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