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 #define MAX_QUEUES_DEFAULT 8
61 static unsigned int xennet_max_queues;
62 module_param_named(max_queues, xennet_max_queues, uint, 0644);
63 MODULE_PARM_DESC(max_queues,
64 "Maximum number of queues per virtual interface");
66 #define XENNET_TIMEOUT (5 * HZ)
68 static const struct ethtool_ops xennet_ethtool_ops;
74 #define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb))
76 #define RX_COPY_THRESHOLD 256
78 #define GRANT_INVALID_REF 0
80 #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, XEN_PAGE_SIZE)
81 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, XEN_PAGE_SIZE)
83 /* Minimum number of Rx slots (includes slot for GSO metadata). */
84 #define NET_RX_SLOTS_MIN (XEN_NETIF_NR_SLOTS_MIN + 1)
86 /* Queue name is interface name with "-qNNN" appended */
87 #define QUEUE_NAME_SIZE (IFNAMSIZ + 6)
89 /* IRQ name is queue name with "-tx" or "-rx" appended */
90 #define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3)
92 static DECLARE_WAIT_QUEUE_HEAD(module_wq);
94 struct netfront_stats {
97 struct u64_stats_sync syncp;
100 struct netfront_info;
102 struct netfront_queue {
103 unsigned int id; /* Queue ID, 0-based */
104 char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */
105 struct netfront_info *info;
107 struct napi_struct napi;
109 /* Split event channels support, tx_* == rx_* when using
110 * single event channel.
112 unsigned int tx_evtchn, rx_evtchn;
113 unsigned int tx_irq, rx_irq;
114 /* Only used when split event channels support is enabled */
115 char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */
116 char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */
119 struct xen_netif_tx_front_ring tx;
123 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
124 * are linked from tx_skb_freelist through skb_entry.link.
126 * NB. Freelist index entries are always going to be less than
127 * PAGE_OFFSET, whereas pointers to skbs will always be equal or
128 * greater than PAGE_OFFSET: we use this property to distinguish
134 } tx_skbs[NET_TX_RING_SIZE];
135 grant_ref_t gref_tx_head;
136 grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
137 struct page *grant_tx_page[NET_TX_RING_SIZE];
138 unsigned tx_skb_freelist;
140 spinlock_t rx_lock ____cacheline_aligned_in_smp;
141 struct xen_netif_rx_front_ring rx;
144 struct timer_list rx_refill_timer;
146 struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
147 grant_ref_t gref_rx_head;
148 grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
151 struct netfront_info {
152 struct list_head list;
153 struct net_device *netdev;
155 struct xenbus_device *xbdev;
157 /* Multi-queue support */
158 struct netfront_queue *queues;
161 struct netfront_stats __percpu *rx_stats;
162 struct netfront_stats __percpu *tx_stats;
164 atomic_t rx_gso_checksum_fixup;
167 struct netfront_rx_info {
168 struct xen_netif_rx_response rx;
169 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
172 static void skb_entry_set_link(union skb_entry *list, unsigned short id)
177 static int skb_entry_is_link(const union skb_entry *list)
179 BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link));
180 return (unsigned long)list->skb < PAGE_OFFSET;
184 * Access macros for acquiring freeing slots in tx_skbs[].
187 static void add_id_to_freelist(unsigned *head, union skb_entry *list,
190 skb_entry_set_link(&list[id], *head);
194 static unsigned short get_id_from_freelist(unsigned *head,
195 union skb_entry *list)
197 unsigned int id = *head;
198 *head = list[id].link;
202 static int xennet_rxidx(RING_IDX idx)
204 return idx & (NET_RX_RING_SIZE - 1);
207 static struct sk_buff *xennet_get_rx_skb(struct netfront_queue *queue,
210 int i = xennet_rxidx(ri);
211 struct sk_buff *skb = queue->rx_skbs[i];
212 queue->rx_skbs[i] = NULL;
216 static grant_ref_t xennet_get_rx_ref(struct netfront_queue *queue,
219 int i = xennet_rxidx(ri);
220 grant_ref_t ref = queue->grant_rx_ref[i];
221 queue->grant_rx_ref[i] = GRANT_INVALID_REF;
226 static const struct attribute_group xennet_dev_group;
229 static bool xennet_can_sg(struct net_device *dev)
231 return dev->features & NETIF_F_SG;
235 static void rx_refill_timeout(unsigned long data)
237 struct netfront_queue *queue = (struct netfront_queue *)data;
238 napi_schedule(&queue->napi);
241 static int netfront_tx_slot_available(struct netfront_queue *queue)
243 return (queue->tx.req_prod_pvt - queue->tx.rsp_cons) <
244 (NET_TX_RING_SIZE - XEN_NETIF_NR_SLOTS_MIN - 1);
247 static void xennet_maybe_wake_tx(struct netfront_queue *queue)
249 struct net_device *dev = queue->info->netdev;
250 struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, queue->id);
252 if (unlikely(netif_tx_queue_stopped(dev_queue)) &&
253 netfront_tx_slot_available(queue) &&
254 likely(netif_running(dev)))
255 netif_tx_wake_queue(netdev_get_tx_queue(dev, queue->id));
259 static struct sk_buff *xennet_alloc_one_rx_buffer(struct netfront_queue *queue)
264 skb = __netdev_alloc_skb(queue->info->netdev,
265 RX_COPY_THRESHOLD + NET_IP_ALIGN,
266 GFP_ATOMIC | __GFP_NOWARN);
270 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
275 skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE);
277 /* Align ip header to a 16 bytes boundary */
278 skb_reserve(skb, NET_IP_ALIGN);
279 skb->dev = queue->info->netdev;
285 static void xennet_alloc_rx_buffers(struct netfront_queue *queue)
287 RING_IDX req_prod = queue->rx.req_prod_pvt;
291 if (unlikely(!netif_carrier_ok(queue->info->netdev)))
294 for (req_prod = queue->rx.req_prod_pvt;
295 req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE;
301 struct xen_netif_rx_request *req;
303 skb = xennet_alloc_one_rx_buffer(queue);
309 id = xennet_rxidx(req_prod);
311 BUG_ON(queue->rx_skbs[id]);
312 queue->rx_skbs[id] = skb;
314 ref = gnttab_claim_grant_reference(&queue->gref_rx_head);
315 WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
316 queue->grant_rx_ref[id] = ref;
318 page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
320 req = RING_GET_REQUEST(&queue->rx, req_prod);
321 gnttab_page_grant_foreign_access_ref_one(ref,
322 queue->info->xbdev->otherend_id,
329 queue->rx.req_prod_pvt = req_prod;
331 /* Try again later if there are not enough requests or skb allocation
333 * Enough requests is quantified as the sum of newly created slots and
334 * the unconsumed slots at the backend.
336 if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN ||
338 mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10));
342 wmb(); /* barrier so backend seens requests */
344 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify);
346 notify_remote_via_irq(queue->rx_irq);
349 static int xennet_open(struct net_device *dev)
351 struct netfront_info *np = netdev_priv(dev);
352 unsigned int num_queues = dev->real_num_tx_queues;
354 struct netfront_queue *queue = NULL;
359 for (i = 0; i < num_queues; ++i) {
360 queue = &np->queues[i];
361 napi_enable(&queue->napi);
363 spin_lock_bh(&queue->rx_lock);
364 if (netif_carrier_ok(dev)) {
365 xennet_alloc_rx_buffers(queue);
366 queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1;
367 if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))
368 napi_schedule(&queue->napi);
370 spin_unlock_bh(&queue->rx_lock);
373 netif_tx_start_all_queues(dev);
378 static void xennet_tx_buf_gc(struct netfront_queue *queue)
385 BUG_ON(!netif_carrier_ok(queue->info->netdev));
388 prod = queue->tx.sring->rsp_prod;
389 rmb(); /* Ensure we see responses up to 'rp'. */
391 for (cons = queue->tx.rsp_cons; cons != prod; cons++) {
392 struct xen_netif_tx_response *txrsp;
394 txrsp = RING_GET_RESPONSE(&queue->tx, cons);
395 if (txrsp->status == XEN_NETIF_RSP_NULL)
399 skb = queue->tx_skbs[id].skb;
400 if (unlikely(gnttab_query_foreign_access(
401 queue->grant_tx_ref[id]) != 0)) {
402 pr_alert("%s: warning -- grant still in use by backend domain\n",
406 gnttab_end_foreign_access_ref(
407 queue->grant_tx_ref[id], GNTMAP_readonly);
408 gnttab_release_grant_reference(
409 &queue->gref_tx_head, queue->grant_tx_ref[id]);
410 queue->grant_tx_ref[id] = GRANT_INVALID_REF;
411 queue->grant_tx_page[id] = NULL;
412 add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, id);
413 dev_kfree_skb_irq(skb);
416 queue->tx.rsp_cons = prod;
418 RING_FINAL_CHECK_FOR_RESPONSES(&queue->tx, more_to_do);
419 } while (more_to_do);
421 xennet_maybe_wake_tx(queue);
424 struct xennet_gnttab_make_txreq {
425 struct netfront_queue *queue;
428 struct xen_netif_tx_request *tx; /* Last request */
432 static void xennet_tx_setup_grant(unsigned long gfn, unsigned int offset,
433 unsigned int len, void *data)
435 struct xennet_gnttab_make_txreq *info = data;
437 struct xen_netif_tx_request *tx;
439 /* convenient aliases */
440 struct page *page = info->page;
441 struct netfront_queue *queue = info->queue;
442 struct sk_buff *skb = info->skb;
444 id = get_id_from_freelist(&queue->tx_skb_freelist, queue->tx_skbs);
445 tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
446 ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
447 WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
449 gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id,
450 gfn, GNTMAP_readonly);
452 queue->tx_skbs[id].skb = skb;
453 queue->grant_tx_page[id] = page;
454 queue->grant_tx_ref[id] = ref;
463 info->size += tx->size;
466 static struct xen_netif_tx_request *xennet_make_first_txreq(
467 struct netfront_queue *queue, struct sk_buff *skb,
468 struct page *page, unsigned int offset, unsigned int len)
470 struct xennet_gnttab_make_txreq info = {
477 gnttab_for_one_grant(page, offset, len, xennet_tx_setup_grant, &info);
482 static void xennet_make_one_txreq(unsigned long gfn, unsigned int offset,
483 unsigned int len, void *data)
485 struct xennet_gnttab_make_txreq *info = data;
487 info->tx->flags |= XEN_NETTXF_more_data;
489 xennet_tx_setup_grant(gfn, offset, len, data);
492 static struct xen_netif_tx_request *xennet_make_txreqs(
493 struct netfront_queue *queue, struct xen_netif_tx_request *tx,
494 struct sk_buff *skb, struct page *page,
495 unsigned int offset, unsigned int len)
497 struct xennet_gnttab_make_txreq info = {
503 /* Skip unused frames from start of page */
504 page += offset >> PAGE_SHIFT;
505 offset &= ~PAGE_MASK;
511 gnttab_foreach_grant_in_range(page, offset, len,
512 xennet_make_one_txreq,
524 * Count how many ring slots are required to send this skb. Each frag
525 * might be a compound page.
527 static int xennet_count_skb_slots(struct sk_buff *skb)
529 int i, frags = skb_shinfo(skb)->nr_frags;
532 slots = gnttab_count_grant(offset_in_page(skb->data),
535 for (i = 0; i < frags; i++) {
536 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
537 unsigned long size = skb_frag_size(frag);
538 unsigned long offset = frag->page_offset;
540 /* Skip unused frames from start of page */
541 offset &= ~PAGE_MASK;
543 slots += gnttab_count_grant(offset, size);
549 static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb,
550 void *accel_priv, select_queue_fallback_t fallback)
552 unsigned int num_queues = dev->real_num_tx_queues;
556 /* First, check if there is only one queue */
557 if (num_queues == 1) {
560 hash = skb_get_hash(skb);
561 queue_idx = hash % num_queues;
567 #define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1)
569 static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
571 struct netfront_info *np = netdev_priv(dev);
572 struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
573 struct xen_netif_tx_request *tx, *first_tx;
581 struct netfront_queue *queue = NULL;
582 unsigned int num_queues = dev->real_num_tx_queues;
584 struct sk_buff *nskb;
586 /* Drop the packet if no queues are set up */
589 /* Determine which queue to transmit this SKB on */
590 queue_index = skb_get_queue_mapping(skb);
591 queue = &np->queues[queue_index];
593 /* If skb->len is too big for wire format, drop skb and alert
594 * user about misconfiguration.
596 if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
597 net_alert_ratelimited(
598 "xennet: skb->len = %u, too big for wire format\n",
603 slots = xennet_count_skb_slots(skb);
604 if (unlikely(slots > MAX_XEN_SKB_FRAGS + 1)) {
605 net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n",
607 if (skb_linearize(skb))
611 page = virt_to_page(skb->data);
612 offset = offset_in_page(skb->data);
614 /* The first req should be at least ETH_HLEN size or the packet will be
615 * dropped by netback.
617 if (unlikely(PAGE_SIZE - offset < ETH_HLEN)) {
618 nskb = skb_copy(skb, GFP_ATOMIC);
621 dev_consume_skb_any(skb);
623 page = virt_to_page(skb->data);
624 offset = offset_in_page(skb->data);
627 len = skb_headlen(skb);
629 spin_lock_irqsave(&queue->tx_lock, flags);
631 if (unlikely(!netif_carrier_ok(dev) ||
632 (slots > 1 && !xennet_can_sg(dev)) ||
633 netif_needs_gso(skb, netif_skb_features(skb)))) {
634 spin_unlock_irqrestore(&queue->tx_lock, flags);
638 /* First request for the linear area. */
639 first_tx = tx = xennet_make_first_txreq(queue, skb,
642 if (offset == PAGE_SIZE) {
648 if (skb->ip_summed == CHECKSUM_PARTIAL)
650 tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
651 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
652 /* remote but checksummed. */
653 tx->flags |= XEN_NETTXF_data_validated;
655 /* Optional extra info after the first request. */
656 if (skb_shinfo(skb)->gso_size) {
657 struct xen_netif_extra_info *gso;
659 gso = (struct xen_netif_extra_info *)
660 RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
662 tx->flags |= XEN_NETTXF_extra_info;
664 gso->u.gso.size = skb_shinfo(skb)->gso_size;
665 gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ?
666 XEN_NETIF_GSO_TYPE_TCPV6 :
667 XEN_NETIF_GSO_TYPE_TCPV4;
669 gso->u.gso.features = 0;
671 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
675 /* Requests for the rest of the linear area. */
676 tx = xennet_make_txreqs(queue, tx, skb, page, offset, len);
678 /* Requests for all the frags. */
679 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
680 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
681 tx = xennet_make_txreqs(queue, tx, skb,
682 skb_frag_page(frag), frag->page_offset,
683 skb_frag_size(frag));
686 /* First request has the packet length. */
687 first_tx->size = skb->len;
689 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
691 notify_remote_via_irq(queue->tx_irq);
693 u64_stats_update_begin(&tx_stats->syncp);
694 tx_stats->bytes += skb->len;
696 u64_stats_update_end(&tx_stats->syncp);
698 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
699 xennet_tx_buf_gc(queue);
701 if (!netfront_tx_slot_available(queue))
702 netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
704 spin_unlock_irqrestore(&queue->tx_lock, flags);
709 dev->stats.tx_dropped++;
710 dev_kfree_skb_any(skb);
714 static int xennet_close(struct net_device *dev)
716 struct netfront_info *np = netdev_priv(dev);
717 unsigned int num_queues = dev->real_num_tx_queues;
719 struct netfront_queue *queue;
720 netif_tx_stop_all_queues(np->netdev);
721 for (i = 0; i < num_queues; ++i) {
722 queue = &np->queues[i];
723 napi_disable(&queue->napi);
728 static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb,
731 int new = xennet_rxidx(queue->rx.req_prod_pvt);
733 BUG_ON(queue->rx_skbs[new]);
734 queue->rx_skbs[new] = skb;
735 queue->grant_rx_ref[new] = ref;
736 RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new;
737 RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref;
738 queue->rx.req_prod_pvt++;
741 static int xennet_get_extras(struct netfront_queue *queue,
742 struct xen_netif_extra_info *extras,
746 struct xen_netif_extra_info *extra;
747 struct device *dev = &queue->info->netdev->dev;
748 RING_IDX cons = queue->rx.rsp_cons;
755 if (unlikely(cons + 1 == rp)) {
757 dev_warn(dev, "Missing extra info\n");
762 extra = (struct xen_netif_extra_info *)
763 RING_GET_RESPONSE(&queue->rx, ++cons);
765 if (unlikely(!extra->type ||
766 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
768 dev_warn(dev, "Invalid extra type: %d\n",
772 memcpy(&extras[extra->type - 1], extra,
776 skb = xennet_get_rx_skb(queue, cons);
777 ref = xennet_get_rx_ref(queue, cons);
778 xennet_move_rx_slot(queue, skb, ref);
779 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
781 queue->rx.rsp_cons = cons;
785 static int xennet_get_responses(struct netfront_queue *queue,
786 struct netfront_rx_info *rinfo, RING_IDX rp,
787 struct sk_buff_head *list)
789 struct xen_netif_rx_response *rx = &rinfo->rx;
790 struct xen_netif_extra_info *extras = rinfo->extras;
791 struct device *dev = &queue->info->netdev->dev;
792 RING_IDX cons = queue->rx.rsp_cons;
793 struct sk_buff *skb = xennet_get_rx_skb(queue, cons);
794 grant_ref_t ref = xennet_get_rx_ref(queue, cons);
795 int max = XEN_NETIF_NR_SLOTS_MIN + (rx->status <= RX_COPY_THRESHOLD);
800 if (rx->flags & XEN_NETRXF_extra_info) {
801 err = xennet_get_extras(queue, extras, rp);
802 cons = queue->rx.rsp_cons;
806 if (unlikely(rx->status < 0 ||
807 rx->offset + rx->status > XEN_PAGE_SIZE)) {
809 dev_warn(dev, "rx->offset: %u, size: %d\n",
810 rx->offset, rx->status);
811 xennet_move_rx_slot(queue, skb, ref);
817 * This definitely indicates a bug, either in this driver or in
818 * the backend driver. In future this should flag the bad
819 * situation to the system controller to reboot the backend.
821 if (ref == GRANT_INVALID_REF) {
823 dev_warn(dev, "Bad rx response id %d.\n",
829 ret = gnttab_end_foreign_access_ref(ref, 0);
832 gnttab_release_grant_reference(&queue->gref_rx_head, ref);
834 __skb_queue_tail(list, skb);
837 if (!(rx->flags & XEN_NETRXF_more_data))
840 if (cons + slots == rp) {
842 dev_warn(dev, "Need more slots\n");
847 rx = RING_GET_RESPONSE(&queue->rx, cons + slots);
848 skb = xennet_get_rx_skb(queue, cons + slots);
849 ref = xennet_get_rx_ref(queue, cons + slots);
853 if (unlikely(slots > max)) {
855 dev_warn(dev, "Too many slots\n");
860 queue->rx.rsp_cons = cons + slots;
865 static int xennet_set_skb_gso(struct sk_buff *skb,
866 struct xen_netif_extra_info *gso)
868 if (!gso->u.gso.size) {
870 pr_warn("GSO size must not be zero\n");
874 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 &&
875 gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) {
877 pr_warn("Bad GSO type %d\n", gso->u.gso.type);
881 skb_shinfo(skb)->gso_size = gso->u.gso.size;
882 skb_shinfo(skb)->gso_type =
883 (gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ?
887 /* Header must be checked, and gso_segs computed. */
888 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
889 skb_shinfo(skb)->gso_segs = 0;
894 static int xennet_fill_frags(struct netfront_queue *queue,
896 struct sk_buff_head *list)
898 RING_IDX cons = queue->rx.rsp_cons;
899 struct sk_buff *nskb;
901 while ((nskb = __skb_dequeue(list))) {
902 struct xen_netif_rx_response *rx =
903 RING_GET_RESPONSE(&queue->rx, ++cons);
904 skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
906 if (skb_shinfo(skb)->nr_frags == MAX_SKB_FRAGS) {
907 unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
909 BUG_ON(pull_to < skb_headlen(skb));
910 __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
912 if (unlikely(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS)) {
913 queue->rx.rsp_cons = ++cons + skb_queue_len(list);
918 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags,
919 skb_frag_page(nfrag),
920 rx->offset, rx->status, PAGE_SIZE);
922 skb_shinfo(nskb)->nr_frags = 0;
926 queue->rx.rsp_cons = cons;
931 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
933 bool recalculate_partial_csum = false;
936 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
937 * peers can fail to set NETRXF_csum_blank when sending a GSO
938 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
939 * recalculate the partial checksum.
941 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
942 struct netfront_info *np = netdev_priv(dev);
943 atomic_inc(&np->rx_gso_checksum_fixup);
944 skb->ip_summed = CHECKSUM_PARTIAL;
945 recalculate_partial_csum = true;
948 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
949 if (skb->ip_summed != CHECKSUM_PARTIAL)
952 return skb_checksum_setup(skb, recalculate_partial_csum);
955 static int handle_incoming_queue(struct netfront_queue *queue,
956 struct sk_buff_head *rxq)
958 struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats);
959 int packets_dropped = 0;
962 while ((skb = __skb_dequeue(rxq)) != NULL) {
963 int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
965 if (pull_to > skb_headlen(skb))
966 __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
968 /* Ethernet work: Delayed to here as it peeks the header. */
969 skb->protocol = eth_type_trans(skb, queue->info->netdev);
970 skb_reset_network_header(skb);
972 if (checksum_setup(queue->info->netdev, skb)) {
975 queue->info->netdev->stats.rx_errors++;
979 u64_stats_update_begin(&rx_stats->syncp);
981 rx_stats->bytes += skb->len;
982 u64_stats_update_end(&rx_stats->syncp);
985 napi_gro_receive(&queue->napi, skb);
988 return packets_dropped;
991 static int xennet_poll(struct napi_struct *napi, int budget)
993 struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi);
994 struct net_device *dev = queue->info->netdev;
996 struct netfront_rx_info rinfo;
997 struct xen_netif_rx_response *rx = &rinfo.rx;
998 struct xen_netif_extra_info *extras = rinfo.extras;
1001 struct sk_buff_head rxq;
1002 struct sk_buff_head errq;
1003 struct sk_buff_head tmpq;
1006 spin_lock(&queue->rx_lock);
1008 skb_queue_head_init(&rxq);
1009 skb_queue_head_init(&errq);
1010 skb_queue_head_init(&tmpq);
1012 rp = queue->rx.sring->rsp_prod;
1013 rmb(); /* Ensure we see queued responses up to 'rp'. */
1015 i = queue->rx.rsp_cons;
1017 while ((i != rp) && (work_done < budget)) {
1018 memcpy(rx, RING_GET_RESPONSE(&queue->rx, i), sizeof(*rx));
1019 memset(extras, 0, sizeof(rinfo.extras));
1021 err = xennet_get_responses(queue, &rinfo, rp, &tmpq);
1023 if (unlikely(err)) {
1025 while ((skb = __skb_dequeue(&tmpq)))
1026 __skb_queue_tail(&errq, skb);
1027 dev->stats.rx_errors++;
1028 i = queue->rx.rsp_cons;
1032 skb = __skb_dequeue(&tmpq);
1034 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1035 struct xen_netif_extra_info *gso;
1036 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1038 if (unlikely(xennet_set_skb_gso(skb, gso))) {
1039 __skb_queue_head(&tmpq, skb);
1040 queue->rx.rsp_cons += skb_queue_len(&tmpq);
1045 NETFRONT_SKB_CB(skb)->pull_to = rx->status;
1046 if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
1047 NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
1049 skb_shinfo(skb)->frags[0].page_offset = rx->offset;
1050 skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
1051 skb->data_len = rx->status;
1052 skb->len += rx->status;
1054 if (unlikely(xennet_fill_frags(queue, skb, &tmpq)))
1057 if (rx->flags & XEN_NETRXF_csum_blank)
1058 skb->ip_summed = CHECKSUM_PARTIAL;
1059 else if (rx->flags & XEN_NETRXF_data_validated)
1060 skb->ip_summed = CHECKSUM_UNNECESSARY;
1062 __skb_queue_tail(&rxq, skb);
1064 i = ++queue->rx.rsp_cons;
1068 __skb_queue_purge(&errq);
1070 work_done -= handle_incoming_queue(queue, &rxq);
1072 xennet_alloc_rx_buffers(queue);
1074 if (work_done < budget) {
1077 napi_complete_done(napi, work_done);
1079 RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do);
1081 napi_schedule(napi);
1084 spin_unlock(&queue->rx_lock);
1089 static int xennet_change_mtu(struct net_device *dev, int mtu)
1091 int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN;
1099 static void xennet_get_stats64(struct net_device *dev,
1100 struct rtnl_link_stats64 *tot)
1102 struct netfront_info *np = netdev_priv(dev);
1105 for_each_possible_cpu(cpu) {
1106 struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu);
1107 struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu);
1108 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1112 start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
1113 tx_packets = tx_stats->packets;
1114 tx_bytes = tx_stats->bytes;
1115 } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
1118 start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
1119 rx_packets = rx_stats->packets;
1120 rx_bytes = rx_stats->bytes;
1121 } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
1123 tot->rx_packets += rx_packets;
1124 tot->tx_packets += tx_packets;
1125 tot->rx_bytes += rx_bytes;
1126 tot->tx_bytes += tx_bytes;
1129 tot->rx_errors = dev->stats.rx_errors;
1130 tot->tx_dropped = dev->stats.tx_dropped;
1133 static void xennet_release_tx_bufs(struct netfront_queue *queue)
1135 struct sk_buff *skb;
1138 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1139 /* Skip over entries which are actually freelist references */
1140 if (skb_entry_is_link(&queue->tx_skbs[i]))
1143 skb = queue->tx_skbs[i].skb;
1144 get_page(queue->grant_tx_page[i]);
1145 gnttab_end_foreign_access(queue->grant_tx_ref[i],
1147 (unsigned long)page_address(queue->grant_tx_page[i]));
1148 queue->grant_tx_page[i] = NULL;
1149 queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1150 add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, i);
1151 dev_kfree_skb_irq(skb);
1155 static void xennet_release_rx_bufs(struct netfront_queue *queue)
1159 spin_lock_bh(&queue->rx_lock);
1161 for (id = 0; id < NET_RX_RING_SIZE; id++) {
1162 struct sk_buff *skb;
1165 skb = queue->rx_skbs[id];
1169 ref = queue->grant_rx_ref[id];
1170 if (ref == GRANT_INVALID_REF)
1173 page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
1175 /* gnttab_end_foreign_access() needs a page ref until
1176 * foreign access is ended (which may be deferred).
1179 gnttab_end_foreign_access(ref, 0,
1180 (unsigned long)page_address(page));
1181 queue->grant_rx_ref[id] = GRANT_INVALID_REF;
1186 spin_unlock_bh(&queue->rx_lock);
1189 static netdev_features_t xennet_fix_features(struct net_device *dev,
1190 netdev_features_t features)
1192 struct netfront_info *np = netdev_priv(dev);
1194 if (features & NETIF_F_SG &&
1195 !xenbus_read_unsigned(np->xbdev->otherend, "feature-sg", 0))
1196 features &= ~NETIF_F_SG;
1198 if (features & NETIF_F_IPV6_CSUM &&
1199 !xenbus_read_unsigned(np->xbdev->otherend,
1200 "feature-ipv6-csum-offload", 0))
1201 features &= ~NETIF_F_IPV6_CSUM;
1203 if (features & NETIF_F_TSO &&
1204 !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv4", 0))
1205 features &= ~NETIF_F_TSO;
1207 if (features & NETIF_F_TSO6 &&
1208 !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv6", 0))
1209 features &= ~NETIF_F_TSO6;
1214 static int xennet_set_features(struct net_device *dev,
1215 netdev_features_t features)
1217 if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1218 netdev_info(dev, "Reducing MTU because no SG offload");
1219 dev->mtu = ETH_DATA_LEN;
1225 static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1227 struct netfront_queue *queue = dev_id;
1228 unsigned long flags;
1230 spin_lock_irqsave(&queue->tx_lock, flags);
1231 xennet_tx_buf_gc(queue);
1232 spin_unlock_irqrestore(&queue->tx_lock, flags);
1237 static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1239 struct netfront_queue *queue = dev_id;
1240 struct net_device *dev = queue->info->netdev;
1242 if (likely(netif_carrier_ok(dev) &&
1243 RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)))
1244 napi_schedule(&queue->napi);
1249 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1251 xennet_tx_interrupt(irq, dev_id);
1252 xennet_rx_interrupt(irq, dev_id);
1256 #ifdef CONFIG_NET_POLL_CONTROLLER
1257 static void xennet_poll_controller(struct net_device *dev)
1259 /* Poll each queue */
1260 struct netfront_info *info = netdev_priv(dev);
1261 unsigned int num_queues = dev->real_num_tx_queues;
1263 for (i = 0; i < num_queues; ++i)
1264 xennet_interrupt(0, &info->queues[i]);
1268 static const struct net_device_ops xennet_netdev_ops = {
1269 .ndo_open = xennet_open,
1270 .ndo_stop = xennet_close,
1271 .ndo_start_xmit = xennet_start_xmit,
1272 .ndo_change_mtu = xennet_change_mtu,
1273 .ndo_get_stats64 = xennet_get_stats64,
1274 .ndo_set_mac_address = eth_mac_addr,
1275 .ndo_validate_addr = eth_validate_addr,
1276 .ndo_fix_features = xennet_fix_features,
1277 .ndo_set_features = xennet_set_features,
1278 .ndo_select_queue = xennet_select_queue,
1279 #ifdef CONFIG_NET_POLL_CONTROLLER
1280 .ndo_poll_controller = xennet_poll_controller,
1284 static void xennet_free_netdev(struct net_device *netdev)
1286 struct netfront_info *np = netdev_priv(netdev);
1288 free_percpu(np->rx_stats);
1289 free_percpu(np->tx_stats);
1290 free_netdev(netdev);
1293 static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1296 struct net_device *netdev;
1297 struct netfront_info *np;
1299 netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues);
1301 return ERR_PTR(-ENOMEM);
1303 np = netdev_priv(netdev);
1309 np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1310 if (np->rx_stats == NULL)
1312 np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1313 if (np->tx_stats == NULL)
1316 netdev->netdev_ops = &xennet_netdev_ops;
1318 netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1320 netdev->hw_features = NETIF_F_SG |
1322 NETIF_F_TSO | NETIF_F_TSO6;
1325 * Assume that all hw features are available for now. This set
1326 * will be adjusted by the call to netdev_update_features() in
1327 * xennet_connect() which is the earliest point where we can
1328 * negotiate with the backend regarding supported features.
1330 netdev->features |= netdev->hw_features;
1332 netdev->ethtool_ops = &xennet_ethtool_ops;
1333 netdev->min_mtu = ETH_MIN_MTU;
1334 netdev->max_mtu = XEN_NETIF_MAX_TX_SIZE;
1335 SET_NETDEV_DEV(netdev, &dev->dev);
1337 np->netdev = netdev;
1339 netif_carrier_off(netdev);
1342 xenbus_switch_state(dev, XenbusStateInitialising);
1343 err = wait_event_timeout(module_wq,
1344 xenbus_read_driver_state(dev->otherend) !=
1345 XenbusStateClosed &&
1346 xenbus_read_driver_state(dev->otherend) !=
1347 XenbusStateUnknown, XENNET_TIMEOUT);
1353 xennet_free_netdev(netdev);
1354 return ERR_PTR(err);
1358 * Entry point to this code when a new device is created. Allocate the basic
1359 * structures and the ring buffers for communication with the backend, and
1360 * inform the backend of the appropriate details for those.
1362 static int netfront_probe(struct xenbus_device *dev,
1363 const struct xenbus_device_id *id)
1366 struct net_device *netdev;
1367 struct netfront_info *info;
1369 netdev = xennet_create_dev(dev);
1370 if (IS_ERR(netdev)) {
1371 err = PTR_ERR(netdev);
1372 xenbus_dev_fatal(dev, err, "creating netdev");
1376 info = netdev_priv(netdev);
1377 dev_set_drvdata(&dev->dev, info);
1379 info->netdev->sysfs_groups[0] = &xennet_dev_group;
1385 static void xennet_end_access(int ref, void *page)
1387 /* This frees the page as a side-effect */
1388 if (ref != GRANT_INVALID_REF)
1389 gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1392 static void xennet_disconnect_backend(struct netfront_info *info)
1395 unsigned int num_queues = info->netdev->real_num_tx_queues;
1397 netif_carrier_off(info->netdev);
1399 for (i = 0; i < num_queues && info->queues; ++i) {
1400 struct netfront_queue *queue = &info->queues[i];
1402 del_timer_sync(&queue->rx_refill_timer);
1404 if (queue->tx_irq && (queue->tx_irq == queue->rx_irq))
1405 unbind_from_irqhandler(queue->tx_irq, queue);
1406 if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) {
1407 unbind_from_irqhandler(queue->tx_irq, queue);
1408 unbind_from_irqhandler(queue->rx_irq, queue);
1410 queue->tx_evtchn = queue->rx_evtchn = 0;
1411 queue->tx_irq = queue->rx_irq = 0;
1413 if (netif_running(info->netdev))
1414 napi_synchronize(&queue->napi);
1416 xennet_release_tx_bufs(queue);
1417 xennet_release_rx_bufs(queue);
1418 gnttab_free_grant_references(queue->gref_tx_head);
1419 gnttab_free_grant_references(queue->gref_rx_head);
1421 /* End access and free the pages */
1422 xennet_end_access(queue->tx_ring_ref, queue->tx.sring);
1423 xennet_end_access(queue->rx_ring_ref, queue->rx.sring);
1425 queue->tx_ring_ref = GRANT_INVALID_REF;
1426 queue->rx_ring_ref = GRANT_INVALID_REF;
1427 queue->tx.sring = NULL;
1428 queue->rx.sring = NULL;
1433 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1434 * driver restart. We tear down our netif structure and recreate it, but
1435 * leave the device-layer structures intact so that this is transparent to the
1436 * rest of the kernel.
1438 static int netfront_resume(struct xenbus_device *dev)
1440 struct netfront_info *info = dev_get_drvdata(&dev->dev);
1442 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1444 xennet_disconnect_backend(info);
1448 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1450 char *s, *e, *macstr;
1453 macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1455 return PTR_ERR(macstr);
1457 for (i = 0; i < ETH_ALEN; i++) {
1458 mac[i] = simple_strtoul(s, &e, 16);
1459 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1470 static int setup_netfront_single(struct netfront_queue *queue)
1474 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1478 err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1480 0, queue->info->netdev->name, queue);
1483 queue->rx_evtchn = queue->tx_evtchn;
1484 queue->rx_irq = queue->tx_irq = err;
1489 xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1490 queue->tx_evtchn = 0;
1495 static int setup_netfront_split(struct netfront_queue *queue)
1499 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1502 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn);
1504 goto alloc_rx_evtchn_fail;
1506 snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
1507 "%s-tx", queue->name);
1508 err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1509 xennet_tx_interrupt,
1510 0, queue->tx_irq_name, queue);
1513 queue->tx_irq = err;
1515 snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
1516 "%s-rx", queue->name);
1517 err = bind_evtchn_to_irqhandler(queue->rx_evtchn,
1518 xennet_rx_interrupt,
1519 0, queue->rx_irq_name, queue);
1522 queue->rx_irq = err;
1527 unbind_from_irqhandler(queue->tx_irq, queue);
1530 xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn);
1531 queue->rx_evtchn = 0;
1532 alloc_rx_evtchn_fail:
1533 xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1534 queue->tx_evtchn = 0;
1539 static int setup_netfront(struct xenbus_device *dev,
1540 struct netfront_queue *queue, unsigned int feature_split_evtchn)
1542 struct xen_netif_tx_sring *txs;
1543 struct xen_netif_rx_sring *rxs;
1547 queue->tx_ring_ref = GRANT_INVALID_REF;
1548 queue->rx_ring_ref = GRANT_INVALID_REF;
1549 queue->rx.sring = NULL;
1550 queue->tx.sring = NULL;
1552 txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1555 xenbus_dev_fatal(dev, err, "allocating tx ring page");
1558 SHARED_RING_INIT(txs);
1559 FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE);
1561 err = xenbus_grant_ring(dev, txs, 1, &gref);
1563 goto grant_tx_ring_fail;
1564 queue->tx_ring_ref = gref;
1566 rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1569 xenbus_dev_fatal(dev, err, "allocating rx ring page");
1570 goto alloc_rx_ring_fail;
1572 SHARED_RING_INIT(rxs);
1573 FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE);
1575 err = xenbus_grant_ring(dev, rxs, 1, &gref);
1577 goto grant_rx_ring_fail;
1578 queue->rx_ring_ref = gref;
1580 if (feature_split_evtchn)
1581 err = setup_netfront_split(queue);
1582 /* setup single event channel if
1583 * a) feature-split-event-channels == 0
1584 * b) feature-split-event-channels == 1 but failed to setup
1586 if (!feature_split_evtchn || (feature_split_evtchn && err))
1587 err = setup_netfront_single(queue);
1590 goto alloc_evtchn_fail;
1594 /* If we fail to setup netfront, it is safe to just revoke access to
1595 * granted pages because backend is not accessing it at this point.
1598 gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0);
1600 free_page((unsigned long)rxs);
1602 gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0);
1604 free_page((unsigned long)txs);
1609 /* Queue-specific initialisation
1610 * This used to be done in xennet_create_dev() but must now
1613 static int xennet_init_queue(struct netfront_queue *queue)
1619 spin_lock_init(&queue->tx_lock);
1620 spin_lock_init(&queue->rx_lock);
1622 setup_timer(&queue->rx_refill_timer, rx_refill_timeout,
1623 (unsigned long)queue);
1625 devid = strrchr(queue->info->xbdev->nodename, '/') + 1;
1626 snprintf(queue->name, sizeof(queue->name), "vif%s-q%u",
1629 /* Initialise tx_skbs as a free chain containing every entry. */
1630 queue->tx_skb_freelist = 0;
1631 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1632 skb_entry_set_link(&queue->tx_skbs[i], i+1);
1633 queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1634 queue->grant_tx_page[i] = NULL;
1637 /* Clear out rx_skbs */
1638 for (i = 0; i < NET_RX_RING_SIZE; i++) {
1639 queue->rx_skbs[i] = NULL;
1640 queue->grant_rx_ref[i] = GRANT_INVALID_REF;
1643 /* A grant for every tx ring slot */
1644 if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
1645 &queue->gref_tx_head) < 0) {
1646 pr_alert("can't alloc tx grant refs\n");
1651 /* A grant for every rx ring slot */
1652 if (gnttab_alloc_grant_references(NET_RX_RING_SIZE,
1653 &queue->gref_rx_head) < 0) {
1654 pr_alert("can't alloc rx grant refs\n");
1662 gnttab_free_grant_references(queue->gref_tx_head);
1667 static int write_queue_xenstore_keys(struct netfront_queue *queue,
1668 struct xenbus_transaction *xbt, int write_hierarchical)
1670 /* Write the queue-specific keys into XenStore in the traditional
1671 * way for a single queue, or in a queue subkeys for multiple
1674 struct xenbus_device *dev = queue->info->xbdev;
1676 const char *message;
1680 /* Choose the correct place to write the keys */
1681 if (write_hierarchical) {
1682 pathsize = strlen(dev->nodename) + 10;
1683 path = kzalloc(pathsize, GFP_KERNEL);
1686 message = "out of memory while writing ring references";
1689 snprintf(path, pathsize, "%s/queue-%u",
1690 dev->nodename, queue->id);
1692 path = (char *)dev->nodename;
1695 /* Write ring references */
1696 err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u",
1697 queue->tx_ring_ref);
1699 message = "writing tx-ring-ref";
1703 err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u",
1704 queue->rx_ring_ref);
1706 message = "writing rx-ring-ref";
1710 /* Write event channels; taking into account both shared
1711 * and split event channel scenarios.
1713 if (queue->tx_evtchn == queue->rx_evtchn) {
1714 /* Shared event channel */
1715 err = xenbus_printf(*xbt, path,
1716 "event-channel", "%u", queue->tx_evtchn);
1718 message = "writing event-channel";
1722 /* Split event channels */
1723 err = xenbus_printf(*xbt, path,
1724 "event-channel-tx", "%u", queue->tx_evtchn);
1726 message = "writing event-channel-tx";
1730 err = xenbus_printf(*xbt, path,
1731 "event-channel-rx", "%u", queue->rx_evtchn);
1733 message = "writing event-channel-rx";
1738 if (write_hierarchical)
1743 if (write_hierarchical)
1745 xenbus_dev_fatal(dev, err, "%s", message);
1749 static void xennet_destroy_queues(struct netfront_info *info)
1753 for (i = 0; i < info->netdev->real_num_tx_queues; i++) {
1754 struct netfront_queue *queue = &info->queues[i];
1756 if (netif_running(info->netdev))
1757 napi_disable(&queue->napi);
1758 netif_napi_del(&queue->napi);
1761 kfree(info->queues);
1762 info->queues = NULL;
1765 static int xennet_create_queues(struct netfront_info *info,
1766 unsigned int *num_queues)
1771 info->queues = kcalloc(*num_queues, sizeof(struct netfront_queue),
1776 for (i = 0; i < *num_queues; i++) {
1777 struct netfront_queue *queue = &info->queues[i];
1782 ret = xennet_init_queue(queue);
1784 dev_warn(&info->xbdev->dev,
1785 "only created %d queues\n", i);
1790 netif_napi_add(queue->info->netdev, &queue->napi,
1792 if (netif_running(info->netdev))
1793 napi_enable(&queue->napi);
1796 netif_set_real_num_tx_queues(info->netdev, *num_queues);
1798 if (*num_queues == 0) {
1799 dev_err(&info->xbdev->dev, "no queues\n");
1805 /* Common code used when first setting up, and when resuming. */
1806 static int talk_to_netback(struct xenbus_device *dev,
1807 struct netfront_info *info)
1809 const char *message;
1810 struct xenbus_transaction xbt;
1812 unsigned int feature_split_evtchn;
1814 unsigned int max_queues = 0;
1815 struct netfront_queue *queue = NULL;
1816 unsigned int num_queues = 1;
1818 info->netdev->irq = 0;
1820 /* Check if backend supports multiple queues */
1821 max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1822 "multi-queue-max-queues", 1);
1823 num_queues = min(max_queues, xennet_max_queues);
1825 /* Check feature-split-event-channels */
1826 feature_split_evtchn = xenbus_read_unsigned(info->xbdev->otherend,
1827 "feature-split-event-channels", 0);
1829 /* Read mac addr. */
1830 err = xen_net_read_mac(dev, info->netdev->dev_addr);
1832 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1838 xennet_destroy_queues(info);
1840 err = xennet_create_queues(info, &num_queues);
1842 xenbus_dev_fatal(dev, err, "creating queues");
1843 kfree(info->queues);
1844 info->queues = NULL;
1849 /* Create shared ring, alloc event channel -- for each queue */
1850 for (i = 0; i < num_queues; ++i) {
1851 queue = &info->queues[i];
1852 err = setup_netfront(dev, queue, feature_split_evtchn);
1858 err = xenbus_transaction_start(&xbt);
1860 xenbus_dev_fatal(dev, err, "starting transaction");
1864 if (xenbus_exists(XBT_NIL,
1865 info->xbdev->otherend, "multi-queue-max-queues")) {
1866 /* Write the number of queues */
1867 err = xenbus_printf(xbt, dev->nodename,
1868 "multi-queue-num-queues", "%u", num_queues);
1870 message = "writing multi-queue-num-queues";
1871 goto abort_transaction_no_dev_fatal;
1875 if (num_queues == 1) {
1876 err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */
1878 goto abort_transaction_no_dev_fatal;
1880 /* Write the keys for each queue */
1881 for (i = 0; i < num_queues; ++i) {
1882 queue = &info->queues[i];
1883 err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */
1885 goto abort_transaction_no_dev_fatal;
1889 /* The remaining keys are not queue-specific */
1890 err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1893 message = "writing request-rx-copy";
1894 goto abort_transaction;
1897 err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1899 message = "writing feature-rx-notify";
1900 goto abort_transaction;
1903 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1905 message = "writing feature-sg";
1906 goto abort_transaction;
1909 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1911 message = "writing feature-gso-tcpv4";
1912 goto abort_transaction;
1915 err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1");
1917 message = "writing feature-gso-tcpv6";
1918 goto abort_transaction;
1921 err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload",
1924 message = "writing feature-ipv6-csum-offload";
1925 goto abort_transaction;
1928 err = xenbus_transaction_end(xbt, 0);
1932 xenbus_dev_fatal(dev, err, "completing transaction");
1939 xenbus_dev_fatal(dev, err, "%s", message);
1940 abort_transaction_no_dev_fatal:
1941 xenbus_transaction_end(xbt, 1);
1943 xennet_disconnect_backend(info);
1945 xennet_destroy_queues(info);
1949 device_unregister(&dev->dev);
1953 static int xennet_connect(struct net_device *dev)
1955 struct netfront_info *np = netdev_priv(dev);
1956 unsigned int num_queues = 0;
1959 struct netfront_queue *queue = NULL;
1961 if (!xenbus_read_unsigned(np->xbdev->otherend, "feature-rx-copy", 0)) {
1963 "backend does not support copying receive path\n");
1967 err = talk_to_netback(np->xbdev, np);
1971 /* talk_to_netback() sets the correct number of queues */
1972 num_queues = dev->real_num_tx_queues;
1974 if (dev->reg_state == NETREG_UNINITIALIZED) {
1975 err = register_netdev(dev);
1977 pr_warn("%s: register_netdev err=%d\n", __func__, err);
1978 device_unregister(&np->xbdev->dev);
1984 netdev_update_features(dev);
1988 * All public and private state should now be sane. Get
1989 * ready to start sending and receiving packets and give the driver
1990 * domain a kick because we've probably just requeued some
1993 netif_carrier_on(np->netdev);
1994 for (j = 0; j < num_queues; ++j) {
1995 queue = &np->queues[j];
1997 notify_remote_via_irq(queue->tx_irq);
1998 if (queue->tx_irq != queue->rx_irq)
1999 notify_remote_via_irq(queue->rx_irq);
2001 spin_lock_irq(&queue->tx_lock);
2002 xennet_tx_buf_gc(queue);
2003 spin_unlock_irq(&queue->tx_lock);
2005 spin_lock_bh(&queue->rx_lock);
2006 xennet_alloc_rx_buffers(queue);
2007 spin_unlock_bh(&queue->rx_lock);
2014 * Callback received when the backend's state changes.
2016 static void netback_changed(struct xenbus_device *dev,
2017 enum xenbus_state backend_state)
2019 struct netfront_info *np = dev_get_drvdata(&dev->dev);
2020 struct net_device *netdev = np->netdev;
2022 dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
2024 wake_up_all(&module_wq);
2026 switch (backend_state) {
2027 case XenbusStateInitialising:
2028 case XenbusStateInitialised:
2029 case XenbusStateReconfiguring:
2030 case XenbusStateReconfigured:
2031 case XenbusStateUnknown:
2034 case XenbusStateInitWait:
2035 if (dev->state != XenbusStateInitialising)
2037 if (xennet_connect(netdev) != 0)
2039 xenbus_switch_state(dev, XenbusStateConnected);
2042 case XenbusStateConnected:
2043 netdev_notify_peers(netdev);
2046 case XenbusStateClosed:
2047 if (dev->state == XenbusStateClosed)
2049 /* Missed the backend's CLOSING state -- fallthrough */
2050 case XenbusStateClosing:
2051 xenbus_frontend_closed(dev);
2056 static const struct xennet_stat {
2057 char name[ETH_GSTRING_LEN];
2059 } xennet_stats[] = {
2061 "rx_gso_checksum_fixup",
2062 offsetof(struct netfront_info, rx_gso_checksum_fixup)
2066 static int xennet_get_sset_count(struct net_device *dev, int string_set)
2068 switch (string_set) {
2070 return ARRAY_SIZE(xennet_stats);
2076 static void xennet_get_ethtool_stats(struct net_device *dev,
2077 struct ethtool_stats *stats, u64 * data)
2079 void *np = netdev_priv(dev);
2082 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2083 data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset));
2086 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
2090 switch (stringset) {
2092 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2093 memcpy(data + i * ETH_GSTRING_LEN,
2094 xennet_stats[i].name, ETH_GSTRING_LEN);
2099 static const struct ethtool_ops xennet_ethtool_ops =
2101 .get_link = ethtool_op_get_link,
2103 .get_sset_count = xennet_get_sset_count,
2104 .get_ethtool_stats = xennet_get_ethtool_stats,
2105 .get_strings = xennet_get_strings,
2109 static ssize_t show_rxbuf(struct device *dev,
2110 struct device_attribute *attr, char *buf)
2112 return sprintf(buf, "%lu\n", NET_RX_RING_SIZE);
2115 static ssize_t store_rxbuf(struct device *dev,
2116 struct device_attribute *attr,
2117 const char *buf, size_t len)
2120 unsigned long target;
2122 if (!capable(CAP_NET_ADMIN))
2125 target = simple_strtoul(buf, &endp, 0);
2129 /* rxbuf_min and rxbuf_max are no longer configurable. */
2134 static DEVICE_ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf);
2135 static DEVICE_ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf);
2136 static DEVICE_ATTR(rxbuf_cur, S_IRUGO, show_rxbuf, NULL);
2138 static struct attribute *xennet_dev_attrs[] = {
2139 &dev_attr_rxbuf_min.attr,
2140 &dev_attr_rxbuf_max.attr,
2141 &dev_attr_rxbuf_cur.attr,
2145 static const struct attribute_group xennet_dev_group = {
2146 .attrs = xennet_dev_attrs
2148 #endif /* CONFIG_SYSFS */
2150 static void xennet_bus_close(struct xenbus_device *dev)
2154 if (xenbus_read_driver_state(dev->otherend) == XenbusStateClosed)
2157 xenbus_switch_state(dev, XenbusStateClosing);
2158 ret = wait_event_timeout(module_wq,
2159 xenbus_read_driver_state(dev->otherend) ==
2160 XenbusStateClosing ||
2161 xenbus_read_driver_state(dev->otherend) ==
2162 XenbusStateClosed ||
2163 xenbus_read_driver_state(dev->otherend) ==
2168 if (xenbus_read_driver_state(dev->otherend) == XenbusStateClosed)
2172 xenbus_switch_state(dev, XenbusStateClosed);
2173 ret = wait_event_timeout(module_wq,
2174 xenbus_read_driver_state(dev->otherend) ==
2175 XenbusStateClosed ||
2176 xenbus_read_driver_state(dev->otherend) ==
2182 static int xennet_remove(struct xenbus_device *dev)
2184 struct netfront_info *info = dev_get_drvdata(&dev->dev);
2186 xennet_bus_close(dev);
2187 xennet_disconnect_backend(info);
2189 if (info->netdev->reg_state == NETREG_REGISTERED)
2190 unregister_netdev(info->netdev);
2194 xennet_destroy_queues(info);
2197 xennet_free_netdev(info->netdev);
2202 static const struct xenbus_device_id netfront_ids[] = {
2207 static struct xenbus_driver netfront_driver = {
2208 .ids = netfront_ids,
2209 .probe = netfront_probe,
2210 .remove = xennet_remove,
2211 .resume = netfront_resume,
2212 .otherend_changed = netback_changed,
2215 static int __init netif_init(void)
2220 if (!xen_has_pv_nic_devices())
2223 pr_info("Initialising Xen virtual ethernet driver\n");
2225 /* Allow as many queues as there are CPUs inut max. 8 if user has not
2226 * specified a value.
2228 if (xennet_max_queues == 0)
2229 xennet_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT,
2232 return xenbus_register_frontend(&netfront_driver);
2234 module_init(netif_init);
2237 static void __exit netif_exit(void)
2239 xenbus_unregister_driver(&netfront_driver);
2241 module_exit(netif_exit);
2243 MODULE_DESCRIPTION("Xen virtual network device frontend");
2244 MODULE_LICENSE("GPL");
2245 MODULE_ALIAS("xen:vif");
2246 MODULE_ALIAS("xennet");