Merge remote-tracking branch 'asoc/topic/arizona' into asoc-next
[deliverable/linux.git] / drivers / usb / host / xhci-mem.c
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
31 /*
32 * Allocates a generic ring segment from the ring pool, sets the dma address,
33 * initializes the segment to zero, and sets the private next pointer to NULL.
34 *
35 * Section 4.11.1.1:
36 * "All components of all Command and Transfer TRBs shall be initialized to '0'"
37 */
38 static struct xhci_segment *xhci_segment_alloc(struct xhci_hcd *xhci,
39 unsigned int cycle_state, gfp_t flags)
40 {
41 struct xhci_segment *seg;
42 dma_addr_t dma;
43 int i;
44
45 seg = kzalloc(sizeof *seg, flags);
46 if (!seg)
47 return NULL;
48
49 seg->trbs = dma_pool_alloc(xhci->segment_pool, flags, &dma);
50 if (!seg->trbs) {
51 kfree(seg);
52 return NULL;
53 }
54
55 memset(seg->trbs, 0, TRB_SEGMENT_SIZE);
56 /* If the cycle state is 0, set the cycle bit to 1 for all the TRBs */
57 if (cycle_state == 0) {
58 for (i = 0; i < TRBS_PER_SEGMENT; i++)
59 seg->trbs[i].link.control |= TRB_CYCLE;
60 }
61 seg->dma = dma;
62 seg->next = NULL;
63
64 return seg;
65 }
66
67 static void xhci_segment_free(struct xhci_hcd *xhci, struct xhci_segment *seg)
68 {
69 if (seg->trbs) {
70 dma_pool_free(xhci->segment_pool, seg->trbs, seg->dma);
71 seg->trbs = NULL;
72 }
73 kfree(seg);
74 }
75
76 static void xhci_free_segments_for_ring(struct xhci_hcd *xhci,
77 struct xhci_segment *first)
78 {
79 struct xhci_segment *seg;
80
81 seg = first->next;
82 while (seg != first) {
83 struct xhci_segment *next = seg->next;
84 xhci_segment_free(xhci, seg);
85 seg = next;
86 }
87 xhci_segment_free(xhci, first);
88 }
89
90 /*
91 * Make the prev segment point to the next segment.
92 *
93 * Change the last TRB in the prev segment to be a Link TRB which points to the
94 * DMA address of the next segment. The caller needs to set any Link TRB
95 * related flags, such as End TRB, Toggle Cycle, and no snoop.
96 */
97 static void xhci_link_segments(struct xhci_hcd *xhci, struct xhci_segment *prev,
98 struct xhci_segment *next, enum xhci_ring_type type)
99 {
100 u32 val;
101
102 if (!prev || !next)
103 return;
104 prev->next = next;
105 if (type != TYPE_EVENT) {
106 prev->trbs[TRBS_PER_SEGMENT-1].link.segment_ptr =
107 cpu_to_le64(next->dma);
108
109 /* Set the last TRB in the segment to have a TRB type ID of Link TRB */
110 val = le32_to_cpu(prev->trbs[TRBS_PER_SEGMENT-1].link.control);
111 val &= ~TRB_TYPE_BITMASK;
112 val |= TRB_TYPE(TRB_LINK);
113 /* Always set the chain bit with 0.95 hardware */
114 /* Set chain bit for isoc rings on AMD 0.96 host */
115 if (xhci_link_trb_quirk(xhci) ||
116 (type == TYPE_ISOC &&
117 (xhci->quirks & XHCI_AMD_0x96_HOST)))
118 val |= TRB_CHAIN;
119 prev->trbs[TRBS_PER_SEGMENT-1].link.control = cpu_to_le32(val);
120 }
121 }
122
123 /*
124 * Link the ring to the new segments.
125 * Set Toggle Cycle for the new ring if needed.
126 */
127 static void xhci_link_rings(struct xhci_hcd *xhci, struct xhci_ring *ring,
128 struct xhci_segment *first, struct xhci_segment *last,
129 unsigned int num_segs)
130 {
131 struct xhci_segment *next;
132
133 if (!ring || !first || !last)
134 return;
135
136 next = ring->enq_seg->next;
137 xhci_link_segments(xhci, ring->enq_seg, first, ring->type);
138 xhci_link_segments(xhci, last, next, ring->type);
139 ring->num_segs += num_segs;
140 ring->num_trbs_free += (TRBS_PER_SEGMENT - 1) * num_segs;
141
142 if (ring->type != TYPE_EVENT && ring->enq_seg == ring->last_seg) {
143 ring->last_seg->trbs[TRBS_PER_SEGMENT-1].link.control
144 &= ~cpu_to_le32(LINK_TOGGLE);
145 last->trbs[TRBS_PER_SEGMENT-1].link.control
146 |= cpu_to_le32(LINK_TOGGLE);
147 ring->last_seg = last;
148 }
149 }
150
151 /* XXX: Do we need the hcd structure in all these functions? */
152 void xhci_ring_free(struct xhci_hcd *xhci, struct xhci_ring *ring)
153 {
154 if (!ring)
155 return;
156
157 if (ring->first_seg)
158 xhci_free_segments_for_ring(xhci, ring->first_seg);
159
160 kfree(ring);
161 }
162
163 static void xhci_initialize_ring_info(struct xhci_ring *ring,
164 unsigned int cycle_state)
165 {
166 /* The ring is empty, so the enqueue pointer == dequeue pointer */
167 ring->enqueue = ring->first_seg->trbs;
168 ring->enq_seg = ring->first_seg;
169 ring->dequeue = ring->enqueue;
170 ring->deq_seg = ring->first_seg;
171 /* The ring is initialized to 0. The producer must write 1 to the cycle
172 * bit to handover ownership of the TRB, so PCS = 1. The consumer must
173 * compare CCS to the cycle bit to check ownership, so CCS = 1.
174 *
175 * New rings are initialized with cycle state equal to 1; if we are
176 * handling ring expansion, set the cycle state equal to the old ring.
177 */
178 ring->cycle_state = cycle_state;
179 /* Not necessary for new rings, but needed for re-initialized rings */
180 ring->enq_updates = 0;
181 ring->deq_updates = 0;
182
183 /*
184 * Each segment has a link TRB, and leave an extra TRB for SW
185 * accounting purpose
186 */
187 ring->num_trbs_free = ring->num_segs * (TRBS_PER_SEGMENT - 1) - 1;
188 }
189
190 /* Allocate segments and link them for a ring */
191 static int xhci_alloc_segments_for_ring(struct xhci_hcd *xhci,
192 struct xhci_segment **first, struct xhci_segment **last,
193 unsigned int num_segs, unsigned int cycle_state,
194 enum xhci_ring_type type, gfp_t flags)
195 {
196 struct xhci_segment *prev;
197
198 prev = xhci_segment_alloc(xhci, cycle_state, flags);
199 if (!prev)
200 return -ENOMEM;
201 num_segs--;
202
203 *first = prev;
204 while (num_segs > 0) {
205 struct xhci_segment *next;
206
207 next = xhci_segment_alloc(xhci, cycle_state, flags);
208 if (!next) {
209 prev = *first;
210 while (prev) {
211 next = prev->next;
212 xhci_segment_free(xhci, prev);
213 prev = next;
214 }
215 return -ENOMEM;
216 }
217 xhci_link_segments(xhci, prev, next, type);
218
219 prev = next;
220 num_segs--;
221 }
222 xhci_link_segments(xhci, prev, *first, type);
223 *last = prev;
224
225 return 0;
226 }
227
228 /**
229 * Create a new ring with zero or more segments.
230 *
231 * Link each segment together into a ring.
232 * Set the end flag and the cycle toggle bit on the last segment.
233 * See section 4.9.1 and figures 15 and 16.
234 */
235 static struct xhci_ring *xhci_ring_alloc(struct xhci_hcd *xhci,
236 unsigned int num_segs, unsigned int cycle_state,
237 enum xhci_ring_type type, gfp_t flags)
238 {
239 struct xhci_ring *ring;
240 int ret;
241
242 ring = kzalloc(sizeof *(ring), flags);
243 if (!ring)
244 return NULL;
245
246 ring->num_segs = num_segs;
247 INIT_LIST_HEAD(&ring->td_list);
248 ring->type = type;
249 if (num_segs == 0)
250 return ring;
251
252 ret = xhci_alloc_segments_for_ring(xhci, &ring->first_seg,
253 &ring->last_seg, num_segs, cycle_state, type, flags);
254 if (ret)
255 goto fail;
256
257 /* Only event ring does not use link TRB */
258 if (type != TYPE_EVENT) {
259 /* See section 4.9.2.1 and 6.4.4.1 */
260 ring->last_seg->trbs[TRBS_PER_SEGMENT - 1].link.control |=
261 cpu_to_le32(LINK_TOGGLE);
262 }
263 xhci_initialize_ring_info(ring, cycle_state);
264 return ring;
265
266 fail:
267 kfree(ring);
268 return NULL;
269 }
270
271 void xhci_free_or_cache_endpoint_ring(struct xhci_hcd *xhci,
272 struct xhci_virt_device *virt_dev,
273 unsigned int ep_index)
274 {
275 int rings_cached;
276
277 rings_cached = virt_dev->num_rings_cached;
278 if (rings_cached < XHCI_MAX_RINGS_CACHED) {
279 virt_dev->ring_cache[rings_cached] =
280 virt_dev->eps[ep_index].ring;
281 virt_dev->num_rings_cached++;
282 xhci_dbg(xhci, "Cached old ring, "
283 "%d ring%s cached\n",
284 virt_dev->num_rings_cached,
285 (virt_dev->num_rings_cached > 1) ? "s" : "");
286 } else {
287 xhci_ring_free(xhci, virt_dev->eps[ep_index].ring);
288 xhci_dbg(xhci, "Ring cache full (%d rings), "
289 "freeing ring\n",
290 virt_dev->num_rings_cached);
291 }
292 virt_dev->eps[ep_index].ring = NULL;
293 }
294
295 /* Zero an endpoint ring (except for link TRBs) and move the enqueue and dequeue
296 * pointers to the beginning of the ring.
297 */
298 static void xhci_reinit_cached_ring(struct xhci_hcd *xhci,
299 struct xhci_ring *ring, unsigned int cycle_state,
300 enum xhci_ring_type type)
301 {
302 struct xhci_segment *seg = ring->first_seg;
303 int i;
304
305 do {
306 memset(seg->trbs, 0,
307 sizeof(union xhci_trb)*TRBS_PER_SEGMENT);
308 if (cycle_state == 0) {
309 for (i = 0; i < TRBS_PER_SEGMENT; i++)
310 seg->trbs[i].link.control |= TRB_CYCLE;
311 }
312 /* All endpoint rings have link TRBs */
313 xhci_link_segments(xhci, seg, seg->next, type);
314 seg = seg->next;
315 } while (seg != ring->first_seg);
316 ring->type = type;
317 xhci_initialize_ring_info(ring, cycle_state);
318 /* td list should be empty since all URBs have been cancelled,
319 * but just in case...
320 */
321 INIT_LIST_HEAD(&ring->td_list);
322 }
323
324 /*
325 * Expand an existing ring.
326 * Look for a cached ring or allocate a new ring which has same segment numbers
327 * and link the two rings.
328 */
329 int xhci_ring_expansion(struct xhci_hcd *xhci, struct xhci_ring *ring,
330 unsigned int num_trbs, gfp_t flags)
331 {
332 struct xhci_segment *first;
333 struct xhci_segment *last;
334 unsigned int num_segs;
335 unsigned int num_segs_needed;
336 int ret;
337
338 num_segs_needed = (num_trbs + (TRBS_PER_SEGMENT - 1) - 1) /
339 (TRBS_PER_SEGMENT - 1);
340
341 /* Allocate number of segments we needed, or double the ring size */
342 num_segs = ring->num_segs > num_segs_needed ?
343 ring->num_segs : num_segs_needed;
344
345 ret = xhci_alloc_segments_for_ring(xhci, &first, &last,
346 num_segs, ring->cycle_state, ring->type, flags);
347 if (ret)
348 return -ENOMEM;
349
350 xhci_link_rings(xhci, ring, first, last, num_segs);
351 xhci_dbg(xhci, "ring expansion succeed, now has %d segments\n",
352 ring->num_segs);
353
354 return 0;
355 }
356
357 #define CTX_SIZE(_hcc) (HCC_64BYTE_CONTEXT(_hcc) ? 64 : 32)
358
359 static struct xhci_container_ctx *xhci_alloc_container_ctx(struct xhci_hcd *xhci,
360 int type, gfp_t flags)
361 {
362 struct xhci_container_ctx *ctx;
363
364 if ((type != XHCI_CTX_TYPE_DEVICE) && (type != XHCI_CTX_TYPE_INPUT))
365 return NULL;
366
367 ctx = kzalloc(sizeof(*ctx), flags);
368 if (!ctx)
369 return NULL;
370
371 ctx->type = type;
372 ctx->size = HCC_64BYTE_CONTEXT(xhci->hcc_params) ? 2048 : 1024;
373 if (type == XHCI_CTX_TYPE_INPUT)
374 ctx->size += CTX_SIZE(xhci->hcc_params);
375
376 ctx->bytes = dma_pool_alloc(xhci->device_pool, flags, &ctx->dma);
377 if (!ctx->bytes) {
378 kfree(ctx);
379 return NULL;
380 }
381 memset(ctx->bytes, 0, ctx->size);
382 return ctx;
383 }
384
385 static void xhci_free_container_ctx(struct xhci_hcd *xhci,
386 struct xhci_container_ctx *ctx)
387 {
388 if (!ctx)
389 return;
390 dma_pool_free(xhci->device_pool, ctx->bytes, ctx->dma);
391 kfree(ctx);
392 }
393
394 struct xhci_input_control_ctx *xhci_get_input_control_ctx(struct xhci_hcd *xhci,
395 struct xhci_container_ctx *ctx)
396 {
397 if (ctx->type != XHCI_CTX_TYPE_INPUT)
398 return NULL;
399
400 return (struct xhci_input_control_ctx *)ctx->bytes;
401 }
402
403 struct xhci_slot_ctx *xhci_get_slot_ctx(struct xhci_hcd *xhci,
404 struct xhci_container_ctx *ctx)
405 {
406 if (ctx->type == XHCI_CTX_TYPE_DEVICE)
407 return (struct xhci_slot_ctx *)ctx->bytes;
408
409 return (struct xhci_slot_ctx *)
410 (ctx->bytes + CTX_SIZE(xhci->hcc_params));
411 }
412
413 struct xhci_ep_ctx *xhci_get_ep_ctx(struct xhci_hcd *xhci,
414 struct xhci_container_ctx *ctx,
415 unsigned int ep_index)
416 {
417 /* increment ep index by offset of start of ep ctx array */
418 ep_index++;
419 if (ctx->type == XHCI_CTX_TYPE_INPUT)
420 ep_index++;
421
422 return (struct xhci_ep_ctx *)
423 (ctx->bytes + (ep_index * CTX_SIZE(xhci->hcc_params)));
424 }
425
426
427 /***************** Streams structures manipulation *************************/
428
429 static void xhci_free_stream_ctx(struct xhci_hcd *xhci,
430 unsigned int num_stream_ctxs,
431 struct xhci_stream_ctx *stream_ctx, dma_addr_t dma)
432 {
433 struct pci_dev *pdev = to_pci_dev(xhci_to_hcd(xhci)->self.controller);
434
435 if (num_stream_ctxs > MEDIUM_STREAM_ARRAY_SIZE)
436 dma_free_coherent(&pdev->dev,
437 sizeof(struct xhci_stream_ctx)*num_stream_ctxs,
438 stream_ctx, dma);
439 else if (num_stream_ctxs <= SMALL_STREAM_ARRAY_SIZE)
440 return dma_pool_free(xhci->small_streams_pool,
441 stream_ctx, dma);
442 else
443 return dma_pool_free(xhci->medium_streams_pool,
444 stream_ctx, dma);
445 }
446
447 /*
448 * The stream context array for each endpoint with bulk streams enabled can
449 * vary in size, based on:
450 * - how many streams the endpoint supports,
451 * - the maximum primary stream array size the host controller supports,
452 * - and how many streams the device driver asks for.
453 *
454 * The stream context array must be a power of 2, and can be as small as
455 * 64 bytes or as large as 1MB.
456 */
457 static struct xhci_stream_ctx *xhci_alloc_stream_ctx(struct xhci_hcd *xhci,
458 unsigned int num_stream_ctxs, dma_addr_t *dma,
459 gfp_t mem_flags)
460 {
461 struct pci_dev *pdev = to_pci_dev(xhci_to_hcd(xhci)->self.controller);
462
463 if (num_stream_ctxs > MEDIUM_STREAM_ARRAY_SIZE)
464 return dma_alloc_coherent(&pdev->dev,
465 sizeof(struct xhci_stream_ctx)*num_stream_ctxs,
466 dma, mem_flags);
467 else if (num_stream_ctxs <= SMALL_STREAM_ARRAY_SIZE)
468 return dma_pool_alloc(xhci->small_streams_pool,
469 mem_flags, dma);
470 else
471 return dma_pool_alloc(xhci->medium_streams_pool,
472 mem_flags, dma);
473 }
474
475 struct xhci_ring *xhci_dma_to_transfer_ring(
476 struct xhci_virt_ep *ep,
477 u64 address)
478 {
479 if (ep->ep_state & EP_HAS_STREAMS)
480 return radix_tree_lookup(&ep->stream_info->trb_address_map,
481 address >> TRB_SEGMENT_SHIFT);
482 return ep->ring;
483 }
484
485 /* Only use this when you know stream_info is valid */
486 #ifdef CONFIG_USB_XHCI_HCD_DEBUGGING
487 static struct xhci_ring *dma_to_stream_ring(
488 struct xhci_stream_info *stream_info,
489 u64 address)
490 {
491 return radix_tree_lookup(&stream_info->trb_address_map,
492 address >> TRB_SEGMENT_SHIFT);
493 }
494 #endif /* CONFIG_USB_XHCI_HCD_DEBUGGING */
495
496 struct xhci_ring *xhci_stream_id_to_ring(
497 struct xhci_virt_device *dev,
498 unsigned int ep_index,
499 unsigned int stream_id)
500 {
501 struct xhci_virt_ep *ep = &dev->eps[ep_index];
502
503 if (stream_id == 0)
504 return ep->ring;
505 if (!ep->stream_info)
506 return NULL;
507
508 if (stream_id > ep->stream_info->num_streams)
509 return NULL;
510 return ep->stream_info->stream_rings[stream_id];
511 }
512
513 #ifdef CONFIG_USB_XHCI_HCD_DEBUGGING
514 static int xhci_test_radix_tree(struct xhci_hcd *xhci,
515 unsigned int num_streams,
516 struct xhci_stream_info *stream_info)
517 {
518 u32 cur_stream;
519 struct xhci_ring *cur_ring;
520 u64 addr;
521
522 for (cur_stream = 1; cur_stream < num_streams; cur_stream++) {
523 struct xhci_ring *mapped_ring;
524 int trb_size = sizeof(union xhci_trb);
525
526 cur_ring = stream_info->stream_rings[cur_stream];
527 for (addr = cur_ring->first_seg->dma;
528 addr < cur_ring->first_seg->dma + TRB_SEGMENT_SIZE;
529 addr += trb_size) {
530 mapped_ring = dma_to_stream_ring(stream_info, addr);
531 if (cur_ring != mapped_ring) {
532 xhci_warn(xhci, "WARN: DMA address 0x%08llx "
533 "didn't map to stream ID %u; "
534 "mapped to ring %p\n",
535 (unsigned long long) addr,
536 cur_stream,
537 mapped_ring);
538 return -EINVAL;
539 }
540 }
541 /* One TRB after the end of the ring segment shouldn't return a
542 * pointer to the current ring (although it may be a part of a
543 * different ring).
544 */
545 mapped_ring = dma_to_stream_ring(stream_info, addr);
546 if (mapped_ring != cur_ring) {
547 /* One TRB before should also fail */
548 addr = cur_ring->first_seg->dma - trb_size;
549 mapped_ring = dma_to_stream_ring(stream_info, addr);
550 }
551 if (mapped_ring == cur_ring) {
552 xhci_warn(xhci, "WARN: Bad DMA address 0x%08llx "
553 "mapped to valid stream ID %u; "
554 "mapped ring = %p\n",
555 (unsigned long long) addr,
556 cur_stream,
557 mapped_ring);
558 return -EINVAL;
559 }
560 }
561 return 0;
562 }
563 #endif /* CONFIG_USB_XHCI_HCD_DEBUGGING */
564
565 /*
566 * Change an endpoint's internal structure so it supports stream IDs. The
567 * number of requested streams includes stream 0, which cannot be used by device
568 * drivers.
569 *
570 * The number of stream contexts in the stream context array may be bigger than
571 * the number of streams the driver wants to use. This is because the number of
572 * stream context array entries must be a power of two.
573 *
574 * We need a radix tree for mapping physical addresses of TRBs to which stream
575 * ID they belong to. We need to do this because the host controller won't tell
576 * us which stream ring the TRB came from. We could store the stream ID in an
577 * event data TRB, but that doesn't help us for the cancellation case, since the
578 * endpoint may stop before it reaches that event data TRB.
579 *
580 * The radix tree maps the upper portion of the TRB DMA address to a ring
581 * segment that has the same upper portion of DMA addresses. For example, say I
582 * have segments of size 1KB, that are always 64-byte aligned. A segment may
583 * start at 0x10c91000 and end at 0x10c913f0. If I use the upper 10 bits, the
584 * key to the stream ID is 0x43244. I can use the DMA address of the TRB to
585 * pass the radix tree a key to get the right stream ID:
586 *
587 * 0x10c90fff >> 10 = 0x43243
588 * 0x10c912c0 >> 10 = 0x43244
589 * 0x10c91400 >> 10 = 0x43245
590 *
591 * Obviously, only those TRBs with DMA addresses that are within the segment
592 * will make the radix tree return the stream ID for that ring.
593 *
594 * Caveats for the radix tree:
595 *
596 * The radix tree uses an unsigned long as a key pair. On 32-bit systems, an
597 * unsigned long will be 32-bits; on a 64-bit system an unsigned long will be
598 * 64-bits. Since we only request 32-bit DMA addresses, we can use that as the
599 * key on 32-bit or 64-bit systems (it would also be fine if we asked for 64-bit
600 * PCI DMA addresses on a 64-bit system). There might be a problem on 32-bit
601 * extended systems (where the DMA address can be bigger than 32-bits),
602 * if we allow the PCI dma mask to be bigger than 32-bits. So don't do that.
603 */
604 struct xhci_stream_info *xhci_alloc_stream_info(struct xhci_hcd *xhci,
605 unsigned int num_stream_ctxs,
606 unsigned int num_streams, gfp_t mem_flags)
607 {
608 struct xhci_stream_info *stream_info;
609 u32 cur_stream;
610 struct xhci_ring *cur_ring;
611 unsigned long key;
612 u64 addr;
613 int ret;
614
615 xhci_dbg(xhci, "Allocating %u streams and %u "
616 "stream context array entries.\n",
617 num_streams, num_stream_ctxs);
618 if (xhci->cmd_ring_reserved_trbs == MAX_RSVD_CMD_TRBS) {
619 xhci_dbg(xhci, "Command ring has no reserved TRBs available\n");
620 return NULL;
621 }
622 xhci->cmd_ring_reserved_trbs++;
623
624 stream_info = kzalloc(sizeof(struct xhci_stream_info), mem_flags);
625 if (!stream_info)
626 goto cleanup_trbs;
627
628 stream_info->num_streams = num_streams;
629 stream_info->num_stream_ctxs = num_stream_ctxs;
630
631 /* Initialize the array of virtual pointers to stream rings. */
632 stream_info->stream_rings = kzalloc(
633 sizeof(struct xhci_ring *)*num_streams,
634 mem_flags);
635 if (!stream_info->stream_rings)
636 goto cleanup_info;
637
638 /* Initialize the array of DMA addresses for stream rings for the HW. */
639 stream_info->stream_ctx_array = xhci_alloc_stream_ctx(xhci,
640 num_stream_ctxs, &stream_info->ctx_array_dma,
641 mem_flags);
642 if (!stream_info->stream_ctx_array)
643 goto cleanup_ctx;
644 memset(stream_info->stream_ctx_array, 0,
645 sizeof(struct xhci_stream_ctx)*num_stream_ctxs);
646
647 /* Allocate everything needed to free the stream rings later */
648 stream_info->free_streams_command =
649 xhci_alloc_command(xhci, true, true, mem_flags);
650 if (!stream_info->free_streams_command)
651 goto cleanup_ctx;
652
653 INIT_RADIX_TREE(&stream_info->trb_address_map, GFP_ATOMIC);
654
655 /* Allocate rings for all the streams that the driver will use,
656 * and add their segment DMA addresses to the radix tree.
657 * Stream 0 is reserved.
658 */
659 for (cur_stream = 1; cur_stream < num_streams; cur_stream++) {
660 stream_info->stream_rings[cur_stream] =
661 xhci_ring_alloc(xhci, 2, 1, TYPE_STREAM, mem_flags);
662 cur_ring = stream_info->stream_rings[cur_stream];
663 if (!cur_ring)
664 goto cleanup_rings;
665 cur_ring->stream_id = cur_stream;
666 /* Set deq ptr, cycle bit, and stream context type */
667 addr = cur_ring->first_seg->dma |
668 SCT_FOR_CTX(SCT_PRI_TR) |
669 cur_ring->cycle_state;
670 stream_info->stream_ctx_array[cur_stream].stream_ring =
671 cpu_to_le64(addr);
672 xhci_dbg(xhci, "Setting stream %d ring ptr to 0x%08llx\n",
673 cur_stream, (unsigned long long) addr);
674
675 key = (unsigned long)
676 (cur_ring->first_seg->dma >> TRB_SEGMENT_SHIFT);
677 ret = radix_tree_insert(&stream_info->trb_address_map,
678 key, cur_ring);
679 if (ret) {
680 xhci_ring_free(xhci, cur_ring);
681 stream_info->stream_rings[cur_stream] = NULL;
682 goto cleanup_rings;
683 }
684 }
685 /* Leave the other unused stream ring pointers in the stream context
686 * array initialized to zero. This will cause the xHC to give us an
687 * error if the device asks for a stream ID we don't have setup (if it
688 * was any other way, the host controller would assume the ring is
689 * "empty" and wait forever for data to be queued to that stream ID).
690 */
691 #if XHCI_DEBUG
692 /* Do a little test on the radix tree to make sure it returns the
693 * correct values.
694 */
695 if (xhci_test_radix_tree(xhci, num_streams, stream_info))
696 goto cleanup_rings;
697 #endif
698
699 return stream_info;
700
701 cleanup_rings:
702 for (cur_stream = 1; cur_stream < num_streams; cur_stream++) {
703 cur_ring = stream_info->stream_rings[cur_stream];
704 if (cur_ring) {
705 addr = cur_ring->first_seg->dma;
706 radix_tree_delete(&stream_info->trb_address_map,
707 addr >> TRB_SEGMENT_SHIFT);
708 xhci_ring_free(xhci, cur_ring);
709 stream_info->stream_rings[cur_stream] = NULL;
710 }
711 }
712 xhci_free_command(xhci, stream_info->free_streams_command);
713 cleanup_ctx:
714 kfree(stream_info->stream_rings);
715 cleanup_info:
716 kfree(stream_info);
717 cleanup_trbs:
718 xhci->cmd_ring_reserved_trbs--;
719 return NULL;
720 }
721 /*
722 * Sets the MaxPStreams field and the Linear Stream Array field.
723 * Sets the dequeue pointer to the stream context array.
724 */
725 void xhci_setup_streams_ep_input_ctx(struct xhci_hcd *xhci,
726 struct xhci_ep_ctx *ep_ctx,
727 struct xhci_stream_info *stream_info)
728 {
729 u32 max_primary_streams;
730 /* MaxPStreams is the number of stream context array entries, not the
731 * number we're actually using. Must be in 2^(MaxPstreams + 1) format.
732 * fls(0) = 0, fls(0x1) = 1, fls(0x10) = 2, fls(0x100) = 3, etc.
733 */
734 max_primary_streams = fls(stream_info->num_stream_ctxs) - 2;
735 xhci_dbg(xhci, "Setting number of stream ctx array entries to %u\n",
736 1 << (max_primary_streams + 1));
737 ep_ctx->ep_info &= cpu_to_le32(~EP_MAXPSTREAMS_MASK);
738 ep_ctx->ep_info |= cpu_to_le32(EP_MAXPSTREAMS(max_primary_streams)
739 | EP_HAS_LSA);
740 ep_ctx->deq = cpu_to_le64(stream_info->ctx_array_dma);
741 }
742
743 /*
744 * Sets the MaxPStreams field and the Linear Stream Array field to 0.
745 * Reinstalls the "normal" endpoint ring (at its previous dequeue mark,
746 * not at the beginning of the ring).
747 */
748 void xhci_setup_no_streams_ep_input_ctx(struct xhci_hcd *xhci,
749 struct xhci_ep_ctx *ep_ctx,
750 struct xhci_virt_ep *ep)
751 {
752 dma_addr_t addr;
753 ep_ctx->ep_info &= cpu_to_le32(~(EP_MAXPSTREAMS_MASK | EP_HAS_LSA));
754 addr = xhci_trb_virt_to_dma(ep->ring->deq_seg, ep->ring->dequeue);
755 ep_ctx->deq = cpu_to_le64(addr | ep->ring->cycle_state);
756 }
757
758 /* Frees all stream contexts associated with the endpoint,
759 *
760 * Caller should fix the endpoint context streams fields.
761 */
762 void xhci_free_stream_info(struct xhci_hcd *xhci,
763 struct xhci_stream_info *stream_info)
764 {
765 int cur_stream;
766 struct xhci_ring *cur_ring;
767 dma_addr_t addr;
768
769 if (!stream_info)
770 return;
771
772 for (cur_stream = 1; cur_stream < stream_info->num_streams;
773 cur_stream++) {
774 cur_ring = stream_info->stream_rings[cur_stream];
775 if (cur_ring) {
776 addr = cur_ring->first_seg->dma;
777 radix_tree_delete(&stream_info->trb_address_map,
778 addr >> TRB_SEGMENT_SHIFT);
779 xhci_ring_free(xhci, cur_ring);
780 stream_info->stream_rings[cur_stream] = NULL;
781 }
782 }
783 xhci_free_command(xhci, stream_info->free_streams_command);
784 xhci->cmd_ring_reserved_trbs--;
785 if (stream_info->stream_ctx_array)
786 xhci_free_stream_ctx(xhci,
787 stream_info->num_stream_ctxs,
788 stream_info->stream_ctx_array,
789 stream_info->ctx_array_dma);
790
791 if (stream_info)
792 kfree(stream_info->stream_rings);
793 kfree(stream_info);
794 }
795
796
797 /***************** Device context manipulation *************************/
798
799 static void xhci_init_endpoint_timer(struct xhci_hcd *xhci,
800 struct xhci_virt_ep *ep)
801 {
802 init_timer(&ep->stop_cmd_timer);
803 ep->stop_cmd_timer.data = (unsigned long) ep;
804 ep->stop_cmd_timer.function = xhci_stop_endpoint_command_watchdog;
805 ep->xhci = xhci;
806 }
807
808 static void xhci_free_tt_info(struct xhci_hcd *xhci,
809 struct xhci_virt_device *virt_dev,
810 int slot_id)
811 {
812 struct list_head *tt_list_head;
813 struct xhci_tt_bw_info *tt_info, *next;
814 bool slot_found = false;
815
816 /* If the device never made it past the Set Address stage,
817 * it may not have the real_port set correctly.
818 */
819 if (virt_dev->real_port == 0 ||
820 virt_dev->real_port > HCS_MAX_PORTS(xhci->hcs_params1)) {
821 xhci_dbg(xhci, "Bad real port.\n");
822 return;
823 }
824
825 tt_list_head = &(xhci->rh_bw[virt_dev->real_port - 1].tts);
826 list_for_each_entry_safe(tt_info, next, tt_list_head, tt_list) {
827 /* Multi-TT hubs will have more than one entry */
828 if (tt_info->slot_id == slot_id) {
829 slot_found = true;
830 list_del(&tt_info->tt_list);
831 kfree(tt_info);
832 } else if (slot_found) {
833 break;
834 }
835 }
836 }
837
838 int xhci_alloc_tt_info(struct xhci_hcd *xhci,
839 struct xhci_virt_device *virt_dev,
840 struct usb_device *hdev,
841 struct usb_tt *tt, gfp_t mem_flags)
842 {
843 struct xhci_tt_bw_info *tt_info;
844 unsigned int num_ports;
845 int i, j;
846
847 if (!tt->multi)
848 num_ports = 1;
849 else
850 num_ports = hdev->maxchild;
851
852 for (i = 0; i < num_ports; i++, tt_info++) {
853 struct xhci_interval_bw_table *bw_table;
854
855 tt_info = kzalloc(sizeof(*tt_info), mem_flags);
856 if (!tt_info)
857 goto free_tts;
858 INIT_LIST_HEAD(&tt_info->tt_list);
859 list_add(&tt_info->tt_list,
860 &xhci->rh_bw[virt_dev->real_port - 1].tts);
861 tt_info->slot_id = virt_dev->udev->slot_id;
862 if (tt->multi)
863 tt_info->ttport = i+1;
864 bw_table = &tt_info->bw_table;
865 for (j = 0; j < XHCI_MAX_INTERVAL; j++)
866 INIT_LIST_HEAD(&bw_table->interval_bw[j].endpoints);
867 }
868 return 0;
869
870 free_tts:
871 xhci_free_tt_info(xhci, virt_dev, virt_dev->udev->slot_id);
872 return -ENOMEM;
873 }
874
875
876 /* All the xhci_tds in the ring's TD list should be freed at this point.
877 * Should be called with xhci->lock held if there is any chance the TT lists
878 * will be manipulated by the configure endpoint, allocate device, or update
879 * hub functions while this function is removing the TT entries from the list.
880 */
881 void xhci_free_virt_device(struct xhci_hcd *xhci, int slot_id)
882 {
883 struct xhci_virt_device *dev;
884 int i;
885 int old_active_eps = 0;
886
887 /* Slot ID 0 is reserved */
888 if (slot_id == 0 || !xhci->devs[slot_id])
889 return;
890
891 dev = xhci->devs[slot_id];
892 xhci->dcbaa->dev_context_ptrs[slot_id] = 0;
893 if (!dev)
894 return;
895
896 if (dev->tt_info)
897 old_active_eps = dev->tt_info->active_eps;
898
899 for (i = 0; i < 31; ++i) {
900 if (dev->eps[i].ring)
901 xhci_ring_free(xhci, dev->eps[i].ring);
902 if (dev->eps[i].stream_info)
903 xhci_free_stream_info(xhci,
904 dev->eps[i].stream_info);
905 /* Endpoints on the TT/root port lists should have been removed
906 * when usb_disable_device() was called for the device.
907 * We can't drop them anyway, because the udev might have gone
908 * away by this point, and we can't tell what speed it was.
909 */
910 if (!list_empty(&dev->eps[i].bw_endpoint_list))
911 xhci_warn(xhci, "Slot %u endpoint %u "
912 "not removed from BW list!\n",
913 slot_id, i);
914 }
915 /* If this is a hub, free the TT(s) from the TT list */
916 xhci_free_tt_info(xhci, dev, slot_id);
917 /* If necessary, update the number of active TTs on this root port */
918 xhci_update_tt_active_eps(xhci, dev, old_active_eps);
919
920 if (dev->ring_cache) {
921 for (i = 0; i < dev->num_rings_cached; i++)
922 xhci_ring_free(xhci, dev->ring_cache[i]);
923 kfree(dev->ring_cache);
924 }
925
926 if (dev->in_ctx)
927 xhci_free_container_ctx(xhci, dev->in_ctx);
928 if (dev->out_ctx)
929 xhci_free_container_ctx(xhci, dev->out_ctx);
930
931 kfree(xhci->devs[slot_id]);
932 xhci->devs[slot_id] = NULL;
933 }
934
935 int xhci_alloc_virt_device(struct xhci_hcd *xhci, int slot_id,
936 struct usb_device *udev, gfp_t flags)
937 {
938 struct xhci_virt_device *dev;
939 int i;
940
941 /* Slot ID 0 is reserved */
942 if (slot_id == 0 || xhci->devs[slot_id]) {
943 xhci_warn(xhci, "Bad Slot ID %d\n", slot_id);
944 return 0;
945 }
946
947 xhci->devs[slot_id] = kzalloc(sizeof(*xhci->devs[slot_id]), flags);
948 if (!xhci->devs[slot_id])
949 return 0;
950 dev = xhci->devs[slot_id];
951
952 /* Allocate the (output) device context that will be used in the HC. */
953 dev->out_ctx = xhci_alloc_container_ctx(xhci, XHCI_CTX_TYPE_DEVICE, flags);
954 if (!dev->out_ctx)
955 goto fail;
956
957 xhci_dbg(xhci, "Slot %d output ctx = 0x%llx (dma)\n", slot_id,
958 (unsigned long long)dev->out_ctx->dma);
959
960 /* Allocate the (input) device context for address device command */
961 dev->in_ctx = xhci_alloc_container_ctx(xhci, XHCI_CTX_TYPE_INPUT, flags);
962 if (!dev->in_ctx)
963 goto fail;
964
965 xhci_dbg(xhci, "Slot %d input ctx = 0x%llx (dma)\n", slot_id,
966 (unsigned long long)dev->in_ctx->dma);
967
968 /* Initialize the cancellation list and watchdog timers for each ep */
969 for (i = 0; i < 31; i++) {
970 xhci_init_endpoint_timer(xhci, &dev->eps[i]);
971 INIT_LIST_HEAD(&dev->eps[i].cancelled_td_list);
972 INIT_LIST_HEAD(&dev->eps[i].bw_endpoint_list);
973 }
974
975 /* Allocate endpoint 0 ring */
976 dev->eps[0].ring = xhci_ring_alloc(xhci, 2, 1, TYPE_CTRL, flags);
977 if (!dev->eps[0].ring)
978 goto fail;
979
980 /* Allocate pointers to the ring cache */
981 dev->ring_cache = kzalloc(
982 sizeof(struct xhci_ring *)*XHCI_MAX_RINGS_CACHED,
983 flags);
984 if (!dev->ring_cache)
985 goto fail;
986 dev->num_rings_cached = 0;
987
988 init_completion(&dev->cmd_completion);
989 INIT_LIST_HEAD(&dev->cmd_list);
990 dev->udev = udev;
991
992 /* Point to output device context in dcbaa. */
993 xhci->dcbaa->dev_context_ptrs[slot_id] = cpu_to_le64(dev->out_ctx->dma);
994 xhci_dbg(xhci, "Set slot id %d dcbaa entry %p to 0x%llx\n",
995 slot_id,
996 &xhci->dcbaa->dev_context_ptrs[slot_id],
997 le64_to_cpu(xhci->dcbaa->dev_context_ptrs[slot_id]));
998
999 return 1;
1000 fail:
1001 xhci_free_virt_device(xhci, slot_id);
1002 return 0;
1003 }
1004
1005 void xhci_copy_ep0_dequeue_into_input_ctx(struct xhci_hcd *xhci,
1006 struct usb_device *udev)
1007 {
1008 struct xhci_virt_device *virt_dev;
1009 struct xhci_ep_ctx *ep0_ctx;
1010 struct xhci_ring *ep_ring;
1011
1012 virt_dev = xhci->devs[udev->slot_id];
1013 ep0_ctx = xhci_get_ep_ctx(xhci, virt_dev->in_ctx, 0);
1014 ep_ring = virt_dev->eps[0].ring;
1015 /*
1016 * FIXME we don't keep track of the dequeue pointer very well after a
1017 * Set TR dequeue pointer, so we're setting the dequeue pointer of the
1018 * host to our enqueue pointer. This should only be called after a
1019 * configured device has reset, so all control transfers should have
1020 * been completed or cancelled before the reset.
1021 */
1022 ep0_ctx->deq = cpu_to_le64(xhci_trb_virt_to_dma(ep_ring->enq_seg,
1023 ep_ring->enqueue)
1024 | ep_ring->cycle_state);
1025 }
1026
1027 /*
1028 * The xHCI roothub may have ports of differing speeds in any order in the port
1029 * status registers. xhci->port_array provides an array of the port speed for
1030 * each offset into the port status registers.
1031 *
1032 * The xHCI hardware wants to know the roothub port number that the USB device
1033 * is attached to (or the roothub port its ancestor hub is attached to). All we
1034 * know is the index of that port under either the USB 2.0 or the USB 3.0
1035 * roothub, but that doesn't give us the real index into the HW port status
1036 * registers. Call xhci_find_raw_port_number() to get real index.
1037 */
1038 static u32 xhci_find_real_port_number(struct xhci_hcd *xhci,
1039 struct usb_device *udev)
1040 {
1041 struct usb_device *top_dev;
1042 struct usb_hcd *hcd;
1043
1044 if (udev->speed == USB_SPEED_SUPER)
1045 hcd = xhci->shared_hcd;
1046 else
1047 hcd = xhci->main_hcd;
1048
1049 for (top_dev = udev; top_dev->parent && top_dev->parent->parent;
1050 top_dev = top_dev->parent)
1051 /* Found device below root hub */;
1052
1053 return xhci_find_raw_port_number(hcd, top_dev->portnum);
1054 }
1055
1056 /* Setup an xHCI virtual device for a Set Address command */
1057 int xhci_setup_addressable_virt_dev(struct xhci_hcd *xhci, struct usb_device *udev)
1058 {
1059 struct xhci_virt_device *dev;
1060 struct xhci_ep_ctx *ep0_ctx;
1061 struct xhci_slot_ctx *slot_ctx;
1062 u32 port_num;
1063 u32 max_packets;
1064 struct usb_device *top_dev;
1065
1066 dev = xhci->devs[udev->slot_id];
1067 /* Slot ID 0 is reserved */
1068 if (udev->slot_id == 0 || !dev) {
1069 xhci_warn(xhci, "Slot ID %d is not assigned to this device\n",
1070 udev->slot_id);
1071 return -EINVAL;
1072 }
1073 ep0_ctx = xhci_get_ep_ctx(xhci, dev->in_ctx, 0);
1074 slot_ctx = xhci_get_slot_ctx(xhci, dev->in_ctx);
1075
1076 /* 3) Only the control endpoint is valid - one endpoint context */
1077 slot_ctx->dev_info |= cpu_to_le32(LAST_CTX(1) | udev->route);
1078 switch (udev->speed) {
1079 case USB_SPEED_SUPER:
1080 slot_ctx->dev_info |= cpu_to_le32(SLOT_SPEED_SS);
1081 max_packets = MAX_PACKET(512);
1082 break;
1083 case USB_SPEED_HIGH:
1084 slot_ctx->dev_info |= cpu_to_le32(SLOT_SPEED_HS);
1085 max_packets = MAX_PACKET(64);
1086 break;
1087 /* USB core guesses at a 64-byte max packet first for FS devices */
1088 case USB_SPEED_FULL:
1089 slot_ctx->dev_info |= cpu_to_le32(SLOT_SPEED_FS);
1090 max_packets = MAX_PACKET(64);
1091 break;
1092 case USB_SPEED_LOW:
1093 slot_ctx->dev_info |= cpu_to_le32(SLOT_SPEED_LS);
1094 max_packets = MAX_PACKET(8);
1095 break;
1096 case USB_SPEED_WIRELESS:
1097 xhci_dbg(xhci, "FIXME xHCI doesn't support wireless speeds\n");
1098 return -EINVAL;
1099 break;
1100 default:
1101 /* Speed was set earlier, this shouldn't happen. */
1102 return -EINVAL;
1103 }
1104 /* Find the root hub port this device is under */
1105 port_num = xhci_find_real_port_number(xhci, udev);
1106 if (!port_num)
1107 return -EINVAL;
1108 slot_ctx->dev_info2 |= cpu_to_le32(ROOT_HUB_PORT(port_num));
1109 /* Set the port number in the virtual_device to the faked port number */
1110 for (top_dev = udev; top_dev->parent && top_dev->parent->parent;
1111 top_dev = top_dev->parent)
1112 /* Found device below root hub */;
1113 dev->fake_port = top_dev->portnum;
1114 dev->real_port = port_num;
1115 xhci_dbg(xhci, "Set root hub portnum to %d\n", port_num);
1116 xhci_dbg(xhci, "Set fake root hub portnum to %d\n", dev->fake_port);
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[port_num - 1].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[port_num - 1];
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 /* Steps 7 and 8 were done in xhci_alloc_virt_device() */
1170
1171 return 0;
1172 }
1173
1174 /*
1175 * Convert interval expressed as 2^(bInterval - 1) == interval into
1176 * straight exponent value 2^n == interval.
1177 *
1178 */
1179 static unsigned int xhci_parse_exponent_interval(struct usb_device *udev,
1180 struct usb_host_endpoint *ep)
1181 {
1182 unsigned int interval;
1183
1184 interval = clamp_val(ep->desc.bInterval, 1, 16) - 1;
1185 if (interval != ep->desc.bInterval - 1)
1186 dev_warn(&udev->dev,
1187 "ep %#x - rounding interval to %d %sframes\n",
1188 ep->desc.bEndpointAddress,
1189 1 << interval,
1190 udev->speed == USB_SPEED_FULL ? "" : "micro");
1191
1192 if (udev->speed == USB_SPEED_FULL) {
1193 /*
1194 * Full speed isoc endpoints specify interval in frames,
1195 * not microframes. We are using microframes everywhere,
1196 * so adjust accordingly.
1197 */
1198 interval += 3; /* 1 frame = 2^3 uframes */
1199 }
1200
1201 return interval;
1202 }
1203
1204 /*
1205 * Convert bInterval expressed in microframes (in 1-255 range) to exponent of
1206 * microframes, rounded down to nearest power of 2.
1207 */
1208 static unsigned int xhci_microframes_to_exponent(struct usb_device *udev,
1209 struct usb_host_endpoint *ep, unsigned int desc_interval,
1210 unsigned int min_exponent, unsigned int max_exponent)
1211 {
1212 unsigned int interval;
1213
1214 interval = fls(desc_interval) - 1;
1215 interval = clamp_val(interval, min_exponent, max_exponent);
1216 if ((1 << interval) != desc_interval)
1217 dev_warn(&udev->dev,
1218 "ep %#x - rounding interval to %d microframes, ep desc says %d microframes\n",
1219 ep->desc.bEndpointAddress,
1220 1 << interval,
1221 desc_interval);
1222
1223 return interval;
1224 }
1225
1226 static unsigned int xhci_parse_microframe_interval(struct usb_device *udev,
1227 struct usb_host_endpoint *ep)
1228 {
1229 if (ep->desc.bInterval == 0)
1230 return 0;
1231 return xhci_microframes_to_exponent(udev, ep,
1232 ep->desc.bInterval, 0, 15);
1233 }
1234
1235
1236 static unsigned int xhci_parse_frame_interval(struct usb_device *udev,
1237 struct usb_host_endpoint *ep)
1238 {
1239 return xhci_microframes_to_exponent(udev, ep,
1240 ep->desc.bInterval * 8, 3, 10);
1241 }
1242
1243 /* Return the polling or NAK interval.
1244 *
1245 * The polling interval is expressed in "microframes". If xHCI's Interval field
1246 * is set to N, it will service the endpoint every 2^(Interval)*125us.
1247 *
1248 * The NAK interval is one NAK per 1 to 255 microframes, or no NAKs if interval
1249 * is set to 0.
1250 */
1251 static unsigned int xhci_get_endpoint_interval(struct usb_device *udev,
1252 struct usb_host_endpoint *ep)
1253 {
1254 unsigned int interval = 0;
1255
1256 switch (udev->speed) {
1257 case USB_SPEED_HIGH:
1258 /* Max NAK rate */
1259 if (usb_endpoint_xfer_control(&ep->desc) ||
1260 usb_endpoint_xfer_bulk(&ep->desc)) {
1261 interval = xhci_parse_microframe_interval(udev, ep);
1262 break;
1263 }
1264 /* Fall through - SS and HS isoc/int have same decoding */
1265
1266 case USB_SPEED_SUPER:
1267 if (usb_endpoint_xfer_int(&ep->desc) ||
1268 usb_endpoint_xfer_isoc(&ep->desc)) {
1269 interval = xhci_parse_exponent_interval(udev, ep);
1270 }
1271 break;
1272
1273 case USB_SPEED_FULL:
1274 if (usb_endpoint_xfer_isoc(&ep->desc)) {
1275 interval = xhci_parse_exponent_interval(udev, ep);
1276 break;
1277 }
1278 /*
1279 * Fall through for interrupt endpoint interval decoding
1280 * since it uses the same rules as low speed interrupt
1281 * endpoints.
1282 */
1283
1284 case USB_SPEED_LOW:
1285 if (usb_endpoint_xfer_int(&ep->desc) ||
1286 usb_endpoint_xfer_isoc(&ep->desc)) {
1287
1288 interval = xhci_parse_frame_interval(udev, ep);
1289 }
1290 break;
1291
1292 default:
1293 BUG();
1294 }
1295 return EP_INTERVAL(interval);
1296 }
1297
1298 /* The "Mult" field in the endpoint context is only set for SuperSpeed isoc eps.
1299 * High speed endpoint descriptors can define "the number of additional
1300 * transaction opportunities per microframe", but that goes in the Max Burst
1301 * endpoint context field.
1302 */
1303 static u32 xhci_get_endpoint_mult(struct usb_device *udev,
1304 struct usb_host_endpoint *ep)
1305 {
1306 if (udev->speed != USB_SPEED_SUPER ||
1307 !usb_endpoint_xfer_isoc(&ep->desc))
1308 return 0;
1309 return ep->ss_ep_comp.bmAttributes;
1310 }
1311
1312 static u32 xhci_get_endpoint_type(struct usb_device *udev,
1313 struct usb_host_endpoint *ep)
1314 {
1315 int in;
1316 u32 type;
1317
1318 in = usb_endpoint_dir_in(&ep->desc);
1319 if (usb_endpoint_xfer_control(&ep->desc)) {
1320 type = EP_TYPE(CTRL_EP);
1321 } else if (usb_endpoint_xfer_bulk(&ep->desc)) {
1322 if (in)
1323 type = EP_TYPE(BULK_IN_EP);
1324 else
1325 type = EP_TYPE(BULK_OUT_EP);
1326 } else if (usb_endpoint_xfer_isoc(&ep->desc)) {
1327 if (in)
1328 type = EP_TYPE(ISOC_IN_EP);
1329 else
1330 type = EP_TYPE(ISOC_OUT_EP);
1331 } else if (usb_endpoint_xfer_int(&ep->desc)) {
1332 if (in)
1333 type = EP_TYPE(INT_IN_EP);
1334 else
1335 type = EP_TYPE(INT_OUT_EP);
1336 } else {
1337 type = 0;
1338 }
1339 return type;
1340 }
1341
1342 /* Return the maximum endpoint service interval time (ESIT) payload.
1343 * Basically, this is the maxpacket size, multiplied by the burst size
1344 * and mult size.
1345 */
1346 static u32 xhci_get_max_esit_payload(struct xhci_hcd *xhci,
1347 struct usb_device *udev,
1348 struct usb_host_endpoint *ep)
1349 {
1350 int max_burst;
1351 int max_packet;
1352
1353 /* Only applies for interrupt or isochronous endpoints */
1354 if (usb_endpoint_xfer_control(&ep->desc) ||
1355 usb_endpoint_xfer_bulk(&ep->desc))
1356 return 0;
1357
1358 if (udev->speed == USB_SPEED_SUPER)
1359 return le16_to_cpu(ep->ss_ep_comp.wBytesPerInterval);
1360
1361 max_packet = GET_MAX_PACKET(usb_endpoint_maxp(&ep->desc));
1362 max_burst = (usb_endpoint_maxp(&ep->desc) & 0x1800) >> 11;
1363 /* A 0 in max burst means 1 transfer per ESIT */
1364 return max_packet * (max_burst + 1);
1365 }
1366
1367 /* Set up an endpoint with one ring segment. Do not allocate stream rings.
1368 * Drivers will have to call usb_alloc_streams() to do that.
1369 */
1370 int xhci_endpoint_init(struct xhci_hcd *xhci,
1371 struct xhci_virt_device *virt_dev,
1372 struct usb_device *udev,
1373 struct usb_host_endpoint *ep,
1374 gfp_t mem_flags)
1375 {
1376 unsigned int ep_index;
1377 struct xhci_ep_ctx *ep_ctx;
1378 struct xhci_ring *ep_ring;
1379 unsigned int max_packet;
1380 unsigned int max_burst;
1381 enum xhci_ring_type type;
1382 u32 max_esit_payload;
1383 u32 endpoint_type;
1384
1385 ep_index = xhci_get_endpoint_index(&ep->desc);
1386 ep_ctx = xhci_get_ep_ctx(xhci, virt_dev->in_ctx, ep_index);
1387
1388 endpoint_type = xhci_get_endpoint_type(udev, ep);
1389 if (!endpoint_type)
1390 return -EINVAL;
1391 ep_ctx->ep_info2 = cpu_to_le32(endpoint_type);
1392
1393 type = usb_endpoint_type(&ep->desc);
1394 /* Set up the endpoint ring */
1395 virt_dev->eps[ep_index].new_ring =
1396 xhci_ring_alloc(xhci, 2, 1, type, mem_flags);
1397 if (!virt_dev->eps[ep_index].new_ring) {
1398 /* Attempt to use the ring cache */
1399 if (virt_dev->num_rings_cached == 0)
1400 return -ENOMEM;
1401 virt_dev->eps[ep_index].new_ring =
1402 virt_dev->ring_cache[virt_dev->num_rings_cached];
1403 virt_dev->ring_cache[virt_dev->num_rings_cached] = NULL;
1404 virt_dev->num_rings_cached--;
1405 xhci_reinit_cached_ring(xhci, virt_dev->eps[ep_index].new_ring,
1406 1, type);
1407 }
1408 virt_dev->eps[ep_index].skip = false;
1409 ep_ring = virt_dev->eps[ep_index].new_ring;
1410 ep_ctx->deq = cpu_to_le64(ep_ring->first_seg->dma | ep_ring->cycle_state);
1411
1412 ep_ctx->ep_info = cpu_to_le32(xhci_get_endpoint_interval(udev, ep)
1413 | EP_MULT(xhci_get_endpoint_mult(udev, ep)));
1414
1415 /* FIXME dig Mult and streams info out of ep companion desc */
1416
1417 /* Allow 3 retries for everything but isoc;
1418 * CErr shall be set to 0 for Isoch endpoints.
1419 */
1420 if (!usb_endpoint_xfer_isoc(&ep->desc))
1421 ep_ctx->ep_info2 |= cpu_to_le32(ERROR_COUNT(3));
1422 else
1423 ep_ctx->ep_info2 |= cpu_to_le32(ERROR_COUNT(0));
1424
1425 /* Set the max packet size and max burst */
1426 max_packet = GET_MAX_PACKET(usb_endpoint_maxp(&ep->desc));
1427 max_burst = 0;
1428 switch (udev->speed) {
1429 case USB_SPEED_SUPER:
1430 /* dig out max burst from ep companion desc */
1431 max_burst = ep->ss_ep_comp.bMaxBurst;
1432 break;
1433 case USB_SPEED_HIGH:
1434 /* Some devices get this wrong */
1435 if (usb_endpoint_xfer_bulk(&ep->desc))
1436 max_packet = 512;
1437 /* bits 11:12 specify the number of additional transaction
1438 * opportunities per microframe (USB 2.0, section 9.6.6)
1439 */
1440 if (usb_endpoint_xfer_isoc(&ep->desc) ||
1441 usb_endpoint_xfer_int(&ep->desc)) {
1442 max_burst = (usb_endpoint_maxp(&ep->desc)
1443 & 0x1800) >> 11;
1444 }
1445 break;
1446 case USB_SPEED_FULL:
1447 case USB_SPEED_LOW:
1448 break;
1449 default:
1450 BUG();
1451 }
1452 ep_ctx->ep_info2 |= cpu_to_le32(MAX_PACKET(max_packet) |
1453 MAX_BURST(max_burst));
1454 max_esit_payload = xhci_get_max_esit_payload(xhci, udev, ep);
1455 ep_ctx->tx_info = cpu_to_le32(MAX_ESIT_PAYLOAD_FOR_EP(max_esit_payload));
1456
1457 /*
1458 * XXX no idea how to calculate the average TRB buffer length for bulk
1459 * endpoints, as the driver gives us no clue how big each scatter gather
1460 * list entry (or buffer) is going to be.
1461 *
1462 * For isochronous and interrupt endpoints, we set it to the max
1463 * available, until we have new API in the USB core to allow drivers to
1464 * declare how much bandwidth they actually need.
1465 *
1466 * Normally, it would be calculated by taking the total of the buffer
1467 * lengths in the TD and then dividing by the number of TRBs in a TD,
1468 * including link TRBs, No-op TRBs, and Event data TRBs. Since we don't
1469 * use Event Data TRBs, and we don't chain in a link TRB on short
1470 * transfers, we're basically dividing by 1.
1471 *
1472 * xHCI 1.0 specification indicates that the Average TRB Length should
1473 * be set to 8 for control endpoints.
1474 */
1475 if (usb_endpoint_xfer_control(&ep->desc) && xhci->hci_version == 0x100)
1476 ep_ctx->tx_info |= cpu_to_le32(AVG_TRB_LENGTH_FOR_EP(8));
1477 else
1478 ep_ctx->tx_info |=
1479 cpu_to_le32(AVG_TRB_LENGTH_FOR_EP(max_esit_payload));
1480
1481 /* FIXME Debug endpoint context */
1482 return 0;
1483 }
1484
1485 void xhci_endpoint_zero(struct xhci_hcd *xhci,
1486 struct xhci_virt_device *virt_dev,
1487 struct usb_host_endpoint *ep)
1488 {
1489 unsigned int ep_index;
1490 struct xhci_ep_ctx *ep_ctx;
1491
1492 ep_index = xhci_get_endpoint_index(&ep->desc);
1493 ep_ctx = xhci_get_ep_ctx(xhci, virt_dev->in_ctx, ep_index);
1494
1495 ep_ctx->ep_info = 0;
1496 ep_ctx->ep_info2 = 0;
1497 ep_ctx->deq = 0;
1498 ep_ctx->tx_info = 0;
1499 /* Don't free the endpoint ring until the set interface or configuration
1500 * request succeeds.
1501 */
1502 }
1503
1504 void xhci_clear_endpoint_bw_info(struct xhci_bw_info *bw_info)
1505 {
1506 bw_info->ep_interval = 0;
1507 bw_info->mult = 0;
1508 bw_info->num_packets = 0;
1509 bw_info->max_packet_size = 0;
1510 bw_info->type = 0;
1511 bw_info->max_esit_payload = 0;
1512 }
1513
1514 void xhci_update_bw_info(struct xhci_hcd *xhci,
1515 struct xhci_container_ctx *in_ctx,
1516 struct xhci_input_control_ctx *ctrl_ctx,
1517 struct xhci_virt_device *virt_dev)
1518 {
1519 struct xhci_bw_info *bw_info;
1520 struct xhci_ep_ctx *ep_ctx;
1521 unsigned int ep_type;
1522 int i;
1523
1524 for (i = 1; i < 31; ++i) {
1525 bw_info = &virt_dev->eps[i].bw_info;
1526
1527 /* We can't tell what endpoint type is being dropped, but
1528 * unconditionally clearing the bandwidth info for non-periodic
1529 * endpoints should be harmless because the info will never be
1530 * set in the first place.
1531 */
1532 if (!EP_IS_ADDED(ctrl_ctx, i) && EP_IS_DROPPED(ctrl_ctx, i)) {
1533 /* Dropped endpoint */
1534 xhci_clear_endpoint_bw_info(bw_info);
1535 continue;
1536 }
1537
1538 if (EP_IS_ADDED(ctrl_ctx, i)) {
1539 ep_ctx = xhci_get_ep_ctx(xhci, in_ctx, i);
1540 ep_type = CTX_TO_EP_TYPE(le32_to_cpu(ep_ctx->ep_info2));
1541
1542 /* Ignore non-periodic endpoints */
1543 if (ep_type != ISOC_OUT_EP && ep_type != INT_OUT_EP &&
1544 ep_type != ISOC_IN_EP &&
1545 ep_type != INT_IN_EP)
1546 continue;
1547
1548 /* Added or changed endpoint */
1549 bw_info->ep_interval = CTX_TO_EP_INTERVAL(
1550 le32_to_cpu(ep_ctx->ep_info));
1551 /* Number of packets and mult are zero-based in the
1552 * input context, but we want one-based for the
1553 * interval table.
1554 */
1555 bw_info->mult = CTX_TO_EP_MULT(
1556 le32_to_cpu(ep_ctx->ep_info)) + 1;
1557 bw_info->num_packets = CTX_TO_MAX_BURST(
1558 le32_to_cpu(ep_ctx->ep_info2)) + 1;
1559 bw_info->max_packet_size = MAX_PACKET_DECODED(
1560 le32_to_cpu(ep_ctx->ep_info2));
1561 bw_info->type = ep_type;
1562 bw_info->max_esit_payload = CTX_TO_MAX_ESIT_PAYLOAD(
1563 le32_to_cpu(ep_ctx->tx_info));
1564 }
1565 }
1566 }
1567
1568 /* Copy output xhci_ep_ctx to the input xhci_ep_ctx copy.
1569 * Useful when you want to change one particular aspect of the endpoint and then
1570 * issue a configure endpoint command.
1571 */
1572 void xhci_endpoint_copy(struct xhci_hcd *xhci,
1573 struct xhci_container_ctx *in_ctx,
1574 struct xhci_container_ctx *out_ctx,
1575 unsigned int ep_index)
1576 {
1577 struct xhci_ep_ctx *out_ep_ctx;
1578 struct xhci_ep_ctx *in_ep_ctx;
1579
1580 out_ep_ctx = xhci_get_ep_ctx(xhci, out_ctx, ep_index);
1581 in_ep_ctx = xhci_get_ep_ctx(xhci, in_ctx, ep_index);
1582
1583 in_ep_ctx->ep_info = out_ep_ctx->ep_info;
1584 in_ep_ctx->ep_info2 = out_ep_ctx->ep_info2;
1585 in_ep_ctx->deq = out_ep_ctx->deq;
1586 in_ep_ctx->tx_info = out_ep_ctx->tx_info;
1587 }
1588
1589 /* Copy output xhci_slot_ctx to the input xhci_slot_ctx.
1590 * Useful when you want to change one particular aspect of the endpoint and then
1591 * issue a configure endpoint command. Only the context entries field matters,
1592 * but we'll copy the whole thing anyway.
1593 */
1594 void xhci_slot_copy(struct xhci_hcd *xhci,
1595 struct xhci_container_ctx *in_ctx,
1596 struct xhci_container_ctx *out_ctx)
1597 {
1598 struct xhci_slot_ctx *in_slot_ctx;
1599 struct xhci_slot_ctx *out_slot_ctx;
1600
1601 in_slot_ctx = xhci_get_slot_ctx(xhci, in_ctx);
1602 out_slot_ctx = xhci_get_slot_ctx(xhci, out_ctx);
1603
1604 in_slot_ctx->dev_info = out_slot_ctx->dev_info;
1605 in_slot_ctx->dev_info2 = out_slot_ctx->dev_info2;
1606 in_slot_ctx->tt_info = out_slot_ctx->tt_info;
1607 in_slot_ctx->dev_state = out_slot_ctx->dev_state;
1608 }
1609
1610 /* Set up the scratchpad buffer array and scratchpad buffers, if needed. */
1611 static int scratchpad_alloc(struct xhci_hcd *xhci, gfp_t flags)
1612 {
1613 int i;
1614 struct device *dev = xhci_to_hcd(xhci)->self.controller;
1615 int num_sp = HCS_MAX_SCRATCHPAD(xhci->hcs_params2);
1616
1617 xhci_dbg(xhci, "Allocating %d scratchpad buffers\n", num_sp);
1618
1619 if (!num_sp)
1620 return 0;
1621
1622 xhci->scratchpad = kzalloc(sizeof(*xhci->scratchpad), flags);
1623 if (!xhci->scratchpad)
1624 goto fail_sp;
1625
1626 xhci->scratchpad->sp_array = dma_alloc_coherent(dev,
1627 num_sp * sizeof(u64),
1628 &xhci->scratchpad->sp_dma, flags);
1629 if (!xhci->scratchpad->sp_array)
1630 goto fail_sp2;
1631
1632 xhci->scratchpad->sp_buffers = kzalloc(sizeof(void *) * num_sp, flags);
1633 if (!xhci->scratchpad->sp_buffers)
1634 goto fail_sp3;
1635
1636 xhci->scratchpad->sp_dma_buffers =
1637 kzalloc(sizeof(dma_addr_t) * num_sp, flags);
1638
1639 if (!xhci->scratchpad->sp_dma_buffers)
1640 goto fail_sp4;
1641
1642 xhci->dcbaa->dev_context_ptrs[0] = cpu_to_le64(xhci->scratchpad->sp_dma);
1643 for (i = 0; i < num_sp; i++) {
1644 dma_addr_t dma;
1645 void *buf = dma_alloc_coherent(dev, xhci->page_size, &dma,
1646 flags);
1647 if (!buf)
1648 goto fail_sp5;
1649
1650 xhci->scratchpad->sp_array[i] = dma;
1651 xhci->scratchpad->sp_buffers[i] = buf;
1652 xhci->scratchpad->sp_dma_buffers[i] = dma;
1653 }
1654
1655 return 0;
1656
1657 fail_sp5:
1658 for (i = i - 1; i >= 0; i--) {
1659 dma_free_coherent(dev, xhci->page_size,
1660 xhci->scratchpad->sp_buffers[i],
1661 xhci->scratchpad->sp_dma_buffers[i]);
1662 }
1663 kfree(xhci->scratchpad->sp_dma_buffers);
1664
1665 fail_sp4:
1666 kfree(xhci->scratchpad->sp_buffers);
1667
1668 fail_sp3:
1669 dma_free_coherent(dev, num_sp * sizeof(u64),
1670 xhci->scratchpad->sp_array,
1671 xhci->scratchpad->sp_dma);
1672
1673 fail_sp2:
1674 kfree(xhci->scratchpad);
1675 xhci->scratchpad = NULL;
1676
1677 fail_sp:
1678 return -ENOMEM;
1679 }
1680
1681 static void scratchpad_free(struct xhci_hcd *xhci)
1682 {
1683 int num_sp;
1684 int i;
1685 struct pci_dev *pdev = to_pci_dev(xhci_to_hcd(xhci)->self.controller);
1686
1687 if (!xhci->scratchpad)
1688 return;
1689
1690 num_sp = HCS_MAX_SCRATCHPAD(xhci->hcs_params2);
1691
1692 for (i = 0; i < num_sp; i++) {
1693 dma_free_coherent(&pdev->dev, xhci->page_size,
1694 xhci->scratchpad->sp_buffers[i],
1695 xhci->scratchpad->sp_dma_buffers[i]);
1696 }
1697 kfree(xhci->scratchpad->sp_dma_buffers);
1698 kfree(xhci->scratchpad->sp_buffers);
1699 dma_free_coherent(&pdev->dev, num_sp * sizeof(u64),
1700 xhci->scratchpad->sp_array,
1701 xhci->scratchpad->sp_dma);
1702 kfree(xhci->scratchpad);
1703 xhci->scratchpad = NULL;
1704 }
1705
1706 struct xhci_command *xhci_alloc_command(struct xhci_hcd *xhci,
1707 bool allocate_in_ctx, bool allocate_completion,
1708 gfp_t mem_flags)
1709 {
1710 struct xhci_command *command;
1711
1712 command = kzalloc(sizeof(*command), mem_flags);
1713 if (!command)
1714 return NULL;
1715
1716 if (allocate_in_ctx) {
1717 command->in_ctx =
1718 xhci_alloc_container_ctx(xhci, XHCI_CTX_TYPE_INPUT,
1719 mem_flags);
1720 if (!command->in_ctx) {
1721 kfree(command);
1722 return NULL;
1723 }
1724 }
1725
1726 if (allocate_completion) {
1727 command->completion =
1728 kzalloc(sizeof(struct completion), mem_flags);
1729 if (!command->completion) {
1730 xhci_free_container_ctx(xhci, command->in_ctx);
1731 kfree(command);
1732 return NULL;
1733 }
1734 init_completion(command->completion);
1735 }
1736
1737 command->status = 0;
1738 INIT_LIST_HEAD(&command->cmd_list);
1739 return command;
1740 }
1741
1742 void xhci_urb_free_priv(struct xhci_hcd *xhci, struct urb_priv *urb_priv)
1743 {
1744 if (urb_priv) {
1745 kfree(urb_priv->td[0]);
1746 kfree(urb_priv);
1747 }
1748 }
1749
1750 void xhci_free_command(struct xhci_hcd *xhci,
1751 struct xhci_command *command)
1752 {
1753 xhci_free_container_ctx(xhci,
1754 command->in_ctx);
1755 kfree(command->completion);
1756 kfree(command);
1757 }
1758
1759 void xhci_mem_cleanup(struct xhci_hcd *xhci)
1760 {
1761 struct pci_dev *pdev = to_pci_dev(xhci_to_hcd(xhci)->self.controller);
1762 struct dev_info *dev_info, *next;
1763 struct xhci_cd *cur_cd, *next_cd;
1764 unsigned long flags;
1765 int size;
1766 int i, j, num_ports;
1767
1768 /* Free the Event Ring Segment Table and the actual Event Ring */
1769 size = sizeof(struct xhci_erst_entry)*(xhci->erst.num_entries);
1770 if (xhci->erst.entries)
1771 dma_free_coherent(&pdev->dev, size,
1772 xhci->erst.entries, xhci->erst.erst_dma_addr);
1773 xhci->erst.entries = NULL;
1774 xhci_dbg(xhci, "Freed ERST\n");
1775 if (xhci->event_ring)
1776 xhci_ring_free(xhci, xhci->event_ring);
1777 xhci->event_ring = NULL;
1778 xhci_dbg(xhci, "Freed event ring\n");
1779
1780 if (xhci->lpm_command)
1781 xhci_free_command(xhci, xhci->lpm_command);
1782 xhci->cmd_ring_reserved_trbs = 0;
1783 if (xhci->cmd_ring)
1784 xhci_ring_free(xhci, xhci->cmd_ring);
1785 xhci->cmd_ring = NULL;
1786 xhci_dbg(xhci, "Freed command ring\n");
1787 list_for_each_entry_safe(cur_cd, next_cd,
1788 &xhci->cancel_cmd_list, cancel_cmd_list) {
1789 list_del(&cur_cd->cancel_cmd_list);
1790 kfree(cur_cd);
1791 }
1792
1793 for (i = 1; i < MAX_HC_SLOTS; ++i)
1794 xhci_free_virt_device(xhci, i);
1795
1796 if (xhci->segment_pool)
1797 dma_pool_destroy(xhci->segment_pool);
1798 xhci->segment_pool = NULL;
1799 xhci_dbg(xhci, "Freed segment pool\n");
1800
1801 if (xhci->device_pool)
1802 dma_pool_destroy(xhci->device_pool);
1803 xhci->device_pool = NULL;
1804 xhci_dbg(xhci, "Freed device context pool\n");
1805
1806 if (xhci->small_streams_pool)
1807 dma_pool_destroy(xhci->small_streams_pool);
1808 xhci->small_streams_pool = NULL;
1809 xhci_dbg(xhci, "Freed small stream array pool\n");
1810
1811 if (xhci->medium_streams_pool)
1812 dma_pool_destroy(xhci->medium_streams_pool);
1813 xhci->medium_streams_pool = NULL;
1814 xhci_dbg(xhci, "Freed medium stream array pool\n");
1815
1816 if (xhci->dcbaa)
1817 dma_free_coherent(&pdev->dev, sizeof(*xhci->dcbaa),
1818 xhci->dcbaa, xhci->dcbaa->dma);
1819 xhci->dcbaa = NULL;
1820
1821 scratchpad_free(xhci);
1822
1823 spin_lock_irqsave(&xhci->lock, flags);
1824 list_for_each_entry_safe(dev_info, next, &xhci->lpm_failed_devs, list) {
1825 list_del(&dev_info->list);
1826 kfree(dev_info);
1827 }
1828 spin_unlock_irqrestore(&xhci->lock, flags);
1829
1830 if (!xhci->rh_bw)
1831 goto no_bw;
1832
1833 num_ports = HCS_MAX_PORTS(xhci->hcs_params1);
1834 for (i = 0; i < num_ports; i++) {
1835 struct xhci_interval_bw_table *bwt = &xhci->rh_bw[i].bw_table;
1836 for (j = 0; j < XHCI_MAX_INTERVAL; j++) {
1837 struct list_head *ep = &bwt->interval_bw[j].endpoints;
1838 while (!list_empty(ep))
1839 list_del_init(ep->next);
1840 }
1841 }
1842
1843 for (i = 0; i < num_ports; i++) {
1844 struct xhci_tt_bw_info *tt, *n;
1845 list_for_each_entry_safe(tt, n, &xhci->rh_bw[i].tts, tt_list) {
1846 list_del(&tt->tt_list);
1847 kfree(tt);
1848 }
1849 }
1850
1851 no_bw:
1852 xhci->num_usb2_ports = 0;
1853 xhci->num_usb3_ports = 0;
1854 xhci->num_active_eps = 0;
1855 kfree(xhci->usb2_ports);
1856 kfree(xhci->usb3_ports);
1857 kfree(xhci->port_array);
1858 kfree(xhci->rh_bw);
1859 kfree(xhci->ext_caps);
1860
1861 xhci->page_size = 0;
1862 xhci->page_shift = 0;
1863 xhci->bus_state[0].bus_suspended = 0;
1864 xhci->bus_state[1].bus_suspended = 0;
1865 }
1866
1867 static int xhci_test_trb_in_td(struct xhci_hcd *xhci,
1868 struct xhci_segment *input_seg,
1869 union xhci_trb *start_trb,
1870 union xhci_trb *end_trb,
1871 dma_addr_t input_dma,
1872 struct xhci_segment *result_seg,
1873 char *test_name, int test_number)
1874 {
1875 unsigned long long start_dma;
1876 unsigned long long end_dma;
1877 struct xhci_segment *seg;
1878
1879 start_dma = xhci_trb_virt_to_dma(input_seg, start_trb);
1880 end_dma = xhci_trb_virt_to_dma(input_seg, end_trb);
1881
1882 seg = trb_in_td(input_seg, start_trb, end_trb, input_dma);
1883 if (seg != result_seg) {
1884 xhci_warn(xhci, "WARN: %s TRB math test %d failed!\n",
1885 test_name, test_number);
1886 xhci_warn(xhci, "Tested TRB math w/ seg %p and "
1887 "input DMA 0x%llx\n",
1888 input_seg,
1889 (unsigned long long) input_dma);
1890 xhci_warn(xhci, "starting TRB %p (0x%llx DMA), "
1891 "ending TRB %p (0x%llx DMA)\n",
1892 start_trb, start_dma,
1893 end_trb, end_dma);
1894 xhci_warn(xhci, "Expected seg %p, got seg %p\n",
1895 result_seg, seg);
1896 return -1;
1897 }
1898 return 0;
1899 }
1900
1901 /* TRB math checks for xhci_trb_in_td(), using the command and event rings. */
1902 static int xhci_check_trb_in_td_math(struct xhci_hcd *xhci, gfp_t mem_flags)
1903 {
1904 struct {
1905 dma_addr_t input_dma;
1906 struct xhci_segment *result_seg;
1907 } simple_test_vector [] = {
1908 /* A zeroed DMA field should fail */
1909 { 0, NULL },
1910 /* One TRB before the ring start should fail */
1911 { xhci->event_ring->first_seg->dma - 16, NULL },
1912 /* One byte before the ring start should fail */
1913 { xhci->event_ring->first_seg->dma - 1, NULL },
1914 /* Starting TRB should succeed */
1915 { xhci->event_ring->first_seg->dma, xhci->event_ring->first_seg },
1916 /* Ending TRB should succeed */
1917 { xhci->event_ring->first_seg->dma + (TRBS_PER_SEGMENT - 1)*16,
1918 xhci->event_ring->first_seg },
1919 /* One byte after the ring end should fail */
1920 { xhci->event_ring->first_seg->dma + (TRBS_PER_SEGMENT - 1)*16 + 1, NULL },
1921 /* One TRB after the ring end should fail */
1922 { xhci->event_ring->first_seg->dma + (TRBS_PER_SEGMENT)*16, NULL },
1923 /* An address of all ones should fail */
1924 { (dma_addr_t) (~0), NULL },
1925 };
1926 struct {
1927 struct xhci_segment *input_seg;
1928 union xhci_trb *start_trb;
1929 union xhci_trb *end_trb;
1930 dma_addr_t input_dma;
1931 struct xhci_segment *result_seg;
1932 } complex_test_vector [] = {
1933 /* Test feeding a valid DMA address from a different ring */
1934 { .input_seg = xhci->event_ring->first_seg,
1935 .start_trb = xhci->event_ring->first_seg->trbs,
1936 .end_trb = &xhci->event_ring->first_seg->trbs[TRBS_PER_SEGMENT - 1],
1937 .input_dma = xhci->cmd_ring->first_seg->dma,
1938 .result_seg = NULL,
1939 },
1940 /* Test feeding a valid end TRB from a different ring */
1941 { .input_seg = xhci->event_ring->first_seg,
1942 .start_trb = xhci->event_ring->first_seg->trbs,
1943 .end_trb = &xhci->cmd_ring->first_seg->trbs[TRBS_PER_SEGMENT - 1],
1944 .input_dma = xhci->cmd_ring->first_seg->dma,
1945 .result_seg = NULL,
1946 },
1947 /* Test feeding a valid start and end TRB from a different ring */
1948 { .input_seg = xhci->event_ring->first_seg,
1949 .start_trb = xhci->cmd_ring->first_seg->trbs,
1950 .end_trb = &xhci->cmd_ring->first_seg->trbs[TRBS_PER_SEGMENT - 1],
1951 .input_dma = xhci->cmd_ring->first_seg->dma,
1952 .result_seg = NULL,
1953 },
1954 /* TRB in this ring, but after this TD */
1955 { .input_seg = xhci->event_ring->first_seg,
1956 .start_trb = &xhci->event_ring->first_seg->trbs[0],
1957 .end_trb = &xhci->event_ring->first_seg->trbs[3],
1958 .input_dma = xhci->event_ring->first_seg->dma + 4*16,
1959 .result_seg = NULL,
1960 },
1961 /* TRB in this ring, but before this TD */
1962 { .input_seg = xhci->event_ring->first_seg,
1963 .start_trb = &xhci->event_ring->first_seg->trbs[3],
1964 .end_trb = &xhci->event_ring->first_seg->trbs[6],
1965 .input_dma = xhci->event_ring->first_seg->dma + 2*16,
1966 .result_seg = NULL,
1967 },
1968 /* TRB in this ring, but after this wrapped TD */
1969 { .input_seg = xhci->event_ring->first_seg,
1970 .start_trb = &xhci->event_ring->first_seg->trbs[TRBS_PER_SEGMENT - 3],
1971 .end_trb = &xhci->event_ring->first_seg->trbs[1],
1972 .input_dma = xhci->event_ring->first_seg->dma + 2*16,
1973 .result_seg = NULL,
1974 },
1975 /* TRB in this ring, but before this wrapped TD */
1976 { .input_seg = xhci->event_ring->first_seg,
1977 .start_trb = &xhci->event_ring->first_seg->trbs[TRBS_PER_SEGMENT - 3],
1978 .end_trb = &xhci->event_ring->first_seg->trbs[1],
1979 .input_dma = xhci->event_ring->first_seg->dma + (TRBS_PER_SEGMENT - 4)*16,
1980 .result_seg = NULL,
1981 },
1982 /* TRB not in this ring, and we have a wrapped TD */
1983 { .input_seg = xhci->event_ring->first_seg,
1984 .start_trb = &xhci->event_ring->first_seg->trbs[TRBS_PER_SEGMENT - 3],
1985 .end_trb = &xhci->event_ring->first_seg->trbs[1],
1986 .input_dma = xhci->cmd_ring->first_seg->dma + 2*16,
1987 .result_seg = NULL,
1988 },
1989 };
1990
1991 unsigned int num_tests;
1992 int i, ret;
1993
1994 num_tests = ARRAY_SIZE(simple_test_vector);
1995 for (i = 0; i < num_tests; i++) {
1996 ret = xhci_test_trb_in_td(xhci,
1997 xhci->event_ring->first_seg,
1998 xhci->event_ring->first_seg->trbs,
1999 &xhci->event_ring->first_seg->trbs[TRBS_PER_SEGMENT - 1],
2000 simple_test_vector[i].input_dma,
2001 simple_test_vector[i].result_seg,
2002 "Simple", i);
2003 if (ret < 0)
2004 return ret;
2005 }
2006
2007 num_tests = ARRAY_SIZE(complex_test_vector);
2008 for (i = 0; i < num_tests; i++) {
2009 ret = xhci_test_trb_in_td(xhci,
2010 complex_test_vector[i].input_seg,
2011 complex_test_vector[i].start_trb,
2012 complex_test_vector[i].end_trb,
2013 complex_test_vector[i].input_dma,
2014 complex_test_vector[i].result_seg,
2015 "Complex", i);
2016 if (ret < 0)
2017 return ret;
2018 }
2019 xhci_dbg(xhci, "TRB math tests passed.\n");
2020 return 0;
2021 }
2022
2023 static void xhci_set_hc_event_deq(struct xhci_hcd *xhci)
2024 {
2025 u64 temp;
2026 dma_addr_t deq;
2027
2028 deq = xhci_trb_virt_to_dma(xhci->event_ring->deq_seg,
2029 xhci->event_ring->dequeue);
2030 if (deq == 0 && !in_interrupt())
2031 xhci_warn(xhci, "WARN something wrong with SW event ring "
2032 "dequeue ptr.\n");
2033 /* Update HC event ring dequeue pointer */
2034 temp = xhci_read_64(xhci, &xhci->ir_set->erst_dequeue);
2035 temp &= ERST_PTR_MASK;
2036 /* Don't clear the EHB bit (which is RW1C) because
2037 * there might be more events to service.
2038 */
2039 temp &= ~ERST_EHB;
2040 xhci_dbg(xhci, "// Write event ring dequeue pointer, "
2041 "preserving EHB bit\n");
2042 xhci_write_64(xhci, ((u64) deq & (u64) ~ERST_PTR_MASK) | temp,
2043 &xhci->ir_set->erst_dequeue);
2044 }
2045
2046 static void xhci_add_in_port(struct xhci_hcd *xhci, unsigned int num_ports,
2047 __le32 __iomem *addr, u8 major_revision, int max_caps)
2048 {
2049 u32 temp, port_offset, port_count;
2050 int i;
2051
2052 if (major_revision > 0x03) {
2053 xhci_warn(xhci, "Ignoring unknown port speed, "
2054 "Ext Cap %p, revision = 0x%x\n",
2055 addr, major_revision);
2056 /* Ignoring port protocol we can't understand. FIXME */
2057 return;
2058 }
2059
2060 /* Port offset and count in the third dword, see section 7.2 */
2061 temp = xhci_readl(xhci, addr + 2);
2062 port_offset = XHCI_EXT_PORT_OFF(temp);
2063 port_count = XHCI_EXT_PORT_COUNT(temp);
2064 xhci_dbg(xhci, "Ext Cap %p, port offset = %u, "
2065 "count = %u, revision = 0x%x\n",
2066 addr, port_offset, port_count, major_revision);
2067 /* Port count includes the current port offset */
2068 if (port_offset == 0 || (port_offset + port_count - 1) > num_ports)
2069 /* WTF? "Valid values are ‘1’ to MaxPorts" */
2070 return;
2071
2072 /* cache usb2 port capabilities */
2073 if (major_revision < 0x03 && xhci->num_ext_caps < max_caps)
2074 xhci->ext_caps[xhci->num_ext_caps++] = temp;
2075
2076 /* Check the host's USB2 LPM capability */
2077 if ((xhci->hci_version == 0x96) && (major_revision != 0x03) &&
2078 (temp & XHCI_L1C)) {
2079 xhci_dbg(xhci, "xHCI 0.96: support USB2 software lpm\n");
2080 xhci->sw_lpm_support = 1;
2081 }
2082
2083 if ((xhci->hci_version >= 0x100) && (major_revision != 0x03)) {
2084 xhci_dbg(xhci, "xHCI 1.0: support USB2 software lpm\n");
2085 xhci->sw_lpm_support = 1;
2086 if (temp & XHCI_HLC) {
2087 xhci_dbg(xhci, "xHCI 1.0: support USB2 hardware lpm\n");
2088 xhci->hw_lpm_support = 1;
2089 }
2090 }
2091
2092 port_offset--;
2093 for (i = port_offset; i < (port_offset + port_count); i++) {
2094 /* Duplicate entry. Ignore the port if the revisions differ. */
2095 if (xhci->port_array[i] != 0) {
2096 xhci_warn(xhci, "Duplicate port entry, Ext Cap %p,"
2097 " port %u\n", addr, i);
2098 xhci_warn(xhci, "Port was marked as USB %u, "
2099 "duplicated as USB %u\n",
2100 xhci->port_array[i], major_revision);
2101 /* Only adjust the roothub port counts if we haven't
2102 * found a similar duplicate.
2103 */
2104 if (xhci->port_array[i] != major_revision &&
2105 xhci->port_array[i] != DUPLICATE_ENTRY) {
2106 if (xhci->port_array[i] == 0x03)
2107 xhci->num_usb3_ports--;
2108 else
2109 xhci->num_usb2_ports--;
2110 xhci->port_array[i] = DUPLICATE_ENTRY;
2111 }
2112 /* FIXME: Should we disable the port? */
2113 continue;
2114 }
2115 xhci->port_array[i] = major_revision;
2116 if (major_revision == 0x03)
2117 xhci->num_usb3_ports++;
2118 else
2119 xhci->num_usb2_ports++;
2120 }
2121 /* FIXME: Should we disable ports not in the Extended Capabilities? */
2122 }
2123
2124 /*
2125 * Scan the Extended Capabilities for the "Supported Protocol Capabilities" that
2126 * specify what speeds each port is supposed to be. We can't count on the port
2127 * speed bits in the PORTSC register being correct until a device is connected,
2128 * but we need to set up the two fake roothubs with the correct number of USB
2129 * 3.0 and USB 2.0 ports at host controller initialization time.
2130 */
2131 static int xhci_setup_port_arrays(struct xhci_hcd *xhci, gfp_t flags)
2132 {
2133 __le32 __iomem *addr, *tmp_addr;
2134 u32 offset, tmp_offset;
2135 unsigned int num_ports;
2136 int i, j, port_index;
2137 int cap_count = 0;
2138
2139 addr = &xhci->cap_regs->hcc_params;
2140 offset = XHCI_HCC_EXT_CAPS(xhci_readl(xhci, addr));
2141 if (offset == 0) {
2142 xhci_err(xhci, "No Extended Capability registers, "
2143 "unable to set up roothub.\n");
2144 return -ENODEV;
2145 }
2146
2147 num_ports = HCS_MAX_PORTS(xhci->hcs_params1);
2148 xhci->port_array = kzalloc(sizeof(*xhci->port_array)*num_ports, flags);
2149 if (!xhci->port_array)
2150 return -ENOMEM;
2151
2152 xhci->rh_bw = kzalloc(sizeof(*xhci->rh_bw)*num_ports, flags);
2153 if (!xhci->rh_bw)
2154 return -ENOMEM;
2155 for (i = 0; i < num_ports; i++) {
2156 struct xhci_interval_bw_table *bw_table;
2157
2158 INIT_LIST_HEAD(&xhci->rh_bw[i].tts);
2159 bw_table = &xhci->rh_bw[i].bw_table;
2160 for (j = 0; j < XHCI_MAX_INTERVAL; j++)
2161 INIT_LIST_HEAD(&bw_table->interval_bw[j].endpoints);
2162 }
2163
2164 /*
2165 * For whatever reason, the first capability offset is from the
2166 * capability register base, not from the HCCPARAMS register.
2167 * See section 5.3.6 for offset calculation.
2168 */
2169 addr = &xhci->cap_regs->hc_capbase + offset;
2170
2171 tmp_addr = addr;
2172 tmp_offset = offset;
2173
2174 /* count extended protocol capability entries for later caching */
2175 do {
2176 u32 cap_id;
2177 cap_id = xhci_readl(xhci, tmp_addr);
2178 if (XHCI_EXT_CAPS_ID(cap_id) == XHCI_EXT_CAPS_PROTOCOL)
2179 cap_count++;
2180 tmp_offset = XHCI_EXT_CAPS_NEXT(cap_id);
2181 tmp_addr += tmp_offset;
2182 } while (tmp_offset);
2183
2184 xhci->ext_caps = kzalloc(sizeof(*xhci->ext_caps) * cap_count, flags);
2185 if (!xhci->ext_caps)
2186 return -ENOMEM;
2187
2188 while (1) {
2189 u32 cap_id;
2190
2191 cap_id = xhci_readl(xhci, addr);
2192 if (XHCI_EXT_CAPS_ID(cap_id) == XHCI_EXT_CAPS_PROTOCOL)
2193 xhci_add_in_port(xhci, num_ports, addr,
2194 (u8) XHCI_EXT_PORT_MAJOR(cap_id),
2195 cap_count);
2196 offset = XHCI_EXT_CAPS_NEXT(cap_id);
2197 if (!offset || (xhci->num_usb2_ports + xhci->num_usb3_ports)
2198 == num_ports)
2199 break;
2200 /*
2201 * Once you're into the Extended Capabilities, the offset is
2202 * always relative to the register holding the offset.
2203 */
2204 addr += offset;
2205 }
2206
2207 if (xhci->num_usb2_ports == 0 && xhci->num_usb3_ports == 0) {
2208 xhci_warn(xhci, "No ports on the roothubs?\n");
2209 return -ENODEV;
2210 }
2211 xhci_dbg(xhci, "Found %u USB 2.0 ports and %u USB 3.0 ports.\n",
2212 xhci->num_usb2_ports, xhci->num_usb3_ports);
2213
2214 /* Place limits on the number of roothub ports so that the hub
2215 * descriptors aren't longer than the USB core will allocate.
2216 */
2217 if (xhci->num_usb3_ports > 15) {
2218 xhci_dbg(xhci, "Limiting USB 3.0 roothub ports to 15.\n");
2219 xhci->num_usb3_ports = 15;
2220 }
2221 if (xhci->num_usb2_ports > USB_MAXCHILDREN) {
2222 xhci_dbg(xhci, "Limiting USB 2.0 roothub ports to %u.\n",
2223 USB_MAXCHILDREN);
2224 xhci->num_usb2_ports = USB_MAXCHILDREN;
2225 }
2226
2227 /*
2228 * Note we could have all USB 3.0 ports, or all USB 2.0 ports.
2229 * Not sure how the USB core will handle a hub with no ports...
2230 */
2231 if (xhci->num_usb2_ports) {
2232 xhci->usb2_ports = kmalloc(sizeof(*xhci->usb2_ports)*
2233 xhci->num_usb2_ports, flags);
2234 if (!xhci->usb2_ports)
2235 return -ENOMEM;
2236
2237 port_index = 0;
2238 for (i = 0; i < num_ports; i++) {
2239 if (xhci->port_array[i] == 0x03 ||
2240 xhci->port_array[i] == 0 ||
2241 xhci->port_array[i] == DUPLICATE_ENTRY)
2242 continue;
2243
2244 xhci->usb2_ports[port_index] =
2245 &xhci->op_regs->port_status_base +
2246 NUM_PORT_REGS*i;
2247 xhci_dbg(xhci, "USB 2.0 port at index %u, "
2248 "addr = %p\n", i,
2249 xhci->usb2_ports[port_index]);
2250 port_index++;
2251 if (port_index == xhci->num_usb2_ports)
2252 break;
2253 }
2254 }
2255 if (xhci->num_usb3_ports) {
2256 xhci->usb3_ports = kmalloc(sizeof(*xhci->usb3_ports)*
2257 xhci->num_usb3_ports, flags);
2258 if (!xhci->usb3_ports)
2259 return -ENOMEM;
2260
2261 port_index = 0;
2262 for (i = 0; i < num_ports; i++)
2263 if (xhci->port_array[i] == 0x03) {
2264 xhci->usb3_ports[port_index] =
2265 &xhci->op_regs->port_status_base +
2266 NUM_PORT_REGS*i;
2267 xhci_dbg(xhci, "USB 3.0 port at index %u, "
2268 "addr = %p\n", i,
2269 xhci->usb3_ports[port_index]);
2270 port_index++;
2271 if (port_index == xhci->num_usb3_ports)
2272 break;
2273 }
2274 }
2275 return 0;
2276 }
2277
2278 int xhci_mem_init(struct xhci_hcd *xhci, gfp_t flags)
2279 {
2280 dma_addr_t dma;
2281 struct device *dev = xhci_to_hcd(xhci)->self.controller;
2282 unsigned int val, val2;
2283 u64 val_64;
2284 struct xhci_segment *seg;
2285 u32 page_size, temp;
2286 int i;
2287
2288 INIT_LIST_HEAD(&xhci->lpm_failed_devs);
2289 INIT_LIST_HEAD(&xhci->cancel_cmd_list);
2290
2291 page_size = xhci_readl(xhci, &xhci->op_regs->page_size);
2292 xhci_dbg(xhci, "Supported page size register = 0x%x\n", page_size);
2293 for (i = 0; i < 16; i++) {
2294 if ((0x1 & page_size) != 0)
2295 break;
2296 page_size = page_size >> 1;
2297 }
2298 if (i < 16)
2299 xhci_dbg(xhci, "Supported page size of %iK\n", (1 << (i+12)) / 1024);
2300 else
2301 xhci_warn(xhci, "WARN: no supported page size\n");
2302 /* Use 4K pages, since that's common and the minimum the HC supports */
2303 xhci->page_shift = 12;
2304 xhci->page_size = 1 << xhci->page_shift;
2305 xhci_dbg(xhci, "HCD page size set to %iK\n", xhci->page_size / 1024);
2306
2307 /*
2308 * Program the Number of Device Slots Enabled field in the CONFIG
2309 * register with the max value of slots the HC can handle.
2310 */
2311 val = HCS_MAX_SLOTS(xhci_readl(xhci, &xhci->cap_regs->hcs_params1));
2312 xhci_dbg(xhci, "// xHC can handle at most %d device slots.\n",
2313 (unsigned int) val);
2314 val2 = xhci_readl(xhci, &xhci->op_regs->config_reg);
2315 val |= (val2 & ~HCS_SLOTS_MASK);
2316 xhci_dbg(xhci, "// Setting Max device slots reg = 0x%x.\n",
2317 (unsigned int) val);
2318 xhci_writel(xhci, val, &xhci->op_regs->config_reg);
2319
2320 /*
2321 * Section 5.4.8 - doorbell array must be
2322 * "physically contiguous and 64-byte (cache line) aligned".
2323 */
2324 xhci->dcbaa = dma_alloc_coherent(dev, sizeof(*xhci->dcbaa), &dma,
2325 GFP_KERNEL);
2326 if (!xhci->dcbaa)
2327 goto fail;
2328 memset(xhci->dcbaa, 0, sizeof *(xhci->dcbaa));
2329 xhci->dcbaa->dma = dma;
2330 xhci_dbg(xhci, "// Device context base array address = 0x%llx (DMA), %p (virt)\n",
2331 (unsigned long long)xhci->dcbaa->dma, xhci->dcbaa);
2332 xhci_write_64(xhci, dma, &xhci->op_regs->dcbaa_ptr);
2333
2334 /*
2335 * Initialize the ring segment pool. The ring must be a contiguous
2336 * structure comprised of TRBs. The TRBs must be 16 byte aligned,
2337 * however, the command ring segment needs 64-byte aligned segments,
2338 * so we pick the greater alignment need.
2339 */
2340 xhci->segment_pool = dma_pool_create("xHCI ring segments", dev,
2341 TRB_SEGMENT_SIZE, 64, xhci->page_size);
2342
2343 /* See Table 46 and Note on Figure 55 */
2344 xhci->device_pool = dma_pool_create("xHCI input/output contexts", dev,
2345 2112, 64, xhci->page_size);
2346 if (!xhci->segment_pool || !xhci->device_pool)
2347 goto fail;
2348
2349 /* Linear stream context arrays don't have any boundary restrictions,
2350 * and only need to be 16-byte aligned.
2351 */
2352 xhci->small_streams_pool =
2353 dma_pool_create("xHCI 256 byte stream ctx arrays",
2354 dev, SMALL_STREAM_ARRAY_SIZE, 16, 0);
2355 xhci->medium_streams_pool =
2356 dma_pool_create("xHCI 1KB stream ctx arrays",
2357 dev, MEDIUM_STREAM_ARRAY_SIZE, 16, 0);
2358 /* Any stream context array bigger than MEDIUM_STREAM_ARRAY_SIZE
2359 * will be allocated with dma_alloc_coherent()
2360 */
2361
2362 if (!xhci->small_streams_pool || !xhci->medium_streams_pool)
2363 goto fail;
2364
2365 /* Set up the command ring to have one segments for now. */
2366 xhci->cmd_ring = xhci_ring_alloc(xhci, 1, 1, TYPE_COMMAND, flags);
2367 if (!xhci->cmd_ring)
2368 goto fail;
2369 xhci_dbg(xhci, "Allocated command ring at %p\n", xhci->cmd_ring);
2370 xhci_dbg(xhci, "First segment DMA is 0x%llx\n",
2371 (unsigned long long)xhci->cmd_ring->first_seg->dma);
2372
2373 /* Set the address in the Command Ring Control register */
2374 val_64 = xhci_read_64(xhci, &xhci->op_regs->cmd_ring);
2375 val_64 = (val_64 & (u64) CMD_RING_RSVD_BITS) |
2376 (xhci->cmd_ring->first_seg->dma & (u64) ~CMD_RING_RSVD_BITS) |
2377 xhci->cmd_ring->cycle_state;
2378 xhci_dbg(xhci, "// Setting command ring address to 0x%x\n", val);
2379 xhci_write_64(xhci, val_64, &xhci->op_regs->cmd_ring);
2380 xhci_dbg_cmd_ptrs(xhci);
2381
2382 xhci->lpm_command = xhci_alloc_command(xhci, true, true, flags);
2383 if (!xhci->lpm_command)
2384 goto fail;
2385
2386 /* Reserve one command ring TRB for disabling LPM.
2387 * Since the USB core grabs the shared usb_bus bandwidth mutex before
2388 * disabling LPM, we only need to reserve one TRB for all devices.
2389 */
2390 xhci->cmd_ring_reserved_trbs++;
2391
2392 val = xhci_readl(xhci, &xhci->cap_regs->db_off);
2393 val &= DBOFF_MASK;
2394 xhci_dbg(xhci, "// Doorbell array is located at offset 0x%x"
2395 " from cap regs base addr\n", val);
2396 xhci->dba = (void __iomem *) xhci->cap_regs + val;
2397 xhci_dbg_regs(xhci);
2398 xhci_print_run_regs(xhci);
2399 /* Set ir_set to interrupt register set 0 */
2400 xhci->ir_set = &xhci->run_regs->ir_set[0];
2401
2402 /*
2403 * Event ring setup: Allocate a normal ring, but also setup
2404 * the event ring segment table (ERST). Section 4.9.3.
2405 */
2406 xhci_dbg(xhci, "// Allocating event ring\n");
2407 xhci->event_ring = xhci_ring_alloc(xhci, ERST_NUM_SEGS, 1, TYPE_EVENT,
2408 flags);
2409 if (!xhci->event_ring)
2410 goto fail;
2411 if (xhci_check_trb_in_td_math(xhci, flags) < 0)
2412 goto fail;
2413
2414 xhci->erst.entries = dma_alloc_coherent(dev,
2415 sizeof(struct xhci_erst_entry) * ERST_NUM_SEGS, &dma,
2416 GFP_KERNEL);
2417 if (!xhci->erst.entries)
2418 goto fail;
2419 xhci_dbg(xhci, "// Allocated event ring segment table at 0x%llx\n",
2420 (unsigned long long)dma);
2421
2422 memset(xhci->erst.entries, 0, sizeof(struct xhci_erst_entry)*ERST_NUM_SEGS);
2423 xhci->erst.num_entries = ERST_NUM_SEGS;
2424 xhci->erst.erst_dma_addr = dma;
2425 xhci_dbg(xhci, "Set ERST to 0; private num segs = %i, virt addr = %p, dma addr = 0x%llx\n",
2426 xhci->erst.num_entries,
2427 xhci->erst.entries,
2428 (unsigned long long)xhci->erst.erst_dma_addr);
2429
2430 /* set ring base address and size for each segment table entry */
2431 for (val = 0, seg = xhci->event_ring->first_seg; val < ERST_NUM_SEGS; val++) {
2432 struct xhci_erst_entry *entry = &xhci->erst.entries[val];
2433 entry->seg_addr = cpu_to_le64(seg->dma);
2434 entry->seg_size = cpu_to_le32(TRBS_PER_SEGMENT);
2435 entry->rsvd = 0;
2436 seg = seg->next;
2437 }
2438
2439 /* set ERST count with the number of entries in the segment table */
2440 val = xhci_readl(xhci, &xhci->ir_set->erst_size);
2441 val &= ERST_SIZE_MASK;
2442 val |= ERST_NUM_SEGS;
2443 xhci_dbg(xhci, "// Write ERST size = %i to ir_set 0 (some bits preserved)\n",
2444 val);
2445 xhci_writel(xhci, val, &xhci->ir_set->erst_size);
2446
2447 xhci_dbg(xhci, "// Set ERST entries to point to event ring.\n");
2448 /* set the segment table base address */
2449 xhci_dbg(xhci, "// Set ERST base address for ir_set 0 = 0x%llx\n",
2450 (unsigned long long)xhci->erst.erst_dma_addr);
2451 val_64 = xhci_read_64(xhci, &xhci->ir_set->erst_base);
2452 val_64 &= ERST_PTR_MASK;
2453 val_64 |= (xhci->erst.erst_dma_addr & (u64) ~ERST_PTR_MASK);
2454 xhci_write_64(xhci, val_64, &xhci->ir_set->erst_base);
2455
2456 /* Set the event ring dequeue address */
2457 xhci_set_hc_event_deq(xhci);
2458 xhci_dbg(xhci, "Wrote ERST address to ir_set 0.\n");
2459 xhci_print_ir_set(xhci, 0);
2460
2461 /*
2462 * XXX: Might need to set the Interrupter Moderation Register to
2463 * something other than the default (~1ms minimum between interrupts).
2464 * See section 5.5.1.2.
2465 */
2466 init_completion(&xhci->addr_dev);
2467 for (i = 0; i < MAX_HC_SLOTS; ++i)
2468 xhci->devs[i] = NULL;
2469 for (i = 0; i < USB_MAXCHILDREN; ++i) {
2470 xhci->bus_state[0].resume_done[i] = 0;
2471 xhci->bus_state[1].resume_done[i] = 0;
2472 }
2473
2474 if (scratchpad_alloc(xhci, flags))
2475 goto fail;
2476 if (xhci_setup_port_arrays(xhci, flags))
2477 goto fail;
2478
2479 /* Enable USB 3.0 device notifications for function remote wake, which
2480 * is necessary for allowing USB 3.0 devices to do remote wakeup from
2481 * U3 (device suspend).
2482 */
2483 temp = xhci_readl(xhci, &xhci->op_regs->dev_notification);
2484 temp &= ~DEV_NOTE_MASK;
2485 temp |= DEV_NOTE_FWAKE;
2486 xhci_writel(xhci, temp, &xhci->op_regs->dev_notification);
2487
2488 return 0;
2489
2490 fail:
2491 xhci_warn(xhci, "Couldn't initialize memory\n");
2492 xhci_halt(xhci);
2493 xhci_reset(xhci);
2494 xhci_mem_cleanup(xhci);
2495 return -ENOMEM;
2496 }
This page took 0.084568 seconds and 5 git commands to generate.