xref: /openbmc/linux/drivers/usb/host/xhci-mem.c (revision 7dd65feb)
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/dmapool.h>
26 
27 #include "xhci.h"
28 
29 /*
30  * Allocates a generic ring segment from the ring pool, sets the dma address,
31  * initializes the segment to zero, and sets the private next pointer to NULL.
32  *
33  * Section 4.11.1.1:
34  * "All components of all Command and Transfer TRBs shall be initialized to '0'"
35  */
36 static struct xhci_segment *xhci_segment_alloc(struct xhci_hcd *xhci, gfp_t flags)
37 {
38 	struct xhci_segment *seg;
39 	dma_addr_t	dma;
40 
41 	seg = kzalloc(sizeof *seg, flags);
42 	if (!seg)
43 		return 0;
44 	xhci_dbg(xhci, "Allocating priv segment structure at %p\n", seg);
45 
46 	seg->trbs = dma_pool_alloc(xhci->segment_pool, flags, &dma);
47 	if (!seg->trbs) {
48 		kfree(seg);
49 		return 0;
50 	}
51 	xhci_dbg(xhci, "// Allocating segment at %p (virtual) 0x%llx (DMA)\n",
52 			seg->trbs, (unsigned long long)dma);
53 
54 	memset(seg->trbs, 0, SEGMENT_SIZE);
55 	seg->dma = dma;
56 	seg->next = NULL;
57 
58 	return seg;
59 }
60 
61 static void xhci_segment_free(struct xhci_hcd *xhci, struct xhci_segment *seg)
62 {
63 	if (!seg)
64 		return;
65 	if (seg->trbs) {
66 		xhci_dbg(xhci, "Freeing DMA segment at %p (virtual) 0x%llx (DMA)\n",
67 				seg->trbs, (unsigned long long)seg->dma);
68 		dma_pool_free(xhci->segment_pool, seg->trbs, seg->dma);
69 		seg->trbs = NULL;
70 	}
71 	xhci_dbg(xhci, "Freeing priv segment structure at %p\n", seg);
72 	kfree(seg);
73 }
74 
75 /*
76  * Make the prev segment point to the next segment.
77  *
78  * Change the last TRB in the prev segment to be a Link TRB which points to the
79  * DMA address of the next segment.  The caller needs to set any Link TRB
80  * related flags, such as End TRB, Toggle Cycle, and no snoop.
81  */
82 static void xhci_link_segments(struct xhci_hcd *xhci, struct xhci_segment *prev,
83 		struct xhci_segment *next, bool link_trbs)
84 {
85 	u32 val;
86 
87 	if (!prev || !next)
88 		return;
89 	prev->next = next;
90 	if (link_trbs) {
91 		prev->trbs[TRBS_PER_SEGMENT-1].link.segment_ptr = next->dma;
92 
93 		/* Set the last TRB in the segment to have a TRB type ID of Link TRB */
94 		val = 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 		if (xhci_link_trb_quirk(xhci))
99 			val |= TRB_CHAIN;
100 		prev->trbs[TRBS_PER_SEGMENT-1].link.control = val;
101 	}
102 	xhci_dbg(xhci, "Linking segment 0x%llx to segment 0x%llx (DMA)\n",
103 			(unsigned long long)prev->dma,
104 			(unsigned long long)next->dma);
105 }
106 
107 /* XXX: Do we need the hcd structure in all these functions? */
108 void xhci_ring_free(struct xhci_hcd *xhci, struct xhci_ring *ring)
109 {
110 	struct xhci_segment *seg;
111 	struct xhci_segment *first_seg;
112 
113 	if (!ring || !ring->first_seg)
114 		return;
115 	first_seg = ring->first_seg;
116 	seg = first_seg->next;
117 	xhci_dbg(xhci, "Freeing ring at %p\n", ring);
118 	while (seg != first_seg) {
119 		struct xhci_segment *next = seg->next;
120 		xhci_segment_free(xhci, seg);
121 		seg = next;
122 	}
123 	xhci_segment_free(xhci, first_seg);
124 	ring->first_seg = NULL;
125 	kfree(ring);
126 }
127 
128 static void xhci_initialize_ring_info(struct xhci_ring *ring)
129 {
130 	/* The ring is empty, so the enqueue pointer == dequeue pointer */
131 	ring->enqueue = ring->first_seg->trbs;
132 	ring->enq_seg = ring->first_seg;
133 	ring->dequeue = ring->enqueue;
134 	ring->deq_seg = ring->first_seg;
135 	/* The ring is initialized to 0. The producer must write 1 to the cycle
136 	 * bit to handover ownership of the TRB, so PCS = 1.  The consumer must
137 	 * compare CCS to the cycle bit to check ownership, so CCS = 1.
138 	 */
139 	ring->cycle_state = 1;
140 	/* Not necessary for new rings, but needed for re-initialized rings */
141 	ring->enq_updates = 0;
142 	ring->deq_updates = 0;
143 }
144 
145 /**
146  * Create a new ring with zero or more segments.
147  *
148  * Link each segment together into a ring.
149  * Set the end flag and the cycle toggle bit on the last segment.
150  * See section 4.9.1 and figures 15 and 16.
151  */
152 static struct xhci_ring *xhci_ring_alloc(struct xhci_hcd *xhci,
153 		unsigned int num_segs, bool link_trbs, gfp_t flags)
154 {
155 	struct xhci_ring	*ring;
156 	struct xhci_segment	*prev;
157 
158 	ring = kzalloc(sizeof *(ring), flags);
159 	xhci_dbg(xhci, "Allocating ring at %p\n", ring);
160 	if (!ring)
161 		return 0;
162 
163 	INIT_LIST_HEAD(&ring->td_list);
164 	if (num_segs == 0)
165 		return ring;
166 
167 	ring->first_seg = xhci_segment_alloc(xhci, flags);
168 	if (!ring->first_seg)
169 		goto fail;
170 	num_segs--;
171 
172 	prev = ring->first_seg;
173 	while (num_segs > 0) {
174 		struct xhci_segment	*next;
175 
176 		next = xhci_segment_alloc(xhci, flags);
177 		if (!next)
178 			goto fail;
179 		xhci_link_segments(xhci, prev, next, link_trbs);
180 
181 		prev = next;
182 		num_segs--;
183 	}
184 	xhci_link_segments(xhci, prev, ring->first_seg, link_trbs);
185 
186 	if (link_trbs) {
187 		/* See section 4.9.2.1 and 6.4.4.1 */
188 		prev->trbs[TRBS_PER_SEGMENT-1].link.control |= (LINK_TOGGLE);
189 		xhci_dbg(xhci, "Wrote link toggle flag to"
190 				" segment %p (virtual), 0x%llx (DMA)\n",
191 				prev, (unsigned long long)prev->dma);
192 	}
193 	xhci_initialize_ring_info(ring);
194 	return ring;
195 
196 fail:
197 	xhci_ring_free(xhci, ring);
198 	return 0;
199 }
200 
201 /* Zero an endpoint ring (except for link TRBs) and move the enqueue and dequeue
202  * pointers to the beginning of the ring.
203  */
204 static void xhci_reinit_cached_ring(struct xhci_hcd *xhci,
205 		struct xhci_ring *ring)
206 {
207 	struct xhci_segment	*seg = ring->first_seg;
208 	do {
209 		memset(seg->trbs, 0,
210 				sizeof(union xhci_trb)*TRBS_PER_SEGMENT);
211 		/* All endpoint rings have link TRBs */
212 		xhci_link_segments(xhci, seg, seg->next, 1);
213 		seg = seg->next;
214 	} while (seg != ring->first_seg);
215 	xhci_initialize_ring_info(ring);
216 	/* td list should be empty since all URBs have been cancelled,
217 	 * but just in case...
218 	 */
219 	INIT_LIST_HEAD(&ring->td_list);
220 }
221 
222 #define CTX_SIZE(_hcc) (HCC_64BYTE_CONTEXT(_hcc) ? 64 : 32)
223 
224 struct xhci_container_ctx *xhci_alloc_container_ctx(struct xhci_hcd *xhci,
225 						    int type, gfp_t flags)
226 {
227 	struct xhci_container_ctx *ctx = kzalloc(sizeof(*ctx), flags);
228 	if (!ctx)
229 		return NULL;
230 
231 	BUG_ON((type != XHCI_CTX_TYPE_DEVICE) && (type != XHCI_CTX_TYPE_INPUT));
232 	ctx->type = type;
233 	ctx->size = HCC_64BYTE_CONTEXT(xhci->hcc_params) ? 2048 : 1024;
234 	if (type == XHCI_CTX_TYPE_INPUT)
235 		ctx->size += CTX_SIZE(xhci->hcc_params);
236 
237 	ctx->bytes = dma_pool_alloc(xhci->device_pool, flags, &ctx->dma);
238 	memset(ctx->bytes, 0, ctx->size);
239 	return ctx;
240 }
241 
242 void xhci_free_container_ctx(struct xhci_hcd *xhci,
243 			     struct xhci_container_ctx *ctx)
244 {
245 	dma_pool_free(xhci->device_pool, ctx->bytes, ctx->dma);
246 	kfree(ctx);
247 }
248 
249 struct xhci_input_control_ctx *xhci_get_input_control_ctx(struct xhci_hcd *xhci,
250 					      struct xhci_container_ctx *ctx)
251 {
252 	BUG_ON(ctx->type != XHCI_CTX_TYPE_INPUT);
253 	return (struct xhci_input_control_ctx *)ctx->bytes;
254 }
255 
256 struct xhci_slot_ctx *xhci_get_slot_ctx(struct xhci_hcd *xhci,
257 					struct xhci_container_ctx *ctx)
258 {
259 	if (ctx->type == XHCI_CTX_TYPE_DEVICE)
260 		return (struct xhci_slot_ctx *)ctx->bytes;
261 
262 	return (struct xhci_slot_ctx *)
263 		(ctx->bytes + CTX_SIZE(xhci->hcc_params));
264 }
265 
266 struct xhci_ep_ctx *xhci_get_ep_ctx(struct xhci_hcd *xhci,
267 				    struct xhci_container_ctx *ctx,
268 				    unsigned int ep_index)
269 {
270 	/* increment ep index by offset of start of ep ctx array */
271 	ep_index++;
272 	if (ctx->type == XHCI_CTX_TYPE_INPUT)
273 		ep_index++;
274 
275 	return (struct xhci_ep_ctx *)
276 		(ctx->bytes + (ep_index * CTX_SIZE(xhci->hcc_params)));
277 }
278 
279 static void xhci_init_endpoint_timer(struct xhci_hcd *xhci,
280 		struct xhci_virt_ep *ep)
281 {
282 	init_timer(&ep->stop_cmd_timer);
283 	ep->stop_cmd_timer.data = (unsigned long) ep;
284 	ep->stop_cmd_timer.function = xhci_stop_endpoint_command_watchdog;
285 	ep->xhci = xhci;
286 }
287 
288 /* All the xhci_tds in the ring's TD list should be freed at this point */
289 void xhci_free_virt_device(struct xhci_hcd *xhci, int slot_id)
290 {
291 	struct xhci_virt_device *dev;
292 	int i;
293 
294 	/* Slot ID 0 is reserved */
295 	if (slot_id == 0 || !xhci->devs[slot_id])
296 		return;
297 
298 	dev = xhci->devs[slot_id];
299 	xhci->dcbaa->dev_context_ptrs[slot_id] = 0;
300 	if (!dev)
301 		return;
302 
303 	for (i = 0; i < 31; ++i)
304 		if (dev->eps[i].ring)
305 			xhci_ring_free(xhci, dev->eps[i].ring);
306 
307 	if (dev->ring_cache) {
308 		for (i = 0; i < dev->num_rings_cached; i++)
309 			xhci_ring_free(xhci, dev->ring_cache[i]);
310 		kfree(dev->ring_cache);
311 	}
312 
313 	if (dev->in_ctx)
314 		xhci_free_container_ctx(xhci, dev->in_ctx);
315 	if (dev->out_ctx)
316 		xhci_free_container_ctx(xhci, dev->out_ctx);
317 
318 	kfree(xhci->devs[slot_id]);
319 	xhci->devs[slot_id] = 0;
320 }
321 
322 int xhci_alloc_virt_device(struct xhci_hcd *xhci, int slot_id,
323 		struct usb_device *udev, gfp_t flags)
324 {
325 	struct xhci_virt_device *dev;
326 	int i;
327 
328 	/* Slot ID 0 is reserved */
329 	if (slot_id == 0 || xhci->devs[slot_id]) {
330 		xhci_warn(xhci, "Bad Slot ID %d\n", slot_id);
331 		return 0;
332 	}
333 
334 	xhci->devs[slot_id] = kzalloc(sizeof(*xhci->devs[slot_id]), flags);
335 	if (!xhci->devs[slot_id])
336 		return 0;
337 	dev = xhci->devs[slot_id];
338 
339 	/* Allocate the (output) device context that will be used in the HC. */
340 	dev->out_ctx = xhci_alloc_container_ctx(xhci, XHCI_CTX_TYPE_DEVICE, flags);
341 	if (!dev->out_ctx)
342 		goto fail;
343 
344 	xhci_dbg(xhci, "Slot %d output ctx = 0x%llx (dma)\n", slot_id,
345 			(unsigned long long)dev->out_ctx->dma);
346 
347 	/* Allocate the (input) device context for address device command */
348 	dev->in_ctx = xhci_alloc_container_ctx(xhci, XHCI_CTX_TYPE_INPUT, flags);
349 	if (!dev->in_ctx)
350 		goto fail;
351 
352 	xhci_dbg(xhci, "Slot %d input ctx = 0x%llx (dma)\n", slot_id,
353 			(unsigned long long)dev->in_ctx->dma);
354 
355 	/* Initialize the cancellation list and watchdog timers for each ep */
356 	for (i = 0; i < 31; i++) {
357 		xhci_init_endpoint_timer(xhci, &dev->eps[i]);
358 		INIT_LIST_HEAD(&dev->eps[i].cancelled_td_list);
359 	}
360 
361 	/* Allocate endpoint 0 ring */
362 	dev->eps[0].ring = xhci_ring_alloc(xhci, 1, true, flags);
363 	if (!dev->eps[0].ring)
364 		goto fail;
365 
366 	/* Allocate pointers to the ring cache */
367 	dev->ring_cache = kzalloc(
368 			sizeof(struct xhci_ring *)*XHCI_MAX_RINGS_CACHED,
369 			flags);
370 	if (!dev->ring_cache)
371 		goto fail;
372 	dev->num_rings_cached = 0;
373 
374 	init_completion(&dev->cmd_completion);
375 	INIT_LIST_HEAD(&dev->cmd_list);
376 
377 	/* Point to output device context in dcbaa. */
378 	xhci->dcbaa->dev_context_ptrs[slot_id] = dev->out_ctx->dma;
379 	xhci_dbg(xhci, "Set slot id %d dcbaa entry %p to 0x%llx\n",
380 			slot_id,
381 			&xhci->dcbaa->dev_context_ptrs[slot_id],
382 			(unsigned long long) xhci->dcbaa->dev_context_ptrs[slot_id]);
383 
384 	return 1;
385 fail:
386 	xhci_free_virt_device(xhci, slot_id);
387 	return 0;
388 }
389 
390 /* Setup an xHCI virtual device for a Set Address command */
391 int xhci_setup_addressable_virt_dev(struct xhci_hcd *xhci, struct usb_device *udev)
392 {
393 	struct xhci_virt_device *dev;
394 	struct xhci_ep_ctx	*ep0_ctx;
395 	struct usb_device	*top_dev;
396 	struct xhci_slot_ctx    *slot_ctx;
397 	struct xhci_input_control_ctx *ctrl_ctx;
398 
399 	dev = xhci->devs[udev->slot_id];
400 	/* Slot ID 0 is reserved */
401 	if (udev->slot_id == 0 || !dev) {
402 		xhci_warn(xhci, "Slot ID %d is not assigned to this device\n",
403 				udev->slot_id);
404 		return -EINVAL;
405 	}
406 	ep0_ctx = xhci_get_ep_ctx(xhci, dev->in_ctx, 0);
407 	ctrl_ctx = xhci_get_input_control_ctx(xhci, dev->in_ctx);
408 	slot_ctx = xhci_get_slot_ctx(xhci, dev->in_ctx);
409 
410 	/* 2) New slot context and endpoint 0 context are valid*/
411 	ctrl_ctx->add_flags = SLOT_FLAG | EP0_FLAG;
412 
413 	/* 3) Only the control endpoint is valid - one endpoint context */
414 	slot_ctx->dev_info |= LAST_CTX(1);
415 
416 	slot_ctx->dev_info |= (u32) udev->route;
417 	switch (udev->speed) {
418 	case USB_SPEED_SUPER:
419 		slot_ctx->dev_info |= (u32) SLOT_SPEED_SS;
420 		break;
421 	case USB_SPEED_HIGH:
422 		slot_ctx->dev_info |= (u32) SLOT_SPEED_HS;
423 		break;
424 	case USB_SPEED_FULL:
425 		slot_ctx->dev_info |= (u32) SLOT_SPEED_FS;
426 		break;
427 	case USB_SPEED_LOW:
428 		slot_ctx->dev_info |= (u32) SLOT_SPEED_LS;
429 		break;
430 	case USB_SPEED_VARIABLE:
431 		xhci_dbg(xhci, "FIXME xHCI doesn't support wireless speeds\n");
432 		return -EINVAL;
433 		break;
434 	default:
435 		/* Speed was set earlier, this shouldn't happen. */
436 		BUG();
437 	}
438 	/* Find the root hub port this device is under */
439 	for (top_dev = udev; top_dev->parent && top_dev->parent->parent;
440 			top_dev = top_dev->parent)
441 		/* Found device below root hub */;
442 	slot_ctx->dev_info2 |= (u32) ROOT_HUB_PORT(top_dev->portnum);
443 	xhci_dbg(xhci, "Set root hub portnum to %d\n", top_dev->portnum);
444 
445 	/* Is this a LS/FS device under a HS hub? */
446 	if ((udev->speed == USB_SPEED_LOW || udev->speed == USB_SPEED_FULL) &&
447 			udev->tt) {
448 		slot_ctx->tt_info = udev->tt->hub->slot_id;
449 		slot_ctx->tt_info |= udev->ttport << 8;
450 		if (udev->tt->multi)
451 			slot_ctx->dev_info |= DEV_MTT;
452 	}
453 	xhci_dbg(xhci, "udev->tt = %p\n", udev->tt);
454 	xhci_dbg(xhci, "udev->ttport = 0x%x\n", udev->ttport);
455 
456 	/* Step 4 - ring already allocated */
457 	/* Step 5 */
458 	ep0_ctx->ep_info2 = EP_TYPE(CTRL_EP);
459 	/*
460 	 * XXX: Not sure about wireless USB devices.
461 	 */
462 	switch (udev->speed) {
463 	case USB_SPEED_SUPER:
464 		ep0_ctx->ep_info2 |= MAX_PACKET(512);
465 		break;
466 	case USB_SPEED_HIGH:
467 	/* USB core guesses at a 64-byte max packet first for FS devices */
468 	case USB_SPEED_FULL:
469 		ep0_ctx->ep_info2 |= MAX_PACKET(64);
470 		break;
471 	case USB_SPEED_LOW:
472 		ep0_ctx->ep_info2 |= MAX_PACKET(8);
473 		break;
474 	case USB_SPEED_VARIABLE:
475 		xhci_dbg(xhci, "FIXME xHCI doesn't support wireless speeds\n");
476 		return -EINVAL;
477 		break;
478 	default:
479 		/* New speed? */
480 		BUG();
481 	}
482 	/* EP 0 can handle "burst" sizes of 1, so Max Burst Size field is 0 */
483 	ep0_ctx->ep_info2 |= MAX_BURST(0);
484 	ep0_ctx->ep_info2 |= ERROR_COUNT(3);
485 
486 	ep0_ctx->deq =
487 		dev->eps[0].ring->first_seg->dma;
488 	ep0_ctx->deq |= dev->eps[0].ring->cycle_state;
489 
490 	/* Steps 7 and 8 were done in xhci_alloc_virt_device() */
491 
492 	return 0;
493 }
494 
495 /* Return the polling or NAK interval.
496  *
497  * The polling interval is expressed in "microframes".  If xHCI's Interval field
498  * is set to N, it will service the endpoint every 2^(Interval)*125us.
499  *
500  * The NAK interval is one NAK per 1 to 255 microframes, or no NAKs if interval
501  * is set to 0.
502  */
503 static inline unsigned int xhci_get_endpoint_interval(struct usb_device *udev,
504 		struct usb_host_endpoint *ep)
505 {
506 	unsigned int interval = 0;
507 
508 	switch (udev->speed) {
509 	case USB_SPEED_HIGH:
510 		/* Max NAK rate */
511 		if (usb_endpoint_xfer_control(&ep->desc) ||
512 				usb_endpoint_xfer_bulk(&ep->desc))
513 			interval = ep->desc.bInterval;
514 		/* Fall through - SS and HS isoc/int have same decoding */
515 	case USB_SPEED_SUPER:
516 		if (usb_endpoint_xfer_int(&ep->desc) ||
517 				usb_endpoint_xfer_isoc(&ep->desc)) {
518 			if (ep->desc.bInterval == 0)
519 				interval = 0;
520 			else
521 				interval = ep->desc.bInterval - 1;
522 			if (interval > 15)
523 				interval = 15;
524 			if (interval != ep->desc.bInterval + 1)
525 				dev_warn(&udev->dev, "ep %#x - rounding interval to %d microframes\n",
526 						ep->desc.bEndpointAddress, 1 << interval);
527 		}
528 		break;
529 	/* Convert bInterval (in 1-255 frames) to microframes and round down to
530 	 * nearest power of 2.
531 	 */
532 	case USB_SPEED_FULL:
533 	case USB_SPEED_LOW:
534 		if (usb_endpoint_xfer_int(&ep->desc) ||
535 				usb_endpoint_xfer_isoc(&ep->desc)) {
536 			interval = fls(8*ep->desc.bInterval) - 1;
537 			if (interval > 10)
538 				interval = 10;
539 			if (interval < 3)
540 				interval = 3;
541 			if ((1 << interval) != 8*ep->desc.bInterval)
542 				dev_warn(&udev->dev, "ep %#x - rounding interval to %d microframes\n",
543 						ep->desc.bEndpointAddress, 1 << interval);
544 		}
545 		break;
546 	default:
547 		BUG();
548 	}
549 	return EP_INTERVAL(interval);
550 }
551 
552 static inline u32 xhci_get_endpoint_type(struct usb_device *udev,
553 		struct usb_host_endpoint *ep)
554 {
555 	int in;
556 	u32 type;
557 
558 	in = usb_endpoint_dir_in(&ep->desc);
559 	if (usb_endpoint_xfer_control(&ep->desc)) {
560 		type = EP_TYPE(CTRL_EP);
561 	} else if (usb_endpoint_xfer_bulk(&ep->desc)) {
562 		if (in)
563 			type = EP_TYPE(BULK_IN_EP);
564 		else
565 			type = EP_TYPE(BULK_OUT_EP);
566 	} else if (usb_endpoint_xfer_isoc(&ep->desc)) {
567 		if (in)
568 			type = EP_TYPE(ISOC_IN_EP);
569 		else
570 			type = EP_TYPE(ISOC_OUT_EP);
571 	} else if (usb_endpoint_xfer_int(&ep->desc)) {
572 		if (in)
573 			type = EP_TYPE(INT_IN_EP);
574 		else
575 			type = EP_TYPE(INT_OUT_EP);
576 	} else {
577 		BUG();
578 	}
579 	return type;
580 }
581 
582 int xhci_endpoint_init(struct xhci_hcd *xhci,
583 		struct xhci_virt_device *virt_dev,
584 		struct usb_device *udev,
585 		struct usb_host_endpoint *ep,
586 		gfp_t mem_flags)
587 {
588 	unsigned int ep_index;
589 	struct xhci_ep_ctx *ep_ctx;
590 	struct xhci_ring *ep_ring;
591 	unsigned int max_packet;
592 	unsigned int max_burst;
593 
594 	ep_index = xhci_get_endpoint_index(&ep->desc);
595 	ep_ctx = xhci_get_ep_ctx(xhci, virt_dev->in_ctx, ep_index);
596 
597 	/* Set up the endpoint ring */
598 	virt_dev->eps[ep_index].new_ring =
599 		xhci_ring_alloc(xhci, 1, true, mem_flags);
600 	if (!virt_dev->eps[ep_index].new_ring) {
601 		/* Attempt to use the ring cache */
602 		if (virt_dev->num_rings_cached == 0)
603 			return -ENOMEM;
604 		virt_dev->eps[ep_index].new_ring =
605 			virt_dev->ring_cache[virt_dev->num_rings_cached];
606 		virt_dev->ring_cache[virt_dev->num_rings_cached] = NULL;
607 		virt_dev->num_rings_cached--;
608 		xhci_reinit_cached_ring(xhci, virt_dev->eps[ep_index].new_ring);
609 	}
610 	ep_ring = virt_dev->eps[ep_index].new_ring;
611 	ep_ctx->deq = ep_ring->first_seg->dma | ep_ring->cycle_state;
612 
613 	ep_ctx->ep_info = xhci_get_endpoint_interval(udev, ep);
614 
615 	/* FIXME dig Mult and streams info out of ep companion desc */
616 
617 	/* Allow 3 retries for everything but isoc;
618 	 * error count = 0 means infinite retries.
619 	 */
620 	if (!usb_endpoint_xfer_isoc(&ep->desc))
621 		ep_ctx->ep_info2 = ERROR_COUNT(3);
622 	else
623 		ep_ctx->ep_info2 = ERROR_COUNT(1);
624 
625 	ep_ctx->ep_info2 |= xhci_get_endpoint_type(udev, ep);
626 
627 	/* Set the max packet size and max burst */
628 	switch (udev->speed) {
629 	case USB_SPEED_SUPER:
630 		max_packet = ep->desc.wMaxPacketSize;
631 		ep_ctx->ep_info2 |= MAX_PACKET(max_packet);
632 		/* dig out max burst from ep companion desc */
633 		if (!ep->ss_ep_comp) {
634 			xhci_warn(xhci, "WARN no SS endpoint companion descriptor.\n");
635 			max_packet = 0;
636 		} else {
637 			max_packet = ep->ss_ep_comp->desc.bMaxBurst;
638 		}
639 		ep_ctx->ep_info2 |= MAX_BURST(max_packet);
640 		break;
641 	case USB_SPEED_HIGH:
642 		/* bits 11:12 specify the number of additional transaction
643 		 * opportunities per microframe (USB 2.0, section 9.6.6)
644 		 */
645 		if (usb_endpoint_xfer_isoc(&ep->desc) ||
646 				usb_endpoint_xfer_int(&ep->desc)) {
647 			max_burst = (ep->desc.wMaxPacketSize & 0x1800) >> 11;
648 			ep_ctx->ep_info2 |= MAX_BURST(max_burst);
649 		}
650 		/* Fall through */
651 	case USB_SPEED_FULL:
652 	case USB_SPEED_LOW:
653 		max_packet = ep->desc.wMaxPacketSize & 0x3ff;
654 		ep_ctx->ep_info2 |= MAX_PACKET(max_packet);
655 		break;
656 	default:
657 		BUG();
658 	}
659 	/* FIXME Debug endpoint context */
660 	return 0;
661 }
662 
663 void xhci_endpoint_zero(struct xhci_hcd *xhci,
664 		struct xhci_virt_device *virt_dev,
665 		struct usb_host_endpoint *ep)
666 {
667 	unsigned int ep_index;
668 	struct xhci_ep_ctx *ep_ctx;
669 
670 	ep_index = xhci_get_endpoint_index(&ep->desc);
671 	ep_ctx = xhci_get_ep_ctx(xhci, virt_dev->in_ctx, ep_index);
672 
673 	ep_ctx->ep_info = 0;
674 	ep_ctx->ep_info2 = 0;
675 	ep_ctx->deq = 0;
676 	ep_ctx->tx_info = 0;
677 	/* Don't free the endpoint ring until the set interface or configuration
678 	 * request succeeds.
679 	 */
680 }
681 
682 /* Copy output xhci_ep_ctx to the input xhci_ep_ctx copy.
683  * Useful when you want to change one particular aspect of the endpoint and then
684  * issue a configure endpoint command.
685  */
686 void xhci_endpoint_copy(struct xhci_hcd *xhci,
687 		struct xhci_container_ctx *in_ctx,
688 		struct xhci_container_ctx *out_ctx,
689 		unsigned int ep_index)
690 {
691 	struct xhci_ep_ctx *out_ep_ctx;
692 	struct xhci_ep_ctx *in_ep_ctx;
693 
694 	out_ep_ctx = xhci_get_ep_ctx(xhci, out_ctx, ep_index);
695 	in_ep_ctx = xhci_get_ep_ctx(xhci, in_ctx, ep_index);
696 
697 	in_ep_ctx->ep_info = out_ep_ctx->ep_info;
698 	in_ep_ctx->ep_info2 = out_ep_ctx->ep_info2;
699 	in_ep_ctx->deq = out_ep_ctx->deq;
700 	in_ep_ctx->tx_info = out_ep_ctx->tx_info;
701 }
702 
703 /* Copy output xhci_slot_ctx to the input xhci_slot_ctx.
704  * Useful when you want to change one particular aspect of the endpoint and then
705  * issue a configure endpoint command.  Only the context entries field matters,
706  * but we'll copy the whole thing anyway.
707  */
708 void xhci_slot_copy(struct xhci_hcd *xhci,
709 		struct xhci_container_ctx *in_ctx,
710 		struct xhci_container_ctx *out_ctx)
711 {
712 	struct xhci_slot_ctx *in_slot_ctx;
713 	struct xhci_slot_ctx *out_slot_ctx;
714 
715 	in_slot_ctx = xhci_get_slot_ctx(xhci, in_ctx);
716 	out_slot_ctx = xhci_get_slot_ctx(xhci, out_ctx);
717 
718 	in_slot_ctx->dev_info = out_slot_ctx->dev_info;
719 	in_slot_ctx->dev_info2 = out_slot_ctx->dev_info2;
720 	in_slot_ctx->tt_info = out_slot_ctx->tt_info;
721 	in_slot_ctx->dev_state = out_slot_ctx->dev_state;
722 }
723 
724 /* Set up the scratchpad buffer array and scratchpad buffers, if needed. */
725 static int scratchpad_alloc(struct xhci_hcd *xhci, gfp_t flags)
726 {
727 	int i;
728 	struct device *dev = xhci_to_hcd(xhci)->self.controller;
729 	int num_sp = HCS_MAX_SCRATCHPAD(xhci->hcs_params2);
730 
731 	xhci_dbg(xhci, "Allocating %d scratchpad buffers\n", num_sp);
732 
733 	if (!num_sp)
734 		return 0;
735 
736 	xhci->scratchpad = kzalloc(sizeof(*xhci->scratchpad), flags);
737 	if (!xhci->scratchpad)
738 		goto fail_sp;
739 
740 	xhci->scratchpad->sp_array =
741 		pci_alloc_consistent(to_pci_dev(dev),
742 				     num_sp * sizeof(u64),
743 				     &xhci->scratchpad->sp_dma);
744 	if (!xhci->scratchpad->sp_array)
745 		goto fail_sp2;
746 
747 	xhci->scratchpad->sp_buffers = kzalloc(sizeof(void *) * num_sp, flags);
748 	if (!xhci->scratchpad->sp_buffers)
749 		goto fail_sp3;
750 
751 	xhci->scratchpad->sp_dma_buffers =
752 		kzalloc(sizeof(dma_addr_t) * num_sp, flags);
753 
754 	if (!xhci->scratchpad->sp_dma_buffers)
755 		goto fail_sp4;
756 
757 	xhci->dcbaa->dev_context_ptrs[0] = xhci->scratchpad->sp_dma;
758 	for (i = 0; i < num_sp; i++) {
759 		dma_addr_t dma;
760 		void *buf = pci_alloc_consistent(to_pci_dev(dev),
761 						 xhci->page_size, &dma);
762 		if (!buf)
763 			goto fail_sp5;
764 
765 		xhci->scratchpad->sp_array[i] = dma;
766 		xhci->scratchpad->sp_buffers[i] = buf;
767 		xhci->scratchpad->sp_dma_buffers[i] = dma;
768 	}
769 
770 	return 0;
771 
772  fail_sp5:
773 	for (i = i - 1; i >= 0; i--) {
774 		pci_free_consistent(to_pci_dev(dev), xhci->page_size,
775 				    xhci->scratchpad->sp_buffers[i],
776 				    xhci->scratchpad->sp_dma_buffers[i]);
777 	}
778 	kfree(xhci->scratchpad->sp_dma_buffers);
779 
780  fail_sp4:
781 	kfree(xhci->scratchpad->sp_buffers);
782 
783  fail_sp3:
784 	pci_free_consistent(to_pci_dev(dev), num_sp * sizeof(u64),
785 			    xhci->scratchpad->sp_array,
786 			    xhci->scratchpad->sp_dma);
787 
788  fail_sp2:
789 	kfree(xhci->scratchpad);
790 	xhci->scratchpad = NULL;
791 
792  fail_sp:
793 	return -ENOMEM;
794 }
795 
796 static void scratchpad_free(struct xhci_hcd *xhci)
797 {
798 	int num_sp;
799 	int i;
800 	struct pci_dev	*pdev = to_pci_dev(xhci_to_hcd(xhci)->self.controller);
801 
802 	if (!xhci->scratchpad)
803 		return;
804 
805 	num_sp = HCS_MAX_SCRATCHPAD(xhci->hcs_params2);
806 
807 	for (i = 0; i < num_sp; i++) {
808 		pci_free_consistent(pdev, xhci->page_size,
809 				    xhci->scratchpad->sp_buffers[i],
810 				    xhci->scratchpad->sp_dma_buffers[i]);
811 	}
812 	kfree(xhci->scratchpad->sp_dma_buffers);
813 	kfree(xhci->scratchpad->sp_buffers);
814 	pci_free_consistent(pdev, num_sp * sizeof(u64),
815 			    xhci->scratchpad->sp_array,
816 			    xhci->scratchpad->sp_dma);
817 	kfree(xhci->scratchpad);
818 	xhci->scratchpad = NULL;
819 }
820 
821 struct xhci_command *xhci_alloc_command(struct xhci_hcd *xhci,
822 		bool allocate_completion, gfp_t mem_flags)
823 {
824 	struct xhci_command *command;
825 
826 	command = kzalloc(sizeof(*command), mem_flags);
827 	if (!command)
828 		return NULL;
829 
830 	command->in_ctx =
831 		xhci_alloc_container_ctx(xhci, XHCI_CTX_TYPE_INPUT, mem_flags);
832 	if (!command->in_ctx) {
833 		kfree(command);
834 		return NULL;
835 	}
836 
837 	if (allocate_completion) {
838 		command->completion =
839 			kzalloc(sizeof(struct completion), mem_flags);
840 		if (!command->completion) {
841 			xhci_free_container_ctx(xhci, command->in_ctx);
842 			kfree(command);
843 			return NULL;
844 		}
845 		init_completion(command->completion);
846 	}
847 
848 	command->status = 0;
849 	INIT_LIST_HEAD(&command->cmd_list);
850 	return command;
851 }
852 
853 void xhci_free_command(struct xhci_hcd *xhci,
854 		struct xhci_command *command)
855 {
856 	xhci_free_container_ctx(xhci,
857 			command->in_ctx);
858 	kfree(command->completion);
859 	kfree(command);
860 }
861 
862 void xhci_mem_cleanup(struct xhci_hcd *xhci)
863 {
864 	struct pci_dev	*pdev = to_pci_dev(xhci_to_hcd(xhci)->self.controller);
865 	int size;
866 	int i;
867 
868 	/* Free the Event Ring Segment Table and the actual Event Ring */
869 	if (xhci->ir_set) {
870 		xhci_writel(xhci, 0, &xhci->ir_set->erst_size);
871 		xhci_write_64(xhci, 0, &xhci->ir_set->erst_base);
872 		xhci_write_64(xhci, 0, &xhci->ir_set->erst_dequeue);
873 	}
874 	size = sizeof(struct xhci_erst_entry)*(xhci->erst.num_entries);
875 	if (xhci->erst.entries)
876 		pci_free_consistent(pdev, size,
877 				xhci->erst.entries, xhci->erst.erst_dma_addr);
878 	xhci->erst.entries = NULL;
879 	xhci_dbg(xhci, "Freed ERST\n");
880 	if (xhci->event_ring)
881 		xhci_ring_free(xhci, xhci->event_ring);
882 	xhci->event_ring = NULL;
883 	xhci_dbg(xhci, "Freed event ring\n");
884 
885 	xhci_write_64(xhci, 0, &xhci->op_regs->cmd_ring);
886 	if (xhci->cmd_ring)
887 		xhci_ring_free(xhci, xhci->cmd_ring);
888 	xhci->cmd_ring = NULL;
889 	xhci_dbg(xhci, "Freed command ring\n");
890 
891 	for (i = 1; i < MAX_HC_SLOTS; ++i)
892 		xhci_free_virt_device(xhci, i);
893 
894 	if (xhci->segment_pool)
895 		dma_pool_destroy(xhci->segment_pool);
896 	xhci->segment_pool = NULL;
897 	xhci_dbg(xhci, "Freed segment pool\n");
898 
899 	if (xhci->device_pool)
900 		dma_pool_destroy(xhci->device_pool);
901 	xhci->device_pool = NULL;
902 	xhci_dbg(xhci, "Freed device context pool\n");
903 
904 	xhci_write_64(xhci, 0, &xhci->op_regs->dcbaa_ptr);
905 	if (xhci->dcbaa)
906 		pci_free_consistent(pdev, sizeof(*xhci->dcbaa),
907 				xhci->dcbaa, xhci->dcbaa->dma);
908 	xhci->dcbaa = NULL;
909 
910 	scratchpad_free(xhci);
911 	xhci->page_size = 0;
912 	xhci->page_shift = 0;
913 }
914 
915 static int xhci_test_trb_in_td(struct xhci_hcd *xhci,
916 		struct xhci_segment *input_seg,
917 		union xhci_trb *start_trb,
918 		union xhci_trb *end_trb,
919 		dma_addr_t input_dma,
920 		struct xhci_segment *result_seg,
921 		char *test_name, int test_number)
922 {
923 	unsigned long long start_dma;
924 	unsigned long long end_dma;
925 	struct xhci_segment *seg;
926 
927 	start_dma = xhci_trb_virt_to_dma(input_seg, start_trb);
928 	end_dma = xhci_trb_virt_to_dma(input_seg, end_trb);
929 
930 	seg = trb_in_td(input_seg, start_trb, end_trb, input_dma);
931 	if (seg != result_seg) {
932 		xhci_warn(xhci, "WARN: %s TRB math test %d failed!\n",
933 				test_name, test_number);
934 		xhci_warn(xhci, "Tested TRB math w/ seg %p and "
935 				"input DMA 0x%llx\n",
936 				input_seg,
937 				(unsigned long long) input_dma);
938 		xhci_warn(xhci, "starting TRB %p (0x%llx DMA), "
939 				"ending TRB %p (0x%llx DMA)\n",
940 				start_trb, start_dma,
941 				end_trb, end_dma);
942 		xhci_warn(xhci, "Expected seg %p, got seg %p\n",
943 				result_seg, seg);
944 		return -1;
945 	}
946 	return 0;
947 }
948 
949 /* TRB math checks for xhci_trb_in_td(), using the command and event rings. */
950 static int xhci_check_trb_in_td_math(struct xhci_hcd *xhci, gfp_t mem_flags)
951 {
952 	struct {
953 		dma_addr_t		input_dma;
954 		struct xhci_segment	*result_seg;
955 	} simple_test_vector [] = {
956 		/* A zeroed DMA field should fail */
957 		{ 0, NULL },
958 		/* One TRB before the ring start should fail */
959 		{ xhci->event_ring->first_seg->dma - 16, NULL },
960 		/* One byte before the ring start should fail */
961 		{ xhci->event_ring->first_seg->dma - 1, NULL },
962 		/* Starting TRB should succeed */
963 		{ xhci->event_ring->first_seg->dma, xhci->event_ring->first_seg },
964 		/* Ending TRB should succeed */
965 		{ xhci->event_ring->first_seg->dma + (TRBS_PER_SEGMENT - 1)*16,
966 			xhci->event_ring->first_seg },
967 		/* One byte after the ring end should fail */
968 		{ xhci->event_ring->first_seg->dma + (TRBS_PER_SEGMENT - 1)*16 + 1, NULL },
969 		/* One TRB after the ring end should fail */
970 		{ xhci->event_ring->first_seg->dma + (TRBS_PER_SEGMENT)*16, NULL },
971 		/* An address of all ones should fail */
972 		{ (dma_addr_t) (~0), NULL },
973 	};
974 	struct {
975 		struct xhci_segment	*input_seg;
976 		union xhci_trb		*start_trb;
977 		union xhci_trb		*end_trb;
978 		dma_addr_t		input_dma;
979 		struct xhci_segment	*result_seg;
980 	} complex_test_vector [] = {
981 		/* Test feeding a valid DMA address from a different ring */
982 		{	.input_seg = xhci->event_ring->first_seg,
983 			.start_trb = xhci->event_ring->first_seg->trbs,
984 			.end_trb = &xhci->event_ring->first_seg->trbs[TRBS_PER_SEGMENT - 1],
985 			.input_dma = xhci->cmd_ring->first_seg->dma,
986 			.result_seg = NULL,
987 		},
988 		/* Test feeding a valid end TRB from a different ring */
989 		{	.input_seg = xhci->event_ring->first_seg,
990 			.start_trb = xhci->event_ring->first_seg->trbs,
991 			.end_trb = &xhci->cmd_ring->first_seg->trbs[TRBS_PER_SEGMENT - 1],
992 			.input_dma = xhci->cmd_ring->first_seg->dma,
993 			.result_seg = NULL,
994 		},
995 		/* Test feeding a valid start and end TRB from a different ring */
996 		{	.input_seg = xhci->event_ring->first_seg,
997 			.start_trb = xhci->cmd_ring->first_seg->trbs,
998 			.end_trb = &xhci->cmd_ring->first_seg->trbs[TRBS_PER_SEGMENT - 1],
999 			.input_dma = xhci->cmd_ring->first_seg->dma,
1000 			.result_seg = NULL,
1001 		},
1002 		/* TRB in this ring, but after this TD */
1003 		{	.input_seg = xhci->event_ring->first_seg,
1004 			.start_trb = &xhci->event_ring->first_seg->trbs[0],
1005 			.end_trb = &xhci->event_ring->first_seg->trbs[3],
1006 			.input_dma = xhci->event_ring->first_seg->dma + 4*16,
1007 			.result_seg = NULL,
1008 		},
1009 		/* TRB in this ring, but before this TD */
1010 		{	.input_seg = xhci->event_ring->first_seg,
1011 			.start_trb = &xhci->event_ring->first_seg->trbs[3],
1012 			.end_trb = &xhci->event_ring->first_seg->trbs[6],
1013 			.input_dma = xhci->event_ring->first_seg->dma + 2*16,
1014 			.result_seg = NULL,
1015 		},
1016 		/* TRB in this ring, but after this wrapped TD */
1017 		{	.input_seg = xhci->event_ring->first_seg,
1018 			.start_trb = &xhci->event_ring->first_seg->trbs[TRBS_PER_SEGMENT - 3],
1019 			.end_trb = &xhci->event_ring->first_seg->trbs[1],
1020 			.input_dma = xhci->event_ring->first_seg->dma + 2*16,
1021 			.result_seg = NULL,
1022 		},
1023 		/* TRB in this ring, but before this wrapped TD */
1024 		{	.input_seg = xhci->event_ring->first_seg,
1025 			.start_trb = &xhci->event_ring->first_seg->trbs[TRBS_PER_SEGMENT - 3],
1026 			.end_trb = &xhci->event_ring->first_seg->trbs[1],
1027 			.input_dma = xhci->event_ring->first_seg->dma + (TRBS_PER_SEGMENT - 4)*16,
1028 			.result_seg = NULL,
1029 		},
1030 		/* TRB not in this ring, and we have a wrapped TD */
1031 		{	.input_seg = xhci->event_ring->first_seg,
1032 			.start_trb = &xhci->event_ring->first_seg->trbs[TRBS_PER_SEGMENT - 3],
1033 			.end_trb = &xhci->event_ring->first_seg->trbs[1],
1034 			.input_dma = xhci->cmd_ring->first_seg->dma + 2*16,
1035 			.result_seg = NULL,
1036 		},
1037 	};
1038 
1039 	unsigned int num_tests;
1040 	int i, ret;
1041 
1042 	num_tests = sizeof(simple_test_vector) / sizeof(simple_test_vector[0]);
1043 	for (i = 0; i < num_tests; i++) {
1044 		ret = xhci_test_trb_in_td(xhci,
1045 				xhci->event_ring->first_seg,
1046 				xhci->event_ring->first_seg->trbs,
1047 				&xhci->event_ring->first_seg->trbs[TRBS_PER_SEGMENT - 1],
1048 				simple_test_vector[i].input_dma,
1049 				simple_test_vector[i].result_seg,
1050 				"Simple", i);
1051 		if (ret < 0)
1052 			return ret;
1053 	}
1054 
1055 	num_tests = sizeof(complex_test_vector) / sizeof(complex_test_vector[0]);
1056 	for (i = 0; i < num_tests; i++) {
1057 		ret = xhci_test_trb_in_td(xhci,
1058 				complex_test_vector[i].input_seg,
1059 				complex_test_vector[i].start_trb,
1060 				complex_test_vector[i].end_trb,
1061 				complex_test_vector[i].input_dma,
1062 				complex_test_vector[i].result_seg,
1063 				"Complex", i);
1064 		if (ret < 0)
1065 			return ret;
1066 	}
1067 	xhci_dbg(xhci, "TRB math tests passed.\n");
1068 	return 0;
1069 }
1070 
1071 
1072 int xhci_mem_init(struct xhci_hcd *xhci, gfp_t flags)
1073 {
1074 	dma_addr_t	dma;
1075 	struct device	*dev = xhci_to_hcd(xhci)->self.controller;
1076 	unsigned int	val, val2;
1077 	u64		val_64;
1078 	struct xhci_segment	*seg;
1079 	u32 page_size;
1080 	int i;
1081 
1082 	page_size = xhci_readl(xhci, &xhci->op_regs->page_size);
1083 	xhci_dbg(xhci, "Supported page size register = 0x%x\n", page_size);
1084 	for (i = 0; i < 16; i++) {
1085 		if ((0x1 & page_size) != 0)
1086 			break;
1087 		page_size = page_size >> 1;
1088 	}
1089 	if (i < 16)
1090 		xhci_dbg(xhci, "Supported page size of %iK\n", (1 << (i+12)) / 1024);
1091 	else
1092 		xhci_warn(xhci, "WARN: no supported page size\n");
1093 	/* Use 4K pages, since that's common and the minimum the HC supports */
1094 	xhci->page_shift = 12;
1095 	xhci->page_size = 1 << xhci->page_shift;
1096 	xhci_dbg(xhci, "HCD page size set to %iK\n", xhci->page_size / 1024);
1097 
1098 	/*
1099 	 * Program the Number of Device Slots Enabled field in the CONFIG
1100 	 * register with the max value of slots the HC can handle.
1101 	 */
1102 	val = HCS_MAX_SLOTS(xhci_readl(xhci, &xhci->cap_regs->hcs_params1));
1103 	xhci_dbg(xhci, "// xHC can handle at most %d device slots.\n",
1104 			(unsigned int) val);
1105 	val2 = xhci_readl(xhci, &xhci->op_regs->config_reg);
1106 	val |= (val2 & ~HCS_SLOTS_MASK);
1107 	xhci_dbg(xhci, "// Setting Max device slots reg = 0x%x.\n",
1108 			(unsigned int) val);
1109 	xhci_writel(xhci, val, &xhci->op_regs->config_reg);
1110 
1111 	/*
1112 	 * Section 5.4.8 - doorbell array must be
1113 	 * "physically contiguous and 64-byte (cache line) aligned".
1114 	 */
1115 	xhci->dcbaa = pci_alloc_consistent(to_pci_dev(dev),
1116 			sizeof(*xhci->dcbaa), &dma);
1117 	if (!xhci->dcbaa)
1118 		goto fail;
1119 	memset(xhci->dcbaa, 0, sizeof *(xhci->dcbaa));
1120 	xhci->dcbaa->dma = dma;
1121 	xhci_dbg(xhci, "// Device context base array address = 0x%llx (DMA), %p (virt)\n",
1122 			(unsigned long long)xhci->dcbaa->dma, xhci->dcbaa);
1123 	xhci_write_64(xhci, dma, &xhci->op_regs->dcbaa_ptr);
1124 
1125 	/*
1126 	 * Initialize the ring segment pool.  The ring must be a contiguous
1127 	 * structure comprised of TRBs.  The TRBs must be 16 byte aligned,
1128 	 * however, the command ring segment needs 64-byte aligned segments,
1129 	 * so we pick the greater alignment need.
1130 	 */
1131 	xhci->segment_pool = dma_pool_create("xHCI ring segments", dev,
1132 			SEGMENT_SIZE, 64, xhci->page_size);
1133 
1134 	/* See Table 46 and Note on Figure 55 */
1135 	xhci->device_pool = dma_pool_create("xHCI input/output contexts", dev,
1136 			2112, 64, xhci->page_size);
1137 	if (!xhci->segment_pool || !xhci->device_pool)
1138 		goto fail;
1139 
1140 	/* Set up the command ring to have one segments for now. */
1141 	xhci->cmd_ring = xhci_ring_alloc(xhci, 1, true, flags);
1142 	if (!xhci->cmd_ring)
1143 		goto fail;
1144 	xhci_dbg(xhci, "Allocated command ring at %p\n", xhci->cmd_ring);
1145 	xhci_dbg(xhci, "First segment DMA is 0x%llx\n",
1146 			(unsigned long long)xhci->cmd_ring->first_seg->dma);
1147 
1148 	/* Set the address in the Command Ring Control register */
1149 	val_64 = xhci_read_64(xhci, &xhci->op_regs->cmd_ring);
1150 	val_64 = (val_64 & (u64) CMD_RING_RSVD_BITS) |
1151 		(xhci->cmd_ring->first_seg->dma & (u64) ~CMD_RING_RSVD_BITS) |
1152 		xhci->cmd_ring->cycle_state;
1153 	xhci_dbg(xhci, "// Setting command ring address to 0x%x\n", val);
1154 	xhci_write_64(xhci, val_64, &xhci->op_regs->cmd_ring);
1155 	xhci_dbg_cmd_ptrs(xhci);
1156 
1157 	val = xhci_readl(xhci, &xhci->cap_regs->db_off);
1158 	val &= DBOFF_MASK;
1159 	xhci_dbg(xhci, "// Doorbell array is located at offset 0x%x"
1160 			" from cap regs base addr\n", val);
1161 	xhci->dba = (void *) xhci->cap_regs + val;
1162 	xhci_dbg_regs(xhci);
1163 	xhci_print_run_regs(xhci);
1164 	/* Set ir_set to interrupt register set 0 */
1165 	xhci->ir_set = (void *) xhci->run_regs->ir_set;
1166 
1167 	/*
1168 	 * Event ring setup: Allocate a normal ring, but also setup
1169 	 * the event ring segment table (ERST).  Section 4.9.3.
1170 	 */
1171 	xhci_dbg(xhci, "// Allocating event ring\n");
1172 	xhci->event_ring = xhci_ring_alloc(xhci, ERST_NUM_SEGS, false, flags);
1173 	if (!xhci->event_ring)
1174 		goto fail;
1175 	if (xhci_check_trb_in_td_math(xhci, flags) < 0)
1176 		goto fail;
1177 
1178 	xhci->erst.entries = pci_alloc_consistent(to_pci_dev(dev),
1179 			sizeof(struct xhci_erst_entry)*ERST_NUM_SEGS, &dma);
1180 	if (!xhci->erst.entries)
1181 		goto fail;
1182 	xhci_dbg(xhci, "// Allocated event ring segment table at 0x%llx\n",
1183 			(unsigned long long)dma);
1184 
1185 	memset(xhci->erst.entries, 0, sizeof(struct xhci_erst_entry)*ERST_NUM_SEGS);
1186 	xhci->erst.num_entries = ERST_NUM_SEGS;
1187 	xhci->erst.erst_dma_addr = dma;
1188 	xhci_dbg(xhci, "Set ERST to 0; private num segs = %i, virt addr = %p, dma addr = 0x%llx\n",
1189 			xhci->erst.num_entries,
1190 			xhci->erst.entries,
1191 			(unsigned long long)xhci->erst.erst_dma_addr);
1192 
1193 	/* set ring base address and size for each segment table entry */
1194 	for (val = 0, seg = xhci->event_ring->first_seg; val < ERST_NUM_SEGS; val++) {
1195 		struct xhci_erst_entry *entry = &xhci->erst.entries[val];
1196 		entry->seg_addr = seg->dma;
1197 		entry->seg_size = TRBS_PER_SEGMENT;
1198 		entry->rsvd = 0;
1199 		seg = seg->next;
1200 	}
1201 
1202 	/* set ERST count with the number of entries in the segment table */
1203 	val = xhci_readl(xhci, &xhci->ir_set->erst_size);
1204 	val &= ERST_SIZE_MASK;
1205 	val |= ERST_NUM_SEGS;
1206 	xhci_dbg(xhci, "// Write ERST size = %i to ir_set 0 (some bits preserved)\n",
1207 			val);
1208 	xhci_writel(xhci, val, &xhci->ir_set->erst_size);
1209 
1210 	xhci_dbg(xhci, "// Set ERST entries to point to event ring.\n");
1211 	/* set the segment table base address */
1212 	xhci_dbg(xhci, "// Set ERST base address for ir_set 0 = 0x%llx\n",
1213 			(unsigned long long)xhci->erst.erst_dma_addr);
1214 	val_64 = xhci_read_64(xhci, &xhci->ir_set->erst_base);
1215 	val_64 &= ERST_PTR_MASK;
1216 	val_64 |= (xhci->erst.erst_dma_addr & (u64) ~ERST_PTR_MASK);
1217 	xhci_write_64(xhci, val_64, &xhci->ir_set->erst_base);
1218 
1219 	/* Set the event ring dequeue address */
1220 	xhci_set_hc_event_deq(xhci);
1221 	xhci_dbg(xhci, "Wrote ERST address to ir_set 0.\n");
1222 	xhci_print_ir_set(xhci, xhci->ir_set, 0);
1223 
1224 	/*
1225 	 * XXX: Might need to set the Interrupter Moderation Register to
1226 	 * something other than the default (~1ms minimum between interrupts).
1227 	 * See section 5.5.1.2.
1228 	 */
1229 	init_completion(&xhci->addr_dev);
1230 	for (i = 0; i < MAX_HC_SLOTS; ++i)
1231 		xhci->devs[i] = 0;
1232 
1233 	if (scratchpad_alloc(xhci, flags))
1234 		goto fail;
1235 
1236 	return 0;
1237 
1238 fail:
1239 	xhci_warn(xhci, "Couldn't initialize memory\n");
1240 	xhci_mem_cleanup(xhci);
1241 	return -ENOMEM;
1242 }
1243