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 skb_frag_struct *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].page_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].page_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].page_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_frags[i].size = 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 skb_frag_struct *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 skb_frag_struct *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);
284 if (!ring->rx_info) {
285 en_err(priv, "Failed allocating rx_info ring\n");
288 en_dbg(DRV, priv, "Allocated rx_info ring at addr:%p size:%d\n",
291 err = mlx4_alloc_hwq_res(mdev->dev, &ring->wqres,
292 ring->buf_size, 2 * PAGE_SIZE);
296 err = mlx4_en_map_buffer(&ring->wqres.buf);
298 en_err(priv, "Failed to map RX buffer\n");
301 ring->buf = ring->wqres.buf.direct.buf;
306 mlx4_free_hwq_res(mdev->dev, &ring->wqres, ring->buf_size);
308 vfree(ring->rx_info);
309 ring->rx_info = NULL;
313 int mlx4_en_activate_rx_rings(struct mlx4_en_priv *priv)
315 struct mlx4_en_rx_ring *ring;
319 int stride = roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc) +
320 DS_SIZE * priv->num_frags);
322 for (ring_ind = 0; ring_ind < priv->rx_ring_num; ring_ind++) {
323 ring = &priv->rx_ring[ring_ind];
327 ring->actual_size = 0;
328 ring->cqn = priv->rx_cq[ring_ind].mcq.cqn;
330 ring->stride = stride;
331 if (ring->stride <= TXBB_SIZE)
332 ring->buf += TXBB_SIZE;
334 ring->log_stride = ffs(ring->stride) - 1;
335 ring->buf_size = ring->size * ring->stride;
337 memset(ring->buf, 0, ring->buf_size);
338 mlx4_en_update_rx_prod_db(ring);
340 /* Initailize all descriptors */
341 for (i = 0; i < ring->size; i++)
342 mlx4_en_init_rx_desc(priv, ring, i);
344 /* Initialize page allocators */
345 err = mlx4_en_init_allocator(priv, ring);
347 en_err(priv, "Failed initializing ring allocator\n");
352 err = mlx4_en_fill_rx_buffers(priv);
356 for (ring_ind = 0; ring_ind < priv->rx_ring_num; ring_ind++) {
357 ring = &priv->rx_ring[ring_ind];
359 ring->size_mask = ring->actual_size - 1;
360 mlx4_en_update_rx_prod_db(ring);
366 for (ring_ind = 0; ring_ind < priv->rx_ring_num; ring_ind++)
367 mlx4_en_free_rx_buf(priv, &priv->rx_ring[ring_ind]);
369 ring_ind = priv->rx_ring_num - 1;
371 while (ring_ind >= 0) {
372 mlx4_en_destroy_allocator(priv, &priv->rx_ring[ring_ind]);
378 void mlx4_en_destroy_rx_ring(struct mlx4_en_priv *priv,
379 struct mlx4_en_rx_ring *ring)
381 struct mlx4_en_dev *mdev = priv->mdev;
383 mlx4_en_unmap_buffer(&ring->wqres.buf);
384 mlx4_free_hwq_res(mdev->dev, &ring->wqres, ring->buf_size + TXBB_SIZE);
385 vfree(ring->rx_info);
386 ring->rx_info = NULL;
389 void mlx4_en_deactivate_rx_ring(struct mlx4_en_priv *priv,
390 struct mlx4_en_rx_ring *ring)
392 mlx4_en_free_rx_buf(priv, ring);
393 if (ring->stride <= TXBB_SIZE)
394 ring->buf -= TXBB_SIZE;
395 mlx4_en_destroy_allocator(priv, ring);
399 /* Unmap a completed descriptor and free unused pages */
400 static int mlx4_en_complete_rx_desc(struct mlx4_en_priv *priv,
401 struct mlx4_en_rx_desc *rx_desc,
402 struct skb_frag_struct *skb_frags,
403 struct skb_frag_struct *skb_frags_rx,
404 struct mlx4_en_rx_alloc *page_alloc,
407 struct mlx4_en_dev *mdev = priv->mdev;
408 struct mlx4_en_frag_info *frag_info;
412 /* Collect used fragments while replacing them in the HW descirptors */
413 for (nr = 0; nr < priv->num_frags; nr++) {
414 frag_info = &priv->frag_info[nr];
415 if (length <= frag_info->frag_prefix_size)
418 /* Save page reference in skb */
419 skb_frags_rx[nr].page = skb_frags[nr].page;
420 skb_frags_rx[nr].size = skb_frags[nr].size;
421 skb_frags_rx[nr].page_offset = skb_frags[nr].page_offset;
422 dma = be64_to_cpu(rx_desc->data[nr].addr);
424 /* Allocate a replacement page */
425 if (mlx4_en_alloc_frag(priv, rx_desc, skb_frags, page_alloc, nr))
429 pci_unmap_single(mdev->pdev, dma, skb_frags_rx[nr].size,
432 /* Adjust size of last fragment to match actual length */
434 skb_frags_rx[nr - 1].size = length -
435 priv->frag_info[nr - 1].frag_prefix_size;
439 /* Drop all accumulated fragments (which have already been replaced in
440 * the descriptor) of this packet; remaining fragments are reused... */
443 put_page(skb_frags_rx[nr].page);
449 static struct sk_buff *mlx4_en_rx_skb(struct mlx4_en_priv *priv,
450 struct mlx4_en_rx_desc *rx_desc,
451 struct skb_frag_struct *skb_frags,
452 struct mlx4_en_rx_alloc *page_alloc,
455 struct mlx4_en_dev *mdev = priv->mdev;
461 skb = dev_alloc_skb(SMALL_PACKET_SIZE + NET_IP_ALIGN);
463 en_dbg(RX_ERR, priv, "Failed allocating skb\n");
466 skb->dev = priv->dev;
467 skb_reserve(skb, NET_IP_ALIGN);
469 skb->truesize = length + sizeof(struct sk_buff);
471 /* Get pointer to first fragment so we could copy the headers into the
472 * (linear part of the) skb */
473 va = page_address(skb_frags[0].page) + skb_frags[0].page_offset;
475 if (length <= SMALL_PACKET_SIZE) {
476 /* We are copying all relevant data to the skb - temporarily
477 * synch buffers for the copy */
478 dma = be64_to_cpu(rx_desc->data[0].addr);
479 dma_sync_single_for_cpu(&mdev->pdev->dev, dma, length,
481 skb_copy_to_linear_data(skb, va, length);
482 dma_sync_single_for_device(&mdev->pdev->dev, dma, length,
487 /* Move relevant fragments to skb */
488 used_frags = mlx4_en_complete_rx_desc(priv, rx_desc, skb_frags,
489 skb_shinfo(skb)->frags,
491 if (unlikely(!used_frags)) {
495 skb_shinfo(skb)->nr_frags = used_frags;
497 /* Copy headers into the skb linear buffer */
498 memcpy(skb->data, va, HEADER_COPY_SIZE);
499 skb->tail += HEADER_COPY_SIZE;
501 /* Skip headers in first fragment */
502 skb_shinfo(skb)->frags[0].page_offset += HEADER_COPY_SIZE;
504 /* Adjust size of first fragment */
505 skb_shinfo(skb)->frags[0].size -= HEADER_COPY_SIZE;
506 skb->data_len = length - HEADER_COPY_SIZE;
511 static void validate_loopback(struct mlx4_en_priv *priv, struct sk_buff *skb)
514 int offset = ETH_HLEN;
516 for (i = 0; i < MLX4_LOOPBACK_TEST_PAYLOAD; i++, offset++) {
517 if (*(skb->data + offset) != (unsigned char) (i & 0xff))
521 priv->loopback_ok = 1;
524 dev_kfree_skb_any(skb);
527 int mlx4_en_process_rx_cq(struct net_device *dev, struct mlx4_en_cq *cq, int budget)
529 struct mlx4_en_priv *priv = netdev_priv(dev);
530 struct mlx4_cqe *cqe;
531 struct mlx4_en_rx_ring *ring = &priv->rx_ring[cq->ring];
532 struct skb_frag_struct *skb_frags;
533 struct mlx4_en_rx_desc *rx_desc;
544 /* We assume a 1:1 mapping between CQEs and Rx descriptors, so Rx
545 * descriptor offset can be deduced from the CQE index instead of
546 * reading 'cqe->index' */
547 index = cq->mcq.cons_index & ring->size_mask;
548 cqe = &cq->buf[index];
550 /* Process all completed CQEs */
551 while (XNOR(cqe->owner_sr_opcode & MLX4_CQE_OWNER_MASK,
552 cq->mcq.cons_index & cq->size)) {
554 skb_frags = ring->rx_info + (index << priv->log_rx_info);
555 rx_desc = ring->buf + (index << ring->log_stride);
558 * make sure we read the CQE after we read the ownership bit
562 /* Drop packet on bad receive or bad checksum */
563 if (unlikely((cqe->owner_sr_opcode & MLX4_CQE_OPCODE_MASK) ==
564 MLX4_CQE_OPCODE_ERROR)) {
565 en_err(priv, "CQE completed in error - vendor "
566 "syndrom:%d syndrom:%d\n",
567 ((struct mlx4_err_cqe *) cqe)->vendor_err_syndrome,
568 ((struct mlx4_err_cqe *) cqe)->syndrome);
571 if (unlikely(cqe->badfcs_enc & MLX4_CQE_BAD_FCS)) {
572 en_dbg(RX_ERR, priv, "Accepted frame with bad FCS\n");
577 * Packet is OK - process it.
579 length = be32_to_cpu(cqe->byte_cnt);
580 ring->bytes += length;
583 if (likely(priv->rx_csum)) {
584 if ((cqe->status & cpu_to_be16(MLX4_CQE_STATUS_IPOK)) &&
585 (cqe->checksum == cpu_to_be16(0xffff))) {
586 priv->port_stats.rx_chksum_good++;
587 /* This packet is eligible for LRO if it is:
588 * - DIX Ethernet (type interpretation)
590 * - without IP options
591 * - not an IP fragment */
592 if (dev->features & NETIF_F_GRO) {
593 struct sk_buff *gro_skb = napi_get_frags(&cq->napi);
595 nr = mlx4_en_complete_rx_desc(
597 skb_frags, skb_shinfo(gro_skb)->frags,
598 ring->page_alloc, length);
602 skb_shinfo(gro_skb)->nr_frags = nr;
603 gro_skb->len = length;
604 gro_skb->data_len = length;
605 gro_skb->truesize += length;
606 gro_skb->ip_summed = CHECKSUM_UNNECESSARY;
608 if (priv->vlgrp && (cqe->vlan_my_qpn &
609 cpu_to_be32(MLX4_CQE_VLAN_PRESENT_MASK)))
610 vlan_gro_frags(&cq->napi, priv->vlgrp, be16_to_cpu(cqe->sl_vid));
612 napi_gro_frags(&cq->napi);
617 /* LRO not possible, complete processing here */
618 ip_summed = CHECKSUM_UNNECESSARY;
620 ip_summed = CHECKSUM_NONE;
621 priv->port_stats.rx_chksum_none++;
624 ip_summed = CHECKSUM_NONE;
625 priv->port_stats.rx_chksum_none++;
628 skb = mlx4_en_rx_skb(priv, rx_desc, skb_frags,
629 ring->page_alloc, length);
631 priv->stats.rx_dropped++;
635 if (unlikely(priv->validate_loopback)) {
636 validate_loopback(priv, skb);
640 skb->ip_summed = ip_summed;
641 skb->protocol = eth_type_trans(skb, dev);
642 skb_record_rx_queue(skb, cq->ring);
644 /* Push it up the stack */
645 if (priv->vlgrp && (be32_to_cpu(cqe->vlan_my_qpn) &
646 MLX4_CQE_VLAN_PRESENT_MASK)) {
647 vlan_hwaccel_receive_skb(skb, priv->vlgrp,
648 be16_to_cpu(cqe->sl_vid));
650 netif_receive_skb(skb);
653 ++cq->mcq.cons_index;
654 index = (cq->mcq.cons_index) & ring->size_mask;
655 cqe = &cq->buf[index];
656 if (++polled == budget) {
657 /* We are here because we reached the NAPI budget -
658 * flush only pending LRO sessions */
664 AVG_PERF_COUNTER(priv->pstats.rx_coal_avg, polled);
665 mlx4_cq_set_ci(&cq->mcq);
666 wmb(); /* ensure HW sees CQ consumer before we post new buffers */
667 ring->cons = cq->mcq.cons_index;
668 ring->prod += polled; /* Polled descriptors were realocated in place */
669 mlx4_en_update_rx_prod_db(ring);
674 void mlx4_en_rx_irq(struct mlx4_cq *mcq)
676 struct mlx4_en_cq *cq = container_of(mcq, struct mlx4_en_cq, mcq);
677 struct mlx4_en_priv *priv = netdev_priv(cq->dev);
680 napi_schedule(&cq->napi);
682 mlx4_en_arm_cq(priv, cq);
685 /* Rx CQ polling - called by NAPI */
686 int mlx4_en_poll_rx_cq(struct napi_struct *napi, int budget)
688 struct mlx4_en_cq *cq = container_of(napi, struct mlx4_en_cq, napi);
689 struct net_device *dev = cq->dev;
690 struct mlx4_en_priv *priv = netdev_priv(dev);
693 done = mlx4_en_process_rx_cq(dev, cq, budget);
695 /* If we used up all the quota - we're probably not done yet... */
697 INC_PERF_COUNTER(priv->pstats.napi_quota);
701 mlx4_en_arm_cq(priv, cq);
707 /* Calculate the last offset position that accomodates a full fragment
708 * (assuming fagment size = stride-align) */
709 static int mlx4_en_last_alloc_offset(struct mlx4_en_priv *priv, u16 stride, u16 align)
711 u16 res = MLX4_EN_ALLOC_SIZE % stride;
712 u16 offset = MLX4_EN_ALLOC_SIZE - stride - res + align;
714 en_dbg(DRV, priv, "Calculated last offset for stride:%d align:%d "
715 "res:%d offset:%d\n", stride, align, res, offset);
720 static int frag_sizes[] = {
727 void mlx4_en_calc_rx_buf(struct net_device *dev)
729 struct mlx4_en_priv *priv = netdev_priv(dev);
730 int eff_mtu = dev->mtu + ETH_HLEN + VLAN_HLEN + ETH_LLC_SNAP_SIZE;
734 while (buf_size < eff_mtu) {
735 priv->frag_info[i].frag_size =
736 (eff_mtu > buf_size + frag_sizes[i]) ?
737 frag_sizes[i] : eff_mtu - buf_size;
738 priv->frag_info[i].frag_prefix_size = buf_size;
740 priv->frag_info[i].frag_align = NET_IP_ALIGN;
741 priv->frag_info[i].frag_stride =
742 ALIGN(frag_sizes[i] + NET_IP_ALIGN, SMP_CACHE_BYTES);
744 priv->frag_info[i].frag_align = 0;
745 priv->frag_info[i].frag_stride =
746 ALIGN(frag_sizes[i], SMP_CACHE_BYTES);
748 priv->frag_info[i].last_offset = mlx4_en_last_alloc_offset(
749 priv, priv->frag_info[i].frag_stride,
750 priv->frag_info[i].frag_align);
751 buf_size += priv->frag_info[i].frag_size;
756 priv->rx_skb_size = eff_mtu;
757 priv->log_rx_info = ROUNDUP_LOG2(i * sizeof(struct skb_frag_struct));
759 en_dbg(DRV, priv, "Rx buffer scatter-list (effective-mtu:%d "
760 "num_frags:%d):\n", eff_mtu, priv->num_frags);
761 for (i = 0; i < priv->num_frags; i++) {
762 en_dbg(DRV, priv, " frag:%d - size:%d prefix:%d align:%d "
763 "stride:%d last_offset:%d\n", i,
764 priv->frag_info[i].frag_size,
765 priv->frag_info[i].frag_prefix_size,
766 priv->frag_info[i].frag_align,
767 priv->frag_info[i].frag_stride,
768 priv->frag_info[i].last_offset);
772 /* RSS related functions */
774 static int mlx4_en_config_rss_qp(struct mlx4_en_priv *priv, int qpn,
775 struct mlx4_en_rx_ring *ring,
776 enum mlx4_qp_state *state,
779 struct mlx4_en_dev *mdev = priv->mdev;
780 struct mlx4_qp_context *context;
783 context = kmalloc(sizeof *context , GFP_KERNEL);
785 en_err(priv, "Failed to allocate qp context\n");
789 err = mlx4_qp_alloc(mdev->dev, qpn, qp);
791 en_err(priv, "Failed to allocate qp #%x\n", qpn);
794 qp->event = mlx4_en_sqp_event;
796 memset(context, 0, sizeof *context);
797 mlx4_en_fill_qp_context(priv, ring->actual_size, ring->stride, 0, 0,
798 qpn, ring->cqn, context);
799 context->db_rec_addr = cpu_to_be64(ring->wqres.db.dma);
801 err = mlx4_qp_to_ready(mdev->dev, &ring->wqres.mtt, context, qp, state);
803 mlx4_qp_remove(mdev->dev, qp);
804 mlx4_qp_free(mdev->dev, qp);
806 mlx4_en_update_rx_prod_db(ring);
812 /* Allocate rx qp's and configure them according to rss map */
813 int mlx4_en_config_rss_steer(struct mlx4_en_priv *priv)
815 struct mlx4_en_dev *mdev = priv->mdev;
816 struct mlx4_en_rss_map *rss_map = &priv->rss_map;
817 struct mlx4_qp_context context;
818 struct mlx4_en_rss_context *rss_context;
825 en_dbg(DRV, priv, "Configuring rss steering\n");
826 err = mlx4_qp_reserve_range(mdev->dev, priv->rx_ring_num,
830 en_err(priv, "Failed reserving %d qps\n", priv->rx_ring_num);
834 for (i = 0; i < priv->rx_ring_num; i++) {
835 qpn = rss_map->base_qpn + i;
836 err = mlx4_en_config_rss_qp(priv, qpn, &priv->rx_ring[i],
845 /* Configure RSS indirection qp */
846 err = mlx4_qp_reserve_range(mdev->dev, 1, 1, &priv->base_qpn);
848 en_err(priv, "Failed to reserve range for RSS "
852 err = mlx4_qp_alloc(mdev->dev, priv->base_qpn, &rss_map->indir_qp);
854 en_err(priv, "Failed to allocate RSS indirection QP\n");
857 rss_map->indir_qp.event = mlx4_en_sqp_event;
858 mlx4_en_fill_qp_context(priv, 0, 0, 0, 1, priv->base_qpn,
859 priv->rx_ring[0].cqn, &context);
861 ptr = ((void *) &context) + 0x3c;
862 rss_context = (struct mlx4_en_rss_context *) ptr;
863 rss_context->base_qpn = cpu_to_be32(ilog2(priv->rx_ring_num) << 24 |
864 (rss_map->base_qpn));
865 rss_context->default_qpn = cpu_to_be32(rss_map->base_qpn);
866 rss_context->flags = rss_mask;
868 if (priv->mdev->profile.udp_rss)
869 rss_context->base_qpn_udp = rss_context->default_qpn;
870 err = mlx4_qp_to_ready(mdev->dev, &priv->res.mtt, &context,
871 &rss_map->indir_qp, &rss_map->indir_state);
878 mlx4_qp_modify(mdev->dev, NULL, rss_map->indir_state,
879 MLX4_QP_STATE_RST, NULL, 0, 0, &rss_map->indir_qp);
880 mlx4_qp_remove(mdev->dev, &rss_map->indir_qp);
881 mlx4_qp_free(mdev->dev, &rss_map->indir_qp);
883 mlx4_qp_release_range(mdev->dev, priv->base_qpn, 1);
885 for (i = 0; i < good_qps; i++) {
886 mlx4_qp_modify(mdev->dev, NULL, rss_map->state[i],
887 MLX4_QP_STATE_RST, NULL, 0, 0, &rss_map->qps[i]);
888 mlx4_qp_remove(mdev->dev, &rss_map->qps[i]);
889 mlx4_qp_free(mdev->dev, &rss_map->qps[i]);
891 mlx4_qp_release_range(mdev->dev, rss_map->base_qpn, priv->rx_ring_num);
895 void mlx4_en_release_rss_steer(struct mlx4_en_priv *priv)
897 struct mlx4_en_dev *mdev = priv->mdev;
898 struct mlx4_en_rss_map *rss_map = &priv->rss_map;
901 mlx4_qp_modify(mdev->dev, NULL, rss_map->indir_state,
902 MLX4_QP_STATE_RST, NULL, 0, 0, &rss_map->indir_qp);
903 mlx4_qp_remove(mdev->dev, &rss_map->indir_qp);
904 mlx4_qp_free(mdev->dev, &rss_map->indir_qp);
905 mlx4_qp_release_range(mdev->dev, priv->base_qpn, 1);
907 for (i = 0; i < priv->rx_ring_num; i++) {
908 mlx4_qp_modify(mdev->dev, NULL, rss_map->state[i],
909 MLX4_QP_STATE_RST, NULL, 0, 0, &rss_map->qps[i]);
910 mlx4_qp_remove(mdev->dev, &rss_map->qps[i]);
911 mlx4_qp_free(mdev->dev, &rss_map->qps[i]);
913 mlx4_qp_release_range(mdev->dev, rss_map->base_qpn, priv->rx_ring_num);