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>
32 #include <linux/filter.h>
33 #include <linux/netdevice.h>
34 #include <linux/pci.h>
35 #include <net/route.h>
37 #include <net/net_failover.h>
39 static int napi_weight
= NAPI_POLL_WEIGHT
;
40 module_param(napi_weight
, int, 0444);
42 static bool csum
= true, gso
= true, napi_tx
;
43 module_param(csum
, bool, 0444);
44 module_param(gso
, bool, 0444);
45 module_param(napi_tx
, bool, 0644);
47 /* FIXME: MTU in config. */
48 #define GOOD_PACKET_LEN (ETH_HLEN + VLAN_HLEN + ETH_DATA_LEN)
49 #define GOOD_COPY_LEN 128
51 #define VIRTNET_RX_PAD (NET_IP_ALIGN + NET_SKB_PAD)
53 /* Amount of XDP headroom to prepend to packets for use by xdp_adjust_head */
54 #define VIRTIO_XDP_HEADROOM 256
56 /* RX packet size EWMA. The average packet size is used to determine the packet
57 * buffer size when refilling RX rings. As the entire RX ring may be refilled
58 * at once, the weight is chosen so that the EWMA will be insensitive to short-
59 * term, transient changes in packet size.
61 DECLARE_EWMA(pkt_len
, 0, 64)
63 #define VIRTNET_DRIVER_VERSION "1.0.0"
65 static const unsigned long guest_offloads
[] = {
66 VIRTIO_NET_F_GUEST_TSO4
,
67 VIRTIO_NET_F_GUEST_TSO6
,
68 VIRTIO_NET_F_GUEST_ECN
,
69 VIRTIO_NET_F_GUEST_UFO
72 struct virtnet_stat_desc
{
73 char desc
[ETH_GSTRING_LEN
];
77 struct virtnet_sq_stats
{
78 struct u64_stats_sync syncp
;
83 struct virtnet_rq_stats
{
84 struct u64_stats_sync syncp
;
89 #define VIRTNET_SQ_STAT(m) offsetof(struct virtnet_sq_stats, m)
90 #define VIRTNET_RQ_STAT(m) offsetof(struct virtnet_rq_stats, m)
92 static const struct virtnet_stat_desc virtnet_sq_stats_desc
[] = {
93 { "packets", VIRTNET_SQ_STAT(packets
) },
94 { "bytes", VIRTNET_SQ_STAT(bytes
) },
97 static const struct virtnet_stat_desc virtnet_rq_stats_desc
[] = {
98 { "packets", VIRTNET_RQ_STAT(packets
) },
99 { "bytes", VIRTNET_RQ_STAT(bytes
) },
102 #define VIRTNET_SQ_STATS_LEN ARRAY_SIZE(virtnet_sq_stats_desc)
103 #define VIRTNET_RQ_STATS_LEN ARRAY_SIZE(virtnet_rq_stats_desc)
105 /* Internal representation of a send virtqueue */
107 /* Virtqueue associated with this send _queue */
108 struct virtqueue
*vq
;
110 /* TX: fragments + linear part + virtio header */
111 struct scatterlist sg
[MAX_SKB_FRAGS
+ 2];
113 /* Name of the send queue: output.$index */
116 struct virtnet_sq_stats stats
;
118 struct napi_struct napi
;
121 /* Internal representation of a receive virtqueue */
122 struct receive_queue
{
123 /* Virtqueue associated with this receive_queue */
124 struct virtqueue
*vq
;
126 struct napi_struct napi
;
128 struct bpf_prog __rcu
*xdp_prog
;
130 struct virtnet_rq_stats stats
;
132 /* Chain pages by the private ptr. */
135 /* Average packet length for mergeable receive buffers. */
136 struct ewma_pkt_len mrg_avg_pkt_len
;
138 /* Page frag for packet buffer allocation. */
139 struct page_frag alloc_frag
;
141 /* RX: fragments + linear part + virtio header */
142 struct scatterlist sg
[MAX_SKB_FRAGS
+ 2];
144 /* Min single buffer size for mergeable buffers case. */
145 unsigned int min_buf_len
;
147 /* Name of this receive queue: input.$index */
150 struct xdp_rxq_info xdp_rxq
;
153 /* Control VQ buffers: protected by the rtnl lock */
155 struct virtio_net_ctrl_hdr hdr
;
156 virtio_net_ctrl_ack status
;
157 struct virtio_net_ctrl_mq mq
;
164 struct virtnet_info
{
165 struct virtio_device
*vdev
;
166 struct virtqueue
*cvq
;
167 struct net_device
*dev
;
168 struct send_queue
*sq
;
169 struct receive_queue
*rq
;
172 /* Max # of queue pairs supported by the device */
175 /* # of queue pairs currently used by the driver */
176 u16 curr_queue_pairs
;
178 /* # of XDP queue pairs currently used by the driver */
181 /* I like... big packets and I cannot lie! */
184 /* Host will merge rx buffers for big packets (shake it! shake it!) */
185 bool mergeable_rx_bufs
;
187 /* Has control virtqueue */
190 /* Host can handle any s/g split between our header and packet data */
193 /* Packet virtio header size */
196 /* Work struct for refilling if we run low on memory. */
197 struct delayed_work refill
;
199 /* Work struct for config space updates */
200 struct work_struct config_work
;
202 /* Does the affinity hint is set for virtqueues? */
203 bool affinity_hint_set
;
205 /* CPU hotplug instances for online & dead */
206 struct hlist_node node
;
207 struct hlist_node node_dead
;
209 struct control_buf
*ctrl
;
211 /* Ethtool settings */
215 unsigned long guest_offloads
;
217 /* failover when STANDBY feature enabled */
218 struct failover
*failover
;
221 struct padded_vnet_hdr
{
222 struct virtio_net_hdr_mrg_rxbuf hdr
;
224 * hdr is in a separate sg buffer, and data sg buffer shares same page
225 * with this header sg. This padding makes next sg 16 byte aligned
231 /* Converting between virtqueue no. and kernel tx/rx queue no.
232 * 0:rx0 1:tx0 2:rx1 3:tx1 ... 2N:rxN 2N+1:txN 2N+2:cvq
234 static int vq2txq(struct virtqueue
*vq
)
236 return (vq
->index
- 1) / 2;
239 static int txq2vq(int txq
)
244 static int vq2rxq(struct virtqueue
*vq
)
246 return vq
->index
/ 2;
249 static int rxq2vq(int rxq
)
254 static inline struct virtio_net_hdr_mrg_rxbuf
*skb_vnet_hdr(struct sk_buff
*skb
)
256 return (struct virtio_net_hdr_mrg_rxbuf
*)skb
->cb
;
260 * private is used to chain pages for big packets, put the whole
261 * most recent used list in the beginning for reuse
263 static void give_pages(struct receive_queue
*rq
, struct page
*page
)
267 /* Find end of list, sew whole thing into vi->rq.pages. */
268 for (end
= page
; end
->private; end
= (struct page
*)end
->private);
269 end
->private = (unsigned long)rq
->pages
;
273 static struct page
*get_a_page(struct receive_queue
*rq
, gfp_t gfp_mask
)
275 struct page
*p
= rq
->pages
;
278 rq
->pages
= (struct page
*)p
->private;
279 /* clear private here, it is used to chain pages */
282 p
= alloc_page(gfp_mask
);
286 static void virtqueue_napi_schedule(struct napi_struct
*napi
,
287 struct virtqueue
*vq
)
289 if (napi_schedule_prep(napi
)) {
290 virtqueue_disable_cb(vq
);
291 __napi_schedule(napi
);
295 static void virtqueue_napi_complete(struct napi_struct
*napi
,
296 struct virtqueue
*vq
, int processed
)
300 opaque
= virtqueue_enable_cb_prepare(vq
);
301 if (napi_complete_done(napi
, processed
)) {
302 if (unlikely(virtqueue_poll(vq
, opaque
)))
303 virtqueue_napi_schedule(napi
, vq
);
305 virtqueue_disable_cb(vq
);
309 static void skb_xmit_done(struct virtqueue
*vq
)
311 struct virtnet_info
*vi
= vq
->vdev
->priv
;
312 struct napi_struct
*napi
= &vi
->sq
[vq2txq(vq
)].napi
;
314 /* Suppress further interrupts. */
315 virtqueue_disable_cb(vq
);
318 virtqueue_napi_schedule(napi
, vq
);
320 /* We were probably waiting for more output buffers. */
321 netif_wake_subqueue(vi
->dev
, vq2txq(vq
));
324 #define MRG_CTX_HEADER_SHIFT 22
325 static void *mergeable_len_to_ctx(unsigned int truesize
,
326 unsigned int headroom
)
328 return (void *)(unsigned long)((headroom
<< MRG_CTX_HEADER_SHIFT
) | truesize
);
331 static unsigned int mergeable_ctx_to_headroom(void *mrg_ctx
)
333 return (unsigned long)mrg_ctx
>> MRG_CTX_HEADER_SHIFT
;
336 static unsigned int mergeable_ctx_to_truesize(void *mrg_ctx
)
338 return (unsigned long)mrg_ctx
& ((1 << MRG_CTX_HEADER_SHIFT
) - 1);
341 /* Called from bottom half context */
342 static struct sk_buff
*page_to_skb(struct virtnet_info
*vi
,
343 struct receive_queue
*rq
,
344 struct page
*page
, unsigned int offset
,
345 unsigned int len
, unsigned int truesize
)
348 struct virtio_net_hdr_mrg_rxbuf
*hdr
;
349 unsigned int copy
, hdr_len
, hdr_padded_len
;
352 p
= page_address(page
) + offset
;
354 /* copy small packet so we can reuse these pages for small data */
355 skb
= napi_alloc_skb(&rq
->napi
, GOOD_COPY_LEN
);
359 hdr
= skb_vnet_hdr(skb
);
361 hdr_len
= vi
->hdr_len
;
362 if (vi
->mergeable_rx_bufs
)
363 hdr_padded_len
= sizeof(*hdr
);
365 hdr_padded_len
= sizeof(struct padded_vnet_hdr
);
367 memcpy(hdr
, p
, hdr_len
);
370 offset
+= hdr_padded_len
;
374 if (copy
> skb_tailroom(skb
))
375 copy
= skb_tailroom(skb
);
376 skb_put_data(skb
, p
, copy
);
381 if (vi
->mergeable_rx_bufs
) {
383 skb_add_rx_frag(skb
, 0, page
, offset
, len
, truesize
);
390 * Verify that we can indeed put this data into a skb.
391 * This is here to handle cases when the device erroneously
392 * tries to receive more than is possible. This is usually
393 * the case of a broken device.
395 if (unlikely(len
> MAX_SKB_FRAGS
* PAGE_SIZE
)) {
396 net_dbg_ratelimited("%s: too much data\n", skb
->dev
->name
);
400 BUG_ON(offset
>= PAGE_SIZE
);
402 unsigned int frag_size
= min((unsigned)PAGE_SIZE
- offset
, len
);
403 skb_add_rx_frag(skb
, skb_shinfo(skb
)->nr_frags
, page
, offset
,
404 frag_size
, truesize
);
406 page
= (struct page
*)page
->private;
411 give_pages(rq
, page
);
416 static int __virtnet_xdp_xmit_one(struct virtnet_info
*vi
,
417 struct send_queue
*sq
,
418 struct xdp_frame
*xdpf
)
420 struct virtio_net_hdr_mrg_rxbuf
*hdr
;
423 /* virtqueue want to use data area in-front of packet */
424 if (unlikely(xdpf
->metasize
> 0))
427 if (unlikely(xdpf
->headroom
< vi
->hdr_len
))
430 /* Make room for virtqueue hdr (also change xdpf->headroom?) */
431 xdpf
->data
-= vi
->hdr_len
;
432 /* Zero header and leave csum up to XDP layers */
434 memset(hdr
, 0, vi
->hdr_len
);
435 xdpf
->len
+= vi
->hdr_len
;
437 sg_init_one(sq
->sg
, xdpf
->data
, xdpf
->len
);
439 err
= virtqueue_add_outbuf(sq
->vq
, sq
->sg
, 1, xdpf
, GFP_ATOMIC
);
441 return -ENOSPC
; /* Caller handle free/refcnt */
446 static int __virtnet_xdp_tx_xmit(struct virtnet_info
*vi
,
447 struct xdp_frame
*xdpf
)
449 struct xdp_frame
*xdpf_sent
;
450 struct send_queue
*sq
;
454 qp
= vi
->curr_queue_pairs
- vi
->xdp_queue_pairs
+ smp_processor_id();
457 /* Free up any pending old buffers before queueing new ones. */
458 while ((xdpf_sent
= virtqueue_get_buf(sq
->vq
, &len
)) != NULL
)
459 xdp_return_frame(xdpf_sent
);
461 return __virtnet_xdp_xmit_one(vi
, sq
, xdpf
);
464 static int virtnet_xdp_xmit(struct net_device
*dev
,
465 int n
, struct xdp_frame
**frames
, u32 flags
)
467 struct virtnet_info
*vi
= netdev_priv(dev
);
468 struct receive_queue
*rq
= vi
->rq
;
469 struct xdp_frame
*xdpf_sent
;
470 struct bpf_prog
*xdp_prog
;
471 struct send_queue
*sq
;
478 if (unlikely(flags
& ~XDP_XMIT_FLAGS_MASK
))
481 qp
= vi
->curr_queue_pairs
- vi
->xdp_queue_pairs
+ smp_processor_id();
484 /* Only allow ndo_xdp_xmit if XDP is loaded on dev, as this
485 * indicate XDP resources have been successfully allocated.
487 xdp_prog
= rcu_dereference(rq
->xdp_prog
);
491 /* Free up any pending old buffers before queueing new ones. */
492 while ((xdpf_sent
= virtqueue_get_buf(sq
->vq
, &len
)) != NULL
)
493 xdp_return_frame(xdpf_sent
);
495 for (i
= 0; i
< n
; i
++) {
496 struct xdp_frame
*xdpf
= frames
[i
];
498 err
= __virtnet_xdp_xmit_one(vi
, sq
, xdpf
);
500 xdp_return_frame_rx_napi(xdpf
);
505 if (flags
& XDP_XMIT_FLUSH
)
506 virtqueue_kick(sq
->vq
);
511 static unsigned int virtnet_get_headroom(struct virtnet_info
*vi
)
513 return vi
->xdp_queue_pairs
? VIRTIO_XDP_HEADROOM
: 0;
516 /* We copy the packet for XDP in the following cases:
518 * 1) Packet is scattered across multiple rx buffers.
519 * 2) Headroom space is insufficient.
521 * This is inefficient but it's a temporary condition that
522 * we hit right after XDP is enabled and until queue is refilled
523 * with large buffers with sufficient headroom - so it should affect
524 * at most queue size packets.
525 * Afterwards, the conditions to enable
526 * XDP should preclude the underlying device from sending packets
527 * across multiple buffers (num_buf > 1), and we make sure buffers
528 * have enough headroom.
530 static struct page
*xdp_linearize_page(struct receive_queue
*rq
,
537 struct page
*page
= alloc_page(GFP_ATOMIC
);
542 memcpy(page_address(page
) + page_off
, page_address(p
) + offset
, *len
);
546 int tailroom
= SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
551 buf
= virtqueue_get_buf(rq
->vq
, &buflen
);
555 p
= virt_to_head_page(buf
);
556 off
= buf
- page_address(p
);
558 /* guard against a misconfigured or uncooperative backend that
559 * is sending packet larger than the MTU.
561 if ((page_off
+ buflen
+ tailroom
) > PAGE_SIZE
) {
566 memcpy(page_address(page
) + page_off
,
567 page_address(p
) + off
, buflen
);
572 /* Headroom does not contribute to packet length */
573 *len
= page_off
- VIRTIO_XDP_HEADROOM
;
576 __free_pages(page
, 0);
580 static struct sk_buff
*receive_small(struct net_device
*dev
,
581 struct virtnet_info
*vi
,
582 struct receive_queue
*rq
,
583 void *buf
, void *ctx
,
588 struct bpf_prog
*xdp_prog
;
589 unsigned int xdp_headroom
= (unsigned long)ctx
;
590 unsigned int header_offset
= VIRTNET_RX_PAD
+ xdp_headroom
;
591 unsigned int headroom
= vi
->hdr_len
+ header_offset
;
592 unsigned int buflen
= SKB_DATA_ALIGN(GOOD_PACKET_LEN
+ headroom
) +
593 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
594 struct page
*page
= virt_to_head_page(buf
);
595 unsigned int delta
= 0;
596 struct page
*xdp_page
;
602 xdp_prog
= rcu_dereference(rq
->xdp_prog
);
604 struct virtio_net_hdr_mrg_rxbuf
*hdr
= buf
+ header_offset
;
605 struct xdp_frame
*xdpf
;
610 if (unlikely(hdr
->hdr
.gso_type
))
613 if (unlikely(xdp_headroom
< virtnet_get_headroom(vi
))) {
614 int offset
= buf
- page_address(page
) + header_offset
;
615 unsigned int tlen
= len
+ vi
->hdr_len
;
618 xdp_headroom
= virtnet_get_headroom(vi
);
619 header_offset
= VIRTNET_RX_PAD
+ xdp_headroom
;
620 headroom
= vi
->hdr_len
+ header_offset
;
621 buflen
= SKB_DATA_ALIGN(GOOD_PACKET_LEN
+ headroom
) +
622 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
623 xdp_page
= xdp_linearize_page(rq
, &num_buf
, page
,
624 offset
, header_offset
,
629 buf
= page_address(xdp_page
);
634 xdp
.data_hard_start
= buf
+ VIRTNET_RX_PAD
+ vi
->hdr_len
;
635 xdp
.data
= xdp
.data_hard_start
+ xdp_headroom
;
636 xdp_set_data_meta_invalid(&xdp
);
637 xdp
.data_end
= xdp
.data
+ len
;
638 xdp
.rxq
= &rq
->xdp_rxq
;
639 orig_data
= xdp
.data
;
640 act
= bpf_prog_run_xdp(xdp_prog
, &xdp
);
644 /* Recalculate length in case bpf program changed it */
645 delta
= orig_data
- xdp
.data
;
646 len
= xdp
.data_end
- xdp
.data
;
649 xdpf
= convert_to_xdp_frame(&xdp
);
652 err
= __virtnet_xdp_tx_xmit(vi
, xdpf
);
654 trace_xdp_exception(vi
->dev
, xdp_prog
, act
);
661 err
= xdp_do_redirect(dev
, &xdp
, xdp_prog
);
668 bpf_warn_invalid_xdp_action(act
);
670 trace_xdp_exception(vi
->dev
, xdp_prog
, act
);
677 skb
= build_skb(buf
, buflen
);
682 skb_reserve(skb
, headroom
- delta
);
685 buf
+= header_offset
;
686 memcpy(skb_vnet_hdr(skb
), buf
, vi
->hdr_len
);
687 } /* keep zeroed vnet hdr since packet was changed by bpf */
694 dev
->stats
.rx_dropped
++;
700 static struct sk_buff
*receive_big(struct net_device
*dev
,
701 struct virtnet_info
*vi
,
702 struct receive_queue
*rq
,
706 struct page
*page
= buf
;
707 struct sk_buff
*skb
= page_to_skb(vi
, rq
, page
, 0, len
, PAGE_SIZE
);
715 dev
->stats
.rx_dropped
++;
716 give_pages(rq
, page
);
720 static struct sk_buff
*receive_mergeable(struct net_device
*dev
,
721 struct virtnet_info
*vi
,
722 struct receive_queue
*rq
,
728 struct virtio_net_hdr_mrg_rxbuf
*hdr
= buf
;
729 u16 num_buf
= virtio16_to_cpu(vi
->vdev
, hdr
->num_buffers
);
730 struct page
*page
= virt_to_head_page(buf
);
731 int offset
= buf
- page_address(page
);
732 struct sk_buff
*head_skb
, *curr_skb
;
733 struct bpf_prog
*xdp_prog
;
734 unsigned int truesize
;
735 unsigned int headroom
= mergeable_ctx_to_headroom(ctx
);
741 xdp_prog
= rcu_dereference(rq
->xdp_prog
);
743 struct xdp_frame
*xdpf
;
744 struct page
*xdp_page
;
749 /* Transient failure which in theory could occur if
750 * in-flight packets from before XDP was enabled reach
751 * the receive path after XDP is loaded.
753 if (unlikely(hdr
->hdr
.gso_type
))
756 /* This happens when rx buffer size is underestimated
757 * or headroom is not enough because of the buffer
758 * was refilled before XDP is set. This should only
759 * happen for the first several packets, so we don't
760 * care much about its performance.
762 if (unlikely(num_buf
> 1 ||
763 headroom
< virtnet_get_headroom(vi
))) {
764 /* linearize data for XDP */
765 xdp_page
= xdp_linearize_page(rq
, &num_buf
,
771 offset
= VIRTIO_XDP_HEADROOM
;
776 /* Allow consuming headroom but reserve enough space to push
777 * the descriptor on if we get an XDP_TX return code.
779 data
= page_address(xdp_page
) + offset
;
780 xdp
.data_hard_start
= data
- VIRTIO_XDP_HEADROOM
+ vi
->hdr_len
;
781 xdp
.data
= data
+ vi
->hdr_len
;
782 xdp_set_data_meta_invalid(&xdp
);
783 xdp
.data_end
= xdp
.data
+ (len
- vi
->hdr_len
);
784 xdp
.rxq
= &rq
->xdp_rxq
;
786 act
= bpf_prog_run_xdp(xdp_prog
, &xdp
);
790 /* recalculate offset to account for any header
791 * adjustments. Note other cases do not build an
792 * skb and avoid using offset
795 page_address(xdp_page
) - vi
->hdr_len
;
797 /* recalculate len if xdp.data or xdp.data_end were
800 len
= xdp
.data_end
- xdp
.data
+ vi
->hdr_len
;
801 /* We can only create skb based on xdp_page. */
802 if (unlikely(xdp_page
!= page
)) {
805 head_skb
= page_to_skb(vi
, rq
, xdp_page
,
806 offset
, len
, PAGE_SIZE
);
811 xdpf
= convert_to_xdp_frame(&xdp
);
814 err
= __virtnet_xdp_tx_xmit(vi
, xdpf
);
816 trace_xdp_exception(vi
->dev
, xdp_prog
, act
);
817 if (unlikely(xdp_page
!= page
))
822 if (unlikely(xdp_page
!= page
))
827 err
= xdp_do_redirect(dev
, &xdp
, xdp_prog
);
829 if (unlikely(xdp_page
!= page
))
834 if (unlikely(xdp_page
!= page
))
839 bpf_warn_invalid_xdp_action(act
);
841 trace_xdp_exception(vi
->dev
, xdp_prog
, act
);
843 if (unlikely(xdp_page
!= page
))
844 __free_pages(xdp_page
, 0);
850 truesize
= mergeable_ctx_to_truesize(ctx
);
851 if (unlikely(len
> truesize
)) {
852 pr_debug("%s: rx error: len %u exceeds truesize %lu\n",
853 dev
->name
, len
, (unsigned long)ctx
);
854 dev
->stats
.rx_length_errors
++;
858 head_skb
= page_to_skb(vi
, rq
, page
, offset
, len
, truesize
);
861 if (unlikely(!curr_skb
))
866 buf
= virtqueue_get_buf_ctx(rq
->vq
, &len
, &ctx
);
867 if (unlikely(!buf
)) {
868 pr_debug("%s: rx error: %d buffers out of %d missing\n",
870 virtio16_to_cpu(vi
->vdev
,
872 dev
->stats
.rx_length_errors
++;
876 page
= virt_to_head_page(buf
);
878 truesize
= mergeable_ctx_to_truesize(ctx
);
879 if (unlikely(len
> truesize
)) {
880 pr_debug("%s: rx error: len %u exceeds truesize %lu\n",
881 dev
->name
, len
, (unsigned long)ctx
);
882 dev
->stats
.rx_length_errors
++;
886 num_skb_frags
= skb_shinfo(curr_skb
)->nr_frags
;
887 if (unlikely(num_skb_frags
== MAX_SKB_FRAGS
)) {
888 struct sk_buff
*nskb
= alloc_skb(0, GFP_ATOMIC
);
892 if (curr_skb
== head_skb
)
893 skb_shinfo(curr_skb
)->frag_list
= nskb
;
895 curr_skb
->next
= nskb
;
897 head_skb
->truesize
+= nskb
->truesize
;
900 if (curr_skb
!= head_skb
) {
901 head_skb
->data_len
+= len
;
902 head_skb
->len
+= len
;
903 head_skb
->truesize
+= truesize
;
905 offset
= buf
- page_address(page
);
906 if (skb_can_coalesce(curr_skb
, num_skb_frags
, page
, offset
)) {
908 skb_coalesce_rx_frag(curr_skb
, num_skb_frags
- 1,
911 skb_add_rx_frag(curr_skb
, num_skb_frags
, page
,
912 offset
, len
, truesize
);
916 ewma_pkt_len_add(&rq
->mrg_avg_pkt_len
, head_skb
->len
);
923 while (num_buf
-- > 1) {
924 buf
= virtqueue_get_buf(rq
->vq
, &len
);
925 if (unlikely(!buf
)) {
926 pr_debug("%s: rx error: %d buffers missing\n",
928 dev
->stats
.rx_length_errors
++;
931 page
= virt_to_head_page(buf
);
935 dev
->stats
.rx_dropped
++;
936 dev_kfree_skb(head_skb
);
941 static int receive_buf(struct virtnet_info
*vi
, struct receive_queue
*rq
,
942 void *buf
, unsigned int len
, void **ctx
, bool *xdp_xmit
)
944 struct net_device
*dev
= vi
->dev
;
946 struct virtio_net_hdr_mrg_rxbuf
*hdr
;
949 if (unlikely(len
< vi
->hdr_len
+ ETH_HLEN
)) {
950 pr_debug("%s: short packet %i\n", dev
->name
, len
);
951 dev
->stats
.rx_length_errors
++;
952 if (vi
->mergeable_rx_bufs
) {
953 put_page(virt_to_head_page(buf
));
954 } else if (vi
->big_packets
) {
957 put_page(virt_to_head_page(buf
));
962 if (vi
->mergeable_rx_bufs
)
963 skb
= receive_mergeable(dev
, vi
, rq
, buf
, ctx
, len
, xdp_xmit
);
964 else if (vi
->big_packets
)
965 skb
= receive_big(dev
, vi
, rq
, buf
, len
);
967 skb
= receive_small(dev
, vi
, rq
, buf
, ctx
, len
, xdp_xmit
);
972 hdr
= skb_vnet_hdr(skb
);
976 if (hdr
->hdr
.flags
& VIRTIO_NET_HDR_F_DATA_VALID
)
977 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
979 if (virtio_net_hdr_to_skb(skb
, &hdr
->hdr
,
980 virtio_is_little_endian(vi
->vdev
))) {
981 net_warn_ratelimited("%s: bad gso: type: %u, size: %u\n",
982 dev
->name
, hdr
->hdr
.gso_type
,
987 skb
->protocol
= eth_type_trans(skb
, dev
);
988 pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
989 ntohs(skb
->protocol
), skb
->len
, skb
->pkt_type
);
991 napi_gro_receive(&rq
->napi
, skb
);
995 dev
->stats
.rx_frame_errors
++;
1000 /* Unlike mergeable buffers, all buffers are allocated to the
1001 * same size, except for the headroom. For this reason we do
1002 * not need to use mergeable_len_to_ctx here - it is enough
1003 * to store the headroom as the context ignoring the truesize.
1005 static int add_recvbuf_small(struct virtnet_info
*vi
, struct receive_queue
*rq
,
1008 struct page_frag
*alloc_frag
= &rq
->alloc_frag
;
1010 unsigned int xdp_headroom
= virtnet_get_headroom(vi
);
1011 void *ctx
= (void *)(unsigned long)xdp_headroom
;
1012 int len
= vi
->hdr_len
+ VIRTNET_RX_PAD
+ GOOD_PACKET_LEN
+ xdp_headroom
;
1015 len
= SKB_DATA_ALIGN(len
) +
1016 SKB_DATA_ALIGN(sizeof(struct skb_shared_info
));
1017 if (unlikely(!skb_page_frag_refill(len
, alloc_frag
, gfp
)))
1020 buf
= (char *)page_address(alloc_frag
->page
) + alloc_frag
->offset
;
1021 get_page(alloc_frag
->page
);
1022 alloc_frag
->offset
+= len
;
1023 sg_init_one(rq
->sg
, buf
+ VIRTNET_RX_PAD
+ xdp_headroom
,
1024 vi
->hdr_len
+ GOOD_PACKET_LEN
);
1025 err
= virtqueue_add_inbuf_ctx(rq
->vq
, rq
->sg
, 1, buf
, ctx
, gfp
);
1027 put_page(virt_to_head_page(buf
));
1031 static int add_recvbuf_big(struct virtnet_info
*vi
, struct receive_queue
*rq
,
1034 struct page
*first
, *list
= NULL
;
1038 sg_init_table(rq
->sg
, MAX_SKB_FRAGS
+ 2);
1040 /* page in rq->sg[MAX_SKB_FRAGS + 1] is list tail */
1041 for (i
= MAX_SKB_FRAGS
+ 1; i
> 1; --i
) {
1042 first
= get_a_page(rq
, gfp
);
1045 give_pages(rq
, list
);
1048 sg_set_buf(&rq
->sg
[i
], page_address(first
), PAGE_SIZE
);
1050 /* chain new page in list head to match sg */
1051 first
->private = (unsigned long)list
;
1055 first
= get_a_page(rq
, gfp
);
1057 give_pages(rq
, list
);
1060 p
= page_address(first
);
1062 /* rq->sg[0], rq->sg[1] share the same page */
1063 /* a separated rq->sg[0] for header - required in case !any_header_sg */
1064 sg_set_buf(&rq
->sg
[0], p
, vi
->hdr_len
);
1066 /* rq->sg[1] for data packet, from offset */
1067 offset
= sizeof(struct padded_vnet_hdr
);
1068 sg_set_buf(&rq
->sg
[1], p
+ offset
, PAGE_SIZE
- offset
);
1070 /* chain first in list head */
1071 first
->private = (unsigned long)list
;
1072 err
= virtqueue_add_inbuf(rq
->vq
, rq
->sg
, MAX_SKB_FRAGS
+ 2,
1075 give_pages(rq
, first
);
1080 static unsigned int get_mergeable_buf_len(struct receive_queue
*rq
,
1081 struct ewma_pkt_len
*avg_pkt_len
,
1084 const size_t hdr_len
= sizeof(struct virtio_net_hdr_mrg_rxbuf
);
1088 return PAGE_SIZE
- room
;
1090 len
= hdr_len
+ clamp_t(unsigned int, ewma_pkt_len_read(avg_pkt_len
),
1091 rq
->min_buf_len
, PAGE_SIZE
- hdr_len
);
1093 return ALIGN(len
, L1_CACHE_BYTES
);
1096 static int add_recvbuf_mergeable(struct virtnet_info
*vi
,
1097 struct receive_queue
*rq
, gfp_t gfp
)
1099 struct page_frag
*alloc_frag
= &rq
->alloc_frag
;
1100 unsigned int headroom
= virtnet_get_headroom(vi
);
1101 unsigned int tailroom
= headroom
? sizeof(struct skb_shared_info
) : 0;
1102 unsigned int room
= SKB_DATA_ALIGN(headroom
+ tailroom
);
1106 unsigned int len
, hole
;
1108 /* Extra tailroom is needed to satisfy XDP's assumption. This
1109 * means rx frags coalescing won't work, but consider we've
1110 * disabled GSO for XDP, it won't be a big issue.
1112 len
= get_mergeable_buf_len(rq
, &rq
->mrg_avg_pkt_len
, room
);
1113 if (unlikely(!skb_page_frag_refill(len
+ room
, alloc_frag
, gfp
)))
1116 buf
= (char *)page_address(alloc_frag
->page
) + alloc_frag
->offset
;
1117 buf
+= headroom
; /* advance address leaving hole at front of pkt */
1118 get_page(alloc_frag
->page
);
1119 alloc_frag
->offset
+= len
+ room
;
1120 hole
= alloc_frag
->size
- alloc_frag
->offset
;
1121 if (hole
< len
+ room
) {
1122 /* To avoid internal fragmentation, if there is very likely not
1123 * enough space for another buffer, add the remaining space to
1124 * the current buffer.
1127 alloc_frag
->offset
+= hole
;
1130 sg_init_one(rq
->sg
, buf
, len
);
1131 ctx
= mergeable_len_to_ctx(len
, headroom
);
1132 err
= virtqueue_add_inbuf_ctx(rq
->vq
, rq
->sg
, 1, buf
, ctx
, gfp
);
1134 put_page(virt_to_head_page(buf
));
1140 * Returns false if we couldn't fill entirely (OOM).
1142 * Normally run in the receive path, but can also be run from ndo_open
1143 * before we're receiving packets, or from refill_work which is
1144 * careful to disable receiving (using napi_disable).
1146 static bool try_fill_recv(struct virtnet_info
*vi
, struct receive_queue
*rq
,
1153 if (vi
->mergeable_rx_bufs
)
1154 err
= add_recvbuf_mergeable(vi
, rq
, gfp
);
1155 else if (vi
->big_packets
)
1156 err
= add_recvbuf_big(vi
, rq
, gfp
);
1158 err
= add_recvbuf_small(vi
, rq
, gfp
);
1160 oom
= err
== -ENOMEM
;
1163 } while (rq
->vq
->num_free
);
1164 virtqueue_kick(rq
->vq
);
1168 static void skb_recv_done(struct virtqueue
*rvq
)
1170 struct virtnet_info
*vi
= rvq
->vdev
->priv
;
1171 struct receive_queue
*rq
= &vi
->rq
[vq2rxq(rvq
)];
1173 virtqueue_napi_schedule(&rq
->napi
, rvq
);
1176 static void virtnet_napi_enable(struct virtqueue
*vq
, struct napi_struct
*napi
)
1180 /* If all buffers were filled by other side before we napi_enabled, we
1181 * won't get another interrupt, so process any outstanding packets now.
1182 * Call local_bh_enable after to trigger softIRQ processing.
1185 virtqueue_napi_schedule(napi
, vq
);
1189 static void virtnet_napi_tx_enable(struct virtnet_info
*vi
,
1190 struct virtqueue
*vq
,
1191 struct napi_struct
*napi
)
1196 /* Tx napi touches cachelines on the cpu handling tx interrupts. Only
1197 * enable the feature if this is likely affine with the transmit path.
1199 if (!vi
->affinity_hint_set
) {
1204 return virtnet_napi_enable(vq
, napi
);
1207 static void virtnet_napi_tx_disable(struct napi_struct
*napi
)
1213 static void refill_work(struct work_struct
*work
)
1215 struct virtnet_info
*vi
=
1216 container_of(work
, struct virtnet_info
, refill
.work
);
1220 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++) {
1221 struct receive_queue
*rq
= &vi
->rq
[i
];
1223 napi_disable(&rq
->napi
);
1224 still_empty
= !try_fill_recv(vi
, rq
, GFP_KERNEL
);
1225 virtnet_napi_enable(rq
->vq
, &rq
->napi
);
1227 /* In theory, this can happen: if we don't get any buffers in
1228 * we will *never* try to fill again.
1231 schedule_delayed_work(&vi
->refill
, HZ
/2);
1235 static int virtnet_receive(struct receive_queue
*rq
, int budget
, bool *xdp_xmit
)
1237 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
1238 unsigned int len
, received
= 0, bytes
= 0;
1241 if (!vi
->big_packets
|| vi
->mergeable_rx_bufs
) {
1244 while (received
< budget
&&
1245 (buf
= virtqueue_get_buf_ctx(rq
->vq
, &len
, &ctx
))) {
1246 bytes
+= receive_buf(vi
, rq
, buf
, len
, ctx
, xdp_xmit
);
1250 while (received
< budget
&&
1251 (buf
= virtqueue_get_buf(rq
->vq
, &len
)) != NULL
) {
1252 bytes
+= receive_buf(vi
, rq
, buf
, len
, NULL
, xdp_xmit
);
1257 if (rq
->vq
->num_free
> virtqueue_get_vring_size(rq
->vq
) / 2) {
1258 if (!try_fill_recv(vi
, rq
, GFP_ATOMIC
))
1259 schedule_delayed_work(&vi
->refill
, 0);
1262 u64_stats_update_begin(&rq
->stats
.syncp
);
1263 rq
->stats
.bytes
+= bytes
;
1264 rq
->stats
.packets
+= received
;
1265 u64_stats_update_end(&rq
->stats
.syncp
);
1270 static void free_old_xmit_skbs(struct send_queue
*sq
)
1272 struct sk_buff
*skb
;
1274 unsigned int packets
= 0;
1275 unsigned int bytes
= 0;
1277 while ((skb
= virtqueue_get_buf(sq
->vq
, &len
)) != NULL
) {
1278 pr_debug("Sent skb %p\n", skb
);
1283 dev_consume_skb_any(skb
);
1286 /* Avoid overhead when no packets have been processed
1287 * happens when called speculatively from start_xmit.
1292 u64_stats_update_begin(&sq
->stats
.syncp
);
1293 sq
->stats
.bytes
+= bytes
;
1294 sq
->stats
.packets
+= packets
;
1295 u64_stats_update_end(&sq
->stats
.syncp
);
1298 static void virtnet_poll_cleantx(struct receive_queue
*rq
)
1300 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
1301 unsigned int index
= vq2rxq(rq
->vq
);
1302 struct send_queue
*sq
= &vi
->sq
[index
];
1303 struct netdev_queue
*txq
= netdev_get_tx_queue(vi
->dev
, index
);
1305 if (!sq
->napi
.weight
)
1308 if (__netif_tx_trylock(txq
)) {
1309 free_old_xmit_skbs(sq
);
1310 __netif_tx_unlock(txq
);
1313 if (sq
->vq
->num_free
>= 2 + MAX_SKB_FRAGS
)
1314 netif_tx_wake_queue(txq
);
1317 static int virtnet_poll(struct napi_struct
*napi
, int budget
)
1319 struct receive_queue
*rq
=
1320 container_of(napi
, struct receive_queue
, napi
);
1321 struct virtnet_info
*vi
= rq
->vq
->vdev
->priv
;
1322 struct send_queue
*sq
;
1323 unsigned int received
, qp
;
1324 bool xdp_xmit
= false;
1326 virtnet_poll_cleantx(rq
);
1328 received
= virtnet_receive(rq
, budget
, &xdp_xmit
);
1330 /* Out of packets? */
1331 if (received
< budget
)
1332 virtqueue_napi_complete(napi
, rq
->vq
, received
);
1335 qp
= vi
->curr_queue_pairs
- vi
->xdp_queue_pairs
+
1338 virtqueue_kick(sq
->vq
);
1345 static int virtnet_open(struct net_device
*dev
)
1347 struct virtnet_info
*vi
= netdev_priv(dev
);
1350 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1351 if (i
< vi
->curr_queue_pairs
)
1352 /* Make sure we have some buffers: if oom use wq. */
1353 if (!try_fill_recv(vi
, &vi
->rq
[i
], GFP_KERNEL
))
1354 schedule_delayed_work(&vi
->refill
, 0);
1356 err
= xdp_rxq_info_reg(&vi
->rq
[i
].xdp_rxq
, dev
, i
);
1360 err
= xdp_rxq_info_reg_mem_model(&vi
->rq
[i
].xdp_rxq
,
1361 MEM_TYPE_PAGE_SHARED
, NULL
);
1363 xdp_rxq_info_unreg(&vi
->rq
[i
].xdp_rxq
);
1367 virtnet_napi_enable(vi
->rq
[i
].vq
, &vi
->rq
[i
].napi
);
1368 virtnet_napi_tx_enable(vi
, vi
->sq
[i
].vq
, &vi
->sq
[i
].napi
);
1374 static int virtnet_poll_tx(struct napi_struct
*napi
, int budget
)
1376 struct send_queue
*sq
= container_of(napi
, struct send_queue
, napi
);
1377 struct virtnet_info
*vi
= sq
->vq
->vdev
->priv
;
1378 struct netdev_queue
*txq
= netdev_get_tx_queue(vi
->dev
, vq2txq(sq
->vq
));
1380 __netif_tx_lock(txq
, raw_smp_processor_id());
1381 free_old_xmit_skbs(sq
);
1382 __netif_tx_unlock(txq
);
1384 virtqueue_napi_complete(napi
, sq
->vq
, 0);
1386 if (sq
->vq
->num_free
>= 2 + MAX_SKB_FRAGS
)
1387 netif_tx_wake_queue(txq
);
1392 static int xmit_skb(struct send_queue
*sq
, struct sk_buff
*skb
)
1394 struct virtio_net_hdr_mrg_rxbuf
*hdr
;
1395 const unsigned char *dest
= ((struct ethhdr
*)skb
->data
)->h_dest
;
1396 struct virtnet_info
*vi
= sq
->vq
->vdev
->priv
;
1398 unsigned hdr_len
= vi
->hdr_len
;
1401 pr_debug("%s: xmit %p %pM\n", vi
->dev
->name
, skb
, dest
);
1403 can_push
= vi
->any_header_sg
&&
1404 !((unsigned long)skb
->data
& (__alignof__(*hdr
) - 1)) &&
1405 !skb_header_cloned(skb
) && skb_headroom(skb
) >= hdr_len
;
1406 /* Even if we can, don't push here yet as this would skew
1407 * csum_start offset below. */
1409 hdr
= (struct virtio_net_hdr_mrg_rxbuf
*)(skb
->data
- hdr_len
);
1411 hdr
= skb_vnet_hdr(skb
);
1413 if (virtio_net_hdr_from_skb(skb
, &hdr
->hdr
,
1414 virtio_is_little_endian(vi
->vdev
), false,
1418 if (vi
->mergeable_rx_bufs
)
1419 hdr
->num_buffers
= 0;
1421 sg_init_table(sq
->sg
, skb_shinfo(skb
)->nr_frags
+ (can_push
? 1 : 2));
1423 __skb_push(skb
, hdr_len
);
1424 num_sg
= skb_to_sgvec(skb
, sq
->sg
, 0, skb
->len
);
1425 if (unlikely(num_sg
< 0))
1427 /* Pull header back to avoid skew in tx bytes calculations. */
1428 __skb_pull(skb
, hdr_len
);
1430 sg_set_buf(sq
->sg
, hdr
, hdr_len
);
1431 num_sg
= skb_to_sgvec(skb
, sq
->sg
+ 1, 0, skb
->len
);
1432 if (unlikely(num_sg
< 0))
1436 return virtqueue_add_outbuf(sq
->vq
, sq
->sg
, num_sg
, skb
, GFP_ATOMIC
);
1439 static netdev_tx_t
start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
1441 struct virtnet_info
*vi
= netdev_priv(dev
);
1442 int qnum
= skb_get_queue_mapping(skb
);
1443 struct send_queue
*sq
= &vi
->sq
[qnum
];
1445 struct netdev_queue
*txq
= netdev_get_tx_queue(dev
, qnum
);
1446 bool kick
= !skb
->xmit_more
;
1447 bool use_napi
= sq
->napi
.weight
;
1449 /* Free up any pending old buffers before queueing new ones. */
1450 free_old_xmit_skbs(sq
);
1452 if (use_napi
&& kick
)
1453 virtqueue_enable_cb_delayed(sq
->vq
);
1455 /* timestamp packet in software */
1456 skb_tx_timestamp(skb
);
1458 /* Try to transmit */
1459 err
= xmit_skb(sq
, skb
);
1461 /* This should not happen! */
1462 if (unlikely(err
)) {
1463 dev
->stats
.tx_fifo_errors
++;
1464 if (net_ratelimit())
1466 "Unexpected TXQ (%d) queue failure: %d\n", qnum
, err
);
1467 dev
->stats
.tx_dropped
++;
1468 dev_kfree_skb_any(skb
);
1469 return NETDEV_TX_OK
;
1472 /* Don't wait up for transmitted skbs to be freed. */
1478 /* If running out of space, stop queue to avoid getting packets that we
1479 * are then unable to transmit.
1480 * An alternative would be to force queuing layer to requeue the skb by
1481 * returning NETDEV_TX_BUSY. However, NETDEV_TX_BUSY should not be
1482 * returned in a normal path of operation: it means that driver is not
1483 * maintaining the TX queue stop/start state properly, and causes
1484 * the stack to do a non-trivial amount of useless work.
1485 * Since most packets only take 1 or 2 ring slots, stopping the queue
1486 * early means 16 slots are typically wasted.
1488 if (sq
->vq
->num_free
< 2+MAX_SKB_FRAGS
) {
1489 netif_stop_subqueue(dev
, qnum
);
1491 unlikely(!virtqueue_enable_cb_delayed(sq
->vq
))) {
1492 /* More just got used, free them then recheck. */
1493 free_old_xmit_skbs(sq
);
1494 if (sq
->vq
->num_free
>= 2+MAX_SKB_FRAGS
) {
1495 netif_start_subqueue(dev
, qnum
);
1496 virtqueue_disable_cb(sq
->vq
);
1501 if (kick
|| netif_xmit_stopped(txq
))
1502 virtqueue_kick(sq
->vq
);
1504 return NETDEV_TX_OK
;
1508 * Send command via the control virtqueue and check status. Commands
1509 * supported by the hypervisor, as indicated by feature bits, should
1510 * never fail unless improperly formatted.
1512 static bool virtnet_send_command(struct virtnet_info
*vi
, u8
class, u8 cmd
,
1513 struct scatterlist
*out
)
1515 struct scatterlist
*sgs
[4], hdr
, stat
;
1516 unsigned out_num
= 0, tmp
;
1518 /* Caller should know better */
1519 BUG_ON(!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
));
1521 vi
->ctrl
->status
= ~0;
1522 vi
->ctrl
->hdr
.class = class;
1523 vi
->ctrl
->hdr
.cmd
= cmd
;
1525 sg_init_one(&hdr
, &vi
->ctrl
->hdr
, sizeof(vi
->ctrl
->hdr
));
1526 sgs
[out_num
++] = &hdr
;
1529 sgs
[out_num
++] = out
;
1531 /* Add return status. */
1532 sg_init_one(&stat
, &vi
->ctrl
->status
, sizeof(vi
->ctrl
->status
));
1533 sgs
[out_num
] = &stat
;
1535 BUG_ON(out_num
+ 1 > ARRAY_SIZE(sgs
));
1536 virtqueue_add_sgs(vi
->cvq
, sgs
, out_num
, 1, vi
, GFP_ATOMIC
);
1538 if (unlikely(!virtqueue_kick(vi
->cvq
)))
1539 return vi
->ctrl
->status
== VIRTIO_NET_OK
;
1541 /* Spin for a response, the kick causes an ioport write, trapping
1542 * into the hypervisor, so the request should be handled immediately.
1544 while (!virtqueue_get_buf(vi
->cvq
, &tmp
) &&
1545 !virtqueue_is_broken(vi
->cvq
))
1548 return vi
->ctrl
->status
== VIRTIO_NET_OK
;
1551 static int virtnet_set_mac_address(struct net_device
*dev
, void *p
)
1553 struct virtnet_info
*vi
= netdev_priv(dev
);
1554 struct virtio_device
*vdev
= vi
->vdev
;
1556 struct sockaddr
*addr
;
1557 struct scatterlist sg
;
1559 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_STANDBY
))
1562 addr
= kmemdup(p
, sizeof(*addr
), GFP_KERNEL
);
1566 ret
= eth_prepare_mac_addr_change(dev
, addr
);
1570 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_MAC_ADDR
)) {
1571 sg_init_one(&sg
, addr
->sa_data
, dev
->addr_len
);
1572 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MAC
,
1573 VIRTIO_NET_CTRL_MAC_ADDR_SET
, &sg
)) {
1574 dev_warn(&vdev
->dev
,
1575 "Failed to set mac address by vq command.\n");
1579 } else if (virtio_has_feature(vdev
, VIRTIO_NET_F_MAC
) &&
1580 !virtio_has_feature(vdev
, VIRTIO_F_VERSION_1
)) {
1583 /* Naturally, this has an atomicity problem. */
1584 for (i
= 0; i
< dev
->addr_len
; i
++)
1585 virtio_cwrite8(vdev
,
1586 offsetof(struct virtio_net_config
, mac
) +
1587 i
, addr
->sa_data
[i
]);
1590 eth_commit_mac_addr_change(dev
, p
);
1598 static void virtnet_stats(struct net_device
*dev
,
1599 struct rtnl_link_stats64
*tot
)
1601 struct virtnet_info
*vi
= netdev_priv(dev
);
1605 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1606 u64 tpackets
, tbytes
, rpackets
, rbytes
;
1607 struct receive_queue
*rq
= &vi
->rq
[i
];
1608 struct send_queue
*sq
= &vi
->sq
[i
];
1611 start
= u64_stats_fetch_begin_irq(&sq
->stats
.syncp
);
1612 tpackets
= sq
->stats
.packets
;
1613 tbytes
= sq
->stats
.bytes
;
1614 } while (u64_stats_fetch_retry_irq(&sq
->stats
.syncp
, start
));
1617 start
= u64_stats_fetch_begin_irq(&rq
->stats
.syncp
);
1618 rpackets
= rq
->stats
.packets
;
1619 rbytes
= rq
->stats
.bytes
;
1620 } while (u64_stats_fetch_retry_irq(&rq
->stats
.syncp
, start
));
1622 tot
->rx_packets
+= rpackets
;
1623 tot
->tx_packets
+= tpackets
;
1624 tot
->rx_bytes
+= rbytes
;
1625 tot
->tx_bytes
+= tbytes
;
1628 tot
->tx_dropped
= dev
->stats
.tx_dropped
;
1629 tot
->tx_fifo_errors
= dev
->stats
.tx_fifo_errors
;
1630 tot
->rx_dropped
= dev
->stats
.rx_dropped
;
1631 tot
->rx_length_errors
= dev
->stats
.rx_length_errors
;
1632 tot
->rx_frame_errors
= dev
->stats
.rx_frame_errors
;
1635 #ifdef CONFIG_NET_POLL_CONTROLLER
1636 static void virtnet_netpoll(struct net_device
*dev
)
1638 struct virtnet_info
*vi
= netdev_priv(dev
);
1641 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++)
1642 napi_schedule(&vi
->rq
[i
].napi
);
1646 static void virtnet_ack_link_announce(struct virtnet_info
*vi
)
1649 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_ANNOUNCE
,
1650 VIRTIO_NET_CTRL_ANNOUNCE_ACK
, NULL
))
1651 dev_warn(&vi
->dev
->dev
, "Failed to ack link announce.\n");
1655 static int _virtnet_set_queues(struct virtnet_info
*vi
, u16 queue_pairs
)
1657 struct scatterlist sg
;
1658 struct net_device
*dev
= vi
->dev
;
1660 if (!vi
->has_cvq
|| !virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_MQ
))
1663 vi
->ctrl
->mq
.virtqueue_pairs
= cpu_to_virtio16(vi
->vdev
, queue_pairs
);
1664 sg_init_one(&sg
, &vi
->ctrl
->mq
, sizeof(vi
->ctrl
->mq
));
1666 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MQ
,
1667 VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET
, &sg
)) {
1668 dev_warn(&dev
->dev
, "Fail to set num of queue pairs to %d\n",
1672 vi
->curr_queue_pairs
= queue_pairs
;
1673 /* virtnet_open() will refill when device is going to up. */
1674 if (dev
->flags
& IFF_UP
)
1675 schedule_delayed_work(&vi
->refill
, 0);
1681 static int virtnet_set_queues(struct virtnet_info
*vi
, u16 queue_pairs
)
1686 err
= _virtnet_set_queues(vi
, queue_pairs
);
1691 static int virtnet_close(struct net_device
*dev
)
1693 struct virtnet_info
*vi
= netdev_priv(dev
);
1696 /* Make sure refill_work doesn't re-enable napi! */
1697 cancel_delayed_work_sync(&vi
->refill
);
1699 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1700 xdp_rxq_info_unreg(&vi
->rq
[i
].xdp_rxq
);
1701 napi_disable(&vi
->rq
[i
].napi
);
1702 virtnet_napi_tx_disable(&vi
->sq
[i
].napi
);
1708 static void virtnet_set_rx_mode(struct net_device
*dev
)
1710 struct virtnet_info
*vi
= netdev_priv(dev
);
1711 struct scatterlist sg
[2];
1712 struct virtio_net_ctrl_mac
*mac_data
;
1713 struct netdev_hw_addr
*ha
;
1719 /* We can't dynamically set ndo_set_rx_mode, so return gracefully */
1720 if (!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_RX
))
1723 vi
->ctrl
->promisc
= ((dev
->flags
& IFF_PROMISC
) != 0);
1724 vi
->ctrl
->allmulti
= ((dev
->flags
& IFF_ALLMULTI
) != 0);
1726 sg_init_one(sg
, &vi
->ctrl
->promisc
, sizeof(vi
->ctrl
->promisc
));
1728 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_RX
,
1729 VIRTIO_NET_CTRL_RX_PROMISC
, sg
))
1730 dev_warn(&dev
->dev
, "Failed to %sable promisc mode.\n",
1731 vi
->ctrl
->promisc
? "en" : "dis");
1733 sg_init_one(sg
, &vi
->ctrl
->allmulti
, sizeof(vi
->ctrl
->allmulti
));
1735 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_RX
,
1736 VIRTIO_NET_CTRL_RX_ALLMULTI
, sg
))
1737 dev_warn(&dev
->dev
, "Failed to %sable allmulti mode.\n",
1738 vi
->ctrl
->allmulti
? "en" : "dis");
1740 uc_count
= netdev_uc_count(dev
);
1741 mc_count
= netdev_mc_count(dev
);
1742 /* MAC filter - use one buffer for both lists */
1743 buf
= kzalloc(((uc_count
+ mc_count
) * ETH_ALEN
) +
1744 (2 * sizeof(mac_data
->entries
)), GFP_ATOMIC
);
1749 sg_init_table(sg
, 2);
1751 /* Store the unicast list and count in the front of the buffer */
1752 mac_data
->entries
= cpu_to_virtio32(vi
->vdev
, uc_count
);
1754 netdev_for_each_uc_addr(ha
, dev
)
1755 memcpy(&mac_data
->macs
[i
++][0], ha
->addr
, ETH_ALEN
);
1757 sg_set_buf(&sg
[0], mac_data
,
1758 sizeof(mac_data
->entries
) + (uc_count
* ETH_ALEN
));
1760 /* multicast list and count fill the end */
1761 mac_data
= (void *)&mac_data
->macs
[uc_count
][0];
1763 mac_data
->entries
= cpu_to_virtio32(vi
->vdev
, mc_count
);
1765 netdev_for_each_mc_addr(ha
, dev
)
1766 memcpy(&mac_data
->macs
[i
++][0], ha
->addr
, ETH_ALEN
);
1768 sg_set_buf(&sg
[1], mac_data
,
1769 sizeof(mac_data
->entries
) + (mc_count
* ETH_ALEN
));
1771 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_MAC
,
1772 VIRTIO_NET_CTRL_MAC_TABLE_SET
, sg
))
1773 dev_warn(&dev
->dev
, "Failed to set MAC filter table.\n");
1778 static int virtnet_vlan_rx_add_vid(struct net_device
*dev
,
1779 __be16 proto
, u16 vid
)
1781 struct virtnet_info
*vi
= netdev_priv(dev
);
1782 struct scatterlist sg
;
1784 vi
->ctrl
->vid
= cpu_to_virtio16(vi
->vdev
, vid
);
1785 sg_init_one(&sg
, &vi
->ctrl
->vid
, sizeof(vi
->ctrl
->vid
));
1787 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_VLAN
,
1788 VIRTIO_NET_CTRL_VLAN_ADD
, &sg
))
1789 dev_warn(&dev
->dev
, "Failed to add VLAN ID %d.\n", vid
);
1793 static int virtnet_vlan_rx_kill_vid(struct net_device
*dev
,
1794 __be16 proto
, u16 vid
)
1796 struct virtnet_info
*vi
= netdev_priv(dev
);
1797 struct scatterlist sg
;
1799 vi
->ctrl
->vid
= cpu_to_virtio16(vi
->vdev
, vid
);
1800 sg_init_one(&sg
, &vi
->ctrl
->vid
, sizeof(vi
->ctrl
->vid
));
1802 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_VLAN
,
1803 VIRTIO_NET_CTRL_VLAN_DEL
, &sg
))
1804 dev_warn(&dev
->dev
, "Failed to kill VLAN ID %d.\n", vid
);
1808 static void virtnet_clean_affinity(struct virtnet_info
*vi
, long hcpu
)
1812 if (vi
->affinity_hint_set
) {
1813 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
1814 virtqueue_set_affinity(vi
->rq
[i
].vq
, -1);
1815 virtqueue_set_affinity(vi
->sq
[i
].vq
, -1);
1818 vi
->affinity_hint_set
= false;
1822 static void virtnet_set_affinity(struct virtnet_info
*vi
)
1827 /* In multiqueue mode, when the number of cpu is equal to the number of
1828 * queue pairs, we let the queue pairs to be private to one cpu by
1829 * setting the affinity hint to eliminate the contention.
1831 if (vi
->curr_queue_pairs
== 1 ||
1832 vi
->max_queue_pairs
!= num_online_cpus()) {
1833 virtnet_clean_affinity(vi
, -1);
1838 for_each_online_cpu(cpu
) {
1839 virtqueue_set_affinity(vi
->rq
[i
].vq
, cpu
);
1840 virtqueue_set_affinity(vi
->sq
[i
].vq
, cpu
);
1841 netif_set_xps_queue(vi
->dev
, cpumask_of(cpu
), i
);
1845 vi
->affinity_hint_set
= true;
1848 static int virtnet_cpu_online(unsigned int cpu
, struct hlist_node
*node
)
1850 struct virtnet_info
*vi
= hlist_entry_safe(node
, struct virtnet_info
,
1852 virtnet_set_affinity(vi
);
1856 static int virtnet_cpu_dead(unsigned int cpu
, struct hlist_node
*node
)
1858 struct virtnet_info
*vi
= hlist_entry_safe(node
, struct virtnet_info
,
1860 virtnet_set_affinity(vi
);
1864 static int virtnet_cpu_down_prep(unsigned int cpu
, struct hlist_node
*node
)
1866 struct virtnet_info
*vi
= hlist_entry_safe(node
, struct virtnet_info
,
1869 virtnet_clean_affinity(vi
, cpu
);
1873 static enum cpuhp_state virtionet_online
;
1875 static int virtnet_cpu_notif_add(struct virtnet_info
*vi
)
1879 ret
= cpuhp_state_add_instance_nocalls(virtionet_online
, &vi
->node
);
1882 ret
= cpuhp_state_add_instance_nocalls(CPUHP_VIRT_NET_DEAD
,
1886 cpuhp_state_remove_instance_nocalls(virtionet_online
, &vi
->node
);
1890 static void virtnet_cpu_notif_remove(struct virtnet_info
*vi
)
1892 cpuhp_state_remove_instance_nocalls(virtionet_online
, &vi
->node
);
1893 cpuhp_state_remove_instance_nocalls(CPUHP_VIRT_NET_DEAD
,
1897 static void virtnet_get_ringparam(struct net_device
*dev
,
1898 struct ethtool_ringparam
*ring
)
1900 struct virtnet_info
*vi
= netdev_priv(dev
);
1902 ring
->rx_max_pending
= virtqueue_get_vring_size(vi
->rq
[0].vq
);
1903 ring
->tx_max_pending
= virtqueue_get_vring_size(vi
->sq
[0].vq
);
1904 ring
->rx_pending
= ring
->rx_max_pending
;
1905 ring
->tx_pending
= ring
->tx_max_pending
;
1909 static void virtnet_get_drvinfo(struct net_device
*dev
,
1910 struct ethtool_drvinfo
*info
)
1912 struct virtnet_info
*vi
= netdev_priv(dev
);
1913 struct virtio_device
*vdev
= vi
->vdev
;
1915 strlcpy(info
->driver
, KBUILD_MODNAME
, sizeof(info
->driver
));
1916 strlcpy(info
->version
, VIRTNET_DRIVER_VERSION
, sizeof(info
->version
));
1917 strlcpy(info
->bus_info
, virtio_bus_name(vdev
), sizeof(info
->bus_info
));
1921 /* TODO: Eliminate OOO packets during switching */
1922 static int virtnet_set_channels(struct net_device
*dev
,
1923 struct ethtool_channels
*channels
)
1925 struct virtnet_info
*vi
= netdev_priv(dev
);
1926 u16 queue_pairs
= channels
->combined_count
;
1929 /* We don't support separate rx/tx channels.
1930 * We don't allow setting 'other' channels.
1932 if (channels
->rx_count
|| channels
->tx_count
|| channels
->other_count
)
1935 if (queue_pairs
> vi
->max_queue_pairs
|| queue_pairs
== 0)
1938 /* For now we don't support modifying channels while XDP is loaded
1939 * also when XDP is loaded all RX queues have XDP programs so we only
1940 * need to check a single RX queue.
1942 if (vi
->rq
[0].xdp_prog
)
1946 err
= _virtnet_set_queues(vi
, queue_pairs
);
1948 netif_set_real_num_tx_queues(dev
, queue_pairs
);
1949 netif_set_real_num_rx_queues(dev
, queue_pairs
);
1951 virtnet_set_affinity(vi
);
1958 static void virtnet_get_strings(struct net_device
*dev
, u32 stringset
, u8
*data
)
1960 struct virtnet_info
*vi
= netdev_priv(dev
);
1961 char *p
= (char *)data
;
1964 switch (stringset
) {
1966 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++) {
1967 for (j
= 0; j
< VIRTNET_RQ_STATS_LEN
; j
++) {
1968 snprintf(p
, ETH_GSTRING_LEN
, "rx_queue_%u_%s",
1969 i
, virtnet_rq_stats_desc
[j
].desc
);
1970 p
+= ETH_GSTRING_LEN
;
1974 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++) {
1975 for (j
= 0; j
< VIRTNET_SQ_STATS_LEN
; j
++) {
1976 snprintf(p
, ETH_GSTRING_LEN
, "tx_queue_%u_%s",
1977 i
, virtnet_sq_stats_desc
[j
].desc
);
1978 p
+= ETH_GSTRING_LEN
;
1985 static int virtnet_get_sset_count(struct net_device
*dev
, int sset
)
1987 struct virtnet_info
*vi
= netdev_priv(dev
);
1991 return vi
->curr_queue_pairs
* (VIRTNET_RQ_STATS_LEN
+
1992 VIRTNET_SQ_STATS_LEN
);
1998 static void virtnet_get_ethtool_stats(struct net_device
*dev
,
1999 struct ethtool_stats
*stats
, u64
*data
)
2001 struct virtnet_info
*vi
= netdev_priv(dev
);
2002 unsigned int idx
= 0, start
, i
, j
;
2003 const u8
*stats_base
;
2006 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++) {
2007 struct receive_queue
*rq
= &vi
->rq
[i
];
2009 stats_base
= (u8
*)&rq
->stats
;
2011 start
= u64_stats_fetch_begin_irq(&rq
->stats
.syncp
);
2012 for (j
= 0; j
< VIRTNET_RQ_STATS_LEN
; j
++) {
2013 offset
= virtnet_rq_stats_desc
[j
].offset
;
2014 data
[idx
+ j
] = *(u64
*)(stats_base
+ offset
);
2016 } while (u64_stats_fetch_retry_irq(&rq
->stats
.syncp
, start
));
2017 idx
+= VIRTNET_RQ_STATS_LEN
;
2020 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++) {
2021 struct send_queue
*sq
= &vi
->sq
[i
];
2023 stats_base
= (u8
*)&sq
->stats
;
2025 start
= u64_stats_fetch_begin_irq(&sq
->stats
.syncp
);
2026 for (j
= 0; j
< VIRTNET_SQ_STATS_LEN
; j
++) {
2027 offset
= virtnet_sq_stats_desc
[j
].offset
;
2028 data
[idx
+ j
] = *(u64
*)(stats_base
+ offset
);
2030 } while (u64_stats_fetch_retry_irq(&sq
->stats
.syncp
, start
));
2031 idx
+= VIRTNET_SQ_STATS_LEN
;
2035 static void virtnet_get_channels(struct net_device
*dev
,
2036 struct ethtool_channels
*channels
)
2038 struct virtnet_info
*vi
= netdev_priv(dev
);
2040 channels
->combined_count
= vi
->curr_queue_pairs
;
2041 channels
->max_combined
= vi
->max_queue_pairs
;
2042 channels
->max_other
= 0;
2043 channels
->rx_count
= 0;
2044 channels
->tx_count
= 0;
2045 channels
->other_count
= 0;
2048 /* Check if the user is trying to change anything besides speed/duplex */
2050 virtnet_validate_ethtool_cmd(const struct ethtool_link_ksettings
*cmd
)
2052 struct ethtool_link_ksettings diff1
= *cmd
;
2053 struct ethtool_link_ksettings diff2
= {};
2055 /* cmd is always set so we need to clear it, validate the port type
2056 * and also without autonegotiation we can ignore advertising
2058 diff1
.base
.speed
= 0;
2059 diff2
.base
.port
= PORT_OTHER
;
2060 ethtool_link_ksettings_zero_link_mode(&diff1
, advertising
);
2061 diff1
.base
.duplex
= 0;
2063 diff1
.base
.link_mode_masks_nwords
= 0;
2065 return !memcmp(&diff1
.base
, &diff2
.base
, sizeof(diff1
.base
)) &&
2066 bitmap_empty(diff1
.link_modes
.supported
,
2067 __ETHTOOL_LINK_MODE_MASK_NBITS
) &&
2068 bitmap_empty(diff1
.link_modes
.advertising
,
2069 __ETHTOOL_LINK_MODE_MASK_NBITS
) &&
2070 bitmap_empty(diff1
.link_modes
.lp_advertising
,
2071 __ETHTOOL_LINK_MODE_MASK_NBITS
);
2074 static int virtnet_set_link_ksettings(struct net_device
*dev
,
2075 const struct ethtool_link_ksettings
*cmd
)
2077 struct virtnet_info
*vi
= netdev_priv(dev
);
2080 speed
= cmd
->base
.speed
;
2081 /* don't allow custom speed and duplex */
2082 if (!ethtool_validate_speed(speed
) ||
2083 !ethtool_validate_duplex(cmd
->base
.duplex
) ||
2084 !virtnet_validate_ethtool_cmd(cmd
))
2087 vi
->duplex
= cmd
->base
.duplex
;
2092 static int virtnet_get_link_ksettings(struct net_device
*dev
,
2093 struct ethtool_link_ksettings
*cmd
)
2095 struct virtnet_info
*vi
= netdev_priv(dev
);
2097 cmd
->base
.speed
= vi
->speed
;
2098 cmd
->base
.duplex
= vi
->duplex
;
2099 cmd
->base
.port
= PORT_OTHER
;
2104 static void virtnet_init_settings(struct net_device
*dev
)
2106 struct virtnet_info
*vi
= netdev_priv(dev
);
2108 vi
->speed
= SPEED_UNKNOWN
;
2109 vi
->duplex
= DUPLEX_UNKNOWN
;
2112 static void virtnet_update_settings(struct virtnet_info
*vi
)
2117 if (!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_SPEED_DUPLEX
))
2120 speed
= virtio_cread32(vi
->vdev
, offsetof(struct virtio_net_config
,
2122 if (ethtool_validate_speed(speed
))
2124 duplex
= virtio_cread8(vi
->vdev
, offsetof(struct virtio_net_config
,
2126 if (ethtool_validate_duplex(duplex
))
2127 vi
->duplex
= duplex
;
2130 static const struct ethtool_ops virtnet_ethtool_ops
= {
2131 .get_drvinfo
= virtnet_get_drvinfo
,
2132 .get_link
= ethtool_op_get_link
,
2133 .get_ringparam
= virtnet_get_ringparam
,
2134 .get_strings
= virtnet_get_strings
,
2135 .get_sset_count
= virtnet_get_sset_count
,
2136 .get_ethtool_stats
= virtnet_get_ethtool_stats
,
2137 .set_channels
= virtnet_set_channels
,
2138 .get_channels
= virtnet_get_channels
,
2139 .get_ts_info
= ethtool_op_get_ts_info
,
2140 .get_link_ksettings
= virtnet_get_link_ksettings
,
2141 .set_link_ksettings
= virtnet_set_link_ksettings
,
2144 static void virtnet_freeze_down(struct virtio_device
*vdev
)
2146 struct virtnet_info
*vi
= vdev
->priv
;
2149 /* Make sure no work handler is accessing the device */
2150 flush_work(&vi
->config_work
);
2152 netif_device_detach(vi
->dev
);
2153 netif_tx_disable(vi
->dev
);
2154 cancel_delayed_work_sync(&vi
->refill
);
2156 if (netif_running(vi
->dev
)) {
2157 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
2158 napi_disable(&vi
->rq
[i
].napi
);
2159 virtnet_napi_tx_disable(&vi
->sq
[i
].napi
);
2164 static int init_vqs(struct virtnet_info
*vi
);
2166 static int virtnet_restore_up(struct virtio_device
*vdev
)
2168 struct virtnet_info
*vi
= vdev
->priv
;
2175 virtio_device_ready(vdev
);
2177 if (netif_running(vi
->dev
)) {
2178 for (i
= 0; i
< vi
->curr_queue_pairs
; i
++)
2179 if (!try_fill_recv(vi
, &vi
->rq
[i
], GFP_KERNEL
))
2180 schedule_delayed_work(&vi
->refill
, 0);
2182 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
2183 virtnet_napi_enable(vi
->rq
[i
].vq
, &vi
->rq
[i
].napi
);
2184 virtnet_napi_tx_enable(vi
, vi
->sq
[i
].vq
,
2189 netif_device_attach(vi
->dev
);
2193 static int virtnet_set_guest_offloads(struct virtnet_info
*vi
, u64 offloads
)
2195 struct scatterlist sg
;
2196 vi
->ctrl
->offloads
= cpu_to_virtio64(vi
->vdev
, offloads
);
2198 sg_init_one(&sg
, &vi
->ctrl
->offloads
, sizeof(vi
->ctrl
->offloads
));
2200 if (!virtnet_send_command(vi
, VIRTIO_NET_CTRL_GUEST_OFFLOADS
,
2201 VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET
, &sg
)) {
2202 dev_warn(&vi
->dev
->dev
, "Fail to set guest offload. \n");
2209 static int virtnet_clear_guest_offloads(struct virtnet_info
*vi
)
2213 if (!vi
->guest_offloads
)
2216 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_GUEST_CSUM
))
2217 offloads
= 1ULL << VIRTIO_NET_F_GUEST_CSUM
;
2219 return virtnet_set_guest_offloads(vi
, offloads
);
2222 static int virtnet_restore_guest_offloads(struct virtnet_info
*vi
)
2224 u64 offloads
= vi
->guest_offloads
;
2226 if (!vi
->guest_offloads
)
2228 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_GUEST_CSUM
))
2229 offloads
|= 1ULL << VIRTIO_NET_F_GUEST_CSUM
;
2231 return virtnet_set_guest_offloads(vi
, offloads
);
2234 static int virtnet_xdp_set(struct net_device
*dev
, struct bpf_prog
*prog
,
2235 struct netlink_ext_ack
*extack
)
2237 unsigned long int max_sz
= PAGE_SIZE
- sizeof(struct padded_vnet_hdr
);
2238 struct virtnet_info
*vi
= netdev_priv(dev
);
2239 struct bpf_prog
*old_prog
;
2240 u16 xdp_qp
= 0, curr_qp
;
2243 if (!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS
)
2244 && (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_GUEST_TSO4
) ||
2245 virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_GUEST_TSO6
) ||
2246 virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_GUEST_ECN
) ||
2247 virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_GUEST_UFO
))) {
2248 NL_SET_ERR_MSG_MOD(extack
, "Can't set XDP while host is implementing LRO, disable LRO first");
2252 if (vi
->mergeable_rx_bufs
&& !vi
->any_header_sg
) {
2253 NL_SET_ERR_MSG_MOD(extack
, "XDP expects header/data in single page, any_header_sg required");
2257 if (dev
->mtu
> max_sz
) {
2258 NL_SET_ERR_MSG_MOD(extack
, "MTU too large to enable XDP");
2259 netdev_warn(dev
, "XDP requires MTU less than %lu\n", max_sz
);
2263 curr_qp
= vi
->curr_queue_pairs
- vi
->xdp_queue_pairs
;
2265 xdp_qp
= nr_cpu_ids
;
2267 /* XDP requires extra queues for XDP_TX */
2268 if (curr_qp
+ xdp_qp
> vi
->max_queue_pairs
) {
2269 NL_SET_ERR_MSG_MOD(extack
, "Too few free TX rings available");
2270 netdev_warn(dev
, "request %i queues but max is %i\n",
2271 curr_qp
+ xdp_qp
, vi
->max_queue_pairs
);
2276 prog
= bpf_prog_add(prog
, vi
->max_queue_pairs
- 1);
2278 return PTR_ERR(prog
);
2281 /* Make sure NAPI is not using any XDP TX queues for RX. */
2282 if (netif_running(dev
))
2283 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
2284 napi_disable(&vi
->rq
[i
].napi
);
2286 netif_set_real_num_rx_queues(dev
, curr_qp
+ xdp_qp
);
2287 err
= _virtnet_set_queues(vi
, curr_qp
+ xdp_qp
);
2290 vi
->xdp_queue_pairs
= xdp_qp
;
2292 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
2293 old_prog
= rtnl_dereference(vi
->rq
[i
].xdp_prog
);
2294 rcu_assign_pointer(vi
->rq
[i
].xdp_prog
, prog
);
2297 virtnet_clear_guest_offloads(vi
);
2299 virtnet_restore_guest_offloads(vi
);
2302 bpf_prog_put(old_prog
);
2303 if (netif_running(dev
))
2304 virtnet_napi_enable(vi
->rq
[i
].vq
, &vi
->rq
[i
].napi
);
2310 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
2311 virtnet_napi_enable(vi
->rq
[i
].vq
, &vi
->rq
[i
].napi
);
2313 bpf_prog_sub(prog
, vi
->max_queue_pairs
- 1);
2317 static u32
virtnet_xdp_query(struct net_device
*dev
)
2319 struct virtnet_info
*vi
= netdev_priv(dev
);
2320 const struct bpf_prog
*xdp_prog
;
2323 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
2324 xdp_prog
= rtnl_dereference(vi
->rq
[i
].xdp_prog
);
2326 return xdp_prog
->aux
->id
;
2331 static int virtnet_xdp(struct net_device
*dev
, struct netdev_bpf
*xdp
)
2333 switch (xdp
->command
) {
2334 case XDP_SETUP_PROG
:
2335 return virtnet_xdp_set(dev
, xdp
->prog
, xdp
->extack
);
2336 case XDP_QUERY_PROG
:
2337 xdp
->prog_id
= virtnet_xdp_query(dev
);
2338 xdp
->prog_attached
= !!xdp
->prog_id
;
2345 static int virtnet_get_phys_port_name(struct net_device
*dev
, char *buf
,
2348 struct virtnet_info
*vi
= netdev_priv(dev
);
2351 if (!virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_STANDBY
))
2354 ret
= snprintf(buf
, len
, "sby");
2361 static const struct net_device_ops virtnet_netdev
= {
2362 .ndo_open
= virtnet_open
,
2363 .ndo_stop
= virtnet_close
,
2364 .ndo_start_xmit
= start_xmit
,
2365 .ndo_validate_addr
= eth_validate_addr
,
2366 .ndo_set_mac_address
= virtnet_set_mac_address
,
2367 .ndo_set_rx_mode
= virtnet_set_rx_mode
,
2368 .ndo_get_stats64
= virtnet_stats
,
2369 .ndo_vlan_rx_add_vid
= virtnet_vlan_rx_add_vid
,
2370 .ndo_vlan_rx_kill_vid
= virtnet_vlan_rx_kill_vid
,
2371 #ifdef CONFIG_NET_POLL_CONTROLLER
2372 .ndo_poll_controller
= virtnet_netpoll
,
2374 .ndo_bpf
= virtnet_xdp
,
2375 .ndo_xdp_xmit
= virtnet_xdp_xmit
,
2376 .ndo_features_check
= passthru_features_check
,
2377 .ndo_get_phys_port_name
= virtnet_get_phys_port_name
,
2380 static void virtnet_config_changed_work(struct work_struct
*work
)
2382 struct virtnet_info
*vi
=
2383 container_of(work
, struct virtnet_info
, config_work
);
2386 if (virtio_cread_feature(vi
->vdev
, VIRTIO_NET_F_STATUS
,
2387 struct virtio_net_config
, status
, &v
) < 0)
2390 if (v
& VIRTIO_NET_S_ANNOUNCE
) {
2391 netdev_notify_peers(vi
->dev
);
2392 virtnet_ack_link_announce(vi
);
2395 /* Ignore unknown (future) status bits */
2396 v
&= VIRTIO_NET_S_LINK_UP
;
2398 if (vi
->status
== v
)
2403 if (vi
->status
& VIRTIO_NET_S_LINK_UP
) {
2404 virtnet_update_settings(vi
);
2405 netif_carrier_on(vi
->dev
);
2406 netif_tx_wake_all_queues(vi
->dev
);
2408 netif_carrier_off(vi
->dev
);
2409 netif_tx_stop_all_queues(vi
->dev
);
2413 static void virtnet_config_changed(struct virtio_device
*vdev
)
2415 struct virtnet_info
*vi
= vdev
->priv
;
2417 schedule_work(&vi
->config_work
);
2420 static void virtnet_free_queues(struct virtnet_info
*vi
)
2424 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
2425 napi_hash_del(&vi
->rq
[i
].napi
);
2426 netif_napi_del(&vi
->rq
[i
].napi
);
2427 netif_napi_del(&vi
->sq
[i
].napi
);
2430 /* We called napi_hash_del() before netif_napi_del(),
2431 * we need to respect an RCU grace period before freeing vi->rq
2440 static void _free_receive_bufs(struct virtnet_info
*vi
)
2442 struct bpf_prog
*old_prog
;
2445 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
2446 while (vi
->rq
[i
].pages
)
2447 __free_pages(get_a_page(&vi
->rq
[i
], GFP_KERNEL
), 0);
2449 old_prog
= rtnl_dereference(vi
->rq
[i
].xdp_prog
);
2450 RCU_INIT_POINTER(vi
->rq
[i
].xdp_prog
, NULL
);
2452 bpf_prog_put(old_prog
);
2456 static void free_receive_bufs(struct virtnet_info
*vi
)
2459 _free_receive_bufs(vi
);
2463 static void free_receive_page_frags(struct virtnet_info
*vi
)
2466 for (i
= 0; i
< vi
->max_queue_pairs
; i
++)
2467 if (vi
->rq
[i
].alloc_frag
.page
)
2468 put_page(vi
->rq
[i
].alloc_frag
.page
);
2471 static bool is_xdp_raw_buffer_queue(struct virtnet_info
*vi
, int q
)
2473 if (q
< (vi
->curr_queue_pairs
- vi
->xdp_queue_pairs
))
2475 else if (q
< vi
->curr_queue_pairs
)
2481 static void free_unused_bufs(struct virtnet_info
*vi
)
2486 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
2487 struct virtqueue
*vq
= vi
->sq
[i
].vq
;
2488 while ((buf
= virtqueue_detach_unused_buf(vq
)) != NULL
) {
2489 if (!is_xdp_raw_buffer_queue(vi
, i
))
2492 put_page(virt_to_head_page(buf
));
2496 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
2497 struct virtqueue
*vq
= vi
->rq
[i
].vq
;
2499 while ((buf
= virtqueue_detach_unused_buf(vq
)) != NULL
) {
2500 if (vi
->mergeable_rx_bufs
) {
2501 put_page(virt_to_head_page(buf
));
2502 } else if (vi
->big_packets
) {
2503 give_pages(&vi
->rq
[i
], buf
);
2505 put_page(virt_to_head_page(buf
));
2511 static void virtnet_del_vqs(struct virtnet_info
*vi
)
2513 struct virtio_device
*vdev
= vi
->vdev
;
2515 virtnet_clean_affinity(vi
, -1);
2517 vdev
->config
->del_vqs(vdev
);
2519 virtnet_free_queues(vi
);
2522 /* How large should a single buffer be so a queue full of these can fit at
2523 * least one full packet?
2524 * Logic below assumes the mergeable buffer header is used.
2526 static unsigned int mergeable_min_buf_len(struct virtnet_info
*vi
, struct virtqueue
*vq
)
2528 const unsigned int hdr_len
= sizeof(struct virtio_net_hdr_mrg_rxbuf
);
2529 unsigned int rq_size
= virtqueue_get_vring_size(vq
);
2530 unsigned int packet_len
= vi
->big_packets
? IP_MAX_MTU
: vi
->dev
->max_mtu
;
2531 unsigned int buf_len
= hdr_len
+ ETH_HLEN
+ VLAN_HLEN
+ packet_len
;
2532 unsigned int min_buf_len
= DIV_ROUND_UP(buf_len
, rq_size
);
2534 return max(max(min_buf_len
, hdr_len
) - hdr_len
,
2535 (unsigned int)GOOD_PACKET_LEN
);
2538 static int virtnet_find_vqs(struct virtnet_info
*vi
)
2540 vq_callback_t
**callbacks
;
2541 struct virtqueue
**vqs
;
2547 /* We expect 1 RX virtqueue followed by 1 TX virtqueue, followed by
2548 * possible N-1 RX/TX queue pairs used in multiqueue mode, followed by
2549 * possible control vq.
2551 total_vqs
= vi
->max_queue_pairs
* 2 +
2552 virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VQ
);
2554 /* Allocate space for find_vqs parameters */
2555 vqs
= kzalloc(total_vqs
* sizeof(*vqs
), GFP_KERNEL
);
2558 callbacks
= kmalloc_array(total_vqs
, sizeof(*callbacks
), GFP_KERNEL
);
2561 names
= kmalloc_array(total_vqs
, sizeof(*names
), GFP_KERNEL
);
2564 if (!vi
->big_packets
|| vi
->mergeable_rx_bufs
) {
2565 ctx
= kzalloc(total_vqs
* sizeof(*ctx
), GFP_KERNEL
);
2572 /* Parameters for control virtqueue, if any */
2574 callbacks
[total_vqs
- 1] = NULL
;
2575 names
[total_vqs
- 1] = "control";
2578 /* Allocate/initialize parameters for send/receive virtqueues */
2579 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
2580 callbacks
[rxq2vq(i
)] = skb_recv_done
;
2581 callbacks
[txq2vq(i
)] = skb_xmit_done
;
2582 sprintf(vi
->rq
[i
].name
, "input.%d", i
);
2583 sprintf(vi
->sq
[i
].name
, "output.%d", i
);
2584 names
[rxq2vq(i
)] = vi
->rq
[i
].name
;
2585 names
[txq2vq(i
)] = vi
->sq
[i
].name
;
2587 ctx
[rxq2vq(i
)] = true;
2590 ret
= vi
->vdev
->config
->find_vqs(vi
->vdev
, total_vqs
, vqs
, callbacks
,
2596 vi
->cvq
= vqs
[total_vqs
- 1];
2597 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_CTRL_VLAN
))
2598 vi
->dev
->features
|= NETIF_F_HW_VLAN_CTAG_FILTER
;
2601 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
2602 vi
->rq
[i
].vq
= vqs
[rxq2vq(i
)];
2603 vi
->rq
[i
].min_buf_len
= mergeable_min_buf_len(vi
, vi
->rq
[i
].vq
);
2604 vi
->sq
[i
].vq
= vqs
[txq2vq(i
)];
2607 /* run here: ret == 0. */
2622 static int virtnet_alloc_queues(struct virtnet_info
*vi
)
2626 vi
->ctrl
= kzalloc(sizeof(*vi
->ctrl
), GFP_KERNEL
);
2629 vi
->sq
= kzalloc(sizeof(*vi
->sq
) * vi
->max_queue_pairs
, GFP_KERNEL
);
2632 vi
->rq
= kzalloc(sizeof(*vi
->rq
) * vi
->max_queue_pairs
, GFP_KERNEL
);
2636 INIT_DELAYED_WORK(&vi
->refill
, refill_work
);
2637 for (i
= 0; i
< vi
->max_queue_pairs
; i
++) {
2638 vi
->rq
[i
].pages
= NULL
;
2639 netif_napi_add(vi
->dev
, &vi
->rq
[i
].napi
, virtnet_poll
,
2641 netif_tx_napi_add(vi
->dev
, &vi
->sq
[i
].napi
, virtnet_poll_tx
,
2642 napi_tx
? napi_weight
: 0);
2644 sg_init_table(vi
->rq
[i
].sg
, ARRAY_SIZE(vi
->rq
[i
].sg
));
2645 ewma_pkt_len_init(&vi
->rq
[i
].mrg_avg_pkt_len
);
2646 sg_init_table(vi
->sq
[i
].sg
, ARRAY_SIZE(vi
->sq
[i
].sg
));
2648 u64_stats_init(&vi
->rq
[i
].stats
.syncp
);
2649 u64_stats_init(&vi
->sq
[i
].stats
.syncp
);
2662 static int init_vqs(struct virtnet_info
*vi
)
2666 /* Allocate send & receive queues */
2667 ret
= virtnet_alloc_queues(vi
);
2671 ret
= virtnet_find_vqs(vi
);
2676 virtnet_set_affinity(vi
);
2682 virtnet_free_queues(vi
);
2688 static ssize_t
mergeable_rx_buffer_size_show(struct netdev_rx_queue
*queue
,
2691 struct virtnet_info
*vi
= netdev_priv(queue
->dev
);
2692 unsigned int queue_index
= get_netdev_rx_queue_index(queue
);
2693 unsigned int headroom
= virtnet_get_headroom(vi
);
2694 unsigned int tailroom
= headroom
? sizeof(struct skb_shared_info
) : 0;
2695 struct ewma_pkt_len
*avg
;
2697 BUG_ON(queue_index
>= vi
->max_queue_pairs
);
2698 avg
= &vi
->rq
[queue_index
].mrg_avg_pkt_len
;
2699 return sprintf(buf
, "%u\n",
2700 get_mergeable_buf_len(&vi
->rq
[queue_index
], avg
,
2701 SKB_DATA_ALIGN(headroom
+ tailroom
)));
2704 static struct rx_queue_attribute mergeable_rx_buffer_size_attribute
=
2705 __ATTR_RO(mergeable_rx_buffer_size
);
2707 static struct attribute
*virtio_net_mrg_rx_attrs
[] = {
2708 &mergeable_rx_buffer_size_attribute
.attr
,
2712 static const struct attribute_group virtio_net_mrg_rx_group
= {
2713 .name
= "virtio_net",
2714 .attrs
= virtio_net_mrg_rx_attrs
2718 static bool virtnet_fail_on_feature(struct virtio_device
*vdev
,
2720 const char *fname
, const char *dname
)
2722 if (!virtio_has_feature(vdev
, fbit
))
2725 dev_err(&vdev
->dev
, "device advertises feature %s but not %s",
2731 #define VIRTNET_FAIL_ON(vdev, fbit, dbit) \
2732 virtnet_fail_on_feature(vdev, fbit, #fbit, dbit)
2734 static bool virtnet_validate_features(struct virtio_device
*vdev
)
2736 if (!virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
) &&
2737 (VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_CTRL_RX
,
2738 "VIRTIO_NET_F_CTRL_VQ") ||
2739 VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_CTRL_VLAN
,
2740 "VIRTIO_NET_F_CTRL_VQ") ||
2741 VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_GUEST_ANNOUNCE
,
2742 "VIRTIO_NET_F_CTRL_VQ") ||
2743 VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_MQ
, "VIRTIO_NET_F_CTRL_VQ") ||
2744 VIRTNET_FAIL_ON(vdev
, VIRTIO_NET_F_CTRL_MAC_ADDR
,
2745 "VIRTIO_NET_F_CTRL_VQ"))) {
2752 #define MIN_MTU ETH_MIN_MTU
2753 #define MAX_MTU ETH_MAX_MTU
2755 static int virtnet_validate(struct virtio_device
*vdev
)
2757 if (!vdev
->config
->get
) {
2758 dev_err(&vdev
->dev
, "%s failure: config access disabled\n",
2763 if (!virtnet_validate_features(vdev
))
2766 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MTU
)) {
2767 int mtu
= virtio_cread16(vdev
,
2768 offsetof(struct virtio_net_config
,
2771 __virtio_clear_bit(vdev
, VIRTIO_NET_F_MTU
);
2777 static int virtnet_probe(struct virtio_device
*vdev
)
2779 int i
, err
= -ENOMEM
;
2780 struct net_device
*dev
;
2781 struct virtnet_info
*vi
;
2782 u16 max_queue_pairs
;
2785 /* Find if host supports multiqueue virtio_net device */
2786 err
= virtio_cread_feature(vdev
, VIRTIO_NET_F_MQ
,
2787 struct virtio_net_config
,
2788 max_virtqueue_pairs
, &max_queue_pairs
);
2790 /* We need at least 2 queue's */
2791 if (err
|| max_queue_pairs
< VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN
||
2792 max_queue_pairs
> VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX
||
2793 !virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
))
2794 max_queue_pairs
= 1;
2796 /* Allocate ourselves a network device with room for our info */
2797 dev
= alloc_etherdev_mq(sizeof(struct virtnet_info
), max_queue_pairs
);
2801 /* Set up network device as normal. */
2802 dev
->priv_flags
|= IFF_UNICAST_FLT
| IFF_LIVE_ADDR_CHANGE
;
2803 dev
->netdev_ops
= &virtnet_netdev
;
2804 dev
->features
= NETIF_F_HIGHDMA
;
2806 dev
->ethtool_ops
= &virtnet_ethtool_ops
;
2807 SET_NETDEV_DEV(dev
, &vdev
->dev
);
2809 /* Do we support "hardware" checksums? */
2810 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CSUM
)) {
2811 /* This opens up the world of extra features. */
2812 dev
->hw_features
|= NETIF_F_HW_CSUM
| NETIF_F_SG
;
2814 dev
->features
|= NETIF_F_HW_CSUM
| NETIF_F_SG
;
2816 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GSO
)) {
2817 dev
->hw_features
|= NETIF_F_TSO
2818 | NETIF_F_TSO_ECN
| NETIF_F_TSO6
;
2820 /* Individual feature bits: what can host handle? */
2821 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_TSO4
))
2822 dev
->hw_features
|= NETIF_F_TSO
;
2823 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_TSO6
))
2824 dev
->hw_features
|= NETIF_F_TSO6
;
2825 if (virtio_has_feature(vdev
, VIRTIO_NET_F_HOST_ECN
))
2826 dev
->hw_features
|= NETIF_F_TSO_ECN
;
2828 dev
->features
|= NETIF_F_GSO_ROBUST
;
2831 dev
->features
|= dev
->hw_features
& NETIF_F_ALL_TSO
;
2832 /* (!csum && gso) case will be fixed by register_netdev() */
2834 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_CSUM
))
2835 dev
->features
|= NETIF_F_RXCSUM
;
2837 dev
->vlan_features
= dev
->features
;
2839 /* MTU range: 68 - 65535 */
2840 dev
->min_mtu
= MIN_MTU
;
2841 dev
->max_mtu
= MAX_MTU
;
2843 /* Configuration may specify what MAC to use. Otherwise random. */
2844 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MAC
))
2845 virtio_cread_bytes(vdev
,
2846 offsetof(struct virtio_net_config
, mac
),
2847 dev
->dev_addr
, dev
->addr_len
);
2849 eth_hw_addr_random(dev
);
2851 /* Set up our device-specific information */
2852 vi
= netdev_priv(dev
);
2857 INIT_WORK(&vi
->config_work
, virtnet_config_changed_work
);
2859 /* If we can receive ANY GSO packets, we must allocate large ones. */
2860 if (virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_TSO4
) ||
2861 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_TSO6
) ||
2862 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_ECN
) ||
2863 virtio_has_feature(vdev
, VIRTIO_NET_F_GUEST_UFO
))
2864 vi
->big_packets
= true;
2866 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MRG_RXBUF
))
2867 vi
->mergeable_rx_bufs
= true;
2869 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MRG_RXBUF
) ||
2870 virtio_has_feature(vdev
, VIRTIO_F_VERSION_1
))
2871 vi
->hdr_len
= sizeof(struct virtio_net_hdr_mrg_rxbuf
);
2873 vi
->hdr_len
= sizeof(struct virtio_net_hdr
);
2875 if (virtio_has_feature(vdev
, VIRTIO_F_ANY_LAYOUT
) ||
2876 virtio_has_feature(vdev
, VIRTIO_F_VERSION_1
))
2877 vi
->any_header_sg
= true;
2879 if (virtio_has_feature(vdev
, VIRTIO_NET_F_CTRL_VQ
))
2882 if (virtio_has_feature(vdev
, VIRTIO_NET_F_MTU
)) {
2883 mtu
= virtio_cread16(vdev
,
2884 offsetof(struct virtio_net_config
,
2886 if (mtu
< dev
->min_mtu
) {
2887 /* Should never trigger: MTU was previously validated
2888 * in virtnet_validate.
2890 dev_err(&vdev
->dev
, "device MTU appears to have changed "
2891 "it is now %d < %d", mtu
, dev
->min_mtu
);
2898 /* TODO: size buffers correctly in this case. */
2899 if (dev
->mtu
> ETH_DATA_LEN
)
2900 vi
->big_packets
= true;
2903 if (vi
->any_header_sg
)
2904 dev
->needed_headroom
= vi
->hdr_len
;
2906 /* Enable multiqueue by default */
2907 if (num_online_cpus() >= max_queue_pairs
)
2908 vi
->curr_queue_pairs
= max_queue_pairs
;
2910 vi
->curr_queue_pairs
= num_online_cpus();
2911 vi
->max_queue_pairs
= max_queue_pairs
;
2913 /* Allocate/initialize the rx/tx queues, and invoke find_vqs */
2919 if (vi
->mergeable_rx_bufs
)
2920 dev
->sysfs_rx_queue_group
= &virtio_net_mrg_rx_group
;
2922 netif_set_real_num_tx_queues(dev
, vi
->curr_queue_pairs
);
2923 netif_set_real_num_rx_queues(dev
, vi
->curr_queue_pairs
);
2925 virtnet_init_settings(dev
);
2927 if (virtio_has_feature(vdev
, VIRTIO_NET_F_STANDBY
)) {
2928 vi
->failover
= net_failover_create(vi
->dev
);
2929 if (IS_ERR(vi
->failover
)) {
2930 err
= PTR_ERR(vi
->failover
);
2935 err
= register_netdev(dev
);
2937 pr_debug("virtio_net: registering device failed\n");
2941 virtio_device_ready(vdev
);
2943 err
= virtnet_cpu_notif_add(vi
);
2945 pr_debug("virtio_net: registering cpu notifier failed\n");
2946 goto free_unregister_netdev
;
2949 virtnet_set_queues(vi
, vi
->curr_queue_pairs
);
2951 /* Assume link up if device can't report link status,
2952 otherwise get link status from config. */
2953 netif_carrier_off(dev
);
2954 if (virtio_has_feature(vi
->vdev
, VIRTIO_NET_F_STATUS
)) {
2955 schedule_work(&vi
->config_work
);
2957 vi
->status
= VIRTIO_NET_S_LINK_UP
;
2958 virtnet_update_settings(vi
);
2959 netif_carrier_on(dev
);
2962 for (i
= 0; i
< ARRAY_SIZE(guest_offloads
); i
++)
2963 if (virtio_has_feature(vi
->vdev
, guest_offloads
[i
]))
2964 set_bit(guest_offloads
[i
], &vi
->guest_offloads
);
2966 pr_debug("virtnet: registered device %s with %d RX and TX vq's\n",
2967 dev
->name
, max_queue_pairs
);
2971 free_unregister_netdev
:
2972 vi
->vdev
->config
->reset(vdev
);
2974 unregister_netdev(dev
);
2976 net_failover_destroy(vi
->failover
);
2978 cancel_delayed_work_sync(&vi
->refill
);
2979 free_receive_page_frags(vi
);
2980 virtnet_del_vqs(vi
);
2986 static void remove_vq_common(struct virtnet_info
*vi
)
2988 vi
->vdev
->config
->reset(vi
->vdev
);
2990 /* Free unused buffers in both send and recv, if any. */
2991 free_unused_bufs(vi
);
2993 free_receive_bufs(vi
);
2995 free_receive_page_frags(vi
);
2997 virtnet_del_vqs(vi
);
3000 static void virtnet_remove(struct virtio_device
*vdev
)
3002 struct virtnet_info
*vi
= vdev
->priv
;
3004 virtnet_cpu_notif_remove(vi
);
3006 /* Make sure no work handler is accessing the device. */
3007 flush_work(&vi
->config_work
);
3009 unregister_netdev(vi
->dev
);
3011 net_failover_destroy(vi
->failover
);
3013 remove_vq_common(vi
);
3015 free_netdev(vi
->dev
);
3018 static __maybe_unused
int virtnet_freeze(struct virtio_device
*vdev
)
3020 struct virtnet_info
*vi
= vdev
->priv
;
3022 virtnet_cpu_notif_remove(vi
);
3023 virtnet_freeze_down(vdev
);
3024 remove_vq_common(vi
);
3029 static __maybe_unused
int virtnet_restore(struct virtio_device
*vdev
)
3031 struct virtnet_info
*vi
= vdev
->priv
;
3034 err
= virtnet_restore_up(vdev
);
3037 virtnet_set_queues(vi
, vi
->curr_queue_pairs
);
3039 err
= virtnet_cpu_notif_add(vi
);
3046 static struct virtio_device_id id_table
[] = {
3047 { VIRTIO_ID_NET
, VIRTIO_DEV_ANY_ID
},
3051 #define VIRTNET_FEATURES \
3052 VIRTIO_NET_F_CSUM, VIRTIO_NET_F_GUEST_CSUM, \
3054 VIRTIO_NET_F_HOST_TSO4, VIRTIO_NET_F_HOST_UFO, VIRTIO_NET_F_HOST_TSO6, \
3055 VIRTIO_NET_F_HOST_ECN, VIRTIO_NET_F_GUEST_TSO4, VIRTIO_NET_F_GUEST_TSO6, \
3056 VIRTIO_NET_F_GUEST_ECN, VIRTIO_NET_F_GUEST_UFO, \
3057 VIRTIO_NET_F_MRG_RXBUF, VIRTIO_NET_F_STATUS, VIRTIO_NET_F_CTRL_VQ, \
3058 VIRTIO_NET_F_CTRL_RX, VIRTIO_NET_F_CTRL_VLAN, \
3059 VIRTIO_NET_F_GUEST_ANNOUNCE, VIRTIO_NET_F_MQ, \
3060 VIRTIO_NET_F_CTRL_MAC_ADDR, \
3061 VIRTIO_NET_F_MTU, VIRTIO_NET_F_CTRL_GUEST_OFFLOADS, \
3062 VIRTIO_NET_F_SPEED_DUPLEX, VIRTIO_NET_F_STANDBY
3064 static unsigned int features
[] = {
3068 static unsigned int features_legacy
[] = {
3071 VIRTIO_F_ANY_LAYOUT
,
3074 static struct virtio_driver virtio_net_driver
= {
3075 .feature_table
= features
,
3076 .feature_table_size
= ARRAY_SIZE(features
),
3077 .feature_table_legacy
= features_legacy
,
3078 .feature_table_size_legacy
= ARRAY_SIZE(features_legacy
),
3079 .driver
.name
= KBUILD_MODNAME
,
3080 .driver
.owner
= THIS_MODULE
,
3081 .id_table
= id_table
,
3082 .validate
= virtnet_validate
,
3083 .probe
= virtnet_probe
,
3084 .remove
= virtnet_remove
,
3085 .config_changed
= virtnet_config_changed
,
3086 #ifdef CONFIG_PM_SLEEP
3087 .freeze
= virtnet_freeze
,
3088 .restore
= virtnet_restore
,
3092 static __init
int virtio_net_driver_init(void)
3096 ret
= cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN
, "virtio/net:online",
3098 virtnet_cpu_down_prep
);
3101 virtionet_online
= ret
;
3102 ret
= cpuhp_setup_state_multi(CPUHP_VIRT_NET_DEAD
, "virtio/net:dead",
3103 NULL
, virtnet_cpu_dead
);
3107 ret
= register_virtio_driver(&virtio_net_driver
);
3112 cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD
);
3114 cpuhp_remove_multi_state(virtionet_online
);
3118 module_init(virtio_net_driver_init
);
3120 static __exit
void virtio_net_driver_exit(void)
3122 unregister_virtio_driver(&virtio_net_driver
);
3123 cpuhp_remove_multi_state(CPUHP_VIRT_NET_DEAD
);
3124 cpuhp_remove_multi_state(virtionet_online
);
3126 module_exit(virtio_net_driver_exit
);
3128 MODULE_DEVICE_TABLE(virtio
, id_table
);
3129 MODULE_DESCRIPTION("Virtio network driver");
3130 MODULE_LICENSE("GPL");