hns_roce_qp.c 25.4 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 34
/*
 * 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.
 */

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

42
#define SQP_NUM				(2 * HNS_ROCE_MAX_PORTS)
43 44 45 46

void hns_roce_qp_event(struct hns_roce_dev *hr_dev, u32 qpn, int event_type)
{
	struct hns_roce_qp_table *qp_table = &hr_dev->qp_table;
47
	struct device *dev = hr_dev->dev;
48 49 50 51 52 53 54 55 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 99 100 101 102 103
	struct hns_roce_qp *qp;

	spin_lock(&qp_table->lock);

	qp = __hns_roce_qp_lookup(hr_dev, qpn);
	if (qp)
		atomic_inc(&qp->refcount);

	spin_unlock(&qp_table->lock);

	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:
104
			dev_dbg(ibqp->device->dev.parent, "roce_ib: Unexpected event type %d on QP %06lx\n",
105 106 107 108 109 110 111 112 113 114 115 116
				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;

117
	return hns_roce_bitmap_alloc_range(&qp_table->bitmap, cnt, align, base);
118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138
}

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;
	}
}
139
EXPORT_SYMBOL_GPL(to_hns_roce_state);
140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156

static int hns_roce_gsi_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;
	int ret;

	if (!qpn)
		return -EINVAL;

	hr_qp->qpn = qpn;

	spin_lock_irq(&qp_table->lock);
	ret = radix_tree_insert(&hr_dev->qp_table_tree,
				hr_qp->qpn & (hr_dev->caps.num_qps - 1), hr_qp);
	spin_unlock_irq(&qp_table->lock);
	if (ret) {
157
		dev_err(hr_dev->dev, "QPC radix_tree_insert failed\n");
158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174
		goto err_put_irrl;
	}

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

	return 0;

err_put_irrl:

	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;
175
	struct device *dev = hr_dev->dev;
176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196
	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;
	}

197 198 199 200 201 202 203 204 205 206
	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;
		}
	}

207 208 209 210 211 212
	spin_lock_irq(&qp_table->lock);
	ret = radix_tree_insert(&hr_dev->qp_table_tree,
				hr_qp->qpn & (hr_dev->caps.num_qps - 1), hr_qp);
	spin_unlock_irq(&qp_table->lock);
	if (ret) {
		dev_err(dev, "QPC radix_tree_insert failed\n");
213
		goto err_put_trrl;
214 215 216 217 218 219 220
	}

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

	return 0;

221 222 223 224
err_put_trrl:
	if (hr_dev->caps.trrl_entry_sz)
		hns_roce_table_put(hr_dev, &qp_table->trrl_table, hr_qp->qpn);

225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244
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)
{
	struct hns_roce_qp_table *qp_table = &hr_dev->qp_table;
	unsigned long flags;

	spin_lock_irqsave(&qp_table->lock, flags);
	radix_tree_delete(&hr_dev->qp_table_tree,
			  hr_qp->qpn & (hr_dev->caps.num_qps - 1));
	spin_unlock_irqrestore(&qp_table->lock, flags);
}
245
EXPORT_SYMBOL_GPL(hns_roce_qp_remove);
246 247 248 249 250 251 252 253 254 255

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) {
256 257 258
		if (hr_dev->caps.trrl_entry_sz)
			hns_roce_table_put(hr_dev, &qp_table->trrl_table,
					   hr_qp->qpn);
259 260 261 262
		hns_roce_table_put(hr_dev, &qp_table->irrl_table, hr_qp->qpn);
		hns_roce_table_put(hr_dev, &qp_table->qp_table, hr_qp->qpn);
	}
}
263
EXPORT_SYMBOL_GPL(hns_roce_qp_free);
264 265 266 267 268 269

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;

270
	if (base_qpn < SQP_NUM)
271 272
		return;

273
	hns_roce_bitmap_free_range(&qp_table->bitmap, base_qpn, cnt, BITMAP_RR);
274
}
275
EXPORT_SYMBOL_GPL(hns_roce_release_range_qp);
276 277 278 279 280

static int hns_roce_set_rq_size(struct hns_roce_dev *hr_dev,
				struct ib_qp_cap *cap, int is_user, int has_srq,
				struct hns_roce_qp *hr_qp)
{
281
	struct device *dev = hr_dev->dev;
282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305
	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;
	}

	/* If srq exit, set zero for relative number of rq */
	if (has_srq) {
		if (cap->max_recv_wr) {
			dev_dbg(dev, "srq no need config max_recv_wr\n");
			return -EINVAL;
		}

		hr_qp->rq.wqe_cnt = hr_qp->rq.max_gs = 0;
	} 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;
		}

306 307 308 309 310
		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;

311 312 313
		hr_qp->rq.wqe_cnt = roundup_pow_of_two(max_cnt);

		if ((u32)hr_qp->rq.wqe_cnt > hr_dev->caps.max_wqes) {
314
			dev_err(dev, "while setting rq size, rq.wqe_cnt too large\n");
315 316 317 318 319
			return -EINVAL;
		}

		max_cnt = max(1U, cap->max_recv_sge);
		hr_qp->rq.max_gs = roundup_pow_of_two(max_cnt);
320 321 322 323 324 325 326
		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);
327 328 329 330 331 332 333 334 335
	}

	cap->max_recv_wr = hr_qp->rq.max_post = hr_qp->rq.wqe_cnt;
	cap->max_recv_sge = hr_qp->rq.max_gs;

	return 0;
}

static int hns_roce_set_user_sq_size(struct hns_roce_dev *hr_dev,
336
				     struct ib_qp_cap *cap,
337 338 339 340 341
				     struct hns_roce_qp *hr_qp,
				     struct hns_roce_ib_create_qp *ucmd)
{
	u32 roundup_sq_stride = roundup_pow_of_two(hr_dev->caps.max_sq_desc_sz);
	u8 max_sq_stride = ilog2(roundup_sq_stride);
342
	u32 page_size;
343
	u32 max_cnt;
344 345 346 347 348

	/* Sanity check SQ size before proceeding */
	if ((u32)(1 << ucmd->log_sq_bb_count) > hr_dev->caps.max_wqes ||
	     ucmd->log_sq_stride > max_sq_stride ||
	     ucmd->log_sq_stride < HNS_ROCE_IB_MIN_SQ_STRIDE) {
349
		dev_err(hr_dev->dev, "check SQ size error!\n");
350 351 352
		return -EINVAL;
	}

353 354 355 356 357 358
	if (cap->max_send_sge > hr_dev->caps.max_sq_sg) {
		dev_err(hr_dev->dev, "SQ sge error! max_send_sge=%d\n",
			cap->max_send_sge);
		return -EINVAL;
	}

359 360 361
	hr_qp->sq.wqe_cnt = 1 << ucmd->log_sq_bb_count;
	hr_qp->sq.wqe_shift = ucmd->log_sq_stride;

362 363 364 365 366 367 368 369 370 371 372
	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));
	hr_qp->sge.sge_shift = 4;

373
	/* Get buf size, SQ and RQ  are aligned to page_szie */
374 375
	if (hr_dev->caps.max_sq_sg <= 2) {
		hr_qp->buff_size = HNS_ROCE_ALOGN_UP((hr_qp->rq.wqe_cnt <<
376
					     hr_qp->rq.wqe_shift), PAGE_SIZE) +
377
				   HNS_ROCE_ALOGN_UP((hr_qp->sq.wqe_cnt <<
378 379
					     hr_qp->sq.wqe_shift), PAGE_SIZE);

380 381
		hr_qp->sq.offset = 0;
		hr_qp->rq.offset = HNS_ROCE_ALOGN_UP((hr_qp->sq.wqe_cnt <<
382
					     hr_qp->sq.wqe_shift), PAGE_SIZE);
383
	} else {
384
		page_size = 1 << (hr_dev->caps.mtt_buf_pg_sz + PAGE_SHIFT);
385
		hr_qp->buff_size = HNS_ROCE_ALOGN_UP((hr_qp->rq.wqe_cnt <<
386
					     hr_qp->rq.wqe_shift), page_size) +
387
				   HNS_ROCE_ALOGN_UP((hr_qp->sge.sge_cnt <<
388
					     hr_qp->sge.sge_shift), page_size) +
389
				   HNS_ROCE_ALOGN_UP((hr_qp->sq.wqe_cnt <<
390
					     hr_qp->sq.wqe_shift), page_size);
391 392 393 394 395 396

		hr_qp->sq.offset = 0;
		if (hr_qp->sge.sge_cnt) {
			hr_qp->sge.offset = HNS_ROCE_ALOGN_UP(
							(hr_qp->sq.wqe_cnt <<
							hr_qp->sq.wqe_shift),
397
							page_size);
398 399 400
			hr_qp->rq.offset = hr_qp->sge.offset +
					HNS_ROCE_ALOGN_UP((hr_qp->sge.sge_cnt <<
						hr_qp->sge.sge_shift),
401
						page_size);
402 403 404 405
		} else {
			hr_qp->rq.offset = HNS_ROCE_ALOGN_UP(
							(hr_qp->sq.wqe_cnt <<
							hr_qp->sq.wqe_shift),
406
							page_size);
407 408
		}
	}
409 410 411 412 413 414 415 416

	return 0;
}

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)
{
417
	struct device *dev = hr_dev->dev;
418
	u32 page_size;
419
	u32 max_cnt;
420
	int size;
421 422 423 424

	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) {
425
		dev_err(dev, "SQ WR or sge or inline data error!\n");
426 427 428 429 430 431 432
		return -EINVAL;
	}

	hr_qp->sq.wqe_shift = ilog2(hr_dev->caps.max_sq_desc_sz);
	hr_qp->sq_max_wqes_per_wr = 1;
	hr_qp->sq_spare_wqes = 0;

433 434 435 436 437
	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;

438 439
	hr_qp->sq.wqe_cnt = roundup_pow_of_two(max_cnt);
	if ((u32)hr_qp->sq.wqe_cnt > hr_dev->caps.max_wqes) {
440
		dev_err(dev, "while setting kernel sq size, sq.wqe_cnt too large\n");
441 442 443 444 445
		return -EINVAL;
	}

	/* Get data_seg numbers */
	max_cnt = max(1U, cap->max_send_sge);
446 447 448 449
	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;
450

451 452 453 454 455 456 457
	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;
	}

	/* Get buf size, SQ and RQ are aligned to PAGE_SIZE */
458
	page_size = 1 << (hr_dev->caps.mtt_buf_pg_sz + PAGE_SHIFT);
459
	hr_qp->sq.offset = 0;
460
	size = HNS_ROCE_ALOGN_UP(hr_qp->sq.wqe_cnt << hr_qp->sq.wqe_shift,
461
				 page_size);
462 463 464 465

	if (hr_dev->caps.max_sq_sg > 2 && hr_qp->sge.sge_cnt) {
		hr_qp->sge.offset = size;
		size += HNS_ROCE_ALOGN_UP(hr_qp->sge.sge_cnt <<
466
					  hr_qp->sge.sge_shift, page_size);
467 468 469 470
	}

	hr_qp->rq.offset = size;
	size += HNS_ROCE_ALOGN_UP((hr_qp->rq.wqe_cnt << hr_qp->rq.wqe_shift),
471
				  page_size);
472
	hr_qp->buff_size = size;
473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489

	/* Get wr and sge number which send */
	cap->max_send_wr = hr_qp->sq.max_post = hr_qp->sq.wqe_cnt;
	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;
}

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)
{
490
	struct device *dev = hr_dev->dev;
491 492 493
	struct hns_roce_ib_create_qp ucmd;
	unsigned long qpn = 0;
	int ret = 0;
494 495
	u32 page_shift;
	u32 npages;
496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521

	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;

	if (init_attr->sq_sig_type == IB_SIGNAL_ALL_WR)
		hr_qp->sq_signal_bits = IB_SIGNAL_ALL_WR;
	else
		hr_qp->sq_signal_bits = IB_SIGNAL_REQ_WR;

	ret = hns_roce_set_rq_size(hr_dev, &init_attr->cap, !!ib_pd->uobject,
				   !!init_attr->srq, hr_qp);
	if (ret) {
		dev_err(dev, "hns_roce_set_rq_size failed\n");
		goto err_out;
	}

	if (ib_pd->uobject) {
		if (ib_copy_from_udata(&ucmd, udata, sizeof(ucmd))) {
			dev_err(dev, "ib_copy_from_udata error for create qp\n");
			ret = -EFAULT;
			goto err_out;
		}

522 523
		ret = hns_roce_set_user_sq_size(hr_dev, &init_attr->cap, hr_qp,
						&ucmd);
524 525 526 527 528 529 530 531 532 533 534 535 536 537
		if (ret) {
			dev_err(dev, "hns_roce_set_user_sq_size error for create qp\n");
			goto err_out;
		}

		hr_qp->umem = ib_umem_get(ib_pd->uobject->context,
					  ucmd.buf_addr, hr_qp->buff_size, 0,
					  0);
		if (IS_ERR(hr_qp->umem)) {
			dev_err(dev, "ib_umem_get error for create qp\n");
			ret = PTR_ERR(hr_qp->umem);
			goto err_out;
		}

538
		hr_qp->mtt.mtt_type = MTT_TYPE_WQE;
539 540 541 542 543 544 545 546 547 548 549 550 551 552
		if (hr_dev->caps.mtt_buf_pg_sz) {
			npages = (ib_umem_page_count(hr_qp->umem) +
				  (1 << hr_dev->caps.mtt_buf_pg_sz) - 1) /
				  (1 << hr_dev->caps.mtt_buf_pg_sz);
			page_shift = PAGE_SHIFT + hr_dev->caps.mtt_buf_pg_sz;
			ret = hns_roce_mtt_init(hr_dev, npages,
				    page_shift,
				    &hr_qp->mtt);
		} else {
			ret = hns_roce_mtt_init(hr_dev,
				    ib_umem_page_count(hr_qp->umem),
				    hr_qp->umem->page_shift,
				    &hr_qp->mtt);
		}
553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579
		if (ret) {
			dev_err(dev, "hns_roce_mtt_init error for create qp\n");
			goto err_buf;
		}

		ret = hns_roce_ib_umem_write_mtt(hr_dev, &hr_qp->mtt,
						 hr_qp->umem);
		if (ret) {
			dev_err(dev, "hns_roce_ib_umem_write_mtt error for create qp\n");
			goto err_mtt;
		}
	} else {
		if (init_attr->create_flags &
		    IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK) {
			dev_err(dev, "init_attr->create_flags error!\n");
			ret = -EINVAL;
			goto err_out;
		}

		if (init_attr->create_flags & IB_QP_CREATE_IPOIB_UD_LSO) {
			dev_err(dev, "init_attr->create_flags error!\n");
			ret = -EINVAL;
			goto err_out;
		}

		/* Set SQ size */
		ret = hns_roce_set_kernel_sq_size(hr_dev, &init_attr->cap,
580
						  hr_qp);
581 582 583 584 585 586
		if (ret) {
			dev_err(dev, "hns_roce_set_kernel_sq_size error!\n");
			goto err_out;
		}

		/* QP doorbell register address */
587
		hr_qp->sq.db_reg_l = hr_dev->reg_base + hr_dev->sdb_offset +
588
				     DB_REG_OFFSET * hr_dev->priv_uar.index;
589
		hr_qp->rq.db_reg_l = hr_dev->reg_base + hr_dev->odb_offset +
590 591 592
				     DB_REG_OFFSET * hr_dev->priv_uar.index;

		/* Allocate QP buf */
593 594 595 596
		page_shift = PAGE_SHIFT + hr_dev->caps.mtt_buf_pg_sz;
		if (hns_roce_buf_alloc(hr_dev, hr_qp->buff_size,
				       (1 << page_shift) * 2,
				       &hr_qp->hr_buf, page_shift)) {
597 598 599 600 601
			dev_err(dev, "hns_roce_buf_alloc error!\n");
			ret = -ENOMEM;
			goto err_out;
		}

602
		hr_qp->mtt.mtt_type = MTT_TYPE_WQE;
603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638
		/* Write MTT */
		ret = hns_roce_mtt_init(hr_dev, hr_qp->hr_buf.npages,
					hr_qp->hr_buf.page_shift, &hr_qp->mtt);
		if (ret) {
			dev_err(dev, "hns_roce_mtt_init error for kernel create qp\n");
			goto err_buf;
		}

		ret = hns_roce_buf_write_mtt(hr_dev, &hr_qp->mtt,
					     &hr_qp->hr_buf);
		if (ret) {
			dev_err(dev, "hns_roce_buf_write_mtt error for kernel create qp\n");
			goto err_mtt;
		}

		hr_qp->sq.wrid = kmalloc_array(hr_qp->sq.wqe_cnt, sizeof(u64),
					       GFP_KERNEL);
		hr_qp->rq.wrid = kmalloc_array(hr_qp->rq.wqe_cnt, sizeof(u64),
					       GFP_KERNEL);
		if (!hr_qp->sq.wrid || !hr_qp->rq.wrid) {
			ret = -ENOMEM;
			goto err_wrid;
		}
	}

	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;
		}
	}

639 640 641
	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 */
642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689
		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
		hr_qp->doorbell_qpn = cpu_to_le64(hr_qp->qpn);

	hr_qp->event = hns_roce_ib_qp_event;

	return 0;

err_qpn:
	if (!sqpn)
		hns_roce_release_range_qp(hr_dev, qpn, 1);

err_wrid:
	kfree(hr_qp->sq.wrid);
	kfree(hr_qp->rq.wrid);

err_mtt:
	hns_roce_mtt_cleanup(hr_dev, &hr_qp->mtt);

err_buf:
	if (ib_pd->uobject)
		ib_umem_release(hr_qp->umem);
	else
		hns_roce_buf_free(hr_dev, hr_qp->buff_size, &hr_qp->hr_buf);

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);
690
	struct device *dev = hr_dev->dev;
691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724
	struct hns_roce_sqp *hr_sqp;
	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) {
			dev_err(dev, "Create RC QP failed\n");
			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: */
		if (pd->uobject) {
			dev_err(dev, "not support usr space GSI\n");
			return ERR_PTR(-EINVAL);
		}

		hr_sqp = kzalloc(sizeof(*hr_sqp), GFP_KERNEL);
		if (!hr_sqp)
			return ERR_PTR(-ENOMEM);

		hr_qp = &hr_sqp->hr_qp;
725 726
		hr_qp->port = init_attr->port_num - 1;
		hr_qp->phy_port = hr_dev->iboe.phy_port[hr_qp->port];
727
		hr_qp->ibqp.qp_num = HNS_ROCE_MAX_PORTS +
728
				     hr_dev->iboe.phy_port[hr_qp->port];
729 730

		ret = hns_roce_create_qp_common(hr_dev, pd, init_attr, udata,
731
						hr_qp->ibqp.qp_num, hr_qp);
732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747
		if (ret) {
			dev_err(dev, "Create GSI QP failed!\n");
			kfree(hr_sqp);
			return ERR_PTR(ret);
		}

		break;
	}
	default:{
		dev_err(dev, "not support QP type %d\n", init_attr->qp_type);
		return ERR_PTR(-EINVAL);
	}
	}

	return &hr_qp->ibqp;
}
748
EXPORT_SYMBOL_GPL(hns_roce_create_qp);
749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766

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;
}
767
EXPORT_SYMBOL_GPL(to_hr_qp_type);
768 769 770 771 772 773 774

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;
775
	struct device *dev = hr_dev->dev;
776 777
	int ret = -EINVAL;
	int p;
L
Lijun Ou 已提交
778
	enum ib_mtu active_mtu;
779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808

	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 (!ib_modify_qp_is_ok(cur_state, new_state, ibqp->qp_type, attr_mask,
				IB_LINK_LAYER_ETHERNET)) {
		dev_err(dev, "ib_modify_qp_is_ok failed\n");
		goto out;
	}

	if ((attr_mask & IB_QP_PORT) &&
	    (attr->port_num == 0 || attr->port_num > hr_dev->caps.num_ports)) {
		dev_err(dev, "attr port_num invalid.attr->port_num=%d\n",
			attr->port_num);
		goto out;
	}

	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]) {
			dev_err(dev, "attr pkey_index invalid.attr->pkey_index=%d\n",
				attr->pkey_index);
			goto out;
		}
	}

L
Lijun Ou 已提交
809 810 811 812
	if (attr_mask & IB_QP_PATH_MTU) {
		p = attr_mask & IB_QP_PORT ? (attr->port_num - 1) : hr_qp->port;
		active_mtu = iboe_get_mtu(hr_dev->iboe.netdevs[p]->mtu);

813 814 815 816
		if ((hr_dev->caps.max_mtu == IB_MTU_4096 &&
		    attr->path_mtu > IB_MTU_4096) ||
		    (hr_dev->caps.max_mtu == IB_MTU_2048 &&
		    attr->path_mtu > IB_MTU_2048) ||
L
Lijun Ou 已提交
817 818 819 820 821 822 823 824
		    attr->path_mtu < IB_MTU_256 ||
		    attr->path_mtu > active_mtu) {
			dev_err(dev, "attr path_mtu(%d)invalid while modify qp",
				attr->path_mtu);
			goto out;
		}
	}

825 826 827 828 829 830 831 832 833 834 835 836 837 838 839
	if (attr_mask & IB_QP_MAX_QP_RD_ATOMIC &&
	    attr->max_rd_atomic > hr_dev->caps.max_qp_init_rdma) {
		dev_err(dev, "attr max_rd_atomic invalid.attr->max_rd_atomic=%d\n",
			attr->max_rd_atomic);
		goto out;
	}

	if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC &&
	    attr->max_dest_rd_atomic > hr_dev->caps.max_qp_dest_rdma) {
		dev_err(dev, "attr max_dest_rd_atomic invalid.attr->max_dest_rd_atomic=%d\n",
			attr->max_dest_rd_atomic);
		goto out;
	}

	if (cur_state == new_state && cur_state == IB_QPS_RESET) {
840
		ret = 0;
841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866
		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);
	}
}
867
EXPORT_SYMBOL_GPL(hns_roce_lock_cqs);
868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883

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);
	}
}
884
EXPORT_SYMBOL_GPL(hns_roce_unlock_cqs);
885 886 887 888 889 890 891 892 893 894 895 896 897

__be32 send_ieth(struct ib_send_wr *wr)
{
	switch (wr->opcode) {
	case IB_WR_SEND_WITH_IMM:
	case IB_WR_RDMA_WRITE_WITH_IMM:
		return cpu_to_le32(wr->ex.imm_data);
	case IB_WR_SEND_WITH_INV:
		return cpu_to_le32(wr->ex.invalidate_rkey);
	default:
		return 0;
	}
}
898
EXPORT_SYMBOL_GPL(send_ieth);
899 900 901 902 903 904 905 906 907 908 909

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));
}
910
EXPORT_SYMBOL_GPL(get_recv_wqe);
911 912 913 914 915

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));
}
916
EXPORT_SYMBOL_GPL(get_send_wqe);
917

918 919 920 921 922 923 924
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));
}
EXPORT_SYMBOL_GPL(get_send_extend_sge);

925 926 927 928 929 930 931 932
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;
	if (likely(cur + nreq < hr_wq->max_post))
933
		return false;
934 935 936 937 938 939 940 941

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

	return cur + nreq >= hr_wq->max_post;
}
942
EXPORT_SYMBOL_GPL(hns_roce_wq_overflow);
943 944 945 946 947 948 949 950 951 952 953 954

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;
	int ret;

	spin_lock_init(&qp_table->lock);
	INIT_RADIX_TREE(&hr_dev->qp_table_tree, GFP_ATOMIC);

	/* A port include two SQP, six port total 12 */
	ret = hns_roce_bitmap_init(&qp_table->bitmap, hr_dev->caps.num_qps,
955
				   hr_dev->caps.num_qps - 1, SQP_NUM,
956 957
				   reserved_from_top);
	if (ret) {
958
		dev_err(hr_dev->dev, "qp bitmap init failed!error=%d\n",
959 960 961 962 963 964 965 966 967 968 969
			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);
}