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 && !hwq_attr->sginfo->nopte) {
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 		goto done;
238 	}
239 
240 	if (npages >= MAX_PBL_LVL_0_PGS) {
241 		if (npages > MAX_PBL_LVL_1_PGS) {
242 			u32 flag = (hwq_attr->type == HWQ_TYPE_L2_CMPL) ?
243 				    0 : PTU_PTE_VALID;
244 			/* 2 levels of indirection */
245 			npbl = npages >> MAX_PBL_LVL_1_PGS_SHIFT;
246 			if (npages % BIT(MAX_PBL_LVL_1_PGS_SHIFT))
247 				npbl++;
248 			npde = npbl >> MAX_PDL_LVL_SHIFT;
249 			if (npbl % BIT(MAX_PDL_LVL_SHIFT))
250 				npde++;
251 			/* Alloc PDE pages */
252 			sginfo.pgsize = npde * pg_size;
253 			sginfo.npages = 1;
254 			rc = __alloc_pbl(res, &hwq->pbl[PBL_LVL_0], &sginfo);
255 
256 			/* Alloc PBL pages */
257 			sginfo.npages = npbl;
258 			sginfo.pgsize = PAGE_SIZE;
259 			rc = __alloc_pbl(res, &hwq->pbl[PBL_LVL_1], &sginfo);
260 			if (rc)
261 				goto fail;
262 			/* Fill PDL with PBL page pointers */
263 			dst_virt_ptr =
264 				(dma_addr_t **)hwq->pbl[PBL_LVL_0].pg_arr;
265 			src_phys_ptr = hwq->pbl[PBL_LVL_1].pg_map_arr;
266 			if (hwq_attr->type == HWQ_TYPE_MR) {
267 			/* For MR it is expected that we supply only 1 contigous
268 			 * page i.e only 1 entry in the PDL that will contain
269 			 * all the PBLs for the user supplied memory region
270 			 */
271 				for (i = 0; i < hwq->pbl[PBL_LVL_1].pg_count;
272 				     i++)
273 					dst_virt_ptr[0][i] = src_phys_ptr[i] |
274 						flag;
275 			} else {
276 				for (i = 0; i < hwq->pbl[PBL_LVL_1].pg_count;
277 				     i++)
278 					dst_virt_ptr[PTR_PG(i)][PTR_IDX(i)] =
279 						src_phys_ptr[i] |
280 						PTU_PDE_VALID;
281 			}
282 			/* Alloc or init PTEs */
283 			rc = __alloc_pbl(res, &hwq->pbl[PBL_LVL_2],
284 					 hwq_attr->sginfo);
285 			if (rc)
286 				goto fail;
287 			hwq->level = PBL_LVL_2;
288 			if (hwq_attr->sginfo->nopte)
289 				goto done;
290 			/* Fill PBLs with PTE pointers */
291 			dst_virt_ptr =
292 				(dma_addr_t **)hwq->pbl[PBL_LVL_1].pg_arr;
293 			src_phys_ptr = hwq->pbl[PBL_LVL_2].pg_map_arr;
294 			for (i = 0; i < hwq->pbl[PBL_LVL_2].pg_count; i++) {
295 				dst_virt_ptr[PTR_PG(i)][PTR_IDX(i)] =
296 					src_phys_ptr[i] | PTU_PTE_VALID;
297 			}
298 			if (hwq_attr->type == HWQ_TYPE_QUEUE) {
299 				/* Find the last pg of the size */
300 				i = hwq->pbl[PBL_LVL_2].pg_count;
301 				dst_virt_ptr[PTR_PG(i - 1)][PTR_IDX(i - 1)] |=
302 								  PTU_PTE_LAST;
303 				if (i > 1)
304 					dst_virt_ptr[PTR_PG(i - 2)]
305 						    [PTR_IDX(i - 2)] |=
306 						    PTU_PTE_NEXT_TO_LAST;
307 			}
308 		} else { /* pages < 512 npbl = 1, npde = 0 */
309 			u32 flag = (hwq_attr->type == HWQ_TYPE_L2_CMPL) ?
310 				    0 : PTU_PTE_VALID;
311 
312 			/* 1 level of indirection */
313 			npbl = npages >> MAX_PBL_LVL_1_PGS_SHIFT;
314 			if (npages % BIT(MAX_PBL_LVL_1_PGS_SHIFT))
315 				npbl++;
316 			sginfo.npages = npbl;
317 			sginfo.pgsize = PAGE_SIZE;
318 			/* Alloc PBL page */
319 			rc = __alloc_pbl(res, &hwq->pbl[PBL_LVL_0], &sginfo);
320 			if (rc)
321 				goto fail;
322 			/* Alloc or init  PTEs */
323 			rc = __alloc_pbl(res, &hwq->pbl[PBL_LVL_1],
324 					 hwq_attr->sginfo);
325 			if (rc)
326 				goto fail;
327 			hwq->level = PBL_LVL_1;
328 			if (hwq_attr->sginfo->nopte)
329 				goto done;
330 			/* Fill PBL with PTE pointers */
331 			dst_virt_ptr =
332 				(dma_addr_t **)hwq->pbl[PBL_LVL_0].pg_arr;
333 			src_phys_ptr = hwq->pbl[PBL_LVL_1].pg_map_arr;
334 			for (i = 0; i < hwq->pbl[PBL_LVL_1].pg_count; i++)
335 				dst_virt_ptr[PTR_PG(i)][PTR_IDX(i)] =
336 					src_phys_ptr[i] | flag;
337 			if (hwq_attr->type == HWQ_TYPE_QUEUE) {
338 				/* Find the last pg of the size */
339 				i = hwq->pbl[PBL_LVL_1].pg_count;
340 				dst_virt_ptr[PTR_PG(i - 1)][PTR_IDX(i - 1)] |=
341 								  PTU_PTE_LAST;
342 				if (i > 1)
343 					dst_virt_ptr[PTR_PG(i - 2)]
344 						    [PTR_IDX(i - 2)] |=
345 						    PTU_PTE_NEXT_TO_LAST;
346 			}
347 		}
348 	}
349 done:
350 	hwq->prod = 0;
351 	hwq->cons = 0;
352 	hwq->pdev = pdev;
353 	hwq->depth = hwq_attr->depth;
354 	hwq->max_elements = depth;
355 	hwq->element_size = stride;
356 	hwq->qe_ppg = pg_size / stride;
357 	/* For direct access to the elements */
358 	lvl = hwq->level;
359 	if (hwq_attr->sginfo->nopte && hwq->level)
360 		lvl = hwq->level - 1;
361 	hwq->pbl_ptr = hwq->pbl[lvl].pg_arr;
362 	hwq->pbl_dma_ptr = hwq->pbl[lvl].pg_map_arr;
363 	spin_lock_init(&hwq->lock);
364 
365 	return 0;
366 fail:
367 	bnxt_qplib_free_hwq(res, hwq);
368 	return -ENOMEM;
369 }
370 
371 /* Context Tables */
372 void bnxt_qplib_free_ctx(struct bnxt_qplib_res *res,
373 			 struct bnxt_qplib_ctx *ctx)
374 {
375 	int i;
376 
377 	bnxt_qplib_free_hwq(res, &ctx->qpc_tbl);
378 	bnxt_qplib_free_hwq(res, &ctx->mrw_tbl);
379 	bnxt_qplib_free_hwq(res, &ctx->srqc_tbl);
380 	bnxt_qplib_free_hwq(res, &ctx->cq_tbl);
381 	bnxt_qplib_free_hwq(res, &ctx->tim_tbl);
382 	for (i = 0; i < MAX_TQM_ALLOC_REQ; i++)
383 		bnxt_qplib_free_hwq(res, &ctx->tqm_ctx.qtbl[i]);
384 	/* restore original pde level before destroy */
385 	ctx->tqm_ctx.pde.level = ctx->tqm_ctx.pde_level;
386 	bnxt_qplib_free_hwq(res, &ctx->tqm_ctx.pde);
387 	bnxt_qplib_free_stats_ctx(res->pdev, &ctx->stats);
388 }
389 
390 static int bnxt_qplib_alloc_tqm_rings(struct bnxt_qplib_res *res,
391 				      struct bnxt_qplib_ctx *ctx)
392 {
393 	struct bnxt_qplib_hwq_attr hwq_attr = {};
394 	struct bnxt_qplib_sg_info sginfo = {};
395 	struct bnxt_qplib_tqm_ctx *tqmctx;
396 	int rc = 0;
397 	int i;
398 
399 	tqmctx = &ctx->tqm_ctx;
400 
401 	sginfo.pgsize = PAGE_SIZE;
402 	sginfo.pgshft = PAGE_SHIFT;
403 	hwq_attr.sginfo = &sginfo;
404 	hwq_attr.res = res;
405 	hwq_attr.type = HWQ_TYPE_CTX;
406 	hwq_attr.depth = 512;
407 	hwq_attr.stride = sizeof(u64);
408 	/* Alloc pdl buffer */
409 	rc = bnxt_qplib_alloc_init_hwq(&tqmctx->pde, &hwq_attr);
410 	if (rc)
411 		goto out;
412 	/* Save original pdl level */
413 	tqmctx->pde_level = tqmctx->pde.level;
414 
415 	hwq_attr.stride = 1;
416 	for (i = 0; i < MAX_TQM_ALLOC_REQ; i++) {
417 		if (!tqmctx->qcount[i])
418 			continue;
419 		hwq_attr.depth = ctx->qpc_count * tqmctx->qcount[i];
420 		rc = bnxt_qplib_alloc_init_hwq(&tqmctx->qtbl[i], &hwq_attr);
421 		if (rc)
422 			goto out;
423 	}
424 out:
425 	return rc;
426 }
427 
428 static void bnxt_qplib_map_tqm_pgtbl(struct bnxt_qplib_tqm_ctx *ctx)
429 {
430 	struct bnxt_qplib_hwq *tbl;
431 	dma_addr_t *dma_ptr;
432 	__le64 **pbl_ptr, *ptr;
433 	int i, j, k;
434 	int fnz_idx = -1;
435 	int pg_count;
436 
437 	pbl_ptr = (__le64 **)ctx->pde.pbl_ptr;
438 
439 	for (i = 0, j = 0; i < MAX_TQM_ALLOC_REQ;
440 	     i++, j += MAX_TQM_ALLOC_BLK_SIZE) {
441 		tbl = &ctx->qtbl[i];
442 		if (!tbl->max_elements)
443 			continue;
444 		if (fnz_idx == -1)
445 			fnz_idx = i; /* first non-zero index */
446 		switch (tbl->level) {
447 		case PBL_LVL_2:
448 			pg_count = tbl->pbl[PBL_LVL_1].pg_count;
449 			for (k = 0; k < pg_count; k++) {
450 				ptr = &pbl_ptr[PTR_PG(j + k)][PTR_IDX(j + k)];
451 				dma_ptr = &tbl->pbl[PBL_LVL_1].pg_map_arr[k];
452 				*ptr = cpu_to_le64(*dma_ptr | PTU_PTE_VALID);
453 			}
454 			break;
455 		case PBL_LVL_1:
456 		case PBL_LVL_0:
457 		default:
458 			ptr = &pbl_ptr[PTR_PG(j)][PTR_IDX(j)];
459 			*ptr = cpu_to_le64(tbl->pbl[PBL_LVL_0].pg_map_arr[0] |
460 					   PTU_PTE_VALID);
461 			break;
462 		}
463 	}
464 	if (fnz_idx == -1)
465 		fnz_idx = 0;
466 	/* update pde level as per page table programming */
467 	ctx->pde.level = (ctx->qtbl[fnz_idx].level == PBL_LVL_2) ? PBL_LVL_2 :
468 			  ctx->qtbl[fnz_idx].level + 1;
469 }
470 
471 static int bnxt_qplib_setup_tqm_rings(struct bnxt_qplib_res *res,
472 				      struct bnxt_qplib_ctx *ctx)
473 {
474 	int rc = 0;
475 
476 	rc = bnxt_qplib_alloc_tqm_rings(res, ctx);
477 	if (rc)
478 		goto fail;
479 
480 	bnxt_qplib_map_tqm_pgtbl(&ctx->tqm_ctx);
481 fail:
482 	return rc;
483 }
484 
485 /*
486  * Routine: bnxt_qplib_alloc_ctx
487  * Description:
488  *     Context tables are memories which are used by the chip fw.
489  *     The 6 tables defined are:
490  *             QPC ctx - holds QP states
491  *             MRW ctx - holds memory region and window
492  *             SRQ ctx - holds shared RQ states
493  *             CQ ctx - holds completion queue states
494  *             TQM ctx - holds Tx Queue Manager context
495  *             TIM ctx - holds timer context
496  *     Depending on the size of the tbl requested, either a 1 Page Buffer List
497  *     or a 1-to-2-stage indirection Page Directory List + 1 PBL is used
498  *     instead.
499  *     Table might be employed as follows:
500  *             For 0      < ctx size <= 1 PAGE, 0 level of ind is used
501  *             For 1 PAGE < ctx size <= 512 entries size, 1 level of ind is used
502  *             For 512    < ctx size <= MAX, 2 levels of ind is used
503  * Returns:
504  *     0 if success, else -ERRORS
505  */
506 int bnxt_qplib_alloc_ctx(struct bnxt_qplib_res *res,
507 			 struct bnxt_qplib_ctx *ctx,
508 			 bool virt_fn, bool is_p5)
509 {
510 	struct bnxt_qplib_hwq_attr hwq_attr = {};
511 	struct bnxt_qplib_sg_info sginfo = {};
512 	int rc = 0;
513 
514 	if (virt_fn || is_p5)
515 		goto stats_alloc;
516 
517 	/* QPC Tables */
518 	sginfo.pgsize = PAGE_SIZE;
519 	sginfo.pgshft = PAGE_SHIFT;
520 	hwq_attr.sginfo = &sginfo;
521 
522 	hwq_attr.res = res;
523 	hwq_attr.depth = ctx->qpc_count;
524 	hwq_attr.stride = BNXT_QPLIB_MAX_QP_CTX_ENTRY_SIZE;
525 	hwq_attr.type = HWQ_TYPE_CTX;
526 	rc = bnxt_qplib_alloc_init_hwq(&ctx->qpc_tbl, &hwq_attr);
527 	if (rc)
528 		goto fail;
529 
530 	/* MRW Tables */
531 	hwq_attr.depth = ctx->mrw_count;
532 	hwq_attr.stride = BNXT_QPLIB_MAX_MRW_CTX_ENTRY_SIZE;
533 	rc = bnxt_qplib_alloc_init_hwq(&ctx->mrw_tbl, &hwq_attr);
534 	if (rc)
535 		goto fail;
536 
537 	/* SRQ Tables */
538 	hwq_attr.depth = ctx->srqc_count;
539 	hwq_attr.stride = BNXT_QPLIB_MAX_SRQ_CTX_ENTRY_SIZE;
540 	rc = bnxt_qplib_alloc_init_hwq(&ctx->srqc_tbl, &hwq_attr);
541 	if (rc)
542 		goto fail;
543 
544 	/* CQ Tables */
545 	hwq_attr.depth = ctx->cq_count;
546 	hwq_attr.stride = BNXT_QPLIB_MAX_CQ_CTX_ENTRY_SIZE;
547 	rc = bnxt_qplib_alloc_init_hwq(&ctx->cq_tbl, &hwq_attr);
548 	if (rc)
549 		goto fail;
550 
551 	/* TQM Buffer */
552 	rc = bnxt_qplib_setup_tqm_rings(res, ctx);
553 	if (rc)
554 		goto fail;
555 	/* TIM Buffer */
556 	ctx->tim_tbl.max_elements = ctx->qpc_count * 16;
557 	hwq_attr.depth = ctx->qpc_count * 16;
558 	hwq_attr.stride = 1;
559 	rc = bnxt_qplib_alloc_init_hwq(&ctx->tim_tbl, &hwq_attr);
560 	if (rc)
561 		goto fail;
562 stats_alloc:
563 	/* Stats */
564 	rc = bnxt_qplib_alloc_stats_ctx(res->pdev, res->cctx, &ctx->stats);
565 	if (rc)
566 		goto fail;
567 
568 	return 0;
569 
570 fail:
571 	bnxt_qplib_free_ctx(res, ctx);
572 	return rc;
573 }
574 
575 static void bnxt_qplib_free_sgid_tbl(struct bnxt_qplib_res *res,
576 				     struct bnxt_qplib_sgid_tbl *sgid_tbl)
577 {
578 	kfree(sgid_tbl->tbl);
579 	kfree(sgid_tbl->hw_id);
580 	kfree(sgid_tbl->ctx);
581 	kfree(sgid_tbl->vlan);
582 	sgid_tbl->tbl = NULL;
583 	sgid_tbl->hw_id = NULL;
584 	sgid_tbl->ctx = NULL;
585 	sgid_tbl->vlan = NULL;
586 	sgid_tbl->max = 0;
587 	sgid_tbl->active = 0;
588 }
589 
590 static int bnxt_qplib_alloc_sgid_tbl(struct bnxt_qplib_res *res,
591 				     struct bnxt_qplib_sgid_tbl *sgid_tbl,
592 				     u16 max)
593 {
594 	sgid_tbl->tbl = kcalloc(max, sizeof(*sgid_tbl->tbl), GFP_KERNEL);
595 	if (!sgid_tbl->tbl)
596 		return -ENOMEM;
597 
598 	sgid_tbl->hw_id = kcalloc(max, sizeof(u16), GFP_KERNEL);
599 	if (!sgid_tbl->hw_id)
600 		goto out_free1;
601 
602 	sgid_tbl->ctx = kcalloc(max, sizeof(void *), GFP_KERNEL);
603 	if (!sgid_tbl->ctx)
604 		goto out_free2;
605 
606 	sgid_tbl->vlan = kcalloc(max, sizeof(u8), GFP_KERNEL);
607 	if (!sgid_tbl->vlan)
608 		goto out_free3;
609 
610 	sgid_tbl->max = max;
611 	return 0;
612 out_free3:
613 	kfree(sgid_tbl->ctx);
614 	sgid_tbl->ctx = NULL;
615 out_free2:
616 	kfree(sgid_tbl->hw_id);
617 	sgid_tbl->hw_id = NULL;
618 out_free1:
619 	kfree(sgid_tbl->tbl);
620 	sgid_tbl->tbl = NULL;
621 	return -ENOMEM;
622 };
623 
624 static void bnxt_qplib_cleanup_sgid_tbl(struct bnxt_qplib_res *res,
625 					struct bnxt_qplib_sgid_tbl *sgid_tbl)
626 {
627 	int i;
628 
629 	for (i = 0; i < sgid_tbl->max; i++) {
630 		if (memcmp(&sgid_tbl->tbl[i], &bnxt_qplib_gid_zero,
631 			   sizeof(bnxt_qplib_gid_zero)))
632 			bnxt_qplib_del_sgid(sgid_tbl, &sgid_tbl->tbl[i].gid,
633 					    sgid_tbl->tbl[i].vlan_id, true);
634 	}
635 	memset(sgid_tbl->tbl, 0, sizeof(*sgid_tbl->tbl) * sgid_tbl->max);
636 	memset(sgid_tbl->hw_id, -1, sizeof(u16) * sgid_tbl->max);
637 	memset(sgid_tbl->vlan, 0, sizeof(u8) * sgid_tbl->max);
638 	sgid_tbl->active = 0;
639 }
640 
641 static void bnxt_qplib_init_sgid_tbl(struct bnxt_qplib_sgid_tbl *sgid_tbl,
642 				     struct net_device *netdev)
643 {
644 	u32 i;
645 
646 	for (i = 0; i < sgid_tbl->max; i++)
647 		sgid_tbl->tbl[i].vlan_id = 0xffff;
648 
649 	memset(sgid_tbl->hw_id, -1, sizeof(u16) * sgid_tbl->max);
650 }
651 
652 static void bnxt_qplib_free_pkey_tbl(struct bnxt_qplib_res *res,
653 				     struct bnxt_qplib_pkey_tbl *pkey_tbl)
654 {
655 	if (!pkey_tbl->tbl)
656 		dev_dbg(&res->pdev->dev, "PKEY tbl not present\n");
657 	else
658 		kfree(pkey_tbl->tbl);
659 
660 	pkey_tbl->tbl = NULL;
661 	pkey_tbl->max = 0;
662 	pkey_tbl->active = 0;
663 }
664 
665 static int bnxt_qplib_alloc_pkey_tbl(struct bnxt_qplib_res *res,
666 				     struct bnxt_qplib_pkey_tbl *pkey_tbl,
667 				     u16 max)
668 {
669 	pkey_tbl->tbl = kcalloc(max, sizeof(u16), GFP_KERNEL);
670 	if (!pkey_tbl->tbl)
671 		return -ENOMEM;
672 
673 	pkey_tbl->max = max;
674 	return 0;
675 };
676 
677 /* PDs */
678 int bnxt_qplib_alloc_pd(struct bnxt_qplib_pd_tbl *pdt, struct bnxt_qplib_pd *pd)
679 {
680 	u32 bit_num;
681 
682 	bit_num = find_first_bit(pdt->tbl, pdt->max);
683 	if (bit_num == pdt->max)
684 		return -ENOMEM;
685 
686 	/* Found unused PD */
687 	clear_bit(bit_num, pdt->tbl);
688 	pd->id = bit_num;
689 	return 0;
690 }
691 
692 int bnxt_qplib_dealloc_pd(struct bnxt_qplib_res *res,
693 			  struct bnxt_qplib_pd_tbl *pdt,
694 			  struct bnxt_qplib_pd *pd)
695 {
696 	if (test_and_set_bit(pd->id, pdt->tbl)) {
697 		dev_warn(&res->pdev->dev, "Freeing an unused PD? pdn = %d\n",
698 			 pd->id);
699 		return -EINVAL;
700 	}
701 	pd->id = 0;
702 	return 0;
703 }
704 
705 static void bnxt_qplib_free_pd_tbl(struct bnxt_qplib_pd_tbl *pdt)
706 {
707 	kfree(pdt->tbl);
708 	pdt->tbl = NULL;
709 	pdt->max = 0;
710 }
711 
712 static int bnxt_qplib_alloc_pd_tbl(struct bnxt_qplib_res *res,
713 				   struct bnxt_qplib_pd_tbl *pdt,
714 				   u32 max)
715 {
716 	u32 bytes;
717 
718 	bytes = max >> 3;
719 	if (!bytes)
720 		bytes = 1;
721 	pdt->tbl = kmalloc(bytes, GFP_KERNEL);
722 	if (!pdt->tbl)
723 		return -ENOMEM;
724 
725 	pdt->max = max;
726 	memset((u8 *)pdt->tbl, 0xFF, bytes);
727 
728 	return 0;
729 }
730 
731 /* DPIs */
732 int bnxt_qplib_alloc_dpi(struct bnxt_qplib_dpi_tbl *dpit,
733 			 struct bnxt_qplib_dpi     *dpi,
734 			 void                      *app)
735 {
736 	u32 bit_num;
737 
738 	bit_num = find_first_bit(dpit->tbl, dpit->max);
739 	if (bit_num == dpit->max)
740 		return -ENOMEM;
741 
742 	/* Found unused DPI */
743 	clear_bit(bit_num, dpit->tbl);
744 	dpit->app_tbl[bit_num] = app;
745 
746 	dpi->dpi = bit_num;
747 	dpi->dbr = dpit->dbr_bar_reg_iomem + (bit_num * PAGE_SIZE);
748 	dpi->umdbr = dpit->unmapped_dbr + (bit_num * PAGE_SIZE);
749 
750 	return 0;
751 }
752 
753 int bnxt_qplib_dealloc_dpi(struct bnxt_qplib_res *res,
754 			   struct bnxt_qplib_dpi_tbl *dpit,
755 			   struct bnxt_qplib_dpi     *dpi)
756 {
757 	if (dpi->dpi >= dpit->max) {
758 		dev_warn(&res->pdev->dev, "Invalid DPI? dpi = %d\n", dpi->dpi);
759 		return -EINVAL;
760 	}
761 	if (test_and_set_bit(dpi->dpi, dpit->tbl)) {
762 		dev_warn(&res->pdev->dev, "Freeing an unused DPI? dpi = %d\n",
763 			 dpi->dpi);
764 		return -EINVAL;
765 	}
766 	if (dpit->app_tbl)
767 		dpit->app_tbl[dpi->dpi] = NULL;
768 	memset(dpi, 0, sizeof(*dpi));
769 
770 	return 0;
771 }
772 
773 static void bnxt_qplib_free_dpi_tbl(struct bnxt_qplib_res     *res,
774 				    struct bnxt_qplib_dpi_tbl *dpit)
775 {
776 	kfree(dpit->tbl);
777 	kfree(dpit->app_tbl);
778 	if (dpit->dbr_bar_reg_iomem)
779 		pci_iounmap(res->pdev, dpit->dbr_bar_reg_iomem);
780 	memset(dpit, 0, sizeof(*dpit));
781 }
782 
783 static int bnxt_qplib_alloc_dpi_tbl(struct bnxt_qplib_res     *res,
784 				    struct bnxt_qplib_dpi_tbl *dpit,
785 				    u32                       dbr_offset)
786 {
787 	u32 dbr_bar_reg = RCFW_DBR_PCI_BAR_REGION;
788 	resource_size_t bar_reg_base;
789 	u32 dbr_len, bytes;
790 
791 	if (dpit->dbr_bar_reg_iomem) {
792 		dev_err(&res->pdev->dev, "DBR BAR region %d already mapped\n",
793 			dbr_bar_reg);
794 		return -EALREADY;
795 	}
796 
797 	bar_reg_base = pci_resource_start(res->pdev, dbr_bar_reg);
798 	if (!bar_reg_base) {
799 		dev_err(&res->pdev->dev, "BAR region %d resc start failed\n",
800 			dbr_bar_reg);
801 		return -ENOMEM;
802 	}
803 
804 	dbr_len = pci_resource_len(res->pdev, dbr_bar_reg) - dbr_offset;
805 	if (!dbr_len || ((dbr_len & (PAGE_SIZE - 1)) != 0)) {
806 		dev_err(&res->pdev->dev, "Invalid DBR length %d\n", dbr_len);
807 		return -ENOMEM;
808 	}
809 
810 	dpit->dbr_bar_reg_iomem = ioremap(bar_reg_base + dbr_offset,
811 						  dbr_len);
812 	if (!dpit->dbr_bar_reg_iomem) {
813 		dev_err(&res->pdev->dev,
814 			"FP: DBR BAR region %d mapping failed\n", dbr_bar_reg);
815 		return -ENOMEM;
816 	}
817 
818 	dpit->unmapped_dbr = bar_reg_base + dbr_offset;
819 	dpit->max = dbr_len / PAGE_SIZE;
820 
821 	dpit->app_tbl = kcalloc(dpit->max, sizeof(void *), GFP_KERNEL);
822 	if (!dpit->app_tbl)
823 		goto unmap_io;
824 
825 	bytes = dpit->max >> 3;
826 	if (!bytes)
827 		bytes = 1;
828 
829 	dpit->tbl = kmalloc(bytes, GFP_KERNEL);
830 	if (!dpit->tbl) {
831 		kfree(dpit->app_tbl);
832 		dpit->app_tbl = NULL;
833 		goto unmap_io;
834 	}
835 
836 	memset((u8 *)dpit->tbl, 0xFF, bytes);
837 
838 	return 0;
839 
840 unmap_io:
841 	pci_iounmap(res->pdev, dpit->dbr_bar_reg_iomem);
842 	dpit->dbr_bar_reg_iomem = NULL;
843 	return -ENOMEM;
844 }
845 
846 /* PKEYs */
847 static void bnxt_qplib_cleanup_pkey_tbl(struct bnxt_qplib_pkey_tbl *pkey_tbl)
848 {
849 	memset(pkey_tbl->tbl, 0, sizeof(u16) * pkey_tbl->max);
850 	pkey_tbl->active = 0;
851 }
852 
853 static void bnxt_qplib_init_pkey_tbl(struct bnxt_qplib_res *res,
854 				     struct bnxt_qplib_pkey_tbl *pkey_tbl)
855 {
856 	u16 pkey = 0xFFFF;
857 
858 	memset(pkey_tbl->tbl, 0, sizeof(u16) * pkey_tbl->max);
859 
860 	/* pkey default = 0xFFFF */
861 	bnxt_qplib_add_pkey(res, pkey_tbl, &pkey, false);
862 }
863 
864 /* Stats */
865 static void bnxt_qplib_free_stats_ctx(struct pci_dev *pdev,
866 				      struct bnxt_qplib_stats *stats)
867 {
868 	if (stats->dma) {
869 		dma_free_coherent(&pdev->dev, stats->size,
870 				  stats->dma, stats->dma_map);
871 	}
872 	memset(stats, 0, sizeof(*stats));
873 	stats->fw_id = -1;
874 }
875 
876 static int bnxt_qplib_alloc_stats_ctx(struct pci_dev *pdev,
877 				      struct bnxt_qplib_chip_ctx *cctx,
878 				      struct bnxt_qplib_stats *stats)
879 {
880 	memset(stats, 0, sizeof(*stats));
881 	stats->fw_id = -1;
882 	stats->size = cctx->hw_stats_size;
883 	stats->dma = dma_alloc_coherent(&pdev->dev, stats->size,
884 					&stats->dma_map, GFP_KERNEL);
885 	if (!stats->dma) {
886 		dev_err(&pdev->dev, "Stats DMA allocation failed\n");
887 		return -ENOMEM;
888 	}
889 	return 0;
890 }
891 
892 void bnxt_qplib_cleanup_res(struct bnxt_qplib_res *res)
893 {
894 	bnxt_qplib_cleanup_pkey_tbl(&res->pkey_tbl);
895 	bnxt_qplib_cleanup_sgid_tbl(res, &res->sgid_tbl);
896 }
897 
898 int bnxt_qplib_init_res(struct bnxt_qplib_res *res)
899 {
900 	bnxt_qplib_init_sgid_tbl(&res->sgid_tbl, res->netdev);
901 	bnxt_qplib_init_pkey_tbl(res, &res->pkey_tbl);
902 
903 	return 0;
904 }
905 
906 void bnxt_qplib_free_res(struct bnxt_qplib_res *res)
907 {
908 	bnxt_qplib_free_pkey_tbl(res, &res->pkey_tbl);
909 	bnxt_qplib_free_sgid_tbl(res, &res->sgid_tbl);
910 	bnxt_qplib_free_pd_tbl(&res->pd_tbl);
911 	bnxt_qplib_free_dpi_tbl(res, &res->dpi_tbl);
912 }
913 
914 int bnxt_qplib_alloc_res(struct bnxt_qplib_res *res, struct pci_dev *pdev,
915 			 struct net_device *netdev,
916 			 struct bnxt_qplib_dev_attr *dev_attr)
917 {
918 	int rc = 0;
919 
920 	res->pdev = pdev;
921 	res->netdev = netdev;
922 
923 	rc = bnxt_qplib_alloc_sgid_tbl(res, &res->sgid_tbl, dev_attr->max_sgid);
924 	if (rc)
925 		goto fail;
926 
927 	rc = bnxt_qplib_alloc_pkey_tbl(res, &res->pkey_tbl, dev_attr->max_pkey);
928 	if (rc)
929 		goto fail;
930 
931 	rc = bnxt_qplib_alloc_pd_tbl(res, &res->pd_tbl, dev_attr->max_pd);
932 	if (rc)
933 		goto fail;
934 
935 	rc = bnxt_qplib_alloc_dpi_tbl(res, &res->dpi_tbl, dev_attr->l2_db_size);
936 	if (rc)
937 		goto fail;
938 
939 	return 0;
940 fail:
941 	bnxt_qplib_free_res(res);
942 	return rc;
943 }
944 
945 int bnxt_qplib_determine_atomics(struct pci_dev *dev)
946 {
947 	int comp;
948 	u16 ctl2;
949 
950 	comp = pci_enable_atomic_ops_to_root(dev,
951 					     PCI_EXP_DEVCAP2_ATOMIC_COMP32);
952 	if (comp)
953 		return -EOPNOTSUPP;
954 	comp = pci_enable_atomic_ops_to_root(dev,
955 					     PCI_EXP_DEVCAP2_ATOMIC_COMP64);
956 	if (comp)
957 		return -EOPNOTSUPP;
958 	pcie_capability_read_word(dev, PCI_EXP_DEVCTL2, &ctl2);
959 	return !(ctl2 & PCI_EXP_DEVCTL2_ATOMIC_REQ);
960 }
961