1 // SPDX-License-Identifier: BSD-3-Clause OR GPL-2.0
2 /* Copyright (c) 2015-2018 Mellanox Technologies. All rights reserved */
3 
4 #include <linux/kernel.h>
5 #include <linux/module.h>
6 #include <linux/export.h>
7 #include <linux/err.h>
8 #include <linux/device.h>
9 #include <linux/pci.h>
10 #include <linux/interrupt.h>
11 #include <linux/wait.h>
12 #include <linux/types.h>
13 #include <linux/skbuff.h>
14 #include <linux/if_vlan.h>
15 #include <linux/log2.h>
16 #include <linux/string.h>
17 
18 #include "pci_hw.h"
19 #include "pci.h"
20 #include "core.h"
21 #include "cmd.h"
22 #include "port.h"
23 #include "resources.h"
24 
25 #define mlxsw_pci_write32(mlxsw_pci, reg, val) \
26 	iowrite32be(val, (mlxsw_pci)->hw_addr + (MLXSW_PCI_ ## reg))
27 #define mlxsw_pci_read32(mlxsw_pci, reg) \
28 	ioread32be((mlxsw_pci)->hw_addr + (MLXSW_PCI_ ## reg))
29 
30 enum mlxsw_pci_queue_type {
31 	MLXSW_PCI_QUEUE_TYPE_SDQ,
32 	MLXSW_PCI_QUEUE_TYPE_RDQ,
33 	MLXSW_PCI_QUEUE_TYPE_CQ,
34 	MLXSW_PCI_QUEUE_TYPE_EQ,
35 };
36 
37 #define MLXSW_PCI_QUEUE_TYPE_COUNT	4
38 
39 static const u16 mlxsw_pci_doorbell_type_offset[] = {
40 	MLXSW_PCI_DOORBELL_SDQ_OFFSET,	/* for type MLXSW_PCI_QUEUE_TYPE_SDQ */
41 	MLXSW_PCI_DOORBELL_RDQ_OFFSET,	/* for type MLXSW_PCI_QUEUE_TYPE_RDQ */
42 	MLXSW_PCI_DOORBELL_CQ_OFFSET,	/* for type MLXSW_PCI_QUEUE_TYPE_CQ */
43 	MLXSW_PCI_DOORBELL_EQ_OFFSET,	/* for type MLXSW_PCI_QUEUE_TYPE_EQ */
44 };
45 
46 static const u16 mlxsw_pci_doorbell_arm_type_offset[] = {
47 	0, /* unused */
48 	0, /* unused */
49 	MLXSW_PCI_DOORBELL_ARM_CQ_OFFSET, /* for type MLXSW_PCI_QUEUE_TYPE_CQ */
50 	MLXSW_PCI_DOORBELL_ARM_EQ_OFFSET, /* for type MLXSW_PCI_QUEUE_TYPE_EQ */
51 };
52 
53 struct mlxsw_pci_mem_item {
54 	char *buf;
55 	dma_addr_t mapaddr;
56 	size_t size;
57 };
58 
59 struct mlxsw_pci_queue_elem_info {
60 	char *elem; /* pointer to actual dma mapped element mem chunk */
61 	union {
62 		struct {
63 			struct sk_buff *skb;
64 		} sdq;
65 		struct {
66 			struct sk_buff *skb;
67 		} rdq;
68 	} u;
69 };
70 
71 struct mlxsw_pci_queue {
72 	spinlock_t lock; /* for queue accesses */
73 	struct mlxsw_pci_mem_item mem_item;
74 	struct mlxsw_pci_queue_elem_info *elem_info;
75 	u16 producer_counter;
76 	u16 consumer_counter;
77 	u16 count; /* number of elements in queue */
78 	u8 num; /* queue number */
79 	u8 elem_size; /* size of one element */
80 	enum mlxsw_pci_queue_type type;
81 	struct tasklet_struct tasklet; /* queue processing tasklet */
82 	struct mlxsw_pci *pci;
83 	union {
84 		struct {
85 			u32 comp_sdq_count;
86 			u32 comp_rdq_count;
87 			enum mlxsw_pci_cqe_v v;
88 		} cq;
89 		struct {
90 			u32 ev_cmd_count;
91 			u32 ev_comp_count;
92 			u32 ev_other_count;
93 		} eq;
94 	} u;
95 };
96 
97 struct mlxsw_pci_queue_type_group {
98 	struct mlxsw_pci_queue *q;
99 	u8 count; /* number of queues in group */
100 };
101 
102 struct mlxsw_pci {
103 	struct pci_dev *pdev;
104 	u8 __iomem *hw_addr;
105 	u64 free_running_clock_offset;
106 	struct mlxsw_pci_queue_type_group queues[MLXSW_PCI_QUEUE_TYPE_COUNT];
107 	u32 doorbell_offset;
108 	struct mlxsw_core *core;
109 	struct {
110 		struct mlxsw_pci_mem_item *items;
111 		unsigned int count;
112 	} fw_area;
113 	struct {
114 		struct mlxsw_pci_mem_item out_mbox;
115 		struct mlxsw_pci_mem_item in_mbox;
116 		struct mutex lock; /* Lock access to command registers */
117 		bool nopoll;
118 		wait_queue_head_t wait;
119 		bool wait_done;
120 		struct {
121 			u8 status;
122 			u64 out_param;
123 		} comp;
124 	} cmd;
125 	struct mlxsw_bus_info bus_info;
126 	const struct pci_device_id *id;
127 	enum mlxsw_pci_cqe_v max_cqe_ver; /* Maximal supported CQE version */
128 	u8 num_sdq_cqs; /* Number of CQs used for SDQs */
129 };
130 
131 static void mlxsw_pci_queue_tasklet_schedule(struct mlxsw_pci_queue *q)
132 {
133 	tasklet_schedule(&q->tasklet);
134 }
135 
136 static char *__mlxsw_pci_queue_elem_get(struct mlxsw_pci_queue *q,
137 					size_t elem_size, int elem_index)
138 {
139 	return q->mem_item.buf + (elem_size * elem_index);
140 }
141 
142 static struct mlxsw_pci_queue_elem_info *
143 mlxsw_pci_queue_elem_info_get(struct mlxsw_pci_queue *q, int elem_index)
144 {
145 	return &q->elem_info[elem_index];
146 }
147 
148 static struct mlxsw_pci_queue_elem_info *
149 mlxsw_pci_queue_elem_info_producer_get(struct mlxsw_pci_queue *q)
150 {
151 	int index = q->producer_counter & (q->count - 1);
152 
153 	if ((u16) (q->producer_counter - q->consumer_counter) == q->count)
154 		return NULL;
155 	return mlxsw_pci_queue_elem_info_get(q, index);
156 }
157 
158 static struct mlxsw_pci_queue_elem_info *
159 mlxsw_pci_queue_elem_info_consumer_get(struct mlxsw_pci_queue *q)
160 {
161 	int index = q->consumer_counter & (q->count - 1);
162 
163 	return mlxsw_pci_queue_elem_info_get(q, index);
164 }
165 
166 static char *mlxsw_pci_queue_elem_get(struct mlxsw_pci_queue *q, int elem_index)
167 {
168 	return mlxsw_pci_queue_elem_info_get(q, elem_index)->elem;
169 }
170 
171 static bool mlxsw_pci_elem_hw_owned(struct mlxsw_pci_queue *q, bool owner_bit)
172 {
173 	return owner_bit != !!(q->consumer_counter & q->count);
174 }
175 
176 static struct mlxsw_pci_queue_type_group *
177 mlxsw_pci_queue_type_group_get(struct mlxsw_pci *mlxsw_pci,
178 			       enum mlxsw_pci_queue_type q_type)
179 {
180 	return &mlxsw_pci->queues[q_type];
181 }
182 
183 static u8 __mlxsw_pci_queue_count(struct mlxsw_pci *mlxsw_pci,
184 				  enum mlxsw_pci_queue_type q_type)
185 {
186 	struct mlxsw_pci_queue_type_group *queue_group;
187 
188 	queue_group = mlxsw_pci_queue_type_group_get(mlxsw_pci, q_type);
189 	return queue_group->count;
190 }
191 
192 static u8 mlxsw_pci_sdq_count(struct mlxsw_pci *mlxsw_pci)
193 {
194 	return __mlxsw_pci_queue_count(mlxsw_pci, MLXSW_PCI_QUEUE_TYPE_SDQ);
195 }
196 
197 static u8 mlxsw_pci_cq_count(struct mlxsw_pci *mlxsw_pci)
198 {
199 	return __mlxsw_pci_queue_count(mlxsw_pci, MLXSW_PCI_QUEUE_TYPE_CQ);
200 }
201 
202 static struct mlxsw_pci_queue *
203 __mlxsw_pci_queue_get(struct mlxsw_pci *mlxsw_pci,
204 		      enum mlxsw_pci_queue_type q_type, u8 q_num)
205 {
206 	return &mlxsw_pci->queues[q_type].q[q_num];
207 }
208 
209 static struct mlxsw_pci_queue *mlxsw_pci_sdq_get(struct mlxsw_pci *mlxsw_pci,
210 						 u8 q_num)
211 {
212 	return __mlxsw_pci_queue_get(mlxsw_pci,
213 				     MLXSW_PCI_QUEUE_TYPE_SDQ, q_num);
214 }
215 
216 static struct mlxsw_pci_queue *mlxsw_pci_rdq_get(struct mlxsw_pci *mlxsw_pci,
217 						 u8 q_num)
218 {
219 	return __mlxsw_pci_queue_get(mlxsw_pci,
220 				     MLXSW_PCI_QUEUE_TYPE_RDQ, q_num);
221 }
222 
223 static struct mlxsw_pci_queue *mlxsw_pci_cq_get(struct mlxsw_pci *mlxsw_pci,
224 						u8 q_num)
225 {
226 	return __mlxsw_pci_queue_get(mlxsw_pci, MLXSW_PCI_QUEUE_TYPE_CQ, q_num);
227 }
228 
229 static struct mlxsw_pci_queue *mlxsw_pci_eq_get(struct mlxsw_pci *mlxsw_pci,
230 						u8 q_num)
231 {
232 	return __mlxsw_pci_queue_get(mlxsw_pci, MLXSW_PCI_QUEUE_TYPE_EQ, q_num);
233 }
234 
235 static void __mlxsw_pci_queue_doorbell_set(struct mlxsw_pci *mlxsw_pci,
236 					   struct mlxsw_pci_queue *q,
237 					   u16 val)
238 {
239 	mlxsw_pci_write32(mlxsw_pci,
240 			  DOORBELL(mlxsw_pci->doorbell_offset,
241 				   mlxsw_pci_doorbell_type_offset[q->type],
242 				   q->num), val);
243 }
244 
245 static void __mlxsw_pci_queue_doorbell_arm_set(struct mlxsw_pci *mlxsw_pci,
246 					       struct mlxsw_pci_queue *q,
247 					       u16 val)
248 {
249 	mlxsw_pci_write32(mlxsw_pci,
250 			  DOORBELL(mlxsw_pci->doorbell_offset,
251 				   mlxsw_pci_doorbell_arm_type_offset[q->type],
252 				   q->num), val);
253 }
254 
255 static void mlxsw_pci_queue_doorbell_producer_ring(struct mlxsw_pci *mlxsw_pci,
256 						   struct mlxsw_pci_queue *q)
257 {
258 	wmb(); /* ensure all writes are done before we ring a bell */
259 	__mlxsw_pci_queue_doorbell_set(mlxsw_pci, q, q->producer_counter);
260 }
261 
262 static void mlxsw_pci_queue_doorbell_consumer_ring(struct mlxsw_pci *mlxsw_pci,
263 						   struct mlxsw_pci_queue *q)
264 {
265 	wmb(); /* ensure all writes are done before we ring a bell */
266 	__mlxsw_pci_queue_doorbell_set(mlxsw_pci, q,
267 				       q->consumer_counter + q->count);
268 }
269 
270 static void
271 mlxsw_pci_queue_doorbell_arm_consumer_ring(struct mlxsw_pci *mlxsw_pci,
272 					   struct mlxsw_pci_queue *q)
273 {
274 	wmb(); /* ensure all writes are done before we ring a bell */
275 	__mlxsw_pci_queue_doorbell_arm_set(mlxsw_pci, q, q->consumer_counter);
276 }
277 
278 static dma_addr_t __mlxsw_pci_queue_page_get(struct mlxsw_pci_queue *q,
279 					     int page_index)
280 {
281 	return q->mem_item.mapaddr + MLXSW_PCI_PAGE_SIZE * page_index;
282 }
283 
284 static int mlxsw_pci_sdq_init(struct mlxsw_pci *mlxsw_pci, char *mbox,
285 			      struct mlxsw_pci_queue *q)
286 {
287 	int tclass;
288 	int i;
289 	int err;
290 
291 	q->producer_counter = 0;
292 	q->consumer_counter = 0;
293 	tclass = q->num == MLXSW_PCI_SDQ_EMAD_INDEX ? MLXSW_PCI_SDQ_EMAD_TC :
294 						      MLXSW_PCI_SDQ_CTL_TC;
295 
296 	/* Set CQ of same number of this SDQ. */
297 	mlxsw_cmd_mbox_sw2hw_dq_cq_set(mbox, q->num);
298 	mlxsw_cmd_mbox_sw2hw_dq_sdq_tclass_set(mbox, tclass);
299 	mlxsw_cmd_mbox_sw2hw_dq_log2_dq_sz_set(mbox, 3); /* 8 pages */
300 	for (i = 0; i < MLXSW_PCI_AQ_PAGES; i++) {
301 		dma_addr_t mapaddr = __mlxsw_pci_queue_page_get(q, i);
302 
303 		mlxsw_cmd_mbox_sw2hw_dq_pa_set(mbox, i, mapaddr);
304 	}
305 
306 	err = mlxsw_cmd_sw2hw_sdq(mlxsw_pci->core, mbox, q->num);
307 	if (err)
308 		return err;
309 	mlxsw_pci_queue_doorbell_producer_ring(mlxsw_pci, q);
310 	return 0;
311 }
312 
313 static void mlxsw_pci_sdq_fini(struct mlxsw_pci *mlxsw_pci,
314 			       struct mlxsw_pci_queue *q)
315 {
316 	mlxsw_cmd_hw2sw_sdq(mlxsw_pci->core, q->num);
317 }
318 
319 static int mlxsw_pci_wqe_frag_map(struct mlxsw_pci *mlxsw_pci, char *wqe,
320 				  int index, char *frag_data, size_t frag_len,
321 				  int direction)
322 {
323 	struct pci_dev *pdev = mlxsw_pci->pdev;
324 	dma_addr_t mapaddr;
325 
326 	mapaddr = dma_map_single(&pdev->dev, frag_data, frag_len, direction);
327 	if (unlikely(dma_mapping_error(&pdev->dev, mapaddr))) {
328 		dev_err_ratelimited(&pdev->dev, "failed to dma map tx frag\n");
329 		return -EIO;
330 	}
331 	mlxsw_pci_wqe_address_set(wqe, index, mapaddr);
332 	mlxsw_pci_wqe_byte_count_set(wqe, index, frag_len);
333 	return 0;
334 }
335 
336 static void mlxsw_pci_wqe_frag_unmap(struct mlxsw_pci *mlxsw_pci, char *wqe,
337 				     int index, int direction)
338 {
339 	struct pci_dev *pdev = mlxsw_pci->pdev;
340 	size_t frag_len = mlxsw_pci_wqe_byte_count_get(wqe, index);
341 	dma_addr_t mapaddr = mlxsw_pci_wqe_address_get(wqe, index);
342 
343 	if (!frag_len)
344 		return;
345 	dma_unmap_single(&pdev->dev, mapaddr, frag_len, direction);
346 }
347 
348 static int mlxsw_pci_rdq_skb_alloc(struct mlxsw_pci *mlxsw_pci,
349 				   struct mlxsw_pci_queue_elem_info *elem_info)
350 {
351 	size_t buf_len = MLXSW_PORT_MAX_MTU;
352 	char *wqe = elem_info->elem;
353 	struct sk_buff *skb;
354 	int err;
355 
356 	skb = netdev_alloc_skb_ip_align(NULL, buf_len);
357 	if (!skb)
358 		return -ENOMEM;
359 
360 	err = mlxsw_pci_wqe_frag_map(mlxsw_pci, wqe, 0, skb->data,
361 				     buf_len, DMA_FROM_DEVICE);
362 	if (err)
363 		goto err_frag_map;
364 
365 	elem_info->u.rdq.skb = skb;
366 	return 0;
367 
368 err_frag_map:
369 	dev_kfree_skb_any(skb);
370 	return err;
371 }
372 
373 static void mlxsw_pci_rdq_skb_free(struct mlxsw_pci *mlxsw_pci,
374 				   struct mlxsw_pci_queue_elem_info *elem_info)
375 {
376 	struct sk_buff *skb;
377 	char *wqe;
378 
379 	skb = elem_info->u.rdq.skb;
380 	wqe = elem_info->elem;
381 
382 	mlxsw_pci_wqe_frag_unmap(mlxsw_pci, wqe, 0, DMA_FROM_DEVICE);
383 	dev_kfree_skb_any(skb);
384 }
385 
386 static int mlxsw_pci_rdq_init(struct mlxsw_pci *mlxsw_pci, char *mbox,
387 			      struct mlxsw_pci_queue *q)
388 {
389 	struct mlxsw_pci_queue_elem_info *elem_info;
390 	u8 sdq_count = mlxsw_pci_sdq_count(mlxsw_pci);
391 	int i;
392 	int err;
393 
394 	q->producer_counter = 0;
395 	q->consumer_counter = 0;
396 
397 	/* Set CQ of same number of this RDQ with base
398 	 * above SDQ count as the lower ones are assigned to SDQs.
399 	 */
400 	mlxsw_cmd_mbox_sw2hw_dq_cq_set(mbox, sdq_count + q->num);
401 	mlxsw_cmd_mbox_sw2hw_dq_log2_dq_sz_set(mbox, 3); /* 8 pages */
402 	for (i = 0; i < MLXSW_PCI_AQ_PAGES; i++) {
403 		dma_addr_t mapaddr = __mlxsw_pci_queue_page_get(q, i);
404 
405 		mlxsw_cmd_mbox_sw2hw_dq_pa_set(mbox, i, mapaddr);
406 	}
407 
408 	err = mlxsw_cmd_sw2hw_rdq(mlxsw_pci->core, mbox, q->num);
409 	if (err)
410 		return err;
411 
412 	mlxsw_pci_queue_doorbell_producer_ring(mlxsw_pci, q);
413 
414 	for (i = 0; i < q->count; i++) {
415 		elem_info = mlxsw_pci_queue_elem_info_producer_get(q);
416 		BUG_ON(!elem_info);
417 		err = mlxsw_pci_rdq_skb_alloc(mlxsw_pci, elem_info);
418 		if (err)
419 			goto rollback;
420 		/* Everything is set up, ring doorbell to pass elem to HW */
421 		q->producer_counter++;
422 		mlxsw_pci_queue_doorbell_producer_ring(mlxsw_pci, q);
423 	}
424 
425 	return 0;
426 
427 rollback:
428 	for (i--; i >= 0; i--) {
429 		elem_info = mlxsw_pci_queue_elem_info_get(q, i);
430 		mlxsw_pci_rdq_skb_free(mlxsw_pci, elem_info);
431 	}
432 	mlxsw_cmd_hw2sw_rdq(mlxsw_pci->core, q->num);
433 
434 	return err;
435 }
436 
437 static void mlxsw_pci_rdq_fini(struct mlxsw_pci *mlxsw_pci,
438 			       struct mlxsw_pci_queue *q)
439 {
440 	struct mlxsw_pci_queue_elem_info *elem_info;
441 	int i;
442 
443 	mlxsw_cmd_hw2sw_rdq(mlxsw_pci->core, q->num);
444 	for (i = 0; i < q->count; i++) {
445 		elem_info = mlxsw_pci_queue_elem_info_get(q, i);
446 		mlxsw_pci_rdq_skb_free(mlxsw_pci, elem_info);
447 	}
448 }
449 
450 static void mlxsw_pci_cq_pre_init(struct mlxsw_pci *mlxsw_pci,
451 				  struct mlxsw_pci_queue *q)
452 {
453 	q->u.cq.v = mlxsw_pci->max_cqe_ver;
454 
455 	/* For SDQ it is pointless to use CQEv2, so use CQEv1 instead */
456 	if (q->u.cq.v == MLXSW_PCI_CQE_V2 &&
457 	    q->num < mlxsw_pci->num_sdq_cqs)
458 		q->u.cq.v = MLXSW_PCI_CQE_V1;
459 }
460 
461 static int mlxsw_pci_cq_init(struct mlxsw_pci *mlxsw_pci, char *mbox,
462 			     struct mlxsw_pci_queue *q)
463 {
464 	int i;
465 	int err;
466 
467 	q->consumer_counter = 0;
468 
469 	for (i = 0; i < q->count; i++) {
470 		char *elem = mlxsw_pci_queue_elem_get(q, i);
471 
472 		mlxsw_pci_cqe_owner_set(q->u.cq.v, elem, 1);
473 	}
474 
475 	if (q->u.cq.v == MLXSW_PCI_CQE_V1)
476 		mlxsw_cmd_mbox_sw2hw_cq_cqe_ver_set(mbox,
477 				MLXSW_CMD_MBOX_SW2HW_CQ_CQE_VER_1);
478 	else if (q->u.cq.v == MLXSW_PCI_CQE_V2)
479 		mlxsw_cmd_mbox_sw2hw_cq_cqe_ver_set(mbox,
480 				MLXSW_CMD_MBOX_SW2HW_CQ_CQE_VER_2);
481 
482 	mlxsw_cmd_mbox_sw2hw_cq_c_eqn_set(mbox, MLXSW_PCI_EQ_COMP_NUM);
483 	mlxsw_cmd_mbox_sw2hw_cq_st_set(mbox, 0);
484 	mlxsw_cmd_mbox_sw2hw_cq_log_cq_size_set(mbox, ilog2(q->count));
485 	for (i = 0; i < MLXSW_PCI_AQ_PAGES; i++) {
486 		dma_addr_t mapaddr = __mlxsw_pci_queue_page_get(q, i);
487 
488 		mlxsw_cmd_mbox_sw2hw_cq_pa_set(mbox, i, mapaddr);
489 	}
490 	err = mlxsw_cmd_sw2hw_cq(mlxsw_pci->core, mbox, q->num);
491 	if (err)
492 		return err;
493 	mlxsw_pci_queue_doorbell_consumer_ring(mlxsw_pci, q);
494 	mlxsw_pci_queue_doorbell_arm_consumer_ring(mlxsw_pci, q);
495 	return 0;
496 }
497 
498 static void mlxsw_pci_cq_fini(struct mlxsw_pci *mlxsw_pci,
499 			      struct mlxsw_pci_queue *q)
500 {
501 	mlxsw_cmd_hw2sw_cq(mlxsw_pci->core, q->num);
502 }
503 
504 static void mlxsw_pci_cqe_sdq_handle(struct mlxsw_pci *mlxsw_pci,
505 				     struct mlxsw_pci_queue *q,
506 				     u16 consumer_counter_limit,
507 				     char *cqe)
508 {
509 	struct pci_dev *pdev = mlxsw_pci->pdev;
510 	struct mlxsw_pci_queue_elem_info *elem_info;
511 	struct mlxsw_tx_info tx_info;
512 	char *wqe;
513 	struct sk_buff *skb;
514 	int i;
515 
516 	spin_lock(&q->lock);
517 	elem_info = mlxsw_pci_queue_elem_info_consumer_get(q);
518 	tx_info = mlxsw_skb_cb(elem_info->u.sdq.skb)->tx_info;
519 	skb = elem_info->u.sdq.skb;
520 	wqe = elem_info->elem;
521 	for (i = 0; i < MLXSW_PCI_WQE_SG_ENTRIES; i++)
522 		mlxsw_pci_wqe_frag_unmap(mlxsw_pci, wqe, i, DMA_TO_DEVICE);
523 
524 	if (unlikely(!tx_info.is_emad &&
525 		     skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
526 		mlxsw_core_ptp_transmitted(mlxsw_pci->core, skb,
527 					   tx_info.local_port);
528 		skb = NULL;
529 	}
530 
531 	if (skb)
532 		dev_kfree_skb_any(skb);
533 	elem_info->u.sdq.skb = NULL;
534 
535 	if (q->consumer_counter++ != consumer_counter_limit)
536 		dev_dbg_ratelimited(&pdev->dev, "Consumer counter does not match limit in SDQ\n");
537 	spin_unlock(&q->lock);
538 }
539 
540 static void mlxsw_pci_cqe_rdq_md_tx_port_init(struct sk_buff *skb,
541 					      const char *cqe)
542 {
543 	struct mlxsw_skb_cb *cb = mlxsw_skb_cb(skb);
544 
545 	if (mlxsw_pci_cqe2_tx_lag_get(cqe)) {
546 		cb->rx_md_info.tx_port_is_lag = true;
547 		cb->rx_md_info.tx_lag_id = mlxsw_pci_cqe2_tx_lag_id_get(cqe);
548 		cb->rx_md_info.tx_lag_port_index =
549 			mlxsw_pci_cqe2_tx_lag_subport_get(cqe);
550 	} else {
551 		cb->rx_md_info.tx_port_is_lag = false;
552 		cb->rx_md_info.tx_sys_port =
553 			mlxsw_pci_cqe2_tx_system_port_get(cqe);
554 	}
555 
556 	if (cb->rx_md_info.tx_sys_port != MLXSW_PCI_CQE2_TX_PORT_MULTI_PORT &&
557 	    cb->rx_md_info.tx_sys_port != MLXSW_PCI_CQE2_TX_PORT_INVALID)
558 		cb->rx_md_info.tx_port_valid = 1;
559 	else
560 		cb->rx_md_info.tx_port_valid = 0;
561 }
562 
563 static void mlxsw_pci_cqe_rdq_md_init(struct sk_buff *skb, const char *cqe)
564 {
565 	struct mlxsw_skb_cb *cb = mlxsw_skb_cb(skb);
566 
567 	cb->rx_md_info.tx_congestion = mlxsw_pci_cqe2_mirror_cong_get(cqe);
568 	if (cb->rx_md_info.tx_congestion != MLXSW_PCI_CQE2_MIRROR_CONG_INVALID)
569 		cb->rx_md_info.tx_congestion_valid = 1;
570 	else
571 		cb->rx_md_info.tx_congestion_valid = 0;
572 	cb->rx_md_info.tx_congestion <<= MLXSW_PCI_CQE2_MIRROR_CONG_SHIFT;
573 
574 	cb->rx_md_info.latency = mlxsw_pci_cqe2_mirror_latency_get(cqe);
575 	if (cb->rx_md_info.latency != MLXSW_PCI_CQE2_MIRROR_LATENCY_INVALID)
576 		cb->rx_md_info.latency_valid = 1;
577 	else
578 		cb->rx_md_info.latency_valid = 0;
579 
580 	cb->rx_md_info.tx_tc = mlxsw_pci_cqe2_mirror_tclass_get(cqe);
581 	if (cb->rx_md_info.tx_tc != MLXSW_PCI_CQE2_MIRROR_TCLASS_INVALID)
582 		cb->rx_md_info.tx_tc_valid = 1;
583 	else
584 		cb->rx_md_info.tx_tc_valid = 0;
585 
586 	mlxsw_pci_cqe_rdq_md_tx_port_init(skb, cqe);
587 }
588 
589 static void mlxsw_pci_cqe_rdq_handle(struct mlxsw_pci *mlxsw_pci,
590 				     struct mlxsw_pci_queue *q,
591 				     u16 consumer_counter_limit,
592 				     enum mlxsw_pci_cqe_v cqe_v, char *cqe)
593 {
594 	struct pci_dev *pdev = mlxsw_pci->pdev;
595 	struct mlxsw_pci_queue_elem_info *elem_info;
596 	struct mlxsw_rx_info rx_info = {};
597 	char wqe[MLXSW_PCI_WQE_SIZE];
598 	struct sk_buff *skb;
599 	u16 byte_count;
600 	int err;
601 
602 	elem_info = mlxsw_pci_queue_elem_info_consumer_get(q);
603 	skb = elem_info->u.rdq.skb;
604 	memcpy(wqe, elem_info->elem, MLXSW_PCI_WQE_SIZE);
605 
606 	if (q->consumer_counter++ != consumer_counter_limit)
607 		dev_dbg_ratelimited(&pdev->dev, "Consumer counter does not match limit in RDQ\n");
608 
609 	err = mlxsw_pci_rdq_skb_alloc(mlxsw_pci, elem_info);
610 	if (err) {
611 		dev_err_ratelimited(&pdev->dev, "Failed to alloc skb for RDQ\n");
612 		goto out;
613 	}
614 
615 	mlxsw_pci_wqe_frag_unmap(mlxsw_pci, wqe, 0, DMA_FROM_DEVICE);
616 
617 	if (mlxsw_pci_cqe_lag_get(cqe_v, cqe)) {
618 		rx_info.is_lag = true;
619 		rx_info.u.lag_id = mlxsw_pci_cqe_lag_id_get(cqe_v, cqe);
620 		rx_info.lag_port_index =
621 			mlxsw_pci_cqe_lag_subport_get(cqe_v, cqe);
622 	} else {
623 		rx_info.is_lag = false;
624 		rx_info.u.sys_port = mlxsw_pci_cqe_system_port_get(cqe);
625 	}
626 
627 	rx_info.trap_id = mlxsw_pci_cqe_trap_id_get(cqe);
628 
629 	if (rx_info.trap_id == MLXSW_TRAP_ID_DISCARD_INGRESS_ACL ||
630 	    rx_info.trap_id == MLXSW_TRAP_ID_DISCARD_EGRESS_ACL) {
631 		u32 cookie_index = 0;
632 
633 		if (mlxsw_pci->max_cqe_ver >= MLXSW_PCI_CQE_V2)
634 			cookie_index = mlxsw_pci_cqe2_user_def_val_orig_pkt_len_get(cqe);
635 		mlxsw_skb_cb(skb)->rx_md_info.cookie_index = cookie_index;
636 	} else if (rx_info.trap_id >= MLXSW_TRAP_ID_MIRROR_SESSION0 &&
637 		   rx_info.trap_id <= MLXSW_TRAP_ID_MIRROR_SESSION7 &&
638 		   mlxsw_pci->max_cqe_ver >= MLXSW_PCI_CQE_V2) {
639 		rx_info.mirror_reason = mlxsw_pci_cqe2_mirror_reason_get(cqe);
640 		mlxsw_pci_cqe_rdq_md_init(skb, cqe);
641 	} else if (rx_info.trap_id == MLXSW_TRAP_ID_PKT_SAMPLE &&
642 		   mlxsw_pci->max_cqe_ver >= MLXSW_PCI_CQE_V2) {
643 		mlxsw_pci_cqe_rdq_md_tx_port_init(skb, cqe);
644 	}
645 
646 	byte_count = mlxsw_pci_cqe_byte_count_get(cqe);
647 	if (mlxsw_pci_cqe_crc_get(cqe_v, cqe))
648 		byte_count -= ETH_FCS_LEN;
649 	skb_put(skb, byte_count);
650 	mlxsw_core_skb_receive(mlxsw_pci->core, skb, &rx_info);
651 
652 out:
653 	/* Everything is set up, ring doorbell to pass elem to HW */
654 	q->producer_counter++;
655 	mlxsw_pci_queue_doorbell_producer_ring(mlxsw_pci, q);
656 	return;
657 }
658 
659 static char *mlxsw_pci_cq_sw_cqe_get(struct mlxsw_pci_queue *q)
660 {
661 	struct mlxsw_pci_queue_elem_info *elem_info;
662 	char *elem;
663 	bool owner_bit;
664 
665 	elem_info = mlxsw_pci_queue_elem_info_consumer_get(q);
666 	elem = elem_info->elem;
667 	owner_bit = mlxsw_pci_cqe_owner_get(q->u.cq.v, elem);
668 	if (mlxsw_pci_elem_hw_owned(q, owner_bit))
669 		return NULL;
670 	q->consumer_counter++;
671 	rmb(); /* make sure we read owned bit before the rest of elem */
672 	return elem;
673 }
674 
675 static void mlxsw_pci_cq_tasklet(struct tasklet_struct *t)
676 {
677 	struct mlxsw_pci_queue *q = from_tasklet(q, t, tasklet);
678 	struct mlxsw_pci *mlxsw_pci = q->pci;
679 	char *cqe;
680 	int items = 0;
681 	int credits = q->count >> 1;
682 
683 	while ((cqe = mlxsw_pci_cq_sw_cqe_get(q))) {
684 		u16 wqe_counter = mlxsw_pci_cqe_wqe_counter_get(cqe);
685 		u8 sendq = mlxsw_pci_cqe_sr_get(q->u.cq.v, cqe);
686 		u8 dqn = mlxsw_pci_cqe_dqn_get(q->u.cq.v, cqe);
687 		char ncqe[MLXSW_PCI_CQE_SIZE_MAX];
688 
689 		memcpy(ncqe, cqe, q->elem_size);
690 		mlxsw_pci_queue_doorbell_consumer_ring(mlxsw_pci, q);
691 
692 		if (sendq) {
693 			struct mlxsw_pci_queue *sdq;
694 
695 			sdq = mlxsw_pci_sdq_get(mlxsw_pci, dqn);
696 			mlxsw_pci_cqe_sdq_handle(mlxsw_pci, sdq,
697 						 wqe_counter, ncqe);
698 			q->u.cq.comp_sdq_count++;
699 		} else {
700 			struct mlxsw_pci_queue *rdq;
701 
702 			rdq = mlxsw_pci_rdq_get(mlxsw_pci, dqn);
703 			mlxsw_pci_cqe_rdq_handle(mlxsw_pci, rdq,
704 						 wqe_counter, q->u.cq.v, ncqe);
705 			q->u.cq.comp_rdq_count++;
706 		}
707 		if (++items == credits)
708 			break;
709 	}
710 	if (items)
711 		mlxsw_pci_queue_doorbell_arm_consumer_ring(mlxsw_pci, q);
712 }
713 
714 static u16 mlxsw_pci_cq_elem_count(const struct mlxsw_pci_queue *q)
715 {
716 	return q->u.cq.v == MLXSW_PCI_CQE_V2 ? MLXSW_PCI_CQE2_COUNT :
717 					       MLXSW_PCI_CQE01_COUNT;
718 }
719 
720 static u8 mlxsw_pci_cq_elem_size(const struct mlxsw_pci_queue *q)
721 {
722 	return q->u.cq.v == MLXSW_PCI_CQE_V2 ? MLXSW_PCI_CQE2_SIZE :
723 					       MLXSW_PCI_CQE01_SIZE;
724 }
725 
726 static int mlxsw_pci_eq_init(struct mlxsw_pci *mlxsw_pci, char *mbox,
727 			     struct mlxsw_pci_queue *q)
728 {
729 	int i;
730 	int err;
731 
732 	q->consumer_counter = 0;
733 
734 	for (i = 0; i < q->count; i++) {
735 		char *elem = mlxsw_pci_queue_elem_get(q, i);
736 
737 		mlxsw_pci_eqe_owner_set(elem, 1);
738 	}
739 
740 	mlxsw_cmd_mbox_sw2hw_eq_int_msix_set(mbox, 1); /* MSI-X used */
741 	mlxsw_cmd_mbox_sw2hw_eq_st_set(mbox, 1); /* armed */
742 	mlxsw_cmd_mbox_sw2hw_eq_log_eq_size_set(mbox, ilog2(q->count));
743 	for (i = 0; i < MLXSW_PCI_AQ_PAGES; i++) {
744 		dma_addr_t mapaddr = __mlxsw_pci_queue_page_get(q, i);
745 
746 		mlxsw_cmd_mbox_sw2hw_eq_pa_set(mbox, i, mapaddr);
747 	}
748 	err = mlxsw_cmd_sw2hw_eq(mlxsw_pci->core, mbox, q->num);
749 	if (err)
750 		return err;
751 	mlxsw_pci_queue_doorbell_consumer_ring(mlxsw_pci, q);
752 	mlxsw_pci_queue_doorbell_arm_consumer_ring(mlxsw_pci, q);
753 	return 0;
754 }
755 
756 static void mlxsw_pci_eq_fini(struct mlxsw_pci *mlxsw_pci,
757 			      struct mlxsw_pci_queue *q)
758 {
759 	mlxsw_cmd_hw2sw_eq(mlxsw_pci->core, q->num);
760 }
761 
762 static void mlxsw_pci_eq_cmd_event(struct mlxsw_pci *mlxsw_pci, char *eqe)
763 {
764 	mlxsw_pci->cmd.comp.status = mlxsw_pci_eqe_cmd_status_get(eqe);
765 	mlxsw_pci->cmd.comp.out_param =
766 		((u64) mlxsw_pci_eqe_cmd_out_param_h_get(eqe)) << 32 |
767 		mlxsw_pci_eqe_cmd_out_param_l_get(eqe);
768 	mlxsw_pci->cmd.wait_done = true;
769 	wake_up(&mlxsw_pci->cmd.wait);
770 }
771 
772 static char *mlxsw_pci_eq_sw_eqe_get(struct mlxsw_pci_queue *q)
773 {
774 	struct mlxsw_pci_queue_elem_info *elem_info;
775 	char *elem;
776 	bool owner_bit;
777 
778 	elem_info = mlxsw_pci_queue_elem_info_consumer_get(q);
779 	elem = elem_info->elem;
780 	owner_bit = mlxsw_pci_eqe_owner_get(elem);
781 	if (mlxsw_pci_elem_hw_owned(q, owner_bit))
782 		return NULL;
783 	q->consumer_counter++;
784 	rmb(); /* make sure we read owned bit before the rest of elem */
785 	return elem;
786 }
787 
788 static void mlxsw_pci_eq_tasklet(struct tasklet_struct *t)
789 {
790 	struct mlxsw_pci_queue *q = from_tasklet(q, t, tasklet);
791 	struct mlxsw_pci *mlxsw_pci = q->pci;
792 	u8 cq_count = mlxsw_pci_cq_count(mlxsw_pci);
793 	unsigned long active_cqns[BITS_TO_LONGS(MLXSW_PCI_CQS_MAX)];
794 	char *eqe;
795 	u8 cqn;
796 	bool cq_handle = false;
797 	int items = 0;
798 	int credits = q->count >> 1;
799 
800 	memset(&active_cqns, 0, sizeof(active_cqns));
801 
802 	while ((eqe = mlxsw_pci_eq_sw_eqe_get(q))) {
803 
804 		/* Command interface completion events are always received on
805 		 * queue MLXSW_PCI_EQ_ASYNC_NUM (EQ0) and completion events
806 		 * are mapped to queue MLXSW_PCI_EQ_COMP_NUM (EQ1).
807 		 */
808 		switch (q->num) {
809 		case MLXSW_PCI_EQ_ASYNC_NUM:
810 			mlxsw_pci_eq_cmd_event(mlxsw_pci, eqe);
811 			q->u.eq.ev_cmd_count++;
812 			break;
813 		case MLXSW_PCI_EQ_COMP_NUM:
814 			cqn = mlxsw_pci_eqe_cqn_get(eqe);
815 			set_bit(cqn, active_cqns);
816 			cq_handle = true;
817 			q->u.eq.ev_comp_count++;
818 			break;
819 		default:
820 			q->u.eq.ev_other_count++;
821 		}
822 		if (++items == credits)
823 			break;
824 	}
825 	if (items) {
826 		mlxsw_pci_queue_doorbell_consumer_ring(mlxsw_pci, q);
827 		mlxsw_pci_queue_doorbell_arm_consumer_ring(mlxsw_pci, q);
828 	}
829 
830 	if (!cq_handle)
831 		return;
832 	for_each_set_bit(cqn, active_cqns, cq_count) {
833 		q = mlxsw_pci_cq_get(mlxsw_pci, cqn);
834 		mlxsw_pci_queue_tasklet_schedule(q);
835 	}
836 }
837 
838 struct mlxsw_pci_queue_ops {
839 	const char *name;
840 	enum mlxsw_pci_queue_type type;
841 	void (*pre_init)(struct mlxsw_pci *mlxsw_pci,
842 			 struct mlxsw_pci_queue *q);
843 	int (*init)(struct mlxsw_pci *mlxsw_pci, char *mbox,
844 		    struct mlxsw_pci_queue *q);
845 	void (*fini)(struct mlxsw_pci *mlxsw_pci,
846 		     struct mlxsw_pci_queue *q);
847 	void (*tasklet)(struct tasklet_struct *t);
848 	u16 (*elem_count_f)(const struct mlxsw_pci_queue *q);
849 	u8 (*elem_size_f)(const struct mlxsw_pci_queue *q);
850 	u16 elem_count;
851 	u8 elem_size;
852 };
853 
854 static const struct mlxsw_pci_queue_ops mlxsw_pci_sdq_ops = {
855 	.type		= MLXSW_PCI_QUEUE_TYPE_SDQ,
856 	.init		= mlxsw_pci_sdq_init,
857 	.fini		= mlxsw_pci_sdq_fini,
858 	.elem_count	= MLXSW_PCI_WQE_COUNT,
859 	.elem_size	= MLXSW_PCI_WQE_SIZE,
860 };
861 
862 static const struct mlxsw_pci_queue_ops mlxsw_pci_rdq_ops = {
863 	.type		= MLXSW_PCI_QUEUE_TYPE_RDQ,
864 	.init		= mlxsw_pci_rdq_init,
865 	.fini		= mlxsw_pci_rdq_fini,
866 	.elem_count	= MLXSW_PCI_WQE_COUNT,
867 	.elem_size	= MLXSW_PCI_WQE_SIZE
868 };
869 
870 static const struct mlxsw_pci_queue_ops mlxsw_pci_cq_ops = {
871 	.type		= MLXSW_PCI_QUEUE_TYPE_CQ,
872 	.pre_init	= mlxsw_pci_cq_pre_init,
873 	.init		= mlxsw_pci_cq_init,
874 	.fini		= mlxsw_pci_cq_fini,
875 	.tasklet	= mlxsw_pci_cq_tasklet,
876 	.elem_count_f	= mlxsw_pci_cq_elem_count,
877 	.elem_size_f	= mlxsw_pci_cq_elem_size
878 };
879 
880 static const struct mlxsw_pci_queue_ops mlxsw_pci_eq_ops = {
881 	.type		= MLXSW_PCI_QUEUE_TYPE_EQ,
882 	.init		= mlxsw_pci_eq_init,
883 	.fini		= mlxsw_pci_eq_fini,
884 	.tasklet	= mlxsw_pci_eq_tasklet,
885 	.elem_count	= MLXSW_PCI_EQE_COUNT,
886 	.elem_size	= MLXSW_PCI_EQE_SIZE
887 };
888 
889 static int mlxsw_pci_queue_init(struct mlxsw_pci *mlxsw_pci, char *mbox,
890 				const struct mlxsw_pci_queue_ops *q_ops,
891 				struct mlxsw_pci_queue *q, u8 q_num)
892 {
893 	struct mlxsw_pci_mem_item *mem_item = &q->mem_item;
894 	int i;
895 	int err;
896 
897 	q->num = q_num;
898 	if (q_ops->pre_init)
899 		q_ops->pre_init(mlxsw_pci, q);
900 
901 	spin_lock_init(&q->lock);
902 	q->count = q_ops->elem_count_f ? q_ops->elem_count_f(q) :
903 					 q_ops->elem_count;
904 	q->elem_size = q_ops->elem_size_f ? q_ops->elem_size_f(q) :
905 					    q_ops->elem_size;
906 	q->type = q_ops->type;
907 	q->pci = mlxsw_pci;
908 
909 	if (q_ops->tasklet)
910 		tasklet_setup(&q->tasklet, q_ops->tasklet);
911 
912 	mem_item->size = MLXSW_PCI_AQ_SIZE;
913 	mem_item->buf = dma_alloc_coherent(&mlxsw_pci->pdev->dev,
914 					   mem_item->size, &mem_item->mapaddr,
915 					   GFP_KERNEL);
916 	if (!mem_item->buf)
917 		return -ENOMEM;
918 
919 	q->elem_info = kcalloc(q->count, sizeof(*q->elem_info), GFP_KERNEL);
920 	if (!q->elem_info) {
921 		err = -ENOMEM;
922 		goto err_elem_info_alloc;
923 	}
924 
925 	/* Initialize dma mapped elements info elem_info for
926 	 * future easy access.
927 	 */
928 	for (i = 0; i < q->count; i++) {
929 		struct mlxsw_pci_queue_elem_info *elem_info;
930 
931 		elem_info = mlxsw_pci_queue_elem_info_get(q, i);
932 		elem_info->elem =
933 			__mlxsw_pci_queue_elem_get(q, q->elem_size, i);
934 	}
935 
936 	mlxsw_cmd_mbox_zero(mbox);
937 	err = q_ops->init(mlxsw_pci, mbox, q);
938 	if (err)
939 		goto err_q_ops_init;
940 	return 0;
941 
942 err_q_ops_init:
943 	kfree(q->elem_info);
944 err_elem_info_alloc:
945 	dma_free_coherent(&mlxsw_pci->pdev->dev, mem_item->size,
946 			  mem_item->buf, mem_item->mapaddr);
947 	return err;
948 }
949 
950 static void mlxsw_pci_queue_fini(struct mlxsw_pci *mlxsw_pci,
951 				 const struct mlxsw_pci_queue_ops *q_ops,
952 				 struct mlxsw_pci_queue *q)
953 {
954 	struct mlxsw_pci_mem_item *mem_item = &q->mem_item;
955 
956 	q_ops->fini(mlxsw_pci, q);
957 	kfree(q->elem_info);
958 	dma_free_coherent(&mlxsw_pci->pdev->dev, mem_item->size,
959 			  mem_item->buf, mem_item->mapaddr);
960 }
961 
962 static int mlxsw_pci_queue_group_init(struct mlxsw_pci *mlxsw_pci, char *mbox,
963 				      const struct mlxsw_pci_queue_ops *q_ops,
964 				      u8 num_qs)
965 {
966 	struct mlxsw_pci_queue_type_group *queue_group;
967 	int i;
968 	int err;
969 
970 	queue_group = mlxsw_pci_queue_type_group_get(mlxsw_pci, q_ops->type);
971 	queue_group->q = kcalloc(num_qs, sizeof(*queue_group->q), GFP_KERNEL);
972 	if (!queue_group->q)
973 		return -ENOMEM;
974 
975 	for (i = 0; i < num_qs; i++) {
976 		err = mlxsw_pci_queue_init(mlxsw_pci, mbox, q_ops,
977 					   &queue_group->q[i], i);
978 		if (err)
979 			goto err_queue_init;
980 	}
981 	queue_group->count = num_qs;
982 
983 	return 0;
984 
985 err_queue_init:
986 	for (i--; i >= 0; i--)
987 		mlxsw_pci_queue_fini(mlxsw_pci, q_ops, &queue_group->q[i]);
988 	kfree(queue_group->q);
989 	return err;
990 }
991 
992 static void mlxsw_pci_queue_group_fini(struct mlxsw_pci *mlxsw_pci,
993 				       const struct mlxsw_pci_queue_ops *q_ops)
994 {
995 	struct mlxsw_pci_queue_type_group *queue_group;
996 	int i;
997 
998 	queue_group = mlxsw_pci_queue_type_group_get(mlxsw_pci, q_ops->type);
999 	for (i = 0; i < queue_group->count; i++)
1000 		mlxsw_pci_queue_fini(mlxsw_pci, q_ops, &queue_group->q[i]);
1001 	kfree(queue_group->q);
1002 }
1003 
1004 static int mlxsw_pci_aqs_init(struct mlxsw_pci *mlxsw_pci, char *mbox)
1005 {
1006 	struct pci_dev *pdev = mlxsw_pci->pdev;
1007 	u8 num_sdqs;
1008 	u8 sdq_log2sz;
1009 	u8 num_rdqs;
1010 	u8 rdq_log2sz;
1011 	u8 num_cqs;
1012 	u8 cq_log2sz;
1013 	u8 cqv2_log2sz;
1014 	u8 num_eqs;
1015 	u8 eq_log2sz;
1016 	int err;
1017 
1018 	mlxsw_cmd_mbox_zero(mbox);
1019 	err = mlxsw_cmd_query_aq_cap(mlxsw_pci->core, mbox);
1020 	if (err)
1021 		return err;
1022 
1023 	num_sdqs = mlxsw_cmd_mbox_query_aq_cap_max_num_sdqs_get(mbox);
1024 	sdq_log2sz = mlxsw_cmd_mbox_query_aq_cap_log_max_sdq_sz_get(mbox);
1025 	num_rdqs = mlxsw_cmd_mbox_query_aq_cap_max_num_rdqs_get(mbox);
1026 	rdq_log2sz = mlxsw_cmd_mbox_query_aq_cap_log_max_rdq_sz_get(mbox);
1027 	num_cqs = mlxsw_cmd_mbox_query_aq_cap_max_num_cqs_get(mbox);
1028 	cq_log2sz = mlxsw_cmd_mbox_query_aq_cap_log_max_cq_sz_get(mbox);
1029 	cqv2_log2sz = mlxsw_cmd_mbox_query_aq_cap_log_max_cqv2_sz_get(mbox);
1030 	num_eqs = mlxsw_cmd_mbox_query_aq_cap_max_num_eqs_get(mbox);
1031 	eq_log2sz = mlxsw_cmd_mbox_query_aq_cap_log_max_eq_sz_get(mbox);
1032 
1033 	if (num_sdqs + num_rdqs > num_cqs ||
1034 	    num_sdqs < MLXSW_PCI_SDQS_MIN ||
1035 	    num_cqs > MLXSW_PCI_CQS_MAX || num_eqs != MLXSW_PCI_EQS_COUNT) {
1036 		dev_err(&pdev->dev, "Unsupported number of queues\n");
1037 		return -EINVAL;
1038 	}
1039 
1040 	if ((1 << sdq_log2sz != MLXSW_PCI_WQE_COUNT) ||
1041 	    (1 << rdq_log2sz != MLXSW_PCI_WQE_COUNT) ||
1042 	    (1 << cq_log2sz != MLXSW_PCI_CQE01_COUNT) ||
1043 	    (mlxsw_pci->max_cqe_ver == MLXSW_PCI_CQE_V2 &&
1044 	     (1 << cqv2_log2sz != MLXSW_PCI_CQE2_COUNT)) ||
1045 	    (1 << eq_log2sz != MLXSW_PCI_EQE_COUNT)) {
1046 		dev_err(&pdev->dev, "Unsupported number of async queue descriptors\n");
1047 		return -EINVAL;
1048 	}
1049 
1050 	mlxsw_pci->num_sdq_cqs = num_sdqs;
1051 
1052 	err = mlxsw_pci_queue_group_init(mlxsw_pci, mbox, &mlxsw_pci_eq_ops,
1053 					 num_eqs);
1054 	if (err) {
1055 		dev_err(&pdev->dev, "Failed to initialize event queues\n");
1056 		return err;
1057 	}
1058 
1059 	err = mlxsw_pci_queue_group_init(mlxsw_pci, mbox, &mlxsw_pci_cq_ops,
1060 					 num_cqs);
1061 	if (err) {
1062 		dev_err(&pdev->dev, "Failed to initialize completion queues\n");
1063 		goto err_cqs_init;
1064 	}
1065 
1066 	err = mlxsw_pci_queue_group_init(mlxsw_pci, mbox, &mlxsw_pci_sdq_ops,
1067 					 num_sdqs);
1068 	if (err) {
1069 		dev_err(&pdev->dev, "Failed to initialize send descriptor queues\n");
1070 		goto err_sdqs_init;
1071 	}
1072 
1073 	err = mlxsw_pci_queue_group_init(mlxsw_pci, mbox, &mlxsw_pci_rdq_ops,
1074 					 num_rdqs);
1075 	if (err) {
1076 		dev_err(&pdev->dev, "Failed to initialize receive descriptor queues\n");
1077 		goto err_rdqs_init;
1078 	}
1079 
1080 	/* We have to poll in command interface until queues are initialized */
1081 	mlxsw_pci->cmd.nopoll = true;
1082 	return 0;
1083 
1084 err_rdqs_init:
1085 	mlxsw_pci_queue_group_fini(mlxsw_pci, &mlxsw_pci_sdq_ops);
1086 err_sdqs_init:
1087 	mlxsw_pci_queue_group_fini(mlxsw_pci, &mlxsw_pci_cq_ops);
1088 err_cqs_init:
1089 	mlxsw_pci_queue_group_fini(mlxsw_pci, &mlxsw_pci_eq_ops);
1090 	return err;
1091 }
1092 
1093 static void mlxsw_pci_aqs_fini(struct mlxsw_pci *mlxsw_pci)
1094 {
1095 	mlxsw_pci->cmd.nopoll = false;
1096 	mlxsw_pci_queue_group_fini(mlxsw_pci, &mlxsw_pci_rdq_ops);
1097 	mlxsw_pci_queue_group_fini(mlxsw_pci, &mlxsw_pci_sdq_ops);
1098 	mlxsw_pci_queue_group_fini(mlxsw_pci, &mlxsw_pci_cq_ops);
1099 	mlxsw_pci_queue_group_fini(mlxsw_pci, &mlxsw_pci_eq_ops);
1100 }
1101 
1102 static void
1103 mlxsw_pci_config_profile_swid_config(struct mlxsw_pci *mlxsw_pci,
1104 				     char *mbox, int index,
1105 				     const struct mlxsw_swid_config *swid)
1106 {
1107 	u8 mask = 0;
1108 
1109 	if (swid->used_type) {
1110 		mlxsw_cmd_mbox_config_profile_swid_config_type_set(
1111 			mbox, index, swid->type);
1112 		mask |= 1;
1113 	}
1114 	if (swid->used_properties) {
1115 		mlxsw_cmd_mbox_config_profile_swid_config_properties_set(
1116 			mbox, index, swid->properties);
1117 		mask |= 2;
1118 	}
1119 	mlxsw_cmd_mbox_config_profile_swid_config_mask_set(mbox, index, mask);
1120 }
1121 
1122 static int
1123 mlxsw_pci_profile_get_kvd_sizes(const struct mlxsw_pci *mlxsw_pci,
1124 				const struct mlxsw_config_profile *profile,
1125 				struct mlxsw_res *res)
1126 {
1127 	u64 single_size, double_size, linear_size;
1128 	int err;
1129 
1130 	err = mlxsw_core_kvd_sizes_get(mlxsw_pci->core, profile,
1131 				       &single_size, &double_size,
1132 				       &linear_size);
1133 	if (err)
1134 		return err;
1135 
1136 	MLXSW_RES_SET(res, KVD_SINGLE_SIZE, single_size);
1137 	MLXSW_RES_SET(res, KVD_DOUBLE_SIZE, double_size);
1138 	MLXSW_RES_SET(res, KVD_LINEAR_SIZE, linear_size);
1139 
1140 	return 0;
1141 }
1142 
1143 static int mlxsw_pci_config_profile(struct mlxsw_pci *mlxsw_pci, char *mbox,
1144 				    const struct mlxsw_config_profile *profile,
1145 				    struct mlxsw_res *res)
1146 {
1147 	int i;
1148 	int err;
1149 
1150 	mlxsw_cmd_mbox_zero(mbox);
1151 
1152 	if (profile->used_max_vepa_channels) {
1153 		mlxsw_cmd_mbox_config_profile_set_max_vepa_channels_set(
1154 			mbox, 1);
1155 		mlxsw_cmd_mbox_config_profile_max_vepa_channels_set(
1156 			mbox, profile->max_vepa_channels);
1157 	}
1158 	if (profile->used_max_mid) {
1159 		mlxsw_cmd_mbox_config_profile_set_max_mid_set(
1160 			mbox, 1);
1161 		mlxsw_cmd_mbox_config_profile_max_mid_set(
1162 			mbox, profile->max_mid);
1163 	}
1164 	if (profile->used_max_pgt) {
1165 		mlxsw_cmd_mbox_config_profile_set_max_pgt_set(
1166 			mbox, 1);
1167 		mlxsw_cmd_mbox_config_profile_max_pgt_set(
1168 			mbox, profile->max_pgt);
1169 	}
1170 	if (profile->used_max_system_port) {
1171 		mlxsw_cmd_mbox_config_profile_set_max_system_port_set(
1172 			mbox, 1);
1173 		mlxsw_cmd_mbox_config_profile_max_system_port_set(
1174 			mbox, profile->max_system_port);
1175 	}
1176 	if (profile->used_max_vlan_groups) {
1177 		mlxsw_cmd_mbox_config_profile_set_max_vlan_groups_set(
1178 			mbox, 1);
1179 		mlxsw_cmd_mbox_config_profile_max_vlan_groups_set(
1180 			mbox, profile->max_vlan_groups);
1181 	}
1182 	if (profile->used_max_regions) {
1183 		mlxsw_cmd_mbox_config_profile_set_max_regions_set(
1184 			mbox, 1);
1185 		mlxsw_cmd_mbox_config_profile_max_regions_set(
1186 			mbox, profile->max_regions);
1187 	}
1188 	if (profile->used_flood_tables) {
1189 		mlxsw_cmd_mbox_config_profile_set_flood_tables_set(
1190 			mbox, 1);
1191 		mlxsw_cmd_mbox_config_profile_max_flood_tables_set(
1192 			mbox, profile->max_flood_tables);
1193 		mlxsw_cmd_mbox_config_profile_max_vid_flood_tables_set(
1194 			mbox, profile->max_vid_flood_tables);
1195 		mlxsw_cmd_mbox_config_profile_max_fid_offset_flood_tables_set(
1196 			mbox, profile->max_fid_offset_flood_tables);
1197 		mlxsw_cmd_mbox_config_profile_fid_offset_flood_table_size_set(
1198 			mbox, profile->fid_offset_flood_table_size);
1199 		mlxsw_cmd_mbox_config_profile_max_fid_flood_tables_set(
1200 			mbox, profile->max_fid_flood_tables);
1201 		mlxsw_cmd_mbox_config_profile_fid_flood_table_size_set(
1202 			mbox, profile->fid_flood_table_size);
1203 	}
1204 	if (profile->used_flood_mode) {
1205 		mlxsw_cmd_mbox_config_profile_set_flood_mode_set(
1206 			mbox, 1);
1207 		mlxsw_cmd_mbox_config_profile_flood_mode_set(
1208 			mbox, profile->flood_mode);
1209 	}
1210 	if (profile->used_max_ib_mc) {
1211 		mlxsw_cmd_mbox_config_profile_set_max_ib_mc_set(
1212 			mbox, 1);
1213 		mlxsw_cmd_mbox_config_profile_max_ib_mc_set(
1214 			mbox, profile->max_ib_mc);
1215 	}
1216 	if (profile->used_max_pkey) {
1217 		mlxsw_cmd_mbox_config_profile_set_max_pkey_set(
1218 			mbox, 1);
1219 		mlxsw_cmd_mbox_config_profile_max_pkey_set(
1220 			mbox, profile->max_pkey);
1221 	}
1222 	if (profile->used_ar_sec) {
1223 		mlxsw_cmd_mbox_config_profile_set_ar_sec_set(
1224 			mbox, 1);
1225 		mlxsw_cmd_mbox_config_profile_ar_sec_set(
1226 			mbox, profile->ar_sec);
1227 	}
1228 	if (profile->used_adaptive_routing_group_cap) {
1229 		mlxsw_cmd_mbox_config_profile_set_adaptive_routing_group_cap_set(
1230 			mbox, 1);
1231 		mlxsw_cmd_mbox_config_profile_adaptive_routing_group_cap_set(
1232 			mbox, profile->adaptive_routing_group_cap);
1233 	}
1234 	if (profile->used_kvd_sizes && MLXSW_RES_VALID(res, KVD_SIZE)) {
1235 		err = mlxsw_pci_profile_get_kvd_sizes(mlxsw_pci, profile, res);
1236 		if (err)
1237 			return err;
1238 
1239 		mlxsw_cmd_mbox_config_profile_set_kvd_linear_size_set(mbox, 1);
1240 		mlxsw_cmd_mbox_config_profile_kvd_linear_size_set(mbox,
1241 					MLXSW_RES_GET(res, KVD_LINEAR_SIZE));
1242 		mlxsw_cmd_mbox_config_profile_set_kvd_hash_single_size_set(mbox,
1243 									   1);
1244 		mlxsw_cmd_mbox_config_profile_kvd_hash_single_size_set(mbox,
1245 					MLXSW_RES_GET(res, KVD_SINGLE_SIZE));
1246 		mlxsw_cmd_mbox_config_profile_set_kvd_hash_double_size_set(
1247 								mbox, 1);
1248 		mlxsw_cmd_mbox_config_profile_kvd_hash_double_size_set(mbox,
1249 					MLXSW_RES_GET(res, KVD_DOUBLE_SIZE));
1250 	}
1251 	if (profile->used_kvh_xlt_cache_mode) {
1252 		mlxsw_cmd_mbox_config_profile_set_kvh_xlt_cache_mode_set(
1253 			mbox, 1);
1254 		mlxsw_cmd_mbox_config_profile_kvh_xlt_cache_mode_set(
1255 			mbox, profile->kvh_xlt_cache_mode);
1256 	}
1257 
1258 	for (i = 0; i < MLXSW_CONFIG_PROFILE_SWID_COUNT; i++)
1259 		mlxsw_pci_config_profile_swid_config(mlxsw_pci, mbox, i,
1260 						     &profile->swid_config[i]);
1261 
1262 	if (mlxsw_pci->max_cqe_ver > MLXSW_PCI_CQE_V0) {
1263 		mlxsw_cmd_mbox_config_profile_set_cqe_version_set(mbox, 1);
1264 		mlxsw_cmd_mbox_config_profile_cqe_version_set(mbox, 1);
1265 	}
1266 
1267 	return mlxsw_cmd_config_profile_set(mlxsw_pci->core, mbox);
1268 }
1269 
1270 static int mlxsw_pci_boardinfo_xm_process(struct mlxsw_pci *mlxsw_pci,
1271 					  struct mlxsw_bus_info *bus_info,
1272 					  char *mbox)
1273 {
1274 	int count = mlxsw_cmd_mbox_boardinfo_xm_num_local_ports_get(mbox);
1275 	int i;
1276 
1277 	if (!mlxsw_cmd_mbox_boardinfo_xm_exists_get(mbox))
1278 		return 0;
1279 
1280 	bus_info->xm_exists = true;
1281 
1282 	if (count > MLXSW_BUS_INFO_XM_LOCAL_PORTS_MAX) {
1283 		dev_err(&mlxsw_pci->pdev->dev, "Invalid number of XM local ports\n");
1284 		return -EINVAL;
1285 	}
1286 	bus_info->xm_local_ports_count = count;
1287 	for (i = 0; i < count; i++)
1288 		bus_info->xm_local_ports[i] =
1289 			mlxsw_cmd_mbox_boardinfo_xm_local_port_entry_get(mbox,
1290 									 i);
1291 	return 0;
1292 }
1293 
1294 static int mlxsw_pci_boardinfo(struct mlxsw_pci *mlxsw_pci, char *mbox)
1295 {
1296 	struct mlxsw_bus_info *bus_info = &mlxsw_pci->bus_info;
1297 	int err;
1298 
1299 	mlxsw_cmd_mbox_zero(mbox);
1300 	err = mlxsw_cmd_boardinfo(mlxsw_pci->core, mbox);
1301 	if (err)
1302 		return err;
1303 	mlxsw_cmd_mbox_boardinfo_vsd_memcpy_from(mbox, bus_info->vsd);
1304 	mlxsw_cmd_mbox_boardinfo_psid_memcpy_from(mbox, bus_info->psid);
1305 
1306 	return mlxsw_pci_boardinfo_xm_process(mlxsw_pci, bus_info, mbox);
1307 }
1308 
1309 static int mlxsw_pci_fw_area_init(struct mlxsw_pci *mlxsw_pci, char *mbox,
1310 				  u16 num_pages)
1311 {
1312 	struct mlxsw_pci_mem_item *mem_item;
1313 	int nent = 0;
1314 	int i;
1315 	int err;
1316 
1317 	mlxsw_pci->fw_area.items = kcalloc(num_pages, sizeof(*mem_item),
1318 					   GFP_KERNEL);
1319 	if (!mlxsw_pci->fw_area.items)
1320 		return -ENOMEM;
1321 	mlxsw_pci->fw_area.count = num_pages;
1322 
1323 	mlxsw_cmd_mbox_zero(mbox);
1324 	for (i = 0; i < num_pages; i++) {
1325 		mem_item = &mlxsw_pci->fw_area.items[i];
1326 
1327 		mem_item->size = MLXSW_PCI_PAGE_SIZE;
1328 		mem_item->buf = dma_alloc_coherent(&mlxsw_pci->pdev->dev,
1329 						   mem_item->size,
1330 						   &mem_item->mapaddr, GFP_KERNEL);
1331 		if (!mem_item->buf) {
1332 			err = -ENOMEM;
1333 			goto err_alloc;
1334 		}
1335 		mlxsw_cmd_mbox_map_fa_pa_set(mbox, nent, mem_item->mapaddr);
1336 		mlxsw_cmd_mbox_map_fa_log2size_set(mbox, nent, 0); /* 1 page */
1337 		if (++nent == MLXSW_CMD_MAP_FA_VPM_ENTRIES_MAX) {
1338 			err = mlxsw_cmd_map_fa(mlxsw_pci->core, mbox, nent);
1339 			if (err)
1340 				goto err_cmd_map_fa;
1341 			nent = 0;
1342 			mlxsw_cmd_mbox_zero(mbox);
1343 		}
1344 	}
1345 
1346 	if (nent) {
1347 		err = mlxsw_cmd_map_fa(mlxsw_pci->core, mbox, nent);
1348 		if (err)
1349 			goto err_cmd_map_fa;
1350 	}
1351 
1352 	return 0;
1353 
1354 err_cmd_map_fa:
1355 err_alloc:
1356 	for (i--; i >= 0; i--) {
1357 		mem_item = &mlxsw_pci->fw_area.items[i];
1358 
1359 		dma_free_coherent(&mlxsw_pci->pdev->dev, mem_item->size,
1360 				  mem_item->buf, mem_item->mapaddr);
1361 	}
1362 	kfree(mlxsw_pci->fw_area.items);
1363 	return err;
1364 }
1365 
1366 static void mlxsw_pci_fw_area_fini(struct mlxsw_pci *mlxsw_pci)
1367 {
1368 	struct mlxsw_pci_mem_item *mem_item;
1369 	int i;
1370 
1371 	mlxsw_cmd_unmap_fa(mlxsw_pci->core);
1372 
1373 	for (i = 0; i < mlxsw_pci->fw_area.count; i++) {
1374 		mem_item = &mlxsw_pci->fw_area.items[i];
1375 
1376 		dma_free_coherent(&mlxsw_pci->pdev->dev, mem_item->size,
1377 				  mem_item->buf, mem_item->mapaddr);
1378 	}
1379 	kfree(mlxsw_pci->fw_area.items);
1380 }
1381 
1382 static irqreturn_t mlxsw_pci_eq_irq_handler(int irq, void *dev_id)
1383 {
1384 	struct mlxsw_pci *mlxsw_pci = dev_id;
1385 	struct mlxsw_pci_queue *q;
1386 	int i;
1387 
1388 	for (i = 0; i < MLXSW_PCI_EQS_COUNT; i++) {
1389 		q = mlxsw_pci_eq_get(mlxsw_pci, i);
1390 		mlxsw_pci_queue_tasklet_schedule(q);
1391 	}
1392 	return IRQ_HANDLED;
1393 }
1394 
1395 static int mlxsw_pci_mbox_alloc(struct mlxsw_pci *mlxsw_pci,
1396 				struct mlxsw_pci_mem_item *mbox)
1397 {
1398 	struct pci_dev *pdev = mlxsw_pci->pdev;
1399 	int err = 0;
1400 
1401 	mbox->size = MLXSW_CMD_MBOX_SIZE;
1402 	mbox->buf = dma_alloc_coherent(&pdev->dev, MLXSW_CMD_MBOX_SIZE,
1403 				       &mbox->mapaddr, GFP_KERNEL);
1404 	if (!mbox->buf) {
1405 		dev_err(&pdev->dev, "Failed allocating memory for mailbox\n");
1406 		err = -ENOMEM;
1407 	}
1408 
1409 	return err;
1410 }
1411 
1412 static void mlxsw_pci_mbox_free(struct mlxsw_pci *mlxsw_pci,
1413 				struct mlxsw_pci_mem_item *mbox)
1414 {
1415 	struct pci_dev *pdev = mlxsw_pci->pdev;
1416 
1417 	dma_free_coherent(&pdev->dev, MLXSW_CMD_MBOX_SIZE, mbox->buf,
1418 			  mbox->mapaddr);
1419 }
1420 
1421 static int mlxsw_pci_sys_ready_wait(struct mlxsw_pci *mlxsw_pci,
1422 				    const struct pci_device_id *id,
1423 				    u32 *p_sys_status)
1424 {
1425 	unsigned long end;
1426 	u32 val;
1427 
1428 	/* We must wait for the HW to become responsive. */
1429 	msleep(MLXSW_PCI_SW_RESET_WAIT_MSECS);
1430 
1431 	end = jiffies + msecs_to_jiffies(MLXSW_PCI_SW_RESET_TIMEOUT_MSECS);
1432 	do {
1433 		val = mlxsw_pci_read32(mlxsw_pci, FW_READY);
1434 		if ((val & MLXSW_PCI_FW_READY_MASK) == MLXSW_PCI_FW_READY_MAGIC)
1435 			return 0;
1436 		cond_resched();
1437 	} while (time_before(jiffies, end));
1438 
1439 	*p_sys_status = val & MLXSW_PCI_FW_READY_MASK;
1440 
1441 	return -EBUSY;
1442 }
1443 
1444 static int mlxsw_pci_sw_reset(struct mlxsw_pci *mlxsw_pci,
1445 			      const struct pci_device_id *id)
1446 {
1447 	struct pci_dev *pdev = mlxsw_pci->pdev;
1448 	char mrsr_pl[MLXSW_REG_MRSR_LEN];
1449 	u32 sys_status;
1450 	int err;
1451 
1452 	err = mlxsw_pci_sys_ready_wait(mlxsw_pci, id, &sys_status);
1453 	if (err) {
1454 		dev_err(&pdev->dev, "Failed to reach system ready status before reset. Status is 0x%x\n",
1455 			sys_status);
1456 		return err;
1457 	}
1458 
1459 	mlxsw_reg_mrsr_pack(mrsr_pl);
1460 	err = mlxsw_reg_write(mlxsw_pci->core, MLXSW_REG(mrsr), mrsr_pl);
1461 	if (err)
1462 		return err;
1463 
1464 	err = mlxsw_pci_sys_ready_wait(mlxsw_pci, id, &sys_status);
1465 	if (err) {
1466 		dev_err(&pdev->dev, "Failed to reach system ready status after reset. Status is 0x%x\n",
1467 			sys_status);
1468 		return err;
1469 	}
1470 
1471 	return 0;
1472 }
1473 
1474 static int mlxsw_pci_alloc_irq_vectors(struct mlxsw_pci *mlxsw_pci)
1475 {
1476 	int err;
1477 
1478 	err = pci_alloc_irq_vectors(mlxsw_pci->pdev, 1, 1, PCI_IRQ_MSIX);
1479 	if (err < 0)
1480 		dev_err(&mlxsw_pci->pdev->dev, "MSI-X init failed\n");
1481 	return err;
1482 }
1483 
1484 static void mlxsw_pci_free_irq_vectors(struct mlxsw_pci *mlxsw_pci)
1485 {
1486 	pci_free_irq_vectors(mlxsw_pci->pdev);
1487 }
1488 
1489 static int mlxsw_pci_init(void *bus_priv, struct mlxsw_core *mlxsw_core,
1490 			  const struct mlxsw_config_profile *profile,
1491 			  struct mlxsw_res *res)
1492 {
1493 	struct mlxsw_pci *mlxsw_pci = bus_priv;
1494 	struct pci_dev *pdev = mlxsw_pci->pdev;
1495 	char *mbox;
1496 	u16 num_pages;
1497 	int err;
1498 
1499 	mlxsw_pci->core = mlxsw_core;
1500 
1501 	mbox = mlxsw_cmd_mbox_alloc();
1502 	if (!mbox)
1503 		return -ENOMEM;
1504 
1505 	err = mlxsw_pci_sw_reset(mlxsw_pci, mlxsw_pci->id);
1506 	if (err)
1507 		goto err_sw_reset;
1508 
1509 	err = mlxsw_pci_alloc_irq_vectors(mlxsw_pci);
1510 	if (err < 0) {
1511 		dev_err(&pdev->dev, "MSI-X init failed\n");
1512 		goto err_alloc_irq;
1513 	}
1514 
1515 	err = mlxsw_cmd_query_fw(mlxsw_core, mbox);
1516 	if (err)
1517 		goto err_query_fw;
1518 
1519 	mlxsw_pci->bus_info.fw_rev.major =
1520 		mlxsw_cmd_mbox_query_fw_fw_rev_major_get(mbox);
1521 	mlxsw_pci->bus_info.fw_rev.minor =
1522 		mlxsw_cmd_mbox_query_fw_fw_rev_minor_get(mbox);
1523 	mlxsw_pci->bus_info.fw_rev.subminor =
1524 		mlxsw_cmd_mbox_query_fw_fw_rev_subminor_get(mbox);
1525 
1526 	if (mlxsw_cmd_mbox_query_fw_cmd_interface_rev_get(mbox) != 1) {
1527 		dev_err(&pdev->dev, "Unsupported cmd interface revision ID queried from hw\n");
1528 		err = -EINVAL;
1529 		goto err_iface_rev;
1530 	}
1531 	if (mlxsw_cmd_mbox_query_fw_doorbell_page_bar_get(mbox) != 0) {
1532 		dev_err(&pdev->dev, "Unsupported doorbell page bar queried from hw\n");
1533 		err = -EINVAL;
1534 		goto err_doorbell_page_bar;
1535 	}
1536 
1537 	mlxsw_pci->doorbell_offset =
1538 		mlxsw_cmd_mbox_query_fw_doorbell_page_offset_get(mbox);
1539 
1540 	if (mlxsw_cmd_mbox_query_fw_fr_rn_clk_bar_get(mbox) != 0) {
1541 		dev_err(&pdev->dev, "Unsupported free running clock BAR queried from hw\n");
1542 		err = -EINVAL;
1543 		goto err_fr_rn_clk_bar;
1544 	}
1545 
1546 	mlxsw_pci->free_running_clock_offset =
1547 		mlxsw_cmd_mbox_query_fw_free_running_clock_offset_get(mbox);
1548 
1549 	num_pages = mlxsw_cmd_mbox_query_fw_fw_pages_get(mbox);
1550 	err = mlxsw_pci_fw_area_init(mlxsw_pci, mbox, num_pages);
1551 	if (err)
1552 		goto err_fw_area_init;
1553 
1554 	err = mlxsw_pci_boardinfo(mlxsw_pci, mbox);
1555 	if (err)
1556 		goto err_boardinfo;
1557 
1558 	err = mlxsw_core_resources_query(mlxsw_core, mbox, res);
1559 	if (err)
1560 		goto err_query_resources;
1561 
1562 	if (MLXSW_CORE_RES_VALID(mlxsw_core, CQE_V2) &&
1563 	    MLXSW_CORE_RES_GET(mlxsw_core, CQE_V2))
1564 		mlxsw_pci->max_cqe_ver = MLXSW_PCI_CQE_V2;
1565 	else if (MLXSW_CORE_RES_VALID(mlxsw_core, CQE_V1) &&
1566 		 MLXSW_CORE_RES_GET(mlxsw_core, CQE_V1))
1567 		mlxsw_pci->max_cqe_ver = MLXSW_PCI_CQE_V1;
1568 	else if ((MLXSW_CORE_RES_VALID(mlxsw_core, CQE_V0) &&
1569 		  MLXSW_CORE_RES_GET(mlxsw_core, CQE_V0)) ||
1570 		 !MLXSW_CORE_RES_VALID(mlxsw_core, CQE_V0)) {
1571 		mlxsw_pci->max_cqe_ver = MLXSW_PCI_CQE_V0;
1572 	} else {
1573 		dev_err(&pdev->dev, "Invalid supported CQE version combination reported\n");
1574 		goto err_cqe_v_check;
1575 	}
1576 
1577 	err = mlxsw_pci_config_profile(mlxsw_pci, mbox, profile, res);
1578 	if (err)
1579 		goto err_config_profile;
1580 
1581 	err = mlxsw_pci_aqs_init(mlxsw_pci, mbox);
1582 	if (err)
1583 		goto err_aqs_init;
1584 
1585 	err = request_irq(pci_irq_vector(pdev, 0),
1586 			  mlxsw_pci_eq_irq_handler, 0,
1587 			  mlxsw_pci->bus_info.device_kind, mlxsw_pci);
1588 	if (err) {
1589 		dev_err(&pdev->dev, "IRQ request failed\n");
1590 		goto err_request_eq_irq;
1591 	}
1592 
1593 	goto mbox_put;
1594 
1595 err_request_eq_irq:
1596 	mlxsw_pci_aqs_fini(mlxsw_pci);
1597 err_aqs_init:
1598 err_config_profile:
1599 err_cqe_v_check:
1600 err_query_resources:
1601 err_boardinfo:
1602 	mlxsw_pci_fw_area_fini(mlxsw_pci);
1603 err_fw_area_init:
1604 err_fr_rn_clk_bar:
1605 err_doorbell_page_bar:
1606 err_iface_rev:
1607 err_query_fw:
1608 	mlxsw_pci_free_irq_vectors(mlxsw_pci);
1609 err_alloc_irq:
1610 err_sw_reset:
1611 mbox_put:
1612 	mlxsw_cmd_mbox_free(mbox);
1613 	return err;
1614 }
1615 
1616 static void mlxsw_pci_fini(void *bus_priv)
1617 {
1618 	struct mlxsw_pci *mlxsw_pci = bus_priv;
1619 
1620 	free_irq(pci_irq_vector(mlxsw_pci->pdev, 0), mlxsw_pci);
1621 	mlxsw_pci_aqs_fini(mlxsw_pci);
1622 	mlxsw_pci_fw_area_fini(mlxsw_pci);
1623 	mlxsw_pci_free_irq_vectors(mlxsw_pci);
1624 }
1625 
1626 static struct mlxsw_pci_queue *
1627 mlxsw_pci_sdq_pick(struct mlxsw_pci *mlxsw_pci,
1628 		   const struct mlxsw_tx_info *tx_info)
1629 {
1630 	u8 ctl_sdq_count = mlxsw_pci_sdq_count(mlxsw_pci) - 1;
1631 	u8 sdqn;
1632 
1633 	if (tx_info->is_emad) {
1634 		sdqn = MLXSW_PCI_SDQ_EMAD_INDEX;
1635 	} else {
1636 		BUILD_BUG_ON(MLXSW_PCI_SDQ_EMAD_INDEX != 0);
1637 		sdqn = 1 + (tx_info->local_port % ctl_sdq_count);
1638 	}
1639 
1640 	return mlxsw_pci_sdq_get(mlxsw_pci, sdqn);
1641 }
1642 
1643 static bool mlxsw_pci_skb_transmit_busy(void *bus_priv,
1644 					const struct mlxsw_tx_info *tx_info)
1645 {
1646 	struct mlxsw_pci *mlxsw_pci = bus_priv;
1647 	struct mlxsw_pci_queue *q = mlxsw_pci_sdq_pick(mlxsw_pci, tx_info);
1648 
1649 	return !mlxsw_pci_queue_elem_info_producer_get(q);
1650 }
1651 
1652 static int mlxsw_pci_skb_transmit(void *bus_priv, struct sk_buff *skb,
1653 				  const struct mlxsw_tx_info *tx_info)
1654 {
1655 	struct mlxsw_pci *mlxsw_pci = bus_priv;
1656 	struct mlxsw_pci_queue *q;
1657 	struct mlxsw_pci_queue_elem_info *elem_info;
1658 	char *wqe;
1659 	int i;
1660 	int err;
1661 
1662 	if (skb_shinfo(skb)->nr_frags > MLXSW_PCI_WQE_SG_ENTRIES - 1) {
1663 		err = skb_linearize(skb);
1664 		if (err)
1665 			return err;
1666 	}
1667 
1668 	q = mlxsw_pci_sdq_pick(mlxsw_pci, tx_info);
1669 	spin_lock_bh(&q->lock);
1670 	elem_info = mlxsw_pci_queue_elem_info_producer_get(q);
1671 	if (!elem_info) {
1672 		/* queue is full */
1673 		err = -EAGAIN;
1674 		goto unlock;
1675 	}
1676 	mlxsw_skb_cb(skb)->tx_info = *tx_info;
1677 	elem_info->u.sdq.skb = skb;
1678 
1679 	wqe = elem_info->elem;
1680 	mlxsw_pci_wqe_c_set(wqe, 1); /* always report completion */
1681 	mlxsw_pci_wqe_lp_set(wqe, !!tx_info->is_emad);
1682 	mlxsw_pci_wqe_type_set(wqe, MLXSW_PCI_WQE_TYPE_ETHERNET);
1683 
1684 	err = mlxsw_pci_wqe_frag_map(mlxsw_pci, wqe, 0, skb->data,
1685 				     skb_headlen(skb), DMA_TO_DEVICE);
1686 	if (err)
1687 		goto unlock;
1688 
1689 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1690 		const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1691 
1692 		err = mlxsw_pci_wqe_frag_map(mlxsw_pci, wqe, i + 1,
1693 					     skb_frag_address(frag),
1694 					     skb_frag_size(frag),
1695 					     DMA_TO_DEVICE);
1696 		if (err)
1697 			goto unmap_frags;
1698 	}
1699 
1700 	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP))
1701 		skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
1702 
1703 	/* Set unused sq entries byte count to zero. */
1704 	for (i++; i < MLXSW_PCI_WQE_SG_ENTRIES; i++)
1705 		mlxsw_pci_wqe_byte_count_set(wqe, i, 0);
1706 
1707 	/* Everything is set up, ring producer doorbell to get HW going */
1708 	q->producer_counter++;
1709 	mlxsw_pci_queue_doorbell_producer_ring(mlxsw_pci, q);
1710 
1711 	goto unlock;
1712 
1713 unmap_frags:
1714 	for (; i >= 0; i--)
1715 		mlxsw_pci_wqe_frag_unmap(mlxsw_pci, wqe, i, DMA_TO_DEVICE);
1716 unlock:
1717 	spin_unlock_bh(&q->lock);
1718 	return err;
1719 }
1720 
1721 static int mlxsw_pci_cmd_exec(void *bus_priv, u16 opcode, u8 opcode_mod,
1722 			      u32 in_mod, bool out_mbox_direct,
1723 			      char *in_mbox, size_t in_mbox_size,
1724 			      char *out_mbox, size_t out_mbox_size,
1725 			      u8 *p_status)
1726 {
1727 	struct mlxsw_pci *mlxsw_pci = bus_priv;
1728 	dma_addr_t in_mapaddr = 0, out_mapaddr = 0;
1729 	bool evreq = mlxsw_pci->cmd.nopoll;
1730 	unsigned long timeout = msecs_to_jiffies(MLXSW_PCI_CIR_TIMEOUT_MSECS);
1731 	bool *p_wait_done = &mlxsw_pci->cmd.wait_done;
1732 	int err;
1733 
1734 	*p_status = MLXSW_CMD_STATUS_OK;
1735 
1736 	err = mutex_lock_interruptible(&mlxsw_pci->cmd.lock);
1737 	if (err)
1738 		return err;
1739 
1740 	if (in_mbox) {
1741 		memcpy(mlxsw_pci->cmd.in_mbox.buf, in_mbox, in_mbox_size);
1742 		in_mapaddr = mlxsw_pci->cmd.in_mbox.mapaddr;
1743 	}
1744 	mlxsw_pci_write32(mlxsw_pci, CIR_IN_PARAM_HI, upper_32_bits(in_mapaddr));
1745 	mlxsw_pci_write32(mlxsw_pci, CIR_IN_PARAM_LO, lower_32_bits(in_mapaddr));
1746 
1747 	if (out_mbox)
1748 		out_mapaddr = mlxsw_pci->cmd.out_mbox.mapaddr;
1749 	mlxsw_pci_write32(mlxsw_pci, CIR_OUT_PARAM_HI, upper_32_bits(out_mapaddr));
1750 	mlxsw_pci_write32(mlxsw_pci, CIR_OUT_PARAM_LO, lower_32_bits(out_mapaddr));
1751 
1752 	mlxsw_pci_write32(mlxsw_pci, CIR_IN_MODIFIER, in_mod);
1753 	mlxsw_pci_write32(mlxsw_pci, CIR_TOKEN, 0);
1754 
1755 	*p_wait_done = false;
1756 
1757 	wmb(); /* all needs to be written before we write control register */
1758 	mlxsw_pci_write32(mlxsw_pci, CIR_CTRL,
1759 			  MLXSW_PCI_CIR_CTRL_GO_BIT |
1760 			  (evreq ? MLXSW_PCI_CIR_CTRL_EVREQ_BIT : 0) |
1761 			  (opcode_mod << MLXSW_PCI_CIR_CTRL_OPCODE_MOD_SHIFT) |
1762 			  opcode);
1763 
1764 	if (!evreq) {
1765 		unsigned long end;
1766 
1767 		end = jiffies + timeout;
1768 		do {
1769 			u32 ctrl = mlxsw_pci_read32(mlxsw_pci, CIR_CTRL);
1770 
1771 			if (!(ctrl & MLXSW_PCI_CIR_CTRL_GO_BIT)) {
1772 				*p_wait_done = true;
1773 				*p_status = ctrl >> MLXSW_PCI_CIR_CTRL_STATUS_SHIFT;
1774 				break;
1775 			}
1776 			cond_resched();
1777 		} while (time_before(jiffies, end));
1778 	} else {
1779 		wait_event_timeout(mlxsw_pci->cmd.wait, *p_wait_done, timeout);
1780 		*p_status = mlxsw_pci->cmd.comp.status;
1781 	}
1782 
1783 	err = 0;
1784 	if (*p_wait_done) {
1785 		if (*p_status)
1786 			err = -EIO;
1787 	} else {
1788 		err = -ETIMEDOUT;
1789 	}
1790 
1791 	if (!err && out_mbox && out_mbox_direct) {
1792 		/* Some commands don't use output param as address to mailbox
1793 		 * but they store output directly into registers. In that case,
1794 		 * copy registers into mbox buffer.
1795 		 */
1796 		__be32 tmp;
1797 
1798 		if (!evreq) {
1799 			tmp = cpu_to_be32(mlxsw_pci_read32(mlxsw_pci,
1800 							   CIR_OUT_PARAM_HI));
1801 			memcpy(out_mbox, &tmp, sizeof(tmp));
1802 			tmp = cpu_to_be32(mlxsw_pci_read32(mlxsw_pci,
1803 							   CIR_OUT_PARAM_LO));
1804 			memcpy(out_mbox + sizeof(tmp), &tmp, sizeof(tmp));
1805 		}
1806 	} else if (!err && out_mbox) {
1807 		memcpy(out_mbox, mlxsw_pci->cmd.out_mbox.buf, out_mbox_size);
1808 	}
1809 
1810 	mutex_unlock(&mlxsw_pci->cmd.lock);
1811 
1812 	return err;
1813 }
1814 
1815 static u32 mlxsw_pci_read_frc_h(void *bus_priv)
1816 {
1817 	struct mlxsw_pci *mlxsw_pci = bus_priv;
1818 	u64 frc_offset;
1819 
1820 	frc_offset = mlxsw_pci->free_running_clock_offset;
1821 	return mlxsw_pci_read32(mlxsw_pci, FREE_RUNNING_CLOCK_H(frc_offset));
1822 }
1823 
1824 static u32 mlxsw_pci_read_frc_l(void *bus_priv)
1825 {
1826 	struct mlxsw_pci *mlxsw_pci = bus_priv;
1827 	u64 frc_offset;
1828 
1829 	frc_offset = mlxsw_pci->free_running_clock_offset;
1830 	return mlxsw_pci_read32(mlxsw_pci, FREE_RUNNING_CLOCK_L(frc_offset));
1831 }
1832 
1833 static const struct mlxsw_bus mlxsw_pci_bus = {
1834 	.kind			= "pci",
1835 	.init			= mlxsw_pci_init,
1836 	.fini			= mlxsw_pci_fini,
1837 	.skb_transmit_busy	= mlxsw_pci_skb_transmit_busy,
1838 	.skb_transmit		= mlxsw_pci_skb_transmit,
1839 	.cmd_exec		= mlxsw_pci_cmd_exec,
1840 	.read_frc_h		= mlxsw_pci_read_frc_h,
1841 	.read_frc_l		= mlxsw_pci_read_frc_l,
1842 	.features		= MLXSW_BUS_F_TXRX | MLXSW_BUS_F_RESET,
1843 };
1844 
1845 static int mlxsw_pci_cmd_init(struct mlxsw_pci *mlxsw_pci)
1846 {
1847 	int err;
1848 
1849 	mutex_init(&mlxsw_pci->cmd.lock);
1850 	init_waitqueue_head(&mlxsw_pci->cmd.wait);
1851 
1852 	err = mlxsw_pci_mbox_alloc(mlxsw_pci, &mlxsw_pci->cmd.in_mbox);
1853 	if (err)
1854 		goto err_in_mbox_alloc;
1855 
1856 	err = mlxsw_pci_mbox_alloc(mlxsw_pci, &mlxsw_pci->cmd.out_mbox);
1857 	if (err)
1858 		goto err_out_mbox_alloc;
1859 
1860 	return 0;
1861 
1862 err_out_mbox_alloc:
1863 	mlxsw_pci_mbox_free(mlxsw_pci, &mlxsw_pci->cmd.in_mbox);
1864 err_in_mbox_alloc:
1865 	mutex_destroy(&mlxsw_pci->cmd.lock);
1866 	return err;
1867 }
1868 
1869 static void mlxsw_pci_cmd_fini(struct mlxsw_pci *mlxsw_pci)
1870 {
1871 	mlxsw_pci_mbox_free(mlxsw_pci, &mlxsw_pci->cmd.out_mbox);
1872 	mlxsw_pci_mbox_free(mlxsw_pci, &mlxsw_pci->cmd.in_mbox);
1873 	mutex_destroy(&mlxsw_pci->cmd.lock);
1874 }
1875 
1876 static int mlxsw_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
1877 {
1878 	const char *driver_name = dev_driver_string(&pdev->dev);
1879 	struct mlxsw_pci *mlxsw_pci;
1880 	int err;
1881 
1882 	mlxsw_pci = kzalloc(sizeof(*mlxsw_pci), GFP_KERNEL);
1883 	if (!mlxsw_pci)
1884 		return -ENOMEM;
1885 
1886 	err = pci_enable_device(pdev);
1887 	if (err) {
1888 		dev_err(&pdev->dev, "pci_enable_device failed\n");
1889 		goto err_pci_enable_device;
1890 	}
1891 
1892 	err = pci_request_regions(pdev, driver_name);
1893 	if (err) {
1894 		dev_err(&pdev->dev, "pci_request_regions failed\n");
1895 		goto err_pci_request_regions;
1896 	}
1897 
1898 	err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
1899 	if (err) {
1900 		err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
1901 		if (err) {
1902 			dev_err(&pdev->dev, "dma_set_mask failed\n");
1903 			goto err_pci_set_dma_mask;
1904 		}
1905 	}
1906 
1907 	if (pci_resource_len(pdev, 0) < MLXSW_PCI_BAR0_SIZE) {
1908 		dev_err(&pdev->dev, "invalid PCI region size\n");
1909 		err = -EINVAL;
1910 		goto err_pci_resource_len_check;
1911 	}
1912 
1913 	mlxsw_pci->hw_addr = ioremap(pci_resource_start(pdev, 0),
1914 				     pci_resource_len(pdev, 0));
1915 	if (!mlxsw_pci->hw_addr) {
1916 		dev_err(&pdev->dev, "ioremap failed\n");
1917 		err = -EIO;
1918 		goto err_ioremap;
1919 	}
1920 	pci_set_master(pdev);
1921 
1922 	mlxsw_pci->pdev = pdev;
1923 	pci_set_drvdata(pdev, mlxsw_pci);
1924 
1925 	err = mlxsw_pci_cmd_init(mlxsw_pci);
1926 	if (err)
1927 		goto err_pci_cmd_init;
1928 
1929 	mlxsw_pci->bus_info.device_kind = driver_name;
1930 	mlxsw_pci->bus_info.device_name = pci_name(mlxsw_pci->pdev);
1931 	mlxsw_pci->bus_info.dev = &pdev->dev;
1932 	mlxsw_pci->bus_info.read_frc_capable = true;
1933 	mlxsw_pci->id = id;
1934 
1935 	err = mlxsw_core_bus_device_register(&mlxsw_pci->bus_info,
1936 					     &mlxsw_pci_bus, mlxsw_pci, false,
1937 					     NULL, NULL);
1938 	if (err) {
1939 		dev_err(&pdev->dev, "cannot register bus device\n");
1940 		goto err_bus_device_register;
1941 	}
1942 
1943 	return 0;
1944 
1945 err_bus_device_register:
1946 	mlxsw_pci_cmd_fini(mlxsw_pci);
1947 err_pci_cmd_init:
1948 	iounmap(mlxsw_pci->hw_addr);
1949 err_ioremap:
1950 err_pci_resource_len_check:
1951 err_pci_set_dma_mask:
1952 	pci_release_regions(pdev);
1953 err_pci_request_regions:
1954 	pci_disable_device(pdev);
1955 err_pci_enable_device:
1956 	kfree(mlxsw_pci);
1957 	return err;
1958 }
1959 
1960 static void mlxsw_pci_remove(struct pci_dev *pdev)
1961 {
1962 	struct mlxsw_pci *mlxsw_pci = pci_get_drvdata(pdev);
1963 
1964 	mlxsw_core_bus_device_unregister(mlxsw_pci->core, false);
1965 	mlxsw_pci_cmd_fini(mlxsw_pci);
1966 	iounmap(mlxsw_pci->hw_addr);
1967 	pci_release_regions(mlxsw_pci->pdev);
1968 	pci_disable_device(mlxsw_pci->pdev);
1969 	kfree(mlxsw_pci);
1970 }
1971 
1972 int mlxsw_pci_driver_register(struct pci_driver *pci_driver)
1973 {
1974 	pci_driver->probe = mlxsw_pci_probe;
1975 	pci_driver->remove = mlxsw_pci_remove;
1976 	return pci_register_driver(pci_driver);
1977 }
1978 EXPORT_SYMBOL(mlxsw_pci_driver_register);
1979 
1980 void mlxsw_pci_driver_unregister(struct pci_driver *pci_driver)
1981 {
1982 	pci_unregister_driver(pci_driver);
1983 }
1984 EXPORT_SYMBOL(mlxsw_pci_driver_unregister);
1985 
1986 static int __init mlxsw_pci_module_init(void)
1987 {
1988 	return 0;
1989 }
1990 
1991 static void __exit mlxsw_pci_module_exit(void)
1992 {
1993 }
1994 
1995 module_init(mlxsw_pci_module_init);
1996 module_exit(mlxsw_pci_module_exit);
1997 
1998 MODULE_LICENSE("Dual BSD/GPL");
1999 MODULE_AUTHOR("Jiri Pirko <jiri@mellanox.com>");
2000 MODULE_DESCRIPTION("Mellanox switch PCI interface driver");
2001