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),
48 					      GFP_KERNEL);
49 	int i;
50 
51 	if (!p_pool)
52 		return -ENOMEM;
53 
54 	INIT_LIST_HEAD(&p_pool->free_list);
55 	for (i = 0; i < PXP_EXTERNAL_BAR_PF_WINDOW_NUM; i++) {
56 		p_pool->ptts[i].idx = i;
57 		p_pool->ptts[i].pxp.offset = QED_BAR_INVALID_OFFSET;
58 		p_pool->ptts[i].pxp.pretend.control = 0;
59 		if (i >= RESERVED_PTT_MAX)
60 			list_add(&p_pool->ptts[i].list_entry,
61 				 &p_pool->free_list);
62 	}
63 
64 	p_hwfn->p_ptt_pool = p_pool;
65 	spin_lock_init(&p_pool->lock);
66 
67 	return 0;
68 }
69 
70 void qed_ptt_invalidate(struct qed_hwfn *p_hwfn)
71 {
72 	struct qed_ptt *p_ptt;
73 	int i;
74 
75 	for (i = 0; i < PXP_EXTERNAL_BAR_PF_WINDOW_NUM; i++) {
76 		p_ptt = &p_hwfn->p_ptt_pool->ptts[i];
77 		p_ptt->pxp.offset = QED_BAR_INVALID_OFFSET;
78 	}
79 }
80 
81 void qed_ptt_pool_free(struct qed_hwfn *p_hwfn)
82 {
83 	kfree(p_hwfn->p_ptt_pool);
84 	p_hwfn->p_ptt_pool = NULL;
85 }
86 
87 struct qed_ptt *qed_ptt_acquire(struct qed_hwfn *p_hwfn)
88 {
89 	struct qed_ptt *p_ptt;
90 	unsigned int i;
91 
92 	/* Take the free PTT from the list */
93 	for (i = 0; i < QED_BAR_ACQUIRE_TIMEOUT; i++) {
94 		spin_lock_bh(&p_hwfn->p_ptt_pool->lock);
95 
96 		if (!list_empty(&p_hwfn->p_ptt_pool->free_list)) {
97 			p_ptt = list_first_entry(&p_hwfn->p_ptt_pool->free_list,
98 						 struct qed_ptt, list_entry);
99 			list_del(&p_ptt->list_entry);
100 
101 			spin_unlock_bh(&p_hwfn->p_ptt_pool->lock);
102 
103 			DP_VERBOSE(p_hwfn, NETIF_MSG_HW,
104 				   "allocated ptt %d\n", p_ptt->idx);
105 			return p_ptt;
106 		}
107 
108 		spin_unlock_bh(&p_hwfn->p_ptt_pool->lock);
109 		usleep_range(1000, 2000);
110 	}
111 
112 	DP_NOTICE(p_hwfn, "PTT acquire timeout - failed to allocate PTT\n");
113 	return NULL;
114 }
115 
116 void qed_ptt_release(struct qed_hwfn *p_hwfn,
117 		     struct qed_ptt *p_ptt)
118 {
119 	spin_lock_bh(&p_hwfn->p_ptt_pool->lock);
120 	list_add(&p_ptt->list_entry, &p_hwfn->p_ptt_pool->free_list);
121 	spin_unlock_bh(&p_hwfn->p_ptt_pool->lock);
122 }
123 
124 u32 qed_ptt_get_hw_addr(struct qed_hwfn *p_hwfn,
125 			struct qed_ptt *p_ptt)
126 {
127 	/* The HW is using DWORDS and we need to translate it to Bytes */
128 	return le32_to_cpu(p_ptt->pxp.offset) << 2;
129 }
130 
131 static u32 qed_ptt_config_addr(struct qed_ptt *p_ptt)
132 {
133 	return PXP_PF_WINDOW_ADMIN_PER_PF_START +
134 	       p_ptt->idx * sizeof(struct pxp_ptt_entry);
135 }
136 
137 u32 qed_ptt_get_bar_addr(struct qed_ptt *p_ptt)
138 {
139 	return PXP_EXTERNAL_BAR_PF_WINDOW_START +
140 	       p_ptt->idx * PXP_EXTERNAL_BAR_PF_WINDOW_SINGLE_SIZE;
141 }
142 
143 void qed_ptt_set_win(struct qed_hwfn *p_hwfn,
144 		     struct qed_ptt *p_ptt,
145 		     u32 new_hw_addr)
146 {
147 	u32 prev_hw_addr;
148 
149 	prev_hw_addr = qed_ptt_get_hw_addr(p_hwfn, p_ptt);
150 
151 	if (new_hw_addr == prev_hw_addr)
152 		return;
153 
154 	/* Update PTT entery in admin window */
155 	DP_VERBOSE(p_hwfn, NETIF_MSG_HW,
156 		   "Updating PTT entry %d to offset 0x%x\n",
157 		   p_ptt->idx, new_hw_addr);
158 
159 	/* The HW is using DWORDS and the address is in Bytes */
160 	p_ptt->pxp.offset = cpu_to_le32(new_hw_addr >> 2);
161 
162 	REG_WR(p_hwfn,
163 	       qed_ptt_config_addr(p_ptt) +
164 	       offsetof(struct pxp_ptt_entry, offset),
165 	       le32_to_cpu(p_ptt->pxp.offset));
166 }
167 
168 static u32 qed_set_ptt(struct qed_hwfn *p_hwfn,
169 		       struct qed_ptt *p_ptt,
170 		       u32 hw_addr)
171 {
172 	u32 win_hw_addr = qed_ptt_get_hw_addr(p_hwfn, p_ptt);
173 	u32 offset;
174 
175 	offset = hw_addr - win_hw_addr;
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,
228 			  u32 hw_addr,
229 			  size_t n,
230 			  bool to_device)
231 {
232 	u32 dw_count, *host_addr, hw_offset;
233 	size_t quota, done = 0;
234 	u32 __iomem *reg_addr;
235 
236 	while (done < n) {
237 		quota = min_t(size_t, n - done,
238 			      PXP_EXTERNAL_BAR_PF_WINDOW_SINGLE_SIZE);
239 
240 		if (IS_PF(p_hwfn->cdev)) {
241 			qed_ptt_set_win(p_hwfn, p_ptt, hw_addr + done);
242 			hw_offset = qed_ptt_get_bar_addr(p_ptt);
243 		} else {
244 			hw_offset = hw_addr + done;
245 		}
246 
247 		dw_count = quota / 4;
248 		host_addr = (u32 *)((u8 *)addr + done);
249 		reg_addr = (u32 __iomem *)REG_ADDR(p_hwfn, hw_offset);
250 		if (to_device)
251 			while (dw_count--)
252 				DIRECT_REG_WR(reg_addr++, *host_addr++);
253 		else
254 			while (dw_count--)
255 				*host_addr++ = DIRECT_REG_RD(reg_addr++);
256 
257 		done += quota;
258 	}
259 }
260 
261 void qed_memcpy_from(struct qed_hwfn *p_hwfn,
262 		     struct qed_ptt *p_ptt,
263 		     void *dest, u32 hw_addr, size_t n)
264 {
265 	DP_VERBOSE(p_hwfn, NETIF_MSG_HW,
266 		   "hw_addr 0x%x, dest %p hw_addr 0x%x, size %lu\n",
267 		   hw_addr, dest, hw_addr, (unsigned long)n);
268 
269 	qed_memcpy_hw(p_hwfn, p_ptt, dest, hw_addr, n, false);
270 }
271 
272 void qed_memcpy_to(struct qed_hwfn *p_hwfn,
273 		   struct qed_ptt *p_ptt,
274 		   u32 hw_addr, void *src, size_t n)
275 {
276 	DP_VERBOSE(p_hwfn, NETIF_MSG_HW,
277 		   "hw_addr 0x%x, hw_addr 0x%x, src %p size %lu\n",
278 		   hw_addr, hw_addr, src, (unsigned long)n);
279 
280 	qed_memcpy_hw(p_hwfn, p_ptt, src, hw_addr, n, true);
281 }
282 
283 void qed_fid_pretend(struct qed_hwfn *p_hwfn,
284 		     struct qed_ptt *p_ptt,
285 		     u16 fid)
286 {
287 	u16 control = 0;
288 
289 	SET_FIELD(control, PXP_PRETEND_CMD_IS_CONCRETE, 1);
290 	SET_FIELD(control, PXP_PRETEND_CMD_PRETEND_FUNCTION, 1);
291 
292 	/* Every pretend undos previous pretends, including
293 	 * previous port pretend.
294 	 */
295 	SET_FIELD(control, PXP_PRETEND_CMD_PORT, 0);
296 	SET_FIELD(control, PXP_PRETEND_CMD_USE_PORT, 0);
297 	SET_FIELD(control, PXP_PRETEND_CMD_PRETEND_PORT, 1);
298 
299 	if (!GET_FIELD(fid, PXP_CONCRETE_FID_VFVALID))
300 		fid = GET_FIELD(fid, PXP_CONCRETE_FID_PFID);
301 
302 	p_ptt->pxp.pretend.control = cpu_to_le16(control);
303 	p_ptt->pxp.pretend.fid.concrete_fid.fid = cpu_to_le16(fid);
304 
305 	REG_WR(p_hwfn,
306 	       qed_ptt_config_addr(p_ptt) +
307 	       offsetof(struct pxp_ptt_entry, pretend),
308 	       *(u32 *)&p_ptt->pxp.pretend);
309 }
310 
311 void qed_port_pretend(struct qed_hwfn *p_hwfn,
312 		      struct qed_ptt *p_ptt,
313 		      u8 port_id)
314 {
315 	u16 control = 0;
316 
317 	SET_FIELD(control, PXP_PRETEND_CMD_PORT, port_id);
318 	SET_FIELD(control, PXP_PRETEND_CMD_USE_PORT, 1);
319 	SET_FIELD(control, PXP_PRETEND_CMD_PRETEND_PORT, 1);
320 
321 	p_ptt->pxp.pretend.control = cpu_to_le16(control);
322 
323 	REG_WR(p_hwfn,
324 	       qed_ptt_config_addr(p_ptt) +
325 	       offsetof(struct pxp_ptt_entry, pretend),
326 	       *(u32 *)&p_ptt->pxp.pretend);
327 }
328 
329 void qed_port_unpretend(struct qed_hwfn *p_hwfn,
330 			struct qed_ptt *p_ptt)
331 {
332 	u16 control = 0;
333 
334 	SET_FIELD(control, PXP_PRETEND_CMD_PORT, 0);
335 	SET_FIELD(control, PXP_PRETEND_CMD_USE_PORT, 0);
336 	SET_FIELD(control, PXP_PRETEND_CMD_PRETEND_PORT, 1);
337 
338 	p_ptt->pxp.pretend.control = cpu_to_le16(control);
339 
340 	REG_WR(p_hwfn,
341 	       qed_ptt_config_addr(p_ptt) +
342 	       offsetof(struct pxp_ptt_entry, pretend),
343 	       *(u32 *)&p_ptt->pxp.pretend);
344 }
345 
346 u32 qed_vfid_to_concrete(struct qed_hwfn *p_hwfn, u8 vfid)
347 {
348 	u32 concrete_fid = 0;
349 
350 	SET_FIELD(concrete_fid, PXP_CONCRETE_FID_PFID, p_hwfn->rel_pf_id);
351 	SET_FIELD(concrete_fid, PXP_CONCRETE_FID_VFID, vfid);
352 	SET_FIELD(concrete_fid, PXP_CONCRETE_FID_VFVALID, 1);
353 
354 	return concrete_fid;
355 }
356 
357 /* DMAE */
358 static void qed_dmae_opcode(struct qed_hwfn *p_hwfn,
359 			    const u8 is_src_type_grc,
360 			    const u8 is_dst_type_grc,
361 			    struct qed_dmae_params *p_params)
362 {
363 	u16 opcode_b = 0;
364 	u32 opcode = 0;
365 
366 	/* Whether the source is the PCIe or the GRC.
367 	 * 0- The source is the PCIe
368 	 * 1- The source is the GRC.
369 	 */
370 	opcode |= (is_src_type_grc ? DMAE_CMD_SRC_MASK_GRC
371 				   : DMAE_CMD_SRC_MASK_PCIE) <<
372 		   DMAE_CMD_SRC_SHIFT;
373 	opcode |= ((p_hwfn->rel_pf_id & DMAE_CMD_SRC_PF_ID_MASK) <<
374 		   DMAE_CMD_SRC_PF_ID_SHIFT);
375 
376 	/* The destination of the DMA can be: 0-None 1-PCIe 2-GRC 3-None */
377 	opcode |= (is_dst_type_grc ? DMAE_CMD_DST_MASK_GRC
378 				   : DMAE_CMD_DST_MASK_PCIE) <<
379 		   DMAE_CMD_DST_SHIFT;
380 	opcode |= ((p_hwfn->rel_pf_id & DMAE_CMD_DST_PF_ID_MASK) <<
381 		   DMAE_CMD_DST_PF_ID_SHIFT);
382 
383 	/* Whether to write a completion word to the completion destination:
384 	 * 0-Do not write a completion word
385 	 * 1-Write the completion word
386 	 */
387 	opcode |= (DMAE_CMD_COMP_WORD_EN_MASK << DMAE_CMD_COMP_WORD_EN_SHIFT);
388 	opcode |= (DMAE_CMD_SRC_ADDR_RESET_MASK <<
389 		   DMAE_CMD_SRC_ADDR_RESET_SHIFT);
390 
391 	if (p_params->flags & QED_DMAE_FLAG_COMPLETION_DST)
392 		opcode |= (1 << DMAE_CMD_COMP_FUNC_SHIFT);
393 
394 	opcode |= (DMAE_CMD_ENDIANITY << DMAE_CMD_ENDIANITY_MODE_SHIFT);
395 
396 	opcode |= ((p_hwfn->port_id) << DMAE_CMD_PORT_ID_SHIFT);
397 
398 	/* reset source address in next go */
399 	opcode |= (DMAE_CMD_SRC_ADDR_RESET_MASK <<
400 		   DMAE_CMD_SRC_ADDR_RESET_SHIFT);
401 
402 	/* reset dest address in next go */
403 	opcode |= (DMAE_CMD_DST_ADDR_RESET_MASK <<
404 		   DMAE_CMD_DST_ADDR_RESET_SHIFT);
405 
406 	/* SRC/DST VFID: all 1's - pf, otherwise VF id */
407 	if (p_params->flags & QED_DMAE_FLAG_VF_SRC) {
408 		opcode |= 1 << DMAE_CMD_SRC_VF_ID_VALID_SHIFT;
409 		opcode_b |= p_params->src_vfid << DMAE_CMD_SRC_VF_ID_SHIFT;
410 	} else {
411 		opcode_b |= DMAE_CMD_SRC_VF_ID_MASK <<
412 			    DMAE_CMD_SRC_VF_ID_SHIFT;
413 	}
414 
415 	if (p_params->flags & QED_DMAE_FLAG_VF_DST) {
416 		opcode |= 1 << DMAE_CMD_DST_VF_ID_VALID_SHIFT;
417 		opcode_b |= p_params->dst_vfid << DMAE_CMD_DST_VF_ID_SHIFT;
418 	} else {
419 		opcode_b |= DMAE_CMD_DST_VF_ID_MASK << DMAE_CMD_DST_VF_ID_SHIFT;
420 	}
421 
422 	p_hwfn->dmae_info.p_dmae_cmd->opcode = cpu_to_le32(opcode);
423 	p_hwfn->dmae_info.p_dmae_cmd->opcode_b = cpu_to_le16(opcode_b);
424 }
425 
426 u32 qed_dmae_idx_to_go_cmd(u8 idx)
427 {
428 	/* All the DMAE 'go' registers form an array in internal memory */
429 	return DMAE_REG_GO_C0 + (idx << 2);
430 }
431 
432 static int
433 qed_dmae_post_command(struct qed_hwfn *p_hwfn,
434 		      struct qed_ptt *p_ptt)
435 {
436 	struct dmae_cmd *command = p_hwfn->dmae_info.p_dmae_cmd;
437 	u8 idx_cmd = p_hwfn->dmae_info.channel, i;
438 	int qed_status = 0;
439 
440 	/* verify address is not NULL */
441 	if ((((command->dst_addr_lo == 0) && (command->dst_addr_hi == 0)) ||
442 	     ((command->src_addr_lo == 0) && (command->src_addr_hi == 0)))) {
443 		DP_NOTICE(p_hwfn,
444 			  "source or destination address 0 idx_cmd=%d\n"
445 			  "opcode = [0x%08x,0x%04x] len=0x%x src=0x%x:%x dst=0x%x:%x\n",
446 			   idx_cmd,
447 			   le32_to_cpu(command->opcode),
448 			   le16_to_cpu(command->opcode_b),
449 			   le16_to_cpu(command->length_dw),
450 			   le32_to_cpu(command->src_addr_hi),
451 			   le32_to_cpu(command->src_addr_lo),
452 			   le32_to_cpu(command->dst_addr_hi),
453 			   le32_to_cpu(command->dst_addr_lo));
454 
455 		return -EINVAL;
456 	}
457 
458 	DP_VERBOSE(p_hwfn,
459 		   NETIF_MSG_HW,
460 		   "Posting DMAE command [idx %d]: opcode = [0x%08x,0x%04x] len=0x%x src=0x%x:%x dst=0x%x:%x\n",
461 		   idx_cmd,
462 		   le32_to_cpu(command->opcode),
463 		   le16_to_cpu(command->opcode_b),
464 		   le16_to_cpu(command->length_dw),
465 		   le32_to_cpu(command->src_addr_hi),
466 		   le32_to_cpu(command->src_addr_lo),
467 		   le32_to_cpu(command->dst_addr_hi),
468 		   le32_to_cpu(command->dst_addr_lo));
469 
470 	/* Copy the command to DMAE - need to do it before every call
471 	 * for source/dest address no reset.
472 	 * The first 9 DWs are the command registers, the 10 DW is the
473 	 * GO register, and the rest are result registers
474 	 * (which are read only by the client).
475 	 */
476 	for (i = 0; i < DMAE_CMD_SIZE; i++) {
477 		u32 data = (i < DMAE_CMD_SIZE_TO_FILL) ?
478 			   *(((u32 *)command) + i) : 0;
479 
480 		qed_wr(p_hwfn, p_ptt,
481 		       DMAE_REG_CMD_MEM +
482 		       (idx_cmd * DMAE_CMD_SIZE * sizeof(u32)) +
483 		       (i * sizeof(u32)), data);
484 	}
485 
486 	qed_wr(p_hwfn, p_ptt,
487 	       qed_dmae_idx_to_go_cmd(idx_cmd),
488 	       DMAE_GO_VALUE);
489 
490 	return qed_status;
491 }
492 
493 int qed_dmae_info_alloc(struct qed_hwfn *p_hwfn)
494 {
495 	dma_addr_t *p_addr = &p_hwfn->dmae_info.completion_word_phys_addr;
496 	struct dmae_cmd **p_cmd = &p_hwfn->dmae_info.p_dmae_cmd;
497 	u32 **p_buff = &p_hwfn->dmae_info.p_intermediate_buffer;
498 	u32 **p_comp = &p_hwfn->dmae_info.p_completion_word;
499 
500 	*p_comp = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
501 				     sizeof(u32),
502 				     p_addr,
503 				     GFP_KERNEL);
504 	if (!*p_comp) {
505 		DP_NOTICE(p_hwfn, "Failed to allocate `p_completion_word'\n");
506 		goto err;
507 	}
508 
509 	p_addr = &p_hwfn->dmae_info.dmae_cmd_phys_addr;
510 	*p_cmd = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
511 				    sizeof(struct dmae_cmd),
512 				    p_addr, GFP_KERNEL);
513 	if (!*p_cmd) {
514 		DP_NOTICE(p_hwfn, "Failed to allocate `struct dmae_cmd'\n");
515 		goto err;
516 	}
517 
518 	p_addr = &p_hwfn->dmae_info.intermediate_buffer_phys_addr;
519 	*p_buff = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
520 				     sizeof(u32) * DMAE_MAX_RW_SIZE,
521 				     p_addr, GFP_KERNEL);
522 	if (!*p_buff) {
523 		DP_NOTICE(p_hwfn, "Failed to allocate `intermediate_buffer'\n");
524 		goto err;
525 	}
526 
527 	p_hwfn->dmae_info.channel = p_hwfn->rel_pf_id;
528 
529 	return 0;
530 err:
531 	qed_dmae_info_free(p_hwfn);
532 	return -ENOMEM;
533 }
534 
535 void qed_dmae_info_free(struct qed_hwfn *p_hwfn)
536 {
537 	dma_addr_t p_phys;
538 
539 	/* Just make sure no one is in the middle */
540 	mutex_lock(&p_hwfn->dmae_info.mutex);
541 
542 	if (p_hwfn->dmae_info.p_completion_word) {
543 		p_phys = p_hwfn->dmae_info.completion_word_phys_addr;
544 		dma_free_coherent(&p_hwfn->cdev->pdev->dev,
545 				  sizeof(u32),
546 				  p_hwfn->dmae_info.p_completion_word,
547 				  p_phys);
548 		p_hwfn->dmae_info.p_completion_word = NULL;
549 	}
550 
551 	if (p_hwfn->dmae_info.p_dmae_cmd) {
552 		p_phys = p_hwfn->dmae_info.dmae_cmd_phys_addr;
553 		dma_free_coherent(&p_hwfn->cdev->pdev->dev,
554 				  sizeof(struct dmae_cmd),
555 				  p_hwfn->dmae_info.p_dmae_cmd,
556 				  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 = 0;
575 	u32 wait_cnt_limit = 10000;
576 
577 	int qed_status = 0;
578 
579 	barrier();
580 	while (*p_hwfn->dmae_info.p_completion_word != DMAE_COMPLETION_VAL) {
581 		udelay(DMAE_MIN_WAIT_TIME);
582 		if (++wait_cnt > wait_cnt_limit) {
583 			DP_NOTICE(p_hwfn->cdev,
584 				  "Timed-out waiting for operation to complete. Completion word is 0x%08x expected 0x%08x.\n",
585 				  *p_hwfn->dmae_info.p_completion_word,
586 				 DMAE_COMPLETION_VAL);
587 			qed_status = -EBUSY;
588 			break;
589 		}
590 
591 		/* to sync the completion_word since we are not
592 		 * using the volatile keyword for p_completion_word
593 		 */
594 		barrier();
595 	}
596 
597 	if (qed_status == 0)
598 		*p_hwfn->dmae_info.p_completion_word = 0;
599 
600 	return qed_status;
601 }
602 
603 static int qed_dmae_execute_sub_operation(struct qed_hwfn *p_hwfn,
604 					  struct qed_ptt *p_ptt,
605 					  u64 src_addr,
606 					  u64 dst_addr,
607 					  u8 src_type,
608 					  u8 dst_type,
609 					  u32 length)
610 {
611 	dma_addr_t phys = p_hwfn->dmae_info.intermediate_buffer_phys_addr;
612 	struct dmae_cmd *cmd = p_hwfn->dmae_info.p_dmae_cmd;
613 	int qed_status = 0;
614 
615 	switch (src_type) {
616 	case QED_DMAE_ADDRESS_GRC:
617 	case QED_DMAE_ADDRESS_HOST_PHYS:
618 		cmd->src_addr_hi = cpu_to_le32(upper_32_bits(src_addr));
619 		cmd->src_addr_lo = cpu_to_le32(lower_32_bits(src_addr));
620 		break;
621 	/* for virtual source addresses we use the intermediate buffer. */
622 	case QED_DMAE_ADDRESS_HOST_VIRT:
623 		cmd->src_addr_hi = cpu_to_le32(upper_32_bits(phys));
624 		cmd->src_addr_lo = cpu_to_le32(lower_32_bits(phys));
625 		memcpy(&p_hwfn->dmae_info.p_intermediate_buffer[0],
626 		       (void *)(uintptr_t)src_addr,
627 		       length * sizeof(u32));
628 		break;
629 	default:
630 		return -EINVAL;
631 	}
632 
633 	switch (dst_type) {
634 	case QED_DMAE_ADDRESS_GRC:
635 	case QED_DMAE_ADDRESS_HOST_PHYS:
636 		cmd->dst_addr_hi = cpu_to_le32(upper_32_bits(dst_addr));
637 		cmd->dst_addr_lo = cpu_to_le32(lower_32_bits(dst_addr));
638 		break;
639 	/* for virtual source addresses we use the intermediate buffer. */
640 	case QED_DMAE_ADDRESS_HOST_VIRT:
641 		cmd->dst_addr_hi = cpu_to_le32(upper_32_bits(phys));
642 		cmd->dst_addr_lo = cpu_to_le32(lower_32_bits(phys));
643 		break;
644 	default:
645 		return -EINVAL;
646 	}
647 
648 	cmd->length_dw = cpu_to_le16((u16)length);
649 
650 	qed_dmae_post_command(p_hwfn, p_ptt);
651 
652 	qed_status = qed_dmae_operation_wait(p_hwfn);
653 
654 	if (qed_status) {
655 		DP_NOTICE(p_hwfn,
656 			  "qed_dmae_host2grc: Wait Failed. source_addr 0x%llx, grc_addr 0x%llx, size_in_dwords 0x%x\n",
657 			  src_addr,
658 			  dst_addr,
659 			  length);
660 		return qed_status;
661 	}
662 
663 	if (dst_type == QED_DMAE_ADDRESS_HOST_VIRT)
664 		memcpy((void *)(uintptr_t)(dst_addr),
665 		       &p_hwfn->dmae_info.p_intermediate_buffer[0],
666 		       length * sizeof(u32));
667 
668 	return 0;
669 }
670 
671 static int qed_dmae_execute_command(struct qed_hwfn *p_hwfn,
672 				    struct qed_ptt *p_ptt,
673 				    u64 src_addr, u64 dst_addr,
674 				    u8 src_type, u8 dst_type,
675 				    u32 size_in_dwords,
676 				    struct qed_dmae_params *p_params)
677 {
678 	dma_addr_t phys = p_hwfn->dmae_info.completion_word_phys_addr;
679 	u16 length_cur = 0, i = 0, cnt_split = 0, length_mod = 0;
680 	struct dmae_cmd *cmd = p_hwfn->dmae_info.p_dmae_cmd;
681 	u64 src_addr_split = 0, dst_addr_split = 0;
682 	u16 length_limit = DMAE_MAX_RW_SIZE;
683 	int qed_status = 0;
684 	u32 offset = 0;
685 
686 	qed_dmae_opcode(p_hwfn,
687 			(src_type == QED_DMAE_ADDRESS_GRC),
688 			(dst_type == QED_DMAE_ADDRESS_GRC),
689 			p_params);
690 
691 	cmd->comp_addr_lo = cpu_to_le32(lower_32_bits(phys));
692 	cmd->comp_addr_hi = cpu_to_le32(upper_32_bits(phys));
693 	cmd->comp_val = cpu_to_le32(DMAE_COMPLETION_VAL);
694 
695 	/* Check if the grc_addr is valid like < MAX_GRC_OFFSET */
696 	cnt_split = size_in_dwords / length_limit;
697 	length_mod = size_in_dwords % length_limit;
698 
699 	src_addr_split = src_addr;
700 	dst_addr_split = dst_addr;
701 
702 	for (i = 0; i <= cnt_split; i++) {
703 		offset = length_limit * i;
704 
705 		if (!(p_params->flags & QED_DMAE_FLAG_RW_REPL_SRC)) {
706 			if (src_type == QED_DMAE_ADDRESS_GRC)
707 				src_addr_split = src_addr + offset;
708 			else
709 				src_addr_split = src_addr + (offset * 4);
710 		}
711 
712 		if (dst_type == QED_DMAE_ADDRESS_GRC)
713 			dst_addr_split = dst_addr + offset;
714 		else
715 			dst_addr_split = dst_addr + (offset * 4);
716 
717 		length_cur = (cnt_split == i) ? length_mod : length_limit;
718 
719 		/* might be zero on last iteration */
720 		if (!length_cur)
721 			continue;
722 
723 		qed_status = qed_dmae_execute_sub_operation(p_hwfn,
724 							    p_ptt,
725 							    src_addr_split,
726 							    dst_addr_split,
727 							    src_type,
728 							    dst_type,
729 							    length_cur);
730 		if (qed_status) {
731 			DP_NOTICE(p_hwfn,
732 				  "qed_dmae_execute_sub_operation Failed with error 0x%x. source_addr 0x%llx, destination addr 0x%llx, size_in_dwords 0x%x\n",
733 				  qed_status,
734 				  src_addr,
735 				  dst_addr,
736 				  length_cur);
737 			break;
738 		}
739 	}
740 
741 	return qed_status;
742 }
743 
744 int qed_dmae_host2grc(struct qed_hwfn *p_hwfn,
745 		      struct qed_ptt *p_ptt,
746 		      u64 source_addr,
747 		      u32 grc_addr,
748 		      u32 size_in_dwords,
749 		      u32 flags)
750 {
751 	u32 grc_addr_in_dw = grc_addr / sizeof(u32);
752 	struct qed_dmae_params params;
753 	int rc;
754 
755 	memset(&params, 0, sizeof(struct qed_dmae_params));
756 	params.flags = flags;
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, &params);
765 
766 	mutex_unlock(&p_hwfn->dmae_info.mutex);
767 
768 	return rc;
769 }
770 
771 int
772 qed_dmae_host2host(struct qed_hwfn *p_hwfn,
773 		   struct qed_ptt *p_ptt,
774 		   dma_addr_t source_addr,
775 		   dma_addr_t dest_addr,
776 		   u32 size_in_dwords, struct qed_dmae_params *p_params)
777 {
778 	int rc;
779 
780 	mutex_lock(&(p_hwfn->dmae_info.mutex));
781 
782 	rc = qed_dmae_execute_command(p_hwfn, p_ptt, source_addr,
783 				      dest_addr,
784 				      QED_DMAE_ADDRESS_HOST_PHYS,
785 				      QED_DMAE_ADDRESS_HOST_PHYS,
786 				      size_in_dwords, p_params);
787 
788 	mutex_unlock(&(p_hwfn->dmae_info.mutex));
789 
790 	return rc;
791 }
792 
793 u16 qed_get_qm_pq(struct qed_hwfn *p_hwfn,
794 		  enum protocol_type proto,
795 		  union qed_qm_pq_params *p_params)
796 {
797 	u16 pq_id = 0;
798 
799 	if ((proto == PROTOCOLID_CORE || proto == PROTOCOLID_ETH) &&
800 	    !p_params) {
801 		DP_NOTICE(p_hwfn,
802 			  "Protocol %d received NULL PQ params\n",
803 			  proto);
804 		return 0;
805 	}
806 
807 	switch (proto) {
808 	case PROTOCOLID_CORE:
809 		if (p_params->core.tc == LB_TC)
810 			pq_id = p_hwfn->qm_info.pure_lb_pq;
811 		else
812 			pq_id = p_hwfn->qm_info.offload_pq;
813 		break;
814 	case PROTOCOLID_ETH:
815 		pq_id = p_params->eth.tc;
816 		if (p_params->eth.is_vf)
817 			pq_id += p_hwfn->qm_info.vf_queues_offset +
818 				 p_params->eth.vf_id;
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