xhci: Fix kernel oops in trace_xhci_free_virt_device
[linux-2.6-microblaze.git] / drivers / usb / host / xhci-mem.c
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/pci.h>
13 #include <linux/slab.h>
14 #include <linux/dmapool.h>
15 #include <linux/dma-mapping.h>
16
17 #include "xhci.h"
18 #include "xhci-trace.h"
19 #include "xhci-debugfs.h"
20
21 /*
22  * Allocates a generic ring segment from the ring pool, sets the dma address,
23  * initializes the segment to zero, and sets the private next pointer to NULL.
24  *
25  * Section 4.11.1.1:
26  * "All components of all Command and Transfer TRBs shall be initialized to '0'"
27  */
28 static struct xhci_segment *xhci_segment_alloc(struct xhci_hcd *xhci,
29                                                unsigned int cycle_state,
30                                                unsigned int max_packet,
31                                                gfp_t flags)
32 {
33         struct xhci_segment *seg;
34         dma_addr_t      dma;
35         int             i;
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         /* 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
68 static 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->bounce_buf);
75         kfree(seg);
76 }
77
78 static void xhci_free_segments_for_ring(struct xhci_hcd *xhci,
79                                 struct xhci_segment *first)
80 {
81         struct xhci_segment *seg;
82
83         seg = first->next;
84         while (seg != first) {
85                 struct xhci_segment *next = seg->next;
86                 xhci_segment_free(xhci, seg);
87                 seg = next;
88         }
89         xhci_segment_free(xhci, first);
90 }
91
92 /*
93  * Make the prev segment point to the next segment.
94  *
95  * Change the last TRB in the prev segment to be a Link TRB which points to the
96  * DMA address of the next segment.  The caller needs to set any Link TRB
97  * related flags, such as End TRB, Toggle Cycle, and no snoop.
98  */
99 static void xhci_link_segments(struct xhci_hcd *xhci, struct xhci_segment *prev,
100                 struct xhci_segment *next, enum xhci_ring_type type)
101 {
102         u32 val;
103
104         if (!prev || !next)
105                 return;
106         prev->next = next;
107         if (type != TYPE_EVENT) {
108                 prev->trbs[TRBS_PER_SEGMENT-1].link.segment_ptr =
109                         cpu_to_le64(next->dma);
110
111                 /* Set the last TRB in the segment to have a TRB type ID of Link TRB */
112                 val = le32_to_cpu(prev->trbs[TRBS_PER_SEGMENT-1].link.control);
113                 val &= ~TRB_TYPE_BITMASK;
114                 val |= TRB_TYPE(TRB_LINK);
115                 /* Always set the chain bit with 0.95 hardware */
116                 /* Set chain bit for isoc rings on AMD 0.96 host */
117                 if (xhci_link_trb_quirk(xhci) ||
118                                 (type == TYPE_ISOC &&
119                                  (xhci->quirks & XHCI_AMD_0x96_HOST)))
120                         val |= TRB_CHAIN;
121                 prev->trbs[TRBS_PER_SEGMENT-1].link.control = cpu_to_le32(val);
122         }
123 }
124
125 /*
126  * Link the ring to the new segments.
127  * Set Toggle Cycle for the new ring if needed.
128  */
129 static void xhci_link_rings(struct xhci_hcd *xhci, struct xhci_ring *ring,
130                 struct xhci_segment *first, struct xhci_segment *last,
131                 unsigned int num_segs)
132 {
133         struct xhci_segment *next;
134
135         if (!ring || !first || !last)
136                 return;
137
138         next = ring->enq_seg->next;
139         xhci_link_segments(xhci, ring->enq_seg, first, ring->type);
140         xhci_link_segments(xhci, last, next, ring->type);
141         ring->num_segs += num_segs;
142         ring->num_trbs_free += (TRBS_PER_SEGMENT - 1) * num_segs;
143
144         if (ring->type != TYPE_EVENT && ring->enq_seg == ring->last_seg) {
145                 ring->last_seg->trbs[TRBS_PER_SEGMENT-1].link.control
146                         &= ~cpu_to_le32(LINK_TOGGLE);
147                 last->trbs[TRBS_PER_SEGMENT-1].link.control
148                         |= cpu_to_le32(LINK_TOGGLE);
149                 ring->last_seg = last;
150         }
151 }
152
153 /*
154  * We need a radix tree for mapping physical addresses of TRBs to which stream
155  * ID they belong to.  We need to do this because the host controller won't tell
156  * us which stream ring the TRB came from.  We could store the stream ID in an
157  * event data TRB, but that doesn't help us for the cancellation case, since the
158  * endpoint may stop before it reaches that event data TRB.
159  *
160  * The radix tree maps the upper portion of the TRB DMA address to a ring
161  * segment that has the same upper portion of DMA addresses.  For example, say I
162  * have segments of size 1KB, that are always 1KB aligned.  A segment may
163  * start at 0x10c91000 and end at 0x10c913f0.  If I use the upper 10 bits, the
164  * key to the stream ID is 0x43244.  I can use the DMA address of the TRB to
165  * pass the radix tree a key to get the right stream ID:
166  *
167  *      0x10c90fff >> 10 = 0x43243
168  *      0x10c912c0 >> 10 = 0x43244
169  *      0x10c91400 >> 10 = 0x43245
170  *
171  * Obviously, only those TRBs with DMA addresses that are within the segment
172  * will make the radix tree return the stream ID for that ring.
173  *
174  * Caveats for the radix tree:
175  *
176  * The radix tree uses an unsigned long as a key pair.  On 32-bit systems, an
177  * unsigned long will be 32-bits; on a 64-bit system an unsigned long will be
178  * 64-bits.  Since we only request 32-bit DMA addresses, we can use that as the
179  * key on 32-bit or 64-bit systems (it would also be fine if we asked for 64-bit
180  * PCI DMA addresses on a 64-bit system).  There might be a problem on 32-bit
181  * extended systems (where the DMA address can be bigger than 32-bits),
182  * if we allow the PCI dma mask to be bigger than 32-bits.  So don't do that.
183  */
184 static int xhci_insert_segment_mapping(struct radix_tree_root *trb_address_map,
185                 struct xhci_ring *ring,
186                 struct xhci_segment *seg,
187                 gfp_t mem_flags)
188 {
189         unsigned long key;
190         int ret;
191
192         key = (unsigned long)(seg->dma >> TRB_SEGMENT_SHIFT);
193         /* Skip any segments that were already added. */
194         if (radix_tree_lookup(trb_address_map, key))
195                 return 0;
196
197         ret = radix_tree_maybe_preload(mem_flags);
198         if (ret)
199                 return ret;
200         ret = radix_tree_insert(trb_address_map,
201                         key, ring);
202         radix_tree_preload_end();
203         return ret;
204 }
205
206 static void xhci_remove_segment_mapping(struct radix_tree_root *trb_address_map,
207                 struct xhci_segment *seg)
208 {
209         unsigned long key;
210
211         key = (unsigned long)(seg->dma >> TRB_SEGMENT_SHIFT);
212         if (radix_tree_lookup(trb_address_map, key))
213                 radix_tree_delete(trb_address_map, key);
214 }
215
216 static int xhci_update_stream_segment_mapping(
217                 struct radix_tree_root *trb_address_map,
218                 struct xhci_ring *ring,
219                 struct xhci_segment *first_seg,
220                 struct xhci_segment *last_seg,
221                 gfp_t mem_flags)
222 {
223         struct xhci_segment *seg;
224         struct xhci_segment *failed_seg;
225         int ret;
226
227         if (WARN_ON_ONCE(trb_address_map == NULL))
228                 return 0;
229
230         seg = first_seg;
231         do {
232                 ret = xhci_insert_segment_mapping(trb_address_map,
233                                 ring, seg, mem_flags);
234                 if (ret)
235                         goto remove_streams;
236                 if (seg == last_seg)
237                         return 0;
238                 seg = seg->next;
239         } while (seg != first_seg);
240
241         return 0;
242
243 remove_streams:
244         failed_seg = seg;
245         seg = first_seg;
246         do {
247                 xhci_remove_segment_mapping(trb_address_map, seg);
248                 if (seg == failed_seg)
249                         return ret;
250                 seg = seg->next;
251         } while (seg != first_seg);
252
253         return ret;
254 }
255
256 static void xhci_remove_stream_mapping(struct xhci_ring *ring)
257 {
258         struct xhci_segment *seg;
259
260         if (WARN_ON_ONCE(ring->trb_address_map == NULL))
261                 return;
262
263         seg = ring->first_seg;
264         do {
265                 xhci_remove_segment_mapping(ring->trb_address_map, seg);
266                 seg = seg->next;
267         } while (seg != ring->first_seg);
268 }
269
270 static int xhci_update_stream_mapping(struct xhci_ring *ring, gfp_t mem_flags)
271 {
272         return xhci_update_stream_segment_mapping(ring->trb_address_map, ring,
273                         ring->first_seg, ring->last_seg, mem_flags);
274 }
275
276 /* XXX: Do we need the hcd structure in all these functions? */
277 void xhci_ring_free(struct xhci_hcd *xhci, struct xhci_ring *ring)
278 {
279         if (!ring)
280                 return;
281
282         trace_xhci_ring_free(ring);
283
284         if (ring->first_seg) {
285                 if (ring->type == TYPE_STREAM)
286                         xhci_remove_stream_mapping(ring);
287                 xhci_free_segments_for_ring(xhci, ring->first_seg);
288         }
289
290         kfree(ring);
291 }
292
293 static void xhci_initialize_ring_info(struct xhci_ring *ring,
294                                         unsigned int cycle_state)
295 {
296         /* The ring is empty, so the enqueue pointer == dequeue pointer */
297         ring->enqueue = ring->first_seg->trbs;
298         ring->enq_seg = ring->first_seg;
299         ring->dequeue = ring->enqueue;
300         ring->deq_seg = ring->first_seg;
301         /* The ring is initialized to 0. The producer must write 1 to the cycle
302          * bit to handover ownership of the TRB, so PCS = 1.  The consumer must
303          * compare CCS to the cycle bit to check ownership, so CCS = 1.
304          *
305          * New rings are initialized with cycle state equal to 1; if we are
306          * handling ring expansion, set the cycle state equal to the old ring.
307          */
308         ring->cycle_state = cycle_state;
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 */
318 static 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, unsigned int max_packet, gfp_t flags)
322 {
323         struct xhci_segment *prev;
324
325         prev = xhci_segment_alloc(xhci, cycle_state, max_packet, 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, max_packet, 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  */
362 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, unsigned int max_packet, gfp_t flags)
365 {
366         struct xhci_ring        *ring;
367         int ret;
368         struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
369
370         ring = kzalloc_node(sizeof(*ring), flags, dev_to_node(dev));
371         if (!ring)
372                 return NULL;
373
374         ring->num_segs = num_segs;
375         ring->bounce_buf_len = max_packet;
376         INIT_LIST_HEAD(&ring->td_list);
377         ring->type = type;
378         if (num_segs == 0)
379                 return ring;
380
381         ret = xhci_alloc_segments_for_ring(xhci, &ring->first_seg,
382                         &ring->last_seg, num_segs, cycle_state, type,
383                         max_packet, flags);
384         if (ret)
385                 goto fail;
386
387         /* Only event ring does not use link TRB */
388         if (type != TYPE_EVENT) {
389                 /* See section 4.9.2.1 and 6.4.4.1 */
390                 ring->last_seg->trbs[TRBS_PER_SEGMENT - 1].link.control |=
391                         cpu_to_le32(LINK_TOGGLE);
392         }
393         xhci_initialize_ring_info(ring, cycle_state);
394         trace_xhci_ring_alloc(ring);
395         return ring;
396
397 fail:
398         kfree(ring);
399         return NULL;
400 }
401
402 void 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  */
414 int xhci_ring_expansion(struct xhci_hcd *xhci, struct xhci_ring *ring,
415                                 unsigned int num_trbs, gfp_t flags)
416 {
417         struct xhci_segment     *first;
418         struct xhci_segment     *last;
419         unsigned int            num_segs;
420         unsigned int            num_segs_needed;
421         int                     ret;
422
423         num_segs_needed = (num_trbs + (TRBS_PER_SEGMENT - 1) - 1) /
424                                 (TRBS_PER_SEGMENT - 1);
425
426         /* Allocate number of segments we needed, or double the ring size */
427         num_segs = ring->num_segs > num_segs_needed ?
428                         ring->num_segs : num_segs_needed;
429
430         ret = xhci_alloc_segments_for_ring(xhci, &first, &last,
431                         num_segs, ring->cycle_state, ring->type,
432                         ring->bounce_buf_len, flags);
433         if (ret)
434                 return -ENOMEM;
435
436         if (ring->type == TYPE_STREAM)
437                 ret = xhci_update_stream_segment_mapping(ring->trb_address_map,
438                                                 ring, first, last, flags);
439         if (ret) {
440                 struct xhci_segment *next;
441                 do {
442                         next = first->next;
443                         xhci_segment_free(xhci, first);
444                         if (first == last)
445                                 break;
446                         first = next;
447                 } while (true);
448                 return ret;
449         }
450
451         xhci_link_rings(xhci, ring, first, last, num_segs);
452         trace_xhci_ring_expansion(ring);
453         xhci_dbg_trace(xhci, trace_xhci_dbg_ring_expansion,
454                         "ring expansion succeed, now has %d segments",
455                         ring->num_segs);
456
457         return 0;
458 }
459
460 struct xhci_container_ctx *xhci_alloc_container_ctx(struct xhci_hcd *xhci,
461                                                     int type, gfp_t flags)
462 {
463         struct xhci_container_ctx *ctx;
464         struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
465
466         if ((type != XHCI_CTX_TYPE_DEVICE) && (type != XHCI_CTX_TYPE_INPUT))
467                 return NULL;
468
469         ctx = kzalloc_node(sizeof(*ctx), flags, dev_to_node(dev));
470         if (!ctx)
471                 return NULL;
472
473         ctx->type = type;
474         ctx->size = HCC_64BYTE_CONTEXT(xhci->hcc_params) ? 2048 : 1024;
475         if (type == XHCI_CTX_TYPE_INPUT)
476                 ctx->size += CTX_SIZE(xhci->hcc_params);
477
478         ctx->bytes = dma_pool_zalloc(xhci->device_pool, flags, &ctx->dma);
479         if (!ctx->bytes) {
480                 kfree(ctx);
481                 return NULL;
482         }
483         return ctx;
484 }
485
486 void xhci_free_container_ctx(struct xhci_hcd *xhci,
487                              struct xhci_container_ctx *ctx)
488 {
489         if (!ctx)
490                 return;
491         dma_pool_free(xhci->device_pool, ctx->bytes, ctx->dma);
492         kfree(ctx);
493 }
494
495 struct xhci_input_control_ctx *xhci_get_input_control_ctx(
496                                               struct xhci_container_ctx *ctx)
497 {
498         if (ctx->type != XHCI_CTX_TYPE_INPUT)
499                 return NULL;
500
501         return (struct xhci_input_control_ctx *)ctx->bytes;
502 }
503
504 struct xhci_slot_ctx *xhci_get_slot_ctx(struct xhci_hcd *xhci,
505                                         struct xhci_container_ctx *ctx)
506 {
507         if (ctx->type == XHCI_CTX_TYPE_DEVICE)
508                 return (struct xhci_slot_ctx *)ctx->bytes;
509
510         return (struct xhci_slot_ctx *)
511                 (ctx->bytes + CTX_SIZE(xhci->hcc_params));
512 }
513
514 struct xhci_ep_ctx *xhci_get_ep_ctx(struct xhci_hcd *xhci,
515                                     struct xhci_container_ctx *ctx,
516                                     unsigned int ep_index)
517 {
518         /* increment ep index by offset of start of ep ctx array */
519         ep_index++;
520         if (ctx->type == XHCI_CTX_TYPE_INPUT)
521                 ep_index++;
522
523         return (struct xhci_ep_ctx *)
524                 (ctx->bytes + (ep_index * CTX_SIZE(xhci->hcc_params)));
525 }
526
527
528 /***************** Streams structures manipulation *************************/
529
530 static void xhci_free_stream_ctx(struct xhci_hcd *xhci,
531                 unsigned int num_stream_ctxs,
532                 struct xhci_stream_ctx *stream_ctx, dma_addr_t dma)
533 {
534         struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
535         size_t size = sizeof(struct xhci_stream_ctx) * num_stream_ctxs;
536
537         if (size > MEDIUM_STREAM_ARRAY_SIZE)
538                 dma_free_coherent(dev, size,
539                                 stream_ctx, dma);
540         else if (size <= SMALL_STREAM_ARRAY_SIZE)
541                 return dma_pool_free(xhci->small_streams_pool,
542                                 stream_ctx, dma);
543         else
544                 return dma_pool_free(xhci->medium_streams_pool,
545                                 stream_ctx, dma);
546 }
547
548 /*
549  * The stream context array for each endpoint with bulk streams enabled can
550  * vary in size, based on:
551  *  - how many streams the endpoint supports,
552  *  - the maximum primary stream array size the host controller supports,
553  *  - and how many streams the device driver asks for.
554  *
555  * The stream context array must be a power of 2, and can be as small as
556  * 64 bytes or as large as 1MB.
557  */
558 static struct xhci_stream_ctx *xhci_alloc_stream_ctx(struct xhci_hcd *xhci,
559                 unsigned int num_stream_ctxs, dma_addr_t *dma,
560                 gfp_t mem_flags)
561 {
562         struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
563         size_t size = sizeof(struct xhci_stream_ctx) * num_stream_ctxs;
564
565         if (size > MEDIUM_STREAM_ARRAY_SIZE)
566                 return dma_alloc_coherent(dev, size,
567                                 dma, mem_flags);
568         else if (size <= SMALL_STREAM_ARRAY_SIZE)
569                 return dma_pool_alloc(xhci->small_streams_pool,
570                                 mem_flags, dma);
571         else
572                 return dma_pool_alloc(xhci->medium_streams_pool,
573                                 mem_flags, dma);
574 }
575
576 struct xhci_ring *xhci_dma_to_transfer_ring(
577                 struct xhci_virt_ep *ep,
578                 u64 address)
579 {
580         if (ep->ep_state & EP_HAS_STREAMS)
581                 return radix_tree_lookup(&ep->stream_info->trb_address_map,
582                                 address >> TRB_SEGMENT_SHIFT);
583         return ep->ring;
584 }
585
586 struct xhci_ring *xhci_stream_id_to_ring(
587                 struct xhci_virt_device *dev,
588                 unsigned int ep_index,
589                 unsigned int stream_id)
590 {
591         struct xhci_virt_ep *ep = &dev->eps[ep_index];
592
593         if (stream_id == 0)
594                 return ep->ring;
595         if (!ep->stream_info)
596                 return NULL;
597
598         if (stream_id > ep->stream_info->num_streams)
599                 return NULL;
600         return ep->stream_info->stream_rings[stream_id];
601 }
602
603 /*
604  * Change an endpoint's internal structure so it supports stream IDs.  The
605  * number of requested streams includes stream 0, which cannot be used by device
606  * drivers.
607  *
608  * The number of stream contexts in the stream context array may be bigger than
609  * the number of streams the driver wants to use.  This is because the number of
610  * stream context array entries must be a power of two.
611  */
612 struct xhci_stream_info *xhci_alloc_stream_info(struct xhci_hcd *xhci,
613                 unsigned int num_stream_ctxs,
614                 unsigned int num_streams,
615                 unsigned int max_packet, gfp_t mem_flags)
616 {
617         struct xhci_stream_info *stream_info;
618         u32 cur_stream;
619         struct xhci_ring *cur_ring;
620         u64 addr;
621         int ret;
622         struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
623
624         xhci_dbg(xhci, "Allocating %u streams and %u "
625                         "stream context array entries.\n",
626                         num_streams, num_stream_ctxs);
627         if (xhci->cmd_ring_reserved_trbs == MAX_RSVD_CMD_TRBS) {
628                 xhci_dbg(xhci, "Command ring has no reserved TRBs available\n");
629                 return NULL;
630         }
631         xhci->cmd_ring_reserved_trbs++;
632
633         stream_info = kzalloc_node(sizeof(*stream_info), mem_flags,
634                         dev_to_node(dev));
635         if (!stream_info)
636                 goto cleanup_trbs;
637
638         stream_info->num_streams = num_streams;
639         stream_info->num_stream_ctxs = num_stream_ctxs;
640
641         /* Initialize the array of virtual pointers to stream rings. */
642         stream_info->stream_rings = kcalloc_node(
643                         num_streams, sizeof(struct xhci_ring *), mem_flags,
644                         dev_to_node(dev));
645         if (!stream_info->stream_rings)
646                 goto cleanup_info;
647
648         /* Initialize the array of DMA addresses for stream rings for the HW. */
649         stream_info->stream_ctx_array = xhci_alloc_stream_ctx(xhci,
650                         num_stream_ctxs, &stream_info->ctx_array_dma,
651                         mem_flags);
652         if (!stream_info->stream_ctx_array)
653                 goto cleanup_ctx;
654         memset(stream_info->stream_ctx_array, 0,
655                         sizeof(struct xhci_stream_ctx)*num_stream_ctxs);
656
657         /* Allocate everything needed to free the stream rings later */
658         stream_info->free_streams_command =
659                 xhci_alloc_command_with_ctx(xhci, true, mem_flags);
660         if (!stream_info->free_streams_command)
661                 goto cleanup_ctx;
662
663         INIT_RADIX_TREE(&stream_info->trb_address_map, GFP_ATOMIC);
664
665         /* Allocate rings for all the streams that the driver will use,
666          * and add their segment DMA addresses to the radix tree.
667          * Stream 0 is reserved.
668          */
669
670         for (cur_stream = 1; cur_stream < num_streams; cur_stream++) {
671                 stream_info->stream_rings[cur_stream] =
672                         xhci_ring_alloc(xhci, 2, 1, TYPE_STREAM, max_packet,
673                                         mem_flags);
674                 cur_ring = stream_info->stream_rings[cur_stream];
675                 if (!cur_ring)
676                         goto cleanup_rings;
677                 cur_ring->stream_id = cur_stream;
678                 cur_ring->trb_address_map = &stream_info->trb_address_map;
679                 /* Set deq ptr, cycle bit, and stream context type */
680                 addr = cur_ring->first_seg->dma |
681                         SCT_FOR_CTX(SCT_PRI_TR) |
682                         cur_ring->cycle_state;
683                 stream_info->stream_ctx_array[cur_stream].stream_ring =
684                         cpu_to_le64(addr);
685                 xhci_dbg(xhci, "Setting stream %d ring ptr to 0x%08llx\n",
686                                 cur_stream, (unsigned long long) addr);
687
688                 ret = xhci_update_stream_mapping(cur_ring, mem_flags);
689                 if (ret) {
690                         xhci_ring_free(xhci, cur_ring);
691                         stream_info->stream_rings[cur_stream] = NULL;
692                         goto cleanup_rings;
693                 }
694         }
695         /* Leave the other unused stream ring pointers in the stream context
696          * array initialized to zero.  This will cause the xHC to give us an
697          * error if the device asks for a stream ID we don't have setup (if it
698          * was any other way, the host controller would assume the ring is
699          * "empty" and wait forever for data to be queued to that stream ID).
700          */
701
702         return stream_info;
703
704 cleanup_rings:
705         for (cur_stream = 1; cur_stream < num_streams; cur_stream++) {
706                 cur_ring = stream_info->stream_rings[cur_stream];
707                 if (cur_ring) {
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_trace(xhci,  trace_xhci_dbg_context_change,
736                         "Setting number of stream ctx array entries to %u",
737                         1 << (max_primary_streams + 1));
738         ep_ctx->ep_info &= cpu_to_le32(~EP_MAXPSTREAMS_MASK);
739         ep_ctx->ep_info |= cpu_to_le32(EP_MAXPSTREAMS(max_primary_streams)
740                                        | EP_HAS_LSA);
741         ep_ctx->deq  = cpu_to_le64(stream_info->ctx_array_dma);
742 }
743
744 /*
745  * Sets the MaxPStreams field and the Linear Stream Array field to 0.
746  * Reinstalls the "normal" endpoint ring (at its previous dequeue mark,
747  * not at the beginning of the ring).
748  */
749 void xhci_setup_no_streams_ep_input_ctx(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
768         if (!stream_info)
769                 return;
770
771         for (cur_stream = 1; cur_stream < stream_info->num_streams;
772                         cur_stream++) {
773                 cur_ring = stream_info->stream_rings[cur_stream];
774                 if (cur_ring) {
775                         xhci_ring_free(xhci, cur_ring);
776                         stream_info->stream_rings[cur_stream] = NULL;
777                 }
778         }
779         xhci_free_command(xhci, stream_info->free_streams_command);
780         xhci->cmd_ring_reserved_trbs--;
781         if (stream_info->stream_ctx_array)
782                 xhci_free_stream_ctx(xhci,
783                                 stream_info->num_stream_ctxs,
784                                 stream_info->stream_ctx_array,
785                                 stream_info->ctx_array_dma);
786
787         kfree(stream_info->stream_rings);
788         kfree(stream_info);
789 }
790
791
792 /***************** Device context manipulation *************************/
793
794 static void xhci_init_endpoint_timer(struct xhci_hcd *xhci,
795                 struct xhci_virt_ep *ep)
796 {
797         timer_setup(&ep->stop_cmd_timer, xhci_stop_endpoint_command_watchdog,
798                     0);
799         ep->xhci = xhci;
800 }
801
802 static void xhci_free_tt_info(struct xhci_hcd *xhci,
803                 struct xhci_virt_device *virt_dev,
804                 int slot_id)
805 {
806         struct list_head *tt_list_head;
807         struct xhci_tt_bw_info *tt_info, *next;
808         bool slot_found = false;
809
810         /* If the device never made it past the Set Address stage,
811          * it may not have the real_port set correctly.
812          */
813         if (virt_dev->real_port == 0 ||
814                         virt_dev->real_port > HCS_MAX_PORTS(xhci->hcs_params1)) {
815                 xhci_dbg(xhci, "Bad real port.\n");
816                 return;
817         }
818
819         tt_list_head = &(xhci->rh_bw[virt_dev->real_port - 1].tts);
820         list_for_each_entry_safe(tt_info, next, tt_list_head, tt_list) {
821                 /* Multi-TT hubs will have more than one entry */
822                 if (tt_info->slot_id == slot_id) {
823                         slot_found = true;
824                         list_del(&tt_info->tt_list);
825                         kfree(tt_info);
826                 } else if (slot_found) {
827                         break;
828                 }
829         }
830 }
831
832 int xhci_alloc_tt_info(struct xhci_hcd *xhci,
833                 struct xhci_virt_device *virt_dev,
834                 struct usb_device *hdev,
835                 struct usb_tt *tt, gfp_t mem_flags)
836 {
837         struct xhci_tt_bw_info          *tt_info;
838         unsigned int                    num_ports;
839         int                             i, j;
840         struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
841
842         if (!tt->multi)
843                 num_ports = 1;
844         else
845                 num_ports = hdev->maxchild;
846
847         for (i = 0; i < num_ports; i++, tt_info++) {
848                 struct xhci_interval_bw_table *bw_table;
849
850                 tt_info = kzalloc_node(sizeof(*tt_info), mem_flags,
851                                 dev_to_node(dev));
852                 if (!tt_info)
853                         goto free_tts;
854                 INIT_LIST_HEAD(&tt_info->tt_list);
855                 list_add(&tt_info->tt_list,
856                                 &xhci->rh_bw[virt_dev->real_port - 1].tts);
857                 tt_info->slot_id = virt_dev->udev->slot_id;
858                 if (tt->multi)
859                         tt_info->ttport = i+1;
860                 bw_table = &tt_info->bw_table;
861                 for (j = 0; j < XHCI_MAX_INTERVAL; j++)
862                         INIT_LIST_HEAD(&bw_table->interval_bw[j].endpoints);
863         }
864         return 0;
865
866 free_tts:
867         xhci_free_tt_info(xhci, virt_dev, virt_dev->udev->slot_id);
868         return -ENOMEM;
869 }
870
871
872 /* All the xhci_tds in the ring's TD list should be freed at this point.
873  * Should be called with xhci->lock held if there is any chance the TT lists
874  * will be manipulated by the configure endpoint, allocate device, or update
875  * hub functions while this function is removing the TT entries from the list.
876  */
877 void xhci_free_virt_device(struct xhci_hcd *xhci, int slot_id)
878 {
879         struct xhci_virt_device *dev;
880         int i;
881         int old_active_eps = 0;
882
883         /* Slot ID 0 is reserved */
884         if (slot_id == 0 || !xhci->devs[slot_id])
885                 return;
886
887         dev = xhci->devs[slot_id];
888
889         xhci->dcbaa->dev_context_ptrs[slot_id] = 0;
890         if (!dev)
891                 return;
892
893         trace_xhci_free_virt_device(dev);
894
895         if (dev->tt_info)
896                 old_active_eps = dev->tt_info->active_eps;
897
898         for (i = 0; i < 31; i++) {
899                 if (dev->eps[i].ring)
900                         xhci_ring_free(xhci, dev->eps[i].ring);
901                 if (dev->eps[i].stream_info)
902                         xhci_free_stream_info(xhci,
903                                         dev->eps[i].stream_info);
904                 /* Endpoints on the TT/root port lists should have been removed
905                  * when usb_disable_device() was called for the device.
906                  * We can't drop them anyway, because the udev might have gone
907                  * away by this point, and we can't tell what speed it was.
908                  */
909                 if (!list_empty(&dev->eps[i].bw_endpoint_list))
910                         xhci_warn(xhci, "Slot %u endpoint %u "
911                                         "not removed from BW list!\n",
912                                         slot_id, i);
913         }
914         /* If this is a hub, free the TT(s) from the TT list */
915         xhci_free_tt_info(xhci, dev, slot_id);
916         /* If necessary, update the number of active TTs on this root port */
917         xhci_update_tt_active_eps(xhci, dev, old_active_eps);
918
919         if (dev->in_ctx)
920                 xhci_free_container_ctx(xhci, dev->in_ctx);
921         if (dev->out_ctx)
922                 xhci_free_container_ctx(xhci, dev->out_ctx);
923
924         if (dev->udev && dev->udev->slot_id)
925                 dev->udev->slot_id = 0;
926         kfree(xhci->devs[slot_id]);
927         xhci->devs[slot_id] = NULL;
928 }
929
930 /*
931  * Free a virt_device structure.
932  * If the virt_device added a tt_info (a hub) and has children pointing to
933  * that tt_info, then free the child first. Recursive.
934  * We can't rely on udev at this point to find child-parent relationships.
935  */
936 void xhci_free_virt_devices_depth_first(struct xhci_hcd *xhci, int slot_id)
937 {
938         struct xhci_virt_device *vdev;
939         struct list_head *tt_list_head;
940         struct xhci_tt_bw_info *tt_info, *next;
941         int i;
942
943         vdev = xhci->devs[slot_id];
944         if (!vdev)
945                 return;
946
947         if (vdev->real_port == 0 ||
948                         vdev->real_port > HCS_MAX_PORTS(xhci->hcs_params1)) {
949                 xhci_dbg(xhci, "Bad vdev->real_port.\n");
950                 goto out;
951         }
952
953         tt_list_head = &(xhci->rh_bw[vdev->real_port - 1].tts);
954         list_for_each_entry_safe(tt_info, next, tt_list_head, tt_list) {
955                 /* is this a hub device that added a tt_info to the tts list */
956                 if (tt_info->slot_id == slot_id) {
957                         /* are any devices using this tt_info? */
958                         for (i = 1; i < HCS_MAX_SLOTS(xhci->hcs_params1); i++) {
959                                 vdev = xhci->devs[i];
960                                 if (vdev && (vdev->tt_info == tt_info))
961                                         xhci_free_virt_devices_depth_first(
962                                                 xhci, i);
963                         }
964                 }
965         }
966 out:
967         /* we are now at a leaf device */
968         xhci_debugfs_remove_slot(xhci, slot_id);
969         xhci_free_virt_device(xhci, slot_id);
970 }
971
972 int xhci_alloc_virt_device(struct xhci_hcd *xhci, int slot_id,
973                 struct usb_device *udev, gfp_t flags)
974 {
975         struct xhci_virt_device *dev;
976         int i;
977
978         /* Slot ID 0 is reserved */
979         if (slot_id == 0 || xhci->devs[slot_id]) {
980                 xhci_warn(xhci, "Bad Slot ID %d\n", slot_id);
981                 return 0;
982         }
983
984         dev = kzalloc(sizeof(*dev), flags);
985         if (!dev)
986                 return 0;
987
988         /* Allocate the (output) device context that will be used in the HC. */
989         dev->out_ctx = xhci_alloc_container_ctx(xhci, XHCI_CTX_TYPE_DEVICE, flags);
990         if (!dev->out_ctx)
991                 goto fail;
992
993         xhci_dbg(xhci, "Slot %d output ctx = 0x%llx (dma)\n", slot_id,
994                         (unsigned long long)dev->out_ctx->dma);
995
996         /* Allocate the (input) device context for address device command */
997         dev->in_ctx = xhci_alloc_container_ctx(xhci, XHCI_CTX_TYPE_INPUT, flags);
998         if (!dev->in_ctx)
999                 goto fail;
1000
1001         xhci_dbg(xhci, "Slot %d input ctx = 0x%llx (dma)\n", slot_id,
1002                         (unsigned long long)dev->in_ctx->dma);
1003
1004         /* Initialize the cancellation list and watchdog timers for each ep */
1005         for (i = 0; i < 31; i++) {
1006                 xhci_init_endpoint_timer(xhci, &dev->eps[i]);
1007                 INIT_LIST_HEAD(&dev->eps[i].cancelled_td_list);
1008                 INIT_LIST_HEAD(&dev->eps[i].bw_endpoint_list);
1009         }
1010
1011         /* Allocate endpoint 0 ring */
1012         dev->eps[0].ring = xhci_ring_alloc(xhci, 2, 1, TYPE_CTRL, 0, flags);
1013         if (!dev->eps[0].ring)
1014                 goto fail;
1015
1016         dev->udev = udev;
1017
1018         /* Point to output device context in dcbaa. */
1019         xhci->dcbaa->dev_context_ptrs[slot_id] = cpu_to_le64(dev->out_ctx->dma);
1020         xhci_dbg(xhci, "Set slot id %d dcbaa entry %p to 0x%llx\n",
1021                  slot_id,
1022                  &xhci->dcbaa->dev_context_ptrs[slot_id],
1023                  le64_to_cpu(xhci->dcbaa->dev_context_ptrs[slot_id]));
1024
1025         trace_xhci_alloc_virt_device(dev);
1026
1027         xhci->devs[slot_id] = dev;
1028
1029         return 1;
1030 fail:
1031
1032         if (dev->in_ctx)
1033                 xhci_free_container_ctx(xhci, dev->in_ctx);
1034         if (dev->out_ctx)
1035                 xhci_free_container_ctx(xhci, dev->out_ctx);
1036         kfree(dev);
1037
1038         return 0;
1039 }
1040
1041 void xhci_copy_ep0_dequeue_into_input_ctx(struct xhci_hcd *xhci,
1042                 struct usb_device *udev)
1043 {
1044         struct xhci_virt_device *virt_dev;
1045         struct xhci_ep_ctx      *ep0_ctx;
1046         struct xhci_ring        *ep_ring;
1047
1048         virt_dev = xhci->devs[udev->slot_id];
1049         ep0_ctx = xhci_get_ep_ctx(xhci, virt_dev->in_ctx, 0);
1050         ep_ring = virt_dev->eps[0].ring;
1051         /*
1052          * FIXME we don't keep track of the dequeue pointer very well after a
1053          * Set TR dequeue pointer, so we're setting the dequeue pointer of the
1054          * host to our enqueue pointer.  This should only be called after a
1055          * configured device has reset, so all control transfers should have
1056          * been completed or cancelled before the reset.
1057          */
1058         ep0_ctx->deq = cpu_to_le64(xhci_trb_virt_to_dma(ep_ring->enq_seg,
1059                                                         ep_ring->enqueue)
1060                                    | ep_ring->cycle_state);
1061 }
1062
1063 /*
1064  * The xHCI roothub may have ports of differing speeds in any order in the port
1065  * status registers.
1066  *
1067  * The xHCI hardware wants to know the roothub port number that the USB device
1068  * is attached to (or the roothub port its ancestor hub is attached to).  All we
1069  * know is the index of that port under either the USB 2.0 or the USB 3.0
1070  * roothub, but that doesn't give us the real index into the HW port status
1071  * registers. Call xhci_find_raw_port_number() to get real index.
1072  */
1073 static u32 xhci_find_real_port_number(struct xhci_hcd *xhci,
1074                 struct usb_device *udev)
1075 {
1076         struct usb_device *top_dev;
1077         struct usb_hcd *hcd;
1078
1079         if (udev->speed >= USB_SPEED_SUPER)
1080                 hcd = xhci->shared_hcd;
1081         else
1082                 hcd = xhci->main_hcd;
1083
1084         for (top_dev = udev; top_dev->parent && top_dev->parent->parent;
1085                         top_dev = top_dev->parent)
1086                 /* Found device below root hub */;
1087
1088         return  xhci_find_raw_port_number(hcd, top_dev->portnum);
1089 }
1090
1091 /* Setup an xHCI virtual device for a Set Address command */
1092 int xhci_setup_addressable_virt_dev(struct xhci_hcd *xhci, struct usb_device *udev)
1093 {
1094         struct xhci_virt_device *dev;
1095         struct xhci_ep_ctx      *ep0_ctx;
1096         struct xhci_slot_ctx    *slot_ctx;
1097         u32                     port_num;
1098         u32                     max_packets;
1099         struct usb_device *top_dev;
1100
1101         dev = xhci->devs[udev->slot_id];
1102         /* Slot ID 0 is reserved */
1103         if (udev->slot_id == 0 || !dev) {
1104                 xhci_warn(xhci, "Slot ID %d is not assigned to this device\n",
1105                                 udev->slot_id);
1106                 return -EINVAL;
1107         }
1108         ep0_ctx = xhci_get_ep_ctx(xhci, dev->in_ctx, 0);
1109         slot_ctx = xhci_get_slot_ctx(xhci, dev->in_ctx);
1110
1111         /* 3) Only the control endpoint is valid - one endpoint context */
1112         slot_ctx->dev_info |= cpu_to_le32(LAST_CTX(1) | udev->route);
1113         switch (udev->speed) {
1114         case USB_SPEED_SUPER_PLUS:
1115                 slot_ctx->dev_info |= cpu_to_le32(SLOT_SPEED_SSP);
1116                 max_packets = MAX_PACKET(512);
1117                 break;
1118         case USB_SPEED_SUPER:
1119                 slot_ctx->dev_info |= cpu_to_le32(SLOT_SPEED_SS);
1120                 max_packets = MAX_PACKET(512);
1121                 break;
1122         case USB_SPEED_HIGH:
1123                 slot_ctx->dev_info |= cpu_to_le32(SLOT_SPEED_HS);
1124                 max_packets = MAX_PACKET(64);
1125                 break;
1126         /* USB core guesses at a 64-byte max packet first for FS devices */
1127         case USB_SPEED_FULL:
1128                 slot_ctx->dev_info |= cpu_to_le32(SLOT_SPEED_FS);
1129                 max_packets = MAX_PACKET(64);
1130                 break;
1131         case USB_SPEED_LOW:
1132                 slot_ctx->dev_info |= cpu_to_le32(SLOT_SPEED_LS);
1133                 max_packets = MAX_PACKET(8);
1134                 break;
1135         case USB_SPEED_WIRELESS:
1136                 xhci_dbg(xhci, "FIXME xHCI doesn't support wireless speeds\n");
1137                 return -EINVAL;
1138                 break;
1139         default:
1140                 /* Speed was set earlier, this shouldn't happen. */
1141                 return -EINVAL;
1142         }
1143         /* Find the root hub port this device is under */
1144         port_num = xhci_find_real_port_number(xhci, udev);
1145         if (!port_num)
1146                 return -EINVAL;
1147         slot_ctx->dev_info2 |= cpu_to_le32(ROOT_HUB_PORT(port_num));
1148         /* Set the port number in the virtual_device to the faked port number */
1149         for (top_dev = udev; top_dev->parent && top_dev->parent->parent;
1150                         top_dev = top_dev->parent)
1151                 /* Found device below root hub */;
1152         dev->fake_port = top_dev->portnum;
1153         dev->real_port = port_num;
1154         xhci_dbg(xhci, "Set root hub portnum to %d\n", port_num);
1155         xhci_dbg(xhci, "Set fake root hub portnum to %d\n", dev->fake_port);
1156
1157         /* Find the right bandwidth table that this device will be a part of.
1158          * If this is a full speed device attached directly to a root port (or a
1159          * decendent of one), it counts as a primary bandwidth domain, not a
1160          * secondary bandwidth domain under a TT.  An xhci_tt_info structure
1161          * will never be created for the HS root hub.
1162          */
1163         if (!udev->tt || !udev->tt->hub->parent) {
1164                 dev->bw_table = &xhci->rh_bw[port_num - 1].bw_table;
1165         } else {
1166                 struct xhci_root_port_bw_info *rh_bw;
1167                 struct xhci_tt_bw_info *tt_bw;
1168
1169                 rh_bw = &xhci->rh_bw[port_num - 1];
1170                 /* Find the right TT. */
1171                 list_for_each_entry(tt_bw, &rh_bw->tts, tt_list) {
1172                         if (tt_bw->slot_id != udev->tt->hub->slot_id)
1173                                 continue;
1174
1175                         if (!dev->udev->tt->multi ||
1176                                         (udev->tt->multi &&
1177                                          tt_bw->ttport == dev->udev->ttport)) {
1178                                 dev->bw_table = &tt_bw->bw_table;
1179                                 dev->tt_info = tt_bw;
1180                                 break;
1181                         }
1182                 }
1183                 if (!dev->tt_info)
1184                         xhci_warn(xhci, "WARN: Didn't find a matching TT\n");
1185         }
1186
1187         /* Is this a LS/FS device under an external HS hub? */
1188         if (udev->tt && udev->tt->hub->parent) {
1189                 slot_ctx->tt_info = cpu_to_le32(udev->tt->hub->slot_id |
1190                                                 (udev->ttport << 8));
1191                 if (udev->tt->multi)
1192                         slot_ctx->dev_info |= cpu_to_le32(DEV_MTT);
1193         }
1194         xhci_dbg(xhci, "udev->tt = %p\n", udev->tt);
1195         xhci_dbg(xhci, "udev->ttport = 0x%x\n", udev->ttport);
1196
1197         /* Step 4 - ring already allocated */
1198         /* Step 5 */
1199         ep0_ctx->ep_info2 = cpu_to_le32(EP_TYPE(CTRL_EP));
1200
1201         /* EP 0 can handle "burst" sizes of 1, so Max Burst Size field is 0 */
1202         ep0_ctx->ep_info2 |= cpu_to_le32(MAX_BURST(0) | ERROR_COUNT(3) |
1203                                          max_packets);
1204
1205         ep0_ctx->deq = cpu_to_le64(dev->eps[0].ring->first_seg->dma |
1206                                    dev->eps[0].ring->cycle_state);
1207
1208         trace_xhci_setup_addressable_virt_device(dev);
1209
1210         /* Steps 7 and 8 were done in xhci_alloc_virt_device() */
1211
1212         return 0;
1213 }
1214
1215 /*
1216  * Convert interval expressed as 2^(bInterval - 1) == interval into
1217  * straight exponent value 2^n == interval.
1218  *
1219  */
1220 static unsigned int xhci_parse_exponent_interval(struct usb_device *udev,
1221                 struct usb_host_endpoint *ep)
1222 {
1223         unsigned int interval;
1224
1225         interval = clamp_val(ep->desc.bInterval, 1, 16) - 1;
1226         if (interval != ep->desc.bInterval - 1)
1227                 dev_warn(&udev->dev,
1228                          "ep %#x - rounding interval to %d %sframes\n",
1229                          ep->desc.bEndpointAddress,
1230                          1 << interval,
1231                          udev->speed == USB_SPEED_FULL ? "" : "micro");
1232
1233         if (udev->speed == USB_SPEED_FULL) {
1234                 /*
1235                  * Full speed isoc endpoints specify interval in frames,
1236                  * not microframes. We are using microframes everywhere,
1237                  * so adjust accordingly.
1238                  */
1239                 interval += 3;  /* 1 frame = 2^3 uframes */
1240         }
1241
1242         return interval;
1243 }
1244
1245 /*
1246  * Convert bInterval expressed in microframes (in 1-255 range) to exponent of
1247  * microframes, rounded down to nearest power of 2.
1248  */
1249 static unsigned int xhci_microframes_to_exponent(struct usb_device *udev,
1250                 struct usb_host_endpoint *ep, unsigned int desc_interval,
1251                 unsigned int min_exponent, unsigned int max_exponent)
1252 {
1253         unsigned int interval;
1254
1255         interval = fls(desc_interval) - 1;
1256         interval = clamp_val(interval, min_exponent, max_exponent);
1257         if ((1 << interval) != desc_interval)
1258                 dev_dbg(&udev->dev,
1259                          "ep %#x - rounding interval to %d microframes, ep desc says %d microframes\n",
1260                          ep->desc.bEndpointAddress,
1261                          1 << interval,
1262                          desc_interval);
1263
1264         return interval;
1265 }
1266
1267 static unsigned int xhci_parse_microframe_interval(struct usb_device *udev,
1268                 struct usb_host_endpoint *ep)
1269 {
1270         if (ep->desc.bInterval == 0)
1271                 return 0;
1272         return xhci_microframes_to_exponent(udev, ep,
1273                         ep->desc.bInterval, 0, 15);
1274 }
1275
1276
1277 static unsigned int xhci_parse_frame_interval(struct usb_device *udev,
1278                 struct usb_host_endpoint *ep)
1279 {
1280         return xhci_microframes_to_exponent(udev, ep,
1281                         ep->desc.bInterval * 8, 3, 10);
1282 }
1283
1284 /* Return the polling or NAK interval.
1285  *
1286  * The polling interval is expressed in "microframes".  If xHCI's Interval field
1287  * is set to N, it will service the endpoint every 2^(Interval)*125us.
1288  *
1289  * The NAK interval is one NAK per 1 to 255 microframes, or no NAKs if interval
1290  * is set to 0.
1291  */
1292 static unsigned int xhci_get_endpoint_interval(struct usb_device *udev,
1293                 struct usb_host_endpoint *ep)
1294 {
1295         unsigned int interval = 0;
1296
1297         switch (udev->speed) {
1298         case USB_SPEED_HIGH:
1299                 /* Max NAK rate */
1300                 if (usb_endpoint_xfer_control(&ep->desc) ||
1301                     usb_endpoint_xfer_bulk(&ep->desc)) {
1302                         interval = xhci_parse_microframe_interval(udev, ep);
1303                         break;
1304                 }
1305                 /* Fall through - SS and HS isoc/int have same decoding */
1306
1307         case USB_SPEED_SUPER_PLUS:
1308         case USB_SPEED_SUPER:
1309                 if (usb_endpoint_xfer_int(&ep->desc) ||
1310                     usb_endpoint_xfer_isoc(&ep->desc)) {
1311                         interval = xhci_parse_exponent_interval(udev, ep);
1312                 }
1313                 break;
1314
1315         case USB_SPEED_FULL:
1316                 if (usb_endpoint_xfer_isoc(&ep->desc)) {
1317                         interval = xhci_parse_exponent_interval(udev, ep);
1318                         break;
1319                 }
1320                 /*
1321                  * Fall through for interrupt endpoint interval decoding
1322                  * since it uses the same rules as low speed interrupt
1323                  * endpoints.
1324                  */
1325                 /* fall through */
1326
1327         case USB_SPEED_LOW:
1328                 if (usb_endpoint_xfer_int(&ep->desc) ||
1329                     usb_endpoint_xfer_isoc(&ep->desc)) {
1330
1331                         interval = xhci_parse_frame_interval(udev, ep);
1332                 }
1333                 break;
1334
1335         default:
1336                 BUG();
1337         }
1338         return interval;
1339 }
1340
1341 /* The "Mult" field in the endpoint context is only set for SuperSpeed isoc eps.
1342  * High speed endpoint descriptors can define "the number of additional
1343  * transaction opportunities per microframe", but that goes in the Max Burst
1344  * endpoint context field.
1345  */
1346 static u32 xhci_get_endpoint_mult(struct usb_device *udev,
1347                 struct usb_host_endpoint *ep)
1348 {
1349         if (udev->speed < USB_SPEED_SUPER ||
1350                         !usb_endpoint_xfer_isoc(&ep->desc))
1351                 return 0;
1352         return ep->ss_ep_comp.bmAttributes;
1353 }
1354
1355 static u32 xhci_get_endpoint_max_burst(struct usb_device *udev,
1356                                        struct usb_host_endpoint *ep)
1357 {
1358         /* Super speed and Plus have max burst in ep companion desc */
1359         if (udev->speed >= USB_SPEED_SUPER)
1360                 return ep->ss_ep_comp.bMaxBurst;
1361
1362         if (udev->speed == USB_SPEED_HIGH &&
1363             (usb_endpoint_xfer_isoc(&ep->desc) ||
1364              usb_endpoint_xfer_int(&ep->desc)))
1365                 return usb_endpoint_maxp_mult(&ep->desc) - 1;
1366
1367         return 0;
1368 }
1369
1370 static u32 xhci_get_endpoint_type(struct usb_host_endpoint *ep)
1371 {
1372         int in;
1373
1374         in = usb_endpoint_dir_in(&ep->desc);
1375
1376         switch (usb_endpoint_type(&ep->desc)) {
1377         case USB_ENDPOINT_XFER_CONTROL:
1378                 return CTRL_EP;
1379         case USB_ENDPOINT_XFER_BULK:
1380                 return in ? BULK_IN_EP : BULK_OUT_EP;
1381         case USB_ENDPOINT_XFER_ISOC:
1382                 return in ? ISOC_IN_EP : ISOC_OUT_EP;
1383         case USB_ENDPOINT_XFER_INT:
1384                 return in ? INT_IN_EP : INT_OUT_EP;
1385         }
1386         return 0;
1387 }
1388
1389 /* Return the maximum endpoint service interval time (ESIT) payload.
1390  * Basically, this is the maxpacket size, multiplied by the burst size
1391  * and mult size.
1392  */
1393 static u32 xhci_get_max_esit_payload(struct usb_device *udev,
1394                 struct usb_host_endpoint *ep)
1395 {
1396         int max_burst;
1397         int max_packet;
1398
1399         /* Only applies for interrupt or isochronous endpoints */
1400         if (usb_endpoint_xfer_control(&ep->desc) ||
1401                         usb_endpoint_xfer_bulk(&ep->desc))
1402                 return 0;
1403
1404         /* SuperSpeedPlus Isoc ep sending over 48k per esit */
1405         if ((udev->speed >= USB_SPEED_SUPER_PLUS) &&
1406             USB_SS_SSP_ISOC_COMP(ep->ss_ep_comp.bmAttributes))
1407                 return le32_to_cpu(ep->ssp_isoc_ep_comp.dwBytesPerInterval);
1408         /* SuperSpeed or SuperSpeedPlus Isoc ep with less than 48k per esit */
1409         else if (udev->speed >= USB_SPEED_SUPER)
1410                 return le16_to_cpu(ep->ss_ep_comp.wBytesPerInterval);
1411
1412         max_packet = usb_endpoint_maxp(&ep->desc);
1413         max_burst = usb_endpoint_maxp_mult(&ep->desc);
1414         /* A 0 in max burst means 1 transfer per ESIT */
1415         return max_packet * max_burst;
1416 }
1417
1418 /* Set up an endpoint with one ring segment.  Do not allocate stream rings.
1419  * Drivers will have to call usb_alloc_streams() to do that.
1420  */
1421 int xhci_endpoint_init(struct xhci_hcd *xhci,
1422                 struct xhci_virt_device *virt_dev,
1423                 struct usb_device *udev,
1424                 struct usb_host_endpoint *ep,
1425                 gfp_t mem_flags)
1426 {
1427         unsigned int ep_index;
1428         struct xhci_ep_ctx *ep_ctx;
1429         struct xhci_ring *ep_ring;
1430         unsigned int max_packet;
1431         enum xhci_ring_type ring_type;
1432         u32 max_esit_payload;
1433         u32 endpoint_type;
1434         unsigned int max_burst;
1435         unsigned int interval;
1436         unsigned int mult;
1437         unsigned int avg_trb_len;
1438         unsigned int err_count = 0;
1439
1440         ep_index = xhci_get_endpoint_index(&ep->desc);
1441         ep_ctx = xhci_get_ep_ctx(xhci, virt_dev->in_ctx, ep_index);
1442
1443         endpoint_type = xhci_get_endpoint_type(ep);
1444         if (!endpoint_type)
1445                 return -EINVAL;
1446
1447         ring_type = usb_endpoint_type(&ep->desc);
1448
1449         /*
1450          * Get values to fill the endpoint context, mostly from ep descriptor.
1451          * The average TRB buffer lengt for bulk endpoints is unclear as we
1452          * have no clue on scatter gather list entry size. For Isoc and Int,
1453          * set it to max available. See xHCI 1.1 spec 4.14.1.1 for details.
1454          */
1455         max_esit_payload = xhci_get_max_esit_payload(udev, ep);
1456         interval = xhci_get_endpoint_interval(udev, ep);
1457
1458         /* Periodic endpoint bInterval limit quirk */
1459         if (usb_endpoint_xfer_int(&ep->desc) ||
1460             usb_endpoint_xfer_isoc(&ep->desc)) {
1461                 if ((xhci->quirks & XHCI_LIMIT_ENDPOINT_INTERVAL_7) &&
1462                     udev->speed >= USB_SPEED_HIGH &&
1463                     interval >= 7) {
1464                         interval = 6;
1465                 }
1466         }
1467
1468         mult = xhci_get_endpoint_mult(udev, ep);
1469         max_packet = usb_endpoint_maxp(&ep->desc);
1470         max_burst = xhci_get_endpoint_max_burst(udev, ep);
1471         avg_trb_len = max_esit_payload;
1472
1473         /* FIXME dig Mult and streams info out of ep companion desc */
1474
1475         /* Allow 3 retries for everything but isoc, set CErr = 3 */
1476         if (!usb_endpoint_xfer_isoc(&ep->desc))
1477                 err_count = 3;
1478         /* Some devices get this wrong */
1479         if (usb_endpoint_xfer_bulk(&ep->desc) && udev->speed == USB_SPEED_HIGH)
1480                 max_packet = 512;
1481         /* xHCI 1.0 and 1.1 indicates that ctrl ep avg TRB Length should be 8 */
1482         if (usb_endpoint_xfer_control(&ep->desc) && xhci->hci_version >= 0x100)
1483                 avg_trb_len = 8;
1484         /* xhci 1.1 with LEC support doesn't use mult field, use RsvdZ */
1485         if ((xhci->hci_version > 0x100) && HCC2_LEC(xhci->hcc_params2))
1486                 mult = 0;
1487
1488         /* Set up the endpoint ring */
1489         virt_dev->eps[ep_index].new_ring =
1490                 xhci_ring_alloc(xhci, 2, 1, ring_type, max_packet, mem_flags);
1491         if (!virt_dev->eps[ep_index].new_ring)
1492                 return -ENOMEM;
1493
1494         virt_dev->eps[ep_index].skip = false;
1495         ep_ring = virt_dev->eps[ep_index].new_ring;
1496
1497         /* Fill the endpoint context */
1498         ep_ctx->ep_info = cpu_to_le32(EP_MAX_ESIT_PAYLOAD_HI(max_esit_payload) |
1499                                       EP_INTERVAL(interval) |
1500                                       EP_MULT(mult));
1501         ep_ctx->ep_info2 = cpu_to_le32(EP_TYPE(endpoint_type) |
1502                                        MAX_PACKET(max_packet) |
1503                                        MAX_BURST(max_burst) |
1504                                        ERROR_COUNT(err_count));
1505         ep_ctx->deq = cpu_to_le64(ep_ring->first_seg->dma |
1506                                   ep_ring->cycle_state);
1507
1508         ep_ctx->tx_info = cpu_to_le32(EP_MAX_ESIT_PAYLOAD_LO(max_esit_payload) |
1509                                       EP_AVG_TRB_LENGTH(avg_trb_len));
1510
1511         return 0;
1512 }
1513
1514 void xhci_endpoint_zero(struct xhci_hcd *xhci,
1515                 struct xhci_virt_device *virt_dev,
1516                 struct usb_host_endpoint *ep)
1517 {
1518         unsigned int ep_index;
1519         struct xhci_ep_ctx *ep_ctx;
1520
1521         ep_index = xhci_get_endpoint_index(&ep->desc);
1522         ep_ctx = xhci_get_ep_ctx(xhci, virt_dev->in_ctx, ep_index);
1523
1524         ep_ctx->ep_info = 0;
1525         ep_ctx->ep_info2 = 0;
1526         ep_ctx->deq = 0;
1527         ep_ctx->tx_info = 0;
1528         /* Don't free the endpoint ring until the set interface or configuration
1529          * request succeeds.
1530          */
1531 }
1532
1533 void xhci_clear_endpoint_bw_info(struct xhci_bw_info *bw_info)
1534 {
1535         bw_info->ep_interval = 0;
1536         bw_info->mult = 0;
1537         bw_info->num_packets = 0;
1538         bw_info->max_packet_size = 0;
1539         bw_info->type = 0;
1540         bw_info->max_esit_payload = 0;
1541 }
1542
1543 void xhci_update_bw_info(struct xhci_hcd *xhci,
1544                 struct xhci_container_ctx *in_ctx,
1545                 struct xhci_input_control_ctx *ctrl_ctx,
1546                 struct xhci_virt_device *virt_dev)
1547 {
1548         struct xhci_bw_info *bw_info;
1549         struct xhci_ep_ctx *ep_ctx;
1550         unsigned int ep_type;
1551         int i;
1552
1553         for (i = 1; i < 31; i++) {
1554                 bw_info = &virt_dev->eps[i].bw_info;
1555
1556                 /* We can't tell what endpoint type is being dropped, but
1557                  * unconditionally clearing the bandwidth info for non-periodic
1558                  * endpoints should be harmless because the info will never be
1559                  * set in the first place.
1560                  */
1561                 if (!EP_IS_ADDED(ctrl_ctx, i) && EP_IS_DROPPED(ctrl_ctx, i)) {
1562                         /* Dropped endpoint */
1563                         xhci_clear_endpoint_bw_info(bw_info);
1564                         continue;
1565                 }
1566
1567                 if (EP_IS_ADDED(ctrl_ctx, i)) {
1568                         ep_ctx = xhci_get_ep_ctx(xhci, in_ctx, i);
1569                         ep_type = CTX_TO_EP_TYPE(le32_to_cpu(ep_ctx->ep_info2));
1570
1571                         /* Ignore non-periodic endpoints */
1572                         if (ep_type != ISOC_OUT_EP && ep_type != INT_OUT_EP &&
1573                                         ep_type != ISOC_IN_EP &&
1574                                         ep_type != INT_IN_EP)
1575                                 continue;
1576
1577                         /* Added or changed endpoint */
1578                         bw_info->ep_interval = CTX_TO_EP_INTERVAL(
1579                                         le32_to_cpu(ep_ctx->ep_info));
1580                         /* Number of packets and mult are zero-based in the
1581                          * input context, but we want one-based for the
1582                          * interval table.
1583                          */
1584                         bw_info->mult = CTX_TO_EP_MULT(
1585                                         le32_to_cpu(ep_ctx->ep_info)) + 1;
1586                         bw_info->num_packets = CTX_TO_MAX_BURST(
1587                                         le32_to_cpu(ep_ctx->ep_info2)) + 1;
1588                         bw_info->max_packet_size = MAX_PACKET_DECODED(
1589                                         le32_to_cpu(ep_ctx->ep_info2));
1590                         bw_info->type = ep_type;
1591                         bw_info->max_esit_payload = CTX_TO_MAX_ESIT_PAYLOAD(
1592                                         le32_to_cpu(ep_ctx->tx_info));
1593                 }
1594         }
1595 }
1596
1597 /* Copy output xhci_ep_ctx to the input xhci_ep_ctx copy.
1598  * Useful when you want to change one particular aspect of the endpoint and then
1599  * issue a configure endpoint command.
1600  */
1601 void xhci_endpoint_copy(struct xhci_hcd *xhci,
1602                 struct xhci_container_ctx *in_ctx,
1603                 struct xhci_container_ctx *out_ctx,
1604                 unsigned int ep_index)
1605 {
1606         struct xhci_ep_ctx *out_ep_ctx;
1607         struct xhci_ep_ctx *in_ep_ctx;
1608
1609         out_ep_ctx = xhci_get_ep_ctx(xhci, out_ctx, ep_index);
1610         in_ep_ctx = xhci_get_ep_ctx(xhci, in_ctx, ep_index);
1611
1612         in_ep_ctx->ep_info = out_ep_ctx->ep_info;
1613         in_ep_ctx->ep_info2 = out_ep_ctx->ep_info2;
1614         in_ep_ctx->deq = out_ep_ctx->deq;
1615         in_ep_ctx->tx_info = out_ep_ctx->tx_info;
1616 }
1617
1618 /* Copy output xhci_slot_ctx to the input xhci_slot_ctx.
1619  * Useful when you want to change one particular aspect of the endpoint and then
1620  * issue a configure endpoint command.  Only the context entries field matters,
1621  * but we'll copy the whole thing anyway.
1622  */
1623 void xhci_slot_copy(struct xhci_hcd *xhci,
1624                 struct xhci_container_ctx *in_ctx,
1625                 struct xhci_container_ctx *out_ctx)
1626 {
1627         struct xhci_slot_ctx *in_slot_ctx;
1628         struct xhci_slot_ctx *out_slot_ctx;
1629
1630         in_slot_ctx = xhci_get_slot_ctx(xhci, in_ctx);
1631         out_slot_ctx = xhci_get_slot_ctx(xhci, out_ctx);
1632
1633         in_slot_ctx->dev_info = out_slot_ctx->dev_info;
1634         in_slot_ctx->dev_info2 = out_slot_ctx->dev_info2;
1635         in_slot_ctx->tt_info = out_slot_ctx->tt_info;
1636         in_slot_ctx->dev_state = out_slot_ctx->dev_state;
1637 }
1638
1639 /* Set up the scratchpad buffer array and scratchpad buffers, if needed. */
1640 static int scratchpad_alloc(struct xhci_hcd *xhci, gfp_t flags)
1641 {
1642         int i;
1643         struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
1644         int num_sp = HCS_MAX_SCRATCHPAD(xhci->hcs_params2);
1645
1646         xhci_dbg_trace(xhci, trace_xhci_dbg_init,
1647                         "Allocating %d scratchpad buffers", num_sp);
1648
1649         if (!num_sp)
1650                 return 0;
1651
1652         xhci->scratchpad = kzalloc_node(sizeof(*xhci->scratchpad), flags,
1653                                 dev_to_node(dev));
1654         if (!xhci->scratchpad)
1655                 goto fail_sp;
1656
1657         xhci->scratchpad->sp_array = dma_alloc_coherent(dev,
1658                                      num_sp * sizeof(u64),
1659                                      &xhci->scratchpad->sp_dma, flags);
1660         if (!xhci->scratchpad->sp_array)
1661                 goto fail_sp2;
1662
1663         xhci->scratchpad->sp_buffers = kcalloc_node(num_sp, sizeof(void *),
1664                                         flags, dev_to_node(dev));
1665         if (!xhci->scratchpad->sp_buffers)
1666                 goto fail_sp3;
1667
1668         xhci->dcbaa->dev_context_ptrs[0] = cpu_to_le64(xhci->scratchpad->sp_dma);
1669         for (i = 0; i < num_sp; i++) {
1670                 dma_addr_t dma;
1671                 void *buf = dma_zalloc_coherent(dev, xhci->page_size, &dma,
1672                                 flags);
1673                 if (!buf)
1674                         goto fail_sp4;
1675
1676                 xhci->scratchpad->sp_array[i] = dma;
1677                 xhci->scratchpad->sp_buffers[i] = buf;
1678         }
1679
1680         return 0;
1681
1682  fail_sp4:
1683         for (i = i - 1; i >= 0; i--) {
1684                 dma_free_coherent(dev, xhci->page_size,
1685                                     xhci->scratchpad->sp_buffers[i],
1686                                     xhci->scratchpad->sp_array[i]);
1687         }
1688
1689         kfree(xhci->scratchpad->sp_buffers);
1690
1691  fail_sp3:
1692         dma_free_coherent(dev, num_sp * sizeof(u64),
1693                             xhci->scratchpad->sp_array,
1694                             xhci->scratchpad->sp_dma);
1695
1696  fail_sp2:
1697         kfree(xhci->scratchpad);
1698         xhci->scratchpad = NULL;
1699
1700  fail_sp:
1701         return -ENOMEM;
1702 }
1703
1704 static void scratchpad_free(struct xhci_hcd *xhci)
1705 {
1706         int num_sp;
1707         int i;
1708         struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
1709
1710         if (!xhci->scratchpad)
1711                 return;
1712
1713         num_sp = HCS_MAX_SCRATCHPAD(xhci->hcs_params2);
1714
1715         for (i = 0; i < num_sp; i++) {
1716                 dma_free_coherent(dev, xhci->page_size,
1717                                     xhci->scratchpad->sp_buffers[i],
1718                                     xhci->scratchpad->sp_array[i]);
1719         }
1720         kfree(xhci->scratchpad->sp_buffers);
1721         dma_free_coherent(dev, num_sp * sizeof(u64),
1722                             xhci->scratchpad->sp_array,
1723                             xhci->scratchpad->sp_dma);
1724         kfree(xhci->scratchpad);
1725         xhci->scratchpad = NULL;
1726 }
1727
1728 struct xhci_command *xhci_alloc_command(struct xhci_hcd *xhci,
1729                 bool allocate_completion, gfp_t mem_flags)
1730 {
1731         struct xhci_command *command;
1732         struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
1733
1734         command = kzalloc_node(sizeof(*command), mem_flags, dev_to_node(dev));
1735         if (!command)
1736                 return NULL;
1737
1738         if (allocate_completion) {
1739                 command->completion =
1740                         kzalloc_node(sizeof(struct completion), mem_flags,
1741                                 dev_to_node(dev));
1742                 if (!command->completion) {
1743                         kfree(command);
1744                         return NULL;
1745                 }
1746                 init_completion(command->completion);
1747         }
1748
1749         command->status = 0;
1750         INIT_LIST_HEAD(&command->cmd_list);
1751         return command;
1752 }
1753
1754 struct xhci_command *xhci_alloc_command_with_ctx(struct xhci_hcd *xhci,
1755                 bool allocate_completion, gfp_t mem_flags)
1756 {
1757         struct xhci_command *command;
1758
1759         command = xhci_alloc_command(xhci, allocate_completion, mem_flags);
1760         if (!command)
1761                 return NULL;
1762
1763         command->in_ctx = xhci_alloc_container_ctx(xhci, XHCI_CTX_TYPE_INPUT,
1764                                                    mem_flags);
1765         if (!command->in_ctx) {
1766                 kfree(command->completion);
1767                 kfree(command);
1768                 return NULL;
1769         }
1770         return command;
1771 }
1772
1773 void xhci_urb_free_priv(struct urb_priv *urb_priv)
1774 {
1775         kfree(urb_priv);
1776 }
1777
1778 void xhci_free_command(struct xhci_hcd *xhci,
1779                 struct xhci_command *command)
1780 {
1781         xhci_free_container_ctx(xhci,
1782                         command->in_ctx);
1783         kfree(command->completion);
1784         kfree(command);
1785 }
1786
1787 int xhci_alloc_erst(struct xhci_hcd *xhci,
1788                     struct xhci_ring *evt_ring,
1789                     struct xhci_erst *erst,
1790                     gfp_t flags)
1791 {
1792         size_t size;
1793         unsigned int val;
1794         struct xhci_segment *seg;
1795         struct xhci_erst_entry *entry;
1796
1797         size = sizeof(struct xhci_erst_entry) * evt_ring->num_segs;
1798         erst->entries = dma_zalloc_coherent(xhci_to_hcd(xhci)->self.sysdev,
1799                                             size, &erst->erst_dma_addr, flags);
1800         if (!erst->entries)
1801                 return -ENOMEM;
1802
1803         erst->num_entries = evt_ring->num_segs;
1804
1805         seg = evt_ring->first_seg;
1806         for (val = 0; val < evt_ring->num_segs; val++) {
1807                 entry = &erst->entries[val];
1808                 entry->seg_addr = cpu_to_le64(seg->dma);
1809                 entry->seg_size = cpu_to_le32(TRBS_PER_SEGMENT);
1810                 entry->rsvd = 0;
1811                 seg = seg->next;
1812         }
1813
1814         return 0;
1815 }
1816
1817 void xhci_free_erst(struct xhci_hcd *xhci, struct xhci_erst *erst)
1818 {
1819         size_t size;
1820         struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
1821
1822         size = sizeof(struct xhci_erst_entry) * (erst->num_entries);
1823         if (erst->entries)
1824                 dma_free_coherent(dev, size,
1825                                 erst->entries,
1826                                 erst->erst_dma_addr);
1827         erst->entries = NULL;
1828 }
1829
1830 void xhci_mem_cleanup(struct xhci_hcd *xhci)
1831 {
1832         struct device   *dev = xhci_to_hcd(xhci)->self.sysdev;
1833         int i, j, num_ports;
1834
1835         cancel_delayed_work_sync(&xhci->cmd_timer);
1836
1837         xhci_free_erst(xhci, &xhci->erst);
1838
1839         if (xhci->event_ring)
1840                 xhci_ring_free(xhci, xhci->event_ring);
1841         xhci->event_ring = NULL;
1842         xhci_dbg_trace(xhci, trace_xhci_dbg_init, "Freed event ring");
1843
1844         if (xhci->lpm_command)
1845                 xhci_free_command(xhci, xhci->lpm_command);
1846         xhci->lpm_command = NULL;
1847         if (xhci->cmd_ring)
1848                 xhci_ring_free(xhci, xhci->cmd_ring);
1849         xhci->cmd_ring = NULL;
1850         xhci_dbg_trace(xhci, trace_xhci_dbg_init, "Freed command ring");
1851         xhci_cleanup_command_queue(xhci);
1852
1853         num_ports = HCS_MAX_PORTS(xhci->hcs_params1);
1854         for (i = 0; i < num_ports && xhci->rh_bw; i++) {
1855                 struct xhci_interval_bw_table *bwt = &xhci->rh_bw[i].bw_table;
1856                 for (j = 0; j < XHCI_MAX_INTERVAL; j++) {
1857                         struct list_head *ep = &bwt->interval_bw[j].endpoints;
1858                         while (!list_empty(ep))
1859                                 list_del_init(ep->next);
1860                 }
1861         }
1862
1863         for (i = HCS_MAX_SLOTS(xhci->hcs_params1); i > 0; i--)
1864                 xhci_free_virt_devices_depth_first(xhci, i);
1865
1866         dma_pool_destroy(xhci->segment_pool);
1867         xhci->segment_pool = NULL;
1868         xhci_dbg_trace(xhci, trace_xhci_dbg_init, "Freed segment pool");
1869
1870         dma_pool_destroy(xhci->device_pool);
1871         xhci->device_pool = NULL;
1872         xhci_dbg_trace(xhci, trace_xhci_dbg_init, "Freed device context pool");
1873
1874         dma_pool_destroy(xhci->small_streams_pool);
1875         xhci->small_streams_pool = NULL;
1876         xhci_dbg_trace(xhci, trace_xhci_dbg_init,
1877                         "Freed small stream array pool");
1878
1879         dma_pool_destroy(xhci->medium_streams_pool);
1880         xhci->medium_streams_pool = NULL;
1881         xhci_dbg_trace(xhci, trace_xhci_dbg_init,
1882                         "Freed medium stream array pool");
1883
1884         if (xhci->dcbaa)
1885                 dma_free_coherent(dev, sizeof(*xhci->dcbaa),
1886                                 xhci->dcbaa, xhci->dcbaa->dma);
1887         xhci->dcbaa = NULL;
1888
1889         scratchpad_free(xhci);
1890
1891         if (!xhci->rh_bw)
1892                 goto no_bw;
1893
1894         for (i = 0; i < num_ports; i++) {
1895                 struct xhci_tt_bw_info *tt, *n;
1896                 list_for_each_entry_safe(tt, n, &xhci->rh_bw[i].tts, tt_list) {
1897                         list_del(&tt->tt_list);
1898                         kfree(tt);
1899                 }
1900         }
1901
1902 no_bw:
1903         xhci->cmd_ring_reserved_trbs = 0;
1904         xhci->usb2_rhub.num_ports = 0;
1905         xhci->usb3_rhub.num_ports = 0;
1906         xhci->num_active_eps = 0;
1907         kfree(xhci->usb2_rhub.ports);
1908         kfree(xhci->usb3_rhub.ports);
1909         kfree(xhci->hw_ports);
1910         kfree(xhci->rh_bw);
1911         kfree(xhci->ext_caps);
1912
1913         xhci->usb2_rhub.ports = NULL;
1914         xhci->usb3_rhub.ports = NULL;
1915         xhci->hw_ports = NULL;
1916         xhci->rh_bw = NULL;
1917         xhci->ext_caps = NULL;
1918
1919         xhci->page_size = 0;
1920         xhci->page_shift = 0;
1921         xhci->bus_state[0].bus_suspended = 0;
1922         xhci->bus_state[1].bus_suspended = 0;
1923 }
1924
1925 static int xhci_test_trb_in_td(struct xhci_hcd *xhci,
1926                 struct xhci_segment *input_seg,
1927                 union xhci_trb *start_trb,
1928                 union xhci_trb *end_trb,
1929                 dma_addr_t input_dma,
1930                 struct xhci_segment *result_seg,
1931                 char *test_name, int test_number)
1932 {
1933         unsigned long long start_dma;
1934         unsigned long long end_dma;
1935         struct xhci_segment *seg;
1936
1937         start_dma = xhci_trb_virt_to_dma(input_seg, start_trb);
1938         end_dma = xhci_trb_virt_to_dma(input_seg, end_trb);
1939
1940         seg = trb_in_td(xhci, input_seg, start_trb, end_trb, input_dma, false);
1941         if (seg != result_seg) {
1942                 xhci_warn(xhci, "WARN: %s TRB math test %d failed!\n",
1943                                 test_name, test_number);
1944                 xhci_warn(xhci, "Tested TRB math w/ seg %p and "
1945                                 "input DMA 0x%llx\n",
1946                                 input_seg,
1947                                 (unsigned long long) input_dma);
1948                 xhci_warn(xhci, "starting TRB %p (0x%llx DMA), "
1949                                 "ending TRB %p (0x%llx DMA)\n",
1950                                 start_trb, start_dma,
1951                                 end_trb, end_dma);
1952                 xhci_warn(xhci, "Expected seg %p, got seg %p\n",
1953                                 result_seg, seg);
1954                 trb_in_td(xhci, input_seg, start_trb, end_trb, input_dma,
1955                           true);
1956                 return -1;
1957         }
1958         return 0;
1959 }
1960
1961 /* TRB math checks for xhci_trb_in_td(), using the command and event rings. */
1962 static int xhci_check_trb_in_td_math(struct xhci_hcd *xhci)
1963 {
1964         struct {
1965                 dma_addr_t              input_dma;
1966                 struct xhci_segment     *result_seg;
1967         } simple_test_vector [] = {
1968                 /* A zeroed DMA field should fail */
1969                 { 0, NULL },
1970                 /* One TRB before the ring start should fail */
1971                 { xhci->event_ring->first_seg->dma - 16, NULL },
1972                 /* One byte before the ring start should fail */
1973                 { xhci->event_ring->first_seg->dma - 1, NULL },
1974                 /* Starting TRB should succeed */
1975                 { xhci->event_ring->first_seg->dma, xhci->event_ring->first_seg },
1976                 /* Ending TRB should succeed */
1977                 { xhci->event_ring->first_seg->dma + (TRBS_PER_SEGMENT - 1)*16,
1978                         xhci->event_ring->first_seg },
1979                 /* One byte after the ring end should fail */
1980                 { xhci->event_ring->first_seg->dma + (TRBS_PER_SEGMENT - 1)*16 + 1, NULL },
1981                 /* One TRB after the ring end should fail */
1982                 { xhci->event_ring->first_seg->dma + (TRBS_PER_SEGMENT)*16, NULL },
1983                 /* An address of all ones should fail */
1984                 { (dma_addr_t) (~0), NULL },
1985         };
1986         struct {
1987                 struct xhci_segment     *input_seg;
1988                 union xhci_trb          *start_trb;
1989                 union xhci_trb          *end_trb;
1990                 dma_addr_t              input_dma;
1991                 struct xhci_segment     *result_seg;
1992         } complex_test_vector [] = {
1993                 /* Test feeding a valid DMA address from a different ring */
1994                 {       .input_seg = xhci->event_ring->first_seg,
1995                         .start_trb = xhci->event_ring->first_seg->trbs,
1996                         .end_trb = &xhci->event_ring->first_seg->trbs[TRBS_PER_SEGMENT - 1],
1997                         .input_dma = xhci->cmd_ring->first_seg->dma,
1998                         .result_seg = NULL,
1999                 },
2000                 /* Test feeding a valid end TRB from a different ring */
2001                 {       .input_seg = xhci->event_ring->first_seg,
2002                         .start_trb = xhci->event_ring->first_seg->trbs,
2003                         .end_trb = &xhci->cmd_ring->first_seg->trbs[TRBS_PER_SEGMENT - 1],
2004                         .input_dma = xhci->cmd_ring->first_seg->dma,
2005                         .result_seg = NULL,
2006                 },
2007                 /* Test feeding a valid start and end TRB from a different ring */
2008                 {       .input_seg = xhci->event_ring->first_seg,
2009                         .start_trb = xhci->cmd_ring->first_seg->trbs,
2010                         .end_trb = &xhci->cmd_ring->first_seg->trbs[TRBS_PER_SEGMENT - 1],
2011                         .input_dma = xhci->cmd_ring->first_seg->dma,
2012                         .result_seg = NULL,
2013                 },
2014                 /* TRB in this ring, but after this TD */
2015                 {       .input_seg = xhci->event_ring->first_seg,
2016                         .start_trb = &xhci->event_ring->first_seg->trbs[0],
2017                         .end_trb = &xhci->event_ring->first_seg->trbs[3],
2018                         .input_dma = xhci->event_ring->first_seg->dma + 4*16,
2019                         .result_seg = NULL,
2020                 },
2021                 /* TRB in this ring, but before this TD */
2022                 {       .input_seg = xhci->event_ring->first_seg,
2023                         .start_trb = &xhci->event_ring->first_seg->trbs[3],
2024                         .end_trb = &xhci->event_ring->first_seg->trbs[6],
2025                         .input_dma = xhci->event_ring->first_seg->dma + 2*16,
2026                         .result_seg = NULL,
2027                 },
2028                 /* TRB in this ring, but after this wrapped TD */
2029                 {       .input_seg = xhci->event_ring->first_seg,
2030                         .start_trb = &xhci->event_ring->first_seg->trbs[TRBS_PER_SEGMENT - 3],
2031                         .end_trb = &xhci->event_ring->first_seg->trbs[1],
2032                         .input_dma = xhci->event_ring->first_seg->dma + 2*16,
2033                         .result_seg = NULL,
2034                 },
2035                 /* TRB in this ring, but before this wrapped TD */
2036                 {       .input_seg = xhci->event_ring->first_seg,
2037                         .start_trb = &xhci->event_ring->first_seg->trbs[TRBS_PER_SEGMENT - 3],
2038                         .end_trb = &xhci->event_ring->first_seg->trbs[1],
2039                         .input_dma = xhci->event_ring->first_seg->dma + (TRBS_PER_SEGMENT - 4)*16,
2040                         .result_seg = NULL,
2041                 },
2042                 /* TRB not in this ring, and we have a wrapped TD */
2043                 {       .input_seg = xhci->event_ring->first_seg,
2044                         .start_trb = &xhci->event_ring->first_seg->trbs[TRBS_PER_SEGMENT - 3],
2045                         .end_trb = &xhci->event_ring->first_seg->trbs[1],
2046                         .input_dma = xhci->cmd_ring->first_seg->dma + 2*16,
2047                         .result_seg = NULL,
2048                 },
2049         };
2050
2051         unsigned int num_tests;
2052         int i, ret;
2053
2054         num_tests = ARRAY_SIZE(simple_test_vector);
2055         for (i = 0; i < num_tests; i++) {
2056                 ret = xhci_test_trb_in_td(xhci,
2057                                 xhci->event_ring->first_seg,
2058                                 xhci->event_ring->first_seg->trbs,
2059                                 &xhci->event_ring->first_seg->trbs[TRBS_PER_SEGMENT - 1],
2060                                 simple_test_vector[i].input_dma,
2061                                 simple_test_vector[i].result_seg,
2062                                 "Simple", i);
2063                 if (ret < 0)
2064                         return ret;
2065         }
2066
2067         num_tests = ARRAY_SIZE(complex_test_vector);
2068         for (i = 0; i < num_tests; i++) {
2069                 ret = xhci_test_trb_in_td(xhci,
2070                                 complex_test_vector[i].input_seg,
2071                                 complex_test_vector[i].start_trb,
2072                                 complex_test_vector[i].end_trb,
2073                                 complex_test_vector[i].input_dma,
2074                                 complex_test_vector[i].result_seg,
2075                                 "Complex", i);
2076                 if (ret < 0)
2077                         return ret;
2078         }
2079         xhci_dbg(xhci, "TRB math tests passed.\n");
2080         return 0;
2081 }
2082
2083 static void xhci_set_hc_event_deq(struct xhci_hcd *xhci)
2084 {
2085         u64 temp;
2086         dma_addr_t deq;
2087
2088         deq = xhci_trb_virt_to_dma(xhci->event_ring->deq_seg,
2089                         xhci->event_ring->dequeue);
2090         if (deq == 0 && !in_interrupt())
2091                 xhci_warn(xhci, "WARN something wrong with SW event ring "
2092                                 "dequeue ptr.\n");
2093         /* Update HC event ring dequeue pointer */
2094         temp = xhci_read_64(xhci, &xhci->ir_set->erst_dequeue);
2095         temp &= ERST_PTR_MASK;
2096         /* Don't clear the EHB bit (which is RW1C) because
2097          * there might be more events to service.
2098          */
2099         temp &= ~ERST_EHB;
2100         xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2101                         "// Write event ring dequeue pointer, "
2102                         "preserving EHB bit");
2103         xhci_write_64(xhci, ((u64) deq & (u64) ~ERST_PTR_MASK) | temp,
2104                         &xhci->ir_set->erst_dequeue);
2105 }
2106
2107 static void xhci_add_in_port(struct xhci_hcd *xhci, unsigned int num_ports,
2108                 __le32 __iomem *addr, int max_caps)
2109 {
2110         u32 temp, port_offset, port_count;
2111         int i;
2112         u8 major_revision, minor_revision;
2113         struct xhci_hub *rhub;
2114         struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
2115
2116         temp = readl(addr);
2117         major_revision = XHCI_EXT_PORT_MAJOR(temp);
2118         minor_revision = XHCI_EXT_PORT_MINOR(temp);
2119
2120         if (major_revision == 0x03) {
2121                 rhub = &xhci->usb3_rhub;
2122         } else if (major_revision <= 0x02) {
2123                 rhub = &xhci->usb2_rhub;
2124         } else {
2125                 xhci_warn(xhci, "Ignoring unknown port speed, "
2126                                 "Ext Cap %p, revision = 0x%x\n",
2127                                 addr, major_revision);
2128                 /* Ignoring port protocol we can't understand. FIXME */
2129                 return;
2130         }
2131         rhub->maj_rev = XHCI_EXT_PORT_MAJOR(temp);
2132
2133         if (rhub->min_rev < minor_revision)
2134                 rhub->min_rev = minor_revision;
2135
2136         /* Port offset and count in the third dword, see section 7.2 */
2137         temp = readl(addr + 2);
2138         port_offset = XHCI_EXT_PORT_OFF(temp);
2139         port_count = XHCI_EXT_PORT_COUNT(temp);
2140         xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2141                         "Ext Cap %p, port offset = %u, "
2142                         "count = %u, revision = 0x%x",
2143                         addr, port_offset, port_count, major_revision);
2144         /* Port count includes the current port offset */
2145         if (port_offset == 0 || (port_offset + port_count - 1) > num_ports)
2146                 /* WTF? "Valid values are â€˜1’ to MaxPorts" */
2147                 return;
2148
2149         rhub->psi_count = XHCI_EXT_PORT_PSIC(temp);
2150         if (rhub->psi_count) {
2151                 rhub->psi = kcalloc_node(rhub->psi_count, sizeof(*rhub->psi),
2152                                     GFP_KERNEL, dev_to_node(dev));
2153                 if (!rhub->psi)
2154                         rhub->psi_count = 0;
2155
2156                 rhub->psi_uid_count++;
2157                 for (i = 0; i < rhub->psi_count; i++) {
2158                         rhub->psi[i] = readl(addr + 4 + i);
2159
2160                         /* count unique ID values, two consecutive entries can
2161                          * have the same ID if link is assymetric
2162                          */
2163                         if (i && (XHCI_EXT_PORT_PSIV(rhub->psi[i]) !=
2164                                   XHCI_EXT_PORT_PSIV(rhub->psi[i - 1])))
2165                                 rhub->psi_uid_count++;
2166
2167                         xhci_dbg(xhci, "PSIV:%d PSIE:%d PLT:%d PFD:%d LP:%d PSIM:%d\n",
2168                                   XHCI_EXT_PORT_PSIV(rhub->psi[i]),
2169                                   XHCI_EXT_PORT_PSIE(rhub->psi[i]),
2170                                   XHCI_EXT_PORT_PLT(rhub->psi[i]),
2171                                   XHCI_EXT_PORT_PFD(rhub->psi[i]),
2172                                   XHCI_EXT_PORT_LP(rhub->psi[i]),
2173                                   XHCI_EXT_PORT_PSIM(rhub->psi[i]));
2174                 }
2175         }
2176         /* cache usb2 port capabilities */
2177         if (major_revision < 0x03 && xhci->num_ext_caps < max_caps)
2178                 xhci->ext_caps[xhci->num_ext_caps++] = temp;
2179
2180         /* Check the host's USB2 LPM capability */
2181         if ((xhci->hci_version == 0x96) && (major_revision != 0x03) &&
2182                         (temp & XHCI_L1C)) {
2183                 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2184                                 "xHCI 0.96: support USB2 software lpm");
2185                 xhci->sw_lpm_support = 1;
2186         }
2187
2188         if ((xhci->hci_version >= 0x100) && (major_revision != 0x03)) {
2189                 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2190                                 "xHCI 1.0: support USB2 software lpm");
2191                 xhci->sw_lpm_support = 1;
2192                 if (temp & XHCI_HLC) {
2193                         xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2194                                         "xHCI 1.0: support USB2 hardware lpm");
2195                         xhci->hw_lpm_support = 1;
2196                 }
2197         }
2198
2199         port_offset--;
2200         for (i = port_offset; i < (port_offset + port_count); i++) {
2201                 struct xhci_port *hw_port = &xhci->hw_ports[i];
2202                 /* Duplicate entry.  Ignore the port if the revisions differ. */
2203                 if (hw_port->rhub) {
2204                         xhci_warn(xhci, "Duplicate port entry, Ext Cap %p,"
2205                                         " port %u\n", addr, i);
2206                         xhci_warn(xhci, "Port was marked as USB %u, "
2207                                         "duplicated as USB %u\n",
2208                                         hw_port->rhub->maj_rev, major_revision);
2209                         /* Only adjust the roothub port counts if we haven't
2210                          * found a similar duplicate.
2211                          */
2212                         if (hw_port->rhub != rhub &&
2213                                  hw_port->hcd_portnum != DUPLICATE_ENTRY) {
2214                                 hw_port->rhub->num_ports--;
2215                                 hw_port->hcd_portnum = DUPLICATE_ENTRY;
2216                         }
2217                         continue;
2218                 }
2219                 hw_port->rhub = rhub;
2220                 rhub->num_ports++;
2221         }
2222         /* FIXME: Should we disable ports not in the Extended Capabilities? */
2223 }
2224
2225 static void xhci_create_rhub_port_array(struct xhci_hcd *xhci,
2226                                         struct xhci_hub *rhub, gfp_t flags)
2227 {
2228         int port_index = 0;
2229         int i;
2230         struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
2231
2232         if (!rhub->num_ports)
2233                 return;
2234         rhub->ports = kcalloc_node(rhub->num_ports, sizeof(rhub->ports), flags,
2235                         dev_to_node(dev));
2236         for (i = 0; i < HCS_MAX_PORTS(xhci->hcs_params1); i++) {
2237                 if (xhci->hw_ports[i].rhub != rhub ||
2238                     xhci->hw_ports[i].hcd_portnum == DUPLICATE_ENTRY)
2239                         continue;
2240                 xhci->hw_ports[i].hcd_portnum = port_index;
2241                 rhub->ports[port_index] = &xhci->hw_ports[i];
2242                 port_index++;
2243                 if (port_index == rhub->num_ports)
2244                         break;
2245         }
2246 }
2247
2248 /*
2249  * Scan the Extended Capabilities for the "Supported Protocol Capabilities" that
2250  * specify what speeds each port is supposed to be.  We can't count on the port
2251  * speed bits in the PORTSC register being correct until a device is connected,
2252  * but we need to set up the two fake roothubs with the correct number of USB
2253  * 3.0 and USB 2.0 ports at host controller initialization time.
2254  */
2255 static int xhci_setup_port_arrays(struct xhci_hcd *xhci, gfp_t flags)
2256 {
2257         void __iomem *base;
2258         u32 offset;
2259         unsigned int num_ports;
2260         int i, j;
2261         int cap_count = 0;
2262         u32 cap_start;
2263         struct device *dev = xhci_to_hcd(xhci)->self.sysdev;
2264
2265         num_ports = HCS_MAX_PORTS(xhci->hcs_params1);
2266         xhci->hw_ports = kcalloc_node(num_ports, sizeof(*xhci->hw_ports),
2267                                 flags, dev_to_node(dev));
2268         if (!xhci->hw_ports)
2269                 return -ENOMEM;
2270
2271         for (i = 0; i < num_ports; i++) {
2272                 xhci->hw_ports[i].addr = &xhci->op_regs->port_status_base +
2273                         NUM_PORT_REGS * i;
2274                 xhci->hw_ports[i].hw_portnum = i;
2275         }
2276
2277         xhci->rh_bw = kcalloc_node(num_ports, sizeof(*xhci->rh_bw), flags,
2278                                    dev_to_node(dev));
2279         if (!xhci->rh_bw)
2280                 return -ENOMEM;
2281         for (i = 0; i < num_ports; i++) {
2282                 struct xhci_interval_bw_table *bw_table;
2283
2284                 INIT_LIST_HEAD(&xhci->rh_bw[i].tts);
2285                 bw_table = &xhci->rh_bw[i].bw_table;
2286                 for (j = 0; j < XHCI_MAX_INTERVAL; j++)
2287                         INIT_LIST_HEAD(&bw_table->interval_bw[j].endpoints);
2288         }
2289         base = &xhci->cap_regs->hc_capbase;
2290
2291         cap_start = xhci_find_next_ext_cap(base, 0, XHCI_EXT_CAPS_PROTOCOL);
2292         if (!cap_start) {
2293                 xhci_err(xhci, "No Extended Capability registers, unable to set up roothub\n");
2294                 return -ENODEV;
2295         }
2296
2297         offset = cap_start;
2298         /* count extended protocol capability entries for later caching */
2299         while (offset) {
2300                 cap_count++;
2301                 offset = xhci_find_next_ext_cap(base, offset,
2302                                                       XHCI_EXT_CAPS_PROTOCOL);
2303         }
2304
2305         xhci->ext_caps = kcalloc_node(cap_count, sizeof(*xhci->ext_caps),
2306                                 flags, dev_to_node(dev));
2307         if (!xhci->ext_caps)
2308                 return -ENOMEM;
2309
2310         offset = cap_start;
2311
2312         while (offset) {
2313                 xhci_add_in_port(xhci, num_ports, base + offset, cap_count);
2314                 if (xhci->usb2_rhub.num_ports + xhci->usb3_rhub.num_ports ==
2315                     num_ports)
2316                         break;
2317                 offset = xhci_find_next_ext_cap(base, offset,
2318                                                 XHCI_EXT_CAPS_PROTOCOL);
2319         }
2320         if (xhci->usb2_rhub.num_ports == 0 && xhci->usb3_rhub.num_ports == 0) {
2321                 xhci_warn(xhci, "No ports on the roothubs?\n");
2322                 return -ENODEV;
2323         }
2324         xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2325                        "Found %u USB 2.0 ports and %u USB 3.0 ports.",
2326                        xhci->usb2_rhub.num_ports, xhci->usb3_rhub.num_ports);
2327
2328         /* Place limits on the number of roothub ports so that the hub
2329          * descriptors aren't longer than the USB core will allocate.
2330          */
2331         if (xhci->usb3_rhub.num_ports > USB_SS_MAXPORTS) {
2332                 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2333                                 "Limiting USB 3.0 roothub ports to %u.",
2334                                 USB_SS_MAXPORTS);
2335                 xhci->usb3_rhub.num_ports = USB_SS_MAXPORTS;
2336         }
2337         if (xhci->usb2_rhub.num_ports > USB_MAXCHILDREN) {
2338                 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2339                                 "Limiting USB 2.0 roothub ports to %u.",
2340                                 USB_MAXCHILDREN);
2341                 xhci->usb2_rhub.num_ports = USB_MAXCHILDREN;
2342         }
2343
2344         /*
2345          * Note we could have all USB 3.0 ports, or all USB 2.0 ports.
2346          * Not sure how the USB core will handle a hub with no ports...
2347          */
2348
2349         xhci_create_rhub_port_array(xhci, &xhci->usb2_rhub, flags);
2350         xhci_create_rhub_port_array(xhci, &xhci->usb3_rhub, flags);
2351
2352         return 0;
2353 }
2354
2355 int xhci_mem_init(struct xhci_hcd *xhci, gfp_t flags)
2356 {
2357         dma_addr_t      dma;
2358         struct device   *dev = xhci_to_hcd(xhci)->self.sysdev;
2359         unsigned int    val, val2;
2360         u64             val_64;
2361         u32             page_size, temp;
2362         int             i, ret;
2363
2364         INIT_LIST_HEAD(&xhci->cmd_list);
2365
2366         /* init command timeout work */
2367         INIT_DELAYED_WORK(&xhci->cmd_timer, xhci_handle_command_timeout);
2368         init_completion(&xhci->cmd_ring_stop_completion);
2369
2370         page_size = readl(&xhci->op_regs->page_size);
2371         xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2372                         "Supported page size register = 0x%x", page_size);
2373         for (i = 0; i < 16; i++) {
2374                 if ((0x1 & page_size) != 0)
2375                         break;
2376                 page_size = page_size >> 1;
2377         }
2378         if (i < 16)
2379                 xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2380                         "Supported page size of %iK", (1 << (i+12)) / 1024);
2381         else
2382                 xhci_warn(xhci, "WARN: no supported page size\n");
2383         /* Use 4K pages, since that's common and the minimum the HC supports */
2384         xhci->page_shift = 12;
2385         xhci->page_size = 1 << xhci->page_shift;
2386         xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2387                         "HCD page size set to %iK", xhci->page_size / 1024);
2388
2389         /*
2390          * Program the Number of Device Slots Enabled field in the CONFIG
2391          * register with the max value of slots the HC can handle.
2392          */
2393         val = HCS_MAX_SLOTS(readl(&xhci->cap_regs->hcs_params1));
2394         xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2395                         "// xHC can handle at most %d device slots.", val);
2396         val2 = readl(&xhci->op_regs->config_reg);
2397         val |= (val2 & ~HCS_SLOTS_MASK);
2398         xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2399                         "// Setting Max device slots reg = 0x%x.", val);
2400         writel(val, &xhci->op_regs->config_reg);
2401
2402         /*
2403          * xHCI section 5.4.6 - doorbell array must be
2404          * "physically contiguous and 64-byte (cache line) aligned".
2405          */
2406         xhci->dcbaa = dma_alloc_coherent(dev, sizeof(*xhci->dcbaa), &dma,
2407                         flags);
2408         if (!xhci->dcbaa)
2409                 goto fail;
2410         memset(xhci->dcbaa, 0, sizeof *(xhci->dcbaa));
2411         xhci->dcbaa->dma = dma;
2412         xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2413                         "// Device context base array address = 0x%llx (DMA), %p (virt)",
2414                         (unsigned long long)xhci->dcbaa->dma, xhci->dcbaa);
2415         xhci_write_64(xhci, dma, &xhci->op_regs->dcbaa_ptr);
2416
2417         /*
2418          * Initialize the ring segment pool.  The ring must be a contiguous
2419          * structure comprised of TRBs.  The TRBs must be 16 byte aligned,
2420          * however, the command ring segment needs 64-byte aligned segments
2421          * and our use of dma addresses in the trb_address_map radix tree needs
2422          * TRB_SEGMENT_SIZE alignment, so we pick the greater alignment need.
2423          */
2424         xhci->segment_pool = dma_pool_create("xHCI ring segments", dev,
2425                         TRB_SEGMENT_SIZE, TRB_SEGMENT_SIZE, xhci->page_size);
2426
2427         /* See Table 46 and Note on Figure 55 */
2428         xhci->device_pool = dma_pool_create("xHCI input/output contexts", dev,
2429                         2112, 64, xhci->page_size);
2430         if (!xhci->segment_pool || !xhci->device_pool)
2431                 goto fail;
2432
2433         /* Linear stream context arrays don't have any boundary restrictions,
2434          * and only need to be 16-byte aligned.
2435          */
2436         xhci->small_streams_pool =
2437                 dma_pool_create("xHCI 256 byte stream ctx arrays",
2438                         dev, SMALL_STREAM_ARRAY_SIZE, 16, 0);
2439         xhci->medium_streams_pool =
2440                 dma_pool_create("xHCI 1KB stream ctx arrays",
2441                         dev, MEDIUM_STREAM_ARRAY_SIZE, 16, 0);
2442         /* Any stream context array bigger than MEDIUM_STREAM_ARRAY_SIZE
2443          * will be allocated with dma_alloc_coherent()
2444          */
2445
2446         if (!xhci->small_streams_pool || !xhci->medium_streams_pool)
2447                 goto fail;
2448
2449         /* Set up the command ring to have one segments for now. */
2450         xhci->cmd_ring = xhci_ring_alloc(xhci, 1, 1, TYPE_COMMAND, 0, flags);
2451         if (!xhci->cmd_ring)
2452                 goto fail;
2453         xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2454                         "Allocated command ring at %p", xhci->cmd_ring);
2455         xhci_dbg_trace(xhci, trace_xhci_dbg_init, "First segment DMA is 0x%llx",
2456                         (unsigned long long)xhci->cmd_ring->first_seg->dma);
2457
2458         /* Set the address in the Command Ring Control register */
2459         val_64 = xhci_read_64(xhci, &xhci->op_regs->cmd_ring);
2460         val_64 = (val_64 & (u64) CMD_RING_RSVD_BITS) |
2461                 (xhci->cmd_ring->first_seg->dma & (u64) ~CMD_RING_RSVD_BITS) |
2462                 xhci->cmd_ring->cycle_state;
2463         xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2464                         "// Setting command ring address to 0x%016llx", val_64);
2465         xhci_write_64(xhci, val_64, &xhci->op_regs->cmd_ring);
2466
2467         xhci->lpm_command = xhci_alloc_command_with_ctx(xhci, true, flags);
2468         if (!xhci->lpm_command)
2469                 goto fail;
2470
2471         /* Reserve one command ring TRB for disabling LPM.
2472          * Since the USB core grabs the shared usb_bus bandwidth mutex before
2473          * disabling LPM, we only need to reserve one TRB for all devices.
2474          */
2475         xhci->cmd_ring_reserved_trbs++;
2476
2477         val = readl(&xhci->cap_regs->db_off);
2478         val &= DBOFF_MASK;
2479         xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2480                         "// Doorbell array is located at offset 0x%x"
2481                         " from cap regs base addr", val);
2482         xhci->dba = (void __iomem *) xhci->cap_regs + val;
2483         /* Set ir_set to interrupt register set 0 */
2484         xhci->ir_set = &xhci->run_regs->ir_set[0];
2485
2486         /*
2487          * Event ring setup: Allocate a normal ring, but also setup
2488          * the event ring segment table (ERST).  Section 4.9.3.
2489          */
2490         xhci_dbg_trace(xhci, trace_xhci_dbg_init, "// Allocating event ring");
2491         xhci->event_ring = xhci_ring_alloc(xhci, ERST_NUM_SEGS, 1, TYPE_EVENT,
2492                                         0, flags);
2493         if (!xhci->event_ring)
2494                 goto fail;
2495         if (xhci_check_trb_in_td_math(xhci) < 0)
2496                 goto fail;
2497
2498         ret = xhci_alloc_erst(xhci, xhci->event_ring, &xhci->erst, flags);
2499         if (ret)
2500                 goto fail;
2501
2502         /* set ERST count with the number of entries in the segment table */
2503         val = readl(&xhci->ir_set->erst_size);
2504         val &= ERST_SIZE_MASK;
2505         val |= ERST_NUM_SEGS;
2506         xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2507                         "// Write ERST size = %i to ir_set 0 (some bits preserved)",
2508                         val);
2509         writel(val, &xhci->ir_set->erst_size);
2510
2511         xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2512                         "// Set ERST entries to point to event ring.");
2513         /* set the segment table base address */
2514         xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2515                         "// Set ERST base address for ir_set 0 = 0x%llx",
2516                         (unsigned long long)xhci->erst.erst_dma_addr);
2517         val_64 = xhci_read_64(xhci, &xhci->ir_set->erst_base);
2518         val_64 &= ERST_PTR_MASK;
2519         val_64 |= (xhci->erst.erst_dma_addr & (u64) ~ERST_PTR_MASK);
2520         xhci_write_64(xhci, val_64, &xhci->ir_set->erst_base);
2521
2522         /* Set the event ring dequeue address */
2523         xhci_set_hc_event_deq(xhci);
2524         xhci_dbg_trace(xhci, trace_xhci_dbg_init,
2525                         "Wrote ERST address to ir_set 0.");
2526
2527         /*
2528          * XXX: Might need to set the Interrupter Moderation Register to
2529          * something other than the default (~1ms minimum between interrupts).
2530          * See section 5.5.1.2.
2531          */
2532         for (i = 0; i < MAX_HC_SLOTS; i++)
2533                 xhci->devs[i] = NULL;
2534         for (i = 0; i < USB_MAXCHILDREN; i++) {
2535                 xhci->bus_state[0].resume_done[i] = 0;
2536                 xhci->bus_state[1].resume_done[i] = 0;
2537                 /* Only the USB 2.0 completions will ever be used. */
2538                 init_completion(&xhci->bus_state[1].rexit_done[i]);
2539         }
2540
2541         if (scratchpad_alloc(xhci, flags))
2542                 goto fail;
2543         if (xhci_setup_port_arrays(xhci, flags))
2544                 goto fail;
2545
2546         /* Enable USB 3.0 device notifications for function remote wake, which
2547          * is necessary for allowing USB 3.0 devices to do remote wakeup from
2548          * U3 (device suspend).
2549          */
2550         temp = readl(&xhci->op_regs->dev_notification);
2551         temp &= ~DEV_NOTE_MASK;
2552         temp |= DEV_NOTE_FWAKE;
2553         writel(temp, &xhci->op_regs->dev_notification);
2554
2555         return 0;
2556
2557 fail:
2558         xhci_halt(xhci);
2559         xhci_reset(xhci);
2560         xhci_mem_cleanup(xhci);
2561         return -ENOMEM;
2562 }