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 #include "qed_sriov.h"
27 
28 #define QED_BAR_ACQUIRE_TIMEOUT 1000
29 
30 /* Invalid values */
31 #define QED_BAR_INVALID_OFFSET          (cpu_to_le32(-1))
32 
33 struct qed_ptt {
34 	struct list_head	list_entry;
35 	unsigned int		idx;
36 	struct pxp_ptt_entry	pxp;
37 };
38 
39 struct qed_ptt_pool {
40 	struct list_head	free_list;
41 	spinlock_t		lock; /* ptt synchronized access */
42 	struct qed_ptt		ptts[PXP_EXTERNAL_BAR_PF_WINDOW_NUM];
43 };
44 
45 int qed_ptt_pool_alloc(struct qed_hwfn *p_hwfn)
46 {
47 	struct qed_ptt_pool *p_pool = kmalloc(sizeof(*p_pool), 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, 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 	/* Verify the address is within the window */
173 	if (hw_addr < win_hw_addr ||
174 	    offset >= PXP_EXTERNAL_BAR_PF_WINDOW_SINGLE_SIZE) {
175 		qed_ptt_set_win(p_hwfn, p_ptt, hw_addr);
176 		offset = 0;
177 	}
178 
179 	return qed_ptt_get_bar_addr(p_ptt) + offset;
180 }
181 
182 struct qed_ptt *qed_get_reserved_ptt(struct qed_hwfn *p_hwfn,
183 				     enum reserved_ptts ptt_idx)
184 {
185 	if (ptt_idx >= RESERVED_PTT_MAX) {
186 		DP_NOTICE(p_hwfn,
187 			  "Requested PTT %d is out of range\n", ptt_idx);
188 		return NULL;
189 	}
190 
191 	return &p_hwfn->p_ptt_pool->ptts[ptt_idx];
192 }
193 
194 void qed_wr(struct qed_hwfn *p_hwfn,
195 	    struct qed_ptt *p_ptt,
196 	    u32 hw_addr, u32 val)
197 {
198 	u32 bar_addr = qed_set_ptt(p_hwfn, p_ptt, hw_addr);
199 
200 	REG_WR(p_hwfn, bar_addr, val);
201 	DP_VERBOSE(p_hwfn, NETIF_MSG_HW,
202 		   "bar_addr 0x%x, hw_addr 0x%x, val 0x%x\n",
203 		   bar_addr, hw_addr, val);
204 }
205 
206 u32 qed_rd(struct qed_hwfn *p_hwfn,
207 	   struct qed_ptt *p_ptt,
208 	   u32 hw_addr)
209 {
210 	u32 bar_addr = qed_set_ptt(p_hwfn, p_ptt, hw_addr);
211 	u32 val = REG_RD(p_hwfn, bar_addr);
212 
213 	DP_VERBOSE(p_hwfn, NETIF_MSG_HW,
214 		   "bar_addr 0x%x, hw_addr 0x%x, val 0x%x\n",
215 		   bar_addr, hw_addr, val);
216 
217 	return val;
218 }
219 
220 static void qed_memcpy_hw(struct qed_hwfn *p_hwfn,
221 			  struct qed_ptt *p_ptt,
222 			  void *addr, u32 hw_addr, size_t n, bool to_device)
223 {
224 	u32 dw_count, *host_addr, hw_offset;
225 	size_t quota, done = 0;
226 	u32 __iomem *reg_addr;
227 
228 	while (done < n) {
229 		quota = min_t(size_t, n - done,
230 			      PXP_EXTERNAL_BAR_PF_WINDOW_SINGLE_SIZE);
231 
232 		if (IS_PF(p_hwfn->cdev)) {
233 			qed_ptt_set_win(p_hwfn, p_ptt, hw_addr + done);
234 			hw_offset = qed_ptt_get_bar_addr(p_ptt);
235 		} else {
236 			hw_offset = hw_addr + done;
237 		}
238 
239 		dw_count = quota / 4;
240 		host_addr = (u32 *)((u8 *)addr + done);
241 		reg_addr = (u32 __iomem *)REG_ADDR(p_hwfn, hw_offset);
242 		if (to_device)
243 			while (dw_count--)
244 				DIRECT_REG_WR(reg_addr++, *host_addr++);
245 		else
246 			while (dw_count--)
247 				*host_addr++ = DIRECT_REG_RD(reg_addr++);
248 
249 		done += quota;
250 	}
251 }
252 
253 void qed_memcpy_from(struct qed_hwfn *p_hwfn,
254 		     struct qed_ptt *p_ptt, void *dest, u32 hw_addr, size_t n)
255 {
256 	DP_VERBOSE(p_hwfn, NETIF_MSG_HW,
257 		   "hw_addr 0x%x, dest %p hw_addr 0x%x, size %lu\n",
258 		   hw_addr, dest, hw_addr, (unsigned long)n);
259 
260 	qed_memcpy_hw(p_hwfn, p_ptt, dest, hw_addr, n, false);
261 }
262 
263 void qed_memcpy_to(struct qed_hwfn *p_hwfn,
264 		   struct qed_ptt *p_ptt, u32 hw_addr, void *src, size_t n)
265 {
266 	DP_VERBOSE(p_hwfn, NETIF_MSG_HW,
267 		   "hw_addr 0x%x, hw_addr 0x%x, src %p size %lu\n",
268 		   hw_addr, hw_addr, src, (unsigned long)n);
269 
270 	qed_memcpy_hw(p_hwfn, p_ptt, src, hw_addr, n, true);
271 }
272 
273 void qed_fid_pretend(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt, u16 fid)
274 {
275 	u16 control = 0;
276 
277 	SET_FIELD(control, PXP_PRETEND_CMD_IS_CONCRETE, 1);
278 	SET_FIELD(control, PXP_PRETEND_CMD_PRETEND_FUNCTION, 1);
279 
280 	/* Every pretend undos previous pretends, including
281 	 * previous port pretend.
282 	 */
283 	SET_FIELD(control, PXP_PRETEND_CMD_PORT, 0);
284 	SET_FIELD(control, PXP_PRETEND_CMD_USE_PORT, 0);
285 	SET_FIELD(control, PXP_PRETEND_CMD_PRETEND_PORT, 1);
286 
287 	if (!GET_FIELD(fid, PXP_CONCRETE_FID_VFVALID))
288 		fid = GET_FIELD(fid, PXP_CONCRETE_FID_PFID);
289 
290 	p_ptt->pxp.pretend.control = cpu_to_le16(control);
291 	p_ptt->pxp.pretend.fid.concrete_fid.fid = cpu_to_le16(fid);
292 
293 	REG_WR(p_hwfn,
294 	       qed_ptt_config_addr(p_ptt) +
295 	       offsetof(struct pxp_ptt_entry, pretend),
296 	       *(u32 *)&p_ptt->pxp.pretend);
297 }
298 
299 void qed_port_pretend(struct qed_hwfn *p_hwfn,
300 		      struct qed_ptt *p_ptt, u8 port_id)
301 {
302 	u16 control = 0;
303 
304 	SET_FIELD(control, PXP_PRETEND_CMD_PORT, port_id);
305 	SET_FIELD(control, PXP_PRETEND_CMD_USE_PORT, 1);
306 	SET_FIELD(control, PXP_PRETEND_CMD_PRETEND_PORT, 1);
307 
308 	p_ptt->pxp.pretend.control = cpu_to_le16(control);
309 
310 	REG_WR(p_hwfn,
311 	       qed_ptt_config_addr(p_ptt) +
312 	       offsetof(struct pxp_ptt_entry, pretend),
313 	       *(u32 *)&p_ptt->pxp.pretend);
314 }
315 
316 void qed_port_unpretend(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
317 {
318 	u16 control = 0;
319 
320 	SET_FIELD(control, PXP_PRETEND_CMD_PORT, 0);
321 	SET_FIELD(control, PXP_PRETEND_CMD_USE_PORT, 0);
322 	SET_FIELD(control, PXP_PRETEND_CMD_PRETEND_PORT, 1);
323 
324 	p_ptt->pxp.pretend.control = cpu_to_le16(control);
325 
326 	REG_WR(p_hwfn,
327 	       qed_ptt_config_addr(p_ptt) +
328 	       offsetof(struct pxp_ptt_entry, pretend),
329 	       *(u32 *)&p_ptt->pxp.pretend);
330 }
331 
332 u32 qed_vfid_to_concrete(struct qed_hwfn *p_hwfn, u8 vfid)
333 {
334 	u32 concrete_fid = 0;
335 
336 	SET_FIELD(concrete_fid, PXP_CONCRETE_FID_PFID, p_hwfn->rel_pf_id);
337 	SET_FIELD(concrete_fid, PXP_CONCRETE_FID_VFID, vfid);
338 	SET_FIELD(concrete_fid, PXP_CONCRETE_FID_VFVALID, 1);
339 
340 	return concrete_fid;
341 }
342 
343 /* DMAE */
344 static void qed_dmae_opcode(struct qed_hwfn *p_hwfn,
345 			    const u8 is_src_type_grc,
346 			    const u8 is_dst_type_grc,
347 			    struct qed_dmae_params *p_params)
348 {
349 	u16 opcode_b = 0;
350 	u32 opcode = 0;
351 
352 	/* Whether the source is the PCIe or the GRC.
353 	 * 0- The source is the PCIe
354 	 * 1- The source is the GRC.
355 	 */
356 	opcode |= (is_src_type_grc ? DMAE_CMD_SRC_MASK_GRC
357 				   : DMAE_CMD_SRC_MASK_PCIE) <<
358 		   DMAE_CMD_SRC_SHIFT;
359 	opcode |= ((p_hwfn->rel_pf_id & DMAE_CMD_SRC_PF_ID_MASK) <<
360 		   DMAE_CMD_SRC_PF_ID_SHIFT);
361 
362 	/* The destination of the DMA can be: 0-None 1-PCIe 2-GRC 3-None */
363 	opcode |= (is_dst_type_grc ? DMAE_CMD_DST_MASK_GRC
364 				   : DMAE_CMD_DST_MASK_PCIE) <<
365 		   DMAE_CMD_DST_SHIFT;
366 	opcode |= ((p_hwfn->rel_pf_id & DMAE_CMD_DST_PF_ID_MASK) <<
367 		   DMAE_CMD_DST_PF_ID_SHIFT);
368 
369 	/* Whether to write a completion word to the completion destination:
370 	 * 0-Do not write a completion word
371 	 * 1-Write the completion word
372 	 */
373 	opcode |= (DMAE_CMD_COMP_WORD_EN_MASK << DMAE_CMD_COMP_WORD_EN_SHIFT);
374 	opcode |= (DMAE_CMD_SRC_ADDR_RESET_MASK <<
375 		   DMAE_CMD_SRC_ADDR_RESET_SHIFT);
376 
377 	if (p_params->flags & QED_DMAE_FLAG_COMPLETION_DST)
378 		opcode |= (1 << DMAE_CMD_COMP_FUNC_SHIFT);
379 
380 	opcode |= (DMAE_CMD_ENDIANITY << DMAE_CMD_ENDIANITY_MODE_SHIFT);
381 
382 	opcode |= ((p_hwfn->port_id) << DMAE_CMD_PORT_ID_SHIFT);
383 
384 	/* reset source address in next go */
385 	opcode |= (DMAE_CMD_SRC_ADDR_RESET_MASK <<
386 		   DMAE_CMD_SRC_ADDR_RESET_SHIFT);
387 
388 	/* reset dest address in next go */
389 	opcode |= (DMAE_CMD_DST_ADDR_RESET_MASK <<
390 		   DMAE_CMD_DST_ADDR_RESET_SHIFT);
391 
392 	/* SRC/DST VFID: all 1's - pf, otherwise VF id */
393 	if (p_params->flags & QED_DMAE_FLAG_VF_SRC) {
394 		opcode |= 1 << DMAE_CMD_SRC_VF_ID_VALID_SHIFT;
395 		opcode_b |= p_params->src_vfid << DMAE_CMD_SRC_VF_ID_SHIFT;
396 	} else {
397 		opcode_b |= DMAE_CMD_SRC_VF_ID_MASK <<
398 			    DMAE_CMD_SRC_VF_ID_SHIFT;
399 	}
400 
401 	if (p_params->flags & QED_DMAE_FLAG_VF_DST) {
402 		opcode |= 1 << DMAE_CMD_DST_VF_ID_VALID_SHIFT;
403 		opcode_b |= p_params->dst_vfid << DMAE_CMD_DST_VF_ID_SHIFT;
404 	} else {
405 		opcode_b |= DMAE_CMD_DST_VF_ID_MASK << DMAE_CMD_DST_VF_ID_SHIFT;
406 	}
407 
408 	p_hwfn->dmae_info.p_dmae_cmd->opcode = cpu_to_le32(opcode);
409 	p_hwfn->dmae_info.p_dmae_cmd->opcode_b = cpu_to_le16(opcode_b);
410 }
411 
412 u32 qed_dmae_idx_to_go_cmd(u8 idx)
413 {
414 	/* All the DMAE 'go' registers form an array in internal memory */
415 	return DMAE_REG_GO_C0 + (idx << 2);
416 }
417 
418 static int qed_dmae_post_command(struct qed_hwfn *p_hwfn,
419 				 struct qed_ptt *p_ptt)
420 {
421 	struct dmae_cmd *p_command = p_hwfn->dmae_info.p_dmae_cmd;
422 	u8 idx_cmd = p_hwfn->dmae_info.channel, i;
423 	int qed_status = 0;
424 
425 	/* verify address is not NULL */
426 	if ((((!p_command->dst_addr_lo) && (!p_command->dst_addr_hi)) ||
427 	     ((!p_command->src_addr_lo) && (!p_command->src_addr_hi)))) {
428 		DP_NOTICE(p_hwfn,
429 			  "source or destination address 0 idx_cmd=%d\n"
430 			  "opcode = [0x%08x,0x%04x] len=0x%x src=0x%x:%x dst=0x%x:%x\n",
431 			  idx_cmd,
432 			  le32_to_cpu(p_command->opcode),
433 			  le16_to_cpu(p_command->opcode_b),
434 			  le16_to_cpu(p_command->length_dw),
435 			  le32_to_cpu(p_command->src_addr_hi),
436 			  le32_to_cpu(p_command->src_addr_lo),
437 			  le32_to_cpu(p_command->dst_addr_hi),
438 			  le32_to_cpu(p_command->dst_addr_lo));
439 
440 		return -EINVAL;
441 	}
442 
443 	DP_VERBOSE(p_hwfn,
444 		   NETIF_MSG_HW,
445 		   "Posting DMAE command [idx %d]: opcode = [0x%08x,0x%04x] len=0x%x src=0x%x:%x dst=0x%x:%x\n",
446 		   idx_cmd,
447 		   le32_to_cpu(p_command->opcode),
448 		   le16_to_cpu(p_command->opcode_b),
449 		   le16_to_cpu(p_command->length_dw),
450 		   le32_to_cpu(p_command->src_addr_hi),
451 		   le32_to_cpu(p_command->src_addr_lo),
452 		   le32_to_cpu(p_command->dst_addr_hi),
453 		   le32_to_cpu(p_command->dst_addr_lo));
454 
455 	/* Copy the command to DMAE - need to do it before every call
456 	 * for source/dest address no reset.
457 	 * The first 9 DWs are the command registers, the 10 DW is the
458 	 * GO register, and the rest are result registers
459 	 * (which are read only by the client).
460 	 */
461 	for (i = 0; i < DMAE_CMD_SIZE; i++) {
462 		u32 data = (i < DMAE_CMD_SIZE_TO_FILL) ?
463 			   *(((u32 *)p_command) + i) : 0;
464 
465 		qed_wr(p_hwfn, p_ptt,
466 		       DMAE_REG_CMD_MEM +
467 		       (idx_cmd * DMAE_CMD_SIZE * sizeof(u32)) +
468 		       (i * sizeof(u32)), data);
469 	}
470 
471 	qed_wr(p_hwfn, p_ptt, qed_dmae_idx_to_go_cmd(idx_cmd), DMAE_GO_VALUE);
472 
473 	return qed_status;
474 }
475 
476 int qed_dmae_info_alloc(struct qed_hwfn *p_hwfn)
477 {
478 	dma_addr_t *p_addr = &p_hwfn->dmae_info.completion_word_phys_addr;
479 	struct dmae_cmd **p_cmd = &p_hwfn->dmae_info.p_dmae_cmd;
480 	u32 **p_buff = &p_hwfn->dmae_info.p_intermediate_buffer;
481 	u32 **p_comp = &p_hwfn->dmae_info.p_completion_word;
482 
483 	*p_comp = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
484 				     sizeof(u32), p_addr, GFP_KERNEL);
485 	if (!*p_comp)
486 		goto err;
487 
488 	p_addr = &p_hwfn->dmae_info.dmae_cmd_phys_addr;
489 	*p_cmd = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
490 				    sizeof(struct dmae_cmd),
491 				    p_addr, GFP_KERNEL);
492 	if (!*p_cmd)
493 		goto err;
494 
495 	p_addr = &p_hwfn->dmae_info.intermediate_buffer_phys_addr;
496 	*p_buff = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
497 				     sizeof(u32) * DMAE_MAX_RW_SIZE,
498 				     p_addr, GFP_KERNEL);
499 	if (!*p_buff)
500 		goto err;
501 
502 	p_hwfn->dmae_info.channel = p_hwfn->rel_pf_id;
503 
504 	return 0;
505 err:
506 	qed_dmae_info_free(p_hwfn);
507 	return -ENOMEM;
508 }
509 
510 void qed_dmae_info_free(struct qed_hwfn *p_hwfn)
511 {
512 	dma_addr_t p_phys;
513 
514 	/* Just make sure no one is in the middle */
515 	mutex_lock(&p_hwfn->dmae_info.mutex);
516 
517 	if (p_hwfn->dmae_info.p_completion_word) {
518 		p_phys = p_hwfn->dmae_info.completion_word_phys_addr;
519 		dma_free_coherent(&p_hwfn->cdev->pdev->dev,
520 				  sizeof(u32),
521 				  p_hwfn->dmae_info.p_completion_word, p_phys);
522 		p_hwfn->dmae_info.p_completion_word = NULL;
523 	}
524 
525 	if (p_hwfn->dmae_info.p_dmae_cmd) {
526 		p_phys = p_hwfn->dmae_info.dmae_cmd_phys_addr;
527 		dma_free_coherent(&p_hwfn->cdev->pdev->dev,
528 				  sizeof(struct dmae_cmd),
529 				  p_hwfn->dmae_info.p_dmae_cmd, p_phys);
530 		p_hwfn->dmae_info.p_dmae_cmd = NULL;
531 	}
532 
533 	if (p_hwfn->dmae_info.p_intermediate_buffer) {
534 		p_phys = p_hwfn->dmae_info.intermediate_buffer_phys_addr;
535 		dma_free_coherent(&p_hwfn->cdev->pdev->dev,
536 				  sizeof(u32) * DMAE_MAX_RW_SIZE,
537 				  p_hwfn->dmae_info.p_intermediate_buffer,
538 				  p_phys);
539 		p_hwfn->dmae_info.p_intermediate_buffer = NULL;
540 	}
541 
542 	mutex_unlock(&p_hwfn->dmae_info.mutex);
543 }
544 
545 static int qed_dmae_operation_wait(struct qed_hwfn *p_hwfn)
546 {
547 	u32 wait_cnt_limit = 10000, wait_cnt = 0;
548 	int qed_status = 0;
549 
550 	barrier();
551 	while (*p_hwfn->dmae_info.p_completion_word != DMAE_COMPLETION_VAL) {
552 		udelay(DMAE_MIN_WAIT_TIME);
553 		if (++wait_cnt > wait_cnt_limit) {
554 			DP_NOTICE(p_hwfn->cdev,
555 				  "Timed-out waiting for operation to complete. Completion word is 0x%08x expected 0x%08x.\n",
556 				  *p_hwfn->dmae_info.p_completion_word,
557 				 DMAE_COMPLETION_VAL);
558 			qed_status = -EBUSY;
559 			break;
560 		}
561 
562 		/* to sync the completion_word since we are not
563 		 * using the volatile keyword for p_completion_word
564 		 */
565 		barrier();
566 	}
567 
568 	if (qed_status == 0)
569 		*p_hwfn->dmae_info.p_completion_word = 0;
570 
571 	return qed_status;
572 }
573 
574 static int qed_dmae_execute_sub_operation(struct qed_hwfn *p_hwfn,
575 					  struct qed_ptt *p_ptt,
576 					  u64 src_addr,
577 					  u64 dst_addr,
578 					  u8 src_type,
579 					  u8 dst_type,
580 					  u32 length_dw)
581 {
582 	dma_addr_t phys = p_hwfn->dmae_info.intermediate_buffer_phys_addr;
583 	struct dmae_cmd *cmd = p_hwfn->dmae_info.p_dmae_cmd;
584 	int qed_status = 0;
585 
586 	switch (src_type) {
587 	case QED_DMAE_ADDRESS_GRC:
588 	case QED_DMAE_ADDRESS_HOST_PHYS:
589 		cmd->src_addr_hi = cpu_to_le32(upper_32_bits(src_addr));
590 		cmd->src_addr_lo = cpu_to_le32(lower_32_bits(src_addr));
591 		break;
592 	/* for virtual source addresses we use the intermediate buffer. */
593 	case QED_DMAE_ADDRESS_HOST_VIRT:
594 		cmd->src_addr_hi = cpu_to_le32(upper_32_bits(phys));
595 		cmd->src_addr_lo = cpu_to_le32(lower_32_bits(phys));
596 		memcpy(&p_hwfn->dmae_info.p_intermediate_buffer[0],
597 		       (void *)(uintptr_t)src_addr,
598 		       length_dw * sizeof(u32));
599 		break;
600 	default:
601 		return -EINVAL;
602 	}
603 
604 	switch (dst_type) {
605 	case QED_DMAE_ADDRESS_GRC:
606 	case QED_DMAE_ADDRESS_HOST_PHYS:
607 		cmd->dst_addr_hi = cpu_to_le32(upper_32_bits(dst_addr));
608 		cmd->dst_addr_lo = cpu_to_le32(lower_32_bits(dst_addr));
609 		break;
610 	/* for virtual source addresses we use the intermediate buffer. */
611 	case QED_DMAE_ADDRESS_HOST_VIRT:
612 		cmd->dst_addr_hi = cpu_to_le32(upper_32_bits(phys));
613 		cmd->dst_addr_lo = cpu_to_le32(lower_32_bits(phys));
614 		break;
615 	default:
616 		return -EINVAL;
617 	}
618 
619 	cmd->length_dw = cpu_to_le16((u16)length_dw);
620 
621 	qed_dmae_post_command(p_hwfn, p_ptt);
622 
623 	qed_status = qed_dmae_operation_wait(p_hwfn);
624 
625 	if (qed_status) {
626 		DP_NOTICE(p_hwfn,
627 			  "qed_dmae_host2grc: Wait Failed. source_addr 0x%llx, grc_addr 0x%llx, size_in_dwords 0x%x\n",
628 			  src_addr, dst_addr, length_dw);
629 		return qed_status;
630 	}
631 
632 	if (dst_type == QED_DMAE_ADDRESS_HOST_VIRT)
633 		memcpy((void *)(uintptr_t)(dst_addr),
634 		       &p_hwfn->dmae_info.p_intermediate_buffer[0],
635 		       length_dw * sizeof(u32));
636 
637 	return 0;
638 }
639 
640 static int qed_dmae_execute_command(struct qed_hwfn *p_hwfn,
641 				    struct qed_ptt *p_ptt,
642 				    u64 src_addr, u64 dst_addr,
643 				    u8 src_type, u8 dst_type,
644 				    u32 size_in_dwords,
645 				    struct qed_dmae_params *p_params)
646 {
647 	dma_addr_t phys = p_hwfn->dmae_info.completion_word_phys_addr;
648 	u16 length_cur = 0, i = 0, cnt_split = 0, length_mod = 0;
649 	struct dmae_cmd *cmd = p_hwfn->dmae_info.p_dmae_cmd;
650 	u64 src_addr_split = 0, dst_addr_split = 0;
651 	u16 length_limit = DMAE_MAX_RW_SIZE;
652 	int qed_status = 0;
653 	u32 offset = 0;
654 
655 	qed_dmae_opcode(p_hwfn,
656 			(src_type == QED_DMAE_ADDRESS_GRC),
657 			(dst_type == QED_DMAE_ADDRESS_GRC),
658 			p_params);
659 
660 	cmd->comp_addr_lo = cpu_to_le32(lower_32_bits(phys));
661 	cmd->comp_addr_hi = cpu_to_le32(upper_32_bits(phys));
662 	cmd->comp_val = cpu_to_le32(DMAE_COMPLETION_VAL);
663 
664 	/* Check if the grc_addr is valid like < MAX_GRC_OFFSET */
665 	cnt_split = size_in_dwords / length_limit;
666 	length_mod = size_in_dwords % length_limit;
667 
668 	src_addr_split = src_addr;
669 	dst_addr_split = dst_addr;
670 
671 	for (i = 0; i <= cnt_split; i++) {
672 		offset = length_limit * i;
673 
674 		if (!(p_params->flags & QED_DMAE_FLAG_RW_REPL_SRC)) {
675 			if (src_type == QED_DMAE_ADDRESS_GRC)
676 				src_addr_split = src_addr + offset;
677 			else
678 				src_addr_split = src_addr + (offset * 4);
679 		}
680 
681 		if (dst_type == QED_DMAE_ADDRESS_GRC)
682 			dst_addr_split = dst_addr + offset;
683 		else
684 			dst_addr_split = dst_addr + (offset * 4);
685 
686 		length_cur = (cnt_split == i) ? length_mod : length_limit;
687 
688 		/* might be zero on last iteration */
689 		if (!length_cur)
690 			continue;
691 
692 		qed_status = qed_dmae_execute_sub_operation(p_hwfn,
693 							    p_ptt,
694 							    src_addr_split,
695 							    dst_addr_split,
696 							    src_type,
697 							    dst_type,
698 							    length_cur);
699 		if (qed_status) {
700 			DP_NOTICE(p_hwfn,
701 				  "qed_dmae_execute_sub_operation Failed with error 0x%x. source_addr 0x%llx, destination addr 0x%llx, size_in_dwords 0x%x\n",
702 				  qed_status, src_addr, dst_addr, length_cur);
703 			break;
704 		}
705 	}
706 
707 	return qed_status;
708 }
709 
710 int qed_dmae_host2grc(struct qed_hwfn *p_hwfn,
711 		      struct qed_ptt *p_ptt,
712 		  u64 source_addr, u32 grc_addr, u32 size_in_dwords, u32 flags)
713 {
714 	u32 grc_addr_in_dw = grc_addr / sizeof(u32);
715 	struct qed_dmae_params params;
716 	int rc;
717 
718 	memset(&params, 0, sizeof(struct qed_dmae_params));
719 	params.flags = flags;
720 
721 	mutex_lock(&p_hwfn->dmae_info.mutex);
722 
723 	rc = qed_dmae_execute_command(p_hwfn, p_ptt, source_addr,
724 				      grc_addr_in_dw,
725 				      QED_DMAE_ADDRESS_HOST_VIRT,
726 				      QED_DMAE_ADDRESS_GRC,
727 				      size_in_dwords, &params);
728 
729 	mutex_unlock(&p_hwfn->dmae_info.mutex);
730 
731 	return rc;
732 }
733 
734 int qed_dmae_grc2host(struct qed_hwfn *p_hwfn,
735 		      struct qed_ptt *p_ptt,
736 		      u32 grc_addr,
737 		      dma_addr_t dest_addr, u32 size_in_dwords, u32 flags)
738 {
739 	u32 grc_addr_in_dw = grc_addr / sizeof(u32);
740 	struct qed_dmae_params params;
741 	int rc;
742 
743 	memset(&params, 0, sizeof(struct qed_dmae_params));
744 	params.flags = flags;
745 
746 	mutex_lock(&p_hwfn->dmae_info.mutex);
747 
748 	rc = qed_dmae_execute_command(p_hwfn, p_ptt, grc_addr_in_dw,
749 				      dest_addr, QED_DMAE_ADDRESS_GRC,
750 				      QED_DMAE_ADDRESS_HOST_VIRT,
751 				      size_in_dwords, &params);
752 
753 	mutex_unlock(&p_hwfn->dmae_info.mutex);
754 
755 	return rc;
756 }
757 
758 int qed_dmae_host2host(struct qed_hwfn *p_hwfn,
759 		       struct qed_ptt *p_ptt,
760 		       dma_addr_t source_addr,
761 		       dma_addr_t dest_addr,
762 		       u32 size_in_dwords, struct qed_dmae_params *p_params)
763 {
764 	int rc;
765 
766 	mutex_lock(&(p_hwfn->dmae_info.mutex));
767 
768 	rc = qed_dmae_execute_command(p_hwfn, p_ptt, source_addr,
769 				      dest_addr,
770 				      QED_DMAE_ADDRESS_HOST_PHYS,
771 				      QED_DMAE_ADDRESS_HOST_PHYS,
772 				      size_in_dwords, p_params);
773 
774 	mutex_unlock(&(p_hwfn->dmae_info.mutex));
775 
776 	return rc;
777 }
778 
779 u16 qed_get_qm_pq(struct qed_hwfn *p_hwfn,
780 		  enum protocol_type proto, union qed_qm_pq_params *p_params)
781 {
782 	u16 pq_id = 0;
783 
784 	if ((proto == PROTOCOLID_CORE ||
785 	     proto == PROTOCOLID_ETH ||
786 	     proto == PROTOCOLID_ISCSI ||
787 	     proto == PROTOCOLID_ROCE) && !p_params) {
788 		DP_NOTICE(p_hwfn,
789 			  "Protocol %d received NULL PQ params\n", proto);
790 		return 0;
791 	}
792 
793 	switch (proto) {
794 	case PROTOCOLID_CORE:
795 		if (p_params->core.tc == LB_TC)
796 			pq_id = p_hwfn->qm_info.pure_lb_pq;
797 		else if (p_params->core.tc == OOO_LB_TC)
798 			pq_id = p_hwfn->qm_info.ooo_pq;
799 		else
800 			pq_id = p_hwfn->qm_info.offload_pq;
801 		break;
802 	case PROTOCOLID_ETH:
803 		pq_id = p_params->eth.tc;
804 		if (p_params->eth.is_vf)
805 			pq_id += p_hwfn->qm_info.vf_queues_offset +
806 				 p_params->eth.vf_id;
807 		break;
808 	case PROTOCOLID_ISCSI:
809 		if (p_params->iscsi.q_idx == 1)
810 			pq_id = p_hwfn->qm_info.pure_ack_pq;
811 		break;
812 	case PROTOCOLID_ROCE:
813 		if (p_params->roce.dcqcn)
814 			pq_id = p_params->roce.qpid;
815 		else
816 			pq_id = p_hwfn->qm_info.offload_pq;
817 		if (pq_id > p_hwfn->qm_info.num_pf_rls)
818 			pq_id = p_hwfn->qm_info.offload_pq;
819 		break;
820 	default:
821 		pq_id = 0;
822 	}
823 
824 	pq_id = CM_TX_PQ_BASE + pq_id + RESC_START(p_hwfn, QED_PQ);
825 
826 	return pq_id;
827 }
828