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>
43
#include <xen/events.h>
44
#include <xen/interface/memory.h>
46
#include <asm/xen/hypercall.h>
47
#include <asm/xen/page.h>
49
struct pending_tx_info {
50
struct xen_netif_tx_request req;
53
typedef unsigned int pending_ring_idx_t;
55
struct netbk_rx_meta {
61
#define MAX_PENDING_REQS 256
63
#define MAX_BUFFER_OFFSET PAGE_SIZE
65
/* extra field used in struct page */
68
#if BITS_PER_LONG < 64
70
#define GROUP_WIDTH (BITS_PER_LONG - IDX_WIDTH)
71
unsigned int group:GROUP_WIDTH;
72
unsigned int idx:IDX_WIDTH;
74
unsigned int group, idx;
82
struct task_struct *task;
84
struct sk_buff_head rx_queue;
85
struct sk_buff_head tx_queue;
87
struct timer_list net_timer;
89
struct page *mmap_pages[MAX_PENDING_REQS];
91
pending_ring_idx_t pending_prod;
92
pending_ring_idx_t pending_cons;
93
struct list_head net_schedule_list;
95
/* Protect the net_schedule_list in netif. */
96
spinlock_t net_schedule_list_lock;
98
atomic_t netfront_count;
100
struct pending_tx_info pending_tx_info[MAX_PENDING_REQS];
101
struct gnttab_copy tx_copy_ops[MAX_PENDING_REQS];
103
u16 pending_ring[MAX_PENDING_REQS];
106
* Given MAX_BUFFER_OFFSET of 4096 the worst case is that each
107
* head/fragment page uses 2 copy operations because it
108
* straddles two buffers in the frontend.
110
struct gnttab_copy grant_copy_op[2*XEN_NETIF_RX_RING_SIZE];
111
struct netbk_rx_meta meta[2*XEN_NETIF_RX_RING_SIZE];
114
static struct xen_netbk *xen_netbk;
115
static int xen_netbk_group_nr;
117
void xen_netbk_add_xenvif(struct xenvif *vif)
120
int min_netfront_count;
122
struct xen_netbk *netbk;
124
min_netfront_count = atomic_read(&xen_netbk[0].netfront_count);
125
for (i = 0; i < xen_netbk_group_nr; i++) {
126
int netfront_count = atomic_read(&xen_netbk[i].netfront_count);
127
if (netfront_count < min_netfront_count) {
129
min_netfront_count = netfront_count;
133
netbk = &xen_netbk[min_group];
136
atomic_inc(&netbk->netfront_count);
139
void xen_netbk_remove_xenvif(struct xenvif *vif)
141
struct xen_netbk *netbk = vif->netbk;
143
atomic_dec(&netbk->netfront_count);
146
static void xen_netbk_idx_release(struct xen_netbk *netbk, u16 pending_idx);
147
static void make_tx_response(struct xenvif *vif,
148
struct xen_netif_tx_request *txp,
150
static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif,
157
static inline unsigned long idx_to_pfn(struct xen_netbk *netbk,
160
return page_to_pfn(netbk->mmap_pages[idx]);
163
static inline unsigned long idx_to_kaddr(struct xen_netbk *netbk,
166
return (unsigned long)pfn_to_kaddr(idx_to_pfn(netbk, idx));
169
/* extra field used in struct page */
170
static inline void set_page_ext(struct page *pg, struct xen_netbk *netbk,
173
unsigned int group = netbk - xen_netbk;
174
union page_ext ext = { .e = { .group = group + 1, .idx = idx } };
176
BUILD_BUG_ON(sizeof(ext) > sizeof(ext.mapping));
177
pg->mapping = ext.mapping;
180
static int get_page_ext(struct page *pg,
181
unsigned int *pgroup, unsigned int *pidx)
183
union page_ext ext = { .mapping = pg->mapping };
184
struct xen_netbk *netbk;
185
unsigned int group, idx;
187
group = ext.e.group - 1;
189
if (group < 0 || group >= xen_netbk_group_nr)
192
netbk = &xen_netbk[group];
196
if ((idx < 0) || (idx >= MAX_PENDING_REQS))
199
if (netbk->mmap_pages[idx] != pg)
209
* This is the amount of packet we copy rather than map, so that the
210
* guest can't fiddle with the contents of the headers while we do
211
* packet processing on them (netfilter, routing, etc).
213
#define PKT_PROT_LEN (ETH_HLEN + \
215
sizeof(struct iphdr) + MAX_IPOPTLEN + \
216
sizeof(struct tcphdr) + MAX_TCP_OPTION_SPACE)
218
static inline pending_ring_idx_t pending_index(unsigned i)
220
return i & (MAX_PENDING_REQS-1);
223
static inline pending_ring_idx_t nr_pending_reqs(struct xen_netbk *netbk)
225
return MAX_PENDING_REQS -
226
netbk->pending_prod + netbk->pending_cons;
229
static void xen_netbk_kick_thread(struct xen_netbk *netbk)
234
static int max_required_rx_slots(struct xenvif *vif)
236
int max = DIV_ROUND_UP(vif->dev->mtu, PAGE_SIZE);
238
if (vif->can_sg || vif->gso || vif->gso_prefix)
239
max += MAX_SKB_FRAGS + 1; /* extra_info + frags */
244
int xen_netbk_rx_ring_full(struct xenvif *vif)
246
RING_IDX peek = vif->rx_req_cons_peek;
247
RING_IDX needed = max_required_rx_slots(vif);
249
return ((vif->rx.sring->req_prod - peek) < needed) ||
250
((vif->rx.rsp_prod_pvt + XEN_NETIF_RX_RING_SIZE - peek) < needed);
253
int xen_netbk_must_stop_queue(struct xenvif *vif)
255
if (!xen_netbk_rx_ring_full(vif))
258
vif->rx.sring->req_event = vif->rx_req_cons_peek +
259
max_required_rx_slots(vif);
260
mb(); /* request notification /then/ check the queue */
262
return xen_netbk_rx_ring_full(vif);
266
* Returns true if we should start a new receive buffer instead of
267
* adding 'size' bytes to a buffer which currently contains 'offset'
270
static bool start_new_rx_buffer(int offset, unsigned long size, int head)
272
/* simple case: we have completely filled the current buffer. */
273
if (offset == MAX_BUFFER_OFFSET)
277
* complex case: start a fresh buffer if the current frag
278
* would overflow the current buffer but only if:
279
* (i) this frag would fit completely in the next buffer
280
* and (ii) there is already some data in the current buffer
281
* and (iii) this is not the head buffer.
284
* - (i) stops us splitting a frag into two copies
285
* unless the frag is too large for a single buffer.
286
* - (ii) stops us from leaving a buffer pointlessly empty.
287
* - (iii) stops us leaving the first buffer
288
* empty. Strictly speaking this is already covered
289
* by (ii) but is explicitly checked because
290
* netfront relies on the first buffer being
291
* non-empty and can crash otherwise.
293
* This means we will effectively linearise small
294
* frags but do not needlessly split large buffers
295
* into multiple copies tend to give large frags their
296
* own buffers as before.
298
if ((offset + size > MAX_BUFFER_OFFSET) &&
299
(size <= MAX_BUFFER_OFFSET) && offset && !head)
306
* Figure out how many ring slots we're going to need to send @skb to
307
* the guest. This function is essentially a dry run of
308
* netbk_gop_frag_copy.
310
unsigned int xen_netbk_count_skb_slots(struct xenvif *vif, struct sk_buff *skb)
315
count = DIV_ROUND_UP(
316
offset_in_page(skb->data)+skb_headlen(skb), PAGE_SIZE);
318
copy_off = skb_headlen(skb) % PAGE_SIZE;
320
if (skb_shinfo(skb)->gso_size)
323
for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
324
unsigned long size = skb_shinfo(skb)->frags[i].size;
327
BUG_ON(copy_off > MAX_BUFFER_OFFSET);
329
if (start_new_rx_buffer(copy_off, size, 0)) {
335
if (copy_off + bytes > MAX_BUFFER_OFFSET)
336
bytes = MAX_BUFFER_OFFSET - copy_off;
345
struct netrx_pending_operations {
346
unsigned copy_prod, copy_cons;
347
unsigned meta_prod, meta_cons;
348
struct gnttab_copy *copy;
349
struct netbk_rx_meta *meta;
351
grant_ref_t copy_gref;
354
static struct netbk_rx_meta *get_next_rx_buffer(struct xenvif *vif,
355
struct netrx_pending_operations *npo)
357
struct netbk_rx_meta *meta;
358
struct xen_netif_rx_request *req;
360
req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++);
362
meta = npo->meta + npo->meta_prod++;
368
npo->copy_gref = req->gref;
374
* Set up the grant operations for this fragment. If it's a flipping
375
* interface, we also set up the unmap request from here.
377
static void netbk_gop_frag_copy(struct xenvif *vif, struct sk_buff *skb,
378
struct netrx_pending_operations *npo,
379
struct page *page, unsigned long size,
380
unsigned long offset, int *head)
382
struct gnttab_copy *copy_gop;
383
struct netbk_rx_meta *meta;
385
* These variables a used iff get_page_ext returns true,
386
* in which case they are guaranteed to be initialized.
388
unsigned int uninitialized_var(group), uninitialized_var(idx);
389
int foreign = get_page_ext(page, &group, &idx);
392
/* Data must not cross a page boundary. */
393
BUG_ON(size + offset > PAGE_SIZE);
395
meta = npo->meta + npo->meta_prod - 1;
398
BUG_ON(npo->copy_off > MAX_BUFFER_OFFSET);
400
if (start_new_rx_buffer(npo->copy_off, size, *head)) {
402
* Netfront requires there to be some data in the head
407
meta = get_next_rx_buffer(vif, npo);
411
if (npo->copy_off + bytes > MAX_BUFFER_OFFSET)
412
bytes = MAX_BUFFER_OFFSET - npo->copy_off;
414
copy_gop = npo->copy + npo->copy_prod++;
415
copy_gop->flags = GNTCOPY_dest_gref;
417
struct xen_netbk *netbk = &xen_netbk[group];
418
struct pending_tx_info *src_pend;
420
src_pend = &netbk->pending_tx_info[idx];
422
copy_gop->source.domid = src_pend->vif->domid;
423
copy_gop->source.u.ref = src_pend->req.gref;
424
copy_gop->flags |= GNTCOPY_source_gref;
426
void *vaddr = page_address(page);
427
copy_gop->source.domid = DOMID_SELF;
428
copy_gop->source.u.gmfn = virt_to_mfn(vaddr);
430
copy_gop->source.offset = offset;
431
copy_gop->dest.domid = vif->domid;
433
copy_gop->dest.offset = npo->copy_off;
434
copy_gop->dest.u.ref = npo->copy_gref;
435
copy_gop->len = bytes;
437
npo->copy_off += bytes;
443
/* Leave a gap for the GSO descriptor. */
444
if (*head && skb_shinfo(skb)->gso_size && !vif->gso_prefix)
447
*head = 0; /* There must be something in this buffer now. */
453
* Prepare an SKB to be transmitted to the frontend.
455
* This function is responsible for allocating grant operations, meta
458
* It returns the number of meta structures consumed. The number of
459
* ring slots used is always equal to the number of meta slots used
460
* plus the number of GSO descriptors used. Currently, we use either
461
* zero GSO descriptors (for non-GSO packets) or one descriptor (for
462
* frontend-side LRO).
464
static int netbk_gop_skb(struct sk_buff *skb,
465
struct netrx_pending_operations *npo)
467
struct xenvif *vif = netdev_priv(skb->dev);
468
int nr_frags = skb_shinfo(skb)->nr_frags;
470
struct xen_netif_rx_request *req;
471
struct netbk_rx_meta *meta;
476
old_meta_prod = npo->meta_prod;
478
/* Set up a GSO prefix descriptor, if necessary */
479
if (skb_shinfo(skb)->gso_size && vif->gso_prefix) {
480
req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++);
481
meta = npo->meta + npo->meta_prod++;
482
meta->gso_size = skb_shinfo(skb)->gso_size;
487
req = RING_GET_REQUEST(&vif->rx, vif->rx.req_cons++);
488
meta = npo->meta + npo->meta_prod++;
490
if (!vif->gso_prefix)
491
meta->gso_size = skb_shinfo(skb)->gso_size;
498
npo->copy_gref = req->gref;
501
while (data < skb_tail_pointer(skb)) {
502
unsigned int offset = offset_in_page(data);
503
unsigned int len = PAGE_SIZE - offset;
505
if (data + len > skb_tail_pointer(skb))
506
len = skb_tail_pointer(skb) - data;
508
netbk_gop_frag_copy(vif, skb, npo,
509
virt_to_page(data), len, offset, &head);
513
for (i = 0; i < nr_frags; i++) {
514
netbk_gop_frag_copy(vif, skb, npo,
515
skb_shinfo(skb)->frags[i].page,
516
skb_shinfo(skb)->frags[i].size,
517
skb_shinfo(skb)->frags[i].page_offset,
521
return npo->meta_prod - old_meta_prod;
525
* This is a twin to netbk_gop_skb. Assume that netbk_gop_skb was
526
* used to set up the operations on the top of
527
* netrx_pending_operations, which have since been done. Check that
528
* they didn't give any errors and advance over them.
530
static int netbk_check_gop(struct xenvif *vif, int nr_meta_slots,
531
struct netrx_pending_operations *npo)
533
struct gnttab_copy *copy_op;
534
int status = XEN_NETIF_RSP_OKAY;
537
for (i = 0; i < nr_meta_slots; i++) {
538
copy_op = npo->copy + npo->copy_cons++;
539
if (copy_op->status != GNTST_okay) {
541
"Bad status %d from copy to DOM%d.\n",
542
copy_op->status, vif->domid);
543
status = XEN_NETIF_RSP_ERROR;
550
static void netbk_add_frag_responses(struct xenvif *vif, int status,
551
struct netbk_rx_meta *meta,
555
unsigned long offset;
557
/* No fragments used */
558
if (nr_meta_slots <= 1)
563
for (i = 0; i < nr_meta_slots; i++) {
565
if (i == nr_meta_slots - 1)
568
flags = XEN_NETRXF_more_data;
571
make_rx_response(vif, meta[i].id, status, offset,
572
meta[i].size, flags);
576
struct skb_cb_overlay {
580
static void xen_netbk_rx_action(struct xen_netbk *netbk)
582
struct xenvif *vif = NULL, *tmp;
585
struct xen_netif_rx_response *resp;
586
struct sk_buff_head rxq;
592
unsigned long offset;
593
struct skb_cb_overlay *sco;
595
struct netrx_pending_operations npo = {
596
.copy = netbk->grant_copy_op,
600
skb_queue_head_init(&rxq);
604
while ((skb = skb_dequeue(&netbk->rx_queue)) != NULL) {
605
vif = netdev_priv(skb->dev);
606
nr_frags = skb_shinfo(skb)->nr_frags;
608
sco = (struct skb_cb_overlay *)skb->cb;
609
sco->meta_slots_used = netbk_gop_skb(skb, &npo);
611
count += nr_frags + 1;
613
__skb_queue_tail(&rxq, skb);
615
/* Filled the batch queue? */
616
if (count + MAX_SKB_FRAGS >= XEN_NETIF_RX_RING_SIZE)
620
BUG_ON(npo.meta_prod > ARRAY_SIZE(netbk->meta));
625
BUG_ON(npo.copy_prod > ARRAY_SIZE(netbk->grant_copy_op));
626
ret = HYPERVISOR_grant_table_op(GNTTABOP_copy, &netbk->grant_copy_op,
630
while ((skb = __skb_dequeue(&rxq)) != NULL) {
631
sco = (struct skb_cb_overlay *)skb->cb;
633
vif = netdev_priv(skb->dev);
635
if (netbk->meta[npo.meta_cons].gso_size && vif->gso_prefix) {
636
resp = RING_GET_RESPONSE(&vif->rx,
637
vif->rx.rsp_prod_pvt++);
639
resp->flags = XEN_NETRXF_gso_prefix | XEN_NETRXF_more_data;
641
resp->offset = netbk->meta[npo.meta_cons].gso_size;
642
resp->id = netbk->meta[npo.meta_cons].id;
643
resp->status = sco->meta_slots_used;
646
sco->meta_slots_used--;
650
vif->dev->stats.tx_bytes += skb->len;
651
vif->dev->stats.tx_packets++;
653
status = netbk_check_gop(vif, sco->meta_slots_used, &npo);
655
if (sco->meta_slots_used == 1)
658
flags = XEN_NETRXF_more_data;
660
if (skb->ip_summed == CHECKSUM_PARTIAL) /* local packet? */
661
flags |= XEN_NETRXF_csum_blank | XEN_NETRXF_data_validated;
662
else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
663
/* remote but checksummed. */
664
flags |= XEN_NETRXF_data_validated;
667
resp = make_rx_response(vif, netbk->meta[npo.meta_cons].id,
669
netbk->meta[npo.meta_cons].size,
672
if (netbk->meta[npo.meta_cons].gso_size && !vif->gso_prefix) {
673
struct xen_netif_extra_info *gso =
674
(struct xen_netif_extra_info *)
675
RING_GET_RESPONSE(&vif->rx,
676
vif->rx.rsp_prod_pvt++);
678
resp->flags |= XEN_NETRXF_extra_info;
680
gso->u.gso.size = netbk->meta[npo.meta_cons].gso_size;
681
gso->u.gso.type = XEN_NETIF_GSO_TYPE_TCPV4;
683
gso->u.gso.features = 0;
685
gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
689
netbk_add_frag_responses(vif, status,
690
netbk->meta + npo.meta_cons + 1,
691
sco->meta_slots_used);
693
RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->rx, ret);
695
if (ret && list_empty(&vif->notify_list))
696
list_add_tail(&vif->notify_list, ¬ify);
698
xenvif_notify_tx_completion(vif);
701
npo.meta_cons += sco->meta_slots_used;
705
list_for_each_entry_safe(vif, tmp, ¬ify, notify_list) {
706
notify_remote_via_irq(vif->irq);
707
list_del_init(&vif->notify_list);
710
/* More work to do? */
711
if (!skb_queue_empty(&netbk->rx_queue) &&
712
!timer_pending(&netbk->net_timer))
713
xen_netbk_kick_thread(netbk);
716
void xen_netbk_queue_tx_skb(struct xenvif *vif, struct sk_buff *skb)
718
struct xen_netbk *netbk = vif->netbk;
720
skb_queue_tail(&netbk->rx_queue, skb);
722
xen_netbk_kick_thread(netbk);
725
static void xen_netbk_alarm(unsigned long data)
727
struct xen_netbk *netbk = (struct xen_netbk *)data;
728
xen_netbk_kick_thread(netbk);
731
static int __on_net_schedule_list(struct xenvif *vif)
733
return !list_empty(&vif->schedule_list);
736
/* Must be called with net_schedule_list_lock held */
737
static void remove_from_net_schedule_list(struct xenvif *vif)
739
if (likely(__on_net_schedule_list(vif))) {
740
list_del_init(&vif->schedule_list);
745
static struct xenvif *poll_net_schedule_list(struct xen_netbk *netbk)
747
struct xenvif *vif = NULL;
749
spin_lock_irq(&netbk->net_schedule_list_lock);
750
if (list_empty(&netbk->net_schedule_list))
753
vif = list_first_entry(&netbk->net_schedule_list,
754
struct xenvif, schedule_list);
760
remove_from_net_schedule_list(vif);
762
spin_unlock_irq(&netbk->net_schedule_list_lock);
766
void xen_netbk_schedule_xenvif(struct xenvif *vif)
769
struct xen_netbk *netbk = vif->netbk;
771
if (__on_net_schedule_list(vif))
774
spin_lock_irqsave(&netbk->net_schedule_list_lock, flags);
775
if (!__on_net_schedule_list(vif) &&
776
likely(xenvif_schedulable(vif))) {
777
list_add_tail(&vif->schedule_list, &netbk->net_schedule_list);
780
spin_unlock_irqrestore(&netbk->net_schedule_list_lock, flags);
784
if ((nr_pending_reqs(netbk) < (MAX_PENDING_REQS/2)) &&
785
!list_empty(&netbk->net_schedule_list))
786
xen_netbk_kick_thread(netbk);
789
void xen_netbk_deschedule_xenvif(struct xenvif *vif)
791
struct xen_netbk *netbk = vif->netbk;
792
spin_lock_irq(&netbk->net_schedule_list_lock);
793
remove_from_net_schedule_list(vif);
794
spin_unlock_irq(&netbk->net_schedule_list_lock);
797
void xen_netbk_check_rx_xenvif(struct xenvif *vif)
801
RING_FINAL_CHECK_FOR_REQUESTS(&vif->tx, more_to_do);
804
xen_netbk_schedule_xenvif(vif);
807
static void tx_add_credit(struct xenvif *vif)
809
unsigned long max_burst, max_credit;
812
* Allow a burst big enough to transmit a jumbo packet of up to 128kB.
813
* Otherwise the interface can seize up due to insufficient credit.
815
max_burst = RING_GET_REQUEST(&vif->tx, vif->tx.req_cons)->size;
816
max_burst = min(max_burst, 131072UL);
817
max_burst = max(max_burst, vif->credit_bytes);
819
/* Take care that adding a new chunk of credit doesn't wrap to zero. */
820
max_credit = vif->remaining_credit + vif->credit_bytes;
821
if (max_credit < vif->remaining_credit)
822
max_credit = ULONG_MAX; /* wrapped: clamp to ULONG_MAX */
824
vif->remaining_credit = min(max_credit, max_burst);
827
static void tx_credit_callback(unsigned long data)
829
struct xenvif *vif = (struct xenvif *)data;
831
xen_netbk_check_rx_xenvif(vif);
834
static void netbk_tx_err(struct xenvif *vif,
835
struct xen_netif_tx_request *txp, RING_IDX end)
837
RING_IDX cons = vif->tx.req_cons;
840
make_tx_response(vif, txp, XEN_NETIF_RSP_ERROR);
843
txp = RING_GET_REQUEST(&vif->tx, cons++);
845
vif->tx.req_cons = cons;
846
xen_netbk_check_rx_xenvif(vif);
850
static int netbk_count_requests(struct xenvif *vif,
851
struct xen_netif_tx_request *first,
852
struct xen_netif_tx_request *txp,
855
RING_IDX cons = vif->tx.req_cons;
858
if (!(first->flags & XEN_NETTXF_more_data))
862
if (frags >= work_to_do) {
863
netdev_dbg(vif->dev, "Need more frags\n");
867
if (unlikely(frags >= MAX_SKB_FRAGS)) {
868
netdev_dbg(vif->dev, "Too many frags\n");
872
memcpy(txp, RING_GET_REQUEST(&vif->tx, cons + frags),
874
if (txp->size > first->size) {
875
netdev_dbg(vif->dev, "Frags galore\n");
879
first->size -= txp->size;
882
if (unlikely((txp->offset + txp->size) > PAGE_SIZE)) {
883
netdev_dbg(vif->dev, "txp->offset: %x, size: %u\n",
884
txp->offset, txp->size);
887
} while ((txp++)->flags & XEN_NETTXF_more_data);
891
static struct page *xen_netbk_alloc_page(struct xen_netbk *netbk,
893
unsigned long pending_idx)
896
page = alloc_page(GFP_KERNEL|__GFP_COLD);
899
set_page_ext(page, netbk, pending_idx);
900
netbk->mmap_pages[pending_idx] = page;
904
static struct gnttab_copy *xen_netbk_get_requests(struct xen_netbk *netbk,
907
struct xen_netif_tx_request *txp,
908
struct gnttab_copy *gop)
910
struct skb_shared_info *shinfo = skb_shinfo(skb);
911
skb_frag_t *frags = shinfo->frags;
912
unsigned long pending_idx = *((u16 *)skb->data);
915
/* Skip first skb fragment if it is on same page as header fragment. */
916
start = ((unsigned long)shinfo->frags[0].page == pending_idx);
918
for (i = start; i < shinfo->nr_frags; i++, txp++) {
920
pending_ring_idx_t index;
921
struct pending_tx_info *pending_tx_info =
922
netbk->pending_tx_info;
924
index = pending_index(netbk->pending_cons++);
925
pending_idx = netbk->pending_ring[index];
926
page = xen_netbk_alloc_page(netbk, skb, pending_idx);
930
netbk->mmap_pages[pending_idx] = page;
932
gop->source.u.ref = txp->gref;
933
gop->source.domid = vif->domid;
934
gop->source.offset = txp->offset;
936
gop->dest.u.gmfn = virt_to_mfn(page_address(page));
937
gop->dest.domid = DOMID_SELF;
938
gop->dest.offset = txp->offset;
940
gop->len = txp->size;
941
gop->flags = GNTCOPY_source_gref;
945
memcpy(&pending_tx_info[pending_idx].req, txp, sizeof(*txp));
947
pending_tx_info[pending_idx].vif = vif;
948
frags[i].page = (void *)pending_idx;
954
static int xen_netbk_tx_check_gop(struct xen_netbk *netbk,
956
struct gnttab_copy **gopp)
958
struct gnttab_copy *gop = *gopp;
959
int pending_idx = *((u16 *)skb->data);
960
struct pending_tx_info *pending_tx_info = netbk->pending_tx_info;
961
struct xenvif *vif = pending_tx_info[pending_idx].vif;
962
struct xen_netif_tx_request *txp;
963
struct skb_shared_info *shinfo = skb_shinfo(skb);
964
int nr_frags = shinfo->nr_frags;
967
/* Check status of header. */
970
pending_ring_idx_t index;
971
index = pending_index(netbk->pending_prod++);
972
txp = &pending_tx_info[pending_idx].req;
973
make_tx_response(vif, txp, XEN_NETIF_RSP_ERROR);
974
netbk->pending_ring[index] = pending_idx;
978
/* Skip first skb fragment if it is on same page as header fragment. */
979
start = ((unsigned long)shinfo->frags[0].page == pending_idx);
981
for (i = start; i < nr_frags; i++) {
983
pending_ring_idx_t index;
985
pending_idx = (unsigned long)shinfo->frags[i].page;
987
/* Check error status: if okay then remember grant handle. */
988
newerr = (++gop)->status;
989
if (likely(!newerr)) {
990
/* Had a previous error? Invalidate this fragment. */
992
xen_netbk_idx_release(netbk, pending_idx);
996
/* Error on this fragment: respond to client with an error. */
997
txp = &netbk->pending_tx_info[pending_idx].req;
998
make_tx_response(vif, txp, XEN_NETIF_RSP_ERROR);
999
index = pending_index(netbk->pending_prod++);
1000
netbk->pending_ring[index] = pending_idx;
1003
/* Not the first error? Preceding frags already invalidated. */
1007
/* First error: invalidate header and preceding fragments. */
1008
pending_idx = *((u16 *)skb->data);
1009
xen_netbk_idx_release(netbk, pending_idx);
1010
for (j = start; j < i; j++) {
1011
pending_idx = (unsigned long)shinfo->frags[i].page;
1012
xen_netbk_idx_release(netbk, pending_idx);
1015
/* Remember the error: invalidate all subsequent fragments. */
1023
static void xen_netbk_fill_frags(struct xen_netbk *netbk, struct sk_buff *skb)
1025
struct skb_shared_info *shinfo = skb_shinfo(skb);
1026
int nr_frags = shinfo->nr_frags;
1029
for (i = 0; i < nr_frags; i++) {
1030
skb_frag_t *frag = shinfo->frags + i;
1031
struct xen_netif_tx_request *txp;
1032
unsigned long pending_idx;
1034
pending_idx = (unsigned long)frag->page;
1036
txp = &netbk->pending_tx_info[pending_idx].req;
1037
frag->page = virt_to_page(idx_to_kaddr(netbk, pending_idx));
1038
frag->size = txp->size;
1039
frag->page_offset = txp->offset;
1041
skb->len += txp->size;
1042
skb->data_len += txp->size;
1043
skb->truesize += txp->size;
1045
/* Take an extra reference to offset xen_netbk_idx_release */
1046
get_page(netbk->mmap_pages[pending_idx]);
1047
xen_netbk_idx_release(netbk, pending_idx);
1051
static int xen_netbk_get_extras(struct xenvif *vif,
1052
struct xen_netif_extra_info *extras,
1055
struct xen_netif_extra_info extra;
1056
RING_IDX cons = vif->tx.req_cons;
1059
if (unlikely(work_to_do-- <= 0)) {
1060
netdev_dbg(vif->dev, "Missing extra info\n");
1064
memcpy(&extra, RING_GET_REQUEST(&vif->tx, cons),
1066
if (unlikely(!extra.type ||
1067
extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
1068
vif->tx.req_cons = ++cons;
1069
netdev_dbg(vif->dev,
1070
"Invalid extra type: %d\n", extra.type);
1074
memcpy(&extras[extra.type - 1], &extra, sizeof(extra));
1075
vif->tx.req_cons = ++cons;
1076
} while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE);
1081
static int netbk_set_skb_gso(struct xenvif *vif,
1082
struct sk_buff *skb,
1083
struct xen_netif_extra_info *gso)
1085
if (!gso->u.gso.size) {
1086
netdev_dbg(vif->dev, "GSO size must not be zero.\n");
1090
/* Currently only TCPv4 S.O. is supported. */
1091
if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4) {
1092
netdev_dbg(vif->dev, "Bad GSO type %d.\n", gso->u.gso.type);
1096
skb_shinfo(skb)->gso_size = gso->u.gso.size;
1097
skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
1099
/* Header must be checked, and gso_segs computed. */
1100
skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1101
skb_shinfo(skb)->gso_segs = 0;
1106
static int checksum_setup(struct xenvif *vif, struct sk_buff *skb)
1111
int recalculate_partial_csum = 0;
1114
* A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
1115
* peers can fail to set NETRXF_csum_blank when sending a GSO
1116
* frame. In this case force the SKB to CHECKSUM_PARTIAL and
1117
* recalculate the partial checksum.
1119
if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
1120
vif->rx_gso_checksum_fixup++;
1121
skb->ip_summed = CHECKSUM_PARTIAL;
1122
recalculate_partial_csum = 1;
1125
/* A non-CHECKSUM_PARTIAL SKB does not require setup. */
1126
if (skb->ip_summed != CHECKSUM_PARTIAL)
1129
if (skb->protocol != htons(ETH_P_IP))
1132
iph = (void *)skb->data;
1133
th = skb->data + 4 * iph->ihl;
1134
if (th >= skb_tail_pointer(skb))
1137
skb->csum_start = th - skb->head;
1138
switch (iph->protocol) {
1140
skb->csum_offset = offsetof(struct tcphdr, check);
1142
if (recalculate_partial_csum) {
1143
struct tcphdr *tcph = (struct tcphdr *)th;
1144
tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
1145
skb->len - iph->ihl*4,
1150
skb->csum_offset = offsetof(struct udphdr, check);
1152
if (recalculate_partial_csum) {
1153
struct udphdr *udph = (struct udphdr *)th;
1154
udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
1155
skb->len - iph->ihl*4,
1160
if (net_ratelimit())
1161
netdev_err(vif->dev,
1162
"Attempting to checksum a non-TCP/UDP packet, dropping a protocol %d packet\n",
1167
if ((th + skb->csum_offset + 2) > skb_tail_pointer(skb))
1176
static bool tx_credit_exceeded(struct xenvif *vif, unsigned size)
1178
unsigned long now = jiffies;
1179
unsigned long next_credit =
1180
vif->credit_timeout.expires +
1181
msecs_to_jiffies(vif->credit_usec / 1000);
1183
/* Timer could already be pending in rare cases. */
1184
if (timer_pending(&vif->credit_timeout))
1187
/* Passed the point where we can replenish credit? */
1188
if (time_after_eq(now, next_credit)) {
1189
vif->credit_timeout.expires = now;
1193
/* Still too big to send right now? Set a callback. */
1194
if (size > vif->remaining_credit) {
1195
vif->credit_timeout.data =
1197
vif->credit_timeout.function =
1199
mod_timer(&vif->credit_timeout,
1208
static unsigned xen_netbk_tx_build_gops(struct xen_netbk *netbk)
1210
struct gnttab_copy *gop = netbk->tx_copy_ops, *request_gop;
1211
struct sk_buff *skb;
1214
while (((nr_pending_reqs(netbk) + MAX_SKB_FRAGS) < MAX_PENDING_REQS) &&
1215
!list_empty(&netbk->net_schedule_list)) {
1217
struct xen_netif_tx_request txreq;
1218
struct xen_netif_tx_request txfrags[MAX_SKB_FRAGS];
1220
struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX-1];
1224
unsigned int data_len;
1225
pending_ring_idx_t index;
1227
/* Get a netif from the list with work to do. */
1228
vif = poll_net_schedule_list(netbk);
1232
RING_FINAL_CHECK_FOR_REQUESTS(&vif->tx, work_to_do);
1238
idx = vif->tx.req_cons;
1239
rmb(); /* Ensure that we see the request before we copy it. */
1240
memcpy(&txreq, RING_GET_REQUEST(&vif->tx, idx), sizeof(txreq));
1242
/* Credit-based scheduling. */
1243
if (txreq.size > vif->remaining_credit &&
1244
tx_credit_exceeded(vif, txreq.size)) {
1249
vif->remaining_credit -= txreq.size;
1252
vif->tx.req_cons = ++idx;
1254
memset(extras, 0, sizeof(extras));
1255
if (txreq.flags & XEN_NETTXF_extra_info) {
1256
work_to_do = xen_netbk_get_extras(vif, extras,
1258
idx = vif->tx.req_cons;
1259
if (unlikely(work_to_do < 0)) {
1260
netbk_tx_err(vif, &txreq, idx);
1265
ret = netbk_count_requests(vif, &txreq, txfrags, work_to_do);
1266
if (unlikely(ret < 0)) {
1267
netbk_tx_err(vif, &txreq, idx - ret);
1272
if (unlikely(txreq.size < ETH_HLEN)) {
1273
netdev_dbg(vif->dev,
1274
"Bad packet size: %d\n", txreq.size);
1275
netbk_tx_err(vif, &txreq, idx);
1279
/* No crossing a page as the payload mustn't fragment. */
1280
if (unlikely((txreq.offset + txreq.size) > PAGE_SIZE)) {
1281
netdev_dbg(vif->dev,
1282
"txreq.offset: %x, size: %u, end: %lu\n",
1283
txreq.offset, txreq.size,
1284
(txreq.offset&~PAGE_MASK) + txreq.size);
1285
netbk_tx_err(vif, &txreq, idx);
1289
index = pending_index(netbk->pending_cons);
1290
pending_idx = netbk->pending_ring[index];
1292
data_len = (txreq.size > PKT_PROT_LEN &&
1293
ret < MAX_SKB_FRAGS) ?
1294
PKT_PROT_LEN : txreq.size;
1296
skb = alloc_skb(data_len + NET_SKB_PAD + NET_IP_ALIGN,
1297
GFP_ATOMIC | __GFP_NOWARN);
1298
if (unlikely(skb == NULL)) {
1299
netdev_dbg(vif->dev,
1300
"Can't allocate a skb in start_xmit.\n");
1301
netbk_tx_err(vif, &txreq, idx);
1305
/* Packets passed to netif_rx() must have some headroom. */
1306
skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
1308
if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1309
struct xen_netif_extra_info *gso;
1310
gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1312
if (netbk_set_skb_gso(vif, skb, gso)) {
1314
netbk_tx_err(vif, &txreq, idx);
1319
/* XXX could copy straight to head */
1320
page = xen_netbk_alloc_page(netbk, skb, pending_idx);
1323
netbk_tx_err(vif, &txreq, idx);
1327
netbk->mmap_pages[pending_idx] = page;
1329
gop->source.u.ref = txreq.gref;
1330
gop->source.domid = vif->domid;
1331
gop->source.offset = txreq.offset;
1333
gop->dest.u.gmfn = virt_to_mfn(page_address(page));
1334
gop->dest.domid = DOMID_SELF;
1335
gop->dest.offset = txreq.offset;
1337
gop->len = txreq.size;
1338
gop->flags = GNTCOPY_source_gref;
1342
memcpy(&netbk->pending_tx_info[pending_idx].req,
1343
&txreq, sizeof(txreq));
1344
netbk->pending_tx_info[pending_idx].vif = vif;
1345
*((u16 *)skb->data) = pending_idx;
1347
__skb_put(skb, data_len);
1349
skb_shinfo(skb)->nr_frags = ret;
1350
if (data_len < txreq.size) {
1351
skb_shinfo(skb)->nr_frags++;
1352
skb_shinfo(skb)->frags[0].page =
1353
(void *)(unsigned long)pending_idx;
1355
/* Discriminate from any valid pending_idx value. */
1356
skb_shinfo(skb)->frags[0].page = (void *)~0UL;
1359
__skb_queue_tail(&netbk->tx_queue, skb);
1361
netbk->pending_cons++;
1363
request_gop = xen_netbk_get_requests(netbk, vif,
1365
if (request_gop == NULL) {
1367
netbk_tx_err(vif, &txreq, idx);
1372
vif->tx.req_cons = idx;
1373
xen_netbk_check_rx_xenvif(vif);
1375
if ((gop-netbk->tx_copy_ops) >= ARRAY_SIZE(netbk->tx_copy_ops))
1379
return gop - netbk->tx_copy_ops;
1382
static void xen_netbk_tx_submit(struct xen_netbk *netbk)
1384
struct gnttab_copy *gop = netbk->tx_copy_ops;
1385
struct sk_buff *skb;
1387
while ((skb = __skb_dequeue(&netbk->tx_queue)) != NULL) {
1388
struct xen_netif_tx_request *txp;
1393
pending_idx = *((u16 *)skb->data);
1394
vif = netbk->pending_tx_info[pending_idx].vif;
1395
txp = &netbk->pending_tx_info[pending_idx].req;
1397
/* Check the remap error code. */
1398
if (unlikely(xen_netbk_tx_check_gop(netbk, skb, &gop))) {
1399
netdev_dbg(vif->dev, "netback grant failed.\n");
1400
skb_shinfo(skb)->nr_frags = 0;
1405
data_len = skb->len;
1407
(void *)(idx_to_kaddr(netbk, pending_idx)|txp->offset),
1409
if (data_len < txp->size) {
1410
/* Append the packet payload as a fragment. */
1411
txp->offset += data_len;
1412
txp->size -= data_len;
1414
/* Schedule a response immediately. */
1415
xen_netbk_idx_release(netbk, pending_idx);
1418
if (txp->flags & XEN_NETTXF_csum_blank)
1419
skb->ip_summed = CHECKSUM_PARTIAL;
1420
else if (txp->flags & XEN_NETTXF_data_validated)
1421
skb->ip_summed = CHECKSUM_UNNECESSARY;
1423
xen_netbk_fill_frags(netbk, skb);
1426
* If the initial fragment was < PKT_PROT_LEN then
1427
* pull through some bytes from the other fragments to
1428
* increase the linear region to PKT_PROT_LEN bytes.
1430
if (skb_headlen(skb) < PKT_PROT_LEN && skb_is_nonlinear(skb)) {
1431
int target = min_t(int, skb->len, PKT_PROT_LEN);
1432
__pskb_pull_tail(skb, target - skb_headlen(skb));
1435
skb->dev = vif->dev;
1436
skb->protocol = eth_type_trans(skb, skb->dev);
1438
if (checksum_setup(vif, skb)) {
1439
netdev_dbg(vif->dev,
1440
"Can't setup checksum in net_tx_action\n");
1445
vif->dev->stats.rx_bytes += skb->len;
1446
vif->dev->stats.rx_packets++;
1448
xenvif_receive_skb(vif, skb);
1452
/* Called after netfront has transmitted */
1453
static void xen_netbk_tx_action(struct xen_netbk *netbk)
1458
nr_gops = xen_netbk_tx_build_gops(netbk);
1462
ret = HYPERVISOR_grant_table_op(GNTTABOP_copy,
1463
netbk->tx_copy_ops, nr_gops);
1466
xen_netbk_tx_submit(netbk);
1470
static void xen_netbk_idx_release(struct xen_netbk *netbk, u16 pending_idx)
1473
struct pending_tx_info *pending_tx_info;
1474
pending_ring_idx_t index;
1476
/* Already complete? */
1477
if (netbk->mmap_pages[pending_idx] == NULL)
1480
pending_tx_info = &netbk->pending_tx_info[pending_idx];
1482
vif = pending_tx_info->vif;
1484
make_tx_response(vif, &pending_tx_info->req, XEN_NETIF_RSP_OKAY);
1486
index = pending_index(netbk->pending_prod++);
1487
netbk->pending_ring[index] = pending_idx;
1491
netbk->mmap_pages[pending_idx]->mapping = 0;
1492
put_page(netbk->mmap_pages[pending_idx]);
1493
netbk->mmap_pages[pending_idx] = NULL;
1496
static void make_tx_response(struct xenvif *vif,
1497
struct xen_netif_tx_request *txp,
1500
RING_IDX i = vif->tx.rsp_prod_pvt;
1501
struct xen_netif_tx_response *resp;
1504
resp = RING_GET_RESPONSE(&vif->tx, i);
1508
if (txp->flags & XEN_NETTXF_extra_info)
1509
RING_GET_RESPONSE(&vif->tx, ++i)->status = XEN_NETIF_RSP_NULL;
1511
vif->tx.rsp_prod_pvt = ++i;
1512
RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->tx, notify);
1514
notify_remote_via_irq(vif->irq);
1517
static struct xen_netif_rx_response *make_rx_response(struct xenvif *vif,
1524
RING_IDX i = vif->rx.rsp_prod_pvt;
1525
struct xen_netif_rx_response *resp;
1527
resp = RING_GET_RESPONSE(&vif->rx, i);
1528
resp->offset = offset;
1529
resp->flags = flags;
1531
resp->status = (s16)size;
1533
resp->status = (s16)st;
1535
vif->rx.rsp_prod_pvt = ++i;
1540
static inline int rx_work_todo(struct xen_netbk *netbk)
1542
return !skb_queue_empty(&netbk->rx_queue);
1545
static inline int tx_work_todo(struct xen_netbk *netbk)
1548
if (((nr_pending_reqs(netbk) + MAX_SKB_FRAGS) < MAX_PENDING_REQS) &&
1549
!list_empty(&netbk->net_schedule_list))
1555
static int xen_netbk_kthread(void *data)
1557
struct xen_netbk *netbk = data;
1558
while (!kthread_should_stop()) {
1559
wait_event_interruptible(netbk->wq,
1560
rx_work_todo(netbk) ||
1561
tx_work_todo(netbk) ||
1562
kthread_should_stop());
1565
if (kthread_should_stop())
1568
if (rx_work_todo(netbk))
1569
xen_netbk_rx_action(netbk);
1571
if (tx_work_todo(netbk))
1572
xen_netbk_tx_action(netbk);
1578
void xen_netbk_unmap_frontend_rings(struct xenvif *vif)
1580
struct gnttab_unmap_grant_ref op;
1582
if (vif->tx.sring) {
1583
gnttab_set_unmap_op(&op, (unsigned long)vif->tx_comms_area->addr,
1584
GNTMAP_host_map, vif->tx_shmem_handle);
1586
if (HYPERVISOR_grant_table_op(GNTTABOP_unmap_grant_ref, &op, 1))
1590
if (vif->rx.sring) {
1591
gnttab_set_unmap_op(&op, (unsigned long)vif->rx_comms_area->addr,
1592
GNTMAP_host_map, vif->rx_shmem_handle);
1594
if (HYPERVISOR_grant_table_op(GNTTABOP_unmap_grant_ref, &op, 1))
1597
if (vif->rx_comms_area)
1598
free_vm_area(vif->rx_comms_area);
1599
if (vif->tx_comms_area)
1600
free_vm_area(vif->tx_comms_area);
1603
int xen_netbk_map_frontend_rings(struct xenvif *vif,
1604
grant_ref_t tx_ring_ref,
1605
grant_ref_t rx_ring_ref)
1607
struct gnttab_map_grant_ref op;
1608
struct xen_netif_tx_sring *txs;
1609
struct xen_netif_rx_sring *rxs;
1613
vif->tx_comms_area = alloc_vm_area(PAGE_SIZE);
1614
if (vif->tx_comms_area == NULL)
1617
vif->rx_comms_area = alloc_vm_area(PAGE_SIZE);
1618
if (vif->rx_comms_area == NULL)
1621
gnttab_set_map_op(&op, (unsigned long)vif->tx_comms_area->addr,
1622
GNTMAP_host_map, tx_ring_ref, vif->domid);
1624
if (HYPERVISOR_grant_table_op(GNTTABOP_map_grant_ref, &op, 1))
1628
netdev_warn(vif->dev,
1629
"failed to map tx ring. err=%d status=%d\n",
1635
vif->tx_shmem_ref = tx_ring_ref;
1636
vif->tx_shmem_handle = op.handle;
1638
txs = (struct xen_netif_tx_sring *)vif->tx_comms_area->addr;
1639
BACK_RING_INIT(&vif->tx, txs, PAGE_SIZE);
1641
gnttab_set_map_op(&op, (unsigned long)vif->rx_comms_area->addr,
1642
GNTMAP_host_map, rx_ring_ref, vif->domid);
1644
if (HYPERVISOR_grant_table_op(GNTTABOP_map_grant_ref, &op, 1))
1648
netdev_warn(vif->dev,
1649
"failed to map rx ring. err=%d status=%d\n",
1655
vif->rx_shmem_ref = rx_ring_ref;
1656
vif->rx_shmem_handle = op.handle;
1657
vif->rx_req_cons_peek = 0;
1659
rxs = (struct xen_netif_rx_sring *)vif->rx_comms_area->addr;
1660
BACK_RING_INIT(&vif->rx, rxs, PAGE_SIZE);
1665
xen_netbk_unmap_frontend_rings(vif);
1669
static int __init netback_init(void)
1675
if (!xen_pv_domain())
1678
xen_netbk_group_nr = num_online_cpus();
1679
xen_netbk = vzalloc(sizeof(struct xen_netbk) * xen_netbk_group_nr);
1681
printk(KERN_ALERT "%s: out of memory\n", __func__);
1685
for (group = 0; group < xen_netbk_group_nr; group++) {
1686
struct xen_netbk *netbk = &xen_netbk[group];
1687
skb_queue_head_init(&netbk->rx_queue);
1688
skb_queue_head_init(&netbk->tx_queue);
1690
init_timer(&netbk->net_timer);
1691
netbk->net_timer.data = (unsigned long)netbk;
1692
netbk->net_timer.function = xen_netbk_alarm;
1694
netbk->pending_cons = 0;
1695
netbk->pending_prod = MAX_PENDING_REQS;
1696
for (i = 0; i < MAX_PENDING_REQS; i++)
1697
netbk->pending_ring[i] = i;
1699
init_waitqueue_head(&netbk->wq);
1700
netbk->task = kthread_create(xen_netbk_kthread,
1702
"netback/%u", group);
1704
if (IS_ERR(netbk->task)) {
1705
printk(KERN_ALERT "kthread_run() fails at netback\n");
1706
del_timer(&netbk->net_timer);
1707
rc = PTR_ERR(netbk->task);
1711
kthread_bind(netbk->task, group);
1713
INIT_LIST_HEAD(&netbk->net_schedule_list);
1715
spin_lock_init(&netbk->net_schedule_list_lock);
1717
atomic_set(&netbk->netfront_count, 0);
1719
wake_up_process(netbk->task);
1722
rc = xenvif_xenbus_init();
1729
while (--group >= 0) {
1730
struct xen_netbk *netbk = &xen_netbk[group];
1731
for (i = 0; i < MAX_PENDING_REQS; i++) {
1732
if (netbk->mmap_pages[i])
1733
__free_page(netbk->mmap_pages[i]);
1735
del_timer(&netbk->net_timer);
1736
kthread_stop(netbk->task);
1743
module_init(netback_init);
1745
MODULE_LICENSE("Dual BSD/GPL");