1 /* A network driver using virtio.
3 * Copyright 2007 Rusty Russell <rusty@rustcorp.com.au> IBM Corporation
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, see <http://www.gnu.org/licenses/>.
19 #include <linux/netdevice.h>
20 #include <linux/etherdevice.h>
21 #include <linux/ethtool.h>
22 #include <linux/module.h>
23 #include <linux/virtio.h>
24 #include <linux/virtio_net.h>
25 #include <linux/bpf.h>
26 #include <linux/bpf_trace.h>
27 #include <linux/scatterlist.h>
28 #include <linux/if_vlan.h>
29 #include <linux/slab.h>
30 #include <linux/cpu.h>
31 #include <linux/average.h>
33 static int napi_weight
= NAPI_POLL_WEIGHT
;
34 module_param(napi_weight
, int, 0444);
36 static bool csum
= true, gso
= true;
37 module_param(csum
, bool, 0444);
38 module_param(gso
, bool, 0444);
40 /* FIXME: MTU in config. */
41 #define GOOD_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
42 #define GOOD_COPY_LEN 128
44 #define VIRTNET_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
46 /* Amount of XDP headroom to prepend to packets for use by xdp_adjust_head */
47 #define VIRTIO_XDP_HEADROOM 256
49 /* RX packet size EWMA. The average packet size is used to determine the packet
50 * buffer size when refilling RX rings. As the entire RX ring may be refilled
51 * at once, the weight is chosen so that the EWMA will be insensitive to short-
52 * term, transient changes in packet size.
54 DECLARE_EWMA(pkt_len
, 0, 64)
56 /* With mergeable buffers we align buffer address and use the low bits to
57 * encode its true size. Buffer size is up to 1 page so we need to align to
58 * square root of page size to ensure we reserve enough bits to encode the true
61 #define MERGEABLE_BUFFER_MIN_ALIGN_SHIFT ((PAGE_SHIFT + 1) / 2)
63 /* Minimum alignment for mergeable packet buffers. */
64 #define MERGEABLE_BUFFER_ALIGN max(L1_CACHE_BYTES, \
65 1 << MERGEABLE_BUFFER_MIN_ALIGN_SHIFT)
67 #define VIRTNET_DRIVER_VERSION "1.0.0"
69 struct virtnet_stats
{
70 struct u64_stats_sync tx_syncp
;
71 struct u64_stats_sync rx_syncp
;
79 /* Internal representation of a send virtqueue */
81 /* Virtqueue associated with this send _queue */
84 /* TX: fragments + linear part + virtio header */
85 struct scatterlist sg
[MAX_SKB_FRAGS
+ 2];
87 /* Name of the send queue: output.$index */
91 /* Internal representation of a receive virtqueue */
92 struct receive_queue
{
93 /* Virtqueue associated with this receive_queue */
96 struct napi_struct napi
;
98 struct bpf_prog __rcu
*xdp_prog
;
100 /* Chain pages by the private ptr. */
103 /* Average packet length for mergeable receive buffers. */
104 struct ewma_pkt_len mrg_avg_pkt_len
;
106 /* Page frag for packet buffer allocation. */
107 struct page_frag alloc_frag
;
109 /* RX: fragments + linear part + virtio header */
110 struct scatterlist sg
[MAX_SKB_FRAGS
+ 2];
112 /* Name of this receive queue: input.$index */
116 struct virtnet_info
{
117 struct virtio_device
*vdev
;
118 struct virtqueue
*cvq
;
119 struct net_device
*dev
;
120 struct send_queue
*sq
;
121 struct receive_queue
*rq
;
124 /* Max # of queue pairs supported by the device */
127 /* # of queue pairs currently used by the driver */
128 u16 curr_queue_pairs
;
130 /* # of XDP queue pairs currently used by the driver */
133 /* I like... big packets and I cannot lie! */
136 /* Host will merge rx buffers for big packets (shake it! shake it!) */
137 bool mergeable_rx_bufs
;
139 /* Has control virtqueue */
142 /* Host can handle any s/g split between our header and packet data */
145 /* Packet virtio header size */
148 /* Active statistics */
149 struct virtnet_stats __percpu
*stats
;
151 /* Work struct for refilling if we run low on memory. */
152 struct delayed_work refill
;
154 /* Work struct for config space updates */
155 struct work_struct config_work
;
157 /* Does the affinity hint is set for virtqueues? */
158 bool affinity_hint_set
;
160 /* CPU hotplug instances for online & dead */
161 struct hlist_node node
;
162 struct hlist_node node_dead
;
164 /* Control VQ buffers: protected by the rtnl lock */
165 struct virtio_net_ctrl_hdr ctrl_hdr
;
166 virtio_net_ctrl_ack ctrl_status
;
167 struct virtio_net_ctrl_mq ctrl_mq
;
172 /* Ethtool settings */
177 struct padded_vnet_hdr
{
178 struct virtio_net_hdr_mrg_rxbuf hdr
;
180 * hdr is in a separate sg buffer, and data sg buffer shares same page
181 * with this header sg. This padding makes next sg 16 byte aligned
187 /* Converting between virtqueue no. and kernel tx/rx queue no.
188 * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq
190 static int vq2txq(struct virtqueue
*vq
)
192 return (vq
->index
- 1) / 2;
195 static int txq2vq(int txq
)
200 static int vq2rxq(struct virtqueue
*vq
)
202 return vq
->index
/ 2;
205 static int rxq2vq(int rxq
)
210 static inline struct virtio_net_hdr_mrg_rxbuf
*skb_vnet_hdr(struct sk_buff
*skb
)
212 return (struct virtio_net_hdr_mrg_rxbuf
*)skb
->cb
;
216 * private is used to chain pages for big packets, put the whole
217 * most recent used list in the beginning for reuse
219 static void give_pages(struct receive_queue
*rq
, struct page
*page
)
223 /* Find end of list, sew whole thing into vi->rq.pages. */
224 for (end
= page
; end
->private; end
= (struct page
*)end
->private);
225 end
->private = (unsigned long)rq
->pages
;
229 static struct page
*get_a_page(struct receive_queue
*rq
, gfp_t gfp_mask
)
231 struct page
*p
= rq
->pages
;
234 rq
->pages
= (struct page
*)p
->private;
235 /* clear private here, it is used to chain pages */
238 p
= alloc_page(gfp_mask
);
242 static void skb_xmit_done(struct virtqueue
*vq
)
244 struct virtnet_info
*vi
= vq
->vdev
->priv
;
246 /* Suppress further interrupts. */
247 virtqueue_disable_cb(vq
);
249 /* We were probably waiting for more output buffers. */
250 netif_wake_subqueue(vi
->dev
, vq2txq(vq
));
253 static unsigned int mergeable_ctx_to_buf_truesize(unsigned long mrg_ctx
)
255 unsigned int truesize
= mrg_ctx
& (MERGEABLE_BUFFER_ALIGN
- 1);
256 return (truesize
+ 1) * MERGEABLE_BUFFER_ALIGN
;
259 static void *mergeable_ctx_to_buf_address(unsigned long mrg_ctx
)
261 return (void *)(mrg_ctx
& -MERGEABLE_BUFFER_ALIGN
);
265 static unsigned long mergeable_buf_to_ctx(void *buf
, unsigned int truesize
)
267 unsigned int size
= truesize
/ MERGEABLE_BUFFER_ALIGN
;
268 return (unsigned long)buf
| (size
- 1);
271 /* Called from bottom half context */
272 static struct sk_buff
*page_to_skb(struct virtnet_info
*vi
,
273 struct receive_queue
*rq
,
274 struct page
*page
, unsigned int offset
,
275 unsigned int len
, unsigned int truesize
)
278 struct virtio_net_hdr_mrg_rxbuf
*hdr
;
279 unsigned int copy
, hdr_len
, hdr_padded_len
;
282 p
= page_address(page
) + offset
;
284 /* copy small packet so we can reuse these pages for small data */
285 skb
= napi_alloc_skb(&rq
->napi
, GOOD_COPY_LEN
);
289 hdr
= skb_vnet_hdr(skb
);
291 hdr_len
= vi
->hdr_len
;
292 if (vi
->mergeable_rx_bufs
)
293 hdr_padded_len
= sizeof *hdr
;
295 hdr_padded_len
= sizeof(struct padded_vnet_hdr
);
297 memcpy(hdr
, p
, hdr_len
);
300 offset
+= hdr_padded_len
;
304 if (copy
> skb_tailroom(skb
))
305 copy
= skb_tailroom(skb
);
306 memcpy(skb_put(skb
, copy
), p
, copy
);
311 if (vi
->mergeable_rx_bufs
) {
313 skb_add_rx_frag(skb
, 0, page
, offset
, len
, truesize
);
320 * Verify that we can indeed put this data into a skb.
321 * This is here to handle cases when the device erroneously
322 * tries to receive more than is possible. This is usually
323 * the case of a broken device.
325 if (unlikely(len
> MAX_SKB_FRAGS
* PAGE_SIZE
)) {
326 net_dbg_ratelimited("%s: too much data\n", skb
->dev
->name
);
330 BUG_ON(offset
>= PAGE_SIZE
);
332 unsigned int frag_size
= min((unsigned)PAGE_SIZE
- offset
, len
);
333 skb_add_rx_frag(skb
, skb_shinfo(skb
)->nr_frags
, page
, offset
,
334 frag_size
, truesize
);
336 page
= (struct page
*)page
->private;
341 give_pages(rq
, page
);
346 static bool virtnet_xdp_xmit(struct virtnet_info
*vi
,
347 struct receive_queue
*rq
,
348 struct xdp_buff
*xdp
)
350 struct virtio_net_hdr_mrg_rxbuf
*hdr
;
352 struct send_queue
*sq
;
357 qp
= vi
->curr_queue_pairs
- vi
->xdp_queue_pairs
+ smp_processor_id();
360 /* Free up any pending old buffers before queueing new ones. */
361 while ((xdp_sent
= virtqueue_get_buf(sq
->vq
, &len
)) != NULL
) {
362 struct page
*sent_page
= virt_to_head_page(xdp_sent
);
367 xdp
->data
-= vi
->hdr_len
;
368 /* Zero header and leave csum up to XDP layers */
370 memset(hdr
, 0, vi
->hdr_len
);
372 sg_init_one(sq
->sg
, xdp
->data
, xdp
->data_end
- xdp
->data
);
374 err
= virtqueue_add_outbuf(sq
->vq
, sq
->sg
, 1, xdp
->data
, GFP_ATOMIC
);
376 struct page
*page
= virt_to_head_page(xdp
->data
);
382 virtqueue_kick(sq
->vq
);
386 static unsigned int virtnet_get_headroom(struct virtnet_info
*vi
)
388 return vi
->xdp_queue_pairs
? VIRTIO_XDP_HEADROOM
: 0;
391 static struct sk_buff
*receive_small(struct net_device
*dev
,
392 struct virtnet_info
*vi
,
393 struct receive_queue
*rq
,
394 void *buf
, unsigned int len
)
397 struct bpf_prog
*xdp_prog
;
398 unsigned int xdp_headroom
= virtnet_get_headroom(vi
);
399 unsigned int header_offset
= VIRTNET_RX_PAD
+ xdp_headroom
;
400 unsigned int headroom
= vi
->hdr_len
+ header_offset
;
401 unsigned int buflen
= SKB_DATA_ALIGN(GOOD_PACKET_LEN
+ headroom
) +
402 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
403 unsigned int delta
= 0;
407 xdp_prog
= rcu_dereference(rq
->xdp_prog
);
409 struct virtio_net_hdr_mrg_rxbuf
*hdr
= buf
+ header_offset
;
414 if (unlikely(hdr
->hdr
.gso_type
|| hdr
->hdr
.flags
))
417 xdp
.data_hard_start
= buf
+ VIRTNET_RX_PAD
+ vi
->hdr_len
;
418 xdp
.data
= xdp
.data_hard_start
+ xdp_headroom
;
419 xdp
.data_end
= xdp
.data
+ len
;
420 orig_data
= xdp
.data
;
421 act
= bpf_prog_run_xdp(xdp_prog
, &xdp
);
425 /* Recalculate length in case bpf program changed it */
426 delta
= orig_data
- xdp
.data
;
429 if (unlikely(!virtnet_xdp_xmit(vi
, rq
, &xdp
)))
430 trace_xdp_exception(vi
->dev
, xdp_prog
, act
);
434 bpf_warn_invalid_xdp_action(act
);
436 trace_xdp_exception(vi
->dev
, xdp_prog
, act
);
443 skb
= build_skb(buf
, buflen
);
445 put_page(virt_to_head_page(buf
));
448 skb_reserve(skb
, headroom
- delta
);
449 skb_put(skb
, len
+ delta
);
451 buf
+= header_offset
;
452 memcpy(skb_vnet_hdr(skb
), buf
, vi
->hdr_len
);
453 } /* keep zeroed vnet hdr since packet was changed by bpf */
460 dev
->stats
.rx_dropped
++;
461 put_page(virt_to_head_page(buf
));
466 static struct sk_buff
*receive_big(struct net_device
*dev
,
467 struct virtnet_info
*vi
,
468 struct receive_queue
*rq
,
472 struct page
*page
= buf
;
473 struct sk_buff
*skb
= page_to_skb(vi
, rq
, page
, 0, len
, PAGE_SIZE
);
481 dev
->stats
.rx_dropped
++;
482 give_pages(rq
, page
);
486 /* The conditions to enable XDP should preclude the underlying device from
487 * sending packets across multiple buffers (num_buf > 1). However per spec
488 * it does not appear to be illegal to do so but rather just against convention.
489 * So in order to avoid making a system unresponsive the packets are pushed
490 * into a page and the XDP program is run. This will be extremely slow and we
491 * push a warning to the user to fix this as soon as possible. Fixing this may
492 * require resolving the underlying hardware to determine why multiple buffers
493 * are being received or simply loading the XDP program in the ingress stack
494 * after the skb is built because there is no advantage to running it here
497 static struct page
*xdp_linearize_page(struct receive_queue
*rq
,
503 struct page
*page
= alloc_page(GFP_ATOMIC
);
504 unsigned int page_off
= VIRTIO_XDP_HEADROOM
;
509 memcpy(page_address(page
) + page_off
, page_address(p
) + offset
, *len
);
518 ctx
= (unsigned long)virtqueue_get_buf(rq
->vq
, &buflen
);
522 buf
= mergeable_ctx_to_buf_address(ctx
);
523 p
= virt_to_head_page(buf
);
524 off
= buf
- page_address(p
);
526 /* guard against a misconfigured or uncooperative backend that
527 * is sending packet larger than the MTU.
529 if ((page_off
+ buflen
) > PAGE_SIZE
) {
534 memcpy(page_address(page
) + page_off
,
535 page_address(p
) + off
, buflen
);
540 /* Headroom does not contribute to packet length */
541 *len
= page_off
- VIRTIO_XDP_HEADROOM
;
544 __free_pages(page
, 0);
548 static struct sk_buff
*receive_mergeable(struct net_device
*dev
,
549 struct virtnet_info
*vi
,
550 struct receive_queue
*rq
,
554 void *buf
= mergeable_ctx_to_buf_address(ctx
);
555 struct virtio_net_hdr_mrg_rxbuf
*hdr
= buf
;
556 u16 num_buf
= virtio16_to_cpu(vi
->vdev
, hdr
->num_buffers
);
557 struct page
*page
= virt_to_head_page(buf
);
558 int offset
= buf
- page_address(page
);
559 struct sk_buff
*head_skb
, *curr_skb
;
560 struct bpf_prog
*xdp_prog
;
561 unsigned int truesize
;
566 xdp_prog
= rcu_dereference(rq
->xdp_prog
);
568 struct page
*xdp_page
;
573 /* This happens when rx buffer size is underestimated */
574 if (unlikely(num_buf
> 1)) {
575 /* linearize data for XDP */
576 xdp_page
= xdp_linearize_page(rq
, &num_buf
,
580 offset
= VIRTIO_XDP_HEADROOM
;
585 /* Transient failure which in theory could occur if
586 * in-flight packets from before XDP was enabled reach
587 * the receive path after XDP is loaded. In practice I
588 * was not able to create this condition.
590 if (unlikely(hdr
->hdr
.gso_type
))
593 /* Allow consuming headroom but reserve enough space to push
594 * the descriptor on if we get an XDP_TX return code.
596 data
= page_address(xdp_page
) + offset
;
597 xdp
.data_hard_start
= data
- VIRTIO_XDP_HEADROOM
+ vi
->hdr_len
;
598 xdp
.data
= data
+ vi
->hdr_len
;
599 xdp
.data_end
= xdp
.data
+ (len
- vi
->hdr_len
);
600 act
= bpf_prog_run_xdp(xdp_prog
, &xdp
);
604 /* recalculate offset to account for any header
605 * adjustments. Note other cases do not build an
606 * skb and avoid using offset
609 page_address(xdp_page
) - vi
->hdr_len
;
611 /* We can only create skb based on xdp_page. */
612 if (unlikely(xdp_page
!= page
)) {
615 head_skb
= page_to_skb(vi
, rq
, xdp_page
,
616 offset
, len
, PAGE_SIZE
);
617 ewma_pkt_len_add(&rq
->mrg_avg_pkt_len
, len
);
622 if (unlikely(!virtnet_xdp_xmit(vi
, rq
, &xdp
)))
623 trace_xdp_exception(vi
->dev
, xdp_prog
, act
);
624 ewma_pkt_len_add(&rq
->mrg_avg_pkt_len
, len
);
625 if (unlikely(xdp_page
!= page
))
630 bpf_warn_invalid_xdp_action(act
);
632 trace_xdp_exception(vi
->dev
, xdp_prog
, act
);
634 if (unlikely(xdp_page
!= page
))
635 __free_pages(xdp_page
, 0);
636 ewma_pkt_len_add(&rq
->mrg_avg_pkt_len
, len
);
642 truesize
= max(len
, mergeable_ctx_to_buf_truesize(ctx
));
643 head_skb
= page_to_skb(vi
, rq
, page
, offset
, len
, truesize
);
646 if (unlikely(!curr_skb
))
651 ctx
= (unsigned long)virtqueue_get_buf(rq
->vq
, &len
);
652 if (unlikely(!ctx
)) {
653 pr_debug("%s: rx error: %d buffers out of %d missing\n",
655 virtio16_to_cpu(vi
->vdev
,
657 dev
->stats
.rx_length_errors
++;
661 buf
= mergeable_ctx_to_buf_address(ctx
);
662 page
= virt_to_head_page(buf
);
664 num_skb_frags
= skb_shinfo(curr_skb
)->nr_frags
;
665 if (unlikely(num_skb_frags
== MAX_SKB_FRAGS
)) {
666 struct sk_buff
*nskb
= alloc_skb(0, GFP_ATOMIC
);
670 if (curr_skb
== head_skb
)
671 skb_shinfo(curr_skb
)->frag_list
= nskb
;
673 curr_skb
->next
= nskb
;
675 head_skb
->truesize
+= nskb
->truesize
;
678 truesize
= max(len
, mergeable_ctx_to_buf_truesize(ctx
));
679 if (curr_skb
!= head_skb
) {
680 head_skb
->data_len
+= len
;
681 head_skb
->len
+= len
;
682 head_skb
->truesize
+= truesize
;
684 offset
= buf
- page_address(page
);
685 if (skb_can_coalesce(curr_skb
, num_skb_frags
, page
, offset
)) {
687 skb_coalesce_rx_frag(curr_skb
, num_skb_frags
- 1,
690 skb_add_rx_frag(curr_skb
, num_skb_frags
, page
,
691 offset
, len
, truesize
);
695 ewma_pkt_len_add(&rq
->mrg_avg_pkt_len
, head_skb
->len
);
703 ctx
= (unsigned long)virtqueue_get_buf(rq
->vq
, &len
);
704 if (unlikely(!ctx
)) {
705 pr_debug("%s: rx error: %d buffers missing\n",
707 dev
->stats
.rx_length_errors
++;
710 page
= virt_to_head_page(mergeable_ctx_to_buf_address(ctx
));
714 dev
->stats
.rx_dropped
++;
715 dev_kfree_skb(head_skb
);
720 static int receive_buf(struct virtnet_info
*vi
, struct receive_queue
*rq
,
721 void *buf
, unsigned int len
)
723 struct net_device
*dev
= vi
->dev
;
725 struct virtio_net_hdr_mrg_rxbuf
*hdr
;
728 if (unlikely(len
< vi
->hdr_len
+ ETH_HLEN
)) {
729 pr_debug("%s: short packet %i\n", dev
->name
, len
);
730 dev
->stats
.rx_length_errors
++;
731 if (vi
->mergeable_rx_bufs
) {
732 unsigned long ctx
= (unsigned long)buf
;
733 void *base
= mergeable_ctx_to_buf_address(ctx
);
734 put_page(virt_to_head_page(base
));
735 } else if (vi
->big_packets
) {
738 put_page(virt_to_head_page(buf
));
743 if (vi
->mergeable_rx_bufs
)
744 skb
= receive_mergeable(dev
, vi
, rq
, (unsigned long)buf
, len
);
745 else if (vi
->big_packets
)
746 skb
= receive_big(dev
, vi
, rq
, buf
, len
);
748 skb
= receive_small(dev
, vi
, rq
, buf
, len
);
753 hdr
= skb_vnet_hdr(skb
);
757 if (hdr
->hdr
.flags
& VIRTIO_NET_HDR_F_DATA_VALID
)
758 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
760 if (virtio_net_hdr_to_skb(skb
, &hdr
->hdr
,
761 virtio_is_little_endian(vi
->vdev
))) {
762 net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n",
763 dev
->name
, hdr
->hdr
.gso_type
,
768 skb
->protocol
= eth_type_trans(skb
, dev
);
769 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
770 ntohs(skb
->protocol
), skb
->len
, skb
->pkt_type
);
772 napi_gro_receive(&rq
->napi
, skb
);
776 dev
->stats
.rx_frame_errors
++;
781 static int add_recvbuf_small(struct virtnet_info
*vi
, struct receive_queue
*rq
,
784 struct page_frag
*alloc_frag
= &rq
->alloc_frag
;
786 unsigned int xdp_headroom
= virtnet_get_headroom(vi
);
787 int len
= vi
->hdr_len
+ VIRTNET_RX_PAD
+ GOOD_PACKET_LEN
+ xdp_headroom
;
790 len
= SKB_DATA_ALIGN(len
) +
791 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
792 if (unlikely(!skb_page_frag_refill(len
, alloc_frag
, gfp
)))
795 buf
= (char *)page_address(alloc_frag
->page
) + alloc_frag
->offset
;
796 get_page(alloc_frag
->page
);
797 alloc_frag
->offset
+= len
;
798 sg_init_one(rq
->sg
, buf
+ VIRTNET_RX_PAD
+ xdp_headroom
,
799 vi
->hdr_len
+ GOOD_PACKET_LEN
);
800 err
= virtqueue_add_inbuf(rq
->vq
, rq
->sg
, 1, buf
, gfp
);
802 put_page(virt_to_head_page(buf
));
807 static int add_recvbuf_big(struct virtnet_info
*vi
, struct receive_queue
*rq
,
810 struct page
*first
, *list
= NULL
;
814 sg_init_table(rq
->sg
, MAX_SKB_FRAGS
+ 2);
816 /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */
817 for (i
= MAX_SKB_FRAGS
+ 1; i
> 1; --i
) {
818 first
= get_a_page(rq
, gfp
);
821 give_pages(rq
, list
);
824 sg_set_buf(&rq
->sg
[i
], page_address(first
), PAGE_SIZE
);
826 /* chain new page in list head to match sg */
827 first
->private = (unsigned long)list
;
831 first
= get_a_page(rq
, gfp
);
833 give_pages(rq
, list
);
836 p
= page_address(first
);
838 /* rq->sg[0], rq->sg[1] share the same page */
839 /* a separated rq->sg[0] for header - required in case !any_header_sg */
840 sg_set_buf(&rq
->sg
[0], p
, vi
->hdr_len
);
842 /* rq->sg[1] for data packet, from offset */
843 offset
= sizeof(struct padded_vnet_hdr
);
844 sg_set_buf(&rq
->sg
[1], p
+ offset
, PAGE_SIZE
- offset
);
846 /* chain first in list head */
847 first
->private = (unsigned long)list
;
848 err
= virtqueue_add_inbuf(rq
->vq
, rq
->sg
, MAX_SKB_FRAGS
+ 2,
851 give_pages(rq
, first
);
856 static unsigned int get_mergeable_buf_len(struct ewma_pkt_len
*avg_pkt_len
)
858 const size_t hdr_len
= sizeof(struct virtio_net_hdr_mrg_rxbuf
);
861 len
= hdr_len
+ clamp_t(unsigned int, ewma_pkt_len_read(avg_pkt_len
),
862 GOOD_PACKET_LEN
, PAGE_SIZE
- hdr_len
);
863 return ALIGN(len
, MERGEABLE_BUFFER_ALIGN
);
866 static int add_recvbuf_mergeable(struct virtnet_info
*vi
,
867 struct receive_queue
*rq
, gfp_t gfp
)
869 struct page_frag
*alloc_frag
= &rq
->alloc_frag
;
870 unsigned int headroom
= virtnet_get_headroom(vi
);
874 unsigned int len
, hole
;
876 len
= get_mergeable_buf_len(&rq
->mrg_avg_pkt_len
);
877 if (unlikely(!skb_page_frag_refill(len
+ headroom
, alloc_frag
, gfp
)))
880 buf
= (char *)page_address(alloc_frag
->page
) + alloc_frag
->offset
;
881 buf
+= headroom
; /* advance address leaving hole at front of pkt */
882 ctx
= mergeable_buf_to_ctx(buf
, len
);
883 get_page(alloc_frag
->page
);
884 alloc_frag
->offset
+= len
+ headroom
;
885 hole
= alloc_frag
->size
- alloc_frag
->offset
;
886 if (hole
< len
+ headroom
) {
887 /* To avoid internal fragmentation, if there is very likely not
888 * enough space for another buffer, add the remaining space to
889 * the current buffer. This extra space is not included in
890 * the truesize stored in ctx.
893 alloc_frag
->offset
+= hole
;
896 sg_init_one(rq
->sg
, buf
, len
);
897 err
= virtqueue_add_inbuf(rq
->vq
, rq
->sg
, 1, (void *)ctx
, gfp
);
899 put_page(virt_to_head_page(buf
));
905 * Returns false if we couldn't fill entirely (OOM).
907 * Normally run in the receive path, but can also be run from ndo_open
908 * before we're receiving packets, or from refill_work which is
909 * careful to disable receiving (using napi_disable).
911 static bool try_fill_recv(struct virtnet_info
*vi
, struct receive_queue
*rq
,
919 if (vi
->mergeable_rx_bufs
)
920 err
= add_recvbuf_mergeable(vi
, rq
, gfp
);
921 else if (vi
->big_packets
)
922 err
= add_recvbuf_big(vi
, rq
, gfp
);
924 err
= add_recvbuf_small(vi
, rq
, gfp
);
926 oom
= err
== -ENOMEM
;
929 } while (rq
->vq
->num_free
);
930 virtqueue_kick(rq
->vq
);
934 static void skb_recv_done(struct virtqueue
*rvq
)
936 struct virtnet_info
*vi
= rvq
->vdev
->priv
;
937 struct receive_queue
*rq
= &vi
->rq
[vq2rxq(rvq
)];
939 /* Schedule NAPI, Suppress further interrupts if successful. */
940 if (napi_schedule_prep(&rq
->napi
)) {
941 virtqueue_disable_cb(rvq
);
942 __napi_schedule(&rq
->napi
);
946 static void virtnet_napi_enable(struct receive_queue
*rq
)
948 napi_enable(&rq
->napi
);
950 /* If all buffers were filled by other side before we napi_enabled, we
951 * won't get another interrupt, so process any outstanding packets
952 * now. virtnet_poll wants re-enable the queue, so we disable here.
953 * We synchronize against interrupts via NAPI_STATE_SCHED */
954 if (napi_schedule_prep(&rq
->napi
)) {
955 virtqueue_disable_cb(rq
->vq
);
957 __napi_schedule(&rq
->napi
);
962 static void refill_work(struct work_struct
*work
)
964 struct virtnet_info
*vi
=
965 container_of(work
, struct virtnet_info
, refill
.work
);
969 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++) {
970 struct receive_queue
*rq
= &vi
->rq
[i
];
972 napi_disable(&rq
->napi
);
973 still_empty
= !try_fill_recv(vi
, rq
, GFP_KERNEL
);
974 virtnet_napi_enable(rq
);
976 /* In theory, this can happen: if we don't get any buffers in
977 * we will *never* try to fill again.
980 schedule_delayed_work(&vi
->refill
, HZ
/2);
984 static int virtnet_receive(struct receive_queue
*rq
, int budget
)
986 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
987 unsigned int len
, received
= 0, bytes
= 0;
989 struct virtnet_stats
*stats
= this_cpu_ptr(vi
->stats
);
991 while (received
< budget
&&
992 (buf
= virtqueue_get_buf(rq
->vq
, &len
)) != NULL
) {
993 bytes
+= receive_buf(vi
, rq
, buf
, len
);
997 if (rq
->vq
->num_free
> virtqueue_get_vring_size(rq
->vq
) / 2) {
998 if (!try_fill_recv(vi
, rq
, GFP_ATOMIC
))
999 schedule_delayed_work(&vi
->refill
, 0);
1002 u64_stats_update_begin(&stats
->rx_syncp
);
1003 stats
->rx_bytes
+= bytes
;
1004 stats
->rx_packets
+= received
;
1005 u64_stats_update_end(&stats
->rx_syncp
);
1010 static int virtnet_poll(struct napi_struct
*napi
, int budget
)
1012 struct receive_queue
*rq
=
1013 container_of(napi
, struct receive_queue
, napi
);
1014 unsigned int r
, received
;
1016 received
= virtnet_receive(rq
, budget
);
1018 /* Out of packets? */
1019 if (received
< budget
) {
1020 r
= virtqueue_enable_cb_prepare(rq
->vq
);
1021 if (napi_complete_done(napi
, received
)) {
1022 if (unlikely(virtqueue_poll(rq
->vq
, r
)) &&
1023 napi_schedule_prep(napi
)) {
1024 virtqueue_disable_cb(rq
->vq
);
1025 __napi_schedule(napi
);
1033 static int virtnet_open(struct net_device
*dev
)
1035 struct virtnet_info
*vi
= netdev_priv(dev
);
1038 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1039 if (i
< vi
->curr_queue_pairs
)
1040 /* Make sure we have some buffers: if oom use wq. */
1041 if (!try_fill_recv(vi
, &vi
->rq
[i
], GFP_KERNEL
))
1042 schedule_delayed_work(&vi
->refill
, 0);
1043 virtnet_napi_enable(&vi
->rq
[i
]);
1049 static void free_old_xmit_skbs(struct send_queue
*sq
)
1051 struct sk_buff
*skb
;
1053 struct virtnet_info
*vi
= sq
->vq
->vdev
->priv
;
1054 struct virtnet_stats
*stats
= this_cpu_ptr(vi
->stats
);
1055 unsigned int packets
= 0;
1056 unsigned int bytes
= 0;
1058 while ((skb
= virtqueue_get_buf(sq
->vq
, &len
)) != NULL
) {
1059 pr_debug("Sent skb %p\n", skb
);
1064 dev_kfree_skb_any(skb
);
1067 /* Avoid overhead when no packets have been processed
1068 * happens when called speculatively from start_xmit.
1073 u64_stats_update_begin(&stats
->tx_syncp
);
1074 stats
->tx_bytes
+= bytes
;
1075 stats
->tx_packets
+= packets
;
1076 u64_stats_update_end(&stats
->tx_syncp
);
1079 static int xmit_skb(struct send_queue
*sq
, struct sk_buff
*skb
)
1081 struct virtio_net_hdr_mrg_rxbuf
*hdr
;
1082 const unsigned char *dest
= ((struct ethhdr
*)skb
->data
)->h_dest
;
1083 struct virtnet_info
*vi
= sq
->vq
->vdev
->priv
;
1085 unsigned hdr_len
= vi
->hdr_len
;
1088 pr_debug("%s: xmit %p %pM\n", vi
->dev
->name
, skb
, dest
);
1090 can_push
= vi
->any_header_sg
&&
1091 !((unsigned long)skb
->data
& (__alignof__(*hdr
) - 1)) &&
1092 !skb_header_cloned(skb
) && skb_headroom(skb
) >= hdr_len
;
1093 /* Even if we can, don't push here yet as this would skew
1094 * csum_start offset below. */
1096 hdr
= (struct virtio_net_hdr_mrg_rxbuf
*)(skb
->data
- hdr_len
);
1098 hdr
= skb_vnet_hdr(skb
);
1100 if (virtio_net_hdr_from_skb(skb
, &hdr
->hdr
,
1101 virtio_is_little_endian(vi
->vdev
), false))
1104 if (vi
->mergeable_rx_bufs
)
1105 hdr
->num_buffers
= 0;
1107 sg_init_table(sq
->sg
, skb_shinfo(skb
)->nr_frags
+ (can_push
? 1 : 2));
1109 __skb_push(skb
, hdr_len
);
1110 num_sg
= skb_to_sgvec(skb
, sq
->sg
, 0, skb
->len
);
1111 /* Pull header back to avoid skew in tx bytes calculations. */
1112 __skb_pull(skb
, hdr_len
);
1114 sg_set_buf(sq
->sg
, hdr
, hdr_len
);
1115 num_sg
= skb_to_sgvec(skb
, sq
->sg
+ 1, 0, skb
->len
) + 1;
1117 return virtqueue_add_outbuf(sq
->vq
, sq
->sg
, num_sg
, skb
, GFP_ATOMIC
);
1120 static netdev_tx_t
start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
1122 struct virtnet_info
*vi
= netdev_priv(dev
);
1123 int qnum
= skb_get_queue_mapping(skb
);
1124 struct send_queue
*sq
= &vi
->sq
[qnum
];
1126 struct netdev_queue
*txq
= netdev_get_tx_queue(dev
, qnum
);
1127 bool kick
= !skb
->xmit_more
;
1129 /* Free up any pending old buffers before queueing new ones. */
1130 free_old_xmit_skbs(sq
);
1132 /* timestamp packet in software */
1133 skb_tx_timestamp(skb
);
1135 /* Try to transmit */
1136 err
= xmit_skb(sq
, skb
);
1138 /* This should not happen! */
1139 if (unlikely(err
)) {
1140 dev
->stats
.tx_fifo_errors
++;
1141 if (net_ratelimit())
1143 "Unexpected TXQ (%d) queue failure: %d\n", qnum
, err
);
1144 dev
->stats
.tx_dropped
++;
1145 dev_kfree_skb_any(skb
);
1146 return NETDEV_TX_OK
;
1149 /* Don't wait up for transmitted skbs to be freed. */
1153 /* If running out of space, stop queue to avoid getting packets that we
1154 * are then unable to transmit.
1155 * An alternative would be to force queuing layer to requeue the skb by
1156 * returning NETDEV_TX_BUSY. However, NETDEV_TX_BUSY should not be
1157 * returned in a normal path of operation: it means that driver is not
1158 * maintaining the TX queue stop/start state properly, and causes
1159 * the stack to do a non-trivial amount of useless work.
1160 * Since most packets only take 1 or 2 ring slots, stopping the queue
1161 * early means 16 slots are typically wasted.
1163 if (sq
->vq
->num_free
< 2+MAX_SKB_FRAGS
) {
1164 netif_stop_subqueue(dev
, qnum
);
1165 if (unlikely(!virtqueue_enable_cb_delayed(sq
->vq
))) {
1166 /* More just got used, free them then recheck. */
1167 free_old_xmit_skbs(sq
);
1168 if (sq
->vq
->num_free
>= 2+MAX_SKB_FRAGS
) {
1169 netif_start_subqueue(dev
, qnum
);
1170 virtqueue_disable_cb(sq
->vq
);
1175 if (kick
|| netif_xmit_stopped(txq
))
1176 virtqueue_kick(sq
->vq
);
1178 return NETDEV_TX_OK
;
1182 * Send command via the control virtqueue and check status. Commands
1183 * supported by the hypervisor, as indicated by feature bits, should
1184 * never fail unless improperly formatted.
1186 static bool virtnet_send_command(struct virtnet_info
*vi
, u8
class, u8 cmd
,
1187 struct scatterlist
*out
)
1189 struct scatterlist
*sgs
[4], hdr
, stat
;
1190 unsigned out_num
= 0, tmp
;
1192 /* Caller should know better */
1193 BUG_ON(!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
));
1195 vi
->ctrl_status
= ~0;
1196 vi
->ctrl_hdr
.class = class;
1197 vi
->ctrl_hdr
.cmd
= cmd
;
1199 sg_init_one(&hdr
, &vi
->ctrl_hdr
, sizeof(vi
->ctrl_hdr
));
1200 sgs
[out_num
++] = &hdr
;
1203 sgs
[out_num
++] = out
;
1205 /* Add return status. */
1206 sg_init_one(&stat
, &vi
->ctrl_status
, sizeof(vi
->ctrl_status
));
1207 sgs
[out_num
] = &stat
;
1209 BUG_ON(out_num
+ 1 > ARRAY_SIZE(sgs
));
1210 virtqueue_add_sgs(vi
->cvq
, sgs
, out_num
, 1, vi
, GFP_ATOMIC
);
1212 if (unlikely(!virtqueue_kick(vi
->cvq
)))
1213 return vi
->ctrl_status
== VIRTIO_NET_OK
;
1215 /* Spin for a response, the kick causes an ioport write, trapping
1216 * into the hypervisor, so the request should be handled immediately.
1218 while (!virtqueue_get_buf(vi
->cvq
, &tmp
) &&
1219 !virtqueue_is_broken(vi
->cvq
))
1222 return vi
->ctrl_status
== VIRTIO_NET_OK
;
1225 static int virtnet_set_mac_address(struct net_device
*dev
, void *p
)
1227 struct virtnet_info
*vi
= netdev_priv(dev
);
1228 struct virtio_device
*vdev
= vi
->vdev
;
1230 struct sockaddr
*addr
;
1231 struct scatterlist sg
;
1233 addr
= kmemdup(p
, sizeof(*addr
), GFP_KERNEL
);
1237 ret
= eth_prepare_mac_addr_change(dev
, addr
);
1241 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_MAC_ADDR
)) {
1242 sg_init_one(&sg
, addr
->sa_data
, dev
->addr_len
);
1243 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MAC
,
1244 VIRTIO_NET_CTRL_MAC_ADDR_SET
, &sg
)) {
1245 dev_warn(&vdev
->dev
,
1246 "Failed to set mac address by vq command.\n");
1250 } else if (virtio_has_feature(vdev
, VIRTIO_NET_F_MAC
) &&
1251 !virtio_has_feature(vdev
, VIRTIO_F_VERSION_1
)) {
1254 /* Naturally, this has an atomicity problem. */
1255 for (i
= 0; i
< dev
->addr_len
; i
++)
1256 virtio_cwrite8(vdev
,
1257 offsetof(struct virtio_net_config
, mac
) +
1258 i
, addr
->sa_data
[i
]);
1261 eth_commit_mac_addr_change(dev
, p
);
1269 static void virtnet_stats(struct net_device
*dev
,
1270 struct rtnl_link_stats64
*tot
)
1272 struct virtnet_info
*vi
= netdev_priv(dev
);
1276 for_each_possible_cpu(cpu
) {
1277 struct virtnet_stats
*stats
= per_cpu_ptr(vi
->stats
, cpu
);
1278 u64 tpackets
, tbytes
, rpackets
, rbytes
;
1281 start
= u64_stats_fetch_begin_irq(&stats
->tx_syncp
);
1282 tpackets
= stats
->tx_packets
;
1283 tbytes
= stats
->tx_bytes
;
1284 } while (u64_stats_fetch_retry_irq(&stats
->tx_syncp
, start
));
1287 start
= u64_stats_fetch_begin_irq(&stats
->rx_syncp
);
1288 rpackets
= stats
->rx_packets
;
1289 rbytes
= stats
->rx_bytes
;
1290 } while (u64_stats_fetch_retry_irq(&stats
->rx_syncp
, start
));
1292 tot
->rx_packets
+= rpackets
;
1293 tot
->tx_packets
+= tpackets
;
1294 tot
->rx_bytes
+= rbytes
;
1295 tot
->tx_bytes
+= tbytes
;
1298 tot
->tx_dropped
= dev
->stats
.tx_dropped
;
1299 tot
->tx_fifo_errors
= dev
->stats
.tx_fifo_errors
;
1300 tot
->rx_dropped
= dev
->stats
.rx_dropped
;
1301 tot
->rx_length_errors
= dev
->stats
.rx_length_errors
;
1302 tot
->rx_frame_errors
= dev
->stats
.rx_frame_errors
;
1305 #ifdef CONFIG_NET_POLL_CONTROLLER
1306 static void virtnet_netpoll(struct net_device
*dev
)
1308 struct virtnet_info
*vi
= netdev_priv(dev
);
1311 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++)
1312 napi_schedule(&vi
->rq
[i
].napi
);
1316 static void virtnet_ack_link_announce(struct virtnet_info
*vi
)
1319 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_ANNOUNCE
,
1320 VIRTIO_NET_CTRL_ANNOUNCE_ACK
, NULL
))
1321 dev_warn(&vi
->dev
->dev
, "Failed to ack link announce.\n");
1325 static int _virtnet_set_queues(struct virtnet_info
*vi
, u16 queue_pairs
)
1327 struct scatterlist sg
;
1328 struct net_device
*dev
= vi
->dev
;
1330 if (!vi
->has_cvq
|| !virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_MQ
))
1333 vi
->ctrl_mq
.virtqueue_pairs
= cpu_to_virtio16(vi
->vdev
, queue_pairs
);
1334 sg_init_one(&sg
, &vi
->ctrl_mq
, sizeof(vi
->ctrl_mq
));
1336 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MQ
,
1337 VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET
, &sg
)) {
1338 dev_warn(&dev
->dev
, "Fail to set num of queue pairs to %d\n",
1342 vi
->curr_queue_pairs
= queue_pairs
;
1343 /* virtnet_open() will refill when device is going to up. */
1344 if (dev
->flags
& IFF_UP
)
1345 schedule_delayed_work(&vi
->refill
, 0);
1351 static int virtnet_set_queues(struct virtnet_info
*vi
, u16 queue_pairs
)
1356 err
= _virtnet_set_queues(vi
, queue_pairs
);
1361 static int virtnet_close(struct net_device
*dev
)
1363 struct virtnet_info
*vi
= netdev_priv(dev
);
1366 /* Make sure refill_work doesn't re-enable napi! */
1367 cancel_delayed_work_sync(&vi
->refill
);
1369 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1370 napi_disable(&vi
->rq
[i
].napi
);
1375 static void virtnet_set_rx_mode(struct net_device
*dev
)
1377 struct virtnet_info
*vi
= netdev_priv(dev
);
1378 struct scatterlist sg
[2];
1379 struct virtio_net_ctrl_mac
*mac_data
;
1380 struct netdev_hw_addr
*ha
;
1386 /* We can't dynamically set ndo_set_rx_mode, so return gracefully */
1387 if (!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_RX
))
1390 vi
->ctrl_promisc
= ((dev
->flags
& IFF_PROMISC
) != 0);
1391 vi
->ctrl_allmulti
= ((dev
->flags
& IFF_ALLMULTI
) != 0);
1393 sg_init_one(sg
, &vi
->ctrl_promisc
, sizeof(vi
->ctrl_promisc
));
1395 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_RX
,
1396 VIRTIO_NET_CTRL_RX_PROMISC
, sg
))
1397 dev_warn(&dev
->dev
, "Failed to %sable promisc mode.\n",
1398 vi
->ctrl_promisc
? "en" : "dis");
1400 sg_init_one(sg
, &vi
->ctrl_allmulti
, sizeof(vi
->ctrl_allmulti
));
1402 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_RX
,
1403 VIRTIO_NET_CTRL_RX_ALLMULTI
, sg
))
1404 dev_warn(&dev
->dev
, "Failed to %sable allmulti mode.\n",
1405 vi
->ctrl_allmulti
? "en" : "dis");
1407 uc_count
= netdev_uc_count(dev
);
1408 mc_count
= netdev_mc_count(dev
);
1409 /* MAC filter - use one buffer for both lists */
1410 buf
= kzalloc(((uc_count
+ mc_count
) * ETH_ALEN
) +
1411 (2 * sizeof(mac_data
->entries
)), GFP_ATOMIC
);
1416 sg_init_table(sg
, 2);
1418 /* Store the unicast list and count in the front of the buffer */
1419 mac_data
->entries
= cpu_to_virtio32(vi
->vdev
, uc_count
);
1421 netdev_for_each_uc_addr(ha
, dev
)
1422 memcpy(&mac_data
->macs
[i
++][0], ha
->addr
, ETH_ALEN
);
1424 sg_set_buf(&sg
[0], mac_data
,
1425 sizeof(mac_data
->entries
) + (uc_count
* ETH_ALEN
));
1427 /* multicast list and count fill the end */
1428 mac_data
= (void *)&mac_data
->macs
[uc_count
][0];
1430 mac_data
->entries
= cpu_to_virtio32(vi
->vdev
, mc_count
);
1432 netdev_for_each_mc_addr(ha
, dev
)
1433 memcpy(&mac_data
->macs
[i
++][0], ha
->addr
, ETH_ALEN
);
1435 sg_set_buf(&sg
[1], mac_data
,
1436 sizeof(mac_data
->entries
) + (mc_count
* ETH_ALEN
));
1438 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MAC
,
1439 VIRTIO_NET_CTRL_MAC_TABLE_SET
, sg
))
1440 dev_warn(&dev
->dev
, "Failed to set MAC filter table.\n");
1445 static int virtnet_vlan_rx_add_vid(struct net_device
*dev
,
1446 __be16 proto
, u16 vid
)
1448 struct virtnet_info
*vi
= netdev_priv(dev
);
1449 struct scatterlist sg
;
1452 sg_init_one(&sg
, &vi
->ctrl_vid
, sizeof(vi
->ctrl_vid
));
1454 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_VLAN
,
1455 VIRTIO_NET_CTRL_VLAN_ADD
, &sg
))
1456 dev_warn(&dev
->dev
, "Failed to add VLAN ID %d.\n", vid
);
1460 static int virtnet_vlan_rx_kill_vid(struct net_device
*dev
,
1461 __be16 proto
, u16 vid
)
1463 struct virtnet_info
*vi
= netdev_priv(dev
);
1464 struct scatterlist sg
;
1467 sg_init_one(&sg
, &vi
->ctrl_vid
, sizeof(vi
->ctrl_vid
));
1469 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_VLAN
,
1470 VIRTIO_NET_CTRL_VLAN_DEL
, &sg
))
1471 dev_warn(&dev
->dev
, "Failed to kill VLAN ID %d.\n", vid
);
1475 static void virtnet_clean_affinity(struct virtnet_info
*vi
, long hcpu
)
1479 if (vi
->affinity_hint_set
) {
1480 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1481 virtqueue_set_affinity(vi
->rq
[i
].vq
, -1);
1482 virtqueue_set_affinity(vi
->sq
[i
].vq
, -1);
1485 vi
->affinity_hint_set
= false;
1489 static void virtnet_set_affinity(struct virtnet_info
*vi
)
1494 /* In multiqueue mode, when the number of cpu is equal to the number of
1495 * queue pairs, we let the queue pairs to be private to one cpu by
1496 * setting the affinity hint to eliminate the contention.
1498 if (vi
->curr_queue_pairs
== 1 ||
1499 vi
->max_queue_pairs
!= num_online_cpus()) {
1500 virtnet_clean_affinity(vi
, -1);
1505 for_each_online_cpu(cpu
) {
1506 virtqueue_set_affinity(vi
->rq
[i
].vq
, cpu
);
1507 virtqueue_set_affinity(vi
->sq
[i
].vq
, cpu
);
1508 netif_set_xps_queue(vi
->dev
, cpumask_of(cpu
), i
);
1512 vi
->affinity_hint_set
= true;
1515 static int virtnet_cpu_online(unsigned int cpu
, struct hlist_node
*node
)
1517 struct virtnet_info
*vi
= hlist_entry_safe(node
, struct virtnet_info
,
1519 virtnet_set_affinity(vi
);
1523 static int virtnet_cpu_dead(unsigned int cpu
, struct hlist_node
*node
)
1525 struct virtnet_info
*vi
= hlist_entry_safe(node
, struct virtnet_info
,
1527 virtnet_set_affinity(vi
);
1531 static int virtnet_cpu_down_prep(unsigned int cpu
, struct hlist_node
*node
)
1533 struct virtnet_info
*vi
= hlist_entry_safe(node
, struct virtnet_info
,
1536 virtnet_clean_affinity(vi
, cpu
);
1540 static enum cpuhp_state virtionet_online
;
1542 static int virtnet_cpu_notif_add(struct virtnet_info
*vi
)
1546 ret
= cpuhp_state_add_instance_nocalls(virtionet_online
, &vi
->node
);
1549 ret
= cpuhp_state_add_instance_nocalls(CPUHP_VIRT_NET_DEAD
,
1553 cpuhp_state_remove_instance_nocalls(virtionet_online
, &vi
->node
);
1557 static void virtnet_cpu_notif_remove(struct virtnet_info
*vi
)
1559 cpuhp_state_remove_instance_nocalls(virtionet_online
, &vi
->node
);
1560 cpuhp_state_remove_instance_nocalls(CPUHP_VIRT_NET_DEAD
,
1564 static void virtnet_get_ringparam(struct net_device
*dev
,
1565 struct ethtool_ringparam
*ring
)
1567 struct virtnet_info
*vi
= netdev_priv(dev
);
1569 ring
->rx_max_pending
= virtqueue_get_vring_size(vi
->rq
[0].vq
);
1570 ring
->tx_max_pending
= virtqueue_get_vring_size(vi
->sq
[0].vq
);
1571 ring
->rx_pending
= ring
->rx_max_pending
;
1572 ring
->tx_pending
= ring
->tx_max_pending
;
1576 static void virtnet_get_drvinfo(struct net_device
*dev
,
1577 struct ethtool_drvinfo
*info
)
1579 struct virtnet_info
*vi
= netdev_priv(dev
);
1580 struct virtio_device
*vdev
= vi
->vdev
;
1582 strlcpy(info
->driver
, KBUILD_MODNAME
, sizeof(info
->driver
));
1583 strlcpy(info
->version
, VIRTNET_DRIVER_VERSION
, sizeof(info
->version
));
1584 strlcpy(info
->bus_info
, virtio_bus_name(vdev
), sizeof(info
->bus_info
));
1588 /* TODO: Eliminate OOO packets during switching */
1589 static int virtnet_set_channels(struct net_device
*dev
,
1590 struct ethtool_channels
*channels
)
1592 struct virtnet_info
*vi
= netdev_priv(dev
);
1593 u16 queue_pairs
= channels
->combined_count
;
1596 /* We don't support separate rx/tx channels.
1597 * We don't allow setting 'other' channels.
1599 if (channels
->rx_count
|| channels
->tx_count
|| channels
->other_count
)
1602 if (queue_pairs
> vi
->max_queue_pairs
|| queue_pairs
== 0)
1605 /* For now we don't support modifying channels while XDP is loaded
1606 * also when XDP is loaded all RX queues have XDP programs so we only
1607 * need to check a single RX queue.
1609 if (vi
->rq
[0].xdp_prog
)
1613 err
= _virtnet_set_queues(vi
, queue_pairs
);
1615 netif_set_real_num_tx_queues(dev
, queue_pairs
);
1616 netif_set_real_num_rx_queues(dev
, queue_pairs
);
1618 virtnet_set_affinity(vi
);
1625 static void virtnet_get_channels(struct net_device
*dev
,
1626 struct ethtool_channels
*channels
)
1628 struct virtnet_info
*vi
= netdev_priv(dev
);
1630 channels
->combined_count
= vi
->curr_queue_pairs
;
1631 channels
->max_combined
= vi
->max_queue_pairs
;
1632 channels
->max_other
= 0;
1633 channels
->rx_count
= 0;
1634 channels
->tx_count
= 0;
1635 channels
->other_count
= 0;
1638 /* Check if the user is trying to change anything besides speed/duplex */
1639 static bool virtnet_validate_ethtool_cmd(const struct ethtool_cmd
*cmd
)
1641 struct ethtool_cmd diff1
= *cmd
;
1642 struct ethtool_cmd diff2
= {};
1644 /* cmd is always set so we need to clear it, validate the port type
1645 * and also without autonegotiation we can ignore advertising
1647 ethtool_cmd_speed_set(&diff1
, 0);
1648 diff2
.port
= PORT_OTHER
;
1649 diff1
.advertising
= 0;
1653 return !memcmp(&diff1
, &diff2
, sizeof(diff1
));
1656 static int virtnet_set_settings(struct net_device
*dev
, struct ethtool_cmd
*cmd
)
1658 struct virtnet_info
*vi
= netdev_priv(dev
);
1661 speed
= ethtool_cmd_speed(cmd
);
1662 /* don't allow custom speed and duplex */
1663 if (!ethtool_validate_speed(speed
) ||
1664 !ethtool_validate_duplex(cmd
->duplex
) ||
1665 !virtnet_validate_ethtool_cmd(cmd
))
1668 vi
->duplex
= cmd
->duplex
;
1673 static int virtnet_get_settings(struct net_device
*dev
, struct ethtool_cmd
*cmd
)
1675 struct virtnet_info
*vi
= netdev_priv(dev
);
1677 ethtool_cmd_speed_set(cmd
, vi
->speed
);
1678 cmd
->duplex
= vi
->duplex
;
1679 cmd
->port
= PORT_OTHER
;
1684 static void virtnet_init_settings(struct net_device
*dev
)
1686 struct virtnet_info
*vi
= netdev_priv(dev
);
1688 vi
->speed
= SPEED_UNKNOWN
;
1689 vi
->duplex
= DUPLEX_UNKNOWN
;
1692 static const struct ethtool_ops virtnet_ethtool_ops
= {
1693 .get_drvinfo
= virtnet_get_drvinfo
,
1694 .get_link
= ethtool_op_get_link
,
1695 .get_ringparam
= virtnet_get_ringparam
,
1696 .set_channels
= virtnet_set_channels
,
1697 .get_channels
= virtnet_get_channels
,
1698 .get_ts_info
= ethtool_op_get_ts_info
,
1699 .get_settings
= virtnet_get_settings
,
1700 .set_settings
= virtnet_set_settings
,
1703 static void virtnet_freeze_down(struct virtio_device
*vdev
)
1705 struct virtnet_info
*vi
= vdev
->priv
;
1708 /* Make sure no work handler is accessing the device */
1709 flush_work(&vi
->config_work
);
1711 netif_device_detach(vi
->dev
);
1712 cancel_delayed_work_sync(&vi
->refill
);
1714 if (netif_running(vi
->dev
)) {
1715 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1716 napi_disable(&vi
->rq
[i
].napi
);
1720 static int init_vqs(struct virtnet_info
*vi
);
1721 static void _remove_vq_common(struct virtnet_info
*vi
);
1723 static int virtnet_restore_up(struct virtio_device
*vdev
)
1725 struct virtnet_info
*vi
= vdev
->priv
;
1732 virtio_device_ready(vdev
);
1734 if (netif_running(vi
->dev
)) {
1735 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++)
1736 if (!try_fill_recv(vi
, &vi
->rq
[i
], GFP_KERNEL
))
1737 schedule_delayed_work(&vi
->refill
, 0);
1739 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1740 virtnet_napi_enable(&vi
->rq
[i
]);
1743 netif_device_attach(vi
->dev
);
1747 static int virtnet_reset(struct virtnet_info
*vi
, int curr_qp
, int xdp_qp
)
1749 struct virtio_device
*dev
= vi
->vdev
;
1752 virtio_config_disable(dev
);
1753 dev
->failed
= dev
->config
->get_status(dev
) & VIRTIO_CONFIG_S_FAILED
;
1754 virtnet_freeze_down(dev
);
1755 _remove_vq_common(vi
);
1757 dev
->config
->reset(dev
);
1758 virtio_add_status(dev
, VIRTIO_CONFIG_S_ACKNOWLEDGE
);
1759 virtio_add_status(dev
, VIRTIO_CONFIG_S_DRIVER
);
1761 ret
= virtio_finalize_features(dev
);
1765 vi
->xdp_queue_pairs
= xdp_qp
;
1766 ret
= virtnet_restore_up(dev
);
1769 ret
= _virtnet_set_queues(vi
, curr_qp
);
1773 virtio_add_status(dev
, VIRTIO_CONFIG_S_DRIVER_OK
);
1774 virtio_config_enable(dev
);
1777 virtio_add_status(dev
, VIRTIO_CONFIG_S_FAILED
);
1781 static int virtnet_xdp_set(struct net_device
*dev
, struct bpf_prog
*prog
)
1783 unsigned long int max_sz
= PAGE_SIZE
- sizeof(struct padded_vnet_hdr
);
1784 struct virtnet_info
*vi
= netdev_priv(dev
);
1785 struct bpf_prog
*old_prog
;
1786 u16 xdp_qp
= 0, curr_qp
;
1789 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_GUEST_TSO4
) ||
1790 virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_GUEST_TSO6
) ||
1791 virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_GUEST_ECN
) ||
1792 virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_GUEST_UFO
)) {
1793 netdev_warn(dev
, "can't set XDP while host is implementing LRO, disable LRO first\n");
1797 if (vi
->mergeable_rx_bufs
&& !vi
->any_header_sg
) {
1798 netdev_warn(dev
, "XDP expects header/data in single page, any_header_sg required\n");
1802 if (dev
->mtu
> max_sz
) {
1803 netdev_warn(dev
, "XDP requires MTU less than %lu\n", max_sz
);
1807 curr_qp
= vi
->curr_queue_pairs
- vi
->xdp_queue_pairs
;
1809 xdp_qp
= nr_cpu_ids
;
1811 /* XDP requires extra queues for XDP_TX */
1812 if (curr_qp
+ xdp_qp
> vi
->max_queue_pairs
) {
1813 netdev_warn(dev
, "request %i queues but max is %i\n",
1814 curr_qp
+ xdp_qp
, vi
->max_queue_pairs
);
1819 prog
= bpf_prog_add(prog
, vi
->max_queue_pairs
- 1);
1821 return PTR_ERR(prog
);
1824 /* Changing the headroom in buffers is a disruptive operation because
1825 * existing buffers must be flushed and reallocated. This will happen
1826 * when a xdp program is initially added or xdp is disabled by removing
1827 * the xdp program resulting in number of XDP queues changing.
1829 if (vi
->xdp_queue_pairs
!= xdp_qp
) {
1830 err
= virtnet_reset(vi
, curr_qp
+ xdp_qp
, xdp_qp
);
1832 dev_warn(&dev
->dev
, "XDP reset failure.\n");
1833 goto virtio_reset_err
;
1837 netif_set_real_num_rx_queues(dev
, curr_qp
+ xdp_qp
);
1839 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1840 old_prog
= rtnl_dereference(vi
->rq
[i
].xdp_prog
);
1841 rcu_assign_pointer(vi
->rq
[i
].xdp_prog
, prog
);
1843 bpf_prog_put(old_prog
);
1849 /* On reset error do our best to unwind XDP changes inflight and return
1850 * error up to user space for resolution. The underlying reset hung on
1851 * us so not much we can do here.
1854 bpf_prog_sub(prog
, vi
->max_queue_pairs
- 1);
1858 static bool virtnet_xdp_query(struct net_device
*dev
)
1860 struct virtnet_info
*vi
= netdev_priv(dev
);
1863 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1864 if (vi
->rq
[i
].xdp_prog
)
1870 static int virtnet_xdp(struct net_device
*dev
, struct netdev_xdp
*xdp
)
1872 switch (xdp
->command
) {
1873 case XDP_SETUP_PROG
:
1874 return virtnet_xdp_set(dev
, xdp
->prog
);
1875 case XDP_QUERY_PROG
:
1876 xdp
->prog_attached
= virtnet_xdp_query(dev
);
1883 static const struct net_device_ops virtnet_netdev
= {
1884 .ndo_open
= virtnet_open
,
1885 .ndo_stop
= virtnet_close
,
1886 .ndo_start_xmit
= start_xmit
,
1887 .ndo_validate_addr
= eth_validate_addr
,
1888 .ndo_set_mac_address
= virtnet_set_mac_address
,
1889 .ndo_set_rx_mode
= virtnet_set_rx_mode
,
1890 .ndo_get_stats64
= virtnet_stats
,
1891 .ndo_vlan_rx_add_vid
= virtnet_vlan_rx_add_vid
,
1892 .ndo_vlan_rx_kill_vid
= virtnet_vlan_rx_kill_vid
,
1893 #ifdef CONFIG_NET_POLL_CONTROLLER
1894 .ndo_poll_controller
= virtnet_netpoll
,
1896 .ndo_xdp
= virtnet_xdp
,
1899 static void virtnet_config_changed_work(struct work_struct
*work
)
1901 struct virtnet_info
*vi
=
1902 container_of(work
, struct virtnet_info
, config_work
);
1905 if (virtio_cread_feature(vi
->vdev
, VIRTIO_NET_F_STATUS
,
1906 struct virtio_net_config
, status
, &v
) < 0)
1909 if (v
& VIRTIO_NET_S_ANNOUNCE
) {
1910 netdev_notify_peers(vi
->dev
);
1911 virtnet_ack_link_announce(vi
);
1914 /* Ignore unknown (future) status bits */
1915 v
&= VIRTIO_NET_S_LINK_UP
;
1917 if (vi
->status
== v
)
1922 if (vi
->status
& VIRTIO_NET_S_LINK_UP
) {
1923 netif_carrier_on(vi
->dev
);
1924 netif_tx_wake_all_queues(vi
->dev
);
1926 netif_carrier_off(vi
->dev
);
1927 netif_tx_stop_all_queues(vi
->dev
);
1931 static void virtnet_config_changed(struct virtio_device
*vdev
)
1933 struct virtnet_info
*vi
= vdev
->priv
;
1935 schedule_work(&vi
->config_work
);
1938 static void virtnet_free_queues(struct virtnet_info
*vi
)
1942 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1943 napi_hash_del(&vi
->rq
[i
].napi
);
1944 netif_napi_del(&vi
->rq
[i
].napi
);
1947 /* We called napi_hash_del() before netif_napi_del(),
1948 * we need to respect an RCU grace period before freeing vi->rq
1956 static void _free_receive_bufs(struct virtnet_info
*vi
)
1958 struct bpf_prog
*old_prog
;
1961 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1962 while (vi
->rq
[i
].pages
)
1963 __free_pages(get_a_page(&vi
->rq
[i
], GFP_KERNEL
), 0);
1965 old_prog
= rtnl_dereference(vi
->rq
[i
].xdp_prog
);
1966 RCU_INIT_POINTER(vi
->rq
[i
].xdp_prog
, NULL
);
1968 bpf_prog_put(old_prog
);
1972 static void free_receive_bufs(struct virtnet_info
*vi
)
1975 _free_receive_bufs(vi
);
1979 static void free_receive_page_frags(struct virtnet_info
*vi
)
1982 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
1983 if (vi
->rq
[i
].alloc_frag
.page
)
1984 put_page(vi
->rq
[i
].alloc_frag
.page
);
1987 static bool is_xdp_raw_buffer_queue(struct virtnet_info
*vi
, int q
)
1989 if (q
< (vi
->curr_queue_pairs
- vi
->xdp_queue_pairs
))
1991 else if (q
< vi
->curr_queue_pairs
)
1997 static void free_unused_bufs(struct virtnet_info
*vi
)
2002 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
2003 struct virtqueue
*vq
= vi
->sq
[i
].vq
;
2004 while ((buf
= virtqueue_detach_unused_buf(vq
)) != NULL
) {
2005 if (!is_xdp_raw_buffer_queue(vi
, i
))
2008 put_page(virt_to_head_page(buf
));
2012 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
2013 struct virtqueue
*vq
= vi
->rq
[i
].vq
;
2015 while ((buf
= virtqueue_detach_unused_buf(vq
)) != NULL
) {
2016 if (vi
->mergeable_rx_bufs
) {
2017 unsigned long ctx
= (unsigned long)buf
;
2018 void *base
= mergeable_ctx_to_buf_address(ctx
);
2019 put_page(virt_to_head_page(base
));
2020 } else if (vi
->big_packets
) {
2021 give_pages(&vi
->rq
[i
], buf
);
2023 put_page(virt_to_head_page(buf
));
2029 static void virtnet_del_vqs(struct virtnet_info
*vi
)
2031 struct virtio_device
*vdev
= vi
->vdev
;
2033 virtnet_clean_affinity(vi
, -1);
2035 vdev
->config
->del_vqs(vdev
);
2037 virtnet_free_queues(vi
);
2040 static int virtnet_find_vqs(struct virtnet_info
*vi
)
2042 vq_callback_t
**callbacks
;
2043 struct virtqueue
**vqs
;
2048 /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
2049 * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
2050 * possible control vq.
2052 total_vqs
= vi
->max_queue_pairs
* 2 +
2053 virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
);
2055 /* Allocate space for find_vqs parameters */
2056 vqs
= kzalloc(total_vqs
* sizeof(*vqs
), GFP_KERNEL
);
2059 callbacks
= kmalloc(total_vqs
* sizeof(*callbacks
), GFP_KERNEL
);
2062 names
= kmalloc(total_vqs
* sizeof(*names
), GFP_KERNEL
);
2066 /* Parameters for control virtqueue, if any */
2068 callbacks
[total_vqs
- 1] = NULL
;
2069 names
[total_vqs
- 1] = "control";
2072 /* Allocate/initialize parameters for send/receive virtqueues */
2073 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
2074 callbacks
[rxq2vq(i
)] = skb_recv_done
;
2075 callbacks
[txq2vq(i
)] = skb_xmit_done
;
2076 sprintf(vi
->rq
[i
].name
, "input.%d", i
);
2077 sprintf(vi
->sq
[i
].name
, "output.%d", i
);
2078 names
[rxq2vq(i
)] = vi
->rq
[i
].name
;
2079 names
[txq2vq(i
)] = vi
->sq
[i
].name
;
2082 ret
= vi
->vdev
->config
->find_vqs(vi
->vdev
, total_vqs
, vqs
, callbacks
,
2088 vi
->cvq
= vqs
[total_vqs
- 1];
2089 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VLAN
))
2090 vi
->dev
->features
|= NETIF_F_HW_VLAN_CTAG_FILTER
;
2093 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
2094 vi
->rq
[i
].vq
= vqs
[rxq2vq(i
)];
2095 vi
->sq
[i
].vq
= vqs
[txq2vq(i
)];
2114 static int virtnet_alloc_queues(struct virtnet_info
*vi
)
2118 vi
->sq
= kzalloc(sizeof(*vi
->sq
) * vi
->max_queue_pairs
, GFP_KERNEL
);
2121 vi
->rq
= kzalloc(sizeof(*vi
->rq
) * vi
->max_queue_pairs
, GFP_KERNEL
);
2125 INIT_DELAYED_WORK(&vi
->refill
, refill_work
);
2126 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
2127 vi
->rq
[i
].pages
= NULL
;
2128 netif_napi_add(vi
->dev
, &vi
->rq
[i
].napi
, virtnet_poll
,
2131 sg_init_table(vi
->rq
[i
].sg
, ARRAY_SIZE(vi
->rq
[i
].sg
));
2132 ewma_pkt_len_init(&vi
->rq
[i
].mrg_avg_pkt_len
);
2133 sg_init_table(vi
->sq
[i
].sg
, ARRAY_SIZE(vi
->sq
[i
].sg
));
2144 static int init_vqs(struct virtnet_info
*vi
)
2148 /* Allocate send & receive queues */
2149 ret
= virtnet_alloc_queues(vi
);
2153 ret
= virtnet_find_vqs(vi
);
2158 virtnet_set_affinity(vi
);
2164 virtnet_free_queues(vi
);
2170 static ssize_t
mergeable_rx_buffer_size_show(struct netdev_rx_queue
*queue
,
2171 struct rx_queue_attribute
*attribute
, char *buf
)
2173 struct virtnet_info
*vi
= netdev_priv(queue
->dev
);
2174 unsigned int queue_index
= get_netdev_rx_queue_index(queue
);
2175 struct ewma_pkt_len
*avg
;
2177 BUG_ON(queue_index
>= vi
->max_queue_pairs
);
2178 avg
= &vi
->rq
[queue_index
].mrg_avg_pkt_len
;
2179 return sprintf(buf
, "%u\n", get_mergeable_buf_len(avg
));
2182 static struct rx_queue_attribute mergeable_rx_buffer_size_attribute
=
2183 __ATTR_RO(mergeable_rx_buffer_size
);
2185 static struct attribute
*virtio_net_mrg_rx_attrs
[] = {
2186 &mergeable_rx_buffer_size_attribute
.attr
,
2190 static const struct attribute_group virtio_net_mrg_rx_group
= {
2191 .name
= "virtio_net",
2192 .attrs
= virtio_net_mrg_rx_attrs
2196 static bool virtnet_fail_on_feature(struct virtio_device
*vdev
,
2198 const char *fname
, const char *dname
)
2200 if (!virtio_has_feature(vdev
, fbit
))
2203 dev_err(&vdev
->dev
, "device advertises feature %s but not %s",
2209 #define VIRTNET_FAIL_ON(vdev, fbit, dbit) \
2210 virtnet_fail_on_feature(vdev, fbit, #fbit, dbit)
2212 static bool virtnet_validate_features(struct virtio_device
*vdev
)
2214 if (!virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
) &&
2215 (VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_CTRL_RX
,
2216 "VIRTIO_NET_F_CTRL_VQ") ||
2217 VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_CTRL_VLAN
,
2218 "VIRTIO_NET_F_CTRL_VQ") ||
2219 VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_GUEST_ANNOUNCE
,
2220 "VIRTIO_NET_F_CTRL_VQ") ||
2221 VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_MQ
, "VIRTIO_NET_F_CTRL_VQ") ||
2222 VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_CTRL_MAC_ADDR
,
2223 "VIRTIO_NET_F_CTRL_VQ"))) {
2230 #define MIN_MTU ETH_MIN_MTU
2231 #define MAX_MTU ETH_MAX_MTU
2233 static int virtnet_validate(struct virtio_device
*vdev
)
2235 if (!vdev
->config
->get
) {
2236 dev_err(&vdev
->dev
, "%s failure: config access disabled\n",
2241 if (!virtnet_validate_features(vdev
))
2244 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MTU
)) {
2245 int mtu
= virtio_cread16(vdev
,
2246 offsetof(struct virtio_net_config
,
2249 __virtio_clear_bit(vdev
, VIRTIO_NET_F_MTU
);
2255 static int virtnet_probe(struct virtio_device
*vdev
)
2258 struct net_device
*dev
;
2259 struct virtnet_info
*vi
;
2260 u16 max_queue_pairs
;
2263 /* Find if host supports multiqueue virtio_net device */
2264 err
= virtio_cread_feature(vdev
, VIRTIO_NET_F_MQ
,
2265 struct virtio_net_config
,
2266 max_virtqueue_pairs
, &max_queue_pairs
);
2268 /* We need at least 2 queue's */
2269 if (err
|| max_queue_pairs
< VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN
||
2270 max_queue_pairs
> VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX
||
2271 !virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
))
2272 max_queue_pairs
= 1;
2274 /* Allocate ourselves a network device with room for our info */
2275 dev
= alloc_etherdev_mq(sizeof(struct virtnet_info
), max_queue_pairs
);
2279 /* Set up network device as normal. */
2280 dev
->priv_flags
|= IFF_UNICAST_FLT
| IFF_LIVE_ADDR_CHANGE
;
2281 dev
->netdev_ops
= &virtnet_netdev
;
2282 dev
->features
= NETIF_F_HIGHDMA
;
2284 dev
->ethtool_ops
= &virtnet_ethtool_ops
;
2285 SET_NETDEV_DEV(dev
, &vdev
->dev
);
2287 /* Do we support "hardware" checksums? */
2288 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CSUM
)) {
2289 /* This opens up the world of extra features. */
2290 dev
->hw_features
|= NETIF_F_HW_CSUM
| NETIF_F_SG
;
2292 dev
->features
|= NETIF_F_HW_CSUM
| NETIF_F_SG
;
2294 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GSO
)) {
2295 dev
->hw_features
|= NETIF_F_TSO
| NETIF_F_UFO
2296 | NETIF_F_TSO_ECN
| NETIF_F_TSO6
;
2298 /* Individual feature bits: what can host handle? */
2299 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_TSO4
))
2300 dev
->hw_features
|= NETIF_F_TSO
;
2301 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_TSO6
))
2302 dev
->hw_features
|= NETIF_F_TSO6
;
2303 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_ECN
))
2304 dev
->hw_features
|= NETIF_F_TSO_ECN
;
2305 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_UFO
))
2306 dev
->hw_features
|= NETIF_F_UFO
;
2308 dev
->features
|= NETIF_F_GSO_ROBUST
;
2311 dev
->features
|= dev
->hw_features
& (NETIF_F_ALL_TSO
|NETIF_F_UFO
);
2312 /* (!csum && gso) case will be fixed by register_netdev() */
2314 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_CSUM
))
2315 dev
->features
|= NETIF_F_RXCSUM
;
2317 dev
->vlan_features
= dev
->features
;
2319 /* MTU range: 68 - 65535 */
2320 dev
->min_mtu
= MIN_MTU
;
2321 dev
->max_mtu
= MAX_MTU
;
2323 /* Configuration may specify what MAC to use. Otherwise random. */
2324 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MAC
))
2325 virtio_cread_bytes(vdev
,
2326 offsetof(struct virtio_net_config
, mac
),
2327 dev
->dev_addr
, dev
->addr_len
);
2329 eth_hw_addr_random(dev
);
2331 /* Set up our device-specific information */
2332 vi
= netdev_priv(dev
);
2336 vi
->stats
= alloc_percpu(struct virtnet_stats
);
2338 if (vi
->stats
== NULL
)
2341 for_each_possible_cpu(i
) {
2342 struct virtnet_stats
*virtnet_stats
;
2343 virtnet_stats
= per_cpu_ptr(vi
->stats
, i
);
2344 u64_stats_init(&virtnet_stats
->tx_syncp
);
2345 u64_stats_init(&virtnet_stats
->rx_syncp
);
2348 INIT_WORK(&vi
->config_work
, virtnet_config_changed_work
);
2350 /* If we can receive ANY GSO packets, we must allocate large ones. */
2351 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_TSO4
) ||
2352 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_TSO6
) ||
2353 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_ECN
) ||
2354 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_UFO
))
2355 vi
->big_packets
= true;
2357 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MRG_RXBUF
))
2358 vi
->mergeable_rx_bufs
= true;
2360 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MRG_RXBUF
) ||
2361 virtio_has_feature(vdev
, VIRTIO_F_VERSION_1
))
2362 vi
->hdr_len
= sizeof(struct virtio_net_hdr_mrg_rxbuf
);
2364 vi
->hdr_len
= sizeof(struct virtio_net_hdr
);
2366 if (virtio_has_feature(vdev
, VIRTIO_F_ANY_LAYOUT
) ||
2367 virtio_has_feature(vdev
, VIRTIO_F_VERSION_1
))
2368 vi
->any_header_sg
= true;
2370 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
))
2373 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MTU
)) {
2374 mtu
= virtio_cread16(vdev
,
2375 offsetof(struct virtio_net_config
,
2377 if (mtu
< dev
->min_mtu
) {
2378 /* Should never trigger: MTU was previously validated
2379 * in virtnet_validate.
2381 dev_err(&vdev
->dev
, "device MTU appears to have changed "
2382 "it is now %d < %d", mtu
, dev
->min_mtu
);
2389 /* TODO: size buffers correctly in this case. */
2390 if (dev
->mtu
> ETH_DATA_LEN
)
2391 vi
->big_packets
= true;
2394 if (vi
->any_header_sg
)
2395 dev
->needed_headroom
= vi
->hdr_len
;
2397 /* Enable multiqueue by default */
2398 if (num_online_cpus() >= max_queue_pairs
)
2399 vi
->curr_queue_pairs
= max_queue_pairs
;
2401 vi
->curr_queue_pairs
= num_online_cpus();
2402 vi
->max_queue_pairs
= max_queue_pairs
;
2404 /* Allocate/initialize the rx/tx queues, and invoke find_vqs */
2410 if (vi
->mergeable_rx_bufs
)
2411 dev
->sysfs_rx_queue_group
= &virtio_net_mrg_rx_group
;
2413 netif_set_real_num_tx_queues(dev
, vi
->curr_queue_pairs
);
2414 netif_set_real_num_rx_queues(dev
, vi
->curr_queue_pairs
);
2416 virtnet_init_settings(dev
);
2418 err
= register_netdev(dev
);
2420 pr_debug("virtio_net: registering device failed\n");
2424 virtio_device_ready(vdev
);
2426 err
= virtnet_cpu_notif_add(vi
);
2428 pr_debug("virtio_net: registering cpu notifier failed\n");
2429 goto free_unregister_netdev
;
2432 virtnet_set_queues(vi
, vi
->curr_queue_pairs
);
2434 /* Assume link up if device can't report link status,
2435 otherwise get link status from config. */
2436 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_STATUS
)) {
2437 netif_carrier_off(dev
);
2438 schedule_work(&vi
->config_work
);
2440 vi
->status
= VIRTIO_NET_S_LINK_UP
;
2441 netif_carrier_on(dev
);
2444 pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
2445 dev
->name
, max_queue_pairs
);
2449 free_unregister_netdev
:
2450 vi
->vdev
->config
->reset(vdev
);
2452 unregister_netdev(dev
);
2454 cancel_delayed_work_sync(&vi
->refill
);
2455 free_receive_page_frags(vi
);
2456 virtnet_del_vqs(vi
);
2458 free_percpu(vi
->stats
);
2464 static void _remove_vq_common(struct virtnet_info
*vi
)
2466 vi
->vdev
->config
->reset(vi
->vdev
);
2467 free_unused_bufs(vi
);
2468 _free_receive_bufs(vi
);
2469 free_receive_page_frags(vi
);
2470 virtnet_del_vqs(vi
);
2473 static void remove_vq_common(struct virtnet_info
*vi
)
2475 vi
->vdev
->config
->reset(vi
->vdev
);
2477 /* Free unused buffers in both send and recv, if any. */
2478 free_unused_bufs(vi
);
2480 free_receive_bufs(vi
);
2482 free_receive_page_frags(vi
);
2484 virtnet_del_vqs(vi
);
2487 static void virtnet_remove(struct virtio_device
*vdev
)
2489 struct virtnet_info
*vi
= vdev
->priv
;
2491 virtnet_cpu_notif_remove(vi
);
2493 /* Make sure no work handler is accessing the device. */
2494 flush_work(&vi
->config_work
);
2496 unregister_netdev(vi
->dev
);
2498 remove_vq_common(vi
);
2500 free_percpu(vi
->stats
);
2501 free_netdev(vi
->dev
);
2504 #ifdef CONFIG_PM_SLEEP
2505 static int virtnet_freeze(struct virtio_device
*vdev
)
2507 struct virtnet_info
*vi
= vdev
->priv
;
2509 virtnet_cpu_notif_remove(vi
);
2510 virtnet_freeze_down(vdev
);
2511 remove_vq_common(vi
);
2516 static int virtnet_restore(struct virtio_device
*vdev
)
2518 struct virtnet_info
*vi
= vdev
->priv
;
2521 err
= virtnet_restore_up(vdev
);
2524 virtnet_set_queues(vi
, vi
->curr_queue_pairs
);
2526 err
= virtnet_cpu_notif_add(vi
);
2534 static struct virtio_device_id id_table
[] = {
2535 { VIRTIO_ID_NET
, VIRTIO_DEV_ANY_ID
},
2539 #define VIRTNET_FEATURES \
2540 VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, \
2542 VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, \
2543 VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, \
2544 VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, \
2545 VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, \
2546 VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, \
2547 VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ, \
2548 VIRTIO_NET_F_CTRL_MAC_ADDR, \
2551 static unsigned int features
[] = {
2555 static unsigned int features_legacy
[] = {
2558 VIRTIO_F_ANY_LAYOUT
,
2561 static struct virtio_driver virtio_net_driver
= {
2562 .feature_table
= features
,
2563 .feature_table_size
= ARRAY_SIZE(features
),
2564 .feature_table_legacy
= features_legacy
,
2565 .feature_table_size_legacy
= ARRAY_SIZE(features_legacy
),
2566 .driver
.name
= KBUILD_MODNAME
,
2567 .driver
.owner
= THIS_MODULE
,
2568 .id_table
= id_table
,
2569 .validate
= virtnet_validate
,
2570 .probe
= virtnet_probe
,
2571 .remove
= virtnet_remove
,
2572 .config_changed
= virtnet_config_changed
,
2573 #ifdef CONFIG_PM_SLEEP
2574 .freeze
= virtnet_freeze
,
2575 .restore
= virtnet_restore
,
2579 static __init
int virtio_net_driver_init(void)
2583 ret
= cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN
, "virtio/net:online",
2585 virtnet_cpu_down_prep
);
2588 virtionet_online
= ret
;
2589 ret
= cpuhp_setup_state_multi(CPUHP_VIRT_NET_DEAD
, "virtio/net:dead",
2590 NULL
, virtnet_cpu_dead
);
2594 ret
= register_virtio_driver(&virtio_net_driver
);
2599 cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD
);
2601 cpuhp_remove_multi_state(virtionet_online
);
2605 module_init(virtio_net_driver_init
);
2607 static __exit
void virtio_net_driver_exit(void)
2609 cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD
);
2610 cpuhp_remove_multi_state(virtionet_online
);
2611 unregister_virtio_driver(&virtio_net_driver
);
2613 module_exit(virtio_net_driver_exit
);
2615 MODULE_DEVICE_TABLE(virtio
, id_table
);
2616 MODULE_DESCRIPTION("Virtio network driver");
2617 MODULE_LICENSE("GPL");