xref: /linux/drivers/usb/host/xhci-mem.c (revision 5d4a2e29fba5b2bef95b96a46b338ec4d76fa4fd)
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)
65 		return;
66 	if (seg->trbs) {
67 		xhci_dbg(xhci, "Freeing DMA segment at %p (virtual) 0x%llx (DMA)\n",
68 				seg->trbs, (unsigned long long)seg->dma);
69 		dma_pool_free(xhci->segment_pool, seg->trbs, seg->dma);
70 		seg->trbs = NULL;
71 	}
72 	xhci_dbg(xhci, "Freeing priv segment structure at %p\n", seg);
73 	kfree(seg);
74 }
75 
76 /*
77  * Make the prev segment point to the next segment.
78  *
79  * Change the last TRB in the prev segment to be a Link TRB which points to the
80  * DMA address of the next segment.  The caller needs to set any Link TRB
81  * related flags, such as End TRB, Toggle Cycle, and no snoop.
82  */
83 static void xhci_link_segments(struct xhci_hcd *xhci, struct xhci_segment *prev,
84 		struct xhci_segment *next, bool link_trbs)
85 {
86 	u32 val;
87 
88 	if (!prev || !next)
89 		return;
90 	prev->next = next;
91 	if (link_trbs) {
92 		prev->trbs[TRBS_PER_SEGMENT-1].link.segment_ptr = next->dma;
93 
94 		/* Set the last TRB in the segment to have a TRB type ID of Link TRB */
95 		val = prev->trbs[TRBS_PER_SEGMENT-1].link.control;
96 		val &= ~TRB_TYPE_BITMASK;
97 		val |= TRB_TYPE(TRB_LINK);
98 		/* Always set the chain bit with 0.95 hardware */
99 		if (xhci_link_trb_quirk(xhci))
100 			val |= TRB_CHAIN;
101 		prev->trbs[TRBS_PER_SEGMENT-1].link.control = val;
102 	}
103 	xhci_dbg(xhci, "Linking segment 0x%llx to segment 0x%llx (DMA)\n",
104 			(unsigned long long)prev->dma,
105 			(unsigned long long)next->dma);
106 }
107 
108 /* XXX: Do we need the hcd structure in all these functions? */
109 void xhci_ring_free(struct xhci_hcd *xhci, struct xhci_ring *ring)
110 {
111 	struct xhci_segment *seg;
112 	struct xhci_segment *first_seg;
113 
114 	if (!ring || !ring->first_seg)
115 		return;
116 	first_seg = ring->first_seg;
117 	seg = first_seg->next;
118 	xhci_dbg(xhci, "Freeing ring at %p\n", ring);
119 	while (seg != first_seg) {
120 		struct xhci_segment *next = seg->next;
121 		xhci_segment_free(xhci, seg);
122 		seg = next;
123 	}
124 	xhci_segment_free(xhci, first_seg);
125 	ring->first_seg = NULL;
126 	kfree(ring);
127 }
128 
129 static void xhci_initialize_ring_info(struct xhci_ring *ring)
130 {
131 	/* The ring is empty, so the enqueue pointer == dequeue pointer */
132 	ring->enqueue = ring->first_seg->trbs;
133 	ring->enq_seg = ring->first_seg;
134 	ring->dequeue = ring->enqueue;
135 	ring->deq_seg = ring->first_seg;
136 	/* The ring is initialized to 0. The producer must write 1 to the cycle
137 	 * bit to handover ownership of the TRB, so PCS = 1.  The consumer must
138 	 * compare CCS to the cycle bit to check ownership, so CCS = 1.
139 	 */
140 	ring->cycle_state = 1;
141 	/* Not necessary for new rings, but needed for re-initialized rings */
142 	ring->enq_updates = 0;
143 	ring->deq_updates = 0;
144 }
145 
146 /**
147  * Create a new ring with zero or more segments.
148  *
149  * Link each segment together into a ring.
150  * Set the end flag and the cycle toggle bit on the last segment.
151  * See section 4.9.1 and figures 15 and 16.
152  */
153 static struct xhci_ring *xhci_ring_alloc(struct xhci_hcd *xhci,
154 		unsigned int num_segs, bool link_trbs, gfp_t flags)
155 {
156 	struct xhci_ring	*ring;
157 	struct xhci_segment	*prev;
158 
159 	ring = kzalloc(sizeof *(ring), flags);
160 	xhci_dbg(xhci, "Allocating ring at %p\n", ring);
161 	if (!ring)
162 		return NULL;
163 
164 	INIT_LIST_HEAD(&ring->td_list);
165 	if (num_segs == 0)
166 		return ring;
167 
168 	ring->first_seg = xhci_segment_alloc(xhci, flags);
169 	if (!ring->first_seg)
170 		goto fail;
171 	num_segs--;
172 
173 	prev = ring->first_seg;
174 	while (num_segs > 0) {
175 		struct xhci_segment	*next;
176 
177 		next = xhci_segment_alloc(xhci, flags);
178 		if (!next)
179 			goto fail;
180 		xhci_link_segments(xhci, prev, next, link_trbs);
181 
182 		prev = next;
183 		num_segs--;
184 	}
185 	xhci_link_segments(xhci, prev, ring->first_seg, link_trbs);
186 
187 	if (link_trbs) {
188 		/* See section 4.9.2.1 and 6.4.4.1 */
189 		prev->trbs[TRBS_PER_SEGMENT-1].link.control |= (LINK_TOGGLE);
190 		xhci_dbg(xhci, "Wrote link toggle flag to"
191 				" segment %p (virtual), 0x%llx (DMA)\n",
192 				prev, (unsigned long long)prev->dma);
193 	}
194 	xhci_initialize_ring_info(ring);
195 	return ring;
196 
197 fail:
198 	xhci_ring_free(xhci, ring);
199 	return NULL;
200 }
201 
202 void xhci_free_or_cache_endpoint_ring(struct xhci_hcd *xhci,
203 		struct xhci_virt_device *virt_dev,
204 		unsigned int ep_index)
205 {
206 	int rings_cached;
207 
208 	rings_cached = virt_dev->num_rings_cached;
209 	if (rings_cached < XHCI_MAX_RINGS_CACHED) {
210 		virt_dev->num_rings_cached++;
211 		rings_cached = virt_dev->num_rings_cached;
212 		virt_dev->ring_cache[rings_cached] =
213 			virt_dev->eps[ep_index].ring;
214 		xhci_dbg(xhci, "Cached old ring, "
215 				"%d ring%s cached\n",
216 				rings_cached,
217 				(rings_cached > 1) ? "s" : "");
218 	} else {
219 		xhci_ring_free(xhci, virt_dev->eps[ep_index].ring);
220 		xhci_dbg(xhci, "Ring cache full (%d rings), "
221 				"freeing ring\n",
222 				virt_dev->num_rings_cached);
223 	}
224 	virt_dev->eps[ep_index].ring = NULL;
225 }
226 
227 /* Zero an endpoint ring (except for link TRBs) and move the enqueue and dequeue
228  * pointers to the beginning of the ring.
229  */
230 static void xhci_reinit_cached_ring(struct xhci_hcd *xhci,
231 		struct xhci_ring *ring)
232 {
233 	struct xhci_segment	*seg = ring->first_seg;
234 	do {
235 		memset(seg->trbs, 0,
236 				sizeof(union xhci_trb)*TRBS_PER_SEGMENT);
237 		/* All endpoint rings have link TRBs */
238 		xhci_link_segments(xhci, seg, seg->next, 1);
239 		seg = seg->next;
240 	} while (seg != ring->first_seg);
241 	xhci_initialize_ring_info(ring);
242 	/* td list should be empty since all URBs have been cancelled,
243 	 * but just in case...
244 	 */
245 	INIT_LIST_HEAD(&ring->td_list);
246 }
247 
248 #define CTX_SIZE(_hcc) (HCC_64BYTE_CONTEXT(_hcc) ? 64 : 32)
249 
250 static struct xhci_container_ctx *xhci_alloc_container_ctx(struct xhci_hcd *xhci,
251 						    int type, gfp_t flags)
252 {
253 	struct xhci_container_ctx *ctx = kzalloc(sizeof(*ctx), flags);
254 	if (!ctx)
255 		return NULL;
256 
257 	BUG_ON((type != XHCI_CTX_TYPE_DEVICE) && (type != XHCI_CTX_TYPE_INPUT));
258 	ctx->type = type;
259 	ctx->size = HCC_64BYTE_CONTEXT(xhci->hcc_params) ? 2048 : 1024;
260 	if (type == XHCI_CTX_TYPE_INPUT)
261 		ctx->size += CTX_SIZE(xhci->hcc_params);
262 
263 	ctx->bytes = dma_pool_alloc(xhci->device_pool, flags, &ctx->dma);
264 	memset(ctx->bytes, 0, ctx->size);
265 	return ctx;
266 }
267 
268 static void xhci_free_container_ctx(struct xhci_hcd *xhci,
269 			     struct xhci_container_ctx *ctx)
270 {
271 	if (!ctx)
272 		return;
273 	dma_pool_free(xhci->device_pool, ctx->bytes, ctx->dma);
274 	kfree(ctx);
275 }
276 
277 struct xhci_input_control_ctx *xhci_get_input_control_ctx(struct xhci_hcd *xhci,
278 					      struct xhci_container_ctx *ctx)
279 {
280 	BUG_ON(ctx->type != XHCI_CTX_TYPE_INPUT);
281 	return (struct xhci_input_control_ctx *)ctx->bytes;
282 }
283 
284 struct xhci_slot_ctx *xhci_get_slot_ctx(struct xhci_hcd *xhci,
285 					struct xhci_container_ctx *ctx)
286 {
287 	if (ctx->type == XHCI_CTX_TYPE_DEVICE)
288 		return (struct xhci_slot_ctx *)ctx->bytes;
289 
290 	return (struct xhci_slot_ctx *)
291 		(ctx->bytes + CTX_SIZE(xhci->hcc_params));
292 }
293 
294 struct xhci_ep_ctx *xhci_get_ep_ctx(struct xhci_hcd *xhci,
295 				    struct xhci_container_ctx *ctx,
296 				    unsigned int ep_index)
297 {
298 	/* increment ep index by offset of start of ep ctx array */
299 	ep_index++;
300 	if (ctx->type == XHCI_CTX_TYPE_INPUT)
301 		ep_index++;
302 
303 	return (struct xhci_ep_ctx *)
304 		(ctx->bytes + (ep_index * CTX_SIZE(xhci->hcc_params)));
305 }
306 
307 
308 /***************** Streams structures manipulation *************************/
309 
310 void xhci_free_stream_ctx(struct xhci_hcd *xhci,
311 		unsigned int num_stream_ctxs,
312 		struct xhci_stream_ctx *stream_ctx, dma_addr_t dma)
313 {
314 	struct pci_dev *pdev = to_pci_dev(xhci_to_hcd(xhci)->self.controller);
315 
316 	if (num_stream_ctxs > MEDIUM_STREAM_ARRAY_SIZE)
317 		pci_free_consistent(pdev,
318 				sizeof(struct xhci_stream_ctx)*num_stream_ctxs,
319 				stream_ctx, dma);
320 	else if (num_stream_ctxs <= SMALL_STREAM_ARRAY_SIZE)
321 		return dma_pool_free(xhci->small_streams_pool,
322 				stream_ctx, dma);
323 	else
324 		return dma_pool_free(xhci->medium_streams_pool,
325 				stream_ctx, dma);
326 }
327 
328 /*
329  * The stream context array for each endpoint with bulk streams enabled can
330  * vary in size, based on:
331  *  - how many streams the endpoint supports,
332  *  - the maximum primary stream array size the host controller supports,
333  *  - and how many streams the device driver asks for.
334  *
335  * The stream context array must be a power of 2, and can be as small as
336  * 64 bytes or as large as 1MB.
337  */
338 struct xhci_stream_ctx *xhci_alloc_stream_ctx(struct xhci_hcd *xhci,
339 		unsigned int num_stream_ctxs, dma_addr_t *dma,
340 		gfp_t mem_flags)
341 {
342 	struct pci_dev *pdev = to_pci_dev(xhci_to_hcd(xhci)->self.controller);
343 
344 	if (num_stream_ctxs > MEDIUM_STREAM_ARRAY_SIZE)
345 		return pci_alloc_consistent(pdev,
346 				sizeof(struct xhci_stream_ctx)*num_stream_ctxs,
347 				dma);
348 	else if (num_stream_ctxs <= SMALL_STREAM_ARRAY_SIZE)
349 		return dma_pool_alloc(xhci->small_streams_pool,
350 				mem_flags, dma);
351 	else
352 		return dma_pool_alloc(xhci->medium_streams_pool,
353 				mem_flags, dma);
354 }
355 
356 struct xhci_ring *xhci_dma_to_transfer_ring(
357 		struct xhci_virt_ep *ep,
358 		u64 address)
359 {
360 	if (ep->ep_state & EP_HAS_STREAMS)
361 		return radix_tree_lookup(&ep->stream_info->trb_address_map,
362 				address >> SEGMENT_SHIFT);
363 	return ep->ring;
364 }
365 
366 /* Only use this when you know stream_info is valid */
367 #ifdef CONFIG_USB_XHCI_HCD_DEBUGGING
368 static struct xhci_ring *dma_to_stream_ring(
369 		struct xhci_stream_info *stream_info,
370 		u64 address)
371 {
372 	return radix_tree_lookup(&stream_info->trb_address_map,
373 			address >> SEGMENT_SHIFT);
374 }
375 #endif	/* CONFIG_USB_XHCI_HCD_DEBUGGING */
376 
377 struct xhci_ring *xhci_stream_id_to_ring(
378 		struct xhci_virt_device *dev,
379 		unsigned int ep_index,
380 		unsigned int stream_id)
381 {
382 	struct xhci_virt_ep *ep = &dev->eps[ep_index];
383 
384 	if (stream_id == 0)
385 		return ep->ring;
386 	if (!ep->stream_info)
387 		return NULL;
388 
389 	if (stream_id > ep->stream_info->num_streams)
390 		return NULL;
391 	return ep->stream_info->stream_rings[stream_id];
392 }
393 
394 struct xhci_ring *xhci_triad_to_transfer_ring(struct xhci_hcd *xhci,
395 		unsigned int slot_id, unsigned int ep_index,
396 		unsigned int stream_id)
397 {
398 	struct xhci_virt_ep *ep;
399 
400 	ep = &xhci->devs[slot_id]->eps[ep_index];
401 	/* Common case: no streams */
402 	if (!(ep->ep_state & EP_HAS_STREAMS))
403 		return ep->ring;
404 
405 	if (stream_id == 0) {
406 		xhci_warn(xhci,
407 				"WARN: Slot ID %u, ep index %u has streams, "
408 				"but URB has no stream ID.\n",
409 				slot_id, ep_index);
410 		return NULL;
411 	}
412 
413 	if (stream_id < ep->stream_info->num_streams)
414 		return ep->stream_info->stream_rings[stream_id];
415 
416 	xhci_warn(xhci,
417 			"WARN: Slot ID %u, ep index %u has "
418 			"stream IDs 1 to %u allocated, "
419 			"but stream ID %u is requested.\n",
420 			slot_id, ep_index,
421 			ep->stream_info->num_streams - 1,
422 			stream_id);
423 	return NULL;
424 }
425 
426 /* Get the right ring for the given URB.
427  * If the endpoint supports streams, boundary check the URB's stream ID.
428  * If the endpoint doesn't support streams, return the singular endpoint ring.
429  */
430 struct xhci_ring *xhci_urb_to_transfer_ring(struct xhci_hcd *xhci,
431 		struct urb *urb)
432 {
433 	return xhci_triad_to_transfer_ring(xhci, urb->dev->slot_id,
434 		xhci_get_endpoint_index(&urb->ep->desc), urb->stream_id);
435 }
436 
437 #ifdef CONFIG_USB_XHCI_HCD_DEBUGGING
438 static int xhci_test_radix_tree(struct xhci_hcd *xhci,
439 		unsigned int num_streams,
440 		struct xhci_stream_info *stream_info)
441 {
442 	u32 cur_stream;
443 	struct xhci_ring *cur_ring;
444 	u64 addr;
445 
446 	for (cur_stream = 1; cur_stream < num_streams; cur_stream++) {
447 		struct xhci_ring *mapped_ring;
448 		int trb_size = sizeof(union xhci_trb);
449 
450 		cur_ring = stream_info->stream_rings[cur_stream];
451 		for (addr = cur_ring->first_seg->dma;
452 				addr < cur_ring->first_seg->dma + SEGMENT_SIZE;
453 				addr += trb_size) {
454 			mapped_ring = dma_to_stream_ring(stream_info, addr);
455 			if (cur_ring != mapped_ring) {
456 				xhci_warn(xhci, "WARN: DMA address 0x%08llx "
457 						"didn't map to stream ID %u; "
458 						"mapped to ring %p\n",
459 						(unsigned long long) addr,
460 						cur_stream,
461 						mapped_ring);
462 				return -EINVAL;
463 			}
464 		}
465 		/* One TRB after the end of the ring segment shouldn't return a
466 		 * pointer to the current ring (although it may be a part of a
467 		 * different ring).
468 		 */
469 		mapped_ring = dma_to_stream_ring(stream_info, addr);
470 		if (mapped_ring != cur_ring) {
471 			/* One TRB before should also fail */
472 			addr = cur_ring->first_seg->dma - trb_size;
473 			mapped_ring = dma_to_stream_ring(stream_info, addr);
474 		}
475 		if (mapped_ring == cur_ring) {
476 			xhci_warn(xhci, "WARN: Bad DMA address 0x%08llx "
477 					"mapped to valid stream ID %u; "
478 					"mapped ring = %p\n",
479 					(unsigned long long) addr,
480 					cur_stream,
481 					mapped_ring);
482 			return -EINVAL;
483 		}
484 	}
485 	return 0;
486 }
487 #endif	/* CONFIG_USB_XHCI_HCD_DEBUGGING */
488 
489 /*
490  * Change an endpoint's internal structure so it supports stream IDs.  The
491  * number of requested streams includes stream 0, which cannot be used by device
492  * drivers.
493  *
494  * The number of stream contexts in the stream context array may be bigger than
495  * the number of streams the driver wants to use.  This is because the number of
496  * stream context array entries must be a power of two.
497  *
498  * We need a radix tree for mapping physical addresses of TRBs to which stream
499  * ID they belong to.  We need to do this because the host controller won't tell
500  * us which stream ring the TRB came from.  We could store the stream ID in an
501  * event data TRB, but that doesn't help us for the cancellation case, since the
502  * endpoint may stop before it reaches that event data TRB.
503  *
504  * The radix tree maps the upper portion of the TRB DMA address to a ring
505  * segment that has the same upper portion of DMA addresses.  For example, say I
506  * have segments of size 1KB, that are always 64-byte aligned.  A segment may
507  * start at 0x10c91000 and end at 0x10c913f0.  If I use the upper 10 bits, the
508  * key to the stream ID is 0x43244.  I can use the DMA address of the TRB to
509  * pass the radix tree a key to get the right stream ID:
510  *
511  * 	0x10c90fff >> 10 = 0x43243
512  * 	0x10c912c0 >> 10 = 0x43244
513  * 	0x10c91400 >> 10 = 0x43245
514  *
515  * Obviously, only those TRBs with DMA addresses that are within the segment
516  * will make the radix tree return the stream ID for that ring.
517  *
518  * Caveats for the radix tree:
519  *
520  * The radix tree uses an unsigned long as a key pair.  On 32-bit systems, an
521  * unsigned long will be 32-bits; on a 64-bit system an unsigned long will be
522  * 64-bits.  Since we only request 32-bit DMA addresses, we can use that as the
523  * key on 32-bit or 64-bit systems (it would also be fine if we asked for 64-bit
524  * PCI DMA addresses on a 64-bit system).  There might be a problem on 32-bit
525  * extended systems (where the DMA address can be bigger than 32-bits),
526  * if we allow the PCI dma mask to be bigger than 32-bits.  So don't do that.
527  */
528 struct xhci_stream_info *xhci_alloc_stream_info(struct xhci_hcd *xhci,
529 		unsigned int num_stream_ctxs,
530 		unsigned int num_streams, gfp_t mem_flags)
531 {
532 	struct xhci_stream_info *stream_info;
533 	u32 cur_stream;
534 	struct xhci_ring *cur_ring;
535 	unsigned long key;
536 	u64 addr;
537 	int ret;
538 
539 	xhci_dbg(xhci, "Allocating %u streams and %u "
540 			"stream context array entries.\n",
541 			num_streams, num_stream_ctxs);
542 	if (xhci->cmd_ring_reserved_trbs == MAX_RSVD_CMD_TRBS) {
543 		xhci_dbg(xhci, "Command ring has no reserved TRBs available\n");
544 		return NULL;
545 	}
546 	xhci->cmd_ring_reserved_trbs++;
547 
548 	stream_info = kzalloc(sizeof(struct xhci_stream_info), mem_flags);
549 	if (!stream_info)
550 		goto cleanup_trbs;
551 
552 	stream_info->num_streams = num_streams;
553 	stream_info->num_stream_ctxs = num_stream_ctxs;
554 
555 	/* Initialize the array of virtual pointers to stream rings. */
556 	stream_info->stream_rings = kzalloc(
557 			sizeof(struct xhci_ring *)*num_streams,
558 			mem_flags);
559 	if (!stream_info->stream_rings)
560 		goto cleanup_info;
561 
562 	/* Initialize the array of DMA addresses for stream rings for the HW. */
563 	stream_info->stream_ctx_array = xhci_alloc_stream_ctx(xhci,
564 			num_stream_ctxs, &stream_info->ctx_array_dma,
565 			mem_flags);
566 	if (!stream_info->stream_ctx_array)
567 		goto cleanup_ctx;
568 	memset(stream_info->stream_ctx_array, 0,
569 			sizeof(struct xhci_stream_ctx)*num_stream_ctxs);
570 
571 	/* Allocate everything needed to free the stream rings later */
572 	stream_info->free_streams_command =
573 		xhci_alloc_command(xhci, true, true, mem_flags);
574 	if (!stream_info->free_streams_command)
575 		goto cleanup_ctx;
576 
577 	INIT_RADIX_TREE(&stream_info->trb_address_map, GFP_ATOMIC);
578 
579 	/* Allocate rings for all the streams that the driver will use,
580 	 * and add their segment DMA addresses to the radix tree.
581 	 * Stream 0 is reserved.
582 	 */
583 	for (cur_stream = 1; cur_stream < num_streams; cur_stream++) {
584 		stream_info->stream_rings[cur_stream] =
585 			xhci_ring_alloc(xhci, 1, true, mem_flags);
586 		cur_ring = stream_info->stream_rings[cur_stream];
587 		if (!cur_ring)
588 			goto cleanup_rings;
589 		cur_ring->stream_id = cur_stream;
590 		/* Set deq ptr, cycle bit, and stream context type */
591 		addr = cur_ring->first_seg->dma |
592 			SCT_FOR_CTX(SCT_PRI_TR) |
593 			cur_ring->cycle_state;
594 		stream_info->stream_ctx_array[cur_stream].stream_ring = addr;
595 		xhci_dbg(xhci, "Setting stream %d ring ptr to 0x%08llx\n",
596 				cur_stream, (unsigned long long) addr);
597 
598 		key = (unsigned long)
599 			(cur_ring->first_seg->dma >> SEGMENT_SHIFT);
600 		ret = radix_tree_insert(&stream_info->trb_address_map,
601 				key, cur_ring);
602 		if (ret) {
603 			xhci_ring_free(xhci, cur_ring);
604 			stream_info->stream_rings[cur_stream] = NULL;
605 			goto cleanup_rings;
606 		}
607 	}
608 	/* Leave the other unused stream ring pointers in the stream context
609 	 * array initialized to zero.  This will cause the xHC to give us an
610 	 * error if the device asks for a stream ID we don't have setup (if it
611 	 * was any other way, the host controller would assume the ring is
612 	 * "empty" and wait forever for data to be queued to that stream ID).
613 	 */
614 #if XHCI_DEBUG
615 	/* Do a little test on the radix tree to make sure it returns the
616 	 * correct values.
617 	 */
618 	if (xhci_test_radix_tree(xhci, num_streams, stream_info))
619 		goto cleanup_rings;
620 #endif
621 
622 	return stream_info;
623 
624 cleanup_rings:
625 	for (cur_stream = 1; cur_stream < num_streams; cur_stream++) {
626 		cur_ring = stream_info->stream_rings[cur_stream];
627 		if (cur_ring) {
628 			addr = cur_ring->first_seg->dma;
629 			radix_tree_delete(&stream_info->trb_address_map,
630 					addr >> SEGMENT_SHIFT);
631 			xhci_ring_free(xhci, cur_ring);
632 			stream_info->stream_rings[cur_stream] = NULL;
633 		}
634 	}
635 	xhci_free_command(xhci, stream_info->free_streams_command);
636 cleanup_ctx:
637 	kfree(stream_info->stream_rings);
638 cleanup_info:
639 	kfree(stream_info);
640 cleanup_trbs:
641 	xhci->cmd_ring_reserved_trbs--;
642 	return NULL;
643 }
644 /*
645  * Sets the MaxPStreams field and the Linear Stream Array field.
646  * Sets the dequeue pointer to the stream context array.
647  */
648 void xhci_setup_streams_ep_input_ctx(struct xhci_hcd *xhci,
649 		struct xhci_ep_ctx *ep_ctx,
650 		struct xhci_stream_info *stream_info)
651 {
652 	u32 max_primary_streams;
653 	/* MaxPStreams is the number of stream context array entries, not the
654 	 * number we're actually using.  Must be in 2^(MaxPstreams + 1) format.
655 	 * fls(0) = 0, fls(0x1) = 1, fls(0x10) = 2, fls(0x100) = 3, etc.
656 	 */
657 	max_primary_streams = fls(stream_info->num_stream_ctxs) - 2;
658 	xhci_dbg(xhci, "Setting number of stream ctx array entries to %u\n",
659 			1 << (max_primary_streams + 1));
660 	ep_ctx->ep_info &= ~EP_MAXPSTREAMS_MASK;
661 	ep_ctx->ep_info |= EP_MAXPSTREAMS(max_primary_streams);
662 	ep_ctx->ep_info |= EP_HAS_LSA;
663 	ep_ctx->deq  = stream_info->ctx_array_dma;
664 }
665 
666 /*
667  * Sets the MaxPStreams field and the Linear Stream Array field to 0.
668  * Reinstalls the "normal" endpoint ring (at its previous dequeue mark,
669  * not at the beginning of the ring).
670  */
671 void xhci_setup_no_streams_ep_input_ctx(struct xhci_hcd *xhci,
672 		struct xhci_ep_ctx *ep_ctx,
673 		struct xhci_virt_ep *ep)
674 {
675 	dma_addr_t addr;
676 	ep_ctx->ep_info &= ~EP_MAXPSTREAMS_MASK;
677 	ep_ctx->ep_info &= ~EP_HAS_LSA;
678 	addr = xhci_trb_virt_to_dma(ep->ring->deq_seg, ep->ring->dequeue);
679 	ep_ctx->deq  = addr | ep->ring->cycle_state;
680 }
681 
682 /* Frees all stream contexts associated with the endpoint,
683  *
684  * Caller should fix the endpoint context streams fields.
685  */
686 void xhci_free_stream_info(struct xhci_hcd *xhci,
687 		struct xhci_stream_info *stream_info)
688 {
689 	int cur_stream;
690 	struct xhci_ring *cur_ring;
691 	dma_addr_t addr;
692 
693 	if (!stream_info)
694 		return;
695 
696 	for (cur_stream = 1; cur_stream < stream_info->num_streams;
697 			cur_stream++) {
698 		cur_ring = stream_info->stream_rings[cur_stream];
699 		if (cur_ring) {
700 			addr = cur_ring->first_seg->dma;
701 			radix_tree_delete(&stream_info->trb_address_map,
702 					addr >> SEGMENT_SHIFT);
703 			xhci_ring_free(xhci, cur_ring);
704 			stream_info->stream_rings[cur_stream] = NULL;
705 		}
706 	}
707 	xhci_free_command(xhci, stream_info->free_streams_command);
708 	xhci->cmd_ring_reserved_trbs--;
709 	if (stream_info->stream_ctx_array)
710 		xhci_free_stream_ctx(xhci,
711 				stream_info->num_stream_ctxs,
712 				stream_info->stream_ctx_array,
713 				stream_info->ctx_array_dma);
714 
715 	if (stream_info)
716 		kfree(stream_info->stream_rings);
717 	kfree(stream_info);
718 }
719 
720 
721 /***************** Device context manipulation *************************/
722 
723 static void xhci_init_endpoint_timer(struct xhci_hcd *xhci,
724 		struct xhci_virt_ep *ep)
725 {
726 	init_timer(&ep->stop_cmd_timer);
727 	ep->stop_cmd_timer.data = (unsigned long) ep;
728 	ep->stop_cmd_timer.function = xhci_stop_endpoint_command_watchdog;
729 	ep->xhci = xhci;
730 }
731 
732 /* All the xhci_tds in the ring's TD list should be freed at this point */
733 void xhci_free_virt_device(struct xhci_hcd *xhci, int slot_id)
734 {
735 	struct xhci_virt_device *dev;
736 	int i;
737 
738 	/* Slot ID 0 is reserved */
739 	if (slot_id == 0 || !xhci->devs[slot_id])
740 		return;
741 
742 	dev = xhci->devs[slot_id];
743 	xhci->dcbaa->dev_context_ptrs[slot_id] = 0;
744 	if (!dev)
745 		return;
746 
747 	for (i = 0; i < 31; ++i) {
748 		if (dev->eps[i].ring)
749 			xhci_ring_free(xhci, dev->eps[i].ring);
750 		if (dev->eps[i].stream_info)
751 			xhci_free_stream_info(xhci,
752 					dev->eps[i].stream_info);
753 	}
754 
755 	if (dev->ring_cache) {
756 		for (i = 0; i < dev->num_rings_cached; i++)
757 			xhci_ring_free(xhci, dev->ring_cache[i]);
758 		kfree(dev->ring_cache);
759 	}
760 
761 	if (dev->in_ctx)
762 		xhci_free_container_ctx(xhci, dev->in_ctx);
763 	if (dev->out_ctx)
764 		xhci_free_container_ctx(xhci, dev->out_ctx);
765 
766 	kfree(xhci->devs[slot_id]);
767 	xhci->devs[slot_id] = NULL;
768 }
769 
770 int xhci_alloc_virt_device(struct xhci_hcd *xhci, int slot_id,
771 		struct usb_device *udev, gfp_t flags)
772 {
773 	struct xhci_virt_device *dev;
774 	int i;
775 
776 	/* Slot ID 0 is reserved */
777 	if (slot_id == 0 || xhci->devs[slot_id]) {
778 		xhci_warn(xhci, "Bad Slot ID %d\n", slot_id);
779 		return 0;
780 	}
781 
782 	xhci->devs[slot_id] = kzalloc(sizeof(*xhci->devs[slot_id]), flags);
783 	if (!xhci->devs[slot_id])
784 		return 0;
785 	dev = xhci->devs[slot_id];
786 
787 	/* Allocate the (output) device context that will be used in the HC. */
788 	dev->out_ctx = xhci_alloc_container_ctx(xhci, XHCI_CTX_TYPE_DEVICE, flags);
789 	if (!dev->out_ctx)
790 		goto fail;
791 
792 	xhci_dbg(xhci, "Slot %d output ctx = 0x%llx (dma)\n", slot_id,
793 			(unsigned long long)dev->out_ctx->dma);
794 
795 	/* Allocate the (input) device context for address device command */
796 	dev->in_ctx = xhci_alloc_container_ctx(xhci, XHCI_CTX_TYPE_INPUT, flags);
797 	if (!dev->in_ctx)
798 		goto fail;
799 
800 	xhci_dbg(xhci, "Slot %d input ctx = 0x%llx (dma)\n", slot_id,
801 			(unsigned long long)dev->in_ctx->dma);
802 
803 	/* Initialize the cancellation list and watchdog timers for each ep */
804 	for (i = 0; i < 31; i++) {
805 		xhci_init_endpoint_timer(xhci, &dev->eps[i]);
806 		INIT_LIST_HEAD(&dev->eps[i].cancelled_td_list);
807 	}
808 
809 	/* Allocate endpoint 0 ring */
810 	dev->eps[0].ring = xhci_ring_alloc(xhci, 1, true, flags);
811 	if (!dev->eps[0].ring)
812 		goto fail;
813 
814 	/* Allocate pointers to the ring cache */
815 	dev->ring_cache = kzalloc(
816 			sizeof(struct xhci_ring *)*XHCI_MAX_RINGS_CACHED,
817 			flags);
818 	if (!dev->ring_cache)
819 		goto fail;
820 	dev->num_rings_cached = 0;
821 
822 	init_completion(&dev->cmd_completion);
823 	INIT_LIST_HEAD(&dev->cmd_list);
824 
825 	/* Point to output device context in dcbaa. */
826 	xhci->dcbaa->dev_context_ptrs[slot_id] = dev->out_ctx->dma;
827 	xhci_dbg(xhci, "Set slot id %d dcbaa entry %p to 0x%llx\n",
828 			slot_id,
829 			&xhci->dcbaa->dev_context_ptrs[slot_id],
830 			(unsigned long long) xhci->dcbaa->dev_context_ptrs[slot_id]);
831 
832 	return 1;
833 fail:
834 	xhci_free_virt_device(xhci, slot_id);
835 	return 0;
836 }
837 
838 /* Setup an xHCI virtual device for a Set Address command */
839 int xhci_setup_addressable_virt_dev(struct xhci_hcd *xhci, struct usb_device *udev)
840 {
841 	struct xhci_virt_device *dev;
842 	struct xhci_ep_ctx	*ep0_ctx;
843 	struct usb_device	*top_dev;
844 	struct xhci_slot_ctx    *slot_ctx;
845 	struct xhci_input_control_ctx *ctrl_ctx;
846 
847 	dev = xhci->devs[udev->slot_id];
848 	/* Slot ID 0 is reserved */
849 	if (udev->slot_id == 0 || !dev) {
850 		xhci_warn(xhci, "Slot ID %d is not assigned to this device\n",
851 				udev->slot_id);
852 		return -EINVAL;
853 	}
854 	ep0_ctx = xhci_get_ep_ctx(xhci, dev->in_ctx, 0);
855 	ctrl_ctx = xhci_get_input_control_ctx(xhci, dev->in_ctx);
856 	slot_ctx = xhci_get_slot_ctx(xhci, dev->in_ctx);
857 
858 	/* 2) New slot context and endpoint 0 context are valid*/
859 	ctrl_ctx->add_flags = SLOT_FLAG | EP0_FLAG;
860 
861 	/* 3) Only the control endpoint is valid - one endpoint context */
862 	slot_ctx->dev_info |= LAST_CTX(1);
863 
864 	slot_ctx->dev_info |= (u32) udev->route;
865 	switch (udev->speed) {
866 	case USB_SPEED_SUPER:
867 		slot_ctx->dev_info |= (u32) SLOT_SPEED_SS;
868 		break;
869 	case USB_SPEED_HIGH:
870 		slot_ctx->dev_info |= (u32) SLOT_SPEED_HS;
871 		break;
872 	case USB_SPEED_FULL:
873 		slot_ctx->dev_info |= (u32) SLOT_SPEED_FS;
874 		break;
875 	case USB_SPEED_LOW:
876 		slot_ctx->dev_info |= (u32) SLOT_SPEED_LS;
877 		break;
878 	case USB_SPEED_WIRELESS:
879 		xhci_dbg(xhci, "FIXME xHCI doesn't support wireless speeds\n");
880 		return -EINVAL;
881 		break;
882 	default:
883 		/* Speed was set earlier, this shouldn't happen. */
884 		BUG();
885 	}
886 	/* Find the root hub port this device is under */
887 	for (top_dev = udev; top_dev->parent && top_dev->parent->parent;
888 			top_dev = top_dev->parent)
889 		/* Found device below root hub */;
890 	slot_ctx->dev_info2 |= (u32) ROOT_HUB_PORT(top_dev->portnum);
891 	xhci_dbg(xhci, "Set root hub portnum to %d\n", top_dev->portnum);
892 
893 	/* Is this a LS/FS device under a HS hub? */
894 	if ((udev->speed == USB_SPEED_LOW || udev->speed == USB_SPEED_FULL) &&
895 			udev->tt) {
896 		slot_ctx->tt_info = udev->tt->hub->slot_id;
897 		slot_ctx->tt_info |= udev->ttport << 8;
898 		if (udev->tt->multi)
899 			slot_ctx->dev_info |= DEV_MTT;
900 	}
901 	xhci_dbg(xhci, "udev->tt = %p\n", udev->tt);
902 	xhci_dbg(xhci, "udev->ttport = 0x%x\n", udev->ttport);
903 
904 	/* Step 4 - ring already allocated */
905 	/* Step 5 */
906 	ep0_ctx->ep_info2 = EP_TYPE(CTRL_EP);
907 	/*
908 	 * XXX: Not sure about wireless USB devices.
909 	 */
910 	switch (udev->speed) {
911 	case USB_SPEED_SUPER:
912 		ep0_ctx->ep_info2 |= MAX_PACKET(512);
913 		break;
914 	case USB_SPEED_HIGH:
915 	/* USB core guesses at a 64-byte max packet first for FS devices */
916 	case USB_SPEED_FULL:
917 		ep0_ctx->ep_info2 |= MAX_PACKET(64);
918 		break;
919 	case USB_SPEED_LOW:
920 		ep0_ctx->ep_info2 |= MAX_PACKET(8);
921 		break;
922 	case USB_SPEED_WIRELESS:
923 		xhci_dbg(xhci, "FIXME xHCI doesn't support wireless speeds\n");
924 		return -EINVAL;
925 		break;
926 	default:
927 		/* New speed? */
928 		BUG();
929 	}
930 	/* EP 0 can handle "burst" sizes of 1, so Max Burst Size field is 0 */
931 	ep0_ctx->ep_info2 |= MAX_BURST(0);
932 	ep0_ctx->ep_info2 |= ERROR_COUNT(3);
933 
934 	ep0_ctx->deq =
935 		dev->eps[0].ring->first_seg->dma;
936 	ep0_ctx->deq |= dev->eps[0].ring->cycle_state;
937 
938 	/* Steps 7 and 8 were done in xhci_alloc_virt_device() */
939 
940 	return 0;
941 }
942 
943 /* Return the polling or NAK interval.
944  *
945  * The polling interval is expressed in "microframes".  If xHCI's Interval field
946  * is set to N, it will service the endpoint every 2^(Interval)*125us.
947  *
948  * The NAK interval is one NAK per 1 to 255 microframes, or no NAKs if interval
949  * is set to 0.
950  */
951 static inline unsigned int xhci_get_endpoint_interval(struct usb_device *udev,
952 		struct usb_host_endpoint *ep)
953 {
954 	unsigned int interval = 0;
955 
956 	switch (udev->speed) {
957 	case USB_SPEED_HIGH:
958 		/* Max NAK rate */
959 		if (usb_endpoint_xfer_control(&ep->desc) ||
960 				usb_endpoint_xfer_bulk(&ep->desc))
961 			interval = ep->desc.bInterval;
962 		/* Fall through - SS and HS isoc/int have same decoding */
963 	case USB_SPEED_SUPER:
964 		if (usb_endpoint_xfer_int(&ep->desc) ||
965 				usb_endpoint_xfer_isoc(&ep->desc)) {
966 			if (ep->desc.bInterval == 0)
967 				interval = 0;
968 			else
969 				interval = ep->desc.bInterval - 1;
970 			if (interval > 15)
971 				interval = 15;
972 			if (interval != ep->desc.bInterval + 1)
973 				dev_warn(&udev->dev, "ep %#x - rounding interval to %d microframes\n",
974 						ep->desc.bEndpointAddress, 1 << interval);
975 		}
976 		break;
977 	/* Convert bInterval (in 1-255 frames) to microframes and round down to
978 	 * nearest power of 2.
979 	 */
980 	case USB_SPEED_FULL:
981 	case USB_SPEED_LOW:
982 		if (usb_endpoint_xfer_int(&ep->desc) ||
983 				usb_endpoint_xfer_isoc(&ep->desc)) {
984 			interval = fls(8*ep->desc.bInterval) - 1;
985 			if (interval > 10)
986 				interval = 10;
987 			if (interval < 3)
988 				interval = 3;
989 			if ((1 << interval) != 8*ep->desc.bInterval)
990 				dev_warn(&udev->dev,
991 						"ep %#x - rounding interval"
992 						" to %d microframes, "
993 						"ep desc says %d microframes\n",
994 						ep->desc.bEndpointAddress,
995 						1 << interval,
996 						8*ep->desc.bInterval);
997 		}
998 		break;
999 	default:
1000 		BUG();
1001 	}
1002 	return EP_INTERVAL(interval);
1003 }
1004 
1005 /* The "Mult" field in the endpoint context is only set for SuperSpeed devices.
1006  * High speed endpoint descriptors can define "the number of additional
1007  * transaction opportunities per microframe", but that goes in the Max Burst
1008  * endpoint context field.
1009  */
1010 static inline u32 xhci_get_endpoint_mult(struct usb_device *udev,
1011 		struct usb_host_endpoint *ep)
1012 {
1013 	if (udev->speed != USB_SPEED_SUPER)
1014 		return 0;
1015 	return ep->ss_ep_comp.bmAttributes;
1016 }
1017 
1018 static inline u32 xhci_get_endpoint_type(struct usb_device *udev,
1019 		struct usb_host_endpoint *ep)
1020 {
1021 	int in;
1022 	u32 type;
1023 
1024 	in = usb_endpoint_dir_in(&ep->desc);
1025 	if (usb_endpoint_xfer_control(&ep->desc)) {
1026 		type = EP_TYPE(CTRL_EP);
1027 	} else if (usb_endpoint_xfer_bulk(&ep->desc)) {
1028 		if (in)
1029 			type = EP_TYPE(BULK_IN_EP);
1030 		else
1031 			type = EP_TYPE(BULK_OUT_EP);
1032 	} else if (usb_endpoint_xfer_isoc(&ep->desc)) {
1033 		if (in)
1034 			type = EP_TYPE(ISOC_IN_EP);
1035 		else
1036 			type = EP_TYPE(ISOC_OUT_EP);
1037 	} else if (usb_endpoint_xfer_int(&ep->desc)) {
1038 		if (in)
1039 			type = EP_TYPE(INT_IN_EP);
1040 		else
1041 			type = EP_TYPE(INT_OUT_EP);
1042 	} else {
1043 		BUG();
1044 	}
1045 	return type;
1046 }
1047 
1048 /* Return the maximum endpoint service interval time (ESIT) payload.
1049  * Basically, this is the maxpacket size, multiplied by the burst size
1050  * and mult size.
1051  */
1052 static inline u32 xhci_get_max_esit_payload(struct xhci_hcd *xhci,
1053 		struct usb_device *udev,
1054 		struct usb_host_endpoint *ep)
1055 {
1056 	int max_burst;
1057 	int max_packet;
1058 
1059 	/* Only applies for interrupt or isochronous endpoints */
1060 	if (usb_endpoint_xfer_control(&ep->desc) ||
1061 			usb_endpoint_xfer_bulk(&ep->desc))
1062 		return 0;
1063 
1064 	if (udev->speed == USB_SPEED_SUPER)
1065 		return ep->ss_ep_comp.wBytesPerInterval;
1066 
1067 	max_packet = ep->desc.wMaxPacketSize & 0x3ff;
1068 	max_burst = (ep->desc.wMaxPacketSize & 0x1800) >> 11;
1069 	/* A 0 in max burst means 1 transfer per ESIT */
1070 	return max_packet * (max_burst + 1);
1071 }
1072 
1073 /* Set up an endpoint with one ring segment.  Do not allocate stream rings.
1074  * Drivers will have to call usb_alloc_streams() to do that.
1075  */
1076 int xhci_endpoint_init(struct xhci_hcd *xhci,
1077 		struct xhci_virt_device *virt_dev,
1078 		struct usb_device *udev,
1079 		struct usb_host_endpoint *ep,
1080 		gfp_t mem_flags)
1081 {
1082 	unsigned int ep_index;
1083 	struct xhci_ep_ctx *ep_ctx;
1084 	struct xhci_ring *ep_ring;
1085 	unsigned int max_packet;
1086 	unsigned int max_burst;
1087 	u32 max_esit_payload;
1088 
1089 	ep_index = xhci_get_endpoint_index(&ep->desc);
1090 	ep_ctx = xhci_get_ep_ctx(xhci, virt_dev->in_ctx, ep_index);
1091 
1092 	/* Set up the endpoint ring */
1093 	virt_dev->eps[ep_index].new_ring =
1094 		xhci_ring_alloc(xhci, 1, true, mem_flags);
1095 	if (!virt_dev->eps[ep_index].new_ring) {
1096 		/* Attempt to use the ring cache */
1097 		if (virt_dev->num_rings_cached == 0)
1098 			return -ENOMEM;
1099 		virt_dev->eps[ep_index].new_ring =
1100 			virt_dev->ring_cache[virt_dev->num_rings_cached];
1101 		virt_dev->ring_cache[virt_dev->num_rings_cached] = NULL;
1102 		virt_dev->num_rings_cached--;
1103 		xhci_reinit_cached_ring(xhci, virt_dev->eps[ep_index].new_ring);
1104 	}
1105 	ep_ring = virt_dev->eps[ep_index].new_ring;
1106 	ep_ctx->deq = ep_ring->first_seg->dma | ep_ring->cycle_state;
1107 
1108 	ep_ctx->ep_info = xhci_get_endpoint_interval(udev, ep);
1109 	ep_ctx->ep_info |= EP_MULT(xhci_get_endpoint_mult(udev, ep));
1110 
1111 	/* FIXME dig Mult and streams info out of ep companion desc */
1112 
1113 	/* Allow 3 retries for everything but isoc;
1114 	 * error count = 0 means infinite retries.
1115 	 */
1116 	if (!usb_endpoint_xfer_isoc(&ep->desc))
1117 		ep_ctx->ep_info2 = ERROR_COUNT(3);
1118 	else
1119 		ep_ctx->ep_info2 = ERROR_COUNT(1);
1120 
1121 	ep_ctx->ep_info2 |= xhci_get_endpoint_type(udev, ep);
1122 
1123 	/* Set the max packet size and max burst */
1124 	switch (udev->speed) {
1125 	case USB_SPEED_SUPER:
1126 		max_packet = ep->desc.wMaxPacketSize;
1127 		ep_ctx->ep_info2 |= MAX_PACKET(max_packet);
1128 		/* dig out max burst from ep companion desc */
1129 		max_packet = ep->ss_ep_comp.bMaxBurst;
1130 		if (!max_packet)
1131 			xhci_warn(xhci, "WARN no SS endpoint bMaxBurst\n");
1132 		ep_ctx->ep_info2 |= MAX_BURST(max_packet);
1133 		break;
1134 	case USB_SPEED_HIGH:
1135 		/* bits 11:12 specify the number of additional transaction
1136 		 * opportunities per microframe (USB 2.0, section 9.6.6)
1137 		 */
1138 		if (usb_endpoint_xfer_isoc(&ep->desc) ||
1139 				usb_endpoint_xfer_int(&ep->desc)) {
1140 			max_burst = (ep->desc.wMaxPacketSize & 0x1800) >> 11;
1141 			ep_ctx->ep_info2 |= MAX_BURST(max_burst);
1142 		}
1143 		/* Fall through */
1144 	case USB_SPEED_FULL:
1145 	case USB_SPEED_LOW:
1146 		max_packet = ep->desc.wMaxPacketSize & 0x3ff;
1147 		ep_ctx->ep_info2 |= MAX_PACKET(max_packet);
1148 		break;
1149 	default:
1150 		BUG();
1151 	}
1152 	max_esit_payload = xhci_get_max_esit_payload(xhci, udev, ep);
1153 	ep_ctx->tx_info = MAX_ESIT_PAYLOAD_FOR_EP(max_esit_payload);
1154 
1155 	/*
1156 	 * XXX no idea how to calculate the average TRB buffer length for bulk
1157 	 * endpoints, as the driver gives us no clue how big each scatter gather
1158 	 * list entry (or buffer) is going to be.
1159 	 *
1160 	 * For isochronous and interrupt endpoints, we set it to the max
1161 	 * available, until we have new API in the USB core to allow drivers to
1162 	 * declare how much bandwidth they actually need.
1163 	 *
1164 	 * Normally, it would be calculated by taking the total of the buffer
1165 	 * lengths in the TD and then dividing by the number of TRBs in a TD,
1166 	 * including link TRBs, No-op TRBs, and Event data TRBs.  Since we don't
1167 	 * use Event Data TRBs, and we don't chain in a link TRB on short
1168 	 * transfers, we're basically dividing by 1.
1169 	 */
1170 	ep_ctx->tx_info |= AVG_TRB_LENGTH_FOR_EP(max_esit_payload);
1171 
1172 	/* FIXME Debug endpoint context */
1173 	return 0;
1174 }
1175 
1176 void xhci_endpoint_zero(struct xhci_hcd *xhci,
1177 		struct xhci_virt_device *virt_dev,
1178 		struct usb_host_endpoint *ep)
1179 {
1180 	unsigned int ep_index;
1181 	struct xhci_ep_ctx *ep_ctx;
1182 
1183 	ep_index = xhci_get_endpoint_index(&ep->desc);
1184 	ep_ctx = xhci_get_ep_ctx(xhci, virt_dev->in_ctx, ep_index);
1185 
1186 	ep_ctx->ep_info = 0;
1187 	ep_ctx->ep_info2 = 0;
1188 	ep_ctx->deq = 0;
1189 	ep_ctx->tx_info = 0;
1190 	/* Don't free the endpoint ring until the set interface or configuration
1191 	 * request succeeds.
1192 	 */
1193 }
1194 
1195 /* Copy output xhci_ep_ctx to the input xhci_ep_ctx copy.
1196  * Useful when you want to change one particular aspect of the endpoint and then
1197  * issue a configure endpoint command.
1198  */
1199 void xhci_endpoint_copy(struct xhci_hcd *xhci,
1200 		struct xhci_container_ctx *in_ctx,
1201 		struct xhci_container_ctx *out_ctx,
1202 		unsigned int ep_index)
1203 {
1204 	struct xhci_ep_ctx *out_ep_ctx;
1205 	struct xhci_ep_ctx *in_ep_ctx;
1206 
1207 	out_ep_ctx = xhci_get_ep_ctx(xhci, out_ctx, ep_index);
1208 	in_ep_ctx = xhci_get_ep_ctx(xhci, in_ctx, ep_index);
1209 
1210 	in_ep_ctx->ep_info = out_ep_ctx->ep_info;
1211 	in_ep_ctx->ep_info2 = out_ep_ctx->ep_info2;
1212 	in_ep_ctx->deq = out_ep_ctx->deq;
1213 	in_ep_ctx->tx_info = out_ep_ctx->tx_info;
1214 }
1215 
1216 /* Copy output xhci_slot_ctx to the input xhci_slot_ctx.
1217  * Useful when you want to change one particular aspect of the endpoint and then
1218  * issue a configure endpoint command.  Only the context entries field matters,
1219  * but we'll copy the whole thing anyway.
1220  */
1221 void xhci_slot_copy(struct xhci_hcd *xhci,
1222 		struct xhci_container_ctx *in_ctx,
1223 		struct xhci_container_ctx *out_ctx)
1224 {
1225 	struct xhci_slot_ctx *in_slot_ctx;
1226 	struct xhci_slot_ctx *out_slot_ctx;
1227 
1228 	in_slot_ctx = xhci_get_slot_ctx(xhci, in_ctx);
1229 	out_slot_ctx = xhci_get_slot_ctx(xhci, out_ctx);
1230 
1231 	in_slot_ctx->dev_info = out_slot_ctx->dev_info;
1232 	in_slot_ctx->dev_info2 = out_slot_ctx->dev_info2;
1233 	in_slot_ctx->tt_info = out_slot_ctx->tt_info;
1234 	in_slot_ctx->dev_state = out_slot_ctx->dev_state;
1235 }
1236 
1237 /* Set up the scratchpad buffer array and scratchpad buffers, if needed. */
1238 static int scratchpad_alloc(struct xhci_hcd *xhci, gfp_t flags)
1239 {
1240 	int i;
1241 	struct device *dev = xhci_to_hcd(xhci)->self.controller;
1242 	int num_sp = HCS_MAX_SCRATCHPAD(xhci->hcs_params2);
1243 
1244 	xhci_dbg(xhci, "Allocating %d scratchpad buffers\n", num_sp);
1245 
1246 	if (!num_sp)
1247 		return 0;
1248 
1249 	xhci->scratchpad = kzalloc(sizeof(*xhci->scratchpad), flags);
1250 	if (!xhci->scratchpad)
1251 		goto fail_sp;
1252 
1253 	xhci->scratchpad->sp_array =
1254 		pci_alloc_consistent(to_pci_dev(dev),
1255 				     num_sp * sizeof(u64),
1256 				     &xhci->scratchpad->sp_dma);
1257 	if (!xhci->scratchpad->sp_array)
1258 		goto fail_sp2;
1259 
1260 	xhci->scratchpad->sp_buffers = kzalloc(sizeof(void *) * num_sp, flags);
1261 	if (!xhci->scratchpad->sp_buffers)
1262 		goto fail_sp3;
1263 
1264 	xhci->scratchpad->sp_dma_buffers =
1265 		kzalloc(sizeof(dma_addr_t) * num_sp, flags);
1266 
1267 	if (!xhci->scratchpad->sp_dma_buffers)
1268 		goto fail_sp4;
1269 
1270 	xhci->dcbaa->dev_context_ptrs[0] = xhci->scratchpad->sp_dma;
1271 	for (i = 0; i < num_sp; i++) {
1272 		dma_addr_t dma;
1273 		void *buf = pci_alloc_consistent(to_pci_dev(dev),
1274 						 xhci->page_size, &dma);
1275 		if (!buf)
1276 			goto fail_sp5;
1277 
1278 		xhci->scratchpad->sp_array[i] = dma;
1279 		xhci->scratchpad->sp_buffers[i] = buf;
1280 		xhci->scratchpad->sp_dma_buffers[i] = dma;
1281 	}
1282 
1283 	return 0;
1284 
1285  fail_sp5:
1286 	for (i = i - 1; i >= 0; i--) {
1287 		pci_free_consistent(to_pci_dev(dev), xhci->page_size,
1288 				    xhci->scratchpad->sp_buffers[i],
1289 				    xhci->scratchpad->sp_dma_buffers[i]);
1290 	}
1291 	kfree(xhci->scratchpad->sp_dma_buffers);
1292 
1293  fail_sp4:
1294 	kfree(xhci->scratchpad->sp_buffers);
1295 
1296  fail_sp3:
1297 	pci_free_consistent(to_pci_dev(dev), num_sp * sizeof(u64),
1298 			    xhci->scratchpad->sp_array,
1299 			    xhci->scratchpad->sp_dma);
1300 
1301  fail_sp2:
1302 	kfree(xhci->scratchpad);
1303 	xhci->scratchpad = NULL;
1304 
1305  fail_sp:
1306 	return -ENOMEM;
1307 }
1308 
1309 static void scratchpad_free(struct xhci_hcd *xhci)
1310 {
1311 	int num_sp;
1312 	int i;
1313 	struct pci_dev	*pdev = to_pci_dev(xhci_to_hcd(xhci)->self.controller);
1314 
1315 	if (!xhci->scratchpad)
1316 		return;
1317 
1318 	num_sp = HCS_MAX_SCRATCHPAD(xhci->hcs_params2);
1319 
1320 	for (i = 0; i < num_sp; i++) {
1321 		pci_free_consistent(pdev, xhci->page_size,
1322 				    xhci->scratchpad->sp_buffers[i],
1323 				    xhci->scratchpad->sp_dma_buffers[i]);
1324 	}
1325 	kfree(xhci->scratchpad->sp_dma_buffers);
1326 	kfree(xhci->scratchpad->sp_buffers);
1327 	pci_free_consistent(pdev, num_sp * sizeof(u64),
1328 			    xhci->scratchpad->sp_array,
1329 			    xhci->scratchpad->sp_dma);
1330 	kfree(xhci->scratchpad);
1331 	xhci->scratchpad = NULL;
1332 }
1333 
1334 struct xhci_command *xhci_alloc_command(struct xhci_hcd *xhci,
1335 		bool allocate_in_ctx, bool allocate_completion,
1336 		gfp_t mem_flags)
1337 {
1338 	struct xhci_command *command;
1339 
1340 	command = kzalloc(sizeof(*command), mem_flags);
1341 	if (!command)
1342 		return NULL;
1343 
1344 	if (allocate_in_ctx) {
1345 		command->in_ctx =
1346 			xhci_alloc_container_ctx(xhci, XHCI_CTX_TYPE_INPUT,
1347 					mem_flags);
1348 		if (!command->in_ctx) {
1349 			kfree(command);
1350 			return NULL;
1351 		}
1352 	}
1353 
1354 	if (allocate_completion) {
1355 		command->completion =
1356 			kzalloc(sizeof(struct completion), mem_flags);
1357 		if (!command->completion) {
1358 			xhci_free_container_ctx(xhci, command->in_ctx);
1359 			kfree(command);
1360 			return NULL;
1361 		}
1362 		init_completion(command->completion);
1363 	}
1364 
1365 	command->status = 0;
1366 	INIT_LIST_HEAD(&command->cmd_list);
1367 	return command;
1368 }
1369 
1370 void xhci_free_command(struct xhci_hcd *xhci,
1371 		struct xhci_command *command)
1372 {
1373 	xhci_free_container_ctx(xhci,
1374 			command->in_ctx);
1375 	kfree(command->completion);
1376 	kfree(command);
1377 }
1378 
1379 void xhci_mem_cleanup(struct xhci_hcd *xhci)
1380 {
1381 	struct pci_dev	*pdev = to_pci_dev(xhci_to_hcd(xhci)->self.controller);
1382 	int size;
1383 	int i;
1384 
1385 	/* Free the Event Ring Segment Table and the actual Event Ring */
1386 	if (xhci->ir_set) {
1387 		xhci_writel(xhci, 0, &xhci->ir_set->erst_size);
1388 		xhci_write_64(xhci, 0, &xhci->ir_set->erst_base);
1389 		xhci_write_64(xhci, 0, &xhci->ir_set->erst_dequeue);
1390 	}
1391 	size = sizeof(struct xhci_erst_entry)*(xhci->erst.num_entries);
1392 	if (xhci->erst.entries)
1393 		pci_free_consistent(pdev, size,
1394 				xhci->erst.entries, xhci->erst.erst_dma_addr);
1395 	xhci->erst.entries = NULL;
1396 	xhci_dbg(xhci, "Freed ERST\n");
1397 	if (xhci->event_ring)
1398 		xhci_ring_free(xhci, xhci->event_ring);
1399 	xhci->event_ring = NULL;
1400 	xhci_dbg(xhci, "Freed event ring\n");
1401 
1402 	xhci_write_64(xhci, 0, &xhci->op_regs->cmd_ring);
1403 	if (xhci->cmd_ring)
1404 		xhci_ring_free(xhci, xhci->cmd_ring);
1405 	xhci->cmd_ring = NULL;
1406 	xhci_dbg(xhci, "Freed command ring\n");
1407 
1408 	for (i = 1; i < MAX_HC_SLOTS; ++i)
1409 		xhci_free_virt_device(xhci, i);
1410 
1411 	if (xhci->segment_pool)
1412 		dma_pool_destroy(xhci->segment_pool);
1413 	xhci->segment_pool = NULL;
1414 	xhci_dbg(xhci, "Freed segment pool\n");
1415 
1416 	if (xhci->device_pool)
1417 		dma_pool_destroy(xhci->device_pool);
1418 	xhci->device_pool = NULL;
1419 	xhci_dbg(xhci, "Freed device context pool\n");
1420 
1421 	if (xhci->small_streams_pool)
1422 		dma_pool_destroy(xhci->small_streams_pool);
1423 	xhci->small_streams_pool = NULL;
1424 	xhci_dbg(xhci, "Freed small stream array pool\n");
1425 
1426 	if (xhci->medium_streams_pool)
1427 		dma_pool_destroy(xhci->medium_streams_pool);
1428 	xhci->medium_streams_pool = NULL;
1429 	xhci_dbg(xhci, "Freed medium stream array pool\n");
1430 
1431 	xhci_write_64(xhci, 0, &xhci->op_regs->dcbaa_ptr);
1432 	if (xhci->dcbaa)
1433 		pci_free_consistent(pdev, sizeof(*xhci->dcbaa),
1434 				xhci->dcbaa, xhci->dcbaa->dma);
1435 	xhci->dcbaa = NULL;
1436 
1437 	scratchpad_free(xhci);
1438 	xhci->page_size = 0;
1439 	xhci->page_shift = 0;
1440 }
1441 
1442 static int xhci_test_trb_in_td(struct xhci_hcd *xhci,
1443 		struct xhci_segment *input_seg,
1444 		union xhci_trb *start_trb,
1445 		union xhci_trb *end_trb,
1446 		dma_addr_t input_dma,
1447 		struct xhci_segment *result_seg,
1448 		char *test_name, int test_number)
1449 {
1450 	unsigned long long start_dma;
1451 	unsigned long long end_dma;
1452 	struct xhci_segment *seg;
1453 
1454 	start_dma = xhci_trb_virt_to_dma(input_seg, start_trb);
1455 	end_dma = xhci_trb_virt_to_dma(input_seg, end_trb);
1456 
1457 	seg = trb_in_td(input_seg, start_trb, end_trb, input_dma);
1458 	if (seg != result_seg) {
1459 		xhci_warn(xhci, "WARN: %s TRB math test %d failed!\n",
1460 				test_name, test_number);
1461 		xhci_warn(xhci, "Tested TRB math w/ seg %p and "
1462 				"input DMA 0x%llx\n",
1463 				input_seg,
1464 				(unsigned long long) input_dma);
1465 		xhci_warn(xhci, "starting TRB %p (0x%llx DMA), "
1466 				"ending TRB %p (0x%llx DMA)\n",
1467 				start_trb, start_dma,
1468 				end_trb, end_dma);
1469 		xhci_warn(xhci, "Expected seg %p, got seg %p\n",
1470 				result_seg, seg);
1471 		return -1;
1472 	}
1473 	return 0;
1474 }
1475 
1476 /* TRB math checks for xhci_trb_in_td(), using the command and event rings. */
1477 static int xhci_check_trb_in_td_math(struct xhci_hcd *xhci, gfp_t mem_flags)
1478 {
1479 	struct {
1480 		dma_addr_t		input_dma;
1481 		struct xhci_segment	*result_seg;
1482 	} simple_test_vector [] = {
1483 		/* A zeroed DMA field should fail */
1484 		{ 0, NULL },
1485 		/* One TRB before the ring start should fail */
1486 		{ xhci->event_ring->first_seg->dma - 16, NULL },
1487 		/* One byte before the ring start should fail */
1488 		{ xhci->event_ring->first_seg->dma - 1, NULL },
1489 		/* Starting TRB should succeed */
1490 		{ xhci->event_ring->first_seg->dma, xhci->event_ring->first_seg },
1491 		/* Ending TRB should succeed */
1492 		{ xhci->event_ring->first_seg->dma + (TRBS_PER_SEGMENT - 1)*16,
1493 			xhci->event_ring->first_seg },
1494 		/* One byte after the ring end should fail */
1495 		{ xhci->event_ring->first_seg->dma + (TRBS_PER_SEGMENT - 1)*16 + 1, NULL },
1496 		/* One TRB after the ring end should fail */
1497 		{ xhci->event_ring->first_seg->dma + (TRBS_PER_SEGMENT)*16, NULL },
1498 		/* An address of all ones should fail */
1499 		{ (dma_addr_t) (~0), NULL },
1500 	};
1501 	struct {
1502 		struct xhci_segment	*input_seg;
1503 		union xhci_trb		*start_trb;
1504 		union xhci_trb		*end_trb;
1505 		dma_addr_t		input_dma;
1506 		struct xhci_segment	*result_seg;
1507 	} complex_test_vector [] = {
1508 		/* Test feeding a valid DMA address from a different ring */
1509 		{	.input_seg = xhci->event_ring->first_seg,
1510 			.start_trb = xhci->event_ring->first_seg->trbs,
1511 			.end_trb = &xhci->event_ring->first_seg->trbs[TRBS_PER_SEGMENT - 1],
1512 			.input_dma = xhci->cmd_ring->first_seg->dma,
1513 			.result_seg = NULL,
1514 		},
1515 		/* Test feeding a valid end TRB from a different ring */
1516 		{	.input_seg = xhci->event_ring->first_seg,
1517 			.start_trb = xhci->event_ring->first_seg->trbs,
1518 			.end_trb = &xhci->cmd_ring->first_seg->trbs[TRBS_PER_SEGMENT - 1],
1519 			.input_dma = xhci->cmd_ring->first_seg->dma,
1520 			.result_seg = NULL,
1521 		},
1522 		/* Test feeding a valid start and end TRB from a different ring */
1523 		{	.input_seg = xhci->event_ring->first_seg,
1524 			.start_trb = xhci->cmd_ring->first_seg->trbs,
1525 			.end_trb = &xhci->cmd_ring->first_seg->trbs[TRBS_PER_SEGMENT - 1],
1526 			.input_dma = xhci->cmd_ring->first_seg->dma,
1527 			.result_seg = NULL,
1528 		},
1529 		/* TRB in this ring, but after this TD */
1530 		{	.input_seg = xhci->event_ring->first_seg,
1531 			.start_trb = &xhci->event_ring->first_seg->trbs[0],
1532 			.end_trb = &xhci->event_ring->first_seg->trbs[3],
1533 			.input_dma = xhci->event_ring->first_seg->dma + 4*16,
1534 			.result_seg = NULL,
1535 		},
1536 		/* TRB in this ring, but before this TD */
1537 		{	.input_seg = xhci->event_ring->first_seg,
1538 			.start_trb = &xhci->event_ring->first_seg->trbs[3],
1539 			.end_trb = &xhci->event_ring->first_seg->trbs[6],
1540 			.input_dma = xhci->event_ring->first_seg->dma + 2*16,
1541 			.result_seg = NULL,
1542 		},
1543 		/* TRB in this ring, but after this wrapped TD */
1544 		{	.input_seg = xhci->event_ring->first_seg,
1545 			.start_trb = &xhci->event_ring->first_seg->trbs[TRBS_PER_SEGMENT - 3],
1546 			.end_trb = &xhci->event_ring->first_seg->trbs[1],
1547 			.input_dma = xhci->event_ring->first_seg->dma + 2*16,
1548 			.result_seg = NULL,
1549 		},
1550 		/* TRB in this ring, but before this wrapped TD */
1551 		{	.input_seg = xhci->event_ring->first_seg,
1552 			.start_trb = &xhci->event_ring->first_seg->trbs[TRBS_PER_SEGMENT - 3],
1553 			.end_trb = &xhci->event_ring->first_seg->trbs[1],
1554 			.input_dma = xhci->event_ring->first_seg->dma + (TRBS_PER_SEGMENT - 4)*16,
1555 			.result_seg = NULL,
1556 		},
1557 		/* TRB not in this ring, and we have a wrapped TD */
1558 		{	.input_seg = xhci->event_ring->first_seg,
1559 			.start_trb = &xhci->event_ring->first_seg->trbs[TRBS_PER_SEGMENT - 3],
1560 			.end_trb = &xhci->event_ring->first_seg->trbs[1],
1561 			.input_dma = xhci->cmd_ring->first_seg->dma + 2*16,
1562 			.result_seg = NULL,
1563 		},
1564 	};
1565 
1566 	unsigned int num_tests;
1567 	int i, ret;
1568 
1569 	num_tests = sizeof(simple_test_vector) / sizeof(simple_test_vector[0]);
1570 	for (i = 0; i < num_tests; i++) {
1571 		ret = xhci_test_trb_in_td(xhci,
1572 				xhci->event_ring->first_seg,
1573 				xhci->event_ring->first_seg->trbs,
1574 				&xhci->event_ring->first_seg->trbs[TRBS_PER_SEGMENT - 1],
1575 				simple_test_vector[i].input_dma,
1576 				simple_test_vector[i].result_seg,
1577 				"Simple", i);
1578 		if (ret < 0)
1579 			return ret;
1580 	}
1581 
1582 	num_tests = sizeof(complex_test_vector) / sizeof(complex_test_vector[0]);
1583 	for (i = 0; i < num_tests; i++) {
1584 		ret = xhci_test_trb_in_td(xhci,
1585 				complex_test_vector[i].input_seg,
1586 				complex_test_vector[i].start_trb,
1587 				complex_test_vector[i].end_trb,
1588 				complex_test_vector[i].input_dma,
1589 				complex_test_vector[i].result_seg,
1590 				"Complex", i);
1591 		if (ret < 0)
1592 			return ret;
1593 	}
1594 	xhci_dbg(xhci, "TRB math tests passed.\n");
1595 	return 0;
1596 }
1597 
1598 
1599 int xhci_mem_init(struct xhci_hcd *xhci, gfp_t flags)
1600 {
1601 	dma_addr_t	dma;
1602 	struct device	*dev = xhci_to_hcd(xhci)->self.controller;
1603 	unsigned int	val, val2;
1604 	u64		val_64;
1605 	struct xhci_segment	*seg;
1606 	u32 page_size;
1607 	int i;
1608 
1609 	page_size = xhci_readl(xhci, &xhci->op_regs->page_size);
1610 	xhci_dbg(xhci, "Supported page size register = 0x%x\n", page_size);
1611 	for (i = 0; i < 16; i++) {
1612 		if ((0x1 & page_size) != 0)
1613 			break;
1614 		page_size = page_size >> 1;
1615 	}
1616 	if (i < 16)
1617 		xhci_dbg(xhci, "Supported page size of %iK\n", (1 << (i+12)) / 1024);
1618 	else
1619 		xhci_warn(xhci, "WARN: no supported page size\n");
1620 	/* Use 4K pages, since that's common and the minimum the HC supports */
1621 	xhci->page_shift = 12;
1622 	xhci->page_size = 1 << xhci->page_shift;
1623 	xhci_dbg(xhci, "HCD page size set to %iK\n", xhci->page_size / 1024);
1624 
1625 	/*
1626 	 * Program the Number of Device Slots Enabled field in the CONFIG
1627 	 * register with the max value of slots the HC can handle.
1628 	 */
1629 	val = HCS_MAX_SLOTS(xhci_readl(xhci, &xhci->cap_regs->hcs_params1));
1630 	xhci_dbg(xhci, "// xHC can handle at most %d device slots.\n",
1631 			(unsigned int) val);
1632 	val2 = xhci_readl(xhci, &xhci->op_regs->config_reg);
1633 	val |= (val2 & ~HCS_SLOTS_MASK);
1634 	xhci_dbg(xhci, "// Setting Max device slots reg = 0x%x.\n",
1635 			(unsigned int) val);
1636 	xhci_writel(xhci, val, &xhci->op_regs->config_reg);
1637 
1638 	/*
1639 	 * Section 5.4.8 - doorbell array must be
1640 	 * "physically contiguous and 64-byte (cache line) aligned".
1641 	 */
1642 	xhci->dcbaa = pci_alloc_consistent(to_pci_dev(dev),
1643 			sizeof(*xhci->dcbaa), &dma);
1644 	if (!xhci->dcbaa)
1645 		goto fail;
1646 	memset(xhci->dcbaa, 0, sizeof *(xhci->dcbaa));
1647 	xhci->dcbaa->dma = dma;
1648 	xhci_dbg(xhci, "// Device context base array address = 0x%llx (DMA), %p (virt)\n",
1649 			(unsigned long long)xhci->dcbaa->dma, xhci->dcbaa);
1650 	xhci_write_64(xhci, dma, &xhci->op_regs->dcbaa_ptr);
1651 
1652 	/*
1653 	 * Initialize the ring segment pool.  The ring must be a contiguous
1654 	 * structure comprised of TRBs.  The TRBs must be 16 byte aligned,
1655 	 * however, the command ring segment needs 64-byte aligned segments,
1656 	 * so we pick the greater alignment need.
1657 	 */
1658 	xhci->segment_pool = dma_pool_create("xHCI ring segments", dev,
1659 			SEGMENT_SIZE, 64, xhci->page_size);
1660 
1661 	/* See Table 46 and Note on Figure 55 */
1662 	xhci->device_pool = dma_pool_create("xHCI input/output contexts", dev,
1663 			2112, 64, xhci->page_size);
1664 	if (!xhci->segment_pool || !xhci->device_pool)
1665 		goto fail;
1666 
1667 	/* Linear stream context arrays don't have any boundary restrictions,
1668 	 * and only need to be 16-byte aligned.
1669 	 */
1670 	xhci->small_streams_pool =
1671 		dma_pool_create("xHCI 256 byte stream ctx arrays",
1672 			dev, SMALL_STREAM_ARRAY_SIZE, 16, 0);
1673 	xhci->medium_streams_pool =
1674 		dma_pool_create("xHCI 1KB stream ctx arrays",
1675 			dev, MEDIUM_STREAM_ARRAY_SIZE, 16, 0);
1676 	/* Any stream context array bigger than MEDIUM_STREAM_ARRAY_SIZE
1677 	 * will be allocated with pci_alloc_consistent()
1678 	 */
1679 
1680 	if (!xhci->small_streams_pool || !xhci->medium_streams_pool)
1681 		goto fail;
1682 
1683 	/* Set up the command ring to have one segments for now. */
1684 	xhci->cmd_ring = xhci_ring_alloc(xhci, 1, true, flags);
1685 	if (!xhci->cmd_ring)
1686 		goto fail;
1687 	xhci_dbg(xhci, "Allocated command ring at %p\n", xhci->cmd_ring);
1688 	xhci_dbg(xhci, "First segment DMA is 0x%llx\n",
1689 			(unsigned long long)xhci->cmd_ring->first_seg->dma);
1690 
1691 	/* Set the address in the Command Ring Control register */
1692 	val_64 = xhci_read_64(xhci, &xhci->op_regs->cmd_ring);
1693 	val_64 = (val_64 & (u64) CMD_RING_RSVD_BITS) |
1694 		(xhci->cmd_ring->first_seg->dma & (u64) ~CMD_RING_RSVD_BITS) |
1695 		xhci->cmd_ring->cycle_state;
1696 	xhci_dbg(xhci, "// Setting command ring address to 0x%x\n", val);
1697 	xhci_write_64(xhci, val_64, &xhci->op_regs->cmd_ring);
1698 	xhci_dbg_cmd_ptrs(xhci);
1699 
1700 	val = xhci_readl(xhci, &xhci->cap_regs->db_off);
1701 	val &= DBOFF_MASK;
1702 	xhci_dbg(xhci, "// Doorbell array is located at offset 0x%x"
1703 			" from cap regs base addr\n", val);
1704 	xhci->dba = (void *) xhci->cap_regs + val;
1705 	xhci_dbg_regs(xhci);
1706 	xhci_print_run_regs(xhci);
1707 	/* Set ir_set to interrupt register set 0 */
1708 	xhci->ir_set = (void *) xhci->run_regs->ir_set;
1709 
1710 	/*
1711 	 * Event ring setup: Allocate a normal ring, but also setup
1712 	 * the event ring segment table (ERST).  Section 4.9.3.
1713 	 */
1714 	xhci_dbg(xhci, "// Allocating event ring\n");
1715 	xhci->event_ring = xhci_ring_alloc(xhci, ERST_NUM_SEGS, false, flags);
1716 	if (!xhci->event_ring)
1717 		goto fail;
1718 	if (xhci_check_trb_in_td_math(xhci, flags) < 0)
1719 		goto fail;
1720 
1721 	xhci->erst.entries = pci_alloc_consistent(to_pci_dev(dev),
1722 			sizeof(struct xhci_erst_entry)*ERST_NUM_SEGS, &dma);
1723 	if (!xhci->erst.entries)
1724 		goto fail;
1725 	xhci_dbg(xhci, "// Allocated event ring segment table at 0x%llx\n",
1726 			(unsigned long long)dma);
1727 
1728 	memset(xhci->erst.entries, 0, sizeof(struct xhci_erst_entry)*ERST_NUM_SEGS);
1729 	xhci->erst.num_entries = ERST_NUM_SEGS;
1730 	xhci->erst.erst_dma_addr = dma;
1731 	xhci_dbg(xhci, "Set ERST to 0; private num segs = %i, virt addr = %p, dma addr = 0x%llx\n",
1732 			xhci->erst.num_entries,
1733 			xhci->erst.entries,
1734 			(unsigned long long)xhci->erst.erst_dma_addr);
1735 
1736 	/* set ring base address and size for each segment table entry */
1737 	for (val = 0, seg = xhci->event_ring->first_seg; val < ERST_NUM_SEGS; val++) {
1738 		struct xhci_erst_entry *entry = &xhci->erst.entries[val];
1739 		entry->seg_addr = seg->dma;
1740 		entry->seg_size = TRBS_PER_SEGMENT;
1741 		entry->rsvd = 0;
1742 		seg = seg->next;
1743 	}
1744 
1745 	/* set ERST count with the number of entries in the segment table */
1746 	val = xhci_readl(xhci, &xhci->ir_set->erst_size);
1747 	val &= ERST_SIZE_MASK;
1748 	val |= ERST_NUM_SEGS;
1749 	xhci_dbg(xhci, "// Write ERST size = %i to ir_set 0 (some bits preserved)\n",
1750 			val);
1751 	xhci_writel(xhci, val, &xhci->ir_set->erst_size);
1752 
1753 	xhci_dbg(xhci, "// Set ERST entries to point to event ring.\n");
1754 	/* set the segment table base address */
1755 	xhci_dbg(xhci, "// Set ERST base address for ir_set 0 = 0x%llx\n",
1756 			(unsigned long long)xhci->erst.erst_dma_addr);
1757 	val_64 = xhci_read_64(xhci, &xhci->ir_set->erst_base);
1758 	val_64 &= ERST_PTR_MASK;
1759 	val_64 |= (xhci->erst.erst_dma_addr & (u64) ~ERST_PTR_MASK);
1760 	xhci_write_64(xhci, val_64, &xhci->ir_set->erst_base);
1761 
1762 	/* Set the event ring dequeue address */
1763 	xhci_set_hc_event_deq(xhci);
1764 	xhci_dbg(xhci, "Wrote ERST address to ir_set 0.\n");
1765 	xhci_print_ir_set(xhci, xhci->ir_set, 0);
1766 
1767 	/*
1768 	 * XXX: Might need to set the Interrupter Moderation Register to
1769 	 * something other than the default (~1ms minimum between interrupts).
1770 	 * See section 5.5.1.2.
1771 	 */
1772 	init_completion(&xhci->addr_dev);
1773 	for (i = 0; i < MAX_HC_SLOTS; ++i)
1774 		xhci->devs[i] = NULL;
1775 
1776 	if (scratchpad_alloc(xhci, flags))
1777 		goto fail;
1778 
1779 	return 0;
1780 
1781 fail:
1782 	xhci_warn(xhci, "Couldn't initialize memory\n");
1783 	xhci_mem_cleanup(xhci);
1784 	return -ENOMEM;
1785 }
1786