1 /*
2  * Broadcom NetXtreme-E RoCE driver.
3  *
4  * Copyright (c) 2016 - 2017, Broadcom. All rights reserved.  The term
5  * Broadcom refers to Broadcom Limited and/or its subsidiaries.
6  *
7  * This software is available to you under a choice of one of two
8  * licenses.  You may choose to be licensed under the terms of the GNU
9  * General Public License (GPL) Version 2, available from the file
10  * COPYING in the main directory of this source tree, or the
11  * BSD license below:
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  *
17  * 1. Redistributions of source code must retain the above copyright
18  *    notice, this list of conditions and the following disclaimer.
19  * 2. Redistributions in binary form must reproduce the above copyright
20  *    notice, this list of conditions and the following disclaimer in
21  *    the documentation and/or other materials provided with the
22  *    distribution.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS''
25  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
26  * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
27  * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS
28  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
29  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
30  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
31  * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
32  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE
33  * OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN
34  * IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
35  *
36  * Description: QPLib resource manager
37  */
38 
39 #define dev_fmt(fmt) "QPLIB: " fmt
40 
41 #include <linux/spinlock.h>
42 #include <linux/pci.h>
43 #include <linux/interrupt.h>
44 #include <linux/inetdevice.h>
45 #include <linux/dma-mapping.h>
46 #include <linux/if_vlan.h>
47 #include <linux/vmalloc.h>
48 #include <rdma/ib_verbs.h>
49 #include <rdma/ib_umem.h>
50 
51 #include "roce_hsi.h"
52 #include "qplib_res.h"
53 #include "qplib_sp.h"
54 #include "qplib_rcfw.h"
55 
56 static void bnxt_qplib_free_stats_ctx(struct pci_dev *pdev,
57 				      struct bnxt_qplib_stats *stats);
58 static int bnxt_qplib_alloc_stats_ctx(struct pci_dev *pdev,
59 				      struct bnxt_qplib_chip_ctx *cctx,
60 				      struct bnxt_qplib_stats *stats);
61 
62 /* PBL */
63 static void __free_pbl(struct bnxt_qplib_res *res, struct bnxt_qplib_pbl *pbl,
64 		       bool is_umem)
65 {
66 	struct pci_dev *pdev = res->pdev;
67 	int i;
68 
69 	if (!is_umem) {
70 		for (i = 0; i < pbl->pg_count; i++) {
71 			if (pbl->pg_arr[i])
72 				dma_free_coherent(&pdev->dev, pbl->pg_size,
73 						  (void *)((unsigned long)
74 						   pbl->pg_arr[i] &
75 						  PAGE_MASK),
76 						  pbl->pg_map_arr[i]);
77 			else
78 				dev_warn(&pdev->dev,
79 					 "PBL free pg_arr[%d] empty?!\n", i);
80 			pbl->pg_arr[i] = NULL;
81 		}
82 	}
83 	vfree(pbl->pg_arr);
84 	pbl->pg_arr = NULL;
85 	vfree(pbl->pg_map_arr);
86 	pbl->pg_map_arr = NULL;
87 	pbl->pg_count = 0;
88 	pbl->pg_size = 0;
89 }
90 
91 static void bnxt_qplib_fill_user_dma_pages(struct bnxt_qplib_pbl *pbl,
92 					   struct bnxt_qplib_sg_info *sginfo)
93 {
94 	struct ib_block_iter biter;
95 	int i = 0;
96 
97 	rdma_umem_for_each_dma_block(sginfo->umem, &biter, sginfo->pgsize) {
98 		pbl->pg_map_arr[i] = rdma_block_iter_dma_address(&biter);
99 		pbl->pg_arr[i] = NULL;
100 		pbl->pg_count++;
101 		i++;
102 	}
103 }
104 
105 static int __alloc_pbl(struct bnxt_qplib_res *res,
106 		       struct bnxt_qplib_pbl *pbl,
107 		       struct bnxt_qplib_sg_info *sginfo)
108 {
109 	struct pci_dev *pdev = res->pdev;
110 	bool is_umem = false;
111 	u32 pages;
112 	int i;
113 
114 	if (sginfo->nopte)
115 		return 0;
116 	if (sginfo->umem)
117 		pages = ib_umem_num_dma_blocks(sginfo->umem, sginfo->pgsize);
118 	else
119 		pages = sginfo->npages;
120 	/* page ptr arrays */
121 	pbl->pg_arr = vmalloc(pages * sizeof(void *));
122 	if (!pbl->pg_arr)
123 		return -ENOMEM;
124 
125 	pbl->pg_map_arr = vmalloc(pages * sizeof(dma_addr_t));
126 	if (!pbl->pg_map_arr) {
127 		vfree(pbl->pg_arr);
128 		pbl->pg_arr = NULL;
129 		return -ENOMEM;
130 	}
131 	pbl->pg_count = 0;
132 	pbl->pg_size = sginfo->pgsize;
133 
134 	if (!sginfo->umem) {
135 		for (i = 0; i < pages; i++) {
136 			pbl->pg_arr[i] = dma_alloc_coherent(&pdev->dev,
137 							    pbl->pg_size,
138 							    &pbl->pg_map_arr[i],
139 							    GFP_KERNEL);
140 			if (!pbl->pg_arr[i])
141 				goto fail;
142 			pbl->pg_count++;
143 		}
144 	} else {
145 		is_umem = true;
146 		bnxt_qplib_fill_user_dma_pages(pbl, sginfo);
147 	}
148 
149 	return 0;
150 fail:
151 	__free_pbl(res, pbl, is_umem);
152 	return -ENOMEM;
153 }
154 
155 /* HWQ */
156 void bnxt_qplib_free_hwq(struct bnxt_qplib_res *res,
157 			 struct bnxt_qplib_hwq *hwq)
158 {
159 	int i;
160 
161 	if (!hwq->max_elements)
162 		return;
163 	if (hwq->level >= PBL_LVL_MAX)
164 		return;
165 
166 	for (i = 0; i < hwq->level + 1; i++) {
167 		if (i == hwq->level)
168 			__free_pbl(res, &hwq->pbl[i], hwq->is_user);
169 		else
170 			__free_pbl(res, &hwq->pbl[i], false);
171 	}
172 
173 	hwq->level = PBL_LVL_MAX;
174 	hwq->max_elements = 0;
175 	hwq->element_size = 0;
176 	hwq->prod = 0;
177 	hwq->cons = 0;
178 	hwq->cp_bit = 0;
179 }
180 
181 /* All HWQs are power of 2 in size */
182 
183 int bnxt_qplib_alloc_init_hwq(struct bnxt_qplib_hwq *hwq,
184 			      struct bnxt_qplib_hwq_attr *hwq_attr)
185 {
186 	u32 npages, aux_slots, pg_size, aux_pages = 0, aux_size = 0;
187 	struct bnxt_qplib_sg_info sginfo = {};
188 	u32 depth, stride, npbl, npde;
189 	dma_addr_t *src_phys_ptr, **dst_virt_ptr;
190 	struct bnxt_qplib_res *res;
191 	struct pci_dev *pdev;
192 	int i, rc, lvl;
193 
194 	res = hwq_attr->res;
195 	pdev = res->pdev;
196 	pg_size = hwq_attr->sginfo->pgsize;
197 	hwq->level = PBL_LVL_MAX;
198 
199 	depth = roundup_pow_of_two(hwq_attr->depth);
200 	stride = roundup_pow_of_two(hwq_attr->stride);
201 	if (hwq_attr->aux_depth) {
202 		aux_slots = hwq_attr->aux_depth;
203 		aux_size = roundup_pow_of_two(hwq_attr->aux_stride);
204 		aux_pages = (aux_slots * aux_size) / pg_size;
205 		if ((aux_slots * aux_size) % pg_size)
206 			aux_pages++;
207 	}
208 
209 	if (!hwq_attr->sginfo->umem) {
210 		hwq->is_user = false;
211 		npages = (depth * stride) / pg_size + aux_pages;
212 		if ((depth * stride) % pg_size)
213 			npages++;
214 		if (!npages)
215 			return -EINVAL;
216 		hwq_attr->sginfo->npages = npages;
217 	} else {
218 		unsigned long sginfo_num_pages = ib_umem_num_dma_blocks(
219 			hwq_attr->sginfo->umem, hwq_attr->sginfo->pgsize);
220 
221 		hwq->is_user = true;
222 		npages = sginfo_num_pages;
223 		npages = (npages * PAGE_SIZE) /
224 			  BIT_ULL(hwq_attr->sginfo->pgshft);
225 		if ((sginfo_num_pages * PAGE_SIZE) %
226 		     BIT_ULL(hwq_attr->sginfo->pgshft))
227 			if (!npages)
228 				npages++;
229 	}
230 
231 	if (npages == MAX_PBL_LVL_0_PGS) {
232 		/* This request is Level 0, map PTE */
233 		rc = __alloc_pbl(res, &hwq->pbl[PBL_LVL_0], hwq_attr->sginfo);
234 		if (rc)
235 			goto fail;
236 		hwq->level = PBL_LVL_0;
237 	}
238 
239 	if (npages > MAX_PBL_LVL_0_PGS) {
240 		if (npages > MAX_PBL_LVL_1_PGS) {
241 			u32 flag = (hwq_attr->type == HWQ_TYPE_L2_CMPL) ?
242 				    0 : PTU_PTE_VALID;
243 			/* 2 levels of indirection */
244 			npbl = npages >> MAX_PBL_LVL_1_PGS_SHIFT;
245 			if (npages % BIT(MAX_PBL_LVL_1_PGS_SHIFT))
246 				npbl++;
247 			npde = npbl >> MAX_PDL_LVL_SHIFT;
248 			if (npbl % BIT(MAX_PDL_LVL_SHIFT))
249 				npde++;
250 			/* Alloc PDE pages */
251 			sginfo.pgsize = npde * pg_size;
252 			sginfo.npages = 1;
253 			rc = __alloc_pbl(res, &hwq->pbl[PBL_LVL_0], &sginfo);
254 
255 			/* Alloc PBL pages */
256 			sginfo.npages = npbl;
257 			sginfo.pgsize = PAGE_SIZE;
258 			rc = __alloc_pbl(res, &hwq->pbl[PBL_LVL_1], &sginfo);
259 			if (rc)
260 				goto fail;
261 			/* Fill PDL with PBL page pointers */
262 			dst_virt_ptr =
263 				(dma_addr_t **)hwq->pbl[PBL_LVL_0].pg_arr;
264 			src_phys_ptr = hwq->pbl[PBL_LVL_1].pg_map_arr;
265 			if (hwq_attr->type == HWQ_TYPE_MR) {
266 			/* For MR it is expected that we supply only 1 contigous
267 			 * page i.e only 1 entry in the PDL that will contain
268 			 * all the PBLs for the user supplied memory region
269 			 */
270 				for (i = 0; i < hwq->pbl[PBL_LVL_1].pg_count;
271 				     i++)
272 					dst_virt_ptr[0][i] = src_phys_ptr[i] |
273 						flag;
274 			} else {
275 				for (i = 0; i < hwq->pbl[PBL_LVL_1].pg_count;
276 				     i++)
277 					dst_virt_ptr[PTR_PG(i)][PTR_IDX(i)] =
278 						src_phys_ptr[i] |
279 						PTU_PDE_VALID;
280 			}
281 			/* Alloc or init PTEs */
282 			rc = __alloc_pbl(res, &hwq->pbl[PBL_LVL_2],
283 					 hwq_attr->sginfo);
284 			if (rc)
285 				goto fail;
286 			hwq->level = PBL_LVL_2;
287 			if (hwq_attr->sginfo->nopte)
288 				goto done;
289 			/* Fill PBLs with PTE pointers */
290 			dst_virt_ptr =
291 				(dma_addr_t **)hwq->pbl[PBL_LVL_1].pg_arr;
292 			src_phys_ptr = hwq->pbl[PBL_LVL_2].pg_map_arr;
293 			for (i = 0; i < hwq->pbl[PBL_LVL_2].pg_count; i++) {
294 				dst_virt_ptr[PTR_PG(i)][PTR_IDX(i)] =
295 					src_phys_ptr[i] | PTU_PTE_VALID;
296 			}
297 			if (hwq_attr->type == HWQ_TYPE_QUEUE) {
298 				/* Find the last pg of the size */
299 				i = hwq->pbl[PBL_LVL_2].pg_count;
300 				dst_virt_ptr[PTR_PG(i - 1)][PTR_IDX(i - 1)] |=
301 								  PTU_PTE_LAST;
302 				if (i > 1)
303 					dst_virt_ptr[PTR_PG(i - 2)]
304 						    [PTR_IDX(i - 2)] |=
305 						    PTU_PTE_NEXT_TO_LAST;
306 			}
307 		} else { /* pages < 512 npbl = 1, npde = 0 */
308 			u32 flag = (hwq_attr->type == HWQ_TYPE_L2_CMPL) ?
309 				    0 : PTU_PTE_VALID;
310 
311 			/* 1 level of indirection */
312 			npbl = npages >> MAX_PBL_LVL_1_PGS_SHIFT;
313 			if (npages % BIT(MAX_PBL_LVL_1_PGS_SHIFT))
314 				npbl++;
315 			sginfo.npages = npbl;
316 			sginfo.pgsize = PAGE_SIZE;
317 			/* Alloc PBL page */
318 			rc = __alloc_pbl(res, &hwq->pbl[PBL_LVL_0], &sginfo);
319 			if (rc)
320 				goto fail;
321 			/* Alloc or init  PTEs */
322 			rc = __alloc_pbl(res, &hwq->pbl[PBL_LVL_1],
323 					 hwq_attr->sginfo);
324 			if (rc)
325 				goto fail;
326 			hwq->level = PBL_LVL_1;
327 			if (hwq_attr->sginfo->nopte)
328 				goto done;
329 			/* Fill PBL with PTE pointers */
330 			dst_virt_ptr =
331 				(dma_addr_t **)hwq->pbl[PBL_LVL_0].pg_arr;
332 			src_phys_ptr = hwq->pbl[PBL_LVL_1].pg_map_arr;
333 			for (i = 0; i < hwq->pbl[PBL_LVL_1].pg_count; i++)
334 				dst_virt_ptr[PTR_PG(i)][PTR_IDX(i)] =
335 					src_phys_ptr[i] | flag;
336 			if (hwq_attr->type == HWQ_TYPE_QUEUE) {
337 				/* Find the last pg of the size */
338 				i = hwq->pbl[PBL_LVL_1].pg_count;
339 				dst_virt_ptr[PTR_PG(i - 1)][PTR_IDX(i - 1)] |=
340 								  PTU_PTE_LAST;
341 				if (i > 1)
342 					dst_virt_ptr[PTR_PG(i - 2)]
343 						    [PTR_IDX(i - 2)] |=
344 						    PTU_PTE_NEXT_TO_LAST;
345 			}
346 		}
347 	}
348 done:
349 	hwq->prod = 0;
350 	hwq->cons = 0;
351 	hwq->pdev = pdev;
352 	hwq->depth = hwq_attr->depth;
353 	hwq->max_elements = depth;
354 	hwq->element_size = stride;
355 	hwq->qe_ppg = pg_size / stride;
356 	/* For direct access to the elements */
357 	lvl = hwq->level;
358 	if (hwq_attr->sginfo->nopte && hwq->level)
359 		lvl = hwq->level - 1;
360 	hwq->pbl_ptr = hwq->pbl[lvl].pg_arr;
361 	hwq->pbl_dma_ptr = hwq->pbl[lvl].pg_map_arr;
362 	spin_lock_init(&hwq->lock);
363 
364 	return 0;
365 fail:
366 	bnxt_qplib_free_hwq(res, hwq);
367 	return -ENOMEM;
368 }
369 
370 /* Context Tables */
371 void bnxt_qplib_free_ctx(struct bnxt_qplib_res *res,
372 			 struct bnxt_qplib_ctx *ctx)
373 {
374 	int i;
375 
376 	bnxt_qplib_free_hwq(res, &ctx->qpc_tbl);
377 	bnxt_qplib_free_hwq(res, &ctx->mrw_tbl);
378 	bnxt_qplib_free_hwq(res, &ctx->srqc_tbl);
379 	bnxt_qplib_free_hwq(res, &ctx->cq_tbl);
380 	bnxt_qplib_free_hwq(res, &ctx->tim_tbl);
381 	for (i = 0; i < MAX_TQM_ALLOC_REQ; i++)
382 		bnxt_qplib_free_hwq(res, &ctx->tqm_ctx.qtbl[i]);
383 	/* restore original pde level before destroy */
384 	ctx->tqm_ctx.pde.level = ctx->tqm_ctx.pde_level;
385 	bnxt_qplib_free_hwq(res, &ctx->tqm_ctx.pde);
386 	bnxt_qplib_free_stats_ctx(res->pdev, &ctx->stats);
387 }
388 
389 static int bnxt_qplib_alloc_tqm_rings(struct bnxt_qplib_res *res,
390 				      struct bnxt_qplib_ctx *ctx)
391 {
392 	struct bnxt_qplib_hwq_attr hwq_attr = {};
393 	struct bnxt_qplib_sg_info sginfo = {};
394 	struct bnxt_qplib_tqm_ctx *tqmctx;
395 	int rc = 0;
396 	int i;
397 
398 	tqmctx = &ctx->tqm_ctx;
399 
400 	sginfo.pgsize = PAGE_SIZE;
401 	sginfo.pgshft = PAGE_SHIFT;
402 	hwq_attr.sginfo = &sginfo;
403 	hwq_attr.res = res;
404 	hwq_attr.type = HWQ_TYPE_CTX;
405 	hwq_attr.depth = 512;
406 	hwq_attr.stride = sizeof(u64);
407 	/* Alloc pdl buffer */
408 	rc = bnxt_qplib_alloc_init_hwq(&tqmctx->pde, &hwq_attr);
409 	if (rc)
410 		goto out;
411 	/* Save original pdl level */
412 	tqmctx->pde_level = tqmctx->pde.level;
413 
414 	hwq_attr.stride = 1;
415 	for (i = 0; i < MAX_TQM_ALLOC_REQ; i++) {
416 		if (!tqmctx->qcount[i])
417 			continue;
418 		hwq_attr.depth = ctx->qpc_count * tqmctx->qcount[i];
419 		rc = bnxt_qplib_alloc_init_hwq(&tqmctx->qtbl[i], &hwq_attr);
420 		if (rc)
421 			goto out;
422 	}
423 out:
424 	return rc;
425 }
426 
427 static void bnxt_qplib_map_tqm_pgtbl(struct bnxt_qplib_tqm_ctx *ctx)
428 {
429 	struct bnxt_qplib_hwq *tbl;
430 	dma_addr_t *dma_ptr;
431 	__le64 **pbl_ptr, *ptr;
432 	int i, j, k;
433 	int fnz_idx = -1;
434 	int pg_count;
435 
436 	pbl_ptr = (__le64 **)ctx->pde.pbl_ptr;
437 
438 	for (i = 0, j = 0; i < MAX_TQM_ALLOC_REQ;
439 	     i++, j += MAX_TQM_ALLOC_BLK_SIZE) {
440 		tbl = &ctx->qtbl[i];
441 		if (!tbl->max_elements)
442 			continue;
443 		if (fnz_idx == -1)
444 			fnz_idx = i; /* first non-zero index */
445 		switch (tbl->level) {
446 		case PBL_LVL_2:
447 			pg_count = tbl->pbl[PBL_LVL_1].pg_count;
448 			for (k = 0; k < pg_count; k++) {
449 				ptr = &pbl_ptr[PTR_PG(j + k)][PTR_IDX(j + k)];
450 				dma_ptr = &tbl->pbl[PBL_LVL_1].pg_map_arr[k];
451 				*ptr = cpu_to_le64(*dma_ptr | PTU_PTE_VALID);
452 			}
453 			break;
454 		case PBL_LVL_1:
455 		case PBL_LVL_0:
456 		default:
457 			ptr = &pbl_ptr[PTR_PG(j)][PTR_IDX(j)];
458 			*ptr = cpu_to_le64(tbl->pbl[PBL_LVL_0].pg_map_arr[0] |
459 					   PTU_PTE_VALID);
460 			break;
461 		}
462 	}
463 	if (fnz_idx == -1)
464 		fnz_idx = 0;
465 	/* update pde level as per page table programming */
466 	ctx->pde.level = (ctx->qtbl[fnz_idx].level == PBL_LVL_2) ? PBL_LVL_2 :
467 			  ctx->qtbl[fnz_idx].level + 1;
468 }
469 
470 static int bnxt_qplib_setup_tqm_rings(struct bnxt_qplib_res *res,
471 				      struct bnxt_qplib_ctx *ctx)
472 {
473 	int rc = 0;
474 
475 	rc = bnxt_qplib_alloc_tqm_rings(res, ctx);
476 	if (rc)
477 		goto fail;
478 
479 	bnxt_qplib_map_tqm_pgtbl(&ctx->tqm_ctx);
480 fail:
481 	return rc;
482 }
483 
484 /*
485  * Routine: bnxt_qplib_alloc_ctx
486  * Description:
487  *     Context tables are memories which are used by the chip fw.
488  *     The 6 tables defined are:
489  *             QPC ctx - holds QP states
490  *             MRW ctx - holds memory region and window
491  *             SRQ ctx - holds shared RQ states
492  *             CQ ctx - holds completion queue states
493  *             TQM ctx - holds Tx Queue Manager context
494  *             TIM ctx - holds timer context
495  *     Depending on the size of the tbl requested, either a 1 Page Buffer List
496  *     or a 1-to-2-stage indirection Page Directory List + 1 PBL is used
497  *     instead.
498  *     Table might be employed as follows:
499  *             For 0      < ctx size <= 1 PAGE, 0 level of ind is used
500  *             For 1 PAGE < ctx size <= 512 entries size, 1 level of ind is used
501  *             For 512    < ctx size <= MAX, 2 levels of ind is used
502  * Returns:
503  *     0 if success, else -ERRORS
504  */
505 int bnxt_qplib_alloc_ctx(struct bnxt_qplib_res *res,
506 			 struct bnxt_qplib_ctx *ctx,
507 			 bool virt_fn, bool is_p5)
508 {
509 	struct bnxt_qplib_hwq_attr hwq_attr = {};
510 	struct bnxt_qplib_sg_info sginfo = {};
511 	int rc = 0;
512 
513 	if (virt_fn || is_p5)
514 		goto stats_alloc;
515 
516 	/* QPC Tables */
517 	sginfo.pgsize = PAGE_SIZE;
518 	sginfo.pgshft = PAGE_SHIFT;
519 	hwq_attr.sginfo = &sginfo;
520 
521 	hwq_attr.res = res;
522 	hwq_attr.depth = ctx->qpc_count;
523 	hwq_attr.stride = BNXT_QPLIB_MAX_QP_CTX_ENTRY_SIZE;
524 	hwq_attr.type = HWQ_TYPE_CTX;
525 	rc = bnxt_qplib_alloc_init_hwq(&ctx->qpc_tbl, &hwq_attr);
526 	if (rc)
527 		goto fail;
528 
529 	/* MRW Tables */
530 	hwq_attr.depth = ctx->mrw_count;
531 	hwq_attr.stride = BNXT_QPLIB_MAX_MRW_CTX_ENTRY_SIZE;
532 	rc = bnxt_qplib_alloc_init_hwq(&ctx->mrw_tbl, &hwq_attr);
533 	if (rc)
534 		goto fail;
535 
536 	/* SRQ Tables */
537 	hwq_attr.depth = ctx->srqc_count;
538 	hwq_attr.stride = BNXT_QPLIB_MAX_SRQ_CTX_ENTRY_SIZE;
539 	rc = bnxt_qplib_alloc_init_hwq(&ctx->srqc_tbl, &hwq_attr);
540 	if (rc)
541 		goto fail;
542 
543 	/* CQ Tables */
544 	hwq_attr.depth = ctx->cq_count;
545 	hwq_attr.stride = BNXT_QPLIB_MAX_CQ_CTX_ENTRY_SIZE;
546 	rc = bnxt_qplib_alloc_init_hwq(&ctx->cq_tbl, &hwq_attr);
547 	if (rc)
548 		goto fail;
549 
550 	/* TQM Buffer */
551 	rc = bnxt_qplib_setup_tqm_rings(res, ctx);
552 	if (rc)
553 		goto fail;
554 	/* TIM Buffer */
555 	ctx->tim_tbl.max_elements = ctx->qpc_count * 16;
556 	hwq_attr.depth = ctx->qpc_count * 16;
557 	hwq_attr.stride = 1;
558 	rc = bnxt_qplib_alloc_init_hwq(&ctx->tim_tbl, &hwq_attr);
559 	if (rc)
560 		goto fail;
561 stats_alloc:
562 	/* Stats */
563 	rc = bnxt_qplib_alloc_stats_ctx(res->pdev, res->cctx, &ctx->stats);
564 	if (rc)
565 		goto fail;
566 
567 	return 0;
568 
569 fail:
570 	bnxt_qplib_free_ctx(res, ctx);
571 	return rc;
572 }
573 
574 /* GUID */
575 void bnxt_qplib_get_guid(u8 *dev_addr, u8 *guid)
576 {
577 	u8 mac[ETH_ALEN];
578 
579 	/* MAC-48 to EUI-64 mapping */
580 	memcpy(mac, dev_addr, ETH_ALEN);
581 	guid[0] = mac[0] ^ 2;
582 	guid[1] = mac[1];
583 	guid[2] = mac[2];
584 	guid[3] = 0xff;
585 	guid[4] = 0xfe;
586 	guid[5] = mac[3];
587 	guid[6] = mac[4];
588 	guid[7] = mac[5];
589 }
590 
591 static void bnxt_qplib_free_sgid_tbl(struct bnxt_qplib_res *res,
592 				     struct bnxt_qplib_sgid_tbl *sgid_tbl)
593 {
594 	kfree(sgid_tbl->tbl);
595 	kfree(sgid_tbl->hw_id);
596 	kfree(sgid_tbl->ctx);
597 	kfree(sgid_tbl->vlan);
598 	sgid_tbl->tbl = NULL;
599 	sgid_tbl->hw_id = NULL;
600 	sgid_tbl->ctx = NULL;
601 	sgid_tbl->vlan = NULL;
602 	sgid_tbl->max = 0;
603 	sgid_tbl->active = 0;
604 }
605 
606 static int bnxt_qplib_alloc_sgid_tbl(struct bnxt_qplib_res *res,
607 				     struct bnxt_qplib_sgid_tbl *sgid_tbl,
608 				     u16 max)
609 {
610 	sgid_tbl->tbl = kcalloc(max, sizeof(*sgid_tbl->tbl), GFP_KERNEL);
611 	if (!sgid_tbl->tbl)
612 		return -ENOMEM;
613 
614 	sgid_tbl->hw_id = kcalloc(max, sizeof(u16), GFP_KERNEL);
615 	if (!sgid_tbl->hw_id)
616 		goto out_free1;
617 
618 	sgid_tbl->ctx = kcalloc(max, sizeof(void *), GFP_KERNEL);
619 	if (!sgid_tbl->ctx)
620 		goto out_free2;
621 
622 	sgid_tbl->vlan = kcalloc(max, sizeof(u8), GFP_KERNEL);
623 	if (!sgid_tbl->vlan)
624 		goto out_free3;
625 
626 	sgid_tbl->max = max;
627 	return 0;
628 out_free3:
629 	kfree(sgid_tbl->ctx);
630 	sgid_tbl->ctx = NULL;
631 out_free2:
632 	kfree(sgid_tbl->hw_id);
633 	sgid_tbl->hw_id = NULL;
634 out_free1:
635 	kfree(sgid_tbl->tbl);
636 	sgid_tbl->tbl = NULL;
637 	return -ENOMEM;
638 };
639 
640 static void bnxt_qplib_cleanup_sgid_tbl(struct bnxt_qplib_res *res,
641 					struct bnxt_qplib_sgid_tbl *sgid_tbl)
642 {
643 	int i;
644 
645 	for (i = 0; i < sgid_tbl->max; i++) {
646 		if (memcmp(&sgid_tbl->tbl[i], &bnxt_qplib_gid_zero,
647 			   sizeof(bnxt_qplib_gid_zero)))
648 			bnxt_qplib_del_sgid(sgid_tbl, &sgid_tbl->tbl[i].gid,
649 					    sgid_tbl->tbl[i].vlan_id, true);
650 	}
651 	memset(sgid_tbl->tbl, 0, sizeof(*sgid_tbl->tbl) * sgid_tbl->max);
652 	memset(sgid_tbl->hw_id, -1, sizeof(u16) * sgid_tbl->max);
653 	memset(sgid_tbl->vlan, 0, sizeof(u8) * sgid_tbl->max);
654 	sgid_tbl->active = 0;
655 }
656 
657 static void bnxt_qplib_init_sgid_tbl(struct bnxt_qplib_sgid_tbl *sgid_tbl,
658 				     struct net_device *netdev)
659 {
660 	u32 i;
661 
662 	for (i = 0; i < sgid_tbl->max; i++)
663 		sgid_tbl->tbl[i].vlan_id = 0xffff;
664 
665 	memset(sgid_tbl->hw_id, -1, sizeof(u16) * sgid_tbl->max);
666 }
667 
668 static void bnxt_qplib_free_pkey_tbl(struct bnxt_qplib_res *res,
669 				     struct bnxt_qplib_pkey_tbl *pkey_tbl)
670 {
671 	if (!pkey_tbl->tbl)
672 		dev_dbg(&res->pdev->dev, "PKEY tbl not present\n");
673 	else
674 		kfree(pkey_tbl->tbl);
675 
676 	pkey_tbl->tbl = NULL;
677 	pkey_tbl->max = 0;
678 	pkey_tbl->active = 0;
679 }
680 
681 static int bnxt_qplib_alloc_pkey_tbl(struct bnxt_qplib_res *res,
682 				     struct bnxt_qplib_pkey_tbl *pkey_tbl,
683 				     u16 max)
684 {
685 	pkey_tbl->tbl = kcalloc(max, sizeof(u16), GFP_KERNEL);
686 	if (!pkey_tbl->tbl)
687 		return -ENOMEM;
688 
689 	pkey_tbl->max = max;
690 	return 0;
691 };
692 
693 /* PDs */
694 int bnxt_qplib_alloc_pd(struct bnxt_qplib_pd_tbl *pdt, struct bnxt_qplib_pd *pd)
695 {
696 	u32 bit_num;
697 
698 	bit_num = find_first_bit(pdt->tbl, pdt->max);
699 	if (bit_num == pdt->max)
700 		return -ENOMEM;
701 
702 	/* Found unused PD */
703 	clear_bit(bit_num, pdt->tbl);
704 	pd->id = bit_num;
705 	return 0;
706 }
707 
708 int bnxt_qplib_dealloc_pd(struct bnxt_qplib_res *res,
709 			  struct bnxt_qplib_pd_tbl *pdt,
710 			  struct bnxt_qplib_pd *pd)
711 {
712 	if (test_and_set_bit(pd->id, pdt->tbl)) {
713 		dev_warn(&res->pdev->dev, "Freeing an unused PD? pdn = %d\n",
714 			 pd->id);
715 		return -EINVAL;
716 	}
717 	pd->id = 0;
718 	return 0;
719 }
720 
721 static void bnxt_qplib_free_pd_tbl(struct bnxt_qplib_pd_tbl *pdt)
722 {
723 	kfree(pdt->tbl);
724 	pdt->tbl = NULL;
725 	pdt->max = 0;
726 }
727 
728 static int bnxt_qplib_alloc_pd_tbl(struct bnxt_qplib_res *res,
729 				   struct bnxt_qplib_pd_tbl *pdt,
730 				   u32 max)
731 {
732 	u32 bytes;
733 
734 	bytes = max >> 3;
735 	if (!bytes)
736 		bytes = 1;
737 	pdt->tbl = kmalloc(bytes, GFP_KERNEL);
738 	if (!pdt->tbl)
739 		return -ENOMEM;
740 
741 	pdt->max = max;
742 	memset((u8 *)pdt->tbl, 0xFF, bytes);
743 
744 	return 0;
745 }
746 
747 /* DPIs */
748 int bnxt_qplib_alloc_dpi(struct bnxt_qplib_dpi_tbl *dpit,
749 			 struct bnxt_qplib_dpi     *dpi,
750 			 void                      *app)
751 {
752 	u32 bit_num;
753 
754 	bit_num = find_first_bit(dpit->tbl, dpit->max);
755 	if (bit_num == dpit->max)
756 		return -ENOMEM;
757 
758 	/* Found unused DPI */
759 	clear_bit(bit_num, dpit->tbl);
760 	dpit->app_tbl[bit_num] = app;
761 
762 	dpi->dpi = bit_num;
763 	dpi->dbr = dpit->dbr_bar_reg_iomem + (bit_num * PAGE_SIZE);
764 	dpi->umdbr = dpit->unmapped_dbr + (bit_num * PAGE_SIZE);
765 
766 	return 0;
767 }
768 
769 int bnxt_qplib_dealloc_dpi(struct bnxt_qplib_res *res,
770 			   struct bnxt_qplib_dpi_tbl *dpit,
771 			   struct bnxt_qplib_dpi     *dpi)
772 {
773 	if (dpi->dpi >= dpit->max) {
774 		dev_warn(&res->pdev->dev, "Invalid DPI? dpi = %d\n", dpi->dpi);
775 		return -EINVAL;
776 	}
777 	if (test_and_set_bit(dpi->dpi, dpit->tbl)) {
778 		dev_warn(&res->pdev->dev, "Freeing an unused DPI? dpi = %d\n",
779 			 dpi->dpi);
780 		return -EINVAL;
781 	}
782 	if (dpit->app_tbl)
783 		dpit->app_tbl[dpi->dpi] = NULL;
784 	memset(dpi, 0, sizeof(*dpi));
785 
786 	return 0;
787 }
788 
789 static void bnxt_qplib_free_dpi_tbl(struct bnxt_qplib_res     *res,
790 				    struct bnxt_qplib_dpi_tbl *dpit)
791 {
792 	kfree(dpit->tbl);
793 	kfree(dpit->app_tbl);
794 	if (dpit->dbr_bar_reg_iomem)
795 		pci_iounmap(res->pdev, dpit->dbr_bar_reg_iomem);
796 	memset(dpit, 0, sizeof(*dpit));
797 }
798 
799 static int bnxt_qplib_alloc_dpi_tbl(struct bnxt_qplib_res     *res,
800 				    struct bnxt_qplib_dpi_tbl *dpit,
801 				    u32                       dbr_offset)
802 {
803 	u32 dbr_bar_reg = RCFW_DBR_PCI_BAR_REGION;
804 	resource_size_t bar_reg_base;
805 	u32 dbr_len, bytes;
806 
807 	if (dpit->dbr_bar_reg_iomem) {
808 		dev_err(&res->pdev->dev, "DBR BAR region %d already mapped\n",
809 			dbr_bar_reg);
810 		return -EALREADY;
811 	}
812 
813 	bar_reg_base = pci_resource_start(res->pdev, dbr_bar_reg);
814 	if (!bar_reg_base) {
815 		dev_err(&res->pdev->dev, "BAR region %d resc start failed\n",
816 			dbr_bar_reg);
817 		return -ENOMEM;
818 	}
819 
820 	dbr_len = pci_resource_len(res->pdev, dbr_bar_reg) - dbr_offset;
821 	if (!dbr_len || ((dbr_len & (PAGE_SIZE - 1)) != 0)) {
822 		dev_err(&res->pdev->dev, "Invalid DBR length %d\n", dbr_len);
823 		return -ENOMEM;
824 	}
825 
826 	dpit->dbr_bar_reg_iomem = ioremap(bar_reg_base + dbr_offset,
827 						  dbr_len);
828 	if (!dpit->dbr_bar_reg_iomem) {
829 		dev_err(&res->pdev->dev,
830 			"FP: DBR BAR region %d mapping failed\n", dbr_bar_reg);
831 		return -ENOMEM;
832 	}
833 
834 	dpit->unmapped_dbr = bar_reg_base + dbr_offset;
835 	dpit->max = dbr_len / PAGE_SIZE;
836 
837 	dpit->app_tbl = kcalloc(dpit->max, sizeof(void *), GFP_KERNEL);
838 	if (!dpit->app_tbl)
839 		goto unmap_io;
840 
841 	bytes = dpit->max >> 3;
842 	if (!bytes)
843 		bytes = 1;
844 
845 	dpit->tbl = kmalloc(bytes, GFP_KERNEL);
846 	if (!dpit->tbl) {
847 		kfree(dpit->app_tbl);
848 		dpit->app_tbl = NULL;
849 		goto unmap_io;
850 	}
851 
852 	memset((u8 *)dpit->tbl, 0xFF, bytes);
853 
854 	return 0;
855 
856 unmap_io:
857 	pci_iounmap(res->pdev, dpit->dbr_bar_reg_iomem);
858 	dpit->dbr_bar_reg_iomem = NULL;
859 	return -ENOMEM;
860 }
861 
862 /* PKEYs */
863 static void bnxt_qplib_cleanup_pkey_tbl(struct bnxt_qplib_pkey_tbl *pkey_tbl)
864 {
865 	memset(pkey_tbl->tbl, 0, sizeof(u16) * pkey_tbl->max);
866 	pkey_tbl->active = 0;
867 }
868 
869 static void bnxt_qplib_init_pkey_tbl(struct bnxt_qplib_res *res,
870 				     struct bnxt_qplib_pkey_tbl *pkey_tbl)
871 {
872 	u16 pkey = 0xFFFF;
873 
874 	memset(pkey_tbl->tbl, 0, sizeof(u16) * pkey_tbl->max);
875 
876 	/* pkey default = 0xFFFF */
877 	bnxt_qplib_add_pkey(res, pkey_tbl, &pkey, false);
878 }
879 
880 /* Stats */
881 static void bnxt_qplib_free_stats_ctx(struct pci_dev *pdev,
882 				      struct bnxt_qplib_stats *stats)
883 {
884 	if (stats->dma) {
885 		dma_free_coherent(&pdev->dev, stats->size,
886 				  stats->dma, stats->dma_map);
887 	}
888 	memset(stats, 0, sizeof(*stats));
889 	stats->fw_id = -1;
890 }
891 
892 static int bnxt_qplib_alloc_stats_ctx(struct pci_dev *pdev,
893 				      struct bnxt_qplib_chip_ctx *cctx,
894 				      struct bnxt_qplib_stats *stats)
895 {
896 	memset(stats, 0, sizeof(*stats));
897 	stats->fw_id = -1;
898 	stats->size = cctx->hw_stats_size;
899 	stats->dma = dma_alloc_coherent(&pdev->dev, stats->size,
900 					&stats->dma_map, GFP_KERNEL);
901 	if (!stats->dma) {
902 		dev_err(&pdev->dev, "Stats DMA allocation failed\n");
903 		return -ENOMEM;
904 	}
905 	return 0;
906 }
907 
908 void bnxt_qplib_cleanup_res(struct bnxt_qplib_res *res)
909 {
910 	bnxt_qplib_cleanup_pkey_tbl(&res->pkey_tbl);
911 	bnxt_qplib_cleanup_sgid_tbl(res, &res->sgid_tbl);
912 }
913 
914 int bnxt_qplib_init_res(struct bnxt_qplib_res *res)
915 {
916 	bnxt_qplib_init_sgid_tbl(&res->sgid_tbl, res->netdev);
917 	bnxt_qplib_init_pkey_tbl(res, &res->pkey_tbl);
918 
919 	return 0;
920 }
921 
922 void bnxt_qplib_free_res(struct bnxt_qplib_res *res)
923 {
924 	bnxt_qplib_free_pkey_tbl(res, &res->pkey_tbl);
925 	bnxt_qplib_free_sgid_tbl(res, &res->sgid_tbl);
926 	bnxt_qplib_free_pd_tbl(&res->pd_tbl);
927 	bnxt_qplib_free_dpi_tbl(res, &res->dpi_tbl);
928 }
929 
930 int bnxt_qplib_alloc_res(struct bnxt_qplib_res *res, struct pci_dev *pdev,
931 			 struct net_device *netdev,
932 			 struct bnxt_qplib_dev_attr *dev_attr)
933 {
934 	int rc = 0;
935 
936 	res->pdev = pdev;
937 	res->netdev = netdev;
938 
939 	rc = bnxt_qplib_alloc_sgid_tbl(res, &res->sgid_tbl, dev_attr->max_sgid);
940 	if (rc)
941 		goto fail;
942 
943 	rc = bnxt_qplib_alloc_pkey_tbl(res, &res->pkey_tbl, dev_attr->max_pkey);
944 	if (rc)
945 		goto fail;
946 
947 	rc = bnxt_qplib_alloc_pd_tbl(res, &res->pd_tbl, dev_attr->max_pd);
948 	if (rc)
949 		goto fail;
950 
951 	rc = bnxt_qplib_alloc_dpi_tbl(res, &res->dpi_tbl, dev_attr->l2_db_size);
952 	if (rc)
953 		goto fail;
954 
955 	return 0;
956 fail:
957 	bnxt_qplib_free_res(res);
958 	return rc;
959 }
960 
961 int bnxt_qplib_determine_atomics(struct pci_dev *dev)
962 {
963 	int comp;
964 	u16 ctl2;
965 
966 	comp = pci_enable_atomic_ops_to_root(dev,
967 					     PCI_EXP_DEVCAP2_ATOMIC_COMP32);
968 	if (comp)
969 		return -EOPNOTSUPP;
970 	comp = pci_enable_atomic_ops_to_root(dev,
971 					     PCI_EXP_DEVCAP2_ATOMIC_COMP64);
972 	if (comp)
973 		return -EOPNOTSUPP;
974 	pcie_capability_read_word(dev, PCI_EXP_DEVCTL2, &ctl2);
975 	return !(ctl2 & PCI_EXP_DEVCTL2_ATOMIC_REQ);
976 }
977