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 <net/busy_poll.h>
35 #include <linux/mlx4/cq.h>
36 #include <linux/slab.h>
37 #include <linux/mlx4/qp.h>
38 #include <linux/skbuff.h>
39 #include <linux/rculist.h>
40 #include <linux/if_ether.h>
41 #include <linux/if_vlan.h>
42 #include <linux/vmalloc.h>
46 static int mlx4_alloc_pages(struct mlx4_en_priv
*priv
,
47 struct mlx4_en_rx_alloc
*page_alloc
,
48 const struct mlx4_en_frag_info
*frag_info
,
55 for (order
= MLX4_EN_ALLOC_PREFER_ORDER
; ;) {
59 gfp
|= __GFP_COMP
| __GFP_NOWARN
;
60 page
= alloc_pages(gfp
, order
);
64 ((PAGE_SIZE
<< order
) < frag_info
->frag_size
))
67 dma
= dma_map_page(priv
->ddev
, page
, 0, PAGE_SIZE
<< order
,
69 if (dma_mapping_error(priv
->ddev
, dma
)) {
73 page_alloc
->page_size
= PAGE_SIZE
<< order
;
74 page_alloc
->page
= page
;
75 page_alloc
->dma
= dma
;
76 page_alloc
->page_offset
= frag_info
->frag_align
;
77 /* Not doing get_page() for each frag is a big win
78 * on asymetric workloads.
80 atomic_set(&page
->_count
,
81 page_alloc
->page_size
/ frag_info
->frag_stride
);
85 static int mlx4_en_alloc_frags(struct mlx4_en_priv
*priv
,
86 struct mlx4_en_rx_desc
*rx_desc
,
87 struct mlx4_en_rx_alloc
*frags
,
88 struct mlx4_en_rx_alloc
*ring_alloc
,
91 struct mlx4_en_rx_alloc page_alloc
[MLX4_EN_MAX_RX_FRAGS
];
92 const struct mlx4_en_frag_info
*frag_info
;
97 for (i
= 0; i
< priv
->num_frags
; i
++) {
98 frag_info
= &priv
->frag_info
[i
];
99 page_alloc
[i
] = ring_alloc
[i
];
100 page_alloc
[i
].page_offset
+= frag_info
->frag_stride
;
102 if (page_alloc
[i
].page_offset
+ frag_info
->frag_stride
<=
103 ring_alloc
[i
].page_size
)
106 if (mlx4_alloc_pages(priv
, &page_alloc
[i
], frag_info
, gfp
))
110 for (i
= 0; i
< priv
->num_frags
; i
++) {
111 frags
[i
] = ring_alloc
[i
];
112 dma
= ring_alloc
[i
].dma
+ ring_alloc
[i
].page_offset
;
113 ring_alloc
[i
] = page_alloc
[i
];
114 rx_desc
->data
[i
].addr
= cpu_to_be64(dma
);
121 frag_info
= &priv
->frag_info
[i
];
122 if (page_alloc
[i
].page
!= ring_alloc
[i
].page
) {
123 dma_unmap_page(priv
->ddev
, page_alloc
[i
].dma
,
124 page_alloc
[i
].page_size
, PCI_DMA_FROMDEVICE
);
125 page
= page_alloc
[i
].page
;
126 atomic_set(&page
->_count
, 1);
133 static void mlx4_en_free_frag(struct mlx4_en_priv
*priv
,
134 struct mlx4_en_rx_alloc
*frags
,
137 const struct mlx4_en_frag_info
*frag_info
= &priv
->frag_info
[i
];
138 u32 next_frag_end
= frags
[i
].page_offset
+ 2 * frag_info
->frag_stride
;
141 if (next_frag_end
> frags
[i
].page_size
)
142 dma_unmap_page(priv
->ddev
, frags
[i
].dma
, frags
[i
].page_size
,
146 put_page(frags
[i
].page
);
149 static int mlx4_en_init_allocator(struct mlx4_en_priv
*priv
,
150 struct mlx4_en_rx_ring
*ring
)
153 struct mlx4_en_rx_alloc
*page_alloc
;
155 for (i
= 0; i
< priv
->num_frags
; i
++) {
156 const struct mlx4_en_frag_info
*frag_info
= &priv
->frag_info
[i
];
158 if (mlx4_alloc_pages(priv
, &ring
->page_alloc
[i
],
159 frag_info
, GFP_KERNEL
))
168 page_alloc
= &ring
->page_alloc
[i
];
169 dma_unmap_page(priv
->ddev
, page_alloc
->dma
,
170 page_alloc
->page_size
, PCI_DMA_FROMDEVICE
);
171 page
= page_alloc
->page
;
172 atomic_set(&page
->_count
, 1);
174 page_alloc
->page
= NULL
;
179 static void mlx4_en_destroy_allocator(struct mlx4_en_priv
*priv
,
180 struct mlx4_en_rx_ring
*ring
)
182 struct mlx4_en_rx_alloc
*page_alloc
;
185 for (i
= 0; i
< priv
->num_frags
; i
++) {
186 const struct mlx4_en_frag_info
*frag_info
= &priv
->frag_info
[i
];
188 page_alloc
= &ring
->page_alloc
[i
];
189 en_dbg(DRV
, priv
, "Freeing allocator:%d count:%d\n",
190 i
, page_count(page_alloc
->page
));
192 dma_unmap_page(priv
->ddev
, page_alloc
->dma
,
193 page_alloc
->page_size
, PCI_DMA_FROMDEVICE
);
194 while (page_alloc
->page_offset
+ frag_info
->frag_stride
<
195 page_alloc
->page_size
) {
196 put_page(page_alloc
->page
);
197 page_alloc
->page_offset
+= frag_info
->frag_stride
;
199 page_alloc
->page
= NULL
;
203 static void mlx4_en_init_rx_desc(struct mlx4_en_priv
*priv
,
204 struct mlx4_en_rx_ring
*ring
, int index
)
206 struct mlx4_en_rx_desc
*rx_desc
= ring
->buf
+ ring
->stride
* index
;
210 /* Set size and memtype fields */
211 for (i
= 0; i
< priv
->num_frags
; i
++) {
212 rx_desc
->data
[i
].byte_count
=
213 cpu_to_be32(priv
->frag_info
[i
].frag_size
);
214 rx_desc
->data
[i
].lkey
= cpu_to_be32(priv
->mdev
->mr
.key
);
217 /* If the number of used fragments does not fill up the ring stride,
218 * remaining (unused) fragments must be padded with null address/size
219 * and a special memory key */
220 possible_frags
= (ring
->stride
- sizeof(struct mlx4_en_rx_desc
)) / DS_SIZE
;
221 for (i
= priv
->num_frags
; i
< possible_frags
; i
++) {
222 rx_desc
->data
[i
].byte_count
= 0;
223 rx_desc
->data
[i
].lkey
= cpu_to_be32(MLX4_EN_MEMTYPE_PAD
);
224 rx_desc
->data
[i
].addr
= 0;
228 static int mlx4_en_prepare_rx_desc(struct mlx4_en_priv
*priv
,
229 struct mlx4_en_rx_ring
*ring
, int index
,
232 struct mlx4_en_rx_desc
*rx_desc
= ring
->buf
+ (index
* ring
->stride
);
233 struct mlx4_en_rx_alloc
*frags
= ring
->rx_info
+
234 (index
<< priv
->log_rx_info
);
236 return mlx4_en_alloc_frags(priv
, rx_desc
, frags
, ring
->page_alloc
, gfp
);
239 static inline void mlx4_en_update_rx_prod_db(struct mlx4_en_rx_ring
*ring
)
241 *ring
->wqres
.db
.db
= cpu_to_be32(ring
->prod
& 0xffff);
244 static void mlx4_en_free_rx_desc(struct mlx4_en_priv
*priv
,
245 struct mlx4_en_rx_ring
*ring
,
248 struct mlx4_en_rx_alloc
*frags
;
251 frags
= ring
->rx_info
+ (index
<< priv
->log_rx_info
);
252 for (nr
= 0; nr
< priv
->num_frags
; nr
++) {
253 en_dbg(DRV
, priv
, "Freeing fragment:%d\n", nr
);
254 mlx4_en_free_frag(priv
, frags
, nr
);
258 static int mlx4_en_fill_rx_buffers(struct mlx4_en_priv
*priv
)
260 struct mlx4_en_rx_ring
*ring
;
265 for (buf_ind
= 0; buf_ind
< priv
->prof
->rx_ring_size
; buf_ind
++) {
266 for (ring_ind
= 0; ring_ind
< priv
->rx_ring_num
; ring_ind
++) {
267 ring
= priv
->rx_ring
[ring_ind
];
269 if (mlx4_en_prepare_rx_desc(priv
, ring
,
272 if (ring
->actual_size
< MLX4_EN_MIN_RX_SIZE
) {
273 en_err(priv
, "Failed to allocate "
274 "enough rx buffers\n");
277 new_size
= rounddown_pow_of_two(ring
->actual_size
);
278 en_warn(priv
, "Only %d buffers allocated "
279 "reducing ring size to %d",
280 ring
->actual_size
, new_size
);
291 for (ring_ind
= 0; ring_ind
< priv
->rx_ring_num
; ring_ind
++) {
292 ring
= priv
->rx_ring
[ring_ind
];
293 while (ring
->actual_size
> new_size
) {
296 mlx4_en_free_rx_desc(priv
, ring
, ring
->actual_size
);
303 static void mlx4_en_free_rx_buf(struct mlx4_en_priv
*priv
,
304 struct mlx4_en_rx_ring
*ring
)
308 en_dbg(DRV
, priv
, "Freeing Rx buf - cons:%d prod:%d\n",
309 ring
->cons
, ring
->prod
);
311 /* Unmap and free Rx buffers */
312 BUG_ON((u32
) (ring
->prod
- ring
->cons
) > ring
->actual_size
);
313 while (ring
->cons
!= ring
->prod
) {
314 index
= ring
->cons
& ring
->size_mask
;
315 en_dbg(DRV
, priv
, "Processing descriptor:%d\n", index
);
316 mlx4_en_free_rx_desc(priv
, ring
, index
);
321 int mlx4_en_create_rx_ring(struct mlx4_en_priv
*priv
,
322 struct mlx4_en_rx_ring
**pring
,
323 u32 size
, u16 stride
, int node
)
325 struct mlx4_en_dev
*mdev
= priv
->mdev
;
326 struct mlx4_en_rx_ring
*ring
;
330 ring
= kzalloc_node(sizeof(*ring
), GFP_KERNEL
, node
);
332 ring
= kzalloc(sizeof(*ring
), GFP_KERNEL
);
334 en_err(priv
, "Failed to allocate RX ring structure\n");
342 ring
->size_mask
= size
- 1;
343 ring
->stride
= stride
;
344 ring
->log_stride
= ffs(ring
->stride
) - 1;
345 ring
->buf_size
= ring
->size
* ring
->stride
+ TXBB_SIZE
;
347 tmp
= size
* roundup_pow_of_two(MLX4_EN_MAX_RX_FRAGS
*
348 sizeof(struct mlx4_en_rx_alloc
));
349 ring
->rx_info
= vmalloc_node(tmp
, node
);
350 if (!ring
->rx_info
) {
351 ring
->rx_info
= vmalloc(tmp
);
352 if (!ring
->rx_info
) {
358 en_dbg(DRV
, priv
, "Allocated rx_info ring at addr:%p size:%d\n",
361 /* Allocate HW buffers on provided NUMA node */
362 set_dev_node(&mdev
->dev
->pdev
->dev
, node
);
363 err
= mlx4_alloc_hwq_res(mdev
->dev
, &ring
->wqres
,
364 ring
->buf_size
, 2 * PAGE_SIZE
);
365 set_dev_node(&mdev
->dev
->pdev
->dev
, mdev
->dev
->numa_node
);
369 err
= mlx4_en_map_buffer(&ring
->wqres
.buf
);
371 en_err(priv
, "Failed to map RX buffer\n");
374 ring
->buf
= ring
->wqres
.buf
.direct
.buf
;
376 ring
->hwtstamp_rx_filter
= priv
->hwtstamp_config
.rx_filter
;
382 mlx4_free_hwq_res(mdev
->dev
, &ring
->wqres
, ring
->buf_size
);
384 vfree(ring
->rx_info
);
385 ring
->rx_info
= NULL
;
393 int mlx4_en_activate_rx_rings(struct mlx4_en_priv
*priv
)
395 struct mlx4_en_rx_ring
*ring
;
399 int stride
= roundup_pow_of_two(sizeof(struct mlx4_en_rx_desc
) +
400 DS_SIZE
* priv
->num_frags
);
402 for (ring_ind
= 0; ring_ind
< priv
->rx_ring_num
; ring_ind
++) {
403 ring
= priv
->rx_ring
[ring_ind
];
407 ring
->actual_size
= 0;
408 ring
->cqn
= priv
->rx_cq
[ring_ind
]->mcq
.cqn
;
410 ring
->stride
= stride
;
411 if (ring
->stride
<= TXBB_SIZE
)
412 ring
->buf
+= TXBB_SIZE
;
414 ring
->log_stride
= ffs(ring
->stride
) - 1;
415 ring
->buf_size
= ring
->size
* ring
->stride
;
417 memset(ring
->buf
, 0, ring
->buf_size
);
418 mlx4_en_update_rx_prod_db(ring
);
420 /* Initialize all descriptors */
421 for (i
= 0; i
< ring
->size
; i
++)
422 mlx4_en_init_rx_desc(priv
, ring
, i
);
424 /* Initialize page allocators */
425 err
= mlx4_en_init_allocator(priv
, ring
);
427 en_err(priv
, "Failed initializing ring allocator\n");
428 if (ring
->stride
<= TXBB_SIZE
)
429 ring
->buf
-= TXBB_SIZE
;
434 err
= mlx4_en_fill_rx_buffers(priv
);
438 for (ring_ind
= 0; ring_ind
< priv
->rx_ring_num
; ring_ind
++) {
439 ring
= priv
->rx_ring
[ring_ind
];
441 ring
->size_mask
= ring
->actual_size
- 1;
442 mlx4_en_update_rx_prod_db(ring
);
448 for (ring_ind
= 0; ring_ind
< priv
->rx_ring_num
; ring_ind
++)
449 mlx4_en_free_rx_buf(priv
, priv
->rx_ring
[ring_ind
]);
451 ring_ind
= priv
->rx_ring_num
- 1;
453 while (ring_ind
>= 0) {
454 if (priv
->rx_ring
[ring_ind
]->stride
<= TXBB_SIZE
)
455 priv
->rx_ring
[ring_ind
]->buf
-= TXBB_SIZE
;
456 mlx4_en_destroy_allocator(priv
, priv
->rx_ring
[ring_ind
]);
462 void mlx4_en_destroy_rx_ring(struct mlx4_en_priv
*priv
,
463 struct mlx4_en_rx_ring
**pring
,
464 u32 size
, u16 stride
)
466 struct mlx4_en_dev
*mdev
= priv
->mdev
;
467 struct mlx4_en_rx_ring
*ring
= *pring
;
469 mlx4_en_unmap_buffer(&ring
->wqres
.buf
);
470 mlx4_free_hwq_res(mdev
->dev
, &ring
->wqres
, size
* stride
+ TXBB_SIZE
);
471 vfree(ring
->rx_info
);
472 ring
->rx_info
= NULL
;
475 #ifdef CONFIG_RFS_ACCEL
476 mlx4_en_cleanup_filters(priv
);
480 void mlx4_en_deactivate_rx_ring(struct mlx4_en_priv
*priv
,
481 struct mlx4_en_rx_ring
*ring
)
483 mlx4_en_free_rx_buf(priv
, ring
);
484 if (ring
->stride
<= TXBB_SIZE
)
485 ring
->buf
-= TXBB_SIZE
;
486 mlx4_en_destroy_allocator(priv
, ring
);
490 static int mlx4_en_complete_rx_desc(struct mlx4_en_priv
*priv
,
491 struct mlx4_en_rx_desc
*rx_desc
,
492 struct mlx4_en_rx_alloc
*frags
,
496 struct skb_frag_struct
*skb_frags_rx
= skb_shinfo(skb
)->frags
;
497 struct mlx4_en_frag_info
*frag_info
;
501 /* Collect used fragments while replacing them in the HW descriptors */
502 for (nr
= 0; nr
< priv
->num_frags
; nr
++) {
503 frag_info
= &priv
->frag_info
[nr
];
504 if (length
<= frag_info
->frag_prefix_size
)
509 dma
= be64_to_cpu(rx_desc
->data
[nr
].addr
);
510 dma_sync_single_for_cpu(priv
->ddev
, dma
, frag_info
->frag_size
,
513 /* Save page reference in skb */
514 __skb_frag_set_page(&skb_frags_rx
[nr
], frags
[nr
].page
);
515 skb_frag_size_set(&skb_frags_rx
[nr
], frag_info
->frag_size
);
516 skb_frags_rx
[nr
].page_offset
= frags
[nr
].page_offset
;
517 skb
->truesize
+= frag_info
->frag_stride
;
518 frags
[nr
].page
= NULL
;
520 /* Adjust size of last fragment to match actual length */
522 skb_frag_size_set(&skb_frags_rx
[nr
- 1],
523 length
- priv
->frag_info
[nr
- 1].frag_prefix_size
);
529 __skb_frag_unref(&skb_frags_rx
[nr
]);
535 static struct sk_buff
*mlx4_en_rx_skb(struct mlx4_en_priv
*priv
,
536 struct mlx4_en_rx_desc
*rx_desc
,
537 struct mlx4_en_rx_alloc
*frags
,
545 skb
= netdev_alloc_skb(priv
->dev
, SMALL_PACKET_SIZE
+ NET_IP_ALIGN
);
547 en_dbg(RX_ERR
, priv
, "Failed allocating skb\n");
550 skb_reserve(skb
, NET_IP_ALIGN
);
553 /* Get pointer to first fragment so we could copy the headers into the
554 * (linear part of the) skb */
555 va
= page_address(frags
[0].page
) + frags
[0].page_offset
;
557 if (length
<= SMALL_PACKET_SIZE
) {
558 /* We are copying all relevant data to the skb - temporarily
559 * sync buffers for the copy */
560 dma
= be64_to_cpu(rx_desc
->data
[0].addr
);
561 dma_sync_single_for_cpu(priv
->ddev
, dma
, length
,
563 skb_copy_to_linear_data(skb
, va
, length
);
566 /* Move relevant fragments to skb */
567 used_frags
= mlx4_en_complete_rx_desc(priv
, rx_desc
, frags
,
569 if (unlikely(!used_frags
)) {
573 skb_shinfo(skb
)->nr_frags
= used_frags
;
575 /* Copy headers into the skb linear buffer */
576 memcpy(skb
->data
, va
, HEADER_COPY_SIZE
);
577 skb
->tail
+= HEADER_COPY_SIZE
;
579 /* Skip headers in first fragment */
580 skb_shinfo(skb
)->frags
[0].page_offset
+= HEADER_COPY_SIZE
;
582 /* Adjust size of first fragment */
583 skb_frag_size_sub(&skb_shinfo(skb
)->frags
[0], HEADER_COPY_SIZE
);
584 skb
->data_len
= length
- HEADER_COPY_SIZE
;
589 static void validate_loopback(struct mlx4_en_priv
*priv
, struct sk_buff
*skb
)
592 int offset
= ETH_HLEN
;
594 for (i
= 0; i
< MLX4_LOOPBACK_TEST_PAYLOAD
; i
++, offset
++) {
595 if (*(skb
->data
+ offset
) != (unsigned char) (i
& 0xff))
599 priv
->loopback_ok
= 1;
602 dev_kfree_skb_any(skb
);
605 static void mlx4_en_refill_rx_buffers(struct mlx4_en_priv
*priv
,
606 struct mlx4_en_rx_ring
*ring
)
608 int index
= ring
->prod
& ring
->size_mask
;
610 while ((u32
) (ring
->prod
- ring
->cons
) < ring
->actual_size
) {
611 if (mlx4_en_prepare_rx_desc(priv
, ring
, index
, GFP_ATOMIC
))
614 index
= ring
->prod
& ring
->size_mask
;
618 int mlx4_en_process_rx_cq(struct net_device
*dev
, struct mlx4_en_cq
*cq
, int budget
)
620 struct mlx4_en_priv
*priv
= netdev_priv(dev
);
621 struct mlx4_en_dev
*mdev
= priv
->mdev
;
622 struct mlx4_cqe
*cqe
;
623 struct mlx4_en_rx_ring
*ring
= priv
->rx_ring
[cq
->ring
];
624 struct mlx4_en_rx_alloc
*frags
;
625 struct mlx4_en_rx_desc
*rx_desc
;
632 int factor
= priv
->cqe_factor
;
638 /* We assume a 1:1 mapping between CQEs and Rx descriptors, so Rx
639 * descriptor offset can be deduced from the CQE index instead of
640 * reading 'cqe->index' */
641 index
= cq
->mcq
.cons_index
& ring
->size_mask
;
642 cqe
= &cq
->buf
[(index
<< factor
) + factor
];
644 /* Process all completed CQEs */
645 while (XNOR(cqe
->owner_sr_opcode
& MLX4_CQE_OWNER_MASK
,
646 cq
->mcq
.cons_index
& cq
->size
)) {
648 frags
= ring
->rx_info
+ (index
<< priv
->log_rx_info
);
649 rx_desc
= ring
->buf
+ (index
<< ring
->log_stride
);
652 * make sure we read the CQE after we read the ownership bit
656 /* Drop packet on bad receive or bad checksum */
657 if (unlikely((cqe
->owner_sr_opcode
& MLX4_CQE_OPCODE_MASK
) ==
658 MLX4_CQE_OPCODE_ERROR
)) {
659 en_err(priv
, "CQE completed in error - vendor "
660 "syndrom:%d syndrom:%d\n",
661 ((struct mlx4_err_cqe
*) cqe
)->vendor_err_syndrome
,
662 ((struct mlx4_err_cqe
*) cqe
)->syndrome
);
665 if (unlikely(cqe
->badfcs_enc
& MLX4_CQE_BAD_FCS
)) {
666 en_dbg(RX_ERR
, priv
, "Accepted frame with bad FCS\n");
670 /* Check if we need to drop the packet if SRIOV is not enabled
671 * and not performing the selftest or flb disabled
673 if (priv
->flags
& MLX4_EN_FLAG_RX_FILTER_NEEDED
) {
676 /* Get pointer to first fragment since we haven't
677 * skb yet and cast it to ethhdr struct
679 dma
= be64_to_cpu(rx_desc
->data
[0].addr
);
680 dma_sync_single_for_cpu(priv
->ddev
, dma
, sizeof(*ethh
),
682 ethh
= (struct ethhdr
*)(page_address(frags
[0].page
) +
683 frags
[0].page_offset
);
685 if (is_multicast_ether_addr(ethh
->h_dest
)) {
686 struct mlx4_mac_entry
*entry
;
687 struct hlist_head
*bucket
;
688 unsigned int mac_hash
;
690 /* Drop the packet, since HW loopback-ed it */
691 mac_hash
= ethh
->h_source
[MLX4_EN_MAC_HASH_IDX
];
692 bucket
= &priv
->mac_hash
[mac_hash
];
694 hlist_for_each_entry_rcu(entry
, bucket
, hlist
) {
695 if (ether_addr_equal_64bits(entry
->mac
,
706 * Packet is OK - process it.
708 length
= be32_to_cpu(cqe
->byte_cnt
);
709 length
-= ring
->fcs_del
;
710 ring
->bytes
+= length
;
713 if (likely(dev
->features
& NETIF_F_RXCSUM
)) {
714 if ((cqe
->status
& cpu_to_be16(MLX4_CQE_STATUS_IPOK
)) &&
715 (cqe
->checksum
== cpu_to_be16(0xffff))) {
717 /* This packet is eligible for GRO if it is:
718 * - DIX Ethernet (type interpretation)
720 * - without IP options
721 * - not an IP fragment
722 * - no LLS polling in progress
724 if (!mlx4_en_cq_ll_polling(cq
) &&
725 (dev
->features
& NETIF_F_GRO
)) {
726 struct sk_buff
*gro_skb
= napi_get_frags(&cq
->napi
);
730 nr
= mlx4_en_complete_rx_desc(priv
,
731 rx_desc
, frags
, gro_skb
,
736 skb_shinfo(gro_skb
)->nr_frags
= nr
;
737 gro_skb
->len
= length
;
738 gro_skb
->data_len
= length
;
739 gro_skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
741 if ((cqe
->vlan_my_qpn
&
742 cpu_to_be32(MLX4_CQE_VLAN_PRESENT_MASK
)) &&
743 (dev
->features
& NETIF_F_HW_VLAN_CTAG_RX
)) {
744 u16 vid
= be16_to_cpu(cqe
->sl_vid
);
746 __vlan_hwaccel_put_tag(gro_skb
, htons(ETH_P_8021Q
), vid
);
749 if (dev
->features
& NETIF_F_RXHASH
)
750 skb_set_hash(gro_skb
,
751 be32_to_cpu(cqe
->immed_rss_invalid
),
754 skb_record_rx_queue(gro_skb
, cq
->ring
);
756 if (ring
->hwtstamp_rx_filter
== HWTSTAMP_FILTER_ALL
) {
757 timestamp
= mlx4_en_get_cqe_ts(cqe
);
758 mlx4_en_fill_hwtstamps(mdev
,
759 skb_hwtstamps(gro_skb
),
763 napi_gro_frags(&cq
->napi
);
767 /* GRO not possible, complete processing here */
768 ip_summed
= CHECKSUM_UNNECESSARY
;
770 ip_summed
= CHECKSUM_NONE
;
774 ip_summed
= CHECKSUM_NONE
;
778 skb
= mlx4_en_rx_skb(priv
, rx_desc
, frags
, length
);
780 priv
->stats
.rx_dropped
++;
784 if (unlikely(priv
->validate_loopback
)) {
785 validate_loopback(priv
, skb
);
789 skb
->ip_summed
= ip_summed
;
790 skb
->protocol
= eth_type_trans(skb
, dev
);
791 skb_record_rx_queue(skb
, cq
->ring
);
793 if (dev
->features
& NETIF_F_RXHASH
)
795 be32_to_cpu(cqe
->immed_rss_invalid
),
798 if ((be32_to_cpu(cqe
->vlan_my_qpn
) &
799 MLX4_CQE_VLAN_PRESENT_MASK
) &&
800 (dev
->features
& NETIF_F_HW_VLAN_CTAG_RX
))
801 __vlan_hwaccel_put_tag(skb
, htons(ETH_P_8021Q
), be16_to_cpu(cqe
->sl_vid
));
803 if (ring
->hwtstamp_rx_filter
== HWTSTAMP_FILTER_ALL
) {
804 timestamp
= mlx4_en_get_cqe_ts(cqe
);
805 mlx4_en_fill_hwtstamps(mdev
, skb_hwtstamps(skb
),
809 skb_mark_napi_id(skb
, &cq
->napi
);
811 /* Push it up the stack */
812 netif_receive_skb(skb
);
815 for (nr
= 0; nr
< priv
->num_frags
; nr
++)
816 mlx4_en_free_frag(priv
, frags
, nr
);
818 ++cq
->mcq
.cons_index
;
819 index
= (cq
->mcq
.cons_index
) & ring
->size_mask
;
820 cqe
= &cq
->buf
[(index
<< factor
) + factor
];
821 if (++polled
== budget
)
826 AVG_PERF_COUNTER(priv
->pstats
.rx_coal_avg
, polled
);
827 mlx4_cq_set_ci(&cq
->mcq
);
828 wmb(); /* ensure HW sees CQ consumer before we post new buffers */
829 ring
->cons
= cq
->mcq
.cons_index
;
830 mlx4_en_refill_rx_buffers(priv
, ring
);
831 mlx4_en_update_rx_prod_db(ring
);
836 void mlx4_en_rx_irq(struct mlx4_cq
*mcq
)
838 struct mlx4_en_cq
*cq
= container_of(mcq
, struct mlx4_en_cq
, mcq
);
839 struct mlx4_en_priv
*priv
= netdev_priv(cq
->dev
);
842 napi_schedule(&cq
->napi
);
844 mlx4_en_arm_cq(priv
, cq
);
847 /* Rx CQ polling - called by NAPI */
848 int mlx4_en_poll_rx_cq(struct napi_struct
*napi
, int budget
)
850 struct mlx4_en_cq
*cq
= container_of(napi
, struct mlx4_en_cq
, napi
);
851 struct net_device
*dev
= cq
->dev
;
852 struct mlx4_en_priv
*priv
= netdev_priv(dev
);
855 if (!mlx4_en_cq_lock_napi(cq
))
858 done
= mlx4_en_process_rx_cq(dev
, cq
, budget
);
860 mlx4_en_cq_unlock_napi(cq
);
862 /* If we used up all the quota - we're probably not done yet... */
864 INC_PERF_COUNTER(priv
->pstats
.napi_quota
);
868 mlx4_en_arm_cq(priv
, cq
);
873 static const int frag_sizes
[] = {
880 void mlx4_en_calc_rx_buf(struct net_device
*dev
)
882 struct mlx4_en_priv
*priv
= netdev_priv(dev
);
883 int eff_mtu
= dev
->mtu
+ ETH_HLEN
+ VLAN_HLEN
+ ETH_LLC_SNAP_SIZE
;
887 while (buf_size
< eff_mtu
) {
888 priv
->frag_info
[i
].frag_size
=
889 (eff_mtu
> buf_size
+ frag_sizes
[i
]) ?
890 frag_sizes
[i
] : eff_mtu
- buf_size
;
891 priv
->frag_info
[i
].frag_prefix_size
= buf_size
;
893 priv
->frag_info
[i
].frag_align
= NET_IP_ALIGN
;
894 priv
->frag_info
[i
].frag_stride
=
895 ALIGN(frag_sizes
[i
] + NET_IP_ALIGN
, SMP_CACHE_BYTES
);
897 priv
->frag_info
[i
].frag_align
= 0;
898 priv
->frag_info
[i
].frag_stride
=
899 ALIGN(frag_sizes
[i
], SMP_CACHE_BYTES
);
901 buf_size
+= priv
->frag_info
[i
].frag_size
;
906 priv
->rx_skb_size
= eff_mtu
;
907 priv
->log_rx_info
= ROUNDUP_LOG2(i
* sizeof(struct mlx4_en_rx_alloc
));
909 en_dbg(DRV
, priv
, "Rx buffer scatter-list (effective-mtu:%d "
910 "num_frags:%d):\n", eff_mtu
, priv
->num_frags
);
911 for (i
= 0; i
< priv
->num_frags
; i
++) {
913 " frag:%d - size:%d prefix:%d align:%d stride:%d\n",
915 priv
->frag_info
[i
].frag_size
,
916 priv
->frag_info
[i
].frag_prefix_size
,
917 priv
->frag_info
[i
].frag_align
,
918 priv
->frag_info
[i
].frag_stride
);
922 /* RSS related functions */
924 static int mlx4_en_config_rss_qp(struct mlx4_en_priv
*priv
, int qpn
,
925 struct mlx4_en_rx_ring
*ring
,
926 enum mlx4_qp_state
*state
,
929 struct mlx4_en_dev
*mdev
= priv
->mdev
;
930 struct mlx4_qp_context
*context
;
933 context
= kmalloc(sizeof(*context
), GFP_KERNEL
);
937 err
= mlx4_qp_alloc(mdev
->dev
, qpn
, qp
);
939 en_err(priv
, "Failed to allocate qp #%x\n", qpn
);
942 qp
->event
= mlx4_en_sqp_event
;
944 memset(context
, 0, sizeof *context
);
945 mlx4_en_fill_qp_context(priv
, ring
->actual_size
, ring
->stride
, 0, 0,
946 qpn
, ring
->cqn
, -1, context
);
947 context
->db_rec_addr
= cpu_to_be64(ring
->wqres
.db
.dma
);
949 /* Cancel FCS removal if FW allows */
950 if (mdev
->dev
->caps
.flags
& MLX4_DEV_CAP_FLAG_FCS_KEEP
) {
951 context
->param3
|= cpu_to_be32(1 << 29);
952 ring
->fcs_del
= ETH_FCS_LEN
;
956 err
= mlx4_qp_to_ready(mdev
->dev
, &ring
->wqres
.mtt
, context
, qp
, state
);
958 mlx4_qp_remove(mdev
->dev
, qp
);
959 mlx4_qp_free(mdev
->dev
, qp
);
961 mlx4_en_update_rx_prod_db(ring
);
967 int mlx4_en_create_drop_qp(struct mlx4_en_priv
*priv
)
972 err
= mlx4_qp_reserve_range(priv
->mdev
->dev
, 1, 1, &qpn
);
974 en_err(priv
, "Failed reserving drop qpn\n");
977 err
= mlx4_qp_alloc(priv
->mdev
->dev
, qpn
, &priv
->drop_qp
);
979 en_err(priv
, "Failed allocating drop qp\n");
980 mlx4_qp_release_range(priv
->mdev
->dev
, qpn
, 1);
987 void mlx4_en_destroy_drop_qp(struct mlx4_en_priv
*priv
)
991 qpn
= priv
->drop_qp
.qpn
;
992 mlx4_qp_remove(priv
->mdev
->dev
, &priv
->drop_qp
);
993 mlx4_qp_free(priv
->mdev
->dev
, &priv
->drop_qp
);
994 mlx4_qp_release_range(priv
->mdev
->dev
, qpn
, 1);
997 /* Allocate rx qp's and configure them according to rss map */
998 int mlx4_en_config_rss_steer(struct mlx4_en_priv
*priv
)
1000 struct mlx4_en_dev
*mdev
= priv
->mdev
;
1001 struct mlx4_en_rss_map
*rss_map
= &priv
->rss_map
;
1002 struct mlx4_qp_context context
;
1003 struct mlx4_rss_context
*rss_context
;
1006 u8 rss_mask
= (MLX4_RSS_IPV4
| MLX4_RSS_TCP_IPV4
| MLX4_RSS_IPV6
|
1011 static const u32 rsskey
[10] = { 0xD181C62C, 0xF7F4DB5B, 0x1983A2FC,
1012 0x943E1ADB, 0xD9389E6B, 0xD1039C2C, 0xA74499AD,
1013 0x593D56D9, 0xF3253C06, 0x2ADC1FFC};
1015 en_dbg(DRV
, priv
, "Configuring rss steering\n");
1016 err
= mlx4_qp_reserve_range(mdev
->dev
, priv
->rx_ring_num
,
1018 &rss_map
->base_qpn
);
1020 en_err(priv
, "Failed reserving %d qps\n", priv
->rx_ring_num
);
1024 for (i
= 0; i
< priv
->rx_ring_num
; i
++) {
1025 qpn
= rss_map
->base_qpn
+ i
;
1026 err
= mlx4_en_config_rss_qp(priv
, qpn
, priv
->rx_ring
[i
],
1035 /* Configure RSS indirection qp */
1036 err
= mlx4_qp_alloc(mdev
->dev
, priv
->base_qpn
, &rss_map
->indir_qp
);
1038 en_err(priv
, "Failed to allocate RSS indirection QP\n");
1041 rss_map
->indir_qp
.event
= mlx4_en_sqp_event
;
1042 mlx4_en_fill_qp_context(priv
, 0, 0, 0, 1, priv
->base_qpn
,
1043 priv
->rx_ring
[0]->cqn
, -1, &context
);
1045 if (!priv
->prof
->rss_rings
|| priv
->prof
->rss_rings
> priv
->rx_ring_num
)
1046 rss_rings
= priv
->rx_ring_num
;
1048 rss_rings
= priv
->prof
->rss_rings
;
1050 ptr
= ((void *) &context
) + offsetof(struct mlx4_qp_context
, pri_path
)
1051 + MLX4_RSS_OFFSET_IN_QPC_PRI_PATH
;
1053 rss_context
->base_qpn
= cpu_to_be32(ilog2(rss_rings
) << 24 |
1054 (rss_map
->base_qpn
));
1055 rss_context
->default_qpn
= cpu_to_be32(rss_map
->base_qpn
);
1056 if (priv
->mdev
->profile
.udp_rss
) {
1057 rss_mask
|= MLX4_RSS_UDP_IPV4
| MLX4_RSS_UDP_IPV6
;
1058 rss_context
->base_qpn_udp
= rss_context
->default_qpn
;
1060 rss_context
->flags
= rss_mask
;
1061 rss_context
->hash_fn
= MLX4_RSS_HASH_TOP
;
1062 for (i
= 0; i
< 10; i
++)
1063 rss_context
->rss_key
[i
] = cpu_to_be32(rsskey
[i
]);
1065 err
= mlx4_qp_to_ready(mdev
->dev
, &priv
->res
.mtt
, &context
,
1066 &rss_map
->indir_qp
, &rss_map
->indir_state
);
1073 mlx4_qp_modify(mdev
->dev
, NULL
, rss_map
->indir_state
,
1074 MLX4_QP_STATE_RST
, NULL
, 0, 0, &rss_map
->indir_qp
);
1075 mlx4_qp_remove(mdev
->dev
, &rss_map
->indir_qp
);
1076 mlx4_qp_free(mdev
->dev
, &rss_map
->indir_qp
);
1078 for (i
= 0; i
< good_qps
; i
++) {
1079 mlx4_qp_modify(mdev
->dev
, NULL
, rss_map
->state
[i
],
1080 MLX4_QP_STATE_RST
, NULL
, 0, 0, &rss_map
->qps
[i
]);
1081 mlx4_qp_remove(mdev
->dev
, &rss_map
->qps
[i
]);
1082 mlx4_qp_free(mdev
->dev
, &rss_map
->qps
[i
]);
1084 mlx4_qp_release_range(mdev
->dev
, rss_map
->base_qpn
, priv
->rx_ring_num
);
1088 void mlx4_en_release_rss_steer(struct mlx4_en_priv
*priv
)
1090 struct mlx4_en_dev
*mdev
= priv
->mdev
;
1091 struct mlx4_en_rss_map
*rss_map
= &priv
->rss_map
;
1094 mlx4_qp_modify(mdev
->dev
, NULL
, rss_map
->indir_state
,
1095 MLX4_QP_STATE_RST
, NULL
, 0, 0, &rss_map
->indir_qp
);
1096 mlx4_qp_remove(mdev
->dev
, &rss_map
->indir_qp
);
1097 mlx4_qp_free(mdev
->dev
, &rss_map
->indir_qp
);
1099 for (i
= 0; i
< priv
->rx_ring_num
; i
++) {
1100 mlx4_qp_modify(mdev
->dev
, NULL
, rss_map
->state
[i
],
1101 MLX4_QP_STATE_RST
, NULL
, 0, 0, &rss_map
->qps
[i
]);
1102 mlx4_qp_remove(mdev
->dev
, &rss_map
->qps
[i
]);
1103 mlx4_qp_free(mdev
->dev
, &rss_map
->qps
[i
]);
1105 mlx4_qp_release_range(mdev
->dev
, rss_map
->base_qpn
, priv
->rx_ring_num
);