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 #include <linux/module.h>
33 #include <linux/kernel.h>
34 #include <linux/netdevice.h>
35 #include <linux/etherdevice.h>
36 #include <linux/skbuff.h>
37 #include <linux/ethtool.h>
38 #include <linux/if_ether.h>
40 #include <linux/udp.h>
41 #include <linux/moduleparam.h>
43 #include <linux/slab.h>
46 #include <asm/xen/page.h>
48 #include <xen/xenbus.h>
49 #include <xen/events.h>
51 #include <xen/platform_pci.h>
52 #include <xen/grant_table.h>
54 #include <xen/interface/io/netif.h>
55 #include <xen/interface/memory.h>
56 #include <xen/interface/grant_table.h>
58 static const struct ethtool_ops xennet_ethtool_ops;
64 #define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb))
66 #define RX_COPY_THRESHOLD 256
68 #define GRANT_INVALID_REF 0
70 #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, PAGE_SIZE)
71 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, PAGE_SIZE)
72 #define TX_MAX_TARGET min_t(int, NET_TX_RING_SIZE, 256)
74 struct netfront_stats {
79 struct u64_stats_sync syncp;
82 struct netfront_info {
83 struct list_head list;
84 struct net_device *netdev;
86 struct napi_struct napi;
88 /* Split event channels support, tx_* == rx_* when using
89 * single event channel.
91 unsigned int tx_evtchn, rx_evtchn;
92 unsigned int tx_irq, rx_irq;
93 /* Only used when split event channels support is enabled */
94 char tx_irq_name[IFNAMSIZ+4]; /* DEVNAME-tx */
95 char rx_irq_name[IFNAMSIZ+4]; /* DEVNAME-rx */
97 struct xenbus_device *xbdev;
100 struct xen_netif_tx_front_ring tx;
104 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
105 * are linked from tx_skb_freelist through skb_entry.link.
107 * NB. Freelist index entries are always going to be less than
108 * PAGE_OFFSET, whereas pointers to skbs will always be equal or
109 * greater than PAGE_OFFSET: we use this property to distinguish
115 } tx_skbs[NET_TX_RING_SIZE];
116 grant_ref_t gref_tx_head;
117 grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
118 unsigned tx_skb_freelist;
120 spinlock_t rx_lock ____cacheline_aligned_in_smp;
121 struct xen_netif_rx_front_ring rx;
124 /* Receive-ring batched refills. */
125 #define RX_MIN_TARGET 8
126 #define RX_DFL_MIN_TARGET 64
127 #define RX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256)
128 unsigned rx_min_target, rx_max_target, rx_target;
129 struct sk_buff_head rx_batch;
131 struct timer_list rx_refill_timer;
133 struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
134 grant_ref_t gref_rx_head;
135 grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
137 unsigned long rx_pfn_array[NET_RX_RING_SIZE];
138 struct multicall_entry rx_mcl[NET_RX_RING_SIZE+1];
139 struct mmu_update rx_mmu[NET_RX_RING_SIZE];
142 struct netfront_stats __percpu *stats;
144 unsigned long rx_gso_checksum_fixup;
147 struct netfront_rx_info {
148 struct xen_netif_rx_response rx;
149 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
152 static void skb_entry_set_link(union skb_entry *list, unsigned short id)
157 static int skb_entry_is_link(const union skb_entry *list)
159 BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link));
160 return (unsigned long)list->skb < PAGE_OFFSET;
164 * Access macros for acquiring freeing slots in tx_skbs[].
167 static void add_id_to_freelist(unsigned *head, union skb_entry *list,
170 skb_entry_set_link(&list[id], *head);
174 static unsigned short get_id_from_freelist(unsigned *head,
175 union skb_entry *list)
177 unsigned int id = *head;
178 *head = list[id].link;
182 static int xennet_rxidx(RING_IDX idx)
184 return idx & (NET_RX_RING_SIZE - 1);
187 static struct sk_buff *xennet_get_rx_skb(struct netfront_info *np,
190 int i = xennet_rxidx(ri);
191 struct sk_buff *skb = np->rx_skbs[i];
192 np->rx_skbs[i] = NULL;
196 static grant_ref_t xennet_get_rx_ref(struct netfront_info *np,
199 int i = xennet_rxidx(ri);
200 grant_ref_t ref = np->grant_rx_ref[i];
201 np->grant_rx_ref[i] = GRANT_INVALID_REF;
206 static int xennet_sysfs_addif(struct net_device *netdev);
207 static void xennet_sysfs_delif(struct net_device *netdev);
208 #else /* !CONFIG_SYSFS */
209 #define xennet_sysfs_addif(dev) (0)
210 #define xennet_sysfs_delif(dev) do { } while (0)
213 static bool xennet_can_sg(struct net_device *dev)
215 return dev->features & NETIF_F_SG;
219 static void rx_refill_timeout(unsigned long data)
221 struct net_device *dev = (struct net_device *)data;
222 struct netfront_info *np = netdev_priv(dev);
223 napi_schedule(&np->napi);
226 static int netfront_tx_slot_available(struct netfront_info *np)
228 return (np->tx.req_prod_pvt - np->tx.rsp_cons) <
229 (TX_MAX_TARGET - MAX_SKB_FRAGS - 2);
232 static void xennet_maybe_wake_tx(struct net_device *dev)
234 struct netfront_info *np = netdev_priv(dev);
236 if (unlikely(netif_queue_stopped(dev)) &&
237 netfront_tx_slot_available(np) &&
238 likely(netif_running(dev)))
239 netif_wake_queue(dev);
242 static void xennet_alloc_rx_buffers(struct net_device *dev)
245 struct netfront_info *np = netdev_priv(dev);
248 int i, batch_target, notify;
249 RING_IDX req_prod = np->rx.req_prod_pvt;
253 struct xen_netif_rx_request *req;
255 if (unlikely(!netif_carrier_ok(dev)))
259 * Allocate skbuffs greedily, even though we batch updates to the
260 * receive ring. This creates a less bursty demand on the memory
261 * allocator, so should reduce the chance of failed allocation requests
262 * both for ourself and for other kernel subsystems.
264 batch_target = np->rx_target - (req_prod - np->rx.rsp_cons);
265 for (i = skb_queue_len(&np->rx_batch); i < batch_target; i++) {
266 skb = __netdev_alloc_skb(dev, RX_COPY_THRESHOLD + NET_IP_ALIGN,
267 GFP_ATOMIC | __GFP_NOWARN);
271 /* Align ip header to a 16 bytes boundary */
272 skb_reserve(skb, NET_IP_ALIGN);
274 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
278 /* Any skbuffs queued for refill? Force them out. */
281 /* Could not allocate any skbuffs. Try again later. */
282 mod_timer(&np->rx_refill_timer,
287 __skb_fill_page_desc(skb, 0, page, 0, 0);
288 skb_shinfo(skb)->nr_frags = 1;
289 __skb_queue_tail(&np->rx_batch, skb);
292 /* Is the batch large enough to be worthwhile? */
293 if (i < (np->rx_target/2)) {
294 if (req_prod > np->rx.sring->req_prod)
299 /* Adjust our fill target if we risked running out of buffers. */
300 if (((req_prod - np->rx.sring->rsp_prod) < (np->rx_target / 4)) &&
301 ((np->rx_target *= 2) > np->rx_max_target))
302 np->rx_target = np->rx_max_target;
306 skb = __skb_dequeue(&np->rx_batch);
312 id = xennet_rxidx(req_prod + i);
314 BUG_ON(np->rx_skbs[id]);
315 np->rx_skbs[id] = skb;
317 ref = gnttab_claim_grant_reference(&np->gref_rx_head);
318 BUG_ON((signed short)ref < 0);
319 np->grant_rx_ref[id] = ref;
321 pfn = page_to_pfn(skb_frag_page(&skb_shinfo(skb)->frags[0]));
322 vaddr = page_address(skb_frag_page(&skb_shinfo(skb)->frags[0]));
324 req = RING_GET_REQUEST(&np->rx, req_prod + i);
325 gnttab_grant_foreign_access_ref(ref,
326 np->xbdev->otherend_id,
334 wmb(); /* barrier so backend seens requests */
336 /* Above is a suitable barrier to ensure backend will see requests. */
337 np->rx.req_prod_pvt = req_prod + i;
339 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->rx, notify);
341 notify_remote_via_irq(np->rx_irq);
344 static int xennet_open(struct net_device *dev)
346 struct netfront_info *np = netdev_priv(dev);
348 napi_enable(&np->napi);
350 spin_lock_bh(&np->rx_lock);
351 if (netif_carrier_ok(dev)) {
352 xennet_alloc_rx_buffers(dev);
353 np->rx.sring->rsp_event = np->rx.rsp_cons + 1;
354 if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx))
355 napi_schedule(&np->napi);
357 spin_unlock_bh(&np->rx_lock);
359 netif_start_queue(dev);
364 static void xennet_tx_buf_gc(struct net_device *dev)
368 struct netfront_info *np = netdev_priv(dev);
371 BUG_ON(!netif_carrier_ok(dev));
374 prod = np->tx.sring->rsp_prod;
375 rmb(); /* Ensure we see responses up to 'rp'. */
377 for (cons = np->tx.rsp_cons; cons != prod; cons++) {
378 struct xen_netif_tx_response *txrsp;
380 txrsp = RING_GET_RESPONSE(&np->tx, cons);
381 if (txrsp->status == XEN_NETIF_RSP_NULL)
385 skb = np->tx_skbs[id].skb;
386 if (unlikely(gnttab_query_foreign_access(
387 np->grant_tx_ref[id]) != 0)) {
388 printk(KERN_ALERT "xennet_tx_buf_gc: warning "
389 "-- grant still in use by backend "
393 gnttab_end_foreign_access_ref(
394 np->grant_tx_ref[id], GNTMAP_readonly);
395 gnttab_release_grant_reference(
396 &np->gref_tx_head, np->grant_tx_ref[id]);
397 np->grant_tx_ref[id] = GRANT_INVALID_REF;
398 add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, id);
399 dev_kfree_skb_irq(skb);
402 np->tx.rsp_cons = prod;
405 * Set a new event, then check for race with update of tx_cons.
406 * Note that it is essential to schedule a callback, no matter
407 * how few buffers are pending. Even if there is space in the
408 * transmit ring, higher layers may be blocked because too much
409 * data is outstanding: in such cases notification from Xen is
410 * likely to be the only kick that we'll get.
412 np->tx.sring->rsp_event =
413 prod + ((np->tx.sring->req_prod - prod) >> 1) + 1;
414 mb(); /* update shared area */
415 } while ((cons == prod) && (prod != np->tx.sring->rsp_prod));
417 xennet_maybe_wake_tx(dev);
420 static void xennet_make_frags(struct sk_buff *skb, struct net_device *dev,
421 struct xen_netif_tx_request *tx)
423 struct netfront_info *np = netdev_priv(dev);
424 char *data = skb->data;
426 RING_IDX prod = np->tx.req_prod_pvt;
427 int frags = skb_shinfo(skb)->nr_frags;
428 unsigned int offset = offset_in_page(data);
429 unsigned int len = skb_headlen(skb);
434 /* While the header overlaps a page boundary (including being
435 larger than a page), split it it into page-sized chunks. */
436 while (len > PAGE_SIZE - offset) {
437 tx->size = PAGE_SIZE - offset;
438 tx->flags |= XEN_NETTXF_more_data;
443 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
444 np->tx_skbs[id].skb = skb_get(skb);
445 tx = RING_GET_REQUEST(&np->tx, prod++);
447 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
448 BUG_ON((signed short)ref < 0);
450 mfn = virt_to_mfn(data);
451 gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id,
452 mfn, GNTMAP_readonly);
454 tx->gref = np->grant_tx_ref[id] = ref;
460 /* Grant backend access to each skb fragment page. */
461 for (i = 0; i < frags; i++) {
462 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
463 struct page *page = skb_frag_page(frag);
465 len = skb_frag_size(frag);
466 offset = frag->page_offset;
468 /* Data must not cross a page boundary. */
469 BUG_ON(len + offset > PAGE_SIZE<<compound_order(page));
471 /* Skip unused frames from start of page */
472 page += offset >> PAGE_SHIFT;
473 offset &= ~PAGE_MASK;
478 BUG_ON(offset >= PAGE_SIZE);
480 bytes = PAGE_SIZE - offset;
484 tx->flags |= XEN_NETTXF_more_data;
486 id = get_id_from_freelist(&np->tx_skb_freelist,
488 np->tx_skbs[id].skb = skb_get(skb);
489 tx = RING_GET_REQUEST(&np->tx, prod++);
491 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
492 BUG_ON((signed short)ref < 0);
494 mfn = pfn_to_mfn(page_to_pfn(page));
495 gnttab_grant_foreign_access_ref(ref,
496 np->xbdev->otherend_id,
497 mfn, GNTMAP_readonly);
499 tx->gref = np->grant_tx_ref[id] = ref;
508 if (offset == PAGE_SIZE && len) {
509 BUG_ON(!PageCompound(page));
516 np->tx.req_prod_pvt = prod;
520 * Count how many ring slots are required to send the frags of this
521 * skb. Each frag might be a compound page.
523 static int xennet_count_skb_frag_slots(struct sk_buff *skb)
525 int i, frags = skb_shinfo(skb)->nr_frags;
528 for (i = 0; i < frags; i++) {
529 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
530 unsigned long size = skb_frag_size(frag);
531 unsigned long offset = frag->page_offset;
533 /* Skip unused frames from start of page */
534 offset &= ~PAGE_MASK;
536 pages += PFN_UP(offset + size);
542 static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
545 struct netfront_info *np = netdev_priv(dev);
546 struct netfront_stats *stats = this_cpu_ptr(np->stats);
547 struct xen_netif_tx_request *tx;
548 char *data = skb->data;
554 unsigned int offset = offset_in_page(data);
555 unsigned int len = skb_headlen(skb);
558 /* If skb->len is too big for wire format, drop skb and alert
559 * user about misconfiguration.
561 if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
562 net_alert_ratelimited(
563 "xennet: skb->len = %u, too big for wire format\n",
568 slots = DIV_ROUND_UP(offset + len, PAGE_SIZE) +
569 xennet_count_skb_frag_slots(skb);
570 if (unlikely(slots > MAX_SKB_FRAGS + 1)) {
571 net_alert_ratelimited(
572 "xennet: skb rides the rocket: %d slots\n", slots);
576 spin_lock_irqsave(&np->tx_lock, flags);
578 if (unlikely(!netif_carrier_ok(dev) ||
579 (slots > 1 && !xennet_can_sg(dev)) ||
580 netif_needs_gso(skb, netif_skb_features(skb)))) {
581 spin_unlock_irqrestore(&np->tx_lock, flags);
585 i = np->tx.req_prod_pvt;
587 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
588 np->tx_skbs[id].skb = skb;
590 tx = RING_GET_REQUEST(&np->tx, i);
593 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
594 BUG_ON((signed short)ref < 0);
595 mfn = virt_to_mfn(data);
596 gnttab_grant_foreign_access_ref(
597 ref, np->xbdev->otherend_id, mfn, GNTMAP_readonly);
598 tx->gref = np->grant_tx_ref[id] = ref;
603 if (skb->ip_summed == CHECKSUM_PARTIAL)
605 tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
606 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
607 /* remote but checksummed. */
608 tx->flags |= XEN_NETTXF_data_validated;
610 if (skb_shinfo(skb)->gso_size) {
611 struct xen_netif_extra_info *gso;
613 gso = (struct xen_netif_extra_info *)
614 RING_GET_REQUEST(&np->tx, ++i);
616 tx->flags |= XEN_NETTXF_extra_info;
618 gso->u.gso.size = skb_shinfo(skb)->gso_size;
619 gso->u.gso.type = XEN_NETIF_GSO_TYPE_TCPV4;
621 gso->u.gso.features = 0;
623 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
627 np->tx.req_prod_pvt = i + 1;
629 xennet_make_frags(skb, dev, tx);
632 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->tx, notify);
634 notify_remote_via_irq(np->tx_irq);
636 u64_stats_update_begin(&stats->syncp);
637 stats->tx_bytes += skb->len;
639 u64_stats_update_end(&stats->syncp);
641 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
642 xennet_tx_buf_gc(dev);
644 if (!netfront_tx_slot_available(np))
645 netif_stop_queue(dev);
647 spin_unlock_irqrestore(&np->tx_lock, flags);
652 dev->stats.tx_dropped++;
657 static int xennet_close(struct net_device *dev)
659 struct netfront_info *np = netdev_priv(dev);
660 netif_stop_queue(np->netdev);
661 napi_disable(&np->napi);
665 static void xennet_move_rx_slot(struct netfront_info *np, struct sk_buff *skb,
668 int new = xennet_rxidx(np->rx.req_prod_pvt);
670 BUG_ON(np->rx_skbs[new]);
671 np->rx_skbs[new] = skb;
672 np->grant_rx_ref[new] = ref;
673 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new;
674 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref;
675 np->rx.req_prod_pvt++;
678 static int xennet_get_extras(struct netfront_info *np,
679 struct xen_netif_extra_info *extras,
683 struct xen_netif_extra_info *extra;
684 struct device *dev = &np->netdev->dev;
685 RING_IDX cons = np->rx.rsp_cons;
692 if (unlikely(cons + 1 == rp)) {
694 dev_warn(dev, "Missing extra info\n");
699 extra = (struct xen_netif_extra_info *)
700 RING_GET_RESPONSE(&np->rx, ++cons);
702 if (unlikely(!extra->type ||
703 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
705 dev_warn(dev, "Invalid extra type: %d\n",
709 memcpy(&extras[extra->type - 1], extra,
713 skb = xennet_get_rx_skb(np, cons);
714 ref = xennet_get_rx_ref(np, cons);
715 xennet_move_rx_slot(np, skb, ref);
716 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
718 np->rx.rsp_cons = cons;
722 static int xennet_get_responses(struct netfront_info *np,
723 struct netfront_rx_info *rinfo, RING_IDX rp,
724 struct sk_buff_head *list)
726 struct xen_netif_rx_response *rx = &rinfo->rx;
727 struct xen_netif_extra_info *extras = rinfo->extras;
728 struct device *dev = &np->netdev->dev;
729 RING_IDX cons = np->rx.rsp_cons;
730 struct sk_buff *skb = xennet_get_rx_skb(np, cons);
731 grant_ref_t ref = xennet_get_rx_ref(np, cons);
732 int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD);
737 if (rx->flags & XEN_NETRXF_extra_info) {
738 err = xennet_get_extras(np, extras, rp);
739 cons = np->rx.rsp_cons;
743 if (unlikely(rx->status < 0 ||
744 rx->offset + rx->status > PAGE_SIZE)) {
746 dev_warn(dev, "rx->offset: %x, size: %u\n",
747 rx->offset, rx->status);
748 xennet_move_rx_slot(np, skb, ref);
754 * This definitely indicates a bug, either in this driver or in
755 * the backend driver. In future this should flag the bad
756 * situation to the system controller to reboot the backend.
758 if (ref == GRANT_INVALID_REF) {
760 dev_warn(dev, "Bad rx response id %d.\n",
766 ret = gnttab_end_foreign_access_ref(ref, 0);
769 gnttab_release_grant_reference(&np->gref_rx_head, ref);
771 __skb_queue_tail(list, skb);
774 if (!(rx->flags & XEN_NETRXF_more_data))
777 if (cons + slots == rp) {
779 dev_warn(dev, "Need more slots\n");
784 rx = RING_GET_RESPONSE(&np->rx, cons + slots);
785 skb = xennet_get_rx_skb(np, cons + slots);
786 ref = xennet_get_rx_ref(np, cons + slots);
790 if (unlikely(slots > max)) {
792 dev_warn(dev, "Too many slots\n");
797 np->rx.rsp_cons = cons + slots;
802 static int xennet_set_skb_gso(struct sk_buff *skb,
803 struct xen_netif_extra_info *gso)
805 if (!gso->u.gso.size) {
807 printk(KERN_WARNING "GSO size must not be zero.\n");
811 /* Currently only TCPv4 S.O. is supported. */
812 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4) {
814 printk(KERN_WARNING "Bad GSO type %d.\n", gso->u.gso.type);
818 skb_shinfo(skb)->gso_size = gso->u.gso.size;
819 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
821 /* Header must be checked, and gso_segs computed. */
822 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
823 skb_shinfo(skb)->gso_segs = 0;
828 static RING_IDX xennet_fill_frags(struct netfront_info *np,
830 struct sk_buff_head *list)
832 struct skb_shared_info *shinfo = skb_shinfo(skb);
833 int nr_frags = shinfo->nr_frags;
834 RING_IDX cons = np->rx.rsp_cons;
835 struct sk_buff *nskb;
837 while ((nskb = __skb_dequeue(list))) {
838 struct xen_netif_rx_response *rx =
839 RING_GET_RESPONSE(&np->rx, ++cons);
840 skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
842 __skb_fill_page_desc(skb, nr_frags,
843 skb_frag_page(nfrag),
844 rx->offset, rx->status);
846 skb->data_len += rx->status;
848 skb_shinfo(nskb)->nr_frags = 0;
854 shinfo->nr_frags = nr_frags;
858 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
863 int recalculate_partial_csum = 0;
866 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
867 * peers can fail to set NETRXF_csum_blank when sending a GSO
868 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
869 * recalculate the partial checksum.
871 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
872 struct netfront_info *np = netdev_priv(dev);
873 np->rx_gso_checksum_fixup++;
874 skb->ip_summed = CHECKSUM_PARTIAL;
875 recalculate_partial_csum = 1;
878 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
879 if (skb->ip_summed != CHECKSUM_PARTIAL)
882 if (skb->protocol != htons(ETH_P_IP))
885 iph = (void *)skb->data;
886 th = skb->data + 4 * iph->ihl;
887 if (th >= skb_tail_pointer(skb))
890 skb->csum_start = th - skb->head;
891 switch (iph->protocol) {
893 skb->csum_offset = offsetof(struct tcphdr, check);
895 if (recalculate_partial_csum) {
896 struct tcphdr *tcph = (struct tcphdr *)th;
897 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
898 skb->len - iph->ihl*4,
903 skb->csum_offset = offsetof(struct udphdr, check);
905 if (recalculate_partial_csum) {
906 struct udphdr *udph = (struct udphdr *)th;
907 udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
908 skb->len - iph->ihl*4,
914 printk(KERN_ERR "Attempting to checksum a non-"
915 "TCP/UDP packet, dropping a protocol"
916 " %d packet", iph->protocol);
920 if ((th + skb->csum_offset + 2) > skb_tail_pointer(skb))
929 static int handle_incoming_queue(struct net_device *dev,
930 struct sk_buff_head *rxq)
932 struct netfront_info *np = netdev_priv(dev);
933 struct netfront_stats *stats = this_cpu_ptr(np->stats);
934 int packets_dropped = 0;
937 while ((skb = __skb_dequeue(rxq)) != NULL) {
938 int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
940 __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
942 /* Ethernet work: Delayed to here as it peeks the header. */
943 skb->protocol = eth_type_trans(skb, dev);
945 if (checksum_setup(dev, skb)) {
948 dev->stats.rx_errors++;
952 u64_stats_update_begin(&stats->syncp);
954 stats->rx_bytes += skb->len;
955 u64_stats_update_end(&stats->syncp);
958 netif_receive_skb(skb);
961 return packets_dropped;
964 static int xennet_poll(struct napi_struct *napi, int budget)
966 struct netfront_info *np = container_of(napi, struct netfront_info, napi);
967 struct net_device *dev = np->netdev;
969 struct netfront_rx_info rinfo;
970 struct xen_netif_rx_response *rx = &rinfo.rx;
971 struct xen_netif_extra_info *extras = rinfo.extras;
974 struct sk_buff_head rxq;
975 struct sk_buff_head errq;
976 struct sk_buff_head tmpq;
980 spin_lock(&np->rx_lock);
982 skb_queue_head_init(&rxq);
983 skb_queue_head_init(&errq);
984 skb_queue_head_init(&tmpq);
986 rp = np->rx.sring->rsp_prod;
987 rmb(); /* Ensure we see queued responses up to 'rp'. */
991 while ((i != rp) && (work_done < budget)) {
992 memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx));
993 memset(extras, 0, sizeof(rinfo.extras));
995 err = xennet_get_responses(np, &rinfo, rp, &tmpq);
999 while ((skb = __skb_dequeue(&tmpq)))
1000 __skb_queue_tail(&errq, skb);
1001 dev->stats.rx_errors++;
1002 i = np->rx.rsp_cons;
1006 skb = __skb_dequeue(&tmpq);
1008 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1009 struct xen_netif_extra_info *gso;
1010 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1012 if (unlikely(xennet_set_skb_gso(skb, gso))) {
1013 __skb_queue_head(&tmpq, skb);
1014 np->rx.rsp_cons += skb_queue_len(&tmpq);
1019 NETFRONT_SKB_CB(skb)->pull_to = rx->status;
1020 if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
1021 NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
1023 skb_shinfo(skb)->frags[0].page_offset = rx->offset;
1024 skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
1025 skb->data_len = rx->status;
1027 i = xennet_fill_frags(np, skb, &tmpq);
1030 * Truesize is the actual allocation size, even if the
1031 * allocation is only partially used.
1033 skb->truesize += PAGE_SIZE * skb_shinfo(skb)->nr_frags;
1034 skb->len += skb->data_len;
1036 if (rx->flags & XEN_NETRXF_csum_blank)
1037 skb->ip_summed = CHECKSUM_PARTIAL;
1038 else if (rx->flags & XEN_NETRXF_data_validated)
1039 skb->ip_summed = CHECKSUM_UNNECESSARY;
1041 __skb_queue_tail(&rxq, skb);
1043 np->rx.rsp_cons = ++i;
1047 __skb_queue_purge(&errq);
1049 work_done -= handle_incoming_queue(dev, &rxq);
1051 /* If we get a callback with very few responses, reduce fill target. */
1052 /* NB. Note exponential increase, linear decrease. */
1053 if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) >
1054 ((3*np->rx_target) / 4)) &&
1055 (--np->rx_target < np->rx_min_target))
1056 np->rx_target = np->rx_min_target;
1058 xennet_alloc_rx_buffers(dev);
1060 if (work_done < budget) {
1063 local_irq_save(flags);
1065 RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, more_to_do);
1067 __napi_complete(napi);
1069 local_irq_restore(flags);
1072 spin_unlock(&np->rx_lock);
1077 static int xennet_change_mtu(struct net_device *dev, int mtu)
1079 int max = xennet_can_sg(dev) ?
1080 XEN_NETIF_MAX_TX_SIZE - MAX_TCP_HEADER : ETH_DATA_LEN;
1088 static struct rtnl_link_stats64 *xennet_get_stats64(struct net_device *dev,
1089 struct rtnl_link_stats64 *tot)
1091 struct netfront_info *np = netdev_priv(dev);
1094 for_each_possible_cpu(cpu) {
1095 struct netfront_stats *stats = per_cpu_ptr(np->stats, cpu);
1096 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1100 start = u64_stats_fetch_begin_bh(&stats->syncp);
1102 rx_packets = stats->rx_packets;
1103 tx_packets = stats->tx_packets;
1104 rx_bytes = stats->rx_bytes;
1105 tx_bytes = stats->tx_bytes;
1106 } while (u64_stats_fetch_retry_bh(&stats->syncp, start));
1108 tot->rx_packets += rx_packets;
1109 tot->tx_packets += tx_packets;
1110 tot->rx_bytes += rx_bytes;
1111 tot->tx_bytes += tx_bytes;
1114 tot->rx_errors = dev->stats.rx_errors;
1115 tot->tx_dropped = dev->stats.tx_dropped;
1120 static void xennet_release_tx_bufs(struct netfront_info *np)
1122 struct sk_buff *skb;
1125 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1126 /* Skip over entries which are actually freelist references */
1127 if (skb_entry_is_link(&np->tx_skbs[i]))
1130 skb = np->tx_skbs[i].skb;
1131 gnttab_end_foreign_access_ref(np->grant_tx_ref[i],
1133 gnttab_release_grant_reference(&np->gref_tx_head,
1134 np->grant_tx_ref[i]);
1135 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1136 add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, i);
1137 dev_kfree_skb_irq(skb);
1141 static void xennet_release_rx_bufs(struct netfront_info *np)
1143 struct mmu_update *mmu = np->rx_mmu;
1144 struct multicall_entry *mcl = np->rx_mcl;
1145 struct sk_buff_head free_list;
1146 struct sk_buff *skb;
1148 int xfer = 0, noxfer = 0, unused = 0;
1151 dev_warn(&np->netdev->dev, "%s: fix me for copying receiver.\n",
1155 skb_queue_head_init(&free_list);
1157 spin_lock_bh(&np->rx_lock);
1159 for (id = 0; id < NET_RX_RING_SIZE; id++) {
1160 ref = np->grant_rx_ref[id];
1161 if (ref == GRANT_INVALID_REF) {
1166 skb = np->rx_skbs[id];
1167 mfn = gnttab_end_foreign_transfer_ref(ref);
1168 gnttab_release_grant_reference(&np->gref_rx_head, ref);
1169 np->grant_rx_ref[id] = GRANT_INVALID_REF;
1172 skb_shinfo(skb)->nr_frags = 0;
1178 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1179 /* Remap the page. */
1180 const struct page *page =
1181 skb_frag_page(&skb_shinfo(skb)->frags[0]);
1182 unsigned long pfn = page_to_pfn(page);
1183 void *vaddr = page_address(page);
1185 MULTI_update_va_mapping(mcl, (unsigned long)vaddr,
1186 mfn_pte(mfn, PAGE_KERNEL),
1189 mmu->ptr = ((u64)mfn << PAGE_SHIFT)
1190 | MMU_MACHPHYS_UPDATE;
1194 set_phys_to_machine(pfn, mfn);
1196 __skb_queue_tail(&free_list, skb);
1200 dev_info(&np->netdev->dev, "%s: %d xfer, %d noxfer, %d unused\n",
1201 __func__, xfer, noxfer, unused);
1204 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1205 /* Do all the remapping work and M2P updates. */
1206 MULTI_mmu_update(mcl, np->rx_mmu, mmu - np->rx_mmu,
1209 HYPERVISOR_multicall(np->rx_mcl, mcl - np->rx_mcl);
1213 __skb_queue_purge(&free_list);
1215 spin_unlock_bh(&np->rx_lock);
1218 static void xennet_uninit(struct net_device *dev)
1220 struct netfront_info *np = netdev_priv(dev);
1221 xennet_release_tx_bufs(np);
1222 xennet_release_rx_bufs(np);
1223 gnttab_free_grant_references(np->gref_tx_head);
1224 gnttab_free_grant_references(np->gref_rx_head);
1227 static netdev_features_t xennet_fix_features(struct net_device *dev,
1228 netdev_features_t features)
1230 struct netfront_info *np = netdev_priv(dev);
1233 if (features & NETIF_F_SG) {
1234 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg",
1239 features &= ~NETIF_F_SG;
1242 if (features & NETIF_F_TSO) {
1243 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1244 "feature-gso-tcpv4", "%d", &val) < 0)
1248 features &= ~NETIF_F_TSO;
1254 static int xennet_set_features(struct net_device *dev,
1255 netdev_features_t features)
1257 if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1258 netdev_info(dev, "Reducing MTU because no SG offload");
1259 dev->mtu = ETH_DATA_LEN;
1265 static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1267 struct netfront_info *np = dev_id;
1268 struct net_device *dev = np->netdev;
1269 unsigned long flags;
1271 spin_lock_irqsave(&np->tx_lock, flags);
1272 xennet_tx_buf_gc(dev);
1273 spin_unlock_irqrestore(&np->tx_lock, flags);
1278 static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1280 struct netfront_info *np = dev_id;
1281 struct net_device *dev = np->netdev;
1283 if (likely(netif_carrier_ok(dev) &&
1284 RING_HAS_UNCONSUMED_RESPONSES(&np->rx)))
1285 napi_schedule(&np->napi);
1290 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1292 xennet_tx_interrupt(irq, dev_id);
1293 xennet_rx_interrupt(irq, dev_id);
1297 #ifdef CONFIG_NET_POLL_CONTROLLER
1298 static void xennet_poll_controller(struct net_device *dev)
1300 xennet_interrupt(0, dev);
1304 static const struct net_device_ops xennet_netdev_ops = {
1305 .ndo_open = xennet_open,
1306 .ndo_uninit = xennet_uninit,
1307 .ndo_stop = xennet_close,
1308 .ndo_start_xmit = xennet_start_xmit,
1309 .ndo_change_mtu = xennet_change_mtu,
1310 .ndo_get_stats64 = xennet_get_stats64,
1311 .ndo_set_mac_address = eth_mac_addr,
1312 .ndo_validate_addr = eth_validate_addr,
1313 .ndo_fix_features = xennet_fix_features,
1314 .ndo_set_features = xennet_set_features,
1315 #ifdef CONFIG_NET_POLL_CONTROLLER
1316 .ndo_poll_controller = xennet_poll_controller,
1320 static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1323 struct net_device *netdev;
1324 struct netfront_info *np;
1326 netdev = alloc_etherdev(sizeof(struct netfront_info));
1328 return ERR_PTR(-ENOMEM);
1330 np = netdev_priv(netdev);
1333 spin_lock_init(&np->tx_lock);
1334 spin_lock_init(&np->rx_lock);
1336 skb_queue_head_init(&np->rx_batch);
1337 np->rx_target = RX_DFL_MIN_TARGET;
1338 np->rx_min_target = RX_DFL_MIN_TARGET;
1339 np->rx_max_target = RX_MAX_TARGET;
1341 init_timer(&np->rx_refill_timer);
1342 np->rx_refill_timer.data = (unsigned long)netdev;
1343 np->rx_refill_timer.function = rx_refill_timeout;
1346 np->stats = alloc_percpu(struct netfront_stats);
1347 if (np->stats == NULL)
1350 /* Initialise tx_skbs as a free chain containing every entry. */
1351 np->tx_skb_freelist = 0;
1352 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1353 skb_entry_set_link(&np->tx_skbs[i], i+1);
1354 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1357 /* Clear out rx_skbs */
1358 for (i = 0; i < NET_RX_RING_SIZE; i++) {
1359 np->rx_skbs[i] = NULL;
1360 np->grant_rx_ref[i] = GRANT_INVALID_REF;
1363 /* A grant for every tx ring slot */
1364 if (gnttab_alloc_grant_references(TX_MAX_TARGET,
1365 &np->gref_tx_head) < 0) {
1366 printk(KERN_ALERT "#### netfront can't alloc tx grant refs\n");
1368 goto exit_free_stats;
1370 /* A grant for every rx ring slot */
1371 if (gnttab_alloc_grant_references(RX_MAX_TARGET,
1372 &np->gref_rx_head) < 0) {
1373 printk(KERN_ALERT "#### netfront can't alloc rx grant refs\n");
1378 netdev->netdev_ops = &xennet_netdev_ops;
1380 netif_napi_add(netdev, &np->napi, xennet_poll, 64);
1381 netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1383 netdev->hw_features = NETIF_F_IP_CSUM | NETIF_F_SG | NETIF_F_TSO;
1386 * Assume that all hw features are available for now. This set
1387 * will be adjusted by the call to netdev_update_features() in
1388 * xennet_connect() which is the earliest point where we can
1389 * negotiate with the backend regarding supported features.
1391 netdev->features |= netdev->hw_features;
1393 SET_ETHTOOL_OPS(netdev, &xennet_ethtool_ops);
1394 SET_NETDEV_DEV(netdev, &dev->dev);
1396 netif_set_gso_max_size(netdev, XEN_NETIF_MAX_TX_SIZE - MAX_TCP_HEADER);
1398 np->netdev = netdev;
1400 netif_carrier_off(netdev);
1405 gnttab_free_grant_references(np->gref_tx_head);
1407 free_percpu(np->stats);
1409 free_netdev(netdev);
1410 return ERR_PTR(err);
1414 * Entry point to this code when a new device is created. Allocate the basic
1415 * structures and the ring buffers for communication with the backend, and
1416 * inform the backend of the appropriate details for those.
1418 static int netfront_probe(struct xenbus_device *dev,
1419 const struct xenbus_device_id *id)
1422 struct net_device *netdev;
1423 struct netfront_info *info;
1425 netdev = xennet_create_dev(dev);
1426 if (IS_ERR(netdev)) {
1427 err = PTR_ERR(netdev);
1428 xenbus_dev_fatal(dev, err, "creating netdev");
1432 info = netdev_priv(netdev);
1433 dev_set_drvdata(&dev->dev, info);
1435 err = register_netdev(info->netdev);
1437 printk(KERN_WARNING "%s: register_netdev err=%d\n",
1442 err = xennet_sysfs_addif(info->netdev);
1444 unregister_netdev(info->netdev);
1445 printk(KERN_WARNING "%s: add sysfs failed err=%d\n",
1453 free_netdev(netdev);
1454 dev_set_drvdata(&dev->dev, NULL);
1458 static void xennet_end_access(int ref, void *page)
1460 /* This frees the page as a side-effect */
1461 if (ref != GRANT_INVALID_REF)
1462 gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1465 static void xennet_disconnect_backend(struct netfront_info *info)
1467 /* Stop old i/f to prevent errors whilst we rebuild the state. */
1468 spin_lock_bh(&info->rx_lock);
1469 spin_lock_irq(&info->tx_lock);
1470 netif_carrier_off(info->netdev);
1471 spin_unlock_irq(&info->tx_lock);
1472 spin_unlock_bh(&info->rx_lock);
1474 if (info->tx_irq && (info->tx_irq == info->rx_irq))
1475 unbind_from_irqhandler(info->tx_irq, info);
1476 if (info->tx_irq && (info->tx_irq != info->rx_irq)) {
1477 unbind_from_irqhandler(info->tx_irq, info);
1478 unbind_from_irqhandler(info->rx_irq, info);
1480 info->tx_evtchn = info->rx_evtchn = 0;
1481 info->tx_irq = info->rx_irq = 0;
1483 /* End access and free the pages */
1484 xennet_end_access(info->tx_ring_ref, info->tx.sring);
1485 xennet_end_access(info->rx_ring_ref, info->rx.sring);
1487 info->tx_ring_ref = GRANT_INVALID_REF;
1488 info->rx_ring_ref = GRANT_INVALID_REF;
1489 info->tx.sring = NULL;
1490 info->rx.sring = NULL;
1494 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1495 * driver restart. We tear down our netif structure and recreate it, but
1496 * leave the device-layer structures intact so that this is transparent to the
1497 * rest of the kernel.
1499 static int netfront_resume(struct xenbus_device *dev)
1501 struct netfront_info *info = dev_get_drvdata(&dev->dev);
1503 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1505 xennet_disconnect_backend(info);
1509 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1511 char *s, *e, *macstr;
1514 macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1516 return PTR_ERR(macstr);
1518 for (i = 0; i < ETH_ALEN; i++) {
1519 mac[i] = simple_strtoul(s, &e, 16);
1520 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1531 static int setup_netfront_single(struct netfront_info *info)
1535 err = xenbus_alloc_evtchn(info->xbdev, &info->tx_evtchn);
1539 err = bind_evtchn_to_irqhandler(info->tx_evtchn,
1541 0, info->netdev->name, info);
1544 info->rx_evtchn = info->tx_evtchn;
1545 info->rx_irq = info->tx_irq = err;
1550 xenbus_free_evtchn(info->xbdev, info->tx_evtchn);
1551 info->tx_evtchn = 0;
1556 static int setup_netfront_split(struct netfront_info *info)
1560 err = xenbus_alloc_evtchn(info->xbdev, &info->tx_evtchn);
1563 err = xenbus_alloc_evtchn(info->xbdev, &info->rx_evtchn);
1565 goto alloc_rx_evtchn_fail;
1567 snprintf(info->tx_irq_name, sizeof(info->tx_irq_name),
1568 "%s-tx", info->netdev->name);
1569 err = bind_evtchn_to_irqhandler(info->tx_evtchn,
1570 xennet_tx_interrupt,
1571 0, info->tx_irq_name, info);
1576 snprintf(info->rx_irq_name, sizeof(info->rx_irq_name),
1577 "%s-rx", info->netdev->name);
1578 err = bind_evtchn_to_irqhandler(info->rx_evtchn,
1579 xennet_rx_interrupt,
1580 0, info->rx_irq_name, info);
1588 unbind_from_irqhandler(info->tx_irq, info);
1591 xenbus_free_evtchn(info->xbdev, info->rx_evtchn);
1592 info->rx_evtchn = 0;
1593 alloc_rx_evtchn_fail:
1594 xenbus_free_evtchn(info->xbdev, info->tx_evtchn);
1595 info->tx_evtchn = 0;
1600 static int setup_netfront(struct xenbus_device *dev, struct netfront_info *info)
1602 struct xen_netif_tx_sring *txs;
1603 struct xen_netif_rx_sring *rxs;
1605 struct net_device *netdev = info->netdev;
1606 unsigned int feature_split_evtchn;
1608 info->tx_ring_ref = GRANT_INVALID_REF;
1609 info->rx_ring_ref = GRANT_INVALID_REF;
1610 info->rx.sring = NULL;
1611 info->tx.sring = NULL;
1614 err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
1615 "feature-split-event-channels", "%u",
1616 &feature_split_evtchn);
1618 feature_split_evtchn = 0;
1620 err = xen_net_read_mac(dev, netdev->dev_addr);
1622 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1626 txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1629 xenbus_dev_fatal(dev, err, "allocating tx ring page");
1632 SHARED_RING_INIT(txs);
1633 FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE);
1635 err = xenbus_grant_ring(dev, virt_to_mfn(txs));
1637 goto grant_tx_ring_fail;
1639 info->tx_ring_ref = err;
1640 rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1643 xenbus_dev_fatal(dev, err, "allocating rx ring page");
1644 goto alloc_rx_ring_fail;
1646 SHARED_RING_INIT(rxs);
1647 FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE);
1649 err = xenbus_grant_ring(dev, virt_to_mfn(rxs));
1651 goto grant_rx_ring_fail;
1652 info->rx_ring_ref = err;
1654 if (feature_split_evtchn)
1655 err = setup_netfront_split(info);
1656 /* setup single event channel if
1657 * a) feature-split-event-channels == 0
1658 * b) feature-split-event-channels == 1 but failed to setup
1660 if (!feature_split_evtchn || (feature_split_evtchn && err))
1661 err = setup_netfront_single(info);
1664 goto alloc_evtchn_fail;
1668 /* If we fail to setup netfront, it is safe to just revoke access to
1669 * granted pages because backend is not accessing it at this point.
1672 gnttab_end_foreign_access_ref(info->rx_ring_ref, 0);
1674 free_page((unsigned long)rxs);
1676 gnttab_end_foreign_access_ref(info->tx_ring_ref, 0);
1678 free_page((unsigned long)txs);
1683 /* Common code used when first setting up, and when resuming. */
1684 static int talk_to_netback(struct xenbus_device *dev,
1685 struct netfront_info *info)
1687 const char *message;
1688 struct xenbus_transaction xbt;
1691 /* Create shared ring, alloc event channel. */
1692 err = setup_netfront(dev, info);
1697 err = xenbus_transaction_start(&xbt);
1699 xenbus_dev_fatal(dev, err, "starting transaction");
1703 err = xenbus_printf(xbt, dev->nodename, "tx-ring-ref", "%u",
1706 message = "writing tx ring-ref";
1707 goto abort_transaction;
1709 err = xenbus_printf(xbt, dev->nodename, "rx-ring-ref", "%u",
1712 message = "writing rx ring-ref";
1713 goto abort_transaction;
1716 if (info->tx_evtchn == info->rx_evtchn) {
1717 err = xenbus_printf(xbt, dev->nodename,
1718 "event-channel", "%u", info->tx_evtchn);
1720 message = "writing event-channel";
1721 goto abort_transaction;
1724 err = xenbus_printf(xbt, dev->nodename,
1725 "event-channel-tx", "%u", info->tx_evtchn);
1727 message = "writing event-channel-tx";
1728 goto abort_transaction;
1730 err = xenbus_printf(xbt, dev->nodename,
1731 "event-channel-rx", "%u", info->rx_evtchn);
1733 message = "writing event-channel-rx";
1734 goto abort_transaction;
1738 err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1741 message = "writing request-rx-copy";
1742 goto abort_transaction;
1745 err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1747 message = "writing feature-rx-notify";
1748 goto abort_transaction;
1751 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1753 message = "writing feature-sg";
1754 goto abort_transaction;
1757 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1759 message = "writing feature-gso-tcpv4";
1760 goto abort_transaction;
1763 err = xenbus_transaction_end(xbt, 0);
1767 xenbus_dev_fatal(dev, err, "completing transaction");
1774 xenbus_transaction_end(xbt, 1);
1775 xenbus_dev_fatal(dev, err, "%s", message);
1777 xennet_disconnect_backend(info);
1782 static int xennet_connect(struct net_device *dev)
1784 struct netfront_info *np = netdev_priv(dev);
1785 int i, requeue_idx, err;
1786 struct sk_buff *skb;
1788 struct xen_netif_rx_request *req;
1789 unsigned int feature_rx_copy;
1791 err = xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1792 "feature-rx-copy", "%u", &feature_rx_copy);
1794 feature_rx_copy = 0;
1796 if (!feature_rx_copy) {
1798 "backend does not support copying receive path\n");
1802 err = talk_to_netback(np->xbdev, np);
1807 netdev_update_features(dev);
1810 spin_lock_bh(&np->rx_lock);
1811 spin_lock_irq(&np->tx_lock);
1813 /* Step 1: Discard all pending TX packet fragments. */
1814 xennet_release_tx_bufs(np);
1816 /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1817 for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) {
1819 const struct page *page;
1820 if (!np->rx_skbs[i])
1823 skb = np->rx_skbs[requeue_idx] = xennet_get_rx_skb(np, i);
1824 ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i);
1825 req = RING_GET_REQUEST(&np->rx, requeue_idx);
1827 frag = &skb_shinfo(skb)->frags[0];
1828 page = skb_frag_page(frag);
1829 gnttab_grant_foreign_access_ref(
1830 ref, np->xbdev->otherend_id,
1831 pfn_to_mfn(page_to_pfn(page)),
1834 req->id = requeue_idx;
1839 np->rx.req_prod_pvt = requeue_idx;
1842 * Step 3: All public and private state should now be sane. Get
1843 * ready to start sending and receiving packets and give the driver
1844 * domain a kick because we've probably just requeued some
1847 netif_carrier_on(np->netdev);
1848 notify_remote_via_irq(np->tx_irq);
1849 if (np->tx_irq != np->rx_irq)
1850 notify_remote_via_irq(np->rx_irq);
1851 xennet_tx_buf_gc(dev);
1852 xennet_alloc_rx_buffers(dev);
1854 spin_unlock_irq(&np->tx_lock);
1855 spin_unlock_bh(&np->rx_lock);
1861 * Callback received when the backend's state changes.
1863 static void netback_changed(struct xenbus_device *dev,
1864 enum xenbus_state backend_state)
1866 struct netfront_info *np = dev_get_drvdata(&dev->dev);
1867 struct net_device *netdev = np->netdev;
1869 dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
1871 switch (backend_state) {
1872 case XenbusStateInitialising:
1873 case XenbusStateInitialised:
1874 case XenbusStateReconfiguring:
1875 case XenbusStateReconfigured:
1876 case XenbusStateUnknown:
1877 case XenbusStateClosed:
1880 case XenbusStateInitWait:
1881 if (dev->state != XenbusStateInitialising)
1883 if (xennet_connect(netdev) != 0)
1885 xenbus_switch_state(dev, XenbusStateConnected);
1888 case XenbusStateConnected:
1889 netdev_notify_peers(netdev);
1892 case XenbusStateClosing:
1893 xenbus_frontend_closed(dev);
1898 static const struct xennet_stat {
1899 char name[ETH_GSTRING_LEN];
1901 } xennet_stats[] = {
1903 "rx_gso_checksum_fixup",
1904 offsetof(struct netfront_info, rx_gso_checksum_fixup)
1908 static int xennet_get_sset_count(struct net_device *dev, int string_set)
1910 switch (string_set) {
1912 return ARRAY_SIZE(xennet_stats);
1918 static void xennet_get_ethtool_stats(struct net_device *dev,
1919 struct ethtool_stats *stats, u64 * data)
1921 void *np = netdev_priv(dev);
1924 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
1925 data[i] = *(unsigned long *)(np + xennet_stats[i].offset);
1928 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
1932 switch (stringset) {
1934 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
1935 memcpy(data + i * ETH_GSTRING_LEN,
1936 xennet_stats[i].name, ETH_GSTRING_LEN);
1941 static const struct ethtool_ops xennet_ethtool_ops =
1943 .get_link = ethtool_op_get_link,
1945 .get_sset_count = xennet_get_sset_count,
1946 .get_ethtool_stats = xennet_get_ethtool_stats,
1947 .get_strings = xennet_get_strings,
1951 static ssize_t show_rxbuf_min(struct device *dev,
1952 struct device_attribute *attr, char *buf)
1954 struct net_device *netdev = to_net_dev(dev);
1955 struct netfront_info *info = netdev_priv(netdev);
1957 return sprintf(buf, "%u\n", info->rx_min_target);
1960 static ssize_t store_rxbuf_min(struct device *dev,
1961 struct device_attribute *attr,
1962 const char *buf, size_t len)
1964 struct net_device *netdev = to_net_dev(dev);
1965 struct netfront_info *np = netdev_priv(netdev);
1967 unsigned long target;
1969 if (!capable(CAP_NET_ADMIN))
1972 target = simple_strtoul(buf, &endp, 0);
1976 if (target < RX_MIN_TARGET)
1977 target = RX_MIN_TARGET;
1978 if (target > RX_MAX_TARGET)
1979 target = RX_MAX_TARGET;
1981 spin_lock_bh(&np->rx_lock);
1982 if (target > np->rx_max_target)
1983 np->rx_max_target = target;
1984 np->rx_min_target = target;
1985 if (target > np->rx_target)
1986 np->rx_target = target;
1988 xennet_alloc_rx_buffers(netdev);
1990 spin_unlock_bh(&np->rx_lock);
1994 static ssize_t show_rxbuf_max(struct device *dev,
1995 struct device_attribute *attr, char *buf)
1997 struct net_device *netdev = to_net_dev(dev);
1998 struct netfront_info *info = netdev_priv(netdev);
2000 return sprintf(buf, "%u\n", info->rx_max_target);
2003 static ssize_t store_rxbuf_max(struct device *dev,
2004 struct device_attribute *attr,
2005 const char *buf, size_t len)
2007 struct net_device *netdev = to_net_dev(dev);
2008 struct netfront_info *np = netdev_priv(netdev);
2010 unsigned long target;
2012 if (!capable(CAP_NET_ADMIN))
2015 target = simple_strtoul(buf, &endp, 0);
2019 if (target < RX_MIN_TARGET)
2020 target = RX_MIN_TARGET;
2021 if (target > RX_MAX_TARGET)
2022 target = RX_MAX_TARGET;
2024 spin_lock_bh(&np->rx_lock);
2025 if (target < np->rx_min_target)
2026 np->rx_min_target = target;
2027 np->rx_max_target = target;
2028 if (target < np->rx_target)
2029 np->rx_target = target;
2031 xennet_alloc_rx_buffers(netdev);
2033 spin_unlock_bh(&np->rx_lock);
2037 static ssize_t show_rxbuf_cur(struct device *dev,
2038 struct device_attribute *attr, char *buf)
2040 struct net_device *netdev = to_net_dev(dev);
2041 struct netfront_info *info = netdev_priv(netdev);
2043 return sprintf(buf, "%u\n", info->rx_target);
2046 static struct device_attribute xennet_attrs[] = {
2047 __ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf_min, store_rxbuf_min),
2048 __ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf_max, store_rxbuf_max),
2049 __ATTR(rxbuf_cur, S_IRUGO, show_rxbuf_cur, NULL),
2052 static int xennet_sysfs_addif(struct net_device *netdev)
2057 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) {
2058 err = device_create_file(&netdev->dev,
2067 device_remove_file(&netdev->dev, &xennet_attrs[i]);
2071 static void xennet_sysfs_delif(struct net_device *netdev)
2075 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++)
2076 device_remove_file(&netdev->dev, &xennet_attrs[i]);
2079 #endif /* CONFIG_SYSFS */
2081 static const struct xenbus_device_id netfront_ids[] = {
2087 static int xennet_remove(struct xenbus_device *dev)
2089 struct netfront_info *info = dev_get_drvdata(&dev->dev);
2091 dev_dbg(&dev->dev, "%s\n", dev->nodename);
2093 xennet_disconnect_backend(info);
2095 xennet_sysfs_delif(info->netdev);
2097 unregister_netdev(info->netdev);
2099 del_timer_sync(&info->rx_refill_timer);
2101 free_percpu(info->stats);
2103 free_netdev(info->netdev);
2108 static DEFINE_XENBUS_DRIVER(netfront, ,
2109 .probe = netfront_probe,
2110 .remove = xennet_remove,
2111 .resume = netfront_resume,
2112 .otherend_changed = netback_changed,
2115 static int __init netif_init(void)
2120 if (xen_hvm_domain() && !xen_platform_pci_unplug)
2123 printk(KERN_INFO "Initialising Xen virtual ethernet driver.\n");
2125 return xenbus_register_frontend(&netfront_driver);
2127 module_init(netif_init);
2130 static void __exit netif_exit(void)
2132 xenbus_unregister_driver(&netfront_driver);
2134 module_exit(netif_exit);
2136 MODULE_DESCRIPTION("Xen virtual network device frontend");
2137 MODULE_LICENSE("GPL");
2138 MODULE_ALIAS("xen:vif");
2139 MODULE_ALIAS("xennet");