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, PAGE_SIZE)
78 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, 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;
285 if (unlikely(!netif_carrier_ok(queue->info->netdev)))
288 for (req_prod = queue->rx.req_prod_pvt;
289 req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE;
295 struct xen_netif_rx_request *req;
297 skb = xennet_alloc_one_rx_buffer(queue);
301 id = xennet_rxidx(req_prod);
303 BUG_ON(queue->rx_skbs[id]);
304 queue->rx_skbs[id] = skb;
306 ref = gnttab_claim_grant_reference(&queue->gref_rx_head);
307 BUG_ON((signed short)ref < 0);
308 queue->grant_rx_ref[id] = ref;
310 pfn = page_to_pfn(skb_frag_page(&skb_shinfo(skb)->frags[0]));
312 req = RING_GET_REQUEST(&queue->rx, req_prod);
313 gnttab_grant_foreign_access_ref(ref,
314 queue->info->xbdev->otherend_id,
322 queue->rx.req_prod_pvt = req_prod;
324 /* Not enough requests? Try again later. */
325 if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN) {
326 mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10));
330 wmb(); /* barrier so backend seens requests */
332 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify);
334 notify_remote_via_irq(queue->rx_irq);
337 static int xennet_open(struct net_device *dev)
339 struct netfront_info *np = netdev_priv(dev);
340 unsigned int num_queues = dev->real_num_tx_queues;
342 struct netfront_queue *queue = NULL;
344 for (i = 0; i < num_queues; ++i) {
345 queue = &np->queues[i];
346 napi_enable(&queue->napi);
348 spin_lock_bh(&queue->rx_lock);
349 if (netif_carrier_ok(dev)) {
350 xennet_alloc_rx_buffers(queue);
351 queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1;
352 if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))
353 napi_schedule(&queue->napi);
355 spin_unlock_bh(&queue->rx_lock);
358 netif_tx_start_all_queues(dev);
363 static void xennet_tx_buf_gc(struct netfront_queue *queue)
369 BUG_ON(!netif_carrier_ok(queue->info->netdev));
372 prod = queue->tx.sring->rsp_prod;
373 rmb(); /* Ensure we see responses up to 'rp'. */
375 for (cons = queue->tx.rsp_cons; cons != prod; cons++) {
376 struct xen_netif_tx_response *txrsp;
378 txrsp = RING_GET_RESPONSE(&queue->tx, cons);
379 if (txrsp->status == XEN_NETIF_RSP_NULL)
383 skb = queue->tx_skbs[id].skb;
384 if (unlikely(gnttab_query_foreign_access(
385 queue->grant_tx_ref[id]) != 0)) {
386 pr_alert("%s: warning -- grant still in use by backend domain\n",
390 gnttab_end_foreign_access_ref(
391 queue->grant_tx_ref[id], GNTMAP_readonly);
392 gnttab_release_grant_reference(
393 &queue->gref_tx_head, queue->grant_tx_ref[id]);
394 queue->grant_tx_ref[id] = GRANT_INVALID_REF;
395 queue->grant_tx_page[id] = NULL;
396 add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, id);
397 dev_kfree_skb_irq(skb);
400 queue->tx.rsp_cons = prod;
403 * Set a new event, then check for race with update of tx_cons.
404 * Note that it is essential to schedule a callback, no matter
405 * how few buffers are pending. Even if there is space in the
406 * transmit ring, higher layers may be blocked because too much
407 * data is outstanding: in such cases notification from Xen is
408 * likely to be the only kick that we'll get.
410 queue->tx.sring->rsp_event =
411 prod + ((queue->tx.sring->req_prod - prod) >> 1) + 1;
412 mb(); /* update shared area */
413 } while ((cons == prod) && (prod != queue->tx.sring->rsp_prod));
415 xennet_maybe_wake_tx(queue);
418 static struct xen_netif_tx_request *xennet_make_one_txreq(
419 struct netfront_queue *queue, struct sk_buff *skb,
420 struct page *page, unsigned int offset, unsigned int len)
423 struct xen_netif_tx_request *tx;
426 len = min_t(unsigned int, PAGE_SIZE - offset, len);
428 id = get_id_from_freelist(&queue->tx_skb_freelist, queue->tx_skbs);
429 tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
430 ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
431 BUG_ON((signed short)ref < 0);
433 gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id,
434 page_to_mfn(page), GNTMAP_readonly);
436 queue->tx_skbs[id].skb = skb;
437 queue->grant_tx_page[id] = page;
438 queue->grant_tx_ref[id] = ref;
449 static struct xen_netif_tx_request *xennet_make_txreqs(
450 struct netfront_queue *queue, struct xen_netif_tx_request *tx,
451 struct sk_buff *skb, struct page *page,
452 unsigned int offset, unsigned int len)
454 /* Skip unused frames from start of page */
455 page += offset >> PAGE_SHIFT;
456 offset &= ~PAGE_MASK;
459 tx->flags |= XEN_NETTXF_more_data;
460 tx = xennet_make_one_txreq(queue, skb_get(skb),
471 * Count how many ring slots are required to send this skb. Each frag
472 * might be a compound page.
474 static int xennet_count_skb_slots(struct sk_buff *skb)
476 int i, frags = skb_shinfo(skb)->nr_frags;
479 pages = PFN_UP(offset_in_page(skb->data) + skb_headlen(skb));
481 for (i = 0; i < frags; i++) {
482 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
483 unsigned long size = skb_frag_size(frag);
484 unsigned long offset = frag->page_offset;
486 /* Skip unused frames from start of page */
487 offset &= ~PAGE_MASK;
489 pages += PFN_UP(offset + size);
495 static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb,
496 void *accel_priv, select_queue_fallback_t fallback)
498 unsigned int num_queues = dev->real_num_tx_queues;
502 /* First, check if there is only one queue */
503 if (num_queues == 1) {
506 hash = skb_get_hash(skb);
507 queue_idx = hash % num_queues;
513 static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
515 struct netfront_info *np = netdev_priv(dev);
516 struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
517 struct xen_netif_tx_request *tx, *first_tx;
525 struct netfront_queue *queue = NULL;
526 unsigned int num_queues = dev->real_num_tx_queues;
529 /* Drop the packet if no queues are set up */
532 /* Determine which queue to transmit this SKB on */
533 queue_index = skb_get_queue_mapping(skb);
534 queue = &np->queues[queue_index];
536 /* If skb->len is too big for wire format, drop skb and alert
537 * user about misconfiguration.
539 if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
540 net_alert_ratelimited(
541 "xennet: skb->len = %u, too big for wire format\n",
546 slots = xennet_count_skb_slots(skb);
547 if (unlikely(slots > MAX_SKB_FRAGS + 1)) {
548 net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n",
550 if (skb_linearize(skb))
554 page = virt_to_page(skb->data);
555 offset = offset_in_page(skb->data);
556 len = skb_headlen(skb);
558 spin_lock_irqsave(&queue->tx_lock, flags);
560 if (unlikely(!netif_carrier_ok(dev) ||
561 (slots > 1 && !xennet_can_sg(dev)) ||
562 netif_needs_gso(skb, netif_skb_features(skb)))) {
563 spin_unlock_irqrestore(&queue->tx_lock, flags);
567 /* First request for the linear area. */
568 first_tx = tx = xennet_make_one_txreq(queue, skb,
574 if (skb->ip_summed == CHECKSUM_PARTIAL)
576 tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
577 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
578 /* remote but checksummed. */
579 tx->flags |= XEN_NETTXF_data_validated;
581 /* Optional extra info after the first request. */
582 if (skb_shinfo(skb)->gso_size) {
583 struct xen_netif_extra_info *gso;
585 gso = (struct xen_netif_extra_info *)
586 RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
588 tx->flags |= XEN_NETTXF_extra_info;
590 gso->u.gso.size = skb_shinfo(skb)->gso_size;
591 gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ?
592 XEN_NETIF_GSO_TYPE_TCPV6 :
593 XEN_NETIF_GSO_TYPE_TCPV4;
595 gso->u.gso.features = 0;
597 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
601 /* Requests for the rest of the linear area. */
602 tx = xennet_make_txreqs(queue, tx, skb, page, offset, len);
604 /* Requests for all the frags. */
605 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
606 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
607 tx = xennet_make_txreqs(queue, tx, skb,
608 skb_frag_page(frag), frag->page_offset,
609 skb_frag_size(frag));
612 /* First request has the packet length. */
613 first_tx->size = skb->len;
615 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
617 notify_remote_via_irq(queue->tx_irq);
619 u64_stats_update_begin(&tx_stats->syncp);
620 tx_stats->bytes += skb->len;
622 u64_stats_update_end(&tx_stats->syncp);
624 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
625 xennet_tx_buf_gc(queue);
627 if (!netfront_tx_slot_available(queue))
628 netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
630 spin_unlock_irqrestore(&queue->tx_lock, flags);
635 dev->stats.tx_dropped++;
636 dev_kfree_skb_any(skb);
640 static int xennet_close(struct net_device *dev)
642 struct netfront_info *np = netdev_priv(dev);
643 unsigned int num_queues = dev->real_num_tx_queues;
645 struct netfront_queue *queue;
646 netif_tx_stop_all_queues(np->netdev);
647 for (i = 0; i < num_queues; ++i) {
648 queue = &np->queues[i];
649 napi_disable(&queue->napi);
654 static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb,
657 int new = xennet_rxidx(queue->rx.req_prod_pvt);
659 BUG_ON(queue->rx_skbs[new]);
660 queue->rx_skbs[new] = skb;
661 queue->grant_rx_ref[new] = ref;
662 RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new;
663 RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref;
664 queue->rx.req_prod_pvt++;
667 static int xennet_get_extras(struct netfront_queue *queue,
668 struct xen_netif_extra_info *extras,
672 struct xen_netif_extra_info *extra;
673 struct device *dev = &queue->info->netdev->dev;
674 RING_IDX cons = queue->rx.rsp_cons;
681 if (unlikely(cons + 1 == rp)) {
683 dev_warn(dev, "Missing extra info\n");
688 extra = (struct xen_netif_extra_info *)
689 RING_GET_RESPONSE(&queue->rx, ++cons);
691 if (unlikely(!extra->type ||
692 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
694 dev_warn(dev, "Invalid extra type: %d\n",
698 memcpy(&extras[extra->type - 1], extra,
702 skb = xennet_get_rx_skb(queue, cons);
703 ref = xennet_get_rx_ref(queue, cons);
704 xennet_move_rx_slot(queue, skb, ref);
705 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
707 queue->rx.rsp_cons = cons;
711 static int xennet_get_responses(struct netfront_queue *queue,
712 struct netfront_rx_info *rinfo, RING_IDX rp,
713 struct sk_buff_head *list)
715 struct xen_netif_rx_response *rx = &rinfo->rx;
716 struct xen_netif_extra_info *extras = rinfo->extras;
717 struct device *dev = &queue->info->netdev->dev;
718 RING_IDX cons = queue->rx.rsp_cons;
719 struct sk_buff *skb = xennet_get_rx_skb(queue, cons);
720 grant_ref_t ref = xennet_get_rx_ref(queue, cons);
721 int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD);
726 if (rx->flags & XEN_NETRXF_extra_info) {
727 err = xennet_get_extras(queue, extras, rp);
728 cons = queue->rx.rsp_cons;
732 if (unlikely(rx->status < 0 ||
733 rx->offset + rx->status > PAGE_SIZE)) {
735 dev_warn(dev, "rx->offset: %u, size: %d\n",
736 rx->offset, rx->status);
737 xennet_move_rx_slot(queue, skb, ref);
743 * This definitely indicates a bug, either in this driver or in
744 * the backend driver. In future this should flag the bad
745 * situation to the system controller to reboot the backend.
747 if (ref == GRANT_INVALID_REF) {
749 dev_warn(dev, "Bad rx response id %d.\n",
755 ret = gnttab_end_foreign_access_ref(ref, 0);
758 gnttab_release_grant_reference(&queue->gref_rx_head, ref);
760 __skb_queue_tail(list, skb);
763 if (!(rx->flags & XEN_NETRXF_more_data))
766 if (cons + slots == rp) {
768 dev_warn(dev, "Need more slots\n");
773 rx = RING_GET_RESPONSE(&queue->rx, cons + slots);
774 skb = xennet_get_rx_skb(queue, cons + slots);
775 ref = xennet_get_rx_ref(queue, cons + slots);
779 if (unlikely(slots > max)) {
781 dev_warn(dev, "Too many slots\n");
786 queue->rx.rsp_cons = cons + slots;
791 static int xennet_set_skb_gso(struct sk_buff *skb,
792 struct xen_netif_extra_info *gso)
794 if (!gso->u.gso.size) {
796 pr_warn("GSO size must not be zero\n");
800 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 &&
801 gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) {
803 pr_warn("Bad GSO type %d\n", gso->u.gso.type);
807 skb_shinfo(skb)->gso_size = gso->u.gso.size;
808 skb_shinfo(skb)->gso_type =
809 (gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ?
813 /* Header must be checked, and gso_segs computed. */
814 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
815 skb_shinfo(skb)->gso_segs = 0;
820 static RING_IDX xennet_fill_frags(struct netfront_queue *queue,
822 struct sk_buff_head *list)
824 struct skb_shared_info *shinfo = skb_shinfo(skb);
825 RING_IDX cons = queue->rx.rsp_cons;
826 struct sk_buff *nskb;
828 while ((nskb = __skb_dequeue(list))) {
829 struct xen_netif_rx_response *rx =
830 RING_GET_RESPONSE(&queue->rx, ++cons);
831 skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
833 if (shinfo->nr_frags == MAX_SKB_FRAGS) {
834 unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
836 BUG_ON(pull_to <= skb_headlen(skb));
837 __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
839 BUG_ON(shinfo->nr_frags >= MAX_SKB_FRAGS);
841 skb_add_rx_frag(skb, shinfo->nr_frags, skb_frag_page(nfrag),
842 rx->offset, rx->status, PAGE_SIZE);
844 skb_shinfo(nskb)->nr_frags = 0;
851 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
853 bool recalculate_partial_csum = false;
856 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
857 * peers can fail to set NETRXF_csum_blank when sending a GSO
858 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
859 * recalculate the partial checksum.
861 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
862 struct netfront_info *np = netdev_priv(dev);
863 atomic_inc(&np->rx_gso_checksum_fixup);
864 skb->ip_summed = CHECKSUM_PARTIAL;
865 recalculate_partial_csum = true;
868 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
869 if (skb->ip_summed != CHECKSUM_PARTIAL)
872 return skb_checksum_setup(skb, recalculate_partial_csum);
875 static int handle_incoming_queue(struct netfront_queue *queue,
876 struct sk_buff_head *rxq)
878 struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats);
879 int packets_dropped = 0;
882 while ((skb = __skb_dequeue(rxq)) != NULL) {
883 int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
885 if (pull_to > skb_headlen(skb))
886 __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
888 /* Ethernet work: Delayed to here as it peeks the header. */
889 skb->protocol = eth_type_trans(skb, queue->info->netdev);
890 skb_reset_network_header(skb);
892 if (checksum_setup(queue->info->netdev, skb)) {
895 queue->info->netdev->stats.rx_errors++;
899 u64_stats_update_begin(&rx_stats->syncp);
901 rx_stats->bytes += skb->len;
902 u64_stats_update_end(&rx_stats->syncp);
905 napi_gro_receive(&queue->napi, skb);
908 return packets_dropped;
911 static int xennet_poll(struct napi_struct *napi, int budget)
913 struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi);
914 struct net_device *dev = queue->info->netdev;
916 struct netfront_rx_info rinfo;
917 struct xen_netif_rx_response *rx = &rinfo.rx;
918 struct xen_netif_extra_info *extras = rinfo.extras;
921 struct sk_buff_head rxq;
922 struct sk_buff_head errq;
923 struct sk_buff_head tmpq;
926 spin_lock(&queue->rx_lock);
928 skb_queue_head_init(&rxq);
929 skb_queue_head_init(&errq);
930 skb_queue_head_init(&tmpq);
932 rp = queue->rx.sring->rsp_prod;
933 rmb(); /* Ensure we see queued responses up to 'rp'. */
935 i = queue->rx.rsp_cons;
937 while ((i != rp) && (work_done < budget)) {
938 memcpy(rx, RING_GET_RESPONSE(&queue->rx, i), sizeof(*rx));
939 memset(extras, 0, sizeof(rinfo.extras));
941 err = xennet_get_responses(queue, &rinfo, rp, &tmpq);
945 while ((skb = __skb_dequeue(&tmpq)))
946 __skb_queue_tail(&errq, skb);
947 dev->stats.rx_errors++;
948 i = queue->rx.rsp_cons;
952 skb = __skb_dequeue(&tmpq);
954 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
955 struct xen_netif_extra_info *gso;
956 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
958 if (unlikely(xennet_set_skb_gso(skb, gso))) {
959 __skb_queue_head(&tmpq, skb);
960 queue->rx.rsp_cons += skb_queue_len(&tmpq);
965 NETFRONT_SKB_CB(skb)->pull_to = rx->status;
966 if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
967 NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
969 skb_shinfo(skb)->frags[0].page_offset = rx->offset;
970 skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
971 skb->data_len = rx->status;
972 skb->len += rx->status;
974 i = xennet_fill_frags(queue, skb, &tmpq);
976 if (rx->flags & XEN_NETRXF_csum_blank)
977 skb->ip_summed = CHECKSUM_PARTIAL;
978 else if (rx->flags & XEN_NETRXF_data_validated)
979 skb->ip_summed = CHECKSUM_UNNECESSARY;
981 __skb_queue_tail(&rxq, skb);
983 queue->rx.rsp_cons = ++i;
987 __skb_queue_purge(&errq);
989 work_done -= handle_incoming_queue(queue, &rxq);
991 xennet_alloc_rx_buffers(queue);
993 if (work_done < budget) {
998 RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do);
1000 napi_schedule(napi);
1003 spin_unlock(&queue->rx_lock);
1008 static int xennet_change_mtu(struct net_device *dev, int mtu)
1010 int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN;
1018 static struct rtnl_link_stats64 *xennet_get_stats64(struct net_device *dev,
1019 struct rtnl_link_stats64 *tot)
1021 struct netfront_info *np = netdev_priv(dev);
1024 for_each_possible_cpu(cpu) {
1025 struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu);
1026 struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu);
1027 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1031 start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
1032 tx_packets = tx_stats->packets;
1033 tx_bytes = tx_stats->bytes;
1034 } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
1037 start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
1038 rx_packets = rx_stats->packets;
1039 rx_bytes = rx_stats->bytes;
1040 } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
1042 tot->rx_packets += rx_packets;
1043 tot->tx_packets += tx_packets;
1044 tot->rx_bytes += rx_bytes;
1045 tot->tx_bytes += tx_bytes;
1048 tot->rx_errors = dev->stats.rx_errors;
1049 tot->tx_dropped = dev->stats.tx_dropped;
1054 static void xennet_release_tx_bufs(struct netfront_queue *queue)
1056 struct sk_buff *skb;
1059 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1060 /* Skip over entries which are actually freelist references */
1061 if (skb_entry_is_link(&queue->tx_skbs[i]))
1064 skb = queue->tx_skbs[i].skb;
1065 get_page(queue->grant_tx_page[i]);
1066 gnttab_end_foreign_access(queue->grant_tx_ref[i],
1068 (unsigned long)page_address(queue->grant_tx_page[i]));
1069 queue->grant_tx_page[i] = NULL;
1070 queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1071 add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, i);
1072 dev_kfree_skb_irq(skb);
1076 static void xennet_release_rx_bufs(struct netfront_queue *queue)
1080 spin_lock_bh(&queue->rx_lock);
1082 for (id = 0; id < NET_RX_RING_SIZE; id++) {
1083 struct sk_buff *skb;
1086 skb = queue->rx_skbs[id];
1090 ref = queue->grant_rx_ref[id];
1091 if (ref == GRANT_INVALID_REF)
1094 page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
1096 /* gnttab_end_foreign_access() needs a page ref until
1097 * foreign access is ended (which may be deferred).
1100 gnttab_end_foreign_access(ref, 0,
1101 (unsigned long)page_address(page));
1102 queue->grant_rx_ref[id] = GRANT_INVALID_REF;
1107 spin_unlock_bh(&queue->rx_lock);
1110 static netdev_features_t xennet_fix_features(struct net_device *dev,
1111 netdev_features_t features)
1113 struct netfront_info *np = netdev_priv(dev);
1116 if (features & NETIF_F_SG) {
1117 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg",
1122 features &= ~NETIF_F_SG;
1125 if (features & NETIF_F_IPV6_CSUM) {
1126 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1127 "feature-ipv6-csum-offload", "%d", &val) < 0)
1131 features &= ~NETIF_F_IPV6_CSUM;
1134 if (features & NETIF_F_TSO) {
1135 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1136 "feature-gso-tcpv4", "%d", &val) < 0)
1140 features &= ~NETIF_F_TSO;
1143 if (features & NETIF_F_TSO6) {
1144 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1145 "feature-gso-tcpv6", "%d", &val) < 0)
1149 features &= ~NETIF_F_TSO6;
1155 static int xennet_set_features(struct net_device *dev,
1156 netdev_features_t features)
1158 if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1159 netdev_info(dev, "Reducing MTU because no SG offload");
1160 dev->mtu = ETH_DATA_LEN;
1166 static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1168 struct netfront_queue *queue = dev_id;
1169 unsigned long flags;
1171 spin_lock_irqsave(&queue->tx_lock, flags);
1172 xennet_tx_buf_gc(queue);
1173 spin_unlock_irqrestore(&queue->tx_lock, flags);
1178 static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1180 struct netfront_queue *queue = dev_id;
1181 struct net_device *dev = queue->info->netdev;
1183 if (likely(netif_carrier_ok(dev) &&
1184 RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)))
1185 napi_schedule(&queue->napi);
1190 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1192 xennet_tx_interrupt(irq, dev_id);
1193 xennet_rx_interrupt(irq, dev_id);
1197 #ifdef CONFIG_NET_POLL_CONTROLLER
1198 static void xennet_poll_controller(struct net_device *dev)
1200 /* Poll each queue */
1201 struct netfront_info *info = netdev_priv(dev);
1202 unsigned int num_queues = dev->real_num_tx_queues;
1204 for (i = 0; i < num_queues; ++i)
1205 xennet_interrupt(0, &info->queues[i]);
1209 static const struct net_device_ops xennet_netdev_ops = {
1210 .ndo_open = xennet_open,
1211 .ndo_stop = xennet_close,
1212 .ndo_start_xmit = xennet_start_xmit,
1213 .ndo_change_mtu = xennet_change_mtu,
1214 .ndo_get_stats64 = xennet_get_stats64,
1215 .ndo_set_mac_address = eth_mac_addr,
1216 .ndo_validate_addr = eth_validate_addr,
1217 .ndo_fix_features = xennet_fix_features,
1218 .ndo_set_features = xennet_set_features,
1219 .ndo_select_queue = xennet_select_queue,
1220 #ifdef CONFIG_NET_POLL_CONTROLLER
1221 .ndo_poll_controller = xennet_poll_controller,
1225 static void xennet_free_netdev(struct net_device *netdev)
1227 struct netfront_info *np = netdev_priv(netdev);
1229 free_percpu(np->rx_stats);
1230 free_percpu(np->tx_stats);
1231 free_netdev(netdev);
1234 static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1237 struct net_device *netdev;
1238 struct netfront_info *np;
1240 netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues);
1242 return ERR_PTR(-ENOMEM);
1244 np = netdev_priv(netdev);
1250 np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1251 if (np->rx_stats == NULL)
1253 np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1254 if (np->tx_stats == NULL)
1257 netdev->netdev_ops = &xennet_netdev_ops;
1259 netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1261 netdev->hw_features = NETIF_F_SG |
1263 NETIF_F_TSO | NETIF_F_TSO6;
1266 * Assume that all hw features are available for now. This set
1267 * will be adjusted by the call to netdev_update_features() in
1268 * xennet_connect() which is the earliest point where we can
1269 * negotiate with the backend regarding supported features.
1271 netdev->features |= netdev->hw_features;
1273 netdev->ethtool_ops = &xennet_ethtool_ops;
1274 SET_NETDEV_DEV(netdev, &dev->dev);
1276 np->netdev = netdev;
1278 netif_carrier_off(netdev);
1283 xennet_free_netdev(netdev);
1284 return ERR_PTR(err);
1288 * Entry point to this code when a new device is created. Allocate the basic
1289 * structures and the ring buffers for communication with the backend, and
1290 * inform the backend of the appropriate details for those.
1292 static int netfront_probe(struct xenbus_device *dev,
1293 const struct xenbus_device_id *id)
1296 struct net_device *netdev;
1297 struct netfront_info *info;
1299 netdev = xennet_create_dev(dev);
1300 if (IS_ERR(netdev)) {
1301 err = PTR_ERR(netdev);
1302 xenbus_dev_fatal(dev, err, "creating netdev");
1306 info = netdev_priv(netdev);
1307 dev_set_drvdata(&dev->dev, info);
1309 info->netdev->sysfs_groups[0] = &xennet_dev_group;
1311 err = register_netdev(info->netdev);
1313 pr_warn("%s: register_netdev err=%d\n", __func__, err);
1320 xennet_free_netdev(netdev);
1321 dev_set_drvdata(&dev->dev, NULL);
1325 static void xennet_end_access(int ref, void *page)
1327 /* This frees the page as a side-effect */
1328 if (ref != GRANT_INVALID_REF)
1329 gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1332 static void xennet_disconnect_backend(struct netfront_info *info)
1335 unsigned int num_queues = info->netdev->real_num_tx_queues;
1337 netif_carrier_off(info->netdev);
1339 for (i = 0; i < num_queues && info->queues; ++i) {
1340 struct netfront_queue *queue = &info->queues[i];
1342 if (queue->tx_irq && (queue->tx_irq == queue->rx_irq))
1343 unbind_from_irqhandler(queue->tx_irq, queue);
1344 if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) {
1345 unbind_from_irqhandler(queue->tx_irq, queue);
1346 unbind_from_irqhandler(queue->rx_irq, queue);
1348 queue->tx_evtchn = queue->rx_evtchn = 0;
1349 queue->tx_irq = queue->rx_irq = 0;
1351 if (netif_running(info->netdev))
1352 napi_synchronize(&queue->napi);
1354 xennet_release_tx_bufs(queue);
1355 xennet_release_rx_bufs(queue);
1356 gnttab_free_grant_references(queue->gref_tx_head);
1357 gnttab_free_grant_references(queue->gref_rx_head);
1359 /* End access and free the pages */
1360 xennet_end_access(queue->tx_ring_ref, queue->tx.sring);
1361 xennet_end_access(queue->rx_ring_ref, queue->rx.sring);
1363 queue->tx_ring_ref = GRANT_INVALID_REF;
1364 queue->rx_ring_ref = GRANT_INVALID_REF;
1365 queue->tx.sring = NULL;
1366 queue->rx.sring = NULL;
1371 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1372 * driver restart. We tear down our netif structure and recreate it, but
1373 * leave the device-layer structures intact so that this is transparent to the
1374 * rest of the kernel.
1376 static int netfront_resume(struct xenbus_device *dev)
1378 struct netfront_info *info = dev_get_drvdata(&dev->dev);
1380 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1382 xennet_disconnect_backend(info);
1386 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1388 char *s, *e, *macstr;
1391 macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1393 return PTR_ERR(macstr);
1395 for (i = 0; i < ETH_ALEN; i++) {
1396 mac[i] = simple_strtoul(s, &e, 16);
1397 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1408 static int setup_netfront_single(struct netfront_queue *queue)
1412 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1416 err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1418 0, queue->info->netdev->name, queue);
1421 queue->rx_evtchn = queue->tx_evtchn;
1422 queue->rx_irq = queue->tx_irq = err;
1427 xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1428 queue->tx_evtchn = 0;
1433 static int setup_netfront_split(struct netfront_queue *queue)
1437 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1440 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn);
1442 goto alloc_rx_evtchn_fail;
1444 snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
1445 "%s-tx", queue->name);
1446 err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1447 xennet_tx_interrupt,
1448 0, queue->tx_irq_name, queue);
1451 queue->tx_irq = err;
1453 snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
1454 "%s-rx", queue->name);
1455 err = bind_evtchn_to_irqhandler(queue->rx_evtchn,
1456 xennet_rx_interrupt,
1457 0, queue->rx_irq_name, queue);
1460 queue->rx_irq = err;
1465 unbind_from_irqhandler(queue->tx_irq, queue);
1468 xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn);
1469 queue->rx_evtchn = 0;
1470 alloc_rx_evtchn_fail:
1471 xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1472 queue->tx_evtchn = 0;
1477 static int setup_netfront(struct xenbus_device *dev,
1478 struct netfront_queue *queue, unsigned int feature_split_evtchn)
1480 struct xen_netif_tx_sring *txs;
1481 struct xen_netif_rx_sring *rxs;
1485 queue->tx_ring_ref = GRANT_INVALID_REF;
1486 queue->rx_ring_ref = GRANT_INVALID_REF;
1487 queue->rx.sring = NULL;
1488 queue->tx.sring = NULL;
1490 txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1493 xenbus_dev_fatal(dev, err, "allocating tx ring page");
1496 SHARED_RING_INIT(txs);
1497 FRONT_RING_INIT(&queue->tx, txs, PAGE_SIZE);
1499 err = xenbus_grant_ring(dev, txs, 1, &gref);
1501 goto grant_tx_ring_fail;
1502 queue->tx_ring_ref = gref;
1504 rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1507 xenbus_dev_fatal(dev, err, "allocating rx ring page");
1508 goto alloc_rx_ring_fail;
1510 SHARED_RING_INIT(rxs);
1511 FRONT_RING_INIT(&queue->rx, rxs, PAGE_SIZE);
1513 err = xenbus_grant_ring(dev, rxs, 1, &gref);
1515 goto grant_rx_ring_fail;
1516 queue->rx_ring_ref = gref;
1518 if (feature_split_evtchn)
1519 err = setup_netfront_split(queue);
1520 /* setup single event channel if
1521 * a) feature-split-event-channels == 0
1522 * b) feature-split-event-channels == 1 but failed to setup
1524 if (!feature_split_evtchn || (feature_split_evtchn && err))
1525 err = setup_netfront_single(queue);
1528 goto alloc_evtchn_fail;
1532 /* If we fail to setup netfront, it is safe to just revoke access to
1533 * granted pages because backend is not accessing it at this point.
1536 gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0);
1538 free_page((unsigned long)rxs);
1540 gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0);
1542 free_page((unsigned long)txs);
1547 /* Queue-specific initialisation
1548 * This used to be done in xennet_create_dev() but must now
1551 static int xennet_init_queue(struct netfront_queue *queue)
1556 spin_lock_init(&queue->tx_lock);
1557 spin_lock_init(&queue->rx_lock);
1559 setup_timer(&queue->rx_refill_timer, rx_refill_timeout,
1560 (unsigned long)queue);
1562 snprintf(queue->name, sizeof(queue->name), "%s-q%u",
1563 queue->info->netdev->name, queue->id);
1565 /* Initialise tx_skbs as a free chain containing every entry. */
1566 queue->tx_skb_freelist = 0;
1567 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1568 skb_entry_set_link(&queue->tx_skbs[i], i+1);
1569 queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1570 queue->grant_tx_page[i] = NULL;
1573 /* Clear out rx_skbs */
1574 for (i = 0; i < NET_RX_RING_SIZE; i++) {
1575 queue->rx_skbs[i] = NULL;
1576 queue->grant_rx_ref[i] = GRANT_INVALID_REF;
1579 /* A grant for every tx ring slot */
1580 if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
1581 &queue->gref_tx_head) < 0) {
1582 pr_alert("can't alloc tx grant refs\n");
1587 /* A grant for every rx ring slot */
1588 if (gnttab_alloc_grant_references(NET_RX_RING_SIZE,
1589 &queue->gref_rx_head) < 0) {
1590 pr_alert("can't alloc rx grant refs\n");
1598 gnttab_free_grant_references(queue->gref_tx_head);
1603 static int write_queue_xenstore_keys(struct netfront_queue *queue,
1604 struct xenbus_transaction *xbt, int write_hierarchical)
1606 /* Write the queue-specific keys into XenStore in the traditional
1607 * way for a single queue, or in a queue subkeys for multiple
1610 struct xenbus_device *dev = queue->info->xbdev;
1612 const char *message;
1616 /* Choose the correct place to write the keys */
1617 if (write_hierarchical) {
1618 pathsize = strlen(dev->nodename) + 10;
1619 path = kzalloc(pathsize, GFP_KERNEL);
1622 message = "out of memory while writing ring references";
1625 snprintf(path, pathsize, "%s/queue-%u",
1626 dev->nodename, queue->id);
1628 path = (char *)dev->nodename;
1631 /* Write ring references */
1632 err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u",
1633 queue->tx_ring_ref);
1635 message = "writing tx-ring-ref";
1639 err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u",
1640 queue->rx_ring_ref);
1642 message = "writing rx-ring-ref";
1646 /* Write event channels; taking into account both shared
1647 * and split event channel scenarios.
1649 if (queue->tx_evtchn == queue->rx_evtchn) {
1650 /* Shared event channel */
1651 err = xenbus_printf(*xbt, path,
1652 "event-channel", "%u", queue->tx_evtchn);
1654 message = "writing event-channel";
1658 /* Split event channels */
1659 err = xenbus_printf(*xbt, path,
1660 "event-channel-tx", "%u", queue->tx_evtchn);
1662 message = "writing event-channel-tx";
1666 err = xenbus_printf(*xbt, path,
1667 "event-channel-rx", "%u", queue->rx_evtchn);
1669 message = "writing event-channel-rx";
1674 if (write_hierarchical)
1679 if (write_hierarchical)
1681 xenbus_dev_fatal(dev, err, "%s", message);
1685 static void xennet_destroy_queues(struct netfront_info *info)
1691 for (i = 0; i < info->netdev->real_num_tx_queues; i++) {
1692 struct netfront_queue *queue = &info->queues[i];
1694 if (netif_running(info->netdev))
1695 napi_disable(&queue->napi);
1696 del_timer_sync(&queue->rx_refill_timer);
1697 netif_napi_del(&queue->napi);
1702 kfree(info->queues);
1703 info->queues = NULL;
1706 static int xennet_create_queues(struct netfront_info *info,
1707 unsigned int num_queues)
1712 info->queues = kcalloc(num_queues, sizeof(struct netfront_queue),
1719 for (i = 0; i < num_queues; i++) {
1720 struct netfront_queue *queue = &info->queues[i];
1725 ret = xennet_init_queue(queue);
1727 dev_warn(&info->netdev->dev,
1728 "only created %d queues\n", i);
1733 netif_napi_add(queue->info->netdev, &queue->napi,
1735 if (netif_running(info->netdev))
1736 napi_enable(&queue->napi);
1739 netif_set_real_num_tx_queues(info->netdev, num_queues);
1743 if (num_queues == 0) {
1744 dev_err(&info->netdev->dev, "no queues\n");
1750 /* Common code used when first setting up, and when resuming. */
1751 static int talk_to_netback(struct xenbus_device *dev,
1752 struct netfront_info *info)
1754 const char *message;
1755 struct xenbus_transaction xbt;
1757 unsigned int feature_split_evtchn;
1759 unsigned int max_queues = 0;
1760 struct netfront_queue *queue = NULL;
1761 unsigned int num_queues = 1;
1763 info->netdev->irq = 0;
1765 /* Check if backend supports multiple queues */
1766 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1767 "multi-queue-max-queues", "%u", &max_queues);
1770 num_queues = min(max_queues, xennet_max_queues);
1772 /* Check feature-split-event-channels */
1773 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1774 "feature-split-event-channels", "%u",
1775 &feature_split_evtchn);
1777 feature_split_evtchn = 0;
1779 /* Read mac addr. */
1780 err = xen_net_read_mac(dev, info->netdev->dev_addr);
1782 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1787 xennet_destroy_queues(info);
1789 err = xennet_create_queues(info, num_queues);
1793 /* Create shared ring, alloc event channel -- for each queue */
1794 for (i = 0; i < num_queues; ++i) {
1795 queue = &info->queues[i];
1796 err = setup_netfront(dev, queue, feature_split_evtchn);
1798 /* setup_netfront() will tidy up the current
1799 * queue on error, but we need to clean up
1800 * those already allocated.
1804 netif_set_real_num_tx_queues(info->netdev, i);
1814 err = xenbus_transaction_start(&xbt);
1816 xenbus_dev_fatal(dev, err, "starting transaction");
1820 if (num_queues == 1) {
1821 err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */
1823 goto abort_transaction_no_dev_fatal;
1825 /* Write the number of queues */
1826 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues",
1829 message = "writing multi-queue-num-queues";
1830 goto abort_transaction_no_dev_fatal;
1833 /* Write the keys for each queue */
1834 for (i = 0; i < num_queues; ++i) {
1835 queue = &info->queues[i];
1836 err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */
1838 goto abort_transaction_no_dev_fatal;
1842 /* The remaining keys are not queue-specific */
1843 err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1846 message = "writing request-rx-copy";
1847 goto abort_transaction;
1850 err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1852 message = "writing feature-rx-notify";
1853 goto abort_transaction;
1856 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1858 message = "writing feature-sg";
1859 goto abort_transaction;
1862 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1864 message = "writing feature-gso-tcpv4";
1865 goto abort_transaction;
1868 err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1");
1870 message = "writing feature-gso-tcpv6";
1871 goto abort_transaction;
1874 err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload",
1877 message = "writing feature-ipv6-csum-offload";
1878 goto abort_transaction;
1881 err = xenbus_transaction_end(xbt, 0);
1885 xenbus_dev_fatal(dev, err, "completing transaction");
1892 xenbus_dev_fatal(dev, err, "%s", message);
1893 abort_transaction_no_dev_fatal:
1894 xenbus_transaction_end(xbt, 1);
1896 xennet_disconnect_backend(info);
1897 kfree(info->queues);
1898 info->queues = NULL;
1903 static int xennet_connect(struct net_device *dev)
1905 struct netfront_info *np = netdev_priv(dev);
1906 unsigned int num_queues = 0;
1908 unsigned int feature_rx_copy;
1910 struct netfront_queue *queue = NULL;
1912 err = xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1913 "feature-rx-copy", "%u", &feature_rx_copy);
1915 feature_rx_copy = 0;
1917 if (!feature_rx_copy) {
1919 "backend does not support copying receive path\n");
1923 err = talk_to_netback(np->xbdev, np);
1927 /* talk_to_netback() sets the correct number of queues */
1928 num_queues = dev->real_num_tx_queues;
1931 netdev_update_features(dev);
1935 * All public and private state should now be sane. Get
1936 * ready to start sending and receiving packets and give the driver
1937 * domain a kick because we've probably just requeued some
1940 netif_carrier_on(np->netdev);
1941 for (j = 0; j < num_queues; ++j) {
1942 queue = &np->queues[j];
1944 notify_remote_via_irq(queue->tx_irq);
1945 if (queue->tx_irq != queue->rx_irq)
1946 notify_remote_via_irq(queue->rx_irq);
1948 spin_lock_irq(&queue->tx_lock);
1949 xennet_tx_buf_gc(queue);
1950 spin_unlock_irq(&queue->tx_lock);
1952 spin_lock_bh(&queue->rx_lock);
1953 xennet_alloc_rx_buffers(queue);
1954 spin_unlock_bh(&queue->rx_lock);
1961 * Callback received when the backend's state changes.
1963 static void netback_changed(struct xenbus_device *dev,
1964 enum xenbus_state backend_state)
1966 struct netfront_info *np = dev_get_drvdata(&dev->dev);
1967 struct net_device *netdev = np->netdev;
1969 dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
1971 switch (backend_state) {
1972 case XenbusStateInitialising:
1973 case XenbusStateInitialised:
1974 case XenbusStateReconfiguring:
1975 case XenbusStateReconfigured:
1976 case XenbusStateUnknown:
1979 case XenbusStateInitWait:
1980 if (dev->state != XenbusStateInitialising)
1982 if (xennet_connect(netdev) != 0)
1984 xenbus_switch_state(dev, XenbusStateConnected);
1987 case XenbusStateConnected:
1988 netdev_notify_peers(netdev);
1991 case XenbusStateClosed:
1992 if (dev->state == XenbusStateClosed)
1994 /* Missed the backend's CLOSING state -- fallthrough */
1995 case XenbusStateClosing:
1996 xenbus_frontend_closed(dev);
2001 static const struct xennet_stat {
2002 char name[ETH_GSTRING_LEN];
2004 } xennet_stats[] = {
2006 "rx_gso_checksum_fixup",
2007 offsetof(struct netfront_info, rx_gso_checksum_fixup)
2011 static int xennet_get_sset_count(struct net_device *dev, int string_set)
2013 switch (string_set) {
2015 return ARRAY_SIZE(xennet_stats);
2021 static void xennet_get_ethtool_stats(struct net_device *dev,
2022 struct ethtool_stats *stats, u64 * data)
2024 void *np = netdev_priv(dev);
2027 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2028 data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset));
2031 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
2035 switch (stringset) {
2037 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2038 memcpy(data + i * ETH_GSTRING_LEN,
2039 xennet_stats[i].name, ETH_GSTRING_LEN);
2044 static const struct ethtool_ops xennet_ethtool_ops =
2046 .get_link = ethtool_op_get_link,
2048 .get_sset_count = xennet_get_sset_count,
2049 .get_ethtool_stats = xennet_get_ethtool_stats,
2050 .get_strings = xennet_get_strings,
2054 static ssize_t show_rxbuf(struct device *dev,
2055 struct device_attribute *attr, char *buf)
2057 return sprintf(buf, "%lu\n", NET_RX_RING_SIZE);
2060 static ssize_t store_rxbuf(struct device *dev,
2061 struct device_attribute *attr,
2062 const char *buf, size_t len)
2065 unsigned long target;
2067 if (!capable(CAP_NET_ADMIN))
2070 target = simple_strtoul(buf, &endp, 0);
2074 /* rxbuf_min and rxbuf_max are no longer configurable. */
2079 static DEVICE_ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf);
2080 static DEVICE_ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf);
2081 static DEVICE_ATTR(rxbuf_cur, S_IRUGO, show_rxbuf, NULL);
2083 static struct attribute *xennet_dev_attrs[] = {
2084 &dev_attr_rxbuf_min.attr,
2085 &dev_attr_rxbuf_max.attr,
2086 &dev_attr_rxbuf_cur.attr,
2090 static const struct attribute_group xennet_dev_group = {
2091 .attrs = xennet_dev_attrs
2093 #endif /* CONFIG_SYSFS */
2095 static int xennet_remove(struct xenbus_device *dev)
2097 struct netfront_info *info = dev_get_drvdata(&dev->dev);
2099 dev_dbg(&dev->dev, "%s\n", dev->nodename);
2101 xennet_disconnect_backend(info);
2103 unregister_netdev(info->netdev);
2106 xennet_destroy_queues(info);
2107 xennet_free_netdev(info->netdev);
2112 static const struct xenbus_device_id netfront_ids[] = {
2117 static struct xenbus_driver netfront_driver = {
2118 .ids = netfront_ids,
2119 .probe = netfront_probe,
2120 .remove = xennet_remove,
2121 .resume = netfront_resume,
2122 .otherend_changed = netback_changed,
2125 static int __init netif_init(void)
2130 if (!xen_has_pv_nic_devices())
2133 pr_info("Initialising Xen virtual ethernet driver\n");
2135 /* Allow as many queues as there are CPUs, by default */
2136 xennet_max_queues = num_online_cpus();
2138 return xenbus_register_frontend(&netfront_driver);
2140 module_init(netif_init);
2143 static void __exit netif_exit(void)
2145 xenbus_unregister_driver(&netfront_driver);
2147 module_exit(netif_exit);
2149 MODULE_DESCRIPTION("Xen virtual network device frontend");
2150 MODULE_LICENSE("GPL");
2151 MODULE_ALIAS("xen:vif");
2152 MODULE_ALIAS("xennet");