1/* SPDX-License-Identifier: (GPL-2.0-only OR BSD-3-Clause) */
2/* QLogic qed NIC Driver
3 * Copyright (c) 2015-2017  QLogic Corporation
4 * Copyright (c) 2019-2020 Marvell International Ltd.
5 */
6
7#ifndef _QED_CHAIN_H
8#define _QED_CHAIN_H
9
10#include <linux/types.h>
11#include <asm/byteorder.h>
12#include <linux/kernel.h>
13#include <linux/list.h>
14#include <linux/sizes.h>
15#include <linux/slab.h>
16#include <linux/qed/common_hsi.h>
17
18enum qed_chain_mode {
19	/* Each Page contains a next pointer at its end */
20	QED_CHAIN_MODE_NEXT_PTR,
21
22	/* Chain is a single page (next ptr) is not required */
23	QED_CHAIN_MODE_SINGLE,
24
25	/* Page pointers are located in a side list */
26	QED_CHAIN_MODE_PBL,
27};
28
29enum qed_chain_use_mode {
30	QED_CHAIN_USE_TO_PRODUCE,			/* Chain starts empty */
31	QED_CHAIN_USE_TO_CONSUME,			/* Chain starts full */
32	QED_CHAIN_USE_TO_CONSUME_PRODUCE,		/* Chain starts empty */
33};
34
35enum qed_chain_cnt_type {
36	/* The chain's size/prod/cons are kept in 16-bit variables */
37	QED_CHAIN_CNT_TYPE_U16,
38
39	/* The chain's size/prod/cons are kept in 32-bit variables  */
40	QED_CHAIN_CNT_TYPE_U32,
41};
42
43struct qed_chain_next {
44	struct regpair					next_phys;
45	void						*next_virt;
46};
47
48struct qed_chain_pbl_u16 {
49	u16						prod_page_idx;
50	u16						cons_page_idx;
51};
52
53struct qed_chain_pbl_u32 {
54	u32						prod_page_idx;
55	u32						cons_page_idx;
56};
57
58struct qed_chain_u16 {
59	/* Cyclic index of next element to produce/consume */
60	u16						prod_idx;
61	u16						cons_idx;
62};
63
64struct qed_chain_u32 {
65	/* Cyclic index of next element to produce/consume */
66	u32						prod_idx;
67	u32						cons_idx;
68};
69
70struct addr_tbl_entry {
71	void						*virt_addr;
72	dma_addr_t					dma_map;
73};
74
75struct qed_chain {
76	/* Fastpath portion of the chain - required for commands such
77	 * as produce / consume.
78	 */
79
80	/* Point to next element to produce/consume */
81	void						*p_prod_elem;
82	void						*p_cons_elem;
83
84	/* Fastpath portions of the PBL [if exists] */
85
86	struct {
87		/* Table for keeping the virtual and physical addresses of the
88		 * chain pages, respectively to the physical addresses
89		 * in the pbl table.
90		 */
91		struct addr_tbl_entry			*pp_addr_tbl;
92
93		union {
94			struct qed_chain_pbl_u16	u16;
95			struct qed_chain_pbl_u32	u32;
96		}					c;
97	}						pbl;
98
99	union {
100		struct qed_chain_u16			chain16;
101		struct qed_chain_u32			chain32;
102	}						u;
103
104	/* Capacity counts only usable elements */
105	u32						capacity;
106	u32						page_cnt;
107
108	enum qed_chain_mode				mode;
109
110	/* Elements information for fast calculations */
111	u16						elem_per_page;
112	u16						elem_per_page_mask;
113	u16						elem_size;
114	u16						next_page_mask;
115	u16						usable_per_page;
116	u8						elem_unusable;
117
118	enum qed_chain_cnt_type				cnt_type;
119
120	/* Slowpath of the chain - required for initialization and destruction,
121	 * but isn't involved in regular functionality.
122	 */
123
124	u32						page_size;
125
126	/* Base address of a pre-allocated buffer for pbl */
127	struct {
128		__le64					*table_virt;
129		dma_addr_t				table_phys;
130		size_t					table_size;
131	}						pbl_sp;
132
133	/* Address of first page of the chain - the address is required
134	 * for fastpath operation [consume/produce] but only for the SINGLE
135	 * flavour which isn't considered fastpath [== SPQ].
136	 */
137	void						*p_virt_addr;
138	dma_addr_t					p_phys_addr;
139
140	/* Total number of elements [for entire chain] */
141	u32						size;
142
143	enum qed_chain_use_mode				intended_use;
144
145	bool						b_external_pbl;
146};
147
148struct qed_chain_init_params {
149	enum qed_chain_mode				mode;
150	enum qed_chain_use_mode				intended_use;
151	enum qed_chain_cnt_type				cnt_type;
152
153	u32						page_size;
154	u32						num_elems;
155	size_t						elem_size;
156
157	void						*ext_pbl_virt;
158	dma_addr_t					ext_pbl_phys;
159};
160
161#define QED_CHAIN_PAGE_SIZE				SZ_4K
162
163#define ELEMS_PER_PAGE(elem_size, page_size)				     \
164	((page_size) / (elem_size))
165
166#define UNUSABLE_ELEMS_PER_PAGE(elem_size, mode)			     \
167	(((mode) == QED_CHAIN_MODE_NEXT_PTR) ?				     \
168	 (u8)(1 + ((sizeof(struct qed_chain_next) - 1) / (elem_size))) :     \
169	 0)
170
171#define USABLE_ELEMS_PER_PAGE(elem_size, page_size, mode)		     \
172	((u32)(ELEMS_PER_PAGE((elem_size), (page_size)) -		     \
173	       UNUSABLE_ELEMS_PER_PAGE((elem_size), (mode))))
174
175#define QED_CHAIN_PAGE_CNT(elem_cnt, elem_size, page_size, mode)	     \
176	DIV_ROUND_UP((elem_cnt),					     \
177		     USABLE_ELEMS_PER_PAGE((elem_size), (page_size), (mode)))
178
179#define is_chain_u16(p)							     \
180	((p)->cnt_type == QED_CHAIN_CNT_TYPE_U16)
181#define is_chain_u32(p)							     \
182	((p)->cnt_type == QED_CHAIN_CNT_TYPE_U32)
183
184/* Accessors */
185
186static inline u16 qed_chain_get_prod_idx(const struct qed_chain *chain)
187{
188	return chain->u.chain16.prod_idx;
189}
190
191static inline u16 qed_chain_get_cons_idx(const struct qed_chain *chain)
192{
193	return chain->u.chain16.cons_idx;
194}
195
196static inline u32 qed_chain_get_prod_idx_u32(const struct qed_chain *chain)
197{
198	return chain->u.chain32.prod_idx;
199}
200
201static inline u32 qed_chain_get_cons_idx_u32(const struct qed_chain *chain)
202{
203	return chain->u.chain32.cons_idx;
204}
205
206static inline u16 qed_chain_get_elem_used(const struct qed_chain *chain)
207{
208	u32 prod = qed_chain_get_prod_idx(chain);
209	u32 cons = qed_chain_get_cons_idx(chain);
210	u16 elem_per_page = chain->elem_per_page;
211	u16 used;
212
213	if (prod < cons)
214		prod += (u32)U16_MAX + 1;
215
216	used = (u16)(prod - cons);
217	if (chain->mode == QED_CHAIN_MODE_NEXT_PTR)
218		used -= (u16)(prod / elem_per_page - cons / elem_per_page);
219
220	return used;
221}
222
223static inline u16 qed_chain_get_elem_left(const struct qed_chain *chain)
224{
225	return (u16)(chain->capacity - qed_chain_get_elem_used(chain));
226}
227
228static inline u32 qed_chain_get_elem_used_u32(const struct qed_chain *chain)
229{
230	u64 prod = qed_chain_get_prod_idx_u32(chain);
231	u64 cons = qed_chain_get_cons_idx_u32(chain);
232	u16 elem_per_page = chain->elem_per_page;
233	u32 used;
234
235	if (prod < cons)
236		prod += (u64)U32_MAX + 1;
237
238	used = (u32)(prod - cons);
239	if (chain->mode == QED_CHAIN_MODE_NEXT_PTR)
240		used -= (u32)(prod / elem_per_page - cons / elem_per_page);
241
242	return used;
243}
244
245static inline u32 qed_chain_get_elem_left_u32(const struct qed_chain *chain)
246{
247	return chain->capacity - qed_chain_get_elem_used_u32(chain);
248}
249
250static inline u16 qed_chain_get_usable_per_page(const struct qed_chain *chain)
251{
252	return chain->usable_per_page;
253}
254
255static inline u8 qed_chain_get_unusable_per_page(const struct qed_chain *chain)
256{
257	return chain->elem_unusable;
258}
259
260static inline u32 qed_chain_get_page_cnt(const struct qed_chain *chain)
261{
262	return chain->page_cnt;
263}
264
265static inline dma_addr_t qed_chain_get_pbl_phys(const struct qed_chain *chain)
266{
267	return chain->pbl_sp.table_phys;
268}
269
270/**
271 * qed_chain_advance_page(): Advance the next element across pages for a
272 *                           linked chain.
273 *
274 * @p_chain: P_chain.
275 * @p_next_elem: P_next_elem.
276 * @idx_to_inc: Idx_to_inc.
277 * @page_to_inc: page_to_inc.
278 *
279 * Return: Void.
280 */
281static inline void
282qed_chain_advance_page(struct qed_chain *p_chain,
283		       void **p_next_elem, void *idx_to_inc, void *page_to_inc)
284{
285	struct qed_chain_next *p_next = NULL;
286	u32 page_index = 0;
287
288	switch (p_chain->mode) {
289	case QED_CHAIN_MODE_NEXT_PTR:
290		p_next = *p_next_elem;
291		*p_next_elem = p_next->next_virt;
292		if (is_chain_u16(p_chain))
293			*(u16 *)idx_to_inc += p_chain->elem_unusable;
294		else
295			*(u32 *)idx_to_inc += p_chain->elem_unusable;
296		break;
297	case QED_CHAIN_MODE_SINGLE:
298		*p_next_elem = p_chain->p_virt_addr;
299		break;
300
301	case QED_CHAIN_MODE_PBL:
302		if (is_chain_u16(p_chain)) {
303			if (++(*(u16 *)page_to_inc) == p_chain->page_cnt)
304				*(u16 *)page_to_inc = 0;
305			page_index = *(u16 *)page_to_inc;
306		} else {
307			if (++(*(u32 *)page_to_inc) == p_chain->page_cnt)
308				*(u32 *)page_to_inc = 0;
309			page_index = *(u32 *)page_to_inc;
310		}
311		*p_next_elem = p_chain->pbl.pp_addr_tbl[page_index].virt_addr;
312	}
313}
314
315#define is_unusable_idx(p, idx)	\
316	(((p)->u.chain16.idx & (p)->elem_per_page_mask) == (p)->usable_per_page)
317
318#define is_unusable_idx_u32(p, idx) \
319	(((p)->u.chain32.idx & (p)->elem_per_page_mask) == (p)->usable_per_page)
320#define is_unusable_next_idx(p, idx)				 \
321	((((p)->u.chain16.idx + 1) & (p)->elem_per_page_mask) == \
322	 (p)->usable_per_page)
323
324#define is_unusable_next_idx_u32(p, idx)			 \
325	((((p)->u.chain32.idx + 1) & (p)->elem_per_page_mask) == \
326	 (p)->usable_per_page)
327
328#define test_and_skip(p, idx)						   \
329	do {						\
330		if (is_chain_u16(p)) {					   \
331			if (is_unusable_idx(p, idx))			   \
332				(p)->u.chain16.idx += (p)->elem_unusable;  \
333		} else {						   \
334			if (is_unusable_idx_u32(p, idx))		   \
335				(p)->u.chain32.idx += (p)->elem_unusable;  \
336		}					\
337	} while (0)
338
339/**
340 * qed_chain_return_produced(): A chain in which the driver "Produces"
341 *                              elements should use this API
342 *                              to indicate previous produced elements
343 *                              are now consumed.
344 *
345 * @p_chain: Chain.
346 *
347 * Return: Void.
348 */
349static inline void qed_chain_return_produced(struct qed_chain *p_chain)
350{
351	if (is_chain_u16(p_chain))
352		p_chain->u.chain16.cons_idx++;
353	else
354		p_chain->u.chain32.cons_idx++;
355	test_and_skip(p_chain, cons_idx);
356}
357
358/**
359 * qed_chain_produce(): A chain in which the driver "Produces"
360 *                      elements should use this to get a pointer to
361 *                      the next element which can be "Produced". It's driver
362 *                      responsibility to validate that the chain has room for
363 *                      new element.
364 *
365 * @p_chain: Chain.
366 *
367 * Return: void*, a pointer to next element.
368 */
369static inline void *qed_chain_produce(struct qed_chain *p_chain)
370{
371	void *p_ret = NULL, *p_prod_idx, *p_prod_page_idx;
372
373	if (is_chain_u16(p_chain)) {
374		if ((p_chain->u.chain16.prod_idx &
375		     p_chain->elem_per_page_mask) == p_chain->next_page_mask) {
376			p_prod_idx = &p_chain->u.chain16.prod_idx;
377			p_prod_page_idx = &p_chain->pbl.c.u16.prod_page_idx;
378			qed_chain_advance_page(p_chain, &p_chain->p_prod_elem,
379					       p_prod_idx, p_prod_page_idx);
380		}
381		p_chain->u.chain16.prod_idx++;
382	} else {
383		if ((p_chain->u.chain32.prod_idx &
384		     p_chain->elem_per_page_mask) == p_chain->next_page_mask) {
385			p_prod_idx = &p_chain->u.chain32.prod_idx;
386			p_prod_page_idx = &p_chain->pbl.c.u32.prod_page_idx;
387			qed_chain_advance_page(p_chain, &p_chain->p_prod_elem,
388					       p_prod_idx, p_prod_page_idx);
389		}
390		p_chain->u.chain32.prod_idx++;
391	}
392
393	p_ret = p_chain->p_prod_elem;
394	p_chain->p_prod_elem = (void *)(((u8 *)p_chain->p_prod_elem) +
395					p_chain->elem_size);
396
397	return p_ret;
398}
399
400/**
401 * qed_chain_get_capacity(): Get the maximum number of BDs in chain
402 *
403 * @p_chain: Chain.
404 *
405 * Return: number of unusable BDs.
406 */
407static inline u32 qed_chain_get_capacity(struct qed_chain *p_chain)
408{
409	return p_chain->capacity;
410}
411
412/**
413 * qed_chain_recycle_consumed(): Returns an element which was
414 *                               previously consumed;
415 *                               Increments producers so they could
416 *                               be written to FW.
417 *
418 * @p_chain: Chain.
419 *
420 * Return: Void.
421 */
422static inline void qed_chain_recycle_consumed(struct qed_chain *p_chain)
423{
424	test_and_skip(p_chain, prod_idx);
425	if (is_chain_u16(p_chain))
426		p_chain->u.chain16.prod_idx++;
427	else
428		p_chain->u.chain32.prod_idx++;
429}
430
431/**
432 * qed_chain_consume(): A Chain in which the driver utilizes data written
433 *                      by a different source (i.e., FW) should use this to
434 *                      access passed buffers.
435 *
436 * @p_chain: Chain.
437 *
438 * Return: void*, a pointer to the next buffer written.
439 */
440static inline void *qed_chain_consume(struct qed_chain *p_chain)
441{
442	void *p_ret = NULL, *p_cons_idx, *p_cons_page_idx;
443
444	if (is_chain_u16(p_chain)) {
445		if ((p_chain->u.chain16.cons_idx &
446		     p_chain->elem_per_page_mask) == p_chain->next_page_mask) {
447			p_cons_idx = &p_chain->u.chain16.cons_idx;
448			p_cons_page_idx = &p_chain->pbl.c.u16.cons_page_idx;
449			qed_chain_advance_page(p_chain, &p_chain->p_cons_elem,
450					       p_cons_idx, p_cons_page_idx);
451		}
452		p_chain->u.chain16.cons_idx++;
453	} else {
454		if ((p_chain->u.chain32.cons_idx &
455		     p_chain->elem_per_page_mask) == p_chain->next_page_mask) {
456			p_cons_idx = &p_chain->u.chain32.cons_idx;
457			p_cons_page_idx = &p_chain->pbl.c.u32.cons_page_idx;
458			qed_chain_advance_page(p_chain, &p_chain->p_cons_elem,
459					       p_cons_idx, p_cons_page_idx);
460		}
461		p_chain->u.chain32.cons_idx++;
462	}
463
464	p_ret = p_chain->p_cons_elem;
465	p_chain->p_cons_elem = (void *)(((u8 *)p_chain->p_cons_elem) +
466					p_chain->elem_size);
467
468	return p_ret;
469}
470
471/**
472 * qed_chain_reset(): Resets the chain to its start state.
473 *
474 * @p_chain: pointer to a previously allocated chain.
475 *
476 * Return Void.
477 */
478static inline void qed_chain_reset(struct qed_chain *p_chain)
479{
480	u32 i;
481
482	if (is_chain_u16(p_chain)) {
483		p_chain->u.chain16.prod_idx = 0;
484		p_chain->u.chain16.cons_idx = 0;
485	} else {
486		p_chain->u.chain32.prod_idx = 0;
487		p_chain->u.chain32.cons_idx = 0;
488	}
489	p_chain->p_cons_elem = p_chain->p_virt_addr;
490	p_chain->p_prod_elem = p_chain->p_virt_addr;
491
492	if (p_chain->mode == QED_CHAIN_MODE_PBL) {
493		/* Use (page_cnt - 1) as a reset value for the prod/cons page's
494		 * indices, to avoid unnecessary page advancing on the first
495		 * call to qed_chain_produce/consume. Instead, the indices
496		 * will be advanced to page_cnt and then will be wrapped to 0.
497		 */
498		u32 reset_val = p_chain->page_cnt - 1;
499
500		if (is_chain_u16(p_chain)) {
501			p_chain->pbl.c.u16.prod_page_idx = (u16)reset_val;
502			p_chain->pbl.c.u16.cons_page_idx = (u16)reset_val;
503		} else {
504			p_chain->pbl.c.u32.prod_page_idx = reset_val;
505			p_chain->pbl.c.u32.cons_page_idx = reset_val;
506		}
507	}
508
509	switch (p_chain->intended_use) {
510	case QED_CHAIN_USE_TO_CONSUME:
511		/* produce empty elements */
512		for (i = 0; i < p_chain->capacity; i++)
513			qed_chain_recycle_consumed(p_chain);
514		break;
515
516	case QED_CHAIN_USE_TO_CONSUME_PRODUCE:
517	case QED_CHAIN_USE_TO_PRODUCE:
518	default:
519		/* Do nothing */
520		break;
521	}
522}
523
524/**
525 * qed_chain_get_last_elem(): Returns a pointer to the last element of the
526 *                            chain.
527 *
528 * @p_chain: Chain.
529 *
530 * Return: void*.
531 */
532static inline void *qed_chain_get_last_elem(struct qed_chain *p_chain)
533{
534	struct qed_chain_next *p_next = NULL;
535	void *p_virt_addr = NULL;
536	u32 size, last_page_idx;
537
538	if (!p_chain->p_virt_addr)
539		goto out;
540
541	switch (p_chain->mode) {
542	case QED_CHAIN_MODE_NEXT_PTR:
543		size = p_chain->elem_size * p_chain->usable_per_page;
544		p_virt_addr = p_chain->p_virt_addr;
545		p_next = (struct qed_chain_next *)((u8 *)p_virt_addr + size);
546		while (p_next->next_virt != p_chain->p_virt_addr) {
547			p_virt_addr = p_next->next_virt;
548			p_next = (struct qed_chain_next *)((u8 *)p_virt_addr +
549							   size);
550		}
551		break;
552	case QED_CHAIN_MODE_SINGLE:
553		p_virt_addr = p_chain->p_virt_addr;
554		break;
555	case QED_CHAIN_MODE_PBL:
556		last_page_idx = p_chain->page_cnt - 1;
557		p_virt_addr = p_chain->pbl.pp_addr_tbl[last_page_idx].virt_addr;
558		break;
559	}
560	/* p_virt_addr points at this stage to the last page of the chain */
561	size = p_chain->elem_size * (p_chain->usable_per_page - 1);
562	p_virt_addr = (u8 *)p_virt_addr + size;
563out:
564	return p_virt_addr;
565}
566
567/**
568 * qed_chain_set_prod(): sets the prod to the given value.
569 *
570 * @p_chain: Chain.
571 * @prod_idx: Prod Idx.
572 * @p_prod_elem: Prod elem.
573 *
574 * Return Void.
575 */
576static inline void qed_chain_set_prod(struct qed_chain *p_chain,
577				      u32 prod_idx, void *p_prod_elem)
578{
579	if (p_chain->mode == QED_CHAIN_MODE_PBL) {
580		u32 cur_prod, page_mask, page_cnt, page_diff;
581
582		cur_prod = is_chain_u16(p_chain) ? p_chain->u.chain16.prod_idx :
583			   p_chain->u.chain32.prod_idx;
584
585		/* Assume that number of elements in a page is power of 2 */
586		page_mask = ~p_chain->elem_per_page_mask;
587
588		/* Use "cur_prod - 1" and "prod_idx - 1" since producer index
589		 * reaches the first element of next page before the page index
590		 * is incremented. See qed_chain_produce().
591		 * Index wrap around is not a problem because the difference
592		 * between current and given producer indices is always
593		 * positive and lower than the chain's capacity.
594		 */
595		page_diff = (((cur_prod - 1) & page_mask) -
596			     ((prod_idx - 1) & page_mask)) /
597			    p_chain->elem_per_page;
598
599		page_cnt = qed_chain_get_page_cnt(p_chain);
600		if (is_chain_u16(p_chain))
601			p_chain->pbl.c.u16.prod_page_idx =
602				(p_chain->pbl.c.u16.prod_page_idx -
603				 page_diff + page_cnt) % page_cnt;
604		else
605			p_chain->pbl.c.u32.prod_page_idx =
606				(p_chain->pbl.c.u32.prod_page_idx -
607				 page_diff + page_cnt) % page_cnt;
608	}
609
610	if (is_chain_u16(p_chain))
611		p_chain->u.chain16.prod_idx = (u16) prod_idx;
612	else
613		p_chain->u.chain32.prod_idx = prod_idx;
614	p_chain->p_prod_elem = p_prod_elem;
615}
616
617/**
618 * qed_chain_pbl_zero_mem(): set chain memory to 0.
619 *
620 * @p_chain: Chain.
621 *
622 * Return: Void.
623 */
624static inline void qed_chain_pbl_zero_mem(struct qed_chain *p_chain)
625{
626	u32 i, page_cnt;
627
628	if (p_chain->mode != QED_CHAIN_MODE_PBL)
629		return;
630
631	page_cnt = qed_chain_get_page_cnt(p_chain);
632
633	for (i = 0; i < page_cnt; i++)
634		memset(p_chain->pbl.pp_addr_tbl[i].virt_addr, 0,
635		       p_chain->page_size);
636}
637
638#endif
639