5018e33e0b34958a3e866d0ab226a3d408914d9f
[pandora-kernel.git] / drivers / usb / host / xhci-mem.c
1 /*
2  * xHCI host controller driver
3  *
4  * Copyright (C) 2008 Intel Corp.
5  *
6  * Author: Sarah Sharp
7  * Some code borrowed from the Linux EHCI driver.
8  *
9  * This program is free software; you can redistribute it and/or modify
10  * it under the terms of the GNU General Public License version 2 as
11  * published by the Free Software Foundation.
12  *
13  * This program is distributed in the hope that it will be useful, but
14  * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
15  * or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
16  * for more details.
17  *
18  * You should have received a copy of the GNU General Public License
19  * along with this program; if not, write to the Free Software Foundation,
20  * Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
21  */
22
23 #include <linux/usb.h>
24 #include <linux/pci.h>
25 #include <linux/slab.h>
26 #include <linux/dmapool.h>
27
28 #include "xhci.h"
29
30 /*
31  * Allocates a generic ring segment from the ring pool, sets the dma address,
32  * initializes the segment to zero, and sets the private next pointer to NULL.
33  *
34  * Section 4.11.1.1:
35  * "All components of all Command and Transfer TRBs shall be initialized to '0'"
36  */
37 static struct xhci_segment *xhci_segment_alloc(struct xhci_hcd *xhci, gfp_t flags)
38 {
39         struct xhci_segment *seg;
40         dma_addr_t      dma;
41
42         seg = kzalloc(sizeof *seg, flags);
43         if (!seg)
44                 return NULL;
45         xhci_dbg(xhci, "Allocating priv segment structure at %p\n", seg);
46
47         seg->trbs = dma_pool_alloc(xhci->segment_pool, flags, &dma);
48         if (!seg->trbs) {
49                 kfree(seg);
50                 return NULL;
51         }
52         xhci_dbg(xhci, "// Allocating segment at %p (virtual) 0x%llx (DMA)\n",
53                         seg->trbs, (unsigned long long)dma);
54
55         memset(seg->trbs, 0, SEGMENT_SIZE);
56         seg->dma = dma;
57         seg->next = NULL;
58
59         return seg;
60 }
61
62 static void xhci_segment_free(struct xhci_hcd *xhci, struct xhci_segment *seg)
63 {
64         if (seg->trbs) {
65                 xhci_dbg(xhci, "Freeing DMA segment at %p (virtual) 0x%llx (DMA)\n",
66                                 seg->trbs, (unsigned long long)seg->dma);
67                 dma_pool_free(xhci->segment_pool, seg->trbs, seg->dma);
68                 seg->trbs = NULL;
69         }
70         xhci_dbg(xhci, "Freeing priv segment structure at %p\n", seg);
71         kfree(seg);
72 }
73
74 /*
75  * Make the prev segment point to the next segment.
76  *
77  * Change the last TRB in the prev segment to be a Link TRB which points to the
78  * DMA address of the next segment.  The caller needs to set any Link TRB
79  * related flags, such as End TRB, Toggle Cycle, and no snoop.
80  */
81 static void xhci_link_segments(struct xhci_hcd *xhci, struct xhci_segment *prev,
82                 struct xhci_segment *next, bool link_trbs, bool isoc)
83 {
84         u32 val;
85
86         if (!prev || !next)
87                 return;
88         prev->next = next;
89         if (link_trbs) {
90                 prev->trbs[TRBS_PER_SEGMENT-1].link.segment_ptr =
91                         cpu_to_le64(next->dma);
92
93                 /* Set the last TRB in the segment to have a TRB type ID of Link TRB */
94                 val = le32_to_cpu(prev->trbs[TRBS_PER_SEGMENT-1].link.control);
95                 val &= ~TRB_TYPE_BITMASK;
96                 val |= TRB_TYPE(TRB_LINK);
97                 /* Always set the chain bit with 0.95 hardware */
98                 /* Set chain bit for isoc rings on AMD 0.96 host */
99                 if (xhci_link_trb_quirk(xhci) ||
100                                 (isoc && (xhci->quirks & XHCI_AMD_0x96_HOST)))
101                         val |= TRB_CHAIN;
102                 prev->trbs[TRBS_PER_SEGMENT-1].link.control = cpu_to_le32(val);
103         }
104         xhci_dbg(xhci, "Linking segment 0x%llx to segment 0x%llx (DMA)\n",
105                         (unsigned long long)prev->dma,
106                         (unsigned long long)next->dma);
107 }
108
109 /* XXX: Do we need the hcd structure in all these functions? */
110 void xhci_ring_free(struct xhci_hcd *xhci, struct xhci_ring *ring)
111 {
112         struct xhci_segment *seg;
113         struct xhci_segment *first_seg;
114
115         if (!ring)
116                 return;
117         if (ring->first_seg) {
118                 first_seg = ring->first_seg;
119                 seg = first_seg->next;
120                 xhci_dbg(xhci, "Freeing ring at %p\n", ring);
121                 while (seg != first_seg) {
122                         struct xhci_segment *next = seg->next;
123                         xhci_segment_free(xhci, seg);
124                         seg = next;
125                 }
126                 xhci_segment_free(xhci, first_seg);
127                 ring->first_seg = NULL;
128         }
129         kfree(ring);
130 }
131
132 static void xhci_initialize_ring_info(struct xhci_ring *ring)
133 {
134         /* The ring is empty, so the enqueue pointer == dequeue pointer */
135         ring->enqueue = ring->first_seg->trbs;
136         ring->enq_seg = ring->first_seg;
137         ring->dequeue = ring->enqueue;
138         ring->deq_seg = ring->first_seg;
139         /* The ring is initialized to 0. The producer must write 1 to the cycle
140          * bit to handover ownership of the TRB, so PCS = 1.  The consumer must
141          * compare CCS to the cycle bit to check ownership, so CCS = 1.
142          */
143         ring->cycle_state = 1;
144         /* Not necessary for new rings, but needed for re-initialized rings */
145         ring->enq_updates = 0;
146         ring->deq_updates = 0;
147 }
148
149 /**
150  * Create a new ring with zero or more segments.
151  *
152  * Link each segment together into a ring.
153  * Set the end flag and the cycle toggle bit on the last segment.
154  * See section 4.9.1 and figures 15 and 16.
155  */
156 static struct xhci_ring *xhci_ring_alloc(struct xhci_hcd *xhci,
157                 unsigned int num_segs, bool link_trbs, bool isoc, gfp_t flags)
158 {
159         struct xhci_ring        *ring;
160         struct xhci_segment     *prev;
161
162         ring = kzalloc(sizeof *(ring), flags);
163         xhci_dbg(xhci, "Allocating ring at %p\n", ring);
164         if (!ring)
165                 return NULL;
166
167         INIT_LIST_HEAD(&ring->td_list);
168         if (num_segs == 0)
169                 return ring;
170
171         ring->first_seg = xhci_segment_alloc(xhci, flags);
172         if (!ring->first_seg)
173                 goto fail;
174         num_segs--;
175
176         prev = ring->first_seg;
177         while (num_segs > 0) {
178                 struct xhci_segment     *next;
179
180                 next = xhci_segment_alloc(xhci, flags);
181                 if (!next) {
182                         prev = ring->first_seg;
183                         while (prev) {
184                                 next = prev->next;
185                                 xhci_segment_free(xhci, prev);
186                                 prev = next;
187                         }
188                         goto fail;
189                 }
190                 xhci_link_segments(xhci, prev, next, link_trbs, isoc);
191
192                 prev = next;
193                 num_segs--;
194         }
195         xhci_link_segments(xhci, prev, ring->first_seg, link_trbs, isoc);
196
197         if (link_trbs) {
198                 /* See section 4.9.2.1 and 6.4.4.1 */
199                 prev->trbs[TRBS_PER_SEGMENT-1].link.control |=
200                         cpu_to_le32(LINK_TOGGLE);
201                 xhci_dbg(xhci, "Wrote link toggle flag to"
202                                 " segment %p (virtual), 0x%llx (DMA)\n",
203                                 prev, (unsigned long long)prev->dma);
204         }
205         xhci_initialize_ring_info(ring);
206         return ring;
207
208 fail:
209         kfree(ring);
210         return NULL;
211 }
212
213 void xhci_free_or_cache_endpoint_ring(struct xhci_hcd *xhci,
214                 struct xhci_virt_device *virt_dev,
215                 unsigned int ep_index)
216 {
217         int rings_cached;
218
219         rings_cached = virt_dev->num_rings_cached;
220         if (rings_cached < XHCI_MAX_RINGS_CACHED) {
221                 virt_dev->ring_cache[rings_cached] =
222                         virt_dev->eps[ep_index].ring;
223                 virt_dev->num_rings_cached++;
224                 xhci_dbg(xhci, "Cached old ring, "
225                                 "%d ring%s cached\n",
226                                 virt_dev->num_rings_cached,
227                                 (virt_dev->num_rings_cached > 1) ? "s" : "");
228         } else {
229                 xhci_ring_free(xhci, virt_dev->eps[ep_index].ring);
230                 xhci_dbg(xhci, "Ring cache full (%d rings), "
231                                 "freeing ring\n",
232                                 virt_dev->num_rings_cached);
233         }
234         virt_dev->eps[ep_index].ring = NULL;
235 }
236
237 /* Zero an endpoint ring (except for link TRBs) and move the enqueue and dequeue
238  * pointers to the beginning of the ring.
239  */
240 static void xhci_reinit_cached_ring(struct xhci_hcd *xhci,
241                 struct xhci_ring *ring, bool isoc)
242 {
243         struct xhci_segment     *seg = ring->first_seg;
244         do {
245                 memset(seg->trbs, 0,
246                                 sizeof(union xhci_trb)*TRBS_PER_SEGMENT);
247                 /* All endpoint rings have link TRBs */
248                 xhci_link_segments(xhci, seg, seg->next, 1, isoc);
249                 seg = seg->next;
250         } while (seg != ring->first_seg);
251         xhci_initialize_ring_info(ring);
252         /* td list should be empty since all URBs have been cancelled,
253          * but just in case...
254          */
255         INIT_LIST_HEAD(&ring->td_list);
256 }
257
258 #define CTX_SIZE(_hcc) (HCC_64BYTE_CONTEXT(_hcc) ? 64 : 32)
259
260 static struct xhci_container_ctx *xhci_alloc_container_ctx(struct xhci_hcd *xhci,
261                                                     int type, gfp_t flags)
262 {
263         struct xhci_container_ctx *ctx = kzalloc(sizeof(*ctx), flags);
264         if (!ctx)
265                 return NULL;
266
267         BUG_ON((type != XHCI_CTX_TYPE_DEVICE) && (type != XHCI_CTX_TYPE_INPUT));
268         ctx->type = type;
269         ctx->size = HCC_64BYTE_CONTEXT(xhci->hcc_params) ? 2048 : 1024;
270         if (type == XHCI_CTX_TYPE_INPUT)
271                 ctx->size += CTX_SIZE(xhci->hcc_params);
272
273         ctx->bytes = dma_pool_alloc(xhci->device_pool, flags, &ctx->dma);
274         memset(ctx->bytes, 0, ctx->size);
275         return ctx;
276 }
277
278 static void xhci_free_container_ctx(struct xhci_hcd *xhci,
279                              struct xhci_container_ctx *ctx)
280 {
281         if (!ctx)
282                 return;
283         dma_pool_free(xhci->device_pool, ctx->bytes, ctx->dma);
284         kfree(ctx);
285 }
286
287 struct xhci_input_control_ctx *xhci_get_input_control_ctx(struct xhci_hcd *xhci,
288                                               struct xhci_container_ctx *ctx)
289 {
290         BUG_ON(ctx->type != XHCI_CTX_TYPE_INPUT);
291         return (struct xhci_input_control_ctx *)ctx->bytes;
292 }
293
294 struct xhci_slot_ctx *xhci_get_slot_ctx(struct xhci_hcd *xhci,
295                                         struct xhci_container_ctx *ctx)
296 {
297         if (ctx->type == XHCI_CTX_TYPE_DEVICE)
298                 return (struct xhci_slot_ctx *)ctx->bytes;
299
300         return (struct xhci_slot_ctx *)
301                 (ctx->bytes + CTX_SIZE(xhci->hcc_params));
302 }
303
304 struct xhci_ep_ctx *xhci_get_ep_ctx(struct xhci_hcd *xhci,
305                                     struct xhci_container_ctx *ctx,
306                                     unsigned int ep_index)
307 {
308         /* increment ep index by offset of start of ep ctx array */
309         ep_index++;
310         if (ctx->type == XHCI_CTX_TYPE_INPUT)
311                 ep_index++;
312
313         return (struct xhci_ep_ctx *)
314                 (ctx->bytes + (ep_index * CTX_SIZE(xhci->hcc_params)));
315 }
316
317
318 /***************** Streams structures manipulation *************************/
319
320 static void xhci_free_stream_ctx(struct xhci_hcd *xhci,
321                 unsigned int num_stream_ctxs,
322                 struct xhci_stream_ctx *stream_ctx, dma_addr_t dma)
323 {
324         struct pci_dev *pdev = to_pci_dev(xhci_to_hcd(xhci)->self.controller);
325
326         if (num_stream_ctxs > MEDIUM_STREAM_ARRAY_SIZE)
327                 dma_free_coherent(&pdev->dev,
328                                 sizeof(struct xhci_stream_ctx)*num_stream_ctxs,
329                                 stream_ctx, dma);
330         else if (num_stream_ctxs <= SMALL_STREAM_ARRAY_SIZE)
331                 return dma_pool_free(xhci->small_streams_pool,
332                                 stream_ctx, dma);
333         else
334                 return dma_pool_free(xhci->medium_streams_pool,
335                                 stream_ctx, dma);
336 }
337
338 /*
339  * The stream context array for each endpoint with bulk streams enabled can
340  * vary in size, based on:
341  *  - how many streams the endpoint supports,
342  *  - the maximum primary stream array size the host controller supports,
343  *  - and how many streams the device driver asks for.
344  *
345  * The stream context array must be a power of 2, and can be as small as
346  * 64 bytes or as large as 1MB.
347  */
348 static struct xhci_stream_ctx *xhci_alloc_stream_ctx(struct xhci_hcd *xhci,
349                 unsigned int num_stream_ctxs, dma_addr_t *dma,
350                 gfp_t mem_flags)
351 {
352         struct pci_dev *pdev = to_pci_dev(xhci_to_hcd(xhci)->self.controller);
353
354         if (num_stream_ctxs > MEDIUM_STREAM_ARRAY_SIZE)
355                 return dma_alloc_coherent(&pdev->dev,
356                                 sizeof(struct xhci_stream_ctx)*num_stream_ctxs,
357                                 dma, mem_flags);
358         else if (num_stream_ctxs <= SMALL_STREAM_ARRAY_SIZE)
359                 return dma_pool_alloc(xhci->small_streams_pool,
360                                 mem_flags, dma);
361         else
362                 return dma_pool_alloc(xhci->medium_streams_pool,
363                                 mem_flags, dma);
364 }
365
366 struct xhci_ring *xhci_dma_to_transfer_ring(
367                 struct xhci_virt_ep *ep,
368                 u64 address)
369 {
370         if (ep->ep_state & EP_HAS_STREAMS)
371                 return radix_tree_lookup(&ep->stream_info->trb_address_map,
372                                 address >> SEGMENT_SHIFT);
373         return ep->ring;
374 }
375
376 /* Only use this when you know stream_info is valid */
377 #ifdef CONFIG_USB_XHCI_HCD_DEBUGGING
378 static struct xhci_ring *dma_to_stream_ring(
379                 struct xhci_stream_info *stream_info,
380                 u64 address)
381 {
382         return radix_tree_lookup(&stream_info->trb_address_map,
383                         address >> SEGMENT_SHIFT);
384 }
385 #endif  /* CONFIG_USB_XHCI_HCD_DEBUGGING */
386
387 struct xhci_ring *xhci_stream_id_to_ring(
388                 struct xhci_virt_device *dev,
389                 unsigned int ep_index,
390                 unsigned int stream_id)
391 {
392         struct xhci_virt_ep *ep = &dev->eps[ep_index];
393
394         if (stream_id == 0)
395                 return ep->ring;
396         if (!ep->stream_info)
397                 return NULL;
398
399         if (stream_id > ep->stream_info->num_streams)
400                 return NULL;
401         return ep->stream_info->stream_rings[stream_id];
402 }
403
404 #ifdef CONFIG_USB_XHCI_HCD_DEBUGGING
405 static int xhci_test_radix_tree(struct xhci_hcd *xhci,
406                 unsigned int num_streams,
407                 struct xhci_stream_info *stream_info)
408 {
409         u32 cur_stream;
410         struct xhci_ring *cur_ring;
411         u64 addr;
412
413         for (cur_stream = 1; cur_stream < num_streams; cur_stream++) {
414                 struct xhci_ring *mapped_ring;
415                 int trb_size = sizeof(union xhci_trb);
416
417                 cur_ring = stream_info->stream_rings[cur_stream];
418                 for (addr = cur_ring->first_seg->dma;
419                                 addr < cur_ring->first_seg->dma + SEGMENT_SIZE;
420                                 addr += trb_size) {
421                         mapped_ring = dma_to_stream_ring(stream_info, addr);
422                         if (cur_ring != mapped_ring) {
423                                 xhci_warn(xhci, "WARN: DMA address 0x%08llx "
424                                                 "didn't map to stream ID %u; "
425                                                 "mapped to ring %p\n",
426                                                 (unsigned long long) addr,
427                                                 cur_stream,
428                                                 mapped_ring);
429                                 return -EINVAL;
430                         }
431                 }
432                 /* One TRB after the end of the ring segment shouldn't return a
433                  * pointer to the current ring (although it may be a part of a
434                  * different ring).
435                  */
436                 mapped_ring = dma_to_stream_ring(stream_info, addr);
437                 if (mapped_ring != cur_ring) {
438                         /* One TRB before should also fail */
439                         addr = cur_ring->first_seg->dma - trb_size;
440                         mapped_ring = dma_to_stream_ring(stream_info, addr);
441                 }
442                 if (mapped_ring == cur_ring) {
443                         xhci_warn(xhci, "WARN: Bad DMA address 0x%08llx "
444                                         "mapped to valid stream ID %u; "
445                                         "mapped ring = %p\n",
446                                         (unsigned long long) addr,
447                                         cur_stream,
448                                         mapped_ring);
449                         return -EINVAL;
450                 }
451         }
452         return 0;
453 }
454 #endif  /* CONFIG_USB_XHCI_HCD_DEBUGGING */
455
456 /*
457  * Change an endpoint's internal structure so it supports stream IDs.  The
458  * number of requested streams includes stream 0, which cannot be used by device
459  * drivers.
460  *
461  * The number of stream contexts in the stream context array may be bigger than
462  * the number of streams the driver wants to use.  This is because the number of
463  * stream context array entries must be a power of two.
464  *
465  * We need a radix tree for mapping physical addresses of TRBs to which stream
466  * ID they belong to.  We need to do this because the host controller won't tell
467  * us which stream ring the TRB came from.  We could store the stream ID in an
468  * event data TRB, but that doesn't help us for the cancellation case, since the
469  * endpoint may stop before it reaches that event data TRB.
470  *
471  * The radix tree maps the upper portion of the TRB DMA address to a ring
472  * segment that has the same upper portion of DMA addresses.  For example, say I
473  * have segments of size 1KB, that are always 64-byte aligned.  A segment may
474  * start at 0x10c91000 and end at 0x10c913f0.  If I use the upper 10 bits, the
475  * key to the stream ID is 0x43244.  I can use the DMA address of the TRB to
476  * pass the radix tree a key to get the right stream ID:
477  *
478  *      0x10c90fff >> 10 = 0x43243
479  *      0x10c912c0 >> 10 = 0x43244
480  *      0x10c91400 >> 10 = 0x43245
481  *
482  * Obviously, only those TRBs with DMA addresses that are within the segment
483  * will make the radix tree return the stream ID for that ring.
484  *
485  * Caveats for the radix tree:
486  *
487  * The radix tree uses an unsigned long as a key pair.  On 32-bit systems, an
488  * unsigned long will be 32-bits; on a 64-bit system an unsigned long will be
489  * 64-bits.  Since we only request 32-bit DMA addresses, we can use that as the
490  * key on 32-bit or 64-bit systems (it would also be fine if we asked for 64-bit
491  * PCI DMA addresses on a 64-bit system).  There might be a problem on 32-bit
492  * extended systems (where the DMA address can be bigger than 32-bits),
493  * if we allow the PCI dma mask to be bigger than 32-bits.  So don't do that.
494  */
495 struct xhci_stream_info *xhci_alloc_stream_info(struct xhci_hcd *xhci,
496                 unsigned int num_stream_ctxs,
497                 unsigned int num_streams, gfp_t mem_flags)
498 {
499         struct xhci_stream_info *stream_info;
500         u32 cur_stream;
501         struct xhci_ring *cur_ring;
502         unsigned long key;
503         u64 addr;
504         int ret;
505
506         xhci_dbg(xhci, "Allocating %u streams and %u "
507                         "stream context array entries.\n",
508                         num_streams, num_stream_ctxs);
509         if (xhci->cmd_ring_reserved_trbs == MAX_RSVD_CMD_TRBS) {
510                 xhci_dbg(xhci, "Command ring has no reserved TRBs available\n");
511                 return NULL;
512         }
513         xhci->cmd_ring_reserved_trbs++;
514
515         stream_info = kzalloc(sizeof(struct xhci_stream_info), mem_flags);
516         if (!stream_info)
517                 goto cleanup_trbs;
518
519         stream_info->num_streams = num_streams;
520         stream_info->num_stream_ctxs = num_stream_ctxs;
521
522         /* Initialize the array of virtual pointers to stream rings. */
523         stream_info->stream_rings = kzalloc(
524                         sizeof(struct xhci_ring *)*num_streams,
525                         mem_flags);
526         if (!stream_info->stream_rings)
527                 goto cleanup_info;
528
529         /* Initialize the array of DMA addresses for stream rings for the HW. */
530         stream_info->stream_ctx_array = xhci_alloc_stream_ctx(xhci,
531                         num_stream_ctxs, &stream_info->ctx_array_dma,
532                         mem_flags);
533         if (!stream_info->stream_ctx_array)
534                 goto cleanup_ctx;
535         memset(stream_info->stream_ctx_array, 0,
536                         sizeof(struct xhci_stream_ctx)*num_stream_ctxs);
537
538         /* Allocate everything needed to free the stream rings later */
539         stream_info->free_streams_command =
540                 xhci_alloc_command(xhci, true, true, mem_flags);
541         if (!stream_info->free_streams_command)
542                 goto cleanup_ctx;
543
544         INIT_RADIX_TREE(&stream_info->trb_address_map, GFP_ATOMIC);
545
546         /* Allocate rings for all the streams that the driver will use,
547          * and add their segment DMA addresses to the radix tree.
548          * Stream 0 is reserved.
549          */
550         for (cur_stream = 1; cur_stream < num_streams; cur_stream++) {
551                 stream_info->stream_rings[cur_stream] =
552                         xhci_ring_alloc(xhci, 1, true, false, mem_flags);
553                 cur_ring = stream_info->stream_rings[cur_stream];
554                 if (!cur_ring)
555                         goto cleanup_rings;
556                 cur_ring->stream_id = cur_stream;
557                 /* Set deq ptr, cycle bit, and stream context type */
558                 addr = cur_ring->first_seg->dma |
559                         SCT_FOR_CTX(SCT_PRI_TR) |
560                         cur_ring->cycle_state;
561                 stream_info->stream_ctx_array[cur_stream].stream_ring =
562                         cpu_to_le64(addr);
563                 xhci_dbg(xhci, "Setting stream %d ring ptr to 0x%08llx\n",
564                                 cur_stream, (unsigned long long) addr);
565
566                 key = (unsigned long)
567                         (cur_ring->first_seg->dma >> SEGMENT_SHIFT);
568                 ret = radix_tree_insert(&stream_info->trb_address_map,
569                                 key, cur_ring);
570                 if (ret) {
571                         xhci_ring_free(xhci, cur_ring);
572                         stream_info->stream_rings[cur_stream] = NULL;
573                         goto cleanup_rings;
574                 }
575         }
576         /* Leave the other unused stream ring pointers in the stream context
577          * array initialized to zero.  This will cause the xHC to give us an
578          * error if the device asks for a stream ID we don't have setup (if it
579          * was any other way, the host controller would assume the ring is
580          * "empty" and wait forever for data to be queued to that stream ID).
581          */
582 #if XHCI_DEBUG
583         /* Do a little test on the radix tree to make sure it returns the
584          * correct values.
585          */
586         if (xhci_test_radix_tree(xhci, num_streams, stream_info))
587                 goto cleanup_rings;
588 #endif
589
590         return stream_info;
591
592 cleanup_rings:
593         for (cur_stream = 1; cur_stream < num_streams; cur_stream++) {
594                 cur_ring = stream_info->stream_rings[cur_stream];
595                 if (cur_ring) {
596                         addr = cur_ring->first_seg->dma;
597                         radix_tree_delete(&stream_info->trb_address_map,
598                                         addr >> SEGMENT_SHIFT);
599                         xhci_ring_free(xhci, cur_ring);
600                         stream_info->stream_rings[cur_stream] = NULL;
601                 }
602         }
603         xhci_free_command(xhci, stream_info->free_streams_command);
604 cleanup_ctx:
605         kfree(stream_info->stream_rings);
606 cleanup_info:
607         kfree(stream_info);
608 cleanup_trbs:
609         xhci->cmd_ring_reserved_trbs--;
610         return NULL;
611 }
612 /*
613  * Sets the MaxPStreams field and the Linear Stream Array field.
614  * Sets the dequeue pointer to the stream context array.
615  */
616 void xhci_setup_streams_ep_input_ctx(struct xhci_hcd *xhci,
617                 struct xhci_ep_ctx *ep_ctx,
618                 struct xhci_stream_info *stream_info)
619 {
620         u32 max_primary_streams;
621         /* MaxPStreams is the number of stream context array entries, not the
622          * number we're actually using.  Must be in 2^(MaxPstreams + 1) format.
623          * fls(0) = 0, fls(0x1) = 1, fls(0x10) = 2, fls(0x100) = 3, etc.
624          */
625         max_primary_streams = fls(stream_info->num_stream_ctxs) - 2;
626         xhci_dbg(xhci, "Setting number of stream ctx array entries to %u\n",
627                         1 << (max_primary_streams + 1));
628         ep_ctx->ep_info &= cpu_to_le32(~EP_MAXPSTREAMS_MASK);
629         ep_ctx->ep_info |= cpu_to_le32(EP_MAXPSTREAMS(max_primary_streams)
630                                        | EP_HAS_LSA);
631         ep_ctx->deq  = cpu_to_le64(stream_info->ctx_array_dma);
632 }
633
634 /*
635  * Sets the MaxPStreams field and the Linear Stream Array field to 0.
636  * Reinstalls the "normal" endpoint ring (at its previous dequeue mark,
637  * not at the beginning of the ring).
638  */
639 void xhci_setup_no_streams_ep_input_ctx(struct xhci_hcd *xhci,
640                 struct xhci_ep_ctx *ep_ctx,
641                 struct xhci_virt_ep *ep)
642 {
643         dma_addr_t addr;
644         ep_ctx->ep_info &= cpu_to_le32(~(EP_MAXPSTREAMS_MASK | EP_HAS_LSA));
645         addr = xhci_trb_virt_to_dma(ep->ring->deq_seg, ep->ring->dequeue);
646         ep_ctx->deq  = cpu_to_le64(addr | ep->ring->cycle_state);
647 }
648
649 /* Frees all stream contexts associated with the endpoint,
650  *
651  * Caller should fix the endpoint context streams fields.
652  */
653 void xhci_free_stream_info(struct xhci_hcd *xhci,
654                 struct xhci_stream_info *stream_info)
655 {
656         int cur_stream;
657         struct xhci_ring *cur_ring;
658         dma_addr_t addr;
659
660         if (!stream_info)
661                 return;
662
663         for (cur_stream = 1; cur_stream < stream_info->num_streams;
664                         cur_stream++) {
665                 cur_ring = stream_info->stream_rings[cur_stream];
666                 if (cur_ring) {
667                         addr = cur_ring->first_seg->dma;
668                         radix_tree_delete(&stream_info->trb_address_map,
669                                         addr >> SEGMENT_SHIFT);
670                         xhci_ring_free(xhci, cur_ring);
671                         stream_info->stream_rings[cur_stream] = NULL;
672                 }
673         }
674         xhci_free_command(xhci, stream_info->free_streams_command);
675         xhci->cmd_ring_reserved_trbs--;
676         if (stream_info->stream_ctx_array)
677                 xhci_free_stream_ctx(xhci,
678                                 stream_info->num_stream_ctxs,
679                                 stream_info->stream_ctx_array,
680                                 stream_info->ctx_array_dma);
681
682         if (stream_info)
683                 kfree(stream_info->stream_rings);
684         kfree(stream_info);
685 }
686
687
688 /***************** Device context manipulation *************************/
689
690 static void xhci_init_endpoint_timer(struct xhci_hcd *xhci,
691                 struct xhci_virt_ep *ep)
692 {
693         init_timer(&ep->stop_cmd_timer);
694         ep->stop_cmd_timer.data = (unsigned long) ep;
695         ep->stop_cmd_timer.function = xhci_stop_endpoint_command_watchdog;
696         ep->xhci = xhci;
697 }
698
699 static void xhci_free_tt_info(struct xhci_hcd *xhci,
700                 struct xhci_virt_device *virt_dev,
701                 int slot_id)
702 {
703         struct list_head *tt_list_head;
704         struct xhci_tt_bw_info *tt_info, *next;
705         bool slot_found = false;
706
707         /* If the device never made it past the Set Address stage,
708          * it may not have the real_port set correctly.
709          */
710         if (virt_dev->real_port == 0 ||
711                         virt_dev->real_port > HCS_MAX_PORTS(xhci->hcs_params1)) {
712                 xhci_dbg(xhci, "Bad real port.\n");
713                 return;
714         }
715
716         tt_list_head = &(xhci->rh_bw[virt_dev->real_port - 1].tts);
717         list_for_each_entry_safe(tt_info, next, tt_list_head, tt_list) {
718                 /* Multi-TT hubs will have more than one entry */
719                 if (tt_info->slot_id == slot_id) {
720                         slot_found = true;
721                         list_del(&tt_info->tt_list);
722                         kfree(tt_info);
723                 } else if (slot_found) {
724                         break;
725                 }
726         }
727 }
728
729 int xhci_alloc_tt_info(struct xhci_hcd *xhci,
730                 struct xhci_virt_device *virt_dev,
731                 struct usb_device *hdev,
732                 struct usb_tt *tt, gfp_t mem_flags)
733 {
734         struct xhci_tt_bw_info          *tt_info;
735         unsigned int                    num_ports;
736         int                             i, j;
737
738         if (!tt->multi)
739                 num_ports = 1;
740         else
741                 num_ports = hdev->maxchild;
742
743         for (i = 0; i < num_ports; i++, tt_info++) {
744                 struct xhci_interval_bw_table *bw_table;
745
746                 tt_info = kzalloc(sizeof(*tt_info), mem_flags);
747                 if (!tt_info)
748                         goto free_tts;
749                 INIT_LIST_HEAD(&tt_info->tt_list);
750                 list_add(&tt_info->tt_list,
751                                 &xhci->rh_bw[virt_dev->real_port - 1].tts);
752                 tt_info->slot_id = virt_dev->udev->slot_id;
753                 if (tt->multi)
754                         tt_info->ttport = i+1;
755                 bw_table = &tt_info->bw_table;
756                 for (j = 0; j < XHCI_MAX_INTERVAL; j++)
757                         INIT_LIST_HEAD(&bw_table->interval_bw[j].endpoints);
758         }
759         return 0;
760
761 free_tts:
762         xhci_free_tt_info(xhci, virt_dev, virt_dev->udev->slot_id);
763         return -ENOMEM;
764 }
765
766
767 /* All the xhci_tds in the ring's TD list should be freed at this point.
768  * Should be called with xhci->lock held if there is any chance the TT lists
769  * will be manipulated by the configure endpoint, allocate device, or update
770  * hub functions while this function is removing the TT entries from the list.
771  */
772 void xhci_free_virt_device(struct xhci_hcd *xhci, int slot_id)
773 {
774         struct xhci_virt_device *dev;
775         int i;
776         int old_active_eps = 0;
777
778         /* Slot ID 0 is reserved */
779         if (slot_id == 0 || !xhci->devs[slot_id])
780                 return;
781
782         dev = xhci->devs[slot_id];
783         xhci->dcbaa->dev_context_ptrs[slot_id] = 0;
784         if (!dev)
785                 return;
786
787         if (dev->tt_info)
788                 old_active_eps = dev->tt_info->active_eps;
789
790         for (i = 0; i < 31; ++i) {
791                 if (dev->eps[i].ring)
792                         xhci_ring_free(xhci, dev->eps[i].ring);
793                 if (dev->eps[i].stream_info)
794                         xhci_free_stream_info(xhci,
795                                         dev->eps[i].stream_info);
796                 /* Endpoints on the TT/root port lists should have been removed
797                  * when usb_disable_device() was called for the device.
798                  * We can't drop them anyway, because the udev might have gone
799                  * away by this point, and we can't tell what speed it was.
800                  */
801                 if (!list_empty(&dev->eps[i].bw_endpoint_list))
802                         xhci_warn(xhci, "Slot %u endpoint %u "
803                                         "not removed from BW list!\n",
804                                         slot_id, i);
805         }
806         /* If this is a hub, free the TT(s) from the TT list */
807         xhci_free_tt_info(xhci, dev, slot_id);
808         /* If necessary, update the number of active TTs on this root port */
809         xhci_update_tt_active_eps(xhci, dev, old_active_eps);
810
811         if (dev->ring_cache) {
812                 for (i = 0; i < dev->num_rings_cached; i++)
813                         xhci_ring_free(xhci, dev->ring_cache[i]);
814                 kfree(dev->ring_cache);
815         }
816
817         if (dev->in_ctx)
818                 xhci_free_container_ctx(xhci, dev->in_ctx);
819         if (dev->out_ctx)
820                 xhci_free_container_ctx(xhci, dev->out_ctx);
821
822         kfree(xhci->devs[slot_id]);
823         xhci->devs[slot_id] = NULL;
824 }
825
826 int xhci_alloc_virt_device(struct xhci_hcd *xhci, int slot_id,
827                 struct usb_device *udev, gfp_t flags)
828 {
829         struct xhci_virt_device *dev;
830         int i;
831
832         /* Slot ID 0 is reserved */
833         if (slot_id == 0 || xhci->devs[slot_id]) {
834                 xhci_warn(xhci, "Bad Slot ID %d\n", slot_id);
835                 return 0;
836         }
837
838         xhci->devs[slot_id] = kzalloc(sizeof(*xhci->devs[slot_id]), flags);
839         if (!xhci->devs[slot_id])
840                 return 0;
841         dev = xhci->devs[slot_id];
842
843         /* Allocate the (output) device context that will be used in the HC. */
844         dev->out_ctx = xhci_alloc_container_ctx(xhci, XHCI_CTX_TYPE_DEVICE, flags);
845         if (!dev->out_ctx)
846                 goto fail;
847
848         xhci_dbg(xhci, "Slot %d output ctx = 0x%llx (dma)\n", slot_id,
849                         (unsigned long long)dev->out_ctx->dma);
850
851         /* Allocate the (input) device context for address device command */
852         dev->in_ctx = xhci_alloc_container_ctx(xhci, XHCI_CTX_TYPE_INPUT, flags);
853         if (!dev->in_ctx)
854                 goto fail;
855
856         xhci_dbg(xhci, "Slot %d input ctx = 0x%llx (dma)\n", slot_id,
857                         (unsigned long long)dev->in_ctx->dma);
858
859         /* Initialize the cancellation list and watchdog timers for each ep */
860         for (i = 0; i < 31; i++) {
861                 xhci_init_endpoint_timer(xhci, &dev->eps[i]);
862                 INIT_LIST_HEAD(&dev->eps[i].cancelled_td_list);
863                 INIT_LIST_HEAD(&dev->eps[i].bw_endpoint_list);
864         }
865
866         /* Allocate endpoint 0 ring */
867         dev->eps[0].ring = xhci_ring_alloc(xhci, 1, true, false, flags);
868         if (!dev->eps[0].ring)
869                 goto fail;
870
871         /* Allocate pointers to the ring cache */
872         dev->ring_cache = kzalloc(
873                         sizeof(struct xhci_ring *)*XHCI_MAX_RINGS_CACHED,
874                         flags);
875         if (!dev->ring_cache)
876                 goto fail;
877         dev->num_rings_cached = 0;
878
879         init_completion(&dev->cmd_completion);
880         INIT_LIST_HEAD(&dev->cmd_list);
881         dev->udev = udev;
882
883         /* Point to output device context in dcbaa. */
884         xhci->dcbaa->dev_context_ptrs[slot_id] = cpu_to_le64(dev->out_ctx->dma);
885         xhci_dbg(xhci, "Set slot id %d dcbaa entry %p to 0x%llx\n",
886                  slot_id,
887                  &xhci->dcbaa->dev_context_ptrs[slot_id],
888                  le64_to_cpu(xhci->dcbaa->dev_context_ptrs[slot_id]));
889
890         return 1;
891 fail:
892         xhci_free_virt_device(xhci, slot_id);
893         return 0;
894 }
895
896 void xhci_copy_ep0_dequeue_into_input_ctx(struct xhci_hcd *xhci,
897                 struct usb_device *udev)
898 {
899         struct xhci_virt_device *virt_dev;
900         struct xhci_ep_ctx      *ep0_ctx;
901         struct xhci_ring        *ep_ring;
902
903         virt_dev = xhci->devs[udev->slot_id];
904         ep0_ctx = xhci_get_ep_ctx(xhci, virt_dev->in_ctx, 0);
905         ep_ring = virt_dev->eps[0].ring;
906         /*
907          * FIXME we don't keep track of the dequeue pointer very well after a
908          * Set TR dequeue pointer, so we're setting the dequeue pointer of the
909          * host to our enqueue pointer.  This should only be called after a
910          * configured device has reset, so all control transfers should have
911          * been completed or cancelled before the reset.
912          */
913         ep0_ctx->deq = cpu_to_le64(xhci_trb_virt_to_dma(ep_ring->enq_seg,
914                                                         ep_ring->enqueue)
915                                    | ep_ring->cycle_state);
916 }
917
918 /*
919  * The xHCI roothub may have ports of differing speeds in any order in the port
920  * status registers.  xhci->port_array provides an array of the port speed for
921  * each offset into the port status registers.
922  *
923  * The xHCI hardware wants to know the roothub port number that the USB device
924  * is attached to (or the roothub port its ancestor hub is attached to).  All we
925  * know is the index of that port under either the USB 2.0 or the USB 3.0
926  * roothub, but that doesn't give us the real index into the HW port status
927  * registers.  Scan through the xHCI roothub port array, looking for the Nth
928  * entry of the correct port speed.  Return the port number of that entry.
929  */
930 static u32 xhci_find_real_port_number(struct xhci_hcd *xhci,
931                 struct usb_device *udev)
932 {
933         struct usb_device *top_dev;
934         unsigned int num_similar_speed_ports;
935         unsigned int faked_port_num;
936         int i;
937
938         for (top_dev = udev; top_dev->parent && top_dev->parent->parent;
939                         top_dev = top_dev->parent)
940                 /* Found device below root hub */;
941         faked_port_num = top_dev->portnum;
942         for (i = 0, num_similar_speed_ports = 0;
943                         i < HCS_MAX_PORTS(xhci->hcs_params1); i++) {
944                 u8 port_speed = xhci->port_array[i];
945
946                 /*
947                  * Skip ports that don't have known speeds, or have duplicate
948                  * Extended Capabilities port speed entries.
949                  */
950                 if (port_speed == 0 || port_speed == DUPLICATE_ENTRY)
951                         continue;
952
953                 /*
954                  * USB 3.0 ports are always under a USB 3.0 hub.  USB 2.0 and
955                  * 1.1 ports are under the USB 2.0 hub.  If the port speed
956                  * matches the device speed, it's a similar speed port.
957                  */
958                 if ((port_speed == 0x03) == (udev->speed == USB_SPEED_SUPER))
959                         num_similar_speed_ports++;
960                 if (num_similar_speed_ports == faked_port_num)
961                         /* Roothub ports are numbered from 1 to N */
962                         return i+1;
963         }
964         return 0;
965 }
966
967 /* Setup an xHCI virtual device for a Set Address command */
968 int xhci_setup_addressable_virt_dev(struct xhci_hcd *xhci, struct usb_device *udev)
969 {
970         struct xhci_virt_device *dev;
971         struct xhci_ep_ctx      *ep0_ctx;
972         struct xhci_slot_ctx    *slot_ctx;
973         u32                     port_num;
974         struct usb_device *top_dev;
975
976         dev = xhci->devs[udev->slot_id];
977         /* Slot ID 0 is reserved */
978         if (udev->slot_id == 0 || !dev) {
979                 xhci_warn(xhci, "Slot ID %d is not assigned to this device\n",
980                                 udev->slot_id);
981                 return -EINVAL;
982         }
983         ep0_ctx = xhci_get_ep_ctx(xhci, dev->in_ctx, 0);
984         slot_ctx = xhci_get_slot_ctx(xhci, dev->in_ctx);
985
986         /* 3) Only the control endpoint is valid - one endpoint context */
987         slot_ctx->dev_info |= cpu_to_le32(LAST_CTX(1) | udev->route);
988         switch (udev->speed) {
989         case USB_SPEED_SUPER:
990                 slot_ctx->dev_info |= cpu_to_le32(SLOT_SPEED_SS);
991                 break;
992         case USB_SPEED_HIGH:
993                 slot_ctx->dev_info |= cpu_to_le32(SLOT_SPEED_HS);
994                 break;
995         case USB_SPEED_FULL:
996                 slot_ctx->dev_info |= cpu_to_le32(SLOT_SPEED_FS);
997                 break;
998         case USB_SPEED_LOW:
999                 slot_ctx->dev_info |= cpu_to_le32(SLOT_SPEED_LS);
1000                 break;
1001         case USB_SPEED_WIRELESS:
1002                 xhci_dbg(xhci, "FIXME xHCI doesn't support wireless speeds\n");
1003                 return -EINVAL;
1004                 break;
1005         default:
1006                 /* Speed was set earlier, this shouldn't happen. */
1007                 BUG();
1008         }
1009         /* Find the root hub port this device is under */
1010         port_num = xhci_find_real_port_number(xhci, udev);
1011         if (!port_num)
1012                 return -EINVAL;
1013         slot_ctx->dev_info2 |= cpu_to_le32(ROOT_HUB_PORT(port_num));
1014         /* Set the port number in the virtual_device to the faked port number */
1015         for (top_dev = udev; top_dev->parent && top_dev->parent->parent;
1016                         top_dev = top_dev->parent)
1017                 /* Found device below root hub */;
1018         dev->fake_port = top_dev->portnum;
1019         dev->real_port = port_num;
1020         xhci_dbg(xhci, "Set root hub portnum to %d\n", port_num);
1021         xhci_dbg(xhci, "Set fake root hub portnum to %d\n", dev->fake_port);
1022
1023         /* Find the right bandwidth table that this device will be a part of.
1024          * If this is a full speed device attached directly to a root port (or a
1025          * decendent of one), it counts as a primary bandwidth domain, not a
1026          * secondary bandwidth domain under a TT.  An xhci_tt_info structure
1027          * will never be created for the HS root hub.
1028          */
1029         if (!udev->tt || !udev->tt->hub->parent) {
1030                 dev->bw_table = &xhci->rh_bw[port_num - 1].bw_table;
1031         } else {
1032                 struct xhci_root_port_bw_info *rh_bw;
1033                 struct xhci_tt_bw_info *tt_bw;
1034
1035                 rh_bw = &xhci->rh_bw[port_num - 1];
1036                 /* Find the right TT. */
1037                 list_for_each_entry(tt_bw, &rh_bw->tts, tt_list) {
1038                         if (tt_bw->slot_id != udev->tt->hub->slot_id)
1039                                 continue;
1040
1041                         if (!dev->udev->tt->multi ||
1042                                         (udev->tt->multi &&
1043                                          tt_bw->ttport == dev->udev->ttport)) {
1044                                 dev->bw_table = &tt_bw->bw_table;
1045                                 dev->tt_info = tt_bw;
1046                                 break;
1047                         }
1048                 }
1049                 if (!dev->tt_info)
1050                         xhci_warn(xhci, "WARN: Didn't find a matching TT\n");
1051         }
1052
1053         /* Is this a LS/FS device under an external HS hub? */
1054         if (udev->tt && udev->tt->hub->parent) {
1055                 slot_ctx->tt_info = cpu_to_le32(udev->tt->hub->slot_id |
1056                                                 (udev->ttport << 8));
1057                 if (udev->tt->multi)
1058                         slot_ctx->dev_info |= cpu_to_le32(DEV_MTT);
1059         }
1060         xhci_dbg(xhci, "udev->tt = %p\n", udev->tt);
1061         xhci_dbg(xhci, "udev->ttport = 0x%x\n", udev->ttport);
1062
1063         /* Step 4 - ring already allocated */
1064         /* Step 5 */
1065         ep0_ctx->ep_info2 = cpu_to_le32(EP_TYPE(CTRL_EP));
1066         /*
1067          * XXX: Not sure about wireless USB devices.
1068          */
1069         switch (udev->speed) {
1070         case USB_SPEED_SUPER:
1071                 ep0_ctx->ep_info2 |= cpu_to_le32(MAX_PACKET(512));
1072                 break;
1073         case USB_SPEED_HIGH:
1074         /* USB core guesses at a 64-byte max packet first for FS devices */
1075         case USB_SPEED_FULL:
1076                 ep0_ctx->ep_info2 |= cpu_to_le32(MAX_PACKET(64));
1077                 break;
1078         case USB_SPEED_LOW:
1079                 ep0_ctx->ep_info2 |= cpu_to_le32(MAX_PACKET(8));
1080                 break;
1081         case USB_SPEED_WIRELESS:
1082                 xhci_dbg(xhci, "FIXME xHCI doesn't support wireless speeds\n");
1083                 return -EINVAL;
1084                 break;
1085         default:
1086                 /* New speed? */
1087                 BUG();
1088         }
1089         /* EP 0 can handle "burst" sizes of 1, so Max Burst Size field is 0 */
1090         ep0_ctx->ep_info2 |= cpu_to_le32(MAX_BURST(0) | ERROR_COUNT(3));
1091
1092         ep0_ctx->deq = cpu_to_le64(dev->eps[0].ring->first_seg->dma |
1093                                    dev->eps[0].ring->cycle_state);
1094
1095         /* Steps 7 and 8 were done in xhci_alloc_virt_device() */
1096
1097         return 0;
1098 }
1099
1100 /*
1101  * Convert interval expressed as 2^(bInterval - 1) == interval into
1102  * straight exponent value 2^n == interval.
1103  *
1104  */
1105 static unsigned int xhci_parse_exponent_interval(struct usb_device *udev,
1106                 struct usb_host_endpoint *ep)
1107 {
1108         unsigned int interval;
1109
1110         interval = clamp_val(ep->desc.bInterval, 1, 16) - 1;
1111         if (interval != ep->desc.bInterval - 1)
1112                 dev_warn(&udev->dev,
1113                          "ep %#x - rounding interval to %d %sframes\n",
1114                          ep->desc.bEndpointAddress,
1115                          1 << interval,
1116                          udev->speed == USB_SPEED_FULL ? "" : "micro");
1117
1118         if (udev->speed == USB_SPEED_FULL) {
1119                 /*
1120                  * Full speed isoc endpoints specify interval in frames,
1121                  * not microframes. We are using microframes everywhere,
1122                  * so adjust accordingly.
1123                  */
1124                 interval += 3;  /* 1 frame = 2^3 uframes */
1125         }
1126
1127         return interval;
1128 }
1129
1130 /*
1131  * Convert bInterval expressed in microframes (in 1-255 range) to exponent of
1132  * microframes, rounded down to nearest power of 2.
1133  */
1134 static unsigned int xhci_microframes_to_exponent(struct usb_device *udev,
1135                 struct usb_host_endpoint *ep, unsigned int desc_interval,
1136                 unsigned int min_exponent, unsigned int max_exponent)
1137 {
1138         unsigned int interval;
1139
1140         interval = fls(desc_interval) - 1;
1141         interval = clamp_val(interval, min_exponent, max_exponent);
1142         if ((1 << interval) != desc_interval)
1143                 dev_warn(&udev->dev,
1144                          "ep %#x - rounding interval to %d microframes, ep desc says %d microframes\n",
1145                          ep->desc.bEndpointAddress,
1146                          1 << interval,
1147                          desc_interval);
1148
1149         return interval;
1150 }
1151
1152 static unsigned int xhci_parse_microframe_interval(struct usb_device *udev,
1153                 struct usb_host_endpoint *ep)
1154 {
1155         if (ep->desc.bInterval == 0)
1156                 return 0;
1157         return xhci_microframes_to_exponent(udev, ep,
1158                         ep->desc.bInterval, 0, 15);
1159 }
1160
1161
1162 static unsigned int xhci_parse_frame_interval(struct usb_device *udev,
1163                 struct usb_host_endpoint *ep)
1164 {
1165         return xhci_microframes_to_exponent(udev, ep,
1166                         ep->desc.bInterval * 8, 3, 10);
1167 }
1168
1169 /* Return the polling or NAK interval.
1170  *
1171  * The polling interval is expressed in "microframes".  If xHCI's Interval field
1172  * is set to N, it will service the endpoint every 2^(Interval)*125us.
1173  *
1174  * The NAK interval is one NAK per 1 to 255 microframes, or no NAKs if interval
1175  * is set to 0.
1176  */
1177 static unsigned int xhci_get_endpoint_interval(struct usb_device *udev,
1178                 struct usb_host_endpoint *ep)
1179 {
1180         unsigned int interval = 0;
1181
1182         switch (udev->speed) {
1183         case USB_SPEED_HIGH:
1184                 /* Max NAK rate */
1185                 if (usb_endpoint_xfer_control(&ep->desc) ||
1186                     usb_endpoint_xfer_bulk(&ep->desc)) {
1187                         interval = xhci_parse_microframe_interval(udev, ep);
1188                         break;
1189                 }
1190                 /* Fall through - SS and HS isoc/int have same decoding */
1191
1192         case USB_SPEED_SUPER:
1193                 if (usb_endpoint_xfer_int(&ep->desc) ||
1194                     usb_endpoint_xfer_isoc(&ep->desc)) {
1195                         interval = xhci_parse_exponent_interval(udev, ep);
1196                 }
1197                 break;
1198
1199         case USB_SPEED_FULL:
1200                 if (usb_endpoint_xfer_isoc(&ep->desc)) {
1201                         interval = xhci_parse_exponent_interval(udev, ep);
1202                         break;
1203                 }
1204                 /*
1205                  * Fall through for interrupt endpoint interval decoding
1206                  * since it uses the same rules as low speed interrupt
1207                  * endpoints.
1208                  */
1209
1210         case USB_SPEED_LOW:
1211                 if (usb_endpoint_xfer_int(&ep->desc) ||
1212                     usb_endpoint_xfer_isoc(&ep->desc)) {
1213
1214                         interval = xhci_parse_frame_interval(udev, ep);
1215                 }
1216                 break;
1217
1218         default:
1219                 BUG();
1220         }
1221         return EP_INTERVAL(interval);
1222 }
1223
1224 /* The "Mult" field in the endpoint context is only set for SuperSpeed isoc eps.
1225  * High speed endpoint descriptors can define "the number of additional
1226  * transaction opportunities per microframe", but that goes in the Max Burst
1227  * endpoint context field.
1228  */
1229 static u32 xhci_get_endpoint_mult(struct usb_device *udev,
1230                 struct usb_host_endpoint *ep)
1231 {
1232         if (udev->speed != USB_SPEED_SUPER ||
1233                         !usb_endpoint_xfer_isoc(&ep->desc))
1234                 return 0;
1235         return ep->ss_ep_comp.bmAttributes;
1236 }
1237
1238 static u32 xhci_get_endpoint_type(struct usb_device *udev,
1239                 struct usb_host_endpoint *ep)
1240 {
1241         int in;
1242         u32 type;
1243
1244         in = usb_endpoint_dir_in(&ep->desc);
1245         if (usb_endpoint_xfer_control(&ep->desc)) {
1246                 type = EP_TYPE(CTRL_EP);
1247         } else if (usb_endpoint_xfer_bulk(&ep->desc)) {
1248                 if (in)
1249                         type = EP_TYPE(BULK_IN_EP);
1250                 else
1251                         type = EP_TYPE(BULK_OUT_EP);
1252         } else if (usb_endpoint_xfer_isoc(&ep->desc)) {
1253                 if (in)
1254                         type = EP_TYPE(ISOC_IN_EP);
1255                 else
1256                         type = EP_TYPE(ISOC_OUT_EP);
1257         } else if (usb_endpoint_xfer_int(&ep->desc)) {
1258                 if (in)
1259                         type = EP_TYPE(INT_IN_EP);
1260                 else
1261                         type = EP_TYPE(INT_OUT_EP);
1262         } else {
1263                 BUG();
1264         }
1265         return type;
1266 }
1267
1268 /* Return the maximum endpoint service interval time (ESIT) payload.
1269  * Basically, this is the maxpacket size, multiplied by the burst size
1270  * and mult size.
1271  */
1272 static u32 xhci_get_max_esit_payload(struct xhci_hcd *xhci,
1273                 struct usb_device *udev,
1274                 struct usb_host_endpoint *ep)
1275 {
1276         int max_burst;
1277         int max_packet;
1278
1279         /* Only applies for interrupt or isochronous endpoints */
1280         if (usb_endpoint_xfer_control(&ep->desc) ||
1281                         usb_endpoint_xfer_bulk(&ep->desc))
1282                 return 0;
1283
1284         if (udev->speed == USB_SPEED_SUPER)
1285                 return le16_to_cpu(ep->ss_ep_comp.wBytesPerInterval);
1286
1287         max_packet = GET_MAX_PACKET(usb_endpoint_maxp(&ep->desc));
1288         max_burst = (usb_endpoint_maxp(&ep->desc) & 0x1800) >> 11;
1289         /* A 0 in max burst means 1 transfer per ESIT */
1290         return max_packet * (max_burst + 1);
1291 }
1292
1293 /* Set up an endpoint with one ring segment.  Do not allocate stream rings.
1294  * Drivers will have to call usb_alloc_streams() to do that.
1295  */
1296 int xhci_endpoint_init(struct xhci_hcd *xhci,
1297                 struct xhci_virt_device *virt_dev,
1298                 struct usb_device *udev,
1299                 struct usb_host_endpoint *ep,
1300                 gfp_t mem_flags)
1301 {
1302         unsigned int ep_index;
1303         struct xhci_ep_ctx *ep_ctx;
1304         struct xhci_ring *ep_ring;
1305         unsigned int max_packet;
1306         unsigned int max_burst;
1307         u32 max_esit_payload;
1308
1309         ep_index = xhci_get_endpoint_index(&ep->desc);
1310         ep_ctx = xhci_get_ep_ctx(xhci, virt_dev->in_ctx, ep_index);
1311
1312         /* Set up the endpoint ring */
1313         /*
1314          * Isochronous endpoint ring needs bigger size because one isoc URB
1315          * carries multiple packets and it will insert multiple tds to the
1316          * ring.
1317          * This should be replaced with dynamic ring resizing in the future.
1318          */
1319         if (usb_endpoint_xfer_isoc(&ep->desc))
1320                 virt_dev->eps[ep_index].new_ring =
1321                         xhci_ring_alloc(xhci, 8, true, true, mem_flags);
1322         else
1323                 virt_dev->eps[ep_index].new_ring =
1324                         xhci_ring_alloc(xhci, 1, true, false, mem_flags);
1325         if (!virt_dev->eps[ep_index].new_ring) {
1326                 /* Attempt to use the ring cache */
1327                 if (virt_dev->num_rings_cached == 0)
1328                         return -ENOMEM;
1329                 virt_dev->eps[ep_index].new_ring =
1330                         virt_dev->ring_cache[virt_dev->num_rings_cached];
1331                 virt_dev->ring_cache[virt_dev->num_rings_cached] = NULL;
1332                 virt_dev->num_rings_cached--;
1333                 xhci_reinit_cached_ring(xhci, virt_dev->eps[ep_index].new_ring,
1334                         usb_endpoint_xfer_isoc(&ep->desc) ? true : false);
1335         }
1336         virt_dev->eps[ep_index].skip = false;
1337         ep_ring = virt_dev->eps[ep_index].new_ring;
1338         ep_ctx->deq = cpu_to_le64(ep_ring->first_seg->dma | ep_ring->cycle_state);
1339
1340         ep_ctx->ep_info = cpu_to_le32(xhci_get_endpoint_interval(udev, ep)
1341                                       | EP_MULT(xhci_get_endpoint_mult(udev, ep)));
1342
1343         /* FIXME dig Mult and streams info out of ep companion desc */
1344
1345         /* Allow 3 retries for everything but isoc;
1346          * CErr shall be set to 0 for Isoch endpoints.
1347          */
1348         if (!usb_endpoint_xfer_isoc(&ep->desc))
1349                 ep_ctx->ep_info2 = cpu_to_le32(ERROR_COUNT(3));
1350         else
1351                 ep_ctx->ep_info2 = cpu_to_le32(ERROR_COUNT(0));
1352
1353         ep_ctx->ep_info2 |= cpu_to_le32(xhci_get_endpoint_type(udev, ep));
1354
1355         /* Set the max packet size and max burst */
1356         max_packet = GET_MAX_PACKET(usb_endpoint_maxp(&ep->desc));
1357         max_burst = 0;
1358         switch (udev->speed) {
1359         case USB_SPEED_SUPER:
1360                 /* dig out max burst from ep companion desc */
1361                 max_burst = ep->ss_ep_comp.bMaxBurst;
1362                 break;
1363         case USB_SPEED_HIGH:
1364                 /* Some devices get this wrong */
1365                 if (usb_endpoint_xfer_bulk(&ep->desc))
1366                         max_packet = 512;
1367                 /* bits 11:12 specify the number of additional transaction
1368                  * opportunities per microframe (USB 2.0, section 9.6.6)
1369                  */
1370                 if (usb_endpoint_xfer_isoc(&ep->desc) ||
1371                                 usb_endpoint_xfer_int(&ep->desc)) {
1372                         max_burst = (usb_endpoint_maxp(&ep->desc)
1373                                      & 0x1800) >> 11;
1374                 }
1375                 break;
1376         case USB_SPEED_FULL:
1377         case USB_SPEED_LOW:
1378                 break;
1379         default:
1380                 BUG();
1381         }
1382         ep_ctx->ep_info2 |= cpu_to_le32(MAX_PACKET(max_packet) |
1383                         MAX_BURST(max_burst));
1384         max_esit_payload = xhci_get_max_esit_payload(xhci, udev, ep);
1385         ep_ctx->tx_info = cpu_to_le32(MAX_ESIT_PAYLOAD_FOR_EP(max_esit_payload));
1386
1387         /*
1388          * XXX no idea how to calculate the average TRB buffer length for bulk
1389          * endpoints, as the driver gives us no clue how big each scatter gather
1390          * list entry (or buffer) is going to be.
1391          *
1392          * For isochronous and interrupt endpoints, we set it to the max
1393          * available, until we have new API in the USB core to allow drivers to
1394          * declare how much bandwidth they actually need.
1395          *
1396          * Normally, it would be calculated by taking the total of the buffer
1397          * lengths in the TD and then dividing by the number of TRBs in a TD,
1398          * including link TRBs, No-op TRBs, and Event data TRBs.  Since we don't
1399          * use Event Data TRBs, and we don't chain in a link TRB on short
1400          * transfers, we're basically dividing by 1.
1401          *
1402          * xHCI 1.0 specification indicates that the Average TRB Length should
1403          * be set to 8 for control endpoints.
1404          */
1405         if (usb_endpoint_xfer_control(&ep->desc) && xhci->hci_version == 0x100)
1406                 ep_ctx->tx_info |= cpu_to_le32(AVG_TRB_LENGTH_FOR_EP(8));
1407         else
1408                 ep_ctx->tx_info |=
1409                          cpu_to_le32(AVG_TRB_LENGTH_FOR_EP(max_esit_payload));
1410
1411         /* FIXME Debug endpoint context */
1412         return 0;
1413 }
1414
1415 void xhci_endpoint_zero(struct xhci_hcd *xhci,
1416                 struct xhci_virt_device *virt_dev,
1417                 struct usb_host_endpoint *ep)
1418 {
1419         unsigned int ep_index;
1420         struct xhci_ep_ctx *ep_ctx;
1421
1422         ep_index = xhci_get_endpoint_index(&ep->desc);
1423         ep_ctx = xhci_get_ep_ctx(xhci, virt_dev->in_ctx, ep_index);
1424
1425         ep_ctx->ep_info = 0;
1426         ep_ctx->ep_info2 = 0;
1427         ep_ctx->deq = 0;
1428         ep_ctx->tx_info = 0;
1429         /* Don't free the endpoint ring until the set interface or configuration
1430          * request succeeds.
1431          */
1432 }
1433
1434 void xhci_clear_endpoint_bw_info(struct xhci_bw_info *bw_info)
1435 {
1436         bw_info->ep_interval = 0;
1437         bw_info->mult = 0;
1438         bw_info->num_packets = 0;
1439         bw_info->max_packet_size = 0;
1440         bw_info->type = 0;
1441         bw_info->max_esit_payload = 0;
1442 }
1443
1444 void xhci_update_bw_info(struct xhci_hcd *xhci,
1445                 struct xhci_container_ctx *in_ctx,
1446                 struct xhci_input_control_ctx *ctrl_ctx,
1447                 struct xhci_virt_device *virt_dev)
1448 {
1449         struct xhci_bw_info *bw_info;
1450         struct xhci_ep_ctx *ep_ctx;
1451         unsigned int ep_type;
1452         int i;
1453
1454         for (i = 1; i < 31; ++i) {
1455                 bw_info = &virt_dev->eps[i].bw_info;
1456
1457                 /* We can't tell what endpoint type is being dropped, but
1458                  * unconditionally clearing the bandwidth info for non-periodic
1459                  * endpoints should be harmless because the info will never be
1460                  * set in the first place.
1461                  */
1462                 if (!EP_IS_ADDED(ctrl_ctx, i) && EP_IS_DROPPED(ctrl_ctx, i)) {
1463                         /* Dropped endpoint */
1464                         xhci_clear_endpoint_bw_info(bw_info);
1465                         continue;
1466                 }
1467
1468                 if (EP_IS_ADDED(ctrl_ctx, i)) {
1469                         ep_ctx = xhci_get_ep_ctx(xhci, in_ctx, i);
1470                         ep_type = CTX_TO_EP_TYPE(le32_to_cpu(ep_ctx->ep_info2));
1471
1472                         /* Ignore non-periodic endpoints */
1473                         if (ep_type != ISOC_OUT_EP && ep_type != INT_OUT_EP &&
1474                                         ep_type != ISOC_IN_EP &&
1475                                         ep_type != INT_IN_EP)
1476                                 continue;
1477
1478                         /* Added or changed endpoint */
1479                         bw_info->ep_interval = CTX_TO_EP_INTERVAL(
1480                                         le32_to_cpu(ep_ctx->ep_info));
1481                         /* Number of packets and mult are zero-based in the
1482                          * input context, but we want one-based for the
1483                          * interval table.
1484                          */
1485                         bw_info->mult = CTX_TO_EP_MULT(
1486                                         le32_to_cpu(ep_ctx->ep_info)) + 1;
1487                         bw_info->num_packets = CTX_TO_MAX_BURST(
1488                                         le32_to_cpu(ep_ctx->ep_info2)) + 1;
1489                         bw_info->max_packet_size = MAX_PACKET_DECODED(
1490                                         le32_to_cpu(ep_ctx->ep_info2));
1491                         bw_info->type = ep_type;
1492                         bw_info->max_esit_payload = CTX_TO_MAX_ESIT_PAYLOAD(
1493                                         le32_to_cpu(ep_ctx->tx_info));
1494                 }
1495         }
1496 }
1497
1498 /* Copy output xhci_ep_ctx to the input xhci_ep_ctx copy.
1499  * Useful when you want to change one particular aspect of the endpoint and then
1500  * issue a configure endpoint command.
1501  */
1502 void xhci_endpoint_copy(struct xhci_hcd *xhci,
1503                 struct xhci_container_ctx *in_ctx,
1504                 struct xhci_container_ctx *out_ctx,
1505                 unsigned int ep_index)
1506 {
1507         struct xhci_ep_ctx *out_ep_ctx;
1508         struct xhci_ep_ctx *in_ep_ctx;
1509
1510         out_ep_ctx = xhci_get_ep_ctx(xhci, out_ctx, ep_index);
1511         in_ep_ctx = xhci_get_ep_ctx(xhci, in_ctx, ep_index);
1512
1513         in_ep_ctx->ep_info = out_ep_ctx->ep_info;
1514         in_ep_ctx->ep_info2 = out_ep_ctx->ep_info2;
1515         in_ep_ctx->deq = out_ep_ctx->deq;
1516         in_ep_ctx->tx_info = out_ep_ctx->tx_info;
1517 }
1518
1519 /* Copy output xhci_slot_ctx to the input xhci_slot_ctx.
1520  * Useful when you want to change one particular aspect of the endpoint and then
1521  * issue a configure endpoint command.  Only the context entries field matters,
1522  * but we'll copy the whole thing anyway.
1523  */
1524 void xhci_slot_copy(struct xhci_hcd *xhci,
1525                 struct xhci_container_ctx *in_ctx,
1526                 struct xhci_container_ctx *out_ctx)
1527 {
1528         struct xhci_slot_ctx *in_slot_ctx;
1529         struct xhci_slot_ctx *out_slot_ctx;
1530
1531         in_slot_ctx = xhci_get_slot_ctx(xhci, in_ctx);
1532         out_slot_ctx = xhci_get_slot_ctx(xhci, out_ctx);
1533
1534         in_slot_ctx->dev_info = out_slot_ctx->dev_info;
1535         in_slot_ctx->dev_info2 = out_slot_ctx->dev_info2;
1536         in_slot_ctx->tt_info = out_slot_ctx->tt_info;
1537         in_slot_ctx->dev_state = out_slot_ctx->dev_state;
1538 }
1539
1540 /* Set up the scratchpad buffer array and scratchpad buffers, if needed. */
1541 static int scratchpad_alloc(struct xhci_hcd *xhci, gfp_t flags)
1542 {
1543         int i;
1544         struct device *dev = xhci_to_hcd(xhci)->self.controller;
1545         int num_sp = HCS_MAX_SCRATCHPAD(xhci->hcs_params2);
1546
1547         xhci_dbg(xhci, "Allocating %d scratchpad buffers\n", num_sp);
1548
1549         if (!num_sp)
1550                 return 0;
1551
1552         xhci->scratchpad = kzalloc(sizeof(*xhci->scratchpad), flags);
1553         if (!xhci->scratchpad)
1554                 goto fail_sp;
1555
1556         xhci->scratchpad->sp_array = dma_alloc_coherent(dev,
1557                                      num_sp * sizeof(u64),
1558                                      &xhci->scratchpad->sp_dma, flags);
1559         if (!xhci->scratchpad->sp_array)
1560                 goto fail_sp2;
1561
1562         xhci->scratchpad->sp_buffers = kzalloc(sizeof(void *) * num_sp, flags);
1563         if (!xhci->scratchpad->sp_buffers)
1564                 goto fail_sp3;
1565
1566         xhci->scratchpad->sp_dma_buffers =
1567                 kzalloc(sizeof(dma_addr_t) * num_sp, flags);
1568
1569         if (!xhci->scratchpad->sp_dma_buffers)
1570                 goto fail_sp4;
1571
1572         xhci->dcbaa->dev_context_ptrs[0] = cpu_to_le64(xhci->scratchpad->sp_dma);
1573         for (i = 0; i < num_sp; i++) {
1574                 dma_addr_t dma;
1575                 void *buf = dma_alloc_coherent(dev, xhci->page_size, &dma,
1576                                 flags);
1577                 if (!buf)
1578                         goto fail_sp5;
1579
1580                 xhci->scratchpad->sp_array[i] = dma;
1581                 xhci->scratchpad->sp_buffers[i] = buf;
1582                 xhci->scratchpad->sp_dma_buffers[i] = dma;
1583         }
1584
1585         return 0;
1586
1587  fail_sp5:
1588         for (i = i - 1; i >= 0; i--) {
1589                 dma_free_coherent(dev, xhci->page_size,
1590                                     xhci->scratchpad->sp_buffers[i],
1591                                     xhci->scratchpad->sp_dma_buffers[i]);
1592         }
1593         kfree(xhci->scratchpad->sp_dma_buffers);
1594
1595  fail_sp4:
1596         kfree(xhci->scratchpad->sp_buffers);
1597
1598  fail_sp3:
1599         dma_free_coherent(dev, num_sp * sizeof(u64),
1600                             xhci->scratchpad->sp_array,
1601                             xhci->scratchpad->sp_dma);
1602
1603  fail_sp2:
1604         kfree(xhci->scratchpad);
1605         xhci->scratchpad = NULL;
1606
1607  fail_sp:
1608         return -ENOMEM;
1609 }
1610
1611 static void scratchpad_free(struct xhci_hcd *xhci)
1612 {
1613         int num_sp;
1614         int i;
1615         struct pci_dev  *pdev = to_pci_dev(xhci_to_hcd(xhci)->self.controller);
1616
1617         if (!xhci->scratchpad)
1618                 return;
1619
1620         num_sp = HCS_MAX_SCRATCHPAD(xhci->hcs_params2);
1621
1622         for (i = 0; i < num_sp; i++) {
1623                 dma_free_coherent(&pdev->dev, xhci->page_size,
1624                                     xhci->scratchpad->sp_buffers[i],
1625                                     xhci->scratchpad->sp_dma_buffers[i]);
1626         }
1627         kfree(xhci->scratchpad->sp_dma_buffers);
1628         kfree(xhci->scratchpad->sp_buffers);
1629         dma_free_coherent(&pdev->dev, num_sp * sizeof(u64),
1630                             xhci->scratchpad->sp_array,
1631                             xhci->scratchpad->sp_dma);
1632         kfree(xhci->scratchpad);
1633         xhci->scratchpad = NULL;
1634 }
1635
1636 struct xhci_command *xhci_alloc_command(struct xhci_hcd *xhci,
1637                 bool allocate_in_ctx, bool allocate_completion,
1638                 gfp_t mem_flags)
1639 {
1640         struct xhci_command *command;
1641
1642         command = kzalloc(sizeof(*command), mem_flags);
1643         if (!command)
1644                 return NULL;
1645
1646         if (allocate_in_ctx) {
1647                 command->in_ctx =
1648                         xhci_alloc_container_ctx(xhci, XHCI_CTX_TYPE_INPUT,
1649                                         mem_flags);
1650                 if (!command->in_ctx) {
1651                         kfree(command);
1652                         return NULL;
1653                 }
1654         }
1655
1656         if (allocate_completion) {
1657                 command->completion =
1658                         kzalloc(sizeof(struct completion), mem_flags);
1659                 if (!command->completion) {
1660                         xhci_free_container_ctx(xhci, command->in_ctx);
1661                         kfree(command);
1662                         return NULL;
1663                 }
1664                 init_completion(command->completion);
1665         }
1666
1667         command->status = 0;
1668         INIT_LIST_HEAD(&command->cmd_list);
1669         return command;
1670 }
1671
1672 void xhci_urb_free_priv(struct xhci_hcd *xhci, struct urb_priv *urb_priv)
1673 {
1674         if (urb_priv) {
1675                 kfree(urb_priv->td[0]);
1676                 kfree(urb_priv);
1677         }
1678 }
1679
1680 void xhci_free_command(struct xhci_hcd *xhci,
1681                 struct xhci_command *command)
1682 {
1683         xhci_free_container_ctx(xhci,
1684                         command->in_ctx);
1685         kfree(command->completion);
1686         kfree(command);
1687 }
1688
1689 void xhci_mem_cleanup(struct xhci_hcd *xhci)
1690 {
1691         struct pci_dev  *pdev = to_pci_dev(xhci_to_hcd(xhci)->self.controller);
1692         struct dev_info *dev_info, *next;
1693         struct xhci_cd  *cur_cd, *next_cd;
1694         unsigned long   flags;
1695         int size;
1696         int i, j, num_ports;
1697
1698         /* Free the Event Ring Segment Table and the actual Event Ring */
1699         size = sizeof(struct xhci_erst_entry)*(xhci->erst.num_entries);
1700         if (xhci->erst.entries)
1701                 dma_free_coherent(&pdev->dev, size,
1702                                 xhci->erst.entries, xhci->erst.erst_dma_addr);
1703         xhci->erst.entries = NULL;
1704         xhci_dbg(xhci, "Freed ERST\n");
1705         if (xhci->event_ring)
1706                 xhci_ring_free(xhci, xhci->event_ring);
1707         xhci->event_ring = NULL;
1708         xhci_dbg(xhci, "Freed event ring\n");
1709
1710         xhci->cmd_ring_reserved_trbs = 0;
1711         if (xhci->cmd_ring)
1712                 xhci_ring_free(xhci, xhci->cmd_ring);
1713         xhci->cmd_ring = NULL;
1714         xhci_dbg(xhci, "Freed command ring\n");
1715         list_for_each_entry_safe(cur_cd, next_cd,
1716                         &xhci->cancel_cmd_list, cancel_cmd_list) {
1717                 list_del(&cur_cd->cancel_cmd_list);
1718                 kfree(cur_cd);
1719         }
1720
1721         for (i = 1; i < MAX_HC_SLOTS; ++i)
1722                 xhci_free_virt_device(xhci, i);
1723
1724         if (xhci->segment_pool)
1725                 dma_pool_destroy(xhci->segment_pool);
1726         xhci->segment_pool = NULL;
1727         xhci_dbg(xhci, "Freed segment pool\n");
1728
1729         if (xhci->device_pool)
1730                 dma_pool_destroy(xhci->device_pool);
1731         xhci->device_pool = NULL;
1732         xhci_dbg(xhci, "Freed device context pool\n");
1733
1734         if (xhci->small_streams_pool)
1735                 dma_pool_destroy(xhci->small_streams_pool);
1736         xhci->small_streams_pool = NULL;
1737         xhci_dbg(xhci, "Freed small stream array pool\n");
1738
1739         if (xhci->medium_streams_pool)
1740                 dma_pool_destroy(xhci->medium_streams_pool);
1741         xhci->medium_streams_pool = NULL;
1742         xhci_dbg(xhci, "Freed medium stream array pool\n");
1743
1744         if (xhci->dcbaa)
1745                 dma_free_coherent(&pdev->dev, sizeof(*xhci->dcbaa),
1746                                 xhci->dcbaa, xhci->dcbaa->dma);
1747         xhci->dcbaa = NULL;
1748
1749         scratchpad_free(xhci);
1750
1751         spin_lock_irqsave(&xhci->lock, flags);
1752         list_for_each_entry_safe(dev_info, next, &xhci->lpm_failed_devs, list) {
1753                 list_del(&dev_info->list);
1754                 kfree(dev_info);
1755         }
1756         spin_unlock_irqrestore(&xhci->lock, flags);
1757
1758         if (!xhci->rh_bw)
1759                 goto no_bw;
1760
1761         num_ports = HCS_MAX_PORTS(xhci->hcs_params1);
1762         for (i = 0; i < num_ports; i++) {
1763                 struct xhci_interval_bw_table *bwt = &xhci->rh_bw[i].bw_table;
1764                 for (j = 0; j < XHCI_MAX_INTERVAL; j++) {
1765                         struct list_head *ep = &bwt->interval_bw[j].endpoints;
1766                         while (!list_empty(ep))
1767                                 list_del_init(ep->next);
1768                 }
1769         }
1770
1771         for (i = 0; i < num_ports; i++) {
1772                 struct xhci_tt_bw_info *tt, *n;
1773                 list_for_each_entry_safe(tt, n, &xhci->rh_bw[i].tts, tt_list) {
1774                         list_del(&tt->tt_list);
1775                         kfree(tt);
1776                 }
1777         }
1778
1779 no_bw:
1780         xhci->num_usb2_ports = 0;
1781         xhci->num_usb3_ports = 0;
1782         xhci->num_active_eps = 0;
1783         kfree(xhci->usb2_ports);
1784         kfree(xhci->usb3_ports);
1785         kfree(xhci->port_array);
1786         kfree(xhci->rh_bw);
1787
1788         xhci->page_size = 0;
1789         xhci->page_shift = 0;
1790         xhci->bus_state[0].bus_suspended = 0;
1791         xhci->bus_state[1].bus_suspended = 0;
1792 }
1793
1794 static int xhci_test_trb_in_td(struct xhci_hcd *xhci,
1795                 struct xhci_segment *input_seg,
1796                 union xhci_trb *start_trb,
1797                 union xhci_trb *end_trb,
1798                 dma_addr_t input_dma,
1799                 struct xhci_segment *result_seg,
1800                 char *test_name, int test_number)
1801 {
1802         unsigned long long start_dma;
1803         unsigned long long end_dma;
1804         struct xhci_segment *seg;
1805
1806         start_dma = xhci_trb_virt_to_dma(input_seg, start_trb);
1807         end_dma = xhci_trb_virt_to_dma(input_seg, end_trb);
1808
1809         seg = trb_in_td(input_seg, start_trb, end_trb, input_dma);
1810         if (seg != result_seg) {
1811                 xhci_warn(xhci, "WARN: %s TRB math test %d failed!\n",
1812                                 test_name, test_number);
1813                 xhci_warn(xhci, "Tested TRB math w/ seg %p and "
1814                                 "input DMA 0x%llx\n",
1815                                 input_seg,
1816                                 (unsigned long long) input_dma);
1817                 xhci_warn(xhci, "starting TRB %p (0x%llx DMA), "
1818                                 "ending TRB %p (0x%llx DMA)\n",
1819                                 start_trb, start_dma,
1820                                 end_trb, end_dma);
1821                 xhci_warn(xhci, "Expected seg %p, got seg %p\n",
1822                                 result_seg, seg);
1823                 return -1;
1824         }
1825         return 0;
1826 }
1827
1828 /* TRB math checks for xhci_trb_in_td(), using the command and event rings. */
1829 static int xhci_check_trb_in_td_math(struct xhci_hcd *xhci, gfp_t mem_flags)
1830 {
1831         struct {
1832                 dma_addr_t              input_dma;
1833                 struct xhci_segment     *result_seg;
1834         } simple_test_vector [] = {
1835                 /* A zeroed DMA field should fail */
1836                 { 0, NULL },
1837                 /* One TRB before the ring start should fail */
1838                 { xhci->event_ring->first_seg->dma - 16, NULL },
1839                 /* One byte before the ring start should fail */
1840                 { xhci->event_ring->first_seg->dma - 1, NULL },
1841                 /* Starting TRB should succeed */
1842                 { xhci->event_ring->first_seg->dma, xhci->event_ring->first_seg },
1843                 /* Ending TRB should succeed */
1844                 { xhci->event_ring->first_seg->dma + (TRBS_PER_SEGMENT - 1)*16,
1845                         xhci->event_ring->first_seg },
1846                 /* One byte after the ring end should fail */
1847                 { xhci->event_ring->first_seg->dma + (TRBS_PER_SEGMENT - 1)*16 + 1, NULL },
1848                 /* One TRB after the ring end should fail */
1849                 { xhci->event_ring->first_seg->dma + (TRBS_PER_SEGMENT)*16, NULL },
1850                 /* An address of all ones should fail */
1851                 { (dma_addr_t) (~0), NULL },
1852         };
1853         struct {
1854                 struct xhci_segment     *input_seg;
1855                 union xhci_trb          *start_trb;
1856                 union xhci_trb          *end_trb;
1857                 dma_addr_t              input_dma;
1858                 struct xhci_segment     *result_seg;
1859         } complex_test_vector [] = {
1860                 /* Test feeding a valid DMA address from a different ring */
1861                 {       .input_seg = xhci->event_ring->first_seg,
1862                         .start_trb = xhci->event_ring->first_seg->trbs,
1863                         .end_trb = &xhci->event_ring->first_seg->trbs[TRBS_PER_SEGMENT - 1],
1864                         .input_dma = xhci->cmd_ring->first_seg->dma,
1865                         .result_seg = NULL,
1866                 },
1867                 /* Test feeding a valid end TRB from a different ring */
1868                 {       .input_seg = xhci->event_ring->first_seg,
1869                         .start_trb = xhci->event_ring->first_seg->trbs,
1870                         .end_trb = &xhci->cmd_ring->first_seg->trbs[TRBS_PER_SEGMENT - 1],
1871                         .input_dma = xhci->cmd_ring->first_seg->dma,
1872                         .result_seg = NULL,
1873                 },
1874                 /* Test feeding a valid start and end TRB from a different ring */
1875                 {       .input_seg = xhci->event_ring->first_seg,
1876                         .start_trb = xhci->cmd_ring->first_seg->trbs,
1877                         .end_trb = &xhci->cmd_ring->first_seg->trbs[TRBS_PER_SEGMENT - 1],
1878                         .input_dma = xhci->cmd_ring->first_seg->dma,
1879                         .result_seg = NULL,
1880                 },
1881                 /* TRB in this ring, but after this TD */
1882                 {       .input_seg = xhci->event_ring->first_seg,
1883                         .start_trb = &xhci->event_ring->first_seg->trbs[0],
1884                         .end_trb = &xhci->event_ring->first_seg->trbs[3],
1885                         .input_dma = xhci->event_ring->first_seg->dma + 4*16,
1886                         .result_seg = NULL,
1887                 },
1888                 /* TRB in this ring, but before this TD */
1889                 {       .input_seg = xhci->event_ring->first_seg,
1890                         .start_trb = &xhci->event_ring->first_seg->trbs[3],
1891                         .end_trb = &xhci->event_ring->first_seg->trbs[6],
1892                         .input_dma = xhci->event_ring->first_seg->dma + 2*16,
1893                         .result_seg = NULL,
1894                 },
1895                 /* TRB in this ring, but after this wrapped TD */
1896                 {       .input_seg = xhci->event_ring->first_seg,
1897                         .start_trb = &xhci->event_ring->first_seg->trbs[TRBS_PER_SEGMENT - 3],
1898                         .end_trb = &xhci->event_ring->first_seg->trbs[1],
1899                         .input_dma = xhci->event_ring->first_seg->dma + 2*16,
1900                         .result_seg = NULL,
1901                 },
1902                 /* TRB in this ring, but before this wrapped TD */
1903                 {       .input_seg = xhci->event_ring->first_seg,
1904                         .start_trb = &xhci->event_ring->first_seg->trbs[TRBS_PER_SEGMENT - 3],
1905                         .end_trb = &xhci->event_ring->first_seg->trbs[1],
1906                         .input_dma = xhci->event_ring->first_seg->dma + (TRBS_PER_SEGMENT - 4)*16,
1907                         .result_seg = NULL,
1908                 },
1909                 /* TRB not in this ring, and we have a wrapped TD */
1910                 {       .input_seg = xhci->event_ring->first_seg,
1911                         .start_trb = &xhci->event_ring->first_seg->trbs[TRBS_PER_SEGMENT - 3],
1912                         .end_trb = &xhci->event_ring->first_seg->trbs[1],
1913                         .input_dma = xhci->cmd_ring->first_seg->dma + 2*16,
1914                         .result_seg = NULL,
1915                 },
1916         };
1917
1918         unsigned int num_tests;
1919         int i, ret;
1920
1921         num_tests = ARRAY_SIZE(simple_test_vector);
1922         for (i = 0; i < num_tests; i++) {
1923                 ret = xhci_test_trb_in_td(xhci,
1924                                 xhci->event_ring->first_seg,
1925                                 xhci->event_ring->first_seg->trbs,
1926                                 &xhci->event_ring->first_seg->trbs[TRBS_PER_SEGMENT - 1],
1927                                 simple_test_vector[i].input_dma,
1928                                 simple_test_vector[i].result_seg,
1929                                 "Simple", i);
1930                 if (ret < 0)
1931                         return ret;
1932         }
1933
1934         num_tests = ARRAY_SIZE(complex_test_vector);
1935         for (i = 0; i < num_tests; i++) {
1936                 ret = xhci_test_trb_in_td(xhci,
1937                                 complex_test_vector[i].input_seg,
1938                                 complex_test_vector[i].start_trb,
1939                                 complex_test_vector[i].end_trb,
1940                                 complex_test_vector[i].input_dma,
1941                                 complex_test_vector[i].result_seg,
1942                                 "Complex", i);
1943                 if (ret < 0)
1944                         return ret;
1945         }
1946         xhci_dbg(xhci, "TRB math tests passed.\n");
1947         return 0;
1948 }
1949
1950 static void xhci_set_hc_event_deq(struct xhci_hcd *xhci)
1951 {
1952         u64 temp;
1953         dma_addr_t deq;
1954
1955         deq = xhci_trb_virt_to_dma(xhci->event_ring->deq_seg,
1956                         xhci->event_ring->dequeue);
1957         if (deq == 0 && !in_interrupt())
1958                 xhci_warn(xhci, "WARN something wrong with SW event ring "
1959                                 "dequeue ptr.\n");
1960         /* Update HC event ring dequeue pointer */
1961         temp = xhci_read_64(xhci, &xhci->ir_set->erst_dequeue);
1962         temp &= ERST_PTR_MASK;
1963         /* Don't clear the EHB bit (which is RW1C) because
1964          * there might be more events to service.
1965          */
1966         temp &= ~ERST_EHB;
1967         xhci_dbg(xhci, "// Write event ring dequeue pointer, "
1968                         "preserving EHB bit\n");
1969         xhci_write_64(xhci, ((u64) deq & (u64) ~ERST_PTR_MASK) | temp,
1970                         &xhci->ir_set->erst_dequeue);
1971 }
1972
1973 static void xhci_add_in_port(struct xhci_hcd *xhci, unsigned int num_ports,
1974                 __le32 __iomem *addr, u8 major_revision)
1975 {
1976         u32 temp, port_offset, port_count;
1977         int i;
1978
1979         if (major_revision > 0x03) {
1980                 xhci_warn(xhci, "Ignoring unknown port speed, "
1981                                 "Ext Cap %p, revision = 0x%x\n",
1982                                 addr, major_revision);
1983                 /* Ignoring port protocol we can't understand. FIXME */
1984                 return;
1985         }
1986
1987         /* Port offset and count in the third dword, see section 7.2 */
1988         temp = xhci_readl(xhci, addr + 2);
1989         port_offset = XHCI_EXT_PORT_OFF(temp);
1990         port_count = XHCI_EXT_PORT_COUNT(temp);
1991         xhci_dbg(xhci, "Ext Cap %p, port offset = %u, "
1992                         "count = %u, revision = 0x%x\n",
1993                         addr, port_offset, port_count, major_revision);
1994         /* Port count includes the current port offset */
1995         if (port_offset == 0 || (port_offset + port_count - 1) > num_ports)
1996                 /* WTF? "Valid values are ‘1’ to MaxPorts" */
1997                 return;
1998
1999         /* Check the host's USB2 LPM capability */
2000         if ((xhci->hci_version == 0x96) && (major_revision != 0x03) &&
2001                         (temp & XHCI_L1C)) {
2002                 xhci_dbg(xhci, "xHCI 0.96: support USB2 software lpm\n");
2003                 xhci->sw_lpm_support = 1;
2004         }
2005
2006         if ((xhci->hci_version >= 0x100) && (major_revision != 0x03)) {
2007                 xhci_dbg(xhci, "xHCI 1.0: support USB2 software lpm\n");
2008                 xhci->sw_lpm_support = 1;
2009                 if (temp & XHCI_HLC) {
2010                         xhci_dbg(xhci, "xHCI 1.0: support USB2 hardware lpm\n");
2011                         xhci->hw_lpm_support = 1;
2012                 }
2013         }
2014
2015         port_offset--;
2016         for (i = port_offset; i < (port_offset + port_count); i++) {
2017                 /* Duplicate entry.  Ignore the port if the revisions differ. */
2018                 if (xhci->port_array[i] != 0) {
2019                         xhci_warn(xhci, "Duplicate port entry, Ext Cap %p,"
2020                                         " port %u\n", addr, i);
2021                         xhci_warn(xhci, "Port was marked as USB %u, "
2022                                         "duplicated as USB %u\n",
2023                                         xhci->port_array[i], major_revision);
2024                         /* Only adjust the roothub port counts if we haven't
2025                          * found a similar duplicate.
2026                          */
2027                         if (xhci->port_array[i] != major_revision &&
2028                                 xhci->port_array[i] != DUPLICATE_ENTRY) {
2029                                 if (xhci->port_array[i] == 0x03)
2030                                         xhci->num_usb3_ports--;
2031                                 else
2032                                         xhci->num_usb2_ports--;
2033                                 xhci->port_array[i] = DUPLICATE_ENTRY;
2034                         }
2035                         /* FIXME: Should we disable the port? */
2036                         continue;
2037                 }
2038                 xhci->port_array[i] = major_revision;
2039                 if (major_revision == 0x03)
2040                         xhci->num_usb3_ports++;
2041                 else
2042                         xhci->num_usb2_ports++;
2043         }
2044         /* FIXME: Should we disable ports not in the Extended Capabilities? */
2045 }
2046
2047 /*
2048  * Scan the Extended Capabilities for the "Supported Protocol Capabilities" that
2049  * specify what speeds each port is supposed to be.  We can't count on the port
2050  * speed bits in the PORTSC register being correct until a device is connected,
2051  * but we need to set up the two fake roothubs with the correct number of USB
2052  * 3.0 and USB 2.0 ports at host controller initialization time.
2053  */
2054 static int xhci_setup_port_arrays(struct xhci_hcd *xhci, gfp_t flags)
2055 {
2056         __le32 __iomem *addr;
2057         u32 offset;
2058         unsigned int num_ports;
2059         int i, j, port_index;
2060
2061         addr = &xhci->cap_regs->hcc_params;
2062         offset = XHCI_HCC_EXT_CAPS(xhci_readl(xhci, addr));
2063         if (offset == 0) {
2064                 xhci_err(xhci, "No Extended Capability registers, "
2065                                 "unable to set up roothub.\n");
2066                 return -ENODEV;
2067         }
2068
2069         num_ports = HCS_MAX_PORTS(xhci->hcs_params1);
2070         xhci->port_array = kzalloc(sizeof(*xhci->port_array)*num_ports, flags);
2071         if (!xhci->port_array)
2072                 return -ENOMEM;
2073
2074         xhci->rh_bw = kzalloc(sizeof(*xhci->rh_bw)*num_ports, flags);
2075         if (!xhci->rh_bw)
2076                 return -ENOMEM;
2077         for (i = 0; i < num_ports; i++) {
2078                 struct xhci_interval_bw_table *bw_table;
2079
2080                 INIT_LIST_HEAD(&xhci->rh_bw[i].tts);
2081                 bw_table = &xhci->rh_bw[i].bw_table;
2082                 for (j = 0; j < XHCI_MAX_INTERVAL; j++)
2083                         INIT_LIST_HEAD(&bw_table->interval_bw[j].endpoints);
2084         }
2085
2086         /*
2087          * For whatever reason, the first capability offset is from the
2088          * capability register base, not from the HCCPARAMS register.
2089          * See section 5.3.6 for offset calculation.
2090          */
2091         addr = &xhci->cap_regs->hc_capbase + offset;
2092         while (1) {
2093                 u32 cap_id;
2094
2095                 cap_id = xhci_readl(xhci, addr);
2096                 if (XHCI_EXT_CAPS_ID(cap_id) == XHCI_EXT_CAPS_PROTOCOL)
2097                         xhci_add_in_port(xhci, num_ports, addr,
2098                                         (u8) XHCI_EXT_PORT_MAJOR(cap_id));
2099                 offset = XHCI_EXT_CAPS_NEXT(cap_id);
2100                 if (!offset || (xhci->num_usb2_ports + xhci->num_usb3_ports)
2101                                 == num_ports)
2102                         break;
2103                 /*
2104                  * Once you're into the Extended Capabilities, the offset is
2105                  * always relative to the register holding the offset.
2106                  */
2107                 addr += offset;
2108         }
2109
2110         if (xhci->num_usb2_ports == 0 && xhci->num_usb3_ports == 0) {
2111                 xhci_warn(xhci, "No ports on the roothubs?\n");
2112                 return -ENODEV;
2113         }
2114         xhci_dbg(xhci, "Found %u USB 2.0 ports and %u USB 3.0 ports.\n",
2115                         xhci->num_usb2_ports, xhci->num_usb3_ports);
2116
2117         /* Place limits on the number of roothub ports so that the hub
2118          * descriptors aren't longer than the USB core will allocate.
2119          */
2120         if (xhci->num_usb3_ports > 15) {
2121                 xhci_dbg(xhci, "Limiting USB 3.0 roothub ports to 15.\n");
2122                 xhci->num_usb3_ports = 15;
2123         }
2124         if (xhci->num_usb2_ports > USB_MAXCHILDREN) {
2125                 xhci_dbg(xhci, "Limiting USB 2.0 roothub ports to %u.\n",
2126                                 USB_MAXCHILDREN);
2127                 xhci->num_usb2_ports = USB_MAXCHILDREN;
2128         }
2129
2130         /*
2131          * Note we could have all USB 3.0 ports, or all USB 2.0 ports.
2132          * Not sure how the USB core will handle a hub with no ports...
2133          */
2134         if (xhci->num_usb2_ports) {
2135                 xhci->usb2_ports = kmalloc(sizeof(*xhci->usb2_ports)*
2136                                 xhci->num_usb2_ports, flags);
2137                 if (!xhci->usb2_ports)
2138                         return -ENOMEM;
2139
2140                 port_index = 0;
2141                 for (i = 0; i < num_ports; i++) {
2142                         if (xhci->port_array[i] == 0x03 ||
2143                                         xhci->port_array[i] == 0 ||
2144                                         xhci->port_array[i] == DUPLICATE_ENTRY)
2145                                 continue;
2146
2147                         xhci->usb2_ports[port_index] =
2148                                 &xhci->op_regs->port_status_base +
2149                                 NUM_PORT_REGS*i;
2150                         xhci_dbg(xhci, "USB 2.0 port at index %u, "
2151                                         "addr = %p\n", i,
2152                                         xhci->usb2_ports[port_index]);
2153                         port_index++;
2154                         if (port_index == xhci->num_usb2_ports)
2155                                 break;
2156                 }
2157         }
2158         if (xhci->num_usb3_ports) {
2159                 xhci->usb3_ports = kmalloc(sizeof(*xhci->usb3_ports)*
2160                                 xhci->num_usb3_ports, flags);
2161                 if (!xhci->usb3_ports)
2162                         return -ENOMEM;
2163
2164                 port_index = 0;
2165                 for (i = 0; i < num_ports; i++)
2166                         if (xhci->port_array[i] == 0x03) {
2167                                 xhci->usb3_ports[port_index] =
2168                                         &xhci->op_regs->port_status_base +
2169                                         NUM_PORT_REGS*i;
2170                                 xhci_dbg(xhci, "USB 3.0 port at index %u, "
2171                                                 "addr = %p\n", i,
2172                                                 xhci->usb3_ports[port_index]);
2173                                 port_index++;
2174                                 if (port_index == xhci->num_usb3_ports)
2175                                         break;
2176                         }
2177         }
2178         return 0;
2179 }
2180
2181 int xhci_mem_init(struct xhci_hcd *xhci, gfp_t flags)
2182 {
2183         dma_addr_t      dma;
2184         struct device   *dev = xhci_to_hcd(xhci)->self.controller;
2185         unsigned int    val, val2;
2186         u64             val_64;
2187         struct xhci_segment     *seg;
2188         u32 page_size;
2189         int i;
2190
2191         INIT_LIST_HEAD(&xhci->lpm_failed_devs);
2192         INIT_LIST_HEAD(&xhci->cancel_cmd_list);
2193
2194         page_size = xhci_readl(xhci, &xhci->op_regs->page_size);
2195         xhci_dbg(xhci, "Supported page size register = 0x%x\n", page_size);
2196         for (i = 0; i < 16; i++) {
2197                 if ((0x1 & page_size) != 0)
2198                         break;
2199                 page_size = page_size >> 1;
2200         }
2201         if (i < 16)
2202                 xhci_dbg(xhci, "Supported page size of %iK\n", (1 << (i+12)) / 1024);
2203         else
2204                 xhci_warn(xhci, "WARN: no supported page size\n");
2205         /* Use 4K pages, since that's common and the minimum the HC supports */
2206         xhci->page_shift = 12;
2207         xhci->page_size = 1 << xhci->page_shift;
2208         xhci_dbg(xhci, "HCD page size set to %iK\n", xhci->page_size / 1024);
2209
2210         /*
2211          * Program the Number of Device Slots Enabled field in the CONFIG
2212          * register with the max value of slots the HC can handle.
2213          */
2214         val = HCS_MAX_SLOTS(xhci_readl(xhci, &xhci->cap_regs->hcs_params1));
2215         xhci_dbg(xhci, "// xHC can handle at most %d device slots.\n",
2216                         (unsigned int) val);
2217         val2 = xhci_readl(xhci, &xhci->op_regs->config_reg);
2218         val |= (val2 & ~HCS_SLOTS_MASK);
2219         xhci_dbg(xhci, "// Setting Max device slots reg = 0x%x.\n",
2220                         (unsigned int) val);
2221         xhci_writel(xhci, val, &xhci->op_regs->config_reg);
2222
2223         /*
2224          * Section 5.4.8 - doorbell array must be
2225          * "physically contiguous and 64-byte (cache line) aligned".
2226          */
2227         xhci->dcbaa = dma_alloc_coherent(dev, sizeof(*xhci->dcbaa), &dma,
2228                         GFP_KERNEL);
2229         if (!xhci->dcbaa)
2230                 goto fail;
2231         memset(xhci->dcbaa, 0, sizeof *(xhci->dcbaa));
2232         xhci->dcbaa->dma = dma;
2233         xhci_dbg(xhci, "// Device context base array address = 0x%llx (DMA), %p (virt)\n",
2234                         (unsigned long long)xhci->dcbaa->dma, xhci->dcbaa);
2235         xhci_write_64(xhci, dma, &xhci->op_regs->dcbaa_ptr);
2236
2237         /*
2238          * Initialize the ring segment pool.  The ring must be a contiguous
2239          * structure comprised of TRBs.  The TRBs must be 16 byte aligned,
2240          * however, the command ring segment needs 64-byte aligned segments,
2241          * so we pick the greater alignment need.
2242          */
2243         xhci->segment_pool = dma_pool_create("xHCI ring segments", dev,
2244                         SEGMENT_SIZE, 64, xhci->page_size);
2245
2246         /* See Table 46 and Note on Figure 55 */
2247         xhci->device_pool = dma_pool_create("xHCI input/output contexts", dev,
2248                         2112, 64, xhci->page_size);
2249         if (!xhci->segment_pool || !xhci->device_pool)
2250                 goto fail;
2251
2252         /* Linear stream context arrays don't have any boundary restrictions,
2253          * and only need to be 16-byte aligned.
2254          */
2255         xhci->small_streams_pool =
2256                 dma_pool_create("xHCI 256 byte stream ctx arrays",
2257                         dev, SMALL_STREAM_ARRAY_SIZE, 16, 0);
2258         xhci->medium_streams_pool =
2259                 dma_pool_create("xHCI 1KB stream ctx arrays",
2260                         dev, MEDIUM_STREAM_ARRAY_SIZE, 16, 0);
2261         /* Any stream context array bigger than MEDIUM_STREAM_ARRAY_SIZE
2262          * will be allocated with dma_alloc_coherent()
2263          */
2264
2265         if (!xhci->small_streams_pool || !xhci->medium_streams_pool)
2266                 goto fail;
2267
2268         /* Set up the command ring to have one segments for now. */
2269         xhci->cmd_ring = xhci_ring_alloc(xhci, 1, true, false, flags);
2270         if (!xhci->cmd_ring)
2271                 goto fail;
2272         xhci_dbg(xhci, "Allocated command ring at %p\n", xhci->cmd_ring);
2273         xhci_dbg(xhci, "First segment DMA is 0x%llx\n",
2274                         (unsigned long long)xhci->cmd_ring->first_seg->dma);
2275
2276         /* Set the address in the Command Ring Control register */
2277         val_64 = xhci_read_64(xhci, &xhci->op_regs->cmd_ring);
2278         val_64 = (val_64 & (u64) CMD_RING_RSVD_BITS) |
2279                 (xhci->cmd_ring->first_seg->dma & (u64) ~CMD_RING_RSVD_BITS) |
2280                 xhci->cmd_ring->cycle_state;
2281         xhci_dbg(xhci, "// Setting command ring address to 0x%x\n", val);
2282         xhci_write_64(xhci, val_64, &xhci->op_regs->cmd_ring);
2283         xhci_dbg_cmd_ptrs(xhci);
2284
2285         val = xhci_readl(xhci, &xhci->cap_regs->db_off);
2286         val &= DBOFF_MASK;
2287         xhci_dbg(xhci, "// Doorbell array is located at offset 0x%x"
2288                         " from cap regs base addr\n", val);
2289         xhci->dba = (void __iomem *) xhci->cap_regs + val;
2290         xhci_dbg_regs(xhci);
2291         xhci_print_run_regs(xhci);
2292         /* Set ir_set to interrupt register set 0 */
2293         xhci->ir_set = &xhci->run_regs->ir_set[0];
2294
2295         /*
2296          * Event ring setup: Allocate a normal ring, but also setup
2297          * the event ring segment table (ERST).  Section 4.9.3.
2298          */
2299         xhci_dbg(xhci, "// Allocating event ring\n");
2300         xhci->event_ring = xhci_ring_alloc(xhci, ERST_NUM_SEGS, false, false,
2301                                                 flags);
2302         if (!xhci->event_ring)
2303                 goto fail;
2304         if (xhci_check_trb_in_td_math(xhci, flags) < 0)
2305                 goto fail;
2306
2307         xhci->erst.entries = dma_alloc_coherent(dev,
2308                         sizeof(struct xhci_erst_entry) * ERST_NUM_SEGS, &dma,
2309                         GFP_KERNEL);
2310         if (!xhci->erst.entries)
2311                 goto fail;
2312         xhci_dbg(xhci, "// Allocated event ring segment table at 0x%llx\n",
2313                         (unsigned long long)dma);
2314
2315         memset(xhci->erst.entries, 0, sizeof(struct xhci_erst_entry)*ERST_NUM_SEGS);
2316         xhci->erst.num_entries = ERST_NUM_SEGS;
2317         xhci->erst.erst_dma_addr = dma;
2318         xhci_dbg(xhci, "Set ERST to 0; private num segs = %i, virt addr = %p, dma addr = 0x%llx\n",
2319                         xhci->erst.num_entries,
2320                         xhci->erst.entries,
2321                         (unsigned long long)xhci->erst.erst_dma_addr);
2322
2323         /* set ring base address and size for each segment table entry */
2324         for (val = 0, seg = xhci->event_ring->first_seg; val < ERST_NUM_SEGS; val++) {
2325                 struct xhci_erst_entry *entry = &xhci->erst.entries[val];
2326                 entry->seg_addr = cpu_to_le64(seg->dma);
2327                 entry->seg_size = cpu_to_le32(TRBS_PER_SEGMENT);
2328                 entry->rsvd = 0;
2329                 seg = seg->next;
2330         }
2331
2332         /* set ERST count with the number of entries in the segment table */
2333         val = xhci_readl(xhci, &xhci->ir_set->erst_size);
2334         val &= ERST_SIZE_MASK;
2335         val |= ERST_NUM_SEGS;
2336         xhci_dbg(xhci, "// Write ERST size = %i to ir_set 0 (some bits preserved)\n",
2337                         val);
2338         xhci_writel(xhci, val, &xhci->ir_set->erst_size);
2339
2340         xhci_dbg(xhci, "// Set ERST entries to point to event ring.\n");
2341         /* set the segment table base address */
2342         xhci_dbg(xhci, "// Set ERST base address for ir_set 0 = 0x%llx\n",
2343                         (unsigned long long)xhci->erst.erst_dma_addr);
2344         val_64 = xhci_read_64(xhci, &xhci->ir_set->erst_base);
2345         val_64 &= ERST_PTR_MASK;
2346         val_64 |= (xhci->erst.erst_dma_addr & (u64) ~ERST_PTR_MASK);
2347         xhci_write_64(xhci, val_64, &xhci->ir_set->erst_base);
2348
2349         /* Set the event ring dequeue address */
2350         xhci_set_hc_event_deq(xhci);
2351         xhci_dbg(xhci, "Wrote ERST address to ir_set 0.\n");
2352         xhci_print_ir_set(xhci, 0);
2353
2354         /*
2355          * XXX: Might need to set the Interrupter Moderation Register to
2356          * something other than the default (~1ms minimum between interrupts).
2357          * See section 5.5.1.2.
2358          */
2359         init_completion(&xhci->addr_dev);
2360         for (i = 0; i < MAX_HC_SLOTS; ++i)
2361                 xhci->devs[i] = NULL;
2362         for (i = 0; i < USB_MAXCHILDREN; ++i) {
2363                 xhci->bus_state[0].resume_done[i] = 0;
2364                 xhci->bus_state[1].resume_done[i] = 0;
2365         }
2366
2367         if (scratchpad_alloc(xhci, flags))
2368                 goto fail;
2369         if (xhci_setup_port_arrays(xhci, flags))
2370                 goto fail;
2371
2372         return 0;
2373
2374 fail:
2375         xhci_warn(xhci, "Couldn't initialize memory\n");
2376         xhci_halt(xhci);
2377         xhci_reset(xhci);
2378         xhci_mem_cleanup(xhci);
2379         return -ENOMEM;
2380 }