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 static const struct ethtool_ops xennet_ethtool_ops;
71 #define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb))
73 #define RX_COPY_THRESHOLD 256
75 #define GRANT_INVALID_REF 0
77 #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, XEN_PAGE_SIZE)
78 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, XEN_PAGE_SIZE)
80 /* Minimum number of Rx slots (includes slot for GSO metadata). */
81 #define NET_RX_SLOTS_MIN (XEN_NETIF_NR_SLOTS_MIN + 1)
83 /* Queue name is interface name with "-qNNN" appended */
84 #define QUEUE_NAME_SIZE (IFNAMSIZ + 6)
86 /* IRQ name is queue name with "-tx" or "-rx" appended */
87 #define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3)
89 struct netfront_stats {
92 struct u64_stats_sync syncp;
97 struct netfront_queue {
98 unsigned int id; /* Queue ID, 0-based */
99 char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */
100 struct netfront_info *info;
102 struct napi_struct napi;
104 /* Split event channels support, tx_* == rx_* when using
105 * single event channel.
107 unsigned int tx_evtchn, rx_evtchn;
108 unsigned int tx_irq, rx_irq;
109 /* Only used when split event channels support is enabled */
110 char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */
111 char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */
114 struct xen_netif_tx_front_ring tx;
118 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
119 * are linked from tx_skb_freelist through skb_entry.link.
121 * NB. Freelist index entries are always going to be less than
122 * PAGE_OFFSET, whereas pointers to skbs will always be equal or
123 * greater than PAGE_OFFSET: we use this property to distinguish
129 } tx_skbs[NET_TX_RING_SIZE];
130 grant_ref_t gref_tx_head;
131 grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
132 struct page *grant_tx_page[NET_TX_RING_SIZE];
133 unsigned tx_skb_freelist;
135 spinlock_t rx_lock ____cacheline_aligned_in_smp;
136 struct xen_netif_rx_front_ring rx;
139 struct timer_list rx_refill_timer;
141 struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
142 grant_ref_t gref_rx_head;
143 grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
146 struct netfront_info {
147 struct list_head list;
148 struct net_device *netdev;
150 struct xenbus_device *xbdev;
152 /* Multi-queue support */
153 struct netfront_queue *queues;
156 struct netfront_stats __percpu *rx_stats;
157 struct netfront_stats __percpu *tx_stats;
159 atomic_t rx_gso_checksum_fixup;
162 struct netfront_rx_info {
163 struct xen_netif_rx_response rx;
164 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
167 static void skb_entry_set_link(union skb_entry *list, unsigned short id)
172 static int skb_entry_is_link(const union skb_entry *list)
174 BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link));
175 return (unsigned long)list->skb < PAGE_OFFSET;
179 * Access macros for acquiring freeing slots in tx_skbs[].
182 static void add_id_to_freelist(unsigned *head, union skb_entry *list,
185 skb_entry_set_link(&list[id], *head);
189 static unsigned short get_id_from_freelist(unsigned *head,
190 union skb_entry *list)
192 unsigned int id = *head;
193 *head = list[id].link;
197 static int xennet_rxidx(RING_IDX idx)
199 return idx & (NET_RX_RING_SIZE - 1);
202 static struct sk_buff *xennet_get_rx_skb(struct netfront_queue *queue,
205 int i = xennet_rxidx(ri);
206 struct sk_buff *skb = queue->rx_skbs[i];
207 queue->rx_skbs[i] = NULL;
211 static grant_ref_t xennet_get_rx_ref(struct netfront_queue *queue,
214 int i = xennet_rxidx(ri);
215 grant_ref_t ref = queue->grant_rx_ref[i];
216 queue->grant_rx_ref[i] = GRANT_INVALID_REF;
221 static const struct attribute_group xennet_dev_group;
224 static bool xennet_can_sg(struct net_device *dev)
226 return dev->features & NETIF_F_SG;
230 static void rx_refill_timeout(unsigned long data)
232 struct netfront_queue *queue = (struct netfront_queue *)data;
233 napi_schedule(&queue->napi);
236 static int netfront_tx_slot_available(struct netfront_queue *queue)
238 return (queue->tx.req_prod_pvt - queue->tx.rsp_cons) <
239 (NET_TX_RING_SIZE - MAX_SKB_FRAGS - 2);
242 static void xennet_maybe_wake_tx(struct netfront_queue *queue)
244 struct net_device *dev = queue->info->netdev;
245 struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, queue->id);
247 if (unlikely(netif_tx_queue_stopped(dev_queue)) &&
248 netfront_tx_slot_available(queue) &&
249 likely(netif_running(dev)))
250 netif_tx_wake_queue(netdev_get_tx_queue(dev, queue->id));
254 static struct sk_buff *xennet_alloc_one_rx_buffer(struct netfront_queue *queue)
259 skb = __netdev_alloc_skb(queue->info->netdev,
260 RX_COPY_THRESHOLD + NET_IP_ALIGN,
261 GFP_ATOMIC | __GFP_NOWARN);
265 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
270 skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE);
272 /* Align ip header to a 16 bytes boundary */
273 skb_reserve(skb, NET_IP_ALIGN);
274 skb->dev = queue->info->netdev;
280 static void xennet_alloc_rx_buffers(struct netfront_queue *queue)
282 RING_IDX req_prod = queue->rx.req_prod_pvt;
286 if (unlikely(!netif_carrier_ok(queue->info->netdev)))
289 for (req_prod = queue->rx.req_prod_pvt;
290 req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE;
296 struct xen_netif_rx_request *req;
298 skb = xennet_alloc_one_rx_buffer(queue);
304 id = xennet_rxidx(req_prod);
306 BUG_ON(queue->rx_skbs[id]);
307 queue->rx_skbs[id] = skb;
309 ref = gnttab_claim_grant_reference(&queue->gref_rx_head);
310 WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
311 queue->grant_rx_ref[id] = ref;
313 page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
315 req = RING_GET_REQUEST(&queue->rx, req_prod);
316 gnttab_page_grant_foreign_access_ref_one(ref,
317 queue->info->xbdev->otherend_id,
324 queue->rx.req_prod_pvt = req_prod;
326 /* Try again later if there are not enough requests or skb allocation
328 * Enough requests is quantified as the sum of newly created slots and
329 * the unconsumed slots at the backend.
331 if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN ||
333 mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10));
337 wmb(); /* barrier so backend seens requests */
339 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify);
341 notify_remote_via_irq(queue->rx_irq);
344 static int xennet_open(struct net_device *dev)
346 struct netfront_info *np = netdev_priv(dev);
347 unsigned int num_queues = dev->real_num_tx_queues;
349 struct netfront_queue *queue = NULL;
351 for (i = 0; i < num_queues; ++i) {
352 queue = &np->queues[i];
353 napi_enable(&queue->napi);
355 spin_lock_bh(&queue->rx_lock);
356 if (netif_carrier_ok(dev)) {
357 xennet_alloc_rx_buffers(queue);
358 queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1;
359 if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))
360 napi_schedule(&queue->napi);
362 spin_unlock_bh(&queue->rx_lock);
365 netif_tx_start_all_queues(dev);
370 static void xennet_tx_buf_gc(struct netfront_queue *queue)
377 BUG_ON(!netif_carrier_ok(queue->info->netdev));
380 prod = queue->tx.sring->rsp_prod;
381 rmb(); /* Ensure we see responses up to 'rp'. */
383 for (cons = queue->tx.rsp_cons; cons != prod; cons++) {
384 struct xen_netif_tx_response *txrsp;
386 txrsp = RING_GET_RESPONSE(&queue->tx, cons);
387 if (txrsp->status == XEN_NETIF_RSP_NULL)
391 skb = queue->tx_skbs[id].skb;
392 if (unlikely(gnttab_query_foreign_access(
393 queue->grant_tx_ref[id]) != 0)) {
394 pr_alert("%s: warning -- grant still in use by backend domain\n",
398 gnttab_end_foreign_access_ref(
399 queue->grant_tx_ref[id], GNTMAP_readonly);
400 gnttab_release_grant_reference(
401 &queue->gref_tx_head, queue->grant_tx_ref[id]);
402 queue->grant_tx_ref[id] = GRANT_INVALID_REF;
403 queue->grant_tx_page[id] = NULL;
404 add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, id);
405 dev_kfree_skb_irq(skb);
408 queue->tx.rsp_cons = prod;
410 RING_FINAL_CHECK_FOR_RESPONSES(&queue->tx, more_to_do);
411 } while (more_to_do);
413 xennet_maybe_wake_tx(queue);
416 struct xennet_gnttab_make_txreq {
417 struct netfront_queue *queue;
420 struct xen_netif_tx_request *tx; /* Last request */
424 static void xennet_tx_setup_grant(unsigned long gfn, unsigned int offset,
425 unsigned int len, void *data)
427 struct xennet_gnttab_make_txreq *info = data;
429 struct xen_netif_tx_request *tx;
431 /* convenient aliases */
432 struct page *page = info->page;
433 struct netfront_queue *queue = info->queue;
434 struct sk_buff *skb = info->skb;
436 id = get_id_from_freelist(&queue->tx_skb_freelist, queue->tx_skbs);
437 tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
438 ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
439 WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
441 gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id,
442 gfn, GNTMAP_readonly);
444 queue->tx_skbs[id].skb = skb;
445 queue->grant_tx_page[id] = page;
446 queue->grant_tx_ref[id] = ref;
455 info->size += tx->size;
458 static struct xen_netif_tx_request *xennet_make_first_txreq(
459 struct netfront_queue *queue, struct sk_buff *skb,
460 struct page *page, unsigned int offset, unsigned int len)
462 struct xennet_gnttab_make_txreq info = {
469 gnttab_for_one_grant(page, offset, len, xennet_tx_setup_grant, &info);
474 static void xennet_make_one_txreq(unsigned long gfn, unsigned int offset,
475 unsigned int len, void *data)
477 struct xennet_gnttab_make_txreq *info = data;
479 info->tx->flags |= XEN_NETTXF_more_data;
481 xennet_tx_setup_grant(gfn, offset, len, data);
484 static struct xen_netif_tx_request *xennet_make_txreqs(
485 struct netfront_queue *queue, struct xen_netif_tx_request *tx,
486 struct sk_buff *skb, struct page *page,
487 unsigned int offset, unsigned int len)
489 struct xennet_gnttab_make_txreq info = {
495 /* Skip unused frames from start of page */
496 page += offset >> PAGE_SHIFT;
497 offset &= ~PAGE_MASK;
503 gnttab_foreach_grant_in_range(page, offset, len,
504 xennet_make_one_txreq,
516 * Count how many ring slots are required to send this skb. Each frag
517 * might be a compound page.
519 static int xennet_count_skb_slots(struct sk_buff *skb)
521 int i, frags = skb_shinfo(skb)->nr_frags;
524 slots = gnttab_count_grant(offset_in_page(skb->data),
527 for (i = 0; i < frags; i++) {
528 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
529 unsigned long size = skb_frag_size(frag);
530 unsigned long offset = frag->page_offset;
532 /* Skip unused frames from start of page */
533 offset &= ~PAGE_MASK;
535 slots += gnttab_count_grant(offset, size);
541 static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb,
542 void *accel_priv, select_queue_fallback_t fallback)
544 unsigned int num_queues = dev->real_num_tx_queues;
548 /* First, check if there is only one queue */
549 if (num_queues == 1) {
552 hash = skb_get_hash(skb);
553 queue_idx = hash % num_queues;
559 #define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1)
561 static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
563 struct netfront_info *np = netdev_priv(dev);
564 struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
565 struct xen_netif_tx_request *tx, *first_tx;
573 struct netfront_queue *queue = NULL;
574 unsigned int num_queues = dev->real_num_tx_queues;
576 struct sk_buff *nskb;
578 /* Drop the packet if no queues are set up */
581 /* Determine which queue to transmit this SKB on */
582 queue_index = skb_get_queue_mapping(skb);
583 queue = &np->queues[queue_index];
585 /* If skb->len is too big for wire format, drop skb and alert
586 * user about misconfiguration.
588 if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
589 net_alert_ratelimited(
590 "xennet: skb->len = %u, too big for wire format\n",
595 slots = xennet_count_skb_slots(skb);
596 if (unlikely(slots > MAX_XEN_SKB_FRAGS + 1)) {
597 net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n",
599 if (skb_linearize(skb))
603 page = virt_to_page(skb->data);
604 offset = offset_in_page(skb->data);
606 /* The first req should be at least ETH_HLEN size or the packet will be
607 * dropped by netback.
609 if (unlikely(PAGE_SIZE - offset < ETH_HLEN)) {
610 nskb = skb_copy(skb, GFP_ATOMIC);
613 dev_kfree_skb_any(skb);
615 page = virt_to_page(skb->data);
616 offset = offset_in_page(skb->data);
619 len = skb_headlen(skb);
621 spin_lock_irqsave(&queue->tx_lock, flags);
623 if (unlikely(!netif_carrier_ok(dev) ||
624 (slots > 1 && !xennet_can_sg(dev)) ||
625 netif_needs_gso(skb, netif_skb_features(skb)))) {
626 spin_unlock_irqrestore(&queue->tx_lock, flags);
630 /* First request for the linear area. */
631 first_tx = tx = xennet_make_first_txreq(queue, skb,
634 if (offset == PAGE_SIZE) {
640 if (skb->ip_summed == CHECKSUM_PARTIAL)
642 tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
643 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
644 /* remote but checksummed. */
645 tx->flags |= XEN_NETTXF_data_validated;
647 /* Optional extra info after the first request. */
648 if (skb_shinfo(skb)->gso_size) {
649 struct xen_netif_extra_info *gso;
651 gso = (struct xen_netif_extra_info *)
652 RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
654 tx->flags |= XEN_NETTXF_extra_info;
656 gso->u.gso.size = skb_shinfo(skb)->gso_size;
657 gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ?
658 XEN_NETIF_GSO_TYPE_TCPV6 :
659 XEN_NETIF_GSO_TYPE_TCPV4;
661 gso->u.gso.features = 0;
663 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
667 /* Requests for the rest of the linear area. */
668 tx = xennet_make_txreqs(queue, tx, skb, page, offset, len);
670 /* Requests for all the frags. */
671 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
672 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
673 tx = xennet_make_txreqs(queue, tx, skb,
674 skb_frag_page(frag), frag->page_offset,
675 skb_frag_size(frag));
678 /* First request has the packet length. */
679 first_tx->size = skb->len;
681 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
683 notify_remote_via_irq(queue->tx_irq);
685 u64_stats_update_begin(&tx_stats->syncp);
686 tx_stats->bytes += skb->len;
688 u64_stats_update_end(&tx_stats->syncp);
690 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
691 xennet_tx_buf_gc(queue);
693 if (!netfront_tx_slot_available(queue))
694 netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
696 spin_unlock_irqrestore(&queue->tx_lock, flags);
701 dev->stats.tx_dropped++;
702 dev_kfree_skb_any(skb);
706 static int xennet_close(struct net_device *dev)
708 struct netfront_info *np = netdev_priv(dev);
709 unsigned int num_queues = dev->real_num_tx_queues;
711 struct netfront_queue *queue;
712 netif_tx_stop_all_queues(np->netdev);
713 for (i = 0; i < num_queues; ++i) {
714 queue = &np->queues[i];
715 napi_disable(&queue->napi);
720 static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb,
723 int new = xennet_rxidx(queue->rx.req_prod_pvt);
725 BUG_ON(queue->rx_skbs[new]);
726 queue->rx_skbs[new] = skb;
727 queue->grant_rx_ref[new] = ref;
728 RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new;
729 RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref;
730 queue->rx.req_prod_pvt++;
733 static int xennet_get_extras(struct netfront_queue *queue,
734 struct xen_netif_extra_info *extras,
738 struct xen_netif_extra_info *extra;
739 struct device *dev = &queue->info->netdev->dev;
740 RING_IDX cons = queue->rx.rsp_cons;
747 if (unlikely(cons + 1 == rp)) {
749 dev_warn(dev, "Missing extra info\n");
754 extra = (struct xen_netif_extra_info *)
755 RING_GET_RESPONSE(&queue->rx, ++cons);
757 if (unlikely(!extra->type ||
758 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
760 dev_warn(dev, "Invalid extra type: %d\n",
764 memcpy(&extras[extra->type - 1], extra,
768 skb = xennet_get_rx_skb(queue, cons);
769 ref = xennet_get_rx_ref(queue, cons);
770 xennet_move_rx_slot(queue, skb, ref);
771 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
773 queue->rx.rsp_cons = cons;
777 static int xennet_get_responses(struct netfront_queue *queue,
778 struct netfront_rx_info *rinfo, RING_IDX rp,
779 struct sk_buff_head *list)
781 struct xen_netif_rx_response *rx = &rinfo->rx;
782 struct xen_netif_extra_info *extras = rinfo->extras;
783 struct device *dev = &queue->info->netdev->dev;
784 RING_IDX cons = queue->rx.rsp_cons;
785 struct sk_buff *skb = xennet_get_rx_skb(queue, cons);
786 grant_ref_t ref = xennet_get_rx_ref(queue, cons);
787 int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD);
792 if (rx->flags & XEN_NETRXF_extra_info) {
793 err = xennet_get_extras(queue, extras, rp);
794 cons = queue->rx.rsp_cons;
798 if (unlikely(rx->status < 0 ||
799 rx->offset + rx->status > XEN_PAGE_SIZE)) {
801 dev_warn(dev, "rx->offset: %u, size: %d\n",
802 rx->offset, rx->status);
803 xennet_move_rx_slot(queue, skb, ref);
809 * This definitely indicates a bug, either in this driver or in
810 * the backend driver. In future this should flag the bad
811 * situation to the system controller to reboot the backend.
813 if (ref == GRANT_INVALID_REF) {
815 dev_warn(dev, "Bad rx response id %d.\n",
821 ret = gnttab_end_foreign_access_ref(ref, 0);
824 gnttab_release_grant_reference(&queue->gref_rx_head, ref);
826 __skb_queue_tail(list, skb);
829 if (!(rx->flags & XEN_NETRXF_more_data))
832 if (cons + slots == rp) {
834 dev_warn(dev, "Need more slots\n");
839 rx = RING_GET_RESPONSE(&queue->rx, cons + slots);
840 skb = xennet_get_rx_skb(queue, cons + slots);
841 ref = xennet_get_rx_ref(queue, cons + slots);
845 if (unlikely(slots > max)) {
847 dev_warn(dev, "Too many slots\n");
852 queue->rx.rsp_cons = cons + slots;
857 static int xennet_set_skb_gso(struct sk_buff *skb,
858 struct xen_netif_extra_info *gso)
860 if (!gso->u.gso.size) {
862 pr_warn("GSO size must not be zero\n");
866 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 &&
867 gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) {
869 pr_warn("Bad GSO type %d\n", gso->u.gso.type);
873 skb_shinfo(skb)->gso_size = gso->u.gso.size;
874 skb_shinfo(skb)->gso_type =
875 (gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ?
879 /* Header must be checked, and gso_segs computed. */
880 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
881 skb_shinfo(skb)->gso_segs = 0;
886 static RING_IDX xennet_fill_frags(struct netfront_queue *queue,
888 struct sk_buff_head *list)
890 struct skb_shared_info *shinfo = skb_shinfo(skb);
891 RING_IDX cons = queue->rx.rsp_cons;
892 struct sk_buff *nskb;
894 while ((nskb = __skb_dequeue(list))) {
895 struct xen_netif_rx_response *rx =
896 RING_GET_RESPONSE(&queue->rx, ++cons);
897 skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
899 if (shinfo->nr_frags == MAX_SKB_FRAGS) {
900 unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
902 BUG_ON(pull_to <= skb_headlen(skb));
903 __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
905 BUG_ON(shinfo->nr_frags >= MAX_SKB_FRAGS);
907 skb_add_rx_frag(skb, shinfo->nr_frags, skb_frag_page(nfrag),
908 rx->offset, rx->status, PAGE_SIZE);
910 skb_shinfo(nskb)->nr_frags = 0;
917 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
919 bool recalculate_partial_csum = false;
922 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
923 * peers can fail to set NETRXF_csum_blank when sending a GSO
924 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
925 * recalculate the partial checksum.
927 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
928 struct netfront_info *np = netdev_priv(dev);
929 atomic_inc(&np->rx_gso_checksum_fixup);
930 skb->ip_summed = CHECKSUM_PARTIAL;
931 recalculate_partial_csum = true;
934 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
935 if (skb->ip_summed != CHECKSUM_PARTIAL)
938 return skb_checksum_setup(skb, recalculate_partial_csum);
941 static int handle_incoming_queue(struct netfront_queue *queue,
942 struct sk_buff_head *rxq)
944 struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats);
945 int packets_dropped = 0;
948 while ((skb = __skb_dequeue(rxq)) != NULL) {
949 int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
951 if (pull_to > skb_headlen(skb))
952 __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
954 /* Ethernet work: Delayed to here as it peeks the header. */
955 skb->protocol = eth_type_trans(skb, queue->info->netdev);
956 skb_reset_network_header(skb);
958 if (checksum_setup(queue->info->netdev, skb)) {
961 queue->info->netdev->stats.rx_errors++;
965 u64_stats_update_begin(&rx_stats->syncp);
967 rx_stats->bytes += skb->len;
968 u64_stats_update_end(&rx_stats->syncp);
971 napi_gro_receive(&queue->napi, skb);
974 return packets_dropped;
977 static int xennet_poll(struct napi_struct *napi, int budget)
979 struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi);
980 struct net_device *dev = queue->info->netdev;
982 struct netfront_rx_info rinfo;
983 struct xen_netif_rx_response *rx = &rinfo.rx;
984 struct xen_netif_extra_info *extras = rinfo.extras;
987 struct sk_buff_head rxq;
988 struct sk_buff_head errq;
989 struct sk_buff_head tmpq;
992 spin_lock(&queue->rx_lock);
994 skb_queue_head_init(&rxq);
995 skb_queue_head_init(&errq);
996 skb_queue_head_init(&tmpq);
998 rp = queue->rx.sring->rsp_prod;
999 rmb(); /* Ensure we see queued responses up to 'rp'. */
1001 i = queue->rx.rsp_cons;
1003 while ((i != rp) && (work_done < budget)) {
1004 memcpy(rx, RING_GET_RESPONSE(&queue->rx, i), sizeof(*rx));
1005 memset(extras, 0, sizeof(rinfo.extras));
1007 err = xennet_get_responses(queue, &rinfo, rp, &tmpq);
1009 if (unlikely(err)) {
1011 while ((skb = __skb_dequeue(&tmpq)))
1012 __skb_queue_tail(&errq, skb);
1013 dev->stats.rx_errors++;
1014 i = queue->rx.rsp_cons;
1018 skb = __skb_dequeue(&tmpq);
1020 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1021 struct xen_netif_extra_info *gso;
1022 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1024 if (unlikely(xennet_set_skb_gso(skb, gso))) {
1025 __skb_queue_head(&tmpq, skb);
1026 queue->rx.rsp_cons += skb_queue_len(&tmpq);
1031 NETFRONT_SKB_CB(skb)->pull_to = rx->status;
1032 if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
1033 NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
1035 skb_shinfo(skb)->frags[0].page_offset = rx->offset;
1036 skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
1037 skb->data_len = rx->status;
1038 skb->len += rx->status;
1040 i = xennet_fill_frags(queue, skb, &tmpq);
1042 if (rx->flags & XEN_NETRXF_csum_blank)
1043 skb->ip_summed = CHECKSUM_PARTIAL;
1044 else if (rx->flags & XEN_NETRXF_data_validated)
1045 skb->ip_summed = CHECKSUM_UNNECESSARY;
1047 __skb_queue_tail(&rxq, skb);
1049 queue->rx.rsp_cons = ++i;
1053 __skb_queue_purge(&errq);
1055 work_done -= handle_incoming_queue(queue, &rxq);
1057 xennet_alloc_rx_buffers(queue);
1059 if (work_done < budget) {
1062 napi_complete_done(napi, work_done);
1064 RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do);
1066 napi_schedule(napi);
1069 spin_unlock(&queue->rx_lock);
1074 static int xennet_change_mtu(struct net_device *dev, int mtu)
1076 int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN;
1084 static void xennet_get_stats64(struct net_device *dev,
1085 struct rtnl_link_stats64 *tot)
1087 struct netfront_info *np = netdev_priv(dev);
1090 for_each_possible_cpu(cpu) {
1091 struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu);
1092 struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu);
1093 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1097 start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
1098 tx_packets = tx_stats->packets;
1099 tx_bytes = tx_stats->bytes;
1100 } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
1103 start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
1104 rx_packets = rx_stats->packets;
1105 rx_bytes = rx_stats->bytes;
1106 } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
1108 tot->rx_packets += rx_packets;
1109 tot->tx_packets += tx_packets;
1110 tot->rx_bytes += rx_bytes;
1111 tot->tx_bytes += tx_bytes;
1114 tot->rx_errors = dev->stats.rx_errors;
1115 tot->tx_dropped = dev->stats.tx_dropped;
1118 static void xennet_release_tx_bufs(struct netfront_queue *queue)
1120 struct sk_buff *skb;
1123 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1124 /* Skip over entries which are actually freelist references */
1125 if (skb_entry_is_link(&queue->tx_skbs[i]))
1128 skb = queue->tx_skbs[i].skb;
1129 get_page(queue->grant_tx_page[i]);
1130 gnttab_end_foreign_access(queue->grant_tx_ref[i],
1132 (unsigned long)page_address(queue->grant_tx_page[i]));
1133 queue->grant_tx_page[i] = NULL;
1134 queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1135 add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, i);
1136 dev_kfree_skb_irq(skb);
1140 static void xennet_release_rx_bufs(struct netfront_queue *queue)
1144 spin_lock_bh(&queue->rx_lock);
1146 for (id = 0; id < NET_RX_RING_SIZE; id++) {
1147 struct sk_buff *skb;
1150 skb = queue->rx_skbs[id];
1154 ref = queue->grant_rx_ref[id];
1155 if (ref == GRANT_INVALID_REF)
1158 page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
1160 /* gnttab_end_foreign_access() needs a page ref until
1161 * foreign access is ended (which may be deferred).
1164 gnttab_end_foreign_access(ref, 0,
1165 (unsigned long)page_address(page));
1166 queue->grant_rx_ref[id] = GRANT_INVALID_REF;
1171 spin_unlock_bh(&queue->rx_lock);
1174 static netdev_features_t xennet_fix_features(struct net_device *dev,
1175 netdev_features_t features)
1177 struct netfront_info *np = netdev_priv(dev);
1179 if (features & NETIF_F_SG &&
1180 !xenbus_read_unsigned(np->xbdev->otherend, "feature-sg", 0))
1181 features &= ~NETIF_F_SG;
1183 if (features & NETIF_F_IPV6_CSUM &&
1184 !xenbus_read_unsigned(np->xbdev->otherend,
1185 "feature-ipv6-csum-offload", 0))
1186 features &= ~NETIF_F_IPV6_CSUM;
1188 if (features & NETIF_F_TSO &&
1189 !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv4", 0))
1190 features &= ~NETIF_F_TSO;
1192 if (features & NETIF_F_TSO6 &&
1193 !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv6", 0))
1194 features &= ~NETIF_F_TSO6;
1199 static int xennet_set_features(struct net_device *dev,
1200 netdev_features_t features)
1202 if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1203 netdev_info(dev, "Reducing MTU because no SG offload");
1204 dev->mtu = ETH_DATA_LEN;
1210 static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1212 struct netfront_queue *queue = dev_id;
1213 unsigned long flags;
1215 spin_lock_irqsave(&queue->tx_lock, flags);
1216 xennet_tx_buf_gc(queue);
1217 spin_unlock_irqrestore(&queue->tx_lock, flags);
1222 static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1224 struct netfront_queue *queue = dev_id;
1225 struct net_device *dev = queue->info->netdev;
1227 if (likely(netif_carrier_ok(dev) &&
1228 RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)))
1229 napi_schedule(&queue->napi);
1234 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1236 xennet_tx_interrupt(irq, dev_id);
1237 xennet_rx_interrupt(irq, dev_id);
1241 #ifdef CONFIG_NET_POLL_CONTROLLER
1242 static void xennet_poll_controller(struct net_device *dev)
1244 /* Poll each queue */
1245 struct netfront_info *info = netdev_priv(dev);
1246 unsigned int num_queues = dev->real_num_tx_queues;
1248 for (i = 0; i < num_queues; ++i)
1249 xennet_interrupt(0, &info->queues[i]);
1253 static const struct net_device_ops xennet_netdev_ops = {
1254 .ndo_open = xennet_open,
1255 .ndo_stop = xennet_close,
1256 .ndo_start_xmit = xennet_start_xmit,
1257 .ndo_change_mtu = xennet_change_mtu,
1258 .ndo_get_stats64 = xennet_get_stats64,
1259 .ndo_set_mac_address = eth_mac_addr,
1260 .ndo_validate_addr = eth_validate_addr,
1261 .ndo_fix_features = xennet_fix_features,
1262 .ndo_set_features = xennet_set_features,
1263 .ndo_select_queue = xennet_select_queue,
1264 #ifdef CONFIG_NET_POLL_CONTROLLER
1265 .ndo_poll_controller = xennet_poll_controller,
1269 static void xennet_free_netdev(struct net_device *netdev)
1271 struct netfront_info *np = netdev_priv(netdev);
1273 free_percpu(np->rx_stats);
1274 free_percpu(np->tx_stats);
1275 free_netdev(netdev);
1278 static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1281 struct net_device *netdev;
1282 struct netfront_info *np;
1284 netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues);
1286 return ERR_PTR(-ENOMEM);
1288 np = netdev_priv(netdev);
1294 np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1295 if (np->rx_stats == NULL)
1297 np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1298 if (np->tx_stats == NULL)
1301 netdev->netdev_ops = &xennet_netdev_ops;
1303 netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1305 netdev->hw_features = NETIF_F_SG |
1307 NETIF_F_TSO | NETIF_F_TSO6;
1310 * Assume that all hw features are available for now. This set
1311 * will be adjusted by the call to netdev_update_features() in
1312 * xennet_connect() which is the earliest point where we can
1313 * negotiate with the backend regarding supported features.
1315 netdev->features |= netdev->hw_features;
1317 netdev->ethtool_ops = &xennet_ethtool_ops;
1318 netdev->min_mtu = 0;
1319 netdev->max_mtu = XEN_NETIF_MAX_TX_SIZE;
1320 SET_NETDEV_DEV(netdev, &dev->dev);
1322 np->netdev = netdev;
1324 netif_carrier_off(netdev);
1329 xennet_free_netdev(netdev);
1330 return ERR_PTR(err);
1334 * Entry point to this code when a new device is created. Allocate the basic
1335 * structures and the ring buffers for communication with the backend, and
1336 * inform the backend of the appropriate details for those.
1338 static int netfront_probe(struct xenbus_device *dev,
1339 const struct xenbus_device_id *id)
1342 struct net_device *netdev;
1343 struct netfront_info *info;
1345 netdev = xennet_create_dev(dev);
1346 if (IS_ERR(netdev)) {
1347 err = PTR_ERR(netdev);
1348 xenbus_dev_fatal(dev, err, "creating netdev");
1352 info = netdev_priv(netdev);
1353 dev_set_drvdata(&dev->dev, info);
1355 info->netdev->sysfs_groups[0] = &xennet_dev_group;
1357 err = register_netdev(info->netdev);
1359 pr_warn("%s: register_netdev err=%d\n", __func__, err);
1366 xennet_free_netdev(netdev);
1367 dev_set_drvdata(&dev->dev, NULL);
1371 static void xennet_end_access(int ref, void *page)
1373 /* This frees the page as a side-effect */
1374 if (ref != GRANT_INVALID_REF)
1375 gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1378 static void xennet_disconnect_backend(struct netfront_info *info)
1381 unsigned int num_queues = info->netdev->real_num_tx_queues;
1383 netif_carrier_off(info->netdev);
1385 for (i = 0; i < num_queues && info->queues; ++i) {
1386 struct netfront_queue *queue = &info->queues[i];
1388 del_timer_sync(&queue->rx_refill_timer);
1390 if (queue->tx_irq && (queue->tx_irq == queue->rx_irq))
1391 unbind_from_irqhandler(queue->tx_irq, queue);
1392 if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) {
1393 unbind_from_irqhandler(queue->tx_irq, queue);
1394 unbind_from_irqhandler(queue->rx_irq, queue);
1396 queue->tx_evtchn = queue->rx_evtchn = 0;
1397 queue->tx_irq = queue->rx_irq = 0;
1399 if (netif_running(info->netdev))
1400 napi_synchronize(&queue->napi);
1402 xennet_release_tx_bufs(queue);
1403 xennet_release_rx_bufs(queue);
1404 gnttab_free_grant_references(queue->gref_tx_head);
1405 gnttab_free_grant_references(queue->gref_rx_head);
1407 /* End access and free the pages */
1408 xennet_end_access(queue->tx_ring_ref, queue->tx.sring);
1409 xennet_end_access(queue->rx_ring_ref, queue->rx.sring);
1411 queue->tx_ring_ref = GRANT_INVALID_REF;
1412 queue->rx_ring_ref = GRANT_INVALID_REF;
1413 queue->tx.sring = NULL;
1414 queue->rx.sring = NULL;
1419 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1420 * driver restart. We tear down our netif structure and recreate it, but
1421 * leave the device-layer structures intact so that this is transparent to the
1422 * rest of the kernel.
1424 static int netfront_resume(struct xenbus_device *dev)
1426 struct netfront_info *info = dev_get_drvdata(&dev->dev);
1428 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1430 xennet_disconnect_backend(info);
1434 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1436 char *s, *e, *macstr;
1439 macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1441 return PTR_ERR(macstr);
1443 for (i = 0; i < ETH_ALEN; i++) {
1444 mac[i] = simple_strtoul(s, &e, 16);
1445 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1456 static int setup_netfront_single(struct netfront_queue *queue)
1460 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1464 err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1466 0, queue->info->netdev->name, queue);
1469 queue->rx_evtchn = queue->tx_evtchn;
1470 queue->rx_irq = queue->tx_irq = err;
1475 xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1476 queue->tx_evtchn = 0;
1481 static int setup_netfront_split(struct netfront_queue *queue)
1485 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1488 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn);
1490 goto alloc_rx_evtchn_fail;
1492 snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
1493 "%s-tx", queue->name);
1494 err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1495 xennet_tx_interrupt,
1496 0, queue->tx_irq_name, queue);
1499 queue->tx_irq = err;
1501 snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
1502 "%s-rx", queue->name);
1503 err = bind_evtchn_to_irqhandler(queue->rx_evtchn,
1504 xennet_rx_interrupt,
1505 0, queue->rx_irq_name, queue);
1508 queue->rx_irq = err;
1513 unbind_from_irqhandler(queue->tx_irq, queue);
1516 xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn);
1517 queue->rx_evtchn = 0;
1518 alloc_rx_evtchn_fail:
1519 xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1520 queue->tx_evtchn = 0;
1525 static int setup_netfront(struct xenbus_device *dev,
1526 struct netfront_queue *queue, unsigned int feature_split_evtchn)
1528 struct xen_netif_tx_sring *txs;
1529 struct xen_netif_rx_sring *rxs;
1533 queue->tx_ring_ref = GRANT_INVALID_REF;
1534 queue->rx_ring_ref = GRANT_INVALID_REF;
1535 queue->rx.sring = NULL;
1536 queue->tx.sring = NULL;
1538 txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1541 xenbus_dev_fatal(dev, err, "allocating tx ring page");
1544 SHARED_RING_INIT(txs);
1545 FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE);
1547 err = xenbus_grant_ring(dev, txs, 1, &gref);
1549 goto grant_tx_ring_fail;
1550 queue->tx_ring_ref = gref;
1552 rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1555 xenbus_dev_fatal(dev, err, "allocating rx ring page");
1556 goto alloc_rx_ring_fail;
1558 SHARED_RING_INIT(rxs);
1559 FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE);
1561 err = xenbus_grant_ring(dev, rxs, 1, &gref);
1563 goto grant_rx_ring_fail;
1564 queue->rx_ring_ref = gref;
1566 if (feature_split_evtchn)
1567 err = setup_netfront_split(queue);
1568 /* setup single event channel if
1569 * a) feature-split-event-channels == 0
1570 * b) feature-split-event-channels == 1 but failed to setup
1572 if (!feature_split_evtchn || (feature_split_evtchn && err))
1573 err = setup_netfront_single(queue);
1576 goto alloc_evtchn_fail;
1580 /* If we fail to setup netfront, it is safe to just revoke access to
1581 * granted pages because backend is not accessing it at this point.
1584 gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0);
1586 free_page((unsigned long)rxs);
1588 gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0);
1590 free_page((unsigned long)txs);
1595 /* Queue-specific initialisation
1596 * This used to be done in xennet_create_dev() but must now
1599 static int xennet_init_queue(struct netfront_queue *queue)
1604 spin_lock_init(&queue->tx_lock);
1605 spin_lock_init(&queue->rx_lock);
1607 setup_timer(&queue->rx_refill_timer, rx_refill_timeout,
1608 (unsigned long)queue);
1610 snprintf(queue->name, sizeof(queue->name), "%s-q%u",
1611 queue->info->netdev->name, queue->id);
1613 /* Initialise tx_skbs as a free chain containing every entry. */
1614 queue->tx_skb_freelist = 0;
1615 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1616 skb_entry_set_link(&queue->tx_skbs[i], i+1);
1617 queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1618 queue->grant_tx_page[i] = NULL;
1621 /* Clear out rx_skbs */
1622 for (i = 0; i < NET_RX_RING_SIZE; i++) {
1623 queue->rx_skbs[i] = NULL;
1624 queue->grant_rx_ref[i] = GRANT_INVALID_REF;
1627 /* A grant for every tx ring slot */
1628 if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
1629 &queue->gref_tx_head) < 0) {
1630 pr_alert("can't alloc tx grant refs\n");
1635 /* A grant for every rx ring slot */
1636 if (gnttab_alloc_grant_references(NET_RX_RING_SIZE,
1637 &queue->gref_rx_head) < 0) {
1638 pr_alert("can't alloc rx grant refs\n");
1646 gnttab_free_grant_references(queue->gref_tx_head);
1651 static int write_queue_xenstore_keys(struct netfront_queue *queue,
1652 struct xenbus_transaction *xbt, int write_hierarchical)
1654 /* Write the queue-specific keys into XenStore in the traditional
1655 * way for a single queue, or in a queue subkeys for multiple
1658 struct xenbus_device *dev = queue->info->xbdev;
1660 const char *message;
1664 /* Choose the correct place to write the keys */
1665 if (write_hierarchical) {
1666 pathsize = strlen(dev->nodename) + 10;
1667 path = kzalloc(pathsize, GFP_KERNEL);
1670 message = "out of memory while writing ring references";
1673 snprintf(path, pathsize, "%s/queue-%u",
1674 dev->nodename, queue->id);
1676 path = (char *)dev->nodename;
1679 /* Write ring references */
1680 err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u",
1681 queue->tx_ring_ref);
1683 message = "writing tx-ring-ref";
1687 err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u",
1688 queue->rx_ring_ref);
1690 message = "writing rx-ring-ref";
1694 /* Write event channels; taking into account both shared
1695 * and split event channel scenarios.
1697 if (queue->tx_evtchn == queue->rx_evtchn) {
1698 /* Shared event channel */
1699 err = xenbus_printf(*xbt, path,
1700 "event-channel", "%u", queue->tx_evtchn);
1702 message = "writing event-channel";
1706 /* Split event channels */
1707 err = xenbus_printf(*xbt, path,
1708 "event-channel-tx", "%u", queue->tx_evtchn);
1710 message = "writing event-channel-tx";
1714 err = xenbus_printf(*xbt, path,
1715 "event-channel-rx", "%u", queue->rx_evtchn);
1717 message = "writing event-channel-rx";
1722 if (write_hierarchical)
1727 if (write_hierarchical)
1729 xenbus_dev_fatal(dev, err, "%s", message);
1733 static void xennet_destroy_queues(struct netfront_info *info)
1739 for (i = 0; i < info->netdev->real_num_tx_queues; i++) {
1740 struct netfront_queue *queue = &info->queues[i];
1742 if (netif_running(info->netdev))
1743 napi_disable(&queue->napi);
1744 netif_napi_del(&queue->napi);
1749 kfree(info->queues);
1750 info->queues = NULL;
1753 static int xennet_create_queues(struct netfront_info *info,
1754 unsigned int *num_queues)
1759 info->queues = kcalloc(*num_queues, sizeof(struct netfront_queue),
1766 for (i = 0; i < *num_queues; i++) {
1767 struct netfront_queue *queue = &info->queues[i];
1772 ret = xennet_init_queue(queue);
1774 dev_warn(&info->netdev->dev,
1775 "only created %d queues\n", i);
1780 netif_napi_add(queue->info->netdev, &queue->napi,
1782 if (netif_running(info->netdev))
1783 napi_enable(&queue->napi);
1786 netif_set_real_num_tx_queues(info->netdev, *num_queues);
1790 if (*num_queues == 0) {
1791 dev_err(&info->netdev->dev, "no queues\n");
1797 /* Common code used when first setting up, and when resuming. */
1798 static int talk_to_netback(struct xenbus_device *dev,
1799 struct netfront_info *info)
1801 const char *message;
1802 struct xenbus_transaction xbt;
1804 unsigned int feature_split_evtchn;
1806 unsigned int max_queues = 0;
1807 struct netfront_queue *queue = NULL;
1808 unsigned int num_queues = 1;
1810 info->netdev->irq = 0;
1812 /* Check if backend supports multiple queues */
1813 max_queues = xenbus_read_unsigned(info->xbdev->otherend,
1814 "multi-queue-max-queues", 1);
1815 num_queues = min(max_queues, xennet_max_queues);
1817 /* Check feature-split-event-channels */
1818 feature_split_evtchn = xenbus_read_unsigned(info->xbdev->otherend,
1819 "feature-split-event-channels", 0);
1821 /* Read mac addr. */
1822 err = xen_net_read_mac(dev, info->netdev->dev_addr);
1824 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1829 xennet_destroy_queues(info);
1831 err = xennet_create_queues(info, &num_queues);
1833 xenbus_dev_fatal(dev, err, "creating queues");
1834 kfree(info->queues);
1835 info->queues = NULL;
1839 /* Create shared ring, alloc event channel -- for each queue */
1840 for (i = 0; i < num_queues; ++i) {
1841 queue = &info->queues[i];
1842 err = setup_netfront(dev, queue, feature_split_evtchn);
1848 err = xenbus_transaction_start(&xbt);
1850 xenbus_dev_fatal(dev, err, "starting transaction");
1854 if (xenbus_exists(XBT_NIL,
1855 info->xbdev->otherend, "multi-queue-max-queues")) {
1856 /* Write the number of queues */
1857 err = xenbus_printf(xbt, dev->nodename,
1858 "multi-queue-num-queues", "%u", num_queues);
1860 message = "writing multi-queue-num-queues";
1861 goto abort_transaction_no_dev_fatal;
1865 if (num_queues == 1) {
1866 err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */
1868 goto abort_transaction_no_dev_fatal;
1870 /* Write the keys for each queue */
1871 for (i = 0; i < num_queues; ++i) {
1872 queue = &info->queues[i];
1873 err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */
1875 goto abort_transaction_no_dev_fatal;
1879 /* The remaining keys are not queue-specific */
1880 err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1883 message = "writing request-rx-copy";
1884 goto abort_transaction;
1887 err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1889 message = "writing feature-rx-notify";
1890 goto abort_transaction;
1893 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1895 message = "writing feature-sg";
1896 goto abort_transaction;
1899 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1901 message = "writing feature-gso-tcpv4";
1902 goto abort_transaction;
1905 err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1");
1907 message = "writing feature-gso-tcpv6";
1908 goto abort_transaction;
1911 err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload",
1914 message = "writing feature-ipv6-csum-offload";
1915 goto abort_transaction;
1918 err = xenbus_transaction_end(xbt, 0);
1922 xenbus_dev_fatal(dev, err, "completing transaction");
1929 xenbus_dev_fatal(dev, err, "%s", message);
1930 abort_transaction_no_dev_fatal:
1931 xenbus_transaction_end(xbt, 1);
1933 xennet_disconnect_backend(info);
1934 xennet_destroy_queues(info);
1936 unregister_netdev(info->netdev);
1937 xennet_free_netdev(info->netdev);
1941 static int xennet_connect(struct net_device *dev)
1943 struct netfront_info *np = netdev_priv(dev);
1944 unsigned int num_queues = 0;
1947 struct netfront_queue *queue = NULL;
1949 if (!xenbus_read_unsigned(np->xbdev->otherend, "feature-rx-copy", 0)) {
1951 "backend does not support copying receive path\n");
1955 err = talk_to_netback(np->xbdev, np);
1959 /* talk_to_netback() sets the correct number of queues */
1960 num_queues = dev->real_num_tx_queues;
1963 netdev_update_features(dev);
1967 * All public and private state should now be sane. Get
1968 * ready to start sending and receiving packets and give the driver
1969 * domain a kick because we've probably just requeued some
1972 netif_carrier_on(np->netdev);
1973 for (j = 0; j < num_queues; ++j) {
1974 queue = &np->queues[j];
1976 notify_remote_via_irq(queue->tx_irq);
1977 if (queue->tx_irq != queue->rx_irq)
1978 notify_remote_via_irq(queue->rx_irq);
1980 spin_lock_irq(&queue->tx_lock);
1981 xennet_tx_buf_gc(queue);
1982 spin_unlock_irq(&queue->tx_lock);
1984 spin_lock_bh(&queue->rx_lock);
1985 xennet_alloc_rx_buffers(queue);
1986 spin_unlock_bh(&queue->rx_lock);
1993 * Callback received when the backend's state changes.
1995 static void netback_changed(struct xenbus_device *dev,
1996 enum xenbus_state backend_state)
1998 struct netfront_info *np = dev_get_drvdata(&dev->dev);
1999 struct net_device *netdev = np->netdev;
2001 dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
2003 switch (backend_state) {
2004 case XenbusStateInitialising:
2005 case XenbusStateInitialised:
2006 case XenbusStateReconfiguring:
2007 case XenbusStateReconfigured:
2008 case XenbusStateUnknown:
2011 case XenbusStateInitWait:
2012 if (dev->state != XenbusStateInitialising)
2014 if (xennet_connect(netdev) != 0)
2016 xenbus_switch_state(dev, XenbusStateConnected);
2019 case XenbusStateConnected:
2020 netdev_notify_peers(netdev);
2023 case XenbusStateClosed:
2024 if (dev->state == XenbusStateClosed)
2026 /* Missed the backend's CLOSING state -- fallthrough */
2027 case XenbusStateClosing:
2028 xenbus_frontend_closed(dev);
2033 static const struct xennet_stat {
2034 char name[ETH_GSTRING_LEN];
2036 } xennet_stats[] = {
2038 "rx_gso_checksum_fixup",
2039 offsetof(struct netfront_info, rx_gso_checksum_fixup)
2043 static int xennet_get_sset_count(struct net_device *dev, int string_set)
2045 switch (string_set) {
2047 return ARRAY_SIZE(xennet_stats);
2053 static void xennet_get_ethtool_stats(struct net_device *dev,
2054 struct ethtool_stats *stats, u64 * data)
2056 void *np = netdev_priv(dev);
2059 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2060 data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset));
2063 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
2067 switch (stringset) {
2069 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2070 memcpy(data + i * ETH_GSTRING_LEN,
2071 xennet_stats[i].name, ETH_GSTRING_LEN);
2076 static const struct ethtool_ops xennet_ethtool_ops =
2078 .get_link = ethtool_op_get_link,
2080 .get_sset_count = xennet_get_sset_count,
2081 .get_ethtool_stats = xennet_get_ethtool_stats,
2082 .get_strings = xennet_get_strings,
2086 static ssize_t show_rxbuf(struct device *dev,
2087 struct device_attribute *attr, char *buf)
2089 return sprintf(buf, "%lu\n", NET_RX_RING_SIZE);
2092 static ssize_t store_rxbuf(struct device *dev,
2093 struct device_attribute *attr,
2094 const char *buf, size_t len)
2097 unsigned long target;
2099 if (!capable(CAP_NET_ADMIN))
2102 target = simple_strtoul(buf, &endp, 0);
2106 /* rxbuf_min and rxbuf_max are no longer configurable. */
2111 static DEVICE_ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf);
2112 static DEVICE_ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf);
2113 static DEVICE_ATTR(rxbuf_cur, S_IRUGO, show_rxbuf, NULL);
2115 static struct attribute *xennet_dev_attrs[] = {
2116 &dev_attr_rxbuf_min.attr,
2117 &dev_attr_rxbuf_max.attr,
2118 &dev_attr_rxbuf_cur.attr,
2122 static const struct attribute_group xennet_dev_group = {
2123 .attrs = xennet_dev_attrs
2125 #endif /* CONFIG_SYSFS */
2127 static int xennet_remove(struct xenbus_device *dev)
2129 struct netfront_info *info = dev_get_drvdata(&dev->dev);
2131 dev_dbg(&dev->dev, "%s\n", dev->nodename);
2133 xennet_disconnect_backend(info);
2135 unregister_netdev(info->netdev);
2138 xennet_destroy_queues(info);
2139 xennet_free_netdev(info->netdev);
2144 static const struct xenbus_device_id netfront_ids[] = {
2149 static struct xenbus_driver netfront_driver = {
2150 .ids = netfront_ids,
2151 .probe = netfront_probe,
2152 .remove = xennet_remove,
2153 .resume = netfront_resume,
2154 .otherend_changed = netback_changed,
2157 static int __init netif_init(void)
2162 if (!xen_has_pv_nic_devices())
2165 pr_info("Initialising Xen virtual ethernet driver\n");
2167 /* Allow as many queues as there are CPUs if user has not
2168 * specified a value.
2170 if (xennet_max_queues == 0)
2171 xennet_max_queues = num_online_cpus();
2173 return xenbus_register_frontend(&netfront_driver);
2175 module_init(netif_init);
2178 static void __exit netif_exit(void)
2180 xenbus_unregister_driver(&netfront_driver);
2182 module_exit(netif_exit);
2184 MODULE_DESCRIPTION("Xen virtual network device frontend");
2185 MODULE_LICENSE("GPL");
2186 MODULE_ALIAS("xen:vif");
2187 MODULE_ALIAS("xennet");