xref: /linux/drivers/net/ethernet/mellanox/mlxsw/pci.c (revision a6a6a98094116b60e5523a571d9443c53325f5b1)
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/types.h>
12 #include <linux/skbuff.h>
13 #include <linux/if_vlan.h>
14 #include <linux/log2.h>
15 #include <linux/string.h>
16 #include <net/page_pool/helpers.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 enum mlxsw_pci_cq_type {
40 	MLXSW_PCI_CQ_SDQ,
41 	MLXSW_PCI_CQ_RDQ,
42 };
43 
44 static const u16 mlxsw_pci_doorbell_type_offset[] = {
45 	MLXSW_PCI_DOORBELL_SDQ_OFFSET,	/* for type MLXSW_PCI_QUEUE_TYPE_SDQ */
46 	MLXSW_PCI_DOORBELL_RDQ_OFFSET,	/* for type MLXSW_PCI_QUEUE_TYPE_RDQ */
47 	MLXSW_PCI_DOORBELL_CQ_OFFSET,	/* for type MLXSW_PCI_QUEUE_TYPE_CQ */
48 	MLXSW_PCI_DOORBELL_EQ_OFFSET,	/* for type MLXSW_PCI_QUEUE_TYPE_EQ */
49 };
50 
51 static const u16 mlxsw_pci_doorbell_arm_type_offset[] = {
52 	0, /* unused */
53 	0, /* unused */
54 	MLXSW_PCI_DOORBELL_ARM_CQ_OFFSET, /* for type MLXSW_PCI_QUEUE_TYPE_CQ */
55 	MLXSW_PCI_DOORBELL_ARM_EQ_OFFSET, /* for type MLXSW_PCI_QUEUE_TYPE_EQ */
56 };
57 
58 struct mlxsw_pci_mem_item {
59 	char *buf;
60 	dma_addr_t mapaddr;
61 	size_t size;
62 };
63 
64 struct mlxsw_pci_queue_elem_info {
65 	struct page *page;
66 	char *elem; /* pointer to actual dma mapped element mem chunk */
67 	struct {
68 		struct sk_buff *skb;
69 	} sdq;
70 };
71 
72 struct mlxsw_pci_queue {
73 	spinlock_t lock; /* for queue accesses */
74 	struct mlxsw_pci_mem_item mem_item;
75 	struct mlxsw_pci_queue_elem_info *elem_info;
76 	u16 producer_counter;
77 	u16 consumer_counter;
78 	u16 count; /* number of elements in queue */
79 	u8 num; /* queue number */
80 	u8 elem_size; /* size of one element */
81 	enum mlxsw_pci_queue_type type;
82 	struct mlxsw_pci *pci;
83 	union {
84 		struct {
85 			enum mlxsw_pci_cqe_v v;
86 			struct mlxsw_pci_queue *dq;
87 			struct napi_struct napi;
88 			struct page_pool *page_pool;
89 		} cq;
90 		struct {
91 			struct tasklet_struct tasklet;
92 		} eq;
93 		struct {
94 			struct mlxsw_pci_queue *cq;
95 		} rdq;
96 	} u;
97 };
98 
99 struct mlxsw_pci_queue_type_group {
100 	struct mlxsw_pci_queue *q;
101 	u8 count; /* number of queues in group */
102 };
103 
104 struct mlxsw_pci {
105 	struct pci_dev *pdev;
106 	u8 __iomem *hw_addr;
107 	u64 free_running_clock_offset;
108 	u64 utc_sec_offset;
109 	u64 utc_nsec_offset;
110 	bool lag_mode_support;
111 	bool cff_support;
112 	enum mlxsw_cmd_mbox_config_profile_lag_mode lag_mode;
113 	enum mlxsw_cmd_mbox_config_profile_flood_mode flood_mode;
114 	struct mlxsw_pci_queue_type_group queues[MLXSW_PCI_QUEUE_TYPE_COUNT];
115 	u32 doorbell_offset;
116 	struct mlxsw_core *core;
117 	struct {
118 		struct mlxsw_pci_mem_item *items;
119 		unsigned int count;
120 	} fw_area;
121 	struct {
122 		struct mlxsw_pci_mem_item out_mbox;
123 		struct mlxsw_pci_mem_item in_mbox;
124 		struct mutex lock; /* Lock access to command registers */
125 		struct {
126 			u8 status;
127 			u64 out_param;
128 		} comp;
129 	} cmd;
130 	struct mlxsw_bus_info bus_info;
131 	const struct pci_device_id *id;
132 	enum mlxsw_pci_cqe_v max_cqe_ver; /* Maximal supported CQE version */
133 	u8 num_cqs; /* Number of CQs */
134 	u8 num_sdqs; /* Number of SDQs */
135 	bool skip_reset;
136 	struct net_device *napi_dev_tx;
137 	struct net_device *napi_dev_rx;
138 };
139 
140 static int mlxsw_pci_napi_devs_init(struct mlxsw_pci *mlxsw_pci)
141 {
142 	int err;
143 
144 	mlxsw_pci->napi_dev_tx = alloc_netdev_dummy(0);
145 	if (!mlxsw_pci->napi_dev_tx)
146 		return -ENOMEM;
147 	strscpy(mlxsw_pci->napi_dev_tx->name, "mlxsw_tx",
148 		sizeof(mlxsw_pci->napi_dev_tx->name));
149 
150 	mlxsw_pci->napi_dev_rx = alloc_netdev_dummy(0);
151 	if (!mlxsw_pci->napi_dev_rx) {
152 		err = -ENOMEM;
153 		goto err_alloc_rx;
154 	}
155 	strscpy(mlxsw_pci->napi_dev_rx->name, "mlxsw_rx",
156 		sizeof(mlxsw_pci->napi_dev_rx->name));
157 	dev_set_threaded(mlxsw_pci->napi_dev_rx, true);
158 
159 	return 0;
160 
161 err_alloc_rx:
162 	free_netdev(mlxsw_pci->napi_dev_tx);
163 	return err;
164 }
165 
166 static void mlxsw_pci_napi_devs_fini(struct mlxsw_pci *mlxsw_pci)
167 {
168 	free_netdev(mlxsw_pci->napi_dev_rx);
169 	free_netdev(mlxsw_pci->napi_dev_tx);
170 }
171 
172 static char *__mlxsw_pci_queue_elem_get(struct mlxsw_pci_queue *q,
173 					size_t elem_size, int elem_index)
174 {
175 	return q->mem_item.buf + (elem_size * elem_index);
176 }
177 
178 static struct mlxsw_pci_queue_elem_info *
179 mlxsw_pci_queue_elem_info_get(struct mlxsw_pci_queue *q, int elem_index)
180 {
181 	return &q->elem_info[elem_index];
182 }
183 
184 static struct mlxsw_pci_queue_elem_info *
185 mlxsw_pci_queue_elem_info_producer_get(struct mlxsw_pci_queue *q)
186 {
187 	int index = q->producer_counter & (q->count - 1);
188 
189 	if ((u16) (q->producer_counter - q->consumer_counter) == q->count)
190 		return NULL;
191 	return mlxsw_pci_queue_elem_info_get(q, index);
192 }
193 
194 static struct mlxsw_pci_queue_elem_info *
195 mlxsw_pci_queue_elem_info_consumer_get(struct mlxsw_pci_queue *q)
196 {
197 	int index = q->consumer_counter & (q->count - 1);
198 
199 	return mlxsw_pci_queue_elem_info_get(q, index);
200 }
201 
202 static char *mlxsw_pci_queue_elem_get(struct mlxsw_pci_queue *q, int elem_index)
203 {
204 	return mlxsw_pci_queue_elem_info_get(q, elem_index)->elem;
205 }
206 
207 static bool mlxsw_pci_elem_hw_owned(struct mlxsw_pci_queue *q, bool owner_bit)
208 {
209 	return owner_bit != !!(q->consumer_counter & q->count);
210 }
211 
212 static struct mlxsw_pci_queue_type_group *
213 mlxsw_pci_queue_type_group_get(struct mlxsw_pci *mlxsw_pci,
214 			       enum mlxsw_pci_queue_type q_type)
215 {
216 	return &mlxsw_pci->queues[q_type];
217 }
218 
219 static struct mlxsw_pci_queue *
220 __mlxsw_pci_queue_get(struct mlxsw_pci *mlxsw_pci,
221 		      enum mlxsw_pci_queue_type q_type, u8 q_num)
222 {
223 	return &mlxsw_pci->queues[q_type].q[q_num];
224 }
225 
226 static struct mlxsw_pci_queue *mlxsw_pci_sdq_get(struct mlxsw_pci *mlxsw_pci,
227 						 u8 q_num)
228 {
229 	return __mlxsw_pci_queue_get(mlxsw_pci,
230 				     MLXSW_PCI_QUEUE_TYPE_SDQ, q_num);
231 }
232 
233 static struct mlxsw_pci_queue *mlxsw_pci_cq_get(struct mlxsw_pci *mlxsw_pci,
234 						u8 q_num)
235 {
236 	return __mlxsw_pci_queue_get(mlxsw_pci, MLXSW_PCI_QUEUE_TYPE_CQ, q_num);
237 }
238 
239 static struct mlxsw_pci_queue *mlxsw_pci_eq_get(struct mlxsw_pci *mlxsw_pci)
240 {
241 	/* There is only one EQ at index 0. */
242 	return __mlxsw_pci_queue_get(mlxsw_pci, MLXSW_PCI_QUEUE_TYPE_EQ, 0);
243 }
244 
245 static void __mlxsw_pci_queue_doorbell_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_type_offset[q->type],
252 				   q->num), val);
253 }
254 
255 static void __mlxsw_pci_queue_doorbell_arm_set(struct mlxsw_pci *mlxsw_pci,
256 					       struct mlxsw_pci_queue *q,
257 					       u16 val)
258 {
259 	mlxsw_pci_write32(mlxsw_pci,
260 			  DOORBELL(mlxsw_pci->doorbell_offset,
261 				   mlxsw_pci_doorbell_arm_type_offset[q->type],
262 				   q->num), val);
263 }
264 
265 static void mlxsw_pci_queue_doorbell_producer_ring(struct mlxsw_pci *mlxsw_pci,
266 						   struct mlxsw_pci_queue *q)
267 {
268 	wmb(); /* ensure all writes are done before we ring a bell */
269 	__mlxsw_pci_queue_doorbell_set(mlxsw_pci, q, q->producer_counter);
270 }
271 
272 static void mlxsw_pci_queue_doorbell_consumer_ring(struct mlxsw_pci *mlxsw_pci,
273 						   struct mlxsw_pci_queue *q)
274 {
275 	wmb(); /* ensure all writes are done before we ring a bell */
276 	__mlxsw_pci_queue_doorbell_set(mlxsw_pci, q,
277 				       q->consumer_counter + q->count);
278 }
279 
280 static void
281 mlxsw_pci_queue_doorbell_arm_consumer_ring(struct mlxsw_pci *mlxsw_pci,
282 					   struct mlxsw_pci_queue *q)
283 {
284 	wmb(); /* ensure all writes are done before we ring a bell */
285 	__mlxsw_pci_queue_doorbell_arm_set(mlxsw_pci, q, q->consumer_counter);
286 }
287 
288 static dma_addr_t __mlxsw_pci_queue_page_get(struct mlxsw_pci_queue *q,
289 					     int page_index)
290 {
291 	return q->mem_item.mapaddr + MLXSW_PCI_PAGE_SIZE * page_index;
292 }
293 
294 static int mlxsw_pci_sdq_init(struct mlxsw_pci *mlxsw_pci, char *mbox,
295 			      struct mlxsw_pci_queue *q)
296 {
297 	struct mlxsw_pci_queue *cq;
298 	int tclass;
299 	u8 cq_num;
300 	int lp;
301 	int i;
302 	int err;
303 
304 	q->producer_counter = 0;
305 	q->consumer_counter = 0;
306 	tclass = q->num == MLXSW_PCI_SDQ_EMAD_INDEX ? MLXSW_PCI_SDQ_EMAD_TC :
307 						      MLXSW_PCI_SDQ_CTL_TC;
308 	lp = q->num == MLXSW_PCI_SDQ_EMAD_INDEX ? MLXSW_CMD_MBOX_SW2HW_DQ_SDQ_LP_IGNORE_WQE :
309 						  MLXSW_CMD_MBOX_SW2HW_DQ_SDQ_LP_WQE;
310 
311 	/* Set CQ of same number of this SDQ. */
312 	cq_num = q->num;
313 	mlxsw_cmd_mbox_sw2hw_dq_cq_set(mbox, cq_num);
314 	mlxsw_cmd_mbox_sw2hw_dq_sdq_lp_set(mbox, lp);
315 	mlxsw_cmd_mbox_sw2hw_dq_sdq_tclass_set(mbox, tclass);
316 	mlxsw_cmd_mbox_sw2hw_dq_log2_dq_sz_set(mbox, 3); /* 8 pages */
317 	for (i = 0; i < MLXSW_PCI_AQ_PAGES; i++) {
318 		dma_addr_t mapaddr = __mlxsw_pci_queue_page_get(q, i);
319 
320 		mlxsw_cmd_mbox_sw2hw_dq_pa_set(mbox, i, mapaddr);
321 	}
322 
323 	err = mlxsw_cmd_sw2hw_sdq(mlxsw_pci->core, mbox, q->num);
324 	if (err)
325 		return err;
326 
327 	cq = mlxsw_pci_cq_get(mlxsw_pci, cq_num);
328 	cq->u.cq.dq = q;
329 	mlxsw_pci_queue_doorbell_producer_ring(mlxsw_pci, q);
330 	return 0;
331 }
332 
333 static void mlxsw_pci_sdq_fini(struct mlxsw_pci *mlxsw_pci,
334 			       struct mlxsw_pci_queue *q)
335 {
336 	mlxsw_cmd_hw2sw_sdq(mlxsw_pci->core, q->num);
337 }
338 
339 #define MLXSW_PCI_SKB_HEADROOM (NET_SKB_PAD + NET_IP_ALIGN)
340 
341 #define MLXSW_PCI_RX_BUF_SW_OVERHEAD		\
342 		(MLXSW_PCI_SKB_HEADROOM +	\
343 		SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
344 
345 static void
346 mlxsw_pci_wqe_rx_frag_set(struct mlxsw_pci *mlxsw_pci, struct page *page,
347 			  char *wqe, int index, size_t frag_len)
348 {
349 	dma_addr_t mapaddr;
350 
351 	mapaddr = page_pool_get_dma_addr(page);
352 	mapaddr += MLXSW_PCI_SKB_HEADROOM;
353 
354 	mlxsw_pci_wqe_address_set(wqe, index, mapaddr);
355 	mlxsw_pci_wqe_byte_count_set(wqe, index, frag_len);
356 }
357 
358 static int mlxsw_pci_wqe_frag_map(struct mlxsw_pci *mlxsw_pci, char *wqe,
359 				  int index, char *frag_data, size_t frag_len,
360 				  int direction)
361 {
362 	struct pci_dev *pdev = mlxsw_pci->pdev;
363 	dma_addr_t mapaddr;
364 
365 	mapaddr = dma_map_single(&pdev->dev, frag_data, frag_len, direction);
366 	if (unlikely(dma_mapping_error(&pdev->dev, mapaddr))) {
367 		dev_err_ratelimited(&pdev->dev, "failed to dma map tx frag\n");
368 		return -EIO;
369 	}
370 	mlxsw_pci_wqe_address_set(wqe, index, mapaddr);
371 	mlxsw_pci_wqe_byte_count_set(wqe, index, frag_len);
372 	return 0;
373 }
374 
375 static void mlxsw_pci_wqe_frag_unmap(struct mlxsw_pci *mlxsw_pci, char *wqe,
376 				     int index, int direction)
377 {
378 	struct pci_dev *pdev = mlxsw_pci->pdev;
379 	size_t frag_len = mlxsw_pci_wqe_byte_count_get(wqe, index);
380 	dma_addr_t mapaddr = mlxsw_pci_wqe_address_get(wqe, index);
381 
382 	if (!frag_len)
383 		return;
384 	dma_unmap_single(&pdev->dev, mapaddr, frag_len, direction);
385 }
386 
387 static struct sk_buff *mlxsw_pci_rdq_build_skb(struct page *page,
388 					       u16 byte_count)
389 {
390 	void *data = page_address(page);
391 	unsigned int allocated_size;
392 	struct sk_buff *skb;
393 
394 	net_prefetch(data);
395 	allocated_size = page_size(page);
396 	skb = napi_build_skb(data, allocated_size);
397 	if (unlikely(!skb))
398 		return ERR_PTR(-ENOMEM);
399 
400 	skb_reserve(skb, MLXSW_PCI_SKB_HEADROOM);
401 	skb_put(skb, byte_count);
402 	return skb;
403 }
404 
405 static int mlxsw_pci_rdq_page_alloc(struct mlxsw_pci_queue *q,
406 				    struct mlxsw_pci_queue_elem_info *elem_info)
407 {
408 	struct mlxsw_pci_queue *cq = q->u.rdq.cq;
409 	size_t buf_len = MLXSW_PORT_MAX_MTU;
410 	char *wqe = elem_info->elem;
411 	struct page *page;
412 
413 	page = page_pool_dev_alloc_pages(cq->u.cq.page_pool);
414 	if (unlikely(!page))
415 		return -ENOMEM;
416 
417 	mlxsw_pci_wqe_rx_frag_set(q->pci, page, wqe, 0, buf_len);
418 	elem_info->page = page;
419 	return 0;
420 }
421 
422 static void mlxsw_pci_rdq_page_free(struct mlxsw_pci_queue *q,
423 				    struct mlxsw_pci_queue_elem_info *elem_info)
424 {
425 	struct mlxsw_pci_queue *cq = q->u.rdq.cq;
426 
427 	page_pool_put_page(cq->u.cq.page_pool, elem_info->page, -1, false);
428 }
429 
430 static int mlxsw_pci_rdq_init(struct mlxsw_pci *mlxsw_pci, char *mbox,
431 			      struct mlxsw_pci_queue *q)
432 {
433 	struct mlxsw_pci_queue_elem_info *elem_info;
434 	u8 sdq_count = mlxsw_pci->num_sdqs;
435 	struct mlxsw_pci_queue *cq;
436 	u8 cq_num;
437 	int i;
438 	int err;
439 
440 	q->producer_counter = 0;
441 	q->consumer_counter = 0;
442 
443 	/* Set CQ of same number of this RDQ with base
444 	 * above SDQ count as the lower ones are assigned to SDQs.
445 	 */
446 	cq_num = sdq_count + q->num;
447 	mlxsw_cmd_mbox_sw2hw_dq_cq_set(mbox, cq_num);
448 	mlxsw_cmd_mbox_sw2hw_dq_log2_dq_sz_set(mbox, 3); /* 8 pages */
449 	for (i = 0; i < MLXSW_PCI_AQ_PAGES; i++) {
450 		dma_addr_t mapaddr = __mlxsw_pci_queue_page_get(q, i);
451 
452 		mlxsw_cmd_mbox_sw2hw_dq_pa_set(mbox, i, mapaddr);
453 	}
454 
455 	err = mlxsw_cmd_sw2hw_rdq(mlxsw_pci->core, mbox, q->num);
456 	if (err)
457 		return err;
458 
459 	cq = mlxsw_pci_cq_get(mlxsw_pci, cq_num);
460 	cq->u.cq.dq = q;
461 	q->u.rdq.cq = cq;
462 
463 	mlxsw_pci_queue_doorbell_producer_ring(mlxsw_pci, q);
464 
465 	for (i = 0; i < q->count; i++) {
466 		elem_info = mlxsw_pci_queue_elem_info_producer_get(q);
467 		BUG_ON(!elem_info);
468 		err = mlxsw_pci_rdq_page_alloc(q, elem_info);
469 		if (err)
470 			goto rollback;
471 		/* Everything is set up, ring doorbell to pass elem to HW */
472 		q->producer_counter++;
473 		mlxsw_pci_queue_doorbell_producer_ring(mlxsw_pci, q);
474 	}
475 
476 	return 0;
477 
478 rollback:
479 	for (i--; i >= 0; i--) {
480 		elem_info = mlxsw_pci_queue_elem_info_get(q, i);
481 		mlxsw_pci_rdq_page_free(q, elem_info);
482 	}
483 	q->u.rdq.cq = NULL;
484 	cq->u.cq.dq = NULL;
485 	mlxsw_cmd_hw2sw_rdq(mlxsw_pci->core, q->num);
486 
487 	return err;
488 }
489 
490 static void mlxsw_pci_rdq_fini(struct mlxsw_pci *mlxsw_pci,
491 			       struct mlxsw_pci_queue *q)
492 {
493 	struct mlxsw_pci_queue_elem_info *elem_info;
494 	int i;
495 
496 	mlxsw_cmd_hw2sw_rdq(mlxsw_pci->core, q->num);
497 	for (i = 0; i < q->count; i++) {
498 		elem_info = mlxsw_pci_queue_elem_info_get(q, i);
499 		mlxsw_pci_rdq_page_free(q, elem_info);
500 	}
501 }
502 
503 static void mlxsw_pci_cq_pre_init(struct mlxsw_pci *mlxsw_pci,
504 				  struct mlxsw_pci_queue *q)
505 {
506 	q->u.cq.v = mlxsw_pci->max_cqe_ver;
507 
508 	if (q->u.cq.v == MLXSW_PCI_CQE_V2 &&
509 	    q->num < mlxsw_pci->num_sdqs &&
510 	    !mlxsw_core_sdq_supports_cqe_v2(mlxsw_pci->core))
511 		q->u.cq.v = MLXSW_PCI_CQE_V1;
512 }
513 
514 static unsigned int mlxsw_pci_read32_off(struct mlxsw_pci *mlxsw_pci,
515 					 ptrdiff_t off)
516 {
517 	return ioread32be(mlxsw_pci->hw_addr + off);
518 }
519 
520 static void mlxsw_pci_skb_cb_ts_set(struct mlxsw_pci *mlxsw_pci,
521 				    struct sk_buff *skb,
522 				    enum mlxsw_pci_cqe_v cqe_v, char *cqe)
523 {
524 	u8 ts_type;
525 
526 	if (cqe_v != MLXSW_PCI_CQE_V2)
527 		return;
528 
529 	ts_type = mlxsw_pci_cqe2_time_stamp_type_get(cqe);
530 
531 	if (ts_type != MLXSW_PCI_CQE_TIME_STAMP_TYPE_UTC &&
532 	    ts_type != MLXSW_PCI_CQE_TIME_STAMP_TYPE_MIRROR_UTC)
533 		return;
534 
535 	mlxsw_skb_cb(skb)->cqe_ts.sec = mlxsw_pci_cqe2_time_stamp_sec_get(cqe);
536 	mlxsw_skb_cb(skb)->cqe_ts.nsec =
537 		mlxsw_pci_cqe2_time_stamp_nsec_get(cqe);
538 }
539 
540 static void mlxsw_pci_cqe_sdq_handle(struct mlxsw_pci *mlxsw_pci,
541 				     struct mlxsw_pci_queue *q,
542 				     u16 consumer_counter_limit,
543 				     enum mlxsw_pci_cqe_v cqe_v,
544 				     char *cqe, int budget)
545 {
546 	struct pci_dev *pdev = mlxsw_pci->pdev;
547 	struct mlxsw_pci_queue_elem_info *elem_info;
548 	struct mlxsw_tx_info tx_info;
549 	char *wqe;
550 	struct sk_buff *skb;
551 	int i;
552 
553 	spin_lock(&q->lock);
554 	elem_info = mlxsw_pci_queue_elem_info_consumer_get(q);
555 	tx_info = mlxsw_skb_cb(elem_info->sdq.skb)->tx_info;
556 	skb = elem_info->sdq.skb;
557 	wqe = elem_info->elem;
558 	for (i = 0; i < MLXSW_PCI_WQE_SG_ENTRIES; i++)
559 		mlxsw_pci_wqe_frag_unmap(mlxsw_pci, wqe, i, DMA_TO_DEVICE);
560 
561 	if (unlikely(!tx_info.is_emad &&
562 		     skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP)) {
563 		mlxsw_pci_skb_cb_ts_set(mlxsw_pci, skb, cqe_v, cqe);
564 		mlxsw_core_ptp_transmitted(mlxsw_pci->core, skb,
565 					   tx_info.local_port);
566 		skb = NULL;
567 	}
568 
569 	if (skb)
570 		napi_consume_skb(skb, budget);
571 	elem_info->sdq.skb = NULL;
572 
573 	if (q->consumer_counter++ != consumer_counter_limit)
574 		dev_dbg_ratelimited(&pdev->dev, "Consumer counter does not match limit in SDQ\n");
575 	spin_unlock(&q->lock);
576 }
577 
578 static void mlxsw_pci_cqe_rdq_md_tx_port_init(struct sk_buff *skb,
579 					      const char *cqe)
580 {
581 	struct mlxsw_skb_cb *cb = mlxsw_skb_cb(skb);
582 
583 	if (mlxsw_pci_cqe2_tx_lag_get(cqe)) {
584 		cb->rx_md_info.tx_port_is_lag = true;
585 		cb->rx_md_info.tx_lag_id = mlxsw_pci_cqe2_tx_lag_id_get(cqe);
586 		cb->rx_md_info.tx_lag_port_index =
587 			mlxsw_pci_cqe2_tx_lag_subport_get(cqe);
588 	} else {
589 		cb->rx_md_info.tx_port_is_lag = false;
590 		cb->rx_md_info.tx_sys_port =
591 			mlxsw_pci_cqe2_tx_system_port_get(cqe);
592 	}
593 
594 	if (cb->rx_md_info.tx_sys_port != MLXSW_PCI_CQE2_TX_PORT_MULTI_PORT &&
595 	    cb->rx_md_info.tx_sys_port != MLXSW_PCI_CQE2_TX_PORT_INVALID)
596 		cb->rx_md_info.tx_port_valid = 1;
597 	else
598 		cb->rx_md_info.tx_port_valid = 0;
599 }
600 
601 static void mlxsw_pci_cqe_rdq_md_init(struct sk_buff *skb, const char *cqe)
602 {
603 	struct mlxsw_skb_cb *cb = mlxsw_skb_cb(skb);
604 
605 	cb->rx_md_info.tx_congestion = mlxsw_pci_cqe2_mirror_cong_get(cqe);
606 	if (cb->rx_md_info.tx_congestion != MLXSW_PCI_CQE2_MIRROR_CONG_INVALID)
607 		cb->rx_md_info.tx_congestion_valid = 1;
608 	else
609 		cb->rx_md_info.tx_congestion_valid = 0;
610 	cb->rx_md_info.tx_congestion <<= MLXSW_PCI_CQE2_MIRROR_CONG_SHIFT;
611 
612 	cb->rx_md_info.latency = mlxsw_pci_cqe2_mirror_latency_get(cqe);
613 	if (cb->rx_md_info.latency != MLXSW_PCI_CQE2_MIRROR_LATENCY_INVALID)
614 		cb->rx_md_info.latency_valid = 1;
615 	else
616 		cb->rx_md_info.latency_valid = 0;
617 
618 	cb->rx_md_info.tx_tc = mlxsw_pci_cqe2_mirror_tclass_get(cqe);
619 	if (cb->rx_md_info.tx_tc != MLXSW_PCI_CQE2_MIRROR_TCLASS_INVALID)
620 		cb->rx_md_info.tx_tc_valid = 1;
621 	else
622 		cb->rx_md_info.tx_tc_valid = 0;
623 
624 	mlxsw_pci_cqe_rdq_md_tx_port_init(skb, cqe);
625 }
626 
627 static void mlxsw_pci_cqe_rdq_handle(struct mlxsw_pci *mlxsw_pci,
628 				     struct mlxsw_pci_queue *q,
629 				     u16 consumer_counter_limit,
630 				     enum mlxsw_pci_cqe_v cqe_v, char *cqe)
631 {
632 	struct pci_dev *pdev = mlxsw_pci->pdev;
633 	struct mlxsw_pci_queue_elem_info *elem_info;
634 	struct mlxsw_pci_queue *cq = q->u.rdq.cq;
635 	struct mlxsw_rx_info rx_info = {};
636 	struct sk_buff *skb;
637 	struct page *page;
638 	u16 byte_count;
639 	int err;
640 
641 	elem_info = mlxsw_pci_queue_elem_info_consumer_get(q);
642 
643 	if (q->consumer_counter++ != consumer_counter_limit)
644 		dev_dbg_ratelimited(&pdev->dev, "Consumer counter does not match limit in RDQ\n");
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 
650 	page = elem_info->page;
651 
652 	err = mlxsw_pci_rdq_page_alloc(q, elem_info);
653 	if (err) {
654 		dev_err_ratelimited(&pdev->dev, "Failed to alloc page\n");
655 		goto out;
656 	}
657 
658 	skb = mlxsw_pci_rdq_build_skb(page, byte_count);
659 	if (IS_ERR(skb)) {
660 		dev_err_ratelimited(&pdev->dev, "Failed to build skb for RDQ\n");
661 		page_pool_recycle_direct(cq->u.cq.page_pool, page);
662 		goto out;
663 	}
664 
665 	skb_mark_for_recycle(skb);
666 
667 	if (mlxsw_pci_cqe_lag_get(cqe_v, cqe)) {
668 		rx_info.is_lag = true;
669 		rx_info.u.lag_id = mlxsw_pci_cqe_lag_id_get(cqe_v, cqe);
670 		rx_info.lag_port_index =
671 			mlxsw_pci_cqe_lag_subport_get(cqe_v, cqe);
672 	} else {
673 		rx_info.is_lag = false;
674 		rx_info.u.sys_port = mlxsw_pci_cqe_system_port_get(cqe);
675 	}
676 
677 	rx_info.trap_id = mlxsw_pci_cqe_trap_id_get(cqe);
678 
679 	if (rx_info.trap_id == MLXSW_TRAP_ID_DISCARD_INGRESS_ACL ||
680 	    rx_info.trap_id == MLXSW_TRAP_ID_DISCARD_EGRESS_ACL) {
681 		u32 cookie_index = 0;
682 
683 		if (mlxsw_pci->max_cqe_ver >= MLXSW_PCI_CQE_V2)
684 			cookie_index = mlxsw_pci_cqe2_user_def_val_orig_pkt_len_get(cqe);
685 		mlxsw_skb_cb(skb)->rx_md_info.cookie_index = cookie_index;
686 	} else if (rx_info.trap_id >= MLXSW_TRAP_ID_MIRROR_SESSION0 &&
687 		   rx_info.trap_id <= MLXSW_TRAP_ID_MIRROR_SESSION7 &&
688 		   mlxsw_pci->max_cqe_ver >= MLXSW_PCI_CQE_V2) {
689 		rx_info.mirror_reason = mlxsw_pci_cqe2_mirror_reason_get(cqe);
690 		mlxsw_pci_cqe_rdq_md_init(skb, cqe);
691 	} else if (rx_info.trap_id == MLXSW_TRAP_ID_PKT_SAMPLE &&
692 		   mlxsw_pci->max_cqe_ver >= MLXSW_PCI_CQE_V2) {
693 		mlxsw_pci_cqe_rdq_md_tx_port_init(skb, cqe);
694 	}
695 
696 	mlxsw_pci_skb_cb_ts_set(mlxsw_pci, skb, cqe_v, cqe);
697 
698 	mlxsw_core_skb_receive(mlxsw_pci->core, skb, &rx_info);
699 
700 out:
701 	q->producer_counter++;
702 	return;
703 }
704 
705 static char *mlxsw_pci_cq_sw_cqe_get(struct mlxsw_pci_queue *q)
706 {
707 	struct mlxsw_pci_queue_elem_info *elem_info;
708 	char *elem;
709 	bool owner_bit;
710 
711 	elem_info = mlxsw_pci_queue_elem_info_consumer_get(q);
712 	elem = elem_info->elem;
713 	owner_bit = mlxsw_pci_cqe_owner_get(q->u.cq.v, elem);
714 	if (mlxsw_pci_elem_hw_owned(q, owner_bit))
715 		return NULL;
716 	q->consumer_counter++;
717 	rmb(); /* make sure we read owned bit before the rest of elem */
718 	return elem;
719 }
720 
721 static bool mlxsw_pci_cq_cqe_to_handle(struct mlxsw_pci_queue *q)
722 {
723 	struct mlxsw_pci_queue_elem_info *elem_info;
724 	bool owner_bit;
725 
726 	elem_info = mlxsw_pci_queue_elem_info_consumer_get(q);
727 	owner_bit = mlxsw_pci_cqe_owner_get(q->u.cq.v, elem_info->elem);
728 	return !mlxsw_pci_elem_hw_owned(q, owner_bit);
729 }
730 
731 static int mlxsw_pci_napi_poll_cq_rx(struct napi_struct *napi, int budget)
732 {
733 	struct mlxsw_pci_queue *q = container_of(napi, struct mlxsw_pci_queue,
734 						 u.cq.napi);
735 	struct mlxsw_pci_queue *rdq = q->u.cq.dq;
736 	struct mlxsw_pci *mlxsw_pci = q->pci;
737 	int work_done = 0;
738 	char *cqe;
739 
740 	/* If the budget is 0, Rx processing should be skipped. */
741 	if (unlikely(!budget))
742 		return 0;
743 
744 	while ((cqe = mlxsw_pci_cq_sw_cqe_get(q))) {
745 		u16 wqe_counter = mlxsw_pci_cqe_wqe_counter_get(cqe);
746 		u8 sendq = mlxsw_pci_cqe_sr_get(q->u.cq.v, cqe);
747 		u8 dqn = mlxsw_pci_cqe_dqn_get(q->u.cq.v, cqe);
748 
749 		if (unlikely(sendq)) {
750 			WARN_ON_ONCE(1);
751 			continue;
752 		}
753 
754 		if (unlikely(dqn != rdq->num)) {
755 			WARN_ON_ONCE(1);
756 			continue;
757 		}
758 
759 		mlxsw_pci_cqe_rdq_handle(mlxsw_pci, rdq,
760 					 wqe_counter, q->u.cq.v, cqe);
761 
762 		if (++work_done == budget)
763 			break;
764 	}
765 
766 	mlxsw_pci_queue_doorbell_consumer_ring(mlxsw_pci, q);
767 	mlxsw_pci_queue_doorbell_producer_ring(mlxsw_pci, rdq);
768 
769 	if (work_done < budget)
770 		goto processing_completed;
771 
772 	/* The driver still has outstanding work to do, budget was exhausted.
773 	 * Return exactly budget. In that case, the NAPI instance will be polled
774 	 * again.
775 	 */
776 	if (mlxsw_pci_cq_cqe_to_handle(q))
777 		goto out;
778 
779 	/* The driver processed all the completions and handled exactly
780 	 * 'budget'. Return 'budget - 1' to distinguish from the case that
781 	 * driver still has completions to handle.
782 	 */
783 	if (work_done == budget)
784 		work_done--;
785 
786 processing_completed:
787 	if (napi_complete_done(napi, work_done))
788 		mlxsw_pci_queue_doorbell_arm_consumer_ring(mlxsw_pci, q);
789 out:
790 	return work_done;
791 }
792 
793 static int mlxsw_pci_napi_poll_cq_tx(struct napi_struct *napi, int budget)
794 {
795 	struct mlxsw_pci_queue *q = container_of(napi, struct mlxsw_pci_queue,
796 						 u.cq.napi);
797 	struct mlxsw_pci_queue *sdq = q->u.cq.dq;
798 	struct mlxsw_pci *mlxsw_pci = q->pci;
799 	int work_done = 0;
800 	char *cqe;
801 
802 	while ((cqe = mlxsw_pci_cq_sw_cqe_get(q))) {
803 		u16 wqe_counter = mlxsw_pci_cqe_wqe_counter_get(cqe);
804 		u8 sendq = mlxsw_pci_cqe_sr_get(q->u.cq.v, cqe);
805 		u8 dqn = mlxsw_pci_cqe_dqn_get(q->u.cq.v, cqe);
806 		char ncqe[MLXSW_PCI_CQE_SIZE_MAX];
807 
808 		if (unlikely(!sendq)) {
809 			WARN_ON_ONCE(1);
810 			continue;
811 		}
812 
813 		if (unlikely(dqn != sdq->num)) {
814 			WARN_ON_ONCE(1);
815 			continue;
816 		}
817 
818 		memcpy(ncqe, cqe, q->elem_size);
819 		mlxsw_pci_queue_doorbell_consumer_ring(mlxsw_pci, q);
820 
821 		mlxsw_pci_cqe_sdq_handle(mlxsw_pci, sdq,
822 					 wqe_counter, q->u.cq.v, ncqe, budget);
823 
824 		work_done++;
825 	}
826 
827 	/* If the budget is 0 napi_complete_done() should never be called. */
828 	if (unlikely(!budget))
829 		goto processing_completed;
830 
831 	work_done = min(work_done, budget - 1);
832 	if (unlikely(!napi_complete_done(napi, work_done)))
833 		goto out;
834 
835 processing_completed:
836 	mlxsw_pci_queue_doorbell_arm_consumer_ring(mlxsw_pci, q);
837 out:
838 	return work_done;
839 }
840 
841 static enum mlxsw_pci_cq_type
842 mlxsw_pci_cq_type(const struct mlxsw_pci *mlxsw_pci,
843 		  const struct mlxsw_pci_queue *q)
844 {
845 	/* Each CQ is mapped to one DQ. The first 'num_sdqs' queues are used
846 	 * for SDQs and the rest are used for RDQs.
847 	 */
848 	if (q->num < mlxsw_pci->num_sdqs)
849 		return MLXSW_PCI_CQ_SDQ;
850 
851 	return MLXSW_PCI_CQ_RDQ;
852 }
853 
854 static void mlxsw_pci_cq_napi_setup(struct mlxsw_pci_queue *q,
855 				    enum mlxsw_pci_cq_type cq_type)
856 {
857 	struct mlxsw_pci *mlxsw_pci = q->pci;
858 
859 	switch (cq_type) {
860 	case MLXSW_PCI_CQ_SDQ:
861 		netif_napi_add(mlxsw_pci->napi_dev_tx, &q->u.cq.napi,
862 			       mlxsw_pci_napi_poll_cq_tx);
863 		break;
864 	case MLXSW_PCI_CQ_RDQ:
865 		netif_napi_add(mlxsw_pci->napi_dev_rx, &q->u.cq.napi,
866 			       mlxsw_pci_napi_poll_cq_rx);
867 		break;
868 	}
869 }
870 
871 static void mlxsw_pci_cq_napi_teardown(struct mlxsw_pci_queue *q)
872 {
873 	netif_napi_del(&q->u.cq.napi);
874 }
875 
876 static int mlxsw_pci_cq_page_pool_init(struct mlxsw_pci_queue *q,
877 				       enum mlxsw_pci_cq_type cq_type)
878 {
879 	struct page_pool_params pp_params = {};
880 	struct mlxsw_pci *mlxsw_pci = q->pci;
881 	struct page_pool *page_pool;
882 	u32 max_pkt_size;
883 
884 	if (cq_type != MLXSW_PCI_CQ_RDQ)
885 		return 0;
886 
887 	max_pkt_size = MLXSW_PORT_MAX_MTU + MLXSW_PCI_RX_BUF_SW_OVERHEAD;
888 	pp_params.order = get_order(max_pkt_size);
889 	pp_params.flags = PP_FLAG_DMA_MAP;
890 	pp_params.pool_size = MLXSW_PCI_WQE_COUNT;
891 	pp_params.nid = dev_to_node(&mlxsw_pci->pdev->dev);
892 	pp_params.dev = &mlxsw_pci->pdev->dev;
893 	pp_params.napi = &q->u.cq.napi;
894 	pp_params.dma_dir = DMA_FROM_DEVICE;
895 
896 	page_pool = page_pool_create(&pp_params);
897 	if (IS_ERR(page_pool))
898 		return PTR_ERR(page_pool);
899 
900 	q->u.cq.page_pool = page_pool;
901 	return 0;
902 }
903 
904 static void mlxsw_pci_cq_page_pool_fini(struct mlxsw_pci_queue *q,
905 					enum mlxsw_pci_cq_type cq_type)
906 {
907 	if (cq_type != MLXSW_PCI_CQ_RDQ)
908 		return;
909 
910 	page_pool_destroy(q->u.cq.page_pool);
911 }
912 
913 static int mlxsw_pci_cq_init(struct mlxsw_pci *mlxsw_pci, char *mbox,
914 			     struct mlxsw_pci_queue *q)
915 {
916 	enum mlxsw_pci_cq_type cq_type = mlxsw_pci_cq_type(mlxsw_pci, q);
917 	int i;
918 	int err;
919 
920 	q->consumer_counter = 0;
921 
922 	for (i = 0; i < q->count; i++) {
923 		char *elem = mlxsw_pci_queue_elem_get(q, i);
924 
925 		mlxsw_pci_cqe_owner_set(q->u.cq.v, elem, 1);
926 	}
927 
928 	if (q->u.cq.v == MLXSW_PCI_CQE_V1)
929 		mlxsw_cmd_mbox_sw2hw_cq_cqe_ver_set(mbox,
930 				MLXSW_CMD_MBOX_SW2HW_CQ_CQE_VER_1);
931 	else if (q->u.cq.v == MLXSW_PCI_CQE_V2)
932 		mlxsw_cmd_mbox_sw2hw_cq_cqe_ver_set(mbox,
933 				MLXSW_CMD_MBOX_SW2HW_CQ_CQE_VER_2);
934 
935 	mlxsw_cmd_mbox_sw2hw_cq_c_eqn_set(mbox, MLXSW_PCI_EQ_COMP_NUM);
936 	mlxsw_cmd_mbox_sw2hw_cq_st_set(mbox, 0);
937 	mlxsw_cmd_mbox_sw2hw_cq_log_cq_size_set(mbox, ilog2(q->count));
938 	for (i = 0; i < MLXSW_PCI_AQ_PAGES; i++) {
939 		dma_addr_t mapaddr = __mlxsw_pci_queue_page_get(q, i);
940 
941 		mlxsw_cmd_mbox_sw2hw_cq_pa_set(mbox, i, mapaddr);
942 	}
943 	err = mlxsw_cmd_sw2hw_cq(mlxsw_pci->core, mbox, q->num);
944 	if (err)
945 		return err;
946 	mlxsw_pci_cq_napi_setup(q, cq_type);
947 
948 	err = mlxsw_pci_cq_page_pool_init(q, cq_type);
949 	if (err)
950 		goto err_page_pool_init;
951 
952 	napi_enable(&q->u.cq.napi);
953 	mlxsw_pci_queue_doorbell_consumer_ring(mlxsw_pci, q);
954 	mlxsw_pci_queue_doorbell_arm_consumer_ring(mlxsw_pci, q);
955 	return 0;
956 
957 err_page_pool_init:
958 	mlxsw_pci_cq_napi_teardown(q);
959 	return err;
960 }
961 
962 static void mlxsw_pci_cq_fini(struct mlxsw_pci *mlxsw_pci,
963 			      struct mlxsw_pci_queue *q)
964 {
965 	enum mlxsw_pci_cq_type cq_type = mlxsw_pci_cq_type(mlxsw_pci, q);
966 
967 	napi_disable(&q->u.cq.napi);
968 	mlxsw_pci_cq_page_pool_fini(q, cq_type);
969 	mlxsw_pci_cq_napi_teardown(q);
970 	mlxsw_cmd_hw2sw_cq(mlxsw_pci->core, q->num);
971 }
972 
973 static u16 mlxsw_pci_cq_elem_count(const struct mlxsw_pci_queue *q)
974 {
975 	return q->u.cq.v == MLXSW_PCI_CQE_V2 ? MLXSW_PCI_CQE2_COUNT :
976 					     MLXSW_PCI_CQE01_COUNT;
977 }
978 
979 static u8 mlxsw_pci_cq_elem_size(const struct mlxsw_pci_queue *q)
980 {
981 	return q->u.cq.v == MLXSW_PCI_CQE_V2 ? MLXSW_PCI_CQE2_SIZE :
982 					       MLXSW_PCI_CQE01_SIZE;
983 }
984 
985 static char *mlxsw_pci_eq_sw_eqe_get(struct mlxsw_pci_queue *q)
986 {
987 	struct mlxsw_pci_queue_elem_info *elem_info;
988 	char *elem;
989 	bool owner_bit;
990 
991 	elem_info = mlxsw_pci_queue_elem_info_consumer_get(q);
992 	elem = elem_info->elem;
993 	owner_bit = mlxsw_pci_eqe_owner_get(elem);
994 	if (mlxsw_pci_elem_hw_owned(q, owner_bit))
995 		return NULL;
996 	q->consumer_counter++;
997 	rmb(); /* make sure we read owned bit before the rest of elem */
998 	return elem;
999 }
1000 
1001 static void mlxsw_pci_eq_tasklet(struct tasklet_struct *t)
1002 {
1003 	unsigned long active_cqns[BITS_TO_LONGS(MLXSW_PCI_CQS_MAX)];
1004 	struct mlxsw_pci_queue *q = from_tasklet(q, t, u.eq.tasklet);
1005 	struct mlxsw_pci *mlxsw_pci = q->pci;
1006 	int credits = q->count >> 1;
1007 	u8 cqn, cq_count;
1008 	int items = 0;
1009 	char *eqe;
1010 
1011 	memset(&active_cqns, 0, sizeof(active_cqns));
1012 
1013 	while ((eqe = mlxsw_pci_eq_sw_eqe_get(q))) {
1014 		cqn = mlxsw_pci_eqe_cqn_get(eqe);
1015 		set_bit(cqn, active_cqns);
1016 
1017 		if (++items == credits)
1018 			break;
1019 	}
1020 
1021 	if (!items)
1022 		return;
1023 
1024 	mlxsw_pci_queue_doorbell_consumer_ring(mlxsw_pci, q);
1025 	mlxsw_pci_queue_doorbell_arm_consumer_ring(mlxsw_pci, q);
1026 
1027 	cq_count = mlxsw_pci->num_cqs;
1028 	for_each_set_bit(cqn, active_cqns, cq_count) {
1029 		q = mlxsw_pci_cq_get(mlxsw_pci, cqn);
1030 		napi_schedule(&q->u.cq.napi);
1031 	}
1032 }
1033 
1034 static int mlxsw_pci_eq_init(struct mlxsw_pci *mlxsw_pci, char *mbox,
1035 			     struct mlxsw_pci_queue *q)
1036 {
1037 	int i;
1038 	int err;
1039 
1040 	/* We expect to initialize only one EQ, which gets num=0 as it is
1041 	 * located at index zero. We use the EQ as EQ1, so set the number for
1042 	 * future use.
1043 	 */
1044 	WARN_ON_ONCE(q->num);
1045 	q->num = MLXSW_PCI_EQ_COMP_NUM;
1046 
1047 	q->consumer_counter = 0;
1048 
1049 	for (i = 0; i < q->count; i++) {
1050 		char *elem = mlxsw_pci_queue_elem_get(q, i);
1051 
1052 		mlxsw_pci_eqe_owner_set(elem, 1);
1053 	}
1054 
1055 	mlxsw_cmd_mbox_sw2hw_eq_int_msix_set(mbox, 1); /* MSI-X used */
1056 	mlxsw_cmd_mbox_sw2hw_eq_st_set(mbox, 1); /* armed */
1057 	mlxsw_cmd_mbox_sw2hw_eq_log_eq_size_set(mbox, ilog2(q->count));
1058 	for (i = 0; i < MLXSW_PCI_AQ_PAGES; i++) {
1059 		dma_addr_t mapaddr = __mlxsw_pci_queue_page_get(q, i);
1060 
1061 		mlxsw_cmd_mbox_sw2hw_eq_pa_set(mbox, i, mapaddr);
1062 	}
1063 	err = mlxsw_cmd_sw2hw_eq(mlxsw_pci->core, mbox, q->num);
1064 	if (err)
1065 		return err;
1066 	tasklet_setup(&q->u.eq.tasklet, mlxsw_pci_eq_tasklet);
1067 	mlxsw_pci_queue_doorbell_consumer_ring(mlxsw_pci, q);
1068 	mlxsw_pci_queue_doorbell_arm_consumer_ring(mlxsw_pci, q);
1069 	return 0;
1070 }
1071 
1072 static void mlxsw_pci_eq_fini(struct mlxsw_pci *mlxsw_pci,
1073 			      struct mlxsw_pci_queue *q)
1074 {
1075 	mlxsw_cmd_hw2sw_eq(mlxsw_pci->core, q->num);
1076 }
1077 
1078 struct mlxsw_pci_queue_ops {
1079 	const char *name;
1080 	enum mlxsw_pci_queue_type type;
1081 	void (*pre_init)(struct mlxsw_pci *mlxsw_pci,
1082 			 struct mlxsw_pci_queue *q);
1083 	int (*init)(struct mlxsw_pci *mlxsw_pci, char *mbox,
1084 		    struct mlxsw_pci_queue *q);
1085 	void (*fini)(struct mlxsw_pci *mlxsw_pci,
1086 		     struct mlxsw_pci_queue *q);
1087 	u16 (*elem_count_f)(const struct mlxsw_pci_queue *q);
1088 	u8 (*elem_size_f)(const struct mlxsw_pci_queue *q);
1089 	u16 elem_count;
1090 	u8 elem_size;
1091 };
1092 
1093 static const struct mlxsw_pci_queue_ops mlxsw_pci_sdq_ops = {
1094 	.type		= MLXSW_PCI_QUEUE_TYPE_SDQ,
1095 	.init		= mlxsw_pci_sdq_init,
1096 	.fini		= mlxsw_pci_sdq_fini,
1097 	.elem_count	= MLXSW_PCI_WQE_COUNT,
1098 	.elem_size	= MLXSW_PCI_WQE_SIZE,
1099 };
1100 
1101 static const struct mlxsw_pci_queue_ops mlxsw_pci_rdq_ops = {
1102 	.type		= MLXSW_PCI_QUEUE_TYPE_RDQ,
1103 	.init		= mlxsw_pci_rdq_init,
1104 	.fini		= mlxsw_pci_rdq_fini,
1105 	.elem_count	= MLXSW_PCI_WQE_COUNT,
1106 	.elem_size	= MLXSW_PCI_WQE_SIZE
1107 };
1108 
1109 static const struct mlxsw_pci_queue_ops mlxsw_pci_cq_ops = {
1110 	.type		= MLXSW_PCI_QUEUE_TYPE_CQ,
1111 	.pre_init	= mlxsw_pci_cq_pre_init,
1112 	.init		= mlxsw_pci_cq_init,
1113 	.fini		= mlxsw_pci_cq_fini,
1114 	.elem_count_f	= mlxsw_pci_cq_elem_count,
1115 	.elem_size_f	= mlxsw_pci_cq_elem_size
1116 };
1117 
1118 static const struct mlxsw_pci_queue_ops mlxsw_pci_eq_ops = {
1119 	.type		= MLXSW_PCI_QUEUE_TYPE_EQ,
1120 	.init		= mlxsw_pci_eq_init,
1121 	.fini		= mlxsw_pci_eq_fini,
1122 	.elem_count	= MLXSW_PCI_EQE_COUNT,
1123 	.elem_size	= MLXSW_PCI_EQE_SIZE
1124 };
1125 
1126 static int mlxsw_pci_queue_init(struct mlxsw_pci *mlxsw_pci, char *mbox,
1127 				const struct mlxsw_pci_queue_ops *q_ops,
1128 				struct mlxsw_pci_queue *q, u8 q_num)
1129 {
1130 	struct mlxsw_pci_mem_item *mem_item = &q->mem_item;
1131 	int i;
1132 	int err;
1133 
1134 	q->num = q_num;
1135 	if (q_ops->pre_init)
1136 		q_ops->pre_init(mlxsw_pci, q);
1137 
1138 	spin_lock_init(&q->lock);
1139 	q->count = q_ops->elem_count_f ? q_ops->elem_count_f(q) :
1140 					 q_ops->elem_count;
1141 	q->elem_size = q_ops->elem_size_f ? q_ops->elem_size_f(q) :
1142 					    q_ops->elem_size;
1143 	q->type = q_ops->type;
1144 	q->pci = mlxsw_pci;
1145 
1146 	mem_item->size = MLXSW_PCI_AQ_SIZE;
1147 	mem_item->buf = dma_alloc_coherent(&mlxsw_pci->pdev->dev,
1148 					   mem_item->size, &mem_item->mapaddr,
1149 					   GFP_KERNEL);
1150 	if (!mem_item->buf)
1151 		return -ENOMEM;
1152 
1153 	q->elem_info = kcalloc(q->count, sizeof(*q->elem_info), GFP_KERNEL);
1154 	if (!q->elem_info) {
1155 		err = -ENOMEM;
1156 		goto err_elem_info_alloc;
1157 	}
1158 
1159 	/* Initialize dma mapped elements info elem_info for
1160 	 * future easy access.
1161 	 */
1162 	for (i = 0; i < q->count; i++) {
1163 		struct mlxsw_pci_queue_elem_info *elem_info;
1164 
1165 		elem_info = mlxsw_pci_queue_elem_info_get(q, i);
1166 		elem_info->elem =
1167 			__mlxsw_pci_queue_elem_get(q, q->elem_size, i);
1168 	}
1169 
1170 	mlxsw_cmd_mbox_zero(mbox);
1171 	err = q_ops->init(mlxsw_pci, mbox, q);
1172 	if (err)
1173 		goto err_q_ops_init;
1174 	return 0;
1175 
1176 err_q_ops_init:
1177 	kfree(q->elem_info);
1178 err_elem_info_alloc:
1179 	dma_free_coherent(&mlxsw_pci->pdev->dev, mem_item->size,
1180 			  mem_item->buf, mem_item->mapaddr);
1181 	return err;
1182 }
1183 
1184 static void mlxsw_pci_queue_fini(struct mlxsw_pci *mlxsw_pci,
1185 				 const struct mlxsw_pci_queue_ops *q_ops,
1186 				 struct mlxsw_pci_queue *q)
1187 {
1188 	struct mlxsw_pci_mem_item *mem_item = &q->mem_item;
1189 
1190 	q_ops->fini(mlxsw_pci, q);
1191 	kfree(q->elem_info);
1192 	dma_free_coherent(&mlxsw_pci->pdev->dev, mem_item->size,
1193 			  mem_item->buf, mem_item->mapaddr);
1194 }
1195 
1196 static int mlxsw_pci_queue_group_init(struct mlxsw_pci *mlxsw_pci, char *mbox,
1197 				      const struct mlxsw_pci_queue_ops *q_ops,
1198 				      u8 num_qs)
1199 {
1200 	struct mlxsw_pci_queue_type_group *queue_group;
1201 	int i;
1202 	int err;
1203 
1204 	queue_group = mlxsw_pci_queue_type_group_get(mlxsw_pci, q_ops->type);
1205 	queue_group->q = kcalloc(num_qs, sizeof(*queue_group->q), GFP_KERNEL);
1206 	if (!queue_group->q)
1207 		return -ENOMEM;
1208 
1209 	for (i = 0; i < num_qs; i++) {
1210 		err = mlxsw_pci_queue_init(mlxsw_pci, mbox, q_ops,
1211 					   &queue_group->q[i], i);
1212 		if (err)
1213 			goto err_queue_init;
1214 	}
1215 	queue_group->count = num_qs;
1216 
1217 	return 0;
1218 
1219 err_queue_init:
1220 	for (i--; i >= 0; i--)
1221 		mlxsw_pci_queue_fini(mlxsw_pci, q_ops, &queue_group->q[i]);
1222 	kfree(queue_group->q);
1223 	return err;
1224 }
1225 
1226 static void mlxsw_pci_queue_group_fini(struct mlxsw_pci *mlxsw_pci,
1227 				       const struct mlxsw_pci_queue_ops *q_ops)
1228 {
1229 	struct mlxsw_pci_queue_type_group *queue_group;
1230 	int i;
1231 
1232 	queue_group = mlxsw_pci_queue_type_group_get(mlxsw_pci, q_ops->type);
1233 	for (i = 0; i < queue_group->count; i++)
1234 		mlxsw_pci_queue_fini(mlxsw_pci, q_ops, &queue_group->q[i]);
1235 	kfree(queue_group->q);
1236 }
1237 
1238 static int mlxsw_pci_aqs_init(struct mlxsw_pci *mlxsw_pci, char *mbox)
1239 {
1240 	struct pci_dev *pdev = mlxsw_pci->pdev;
1241 	u8 num_sdqs;
1242 	u8 sdq_log2sz;
1243 	u8 num_rdqs;
1244 	u8 rdq_log2sz;
1245 	u8 num_cqs;
1246 	u8 cq_log2sz;
1247 	u8 cqv2_log2sz;
1248 	u8 num_eqs;
1249 	u8 eq_log2sz;
1250 	int err;
1251 
1252 	mlxsw_cmd_mbox_zero(mbox);
1253 	err = mlxsw_cmd_query_aq_cap(mlxsw_pci->core, mbox);
1254 	if (err)
1255 		return err;
1256 
1257 	num_sdqs = mlxsw_cmd_mbox_query_aq_cap_max_num_sdqs_get(mbox);
1258 	sdq_log2sz = mlxsw_cmd_mbox_query_aq_cap_log_max_sdq_sz_get(mbox);
1259 	num_rdqs = mlxsw_cmd_mbox_query_aq_cap_max_num_rdqs_get(mbox);
1260 	rdq_log2sz = mlxsw_cmd_mbox_query_aq_cap_log_max_rdq_sz_get(mbox);
1261 	num_cqs = mlxsw_cmd_mbox_query_aq_cap_max_num_cqs_get(mbox);
1262 	cq_log2sz = mlxsw_cmd_mbox_query_aq_cap_log_max_cq_sz_get(mbox);
1263 	cqv2_log2sz = mlxsw_cmd_mbox_query_aq_cap_log_max_cqv2_sz_get(mbox);
1264 	num_eqs = mlxsw_cmd_mbox_query_aq_cap_max_num_eqs_get(mbox);
1265 	eq_log2sz = mlxsw_cmd_mbox_query_aq_cap_log_max_eq_sz_get(mbox);
1266 
1267 	if (num_sdqs + num_rdqs > num_cqs ||
1268 	    num_sdqs < MLXSW_PCI_SDQS_MIN ||
1269 	    num_cqs > MLXSW_PCI_CQS_MAX || num_eqs != MLXSW_PCI_EQS_MAX) {
1270 		dev_err(&pdev->dev, "Unsupported number of queues\n");
1271 		return -EINVAL;
1272 	}
1273 
1274 	if ((1 << sdq_log2sz != MLXSW_PCI_WQE_COUNT) ||
1275 	    (1 << rdq_log2sz != MLXSW_PCI_WQE_COUNT) ||
1276 	    (1 << cq_log2sz != MLXSW_PCI_CQE01_COUNT) ||
1277 	    (mlxsw_pci->max_cqe_ver == MLXSW_PCI_CQE_V2 &&
1278 	     (1 << cqv2_log2sz != MLXSW_PCI_CQE2_COUNT)) ||
1279 	    (1 << eq_log2sz != MLXSW_PCI_EQE_COUNT)) {
1280 		dev_err(&pdev->dev, "Unsupported number of async queue descriptors\n");
1281 		return -EINVAL;
1282 	}
1283 
1284 	mlxsw_pci->num_cqs = num_cqs;
1285 	mlxsw_pci->num_sdqs = num_sdqs;
1286 
1287 	err = mlxsw_pci_queue_group_init(mlxsw_pci, mbox, &mlxsw_pci_eq_ops,
1288 					 MLXSW_PCI_EQS_COUNT);
1289 	if (err) {
1290 		dev_err(&pdev->dev, "Failed to initialize event queues\n");
1291 		return err;
1292 	}
1293 
1294 	err = mlxsw_pci_queue_group_init(mlxsw_pci, mbox, &mlxsw_pci_cq_ops,
1295 					 num_cqs);
1296 	if (err) {
1297 		dev_err(&pdev->dev, "Failed to initialize completion queues\n");
1298 		goto err_cqs_init;
1299 	}
1300 
1301 	err = mlxsw_pci_queue_group_init(mlxsw_pci, mbox, &mlxsw_pci_sdq_ops,
1302 					 num_sdqs);
1303 	if (err) {
1304 		dev_err(&pdev->dev, "Failed to initialize send descriptor queues\n");
1305 		goto err_sdqs_init;
1306 	}
1307 
1308 	err = mlxsw_pci_queue_group_init(mlxsw_pci, mbox, &mlxsw_pci_rdq_ops,
1309 					 num_rdqs);
1310 	if (err) {
1311 		dev_err(&pdev->dev, "Failed to initialize receive descriptor queues\n");
1312 		goto err_rdqs_init;
1313 	}
1314 
1315 	return 0;
1316 
1317 err_rdqs_init:
1318 	mlxsw_pci_queue_group_fini(mlxsw_pci, &mlxsw_pci_sdq_ops);
1319 err_sdqs_init:
1320 	mlxsw_pci_queue_group_fini(mlxsw_pci, &mlxsw_pci_cq_ops);
1321 err_cqs_init:
1322 	mlxsw_pci_queue_group_fini(mlxsw_pci, &mlxsw_pci_eq_ops);
1323 	return err;
1324 }
1325 
1326 static void mlxsw_pci_aqs_fini(struct mlxsw_pci *mlxsw_pci)
1327 {
1328 	mlxsw_pci_queue_group_fini(mlxsw_pci, &mlxsw_pci_rdq_ops);
1329 	mlxsw_pci_queue_group_fini(mlxsw_pci, &mlxsw_pci_sdq_ops);
1330 	mlxsw_pci_queue_group_fini(mlxsw_pci, &mlxsw_pci_cq_ops);
1331 	mlxsw_pci_queue_group_fini(mlxsw_pci, &mlxsw_pci_eq_ops);
1332 }
1333 
1334 static void
1335 mlxsw_pci_config_profile_swid_config(struct mlxsw_pci *mlxsw_pci,
1336 				     char *mbox, int index,
1337 				     const struct mlxsw_swid_config *swid)
1338 {
1339 	u8 mask = 0;
1340 
1341 	if (swid->used_type) {
1342 		mlxsw_cmd_mbox_config_profile_swid_config_type_set(
1343 			mbox, index, swid->type);
1344 		mask |= 1;
1345 	}
1346 	if (swid->used_properties) {
1347 		mlxsw_cmd_mbox_config_profile_swid_config_properties_set(
1348 			mbox, index, swid->properties);
1349 		mask |= 2;
1350 	}
1351 	mlxsw_cmd_mbox_config_profile_swid_config_mask_set(mbox, index, mask);
1352 }
1353 
1354 static int
1355 mlxsw_pci_profile_get_kvd_sizes(const struct mlxsw_pci *mlxsw_pci,
1356 				const struct mlxsw_config_profile *profile,
1357 				struct mlxsw_res *res)
1358 {
1359 	u64 single_size, double_size, linear_size;
1360 	int err;
1361 
1362 	err = mlxsw_core_kvd_sizes_get(mlxsw_pci->core, profile,
1363 				       &single_size, &double_size,
1364 				       &linear_size);
1365 	if (err)
1366 		return err;
1367 
1368 	MLXSW_RES_SET(res, KVD_SINGLE_SIZE, single_size);
1369 	MLXSW_RES_SET(res, KVD_DOUBLE_SIZE, double_size);
1370 	MLXSW_RES_SET(res, KVD_LINEAR_SIZE, linear_size);
1371 
1372 	return 0;
1373 }
1374 
1375 static int mlxsw_pci_config_profile(struct mlxsw_pci *mlxsw_pci, char *mbox,
1376 				    const struct mlxsw_config_profile *profile,
1377 				    struct mlxsw_res *res)
1378 {
1379 	int i;
1380 	int err;
1381 
1382 	mlxsw_cmd_mbox_zero(mbox);
1383 
1384 	if (profile->used_max_vepa_channels) {
1385 		mlxsw_cmd_mbox_config_profile_set_max_vepa_channels_set(
1386 			mbox, 1);
1387 		mlxsw_cmd_mbox_config_profile_max_vepa_channels_set(
1388 			mbox, profile->max_vepa_channels);
1389 	}
1390 	if (profile->used_max_lag) {
1391 		mlxsw_cmd_mbox_config_profile_set_max_lag_set(mbox, 1);
1392 		mlxsw_cmd_mbox_config_profile_max_lag_set(mbox,
1393 							  profile->max_lag);
1394 	}
1395 	if (profile->used_max_mid) {
1396 		mlxsw_cmd_mbox_config_profile_set_max_mid_set(
1397 			mbox, 1);
1398 		mlxsw_cmd_mbox_config_profile_max_mid_set(
1399 			mbox, profile->max_mid);
1400 	}
1401 	if (profile->used_max_pgt) {
1402 		mlxsw_cmd_mbox_config_profile_set_max_pgt_set(
1403 			mbox, 1);
1404 		mlxsw_cmd_mbox_config_profile_max_pgt_set(
1405 			mbox, profile->max_pgt);
1406 	}
1407 	if (profile->used_max_system_port) {
1408 		mlxsw_cmd_mbox_config_profile_set_max_system_port_set(
1409 			mbox, 1);
1410 		mlxsw_cmd_mbox_config_profile_max_system_port_set(
1411 			mbox, profile->max_system_port);
1412 	}
1413 	if (profile->used_max_vlan_groups) {
1414 		mlxsw_cmd_mbox_config_profile_set_max_vlan_groups_set(
1415 			mbox, 1);
1416 		mlxsw_cmd_mbox_config_profile_max_vlan_groups_set(
1417 			mbox, profile->max_vlan_groups);
1418 	}
1419 	if (profile->used_max_regions) {
1420 		mlxsw_cmd_mbox_config_profile_set_max_regions_set(
1421 			mbox, 1);
1422 		mlxsw_cmd_mbox_config_profile_max_regions_set(
1423 			mbox, profile->max_regions);
1424 	}
1425 	if (profile->used_flood_tables) {
1426 		mlxsw_cmd_mbox_config_profile_set_flood_tables_set(
1427 			mbox, 1);
1428 		mlxsw_cmd_mbox_config_profile_max_flood_tables_set(
1429 			mbox, profile->max_flood_tables);
1430 		mlxsw_cmd_mbox_config_profile_max_vid_flood_tables_set(
1431 			mbox, profile->max_vid_flood_tables);
1432 		mlxsw_cmd_mbox_config_profile_max_fid_offset_flood_tables_set(
1433 			mbox, profile->max_fid_offset_flood_tables);
1434 		mlxsw_cmd_mbox_config_profile_fid_offset_flood_table_size_set(
1435 			mbox, profile->fid_offset_flood_table_size);
1436 		mlxsw_cmd_mbox_config_profile_max_fid_flood_tables_set(
1437 			mbox, profile->max_fid_flood_tables);
1438 		mlxsw_cmd_mbox_config_profile_fid_flood_table_size_set(
1439 			mbox, profile->fid_flood_table_size);
1440 	}
1441 	if (profile->flood_mode_prefer_cff && mlxsw_pci->cff_support) {
1442 		enum mlxsw_cmd_mbox_config_profile_flood_mode flood_mode =
1443 			MLXSW_CMD_MBOX_CONFIG_PROFILE_FLOOD_MODE_CFF;
1444 
1445 		mlxsw_cmd_mbox_config_profile_set_flood_mode_set(mbox, 1);
1446 		mlxsw_cmd_mbox_config_profile_flood_mode_set(mbox, flood_mode);
1447 		mlxsw_pci->flood_mode = flood_mode;
1448 	} else if (profile->used_flood_mode) {
1449 		mlxsw_cmd_mbox_config_profile_set_flood_mode_set(
1450 			mbox, 1);
1451 		mlxsw_cmd_mbox_config_profile_flood_mode_set(
1452 			mbox, profile->flood_mode);
1453 		mlxsw_pci->flood_mode = profile->flood_mode;
1454 	} else {
1455 		WARN_ON(1);
1456 		return -EINVAL;
1457 	}
1458 	if (profile->used_max_ib_mc) {
1459 		mlxsw_cmd_mbox_config_profile_set_max_ib_mc_set(
1460 			mbox, 1);
1461 		mlxsw_cmd_mbox_config_profile_max_ib_mc_set(
1462 			mbox, profile->max_ib_mc);
1463 	}
1464 	if (profile->used_max_pkey) {
1465 		mlxsw_cmd_mbox_config_profile_set_max_pkey_set(
1466 			mbox, 1);
1467 		mlxsw_cmd_mbox_config_profile_max_pkey_set(
1468 			mbox, profile->max_pkey);
1469 	}
1470 	if (profile->used_ar_sec) {
1471 		mlxsw_cmd_mbox_config_profile_set_ar_sec_set(
1472 			mbox, 1);
1473 		mlxsw_cmd_mbox_config_profile_ar_sec_set(
1474 			mbox, profile->ar_sec);
1475 	}
1476 	if (profile->used_adaptive_routing_group_cap) {
1477 		mlxsw_cmd_mbox_config_profile_set_adaptive_routing_group_cap_set(
1478 			mbox, 1);
1479 		mlxsw_cmd_mbox_config_profile_adaptive_routing_group_cap_set(
1480 			mbox, profile->adaptive_routing_group_cap);
1481 	}
1482 	if (profile->used_ubridge) {
1483 		mlxsw_cmd_mbox_config_profile_set_ubridge_set(mbox, 1);
1484 		mlxsw_cmd_mbox_config_profile_ubridge_set(mbox,
1485 							  profile->ubridge);
1486 	}
1487 	if (profile->used_kvd_sizes && MLXSW_RES_VALID(res, KVD_SIZE)) {
1488 		err = mlxsw_pci_profile_get_kvd_sizes(mlxsw_pci, profile, res);
1489 		if (err)
1490 			return err;
1491 
1492 		mlxsw_cmd_mbox_config_profile_set_kvd_linear_size_set(mbox, 1);
1493 		mlxsw_cmd_mbox_config_profile_kvd_linear_size_set(mbox,
1494 					MLXSW_RES_GET(res, KVD_LINEAR_SIZE));
1495 		mlxsw_cmd_mbox_config_profile_set_kvd_hash_single_size_set(mbox,
1496 									   1);
1497 		mlxsw_cmd_mbox_config_profile_kvd_hash_single_size_set(mbox,
1498 					MLXSW_RES_GET(res, KVD_SINGLE_SIZE));
1499 		mlxsw_cmd_mbox_config_profile_set_kvd_hash_double_size_set(
1500 								mbox, 1);
1501 		mlxsw_cmd_mbox_config_profile_kvd_hash_double_size_set(mbox,
1502 					MLXSW_RES_GET(res, KVD_DOUBLE_SIZE));
1503 	}
1504 
1505 	for (i = 0; i < MLXSW_CONFIG_PROFILE_SWID_COUNT; i++)
1506 		mlxsw_pci_config_profile_swid_config(mlxsw_pci, mbox, i,
1507 						     &profile->swid_config[i]);
1508 
1509 	if (mlxsw_pci->max_cqe_ver > MLXSW_PCI_CQE_V0) {
1510 		mlxsw_cmd_mbox_config_profile_set_cqe_version_set(mbox, 1);
1511 		mlxsw_cmd_mbox_config_profile_cqe_version_set(mbox, 1);
1512 	}
1513 
1514 	if (profile->used_cqe_time_stamp_type) {
1515 		mlxsw_cmd_mbox_config_profile_set_cqe_time_stamp_type_set(mbox,
1516 									  1);
1517 		mlxsw_cmd_mbox_config_profile_cqe_time_stamp_type_set(mbox,
1518 					profile->cqe_time_stamp_type);
1519 	}
1520 
1521 	if (profile->lag_mode_prefer_sw && mlxsw_pci->lag_mode_support) {
1522 		enum mlxsw_cmd_mbox_config_profile_lag_mode lag_mode =
1523 			MLXSW_CMD_MBOX_CONFIG_PROFILE_LAG_MODE_SW;
1524 
1525 		mlxsw_cmd_mbox_config_profile_set_lag_mode_set(mbox, 1);
1526 		mlxsw_cmd_mbox_config_profile_lag_mode_set(mbox, lag_mode);
1527 		mlxsw_pci->lag_mode = lag_mode;
1528 	} else {
1529 		mlxsw_pci->lag_mode = MLXSW_CMD_MBOX_CONFIG_PROFILE_LAG_MODE_FW;
1530 	}
1531 	return mlxsw_cmd_config_profile_set(mlxsw_pci->core, mbox);
1532 }
1533 
1534 static int mlxsw_pci_boardinfo(struct mlxsw_pci *mlxsw_pci, char *mbox)
1535 {
1536 	struct mlxsw_bus_info *bus_info = &mlxsw_pci->bus_info;
1537 	int err;
1538 
1539 	mlxsw_cmd_mbox_zero(mbox);
1540 	err = mlxsw_cmd_boardinfo(mlxsw_pci->core, mbox);
1541 	if (err)
1542 		return err;
1543 	mlxsw_cmd_mbox_boardinfo_vsd_memcpy_from(mbox, bus_info->vsd);
1544 	mlxsw_cmd_mbox_boardinfo_psid_memcpy_from(mbox, bus_info->psid);
1545 	return 0;
1546 }
1547 
1548 static int mlxsw_pci_fw_area_init(struct mlxsw_pci *mlxsw_pci, char *mbox,
1549 				  u16 num_pages)
1550 {
1551 	struct mlxsw_pci_mem_item *mem_item;
1552 	int nent = 0;
1553 	int i;
1554 	int err;
1555 
1556 	mlxsw_pci->fw_area.items = kcalloc(num_pages, sizeof(*mem_item),
1557 					   GFP_KERNEL);
1558 	if (!mlxsw_pci->fw_area.items)
1559 		return -ENOMEM;
1560 	mlxsw_pci->fw_area.count = num_pages;
1561 
1562 	mlxsw_cmd_mbox_zero(mbox);
1563 	for (i = 0; i < num_pages; i++) {
1564 		mem_item = &mlxsw_pci->fw_area.items[i];
1565 
1566 		mem_item->size = MLXSW_PCI_PAGE_SIZE;
1567 		mem_item->buf = dma_alloc_coherent(&mlxsw_pci->pdev->dev,
1568 						   mem_item->size,
1569 						   &mem_item->mapaddr, GFP_KERNEL);
1570 		if (!mem_item->buf) {
1571 			err = -ENOMEM;
1572 			goto err_alloc;
1573 		}
1574 		mlxsw_cmd_mbox_map_fa_pa_set(mbox, nent, mem_item->mapaddr);
1575 		mlxsw_cmd_mbox_map_fa_log2size_set(mbox, nent, 0); /* 1 page */
1576 		if (++nent == MLXSW_CMD_MAP_FA_VPM_ENTRIES_MAX) {
1577 			err = mlxsw_cmd_map_fa(mlxsw_pci->core, mbox, nent);
1578 			if (err)
1579 				goto err_cmd_map_fa;
1580 			nent = 0;
1581 			mlxsw_cmd_mbox_zero(mbox);
1582 		}
1583 	}
1584 
1585 	if (nent) {
1586 		err = mlxsw_cmd_map_fa(mlxsw_pci->core, mbox, nent);
1587 		if (err)
1588 			goto err_cmd_map_fa;
1589 	}
1590 
1591 	return 0;
1592 
1593 err_cmd_map_fa:
1594 err_alloc:
1595 	for (i--; i >= 0; i--) {
1596 		mem_item = &mlxsw_pci->fw_area.items[i];
1597 
1598 		dma_free_coherent(&mlxsw_pci->pdev->dev, mem_item->size,
1599 				  mem_item->buf, mem_item->mapaddr);
1600 	}
1601 	kfree(mlxsw_pci->fw_area.items);
1602 	return err;
1603 }
1604 
1605 static void mlxsw_pci_fw_area_fini(struct mlxsw_pci *mlxsw_pci)
1606 {
1607 	struct mlxsw_pci_mem_item *mem_item;
1608 	int i;
1609 
1610 	mlxsw_cmd_unmap_fa(mlxsw_pci->core);
1611 
1612 	for (i = 0; i < mlxsw_pci->fw_area.count; i++) {
1613 		mem_item = &mlxsw_pci->fw_area.items[i];
1614 
1615 		dma_free_coherent(&mlxsw_pci->pdev->dev, mem_item->size,
1616 				  mem_item->buf, mem_item->mapaddr);
1617 	}
1618 	kfree(mlxsw_pci->fw_area.items);
1619 }
1620 
1621 static irqreturn_t mlxsw_pci_eq_irq_handler(int irq, void *dev_id)
1622 {
1623 	struct mlxsw_pci *mlxsw_pci = dev_id;
1624 	struct mlxsw_pci_queue *q;
1625 
1626 	q = mlxsw_pci_eq_get(mlxsw_pci);
1627 	tasklet_schedule(&q->u.eq.tasklet);
1628 	return IRQ_HANDLED;
1629 }
1630 
1631 static int mlxsw_pci_mbox_alloc(struct mlxsw_pci *mlxsw_pci,
1632 				struct mlxsw_pci_mem_item *mbox)
1633 {
1634 	struct pci_dev *pdev = mlxsw_pci->pdev;
1635 	int err = 0;
1636 
1637 	mbox->size = MLXSW_CMD_MBOX_SIZE;
1638 	mbox->buf = dma_alloc_coherent(&pdev->dev, MLXSW_CMD_MBOX_SIZE,
1639 				       &mbox->mapaddr, GFP_KERNEL);
1640 	if (!mbox->buf) {
1641 		dev_err(&pdev->dev, "Failed allocating memory for mailbox\n");
1642 		err = -ENOMEM;
1643 	}
1644 
1645 	return err;
1646 }
1647 
1648 static void mlxsw_pci_mbox_free(struct mlxsw_pci *mlxsw_pci,
1649 				struct mlxsw_pci_mem_item *mbox)
1650 {
1651 	struct pci_dev *pdev = mlxsw_pci->pdev;
1652 
1653 	dma_free_coherent(&pdev->dev, MLXSW_CMD_MBOX_SIZE, mbox->buf,
1654 			  mbox->mapaddr);
1655 }
1656 
1657 static int mlxsw_pci_sys_ready_wait(struct mlxsw_pci *mlxsw_pci,
1658 				    const struct pci_device_id *id,
1659 				    u32 *p_sys_status)
1660 {
1661 	unsigned long end;
1662 	u32 val;
1663 
1664 	/* We must wait for the HW to become responsive. */
1665 	msleep(MLXSW_PCI_SW_RESET_WAIT_MSECS);
1666 
1667 	end = jiffies + msecs_to_jiffies(MLXSW_PCI_SW_RESET_TIMEOUT_MSECS);
1668 	do {
1669 		val = mlxsw_pci_read32(mlxsw_pci, FW_READY);
1670 		if ((val & MLXSW_PCI_FW_READY_MASK) == MLXSW_PCI_FW_READY_MAGIC)
1671 			return 0;
1672 		cond_resched();
1673 	} while (time_before(jiffies, end));
1674 
1675 	*p_sys_status = val & MLXSW_PCI_FW_READY_MASK;
1676 
1677 	return -EBUSY;
1678 }
1679 
1680 static int mlxsw_pci_reset_at_pci_disable(struct mlxsw_pci *mlxsw_pci)
1681 {
1682 	struct pci_dev *pdev = mlxsw_pci->pdev;
1683 	char mrsr_pl[MLXSW_REG_MRSR_LEN];
1684 	int err;
1685 
1686 	mlxsw_reg_mrsr_pack(mrsr_pl,
1687 			    MLXSW_REG_MRSR_COMMAND_RESET_AT_PCI_DISABLE);
1688 	err = mlxsw_reg_write(mlxsw_pci->core, MLXSW_REG(mrsr), mrsr_pl);
1689 	if (err)
1690 		return err;
1691 
1692 	device_lock_assert(&pdev->dev);
1693 
1694 	pci_cfg_access_lock(pdev);
1695 	pci_save_state(pdev);
1696 
1697 	err = __pci_reset_function_locked(pdev);
1698 	if (err)
1699 		pci_err(pdev, "PCI function reset failed with %d\n", err);
1700 
1701 	pci_restore_state(pdev);
1702 	pci_cfg_access_unlock(pdev);
1703 
1704 	return err;
1705 }
1706 
1707 static int mlxsw_pci_reset_sw(struct mlxsw_pci *mlxsw_pci)
1708 {
1709 	char mrsr_pl[MLXSW_REG_MRSR_LEN];
1710 
1711 	mlxsw_reg_mrsr_pack(mrsr_pl, MLXSW_REG_MRSR_COMMAND_SOFTWARE_RESET);
1712 	return mlxsw_reg_write(mlxsw_pci->core, MLXSW_REG(mrsr), mrsr_pl);
1713 }
1714 
1715 static int
1716 mlxsw_pci_reset(struct mlxsw_pci *mlxsw_pci, const struct pci_device_id *id)
1717 {
1718 	struct pci_dev *pdev = mlxsw_pci->pdev;
1719 	char mcam_pl[MLXSW_REG_MCAM_LEN];
1720 	bool pci_reset_supported = false;
1721 	u32 sys_status;
1722 	int err;
1723 
1724 	err = mlxsw_pci_sys_ready_wait(mlxsw_pci, id, &sys_status);
1725 	if (err) {
1726 		dev_err(&pdev->dev, "Failed to reach system ready status before reset. Status is 0x%x\n",
1727 			sys_status);
1728 		return err;
1729 	}
1730 
1731 	/* PCI core already issued a PCI reset, do not issue another reset. */
1732 	if (mlxsw_pci->skip_reset)
1733 		return 0;
1734 
1735 	mlxsw_reg_mcam_pack(mcam_pl,
1736 			    MLXSW_REG_MCAM_FEATURE_GROUP_ENHANCED_FEATURES);
1737 	err = mlxsw_reg_query(mlxsw_pci->core, MLXSW_REG(mcam), mcam_pl);
1738 	if (!err)
1739 		mlxsw_reg_mcam_unpack(mcam_pl, MLXSW_REG_MCAM_PCI_RESET,
1740 				      &pci_reset_supported);
1741 
1742 	if (pci_reset_supported) {
1743 		pci_dbg(pdev, "Starting PCI reset flow\n");
1744 		err = mlxsw_pci_reset_at_pci_disable(mlxsw_pci);
1745 	} else {
1746 		pci_dbg(pdev, "Starting software reset flow\n");
1747 		err = mlxsw_pci_reset_sw(mlxsw_pci);
1748 	}
1749 	if (err)
1750 		return err;
1751 
1752 	err = mlxsw_pci_sys_ready_wait(mlxsw_pci, id, &sys_status);
1753 	if (err) {
1754 		dev_err(&pdev->dev, "Failed to reach system ready status after reset. Status is 0x%x\n",
1755 			sys_status);
1756 		return err;
1757 	}
1758 
1759 	return 0;
1760 }
1761 
1762 static int mlxsw_pci_alloc_irq_vectors(struct mlxsw_pci *mlxsw_pci)
1763 {
1764 	int err;
1765 
1766 	err = pci_alloc_irq_vectors(mlxsw_pci->pdev, 1, 1, PCI_IRQ_MSIX);
1767 	if (err < 0)
1768 		dev_err(&mlxsw_pci->pdev->dev, "MSI-X init failed\n");
1769 	return err;
1770 }
1771 
1772 static void mlxsw_pci_free_irq_vectors(struct mlxsw_pci *mlxsw_pci)
1773 {
1774 	pci_free_irq_vectors(mlxsw_pci->pdev);
1775 }
1776 
1777 static int mlxsw_pci_init(void *bus_priv, struct mlxsw_core *mlxsw_core,
1778 			  const struct mlxsw_config_profile *profile,
1779 			  struct mlxsw_res *res)
1780 {
1781 	struct mlxsw_pci *mlxsw_pci = bus_priv;
1782 	struct pci_dev *pdev = mlxsw_pci->pdev;
1783 	char *mbox;
1784 	u16 num_pages;
1785 	int err;
1786 
1787 	mlxsw_pci->core = mlxsw_core;
1788 
1789 	mbox = mlxsw_cmd_mbox_alloc();
1790 	if (!mbox)
1791 		return -ENOMEM;
1792 
1793 	err = mlxsw_pci_reset(mlxsw_pci, mlxsw_pci->id);
1794 	if (err)
1795 		goto err_reset;
1796 
1797 	err = mlxsw_pci_alloc_irq_vectors(mlxsw_pci);
1798 	if (err < 0) {
1799 		dev_err(&pdev->dev, "MSI-X init failed\n");
1800 		goto err_alloc_irq;
1801 	}
1802 
1803 	err = mlxsw_cmd_query_fw(mlxsw_core, mbox);
1804 	if (err)
1805 		goto err_query_fw;
1806 
1807 	mlxsw_pci->bus_info.fw_rev.major =
1808 		mlxsw_cmd_mbox_query_fw_fw_rev_major_get(mbox);
1809 	mlxsw_pci->bus_info.fw_rev.minor =
1810 		mlxsw_cmd_mbox_query_fw_fw_rev_minor_get(mbox);
1811 	mlxsw_pci->bus_info.fw_rev.subminor =
1812 		mlxsw_cmd_mbox_query_fw_fw_rev_subminor_get(mbox);
1813 
1814 	if (mlxsw_cmd_mbox_query_fw_cmd_interface_rev_get(mbox) != 1) {
1815 		dev_err(&pdev->dev, "Unsupported cmd interface revision ID queried from hw\n");
1816 		err = -EINVAL;
1817 		goto err_iface_rev;
1818 	}
1819 	if (mlxsw_cmd_mbox_query_fw_doorbell_page_bar_get(mbox) != 0) {
1820 		dev_err(&pdev->dev, "Unsupported doorbell page bar queried from hw\n");
1821 		err = -EINVAL;
1822 		goto err_doorbell_page_bar;
1823 	}
1824 
1825 	mlxsw_pci->doorbell_offset =
1826 		mlxsw_cmd_mbox_query_fw_doorbell_page_offset_get(mbox);
1827 
1828 	if (mlxsw_cmd_mbox_query_fw_fr_rn_clk_bar_get(mbox) != 0) {
1829 		dev_err(&pdev->dev, "Unsupported free running clock BAR queried from hw\n");
1830 		err = -EINVAL;
1831 		goto err_fr_rn_clk_bar;
1832 	}
1833 
1834 	mlxsw_pci->free_running_clock_offset =
1835 		mlxsw_cmd_mbox_query_fw_free_running_clock_offset_get(mbox);
1836 
1837 	if (mlxsw_cmd_mbox_query_fw_utc_sec_bar_get(mbox) != 0) {
1838 		dev_err(&pdev->dev, "Unsupported UTC sec BAR queried from hw\n");
1839 		err = -EINVAL;
1840 		goto err_utc_sec_bar;
1841 	}
1842 
1843 	mlxsw_pci->utc_sec_offset =
1844 		mlxsw_cmd_mbox_query_fw_utc_sec_offset_get(mbox);
1845 
1846 	if (mlxsw_cmd_mbox_query_fw_utc_nsec_bar_get(mbox) != 0) {
1847 		dev_err(&pdev->dev, "Unsupported UTC nsec BAR queried from hw\n");
1848 		err = -EINVAL;
1849 		goto err_utc_nsec_bar;
1850 	}
1851 
1852 	mlxsw_pci->utc_nsec_offset =
1853 		mlxsw_cmd_mbox_query_fw_utc_nsec_offset_get(mbox);
1854 
1855 	mlxsw_pci->lag_mode_support =
1856 		mlxsw_cmd_mbox_query_fw_lag_mode_support_get(mbox);
1857 	mlxsw_pci->cff_support =
1858 		mlxsw_cmd_mbox_query_fw_cff_support_get(mbox);
1859 
1860 	num_pages = mlxsw_cmd_mbox_query_fw_fw_pages_get(mbox);
1861 	err = mlxsw_pci_fw_area_init(mlxsw_pci, mbox, num_pages);
1862 	if (err)
1863 		goto err_fw_area_init;
1864 
1865 	err = mlxsw_pci_boardinfo(mlxsw_pci, mbox);
1866 	if (err)
1867 		goto err_boardinfo;
1868 
1869 	err = mlxsw_core_resources_query(mlxsw_core, mbox, res);
1870 	if (err)
1871 		goto err_query_resources;
1872 
1873 	if (MLXSW_CORE_RES_VALID(mlxsw_core, CQE_V2) &&
1874 	    MLXSW_CORE_RES_GET(mlxsw_core, CQE_V2))
1875 		mlxsw_pci->max_cqe_ver = MLXSW_PCI_CQE_V2;
1876 	else if (MLXSW_CORE_RES_VALID(mlxsw_core, CQE_V1) &&
1877 		 MLXSW_CORE_RES_GET(mlxsw_core, CQE_V1))
1878 		mlxsw_pci->max_cqe_ver = MLXSW_PCI_CQE_V1;
1879 	else if ((MLXSW_CORE_RES_VALID(mlxsw_core, CQE_V0) &&
1880 		  MLXSW_CORE_RES_GET(mlxsw_core, CQE_V0)) ||
1881 		 !MLXSW_CORE_RES_VALID(mlxsw_core, CQE_V0)) {
1882 		mlxsw_pci->max_cqe_ver = MLXSW_PCI_CQE_V0;
1883 	} else {
1884 		dev_err(&pdev->dev, "Invalid supported CQE version combination reported\n");
1885 		goto err_cqe_v_check;
1886 	}
1887 
1888 	err = mlxsw_pci_config_profile(mlxsw_pci, mbox, profile, res);
1889 	if (err)
1890 		goto err_config_profile;
1891 
1892 	/* Some resources depend on details of config_profile, such as unified
1893 	 * bridge model. Query the resources again to get correct values.
1894 	 */
1895 	err = mlxsw_core_resources_query(mlxsw_core, mbox, res);
1896 	if (err)
1897 		goto err_requery_resources;
1898 
1899 	err = mlxsw_pci_napi_devs_init(mlxsw_pci);
1900 	if (err)
1901 		goto err_napi_devs_init;
1902 
1903 	err = mlxsw_pci_aqs_init(mlxsw_pci, mbox);
1904 	if (err)
1905 		goto err_aqs_init;
1906 
1907 	err = request_irq(pci_irq_vector(pdev, 0),
1908 			  mlxsw_pci_eq_irq_handler, 0,
1909 			  mlxsw_pci->bus_info.device_kind, mlxsw_pci);
1910 	if (err) {
1911 		dev_err(&pdev->dev, "IRQ request failed\n");
1912 		goto err_request_eq_irq;
1913 	}
1914 
1915 	goto mbox_put;
1916 
1917 err_request_eq_irq:
1918 	mlxsw_pci_aqs_fini(mlxsw_pci);
1919 err_aqs_init:
1920 	mlxsw_pci_napi_devs_fini(mlxsw_pci);
1921 err_napi_devs_init:
1922 err_requery_resources:
1923 err_config_profile:
1924 err_cqe_v_check:
1925 err_query_resources:
1926 err_boardinfo:
1927 	mlxsw_pci_fw_area_fini(mlxsw_pci);
1928 err_fw_area_init:
1929 err_utc_nsec_bar:
1930 err_utc_sec_bar:
1931 err_fr_rn_clk_bar:
1932 err_doorbell_page_bar:
1933 err_iface_rev:
1934 err_query_fw:
1935 	mlxsw_pci_free_irq_vectors(mlxsw_pci);
1936 err_alloc_irq:
1937 err_reset:
1938 mbox_put:
1939 	mlxsw_cmd_mbox_free(mbox);
1940 	return err;
1941 }
1942 
1943 static void mlxsw_pci_fini(void *bus_priv)
1944 {
1945 	struct mlxsw_pci *mlxsw_pci = bus_priv;
1946 
1947 	free_irq(pci_irq_vector(mlxsw_pci->pdev, 0), mlxsw_pci);
1948 	mlxsw_pci_aqs_fini(mlxsw_pci);
1949 	mlxsw_pci_napi_devs_fini(mlxsw_pci);
1950 	mlxsw_pci_fw_area_fini(mlxsw_pci);
1951 	mlxsw_pci_free_irq_vectors(mlxsw_pci);
1952 }
1953 
1954 static struct mlxsw_pci_queue *
1955 mlxsw_pci_sdq_pick(struct mlxsw_pci *mlxsw_pci,
1956 		   const struct mlxsw_tx_info *tx_info)
1957 {
1958 	u8 ctl_sdq_count = mlxsw_pci->num_sdqs - 1;
1959 	u8 sdqn;
1960 
1961 	if (tx_info->is_emad) {
1962 		sdqn = MLXSW_PCI_SDQ_EMAD_INDEX;
1963 	} else {
1964 		BUILD_BUG_ON(MLXSW_PCI_SDQ_EMAD_INDEX != 0);
1965 		sdqn = 1 + (tx_info->local_port % ctl_sdq_count);
1966 	}
1967 
1968 	return mlxsw_pci_sdq_get(mlxsw_pci, sdqn);
1969 }
1970 
1971 static bool mlxsw_pci_skb_transmit_busy(void *bus_priv,
1972 					const struct mlxsw_tx_info *tx_info)
1973 {
1974 	struct mlxsw_pci *mlxsw_pci = bus_priv;
1975 	struct mlxsw_pci_queue *q = mlxsw_pci_sdq_pick(mlxsw_pci, tx_info);
1976 
1977 	return !mlxsw_pci_queue_elem_info_producer_get(q);
1978 }
1979 
1980 static int mlxsw_pci_skb_transmit(void *bus_priv, struct sk_buff *skb,
1981 				  const struct mlxsw_tx_info *tx_info)
1982 {
1983 	struct mlxsw_pci *mlxsw_pci = bus_priv;
1984 	struct mlxsw_pci_queue *q;
1985 	struct mlxsw_pci_queue_elem_info *elem_info;
1986 	char *wqe;
1987 	int i;
1988 	int err;
1989 
1990 	if (skb_shinfo(skb)->nr_frags > MLXSW_PCI_WQE_SG_ENTRIES - 1) {
1991 		err = skb_linearize(skb);
1992 		if (err)
1993 			return err;
1994 	}
1995 
1996 	q = mlxsw_pci_sdq_pick(mlxsw_pci, tx_info);
1997 	spin_lock_bh(&q->lock);
1998 	elem_info = mlxsw_pci_queue_elem_info_producer_get(q);
1999 	if (!elem_info) {
2000 		/* queue is full */
2001 		err = -EAGAIN;
2002 		goto unlock;
2003 	}
2004 	mlxsw_skb_cb(skb)->tx_info = *tx_info;
2005 	elem_info->sdq.skb = skb;
2006 
2007 	wqe = elem_info->elem;
2008 	mlxsw_pci_wqe_c_set(wqe, 1); /* always report completion */
2009 	mlxsw_pci_wqe_lp_set(wqe, 0);
2010 	mlxsw_pci_wqe_type_set(wqe, MLXSW_PCI_WQE_TYPE_ETHERNET);
2011 
2012 	err = mlxsw_pci_wqe_frag_map(mlxsw_pci, wqe, 0, skb->data,
2013 				     skb_headlen(skb), DMA_TO_DEVICE);
2014 	if (err)
2015 		goto unlock;
2016 
2017 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2018 		const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2019 
2020 		err = mlxsw_pci_wqe_frag_map(mlxsw_pci, wqe, i + 1,
2021 					     skb_frag_address(frag),
2022 					     skb_frag_size(frag),
2023 					     DMA_TO_DEVICE);
2024 		if (err)
2025 			goto unmap_frags;
2026 	}
2027 
2028 	if (unlikely(skb_shinfo(skb)->tx_flags & SKBTX_HW_TSTAMP))
2029 		skb_shinfo(skb)->tx_flags |= SKBTX_IN_PROGRESS;
2030 
2031 	/* Set unused sq entries byte count to zero. */
2032 	for (i++; i < MLXSW_PCI_WQE_SG_ENTRIES; i++)
2033 		mlxsw_pci_wqe_byte_count_set(wqe, i, 0);
2034 
2035 	/* Everything is set up, ring producer doorbell to get HW going */
2036 	q->producer_counter++;
2037 	mlxsw_pci_queue_doorbell_producer_ring(mlxsw_pci, q);
2038 
2039 	goto unlock;
2040 
2041 unmap_frags:
2042 	for (; i >= 0; i--)
2043 		mlxsw_pci_wqe_frag_unmap(mlxsw_pci, wqe, i, DMA_TO_DEVICE);
2044 unlock:
2045 	spin_unlock_bh(&q->lock);
2046 	return err;
2047 }
2048 
2049 static int mlxsw_pci_cmd_exec(void *bus_priv, u16 opcode, u8 opcode_mod,
2050 			      u32 in_mod, bool out_mbox_direct,
2051 			      char *in_mbox, size_t in_mbox_size,
2052 			      char *out_mbox, size_t out_mbox_size,
2053 			      u8 *p_status)
2054 {
2055 	struct mlxsw_pci *mlxsw_pci = bus_priv;
2056 	dma_addr_t in_mapaddr = 0, out_mapaddr = 0;
2057 	unsigned long timeout = msecs_to_jiffies(MLXSW_PCI_CIR_TIMEOUT_MSECS);
2058 	unsigned long end;
2059 	bool wait_done;
2060 	int err;
2061 
2062 	*p_status = MLXSW_CMD_STATUS_OK;
2063 
2064 	err = mutex_lock_interruptible(&mlxsw_pci->cmd.lock);
2065 	if (err)
2066 		return err;
2067 
2068 	if (in_mbox) {
2069 		memcpy(mlxsw_pci->cmd.in_mbox.buf, in_mbox, in_mbox_size);
2070 		in_mapaddr = mlxsw_pci->cmd.in_mbox.mapaddr;
2071 	}
2072 	mlxsw_pci_write32(mlxsw_pci, CIR_IN_PARAM_HI, upper_32_bits(in_mapaddr));
2073 	mlxsw_pci_write32(mlxsw_pci, CIR_IN_PARAM_LO, lower_32_bits(in_mapaddr));
2074 
2075 	if (out_mbox)
2076 		out_mapaddr = mlxsw_pci->cmd.out_mbox.mapaddr;
2077 	mlxsw_pci_write32(mlxsw_pci, CIR_OUT_PARAM_HI, upper_32_bits(out_mapaddr));
2078 	mlxsw_pci_write32(mlxsw_pci, CIR_OUT_PARAM_LO, lower_32_bits(out_mapaddr));
2079 
2080 	mlxsw_pci_write32(mlxsw_pci, CIR_IN_MODIFIER, in_mod);
2081 	mlxsw_pci_write32(mlxsw_pci, CIR_TOKEN, 0);
2082 
2083 	wait_done = false;
2084 
2085 	wmb(); /* all needs to be written before we write control register */
2086 	mlxsw_pci_write32(mlxsw_pci, CIR_CTRL,
2087 			  MLXSW_PCI_CIR_CTRL_GO_BIT |
2088 			  (opcode_mod << MLXSW_PCI_CIR_CTRL_OPCODE_MOD_SHIFT) |
2089 			  opcode);
2090 
2091 	end = jiffies + timeout;
2092 	do {
2093 		u32 ctrl = mlxsw_pci_read32(mlxsw_pci, CIR_CTRL);
2094 
2095 		if (!(ctrl & MLXSW_PCI_CIR_CTRL_GO_BIT)) {
2096 			wait_done = true;
2097 			*p_status = ctrl >> MLXSW_PCI_CIR_CTRL_STATUS_SHIFT;
2098 			break;
2099 		}
2100 		cond_resched();
2101 	} while (time_before(jiffies, end));
2102 
2103 	err = 0;
2104 	if (wait_done) {
2105 		if (*p_status)
2106 			err = -EIO;
2107 	} else {
2108 		err = -ETIMEDOUT;
2109 	}
2110 
2111 	if (!err && out_mbox && out_mbox_direct) {
2112 		/* Some commands don't use output param as address to mailbox
2113 		 * but they store output directly into registers. In that case,
2114 		 * copy registers into mbox buffer.
2115 		 */
2116 		__be32 tmp;
2117 
2118 		tmp = cpu_to_be32(mlxsw_pci_read32(mlxsw_pci,
2119 						   CIR_OUT_PARAM_HI));
2120 		memcpy(out_mbox, &tmp, sizeof(tmp));
2121 		tmp = cpu_to_be32(mlxsw_pci_read32(mlxsw_pci,
2122 						   CIR_OUT_PARAM_LO));
2123 		memcpy(out_mbox + sizeof(tmp), &tmp, sizeof(tmp));
2124 	} else if (!err && out_mbox) {
2125 		memcpy(out_mbox, mlxsw_pci->cmd.out_mbox.buf, out_mbox_size);
2126 	}
2127 
2128 	mutex_unlock(&mlxsw_pci->cmd.lock);
2129 
2130 	return err;
2131 }
2132 
2133 static u32 mlxsw_pci_read_frc_h(void *bus_priv)
2134 {
2135 	struct mlxsw_pci *mlxsw_pci = bus_priv;
2136 	u64 frc_offset_h;
2137 
2138 	frc_offset_h = mlxsw_pci->free_running_clock_offset;
2139 	return mlxsw_pci_read32_off(mlxsw_pci, frc_offset_h);
2140 }
2141 
2142 static u32 mlxsw_pci_read_frc_l(void *bus_priv)
2143 {
2144 	struct mlxsw_pci *mlxsw_pci = bus_priv;
2145 	u64 frc_offset_l;
2146 
2147 	frc_offset_l = mlxsw_pci->free_running_clock_offset + 4;
2148 	return mlxsw_pci_read32_off(mlxsw_pci, frc_offset_l);
2149 }
2150 
2151 static u32 mlxsw_pci_read_utc_sec(void *bus_priv)
2152 {
2153 	struct mlxsw_pci *mlxsw_pci = bus_priv;
2154 
2155 	return mlxsw_pci_read32_off(mlxsw_pci, mlxsw_pci->utc_sec_offset);
2156 }
2157 
2158 static u32 mlxsw_pci_read_utc_nsec(void *bus_priv)
2159 {
2160 	struct mlxsw_pci *mlxsw_pci = bus_priv;
2161 
2162 	return mlxsw_pci_read32_off(mlxsw_pci, mlxsw_pci->utc_nsec_offset);
2163 }
2164 
2165 static enum mlxsw_cmd_mbox_config_profile_lag_mode
2166 mlxsw_pci_lag_mode(void *bus_priv)
2167 {
2168 	struct mlxsw_pci *mlxsw_pci = bus_priv;
2169 
2170 	return mlxsw_pci->lag_mode;
2171 }
2172 
2173 static enum mlxsw_cmd_mbox_config_profile_flood_mode
2174 mlxsw_pci_flood_mode(void *bus_priv)
2175 {
2176 	struct mlxsw_pci *mlxsw_pci = bus_priv;
2177 
2178 	return mlxsw_pci->flood_mode;
2179 }
2180 
2181 static const struct mlxsw_bus mlxsw_pci_bus = {
2182 	.kind			= "pci",
2183 	.init			= mlxsw_pci_init,
2184 	.fini			= mlxsw_pci_fini,
2185 	.skb_transmit_busy	= mlxsw_pci_skb_transmit_busy,
2186 	.skb_transmit		= mlxsw_pci_skb_transmit,
2187 	.cmd_exec		= mlxsw_pci_cmd_exec,
2188 	.read_frc_h		= mlxsw_pci_read_frc_h,
2189 	.read_frc_l		= mlxsw_pci_read_frc_l,
2190 	.read_utc_sec		= mlxsw_pci_read_utc_sec,
2191 	.read_utc_nsec		= mlxsw_pci_read_utc_nsec,
2192 	.lag_mode		= mlxsw_pci_lag_mode,
2193 	.flood_mode		= mlxsw_pci_flood_mode,
2194 	.features		= MLXSW_BUS_F_TXRX | MLXSW_BUS_F_RESET,
2195 };
2196 
2197 static int mlxsw_pci_cmd_init(struct mlxsw_pci *mlxsw_pci)
2198 {
2199 	int err;
2200 
2201 	mutex_init(&mlxsw_pci->cmd.lock);
2202 
2203 	err = mlxsw_pci_mbox_alloc(mlxsw_pci, &mlxsw_pci->cmd.in_mbox);
2204 	if (err)
2205 		goto err_in_mbox_alloc;
2206 
2207 	err = mlxsw_pci_mbox_alloc(mlxsw_pci, &mlxsw_pci->cmd.out_mbox);
2208 	if (err)
2209 		goto err_out_mbox_alloc;
2210 
2211 	return 0;
2212 
2213 err_out_mbox_alloc:
2214 	mlxsw_pci_mbox_free(mlxsw_pci, &mlxsw_pci->cmd.in_mbox);
2215 err_in_mbox_alloc:
2216 	mutex_destroy(&mlxsw_pci->cmd.lock);
2217 	return err;
2218 }
2219 
2220 static void mlxsw_pci_cmd_fini(struct mlxsw_pci *mlxsw_pci)
2221 {
2222 	mlxsw_pci_mbox_free(mlxsw_pci, &mlxsw_pci->cmd.out_mbox);
2223 	mlxsw_pci_mbox_free(mlxsw_pci, &mlxsw_pci->cmd.in_mbox);
2224 	mutex_destroy(&mlxsw_pci->cmd.lock);
2225 }
2226 
2227 static int mlxsw_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
2228 {
2229 	const char *driver_name = dev_driver_string(&pdev->dev);
2230 	struct mlxsw_pci *mlxsw_pci;
2231 	int err;
2232 
2233 	mlxsw_pci = kzalloc(sizeof(*mlxsw_pci), GFP_KERNEL);
2234 	if (!mlxsw_pci)
2235 		return -ENOMEM;
2236 
2237 	err = pci_enable_device(pdev);
2238 	if (err) {
2239 		dev_err(&pdev->dev, "pci_enable_device failed\n");
2240 		goto err_pci_enable_device;
2241 	}
2242 
2243 	err = pci_request_regions(pdev, driver_name);
2244 	if (err) {
2245 		dev_err(&pdev->dev, "pci_request_regions failed\n");
2246 		goto err_pci_request_regions;
2247 	}
2248 
2249 	err = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
2250 	if (err) {
2251 		err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(32));
2252 		if (err) {
2253 			dev_err(&pdev->dev, "dma_set_mask failed\n");
2254 			goto err_pci_set_dma_mask;
2255 		}
2256 	}
2257 
2258 	if (pci_resource_len(pdev, 0) < MLXSW_PCI_BAR0_SIZE) {
2259 		dev_err(&pdev->dev, "invalid PCI region size\n");
2260 		err = -EINVAL;
2261 		goto err_pci_resource_len_check;
2262 	}
2263 
2264 	mlxsw_pci->hw_addr = ioremap(pci_resource_start(pdev, 0),
2265 				     pci_resource_len(pdev, 0));
2266 	if (!mlxsw_pci->hw_addr) {
2267 		dev_err(&pdev->dev, "ioremap failed\n");
2268 		err = -EIO;
2269 		goto err_ioremap;
2270 	}
2271 	pci_set_master(pdev);
2272 
2273 	mlxsw_pci->pdev = pdev;
2274 	pci_set_drvdata(pdev, mlxsw_pci);
2275 
2276 	err = mlxsw_pci_cmd_init(mlxsw_pci);
2277 	if (err)
2278 		goto err_pci_cmd_init;
2279 
2280 	mlxsw_pci->bus_info.device_kind = driver_name;
2281 	mlxsw_pci->bus_info.device_name = pci_name(mlxsw_pci->pdev);
2282 	mlxsw_pci->bus_info.dev = &pdev->dev;
2283 	mlxsw_pci->bus_info.read_clock_capable = true;
2284 	mlxsw_pci->id = id;
2285 
2286 	err = mlxsw_core_bus_device_register(&mlxsw_pci->bus_info,
2287 					     &mlxsw_pci_bus, mlxsw_pci, false,
2288 					     NULL, NULL);
2289 	if (err) {
2290 		dev_err(&pdev->dev, "cannot register bus device\n");
2291 		goto err_bus_device_register;
2292 	}
2293 
2294 	return 0;
2295 
2296 err_bus_device_register:
2297 	mlxsw_pci_cmd_fini(mlxsw_pci);
2298 err_pci_cmd_init:
2299 	iounmap(mlxsw_pci->hw_addr);
2300 err_ioremap:
2301 err_pci_resource_len_check:
2302 err_pci_set_dma_mask:
2303 	pci_release_regions(pdev);
2304 err_pci_request_regions:
2305 	pci_disable_device(pdev);
2306 err_pci_enable_device:
2307 	kfree(mlxsw_pci);
2308 	return err;
2309 }
2310 
2311 static void mlxsw_pci_remove(struct pci_dev *pdev)
2312 {
2313 	struct mlxsw_pci *mlxsw_pci = pci_get_drvdata(pdev);
2314 
2315 	mlxsw_core_bus_device_unregister(mlxsw_pci->core, false);
2316 	mlxsw_pci_cmd_fini(mlxsw_pci);
2317 	iounmap(mlxsw_pci->hw_addr);
2318 	pci_release_regions(mlxsw_pci->pdev);
2319 	pci_disable_device(mlxsw_pci->pdev);
2320 	kfree(mlxsw_pci);
2321 }
2322 
2323 static void mlxsw_pci_reset_prepare(struct pci_dev *pdev)
2324 {
2325 	struct mlxsw_pci *mlxsw_pci = pci_get_drvdata(pdev);
2326 
2327 	mlxsw_core_bus_device_unregister(mlxsw_pci->core, false);
2328 }
2329 
2330 static void mlxsw_pci_reset_done(struct pci_dev *pdev)
2331 {
2332 	struct mlxsw_pci *mlxsw_pci = pci_get_drvdata(pdev);
2333 
2334 	mlxsw_pci->skip_reset = true;
2335 	mlxsw_core_bus_device_register(&mlxsw_pci->bus_info, &mlxsw_pci_bus,
2336 				       mlxsw_pci, false, NULL, NULL);
2337 	mlxsw_pci->skip_reset = false;
2338 }
2339 
2340 static const struct pci_error_handlers mlxsw_pci_err_handler = {
2341 	.reset_prepare = mlxsw_pci_reset_prepare,
2342 	.reset_done = mlxsw_pci_reset_done,
2343 };
2344 
2345 int mlxsw_pci_driver_register(struct pci_driver *pci_driver)
2346 {
2347 	pci_driver->probe = mlxsw_pci_probe;
2348 	pci_driver->remove = mlxsw_pci_remove;
2349 	pci_driver->shutdown = mlxsw_pci_remove;
2350 	pci_driver->err_handler = &mlxsw_pci_err_handler;
2351 	return pci_register_driver(pci_driver);
2352 }
2353 EXPORT_SYMBOL(mlxsw_pci_driver_register);
2354 
2355 void mlxsw_pci_driver_unregister(struct pci_driver *pci_driver)
2356 {
2357 	pci_unregister_driver(pci_driver);
2358 }
2359 EXPORT_SYMBOL(mlxsw_pci_driver_unregister);
2360 
2361 static int __init mlxsw_pci_module_init(void)
2362 {
2363 	return 0;
2364 }
2365 
2366 static void __exit mlxsw_pci_module_exit(void)
2367 {
2368 }
2369 
2370 module_init(mlxsw_pci_module_init);
2371 module_exit(mlxsw_pci_module_exit);
2372 
2373 MODULE_LICENSE("Dual BSD/GPL");
2374 MODULE_AUTHOR("Jiri Pirko <jiri@mellanox.com>");
2375 MODULE_DESCRIPTION("Mellanox switch PCI interface driver");
2376