2 * Copyright (c) 2007 Mellanox Technologies. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
34 #include <linux/mlx4/cq.h>
35 #include <linux/slab.h>
36 #include <linux/mlx4/qp.h>
37 #include <linux/skbuff.h>
38 #include <linux/if_ether.h>
39 #include <linux/if_vlan.h>
40 #include <linux/vmalloc.h>
45 static int mlx4_en_alloc_frag(struct mlx4_en_priv *priv,
46 struct mlx4_en_rx_desc *rx_desc,
47 struct page_frag *skb_frags,
48 struct mlx4_en_rx_alloc *ring_alloc,
51 struct mlx4_en_dev *mdev = priv->mdev;
52 struct mlx4_en_frag_info *frag_info = &priv->frag_info[i];
53 struct mlx4_en_rx_alloc *page_alloc = &ring_alloc[i];
57 if (page_alloc->offset == frag_info->last_offset) {
58 /* Allocate new page */
59 page = alloc_pages(GFP_ATOMIC | __GFP_COMP, MLX4_EN_ALLOC_ORDER);
63 skb_frags[i].page = page_alloc->page;
64 skb_frags[i].offset = page_alloc->offset;
65 page_alloc->page = page;
66 page_alloc->offset = frag_info->frag_align;
68 page = page_alloc->page;
71 skb_frags[i].page = page;
72 skb_frags[i].offset = page_alloc->offset;
73 page_alloc->offset += frag_info->frag_stride;
75 dma = pci_map_single(mdev->pdev, page_address(skb_frags[i].page) +
76 skb_frags[i].offset, frag_info->frag_size,
78 rx_desc->data[i].addr = cpu_to_be64(dma);
82 static int mlx4_en_init_allocator(struct mlx4_en_priv *priv,
83 struct mlx4_en_rx_ring *ring)
85 struct mlx4_en_rx_alloc *page_alloc;
88 for (i = 0; i < priv->num_frags; i++) {
89 page_alloc = &ring->page_alloc[i];
90 page_alloc->page = alloc_pages(GFP_ATOMIC | __GFP_COMP,
92 if (!page_alloc->page)
95 page_alloc->offset = priv->frag_info[i].frag_align;
96 en_dbg(DRV, priv, "Initialized allocator:%d with page:%p\n",
103 page_alloc = &ring->page_alloc[i];
104 put_page(page_alloc->page);
105 page_alloc->page = NULL;
110 static void mlx4_en_destroy_allocator(struct mlx4_en_priv *priv,
111 struct mlx4_en_rx_ring *ring)
113 struct mlx4_en_rx_alloc *page_alloc;
116 for (i = 0; i < priv->num_frags; i++) {
117 page_alloc = &ring->page_alloc[i];
118 en_dbg(DRV, priv, "Freeing allocator:%d count:%d\n",
119 i, page_count(page_alloc->page));
121 put_page(page_alloc->page);
122 page_alloc->page = NULL;
127 static void mlx4_en_init_rx_desc(struct mlx4_en_priv *priv,
128 struct mlx4_en_rx_ring *ring, int index)
130 struct mlx4_en_rx_desc *rx_desc = ring->buf + ring->stride * index;
131 struct skb_frag_struct *skb_frags = ring->rx_info +
132 (index << priv->log_rx_info);
136 /* Set size and memtype fields */
137 for (i = 0; i < priv->num_frags; i++) {
138 skb_frag_size_set(&skb_frags[i], priv->frag_info[i].frag_size);
139 rx_desc->data[i].byte_count =
140 cpu_to_be32(priv->frag_info[i].frag_size);
141 rx_desc->data[i].lkey = cpu_to_be32(priv->mdev->mr.key);
144 /* If the number of used fragments does not fill up the ring stride,
145 * remaining (unused) fragments must be padded with null address/size
146 * and a special memory key */
147 possible_frags = (ring->stride - sizeof(struct mlx4_en_rx_desc)) / DS_SIZE;
148 for (i = priv->num_frags; i < possible_frags; i++) {
149 rx_desc->data[i].byte_count = 0;
150 rx_desc->data[i].lkey = cpu_to_be32(MLX4_EN_MEMTYPE_PAD);
151 rx_desc->data[i].addr = 0;
156 static int mlx4_en_prepare_rx_desc(struct mlx4_en_priv *priv,
157 struct mlx4_en_rx_ring *ring, int index)
159 struct mlx4_en_rx_desc *rx_desc = ring->buf + (index * ring->stride);
160 struct page_frag *skb_frags = ring->rx_info +
161 (index << priv->log_rx_info);
164 for (i = 0; i < priv->num_frags; i++)
165 if (mlx4_en_alloc_frag(priv, rx_desc, skb_frags, ring->page_alloc, i))
172 put_page(skb_frags[i].page);
176 static inline void mlx4_en_update_rx_prod_db(struct mlx4_en_rx_ring *ring)
178 *ring->wqres.db.db = cpu_to_be32(ring->prod & 0xffff);
181 static void mlx4_en_free_rx_desc(struct mlx4_en_priv *priv,
182 struct mlx4_en_rx_ring *ring,
185 struct mlx4_en_dev *mdev = priv->mdev;
186 struct page_frag *skb_frags;
187 struct mlx4_en_rx_desc *rx_desc = ring->buf + (index << ring->log_stride);
191 skb_frags = ring->rx_info + (index << priv->log_rx_info);
192 for (nr = 0; nr < priv->num_frags; nr++) {
193 en_dbg(DRV, priv, "Freeing fragment:%d\n", nr);
194 dma = be64_to_cpu(rx_desc->data[nr].addr);
196 en_dbg(DRV, priv, "Unmapping buffer at dma:0x%llx\n", (u64) dma);
197 pci_unmap_single(mdev->pdev, dma, skb_frags[nr].size,
199 put_page(skb_frags[nr].page);
203 static int mlx4_en_fill_rx_buffers(struct mlx4_en_priv *priv)
205 struct mlx4_en_rx_ring *ring;
210 for (buf_ind = 0; buf_ind < priv->prof->rx_ring_size; buf_ind++) {
211 for (ring_ind = 0; ring_ind < priv->rx_ring_num; ring_ind++) {
212 ring = &priv->rx_ring[ring_ind];
214 if (mlx4_en_prepare_rx_desc(priv, ring,
215 ring->actual_size)) {
216 if (ring->actual_size < MLX4_EN_MIN_RX_SIZE) {
217 en_err(priv, "Failed to allocate "
218 "enough rx buffers\n");
221 new_size = rounddown_pow_of_two(ring->actual_size);
222 en_warn(priv, "Only %d buffers allocated "
223 "reducing ring size to %d",
224 ring->actual_size, new_size);
235 for (ring_ind = 0; ring_ind < priv->rx_ring_num; ring_ind++) {
236 ring = &priv->rx_ring[ring_ind];
237 while (ring->actual_size > new_size) {
240 mlx4_en_free_rx_desc(priv, ring, ring->actual_size);
247 static void mlx4_en_free_rx_buf(struct mlx4_en_priv *priv,
248 struct mlx4_en_rx_ring *ring)
252 en_dbg(DRV, priv, "Freeing Rx buf - cons:%d prod:%d\n",
253 ring->cons, ring->prod);
255 /* Unmap and free Rx buffers */
256 BUG_ON((u32) (ring->prod - ring->cons) > ring->actual_size);
257 while (ring->cons != ring->prod) {
258 index = ring->cons & ring->size_mask;
259 en_dbg(DRV, priv, "Processing descriptor:%d\n", index);
260 mlx4_en_free_rx_desc(priv, ring, index);
265 int mlx4_en_create_rx_ring(struct mlx4_en_priv *priv,
266 struct mlx4_en_rx_ring *ring, u32 size, u16 stride)
268 struct mlx4_en_dev *mdev = priv->mdev;
276 ring->size_mask = size - 1;
277 ring->stride = stride;
278 ring->log_stride = ffs(ring->stride) - 1;
279 ring->buf_size = ring->size * ring->stride + TXBB_SIZE;
281 tmp = size * roundup_pow_of_two(MLX4_EN_MAX_RX_FRAGS *
282 sizeof(struct skb_frag_struct));
283 ring->rx_info = vmalloc(tmp);
287 en_dbg(DRV, priv, "Allocated rx_info ring at addr:%p size:%d\n",
290 err = mlx4_alloc_hwq_res(mdev->dev, &ring->wqres,
291 ring->buf_size, 2 * PAGE_SIZE);
295 err = mlx4_en_map_buffer(&ring->wqres.buf);
297 en_err(priv, "Failed to map RX buffer\n");
300 ring->buf = ring->wqres.buf.direct.buf;
305 mlx4_free_hwq_res(mdev->dev, &ring->wqres, ring->buf_size);
307 vfree(ring->rx_info);
308 ring->rx_info = NULL;
312 int mlx4_en_activate_rx_rings(struct mlx4_en_priv *priv)
314 struct mlx4_en_rx_ring *ring;
318 int stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
319 DS_SIZE * priv->num_frags);
321 for (ring_ind = 0; ring_ind < priv->rx_ring_num; ring_ind++) {
322 ring = &priv->rx_ring[ring_ind];
326 ring->actual_size = 0;
327 ring->cqn = priv->rx_cq[ring_ind].mcq.cqn;
329 ring->stride = stride;
330 if (ring->stride <= TXBB_SIZE)
331 ring->buf += TXBB_SIZE;
333 ring->log_stride = ffs(ring->stride) - 1;
334 ring->buf_size = ring->size * ring->stride;
336 memset(ring->buf, 0, ring->buf_size);
337 mlx4_en_update_rx_prod_db(ring);
339 /* Initailize all descriptors */
340 for (i = 0; i < ring->size; i++)
341 mlx4_en_init_rx_desc(priv, ring, i);
343 /* Initialize page allocators */
344 err = mlx4_en_init_allocator(priv, ring);
346 en_err(priv, "Failed initializing ring allocator\n");
347 if (ring->stride <= TXBB_SIZE)
348 ring->buf -= TXBB_SIZE;
353 err = mlx4_en_fill_rx_buffers(priv);
357 for (ring_ind = 0; ring_ind < priv->rx_ring_num; ring_ind++) {
358 ring = &priv->rx_ring[ring_ind];
360 ring->size_mask = ring->actual_size - 1;
361 mlx4_en_update_rx_prod_db(ring);
367 for (ring_ind = 0; ring_ind < priv->rx_ring_num; ring_ind++)
368 mlx4_en_free_rx_buf(priv, &priv->rx_ring[ring_ind]);
370 ring_ind = priv->rx_ring_num - 1;
372 while (ring_ind >= 0) {
373 if (priv->rx_ring[ring_ind].stride <= TXBB_SIZE)
374 priv->rx_ring[ring_ind].buf -= TXBB_SIZE;
375 mlx4_en_destroy_allocator(priv, &priv->rx_ring[ring_ind]);
381 void mlx4_en_destroy_rx_ring(struct mlx4_en_priv *priv,
382 struct mlx4_en_rx_ring *ring)
384 struct mlx4_en_dev *mdev = priv->mdev;
386 mlx4_en_unmap_buffer(&ring->wqres.buf);
387 mlx4_free_hwq_res(mdev->dev, &ring->wqres, ring->buf_size + TXBB_SIZE);
388 vfree(ring->rx_info);
389 ring->rx_info = NULL;
392 void mlx4_en_deactivate_rx_ring(struct mlx4_en_priv *priv,
393 struct mlx4_en_rx_ring *ring)
395 mlx4_en_free_rx_buf(priv, ring);
396 if (ring->stride <= TXBB_SIZE)
397 ring->buf -= TXBB_SIZE;
398 mlx4_en_destroy_allocator(priv, ring);
402 /* Unmap a completed descriptor and free unused pages */
403 static int mlx4_en_complete_rx_desc(struct mlx4_en_priv *priv,
404 struct mlx4_en_rx_desc *rx_desc,
405 struct page_frag *skb_frags,
407 struct mlx4_en_rx_alloc *page_alloc,
410 struct skb_frag_struct *skb_frags_rx = skb_shinfo(skb)->frags;
411 struct mlx4_en_dev *mdev = priv->mdev;
412 struct mlx4_en_frag_info *frag_info;
416 /* Collect used fragments while replacing them in the HW descirptors */
417 for (nr = 0; nr < priv->num_frags; nr++) {
418 frag_info = &priv->frag_info[nr];
419 if (length <= frag_info->frag_prefix_size)
422 /* Save page reference in skb */
423 __skb_frag_set_page(&skb_frags_rx[nr], skb_frags[nr].page);
424 skb_frag_size_set(&skb_frags_rx[nr], skb_frags[nr].size);
425 skb_frags_rx[nr].page_offset = skb_frags[nr].offset;
426 skb->truesize += frag_info->frag_stride;
427 dma = be64_to_cpu(rx_desc->data[nr].addr);
429 /* Allocate a replacement page */
430 if (mlx4_en_alloc_frag(priv, rx_desc, skb_frags, page_alloc, nr))
434 pci_unmap_single(mdev->pdev, dma, skb_frag_size(&skb_frags_rx[nr]),
437 /* Adjust size of last fragment to match actual length */
439 skb_frag_size_set(&skb_frags_rx[nr - 1],
440 length - priv->frag_info[nr - 1].frag_prefix_size);
444 /* Drop all accumulated fragments (which have already been replaced in
445 * the descriptor) of this packet; remaining fragments are reused... */
448 __skb_frag_unref(&skb_frags_rx[nr]);
454 static struct sk_buff *mlx4_en_rx_skb(struct mlx4_en_priv *priv,
455 struct mlx4_en_rx_desc *rx_desc,
456 struct page_frag *skb_frags,
457 struct mlx4_en_rx_alloc *page_alloc,
460 struct mlx4_en_dev *mdev = priv->mdev;
466 skb = dev_alloc_skb(SMALL_PACKET_SIZE + NET_IP_ALIGN);
468 en_dbg(RX_ERR, priv, "Failed allocating skb\n");
471 skb->dev = priv->dev;
472 skb_reserve(skb, NET_IP_ALIGN);
475 /* Get pointer to first fragment so we could copy the headers into the
476 * (linear part of the) skb */
477 va = page_address(skb_frags[0].page) + skb_frags[0].offset;
479 if (length <= SMALL_PACKET_SIZE) {
480 /* We are copying all relevant data to the skb - temporarily
481 * synch buffers for the copy */
482 dma = be64_to_cpu(rx_desc->data[0].addr);
483 dma_sync_single_for_cpu(&mdev->pdev->dev, dma, length,
485 skb_copy_to_linear_data(skb, va, length);
486 dma_sync_single_for_device(&mdev->pdev->dev, dma, length,
491 /* Move relevant fragments to skb */
492 used_frags = mlx4_en_complete_rx_desc(priv, rx_desc, skb_frags,
493 skb, page_alloc, length);
494 if (unlikely(!used_frags)) {
498 skb_shinfo(skb)->nr_frags = used_frags;
500 /* Copy headers into the skb linear buffer */
501 memcpy(skb->data, va, HEADER_COPY_SIZE);
502 skb->tail += HEADER_COPY_SIZE;
504 /* Skip headers in first fragment */
505 skb_shinfo(skb)->frags[0].page_offset += HEADER_COPY_SIZE;
507 /* Adjust size of first fragment */
508 skb_frag_size_sub(&skb_shinfo(skb)->frags[0], HEADER_COPY_SIZE);
509 skb->data_len = length - HEADER_COPY_SIZE;
514 static void validate_loopback(struct mlx4_en_priv *priv, struct sk_buff *skb)
517 int offset = ETH_HLEN;
519 for (i = 0; i < MLX4_LOOPBACK_TEST_PAYLOAD; i++, offset++) {
520 if (*(skb->data + offset) != (unsigned char) (i & 0xff))
524 priv->loopback_ok = 1;
527 dev_kfree_skb_any(skb);
530 int mlx4_en_process_rx_cq(struct net_device *dev, struct mlx4_en_cq *cq, int budget)
532 struct mlx4_en_priv *priv = netdev_priv(dev);
533 struct mlx4_cqe *cqe;
534 struct mlx4_en_rx_ring *ring = &priv->rx_ring[cq->ring];
535 struct page_frag *skb_frags;
536 struct mlx4_en_rx_desc *rx_desc;
549 /* We assume a 1:1 mapping between CQEs and Rx descriptors, so Rx
550 * descriptor offset can be deduced from the CQE index instead of
551 * reading 'cqe->index' */
552 index = cq->mcq.cons_index & ring->size_mask;
553 cqe = &cq->buf[index];
555 /* Process all completed CQEs */
556 while (XNOR(cqe->owner_sr_opcode & MLX4_CQE_OWNER_MASK,
557 cq->mcq.cons_index & cq->size)) {
559 skb_frags = ring->rx_info + (index << priv->log_rx_info);
560 rx_desc = ring->buf + (index << ring->log_stride);
563 * make sure we read the CQE after we read the ownership bit
567 /* Drop packet on bad receive or bad checksum */
568 if (unlikely((cqe->owner_sr_opcode & MLX4_CQE_OPCODE_MASK) ==
569 MLX4_CQE_OPCODE_ERROR)) {
570 en_err(priv, "CQE completed in error - vendor "
571 "syndrom:%d syndrom:%d\n",
572 ((struct mlx4_err_cqe *) cqe)->vendor_err_syndrome,
573 ((struct mlx4_err_cqe *) cqe)->syndrome);
576 if (unlikely(cqe->badfcs_enc & MLX4_CQE_BAD_FCS)) {
577 en_dbg(RX_ERR, priv, "Accepted frame with bad FCS\n");
581 /* Get pointer to first fragment since we haven't skb yet and
582 * cast it to ethhdr struct */
583 ethh = (struct ethhdr *)(page_address(skb_frags[0].page) +
584 skb_frags[0].offset);
585 s_mac = mlx4_en_mac_to_u64(ethh->h_source);
587 /* If source MAC is equal to our own MAC and not performing
588 * the selftest or flb disabled - drop the packet */
589 if (s_mac == priv->mac &&
590 (!(dev->features & NETIF_F_LOOPBACK) ||
591 !priv->validate_loopback))
595 * Packet is OK - process it.
597 length = be32_to_cpu(cqe->byte_cnt);
598 length -= ring->fcs_del;
599 ring->bytes += length;
602 if (likely(dev->features & NETIF_F_RXCSUM)) {
603 if ((cqe->status & cpu_to_be16(MLX4_CQE_STATUS_IPOK)) &&
604 (cqe->checksum == cpu_to_be16(0xffff))) {
606 /* This packet is eligible for LRO if it is:
607 * - DIX Ethernet (type interpretation)
609 * - without IP options
610 * - not an IP fragment */
611 if (dev->features & NETIF_F_GRO) {
612 struct sk_buff *gro_skb = napi_get_frags(&cq->napi);
616 nr = mlx4_en_complete_rx_desc(
619 ring->page_alloc, length);
623 skb_shinfo(gro_skb)->nr_frags = nr;
624 gro_skb->len = length;
625 gro_skb->data_len = length;
626 gro_skb->ip_summed = CHECKSUM_UNNECESSARY;
628 if (cqe->vlan_my_qpn &
629 cpu_to_be32(MLX4_CQE_VLAN_PRESENT_MASK)) {
630 u16 vid = be16_to_cpu(cqe->sl_vid);
632 __vlan_hwaccel_put_tag(gro_skb, vid);
635 if (dev->features & NETIF_F_RXHASH)
636 gro_skb->rxhash = be32_to_cpu(cqe->immed_rss_invalid);
638 skb_record_rx_queue(gro_skb, cq->ring);
639 napi_gro_frags(&cq->napi);
644 /* LRO not possible, complete processing here */
645 ip_summed = CHECKSUM_UNNECESSARY;
647 ip_summed = CHECKSUM_NONE;
651 ip_summed = CHECKSUM_NONE;
655 skb = mlx4_en_rx_skb(priv, rx_desc, skb_frags,
656 ring->page_alloc, length);
658 priv->stats.rx_dropped++;
662 if (unlikely(priv->validate_loopback)) {
663 validate_loopback(priv, skb);
667 skb->ip_summed = ip_summed;
668 skb->protocol = eth_type_trans(skb, dev);
669 skb_record_rx_queue(skb, cq->ring);
671 if (dev->features & NETIF_F_RXHASH)
672 skb->rxhash = be32_to_cpu(cqe->immed_rss_invalid);
674 if (be32_to_cpu(cqe->vlan_my_qpn) &
675 MLX4_CQE_VLAN_PRESENT_MASK)
676 __vlan_hwaccel_put_tag(skb, be16_to_cpu(cqe->sl_vid));
678 /* Push it up the stack */
679 netif_receive_skb(skb);
682 ++cq->mcq.cons_index;
683 index = (cq->mcq.cons_index) & ring->size_mask;
684 cqe = &cq->buf[index];
685 if (++polled == budget) {
686 /* We are here because we reached the NAPI budget -
687 * flush only pending LRO sessions */
693 AVG_PERF_COUNTER(priv->pstats.rx_coal_avg, polled);
694 mlx4_cq_set_ci(&cq->mcq);
695 wmb(); /* ensure HW sees CQ consumer before we post new buffers */
696 ring->cons = cq->mcq.cons_index;
697 ring->prod += polled; /* Polled descriptors were realocated in place */
698 mlx4_en_update_rx_prod_db(ring);
703 void mlx4_en_rx_irq(struct mlx4_cq *mcq)
705 struct mlx4_en_cq *cq = container_of(mcq, struct mlx4_en_cq, mcq);
706 struct mlx4_en_priv *priv = netdev_priv(cq->dev);
709 napi_schedule(&cq->napi);
711 mlx4_en_arm_cq(priv, cq);
714 /* Rx CQ polling - called by NAPI */
715 int mlx4_en_poll_rx_cq(struct napi_struct *napi, int budget)
717 struct mlx4_en_cq *cq = container_of(napi, struct mlx4_en_cq, napi);
718 struct net_device *dev = cq->dev;
719 struct mlx4_en_priv *priv = netdev_priv(dev);
722 done = mlx4_en_process_rx_cq(dev, cq, budget);
724 /* If we used up all the quota - we're probably not done yet... */
726 INC_PERF_COUNTER(priv->pstats.napi_quota);
730 mlx4_en_arm_cq(priv, cq);
736 /* Calculate the last offset position that accommodates a full fragment
737 * (assuming fagment size = stride-align) */
738 static int mlx4_en_last_alloc_offset(struct mlx4_en_priv *priv, u16 stride, u16 align)
740 u16 res = MLX4_EN_ALLOC_SIZE % stride;
741 u16 offset = MLX4_EN_ALLOC_SIZE - stride - res + align;
743 en_dbg(DRV, priv, "Calculated last offset for stride:%d align:%d "
744 "res:%d offset:%d\n", stride, align, res, offset);
749 static int frag_sizes[] = {
756 void mlx4_en_calc_rx_buf(struct net_device *dev)
758 struct mlx4_en_priv *priv = netdev_priv(dev);
759 int eff_mtu = dev->mtu + ETH_HLEN + VLAN_HLEN + ETH_LLC_SNAP_SIZE;
763 while (buf_size < eff_mtu) {
764 priv->frag_info[i].frag_size =
765 (eff_mtu > buf_size + frag_sizes[i]) ?
766 frag_sizes[i] : eff_mtu - buf_size;
767 priv->frag_info[i].frag_prefix_size = buf_size;
769 priv->frag_info[i].frag_align = NET_IP_ALIGN;
770 priv->frag_info[i].frag_stride =
771 ALIGN(frag_sizes[i] + NET_IP_ALIGN, SMP_CACHE_BYTES);
773 priv->frag_info[i].frag_align = 0;
774 priv->frag_info[i].frag_stride =
775 ALIGN(frag_sizes[i], SMP_CACHE_BYTES);
777 priv->frag_info[i].last_offset = mlx4_en_last_alloc_offset(
778 priv, priv->frag_info[i].frag_stride,
779 priv->frag_info[i].frag_align);
780 buf_size += priv->frag_info[i].frag_size;
785 priv->rx_skb_size = eff_mtu;
786 priv->log_rx_info = ROUNDUP_LOG2(i * sizeof(struct skb_frag_struct));
788 en_dbg(DRV, priv, "Rx buffer scatter-list (effective-mtu:%d "
789 "num_frags:%d):\n", eff_mtu, priv->num_frags);
790 for (i = 0; i < priv->num_frags; i++) {
791 en_dbg(DRV, priv, " frag:%d - size:%d prefix:%d align:%d "
792 "stride:%d last_offset:%d\n", i,
793 priv->frag_info[i].frag_size,
794 priv->frag_info[i].frag_prefix_size,
795 priv->frag_info[i].frag_align,
796 priv->frag_info[i].frag_stride,
797 priv->frag_info[i].last_offset);
801 /* RSS related functions */
803 static int mlx4_en_config_rss_qp(struct mlx4_en_priv *priv, int qpn,
804 struct mlx4_en_rx_ring *ring,
805 enum mlx4_qp_state *state,
808 struct mlx4_en_dev *mdev = priv->mdev;
809 struct mlx4_qp_context *context;
812 context = kmalloc(sizeof *context , GFP_KERNEL);
814 en_err(priv, "Failed to allocate qp context\n");
818 err = mlx4_qp_alloc(mdev->dev, qpn, qp);
820 en_err(priv, "Failed to allocate qp #%x\n", qpn);
823 qp->event = mlx4_en_sqp_event;
825 memset(context, 0, sizeof *context);
826 mlx4_en_fill_qp_context(priv, ring->actual_size, ring->stride, 0, 0,
827 qpn, ring->cqn, context);
828 context->db_rec_addr = cpu_to_be64(ring->wqres.db.dma);
830 /* Cancel FCS removal if FW allows */
831 if (mdev->dev->caps.flags & MLX4_DEV_CAP_FLAG_FCS_KEEP) {
832 context->param3 |= cpu_to_be32(1 << 29);
833 ring->fcs_del = ETH_FCS_LEN;
837 err = mlx4_qp_to_ready(mdev->dev, &ring->wqres.mtt, context, qp, state);
839 mlx4_qp_remove(mdev->dev, qp);
840 mlx4_qp_free(mdev->dev, qp);
842 mlx4_en_update_rx_prod_db(ring);
848 /* Allocate rx qp's and configure them according to rss map */
849 int mlx4_en_config_rss_steer(struct mlx4_en_priv *priv)
851 struct mlx4_en_dev *mdev = priv->mdev;
852 struct mlx4_en_rss_map *rss_map = &priv->rss_map;
853 struct mlx4_qp_context context;
854 struct mlx4_rss_context *rss_context;
857 u8 rss_mask = (MLX4_RSS_IPV4 | MLX4_RSS_TCP_IPV4 | MLX4_RSS_IPV6 |
862 static const u32 rsskey[10] = { 0xD181C62C, 0xF7F4DB5B, 0x1983A2FC,
863 0x943E1ADB, 0xD9389E6B, 0xD1039C2C, 0xA74499AD,
864 0x593D56D9, 0xF3253C06, 0x2ADC1FFC};
866 en_dbg(DRV, priv, "Configuring rss steering\n");
867 err = mlx4_qp_reserve_range(mdev->dev, priv->rx_ring_num,
871 en_err(priv, "Failed reserving %d qps\n", priv->rx_ring_num);
875 for (i = 0; i < priv->rx_ring_num; i++) {
876 qpn = rss_map->base_qpn + i;
877 err = mlx4_en_config_rss_qp(priv, qpn, &priv->rx_ring[i],
886 /* Configure RSS indirection qp */
887 err = mlx4_qp_alloc(mdev->dev, priv->base_qpn, &rss_map->indir_qp);
889 en_err(priv, "Failed to allocate RSS indirection QP\n");
892 rss_map->indir_qp.event = mlx4_en_sqp_event;
893 mlx4_en_fill_qp_context(priv, 0, 0, 0, 1, priv->base_qpn,
894 priv->rx_ring[0].cqn, &context);
896 if (!priv->prof->rss_rings || priv->prof->rss_rings > priv->rx_ring_num)
897 rss_rings = priv->rx_ring_num;
899 rss_rings = priv->prof->rss_rings;
901 ptr = ((void *) &context) + offsetof(struct mlx4_qp_context, pri_path)
902 + MLX4_RSS_OFFSET_IN_QPC_PRI_PATH;
904 rss_context->base_qpn = cpu_to_be32(ilog2(rss_rings) << 24 |
905 (rss_map->base_qpn));
906 rss_context->default_qpn = cpu_to_be32(rss_map->base_qpn);
907 if (priv->mdev->profile.udp_rss) {
908 rss_mask |= MLX4_RSS_UDP_IPV4 | MLX4_RSS_UDP_IPV6;
909 rss_context->base_qpn_udp = rss_context->default_qpn;
911 rss_context->flags = rss_mask;
912 rss_context->hash_fn = MLX4_RSS_HASH_TOP;
913 for (i = 0; i < 10; i++)
914 rss_context->rss_key[i] = rsskey[i];
916 err = mlx4_qp_to_ready(mdev->dev, &priv->res.mtt, &context,
917 &rss_map->indir_qp, &rss_map->indir_state);
924 mlx4_qp_modify(mdev->dev, NULL, rss_map->indir_state,
925 MLX4_QP_STATE_RST, NULL, 0, 0, &rss_map->indir_qp);
926 mlx4_qp_remove(mdev->dev, &rss_map->indir_qp);
927 mlx4_qp_free(mdev->dev, &rss_map->indir_qp);
929 for (i = 0; i < good_qps; i++) {
930 mlx4_qp_modify(mdev->dev, NULL, rss_map->state[i],
931 MLX4_QP_STATE_RST, NULL, 0, 0, &rss_map->qps[i]);
932 mlx4_qp_remove(mdev->dev, &rss_map->qps[i]);
933 mlx4_qp_free(mdev->dev, &rss_map->qps[i]);
935 mlx4_qp_release_range(mdev->dev, rss_map->base_qpn, priv->rx_ring_num);
939 void mlx4_en_release_rss_steer(struct mlx4_en_priv *priv)
941 struct mlx4_en_dev *mdev = priv->mdev;
942 struct mlx4_en_rss_map *rss_map = &priv->rss_map;
945 mlx4_qp_modify(mdev->dev, NULL, rss_map->indir_state,
946 MLX4_QP_STATE_RST, NULL, 0, 0, &rss_map->indir_qp);
947 mlx4_qp_remove(mdev->dev, &rss_map->indir_qp);
948 mlx4_qp_free(mdev->dev, &rss_map->indir_qp);
950 for (i = 0; i < priv->rx_ring_num; i++) {
951 mlx4_qp_modify(mdev->dev, NULL, rss_map->state[i],
952 MLX4_QP_STATE_RST, NULL, 0, 0, &rss_map->qps[i]);
953 mlx4_qp_remove(mdev->dev, &rss_map->qps[i]);
954 mlx4_qp_free(mdev->dev, &rss_map->qps[i]);
956 mlx4_qp_release_range(mdev->dev, rss_map->base_qpn, priv->rx_ring_num);