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>
47 #include <xen/xenbus.h>
48 #include <xen/events.h>
50 #include <xen/grant_table.h>
52 #include <xen/interface/io/netif.h>
53 #include <xen/interface/memory.h>
54 #include <xen/interface/grant_table.h>
56 static const struct ethtool_ops xennet_ethtool_ops;
63 #define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb))
65 #define RX_COPY_THRESHOLD 256
67 #define GRANT_INVALID_REF 0
69 #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, PAGE_SIZE)
70 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, PAGE_SIZE)
71 #define TX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256)
73 struct netfront_stats {
78 struct u64_stats_sync syncp;
81 struct netfront_info {
82 struct list_head list;
83 struct net_device *netdev;
85 struct napi_struct napi;
88 struct xenbus_device *xbdev;
91 struct xen_netif_tx_front_ring tx;
95 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
96 * are linked from tx_skb_freelist through skb_entry.link.
98 * NB. Freelist index entries are always going to be less than
99 * PAGE_OFFSET, whereas pointers to skbs will always be equal or
100 * greater than PAGE_OFFSET: we use this property to distinguish
106 } tx_skbs[NET_TX_RING_SIZE];
107 grant_ref_t gref_tx_head;
108 grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
109 unsigned tx_skb_freelist;
111 spinlock_t rx_lock ____cacheline_aligned_in_smp;
112 struct xen_netif_rx_front_ring rx;
115 /* Receive-ring batched refills. */
116 #define RX_MIN_TARGET 8
117 #define RX_DFL_MIN_TARGET 64
118 #define RX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256)
119 unsigned rx_min_target, rx_max_target, rx_target;
120 struct sk_buff_head rx_batch;
122 struct timer_list rx_refill_timer;
124 struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
125 grant_ref_t gref_rx_head;
126 grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
128 unsigned long rx_pfn_array[NET_RX_RING_SIZE];
129 struct multicall_entry rx_mcl[NET_RX_RING_SIZE+1];
130 struct mmu_update rx_mmu[NET_RX_RING_SIZE];
133 struct netfront_stats __percpu *stats;
135 unsigned long rx_gso_checksum_fixup;
138 struct netfront_rx_info {
139 struct xen_netif_rx_response rx;
140 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
143 static void skb_entry_set_link(union skb_entry *list, unsigned short id)
148 static int skb_entry_is_link(const union skb_entry *list)
150 BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link));
151 return (unsigned long)list->skb < PAGE_OFFSET;
155 * Access macros for acquiring freeing slots in tx_skbs[].
158 static void add_id_to_freelist(unsigned *head, union skb_entry *list,
161 skb_entry_set_link(&list[id], *head);
165 static unsigned short get_id_from_freelist(unsigned *head,
166 union skb_entry *list)
168 unsigned int id = *head;
169 *head = list[id].link;
173 static int xennet_rxidx(RING_IDX idx)
175 return idx & (NET_RX_RING_SIZE - 1);
178 static struct sk_buff *xennet_get_rx_skb(struct netfront_info *np,
181 int i = xennet_rxidx(ri);
182 struct sk_buff *skb = np->rx_skbs[i];
183 np->rx_skbs[i] = NULL;
187 static grant_ref_t xennet_get_rx_ref(struct netfront_info *np,
190 int i = xennet_rxidx(ri);
191 grant_ref_t ref = np->grant_rx_ref[i];
192 np->grant_rx_ref[i] = GRANT_INVALID_REF;
197 static int xennet_sysfs_addif(struct net_device *netdev);
198 static void xennet_sysfs_delif(struct net_device *netdev);
199 #else /* !CONFIG_SYSFS */
200 #define xennet_sysfs_addif(dev) (0)
201 #define xennet_sysfs_delif(dev) do { } while (0)
204 static int xennet_can_sg(struct net_device *dev)
206 return dev->features & NETIF_F_SG;
210 static void rx_refill_timeout(unsigned long data)
212 struct net_device *dev = (struct net_device *)data;
213 struct netfront_info *np = netdev_priv(dev);
214 napi_schedule(&np->napi);
217 static int netfront_tx_slot_available(struct netfront_info *np)
219 return (np->tx.req_prod_pvt - np->tx.rsp_cons) <
220 (TX_MAX_TARGET - MAX_SKB_FRAGS - 2);
223 static void xennet_maybe_wake_tx(struct net_device *dev)
225 struct netfront_info *np = netdev_priv(dev);
227 if (unlikely(netif_queue_stopped(dev)) &&
228 netfront_tx_slot_available(np) &&
229 likely(netif_running(dev)))
230 netif_wake_queue(dev);
233 static void xennet_alloc_rx_buffers(struct net_device *dev)
236 struct netfront_info *np = netdev_priv(dev);
239 int i, batch_target, notify;
240 RING_IDX req_prod = np->rx.req_prod_pvt;
244 struct xen_netif_rx_request *req;
246 if (unlikely(!netif_carrier_ok(dev)))
250 * Allocate skbuffs greedily, even though we batch updates to the
251 * receive ring. This creates a less bursty demand on the memory
252 * allocator, so should reduce the chance of failed allocation requests
253 * both for ourself and for other kernel subsystems.
255 batch_target = np->rx_target - (req_prod - np->rx.rsp_cons);
256 for (i = skb_queue_len(&np->rx_batch); i < batch_target; i++) {
257 skb = __netdev_alloc_skb(dev, RX_COPY_THRESHOLD + NET_IP_ALIGN,
258 GFP_ATOMIC | __GFP_NOWARN);
262 /* Align ip header to a 16 bytes boundary */
263 skb_reserve(skb, NET_IP_ALIGN);
265 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
269 /* Any skbuffs queued for refill? Force them out. */
272 /* Could not allocate any skbuffs. Try again later. */
273 mod_timer(&np->rx_refill_timer,
278 __skb_fill_page_desc(skb, 0, page, 0, 0);
279 skb_shinfo(skb)->nr_frags = 1;
280 __skb_queue_tail(&np->rx_batch, skb);
283 /* Is the batch large enough to be worthwhile? */
284 if (i < (np->rx_target/2)) {
285 if (req_prod > np->rx.sring->req_prod)
290 /* Adjust our fill target if we risked running out of buffers. */
291 if (((req_prod - np->rx.sring->rsp_prod) < (np->rx_target / 4)) &&
292 ((np->rx_target *= 2) > np->rx_max_target))
293 np->rx_target = np->rx_max_target;
297 skb = __skb_dequeue(&np->rx_batch);
303 id = xennet_rxidx(req_prod + i);
305 BUG_ON(np->rx_skbs[id]);
306 np->rx_skbs[id] = skb;
308 ref = gnttab_claim_grant_reference(&np->gref_rx_head);
309 BUG_ON((signed short)ref < 0);
310 np->grant_rx_ref[id] = ref;
312 pfn = page_to_pfn(skb_frag_page(&skb_shinfo(skb)->frags[0]));
313 vaddr = page_address(skb_frag_page(&skb_shinfo(skb)->frags[0]));
315 req = RING_GET_REQUEST(&np->rx, req_prod + i);
316 gnttab_grant_foreign_access_ref(ref,
317 np->xbdev->otherend_id,
325 wmb(); /* barrier so backend seens requests */
327 /* Above is a suitable barrier to ensure backend will see requests. */
328 np->rx.req_prod_pvt = req_prod + i;
330 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->rx, notify);
332 notify_remote_via_irq(np->netdev->irq);
335 static int xennet_open(struct net_device *dev)
337 struct netfront_info *np = netdev_priv(dev);
339 napi_enable(&np->napi);
341 spin_lock_bh(&np->rx_lock);
342 if (netif_carrier_ok(dev)) {
343 xennet_alloc_rx_buffers(dev);
344 np->rx.sring->rsp_event = np->rx.rsp_cons + 1;
345 if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx))
346 napi_schedule(&np->napi);
348 spin_unlock_bh(&np->rx_lock);
350 netif_start_queue(dev);
355 static void xennet_tx_buf_gc(struct net_device *dev)
359 struct netfront_info *np = netdev_priv(dev);
362 BUG_ON(!netif_carrier_ok(dev));
365 prod = np->tx.sring->rsp_prod;
366 rmb(); /* Ensure we see responses up to 'rp'. */
368 for (cons = np->tx.rsp_cons; cons != prod; cons++) {
369 struct xen_netif_tx_response *txrsp;
371 txrsp = RING_GET_RESPONSE(&np->tx, cons);
372 if (txrsp->status == XEN_NETIF_RSP_NULL)
376 skb = np->tx_skbs[id].skb;
377 if (unlikely(gnttab_query_foreign_access(
378 np->grant_tx_ref[id]) != 0)) {
379 printk(KERN_ALERT "xennet_tx_buf_gc: warning "
380 "-- grant still in use by backend "
384 gnttab_end_foreign_access_ref(
385 np->grant_tx_ref[id], GNTMAP_readonly);
386 gnttab_release_grant_reference(
387 &np->gref_tx_head, np->grant_tx_ref[id]);
388 np->grant_tx_ref[id] = GRANT_INVALID_REF;
389 add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, id);
390 dev_kfree_skb_irq(skb);
393 np->tx.rsp_cons = prod;
396 * Set a new event, then check for race with update of tx_cons.
397 * Note that it is essential to schedule a callback, no matter
398 * how few buffers are pending. Even if there is space in the
399 * transmit ring, higher layers may be blocked because too much
400 * data is outstanding: in such cases notification from Xen is
401 * likely to be the only kick that we'll get.
403 np->tx.sring->rsp_event =
404 prod + ((np->tx.sring->req_prod - prod) >> 1) + 1;
405 mb(); /* update shared area */
406 } while ((cons == prod) && (prod != np->tx.sring->rsp_prod));
408 xennet_maybe_wake_tx(dev);
411 static void xennet_make_frags(struct sk_buff *skb, struct net_device *dev,
412 struct xen_netif_tx_request *tx)
414 struct netfront_info *np = netdev_priv(dev);
415 char *data = skb->data;
417 RING_IDX prod = np->tx.req_prod_pvt;
418 int frags = skb_shinfo(skb)->nr_frags;
419 unsigned int offset = offset_in_page(data);
420 unsigned int len = skb_headlen(skb);
425 /* While the header overlaps a page boundary (including being
426 larger than a page), split it it into page-sized chunks. */
427 while (len > PAGE_SIZE - offset) {
428 tx->size = PAGE_SIZE - offset;
429 tx->flags |= XEN_NETTXF_more_data;
434 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
435 np->tx_skbs[id].skb = skb_get(skb);
436 tx = RING_GET_REQUEST(&np->tx, prod++);
438 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
439 BUG_ON((signed short)ref < 0);
441 mfn = virt_to_mfn(data);
442 gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id,
443 mfn, GNTMAP_readonly);
445 tx->gref = np->grant_tx_ref[id] = ref;
451 /* Grant backend access to each skb fragment page. */
452 for (i = 0; i < frags; i++) {
453 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
455 tx->flags |= XEN_NETTXF_more_data;
457 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
458 np->tx_skbs[id].skb = skb_get(skb);
459 tx = RING_GET_REQUEST(&np->tx, prod++);
461 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
462 BUG_ON((signed short)ref < 0);
464 mfn = pfn_to_mfn(page_to_pfn(skb_frag_page(frag)));
465 gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id,
466 mfn, GNTMAP_readonly);
468 tx->gref = np->grant_tx_ref[id] = ref;
469 tx->offset = frag->page_offset;
470 tx->size = skb_frag_size(frag);
474 np->tx.req_prod_pvt = prod;
477 static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
480 struct netfront_info *np = netdev_priv(dev);
481 struct netfront_stats *stats = this_cpu_ptr(np->stats);
482 struct xen_netif_tx_request *tx;
483 struct xen_netif_extra_info *extra;
484 char *data = skb->data;
489 int frags = skb_shinfo(skb)->nr_frags;
490 unsigned int offset = offset_in_page(data);
491 unsigned int len = skb_headlen(skb);
493 /* If skb->len is too big for wire format, drop skb and alert
494 * user about misconfiguration.
496 if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
497 net_alert_ratelimited(
498 "xennet: skb->len = %u, too big for wire format\n",
503 frags += DIV_ROUND_UP(offset + len, PAGE_SIZE);
504 if (unlikely(frags > MAX_SKB_FRAGS + 1)) {
505 printk(KERN_ALERT "xennet: skb rides the rocket: %d frags\n",
511 spin_lock_irq(&np->tx_lock);
513 if (unlikely(!netif_carrier_ok(dev) ||
514 (frags > 1 && !xennet_can_sg(dev)) ||
515 netif_needs_gso(skb, netif_skb_features(skb)))) {
516 spin_unlock_irq(&np->tx_lock);
520 i = np->tx.req_prod_pvt;
522 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
523 np->tx_skbs[id].skb = skb;
525 tx = RING_GET_REQUEST(&np->tx, i);
528 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
529 BUG_ON((signed short)ref < 0);
530 mfn = virt_to_mfn(data);
531 gnttab_grant_foreign_access_ref(
532 ref, np->xbdev->otherend_id, mfn, GNTMAP_readonly);
533 tx->gref = np->grant_tx_ref[id] = ref;
539 if (skb->ip_summed == CHECKSUM_PARTIAL)
541 tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
542 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
543 /* remote but checksummed. */
544 tx->flags |= XEN_NETTXF_data_validated;
546 if (skb_shinfo(skb)->gso_size) {
547 struct xen_netif_extra_info *gso;
549 gso = (struct xen_netif_extra_info *)
550 RING_GET_REQUEST(&np->tx, ++i);
553 extra->flags |= XEN_NETIF_EXTRA_FLAG_MORE;
555 tx->flags |= XEN_NETTXF_extra_info;
557 gso->u.gso.size = skb_shinfo(skb)->gso_size;
558 gso->u.gso.type = XEN_NETIF_GSO_TYPE_TCPV4;
560 gso->u.gso.features = 0;
562 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
567 np->tx.req_prod_pvt = i + 1;
569 xennet_make_frags(skb, dev, tx);
572 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->tx, notify);
574 notify_remote_via_irq(np->netdev->irq);
576 u64_stats_update_begin(&stats->syncp);
577 stats->tx_bytes += skb->len;
579 u64_stats_update_end(&stats->syncp);
581 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
582 xennet_tx_buf_gc(dev);
584 if (!netfront_tx_slot_available(np))
585 netif_stop_queue(dev);
587 spin_unlock_irq(&np->tx_lock);
592 dev->stats.tx_dropped++;
597 static int xennet_close(struct net_device *dev)
599 struct netfront_info *np = netdev_priv(dev);
600 netif_stop_queue(np->netdev);
601 napi_disable(&np->napi);
605 static void xennet_move_rx_slot(struct netfront_info *np, struct sk_buff *skb,
608 int new = xennet_rxidx(np->rx.req_prod_pvt);
610 BUG_ON(np->rx_skbs[new]);
611 np->rx_skbs[new] = skb;
612 np->grant_rx_ref[new] = ref;
613 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new;
614 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref;
615 np->rx.req_prod_pvt++;
618 static int xennet_get_extras(struct netfront_info *np,
619 struct xen_netif_extra_info *extras,
623 struct xen_netif_extra_info *extra;
624 struct device *dev = &np->netdev->dev;
625 RING_IDX cons = np->rx.rsp_cons;
632 if (unlikely(cons + 1 == rp)) {
634 dev_warn(dev, "Missing extra info\n");
639 extra = (struct xen_netif_extra_info *)
640 RING_GET_RESPONSE(&np->rx, ++cons);
642 if (unlikely(!extra->type ||
643 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
645 dev_warn(dev, "Invalid extra type: %d\n",
649 memcpy(&extras[extra->type - 1], extra,
653 skb = xennet_get_rx_skb(np, cons);
654 ref = xennet_get_rx_ref(np, cons);
655 xennet_move_rx_slot(np, skb, ref);
656 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
658 np->rx.rsp_cons = cons;
662 static int xennet_get_responses(struct netfront_info *np,
663 struct netfront_rx_info *rinfo, RING_IDX rp,
664 struct sk_buff_head *list)
666 struct xen_netif_rx_response *rx = &rinfo->rx;
667 struct xen_netif_extra_info *extras = rinfo->extras;
668 struct device *dev = &np->netdev->dev;
669 RING_IDX cons = np->rx.rsp_cons;
670 struct sk_buff *skb = xennet_get_rx_skb(np, cons);
671 grant_ref_t ref = xennet_get_rx_ref(np, cons);
672 int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD);
677 if (rx->flags & XEN_NETRXF_extra_info) {
678 err = xennet_get_extras(np, extras, rp);
679 cons = np->rx.rsp_cons;
683 if (unlikely(rx->status < 0 ||
684 rx->offset + rx->status > PAGE_SIZE)) {
686 dev_warn(dev, "rx->offset: %x, size: %u\n",
687 rx->offset, rx->status);
688 xennet_move_rx_slot(np, skb, ref);
694 * This definitely indicates a bug, either in this driver or in
695 * the backend driver. In future this should flag the bad
696 * situation to the system controller to reboot the backed.
698 if (ref == GRANT_INVALID_REF) {
700 dev_warn(dev, "Bad rx response id %d.\n",
706 ret = gnttab_end_foreign_access_ref(ref, 0);
709 gnttab_release_grant_reference(&np->gref_rx_head, ref);
711 __skb_queue_tail(list, skb);
714 if (!(rx->flags & XEN_NETRXF_more_data))
717 if (cons + frags == rp) {
719 dev_warn(dev, "Need more frags\n");
724 rx = RING_GET_RESPONSE(&np->rx, cons + frags);
725 skb = xennet_get_rx_skb(np, cons + frags);
726 ref = xennet_get_rx_ref(np, cons + frags);
730 if (unlikely(frags > max)) {
732 dev_warn(dev, "Too many frags\n");
737 np->rx.rsp_cons = cons + frags;
742 static int xennet_set_skb_gso(struct sk_buff *skb,
743 struct xen_netif_extra_info *gso)
745 if (!gso->u.gso.size) {
747 printk(KERN_WARNING "GSO size must not be zero.\n");
751 /* Currently only TCPv4 S.O. is supported. */
752 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4) {
754 printk(KERN_WARNING "Bad GSO type %d.\n", gso->u.gso.type);
758 skb_shinfo(skb)->gso_size = gso->u.gso.size;
759 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
761 /* Header must be checked, and gso_segs computed. */
762 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
763 skb_shinfo(skb)->gso_segs = 0;
768 static RING_IDX xennet_fill_frags(struct netfront_info *np,
770 struct sk_buff_head *list)
772 struct skb_shared_info *shinfo = skb_shinfo(skb);
773 int nr_frags = shinfo->nr_frags;
774 RING_IDX cons = np->rx.rsp_cons;
775 struct sk_buff *nskb;
777 while ((nskb = __skb_dequeue(list))) {
778 struct xen_netif_rx_response *rx =
779 RING_GET_RESPONSE(&np->rx, ++cons);
780 skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
782 __skb_fill_page_desc(skb, nr_frags,
783 skb_frag_page(nfrag),
784 rx->offset, rx->status);
786 skb->data_len += rx->status;
788 skb_shinfo(nskb)->nr_frags = 0;
794 shinfo->nr_frags = nr_frags;
798 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
803 int recalculate_partial_csum = 0;
806 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
807 * peers can fail to set NETRXF_csum_blank when sending a GSO
808 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
809 * recalculate the partial checksum.
811 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
812 struct netfront_info *np = netdev_priv(dev);
813 np->rx_gso_checksum_fixup++;
814 skb->ip_summed = CHECKSUM_PARTIAL;
815 recalculate_partial_csum = 1;
818 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
819 if (skb->ip_summed != CHECKSUM_PARTIAL)
822 if (skb->protocol != htons(ETH_P_IP))
825 iph = (void *)skb->data;
826 th = skb->data + 4 * iph->ihl;
827 if (th >= skb_tail_pointer(skb))
830 skb->csum_start = th - skb->head;
831 switch (iph->protocol) {
833 skb->csum_offset = offsetof(struct tcphdr, check);
835 if (recalculate_partial_csum) {
836 struct tcphdr *tcph = (struct tcphdr *)th;
837 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
838 skb->len - iph->ihl*4,
843 skb->csum_offset = offsetof(struct udphdr, check);
845 if (recalculate_partial_csum) {
846 struct udphdr *udph = (struct udphdr *)th;
847 udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
848 skb->len - iph->ihl*4,
854 printk(KERN_ERR "Attempting to checksum a non-"
855 "TCP/UDP packet, dropping a protocol"
856 " %d packet", iph->protocol);
860 if ((th + skb->csum_offset + 2) > skb_tail_pointer(skb))
869 static int handle_incoming_queue(struct net_device *dev,
870 struct sk_buff_head *rxq)
872 struct netfront_info *np = netdev_priv(dev);
873 struct netfront_stats *stats = this_cpu_ptr(np->stats);
874 int packets_dropped = 0;
877 while ((skb = __skb_dequeue(rxq)) != NULL) {
878 struct page *page = NETFRONT_SKB_CB(skb)->page;
879 void *vaddr = page_address(page);
880 unsigned offset = NETFRONT_SKB_CB(skb)->offset;
882 memcpy(skb->data, vaddr + offset,
885 if (page != skb_frag_page(&skb_shinfo(skb)->frags[0]))
888 /* Ethernet work: Delayed to here as it peeks the header. */
889 skb->protocol = eth_type_trans(skb, dev);
891 if (checksum_setup(dev, skb)) {
894 dev->stats.rx_errors++;
898 u64_stats_update_begin(&stats->syncp);
900 stats->rx_bytes += skb->len;
901 u64_stats_update_end(&stats->syncp);
904 netif_receive_skb(skb);
907 return packets_dropped;
910 static int xennet_poll(struct napi_struct *napi, int budget)
912 struct netfront_info *np = container_of(napi, struct netfront_info, napi);
913 struct net_device *dev = np->netdev;
915 struct netfront_rx_info rinfo;
916 struct xen_netif_rx_response *rx = &rinfo.rx;
917 struct xen_netif_extra_info *extras = rinfo.extras;
920 struct sk_buff_head rxq;
921 struct sk_buff_head errq;
922 struct sk_buff_head tmpq;
927 spin_lock(&np->rx_lock);
929 skb_queue_head_init(&rxq);
930 skb_queue_head_init(&errq);
931 skb_queue_head_init(&tmpq);
933 rp = np->rx.sring->rsp_prod;
934 rmb(); /* Ensure we see queued responses up to 'rp'. */
938 while ((i != rp) && (work_done < budget)) {
939 memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx));
940 memset(extras, 0, sizeof(rinfo.extras));
942 err = xennet_get_responses(np, &rinfo, rp, &tmpq);
946 while ((skb = __skb_dequeue(&tmpq)))
947 __skb_queue_tail(&errq, skb);
948 dev->stats.rx_errors++;
953 skb = __skb_dequeue(&tmpq);
955 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
956 struct xen_netif_extra_info *gso;
957 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
959 if (unlikely(xennet_set_skb_gso(skb, gso))) {
960 __skb_queue_head(&tmpq, skb);
961 np->rx.rsp_cons += skb_queue_len(&tmpq);
966 NETFRONT_SKB_CB(skb)->page =
967 skb_frag_page(&skb_shinfo(skb)->frags[0]);
968 NETFRONT_SKB_CB(skb)->offset = rx->offset;
971 if (len > RX_COPY_THRESHOLD)
972 len = RX_COPY_THRESHOLD;
975 if (rx->status > len) {
976 skb_shinfo(skb)->frags[0].page_offset =
978 skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status - len);
979 skb->data_len = rx->status - len;
981 __skb_fill_page_desc(skb, 0, NULL, 0, 0);
982 skb_shinfo(skb)->nr_frags = 0;
985 i = xennet_fill_frags(np, skb, &tmpq);
988 * Truesize approximates the size of true data plus
989 * any supervisor overheads. Adding hypervisor
990 * overheads has been shown to significantly reduce
991 * achievable bandwidth with the default receive
992 * buffer size. It is therefore not wise to account
995 * After alloc_skb(RX_COPY_THRESHOLD), truesize is set
996 * to RX_COPY_THRESHOLD + the supervisor
997 * overheads. Here, we add the size of the data pulled
998 * in xennet_fill_frags().
1000 * We also adjust for any unused space in the main
1001 * data area by subtracting (RX_COPY_THRESHOLD -
1002 * len). This is especially important with drivers
1003 * which split incoming packets into header and data,
1004 * using only 66 bytes of the main data area (see the
1005 * e1000 driver for example.) On such systems,
1006 * without this last adjustement, our achievable
1007 * receive throughout using the standard receive
1008 * buffer size was cut by 25%(!!!).
1010 skb->truesize += skb->data_len - (RX_COPY_THRESHOLD - len);
1011 skb->len += skb->data_len;
1013 if (rx->flags & XEN_NETRXF_csum_blank)
1014 skb->ip_summed = CHECKSUM_PARTIAL;
1015 else if (rx->flags & XEN_NETRXF_data_validated)
1016 skb->ip_summed = CHECKSUM_UNNECESSARY;
1018 __skb_queue_tail(&rxq, skb);
1020 np->rx.rsp_cons = ++i;
1024 __skb_queue_purge(&errq);
1026 work_done -= handle_incoming_queue(dev, &rxq);
1028 /* If we get a callback with very few responses, reduce fill target. */
1029 /* NB. Note exponential increase, linear decrease. */
1030 if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) >
1031 ((3*np->rx_target) / 4)) &&
1032 (--np->rx_target < np->rx_min_target))
1033 np->rx_target = np->rx_min_target;
1035 xennet_alloc_rx_buffers(dev);
1037 if (work_done < budget) {
1040 local_irq_save(flags);
1042 RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, more_to_do);
1044 __napi_complete(napi);
1046 local_irq_restore(flags);
1049 spin_unlock(&np->rx_lock);
1054 static int xennet_change_mtu(struct net_device *dev, int mtu)
1056 int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN;
1064 static struct rtnl_link_stats64 *xennet_get_stats64(struct net_device *dev,
1065 struct rtnl_link_stats64 *tot)
1067 struct netfront_info *np = netdev_priv(dev);
1070 for_each_possible_cpu(cpu) {
1071 struct netfront_stats *stats = per_cpu_ptr(np->stats, cpu);
1072 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1076 start = u64_stats_fetch_begin_bh(&stats->syncp);
1078 rx_packets = stats->rx_packets;
1079 tx_packets = stats->tx_packets;
1080 rx_bytes = stats->rx_bytes;
1081 tx_bytes = stats->tx_bytes;
1082 } while (u64_stats_fetch_retry_bh(&stats->syncp, start));
1084 tot->rx_packets += rx_packets;
1085 tot->tx_packets += tx_packets;
1086 tot->rx_bytes += rx_bytes;
1087 tot->tx_bytes += tx_bytes;
1090 tot->rx_errors = dev->stats.rx_errors;
1091 tot->tx_dropped = dev->stats.tx_dropped;
1096 static void xennet_release_tx_bufs(struct netfront_info *np)
1098 struct sk_buff *skb;
1101 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1102 /* Skip over entries which are actually freelist references */
1103 if (skb_entry_is_link(&np->tx_skbs[i]))
1106 skb = np->tx_skbs[i].skb;
1107 gnttab_end_foreign_access_ref(np->grant_tx_ref[i],
1109 gnttab_release_grant_reference(&np->gref_tx_head,
1110 np->grant_tx_ref[i]);
1111 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1112 add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, i);
1113 dev_kfree_skb_irq(skb);
1117 static void xennet_release_rx_bufs(struct netfront_info *np)
1119 struct mmu_update *mmu = np->rx_mmu;
1120 struct multicall_entry *mcl = np->rx_mcl;
1121 struct sk_buff_head free_list;
1122 struct sk_buff *skb;
1124 int xfer = 0, noxfer = 0, unused = 0;
1127 dev_warn(&np->netdev->dev, "%s: fix me for copying receiver.\n",
1131 skb_queue_head_init(&free_list);
1133 spin_lock_bh(&np->rx_lock);
1135 for (id = 0; id < NET_RX_RING_SIZE; id++) {
1136 ref = np->grant_rx_ref[id];
1137 if (ref == GRANT_INVALID_REF) {
1142 skb = np->rx_skbs[id];
1143 mfn = gnttab_end_foreign_transfer_ref(ref);
1144 gnttab_release_grant_reference(&np->gref_rx_head, ref);
1145 np->grant_rx_ref[id] = GRANT_INVALID_REF;
1148 skb_shinfo(skb)->nr_frags = 0;
1154 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1155 /* Remap the page. */
1156 const struct page *page =
1157 skb_frag_page(&skb_shinfo(skb)->frags[0]);
1158 unsigned long pfn = page_to_pfn(page);
1159 void *vaddr = page_address(page);
1161 MULTI_update_va_mapping(mcl, (unsigned long)vaddr,
1162 mfn_pte(mfn, PAGE_KERNEL),
1165 mmu->ptr = ((u64)mfn << PAGE_SHIFT)
1166 | MMU_MACHPHYS_UPDATE;
1170 set_phys_to_machine(pfn, mfn);
1172 __skb_queue_tail(&free_list, skb);
1176 dev_info(&np->netdev->dev, "%s: %d xfer, %d noxfer, %d unused\n",
1177 __func__, xfer, noxfer, unused);
1180 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1181 /* Do all the remapping work and M2P updates. */
1182 MULTI_mmu_update(mcl, np->rx_mmu, mmu - np->rx_mmu,
1185 HYPERVISOR_multicall(np->rx_mcl, mcl - np->rx_mcl);
1189 __skb_queue_purge(&free_list);
1191 spin_unlock_bh(&np->rx_lock);
1194 static void xennet_uninit(struct net_device *dev)
1196 struct netfront_info *np = netdev_priv(dev);
1197 xennet_release_tx_bufs(np);
1198 xennet_release_rx_bufs(np);
1199 gnttab_free_grant_references(np->gref_tx_head);
1200 gnttab_free_grant_references(np->gref_rx_head);
1203 static u32 xennet_fix_features(struct net_device *dev, u32 features)
1205 struct netfront_info *np = netdev_priv(dev);
1208 if (features & NETIF_F_SG) {
1209 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg",
1214 features &= ~NETIF_F_SG;
1217 if (features & NETIF_F_TSO) {
1218 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1219 "feature-gso-tcpv4", "%d", &val) < 0)
1223 features &= ~NETIF_F_TSO;
1229 static int xennet_set_features(struct net_device *dev, u32 features)
1231 if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1232 netdev_info(dev, "Reducing MTU because no SG offload");
1233 dev->mtu = ETH_DATA_LEN;
1239 static const struct net_device_ops xennet_netdev_ops = {
1240 .ndo_open = xennet_open,
1241 .ndo_uninit = xennet_uninit,
1242 .ndo_stop = xennet_close,
1243 .ndo_start_xmit = xennet_start_xmit,
1244 .ndo_change_mtu = xennet_change_mtu,
1245 .ndo_get_stats64 = xennet_get_stats64,
1246 .ndo_set_mac_address = eth_mac_addr,
1247 .ndo_validate_addr = eth_validate_addr,
1248 .ndo_fix_features = xennet_fix_features,
1249 .ndo_set_features = xennet_set_features,
1252 static struct net_device * __devinit xennet_create_dev(struct xenbus_device *dev)
1255 struct net_device *netdev;
1256 struct netfront_info *np;
1258 netdev = alloc_etherdev(sizeof(struct netfront_info));
1260 printk(KERN_WARNING "%s> alloc_etherdev failed.\n",
1262 return ERR_PTR(-ENOMEM);
1265 np = netdev_priv(netdev);
1268 spin_lock_init(&np->tx_lock);
1269 spin_lock_init(&np->rx_lock);
1271 skb_queue_head_init(&np->rx_batch);
1272 np->rx_target = RX_DFL_MIN_TARGET;
1273 np->rx_min_target = RX_DFL_MIN_TARGET;
1274 np->rx_max_target = RX_MAX_TARGET;
1276 init_timer(&np->rx_refill_timer);
1277 np->rx_refill_timer.data = (unsigned long)netdev;
1278 np->rx_refill_timer.function = rx_refill_timeout;
1281 np->stats = alloc_percpu(struct netfront_stats);
1282 if (np->stats == NULL)
1285 /* Initialise tx_skbs as a free chain containing every entry. */
1286 np->tx_skb_freelist = 0;
1287 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1288 skb_entry_set_link(&np->tx_skbs[i], i+1);
1289 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1292 /* Clear out rx_skbs */
1293 for (i = 0; i < NET_RX_RING_SIZE; i++) {
1294 np->rx_skbs[i] = NULL;
1295 np->grant_rx_ref[i] = GRANT_INVALID_REF;
1298 /* A grant for every tx ring slot */
1299 if (gnttab_alloc_grant_references(TX_MAX_TARGET,
1300 &np->gref_tx_head) < 0) {
1301 printk(KERN_ALERT "#### netfront can't alloc tx grant refs\n");
1303 goto exit_free_stats;
1305 /* A grant for every rx ring slot */
1306 if (gnttab_alloc_grant_references(RX_MAX_TARGET,
1307 &np->gref_rx_head) < 0) {
1308 printk(KERN_ALERT "#### netfront can't alloc rx grant refs\n");
1313 netdev->netdev_ops = &xennet_netdev_ops;
1315 netif_napi_add(netdev, &np->napi, xennet_poll, 64);
1316 netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1318 netdev->hw_features = NETIF_F_IP_CSUM | NETIF_F_SG | NETIF_F_TSO;
1321 * Assume that all hw features are available for now. This set
1322 * will be adjusted by the call to netdev_update_features() in
1323 * xennet_connect() which is the earliest point where we can
1324 * negotiate with the backend regarding supported features.
1326 netdev->features |= netdev->hw_features;
1328 SET_ETHTOOL_OPS(netdev, &xennet_ethtool_ops);
1329 SET_NETDEV_DEV(netdev, &dev->dev);
1331 np->netdev = netdev;
1333 netif_carrier_off(netdev);
1338 gnttab_free_grant_references(np->gref_tx_head);
1340 free_percpu(np->stats);
1342 free_netdev(netdev);
1343 return ERR_PTR(err);
1347 * Entry point to this code when a new device is created. Allocate the basic
1348 * structures and the ring buffers for communication with the backend, and
1349 * inform the backend of the appropriate details for those.
1351 static int __devinit netfront_probe(struct xenbus_device *dev,
1352 const struct xenbus_device_id *id)
1355 struct net_device *netdev;
1356 struct netfront_info *info;
1358 netdev = xennet_create_dev(dev);
1359 if (IS_ERR(netdev)) {
1360 err = PTR_ERR(netdev);
1361 xenbus_dev_fatal(dev, err, "creating netdev");
1365 info = netdev_priv(netdev);
1366 dev_set_drvdata(&dev->dev, info);
1368 err = register_netdev(info->netdev);
1370 printk(KERN_WARNING "%s: register_netdev err=%d\n",
1375 err = xennet_sysfs_addif(info->netdev);
1377 unregister_netdev(info->netdev);
1378 printk(KERN_WARNING "%s: add sysfs failed err=%d\n",
1386 free_netdev(netdev);
1387 dev_set_drvdata(&dev->dev, NULL);
1391 static void xennet_end_access(int ref, void *page)
1393 /* This frees the page as a side-effect */
1394 if (ref != GRANT_INVALID_REF)
1395 gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1398 static void xennet_disconnect_backend(struct netfront_info *info)
1400 /* Stop old i/f to prevent errors whilst we rebuild the state. */
1401 spin_lock_bh(&info->rx_lock);
1402 spin_lock_irq(&info->tx_lock);
1403 netif_carrier_off(info->netdev);
1404 spin_unlock_irq(&info->tx_lock);
1405 spin_unlock_bh(&info->rx_lock);
1407 if (info->netdev->irq)
1408 unbind_from_irqhandler(info->netdev->irq, info->netdev);
1409 info->evtchn = info->netdev->irq = 0;
1411 /* End access and free the pages */
1412 xennet_end_access(info->tx_ring_ref, info->tx.sring);
1413 xennet_end_access(info->rx_ring_ref, info->rx.sring);
1415 info->tx_ring_ref = GRANT_INVALID_REF;
1416 info->rx_ring_ref = GRANT_INVALID_REF;
1417 info->tx.sring = NULL;
1418 info->rx.sring = NULL;
1422 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1423 * driver restart. We tear down our netif structure and recreate it, but
1424 * leave the device-layer structures intact so that this is transparent to the
1425 * rest of the kernel.
1427 static int netfront_resume(struct xenbus_device *dev)
1429 struct netfront_info *info = dev_get_drvdata(&dev->dev);
1431 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1433 xennet_disconnect_backend(info);
1437 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1439 char *s, *e, *macstr;
1442 macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1444 return PTR_ERR(macstr);
1446 for (i = 0; i < ETH_ALEN; i++) {
1447 mac[i] = simple_strtoul(s, &e, 16);
1448 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1459 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1461 struct net_device *dev = dev_id;
1462 struct netfront_info *np = netdev_priv(dev);
1463 unsigned long flags;
1465 spin_lock_irqsave(&np->tx_lock, flags);
1467 if (likely(netif_carrier_ok(dev))) {
1468 xennet_tx_buf_gc(dev);
1469 /* Under tx_lock: protects access to rx shared-ring indexes. */
1470 if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx))
1471 napi_schedule(&np->napi);
1474 spin_unlock_irqrestore(&np->tx_lock, flags);
1479 static int setup_netfront(struct xenbus_device *dev, struct netfront_info *info)
1481 struct xen_netif_tx_sring *txs;
1482 struct xen_netif_rx_sring *rxs;
1484 struct net_device *netdev = info->netdev;
1486 info->tx_ring_ref = GRANT_INVALID_REF;
1487 info->rx_ring_ref = GRANT_INVALID_REF;
1488 info->rx.sring = NULL;
1489 info->tx.sring = NULL;
1492 err = xen_net_read_mac(dev, netdev->dev_addr);
1494 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1498 txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1501 xenbus_dev_fatal(dev, err, "allocating tx ring page");
1504 SHARED_RING_INIT(txs);
1505 FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE);
1507 err = xenbus_grant_ring(dev, virt_to_mfn(txs));
1509 free_page((unsigned long)txs);
1513 info->tx_ring_ref = err;
1514 rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1517 xenbus_dev_fatal(dev, err, "allocating rx ring page");
1520 SHARED_RING_INIT(rxs);
1521 FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE);
1523 err = xenbus_grant_ring(dev, virt_to_mfn(rxs));
1525 free_page((unsigned long)rxs);
1528 info->rx_ring_ref = err;
1530 err = xenbus_alloc_evtchn(dev, &info->evtchn);
1534 err = bind_evtchn_to_irqhandler(info->evtchn, xennet_interrupt,
1535 0, netdev->name, netdev);
1545 /* Common code used when first setting up, and when resuming. */
1546 static int talk_to_netback(struct xenbus_device *dev,
1547 struct netfront_info *info)
1549 const char *message;
1550 struct xenbus_transaction xbt;
1553 /* Create shared ring, alloc event channel. */
1554 err = setup_netfront(dev, info);
1559 err = xenbus_transaction_start(&xbt);
1561 xenbus_dev_fatal(dev, err, "starting transaction");
1565 err = xenbus_printf(xbt, dev->nodename, "tx-ring-ref", "%u",
1568 message = "writing tx ring-ref";
1569 goto abort_transaction;
1571 err = xenbus_printf(xbt, dev->nodename, "rx-ring-ref", "%u",
1574 message = "writing rx ring-ref";
1575 goto abort_transaction;
1577 err = xenbus_printf(xbt, dev->nodename,
1578 "event-channel", "%u", info->evtchn);
1580 message = "writing event-channel";
1581 goto abort_transaction;
1584 err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1587 message = "writing request-rx-copy";
1588 goto abort_transaction;
1591 err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1593 message = "writing feature-rx-notify";
1594 goto abort_transaction;
1597 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1599 message = "writing feature-sg";
1600 goto abort_transaction;
1603 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1605 message = "writing feature-gso-tcpv4";
1606 goto abort_transaction;
1609 err = xenbus_transaction_end(xbt, 0);
1613 xenbus_dev_fatal(dev, err, "completing transaction");
1620 xenbus_transaction_end(xbt, 1);
1621 xenbus_dev_fatal(dev, err, "%s", message);
1623 xennet_disconnect_backend(info);
1628 static int xennet_connect(struct net_device *dev)
1630 struct netfront_info *np = netdev_priv(dev);
1631 int i, requeue_idx, err;
1632 struct sk_buff *skb;
1634 struct xen_netif_rx_request *req;
1635 unsigned int feature_rx_copy;
1637 err = xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1638 "feature-rx-copy", "%u", &feature_rx_copy);
1640 feature_rx_copy = 0;
1642 if (!feature_rx_copy) {
1644 "backend does not support copying receive path\n");
1648 err = talk_to_netback(np->xbdev, np);
1653 netdev_update_features(dev);
1656 spin_lock_bh(&np->rx_lock);
1657 spin_lock_irq(&np->tx_lock);
1659 /* Step 1: Discard all pending TX packet fragments. */
1660 xennet_release_tx_bufs(np);
1662 /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1663 for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) {
1665 const struct page *page;
1666 if (!np->rx_skbs[i])
1669 skb = np->rx_skbs[requeue_idx] = xennet_get_rx_skb(np, i);
1670 ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i);
1671 req = RING_GET_REQUEST(&np->rx, requeue_idx);
1673 frag = &skb_shinfo(skb)->frags[0];
1674 page = skb_frag_page(frag);
1675 gnttab_grant_foreign_access_ref(
1676 ref, np->xbdev->otherend_id,
1677 pfn_to_mfn(page_to_pfn(page)),
1680 req->id = requeue_idx;
1685 np->rx.req_prod_pvt = requeue_idx;
1688 * Step 3: All public and private state should now be sane. Get
1689 * ready to start sending and receiving packets and give the driver
1690 * domain a kick because we've probably just requeued some
1693 netif_carrier_on(np->netdev);
1694 notify_remote_via_irq(np->netdev->irq);
1695 xennet_tx_buf_gc(dev);
1696 xennet_alloc_rx_buffers(dev);
1698 spin_unlock_irq(&np->tx_lock);
1699 spin_unlock_bh(&np->rx_lock);
1705 * Callback received when the backend's state changes.
1707 static void netback_changed(struct xenbus_device *dev,
1708 enum xenbus_state backend_state)
1710 struct netfront_info *np = dev_get_drvdata(&dev->dev);
1711 struct net_device *netdev = np->netdev;
1713 dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
1715 switch (backend_state) {
1716 case XenbusStateInitialising:
1717 case XenbusStateInitialised:
1718 case XenbusStateReconfiguring:
1719 case XenbusStateReconfigured:
1720 case XenbusStateUnknown:
1721 case XenbusStateClosed:
1724 case XenbusStateInitWait:
1725 if (dev->state != XenbusStateInitialising)
1727 if (xennet_connect(netdev) != 0)
1729 xenbus_switch_state(dev, XenbusStateConnected);
1732 case XenbusStateConnected:
1733 netif_notify_peers(netdev);
1736 case XenbusStateClosing:
1737 xenbus_frontend_closed(dev);
1742 static const struct xennet_stat {
1743 char name[ETH_GSTRING_LEN];
1745 } xennet_stats[] = {
1747 "rx_gso_checksum_fixup",
1748 offsetof(struct netfront_info, rx_gso_checksum_fixup)
1752 static int xennet_get_sset_count(struct net_device *dev, int string_set)
1754 switch (string_set) {
1756 return ARRAY_SIZE(xennet_stats);
1762 static void xennet_get_ethtool_stats(struct net_device *dev,
1763 struct ethtool_stats *stats, u64 * data)
1765 void *np = netdev_priv(dev);
1768 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
1769 data[i] = *(unsigned long *)(np + xennet_stats[i].offset);
1772 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
1776 switch (stringset) {
1778 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
1779 memcpy(data + i * ETH_GSTRING_LEN,
1780 xennet_stats[i].name, ETH_GSTRING_LEN);
1785 static const struct ethtool_ops xennet_ethtool_ops =
1787 .get_link = ethtool_op_get_link,
1789 .get_sset_count = xennet_get_sset_count,
1790 .get_ethtool_stats = xennet_get_ethtool_stats,
1791 .get_strings = xennet_get_strings,
1795 static ssize_t show_rxbuf_min(struct device *dev,
1796 struct device_attribute *attr, char *buf)
1798 struct net_device *netdev = to_net_dev(dev);
1799 struct netfront_info *info = netdev_priv(netdev);
1801 return sprintf(buf, "%u\n", info->rx_min_target);
1804 static ssize_t store_rxbuf_min(struct device *dev,
1805 struct device_attribute *attr,
1806 const char *buf, size_t len)
1808 struct net_device *netdev = to_net_dev(dev);
1809 struct netfront_info *np = netdev_priv(netdev);
1811 unsigned long target;
1813 if (!capable(CAP_NET_ADMIN))
1816 target = simple_strtoul(buf, &endp, 0);
1820 if (target < RX_MIN_TARGET)
1821 target = RX_MIN_TARGET;
1822 if (target > RX_MAX_TARGET)
1823 target = RX_MAX_TARGET;
1825 spin_lock_bh(&np->rx_lock);
1826 if (target > np->rx_max_target)
1827 np->rx_max_target = target;
1828 np->rx_min_target = target;
1829 if (target > np->rx_target)
1830 np->rx_target = target;
1832 xennet_alloc_rx_buffers(netdev);
1834 spin_unlock_bh(&np->rx_lock);
1838 static ssize_t show_rxbuf_max(struct device *dev,
1839 struct device_attribute *attr, char *buf)
1841 struct net_device *netdev = to_net_dev(dev);
1842 struct netfront_info *info = netdev_priv(netdev);
1844 return sprintf(buf, "%u\n", info->rx_max_target);
1847 static ssize_t store_rxbuf_max(struct device *dev,
1848 struct device_attribute *attr,
1849 const char *buf, size_t len)
1851 struct net_device *netdev = to_net_dev(dev);
1852 struct netfront_info *np = netdev_priv(netdev);
1854 unsigned long target;
1856 if (!capable(CAP_NET_ADMIN))
1859 target = simple_strtoul(buf, &endp, 0);
1863 if (target < RX_MIN_TARGET)
1864 target = RX_MIN_TARGET;
1865 if (target > RX_MAX_TARGET)
1866 target = RX_MAX_TARGET;
1868 spin_lock_bh(&np->rx_lock);
1869 if (target < np->rx_min_target)
1870 np->rx_min_target = target;
1871 np->rx_max_target = target;
1872 if (target < np->rx_target)
1873 np->rx_target = target;
1875 xennet_alloc_rx_buffers(netdev);
1877 spin_unlock_bh(&np->rx_lock);
1881 static ssize_t show_rxbuf_cur(struct device *dev,
1882 struct device_attribute *attr, char *buf)
1884 struct net_device *netdev = to_net_dev(dev);
1885 struct netfront_info *info = netdev_priv(netdev);
1887 return sprintf(buf, "%u\n", info->rx_target);
1890 static struct device_attribute xennet_attrs[] = {
1891 __ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf_min, store_rxbuf_min),
1892 __ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf_max, store_rxbuf_max),
1893 __ATTR(rxbuf_cur, S_IRUGO, show_rxbuf_cur, NULL),
1896 static int xennet_sysfs_addif(struct net_device *netdev)
1901 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) {
1902 err = device_create_file(&netdev->dev,
1911 device_remove_file(&netdev->dev, &xennet_attrs[i]);
1915 static void xennet_sysfs_delif(struct net_device *netdev)
1919 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++)
1920 device_remove_file(&netdev->dev, &xennet_attrs[i]);
1923 #endif /* CONFIG_SYSFS */
1925 static struct xenbus_device_id netfront_ids[] = {
1931 static int __devexit xennet_remove(struct xenbus_device *dev)
1933 struct netfront_info *info = dev_get_drvdata(&dev->dev);
1935 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1937 xennet_disconnect_backend(info);
1939 xennet_sysfs_delif(info->netdev);
1941 unregister_netdev(info->netdev);
1943 del_timer_sync(&info->rx_refill_timer);
1945 free_percpu(info->stats);
1947 free_netdev(info->netdev);
1952 static struct xenbus_driver netfront_driver = {
1954 .owner = THIS_MODULE,
1955 .ids = netfront_ids,
1956 .probe = netfront_probe,
1957 .remove = __devexit_p(xennet_remove),
1958 .resume = netfront_resume,
1959 .otherend_changed = netback_changed,
1962 static int __init netif_init(void)
1967 if (xen_initial_domain())
1970 printk(KERN_INFO "Initialising Xen virtual ethernet driver.\n");
1972 return xenbus_register_frontend(&netfront_driver);
1974 module_init(netif_init);
1977 static void __exit netif_exit(void)
1979 if (xen_initial_domain())
1982 xenbus_unregister_driver(&netfront_driver);
1984 module_exit(netif_exit);
1986 MODULE_DESCRIPTION("Xen virtual network device frontend");
1987 MODULE_LICENSE("GPL");
1988 MODULE_ALIAS("xen:vif");
1989 MODULE_ALIAS("xennet");