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)
80 #define TX_MAX_TARGET min_t(int, NET_TX_RING_SIZE, 256)
82 /* Queue name is interface name with "-qNNN" appended */
83 #define QUEUE_NAME_SIZE (IFNAMSIZ + 6)
85 /* IRQ name is queue name with "-tx" or "-rx" appended */
86 #define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3)
88 struct netfront_stats {
93 struct u64_stats_sync syncp;
98 struct netfront_queue {
99 unsigned int id; /* Queue ID, 0-based */
100 char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */
101 struct netfront_info *info;
103 struct napi_struct napi;
105 /* Split event channels support, tx_* == rx_* when using
106 * single event channel.
108 unsigned int tx_evtchn, rx_evtchn;
109 unsigned int tx_irq, rx_irq;
110 /* Only used when split event channels support is enabled */
111 char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */
112 char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */
115 struct xen_netif_tx_front_ring tx;
119 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
120 * are linked from tx_skb_freelist through skb_entry.link.
122 * NB. Freelist index entries are always going to be less than
123 * PAGE_OFFSET, whereas pointers to skbs will always be equal or
124 * greater than PAGE_OFFSET: we use this property to distinguish
130 } tx_skbs[NET_TX_RING_SIZE];
131 grant_ref_t gref_tx_head;
132 grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
133 struct page *grant_tx_page[NET_TX_RING_SIZE];
134 unsigned tx_skb_freelist;
136 spinlock_t rx_lock ____cacheline_aligned_in_smp;
137 struct xen_netif_rx_front_ring rx;
140 /* Receive-ring batched refills. */
141 #define RX_MIN_TARGET 8
142 #define RX_DFL_MIN_TARGET 64
143 #define RX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256)
144 unsigned rx_min_target, rx_max_target, rx_target;
145 struct sk_buff_head rx_batch;
147 struct timer_list rx_refill_timer;
149 struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
150 grant_ref_t gref_rx_head;
151 grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
153 unsigned long rx_pfn_array[NET_RX_RING_SIZE];
154 struct multicall_entry rx_mcl[NET_RX_RING_SIZE+1];
155 struct mmu_update rx_mmu[NET_RX_RING_SIZE];
158 struct netfront_info {
159 struct list_head list;
160 struct net_device *netdev;
162 struct xenbus_device *xbdev;
164 /* Multi-queue support */
165 struct netfront_queue *queues;
168 struct netfront_stats __percpu *stats;
170 atomic_t rx_gso_checksum_fixup;
173 struct netfront_rx_info {
174 struct xen_netif_rx_response rx;
175 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
178 static void skb_entry_set_link(union skb_entry *list, unsigned short id)
183 static int skb_entry_is_link(const union skb_entry *list)
185 BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link));
186 return (unsigned long)list->skb < PAGE_OFFSET;
190 * Access macros for acquiring freeing slots in tx_skbs[].
193 static void add_id_to_freelist(unsigned *head, union skb_entry *list,
196 skb_entry_set_link(&list[id], *head);
200 static unsigned short get_id_from_freelist(unsigned *head,
201 union skb_entry *list)
203 unsigned int id = *head;
204 *head = list[id].link;
208 static int xennet_rxidx(RING_IDX idx)
210 return idx & (NET_RX_RING_SIZE - 1);
213 static struct sk_buff *xennet_get_rx_skb(struct netfront_queue *queue,
216 int i = xennet_rxidx(ri);
217 struct sk_buff *skb = queue->rx_skbs[i];
218 queue->rx_skbs[i] = NULL;
222 static grant_ref_t xennet_get_rx_ref(struct netfront_queue *queue,
225 int i = xennet_rxidx(ri);
226 grant_ref_t ref = queue->grant_rx_ref[i];
227 queue->grant_rx_ref[i] = GRANT_INVALID_REF;
232 static int xennet_sysfs_addif(struct net_device *netdev);
233 static void xennet_sysfs_delif(struct net_device *netdev);
234 #else /* !CONFIG_SYSFS */
235 #define xennet_sysfs_addif(dev) (0)
236 #define xennet_sysfs_delif(dev) do { } while (0)
239 static bool xennet_can_sg(struct net_device *dev)
241 return dev->features & NETIF_F_SG;
245 static void rx_refill_timeout(unsigned long data)
247 struct netfront_queue *queue = (struct netfront_queue *)data;
248 napi_schedule(&queue->napi);
251 static int netfront_tx_slot_available(struct netfront_queue *queue)
253 return (queue->tx.req_prod_pvt - queue->tx.rsp_cons) <
254 (TX_MAX_TARGET - MAX_SKB_FRAGS - 2);
257 static void xennet_maybe_wake_tx(struct netfront_queue *queue)
259 struct net_device *dev = queue->info->netdev;
260 struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, queue->id);
262 if (unlikely(netif_tx_queue_stopped(dev_queue)) &&
263 netfront_tx_slot_available(queue) &&
264 likely(netif_running(dev)))
265 netif_tx_wake_queue(netdev_get_tx_queue(dev, queue->id));
268 static void xennet_alloc_rx_buffers(struct netfront_queue *queue)
273 int i, batch_target, notify;
274 RING_IDX req_prod = queue->rx.req_prod_pvt;
278 struct xen_netif_rx_request *req;
280 if (unlikely(!netif_carrier_ok(queue->info->netdev)))
284 * Allocate skbuffs greedily, even though we batch updates to the
285 * receive ring. This creates a less bursty demand on the memory
286 * allocator, so should reduce the chance of failed allocation requests
287 * both for ourself and for other kernel subsystems.
289 batch_target = queue->rx_target - (req_prod - queue->rx.rsp_cons);
290 for (i = skb_queue_len(&queue->rx_batch); i < batch_target; i++) {
291 skb = __netdev_alloc_skb(queue->info->netdev,
292 RX_COPY_THRESHOLD + NET_IP_ALIGN,
293 GFP_ATOMIC | __GFP_NOWARN);
297 /* Align ip header to a 16 bytes boundary */
298 skb_reserve(skb, NET_IP_ALIGN);
300 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
304 /* Could not allocate any skbuffs. Try again later. */
305 mod_timer(&queue->rx_refill_timer,
308 /* Any skbuffs queued for refill? Force them out. */
314 skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE);
315 __skb_queue_tail(&queue->rx_batch, skb);
318 /* Is the batch large enough to be worthwhile? */
319 if (i < (queue->rx_target/2)) {
320 if (req_prod > queue->rx.sring->req_prod)
325 /* Adjust our fill target if we risked running out of buffers. */
326 if (((req_prod - queue->rx.sring->rsp_prod) < (queue->rx_target / 4)) &&
327 ((queue->rx_target *= 2) > queue->rx_max_target))
328 queue->rx_target = queue->rx_max_target;
332 skb = __skb_dequeue(&queue->rx_batch);
336 skb->dev = queue->info->netdev;
338 id = xennet_rxidx(req_prod + i);
340 BUG_ON(queue->rx_skbs[id]);
341 queue->rx_skbs[id] = skb;
343 ref = gnttab_claim_grant_reference(&queue->gref_rx_head);
344 BUG_ON((signed short)ref < 0);
345 queue->grant_rx_ref[id] = ref;
347 pfn = page_to_pfn(skb_frag_page(&skb_shinfo(skb)->frags[0]));
348 vaddr = page_address(skb_frag_page(&skb_shinfo(skb)->frags[0]));
350 req = RING_GET_REQUEST(&queue->rx, req_prod + i);
351 gnttab_grant_foreign_access_ref(ref,
352 queue->info->xbdev->otherend_id,
360 wmb(); /* barrier so backend seens requests */
362 /* Above is a suitable barrier to ensure backend will see requests. */
363 queue->rx.req_prod_pvt = req_prod + i;
365 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify);
367 notify_remote_via_irq(queue->rx_irq);
370 static int xennet_open(struct net_device *dev)
372 struct netfront_info *np = netdev_priv(dev);
373 unsigned int num_queues = dev->real_num_tx_queues;
375 struct netfront_queue *queue = NULL;
377 for (i = 0; i < num_queues; ++i) {
378 queue = &np->queues[i];
379 napi_enable(&queue->napi);
381 spin_lock_bh(&queue->rx_lock);
382 if (netif_carrier_ok(dev)) {
383 xennet_alloc_rx_buffers(queue);
384 queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1;
385 if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))
386 napi_schedule(&queue->napi);
388 spin_unlock_bh(&queue->rx_lock);
391 netif_tx_start_all_queues(dev);
396 static void xennet_tx_buf_gc(struct netfront_queue *queue)
402 BUG_ON(!netif_carrier_ok(queue->info->netdev));
405 prod = queue->tx.sring->rsp_prod;
406 rmb(); /* Ensure we see responses up to 'rp'. */
408 for (cons = queue->tx.rsp_cons; cons != prod; cons++) {
409 struct xen_netif_tx_response *txrsp;
411 txrsp = RING_GET_RESPONSE(&queue->tx, cons);
412 if (txrsp->status == XEN_NETIF_RSP_NULL)
416 skb = queue->tx_skbs[id].skb;
417 if (unlikely(gnttab_query_foreign_access(
418 queue->grant_tx_ref[id]) != 0)) {
419 pr_alert("%s: warning -- grant still in use by backend domain\n",
423 gnttab_end_foreign_access_ref(
424 queue->grant_tx_ref[id], GNTMAP_readonly);
425 gnttab_release_grant_reference(
426 &queue->gref_tx_head, queue->grant_tx_ref[id]);
427 queue->grant_tx_ref[id] = GRANT_INVALID_REF;
428 queue->grant_tx_page[id] = NULL;
429 add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, id);
430 dev_kfree_skb_irq(skb);
433 queue->tx.rsp_cons = prod;
436 * Set a new event, then check for race with update of tx_cons.
437 * Note that it is essential to schedule a callback, no matter
438 * how few buffers are pending. Even if there is space in the
439 * transmit ring, higher layers may be blocked because too much
440 * data is outstanding: in such cases notification from Xen is
441 * likely to be the only kick that we'll get.
443 queue->tx.sring->rsp_event =
444 prod + ((queue->tx.sring->req_prod - prod) >> 1) + 1;
445 mb(); /* update shared area */
446 } while ((cons == prod) && (prod != queue->tx.sring->rsp_prod));
448 xennet_maybe_wake_tx(queue);
451 static void xennet_make_frags(struct sk_buff *skb, struct netfront_queue *queue,
452 struct xen_netif_tx_request *tx)
454 char *data = skb->data;
456 RING_IDX prod = queue->tx.req_prod_pvt;
457 int frags = skb_shinfo(skb)->nr_frags;
458 unsigned int offset = offset_in_page(data);
459 unsigned int len = skb_headlen(skb);
464 /* While the header overlaps a page boundary (including being
465 larger than a page), split it it into page-sized chunks. */
466 while (len > PAGE_SIZE - offset) {
467 tx->size = PAGE_SIZE - offset;
468 tx->flags |= XEN_NETTXF_more_data;
473 id = get_id_from_freelist(&queue->tx_skb_freelist, queue->tx_skbs);
474 queue->tx_skbs[id].skb = skb_get(skb);
475 tx = RING_GET_REQUEST(&queue->tx, prod++);
477 ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
478 BUG_ON((signed short)ref < 0);
480 mfn = virt_to_mfn(data);
481 gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id,
482 mfn, GNTMAP_readonly);
484 queue->grant_tx_page[id] = virt_to_page(data);
485 tx->gref = queue->grant_tx_ref[id] = ref;
491 /* Grant backend access to each skb fragment page. */
492 for (i = 0; i < frags; i++) {
493 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
494 struct page *page = skb_frag_page(frag);
496 len = skb_frag_size(frag);
497 offset = frag->page_offset;
499 /* Data must not cross a page boundary. */
500 BUG_ON(len + offset > PAGE_SIZE<<compound_order(page));
502 /* Skip unused frames from start of page */
503 page += offset >> PAGE_SHIFT;
504 offset &= ~PAGE_MASK;
509 BUG_ON(offset >= PAGE_SIZE);
511 bytes = PAGE_SIZE - offset;
515 tx->flags |= XEN_NETTXF_more_data;
517 id = get_id_from_freelist(&queue->tx_skb_freelist,
519 queue->tx_skbs[id].skb = skb_get(skb);
520 tx = RING_GET_REQUEST(&queue->tx, prod++);
522 ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
523 BUG_ON((signed short)ref < 0);
525 mfn = pfn_to_mfn(page_to_pfn(page));
526 gnttab_grant_foreign_access_ref(ref,
527 queue->info->xbdev->otherend_id,
528 mfn, GNTMAP_readonly);
530 queue->grant_tx_page[id] = page;
531 tx->gref = queue->grant_tx_ref[id] = ref;
540 if (offset == PAGE_SIZE && len) {
541 BUG_ON(!PageCompound(page));
548 queue->tx.req_prod_pvt = prod;
552 * Count how many ring slots are required to send the frags of this
553 * skb. Each frag might be a compound page.
555 static int xennet_count_skb_frag_slots(struct sk_buff *skb)
557 int i, frags = skb_shinfo(skb)->nr_frags;
560 for (i = 0; i < frags; i++) {
561 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
562 unsigned long size = skb_frag_size(frag);
563 unsigned long offset = frag->page_offset;
565 /* Skip unused frames from start of page */
566 offset &= ~PAGE_MASK;
568 pages += PFN_UP(offset + size);
574 static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb,
575 void *accel_priv, select_queue_fallback_t fallback)
577 unsigned int num_queues = dev->real_num_tx_queues;
581 /* First, check if there is only one queue */
582 if (num_queues == 1) {
585 hash = skb_get_hash(skb);
586 queue_idx = hash % num_queues;
592 static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
595 struct netfront_info *np = netdev_priv(dev);
596 struct netfront_stats *stats = this_cpu_ptr(np->stats);
597 struct xen_netif_tx_request *tx;
598 char *data = skb->data;
604 unsigned int offset = offset_in_page(data);
605 unsigned int len = skb_headlen(skb);
607 struct netfront_queue *queue = NULL;
608 unsigned int num_queues = dev->real_num_tx_queues;
611 /* Drop the packet if no queues are set up */
614 /* Determine which queue to transmit this SKB on */
615 queue_index = skb_get_queue_mapping(skb);
616 queue = &np->queues[queue_index];
618 /* If skb->len is too big for wire format, drop skb and alert
619 * user about misconfiguration.
621 if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
622 net_alert_ratelimited(
623 "xennet: skb->len = %u, too big for wire format\n",
628 slots = DIV_ROUND_UP(offset + len, PAGE_SIZE) +
629 xennet_count_skb_frag_slots(skb);
630 if (unlikely(slots > MAX_SKB_FRAGS + 1)) {
631 net_alert_ratelimited(
632 "xennet: skb rides the rocket: %d slots\n", slots);
636 spin_lock_irqsave(&queue->tx_lock, flags);
638 if (unlikely(!netif_carrier_ok(dev) ||
639 (slots > 1 && !xennet_can_sg(dev)) ||
640 netif_needs_gso(skb, netif_skb_features(skb)))) {
641 spin_unlock_irqrestore(&queue->tx_lock, flags);
645 i = queue->tx.req_prod_pvt;
647 id = get_id_from_freelist(&queue->tx_skb_freelist, queue->tx_skbs);
648 queue->tx_skbs[id].skb = skb;
650 tx = RING_GET_REQUEST(&queue->tx, i);
653 ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
654 BUG_ON((signed short)ref < 0);
655 mfn = virt_to_mfn(data);
656 gnttab_grant_foreign_access_ref(
657 ref, queue->info->xbdev->otherend_id, mfn, GNTMAP_readonly);
658 queue->grant_tx_page[id] = virt_to_page(data);
659 tx->gref = queue->grant_tx_ref[id] = ref;
664 if (skb->ip_summed == CHECKSUM_PARTIAL)
666 tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
667 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
668 /* remote but checksummed. */
669 tx->flags |= XEN_NETTXF_data_validated;
671 if (skb_shinfo(skb)->gso_size) {
672 struct xen_netif_extra_info *gso;
674 gso = (struct xen_netif_extra_info *)
675 RING_GET_REQUEST(&queue->tx, ++i);
677 tx->flags |= XEN_NETTXF_extra_info;
679 gso->u.gso.size = skb_shinfo(skb)->gso_size;
680 gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ?
681 XEN_NETIF_GSO_TYPE_TCPV6 :
682 XEN_NETIF_GSO_TYPE_TCPV4;
684 gso->u.gso.features = 0;
686 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
690 queue->tx.req_prod_pvt = i + 1;
692 xennet_make_frags(skb, queue, tx);
695 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
697 notify_remote_via_irq(queue->tx_irq);
699 u64_stats_update_begin(&stats->syncp);
700 stats->tx_bytes += skb->len;
702 u64_stats_update_end(&stats->syncp);
704 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
705 xennet_tx_buf_gc(queue);
707 if (!netfront_tx_slot_available(queue))
708 netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
710 spin_unlock_irqrestore(&queue->tx_lock, flags);
715 dev->stats.tx_dropped++;
716 dev_kfree_skb_any(skb);
720 static int xennet_close(struct net_device *dev)
722 struct netfront_info *np = netdev_priv(dev);
723 unsigned int num_queues = dev->real_num_tx_queues;
725 struct netfront_queue *queue;
726 netif_tx_stop_all_queues(np->netdev);
727 for (i = 0; i < num_queues; ++i) {
728 queue = &np->queues[i];
729 napi_disable(&queue->napi);
734 static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb,
737 int new = xennet_rxidx(queue->rx.req_prod_pvt);
739 BUG_ON(queue->rx_skbs[new]);
740 queue->rx_skbs[new] = skb;
741 queue->grant_rx_ref[new] = ref;
742 RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new;
743 RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref;
744 queue->rx.req_prod_pvt++;
747 static int xennet_get_extras(struct netfront_queue *queue,
748 struct xen_netif_extra_info *extras,
752 struct xen_netif_extra_info *extra;
753 struct device *dev = &queue->info->netdev->dev;
754 RING_IDX cons = queue->rx.rsp_cons;
761 if (unlikely(cons + 1 == rp)) {
763 dev_warn(dev, "Missing extra info\n");
768 extra = (struct xen_netif_extra_info *)
769 RING_GET_RESPONSE(&queue->rx, ++cons);
771 if (unlikely(!extra->type ||
772 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
774 dev_warn(dev, "Invalid extra type: %d\n",
778 memcpy(&extras[extra->type - 1], extra,
782 skb = xennet_get_rx_skb(queue, cons);
783 ref = xennet_get_rx_ref(queue, cons);
784 xennet_move_rx_slot(queue, skb, ref);
785 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
787 queue->rx.rsp_cons = cons;
791 static int xennet_get_responses(struct netfront_queue *queue,
792 struct netfront_rx_info *rinfo, RING_IDX rp,
793 struct sk_buff_head *list)
795 struct xen_netif_rx_response *rx = &rinfo->rx;
796 struct xen_netif_extra_info *extras = rinfo->extras;
797 struct device *dev = &queue->info->netdev->dev;
798 RING_IDX cons = queue->rx.rsp_cons;
799 struct sk_buff *skb = xennet_get_rx_skb(queue, cons);
800 grant_ref_t ref = xennet_get_rx_ref(queue, cons);
801 int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD);
806 if (rx->flags & XEN_NETRXF_extra_info) {
807 err = xennet_get_extras(queue, extras, rp);
808 cons = queue->rx.rsp_cons;
812 if (unlikely(rx->status < 0 ||
813 rx->offset + rx->status > PAGE_SIZE)) {
815 dev_warn(dev, "rx->offset: %x, size: %u\n",
816 rx->offset, rx->status);
817 xennet_move_rx_slot(queue, skb, ref);
823 * This definitely indicates a bug, either in this driver or in
824 * the backend driver. In future this should flag the bad
825 * situation to the system controller to reboot the backend.
827 if (ref == GRANT_INVALID_REF) {
829 dev_warn(dev, "Bad rx response id %d.\n",
835 ret = gnttab_end_foreign_access_ref(ref, 0);
838 gnttab_release_grant_reference(&queue->gref_rx_head, ref);
840 __skb_queue_tail(list, skb);
843 if (!(rx->flags & XEN_NETRXF_more_data))
846 if (cons + slots == rp) {
848 dev_warn(dev, "Need more slots\n");
853 rx = RING_GET_RESPONSE(&queue->rx, cons + slots);
854 skb = xennet_get_rx_skb(queue, cons + slots);
855 ref = xennet_get_rx_ref(queue, cons + slots);
859 if (unlikely(slots > max)) {
861 dev_warn(dev, "Too many slots\n");
866 queue->rx.rsp_cons = cons + slots;
871 static int xennet_set_skb_gso(struct sk_buff *skb,
872 struct xen_netif_extra_info *gso)
874 if (!gso->u.gso.size) {
876 pr_warn("GSO size must not be zero\n");
880 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 &&
881 gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) {
883 pr_warn("Bad GSO type %d\n", gso->u.gso.type);
887 skb_shinfo(skb)->gso_size = gso->u.gso.size;
888 skb_shinfo(skb)->gso_type =
889 (gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ?
893 /* Header must be checked, and gso_segs computed. */
894 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
895 skb_shinfo(skb)->gso_segs = 0;
900 static RING_IDX xennet_fill_frags(struct netfront_queue *queue,
902 struct sk_buff_head *list)
904 struct skb_shared_info *shinfo = skb_shinfo(skb);
905 RING_IDX cons = queue->rx.rsp_cons;
906 struct sk_buff *nskb;
908 while ((nskb = __skb_dequeue(list))) {
909 struct xen_netif_rx_response *rx =
910 RING_GET_RESPONSE(&queue->rx, ++cons);
911 skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
913 if (shinfo->nr_frags == MAX_SKB_FRAGS) {
914 unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
916 BUG_ON(pull_to <= skb_headlen(skb));
917 __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
919 BUG_ON(shinfo->nr_frags >= MAX_SKB_FRAGS);
921 skb_add_rx_frag(skb, shinfo->nr_frags, skb_frag_page(nfrag),
922 rx->offset, rx->status, PAGE_SIZE);
924 skb_shinfo(nskb)->nr_frags = 0;
931 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
933 bool recalculate_partial_csum = false;
936 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
937 * peers can fail to set NETRXF_csum_blank when sending a GSO
938 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
939 * recalculate the partial checksum.
941 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
942 struct netfront_info *np = netdev_priv(dev);
943 atomic_inc(&np->rx_gso_checksum_fixup);
944 skb->ip_summed = CHECKSUM_PARTIAL;
945 recalculate_partial_csum = true;
948 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
949 if (skb->ip_summed != CHECKSUM_PARTIAL)
952 return skb_checksum_setup(skb, recalculate_partial_csum);
955 static int handle_incoming_queue(struct netfront_queue *queue,
956 struct sk_buff_head *rxq)
958 struct netfront_stats *stats = this_cpu_ptr(queue->info->stats);
959 int packets_dropped = 0;
962 while ((skb = __skb_dequeue(rxq)) != NULL) {
963 int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
965 if (pull_to > skb_headlen(skb))
966 __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
968 /* Ethernet work: Delayed to here as it peeks the header. */
969 skb->protocol = eth_type_trans(skb, queue->info->netdev);
970 skb_reset_network_header(skb);
972 if (checksum_setup(queue->info->netdev, skb)) {
975 queue->info->netdev->stats.rx_errors++;
979 u64_stats_update_begin(&stats->syncp);
981 stats->rx_bytes += skb->len;
982 u64_stats_update_end(&stats->syncp);
985 napi_gro_receive(&queue->napi, skb);
988 return packets_dropped;
991 static int xennet_poll(struct napi_struct *napi, int budget)
993 struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi);
994 struct net_device *dev = queue->info->netdev;
996 struct netfront_rx_info rinfo;
997 struct xen_netif_rx_response *rx = &rinfo.rx;
998 struct xen_netif_extra_info *extras = rinfo.extras;
1001 struct sk_buff_head rxq;
1002 struct sk_buff_head errq;
1003 struct sk_buff_head tmpq;
1004 unsigned long flags;
1007 spin_lock(&queue->rx_lock);
1009 skb_queue_head_init(&rxq);
1010 skb_queue_head_init(&errq);
1011 skb_queue_head_init(&tmpq);
1013 rp = queue->rx.sring->rsp_prod;
1014 rmb(); /* Ensure we see queued responses up to 'rp'. */
1016 i = queue->rx.rsp_cons;
1018 while ((i != rp) && (work_done < budget)) {
1019 memcpy(rx, RING_GET_RESPONSE(&queue->rx, i), sizeof(*rx));
1020 memset(extras, 0, sizeof(rinfo.extras));
1022 err = xennet_get_responses(queue, &rinfo, rp, &tmpq);
1024 if (unlikely(err)) {
1026 while ((skb = __skb_dequeue(&tmpq)))
1027 __skb_queue_tail(&errq, skb);
1028 dev->stats.rx_errors++;
1029 i = queue->rx.rsp_cons;
1033 skb = __skb_dequeue(&tmpq);
1035 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1036 struct xen_netif_extra_info *gso;
1037 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1039 if (unlikely(xennet_set_skb_gso(skb, gso))) {
1040 __skb_queue_head(&tmpq, skb);
1041 queue->rx.rsp_cons += skb_queue_len(&tmpq);
1046 NETFRONT_SKB_CB(skb)->pull_to = rx->status;
1047 if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
1048 NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
1050 skb_shinfo(skb)->frags[0].page_offset = rx->offset;
1051 skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
1052 skb->data_len = rx->status;
1053 skb->len += rx->status;
1055 i = xennet_fill_frags(queue, skb, &tmpq);
1057 if (rx->flags & XEN_NETRXF_csum_blank)
1058 skb->ip_summed = CHECKSUM_PARTIAL;
1059 else if (rx->flags & XEN_NETRXF_data_validated)
1060 skb->ip_summed = CHECKSUM_UNNECESSARY;
1062 __skb_queue_tail(&rxq, skb);
1064 queue->rx.rsp_cons = ++i;
1068 __skb_queue_purge(&errq);
1070 work_done -= handle_incoming_queue(queue, &rxq);
1072 /* If we get a callback with very few responses, reduce fill target. */
1073 /* NB. Note exponential increase, linear decrease. */
1074 if (((queue->rx.req_prod_pvt - queue->rx.sring->rsp_prod) >
1075 ((3*queue->rx_target) / 4)) &&
1076 (--queue->rx_target < queue->rx_min_target))
1077 queue->rx_target = queue->rx_min_target;
1079 xennet_alloc_rx_buffers(queue);
1081 if (work_done < budget) {
1084 napi_gro_flush(napi, false);
1086 local_irq_save(flags);
1088 RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do);
1090 __napi_complete(napi);
1092 local_irq_restore(flags);
1095 spin_unlock(&queue->rx_lock);
1100 static int xennet_change_mtu(struct net_device *dev, int mtu)
1102 int max = xennet_can_sg(dev) ?
1103 XEN_NETIF_MAX_TX_SIZE - MAX_TCP_HEADER : ETH_DATA_LEN;
1111 static struct rtnl_link_stats64 *xennet_get_stats64(struct net_device *dev,
1112 struct rtnl_link_stats64 *tot)
1114 struct netfront_info *np = netdev_priv(dev);
1117 for_each_possible_cpu(cpu) {
1118 struct netfront_stats *stats = per_cpu_ptr(np->stats, cpu);
1119 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1123 start = u64_stats_fetch_begin_irq(&stats->syncp);
1125 rx_packets = stats->rx_packets;
1126 tx_packets = stats->tx_packets;
1127 rx_bytes = stats->rx_bytes;
1128 tx_bytes = stats->tx_bytes;
1129 } while (u64_stats_fetch_retry_irq(&stats->syncp, start));
1131 tot->rx_packets += rx_packets;
1132 tot->tx_packets += tx_packets;
1133 tot->rx_bytes += rx_bytes;
1134 tot->tx_bytes += tx_bytes;
1137 tot->rx_errors = dev->stats.rx_errors;
1138 tot->tx_dropped = dev->stats.tx_dropped;
1143 static void xennet_release_tx_bufs(struct netfront_queue *queue)
1145 struct sk_buff *skb;
1148 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1149 /* Skip over entries which are actually freelist references */
1150 if (skb_entry_is_link(&queue->tx_skbs[i]))
1153 skb = queue->tx_skbs[i].skb;
1154 get_page(queue->grant_tx_page[i]);
1155 gnttab_end_foreign_access(queue->grant_tx_ref[i],
1157 (unsigned long)page_address(queue->grant_tx_page[i]));
1158 queue->grant_tx_page[i] = NULL;
1159 queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1160 add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, i);
1161 dev_kfree_skb_irq(skb);
1165 static void xennet_release_rx_bufs(struct netfront_queue *queue)
1169 spin_lock_bh(&queue->rx_lock);
1171 for (id = 0; id < NET_RX_RING_SIZE; id++) {
1172 struct sk_buff *skb;
1175 skb = queue->rx_skbs[id];
1179 ref = queue->grant_rx_ref[id];
1180 if (ref == GRANT_INVALID_REF)
1183 page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
1185 /* gnttab_end_foreign_access() needs a page ref until
1186 * foreign access is ended (which may be deferred).
1189 gnttab_end_foreign_access(ref, 0,
1190 (unsigned long)page_address(page));
1191 queue->grant_rx_ref[id] = GRANT_INVALID_REF;
1196 spin_unlock_bh(&queue->rx_lock);
1199 static netdev_features_t xennet_fix_features(struct net_device *dev,
1200 netdev_features_t features)
1202 struct netfront_info *np = netdev_priv(dev);
1205 if (features & NETIF_F_SG) {
1206 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg",
1211 features &= ~NETIF_F_SG;
1214 if (features & NETIF_F_IPV6_CSUM) {
1215 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1216 "feature-ipv6-csum-offload", "%d", &val) < 0)
1220 features &= ~NETIF_F_IPV6_CSUM;
1223 if (features & NETIF_F_TSO) {
1224 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1225 "feature-gso-tcpv4", "%d", &val) < 0)
1229 features &= ~NETIF_F_TSO;
1232 if (features & NETIF_F_TSO6) {
1233 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1234 "feature-gso-tcpv6", "%d", &val) < 0)
1238 features &= ~NETIF_F_TSO6;
1244 static int xennet_set_features(struct net_device *dev,
1245 netdev_features_t features)
1247 if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1248 netdev_info(dev, "Reducing MTU because no SG offload");
1249 dev->mtu = ETH_DATA_LEN;
1255 static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1257 struct netfront_queue *queue = dev_id;
1258 unsigned long flags;
1260 spin_lock_irqsave(&queue->tx_lock, flags);
1261 xennet_tx_buf_gc(queue);
1262 spin_unlock_irqrestore(&queue->tx_lock, flags);
1267 static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1269 struct netfront_queue *queue = dev_id;
1270 struct net_device *dev = queue->info->netdev;
1272 if (likely(netif_carrier_ok(dev) &&
1273 RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)))
1274 napi_schedule(&queue->napi);
1279 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1281 xennet_tx_interrupt(irq, dev_id);
1282 xennet_rx_interrupt(irq, dev_id);
1286 #ifdef CONFIG_NET_POLL_CONTROLLER
1287 static void xennet_poll_controller(struct net_device *dev)
1289 /* Poll each queue */
1290 struct netfront_info *info = netdev_priv(dev);
1291 unsigned int num_queues = dev->real_num_tx_queues;
1293 for (i = 0; i < num_queues; ++i)
1294 xennet_interrupt(0, &info->queues[i]);
1298 static const struct net_device_ops xennet_netdev_ops = {
1299 .ndo_open = xennet_open,
1300 .ndo_stop = xennet_close,
1301 .ndo_start_xmit = xennet_start_xmit,
1302 .ndo_change_mtu = xennet_change_mtu,
1303 .ndo_get_stats64 = xennet_get_stats64,
1304 .ndo_set_mac_address = eth_mac_addr,
1305 .ndo_validate_addr = eth_validate_addr,
1306 .ndo_fix_features = xennet_fix_features,
1307 .ndo_set_features = xennet_set_features,
1308 .ndo_select_queue = xennet_select_queue,
1309 #ifdef CONFIG_NET_POLL_CONTROLLER
1310 .ndo_poll_controller = xennet_poll_controller,
1314 static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1317 struct net_device *netdev;
1318 struct netfront_info *np;
1320 netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues);
1322 return ERR_PTR(-ENOMEM);
1324 np = netdev_priv(netdev);
1327 /* No need to use rtnl_lock() before the call below as it
1328 * happens before register_netdev().
1330 netif_set_real_num_tx_queues(netdev, 0);
1334 np->stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1335 if (np->stats == NULL)
1338 netdev->netdev_ops = &xennet_netdev_ops;
1340 netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1342 netdev->hw_features = NETIF_F_SG |
1344 NETIF_F_TSO | NETIF_F_TSO6;
1347 * Assume that all hw features are available for now. This set
1348 * will be adjusted by the call to netdev_update_features() in
1349 * xennet_connect() which is the earliest point where we can
1350 * negotiate with the backend regarding supported features.
1352 netdev->features |= netdev->hw_features;
1354 netdev->ethtool_ops = &xennet_ethtool_ops;
1355 SET_NETDEV_DEV(netdev, &dev->dev);
1357 netif_set_gso_max_size(netdev, XEN_NETIF_MAX_TX_SIZE - MAX_TCP_HEADER);
1359 np->netdev = netdev;
1361 netif_carrier_off(netdev);
1366 free_netdev(netdev);
1367 return ERR_PTR(err);
1371 * Entry point to this code when a new device is created. Allocate the basic
1372 * structures and the ring buffers for communication with the backend, and
1373 * inform the backend of the appropriate details for those.
1375 static int netfront_probe(struct xenbus_device *dev,
1376 const struct xenbus_device_id *id)
1379 struct net_device *netdev;
1380 struct netfront_info *info;
1382 netdev = xennet_create_dev(dev);
1383 if (IS_ERR(netdev)) {
1384 err = PTR_ERR(netdev);
1385 xenbus_dev_fatal(dev, err, "creating netdev");
1389 info = netdev_priv(netdev);
1390 dev_set_drvdata(&dev->dev, info);
1392 err = register_netdev(info->netdev);
1394 pr_warn("%s: register_netdev err=%d\n", __func__, err);
1398 err = xennet_sysfs_addif(info->netdev);
1400 unregister_netdev(info->netdev);
1401 pr_warn("%s: add sysfs failed err=%d\n", __func__, err);
1408 free_netdev(netdev);
1409 dev_set_drvdata(&dev->dev, NULL);
1413 static void xennet_end_access(int ref, void *page)
1415 /* This frees the page as a side-effect */
1416 if (ref != GRANT_INVALID_REF)
1417 gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1420 static void xennet_disconnect_backend(struct netfront_info *info)
1423 unsigned int num_queues = info->netdev->real_num_tx_queues;
1425 netif_carrier_off(info->netdev);
1427 for (i = 0; i < num_queues; ++i) {
1428 struct netfront_queue *queue = &info->queues[i];
1430 if (queue->tx_irq && (queue->tx_irq == queue->rx_irq))
1431 unbind_from_irqhandler(queue->tx_irq, queue);
1432 if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) {
1433 unbind_from_irqhandler(queue->tx_irq, queue);
1434 unbind_from_irqhandler(queue->rx_irq, queue);
1436 queue->tx_evtchn = queue->rx_evtchn = 0;
1437 queue->tx_irq = queue->rx_irq = 0;
1439 napi_synchronize(&queue->napi);
1441 xennet_release_tx_bufs(queue);
1442 xennet_release_rx_bufs(queue);
1443 gnttab_free_grant_references(queue->gref_tx_head);
1444 gnttab_free_grant_references(queue->gref_rx_head);
1446 /* End access and free the pages */
1447 xennet_end_access(queue->tx_ring_ref, queue->tx.sring);
1448 xennet_end_access(queue->rx_ring_ref, queue->rx.sring);
1450 queue->tx_ring_ref = GRANT_INVALID_REF;
1451 queue->rx_ring_ref = GRANT_INVALID_REF;
1452 queue->tx.sring = NULL;
1453 queue->rx.sring = NULL;
1458 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1459 * driver restart. We tear down our netif structure and recreate it, but
1460 * leave the device-layer structures intact so that this is transparent to the
1461 * rest of the kernel.
1463 static int netfront_resume(struct xenbus_device *dev)
1465 struct netfront_info *info = dev_get_drvdata(&dev->dev);
1467 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1469 xennet_disconnect_backend(info);
1473 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1475 char *s, *e, *macstr;
1478 macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1480 return PTR_ERR(macstr);
1482 for (i = 0; i < ETH_ALEN; i++) {
1483 mac[i] = simple_strtoul(s, &e, 16);
1484 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1495 static int setup_netfront_single(struct netfront_queue *queue)
1499 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1503 err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1505 0, queue->info->netdev->name, queue);
1508 queue->rx_evtchn = queue->tx_evtchn;
1509 queue->rx_irq = queue->tx_irq = err;
1514 xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1515 queue->tx_evtchn = 0;
1520 static int setup_netfront_split(struct netfront_queue *queue)
1524 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1527 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn);
1529 goto alloc_rx_evtchn_fail;
1531 snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
1532 "%s-tx", queue->name);
1533 err = bind_evtchn_to_irqhandler(queue->tx_evtchn,
1534 xennet_tx_interrupt,
1535 0, queue->tx_irq_name, queue);
1538 queue->tx_irq = err;
1540 snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
1541 "%s-rx", queue->name);
1542 err = bind_evtchn_to_irqhandler(queue->rx_evtchn,
1543 xennet_rx_interrupt,
1544 0, queue->rx_irq_name, queue);
1547 queue->rx_irq = err;
1552 unbind_from_irqhandler(queue->tx_irq, queue);
1555 xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn);
1556 queue->rx_evtchn = 0;
1557 alloc_rx_evtchn_fail:
1558 xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1559 queue->tx_evtchn = 0;
1564 static int setup_netfront(struct xenbus_device *dev,
1565 struct netfront_queue *queue, unsigned int feature_split_evtchn)
1567 struct xen_netif_tx_sring *txs;
1568 struct xen_netif_rx_sring *rxs;
1571 queue->tx_ring_ref = GRANT_INVALID_REF;
1572 queue->rx_ring_ref = GRANT_INVALID_REF;
1573 queue->rx.sring = NULL;
1574 queue->tx.sring = NULL;
1576 txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1579 xenbus_dev_fatal(dev, err, "allocating tx ring page");
1582 SHARED_RING_INIT(txs);
1583 FRONT_RING_INIT(&queue->tx, txs, PAGE_SIZE);
1585 err = xenbus_grant_ring(dev, virt_to_mfn(txs));
1587 goto grant_tx_ring_fail;
1588 queue->tx_ring_ref = err;
1590 rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1593 xenbus_dev_fatal(dev, err, "allocating rx ring page");
1594 goto alloc_rx_ring_fail;
1596 SHARED_RING_INIT(rxs);
1597 FRONT_RING_INIT(&queue->rx, rxs, PAGE_SIZE);
1599 err = xenbus_grant_ring(dev, virt_to_mfn(rxs));
1601 goto grant_rx_ring_fail;
1602 queue->rx_ring_ref = err;
1604 if (feature_split_evtchn)
1605 err = setup_netfront_split(queue);
1606 /* setup single event channel if
1607 * a) feature-split-event-channels == 0
1608 * b) feature-split-event-channels == 1 but failed to setup
1610 if (!feature_split_evtchn || (feature_split_evtchn && err))
1611 err = setup_netfront_single(queue);
1614 goto alloc_evtchn_fail;
1618 /* If we fail to setup netfront, it is safe to just revoke access to
1619 * granted pages because backend is not accessing it at this point.
1622 gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0);
1624 free_page((unsigned long)rxs);
1626 gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0);
1628 free_page((unsigned long)txs);
1633 /* Queue-specific initialisation
1634 * This used to be done in xennet_create_dev() but must now
1637 static int xennet_init_queue(struct netfront_queue *queue)
1642 spin_lock_init(&queue->tx_lock);
1643 spin_lock_init(&queue->rx_lock);
1645 skb_queue_head_init(&queue->rx_batch);
1646 queue->rx_target = RX_DFL_MIN_TARGET;
1647 queue->rx_min_target = RX_DFL_MIN_TARGET;
1648 queue->rx_max_target = RX_MAX_TARGET;
1650 init_timer(&queue->rx_refill_timer);
1651 queue->rx_refill_timer.data = (unsigned long)queue;
1652 queue->rx_refill_timer.function = rx_refill_timeout;
1654 snprintf(queue->name, sizeof(queue->name), "%s-q%u",
1655 queue->info->netdev->name, queue->id);
1657 /* Initialise tx_skbs as a free chain containing every entry. */
1658 queue->tx_skb_freelist = 0;
1659 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1660 skb_entry_set_link(&queue->tx_skbs[i], i+1);
1661 queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1662 queue->grant_tx_page[i] = NULL;
1665 /* Clear out rx_skbs */
1666 for (i = 0; i < NET_RX_RING_SIZE; i++) {
1667 queue->rx_skbs[i] = NULL;
1668 queue->grant_rx_ref[i] = GRANT_INVALID_REF;
1671 /* A grant for every tx ring slot */
1672 if (gnttab_alloc_grant_references(TX_MAX_TARGET,
1673 &queue->gref_tx_head) < 0) {
1674 pr_alert("can't alloc tx grant refs\n");
1679 /* A grant for every rx ring slot */
1680 if (gnttab_alloc_grant_references(RX_MAX_TARGET,
1681 &queue->gref_rx_head) < 0) {
1682 pr_alert("can't alloc rx grant refs\n");
1690 gnttab_free_grant_references(queue->gref_tx_head);
1695 static int write_queue_xenstore_keys(struct netfront_queue *queue,
1696 struct xenbus_transaction *xbt, int write_hierarchical)
1698 /* Write the queue-specific keys into XenStore in the traditional
1699 * way for a single queue, or in a queue subkeys for multiple
1702 struct xenbus_device *dev = queue->info->xbdev;
1704 const char *message;
1708 /* Choose the correct place to write the keys */
1709 if (write_hierarchical) {
1710 pathsize = strlen(dev->nodename) + 10;
1711 path = kzalloc(pathsize, GFP_KERNEL);
1714 message = "out of memory while writing ring references";
1717 snprintf(path, pathsize, "%s/queue-%u",
1718 dev->nodename, queue->id);
1720 path = (char *)dev->nodename;
1723 /* Write ring references */
1724 err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u",
1725 queue->tx_ring_ref);
1727 message = "writing tx-ring-ref";
1731 err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u",
1732 queue->rx_ring_ref);
1734 message = "writing rx-ring-ref";
1738 /* Write event channels; taking into account both shared
1739 * and split event channel scenarios.
1741 if (queue->tx_evtchn == queue->rx_evtchn) {
1742 /* Shared event channel */
1743 err = xenbus_printf(*xbt, path,
1744 "event-channel", "%u", queue->tx_evtchn);
1746 message = "writing event-channel";
1750 /* Split event channels */
1751 err = xenbus_printf(*xbt, path,
1752 "event-channel-tx", "%u", queue->tx_evtchn);
1754 message = "writing event-channel-tx";
1758 err = xenbus_printf(*xbt, path,
1759 "event-channel-rx", "%u", queue->rx_evtchn);
1761 message = "writing event-channel-rx";
1766 if (write_hierarchical)
1771 if (write_hierarchical)
1773 xenbus_dev_fatal(dev, err, "%s", message);
1777 static void xennet_destroy_queues(struct netfront_info *info)
1783 for (i = 0; i < info->netdev->real_num_tx_queues; i++) {
1784 struct netfront_queue *queue = &info->queues[i];
1786 if (netif_running(info->netdev))
1787 napi_disable(&queue->napi);
1788 netif_napi_del(&queue->napi);
1793 kfree(info->queues);
1794 info->queues = NULL;
1797 static int xennet_create_queues(struct netfront_info *info,
1798 unsigned int num_queues)
1803 info->queues = kcalloc(num_queues, sizeof(struct netfront_queue),
1810 for (i = 0; i < num_queues; i++) {
1811 struct netfront_queue *queue = &info->queues[i];
1816 ret = xennet_init_queue(queue);
1818 dev_warn(&info->netdev->dev,
1819 "only created %d queues\n", i);
1824 netif_napi_add(queue->info->netdev, &queue->napi,
1826 if (netif_running(info->netdev))
1827 napi_enable(&queue->napi);
1830 netif_set_real_num_tx_queues(info->netdev, num_queues);
1834 if (num_queues == 0) {
1835 dev_err(&info->netdev->dev, "no queues\n");
1841 /* Common code used when first setting up, and when resuming. */
1842 static int talk_to_netback(struct xenbus_device *dev,
1843 struct netfront_info *info)
1845 const char *message;
1846 struct xenbus_transaction xbt;
1848 unsigned int feature_split_evtchn;
1850 unsigned int max_queues = 0;
1851 struct netfront_queue *queue = NULL;
1852 unsigned int num_queues = 1;
1854 info->netdev->irq = 0;
1856 /* Check if backend supports multiple queues */
1857 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1858 "multi-queue-max-queues", "%u", &max_queues);
1861 num_queues = min(max_queues, xennet_max_queues);
1863 /* Check feature-split-event-channels */
1864 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1865 "feature-split-event-channels", "%u",
1866 &feature_split_evtchn);
1868 feature_split_evtchn = 0;
1870 /* Read mac addr. */
1871 err = xen_net_read_mac(dev, info->netdev->dev_addr);
1873 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1878 xennet_destroy_queues(info);
1880 err = xennet_create_queues(info, num_queues);
1884 /* Create shared ring, alloc event channel -- for each queue */
1885 for (i = 0; i < num_queues; ++i) {
1886 queue = &info->queues[i];
1887 err = setup_netfront(dev, queue, feature_split_evtchn);
1889 /* setup_netfront() will tidy up the current
1890 * queue on error, but we need to clean up
1891 * those already allocated.
1895 netif_set_real_num_tx_queues(info->netdev, i);
1905 err = xenbus_transaction_start(&xbt);
1907 xenbus_dev_fatal(dev, err, "starting transaction");
1911 if (num_queues == 1) {
1912 err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */
1914 goto abort_transaction_no_dev_fatal;
1916 /* Write the number of queues */
1917 err = xenbus_printf(xbt, dev->nodename, "multi-queue-num-queues",
1920 message = "writing multi-queue-num-queues";
1921 goto abort_transaction_no_dev_fatal;
1924 /* Write the keys for each queue */
1925 for (i = 0; i < num_queues; ++i) {
1926 queue = &info->queues[i];
1927 err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */
1929 goto abort_transaction_no_dev_fatal;
1933 /* The remaining keys are not queue-specific */
1934 err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1937 message = "writing request-rx-copy";
1938 goto abort_transaction;
1941 err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1943 message = "writing feature-rx-notify";
1944 goto abort_transaction;
1947 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1949 message = "writing feature-sg";
1950 goto abort_transaction;
1953 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1955 message = "writing feature-gso-tcpv4";
1956 goto abort_transaction;
1959 err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1");
1961 message = "writing feature-gso-tcpv6";
1962 goto abort_transaction;
1965 err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload",
1968 message = "writing feature-ipv6-csum-offload";
1969 goto abort_transaction;
1972 err = xenbus_transaction_end(xbt, 0);
1976 xenbus_dev_fatal(dev, err, "completing transaction");
1983 xenbus_dev_fatal(dev, err, "%s", message);
1984 abort_transaction_no_dev_fatal:
1985 xenbus_transaction_end(xbt, 1);
1987 xennet_disconnect_backend(info);
1988 kfree(info->queues);
1989 info->queues = NULL;
1991 netif_set_real_num_tx_queues(info->netdev, 0);
1997 static int xennet_connect(struct net_device *dev)
1999 struct netfront_info *np = netdev_priv(dev);
2000 unsigned int num_queues = 0;
2002 unsigned int feature_rx_copy;
2004 struct netfront_queue *queue = NULL;
2006 err = xenbus_scanf(XBT_NIL, np->xbdev->otherend,
2007 "feature-rx-copy", "%u", &feature_rx_copy);
2009 feature_rx_copy = 0;
2011 if (!feature_rx_copy) {
2013 "backend does not support copying receive path\n");
2017 err = talk_to_netback(np->xbdev, np);
2021 /* talk_to_netback() sets the correct number of queues */
2022 num_queues = dev->real_num_tx_queues;
2025 netdev_update_features(dev);
2029 * All public and private state should now be sane. Get
2030 * ready to start sending and receiving packets and give the driver
2031 * domain a kick because we've probably just requeued some
2034 netif_carrier_on(np->netdev);
2035 for (j = 0; j < num_queues; ++j) {
2036 queue = &np->queues[j];
2038 notify_remote_via_irq(queue->tx_irq);
2039 if (queue->tx_irq != queue->rx_irq)
2040 notify_remote_via_irq(queue->rx_irq);
2042 spin_lock_irq(&queue->tx_lock);
2043 xennet_tx_buf_gc(queue);
2044 spin_unlock_irq(&queue->tx_lock);
2046 spin_lock_bh(&queue->rx_lock);
2047 xennet_alloc_rx_buffers(queue);
2048 spin_unlock_bh(&queue->rx_lock);
2055 * Callback received when the backend's state changes.
2057 static void netback_changed(struct xenbus_device *dev,
2058 enum xenbus_state backend_state)
2060 struct netfront_info *np = dev_get_drvdata(&dev->dev);
2061 struct net_device *netdev = np->netdev;
2063 dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
2065 switch (backend_state) {
2066 case XenbusStateInitialising:
2067 case XenbusStateInitialised:
2068 case XenbusStateReconfiguring:
2069 case XenbusStateReconfigured:
2070 case XenbusStateUnknown:
2073 case XenbusStateInitWait:
2074 if (dev->state != XenbusStateInitialising)
2076 if (xennet_connect(netdev) != 0)
2078 xenbus_switch_state(dev, XenbusStateConnected);
2081 case XenbusStateConnected:
2082 netdev_notify_peers(netdev);
2085 case XenbusStateClosed:
2086 if (dev->state == XenbusStateClosed)
2088 /* Missed the backend's CLOSING state -- fallthrough */
2089 case XenbusStateClosing:
2090 xenbus_frontend_closed(dev);
2095 static const struct xennet_stat {
2096 char name[ETH_GSTRING_LEN];
2098 } xennet_stats[] = {
2100 "rx_gso_checksum_fixup",
2101 offsetof(struct netfront_info, rx_gso_checksum_fixup)
2105 static int xennet_get_sset_count(struct net_device *dev, int string_set)
2107 switch (string_set) {
2109 return ARRAY_SIZE(xennet_stats);
2115 static void xennet_get_ethtool_stats(struct net_device *dev,
2116 struct ethtool_stats *stats, u64 * data)
2118 void *np = netdev_priv(dev);
2121 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2122 data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset));
2125 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
2129 switch (stringset) {
2131 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2132 memcpy(data + i * ETH_GSTRING_LEN,
2133 xennet_stats[i].name, ETH_GSTRING_LEN);
2138 static const struct ethtool_ops xennet_ethtool_ops =
2140 .get_link = ethtool_op_get_link,
2142 .get_sset_count = xennet_get_sset_count,
2143 .get_ethtool_stats = xennet_get_ethtool_stats,
2144 .get_strings = xennet_get_strings,
2148 static ssize_t show_rxbuf_min(struct device *dev,
2149 struct device_attribute *attr, char *buf)
2151 struct net_device *netdev = to_net_dev(dev);
2152 struct netfront_info *info = netdev_priv(netdev);
2153 unsigned int num_queues = netdev->real_num_tx_queues;
2156 return sprintf(buf, "%u\n", info->queues[0].rx_min_target);
2158 return sprintf(buf, "%u\n", RX_MIN_TARGET);
2161 static ssize_t store_rxbuf_min(struct device *dev,
2162 struct device_attribute *attr,
2163 const char *buf, size_t len)
2165 struct net_device *netdev = to_net_dev(dev);
2166 struct netfront_info *np = netdev_priv(netdev);
2167 unsigned int num_queues = netdev->real_num_tx_queues;
2169 unsigned long target;
2171 struct netfront_queue *queue;
2173 if (!capable(CAP_NET_ADMIN))
2176 target = simple_strtoul(buf, &endp, 0);
2180 if (target < RX_MIN_TARGET)
2181 target = RX_MIN_TARGET;
2182 if (target > RX_MAX_TARGET)
2183 target = RX_MAX_TARGET;
2185 for (i = 0; i < num_queues; ++i) {
2186 queue = &np->queues[i];
2187 spin_lock_bh(&queue->rx_lock);
2188 if (target > queue->rx_max_target)
2189 queue->rx_max_target = target;
2190 queue->rx_min_target = target;
2191 if (target > queue->rx_target)
2192 queue->rx_target = target;
2194 xennet_alloc_rx_buffers(queue);
2196 spin_unlock_bh(&queue->rx_lock);
2201 static ssize_t show_rxbuf_max(struct device *dev,
2202 struct device_attribute *attr, char *buf)
2204 struct net_device *netdev = to_net_dev(dev);
2205 struct netfront_info *info = netdev_priv(netdev);
2206 unsigned int num_queues = netdev->real_num_tx_queues;
2209 return sprintf(buf, "%u\n", info->queues[0].rx_max_target);
2211 return sprintf(buf, "%u\n", RX_MAX_TARGET);
2214 static ssize_t store_rxbuf_max(struct device *dev,
2215 struct device_attribute *attr,
2216 const char *buf, size_t len)
2218 struct net_device *netdev = to_net_dev(dev);
2219 struct netfront_info *np = netdev_priv(netdev);
2220 unsigned int num_queues = netdev->real_num_tx_queues;
2222 unsigned long target;
2224 struct netfront_queue *queue = NULL;
2226 if (!capable(CAP_NET_ADMIN))
2229 target = simple_strtoul(buf, &endp, 0);
2233 if (target < RX_MIN_TARGET)
2234 target = RX_MIN_TARGET;
2235 if (target > RX_MAX_TARGET)
2236 target = RX_MAX_TARGET;
2238 for (i = 0; i < num_queues; ++i) {
2239 queue = &np->queues[i];
2240 spin_lock_bh(&queue->rx_lock);
2241 if (target < queue->rx_min_target)
2242 queue->rx_min_target = target;
2243 queue->rx_max_target = target;
2244 if (target < queue->rx_target)
2245 queue->rx_target = target;
2247 xennet_alloc_rx_buffers(queue);
2249 spin_unlock_bh(&queue->rx_lock);
2254 static ssize_t show_rxbuf_cur(struct device *dev,
2255 struct device_attribute *attr, char *buf)
2257 struct net_device *netdev = to_net_dev(dev);
2258 struct netfront_info *info = netdev_priv(netdev);
2259 unsigned int num_queues = netdev->real_num_tx_queues;
2262 return sprintf(buf, "%u\n", info->queues[0].rx_target);
2264 return sprintf(buf, "0\n");
2267 static struct device_attribute xennet_attrs[] = {
2268 __ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf_min, store_rxbuf_min),
2269 __ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf_max, store_rxbuf_max),
2270 __ATTR(rxbuf_cur, S_IRUGO, show_rxbuf_cur, NULL),
2273 static int xennet_sysfs_addif(struct net_device *netdev)
2278 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) {
2279 err = device_create_file(&netdev->dev,
2288 device_remove_file(&netdev->dev, &xennet_attrs[i]);
2292 static void xennet_sysfs_delif(struct net_device *netdev)
2296 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++)
2297 device_remove_file(&netdev->dev, &xennet_attrs[i]);
2300 #endif /* CONFIG_SYSFS */
2302 static const struct xenbus_device_id netfront_ids[] = {
2308 static int xennet_remove(struct xenbus_device *dev)
2310 struct netfront_info *info = dev_get_drvdata(&dev->dev);
2311 unsigned int num_queues = info->netdev->real_num_tx_queues;
2312 struct netfront_queue *queue = NULL;
2315 dev_dbg(&dev->dev, "%s\n", dev->nodename);
2317 xennet_disconnect_backend(info);
2319 xennet_sysfs_delif(info->netdev);
2321 unregister_netdev(info->netdev);
2323 for (i = 0; i < num_queues; ++i) {
2324 queue = &info->queues[i];
2325 del_timer_sync(&queue->rx_refill_timer);
2329 kfree(info->queues);
2330 info->queues = NULL;
2333 free_percpu(info->stats);
2335 free_netdev(info->netdev);
2340 static DEFINE_XENBUS_DRIVER(netfront, ,
2341 .probe = netfront_probe,
2342 .remove = xennet_remove,
2343 .resume = netfront_resume,
2344 .otherend_changed = netback_changed,
2347 static int __init netif_init(void)
2352 if (!xen_has_pv_nic_devices())
2355 pr_info("Initialising Xen virtual ethernet driver\n");
2357 /* Allow as many queues as there are CPUs, by default */
2358 xennet_max_queues = num_online_cpus();
2360 return xenbus_register_frontend(&netfront_driver);
2362 module_init(netif_init);
2365 static void __exit netif_exit(void)
2367 xenbus_unregister_driver(&netfront_driver);
2369 module_exit(netif_exit);
2371 MODULE_DESCRIPTION("Xen virtual network device frontend");
2372 MODULE_LICENSE("GPL");
2373 MODULE_ALIAS("xen:vif");
2374 MODULE_ALIAS("xennet");