en_rx.c 32.9 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) 2007 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 <net/busy_poll.h>
35
#include <linux/bpf.h>
36
#include <linux/bpf_trace.h>
37
#include <linux/mlx4/cq.h>
38
#include <linux/slab.h>
39 40
#include <linux/mlx4/qp.h>
#include <linux/skbuff.h>
41
#include <linux/rculist.h>
42 43 44
#include <linux/if_ether.h>
#include <linux/if_vlan.h>
#include <linux/vmalloc.h>
45
#include <linux/irq.h>
46

47 48 49 50
#if IS_ENABLED(CONFIG_IPV6)
#include <net/ip6_checksum.h>
#endif

51 52
#include "mlx4_en.h"

53 54 55
static int mlx4_alloc_page(struct mlx4_en_priv *priv,
			   struct mlx4_en_rx_alloc *frag,
			   gfp_t gfp)
56 57 58 59
{
	struct page *page;
	dma_addr_t dma;

E
Eric Dumazet 已提交
60 61 62 63
	page = alloc_page(gfp);
	if (unlikely(!page))
		return -ENOMEM;
	dma = dma_map_page(priv->ddev, page, 0, PAGE_SIZE, priv->dma_dir);
64
	if (unlikely(dma_mapping_error(priv->ddev, dma))) {
65
		__free_page(page);
66 67
		return -ENOMEM;
	}
68 69 70
	frag->page = page;
	frag->dma = dma;
	frag->page_offset = priv->rx_headroom;
71 72 73
	return 0;
}

74
static int mlx4_en_alloc_frags(struct mlx4_en_priv *priv,
75
			       struct mlx4_en_rx_ring *ring,
76 77
			       struct mlx4_en_rx_desc *rx_desc,
			       struct mlx4_en_rx_alloc *frags,
78
			       gfp_t gfp)
79
{
80
	int i;
81

82
	for (i = 0; i < priv->num_frags; i++, frags++) {
83 84 85 86 87
		if (!frags->page) {
			if (mlx4_alloc_page(priv, frags, gfp))
				return -ENOMEM;
			ring->rx_alloc_pages++;
		}
88 89
		rx_desc->data[i].addr = cpu_to_be64(frags->dma +
						    frags->page_offset);
90 91 92 93
	}
	return 0;
}

94 95
static void mlx4_en_free_frag(const struct mlx4_en_priv *priv,
			      struct mlx4_en_rx_alloc *frag)
96
{
97 98
	if (frag->page) {
		dma_unmap_page(priv->ddev, frag->dma,
E
Eric Dumazet 已提交
99
			       PAGE_SIZE, priv->dma_dir);
100
		__free_page(frag->page);
101
	}
102 103 104 105
	/* We need to clear all fields, otherwise a change of priv->log_rx_info
	 * could lead to see garbage later in frag->page.
	 */
	memset(frag, 0, sizeof(*frag));
106 107
}

108
static void mlx4_en_init_rx_desc(const struct mlx4_en_priv *priv,
109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133
				 struct mlx4_en_rx_ring *ring, int index)
{
	struct mlx4_en_rx_desc *rx_desc = ring->buf + ring->stride * index;
	int possible_frags;
	int i;

	/* Set size and memtype fields */
	for (i = 0; i < priv->num_frags; i++) {
		rx_desc->data[i].byte_count =
			cpu_to_be32(priv->frag_info[i].frag_size);
		rx_desc->data[i].lkey = cpu_to_be32(priv->mdev->mr.key);
	}

	/* If the number of used fragments does not fill up the ring stride,
	 * remaining (unused) fragments must be padded with null address/size
	 * and a special memory key */
	possible_frags = (ring->stride - sizeof(struct mlx4_en_rx_desc)) / DS_SIZE;
	for (i = priv->num_frags; i < possible_frags; i++) {
		rx_desc->data[i].byte_count = 0;
		rx_desc->data[i].lkey = cpu_to_be32(MLX4_EN_MEMTYPE_PAD);
		rx_desc->data[i].addr = 0;
	}
}

static int mlx4_en_prepare_rx_desc(struct mlx4_en_priv *priv,
134 135
				   struct mlx4_en_rx_ring *ring, int index,
				   gfp_t gfp)
136 137
{
	struct mlx4_en_rx_desc *rx_desc = ring->buf + (index * ring->stride);
138 139
	struct mlx4_en_rx_alloc *frags = ring->rx_info +
					(index << priv->log_rx_info);
140
	if (ring->page_cache.index > 0) {
141 142 143 144 145 146 147 148 149
		/* XDP uses a single page per frame */
		if (!frags->page) {
			ring->page_cache.index--;
			frags->page = ring->page_cache.buf[ring->page_cache.index].page;
			frags->dma  = ring->page_cache.buf[ring->page_cache.index].dma;
		}
		frags->page_offset = XDP_PACKET_HEADROOM;
		rx_desc->data[0].addr = cpu_to_be64(frags->dma +
						    XDP_PACKET_HEADROOM);
150 151 152
		return 0;
	}

153
	return mlx4_en_alloc_frags(priv, ring, rx_desc, frags, gfp);
154 155
}

156
static bool mlx4_en_is_ring_empty(const struct mlx4_en_rx_ring *ring)
157 158 159 160
{
	return ring->prod == ring->cons;
}

161 162 163 164 165
static inline void mlx4_en_update_rx_prod_db(struct mlx4_en_rx_ring *ring)
{
	*ring->wqres.db.db = cpu_to_be32(ring->prod & 0xffff);
}

166 167
/* slow path */
static void mlx4_en_free_rx_desc(const struct mlx4_en_priv *priv,
168 169 170
				 struct mlx4_en_rx_ring *ring,
				 int index)
{
171
	struct mlx4_en_rx_alloc *frags;
172 173
	int nr;

174
	frags = ring->rx_info + (index << priv->log_rx_info);
175
	for (nr = 0; nr < priv->num_frags; nr++) {
176
		en_dbg(DRV, priv, "Freeing fragment:%d\n", nr);
177
		mlx4_en_free_frag(priv, frags + nr);
178 179 180
	}
}

181 182 183 184 185
static int mlx4_en_fill_rx_buffers(struct mlx4_en_priv *priv)
{
	struct mlx4_en_rx_ring *ring;
	int ring_ind;
	int buf_ind;
186
	int new_size;
187 188 189

	for (buf_ind = 0; buf_ind < priv->prof->rx_ring_size; buf_ind++) {
		for (ring_ind = 0; ring_ind < priv->rx_ring_num; ring_ind++) {
190
			ring = priv->rx_ring[ring_ind];
191 192

			if (mlx4_en_prepare_rx_desc(priv, ring,
193
						    ring->actual_size,
194
						    GFP_KERNEL | __GFP_COLD)) {
195
				if (ring->actual_size < MLX4_EN_MIN_RX_SIZE) {
J
Joe Perches 已提交
196
					en_err(priv, "Failed to allocate enough rx buffers\n");
197 198
					return -ENOMEM;
				} else {
199
					new_size = rounddown_pow_of_two(ring->actual_size);
J
Joe Perches 已提交
200
					en_warn(priv, "Only %d buffers allocated reducing ring size to %d\n",
201
						ring->actual_size, new_size);
202
					goto reduce_rings;
203 204 205 206 207 208
				}
			}
			ring->actual_size++;
			ring->prod++;
		}
	}
209 210 211 212
	return 0;

reduce_rings:
	for (ring_ind = 0; ring_ind < priv->rx_ring_num; ring_ind++) {
213
		ring = priv->rx_ring[ring_ind];
214 215 216 217 218 219 220
		while (ring->actual_size > new_size) {
			ring->actual_size--;
			ring->prod--;
			mlx4_en_free_rx_desc(priv, ring, ring->actual_size);
		}
	}

221 222 223 224 225 226 227 228
	return 0;
}

static void mlx4_en_free_rx_buf(struct mlx4_en_priv *priv,
				struct mlx4_en_rx_ring *ring)
{
	int index;

229 230
	en_dbg(DRV, priv, "Freeing Rx buf - cons:%d prod:%d\n",
	       ring->cons, ring->prod);
231 232

	/* Unmap and free Rx buffers */
233
	for (index = 0; index < ring->size; index++) {
234
		en_dbg(DRV, priv, "Processing descriptor:%d\n", index);
235
		mlx4_en_free_rx_desc(priv, ring, index);
236
	}
237 238
	ring->cons = 0;
	ring->prod = 0;
239 240
}

241 242 243 244
void mlx4_en_set_num_rx_rings(struct mlx4_en_dev *mdev)
{
	int i;
	int num_of_eqs;
245
	int num_rx_rings;
246 247 248
	struct mlx4_dev *dev = mdev->dev;

	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH) {
M
Matan Barak 已提交
249 250 251 252
		num_of_eqs = max_t(int, MIN_RX_RINGS,
				   min_t(int,
					 mlx4_get_eqs_per_port(mdev->dev, i),
					 DEF_RX_RINGS));
253

254 255 256
		num_rx_rings = mlx4_low_memory_profile() ? MIN_RX_RINGS :
			min_t(int, num_of_eqs,
			      netif_get_num_default_rss_queues());
257
		mdev->profile.prof[i].rx_ring_num =
258
			rounddown_pow_of_two(num_rx_rings);
259 260 261
	}
}

262
int mlx4_en_create_rx_ring(struct mlx4_en_priv *priv,
263
			   struct mlx4_en_rx_ring **pring,
264
			   u32 size, u16 stride, int node)
265 266
{
	struct mlx4_en_dev *mdev = priv->mdev;
267
	struct mlx4_en_rx_ring *ring;
268
	int err = -ENOMEM;
269 270
	int tmp;

271
	ring = kzalloc_node(sizeof(*ring), GFP_KERNEL, node);
272
	if (!ring) {
273 274 275 276 277
		ring = kzalloc(sizeof(*ring), GFP_KERNEL);
		if (!ring) {
			en_err(priv, "Failed to allocate RX ring structure\n");
			return -ENOMEM;
		}
278 279
	}

280 281 282 283 284 285
	ring->prod = 0;
	ring->cons = 0;
	ring->size = size;
	ring->size_mask = size - 1;
	ring->stride = stride;
	ring->log_stride = ffs(ring->stride) - 1;
286
	ring->buf_size = ring->size * ring->stride + TXBB_SIZE;
287 288

	tmp = size * roundup_pow_of_two(MLX4_EN_MAX_RX_FRAGS *
289
					sizeof(struct mlx4_en_rx_alloc));
290
	ring->rx_info = vzalloc_node(tmp, node);
291
	if (!ring->rx_info) {
292
		ring->rx_info = vzalloc(tmp);
293 294 295 296
		if (!ring->rx_info) {
			err = -ENOMEM;
			goto err_ring;
		}
297
	}
298

299
	en_dbg(DRV, priv, "Allocated rx_info ring at addr:%p size:%d\n",
300 301
		 ring->rx_info, tmp);

302
	/* Allocate HW buffers on provided NUMA node */
303
	set_dev_node(&mdev->dev->persist->pdev->dev, node);
304
	err = mlx4_alloc_hwq_res(mdev->dev, &ring->wqres, ring->buf_size);
305
	set_dev_node(&mdev->dev->persist->pdev->dev, mdev->dev->numa_node);
306
	if (err)
307
		goto err_info;
308 309 310

	ring->buf = ring->wqres.buf.direct.buf;

311 312
	ring->hwtstamp_rx_filter = priv->hwtstamp_config.rx_filter;

313
	*pring = ring;
314 315
	return 0;

316
err_info:
317 318
	vfree(ring->rx_info);
	ring->rx_info = NULL;
319 320 321 322
err_ring:
	kfree(ring);
	*pring = NULL;

323 324 325 326 327 328 329 330 331 332 333 334 335
	return err;
}

int mlx4_en_activate_rx_rings(struct mlx4_en_priv *priv)
{
	struct mlx4_en_rx_ring *ring;
	int i;
	int ring_ind;
	int err;
	int stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
					DS_SIZE * priv->num_frags);

	for (ring_ind = 0; ring_ind < priv->rx_ring_num; ring_ind++) {
336
		ring = priv->rx_ring[ring_ind];
337 338 339 340

		ring->prod = 0;
		ring->cons = 0;
		ring->actual_size = 0;
341
		ring->cqn = priv->rx_cq[ring_ind]->mcq.cqn;
342 343

		ring->stride = stride;
E
Eugenia Emantayev 已提交
344 345 346 347 348 349
		if (ring->stride <= TXBB_SIZE) {
			/* Stamp first unused send wqe */
			__be32 *ptr = (__be32 *)ring->buf;
			__be32 stamp = cpu_to_be32(1 << STAMP_SHIFT);
			*ptr = stamp;
			/* Move pointer to start of rx section */
350
			ring->buf += TXBB_SIZE;
E
Eugenia Emantayev 已提交
351
		}
352

353 354 355 356 357 358
		ring->log_stride = ffs(ring->stride) - 1;
		ring->buf_size = ring->size * ring->stride;

		memset(ring->buf, 0, ring->buf_size);
		mlx4_en_update_rx_prod_db(ring);

359
		/* Initialize all descriptors */
360 361 362
		for (i = 0; i < ring->size; i++)
			mlx4_en_init_rx_desc(priv, ring, i);
	}
363 364
	err = mlx4_en_fill_rx_buffers(priv);
	if (err)
365 366 367
		goto err_buffers;

	for (ring_ind = 0; ring_ind < priv->rx_ring_num; ring_ind++) {
368
		ring = priv->rx_ring[ring_ind];
369

370
		ring->size_mask = ring->actual_size - 1;
371 372 373 374 375 376 377
		mlx4_en_update_rx_prod_db(ring);
	}

	return 0;

err_buffers:
	for (ring_ind = 0; ring_ind < priv->rx_ring_num; ring_ind++)
378
		mlx4_en_free_rx_buf(priv, priv->rx_ring[ring_ind]);
379 380 381

	ring_ind = priv->rx_ring_num - 1;
	while (ring_ind >= 0) {
382 383
		if (priv->rx_ring[ring_ind]->stride <= TXBB_SIZE)
			priv->rx_ring[ring_ind]->buf -= TXBB_SIZE;
384 385 386 387 388
		ring_ind--;
	}
	return err;
}

389 390 391 392 393 394 395 396 397 398 399 400
/* We recover from out of memory by scheduling our napi poll
 * function (mlx4_en_process_cq), which tries to allocate
 * all missing RX buffers (call to mlx4_en_refill_rx_buffers).
 */
void mlx4_en_recover_from_oom(struct mlx4_en_priv *priv)
{
	int ring;

	if (!priv->port_up)
		return;

	for (ring = 0; ring < priv->rx_ring_num; ring++) {
401 402
		if (mlx4_en_is_ring_empty(priv->rx_ring[ring])) {
			local_bh_disable();
403
			napi_reschedule(&priv->rx_cq[ring]->napi);
404 405
			local_bh_enable();
		}
406 407 408
	}
}

409 410 411 412 413 414 415 416 417 418 419 420 421 422
/* When the rx ring is running in page-per-packet mode, a released frame can go
 * directly into a small cache, to avoid unmapping or touching the page
 * allocator. In bpf prog performance scenarios, buffers are either forwarded
 * or dropped, never converted to skbs, so every page can come directly from
 * this cache when it is sized to be a multiple of the napi budget.
 */
bool mlx4_en_rx_recycle(struct mlx4_en_rx_ring *ring,
			struct mlx4_en_rx_alloc *frame)
{
	struct mlx4_en_page_cache *cache = &ring->page_cache;

	if (cache->index >= MLX4_EN_CACHE_SIZE)
		return false;

423 424 425
	cache->buf[cache->index].page = frame->page;
	cache->buf[cache->index].dma = frame->dma;
	cache->index++;
426 427 428
	return true;
}

429
void mlx4_en_destroy_rx_ring(struct mlx4_en_priv *priv,
430 431
			     struct mlx4_en_rx_ring **pring,
			     u32 size, u16 stride)
432 433
{
	struct mlx4_en_dev *mdev = priv->mdev;
434
	struct mlx4_en_rx_ring *ring = *pring;
435
	struct bpf_prog *old_prog;
436

437 438 439
	old_prog = rcu_dereference_protected(
					ring->xdp_prog,
					lockdep_is_held(&mdev->state_lock));
440 441
	if (old_prog)
		bpf_prog_put(old_prog);
442
	mlx4_free_hwq_res(mdev->dev, &ring->wqres, size * stride + TXBB_SIZE);
443 444
	vfree(ring->rx_info);
	ring->rx_info = NULL;
445 446
	kfree(ring);
	*pring = NULL;
447 448 449 450 451
}

void mlx4_en_deactivate_rx_ring(struct mlx4_en_priv *priv,
				struct mlx4_en_rx_ring *ring)
{
452 453 454
	int i;

	for (i = 0; i < ring->page_cache.index; i++) {
455 456 457
		dma_unmap_page(priv->ddev, ring->page_cache.buf[i].dma,
			       PAGE_SIZE, priv->dma_dir);
		put_page(ring->page_cache.buf[i].page);
458 459
	}
	ring->page_cache.index = 0;
460
	mlx4_en_free_rx_buf(priv, ring);
461 462
	if (ring->stride <= TXBB_SIZE)
		ring->buf -= TXBB_SIZE;
463 464 465 466
}


static int mlx4_en_complete_rx_desc(struct mlx4_en_priv *priv,
467
				    struct mlx4_en_rx_alloc *frags,
468
				    struct sk_buff *skb,
469 470
				    int length)
{
471 472
	const struct mlx4_en_frag_info *frag_info = priv->frag_info;
	unsigned int truesize = 0;
E
Eric Dumazet 已提交
473
	int nr, frag_size;
474
	struct page *page;
475
	dma_addr_t dma;
476
	bool release;
477

478
	/* Collect used fragments while replacing them in the HW descriptors */
479
	for (nr = 0;; frags++) {
E
Eric Dumazet 已提交
480 481
		frag_size = min_t(int, length, frag_info->frag_size);

482 483
		page = frags->page;
		if (unlikely(!page))
484
			goto fail;
485

486 487 488
		dma = frags->dma;
		dma_sync_single_range_for_cpu(priv->ddev, dma, frags->page_offset,
					      frag_size, priv->dma_dir);
489

490
		__skb_fill_page_desc(skb, nr, page, frags->page_offset,
E
Eric Dumazet 已提交
491
				     frag_size);
492

493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511
		truesize += frag_info->frag_stride;
		if (frag_info->frag_stride == PAGE_SIZE / 2) {
			frags->page_offset ^= PAGE_SIZE / 2;
			release = page_count(page) != 1 ||
				  page_is_pfmemalloc(page) ||
				  page_to_nid(page) != numa_mem_id();
		} else {
			u32 sz_align = ALIGN(frag_size, SMP_CACHE_BYTES);

			frags->page_offset += sz_align;
			release = frags->page_offset + frag_info->frag_size > PAGE_SIZE;
		}
		if (release) {
			dma_unmap_page(priv->ddev, dma, PAGE_SIZE, priv->dma_dir);
			frags->page = NULL;
		} else {
			page_ref_inc(page);
		}

E
Eric Dumazet 已提交
512 513 514 515 516
		nr++;
		length -= frag_size;
		if (!length)
			break;
		frag_info++;
517
	}
518
	skb->truesize += truesize;
519 520 521 522 523
	return nr;

fail:
	while (nr > 0) {
		nr--;
524
		__skb_frag_unref(skb_shinfo(skb)->frags + nr);
525 526 527 528
	}
	return 0;
}

529
static void validate_loopback(struct mlx4_en_priv *priv, void *va)
530
{
531
	const unsigned char *data = va + ETH_HLEN;
532 533
	int i;

534 535 536
	for (i = 0; i < MLX4_LOOPBACK_TEST_PAYLOAD; i++) {
		if (data[i] != (unsigned char)i)
			return;
537 538 539 540
	}
	/* Loopback found */
	priv->loopback_ok = 1;
}
541

542 543
static bool mlx4_en_refill_rx_buffers(struct mlx4_en_priv *priv,
				      struct mlx4_en_rx_ring *ring)
544
{
545
	u32 missing = ring->actual_size - (ring->prod - ring->cons);
546

547 548 549 550 551 552
	/* Try to batch allocations, but not too much. */
	if (missing < 8)
		return false;
	do {
		if (mlx4_en_prepare_rx_desc(priv, ring,
					    ring->prod & ring->size_mask,
553 554
					    GFP_ATOMIC | __GFP_COLD |
					    __GFP_MEMALLOC))
555 556
			break;
		ring->prod++;
557 558 559
	} while (--missing);

	return true;
560 561
}

562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596
/* When hardware doesn't strip the vlan, we need to calculate the checksum
 * over it and add it to the hardware's checksum calculation
 */
static inline __wsum get_fixed_vlan_csum(__wsum hw_checksum,
					 struct vlan_hdr *vlanh)
{
	return csum_add(hw_checksum, *(__wsum *)vlanh);
}

/* Although the stack expects checksum which doesn't include the pseudo
 * header, the HW adds it. To address that, we are subtracting the pseudo
 * header checksum from the checksum value provided by the HW.
 */
static void get_fixed_ipv4_csum(__wsum hw_checksum, struct sk_buff *skb,
				struct iphdr *iph)
{
	__u16 length_for_csum = 0;
	__wsum csum_pseudo_header = 0;

	length_for_csum = (be16_to_cpu(iph->tot_len) - (iph->ihl << 2));
	csum_pseudo_header = csum_tcpudp_nofold(iph->saddr, iph->daddr,
						length_for_csum, iph->protocol, 0);
	skb->csum = csum_sub(hw_checksum, csum_pseudo_header);
}

#if IS_ENABLED(CONFIG_IPV6)
/* In IPv6 packets, besides subtracting the pseudo header checksum,
 * we also compute/add the IP header checksum which
 * is not added by the HW.
 */
static int get_fixed_ipv6_csum(__wsum hw_checksum, struct sk_buff *skb,
			       struct ipv6hdr *ipv6h)
{
	__wsum csum_pseudo_hdr = 0;

597 598
	if (unlikely(ipv6h->nexthdr == IPPROTO_FRAGMENT ||
		     ipv6h->nexthdr == IPPROTO_HOPOPTS))
599
		return -1;
600
	hw_checksum = csum_add(hw_checksum, (__force __wsum)htons(ipv6h->nexthdr));
601 602 603 604 605 606 607 608 609 610 611 612

	csum_pseudo_hdr = csum_partial(&ipv6h->saddr,
				       sizeof(ipv6h->saddr) + sizeof(ipv6h->daddr), 0);
	csum_pseudo_hdr = csum_add(csum_pseudo_hdr, (__force __wsum)ipv6h->payload_len);
	csum_pseudo_hdr = csum_add(csum_pseudo_hdr, (__force __wsum)ntohs(ipv6h->nexthdr));

	skb->csum = csum_sub(hw_checksum, csum_pseudo_hdr);
	skb->csum = csum_add(skb->csum, csum_partial(ipv6h, sizeof(struct ipv6hdr), 0));
	return 0;
}
#endif
static int check_csum(struct mlx4_cqe *cqe, struct sk_buff *skb, void *va,
613
		      netdev_features_t dev_features)
614 615 616 617 618 619 620
{
	__wsum hw_checksum = 0;

	void *hdr = (u8 *)va + sizeof(struct ethhdr);

	hw_checksum = csum_unfold((__force __sum16)cqe->checksum);

621
	if (cqe->vlan_my_qpn & cpu_to_be32(MLX4_CQE_CVLAN_PRESENT_MASK) &&
622
	    !(dev_features & NETIF_F_HW_VLAN_CTAG_RX)) {
623 624 625 626 627 628 629 630
		hw_checksum = get_fixed_vlan_csum(hw_checksum, hdr);
		hdr += sizeof(struct vlan_hdr);
	}

	if (cqe->status & cpu_to_be16(MLX4_CQE_STATUS_IPV4))
		get_fixed_ipv4_csum(hw_checksum, skb, hdr);
#if IS_ENABLED(CONFIG_IPV6)
	else if (cqe->status & cpu_to_be16(MLX4_CQE_STATUS_IPV6))
631
		if (unlikely(get_fixed_ipv6_csum(hw_checksum, skb, hdr)))
632 633 634 635 636
			return -1;
#endif
	return 0;
}

637 638 639
int mlx4_en_process_rx_cq(struct net_device *dev, struct mlx4_en_cq *cq, int budget)
{
	struct mlx4_en_priv *priv = netdev_priv(dev);
640
	struct mlx4_en_dev *mdev = priv->mdev;
641
	struct mlx4_cqe *cqe;
642
	struct mlx4_en_rx_ring *ring = priv->rx_ring[cq->ring];
643
	struct mlx4_en_rx_alloc *frags;
644
	struct bpf_prog *xdp_prog;
645
	int doorbell_pending;
646 647 648 649 650 651
	struct sk_buff *skb;
	int index;
	int nr;
	unsigned int length;
	int polled = 0;
	int ip_summed;
O
Or Gerlitz 已提交
652
	int factor = priv->cqe_factor;
653
	u64 timestamp;
654
	bool l2_tunnel;
655

656
	if (unlikely(!priv->port_up))
657 658
		return 0;

659
	if (unlikely(budget <= 0))
660 661
		return polled;

662 663 664
	/* Protect accesses to: ring->xdp_prog, priv->mac_hash list */
	rcu_read_lock();
	xdp_prog = rcu_dereference(ring->xdp_prog);
665
	doorbell_pending = 0;
666

667 668 669 670
	/* We assume a 1:1 mapping between CQEs and Rx descriptors, so Rx
	 * descriptor offset can be deduced from the CQE index instead of
	 * reading 'cqe->index' */
	index = cq->mcq.cons_index & ring->size_mask;
671
	cqe = mlx4_en_get_cqe(cq->buf, index, priv->cqe_size) + factor;
672 673 674 675

	/* Process all completed CQEs */
	while (XNOR(cqe->owner_sr_opcode & MLX4_CQE_OWNER_MASK,
		    cq->mcq.cons_index & cq->size)) {
676
		void *va;
677

678
		frags = ring->rx_info + (index << priv->log_rx_info);
679
		va = page_address(frags[0].page) + frags[0].page_offset;
680 681 682
		/*
		 * make sure we read the CQE after we read the ownership bit
		 */
683
		dma_rmb();
684 685 686 687

		/* Drop packet on bad receive or bad checksum */
		if (unlikely((cqe->owner_sr_opcode & MLX4_CQE_OPCODE_MASK) ==
						MLX4_CQE_OPCODE_ERROR)) {
J
Joe Perches 已提交
688 689 690
			en_err(priv, "CQE completed in error - vendor syndrom:%d syndrom:%d\n",
			       ((struct mlx4_err_cqe *)cqe)->vendor_err_syndrome,
			       ((struct mlx4_err_cqe *)cqe)->syndrome);
691 692 693
			goto next;
		}
		if (unlikely(cqe->badfcs_enc & MLX4_CQE_BAD_FCS)) {
694
			en_dbg(RX_ERR, priv, "Accepted frame with bad FCS\n");
695 696 697
			goto next;
		}

698 699 700 701
		/* Check if we need to drop the packet if SRIOV is not enabled
		 * and not performing the selftest or flb disabled
		 */
		if (priv->flags & MLX4_EN_FLAG_RX_FILTER_NEEDED) {
702
			const struct ethhdr *ethh = va;
703 704 705 706
			dma_addr_t dma;
			/* Get pointer to first fragment since we haven't
			 * skb yet and cast it to ethhdr struct
			 */
707
			dma = frags[0].dma + frags[0].page_offset;
708 709 710
			dma_sync_single_for_cpu(priv->ddev, dma, sizeof(*ethh),
						DMA_FROM_DEVICE);

711 712 713 714 715 716 717 718
			if (is_multicast_ether_addr(ethh->h_dest)) {
				struct mlx4_mac_entry *entry;
				struct hlist_head *bucket;
				unsigned int mac_hash;

				/* Drop the packet, since HW loopback-ed it */
				mac_hash = ethh->h_source[MLX4_EN_MAC_HASH_IDX];
				bucket = &priv->mac_hash[mac_hash];
719
				hlist_for_each_entry_rcu(entry, bucket, hlist) {
720
					if (ether_addr_equal_64bits(entry->mac,
721
								    ethh->h_source))
722 723 724
						goto next;
				}
			}
725
		}
726

727 728 729 730 731
		if (unlikely(priv->validate_loopback)) {
			validate_loopback(priv, va);
			goto next;
		}

732 733 734 735
		/*
		 * Packet is OK - process it.
		 */
		length = be32_to_cpu(cqe->byte_cnt);
736
		length -= ring->fcs_del;
737

738 739 740 741 742 743
		/* A bpf program gets first chance to drop the packet. It may
		 * read bytes but not past the end of the frag.
		 */
		if (xdp_prog) {
			struct xdp_buff xdp;
			dma_addr_t dma;
744
			void *orig_data;
745 746
			u32 act;

747
			dma = frags[0].dma + frags[0].page_offset;
748 749 750 751
			dma_sync_single_for_cpu(priv->ddev, dma,
						priv->frag_info[0].frag_size,
						DMA_FROM_DEVICE);

752 753
			xdp.data_hard_start = va - frags[0].page_offset;
			xdp.data = va;
754
			xdp.data_end = xdp.data + length;
755
			orig_data = xdp.data;
756 757

			act = bpf_prog_run_xdp(xdp_prog, &xdp);
758 759 760 761 762

			if (xdp.data != orig_data) {
				length = xdp.data_end - xdp.data;
				frags[0].page_offset = xdp.data -
					xdp.data_hard_start;
763
				va = xdp.data;
764 765
			}

766 767 768
			switch (act) {
			case XDP_PASS:
				break;
769
			case XDP_TX:
770
				if (likely(!mlx4_en_xmit_frame(ring, frags, dev,
771
							length, cq->ring,
772 773 774 775
							&doorbell_pending))) {
					frags[0].page = NULL;
					goto next;
				}
776
				trace_xdp_exception(dev, xdp_prog, act);
777
				goto xdp_drop_no_cnt; /* Drop on xmit failure */
778 779 780
			default:
				bpf_warn_invalid_xdp_action(act);
			case XDP_ABORTED:
781
				trace_xdp_exception(dev, xdp_prog, act);
782
			case XDP_DROP:
783 784
				ring->xdp_drop++;
xdp_drop_no_cnt:
785 786 787 788
				goto next;
			}
		}

789 790 791
		ring->bytes += length;
		ring->packets++;

792 793 794 795 796 797 798 799 800 801 802
		skb = napi_get_frags(&cq->napi);
		if (!skb)
			goto next;

		if (unlikely(ring->hwtstamp_rx_filter == HWTSTAMP_FILTER_ALL)) {
			timestamp = mlx4_en_get_cqe_ts(cqe);
			mlx4_en_fill_hwtstamps(mdev, skb_hwtstamps(skb),
					       timestamp);
		}
		skb_record_rx_queue(skb, cq->ring);

803
		if (likely(dev->features & NETIF_F_RXCSUM)) {
804 805 806 807 808
			if (cqe->status & cpu_to_be16(MLX4_CQE_STATUS_TCP |
						      MLX4_CQE_STATUS_UDP)) {
				if ((cqe->status & cpu_to_be16(MLX4_CQE_STATUS_IPOK)) &&
				    cqe->checksum == cpu_to_be16(0xffff)) {
					ip_summed = CHECKSUM_UNNECESSARY;
809 810 811 812
					l2_tunnel = (dev->hw_enc_features & NETIF_F_RXCSUM) &&
						(cqe->vlan_my_qpn & cpu_to_be32(MLX4_CQE_L2_TUNNEL));
					if (l2_tunnel)
						skb->csum_level = 1;
813 814
					ring->csum_ok++;
				} else {
815
					goto csum_none;
816
				}
817
			} else {
818 819 820
				if (priv->flags & MLX4_EN_FLAG_RX_CSUM_NON_TCP_UDP &&
				    (cqe->status & cpu_to_be16(MLX4_CQE_STATUS_IPV4 |
							       MLX4_CQE_STATUS_IPV6))) {
821 822 823 824 825 826
					if (check_csum(cqe, skb, va, dev->features)) {
						goto csum_none;
					} else {
						ip_summed = CHECKSUM_COMPLETE;
						ring->csum_complete++;
					}
827
				} else {
828
					goto csum_none;
829
				}
830 831
			}
		} else {
832
csum_none:
833
			ip_summed = CHECKSUM_NONE;
834
			ring->csum_none++;
835 836
		}
		skb->ip_summed = ip_summed;
Y
Yevgeny Petrilin 已提交
837
		if (dev->features & NETIF_F_RXHASH)
T
Tom Herbert 已提交
838 839
			skb_set_hash(skb,
				     be32_to_cpu(cqe->immed_rss_invalid),
840 841 842
				     (ip_summed == CHECKSUM_UNNECESSARY) ?
					PKT_HASH_TYPE_L4 :
					PKT_HASH_TYPE_L3);
Y
Yevgeny Petrilin 已提交
843

844 845 846

		if ((cqe->vlan_my_qpn &
		     cpu_to_be32(MLX4_CQE_CVLAN_PRESENT_MASK)) &&
847
		    (dev->features & NETIF_F_HW_VLAN_CTAG_RX))
848 849 850 851
			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
					       be16_to_cpu(cqe->sl_vid));
		else if ((cqe->vlan_my_qpn &
			  cpu_to_be32(MLX4_CQE_SVLAN_PRESENT_MASK)) &&
852 853 854
			 (dev->features & NETIF_F_HW_VLAN_STAG_RX))
			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021AD),
					       be16_to_cpu(cqe->sl_vid));
J
Jiri Pirko 已提交
855

856 857 858 859 860 861 862 863 864
		nr = mlx4_en_complete_rx_desc(priv, frags, skb, length);
		if (likely(nr)) {
			skb_shinfo(skb)->nr_frags = nr;
			skb->len = length;
			skb->data_len = length;
			napi_gro_frags(&cq->napi);
		} else {
			skb->vlan_tci = 0;
			skb_clear_hash(skb);
865
		}
866 867 868
next:
		++cq->mcq.cons_index;
		index = (cq->mcq.cons_index) & ring->size_mask;
869
		cqe = mlx4_en_get_cqe(cq->buf, index, priv->cqe_size) + factor;
870
		if (++polled == budget)
871
			break;
872 873
	}

874
	rcu_read_unlock();
875

876 877 878 879 880 881 882 883
	if (polled) {
		if (doorbell_pending)
			mlx4_en_xmit_doorbell(priv->tx_ring[TX_XDP][cq->ring]);

		mlx4_cq_set_ci(&cq->mcq);
		wmb(); /* ensure HW sees CQ consumer before we post new buffers */
		ring->cons = cq->mcq.cons_index;
	}
884
	AVG_PERF_COUNTER(priv->pstats.rx_coal_avg, polled);
885 886 887 888

	if (mlx4_en_refill_rx_buffers(priv, ring))
		mlx4_en_update_rx_prod_db(ring);

889 890 891 892 893 894 895 896 897
	return polled;
}


void mlx4_en_rx_irq(struct mlx4_cq *mcq)
{
	struct mlx4_en_cq *cq = container_of(mcq, struct mlx4_en_cq, mcq);
	struct mlx4_en_priv *priv = netdev_priv(cq->dev);

E
Eric Dumazet 已提交
898 899
	if (likely(priv->port_up))
		napi_schedule_irqoff(&cq->napi);
900 901 902 903 904 905 906 907 908 909 910 911 912 913 914
	else
		mlx4_en_arm_cq(priv, cq);
}

/* Rx CQ polling - called by NAPI */
int mlx4_en_poll_rx_cq(struct napi_struct *napi, int budget)
{
	struct mlx4_en_cq *cq = container_of(napi, struct mlx4_en_cq, napi);
	struct net_device *dev = cq->dev;
	struct mlx4_en_priv *priv = netdev_priv(dev);
	int done;

	done = mlx4_en_process_rx_cq(dev, cq, budget);

	/* If we used up all the quota - we're probably not done yet... */
915
	if (done == budget) {
916
		const struct cpumask *aff;
917 918
		struct irq_data *idata;
		int cpu_curr;
919

920
		INC_PERF_COUNTER(priv->pstats.napi_quota);
921 922

		cpu_curr = smp_processor_id();
923 924
		idata = irq_desc_get_irq_data(cq->irq_desc);
		aff = irq_data_get_affinity_mask(idata);
925

926 927 928 929
		if (likely(cpumask_test_cpu(cpu_curr, aff)))
			return budget;

		/* Current cpu is not according to smp_irq_affinity -
930 931 932 933
		 * probably affinity changed. Need to stop this NAPI
		 * poll, and restart it on the right CPU.
		 * Try to avoid returning a too small value (like 0),
		 * to not fool net_rx_action() and its netdev_budget
934
		 */
935 936
		if (done)
			done--;
937
	}
E
Eric Dumazet 已提交
938
	/* Done for now */
939 940
	if (napi_complete_done(napi, done))
		mlx4_en_arm_cq(priv, cq);
941 942 943 944 945 946
	return done;
}

void mlx4_en_calc_rx_buf(struct net_device *dev)
{
	struct mlx4_en_priv *priv = netdev_priv(dev);
947
	int eff_mtu = MLX4_EN_EFF_MTU(dev->mtu);
948 949
	int i = 0;

950 951 952
	/* bpf requires buffers to be set up as 1 packet per page.
	 * This only works when num_frags == 1.
	 */
953
	if (priv->tx_ring_num[TX_XDP]) {
954 955 956
		priv->frag_info[0].frag_size = eff_mtu;
		/* This will gain efficient xdp frame recycling at the
		 * expense of more costly truesize accounting
957
		 */
958
		priv->frag_info[0].frag_stride = PAGE_SIZE;
959
		priv->dma_dir = PCI_DMA_BIDIRECTIONAL;
960
		priv->rx_headroom = XDP_PACKET_HEADROOM;
961 962
		i = 1;
	} else {
E
Eric Dumazet 已提交
963 964 965 966 967
		int frag_size_max = 2048, buf_size = 0;

		/* should not happen, right ? */
		if (eff_mtu > PAGE_SIZE + (MLX4_EN_MAX_RX_FRAGS - 1) * 2048)
			frag_size_max = PAGE_SIZE;
968 969

		while (buf_size < eff_mtu) {
E
Eric Dumazet 已提交
970 971
			int frag_stride, frag_size = eff_mtu - buf_size;
			int pad, nb;
E
Eric Dumazet 已提交
972 973

			if (i < MLX4_EN_MAX_RX_FRAGS - 1)
E
Eric Dumazet 已提交
974
				frag_size = min(frag_size, frag_size_max);
E
Eric Dumazet 已提交
975 976

			priv->frag_info[i].frag_size = frag_size;
E
Eric Dumazet 已提交
977 978 979 980 981 982 983 984
			frag_stride = ALIGN(frag_size, SMP_CACHE_BYTES);
			/* We can only pack 2 1536-bytes frames in on 4K page
			 * Therefore, each frame would consume more bytes (truesize)
			 */
			nb = PAGE_SIZE / frag_stride;
			pad = (PAGE_SIZE - nb * frag_stride) / nb;
			pad &= ~(SMP_CACHE_BYTES - 1);
			priv->frag_info[i].frag_stride = frag_stride + pad;
E
Eric Dumazet 已提交
985 986

			buf_size += frag_size;
987 988
			i++;
		}
989
		priv->dma_dir = PCI_DMA_FROMDEVICE;
990
		priv->rx_headroom = 0;
991 992 993 994
	}

	priv->num_frags = i;
	priv->rx_skb_size = eff_mtu;
995
	priv->log_rx_info = ROUNDUP_LOG2(i * sizeof(struct mlx4_en_rx_alloc));
996

J
Joe Perches 已提交
997 998
	en_dbg(DRV, priv, "Rx buffer scatter-list (effective-mtu:%d num_frags:%d):\n",
	       eff_mtu, priv->num_frags);
999
	for (i = 0; i < priv->num_frags; i++) {
1000 1001
		en_dbg(DRV,
		       priv,
E
Eric Dumazet 已提交
1002
		       "  frag:%d - size:%d stride:%d\n",
1003 1004 1005
		       i,
		       priv->frag_info[i].frag_size,
		       priv->frag_info[i].frag_stride);
1006 1007 1008 1009 1010
	}
}

/* RSS related functions */

1011 1012
static int mlx4_en_config_rss_qp(struct mlx4_en_priv *priv, int qpn,
				 struct mlx4_en_rx_ring *ring,
1013 1014 1015 1016 1017 1018 1019
				 enum mlx4_qp_state *state,
				 struct mlx4_qp *qp)
{
	struct mlx4_en_dev *mdev = priv->mdev;
	struct mlx4_qp_context *context;
	int err = 0;

1020 1021
	context = kmalloc(sizeof(*context), GFP_KERNEL);
	if (!context)
1022 1023
		return -ENOMEM;

1024
	err = mlx4_qp_alloc(mdev->dev, qpn, qp, GFP_KERNEL);
1025
	if (err) {
1026
		en_err(priv, "Failed to allocate qp #%x\n", qpn);
1027 1028 1029 1030 1031
		goto out;
	}
	qp->event = mlx4_en_sqp_event;

	memset(context, 0, sizeof *context);
1032
	mlx4_en_fill_qp_context(priv, ring->actual_size, ring->stride, 0, 0,
1033
				qpn, ring->cqn, -1, context);
1034
	context->db_rec_addr = cpu_to_be64(ring->wqres.db.dma);
1035

1036
	/* Cancel FCS removal if FW allows */
1037
	if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP) {
1038
		context->param3 |= cpu_to_be32(1 << 29);
1039 1040 1041 1042
		if (priv->dev->features & NETIF_F_RXFCS)
			ring->fcs_del = 0;
		else
			ring->fcs_del = ETH_FCS_LEN;
1043 1044
	} else
		ring->fcs_del = 0;
1045

1046
	err = mlx4_qp_to_ready(mdev->dev, &ring->wqres.mtt, context, qp, state);
1047 1048 1049 1050
	if (err) {
		mlx4_qp_remove(mdev->dev, qp);
		mlx4_qp_free(mdev->dev, qp);
	}
1051
	mlx4_en_update_rx_prod_db(ring);
1052 1053 1054 1055 1056
out:
	kfree(context);
	return err;
}

1057 1058 1059 1060 1061
int mlx4_en_create_drop_qp(struct mlx4_en_priv *priv)
{
	int err;
	u32 qpn;

M
Matan Barak 已提交
1062 1063
	err = mlx4_qp_reserve_range(priv->mdev->dev, 1, 1, &qpn,
				    MLX4_RESERVE_A0_QP);
1064 1065 1066 1067
	if (err) {
		en_err(priv, "Failed reserving drop qpn\n");
		return err;
	}
1068
	err = mlx4_qp_alloc(priv->mdev->dev, qpn, &priv->drop_qp, GFP_KERNEL);
1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087
	if (err) {
		en_err(priv, "Failed allocating drop qp\n");
		mlx4_qp_release_range(priv->mdev->dev, qpn, 1);
		return err;
	}

	return 0;
}

void mlx4_en_destroy_drop_qp(struct mlx4_en_priv *priv)
{
	u32 qpn;

	qpn = priv->drop_qp.qpn;
	mlx4_qp_remove(priv->mdev->dev, &priv->drop_qp);
	mlx4_qp_free(priv->mdev->dev, &priv->drop_qp);
	mlx4_qp_release_range(priv->mdev->dev, qpn, 1);
}

1088 1089 1090 1091 1092 1093
/* Allocate rx qp's and configure them according to rss map */
int mlx4_en_config_rss_steer(struct mlx4_en_priv *priv)
{
	struct mlx4_en_dev *mdev = priv->mdev;
	struct mlx4_en_rss_map *rss_map = &priv->rss_map;
	struct mlx4_qp_context context;
1094
	struct mlx4_rss_context *rss_context;
1095
	int rss_rings;
1096
	void *ptr;
1097
	u8 rss_mask = (MLX4_RSS_IPV4 | MLX4_RSS_TCP_IPV4 | MLX4_RSS_IPV6 |
1098
			MLX4_RSS_TCP_IPV6);
1099
	int i, qpn;
1100 1101 1102
	int err = 0;
	int good_qps = 0;

1103
	en_dbg(DRV, priv, "Configuring rss steering\n");
1104 1105
	err = mlx4_qp_reserve_range(mdev->dev, priv->rx_ring_num,
				    priv->rx_ring_num,
1106
				    &rss_map->base_qpn, 0);
1107
	if (err) {
1108
		en_err(priv, "Failed reserving %d qps\n", priv->rx_ring_num);
1109 1110 1111
		return err;
	}

1112
	for (i = 0; i < priv->rx_ring_num; i++) {
1113
		qpn = rss_map->base_qpn + i;
1114
		err = mlx4_en_config_rss_qp(priv, qpn, priv->rx_ring[i],
1115 1116 1117 1118 1119 1120 1121 1122 1123
					    &rss_map->state[i],
					    &rss_map->qps[i]);
		if (err)
			goto rss_err;

		++good_qps;
	}

	/* Configure RSS indirection qp */
1124
	err = mlx4_qp_alloc(mdev->dev, priv->base_qpn, &rss_map->indir_qp, GFP_KERNEL);
1125
	if (err) {
1126
		en_err(priv, "Failed to allocate RSS indirection QP\n");
1127
		goto rss_err;
1128 1129 1130
	}
	rss_map->indir_qp.event = mlx4_en_sqp_event;
	mlx4_en_fill_qp_context(priv, 0, 0, 0, 1, priv->base_qpn,
1131
				priv->rx_ring[0]->cqn, -1, &context);
1132

1133 1134 1135 1136 1137
	if (!priv->prof->rss_rings || priv->prof->rss_rings > priv->rx_ring_num)
		rss_rings = priv->rx_ring_num;
	else
		rss_rings = priv->prof->rss_rings;

1138 1139
	ptr = ((void *) &context) + offsetof(struct mlx4_qp_context, pri_path)
					+ MLX4_RSS_OFFSET_IN_QPC_PRI_PATH;
1140
	rss_context = ptr;
1141
	rss_context->base_qpn = cpu_to_be32(ilog2(rss_rings) << 24 |
1142
					    (rss_map->base_qpn));
1143
	rss_context->default_qpn = cpu_to_be32(rss_map->base_qpn);
1144 1145 1146 1147
	if (priv->mdev->profile.udp_rss) {
		rss_mask |=  MLX4_RSS_UDP_IPV4 | MLX4_RSS_UDP_IPV6;
		rss_context->base_qpn_udp = rss_context->default_qpn;
	}
1148 1149 1150 1151 1152 1153

	if (mdev->dev->caps.tunnel_offload_mode == MLX4_TUNNEL_OFFLOAD_MODE_VXLAN) {
		en_info(priv, "Setting RSS context tunnel type to RSS on inner headers\n");
		rss_mask |= MLX4_RSS_BY_INNER_HEADERS;
	}

Y
Yevgeny Petrilin 已提交
1154
	rss_context->flags = rss_mask;
1155
	rss_context->hash_fn = MLX4_RSS_HASH_TOP;
1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166
	if (priv->rss_hash_fn == ETH_RSS_HASH_XOR) {
		rss_context->hash_fn = MLX4_RSS_HASH_XOR;
	} else if (priv->rss_hash_fn == ETH_RSS_HASH_TOP) {
		rss_context->hash_fn = MLX4_RSS_HASH_TOP;
		memcpy(rss_context->rss_key, priv->rss_key,
		       MLX4_EN_RSS_KEY_SIZE);
	} else {
		en_err(priv, "Unknown RSS hash function requested\n");
		err = -EINVAL;
		goto indir_err;
	}
1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185
	err = mlx4_qp_to_ready(mdev->dev, &priv->res.mtt, &context,
			       &rss_map->indir_qp, &rss_map->indir_state);
	if (err)
		goto indir_err;

	return 0;

indir_err:
	mlx4_qp_modify(mdev->dev, NULL, rss_map->indir_state,
		       MLX4_QP_STATE_RST, NULL, 0, 0, &rss_map->indir_qp);
	mlx4_qp_remove(mdev->dev, &rss_map->indir_qp);
	mlx4_qp_free(mdev->dev, &rss_map->indir_qp);
rss_err:
	for (i = 0; i < good_qps; i++) {
		mlx4_qp_modify(mdev->dev, NULL, rss_map->state[i],
			       MLX4_QP_STATE_RST, NULL, 0, 0, &rss_map->qps[i]);
		mlx4_qp_remove(mdev->dev, &rss_map->qps[i]);
		mlx4_qp_free(mdev->dev, &rss_map->qps[i]);
	}
1186
	mlx4_qp_release_range(mdev->dev, rss_map->base_qpn, priv->rx_ring_num);
1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200
	return err;
}

void mlx4_en_release_rss_steer(struct mlx4_en_priv *priv)
{
	struct mlx4_en_dev *mdev = priv->mdev;
	struct mlx4_en_rss_map *rss_map = &priv->rss_map;
	int i;

	mlx4_qp_modify(mdev->dev, NULL, rss_map->indir_state,
		       MLX4_QP_STATE_RST, NULL, 0, 0, &rss_map->indir_qp);
	mlx4_qp_remove(mdev->dev, &rss_map->indir_qp);
	mlx4_qp_free(mdev->dev, &rss_map->indir_qp);

1201
	for (i = 0; i < priv->rx_ring_num; i++) {
1202 1203 1204 1205 1206
		mlx4_qp_modify(mdev->dev, NULL, rss_map->state[i],
			       MLX4_QP_STATE_RST, NULL, 0, 0, &rss_map->qps[i]);
		mlx4_qp_remove(mdev->dev, &rss_map->qps[i]);
		mlx4_qp_free(mdev->dev, &rss_map->qps[i]);
	}
1207
	mlx4_qp_release_range(mdev->dev, rss_map->base_qpn, priv->rx_ring_num);
1208
}