qp.c 46.7 KB
Newer Older
1
/*
2
 * Copyright(c) 2016 Intel Corporation.
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 35 36 37 38 39 40 41 42 43 44 45 46 47
 *
 * This file is provided under a dual BSD/GPLv2 license.  When using or
 * redistributing this file, you may do so under either license.
 *
 * GPL LICENSE SUMMARY
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of version 2 of the GNU General Public License as
 * published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful, but
 * WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * BSD LICENSE
 *
 * 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.
 *  - Neither the name of Intel Corporation nor the names of its
 *    contributors may be used to endorse or promote products derived
 *    from this software without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *
 */

48
#include <linux/hash.h>
49 50
#include <linux/bitops.h>
#include <linux/lockdep.h>
51 52 53
#include <linux/vmalloc.h>
#include <linux/slab.h>
#include <rdma/ib_verbs.h>
54
#include "qp.h"
55
#include "vt.h"
56
#include "trace.h"
57

58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78
/*
 * Note that it is OK to post send work requests in the SQE and ERR
 * states; rvt_do_send() will process them and generate error
 * completions as per IB 1.2 C10-96.
 */
const int ib_rvt_state_ops[IB_QPS_ERR + 1] = {
	[IB_QPS_RESET] = 0,
	[IB_QPS_INIT] = RVT_POST_RECV_OK,
	[IB_QPS_RTR] = RVT_POST_RECV_OK | RVT_PROCESS_RECV_OK,
	[IB_QPS_RTS] = RVT_POST_RECV_OK | RVT_PROCESS_RECV_OK |
	    RVT_POST_SEND_OK | RVT_PROCESS_SEND_OK |
	    RVT_PROCESS_NEXT_SEND_OK,
	[IB_QPS_SQD] = RVT_POST_RECV_OK | RVT_PROCESS_RECV_OK |
	    RVT_POST_SEND_OK | RVT_PROCESS_SEND_OK,
	[IB_QPS_SQE] = RVT_POST_RECV_OK | RVT_PROCESS_RECV_OK |
	    RVT_POST_SEND_OK | RVT_FLUSH_SEND,
	[IB_QPS_ERR] = RVT_POST_RECV_OK | RVT_FLUSH_RECV |
	    RVT_POST_SEND_OK | RVT_FLUSH_SEND,
};
EXPORT_SYMBOL(ib_rvt_state_ops);

79 80 81
static void get_map_page(struct rvt_qpn_table *qpt,
			 struct rvt_qpn_map *map,
			 gfp_t gfp)
82
{
83
	unsigned long page = get_zeroed_page(gfp);
84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106

	/*
	 * Free the page if someone raced with us installing it.
	 */

	spin_lock(&qpt->lock);
	if (map->page)
		free_page(page);
	else
		map->page = (void *)page;
	spin_unlock(&qpt->lock);
}

/**
 * init_qpn_table - initialize the QP number table for a device
 * @qpt: the QPN table
 */
static int init_qpn_table(struct rvt_dev_info *rdi, struct rvt_qpn_table *qpt)
{
	u32 offset, i;
	struct rvt_qpn_map *map;
	int ret = 0;

107
	if (!(rdi->dparms.qpn_res_end >= rdi->dparms.qpn_res_start))
108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132
		return -EINVAL;

	spin_lock_init(&qpt->lock);

	qpt->last = rdi->dparms.qpn_start;
	qpt->incr = rdi->dparms.qpn_inc << rdi->dparms.qos_shift;

	/*
	 * Drivers may want some QPs beyond what we need for verbs let them use
	 * our qpn table. No need for two. Lets go ahead and mark the bitmaps
	 * for those. The reserved range must be *after* the range which verbs
	 * will pick from.
	 */

	/* Figure out number of bit maps needed before reserved range */
	qpt->nmaps = rdi->dparms.qpn_res_start / RVT_BITS_PER_PAGE;

	/* This should always be zero */
	offset = rdi->dparms.qpn_res_start & RVT_BITS_PER_PAGE_MASK;

	/* Starting with the first reserved bit map */
	map = &qpt->map[qpt->nmaps];

	rvt_pr_info(rdi, "Reserving QPNs from 0x%x to 0x%x for non-verbs use\n",
		    rdi->dparms.qpn_res_start, rdi->dparms.qpn_res_end);
133
	for (i = rdi->dparms.qpn_res_start; i <= rdi->dparms.qpn_res_end; i++) {
134
		if (!map->page) {
135
			get_map_page(qpt, map, GFP_KERNEL);
136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164
			if (!map->page) {
				ret = -ENOMEM;
				break;
			}
		}
		set_bit(offset, map->page);
		offset++;
		if (offset == RVT_BITS_PER_PAGE) {
			/* next page */
			qpt->nmaps++;
			map++;
			offset = 0;
		}
	}
	return ret;
}

/**
 * free_qpn_table - free the QP number table for a device
 * @qpt: the QPN table
 */
static void free_qpn_table(struct rvt_qpn_table *qpt)
{
	int i;

	for (i = 0; i < ARRAY_SIZE(qpt->map); i++)
		free_page((unsigned long)qpt->map[i].page);
}

165 166 167 168 169 170
/**
 * rvt_driver_qp_init - Init driver qp resources
 * @rdi: rvt dev strucutre
 *
 * Return: 0 on success
 */
171 172 173 174 175 176 177 178 179 180 181 182
int rvt_driver_qp_init(struct rvt_dev_info *rdi)
{
	int i;
	int ret = -ENOMEM;

	if (!rdi->dparms.qp_table_size)
		return -EINVAL;

	/*
	 * If driver is not doing any QP allocation then make sure it is
	 * providing the necessary QP functions.
	 */
183 184 185 186
	if (!rdi->driver_f.free_all_qps ||
	    !rdi->driver_f.qp_priv_alloc ||
	    !rdi->driver_f.qp_priv_free ||
	    !rdi->driver_f.notify_qp_reset)
187 188 189
		return -EINVAL;

	/* allocate parent object */
190 191
	rdi->qp_dev = kzalloc_node(sizeof(*rdi->qp_dev), GFP_KERNEL,
				   rdi->dparms.node);
192 193 194 195 196 197 198
	if (!rdi->qp_dev)
		return -ENOMEM;

	/* allocate hash table */
	rdi->qp_dev->qp_table_size = rdi->dparms.qp_table_size;
	rdi->qp_dev->qp_table_bits = ilog2(rdi->dparms.qp_table_size);
	rdi->qp_dev->qp_table =
199 200 201
		kmalloc_node(rdi->qp_dev->qp_table_size *
			     sizeof(*rdi->qp_dev->qp_table),
			     GFP_KERNEL, rdi->dparms.node);
202 203 204 205 206 207 208 209 210 211 212 213
	if (!rdi->qp_dev->qp_table)
		goto no_qp_table;

	for (i = 0; i < rdi->qp_dev->qp_table_size; i++)
		RCU_INIT_POINTER(rdi->qp_dev->qp_table[i], NULL);

	spin_lock_init(&rdi->qp_dev->qpt_lock);

	/* initialize qpn map */
	if (init_qpn_table(rdi, &rdi->qp_dev->qpn_table))
		goto fail_table;

214 215 216
	spin_lock_init(&rdi->n_qps_lock);

	return 0;
217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234

fail_table:
	kfree(rdi->qp_dev->qp_table);
	free_qpn_table(&rdi->qp_dev->qpn_table);

no_qp_table:
	kfree(rdi->qp_dev);

	return ret;
}

/**
 * free_all_qps - check for QPs still in use
 * @qpt: the QP table to empty
 *
 * There should not be any QPs still in use.
 * Free memory for table.
 */
235
static unsigned rvt_free_all_qps(struct rvt_dev_info *rdi)
236 237 238 239 240 241
{
	unsigned long flags;
	struct rvt_qp *qp;
	unsigned n, qp_inuse = 0;
	spinlock_t *ql; /* work around too long line below */

242 243
	if (rdi->driver_f.free_all_qps)
		qp_inuse = rdi->driver_f.free_all_qps(rdi);
244

245 246
	qp_inuse += rvt_mcast_tree_empty(rdi);

247
	if (!rdi->qp_dev)
248
		return qp_inuse;
249 250

	ql = &rdi->qp_dev->qpt_lock;
251
	spin_lock_irqsave(ql, flags);
252 253 254 255
	for (n = 0; n < rdi->qp_dev->qp_table_size; n++) {
		qp = rcu_dereference_protected(rdi->qp_dev->qp_table[n],
					       lockdep_is_held(ql));
		RCU_INIT_POINTER(rdi->qp_dev->qp_table[n], NULL);
256 257 258

		for (; qp; qp = rcu_dereference_protected(qp->next,
							  lockdep_is_held(ql)))
259 260 261 262 263 264 265
			qp_inuse++;
	}
	spin_unlock_irqrestore(ql, flags);
	synchronize_rcu();
	return qp_inuse;
}

266 267 268 269 270 271
/**
 * rvt_qp_exit - clean up qps on device exit
 * @rdi: rvt dev structure
 *
 * Check for qp leaks and free resources.
 */
272 273
void rvt_qp_exit(struct rvt_dev_info *rdi)
{
274
	u32 qps_inuse = rvt_free_all_qps(rdi);
275 276 277 278 279 280 281 282 283 284 285 286

	if (qps_inuse)
		rvt_pr_err(rdi, "QP memory leak! %u still in use\n",
			   qps_inuse);
	if (!rdi->qp_dev)
		return;

	kfree(rdi->qp_dev->qp_table);
	free_qpn_table(&rdi->qp_dev->qpn_table);
	kfree(rdi->qp_dev);
}

287 288 289 290 291 292
static inline unsigned mk_qpn(struct rvt_qpn_table *qpt,
			      struct rvt_qpn_map *map, unsigned off)
{
	return (map - qpt->map) * RVT_BITS_PER_PAGE + off;
}

293 294 295 296 297 298 299 300
/**
 * alloc_qpn - Allocate the next available qpn or zero/one for QP type
 *	       IB_QPT_SMI/IB_QPT_GSI
 *@rdi:	rvt device info structure
 *@qpt: queue pair number table pointer
 *@port_num: IB port number, 1 based, comes from core
 *
 * Return: The queue pair number
301 302
 */
static int alloc_qpn(struct rvt_dev_info *rdi, struct rvt_qpn_table *qpt,
303
		     enum ib_qp_type type, u8 port_num, gfp_t gfp)
304 305 306 307 308 309
{
	u32 i, offset, max_scan, qpn;
	struct rvt_qpn_map *map;
	u32 ret;

	if (rdi->driver_f.alloc_qpn)
310
		return rdi->driver_f.alloc_qpn(rdi, qpt, type, port_num, gfp);
311 312 313 314 315

	if (type == IB_QPT_SMI || type == IB_QPT_GSI) {
		unsigned n;

		ret = type == IB_QPT_GSI;
316
		n = 1 << (ret + 2 * (port_num - 1));
317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334
		spin_lock(&qpt->lock);
		if (qpt->flags & n)
			ret = -EINVAL;
		else
			qpt->flags |= n;
		spin_unlock(&qpt->lock);
		goto bail;
	}

	qpn = qpt->last + qpt->incr;
	if (qpn >= RVT_QPN_MAX)
		qpn = qpt->incr | ((qpt->last & 1) ^ 1);
	/* offset carries bit 0 */
	offset = qpn & RVT_BITS_PER_PAGE_MASK;
	map = &qpt->map[qpn / RVT_BITS_PER_PAGE];
	max_scan = qpt->nmaps - !offset;
	for (i = 0;;) {
		if (unlikely(!map->page)) {
335
			get_map_page(qpt, map, gfp);
336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371
			if (unlikely(!map->page))
				break;
		}
		do {
			if (!test_and_set_bit(offset, map->page)) {
				qpt->last = qpn;
				ret = qpn;
				goto bail;
			}
			offset += qpt->incr;
			/*
			 * This qpn might be bogus if offset >= BITS_PER_PAGE.
			 * That is OK.   It gets re-assigned below
			 */
			qpn = mk_qpn(qpt, map, offset);
		} while (offset < RVT_BITS_PER_PAGE && qpn < RVT_QPN_MAX);
		/*
		 * In order to keep the number of pages allocated to a
		 * minimum, we scan the all existing pages before increasing
		 * the size of the bitmap table.
		 */
		if (++i > max_scan) {
			if (qpt->nmaps == RVT_QPNMAP_ENTRIES)
				break;
			map = &qpt->map[qpt->nmaps++];
			/* start at incr with current bit 0 */
			offset = qpt->incr | (offset & 1);
		} else if (map < &qpt->map[qpt->nmaps]) {
			++map;
			/* start at incr with current bit 0 */
			offset = qpt->incr | (offset & 1);
		} else {
			map = &qpt->map[0];
			/* wrap to first map page, invert bit 0 */
			offset = qpt->incr | ((offset & 1) ^ 1);
		}
372 373
		/* there can be no set bits in low-order QoS bits */
		WARN_ON(offset & (BIT(rdi->dparms.qos_shift) - 1));
374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391
		qpn = mk_qpn(qpt, map, offset);
	}

	ret = -ENOMEM;

bail:
	return ret;
}

static void free_qpn(struct rvt_qpn_table *qpt, u32 qpn)
{
	struct rvt_qpn_map *map;

	map = qpt->map + qpn / RVT_BITS_PER_PAGE;
	if (map->page)
		clear_bit(qpn & RVT_BITS_PER_PAGE_MASK, map->page);
}

392 393 394 395 396 397 398 399
/**
 * rvt_clear_mr_refs - Drop help mr refs
 * @qp: rvt qp data structure
 * @clr_sends: If shoudl clear send side or not
 */
static void rvt_clear_mr_refs(struct rvt_qp *qp, int clr_sends)
{
	unsigned n;
400
	struct rvt_dev_info *rdi = ib_to_rvt(qp->ibqp.device);
401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434

	if (test_and_clear_bit(RVT_R_REWIND_SGE, &qp->r_aflags))
		rvt_put_ss(&qp->s_rdma_read_sge);

	rvt_put_ss(&qp->r_sge);

	if (clr_sends) {
		while (qp->s_last != qp->s_head) {
			struct rvt_swqe *wqe = rvt_get_swqe_ptr(qp, qp->s_last);
			unsigned i;

			for (i = 0; i < wqe->wr.num_sge; i++) {
				struct rvt_sge *sge = &wqe->sg_list[i];

				rvt_put_mr(sge->mr);
			}
			if (qp->ibqp.qp_type == IB_QPT_UD ||
			    qp->ibqp.qp_type == IB_QPT_SMI ||
			    qp->ibqp.qp_type == IB_QPT_GSI)
				atomic_dec(&ibah_to_rvtah(
						wqe->ud_wr.ah)->refcount);
			if (++qp->s_last >= qp->s_size)
				qp->s_last = 0;
			smp_wmb(); /* see qp_set_savail */
		}
		if (qp->s_rdma_mr) {
			rvt_put_mr(qp->s_rdma_mr);
			qp->s_rdma_mr = NULL;
		}
	}

	if (qp->ibqp.qp_type != IB_QPT_RC)
		return;

435
	for (n = 0; n < rvt_max_atomic(rdi); n++) {
436 437
		struct rvt_ack_entry *e = &qp->s_ack_queue[n];

438
		if (e->rdma_sge.mr) {
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
			rvt_put_mr(e->rdma_sge.mr);
			e->rdma_sge.mr = NULL;
		}
	}
}

/**
 * rvt_remove_qp - remove qp form table
 * @rdi: rvt dev struct
 * @qp: qp to remove
 *
 * Remove the QP from the table so it can't be found asynchronously by
 * the receive routine.
 */
static void rvt_remove_qp(struct rvt_dev_info *rdi, struct rvt_qp *qp)
{
	struct rvt_ibport *rvp = rdi->ports[qp->port_num - 1];
	u32 n = hash_32(qp->ibqp.qp_num, rdi->qp_dev->qp_table_bits);
	unsigned long flags;
	int removed = 1;

	spin_lock_irqsave(&rdi->qp_dev->qpt_lock, flags);

	if (rcu_dereference_protected(rvp->qp[0],
			lockdep_is_held(&rdi->qp_dev->qpt_lock)) == qp) {
		RCU_INIT_POINTER(rvp->qp[0], NULL);
	} else if (rcu_dereference_protected(rvp->qp[1],
			lockdep_is_held(&rdi->qp_dev->qpt_lock)) == qp) {
		RCU_INIT_POINTER(rvp->qp[1], NULL);
	} else {
		struct rvt_qp *q;
		struct rvt_qp __rcu **qpp;

		removed = 0;
		qpp = &rdi->qp_dev->qp_table[n];
		for (; (q = rcu_dereference_protected(*qpp,
			lockdep_is_held(&rdi->qp_dev->qpt_lock))) != NULL;
			qpp = &q->next) {
			if (q == qp) {
				RCU_INIT_POINTER(*qpp,
				     rcu_dereference_protected(qp->next,
				     lockdep_is_held(&rdi->qp_dev->qpt_lock)));
				removed = 1;
				trace_rvt_qpremove(qp, n);
				break;
			}
		}
	}

	spin_unlock_irqrestore(&rdi->qp_dev->qpt_lock, flags);
	if (removed) {
		synchronize_rcu();
		if (atomic_dec_and_test(&qp->refcount))
			wake_up(&qp->wait);
	}
}

496 497 498 499
/**
 * reset_qp - initialize the QP state to the reset state
 * @qp: the QP to reset
 * @type: the QP type
500
 * r and s lock are required to be held by the caller
501
 */
502
static void rvt_reset_qp(struct rvt_dev_info *rdi, struct rvt_qp *qp,
503
		  enum ib_qp_type type)
504 505 506 507 508 509
	__releases(&qp->s_lock)
	__releases(&qp->s_hlock)
	__releases(&qp->r_lock)
	__acquires(&qp->r_lock)
	__acquires(&qp->s_hlock)
	__acquires(&qp->s_lock)
510
{
511 512 513 514 515 516 517
	if (qp->state != IB_QPS_RESET) {
		qp->state = IB_QPS_RESET;

		/* Let drivers flush their waitlist */
		rdi->driver_f.flush_qp_waiters(qp);
		qp->s_flags &= ~(RVT_S_TIMER | RVT_S_ANY_WAIT);
		spin_unlock(&qp->s_lock);
518
		spin_unlock(&qp->s_hlock);
519 520 521 522 523 524 525 526 527 528 529 530 531 532
		spin_unlock_irq(&qp->r_lock);

		/* Stop the send queue and the retry timer */
		rdi->driver_f.stop_send_queue(qp);

		/* Wait for things to stop */
		rdi->driver_f.quiesce_qp(qp);

		/* take qp out the hash and wait for it to be unused */
		rvt_remove_qp(rdi, qp);
		wait_event(qp->wait, !atomic_read(&qp->refcount));

		/* grab the lock b/c it was locked at call time */
		spin_lock_irq(&qp->r_lock);
533
		spin_lock(&qp->s_hlock);
534 535 536 537
		spin_lock(&qp->s_lock);

		rvt_clear_mr_refs(qp, 1);
	}
538 539

	/*
540 541
	 * Let the driver do any tear down it needs to for a qp
	 * that has been reset
542 543 544
	 */
	rdi->driver_f.notify_qp_reset(qp);

545 546 547
	qp->remote_qpn = 0;
	qp->qkey = 0;
	qp->qp_access_flags = 0;
548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585
	qp->s_flags &= RVT_S_SIGNAL_REQ_WR;
	qp->s_hdrwords = 0;
	qp->s_wqe = NULL;
	qp->s_draining = 0;
	qp->s_next_psn = 0;
	qp->s_last_psn = 0;
	qp->s_sending_psn = 0;
	qp->s_sending_hpsn = 0;
	qp->s_psn = 0;
	qp->r_psn = 0;
	qp->r_msn = 0;
	if (type == IB_QPT_RC) {
		qp->s_state = IB_OPCODE_RC_SEND_LAST;
		qp->r_state = IB_OPCODE_RC_SEND_LAST;
	} else {
		qp->s_state = IB_OPCODE_UC_SEND_LAST;
		qp->r_state = IB_OPCODE_UC_SEND_LAST;
	}
	qp->s_ack_state = IB_OPCODE_RC_ACKNOWLEDGE;
	qp->r_nak_state = 0;
	qp->r_aflags = 0;
	qp->r_flags = 0;
	qp->s_head = 0;
	qp->s_tail = 0;
	qp->s_cur = 0;
	qp->s_acked = 0;
	qp->s_last = 0;
	qp->s_ssn = 1;
	qp->s_lsn = 0;
	qp->s_mig_state = IB_MIG_MIGRATED;
	qp->r_head_ack_queue = 0;
	qp->s_tail_ack_queue = 0;
	qp->s_num_rd_atomic = 0;
	if (qp->r_rq.wq) {
		qp->r_rq.wq->head = 0;
		qp->r_rq.wq->tail = 0;
	}
	qp->r_sge.num_sge = 0;
586
	atomic_set(&qp->s_reserved_used, 0);
587 588
}

589 590 591 592 593 594
/**
 * rvt_create_qp - create a queue pair for a device
 * @ibpd: the protection domain who's device we create the queue pair for
 * @init_attr: the attributes of the queue pair
 * @udata: user data for libibverbs.so
 *
595 596 597 598
 * Queue pair creation is mostly an rvt issue. However, drivers have their own
 * unique idea of what queue pair numbers mean. For instance there is a reserved
 * range for PSM.
 *
599
 * Return: the queue pair on success, otherwise returns an errno.
600 601 602 603 604 605 606
 *
 * Called by the ib_create_qp() core verbs function.
 */
struct ib_qp *rvt_create_qp(struct ib_pd *ibpd,
			    struct ib_qp_init_attr *init_attr,
			    struct ib_udata *udata)
{
607 608 609 610 611 612 613 614
	struct rvt_qp *qp;
	int err;
	struct rvt_swqe *swq = NULL;
	size_t sz;
	size_t sg_list_sz;
	struct ib_qp *ret = ERR_PTR(-ENOMEM);
	struct rvt_dev_info *rdi = ib_to_rvt(ibpd->device);
	void *priv = NULL;
615
	gfp_t gfp;
616
	size_t sqsize;
617 618 619 620 621 622

	if (!rdi)
		return ERR_PTR(-EINVAL);

	if (init_attr->cap.max_send_sge > rdi->dparms.props.max_sge ||
	    init_attr->cap.max_send_wr > rdi->dparms.props.max_qp_wr ||
623
	    init_attr->create_flags & ~(IB_QP_CREATE_USE_GFP_NOIO))
624 625
		return ERR_PTR(-EINVAL);

626 627 628 629 630 631 632 633 634
	/* GFP_NOIO is applicable to RC QP's only */

	if (init_attr->create_flags & IB_QP_CREATE_USE_GFP_NOIO &&
	    init_attr->qp_type != IB_QPT_RC)
		return ERR_PTR(-EINVAL);

	gfp = init_attr->create_flags & IB_QP_CREATE_USE_GFP_NOIO ?
						GFP_NOIO : GFP_KERNEL;

635 636 637 638 639 640 641 642 643 644 645 646
	/* Check receive queue parameters if no SRQ is specified. */
	if (!init_attr->srq) {
		if (init_attr->cap.max_recv_sge > rdi->dparms.props.max_sge ||
		    init_attr->cap.max_recv_wr > rdi->dparms.props.max_qp_wr)
			return ERR_PTR(-EINVAL);

		if (init_attr->cap.max_send_sge +
		    init_attr->cap.max_send_wr +
		    init_attr->cap.max_recv_sge +
		    init_attr->cap.max_recv_wr == 0)
			return ERR_PTR(-EINVAL);
	}
647
	sqsize =
648 649
		init_attr->cap.max_send_wr + 1 +
		rdi->dparms.reserved_operations;
650 651 652 653 654 655 656 657 658 659 660 661
	switch (init_attr->qp_type) {
	case IB_QPT_SMI:
	case IB_QPT_GSI:
		if (init_attr->port_num == 0 ||
		    init_attr->port_num > ibpd->device->phys_port_cnt)
			return ERR_PTR(-EINVAL);
	case IB_QPT_UC:
	case IB_QPT_RC:
	case IB_QPT_UD:
		sz = sizeof(struct rvt_sge) *
			init_attr->cap.max_send_sge +
			sizeof(struct rvt_swqe);
662 663
		if (gfp == GFP_NOIO)
			swq = __vmalloc(
664
				sqsize * sz,
665
				gfp | __GFP_ZERO, PAGE_KERNEL);
666
		else
667
			swq = vzalloc_node(
668
				sqsize * sz,
669
				rdi->dparms.node);
670 671 672 673 674 675 676 677 678 679 680 681 682 683
		if (!swq)
			return ERR_PTR(-ENOMEM);

		sz = sizeof(*qp);
		sg_list_sz = 0;
		if (init_attr->srq) {
			struct rvt_srq *srq = ibsrq_to_rvtsrq(init_attr->srq);

			if (srq->rq.max_sge > 1)
				sg_list_sz = sizeof(*qp->r_sg_list) *
					(srq->rq.max_sge - 1);
		} else if (init_attr->cap.max_recv_sge > 1)
			sg_list_sz = sizeof(*qp->r_sg_list) *
				(init_attr->cap.max_recv_sge - 1);
684
		qp = kzalloc_node(sz + sg_list_sz, gfp, rdi->dparms.node);
685 686 687 688
		if (!qp)
			goto bail_swq;

		RCU_INIT_POINTER(qp->next, NULL);
689 690 691 692 693 694 695 696 697 698
		if (init_attr->qp_type == IB_QPT_RC) {
			qp->s_ack_queue =
				kzalloc_node(
					sizeof(*qp->s_ack_queue) *
					 rvt_max_atomic(rdi),
					gfp,
					rdi->dparms.node);
			if (!qp->s_ack_queue)
				goto bail_qp;
		}
699 700 701 702 703

		/*
		 * Driver needs to set up it's private QP structure and do any
		 * initialization that is needed.
		 */
704
		priv = rdi->driver_f.qp_priv_alloc(rdi, qp, gfp);
705 706
		if (IS_ERR(priv)) {
			ret = priv;
707
			goto bail_qp;
708
		}
709 710 711 712 713 714 715 716 717 718 719
		qp->priv = priv;
		qp->timeout_jiffies =
			usecs_to_jiffies((4096UL * (1UL << qp->timeout)) /
				1000UL);
		if (init_attr->srq) {
			sz = 0;
		} else {
			qp->r_rq.size = init_attr->cap.max_recv_wr + 1;
			qp->r_rq.max_sge = init_attr->cap.max_recv_sge;
			sz = (sizeof(struct ib_sge) * qp->r_rq.max_sge) +
				sizeof(struct rvt_rwqe);
720 721 722 723 724 725 726 727
			if (udata)
				qp->r_rq.wq = vmalloc_user(
						sizeof(struct rvt_rwq) +
						qp->r_rq.size * sz);
			else if (gfp == GFP_NOIO)
				qp->r_rq.wq = __vmalloc(
						sizeof(struct rvt_rwq) +
						qp->r_rq.size * sz,
728
						gfp | __GFP_ZERO, PAGE_KERNEL);
729
			else
730
				qp->r_rq.wq = vzalloc_node(
731
						sizeof(struct rvt_rwq) +
732 733
						qp->r_rq.size * sz,
						rdi->dparms.node);
734 735 736 737 738 739 740 741 742
			if (!qp->r_rq.wq)
				goto bail_driver_priv;
		}

		/*
		 * ib_create_qp() will initialize qp->ibqp
		 * except for qp->ibqp.qp_num.
		 */
		spin_lock_init(&qp->r_lock);
743
		spin_lock_init(&qp->s_hlock);
744 745 746
		spin_lock_init(&qp->s_lock);
		spin_lock_init(&qp->r_rq.lock);
		atomic_set(&qp->refcount, 0);
747
		atomic_set(&qp->local_ops_pending, 0);
748 749 750 751 752 753
		init_waitqueue_head(&qp->wait);
		init_timer(&qp->s_timer);
		qp->s_timer.data = (unsigned long)qp;
		INIT_LIST_HEAD(&qp->rspwait);
		qp->state = IB_QPS_RESET;
		qp->s_wq = swq;
754
		qp->s_size = sqsize;
755
		qp->s_avail = init_attr->cap.max_send_wr;
756 757 758 759 760 761
		qp->s_max_sge = init_attr->cap.max_send_sge;
		if (init_attr->sq_sig_type == IB_SIGNAL_REQ_WR)
			qp->s_flags = RVT_S_SIGNAL_REQ_WR;

		err = alloc_qpn(rdi, &rdi->qp_dev->qpn_table,
				init_attr->qp_type,
762
				init_attr->port_num, gfp);
763 764 765 766 767 768
		if (err < 0) {
			ret = ERR_PTR(err);
			goto bail_rq_wq;
		}
		qp->ibqp.qp_num = err;
		qp->port_num = init_attr->port_num;
769
		rvt_reset_qp(rdi, qp, init_attr->qp_type);
770 771 772 773 774 775 776 777 778 779 780
		break;

	default:
		/* Don't support raw QPs */
		return ERR_PTR(-EINVAL);
	}

	init_attr->cap.max_inline_data = 0;

	/*
	 * Return the address of the RWQ as the offset to mmap.
781
	 * See rvt_mmap() for details.
782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810
	 */
	if (udata && udata->outlen >= sizeof(__u64)) {
		if (!qp->r_rq.wq) {
			__u64 offset = 0;

			err = ib_copy_to_udata(udata, &offset,
					       sizeof(offset));
			if (err) {
				ret = ERR_PTR(err);
				goto bail_qpn;
			}
		} else {
			u32 s = sizeof(struct rvt_rwq) + qp->r_rq.size * sz;

			qp->ip = rvt_create_mmap_info(rdi, s,
						      ibpd->uobject->context,
						      qp->r_rq.wq);
			if (!qp->ip) {
				ret = ERR_PTR(-ENOMEM);
				goto bail_qpn;
			}

			err = ib_copy_to_udata(udata, &qp->ip->offset,
					       sizeof(qp->ip->offset));
			if (err) {
				ret = ERR_PTR(err);
				goto bail_ip;
			}
		}
811
		qp->pid = current->pid;
812 813 814 815 816 817 818 819 820 821
	}

	spin_lock(&rdi->n_qps_lock);
	if (rdi->n_qps_allocated == rdi->dparms.props.max_qp) {
		spin_unlock(&rdi->n_qps_lock);
		ret = ERR_PTR(-ENOMEM);
		goto bail_ip;
	}

	rdi->n_qps_allocated++;
822 823 824 825 826 827 828 829 830 831 832 833 834
	/*
	 * Maintain a busy_jiffies variable that will be added to the timeout
	 * period in mod_retry_timer and add_retry_timer. This busy jiffies
	 * is scaled by the number of rc qps created for the device to reduce
	 * the number of timeouts occurring when there is a large number of
	 * qps. busy_jiffies is incremented every rc qp scaling interval.
	 * The scaling interval is selected based on extensive performance
	 * evaluation of targeted workloads.
	 */
	if (init_attr->qp_type == IB_QPT_RC) {
		rdi->n_rc_qps++;
		rdi->busy_jiffies = rdi->n_rc_qps / RC_QP_SCALING_INTERVAL;
	}
835 836 837 838 839 840 841 842 843 844
	spin_unlock(&rdi->n_qps_lock);

	if (qp->ip) {
		spin_lock_irq(&rdi->pending_lock);
		list_add(&qp->ip->pending_mmaps, &rdi->pending_mmaps);
		spin_unlock_irq(&rdi->pending_lock);
	}

	ret = &qp->ibqp;

845
	/*
846 847 848
	 * We have our QP and its good, now keep track of what types of opcodes
	 * can be processed on this QP. We do this by keeping track of what the
	 * 3 high order bits of the opcode are.
849
	 */
850 851 852 853
	switch (init_attr->qp_type) {
	case IB_QPT_SMI:
	case IB_QPT_GSI:
	case IB_QPT_UD:
854
		qp->allowed_ops = IB_OPCODE_UD;
855 856
		break;
	case IB_QPT_RC:
857
		qp->allowed_ops = IB_OPCODE_RC;
858 859
		break;
	case IB_QPT_UC:
860
		qp->allowed_ops = IB_OPCODE_UC;
861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881
		break;
	default:
		ret = ERR_PTR(-EINVAL);
		goto bail_ip;
	}

	return ret;

bail_ip:
	kref_put(&qp->ip->ref, rvt_release_mmap_info);

bail_qpn:
	free_qpn(&rdi->qp_dev->qpn_table, qp->ibqp.qp_num);

bail_rq_wq:
	vfree(qp->r_rq.wq);

bail_driver_priv:
	rdi->driver_f.qp_priv_free(rdi, qp);

bail_qp:
882
	kfree(qp->s_ack_queue);
883 884 885 886 887 888
	kfree(qp);

bail_swq:
	vfree(swq);

	return ret;
889 890
}

891 892 893 894 895 896
/**
 * rvt_error_qp - put a QP into the error state
 * @qp: the QP to put into the error state
 * @err: the receive completion error to signal if a RWQE is active
 *
 * Flushes both send and receive work queues.
897 898
 *
 * Return: true if last WQE event should be generated.
899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923
 * The QP r_lock and s_lock should be held and interrupts disabled.
 * If we are already in error state, just return.
 */
int rvt_error_qp(struct rvt_qp *qp, enum ib_wc_status err)
{
	struct ib_wc wc;
	int ret = 0;
	struct rvt_dev_info *rdi = ib_to_rvt(qp->ibqp.device);

	if (qp->state == IB_QPS_ERR || qp->state == IB_QPS_RESET)
		goto bail;

	qp->state = IB_QPS_ERR;

	if (qp->s_flags & (RVT_S_TIMER | RVT_S_WAIT_RNR)) {
		qp->s_flags &= ~(RVT_S_TIMER | RVT_S_WAIT_RNR);
		del_timer(&qp->s_timer);
	}

	if (qp->s_flags & RVT_S_ANY_WAIT_SEND)
		qp->s_flags &= ~RVT_S_ANY_WAIT_SEND;

	rdi->driver_f.notify_error_qp(qp);

	/* Schedule the sending tasklet to drain the send work queue. */
924
	if (ACCESS_ONCE(qp->s_last) != qp->s_head)
925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997
		rdi->driver_f.schedule_send(qp);

	rvt_clear_mr_refs(qp, 0);

	memset(&wc, 0, sizeof(wc));
	wc.qp = &qp->ibqp;
	wc.opcode = IB_WC_RECV;

	if (test_and_clear_bit(RVT_R_WRID_VALID, &qp->r_aflags)) {
		wc.wr_id = qp->r_wr_id;
		wc.status = err;
		rvt_cq_enter(ibcq_to_rvtcq(qp->ibqp.recv_cq), &wc, 1);
	}
	wc.status = IB_WC_WR_FLUSH_ERR;

	if (qp->r_rq.wq) {
		struct rvt_rwq *wq;
		u32 head;
		u32 tail;

		spin_lock(&qp->r_rq.lock);

		/* sanity check pointers before trusting them */
		wq = qp->r_rq.wq;
		head = wq->head;
		if (head >= qp->r_rq.size)
			head = 0;
		tail = wq->tail;
		if (tail >= qp->r_rq.size)
			tail = 0;
		while (tail != head) {
			wc.wr_id = rvt_get_rwqe_ptr(&qp->r_rq, tail)->wr_id;
			if (++tail >= qp->r_rq.size)
				tail = 0;
			rvt_cq_enter(ibcq_to_rvtcq(qp->ibqp.recv_cq), &wc, 1);
		}
		wq->tail = tail;

		spin_unlock(&qp->r_rq.lock);
	} else if (qp->ibqp.event_handler) {
		ret = 1;
	}

bail:
	return ret;
}
EXPORT_SYMBOL(rvt_error_qp);

/*
 * Put the QP into the hash table.
 * The hash table holds a reference to the QP.
 */
static void rvt_insert_qp(struct rvt_dev_info *rdi, struct rvt_qp *qp)
{
	struct rvt_ibport *rvp = rdi->ports[qp->port_num - 1];
	unsigned long flags;

	atomic_inc(&qp->refcount);
	spin_lock_irqsave(&rdi->qp_dev->qpt_lock, flags);

	if (qp->ibqp.qp_num <= 1) {
		rcu_assign_pointer(rvp->qp[qp->ibqp.qp_num], qp);
	} else {
		u32 n = hash_32(qp->ibqp.qp_num, rdi->qp_dev->qp_table_bits);

		qp->next = rdi->qp_dev->qp_table[n];
		rcu_assign_pointer(rdi->qp_dev->qp_table[n], qp);
		trace_rvt_qpinsert(qp, n);
	}

	spin_unlock_irqrestore(&rdi->qp_dev->qpt_lock, flags);
}

998 999 1000 1001 1002 1003 1004
/**
 * qib_modify_qp - modify the attributes of a queue pair
 * @ibqp: the queue pair who's attributes we're modifying
 * @attr: the new attributes
 * @attr_mask: the mask of attributes to modify
 * @udata: user data for libibverbs.so
 *
1005
 * Return: 0 on success, otherwise returns an errno.
1006 1007 1008 1009
 */
int rvt_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
		  int attr_mask, struct ib_udata *udata)
{
1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021
	struct rvt_dev_info *rdi = ib_to_rvt(ibqp->device);
	struct rvt_qp *qp = ibqp_to_rvtqp(ibqp);
	enum ib_qp_state cur_state, new_state;
	struct ib_event ev;
	int lastwqe = 0;
	int mig = 0;
	int pmtu = 0; /* for gcc warning only */
	enum rdma_link_layer link;

	link = rdma_port_get_link_layer(ibqp->device, qp->port_num);

	spin_lock_irq(&qp->r_lock);
1022
	spin_lock(&qp->s_hlock);
1023 1024 1025 1026 1027 1028 1029 1030 1031 1032
	spin_lock(&qp->s_lock);

	cur_state = attr_mask & IB_QP_CUR_STATE ?
		attr->cur_qp_state : 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, link))
		goto inval;

1033 1034 1035 1036
	if (rdi->driver_f.check_modify_qp &&
	    rdi->driver_f.check_modify_qp(qp, attr, attr_mask, udata))
		goto inval;

1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080
	if (attr_mask & IB_QP_AV) {
		if (attr->ah_attr.dlid >= be16_to_cpu(IB_MULTICAST_LID_BASE))
			goto inval;
		if (rvt_check_ah(qp->ibqp.device, &attr->ah_attr))
			goto inval;
	}

	if (attr_mask & IB_QP_ALT_PATH) {
		if (attr->alt_ah_attr.dlid >=
		    be16_to_cpu(IB_MULTICAST_LID_BASE))
			goto inval;
		if (rvt_check_ah(qp->ibqp.device, &attr->alt_ah_attr))
			goto inval;
		if (attr->alt_pkey_index >= rvt_get_npkeys(rdi))
			goto inval;
	}

	if (attr_mask & IB_QP_PKEY_INDEX)
		if (attr->pkey_index >= rvt_get_npkeys(rdi))
			goto inval;

	if (attr_mask & IB_QP_MIN_RNR_TIMER)
		if (attr->min_rnr_timer > 31)
			goto inval;

	if (attr_mask & IB_QP_PORT)
		if (qp->ibqp.qp_type == IB_QPT_SMI ||
		    qp->ibqp.qp_type == IB_QPT_GSI ||
		    attr->port_num == 0 ||
		    attr->port_num > ibqp->device->phys_port_cnt)
			goto inval;

	if (attr_mask & IB_QP_DEST_QPN)
		if (attr->dest_qp_num > RVT_QPN_MASK)
			goto inval;

	if (attr_mask & IB_QP_RETRY_CNT)
		if (attr->retry_cnt > 7)
			goto inval;

	if (attr_mask & IB_QP_RNR_RETRY)
		if (attr->rnr_retry > 7)
			goto inval;

1081
	/*
1082 1083 1084 1085 1086 1087
	 * Don't allow invalid path_mtu values.  OK to set greater
	 * than the active mtu (or even the max_cap, if we have tuned
	 * that to a small mtu.  We'll set qp->path_mtu
	 * to the lesser of requested attribute mtu and active,
	 * for packetizing messages.
	 * Note that the QP port has to be set in INIT and MTU in RTR.
1088
	 */
1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 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 1194
	if (attr_mask & IB_QP_PATH_MTU) {
		pmtu = rdi->driver_f.get_pmtu_from_attr(rdi, qp, attr);
		if (pmtu < 0)
			goto inval;
	}

	if (attr_mask & IB_QP_PATH_MIG_STATE) {
		if (attr->path_mig_state == IB_MIG_REARM) {
			if (qp->s_mig_state == IB_MIG_ARMED)
				goto inval;
			if (new_state != IB_QPS_RTS)
				goto inval;
		} else if (attr->path_mig_state == IB_MIG_MIGRATED) {
			if (qp->s_mig_state == IB_MIG_REARM)
				goto inval;
			if (new_state != IB_QPS_RTS && new_state != IB_QPS_SQD)
				goto inval;
			if (qp->s_mig_state == IB_MIG_ARMED)
				mig = 1;
		} else {
			goto inval;
		}
	}

	if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC)
		if (attr->max_dest_rd_atomic > rdi->dparms.max_rdma_atomic)
			goto inval;

	switch (new_state) {
	case IB_QPS_RESET:
		if (qp->state != IB_QPS_RESET)
			rvt_reset_qp(rdi, qp, ibqp->qp_type);
		break;

	case IB_QPS_RTR:
		/* Allow event to re-trigger if QP set to RTR more than once */
		qp->r_flags &= ~RVT_R_COMM_EST;
		qp->state = new_state;
		break;

	case IB_QPS_SQD:
		qp->s_draining = qp->s_last != qp->s_cur;
		qp->state = new_state;
		break;

	case IB_QPS_SQE:
		if (qp->ibqp.qp_type == IB_QPT_RC)
			goto inval;
		qp->state = new_state;
		break;

	case IB_QPS_ERR:
		lastwqe = rvt_error_qp(qp, IB_WC_WR_FLUSH_ERR);
		break;

	default:
		qp->state = new_state;
		break;
	}

	if (attr_mask & IB_QP_PKEY_INDEX)
		qp->s_pkey_index = attr->pkey_index;

	if (attr_mask & IB_QP_PORT)
		qp->port_num = attr->port_num;

	if (attr_mask & IB_QP_DEST_QPN)
		qp->remote_qpn = attr->dest_qp_num;

	if (attr_mask & IB_QP_SQ_PSN) {
		qp->s_next_psn = attr->sq_psn & rdi->dparms.psn_modify_mask;
		qp->s_psn = qp->s_next_psn;
		qp->s_sending_psn = qp->s_next_psn;
		qp->s_last_psn = qp->s_next_psn - 1;
		qp->s_sending_hpsn = qp->s_last_psn;
	}

	if (attr_mask & IB_QP_RQ_PSN)
		qp->r_psn = attr->rq_psn & rdi->dparms.psn_modify_mask;

	if (attr_mask & IB_QP_ACCESS_FLAGS)
		qp->qp_access_flags = attr->qp_access_flags;

	if (attr_mask & IB_QP_AV) {
		qp->remote_ah_attr = attr->ah_attr;
		qp->s_srate = attr->ah_attr.static_rate;
		qp->srate_mbps = ib_rate_to_mbps(qp->s_srate);
	}

	if (attr_mask & IB_QP_ALT_PATH) {
		qp->alt_ah_attr = attr->alt_ah_attr;
		qp->s_alt_pkey_index = attr->alt_pkey_index;
	}

	if (attr_mask & IB_QP_PATH_MIG_STATE) {
		qp->s_mig_state = attr->path_mig_state;
		if (mig) {
			qp->remote_ah_attr = qp->alt_ah_attr;
			qp->port_num = qp->alt_ah_attr.port_num;
			qp->s_pkey_index = qp->s_alt_pkey_index;
		}
	}

	if (attr_mask & IB_QP_PATH_MTU) {
		qp->pmtu = rdi->driver_f.mtu_from_qp(rdi, qp, pmtu);
		qp->path_mtu = rdi->driver_f.mtu_to_path_mtu(qp->pmtu);
1195
		qp->log_pmtu = ilog2(qp->pmtu);
1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226
	}

	if (attr_mask & IB_QP_RETRY_CNT) {
		qp->s_retry_cnt = attr->retry_cnt;
		qp->s_retry = attr->retry_cnt;
	}

	if (attr_mask & IB_QP_RNR_RETRY) {
		qp->s_rnr_retry_cnt = attr->rnr_retry;
		qp->s_rnr_retry = attr->rnr_retry;
	}

	if (attr_mask & IB_QP_MIN_RNR_TIMER)
		qp->r_min_rnr_timer = attr->min_rnr_timer;

	if (attr_mask & IB_QP_TIMEOUT) {
		qp->timeout = attr->timeout;
		qp->timeout_jiffies =
			usecs_to_jiffies((4096UL * (1UL << qp->timeout)) /
				1000UL);
	}

	if (attr_mask & IB_QP_QKEY)
		qp->qkey = attr->qkey;

	if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC)
		qp->r_max_rd_atomic = attr->max_dest_rd_atomic;

	if (attr_mask & IB_QP_MAX_QP_RD_ATOMIC)
		qp->s_max_rd_atomic = attr->max_rd_atomic;

1227 1228 1229
	if (rdi->driver_f.modify_qp)
		rdi->driver_f.modify_qp(qp, attr, attr_mask, udata);

1230
	spin_unlock(&qp->s_lock);
1231
	spin_unlock(&qp->s_hlock);
1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252
	spin_unlock_irq(&qp->r_lock);

	if (cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT)
		rvt_insert_qp(rdi, qp);

	if (lastwqe) {
		ev.device = qp->ibqp.device;
		ev.element.qp = &qp->ibqp;
		ev.event = IB_EVENT_QP_LAST_WQE_REACHED;
		qp->ibqp.event_handler(&ev, qp->ibqp.qp_context);
	}
	if (mig) {
		ev.device = qp->ibqp.device;
		ev.element.qp = &qp->ibqp;
		ev.event = IB_EVENT_PATH_MIG;
		qp->ibqp.event_handler(&ev, qp->ibqp.qp_context);
	}
	return 0;

inval:
	spin_unlock(&qp->s_lock);
1253
	spin_unlock(&qp->s_hlock);
1254 1255
	spin_unlock_irq(&qp->r_lock);
	return -EINVAL;
1256 1257
}

1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270
/** rvt_free_qpn - Free a qpn from the bit map
 * @qpt: QP table
 * @qpn: queue pair number to free
 */
static void rvt_free_qpn(struct rvt_qpn_table *qpt, u32 qpn)
{
	struct rvt_qpn_map *map;

	map = qpt->map + qpn / RVT_BITS_PER_PAGE;
	if (map->page)
		clear_bit(qpn & RVT_BITS_PER_PAGE_MASK, map->page);
}

1271 1272 1273 1274 1275 1276
/**
 * rvt_destroy_qp - destroy a queue pair
 * @ibqp: the queue pair to destroy
 *
 * Note that this can be called while the QP is actively sending or
 * receiving!
1277 1278
 *
 * Return: 0 on success.
1279 1280 1281
 */
int rvt_destroy_qp(struct ib_qp *ibqp)
{
1282 1283
	struct rvt_qp *qp = ibqp_to_rvtqp(ibqp);
	struct rvt_dev_info *rdi = ib_to_rvt(ibqp->device);
1284

1285
	spin_lock_irq(&qp->r_lock);
1286
	spin_lock(&qp->s_hlock);
1287 1288 1289
	spin_lock(&qp->s_lock);
	rvt_reset_qp(rdi, qp, ibqp->qp_type);
	spin_unlock(&qp->s_lock);
1290
	spin_unlock(&qp->s_hlock);
1291 1292 1293 1294 1295 1296 1297
	spin_unlock_irq(&qp->r_lock);

	/* qpn is now available for use again */
	rvt_free_qpn(&rdi->qp_dev->qpn_table, qp->ibqp.qp_num);

	spin_lock(&rdi->n_qps_lock);
	rdi->n_qps_allocated--;
1298 1299 1300 1301
	if (qp->ibqp.qp_type == IB_QPT_RC) {
		rdi->n_rc_qps--;
		rdi->busy_jiffies = rdi->n_rc_qps / RC_QP_SCALING_INTERVAL;
	}
1302 1303 1304 1305 1306 1307 1308 1309
	spin_unlock(&rdi->n_qps_lock);

	if (qp->ip)
		kref_put(&qp->ip->ref, rvt_release_mmap_info);
	else
		vfree(qp->r_rq.wq);
	vfree(qp->s_wq);
	rdi->driver_f.qp_priv_free(rdi, qp);
1310
	kfree(qp->s_ack_queue);
1311 1312
	kfree(qp);
	return 0;
1313 1314
}

1315 1316 1317 1318 1319 1320 1321 1322 1323
/**
 * rvt_query_qp - query an ipbq
 * @ibqp: IB qp to query
 * @attr: attr struct to fill in
 * @attr_mask: attr mask ignored
 * @init_attr: struct to fill in
 *
 * Return: always 0
 */
1324 1325 1326
int rvt_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
		 int attr_mask, struct ib_qp_init_attr *init_attr)
{
1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338
	struct rvt_qp *qp = ibqp_to_rvtqp(ibqp);
	struct rvt_dev_info *rdi = ib_to_rvt(ibqp->device);

	attr->qp_state = qp->state;
	attr->cur_qp_state = attr->qp_state;
	attr->path_mtu = qp->path_mtu;
	attr->path_mig_state = qp->s_mig_state;
	attr->qkey = qp->qkey;
	attr->rq_psn = qp->r_psn & rdi->dparms.psn_mask;
	attr->sq_psn = qp->s_next_psn & rdi->dparms.psn_mask;
	attr->dest_qp_num = qp->remote_qpn;
	attr->qp_access_flags = qp->qp_access_flags;
1339 1340
	attr->cap.max_send_wr = qp->s_size - 1 -
		rdi->dparms.reserved_operations;
1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373
	attr->cap.max_recv_wr = qp->ibqp.srq ? 0 : qp->r_rq.size - 1;
	attr->cap.max_send_sge = qp->s_max_sge;
	attr->cap.max_recv_sge = qp->r_rq.max_sge;
	attr->cap.max_inline_data = 0;
	attr->ah_attr = qp->remote_ah_attr;
	attr->alt_ah_attr = qp->alt_ah_attr;
	attr->pkey_index = qp->s_pkey_index;
	attr->alt_pkey_index = qp->s_alt_pkey_index;
	attr->en_sqd_async_notify = 0;
	attr->sq_draining = qp->s_draining;
	attr->max_rd_atomic = qp->s_max_rd_atomic;
	attr->max_dest_rd_atomic = qp->r_max_rd_atomic;
	attr->min_rnr_timer = qp->r_min_rnr_timer;
	attr->port_num = qp->port_num;
	attr->timeout = qp->timeout;
	attr->retry_cnt = qp->s_retry_cnt;
	attr->rnr_retry = qp->s_rnr_retry_cnt;
	attr->alt_port_num = qp->alt_ah_attr.port_num;
	attr->alt_timeout = qp->alt_timeout;

	init_attr->event_handler = qp->ibqp.event_handler;
	init_attr->qp_context = qp->ibqp.qp_context;
	init_attr->send_cq = qp->ibqp.send_cq;
	init_attr->recv_cq = qp->ibqp.recv_cq;
	init_attr->srq = qp->ibqp.srq;
	init_attr->cap = attr->cap;
	if (qp->s_flags & RVT_S_SIGNAL_REQ_WR)
		init_attr->sq_sig_type = IB_SIGNAL_REQ_WR;
	else
		init_attr->sq_sig_type = IB_SIGNAL_ALL_WR;
	init_attr->qp_type = qp->ibqp.qp_type;
	init_attr->port_num = qp->port_num;
	return 0;
1374
}
1375 1376 1377 1378 1379 1380 1381 1382

/**
 * rvt_post_receive - post a receive on a QP
 * @ibqp: the QP to post the receive on
 * @wr: the WR to post
 * @bad_wr: the first bad WR is put here
 *
 * This may be called from interrupt context.
1383 1384
 *
 * Return: 0 on success otherwise errno
1385 1386 1387 1388
 */
int rvt_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
		  struct ib_recv_wr **bad_wr)
{
1389 1390 1391
	struct rvt_qp *qp = ibqp_to_rvtqp(ibqp);
	struct rvt_rwq *wq = qp->r_rq.wq;
	unsigned long flags;
1392 1393
	int qp_err_flush = (ib_rvt_state_ops[qp->state] & RVT_FLUSH_RECV) &&
				!qp->ibqp.srq;
1394

1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419
	/* Check that state is OK to post receive. */
	if (!(ib_rvt_state_ops[qp->state] & RVT_POST_RECV_OK) || !wq) {
		*bad_wr = wr;
		return -EINVAL;
	}

	for (; wr; wr = wr->next) {
		struct rvt_rwqe *wqe;
		u32 next;
		int i;

		if ((unsigned)wr->num_sge > qp->r_rq.max_sge) {
			*bad_wr = wr;
			return -EINVAL;
		}

		spin_lock_irqsave(&qp->r_rq.lock, flags);
		next = wq->head + 1;
		if (next >= qp->r_rq.size)
			next = 0;
		if (next == wq->tail) {
			spin_unlock_irqrestore(&qp->r_rq.lock, flags);
			*bad_wr = wr;
			return -ENOMEM;
		}
1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441
		if (unlikely(qp_err_flush)) {
			struct ib_wc wc;

			memset(&wc, 0, sizeof(wc));
			wc.qp = &qp->ibqp;
			wc.opcode = IB_WC_RECV;
			wc.wr_id = wr->wr_id;
			wc.status = IB_WC_WR_FLUSH_ERR;
			rvt_cq_enter(ibcq_to_rvtcq(qp->ibqp.recv_cq), &wc, 1);
		} else {
			wqe = rvt_get_rwqe_ptr(&qp->r_rq, wq->head);
			wqe->wr_id = wr->wr_id;
			wqe->num_sge = wr->num_sge;
			for (i = 0; i < wr->num_sge; i++)
				wqe->sg_list[i] = wr->sg_list[i];
			/*
			 * Make sure queue entry is written
			 * before the head index.
			 */
			smp_wmb();
			wq->head = next;
		}
1442 1443 1444
		spin_unlock_irqrestore(&qp->r_rq.lock, flags);
	}
	return 0;
1445 1446
}

1447
/**
1448 1449 1450 1451
 * rvt_qp_valid_operation - validate post send wr request
 * @qp - the qp
 * @post-parms - the post send table for the driver
 * @wr - the work request
1452
 *
1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498
 * The routine validates the operation based on the
 * validation table an returns the length of the operation
 * which can extend beyond the ib_send_bw.  Operation
 * dependent flags key atomic operation validation.
 *
 * There is an exception for UD qps that validates the pd and
 * overrides the length to include the additional UD specific
 * length.
 *
 * Returns a negative error or the length of the work request
 * for building the swqe.
 */
static inline int rvt_qp_valid_operation(
	struct rvt_qp *qp,
	const struct rvt_operation_params *post_parms,
	struct ib_send_wr *wr)
{
	int len;

	if (wr->opcode >= RVT_OPERATION_MAX || !post_parms[wr->opcode].length)
		return -EINVAL;
	if (!(post_parms[wr->opcode].qpt_support & BIT(qp->ibqp.qp_type)))
		return -EINVAL;
	if ((post_parms[wr->opcode].flags & RVT_OPERATION_PRIV) &&
	    ibpd_to_rvtpd(qp->ibqp.pd)->user)
		return -EINVAL;
	if (post_parms[wr->opcode].flags & RVT_OPERATION_ATOMIC_SGE &&
	    (wr->num_sge == 0 ||
	     wr->sg_list[0].length < sizeof(u64) ||
	     wr->sg_list[0].addr & (sizeof(u64) - 1)))
		return -EINVAL;
	if (post_parms[wr->opcode].flags & RVT_OPERATION_ATOMIC &&
	    !qp->s_max_rd_atomic)
		return -EINVAL;
	len = post_parms[wr->opcode].length;
	/* UD specific */
	if (qp->ibqp.qp_type != IB_QPT_UC &&
	    qp->ibqp.qp_type != IB_QPT_RC) {
		if (qp->ibqp.pd != ud_wr(wr)->ah->pd)
			return -EINVAL;
		len = sizeof(struct ib_ud_wr);
	}
	return len;
}

/**
1499
 * rvt_qp_is_avail - determine queue capacity
1500
 * @qp - the qp
1501 1502
 * @rdi - the rdmavt device
 * @reserved_op - is reserved operation
1503 1504 1505
 *
 * This assumes the s_hlock is held but the s_last
 * qp variable is uncontrolled.
1506
 *
1507 1508 1509 1510
 * For non reserved operations, the qp->s_avail
 * may be changed.
 *
 * The return value is zero or a -ENOMEM.
1511
 */
1512 1513 1514 1515
static inline int rvt_qp_is_avail(
	struct rvt_qp *qp,
	struct rvt_dev_info *rdi,
	bool reserved_op)
1516 1517
{
	u32 slast;
1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533
	u32 avail;
	u32 reserved_used;

	/* see rvt_qp_wqe_unreserve() */
	smp_mb__before_atomic();
	reserved_used = atomic_read(&qp->s_reserved_used);
	if (unlikely(reserved_op)) {
		/* see rvt_qp_wqe_unreserve() */
		smp_mb__before_atomic();
		if (reserved_used >= rdi->dparms.reserved_operations)
			return -ENOMEM;
		return 0;
	}
	/* non-reserved operations */
	if (likely(qp->s_avail))
		return 0;
1534 1535 1536
	smp_read_barrier_depends(); /* see rc.c */
	slast = ACCESS_ONCE(qp->s_last);
	if (qp->s_head >= slast)
1537
		avail = qp->s_size - (qp->s_head - slast);
1538
	else
1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557
		avail = slast - qp->s_head;

	/* see rvt_qp_wqe_unreserve() */
	smp_mb__before_atomic();
	reserved_used = atomic_read(&qp->s_reserved_used);
	avail =  avail - 1 -
		(rdi->dparms.reserved_operations - reserved_used);
	/* insure we don't assign a negative s_avail */
	if ((s32)avail <= 0)
		return -ENOMEM;
	qp->s_avail = avail;
	if (WARN_ON(qp->s_avail >
		    (qp->s_size - 1 - rdi->dparms.reserved_operations)))
		rvt_pr_err(rdi,
			   "More avail entries than QP RB size.\nQP: %u, size: %u, avail: %u\nhead: %u, tail: %u, cur: %u, acked: %u, last: %u",
			   qp->ibqp.qp_num, qp->s_size, qp->s_avail,
			   qp->s_head, qp->s_tail, qp->s_cur,
			   qp->s_acked, qp->s_last);
	return 0;
1558 1559
}

1560 1561 1562 1563 1564
/**
 * rvt_post_one_wr - post one RC, UC, or UD send work request
 * @qp: the QP to post on
 * @wr: the work request to send
 */
1565 1566 1567
static int rvt_post_one_wr(struct rvt_qp *qp,
			   struct ib_send_wr *wr,
			   int *call_send)
1568 1569 1570 1571 1572 1573 1574 1575 1576
{
	struct rvt_swqe *wqe;
	u32 next;
	int i;
	int j;
	int acc;
	struct rvt_lkey_table *rkt;
	struct rvt_pd *pd;
	struct rvt_dev_info *rdi = ib_to_rvt(qp->ibqp.device);
1577 1578
	u8 log_pmtu;
	int ret;
1579
	size_t cplen;
1580
	bool reserved_op;
1581
	int local_ops_delayed = 0;
1582

1583 1584
	BUILD_BUG_ON(IB_QPT_MAX >= (sizeof(u32) * BITS_PER_BYTE));

1585 1586 1587 1588
	/* IB spec says that num_sge == 0 is OK. */
	if (unlikely(wr->num_sge > qp->s_max_sge))
		return -EINVAL;

1589 1590 1591 1592 1593
	ret = rvt_qp_valid_operation(qp, rdi->post_parms, wr);
	if (ret < 0)
		return ret;
	cplen = ret;

1594
	/*
1595 1596 1597 1598 1599 1600 1601 1602 1603
	 * Local operations include fast register and local invalidate.
	 * Fast register needs to be processed immediately because the
	 * registered lkey may be used by following work requests and the
	 * lkey needs to be valid at the time those requests are posted.
	 * Local invalidate can be processed immediately if fencing is
	 * not required and no previous local invalidate ops are pending.
	 * Signaled local operations that have been processed immediately
	 * need to have requests with "completion only" flags set posted
	 * to the send queue in order to generate completions.
1604
	 */
1605
	if ((rdi->post_parms[wr->opcode].flags & RVT_OPERATION_LOCAL)) {
1606 1607
		switch (wr->opcode) {
		case IB_WR_REG_MR:
1608 1609 1610 1611 1612 1613 1614
			ret = rvt_fast_reg_mr(qp,
					      reg_wr(wr)->mr,
					      reg_wr(wr)->key,
					      reg_wr(wr)->access);
			if (ret || !(wr->send_flags & IB_SEND_SIGNALED))
				return ret;
			break;
1615
		case IB_WR_LOCAL_INV:
1616 1617 1618 1619 1620 1621 1622 1623 1624 1625
			if ((wr->send_flags & IB_SEND_FENCE) ||
			    atomic_read(&qp->local_ops_pending)) {
				local_ops_delayed = 1;
			} else {
				ret = rvt_invalidate_rkey(
					qp, wr->ex.invalidate_rkey);
				if (ret || !(wr->send_flags & IB_SEND_SIGNALED))
					return ret;
			}
			break;
1626 1627 1628 1629 1630
		default:
			return -EINVAL;
		}
	}

1631 1632
	reserved_op = rdi->post_parms[wr->opcode].flags &
			RVT_OPERATION_USE_RESERVE;
1633
	/* check for avail */
1634 1635 1636
	ret = rvt_qp_is_avail(qp, rdi, reserved_op);
	if (ret)
		return ret;
1637 1638 1639
	next = qp->s_head + 1;
	if (next >= qp->s_size)
		next = 0;
1640

1641 1642 1643 1644
	rkt = &rdi->lkey_table;
	pd = ibpd_to_rvtpd(qp->ibqp.pd);
	wqe = rvt_get_swqe_ptr(qp, qp->s_head);

1645 1646
	/* cplen has length from above */
	memcpy(&wqe->wr, wr, cplen);
1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660

	wqe->length = 0;
	j = 0;
	if (wr->num_sge) {
		acc = wr->opcode >= IB_WR_RDMA_READ ?
			IB_ACCESS_LOCAL_WRITE : 0;
		for (i = 0; i < wr->num_sge; i++) {
			u32 length = wr->sg_list[i].length;
			int ok;

			if (length == 0)
				continue;
			ok = rvt_lkey_ok(rkt, pd, &wqe->sg_list[j],
					 &wr->sg_list[i], acc);
1661 1662
			if (!ok) {
				ret = -EINVAL;
1663
				goto bail_inval_free;
1664
			}
1665 1666 1667 1668 1669
			wqe->length += length;
			j++;
		}
		wqe->wr.num_sge = j;
	}
1670 1671 1672 1673

	/* general part of wqe valid - allow for driver checks */
	if (rdi->driver_f.check_send_wqe) {
		ret = rdi->driver_f.check_send_wqe(qp, wqe);
1674
		if (ret < 0)
1675
			goto bail_inval_free;
1676 1677
		if (ret)
			*call_send = ret;
1678 1679 1680 1681 1682 1683 1684 1685
	}

	log_pmtu = qp->log_pmtu;
	if (qp->ibqp.qp_type != IB_QPT_UC &&
	    qp->ibqp.qp_type != IB_QPT_RC) {
		struct rvt_ah *ah = ibah_to_rvtah(wqe->ud_wr.ah);

		log_pmtu = ah->log_pmtu;
1686 1687
		atomic_inc(&ibah_to_rvtah(ud_wr(wr)->ah)->refcount);
	}
1688

1689
	if (rdi->post_parms[wr->opcode].flags & RVT_OPERATION_LOCAL) {
1690 1691 1692 1693
		if (local_ops_delayed)
			atomic_inc(&qp->local_ops_pending);
		else
			wqe->wr.send_flags |= RVT_SEND_COMPLETION_ONLY;
1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705
		wqe->ssn = 0;
		wqe->psn = 0;
		wqe->lpsn = 0;
	} else {
		wqe->ssn = qp->s_ssn++;
		wqe->psn = qp->s_next_psn;
		wqe->lpsn = wqe->psn +
				(wqe->length ?
					((wqe->length - 1) >> log_pmtu) :
					0);
		qp->s_next_psn = wqe->lpsn + 1;
	}
1706
	trace_rvt_post_one_wr(qp, wqe);
1707 1708 1709 1710
	if (unlikely(reserved_op))
		rvt_qp_wqe_reserve(qp, wqe);
	else
		qp->s_avail--;
1711
	smp_wmb(); /* see request builders */
1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722
	qp->s_head = next;

	return 0;

bail_inval_free:
	/* release mr holds */
	while (j) {
		struct rvt_sge *sge = &wqe->sg_list[--j];

		rvt_put_mr(sge->mr);
	}
1723
	return ret;
1724 1725
}

1726 1727 1728 1729 1730 1731 1732
/**
 * rvt_post_send - post a send on a QP
 * @ibqp: the QP to post the send on
 * @wr: the list of work requests to post
 * @bad_wr: the first bad WR is put here
 *
 * This may be called from interrupt context.
1733 1734
 *
 * Return: 0 on success else errno
1735 1736 1737 1738
 */
int rvt_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
		  struct ib_send_wr **bad_wr)
{
1739 1740 1741 1742 1743 1744 1745
	struct rvt_qp *qp = ibqp_to_rvtqp(ibqp);
	struct rvt_dev_info *rdi = ib_to_rvt(ibqp->device);
	unsigned long flags = 0;
	int call_send;
	unsigned nreq = 0;
	int err = 0;

1746
	spin_lock_irqsave(&qp->s_hlock, flags);
1747

1748
	/*
1749 1750
	 * Ensure QP state is such that we can send. If not bail out early,
	 * there is no need to do this every time we post a send.
1751
	 */
1752
	if (unlikely(!(ib_rvt_state_ops[qp->state] & RVT_POST_SEND_OK))) {
1753
		spin_unlock_irqrestore(&qp->s_hlock, flags);
1754 1755
		return -EINVAL;
	}
1756

1757 1758 1759 1760 1761 1762 1763 1764
	/*
	 * If the send queue is empty, and we only have a single WR then just go
	 * ahead and kick the send engine into gear. Otherwise we will always
	 * just schedule the send to happen later.
	 */
	call_send = qp->s_head == ACCESS_ONCE(qp->s_last) && !wr->next;

	for (; wr; wr = wr->next) {
1765
		err = rvt_post_one_wr(qp, wr, &call_send);
1766 1767 1768 1769 1770 1771 1772
		if (unlikely(err)) {
			*bad_wr = wr;
			goto bail;
		}
		nreq++;
	}
bail:
1773 1774 1775 1776
	spin_unlock_irqrestore(&qp->s_hlock, flags);
	if (nreq) {
		if (call_send)
			rdi->driver_f.do_send(qp);
1777 1778
		else
			rdi->driver_f.schedule_send_no_lock(qp);
1779
	}
1780
	return err;
1781 1782 1783 1784 1785 1786 1787 1788 1789
}

/**
 * rvt_post_srq_receive - post a receive on a shared receive queue
 * @ibsrq: the SRQ to post the receive on
 * @wr: the list of work requests to post
 * @bad_wr: A pointer to the first WR to cause a problem is put here
 *
 * This may be called from interrupt context.
1790 1791
 *
 * Return: 0 on success else errno
1792 1793 1794 1795
 */
int rvt_post_srq_recv(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
		      struct ib_recv_wr **bad_wr)
{
1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831
	struct rvt_srq *srq = ibsrq_to_rvtsrq(ibsrq);
	struct rvt_rwq *wq;
	unsigned long flags;

	for (; wr; wr = wr->next) {
		struct rvt_rwqe *wqe;
		u32 next;
		int i;

		if ((unsigned)wr->num_sge > srq->rq.max_sge) {
			*bad_wr = wr;
			return -EINVAL;
		}

		spin_lock_irqsave(&srq->rq.lock, flags);
		wq = srq->rq.wq;
		next = wq->head + 1;
		if (next >= srq->rq.size)
			next = 0;
		if (next == wq->tail) {
			spin_unlock_irqrestore(&srq->rq.lock, flags);
			*bad_wr = wr;
			return -ENOMEM;
		}

		wqe = rvt_get_rwqe_ptr(&srq->rq, wq->head);
		wqe->wr_id = wr->wr_id;
		wqe->num_sge = wr->num_sge;
		for (i = 0; i < wr->num_sge; i++)
			wqe->sg_list[i] = wr->sg_list[i];
		/* Make sure queue entry is written before the head index. */
		smp_wmb();
		wq->head = next;
		spin_unlock_irqrestore(&srq->rq.lock, flags);
	}
	return 0;
1832
}