2 * Back-end of the driver for virtual network devices. This portion of the
3 * driver exports a 'unified' network-device interface that can be accessed
4 * by any operating system that implements a compatible front end. A
5 * reference front-end implementation can be found in:
6 * drivers/net/xen-netfront.c
8 * Copyright (c) 2002-2005, K A Fraser
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License version 2
12 * as published by the Free Software Foundation; or, when distributed
13 * separately from the Linux kernel or incorporated into other
14 * software packages, subject to the following license:
16 * Permission is hereby granted, free of charge, to any person obtaining a copy
17 * of this source file (the "Software"), to deal in the Software without
18 * restriction, including without limitation the rights to use, copy, modify,
19 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
20 * and to permit persons to whom the Software is furnished to do so, subject to
21 * the following conditions:
23 * The above copyright notice and this permission notice shall be included in
24 * all copies or substantial portions of the Software.
26 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
27 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
28 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
29 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
30 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
31 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
37 #include <linux/kthread.h>
38 #include <linux/if_vlan.h>
39 #include <linux/udp.h>
42 #include <net/flow_keys.h>
45 #include <xen/events.h>
46 #include <xen/interface/memory.h>
48 #include <asm/xen/hypercall.h>
49 #include <asm/xen/page.h>
51 struct pending_tx_info {
52 struct xen_netif_tx_request req;
55 typedef unsigned int pending_ring_idx_t;
57 struct netbk_rx_meta {
63 #define MAX_PENDING_REQS 256
65 /* Discriminate from any valid pending_idx value. */
66 #define INVALID_PENDING_IDX 0xFFFF
68 #define MAX_BUFFER_OFFSET PAGE_SIZE
70 /* extra field used in struct page */
73 #if BITS_PER_LONG < 64
75 #define GROUP_WIDTH (BITS_PER_LONG - IDX_WIDTH)
76 unsigned int group:GROUP_WIDTH;
77 unsigned int idx:IDX_WIDTH;
79 unsigned int group, idx;
87 struct task_struct *task;
89 struct sk_buff_head rx_queue;
90 struct sk_buff_head tx_queue;
92 struct timer_list net_timer;
94 struct page *mmap_pages[MAX_PENDING_REQS];
96 pending_ring_idx_t pending_prod;
97 pending_ring_idx_t pending_cons;
98 struct list_head net_schedule_list;
100 /* Protect the net_schedule_list in netif. */
101 spinlock_t net_schedule_list_lock;
103 atomic_t netfront_count;
105 struct pending_tx_info pending_tx_info[MAX_PENDING_REQS];
106 struct gnttab_copy tx_copy_ops[MAX_PENDING_REQS];
108 u16 pending_ring[MAX_PENDING_REQS];
111 * Given MAX_BUFFER_OFFSET of 4096 the worst case is that each
112 * head/fragment page uses 2 copy operations because it
113 * straddles two buffers in the frontend.
115 struct gnttab_copy grant_copy_op[2*XEN_NETIF_RX_RING_SIZE];
116 struct netbk_rx_meta meta[2*XEN_NETIF_RX_RING_SIZE];
119 static struct xen_netbk *xen_netbk;
120 static int xen_netbk_group_nr;
122 void xen_netbk_add_xenvif(struct xenvif *vif)
125 int min_netfront_count;
127 struct xen_netbk *netbk;
129 min_netfront_count = atomic_read(&xen_netbk[0].netfront_count);
130 for (i = 0; i < xen_netbk_group_nr; i++) {
131 int netfront_count = atomic_read(&xen_netbk[i].netfront_count);
132 if (netfront_count < min_netfront_count) {
134 min_netfront_count = netfront_count;
138 netbk = &xen_netbk[min_group];
141 atomic_inc(&netbk->netfront_count);
144 void xen_netbk_remove_xenvif(struct xenvif *vif)
146 struct xen_netbk *netbk = vif->netbk;
148 atomic_dec(&netbk->netfront_count);
151 static void xen_netbk_idx_release(struct xen_netbk *netbk, u16 pending_idx,
153 static void make_tx_response(struct xenvif *vif,
154 struct xen_netif_tx_request *txp,
156 static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif,
163 static inline unsigned long idx_to_pfn(struct xen_netbk *netbk,
166 return page_to_pfn(netbk->mmap_pages[idx]);
169 static inline unsigned long idx_to_kaddr(struct xen_netbk *netbk,
172 return (unsigned long)pfn_to_kaddr(idx_to_pfn(netbk, idx));
175 /* extra field used in struct page */
176 static inline void set_page_ext(struct page *pg, struct xen_netbk *netbk,
179 unsigned int group = netbk - xen_netbk;
180 union page_ext ext = { .e = { .group = group + 1, .idx = idx } };
182 BUILD_BUG_ON(sizeof(ext) > sizeof(ext.mapping));
183 pg->mapping = ext.mapping;
186 static int get_page_ext(struct page *pg,
187 unsigned int *pgroup, unsigned int *pidx)
189 union page_ext ext = { .mapping = pg->mapping };
190 struct xen_netbk *netbk;
191 unsigned int group, idx;
193 group = ext.e.group - 1;
195 if (group < 0 || group >= xen_netbk_group_nr)
198 netbk = &xen_netbk[group];
202 if ((idx < 0) || (idx >= MAX_PENDING_REQS))
205 if (netbk->mmap_pages[idx] != pg)
215 * This is the amount of packet we copy rather than map, so that the
216 * guest can't fiddle with the contents of the headers while we do
217 * packet processing on them (netfilter, routing, etc).
219 #define PKT_PROT_LEN (ETH_HLEN + \
221 sizeof(struct iphdr) + MAX_IPOPTLEN + \
222 sizeof(struct tcphdr) + MAX_TCP_OPTION_SPACE)
224 static u16 frag_get_pending_idx(skb_frag_t *frag)
226 return (u16)frag->page_offset;
229 static void frag_set_pending_idx(skb_frag_t *frag, u16 pending_idx)
231 frag->page_offset = pending_idx;
234 static inline pending_ring_idx_t pending_index(unsigned i)
236 return i & (MAX_PENDING_REQS-1);
239 static inline pending_ring_idx_t nr_pending_reqs(struct xen_netbk *netbk)
241 return MAX_PENDING_REQS -
242 netbk->pending_prod + netbk->pending_cons;
245 static void xen_netbk_kick_thread(struct xen_netbk *netbk)
250 static int max_required_rx_slots(struct xenvif *vif)
252 int max = DIV_ROUND_UP(vif->dev->mtu, PAGE_SIZE);
254 if (vif->can_sg || vif->gso || vif->gso_prefix)
255 max += MAX_SKB_FRAGS + 1; /* extra_info + frags */
260 int xen_netbk_rx_ring_full(struct xenvif *vif)
262 RING_IDX peek = vif->rx_req_cons_peek;
263 RING_IDX needed = max_required_rx_slots(vif);
265 return ((vif->rx.sring->req_prod - peek) < needed) ||
266 ((vif->rx.rsp_prod_pvt + XEN_NETIF_RX_RING_SIZE - peek) < needed);
269 int xen_netbk_must_stop_queue(struct xenvif *vif)
271 if (!xen_netbk_rx_ring_full(vif))
274 vif->rx.sring->req_event = vif->rx_req_cons_peek +
275 max_required_rx_slots(vif);
276 mb(); /* request notification /then/ check the queue */
278 return xen_netbk_rx_ring_full(vif);
282 * Returns true if we should start a new receive buffer instead of
283 * adding 'size' bytes to a buffer which currently contains 'offset'
286 static bool start_new_rx_buffer(int offset, unsigned long size, int head)
288 /* simple case: we have completely filled the current buffer. */
289 if (offset == MAX_BUFFER_OFFSET)
293 * complex case: start a fresh buffer if the current frag
294 * would overflow the current buffer but only if:
295 * (i) this frag would fit completely in the next buffer
296 * and (ii) there is already some data in the current buffer
297 * and (iii) this is not the head buffer.
300 * - (i) stops us splitting a frag into two copies
301 * unless the frag is too large for a single buffer.
302 * - (ii) stops us from leaving a buffer pointlessly empty.
303 * - (iii) stops us leaving the first buffer
304 * empty. Strictly speaking this is already covered
305 * by (ii) but is explicitly checked because
306 * netfront relies on the first buffer being
307 * non-empty and can crash otherwise.
309 * This means we will effectively linearise small
310 * frags but do not needlessly split large buffers
311 * into multiple copies tend to give large frags their
312 * own buffers as before.
314 if ((offset + size > MAX_BUFFER_OFFSET) &&
315 (size <= MAX_BUFFER_OFFSET) && offset && !head)
322 * Figure out how many ring slots we're going to need to send @skb to
323 * the guest. This function is essentially a dry run of
324 * netbk_gop_frag_copy.
326 unsigned int xen_netbk_count_skb_slots(struct xenvif *vif, struct sk_buff *skb)
331 count = DIV_ROUND_UP(skb_headlen(skb), PAGE_SIZE);
333 copy_off = skb_headlen(skb) % PAGE_SIZE;
335 if (skb_shinfo(skb)->gso_size)
338 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
339 unsigned long size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
340 unsigned long offset = skb_shinfo(skb)->frags[i].page_offset;
343 offset &= ~PAGE_MASK;
346 BUG_ON(offset >= PAGE_SIZE);
347 BUG_ON(copy_off > MAX_BUFFER_OFFSET);
349 bytes = PAGE_SIZE - offset;
354 if (start_new_rx_buffer(copy_off, bytes, 0)) {
359 if (copy_off + bytes > MAX_BUFFER_OFFSET)
360 bytes = MAX_BUFFER_OFFSET - copy_off;
367 if (offset == PAGE_SIZE)
374 struct netrx_pending_operations {
375 unsigned copy_prod, copy_cons;
376 unsigned meta_prod, meta_cons;
377 struct gnttab_copy *copy;
378 struct netbk_rx_meta *meta;
380 grant_ref_t copy_gref;
383 static struct netbk_rx_meta *get_next_rx_buffer(struct xenvif *vif,
384 struct netrx_pending_operations *npo)
386 struct netbk_rx_meta *meta;
387 struct xen_netif_rx_request *req;
389 req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++);
391 meta = npo->meta + npo->meta_prod++;
397 npo->copy_gref = req->gref;
403 * Set up the grant operations for this fragment. If it's a flipping
404 * interface, we also set up the unmap request from here.
406 static void netbk_gop_frag_copy(struct xenvif *vif, struct sk_buff *skb,
407 struct netrx_pending_operations *npo,
408 struct page *page, unsigned long size,
409 unsigned long offset, int *head)
411 struct gnttab_copy *copy_gop;
412 struct netbk_rx_meta *meta;
414 * These variables are used iff get_page_ext returns true,
415 * in which case they are guaranteed to be initialized.
417 unsigned int uninitialized_var(group), uninitialized_var(idx);
418 int foreign = get_page_ext(page, &group, &idx);
421 /* Data must not cross a page boundary. */
422 BUG_ON(size + offset > PAGE_SIZE<<compound_order(page));
424 meta = npo->meta + npo->meta_prod - 1;
426 /* Skip unused frames from start of page */
427 page += offset >> PAGE_SHIFT;
428 offset &= ~PAGE_MASK;
431 BUG_ON(offset >= PAGE_SIZE);
432 BUG_ON(npo->copy_off > MAX_BUFFER_OFFSET);
434 bytes = PAGE_SIZE - offset;
439 if (start_new_rx_buffer(npo->copy_off, bytes, *head)) {
441 * Netfront requires there to be some data in the head
446 meta = get_next_rx_buffer(vif, npo);
449 if (npo->copy_off + bytes > MAX_BUFFER_OFFSET)
450 bytes = MAX_BUFFER_OFFSET - npo->copy_off;
452 copy_gop = npo->copy + npo->copy_prod++;
453 copy_gop->flags = GNTCOPY_dest_gref;
455 struct xen_netbk *netbk = &xen_netbk[group];
456 struct pending_tx_info *src_pend;
458 src_pend = &netbk->pending_tx_info[idx];
460 copy_gop->source.domid = src_pend->vif->domid;
461 copy_gop->source.u.ref = src_pend->req.gref;
462 copy_gop->flags |= GNTCOPY_source_gref;
464 void *vaddr = page_address(page);
465 copy_gop->source.domid = DOMID_SELF;
466 copy_gop->source.u.gmfn = virt_to_mfn(vaddr);
468 copy_gop->source.offset = offset;
469 copy_gop->dest.domid = vif->domid;
471 copy_gop->dest.offset = npo->copy_off;
472 copy_gop->dest.u.ref = npo->copy_gref;
473 copy_gop->len = bytes;
475 npo->copy_off += bytes;
482 if (offset == PAGE_SIZE && size) {
483 BUG_ON(!PageCompound(page));
488 /* Leave a gap for the GSO descriptor. */
489 if (*head && skb_shinfo(skb)->gso_size && !vif->gso_prefix)
492 *head = 0; /* There must be something in this buffer now. */
498 * Prepare an SKB to be transmitted to the frontend.
500 * This function is responsible for allocating grant operations, meta
503 * It returns the number of meta structures consumed. The number of
504 * ring slots used is always equal to the number of meta slots used
505 * plus the number of GSO descriptors used. Currently, we use either
506 * zero GSO descriptors (for non-GSO packets) or one descriptor (for
507 * frontend-side LRO).
509 static int netbk_gop_skb(struct sk_buff *skb,
510 struct netrx_pending_operations *npo)
512 struct xenvif *vif = netdev_priv(skb->dev);
513 int nr_frags = skb_shinfo(skb)->nr_frags;
515 struct xen_netif_rx_request *req;
516 struct netbk_rx_meta *meta;
521 old_meta_prod = npo->meta_prod;
523 /* Set up a GSO prefix descriptor, if necessary */
524 if (skb_shinfo(skb)->gso_size && vif->gso_prefix) {
525 req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++);
526 meta = npo->meta + npo->meta_prod++;
527 meta->gso_size = skb_shinfo(skb)->gso_size;
532 req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++);
533 meta = npo->meta + npo->meta_prod++;
535 if (!vif->gso_prefix)
536 meta->gso_size = skb_shinfo(skb)->gso_size;
543 npo->copy_gref = req->gref;
546 while (data < skb_tail_pointer(skb)) {
547 unsigned int offset = offset_in_page(data);
548 unsigned int len = PAGE_SIZE - offset;
550 if (data + len > skb_tail_pointer(skb))
551 len = skb_tail_pointer(skb) - data;
553 netbk_gop_frag_copy(vif, skb, npo,
554 virt_to_page(data), len, offset, &head);
558 for (i = 0; i < nr_frags; i++) {
559 netbk_gop_frag_copy(vif, skb, npo,
560 skb_frag_page(&skb_shinfo(skb)->frags[i]),
561 skb_frag_size(&skb_shinfo(skb)->frags[i]),
562 skb_shinfo(skb)->frags[i].page_offset,
566 return npo->meta_prod - old_meta_prod;
570 * This is a twin to netbk_gop_skb. Assume that netbk_gop_skb was
571 * used to set up the operations on the top of
572 * netrx_pending_operations, which have since been done. Check that
573 * they didn't give any errors and advance over them.
575 static int netbk_check_gop(struct xenvif *vif, int nr_meta_slots,
576 struct netrx_pending_operations *npo)
578 struct gnttab_copy *copy_op;
579 int status = XEN_NETIF_RSP_OKAY;
582 for (i = 0; i < nr_meta_slots; i++) {
583 copy_op = npo->copy + npo->copy_cons++;
584 if (copy_op->status != GNTST_okay) {
586 "Bad status %d from copy to DOM%d.\n",
587 copy_op->status, vif->domid);
588 status = XEN_NETIF_RSP_ERROR;
595 static void netbk_add_frag_responses(struct xenvif *vif, int status,
596 struct netbk_rx_meta *meta,
600 unsigned long offset;
602 /* No fragments used */
603 if (nr_meta_slots <= 1)
608 for (i = 0; i < nr_meta_slots; i++) {
610 if (i == nr_meta_slots - 1)
613 flags = XEN_NETRXF_more_data;
616 make_rx_response(vif, meta[i].id, status, offset,
617 meta[i].size, flags);
621 struct skb_cb_overlay {
625 static void xen_netbk_rx_action(struct xen_netbk *netbk)
627 struct xenvif *vif = NULL, *tmp;
630 struct xen_netif_rx_response *resp;
631 struct sk_buff_head rxq;
637 unsigned long offset;
638 struct skb_cb_overlay *sco;
640 struct netrx_pending_operations npo = {
641 .copy = netbk->grant_copy_op,
645 skb_queue_head_init(&rxq);
649 while ((skb = skb_dequeue(&netbk->rx_queue)) != NULL) {
650 vif = netdev_priv(skb->dev);
651 nr_frags = skb_shinfo(skb)->nr_frags;
653 sco = (struct skb_cb_overlay *)skb->cb;
654 sco->meta_slots_used = netbk_gop_skb(skb, &npo);
656 count += nr_frags + 1;
658 __skb_queue_tail(&rxq, skb);
660 /* Filled the batch queue? */
661 if (count + MAX_SKB_FRAGS >= XEN_NETIF_RX_RING_SIZE)
665 BUG_ON(npo.meta_prod > ARRAY_SIZE(netbk->meta));
670 BUG_ON(npo.copy_prod > ARRAY_SIZE(netbk->grant_copy_op));
671 gnttab_batch_copy(netbk->grant_copy_op, npo.copy_prod);
673 while ((skb = __skb_dequeue(&rxq)) != NULL) {
674 sco = (struct skb_cb_overlay *)skb->cb;
676 vif = netdev_priv(skb->dev);
678 if (netbk->meta[npo.meta_cons].gso_size && vif->gso_prefix) {
679 resp = RING_GET_RESPONSE(&vif->rx,
680 vif->rx.rsp_prod_pvt++);
682 resp->flags = XEN_NETRXF_gso_prefix | XEN_NETRXF_more_data;
684 resp->offset = netbk->meta[npo.meta_cons].gso_size;
685 resp->id = netbk->meta[npo.meta_cons].id;
686 resp->status = sco->meta_slots_used;
689 sco->meta_slots_used--;
693 vif->dev->stats.tx_bytes += skb->len;
694 vif->dev->stats.tx_packets++;
696 status = netbk_check_gop(vif, sco->meta_slots_used, &npo);
698 if (sco->meta_slots_used == 1)
701 flags = XEN_NETRXF_more_data;
703 if (skb->ip_summed == CHECKSUM_PARTIAL) /* local packet? */
704 flags |= XEN_NETRXF_csum_blank | XEN_NETRXF_data_validated;
705 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
706 /* remote but checksummed. */
707 flags |= XEN_NETRXF_data_validated;
710 resp = make_rx_response(vif, netbk->meta[npo.meta_cons].id,
712 netbk->meta[npo.meta_cons].size,
715 if (netbk->meta[npo.meta_cons].gso_size && !vif->gso_prefix) {
716 struct xen_netif_extra_info *gso =
717 (struct xen_netif_extra_info *)
718 RING_GET_RESPONSE(&vif->rx,
719 vif->rx.rsp_prod_pvt++);
721 resp->flags |= XEN_NETRXF_extra_info;
723 gso->u.gso.size = netbk->meta[npo.meta_cons].gso_size;
724 gso->u.gso.type = XEN_NETIF_GSO_TYPE_TCPV4;
726 gso->u.gso.features = 0;
728 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
732 netbk_add_frag_responses(vif, status,
733 netbk->meta + npo.meta_cons + 1,
734 sco->meta_slots_used);
736 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->rx, ret);
738 if (ret && list_empty(&vif->notify_list))
739 list_add_tail(&vif->notify_list, ¬ify);
741 xenvif_notify_tx_completion(vif);
744 npo.meta_cons += sco->meta_slots_used;
748 list_for_each_entry_safe(vif, tmp, ¬ify, notify_list) {
749 notify_remote_via_irq(vif->irq);
750 list_del_init(&vif->notify_list);
753 /* More work to do? */
754 if (!skb_queue_empty(&netbk->rx_queue) &&
755 !timer_pending(&netbk->net_timer))
756 xen_netbk_kick_thread(netbk);
759 void xen_netbk_queue_tx_skb(struct xenvif *vif, struct sk_buff *skb)
761 struct xen_netbk *netbk = vif->netbk;
763 skb_queue_tail(&netbk->rx_queue, skb);
765 xen_netbk_kick_thread(netbk);
768 static void xen_netbk_alarm(unsigned long data)
770 struct xen_netbk *netbk = (struct xen_netbk *)data;
771 xen_netbk_kick_thread(netbk);
774 static int __on_net_schedule_list(struct xenvif *vif)
776 return !list_empty(&vif->schedule_list);
779 /* Must be called with net_schedule_list_lock held */
780 static void remove_from_net_schedule_list(struct xenvif *vif)
782 if (likely(__on_net_schedule_list(vif))) {
783 list_del_init(&vif->schedule_list);
788 static struct xenvif *poll_net_schedule_list(struct xen_netbk *netbk)
790 struct xenvif *vif = NULL;
792 spin_lock_irq(&netbk->net_schedule_list_lock);
793 if (list_empty(&netbk->net_schedule_list))
796 vif = list_first_entry(&netbk->net_schedule_list,
797 struct xenvif, schedule_list);
803 remove_from_net_schedule_list(vif);
805 spin_unlock_irq(&netbk->net_schedule_list_lock);
809 void xen_netbk_schedule_xenvif(struct xenvif *vif)
812 struct xen_netbk *netbk = vif->netbk;
814 if (__on_net_schedule_list(vif))
817 spin_lock_irqsave(&netbk->net_schedule_list_lock, flags);
818 if (!__on_net_schedule_list(vif) &&
819 likely(xenvif_schedulable(vif))) {
820 list_add_tail(&vif->schedule_list, &netbk->net_schedule_list);
823 spin_unlock_irqrestore(&netbk->net_schedule_list_lock, flags);
827 if ((nr_pending_reqs(netbk) < (MAX_PENDING_REQS/2)) &&
828 !list_empty(&netbk->net_schedule_list))
829 xen_netbk_kick_thread(netbk);
832 void xen_netbk_deschedule_xenvif(struct xenvif *vif)
834 struct xen_netbk *netbk = vif->netbk;
835 spin_lock_irq(&netbk->net_schedule_list_lock);
836 remove_from_net_schedule_list(vif);
837 spin_unlock_irq(&netbk->net_schedule_list_lock);
840 void xen_netbk_check_rx_xenvif(struct xenvif *vif)
844 RING_FINAL_CHECK_FOR_REQUESTS(&vif->tx, more_to_do);
847 xen_netbk_schedule_xenvif(vif);
850 static void tx_add_credit(struct xenvif *vif)
852 unsigned long max_burst, max_credit;
855 * Allow a burst big enough to transmit a jumbo packet of up to 128kB.
856 * Otherwise the interface can seize up due to insufficient credit.
858 max_burst = RING_GET_REQUEST(&vif->tx, vif->tx.req_cons)->size;
859 max_burst = min(max_burst, 131072UL);
860 max_burst = max(max_burst, vif->credit_bytes);
862 /* Take care that adding a new chunk of credit doesn't wrap to zero. */
863 max_credit = vif->remaining_credit + vif->credit_bytes;
864 if (max_credit < vif->remaining_credit)
865 max_credit = ULONG_MAX; /* wrapped: clamp to ULONG_MAX */
867 vif->remaining_credit = min(max_credit, max_burst);
870 static void tx_credit_callback(unsigned long data)
872 struct xenvif *vif = (struct xenvif *)data;
874 xen_netbk_check_rx_xenvif(vif);
877 static void netbk_tx_err(struct xenvif *vif,
878 struct xen_netif_tx_request *txp, RING_IDX end)
880 RING_IDX cons = vif->tx.req_cons;
883 make_tx_response(vif, txp, XEN_NETIF_RSP_ERROR);
886 txp = RING_GET_REQUEST(&vif->tx, cons++);
888 vif->tx.req_cons = cons;
889 xen_netbk_check_rx_xenvif(vif);
893 static void netbk_fatal_tx_err(struct xenvif *vif)
895 netdev_err(vif->dev, "fatal error; disabling device\n");
896 xenvif_carrier_off(vif);
900 static int netbk_count_requests(struct xenvif *vif,
901 struct xen_netif_tx_request *first,
902 struct xen_netif_tx_request *txp,
905 RING_IDX cons = vif->tx.req_cons;
908 if (!(first->flags & XEN_NETTXF_more_data))
912 if (frags >= work_to_do) {
913 netdev_err(vif->dev, "Need more frags\n");
914 netbk_fatal_tx_err(vif);
918 if (unlikely(frags >= MAX_SKB_FRAGS)) {
919 netdev_err(vif->dev, "Too many frags\n");
920 netbk_fatal_tx_err(vif);
924 memcpy(txp, RING_GET_REQUEST(&vif->tx, cons + frags),
926 if (txp->size > first->size) {
927 netdev_err(vif->dev, "Frag is bigger than frame.\n");
928 netbk_fatal_tx_err(vif);
932 first->size -= txp->size;
935 if (unlikely((txp->offset + txp->size) > PAGE_SIZE)) {
936 netdev_err(vif->dev, "txp->offset: %x, size: %u\n",
937 txp->offset, txp->size);
938 netbk_fatal_tx_err(vif);
941 } while ((txp++)->flags & XEN_NETTXF_more_data);
945 static struct page *xen_netbk_alloc_page(struct xen_netbk *netbk,
949 page = alloc_page(GFP_KERNEL|__GFP_COLD);
952 set_page_ext(page, netbk, pending_idx);
953 netbk->mmap_pages[pending_idx] = page;
957 static struct gnttab_copy *xen_netbk_get_requests(struct xen_netbk *netbk,
960 struct xen_netif_tx_request *txp,
961 struct gnttab_copy *gop)
963 struct skb_shared_info *shinfo = skb_shinfo(skb);
964 skb_frag_t *frags = shinfo->frags;
965 u16 pending_idx = *((u16 *)skb->data);
968 /* Skip first skb fragment if it is on same page as header fragment. */
969 start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx);
971 for (i = start; i < shinfo->nr_frags; i++, txp++) {
973 pending_ring_idx_t index;
974 struct pending_tx_info *pending_tx_info =
975 netbk->pending_tx_info;
977 index = pending_index(netbk->pending_cons++);
978 pending_idx = netbk->pending_ring[index];
979 page = xen_netbk_alloc_page(netbk, pending_idx);
983 gop->source.u.ref = txp->gref;
984 gop->source.domid = vif->domid;
985 gop->source.offset = txp->offset;
987 gop->dest.u.gmfn = virt_to_mfn(page_address(page));
988 gop->dest.domid = DOMID_SELF;
989 gop->dest.offset = txp->offset;
991 gop->len = txp->size;
992 gop->flags = GNTCOPY_source_gref;
996 memcpy(&pending_tx_info[pending_idx].req, txp, sizeof(*txp));
998 pending_tx_info[pending_idx].vif = vif;
999 frag_set_pending_idx(&frags[i], pending_idx);
1004 /* Unwind, freeing all pages and sending error responses. */
1005 while (i-- > start) {
1006 xen_netbk_idx_release(netbk, frag_get_pending_idx(&frags[i]),
1007 XEN_NETIF_RSP_ERROR);
1009 /* The head too, if necessary. */
1011 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_ERROR);
1016 static int xen_netbk_tx_check_gop(struct xen_netbk *netbk,
1017 struct sk_buff *skb,
1018 struct gnttab_copy **gopp)
1020 struct gnttab_copy *gop = *gopp;
1021 u16 pending_idx = *((u16 *)skb->data);
1022 struct skb_shared_info *shinfo = skb_shinfo(skb);
1023 int nr_frags = shinfo->nr_frags;
1026 /* Check status of header. */
1029 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_ERROR);
1031 /* Skip first skb fragment if it is on same page as header fragment. */
1032 start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx);
1034 for (i = start; i < nr_frags; i++) {
1037 pending_idx = frag_get_pending_idx(&shinfo->frags[i]);
1039 /* Check error status: if okay then remember grant handle. */
1040 newerr = (++gop)->status;
1041 if (likely(!newerr)) {
1042 /* Had a previous error? Invalidate this fragment. */
1044 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_OKAY);
1048 /* Error on this fragment: respond to client with an error. */
1049 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_ERROR);
1051 /* Not the first error? Preceding frags already invalidated. */
1055 /* First error: invalidate header and preceding fragments. */
1056 pending_idx = *((u16 *)skb->data);
1057 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_OKAY);
1058 for (j = start; j < i; j++) {
1059 pending_idx = frag_get_pending_idx(&shinfo->frags[j]);
1060 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_OKAY);
1063 /* Remember the error: invalidate all subsequent fragments. */
1071 static void xen_netbk_fill_frags(struct xen_netbk *netbk, struct sk_buff *skb)
1073 struct skb_shared_info *shinfo = skb_shinfo(skb);
1074 int nr_frags = shinfo->nr_frags;
1077 for (i = 0; i < nr_frags; i++) {
1078 skb_frag_t *frag = shinfo->frags + i;
1079 struct xen_netif_tx_request *txp;
1083 pending_idx = frag_get_pending_idx(frag);
1085 txp = &netbk->pending_tx_info[pending_idx].req;
1086 page = virt_to_page(idx_to_kaddr(netbk, pending_idx));
1087 __skb_fill_page_desc(skb, i, page, txp->offset, txp->size);
1088 skb->len += txp->size;
1089 skb->data_len += txp->size;
1090 skb->truesize += txp->size;
1092 /* Take an extra reference to offset xen_netbk_idx_release */
1093 get_page(netbk->mmap_pages[pending_idx]);
1094 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_OKAY);
1098 static int xen_netbk_get_extras(struct xenvif *vif,
1099 struct xen_netif_extra_info *extras,
1102 struct xen_netif_extra_info extra;
1103 RING_IDX cons = vif->tx.req_cons;
1106 if (unlikely(work_to_do-- <= 0)) {
1107 netdev_err(vif->dev, "Missing extra info\n");
1108 netbk_fatal_tx_err(vif);
1112 memcpy(&extra, RING_GET_REQUEST(&vif->tx, cons),
1114 if (unlikely(!extra.type ||
1115 extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
1116 vif->tx.req_cons = ++cons;
1117 netdev_err(vif->dev,
1118 "Invalid extra type: %d\n", extra.type);
1119 netbk_fatal_tx_err(vif);
1123 memcpy(&extras[extra.type - 1], &extra, sizeof(extra));
1124 vif->tx.req_cons = ++cons;
1125 } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE);
1130 static int netbk_set_skb_gso(struct xenvif *vif,
1131 struct sk_buff *skb,
1132 struct xen_netif_extra_info *gso)
1134 if (!gso->u.gso.size) {
1135 netdev_err(vif->dev, "GSO size must not be zero.\n");
1136 netbk_fatal_tx_err(vif);
1140 /* Currently only TCPv4 S.O. is supported. */
1141 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4) {
1142 netdev_err(vif->dev, "Bad GSO type %d.\n", gso->u.gso.type);
1143 netbk_fatal_tx_err(vif);
1147 skb_shinfo(skb)->gso_size = gso->u.gso.size;
1148 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1150 /* Header must be checked, and gso_segs computed. */
1151 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1152 skb_shinfo(skb)->gso_segs = 0;
1157 static int checksum_setup(struct xenvif *vif, struct sk_buff *skb)
1162 int recalculate_partial_csum = 0;
1165 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
1166 * peers can fail to set NETRXF_csum_blank when sending a GSO
1167 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
1168 * recalculate the partial checksum.
1170 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
1171 vif->rx_gso_checksum_fixup++;
1172 skb->ip_summed = CHECKSUM_PARTIAL;
1173 recalculate_partial_csum = 1;
1176 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
1177 if (skb->ip_summed != CHECKSUM_PARTIAL)
1180 if (skb->protocol != htons(ETH_P_IP))
1183 iph = (void *)skb->data;
1184 th = skb->data + 4 * iph->ihl;
1185 if (th >= skb_tail_pointer(skb))
1188 skb_set_transport_header(skb, 4 * iph->ihl);
1189 skb->csum_start = th - skb->head;
1190 switch (iph->protocol) {
1192 skb->csum_offset = offsetof(struct tcphdr, check);
1194 if (recalculate_partial_csum) {
1195 struct tcphdr *tcph = (struct tcphdr *)th;
1196 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
1197 skb->len - iph->ihl*4,
1202 skb->csum_offset = offsetof(struct udphdr, check);
1204 if (recalculate_partial_csum) {
1205 struct udphdr *udph = (struct udphdr *)th;
1206 udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
1207 skb->len - iph->ihl*4,
1212 if (net_ratelimit())
1213 netdev_err(vif->dev,
1214 "Attempting to checksum a non-TCP/UDP packet, dropping a protocol %d packet\n",
1219 if ((th + skb->csum_offset + 2) > skb_tail_pointer(skb))
1228 static bool tx_credit_exceeded(struct xenvif *vif, unsigned size)
1230 unsigned long now = jiffies;
1231 unsigned long next_credit =
1232 vif->credit_timeout.expires +
1233 msecs_to_jiffies(vif->credit_usec / 1000);
1235 /* Timer could already be pending in rare cases. */
1236 if (timer_pending(&vif->credit_timeout))
1239 /* Passed the point where we can replenish credit? */
1240 if (time_after_eq(now, next_credit)) {
1241 vif->credit_timeout.expires = now;
1245 /* Still too big to send right now? Set a callback. */
1246 if (size > vif->remaining_credit) {
1247 vif->credit_timeout.data =
1249 vif->credit_timeout.function =
1251 mod_timer(&vif->credit_timeout,
1260 static unsigned xen_netbk_tx_build_gops(struct xen_netbk *netbk)
1262 struct gnttab_copy *gop = netbk->tx_copy_ops, *request_gop;
1263 struct sk_buff *skb;
1266 while (((nr_pending_reqs(netbk) + MAX_SKB_FRAGS) < MAX_PENDING_REQS) &&
1267 !list_empty(&netbk->net_schedule_list)) {
1269 struct xen_netif_tx_request txreq;
1270 struct xen_netif_tx_request txfrags[MAX_SKB_FRAGS];
1272 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX-1];
1276 unsigned int data_len;
1277 pending_ring_idx_t index;
1279 /* Get a netif from the list with work to do. */
1280 vif = poll_net_schedule_list(netbk);
1281 /* This can sometimes happen because the test of
1282 * list_empty(net_schedule_list) at the top of the
1283 * loop is unlocked. Just go back and have another
1289 if (vif->tx.sring->req_prod - vif->tx.req_cons >
1290 XEN_NETIF_TX_RING_SIZE) {
1291 netdev_err(vif->dev,
1292 "Impossible number of requests. "
1293 "req_prod %d, req_cons %d, size %ld\n",
1294 vif->tx.sring->req_prod, vif->tx.req_cons,
1295 XEN_NETIF_TX_RING_SIZE);
1296 netbk_fatal_tx_err(vif);
1300 RING_FINAL_CHECK_FOR_REQUESTS(&vif->tx, work_to_do);
1306 idx = vif->tx.req_cons;
1307 rmb(); /* Ensure that we see the request before we copy it. */
1308 memcpy(&txreq, RING_GET_REQUEST(&vif->tx, idx), sizeof(txreq));
1310 /* Credit-based scheduling. */
1311 if (txreq.size > vif->remaining_credit &&
1312 tx_credit_exceeded(vif, txreq.size)) {
1317 vif->remaining_credit -= txreq.size;
1320 vif->tx.req_cons = ++idx;
1322 memset(extras, 0, sizeof(extras));
1323 if (txreq.flags & XEN_NETTXF_extra_info) {
1324 work_to_do = xen_netbk_get_extras(vif, extras,
1326 idx = vif->tx.req_cons;
1327 if (unlikely(work_to_do < 0))
1331 ret = netbk_count_requests(vif, &txreq, txfrags, work_to_do);
1332 if (unlikely(ret < 0))
1337 if (unlikely(txreq.size < ETH_HLEN)) {
1338 netdev_dbg(vif->dev,
1339 "Bad packet size: %d\n", txreq.size);
1340 netbk_tx_err(vif, &txreq, idx);
1344 /* No crossing a page as the payload mustn't fragment. */
1345 if (unlikely((txreq.offset + txreq.size) > PAGE_SIZE)) {
1346 netdev_err(vif->dev,
1347 "txreq.offset: %x, size: %u, end: %lu\n",
1348 txreq.offset, txreq.size,
1349 (txreq.offset&~PAGE_MASK) + txreq.size);
1350 netbk_fatal_tx_err(vif);
1354 index = pending_index(netbk->pending_cons);
1355 pending_idx = netbk->pending_ring[index];
1357 data_len = (txreq.size > PKT_PROT_LEN &&
1358 ret < MAX_SKB_FRAGS) ?
1359 PKT_PROT_LEN : txreq.size;
1361 skb = alloc_skb(data_len + NET_SKB_PAD + NET_IP_ALIGN,
1362 GFP_ATOMIC | __GFP_NOWARN);
1363 if (unlikely(skb == NULL)) {
1364 netdev_dbg(vif->dev,
1365 "Can't allocate a skb in start_xmit.\n");
1366 netbk_tx_err(vif, &txreq, idx);
1370 /* Packets passed to netif_rx() must have some headroom. */
1371 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
1373 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1374 struct xen_netif_extra_info *gso;
1375 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1377 if (netbk_set_skb_gso(vif, skb, gso)) {
1378 /* Failure in netbk_set_skb_gso is fatal. */
1384 /* XXX could copy straight to head */
1385 page = xen_netbk_alloc_page(netbk, pending_idx);
1388 netbk_tx_err(vif, &txreq, idx);
1392 gop->source.u.ref = txreq.gref;
1393 gop->source.domid = vif->domid;
1394 gop->source.offset = txreq.offset;
1396 gop->dest.u.gmfn = virt_to_mfn(page_address(page));
1397 gop->dest.domid = DOMID_SELF;
1398 gop->dest.offset = txreq.offset;
1400 gop->len = txreq.size;
1401 gop->flags = GNTCOPY_source_gref;
1405 memcpy(&netbk->pending_tx_info[pending_idx].req,
1406 &txreq, sizeof(txreq));
1407 netbk->pending_tx_info[pending_idx].vif = vif;
1408 *((u16 *)skb->data) = pending_idx;
1410 __skb_put(skb, data_len);
1412 skb_shinfo(skb)->nr_frags = ret;
1413 if (data_len < txreq.size) {
1414 skb_shinfo(skb)->nr_frags++;
1415 frag_set_pending_idx(&skb_shinfo(skb)->frags[0],
1418 frag_set_pending_idx(&skb_shinfo(skb)->frags[0],
1419 INVALID_PENDING_IDX);
1422 netbk->pending_cons++;
1424 request_gop = xen_netbk_get_requests(netbk, vif,
1426 if (request_gop == NULL) {
1428 netbk_tx_err(vif, &txreq, idx);
1433 __skb_queue_tail(&netbk->tx_queue, skb);
1435 vif->tx.req_cons = idx;
1436 xen_netbk_check_rx_xenvif(vif);
1438 if ((gop-netbk->tx_copy_ops) >= ARRAY_SIZE(netbk->tx_copy_ops))
1442 return gop - netbk->tx_copy_ops;
1445 static void xen_netbk_tx_submit(struct xen_netbk *netbk)
1447 struct gnttab_copy *gop = netbk->tx_copy_ops;
1448 struct sk_buff *skb;
1450 while ((skb = __skb_dequeue(&netbk->tx_queue)) != NULL) {
1451 struct xen_netif_tx_request *txp;
1456 pending_idx = *((u16 *)skb->data);
1457 vif = netbk->pending_tx_info[pending_idx].vif;
1458 txp = &netbk->pending_tx_info[pending_idx].req;
1460 /* Check the remap error code. */
1461 if (unlikely(xen_netbk_tx_check_gop(netbk, skb, &gop))) {
1462 netdev_dbg(vif->dev, "netback grant failed.\n");
1463 skb_shinfo(skb)->nr_frags = 0;
1468 data_len = skb->len;
1470 (void *)(idx_to_kaddr(netbk, pending_idx)|txp->offset),
1472 if (data_len < txp->size) {
1473 /* Append the packet payload as a fragment. */
1474 txp->offset += data_len;
1475 txp->size -= data_len;
1477 /* Schedule a response immediately. */
1478 xen_netbk_idx_release(netbk, pending_idx, XEN_NETIF_RSP_OKAY);
1481 if (txp->flags & XEN_NETTXF_csum_blank)
1482 skb->ip_summed = CHECKSUM_PARTIAL;
1483 else if (txp->flags & XEN_NETTXF_data_validated)
1484 skb->ip_summed = CHECKSUM_UNNECESSARY;
1486 xen_netbk_fill_frags(netbk, skb);
1489 * If the initial fragment was < PKT_PROT_LEN then
1490 * pull through some bytes from the other fragments to
1491 * increase the linear region to PKT_PROT_LEN bytes.
1493 if (skb_headlen(skb) < PKT_PROT_LEN && skb_is_nonlinear(skb)) {
1494 int target = min_t(int, skb->len, PKT_PROT_LEN);
1495 __pskb_pull_tail(skb, target - skb_headlen(skb));
1498 skb->dev = vif->dev;
1499 skb->protocol = eth_type_trans(skb, skb->dev);
1500 skb_reset_network_header(skb);
1502 if (checksum_setup(vif, skb)) {
1503 netdev_dbg(vif->dev,
1504 "Can't setup checksum in net_tx_action\n");
1509 if (!skb_transport_header_was_set(skb)) {
1510 struct flow_keys keys;
1512 if (skb_flow_dissect(skb, &keys))
1513 skb_set_transport_header(skb, keys.thoff);
1515 skb_reset_transport_header(skb);
1518 vif->dev->stats.rx_bytes += skb->len;
1519 vif->dev->stats.rx_packets++;
1521 xenvif_receive_skb(vif, skb);
1525 /* Called after netfront has transmitted */
1526 static void xen_netbk_tx_action(struct xen_netbk *netbk)
1530 nr_gops = xen_netbk_tx_build_gops(netbk);
1535 gnttab_batch_copy(netbk->tx_copy_ops, nr_gops);
1537 xen_netbk_tx_submit(netbk);
1540 static void xen_netbk_idx_release(struct xen_netbk *netbk, u16 pending_idx,
1544 struct pending_tx_info *pending_tx_info;
1545 pending_ring_idx_t index;
1547 /* Already complete? */
1548 if (netbk->mmap_pages[pending_idx] == NULL)
1551 pending_tx_info = &netbk->pending_tx_info[pending_idx];
1553 vif = pending_tx_info->vif;
1555 make_tx_response(vif, &pending_tx_info->req, status);
1557 index = pending_index(netbk->pending_prod++);
1558 netbk->pending_ring[index] = pending_idx;
1562 netbk->mmap_pages[pending_idx]->mapping = 0;
1563 put_page(netbk->mmap_pages[pending_idx]);
1564 netbk->mmap_pages[pending_idx] = NULL;
1567 static void make_tx_response(struct xenvif *vif,
1568 struct xen_netif_tx_request *txp,
1571 RING_IDX i = vif->tx.rsp_prod_pvt;
1572 struct xen_netif_tx_response *resp;
1575 resp = RING_GET_RESPONSE(&vif->tx, i);
1579 if (txp->flags & XEN_NETTXF_extra_info)
1580 RING_GET_RESPONSE(&vif->tx, ++i)->status = XEN_NETIF_RSP_NULL;
1582 vif->tx.rsp_prod_pvt = ++i;
1583 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->tx, notify);
1585 notify_remote_via_irq(vif->irq);
1588 static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif,
1595 RING_IDX i = vif->rx.rsp_prod_pvt;
1596 struct xen_netif_rx_response *resp;
1598 resp = RING_GET_RESPONSE(&vif->rx, i);
1599 resp->offset = offset;
1600 resp->flags = flags;
1602 resp->status = (s16)size;
1604 resp->status = (s16)st;
1606 vif->rx.rsp_prod_pvt = ++i;
1611 static inline int rx_work_todo(struct xen_netbk *netbk)
1613 return !skb_queue_empty(&netbk->rx_queue);
1616 static inline int tx_work_todo(struct xen_netbk *netbk)
1619 if (((nr_pending_reqs(netbk) + MAX_SKB_FRAGS) < MAX_PENDING_REQS) &&
1620 !list_empty(&netbk->net_schedule_list))
1626 static int xen_netbk_kthread(void *data)
1628 struct xen_netbk *netbk = data;
1629 while (!kthread_should_stop()) {
1630 wait_event_interruptible(netbk->wq,
1631 rx_work_todo(netbk) ||
1632 tx_work_todo(netbk) ||
1633 kthread_should_stop());
1636 if (kthread_should_stop())
1639 if (rx_work_todo(netbk))
1640 xen_netbk_rx_action(netbk);
1642 if (tx_work_todo(netbk))
1643 xen_netbk_tx_action(netbk);
1649 void xen_netbk_unmap_frontend_rings(struct xenvif *vif)
1652 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
1655 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
1659 int xen_netbk_map_frontend_rings(struct xenvif *vif,
1660 grant_ref_t tx_ring_ref,
1661 grant_ref_t rx_ring_ref)
1664 struct xen_netif_tx_sring *txs;
1665 struct xen_netif_rx_sring *rxs;
1669 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif),
1670 tx_ring_ref, &addr);
1674 txs = (struct xen_netif_tx_sring *)addr;
1675 BACK_RING_INIT(&vif->tx, txs, PAGE_SIZE);
1677 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(vif),
1678 rx_ring_ref, &addr);
1682 rxs = (struct xen_netif_rx_sring *)addr;
1683 BACK_RING_INIT(&vif->rx, rxs, PAGE_SIZE);
1685 vif->rx_req_cons_peek = 0;
1690 xen_netbk_unmap_frontend_rings(vif);
1694 static int __init netback_init(void)
1703 xen_netbk_group_nr = num_online_cpus();
1704 xen_netbk = vzalloc(sizeof(struct xen_netbk) * xen_netbk_group_nr);
1708 for (group = 0; group < xen_netbk_group_nr; group++) {
1709 struct xen_netbk *netbk = &xen_netbk[group];
1710 skb_queue_head_init(&netbk->rx_queue);
1711 skb_queue_head_init(&netbk->tx_queue);
1713 init_timer(&netbk->net_timer);
1714 netbk->net_timer.data = (unsigned long)netbk;
1715 netbk->net_timer.function = xen_netbk_alarm;
1717 netbk->pending_cons = 0;
1718 netbk->pending_prod = MAX_PENDING_REQS;
1719 for (i = 0; i < MAX_PENDING_REQS; i++)
1720 netbk->pending_ring[i] = i;
1722 init_waitqueue_head(&netbk->wq);
1723 netbk->task = kthread_create(xen_netbk_kthread,
1725 "netback/%u", group);
1727 if (IS_ERR(netbk->task)) {
1728 printk(KERN_ALERT "kthread_create() fails at netback\n");
1729 del_timer(&netbk->net_timer);
1730 rc = PTR_ERR(netbk->task);
1734 kthread_bind(netbk->task, group);
1736 INIT_LIST_HEAD(&netbk->net_schedule_list);
1738 spin_lock_init(&netbk->net_schedule_list_lock);
1740 atomic_set(&netbk->netfront_count, 0);
1742 wake_up_process(netbk->task);
1745 rc = xenvif_xenbus_init();
1752 while (--group >= 0) {
1753 struct xen_netbk *netbk = &xen_netbk[group];
1754 for (i = 0; i < MAX_PENDING_REQS; i++) {
1755 if (netbk->mmap_pages[i])
1756 __free_page(netbk->mmap_pages[i]);
1758 del_timer(&netbk->net_timer);
1759 kthread_stop(netbk->task);
1766 module_init(netback_init);
1768 MODULE_LICENSE("Dual BSD/GPL");
1769 MODULE_ALIAS("xen-backend:vif");