1 /* QLogic qed NIC Driver
2  * Copyright (c) 2015 QLogic Corporation
3  *
4  * This software is available under the terms of the GNU General Public License
5  * (GPL) Version 2, available from the file COPYING in the main directory of
6  * this source tree.
7  */
8 
9 #include <linux/types.h>
10 #include <linux/io.h>
11 #include <linux/delay.h>
12 #include <linux/dma-mapping.h>
13 #include <linux/errno.h>
14 #include <linux/kernel.h>
15 #include <linux/list.h>
16 #include <linux/mutex.h>
17 #include <linux/pci.h>
18 #include <linux/slab.h>
19 #include <linux/spinlock.h>
20 #include <linux/string.h>
21 #include <linux/qed/qed_chain.h>
22 #include "qed.h"
23 #include "qed_hsi.h"
24 #include "qed_hw.h"
25 #include "qed_reg_addr.h"
26 
27 #define QED_BAR_ACQUIRE_TIMEOUT 1000
28 
29 /* Invalid values */
30 #define QED_BAR_INVALID_OFFSET          (cpu_to_le32(-1))
31 
32 struct qed_ptt {
33 	struct list_head	list_entry;
34 	unsigned int		idx;
35 	struct pxp_ptt_entry	pxp;
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),
47 					      GFP_KERNEL);
48 	int i;
49 
50 	if (!p_pool)
51 		return -ENOMEM;
52 
53 	INIT_LIST_HEAD(&p_pool->free_list);
54 	for (i = 0; i < PXP_EXTERNAL_BAR_PF_WINDOW_NUM; i++) {
55 		p_pool->ptts[i].idx = i;
56 		p_pool->ptts[i].pxp.offset = QED_BAR_INVALID_OFFSET;
57 		p_pool->ptts[i].pxp.pretend.control = 0;
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,
116 		     struct qed_ptt *p_ptt)
117 {
118 	spin_lock_bh(&p_hwfn->p_ptt_pool->lock);
119 	list_add(&p_ptt->list_entry, &p_hwfn->p_ptt_pool->free_list);
120 	spin_unlock_bh(&p_hwfn->p_ptt_pool->lock);
121 }
122 
123 u32 qed_ptt_get_hw_addr(struct qed_hwfn *p_hwfn,
124 			struct qed_ptt *p_ptt)
125 {
126 	/* The HW is using DWORDS and we need to translate it to Bytes */
127 	return le32_to_cpu(p_ptt->pxp.offset) << 2;
128 }
129 
130 static u32 qed_ptt_config_addr(struct qed_ptt *p_ptt)
131 {
132 	return PXP_PF_WINDOW_ADMIN_PER_PF_START +
133 	       p_ptt->idx * sizeof(struct pxp_ptt_entry);
134 }
135 
136 u32 qed_ptt_get_bar_addr(struct qed_ptt *p_ptt)
137 {
138 	return PXP_EXTERNAL_BAR_PF_WINDOW_START +
139 	       p_ptt->idx * PXP_EXTERNAL_BAR_PF_WINDOW_SINGLE_SIZE;
140 }
141 
142 void qed_ptt_set_win(struct qed_hwfn *p_hwfn,
143 		     struct qed_ptt *p_ptt,
144 		     u32 new_hw_addr)
145 {
146 	u32 prev_hw_addr;
147 
148 	prev_hw_addr = qed_ptt_get_hw_addr(p_hwfn, p_ptt);
149 
150 	if (new_hw_addr == prev_hw_addr)
151 		return;
152 
153 	/* Update PTT entery in admin window */
154 	DP_VERBOSE(p_hwfn, NETIF_MSG_HW,
155 		   "Updating PTT entry %d to offset 0x%x\n",
156 		   p_ptt->idx, new_hw_addr);
157 
158 	/* The HW is using DWORDS and the address is in Bytes */
159 	p_ptt->pxp.offset = cpu_to_le32(new_hw_addr >> 2);
160 
161 	REG_WR(p_hwfn,
162 	       qed_ptt_config_addr(p_ptt) +
163 	       offsetof(struct pxp_ptt_entry, offset),
164 	       le32_to_cpu(p_ptt->pxp.offset));
165 }
166 
167 static u32 qed_set_ptt(struct qed_hwfn *p_hwfn,
168 		       struct qed_ptt *p_ptt,
169 		       u32 hw_addr)
170 {
171 	u32 win_hw_addr = qed_ptt_get_hw_addr(p_hwfn, p_ptt);
172 	u32 offset;
173 
174 	offset = hw_addr - win_hw_addr;
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,
227 			  u32 hw_addr,
228 			  size_t n,
229 			  bool to_device)
230 {
231 	u32 dw_count, *host_addr, hw_offset;
232 	size_t quota, done = 0;
233 	u32 __iomem *reg_addr;
234 
235 	while (done < n) {
236 		quota = min_t(size_t, n - done,
237 			      PXP_EXTERNAL_BAR_PF_WINDOW_SINGLE_SIZE);
238 
239 		qed_ptt_set_win(p_hwfn, p_ptt, hw_addr + done);
240 		hw_offset = qed_ptt_get_bar_addr(p_ptt);
241 
242 		dw_count = quota / 4;
243 		host_addr = (u32 *)((u8 *)addr + done);
244 		reg_addr = (u32 __iomem *)REG_ADDR(p_hwfn, hw_offset);
245 		if (to_device)
246 			while (dw_count--)
247 				DIRECT_REG_WR(reg_addr++, *host_addr++);
248 		else
249 			while (dw_count--)
250 				*host_addr++ = DIRECT_REG_RD(reg_addr++);
251 
252 		done += quota;
253 	}
254 }
255 
256 void qed_memcpy_from(struct qed_hwfn *p_hwfn,
257 		     struct qed_ptt *p_ptt,
258 		     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,
269 		   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,
279 		     struct qed_ptt *p_ptt,
280 		     u16 fid)
281 {
282 	u16 control = 0;
283 
284 	SET_FIELD(control, PXP_PRETEND_CMD_IS_CONCRETE, 1);
285 	SET_FIELD(control, PXP_PRETEND_CMD_PRETEND_FUNCTION, 1);
286 
287 	/* Every pretend undos previous pretends, including
288 	 * previous port pretend.
289 	 */
290 	SET_FIELD(control, PXP_PRETEND_CMD_PORT, 0);
291 	SET_FIELD(control, PXP_PRETEND_CMD_USE_PORT, 0);
292 	SET_FIELD(control, PXP_PRETEND_CMD_PRETEND_PORT, 1);
293 
294 	if (!GET_FIELD(fid, PXP_CONCRETE_FID_VFVALID))
295 		fid = GET_FIELD(fid, PXP_CONCRETE_FID_PFID);
296 
297 	p_ptt->pxp.pretend.control = cpu_to_le16(control);
298 	p_ptt->pxp.pretend.fid.concrete_fid.fid = cpu_to_le16(fid);
299 
300 	REG_WR(p_hwfn,
301 	       qed_ptt_config_addr(p_ptt) +
302 	       offsetof(struct pxp_ptt_entry, pretend),
303 	       *(u32 *)&p_ptt->pxp.pretend);
304 }
305 
306 void qed_port_pretend(struct qed_hwfn *p_hwfn,
307 		      struct qed_ptt *p_ptt,
308 		      u8 port_id)
309 {
310 	u16 control = 0;
311 
312 	SET_FIELD(control, PXP_PRETEND_CMD_PORT, port_id);
313 	SET_FIELD(control, PXP_PRETEND_CMD_USE_PORT, 1);
314 	SET_FIELD(control, PXP_PRETEND_CMD_PRETEND_PORT, 1);
315 
316 	p_ptt->pxp.pretend.control = cpu_to_le16(control);
317 
318 	REG_WR(p_hwfn,
319 	       qed_ptt_config_addr(p_ptt) +
320 	       offsetof(struct pxp_ptt_entry, pretend),
321 	       *(u32 *)&p_ptt->pxp.pretend);
322 }
323 
324 void qed_port_unpretend(struct qed_hwfn *p_hwfn,
325 			struct qed_ptt *p_ptt)
326 {
327 	u16 control = 0;
328 
329 	SET_FIELD(control, PXP_PRETEND_CMD_PORT, 0);
330 	SET_FIELD(control, PXP_PRETEND_CMD_USE_PORT, 0);
331 	SET_FIELD(control, PXP_PRETEND_CMD_PRETEND_PORT, 1);
332 
333 	p_ptt->pxp.pretend.control = cpu_to_le16(control);
334 
335 	REG_WR(p_hwfn,
336 	       qed_ptt_config_addr(p_ptt) +
337 	       offsetof(struct pxp_ptt_entry, pretend),
338 	       *(u32 *)&p_ptt->pxp.pretend);
339 }
340 
341 u32 qed_vfid_to_concrete(struct qed_hwfn *p_hwfn, u8 vfid)
342 {
343 	u32 concrete_fid = 0;
344 
345 	SET_FIELD(concrete_fid, PXP_CONCRETE_FID_PFID, p_hwfn->rel_pf_id);
346 	SET_FIELD(concrete_fid, PXP_CONCRETE_FID_VFID, vfid);
347 	SET_FIELD(concrete_fid, PXP_CONCRETE_FID_VFVALID, 1);
348 
349 	return concrete_fid;
350 }
351 
352 /* DMAE */
353 static void qed_dmae_opcode(struct qed_hwfn *p_hwfn,
354 			    const u8 is_src_type_grc,
355 			    const u8 is_dst_type_grc,
356 			    struct qed_dmae_params *p_params)
357 {
358 	u16 opcode_b = 0;
359 	u32 opcode = 0;
360 
361 	/* Whether the source is the PCIe or the GRC.
362 	 * 0- The source is the PCIe
363 	 * 1- The source is the GRC.
364 	 */
365 	opcode |= (is_src_type_grc ? DMAE_CMD_SRC_MASK_GRC
366 				   : DMAE_CMD_SRC_MASK_PCIE) <<
367 		   DMAE_CMD_SRC_SHIFT;
368 	opcode |= ((p_hwfn->rel_pf_id & DMAE_CMD_SRC_PF_ID_MASK) <<
369 		   DMAE_CMD_SRC_PF_ID_SHIFT);
370 
371 	/* The destination of the DMA can be: 0-None 1-PCIe 2-GRC 3-None */
372 	opcode |= (is_dst_type_grc ? DMAE_CMD_DST_MASK_GRC
373 				   : DMAE_CMD_DST_MASK_PCIE) <<
374 		   DMAE_CMD_DST_SHIFT;
375 	opcode |= ((p_hwfn->rel_pf_id & DMAE_CMD_DST_PF_ID_MASK) <<
376 		   DMAE_CMD_DST_PF_ID_SHIFT);
377 
378 	/* Whether to write a completion word to the completion destination:
379 	 * 0-Do not write a completion word
380 	 * 1-Write the completion word
381 	 */
382 	opcode |= (DMAE_CMD_COMP_WORD_EN_MASK << DMAE_CMD_COMP_WORD_EN_SHIFT);
383 	opcode |= (DMAE_CMD_SRC_ADDR_RESET_MASK <<
384 		   DMAE_CMD_SRC_ADDR_RESET_SHIFT);
385 
386 	if (p_params->flags & QED_DMAE_FLAG_COMPLETION_DST)
387 		opcode |= (1 << DMAE_CMD_COMP_FUNC_SHIFT);
388 
389 	opcode |= (DMAE_CMD_ENDIANITY << DMAE_CMD_ENDIANITY_MODE_SHIFT);
390 
391 	opcode |= ((p_hwfn->port_id) << DMAE_CMD_PORT_ID_SHIFT);
392 
393 	/* reset source address in next go */
394 	opcode |= (DMAE_CMD_SRC_ADDR_RESET_MASK <<
395 		   DMAE_CMD_SRC_ADDR_RESET_SHIFT);
396 
397 	/* reset dest address in next go */
398 	opcode |= (DMAE_CMD_DST_ADDR_RESET_MASK <<
399 		   DMAE_CMD_DST_ADDR_RESET_SHIFT);
400 
401 	/* SRC/DST VFID: all 1's - pf, otherwise VF id */
402 	if (p_params->flags & QED_DMAE_FLAG_VF_SRC) {
403 		opcode |= 1 << DMAE_CMD_SRC_VF_ID_VALID_SHIFT;
404 		opcode_b |= p_params->src_vfid << DMAE_CMD_SRC_VF_ID_SHIFT;
405 	} else {
406 		opcode_b |= DMAE_CMD_SRC_VF_ID_MASK <<
407 			    DMAE_CMD_SRC_VF_ID_SHIFT;
408 	}
409 
410 	if (p_params->flags & QED_DMAE_FLAG_VF_DST) {
411 		opcode |= 1 << DMAE_CMD_DST_VF_ID_VALID_SHIFT;
412 		opcode_b |= p_params->dst_vfid << DMAE_CMD_DST_VF_ID_SHIFT;
413 	} else {
414 		opcode_b |= DMAE_CMD_DST_VF_ID_MASK << DMAE_CMD_DST_VF_ID_SHIFT;
415 	}
416 
417 	p_hwfn->dmae_info.p_dmae_cmd->opcode = cpu_to_le32(opcode);
418 	p_hwfn->dmae_info.p_dmae_cmd->opcode_b = cpu_to_le16(opcode_b);
419 }
420 
421 u32 qed_dmae_idx_to_go_cmd(u8 idx)
422 {
423 	/* All the DMAE 'go' registers form an array in internal memory */
424 	return DMAE_REG_GO_C0 + (idx << 2);
425 }
426 
427 static int
428 qed_dmae_post_command(struct qed_hwfn *p_hwfn,
429 		      struct qed_ptt *p_ptt)
430 {
431 	struct dmae_cmd *command = p_hwfn->dmae_info.p_dmae_cmd;
432 	u8 idx_cmd = p_hwfn->dmae_info.channel, i;
433 	int qed_status = 0;
434 
435 	/* verify address is not NULL */
436 	if ((((command->dst_addr_lo == 0) && (command->dst_addr_hi == 0)) ||
437 	     ((command->src_addr_lo == 0) && (command->src_addr_hi == 0)))) {
438 		DP_NOTICE(p_hwfn,
439 			  "source or destination address 0 idx_cmd=%d\n"
440 			  "opcode = [0x%08x,0x%04x] len=0x%x src=0x%x:%x dst=0x%x:%x\n",
441 			   idx_cmd,
442 			   le32_to_cpu(command->opcode),
443 			   le16_to_cpu(command->opcode_b),
444 			   le16_to_cpu(command->length),
445 			   le32_to_cpu(command->src_addr_hi),
446 			   le32_to_cpu(command->src_addr_lo),
447 			   le32_to_cpu(command->dst_addr_hi),
448 			   le32_to_cpu(command->dst_addr_lo));
449 
450 		return -EINVAL;
451 	}
452 
453 	DP_VERBOSE(p_hwfn,
454 		   NETIF_MSG_HW,
455 		   "Posting DMAE command [idx %d]: opcode = [0x%08x,0x%04x] len=0x%x src=0x%x:%x dst=0x%x:%x\n",
456 		   idx_cmd,
457 		   le32_to_cpu(command->opcode),
458 		   le16_to_cpu(command->opcode_b),
459 		   le16_to_cpu(command->length),
460 		   le32_to_cpu(command->src_addr_hi),
461 		   le32_to_cpu(command->src_addr_lo),
462 		   le32_to_cpu(command->dst_addr_hi),
463 		   le32_to_cpu(command->dst_addr_lo));
464 
465 	/* Copy the command to DMAE - need to do it before every call
466 	 * for source/dest address no reset.
467 	 * The first 9 DWs are the command registers, the 10 DW is the
468 	 * GO register, and the rest are result registers
469 	 * (which are read only by the client).
470 	 */
471 	for (i = 0; i < DMAE_CMD_SIZE; i++) {
472 		u32 data = (i < DMAE_CMD_SIZE_TO_FILL) ?
473 			   *(((u32 *)command) + i) : 0;
474 
475 		qed_wr(p_hwfn, p_ptt,
476 		       DMAE_REG_CMD_MEM +
477 		       (idx_cmd * DMAE_CMD_SIZE * sizeof(u32)) +
478 		       (i * sizeof(u32)), data);
479 	}
480 
481 	qed_wr(p_hwfn, p_ptt,
482 	       qed_dmae_idx_to_go_cmd(idx_cmd),
483 	       DMAE_GO_VALUE);
484 
485 	return qed_status;
486 }
487 
488 int qed_dmae_info_alloc(struct qed_hwfn *p_hwfn)
489 {
490 	dma_addr_t *p_addr = &p_hwfn->dmae_info.completion_word_phys_addr;
491 	struct dmae_cmd **p_cmd = &p_hwfn->dmae_info.p_dmae_cmd;
492 	u32 **p_buff = &p_hwfn->dmae_info.p_intermediate_buffer;
493 	u32 **p_comp = &p_hwfn->dmae_info.p_completion_word;
494 
495 	*p_comp = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
496 				     sizeof(u32),
497 				     p_addr,
498 				     GFP_KERNEL);
499 	if (!*p_comp) {
500 		DP_NOTICE(p_hwfn, "Failed to allocate `p_completion_word'\n");
501 		goto err;
502 	}
503 
504 	p_addr = &p_hwfn->dmae_info.dmae_cmd_phys_addr;
505 	*p_cmd = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
506 				    sizeof(struct dmae_cmd),
507 				    p_addr, GFP_KERNEL);
508 	if (!*p_cmd) {
509 		DP_NOTICE(p_hwfn, "Failed to allocate `struct dmae_cmd'\n");
510 		goto err;
511 	}
512 
513 	p_addr = &p_hwfn->dmae_info.intermediate_buffer_phys_addr;
514 	*p_buff = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
515 				     sizeof(u32) * DMAE_MAX_RW_SIZE,
516 				     p_addr, GFP_KERNEL);
517 	if (!*p_buff) {
518 		DP_NOTICE(p_hwfn, "Failed to allocate `intermediate_buffer'\n");
519 		goto err;
520 	}
521 
522 	p_hwfn->dmae_info.channel = p_hwfn->rel_pf_id;
523 
524 	return 0;
525 err:
526 	qed_dmae_info_free(p_hwfn);
527 	return -ENOMEM;
528 }
529 
530 void qed_dmae_info_free(struct qed_hwfn *p_hwfn)
531 {
532 	dma_addr_t p_phys;
533 
534 	/* Just make sure no one is in the middle */
535 	mutex_lock(&p_hwfn->dmae_info.mutex);
536 
537 	if (p_hwfn->dmae_info.p_completion_word) {
538 		p_phys = p_hwfn->dmae_info.completion_word_phys_addr;
539 		dma_free_coherent(&p_hwfn->cdev->pdev->dev,
540 				  sizeof(u32),
541 				  p_hwfn->dmae_info.p_completion_word,
542 				  p_phys);
543 		p_hwfn->dmae_info.p_completion_word = NULL;
544 	}
545 
546 	if (p_hwfn->dmae_info.p_dmae_cmd) {
547 		p_phys = p_hwfn->dmae_info.dmae_cmd_phys_addr;
548 		dma_free_coherent(&p_hwfn->cdev->pdev->dev,
549 				  sizeof(struct dmae_cmd),
550 				  p_hwfn->dmae_info.p_dmae_cmd,
551 				  p_phys);
552 		p_hwfn->dmae_info.p_dmae_cmd = NULL;
553 	}
554 
555 	if (p_hwfn->dmae_info.p_intermediate_buffer) {
556 		p_phys = p_hwfn->dmae_info.intermediate_buffer_phys_addr;
557 		dma_free_coherent(&p_hwfn->cdev->pdev->dev,
558 				  sizeof(u32) * DMAE_MAX_RW_SIZE,
559 				  p_hwfn->dmae_info.p_intermediate_buffer,
560 				  p_phys);
561 		p_hwfn->dmae_info.p_intermediate_buffer = NULL;
562 	}
563 
564 	mutex_unlock(&p_hwfn->dmae_info.mutex);
565 }
566 
567 static int qed_dmae_operation_wait(struct qed_hwfn *p_hwfn)
568 {
569 	u32 wait_cnt = 0;
570 	u32 wait_cnt_limit = 10000;
571 
572 	int qed_status = 0;
573 
574 	barrier();
575 	while (*p_hwfn->dmae_info.p_completion_word != DMAE_COMPLETION_VAL) {
576 		udelay(DMAE_MIN_WAIT_TIME);
577 		if (++wait_cnt > wait_cnt_limit) {
578 			DP_NOTICE(p_hwfn->cdev,
579 				  "Timed-out waiting for operation to complete. Completion word is 0x%08x expected 0x%08x.\n",
580 				  *p_hwfn->dmae_info.p_completion_word,
581 				 DMAE_COMPLETION_VAL);
582 			qed_status = -EBUSY;
583 			break;
584 		}
585 
586 		/* to sync the completion_word since we are not
587 		 * using the volatile keyword for p_completion_word
588 		 */
589 		barrier();
590 	}
591 
592 	if (qed_status == 0)
593 		*p_hwfn->dmae_info.p_completion_word = 0;
594 
595 	return qed_status;
596 }
597 
598 static int qed_dmae_execute_sub_operation(struct qed_hwfn *p_hwfn,
599 					  struct qed_ptt *p_ptt,
600 					  u64 src_addr,
601 					  u64 dst_addr,
602 					  u8 src_type,
603 					  u8 dst_type,
604 					  u32 length)
605 {
606 	dma_addr_t phys = p_hwfn->dmae_info.intermediate_buffer_phys_addr;
607 	struct dmae_cmd *cmd = p_hwfn->dmae_info.p_dmae_cmd;
608 	int qed_status = 0;
609 
610 	switch (src_type) {
611 	case QED_DMAE_ADDRESS_GRC:
612 	case QED_DMAE_ADDRESS_HOST_PHYS:
613 		cmd->src_addr_hi = cpu_to_le32(upper_32_bits(src_addr));
614 		cmd->src_addr_lo = cpu_to_le32(lower_32_bits(src_addr));
615 		break;
616 	/* for virtual source addresses we use the intermediate buffer. */
617 	case QED_DMAE_ADDRESS_HOST_VIRT:
618 		cmd->src_addr_hi = cpu_to_le32(upper_32_bits(phys));
619 		cmd->src_addr_lo = cpu_to_le32(lower_32_bits(phys));
620 		memcpy(&p_hwfn->dmae_info.p_intermediate_buffer[0],
621 		       (void *)(uintptr_t)src_addr,
622 		       length * sizeof(u32));
623 		break;
624 	default:
625 		return -EINVAL;
626 	}
627 
628 	switch (dst_type) {
629 	case QED_DMAE_ADDRESS_GRC:
630 	case QED_DMAE_ADDRESS_HOST_PHYS:
631 		cmd->dst_addr_hi = cpu_to_le32(upper_32_bits(dst_addr));
632 		cmd->dst_addr_lo = cpu_to_le32(lower_32_bits(dst_addr));
633 		break;
634 	/* for virtual source addresses we use the intermediate buffer. */
635 	case QED_DMAE_ADDRESS_HOST_VIRT:
636 		cmd->dst_addr_hi = cpu_to_le32(upper_32_bits(phys));
637 		cmd->dst_addr_lo = cpu_to_le32(lower_32_bits(phys));
638 		break;
639 	default:
640 		return -EINVAL;
641 	}
642 
643 	cmd->length = cpu_to_le16((u16)length);
644 
645 	qed_dmae_post_command(p_hwfn, p_ptt);
646 
647 	qed_status = qed_dmae_operation_wait(p_hwfn);
648 
649 	if (qed_status) {
650 		DP_NOTICE(p_hwfn,
651 			  "qed_dmae_host2grc: Wait Failed. source_addr 0x%llx, grc_addr 0x%llx, size_in_dwords 0x%x\n",
652 			  src_addr,
653 			  dst_addr,
654 			  length);
655 		return qed_status;
656 	}
657 
658 	if (dst_type == QED_DMAE_ADDRESS_HOST_VIRT)
659 		memcpy((void *)(uintptr_t)(dst_addr),
660 		       &p_hwfn->dmae_info.p_intermediate_buffer[0],
661 		       length * sizeof(u32));
662 
663 	return 0;
664 }
665 
666 static int qed_dmae_execute_command(struct qed_hwfn *p_hwfn,
667 				    struct qed_ptt *p_ptt,
668 				    u64 src_addr, u64 dst_addr,
669 				    u8 src_type, u8 dst_type,
670 				    u32 size_in_dwords,
671 				    struct qed_dmae_params *p_params)
672 {
673 	dma_addr_t phys = p_hwfn->dmae_info.completion_word_phys_addr;
674 	u16 length_cur = 0, i = 0, cnt_split = 0, length_mod = 0;
675 	struct dmae_cmd *cmd = p_hwfn->dmae_info.p_dmae_cmd;
676 	u64 src_addr_split = 0, dst_addr_split = 0;
677 	u16 length_limit = DMAE_MAX_RW_SIZE;
678 	int qed_status = 0;
679 	u32 offset = 0;
680 
681 	qed_dmae_opcode(p_hwfn,
682 			(src_type == QED_DMAE_ADDRESS_GRC),
683 			(dst_type == QED_DMAE_ADDRESS_GRC),
684 			p_params);
685 
686 	cmd->comp_addr_lo = cpu_to_le32(lower_32_bits(phys));
687 	cmd->comp_addr_hi = cpu_to_le32(upper_32_bits(phys));
688 	cmd->comp_val = cpu_to_le32(DMAE_COMPLETION_VAL);
689 
690 	/* Check if the grc_addr is valid like < MAX_GRC_OFFSET */
691 	cnt_split = size_in_dwords / length_limit;
692 	length_mod = size_in_dwords % length_limit;
693 
694 	src_addr_split = src_addr;
695 	dst_addr_split = dst_addr;
696 
697 	for (i = 0; i <= cnt_split; i++) {
698 		offset = length_limit * i;
699 
700 		if (!(p_params->flags & QED_DMAE_FLAG_RW_REPL_SRC)) {
701 			if (src_type == QED_DMAE_ADDRESS_GRC)
702 				src_addr_split = src_addr + offset;
703 			else
704 				src_addr_split = src_addr + (offset * 4);
705 		}
706 
707 		if (dst_type == QED_DMAE_ADDRESS_GRC)
708 			dst_addr_split = dst_addr + offset;
709 		else
710 			dst_addr_split = dst_addr + (offset * 4);
711 
712 		length_cur = (cnt_split == i) ? length_mod : length_limit;
713 
714 		/* might be zero on last iteration */
715 		if (!length_cur)
716 			continue;
717 
718 		qed_status = qed_dmae_execute_sub_operation(p_hwfn,
719 							    p_ptt,
720 							    src_addr_split,
721 							    dst_addr_split,
722 							    src_type,
723 							    dst_type,
724 							    length_cur);
725 		if (qed_status) {
726 			DP_NOTICE(p_hwfn,
727 				  "qed_dmae_execute_sub_operation Failed with error 0x%x. source_addr 0x%llx, destination addr 0x%llx, size_in_dwords 0x%x\n",
728 				  qed_status,
729 				  src_addr,
730 				  dst_addr,
731 				  length_cur);
732 			break;
733 		}
734 	}
735 
736 	return qed_status;
737 }
738 
739 int qed_dmae_host2grc(struct qed_hwfn *p_hwfn,
740 		      struct qed_ptt *p_ptt,
741 		      u64 source_addr,
742 		      u32 grc_addr,
743 		      u32 size_in_dwords,
744 		      u32 flags)
745 {
746 	u32 grc_addr_in_dw = grc_addr / sizeof(u32);
747 	struct qed_dmae_params params;
748 	int rc;
749 
750 	memset(&params, 0, sizeof(struct qed_dmae_params));
751 	params.flags = flags;
752 
753 	mutex_lock(&p_hwfn->dmae_info.mutex);
754 
755 	rc = qed_dmae_execute_command(p_hwfn, p_ptt, source_addr,
756 				      grc_addr_in_dw,
757 				      QED_DMAE_ADDRESS_HOST_VIRT,
758 				      QED_DMAE_ADDRESS_GRC,
759 				      size_in_dwords, &params);
760 
761 	mutex_unlock(&p_hwfn->dmae_info.mutex);
762 
763 	return rc;
764 }
765 
766 int
767 qed_dmae_host2host(struct qed_hwfn *p_hwfn,
768 		   struct qed_ptt *p_ptt,
769 		   dma_addr_t source_addr,
770 		   dma_addr_t dest_addr,
771 		   u32 size_in_dwords, struct qed_dmae_params *p_params)
772 {
773 	int rc;
774 
775 	mutex_lock(&(p_hwfn->dmae_info.mutex));
776 
777 	rc = qed_dmae_execute_command(p_hwfn, p_ptt, source_addr,
778 				      dest_addr,
779 				      QED_DMAE_ADDRESS_HOST_PHYS,
780 				      QED_DMAE_ADDRESS_HOST_PHYS,
781 				      size_in_dwords, p_params);
782 
783 	mutex_unlock(&(p_hwfn->dmae_info.mutex));
784 
785 	return rc;
786 }
787 
788 u16 qed_get_qm_pq(struct qed_hwfn *p_hwfn,
789 		  enum protocol_type proto,
790 		  union qed_qm_pq_params *p_params)
791 {
792 	u16 pq_id = 0;
793 
794 	if ((proto == PROTOCOLID_CORE || proto == PROTOCOLID_ETH) &&
795 	    !p_params) {
796 		DP_NOTICE(p_hwfn,
797 			  "Protocol %d received NULL PQ params\n",
798 			  proto);
799 		return 0;
800 	}
801 
802 	switch (proto) {
803 	case PROTOCOLID_CORE:
804 		if (p_params->core.tc == LB_TC)
805 			pq_id = p_hwfn->qm_info.pure_lb_pq;
806 		else
807 			pq_id = p_hwfn->qm_info.offload_pq;
808 		break;
809 	case PROTOCOLID_ETH:
810 		pq_id = p_params->eth.tc;
811 		break;
812 	default:
813 		pq_id = 0;
814 	}
815 
816 	pq_id = CM_TX_PQ_BASE + pq_id + RESC_START(p_hwfn, QED_PQ);
817 
818 	return pq_id;
819 }
820