Loading...
1/*
2 * xHCI host controller driver
3 *
4 * Copyright (C) 2008 Intel Corp.
5 *
6 * Author: Sarah Sharp
7 * Some code borrowed from the Linux EHCI driver.
8 *
9 * This program is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License version 2 as
11 * published by the Free Software Foundation.
12 *
13 * This program is distributed in the hope that it will be useful, but
14 * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
15 * or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
16 * for more details.
17 *
18 * You should have received a copy of the GNU General Public License
19 * along with this program; if not, write to the Free Software Foundation,
20 * Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
21 */
22
23#include <linux/usb.h>
24#include <linux/pci.h>
25#include <linux/slab.h>
26#include <linux/dmapool.h>
27#include <linux/dma-mapping.h>
28
29#include "xhci.h"
30#include "xhci-trace.h"
31
32/*
33 * Allocates a generic ring segment from the ring pool, sets the dma address,
34 * initializes the segment to zero, and sets the private next pointer to NULL.
35 *
36 * Section 4.11.1.1:
37 * "All components of all Command and Transfer TRBs shall be initialized to '0'"
38 */
39static struct xhci_segment *xhci_segment_alloc(struct xhci_hcd *xhci,
40 unsigned int cycle_state, gfp_t flags)
41{
42 struct xhci_segment *seg;
43 dma_addr_t dma;
44 int i;
45
46 seg = kzalloc(sizeof *seg, flags);
47 if (!seg)
48 return NULL;
49
50 seg->trbs = dma_pool_alloc(xhci->segment_pool, flags, &dma);
51 if (!seg->trbs) {
52 kfree(seg);
53 return NULL;
54 }
55
56 memset(seg->trbs, 0, TRB_SEGMENT_SIZE);
57 /* If the cycle state is 0, set the cycle bit to 1 for all the TRBs */
58 if (cycle_state == 0) {
59 for (i = 0; i < TRBS_PER_SEGMENT; i++)
60 seg->trbs[i].link.control |= cpu_to_le32(TRB_CYCLE);
61 }
62 seg->dma = dma;
63 seg->next = NULL;
64
65 return seg;
66}
67
68static void xhci_segment_free(struct xhci_hcd *xhci, struct xhci_segment *seg)
69{
70 if (seg->trbs) {
71 dma_pool_free(xhci->segment_pool, seg->trbs, seg->dma);
72 seg->trbs = NULL;
73 }
74 kfree(seg);
75}
76
77static void xhci_free_segments_for_ring(struct xhci_hcd *xhci,
78 struct xhci_segment *first)
79{
80 struct xhci_segment *seg;
81
82 seg = first->next;
83 while (seg != first) {
84 struct xhci_segment *next = seg->next;
85 xhci_segment_free(xhci, seg);
86 seg = next;
87 }
88 xhci_segment_free(xhci, first);
89}
90
91/*
92 * Make the prev segment point to the next segment.
93 *
94 * Change the last TRB in the prev segment to be a Link TRB which points to the
95 * DMA address of the next segment. The caller needs to set any Link TRB
96 * related flags, such as End TRB, Toggle Cycle, and no snoop.
97 */
98static void xhci_link_segments(struct xhci_hcd *xhci, struct xhci_segment *prev,
99 struct xhci_segment *next, enum xhci_ring_type type)
100{
101 u32 val;
102
103 if (!prev || !next)
104 return;
105 prev->next = next;
106 if (type != TYPE_EVENT) {
107 prev->trbs[TRBS_PER_SEGMENT-1].link.segment_ptr =
108 cpu_to_le64(next->dma);
109
110 /* Set the last TRB in the segment to have a TRB type ID of Link TRB */
111 val = le32_to_cpu(prev->trbs[TRBS_PER_SEGMENT-1].link.control);
112 val &= ~TRB_TYPE_BITMASK;
113 val |= TRB_TYPE(TRB_LINK);
114 /* Always set the chain bit with 0.95 hardware */
115 /* Set chain bit for isoc rings on AMD 0.96 host */
116 if (xhci_link_trb_quirk(xhci) ||
117 (type == TYPE_ISOC &&
118 (xhci->quirks & XHCI_AMD_0x96_HOST)))
119 val |= TRB_CHAIN;
120 prev->trbs[TRBS_PER_SEGMENT-1].link.control = cpu_to_le32(val);
121 }
122}
123
124/*
125 * Link the ring to the new segments.
126 * Set Toggle Cycle for the new ring if needed.
127 */
128static void xhci_link_rings(struct xhci_hcd *xhci, struct xhci_ring *ring,
129 struct xhci_segment *first, struct xhci_segment *last,
130 unsigned int num_segs)
131{
132 struct xhci_segment *next;
133
134 if (!ring || !first || !last)
135 return;
136
137 next = ring->enq_seg->next;
138 xhci_link_segments(xhci, ring->enq_seg, first, ring->type);
139 xhci_link_segments(xhci, last, next, ring->type);
140 ring->num_segs += num_segs;
141 ring->num_trbs_free += (TRBS_PER_SEGMENT - 1) * num_segs;
142
143 if (ring->type != TYPE_EVENT && ring->enq_seg == ring->last_seg) {
144 ring->last_seg->trbs[TRBS_PER_SEGMENT-1].link.control
145 &= ~cpu_to_le32(LINK_TOGGLE);
146 last->trbs[TRBS_PER_SEGMENT-1].link.control
147 |= cpu_to_le32(LINK_TOGGLE);
148 ring->last_seg = last;
149 }
150}
151
152/*
153 * We need a radix tree for mapping physical addresses of TRBs to which stream
154 * ID they belong to. We need to do this because the host controller won't tell
155 * us which stream ring the TRB came from. We could store the stream ID in an
156 * event data TRB, but that doesn't help us for the cancellation case, since the
157 * endpoint may stop before it reaches that event data TRB.
158 *
159 * The radix tree maps the upper portion of the TRB DMA address to a ring
160 * segment that has the same upper portion of DMA addresses. For example, say I
161 * have segments of size 1KB, that are always 1KB aligned. A segment may
162 * start at 0x10c91000 and end at 0x10c913f0. If I use the upper 10 bits, the
163 * key to the stream ID is 0x43244. I can use the DMA address of the TRB to
164 * pass the radix tree a key to get the right stream ID:
165 *
166 * 0x10c90fff >> 10 = 0x43243
167 * 0x10c912c0 >> 10 = 0x43244
168 * 0x10c91400 >> 10 = 0x43245
169 *
170 * Obviously, only those TRBs with DMA addresses that are within the segment
171 * will make the radix tree return the stream ID for that ring.
172 *
173 * Caveats for the radix tree:
174 *
175 * The radix tree uses an unsigned long as a key pair. On 32-bit systems, an
176 * unsigned long will be 32-bits; on a 64-bit system an unsigned long will be
177 * 64-bits. Since we only request 32-bit DMA addresses, we can use that as the
178 * key on 32-bit or 64-bit systems (it would also be fine if we asked for 64-bit
179 * PCI DMA addresses on a 64-bit system). There might be a problem on 32-bit
180 * extended systems (where the DMA address can be bigger than 32-bits),
181 * if we allow the PCI dma mask to be bigger than 32-bits. So don't do that.
182 */
183static int xhci_insert_segment_mapping(struct radix_tree_root *trb_address_map,
184 struct xhci_ring *ring,
185 struct xhci_segment *seg,
186 gfp_t mem_flags)
187{
188 unsigned long key;
189 int ret;
190
191 key = (unsigned long)(seg->dma >> TRB_SEGMENT_SHIFT);
192 /* Skip any segments that were already added. */
193 if (radix_tree_lookup(trb_address_map, key))
194 return 0;
195
196 ret = radix_tree_maybe_preload(mem_flags);
197 if (ret)
198 return ret;
199 ret = radix_tree_insert(trb_address_map,
200 key, ring);
201 radix_tree_preload_end();
202 return ret;
203}
204
205static void xhci_remove_segment_mapping(struct radix_tree_root *trb_address_map,
206 struct xhci_segment *seg)
207{
208 unsigned long key;
209
210 key = (unsigned long)(seg->dma >> TRB_SEGMENT_SHIFT);
211 if (radix_tree_lookup(trb_address_map, key))
212 radix_tree_delete(trb_address_map, key);
213}
214
215static int xhci_update_stream_segment_mapping(
216 struct radix_tree_root *trb_address_map,
217 struct xhci_ring *ring,
218 struct xhci_segment *first_seg,
219 struct xhci_segment *last_seg,
220 gfp_t mem_flags)
221{
222 struct xhci_segment *seg;
223 struct xhci_segment *failed_seg;
224 int ret;
225
226 if (WARN_ON_ONCE(trb_address_map == NULL))
227 return 0;
228
229 seg = first_seg;
230 do {
231 ret = xhci_insert_segment_mapping(trb_address_map,
232 ring, seg, mem_flags);
233 if (ret)
234 goto remove_streams;
235 if (seg == last_seg)
236 return 0;
237 seg = seg->next;
238 } while (seg != first_seg);
239
240 return 0;
241
242remove_streams:
243 failed_seg = seg;
244 seg = first_seg;
245 do {
246 xhci_remove_segment_mapping(trb_address_map, seg);
247 if (seg == failed_seg)
248 return ret;
249 seg = seg->next;
250 } while (seg != first_seg);
251
252 return ret;
253}
254
255static void xhci_remove_stream_mapping(struct xhci_ring *ring)
256{
257 struct xhci_segment *seg;
258
259 if (WARN_ON_ONCE(ring->trb_address_map == NULL))
260 return;
261
262 seg = ring->first_seg;
263 do {
264 xhci_remove_segment_mapping(ring->trb_address_map, seg);
265 seg = seg->next;
266 } while (seg != ring->first_seg);
267}
268
269static int xhci_update_stream_mapping(struct xhci_ring *ring, gfp_t mem_flags)
270{
271 return xhci_update_stream_segment_mapping(ring->trb_address_map, ring,
272 ring->first_seg, ring->last_seg, mem_flags);
273}
274
275/* XXX: Do we need the hcd structure in all these functions? */
276void xhci_ring_free(struct xhci_hcd *xhci, struct xhci_ring *ring)
277{
278 if (!ring)
279 return;
280
281 if (ring->first_seg) {
282 if (ring->type == TYPE_STREAM)
283 xhci_remove_stream_mapping(ring);
284 xhci_free_segments_for_ring(xhci, ring->first_seg);
285 }
286
287 kfree(ring);
288}
289
290static void xhci_initialize_ring_info(struct xhci_ring *ring,
291 unsigned int cycle_state)
292{
293 /* The ring is empty, so the enqueue pointer == dequeue pointer */
294 ring->enqueue = ring->first_seg->trbs;
295 ring->enq_seg = ring->first_seg;
296 ring->dequeue = ring->enqueue;
297 ring->deq_seg = ring->first_seg;
298 /* The ring is initialized to 0. The producer must write 1 to the cycle
299 * bit to handover ownership of the TRB, so PCS = 1. The consumer must
300 * compare CCS to the cycle bit to check ownership, so CCS = 1.
301 *
302 * New rings are initialized with cycle state equal to 1; if we are
303 * handling ring expansion, set the cycle state equal to the old ring.
304 */
305 ring->cycle_state = cycle_state;
306 /* Not necessary for new rings, but needed for re-initialized rings */
307 ring->enq_updates = 0;
308 ring->deq_updates = 0;
309
310 /*
311 * Each segment has a link TRB, and leave an extra TRB for SW
312 * accounting purpose
313 */
314 ring->num_trbs_free = ring->num_segs * (TRBS_PER_SEGMENT - 1) - 1;
315}
316
317/* Allocate segments and link them for a ring */
318static int xhci_alloc_segments_for_ring(struct xhci_hcd *xhci,
319 struct xhci_segment **first, struct xhci_segment **last,
320 unsigned int num_segs, unsigned int cycle_state,
321 enum xhci_ring_type type, gfp_t flags)
322{
323 struct xhci_segment *prev;
324
325 prev = xhci_segment_alloc(xhci, cycle_state, flags);
326 if (!prev)
327 return -ENOMEM;
328 num_segs--;
329
330 *first = prev;
331 while (num_segs > 0) {
332 struct xhci_segment *next;
333
334 next = xhci_segment_alloc(xhci, cycle_state, flags);
335 if (!next) {
336 prev = *first;
337 while (prev) {
338 next = prev->next;
339 xhci_segment_free(xhci, prev);
340 prev = next;
341 }
342 return -ENOMEM;
343 }
344 xhci_link_segments(xhci, prev, next, type);
345
346 prev = next;
347 num_segs--;
348 }
349 xhci_link_segments(xhci, prev, *first, type);
350 *last = prev;
351
352 return 0;
353}
354
355/**
356 * Create a new ring with zero or more segments.
357 *
358 * Link each segment together into a ring.
359 * Set the end flag and the cycle toggle bit on the last segment.
360 * See section 4.9.1 and figures 15 and 16.
361 */
362static struct xhci_ring *xhci_ring_alloc(struct xhci_hcd *xhci,
363 unsigned int num_segs, unsigned int cycle_state,
364 enum xhci_ring_type type, gfp_t flags)
365{
366 struct xhci_ring *ring;
367 int ret;
368
369 ring = kzalloc(sizeof *(ring), flags);
370 if (!ring)
371 return NULL;
372
373 ring->num_segs = num_segs;
374 INIT_LIST_HEAD(&ring->td_list);
375 ring->type = type;
376 if (num_segs == 0)
377 return ring;
378
379 ret = xhci_alloc_segments_for_ring(xhci, &ring->first_seg,
380 &ring->last_seg, num_segs, cycle_state, type, flags);
381 if (ret)
382 goto fail;
383
384 /* Only event ring does not use link TRB */
385 if (type != TYPE_EVENT) {
386 /* See section 4.9.2.1 and 6.4.4.1 */
387 ring->last_seg->trbs[TRBS_PER_SEGMENT - 1].link.control |=
388 cpu_to_le32(LINK_TOGGLE);
389 }
390 xhci_initialize_ring_info(ring, cycle_state);
391 return ring;
392
393fail:
394 kfree(ring);
395 return NULL;
396}
397
398void xhci_free_or_cache_endpoint_ring(struct xhci_hcd *xhci,
399 struct xhci_virt_device *virt_dev,
400 unsigned int ep_index)
401{
402 int rings_cached;
403
404 rings_cached = virt_dev->num_rings_cached;
405 if (rings_cached < XHCI_MAX_RINGS_CACHED) {
406 virt_dev->ring_cache[rings_cached] =
407 virt_dev->eps[ep_index].ring;
408 virt_dev->num_rings_cached++;
409 xhci_dbg(xhci, "Cached old ring, "
410 "%d ring%s cached\n",
411 virt_dev->num_rings_cached,
412 (virt_dev->num_rings_cached > 1) ? "s" : "");
413 } else {
414 xhci_ring_free(xhci, virt_dev->eps[ep_index].ring);
415 xhci_dbg(xhci, "Ring cache full (%d rings), "
416 "freeing ring\n",
417 virt_dev->num_rings_cached);
418 }
419 virt_dev->eps[ep_index].ring = NULL;
420}
421
422/* Zero an endpoint ring (except for link TRBs) and move the enqueue and dequeue
423 * pointers to the beginning of the ring.
424 */
425static void xhci_reinit_cached_ring(struct xhci_hcd *xhci,
426 struct xhci_ring *ring, unsigned int cycle_state,
427 enum xhci_ring_type type)
428{
429 struct xhci_segment *seg = ring->first_seg;
430 int i;
431
432 do {
433 memset(seg->trbs, 0,
434 sizeof(union xhci_trb)*TRBS_PER_SEGMENT);
435 if (cycle_state == 0) {
436 for (i = 0; i < TRBS_PER_SEGMENT; i++)
437 seg->trbs[i].link.control |=
438 cpu_to_le32(TRB_CYCLE);
439 }
440 /* All endpoint rings have link TRBs */
441 xhci_link_segments(xhci, seg, seg->next, type);
442 seg = seg->next;
443 } while (seg != ring->first_seg);
444 ring->type = type;
445 xhci_initialize_ring_info(ring, cycle_state);
446 /* td list should be empty since all URBs have been cancelled,
447 * but just in case...
448 */
449 INIT_LIST_HEAD(&ring->td_list);
450}
451
452/*
453 * Expand an existing ring.
454 * Look for a cached ring or allocate a new ring which has same segment numbers
455 * and link the two rings.
456 */
457int xhci_ring_expansion(struct xhci_hcd *xhci, struct xhci_ring *ring,
458 unsigned int num_trbs, gfp_t flags)
459{
460 struct xhci_segment *first;
461 struct xhci_segment *last;
462 unsigned int num_segs;
463 unsigned int num_segs_needed;
464 int ret;
465
466 num_segs_needed = (num_trbs + (TRBS_PER_SEGMENT - 1) - 1) /
467 (TRBS_PER_SEGMENT - 1);
468
469 /* Allocate number of segments we needed, or double the ring size */
470 num_segs = ring->num_segs > num_segs_needed ?
471 ring->num_segs : num_segs_needed;
472
473 ret = xhci_alloc_segments_for_ring(xhci, &first, &last,
474 num_segs, ring->cycle_state, ring->type, flags);
475 if (ret)
476 return -ENOMEM;
477
478 if (ring->type == TYPE_STREAM)
479 ret = xhci_update_stream_segment_mapping(ring->trb_address_map,
480 ring, first, last, flags);
481 if (ret) {
482 struct xhci_segment *next;
483 do {
484 next = first->next;
485 xhci_segment_free(xhci, first);
486 if (first == last)
487 break;
488 first = next;
489 } while (true);
490 return ret;
491 }
492
493 xhci_link_rings(xhci, ring, first, last, num_segs);
494 xhci_dbg_trace(xhci, trace_xhci_dbg_ring_expansion,
495 "ring expansion succeed, now has %d segments",
496 ring->num_segs);
497
498 return 0;
499}
500
501#define CTX_SIZE(_hcc) (HCC_64BYTE_CONTEXT(_hcc) ? 64 : 32)
502
503static struct xhci_container_ctx *xhci_alloc_container_ctx(struct xhci_hcd *xhci,
504 int type, gfp_t flags)
505{
506 struct xhci_container_ctx *ctx;
507
508 if ((type != XHCI_CTX_TYPE_DEVICE) && (type != XHCI_CTX_TYPE_INPUT))
509 return NULL;
510
511 ctx = kzalloc(sizeof(*ctx), flags);
512 if (!ctx)
513 return NULL;
514
515 ctx->type = type;
516 ctx->size = HCC_64BYTE_CONTEXT(xhci->hcc_params) ? 2048 : 1024;
517 if (type == XHCI_CTX_TYPE_INPUT)
518 ctx->size += CTX_SIZE(xhci->hcc_params);
519
520 ctx->bytes = dma_pool_alloc(xhci->device_pool, flags, &ctx->dma);
521 if (!ctx->bytes) {
522 kfree(ctx);
523 return NULL;
524 }
525 memset(ctx->bytes, 0, ctx->size);
526 return ctx;
527}
528
529static void xhci_free_container_ctx(struct xhci_hcd *xhci,
530 struct xhci_container_ctx *ctx)
531{
532 if (!ctx)
533 return;
534 dma_pool_free(xhci->device_pool, ctx->bytes, ctx->dma);
535 kfree(ctx);
536}
537
538struct xhci_input_control_ctx *xhci_get_input_control_ctx(struct xhci_hcd *xhci,
539 struct xhci_container_ctx *ctx)
540{
541 if (ctx->type != XHCI_CTX_TYPE_INPUT)
542 return NULL;
543
544 return (struct xhci_input_control_ctx *)ctx->bytes;
545}
546
547struct xhci_slot_ctx *xhci_get_slot_ctx(struct xhci_hcd *xhci,
548 struct xhci_container_ctx *ctx)
549{
550 if (ctx->type == XHCI_CTX_TYPE_DEVICE)
551 return (struct xhci_slot_ctx *)ctx->bytes;
552
553 return (struct xhci_slot_ctx *)
554 (ctx->bytes + CTX_SIZE(xhci->hcc_params));
555}
556
557struct xhci_ep_ctx *xhci_get_ep_ctx(struct xhci_hcd *xhci,
558 struct xhci_container_ctx *ctx,
559 unsigned int ep_index)
560{
561 /* increment ep index by offset of start of ep ctx array */
562 ep_index++;
563 if (ctx->type == XHCI_CTX_TYPE_INPUT)
564 ep_index++;
565
566 return (struct xhci_ep_ctx *)
567 (ctx->bytes + (ep_index * CTX_SIZE(xhci->hcc_params)));
568}
569
570
571/***************** Streams structures manipulation *************************/
572
573static void xhci_free_stream_ctx(struct xhci_hcd *xhci,
574 unsigned int num_stream_ctxs,
575 struct xhci_stream_ctx *stream_ctx, dma_addr_t dma)
576{
577 struct device *dev = xhci_to_hcd(xhci)->self.controller;
578 size_t size = sizeof(struct xhci_stream_ctx) * num_stream_ctxs;
579
580 if (size > MEDIUM_STREAM_ARRAY_SIZE)
581 dma_free_coherent(dev, size,
582 stream_ctx, dma);
583 else if (size <= SMALL_STREAM_ARRAY_SIZE)
584 return dma_pool_free(xhci->small_streams_pool,
585 stream_ctx, dma);
586 else
587 return dma_pool_free(xhci->medium_streams_pool,
588 stream_ctx, dma);
589}
590
591/*
592 * The stream context array for each endpoint with bulk streams enabled can
593 * vary in size, based on:
594 * - how many streams the endpoint supports,
595 * - the maximum primary stream array size the host controller supports,
596 * - and how many streams the device driver asks for.
597 *
598 * The stream context array must be a power of 2, and can be as small as
599 * 64 bytes or as large as 1MB.
600 */
601static struct xhci_stream_ctx *xhci_alloc_stream_ctx(struct xhci_hcd *xhci,
602 unsigned int num_stream_ctxs, dma_addr_t *dma,
603 gfp_t mem_flags)
604{
605 struct device *dev = xhci_to_hcd(xhci)->self.controller;
606 size_t size = sizeof(struct xhci_stream_ctx) * num_stream_ctxs;
607
608 if (size > MEDIUM_STREAM_ARRAY_SIZE)
609 return dma_alloc_coherent(dev, size,
610 dma, mem_flags);
611 else if (size <= SMALL_STREAM_ARRAY_SIZE)
612 return dma_pool_alloc(xhci->small_streams_pool,
613 mem_flags, dma);
614 else
615 return dma_pool_alloc(xhci->medium_streams_pool,
616 mem_flags, dma);
617}
618
619struct xhci_ring *xhci_dma_to_transfer_ring(
620 struct xhci_virt_ep *ep,
621 u64 address)
622{
623 if (ep->ep_state & EP_HAS_STREAMS)
624 return radix_tree_lookup(&ep->stream_info->trb_address_map,
625 address >> TRB_SEGMENT_SHIFT);
626 return ep->ring;
627}
628
629struct xhci_ring *xhci_stream_id_to_ring(
630 struct xhci_virt_device *dev,
631 unsigned int ep_index,
632 unsigned int stream_id)
633{
634 struct xhci_virt_ep *ep = &dev->eps[ep_index];
635
636 if (stream_id == 0)
637 return ep->ring;
638 if (!ep->stream_info)
639 return NULL;
640
641 if (stream_id > ep->stream_info->num_streams)
642 return NULL;
643 return ep->stream_info->stream_rings[stream_id];
644}
645
646/*
647 * Change an endpoint's internal structure so it supports stream IDs. The
648 * number of requested streams includes stream 0, which cannot be used by device
649 * drivers.
650 *
651 * The number of stream contexts in the stream context array may be bigger than
652 * the number of streams the driver wants to use. This is because the number of
653 * stream context array entries must be a power of two.
654 */
655struct xhci_stream_info *xhci_alloc_stream_info(struct xhci_hcd *xhci,
656 unsigned int num_stream_ctxs,
657 unsigned int num_streams, gfp_t mem_flags)
658{
659 struct xhci_stream_info *stream_info;
660 u32 cur_stream;
661 struct xhci_ring *cur_ring;
662 u64 addr;
663 int ret;
664
665 xhci_dbg(xhci, "Allocating %u streams and %u "
666 "stream context array entries.\n",
667 num_streams, num_stream_ctxs);
668 if (xhci->cmd_ring_reserved_trbs == MAX_RSVD_CMD_TRBS) {
669 xhci_dbg(xhci, "Command ring has no reserved TRBs available\n");
670 return NULL;
671 }
672 xhci->cmd_ring_reserved_trbs++;
673
674 stream_info = kzalloc(sizeof(struct xhci_stream_info), mem_flags);
675 if (!stream_info)
676 goto cleanup_trbs;
677
678 stream_info->num_streams = num_streams;
679 stream_info->num_stream_ctxs = num_stream_ctxs;
680
681 /* Initialize the array of virtual pointers to stream rings. */
682 stream_info->stream_rings = kzalloc(
683 sizeof(struct xhci_ring *)*num_streams,
684 mem_flags);
685 if (!stream_info->stream_rings)
686 goto cleanup_info;
687
688 /* Initialize the array of DMA addresses for stream rings for the HW. */
689 stream_info->stream_ctx_array = xhci_alloc_stream_ctx(xhci,
690 num_stream_ctxs, &stream_info->ctx_array_dma,
691 mem_flags);
692 if (!stream_info->stream_ctx_array)
693 goto cleanup_ctx;
694 memset(stream_info->stream_ctx_array, 0,
695 sizeof(struct xhci_stream_ctx)*num_stream_ctxs);
696
697 /* Allocate everything needed to free the stream rings later */
698 stream_info->free_streams_command =
699 xhci_alloc_command(xhci, true, true, mem_flags);
700 if (!stream_info->free_streams_command)
701 goto cleanup_ctx;
702
703 INIT_RADIX_TREE(&stream_info->trb_address_map, GFP_ATOMIC);
704
705 /* Allocate rings for all the streams that the driver will use,
706 * and add their segment DMA addresses to the radix tree.
707 * Stream 0 is reserved.
708 */
709 for (cur_stream = 1; cur_stream < num_streams; cur_stream++) {
710 stream_info->stream_rings[cur_stream] =
711 xhci_ring_alloc(xhci, 2, 1, TYPE_STREAM, mem_flags);
712 cur_ring = stream_info->stream_rings[cur_stream];
713 if (!cur_ring)
714 goto cleanup_rings;
715 cur_ring->stream_id = cur_stream;
716 cur_ring->trb_address_map = &stream_info->trb_address_map;
717 /* Set deq ptr, cycle bit, and stream context type */
718 addr = cur_ring->first_seg->dma |
719 SCT_FOR_CTX(SCT_PRI_TR) |
720 cur_ring->cycle_state;
721 stream_info->stream_ctx_array[cur_stream].stream_ring =
722 cpu_to_le64(addr);
723 xhci_dbg(xhci, "Setting stream %d ring ptr to 0x%08llx\n",
724 cur_stream, (unsigned long long) addr);
725
726 ret = xhci_update_stream_mapping(cur_ring, mem_flags);
727 if (ret) {
728 xhci_ring_free(xhci, cur_ring);
729 stream_info->stream_rings[cur_stream] = NULL;
730 goto cleanup_rings;
731 }
732 }
733 /* Leave the other unused stream ring pointers in the stream context
734 * array initialized to zero. This will cause the xHC to give us an
735 * error if the device asks for a stream ID we don't have setup (if it
736 * was any other way, the host controller would assume the ring is
737 * "empty" and wait forever for data to be queued to that stream ID).
738 */
739
740 return stream_info;
741
742cleanup_rings:
743 for (cur_stream = 1; cur_stream < num_streams; cur_stream++) {
744 cur_ring = stream_info->stream_rings[cur_stream];
745 if (cur_ring) {
746 xhci_ring_free(xhci, cur_ring);
747 stream_info->stream_rings[cur_stream] = NULL;
748 }
749 }
750 xhci_free_command(xhci, stream_info->free_streams_command);
751cleanup_ctx:
752 kfree(stream_info->stream_rings);
753cleanup_info:
754 kfree(stream_info);
755cleanup_trbs:
756 xhci->cmd_ring_reserved_trbs--;
757 return NULL;
758}
759/*
760 * Sets the MaxPStreams field and the Linear Stream Array field.
761 * Sets the dequeue pointer to the stream context array.
762 */
763void xhci_setup_streams_ep_input_ctx(struct xhci_hcd *xhci,
764 struct xhci_ep_ctx *ep_ctx,
765 struct xhci_stream_info *stream_info)
766{
767 u32 max_primary_streams;
768 /* MaxPStreams is the number of stream context array entries, not the
769 * number we're actually using. Must be in 2^(MaxPstreams + 1) format.
770 * fls(0) = 0, fls(0x1) = 1, fls(0x10) = 2, fls(0x100) = 3, etc.
771 */
772 max_primary_streams = fls(stream_info->num_stream_ctxs) - 2;
773 xhci_dbg_trace(xhci, trace_xhci_dbg_context_change,
774 "Setting number of stream ctx array entries to %u",
775 1 << (max_primary_streams + 1));
776 ep_ctx->ep_info &= cpu_to_le32(~EP_MAXPSTREAMS_MASK);
777 ep_ctx->ep_info |= cpu_to_le32(EP_MAXPSTREAMS(max_primary_streams)
778 | EP_HAS_LSA);
779 ep_ctx->deq = cpu_to_le64(stream_info->ctx_array_dma);
780}
781
782/*
783 * Sets the MaxPStreams field and the Linear Stream Array field to 0.
784 * Reinstalls the "normal" endpoint ring (at its previous dequeue mark,
785 * not at the beginning of the ring).
786 */
787void xhci_setup_no_streams_ep_input_ctx(struct xhci_hcd *xhci,
788 struct xhci_ep_ctx *ep_ctx,
789 struct xhci_virt_ep *ep)
790{
791 dma_addr_t addr;
792 ep_ctx->ep_info &= cpu_to_le32(~(EP_MAXPSTREAMS_MASK | EP_HAS_LSA));
793 addr = xhci_trb_virt_to_dma(ep->ring->deq_seg, ep->ring->dequeue);
794 ep_ctx->deq = cpu_to_le64(addr | ep->ring->cycle_state);
795}
796
797/* Frees all stream contexts associated with the endpoint,
798 *
799 * Caller should fix the endpoint context streams fields.
800 */
801void xhci_free_stream_info(struct xhci_hcd *xhci,
802 struct xhci_stream_info *stream_info)
803{
804 int cur_stream;
805 struct xhci_ring *cur_ring;
806
807 if (!stream_info)
808 return;
809
810 for (cur_stream = 1; cur_stream < stream_info->num_streams;
811 cur_stream++) {
812 cur_ring = stream_info->stream_rings[cur_stream];
813 if (cur_ring) {
814 xhci_ring_free(xhci, cur_ring);
815 stream_info->stream_rings[cur_stream] = NULL;
816 }
817 }
818 xhci_free_command(xhci, stream_info->free_streams_command);
819 xhci->cmd_ring_reserved_trbs--;
820 if (stream_info->stream_ctx_array)
821 xhci_free_stream_ctx(xhci,
822 stream_info->num_stream_ctxs,
823 stream_info->stream_ctx_array,
824 stream_info->ctx_array_dma);
825
826 kfree(stream_info->stream_rings);
827 kfree(stream_info);
828}
829
830
831/***************** Device context manipulation *************************/
832
833static void xhci_init_endpoint_timer(struct xhci_hcd *xhci,
834 struct xhci_virt_ep *ep)
835{
836 init_timer(&ep->stop_cmd_timer);
837 ep->stop_cmd_timer.data = (unsigned long) ep;
838 ep->stop_cmd_timer.function = xhci_stop_endpoint_command_watchdog;
839 ep->xhci = xhci;
840}
841
842static void xhci_free_tt_info(struct xhci_hcd *xhci,
843 struct xhci_virt_device *virt_dev,
844 int slot_id)
845{
846 struct list_head *tt_list_head;
847 struct xhci_tt_bw_info *tt_info, *next;
848 bool slot_found = false;
849
850 /* If the device never made it past the Set Address stage,
851 * it may not have the real_port set correctly.
852 */
853 if (virt_dev->real_port == 0 ||
854 virt_dev->real_port > HCS_MAX_PORTS(xhci->hcs_params1)) {
855 xhci_dbg(xhci, "Bad real port.\n");
856 return;
857 }
858
859 tt_list_head = &(xhci->rh_bw[virt_dev->real_port - 1].tts);
860 list_for_each_entry_safe(tt_info, next, tt_list_head, tt_list) {
861 /* Multi-TT hubs will have more than one entry */
862 if (tt_info->slot_id == slot_id) {
863 slot_found = true;
864 list_del(&tt_info->tt_list);
865 kfree(tt_info);
866 } else if (slot_found) {
867 break;
868 }
869 }
870}
871
872int xhci_alloc_tt_info(struct xhci_hcd *xhci,
873 struct xhci_virt_device *virt_dev,
874 struct usb_device *hdev,
875 struct usb_tt *tt, gfp_t mem_flags)
876{
877 struct xhci_tt_bw_info *tt_info;
878 unsigned int num_ports;
879 int i, j;
880
881 if (!tt->multi)
882 num_ports = 1;
883 else
884 num_ports = hdev->maxchild;
885
886 for (i = 0; i < num_ports; i++, tt_info++) {
887 struct xhci_interval_bw_table *bw_table;
888
889 tt_info = kzalloc(sizeof(*tt_info), mem_flags);
890 if (!tt_info)
891 goto free_tts;
892 INIT_LIST_HEAD(&tt_info->tt_list);
893 list_add(&tt_info->tt_list,
894 &xhci->rh_bw[virt_dev->real_port - 1].tts);
895 tt_info->slot_id = virt_dev->udev->slot_id;
896 if (tt->multi)
897 tt_info->ttport = i+1;
898 bw_table = &tt_info->bw_table;
899 for (j = 0; j < XHCI_MAX_INTERVAL; j++)
900 INIT_LIST_HEAD(&bw_table->interval_bw[j].endpoints);
901 }
902 return 0;
903
904free_tts:
905 xhci_free_tt_info(xhci, virt_dev, virt_dev->udev->slot_id);
906 return -ENOMEM;
907}
908
909
910/* All the xhci_tds in the ring's TD list should be freed at this point.
911 * Should be called with xhci->lock held if there is any chance the TT lists
912 * will be manipulated by the configure endpoint, allocate device, or update
913 * hub functions while this function is removing the TT entries from the list.
914 */
915void xhci_free_virt_device(struct xhci_hcd *xhci, int slot_id)
916{
917 struct xhci_virt_device *dev;
918 int i;
919 int old_active_eps = 0;
920
921 /* Slot ID 0 is reserved */
922 if (slot_id == 0 || !xhci->devs[slot_id])
923 return;
924
925 dev = xhci->devs[slot_id];
926 xhci->dcbaa->dev_context_ptrs[slot_id] = 0;
927 if (!dev)
928 return;
929
930 if (dev->tt_info)
931 old_active_eps = dev->tt_info->active_eps;
932
933 for (i = 0; i < 31; ++i) {
934 if (dev->eps[i].ring)
935 xhci_ring_free(xhci, dev->eps[i].ring);
936 if (dev->eps[i].stream_info)
937 xhci_free_stream_info(xhci,
938 dev->eps[i].stream_info);
939 /* Endpoints on the TT/root port lists should have been removed
940 * when usb_disable_device() was called for the device.
941 * We can't drop them anyway, because the udev might have gone
942 * away by this point, and we can't tell what speed it was.
943 */
944 if (!list_empty(&dev->eps[i].bw_endpoint_list))
945 xhci_warn(xhci, "Slot %u endpoint %u "
946 "not removed from BW list!\n",
947 slot_id, i);
948 }
949 /* If this is a hub, free the TT(s) from the TT list */
950 xhci_free_tt_info(xhci, dev, slot_id);
951 /* If necessary, update the number of active TTs on this root port */
952 xhci_update_tt_active_eps(xhci, dev, old_active_eps);
953
954 if (dev->ring_cache) {
955 for (i = 0; i < dev->num_rings_cached; i++)
956 xhci_ring_free(xhci, dev->ring_cache[i]);
957 kfree(dev->ring_cache);
958 }
959
960 if (dev->in_ctx)
961 xhci_free_container_ctx(xhci, dev->in_ctx);
962 if (dev->out_ctx)
963 xhci_free_container_ctx(xhci, dev->out_ctx);
964
965 kfree(xhci->devs[slot_id]);
966 xhci->devs[slot_id] = NULL;
967}
968
969int xhci_alloc_virt_device(struct xhci_hcd *xhci, int slot_id,
970 struct usb_device *udev, gfp_t flags)
971{
972 struct xhci_virt_device *dev;
973 int i;
974
975 /* Slot ID 0 is reserved */
976 if (slot_id == 0 || xhci->devs[slot_id]) {
977 xhci_warn(xhci, "Bad Slot ID %d\n", slot_id);
978 return 0;
979 }
980
981 xhci->devs[slot_id] = kzalloc(sizeof(*xhci->devs[slot_id]), flags);
982 if (!xhci->devs[slot_id])
983 return 0;
984 dev = xhci->devs[slot_id];
985
986 /* Allocate the (output) device context that will be used in the HC. */
987 dev->out_ctx = xhci_alloc_container_ctx(xhci, XHCI_CTX_TYPE_DEVICE, flags);
988 if (!dev->out_ctx)
989 goto fail;
990
991 xhci_dbg(xhci, "Slot %d output ctx = 0x%llx (dma)\n", slot_id,
992 (unsigned long long)dev->out_ctx->dma);
993
994 /* Allocate the (input) device context for address device command */
995 dev->in_ctx = xhci_alloc_container_ctx(xhci, XHCI_CTX_TYPE_INPUT, flags);
996 if (!dev->in_ctx)
997 goto fail;
998
999 xhci_dbg(xhci, "Slot %d input ctx = 0x%llx (dma)\n", slot_id,
1000 (unsigned long long)dev->in_ctx->dma);
1001
1002 /* Initialize the cancellation list and watchdog timers for each ep */
1003 for (i = 0; i < 31; i++) {
1004 xhci_init_endpoint_timer(xhci, &dev->eps[i]);
1005 INIT_LIST_HEAD(&dev->eps[i].cancelled_td_list);
1006 INIT_LIST_HEAD(&dev->eps[i].bw_endpoint_list);
1007 }
1008
1009 /* Allocate endpoint 0 ring */
1010 dev->eps[0].ring = xhci_ring_alloc(xhci, 2, 1, TYPE_CTRL, flags);
1011 if (!dev->eps[0].ring)
1012 goto fail;
1013
1014 /* Allocate pointers to the ring cache */
1015 dev->ring_cache = kzalloc(
1016 sizeof(struct xhci_ring *)*XHCI_MAX_RINGS_CACHED,
1017 flags);
1018 if (!dev->ring_cache)
1019 goto fail;
1020 dev->num_rings_cached = 0;
1021
1022 init_completion(&dev->cmd_completion);
1023 INIT_LIST_HEAD(&dev->cmd_list);
1024 dev->udev = udev;
1025
1026 /* Point to output device context in dcbaa. */
1027 xhci->dcbaa->dev_context_ptrs[slot_id] = cpu_to_le64(dev->out_ctx->dma);
1028 xhci_dbg(xhci, "Set slot id %d dcbaa entry %p to 0x%llx\n",
1029 slot_id,
1030 &xhci->dcbaa->dev_context_ptrs[slot_id],
1031 le64_to_cpu(xhci->dcbaa->dev_context_ptrs[slot_id]));
1032
1033 return 1;
1034fail:
1035 xhci_free_virt_device(xhci, slot_id);
1036 return 0;
1037}
1038
1039void xhci_copy_ep0_dequeue_into_input_ctx(struct xhci_hcd *xhci,
1040 struct usb_device *udev)
1041{
1042 struct xhci_virt_device *virt_dev;
1043 struct xhci_ep_ctx *ep0_ctx;
1044 struct xhci_ring *ep_ring;
1045
1046 virt_dev = xhci->devs[udev->slot_id];
1047 ep0_ctx = xhci_get_ep_ctx(xhci, virt_dev->in_ctx, 0);
1048 ep_ring = virt_dev->eps[0].ring;
1049 /*
1050 * FIXME we don't keep track of the dequeue pointer very well after a
1051 * Set TR dequeue pointer, so we're setting the dequeue pointer of the
1052 * host to our enqueue pointer. This should only be called after a
1053 * configured device has reset, so all control transfers should have
1054 * been completed or cancelled before the reset.
1055 */
1056 ep0_ctx->deq = cpu_to_le64(xhci_trb_virt_to_dma(ep_ring->enq_seg,
1057 ep_ring->enqueue)
1058 | ep_ring->cycle_state);
1059}
1060
1061/*
1062 * The xHCI roothub may have ports of differing speeds in any order in the port
1063 * status registers. xhci->port_array provides an array of the port speed for
1064 * each offset into the port status registers.
1065 *
1066 * The xHCI hardware wants to know the roothub port number that the USB device
1067 * is attached to (or the roothub port its ancestor hub is attached to). All we
1068 * know is the index of that port under either the USB 2.0 or the USB 3.0
1069 * roothub, but that doesn't give us the real index into the HW port status
1070 * registers. Call xhci_find_raw_port_number() to get real index.
1071 */
1072static u32 xhci_find_real_port_number(struct xhci_hcd *xhci,
1073 struct usb_device *udev)
1074{
1075 struct usb_device *top_dev;
1076 struct usb_hcd *hcd;
1077
1078 if (udev->speed == USB_SPEED_SUPER)
1079 hcd = xhci->shared_hcd;
1080 else
1081 hcd = xhci->main_hcd;
1082
1083 for (top_dev = udev; top_dev->parent && top_dev->parent->parent;
1084 top_dev = top_dev->parent)
1085 /* Found device below root hub */;
1086
1087 return xhci_find_raw_port_number(hcd, top_dev->portnum);
1088}
1089
1090/* Setup an xHCI virtual device for a Set Address command */
1091int xhci_setup_addressable_virt_dev(struct xhci_hcd *xhci, struct usb_device *udev)
1092{
1093 struct xhci_virt_device *dev;
1094 struct xhci_ep_ctx *ep0_ctx;
1095 struct xhci_slot_ctx *slot_ctx;
1096 u32 port_num;
1097 u32 max_packets;
1098 struct usb_device *top_dev;
1099
1100 dev = xhci->devs[udev->slot_id];
1101 /* Slot ID 0 is reserved */
1102 if (udev->slot_id == 0 || !dev) {
1103 xhci_warn(xhci, "Slot ID %d is not assigned to this device\n",
1104 udev->slot_id);
1105 return -EINVAL;
1106 }
1107 ep0_ctx = xhci_get_ep_ctx(xhci, dev->in_ctx, 0);
1108 slot_ctx = xhci_get_slot_ctx(xhci, dev->in_ctx);
1109
1110 /* 3) Only the control endpoint is valid - one endpoint context */
1111 slot_ctx->dev_info |= cpu_to_le32(LAST_CTX(1) | udev->route);
1112 switch (udev->speed) {
1113 case USB_SPEED_SUPER:
1114 slot_ctx->dev_info |= cpu_to_le32(SLOT_SPEED_SS);
1115 max_packets = MAX_PACKET(512);
1116 break;
1117 case USB_SPEED_HIGH:
1118 slot_ctx->dev_info |= cpu_to_le32(SLOT_SPEED_HS);
1119 max_packets = MAX_PACKET(64);
1120 break;
1121 /* USB core guesses at a 64-byte max packet first for FS devices */
1122 case USB_SPEED_FULL:
1123 slot_ctx->dev_info |= cpu_to_le32(SLOT_SPEED_FS);
1124 max_packets = MAX_PACKET(64);
1125 break;
1126 case USB_SPEED_LOW:
1127 slot_ctx->dev_info |= cpu_to_le32(SLOT_SPEED_LS);
1128 max_packets = MAX_PACKET(8);
1129 break;
1130 case USB_SPEED_WIRELESS:
1131 xhci_dbg(xhci, "FIXME xHCI doesn't support wireless speeds\n");
1132 return -EINVAL;
1133 break;
1134 default:
1135 /* Speed was set earlier, this shouldn't happen. */
1136 return -EINVAL;
1137 }
1138 /* Find the root hub port this device is under */
1139 port_num = xhci_find_real_port_number(xhci, udev);
1140 if (!port_num)
1141 return -EINVAL;
1142 slot_ctx->dev_info2 |= cpu_to_le32(ROOT_HUB_PORT(port_num));
1143 /* Set the port number in the virtual_device to the faked port number */
1144 for (top_dev = udev; top_dev->parent && top_dev->parent->parent;
1145 top_dev = top_dev->parent)
1146 /* Found device below root hub */;
1147 dev->fake_port = top_dev->portnum;
1148 dev->real_port = port_num;
1149 xhci_dbg(xhci, "Set root hub portnum to %d\n", port_num);
1150 xhci_dbg(xhci, "Set fake root hub portnum to %d\n", dev->fake_port);
1151
1152 /* Find the right bandwidth table that this device will be a part of.
1153 * If this is a full speed device attached directly to a root port (or a
1154 * decendent of one), it counts as a primary bandwidth domain, not a
1155 * secondary bandwidth domain under a TT. An xhci_tt_info structure
1156 * will never be created for the HS root hub.
1157 */
1158 if (!udev->tt || !udev->tt->hub->parent) {
1159 dev->bw_table = &xhci->rh_bw[port_num - 1].bw_table;
1160 } else {
1161 struct xhci_root_port_bw_info *rh_bw;
1162 struct xhci_tt_bw_info *tt_bw;
1163
1164 rh_bw = &xhci->rh_bw[port_num - 1];
1165 /* Find the right TT. */
1166 list_for_each_entry(tt_bw, &rh_bw->tts, tt_list) {
1167 if (tt_bw->slot_id != udev->tt->hub->slot_id)
1168 continue;
1169
1170 if (!dev->udev->tt->multi ||
1171 (udev->tt->multi &&
1172 tt_bw->ttport == dev->udev->ttport)) {
1173 dev->bw_table = &tt_bw->bw_table;
1174 dev->tt_info = tt_bw;
1175 break;
1176 }
1177 }
1178 if (!dev->tt_info)
1179 xhci_warn(xhci, "WARN: Didn't find a matching TT\n");
1180 }
1181
1182 /* Is this a LS/FS device under an external HS hub? */
1183 if (udev->tt && udev->tt->hub->parent) {
1184 slot_ctx->tt_info = cpu_to_le32(udev->tt->hub->slot_id |
1185 (udev->ttport << 8));
1186 if (udev->tt->multi)
1187 slot_ctx->dev_info |= cpu_to_le32(DEV_MTT);
1188 }
1189 xhci_dbg(xhci, "udev->tt = %p\n", udev->tt);
1190 xhci_dbg(xhci, "udev->ttport = 0x%x\n", udev->ttport);
1191
1192 /* Step 4 - ring already allocated */
1193 /* Step 5 */
1194 ep0_ctx->ep_info2 = cpu_to_le32(EP_TYPE(CTRL_EP));
1195
1196 /* EP 0 can handle "burst" sizes of 1, so Max Burst Size field is 0 */
1197 ep0_ctx->ep_info2 |= cpu_to_le32(MAX_BURST(0) | ERROR_COUNT(3) |
1198 max_packets);
1199
1200 ep0_ctx->deq = cpu_to_le64(dev->eps[0].ring->first_seg->dma |
1201 dev->eps[0].ring->cycle_state);
1202
1203 /* Steps 7 and 8 were done in xhci_alloc_virt_device() */
1204
1205 return 0;
1206}
1207
1208/*
1209 * Convert interval expressed as 2^(bInterval - 1) == interval into
1210 * straight exponent value 2^n == interval.
1211 *
1212 */
1213static unsigned int xhci_parse_exponent_interval(struct usb_device *udev,
1214 struct usb_host_endpoint *ep)
1215{
1216 unsigned int interval;
1217
1218 interval = clamp_val(ep->desc.bInterval, 1, 16) - 1;
1219 if (interval != ep->desc.bInterval - 1)
1220 dev_warn(&udev->dev,
1221 "ep %#x - rounding interval to %d %sframes\n",
1222 ep->desc.bEndpointAddress,
1223 1 << interval,
1224 udev->speed == USB_SPEED_FULL ? "" : "micro");
1225
1226 if (udev->speed == USB_SPEED_FULL) {
1227 /*
1228 * Full speed isoc endpoints specify interval in frames,
1229 * not microframes. We are using microframes everywhere,
1230 * so adjust accordingly.
1231 */
1232 interval += 3; /* 1 frame = 2^3 uframes */
1233 }
1234
1235 return interval;
1236}
1237
1238/*
1239 * Convert bInterval expressed in microframes (in 1-255 range) to exponent of
1240 * microframes, rounded down to nearest power of 2.
1241 */
1242static unsigned int xhci_microframes_to_exponent(struct usb_device *udev,
1243 struct usb_host_endpoint *ep, unsigned int desc_interval,
1244 unsigned int min_exponent, unsigned int max_exponent)
1245{
1246 unsigned int interval;
1247
1248 interval = fls(desc_interval) - 1;
1249 interval = clamp_val(interval, min_exponent, max_exponent);
1250 if ((1 << interval) != desc_interval)
1251 dev_warn(&udev->dev,
1252 "ep %#x - rounding interval to %d microframes, ep desc says %d microframes\n",
1253 ep->desc.bEndpointAddress,
1254 1 << interval,
1255 desc_interval);
1256
1257 return interval;
1258}
1259
1260static unsigned int xhci_parse_microframe_interval(struct usb_device *udev,
1261 struct usb_host_endpoint *ep)
1262{
1263 if (ep->desc.bInterval == 0)
1264 return 0;
1265 return xhci_microframes_to_exponent(udev, ep,
1266 ep->desc.bInterval, 0, 15);
1267}
1268
1269
1270static unsigned int xhci_parse_frame_interval(struct usb_device *udev,
1271 struct usb_host_endpoint *ep)
1272{
1273 return xhci_microframes_to_exponent(udev, ep,
1274 ep->desc.bInterval * 8, 3, 10);
1275}
1276
1277/* Return the polling or NAK interval.
1278 *
1279 * The polling interval is expressed in "microframes". If xHCI's Interval field
1280 * is set to N, it will service the endpoint every 2^(Interval)*125us.
1281 *
1282 * The NAK interval is one NAK per 1 to 255 microframes, or no NAKs if interval
1283 * is set to 0.
1284 */
1285static unsigned int xhci_get_endpoint_interval(struct usb_device *udev,
1286 struct usb_host_endpoint *ep)
1287{
1288 unsigned int interval = 0;
1289
1290 switch (udev->speed) {
1291 case USB_SPEED_HIGH:
1292 /* Max NAK rate */
1293 if (usb_endpoint_xfer_control(&ep->desc) ||
1294 usb_endpoint_xfer_bulk(&ep->desc)) {
1295 interval = xhci_parse_microframe_interval(udev, ep);
1296 break;
1297 }
1298 /* Fall through - SS and HS isoc/int have same decoding */
1299
1300 case USB_SPEED_SUPER:
1301 if (usb_endpoint_xfer_int(&ep->desc) ||
1302 usb_endpoint_xfer_isoc(&ep->desc)) {
1303 interval = xhci_parse_exponent_interval(udev, ep);
1304 }
1305 break;
1306
1307 case USB_SPEED_FULL:
1308 if (usb_endpoint_xfer_isoc(&ep->desc)) {
1309 interval = xhci_parse_exponent_interval(udev, ep);
1310 break;
1311 }
1312 /*
1313 * Fall through for interrupt endpoint interval decoding
1314 * since it uses the same rules as low speed interrupt
1315 * endpoints.
1316 */
1317
1318 case USB_SPEED_LOW:
1319 if (usb_endpoint_xfer_int(&ep->desc) ||
1320 usb_endpoint_xfer_isoc(&ep->desc)) {
1321
1322 interval = xhci_parse_frame_interval(udev, ep);
1323 }
1324 break;
1325
1326 default:
1327 BUG();
1328 }
1329 return EP_INTERVAL(interval);
1330}
1331
1332/* The "Mult" field in the endpoint context is only set for SuperSpeed isoc eps.
1333 * High speed endpoint descriptors can define "the number of additional
1334 * transaction opportunities per microframe", but that goes in the Max Burst
1335 * endpoint context field.
1336 */
1337static u32 xhci_get_endpoint_mult(struct usb_device *udev,
1338 struct usb_host_endpoint *ep)
1339{
1340 if (udev->speed != USB_SPEED_SUPER ||
1341 !usb_endpoint_xfer_isoc(&ep->desc))
1342 return 0;
1343 return ep->ss_ep_comp.bmAttributes;
1344}
1345
1346static u32 xhci_get_endpoint_type(struct usb_device *udev,
1347 struct usb_host_endpoint *ep)
1348{
1349 int in;
1350 u32 type;
1351
1352 in = usb_endpoint_dir_in(&ep->desc);
1353 if (usb_endpoint_xfer_control(&ep->desc)) {
1354 type = EP_TYPE(CTRL_EP);
1355 } else if (usb_endpoint_xfer_bulk(&ep->desc)) {
1356 if (in)
1357 type = EP_TYPE(BULK_IN_EP);
1358 else
1359 type = EP_TYPE(BULK_OUT_EP);
1360 } else if (usb_endpoint_xfer_isoc(&ep->desc)) {
1361 if (in)
1362 type = EP_TYPE(ISOC_IN_EP);
1363 else
1364 type = EP_TYPE(ISOC_OUT_EP);
1365 } else if (usb_endpoint_xfer_int(&ep->desc)) {
1366 if (in)
1367 type = EP_TYPE(INT_IN_EP);
1368 else
1369 type = EP_TYPE(INT_OUT_EP);
1370 } else {
1371 type = 0;
1372 }
1373 return type;
1374}
1375
1376/* Return the maximum endpoint service interval time (ESIT) payload.
1377 * Basically, this is the maxpacket size, multiplied by the burst size
1378 * and mult size.
1379 */
1380static u32 xhci_get_max_esit_payload(struct xhci_hcd *xhci,
1381 struct usb_device *udev,
1382 struct usb_host_endpoint *ep)
1383{
1384 int max_burst;
1385 int max_packet;
1386
1387 /* Only applies for interrupt or isochronous endpoints */
1388 if (usb_endpoint_xfer_control(&ep->desc) ||
1389 usb_endpoint_xfer_bulk(&ep->desc))
1390 return 0;
1391
1392 if (udev->speed == USB_SPEED_SUPER)
1393 return le16_to_cpu(ep->ss_ep_comp.wBytesPerInterval);
1394
1395 max_packet = GET_MAX_PACKET(usb_endpoint_maxp(&ep->desc));
1396 max_burst = (usb_endpoint_maxp(&ep->desc) & 0x1800) >> 11;
1397 /* A 0 in max burst means 1 transfer per ESIT */
1398 return max_packet * (max_burst + 1);
1399}
1400
1401/* Set up an endpoint with one ring segment. Do not allocate stream rings.
1402 * Drivers will have to call usb_alloc_streams() to do that.
1403 */
1404int xhci_endpoint_init(struct xhci_hcd *xhci,
1405 struct xhci_virt_device *virt_dev,
1406 struct usb_device *udev,
1407 struct usb_host_endpoint *ep,
1408 gfp_t mem_flags)
1409{
1410 unsigned int ep_index;
1411 struct xhci_ep_ctx *ep_ctx;
1412 struct xhci_ring *ep_ring;
1413 unsigned int max_packet;
1414 unsigned int max_burst;
1415 enum xhci_ring_type type;
1416 u32 max_esit_payload;
1417 u32 endpoint_type;
1418
1419 ep_index = xhci_get_endpoint_index(&ep->desc);
1420 ep_ctx = xhci_get_ep_ctx(xhci, virt_dev->in_ctx, ep_index);
1421
1422 endpoint_type = xhci_get_endpoint_type(udev, ep);
1423 if (!endpoint_type)
1424 return -EINVAL;
1425 ep_ctx->ep_info2 = cpu_to_le32(endpoint_type);
1426
1427 type = usb_endpoint_type(&ep->desc);
1428 /* Set up the endpoint ring */
1429 virt_dev->eps[ep_index].new_ring =
1430 xhci_ring_alloc(xhci, 2, 1, type, mem_flags);
1431 if (!virt_dev->eps[ep_index].new_ring) {
1432 /* Attempt to use the ring cache */
1433 if (virt_dev->num_rings_cached == 0)
1434 return -ENOMEM;
1435 virt_dev->eps[ep_index].new_ring =
1436 virt_dev->ring_cache[virt_dev->num_rings_cached];
1437 virt_dev->ring_cache[virt_dev->num_rings_cached] = NULL;
1438 virt_dev->num_rings_cached--;
1439 xhci_reinit_cached_ring(xhci, virt_dev->eps[ep_index].new_ring,
1440 1, type);
1441 }
1442 virt_dev->eps[ep_index].skip = false;
1443 ep_ring = virt_dev->eps[ep_index].new_ring;
1444 ep_ctx->deq = cpu_to_le64(ep_ring->first_seg->dma | ep_ring->cycle_state);
1445
1446 ep_ctx->ep_info = cpu_to_le32(xhci_get_endpoint_interval(udev, ep)
1447 | EP_MULT(xhci_get_endpoint_mult(udev, ep)));
1448
1449 /* FIXME dig Mult and streams info out of ep companion desc */
1450
1451 /* Allow 3 retries for everything but isoc;
1452 * CErr shall be set to 0 for Isoch endpoints.
1453 */
1454 if (!usb_endpoint_xfer_isoc(&ep->desc))
1455 ep_ctx->ep_info2 |= cpu_to_le32(ERROR_COUNT(3));
1456 else
1457 ep_ctx->ep_info2 |= cpu_to_le32(ERROR_COUNT(0));
1458
1459 /* Set the max packet size and max burst */
1460 max_packet = GET_MAX_PACKET(usb_endpoint_maxp(&ep->desc));
1461 max_burst = 0;
1462 switch (udev->speed) {
1463 case USB_SPEED_SUPER:
1464 /* dig out max burst from ep companion desc */
1465 max_burst = ep->ss_ep_comp.bMaxBurst;
1466 break;
1467 case USB_SPEED_HIGH:
1468 /* Some devices get this wrong */
1469 if (usb_endpoint_xfer_bulk(&ep->desc))
1470 max_packet = 512;
1471 /* bits 11:12 specify the number of additional transaction
1472 * opportunities per microframe (USB 2.0, section 9.6.6)
1473 */
1474 if (usb_endpoint_xfer_isoc(&ep->desc) ||
1475 usb_endpoint_xfer_int(&ep->desc)) {
1476 max_burst = (usb_endpoint_maxp(&ep->desc)
1477 & 0x1800) >> 11;
1478 }
1479 break;
1480 case USB_SPEED_FULL:
1481 case USB_SPEED_LOW:
1482 break;
1483 default:
1484 BUG();
1485 }
1486 ep_ctx->ep_info2 |= cpu_to_le32(MAX_PACKET(max_packet) |
1487 MAX_BURST(max_burst));
1488 max_esit_payload = xhci_get_max_esit_payload(xhci, udev, ep);
1489 ep_ctx->tx_info = cpu_to_le32(MAX_ESIT_PAYLOAD_FOR_EP(max_esit_payload));
1490
1491 /*
1492 * XXX no idea how to calculate the average TRB buffer length for bulk
1493 * endpoints, as the driver gives us no clue how big each scatter gather
1494 * list entry (or buffer) is going to be.
1495 *
1496 * For isochronous and interrupt endpoints, we set it to the max
1497 * available, until we have new API in the USB core to allow drivers to
1498 * declare how much bandwidth they actually need.
1499 *
1500 * Normally, it would be calculated by taking the total of the buffer
1501 * lengths in the TD and then dividing by the number of TRBs in a TD,
1502 * including link TRBs, No-op TRBs, and Event data TRBs. Since we don't
1503 * use Event Data TRBs, and we don't chain in a link TRB on short
1504 * transfers, we're basically dividing by 1.
1505 *
1506 * xHCI 1.0 specification indicates that the Average TRB Length should
1507 * be set to 8 for control endpoints.
1508 */
1509 if (usb_endpoint_xfer_control(&ep->desc) && xhci->hci_version == 0x100)
1510 ep_ctx->tx_info |= cpu_to_le32(AVG_TRB_LENGTH_FOR_EP(8));
1511 else
1512 ep_ctx->tx_info |=
1513 cpu_to_le32(AVG_TRB_LENGTH_FOR_EP(max_esit_payload));
1514
1515 /* FIXME Debug endpoint context */
1516 return 0;
1517}
1518
1519void xhci_endpoint_zero(struct xhci_hcd *xhci,
1520 struct xhci_virt_device *virt_dev,
1521 struct usb_host_endpoint *ep)
1522{
1523 unsigned int ep_index;
1524 struct xhci_ep_ctx *ep_ctx;
1525
1526 ep_index = xhci_get_endpoint_index(&ep->desc);
1527 ep_ctx = xhci_get_ep_ctx(xhci, virt_dev->in_ctx, ep_index);
1528
1529 ep_ctx->ep_info = 0;
1530 ep_ctx->ep_info2 = 0;
1531 ep_ctx->deq = 0;
1532 ep_ctx->tx_info = 0;
1533 /* Don't free the endpoint ring until the set interface or configuration
1534 * request succeeds.
1535 */
1536}
1537
1538void xhci_clear_endpoint_bw_info(struct xhci_bw_info *bw_info)
1539{
1540 bw_info->ep_interval = 0;
1541 bw_info->mult = 0;
1542 bw_info->num_packets = 0;
1543 bw_info->max_packet_size = 0;
1544 bw_info->type = 0;
1545 bw_info->max_esit_payload = 0;
1546}
1547
1548void xhci_update_bw_info(struct xhci_hcd *xhci,
1549 struct xhci_container_ctx *in_ctx,
1550 struct xhci_input_control_ctx *ctrl_ctx,
1551 struct xhci_virt_device *virt_dev)
1552{
1553 struct xhci_bw_info *bw_info;
1554 struct xhci_ep_ctx *ep_ctx;
1555 unsigned int ep_type;
1556 int i;
1557
1558 for (i = 1; i < 31; ++i) {
1559 bw_info = &virt_dev->eps[i].bw_info;
1560
1561 /* We can't tell what endpoint type is being dropped, but
1562 * unconditionally clearing the bandwidth info for non-periodic
1563 * endpoints should be harmless because the info will never be
1564 * set in the first place.
1565 */
1566 if (!EP_IS_ADDED(ctrl_ctx, i) && EP_IS_DROPPED(ctrl_ctx, i)) {
1567 /* Dropped endpoint */
1568 xhci_clear_endpoint_bw_info(bw_info);
1569 continue;
1570 }
1571
1572 if (EP_IS_ADDED(ctrl_ctx, i)) {
1573 ep_ctx = xhci_get_ep_ctx(xhci, in_ctx, i);
1574 ep_type = CTX_TO_EP_TYPE(le32_to_cpu(ep_ctx->ep_info2));
1575
1576 /* Ignore non-periodic endpoints */
1577 if (ep_type != ISOC_OUT_EP && ep_type != INT_OUT_EP &&
1578 ep_type != ISOC_IN_EP &&
1579 ep_type != INT_IN_EP)
1580 continue;
1581
1582 /* Added or changed endpoint */
1583 bw_info->ep_interval = CTX_TO_EP_INTERVAL(
1584 le32_to_cpu(ep_ctx->ep_info));
1585 /* Number of packets and mult are zero-based in the
1586 * input context, but we want one-based for the
1587 * interval table.
1588 */
1589 bw_info->mult = CTX_TO_EP_MULT(
1590 le32_to_cpu(ep_ctx->ep_info)) + 1;
1591 bw_info->num_packets = CTX_TO_MAX_BURST(
1592 le32_to_cpu(ep_ctx->ep_info2)) + 1;
1593 bw_info->max_packet_size = MAX_PACKET_DECODED(
1594 le32_to_cpu(ep_ctx->ep_info2));
1595 bw_info->type = ep_type;
1596 bw_info->max_esit_payload = CTX_TO_MAX_ESIT_PAYLOAD(
1597 le32_to_cpu(ep_ctx->tx_info));
1598 }
1599 }
1600}
1601
1602/* Copy output xhci_ep_ctx to the input xhci_ep_ctx copy.
1603 * Useful when you want to change one particular aspect of the endpoint and then
1604 * issue a configure endpoint command.
1605 */
1606void xhci_endpoint_copy(struct xhci_hcd *xhci,
1607 struct xhci_container_ctx *in_ctx,
1608 struct xhci_container_ctx *out_ctx,
1609 unsigned int ep_index)
1610{
1611 struct xhci_ep_ctx *out_ep_ctx;
1612 struct xhci_ep_ctx *in_ep_ctx;
1613
1614 out_ep_ctx = xhci_get_ep_ctx(xhci, out_ctx, ep_index);
1615 in_ep_ctx = xhci_get_ep_ctx(xhci, in_ctx, ep_index);
1616
1617 in_ep_ctx->ep_info = out_ep_ctx->ep_info;
1618 in_ep_ctx->ep_info2 = out_ep_ctx->ep_info2;
1619 in_ep_ctx->deq = out_ep_ctx->deq;
1620 in_ep_ctx->tx_info = out_ep_ctx->tx_info;
1621}
1622
1623/* Copy output xhci_slot_ctx to the input xhci_slot_ctx.
1624 * Useful when you want to change one particular aspect of the endpoint and then
1625 * issue a configure endpoint command. Only the context entries field matters,
1626 * but we'll copy the whole thing anyway.
1627 */
1628void xhci_slot_copy(struct xhci_hcd *xhci,
1629 struct xhci_container_ctx *in_ctx,
1630 struct xhci_container_ctx *out_ctx)
1631{
1632 struct xhci_slot_ctx *in_slot_ctx;
1633 struct xhci_slot_ctx *out_slot_ctx;
1634
1635 in_slot_ctx = xhci_get_slot_ctx(xhci, in_ctx);
1636 out_slot_ctx = xhci_get_slot_ctx(xhci, out_ctx);
1637
1638 in_slot_ctx->dev_info = out_slot_ctx->dev_info;
1639 in_slot_ctx->dev_info2 = out_slot_ctx->dev_info2;
1640 in_slot_ctx->tt_info = out_slot_ctx->tt_info;
1641 in_slot_ctx->dev_state = out_slot_ctx->dev_state;
1642}
1643
1644/* Set up the scratchpad buffer array and scratchpad buffers, if needed. */
1645static int scratchpad_alloc(struct xhci_hcd *xhci, gfp_t flags)
1646{
1647 int i;
1648 struct device *dev = xhci_to_hcd(xhci)->self.controller;
1649 int num_sp = HCS_MAX_SCRATCHPAD(xhci->hcs_params2);
1650
1651 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
1652 "Allocating %d scratchpad buffers", num_sp);
1653
1654 if (!num_sp)
1655 return 0;
1656
1657 xhci->scratchpad = kzalloc(sizeof(*xhci->scratchpad), flags);
1658 if (!xhci->scratchpad)
1659 goto fail_sp;
1660
1661 xhci->scratchpad->sp_array = dma_alloc_coherent(dev,
1662 num_sp * sizeof(u64),
1663 &xhci->scratchpad->sp_dma, flags);
1664 if (!xhci->scratchpad->sp_array)
1665 goto fail_sp2;
1666
1667 xhci->scratchpad->sp_buffers = kzalloc(sizeof(void *) * num_sp, flags);
1668 if (!xhci->scratchpad->sp_buffers)
1669 goto fail_sp3;
1670
1671 xhci->scratchpad->sp_dma_buffers =
1672 kzalloc(sizeof(dma_addr_t) * num_sp, flags);
1673
1674 if (!xhci->scratchpad->sp_dma_buffers)
1675 goto fail_sp4;
1676
1677 xhci->dcbaa->dev_context_ptrs[0] = cpu_to_le64(xhci->scratchpad->sp_dma);
1678 for (i = 0; i < num_sp; i++) {
1679 dma_addr_t dma;
1680 void *buf = dma_alloc_coherent(dev, xhci->page_size, &dma,
1681 flags);
1682 if (!buf)
1683 goto fail_sp5;
1684
1685 xhci->scratchpad->sp_array[i] = dma;
1686 xhci->scratchpad->sp_buffers[i] = buf;
1687 xhci->scratchpad->sp_dma_buffers[i] = dma;
1688 }
1689
1690 return 0;
1691
1692 fail_sp5:
1693 for (i = i - 1; i >= 0; i--) {
1694 dma_free_coherent(dev, xhci->page_size,
1695 xhci->scratchpad->sp_buffers[i],
1696 xhci->scratchpad->sp_dma_buffers[i]);
1697 }
1698 kfree(xhci->scratchpad->sp_dma_buffers);
1699
1700 fail_sp4:
1701 kfree(xhci->scratchpad->sp_buffers);
1702
1703 fail_sp3:
1704 dma_free_coherent(dev, num_sp * sizeof(u64),
1705 xhci->scratchpad->sp_array,
1706 xhci->scratchpad->sp_dma);
1707
1708 fail_sp2:
1709 kfree(xhci->scratchpad);
1710 xhci->scratchpad = NULL;
1711
1712 fail_sp:
1713 return -ENOMEM;
1714}
1715
1716static void scratchpad_free(struct xhci_hcd *xhci)
1717{
1718 int num_sp;
1719 int i;
1720 struct device *dev = xhci_to_hcd(xhci)->self.controller;
1721
1722 if (!xhci->scratchpad)
1723 return;
1724
1725 num_sp = HCS_MAX_SCRATCHPAD(xhci->hcs_params2);
1726
1727 for (i = 0; i < num_sp; i++) {
1728 dma_free_coherent(dev, xhci->page_size,
1729 xhci->scratchpad->sp_buffers[i],
1730 xhci->scratchpad->sp_dma_buffers[i]);
1731 }
1732 kfree(xhci->scratchpad->sp_dma_buffers);
1733 kfree(xhci->scratchpad->sp_buffers);
1734 dma_free_coherent(dev, num_sp * sizeof(u64),
1735 xhci->scratchpad->sp_array,
1736 xhci->scratchpad->sp_dma);
1737 kfree(xhci->scratchpad);
1738 xhci->scratchpad = NULL;
1739}
1740
1741struct xhci_command *xhci_alloc_command(struct xhci_hcd *xhci,
1742 bool allocate_in_ctx, bool allocate_completion,
1743 gfp_t mem_flags)
1744{
1745 struct xhci_command *command;
1746
1747 command = kzalloc(sizeof(*command), mem_flags);
1748 if (!command)
1749 return NULL;
1750
1751 if (allocate_in_ctx) {
1752 command->in_ctx =
1753 xhci_alloc_container_ctx(xhci, XHCI_CTX_TYPE_INPUT,
1754 mem_flags);
1755 if (!command->in_ctx) {
1756 kfree(command);
1757 return NULL;
1758 }
1759 }
1760
1761 if (allocate_completion) {
1762 command->completion =
1763 kzalloc(sizeof(struct completion), mem_flags);
1764 if (!command->completion) {
1765 xhci_free_container_ctx(xhci, command->in_ctx);
1766 kfree(command);
1767 return NULL;
1768 }
1769 init_completion(command->completion);
1770 }
1771
1772 command->status = 0;
1773 INIT_LIST_HEAD(&command->cmd_list);
1774 return command;
1775}
1776
1777void xhci_urb_free_priv(struct xhci_hcd *xhci, struct urb_priv *urb_priv)
1778{
1779 if (urb_priv) {
1780 kfree(urb_priv->td[0]);
1781 kfree(urb_priv);
1782 }
1783}
1784
1785void xhci_free_command(struct xhci_hcd *xhci,
1786 struct xhci_command *command)
1787{
1788 xhci_free_container_ctx(xhci,
1789 command->in_ctx);
1790 kfree(command->completion);
1791 kfree(command);
1792}
1793
1794void xhci_mem_cleanup(struct xhci_hcd *xhci)
1795{
1796 struct device *dev = xhci_to_hcd(xhci)->self.controller;
1797 struct xhci_cd *cur_cd, *next_cd;
1798 int size;
1799 int i, j, num_ports;
1800
1801 /* Free the Event Ring Segment Table and the actual Event Ring */
1802 size = sizeof(struct xhci_erst_entry)*(xhci->erst.num_entries);
1803 if (xhci->erst.entries)
1804 dma_free_coherent(dev, size,
1805 xhci->erst.entries, xhci->erst.erst_dma_addr);
1806 xhci->erst.entries = NULL;
1807 xhci_dbg_trace(xhci, trace_xhci_dbg_init, "Freed ERST");
1808 if (xhci->event_ring)
1809 xhci_ring_free(xhci, xhci->event_ring);
1810 xhci->event_ring = NULL;
1811 xhci_dbg_trace(xhci, trace_xhci_dbg_init, "Freed event ring");
1812
1813 if (xhci->lpm_command)
1814 xhci_free_command(xhci, xhci->lpm_command);
1815 if (xhci->cmd_ring)
1816 xhci_ring_free(xhci, xhci->cmd_ring);
1817 xhci->cmd_ring = NULL;
1818 xhci_dbg_trace(xhci, trace_xhci_dbg_init, "Freed command ring");
1819 list_for_each_entry_safe(cur_cd, next_cd,
1820 &xhci->cancel_cmd_list, cancel_cmd_list) {
1821 list_del(&cur_cd->cancel_cmd_list);
1822 kfree(cur_cd);
1823 }
1824
1825 num_ports = HCS_MAX_PORTS(xhci->hcs_params1);
1826 for (i = 0; i < num_ports; i++) {
1827 struct xhci_interval_bw_table *bwt = &xhci->rh_bw[i].bw_table;
1828 for (j = 0; j < XHCI_MAX_INTERVAL; j++) {
1829 struct list_head *ep = &bwt->interval_bw[j].endpoints;
1830 while (!list_empty(ep))
1831 list_del_init(ep->next);
1832 }
1833 }
1834
1835 for (i = 1; i < MAX_HC_SLOTS; ++i)
1836 xhci_free_virt_device(xhci, i);
1837
1838 if (xhci->segment_pool)
1839 dma_pool_destroy(xhci->segment_pool);
1840 xhci->segment_pool = NULL;
1841 xhci_dbg_trace(xhci, trace_xhci_dbg_init, "Freed segment pool");
1842
1843 if (xhci->device_pool)
1844 dma_pool_destroy(xhci->device_pool);
1845 xhci->device_pool = NULL;
1846 xhci_dbg_trace(xhci, trace_xhci_dbg_init, "Freed device context pool");
1847
1848 if (xhci->small_streams_pool)
1849 dma_pool_destroy(xhci->small_streams_pool);
1850 xhci->small_streams_pool = NULL;
1851 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
1852 "Freed small stream array pool");
1853
1854 if (xhci->medium_streams_pool)
1855 dma_pool_destroy(xhci->medium_streams_pool);
1856 xhci->medium_streams_pool = NULL;
1857 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
1858 "Freed medium stream array pool");
1859
1860 if (xhci->dcbaa)
1861 dma_free_coherent(dev, sizeof(*xhci->dcbaa),
1862 xhci->dcbaa, xhci->dcbaa->dma);
1863 xhci->dcbaa = NULL;
1864
1865 scratchpad_free(xhci);
1866
1867 if (!xhci->rh_bw)
1868 goto no_bw;
1869
1870 for (i = 0; i < num_ports; i++) {
1871 struct xhci_tt_bw_info *tt, *n;
1872 list_for_each_entry_safe(tt, n, &xhci->rh_bw[i].tts, tt_list) {
1873 list_del(&tt->tt_list);
1874 kfree(tt);
1875 }
1876 }
1877
1878no_bw:
1879 xhci->cmd_ring_reserved_trbs = 0;
1880 xhci->num_usb2_ports = 0;
1881 xhci->num_usb3_ports = 0;
1882 xhci->num_active_eps = 0;
1883 kfree(xhci->usb2_ports);
1884 kfree(xhci->usb3_ports);
1885 kfree(xhci->port_array);
1886 kfree(xhci->rh_bw);
1887 kfree(xhci->ext_caps);
1888
1889 xhci->page_size = 0;
1890 xhci->page_shift = 0;
1891 xhci->bus_state[0].bus_suspended = 0;
1892 xhci->bus_state[1].bus_suspended = 0;
1893}
1894
1895static int xhci_test_trb_in_td(struct xhci_hcd *xhci,
1896 struct xhci_segment *input_seg,
1897 union xhci_trb *start_trb,
1898 union xhci_trb *end_trb,
1899 dma_addr_t input_dma,
1900 struct xhci_segment *result_seg,
1901 char *test_name, int test_number)
1902{
1903 unsigned long long start_dma;
1904 unsigned long long end_dma;
1905 struct xhci_segment *seg;
1906
1907 start_dma = xhci_trb_virt_to_dma(input_seg, start_trb);
1908 end_dma = xhci_trb_virt_to_dma(input_seg, end_trb);
1909
1910 seg = trb_in_td(input_seg, start_trb, end_trb, input_dma);
1911 if (seg != result_seg) {
1912 xhci_warn(xhci, "WARN: %s TRB math test %d failed!\n",
1913 test_name, test_number);
1914 xhci_warn(xhci, "Tested TRB math w/ seg %p and "
1915 "input DMA 0x%llx\n",
1916 input_seg,
1917 (unsigned long long) input_dma);
1918 xhci_warn(xhci, "starting TRB %p (0x%llx DMA), "
1919 "ending TRB %p (0x%llx DMA)\n",
1920 start_trb, start_dma,
1921 end_trb, end_dma);
1922 xhci_warn(xhci, "Expected seg %p, got seg %p\n",
1923 result_seg, seg);
1924 return -1;
1925 }
1926 return 0;
1927}
1928
1929/* TRB math checks for xhci_trb_in_td(), using the command and event rings. */
1930static int xhci_check_trb_in_td_math(struct xhci_hcd *xhci, gfp_t mem_flags)
1931{
1932 struct {
1933 dma_addr_t input_dma;
1934 struct xhci_segment *result_seg;
1935 } simple_test_vector [] = {
1936 /* A zeroed DMA field should fail */
1937 { 0, NULL },
1938 /* One TRB before the ring start should fail */
1939 { xhci->event_ring->first_seg->dma - 16, NULL },
1940 /* One byte before the ring start should fail */
1941 { xhci->event_ring->first_seg->dma - 1, NULL },
1942 /* Starting TRB should succeed */
1943 { xhci->event_ring->first_seg->dma, xhci->event_ring->first_seg },
1944 /* Ending TRB should succeed */
1945 { xhci->event_ring->first_seg->dma + (TRBS_PER_SEGMENT - 1)*16,
1946 xhci->event_ring->first_seg },
1947 /* One byte after the ring end should fail */
1948 { xhci->event_ring->first_seg->dma + (TRBS_PER_SEGMENT - 1)*16 + 1, NULL },
1949 /* One TRB after the ring end should fail */
1950 { xhci->event_ring->first_seg->dma + (TRBS_PER_SEGMENT)*16, NULL },
1951 /* An address of all ones should fail */
1952 { (dma_addr_t) (~0), NULL },
1953 };
1954 struct {
1955 struct xhci_segment *input_seg;
1956 union xhci_trb *start_trb;
1957 union xhci_trb *end_trb;
1958 dma_addr_t input_dma;
1959 struct xhci_segment *result_seg;
1960 } complex_test_vector [] = {
1961 /* Test feeding a valid DMA address from a different ring */
1962 { .input_seg = xhci->event_ring->first_seg,
1963 .start_trb = xhci->event_ring->first_seg->trbs,
1964 .end_trb = &xhci->event_ring->first_seg->trbs[TRBS_PER_SEGMENT - 1],
1965 .input_dma = xhci->cmd_ring->first_seg->dma,
1966 .result_seg = NULL,
1967 },
1968 /* Test feeding a valid end TRB from a different ring */
1969 { .input_seg = xhci->event_ring->first_seg,
1970 .start_trb = xhci->event_ring->first_seg->trbs,
1971 .end_trb = &xhci->cmd_ring->first_seg->trbs[TRBS_PER_SEGMENT - 1],
1972 .input_dma = xhci->cmd_ring->first_seg->dma,
1973 .result_seg = NULL,
1974 },
1975 /* Test feeding a valid start and end TRB from a different ring */
1976 { .input_seg = xhci->event_ring->first_seg,
1977 .start_trb = xhci->cmd_ring->first_seg->trbs,
1978 .end_trb = &xhci->cmd_ring->first_seg->trbs[TRBS_PER_SEGMENT - 1],
1979 .input_dma = xhci->cmd_ring->first_seg->dma,
1980 .result_seg = NULL,
1981 },
1982 /* TRB in this ring, but after this TD */
1983 { .input_seg = xhci->event_ring->first_seg,
1984 .start_trb = &xhci->event_ring->first_seg->trbs[0],
1985 .end_trb = &xhci->event_ring->first_seg->trbs[3],
1986 .input_dma = xhci->event_ring->first_seg->dma + 4*16,
1987 .result_seg = NULL,
1988 },
1989 /* TRB in this ring, but before this TD */
1990 { .input_seg = xhci->event_ring->first_seg,
1991 .start_trb = &xhci->event_ring->first_seg->trbs[3],
1992 .end_trb = &xhci->event_ring->first_seg->trbs[6],
1993 .input_dma = xhci->event_ring->first_seg->dma + 2*16,
1994 .result_seg = NULL,
1995 },
1996 /* TRB in this ring, but after this wrapped TD */
1997 { .input_seg = xhci->event_ring->first_seg,
1998 .start_trb = &xhci->event_ring->first_seg->trbs[TRBS_PER_SEGMENT - 3],
1999 .end_trb = &xhci->event_ring->first_seg->trbs[1],
2000 .input_dma = xhci->event_ring->first_seg->dma + 2*16,
2001 .result_seg = NULL,
2002 },
2003 /* TRB in this ring, but before this wrapped TD */
2004 { .input_seg = xhci->event_ring->first_seg,
2005 .start_trb = &xhci->event_ring->first_seg->trbs[TRBS_PER_SEGMENT - 3],
2006 .end_trb = &xhci->event_ring->first_seg->trbs[1],
2007 .input_dma = xhci->event_ring->first_seg->dma + (TRBS_PER_SEGMENT - 4)*16,
2008 .result_seg = NULL,
2009 },
2010 /* TRB not in this ring, and we have a wrapped TD */
2011 { .input_seg = xhci->event_ring->first_seg,
2012 .start_trb = &xhci->event_ring->first_seg->trbs[TRBS_PER_SEGMENT - 3],
2013 .end_trb = &xhci->event_ring->first_seg->trbs[1],
2014 .input_dma = xhci->cmd_ring->first_seg->dma + 2*16,
2015 .result_seg = NULL,
2016 },
2017 };
2018
2019 unsigned int num_tests;
2020 int i, ret;
2021
2022 num_tests = ARRAY_SIZE(simple_test_vector);
2023 for (i = 0; i < num_tests; i++) {
2024 ret = xhci_test_trb_in_td(xhci,
2025 xhci->event_ring->first_seg,
2026 xhci->event_ring->first_seg->trbs,
2027 &xhci->event_ring->first_seg->trbs[TRBS_PER_SEGMENT - 1],
2028 simple_test_vector[i].input_dma,
2029 simple_test_vector[i].result_seg,
2030 "Simple", i);
2031 if (ret < 0)
2032 return ret;
2033 }
2034
2035 num_tests = ARRAY_SIZE(complex_test_vector);
2036 for (i = 0; i < num_tests; i++) {
2037 ret = xhci_test_trb_in_td(xhci,
2038 complex_test_vector[i].input_seg,
2039 complex_test_vector[i].start_trb,
2040 complex_test_vector[i].end_trb,
2041 complex_test_vector[i].input_dma,
2042 complex_test_vector[i].result_seg,
2043 "Complex", i);
2044 if (ret < 0)
2045 return ret;
2046 }
2047 xhci_dbg(xhci, "TRB math tests passed.\n");
2048 return 0;
2049}
2050
2051static void xhci_set_hc_event_deq(struct xhci_hcd *xhci)
2052{
2053 u64 temp;
2054 dma_addr_t deq;
2055
2056 deq = xhci_trb_virt_to_dma(xhci->event_ring->deq_seg,
2057 xhci->event_ring->dequeue);
2058 if (deq == 0 && !in_interrupt())
2059 xhci_warn(xhci, "WARN something wrong with SW event ring "
2060 "dequeue ptr.\n");
2061 /* Update HC event ring dequeue pointer */
2062 temp = xhci_read_64(xhci, &xhci->ir_set->erst_dequeue);
2063 temp &= ERST_PTR_MASK;
2064 /* Don't clear the EHB bit (which is RW1C) because
2065 * there might be more events to service.
2066 */
2067 temp &= ~ERST_EHB;
2068 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2069 "// Write event ring dequeue pointer, "
2070 "preserving EHB bit");
2071 xhci_write_64(xhci, ((u64) deq & (u64) ~ERST_PTR_MASK) | temp,
2072 &xhci->ir_set->erst_dequeue);
2073}
2074
2075static void xhci_add_in_port(struct xhci_hcd *xhci, unsigned int num_ports,
2076 __le32 __iomem *addr, u8 major_revision, int max_caps)
2077{
2078 u32 temp, port_offset, port_count;
2079 int i;
2080
2081 if (major_revision > 0x03) {
2082 xhci_warn(xhci, "Ignoring unknown port speed, "
2083 "Ext Cap %p, revision = 0x%x\n",
2084 addr, major_revision);
2085 /* Ignoring port protocol we can't understand. FIXME */
2086 return;
2087 }
2088
2089 /* Port offset and count in the third dword, see section 7.2 */
2090 temp = readl(addr + 2);
2091 port_offset = XHCI_EXT_PORT_OFF(temp);
2092 port_count = XHCI_EXT_PORT_COUNT(temp);
2093 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2094 "Ext Cap %p, port offset = %u, "
2095 "count = %u, revision = 0x%x",
2096 addr, port_offset, port_count, major_revision);
2097 /* Port count includes the current port offset */
2098 if (port_offset == 0 || (port_offset + port_count - 1) > num_ports)
2099 /* WTF? "Valid values are ‘1’ to MaxPorts" */
2100 return;
2101
2102 /* cache usb2 port capabilities */
2103 if (major_revision < 0x03 && xhci->num_ext_caps < max_caps)
2104 xhci->ext_caps[xhci->num_ext_caps++] = temp;
2105
2106 /* Check the host's USB2 LPM capability */
2107 if ((xhci->hci_version == 0x96) && (major_revision != 0x03) &&
2108 (temp & XHCI_L1C)) {
2109 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2110 "xHCI 0.96: support USB2 software lpm");
2111 xhci->sw_lpm_support = 1;
2112 }
2113
2114 if ((xhci->hci_version >= 0x100) && (major_revision != 0x03)) {
2115 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2116 "xHCI 1.0: support USB2 software lpm");
2117 xhci->sw_lpm_support = 1;
2118 if (temp & XHCI_HLC) {
2119 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2120 "xHCI 1.0: support USB2 hardware lpm");
2121 xhci->hw_lpm_support = 1;
2122 }
2123 }
2124
2125 port_offset--;
2126 for (i = port_offset; i < (port_offset + port_count); i++) {
2127 /* Duplicate entry. Ignore the port if the revisions differ. */
2128 if (xhci->port_array[i] != 0) {
2129 xhci_warn(xhci, "Duplicate port entry, Ext Cap %p,"
2130 " port %u\n", addr, i);
2131 xhci_warn(xhci, "Port was marked as USB %u, "
2132 "duplicated as USB %u\n",
2133 xhci->port_array[i], major_revision);
2134 /* Only adjust the roothub port counts if we haven't
2135 * found a similar duplicate.
2136 */
2137 if (xhci->port_array[i] != major_revision &&
2138 xhci->port_array[i] != DUPLICATE_ENTRY) {
2139 if (xhci->port_array[i] == 0x03)
2140 xhci->num_usb3_ports--;
2141 else
2142 xhci->num_usb2_ports--;
2143 xhci->port_array[i] = DUPLICATE_ENTRY;
2144 }
2145 /* FIXME: Should we disable the port? */
2146 continue;
2147 }
2148 xhci->port_array[i] = major_revision;
2149 if (major_revision == 0x03)
2150 xhci->num_usb3_ports++;
2151 else
2152 xhci->num_usb2_ports++;
2153 }
2154 /* FIXME: Should we disable ports not in the Extended Capabilities? */
2155}
2156
2157/*
2158 * Scan the Extended Capabilities for the "Supported Protocol Capabilities" that
2159 * specify what speeds each port is supposed to be. We can't count on the port
2160 * speed bits in the PORTSC register being correct until a device is connected,
2161 * but we need to set up the two fake roothubs with the correct number of USB
2162 * 3.0 and USB 2.0 ports at host controller initialization time.
2163 */
2164static int xhci_setup_port_arrays(struct xhci_hcd *xhci, gfp_t flags)
2165{
2166 __le32 __iomem *addr, *tmp_addr;
2167 u32 offset, tmp_offset;
2168 unsigned int num_ports;
2169 int i, j, port_index;
2170 int cap_count = 0;
2171
2172 addr = &xhci->cap_regs->hcc_params;
2173 offset = XHCI_HCC_EXT_CAPS(readl(addr));
2174 if (offset == 0) {
2175 xhci_err(xhci, "No Extended Capability registers, "
2176 "unable to set up roothub.\n");
2177 return -ENODEV;
2178 }
2179
2180 num_ports = HCS_MAX_PORTS(xhci->hcs_params1);
2181 xhci->port_array = kzalloc(sizeof(*xhci->port_array)*num_ports, flags);
2182 if (!xhci->port_array)
2183 return -ENOMEM;
2184
2185 xhci->rh_bw = kzalloc(sizeof(*xhci->rh_bw)*num_ports, flags);
2186 if (!xhci->rh_bw)
2187 return -ENOMEM;
2188 for (i = 0; i < num_ports; i++) {
2189 struct xhci_interval_bw_table *bw_table;
2190
2191 INIT_LIST_HEAD(&xhci->rh_bw[i].tts);
2192 bw_table = &xhci->rh_bw[i].bw_table;
2193 for (j = 0; j < XHCI_MAX_INTERVAL; j++)
2194 INIT_LIST_HEAD(&bw_table->interval_bw[j].endpoints);
2195 }
2196
2197 /*
2198 * For whatever reason, the first capability offset is from the
2199 * capability register base, not from the HCCPARAMS register.
2200 * See section 5.3.6 for offset calculation.
2201 */
2202 addr = &xhci->cap_regs->hc_capbase + offset;
2203
2204 tmp_addr = addr;
2205 tmp_offset = offset;
2206
2207 /* count extended protocol capability entries for later caching */
2208 do {
2209 u32 cap_id;
2210 cap_id = readl(tmp_addr);
2211 if (XHCI_EXT_CAPS_ID(cap_id) == XHCI_EXT_CAPS_PROTOCOL)
2212 cap_count++;
2213 tmp_offset = XHCI_EXT_CAPS_NEXT(cap_id);
2214 tmp_addr += tmp_offset;
2215 } while (tmp_offset);
2216
2217 xhci->ext_caps = kzalloc(sizeof(*xhci->ext_caps) * cap_count, flags);
2218 if (!xhci->ext_caps)
2219 return -ENOMEM;
2220
2221 while (1) {
2222 u32 cap_id;
2223
2224 cap_id = readl(addr);
2225 if (XHCI_EXT_CAPS_ID(cap_id) == XHCI_EXT_CAPS_PROTOCOL)
2226 xhci_add_in_port(xhci, num_ports, addr,
2227 (u8) XHCI_EXT_PORT_MAJOR(cap_id),
2228 cap_count);
2229 offset = XHCI_EXT_CAPS_NEXT(cap_id);
2230 if (!offset || (xhci->num_usb2_ports + xhci->num_usb3_ports)
2231 == num_ports)
2232 break;
2233 /*
2234 * Once you're into the Extended Capabilities, the offset is
2235 * always relative to the register holding the offset.
2236 */
2237 addr += offset;
2238 }
2239
2240 if (xhci->num_usb2_ports == 0 && xhci->num_usb3_ports == 0) {
2241 xhci_warn(xhci, "No ports on the roothubs?\n");
2242 return -ENODEV;
2243 }
2244 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2245 "Found %u USB 2.0 ports and %u USB 3.0 ports.",
2246 xhci->num_usb2_ports, xhci->num_usb3_ports);
2247
2248 /* Place limits on the number of roothub ports so that the hub
2249 * descriptors aren't longer than the USB core will allocate.
2250 */
2251 if (xhci->num_usb3_ports > 15) {
2252 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2253 "Limiting USB 3.0 roothub ports to 15.");
2254 xhci->num_usb3_ports = 15;
2255 }
2256 if (xhci->num_usb2_ports > USB_MAXCHILDREN) {
2257 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2258 "Limiting USB 2.0 roothub ports to %u.",
2259 USB_MAXCHILDREN);
2260 xhci->num_usb2_ports = USB_MAXCHILDREN;
2261 }
2262
2263 /*
2264 * Note we could have all USB 3.0 ports, or all USB 2.0 ports.
2265 * Not sure how the USB core will handle a hub with no ports...
2266 */
2267 if (xhci->num_usb2_ports) {
2268 xhci->usb2_ports = kmalloc(sizeof(*xhci->usb2_ports)*
2269 xhci->num_usb2_ports, flags);
2270 if (!xhci->usb2_ports)
2271 return -ENOMEM;
2272
2273 port_index = 0;
2274 for (i = 0; i < num_ports; i++) {
2275 if (xhci->port_array[i] == 0x03 ||
2276 xhci->port_array[i] == 0 ||
2277 xhci->port_array[i] == DUPLICATE_ENTRY)
2278 continue;
2279
2280 xhci->usb2_ports[port_index] =
2281 &xhci->op_regs->port_status_base +
2282 NUM_PORT_REGS*i;
2283 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2284 "USB 2.0 port at index %u, "
2285 "addr = %p", i,
2286 xhci->usb2_ports[port_index]);
2287 port_index++;
2288 if (port_index == xhci->num_usb2_ports)
2289 break;
2290 }
2291 }
2292 if (xhci->num_usb3_ports) {
2293 xhci->usb3_ports = kmalloc(sizeof(*xhci->usb3_ports)*
2294 xhci->num_usb3_ports, flags);
2295 if (!xhci->usb3_ports)
2296 return -ENOMEM;
2297
2298 port_index = 0;
2299 for (i = 0; i < num_ports; i++)
2300 if (xhci->port_array[i] == 0x03) {
2301 xhci->usb3_ports[port_index] =
2302 &xhci->op_regs->port_status_base +
2303 NUM_PORT_REGS*i;
2304 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2305 "USB 3.0 port at index %u, "
2306 "addr = %p", i,
2307 xhci->usb3_ports[port_index]);
2308 port_index++;
2309 if (port_index == xhci->num_usb3_ports)
2310 break;
2311 }
2312 }
2313 return 0;
2314}
2315
2316int xhci_mem_init(struct xhci_hcd *xhci, gfp_t flags)
2317{
2318 dma_addr_t dma;
2319 struct device *dev = xhci_to_hcd(xhci)->self.controller;
2320 unsigned int val, val2;
2321 u64 val_64;
2322 struct xhci_segment *seg;
2323 u32 page_size, temp;
2324 int i;
2325
2326 INIT_LIST_HEAD(&xhci->cancel_cmd_list);
2327
2328 page_size = readl(&xhci->op_regs->page_size);
2329 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2330 "Supported page size register = 0x%x", page_size);
2331 for (i = 0; i < 16; i++) {
2332 if ((0x1 & page_size) != 0)
2333 break;
2334 page_size = page_size >> 1;
2335 }
2336 if (i < 16)
2337 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2338 "Supported page size of %iK", (1 << (i+12)) / 1024);
2339 else
2340 xhci_warn(xhci, "WARN: no supported page size\n");
2341 /* Use 4K pages, since that's common and the minimum the HC supports */
2342 xhci->page_shift = 12;
2343 xhci->page_size = 1 << xhci->page_shift;
2344 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2345 "HCD page size set to %iK", xhci->page_size / 1024);
2346
2347 /*
2348 * Program the Number of Device Slots Enabled field in the CONFIG
2349 * register with the max value of slots the HC can handle.
2350 */
2351 val = HCS_MAX_SLOTS(readl(&xhci->cap_regs->hcs_params1));
2352 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2353 "// xHC can handle at most %d device slots.", val);
2354 val2 = readl(&xhci->op_regs->config_reg);
2355 val |= (val2 & ~HCS_SLOTS_MASK);
2356 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2357 "// Setting Max device slots reg = 0x%x.", val);
2358 writel(val, &xhci->op_regs->config_reg);
2359
2360 /*
2361 * Section 5.4.8 - doorbell array must be
2362 * "physically contiguous and 64-byte (cache line) aligned".
2363 */
2364 xhci->dcbaa = dma_alloc_coherent(dev, sizeof(*xhci->dcbaa), &dma,
2365 GFP_KERNEL);
2366 if (!xhci->dcbaa)
2367 goto fail;
2368 memset(xhci->dcbaa, 0, sizeof *(xhci->dcbaa));
2369 xhci->dcbaa->dma = dma;
2370 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2371 "// Device context base array address = 0x%llx (DMA), %p (virt)",
2372 (unsigned long long)xhci->dcbaa->dma, xhci->dcbaa);
2373 xhci_write_64(xhci, dma, &xhci->op_regs->dcbaa_ptr);
2374
2375 /*
2376 * Initialize the ring segment pool. The ring must be a contiguous
2377 * structure comprised of TRBs. The TRBs must be 16 byte aligned,
2378 * however, the command ring segment needs 64-byte aligned segments
2379 * and our use of dma addresses in the trb_address_map radix tree needs
2380 * TRB_SEGMENT_SIZE alignment, so we pick the greater alignment need.
2381 */
2382 xhci->segment_pool = dma_pool_create("xHCI ring segments", dev,
2383 TRB_SEGMENT_SIZE, TRB_SEGMENT_SIZE, xhci->page_size);
2384
2385 /* See Table 46 and Note on Figure 55 */
2386 xhci->device_pool = dma_pool_create("xHCI input/output contexts", dev,
2387 2112, 64, xhci->page_size);
2388 if (!xhci->segment_pool || !xhci->device_pool)
2389 goto fail;
2390
2391 /* Linear stream context arrays don't have any boundary restrictions,
2392 * and only need to be 16-byte aligned.
2393 */
2394 xhci->small_streams_pool =
2395 dma_pool_create("xHCI 256 byte stream ctx arrays",
2396 dev, SMALL_STREAM_ARRAY_SIZE, 16, 0);
2397 xhci->medium_streams_pool =
2398 dma_pool_create("xHCI 1KB stream ctx arrays",
2399 dev, MEDIUM_STREAM_ARRAY_SIZE, 16, 0);
2400 /* Any stream context array bigger than MEDIUM_STREAM_ARRAY_SIZE
2401 * will be allocated with dma_alloc_coherent()
2402 */
2403
2404 if (!xhci->small_streams_pool || !xhci->medium_streams_pool)
2405 goto fail;
2406
2407 /* Set up the command ring to have one segments for now. */
2408 xhci->cmd_ring = xhci_ring_alloc(xhci, 1, 1, TYPE_COMMAND, flags);
2409 if (!xhci->cmd_ring)
2410 goto fail;
2411 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2412 "Allocated command ring at %p", xhci->cmd_ring);
2413 xhci_dbg_trace(xhci, trace_xhci_dbg_init, "First segment DMA is 0x%llx",
2414 (unsigned long long)xhci->cmd_ring->first_seg->dma);
2415
2416 /* Set the address in the Command Ring Control register */
2417 val_64 = xhci_read_64(xhci, &xhci->op_regs->cmd_ring);
2418 val_64 = (val_64 & (u64) CMD_RING_RSVD_BITS) |
2419 (xhci->cmd_ring->first_seg->dma & (u64) ~CMD_RING_RSVD_BITS) |
2420 xhci->cmd_ring->cycle_state;
2421 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2422 "// Setting command ring address to 0x%x", val);
2423 xhci_write_64(xhci, val_64, &xhci->op_regs->cmd_ring);
2424 xhci_dbg_cmd_ptrs(xhci);
2425
2426 xhci->lpm_command = xhci_alloc_command(xhci, true, true, flags);
2427 if (!xhci->lpm_command)
2428 goto fail;
2429
2430 /* Reserve one command ring TRB for disabling LPM.
2431 * Since the USB core grabs the shared usb_bus bandwidth mutex before
2432 * disabling LPM, we only need to reserve one TRB for all devices.
2433 */
2434 xhci->cmd_ring_reserved_trbs++;
2435
2436 val = readl(&xhci->cap_regs->db_off);
2437 val &= DBOFF_MASK;
2438 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2439 "// Doorbell array is located at offset 0x%x"
2440 " from cap regs base addr", val);
2441 xhci->dba = (void __iomem *) xhci->cap_regs + val;
2442 xhci_dbg_regs(xhci);
2443 xhci_print_run_regs(xhci);
2444 /* Set ir_set to interrupt register set 0 */
2445 xhci->ir_set = &xhci->run_regs->ir_set[0];
2446
2447 /*
2448 * Event ring setup: Allocate a normal ring, but also setup
2449 * the event ring segment table (ERST). Section 4.9.3.
2450 */
2451 xhci_dbg_trace(xhci, trace_xhci_dbg_init, "// Allocating event ring");
2452 xhci->event_ring = xhci_ring_alloc(xhci, ERST_NUM_SEGS, 1, TYPE_EVENT,
2453 flags);
2454 if (!xhci->event_ring)
2455 goto fail;
2456 if (xhci_check_trb_in_td_math(xhci, flags) < 0)
2457 goto fail;
2458
2459 xhci->erst.entries = dma_alloc_coherent(dev,
2460 sizeof(struct xhci_erst_entry) * ERST_NUM_SEGS, &dma,
2461 GFP_KERNEL);
2462 if (!xhci->erst.entries)
2463 goto fail;
2464 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2465 "// Allocated event ring segment table at 0x%llx",
2466 (unsigned long long)dma);
2467
2468 memset(xhci->erst.entries, 0, sizeof(struct xhci_erst_entry)*ERST_NUM_SEGS);
2469 xhci->erst.num_entries = ERST_NUM_SEGS;
2470 xhci->erst.erst_dma_addr = dma;
2471 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2472 "Set ERST to 0; private num segs = %i, virt addr = %p, dma addr = 0x%llx",
2473 xhci->erst.num_entries,
2474 xhci->erst.entries,
2475 (unsigned long long)xhci->erst.erst_dma_addr);
2476
2477 /* set ring base address and size for each segment table entry */
2478 for (val = 0, seg = xhci->event_ring->first_seg; val < ERST_NUM_SEGS; val++) {
2479 struct xhci_erst_entry *entry = &xhci->erst.entries[val];
2480 entry->seg_addr = cpu_to_le64(seg->dma);
2481 entry->seg_size = cpu_to_le32(TRBS_PER_SEGMENT);
2482 entry->rsvd = 0;
2483 seg = seg->next;
2484 }
2485
2486 /* set ERST count with the number of entries in the segment table */
2487 val = readl(&xhci->ir_set->erst_size);
2488 val &= ERST_SIZE_MASK;
2489 val |= ERST_NUM_SEGS;
2490 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2491 "// Write ERST size = %i to ir_set 0 (some bits preserved)",
2492 val);
2493 writel(val, &xhci->ir_set->erst_size);
2494
2495 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2496 "// Set ERST entries to point to event ring.");
2497 /* set the segment table base address */
2498 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2499 "// Set ERST base address for ir_set 0 = 0x%llx",
2500 (unsigned long long)xhci->erst.erst_dma_addr);
2501 val_64 = xhci_read_64(xhci, &xhci->ir_set->erst_base);
2502 val_64 &= ERST_PTR_MASK;
2503 val_64 |= (xhci->erst.erst_dma_addr & (u64) ~ERST_PTR_MASK);
2504 xhci_write_64(xhci, val_64, &xhci->ir_set->erst_base);
2505
2506 /* Set the event ring dequeue address */
2507 xhci_set_hc_event_deq(xhci);
2508 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2509 "Wrote ERST address to ir_set 0.");
2510 xhci_print_ir_set(xhci, 0);
2511
2512 /*
2513 * XXX: Might need to set the Interrupter Moderation Register to
2514 * something other than the default (~1ms minimum between interrupts).
2515 * See section 5.5.1.2.
2516 */
2517 init_completion(&xhci->addr_dev);
2518 for (i = 0; i < MAX_HC_SLOTS; ++i)
2519 xhci->devs[i] = NULL;
2520 for (i = 0; i < USB_MAXCHILDREN; ++i) {
2521 xhci->bus_state[0].resume_done[i] = 0;
2522 xhci->bus_state[1].resume_done[i] = 0;
2523 /* Only the USB 2.0 completions will ever be used. */
2524 init_completion(&xhci->bus_state[1].rexit_done[i]);
2525 }
2526
2527 if (scratchpad_alloc(xhci, flags))
2528 goto fail;
2529 if (xhci_setup_port_arrays(xhci, flags))
2530 goto fail;
2531
2532 /* Enable USB 3.0 device notifications for function remote wake, which
2533 * is necessary for allowing USB 3.0 devices to do remote wakeup from
2534 * U3 (device suspend).
2535 */
2536 temp = readl(&xhci->op_regs->dev_notification);
2537 temp &= ~DEV_NOTE_MASK;
2538 temp |= DEV_NOTE_FWAKE;
2539 writel(temp, &xhci->op_regs->dev_notification);
2540
2541 return 0;
2542
2543fail:
2544 xhci_warn(xhci, "Couldn't initialize memory\n");
2545 xhci_halt(xhci);
2546 xhci_reset(xhci);
2547 xhci_mem_cleanup(xhci);
2548 return -ENOMEM;
2549}
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 max_packet,
31 unsigned int num,
32 gfp_t flags)
33{
34 struct xhci_segment *seg;
35 dma_addr_t dma;
36 struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
37
38 seg = kzalloc_node(sizeof(*seg), flags, dev_to_node(dev));
39 if (!seg)
40 return NULL;
41
42 seg->trbs = dma_pool_zalloc(xhci->segment_pool, flags, &dma);
43 if (!seg->trbs) {
44 kfree(seg);
45 return NULL;
46 }
47
48 if (max_packet) {
49 seg->bounce_buf = kzalloc_node(max_packet, flags,
50 dev_to_node(dev));
51 if (!seg->bounce_buf) {
52 dma_pool_free(xhci->segment_pool, seg->trbs, dma);
53 kfree(seg);
54 return NULL;
55 }
56 }
57 seg->num = num;
58 seg->dma = dma;
59 seg->next = NULL;
60
61 return seg;
62}
63
64static void xhci_segment_free(struct xhci_hcd *xhci, struct xhci_segment *seg)
65{
66 if (seg->trbs) {
67 dma_pool_free(xhci->segment_pool, seg->trbs, seg->dma);
68 seg->trbs = NULL;
69 }
70 kfree(seg->bounce_buf);
71 kfree(seg);
72}
73
74static void xhci_ring_segments_free(struct xhci_hcd *xhci, struct xhci_ring *ring)
75{
76 struct xhci_segment *seg, *next;
77
78 ring->last_seg->next = NULL;
79 seg = ring->first_seg;
80
81 while (seg) {
82 next = seg->next;
83 xhci_segment_free(xhci, seg);
84 seg = next;
85 }
86}
87
88/*
89 * Only for transfer and command rings where driver is the producer, not for
90 * event rings.
91 *
92 * Change the last TRB in the segment to be a Link TRB which points to the
93 * DMA address of the next segment. The caller needs to set any Link TRB
94 * related flags, such as End TRB, Toggle Cycle, and no snoop.
95 */
96static void xhci_set_link_trb(struct xhci_segment *seg, bool chain_links)
97{
98 union xhci_trb *trb;
99 u32 val;
100
101 if (!seg || !seg->next)
102 return;
103
104 trb = &seg->trbs[TRBS_PER_SEGMENT - 1];
105
106 /* Set the last TRB in the segment to have a TRB type ID of Link TRB */
107 val = le32_to_cpu(trb->link.control);
108 val &= ~TRB_TYPE_BITMASK;
109 val |= TRB_TYPE(TRB_LINK);
110 if (chain_links)
111 val |= TRB_CHAIN;
112 trb->link.control = cpu_to_le32(val);
113 trb->link.segment_ptr = cpu_to_le64(seg->next->dma);
114}
115
116static void xhci_initialize_ring_segments(struct xhci_hcd *xhci, struct xhci_ring *ring)
117{
118 struct xhci_segment *seg;
119 bool chain_links;
120
121 if (ring->type == TYPE_EVENT)
122 return;
123
124 chain_links = xhci_link_chain_quirk(xhci, ring->type);
125 xhci_for_each_ring_seg(ring->first_seg, seg)
126 xhci_set_link_trb(seg, chain_links);
127
128 /* See section 4.9.2.1 and 6.4.4.1 */
129 ring->last_seg->trbs[TRBS_PER_SEGMENT - 1].link.control |= cpu_to_le32(LINK_TOGGLE);
130}
131
132/*
133 * Link the src ring segments to the dst ring.
134 * Set Toggle Cycle for the new ring if needed.
135 */
136static void xhci_link_rings(struct xhci_hcd *xhci, struct xhci_ring *src, struct xhci_ring *dst)
137{
138 struct xhci_segment *seg;
139 bool chain_links;
140
141 if (!src || !dst)
142 return;
143
144 /* If the cycle state is 0, set the cycle bit to 1 for all the TRBs */
145 if (dst->cycle_state == 0) {
146 xhci_for_each_ring_seg(src->first_seg, seg) {
147 for (int i = 0; i < TRBS_PER_SEGMENT; i++)
148 seg->trbs[i].link.control |= cpu_to_le32(TRB_CYCLE);
149 }
150 }
151
152 src->last_seg->next = dst->enq_seg->next;
153 dst->enq_seg->next = src->first_seg;
154 if (dst->type != TYPE_EVENT) {
155 chain_links = xhci_link_chain_quirk(xhci, dst->type);
156 xhci_set_link_trb(dst->enq_seg, chain_links);
157 xhci_set_link_trb(src->last_seg, chain_links);
158 }
159 dst->num_segs += src->num_segs;
160
161 if (dst->enq_seg == dst->last_seg) {
162 if (dst->type != TYPE_EVENT)
163 dst->last_seg->trbs[TRBS_PER_SEGMENT-1].link.control
164 &= ~cpu_to_le32(LINK_TOGGLE);
165
166 dst->last_seg = src->last_seg;
167 } else if (dst->type != TYPE_EVENT) {
168 src->last_seg->trbs[TRBS_PER_SEGMENT-1].link.control &= ~cpu_to_le32(LINK_TOGGLE);
169 }
170
171 for (seg = dst->enq_seg; seg != dst->last_seg; seg = seg->next)
172 seg->next->num = seg->num + 1;
173}
174
175/*
176 * We need a radix tree for mapping physical addresses of TRBs to which stream
177 * ID they belong to. We need to do this because the host controller won't tell
178 * us which stream ring the TRB came from. We could store the stream ID in an
179 * event data TRB, but that doesn't help us for the cancellation case, since the
180 * endpoint may stop before it reaches that event data TRB.
181 *
182 * The radix tree maps the upper portion of the TRB DMA address to a ring
183 * segment that has the same upper portion of DMA addresses. For example, say I
184 * have segments of size 1KB, that are always 1KB aligned. A segment may
185 * start at 0x10c91000 and end at 0x10c913f0. If I use the upper 10 bits, the
186 * key to the stream ID is 0x43244. I can use the DMA address of the TRB to
187 * pass the radix tree a key to get the right stream ID:
188 *
189 * 0x10c90fff >> 10 = 0x43243
190 * 0x10c912c0 >> 10 = 0x43244
191 * 0x10c91400 >> 10 = 0x43245
192 *
193 * Obviously, only those TRBs with DMA addresses that are within the segment
194 * will make the radix tree return the stream ID for that ring.
195 *
196 * Caveats for the radix tree:
197 *
198 * The radix tree uses an unsigned long as a key pair. On 32-bit systems, an
199 * unsigned long will be 32-bits; on a 64-bit system an unsigned long will be
200 * 64-bits. Since we only request 32-bit DMA addresses, we can use that as the
201 * key on 32-bit or 64-bit systems (it would also be fine if we asked for 64-bit
202 * PCI DMA addresses on a 64-bit system). There might be a problem on 32-bit
203 * extended systems (where the DMA address can be bigger than 32-bits),
204 * if we allow the PCI dma mask to be bigger than 32-bits. So don't do that.
205 */
206static int xhci_insert_segment_mapping(struct radix_tree_root *trb_address_map,
207 struct xhci_ring *ring,
208 struct xhci_segment *seg,
209 gfp_t mem_flags)
210{
211 unsigned long key;
212 int ret;
213
214 key = (unsigned long)(seg->dma >> TRB_SEGMENT_SHIFT);
215 /* Skip any segments that were already added. */
216 if (radix_tree_lookup(trb_address_map, key))
217 return 0;
218
219 ret = radix_tree_maybe_preload(mem_flags);
220 if (ret)
221 return ret;
222 ret = radix_tree_insert(trb_address_map,
223 key, ring);
224 radix_tree_preload_end();
225 return ret;
226}
227
228static void xhci_remove_segment_mapping(struct radix_tree_root *trb_address_map,
229 struct xhci_segment *seg)
230{
231 unsigned long key;
232
233 key = (unsigned long)(seg->dma >> TRB_SEGMENT_SHIFT);
234 if (radix_tree_lookup(trb_address_map, key))
235 radix_tree_delete(trb_address_map, key);
236}
237
238static int xhci_update_stream_segment_mapping(
239 struct radix_tree_root *trb_address_map,
240 struct xhci_ring *ring,
241 struct xhci_segment *first_seg,
242 gfp_t mem_flags)
243{
244 struct xhci_segment *seg;
245 struct xhci_segment *failed_seg;
246 int ret;
247
248 if (WARN_ON_ONCE(trb_address_map == NULL))
249 return 0;
250
251 xhci_for_each_ring_seg(first_seg, seg) {
252 ret = xhci_insert_segment_mapping(trb_address_map,
253 ring, seg, mem_flags);
254 if (ret)
255 goto remove_streams;
256 }
257
258 return 0;
259
260remove_streams:
261 failed_seg = seg;
262 xhci_for_each_ring_seg(first_seg, seg) {
263 xhci_remove_segment_mapping(trb_address_map, seg);
264 if (seg == failed_seg)
265 return ret;
266 }
267
268 return ret;
269}
270
271static void xhci_remove_stream_mapping(struct xhci_ring *ring)
272{
273 struct xhci_segment *seg;
274
275 if (WARN_ON_ONCE(ring->trb_address_map == NULL))
276 return;
277
278 xhci_for_each_ring_seg(ring->first_seg, seg)
279 xhci_remove_segment_mapping(ring->trb_address_map, seg);
280}
281
282static int xhci_update_stream_mapping(struct xhci_ring *ring, gfp_t mem_flags)
283{
284 return xhci_update_stream_segment_mapping(ring->trb_address_map, ring,
285 ring->first_seg, mem_flags);
286}
287
288/* XXX: Do we need the hcd structure in all these functions? */
289void xhci_ring_free(struct xhci_hcd *xhci, struct xhci_ring *ring)
290{
291 if (!ring)
292 return;
293
294 trace_xhci_ring_free(ring);
295
296 if (ring->first_seg) {
297 if (ring->type == TYPE_STREAM)
298 xhci_remove_stream_mapping(ring);
299 xhci_ring_segments_free(xhci, ring);
300 }
301
302 kfree(ring);
303}
304
305void xhci_initialize_ring_info(struct xhci_ring *ring)
306{
307 /* The ring is empty, so the enqueue pointer == dequeue pointer */
308 ring->enqueue = ring->first_seg->trbs;
309 ring->enq_seg = ring->first_seg;
310 ring->dequeue = ring->enqueue;
311 ring->deq_seg = ring->first_seg;
312 /* The ring is initialized to 0. The producer must write 1 to the cycle
313 * bit to handover ownership of the TRB, so PCS = 1. The consumer must
314 * compare CCS to the cycle bit to check ownership, so CCS = 1.
315 *
316 * New rings are initialized with cycle state equal to 1; if we are
317 * handling ring expansion, set the cycle state equal to the old ring.
318 */
319 ring->cycle_state = 1;
320
321 /*
322 * Each segment has a link TRB, and leave an extra TRB for SW
323 * accounting purpose
324 */
325 ring->num_trbs_free = ring->num_segs * (TRBS_PER_SEGMENT - 1) - 1;
326}
327EXPORT_SYMBOL_GPL(xhci_initialize_ring_info);
328
329/* Allocate segments and link them for a ring */
330static int xhci_alloc_segments_for_ring(struct xhci_hcd *xhci, struct xhci_ring *ring, gfp_t flags)
331{
332 struct xhci_segment *prev;
333 unsigned int num = 0;
334
335 prev = xhci_segment_alloc(xhci, ring->bounce_buf_len, num, flags);
336 if (!prev)
337 return -ENOMEM;
338 num++;
339
340 ring->first_seg = prev;
341 while (num < ring->num_segs) {
342 struct xhci_segment *next;
343
344 next = xhci_segment_alloc(xhci, ring->bounce_buf_len, num, flags);
345 if (!next)
346 goto free_segments;
347
348 prev->next = next;
349 prev = next;
350 num++;
351 }
352 ring->last_seg = prev;
353
354 ring->last_seg->next = ring->first_seg;
355 return 0;
356
357free_segments:
358 ring->last_seg = prev;
359 xhci_ring_segments_free(xhci, ring);
360 return -ENOMEM;
361}
362
363/*
364 * Create a new ring with zero or more segments.
365 *
366 * Link each segment together into a ring.
367 * Set the end flag and the cycle toggle bit on the last segment.
368 * See section 4.9.1 and figures 15 and 16.
369 */
370struct xhci_ring *xhci_ring_alloc(struct xhci_hcd *xhci, unsigned int num_segs,
371 enum xhci_ring_type type, unsigned int max_packet, gfp_t flags)
372{
373 struct xhci_ring *ring;
374 int ret;
375 struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
376
377 ring = kzalloc_node(sizeof(*ring), flags, dev_to_node(dev));
378 if (!ring)
379 return NULL;
380
381 ring->num_segs = num_segs;
382 ring->bounce_buf_len = max_packet;
383 INIT_LIST_HEAD(&ring->td_list);
384 ring->type = type;
385 if (num_segs == 0)
386 return ring;
387
388 ret = xhci_alloc_segments_for_ring(xhci, ring, flags);
389 if (ret)
390 goto fail;
391
392 xhci_initialize_ring_segments(xhci, ring);
393 xhci_initialize_ring_info(ring);
394 trace_xhci_ring_alloc(ring);
395 return ring;
396
397fail:
398 kfree(ring);
399 return NULL;
400}
401
402void xhci_free_endpoint_ring(struct xhci_hcd *xhci,
403 struct xhci_virt_device *virt_dev,
404 unsigned int ep_index)
405{
406 xhci_ring_free(xhci, virt_dev->eps[ep_index].ring);
407 virt_dev->eps[ep_index].ring = NULL;
408}
409
410/*
411 * Expand an existing ring.
412 * Allocate a new ring which has same segment numbers and link the two rings.
413 */
414int xhci_ring_expansion(struct xhci_hcd *xhci, struct xhci_ring *ring,
415 unsigned int num_new_segs, gfp_t flags)
416{
417 struct xhci_ring new_ring;
418 int ret;
419
420 if (num_new_segs == 0)
421 return 0;
422
423 new_ring.num_segs = num_new_segs;
424 new_ring.bounce_buf_len = ring->bounce_buf_len;
425 new_ring.type = ring->type;
426 ret = xhci_alloc_segments_for_ring(xhci, &new_ring, flags);
427 if (ret)
428 return -ENOMEM;
429
430 xhci_initialize_ring_segments(xhci, &new_ring);
431
432 if (ring->type == TYPE_STREAM) {
433 ret = xhci_update_stream_segment_mapping(ring->trb_address_map, ring,
434 new_ring.first_seg, flags);
435 if (ret)
436 goto free_segments;
437 }
438
439 xhci_link_rings(xhci, &new_ring, ring);
440 trace_xhci_ring_expansion(ring);
441 xhci_dbg_trace(xhci, trace_xhci_dbg_ring_expansion,
442 "ring expansion succeed, now has %d segments",
443 ring->num_segs);
444
445 return 0;
446
447free_segments:
448 xhci_ring_segments_free(xhci, &new_ring);
449 return ret;
450}
451
452struct xhci_container_ctx *xhci_alloc_container_ctx(struct xhci_hcd *xhci,
453 int type, gfp_t flags)
454{
455 struct xhci_container_ctx *ctx;
456 struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
457
458 if ((type != XHCI_CTX_TYPE_DEVICE) && (type != XHCI_CTX_TYPE_INPUT))
459 return NULL;
460
461 ctx = kzalloc_node(sizeof(*ctx), flags, dev_to_node(dev));
462 if (!ctx)
463 return NULL;
464
465 ctx->type = type;
466 ctx->size = HCC_64BYTE_CONTEXT(xhci->hcc_params) ? 2048 : 1024;
467 if (type == XHCI_CTX_TYPE_INPUT)
468 ctx->size += CTX_SIZE(xhci->hcc_params);
469
470 ctx->bytes = dma_pool_zalloc(xhci->device_pool, flags, &ctx->dma);
471 if (!ctx->bytes) {
472 kfree(ctx);
473 return NULL;
474 }
475 return ctx;
476}
477
478void xhci_free_container_ctx(struct xhci_hcd *xhci,
479 struct xhci_container_ctx *ctx)
480{
481 if (!ctx)
482 return;
483 dma_pool_free(xhci->device_pool, ctx->bytes, ctx->dma);
484 kfree(ctx);
485}
486
487struct xhci_input_control_ctx *xhci_get_input_control_ctx(
488 struct xhci_container_ctx *ctx)
489{
490 if (ctx->type != XHCI_CTX_TYPE_INPUT)
491 return NULL;
492
493 return (struct xhci_input_control_ctx *)ctx->bytes;
494}
495
496struct xhci_slot_ctx *xhci_get_slot_ctx(struct xhci_hcd *xhci,
497 struct xhci_container_ctx *ctx)
498{
499 if (ctx->type == XHCI_CTX_TYPE_DEVICE)
500 return (struct xhci_slot_ctx *)ctx->bytes;
501
502 return (struct xhci_slot_ctx *)
503 (ctx->bytes + CTX_SIZE(xhci->hcc_params));
504}
505
506struct xhci_ep_ctx *xhci_get_ep_ctx(struct xhci_hcd *xhci,
507 struct xhci_container_ctx *ctx,
508 unsigned int ep_index)
509{
510 /* increment ep index by offset of start of ep ctx array */
511 ep_index++;
512 if (ctx->type == XHCI_CTX_TYPE_INPUT)
513 ep_index++;
514
515 return (struct xhci_ep_ctx *)
516 (ctx->bytes + (ep_index * CTX_SIZE(xhci->hcc_params)));
517}
518EXPORT_SYMBOL_GPL(xhci_get_ep_ctx);
519
520/***************** Streams structures manipulation *************************/
521
522static void xhci_free_stream_ctx(struct xhci_hcd *xhci,
523 unsigned int num_stream_ctxs,
524 struct xhci_stream_ctx *stream_ctx, dma_addr_t dma)
525{
526 struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
527 size_t size = array_size(sizeof(struct xhci_stream_ctx), num_stream_ctxs);
528
529 if (size > MEDIUM_STREAM_ARRAY_SIZE)
530 dma_free_coherent(dev, size, stream_ctx, dma);
531 else if (size > SMALL_STREAM_ARRAY_SIZE)
532 dma_pool_free(xhci->medium_streams_pool, stream_ctx, dma);
533 else
534 dma_pool_free(xhci->small_streams_pool, stream_ctx, dma);
535}
536
537/*
538 * The stream context array for each endpoint with bulk streams enabled can
539 * vary in size, based on:
540 * - how many streams the endpoint supports,
541 * - the maximum primary stream array size the host controller supports,
542 * - and how many streams the device driver asks for.
543 *
544 * The stream context array must be a power of 2, and can be as small as
545 * 64 bytes or as large as 1MB.
546 */
547static struct xhci_stream_ctx *xhci_alloc_stream_ctx(struct xhci_hcd *xhci,
548 unsigned int num_stream_ctxs, dma_addr_t *dma,
549 gfp_t mem_flags)
550{
551 struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
552 size_t size = array_size(sizeof(struct xhci_stream_ctx), num_stream_ctxs);
553
554 if (size > MEDIUM_STREAM_ARRAY_SIZE)
555 return dma_alloc_coherent(dev, size, dma, mem_flags);
556 if (size > SMALL_STREAM_ARRAY_SIZE)
557 return dma_pool_zalloc(xhci->medium_streams_pool, mem_flags, dma);
558 else
559 return dma_pool_zalloc(xhci->small_streams_pool, mem_flags, dma);
560}
561
562struct xhci_ring *xhci_dma_to_transfer_ring(
563 struct xhci_virt_ep *ep,
564 u64 address)
565{
566 if (ep->ep_state & EP_HAS_STREAMS)
567 return radix_tree_lookup(&ep->stream_info->trb_address_map,
568 address >> TRB_SEGMENT_SHIFT);
569 return ep->ring;
570}
571
572/*
573 * Change an endpoint's internal structure so it supports stream IDs. The
574 * number of requested streams includes stream 0, which cannot be used by device
575 * drivers.
576 *
577 * The number of stream contexts in the stream context array may be bigger than
578 * the number of streams the driver wants to use. This is because the number of
579 * stream context array entries must be a power of two.
580 */
581struct xhci_stream_info *xhci_alloc_stream_info(struct xhci_hcd *xhci,
582 unsigned int num_stream_ctxs,
583 unsigned int num_streams,
584 unsigned int max_packet, gfp_t mem_flags)
585{
586 struct xhci_stream_info *stream_info;
587 u32 cur_stream;
588 struct xhci_ring *cur_ring;
589 u64 addr;
590 int ret;
591 struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
592
593 xhci_dbg(xhci, "Allocating %u streams and %u stream context array entries.\n",
594 num_streams, num_stream_ctxs);
595 if (xhci->cmd_ring_reserved_trbs == MAX_RSVD_CMD_TRBS) {
596 xhci_dbg(xhci, "Command ring has no reserved TRBs available\n");
597 return NULL;
598 }
599 xhci->cmd_ring_reserved_trbs++;
600
601 stream_info = kzalloc_node(sizeof(*stream_info), mem_flags,
602 dev_to_node(dev));
603 if (!stream_info)
604 goto cleanup_trbs;
605
606 stream_info->num_streams = num_streams;
607 stream_info->num_stream_ctxs = num_stream_ctxs;
608
609 /* Initialize the array of virtual pointers to stream rings. */
610 stream_info->stream_rings = kcalloc_node(
611 num_streams, sizeof(struct xhci_ring *), mem_flags,
612 dev_to_node(dev));
613 if (!stream_info->stream_rings)
614 goto cleanup_info;
615
616 /* Initialize the array of DMA addresses for stream rings for the HW. */
617 stream_info->stream_ctx_array = xhci_alloc_stream_ctx(xhci,
618 num_stream_ctxs, &stream_info->ctx_array_dma,
619 mem_flags);
620 if (!stream_info->stream_ctx_array)
621 goto cleanup_ring_array;
622
623 /* Allocate everything needed to free the stream rings later */
624 stream_info->free_streams_command =
625 xhci_alloc_command_with_ctx(xhci, true, mem_flags);
626 if (!stream_info->free_streams_command)
627 goto cleanup_ctx;
628
629 INIT_RADIX_TREE(&stream_info->trb_address_map, GFP_ATOMIC);
630
631 /* Allocate rings for all the streams that the driver will use,
632 * and add their segment DMA addresses to the radix tree.
633 * Stream 0 is reserved.
634 */
635
636 for (cur_stream = 1; cur_stream < num_streams; cur_stream++) {
637 stream_info->stream_rings[cur_stream] =
638 xhci_ring_alloc(xhci, 2, TYPE_STREAM, max_packet, mem_flags);
639 cur_ring = stream_info->stream_rings[cur_stream];
640 if (!cur_ring)
641 goto cleanup_rings;
642 cur_ring->stream_id = cur_stream;
643 cur_ring->trb_address_map = &stream_info->trb_address_map;
644 /* Set deq ptr, cycle bit, and stream context type */
645 addr = cur_ring->first_seg->dma |
646 SCT_FOR_CTX(SCT_PRI_TR) |
647 cur_ring->cycle_state;
648 stream_info->stream_ctx_array[cur_stream].stream_ring =
649 cpu_to_le64(addr);
650 xhci_dbg(xhci, "Setting stream %d ring ptr to 0x%08llx\n", cur_stream, addr);
651
652 ret = xhci_update_stream_mapping(cur_ring, mem_flags);
653
654 trace_xhci_alloc_stream_info_ctx(stream_info, cur_stream);
655 if (ret) {
656 xhci_ring_free(xhci, cur_ring);
657 stream_info->stream_rings[cur_stream] = NULL;
658 goto cleanup_rings;
659 }
660 }
661 /* Leave the other unused stream ring pointers in the stream context
662 * array initialized to zero. This will cause the xHC to give us an
663 * error if the device asks for a stream ID we don't have setup (if it
664 * was any other way, the host controller would assume the ring is
665 * "empty" and wait forever for data to be queued to that stream ID).
666 */
667
668 return stream_info;
669
670cleanup_rings:
671 for (cur_stream = 1; cur_stream < num_streams; cur_stream++) {
672 cur_ring = stream_info->stream_rings[cur_stream];
673 if (cur_ring) {
674 xhci_ring_free(xhci, cur_ring);
675 stream_info->stream_rings[cur_stream] = NULL;
676 }
677 }
678 xhci_free_command(xhci, stream_info->free_streams_command);
679cleanup_ctx:
680 xhci_free_stream_ctx(xhci,
681 stream_info->num_stream_ctxs,
682 stream_info->stream_ctx_array,
683 stream_info->ctx_array_dma);
684cleanup_ring_array:
685 kfree(stream_info->stream_rings);
686cleanup_info:
687 kfree(stream_info);
688cleanup_trbs:
689 xhci->cmd_ring_reserved_trbs--;
690 return NULL;
691}
692/*
693 * Sets the MaxPStreams field and the Linear Stream Array field.
694 * Sets the dequeue pointer to the stream context array.
695 */
696void xhci_setup_streams_ep_input_ctx(struct xhci_hcd *xhci,
697 struct xhci_ep_ctx *ep_ctx,
698 struct xhci_stream_info *stream_info)
699{
700 u32 max_primary_streams;
701 /* MaxPStreams is the number of stream context array entries, not the
702 * number we're actually using. Must be in 2^(MaxPstreams + 1) format.
703 * fls(0) = 0, fls(0x1) = 1, fls(0x10) = 2, fls(0x100) = 3, etc.
704 */
705 max_primary_streams = fls(stream_info->num_stream_ctxs) - 2;
706 xhci_dbg_trace(xhci, trace_xhci_dbg_context_change,
707 "Setting number of stream ctx array entries to %u",
708 1 << (max_primary_streams + 1));
709 ep_ctx->ep_info &= cpu_to_le32(~EP_MAXPSTREAMS_MASK);
710 ep_ctx->ep_info |= cpu_to_le32(EP_MAXPSTREAMS(max_primary_streams)
711 | EP_HAS_LSA);
712 ep_ctx->deq = cpu_to_le64(stream_info->ctx_array_dma);
713}
714
715/*
716 * Sets the MaxPStreams field and the Linear Stream Array field to 0.
717 * Reinstalls the "normal" endpoint ring (at its previous dequeue mark,
718 * not at the beginning of the ring).
719 */
720void xhci_setup_no_streams_ep_input_ctx(struct xhci_ep_ctx *ep_ctx,
721 struct xhci_virt_ep *ep)
722{
723 dma_addr_t addr;
724 ep_ctx->ep_info &= cpu_to_le32(~(EP_MAXPSTREAMS_MASK | EP_HAS_LSA));
725 addr = xhci_trb_virt_to_dma(ep->ring->deq_seg, ep->ring->dequeue);
726 ep_ctx->deq = cpu_to_le64(addr | ep->ring->cycle_state);
727}
728
729/* Frees all stream contexts associated with the endpoint,
730 *
731 * Caller should fix the endpoint context streams fields.
732 */
733void xhci_free_stream_info(struct xhci_hcd *xhci,
734 struct xhci_stream_info *stream_info)
735{
736 int cur_stream;
737 struct xhci_ring *cur_ring;
738
739 if (!stream_info)
740 return;
741
742 for (cur_stream = 1; cur_stream < stream_info->num_streams;
743 cur_stream++) {
744 cur_ring = stream_info->stream_rings[cur_stream];
745 if (cur_ring) {
746 xhci_ring_free(xhci, cur_ring);
747 stream_info->stream_rings[cur_stream] = NULL;
748 }
749 }
750 xhci_free_command(xhci, stream_info->free_streams_command);
751 xhci->cmd_ring_reserved_trbs--;
752 if (stream_info->stream_ctx_array)
753 xhci_free_stream_ctx(xhci,
754 stream_info->num_stream_ctxs,
755 stream_info->stream_ctx_array,
756 stream_info->ctx_array_dma);
757
758 kfree(stream_info->stream_rings);
759 kfree(stream_info);
760}
761
762
763/***************** Device context manipulation *************************/
764
765static void xhci_free_tt_info(struct xhci_hcd *xhci,
766 struct xhci_virt_device *virt_dev,
767 int slot_id)
768{
769 struct list_head *tt_list_head;
770 struct xhci_tt_bw_info *tt_info, *next;
771 bool slot_found = false;
772
773 /* If the device never made it past the Set Address stage,
774 * it may not have the root hub port pointer set correctly.
775 */
776 if (!virt_dev->rhub_port) {
777 xhci_dbg(xhci, "Bad rhub port.\n");
778 return;
779 }
780
781 tt_list_head = &(xhci->rh_bw[virt_dev->rhub_port->hw_portnum].tts);
782 list_for_each_entry_safe(tt_info, next, tt_list_head, tt_list) {
783 /* Multi-TT hubs will have more than one entry */
784 if (tt_info->slot_id == slot_id) {
785 slot_found = true;
786 list_del(&tt_info->tt_list);
787 kfree(tt_info);
788 } else if (slot_found) {
789 break;
790 }
791 }
792}
793
794int xhci_alloc_tt_info(struct xhci_hcd *xhci,
795 struct xhci_virt_device *virt_dev,
796 struct usb_device *hdev,
797 struct usb_tt *tt, gfp_t mem_flags)
798{
799 struct xhci_tt_bw_info *tt_info;
800 unsigned int num_ports;
801 int i, j;
802 struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
803
804 if (!tt->multi)
805 num_ports = 1;
806 else
807 num_ports = hdev->maxchild;
808
809 for (i = 0; i < num_ports; i++, tt_info++) {
810 struct xhci_interval_bw_table *bw_table;
811
812 tt_info = kzalloc_node(sizeof(*tt_info), mem_flags,
813 dev_to_node(dev));
814 if (!tt_info)
815 goto free_tts;
816 INIT_LIST_HEAD(&tt_info->tt_list);
817 list_add(&tt_info->tt_list,
818 &xhci->rh_bw[virt_dev->rhub_port->hw_portnum].tts);
819 tt_info->slot_id = virt_dev->udev->slot_id;
820 if (tt->multi)
821 tt_info->ttport = i+1;
822 bw_table = &tt_info->bw_table;
823 for (j = 0; j < XHCI_MAX_INTERVAL; j++)
824 INIT_LIST_HEAD(&bw_table->interval_bw[j].endpoints);
825 }
826 return 0;
827
828free_tts:
829 xhci_free_tt_info(xhci, virt_dev, virt_dev->udev->slot_id);
830 return -ENOMEM;
831}
832
833
834/* All the xhci_tds in the ring's TD list should be freed at this point.
835 * Should be called with xhci->lock held if there is any chance the TT lists
836 * will be manipulated by the configure endpoint, allocate device, or update
837 * hub functions while this function is removing the TT entries from the list.
838 */
839void xhci_free_virt_device(struct xhci_hcd *xhci, int slot_id)
840{
841 struct xhci_virt_device *dev;
842 int i;
843 int old_active_eps = 0;
844
845 /* Slot ID 0 is reserved */
846 if (slot_id == 0 || !xhci->devs[slot_id])
847 return;
848
849 dev = xhci->devs[slot_id];
850
851 xhci->dcbaa->dev_context_ptrs[slot_id] = 0;
852 if (!dev)
853 return;
854
855 trace_xhci_free_virt_device(dev);
856
857 if (dev->tt_info)
858 old_active_eps = dev->tt_info->active_eps;
859
860 for (i = 0; i < 31; i++) {
861 if (dev->eps[i].ring)
862 xhci_ring_free(xhci, dev->eps[i].ring);
863 if (dev->eps[i].stream_info)
864 xhci_free_stream_info(xhci,
865 dev->eps[i].stream_info);
866 /*
867 * Endpoints are normally deleted from the bandwidth list when
868 * endpoints are dropped, before device is freed.
869 * If host is dying or being removed then endpoints aren't
870 * dropped cleanly, so delete the endpoint from list here.
871 * Only applicable for hosts with software bandwidth checking.
872 */
873
874 if (!list_empty(&dev->eps[i].bw_endpoint_list)) {
875 list_del_init(&dev->eps[i].bw_endpoint_list);
876 xhci_dbg(xhci, "Slot %u endpoint %u not removed from BW list!\n",
877 slot_id, i);
878 }
879 }
880 /* If this is a hub, free the TT(s) from the TT list */
881 xhci_free_tt_info(xhci, dev, slot_id);
882 /* If necessary, update the number of active TTs on this root port */
883 xhci_update_tt_active_eps(xhci, dev, old_active_eps);
884
885 if (dev->in_ctx)
886 xhci_free_container_ctx(xhci, dev->in_ctx);
887 if (dev->out_ctx)
888 xhci_free_container_ctx(xhci, dev->out_ctx);
889
890 if (dev->udev && dev->udev->slot_id)
891 dev->udev->slot_id = 0;
892 if (dev->rhub_port && dev->rhub_port->slot_id == slot_id)
893 dev->rhub_port->slot_id = 0;
894 kfree(xhci->devs[slot_id]);
895 xhci->devs[slot_id] = NULL;
896}
897
898/*
899 * Free a virt_device structure.
900 * If the virt_device added a tt_info (a hub) and has children pointing to
901 * that tt_info, then free the child first. Recursive.
902 * We can't rely on udev at this point to find child-parent relationships.
903 */
904static void xhci_free_virt_devices_depth_first(struct xhci_hcd *xhci, int slot_id)
905{
906 struct xhci_virt_device *vdev;
907 struct list_head *tt_list_head;
908 struct xhci_tt_bw_info *tt_info, *next;
909 int i;
910
911 vdev = xhci->devs[slot_id];
912 if (!vdev)
913 return;
914
915 if (!vdev->rhub_port) {
916 xhci_dbg(xhci, "Bad rhub port.\n");
917 goto out;
918 }
919
920 tt_list_head = &(xhci->rh_bw[vdev->rhub_port->hw_portnum].tts);
921 list_for_each_entry_safe(tt_info, next, tt_list_head, tt_list) {
922 /* is this a hub device that added a tt_info to the tts list */
923 if (tt_info->slot_id == slot_id) {
924 /* are any devices using this tt_info? */
925 for (i = 1; i < HCS_MAX_SLOTS(xhci->hcs_params1); i++) {
926 vdev = xhci->devs[i];
927 if (vdev && (vdev->tt_info == tt_info))
928 xhci_free_virt_devices_depth_first(
929 xhci, i);
930 }
931 }
932 }
933out:
934 /* we are now at a leaf device */
935 xhci_debugfs_remove_slot(xhci, slot_id);
936 xhci_free_virt_device(xhci, slot_id);
937}
938
939int xhci_alloc_virt_device(struct xhci_hcd *xhci, int slot_id,
940 struct usb_device *udev, gfp_t flags)
941{
942 struct xhci_virt_device *dev;
943 int i;
944
945 /* Slot ID 0 is reserved */
946 if (slot_id == 0 || xhci->devs[slot_id]) {
947 xhci_warn(xhci, "Bad Slot ID %d\n", slot_id);
948 return 0;
949 }
950
951 dev = kzalloc(sizeof(*dev), flags);
952 if (!dev)
953 return 0;
954
955 dev->slot_id = slot_id;
956
957 /* Allocate the (output) device context that will be used in the HC. */
958 dev->out_ctx = xhci_alloc_container_ctx(xhci, XHCI_CTX_TYPE_DEVICE, flags);
959 if (!dev->out_ctx)
960 goto fail;
961
962 xhci_dbg(xhci, "Slot %d output ctx = 0x%pad (dma)\n", slot_id, &dev->out_ctx->dma);
963
964 /* Allocate the (input) device context for address device command */
965 dev->in_ctx = xhci_alloc_container_ctx(xhci, XHCI_CTX_TYPE_INPUT, flags);
966 if (!dev->in_ctx)
967 goto fail;
968
969 xhci_dbg(xhci, "Slot %d input ctx = 0x%pad (dma)\n", slot_id, &dev->in_ctx->dma);
970
971 /* Initialize the cancellation and bandwidth list for each ep */
972 for (i = 0; i < 31; i++) {
973 dev->eps[i].ep_index = i;
974 dev->eps[i].vdev = dev;
975 dev->eps[i].xhci = xhci;
976 INIT_LIST_HEAD(&dev->eps[i].cancelled_td_list);
977 INIT_LIST_HEAD(&dev->eps[i].bw_endpoint_list);
978 }
979
980 /* Allocate endpoint 0 ring */
981 dev->eps[0].ring = xhci_ring_alloc(xhci, 2, TYPE_CTRL, 0, flags);
982 if (!dev->eps[0].ring)
983 goto fail;
984
985 dev->udev = udev;
986
987 /* Point to output device context in dcbaa. */
988 xhci->dcbaa->dev_context_ptrs[slot_id] = cpu_to_le64(dev->out_ctx->dma);
989 xhci_dbg(xhci, "Set slot id %d dcbaa entry %p to 0x%llx\n",
990 slot_id,
991 &xhci->dcbaa->dev_context_ptrs[slot_id],
992 le64_to_cpu(xhci->dcbaa->dev_context_ptrs[slot_id]));
993
994 trace_xhci_alloc_virt_device(dev);
995
996 xhci->devs[slot_id] = dev;
997
998 return 1;
999fail:
1000
1001 if (dev->in_ctx)
1002 xhci_free_container_ctx(xhci, dev->in_ctx);
1003 if (dev->out_ctx)
1004 xhci_free_container_ctx(xhci, dev->out_ctx);
1005 kfree(dev);
1006
1007 return 0;
1008}
1009
1010void xhci_copy_ep0_dequeue_into_input_ctx(struct xhci_hcd *xhci,
1011 struct usb_device *udev)
1012{
1013 struct xhci_virt_device *virt_dev;
1014 struct xhci_ep_ctx *ep0_ctx;
1015 struct xhci_ring *ep_ring;
1016
1017 virt_dev = xhci->devs[udev->slot_id];
1018 ep0_ctx = xhci_get_ep_ctx(xhci, virt_dev->in_ctx, 0);
1019 ep_ring = virt_dev->eps[0].ring;
1020 /*
1021 * FIXME we don't keep track of the dequeue pointer very well after a
1022 * Set TR dequeue pointer, so we're setting the dequeue pointer of the
1023 * host to our enqueue pointer. This should only be called after a
1024 * configured device has reset, so all control transfers should have
1025 * been completed or cancelled before the reset.
1026 */
1027 ep0_ctx->deq = cpu_to_le64(xhci_trb_virt_to_dma(ep_ring->enq_seg,
1028 ep_ring->enqueue)
1029 | ep_ring->cycle_state);
1030}
1031
1032/*
1033 * The xHCI roothub may have ports of differing speeds in any order in the port
1034 * status registers.
1035 *
1036 * The xHCI hardware wants to know the roothub port that the USB device
1037 * is attached to (or the roothub port its ancestor hub is attached to). All we
1038 * know is the index of that port under either the USB 2.0 or the USB 3.0
1039 * roothub, but that doesn't give us the real index into the HW port status
1040 * registers.
1041 */
1042static struct xhci_port *xhci_find_rhub_port(struct xhci_hcd *xhci, struct usb_device *udev)
1043{
1044 struct usb_device *top_dev;
1045 struct xhci_hub *rhub;
1046 struct usb_hcd *hcd;
1047
1048 if (udev->speed >= USB_SPEED_SUPER)
1049 hcd = xhci_get_usb3_hcd(xhci);
1050 else
1051 hcd = xhci->main_hcd;
1052
1053 for (top_dev = udev; top_dev->parent && top_dev->parent->parent;
1054 top_dev = top_dev->parent)
1055 /* Found device below root hub */;
1056
1057 rhub = xhci_get_rhub(hcd);
1058 return rhub->ports[top_dev->portnum - 1];
1059}
1060
1061/* Setup an xHCI virtual device for a Set Address command */
1062int xhci_setup_addressable_virt_dev(struct xhci_hcd *xhci, struct usb_device *udev)
1063{
1064 struct xhci_virt_device *dev;
1065 struct xhci_ep_ctx *ep0_ctx;
1066 struct xhci_slot_ctx *slot_ctx;
1067 u32 max_packets;
1068
1069 dev = xhci->devs[udev->slot_id];
1070 /* Slot ID 0 is reserved */
1071 if (udev->slot_id == 0 || !dev) {
1072 xhci_warn(xhci, "Slot ID %d is not assigned to this device\n",
1073 udev->slot_id);
1074 return -EINVAL;
1075 }
1076 ep0_ctx = xhci_get_ep_ctx(xhci, dev->in_ctx, 0);
1077 slot_ctx = xhci_get_slot_ctx(xhci, dev->in_ctx);
1078
1079 /* 3) Only the control endpoint is valid - one endpoint context */
1080 slot_ctx->dev_info |= cpu_to_le32(LAST_CTX(1) | udev->route);
1081 switch (udev->speed) {
1082 case USB_SPEED_SUPER_PLUS:
1083 slot_ctx->dev_info |= cpu_to_le32(SLOT_SPEED_SSP);
1084 max_packets = MAX_PACKET(512);
1085 break;
1086 case USB_SPEED_SUPER:
1087 slot_ctx->dev_info |= cpu_to_le32(SLOT_SPEED_SS);
1088 max_packets = MAX_PACKET(512);
1089 break;
1090 case USB_SPEED_HIGH:
1091 slot_ctx->dev_info |= cpu_to_le32(SLOT_SPEED_HS);
1092 max_packets = MAX_PACKET(64);
1093 break;
1094 /* USB core guesses at a 64-byte max packet first for FS devices */
1095 case USB_SPEED_FULL:
1096 slot_ctx->dev_info |= cpu_to_le32(SLOT_SPEED_FS);
1097 max_packets = MAX_PACKET(64);
1098 break;
1099 case USB_SPEED_LOW:
1100 slot_ctx->dev_info |= cpu_to_le32(SLOT_SPEED_LS);
1101 max_packets = MAX_PACKET(8);
1102 break;
1103 default:
1104 /* Speed was set earlier, this shouldn't happen. */
1105 return -EINVAL;
1106 }
1107 /* Find the root hub port this device is under */
1108 dev->rhub_port = xhci_find_rhub_port(xhci, udev);
1109 if (!dev->rhub_port)
1110 return -EINVAL;
1111 /* Slot ID is set to the device directly below the root hub */
1112 if (!udev->parent->parent)
1113 dev->rhub_port->slot_id = udev->slot_id;
1114 slot_ctx->dev_info2 |= cpu_to_le32(ROOT_HUB_PORT(dev->rhub_port->hw_portnum + 1));
1115 xhci_dbg(xhci, "Slot ID %d: HW portnum %d, hcd portnum %d\n",
1116 udev->slot_id, dev->rhub_port->hw_portnum, dev->rhub_port->hcd_portnum);
1117
1118 /* Find the right bandwidth table that this device will be a part of.
1119 * If this is a full speed device attached directly to a root port (or a
1120 * decendent of one), it counts as a primary bandwidth domain, not a
1121 * secondary bandwidth domain under a TT. An xhci_tt_info structure
1122 * will never be created for the HS root hub.
1123 */
1124 if (!udev->tt || !udev->tt->hub->parent) {
1125 dev->bw_table = &xhci->rh_bw[dev->rhub_port->hw_portnum].bw_table;
1126 } else {
1127 struct xhci_root_port_bw_info *rh_bw;
1128 struct xhci_tt_bw_info *tt_bw;
1129
1130 rh_bw = &xhci->rh_bw[dev->rhub_port->hw_portnum];
1131 /* Find the right TT. */
1132 list_for_each_entry(tt_bw, &rh_bw->tts, tt_list) {
1133 if (tt_bw->slot_id != udev->tt->hub->slot_id)
1134 continue;
1135
1136 if (!dev->udev->tt->multi ||
1137 (udev->tt->multi &&
1138 tt_bw->ttport == dev->udev->ttport)) {
1139 dev->bw_table = &tt_bw->bw_table;
1140 dev->tt_info = tt_bw;
1141 break;
1142 }
1143 }
1144 if (!dev->tt_info)
1145 xhci_warn(xhci, "WARN: Didn't find a matching TT\n");
1146 }
1147
1148 /* Is this a LS/FS device under an external HS hub? */
1149 if (udev->tt && udev->tt->hub->parent) {
1150 slot_ctx->tt_info = cpu_to_le32(udev->tt->hub->slot_id |
1151 (udev->ttport << 8));
1152 if (udev->tt->multi)
1153 slot_ctx->dev_info |= cpu_to_le32(DEV_MTT);
1154 }
1155 xhci_dbg(xhci, "udev->tt = %p\n", udev->tt);
1156 xhci_dbg(xhci, "udev->ttport = 0x%x\n", udev->ttport);
1157
1158 /* Step 4 - ring already allocated */
1159 /* Step 5 */
1160 ep0_ctx->ep_info2 = cpu_to_le32(EP_TYPE(CTRL_EP));
1161
1162 /* EP 0 can handle "burst" sizes of 1, so Max Burst Size field is 0 */
1163 ep0_ctx->ep_info2 |= cpu_to_le32(MAX_BURST(0) | ERROR_COUNT(3) |
1164 max_packets);
1165
1166 ep0_ctx->deq = cpu_to_le64(dev->eps[0].ring->first_seg->dma |
1167 dev->eps[0].ring->cycle_state);
1168
1169 trace_xhci_setup_addressable_virt_device(dev);
1170
1171 /* Steps 7 and 8 were done in xhci_alloc_virt_device() */
1172
1173 return 0;
1174}
1175
1176/*
1177 * Convert interval expressed as 2^(bInterval - 1) == interval into
1178 * straight exponent value 2^n == interval.
1179 *
1180 */
1181static unsigned int xhci_parse_exponent_interval(struct usb_device *udev,
1182 struct usb_host_endpoint *ep)
1183{
1184 unsigned int interval;
1185
1186 interval = clamp_val(ep->desc.bInterval, 1, 16) - 1;
1187 if (interval != ep->desc.bInterval - 1)
1188 dev_warn(&udev->dev,
1189 "ep %#x - rounding interval to %d %sframes\n",
1190 ep->desc.bEndpointAddress,
1191 1 << interval,
1192 udev->speed == USB_SPEED_FULL ? "" : "micro");
1193
1194 if (udev->speed == USB_SPEED_FULL) {
1195 /*
1196 * Full speed isoc endpoints specify interval in frames,
1197 * not microframes. We are using microframes everywhere,
1198 * so adjust accordingly.
1199 */
1200 interval += 3; /* 1 frame = 2^3 uframes */
1201 }
1202
1203 return interval;
1204}
1205
1206/*
1207 * Convert bInterval expressed in microframes (in 1-255 range) to exponent of
1208 * microframes, rounded down to nearest power of 2.
1209 */
1210static unsigned int xhci_microframes_to_exponent(struct usb_device *udev,
1211 struct usb_host_endpoint *ep, unsigned int desc_interval,
1212 unsigned int min_exponent, unsigned int max_exponent)
1213{
1214 unsigned int interval;
1215
1216 interval = fls(desc_interval) - 1;
1217 interval = clamp_val(interval, min_exponent, max_exponent);
1218 if ((1 << interval) != desc_interval)
1219 dev_dbg(&udev->dev,
1220 "ep %#x - rounding interval to %d microframes, ep desc says %d microframes\n",
1221 ep->desc.bEndpointAddress,
1222 1 << interval,
1223 desc_interval);
1224
1225 return interval;
1226}
1227
1228static unsigned int xhci_parse_microframe_interval(struct usb_device *udev,
1229 struct usb_host_endpoint *ep)
1230{
1231 if (ep->desc.bInterval == 0)
1232 return 0;
1233 return xhci_microframes_to_exponent(udev, ep,
1234 ep->desc.bInterval, 0, 15);
1235}
1236
1237
1238static unsigned int xhci_parse_frame_interval(struct usb_device *udev,
1239 struct usb_host_endpoint *ep)
1240{
1241 return xhci_microframes_to_exponent(udev, ep,
1242 ep->desc.bInterval * 8, 3, 10);
1243}
1244
1245/* Return the polling or NAK interval.
1246 *
1247 * The polling interval is expressed in "microframes". If xHCI's Interval field
1248 * is set to N, it will service the endpoint every 2^(Interval)*125us.
1249 *
1250 * The NAK interval is one NAK per 1 to 255 microframes, or no NAKs if interval
1251 * is set to 0.
1252 */
1253static unsigned int xhci_get_endpoint_interval(struct usb_device *udev,
1254 struct usb_host_endpoint *ep)
1255{
1256 unsigned int interval = 0;
1257
1258 switch (udev->speed) {
1259 case USB_SPEED_HIGH:
1260 /* Max NAK rate */
1261 if (usb_endpoint_xfer_control(&ep->desc) ||
1262 usb_endpoint_xfer_bulk(&ep->desc)) {
1263 interval = xhci_parse_microframe_interval(udev, ep);
1264 break;
1265 }
1266 fallthrough; /* SS and HS isoc/int have same decoding */
1267
1268 case USB_SPEED_SUPER_PLUS:
1269 case USB_SPEED_SUPER:
1270 if (usb_endpoint_xfer_int(&ep->desc) ||
1271 usb_endpoint_xfer_isoc(&ep->desc)) {
1272 interval = xhci_parse_exponent_interval(udev, ep);
1273 }
1274 break;
1275
1276 case USB_SPEED_FULL:
1277 if (usb_endpoint_xfer_isoc(&ep->desc)) {
1278 interval = xhci_parse_exponent_interval(udev, ep);
1279 break;
1280 }
1281 /*
1282 * Fall through for interrupt endpoint interval decoding
1283 * since it uses the same rules as low speed interrupt
1284 * endpoints.
1285 */
1286 fallthrough;
1287
1288 case USB_SPEED_LOW:
1289 if (usb_endpoint_xfer_int(&ep->desc) ||
1290 usb_endpoint_xfer_isoc(&ep->desc)) {
1291
1292 interval = xhci_parse_frame_interval(udev, ep);
1293 }
1294 break;
1295
1296 default:
1297 BUG();
1298 }
1299 return interval;
1300}
1301
1302/* The "Mult" field in the endpoint context is only set for SuperSpeed isoc eps.
1303 * High speed endpoint descriptors can define "the number of additional
1304 * transaction opportunities per microframe", but that goes in the Max Burst
1305 * endpoint context field.
1306 */
1307static u32 xhci_get_endpoint_mult(struct usb_device *udev,
1308 struct usb_host_endpoint *ep)
1309{
1310 if (udev->speed < USB_SPEED_SUPER ||
1311 !usb_endpoint_xfer_isoc(&ep->desc))
1312 return 0;
1313 return ep->ss_ep_comp.bmAttributes;
1314}
1315
1316static u32 xhci_get_endpoint_max_burst(struct usb_device *udev,
1317 struct usb_host_endpoint *ep)
1318{
1319 /* Super speed and Plus have max burst in ep companion desc */
1320 if (udev->speed >= USB_SPEED_SUPER)
1321 return ep->ss_ep_comp.bMaxBurst;
1322
1323 if (udev->speed == USB_SPEED_HIGH &&
1324 (usb_endpoint_xfer_isoc(&ep->desc) ||
1325 usb_endpoint_xfer_int(&ep->desc)))
1326 return usb_endpoint_maxp_mult(&ep->desc) - 1;
1327
1328 return 0;
1329}
1330
1331static u32 xhci_get_endpoint_type(struct usb_host_endpoint *ep)
1332{
1333 int in;
1334
1335 in = usb_endpoint_dir_in(&ep->desc);
1336
1337 switch (usb_endpoint_type(&ep->desc)) {
1338 case USB_ENDPOINT_XFER_CONTROL:
1339 return CTRL_EP;
1340 case USB_ENDPOINT_XFER_BULK:
1341 return in ? BULK_IN_EP : BULK_OUT_EP;
1342 case USB_ENDPOINT_XFER_ISOC:
1343 return in ? ISOC_IN_EP : ISOC_OUT_EP;
1344 case USB_ENDPOINT_XFER_INT:
1345 return in ? INT_IN_EP : INT_OUT_EP;
1346 }
1347 return 0;
1348}
1349
1350/* Return the maximum endpoint service interval time (ESIT) payload.
1351 * Basically, this is the maxpacket size, multiplied by the burst size
1352 * and mult size.
1353 */
1354static u32 xhci_get_max_esit_payload(struct usb_device *udev,
1355 struct usb_host_endpoint *ep)
1356{
1357 int max_burst;
1358 int max_packet;
1359
1360 /* Only applies for interrupt or isochronous endpoints */
1361 if (usb_endpoint_xfer_control(&ep->desc) ||
1362 usb_endpoint_xfer_bulk(&ep->desc))
1363 return 0;
1364
1365 /* SuperSpeedPlus Isoc ep sending over 48k per esit */
1366 if ((udev->speed >= USB_SPEED_SUPER_PLUS) &&
1367 USB_SS_SSP_ISOC_COMP(ep->ss_ep_comp.bmAttributes))
1368 return le32_to_cpu(ep->ssp_isoc_ep_comp.dwBytesPerInterval);
1369
1370 /* SuperSpeed or SuperSpeedPlus Isoc ep with less than 48k per esit */
1371 if (udev->speed >= USB_SPEED_SUPER)
1372 return le16_to_cpu(ep->ss_ep_comp.wBytesPerInterval);
1373
1374 max_packet = usb_endpoint_maxp(&ep->desc);
1375 max_burst = usb_endpoint_maxp_mult(&ep->desc);
1376 /* A 0 in max burst means 1 transfer per ESIT */
1377 return max_packet * max_burst;
1378}
1379
1380/* Set up an endpoint with one ring segment. Do not allocate stream rings.
1381 * Drivers will have to call usb_alloc_streams() to do that.
1382 */
1383int xhci_endpoint_init(struct xhci_hcd *xhci,
1384 struct xhci_virt_device *virt_dev,
1385 struct usb_device *udev,
1386 struct usb_host_endpoint *ep,
1387 gfp_t mem_flags)
1388{
1389 unsigned int ep_index;
1390 struct xhci_ep_ctx *ep_ctx;
1391 struct xhci_ring *ep_ring;
1392 unsigned int max_packet;
1393 enum xhci_ring_type ring_type;
1394 u32 max_esit_payload;
1395 u32 endpoint_type;
1396 unsigned int max_burst;
1397 unsigned int interval;
1398 unsigned int mult;
1399 unsigned int avg_trb_len;
1400 unsigned int err_count = 0;
1401
1402 ep_index = xhci_get_endpoint_index(&ep->desc);
1403 ep_ctx = xhci_get_ep_ctx(xhci, virt_dev->in_ctx, ep_index);
1404
1405 endpoint_type = xhci_get_endpoint_type(ep);
1406 if (!endpoint_type)
1407 return -EINVAL;
1408
1409 ring_type = usb_endpoint_type(&ep->desc);
1410
1411 /*
1412 * Get values to fill the endpoint context, mostly from ep descriptor.
1413 * The average TRB buffer lengt for bulk endpoints is unclear as we
1414 * have no clue on scatter gather list entry size. For Isoc and Int,
1415 * set it to max available. See xHCI 1.1 spec 4.14.1.1 for details.
1416 */
1417 max_esit_payload = xhci_get_max_esit_payload(udev, ep);
1418 interval = xhci_get_endpoint_interval(udev, ep);
1419
1420 /* Periodic endpoint bInterval limit quirk */
1421 if (usb_endpoint_xfer_int(&ep->desc) ||
1422 usb_endpoint_xfer_isoc(&ep->desc)) {
1423 if ((xhci->quirks & XHCI_LIMIT_ENDPOINT_INTERVAL_7) &&
1424 udev->speed >= USB_SPEED_HIGH &&
1425 interval >= 7) {
1426 interval = 6;
1427 }
1428 }
1429
1430 mult = xhci_get_endpoint_mult(udev, ep);
1431 max_packet = usb_endpoint_maxp(&ep->desc);
1432 max_burst = xhci_get_endpoint_max_burst(udev, ep);
1433 avg_trb_len = max_esit_payload;
1434
1435 /* FIXME dig Mult and streams info out of ep companion desc */
1436
1437 /* Allow 3 retries for everything but isoc, set CErr = 3 */
1438 if (!usb_endpoint_xfer_isoc(&ep->desc))
1439 err_count = 3;
1440 /* HS bulk max packet should be 512, FS bulk supports 8, 16, 32 or 64 */
1441 if (usb_endpoint_xfer_bulk(&ep->desc)) {
1442 if (udev->speed == USB_SPEED_HIGH)
1443 max_packet = 512;
1444 if (udev->speed == USB_SPEED_FULL) {
1445 max_packet = rounddown_pow_of_two(max_packet);
1446 max_packet = clamp_val(max_packet, 8, 64);
1447 }
1448 }
1449 /* xHCI 1.0 and 1.1 indicates that ctrl ep avg TRB Length should be 8 */
1450 if (usb_endpoint_xfer_control(&ep->desc) && xhci->hci_version >= 0x100)
1451 avg_trb_len = 8;
1452 /* xhci 1.1 with LEC support doesn't use mult field, use RsvdZ */
1453 if ((xhci->hci_version > 0x100) && HCC2_LEC(xhci->hcc_params2))
1454 mult = 0;
1455
1456 /* Set up the endpoint ring */
1457 virt_dev->eps[ep_index].new_ring =
1458 xhci_ring_alloc(xhci, 2, ring_type, max_packet, mem_flags);
1459 if (!virt_dev->eps[ep_index].new_ring)
1460 return -ENOMEM;
1461
1462 virt_dev->eps[ep_index].skip = false;
1463 ep_ring = virt_dev->eps[ep_index].new_ring;
1464
1465 /* Fill the endpoint context */
1466 ep_ctx->ep_info = cpu_to_le32(EP_MAX_ESIT_PAYLOAD_HI(max_esit_payload) |
1467 EP_INTERVAL(interval) |
1468 EP_MULT(mult));
1469 ep_ctx->ep_info2 = cpu_to_le32(EP_TYPE(endpoint_type) |
1470 MAX_PACKET(max_packet) |
1471 MAX_BURST(max_burst) |
1472 ERROR_COUNT(err_count));
1473 ep_ctx->deq = cpu_to_le64(ep_ring->first_seg->dma |
1474 ep_ring->cycle_state);
1475
1476 ep_ctx->tx_info = cpu_to_le32(EP_MAX_ESIT_PAYLOAD_LO(max_esit_payload) |
1477 EP_AVG_TRB_LENGTH(avg_trb_len));
1478
1479 return 0;
1480}
1481
1482void xhci_endpoint_zero(struct xhci_hcd *xhci,
1483 struct xhci_virt_device *virt_dev,
1484 struct usb_host_endpoint *ep)
1485{
1486 unsigned int ep_index;
1487 struct xhci_ep_ctx *ep_ctx;
1488
1489 ep_index = xhci_get_endpoint_index(&ep->desc);
1490 ep_ctx = xhci_get_ep_ctx(xhci, virt_dev->in_ctx, ep_index);
1491
1492 ep_ctx->ep_info = 0;
1493 ep_ctx->ep_info2 = 0;
1494 ep_ctx->deq = 0;
1495 ep_ctx->tx_info = 0;
1496 /* Don't free the endpoint ring until the set interface or configuration
1497 * request succeeds.
1498 */
1499}
1500
1501void xhci_clear_endpoint_bw_info(struct xhci_bw_info *bw_info)
1502{
1503 bw_info->ep_interval = 0;
1504 bw_info->mult = 0;
1505 bw_info->num_packets = 0;
1506 bw_info->max_packet_size = 0;
1507 bw_info->type = 0;
1508 bw_info->max_esit_payload = 0;
1509}
1510
1511void xhci_update_bw_info(struct xhci_hcd *xhci,
1512 struct xhci_container_ctx *in_ctx,
1513 struct xhci_input_control_ctx *ctrl_ctx,
1514 struct xhci_virt_device *virt_dev)
1515{
1516 struct xhci_bw_info *bw_info;
1517 struct xhci_ep_ctx *ep_ctx;
1518 unsigned int ep_type;
1519 int i;
1520
1521 for (i = 1; i < 31; i++) {
1522 bw_info = &virt_dev->eps[i].bw_info;
1523
1524 /* We can't tell what endpoint type is being dropped, but
1525 * unconditionally clearing the bandwidth info for non-periodic
1526 * endpoints should be harmless because the info will never be
1527 * set in the first place.
1528 */
1529 if (!EP_IS_ADDED(ctrl_ctx, i) && EP_IS_DROPPED(ctrl_ctx, i)) {
1530 /* Dropped endpoint */
1531 xhci_clear_endpoint_bw_info(bw_info);
1532 continue;
1533 }
1534
1535 if (EP_IS_ADDED(ctrl_ctx, i)) {
1536 ep_ctx = xhci_get_ep_ctx(xhci, in_ctx, i);
1537 ep_type = CTX_TO_EP_TYPE(le32_to_cpu(ep_ctx->ep_info2));
1538
1539 /* Ignore non-periodic endpoints */
1540 if (ep_type != ISOC_OUT_EP && ep_type != INT_OUT_EP &&
1541 ep_type != ISOC_IN_EP &&
1542 ep_type != INT_IN_EP)
1543 continue;
1544
1545 /* Added or changed endpoint */
1546 bw_info->ep_interval = CTX_TO_EP_INTERVAL(
1547 le32_to_cpu(ep_ctx->ep_info));
1548 /* Number of packets and mult are zero-based in the
1549 * input context, but we want one-based for the
1550 * interval table.
1551 */
1552 bw_info->mult = CTX_TO_EP_MULT(
1553 le32_to_cpu(ep_ctx->ep_info)) + 1;
1554 bw_info->num_packets = CTX_TO_MAX_BURST(
1555 le32_to_cpu(ep_ctx->ep_info2)) + 1;
1556 bw_info->max_packet_size = MAX_PACKET_DECODED(
1557 le32_to_cpu(ep_ctx->ep_info2));
1558 bw_info->type = ep_type;
1559 bw_info->max_esit_payload = CTX_TO_MAX_ESIT_PAYLOAD(
1560 le32_to_cpu(ep_ctx->tx_info));
1561 }
1562 }
1563}
1564
1565/* Copy output xhci_ep_ctx to the input xhci_ep_ctx copy.
1566 * Useful when you want to change one particular aspect of the endpoint and then
1567 * issue a configure endpoint command.
1568 */
1569void xhci_endpoint_copy(struct xhci_hcd *xhci,
1570 struct xhci_container_ctx *in_ctx,
1571 struct xhci_container_ctx *out_ctx,
1572 unsigned int ep_index)
1573{
1574 struct xhci_ep_ctx *out_ep_ctx;
1575 struct xhci_ep_ctx *in_ep_ctx;
1576
1577 out_ep_ctx = xhci_get_ep_ctx(xhci, out_ctx, ep_index);
1578 in_ep_ctx = xhci_get_ep_ctx(xhci, in_ctx, ep_index);
1579
1580 in_ep_ctx->ep_info = out_ep_ctx->ep_info;
1581 in_ep_ctx->ep_info2 = out_ep_ctx->ep_info2;
1582 in_ep_ctx->deq = out_ep_ctx->deq;
1583 in_ep_ctx->tx_info = out_ep_ctx->tx_info;
1584 if (xhci->quirks & XHCI_MTK_HOST) {
1585 in_ep_ctx->reserved[0] = out_ep_ctx->reserved[0];
1586 in_ep_ctx->reserved[1] = out_ep_ctx->reserved[1];
1587 }
1588}
1589
1590/* Copy output xhci_slot_ctx to the input xhci_slot_ctx.
1591 * Useful when you want to change one particular aspect of the endpoint and then
1592 * issue a configure endpoint command. Only the context entries field matters,
1593 * but we'll copy the whole thing anyway.
1594 */
1595void xhci_slot_copy(struct xhci_hcd *xhci,
1596 struct xhci_container_ctx *in_ctx,
1597 struct xhci_container_ctx *out_ctx)
1598{
1599 struct xhci_slot_ctx *in_slot_ctx;
1600 struct xhci_slot_ctx *out_slot_ctx;
1601
1602 in_slot_ctx = xhci_get_slot_ctx(xhci, in_ctx);
1603 out_slot_ctx = xhci_get_slot_ctx(xhci, out_ctx);
1604
1605 in_slot_ctx->dev_info = out_slot_ctx->dev_info;
1606 in_slot_ctx->dev_info2 = out_slot_ctx->dev_info2;
1607 in_slot_ctx->tt_info = out_slot_ctx->tt_info;
1608 in_slot_ctx->dev_state = out_slot_ctx->dev_state;
1609}
1610
1611/* Set up the scratchpad buffer array and scratchpad buffers, if needed. */
1612static int scratchpad_alloc(struct xhci_hcd *xhci, gfp_t flags)
1613{
1614 int i;
1615 struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
1616 int num_sp = HCS_MAX_SCRATCHPAD(xhci->hcs_params2);
1617
1618 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
1619 "Allocating %d scratchpad buffers", num_sp);
1620
1621 if (!num_sp)
1622 return 0;
1623
1624 xhci->scratchpad = kzalloc_node(sizeof(*xhci->scratchpad), flags,
1625 dev_to_node(dev));
1626 if (!xhci->scratchpad)
1627 goto fail_sp;
1628
1629 xhci->scratchpad->sp_array = dma_alloc_coherent(dev,
1630 array_size(sizeof(u64), num_sp),
1631 &xhci->scratchpad->sp_dma, flags);
1632 if (!xhci->scratchpad->sp_array)
1633 goto fail_sp2;
1634
1635 xhci->scratchpad->sp_buffers = kcalloc_node(num_sp, sizeof(void *),
1636 flags, dev_to_node(dev));
1637 if (!xhci->scratchpad->sp_buffers)
1638 goto fail_sp3;
1639
1640 xhci->dcbaa->dev_context_ptrs[0] = cpu_to_le64(xhci->scratchpad->sp_dma);
1641 for (i = 0; i < num_sp; i++) {
1642 dma_addr_t dma;
1643 void *buf = dma_alloc_coherent(dev, xhci->page_size, &dma,
1644 flags);
1645 if (!buf)
1646 goto fail_sp4;
1647
1648 xhci->scratchpad->sp_array[i] = dma;
1649 xhci->scratchpad->sp_buffers[i] = buf;
1650 }
1651
1652 return 0;
1653
1654 fail_sp4:
1655 while (i--)
1656 dma_free_coherent(dev, xhci->page_size,
1657 xhci->scratchpad->sp_buffers[i],
1658 xhci->scratchpad->sp_array[i]);
1659
1660 kfree(xhci->scratchpad->sp_buffers);
1661
1662 fail_sp3:
1663 dma_free_coherent(dev, array_size(sizeof(u64), num_sp),
1664 xhci->scratchpad->sp_array,
1665 xhci->scratchpad->sp_dma);
1666
1667 fail_sp2:
1668 kfree(xhci->scratchpad);
1669 xhci->scratchpad = NULL;
1670
1671 fail_sp:
1672 return -ENOMEM;
1673}
1674
1675static void scratchpad_free(struct xhci_hcd *xhci)
1676{
1677 int num_sp;
1678 int i;
1679 struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
1680
1681 if (!xhci->scratchpad)
1682 return;
1683
1684 num_sp = HCS_MAX_SCRATCHPAD(xhci->hcs_params2);
1685
1686 for (i = 0; i < num_sp; i++) {
1687 dma_free_coherent(dev, xhci->page_size,
1688 xhci->scratchpad->sp_buffers[i],
1689 xhci->scratchpad->sp_array[i]);
1690 }
1691 kfree(xhci->scratchpad->sp_buffers);
1692 dma_free_coherent(dev, array_size(sizeof(u64), num_sp),
1693 xhci->scratchpad->sp_array,
1694 xhci->scratchpad->sp_dma);
1695 kfree(xhci->scratchpad);
1696 xhci->scratchpad = NULL;
1697}
1698
1699struct xhci_command *xhci_alloc_command(struct xhci_hcd *xhci,
1700 bool allocate_completion, gfp_t mem_flags)
1701{
1702 struct xhci_command *command;
1703 struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
1704
1705 command = kzalloc_node(sizeof(*command), mem_flags, dev_to_node(dev));
1706 if (!command)
1707 return NULL;
1708
1709 if (allocate_completion) {
1710 command->completion =
1711 kzalloc_node(sizeof(struct completion), mem_flags,
1712 dev_to_node(dev));
1713 if (!command->completion) {
1714 kfree(command);
1715 return NULL;
1716 }
1717 init_completion(command->completion);
1718 }
1719
1720 command->status = 0;
1721 /* set default timeout to 5000 ms */
1722 command->timeout_ms = XHCI_CMD_DEFAULT_TIMEOUT;
1723 INIT_LIST_HEAD(&command->cmd_list);
1724 return command;
1725}
1726
1727struct xhci_command *xhci_alloc_command_with_ctx(struct xhci_hcd *xhci,
1728 bool allocate_completion, gfp_t mem_flags)
1729{
1730 struct xhci_command *command;
1731
1732 command = xhci_alloc_command(xhci, allocate_completion, mem_flags);
1733 if (!command)
1734 return NULL;
1735
1736 command->in_ctx = xhci_alloc_container_ctx(xhci, XHCI_CTX_TYPE_INPUT,
1737 mem_flags);
1738 if (!command->in_ctx) {
1739 kfree(command->completion);
1740 kfree(command);
1741 return NULL;
1742 }
1743 return command;
1744}
1745
1746void xhci_urb_free_priv(struct urb_priv *urb_priv)
1747{
1748 kfree(urb_priv);
1749}
1750
1751void xhci_free_command(struct xhci_hcd *xhci,
1752 struct xhci_command *command)
1753{
1754 xhci_free_container_ctx(xhci,
1755 command->in_ctx);
1756 kfree(command->completion);
1757 kfree(command);
1758}
1759
1760static int xhci_alloc_erst(struct xhci_hcd *xhci,
1761 struct xhci_ring *evt_ring,
1762 struct xhci_erst *erst,
1763 gfp_t flags)
1764{
1765 size_t size;
1766 unsigned int val;
1767 struct xhci_segment *seg;
1768 struct xhci_erst_entry *entry;
1769
1770 size = array_size(sizeof(struct xhci_erst_entry), evt_ring->num_segs);
1771 erst->entries = dma_alloc_coherent(xhci_to_hcd(xhci)->self.sysdev,
1772 size, &erst->erst_dma_addr, flags);
1773 if (!erst->entries)
1774 return -ENOMEM;
1775
1776 erst->num_entries = evt_ring->num_segs;
1777
1778 seg = evt_ring->first_seg;
1779 for (val = 0; val < evt_ring->num_segs; val++) {
1780 entry = &erst->entries[val];
1781 entry->seg_addr = cpu_to_le64(seg->dma);
1782 entry->seg_size = cpu_to_le32(TRBS_PER_SEGMENT);
1783 entry->rsvd = 0;
1784 seg = seg->next;
1785 }
1786
1787 return 0;
1788}
1789
1790static void
1791xhci_remove_interrupter(struct xhci_hcd *xhci, struct xhci_interrupter *ir)
1792{
1793 u32 tmp;
1794
1795 if (!ir)
1796 return;
1797
1798 /*
1799 * Clean out interrupter registers except ERSTBA. Clearing either the
1800 * low or high 32 bits of ERSTBA immediately causes the controller to
1801 * dereference the partially cleared 64 bit address, causing IOMMU error.
1802 */
1803 if (ir->ir_set) {
1804 tmp = readl(&ir->ir_set->erst_size);
1805 tmp &= ERST_SIZE_MASK;
1806 writel(tmp, &ir->ir_set->erst_size);
1807
1808 xhci_write_64(xhci, ERST_EHB, &ir->ir_set->erst_dequeue);
1809 }
1810}
1811
1812static void
1813xhci_free_interrupter(struct xhci_hcd *xhci, struct xhci_interrupter *ir)
1814{
1815 struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
1816 size_t erst_size;
1817
1818 if (!ir)
1819 return;
1820
1821 erst_size = array_size(sizeof(struct xhci_erst_entry), ir->erst.num_entries);
1822 if (ir->erst.entries)
1823 dma_free_coherent(dev, erst_size,
1824 ir->erst.entries,
1825 ir->erst.erst_dma_addr);
1826 ir->erst.entries = NULL;
1827
1828 /* free interrupter event ring */
1829 if (ir->event_ring)
1830 xhci_ring_free(xhci, ir->event_ring);
1831
1832 ir->event_ring = NULL;
1833
1834 kfree(ir);
1835}
1836
1837void xhci_remove_secondary_interrupter(struct usb_hcd *hcd, struct xhci_interrupter *ir)
1838{
1839 struct xhci_hcd *xhci = hcd_to_xhci(hcd);
1840 unsigned int intr_num;
1841
1842 spin_lock_irq(&xhci->lock);
1843
1844 /* interrupter 0 is primary interrupter, don't touch it */
1845 if (!ir || !ir->intr_num || ir->intr_num >= xhci->max_interrupters) {
1846 xhci_dbg(xhci, "Invalid secondary interrupter, can't remove\n");
1847 spin_unlock_irq(&xhci->lock);
1848 return;
1849 }
1850
1851 intr_num = ir->intr_num;
1852
1853 xhci_remove_interrupter(xhci, ir);
1854 xhci->interrupters[intr_num] = NULL;
1855
1856 spin_unlock_irq(&xhci->lock);
1857
1858 xhci_free_interrupter(xhci, ir);
1859}
1860EXPORT_SYMBOL_GPL(xhci_remove_secondary_interrupter);
1861
1862void xhci_mem_cleanup(struct xhci_hcd *xhci)
1863{
1864 struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
1865 int i, j, num_ports;
1866
1867 cancel_delayed_work_sync(&xhci->cmd_timer);
1868
1869 for (i = 0; xhci->interrupters && i < xhci->max_interrupters; i++) {
1870 if (xhci->interrupters[i]) {
1871 xhci_remove_interrupter(xhci, xhci->interrupters[i]);
1872 xhci_free_interrupter(xhci, xhci->interrupters[i]);
1873 xhci->interrupters[i] = NULL;
1874 }
1875 }
1876 xhci_dbg_trace(xhci, trace_xhci_dbg_init, "Freed interrupters");
1877
1878 if (xhci->cmd_ring)
1879 xhci_ring_free(xhci, xhci->cmd_ring);
1880 xhci->cmd_ring = NULL;
1881 xhci_dbg_trace(xhci, trace_xhci_dbg_init, "Freed command ring");
1882 xhci_cleanup_command_queue(xhci);
1883
1884 num_ports = HCS_MAX_PORTS(xhci->hcs_params1);
1885 for (i = 0; i < num_ports && xhci->rh_bw; i++) {
1886 struct xhci_interval_bw_table *bwt = &xhci->rh_bw[i].bw_table;
1887 for (j = 0; j < XHCI_MAX_INTERVAL; j++) {
1888 struct list_head *ep = &bwt->interval_bw[j].endpoints;
1889 while (!list_empty(ep))
1890 list_del_init(ep->next);
1891 }
1892 }
1893
1894 for (i = HCS_MAX_SLOTS(xhci->hcs_params1); i > 0; i--)
1895 xhci_free_virt_devices_depth_first(xhci, i);
1896
1897 dma_pool_destroy(xhci->segment_pool);
1898 xhci->segment_pool = NULL;
1899 xhci_dbg_trace(xhci, trace_xhci_dbg_init, "Freed segment pool");
1900
1901 dma_pool_destroy(xhci->device_pool);
1902 xhci->device_pool = NULL;
1903 xhci_dbg_trace(xhci, trace_xhci_dbg_init, "Freed device context pool");
1904
1905 dma_pool_destroy(xhci->small_streams_pool);
1906 xhci->small_streams_pool = NULL;
1907 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
1908 "Freed small stream array pool");
1909
1910 dma_pool_destroy(xhci->medium_streams_pool);
1911 xhci->medium_streams_pool = NULL;
1912 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
1913 "Freed medium stream array pool");
1914
1915 if (xhci->dcbaa)
1916 dma_free_coherent(dev, sizeof(*xhci->dcbaa),
1917 xhci->dcbaa, xhci->dcbaa->dma);
1918 xhci->dcbaa = NULL;
1919
1920 scratchpad_free(xhci);
1921
1922 if (!xhci->rh_bw)
1923 goto no_bw;
1924
1925 for (i = 0; i < num_ports; i++) {
1926 struct xhci_tt_bw_info *tt, *n;
1927 list_for_each_entry_safe(tt, n, &xhci->rh_bw[i].tts, tt_list) {
1928 list_del(&tt->tt_list);
1929 kfree(tt);
1930 }
1931 }
1932
1933no_bw:
1934 xhci->cmd_ring_reserved_trbs = 0;
1935 xhci->usb2_rhub.num_ports = 0;
1936 xhci->usb3_rhub.num_ports = 0;
1937 xhci->num_active_eps = 0;
1938 kfree(xhci->usb2_rhub.ports);
1939 kfree(xhci->usb3_rhub.ports);
1940 kfree(xhci->hw_ports);
1941 kfree(xhci->rh_bw);
1942 for (i = 0; i < xhci->num_port_caps; i++)
1943 kfree(xhci->port_caps[i].psi);
1944 kfree(xhci->port_caps);
1945 kfree(xhci->interrupters);
1946 xhci->num_port_caps = 0;
1947
1948 xhci->usb2_rhub.ports = NULL;
1949 xhci->usb3_rhub.ports = NULL;
1950 xhci->hw_ports = NULL;
1951 xhci->rh_bw = NULL;
1952 xhci->port_caps = NULL;
1953 xhci->interrupters = NULL;
1954
1955 xhci->page_size = 0;
1956 xhci->page_shift = 0;
1957 xhci->usb2_rhub.bus_state.bus_suspended = 0;
1958 xhci->usb3_rhub.bus_state.bus_suspended = 0;
1959}
1960
1961static void xhci_set_hc_event_deq(struct xhci_hcd *xhci, struct xhci_interrupter *ir)
1962{
1963 dma_addr_t deq;
1964
1965 deq = xhci_trb_virt_to_dma(ir->event_ring->deq_seg,
1966 ir->event_ring->dequeue);
1967 if (!deq)
1968 xhci_warn(xhci, "WARN something wrong with SW event ring dequeue ptr.\n");
1969 /* Update HC event ring dequeue pointer */
1970 /* Don't clear the EHB bit (which is RW1C) because
1971 * there might be more events to service.
1972 */
1973 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
1974 "// Write event ring dequeue pointer, preserving EHB bit");
1975 xhci_write_64(xhci, deq & ERST_PTR_MASK, &ir->ir_set->erst_dequeue);
1976}
1977
1978static void xhci_add_in_port(struct xhci_hcd *xhci, unsigned int num_ports,
1979 __le32 __iomem *addr, int max_caps)
1980{
1981 u32 temp, port_offset, port_count;
1982 int i;
1983 u8 major_revision, minor_revision, tmp_minor_revision;
1984 struct xhci_hub *rhub;
1985 struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
1986 struct xhci_port_cap *port_cap;
1987
1988 temp = readl(addr);
1989 major_revision = XHCI_EXT_PORT_MAJOR(temp);
1990 minor_revision = XHCI_EXT_PORT_MINOR(temp);
1991
1992 if (major_revision == 0x03) {
1993 rhub = &xhci->usb3_rhub;
1994 /*
1995 * Some hosts incorrectly use sub-minor version for minor
1996 * version (i.e. 0x02 instead of 0x20 for bcdUSB 0x320 and 0x01
1997 * for bcdUSB 0x310). Since there is no USB release with sub
1998 * minor version 0x301 to 0x309, we can assume that they are
1999 * incorrect and fix it here.
2000 */
2001 if (minor_revision > 0x00 && minor_revision < 0x10)
2002 minor_revision <<= 4;
2003 /*
2004 * Some zhaoxin's xHCI controller that follow usb3.1 spec
2005 * but only support Gen1.
2006 */
2007 if (xhci->quirks & XHCI_ZHAOXIN_HOST) {
2008 tmp_minor_revision = minor_revision;
2009 minor_revision = 0;
2010 }
2011
2012 } else if (major_revision <= 0x02) {
2013 rhub = &xhci->usb2_rhub;
2014 } else {
2015 xhci_warn(xhci, "Ignoring unknown port speed, Ext Cap %p, revision = 0x%x\n",
2016 addr, major_revision);
2017 /* Ignoring port protocol we can't understand. FIXME */
2018 return;
2019 }
2020
2021 /* Port offset and count in the third dword, see section 7.2 */
2022 temp = readl(addr + 2);
2023 port_offset = XHCI_EXT_PORT_OFF(temp);
2024 port_count = XHCI_EXT_PORT_COUNT(temp);
2025 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2026 "Ext Cap %p, port offset = %u, count = %u, revision = 0x%x",
2027 addr, port_offset, port_count, major_revision);
2028 /* Port count includes the current port offset */
2029 if (port_offset == 0 || (port_offset + port_count - 1) > num_ports)
2030 /* WTF? "Valid values are ‘1’ to MaxPorts" */
2031 return;
2032
2033 port_cap = &xhci->port_caps[xhci->num_port_caps++];
2034 if (xhci->num_port_caps > max_caps)
2035 return;
2036
2037 port_cap->psi_count = XHCI_EXT_PORT_PSIC(temp);
2038
2039 if (port_cap->psi_count) {
2040 port_cap->psi = kcalloc_node(port_cap->psi_count,
2041 sizeof(*port_cap->psi),
2042 GFP_KERNEL, dev_to_node(dev));
2043 if (!port_cap->psi)
2044 port_cap->psi_count = 0;
2045
2046 port_cap->psi_uid_count++;
2047 for (i = 0; i < port_cap->psi_count; i++) {
2048 port_cap->psi[i] = readl(addr + 4 + i);
2049
2050 /* count unique ID values, two consecutive entries can
2051 * have the same ID if link is assymetric
2052 */
2053 if (i && (XHCI_EXT_PORT_PSIV(port_cap->psi[i]) !=
2054 XHCI_EXT_PORT_PSIV(port_cap->psi[i - 1])))
2055 port_cap->psi_uid_count++;
2056
2057 if (xhci->quirks & XHCI_ZHAOXIN_HOST &&
2058 major_revision == 0x03 &&
2059 XHCI_EXT_PORT_PSIV(port_cap->psi[i]) >= 5)
2060 minor_revision = tmp_minor_revision;
2061
2062 xhci_dbg(xhci, "PSIV:%d PSIE:%d PLT:%d PFD:%d LP:%d PSIM:%d\n",
2063 XHCI_EXT_PORT_PSIV(port_cap->psi[i]),
2064 XHCI_EXT_PORT_PSIE(port_cap->psi[i]),
2065 XHCI_EXT_PORT_PLT(port_cap->psi[i]),
2066 XHCI_EXT_PORT_PFD(port_cap->psi[i]),
2067 XHCI_EXT_PORT_LP(port_cap->psi[i]),
2068 XHCI_EXT_PORT_PSIM(port_cap->psi[i]));
2069 }
2070 }
2071
2072 rhub->maj_rev = major_revision;
2073
2074 if (rhub->min_rev < minor_revision)
2075 rhub->min_rev = minor_revision;
2076
2077 port_cap->maj_rev = major_revision;
2078 port_cap->min_rev = minor_revision;
2079 port_cap->protocol_caps = temp;
2080
2081 if ((xhci->hci_version >= 0x100) && (major_revision != 0x03) &&
2082 (temp & XHCI_HLC)) {
2083 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2084 "xHCI 1.0: support USB2 hardware lpm");
2085 xhci->hw_lpm_support = 1;
2086 }
2087
2088 port_offset--;
2089 for (i = port_offset; i < (port_offset + port_count); i++) {
2090 struct xhci_port *hw_port = &xhci->hw_ports[i];
2091 /* Duplicate entry. Ignore the port if the revisions differ. */
2092 if (hw_port->rhub) {
2093 xhci_warn(xhci, "Duplicate port entry, Ext Cap %p, port %u\n", addr, i);
2094 xhci_warn(xhci, "Port was marked as USB %u, duplicated as USB %u\n",
2095 hw_port->rhub->maj_rev, major_revision);
2096 /* Only adjust the roothub port counts if we haven't
2097 * found a similar duplicate.
2098 */
2099 if (hw_port->rhub != rhub &&
2100 hw_port->hcd_portnum != DUPLICATE_ENTRY) {
2101 hw_port->rhub->num_ports--;
2102 hw_port->hcd_portnum = DUPLICATE_ENTRY;
2103 }
2104 continue;
2105 }
2106 hw_port->rhub = rhub;
2107 hw_port->port_cap = port_cap;
2108 rhub->num_ports++;
2109 }
2110 /* FIXME: Should we disable ports not in the Extended Capabilities? */
2111}
2112
2113static void xhci_create_rhub_port_array(struct xhci_hcd *xhci,
2114 struct xhci_hub *rhub, gfp_t flags)
2115{
2116 int port_index = 0;
2117 int i;
2118 struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
2119
2120 if (!rhub->num_ports)
2121 return;
2122 rhub->ports = kcalloc_node(rhub->num_ports, sizeof(*rhub->ports),
2123 flags, dev_to_node(dev));
2124 if (!rhub->ports)
2125 return;
2126
2127 for (i = 0; i < HCS_MAX_PORTS(xhci->hcs_params1); i++) {
2128 if (xhci->hw_ports[i].rhub != rhub ||
2129 xhci->hw_ports[i].hcd_portnum == DUPLICATE_ENTRY)
2130 continue;
2131 xhci->hw_ports[i].hcd_portnum = port_index;
2132 rhub->ports[port_index] = &xhci->hw_ports[i];
2133 port_index++;
2134 if (port_index == rhub->num_ports)
2135 break;
2136 }
2137}
2138
2139/*
2140 * Scan the Extended Capabilities for the "Supported Protocol Capabilities" that
2141 * specify what speeds each port is supposed to be. We can't count on the port
2142 * speed bits in the PORTSC register being correct until a device is connected,
2143 * but we need to set up the two fake roothubs with the correct number of USB
2144 * 3.0 and USB 2.0 ports at host controller initialization time.
2145 */
2146static int xhci_setup_port_arrays(struct xhci_hcd *xhci, gfp_t flags)
2147{
2148 void __iomem *base;
2149 u32 offset;
2150 unsigned int num_ports;
2151 int i, j;
2152 int cap_count = 0;
2153 u32 cap_start;
2154 struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
2155
2156 num_ports = HCS_MAX_PORTS(xhci->hcs_params1);
2157 xhci->hw_ports = kcalloc_node(num_ports, sizeof(*xhci->hw_ports),
2158 flags, dev_to_node(dev));
2159 if (!xhci->hw_ports)
2160 return -ENOMEM;
2161
2162 for (i = 0; i < num_ports; i++) {
2163 xhci->hw_ports[i].addr = &xhci->op_regs->port_status_base +
2164 NUM_PORT_REGS * i;
2165 xhci->hw_ports[i].hw_portnum = i;
2166
2167 init_completion(&xhci->hw_ports[i].rexit_done);
2168 init_completion(&xhci->hw_ports[i].u3exit_done);
2169 }
2170
2171 xhci->rh_bw = kcalloc_node(num_ports, sizeof(*xhci->rh_bw), flags,
2172 dev_to_node(dev));
2173 if (!xhci->rh_bw)
2174 return -ENOMEM;
2175 for (i = 0; i < num_ports; i++) {
2176 struct xhci_interval_bw_table *bw_table;
2177
2178 INIT_LIST_HEAD(&xhci->rh_bw[i].tts);
2179 bw_table = &xhci->rh_bw[i].bw_table;
2180 for (j = 0; j < XHCI_MAX_INTERVAL; j++)
2181 INIT_LIST_HEAD(&bw_table->interval_bw[j].endpoints);
2182 }
2183 base = &xhci->cap_regs->hc_capbase;
2184
2185 cap_start = xhci_find_next_ext_cap(base, 0, XHCI_EXT_CAPS_PROTOCOL);
2186 if (!cap_start) {
2187 xhci_err(xhci, "No Extended Capability registers, unable to set up roothub\n");
2188 return -ENODEV;
2189 }
2190
2191 offset = cap_start;
2192 /* count extended protocol capability entries for later caching */
2193 while (offset) {
2194 cap_count++;
2195 offset = xhci_find_next_ext_cap(base, offset,
2196 XHCI_EXT_CAPS_PROTOCOL);
2197 }
2198
2199 xhci->port_caps = kcalloc_node(cap_count, sizeof(*xhci->port_caps),
2200 flags, dev_to_node(dev));
2201 if (!xhci->port_caps)
2202 return -ENOMEM;
2203
2204 offset = cap_start;
2205
2206 while (offset) {
2207 xhci_add_in_port(xhci, num_ports, base + offset, cap_count);
2208 if (xhci->usb2_rhub.num_ports + xhci->usb3_rhub.num_ports ==
2209 num_ports)
2210 break;
2211 offset = xhci_find_next_ext_cap(base, offset,
2212 XHCI_EXT_CAPS_PROTOCOL);
2213 }
2214 if (xhci->usb2_rhub.num_ports == 0 && xhci->usb3_rhub.num_ports == 0) {
2215 xhci_warn(xhci, "No ports on the roothubs?\n");
2216 return -ENODEV;
2217 }
2218 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2219 "Found %u USB 2.0 ports and %u USB 3.0 ports.",
2220 xhci->usb2_rhub.num_ports, xhci->usb3_rhub.num_ports);
2221
2222 /* Place limits on the number of roothub ports so that the hub
2223 * descriptors aren't longer than the USB core will allocate.
2224 */
2225 if (xhci->usb3_rhub.num_ports > USB_SS_MAXPORTS) {
2226 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2227 "Limiting USB 3.0 roothub ports to %u.",
2228 USB_SS_MAXPORTS);
2229 xhci->usb3_rhub.num_ports = USB_SS_MAXPORTS;
2230 }
2231 if (xhci->usb2_rhub.num_ports > USB_MAXCHILDREN) {
2232 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2233 "Limiting USB 2.0 roothub ports to %u.",
2234 USB_MAXCHILDREN);
2235 xhci->usb2_rhub.num_ports = USB_MAXCHILDREN;
2236 }
2237
2238 if (!xhci->usb2_rhub.num_ports)
2239 xhci_info(xhci, "USB2 root hub has no ports\n");
2240
2241 if (!xhci->usb3_rhub.num_ports)
2242 xhci_info(xhci, "USB3 root hub has no ports\n");
2243
2244 xhci_create_rhub_port_array(xhci, &xhci->usb2_rhub, flags);
2245 xhci_create_rhub_port_array(xhci, &xhci->usb3_rhub, flags);
2246
2247 return 0;
2248}
2249
2250static struct xhci_interrupter *
2251xhci_alloc_interrupter(struct xhci_hcd *xhci, unsigned int segs, gfp_t flags)
2252{
2253 struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
2254 struct xhci_interrupter *ir;
2255 unsigned int max_segs;
2256 int ret;
2257
2258 if (!segs)
2259 segs = ERST_DEFAULT_SEGS;
2260
2261 max_segs = BIT(HCS_ERST_MAX(xhci->hcs_params2));
2262 segs = min(segs, max_segs);
2263
2264 ir = kzalloc_node(sizeof(*ir), flags, dev_to_node(dev));
2265 if (!ir)
2266 return NULL;
2267
2268 ir->event_ring = xhci_ring_alloc(xhci, segs, TYPE_EVENT, 0, flags);
2269 if (!ir->event_ring) {
2270 xhci_warn(xhci, "Failed to allocate interrupter event ring\n");
2271 kfree(ir);
2272 return NULL;
2273 }
2274
2275 ret = xhci_alloc_erst(xhci, ir->event_ring, &ir->erst, flags);
2276 if (ret) {
2277 xhci_warn(xhci, "Failed to allocate interrupter erst\n");
2278 xhci_ring_free(xhci, ir->event_ring);
2279 kfree(ir);
2280 return NULL;
2281 }
2282
2283 return ir;
2284}
2285
2286static int
2287xhci_add_interrupter(struct xhci_hcd *xhci, struct xhci_interrupter *ir,
2288 unsigned int intr_num)
2289{
2290 u64 erst_base;
2291 u32 erst_size;
2292
2293 if (intr_num >= xhci->max_interrupters) {
2294 xhci_warn(xhci, "Can't add interrupter %d, max interrupters %d\n",
2295 intr_num, xhci->max_interrupters);
2296 return -EINVAL;
2297 }
2298
2299 if (xhci->interrupters[intr_num]) {
2300 xhci_warn(xhci, "Interrupter %d\n already set up", intr_num);
2301 return -EINVAL;
2302 }
2303
2304 xhci->interrupters[intr_num] = ir;
2305 ir->intr_num = intr_num;
2306 ir->ir_set = &xhci->run_regs->ir_set[intr_num];
2307
2308 /* set ERST count with the number of entries in the segment table */
2309 erst_size = readl(&ir->ir_set->erst_size);
2310 erst_size &= ERST_SIZE_MASK;
2311 erst_size |= ir->event_ring->num_segs;
2312 writel(erst_size, &ir->ir_set->erst_size);
2313
2314 erst_base = xhci_read_64(xhci, &ir->ir_set->erst_base);
2315 erst_base &= ERST_BASE_RSVDP;
2316 erst_base |= ir->erst.erst_dma_addr & ~ERST_BASE_RSVDP;
2317 if (xhci->quirks & XHCI_WRITE_64_HI_LO)
2318 hi_lo_writeq(erst_base, &ir->ir_set->erst_base);
2319 else
2320 xhci_write_64(xhci, erst_base, &ir->ir_set->erst_base);
2321
2322 /* Set the event ring dequeue address of this interrupter */
2323 xhci_set_hc_event_deq(xhci, ir);
2324
2325 return 0;
2326}
2327
2328struct xhci_interrupter *
2329xhci_create_secondary_interrupter(struct usb_hcd *hcd, unsigned int segs,
2330 u32 imod_interval)
2331{
2332 struct xhci_hcd *xhci = hcd_to_xhci(hcd);
2333 struct xhci_interrupter *ir;
2334 unsigned int i;
2335 int err = -ENOSPC;
2336
2337 if (!xhci->interrupters || xhci->max_interrupters <= 1)
2338 return NULL;
2339
2340 ir = xhci_alloc_interrupter(xhci, segs, GFP_KERNEL);
2341 if (!ir)
2342 return NULL;
2343
2344 spin_lock_irq(&xhci->lock);
2345
2346 /* Find available secondary interrupter, interrupter 0 is reserved for primary */
2347 for (i = 1; i < xhci->max_interrupters; i++) {
2348 if (xhci->interrupters[i] == NULL) {
2349 err = xhci_add_interrupter(xhci, ir, i);
2350 break;
2351 }
2352 }
2353
2354 spin_unlock_irq(&xhci->lock);
2355
2356 if (err) {
2357 xhci_warn(xhci, "Failed to add secondary interrupter, max interrupters %d\n",
2358 xhci->max_interrupters);
2359 xhci_free_interrupter(xhci, ir);
2360 return NULL;
2361 }
2362
2363 err = xhci_set_interrupter_moderation(ir, imod_interval);
2364 if (err)
2365 xhci_warn(xhci, "Failed to set interrupter %d moderation to %uns\n",
2366 i, imod_interval);
2367
2368 xhci_dbg(xhci, "Add secondary interrupter %d, max interrupters %d\n",
2369 i, xhci->max_interrupters);
2370
2371 return ir;
2372}
2373EXPORT_SYMBOL_GPL(xhci_create_secondary_interrupter);
2374
2375int xhci_mem_init(struct xhci_hcd *xhci, gfp_t flags)
2376{
2377 struct xhci_interrupter *ir;
2378 struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
2379 dma_addr_t dma;
2380 unsigned int val, val2;
2381 u64 val_64;
2382 u32 page_size, temp;
2383 int i;
2384
2385 INIT_LIST_HEAD(&xhci->cmd_list);
2386
2387 /* init command timeout work */
2388 INIT_DELAYED_WORK(&xhci->cmd_timer, xhci_handle_command_timeout);
2389 init_completion(&xhci->cmd_ring_stop_completion);
2390
2391 page_size = readl(&xhci->op_regs->page_size);
2392 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2393 "Supported page size register = 0x%x", page_size);
2394 i = ffs(page_size);
2395 if (i < 16)
2396 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2397 "Supported page size of %iK", (1 << (i+12)) / 1024);
2398 else
2399 xhci_warn(xhci, "WARN: no supported page size\n");
2400 /* Use 4K pages, since that's common and the minimum the HC supports */
2401 xhci->page_shift = 12;
2402 xhci->page_size = 1 << xhci->page_shift;
2403 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2404 "HCD page size set to %iK", xhci->page_size / 1024);
2405
2406 /*
2407 * Program the Number of Device Slots Enabled field in the CONFIG
2408 * register with the max value of slots the HC can handle.
2409 */
2410 val = HCS_MAX_SLOTS(readl(&xhci->cap_regs->hcs_params1));
2411 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2412 "// xHC can handle at most %d device slots.", val);
2413 val2 = readl(&xhci->op_regs->config_reg);
2414 val |= (val2 & ~HCS_SLOTS_MASK);
2415 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2416 "// Setting Max device slots reg = 0x%x.", val);
2417 writel(val, &xhci->op_regs->config_reg);
2418
2419 /*
2420 * xHCI section 5.4.6 - Device Context array must be
2421 * "physically contiguous and 64-byte (cache line) aligned".
2422 */
2423 xhci->dcbaa = dma_alloc_coherent(dev, sizeof(*xhci->dcbaa), &dma,
2424 flags);
2425 if (!xhci->dcbaa)
2426 goto fail;
2427 xhci->dcbaa->dma = dma;
2428 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2429 "// Device context base array address = 0x%pad (DMA), %p (virt)",
2430 &xhci->dcbaa->dma, xhci->dcbaa);
2431 xhci_write_64(xhci, dma, &xhci->op_regs->dcbaa_ptr);
2432
2433 /*
2434 * Initialize the ring segment pool. The ring must be a contiguous
2435 * structure comprised of TRBs. The TRBs must be 16 byte aligned,
2436 * however, the command ring segment needs 64-byte aligned segments
2437 * and our use of dma addresses in the trb_address_map radix tree needs
2438 * TRB_SEGMENT_SIZE alignment, so we pick the greater alignment need.
2439 */
2440 if (xhci->quirks & XHCI_TRB_OVERFETCH)
2441 /* Buggy HC prefetches beyond segment bounds - allocate dummy space at the end */
2442 xhci->segment_pool = dma_pool_create("xHCI ring segments", dev,
2443 TRB_SEGMENT_SIZE * 2, TRB_SEGMENT_SIZE * 2, xhci->page_size * 2);
2444 else
2445 xhci->segment_pool = dma_pool_create("xHCI ring segments", dev,
2446 TRB_SEGMENT_SIZE, TRB_SEGMENT_SIZE, xhci->page_size);
2447
2448 /* See Table 46 and Note on Figure 55 */
2449 xhci->device_pool = dma_pool_create("xHCI input/output contexts", dev,
2450 2112, 64, xhci->page_size);
2451 if (!xhci->segment_pool || !xhci->device_pool)
2452 goto fail;
2453
2454 /* Linear stream context arrays don't have any boundary restrictions,
2455 * and only need to be 16-byte aligned.
2456 */
2457 xhci->small_streams_pool =
2458 dma_pool_create("xHCI 256 byte stream ctx arrays",
2459 dev, SMALL_STREAM_ARRAY_SIZE, 16, 0);
2460 xhci->medium_streams_pool =
2461 dma_pool_create("xHCI 1KB stream ctx arrays",
2462 dev, MEDIUM_STREAM_ARRAY_SIZE, 16, 0);
2463 /* Any stream context array bigger than MEDIUM_STREAM_ARRAY_SIZE
2464 * will be allocated with dma_alloc_coherent()
2465 */
2466
2467 if (!xhci->small_streams_pool || !xhci->medium_streams_pool)
2468 goto fail;
2469
2470 /* Set up the command ring to have one segments for now. */
2471 xhci->cmd_ring = xhci_ring_alloc(xhci, 1, TYPE_COMMAND, 0, flags);
2472 if (!xhci->cmd_ring)
2473 goto fail;
2474 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2475 "Allocated command ring at %p", xhci->cmd_ring);
2476 xhci_dbg_trace(xhci, trace_xhci_dbg_init, "First segment DMA is 0x%pad",
2477 &xhci->cmd_ring->first_seg->dma);
2478
2479 /* Set the address in the Command Ring Control register */
2480 val_64 = xhci_read_64(xhci, &xhci->op_regs->cmd_ring);
2481 val_64 = (val_64 & (u64) CMD_RING_RSVD_BITS) |
2482 (xhci->cmd_ring->first_seg->dma & (u64) ~CMD_RING_RSVD_BITS) |
2483 xhci->cmd_ring->cycle_state;
2484 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2485 "// Setting command ring address to 0x%016llx", val_64);
2486 xhci_write_64(xhci, val_64, &xhci->op_regs->cmd_ring);
2487
2488 /* Reserve one command ring TRB for disabling LPM.
2489 * Since the USB core grabs the shared usb_bus bandwidth mutex before
2490 * disabling LPM, we only need to reserve one TRB for all devices.
2491 */
2492 xhci->cmd_ring_reserved_trbs++;
2493
2494 val = readl(&xhci->cap_regs->db_off);
2495 val &= DBOFF_MASK;
2496 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2497 "// Doorbell array is located at offset 0x%x from cap regs base addr",
2498 val);
2499 xhci->dba = (void __iomem *) xhci->cap_regs + val;
2500
2501 /* Allocate and set up primary interrupter 0 with an event ring. */
2502 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2503 "Allocating primary event ring");
2504 xhci->interrupters = kcalloc_node(xhci->max_interrupters, sizeof(*xhci->interrupters),
2505 flags, dev_to_node(dev));
2506
2507 ir = xhci_alloc_interrupter(xhci, 0, flags);
2508 if (!ir)
2509 goto fail;
2510
2511 if (xhci_add_interrupter(xhci, ir, 0))
2512 goto fail;
2513
2514 ir->isoc_bei_interval = AVOID_BEI_INTERVAL_MAX;
2515
2516 for (i = 0; i < MAX_HC_SLOTS; i++)
2517 xhci->devs[i] = NULL;
2518
2519 if (scratchpad_alloc(xhci, flags))
2520 goto fail;
2521 if (xhci_setup_port_arrays(xhci, flags))
2522 goto fail;
2523
2524 /* Enable USB 3.0 device notifications for function remote wake, which
2525 * is necessary for allowing USB 3.0 devices to do remote wakeup from
2526 * U3 (device suspend).
2527 */
2528 temp = readl(&xhci->op_regs->dev_notification);
2529 temp &= ~DEV_NOTE_MASK;
2530 temp |= DEV_NOTE_FWAKE;
2531 writel(temp, &xhci->op_regs->dev_notification);
2532
2533 return 0;
2534
2535fail:
2536 xhci_halt(xhci);
2537 xhci_reset(xhci, XHCI_RESET_SHORT_USEC);
2538 xhci_mem_cleanup(xhci);
2539 return -ENOMEM;
2540}