xref: /linux/drivers/usb/host/xhci-mtk-sch.c (revision 8e1bb4a41aa78d6105e59186af3dcd545fc66e70)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2015 MediaTek Inc.
4  * Author:
5  *  Zhigang.Wei <zhigang.wei@mediatek.com>
6  *  Chunfeng.Yun <chunfeng.yun@mediatek.com>
7  */
8 
9 #include <linux/kernel.h>
10 #include <linux/module.h>
11 #include <linux/slab.h>
12 
13 #include "xhci.h"
14 #include "xhci-mtk.h"
15 
16 #define SSP_BW_BOUNDARY	130000
17 #define SS_BW_BOUNDARY	51000
18 /* table 5-5. High-speed Isoc Transaction Limits in usb_20 spec */
19 #define HS_BW_BOUNDARY	6144
20 /* usb2 spec section11.18.1: at most 188 FS bytes per microframe */
21 #define FS_PAYLOAD_MAX 188
22 #define LS_PAYLOAD_MAX 18
23 /* section 11.18.1, per fs frame */
24 #define FS_BW_BOUNDARY	1157
25 #define LS_BW_BOUNDARY	144
26 
27 /*
28  * max number of microframes for split transfer, assume extra-cs budget is 0
29  * for fs isoc in : 1 ss + 1 idle + 6 cs (roundup(1023/188))
30  */
31 #define TT_MICROFRAMES_MAX	8
32 /* offset from SS for fs/ls isoc/intr ep (ss + idle) */
33 #define CS_OFFSET	2
34 
35 #define DBG_BUF_EN	64
36 
37 /* schedule error type */
38 #define ESCH_SS_Y6		1001
39 #define ESCH_SS_OVERLAP		1002
40 #define ESCH_CS_OVERFLOW	1003
41 #define ESCH_BW_OVERFLOW	1004
42 #define ESCH_FIXME		1005
43 
44 /* mtk scheduler bitmasks */
45 #define EP_BPKTS(p)	((p) & 0x7f)
46 #define EP_BCSCOUNT(p)	(((p) & 0x7) << 8)
47 #define EP_BBM(p)	((p) << 11)
48 #define EP_BOFFSET(p)	((p) & 0x3fff)
49 #define EP_BREPEAT(p)	(((p) & 0x7fff) << 16)
50 
51 static char *sch_error_string(int err_num)
52 {
53 	switch (err_num) {
54 	case ESCH_SS_Y6:
55 		return "Can't schedule Start-Split in Y6";
56 	case ESCH_SS_OVERLAP:
57 		return "Can't find a suitable Start-Split location";
58 	case ESCH_CS_OVERFLOW:
59 		return "The last Complete-Split is greater than 7";
60 	case ESCH_BW_OVERFLOW:
61 		return "Bandwidth exceeds the maximum limit";
62 	case ESCH_FIXME:
63 		return "FIXME, to be resolved";
64 	default:
65 		return "Unknown";
66 	}
67 }
68 
69 static int is_fs_or_ls(enum usb_device_speed speed)
70 {
71 	return speed == USB_SPEED_FULL || speed == USB_SPEED_LOW;
72 }
73 
74 static const char *
75 decode_ep(struct usb_host_endpoint *ep, enum usb_device_speed speed)
76 {
77 	static char buf[DBG_BUF_EN];
78 	struct usb_endpoint_descriptor *epd = &ep->desc;
79 	unsigned int interval;
80 	const char *unit;
81 
82 	interval = usb_decode_interval(epd, speed);
83 	if (interval % 1000) {
84 		unit = "us";
85 	} else {
86 		unit = "ms";
87 		interval /= 1000;
88 	}
89 
90 	snprintf(buf, DBG_BUF_EN, "%s ep%d%s %s, mpkt:%d, interval:%d/%d%s",
91 		 usb_speed_string(speed), usb_endpoint_num(epd),
92 		 usb_endpoint_dir_in(epd) ? "in" : "out",
93 		 usb_ep_type_string(usb_endpoint_type(epd)),
94 		 usb_endpoint_maxp(epd), epd->bInterval, interval, unit);
95 
96 	return buf;
97 }
98 
99 static u32 get_bw_boundary(enum usb_device_speed speed)
100 {
101 	u32 boundary;
102 
103 	switch (speed) {
104 	case USB_SPEED_SUPER_PLUS:
105 		boundary = SSP_BW_BOUNDARY;
106 		break;
107 	case USB_SPEED_SUPER:
108 		boundary = SS_BW_BOUNDARY;
109 		break;
110 	default:
111 		boundary = HS_BW_BOUNDARY;
112 		break;
113 	}
114 
115 	return boundary;
116 }
117 
118 /*
119 * get the bandwidth domain which @ep belongs to.
120 *
121 * the bandwidth domain array is saved to @sch_array of struct xhci_hcd_mtk,
122 * each HS root port is treated as a single bandwidth domain,
123 * but each SS root port is treated as two bandwidth domains, one for IN eps,
124 * one for OUT eps.
125 */
126 static struct mu3h_sch_bw_info *
127 get_bw_info(struct xhci_hcd_mtk *mtk, struct usb_device *udev,
128 	    struct usb_host_endpoint *ep)
129 {
130 	struct xhci_hcd *xhci = hcd_to_xhci(mtk->hcd);
131 	struct xhci_virt_device *virt_dev;
132 	int bw_index;
133 
134 	virt_dev = xhci->devs[udev->slot_id];
135 	if (!virt_dev->rhub_port) {
136 		WARN_ONCE(1, "%s invalid rhub port\n", dev_name(&udev->dev));
137 		return NULL;
138 	}
139 
140 	if (udev->speed >= USB_SPEED_SUPER) {
141 		if (usb_endpoint_dir_out(&ep->desc))
142 			bw_index = (virt_dev->rhub_port->hw_portnum) * 2;
143 		else
144 			bw_index = (virt_dev->rhub_port->hw_portnum) * 2 + 1;
145 	} else {
146 		/* add one more for each SS port */
147 		bw_index = virt_dev->rhub_port->hw_portnum + xhci->usb3_rhub.num_ports;
148 	}
149 
150 	return &mtk->sch_array[bw_index];
151 }
152 
153 static u32 get_esit(struct xhci_ep_ctx *ep_ctx)
154 {
155 	u32 esit;
156 
157 	esit = 1 << CTX_TO_EP_INTERVAL(le32_to_cpu(ep_ctx->ep_info));
158 	if (esit > XHCI_MTK_MAX_ESIT)
159 		esit = XHCI_MTK_MAX_ESIT;
160 
161 	return esit;
162 }
163 
164 static struct mu3h_sch_tt *find_tt(struct usb_device *udev)
165 {
166 	struct usb_tt *utt = udev->tt;
167 	struct mu3h_sch_tt *tt, **tt_index, **ptt;
168 	bool allocated_index = false;
169 
170 	if (!utt)
171 		return NULL;	/* Not below a TT */
172 
173 	/*
174 	 * Find/create our data structure.
175 	 * For hubs with a single TT, we get it directly.
176 	 * For hubs with multiple TTs, there's an extra level of pointers.
177 	 */
178 	tt_index = NULL;
179 	if (utt->multi) {
180 		tt_index = utt->hcpriv;
181 		if (!tt_index) {	/* Create the index array */
182 			tt_index = kcalloc(utt->hub->maxchild,
183 					sizeof(*tt_index), GFP_KERNEL);
184 			if (!tt_index)
185 				return ERR_PTR(-ENOMEM);
186 			utt->hcpriv = tt_index;
187 			allocated_index = true;
188 		}
189 		ptt = &tt_index[udev->ttport - 1];
190 	} else {
191 		ptt = (struct mu3h_sch_tt **) &utt->hcpriv;
192 	}
193 
194 	tt = *ptt;
195 	if (!tt) {	/* Create the mu3h_sch_tt */
196 		tt = kzalloc(sizeof(*tt), GFP_KERNEL);
197 		if (!tt) {
198 			if (allocated_index) {
199 				utt->hcpriv = NULL;
200 				kfree(tt_index);
201 			}
202 			return ERR_PTR(-ENOMEM);
203 		}
204 		INIT_LIST_HEAD(&tt->ep_list);
205 		*ptt = tt;
206 	}
207 
208 	return tt;
209 }
210 
211 /* Release the TT above udev, if it's not in use */
212 static void drop_tt(struct usb_device *udev)
213 {
214 	struct usb_tt *utt = udev->tt;
215 	struct mu3h_sch_tt *tt, **tt_index, **ptt;
216 	int i, cnt;
217 
218 	if (!utt || !utt->hcpriv)
219 		return;		/* Not below a TT, or never allocated */
220 
221 	cnt = 0;
222 	if (utt->multi) {
223 		tt_index = utt->hcpriv;
224 		ptt = &tt_index[udev->ttport - 1];
225 		/*  How many entries are left in tt_index? */
226 		for (i = 0; i < utt->hub->maxchild; ++i)
227 			cnt += !!tt_index[i];
228 	} else {
229 		tt_index = NULL;
230 		ptt = (struct mu3h_sch_tt **)&utt->hcpriv;
231 	}
232 
233 	tt = *ptt;
234 	if (!tt || !list_empty(&tt->ep_list))
235 		return;		/* never allocated , or still in use*/
236 
237 	*ptt = NULL;
238 	kfree(tt);
239 
240 	if (cnt == 1) {
241 		utt->hcpriv = NULL;
242 		kfree(tt_index);
243 	}
244 }
245 
246 static struct mu3h_sch_ep_info *
247 create_sch_ep(struct xhci_hcd_mtk *mtk, struct usb_device *udev,
248 	      struct usb_host_endpoint *ep, struct xhci_ep_ctx *ep_ctx)
249 {
250 	struct mu3h_sch_ep_info *sch_ep;
251 	struct mu3h_sch_bw_info *bw_info;
252 	struct mu3h_sch_tt *tt = NULL;
253 	u32 len;
254 
255 	bw_info = get_bw_info(mtk, udev, ep);
256 	if (!bw_info)
257 		return ERR_PTR(-ENODEV);
258 
259 	if (is_fs_or_ls(udev->speed))
260 		len = TT_MICROFRAMES_MAX;
261 	else if ((udev->speed >= USB_SPEED_SUPER) &&
262 		 usb_endpoint_xfer_isoc(&ep->desc))
263 		len = get_esit(ep_ctx);
264 	else
265 		len = 1;
266 
267 	sch_ep = kzalloc(struct_size(sch_ep, bw_budget_table, len), GFP_KERNEL);
268 	if (!sch_ep)
269 		return ERR_PTR(-ENOMEM);
270 
271 	if (is_fs_or_ls(udev->speed)) {
272 		tt = find_tt(udev);
273 		if (IS_ERR(tt)) {
274 			kfree(sch_ep);
275 			return ERR_PTR(-ENOMEM);
276 		}
277 	}
278 
279 	sch_ep->bw_info = bw_info;
280 	sch_ep->sch_tt = tt;
281 	sch_ep->ep = ep;
282 	sch_ep->speed = udev->speed;
283 	INIT_LIST_HEAD(&sch_ep->endpoint);
284 	INIT_LIST_HEAD(&sch_ep->tt_endpoint);
285 	INIT_HLIST_NODE(&sch_ep->hentry);
286 
287 	return sch_ep;
288 }
289 
290 static void setup_sch_info(struct xhci_ep_ctx *ep_ctx,
291 			   struct mu3h_sch_ep_info *sch_ep)
292 {
293 	u32 ep_type;
294 	u32 maxpkt;
295 	u32 max_burst;
296 	u32 mult;
297 	u32 esit_pkts;
298 	u32 max_esit_payload;
299 	u32 bw_per_microframe;
300 	u32 *bwb_table;
301 	int i;
302 
303 	bwb_table = sch_ep->bw_budget_table;
304 	ep_type = CTX_TO_EP_TYPE(le32_to_cpu(ep_ctx->ep_info2));
305 	maxpkt = MAX_PACKET_DECODED(le32_to_cpu(ep_ctx->ep_info2));
306 	max_burst = CTX_TO_MAX_BURST(le32_to_cpu(ep_ctx->ep_info2));
307 	mult = CTX_TO_EP_MULT(le32_to_cpu(ep_ctx->ep_info));
308 	max_esit_payload =
309 		(CTX_TO_MAX_ESIT_PAYLOAD_HI(
310 			le32_to_cpu(ep_ctx->ep_info)) << 16) |
311 		 CTX_TO_MAX_ESIT_PAYLOAD(le32_to_cpu(ep_ctx->tx_info));
312 
313 	sch_ep->esit = get_esit(ep_ctx);
314 	sch_ep->num_esit = XHCI_MTK_MAX_ESIT / sch_ep->esit;
315 	sch_ep->ep_type = ep_type;
316 	sch_ep->maxpkt = maxpkt;
317 	sch_ep->offset = 0;
318 	sch_ep->burst_mode = 0;
319 	sch_ep->repeat = 0;
320 
321 	if (sch_ep->speed == USB_SPEED_HIGH) {
322 		sch_ep->cs_count = 0;
323 
324 		/*
325 		 * usb_20 spec section5.9
326 		 * a single microframe is enough for HS synchromous endpoints
327 		 * in a interval
328 		 */
329 		sch_ep->num_budget_microframes = 1;
330 
331 		/*
332 		 * xHCI spec section6.2.3.4
333 		 * @max_burst is the number of additional transactions
334 		 * opportunities per microframe
335 		 */
336 		sch_ep->pkts = max_burst + 1;
337 		bwb_table[0] = maxpkt * sch_ep->pkts;
338 	} else if (sch_ep->speed >= USB_SPEED_SUPER) {
339 		/* usb3_r1 spec section4.4.7 & 4.4.8 */
340 		sch_ep->cs_count = 0;
341 		sch_ep->burst_mode = 1;
342 		/*
343 		 * some device's (d)wBytesPerInterval is set as 0,
344 		 * then max_esit_payload is 0, so evaluate esit_pkts from
345 		 * mult and burst
346 		 */
347 		esit_pkts = DIV_ROUND_UP(max_esit_payload, maxpkt);
348 		if (esit_pkts == 0)
349 			esit_pkts = (mult + 1) * (max_burst + 1);
350 
351 		if (ep_type == INT_IN_EP || ep_type == INT_OUT_EP) {
352 			sch_ep->pkts = esit_pkts;
353 			sch_ep->num_budget_microframes = 1;
354 			bwb_table[0] = maxpkt * sch_ep->pkts;
355 		}
356 
357 		if (ep_type == ISOC_IN_EP || ep_type == ISOC_OUT_EP) {
358 
359 			if (sch_ep->esit == 1)
360 				sch_ep->pkts = esit_pkts;
361 			else if (esit_pkts <= sch_ep->esit)
362 				sch_ep->pkts = 1;
363 			else
364 				sch_ep->pkts = roundup_pow_of_two(esit_pkts)
365 					/ sch_ep->esit;
366 
367 			sch_ep->num_budget_microframes =
368 				DIV_ROUND_UP(esit_pkts, sch_ep->pkts);
369 
370 			sch_ep->repeat = !!(sch_ep->num_budget_microframes > 1);
371 			bw_per_microframe = maxpkt * sch_ep->pkts;
372 
373 			for (i = 0; i < sch_ep->num_budget_microframes - 1; i++)
374 				bwb_table[i] = bw_per_microframe;
375 
376 			/* last one <= bw_per_microframe */
377 			bwb_table[i] = maxpkt * esit_pkts - i * bw_per_microframe;
378 		}
379 	} else if (is_fs_or_ls(sch_ep->speed)) {
380 		sch_ep->pkts = 1; /* at most one packet for each microframe */
381 
382 		/*
383 		 * @cs_count will be updated to add extra-cs when
384 		 * check TT for INT_OUT_EP, ISOC/INT_IN_EP type
385 		 * @maxpkt <= 1023;
386 		 */
387 		sch_ep->cs_count = DIV_ROUND_UP(maxpkt, FS_PAYLOAD_MAX);
388 		sch_ep->num_budget_microframes = sch_ep->cs_count;
389 
390 		/* init budget table */
391 		if (ep_type == ISOC_OUT_EP) {
392 			for (i = 0; i < sch_ep->cs_count - 1; i++)
393 				bwb_table[i] = FS_PAYLOAD_MAX;
394 
395 			bwb_table[i] = maxpkt - i * FS_PAYLOAD_MAX;
396 		} else if (ep_type == INT_OUT_EP) {
397 			/* only first one used (maxpkt <= 64), others zero */
398 			bwb_table[0] = maxpkt;
399 		} else { /* INT_IN_EP or ISOC_IN_EP */
400 			bwb_table[0] = 0; /* start split */
401 			bwb_table[1] = 0; /* idle */
402 			/*
403 			 * @cs_count will be updated according to cs position
404 			 * (add 1 or 2 extra-cs), but assume only first
405 			 * @num_budget_microframes elements will be used later,
406 			 * although in fact it does not (extra-cs budget many receive
407 			 * some data for IN ep);
408 			 * @cs_count is 1 for INT_IN_EP (maxpkt <= 64);
409 			 */
410 			for (i = 0; i < sch_ep->cs_count - 1; i++)
411 				bwb_table[i + CS_OFFSET] = FS_PAYLOAD_MAX;
412 
413 			bwb_table[i + CS_OFFSET] = maxpkt - i * FS_PAYLOAD_MAX;
414 			/* ss + idle */
415 			sch_ep->num_budget_microframes += CS_OFFSET;
416 		}
417 	}
418 }
419 
420 /* Get maximum bandwidth when we schedule at offset slot. */
421 static u32 get_max_bw(struct mu3h_sch_bw_info *sch_bw,
422 	struct mu3h_sch_ep_info *sch_ep, u32 offset)
423 {
424 	u32 max_bw = 0;
425 	u32 bw;
426 	int i, j, k;
427 
428 	for (i = 0; i < sch_ep->num_esit; i++) {
429 		u32 base = offset + i * sch_ep->esit;
430 
431 		for (j = 0; j < sch_ep->num_budget_microframes; j++) {
432 			k = XHCI_MTK_BW_INDEX(base + j);
433 			bw = sch_bw->bus_bw[k] + sch_ep->bw_budget_table[j];
434 			if (bw > max_bw)
435 				max_bw = bw;
436 		}
437 	}
438 	return max_bw;
439 }
440 
441 /*
442  * for OUT: get first SS consumed bw;
443  * for IN: get first CS consumed bw;
444  */
445 static u16 get_fs_bw(struct mu3h_sch_ep_info *sch_ep, int offset)
446 {
447 	struct mu3h_sch_tt *tt = sch_ep->sch_tt;
448 	u16 fs_bw;
449 
450 	if (sch_ep->ep_type == ISOC_OUT_EP || sch_ep->ep_type == INT_OUT_EP)
451 		fs_bw = tt->fs_bus_bw_out[XHCI_MTK_BW_INDEX(offset)];
452 	else	/* skip ss + idle */
453 		fs_bw = tt->fs_bus_bw_in[XHCI_MTK_BW_INDEX(offset + CS_OFFSET)];
454 
455 	return fs_bw;
456 }
457 
458 static void update_bus_bw(struct mu3h_sch_bw_info *sch_bw,
459 	struct mu3h_sch_ep_info *sch_ep, bool used)
460 {
461 	u32 base;
462 	int i, j, k;
463 
464 	for (i = 0; i < sch_ep->num_esit; i++) {
465 		base = sch_ep->offset + i * sch_ep->esit;
466 		for (j = 0; j < sch_ep->num_budget_microframes; j++) {
467 			k = XHCI_MTK_BW_INDEX(base + j);
468 			if (used)
469 				sch_bw->bus_bw[k] += sch_ep->bw_budget_table[j];
470 			else
471 				sch_bw->bus_bw[k] -= sch_ep->bw_budget_table[j];
472 		}
473 	}
474 }
475 
476 static int check_ls_budget_microframes(struct mu3h_sch_ep_info *sch_ep, int offset)
477 {
478 	struct mu3h_sch_tt *tt = sch_ep->sch_tt;
479 	int i;
480 
481 	if (sch_ep->speed != USB_SPEED_LOW)
482 		return 0;
483 
484 	if (sch_ep->ep_type == INT_OUT_EP)
485 		i = XHCI_MTK_BW_INDEX(offset);
486 	else if (sch_ep->ep_type == INT_IN_EP)
487 		i = XHCI_MTK_BW_INDEX(offset + CS_OFFSET); /* skip ss + idle */
488 	else
489 		return -EINVAL;
490 
491 	if (tt->ls_bus_bw[i] + sch_ep->maxpkt > LS_PAYLOAD_MAX)
492 		return -ESCH_BW_OVERFLOW;
493 
494 	return 0;
495 }
496 
497 static int check_fs_budget_microframes(struct mu3h_sch_ep_info *sch_ep, int offset)
498 {
499 	struct mu3h_sch_tt *tt = sch_ep->sch_tt;
500 	u32 tmp;
501 	int i, k;
502 
503 	/*
504 	 * for OUT eps, will transfer exactly assigned length of data,
505 	 * so can't allocate more than 188 bytes;
506 	 * but it's not for IN eps, usually it can't receive full
507 	 * 188 bytes in a uframe, if it not assign full 188 bytes,
508 	 * can add another one;
509 	 */
510 	for (i = 0; i < sch_ep->num_budget_microframes; i++) {
511 		k = XHCI_MTK_BW_INDEX(offset + i);
512 		if (sch_ep->ep_type == ISOC_OUT_EP || sch_ep->ep_type == INT_OUT_EP)
513 			tmp = tt->fs_bus_bw_out[k] + sch_ep->bw_budget_table[i];
514 		else /* ep_type : ISOC IN / INTR IN */
515 			tmp = tt->fs_bus_bw_in[k];
516 
517 		if (tmp > FS_PAYLOAD_MAX)
518 			return -ESCH_BW_OVERFLOW;
519 	}
520 
521 	return 0;
522 }
523 
524 static int check_fs_budget_frames(struct mu3h_sch_ep_info *sch_ep, int offset)
525 {
526 	struct mu3h_sch_tt *tt = sch_ep->sch_tt;
527 	u32 head, tail;
528 	int i, j, k;
529 
530 	/* bugdet scheduled may cross at most two fs frames */
531 	j = XHCI_MTK_BW_INDEX(offset) / UFRAMES_PER_FRAME;
532 	k = XHCI_MTK_BW_INDEX(offset + sch_ep->num_budget_microframes - 1) / UFRAMES_PER_FRAME;
533 
534 	if (j != k) {
535 		head = tt->fs_frame_bw[j];
536 		tail = tt->fs_frame_bw[k];
537 	} else {
538 		head = tt->fs_frame_bw[j];
539 		tail = 0;
540 	}
541 
542 	j = roundup(offset, UFRAMES_PER_FRAME);
543 	for (i = 0; i < sch_ep->num_budget_microframes; i++) {
544 		if ((offset + i) < j)
545 			head += sch_ep->bw_budget_table[i];
546 		else
547 			tail += sch_ep->bw_budget_table[i];
548 	}
549 
550 	if (head > FS_BW_BOUNDARY || tail > FS_BW_BOUNDARY)
551 		return -ESCH_BW_OVERFLOW;
552 
553 	return 0;
554 }
555 
556 static int check_fs_bus_bw(struct mu3h_sch_ep_info *sch_ep, int offset)
557 {
558 	int i, base;
559 	int ret = 0;
560 
561 	for (i = 0; i < sch_ep->num_esit; i++) {
562 		base = offset + i * sch_ep->esit;
563 
564 		ret = check_ls_budget_microframes(sch_ep, base);
565 		if (ret)
566 			goto err;
567 
568 		ret = check_fs_budget_microframes(sch_ep, base);
569 		if (ret)
570 			goto err;
571 
572 		ret = check_fs_budget_frames(sch_ep, base);
573 		if (ret)
574 			goto err;
575 	}
576 
577 err:
578 	return ret;
579 }
580 
581 static int check_ss_and_cs(struct mu3h_sch_ep_info *sch_ep, u32 offset)
582 {
583 	u32 start_ss, last_ss;
584 	u32 start_cs, last_cs;
585 
586 	start_ss = offset % UFRAMES_PER_FRAME;
587 
588 	if (sch_ep->ep_type == ISOC_OUT_EP) {
589 		last_ss = start_ss + sch_ep->cs_count - 1;
590 
591 		/*
592 		 * usb_20 spec section11.18:
593 		 * must never schedule Start-Split in Y6
594 		 */
595 		if (!(start_ss == 7 || last_ss < 6))
596 			return -ESCH_SS_Y6;
597 
598 	} else {
599 		/* maxpkt <= 1023, cs <= 6 */
600 		u32 cs_count = DIV_ROUND_UP(sch_ep->maxpkt, FS_PAYLOAD_MAX);
601 
602 		/*
603 		 * usb_20 spec section11.18:
604 		 * must never schedule Start-Split in Y6
605 		 */
606 		if (start_ss == 6)
607 			return -ESCH_SS_Y6;
608 
609 		/* one uframe for ss + one uframe for idle */
610 		start_cs = (start_ss + CS_OFFSET) % UFRAMES_PER_FRAME;
611 		last_cs = start_cs + cs_count - 1;
612 		if (last_cs > 7)
613 			return -ESCH_CS_OVERFLOW;
614 
615 		/* add extra-cs */
616 		cs_count += (last_cs == 7) ? 1 : 2;
617 		if (cs_count > 7)
618 			cs_count = 7; /* HW limit */
619 
620 		sch_ep->cs_count = cs_count;
621 
622 	}
623 
624 	return 0;
625 }
626 
627 /*
628  * when isoc-out transfers 188 bytes in a uframe, and send isoc/intr's
629  * ss token in the uframe, may cause 'bit stuff error' in downstream
630  * port;
631  * when isoc-out transfer less than 188 bytes in a uframe, shall send
632  * isoc-in's ss after isoc-out's ss (but hw can't ensure the sequence,
633  * so just avoid overlap).
634  */
635 static int check_isoc_ss_overlap(struct mu3h_sch_ep_info *sch_ep, u32 offset)
636 {
637 	struct mu3h_sch_tt *tt = sch_ep->sch_tt;
638 	int base;
639 	int i, j, k;
640 
641 	if (!tt)
642 		return 0;
643 
644 	for (i = 0; i < sch_ep->num_esit; i++) {
645 		base = offset + i * sch_ep->esit;
646 
647 		if (sch_ep->ep_type == ISOC_OUT_EP) {
648 			for (j = 0; j < sch_ep->num_budget_microframes; j++) {
649 				k = XHCI_MTK_BW_INDEX(base + j);
650 				if (tt->in_ss_cnt[k])
651 					return -ESCH_SS_OVERLAP;
652 			}
653 		} else if (sch_ep->ep_type == ISOC_IN_EP || sch_ep->ep_type == INT_IN_EP) {
654 			k = XHCI_MTK_BW_INDEX(base);
655 			/* only check IN's ss */
656 			if (tt->fs_bus_bw_out[k])
657 				return -ESCH_SS_OVERLAP;
658 		}
659 	}
660 
661 	return 0;
662 }
663 
664 static int check_sch_tt_budget(struct mu3h_sch_ep_info *sch_ep, u32 offset)
665 {
666 	int ret;
667 
668 	ret = check_ss_and_cs(sch_ep, offset);
669 	if (ret)
670 		return ret;
671 
672 	ret = check_isoc_ss_overlap(sch_ep, offset);
673 	if (ret)
674 		return ret;
675 
676 	return check_fs_bus_bw(sch_ep, offset);
677 }
678 
679 /* allocate microframes in the ls/fs frame */
680 static int alloc_sch_portion_of_frame(struct mu3h_sch_ep_info *sch_ep)
681 {
682 	struct mu3h_sch_bw_info *sch_bw = sch_ep->bw_info;
683 	const u32 bw_boundary = get_bw_boundary(sch_ep->speed);
684 	u32 bw_max, fs_bw_min;
685 	u32 offset, offset_min;
686 	u16 fs_bw;
687 	int frames;
688 	int i, j;
689 	int ret;
690 
691 	frames = sch_ep->esit / UFRAMES_PER_FRAME;
692 
693 	for (i = 0; i < UFRAMES_PER_FRAME; i++) {
694 		fs_bw_min = FS_PAYLOAD_MAX;
695 		offset_min = XHCI_MTK_MAX_ESIT;
696 
697 		for (j = 0; j < frames; j++) {
698 			offset = (i + j * UFRAMES_PER_FRAME) % sch_ep->esit;
699 
700 			ret = check_sch_tt_budget(sch_ep, offset);
701 			if (ret)
702 				continue;
703 
704 			/* check hs bw domain */
705 			bw_max = get_max_bw(sch_bw, sch_ep, offset);
706 			if (bw_max > bw_boundary) {
707 				ret = -ESCH_BW_OVERFLOW;
708 				continue;
709 			}
710 
711 			/* use best-fit between frames */
712 			fs_bw = get_fs_bw(sch_ep, offset);
713 			if (fs_bw < fs_bw_min) {
714 				fs_bw_min = fs_bw;
715 				offset_min = offset;
716 			}
717 
718 			if (!fs_bw_min)
719 				break;
720 		}
721 
722 		/* use first-fit between microframes in a frame */
723 		if (offset_min < XHCI_MTK_MAX_ESIT)
724 			break;
725 	}
726 
727 	if (offset_min == XHCI_MTK_MAX_ESIT)
728 		return -ESCH_BW_OVERFLOW;
729 
730 	sch_ep->offset = offset_min;
731 
732 	return 0;
733 }
734 
735 static void update_sch_tt(struct mu3h_sch_ep_info *sch_ep, bool used)
736 {
737 	struct mu3h_sch_tt *tt = sch_ep->sch_tt;
738 	u16 *fs_bus_bw;
739 	u32 base;
740 	int i, j, k, f;
741 
742 	if (sch_ep->ep_type == ISOC_OUT_EP || sch_ep->ep_type == INT_OUT_EP)
743 		fs_bus_bw = tt->fs_bus_bw_out;
744 	else
745 		fs_bus_bw = tt->fs_bus_bw_in;
746 
747 	for (i = 0; i < sch_ep->num_esit; i++) {
748 		base = sch_ep->offset + i * sch_ep->esit;
749 
750 		for (j = 0; j < sch_ep->num_budget_microframes; j++) {
751 			k = XHCI_MTK_BW_INDEX(base + j);
752 			f = k / UFRAMES_PER_FRAME;
753 			if (used) {
754 				if (sch_ep->speed == USB_SPEED_LOW)
755 					tt->ls_bus_bw[k] += (u8)sch_ep->bw_budget_table[j];
756 
757 				fs_bus_bw[k] += (u16)sch_ep->bw_budget_table[j];
758 				tt->fs_frame_bw[f] += (u16)sch_ep->bw_budget_table[j];
759 			} else {
760 				if (sch_ep->speed == USB_SPEED_LOW)
761 					tt->ls_bus_bw[k] -= (u8)sch_ep->bw_budget_table[j];
762 
763 				fs_bus_bw[k] -= (u16)sch_ep->bw_budget_table[j];
764 				tt->fs_frame_bw[f] -= (u16)sch_ep->bw_budget_table[j];
765 			}
766 		}
767 
768 		if (sch_ep->ep_type == ISOC_IN_EP || sch_ep->ep_type == INT_IN_EP) {
769 			k = XHCI_MTK_BW_INDEX(base);
770 			if (used)
771 				tt->in_ss_cnt[k]++;
772 			else
773 				tt->in_ss_cnt[k]--;
774 		}
775 	}
776 
777 	if (used)
778 		list_add_tail(&sch_ep->tt_endpoint, &tt->ep_list);
779 	else
780 		list_del(&sch_ep->tt_endpoint);
781 }
782 
783 static int load_ep_bw(struct mu3h_sch_bw_info *sch_bw,
784 		      struct mu3h_sch_ep_info *sch_ep, bool loaded)
785 {
786 	if (sch_ep->sch_tt)
787 		update_sch_tt(sch_ep, loaded);
788 
789 	/* update bus bandwidth info */
790 	update_bus_bw(sch_bw, sch_ep, loaded);
791 	sch_ep->allocated = loaded;
792 
793 	return 0;
794 }
795 
796 /* allocate microframes for hs/ss/ssp */
797 static int alloc_sch_microframes(struct mu3h_sch_ep_info *sch_ep)
798 {
799 	struct mu3h_sch_bw_info *sch_bw = sch_ep->bw_info;
800 	const u32 bw_boundary = get_bw_boundary(sch_ep->speed);
801 	u32 offset;
802 	u32 worst_bw;
803 	u32 min_bw = ~0;
804 	int min_index = -1;
805 
806 	/*
807 	 * Search through all possible schedule microframes.
808 	 * and find a microframe where its worst bandwidth is minimum.
809 	 */
810 	for (offset = 0; offset < sch_ep->esit; offset++) {
811 
812 		worst_bw = get_max_bw(sch_bw, sch_ep, offset);
813 		if (worst_bw > bw_boundary)
814 			continue;
815 
816 		if (min_bw > worst_bw) {
817 			min_bw = worst_bw;
818 			min_index = offset;
819 		}
820 	}
821 
822 	if (min_index < 0)
823 		return -ESCH_BW_OVERFLOW;
824 
825 	sch_ep->offset = min_index;
826 
827 	return 0;
828 }
829 
830 static int check_sch_bw(struct mu3h_sch_ep_info *sch_ep)
831 {
832 	int ret;
833 
834 	if (sch_ep->sch_tt)
835 		ret = alloc_sch_portion_of_frame(sch_ep);
836 	else
837 		ret = alloc_sch_microframes(sch_ep);
838 
839 	if (ret)
840 		return ret;
841 
842 	return load_ep_bw(sch_ep->bw_info, sch_ep, true);
843 }
844 
845 static void destroy_sch_ep(struct xhci_hcd_mtk *mtk, struct usb_device *udev,
846 			   struct mu3h_sch_ep_info *sch_ep)
847 {
848 	/* only release ep bw check passed by check_sch_bw() */
849 	if (sch_ep->allocated)
850 		load_ep_bw(sch_ep->bw_info, sch_ep, false);
851 
852 	if (sch_ep->sch_tt)
853 		drop_tt(udev);
854 
855 	list_del(&sch_ep->endpoint);
856 	hlist_del(&sch_ep->hentry);
857 	kfree(sch_ep);
858 }
859 
860 static bool need_bw_sch(struct usb_device *udev,
861 			struct usb_host_endpoint *ep)
862 {
863 	bool has_tt = udev->tt && udev->tt->hub->parent;
864 
865 	/* only for periodic endpoints */
866 	if (usb_endpoint_xfer_control(&ep->desc)
867 		|| usb_endpoint_xfer_bulk(&ep->desc))
868 		return false;
869 
870 	/*
871 	 * for LS & FS periodic endpoints which its device is not behind
872 	 * a TT are also ignored, root-hub will schedule them directly,
873 	 * but need set @bpkts field of endpoint context to 1.
874 	 */
875 	if (is_fs_or_ls(udev->speed) && !has_tt)
876 		return false;
877 
878 	/* skip endpoint with zero maxpkt */
879 	if (usb_endpoint_maxp(&ep->desc) == 0)
880 		return false;
881 
882 	return true;
883 }
884 
885 int xhci_mtk_sch_init(struct xhci_hcd_mtk *mtk)
886 {
887 	struct xhci_hcd *xhci = hcd_to_xhci(mtk->hcd);
888 	struct mu3h_sch_bw_info *sch_array;
889 	int num_usb_bus;
890 
891 	/* ss IN and OUT are separated */
892 	num_usb_bus = xhci->usb3_rhub.num_ports * 2 + xhci->usb2_rhub.num_ports;
893 
894 	sch_array = kcalloc(num_usb_bus, sizeof(*sch_array), GFP_KERNEL);
895 	if (sch_array == NULL)
896 		return -ENOMEM;
897 
898 	mtk->sch_array = sch_array;
899 
900 	INIT_LIST_HEAD(&mtk->bw_ep_chk_list);
901 	hash_init(mtk->sch_ep_hash);
902 
903 	return 0;
904 }
905 
906 void xhci_mtk_sch_exit(struct xhci_hcd_mtk *mtk)
907 {
908 	kfree(mtk->sch_array);
909 }
910 
911 static int add_ep_quirk(struct usb_hcd *hcd, struct usb_device *udev,
912 			struct usb_host_endpoint *ep)
913 {
914 	struct xhci_hcd_mtk *mtk = hcd_to_mtk(hcd);
915 	struct xhci_hcd *xhci = hcd_to_xhci(hcd);
916 	struct xhci_ep_ctx *ep_ctx;
917 	struct xhci_virt_device *virt_dev;
918 	struct mu3h_sch_ep_info *sch_ep;
919 	unsigned int ep_index;
920 
921 	virt_dev = xhci->devs[udev->slot_id];
922 	ep_index = xhci_get_endpoint_index(&ep->desc);
923 	ep_ctx = xhci_get_ep_ctx(xhci, virt_dev->in_ctx, ep_index);
924 
925 	if (!need_bw_sch(udev, ep)) {
926 		/*
927 		 * set @bpkts to 1 if it is LS or FS periodic endpoint, and its
928 		 * device does not connected through an external HS hub
929 		 */
930 		if (usb_endpoint_xfer_int(&ep->desc)
931 			|| usb_endpoint_xfer_isoc(&ep->desc))
932 			ep_ctx->reserved[0] = cpu_to_le32(EP_BPKTS(1));
933 
934 		return 0;
935 	}
936 
937 	xhci_dbg(xhci, "%s %s\n", __func__, decode_ep(ep, udev->speed));
938 
939 	sch_ep = create_sch_ep(mtk, udev, ep, ep_ctx);
940 	if (IS_ERR_OR_NULL(sch_ep))
941 		return -ENOMEM;
942 
943 	setup_sch_info(ep_ctx, sch_ep);
944 
945 	list_add_tail(&sch_ep->endpoint, &mtk->bw_ep_chk_list);
946 	hash_add(mtk->sch_ep_hash, &sch_ep->hentry, (unsigned long)ep);
947 
948 	return 0;
949 }
950 
951 static void drop_ep_quirk(struct usb_hcd *hcd, struct usb_device *udev,
952 			  struct usb_host_endpoint *ep)
953 {
954 	struct xhci_hcd_mtk *mtk = hcd_to_mtk(hcd);
955 	struct xhci_hcd *xhci = hcd_to_xhci(hcd);
956 	struct mu3h_sch_ep_info *sch_ep;
957 	struct hlist_node *hn;
958 
959 	if (!need_bw_sch(udev, ep))
960 		return;
961 
962 	xhci_dbg(xhci, "%s %s\n", __func__, decode_ep(ep, udev->speed));
963 
964 	hash_for_each_possible_safe(mtk->sch_ep_hash, sch_ep,
965 				    hn, hentry, (unsigned long)ep) {
966 		if (sch_ep->ep == ep) {
967 			destroy_sch_ep(mtk, udev, sch_ep);
968 			break;
969 		}
970 	}
971 }
972 
973 int xhci_mtk_check_bandwidth(struct usb_hcd *hcd, struct usb_device *udev)
974 {
975 	struct xhci_hcd_mtk *mtk = hcd_to_mtk(hcd);
976 	struct xhci_hcd *xhci = hcd_to_xhci(hcd);
977 	struct xhci_virt_device *virt_dev = xhci->devs[udev->slot_id];
978 	struct mu3h_sch_ep_info *sch_ep;
979 	int ret;
980 
981 	xhci_dbg(xhci, "%s() udev %s\n", __func__, dev_name(&udev->dev));
982 
983 	list_for_each_entry(sch_ep, &mtk->bw_ep_chk_list, endpoint) {
984 		struct xhci_ep_ctx *ep_ctx;
985 		struct usb_host_endpoint *ep = sch_ep->ep;
986 		unsigned int ep_index = xhci_get_endpoint_index(&ep->desc);
987 
988 		ret = check_sch_bw(sch_ep);
989 		if (ret) {
990 			xhci_err(xhci, "Not enough bandwidth! (%s)\n",
991 				 sch_error_string(-ret));
992 			return -ENOSPC;
993 		}
994 
995 		ep_ctx = xhci_get_ep_ctx(xhci, virt_dev->in_ctx, ep_index);
996 		ep_ctx->reserved[0] = cpu_to_le32(EP_BPKTS(sch_ep->pkts)
997 			| EP_BCSCOUNT(sch_ep->cs_count)
998 			| EP_BBM(sch_ep->burst_mode));
999 		ep_ctx->reserved[1] = cpu_to_le32(EP_BOFFSET(sch_ep->offset)
1000 			| EP_BREPEAT(sch_ep->repeat));
1001 
1002 		xhci_dbg(xhci, " PKTS:%x, CSCOUNT:%x, BM:%x, OFFSET:%x, REPEAT:%x\n",
1003 			sch_ep->pkts, sch_ep->cs_count, sch_ep->burst_mode,
1004 			sch_ep->offset, sch_ep->repeat);
1005 	}
1006 
1007 	ret = xhci_check_bandwidth(hcd, udev);
1008 	if (!ret)
1009 		list_del_init(&mtk->bw_ep_chk_list);
1010 
1011 	return ret;
1012 }
1013 
1014 void xhci_mtk_reset_bandwidth(struct usb_hcd *hcd, struct usb_device *udev)
1015 {
1016 	struct xhci_hcd_mtk *mtk = hcd_to_mtk(hcd);
1017 	struct xhci_hcd *xhci = hcd_to_xhci(hcd);
1018 	struct mu3h_sch_ep_info *sch_ep, *tmp;
1019 
1020 	xhci_dbg(xhci, "%s() udev %s\n", __func__, dev_name(&udev->dev));
1021 
1022 	list_for_each_entry_safe(sch_ep, tmp, &mtk->bw_ep_chk_list, endpoint)
1023 		destroy_sch_ep(mtk, udev, sch_ep);
1024 
1025 	xhci_reset_bandwidth(hcd, udev);
1026 }
1027 
1028 int xhci_mtk_add_ep(struct usb_hcd *hcd, struct usb_device *udev,
1029 		    struct usb_host_endpoint *ep)
1030 {
1031 	int ret;
1032 
1033 	ret = xhci_add_endpoint(hcd, udev, ep);
1034 	if (ret)
1035 		return ret;
1036 
1037 	if (ep->hcpriv)
1038 		ret = add_ep_quirk(hcd, udev, ep);
1039 
1040 	return ret;
1041 }
1042 
1043 int xhci_mtk_drop_ep(struct usb_hcd *hcd, struct usb_device *udev,
1044 		     struct usb_host_endpoint *ep)
1045 {
1046 	int ret;
1047 
1048 	ret = xhci_drop_endpoint(hcd, udev, ep);
1049 	if (ret)
1050 		return ret;
1051 
1052 	/* needn't check @ep->hcpriv, xhci_endpoint_disable set it NULL */
1053 	drop_ep_quirk(hcd, udev, ep);
1054 
1055 	return 0;
1056 }
1057