1 // SPDX-License-Identifier: (GPL-2.0-only OR BSD-3-Clause)
2 /* QLogic qed NIC Driver
3  * Copyright (c) 2015-2017  QLogic Corporation
4  * Copyright (c) 2019-2020 Marvell International Ltd.
5  */
6 
7 #include <linux/types.h>
8 #include <linux/io.h>
9 #include <linux/delay.h>
10 #include <linux/dma-mapping.h>
11 #include <linux/errno.h>
12 #include <linux/kernel.h>
13 #include <linux/list.h>
14 #include <linux/mutex.h>
15 #include <linux/pci.h>
16 #include <linux/slab.h>
17 #include <linux/spinlock.h>
18 #include <linux/string.h>
19 #include <linux/qed/qed_chain.h>
20 #include "qed.h"
21 #include "qed_hsi.h"
22 #include "qed_hw.h"
23 #include "qed_reg_addr.h"
24 #include "qed_sriov.h"
25 
26 #define QED_BAR_ACQUIRE_TIMEOUT 1000
27 
28 /* Invalid values */
29 #define QED_BAR_INVALID_OFFSET          (cpu_to_le32(-1))
30 
31 struct qed_ptt {
32 	struct list_head	list_entry;
33 	unsigned int		idx;
34 	struct pxp_ptt_entry	pxp;
35 	u8			hwfn_id;
36 };
37 
38 struct qed_ptt_pool {
39 	struct list_head	free_list;
40 	spinlock_t		lock; /* ptt synchronized access */
41 	struct qed_ptt		ptts[PXP_EXTERNAL_BAR_PF_WINDOW_NUM];
42 };
43 
44 int qed_ptt_pool_alloc(struct qed_hwfn *p_hwfn)
45 {
46 	struct qed_ptt_pool *p_pool = kmalloc(sizeof(*p_pool), GFP_KERNEL);
47 	int i;
48 
49 	if (!p_pool)
50 		return -ENOMEM;
51 
52 	INIT_LIST_HEAD(&p_pool->free_list);
53 	for (i = 0; i < PXP_EXTERNAL_BAR_PF_WINDOW_NUM; i++) {
54 		p_pool->ptts[i].idx = i;
55 		p_pool->ptts[i].pxp.offset = QED_BAR_INVALID_OFFSET;
56 		p_pool->ptts[i].pxp.pretend.control = 0;
57 		p_pool->ptts[i].hwfn_id = p_hwfn->my_id;
58 		if (i >= RESERVED_PTT_MAX)
59 			list_add(&p_pool->ptts[i].list_entry,
60 				 &p_pool->free_list);
61 	}
62 
63 	p_hwfn->p_ptt_pool = p_pool;
64 	spin_lock_init(&p_pool->lock);
65 
66 	return 0;
67 }
68 
69 void qed_ptt_invalidate(struct qed_hwfn *p_hwfn)
70 {
71 	struct qed_ptt *p_ptt;
72 	int i;
73 
74 	for (i = 0; i < PXP_EXTERNAL_BAR_PF_WINDOW_NUM; i++) {
75 		p_ptt = &p_hwfn->p_ptt_pool->ptts[i];
76 		p_ptt->pxp.offset = QED_BAR_INVALID_OFFSET;
77 	}
78 }
79 
80 void qed_ptt_pool_free(struct qed_hwfn *p_hwfn)
81 {
82 	kfree(p_hwfn->p_ptt_pool);
83 	p_hwfn->p_ptt_pool = NULL;
84 }
85 
86 struct qed_ptt *qed_ptt_acquire(struct qed_hwfn *p_hwfn)
87 {
88 	struct qed_ptt *p_ptt;
89 	unsigned int i;
90 
91 	/* Take the free PTT from the list */
92 	for (i = 0; i < QED_BAR_ACQUIRE_TIMEOUT; i++) {
93 		spin_lock_bh(&p_hwfn->p_ptt_pool->lock);
94 
95 		if (!list_empty(&p_hwfn->p_ptt_pool->free_list)) {
96 			p_ptt = list_first_entry(&p_hwfn->p_ptt_pool->free_list,
97 						 struct qed_ptt, list_entry);
98 			list_del(&p_ptt->list_entry);
99 
100 			spin_unlock_bh(&p_hwfn->p_ptt_pool->lock);
101 
102 			DP_VERBOSE(p_hwfn, NETIF_MSG_HW,
103 				   "allocated ptt %d\n", p_ptt->idx);
104 			return p_ptt;
105 		}
106 
107 		spin_unlock_bh(&p_hwfn->p_ptt_pool->lock);
108 		usleep_range(1000, 2000);
109 	}
110 
111 	DP_NOTICE(p_hwfn, "PTT acquire timeout - failed to allocate PTT\n");
112 	return NULL;
113 }
114 
115 void qed_ptt_release(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
116 {
117 	spin_lock_bh(&p_hwfn->p_ptt_pool->lock);
118 	list_add(&p_ptt->list_entry, &p_hwfn->p_ptt_pool->free_list);
119 	spin_unlock_bh(&p_hwfn->p_ptt_pool->lock);
120 }
121 
122 u32 qed_ptt_get_hw_addr(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
123 {
124 	/* The HW is using DWORDS and we need to translate it to Bytes */
125 	return le32_to_cpu(p_ptt->pxp.offset) << 2;
126 }
127 
128 static u32 qed_ptt_config_addr(struct qed_ptt *p_ptt)
129 {
130 	return PXP_PF_WINDOW_ADMIN_PER_PF_START +
131 	       p_ptt->idx * sizeof(struct pxp_ptt_entry);
132 }
133 
134 u32 qed_ptt_get_bar_addr(struct qed_ptt *p_ptt)
135 {
136 	return PXP_EXTERNAL_BAR_PF_WINDOW_START +
137 	       p_ptt->idx * PXP_EXTERNAL_BAR_PF_WINDOW_SINGLE_SIZE;
138 }
139 
140 void qed_ptt_set_win(struct qed_hwfn *p_hwfn,
141 		     struct qed_ptt *p_ptt, u32 new_hw_addr)
142 {
143 	u32 prev_hw_addr;
144 
145 	prev_hw_addr = qed_ptt_get_hw_addr(p_hwfn, p_ptt);
146 
147 	if (new_hw_addr == prev_hw_addr)
148 		return;
149 
150 	/* Update PTT entery in admin window */
151 	DP_VERBOSE(p_hwfn, NETIF_MSG_HW,
152 		   "Updating PTT entry %d to offset 0x%x\n",
153 		   p_ptt->idx, new_hw_addr);
154 
155 	/* The HW is using DWORDS and the address is in Bytes */
156 	p_ptt->pxp.offset = cpu_to_le32(new_hw_addr >> 2);
157 
158 	REG_WR(p_hwfn,
159 	       qed_ptt_config_addr(p_ptt) +
160 	       offsetof(struct pxp_ptt_entry, offset),
161 	       le32_to_cpu(p_ptt->pxp.offset));
162 }
163 
164 static u32 qed_set_ptt(struct qed_hwfn *p_hwfn,
165 		       struct qed_ptt *p_ptt, u32 hw_addr)
166 {
167 	u32 win_hw_addr = qed_ptt_get_hw_addr(p_hwfn, p_ptt);
168 	u32 offset;
169 
170 	offset = hw_addr - win_hw_addr;
171 
172 	if (p_ptt->hwfn_id != p_hwfn->my_id)
173 		DP_NOTICE(p_hwfn,
174 			  "ptt[%d] of hwfn[%02x] is used by hwfn[%02x]!\n",
175 			  p_ptt->idx, p_ptt->hwfn_id, p_hwfn->my_id);
176 
177 	/* Verify the address is within the window */
178 	if (hw_addr < win_hw_addr ||
179 	    offset >= PXP_EXTERNAL_BAR_PF_WINDOW_SINGLE_SIZE) {
180 		qed_ptt_set_win(p_hwfn, p_ptt, hw_addr);
181 		offset = 0;
182 	}
183 
184 	return qed_ptt_get_bar_addr(p_ptt) + offset;
185 }
186 
187 struct qed_ptt *qed_get_reserved_ptt(struct qed_hwfn *p_hwfn,
188 				     enum reserved_ptts ptt_idx)
189 {
190 	if (ptt_idx >= RESERVED_PTT_MAX) {
191 		DP_NOTICE(p_hwfn,
192 			  "Requested PTT %d is out of range\n", ptt_idx);
193 		return NULL;
194 	}
195 
196 	return &p_hwfn->p_ptt_pool->ptts[ptt_idx];
197 }
198 
199 void qed_wr(struct qed_hwfn *p_hwfn,
200 	    struct qed_ptt *p_ptt,
201 	    u32 hw_addr, u32 val)
202 {
203 	u32 bar_addr = qed_set_ptt(p_hwfn, p_ptt, hw_addr);
204 
205 	REG_WR(p_hwfn, bar_addr, val);
206 	DP_VERBOSE(p_hwfn, NETIF_MSG_HW,
207 		   "bar_addr 0x%x, hw_addr 0x%x, val 0x%x\n",
208 		   bar_addr, hw_addr, val);
209 }
210 
211 u32 qed_rd(struct qed_hwfn *p_hwfn,
212 	   struct qed_ptt *p_ptt,
213 	   u32 hw_addr)
214 {
215 	u32 bar_addr = qed_set_ptt(p_hwfn, p_ptt, hw_addr);
216 	u32 val = REG_RD(p_hwfn, bar_addr);
217 
218 	DP_VERBOSE(p_hwfn, NETIF_MSG_HW,
219 		   "bar_addr 0x%x, hw_addr 0x%x, val 0x%x\n",
220 		   bar_addr, hw_addr, val);
221 
222 	return val;
223 }
224 
225 static void qed_memcpy_hw(struct qed_hwfn *p_hwfn,
226 			  struct qed_ptt *p_ptt,
227 			  void *addr, u32 hw_addr, size_t n, bool to_device)
228 {
229 	u32 dw_count, *host_addr, hw_offset;
230 	size_t quota, done = 0;
231 	u32 __iomem *reg_addr;
232 
233 	while (done < n) {
234 		quota = min_t(size_t, n - done,
235 			      PXP_EXTERNAL_BAR_PF_WINDOW_SINGLE_SIZE);
236 
237 		if (IS_PF(p_hwfn->cdev)) {
238 			qed_ptt_set_win(p_hwfn, p_ptt, hw_addr + done);
239 			hw_offset = qed_ptt_get_bar_addr(p_ptt);
240 		} else {
241 			hw_offset = hw_addr + done;
242 		}
243 
244 		dw_count = quota / 4;
245 		host_addr = (u32 *)((u8 *)addr + done);
246 		reg_addr = (u32 __iomem *)REG_ADDR(p_hwfn, hw_offset);
247 		if (to_device)
248 			while (dw_count--)
249 				DIRECT_REG_WR(reg_addr++, *host_addr++);
250 		else
251 			while (dw_count--)
252 				*host_addr++ = DIRECT_REG_RD(reg_addr++);
253 
254 		done += quota;
255 	}
256 }
257 
258 void qed_memcpy_from(struct qed_hwfn *p_hwfn,
259 		     struct qed_ptt *p_ptt, void *dest, u32 hw_addr, size_t n)
260 {
261 	DP_VERBOSE(p_hwfn, NETIF_MSG_HW,
262 		   "hw_addr 0x%x, dest %p hw_addr 0x%x, size %lu\n",
263 		   hw_addr, dest, hw_addr, (unsigned long)n);
264 
265 	qed_memcpy_hw(p_hwfn, p_ptt, dest, hw_addr, n, false);
266 }
267 
268 void qed_memcpy_to(struct qed_hwfn *p_hwfn,
269 		   struct qed_ptt *p_ptt, u32 hw_addr, void *src, size_t n)
270 {
271 	DP_VERBOSE(p_hwfn, NETIF_MSG_HW,
272 		   "hw_addr 0x%x, hw_addr 0x%x, src %p size %lu\n",
273 		   hw_addr, hw_addr, src, (unsigned long)n);
274 
275 	qed_memcpy_hw(p_hwfn, p_ptt, src, hw_addr, n, true);
276 }
277 
278 void qed_fid_pretend(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt, u16 fid)
279 {
280 	u16 control = 0;
281 
282 	SET_FIELD(control, PXP_PRETEND_CMD_IS_CONCRETE, 1);
283 	SET_FIELD(control, PXP_PRETEND_CMD_PRETEND_FUNCTION, 1);
284 
285 	/* Every pretend undos previous pretends, including
286 	 * previous port pretend.
287 	 */
288 	SET_FIELD(control, PXP_PRETEND_CMD_PORT, 0);
289 	SET_FIELD(control, PXP_PRETEND_CMD_USE_PORT, 0);
290 	SET_FIELD(control, PXP_PRETEND_CMD_PRETEND_PORT, 1);
291 
292 	if (!GET_FIELD(fid, PXP_CONCRETE_FID_VFVALID))
293 		fid = GET_FIELD(fid, PXP_CONCRETE_FID_PFID);
294 
295 	p_ptt->pxp.pretend.control = cpu_to_le16(control);
296 	p_ptt->pxp.pretend.fid.concrete_fid.fid = cpu_to_le16(fid);
297 
298 	REG_WR(p_hwfn,
299 	       qed_ptt_config_addr(p_ptt) +
300 	       offsetof(struct pxp_ptt_entry, pretend),
301 	       *(u32 *)&p_ptt->pxp.pretend);
302 }
303 
304 void qed_port_pretend(struct qed_hwfn *p_hwfn,
305 		      struct qed_ptt *p_ptt, u8 port_id)
306 {
307 	u16 control = 0;
308 
309 	SET_FIELD(control, PXP_PRETEND_CMD_PORT, port_id);
310 	SET_FIELD(control, PXP_PRETEND_CMD_USE_PORT, 1);
311 	SET_FIELD(control, PXP_PRETEND_CMD_PRETEND_PORT, 1);
312 
313 	p_ptt->pxp.pretend.control = cpu_to_le16(control);
314 
315 	REG_WR(p_hwfn,
316 	       qed_ptt_config_addr(p_ptt) +
317 	       offsetof(struct pxp_ptt_entry, pretend),
318 	       *(u32 *)&p_ptt->pxp.pretend);
319 }
320 
321 void qed_port_unpretend(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
322 {
323 	u16 control = 0;
324 
325 	SET_FIELD(control, PXP_PRETEND_CMD_PORT, 0);
326 	SET_FIELD(control, PXP_PRETEND_CMD_USE_PORT, 0);
327 	SET_FIELD(control, PXP_PRETEND_CMD_PRETEND_PORT, 1);
328 
329 	p_ptt->pxp.pretend.control = cpu_to_le16(control);
330 
331 	REG_WR(p_hwfn,
332 	       qed_ptt_config_addr(p_ptt) +
333 	       offsetof(struct pxp_ptt_entry, pretend),
334 	       *(u32 *)&p_ptt->pxp.pretend);
335 }
336 
337 void qed_port_fid_pretend(struct qed_hwfn *p_hwfn,
338 			  struct qed_ptt *p_ptt, u8 port_id, u16 fid)
339 {
340 	u16 control = 0;
341 
342 	SET_FIELD(control, PXP_PRETEND_CMD_PORT, port_id);
343 	SET_FIELD(control, PXP_PRETEND_CMD_USE_PORT, 1);
344 	SET_FIELD(control, PXP_PRETEND_CMD_PRETEND_PORT, 1);
345 	SET_FIELD(control, PXP_PRETEND_CMD_IS_CONCRETE, 1);
346 	SET_FIELD(control, PXP_PRETEND_CMD_PRETEND_FUNCTION, 1);
347 	if (!GET_FIELD(fid, PXP_CONCRETE_FID_VFVALID))
348 		fid = GET_FIELD(fid, PXP_CONCRETE_FID_PFID);
349 	p_ptt->pxp.pretend.control = cpu_to_le16(control);
350 	p_ptt->pxp.pretend.fid.concrete_fid.fid = cpu_to_le16(fid);
351 	REG_WR(p_hwfn,
352 	       qed_ptt_config_addr(p_ptt) +
353 	       offsetof(struct pxp_ptt_entry, pretend),
354 	       *(u32 *)&p_ptt->pxp.pretend);
355 }
356 
357 u32 qed_vfid_to_concrete(struct qed_hwfn *p_hwfn, u8 vfid)
358 {
359 	u32 concrete_fid = 0;
360 
361 	SET_FIELD(concrete_fid, PXP_CONCRETE_FID_PFID, p_hwfn->rel_pf_id);
362 	SET_FIELD(concrete_fid, PXP_CONCRETE_FID_VFID, vfid);
363 	SET_FIELD(concrete_fid, PXP_CONCRETE_FID_VFVALID, 1);
364 
365 	return concrete_fid;
366 }
367 
368 /* DMAE */
369 #define QED_DMAE_FLAGS_IS_SET(params, flag) \
370 	((params) != NULL && GET_FIELD((params)->flags, QED_DMAE_PARAMS_##flag))
371 
372 static void qed_dmae_opcode(struct qed_hwfn *p_hwfn,
373 			    const u8 is_src_type_grc,
374 			    const u8 is_dst_type_grc,
375 			    struct qed_dmae_params *p_params)
376 {
377 	u8 src_pfid, dst_pfid, port_id;
378 	u16 opcode_b = 0;
379 	u32 opcode = 0;
380 
381 	/* Whether the source is the PCIe or the GRC.
382 	 * 0- The source is the PCIe
383 	 * 1- The source is the GRC.
384 	 */
385 	SET_FIELD(opcode, DMAE_CMD_SRC,
386 		  (is_src_type_grc ? dmae_cmd_src_grc : dmae_cmd_src_pcie));
387 	src_pfid = QED_DMAE_FLAGS_IS_SET(p_params, SRC_PF_VALID) ?
388 	    p_params->src_pfid : p_hwfn->rel_pf_id;
389 	SET_FIELD(opcode, DMAE_CMD_SRC_PF_ID, src_pfid);
390 
391 	/* The destination of the DMA can be: 0-None 1-PCIe 2-GRC 3-None */
392 	SET_FIELD(opcode, DMAE_CMD_DST,
393 		  (is_dst_type_grc ? dmae_cmd_dst_grc : dmae_cmd_dst_pcie));
394 	dst_pfid = QED_DMAE_FLAGS_IS_SET(p_params, DST_PF_VALID) ?
395 	    p_params->dst_pfid : p_hwfn->rel_pf_id;
396 	SET_FIELD(opcode, DMAE_CMD_DST_PF_ID, dst_pfid);
397 
398 
399 	/* Whether to write a completion word to the completion destination:
400 	 * 0-Do not write a completion word
401 	 * 1-Write the completion word
402 	 */
403 	SET_FIELD(opcode, DMAE_CMD_COMP_WORD_EN, 1);
404 	SET_FIELD(opcode, DMAE_CMD_SRC_ADDR_RESET, 1);
405 
406 	if (QED_DMAE_FLAGS_IS_SET(p_params, COMPLETION_DST))
407 		SET_FIELD(opcode, DMAE_CMD_COMP_FUNC, 1);
408 
409 	/* swapping mode 3 - big endian */
410 	SET_FIELD(opcode, DMAE_CMD_ENDIANITY_MODE, DMAE_CMD_ENDIANITY);
411 
412 	port_id = (QED_DMAE_FLAGS_IS_SET(p_params, PORT_VALID)) ?
413 	    p_params->port_id : p_hwfn->port_id;
414 	SET_FIELD(opcode, DMAE_CMD_PORT_ID, port_id);
415 
416 	/* reset source address in next go */
417 	SET_FIELD(opcode, DMAE_CMD_SRC_ADDR_RESET, 1);
418 
419 	/* reset dest address in next go */
420 	SET_FIELD(opcode, DMAE_CMD_DST_ADDR_RESET, 1);
421 
422 	/* SRC/DST VFID: all 1's - pf, otherwise VF id */
423 	if (QED_DMAE_FLAGS_IS_SET(p_params, SRC_VF_VALID)) {
424 		SET_FIELD(opcode, DMAE_CMD_SRC_VF_ID_VALID, 1);
425 		SET_FIELD(opcode_b, DMAE_CMD_SRC_VF_ID, p_params->src_vfid);
426 	} else {
427 		SET_FIELD(opcode_b, DMAE_CMD_SRC_VF_ID, 0xFF);
428 	}
429 	if (QED_DMAE_FLAGS_IS_SET(p_params, DST_VF_VALID)) {
430 		SET_FIELD(opcode, DMAE_CMD_DST_VF_ID_VALID, 1);
431 		SET_FIELD(opcode_b, DMAE_CMD_DST_VF_ID, p_params->dst_vfid);
432 	} else {
433 		SET_FIELD(opcode_b, DMAE_CMD_DST_VF_ID, 0xFF);
434 	}
435 
436 	p_hwfn->dmae_info.p_dmae_cmd->opcode = cpu_to_le32(opcode);
437 	p_hwfn->dmae_info.p_dmae_cmd->opcode_b = cpu_to_le16(opcode_b);
438 }
439 
440 u32 qed_dmae_idx_to_go_cmd(u8 idx)
441 {
442 	/* All the DMAE 'go' registers form an array in internal memory */
443 	return DMAE_REG_GO_C0 + (idx << 2);
444 }
445 
446 static int qed_dmae_post_command(struct qed_hwfn *p_hwfn,
447 				 struct qed_ptt *p_ptt)
448 {
449 	struct dmae_cmd *p_command = p_hwfn->dmae_info.p_dmae_cmd;
450 	u8 idx_cmd = p_hwfn->dmae_info.channel, i;
451 	int qed_status = 0;
452 
453 	/* verify address is not NULL */
454 	if ((((!p_command->dst_addr_lo) && (!p_command->dst_addr_hi)) ||
455 	     ((!p_command->src_addr_lo) && (!p_command->src_addr_hi)))) {
456 		DP_NOTICE(p_hwfn,
457 			  "source or destination address 0 idx_cmd=%d\n"
458 			  "opcode = [0x%08x,0x%04x] len=0x%x src=0x%x:%x dst=0x%x:%x\n",
459 			  idx_cmd,
460 			  le32_to_cpu(p_command->opcode),
461 			  le16_to_cpu(p_command->opcode_b),
462 			  le16_to_cpu(p_command->length_dw),
463 			  le32_to_cpu(p_command->src_addr_hi),
464 			  le32_to_cpu(p_command->src_addr_lo),
465 			  le32_to_cpu(p_command->dst_addr_hi),
466 			  le32_to_cpu(p_command->dst_addr_lo));
467 
468 		return -EINVAL;
469 	}
470 
471 	DP_VERBOSE(p_hwfn,
472 		   NETIF_MSG_HW,
473 		   "Posting DMAE command [idx %d]: opcode = [0x%08x,0x%04x] len=0x%x src=0x%x:%x dst=0x%x:%x\n",
474 		   idx_cmd,
475 		   le32_to_cpu(p_command->opcode),
476 		   le16_to_cpu(p_command->opcode_b),
477 		   le16_to_cpu(p_command->length_dw),
478 		   le32_to_cpu(p_command->src_addr_hi),
479 		   le32_to_cpu(p_command->src_addr_lo),
480 		   le32_to_cpu(p_command->dst_addr_hi),
481 		   le32_to_cpu(p_command->dst_addr_lo));
482 
483 	/* Copy the command to DMAE - need to do it before every call
484 	 * for source/dest address no reset.
485 	 * The first 9 DWs are the command registers, the 10 DW is the
486 	 * GO register, and the rest are result registers
487 	 * (which are read only by the client).
488 	 */
489 	for (i = 0; i < DMAE_CMD_SIZE; i++) {
490 		u32 data = (i < DMAE_CMD_SIZE_TO_FILL) ?
491 			   *(((u32 *)p_command) + i) : 0;
492 
493 		qed_wr(p_hwfn, p_ptt,
494 		       DMAE_REG_CMD_MEM +
495 		       (idx_cmd * DMAE_CMD_SIZE * sizeof(u32)) +
496 		       (i * sizeof(u32)), data);
497 	}
498 
499 	qed_wr(p_hwfn, p_ptt, qed_dmae_idx_to_go_cmd(idx_cmd), DMAE_GO_VALUE);
500 
501 	return qed_status;
502 }
503 
504 int qed_dmae_info_alloc(struct qed_hwfn *p_hwfn)
505 {
506 	dma_addr_t *p_addr = &p_hwfn->dmae_info.completion_word_phys_addr;
507 	struct dmae_cmd **p_cmd = &p_hwfn->dmae_info.p_dmae_cmd;
508 	u32 **p_buff = &p_hwfn->dmae_info.p_intermediate_buffer;
509 	u32 **p_comp = &p_hwfn->dmae_info.p_completion_word;
510 
511 	*p_comp = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
512 				     sizeof(u32), p_addr, GFP_KERNEL);
513 	if (!*p_comp)
514 		goto err;
515 
516 	p_addr = &p_hwfn->dmae_info.dmae_cmd_phys_addr;
517 	*p_cmd = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
518 				    sizeof(struct dmae_cmd),
519 				    p_addr, GFP_KERNEL);
520 	if (!*p_cmd)
521 		goto err;
522 
523 	p_addr = &p_hwfn->dmae_info.intermediate_buffer_phys_addr;
524 	*p_buff = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
525 				     sizeof(u32) * DMAE_MAX_RW_SIZE,
526 				     p_addr, GFP_KERNEL);
527 	if (!*p_buff)
528 		goto err;
529 
530 	p_hwfn->dmae_info.channel = p_hwfn->rel_pf_id;
531 
532 	return 0;
533 err:
534 	qed_dmae_info_free(p_hwfn);
535 	return -ENOMEM;
536 }
537 
538 void qed_dmae_info_free(struct qed_hwfn *p_hwfn)
539 {
540 	dma_addr_t p_phys;
541 
542 	/* Just make sure no one is in the middle */
543 	mutex_lock(&p_hwfn->dmae_info.mutex);
544 
545 	if (p_hwfn->dmae_info.p_completion_word) {
546 		p_phys = p_hwfn->dmae_info.completion_word_phys_addr;
547 		dma_free_coherent(&p_hwfn->cdev->pdev->dev,
548 				  sizeof(u32),
549 				  p_hwfn->dmae_info.p_completion_word, p_phys);
550 		p_hwfn->dmae_info.p_completion_word = NULL;
551 	}
552 
553 	if (p_hwfn->dmae_info.p_dmae_cmd) {
554 		p_phys = p_hwfn->dmae_info.dmae_cmd_phys_addr;
555 		dma_free_coherent(&p_hwfn->cdev->pdev->dev,
556 				  sizeof(struct dmae_cmd),
557 				  p_hwfn->dmae_info.p_dmae_cmd, p_phys);
558 		p_hwfn->dmae_info.p_dmae_cmd = NULL;
559 	}
560 
561 	if (p_hwfn->dmae_info.p_intermediate_buffer) {
562 		p_phys = p_hwfn->dmae_info.intermediate_buffer_phys_addr;
563 		dma_free_coherent(&p_hwfn->cdev->pdev->dev,
564 				  sizeof(u32) * DMAE_MAX_RW_SIZE,
565 				  p_hwfn->dmae_info.p_intermediate_buffer,
566 				  p_phys);
567 		p_hwfn->dmae_info.p_intermediate_buffer = NULL;
568 	}
569 
570 	mutex_unlock(&p_hwfn->dmae_info.mutex);
571 }
572 
573 static int qed_dmae_operation_wait(struct qed_hwfn *p_hwfn)
574 {
575 	u32 wait_cnt_limit = 10000, wait_cnt = 0;
576 	int qed_status = 0;
577 
578 	barrier();
579 	while (*p_hwfn->dmae_info.p_completion_word != DMAE_COMPLETION_VAL) {
580 		udelay(DMAE_MIN_WAIT_TIME);
581 		if (++wait_cnt > wait_cnt_limit) {
582 			DP_NOTICE(p_hwfn->cdev,
583 				  "Timed-out waiting for operation to complete. Completion word is 0x%08x expected 0x%08x.\n",
584 				  *p_hwfn->dmae_info.p_completion_word,
585 				 DMAE_COMPLETION_VAL);
586 			qed_status = -EBUSY;
587 			break;
588 		}
589 
590 		/* to sync the completion_word since we are not
591 		 * using the volatile keyword for p_completion_word
592 		 */
593 		barrier();
594 	}
595 
596 	if (qed_status == 0)
597 		*p_hwfn->dmae_info.p_completion_word = 0;
598 
599 	return qed_status;
600 }
601 
602 static int qed_dmae_execute_sub_operation(struct qed_hwfn *p_hwfn,
603 					  struct qed_ptt *p_ptt,
604 					  u64 src_addr,
605 					  u64 dst_addr,
606 					  u8 src_type,
607 					  u8 dst_type,
608 					  u32 length_dw)
609 {
610 	dma_addr_t phys = p_hwfn->dmae_info.intermediate_buffer_phys_addr;
611 	struct dmae_cmd *cmd = p_hwfn->dmae_info.p_dmae_cmd;
612 	int qed_status = 0;
613 
614 	switch (src_type) {
615 	case QED_DMAE_ADDRESS_GRC:
616 	case QED_DMAE_ADDRESS_HOST_PHYS:
617 		cmd->src_addr_hi = cpu_to_le32(upper_32_bits(src_addr));
618 		cmd->src_addr_lo = cpu_to_le32(lower_32_bits(src_addr));
619 		break;
620 	/* for virtual source addresses we use the intermediate buffer. */
621 	case QED_DMAE_ADDRESS_HOST_VIRT:
622 		cmd->src_addr_hi = cpu_to_le32(upper_32_bits(phys));
623 		cmd->src_addr_lo = cpu_to_le32(lower_32_bits(phys));
624 		memcpy(&p_hwfn->dmae_info.p_intermediate_buffer[0],
625 		       (void *)(uintptr_t)src_addr,
626 		       length_dw * sizeof(u32));
627 		break;
628 	default:
629 		return -EINVAL;
630 	}
631 
632 	switch (dst_type) {
633 	case QED_DMAE_ADDRESS_GRC:
634 	case QED_DMAE_ADDRESS_HOST_PHYS:
635 		cmd->dst_addr_hi = cpu_to_le32(upper_32_bits(dst_addr));
636 		cmd->dst_addr_lo = cpu_to_le32(lower_32_bits(dst_addr));
637 		break;
638 	/* for virtual source addresses we use the intermediate buffer. */
639 	case QED_DMAE_ADDRESS_HOST_VIRT:
640 		cmd->dst_addr_hi = cpu_to_le32(upper_32_bits(phys));
641 		cmd->dst_addr_lo = cpu_to_le32(lower_32_bits(phys));
642 		break;
643 	default:
644 		return -EINVAL;
645 	}
646 
647 	cmd->length_dw = cpu_to_le16((u16)length_dw);
648 
649 	qed_dmae_post_command(p_hwfn, p_ptt);
650 
651 	qed_status = qed_dmae_operation_wait(p_hwfn);
652 
653 	if (qed_status) {
654 		DP_NOTICE(p_hwfn,
655 			  "qed_dmae_host2grc: Wait Failed. source_addr 0x%llx, grc_addr 0x%llx, size_in_dwords 0x%x\n",
656 			  src_addr, dst_addr, length_dw);
657 		return qed_status;
658 	}
659 
660 	if (dst_type == QED_DMAE_ADDRESS_HOST_VIRT)
661 		memcpy((void *)(uintptr_t)(dst_addr),
662 		       &p_hwfn->dmae_info.p_intermediate_buffer[0],
663 		       length_dw * sizeof(u32));
664 
665 	return 0;
666 }
667 
668 static int qed_dmae_execute_command(struct qed_hwfn *p_hwfn,
669 				    struct qed_ptt *p_ptt,
670 				    u64 src_addr, u64 dst_addr,
671 				    u8 src_type, u8 dst_type,
672 				    u32 size_in_dwords,
673 				    struct qed_dmae_params *p_params)
674 {
675 	dma_addr_t phys = p_hwfn->dmae_info.completion_word_phys_addr;
676 	u16 length_cur = 0, i = 0, cnt_split = 0, length_mod = 0;
677 	struct dmae_cmd *cmd = p_hwfn->dmae_info.p_dmae_cmd;
678 	u64 src_addr_split = 0, dst_addr_split = 0;
679 	u16 length_limit = DMAE_MAX_RW_SIZE;
680 	int qed_status = 0;
681 	u32 offset = 0;
682 
683 	if (p_hwfn->cdev->recov_in_prog) {
684 		DP_VERBOSE(p_hwfn,
685 			   NETIF_MSG_HW,
686 			   "Recovery is in progress. Avoid DMAE transaction [{src: addr 0x%llx, type %d}, {dst: addr 0x%llx, type %d}, size %d].\n",
687 			   src_addr, src_type, dst_addr, dst_type,
688 			   size_in_dwords);
689 
690 		/* Let the flow complete w/o any error handling */
691 		return 0;
692 	}
693 
694 	qed_dmae_opcode(p_hwfn,
695 			(src_type == QED_DMAE_ADDRESS_GRC),
696 			(dst_type == QED_DMAE_ADDRESS_GRC),
697 			p_params);
698 
699 	cmd->comp_addr_lo = cpu_to_le32(lower_32_bits(phys));
700 	cmd->comp_addr_hi = cpu_to_le32(upper_32_bits(phys));
701 	cmd->comp_val = cpu_to_le32(DMAE_COMPLETION_VAL);
702 
703 	/* Check if the grc_addr is valid like < MAX_GRC_OFFSET */
704 	cnt_split = size_in_dwords / length_limit;
705 	length_mod = size_in_dwords % length_limit;
706 
707 	src_addr_split = src_addr;
708 	dst_addr_split = dst_addr;
709 
710 	for (i = 0; i <= cnt_split; i++) {
711 		offset = length_limit * i;
712 
713 		if (!QED_DMAE_FLAGS_IS_SET(p_params, RW_REPL_SRC)) {
714 			if (src_type == QED_DMAE_ADDRESS_GRC)
715 				src_addr_split = src_addr + offset;
716 			else
717 				src_addr_split = src_addr + (offset * 4);
718 		}
719 
720 		if (dst_type == QED_DMAE_ADDRESS_GRC)
721 			dst_addr_split = dst_addr + offset;
722 		else
723 			dst_addr_split = dst_addr + (offset * 4);
724 
725 		length_cur = (cnt_split == i) ? length_mod : length_limit;
726 
727 		/* might be zero on last iteration */
728 		if (!length_cur)
729 			continue;
730 
731 		qed_status = qed_dmae_execute_sub_operation(p_hwfn,
732 							    p_ptt,
733 							    src_addr_split,
734 							    dst_addr_split,
735 							    src_type,
736 							    dst_type,
737 							    length_cur);
738 		if (qed_status) {
739 			qed_hw_err_notify(p_hwfn, p_ptt, QED_HW_ERR_DMAE_FAIL,
740 					  "qed_dmae_execute_sub_operation Failed with error 0x%x. source_addr 0x%llx, destination addr 0x%llx, size_in_dwords 0x%x\n",
741 					  qed_status, src_addr,
742 					  dst_addr, length_cur);
743 			break;
744 		}
745 	}
746 
747 	return qed_status;
748 }
749 
750 int qed_dmae_host2grc(struct qed_hwfn *p_hwfn,
751 		      struct qed_ptt *p_ptt,
752 		      u64 source_addr, u32 grc_addr, u32 size_in_dwords,
753 		      struct qed_dmae_params *p_params)
754 {
755 	u32 grc_addr_in_dw = grc_addr / sizeof(u32);
756 	int rc;
757 
758 
759 	mutex_lock(&p_hwfn->dmae_info.mutex);
760 
761 	rc = qed_dmae_execute_command(p_hwfn, p_ptt, source_addr,
762 				      grc_addr_in_dw,
763 				      QED_DMAE_ADDRESS_HOST_VIRT,
764 				      QED_DMAE_ADDRESS_GRC,
765 				      size_in_dwords, p_params);
766 
767 	mutex_unlock(&p_hwfn->dmae_info.mutex);
768 
769 	return rc;
770 }
771 
772 int qed_dmae_grc2host(struct qed_hwfn *p_hwfn,
773 		      struct qed_ptt *p_ptt,
774 		      u32 grc_addr,
775 		      dma_addr_t dest_addr, u32 size_in_dwords,
776 		      struct qed_dmae_params *p_params)
777 {
778 	u32 grc_addr_in_dw = grc_addr / sizeof(u32);
779 	int rc;
780 
781 
782 	mutex_lock(&p_hwfn->dmae_info.mutex);
783 
784 	rc = qed_dmae_execute_command(p_hwfn, p_ptt, grc_addr_in_dw,
785 				      dest_addr, QED_DMAE_ADDRESS_GRC,
786 				      QED_DMAE_ADDRESS_HOST_VIRT,
787 				      size_in_dwords, p_params);
788 
789 	mutex_unlock(&p_hwfn->dmae_info.mutex);
790 
791 	return rc;
792 }
793 
794 int qed_dmae_host2host(struct qed_hwfn *p_hwfn,
795 		       struct qed_ptt *p_ptt,
796 		       dma_addr_t source_addr,
797 		       dma_addr_t dest_addr,
798 		       u32 size_in_dwords, struct qed_dmae_params *p_params)
799 {
800 	int rc;
801 
802 	mutex_lock(&(p_hwfn->dmae_info.mutex));
803 
804 	rc = qed_dmae_execute_command(p_hwfn, p_ptt, source_addr,
805 				      dest_addr,
806 				      QED_DMAE_ADDRESS_HOST_PHYS,
807 				      QED_DMAE_ADDRESS_HOST_PHYS,
808 				      size_in_dwords, p_params);
809 
810 	mutex_unlock(&(p_hwfn->dmae_info.mutex));
811 
812 	return rc;
813 }
814 
815 void qed_hw_err_notify(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt,
816 		       enum qed_hw_err_type err_type, const char *fmt, ...)
817 {
818 	char buf[QED_HW_ERR_MAX_STR_SIZE];
819 	va_list vl;
820 	int len;
821 
822 	if (fmt) {
823 		va_start(vl, fmt);
824 		len = vsnprintf(buf, QED_HW_ERR_MAX_STR_SIZE, fmt, vl);
825 		va_end(vl);
826 
827 		if (len > QED_HW_ERR_MAX_STR_SIZE - 1)
828 			len = QED_HW_ERR_MAX_STR_SIZE - 1;
829 
830 		DP_NOTICE(p_hwfn, "%s", buf);
831 	}
832 
833 	/* Fan failure cannot be masked by handling of another HW error */
834 	if (p_hwfn->cdev->recov_in_prog &&
835 	    err_type != QED_HW_ERR_FAN_FAIL) {
836 		DP_VERBOSE(p_hwfn,
837 			   NETIF_MSG_DRV,
838 			   "Recovery is in progress. Avoid notifying about HW error %d.\n",
839 			   err_type);
840 		return;
841 	}
842 
843 	qed_hw_error_occurred(p_hwfn, err_type);
844 
845 	if (fmt)
846 		qed_mcp_send_raw_debug_data(p_hwfn, p_ptt, buf, len);
847 }
848 
849 int qed_dmae_sanity(struct qed_hwfn *p_hwfn,
850 		    struct qed_ptt *p_ptt, const char *phase)
851 {
852 	u32 size = PAGE_SIZE / 2, val;
853 	int rc = 0;
854 	dma_addr_t p_phys;
855 	void *p_virt;
856 	u32 *p_tmp;
857 
858 	p_virt = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
859 				    2 * size, &p_phys, GFP_KERNEL);
860 	if (!p_virt) {
861 		DP_NOTICE(p_hwfn,
862 			  "DMAE sanity [%s]: failed to allocate memory\n",
863 			  phase);
864 		return -ENOMEM;
865 	}
866 
867 	/* Fill the bottom half of the allocated memory with a known pattern */
868 	for (p_tmp = (u32 *)p_virt;
869 	     p_tmp < (u32 *)((u8 *)p_virt + size); p_tmp++) {
870 		/* Save the address itself as the value */
871 		val = (u32)(uintptr_t)p_tmp;
872 		*p_tmp = val;
873 	}
874 
875 	/* Zero the top half of the allocated memory */
876 	memset((u8 *)p_virt + size, 0, size);
877 
878 	DP_VERBOSE(p_hwfn,
879 		   QED_MSG_SP,
880 		   "DMAE sanity [%s]: src_addr={phys 0x%llx, virt %p}, dst_addr={phys 0x%llx, virt %p}, size 0x%x\n",
881 		   phase,
882 		   (u64)p_phys,
883 		   p_virt, (u64)(p_phys + size), (u8 *)p_virt + size, size);
884 
885 	rc = qed_dmae_host2host(p_hwfn, p_ptt, p_phys, p_phys + size,
886 				size / 4, NULL);
887 	if (rc) {
888 		DP_NOTICE(p_hwfn,
889 			  "DMAE sanity [%s]: qed_dmae_host2host() failed. rc = %d.\n",
890 			  phase, rc);
891 		goto out;
892 	}
893 
894 	/* Verify that the top half of the allocated memory has the pattern */
895 	for (p_tmp = (u32 *)((u8 *)p_virt + size);
896 	     p_tmp < (u32 *)((u8 *)p_virt + (2 * size)); p_tmp++) {
897 		/* The corresponding address in the bottom half */
898 		val = (u32)(uintptr_t)p_tmp - size;
899 
900 		if (*p_tmp != val) {
901 			DP_NOTICE(p_hwfn,
902 				  "DMAE sanity [%s]: addr={phys 0x%llx, virt %p}, read_val 0x%08x, expected_val 0x%08x\n",
903 				  phase,
904 				  (u64)p_phys + ((u8 *)p_tmp - (u8 *)p_virt),
905 				  p_tmp, *p_tmp, val);
906 			rc = -EINVAL;
907 			goto out;
908 		}
909 	}
910 
911 out:
912 	dma_free_coherent(&p_hwfn->cdev->pdev->dev, 2 * size, p_virt, p_phys);
913 	return rc;
914 }
915