2 * Virtual network driver for conversing with remote driver backends.
4 * Copyright (c) 2002-2005, K A Fraser
5 * Copyright (c) 2005, XenSource Ltd
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License version 2
9 * as published by the Free Software Foundation; or, when distributed
10 * separately from the Linux kernel or incorporated into other
11 * software packages, subject to the following license:
13 * Permission is hereby granted, free of charge, to any person obtaining a copy
14 * of this source file (the "Software"), to deal in the Software without
15 * restriction, including without limitation the rights to use, copy, modify,
16 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
17 * and to permit persons to whom the Software is furnished to do so, subject to
18 * the following conditions:
20 * The above copyright notice and this permission notice shall be included in
21 * all copies or substantial portions of the Software.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
24 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
25 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
26 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
27 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
28 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
32 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
34 #include <linux/module.h>
35 #include <linux/kernel.h>
36 #include <linux/netdevice.h>
37 #include <linux/etherdevice.h>
38 #include <linux/skbuff.h>
39 #include <linux/ethtool.h>
40 #include <linux/if_ether.h>
42 #include <linux/udp.h>
43 #include <linux/moduleparam.h>
45 #include <linux/slab.h>
47 #include <linux/bpf.h>
48 #include <net/page_pool.h>
49 #include <linux/bpf_trace.h>
52 #include <xen/xenbus.h>
53 #include <xen/events.h>
55 #include <xen/platform_pci.h>
56 #include <xen/grant_table.h>
58 #include <xen/interface/io/netif.h>
59 #include <xen/interface/memory.h>
60 #include <xen/interface/grant_table.h>
62 /* Module parameters */
63 #define MAX_QUEUES_DEFAULT 8
64 static unsigned int xennet_max_queues;
65 module_param_named(max_queues, xennet_max_queues, uint, 0644);
66 MODULE_PARM_DESC(max_queues,
67 "Maximum number of queues per virtual interface");
69 #define XENNET_TIMEOUT (5 * HZ)
71 static const struct ethtool_ops xennet_ethtool_ops;
77 #define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb))
79 #define RX_COPY_THRESHOLD 256
81 #define GRANT_INVALID_REF 0
83 #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, XEN_PAGE_SIZE)
84 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, XEN_PAGE_SIZE)
86 /* Minimum number of Rx slots (includes slot for GSO metadata). */
87 #define NET_RX_SLOTS_MIN (XEN_NETIF_NR_SLOTS_MIN + 1)
89 /* Queue name is interface name with "-qNNN" appended */
90 #define QUEUE_NAME_SIZE (IFNAMSIZ + 6)
92 /* IRQ name is queue name with "-tx" or "-rx" appended */
93 #define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3)
95 static DECLARE_WAIT_QUEUE_HEAD(module_wq);
97 struct netfront_stats {
100 struct u64_stats_sync syncp;
103 struct netfront_info;
105 struct netfront_queue {
106 unsigned int id; /* Queue ID, 0-based */
107 char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */
108 struct netfront_info *info;
110 struct bpf_prog __rcu *xdp_prog;
112 struct napi_struct napi;
114 /* Split event channels support, tx_* == rx_* when using
115 * single event channel.
117 unsigned int tx_evtchn, rx_evtchn;
118 unsigned int tx_irq, rx_irq;
119 /* Only used when split event channels support is enabled */
120 char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */
121 char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */
124 struct xen_netif_tx_front_ring tx;
128 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
129 * are linked from tx_skb_freelist through skb_entry.link.
131 * NB. Freelist index entries are always going to be less than
132 * PAGE_OFFSET, whereas pointers to skbs will always be equal or
133 * greater than PAGE_OFFSET: we use this property to distinguish
139 } tx_skbs[NET_TX_RING_SIZE];
140 grant_ref_t gref_tx_head;
141 grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
142 struct page *grant_tx_page[NET_TX_RING_SIZE];
143 unsigned tx_skb_freelist;
145 spinlock_t rx_lock ____cacheline_aligned_in_smp;
146 struct xen_netif_rx_front_ring rx;
149 struct timer_list rx_refill_timer;
151 struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
152 grant_ref_t gref_rx_head;
153 grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
155 struct page_pool *page_pool;
156 struct xdp_rxq_info xdp_rxq;
159 struct netfront_info {
160 struct list_head list;
161 struct net_device *netdev;
163 struct xenbus_device *xbdev;
165 /* Multi-queue support */
166 struct netfront_queue *queues;
169 struct netfront_stats __percpu *rx_stats;
170 struct netfront_stats __percpu *tx_stats;
173 bool netback_has_xdp_headroom;
174 bool netfront_xdp_enabled;
176 atomic_t rx_gso_checksum_fixup;
179 struct netfront_rx_info {
180 struct xen_netif_rx_response rx;
181 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
184 static void skb_entry_set_link(union skb_entry *list, unsigned short id)
189 static int skb_entry_is_link(const union skb_entry *list)
191 BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link));
192 return (unsigned long)list->skb < PAGE_OFFSET;
196 * Access macros for acquiring freeing slots in tx_skbs[].
199 static void add_id_to_freelist(unsigned *head, union skb_entry *list,
202 skb_entry_set_link(&list[id], *head);
206 static unsigned short get_id_from_freelist(unsigned *head,
207 union skb_entry *list)
209 unsigned int id = *head;
210 *head = list[id].link;
214 static int xennet_rxidx(RING_IDX idx)
216 return idx & (NET_RX_RING_SIZE - 1);
219 static struct sk_buff *xennet_get_rx_skb(struct netfront_queue *queue,
222 int i = xennet_rxidx(ri);
223 struct sk_buff *skb = queue->rx_skbs[i];
224 queue->rx_skbs[i] = NULL;
228 static grant_ref_t xennet_get_rx_ref(struct netfront_queue *queue,
231 int i = xennet_rxidx(ri);
232 grant_ref_t ref = queue->grant_rx_ref[i];
233 queue->grant_rx_ref[i] = GRANT_INVALID_REF;
238 static const struct attribute_group xennet_dev_group;
241 static bool xennet_can_sg(struct net_device *dev)
243 return dev->features & NETIF_F_SG;
247 static void rx_refill_timeout(struct timer_list *t)
249 struct netfront_queue *queue = from_timer(queue, t, rx_refill_timer);
250 napi_schedule(&queue->napi);
253 static int netfront_tx_slot_available(struct netfront_queue *queue)
255 return (queue->tx.req_prod_pvt - queue->tx.rsp_cons) <
256 (NET_TX_RING_SIZE - XEN_NETIF_NR_SLOTS_MIN - 1);
259 static void xennet_maybe_wake_tx(struct netfront_queue *queue)
261 struct net_device *dev = queue->info->netdev;
262 struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, queue->id);
264 if (unlikely(netif_tx_queue_stopped(dev_queue)) &&
265 netfront_tx_slot_available(queue) &&
266 likely(netif_running(dev)))
267 netif_tx_wake_queue(netdev_get_tx_queue(dev, queue->id));
271 static struct sk_buff *xennet_alloc_one_rx_buffer(struct netfront_queue *queue)
276 skb = __netdev_alloc_skb(queue->info->netdev,
277 RX_COPY_THRESHOLD + NET_IP_ALIGN,
278 GFP_ATOMIC | __GFP_NOWARN);
282 page = page_pool_dev_alloc_pages(queue->page_pool);
283 if (unlikely(!page)) {
287 skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE);
289 /* Align ip header to a 16 bytes boundary */
290 skb_reserve(skb, NET_IP_ALIGN);
291 skb->dev = queue->info->netdev;
297 static void xennet_alloc_rx_buffers(struct netfront_queue *queue)
299 RING_IDX req_prod = queue->rx.req_prod_pvt;
303 if (unlikely(!netif_carrier_ok(queue->info->netdev)))
306 for (req_prod = queue->rx.req_prod_pvt;
307 req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE;
313 struct xen_netif_rx_request *req;
315 skb = xennet_alloc_one_rx_buffer(queue);
321 id = xennet_rxidx(req_prod);
323 BUG_ON(queue->rx_skbs[id]);
324 queue->rx_skbs[id] = skb;
326 ref = gnttab_claim_grant_reference(&queue->gref_rx_head);
327 WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
328 queue->grant_rx_ref[id] = ref;
330 page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
332 req = RING_GET_REQUEST(&queue->rx, req_prod);
333 gnttab_page_grant_foreign_access_ref_one(ref,
334 queue->info->xbdev->otherend_id,
341 queue->rx.req_prod_pvt = req_prod;
343 /* Try again later if there are not enough requests or skb allocation
345 * Enough requests is quantified as the sum of newly created slots and
346 * the unconsumed slots at the backend.
348 if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN ||
350 mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10));
354 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify);
356 notify_remote_via_irq(queue->rx_irq);
359 static int xennet_open(struct net_device *dev)
361 struct netfront_info *np = netdev_priv(dev);
362 unsigned int num_queues = dev->real_num_tx_queues;
364 struct netfront_queue *queue = NULL;
369 for (i = 0; i < num_queues; ++i) {
370 queue = &np->queues[i];
371 napi_enable(&queue->napi);
373 spin_lock_bh(&queue->rx_lock);
374 if (netif_carrier_ok(dev)) {
375 xennet_alloc_rx_buffers(queue);
376 queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1;
377 if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))
378 napi_schedule(&queue->napi);
380 spin_unlock_bh(&queue->rx_lock);
383 netif_tx_start_all_queues(dev);
388 static void xennet_tx_buf_gc(struct netfront_queue *queue)
395 BUG_ON(!netif_carrier_ok(queue->info->netdev));
398 prod = queue->tx.sring->rsp_prod;
399 rmb(); /* Ensure we see responses up to 'rp'. */
401 for (cons = queue->tx.rsp_cons; cons != prod; cons++) {
402 struct xen_netif_tx_response *txrsp;
404 txrsp = RING_GET_RESPONSE(&queue->tx, cons);
405 if (txrsp->status == XEN_NETIF_RSP_NULL)
409 skb = queue->tx_skbs[id].skb;
410 if (unlikely(gnttab_query_foreign_access(
411 queue->grant_tx_ref[id]) != 0)) {
412 pr_alert("%s: warning -- grant still in use by backend domain\n",
416 gnttab_end_foreign_access_ref(
417 queue->grant_tx_ref[id], GNTMAP_readonly);
418 gnttab_release_grant_reference(
419 &queue->gref_tx_head, queue->grant_tx_ref[id]);
420 queue->grant_tx_ref[id] = GRANT_INVALID_REF;
421 queue->grant_tx_page[id] = NULL;
422 add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, id);
423 dev_kfree_skb_irq(skb);
426 queue->tx.rsp_cons = prod;
428 RING_FINAL_CHECK_FOR_RESPONSES(&queue->tx, more_to_do);
429 } while (more_to_do);
431 xennet_maybe_wake_tx(queue);
434 struct xennet_gnttab_make_txreq {
435 struct netfront_queue *queue;
438 struct xen_netif_tx_request *tx; /* Last request */
442 static void xennet_tx_setup_grant(unsigned long gfn, unsigned int offset,
443 unsigned int len, void *data)
445 struct xennet_gnttab_make_txreq *info = data;
447 struct xen_netif_tx_request *tx;
449 /* convenient aliases */
450 struct page *page = info->page;
451 struct netfront_queue *queue = info->queue;
452 struct sk_buff *skb = info->skb;
454 id = get_id_from_freelist(&queue->tx_skb_freelist, queue->tx_skbs);
455 tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
456 ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
457 WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
459 gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id,
460 gfn, GNTMAP_readonly);
462 queue->tx_skbs[id].skb = skb;
463 queue->grant_tx_page[id] = page;
464 queue->grant_tx_ref[id] = ref;
473 info->size += tx->size;
476 static struct xen_netif_tx_request *xennet_make_first_txreq(
477 struct netfront_queue *queue, struct sk_buff *skb,
478 struct page *page, unsigned int offset, unsigned int len)
480 struct xennet_gnttab_make_txreq info = {
487 gnttab_for_one_grant(page, offset, len, xennet_tx_setup_grant, &info);
492 static void xennet_make_one_txreq(unsigned long gfn, unsigned int offset,
493 unsigned int len, void *data)
495 struct xennet_gnttab_make_txreq *info = data;
497 info->tx->flags |= XEN_NETTXF_more_data;
499 xennet_tx_setup_grant(gfn, offset, len, data);
502 static struct xen_netif_tx_request *xennet_make_txreqs(
503 struct netfront_queue *queue, struct xen_netif_tx_request *tx,
504 struct sk_buff *skb, struct page *page,
505 unsigned int offset, unsigned int len)
507 struct xennet_gnttab_make_txreq info = {
513 /* Skip unused frames from start of page */
514 page += offset >> PAGE_SHIFT;
515 offset &= ~PAGE_MASK;
521 gnttab_foreach_grant_in_range(page, offset, len,
522 xennet_make_one_txreq,
534 * Count how many ring slots are required to send this skb. Each frag
535 * might be a compound page.
537 static int xennet_count_skb_slots(struct sk_buff *skb)
539 int i, frags = skb_shinfo(skb)->nr_frags;
542 slots = gnttab_count_grant(offset_in_page(skb->data),
545 for (i = 0; i < frags; i++) {
546 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
547 unsigned long size = skb_frag_size(frag);
548 unsigned long offset = skb_frag_off(frag);
550 /* Skip unused frames from start of page */
551 offset &= ~PAGE_MASK;
553 slots += gnttab_count_grant(offset, size);
559 static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb,
560 struct net_device *sb_dev)
562 unsigned int num_queues = dev->real_num_tx_queues;
566 /* First, check if there is only one queue */
567 if (num_queues == 1) {
570 hash = skb_get_hash(skb);
571 queue_idx = hash % num_queues;
577 static int xennet_xdp_xmit_one(struct net_device *dev,
578 struct netfront_queue *queue,
579 struct xdp_frame *xdpf)
581 struct netfront_info *np = netdev_priv(dev);
582 struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
585 xennet_make_first_txreq(queue, NULL,
586 virt_to_page(xdpf->data),
587 offset_in_page(xdpf->data),
590 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
592 notify_remote_via_irq(queue->tx_irq);
594 u64_stats_update_begin(&tx_stats->syncp);
595 tx_stats->bytes += xdpf->len;
597 u64_stats_update_end(&tx_stats->syncp);
599 xennet_tx_buf_gc(queue);
604 static int xennet_xdp_xmit(struct net_device *dev, int n,
605 struct xdp_frame **frames, u32 flags)
607 unsigned int num_queues = dev->real_num_tx_queues;
608 struct netfront_info *np = netdev_priv(dev);
609 struct netfront_queue *queue = NULL;
610 unsigned long irq_flags;
614 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
617 queue = &np->queues[smp_processor_id() % num_queues];
619 spin_lock_irqsave(&queue->tx_lock, irq_flags);
620 for (i = 0; i < n; i++) {
621 struct xdp_frame *xdpf = frames[i];
625 err = xennet_xdp_xmit_one(dev, queue, xdpf);
627 xdp_return_frame_rx_napi(xdpf);
631 spin_unlock_irqrestore(&queue->tx_lock, irq_flags);
637 #define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1)
639 static netdev_tx_t xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
641 struct netfront_info *np = netdev_priv(dev);
642 struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
643 struct xen_netif_tx_request *tx, *first_tx;
651 struct netfront_queue *queue = NULL;
652 unsigned int num_queues = dev->real_num_tx_queues;
654 struct sk_buff *nskb;
656 /* Drop the packet if no queues are set up */
659 /* Determine which queue to transmit this SKB on */
660 queue_index = skb_get_queue_mapping(skb);
661 queue = &np->queues[queue_index];
663 /* If skb->len is too big for wire format, drop skb and alert
664 * user about misconfiguration.
666 if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
667 net_alert_ratelimited(
668 "xennet: skb->len = %u, too big for wire format\n",
673 slots = xennet_count_skb_slots(skb);
674 if (unlikely(slots > MAX_XEN_SKB_FRAGS + 1)) {
675 net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n",
677 if (skb_linearize(skb))
681 page = virt_to_page(skb->data);
682 offset = offset_in_page(skb->data);
684 /* The first req should be at least ETH_HLEN size or the packet will be
685 * dropped by netback.
687 if (unlikely(PAGE_SIZE - offset < ETH_HLEN)) {
688 nskb = skb_copy(skb, GFP_ATOMIC);
691 dev_consume_skb_any(skb);
693 page = virt_to_page(skb->data);
694 offset = offset_in_page(skb->data);
697 len = skb_headlen(skb);
699 spin_lock_irqsave(&queue->tx_lock, flags);
701 if (unlikely(!netif_carrier_ok(dev) ||
702 (slots > 1 && !xennet_can_sg(dev)) ||
703 netif_needs_gso(skb, netif_skb_features(skb)))) {
704 spin_unlock_irqrestore(&queue->tx_lock, flags);
708 /* First request for the linear area. */
709 first_tx = tx = xennet_make_first_txreq(queue, skb,
712 if (offset == PAGE_SIZE) {
718 if (skb->ip_summed == CHECKSUM_PARTIAL)
720 tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
721 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
722 /* remote but checksummed. */
723 tx->flags |= XEN_NETTXF_data_validated;
725 /* Optional extra info after the first request. */
726 if (skb_shinfo(skb)->gso_size) {
727 struct xen_netif_extra_info *gso;
729 gso = (struct xen_netif_extra_info *)
730 RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
732 tx->flags |= XEN_NETTXF_extra_info;
734 gso->u.gso.size = skb_shinfo(skb)->gso_size;
735 gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ?
736 XEN_NETIF_GSO_TYPE_TCPV6 :
737 XEN_NETIF_GSO_TYPE_TCPV4;
739 gso->u.gso.features = 0;
741 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
745 /* Requests for the rest of the linear area. */
746 tx = xennet_make_txreqs(queue, tx, skb, page, offset, len);
748 /* Requests for all the frags. */
749 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
750 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
751 tx = xennet_make_txreqs(queue, tx, skb, skb_frag_page(frag),
753 skb_frag_size(frag));
756 /* First request has the packet length. */
757 first_tx->size = skb->len;
759 /* timestamp packet in software */
760 skb_tx_timestamp(skb);
762 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
764 notify_remote_via_irq(queue->tx_irq);
766 u64_stats_update_begin(&tx_stats->syncp);
767 tx_stats->bytes += skb->len;
769 u64_stats_update_end(&tx_stats->syncp);
771 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
772 xennet_tx_buf_gc(queue);
774 if (!netfront_tx_slot_available(queue))
775 netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
777 spin_unlock_irqrestore(&queue->tx_lock, flags);
782 dev->stats.tx_dropped++;
783 dev_kfree_skb_any(skb);
787 static int xennet_close(struct net_device *dev)
789 struct netfront_info *np = netdev_priv(dev);
790 unsigned int num_queues = dev->real_num_tx_queues;
792 struct netfront_queue *queue;
793 netif_tx_stop_all_queues(np->netdev);
794 for (i = 0; i < num_queues; ++i) {
795 queue = &np->queues[i];
796 napi_disable(&queue->napi);
801 static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb,
804 int new = xennet_rxidx(queue->rx.req_prod_pvt);
806 BUG_ON(queue->rx_skbs[new]);
807 queue->rx_skbs[new] = skb;
808 queue->grant_rx_ref[new] = ref;
809 RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new;
810 RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref;
811 queue->rx.req_prod_pvt++;
814 static int xennet_get_extras(struct netfront_queue *queue,
815 struct xen_netif_extra_info *extras,
819 struct xen_netif_extra_info *extra;
820 struct device *dev = &queue->info->netdev->dev;
821 RING_IDX cons = queue->rx.rsp_cons;
828 if (unlikely(cons + 1 == rp)) {
830 dev_warn(dev, "Missing extra info\n");
835 extra = (struct xen_netif_extra_info *)
836 RING_GET_RESPONSE(&queue->rx, ++cons);
838 if (unlikely(!extra->type ||
839 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
841 dev_warn(dev, "Invalid extra type: %d\n",
845 memcpy(&extras[extra->type - 1], extra,
849 skb = xennet_get_rx_skb(queue, cons);
850 ref = xennet_get_rx_ref(queue, cons);
851 xennet_move_rx_slot(queue, skb, ref);
852 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
854 queue->rx.rsp_cons = cons;
858 static u32 xennet_run_xdp(struct netfront_queue *queue, struct page *pdata,
859 struct xen_netif_rx_response *rx, struct bpf_prog *prog,
860 struct xdp_buff *xdp, bool *need_xdp_flush)
862 struct xdp_frame *xdpf;
863 u32 len = rx->status;
867 xdp->data_hard_start = page_address(pdata);
868 xdp->data = xdp->data_hard_start + XDP_PACKET_HEADROOM;
869 xdp_set_data_meta_invalid(xdp);
870 xdp->data_end = xdp->data + len;
871 xdp->rxq = &queue->xdp_rxq;
872 xdp->frame_sz = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM;
874 act = bpf_prog_run_xdp(prog, xdp);
878 xdpf = xdp_convert_buff_to_frame(xdp);
879 err = xennet_xdp_xmit(queue->info->netdev, 1, &xdpf, 0);
880 if (unlikely(err < 0))
881 trace_xdp_exception(queue->info->netdev, prog, act);
885 err = xdp_do_redirect(queue->info->netdev, xdp, prog);
886 *need_xdp_flush = true;
888 trace_xdp_exception(queue->info->netdev, prog, act);
895 trace_xdp_exception(queue->info->netdev, prog, act);
899 bpf_warn_invalid_xdp_action(act);
905 static int xennet_get_responses(struct netfront_queue *queue,
906 struct netfront_rx_info *rinfo, RING_IDX rp,
907 struct sk_buff_head *list,
908 bool *need_xdp_flush)
910 struct xen_netif_rx_response *rx = &rinfo->rx;
911 int max = XEN_NETIF_NR_SLOTS_MIN + (rx->status <= RX_COPY_THRESHOLD);
912 RING_IDX cons = queue->rx.rsp_cons;
913 struct sk_buff *skb = xennet_get_rx_skb(queue, cons);
914 struct xen_netif_extra_info *extras = rinfo->extras;
915 grant_ref_t ref = xennet_get_rx_ref(queue, cons);
916 struct device *dev = &queue->info->netdev->dev;
917 struct bpf_prog *xdp_prog;
924 if (rx->flags & XEN_NETRXF_extra_info) {
925 err = xennet_get_extras(queue, extras, rp);
927 if (extras[XEN_NETIF_EXTRA_TYPE_XDP - 1].type) {
928 struct xen_netif_extra_info *xdp;
930 xdp = &extras[XEN_NETIF_EXTRA_TYPE_XDP - 1];
931 rx->offset = xdp->u.xdp.headroom;
934 cons = queue->rx.rsp_cons;
938 if (unlikely(rx->status < 0 ||
939 rx->offset + rx->status > XEN_PAGE_SIZE)) {
941 dev_warn(dev, "rx->offset: %u, size: %d\n",
942 rx->offset, rx->status);
943 xennet_move_rx_slot(queue, skb, ref);
949 * This definitely indicates a bug, either in this driver or in
950 * the backend driver. In future this should flag the bad
951 * situation to the system controller to reboot the backend.
953 if (ref == GRANT_INVALID_REF) {
955 dev_warn(dev, "Bad rx response id %d.\n",
961 ret = gnttab_end_foreign_access_ref(ref, 0);
964 gnttab_release_grant_reference(&queue->gref_rx_head, ref);
967 xdp_prog = rcu_dereference(queue->xdp_prog);
969 if (!(rx->flags & XEN_NETRXF_more_data)) {
970 /* currently only a single page contains data */
971 verdict = xennet_run_xdp(queue,
972 skb_frag_page(&skb_shinfo(skb)->frags[0]),
973 rx, xdp_prog, &xdp, need_xdp_flush);
974 if (verdict != XDP_PASS)
983 __skb_queue_tail(list, skb);
984 if (!(rx->flags & XEN_NETRXF_more_data))
987 if (cons + slots == rp) {
989 dev_warn(dev, "Need more slots\n");
994 rx = RING_GET_RESPONSE(&queue->rx, cons + slots);
995 skb = xennet_get_rx_skb(queue, cons + slots);
996 ref = xennet_get_rx_ref(queue, cons + slots);
1000 if (unlikely(slots > max)) {
1001 if (net_ratelimit())
1002 dev_warn(dev, "Too many slots\n");
1007 queue->rx.rsp_cons = cons + slots;
1012 static int xennet_set_skb_gso(struct sk_buff *skb,
1013 struct xen_netif_extra_info *gso)
1015 if (!gso->u.gso.size) {
1016 if (net_ratelimit())
1017 pr_warn("GSO size must not be zero\n");
1021 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 &&
1022 gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) {
1023 if (net_ratelimit())
1024 pr_warn("Bad GSO type %d\n", gso->u.gso.type);
1028 skb_shinfo(skb)->gso_size = gso->u.gso.size;
1029 skb_shinfo(skb)->gso_type =
1030 (gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ?
1034 /* Header must be checked, and gso_segs computed. */
1035 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1036 skb_shinfo(skb)->gso_segs = 0;
1041 static int xennet_fill_frags(struct netfront_queue *queue,
1042 struct sk_buff *skb,
1043 struct sk_buff_head *list)
1045 RING_IDX cons = queue->rx.rsp_cons;
1046 struct sk_buff *nskb;
1048 while ((nskb = __skb_dequeue(list))) {
1049 struct xen_netif_rx_response *rx =
1050 RING_GET_RESPONSE(&queue->rx, ++cons);
1051 skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
1053 if (skb_shinfo(skb)->nr_frags == MAX_SKB_FRAGS) {
1054 unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
1056 BUG_ON(pull_to < skb_headlen(skb));
1057 __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
1059 if (unlikely(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS)) {
1060 queue->rx.rsp_cons = ++cons + skb_queue_len(list);
1065 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags,
1066 skb_frag_page(nfrag),
1067 rx->offset, rx->status, PAGE_SIZE);
1069 skb_shinfo(nskb)->nr_frags = 0;
1073 queue->rx.rsp_cons = cons;
1078 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
1080 bool recalculate_partial_csum = false;
1083 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
1084 * peers can fail to set NETRXF_csum_blank when sending a GSO
1085 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
1086 * recalculate the partial checksum.
1088 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
1089 struct netfront_info *np = netdev_priv(dev);
1090 atomic_inc(&np->rx_gso_checksum_fixup);
1091 skb->ip_summed = CHECKSUM_PARTIAL;
1092 recalculate_partial_csum = true;
1095 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
1096 if (skb->ip_summed != CHECKSUM_PARTIAL)
1099 return skb_checksum_setup(skb, recalculate_partial_csum);
1102 static int handle_incoming_queue(struct netfront_queue *queue,
1103 struct sk_buff_head *rxq)
1105 struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats);
1106 int packets_dropped = 0;
1107 struct sk_buff *skb;
1109 while ((skb = __skb_dequeue(rxq)) != NULL) {
1110 int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
1112 if (pull_to > skb_headlen(skb))
1113 __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
1115 /* Ethernet work: Delayed to here as it peeks the header. */
1116 skb->protocol = eth_type_trans(skb, queue->info->netdev);
1117 skb_reset_network_header(skb);
1119 if (checksum_setup(queue->info->netdev, skb)) {
1122 queue->info->netdev->stats.rx_errors++;
1126 u64_stats_update_begin(&rx_stats->syncp);
1127 rx_stats->packets++;
1128 rx_stats->bytes += skb->len;
1129 u64_stats_update_end(&rx_stats->syncp);
1132 napi_gro_receive(&queue->napi, skb);
1135 return packets_dropped;
1138 static int xennet_poll(struct napi_struct *napi, int budget)
1140 struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi);
1141 struct net_device *dev = queue->info->netdev;
1142 struct sk_buff *skb;
1143 struct netfront_rx_info rinfo;
1144 struct xen_netif_rx_response *rx = &rinfo.rx;
1145 struct xen_netif_extra_info *extras = rinfo.extras;
1148 struct sk_buff_head rxq;
1149 struct sk_buff_head errq;
1150 struct sk_buff_head tmpq;
1152 bool need_xdp_flush = false;
1154 spin_lock(&queue->rx_lock);
1156 skb_queue_head_init(&rxq);
1157 skb_queue_head_init(&errq);
1158 skb_queue_head_init(&tmpq);
1160 rp = queue->rx.sring->rsp_prod;
1161 rmb(); /* Ensure we see queued responses up to 'rp'. */
1163 i = queue->rx.rsp_cons;
1165 while ((i != rp) && (work_done < budget)) {
1166 memcpy(rx, RING_GET_RESPONSE(&queue->rx, i), sizeof(*rx));
1167 memset(extras, 0, sizeof(rinfo.extras));
1169 err = xennet_get_responses(queue, &rinfo, rp, &tmpq,
1172 if (unlikely(err)) {
1174 while ((skb = __skb_dequeue(&tmpq)))
1175 __skb_queue_tail(&errq, skb);
1176 dev->stats.rx_errors++;
1177 i = queue->rx.rsp_cons;
1181 skb = __skb_dequeue(&tmpq);
1183 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1184 struct xen_netif_extra_info *gso;
1185 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1187 if (unlikely(xennet_set_skb_gso(skb, gso))) {
1188 __skb_queue_head(&tmpq, skb);
1189 queue->rx.rsp_cons += skb_queue_len(&tmpq);
1194 NETFRONT_SKB_CB(skb)->pull_to = rx->status;
1195 if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
1196 NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
1198 skb_frag_off_set(&skb_shinfo(skb)->frags[0], rx->offset);
1199 skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
1200 skb->data_len = rx->status;
1201 skb->len += rx->status;
1203 if (unlikely(xennet_fill_frags(queue, skb, &tmpq)))
1206 if (rx->flags & XEN_NETRXF_csum_blank)
1207 skb->ip_summed = CHECKSUM_PARTIAL;
1208 else if (rx->flags & XEN_NETRXF_data_validated)
1209 skb->ip_summed = CHECKSUM_UNNECESSARY;
1211 __skb_queue_tail(&rxq, skb);
1213 i = ++queue->rx.rsp_cons;
1219 __skb_queue_purge(&errq);
1221 work_done -= handle_incoming_queue(queue, &rxq);
1223 xennet_alloc_rx_buffers(queue);
1225 if (work_done < budget) {
1228 napi_complete_done(napi, work_done);
1230 RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do);
1232 napi_schedule(napi);
1235 spin_unlock(&queue->rx_lock);
1240 static int xennet_change_mtu(struct net_device *dev, int mtu)
1242 int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN;
1250 static void xennet_get_stats64(struct net_device *dev,
1251 struct rtnl_link_stats64 *tot)
1253 struct netfront_info *np = netdev_priv(dev);
1256 for_each_possible_cpu(cpu) {
1257 struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu);
1258 struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu);
1259 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1263 start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
1264 tx_packets = tx_stats->packets;
1265 tx_bytes = tx_stats->bytes;
1266 } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
1269 start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
1270 rx_packets = rx_stats->packets;
1271 rx_bytes = rx_stats->bytes;
1272 } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
1274 tot->rx_packets += rx_packets;
1275 tot->tx_packets += tx_packets;
1276 tot->rx_bytes += rx_bytes;
1277 tot->tx_bytes += tx_bytes;
1280 tot->rx_errors = dev->stats.rx_errors;
1281 tot->tx_dropped = dev->stats.tx_dropped;
1284 static void xennet_release_tx_bufs(struct netfront_queue *queue)
1286 struct sk_buff *skb;
1289 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1290 /* Skip over entries which are actually freelist references */
1291 if (skb_entry_is_link(&queue->tx_skbs[i]))
1294 skb = queue->tx_skbs[i].skb;
1295 get_page(queue->grant_tx_page[i]);
1296 gnttab_end_foreign_access(queue->grant_tx_ref[i],
1298 (unsigned long)page_address(queue->grant_tx_page[i]));
1299 queue->grant_tx_page[i] = NULL;
1300 queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1301 add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, i);
1302 dev_kfree_skb_irq(skb);
1306 static void xennet_release_rx_bufs(struct netfront_queue *queue)
1310 spin_lock_bh(&queue->rx_lock);
1312 for (id = 0; id < NET_RX_RING_SIZE; id++) {
1313 struct sk_buff *skb;
1316 skb = queue->rx_skbs[id];
1320 ref = queue->grant_rx_ref[id];
1321 if (ref == GRANT_INVALID_REF)
1324 page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
1326 /* gnttab_end_foreign_access() needs a page ref until
1327 * foreign access is ended (which may be deferred).
1330 gnttab_end_foreign_access(ref, 0,
1331 (unsigned long)page_address(page));
1332 queue->grant_rx_ref[id] = GRANT_INVALID_REF;
1337 spin_unlock_bh(&queue->rx_lock);
1340 static netdev_features_t xennet_fix_features(struct net_device *dev,
1341 netdev_features_t features)
1343 struct netfront_info *np = netdev_priv(dev);
1345 if (features & NETIF_F_SG &&
1346 !xenbus_read_unsigned(np->xbdev->otherend, "feature-sg", 0))
1347 features &= ~NETIF_F_SG;
1349 if (features & NETIF_F_IPV6_CSUM &&
1350 !xenbus_read_unsigned(np->xbdev->otherend,
1351 "feature-ipv6-csum-offload", 0))
1352 features &= ~NETIF_F_IPV6_CSUM;
1354 if (features & NETIF_F_TSO &&
1355 !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv4", 0))
1356 features &= ~NETIF_F_TSO;
1358 if (features & NETIF_F_TSO6 &&
1359 !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv6", 0))
1360 features &= ~NETIF_F_TSO6;
1365 static int xennet_set_features(struct net_device *dev,
1366 netdev_features_t features)
1368 if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1369 netdev_info(dev, "Reducing MTU because no SG offload");
1370 dev->mtu = ETH_DATA_LEN;
1376 static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1378 struct netfront_queue *queue = dev_id;
1379 unsigned long flags;
1381 spin_lock_irqsave(&queue->tx_lock, flags);
1382 xennet_tx_buf_gc(queue);
1383 spin_unlock_irqrestore(&queue->tx_lock, flags);
1388 static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1390 struct netfront_queue *queue = dev_id;
1391 struct net_device *dev = queue->info->netdev;
1393 if (likely(netif_carrier_ok(dev) &&
1394 RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)))
1395 napi_schedule(&queue->napi);
1400 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1402 xennet_tx_interrupt(irq, dev_id);
1403 xennet_rx_interrupt(irq, dev_id);
1407 #ifdef CONFIG_NET_POLL_CONTROLLER
1408 static void xennet_poll_controller(struct net_device *dev)
1410 /* Poll each queue */
1411 struct netfront_info *info = netdev_priv(dev);
1412 unsigned int num_queues = dev->real_num_tx_queues;
1414 for (i = 0; i < num_queues; ++i)
1415 xennet_interrupt(0, &info->queues[i]);
1419 #define NETBACK_XDP_HEADROOM_DISABLE 0
1420 #define NETBACK_XDP_HEADROOM_ENABLE 1
1422 static int talk_to_netback_xdp(struct netfront_info *np, int xdp)
1425 unsigned short headroom;
1427 headroom = xdp ? XDP_PACKET_HEADROOM : 0;
1428 err = xenbus_printf(XBT_NIL, np->xbdev->nodename,
1429 "xdp-headroom", "%hu",
1432 pr_warn("Error writing xdp-headroom\n");
1437 static int xennet_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1438 struct netlink_ext_ack *extack)
1440 unsigned long max_mtu = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM;
1441 struct netfront_info *np = netdev_priv(dev);
1442 struct bpf_prog *old_prog;
1443 unsigned int i, err;
1445 if (dev->mtu > max_mtu) {
1446 netdev_warn(dev, "XDP requires MTU less than %lu\n", max_mtu);
1450 if (!np->netback_has_xdp_headroom)
1453 xenbus_switch_state(np->xbdev, XenbusStateReconfiguring);
1455 err = talk_to_netback_xdp(np, prog ? NETBACK_XDP_HEADROOM_ENABLE :
1456 NETBACK_XDP_HEADROOM_DISABLE);
1460 /* avoid the race with XDP headroom adjustment */
1461 wait_event(module_wq,
1462 xenbus_read_driver_state(np->xbdev->otherend) ==
1463 XenbusStateReconfigured);
1464 np->netfront_xdp_enabled = true;
1466 old_prog = rtnl_dereference(np->queues[0].xdp_prog);
1469 bpf_prog_add(prog, dev->real_num_tx_queues);
1471 for (i = 0; i < dev->real_num_tx_queues; ++i)
1472 rcu_assign_pointer(np->queues[i].xdp_prog, prog);
1475 for (i = 0; i < dev->real_num_tx_queues; ++i)
1476 bpf_prog_put(old_prog);
1478 xenbus_switch_state(np->xbdev, XenbusStateConnected);
1483 static u32 xennet_xdp_query(struct net_device *dev)
1485 unsigned int num_queues = dev->real_num_tx_queues;
1486 struct netfront_info *np = netdev_priv(dev);
1487 const struct bpf_prog *xdp_prog;
1488 struct netfront_queue *queue;
1491 for (i = 0; i < num_queues; ++i) {
1492 queue = &np->queues[i];
1493 xdp_prog = rtnl_dereference(queue->xdp_prog);
1495 return xdp_prog->aux->id;
1501 static int xennet_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1503 switch (xdp->command) {
1504 case XDP_SETUP_PROG:
1505 return xennet_xdp_set(dev, xdp->prog, xdp->extack);
1506 case XDP_QUERY_PROG:
1507 xdp->prog_id = xennet_xdp_query(dev);
1514 static const struct net_device_ops xennet_netdev_ops = {
1515 .ndo_open = xennet_open,
1516 .ndo_stop = xennet_close,
1517 .ndo_start_xmit = xennet_start_xmit,
1518 .ndo_change_mtu = xennet_change_mtu,
1519 .ndo_get_stats64 = xennet_get_stats64,
1520 .ndo_set_mac_address = eth_mac_addr,
1521 .ndo_validate_addr = eth_validate_addr,
1522 .ndo_fix_features = xennet_fix_features,
1523 .ndo_set_features = xennet_set_features,
1524 .ndo_select_queue = xennet_select_queue,
1525 .ndo_bpf = xennet_xdp,
1526 .ndo_xdp_xmit = xennet_xdp_xmit,
1527 #ifdef CONFIG_NET_POLL_CONTROLLER
1528 .ndo_poll_controller = xennet_poll_controller,
1532 static void xennet_free_netdev(struct net_device *netdev)
1534 struct netfront_info *np = netdev_priv(netdev);
1536 free_percpu(np->rx_stats);
1537 free_percpu(np->tx_stats);
1538 free_netdev(netdev);
1541 static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1544 struct net_device *netdev;
1545 struct netfront_info *np;
1547 netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues);
1549 return ERR_PTR(-ENOMEM);
1551 np = netdev_priv(netdev);
1557 np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1558 if (np->rx_stats == NULL)
1560 np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1561 if (np->tx_stats == NULL)
1564 netdev->netdev_ops = &xennet_netdev_ops;
1566 netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1568 netdev->hw_features = NETIF_F_SG |
1570 NETIF_F_TSO | NETIF_F_TSO6;
1573 * Assume that all hw features are available for now. This set
1574 * will be adjusted by the call to netdev_update_features() in
1575 * xennet_connect() which is the earliest point where we can
1576 * negotiate with the backend regarding supported features.
1578 netdev->features |= netdev->hw_features;
1580 netdev->ethtool_ops = &xennet_ethtool_ops;
1581 netdev->min_mtu = ETH_MIN_MTU;
1582 netdev->max_mtu = XEN_NETIF_MAX_TX_SIZE;
1583 SET_NETDEV_DEV(netdev, &dev->dev);
1585 np->netdev = netdev;
1586 np->netfront_xdp_enabled = false;
1588 netif_carrier_off(netdev);
1591 xenbus_switch_state(dev, XenbusStateInitialising);
1592 err = wait_event_timeout(module_wq,
1593 xenbus_read_driver_state(dev->otherend) !=
1594 XenbusStateClosed &&
1595 xenbus_read_driver_state(dev->otherend) !=
1596 XenbusStateUnknown, XENNET_TIMEOUT);
1602 xennet_free_netdev(netdev);
1603 return ERR_PTR(err);
1607 * Entry point to this code when a new device is created. Allocate the basic
1608 * structures and the ring buffers for communication with the backend, and
1609 * inform the backend of the appropriate details for those.
1611 static int netfront_probe(struct xenbus_device *dev,
1612 const struct xenbus_device_id *id)
1615 struct net_device *netdev;
1616 struct netfront_info *info;
1618 netdev = xennet_create_dev(dev);
1619 if (IS_ERR(netdev)) {
1620 err = PTR_ERR(netdev);
1621 xenbus_dev_fatal(dev, err, "creating netdev");
1625 info = netdev_priv(netdev);
1626 dev_set_drvdata(&dev->dev, info);
1628 info->netdev->sysfs_groups[0] = &xennet_dev_group;
1634 static void xennet_end_access(int ref, void *page)
1636 /* This frees the page as a side-effect */
1637 if (ref != GRANT_INVALID_REF)
1638 gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1641 static void xennet_disconnect_backend(struct netfront_info *info)
1644 unsigned int num_queues = info->netdev->real_num_tx_queues;
1646 netif_carrier_off(info->netdev);
1648 for (i = 0; i < num_queues && info->queues; ++i) {
1649 struct netfront_queue *queue = &info->queues[i];
1651 del_timer_sync(&queue->rx_refill_timer);
1653 if (queue->tx_irq && (queue->tx_irq == queue->rx_irq))
1654 unbind_from_irqhandler(queue->tx_irq, queue);
1655 if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) {
1656 unbind_from_irqhandler(queue->tx_irq, queue);
1657 unbind_from_irqhandler(queue->rx_irq, queue);
1659 queue->tx_evtchn = queue->rx_evtchn = 0;
1660 queue->tx_irq = queue->rx_irq = 0;
1662 if (netif_running(info->netdev))
1663 napi_synchronize(&queue->napi);
1665 xennet_release_tx_bufs(queue);
1666 xennet_release_rx_bufs(queue);
1667 gnttab_free_grant_references(queue->gref_tx_head);
1668 gnttab_free_grant_references(queue->gref_rx_head);
1670 /* End access and free the pages */
1671 xennet_end_access(queue->tx_ring_ref, queue->tx.sring);
1672 xennet_end_access(queue->rx_ring_ref, queue->rx.sring);
1674 queue->tx_ring_ref = GRANT_INVALID_REF;
1675 queue->rx_ring_ref = GRANT_INVALID_REF;
1676 queue->tx.sring = NULL;
1677 queue->rx.sring = NULL;
1679 page_pool_destroy(queue->page_pool);
1684 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1685 * driver restart. We tear down our netif structure and recreate it, but
1686 * leave the device-layer structures intact so that this is transparent to the
1687 * rest of the kernel.
1689 static int netfront_resume(struct xenbus_device *dev)
1691 struct netfront_info *info = dev_get_drvdata(&dev->dev);
1693 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1695 xennet_disconnect_backend(info);
1699 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1701 char *s, *e, *macstr;
1704 macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1706 return PTR_ERR(macstr);
1708 for (i = 0; i < ETH_ALEN; i++) {
1709 mac[i] = simple_strtoul(s, &e, 16);
1710 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1721 static int setup_netfront_single(struct netfront_queue *queue)
1725 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1729 err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1731 0, queue->info->netdev->name, queue);
1734 queue->rx_evtchn = queue->tx_evtchn;
1735 queue->rx_irq = queue->tx_irq = err;
1740 xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1741 queue->tx_evtchn = 0;
1746 static int setup_netfront_split(struct netfront_queue *queue)
1750 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1753 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn);
1755 goto alloc_rx_evtchn_fail;
1757 snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
1758 "%s-tx", queue->name);
1759 err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1760 xennet_tx_interrupt,
1761 0, queue->tx_irq_name, queue);
1764 queue->tx_irq = err;
1766 snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
1767 "%s-rx", queue->name);
1768 err = bind_evtchn_to_irqhandler(queue->rx_evtchn,
1769 xennet_rx_interrupt,
1770 0, queue->rx_irq_name, queue);
1773 queue->rx_irq = err;
1778 unbind_from_irqhandler(queue->tx_irq, queue);
1781 xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn);
1782 queue->rx_evtchn = 0;
1783 alloc_rx_evtchn_fail:
1784 xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1785 queue->tx_evtchn = 0;
1790 static int setup_netfront(struct xenbus_device *dev,
1791 struct netfront_queue *queue, unsigned int feature_split_evtchn)
1793 struct xen_netif_tx_sring *txs;
1794 struct xen_netif_rx_sring *rxs;
1798 queue->tx_ring_ref = GRANT_INVALID_REF;
1799 queue->rx_ring_ref = GRANT_INVALID_REF;
1800 queue->rx.sring = NULL;
1801 queue->tx.sring = NULL;
1803 txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1806 xenbus_dev_fatal(dev, err, "allocating tx ring page");
1809 SHARED_RING_INIT(txs);
1810 FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE);
1812 err = xenbus_grant_ring(dev, txs, 1, &gref);
1814 goto grant_tx_ring_fail;
1815 queue->tx_ring_ref = gref;
1817 rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1820 xenbus_dev_fatal(dev, err, "allocating rx ring page");
1821 goto alloc_rx_ring_fail;
1823 SHARED_RING_INIT(rxs);
1824 FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE);
1826 err = xenbus_grant_ring(dev, rxs, 1, &gref);
1828 goto grant_rx_ring_fail;
1829 queue->rx_ring_ref = gref;
1831 if (feature_split_evtchn)
1832 err = setup_netfront_split(queue);
1833 /* setup single event channel if
1834 * a) feature-split-event-channels == 0
1835 * b) feature-split-event-channels == 1 but failed to setup
1837 if (!feature_split_evtchn || (feature_split_evtchn && err))
1838 err = setup_netfront_single(queue);
1841 goto alloc_evtchn_fail;
1845 /* If we fail to setup netfront, it is safe to just revoke access to
1846 * granted pages because backend is not accessing it at this point.
1849 gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0);
1851 free_page((unsigned long)rxs);
1853 gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0);
1855 free_page((unsigned long)txs);
1860 /* Queue-specific initialisation
1861 * This used to be done in xennet_create_dev() but must now
1864 static int xennet_init_queue(struct netfront_queue *queue)
1870 spin_lock_init(&queue->tx_lock);
1871 spin_lock_init(&queue->rx_lock);
1873 timer_setup(&queue->rx_refill_timer, rx_refill_timeout, 0);
1875 devid = strrchr(queue->info->xbdev->nodename, '/') + 1;
1876 snprintf(queue->name, sizeof(queue->name), "vif%s-q%u",
1879 /* Initialise tx_skbs as a free chain containing every entry. */
1880 queue->tx_skb_freelist = 0;
1881 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1882 skb_entry_set_link(&queue->tx_skbs[i], i+1);
1883 queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1884 queue->grant_tx_page[i] = NULL;
1887 /* Clear out rx_skbs */
1888 for (i = 0; i < NET_RX_RING_SIZE; i++) {
1889 queue->rx_skbs[i] = NULL;
1890 queue->grant_rx_ref[i] = GRANT_INVALID_REF;
1893 /* A grant for every tx ring slot */
1894 if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
1895 &queue->gref_tx_head) < 0) {
1896 pr_alert("can't alloc tx grant refs\n");
1901 /* A grant for every rx ring slot */
1902 if (gnttab_alloc_grant_references(NET_RX_RING_SIZE,
1903 &queue->gref_rx_head) < 0) {
1904 pr_alert("can't alloc rx grant refs\n");
1912 gnttab_free_grant_references(queue->gref_tx_head);
1917 static int write_queue_xenstore_keys(struct netfront_queue *queue,
1918 struct xenbus_transaction *xbt, int write_hierarchical)
1920 /* Write the queue-specific keys into XenStore in the traditional
1921 * way for a single queue, or in a queue subkeys for multiple
1924 struct xenbus_device *dev = queue->info->xbdev;
1926 const char *message;
1930 /* Choose the correct place to write the keys */
1931 if (write_hierarchical) {
1932 pathsize = strlen(dev->nodename) + 10;
1933 path = kzalloc(pathsize, GFP_KERNEL);
1936 message = "out of memory while writing ring references";
1939 snprintf(path, pathsize, "%s/queue-%u",
1940 dev->nodename, queue->id);
1942 path = (char *)dev->nodename;
1945 /* Write ring references */
1946 err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u",
1947 queue->tx_ring_ref);
1949 message = "writing tx-ring-ref";
1953 err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u",
1954 queue->rx_ring_ref);
1956 message = "writing rx-ring-ref";
1960 /* Write event channels; taking into account both shared
1961 * and split event channel scenarios.
1963 if (queue->tx_evtchn == queue->rx_evtchn) {
1964 /* Shared event channel */
1965 err = xenbus_printf(*xbt, path,
1966 "event-channel", "%u", queue->tx_evtchn);
1968 message = "writing event-channel";
1972 /* Split event channels */
1973 err = xenbus_printf(*xbt, path,
1974 "event-channel-tx", "%u", queue->tx_evtchn);
1976 message = "writing event-channel-tx";
1980 err = xenbus_printf(*xbt, path,
1981 "event-channel-rx", "%u", queue->rx_evtchn);
1983 message = "writing event-channel-rx";
1988 if (write_hierarchical)
1993 if (write_hierarchical)
1995 xenbus_dev_fatal(dev, err, "%s", message);
1999 static void xennet_destroy_queues(struct netfront_info *info)
2003 for (i = 0; i < info->netdev->real_num_tx_queues; i++) {
2004 struct netfront_queue *queue = &info->queues[i];
2006 if (netif_running(info->netdev))
2007 napi_disable(&queue->napi);
2008 netif_napi_del(&queue->napi);
2011 kfree(info->queues);
2012 info->queues = NULL;
2017 static int xennet_create_page_pool(struct netfront_queue *queue)
2020 struct page_pool_params pp_params = {
2023 .pool_size = NET_RX_RING_SIZE,
2024 .nid = NUMA_NO_NODE,
2025 .dev = &queue->info->netdev->dev,
2026 .offset = XDP_PACKET_HEADROOM,
2027 .max_len = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM,
2030 queue->page_pool = page_pool_create(&pp_params);
2031 if (IS_ERR(queue->page_pool)) {
2032 err = PTR_ERR(queue->page_pool);
2033 queue->page_pool = NULL;
2037 err = xdp_rxq_info_reg(&queue->xdp_rxq, queue->info->netdev,
2040 netdev_err(queue->info->netdev, "xdp_rxq_info_reg failed\n");
2044 err = xdp_rxq_info_reg_mem_model(&queue->xdp_rxq,
2045 MEM_TYPE_PAGE_POOL, queue->page_pool);
2047 netdev_err(queue->info->netdev, "xdp_rxq_info_reg_mem_model failed\n");
2048 goto err_unregister_rxq;
2053 xdp_rxq_info_unreg(&queue->xdp_rxq);
2055 page_pool_destroy(queue->page_pool);
2056 queue->page_pool = NULL;
2060 static int xennet_create_queues(struct netfront_info *info,
2061 unsigned int *num_queues)
2066 info->queues = kcalloc(*num_queues, sizeof(struct netfront_queue),
2071 for (i = 0; i < *num_queues; i++) {
2072 struct netfront_queue *queue = &info->queues[i];
2077 ret = xennet_init_queue(queue);
2079 dev_warn(&info->xbdev->dev,
2080 "only created %d queues\n", i);
2085 /* use page pool recycling instead of buddy allocator */
2086 ret = xennet_create_page_pool(queue);
2088 dev_err(&info->xbdev->dev, "can't allocate page pool\n");
2093 netif_napi_add(queue->info->netdev, &queue->napi,
2095 if (netif_running(info->netdev))
2096 napi_enable(&queue->napi);
2099 netif_set_real_num_tx_queues(info->netdev, *num_queues);
2101 if (*num_queues == 0) {
2102 dev_err(&info->xbdev->dev, "no queues\n");
2108 /* Common code used when first setting up, and when resuming. */
2109 static int talk_to_netback(struct xenbus_device *dev,
2110 struct netfront_info *info)
2112 const char *message;
2113 struct xenbus_transaction xbt;
2115 unsigned int feature_split_evtchn;
2117 unsigned int max_queues = 0;
2118 struct netfront_queue *queue = NULL;
2119 unsigned int num_queues = 1;
2121 info->netdev->irq = 0;
2123 /* Check if backend supports multiple queues */
2124 max_queues = xenbus_read_unsigned(info->xbdev->otherend,
2125 "multi-queue-max-queues", 1);
2126 num_queues = min(max_queues, xennet_max_queues);
2128 /* Check feature-split-event-channels */
2129 feature_split_evtchn = xenbus_read_unsigned(info->xbdev->otherend,
2130 "feature-split-event-channels", 0);
2132 /* Read mac addr. */
2133 err = xen_net_read_mac(dev, info->netdev->dev_addr);
2135 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
2139 info->netback_has_xdp_headroom = xenbus_read_unsigned(info->xbdev->otherend,
2140 "feature-xdp-headroom", 0);
2141 if (info->netback_has_xdp_headroom) {
2142 /* set the current xen-netfront xdp state */
2143 err = talk_to_netback_xdp(info, info->netfront_xdp_enabled ?
2144 NETBACK_XDP_HEADROOM_ENABLE :
2145 NETBACK_XDP_HEADROOM_DISABLE);
2152 xennet_destroy_queues(info);
2154 err = xennet_create_queues(info, &num_queues);
2156 xenbus_dev_fatal(dev, err, "creating queues");
2157 kfree(info->queues);
2158 info->queues = NULL;
2163 /* Create shared ring, alloc event channel -- for each queue */
2164 for (i = 0; i < num_queues; ++i) {
2165 queue = &info->queues[i];
2166 err = setup_netfront(dev, queue, feature_split_evtchn);
2172 err = xenbus_transaction_start(&xbt);
2174 xenbus_dev_fatal(dev, err, "starting transaction");
2178 if (xenbus_exists(XBT_NIL,
2179 info->xbdev->otherend, "multi-queue-max-queues")) {
2180 /* Write the number of queues */
2181 err = xenbus_printf(xbt, dev->nodename,
2182 "multi-queue-num-queues", "%u", num_queues);
2184 message = "writing multi-queue-num-queues";
2185 goto abort_transaction_no_dev_fatal;
2189 if (num_queues == 1) {
2190 err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */
2192 goto abort_transaction_no_dev_fatal;
2194 /* Write the keys for each queue */
2195 for (i = 0; i < num_queues; ++i) {
2196 queue = &info->queues[i];
2197 err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */
2199 goto abort_transaction_no_dev_fatal;
2203 /* The remaining keys are not queue-specific */
2204 err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
2207 message = "writing request-rx-copy";
2208 goto abort_transaction;
2211 err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
2213 message = "writing feature-rx-notify";
2214 goto abort_transaction;
2217 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
2219 message = "writing feature-sg";
2220 goto abort_transaction;
2223 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
2225 message = "writing feature-gso-tcpv4";
2226 goto abort_transaction;
2229 err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1");
2231 message = "writing feature-gso-tcpv6";
2232 goto abort_transaction;
2235 err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload",
2238 message = "writing feature-ipv6-csum-offload";
2239 goto abort_transaction;
2242 err = xenbus_transaction_end(xbt, 0);
2246 xenbus_dev_fatal(dev, err, "completing transaction");
2253 xenbus_dev_fatal(dev, err, "%s", message);
2254 abort_transaction_no_dev_fatal:
2255 xenbus_transaction_end(xbt, 1);
2257 xennet_disconnect_backend(info);
2259 xennet_destroy_queues(info);
2263 device_unregister(&dev->dev);
2267 static int xennet_connect(struct net_device *dev)
2269 struct netfront_info *np = netdev_priv(dev);
2270 unsigned int num_queues = 0;
2273 struct netfront_queue *queue = NULL;
2275 if (!xenbus_read_unsigned(np->xbdev->otherend, "feature-rx-copy", 0)) {
2277 "backend does not support copying receive path\n");
2281 err = talk_to_netback(np->xbdev, np);
2284 if (np->netback_has_xdp_headroom)
2285 pr_info("backend supports XDP headroom\n");
2287 /* talk_to_netback() sets the correct number of queues */
2288 num_queues = dev->real_num_tx_queues;
2290 if (dev->reg_state == NETREG_UNINITIALIZED) {
2291 err = register_netdev(dev);
2293 pr_warn("%s: register_netdev err=%d\n", __func__, err);
2294 device_unregister(&np->xbdev->dev);
2300 netdev_update_features(dev);
2304 * All public and private state should now be sane. Get
2305 * ready to start sending and receiving packets and give the driver
2306 * domain a kick because we've probably just requeued some
2309 netif_carrier_on(np->netdev);
2310 for (j = 0; j < num_queues; ++j) {
2311 queue = &np->queues[j];
2313 notify_remote_via_irq(queue->tx_irq);
2314 if (queue->tx_irq != queue->rx_irq)
2315 notify_remote_via_irq(queue->rx_irq);
2317 spin_lock_irq(&queue->tx_lock);
2318 xennet_tx_buf_gc(queue);
2319 spin_unlock_irq(&queue->tx_lock);
2321 spin_lock_bh(&queue->rx_lock);
2322 xennet_alloc_rx_buffers(queue);
2323 spin_unlock_bh(&queue->rx_lock);
2330 * Callback received when the backend's state changes.
2332 static void netback_changed(struct xenbus_device *dev,
2333 enum xenbus_state backend_state)
2335 struct netfront_info *np = dev_get_drvdata(&dev->dev);
2336 struct net_device *netdev = np->netdev;
2338 dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
2340 wake_up_all(&module_wq);
2342 switch (backend_state) {
2343 case XenbusStateInitialising:
2344 case XenbusStateInitialised:
2345 case XenbusStateReconfiguring:
2346 case XenbusStateReconfigured:
2347 case XenbusStateUnknown:
2350 case XenbusStateInitWait:
2351 if (dev->state != XenbusStateInitialising)
2353 if (xennet_connect(netdev) != 0)
2355 xenbus_switch_state(dev, XenbusStateConnected);
2358 case XenbusStateConnected:
2359 netdev_notify_peers(netdev);
2362 case XenbusStateClosed:
2363 if (dev->state == XenbusStateClosed)
2365 /* Fall through - Missed the backend's CLOSING state. */
2366 case XenbusStateClosing:
2367 xenbus_frontend_closed(dev);
2372 static const struct xennet_stat {
2373 char name[ETH_GSTRING_LEN];
2375 } xennet_stats[] = {
2377 "rx_gso_checksum_fixup",
2378 offsetof(struct netfront_info, rx_gso_checksum_fixup)
2382 static int xennet_get_sset_count(struct net_device *dev, int string_set)
2384 switch (string_set) {
2386 return ARRAY_SIZE(xennet_stats);
2392 static void xennet_get_ethtool_stats(struct net_device *dev,
2393 struct ethtool_stats *stats, u64 * data)
2395 void *np = netdev_priv(dev);
2398 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2399 data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset));
2402 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
2406 switch (stringset) {
2408 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2409 memcpy(data + i * ETH_GSTRING_LEN,
2410 xennet_stats[i].name, ETH_GSTRING_LEN);
2415 static const struct ethtool_ops xennet_ethtool_ops =
2417 .get_link = ethtool_op_get_link,
2419 .get_sset_count = xennet_get_sset_count,
2420 .get_ethtool_stats = xennet_get_ethtool_stats,
2421 .get_strings = xennet_get_strings,
2422 .get_ts_info = ethtool_op_get_ts_info,
2426 static ssize_t show_rxbuf(struct device *dev,
2427 struct device_attribute *attr, char *buf)
2429 return sprintf(buf, "%lu\n", NET_RX_RING_SIZE);
2432 static ssize_t store_rxbuf(struct device *dev,
2433 struct device_attribute *attr,
2434 const char *buf, size_t len)
2437 unsigned long target;
2439 if (!capable(CAP_NET_ADMIN))
2442 target = simple_strtoul(buf, &endp, 0);
2446 /* rxbuf_min and rxbuf_max are no longer configurable. */
2451 static DEVICE_ATTR(rxbuf_min, 0644, show_rxbuf, store_rxbuf);
2452 static DEVICE_ATTR(rxbuf_max, 0644, show_rxbuf, store_rxbuf);
2453 static DEVICE_ATTR(rxbuf_cur, 0444, show_rxbuf, NULL);
2455 static struct attribute *xennet_dev_attrs[] = {
2456 &dev_attr_rxbuf_min.attr,
2457 &dev_attr_rxbuf_max.attr,
2458 &dev_attr_rxbuf_cur.attr,
2462 static const struct attribute_group xennet_dev_group = {
2463 .attrs = xennet_dev_attrs
2465 #endif /* CONFIG_SYSFS */
2467 static void xennet_bus_close(struct xenbus_device *dev)
2471 if (xenbus_read_driver_state(dev->otherend) == XenbusStateClosed)
2474 xenbus_switch_state(dev, XenbusStateClosing);
2475 ret = wait_event_timeout(module_wq,
2476 xenbus_read_driver_state(dev->otherend) ==
2477 XenbusStateClosing ||
2478 xenbus_read_driver_state(dev->otherend) ==
2479 XenbusStateClosed ||
2480 xenbus_read_driver_state(dev->otherend) ==
2485 if (xenbus_read_driver_state(dev->otherend) == XenbusStateClosed)
2489 xenbus_switch_state(dev, XenbusStateClosed);
2490 ret = wait_event_timeout(module_wq,
2491 xenbus_read_driver_state(dev->otherend) ==
2492 XenbusStateClosed ||
2493 xenbus_read_driver_state(dev->otherend) ==
2499 static int xennet_remove(struct xenbus_device *dev)
2501 struct netfront_info *info = dev_get_drvdata(&dev->dev);
2503 xennet_bus_close(dev);
2504 xennet_disconnect_backend(info);
2506 if (info->netdev->reg_state == NETREG_REGISTERED)
2507 unregister_netdev(info->netdev);
2511 xennet_destroy_queues(info);
2514 xennet_free_netdev(info->netdev);
2519 static const struct xenbus_device_id netfront_ids[] = {
2524 static struct xenbus_driver netfront_driver = {
2525 .ids = netfront_ids,
2526 .probe = netfront_probe,
2527 .remove = xennet_remove,
2528 .resume = netfront_resume,
2529 .otherend_changed = netback_changed,
2532 static int __init netif_init(void)
2537 if (!xen_has_pv_nic_devices())
2540 pr_info("Initialising Xen virtual ethernet driver\n");
2542 /* Allow as many queues as there are CPUs inut max. 8 if user has not
2543 * specified a value.
2545 if (xennet_max_queues == 0)
2546 xennet_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT,
2549 return xenbus_register_frontend(&netfront_driver);
2551 module_init(netif_init);
2554 static void __exit netif_exit(void)
2556 xenbus_unregister_driver(&netfront_driver);
2558 module_exit(netif_exit);
2560 MODULE_DESCRIPTION("Xen virtual network device frontend");
2561 MODULE_LICENSE("GPL");
2562 MODULE_ALIAS("xen:vif");
2563 MODULE_ALIAS("xennet");