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;
89 struct xenbus_device *xbdev;
92 struct xen_netif_tx_front_ring tx;
96 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
97 * are linked from tx_skb_freelist through skb_entry.link.
99 * NB. Freelist index entries are always going to be less than
100 * PAGE_OFFSET, whereas pointers to skbs will always be equal or
101 * greater than PAGE_OFFSET: we use this property to distinguish
107 } tx_skbs[NET_TX_RING_SIZE];
108 grant_ref_t gref_tx_head;
109 grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
110 struct page *grant_tx_page[NET_TX_RING_SIZE];
111 unsigned tx_skb_freelist;
113 spinlock_t rx_lock ____cacheline_aligned_in_smp;
114 struct xen_netif_rx_front_ring rx;
117 /* Receive-ring batched refills. */
118 #define RX_MIN_TARGET 8
119 #define RX_DFL_MIN_TARGET 64
120 #define RX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256)
121 unsigned rx_min_target, rx_max_target, rx_target;
122 struct sk_buff_head rx_batch;
124 struct timer_list rx_refill_timer;
126 struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
127 grant_ref_t gref_rx_head;
128 grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
130 unsigned long rx_pfn_array[NET_RX_RING_SIZE];
131 struct multicall_entry rx_mcl[NET_RX_RING_SIZE+1];
132 struct mmu_update rx_mmu[NET_RX_RING_SIZE];
135 struct netfront_stats __percpu *stats;
137 unsigned long rx_gso_checksum_fixup;
140 struct netfront_rx_info {
141 struct xen_netif_rx_response rx;
142 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
145 static void skb_entry_set_link(union skb_entry *list, unsigned short id)
150 static int skb_entry_is_link(const union skb_entry *list)
152 BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link));
153 return (unsigned long)list->skb < PAGE_OFFSET;
157 * Access macros for acquiring freeing slots in tx_skbs[].
160 static void add_id_to_freelist(unsigned *head, union skb_entry *list,
163 skb_entry_set_link(&list[id], *head);
167 static unsigned short get_id_from_freelist(unsigned *head,
168 union skb_entry *list)
170 unsigned int id = *head;
171 *head = list[id].link;
175 static int xennet_rxidx(RING_IDX idx)
177 return idx & (NET_RX_RING_SIZE - 1);
180 static struct sk_buff *xennet_get_rx_skb(struct netfront_info *np,
183 int i = xennet_rxidx(ri);
184 struct sk_buff *skb = np->rx_skbs[i];
185 np->rx_skbs[i] = NULL;
189 static grant_ref_t xennet_get_rx_ref(struct netfront_info *np,
192 int i = xennet_rxidx(ri);
193 grant_ref_t ref = np->grant_rx_ref[i];
194 np->grant_rx_ref[i] = GRANT_INVALID_REF;
199 static int xennet_sysfs_addif(struct net_device *netdev);
200 static void xennet_sysfs_delif(struct net_device *netdev);
201 #else /* !CONFIG_SYSFS */
202 #define xennet_sysfs_addif(dev) (0)
203 #define xennet_sysfs_delif(dev) do { } while (0)
206 static bool xennet_can_sg(struct net_device *dev)
208 return dev->features & NETIF_F_SG;
212 static void rx_refill_timeout(unsigned long data)
214 struct net_device *dev = (struct net_device *)data;
215 struct netfront_info *np = netdev_priv(dev);
216 napi_schedule(&np->napi);
219 static int netfront_tx_slot_available(struct netfront_info *np)
221 return (np->tx.req_prod_pvt - np->tx.rsp_cons) <
222 (TX_MAX_TARGET - MAX_SKB_FRAGS - 2);
225 static void xennet_maybe_wake_tx(struct net_device *dev)
227 struct netfront_info *np = netdev_priv(dev);
229 if (unlikely(netif_queue_stopped(dev)) &&
230 netfront_tx_slot_available(np) &&
231 likely(netif_running(dev)))
232 netif_wake_queue(dev);
235 static void xennet_alloc_rx_buffers(struct net_device *dev)
238 struct netfront_info *np = netdev_priv(dev);
241 int i, batch_target, notify;
242 RING_IDX req_prod = np->rx.req_prod_pvt;
246 struct xen_netif_rx_request *req;
248 if (unlikely(!netif_carrier_ok(dev)))
252 * Allocate skbuffs greedily, even though we batch updates to the
253 * receive ring. This creates a less bursty demand on the memory
254 * allocator, so should reduce the chance of failed allocation requests
255 * both for ourself and for other kernel subsystems.
257 batch_target = np->rx_target - (req_prod - np->rx.rsp_cons);
258 for (i = skb_queue_len(&np->rx_batch); i < batch_target; i++) {
259 skb = __netdev_alloc_skb(dev, RX_COPY_THRESHOLD + NET_IP_ALIGN,
260 GFP_ATOMIC | __GFP_NOWARN);
264 /* Align ip header to a 16 bytes boundary */
265 skb_reserve(skb, NET_IP_ALIGN);
267 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
271 /* Any skbuffs queued for refill? Force them out. */
274 /* Could not allocate any skbuffs. Try again later. */
275 mod_timer(&np->rx_refill_timer,
280 skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE);
281 __skb_queue_tail(&np->rx_batch, skb);
284 /* Is the batch large enough to be worthwhile? */
285 if (i < (np->rx_target/2)) {
286 if (req_prod > np->rx.sring->req_prod)
291 /* Adjust our fill target if we risked running out of buffers. */
292 if (((req_prod - np->rx.sring->rsp_prod) < (np->rx_target / 4)) &&
293 ((np->rx_target *= 2) > np->rx_max_target))
294 np->rx_target = np->rx_max_target;
298 skb = __skb_dequeue(&np->rx_batch);
304 id = xennet_rxidx(req_prod + i);
306 BUG_ON(np->rx_skbs[id]);
307 np->rx_skbs[id] = skb;
309 ref = gnttab_claim_grant_reference(&np->gref_rx_head);
310 BUG_ON((signed short)ref < 0);
311 np->grant_rx_ref[id] = ref;
313 pfn = page_to_pfn(skb_frag_page(&skb_shinfo(skb)->frags[0]));
314 vaddr = page_address(skb_frag_page(&skb_shinfo(skb)->frags[0]));
316 req = RING_GET_REQUEST(&np->rx, req_prod + i);
317 gnttab_grant_foreign_access_ref(ref,
318 np->xbdev->otherend_id,
326 wmb(); /* barrier so backend seens requests */
328 /* Above is a suitable barrier to ensure backend will see requests. */
329 np->rx.req_prod_pvt = req_prod + i;
331 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->rx, notify);
333 notify_remote_via_irq(np->netdev->irq);
336 static int xennet_open(struct net_device *dev)
338 struct netfront_info *np = netdev_priv(dev);
340 napi_enable(&np->napi);
342 spin_lock_bh(&np->rx_lock);
343 if (netif_carrier_ok(dev)) {
344 xennet_alloc_rx_buffers(dev);
345 np->rx.sring->rsp_event = np->rx.rsp_cons + 1;
346 if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx))
347 napi_schedule(&np->napi);
349 spin_unlock_bh(&np->rx_lock);
351 netif_start_queue(dev);
356 static void xennet_tx_buf_gc(struct net_device *dev)
360 struct netfront_info *np = netdev_priv(dev);
363 BUG_ON(!netif_carrier_ok(dev));
366 prod = np->tx.sring->rsp_prod;
367 rmb(); /* Ensure we see responses up to 'rp'. */
369 for (cons = np->tx.rsp_cons; cons != prod; cons++) {
370 struct xen_netif_tx_response *txrsp;
372 txrsp = RING_GET_RESPONSE(&np->tx, cons);
373 if (txrsp->status == XEN_NETIF_RSP_NULL)
377 skb = np->tx_skbs[id].skb;
378 if (unlikely(gnttab_query_foreign_access(
379 np->grant_tx_ref[id]) != 0)) {
380 printk(KERN_ALERT "xennet_tx_buf_gc: warning "
381 "-- grant still in use by backend "
385 gnttab_end_foreign_access_ref(
386 np->grant_tx_ref[id], GNTMAP_readonly);
387 gnttab_release_grant_reference(
388 &np->gref_tx_head, np->grant_tx_ref[id]);
389 np->grant_tx_ref[id] = GRANT_INVALID_REF;
390 np->grant_tx_page[id] = NULL;
391 add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, id);
392 dev_kfree_skb_irq(skb);
395 np->tx.rsp_cons = prod;
398 * Set a new event, then check for race with update of tx_cons.
399 * Note that it is essential to schedule a callback, no matter
400 * how few buffers are pending. Even if there is space in the
401 * transmit ring, higher layers may be blocked because too much
402 * data is outstanding: in such cases notification from Xen is
403 * likely to be the only kick that we'll get.
405 np->tx.sring->rsp_event =
406 prod + ((np->tx.sring->req_prod - prod) >> 1) + 1;
407 mb(); /* update shared area */
408 } while ((cons == prod) && (prod != np->tx.sring->rsp_prod));
410 xennet_maybe_wake_tx(dev);
413 static void xennet_make_frags(struct sk_buff *skb, struct net_device *dev,
414 struct xen_netif_tx_request *tx)
416 struct netfront_info *np = netdev_priv(dev);
417 char *data = skb->data;
419 RING_IDX prod = np->tx.req_prod_pvt;
420 int frags = skb_shinfo(skb)->nr_frags;
421 unsigned int offset = offset_in_page(data);
422 unsigned int len = skb_headlen(skb);
427 /* While the header overlaps a page boundary (including being
428 larger than a page), split it it into page-sized chunks. */
429 while (len > PAGE_SIZE - offset) {
430 tx->size = PAGE_SIZE - offset;
431 tx->flags |= XEN_NETTXF_more_data;
436 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
437 np->tx_skbs[id].skb = skb_get(skb);
438 tx = RING_GET_REQUEST(&np->tx, prod++);
440 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
441 BUG_ON((signed short)ref < 0);
443 mfn = virt_to_mfn(data);
444 gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id,
445 mfn, GNTMAP_readonly);
447 np->grant_tx_page[id] = virt_to_page(data);
448 tx->gref = np->grant_tx_ref[id] = ref;
454 /* Grant backend access to each skb fragment page. */
455 for (i = 0; i < frags; i++) {
456 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
457 struct page *page = skb_frag_page(frag);
459 len = skb_frag_size(frag);
460 offset = frag->page_offset;
462 /* Data must not cross a page boundary. */
463 BUG_ON(len + offset > PAGE_SIZE<<compound_order(page));
465 /* Skip unused frames from start of page */
466 page += offset >> PAGE_SHIFT;
467 offset &= ~PAGE_MASK;
472 BUG_ON(offset >= PAGE_SIZE);
474 bytes = PAGE_SIZE - offset;
478 tx->flags |= XEN_NETTXF_more_data;
480 id = get_id_from_freelist(&np->tx_skb_freelist,
482 np->tx_skbs[id].skb = skb_get(skb);
483 tx = RING_GET_REQUEST(&np->tx, prod++);
485 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
486 BUG_ON((signed short)ref < 0);
488 mfn = pfn_to_mfn(page_to_pfn(page));
489 gnttab_grant_foreign_access_ref(ref,
490 np->xbdev->otherend_id,
491 mfn, GNTMAP_readonly);
493 np->grant_tx_page[id] = page;
494 tx->gref = np->grant_tx_ref[id] = ref;
503 if (offset == PAGE_SIZE && len) {
504 BUG_ON(!PageCompound(page));
511 np->tx.req_prod_pvt = prod;
515 * Count how many ring slots are required to send the frags of this
516 * skb. Each frag might be a compound page.
518 static int xennet_count_skb_frag_slots(struct sk_buff *skb)
520 int i, frags = skb_shinfo(skb)->nr_frags;
523 for (i = 0; i < frags; i++) {
524 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
525 unsigned long size = skb_frag_size(frag);
526 unsigned long offset = frag->page_offset;
528 /* Skip unused frames from start of page */
529 offset &= ~PAGE_MASK;
531 pages += PFN_UP(offset + size);
537 static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
540 struct netfront_info *np = netdev_priv(dev);
541 struct netfront_stats *stats = this_cpu_ptr(np->stats);
542 struct xen_netif_tx_request *tx;
543 char *data = skb->data;
549 unsigned int offset = offset_in_page(data);
550 unsigned int len = skb_headlen(skb);
553 /* If skb->len is too big for wire format, drop skb and alert
554 * user about misconfiguration.
556 if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
557 net_alert_ratelimited(
558 "xennet: skb->len = %u, too big for wire format\n",
563 slots = DIV_ROUND_UP(offset + len, PAGE_SIZE) +
564 xennet_count_skb_frag_slots(skb);
565 if (unlikely(slots > MAX_SKB_FRAGS + 1)) {
566 net_alert_ratelimited(
567 "xennet: skb rides the rocket: %d slots\n", slots);
571 spin_lock_irqsave(&np->tx_lock, flags);
573 if (unlikely(!netif_carrier_ok(dev) ||
574 (slots > 1 && !xennet_can_sg(dev)) ||
575 netif_needs_gso(skb, netif_skb_features(skb)))) {
576 spin_unlock_irqrestore(&np->tx_lock, flags);
580 i = np->tx.req_prod_pvt;
582 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
583 np->tx_skbs[id].skb = skb;
585 tx = RING_GET_REQUEST(&np->tx, i);
588 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
589 BUG_ON((signed short)ref < 0);
590 mfn = virt_to_mfn(data);
591 gnttab_grant_foreign_access_ref(
592 ref, np->xbdev->otherend_id, mfn, GNTMAP_readonly);
593 np->grant_tx_page[id] = virt_to_page(data);
594 tx->gref = np->grant_tx_ref[id] = ref;
599 if (skb->ip_summed == CHECKSUM_PARTIAL)
601 tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
602 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
603 /* remote but checksummed. */
604 tx->flags |= XEN_NETTXF_data_validated;
606 if (skb_shinfo(skb)->gso_size) {
607 struct xen_netif_extra_info *gso;
609 gso = (struct xen_netif_extra_info *)
610 RING_GET_REQUEST(&np->tx, ++i);
612 tx->flags |= XEN_NETTXF_extra_info;
614 gso->u.gso.size = skb_shinfo(skb)->gso_size;
615 gso->u.gso.type = XEN_NETIF_GSO_TYPE_TCPV4;
617 gso->u.gso.features = 0;
619 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
623 np->tx.req_prod_pvt = i + 1;
625 xennet_make_frags(skb, dev, tx);
628 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->tx, notify);
630 notify_remote_via_irq(np->netdev->irq);
632 u64_stats_update_begin(&stats->syncp);
633 stats->tx_bytes += skb->len;
635 u64_stats_update_end(&stats->syncp);
637 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
638 xennet_tx_buf_gc(dev);
640 if (!netfront_tx_slot_available(np))
641 netif_stop_queue(dev);
643 spin_unlock_irqrestore(&np->tx_lock, flags);
648 dev->stats.tx_dropped++;
653 static int xennet_close(struct net_device *dev)
655 struct netfront_info *np = netdev_priv(dev);
656 netif_stop_queue(np->netdev);
657 napi_disable(&np->napi);
661 static void xennet_move_rx_slot(struct netfront_info *np, struct sk_buff *skb,
664 int new = xennet_rxidx(np->rx.req_prod_pvt);
666 BUG_ON(np->rx_skbs[new]);
667 np->rx_skbs[new] = skb;
668 np->grant_rx_ref[new] = ref;
669 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new;
670 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref;
671 np->rx.req_prod_pvt++;
674 static int xennet_get_extras(struct netfront_info *np,
675 struct xen_netif_extra_info *extras,
679 struct xen_netif_extra_info *extra;
680 struct device *dev = &np->netdev->dev;
681 RING_IDX cons = np->rx.rsp_cons;
688 if (unlikely(cons + 1 == rp)) {
690 dev_warn(dev, "Missing extra info\n");
695 extra = (struct xen_netif_extra_info *)
696 RING_GET_RESPONSE(&np->rx, ++cons);
698 if (unlikely(!extra->type ||
699 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
701 dev_warn(dev, "Invalid extra type: %d\n",
705 memcpy(&extras[extra->type - 1], extra,
709 skb = xennet_get_rx_skb(np, cons);
710 ref = xennet_get_rx_ref(np, cons);
711 xennet_move_rx_slot(np, skb, ref);
712 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
714 np->rx.rsp_cons = cons;
718 static int xennet_get_responses(struct netfront_info *np,
719 struct netfront_rx_info *rinfo, RING_IDX rp,
720 struct sk_buff_head *list)
722 struct xen_netif_rx_response *rx = &rinfo->rx;
723 struct xen_netif_extra_info *extras = rinfo->extras;
724 struct device *dev = &np->netdev->dev;
725 RING_IDX cons = np->rx.rsp_cons;
726 struct sk_buff *skb = xennet_get_rx_skb(np, cons);
727 grant_ref_t ref = xennet_get_rx_ref(np, cons);
728 int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD);
733 if (rx->flags & XEN_NETRXF_extra_info) {
734 err = xennet_get_extras(np, extras, rp);
735 cons = np->rx.rsp_cons;
739 if (unlikely(rx->status < 0 ||
740 rx->offset + rx->status > PAGE_SIZE)) {
742 dev_warn(dev, "rx->offset: %x, size: %u\n",
743 rx->offset, rx->status);
744 xennet_move_rx_slot(np, skb, ref);
750 * This definitely indicates a bug, either in this driver or in
751 * the backend driver. In future this should flag the bad
752 * situation to the system controller to reboot the backend.
754 if (ref == GRANT_INVALID_REF) {
756 dev_warn(dev, "Bad rx response id %d.\n",
762 ret = gnttab_end_foreign_access_ref(ref, 0);
765 gnttab_release_grant_reference(&np->gref_rx_head, ref);
767 __skb_queue_tail(list, skb);
770 if (!(rx->flags & XEN_NETRXF_more_data))
773 if (cons + slots == rp) {
775 dev_warn(dev, "Need more slots\n");
780 rx = RING_GET_RESPONSE(&np->rx, cons + slots);
781 skb = xennet_get_rx_skb(np, cons + slots);
782 ref = xennet_get_rx_ref(np, cons + slots);
786 if (unlikely(slots > max)) {
788 dev_warn(dev, "Too many slots\n");
793 np->rx.rsp_cons = cons + slots;
798 static int xennet_set_skb_gso(struct sk_buff *skb,
799 struct xen_netif_extra_info *gso)
801 if (!gso->u.gso.size) {
803 printk(KERN_WARNING "GSO size must not be zero.\n");
807 /* Currently only TCPv4 S.O. is supported. */
808 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4) {
810 printk(KERN_WARNING "Bad GSO type %d.\n", gso->u.gso.type);
814 skb_shinfo(skb)->gso_size = gso->u.gso.size;
815 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
817 /* Header must be checked, and gso_segs computed. */
818 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
819 skb_shinfo(skb)->gso_segs = 0;
824 static RING_IDX xennet_fill_frags(struct netfront_info *np,
826 struct sk_buff_head *list)
828 struct skb_shared_info *shinfo = skb_shinfo(skb);
829 RING_IDX cons = np->rx.rsp_cons;
830 struct sk_buff *nskb;
832 while ((nskb = __skb_dequeue(list))) {
833 struct xen_netif_rx_response *rx =
834 RING_GET_RESPONSE(&np->rx, ++cons);
835 skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
837 if (shinfo->nr_frags == MAX_SKB_FRAGS) {
838 unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
840 BUG_ON(pull_to <= skb_headlen(skb));
841 __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
843 BUG_ON(shinfo->nr_frags >= MAX_SKB_FRAGS);
845 skb_add_rx_frag(skb, shinfo->nr_frags, skb_frag_page(nfrag),
846 rx->offset, rx->status, PAGE_SIZE);
848 skb_shinfo(nskb)->nr_frags = 0;
855 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
860 int recalculate_partial_csum = 0;
863 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
864 * peers can fail to set NETRXF_csum_blank when sending a GSO
865 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
866 * recalculate the partial checksum.
868 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
869 struct netfront_info *np = netdev_priv(dev);
870 np->rx_gso_checksum_fixup++;
871 skb->ip_summed = CHECKSUM_PARTIAL;
872 recalculate_partial_csum = 1;
875 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
876 if (skb->ip_summed != CHECKSUM_PARTIAL)
879 if (skb->protocol != htons(ETH_P_IP))
882 iph = (void *)skb->data;
883 th = skb->data + 4 * iph->ihl;
884 if (th >= skb_tail_pointer(skb))
887 skb->csum_start = th - skb->head;
888 switch (iph->protocol) {
890 skb->csum_offset = offsetof(struct tcphdr, check);
892 if (recalculate_partial_csum) {
893 struct tcphdr *tcph = (struct tcphdr *)th;
894 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
895 skb->len - iph->ihl*4,
900 skb->csum_offset = offsetof(struct udphdr, check);
902 if (recalculate_partial_csum) {
903 struct udphdr *udph = (struct udphdr *)th;
904 udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
905 skb->len - iph->ihl*4,
911 printk(KERN_ERR "Attempting to checksum a non-"
912 "TCP/UDP packet, dropping a protocol"
913 " %d packet", iph->protocol);
917 if ((th + skb->csum_offset + 2) > skb_tail_pointer(skb))
926 static int handle_incoming_queue(struct net_device *dev,
927 struct sk_buff_head *rxq)
929 struct netfront_info *np = netdev_priv(dev);
930 struct netfront_stats *stats = this_cpu_ptr(np->stats);
931 int packets_dropped = 0;
934 while ((skb = __skb_dequeue(rxq)) != NULL) {
935 int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
937 if (pull_to > skb_headlen(skb))
938 __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
940 /* Ethernet work: Delayed to here as it peeks the header. */
941 skb->protocol = eth_type_trans(skb, dev);
943 if (checksum_setup(dev, skb)) {
946 dev->stats.rx_errors++;
950 u64_stats_update_begin(&stats->syncp);
952 stats->rx_bytes += skb->len;
953 u64_stats_update_end(&stats->syncp);
956 netif_receive_skb(skb);
959 return packets_dropped;
962 static int xennet_poll(struct napi_struct *napi, int budget)
964 struct netfront_info *np = container_of(napi, struct netfront_info, napi);
965 struct net_device *dev = np->netdev;
967 struct netfront_rx_info rinfo;
968 struct xen_netif_rx_response *rx = &rinfo.rx;
969 struct xen_netif_extra_info *extras = rinfo.extras;
972 struct sk_buff_head rxq;
973 struct sk_buff_head errq;
974 struct sk_buff_head tmpq;
978 spin_lock(&np->rx_lock);
980 skb_queue_head_init(&rxq);
981 skb_queue_head_init(&errq);
982 skb_queue_head_init(&tmpq);
984 rp = np->rx.sring->rsp_prod;
985 rmb(); /* Ensure we see queued responses up to 'rp'. */
989 while ((i != rp) && (work_done < budget)) {
990 memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx));
991 memset(extras, 0, sizeof(rinfo.extras));
993 err = xennet_get_responses(np, &rinfo, rp, &tmpq);
997 while ((skb = __skb_dequeue(&tmpq)))
998 __skb_queue_tail(&errq, skb);
999 dev->stats.rx_errors++;
1000 i = np->rx.rsp_cons;
1004 skb = __skb_dequeue(&tmpq);
1006 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1007 struct xen_netif_extra_info *gso;
1008 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1010 if (unlikely(xennet_set_skb_gso(skb, gso))) {
1011 __skb_queue_head(&tmpq, skb);
1012 np->rx.rsp_cons += skb_queue_len(&tmpq);
1017 NETFRONT_SKB_CB(skb)->pull_to = rx->status;
1018 if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
1019 NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
1021 skb_shinfo(skb)->frags[0].page_offset = rx->offset;
1022 skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
1023 skb->data_len = rx->status;
1024 skb->len += rx->status;
1026 i = xennet_fill_frags(np, skb, &tmpq);
1028 if (rx->flags & XEN_NETRXF_csum_blank)
1029 skb->ip_summed = CHECKSUM_PARTIAL;
1030 else if (rx->flags & XEN_NETRXF_data_validated)
1031 skb->ip_summed = CHECKSUM_UNNECESSARY;
1033 __skb_queue_tail(&rxq, skb);
1035 np->rx.rsp_cons = ++i;
1039 __skb_queue_purge(&errq);
1041 work_done -= handle_incoming_queue(dev, &rxq);
1043 /* If we get a callback with very few responses, reduce fill target. */
1044 /* NB. Note exponential increase, linear decrease. */
1045 if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) >
1046 ((3*np->rx_target) / 4)) &&
1047 (--np->rx_target < np->rx_min_target))
1048 np->rx_target = np->rx_min_target;
1050 xennet_alloc_rx_buffers(dev);
1052 if (work_done < budget) {
1055 local_irq_save(flags);
1057 RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, more_to_do);
1059 __napi_complete(napi);
1061 local_irq_restore(flags);
1064 spin_unlock(&np->rx_lock);
1069 static int xennet_change_mtu(struct net_device *dev, int mtu)
1071 int max = xennet_can_sg(dev) ?
1072 XEN_NETIF_MAX_TX_SIZE - MAX_TCP_HEADER : ETH_DATA_LEN;
1080 static struct rtnl_link_stats64 *xennet_get_stats64(struct net_device *dev,
1081 struct rtnl_link_stats64 *tot)
1083 struct netfront_info *np = netdev_priv(dev);
1086 for_each_possible_cpu(cpu) {
1087 struct netfront_stats *stats = per_cpu_ptr(np->stats, cpu);
1088 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1092 start = u64_stats_fetch_begin_bh(&stats->syncp);
1094 rx_packets = stats->rx_packets;
1095 tx_packets = stats->tx_packets;
1096 rx_bytes = stats->rx_bytes;
1097 tx_bytes = stats->tx_bytes;
1098 } while (u64_stats_fetch_retry_bh(&stats->syncp, start));
1100 tot->rx_packets += rx_packets;
1101 tot->tx_packets += tx_packets;
1102 tot->rx_bytes += rx_bytes;
1103 tot->tx_bytes += tx_bytes;
1106 tot->rx_errors = dev->stats.rx_errors;
1107 tot->tx_dropped = dev->stats.tx_dropped;
1112 static void xennet_release_tx_bufs(struct netfront_info *np)
1114 struct sk_buff *skb;
1117 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1118 /* Skip over entries which are actually freelist references */
1119 if (skb_entry_is_link(&np->tx_skbs[i]))
1122 skb = np->tx_skbs[i].skb;
1123 get_page(np->grant_tx_page[i]);
1124 gnttab_end_foreign_access(np->grant_tx_ref[i],
1126 (unsigned long)page_address(np->grant_tx_page[i]));
1127 np->grant_tx_page[i] = NULL;
1128 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1129 add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, i);
1130 dev_kfree_skb_irq(skb);
1134 static void xennet_release_rx_bufs(struct netfront_info *np)
1138 spin_lock_bh(&np->rx_lock);
1140 for (id = 0; id < NET_RX_RING_SIZE; id++) {
1141 struct sk_buff *skb;
1144 skb = np->rx_skbs[id];
1148 ref = np->grant_rx_ref[id];
1149 if (ref == GRANT_INVALID_REF)
1152 page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
1154 /* gnttab_end_foreign_access() needs a page ref until
1155 * foreign access is ended (which may be deferred).
1158 gnttab_end_foreign_access(ref, 0,
1159 (unsigned long)page_address(page));
1160 np->grant_rx_ref[id] = GRANT_INVALID_REF;
1165 spin_unlock_bh(&np->rx_lock);
1168 static void xennet_uninit(struct net_device *dev)
1170 struct netfront_info *np = netdev_priv(dev);
1171 xennet_release_tx_bufs(np);
1172 xennet_release_rx_bufs(np);
1173 gnttab_free_grant_references(np->gref_tx_head);
1174 gnttab_free_grant_references(np->gref_rx_head);
1177 static netdev_features_t xennet_fix_features(struct net_device *dev,
1178 netdev_features_t features)
1180 struct netfront_info *np = netdev_priv(dev);
1183 if (features & NETIF_F_SG) {
1184 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg",
1189 features &= ~NETIF_F_SG;
1192 if (features & NETIF_F_TSO) {
1193 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1194 "feature-gso-tcpv4", "%d", &val) < 0)
1198 features &= ~NETIF_F_TSO;
1204 static int xennet_set_features(struct net_device *dev,
1205 netdev_features_t features)
1207 if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1208 netdev_info(dev, "Reducing MTU because no SG offload");
1209 dev->mtu = ETH_DATA_LEN;
1215 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1217 struct net_device *dev = dev_id;
1218 struct netfront_info *np = netdev_priv(dev);
1219 unsigned long flags;
1221 spin_lock_irqsave(&np->tx_lock, flags);
1223 if (likely(netif_carrier_ok(dev))) {
1224 xennet_tx_buf_gc(dev);
1225 /* Under tx_lock: protects access to rx shared-ring indexes. */
1226 if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx))
1227 napi_schedule(&np->napi);
1230 spin_unlock_irqrestore(&np->tx_lock, flags);
1235 #ifdef CONFIG_NET_POLL_CONTROLLER
1236 static void xennet_poll_controller(struct net_device *dev)
1238 xennet_interrupt(0, dev);
1242 static const struct net_device_ops xennet_netdev_ops = {
1243 .ndo_open = xennet_open,
1244 .ndo_uninit = xennet_uninit,
1245 .ndo_stop = xennet_close,
1246 .ndo_start_xmit = xennet_start_xmit,
1247 .ndo_change_mtu = xennet_change_mtu,
1248 .ndo_get_stats64 = xennet_get_stats64,
1249 .ndo_set_mac_address = eth_mac_addr,
1250 .ndo_validate_addr = eth_validate_addr,
1251 .ndo_fix_features = xennet_fix_features,
1252 .ndo_set_features = xennet_set_features,
1253 #ifdef CONFIG_NET_POLL_CONTROLLER
1254 .ndo_poll_controller = xennet_poll_controller,
1258 static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1261 struct net_device *netdev;
1262 struct netfront_info *np;
1264 netdev = alloc_etherdev(sizeof(struct netfront_info));
1266 return ERR_PTR(-ENOMEM);
1268 np = netdev_priv(netdev);
1271 spin_lock_init(&np->tx_lock);
1272 spin_lock_init(&np->rx_lock);
1274 skb_queue_head_init(&np->rx_batch);
1275 np->rx_target = RX_DFL_MIN_TARGET;
1276 np->rx_min_target = RX_DFL_MIN_TARGET;
1277 np->rx_max_target = RX_MAX_TARGET;
1279 init_timer(&np->rx_refill_timer);
1280 np->rx_refill_timer.data = (unsigned long)netdev;
1281 np->rx_refill_timer.function = rx_refill_timeout;
1284 np->stats = alloc_percpu(struct netfront_stats);
1285 if (np->stats == NULL)
1288 /* Initialise tx_skbs as a free chain containing every entry. */
1289 np->tx_skb_freelist = 0;
1290 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1291 skb_entry_set_link(&np->tx_skbs[i], i+1);
1292 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1295 /* Clear out rx_skbs */
1296 for (i = 0; i < NET_RX_RING_SIZE; i++) {
1297 np->rx_skbs[i] = NULL;
1298 np->grant_rx_ref[i] = GRANT_INVALID_REF;
1299 np->grant_tx_page[i] = NULL;
1302 /* A grant for every tx ring slot */
1303 if (gnttab_alloc_grant_references(TX_MAX_TARGET,
1304 &np->gref_tx_head) < 0) {
1305 printk(KERN_ALERT "#### netfront can't alloc tx grant refs\n");
1307 goto exit_free_stats;
1309 /* A grant for every rx ring slot */
1310 if (gnttab_alloc_grant_references(RX_MAX_TARGET,
1311 &np->gref_rx_head) < 0) {
1312 printk(KERN_ALERT "#### netfront can't alloc rx grant refs\n");
1317 netdev->netdev_ops = &xennet_netdev_ops;
1319 netif_napi_add(netdev, &np->napi, xennet_poll, 64);
1320 netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1322 netdev->hw_features = NETIF_F_IP_CSUM | NETIF_F_SG | NETIF_F_TSO;
1325 * Assume that all hw features are available for now. This set
1326 * will be adjusted by the call to netdev_update_features() in
1327 * xennet_connect() which is the earliest point where we can
1328 * negotiate with the backend regarding supported features.
1330 netdev->features |= netdev->hw_features;
1332 SET_ETHTOOL_OPS(netdev, &xennet_ethtool_ops);
1333 SET_NETDEV_DEV(netdev, &dev->dev);
1335 netif_set_gso_max_size(netdev, XEN_NETIF_MAX_TX_SIZE - MAX_TCP_HEADER);
1337 np->netdev = netdev;
1339 netif_carrier_off(netdev);
1344 gnttab_free_grant_references(np->gref_tx_head);
1346 free_percpu(np->stats);
1348 free_netdev(netdev);
1349 return ERR_PTR(err);
1353 * Entry point to this code when a new device is created. Allocate the basic
1354 * structures and the ring buffers for communication with the backend, and
1355 * inform the backend of the appropriate details for those.
1357 static int netfront_probe(struct xenbus_device *dev,
1358 const struct xenbus_device_id *id)
1361 struct net_device *netdev;
1362 struct netfront_info *info;
1364 netdev = xennet_create_dev(dev);
1365 if (IS_ERR(netdev)) {
1366 err = PTR_ERR(netdev);
1367 xenbus_dev_fatal(dev, err, "creating netdev");
1371 info = netdev_priv(netdev);
1372 dev_set_drvdata(&dev->dev, info);
1374 err = register_netdev(info->netdev);
1376 printk(KERN_WARNING "%s: register_netdev err=%d\n",
1381 err = xennet_sysfs_addif(info->netdev);
1383 unregister_netdev(info->netdev);
1384 printk(KERN_WARNING "%s: add sysfs failed err=%d\n",
1392 free_netdev(netdev);
1393 dev_set_drvdata(&dev->dev, NULL);
1397 static void xennet_end_access(int ref, void *page)
1399 /* This frees the page as a side-effect */
1400 if (ref != GRANT_INVALID_REF)
1401 gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1404 static void xennet_disconnect_backend(struct netfront_info *info)
1406 /* Stop old i/f to prevent errors whilst we rebuild the state. */
1407 spin_lock_bh(&info->rx_lock);
1408 spin_lock_irq(&info->tx_lock);
1409 netif_carrier_off(info->netdev);
1410 spin_unlock_irq(&info->tx_lock);
1411 spin_unlock_bh(&info->rx_lock);
1413 if (info->netdev->irq)
1414 unbind_from_irqhandler(info->netdev->irq, info->netdev);
1415 info->evtchn = info->netdev->irq = 0;
1417 /* End access and free the pages */
1418 xennet_end_access(info->tx_ring_ref, info->tx.sring);
1419 xennet_end_access(info->rx_ring_ref, info->rx.sring);
1421 info->tx_ring_ref = GRANT_INVALID_REF;
1422 info->rx_ring_ref = GRANT_INVALID_REF;
1423 info->tx.sring = NULL;
1424 info->rx.sring = NULL;
1428 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1429 * driver restart. We tear down our netif structure and recreate it, but
1430 * leave the device-layer structures intact so that this is transparent to the
1431 * rest of the kernel.
1433 static int netfront_resume(struct xenbus_device *dev)
1435 struct netfront_info *info = dev_get_drvdata(&dev->dev);
1437 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1439 xennet_disconnect_backend(info);
1443 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1445 char *s, *e, *macstr;
1448 macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1450 return PTR_ERR(macstr);
1452 for (i = 0; i < ETH_ALEN; i++) {
1453 mac[i] = simple_strtoul(s, &e, 16);
1454 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1465 static int setup_netfront(struct xenbus_device *dev, struct netfront_info *info)
1467 struct xen_netif_tx_sring *txs;
1468 struct xen_netif_rx_sring *rxs;
1470 struct net_device *netdev = info->netdev;
1472 info->tx_ring_ref = GRANT_INVALID_REF;
1473 info->rx_ring_ref = GRANT_INVALID_REF;
1474 info->rx.sring = NULL;
1475 info->tx.sring = NULL;
1478 err = xen_net_read_mac(dev, netdev->dev_addr);
1480 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1484 txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1487 xenbus_dev_fatal(dev, err, "allocating tx ring page");
1490 SHARED_RING_INIT(txs);
1491 FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE);
1493 err = xenbus_grant_ring(dev, virt_to_mfn(txs));
1495 free_page((unsigned long)txs);
1499 info->tx_ring_ref = err;
1500 rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1503 xenbus_dev_fatal(dev, err, "allocating rx ring page");
1506 SHARED_RING_INIT(rxs);
1507 FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE);
1509 err = xenbus_grant_ring(dev, virt_to_mfn(rxs));
1511 free_page((unsigned long)rxs);
1514 info->rx_ring_ref = err;
1516 err = xenbus_alloc_evtchn(dev, &info->evtchn);
1520 err = bind_evtchn_to_irqhandler(info->evtchn, xennet_interrupt,
1521 0, netdev->name, netdev);
1531 /* Common code used when first setting up, and when resuming. */
1532 static int talk_to_netback(struct xenbus_device *dev,
1533 struct netfront_info *info)
1535 const char *message;
1536 struct xenbus_transaction xbt;
1539 /* Create shared ring, alloc event channel. */
1540 err = setup_netfront(dev, info);
1545 err = xenbus_transaction_start(&xbt);
1547 xenbus_dev_fatal(dev, err, "starting transaction");
1551 err = xenbus_printf(xbt, dev->nodename, "tx-ring-ref", "%u",
1554 message = "writing tx ring-ref";
1555 goto abort_transaction;
1557 err = xenbus_printf(xbt, dev->nodename, "rx-ring-ref", "%u",
1560 message = "writing rx ring-ref";
1561 goto abort_transaction;
1563 err = xenbus_printf(xbt, dev->nodename,
1564 "event-channel", "%u", info->evtchn);
1566 message = "writing event-channel";
1567 goto abort_transaction;
1570 err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1573 message = "writing request-rx-copy";
1574 goto abort_transaction;
1577 err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1579 message = "writing feature-rx-notify";
1580 goto abort_transaction;
1583 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1585 message = "writing feature-sg";
1586 goto abort_transaction;
1589 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1591 message = "writing feature-gso-tcpv4";
1592 goto abort_transaction;
1595 err = xenbus_transaction_end(xbt, 0);
1599 xenbus_dev_fatal(dev, err, "completing transaction");
1606 xenbus_transaction_end(xbt, 1);
1607 xenbus_dev_fatal(dev, err, "%s", message);
1609 xennet_disconnect_backend(info);
1614 static int xennet_connect(struct net_device *dev)
1616 struct netfront_info *np = netdev_priv(dev);
1617 int i, requeue_idx, err;
1618 struct sk_buff *skb;
1620 struct xen_netif_rx_request *req;
1621 unsigned int feature_rx_copy;
1623 err = xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1624 "feature-rx-copy", "%u", &feature_rx_copy);
1626 feature_rx_copy = 0;
1628 if (!feature_rx_copy) {
1630 "backend does not support copying receive path\n");
1634 err = talk_to_netback(np->xbdev, np);
1639 netdev_update_features(dev);
1642 spin_lock_bh(&np->rx_lock);
1643 spin_lock_irq(&np->tx_lock);
1645 /* Step 1: Discard all pending TX packet fragments. */
1646 xennet_release_tx_bufs(np);
1648 /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1649 for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) {
1651 const struct page *page;
1652 if (!np->rx_skbs[i])
1655 skb = np->rx_skbs[requeue_idx] = xennet_get_rx_skb(np, i);
1656 ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i);
1657 req = RING_GET_REQUEST(&np->rx, requeue_idx);
1659 frag = &skb_shinfo(skb)->frags[0];
1660 page = skb_frag_page(frag);
1661 gnttab_grant_foreign_access_ref(
1662 ref, np->xbdev->otherend_id,
1663 pfn_to_mfn(page_to_pfn(page)),
1666 req->id = requeue_idx;
1671 np->rx.req_prod_pvt = requeue_idx;
1674 * Step 3: All public and private state should now be sane. Get
1675 * ready to start sending and receiving packets and give the driver
1676 * domain a kick because we've probably just requeued some
1679 netif_carrier_on(np->netdev);
1680 notify_remote_via_irq(np->netdev->irq);
1681 xennet_tx_buf_gc(dev);
1682 xennet_alloc_rx_buffers(dev);
1684 spin_unlock_irq(&np->tx_lock);
1685 spin_unlock_bh(&np->rx_lock);
1691 * Callback received when the backend's state changes.
1693 static void netback_changed(struct xenbus_device *dev,
1694 enum xenbus_state backend_state)
1696 struct netfront_info *np = dev_get_drvdata(&dev->dev);
1697 struct net_device *netdev = np->netdev;
1699 dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
1701 switch (backend_state) {
1702 case XenbusStateInitialising:
1703 case XenbusStateInitialised:
1704 case XenbusStateReconfiguring:
1705 case XenbusStateReconfigured:
1706 case XenbusStateUnknown:
1707 case XenbusStateClosed:
1710 case XenbusStateInitWait:
1711 if (dev->state != XenbusStateInitialising)
1713 if (xennet_connect(netdev) != 0)
1715 xenbus_switch_state(dev, XenbusStateConnected);
1718 case XenbusStateConnected:
1719 netdev_notify_peers(netdev);
1722 case XenbusStateClosing:
1723 xenbus_frontend_closed(dev);
1728 static const struct xennet_stat {
1729 char name[ETH_GSTRING_LEN];
1731 } xennet_stats[] = {
1733 "rx_gso_checksum_fixup",
1734 offsetof(struct netfront_info, rx_gso_checksum_fixup)
1738 static int xennet_get_sset_count(struct net_device *dev, int string_set)
1740 switch (string_set) {
1742 return ARRAY_SIZE(xennet_stats);
1748 static void xennet_get_ethtool_stats(struct net_device *dev,
1749 struct ethtool_stats *stats, u64 * data)
1751 void *np = netdev_priv(dev);
1754 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
1755 data[i] = *(unsigned long *)(np + xennet_stats[i].offset);
1758 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
1762 switch (stringset) {
1764 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
1765 memcpy(data + i * ETH_GSTRING_LEN,
1766 xennet_stats[i].name, ETH_GSTRING_LEN);
1771 static const struct ethtool_ops xennet_ethtool_ops =
1773 .get_link = ethtool_op_get_link,
1775 .get_sset_count = xennet_get_sset_count,
1776 .get_ethtool_stats = xennet_get_ethtool_stats,
1777 .get_strings = xennet_get_strings,
1781 static ssize_t show_rxbuf_min(struct device *dev,
1782 struct device_attribute *attr, char *buf)
1784 struct net_device *netdev = to_net_dev(dev);
1785 struct netfront_info *info = netdev_priv(netdev);
1787 return sprintf(buf, "%u\n", info->rx_min_target);
1790 static ssize_t store_rxbuf_min(struct device *dev,
1791 struct device_attribute *attr,
1792 const char *buf, size_t len)
1794 struct net_device *netdev = to_net_dev(dev);
1795 struct netfront_info *np = netdev_priv(netdev);
1797 unsigned long target;
1799 if (!capable(CAP_NET_ADMIN))
1802 target = simple_strtoul(buf, &endp, 0);
1806 if (target < RX_MIN_TARGET)
1807 target = RX_MIN_TARGET;
1808 if (target > RX_MAX_TARGET)
1809 target = RX_MAX_TARGET;
1811 spin_lock_bh(&np->rx_lock);
1812 if (target > np->rx_max_target)
1813 np->rx_max_target = target;
1814 np->rx_min_target = target;
1815 if (target > np->rx_target)
1816 np->rx_target = target;
1818 xennet_alloc_rx_buffers(netdev);
1820 spin_unlock_bh(&np->rx_lock);
1824 static ssize_t show_rxbuf_max(struct device *dev,
1825 struct device_attribute *attr, char *buf)
1827 struct net_device *netdev = to_net_dev(dev);
1828 struct netfront_info *info = netdev_priv(netdev);
1830 return sprintf(buf, "%u\n", info->rx_max_target);
1833 static ssize_t store_rxbuf_max(struct device *dev,
1834 struct device_attribute *attr,
1835 const char *buf, size_t len)
1837 struct net_device *netdev = to_net_dev(dev);
1838 struct netfront_info *np = netdev_priv(netdev);
1840 unsigned long target;
1842 if (!capable(CAP_NET_ADMIN))
1845 target = simple_strtoul(buf, &endp, 0);
1849 if (target < RX_MIN_TARGET)
1850 target = RX_MIN_TARGET;
1851 if (target > RX_MAX_TARGET)
1852 target = RX_MAX_TARGET;
1854 spin_lock_bh(&np->rx_lock);
1855 if (target < np->rx_min_target)
1856 np->rx_min_target = target;
1857 np->rx_max_target = target;
1858 if (target < np->rx_target)
1859 np->rx_target = target;
1861 xennet_alloc_rx_buffers(netdev);
1863 spin_unlock_bh(&np->rx_lock);
1867 static ssize_t show_rxbuf_cur(struct device *dev,
1868 struct device_attribute *attr, char *buf)
1870 struct net_device *netdev = to_net_dev(dev);
1871 struct netfront_info *info = netdev_priv(netdev);
1873 return sprintf(buf, "%u\n", info->rx_target);
1876 static struct device_attribute xennet_attrs[] = {
1877 __ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf_min, store_rxbuf_min),
1878 __ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf_max, store_rxbuf_max),
1879 __ATTR(rxbuf_cur, S_IRUGO, show_rxbuf_cur, NULL),
1882 static int xennet_sysfs_addif(struct net_device *netdev)
1887 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) {
1888 err = device_create_file(&netdev->dev,
1897 device_remove_file(&netdev->dev, &xennet_attrs[i]);
1901 static void xennet_sysfs_delif(struct net_device *netdev)
1905 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++)
1906 device_remove_file(&netdev->dev, &xennet_attrs[i]);
1909 #endif /* CONFIG_SYSFS */
1911 static const struct xenbus_device_id netfront_ids[] = {
1917 static int xennet_remove(struct xenbus_device *dev)
1919 struct netfront_info *info = dev_get_drvdata(&dev->dev);
1921 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1923 xennet_disconnect_backend(info);
1925 xennet_sysfs_delif(info->netdev);
1927 unregister_netdev(info->netdev);
1929 del_timer_sync(&info->rx_refill_timer);
1931 free_percpu(info->stats);
1933 free_netdev(info->netdev);
1938 static DEFINE_XENBUS_DRIVER(netfront, ,
1939 .probe = netfront_probe,
1940 .remove = xennet_remove,
1941 .resume = netfront_resume,
1942 .otherend_changed = netback_changed,
1945 static int __init netif_init(void)
1950 if (xen_hvm_domain() && !xen_platform_pci_unplug)
1953 printk(KERN_INFO "Initialising Xen virtual ethernet driver.\n");
1955 return xenbus_register_frontend(&netfront_driver);
1957 module_init(netif_init);
1960 static void __exit netif_exit(void)
1962 xenbus_unregister_driver(&netfront_driver);
1964 module_exit(netif_exit);
1966 MODULE_DESCRIPTION("Xen virtual network device frontend");
1967 MODULE_LICENSE("GPL");
1968 MODULE_ALIAS("xen:vif");
1969 MODULE_ALIAS("xennet");