1763		struct xhci_command *command)
1764{
1765	xhci_free_container_ctx(xhci,
1766			command->in_ctx);
1767	kfree(command->completion);
1768	kfree(command);
1769}
1770
1771static int xhci_alloc_erst(struct xhci_hcd *xhci,
1772		    struct xhci_ring *evt_ring,
1773		    struct xhci_erst *erst,
1774		    gfp_t flags)
1775{
1776	size_t size;
1777	unsigned int val;
1778	struct xhci_segment *seg;
1779	struct xhci_erst_entry *entry;
1780
1781	size = size_mul(sizeof(struct xhci_erst_entry), evt_ring->num_segs);
1782	erst->entries = dma_alloc_coherent(xhci_to_hcd(xhci)->self.sysdev,
1783					   size, &erst->erst_dma_addr, flags);
1784	if (!erst->entries)
1785		return -ENOMEM;
1786
1787	erst->num_entries = evt_ring->num_segs;
1788
1789	seg = evt_ring->first_seg;
1790	for (val = 0; val < evt_ring->num_segs; val++) {
1791		entry = &erst->entries[val];
1792		entry->seg_addr = cpu_to_le64(seg->dma);
1793		entry->seg_size = cpu_to_le32(TRBS_PER_SEGMENT);
1794		entry->rsvd = 0;
1795		seg = seg->next;
1796	}
1797
1798	return 0;
1799}
1800
1801static void
1802xhci_remove_interrupter(struct xhci_hcd *xhci, struct xhci_interrupter *ir)
1803{
1804	u32 tmp;
1805
1806	if (!ir)
1807		return;
1808
1809	/*
1810	 * Clean out interrupter registers except ERSTBA. Clearing either the
1811	 * low or high 32 bits of ERSTBA immediately causes the controller to
1812	 * dereference the partially cleared 64 bit address, causing IOMMU error.
1813	 */
1814	if (ir->ir_set) {
1815		tmp = readl(&ir->ir_set->erst_size);
1816		tmp &= ERST_SIZE_MASK;
1817		writel(tmp, &ir->ir_set->erst_size);
1818
1819		xhci_write_64(xhci, ERST_EHB, &ir->ir_set->erst_dequeue);
1820	}
1821}
1822
1823static void
1824xhci_free_interrupter(struct xhci_hcd *xhci, struct xhci_interrupter *ir)
1825{
1826	struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
1827	size_t erst_size;
1828
1829	if (!ir)
1830		return;
1831
1832	erst_size = sizeof(struct xhci_erst_entry) * ir->erst.num_entries;
1833	if (ir->erst.entries)
1834		dma_free_coherent(dev, erst_size,
1835				  ir->erst.entries,
1836				  ir->erst.erst_dma_addr);
1837	ir->erst.entries = NULL;
1838
1839	/* free interrupter event ring */
1840	if (ir->event_ring)
1841		xhci_ring_free(xhci, ir->event_ring);
1842
1843	ir->event_ring = NULL;
1844
1845	kfree(ir);
1846}
1847
1848void xhci_remove_secondary_interrupter(struct usb_hcd *hcd, struct xhci_interrupter *ir)
1849{
1850	struct xhci_hcd *xhci = hcd_to_xhci(hcd);
1851	unsigned int intr_num;
1852
1853	spin_lock_irq(&xhci->lock);
1854
1855	/* interrupter 0 is primary interrupter, don't touch it */
1856	if (!ir || !ir->intr_num || ir->intr_num >= xhci->max_interrupters) {
1857		xhci_dbg(xhci, "Invalid secondary interrupter, can't remove\n");
1858		spin_unlock_irq(&xhci->lock);
1859		return;
1860	}
1861
1862	intr_num = ir->intr_num;
1863
1864	xhci_remove_interrupter(xhci, ir);
1865	xhci->interrupters[intr_num] = NULL;
1866
1867	spin_unlock_irq(&xhci->lock);
1868
1869	xhci_free_interrupter(xhci, ir);
1870}
1871EXPORT_SYMBOL_GPL(xhci_remove_secondary_interrupter);
1872
1873void xhci_mem_cleanup(struct xhci_hcd *xhci)
1874{
1875	struct device	*dev = xhci_to_hcd(xhci)->self.sysdev;
1876	int i, j, num_ports;
1877
1878	cancel_delayed_work_sync(&xhci->cmd_timer);
1879
1880	for (i = 0; i < xhci->max_interrupters; i++) {
1881		if (xhci->interrupters[i]) {
1882			xhci_remove_interrupter(xhci, xhci->interrupters[i]);
1883			xhci_free_interrupter(xhci, xhci->interrupters[i]);
1884			xhci->interrupters[i] = NULL;
1885		}
1886	}
1887	xhci_dbg_trace(xhci, trace_xhci_dbg_init, "Freed interrupters");
1888
1889	if (xhci->cmd_ring)
1890		xhci_ring_free(xhci, xhci->cmd_ring);
1891	xhci->cmd_ring = NULL;
1892	xhci_dbg_trace(xhci, trace_xhci_dbg_init, "Freed command ring");
1893	xhci_cleanup_command_queue(xhci);
1894
1895	num_ports = HCS_MAX_PORTS(xhci->hcs_params1);
1896	for (i = 0; i < num_ports && xhci->rh_bw; i++) {
1897		struct xhci_interval_bw_table *bwt = &xhci->rh_bw[i].bw_table;
1898		for (j = 0; j < XHCI_MAX_INTERVAL; j++) {
1899			struct list_head *ep = &bwt->interval_bw[j].endpoints;
1900			while (!list_empty(ep))
1901				list_del_init(ep->next);
1902		}
1903	}
1904
1905	for (i = HCS_MAX_SLOTS(xhci->hcs_params1); i > 0; i--)
1906		xhci_free_virt_devices_depth_first(xhci, i);
1907
1908	dma_pool_destroy(xhci->segment_pool);
1909	xhci->segment_pool = NULL;
1910	xhci_dbg_trace(xhci, trace_xhci_dbg_init, "Freed segment pool");
1911
1912	dma_pool_destroy(xhci->device_pool);
1913	xhci->device_pool = NULL;
1914	xhci_dbg_trace(xhci, trace_xhci_dbg_init, "Freed device context pool");
1915
1916	dma_pool_destroy(xhci->small_streams_pool);
1917	xhci->small_streams_pool = NULL;
1918	xhci_dbg_trace(xhci, trace_xhci_dbg_init,
1919			"Freed small stream array pool");
1920
1921	dma_pool_destroy(xhci->medium_streams_pool);
1922	xhci->medium_streams_pool = NULL;
1923	xhci_dbg_trace(xhci, trace_xhci_dbg_init,
1924			"Freed medium stream array pool");
1925
1926	if (xhci->dcbaa)
1927		dma_free_coherent(dev, sizeof(*xhci->dcbaa),
1928				xhci->dcbaa, xhci->dcbaa->dma);
1929	xhci->dcbaa = NULL;
1930
1931	scratchpad_free(xhci);
1932
1933	if (!xhci->rh_bw)
1934		goto no_bw;
1935
1936	for (i = 0; i < num_ports; i++) {
1937		struct xhci_tt_bw_info *tt, *n;
1938		list_for_each_entry_safe(tt, n, &xhci->rh_bw[i].tts, tt_list) {
1939			list_del(&tt->tt_list);
1940			kfree(tt);
1941		}
1942	}
1943
1944no_bw:
1945	xhci->cmd_ring_reserved_trbs = 0;
1946	xhci->usb2_rhub.num_ports = 0;
1947	xhci->usb3_rhub.num_ports = 0;
1948	xhci->num_active_eps = 0;
1949	kfree(xhci->usb2_rhub.ports);
1950	kfree(xhci->usb3_rhub.ports);
1951	kfree(xhci->hw_ports);
1952	kfree(xhci->rh_bw);
1953	kfree(xhci->ext_caps);
1954	for (i = 0; i < xhci->num_port_caps; i++)
1955		kfree(xhci->port_caps[i].psi);
1956	kfree(xhci->port_caps);
1957	kfree(xhci->interrupters);
1958	xhci->num_port_caps = 0;
1959
1960	xhci->usb2_rhub.ports = NULL;
1961	xhci->usb3_rhub.ports = NULL;
1962	xhci->hw_ports = NULL;
1963	xhci->rh_bw = NULL;
1964	xhci->ext_caps = NULL;
1965	xhci->port_caps = NULL;
1966	xhci->interrupters = NULL;
1967
1968	xhci->page_size = 0;
1969	xhci->page_shift = 0;
1970	xhci->usb2_rhub.bus_state.bus_suspended = 0;
1971	xhci->usb3_rhub.bus_state.bus_suspended = 0;
1972}
1973
1974static void xhci_set_hc_event_deq(struct xhci_hcd *xhci, struct xhci_interrupter *ir)
1975{
1976	dma_addr_t deq;
1977
1978	deq = xhci_trb_virt_to_dma(ir->event_ring->deq_seg,
1979			ir->event_ring->dequeue);
1980	if (!deq)
1981		xhci_warn(xhci, "WARN something wrong with SW event ring dequeue ptr.\n");
1982	/* Update HC event ring dequeue pointer */
1983	/* Don't clear the EHB bit (which is RW1C) because
1984	 * there might be more events to service.
1985	 */
1986	xhci_dbg_trace(xhci, trace_xhci_dbg_init,
1987		       "// Write event ring dequeue pointer, preserving EHB bit");
1988	xhci_write_64(xhci, deq & ERST_PTR_MASK, &ir->ir_set->erst_dequeue);
1989}
1990
1991static void xhci_add_in_port(struct xhci_hcd *xhci, unsigned int num_ports,
1992		__le32 __iomem *addr, int max_caps)
1993{
1994	u32 temp, port_offset, port_count;
1995	int i;
1996	u8 major_revision, minor_revision, tmp_minor_revision;
1997	struct xhci_hub *rhub;
1998	struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
1999	struct xhci_port_cap *port_cap;
2000
2001	temp = readl(addr);
2002	major_revision = XHCI_EXT_PORT_MAJOR(temp);
2003	minor_revision = XHCI_EXT_PORT_MINOR(temp);
2004
2005	if (major_revision == 0x03) {
2006		rhub = &xhci->usb3_rhub;
2007		/*
2008		 * Some hosts incorrectly use sub-minor version for minor
2009		 * version (i.e. 0x02 instead of 0x20 for bcdUSB 0x320 and 0x01
2010		 * for bcdUSB 0x310). Since there is no USB release with sub
2011		 * minor version 0x301 to 0x309, we can assume that they are
2012		 * incorrect and fix it here.
2013		 */
2014		if (minor_revision > 0x00 && minor_revision < 0x10)
2015			minor_revision <<= 4;
2016		/*
2017		 * Some zhaoxin's xHCI controller that follow usb3.1 spec
2018		 * but only support Gen1.
2019		 */
2020		if (xhci->quirks & XHCI_ZHAOXIN_HOST) {
2021			tmp_minor_revision = minor_revision;
2022			minor_revision = 0;
2023		}
2024
2025	} else if (major_revision <= 0x02) {
2026		rhub = &xhci->usb2_rhub;
2027	} else {
2028		xhci_warn(xhci, "Ignoring unknown port speed, Ext Cap %p, revision = 0x%x\n",
2029				addr, major_revision);
2030		/* Ignoring port protocol we can't understand. FIXME */
2031		return;
2032	}
2033
2034	/* Port offset and count in the third dword, see section 7.2 */
2035	temp = readl(addr + 2);
2036	port_offset = XHCI_EXT_PORT_OFF(temp);
2037	port_count = XHCI_EXT_PORT_COUNT(temp);
2038	xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2039		       "Ext Cap %p, port offset = %u, count = %u, revision = 0x%x",
2040		       addr, port_offset, port_count, major_revision);
2041	/* Port count includes the current port offset */
2042	if (port_offset == 0 || (port_offset + port_count - 1) > num_ports)
2043		/* WTF? "Valid values are ���1��� to MaxPorts" */
2044		return;
2045
2046	port_cap = &xhci->port_caps[xhci->num_port_caps++];
2047	if (xhci->num_port_caps > max_caps)
2048		return;
2049
2050	port_cap->psi_count = XHCI_EXT_PORT_PSIC(temp);
2051
2052	if (port_cap->psi_count) {
2053		port_cap->psi = kcalloc_node(port_cap->psi_count,
2054					     sizeof(*port_cap->psi),
2055					     GFP_KERNEL, dev_to_node(dev));
2056		if (!port_cap->psi)
2057			port_cap->psi_count = 0;
2058
2059		port_cap->psi_uid_count++;
2060		for (i = 0; i < port_cap->psi_count; i++) {
2061			port_cap->psi[i] = readl(addr + 4 + i);
2062
2063			/* count unique ID values, two consecutive entries can
2064			 * have the same ID if link is assymetric
2065			 */
2066			if (i && (XHCI_EXT_PORT_PSIV(port_cap->psi[i]) !=
2067				  XHCI_EXT_PORT_PSIV(port_cap->psi[i - 1])))
2068				port_cap->psi_uid_count++;
2069
2070			if (xhci->quirks & XHCI_ZHAOXIN_HOST &&
2071			    major_revision == 0x03 &&
2072			    XHCI_EXT_PORT_PSIV(port_cap->psi[i]) >= 5)
2073				minor_revision = tmp_minor_revision;
2074
2075			xhci_dbg(xhci, "PSIV:%d PSIE:%d PLT:%d PFD:%d LP:%d PSIM:%d\n",
2076				  XHCI_EXT_PORT_PSIV(port_cap->psi[i]),
2077				  XHCI_EXT_PORT_PSIE(port_cap->psi[i]),
2078				  XHCI_EXT_PORT_PLT(port_cap->psi[i]),
2079				  XHCI_EXT_PORT_PFD(port_cap->psi[i]),
2080				  XHCI_EXT_PORT_LP(port_cap->psi[i]),
2081				  XHCI_EXT_PORT_PSIM(port_cap->psi[i]));
2082		}
2083	}
2084
2085	rhub->maj_rev = major_revision;
2086
2087	if (rhub->min_rev < minor_revision)
2088		rhub->min_rev = minor_revision;
2089
2090	port_cap->maj_rev = major_revision;
2091	port_cap->min_rev = minor_revision;
2092
2093	/* cache usb2 port capabilities */
2094	if (major_revision < 0x03 && xhci->num_ext_caps < max_caps)
2095		xhci->ext_caps[xhci->num_ext_caps++] = temp;
2096
2097	if ((xhci->hci_version >= 0x100) && (major_revision != 0x03) &&
2098		 (temp & XHCI_HLC)) {
2099		xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2100			       "xHCI 1.0: support USB2 hardware lpm");
2101		xhci->hw_lpm_support = 1;
2102	}
2103
2104	port_offset--;
2105	for (i = port_offset; i < (port_offset + port_count); i++) {
2106		struct xhci_port *hw_port = &xhci->hw_ports[i];
2107		/* Duplicate entry.  Ignore the port if the revisions differ. */
2108		if (hw_port->rhub) {
2109			xhci_warn(xhci, "Duplicate port entry, Ext Cap %p, port %u\n", addr, i);
2110			xhci_warn(xhci, "Port was marked as USB %u, duplicated as USB %u\n",
2111					hw_port->rhub->maj_rev, major_revision);
2112			/* Only adjust the roothub port counts if we haven't
2113			 * found a similar duplicate.
2114			 */
2115			if (hw_port->rhub != rhub &&
2116				 hw_port->hcd_portnum != DUPLICATE_ENTRY) {
2117				hw_port->rhub->num_ports--;
2118				hw_port->hcd_portnum = DUPLICATE_ENTRY;
2119			}
2120			continue;
2121		}
2122		hw_port->rhub = rhub;
2123		hw_port->port_cap = port_cap;
2124		rhub->num_ports++;
2125	}
2126	/* FIXME: Should we disable ports not in the Extended Capabilities? */
2127}
2128
2129static void xhci_create_rhub_port_array(struct xhci_hcd *xhci,
2130					struct xhci_hub *rhub, gfp_t flags)
2131{
2132	int port_index = 0;
2133	int i;
2134	struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
2135
2136	if (!rhub->num_ports)
2137		return;
2138	rhub->ports = kcalloc_node(rhub->num_ports, sizeof(*rhub->ports),
2139			flags, dev_to_node(dev));
2140	if (!rhub->ports)
2141		return;
2142
2143	for (i = 0; i < HCS_MAX_PORTS(xhci->hcs_params1); i++) {
2144		if (xhci->hw_ports[i].rhub != rhub ||
2145		    xhci->hw_ports[i].hcd_portnum == DUPLICATE_ENTRY)
2146			continue;
2147		xhci->hw_ports[i].hcd_portnum = port_index;
2148		rhub->ports[port_index] = &xhci->hw_ports[i];
2149		port_index++;
2150		if (port_index == rhub->num_ports)
2151			break;
2152	}
2153}
2154
2155/*
2156 * Scan the Extended Capabilities for the "Supported Protocol Capabilities" that
2157 * specify what speeds each port is supposed to be.  We can't count on the port
2158 * speed bits in the PORTSC register being correct until a device is connected,
2159 * but we need to set up the two fake roothubs with the correct number of USB
2160 * 3.0 and USB 2.0 ports at host controller initialization time.
2161 */
2162static int xhci_setup_port_arrays(struct xhci_hcd *xhci, gfp_t flags)
2163{
2164	void __iomem *base;
2165	u32 offset;
2166	unsigned int num_ports;
2167	int i, j;
2168	int cap_count = 0;
2169	u32 cap_start;
2170	struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
2171
2172	num_ports = HCS_MAX_PORTS(xhci->hcs_params1);
2173	xhci->hw_ports = kcalloc_node(num_ports, sizeof(*xhci->hw_ports),
2174				flags, dev_to_node(dev));
2175	if (!xhci->hw_ports)
2176		return -ENOMEM;
2177
2178	for (i = 0; i < num_ports; i++) {
2179		xhci->hw_ports[i].addr = &xhci->op_regs->port_status_base +
2180			NUM_PORT_REGS * i;
2181		xhci->hw_ports[i].hw_portnum = i;
2182
2183		init_completion(&xhci->hw_ports[i].rexit_done);
2184		init_completion(&xhci->hw_ports[i].u3exit_done);
2185	}
2186
2187	xhci->rh_bw = kcalloc_node(num_ports, sizeof(*xhci->rh_bw), flags,
2188				   dev_to_node(dev));
2189	if (!xhci->rh_bw)
2190		return -ENOMEM;
2191	for (i = 0; i < num_ports; i++) {
2192		struct xhci_interval_bw_table *bw_table;
2193
2194		INIT_LIST_HEAD(&xhci->rh_bw[i].tts);
2195		bw_table = &xhci->rh_bw[i].bw_table;
2196		for (j = 0; j < XHCI_MAX_INTERVAL; j++)
2197			INIT_LIST_HEAD(&bw_table->interval_bw[j].endpoints);
2198	}
2199	base = &xhci->cap_regs->hc_capbase;
2200
2201	cap_start = xhci_find_next_ext_cap(base, 0, XHCI_EXT_CAPS_PROTOCOL);
2202	if (!cap_start) {
2203		xhci_err(xhci, "No Extended Capability registers, unable to set up roothub\n");
2204		return -ENODEV;
2205	}
2206
2207	offset = cap_start;
2208	/* count extended protocol capability entries for later caching */
2209	while (offset) {
2210		cap_count++;
2211		offset = xhci_find_next_ext_cap(base, offset,
2212						      XHCI_EXT_CAPS_PROTOCOL);
2213	}
2214
2215	xhci->ext_caps = kcalloc_node(cap_count, sizeof(*xhci->ext_caps),
2216				flags, dev_to_node(dev));
2217	if (!xhci->ext_caps)
2218		return -ENOMEM;
2219
2220	xhci->port_caps = kcalloc_node(cap_count, sizeof(*xhci->port_caps),
2221				flags, dev_to_node(dev));
2222	if (!xhci->port_caps)
2223		return -ENOMEM;
2224
2225	offset = cap_start;
2226
2227	while (offset) {
2228		xhci_add_in_port(xhci, num_ports, base + offset, cap_count);
2229		if (xhci->usb2_rhub.num_ports + xhci->usb3_rhub.num_ports ==
2230		    num_ports)
2231			break;
2232		offset = xhci_find_next_ext_cap(base, offset,
2233						XHCI_EXT_CAPS_PROTOCOL);
2234	}
2235	if (xhci->usb2_rhub.num_ports == 0 && xhci->usb3_rhub.num_ports == 0) {
2236		xhci_warn(xhci, "No ports on the roothubs?\n");
2237		return -ENODEV;
2238	}
2239	xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2240		       "Found %u USB 2.0 ports and %u USB 3.0 ports.",
2241		       xhci->usb2_rhub.num_ports, xhci->usb3_rhub.num_ports);
2242
2243	/* Place limits on the number of roothub ports so that the hub
2244	 * descriptors aren't longer than the USB core will allocate.
2245	 */
2246	if (xhci->usb3_rhub.num_ports > USB_SS_MAXPORTS) {
2247		xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2248				"Limiting USB 3.0 roothub ports to %u.",
2249				USB_SS_MAXPORTS);
2250		xhci->usb3_rhub.num_ports = USB_SS_MAXPORTS;
2251	}
2252	if (xhci->usb2_rhub.num_ports > USB_MAXCHILDREN) {
2253		xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2254				"Limiting USB 2.0 roothub ports to %u.",
2255				USB_MAXCHILDREN);
2256		xhci->usb2_rhub.num_ports = USB_MAXCHILDREN;
2257	}
2258
2259	if (!xhci->usb2_rhub.num_ports)
2260		xhci_info(xhci, "USB2 root hub has no ports\n");
2261
2262	if (!xhci->usb3_rhub.num_ports)
2263		xhci_info(xhci, "USB3 root hub has no ports\n");
2264
2265	xhci_create_rhub_port_array(xhci, &xhci->usb2_rhub, flags);
2266	xhci_create_rhub_port_array(xhci, &xhci->usb3_rhub, flags);
2267
2268	return 0;
2269}
2270
2271static struct xhci_interrupter *
2272xhci_alloc_interrupter(struct xhci_hcd *xhci, int segs, gfp_t flags)
2273{
2274	struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
2275	struct xhci_interrupter *ir;
2276	unsigned int num_segs = segs;
2277	int ret;
2278
2279	ir = kzalloc_node(sizeof(*ir), flags, dev_to_node(dev));
2280	if (!ir)
2281		return NULL;
2282
2283	/* number of ring segments should be greater than 0 */
2284	if (segs <= 0)
2285		num_segs = min_t(unsigned int, 1 << HCS_ERST_MAX(xhci->hcs_params2),
2286			 ERST_MAX_SEGS);
2287
2288	ir->event_ring = xhci_ring_alloc(xhci, num_segs, 1, TYPE_EVENT, 0,
2289					 flags);
2290	if (!ir->event_ring) {
2291		xhci_warn(xhci, "Failed to allocate interrupter event ring\n");
2292		kfree(ir);
2293		return NULL;
2294	}
2295
2296	ret = xhci_alloc_erst(xhci, ir->event_ring, &ir->erst, flags);
2297	if (ret) {
2298		xhci_warn(xhci, "Failed to allocate interrupter erst\n");
2299		xhci_ring_free(xhci, ir->event_ring);
2300		kfree(ir);
2301		return NULL;
2302	}
2303
2304	return ir;
2305}
2306
2307static int
2308xhci_add_interrupter(struct xhci_hcd *xhci, struct xhci_interrupter *ir,
2309		     unsigned int intr_num)
2310{
2311	u64 erst_base;
2312	u32 erst_size;
2313
2314	if (intr_num >= xhci->max_interrupters) {
2315		xhci_warn(xhci, "Can't add interrupter %d, max interrupters %d\n",
2316			  intr_num, xhci->max_interrupters);
2317		return -EINVAL;
2318	}
2319
2320	if (xhci->interrupters[intr_num]) {
2321		xhci_warn(xhci, "Interrupter %d\n already set up", intr_num);
2322		return -EINVAL;
2323	}
2324
2325	xhci->interrupters[intr_num] = ir;
2326	ir->intr_num = intr_num;
2327	ir->ir_set = &xhci->run_regs->ir_set[intr_num];
2328
2329	/* set ERST count with the number of entries in the segment table */
2330	erst_size = readl(&ir->ir_set->erst_size);
2331	erst_size &= ERST_SIZE_MASK;
2332	erst_size |= ir->event_ring->num_segs;
2333	writel(erst_size, &ir->ir_set->erst_size);
2334
2335	erst_base = xhci_read_64(xhci, &ir->ir_set->erst_base);
2336	erst_base &= ERST_BASE_RSVDP;
2337	erst_base |= ir->erst.erst_dma_addr & ~ERST_BASE_RSVDP;
2338	xhci_write_64(xhci, erst_base, &ir->ir_set->erst_base);
2339
2340	/* Set the event ring dequeue address of this interrupter */
2341	xhci_set_hc_event_deq(xhci, ir);
2342
2343	return 0;
2344}
2345
2346struct xhci_interrupter *
2347xhci_create_secondary_interrupter(struct usb_hcd *hcd, int num_seg)
2348{
2349	struct xhci_hcd *xhci = hcd_to_xhci(hcd);
2350	struct xhci_interrupter *ir;
2351	unsigned int i;
2352	int err = -ENOSPC;
2353
2354	if (!xhci->interrupters || xhci->max_interrupters <= 1)
2355		return NULL;
2356
2357	ir = xhci_alloc_interrupter(xhci, num_seg, GFP_KERNEL);
2358	if (!ir)
2359		return NULL;
2360
2361	spin_lock_irq(&xhci->lock);
2362
2363	/* Find available secondary interrupter, interrupter 0 is reserved for primary */
2364	for (i = 1; i < xhci->max_interrupters; i++) {
2365		if (xhci->interrupters[i] == NULL) {
2366			err = xhci_add_interrupter(xhci, ir, i);
2367			break;
2368		}
2369	}
2370
2371	spin_unlock_irq(&xhci->lock);
2372
2373	if (err) {
2374		xhci_warn(xhci, "Failed to add secondary interrupter, max interrupters %d\n",
2375			  xhci->max_interrupters);
2376		xhci_free_interrupter(xhci, ir);
2377		return NULL;
2378	}
2379
2380	xhci_dbg(xhci, "Add secondary interrupter %d, max interrupters %d\n",
2381		 i, xhci->max_interrupters);
2382
2383	return ir;
2384}
2385EXPORT_SYMBOL_GPL(xhci_create_secondary_interrupter);
2386
2387int xhci_mem_init(struct xhci_hcd *xhci, gfp_t flags)
2388{
2389	struct xhci_interrupter *ir;
2390	struct device	*dev = xhci_to_hcd(xhci)->self.sysdev;
2391	dma_addr_t	dma;
2392	unsigned int	val, val2;
2393	u64		val_64;
2394	u32		page_size, temp;
2395	int		i;
2396
2397	INIT_LIST_HEAD(&xhci->cmd_list);
2398
2399	/* init command timeout work */
2400	INIT_DELAYED_WORK(&xhci->cmd_timer, xhci_handle_command_timeout);
2401	init_completion(&xhci->cmd_ring_stop_completion);
2402
2403	page_size = readl(&xhci->op_regs->page_size);
2404	xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2405			"Supported page size register = 0x%x", page_size);
2406	i = ffs(page_size);
2407	if (i < 16)
2408		xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2409			"Supported page size of %iK", (1 << (i+12)) / 1024);
2410	else
2411		xhci_warn(xhci, "WARN: no supported page size\n");
2412	/* Use 4K pages, since that's common and the minimum the HC supports */
2413	xhci->page_shift = 12;
2414	xhci->page_size = 1 << xhci->page_shift;
2415	xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2416			"HCD page size set to %iK", xhci->page_size / 1024);
2417
2418	/*
2419	 * Program the Number of Device Slots Enabled field in the CONFIG
2420	 * register with the max value of slots the HC can handle.
2421	 */
2422	val = HCS_MAX_SLOTS(readl(&xhci->cap_regs->hcs_params1));
2423	xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2424			"// xHC can handle at most %d device slots.", val);
2425	val2 = readl(&xhci->op_regs->config_reg);
2426	val |= (val2 & ~HCS_SLOTS_MASK);
2427	xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2428			"// Setting Max device slots reg = 0x%x.", val);
2429	writel(val, &xhci->op_regs->config_reg);
2430
2431	/*
2432	 * xHCI section 5.4.6 - Device Context array must be
2433	 * "physically contiguous and 64-byte (cache line) aligned".
2434	 */
2435	xhci->dcbaa = dma_alloc_coherent(dev, sizeof(*xhci->dcbaa), &dma,
2436			flags);
2437	if (!xhci->dcbaa)
2438		goto fail;
2439	xhci->dcbaa->dma = dma;
2440	xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2441			"// Device context base array address = 0x%pad (DMA), %p (virt)",
2442			&xhci->dcbaa->dma, xhci->dcbaa);
2443	xhci_write_64(xhci, dma, &xhci->op_regs->dcbaa_ptr);
2444
2445	/*
2446	 * Initialize the ring segment pool.  The ring must be a contiguous
2447	 * structure comprised of TRBs.  The TRBs must be 16 byte aligned,
2448	 * however, the command ring segment needs 64-byte aligned segments
2449	 * and our use of dma addresses in the trb_address_map radix tree needs
2450	 * TRB_SEGMENT_SIZE alignment, so we pick the greater alignment need.
2451	 */
2452	if (xhci->quirks & XHCI_ZHAOXIN_TRB_FETCH)
2453		xhci->segment_pool = dma_pool_create("xHCI ring segments", dev,
2454				TRB_SEGMENT_SIZE * 2, TRB_SEGMENT_SIZE * 2, xhci->page_size * 2);
2455	else
2456		xhci->segment_pool = dma_pool_create("xHCI ring segments", dev,
2457				TRB_SEGMENT_SIZE, TRB_SEGMENT_SIZE, xhci->page_size);
2458
2459	/* See Table 46 and Note on Figure 55 */
2460	xhci->device_pool = dma_pool_create("xHCI input/output contexts", dev,
2461			2112, 64, xhci->page_size);
2462	if (!xhci->segment_pool || !xhci->device_pool)
2463		goto fail;
2464
2465	/* Linear stream context arrays don't have any boundary restrictions,
2466	 * and only need to be 16-byte aligned.
2467	 */
2468	xhci->small_streams_pool =
2469		dma_pool_create("xHCI 256 byte stream ctx arrays",
2470			dev, SMALL_STREAM_ARRAY_SIZE, 16, 0);
2471	xhci->medium_streams_pool =
2472		dma_pool_create("xHCI 1KB stream ctx arrays",
2473			dev, MEDIUM_STREAM_ARRAY_SIZE, 16, 0);
2474	/* Any stream context array bigger than MEDIUM_STREAM_ARRAY_SIZE
2475	 * will be allocated with dma_alloc_coherent()
2476	 */
2477
2478	if (!xhci->small_streams_pool || !xhci->medium_streams_pool)
2479		goto fail;
2480
2481	/* Set up the command ring to have one segments for now. */
2482	xhci->cmd_ring = xhci_ring_alloc(xhci, 1, 1, TYPE_COMMAND, 0, flags);
2483	if (!xhci->cmd_ring)
2484		goto fail;
2485	xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2486			"Allocated command ring at %p", xhci->cmd_ring);
2487	xhci_dbg_trace(xhci, trace_xhci_dbg_init, "First segment DMA is 0x%pad",
2488			&xhci->cmd_ring->first_seg->dma);
2489
2490	/* Set the address in the Command Ring Control register */
2491	val_64 = xhci_read_64(xhci, &xhci->op_regs->cmd_ring);
2492	val_64 = (val_64 & (u64) CMD_RING_RSVD_BITS) |
2493		(xhci->cmd_ring->first_seg->dma & (u64) ~CMD_RING_RSVD_BITS) |
2494		xhci->cmd_ring->cycle_state;
2495	xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2496			"// Setting command ring address to 0x%016llx", val_64);
2497	xhci_write_64(xhci, val_64, &xhci->op_regs->cmd_ring);
2498
2499	/* Reserve one command ring TRB for disabling LPM.
2500	 * Since the USB core grabs the shared usb_bus bandwidth mutex before
2501	 * disabling LPM, we only need to reserve one TRB for all devices.
2502	 */
2503	xhci->cmd_ring_reserved_trbs++;
2504
2505	val = readl(&xhci->cap_regs->db_off);
2506	val &= DBOFF_MASK;
2507	xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2508		       "// Doorbell array is located at offset 0x%x from cap regs base addr",
2509		       val);
2510	xhci->dba = (void __iomem *) xhci->cap_regs + val;
2511
2512	/* Allocate and set up primary interrupter 0 with an event ring. */
2513	xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2514		       "Allocating primary event ring");
2515	xhci->interrupters = kcalloc_node(xhci->max_interrupters, sizeof(*xhci->interrupters),
2516					  flags, dev_to_node(dev));
2517
2518	ir = xhci_alloc_interrupter(xhci, 0, flags);
2519	if (!ir)
2520		goto fail;
2521
2522	if (xhci_add_interrupter(xhci, ir, 0))
2523		goto fail;
2524
2525	ir->isoc_bei_interval = AVOID_BEI_INTERVAL_MAX;
2526
2527	/*
2528	 * XXX: Might need to set the Interrupter Moderation Register to
2529	 * something other than the default (~1ms minimum between interrupts).
2530	 * See section 5.5.1.2.
2531	 */
2532	for (i = 0; i < MAX_HC_SLOTS; i++)
2533		xhci->devs[i] = NULL;
2534
2535	if (scratchpad_alloc(xhci, flags))
2536		goto fail;
2537	if (xhci_setup_port_arrays(xhci, flags))
2538		goto fail;
2539
2540	/* Enable USB 3.0 device notifications for function remote wake, which
2541	 * is necessary for allowing USB 3.0 devices to do remote wakeup from
2542	 * U3 (device suspend).
2543	 */
2544	temp = readl(&xhci->op_regs->dev_notification);
2545	temp &= ~DEV_NOTE_MASK;
2546	temp |= DEV_NOTE_FWAKE;
2547	writel(temp, &xhci->op_regs->dev_notification);
2548
2549	return 0;
2550
2551fail:
2552	xhci_halt(xhci);
2553	xhci_reset(xhci, XHCI_RESET_SHORT_USEC);
2554	xhci_mem_cleanup(xhci);
2555	return -ENOMEM;
2556}
2557