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>
49 #include <xen/xenbus.h>
50 #include <xen/events.h>
52 #include <xen/platform_pci.h>
53 #include <xen/grant_table.h>
55 #include <xen/interface/io/netif.h>
56 #include <xen/interface/memory.h>
57 #include <xen/interface/grant_table.h>
59 /* Module parameters */
60 static unsigned int xennet_max_queues;
61 module_param_named(max_queues, xennet_max_queues, uint, 0644);
62 MODULE_PARM_DESC(max_queues,
63 "Maximum number of queues per virtual interface");
65 #define XENNET_TIMEOUT (5 * HZ)
67 static const struct ethtool_ops xennet_ethtool_ops;
73 #define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb))
75 #define RX_COPY_THRESHOLD 256
77 #define GRANT_INVALID_REF 0
79 #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, XEN_PAGE_SIZE)
80 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, XEN_PAGE_SIZE)
82 /* Minimum number of Rx slots (includes slot for GSO metadata). */
83 #define NET_RX_SLOTS_MIN (XEN_NETIF_NR_SLOTS_MIN + 1)
85 /* Queue name is interface name with "-qNNN" appended */
86 #define QUEUE_NAME_SIZE (IFNAMSIZ + 6)
88 /* IRQ name is queue name with "-tx" or "-rx" appended */
89 #define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3)
91 static DECLARE_WAIT_QUEUE_HEAD(module_wq);
93 struct netfront_stats {
96 struct u64_stats_sync syncp;
101 struct netfront_queue {
102 unsigned int id; /* Queue ID, 0-based */
103 char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */
104 struct netfront_info *info;
106 struct napi_struct napi;
108 /* Split event channels support, tx_* == rx_* when using
109 * single event channel.
111 unsigned int tx_evtchn, rx_evtchn;
112 unsigned int tx_irq, rx_irq;
113 /* Only used when split event channels support is enabled */
114 char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */
115 char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */
118 struct xen_netif_tx_front_ring tx;
122 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
123 * are linked from tx_skb_freelist through skb_entry.link.
125 * NB. Freelist index entries are always going to be less than
126 * PAGE_OFFSET, whereas pointers to skbs will always be equal or
127 * greater than PAGE_OFFSET: we use this property to distinguish
133 } tx_skbs[NET_TX_RING_SIZE];
134 grant_ref_t gref_tx_head;
135 grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
136 struct page *grant_tx_page[NET_TX_RING_SIZE];
137 unsigned tx_skb_freelist;
139 spinlock_t rx_lock ____cacheline_aligned_in_smp;
140 struct xen_netif_rx_front_ring rx;
143 struct timer_list rx_refill_timer;
145 struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
146 grant_ref_t gref_rx_head;
147 grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
150 struct netfront_info {
151 struct list_head list;
152 struct net_device *netdev;
154 struct xenbus_device *xbdev;
156 /* Multi-queue support */
157 struct netfront_queue *queues;
160 struct netfront_stats __percpu *rx_stats;
161 struct netfront_stats __percpu *tx_stats;
163 atomic_t rx_gso_checksum_fixup;
166 struct netfront_rx_info {
167 struct xen_netif_rx_response rx;
168 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
171 static void skb_entry_set_link(union skb_entry *list, unsigned short id)
176 static int skb_entry_is_link(const union skb_entry *list)
178 BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link));
179 return (unsigned long)list->skb < PAGE_OFFSET;
183 * Access macros for acquiring freeing slots in tx_skbs[].
186 static void add_id_to_freelist(unsigned *head, union skb_entry *list,
189 skb_entry_set_link(&list[id], *head);
193 static unsigned short get_id_from_freelist(unsigned *head,
194 union skb_entry *list)
196 unsigned int id = *head;
197 *head = list[id].link;
201 static int xennet_rxidx(RING_IDX idx)
203 return idx & (NET_RX_RING_SIZE - 1);
206 static struct sk_buff *xennet_get_rx_skb(struct netfront_queue *queue,
209 int i = xennet_rxidx(ri);
210 struct sk_buff *skb = queue->rx_skbs[i];
211 queue->rx_skbs[i] = NULL;
215 static grant_ref_t xennet_get_rx_ref(struct netfront_queue *queue,
218 int i = xennet_rxidx(ri);
219 grant_ref_t ref = queue->grant_rx_ref[i];
220 queue->grant_rx_ref[i] = GRANT_INVALID_REF;
225 static const struct attribute_group xennet_dev_group;
228 static bool xennet_can_sg(struct net_device *dev)
230 return dev->features & NETIF_F_SG;
234 static void rx_refill_timeout(unsigned long data)
236 struct netfront_queue *queue = (struct netfront_queue *)data;
237 napi_schedule(&queue->napi);
240 static int netfront_tx_slot_available(struct netfront_queue *queue)
242 return (queue->tx.req_prod_pvt - queue->tx.rsp_cons) <
243 (NET_TX_RING_SIZE - XEN_NETIF_NR_SLOTS_MIN - 1);
246 static void xennet_maybe_wake_tx(struct netfront_queue *queue)
248 struct net_device *dev = queue->info->netdev;
249 struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, queue->id);
251 if (unlikely(netif_tx_queue_stopped(dev_queue)) &&
252 netfront_tx_slot_available(queue) &&
253 likely(netif_running(dev)))
254 netif_tx_wake_queue(netdev_get_tx_queue(dev, queue->id));
258 static struct sk_buff *xennet_alloc_one_rx_buffer(struct netfront_queue *queue)
263 skb = __netdev_alloc_skb(queue->info->netdev,
264 RX_COPY_THRESHOLD + NET_IP_ALIGN,
265 GFP_ATOMIC | __GFP_NOWARN);
269 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
274 skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE);
276 /* Align ip header to a 16 bytes boundary */
277 skb_reserve(skb, NET_IP_ALIGN);
278 skb->dev = queue->info->netdev;
284 static void xennet_alloc_rx_buffers(struct netfront_queue *queue)
286 RING_IDX req_prod = queue->rx.req_prod_pvt;
290 if (unlikely(!netif_carrier_ok(queue->info->netdev)))
293 for (req_prod = queue->rx.req_prod_pvt;
294 req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE;
300 struct xen_netif_rx_request *req;
302 skb = xennet_alloc_one_rx_buffer(queue);
308 id = xennet_rxidx(req_prod);
310 BUG_ON(queue->rx_skbs[id]);
311 queue->rx_skbs[id] = skb;
313 ref = gnttab_claim_grant_reference(&queue->gref_rx_head);
314 WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
315 queue->grant_rx_ref[id] = ref;
317 page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
319 req = RING_GET_REQUEST(&queue->rx, req_prod);
320 gnttab_page_grant_foreign_access_ref_one(ref,
321 queue->info->xbdev->otherend_id,
328 queue->rx.req_prod_pvt = req_prod;
330 /* Try again later if there are not enough requests or skb allocation
332 * Enough requests is quantified as the sum of newly created slots and
333 * the unconsumed slots at the backend.
335 if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN ||
337 mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10));
341 wmb(); /* barrier so backend seens requests */
343 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify);
345 notify_remote_via_irq(queue->rx_irq);
348 static int xennet_open(struct net_device *dev)
350 struct netfront_info *np = netdev_priv(dev);
351 unsigned int num_queues = dev->real_num_tx_queues;
353 struct netfront_queue *queue = NULL;
358 for (i = 0; i < num_queues; ++i) {
359 queue = &np->queues[i];
360 napi_enable(&queue->napi);
362 spin_lock_bh(&queue->rx_lock);
363 if (netif_carrier_ok(dev)) {
364 xennet_alloc_rx_buffers(queue);
365 queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1;
366 if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))
367 napi_schedule(&queue->napi);
369 spin_unlock_bh(&queue->rx_lock);
372 netif_tx_start_all_queues(dev);
377 static void xennet_tx_buf_gc(struct netfront_queue *queue)
384 BUG_ON(!netif_carrier_ok(queue->info->netdev));
387 prod = queue->tx.sring->rsp_prod;
388 rmb(); /* Ensure we see responses up to 'rp'. */
390 for (cons = queue->tx.rsp_cons; cons != prod; cons++) {
391 struct xen_netif_tx_response *txrsp;
393 txrsp = RING_GET_RESPONSE(&queue->tx, cons);
394 if (txrsp->status == XEN_NETIF_RSP_NULL)
398 skb = queue->tx_skbs[id].skb;
399 if (unlikely(gnttab_query_foreign_access(
400 queue->grant_tx_ref[id]) != 0)) {
401 pr_alert("%s: warning -- grant still in use by backend domain\n",
405 gnttab_end_foreign_access_ref(
406 queue->grant_tx_ref[id], GNTMAP_readonly);
407 gnttab_release_grant_reference(
408 &queue->gref_tx_head, queue->grant_tx_ref[id]);
409 queue->grant_tx_ref[id] = GRANT_INVALID_REF;
410 queue->grant_tx_page[id] = NULL;
411 add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, id);
412 dev_kfree_skb_irq(skb);
415 queue->tx.rsp_cons = prod;
417 RING_FINAL_CHECK_FOR_RESPONSES(&queue->tx, more_to_do);
418 } while (more_to_do);
420 xennet_maybe_wake_tx(queue);
423 struct xennet_gnttab_make_txreq {
424 struct netfront_queue *queue;
427 struct xen_netif_tx_request *tx; /* Last request */
431 static void xennet_tx_setup_grant(unsigned long gfn, unsigned int offset,
432 unsigned int len, void *data)
434 struct xennet_gnttab_make_txreq *info = data;
436 struct xen_netif_tx_request *tx;
438 /* convenient aliases */
439 struct page *page = info->page;
440 struct netfront_queue *queue = info->queue;
441 struct sk_buff *skb = info->skb;
443 id = get_id_from_freelist(&queue->tx_skb_freelist, queue->tx_skbs);
444 tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
445 ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
446 WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
448 gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id,
449 gfn, GNTMAP_readonly);
451 queue->tx_skbs[id].skb = skb;
452 queue->grant_tx_page[id] = page;
453 queue->grant_tx_ref[id] = ref;
462 info->size += tx->size;
465 static struct xen_netif_tx_request *xennet_make_first_txreq(
466 struct netfront_queue *queue, struct sk_buff *skb,
467 struct page *page, unsigned int offset, unsigned int len)
469 struct xennet_gnttab_make_txreq info = {
476 gnttab_for_one_grant(page, offset, len, xennet_tx_setup_grant, &info);
481 static void xennet_make_one_txreq(unsigned long gfn, unsigned int offset,
482 unsigned int len, void *data)
484 struct xennet_gnttab_make_txreq *info = data;
486 info->tx->flags |= XEN_NETTXF_more_data;
488 xennet_tx_setup_grant(gfn, offset, len, data);
491 static struct xen_netif_tx_request *xennet_make_txreqs(
492 struct netfront_queue *queue, struct xen_netif_tx_request *tx,
493 struct sk_buff *skb, struct page *page,
494 unsigned int offset, unsigned int len)
496 struct xennet_gnttab_make_txreq info = {
502 /* Skip unused frames from start of page */
503 page += offset >> PAGE_SHIFT;
504 offset &= ~PAGE_MASK;
510 gnttab_foreach_grant_in_range(page, offset, len,
511 xennet_make_one_txreq,
523 * Count how many ring slots are required to send this skb. Each frag
524 * might be a compound page.
526 static int xennet_count_skb_slots(struct sk_buff *skb)
528 int i, frags = skb_shinfo(skb)->nr_frags;
531 slots = gnttab_count_grant(offset_in_page(skb->data),
534 for (i = 0; i < frags; i++) {
535 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
536 unsigned long size = skb_frag_size(frag);
537 unsigned long offset = frag->page_offset;
539 /* Skip unused frames from start of page */
540 offset &= ~PAGE_MASK;
542 slots += gnttab_count_grant(offset, size);
548 static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb,
549 void *accel_priv, select_queue_fallback_t fallback)
551 unsigned int num_queues = dev->real_num_tx_queues;
555 /* First, check if there is only one queue */
556 if (num_queues == 1) {
559 hash = skb_get_hash(skb);
560 queue_idx = hash % num_queues;
566 #define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1)
568 static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
570 struct netfront_info *np = netdev_priv(dev);
571 struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
572 struct xen_netif_tx_request *tx, *first_tx;
580 struct netfront_queue *queue = NULL;
581 unsigned int num_queues = dev->real_num_tx_queues;
583 struct sk_buff *nskb;
585 /* Drop the packet if no queues are set up */
588 /* Determine which queue to transmit this SKB on */
589 queue_index = skb_get_queue_mapping(skb);
590 queue = &np->queues[queue_index];
592 /* If skb->len is too big for wire format, drop skb and alert
593 * user about misconfiguration.
595 if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
596 net_alert_ratelimited(
597 "xennet: skb->len = %u, too big for wire format\n",
602 slots = xennet_count_skb_slots(skb);
603 if (unlikely(slots > MAX_XEN_SKB_FRAGS + 1)) {
604 net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n",
606 if (skb_linearize(skb))
610 page = virt_to_page(skb->data);
611 offset = offset_in_page(skb->data);
613 /* The first req should be at least ETH_HLEN size or the packet will be
614 * dropped by netback.
616 if (unlikely(PAGE_SIZE - offset < ETH_HLEN)) {
617 nskb = skb_copy(skb, GFP_ATOMIC);
620 dev_kfree_skb_any(skb);
622 page = virt_to_page(skb->data);
623 offset = offset_in_page(skb->data);
626 len = skb_headlen(skb);
628 spin_lock_irqsave(&queue->tx_lock, flags);
630 if (unlikely(!netif_carrier_ok(dev) ||
631 (slots > 1 && !xennet_can_sg(dev)) ||
632 netif_needs_gso(skb, netif_skb_features(skb)))) {
633 spin_unlock_irqrestore(&queue->tx_lock, flags);
637 /* First request for the linear area. */
638 first_tx = tx = xennet_make_first_txreq(queue, skb,
641 if (offset == PAGE_SIZE) {
647 if (skb->ip_summed == CHECKSUM_PARTIAL)
649 tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
650 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
651 /* remote but checksummed. */
652 tx->flags |= XEN_NETTXF_data_validated;
654 /* Optional extra info after the first request. */
655 if (skb_shinfo(skb)->gso_size) {
656 struct xen_netif_extra_info *gso;
658 gso = (struct xen_netif_extra_info *)
659 RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
661 tx->flags |= XEN_NETTXF_extra_info;
663 gso->u.gso.size = skb_shinfo(skb)->gso_size;
664 gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ?
665 XEN_NETIF_GSO_TYPE_TCPV6 :
666 XEN_NETIF_GSO_TYPE_TCPV4;
668 gso->u.gso.features = 0;
670 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
674 /* Requests for the rest of the linear area. */
675 tx = xennet_make_txreqs(queue, tx, skb, page, offset, len);
677 /* Requests for all the frags. */
678 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
679 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
680 tx = xennet_make_txreqs(queue, tx, skb,
681 skb_frag_page(frag), frag->page_offset,
682 skb_frag_size(frag));
685 /* First request has the packet length. */
686 first_tx->size = skb->len;
688 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
690 notify_remote_via_irq(queue->tx_irq);
692 u64_stats_update_begin(&tx_stats->syncp);
693 tx_stats->bytes += skb->len;
695 u64_stats_update_end(&tx_stats->syncp);
697 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
698 xennet_tx_buf_gc(queue);
700 if (!netfront_tx_slot_available(queue))
701 netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
703 spin_unlock_irqrestore(&queue->tx_lock, flags);
708 dev->stats.tx_dropped++;
709 dev_kfree_skb_any(skb);
713 static int xennet_close(struct net_device *dev)
715 struct netfront_info *np = netdev_priv(dev);
716 unsigned int num_queues = dev->real_num_tx_queues;
718 struct netfront_queue *queue;
719 netif_tx_stop_all_queues(np->netdev);
720 for (i = 0; i < num_queues; ++i) {
721 queue = &np->queues[i];
722 napi_disable(&queue->napi);
727 static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb,
730 int new = xennet_rxidx(queue->rx.req_prod_pvt);
732 BUG_ON(queue->rx_skbs[new]);
733 queue->rx_skbs[new] = skb;
734 queue->grant_rx_ref[new] = ref;
735 RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new;
736 RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref;
737 queue->rx.req_prod_pvt++;
740 static int xennet_get_extras(struct netfront_queue *queue,
741 struct xen_netif_extra_info *extras,
745 struct xen_netif_extra_info *extra;
746 struct device *dev = &queue->info->netdev->dev;
747 RING_IDX cons = queue->rx.rsp_cons;
754 if (unlikely(cons + 1 == rp)) {
756 dev_warn(dev, "Missing extra info\n");
761 extra = (struct xen_netif_extra_info *)
762 RING_GET_RESPONSE(&queue->rx, ++cons);
764 if (unlikely(!extra->type ||
765 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
767 dev_warn(dev, "Invalid extra type: %d\n",
771 memcpy(&extras[extra->type - 1], extra,
775 skb = xennet_get_rx_skb(queue, cons);
776 ref = xennet_get_rx_ref(queue, cons);
777 xennet_move_rx_slot(queue, skb, ref);
778 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
780 queue->rx.rsp_cons = cons;
784 static int xennet_get_responses(struct netfront_queue *queue,
785 struct netfront_rx_info *rinfo, RING_IDX rp,
786 struct sk_buff_head *list)
788 struct xen_netif_rx_response *rx = &rinfo->rx;
789 struct xen_netif_extra_info *extras = rinfo->extras;
790 struct device *dev = &queue->info->netdev->dev;
791 RING_IDX cons = queue->rx.rsp_cons;
792 struct sk_buff *skb = xennet_get_rx_skb(queue, cons);
793 grant_ref_t ref = xennet_get_rx_ref(queue, cons);
794 int max = XEN_NETIF_NR_SLOTS_MIN + (rx->status <= RX_COPY_THRESHOLD);
799 if (rx->flags & XEN_NETRXF_extra_info) {
800 err = xennet_get_extras(queue, extras, rp);
801 cons = queue->rx.rsp_cons;
805 if (unlikely(rx->status < 0 ||
806 rx->offset + rx->status > XEN_PAGE_SIZE)) {
808 dev_warn(dev, "rx->offset: %u, size: %d\n",
809 rx->offset, rx->status);
810 xennet_move_rx_slot(queue, skb, ref);
816 * This definitely indicates a bug, either in this driver or in
817 * the backend driver. In future this should flag the bad
818 * situation to the system controller to reboot the backend.
820 if (ref == GRANT_INVALID_REF) {
822 dev_warn(dev, "Bad rx response id %d.\n",
828 ret = gnttab_end_foreign_access_ref(ref, 0);
831 gnttab_release_grant_reference(&queue->gref_rx_head, ref);
833 __skb_queue_tail(list, skb);
836 if (!(rx->flags & XEN_NETRXF_more_data))
839 if (cons + slots == rp) {
841 dev_warn(dev, "Need more slots\n");
846 rx = RING_GET_RESPONSE(&queue->rx, cons + slots);
847 skb = xennet_get_rx_skb(queue, cons + slots);
848 ref = xennet_get_rx_ref(queue, cons + slots);
852 if (unlikely(slots > max)) {
854 dev_warn(dev, "Too many slots\n");
859 queue->rx.rsp_cons = cons + slots;
864 static int xennet_set_skb_gso(struct sk_buff *skb,
865 struct xen_netif_extra_info *gso)
867 if (!gso->u.gso.size) {
869 pr_warn("GSO size must not be zero\n");
873 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 &&
874 gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) {
876 pr_warn("Bad GSO type %d\n", gso->u.gso.type);
880 skb_shinfo(skb)->gso_size = gso->u.gso.size;
881 skb_shinfo(skb)->gso_type =
882 (gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ?
886 /* Header must be checked, and gso_segs computed. */
887 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
888 skb_shinfo(skb)->gso_segs = 0;
893 static int xennet_fill_frags(struct netfront_queue *queue,
895 struct sk_buff_head *list)
897 RING_IDX cons = queue->rx.rsp_cons;
898 struct sk_buff *nskb;
900 while ((nskb = __skb_dequeue(list))) {
901 struct xen_netif_rx_response *rx =
902 RING_GET_RESPONSE(&queue->rx, ++cons);
903 skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
905 if (skb_shinfo(skb)->nr_frags == MAX_SKB_FRAGS) {
906 unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
908 BUG_ON(pull_to < skb_headlen(skb));
909 __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
911 if (unlikely(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS)) {
912 queue->rx.rsp_cons = ++cons + skb_queue_len(list);
917 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags,
918 skb_frag_page(nfrag),
919 rx->offset, rx->status, PAGE_SIZE);
921 skb_shinfo(nskb)->nr_frags = 0;
925 queue->rx.rsp_cons = cons;
930 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
932 bool recalculate_partial_csum = false;
935 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
936 * peers can fail to set NETRXF_csum_blank when sending a GSO
937 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
938 * recalculate the partial checksum.
940 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
941 struct netfront_info *np = netdev_priv(dev);
942 atomic_inc(&np->rx_gso_checksum_fixup);
943 skb->ip_summed = CHECKSUM_PARTIAL;
944 recalculate_partial_csum = true;
947 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
948 if (skb->ip_summed != CHECKSUM_PARTIAL)
951 return skb_checksum_setup(skb, recalculate_partial_csum);
954 static int handle_incoming_queue(struct netfront_queue *queue,
955 struct sk_buff_head *rxq)
957 struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats);
958 int packets_dropped = 0;
961 while ((skb = __skb_dequeue(rxq)) != NULL) {
962 int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
964 if (pull_to > skb_headlen(skb))
965 __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
967 /* Ethernet work: Delayed to here as it peeks the header. */
968 skb->protocol = eth_type_trans(skb, queue->info->netdev);
969 skb_reset_network_header(skb);
971 if (checksum_setup(queue->info->netdev, skb)) {
974 queue->info->netdev->stats.rx_errors++;
978 u64_stats_update_begin(&rx_stats->syncp);
980 rx_stats->bytes += skb->len;
981 u64_stats_update_end(&rx_stats->syncp);
984 napi_gro_receive(&queue->napi, skb);
987 return packets_dropped;
990 static int xennet_poll(struct napi_struct *napi, int budget)
992 struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi);
993 struct net_device *dev = queue->info->netdev;
995 struct netfront_rx_info rinfo;
996 struct xen_netif_rx_response *rx = &rinfo.rx;
997 struct xen_netif_extra_info *extras = rinfo.extras;
1000 struct sk_buff_head rxq;
1001 struct sk_buff_head errq;
1002 struct sk_buff_head tmpq;
1005 spin_lock(&queue->rx_lock);
1007 skb_queue_head_init(&rxq);
1008 skb_queue_head_init(&errq);
1009 skb_queue_head_init(&tmpq);
1011 rp = queue->rx.sring->rsp_prod;
1012 rmb(); /* Ensure we see queued responses up to 'rp'. */
1014 i = queue->rx.rsp_cons;
1016 while ((i != rp) && (work_done < budget)) {
1017 memcpy(rx, RING_GET_RESPONSE(&queue->rx, i), sizeof(*rx));
1018 memset(extras, 0, sizeof(rinfo.extras));
1020 err = xennet_get_responses(queue, &rinfo, rp, &tmpq);
1022 if (unlikely(err)) {
1024 while ((skb = __skb_dequeue(&tmpq)))
1025 __skb_queue_tail(&errq, skb);
1026 dev->stats.rx_errors++;
1027 i = queue->rx.rsp_cons;
1031 skb = __skb_dequeue(&tmpq);
1033 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1034 struct xen_netif_extra_info *gso;
1035 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1037 if (unlikely(xennet_set_skb_gso(skb, gso))) {
1038 __skb_queue_head(&tmpq, skb);
1039 queue->rx.rsp_cons += skb_queue_len(&tmpq);
1044 NETFRONT_SKB_CB(skb)->pull_to = rx->status;
1045 if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
1046 NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
1048 skb_shinfo(skb)->frags[0].page_offset = rx->offset;
1049 skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
1050 skb->data_len = rx->status;
1051 skb->len += rx->status;
1053 if (unlikely(xennet_fill_frags(queue, skb, &tmpq)))
1056 if (rx->flags & XEN_NETRXF_csum_blank)
1057 skb->ip_summed = CHECKSUM_PARTIAL;
1058 else if (rx->flags & XEN_NETRXF_data_validated)
1059 skb->ip_summed = CHECKSUM_UNNECESSARY;
1061 __skb_queue_tail(&rxq, skb);
1063 i = ++queue->rx.rsp_cons;
1067 __skb_queue_purge(&errq);
1069 work_done -= handle_incoming_queue(queue, &rxq);
1071 xennet_alloc_rx_buffers(queue);
1073 if (work_done < budget) {
1076 napi_complete(napi);
1078 RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do);
1080 napi_schedule(napi);
1083 spin_unlock(&queue->rx_lock);
1088 static int xennet_change_mtu(struct net_device *dev, int mtu)
1090 int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN;
1098 static struct rtnl_link_stats64 *xennet_get_stats64(struct net_device *dev,
1099 struct rtnl_link_stats64 *tot)
1101 struct netfront_info *np = netdev_priv(dev);
1104 for_each_possible_cpu(cpu) {
1105 struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu);
1106 struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu);
1107 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1111 start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
1112 tx_packets = tx_stats->packets;
1113 tx_bytes = tx_stats->bytes;
1114 } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
1117 start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
1118 rx_packets = rx_stats->packets;
1119 rx_bytes = rx_stats->bytes;
1120 } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
1122 tot->rx_packets += rx_packets;
1123 tot->tx_packets += tx_packets;
1124 tot->rx_bytes += rx_bytes;
1125 tot->tx_bytes += tx_bytes;
1128 tot->rx_errors = dev->stats.rx_errors;
1129 tot->tx_dropped = dev->stats.tx_dropped;
1134 static void xennet_release_tx_bufs(struct netfront_queue *queue)
1136 struct sk_buff *skb;
1139 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1140 /* Skip over entries which are actually freelist references */
1141 if (skb_entry_is_link(&queue->tx_skbs[i]))
1144 skb = queue->tx_skbs[i].skb;
1145 get_page(queue->grant_tx_page[i]);
1146 gnttab_end_foreign_access(queue->grant_tx_ref[i],
1148 (unsigned long)page_address(queue->grant_tx_page[i]));
1149 queue->grant_tx_page[i] = NULL;
1150 queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1151 add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, i);
1152 dev_kfree_skb_irq(skb);
1156 static void xennet_release_rx_bufs(struct netfront_queue *queue)
1160 spin_lock_bh(&queue->rx_lock);
1162 for (id = 0; id < NET_RX_RING_SIZE; id++) {
1163 struct sk_buff *skb;
1166 skb = queue->rx_skbs[id];
1170 ref = queue->grant_rx_ref[id];
1171 if (ref == GRANT_INVALID_REF)
1174 page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
1176 /* gnttab_end_foreign_access() needs a page ref until
1177 * foreign access is ended (which may be deferred).
1180 gnttab_end_foreign_access(ref, 0,
1181 (unsigned long)page_address(page));
1182 queue->grant_rx_ref[id] = GRANT_INVALID_REF;
1187 spin_unlock_bh(&queue->rx_lock);
1190 static netdev_features_t xennet_fix_features(struct net_device *dev,
1191 netdev_features_t features)
1193 struct netfront_info *np = netdev_priv(dev);
1196 if (features & NETIF_F_SG) {
1197 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg",
1202 features &= ~NETIF_F_SG;
1205 if (features & NETIF_F_IPV6_CSUM) {
1206 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1207 "feature-ipv6-csum-offload", "%d", &val) < 0)
1211 features &= ~NETIF_F_IPV6_CSUM;
1214 if (features & NETIF_F_TSO) {
1215 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1216 "feature-gso-tcpv4", "%d", &val) < 0)
1220 features &= ~NETIF_F_TSO;
1223 if (features & NETIF_F_TSO6) {
1224 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1225 "feature-gso-tcpv6", "%d", &val) < 0)
1229 features &= ~NETIF_F_TSO6;
1235 static int xennet_set_features(struct net_device *dev,
1236 netdev_features_t features)
1238 if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1239 netdev_info(dev, "Reducing MTU because no SG offload");
1240 dev->mtu = ETH_DATA_LEN;
1246 static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1248 struct netfront_queue *queue = dev_id;
1249 unsigned long flags;
1251 spin_lock_irqsave(&queue->tx_lock, flags);
1252 xennet_tx_buf_gc(queue);
1253 spin_unlock_irqrestore(&queue->tx_lock, flags);
1258 static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1260 struct netfront_queue *queue = dev_id;
1261 struct net_device *dev = queue->info->netdev;
1263 if (likely(netif_carrier_ok(dev) &&
1264 RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)))
1265 napi_schedule(&queue->napi);
1270 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1272 xennet_tx_interrupt(irq, dev_id);
1273 xennet_rx_interrupt(irq, dev_id);
1277 #ifdef CONFIG_NET_POLL_CONTROLLER
1278 static void xennet_poll_controller(struct net_device *dev)
1280 /* Poll each queue */
1281 struct netfront_info *info = netdev_priv(dev);
1282 unsigned int num_queues = dev->real_num_tx_queues;
1284 for (i = 0; i < num_queues; ++i)
1285 xennet_interrupt(0, &info->queues[i]);
1289 static const struct net_device_ops xennet_netdev_ops = {
1290 .ndo_open = xennet_open,
1291 .ndo_stop = xennet_close,
1292 .ndo_start_xmit = xennet_start_xmit,
1293 .ndo_change_mtu = xennet_change_mtu,
1294 .ndo_get_stats64 = xennet_get_stats64,
1295 .ndo_set_mac_address = eth_mac_addr,
1296 .ndo_validate_addr = eth_validate_addr,
1297 .ndo_fix_features = xennet_fix_features,
1298 .ndo_set_features = xennet_set_features,
1299 .ndo_select_queue = xennet_select_queue,
1300 #ifdef CONFIG_NET_POLL_CONTROLLER
1301 .ndo_poll_controller = xennet_poll_controller,
1305 static void xennet_free_netdev(struct net_device *netdev)
1307 struct netfront_info *np = netdev_priv(netdev);
1309 free_percpu(np->rx_stats);
1310 free_percpu(np->tx_stats);
1311 free_netdev(netdev);
1314 static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1317 struct net_device *netdev;
1318 struct netfront_info *np;
1320 netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues);
1322 return ERR_PTR(-ENOMEM);
1324 np = netdev_priv(netdev);
1330 np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1331 if (np->rx_stats == NULL)
1333 np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1334 if (np->tx_stats == NULL)
1337 netdev->netdev_ops = &xennet_netdev_ops;
1339 netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1341 netdev->hw_features = NETIF_F_SG |
1343 NETIF_F_TSO | NETIF_F_TSO6;
1346 * Assume that all hw features are available for now. This set
1347 * will be adjusted by the call to netdev_update_features() in
1348 * xennet_connect() which is the earliest point where we can
1349 * negotiate with the backend regarding supported features.
1351 netdev->features |= netdev->hw_features;
1353 netdev->ethtool_ops = &xennet_ethtool_ops;
1354 SET_NETDEV_DEV(netdev, &dev->dev);
1356 np->netdev = netdev;
1358 netif_carrier_off(netdev);
1361 xenbus_switch_state(dev, XenbusStateInitialising);
1362 err = wait_event_timeout(module_wq,
1363 xenbus_read_driver_state(dev->otherend) !=
1364 XenbusStateClosed &&
1365 xenbus_read_driver_state(dev->otherend) !=
1366 XenbusStateUnknown, XENNET_TIMEOUT);
1372 xennet_free_netdev(netdev);
1373 return ERR_PTR(err);
1377 * Entry point to this code when a new device is created. Allocate the basic
1378 * structures and the ring buffers for communication with the backend, and
1379 * inform the backend of the appropriate details for those.
1381 static int netfront_probe(struct xenbus_device *dev,
1382 const struct xenbus_device_id *id)
1385 struct net_device *netdev;
1386 struct netfront_info *info;
1388 netdev = xennet_create_dev(dev);
1389 if (IS_ERR(netdev)) {
1390 err = PTR_ERR(netdev);
1391 xenbus_dev_fatal(dev, err, "creating netdev");
1395 info = netdev_priv(netdev);
1396 dev_set_drvdata(&dev->dev, info);
1398 info->netdev->sysfs_groups[0] = &xennet_dev_group;
1404 static void xennet_end_access(int ref, void *page)
1406 /* This frees the page as a side-effect */
1407 if (ref != GRANT_INVALID_REF)
1408 gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1411 static void xennet_disconnect_backend(struct netfront_info *info)
1414 unsigned int num_queues = info->netdev->real_num_tx_queues;
1416 netif_carrier_off(info->netdev);
1418 for (i = 0; i < num_queues && info->queues; ++i) {
1419 struct netfront_queue *queue = &info->queues[i];
1421 del_timer_sync(&queue->rx_refill_timer);
1423 if (queue->tx_irq && (queue->tx_irq == queue->rx_irq))
1424 unbind_from_irqhandler(queue->tx_irq, queue);
1425 if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) {
1426 unbind_from_irqhandler(queue->tx_irq, queue);
1427 unbind_from_irqhandler(queue->rx_irq, queue);
1429 queue->tx_evtchn = queue->rx_evtchn = 0;
1430 queue->tx_irq = queue->rx_irq = 0;
1432 if (netif_running(info->netdev))
1433 napi_synchronize(&queue->napi);
1435 xennet_release_tx_bufs(queue);
1436 xennet_release_rx_bufs(queue);
1437 gnttab_free_grant_references(queue->gref_tx_head);
1438 gnttab_free_grant_references(queue->gref_rx_head);
1440 /* End access and free the pages */
1441 xennet_end_access(queue->tx_ring_ref, queue->tx.sring);
1442 xennet_end_access(queue->rx_ring_ref, queue->rx.sring);
1444 queue->tx_ring_ref = GRANT_INVALID_REF;
1445 queue->rx_ring_ref = GRANT_INVALID_REF;
1446 queue->tx.sring = NULL;
1447 queue->rx.sring = NULL;
1452 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1453 * driver restart. We tear down our netif structure and recreate it, but
1454 * leave the device-layer structures intact so that this is transparent to the
1455 * rest of the kernel.
1457 static int netfront_resume(struct xenbus_device *dev)
1459 struct netfront_info *info = dev_get_drvdata(&dev->dev);
1461 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1463 xennet_disconnect_backend(info);
1467 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1469 char *s, *e, *macstr;
1472 macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1474 return PTR_ERR(macstr);
1476 for (i = 0; i < ETH_ALEN; i++) {
1477 mac[i] = simple_strtoul(s, &e, 16);
1478 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1489 static int setup_netfront_single(struct netfront_queue *queue)
1493 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1497 err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1499 0, queue->info->netdev->name, queue);
1502 queue->rx_evtchn = queue->tx_evtchn;
1503 queue->rx_irq = queue->tx_irq = err;
1508 xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1509 queue->tx_evtchn = 0;
1514 static int setup_netfront_split(struct netfront_queue *queue)
1518 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1521 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn);
1523 goto alloc_rx_evtchn_fail;
1525 snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
1526 "%s-tx", queue->name);
1527 err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1528 xennet_tx_interrupt,
1529 0, queue->tx_irq_name, queue);
1532 queue->tx_irq = err;
1534 snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
1535 "%s-rx", queue->name);
1536 err = bind_evtchn_to_irqhandler(queue->rx_evtchn,
1537 xennet_rx_interrupt,
1538 0, queue->rx_irq_name, queue);
1541 queue->rx_irq = err;
1546 unbind_from_irqhandler(queue->tx_irq, queue);
1549 xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn);
1550 queue->rx_evtchn = 0;
1551 alloc_rx_evtchn_fail:
1552 xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1553 queue->tx_evtchn = 0;
1558 static int setup_netfront(struct xenbus_device *dev,
1559 struct netfront_queue *queue, unsigned int feature_split_evtchn)
1561 struct xen_netif_tx_sring *txs;
1562 struct xen_netif_rx_sring *rxs;
1566 queue->tx_ring_ref = GRANT_INVALID_REF;
1567 queue->rx_ring_ref = GRANT_INVALID_REF;
1568 queue->rx.sring = NULL;
1569 queue->tx.sring = NULL;
1571 txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1574 xenbus_dev_fatal(dev, err, "allocating tx ring page");
1577 SHARED_RING_INIT(txs);
1578 FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE);
1580 err = xenbus_grant_ring(dev, txs, 1, &gref);
1582 goto grant_tx_ring_fail;
1583 queue->tx_ring_ref = gref;
1585 rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1588 xenbus_dev_fatal(dev, err, "allocating rx ring page");
1589 goto alloc_rx_ring_fail;
1591 SHARED_RING_INIT(rxs);
1592 FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE);
1594 err = xenbus_grant_ring(dev, rxs, 1, &gref);
1596 goto grant_rx_ring_fail;
1597 queue->rx_ring_ref = gref;
1599 if (feature_split_evtchn)
1600 err = setup_netfront_split(queue);
1601 /* setup single event channel if
1602 * a) feature-split-event-channels == 0
1603 * b) feature-split-event-channels == 1 but failed to setup
1605 if (!feature_split_evtchn || (feature_split_evtchn && err))
1606 err = setup_netfront_single(queue);
1609 goto alloc_evtchn_fail;
1613 /* If we fail to setup netfront, it is safe to just revoke access to
1614 * granted pages because backend is not accessing it at this point.
1617 gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0);
1619 free_page((unsigned long)rxs);
1621 gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0);
1623 free_page((unsigned long)txs);
1628 /* Queue-specific initialisation
1629 * This used to be done in xennet_create_dev() but must now
1632 static int xennet_init_queue(struct netfront_queue *queue)
1638 spin_lock_init(&queue->tx_lock);
1639 spin_lock_init(&queue->rx_lock);
1641 setup_timer(&queue->rx_refill_timer, rx_refill_timeout,
1642 (unsigned long)queue);
1644 devid = strrchr(queue->info->xbdev->nodename, '/') + 1;
1645 snprintf(queue->name, sizeof(queue->name), "vif%s-q%u",
1648 /* Initialise tx_skbs as a free chain containing every entry. */
1649 queue->tx_skb_freelist = 0;
1650 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1651 skb_entry_set_link(&queue->tx_skbs[i], i+1);
1652 queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1653 queue->grant_tx_page[i] = NULL;
1656 /* Clear out rx_skbs */
1657 for (i = 0; i < NET_RX_RING_SIZE; i++) {
1658 queue->rx_skbs[i] = NULL;
1659 queue->grant_rx_ref[i] = GRANT_INVALID_REF;
1662 /* A grant for every tx ring slot */
1663 if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
1664 &queue->gref_tx_head) < 0) {
1665 pr_alert("can't alloc tx grant refs\n");
1670 /* A grant for every rx ring slot */
1671 if (gnttab_alloc_grant_references(NET_RX_RING_SIZE,
1672 &queue->gref_rx_head) < 0) {
1673 pr_alert("can't alloc rx grant refs\n");
1681 gnttab_free_grant_references(queue->gref_tx_head);
1686 static int write_queue_xenstore_keys(struct netfront_queue *queue,
1687 struct xenbus_transaction *xbt, int write_hierarchical)
1689 /* Write the queue-specific keys into XenStore in the traditional
1690 * way for a single queue, or in a queue subkeys for multiple
1693 struct xenbus_device *dev = queue->info->xbdev;
1695 const char *message;
1699 /* Choose the correct place to write the keys */
1700 if (write_hierarchical) {
1701 pathsize = strlen(dev->nodename) + 10;
1702 path = kzalloc(pathsize, GFP_KERNEL);
1705 message = "out of memory while writing ring references";
1708 snprintf(path, pathsize, "%s/queue-%u",
1709 dev->nodename, queue->id);
1711 path = (char *)dev->nodename;
1714 /* Write ring references */
1715 err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u",
1716 queue->tx_ring_ref);
1718 message = "writing tx-ring-ref";
1722 err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u",
1723 queue->rx_ring_ref);
1725 message = "writing rx-ring-ref";
1729 /* Write event channels; taking into account both shared
1730 * and split event channel scenarios.
1732 if (queue->tx_evtchn == queue->rx_evtchn) {
1733 /* Shared event channel */
1734 err = xenbus_printf(*xbt, path,
1735 "event-channel", "%u", queue->tx_evtchn);
1737 message = "writing event-channel";
1741 /* Split event channels */
1742 err = xenbus_printf(*xbt, path,
1743 "event-channel-tx", "%u", queue->tx_evtchn);
1745 message = "writing event-channel-tx";
1749 err = xenbus_printf(*xbt, path,
1750 "event-channel-rx", "%u", queue->rx_evtchn);
1752 message = "writing event-channel-rx";
1757 if (write_hierarchical)
1762 if (write_hierarchical)
1764 xenbus_dev_fatal(dev, err, "%s", message);
1768 static void xennet_destroy_queues(struct netfront_info *info)
1772 for (i = 0; i < info->netdev->real_num_tx_queues; i++) {
1773 struct netfront_queue *queue = &info->queues[i];
1775 if (netif_running(info->netdev))
1776 napi_disable(&queue->napi);
1777 netif_napi_del(&queue->napi);
1780 kfree(info->queues);
1781 info->queues = NULL;
1784 static int xennet_create_queues(struct netfront_info *info,
1785 unsigned int *num_queues)
1790 info->queues = kcalloc(*num_queues, sizeof(struct netfront_queue),
1795 for (i = 0; i < *num_queues; i++) {
1796 struct netfront_queue *queue = &info->queues[i];
1801 ret = xennet_init_queue(queue);
1803 dev_warn(&info->xbdev->dev,
1804 "only created %d queues\n", i);
1809 netif_napi_add(queue->info->netdev, &queue->napi,
1811 if (netif_running(info->netdev))
1812 napi_enable(&queue->napi);
1815 netif_set_real_num_tx_queues(info->netdev, *num_queues);
1817 if (*num_queues == 0) {
1818 dev_err(&info->xbdev->dev, "no queues\n");
1824 /* Common code used when first setting up, and when resuming. */
1825 static int talk_to_netback(struct xenbus_device *dev,
1826 struct netfront_info *info)
1828 const char *message;
1829 struct xenbus_transaction xbt;
1831 unsigned int feature_split_evtchn;
1833 unsigned int max_queues = 0;
1834 struct netfront_queue *queue = NULL;
1835 unsigned int num_queues = 1;
1837 info->netdev->irq = 0;
1839 /* Check if backend supports multiple queues */
1840 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1841 "multi-queue-max-queues", "%u", &max_queues);
1844 num_queues = min(max_queues, xennet_max_queues);
1846 /* Check feature-split-event-channels */
1847 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1848 "feature-split-event-channels", "%u",
1849 &feature_split_evtchn);
1851 feature_split_evtchn = 0;
1853 /* Read mac addr. */
1854 err = xen_net_read_mac(dev, info->netdev->dev_addr);
1856 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1862 xennet_destroy_queues(info);
1864 err = xennet_create_queues(info, &num_queues);
1866 xenbus_dev_fatal(dev, err, "creating queues");
1867 kfree(info->queues);
1868 info->queues = NULL;
1873 /* Create shared ring, alloc event channel -- for each queue */
1874 for (i = 0; i < num_queues; ++i) {
1875 queue = &info->queues[i];
1876 err = setup_netfront(dev, queue, feature_split_evtchn);
1882 err = xenbus_transaction_start(&xbt);
1884 xenbus_dev_fatal(dev, err, "starting transaction");
1888 if (xenbus_exists(XBT_NIL,
1889 info->xbdev->otherend, "multi-queue-max-queues")) {
1890 /* Write the number of queues */
1891 err = xenbus_printf(xbt, dev->nodename,
1892 "multi-queue-num-queues", "%u", num_queues);
1894 message = "writing multi-queue-num-queues";
1895 goto abort_transaction_no_dev_fatal;
1899 if (num_queues == 1) {
1900 err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */
1902 goto abort_transaction_no_dev_fatal;
1904 /* Write the keys for each queue */
1905 for (i = 0; i < num_queues; ++i) {
1906 queue = &info->queues[i];
1907 err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */
1909 goto abort_transaction_no_dev_fatal;
1913 /* The remaining keys are not queue-specific */
1914 err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1917 message = "writing request-rx-copy";
1918 goto abort_transaction;
1921 err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1923 message = "writing feature-rx-notify";
1924 goto abort_transaction;
1927 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1929 message = "writing feature-sg";
1930 goto abort_transaction;
1933 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1935 message = "writing feature-gso-tcpv4";
1936 goto abort_transaction;
1939 err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1");
1941 message = "writing feature-gso-tcpv6";
1942 goto abort_transaction;
1945 err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload",
1948 message = "writing feature-ipv6-csum-offload";
1949 goto abort_transaction;
1952 err = xenbus_transaction_end(xbt, 0);
1956 xenbus_dev_fatal(dev, err, "completing transaction");
1963 xenbus_dev_fatal(dev, err, "%s", message);
1964 abort_transaction_no_dev_fatal:
1965 xenbus_transaction_end(xbt, 1);
1967 xennet_disconnect_backend(info);
1969 xennet_destroy_queues(info);
1973 device_unregister(&dev->dev);
1977 static int xennet_connect(struct net_device *dev)
1979 struct netfront_info *np = netdev_priv(dev);
1980 unsigned int num_queues = 0;
1982 unsigned int feature_rx_copy;
1984 struct netfront_queue *queue = NULL;
1986 err = xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1987 "feature-rx-copy", "%u", &feature_rx_copy);
1989 feature_rx_copy = 0;
1991 if (!feature_rx_copy) {
1993 "backend does not support copying receive path\n");
1997 err = talk_to_netback(np->xbdev, np);
2001 /* talk_to_netback() sets the correct number of queues */
2002 num_queues = dev->real_num_tx_queues;
2004 if (dev->reg_state == NETREG_UNINITIALIZED) {
2005 err = register_netdev(dev);
2007 pr_warn("%s: register_netdev err=%d\n", __func__, err);
2008 device_unregister(&np->xbdev->dev);
2014 netdev_update_features(dev);
2018 * All public and private state should now be sane. Get
2019 * ready to start sending and receiving packets and give the driver
2020 * domain a kick because we've probably just requeued some
2023 netif_carrier_on(np->netdev);
2024 for (j = 0; j < num_queues; ++j) {
2025 queue = &np->queues[j];
2027 notify_remote_via_irq(queue->tx_irq);
2028 if (queue->tx_irq != queue->rx_irq)
2029 notify_remote_via_irq(queue->rx_irq);
2031 spin_lock_irq(&queue->tx_lock);
2032 xennet_tx_buf_gc(queue);
2033 spin_unlock_irq(&queue->tx_lock);
2035 spin_lock_bh(&queue->rx_lock);
2036 xennet_alloc_rx_buffers(queue);
2037 spin_unlock_bh(&queue->rx_lock);
2044 * Callback received when the backend's state changes.
2046 static void netback_changed(struct xenbus_device *dev,
2047 enum xenbus_state backend_state)
2049 struct netfront_info *np = dev_get_drvdata(&dev->dev);
2050 struct net_device *netdev = np->netdev;
2052 dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
2054 wake_up_all(&module_wq);
2056 switch (backend_state) {
2057 case XenbusStateInitialising:
2058 case XenbusStateInitialised:
2059 case XenbusStateReconfiguring:
2060 case XenbusStateReconfigured:
2061 case XenbusStateUnknown:
2064 case XenbusStateInitWait:
2065 if (dev->state != XenbusStateInitialising)
2067 if (xennet_connect(netdev) != 0)
2069 xenbus_switch_state(dev, XenbusStateConnected);
2072 case XenbusStateConnected:
2073 netdev_notify_peers(netdev);
2076 case XenbusStateClosed:
2077 if (dev->state == XenbusStateClosed)
2079 /* Missed the backend's CLOSING state -- fallthrough */
2080 case XenbusStateClosing:
2081 xenbus_frontend_closed(dev);
2086 static const struct xennet_stat {
2087 char name[ETH_GSTRING_LEN];
2089 } xennet_stats[] = {
2091 "rx_gso_checksum_fixup",
2092 offsetof(struct netfront_info, rx_gso_checksum_fixup)
2096 static int xennet_get_sset_count(struct net_device *dev, int string_set)
2098 switch (string_set) {
2100 return ARRAY_SIZE(xennet_stats);
2106 static void xennet_get_ethtool_stats(struct net_device *dev,
2107 struct ethtool_stats *stats, u64 * data)
2109 void *np = netdev_priv(dev);
2112 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2113 data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset));
2116 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
2120 switch (stringset) {
2122 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2123 memcpy(data + i * ETH_GSTRING_LEN,
2124 xennet_stats[i].name, ETH_GSTRING_LEN);
2129 static const struct ethtool_ops xennet_ethtool_ops =
2131 .get_link = ethtool_op_get_link,
2133 .get_sset_count = xennet_get_sset_count,
2134 .get_ethtool_stats = xennet_get_ethtool_stats,
2135 .get_strings = xennet_get_strings,
2139 static ssize_t show_rxbuf(struct device *dev,
2140 struct device_attribute *attr, char *buf)
2142 return sprintf(buf, "%lu\n", NET_RX_RING_SIZE);
2145 static ssize_t store_rxbuf(struct device *dev,
2146 struct device_attribute *attr,
2147 const char *buf, size_t len)
2150 unsigned long target;
2152 if (!capable(CAP_NET_ADMIN))
2155 target = simple_strtoul(buf, &endp, 0);
2159 /* rxbuf_min and rxbuf_max are no longer configurable. */
2164 static DEVICE_ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf);
2165 static DEVICE_ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf);
2166 static DEVICE_ATTR(rxbuf_cur, S_IRUGO, show_rxbuf, NULL);
2168 static struct attribute *xennet_dev_attrs[] = {
2169 &dev_attr_rxbuf_min.attr,
2170 &dev_attr_rxbuf_max.attr,
2171 &dev_attr_rxbuf_cur.attr,
2175 static const struct attribute_group xennet_dev_group = {
2176 .attrs = xennet_dev_attrs
2178 #endif /* CONFIG_SYSFS */
2180 static void xennet_bus_close(struct xenbus_device *dev)
2184 if (xenbus_read_driver_state(dev->otherend) == XenbusStateClosed)
2187 xenbus_switch_state(dev, XenbusStateClosing);
2188 ret = wait_event_timeout(module_wq,
2189 xenbus_read_driver_state(dev->otherend) ==
2190 XenbusStateClosing ||
2191 xenbus_read_driver_state(dev->otherend) ==
2192 XenbusStateClosed ||
2193 xenbus_read_driver_state(dev->otherend) ==
2198 if (xenbus_read_driver_state(dev->otherend) == XenbusStateClosed)
2202 xenbus_switch_state(dev, XenbusStateClosed);
2203 ret = wait_event_timeout(module_wq,
2204 xenbus_read_driver_state(dev->otherend) ==
2205 XenbusStateClosed ||
2206 xenbus_read_driver_state(dev->otherend) ==
2212 static int xennet_remove(struct xenbus_device *dev)
2214 struct netfront_info *info = dev_get_drvdata(&dev->dev);
2216 xennet_bus_close(dev);
2217 xennet_disconnect_backend(info);
2219 if (info->netdev->reg_state == NETREG_REGISTERED)
2220 unregister_netdev(info->netdev);
2224 xennet_destroy_queues(info);
2227 xennet_free_netdev(info->netdev);
2232 static const struct xenbus_device_id netfront_ids[] = {
2237 static struct xenbus_driver netfront_driver = {
2238 .ids = netfront_ids,
2239 .probe = netfront_probe,
2240 .remove = xennet_remove,
2241 .resume = netfront_resume,
2242 .otherend_changed = netback_changed,
2245 static int __init netif_init(void)
2250 if (!xen_has_pv_nic_devices())
2253 pr_info("Initialising Xen virtual ethernet driver\n");
2255 /* Allow as many queues as there are CPUs if user has not
2256 * specified a value.
2258 if (xennet_max_queues == 0)
2259 xennet_max_queues = num_online_cpus();
2261 return xenbus_register_frontend(&netfront_driver);
2263 module_init(netif_init);
2266 static void __exit netif_exit(void)
2268 xenbus_unregister_driver(&netfront_driver);
2270 module_exit(netif_exit);
2272 MODULE_DESCRIPTION("Xen virtual network device frontend");
2273 MODULE_LICENSE("GPL");
2274 MODULE_ALIAS("xen:vif");
2275 MODULE_ALIAS("xennet");