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>
48 #include <asm/xen/page.h>
50 #include <xen/xenbus.h>
51 #include <xen/events.h>
53 #include <xen/platform_pci.h>
54 #include <xen/grant_table.h>
56 #include <xen/interface/io/netif.h>
57 #include <xen/interface/memory.h>
58 #include <xen/interface/grant_table.h>
60 /* Module parameters */
61 static unsigned int xennet_max_queues;
62 module_param_named(max_queues, xennet_max_queues, uint, 0644);
63 MODULE_PARM_DESC(max_queues,
64 "Maximum number of queues per virtual interface");
66 static const struct ethtool_ops xennet_ethtool_ops;
72 #define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb))
74 #define RX_COPY_THRESHOLD 256
76 #define GRANT_INVALID_REF 0
78 #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, PAGE_SIZE)
79 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, PAGE_SIZE)
81 /* Minimum number of Rx slots (includes slot for GSO metadata). */
82 #define NET_RX_SLOTS_MIN (XEN_NETIF_NR_SLOTS_MIN + 1)
84 /* Queue name is interface name with "-qNNN" appended */
85 #define QUEUE_NAME_SIZE (IFNAMSIZ + 6)
87 /* IRQ name is queue name with "-tx" or "-rx" appended */
88 #define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3)
90 struct netfront_stats {
95 struct u64_stats_sync syncp;
100 struct netfront_queue {
101 unsigned int id; /* Queue ID, 0-based */
102 char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */
103 struct netfront_info *info;
105 struct napi_struct napi;
107 /* Split event channels support, tx_* == rx_* when using
108 * single event channel.
110 unsigned int tx_evtchn, rx_evtchn;
111 unsigned int tx_irq, rx_irq;
112 /* Only used when split event channels support is enabled */
113 char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */
114 char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */
117 struct xen_netif_tx_front_ring tx;
121 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
122 * are linked from tx_skb_freelist through skb_entry.link.
124 * NB. Freelist index entries are always going to be less than
125 * PAGE_OFFSET, whereas pointers to skbs will always be equal or
126 * greater than PAGE_OFFSET: we use this property to distinguish
132 } tx_skbs[NET_TX_RING_SIZE];
133 grant_ref_t gref_tx_head;
134 grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
135 struct page *grant_tx_page[NET_TX_RING_SIZE];
136 unsigned tx_skb_freelist;
138 spinlock_t rx_lock ____cacheline_aligned_in_smp;
139 struct xen_netif_rx_front_ring rx;
142 struct timer_list rx_refill_timer;
144 struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
145 grant_ref_t gref_rx_head;
146 grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
148 unsigned long rx_pfn_array[NET_RX_RING_SIZE];
149 struct multicall_entry rx_mcl[NET_RX_RING_SIZE+1];
150 struct mmu_update rx_mmu[NET_RX_RING_SIZE];
153 struct netfront_info {
154 struct list_head list;
155 struct net_device *netdev;
157 struct xenbus_device *xbdev;
159 /* Multi-queue support */
160 struct netfront_queue *queues;
163 struct netfront_stats __percpu *stats;
165 atomic_t rx_gso_checksum_fixup;
168 struct netfront_rx_info {
169 struct xen_netif_rx_response rx;
170 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
173 static void skb_entry_set_link(union skb_entry *list, unsigned short id)
178 static int skb_entry_is_link(const union skb_entry *list)
180 BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link));
181 return (unsigned long)list->skb < PAGE_OFFSET;
185 * Access macros for acquiring freeing slots in tx_skbs[].
188 static void add_id_to_freelist(unsigned *head, union skb_entry *list,
191 skb_entry_set_link(&list[id], *head);
195 static unsigned short get_id_from_freelist(unsigned *head,
196 union skb_entry *list)
198 unsigned int id = *head;
199 *head = list[id].link;
203 static int xennet_rxidx(RING_IDX idx)
205 return idx & (NET_RX_RING_SIZE - 1);
208 static struct sk_buff *xennet_get_rx_skb(struct netfront_queue *queue,
211 int i = xennet_rxidx(ri);
212 struct sk_buff *skb = queue->rx_skbs[i];
213 queue->rx_skbs[i] = NULL;
217 static grant_ref_t xennet_get_rx_ref(struct netfront_queue *queue,
220 int i = xennet_rxidx(ri);
221 grant_ref_t ref = queue->grant_rx_ref[i];
222 queue->grant_rx_ref[i] = GRANT_INVALID_REF;
227 static int xennet_sysfs_addif(struct net_device *netdev);
228 static void xennet_sysfs_delif(struct net_device *netdev);
229 #else /* !CONFIG_SYSFS */
230 #define xennet_sysfs_addif(dev) (0)
231 #define xennet_sysfs_delif(dev) do { } while (0)
234 static bool xennet_can_sg(struct net_device *dev)
236 return dev->features & NETIF_F_SG;
240 static void rx_refill_timeout(unsigned long data)
242 struct netfront_queue *queue = (struct netfront_queue *)data;
243 napi_schedule(&queue->napi);
246 static int netfront_tx_slot_available(struct netfront_queue *queue)
248 return (queue->tx.req_prod_pvt - queue->tx.rsp_cons) <
249 (NET_TX_RING_SIZE - MAX_SKB_FRAGS - 2);
252 static void xennet_maybe_wake_tx(struct netfront_queue *queue)
254 struct net_device *dev = queue->info->netdev;
255 struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, queue->id);
257 if (unlikely(netif_tx_queue_stopped(dev_queue)) &&
258 netfront_tx_slot_available(queue) &&
259 likely(netif_running(dev)))
260 netif_tx_wake_queue(netdev_get_tx_queue(dev, queue->id));
264 static struct sk_buff *xennet_alloc_one_rx_buffer(struct netfront_queue *queue)
269 skb = __netdev_alloc_skb(queue->info->netdev,
270 RX_COPY_THRESHOLD + NET_IP_ALIGN,
271 GFP_ATOMIC | __GFP_NOWARN);
275 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
280 skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE);
282 /* Align ip header to a 16 bytes boundary */
283 skb_reserve(skb, NET_IP_ALIGN);
284 skb->dev = queue->info->netdev;
290 static void xennet_alloc_rx_buffers(struct netfront_queue *queue)
292 RING_IDX req_prod = queue->rx.req_prod_pvt;
295 if (unlikely(!netif_carrier_ok(queue->info->netdev)))
298 for (req_prod = queue->rx.req_prod_pvt;
299 req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE;
305 struct xen_netif_rx_request *req;
307 skb = xennet_alloc_one_rx_buffer(queue);
311 id = xennet_rxidx(req_prod);
313 BUG_ON(queue->rx_skbs[id]);
314 queue->rx_skbs[id] = skb;
316 ref = gnttab_claim_grant_reference(&queue->gref_rx_head);
317 BUG_ON((signed short)ref < 0);
318 queue->grant_rx_ref[id] = ref;
320 pfn = page_to_pfn(skb_frag_page(&skb_shinfo(skb)->frags[0]));
322 req = RING_GET_REQUEST(&queue->rx, req_prod);
323 gnttab_grant_foreign_access_ref(ref,
324 queue->info->xbdev->otherend_id,
332 queue->rx.req_prod_pvt = req_prod;
334 /* Not enough requests? Try again later. */
335 if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN) {
336 mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10));
340 wmb(); /* barrier so backend seens requests */
342 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify);
344 notify_remote_via_irq(queue->rx_irq);
347 static int xennet_open(struct net_device *dev)
349 struct netfront_info *np = netdev_priv(dev);
350 unsigned int num_queues = dev->real_num_tx_queues;
352 struct netfront_queue *queue = NULL;
354 for (i = 0; i < num_queues; ++i) {
355 queue = &np->queues[i];
356 napi_enable(&queue->napi);
358 spin_lock_bh(&queue->rx_lock);
359 if (netif_carrier_ok(dev)) {
360 xennet_alloc_rx_buffers(queue);
361 queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1;
362 if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))
363 napi_schedule(&queue->napi);
365 spin_unlock_bh(&queue->rx_lock);
368 netif_tx_start_all_queues(dev);
373 static void xennet_tx_buf_gc(struct netfront_queue *queue)
379 BUG_ON(!netif_carrier_ok(queue->info->netdev));
382 prod = queue->tx.sring->rsp_prod;
383 rmb(); /* Ensure we see responses up to 'rp'. */
385 for (cons = queue->tx.rsp_cons; cons != prod; cons++) {
386 struct xen_netif_tx_response *txrsp;
388 txrsp = RING_GET_RESPONSE(&queue->tx, cons);
389 if (txrsp->status == XEN_NETIF_RSP_NULL)
393 skb = queue->tx_skbs[id].skb;
394 if (unlikely(gnttab_query_foreign_access(
395 queue->grant_tx_ref[id]) != 0)) {
396 pr_alert("%s: warning -- grant still in use by backend domain\n",
400 gnttab_end_foreign_access_ref(
401 queue->grant_tx_ref[id], GNTMAP_readonly);
402 gnttab_release_grant_reference(
403 &queue->gref_tx_head, queue->grant_tx_ref[id]);
404 queue->grant_tx_ref[id] = GRANT_INVALID_REF;
405 queue->grant_tx_page[id] = NULL;
406 add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, id);
407 dev_kfree_skb_irq(skb);
410 queue->tx.rsp_cons = prod;
413 * Set a new event, then check for race with update of tx_cons.
414 * Note that it is essential to schedule a callback, no matter
415 * how few buffers are pending. Even if there is space in the
416 * transmit ring, higher layers may be blocked because too much
417 * data is outstanding: in such cases notification from Xen is
418 * likely to be the only kick that we'll get.
420 queue->tx.sring->rsp_event =
421 prod + ((queue->tx.sring->req_prod - prod) >> 1) + 1;
422 mb(); /* update shared area */
423 } while ((cons == prod) && (prod != queue->tx.sring->rsp_prod));
425 xennet_maybe_wake_tx(queue);
428 static void xennet_make_frags(struct sk_buff *skb, struct netfront_queue *queue,
429 struct xen_netif_tx_request *tx)
431 char *data = skb->data;
433 RING_IDX prod = queue->tx.req_prod_pvt;
434 int frags = skb_shinfo(skb)->nr_frags;
435 unsigned int offset = offset_in_page(data);
436 unsigned int len = skb_headlen(skb);
441 /* While the header overlaps a page boundary (including being
442 larger than a page), split it it into page-sized chunks. */
443 while (len > PAGE_SIZE - offset) {
444 tx->size = PAGE_SIZE - offset;
445 tx->flags |= XEN_NETTXF_more_data;
450 id = get_id_from_freelist(&queue->tx_skb_freelist, queue->tx_skbs);
451 queue->tx_skbs[id].skb = skb_get(skb);
452 tx = RING_GET_REQUEST(&queue->tx, prod++);
454 ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
455 BUG_ON((signed short)ref < 0);
457 mfn = virt_to_mfn(data);
458 gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id,
459 mfn, GNTMAP_readonly);
461 queue->grant_tx_page[id] = virt_to_page(data);
462 tx->gref = queue->grant_tx_ref[id] = ref;
468 /* Grant backend access to each skb fragment page. */
469 for (i = 0; i < frags; i++) {
470 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
471 struct page *page = skb_frag_page(frag);
473 len = skb_frag_size(frag);
474 offset = frag->page_offset;
476 /* Skip unused frames from start of page */
477 page += offset >> PAGE_SHIFT;
478 offset &= ~PAGE_MASK;
483 bytes = PAGE_SIZE - offset;
487 tx->flags |= XEN_NETTXF_more_data;
489 id = get_id_from_freelist(&queue->tx_skb_freelist,
491 queue->tx_skbs[id].skb = skb_get(skb);
492 tx = RING_GET_REQUEST(&queue->tx, prod++);
494 ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
495 BUG_ON((signed short)ref < 0);
497 mfn = pfn_to_mfn(page_to_pfn(page));
498 gnttab_grant_foreign_access_ref(ref,
499 queue->info->xbdev->otherend_id,
500 mfn, GNTMAP_readonly);
502 queue->grant_tx_page[id] = page;
503 tx->gref = queue->grant_tx_ref[id] = ref;
512 if (offset == PAGE_SIZE && len) {
513 BUG_ON(!PageCompound(page));
520 queue->tx.req_prod_pvt = prod;
524 * Count how many ring slots are required to send the frags of this
525 * skb. Each frag might be a compound page.
527 static int xennet_count_skb_frag_slots(struct sk_buff *skb)
529 int i, frags = skb_shinfo(skb)->nr_frags;
532 for (i = 0; i < frags; i++) {
533 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
534 unsigned long size = skb_frag_size(frag);
535 unsigned long offset = frag->page_offset;
537 /* Skip unused frames from start of page */
538 offset &= ~PAGE_MASK;
540 pages += PFN_UP(offset + size);
546 static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb,
547 void *accel_priv, select_queue_fallback_t fallback)
549 unsigned int num_queues = dev->real_num_tx_queues;
553 /* First, check if there is only one queue */
554 if (num_queues == 1) {
557 hash = skb_get_hash(skb);
558 queue_idx = hash % num_queues;
564 static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
567 struct netfront_info *np = netdev_priv(dev);
568 struct netfront_stats *stats = this_cpu_ptr(np->stats);
569 struct xen_netif_tx_request *tx;
570 char *data = skb->data;
576 unsigned int offset = offset_in_page(data);
577 unsigned int len = skb_headlen(skb);
579 struct netfront_queue *queue = NULL;
580 unsigned int num_queues = dev->real_num_tx_queues;
583 /* Drop the packet if no queues are set up */
586 /* Determine which queue to transmit this SKB on */
587 queue_index = skb_get_queue_mapping(skb);
588 queue = &np->queues[queue_index];
590 /* If skb->len is too big for wire format, drop skb and alert
591 * user about misconfiguration.
593 if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
594 net_alert_ratelimited(
595 "xennet: skb->len = %u, too big for wire format\n",
600 slots = DIV_ROUND_UP(offset + len, PAGE_SIZE) +
601 xennet_count_skb_frag_slots(skb);
602 if (unlikely(slots > MAX_SKB_FRAGS + 1)) {
603 net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n",
605 if (skb_linearize(skb))
608 offset = offset_in_page(data);
609 len = skb_headlen(skb);
612 spin_lock_irqsave(&queue->tx_lock, flags);
614 if (unlikely(!netif_carrier_ok(dev) ||
615 (slots > 1 && !xennet_can_sg(dev)) ||
616 netif_needs_gso(dev, skb, netif_skb_features(skb)))) {
617 spin_unlock_irqrestore(&queue->tx_lock, flags);
621 i = queue->tx.req_prod_pvt;
623 id = get_id_from_freelist(&queue->tx_skb_freelist, queue->tx_skbs);
624 queue->tx_skbs[id].skb = skb;
626 tx = RING_GET_REQUEST(&queue->tx, i);
629 ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
630 BUG_ON((signed short)ref < 0);
631 mfn = virt_to_mfn(data);
632 gnttab_grant_foreign_access_ref(
633 ref, queue->info->xbdev->otherend_id, mfn, GNTMAP_readonly);
634 queue->grant_tx_page[id] = virt_to_page(data);
635 tx->gref = queue->grant_tx_ref[id] = ref;
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 if (skb_shinfo(skb)->gso_size) {
648 struct xen_netif_extra_info *gso;
650 gso = (struct xen_netif_extra_info *)
651 RING_GET_REQUEST(&queue->tx, ++i);
653 tx->flags |= XEN_NETTXF_extra_info;
655 gso->u.gso.size = skb_shinfo(skb)->gso_size;
656 gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ?
657 XEN_NETIF_GSO_TYPE_TCPV6 :
658 XEN_NETIF_GSO_TYPE_TCPV4;
660 gso->u.gso.features = 0;
662 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
666 queue->tx.req_prod_pvt = i + 1;
668 xennet_make_frags(skb, queue, tx);
671 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
673 notify_remote_via_irq(queue->tx_irq);
675 u64_stats_update_begin(&stats->syncp);
676 stats->tx_bytes += skb->len;
678 u64_stats_update_end(&stats->syncp);
680 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
681 xennet_tx_buf_gc(queue);
683 if (!netfront_tx_slot_available(queue))
684 netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
686 spin_unlock_irqrestore(&queue->tx_lock, flags);
691 dev->stats.tx_dropped++;
692 dev_kfree_skb_any(skb);
696 static int xennet_close(struct net_device *dev)
698 struct netfront_info *np = netdev_priv(dev);
699 unsigned int num_queues = dev->real_num_tx_queues;
701 struct netfront_queue *queue;
702 netif_tx_stop_all_queues(np->netdev);
703 for (i = 0; i < num_queues; ++i) {
704 queue = &np->queues[i];
705 napi_disable(&queue->napi);
710 static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb,
713 int new = xennet_rxidx(queue->rx.req_prod_pvt);
715 BUG_ON(queue->rx_skbs[new]);
716 queue->rx_skbs[new] = skb;
717 queue->grant_rx_ref[new] = ref;
718 RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new;
719 RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref;
720 queue->rx.req_prod_pvt++;
723 static int xennet_get_extras(struct netfront_queue *queue,
724 struct xen_netif_extra_info *extras,
728 struct xen_netif_extra_info *extra;
729 struct device *dev = &queue->info->netdev->dev;
730 RING_IDX cons = queue->rx.rsp_cons;
737 if (unlikely(cons + 1 == rp)) {
739 dev_warn(dev, "Missing extra info\n");
744 extra = (struct xen_netif_extra_info *)
745 RING_GET_RESPONSE(&queue->rx, ++cons);
747 if (unlikely(!extra->type ||
748 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
750 dev_warn(dev, "Invalid extra type: %d\n",
754 memcpy(&extras[extra->type - 1], extra,
758 skb = xennet_get_rx_skb(queue, cons);
759 ref = xennet_get_rx_ref(queue, cons);
760 xennet_move_rx_slot(queue, skb, ref);
761 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
763 queue->rx.rsp_cons = cons;
767 static int xennet_get_responses(struct netfront_queue *queue,
768 struct netfront_rx_info *rinfo, RING_IDX rp,
769 struct sk_buff_head *list)
771 struct xen_netif_rx_response *rx = &rinfo->rx;
772 struct xen_netif_extra_info *extras = rinfo->extras;
773 struct device *dev = &queue->info->netdev->dev;
774 RING_IDX cons = queue->rx.rsp_cons;
775 struct sk_buff *skb = xennet_get_rx_skb(queue, cons);
776 grant_ref_t ref = xennet_get_rx_ref(queue, cons);
777 int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD);
782 if (rx->flags & XEN_NETRXF_extra_info) {
783 err = xennet_get_extras(queue, extras, rp);
784 cons = queue->rx.rsp_cons;
788 if (unlikely(rx->status < 0 ||
789 rx->offset + rx->status > PAGE_SIZE)) {
791 dev_warn(dev, "rx->offset: %x, size: %u\n",
792 rx->offset, rx->status);
793 xennet_move_rx_slot(queue, skb, ref);
799 * This definitely indicates a bug, either in this driver or in
800 * the backend driver. In future this should flag the bad
801 * situation to the system controller to reboot the backend.
803 if (ref == GRANT_INVALID_REF) {
805 dev_warn(dev, "Bad rx response id %d.\n",
811 ret = gnttab_end_foreign_access_ref(ref, 0);
814 gnttab_release_grant_reference(&queue->gref_rx_head, ref);
816 __skb_queue_tail(list, skb);
819 if (!(rx->flags & XEN_NETRXF_more_data))
822 if (cons + slots == rp) {
824 dev_warn(dev, "Need more slots\n");
829 rx = RING_GET_RESPONSE(&queue->rx, cons + slots);
830 skb = xennet_get_rx_skb(queue, cons + slots);
831 ref = xennet_get_rx_ref(queue, cons + slots);
835 if (unlikely(slots > max)) {
837 dev_warn(dev, "Too many slots\n");
842 queue->rx.rsp_cons = cons + slots;
847 static int xennet_set_skb_gso(struct sk_buff *skb,
848 struct xen_netif_extra_info *gso)
850 if (!gso->u.gso.size) {
852 pr_warn("GSO size must not be zero\n");
856 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 &&
857 gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) {
859 pr_warn("Bad GSO type %d\n", gso->u.gso.type);
863 skb_shinfo(skb)->gso_size = gso->u.gso.size;
864 skb_shinfo(skb)->gso_type =
865 (gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ?
869 /* Header must be checked, and gso_segs computed. */
870 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
871 skb_shinfo(skb)->gso_segs = 0;
876 static RING_IDX xennet_fill_frags(struct netfront_queue *queue,
878 struct sk_buff_head *list)
880 struct skb_shared_info *shinfo = skb_shinfo(skb);
881 RING_IDX cons = queue->rx.rsp_cons;
882 struct sk_buff *nskb;
884 while ((nskb = __skb_dequeue(list))) {
885 struct xen_netif_rx_response *rx =
886 RING_GET_RESPONSE(&queue->rx, ++cons);
887 skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
889 if (shinfo->nr_frags == MAX_SKB_FRAGS) {
890 unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
892 BUG_ON(pull_to <= skb_headlen(skb));
893 __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
895 BUG_ON(shinfo->nr_frags >= MAX_SKB_FRAGS);
897 skb_add_rx_frag(skb, shinfo->nr_frags, skb_frag_page(nfrag),
898 rx->offset, rx->status, PAGE_SIZE);
900 skb_shinfo(nskb)->nr_frags = 0;
907 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
909 bool recalculate_partial_csum = false;
912 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
913 * peers can fail to set NETRXF_csum_blank when sending a GSO
914 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
915 * recalculate the partial checksum.
917 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
918 struct netfront_info *np = netdev_priv(dev);
919 atomic_inc(&np->rx_gso_checksum_fixup);
920 skb->ip_summed = CHECKSUM_PARTIAL;
921 recalculate_partial_csum = true;
924 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
925 if (skb->ip_summed != CHECKSUM_PARTIAL)
928 return skb_checksum_setup(skb, recalculate_partial_csum);
931 static int handle_incoming_queue(struct netfront_queue *queue,
932 struct sk_buff_head *rxq)
934 struct netfront_stats *stats = this_cpu_ptr(queue->info->stats);
935 int packets_dropped = 0;
938 while ((skb = __skb_dequeue(rxq)) != NULL) {
939 int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
941 if (pull_to > skb_headlen(skb))
942 __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
944 /* Ethernet work: Delayed to here as it peeks the header. */
945 skb->protocol = eth_type_trans(skb, queue->info->netdev);
946 skb_reset_network_header(skb);
948 if (checksum_setup(queue->info->netdev, skb)) {
951 queue->info->netdev->stats.rx_errors++;
955 u64_stats_update_begin(&stats->syncp);
957 stats->rx_bytes += skb->len;
958 u64_stats_update_end(&stats->syncp);
961 napi_gro_receive(&queue->napi, skb);
964 return packets_dropped;
967 static int xennet_poll(struct napi_struct *napi, int budget)
969 struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi);
970 struct net_device *dev = queue->info->netdev;
972 struct netfront_rx_info rinfo;
973 struct xen_netif_rx_response *rx = &rinfo.rx;
974 struct xen_netif_extra_info *extras = rinfo.extras;
977 struct sk_buff_head rxq;
978 struct sk_buff_head errq;
979 struct sk_buff_head tmpq;
983 spin_lock(&queue->rx_lock);
985 skb_queue_head_init(&rxq);
986 skb_queue_head_init(&errq);
987 skb_queue_head_init(&tmpq);
989 rp = queue->rx.sring->rsp_prod;
990 rmb(); /* Ensure we see queued responses up to 'rp'. */
992 i = queue->rx.rsp_cons;
994 while ((i != rp) && (work_done < budget)) {
995 memcpy(rx, RING_GET_RESPONSE(&queue->rx, i), sizeof(*rx));
996 memset(extras, 0, sizeof(rinfo.extras));
998 err = xennet_get_responses(queue, &rinfo, rp, &tmpq);
1000 if (unlikely(err)) {
1002 while ((skb = __skb_dequeue(&tmpq)))
1003 __skb_queue_tail(&errq, skb);
1004 dev->stats.rx_errors++;
1005 i = queue->rx.rsp_cons;
1009 skb = __skb_dequeue(&tmpq);
1011 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1012 struct xen_netif_extra_info *gso;
1013 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1015 if (unlikely(xennet_set_skb_gso(skb, gso))) {
1016 __skb_queue_head(&tmpq, skb);
1017 queue->rx.rsp_cons += skb_queue_len(&tmpq);
1022 NETFRONT_SKB_CB(skb)->pull_to = rx->status;
1023 if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
1024 NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
1026 skb_shinfo(skb)->frags[0].page_offset = rx->offset;
1027 skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
1028 skb->data_len = rx->status;
1029 skb->len += rx->status;
1031 i = xennet_fill_frags(queue, skb, &tmpq);
1033 if (rx->flags & XEN_NETRXF_csum_blank)
1034 skb->ip_summed = CHECKSUM_PARTIAL;
1035 else if (rx->flags & XEN_NETRXF_data_validated)
1036 skb->ip_summed = CHECKSUM_UNNECESSARY;
1038 __skb_queue_tail(&rxq, skb);
1040 queue->rx.rsp_cons = ++i;
1044 __skb_queue_purge(&errq);
1046 work_done -= handle_incoming_queue(queue, &rxq);
1048 xennet_alloc_rx_buffers(queue);
1050 if (work_done < budget) {
1053 napi_gro_flush(napi, false);
1055 local_irq_save(flags);
1057 RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do);
1059 __napi_complete(napi);
1061 local_irq_restore(flags);
1064 spin_unlock(&queue->rx_lock);
1069 static int xennet_change_mtu(struct net_device *dev, int mtu)
1071 int max = xennet_can_sg(dev) ?
1072 XEN_NETIF_MAX_TX_SIZE - MAX_TCP_HEADER : ETH_DATA_LEN;
1080 static struct rtnl_link_stats64 *xennet_get_stats64(struct net_device *dev,
1081 struct rtnl_link_stats64 *tot)
1083 struct netfront_info *np = netdev_priv(dev);
1086 for_each_possible_cpu(cpu) {
1087 struct netfront_stats *stats = per_cpu_ptr(np->stats, cpu);
1088 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1092 start = u64_stats_fetch_begin_irq(&stats->syncp);
1094 rx_packets = stats->rx_packets;
1095 tx_packets = stats->tx_packets;
1096 rx_bytes = stats->rx_bytes;
1097 tx_bytes = stats->tx_bytes;
1098 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1100 tot->rx_packets += rx_packets;
1101 tot->tx_packets += tx_packets;
1102 tot->rx_bytes += rx_bytes;
1103 tot->tx_bytes += tx_bytes;
1106 tot->rx_errors = dev->stats.rx_errors;
1107 tot->tx_dropped = dev->stats.tx_dropped;
1112 static void xennet_release_tx_bufs(struct netfront_queue *queue)
1114 struct sk_buff *skb;
1117 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1118 /* Skip over entries which are actually freelist references */
1119 if (skb_entry_is_link(&queue->tx_skbs[i]))
1122 skb = queue->tx_skbs[i].skb;
1123 get_page(queue->grant_tx_page[i]);
1124 gnttab_end_foreign_access(queue->grant_tx_ref[i],
1126 (unsigned long)page_address(queue->grant_tx_page[i]));
1127 queue->grant_tx_page[i] = NULL;
1128 queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1129 add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, i);
1130 dev_kfree_skb_irq(skb);
1134 static void xennet_release_rx_bufs(struct netfront_queue *queue)
1138 spin_lock_bh(&queue->rx_lock);
1140 for (id = 0; id < NET_RX_RING_SIZE; id++) {
1141 struct sk_buff *skb;
1144 skb = queue->rx_skbs[id];
1148 ref = queue->grant_rx_ref[id];
1149 if (ref == GRANT_INVALID_REF)
1152 page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
1154 /* gnttab_end_foreign_access() needs a page ref until
1155 * foreign access is ended (which may be deferred).
1158 gnttab_end_foreign_access(ref, 0,
1159 (unsigned long)page_address(page));
1160 queue->grant_rx_ref[id] = GRANT_INVALID_REF;
1165 spin_unlock_bh(&queue->rx_lock);
1168 static netdev_features_t xennet_fix_features(struct net_device *dev,
1169 netdev_features_t features)
1171 struct netfront_info *np = netdev_priv(dev);
1174 if (features & NETIF_F_SG) {
1175 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg",
1180 features &= ~NETIF_F_SG;
1183 if (features & NETIF_F_IPV6_CSUM) {
1184 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1185 "feature-ipv6-csum-offload", "%d", &val) < 0)
1189 features &= ~NETIF_F_IPV6_CSUM;
1192 if (features & NETIF_F_TSO) {
1193 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1194 "feature-gso-tcpv4", "%d", &val) < 0)
1198 features &= ~NETIF_F_TSO;
1201 if (features & NETIF_F_TSO6) {
1202 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1203 "feature-gso-tcpv6", "%d", &val) < 0)
1207 features &= ~NETIF_F_TSO6;
1213 static int xennet_set_features(struct net_device *dev,
1214 netdev_features_t features)
1216 if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1217 netdev_info(dev, "Reducing MTU because no SG offload");
1218 dev->mtu = ETH_DATA_LEN;
1224 static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1226 struct netfront_queue *queue = dev_id;
1227 unsigned long flags;
1229 spin_lock_irqsave(&queue->tx_lock, flags);
1230 xennet_tx_buf_gc(queue);
1231 spin_unlock_irqrestore(&queue->tx_lock, flags);
1236 static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1238 struct netfront_queue *queue = dev_id;
1239 struct net_device *dev = queue->info->netdev;
1241 if (likely(netif_carrier_ok(dev) &&
1242 RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)))
1243 napi_schedule(&queue->napi);
1248 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1250 xennet_tx_interrupt(irq, dev_id);
1251 xennet_rx_interrupt(irq, dev_id);
1255 #ifdef CONFIG_NET_POLL_CONTROLLER
1256 static void xennet_poll_controller(struct net_device *dev)
1258 /* Poll each queue */
1259 struct netfront_info *info = netdev_priv(dev);
1260 unsigned int num_queues = dev->real_num_tx_queues;
1262 for (i = 0; i < num_queues; ++i)
1263 xennet_interrupt(0, &info->queues[i]);
1267 static const struct net_device_ops xennet_netdev_ops = {
1268 .ndo_open = xennet_open,
1269 .ndo_stop = xennet_close,
1270 .ndo_start_xmit = xennet_start_xmit,
1271 .ndo_change_mtu = xennet_change_mtu,
1272 .ndo_get_stats64 = xennet_get_stats64,
1273 .ndo_set_mac_address = eth_mac_addr,
1274 .ndo_validate_addr = eth_validate_addr,
1275 .ndo_fix_features = xennet_fix_features,
1276 .ndo_set_features = xennet_set_features,
1277 .ndo_select_queue = xennet_select_queue,
1278 #ifdef CONFIG_NET_POLL_CONTROLLER
1279 .ndo_poll_controller = xennet_poll_controller,
1283 static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1286 struct net_device *netdev;
1287 struct netfront_info *np;
1289 netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues);
1291 return ERR_PTR(-ENOMEM);
1293 np = netdev_priv(netdev);
1296 /* No need to use rtnl_lock() before the call below as it
1297 * happens before register_netdev().
1299 netif_set_real_num_tx_queues(netdev, 0);
1303 np->stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1304 if (np->stats == NULL)
1307 netdev->netdev_ops = &xennet_netdev_ops;
1309 netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1311 netdev->hw_features = NETIF_F_SG |
1313 NETIF_F_TSO | NETIF_F_TSO6;
1316 * Assume that all hw features are available for now. This set
1317 * will be adjusted by the call to netdev_update_features() in
1318 * xennet_connect() which is the earliest point where we can
1319 * negotiate with the backend regarding supported features.
1321 netdev->features |= netdev->hw_features;
1323 netdev->ethtool_ops = &xennet_ethtool_ops;
1324 SET_NETDEV_DEV(netdev, &dev->dev);
1326 netif_set_gso_max_size(netdev, XEN_NETIF_MAX_TX_SIZE - MAX_TCP_HEADER);
1328 np->netdev = netdev;
1330 netif_carrier_off(netdev);
1335 free_netdev(netdev);
1336 return ERR_PTR(err);
1340 * Entry point to this code when a new device is created. Allocate the basic
1341 * structures and the ring buffers for communication with the backend, and
1342 * inform the backend of the appropriate details for those.
1344 static int netfront_probe(struct xenbus_device *dev,
1345 const struct xenbus_device_id *id)
1348 struct net_device *netdev;
1349 struct netfront_info *info;
1351 netdev = xennet_create_dev(dev);
1352 if (IS_ERR(netdev)) {
1353 err = PTR_ERR(netdev);
1354 xenbus_dev_fatal(dev, err, "creating netdev");
1358 info = netdev_priv(netdev);
1359 dev_set_drvdata(&dev->dev, info);
1361 err = register_netdev(info->netdev);
1363 pr_warn("%s: register_netdev err=%d\n", __func__, err);
1367 err = xennet_sysfs_addif(info->netdev);
1369 unregister_netdev(info->netdev);
1370 pr_warn("%s: add sysfs failed err=%d\n", __func__, err);
1377 free_netdev(netdev);
1378 dev_set_drvdata(&dev->dev, NULL);
1382 static void xennet_end_access(int ref, void *page)
1384 /* This frees the page as a side-effect */
1385 if (ref != GRANT_INVALID_REF)
1386 gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1389 static void xennet_disconnect_backend(struct netfront_info *info)
1392 unsigned int num_queues = info->netdev->real_num_tx_queues;
1394 netif_carrier_off(info->netdev);
1396 for (i = 0; i < num_queues; ++i) {
1397 struct netfront_queue *queue = &info->queues[i];
1399 if (queue->tx_irq && (queue->tx_irq == queue->rx_irq))
1400 unbind_from_irqhandler(queue->tx_irq, queue);
1401 if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) {
1402 unbind_from_irqhandler(queue->tx_irq, queue);
1403 unbind_from_irqhandler(queue->rx_irq, queue);
1405 queue->tx_evtchn = queue->rx_evtchn = 0;
1406 queue->tx_irq = queue->rx_irq = 0;
1408 napi_synchronize(&queue->napi);
1410 xennet_release_tx_bufs(queue);
1411 xennet_release_rx_bufs(queue);
1412 gnttab_free_grant_references(queue->gref_tx_head);
1413 gnttab_free_grant_references(queue->gref_rx_head);
1415 /* End access and free the pages */
1416 xennet_end_access(queue->tx_ring_ref, queue->tx.sring);
1417 xennet_end_access(queue->rx_ring_ref, queue->rx.sring);
1419 queue->tx_ring_ref = GRANT_INVALID_REF;
1420 queue->rx_ring_ref = GRANT_INVALID_REF;
1421 queue->tx.sring = NULL;
1422 queue->rx.sring = NULL;
1427 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1428 * driver restart. We tear down our netif structure and recreate it, but
1429 * leave the device-layer structures intact so that this is transparent to the
1430 * rest of the kernel.
1432 static int netfront_resume(struct xenbus_device *dev)
1434 struct netfront_info *info = dev_get_drvdata(&dev->dev);
1436 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1438 xennet_disconnect_backend(info);
1442 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1444 char *s, *e, *macstr;
1447 macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1449 return PTR_ERR(macstr);
1451 for (i = 0; i < ETH_ALEN; i++) {
1452 mac[i] = simple_strtoul(s, &e, 16);
1453 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1464 static int setup_netfront_single(struct netfront_queue *queue)
1468 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1472 err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1474 0, queue->info->netdev->name, queue);
1477 queue->rx_evtchn = queue->tx_evtchn;
1478 queue->rx_irq = queue->tx_irq = err;
1483 xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1484 queue->tx_evtchn = 0;
1489 static int setup_netfront_split(struct netfront_queue *queue)
1493 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1496 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn);
1498 goto alloc_rx_evtchn_fail;
1500 snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
1501 "%s-tx", queue->name);
1502 err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1503 xennet_tx_interrupt,
1504 0, queue->tx_irq_name, queue);
1507 queue->tx_irq = err;
1509 snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
1510 "%s-rx", queue->name);
1511 err = bind_evtchn_to_irqhandler(queue->rx_evtchn,
1512 xennet_rx_interrupt,
1513 0, queue->rx_irq_name, queue);
1516 queue->rx_irq = err;
1521 unbind_from_irqhandler(queue->tx_irq, queue);
1524 xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn);
1525 queue->rx_evtchn = 0;
1526 alloc_rx_evtchn_fail:
1527 xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1528 queue->tx_evtchn = 0;
1533 static int setup_netfront(struct xenbus_device *dev,
1534 struct netfront_queue *queue, unsigned int feature_split_evtchn)
1536 struct xen_netif_tx_sring *txs;
1537 struct xen_netif_rx_sring *rxs;
1540 queue->tx_ring_ref = GRANT_INVALID_REF;
1541 queue->rx_ring_ref = GRANT_INVALID_REF;
1542 queue->rx.sring = NULL;
1543 queue->tx.sring = NULL;
1545 txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1548 xenbus_dev_fatal(dev, err, "allocating tx ring page");
1551 SHARED_RING_INIT(txs);
1552 FRONT_RING_INIT(&queue->tx, txs, PAGE_SIZE);
1554 err = xenbus_grant_ring(dev, virt_to_mfn(txs));
1556 goto grant_tx_ring_fail;
1557 queue->tx_ring_ref = err;
1559 rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1562 xenbus_dev_fatal(dev, err, "allocating rx ring page");
1563 goto alloc_rx_ring_fail;
1565 SHARED_RING_INIT(rxs);
1566 FRONT_RING_INIT(&queue->rx, rxs, PAGE_SIZE);
1568 err = xenbus_grant_ring(dev, virt_to_mfn(rxs));
1570 goto grant_rx_ring_fail;
1571 queue->rx_ring_ref = err;
1573 if (feature_split_evtchn)
1574 err = setup_netfront_split(queue);
1575 /* setup single event channel if
1576 * a) feature-split-event-channels == 0
1577 * b) feature-split-event-channels == 1 but failed to setup
1579 if (!feature_split_evtchn || (feature_split_evtchn && err))
1580 err = setup_netfront_single(queue);
1583 goto alloc_evtchn_fail;
1587 /* If we fail to setup netfront, it is safe to just revoke access to
1588 * granted pages because backend is not accessing it at this point.
1591 gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0);
1593 free_page((unsigned long)rxs);
1595 gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0);
1597 free_page((unsigned long)txs);
1602 /* Queue-specific initialisation
1603 * This used to be done in xennet_create_dev() but must now
1606 static int xennet_init_queue(struct netfront_queue *queue)
1611 spin_lock_init(&queue->tx_lock);
1612 spin_lock_init(&queue->rx_lock);
1614 init_timer(&queue->rx_refill_timer);
1615 queue->rx_refill_timer.data = (unsigned long)queue;
1616 queue->rx_refill_timer.function = rx_refill_timeout;
1618 snprintf(queue->name, sizeof(queue->name), "%s-q%u",
1619 queue->info->netdev->name, queue->id);
1621 /* Initialise tx_skbs as a free chain containing every entry. */
1622 queue->tx_skb_freelist = 0;
1623 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1624 skb_entry_set_link(&queue->tx_skbs[i], i+1);
1625 queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1626 queue->grant_tx_page[i] = NULL;
1629 /* Clear out rx_skbs */
1630 for (i = 0; i < NET_RX_RING_SIZE; i++) {
1631 queue->rx_skbs[i] = NULL;
1632 queue->grant_rx_ref[i] = GRANT_INVALID_REF;
1635 /* A grant for every tx ring slot */
1636 if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
1637 &queue->gref_tx_head) < 0) {
1638 pr_alert("can't alloc tx grant refs\n");
1643 /* A grant for every rx ring slot */
1644 if (gnttab_alloc_grant_references(NET_RX_RING_SIZE,
1645 &queue->gref_rx_head) < 0) {
1646 pr_alert("can't alloc rx grant refs\n");
1654 gnttab_free_grant_references(queue->gref_tx_head);
1659 static int write_queue_xenstore_keys(struct netfront_queue *queue,
1660 struct xenbus_transaction *xbt, int write_hierarchical)
1662 /* Write the queue-specific keys into XenStore in the traditional
1663 * way for a single queue, or in a queue subkeys for multiple
1666 struct xenbus_device *dev = queue->info->xbdev;
1668 const char *message;
1672 /* Choose the correct place to write the keys */
1673 if (write_hierarchical) {
1674 pathsize = strlen(dev->nodename) + 10;
1675 path = kzalloc(pathsize, GFP_KERNEL);
1678 message = "out of memory while writing ring references";
1681 snprintf(path, pathsize, "%s/queue-%u",
1682 dev->nodename, queue->id);
1684 path = (char *)dev->nodename;
1687 /* Write ring references */
1688 err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u",
1689 queue->tx_ring_ref);
1691 message = "writing tx-ring-ref";
1695 err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u",
1696 queue->rx_ring_ref);
1698 message = "writing rx-ring-ref";
1702 /* Write event channels; taking into account both shared
1703 * and split event channel scenarios.
1705 if (queue->tx_evtchn == queue->rx_evtchn) {
1706 /* Shared event channel */
1707 err = xenbus_printf(*xbt, path,
1708 "event-channel", "%u", queue->tx_evtchn);
1710 message = "writing event-channel";
1714 /* Split event channels */
1715 err = xenbus_printf(*xbt, path,
1716 "event-channel-tx", "%u", queue->tx_evtchn);
1718 message = "writing event-channel-tx";
1722 err = xenbus_printf(*xbt, path,
1723 "event-channel-rx", "%u", queue->rx_evtchn);
1725 message = "writing event-channel-rx";
1730 if (write_hierarchical)
1735 if (write_hierarchical)
1737 xenbus_dev_fatal(dev, err, "%s", message);
1741 static void xennet_destroy_queues(struct netfront_info *info)
1747 for (i = 0; i < info->netdev->real_num_tx_queues; i++) {
1748 struct netfront_queue *queue = &info->queues[i];
1750 if (netif_running(info->netdev))
1751 napi_disable(&queue->napi);
1752 netif_napi_del(&queue->napi);
1757 kfree(info->queues);
1758 info->queues = NULL;
1761 static int xennet_create_queues(struct netfront_info *info,
1762 unsigned int num_queues)
1767 info->queues = kcalloc(num_queues, sizeof(struct netfront_queue),
1774 for (i = 0; i < num_queues; i++) {
1775 struct netfront_queue *queue = &info->queues[i];
1780 ret = xennet_init_queue(queue);
1782 dev_warn(&info->netdev->dev,
1783 "only created %d queues\n", i);
1788 netif_napi_add(queue->info->netdev, &queue->napi,
1790 if (netif_running(info->netdev))
1791 napi_enable(&queue->napi);
1794 netif_set_real_num_tx_queues(info->netdev, num_queues);
1798 if (num_queues == 0) {
1799 dev_err(&info->netdev->dev, "no queues\n");
1805 /* Common code used when first setting up, and when resuming. */
1806 static int talk_to_netback(struct xenbus_device *dev,
1807 struct netfront_info *info)
1809 const char *message;
1810 struct xenbus_transaction xbt;
1812 unsigned int feature_split_evtchn;
1814 unsigned int max_queues = 0;
1815 struct netfront_queue *queue = NULL;
1816 unsigned int num_queues = 1;
1818 info->netdev->irq = 0;
1820 /* Check if backend supports multiple queues */
1821 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1822 "multi-queue-max-queues", "%u", &max_queues);
1825 num_queues = min(max_queues, xennet_max_queues);
1827 /* Check feature-split-event-channels */
1828 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1829 "feature-split-event-channels", "%u",
1830 &feature_split_evtchn);
1832 feature_split_evtchn = 0;
1834 /* Read mac addr. */
1835 err = xen_net_read_mac(dev, info->netdev->dev_addr);
1837 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1842 xennet_destroy_queues(info);
1844 err = xennet_create_queues(info, num_queues);
1848 /* Create shared ring, alloc event channel -- for each queue */
1849 for (i = 0; i < num_queues; ++i) {
1850 queue = &info->queues[i];
1851 err = setup_netfront(dev, queue, feature_split_evtchn);
1853 /* setup_netfront() will tidy up the current
1854 * queue on error, but we need to clean up
1855 * those already allocated.
1859 netif_set_real_num_tx_queues(info->netdev, i);
1869 err = xenbus_transaction_start(&xbt);
1871 xenbus_dev_fatal(dev, err, "starting transaction");
1875 if (num_queues == 1) {
1876 err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */
1878 goto abort_transaction_no_dev_fatal;
1880 /* Write the number of queues */
1881 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues",
1884 message = "writing multi-queue-num-queues";
1885 goto abort_transaction_no_dev_fatal;
1888 /* Write the keys for each queue */
1889 for (i = 0; i < num_queues; ++i) {
1890 queue = &info->queues[i];
1891 err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */
1893 goto abort_transaction_no_dev_fatal;
1897 /* The remaining keys are not queue-specific */
1898 err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1901 message = "writing request-rx-copy";
1902 goto abort_transaction;
1905 err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1907 message = "writing feature-rx-notify";
1908 goto abort_transaction;
1911 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1913 message = "writing feature-sg";
1914 goto abort_transaction;
1917 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1919 message = "writing feature-gso-tcpv4";
1920 goto abort_transaction;
1923 err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1");
1925 message = "writing feature-gso-tcpv6";
1926 goto abort_transaction;
1929 err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload",
1932 message = "writing feature-ipv6-csum-offload";
1933 goto abort_transaction;
1936 err = xenbus_transaction_end(xbt, 0);
1940 xenbus_dev_fatal(dev, err, "completing transaction");
1947 xenbus_dev_fatal(dev, err, "%s", message);
1948 abort_transaction_no_dev_fatal:
1949 xenbus_transaction_end(xbt, 1);
1951 xennet_disconnect_backend(info);
1952 kfree(info->queues);
1953 info->queues = NULL;
1955 netif_set_real_num_tx_queues(info->netdev, 0);
1961 static int xennet_connect(struct net_device *dev)
1963 struct netfront_info *np = netdev_priv(dev);
1964 unsigned int num_queues = 0;
1966 unsigned int feature_rx_copy;
1968 struct netfront_queue *queue = NULL;
1970 err = xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1971 "feature-rx-copy", "%u", &feature_rx_copy);
1973 feature_rx_copy = 0;
1975 if (!feature_rx_copy) {
1977 "backend does not support copying receive path\n");
1981 err = talk_to_netback(np->xbdev, np);
1985 /* talk_to_netback() sets the correct number of queues */
1986 num_queues = dev->real_num_tx_queues;
1989 netdev_update_features(dev);
1993 * All public and private state should now be sane. Get
1994 * ready to start sending and receiving packets and give the driver
1995 * domain a kick because we've probably just requeued some
1998 netif_carrier_on(np->netdev);
1999 for (j = 0; j < num_queues; ++j) {
2000 queue = &np->queues[j];
2002 notify_remote_via_irq(queue->tx_irq);
2003 if (queue->tx_irq != queue->rx_irq)
2004 notify_remote_via_irq(queue->rx_irq);
2006 spin_lock_irq(&queue->tx_lock);
2007 xennet_tx_buf_gc(queue);
2008 spin_unlock_irq(&queue->tx_lock);
2010 spin_lock_bh(&queue->rx_lock);
2011 xennet_alloc_rx_buffers(queue);
2012 spin_unlock_bh(&queue->rx_lock);
2019 * Callback received when the backend's state changes.
2021 static void netback_changed(struct xenbus_device *dev,
2022 enum xenbus_state backend_state)
2024 struct netfront_info *np = dev_get_drvdata(&dev->dev);
2025 struct net_device *netdev = np->netdev;
2027 dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
2029 switch (backend_state) {
2030 case XenbusStateInitialising:
2031 case XenbusStateInitialised:
2032 case XenbusStateReconfiguring:
2033 case XenbusStateReconfigured:
2034 case XenbusStateUnknown:
2037 case XenbusStateInitWait:
2038 if (dev->state != XenbusStateInitialising)
2040 if (xennet_connect(netdev) != 0)
2042 xenbus_switch_state(dev, XenbusStateConnected);
2045 case XenbusStateConnected:
2046 netdev_notify_peers(netdev);
2049 case XenbusStateClosed:
2050 if (dev->state == XenbusStateClosed)
2052 /* Missed the backend's CLOSING state -- fallthrough */
2053 case XenbusStateClosing:
2054 xenbus_frontend_closed(dev);
2059 static const struct xennet_stat {
2060 char name[ETH_GSTRING_LEN];
2062 } xennet_stats[] = {
2064 "rx_gso_checksum_fixup",
2065 offsetof(struct netfront_info, rx_gso_checksum_fixup)
2069 static int xennet_get_sset_count(struct net_device *dev, int string_set)
2071 switch (string_set) {
2073 return ARRAY_SIZE(xennet_stats);
2079 static void xennet_get_ethtool_stats(struct net_device *dev,
2080 struct ethtool_stats *stats, u64 * data)
2082 void *np = netdev_priv(dev);
2085 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2086 data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset));
2089 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
2093 switch (stringset) {
2095 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2096 memcpy(data + i * ETH_GSTRING_LEN,
2097 xennet_stats[i].name, ETH_GSTRING_LEN);
2102 static const struct ethtool_ops xennet_ethtool_ops =
2104 .get_link = ethtool_op_get_link,
2106 .get_sset_count = xennet_get_sset_count,
2107 .get_ethtool_stats = xennet_get_ethtool_stats,
2108 .get_strings = xennet_get_strings,
2112 static ssize_t show_rxbuf(struct device *dev,
2113 struct device_attribute *attr, char *buf)
2115 return sprintf(buf, "%lu\n", NET_RX_RING_SIZE);
2118 static ssize_t store_rxbuf(struct device *dev,
2119 struct device_attribute *attr,
2120 const char *buf, size_t len)
2123 unsigned long target;
2125 if (!capable(CAP_NET_ADMIN))
2128 target = simple_strtoul(buf, &endp, 0);
2132 /* rxbuf_min and rxbuf_max are no longer configurable. */
2137 static struct device_attribute xennet_attrs[] = {
2138 __ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf),
2139 __ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf, store_rxbuf),
2140 __ATTR(rxbuf_cur, S_IRUGO, show_rxbuf, NULL),
2143 static int xennet_sysfs_addif(struct net_device *netdev)
2148 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) {
2149 err = device_create_file(&netdev->dev,
2158 device_remove_file(&netdev->dev, &xennet_attrs[i]);
2162 static void xennet_sysfs_delif(struct net_device *netdev)
2166 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++)
2167 device_remove_file(&netdev->dev, &xennet_attrs[i]);
2170 #endif /* CONFIG_SYSFS */
2172 static int xennet_remove(struct xenbus_device *dev)
2174 struct netfront_info *info = dev_get_drvdata(&dev->dev);
2175 unsigned int num_queues = info->netdev->real_num_tx_queues;
2176 struct netfront_queue *queue = NULL;
2179 dev_dbg(&dev->dev, "%s\n", dev->nodename);
2181 xennet_disconnect_backend(info);
2183 xennet_sysfs_delif(info->netdev);
2185 unregister_netdev(info->netdev);
2187 for (i = 0; i < num_queues; ++i) {
2188 queue = &info->queues[i];
2189 del_timer_sync(&queue->rx_refill_timer);
2193 kfree(info->queues);
2194 info->queues = NULL;
2197 free_percpu(info->stats);
2199 free_netdev(info->netdev);
2204 static const struct xenbus_device_id netfront_ids[] = {
2209 static struct xenbus_driver netfront_driver = {
2210 .ids = netfront_ids,
2211 .probe = netfront_probe,
2212 .remove = xennet_remove,
2213 .resume = netfront_resume,
2214 .otherend_changed = netback_changed,
2217 static int __init netif_init(void)
2222 if (!xen_has_pv_nic_devices())
2225 pr_info("Initialising Xen virtual ethernet driver\n");
2227 /* Allow as many queues as there are CPUs, by default */
2228 xennet_max_queues = num_online_cpus();
2230 return xenbus_register_frontend(&netfront_driver);
2232 module_init(netif_init);
2235 static void __exit netif_exit(void)
2237 xenbus_unregister_driver(&netfront_driver);
2239 module_exit(netif_exit);
2241 MODULE_DESCRIPTION("Xen virtual network device frontend");
2242 MODULE_LICENSE("GPL");
2243 MODULE_ALIAS("xen:vif");
2244 MODULE_ALIAS("xennet");