hns_roce_qp.c 34.2 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33
/*
 * Copyright (c) 2016 Hisilicon Limited.
 * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
 *
 * This software is available to you under a choice of one of two
 * licenses.  You may choose to be licensed under the terms of the GNU
 * General Public License (GPL) Version 2, available from the file
 * COPYING in the main directory of this source tree, or the
 * OpenIB.org BSD license below:
 *
 *     Redistribution and use in source and binary forms, with or
 *     without modification, are permitted provided that the following
 *     conditions are met:
 *
 *      - Redistributions of source code must retain the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer.
 *
 *      - Redistributions in binary form must reproduce the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer in the documentation and/or other materials
 *        provided with the distribution.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 */

34
#include <linux/pci.h>
35
#include <linux/platform_device.h>
L
Lijun Ou 已提交
36
#include <rdma/ib_addr.h>
37
#include <rdma/ib_umem.h>
38
#include <rdma/uverbs_ioctl.h>
39 40 41
#include "hns_roce_common.h"
#include "hns_roce_device.h"
#include "hns_roce_hem.h"
42
#include <rdma/hns-abi.h>
43

44
#define SQP_NUM				(2 * HNS_ROCE_MAX_PORTS)
45 46 47

void hns_roce_qp_event(struct hns_roce_dev *hr_dev, u32 qpn, int event_type)
{
48
	struct device *dev = hr_dev->dev;
49 50
	struct hns_roce_qp *qp;

51
	xa_lock(&hr_dev->qp_table_xa);
52 53 54
	qp = __hns_roce_qp_lookup(hr_dev, qpn);
	if (qp)
		atomic_inc(&qp->refcount);
55
	xa_unlock(&hr_dev->qp_table_xa);
56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102

	if (!qp) {
		dev_warn(dev, "Async event for bogus QP %08x\n", qpn);
		return;
	}

	qp->event(qp, (enum hns_roce_event)event_type);

	if (atomic_dec_and_test(&qp->refcount))
		complete(&qp->free);
}

static void hns_roce_ib_qp_event(struct hns_roce_qp *hr_qp,
				 enum hns_roce_event type)
{
	struct ib_event event;
	struct ib_qp *ibqp = &hr_qp->ibqp;

	if (ibqp->event_handler) {
		event.device = ibqp->device;
		event.element.qp = ibqp;
		switch (type) {
		case HNS_ROCE_EVENT_TYPE_PATH_MIG:
			event.event = IB_EVENT_PATH_MIG;
			break;
		case HNS_ROCE_EVENT_TYPE_COMM_EST:
			event.event = IB_EVENT_COMM_EST;
			break;
		case HNS_ROCE_EVENT_TYPE_SQ_DRAINED:
			event.event = IB_EVENT_SQ_DRAINED;
			break;
		case HNS_ROCE_EVENT_TYPE_SRQ_LAST_WQE_REACH:
			event.event = IB_EVENT_QP_LAST_WQE_REACHED;
			break;
		case HNS_ROCE_EVENT_TYPE_WQ_CATAS_ERROR:
			event.event = IB_EVENT_QP_FATAL;
			break;
		case HNS_ROCE_EVENT_TYPE_PATH_MIG_FAILED:
			event.event = IB_EVENT_PATH_MIG_ERR;
			break;
		case HNS_ROCE_EVENT_TYPE_INV_REQ_LOCAL_WQ_ERROR:
			event.event = IB_EVENT_QP_REQ_ERR;
			break;
		case HNS_ROCE_EVENT_TYPE_LOCAL_WQ_ACCESS_ERROR:
			event.event = IB_EVENT_QP_ACCESS_ERR;
			break;
		default:
103
			dev_dbg(ibqp->device->dev.parent, "roce_ib: Unexpected event type %d on QP %06lx\n",
104 105 106 107 108 109 110 111 112 113 114 115
				type, hr_qp->qpn);
			return;
		}
		ibqp->event_handler(&event, ibqp->qp_context);
	}
}

static int hns_roce_reserve_range_qp(struct hns_roce_dev *hr_dev, int cnt,
				     int align, unsigned long *base)
{
	struct hns_roce_qp_table *qp_table = &hr_dev->qp_table;

116 117 118 119
	return hns_roce_bitmap_alloc_range(&qp_table->bitmap, cnt, align,
					   base) ?
		       -ENOMEM :
		       0;
120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144
}

enum hns_roce_qp_state to_hns_roce_state(enum ib_qp_state state)
{
	switch (state) {
	case IB_QPS_RESET:
		return HNS_ROCE_QP_STATE_RST;
	case IB_QPS_INIT:
		return HNS_ROCE_QP_STATE_INIT;
	case IB_QPS_RTR:
		return HNS_ROCE_QP_STATE_RTR;
	case IB_QPS_RTS:
		return HNS_ROCE_QP_STATE_RTS;
	case IB_QPS_SQD:
		return HNS_ROCE_QP_STATE_SQD;
	case IB_QPS_ERR:
		return HNS_ROCE_QP_STATE_ERR;
	default:
		return HNS_ROCE_QP_NUM_STATE;
	}
}

static int hns_roce_gsi_qp_alloc(struct hns_roce_dev *hr_dev, unsigned long qpn,
				 struct hns_roce_qp *hr_qp)
{
145
	struct xarray *xa = &hr_dev->qp_table_xa;
146 147 148 149 150 151 152 153 154
	int ret;

	if (!qpn)
		return -EINVAL;

	hr_qp->qpn = qpn;
	atomic_set(&hr_qp->refcount, 1);
	init_completion(&hr_qp->free);

155 156 157 158
	ret = xa_err(xa_store_irq(xa, hr_qp->qpn & (hr_dev->caps.num_qps - 1),
				hr_qp, GFP_KERNEL));
	if (ret)
		dev_err(hr_dev->dev, "QPC xa_store failed\n");
159 160 161 162 163 164 165 166

	return ret;
}

static int hns_roce_qp_alloc(struct hns_roce_dev *hr_dev, unsigned long qpn,
			     struct hns_roce_qp *hr_qp)
{
	struct hns_roce_qp_table *qp_table = &hr_dev->qp_table;
167
	struct device *dev = hr_dev->dev;
168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188
	int ret;

	if (!qpn)
		return -EINVAL;

	hr_qp->qpn = qpn;

	/* Alloc memory for QPC */
	ret = hns_roce_table_get(hr_dev, &qp_table->qp_table, hr_qp->qpn);
	if (ret) {
		dev_err(dev, "QPC table get failed\n");
		goto err_out;
	}

	/* Alloc memory for IRRL */
	ret = hns_roce_table_get(hr_dev, &qp_table->irrl_table, hr_qp->qpn);
	if (ret) {
		dev_err(dev, "IRRL table get failed\n");
		goto err_put_qp;
	}

189 190 191 192 193 194 195 196 197 198
	if (hr_dev->caps.trrl_entry_sz) {
		/* Alloc memory for TRRL */
		ret = hns_roce_table_get(hr_dev, &qp_table->trrl_table,
					 hr_qp->qpn);
		if (ret) {
			dev_err(dev, "TRRL table get failed\n");
			goto err_put_irrl;
		}
	}

199 200 201 202 203 204 205 206 207 208
	if (hr_dev->caps.sccc_entry_sz) {
		/* Alloc memory for SCC CTX */
		ret = hns_roce_table_get(hr_dev, &qp_table->sccc_table,
					 hr_qp->qpn);
		if (ret) {
			dev_err(dev, "SCC CTX table get failed\n");
			goto err_put_trrl;
		}
	}

209 210
	ret = hns_roce_gsi_qp_alloc(hr_dev, qpn, hr_qp);
	if (ret)
211
		goto err_put_sccc;
212 213 214

	return 0;

215 216 217 218 219
err_put_sccc:
	if (hr_dev->caps.sccc_entry_sz)
		hns_roce_table_put(hr_dev, &qp_table->sccc_table,
				   hr_qp->qpn);

220 221 222 223
err_put_trrl:
	if (hr_dev->caps.trrl_entry_sz)
		hns_roce_table_put(hr_dev, &qp_table->trrl_table, hr_qp->qpn);

224 225 226 227 228 229 230 231 232 233 234 235
err_put_irrl:
	hns_roce_table_put(hr_dev, &qp_table->irrl_table, hr_qp->qpn);

err_put_qp:
	hns_roce_table_put(hr_dev, &qp_table->qp_table, hr_qp->qpn);

err_out:
	return ret;
}

void hns_roce_qp_remove(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp)
{
236
	struct xarray *xa = &hr_dev->qp_table_xa;
237 238
	unsigned long flags;

239 240 241
	xa_lock_irqsave(xa, flags);
	__xa_erase(xa, hr_qp->qpn & (hr_dev->caps.num_qps - 1));
	xa_unlock_irqrestore(xa, flags);
242 243 244 245 246 247 248 249 250 251 252
}

void hns_roce_qp_free(struct hns_roce_dev *hr_dev, struct hns_roce_qp *hr_qp)
{
	struct hns_roce_qp_table *qp_table = &hr_dev->qp_table;

	if (atomic_dec_and_test(&hr_qp->refcount))
		complete(&hr_qp->free);
	wait_for_completion(&hr_qp->free);

	if ((hr_qp->ibqp.qp_type) != IB_QPT_GSI) {
253 254 255
		if (hr_dev->caps.trrl_entry_sz)
			hns_roce_table_put(hr_dev, &qp_table->trrl_table,
					   hr_qp->qpn);
256 257 258 259 260 261 262 263 264
		hns_roce_table_put(hr_dev, &qp_table->irrl_table, hr_qp->qpn);
	}
}

void hns_roce_release_range_qp(struct hns_roce_dev *hr_dev, int base_qpn,
			       int cnt)
{
	struct hns_roce_qp_table *qp_table = &hr_dev->qp_table;

C
chenglang 已提交
265
	if (base_qpn < hr_dev->caps.reserved_qps)
266 267
		return;

268
	hns_roce_bitmap_free_range(&qp_table->bitmap, base_qpn, cnt, BITMAP_RR);
269 270 271
}

static int hns_roce_set_rq_size(struct hns_roce_dev *hr_dev,
272
				struct ib_qp_cap *cap, bool is_user, int has_rq,
273 274
				struct hns_roce_qp *hr_qp)
{
275
	struct device *dev = hr_dev->dev;
276 277 278 279 280 281 282 283 284 285
	u32 max_cnt;

	/* Check the validity of QP support capacity */
	if (cap->max_recv_wr > hr_dev->caps.max_wqes ||
	    cap->max_recv_sge > hr_dev->caps.max_rq_sg) {
		dev_err(dev, "RQ WR or sge error!max_recv_wr=%d max_recv_sge=%d\n",
			cap->max_recv_wr, cap->max_recv_sge);
		return -EINVAL;
	}

286 287 288 289 290 291
	/* If srq exist, set zero for relative number of rq */
	if (!has_rq) {
		hr_qp->rq.wqe_cnt = 0;
		hr_qp->rq.max_gs = 0;
		cap->max_recv_wr = 0;
		cap->max_recv_sge = 0;
292 293 294 295 296 297
	} else {
		if (is_user && (!cap->max_recv_wr || !cap->max_recv_sge)) {
			dev_err(dev, "user space no need config max_recv_wr max_recv_sge\n");
			return -EINVAL;
		}

298 299 300 301 302
		if (hr_dev->caps.min_wqes)
			max_cnt = max(cap->max_recv_wr, hr_dev->caps.min_wqes);
		else
			max_cnt = cap->max_recv_wr;

303 304 305
		hr_qp->rq.wqe_cnt = roundup_pow_of_two(max_cnt);

		if ((u32)hr_qp->rq.wqe_cnt > hr_dev->caps.max_wqes) {
306
			dev_err(dev, "while setting rq size, rq.wqe_cnt too large\n");
307 308 309 310 311
			return -EINVAL;
		}

		max_cnt = max(1U, cap->max_recv_sge);
		hr_qp->rq.max_gs = roundup_pow_of_two(max_cnt);
312 313 314 315 316 317 318
		if (hr_dev->caps.max_rq_sg <= 2)
			hr_qp->rq.wqe_shift =
					ilog2(hr_dev->caps.max_rq_desc_sz);
		else
			hr_qp->rq.wqe_shift =
					ilog2(hr_dev->caps.max_rq_desc_sz
					      * hr_qp->rq.max_gs);
319 320
	}

321
	cap->max_recv_wr = hr_qp->rq.wqe_cnt;
322 323 324 325 326
	cap->max_recv_sge = hr_qp->rq.max_gs;

	return 0;
}

327 328 329
static int check_sq_size_with_integrity(struct hns_roce_dev *hr_dev,
					struct ib_qp_cap *cap,
					struct hns_roce_ib_create_qp *ucmd)
330 331 332 333 334
{
	u32 roundup_sq_stride = roundup_pow_of_two(hr_dev->caps.max_sq_desc_sz);
	u8 max_sq_stride = ilog2(roundup_sq_stride);

	/* Sanity check SQ size before proceeding */
335 336
	if (ucmd->log_sq_stride > max_sq_stride ||
	    ucmd->log_sq_stride < HNS_ROCE_IB_MIN_SQ_STRIDE) {
337
		ibdev_err(&hr_dev->ib_dev, "check SQ size error!\n");
338 339 340
		return -EINVAL;
	}

341
	if (cap->max_send_sge > hr_dev->caps.max_sq_sg) {
342 343
		ibdev_err(&hr_dev->ib_dev, "SQ sge error! max_send_sge=%d\n",
			  cap->max_send_sge);
344 345 346
		return -EINVAL;
	}

347 348 349 350 351 352 353 354 355 356 357 358 359
	return 0;
}

static int hns_roce_set_user_sq_size(struct hns_roce_dev *hr_dev,
				     struct ib_qp_cap *cap,
				     struct hns_roce_qp *hr_qp,
				     struct hns_roce_ib_create_qp *ucmd)
{
	u32 ex_sge_num;
	u32 page_size;
	u32 max_cnt;
	int ret;

360 361 362 363
	if (check_shl_overflow(1, ucmd->log_sq_bb_count, &hr_qp->sq.wqe_cnt) ||
	    hr_qp->sq.wqe_cnt > hr_dev->caps.max_wqes)
		return -EINVAL;

364 365 366 367 368 369
	ret = check_sq_size_with_integrity(hr_dev, cap, ucmd);
	if (ret) {
		ibdev_err(&hr_dev->ib_dev, "Sanity check sq size failed\n");
		return ret;
	}

370 371
	hr_qp->sq.wqe_shift = ucmd->log_sq_stride;

372 373 374 375 376 377 378 379 380
	max_cnt = max(1U, cap->max_send_sge);
	if (hr_dev->caps.max_sq_sg <= 2)
		hr_qp->sq.max_gs = roundup_pow_of_two(max_cnt);
	else
		hr_qp->sq.max_gs = max_cnt;

	if (hr_qp->sq.max_gs > 2)
		hr_qp->sge.sge_cnt = roundup_pow_of_two(hr_qp->sq.wqe_cnt *
							(hr_qp->sq.max_gs - 2));
381 382 383 384 385 386 387 388 389 390

	if ((hr_qp->sq.max_gs > 2) && (hr_dev->pci_dev->revision == 0x20)) {
		if (hr_qp->sge.sge_cnt > hr_dev->caps.max_extend_sg) {
			dev_err(hr_dev->dev,
				"The extended sge cnt error! sge_cnt=%d\n",
				hr_qp->sge.sge_cnt);
			return -EINVAL;
		}
	}

391
	hr_qp->sge.sge_shift = 4;
L
Lijun Ou 已提交
392
	ex_sge_num = hr_qp->sge.sge_cnt;
393

394
	/* Get buf size, SQ and RQ  are aligned to page_szie */
395
	if (hr_dev->caps.max_sq_sg <= 2) {
396
		hr_qp->buff_size = HNS_ROCE_ALIGN_UP((hr_qp->rq.wqe_cnt <<
397
					     hr_qp->rq.wqe_shift), PAGE_SIZE) +
398
				   HNS_ROCE_ALIGN_UP((hr_qp->sq.wqe_cnt <<
399 400
					     hr_qp->sq.wqe_shift), PAGE_SIZE);

401
		hr_qp->sq.offset = 0;
402
		hr_qp->rq.offset = HNS_ROCE_ALIGN_UP((hr_qp->sq.wqe_cnt <<
403
					     hr_qp->sq.wqe_shift), PAGE_SIZE);
404
	} else {
405
		page_size = 1 << (hr_dev->caps.mtt_buf_pg_sz + PAGE_SHIFT);
406 407
		hr_qp->sge.sge_cnt = ex_sge_num ?
		   max(page_size / (1 << hr_qp->sge.sge_shift), ex_sge_num) : 0;
408
		hr_qp->buff_size = HNS_ROCE_ALIGN_UP((hr_qp->rq.wqe_cnt <<
409
					     hr_qp->rq.wqe_shift), page_size) +
410
				   HNS_ROCE_ALIGN_UP((hr_qp->sge.sge_cnt <<
411
					     hr_qp->sge.sge_shift), page_size) +
412
				   HNS_ROCE_ALIGN_UP((hr_qp->sq.wqe_cnt <<
413
					     hr_qp->sq.wqe_shift), page_size);
414 415

		hr_qp->sq.offset = 0;
L
Lijun Ou 已提交
416
		if (ex_sge_num) {
417
			hr_qp->sge.offset = HNS_ROCE_ALIGN_UP(
418 419
							(hr_qp->sq.wqe_cnt <<
							hr_qp->sq.wqe_shift),
420
							page_size);
421
			hr_qp->rq.offset = hr_qp->sge.offset +
422
					HNS_ROCE_ALIGN_UP((hr_qp->sge.sge_cnt <<
423
						hr_qp->sge.sge_shift),
424
						page_size);
425
		} else {
426
			hr_qp->rq.offset = HNS_ROCE_ALIGN_UP(
427 428
							(hr_qp->sq.wqe_cnt <<
							hr_qp->sq.wqe_shift),
429
							page_size);
430 431
		}
	}
432 433 434 435

	return 0;
}

436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520
static int split_wqe_buf_region(struct hns_roce_dev *hr_dev,
				struct hns_roce_qp *hr_qp,
				struct hns_roce_buf_region *regions,
				int region_max, int page_shift)
{
	int page_size = 1 << page_shift;
	bool is_extend_sge;
	int region_cnt = 0;
	int buf_size;
	int buf_cnt;

	if (hr_qp->buff_size < 1 || region_max < 1)
		return region_cnt;

	if (hr_qp->sge.sge_cnt > 0)
		is_extend_sge = true;
	else
		is_extend_sge = false;

	/* sq region */
	if (is_extend_sge)
		buf_size = hr_qp->sge.offset - hr_qp->sq.offset;
	else
		buf_size = hr_qp->rq.offset - hr_qp->sq.offset;

	if (buf_size > 0 && region_cnt < region_max) {
		buf_cnt = DIV_ROUND_UP(buf_size, page_size);
		hns_roce_init_buf_region(&regions[region_cnt],
					 hr_dev->caps.wqe_sq_hop_num,
					 hr_qp->sq.offset / page_size,
					 buf_cnt);
		region_cnt++;
	}

	/* sge region */
	if (is_extend_sge) {
		buf_size = hr_qp->rq.offset - hr_qp->sge.offset;
		if (buf_size > 0 && region_cnt < region_max) {
			buf_cnt = DIV_ROUND_UP(buf_size, page_size);
			hns_roce_init_buf_region(&regions[region_cnt],
						 hr_dev->caps.wqe_sge_hop_num,
						 hr_qp->sge.offset / page_size,
						 buf_cnt);
			region_cnt++;
		}
	}

	/* rq region */
	buf_size = hr_qp->buff_size - hr_qp->rq.offset;
	if (buf_size > 0) {
		buf_cnt = DIV_ROUND_UP(buf_size, page_size);
		hns_roce_init_buf_region(&regions[region_cnt],
					 hr_dev->caps.wqe_rq_hop_num,
					 hr_qp->rq.offset / page_size,
					 buf_cnt);
		region_cnt++;
	}

	return region_cnt;
}

static int calc_wqe_bt_page_shift(struct hns_roce_dev *hr_dev,
				  struct hns_roce_buf_region *regions,
				  int region_cnt)
{
	int bt_pg_shift;
	int ba_num;
	int ret;

	bt_pg_shift = PAGE_SHIFT + hr_dev->caps.mtt_ba_pg_sz;

	/* all root ba entries must in one bt page */
	do {
		ba_num = (1 << bt_pg_shift) / BA_BYTE_LEN;
		ret = hns_roce_hem_list_calc_root_ba(regions, region_cnt,
						     ba_num);
		if (ret <= ba_num)
			break;

		bt_pg_shift++;
	} while (ret > ba_num);

	return bt_pg_shift - PAGE_SHIFT;
}

521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549
static int set_extend_sge_param(struct hns_roce_dev *hr_dev,
				struct hns_roce_qp *hr_qp)
{
	struct device *dev = hr_dev->dev;

	if (hr_qp->sq.max_gs > 2) {
		hr_qp->sge.sge_cnt = roundup_pow_of_two(hr_qp->sq.wqe_cnt *
				     (hr_qp->sq.max_gs - 2));
		hr_qp->sge.sge_shift = 4;
	}

	/* ud sqwqe's sge use extend sge */
	if (hr_dev->caps.max_sq_sg > 2 && hr_qp->ibqp.qp_type == IB_QPT_GSI) {
		hr_qp->sge.sge_cnt = roundup_pow_of_two(hr_qp->sq.wqe_cnt *
				     hr_qp->sq.max_gs);
		hr_qp->sge.sge_shift = 4;
	}

	if ((hr_qp->sq.max_gs > 2) && hr_dev->pci_dev->revision == 0x20) {
		if (hr_qp->sge.sge_cnt > hr_dev->caps.max_extend_sg) {
			dev_err(dev, "The extended sge cnt error! sge_cnt=%d\n",
				hr_qp->sge.sge_cnt);
			return -EINVAL;
		}
	}

	return 0;
}

550 551 552 553
static int hns_roce_set_kernel_sq_size(struct hns_roce_dev *hr_dev,
				       struct ib_qp_cap *cap,
				       struct hns_roce_qp *hr_qp)
{
554
	struct device *dev = hr_dev->dev;
555
	u32 page_size;
556
	u32 max_cnt;
557
	int size;
558
	int ret;
559 560 561 562

	if (cap->max_send_wr  > hr_dev->caps.max_wqes  ||
	    cap->max_send_sge > hr_dev->caps.max_sq_sg ||
	    cap->max_inline_data > hr_dev->caps.max_sq_inline) {
563
		dev_err(dev, "SQ WR or sge or inline data error!\n");
564 565 566 567 568
		return -EINVAL;
	}

	hr_qp->sq.wqe_shift = ilog2(hr_dev->caps.max_sq_desc_sz);

569 570 571 572 573
	if (hr_dev->caps.min_wqes)
		max_cnt = max(cap->max_send_wr, hr_dev->caps.min_wqes);
	else
		max_cnt = cap->max_send_wr;

574 575
	hr_qp->sq.wqe_cnt = roundup_pow_of_two(max_cnt);
	if ((u32)hr_qp->sq.wqe_cnt > hr_dev->caps.max_wqes) {
576
		dev_err(dev, "while setting kernel sq size, sq.wqe_cnt too large\n");
577 578 579 580 581
		return -EINVAL;
	}

	/* Get data_seg numbers */
	max_cnt = max(1U, cap->max_send_sge);
582 583 584 585
	if (hr_dev->caps.max_sq_sg <= 2)
		hr_qp->sq.max_gs = roundup_pow_of_two(max_cnt);
	else
		hr_qp->sq.max_gs = max_cnt;
586

587 588 589 590
	ret = set_extend_sge_param(hr_dev, hr_qp);
	if (ret) {
		dev_err(dev, "set extend sge parameters fail\n");
		return ret;
591 592
	}

593
	/* Get buf size, SQ and RQ are aligned to PAGE_SIZE */
594
	page_size = 1 << (hr_dev->caps.mtt_buf_pg_sz + PAGE_SHIFT);
595
	hr_qp->sq.offset = 0;
596
	size = HNS_ROCE_ALIGN_UP(hr_qp->sq.wqe_cnt << hr_qp->sq.wqe_shift,
597
				 page_size);
598 599

	if (hr_dev->caps.max_sq_sg > 2 && hr_qp->sge.sge_cnt) {
L
Lijun Ou 已提交
600 601
		hr_qp->sge.sge_cnt = max(page_size/(1 << hr_qp->sge.sge_shift),
					(u32)hr_qp->sge.sge_cnt);
602
		hr_qp->sge.offset = size;
603
		size += HNS_ROCE_ALIGN_UP(hr_qp->sge.sge_cnt <<
604
					  hr_qp->sge.sge_shift, page_size);
605 606 607
	}

	hr_qp->rq.offset = size;
608
	size += HNS_ROCE_ALIGN_UP((hr_qp->rq.wqe_cnt << hr_qp->rq.wqe_shift),
609
				  page_size);
610
	hr_qp->buff_size = size;
611 612

	/* Get wr and sge number which send */
613
	cap->max_send_wr = hr_qp->sq.wqe_cnt;
614 615 616 617 618 619 620 621
	cap->max_send_sge = hr_qp->sq.max_gs;

	/* We don't support inline sends for kernel QPs (yet) */
	cap->max_inline_data = 0;

	return 0;
}

622 623
static int hns_roce_qp_has_sq(struct ib_qp_init_attr *attr)
{
L
Lijun Ou 已提交
624
	if (attr->qp_type == IB_QPT_XRC_TGT || !attr->cap.max_send_wr)
625 626 627 628 629
		return 0;

	return 1;
}

630 631 632
static int hns_roce_qp_has_rq(struct ib_qp_init_attr *attr)
{
	if (attr->qp_type == IB_QPT_XRC_INI ||
633 634
	    attr->qp_type == IB_QPT_XRC_TGT || attr->srq ||
	    !attr->cap.max_recv_wr)
635 636 637 638 639
		return 0;

	return 1;
}

640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683
static int alloc_rq_inline_buf(struct hns_roce_qp *hr_qp,
			       struct ib_qp_init_attr *init_attr)
{
	u32 max_recv_sge = init_attr->cap.max_recv_sge;
	struct hns_roce_rinl_wqe *wqe_list;
	u32 wqe_cnt = hr_qp->rq.wqe_cnt;
	int i;

	/* allocate recv inline buf */
	wqe_list = kcalloc(wqe_cnt, sizeof(struct hns_roce_rinl_wqe),
			   GFP_KERNEL);

	if (!wqe_list)
		goto err;

	/* Allocate a continuous buffer for all inline sge we need */
	wqe_list[0].sg_list = kcalloc(wqe_cnt, (max_recv_sge *
				      sizeof(struct hns_roce_rinl_sge)),
				      GFP_KERNEL);
	if (!wqe_list[0].sg_list)
		goto err_wqe_list;

	/* Assign buffers of sg_list to each inline wqe */
	for (i = 1; i < wqe_cnt; i++)
		wqe_list[i].sg_list = &wqe_list[0].sg_list[i * max_recv_sge];

	hr_qp->rq_inl_buf.wqe_list = wqe_list;
	hr_qp->rq_inl_buf.wqe_cnt = wqe_cnt;

	return 0;

err_wqe_list:
	kfree(wqe_list);

err:
	return -ENOMEM;
}

static void free_rq_inline_buf(struct hns_roce_qp *hr_qp)
{
	kfree(hr_qp->rq_inl_buf.wqe_list[0].sg_list);
	kfree(hr_qp->rq_inl_buf.wqe_list);
}

684 685 686 687 688 689
static int hns_roce_create_qp_common(struct hns_roce_dev *hr_dev,
				     struct ib_pd *ib_pd,
				     struct ib_qp_init_attr *init_attr,
				     struct ib_udata *udata, unsigned long sqpn,
				     struct hns_roce_qp *hr_qp)
{
690
	dma_addr_t *buf_list[ARRAY_SIZE(hr_qp->regions)] = { NULL };
691
	struct device *dev = hr_dev->dev;
692
	struct hns_roce_ib_create_qp ucmd;
693
	struct hns_roce_ib_create_qp_resp resp = {};
694 695
	struct hns_roce_ucontext *uctx = rdma_udata_to_drv_context(
		udata, struct hns_roce_ucontext, ibucontext);
696
	struct hns_roce_buf_region *r;
697
	unsigned long qpn = 0;
698
	u32 page_shift;
699 700
	int buf_count;
	int ret;
701
	int i;
702 703 704 705 706 707 708

	mutex_init(&hr_qp->mutex);
	spin_lock_init(&hr_qp->sq.lock);
	spin_lock_init(&hr_qp->rq.lock);

	hr_qp->state = IB_QPS_RESET;

O
oulijun 已提交
709 710
	hr_qp->ibqp.qp_type = init_attr->qp_type;

711
	if (init_attr->sq_sig_type == IB_SIGNAL_ALL_WR)
712
		hr_qp->sq_signal_bits = IB_SIGNAL_ALL_WR;
713
	else
714
		hr_qp->sq_signal_bits = IB_SIGNAL_REQ_WR;
715

716
	ret = hns_roce_set_rq_size(hr_dev, &init_attr->cap, udata,
717
				   hns_roce_qp_has_rq(init_attr), hr_qp);
718 719 720 721 722
	if (ret) {
		dev_err(dev, "hns_roce_set_rq_size failed\n");
		goto err_out;
	}

723 724
	if ((hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_RQ_INLINE) &&
	    hns_roce_qp_has_rq(init_attr)) {
725 726 727
		ret = alloc_rq_inline_buf(hr_qp, init_attr);
		if (ret) {
			dev_err(dev, "allocate receive inline buffer failed\n");
728 729 730 731
			goto err_out;
		}
	}

732
	page_shift = PAGE_SHIFT + hr_dev->caps.mtt_buf_pg_sz;
733
	if (udata) {
734 735 736
		if (ib_copy_from_udata(&ucmd, udata, sizeof(ucmd))) {
			dev_err(dev, "ib_copy_from_udata error for create qp\n");
			ret = -EFAULT;
737
			goto err_alloc_rq_inline_buf;
738 739
		}

740 741
		ret = hns_roce_set_user_sq_size(hr_dev, &init_attr->cap, hr_qp,
						&ucmd);
742 743
		if (ret) {
			dev_err(dev, "hns_roce_set_user_sq_size error for create qp\n");
744
			goto err_alloc_rq_inline_buf;
745 746
		}

747
		hr_qp->umem = ib_umem_get(udata, ucmd.buf_addr,
748
					  hr_qp->buff_size, 0);
749 750 751
		if (IS_ERR(hr_qp->umem)) {
			dev_err(dev, "ib_umem_get error for create qp\n");
			ret = PTR_ERR(hr_qp->umem);
752
			goto err_alloc_rq_inline_buf;
753
		}
754 755 756 757 758
		hr_qp->region_cnt = split_wqe_buf_region(hr_dev, hr_qp,
				hr_qp->regions, ARRAY_SIZE(hr_qp->regions),
				page_shift);
		ret = hns_roce_alloc_buf_list(hr_qp->regions, buf_list,
					      hr_qp->region_cnt);
759
		if (ret) {
760 761
			dev_err(dev, "alloc buf_list error for create qp\n");
			goto err_alloc_list;
762 763
		}

764 765 766 767 768 769 770 771 772 773 774 775
		for (i = 0; i < hr_qp->region_cnt; i++) {
			r = &hr_qp->regions[i];
			buf_count = hns_roce_get_umem_bufs(hr_dev,
					buf_list[i], r->count, r->offset,
					hr_qp->umem, page_shift);
			if (buf_count != r->count) {
				dev_err(dev,
					"get umem buf err, expect %d,ret %d.\n",
					r->count, buf_count);
				ret = -ENOBUFS;
				goto err_get_bufs;
			}
776
		}
777

778 779 780 781
		if ((hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SQ_RECORD_DB) &&
		    (udata->inlen >= sizeof(ucmd)) &&
		    (udata->outlen >= sizeof(resp)) &&
		    hns_roce_qp_has_sq(init_attr)) {
782 783
			ret = hns_roce_db_map_user(uctx, udata, ucmd.sdb_addr,
						   &hr_qp->sdb);
784 785
			if (ret) {
				dev_err(dev, "sq record doorbell map failed!\n");
786
				goto err_get_bufs;
787 788 789 790 791 792 793
			}

			/* indicate kernel supports sq record db */
			resp.cap_flags |= HNS_ROCE_SUPPORT_SQ_RECORD_DB;
			hr_qp->sdb_en = 1;
		}

794
		if ((hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_RECORD_DB) &&
795
		    (udata->outlen >= sizeof(resp)) &&
796
		    hns_roce_qp_has_rq(init_attr)) {
797 798
			ret = hns_roce_db_map_user(uctx, udata, ucmd.db_addr,
						   &hr_qp->rdb);
799
			if (ret) {
O
oulijun 已提交
800
				dev_err(dev, "rq record doorbell map failed!\n");
801
				goto err_sq_dbmap;
802
			}
803 804 805 806

			/* indicate kernel supports rq record db */
			resp.cap_flags |= HNS_ROCE_SUPPORT_RQ_RECORD_DB;
			hr_qp->rdb_en = 1;
807
		}
808 809 810 811 812
	} else {
		if (init_attr->create_flags &
		    IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK) {
			dev_err(dev, "init_attr->create_flags error!\n");
			ret = -EINVAL;
813
			goto err_alloc_rq_inline_buf;
814 815 816 817 818
		}

		if (init_attr->create_flags & IB_QP_CREATE_IPOIB_UD_LSO) {
			dev_err(dev, "init_attr->create_flags error!\n");
			ret = -EINVAL;
819
			goto err_alloc_rq_inline_buf;
820 821 822 823
		}

		/* Set SQ size */
		ret = hns_roce_set_kernel_sq_size(hr_dev, &init_attr->cap,
824
						  hr_qp);
825 826
		if (ret) {
			dev_err(dev, "hns_roce_set_kernel_sq_size error!\n");
827
			goto err_alloc_rq_inline_buf;
828 829 830
		}

		/* QP doorbell register address */
831
		hr_qp->sq.db_reg_l = hr_dev->reg_base + hr_dev->sdb_offset +
832
				     DB_REG_OFFSET * hr_dev->priv_uar.index;
833
		hr_qp->rq.db_reg_l = hr_dev->reg_base + hr_dev->odb_offset +
834 835
				     DB_REG_OFFSET * hr_dev->priv_uar.index;

836 837 838 839 840
		if ((hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_RECORD_DB) &&
		    hns_roce_qp_has_rq(init_attr)) {
			ret = hns_roce_alloc_db(hr_dev, &hr_qp->rdb, 0);
			if (ret) {
				dev_err(dev, "rq record doorbell alloc failed!\n");
841
				goto err_alloc_rq_inline_buf;
842 843
			}
			*hr_qp->rdb.db_record = 0;
844
			hr_qp->rdb_en = 1;
845 846
		}

847
		/* Allocate QP buf */
848 849 850
		if (hns_roce_buf_alloc(hr_dev, hr_qp->buff_size,
				       (1 << page_shift) * 2,
				       &hr_qp->hr_buf, page_shift)) {
851 852
			dev_err(dev, "hns_roce_buf_alloc error!\n");
			ret = -ENOMEM;
853
			goto err_db;
854
		}
855 856 857 858 859
		hr_qp->region_cnt = split_wqe_buf_region(hr_dev, hr_qp,
				hr_qp->regions, ARRAY_SIZE(hr_qp->regions),
				page_shift);
		ret = hns_roce_alloc_buf_list(hr_qp->regions, buf_list,
					      hr_qp->region_cnt);
860
		if (ret) {
861 862
			dev_err(dev, "alloc buf_list error for create qp!\n");
			goto err_alloc_list;
863 864
		}

865 866 867 868 869 870 871 872 873 874 875 876
		for (i = 0; i < hr_qp->region_cnt; i++) {
			r = &hr_qp->regions[i];
			buf_count = hns_roce_get_kmem_bufs(hr_dev,
					buf_list[i], r->count, r->offset,
					&hr_qp->hr_buf);
			if (buf_count != r->count) {
				dev_err(dev,
					"get kmem buf err, expect %d,ret %d.\n",
					r->count, buf_count);
				ret = -ENOBUFS;
				goto err_get_bufs;
			}
877 878
		}

879 880
		hr_qp->sq.wrid = kcalloc(hr_qp->sq.wqe_cnt, sizeof(u64),
					 GFP_KERNEL);
881
		if (ZERO_OR_NULL_PTR(hr_qp->sq.wrid)) {
882
			ret = -ENOMEM;
883 884 885 886 887 888 889 890 891 892
			goto err_get_bufs;
		}

		if (hr_qp->rq.wqe_cnt) {
			hr_qp->rq.wrid = kcalloc(hr_qp->rq.wqe_cnt, sizeof(u64),
						 GFP_KERNEL);
			if (ZERO_OR_NULL_PTR(hr_qp->rq.wrid)) {
				ret = -ENOMEM;
				goto err_sq_wrid;
			}
893 894 895 896 897 898 899 900 901 902 903 904 905 906
		}
	}

	if (sqpn) {
		qpn = sqpn;
	} else {
		/* Get QPN */
		ret = hns_roce_reserve_range_qp(hr_dev, 1, 1, &qpn);
		if (ret) {
			dev_err(dev, "hns_roce_reserve_range_qp alloc qpn error\n");
			goto err_wrid;
		}
	}

907 908 909 910 911 912 913
	hr_qp->wqe_bt_pg_shift = calc_wqe_bt_page_shift(hr_dev, hr_qp->regions,
							hr_qp->region_cnt);
	hns_roce_mtr_init(&hr_qp->mtr, PAGE_SHIFT + hr_qp->wqe_bt_pg_shift,
			  page_shift);
	ret = hns_roce_mtr_attach(hr_dev, &hr_qp->mtr, buf_list,
				  hr_qp->regions, hr_qp->region_cnt);
	if (ret) {
914
		dev_err(dev, "mtr attach error for create qp\n");
915 916 917
		goto err_mtr;
	}

918 919 920
	if (init_attr->qp_type == IB_QPT_GSI &&
	    hr_dev->hw_rev == HNS_ROCE_HW_VER1) {
		/* In v1 engine, GSI QP context in RoCE engine's register */
921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936
		ret = hns_roce_gsi_qp_alloc(hr_dev, qpn, hr_qp);
		if (ret) {
			dev_err(dev, "hns_roce_qp_alloc failed!\n");
			goto err_qpn;
		}
	} else {
		ret = hns_roce_qp_alloc(hr_dev, qpn, hr_qp);
		if (ret) {
			dev_err(dev, "hns_roce_qp_alloc failed!\n");
			goto err_qpn;
		}
	}

	if (sqpn)
		hr_qp->doorbell_qpn = 1;
	else
937
		hr_qp->doorbell_qpn = (u32)hr_qp->qpn;
938

939 940 941
	if (udata) {
		ret = ib_copy_to_udata(udata, &resp,
				       min(udata->outlen, sizeof(resp)));
942 943 944
		if (ret)
			goto err_qp;
	}
945 946 947 948 949 950 951

	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_QP_FLOW_CTRL) {
		ret = hr_dev->hw->qp_flow_control_init(hr_dev, hr_qp);
		if (ret)
			goto err_qp;
	}

952
	hr_qp->event = hns_roce_ib_qp_event;
953
	hns_roce_free_buf_list(buf_list, hr_qp->region_cnt);
954 955 956

	return 0;

957 958 959 960 961 962 963
err_qp:
	if (init_attr->qp_type == IB_QPT_GSI &&
		hr_dev->hw_rev == HNS_ROCE_HW_VER1)
		hns_roce_qp_remove(hr_dev, hr_qp);
	else
		hns_roce_qp_free(hr_dev, hr_qp);

964 965 966 967
err_qpn:
	if (!sqpn)
		hns_roce_release_range_qp(hr_dev, qpn, 1);

968 969 970
err_mtr:
	hns_roce_mtr_cleanup(hr_dev, &hr_qp->mtr);

971
err_wrid:
972
	if (udata) {
973
		if ((hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_RECORD_DB) &&
974
		    (udata->outlen >= sizeof(resp)) &&
975
		    hns_roce_qp_has_rq(init_attr))
976
			hns_roce_db_unmap_user(uctx, &hr_qp->rdb);
977
	} else {
978 979
		if (hr_qp->rq.wqe_cnt)
			kfree(hr_qp->rq.wrid);
980
	}
981

982
err_sq_dbmap:
983
	if (udata)
984 985 986 987
		if ((hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SQ_RECORD_DB) &&
		    (udata->inlen >= sizeof(ucmd)) &&
		    (udata->outlen >= sizeof(resp)) &&
		    hns_roce_qp_has_sq(init_attr))
988
			hns_roce_db_unmap_user(uctx, &hr_qp->sdb);
989

990 991 992 993
err_sq_wrid:
	if (!udata)
		kfree(hr_qp->sq.wrid);

994 995
err_get_bufs:
	hns_roce_free_buf_list(buf_list, hr_qp->region_cnt);
996

997
err_alloc_list:
998
	if (!hr_qp->umem)
999
		hns_roce_buf_free(hr_dev, hr_qp->buff_size, &hr_qp->hr_buf);
1000
	ib_umem_release(hr_qp->umem);
1001

1002
err_db:
1003
	if (!udata && hns_roce_qp_has_rq(init_attr) &&
1004 1005 1006
	    (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_RECORD_DB))
		hns_roce_free_db(hr_dev, &hr_qp->rdb);

1007
err_alloc_rq_inline_buf:
1008 1009
	if ((hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_RQ_INLINE) &&
	     hns_roce_qp_has_rq(init_attr))
1010
		free_rq_inline_buf(hr_qp);
1011

1012 1013 1014 1015 1016 1017 1018 1019 1020
err_out:
	return ret;
}

struct ib_qp *hns_roce_create_qp(struct ib_pd *pd,
				 struct ib_qp_init_attr *init_attr,
				 struct ib_udata *udata)
{
	struct hns_roce_dev *hr_dev = to_hr_dev(pd->device);
1021
	struct ib_device *ibdev = &hr_dev->ib_dev;
1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033
	struct hns_roce_qp *hr_qp;
	int ret;

	switch (init_attr->qp_type) {
	case IB_QPT_RC: {
		hr_qp = kzalloc(sizeof(*hr_qp), GFP_KERNEL);
		if (!hr_qp)
			return ERR_PTR(-ENOMEM);

		ret = hns_roce_create_qp_common(hr_dev, pd, init_attr, udata, 0,
						hr_qp);
		if (ret) {
1034
			ibdev_err(ibdev, "Create QP 0x%06lx failed(%d)\n",
1035
				  hr_qp->qpn, ret);
1036 1037 1038 1039 1040 1041 1042 1043 1044 1045
			kfree(hr_qp);
			return ERR_PTR(ret);
		}

		hr_qp->ibqp.qp_num = hr_qp->qpn;

		break;
	}
	case IB_QPT_GSI: {
		/* Userspace is not allowed to create special QPs: */
1046
		if (udata) {
1047
			ibdev_err(ibdev, "not support usr space GSI\n");
1048 1049 1050
			return ERR_PTR(-EINVAL);
		}

1051 1052
		hr_qp = kzalloc(sizeof(*hr_qp), GFP_KERNEL);
		if (!hr_qp)
1053 1054
			return ERR_PTR(-ENOMEM);

1055 1056
		hr_qp->port = init_attr->port_num - 1;
		hr_qp->phy_port = hr_dev->iboe.phy_port[hr_qp->port];
O
oulijun 已提交
1057 1058 1059 1060 1061 1062 1063

		/* when hw version is v1, the sqpn is allocated */
		if (hr_dev->caps.max_sq_sg <= 2)
			hr_qp->ibqp.qp_num = HNS_ROCE_MAX_PORTS +
					     hr_dev->iboe.phy_port[hr_qp->port];
		else
			hr_qp->ibqp.qp_num = 1;
1064 1065

		ret = hns_roce_create_qp_common(hr_dev, pd, init_attr, udata,
1066
						hr_qp->ibqp.qp_num, hr_qp);
1067
		if (ret) {
1068
			ibdev_err(ibdev, "Create GSI QP failed!\n");
1069
			kfree(hr_qp);
1070 1071 1072 1073 1074 1075
			return ERR_PTR(ret);
		}

		break;
	}
	default:{
1076 1077
		ibdev_err(ibdev, "not support QP type %d\n",
			  init_attr->qp_type);
1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102
		return ERR_PTR(-EINVAL);
	}
	}

	return &hr_qp->ibqp;
}

int to_hr_qp_type(int qp_type)
{
	int transport_type;

	if (qp_type == IB_QPT_RC)
		transport_type = SERV_TYPE_RC;
	else if (qp_type == IB_QPT_UC)
		transport_type = SERV_TYPE_UC;
	else if (qp_type == IB_QPT_UD)
		transport_type = SERV_TYPE_UD;
	else if (qp_type == IB_QPT_GSI)
		transport_type = SERV_TYPE_UD;
	else
		transport_type = -1;

	return transport_type;
}

1103 1104 1105
static int check_mtu_validate(struct hns_roce_dev *hr_dev,
			      struct hns_roce_qp *hr_qp,
			      struct ib_qp_attr *attr, int attr_mask)
1106
{
L
Lijun Ou 已提交
1107
	enum ib_mtu active_mtu;
1108
	int p;
1109

1110
	p = attr_mask & IB_QP_PORT ? (attr->port_num - 1) : hr_qp->port;
1111
	active_mtu = iboe_get_mtu(hr_dev->iboe.netdevs[p]->mtu);
1112

1113 1114 1115
	if ((hr_dev->caps.max_mtu >= IB_MTU_2048 &&
	    attr->path_mtu > hr_dev->caps.max_mtu) ||
	    attr->path_mtu < IB_MTU_256 || attr->path_mtu > active_mtu) {
1116 1117
		ibdev_err(&hr_dev->ib_dev,
			"attr path_mtu(%d)invalid while modify qp",
1118 1119
			attr->path_mtu);
		return -EINVAL;
1120 1121
	}

1122 1123 1124 1125 1126 1127 1128 1129 1130
	return 0;
}

static int hns_roce_check_qp_attr(struct ib_qp *ibqp, struct ib_qp_attr *attr,
				  int attr_mask)
{
	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
	int p;
1131 1132 1133

	if ((attr_mask & IB_QP_PORT) &&
	    (attr->port_num == 0 || attr->port_num > hr_dev->caps.num_ports)) {
1134 1135
		ibdev_err(&hr_dev->ib_dev,
			"attr port_num invalid.attr->port_num=%d\n",
1136
			attr->port_num);
1137
		return -EINVAL;
1138 1139 1140 1141 1142
	}

	if (attr_mask & IB_QP_PKEY_INDEX) {
		p = attr_mask & IB_QP_PORT ? (attr->port_num - 1) : hr_qp->port;
		if (attr->pkey_index >= hr_dev->caps.pkey_table_len[p]) {
1143 1144
			ibdev_err(&hr_dev->ib_dev,
				"attr pkey_index invalid.attr->pkey_index=%d\n",
1145
				attr->pkey_index);
1146
			return -EINVAL;
L
Lijun Ou 已提交
1147 1148 1149
		}
	}

1150 1151
	if (attr_mask & IB_QP_MAX_QP_RD_ATOMIC &&
	    attr->max_rd_atomic > hr_dev->caps.max_qp_init_rdma) {
1152 1153
		ibdev_err(&hr_dev->ib_dev,
			"attr max_rd_atomic invalid.attr->max_rd_atomic=%d\n",
1154
			attr->max_rd_atomic);
1155
		return -EINVAL;
1156 1157 1158 1159
	}

	if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC &&
	    attr->max_dest_rd_atomic > hr_dev->caps.max_qp_dest_rdma) {
1160 1161
		ibdev_err(&hr_dev->ib_dev,
			"attr max_dest_rd_atomic invalid.attr->max_dest_rd_atomic=%d\n",
1162
			attr->max_dest_rd_atomic);
1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193
		return -EINVAL;
	}

	if (attr_mask & IB_QP_PATH_MTU)
		return check_mtu_validate(hr_dev, hr_qp, attr, attr_mask);

	return 0;
}

int hns_roce_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
		       int attr_mask, struct ib_udata *udata)
{
	struct hns_roce_dev *hr_dev = to_hr_dev(ibqp->device);
	struct hns_roce_qp *hr_qp = to_hr_qp(ibqp);
	enum ib_qp_state cur_state, new_state;
	int ret = -EINVAL;

	mutex_lock(&hr_qp->mutex);

	cur_state = attr_mask & IB_QP_CUR_STATE ?
		    attr->cur_qp_state : (enum ib_qp_state)hr_qp->state;
	new_state = attr_mask & IB_QP_STATE ? attr->qp_state : cur_state;

	if (ibqp->uobject &&
	    (attr_mask & IB_QP_STATE) && new_state == IB_QPS_ERR) {
		if (hr_qp->sdb_en == 1) {
			hr_qp->sq.head = *(int *)(hr_qp->sdb.virt_addr);

			if (hr_qp->rdb_en == 1)
				hr_qp->rq.head = *(int *)(hr_qp->rdb.virt_addr);
		} else {
1194 1195
			ibdev_warn(&hr_dev->ib_dev,
				  "flush cqe is not supported in userspace!\n");
1196 1197 1198 1199 1200 1201
			goto out;
		}
	}

	if (!ib_modify_qp_is_ok(cur_state, new_state, ibqp->qp_type,
				attr_mask)) {
1202
		ibdev_err(&hr_dev->ib_dev, "ib_modify_qp_is_ok failed\n");
1203 1204 1205
		goto out;
	}

1206 1207 1208 1209
	ret = hns_roce_check_qp_attr(ibqp, attr, attr_mask);
	if (ret)
		goto out;

1210
	if (cur_state == new_state && cur_state == IB_QPS_RESET) {
1211 1212
		if (hr_dev->caps.min_wqes) {
			ret = -EPERM;
1213 1214
			ibdev_err(&hr_dev->ib_dev,
				"cur_state=%d new_state=%d\n", cur_state,
1215 1216 1217 1218 1219
				new_state);
		} else {
			ret = 0;
		}

1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278
		goto out;
	}

	ret = hr_dev->hw->modify_qp(ibqp, attr, attr_mask, cur_state,
				    new_state);

out:
	mutex_unlock(&hr_qp->mutex);

	return ret;
}

void hns_roce_lock_cqs(struct hns_roce_cq *send_cq, struct hns_roce_cq *recv_cq)
		       __acquires(&send_cq->lock) __acquires(&recv_cq->lock)
{
	if (send_cq == recv_cq) {
		spin_lock_irq(&send_cq->lock);
		__acquire(&recv_cq->lock);
	} else if (send_cq->cqn < recv_cq->cqn) {
		spin_lock_irq(&send_cq->lock);
		spin_lock_nested(&recv_cq->lock, SINGLE_DEPTH_NESTING);
	} else {
		spin_lock_irq(&recv_cq->lock);
		spin_lock_nested(&send_cq->lock, SINGLE_DEPTH_NESTING);
	}
}

void hns_roce_unlock_cqs(struct hns_roce_cq *send_cq,
			 struct hns_roce_cq *recv_cq) __releases(&send_cq->lock)
			 __releases(&recv_cq->lock)
{
	if (send_cq == recv_cq) {
		__release(&recv_cq->lock);
		spin_unlock_irq(&send_cq->lock);
	} else if (send_cq->cqn < recv_cq->cqn) {
		spin_unlock(&recv_cq->lock);
		spin_unlock_irq(&send_cq->lock);
	} else {
		spin_unlock(&send_cq->lock);
		spin_unlock_irq(&recv_cq->lock);
	}
}

static void *get_wqe(struct hns_roce_qp *hr_qp, int offset)
{

	return hns_roce_buf_offset(&hr_qp->hr_buf, offset);
}

void *get_recv_wqe(struct hns_roce_qp *hr_qp, int n)
{
	return get_wqe(hr_qp, hr_qp->rq.offset + (n << hr_qp->rq.wqe_shift));
}

void *get_send_wqe(struct hns_roce_qp *hr_qp, int n)
{
	return get_wqe(hr_qp, hr_qp->sq.offset + (n << hr_qp->sq.wqe_shift));
}

1279 1280 1281 1282 1283 1284
void *get_send_extend_sge(struct hns_roce_qp *hr_qp, int n)
{
	return hns_roce_buf_offset(&hr_qp->hr_buf, hr_qp->sge.offset +
					(n << hr_qp->sge.sge_shift));
}

1285 1286 1287 1288 1289 1290 1291
bool hns_roce_wq_overflow(struct hns_roce_wq *hr_wq, int nreq,
			  struct ib_cq *ib_cq)
{
	struct hns_roce_cq *hr_cq;
	u32 cur;

	cur = hr_wq->head - hr_wq->tail;
1292
	if (likely(cur + nreq < hr_wq->wqe_cnt))
1293
		return false;
1294 1295 1296 1297 1298 1299

	hr_cq = to_hr_cq(ib_cq);
	spin_lock(&hr_cq->lock);
	cur = hr_wq->head - hr_wq->tail;
	spin_unlock(&hr_cq->lock);

1300
	return cur + nreq >= hr_wq->wqe_cnt;
1301 1302 1303 1304 1305 1306
}

int hns_roce_init_qp_table(struct hns_roce_dev *hr_dev)
{
	struct hns_roce_qp_table *qp_table = &hr_dev->qp_table;
	int reserved_from_top = 0;
L
Lijun Ou 已提交
1307
	int reserved_from_bot;
1308 1309
	int ret;

1310
	mutex_init(&qp_table->scc_mutex);
1311
	xa_init(&hr_dev->qp_table_xa);
1312

C
chenglang 已提交
1313
	reserved_from_bot = hr_dev->caps.reserved_qps;
L
Lijun Ou 已提交
1314

1315
	ret = hns_roce_bitmap_init(&qp_table->bitmap, hr_dev->caps.num_qps,
L
Lijun Ou 已提交
1316
				   hr_dev->caps.num_qps - 1, reserved_from_bot,
1317 1318
				   reserved_from_top);
	if (ret) {
1319
		dev_err(hr_dev->dev, "qp bitmap init failed!error=%d\n",
1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330
			ret);
		return ret;
	}

	return 0;
}

void hns_roce_cleanup_qp_table(struct hns_roce_dev *hr_dev)
{
	hns_roce_bitmap_cleanup(&hr_dev->qp_table.bitmap);
}