xref: /linux/drivers/infiniband/hw/bnxt_re/qplib_res.c (revision 8d765af51a099884bab37a51e211c7047f67f1f3)
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 /* PBL */
57 static void __free_pbl(struct bnxt_qplib_res *res, struct bnxt_qplib_pbl *pbl,
58 		       bool is_umem)
59 {
60 	struct pci_dev *pdev = res->pdev;
61 	int i;
62 
63 	if (!is_umem) {
64 		for (i = 0; i < pbl->pg_count; i++) {
65 			if (pbl->pg_arr[i])
66 				dma_free_coherent(&pdev->dev, pbl->pg_size,
67 						  (void *)((unsigned long)
68 						   pbl->pg_arr[i] &
69 						  PAGE_MASK),
70 						  pbl->pg_map_arr[i]);
71 			else
72 				dev_warn(&pdev->dev,
73 					 "PBL free pg_arr[%d] empty?!\n", i);
74 			pbl->pg_arr[i] = NULL;
75 		}
76 	}
77 	vfree(pbl->pg_arr);
78 	pbl->pg_arr = NULL;
79 	vfree(pbl->pg_map_arr);
80 	pbl->pg_map_arr = NULL;
81 	pbl->pg_count = 0;
82 	pbl->pg_size = 0;
83 }
84 
85 static void bnxt_qplib_fill_user_dma_pages(struct bnxt_qplib_pbl *pbl,
86 					   struct bnxt_qplib_sg_info *sginfo)
87 {
88 	struct ib_block_iter biter;
89 	int i = 0;
90 
91 	rdma_umem_for_each_dma_block(sginfo->umem, &biter, sginfo->pgsize) {
92 		pbl->pg_map_arr[i] = rdma_block_iter_dma_address(&biter);
93 		pbl->pg_arr[i] = NULL;
94 		pbl->pg_count++;
95 		i++;
96 	}
97 }
98 
99 static int __alloc_pbl(struct bnxt_qplib_res *res,
100 		       struct bnxt_qplib_pbl *pbl,
101 		       struct bnxt_qplib_sg_info *sginfo)
102 {
103 	struct pci_dev *pdev = res->pdev;
104 	bool is_umem = false;
105 	u32 pages;
106 	int i;
107 
108 	if (sginfo->nopte)
109 		return 0;
110 	if (sginfo->umem)
111 		pages = ib_umem_num_dma_blocks(sginfo->umem, sginfo->pgsize);
112 	else
113 		pages = sginfo->npages;
114 	/* page ptr arrays */
115 	pbl->pg_arr = vmalloc_array(pages, sizeof(void *));
116 	if (!pbl->pg_arr)
117 		return -ENOMEM;
118 
119 	pbl->pg_map_arr = vmalloc_array(pages, sizeof(dma_addr_t));
120 	if (!pbl->pg_map_arr) {
121 		vfree(pbl->pg_arr);
122 		pbl->pg_arr = NULL;
123 		return -ENOMEM;
124 	}
125 	pbl->pg_count = 0;
126 	pbl->pg_size = sginfo->pgsize;
127 
128 	if (!sginfo->umem) {
129 		for (i = 0; i < pages; i++) {
130 			pbl->pg_arr[i] = dma_alloc_coherent(&pdev->dev,
131 							    pbl->pg_size,
132 							    &pbl->pg_map_arr[i],
133 							    GFP_KERNEL);
134 			if (!pbl->pg_arr[i])
135 				goto fail;
136 			pbl->pg_count++;
137 		}
138 	} else {
139 		is_umem = true;
140 		bnxt_qplib_fill_user_dma_pages(pbl, sginfo);
141 	}
142 
143 	return 0;
144 fail:
145 	__free_pbl(res, pbl, is_umem);
146 	return -ENOMEM;
147 }
148 
149 /* HWQ */
150 void bnxt_qplib_free_hwq(struct bnxt_qplib_res *res,
151 			 struct bnxt_qplib_hwq *hwq)
152 {
153 	int i;
154 
155 	if (!hwq->max_elements)
156 		return;
157 	if (hwq->level >= PBL_LVL_MAX)
158 		return;
159 
160 	for (i = 0; i < hwq->level + 1; i++) {
161 		if (i == hwq->level)
162 			__free_pbl(res, &hwq->pbl[i], hwq->is_user);
163 		else
164 			__free_pbl(res, &hwq->pbl[i], false);
165 	}
166 
167 	hwq->level = PBL_LVL_MAX;
168 	hwq->max_elements = 0;
169 	hwq->element_size = 0;
170 	hwq->prod = 0;
171 	hwq->cons = 0;
172 	hwq->cp_bit = 0;
173 }
174 
175 /* All HWQs are power of 2 in size */
176 
177 int bnxt_qplib_alloc_init_hwq(struct bnxt_qplib_hwq *hwq,
178 			      struct bnxt_qplib_hwq_attr *hwq_attr)
179 {
180 	u32 npages, aux_slots, pg_size, aux_pages = 0, aux_size = 0;
181 	struct bnxt_qplib_sg_info sginfo = {};
182 	u32 depth, stride, npbl, npde;
183 	dma_addr_t *src_phys_ptr, **dst_virt_ptr;
184 	struct bnxt_qplib_res *res;
185 	struct pci_dev *pdev;
186 	int i, rc, lvl;
187 
188 	res = hwq_attr->res;
189 	pdev = res->pdev;
190 	pg_size = hwq_attr->sginfo->pgsize;
191 	hwq->level = PBL_LVL_MAX;
192 
193 	depth = roundup_pow_of_two(hwq_attr->depth);
194 	stride = roundup_pow_of_two(hwq_attr->stride);
195 	if (hwq_attr->aux_depth) {
196 		aux_slots = hwq_attr->aux_depth;
197 		aux_size = roundup_pow_of_two(hwq_attr->aux_stride);
198 		aux_pages = (aux_slots * aux_size) / pg_size;
199 		if ((aux_slots * aux_size) % pg_size)
200 			aux_pages++;
201 	}
202 
203 	if (!hwq_attr->sginfo->umem) {
204 		hwq->is_user = false;
205 		npages = (depth * stride) / pg_size + aux_pages;
206 		if ((depth * stride) % pg_size)
207 			npages++;
208 		if (!npages)
209 			return -EINVAL;
210 		hwq_attr->sginfo->npages = npages;
211 	} else {
212 		npages = ib_umem_num_dma_blocks(hwq_attr->sginfo->umem,
213 						hwq_attr->sginfo->pgsize);
214 		hwq->is_user = true;
215 	}
216 
217 	if (npages == MAX_PBL_LVL_0_PGS && !hwq_attr->sginfo->nopte) {
218 		/* This request is Level 0, map PTE */
219 		rc = __alloc_pbl(res, &hwq->pbl[PBL_LVL_0], hwq_attr->sginfo);
220 		if (rc)
221 			goto fail;
222 		hwq->level = PBL_LVL_0;
223 		goto done;
224 	}
225 
226 	if (npages >= MAX_PBL_LVL_0_PGS) {
227 		if (npages > MAX_PBL_LVL_1_PGS) {
228 			u32 flag = (hwq_attr->type == HWQ_TYPE_L2_CMPL) ?
229 				    0 : PTU_PTE_VALID;
230 			/* 2 levels of indirection */
231 			npbl = npages >> MAX_PBL_LVL_1_PGS_SHIFT;
232 			if (npages % BIT(MAX_PBL_LVL_1_PGS_SHIFT))
233 				npbl++;
234 			npde = npbl >> MAX_PDL_LVL_SHIFT;
235 			if (npbl % BIT(MAX_PDL_LVL_SHIFT))
236 				npde++;
237 			/* Alloc PDE pages */
238 			sginfo.pgsize = npde * pg_size;
239 			sginfo.npages = 1;
240 			rc = __alloc_pbl(res, &hwq->pbl[PBL_LVL_0], &sginfo);
241 			if (rc)
242 				goto fail;
243 
244 			/* Alloc PBL pages */
245 			sginfo.npages = npbl;
246 			sginfo.pgsize = PAGE_SIZE;
247 			rc = __alloc_pbl(res, &hwq->pbl[PBL_LVL_1], &sginfo);
248 			if (rc)
249 				goto fail;
250 			/* Fill PDL with PBL page pointers */
251 			dst_virt_ptr =
252 				(dma_addr_t **)hwq->pbl[PBL_LVL_0].pg_arr;
253 			src_phys_ptr = hwq->pbl[PBL_LVL_1].pg_map_arr;
254 			for (i = 0; i < hwq->pbl[PBL_LVL_1].pg_count; i++)
255 				dst_virt_ptr[0][i] = src_phys_ptr[i] | flag;
256 
257 			/* Alloc or init PTEs */
258 			rc = __alloc_pbl(res, &hwq->pbl[PBL_LVL_2],
259 					 hwq_attr->sginfo);
260 			if (rc)
261 				goto fail;
262 			hwq->level = PBL_LVL_2;
263 			if (hwq_attr->sginfo->nopte)
264 				goto done;
265 			/* Fill PBLs with PTE pointers */
266 			dst_virt_ptr =
267 				(dma_addr_t **)hwq->pbl[PBL_LVL_1].pg_arr;
268 			src_phys_ptr = hwq->pbl[PBL_LVL_2].pg_map_arr;
269 			for (i = 0; i < hwq->pbl[PBL_LVL_2].pg_count; i++) {
270 				dst_virt_ptr[PTR_PG(i)][PTR_IDX(i)] =
271 					src_phys_ptr[i] | PTU_PTE_VALID;
272 			}
273 			if (hwq_attr->type == HWQ_TYPE_QUEUE) {
274 				/* Find the last pg of the size */
275 				i = hwq->pbl[PBL_LVL_2].pg_count;
276 				dst_virt_ptr[PTR_PG(i - 1)][PTR_IDX(i - 1)] |=
277 								  PTU_PTE_LAST;
278 				if (i > 1)
279 					dst_virt_ptr[PTR_PG(i - 2)]
280 						    [PTR_IDX(i - 2)] |=
281 						    PTU_PTE_NEXT_TO_LAST;
282 			}
283 		} else { /* pages < 512 npbl = 1, npde = 0 */
284 			u32 flag = (hwq_attr->type == HWQ_TYPE_L2_CMPL) ?
285 				    0 : PTU_PTE_VALID;
286 
287 			/* 1 level of indirection */
288 			npbl = npages >> MAX_PBL_LVL_1_PGS_SHIFT;
289 			if (npages % BIT(MAX_PBL_LVL_1_PGS_SHIFT))
290 				npbl++;
291 			sginfo.npages = npbl;
292 			sginfo.pgsize = PAGE_SIZE;
293 			/* Alloc PBL page */
294 			rc = __alloc_pbl(res, &hwq->pbl[PBL_LVL_0], &sginfo);
295 			if (rc)
296 				goto fail;
297 			/* Alloc or init  PTEs */
298 			rc = __alloc_pbl(res, &hwq->pbl[PBL_LVL_1],
299 					 hwq_attr->sginfo);
300 			if (rc)
301 				goto fail;
302 			hwq->level = PBL_LVL_1;
303 			if (hwq_attr->sginfo->nopte)
304 				goto done;
305 			/* Fill PBL with PTE pointers */
306 			dst_virt_ptr =
307 				(dma_addr_t **)hwq->pbl[PBL_LVL_0].pg_arr;
308 			src_phys_ptr = hwq->pbl[PBL_LVL_1].pg_map_arr;
309 			for (i = 0; i < hwq->pbl[PBL_LVL_1].pg_count; i++)
310 				dst_virt_ptr[PTR_PG(i)][PTR_IDX(i)] =
311 					src_phys_ptr[i] | flag;
312 			if (hwq_attr->type == HWQ_TYPE_QUEUE) {
313 				/* Find the last pg of the size */
314 				i = hwq->pbl[PBL_LVL_1].pg_count;
315 				dst_virt_ptr[PTR_PG(i - 1)][PTR_IDX(i - 1)] |=
316 								  PTU_PTE_LAST;
317 				if (i > 1)
318 					dst_virt_ptr[PTR_PG(i - 2)]
319 						    [PTR_IDX(i - 2)] |=
320 						    PTU_PTE_NEXT_TO_LAST;
321 			}
322 		}
323 	}
324 done:
325 	hwq->prod = 0;
326 	hwq->cons = 0;
327 	hwq->pdev = pdev;
328 	hwq->depth = hwq_attr->depth;
329 	hwq->max_elements = hwq->depth;
330 	hwq->element_size = stride;
331 	hwq->qe_ppg = pg_size / stride;
332 	/* For direct access to the elements */
333 	lvl = hwq->level;
334 	if (hwq_attr->sginfo->nopte && hwq->level)
335 		lvl = hwq->level - 1;
336 	hwq->pbl_ptr = hwq->pbl[lvl].pg_arr;
337 	hwq->pbl_dma_ptr = hwq->pbl[lvl].pg_map_arr;
338 	spin_lock_init(&hwq->lock);
339 
340 	return 0;
341 fail:
342 	bnxt_qplib_free_hwq(res, hwq);
343 	return -ENOMEM;
344 }
345 
346 /* Context Tables */
347 void bnxt_qplib_free_hwctx(struct bnxt_qplib_res *res,
348 			   struct bnxt_qplib_ctx *ctx)
349 {
350 	int i;
351 
352 	bnxt_qplib_free_hwq(res, &ctx->qpc_tbl);
353 	bnxt_qplib_free_hwq(res, &ctx->mrw_tbl);
354 	bnxt_qplib_free_hwq(res, &ctx->srqc_tbl);
355 	bnxt_qplib_free_hwq(res, &ctx->cq_tbl);
356 	bnxt_qplib_free_hwq(res, &ctx->tim_tbl);
357 	for (i = 0; i < MAX_TQM_ALLOC_REQ; i++)
358 		bnxt_qplib_free_hwq(res, &ctx->tqm_ctx.qtbl[i]);
359 	/* restore original pde level before destroy */
360 	ctx->tqm_ctx.pde.level = ctx->tqm_ctx.pde_level;
361 	bnxt_qplib_free_hwq(res, &ctx->tqm_ctx.pde);
362 }
363 
364 static int bnxt_qplib_alloc_tqm_rings(struct bnxt_qplib_res *res,
365 				      struct bnxt_qplib_ctx *ctx)
366 {
367 	struct bnxt_qplib_hwq_attr hwq_attr = {};
368 	struct bnxt_qplib_sg_info sginfo = {};
369 	struct bnxt_qplib_tqm_ctx *tqmctx;
370 	int rc;
371 	int i;
372 
373 	tqmctx = &ctx->tqm_ctx;
374 
375 	sginfo.pgsize = PAGE_SIZE;
376 	sginfo.pgshft = PAGE_SHIFT;
377 	hwq_attr.sginfo = &sginfo;
378 	hwq_attr.res = res;
379 	hwq_attr.type = HWQ_TYPE_CTX;
380 	hwq_attr.depth = 512;
381 	hwq_attr.stride = sizeof(u64);
382 	/* Alloc pdl buffer */
383 	rc = bnxt_qplib_alloc_init_hwq(&tqmctx->pde, &hwq_attr);
384 	if (rc)
385 		goto out;
386 	/* Save original pdl level */
387 	tqmctx->pde_level = tqmctx->pde.level;
388 
389 	hwq_attr.stride = 1;
390 	for (i = 0; i < MAX_TQM_ALLOC_REQ; i++) {
391 		if (!tqmctx->qcount[i])
392 			continue;
393 		hwq_attr.depth = ctx->qpc_count * tqmctx->qcount[i];
394 		rc = bnxt_qplib_alloc_init_hwq(&tqmctx->qtbl[i], &hwq_attr);
395 		if (rc)
396 			goto out;
397 	}
398 out:
399 	return rc;
400 }
401 
402 static void bnxt_qplib_map_tqm_pgtbl(struct bnxt_qplib_tqm_ctx *ctx)
403 {
404 	struct bnxt_qplib_hwq *tbl;
405 	dma_addr_t *dma_ptr;
406 	__le64 **pbl_ptr, *ptr;
407 	int i, j, k;
408 	int fnz_idx = -1;
409 	int pg_count;
410 
411 	pbl_ptr = (__le64 **)ctx->pde.pbl_ptr;
412 
413 	for (i = 0, j = 0; i < MAX_TQM_ALLOC_REQ;
414 	     i++, j += MAX_TQM_ALLOC_BLK_SIZE) {
415 		tbl = &ctx->qtbl[i];
416 		if (!tbl->max_elements)
417 			continue;
418 		if (fnz_idx == -1)
419 			fnz_idx = i; /* first non-zero index */
420 		switch (tbl->level) {
421 		case PBL_LVL_2:
422 			pg_count = tbl->pbl[PBL_LVL_1].pg_count;
423 			for (k = 0; k < pg_count; k++) {
424 				ptr = &pbl_ptr[PTR_PG(j + k)][PTR_IDX(j + k)];
425 				dma_ptr = &tbl->pbl[PBL_LVL_1].pg_map_arr[k];
426 				*ptr = cpu_to_le64(*dma_ptr | PTU_PTE_VALID);
427 			}
428 			break;
429 		case PBL_LVL_1:
430 		case PBL_LVL_0:
431 		default:
432 			ptr = &pbl_ptr[PTR_PG(j)][PTR_IDX(j)];
433 			*ptr = cpu_to_le64(tbl->pbl[PBL_LVL_0].pg_map_arr[0] |
434 					   PTU_PTE_VALID);
435 			break;
436 		}
437 	}
438 	if (fnz_idx == -1)
439 		fnz_idx = 0;
440 	/* update pde level as per page table programming */
441 	ctx->pde.level = (ctx->qtbl[fnz_idx].level == PBL_LVL_2) ? PBL_LVL_2 :
442 			  ctx->qtbl[fnz_idx].level + 1;
443 }
444 
445 static int bnxt_qplib_setup_tqm_rings(struct bnxt_qplib_res *res,
446 				      struct bnxt_qplib_ctx *ctx)
447 {
448 	int rc;
449 
450 	rc = bnxt_qplib_alloc_tqm_rings(res, ctx);
451 	if (rc)
452 		goto fail;
453 
454 	bnxt_qplib_map_tqm_pgtbl(&ctx->tqm_ctx);
455 fail:
456 	return rc;
457 }
458 
459 /*
460  * Routine: bnxt_qplib_alloc_hwctx
461  * Description:
462  *     Context tables are memories which are used by the chip fw.
463  *     The 6 tables defined are:
464  *             QPC ctx - holds QP states
465  *             MRW ctx - holds memory region and window
466  *             SRQ ctx - holds shared RQ states
467  *             CQ ctx - holds completion queue states
468  *             TQM ctx - holds Tx Queue Manager context
469  *             TIM ctx - holds timer context
470  *     Depending on the size of the tbl requested, either a 1 Page Buffer List
471  *     or a 1-to-2-stage indirection Page Directory List + 1 PBL is used
472  *     instead.
473  *     Table might be employed as follows:
474  *             For 0      < ctx size <= 1 PAGE, 0 level of ind is used
475  *             For 1 PAGE < ctx size <= 512 entries size, 1 level of ind is used
476  *             For 512    < ctx size <= MAX, 2 levels of ind is used
477  * Returns:
478  *     0 if success, else -ERRORS
479  */
480 int bnxt_qplib_alloc_hwctx(struct bnxt_qplib_res *res,
481 			   struct bnxt_qplib_ctx *ctx)
482 {
483 	struct bnxt_qplib_hwq_attr hwq_attr = {};
484 	struct bnxt_qplib_sg_info sginfo = {};
485 	int rc;
486 
487 	/* QPC Tables */
488 	sginfo.pgsize = PAGE_SIZE;
489 	sginfo.pgshft = PAGE_SHIFT;
490 	hwq_attr.sginfo = &sginfo;
491 
492 	hwq_attr.res = res;
493 	hwq_attr.depth = ctx->qpc_count;
494 	hwq_attr.stride = BNXT_QPLIB_MAX_QP_CTX_ENTRY_SIZE;
495 	hwq_attr.type = HWQ_TYPE_CTX;
496 	rc = bnxt_qplib_alloc_init_hwq(&ctx->qpc_tbl, &hwq_attr);
497 	if (rc)
498 		goto fail;
499 
500 	/* MRW Tables */
501 	hwq_attr.depth = ctx->mrw_count;
502 	hwq_attr.stride = BNXT_QPLIB_MAX_MRW_CTX_ENTRY_SIZE;
503 	rc = bnxt_qplib_alloc_init_hwq(&ctx->mrw_tbl, &hwq_attr);
504 	if (rc)
505 		goto fail;
506 
507 	/* SRQ Tables */
508 	hwq_attr.depth = ctx->srqc_count;
509 	hwq_attr.stride = BNXT_QPLIB_MAX_SRQ_CTX_ENTRY_SIZE;
510 	rc = bnxt_qplib_alloc_init_hwq(&ctx->srqc_tbl, &hwq_attr);
511 	if (rc)
512 		goto fail;
513 
514 	/* CQ Tables */
515 	hwq_attr.depth = ctx->cq_count;
516 	hwq_attr.stride = BNXT_QPLIB_MAX_CQ_CTX_ENTRY_SIZE;
517 	rc = bnxt_qplib_alloc_init_hwq(&ctx->cq_tbl, &hwq_attr);
518 	if (rc)
519 		goto fail;
520 
521 	/* TQM Buffer */
522 	rc = bnxt_qplib_setup_tqm_rings(res, ctx);
523 	if (rc)
524 		goto fail;
525 	/* TIM Buffer */
526 	ctx->tim_tbl.max_elements = ctx->qpc_count * 16;
527 	hwq_attr.depth = ctx->qpc_count * 16;
528 	hwq_attr.stride = 1;
529 	rc = bnxt_qplib_alloc_init_hwq(&ctx->tim_tbl, &hwq_attr);
530 	if (rc)
531 		goto fail;
532 
533 	return 0;
534 
535 fail:
536 	bnxt_qplib_free_hwctx(res, ctx);
537 	return rc;
538 }
539 
540 static void bnxt_qplib_free_sgid_tbl(struct bnxt_qplib_res *res,
541 				     struct bnxt_qplib_sgid_tbl *sgid_tbl)
542 {
543 	kfree(sgid_tbl->tbl);
544 	kfree(sgid_tbl->hw_id);
545 	kfree(sgid_tbl->ctx);
546 	kfree(sgid_tbl->vlan);
547 	sgid_tbl->tbl = NULL;
548 	sgid_tbl->hw_id = NULL;
549 	sgid_tbl->ctx = NULL;
550 	sgid_tbl->vlan = NULL;
551 	sgid_tbl->max = 0;
552 	sgid_tbl->active = 0;
553 }
554 
555 static int bnxt_qplib_alloc_sgid_tbl(struct bnxt_qplib_res *res,
556 				     struct bnxt_qplib_sgid_tbl *sgid_tbl,
557 				     u16 max)
558 {
559 	sgid_tbl->tbl = kcalloc(max, sizeof(*sgid_tbl->tbl), GFP_KERNEL);
560 	if (!sgid_tbl->tbl)
561 		return -ENOMEM;
562 
563 	sgid_tbl->hw_id = kcalloc(max, sizeof(u16), GFP_KERNEL);
564 	if (!sgid_tbl->hw_id)
565 		goto out_free1;
566 
567 	sgid_tbl->ctx = kcalloc(max, sizeof(void *), GFP_KERNEL);
568 	if (!sgid_tbl->ctx)
569 		goto out_free2;
570 
571 	sgid_tbl->vlan = kcalloc(max, sizeof(u8), GFP_KERNEL);
572 	if (!sgid_tbl->vlan)
573 		goto out_free3;
574 
575 	sgid_tbl->max = max;
576 	return 0;
577 out_free3:
578 	kfree(sgid_tbl->ctx);
579 	sgid_tbl->ctx = NULL;
580 out_free2:
581 	kfree(sgid_tbl->hw_id);
582 	sgid_tbl->hw_id = NULL;
583 out_free1:
584 	kfree(sgid_tbl->tbl);
585 	sgid_tbl->tbl = NULL;
586 	return -ENOMEM;
587 };
588 
589 static void bnxt_qplib_cleanup_sgid_tbl(struct bnxt_qplib_res *res,
590 					struct bnxt_qplib_sgid_tbl *sgid_tbl)
591 {
592 	int i;
593 
594 	for (i = 0; i < sgid_tbl->max; i++) {
595 		if (memcmp(&sgid_tbl->tbl[i], &bnxt_qplib_gid_zero,
596 			   sizeof(bnxt_qplib_gid_zero)))
597 			bnxt_qplib_del_sgid(sgid_tbl, &sgid_tbl->tbl[i].gid,
598 					    sgid_tbl->tbl[i].vlan_id, true);
599 	}
600 	memset(sgid_tbl->tbl, 0, sizeof(*sgid_tbl->tbl) * sgid_tbl->max);
601 	memset(sgid_tbl->hw_id, -1, sizeof(u16) * sgid_tbl->max);
602 	memset(sgid_tbl->vlan, 0, sizeof(u8) * sgid_tbl->max);
603 	sgid_tbl->active = 0;
604 }
605 
606 static void bnxt_qplib_init_sgid_tbl(struct bnxt_qplib_sgid_tbl *sgid_tbl,
607 				     struct net_device *netdev)
608 {
609 	u32 i;
610 
611 	for (i = 0; i < sgid_tbl->max; i++)
612 		sgid_tbl->tbl[i].vlan_id = 0xffff;
613 
614 	memset(sgid_tbl->hw_id, -1, sizeof(u16) * sgid_tbl->max);
615 }
616 
617 /* PDs */
618 int bnxt_qplib_alloc_pd(struct bnxt_qplib_res  *res, struct bnxt_qplib_pd *pd)
619 {
620 	struct bnxt_qplib_pd_tbl *pdt = &res->pd_tbl;
621 	u32 bit_num;
622 	int rc = 0;
623 
624 	mutex_lock(&res->pd_tbl_lock);
625 	bit_num = find_first_bit(pdt->tbl, pdt->max);
626 	if (bit_num == pdt->max) {
627 		rc = -ENOMEM;
628 		goto exit;
629 	}
630 
631 	/* Found unused PD */
632 	clear_bit(bit_num, pdt->tbl);
633 	pd->id = bit_num;
634 exit:
635 	mutex_unlock(&res->pd_tbl_lock);
636 	return rc;
637 }
638 
639 int bnxt_qplib_dealloc_pd(struct bnxt_qplib_res *res,
640 			  struct bnxt_qplib_pd_tbl *pdt,
641 			  struct bnxt_qplib_pd *pd)
642 {
643 	int rc = 0;
644 
645 	mutex_lock(&res->pd_tbl_lock);
646 	if (test_and_set_bit(pd->id, pdt->tbl)) {
647 		dev_warn(&res->pdev->dev, "Freeing an unused PD? pdn = %d\n",
648 			 pd->id);
649 		rc = -EINVAL;
650 		goto exit;
651 	}
652 	pd->id = 0;
653 exit:
654 	mutex_unlock(&res->pd_tbl_lock);
655 	return rc;
656 }
657 
658 static void bnxt_qplib_free_pd_tbl(struct bnxt_qplib_pd_tbl *pdt)
659 {
660 	kfree(pdt->tbl);
661 	pdt->tbl = NULL;
662 	pdt->max = 0;
663 }
664 
665 static int bnxt_qplib_alloc_pd_tbl(struct bnxt_qplib_res *res,
666 				   struct bnxt_qplib_pd_tbl *pdt,
667 				   u32 max)
668 {
669 	u32 bytes;
670 
671 	bytes = max >> 3;
672 	if (!bytes)
673 		bytes = 1;
674 	pdt->tbl = kmalloc(bytes, GFP_KERNEL);
675 	if (!pdt->tbl)
676 		return -ENOMEM;
677 
678 	pdt->max = max;
679 	memset((u8 *)pdt->tbl, 0xFF, bytes);
680 	mutex_init(&res->pd_tbl_lock);
681 
682 	return 0;
683 }
684 
685 /* DPIs */
686 int bnxt_qplib_alloc_dpi(struct bnxt_qplib_res *res,
687 			 struct bnxt_qplib_dpi *dpi,
688 			 void *app, u8 type)
689 {
690 	struct bnxt_qplib_dpi_tbl *dpit = &res->dpi_tbl;
691 	struct bnxt_qplib_reg_desc *reg;
692 	u32 bit_num;
693 	u64 umaddr;
694 
695 	reg = &dpit->wcreg;
696 	mutex_lock(&res->dpi_tbl_lock);
697 
698 	bit_num = find_first_bit(dpit->tbl, dpit->max);
699 	if (bit_num == dpit->max) {
700 		mutex_unlock(&res->dpi_tbl_lock);
701 		return -ENOMEM;
702 	}
703 
704 	/* Found unused DPI */
705 	clear_bit(bit_num, dpit->tbl);
706 	dpit->app_tbl[bit_num] = app;
707 
708 	dpi->bit = bit_num;
709 	dpi->dpi = bit_num + (reg->offset - dpit->ucreg.offset) / PAGE_SIZE;
710 
711 	umaddr = reg->bar_base + reg->offset + bit_num * PAGE_SIZE;
712 	dpi->umdbr = umaddr;
713 
714 	switch (type) {
715 	case BNXT_QPLIB_DPI_TYPE_KERNEL:
716 		/* privileged dbr was already mapped just initialize it. */
717 		dpi->umdbr = dpit->ucreg.bar_base +
718 			     dpit->ucreg.offset + bit_num * PAGE_SIZE;
719 		dpi->dbr = dpit->priv_db;
720 		dpi->dpi = dpi->bit;
721 		break;
722 	case BNXT_QPLIB_DPI_TYPE_WC:
723 		dpi->dbr = ioremap_wc(umaddr, PAGE_SIZE);
724 		break;
725 	default:
726 		dpi->dbr = ioremap(umaddr, PAGE_SIZE);
727 		break;
728 	}
729 
730 	dpi->type = type;
731 	mutex_unlock(&res->dpi_tbl_lock);
732 	return 0;
733 
734 }
735 
736 int bnxt_qplib_dealloc_dpi(struct bnxt_qplib_res *res,
737 			   struct bnxt_qplib_dpi *dpi)
738 {
739 	struct bnxt_qplib_dpi_tbl *dpit = &res->dpi_tbl;
740 
741 	mutex_lock(&res->dpi_tbl_lock);
742 	if (dpi->dpi && dpi->type != BNXT_QPLIB_DPI_TYPE_KERNEL)
743 		pci_iounmap(res->pdev, dpi->dbr);
744 
745 	if (test_and_set_bit(dpi->bit, dpit->tbl)) {
746 		dev_warn(&res->pdev->dev,
747 			 "Freeing an unused DPI? dpi = %d, bit = %d\n",
748 				dpi->dpi, dpi->bit);
749 		mutex_unlock(&res->dpi_tbl_lock);
750 		return -EINVAL;
751 	}
752 	if (dpit->app_tbl)
753 		dpit->app_tbl[dpi->bit] = NULL;
754 	memset(dpi, 0, sizeof(*dpi));
755 	mutex_unlock(&res->dpi_tbl_lock);
756 	return 0;
757 }
758 
759 static void bnxt_qplib_free_dpi_tbl(struct bnxt_qplib_res     *res,
760 				    struct bnxt_qplib_dpi_tbl *dpit)
761 {
762 	kfree(dpit->tbl);
763 	kfree(dpit->app_tbl);
764 	dpit->tbl = NULL;
765 	dpit->app_tbl = NULL;
766 	dpit->max = 0;
767 }
768 
769 static int bnxt_qplib_alloc_dpi_tbl(struct bnxt_qplib_res *res,
770 				    struct bnxt_qplib_dev_attr *dev_attr)
771 {
772 	struct bnxt_qplib_dpi_tbl *dpit;
773 	struct bnxt_qplib_reg_desc *reg;
774 	unsigned long bar_len;
775 	u32 dbr_offset;
776 	u32 bytes;
777 
778 	dpit = &res->dpi_tbl;
779 	reg = &dpit->wcreg;
780 
781 	if (!bnxt_qplib_is_chip_gen_p5_p7(res->cctx)) {
782 		/* Offest should come from L2 driver */
783 		dbr_offset = dev_attr->l2_db_size;
784 		dpit->ucreg.offset = dbr_offset;
785 		dpit->wcreg.offset = dbr_offset;
786 	}
787 
788 	bar_len = pci_resource_len(res->pdev, reg->bar_id);
789 	dpit->max = (bar_len - reg->offset) / PAGE_SIZE;
790 	if (dev_attr->max_dpi)
791 		dpit->max = min_t(u32, dpit->max, dev_attr->max_dpi);
792 
793 	dpit->app_tbl = kcalloc(dpit->max,  sizeof(void *), GFP_KERNEL);
794 	if (!dpit->app_tbl)
795 		return -ENOMEM;
796 
797 	bytes = dpit->max >> 3;
798 	if (!bytes)
799 		bytes = 1;
800 
801 	dpit->tbl = kmalloc(bytes, GFP_KERNEL);
802 	if (!dpit->tbl) {
803 		kfree(dpit->app_tbl);
804 		dpit->app_tbl = NULL;
805 		return -ENOMEM;
806 	}
807 
808 	memset((u8 *)dpit->tbl, 0xFF, bytes);
809 	mutex_init(&res->dpi_tbl_lock);
810 	dpit->priv_db = dpit->ucreg.bar_reg + dpit->ucreg.offset;
811 
812 	return 0;
813 
814 }
815 
816 /* Stats */
817 void bnxt_qplib_free_stats_ctx(struct pci_dev *pdev,
818 			       struct bnxt_qplib_stats *stats)
819 {
820 	if (stats->dma) {
821 		dma_free_coherent(&pdev->dev, stats->size,
822 				  stats->dma, stats->dma_map);
823 	}
824 	memset(stats, 0, sizeof(*stats));
825 	stats->fw_id = -1;
826 }
827 
828 int bnxt_qplib_alloc_stats_ctx(struct pci_dev *pdev,
829 			       struct bnxt_qplib_chip_ctx *cctx,
830 			       struct bnxt_qplib_stats *stats)
831 {
832 	memset(stats, 0, sizeof(*stats));
833 	stats->fw_id = -1;
834 	stats->size = cctx->hw_stats_size;
835 	stats->dma = dma_alloc_coherent(&pdev->dev, stats->size,
836 					&stats->dma_map, GFP_KERNEL);
837 	if (!stats->dma) {
838 		dev_err(&pdev->dev, "Stats DMA allocation failed\n");
839 		return -ENOMEM;
840 	}
841 	return 0;
842 }
843 
844 void bnxt_qplib_cleanup_res(struct bnxt_qplib_res *res)
845 {
846 	bnxt_qplib_cleanup_sgid_tbl(res, &res->sgid_tbl);
847 }
848 
849 int bnxt_qplib_init_res(struct bnxt_qplib_res *res)
850 {
851 	bnxt_qplib_init_sgid_tbl(&res->sgid_tbl, res->netdev);
852 
853 	return 0;
854 }
855 
856 void bnxt_qplib_free_res(struct bnxt_qplib_res *res)
857 {
858 	kfree(res->rcfw->qp_tbl);
859 	bnxt_qplib_free_sgid_tbl(res, &res->sgid_tbl);
860 	bnxt_qplib_free_pd_tbl(&res->pd_tbl);
861 	bnxt_qplib_free_dpi_tbl(res, &res->dpi_tbl);
862 }
863 
864 int bnxt_qplib_alloc_res(struct bnxt_qplib_res *res, struct net_device *netdev)
865 {
866 	struct bnxt_qplib_rcfw *rcfw = res->rcfw;
867 	struct bnxt_qplib_dev_attr *dev_attr;
868 	int rc;
869 
870 	res->netdev = netdev;
871 	dev_attr = res->dattr;
872 
873 	/* Allocate one extra to hold the QP1 entries */
874 	rcfw->qp_tbl_size = max_t(u32, BNXT_RE_MAX_QPC_COUNT + 1, dev_attr->max_qp);
875 	rcfw->qp_tbl = kcalloc(rcfw->qp_tbl_size, sizeof(struct bnxt_qplib_qp_node),
876 			       GFP_KERNEL);
877 	if (!rcfw->qp_tbl)
878 		return -ENOMEM;
879 
880 	rc = bnxt_qplib_alloc_sgid_tbl(res, &res->sgid_tbl, dev_attr->max_sgid);
881 	if (rc)
882 		goto fail;
883 
884 	rc = bnxt_qplib_alloc_pd_tbl(res, &res->pd_tbl, dev_attr->max_pd);
885 	if (rc)
886 		goto fail;
887 
888 	rc = bnxt_qplib_alloc_dpi_tbl(res, dev_attr);
889 	if (rc)
890 		goto fail;
891 
892 	return 0;
893 fail:
894 	bnxt_qplib_free_res(res);
895 	return rc;
896 }
897 
898 void bnxt_qplib_unmap_db_bar(struct bnxt_qplib_res *res)
899 {
900 	struct bnxt_qplib_reg_desc *reg;
901 
902 	reg = &res->dpi_tbl.ucreg;
903 	if (reg->bar_reg)
904 		pci_iounmap(res->pdev, reg->bar_reg);
905 	reg->bar_reg = NULL;
906 	reg->bar_base = 0;
907 	reg->len = 0;
908 	reg->bar_id = 0;
909 }
910 
911 int bnxt_qplib_map_db_bar(struct bnxt_qplib_res *res)
912 {
913 	struct bnxt_qplib_reg_desc *ucreg;
914 	struct bnxt_qplib_reg_desc *wcreg;
915 
916 	wcreg = &res->dpi_tbl.wcreg;
917 	wcreg->bar_id = RCFW_DBR_PCI_BAR_REGION;
918 	wcreg->bar_base = pci_resource_start(res->pdev, wcreg->bar_id);
919 
920 	ucreg = &res->dpi_tbl.ucreg;
921 	ucreg->bar_id = RCFW_DBR_PCI_BAR_REGION;
922 	ucreg->bar_base = pci_resource_start(res->pdev, ucreg->bar_id);
923 	ucreg->len = ucreg->offset + PAGE_SIZE;
924 	if (!ucreg->len || ((ucreg->len & (PAGE_SIZE - 1)) != 0)) {
925 		dev_err(&res->pdev->dev, "QPLIB: invalid dbr length %d",
926 			(int)ucreg->len);
927 		return -EINVAL;
928 	}
929 	ucreg->bar_reg = ioremap(ucreg->bar_base, ucreg->len);
930 	if (!ucreg->bar_reg) {
931 		dev_err(&res->pdev->dev, "privileged dpi map failed!");
932 		return -ENOMEM;
933 	}
934 
935 	return 0;
936 }
937 
938 int bnxt_qplib_determine_atomics(struct pci_dev *dev)
939 {
940 	int comp;
941 	u16 ctl2;
942 
943 	comp = pci_enable_atomic_ops_to_root(dev,
944 					     PCI_EXP_DEVCAP2_ATOMIC_COMP32);
945 	if (comp)
946 		return -EOPNOTSUPP;
947 	comp = pci_enable_atomic_ops_to_root(dev,
948 					     PCI_EXP_DEVCAP2_ATOMIC_COMP64);
949 	if (comp)
950 		return -EOPNOTSUPP;
951 	pcie_capability_read_word(dev, PCI_EXP_DEVCTL2, &ctl2);
952 	return !(ctl2 & PCI_EXP_DEVCTL2_ATOMIC_REQ);
953 }
954