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>
39 #include <linux/tcp.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/platform_pci.h>
51 #include <xen/grant_table.h>
53 #include <xen/interface/io/netif.h>
54 #include <xen/interface/memory.h>
55 #include <xen/interface/grant_table.h>
57 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_TX_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 bool 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
);
494 frags
+= DIV_ROUND_UP(offset
+ len
, PAGE_SIZE
);
495 if (unlikely(frags
> MAX_SKB_FRAGS
+ 1)) {
496 printk(KERN_ALERT
"xennet: skb rides the rocket: %d frags\n",
502 spin_lock_irqsave(&np
->tx_lock
, flags
);
504 if (unlikely(!netif_carrier_ok(dev
) ||
505 (frags
> 1 && !xennet_can_sg(dev
)) ||
506 netif_needs_gso(skb
, netif_skb_features(skb
)))) {
507 spin_unlock_irqrestore(&np
->tx_lock
, flags
);
511 i
= np
->tx
.req_prod_pvt
;
513 id
= get_id_from_freelist(&np
->tx_skb_freelist
, np
->tx_skbs
);
514 np
->tx_skbs
[id
].skb
= skb
;
516 tx
= RING_GET_REQUEST(&np
->tx
, i
);
519 ref
= gnttab_claim_grant_reference(&np
->gref_tx_head
);
520 BUG_ON((signed short)ref
< 0);
521 mfn
= virt_to_mfn(data
);
522 gnttab_grant_foreign_access_ref(
523 ref
, np
->xbdev
->otherend_id
, mfn
, GNTMAP_readonly
);
524 tx
->gref
= np
->grant_tx_ref
[id
] = ref
;
530 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
532 tx
->flags
|= XEN_NETTXF_csum_blank
| XEN_NETTXF_data_validated
;
533 else if (skb
->ip_summed
== CHECKSUM_UNNECESSARY
)
534 /* remote but checksummed. */
535 tx
->flags
|= XEN_NETTXF_data_validated
;
537 if (skb_shinfo(skb
)->gso_size
) {
538 struct xen_netif_extra_info
*gso
;
540 gso
= (struct xen_netif_extra_info
*)
541 RING_GET_REQUEST(&np
->tx
, ++i
);
544 extra
->flags
|= XEN_NETIF_EXTRA_FLAG_MORE
;
546 tx
->flags
|= XEN_NETTXF_extra_info
;
548 gso
->u
.gso
.size
= skb_shinfo(skb
)->gso_size
;
549 gso
->u
.gso
.type
= XEN_NETIF_GSO_TYPE_TCPV4
;
551 gso
->u
.gso
.features
= 0;
553 gso
->type
= XEN_NETIF_EXTRA_TYPE_GSO
;
558 np
->tx
.req_prod_pvt
= i
+ 1;
560 xennet_make_frags(skb
, dev
, tx
);
563 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np
->tx
, notify
);
565 notify_remote_via_irq(np
->netdev
->irq
);
567 u64_stats_update_begin(&stats
->syncp
);
568 stats
->tx_bytes
+= skb
->len
;
570 u64_stats_update_end(&stats
->syncp
);
572 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
573 xennet_tx_buf_gc(dev
);
575 if (!netfront_tx_slot_available(np
))
576 netif_stop_queue(dev
);
578 spin_unlock_irqrestore(&np
->tx_lock
, flags
);
583 dev
->stats
.tx_dropped
++;
588 static int xennet_close(struct net_device
*dev
)
590 struct netfront_info
*np
= netdev_priv(dev
);
591 netif_stop_queue(np
->netdev
);
592 napi_disable(&np
->napi
);
596 static void xennet_move_rx_slot(struct netfront_info
*np
, struct sk_buff
*skb
,
599 int new = xennet_rxidx(np
->rx
.req_prod_pvt
);
601 BUG_ON(np
->rx_skbs
[new]);
602 np
->rx_skbs
[new] = skb
;
603 np
->grant_rx_ref
[new] = ref
;
604 RING_GET_REQUEST(&np
->rx
, np
->rx
.req_prod_pvt
)->id
= new;
605 RING_GET_REQUEST(&np
->rx
, np
->rx
.req_prod_pvt
)->gref
= ref
;
606 np
->rx
.req_prod_pvt
++;
609 static int xennet_get_extras(struct netfront_info
*np
,
610 struct xen_netif_extra_info
*extras
,
614 struct xen_netif_extra_info
*extra
;
615 struct device
*dev
= &np
->netdev
->dev
;
616 RING_IDX cons
= np
->rx
.rsp_cons
;
623 if (unlikely(cons
+ 1 == rp
)) {
625 dev_warn(dev
, "Missing extra info\n");
630 extra
= (struct xen_netif_extra_info
*)
631 RING_GET_RESPONSE(&np
->rx
, ++cons
);
633 if (unlikely(!extra
->type
||
634 extra
->type
>= XEN_NETIF_EXTRA_TYPE_MAX
)) {
636 dev_warn(dev
, "Invalid extra type: %d\n",
640 memcpy(&extras
[extra
->type
- 1], extra
,
644 skb
= xennet_get_rx_skb(np
, cons
);
645 ref
= xennet_get_rx_ref(np
, cons
);
646 xennet_move_rx_slot(np
, skb
, ref
);
647 } while (extra
->flags
& XEN_NETIF_EXTRA_FLAG_MORE
);
649 np
->rx
.rsp_cons
= cons
;
653 static int xennet_get_responses(struct netfront_info
*np
,
654 struct netfront_rx_info
*rinfo
, RING_IDX rp
,
655 struct sk_buff_head
*list
)
657 struct xen_netif_rx_response
*rx
= &rinfo
->rx
;
658 struct xen_netif_extra_info
*extras
= rinfo
->extras
;
659 struct device
*dev
= &np
->netdev
->dev
;
660 RING_IDX cons
= np
->rx
.rsp_cons
;
661 struct sk_buff
*skb
= xennet_get_rx_skb(np
, cons
);
662 grant_ref_t ref
= xennet_get_rx_ref(np
, cons
);
663 int max
= MAX_SKB_FRAGS
+ (rx
->status
<= RX_COPY_THRESHOLD
);
668 if (rx
->flags
& XEN_NETRXF_extra_info
) {
669 err
= xennet_get_extras(np
, extras
, rp
);
670 cons
= np
->rx
.rsp_cons
;
674 if (unlikely(rx
->status
< 0 ||
675 rx
->offset
+ rx
->status
> PAGE_SIZE
)) {
677 dev_warn(dev
, "rx->offset: %x, size: %u\n",
678 rx
->offset
, rx
->status
);
679 xennet_move_rx_slot(np
, skb
, ref
);
685 * This definitely indicates a bug, either in this driver or in
686 * the backend driver. In future this should flag the bad
687 * situation to the system controller to reboot the backed.
689 if (ref
== GRANT_INVALID_REF
) {
691 dev_warn(dev
, "Bad rx response id %d.\n",
697 ret
= gnttab_end_foreign_access_ref(ref
, 0);
700 gnttab_release_grant_reference(&np
->gref_rx_head
, ref
);
702 __skb_queue_tail(list
, skb
);
705 if (!(rx
->flags
& XEN_NETRXF_more_data
))
708 if (cons
+ frags
== rp
) {
710 dev_warn(dev
, "Need more frags\n");
715 rx
= RING_GET_RESPONSE(&np
->rx
, cons
+ frags
);
716 skb
= xennet_get_rx_skb(np
, cons
+ frags
);
717 ref
= xennet_get_rx_ref(np
, cons
+ frags
);
721 if (unlikely(frags
> max
)) {
723 dev_warn(dev
, "Too many frags\n");
728 np
->rx
.rsp_cons
= cons
+ frags
;
733 static int xennet_set_skb_gso(struct sk_buff
*skb
,
734 struct xen_netif_extra_info
*gso
)
736 if (!gso
->u
.gso
.size
) {
738 printk(KERN_WARNING
"GSO size must not be zero.\n");
742 /* Currently only TCPv4 S.O. is supported. */
743 if (gso
->u
.gso
.type
!= XEN_NETIF_GSO_TYPE_TCPV4
) {
745 printk(KERN_WARNING
"Bad GSO type %d.\n", gso
->u
.gso
.type
);
749 skb_shinfo(skb
)->gso_size
= gso
->u
.gso
.size
;
750 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV4
;
752 /* Header must be checked, and gso_segs computed. */
753 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
754 skb_shinfo(skb
)->gso_segs
= 0;
759 static RING_IDX
xennet_fill_frags(struct netfront_info
*np
,
761 struct sk_buff_head
*list
)
763 struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
764 int nr_frags
= shinfo
->nr_frags
;
765 RING_IDX cons
= np
->rx
.rsp_cons
;
766 struct sk_buff
*nskb
;
768 while ((nskb
= __skb_dequeue(list
))) {
769 struct xen_netif_rx_response
*rx
=
770 RING_GET_RESPONSE(&np
->rx
, ++cons
);
771 skb_frag_t
*nfrag
= &skb_shinfo(nskb
)->frags
[0];
773 __skb_fill_page_desc(skb
, nr_frags
,
774 skb_frag_page(nfrag
),
775 rx
->offset
, rx
->status
);
777 skb
->data_len
+= rx
->status
;
779 skb_shinfo(nskb
)->nr_frags
= 0;
785 shinfo
->nr_frags
= nr_frags
;
789 static int checksum_setup(struct net_device
*dev
, struct sk_buff
*skb
)
794 int recalculate_partial_csum
= 0;
797 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
798 * peers can fail to set NETRXF_csum_blank when sending a GSO
799 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
800 * recalculate the partial checksum.
802 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
&& skb_is_gso(skb
)) {
803 struct netfront_info
*np
= netdev_priv(dev
);
804 np
->rx_gso_checksum_fixup
++;
805 skb
->ip_summed
= CHECKSUM_PARTIAL
;
806 recalculate_partial_csum
= 1;
809 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
810 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
)
813 if (skb
->protocol
!= htons(ETH_P_IP
))
816 iph
= (void *)skb
->data
;
817 th
= skb
->data
+ 4 * iph
->ihl
;
818 if (th
>= skb_tail_pointer(skb
))
821 skb
->csum_start
= th
- skb
->head
;
822 switch (iph
->protocol
) {
824 skb
->csum_offset
= offsetof(struct tcphdr
, check
);
826 if (recalculate_partial_csum
) {
827 struct tcphdr
*tcph
= (struct tcphdr
*)th
;
828 tcph
->check
= ~csum_tcpudp_magic(iph
->saddr
, iph
->daddr
,
829 skb
->len
- iph
->ihl
*4,
834 skb
->csum_offset
= offsetof(struct udphdr
, check
);
836 if (recalculate_partial_csum
) {
837 struct udphdr
*udph
= (struct udphdr
*)th
;
838 udph
->check
= ~csum_tcpudp_magic(iph
->saddr
, iph
->daddr
,
839 skb
->len
- iph
->ihl
*4,
845 printk(KERN_ERR
"Attempting to checksum a non-"
846 "TCP/UDP packet, dropping a protocol"
847 " %d packet", iph
->protocol
);
851 if ((th
+ skb
->csum_offset
+ 2) > skb_tail_pointer(skb
))
860 static int handle_incoming_queue(struct net_device
*dev
,
861 struct sk_buff_head
*rxq
)
863 struct netfront_info
*np
= netdev_priv(dev
);
864 struct netfront_stats
*stats
= this_cpu_ptr(np
->stats
);
865 int packets_dropped
= 0;
868 while ((skb
= __skb_dequeue(rxq
)) != NULL
) {
869 int pull_to
= NETFRONT_SKB_CB(skb
)->pull_to
;
871 __pskb_pull_tail(skb
, pull_to
- skb_headlen(skb
));
873 /* Ethernet work: Delayed to here as it peeks the header. */
874 skb
->protocol
= eth_type_trans(skb
, dev
);
876 if (checksum_setup(dev
, skb
)) {
879 dev
->stats
.rx_errors
++;
883 u64_stats_update_begin(&stats
->syncp
);
885 stats
->rx_bytes
+= skb
->len
;
886 u64_stats_update_end(&stats
->syncp
);
889 netif_receive_skb(skb
);
892 return packets_dropped
;
895 static int xennet_poll(struct napi_struct
*napi
, int budget
)
897 struct netfront_info
*np
= container_of(napi
, struct netfront_info
, napi
);
898 struct net_device
*dev
= np
->netdev
;
900 struct netfront_rx_info rinfo
;
901 struct xen_netif_rx_response
*rx
= &rinfo
.rx
;
902 struct xen_netif_extra_info
*extras
= rinfo
.extras
;
905 struct sk_buff_head rxq
;
906 struct sk_buff_head errq
;
907 struct sk_buff_head tmpq
;
911 spin_lock(&np
->rx_lock
);
913 skb_queue_head_init(&rxq
);
914 skb_queue_head_init(&errq
);
915 skb_queue_head_init(&tmpq
);
917 rp
= np
->rx
.sring
->rsp_prod
;
918 rmb(); /* Ensure we see queued responses up to 'rp'. */
922 while ((i
!= rp
) && (work_done
< budget
)) {
923 memcpy(rx
, RING_GET_RESPONSE(&np
->rx
, i
), sizeof(*rx
));
924 memset(extras
, 0, sizeof(rinfo
.extras
));
926 err
= xennet_get_responses(np
, &rinfo
, rp
, &tmpq
);
930 while ((skb
= __skb_dequeue(&tmpq
)))
931 __skb_queue_tail(&errq
, skb
);
932 dev
->stats
.rx_errors
++;
937 skb
= __skb_dequeue(&tmpq
);
939 if (extras
[XEN_NETIF_EXTRA_TYPE_GSO
- 1].type
) {
940 struct xen_netif_extra_info
*gso
;
941 gso
= &extras
[XEN_NETIF_EXTRA_TYPE_GSO
- 1];
943 if (unlikely(xennet_set_skb_gso(skb
, gso
))) {
944 __skb_queue_head(&tmpq
, skb
);
945 np
->rx
.rsp_cons
+= skb_queue_len(&tmpq
);
950 NETFRONT_SKB_CB(skb
)->pull_to
= rx
->status
;
951 if (NETFRONT_SKB_CB(skb
)->pull_to
> RX_COPY_THRESHOLD
)
952 NETFRONT_SKB_CB(skb
)->pull_to
= RX_COPY_THRESHOLD
;
954 skb_shinfo(skb
)->frags
[0].page_offset
= rx
->offset
;
955 skb_frag_size_set(&skb_shinfo(skb
)->frags
[0], rx
->status
);
956 skb
->data_len
= rx
->status
;
958 i
= xennet_fill_frags(np
, skb
, &tmpq
);
961 * Truesize approximates the size of true data plus
962 * any supervisor overheads. Adding hypervisor
963 * overheads has been shown to significantly reduce
964 * achievable bandwidth with the default receive
965 * buffer size. It is therefore not wise to account
968 * After alloc_skb(RX_COPY_THRESHOLD), truesize is set
969 * to RX_COPY_THRESHOLD + the supervisor
970 * overheads. Here, we add the size of the data pulled
971 * in xennet_fill_frags().
973 * We also adjust for any unused space in the main
974 * data area by subtracting (RX_COPY_THRESHOLD -
975 * len). This is especially important with drivers
976 * which split incoming packets into header and data,
977 * using only 66 bytes of the main data area (see the
978 * e1000 driver for example.) On such systems,
979 * without this last adjustement, our achievable
980 * receive throughout using the standard receive
981 * buffer size was cut by 25%(!!!).
983 skb
->truesize
+= skb
->data_len
- RX_COPY_THRESHOLD
;
984 skb
->len
+= skb
->data_len
;
986 if (rx
->flags
& XEN_NETRXF_csum_blank
)
987 skb
->ip_summed
= CHECKSUM_PARTIAL
;
988 else if (rx
->flags
& XEN_NETRXF_data_validated
)
989 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
991 __skb_queue_tail(&rxq
, skb
);
993 np
->rx
.rsp_cons
= ++i
;
997 __skb_queue_purge(&errq
);
999 work_done
-= handle_incoming_queue(dev
, &rxq
);
1001 /* If we get a callback with very few responses, reduce fill target. */
1002 /* NB. Note exponential increase, linear decrease. */
1003 if (((np
->rx
.req_prod_pvt
- np
->rx
.sring
->rsp_prod
) >
1004 ((3*np
->rx_target
) / 4)) &&
1005 (--np
->rx_target
< np
->rx_min_target
))
1006 np
->rx_target
= np
->rx_min_target
;
1008 xennet_alloc_rx_buffers(dev
);
1010 if (work_done
< budget
) {
1013 local_irq_save(flags
);
1015 RING_FINAL_CHECK_FOR_RESPONSES(&np
->rx
, more_to_do
);
1017 __napi_complete(napi
);
1019 local_irq_restore(flags
);
1022 spin_unlock(&np
->rx_lock
);
1027 static int xennet_change_mtu(struct net_device
*dev
, int mtu
)
1029 int max
= xennet_can_sg(dev
) ? 65535 - ETH_HLEN
: ETH_DATA_LEN
;
1037 static struct rtnl_link_stats64
*xennet_get_stats64(struct net_device
*dev
,
1038 struct rtnl_link_stats64
*tot
)
1040 struct netfront_info
*np
= netdev_priv(dev
);
1043 for_each_possible_cpu(cpu
) {
1044 struct netfront_stats
*stats
= per_cpu_ptr(np
->stats
, cpu
);
1045 u64 rx_packets
, rx_bytes
, tx_packets
, tx_bytes
;
1049 start
= u64_stats_fetch_begin_bh(&stats
->syncp
);
1051 rx_packets
= stats
->rx_packets
;
1052 tx_packets
= stats
->tx_packets
;
1053 rx_bytes
= stats
->rx_bytes
;
1054 tx_bytes
= stats
->tx_bytes
;
1055 } while (u64_stats_fetch_retry_bh(&stats
->syncp
, start
));
1057 tot
->rx_packets
+= rx_packets
;
1058 tot
->tx_packets
+= tx_packets
;
1059 tot
->rx_bytes
+= rx_bytes
;
1060 tot
->tx_bytes
+= tx_bytes
;
1063 tot
->rx_errors
= dev
->stats
.rx_errors
;
1064 tot
->tx_dropped
= dev
->stats
.tx_dropped
;
1069 static void xennet_release_tx_bufs(struct netfront_info
*np
)
1071 struct sk_buff
*skb
;
1074 for (i
= 0; i
< NET_TX_RING_SIZE
; i
++) {
1075 /* Skip over entries which are actually freelist references */
1076 if (skb_entry_is_link(&np
->tx_skbs
[i
]))
1079 skb
= np
->tx_skbs
[i
].skb
;
1080 gnttab_end_foreign_access_ref(np
->grant_tx_ref
[i
],
1082 gnttab_release_grant_reference(&np
->gref_tx_head
,
1083 np
->grant_tx_ref
[i
]);
1084 np
->grant_tx_ref
[i
] = GRANT_INVALID_REF
;
1085 add_id_to_freelist(&np
->tx_skb_freelist
, np
->tx_skbs
, i
);
1086 dev_kfree_skb_irq(skb
);
1090 static void xennet_release_rx_bufs(struct netfront_info
*np
)
1092 struct mmu_update
*mmu
= np
->rx_mmu
;
1093 struct multicall_entry
*mcl
= np
->rx_mcl
;
1094 struct sk_buff_head free_list
;
1095 struct sk_buff
*skb
;
1097 int xfer
= 0, noxfer
= 0, unused
= 0;
1100 dev_warn(&np
->netdev
->dev
, "%s: fix me for copying receiver.\n",
1104 skb_queue_head_init(&free_list
);
1106 spin_lock_bh(&np
->rx_lock
);
1108 for (id
= 0; id
< NET_RX_RING_SIZE
; id
++) {
1109 ref
= np
->grant_rx_ref
[id
];
1110 if (ref
== GRANT_INVALID_REF
) {
1115 skb
= np
->rx_skbs
[id
];
1116 mfn
= gnttab_end_foreign_transfer_ref(ref
);
1117 gnttab_release_grant_reference(&np
->gref_rx_head
, ref
);
1118 np
->grant_rx_ref
[id
] = GRANT_INVALID_REF
;
1121 skb_shinfo(skb
)->nr_frags
= 0;
1127 if (!xen_feature(XENFEAT_auto_translated_physmap
)) {
1128 /* Remap the page. */
1129 const struct page
*page
=
1130 skb_frag_page(&skb_shinfo(skb
)->frags
[0]);
1131 unsigned long pfn
= page_to_pfn(page
);
1132 void *vaddr
= page_address(page
);
1134 MULTI_update_va_mapping(mcl
, (unsigned long)vaddr
,
1135 mfn_pte(mfn
, PAGE_KERNEL
),
1138 mmu
->ptr
= ((u64
)mfn
<< PAGE_SHIFT
)
1139 | MMU_MACHPHYS_UPDATE
;
1143 set_phys_to_machine(pfn
, mfn
);
1145 __skb_queue_tail(&free_list
, skb
);
1149 dev_info(&np
->netdev
->dev
, "%s: %d xfer, %d noxfer, %d unused\n",
1150 __func__
, xfer
, noxfer
, unused
);
1153 if (!xen_feature(XENFEAT_auto_translated_physmap
)) {
1154 /* Do all the remapping work and M2P updates. */
1155 MULTI_mmu_update(mcl
, np
->rx_mmu
, mmu
- np
->rx_mmu
,
1158 HYPERVISOR_multicall(np
->rx_mcl
, mcl
- np
->rx_mcl
);
1162 __skb_queue_purge(&free_list
);
1164 spin_unlock_bh(&np
->rx_lock
);
1167 static void xennet_uninit(struct net_device
*dev
)
1169 struct netfront_info
*np
= netdev_priv(dev
);
1170 xennet_release_tx_bufs(np
);
1171 xennet_release_rx_bufs(np
);
1172 gnttab_free_grant_references(np
->gref_tx_head
);
1173 gnttab_free_grant_references(np
->gref_rx_head
);
1176 static netdev_features_t
xennet_fix_features(struct net_device
*dev
,
1177 netdev_features_t features
)
1179 struct netfront_info
*np
= netdev_priv(dev
);
1182 if (features
& NETIF_F_SG
) {
1183 if (xenbus_scanf(XBT_NIL
, np
->xbdev
->otherend
, "feature-sg",
1188 features
&= ~NETIF_F_SG
;
1191 if (features
& NETIF_F_TSO
) {
1192 if (xenbus_scanf(XBT_NIL
, np
->xbdev
->otherend
,
1193 "feature-gso-tcpv4", "%d", &val
) < 0)
1197 features
&= ~NETIF_F_TSO
;
1203 static int xennet_set_features(struct net_device
*dev
,
1204 netdev_features_t features
)
1206 if (!(features
& NETIF_F_SG
) && dev
->mtu
> ETH_DATA_LEN
) {
1207 netdev_info(dev
, "Reducing MTU because no SG offload");
1208 dev
->mtu
= ETH_DATA_LEN
;
1214 static irqreturn_t
xennet_interrupt(int irq
, void *dev_id
)
1216 struct net_device
*dev
= dev_id
;
1217 struct netfront_info
*np
= netdev_priv(dev
);
1218 unsigned long flags
;
1220 spin_lock_irqsave(&np
->tx_lock
, flags
);
1222 if (likely(netif_carrier_ok(dev
))) {
1223 xennet_tx_buf_gc(dev
);
1224 /* Under tx_lock: protects access to rx shared-ring indexes. */
1225 if (RING_HAS_UNCONSUMED_RESPONSES(&np
->rx
))
1226 napi_schedule(&np
->napi
);
1229 spin_unlock_irqrestore(&np
->tx_lock
, flags
);
1234 #ifdef CONFIG_NET_POLL_CONTROLLER
1235 static void xennet_poll_controller(struct net_device
*dev
)
1237 xennet_interrupt(0, dev
);
1241 static const struct net_device_ops xennet_netdev_ops
= {
1242 .ndo_open
= xennet_open
,
1243 .ndo_uninit
= xennet_uninit
,
1244 .ndo_stop
= xennet_close
,
1245 .ndo_start_xmit
= xennet_start_xmit
,
1246 .ndo_change_mtu
= xennet_change_mtu
,
1247 .ndo_get_stats64
= xennet_get_stats64
,
1248 .ndo_set_mac_address
= eth_mac_addr
,
1249 .ndo_validate_addr
= eth_validate_addr
,
1250 .ndo_fix_features
= xennet_fix_features
,
1251 .ndo_set_features
= xennet_set_features
,
1252 #ifdef CONFIG_NET_POLL_CONTROLLER
1253 .ndo_poll_controller
= xennet_poll_controller
,
1257 static struct net_device
* __devinit
xennet_create_dev(struct xenbus_device
*dev
)
1260 struct net_device
*netdev
;
1261 struct netfront_info
*np
;
1263 netdev
= alloc_etherdev(sizeof(struct netfront_info
));
1265 return ERR_PTR(-ENOMEM
);
1267 np
= netdev_priv(netdev
);
1270 spin_lock_init(&np
->tx_lock
);
1271 spin_lock_init(&np
->rx_lock
);
1273 skb_queue_head_init(&np
->rx_batch
);
1274 np
->rx_target
= RX_DFL_MIN_TARGET
;
1275 np
->rx_min_target
= RX_DFL_MIN_TARGET
;
1276 np
->rx_max_target
= RX_MAX_TARGET
;
1278 init_timer(&np
->rx_refill_timer
);
1279 np
->rx_refill_timer
.data
= (unsigned long)netdev
;
1280 np
->rx_refill_timer
.function
= rx_refill_timeout
;
1283 np
->stats
= alloc_percpu(struct netfront_stats
);
1284 if (np
->stats
== NULL
)
1287 /* Initialise tx_skbs as a free chain containing every entry. */
1288 np
->tx_skb_freelist
= 0;
1289 for (i
= 0; i
< NET_TX_RING_SIZE
; i
++) {
1290 skb_entry_set_link(&np
->tx_skbs
[i
], i
+1);
1291 np
->grant_tx_ref
[i
] = GRANT_INVALID_REF
;
1294 /* Clear out rx_skbs */
1295 for (i
= 0; i
< NET_RX_RING_SIZE
; i
++) {
1296 np
->rx_skbs
[i
] = NULL
;
1297 np
->grant_rx_ref
[i
] = GRANT_INVALID_REF
;
1300 /* A grant for every tx ring slot */
1301 if (gnttab_alloc_grant_references(TX_MAX_TARGET
,
1302 &np
->gref_tx_head
) < 0) {
1303 printk(KERN_ALERT
"#### netfront can't alloc tx grant refs\n");
1305 goto exit_free_stats
;
1307 /* A grant for every rx ring slot */
1308 if (gnttab_alloc_grant_references(RX_MAX_TARGET
,
1309 &np
->gref_rx_head
) < 0) {
1310 printk(KERN_ALERT
"#### netfront can't alloc rx grant refs\n");
1315 netdev
->netdev_ops
= &xennet_netdev_ops
;
1317 netif_napi_add(netdev
, &np
->napi
, xennet_poll
, 64);
1318 netdev
->features
= NETIF_F_IP_CSUM
| NETIF_F_RXCSUM
|
1320 netdev
->hw_features
= NETIF_F_IP_CSUM
| NETIF_F_SG
| NETIF_F_TSO
;
1323 * Assume that all hw features are available for now. This set
1324 * will be adjusted by the call to netdev_update_features() in
1325 * xennet_connect() which is the earliest point where we can
1326 * negotiate with the backend regarding supported features.
1328 netdev
->features
|= netdev
->hw_features
;
1330 SET_ETHTOOL_OPS(netdev
, &xennet_ethtool_ops
);
1331 SET_NETDEV_DEV(netdev
, &dev
->dev
);
1333 np
->netdev
= netdev
;
1335 netif_carrier_off(netdev
);
1340 gnttab_free_grant_references(np
->gref_tx_head
);
1342 free_percpu(np
->stats
);
1344 free_netdev(netdev
);
1345 return ERR_PTR(err
);
1349 * Entry point to this code when a new device is created. Allocate the basic
1350 * structures and the ring buffers for communication with the backend, and
1351 * inform the backend of the appropriate details for those.
1353 static int __devinit
netfront_probe(struct xenbus_device
*dev
,
1354 const struct xenbus_device_id
*id
)
1357 struct net_device
*netdev
;
1358 struct netfront_info
*info
;
1360 netdev
= xennet_create_dev(dev
);
1361 if (IS_ERR(netdev
)) {
1362 err
= PTR_ERR(netdev
);
1363 xenbus_dev_fatal(dev
, err
, "creating netdev");
1367 info
= netdev_priv(netdev
);
1368 dev_set_drvdata(&dev
->dev
, info
);
1370 err
= register_netdev(info
->netdev
);
1372 printk(KERN_WARNING
"%s: register_netdev err=%d\n",
1377 err
= xennet_sysfs_addif(info
->netdev
);
1379 unregister_netdev(info
->netdev
);
1380 printk(KERN_WARNING
"%s: add sysfs failed err=%d\n",
1388 free_netdev(netdev
);
1389 dev_set_drvdata(&dev
->dev
, NULL
);
1393 static void xennet_end_access(int ref
, void *page
)
1395 /* This frees the page as a side-effect */
1396 if (ref
!= GRANT_INVALID_REF
)
1397 gnttab_end_foreign_access(ref
, 0, (unsigned long)page
);
1400 static void xennet_disconnect_backend(struct netfront_info
*info
)
1402 /* Stop old i/f to prevent errors whilst we rebuild the state. */
1403 spin_lock_bh(&info
->rx_lock
);
1404 spin_lock_irq(&info
->tx_lock
);
1405 netif_carrier_off(info
->netdev
);
1406 spin_unlock_irq(&info
->tx_lock
);
1407 spin_unlock_bh(&info
->rx_lock
);
1409 if (info
->netdev
->irq
)
1410 unbind_from_irqhandler(info
->netdev
->irq
, info
->netdev
);
1411 info
->evtchn
= info
->netdev
->irq
= 0;
1413 /* End access and free the pages */
1414 xennet_end_access(info
->tx_ring_ref
, info
->tx
.sring
);
1415 xennet_end_access(info
->rx_ring_ref
, info
->rx
.sring
);
1417 info
->tx_ring_ref
= GRANT_INVALID_REF
;
1418 info
->rx_ring_ref
= GRANT_INVALID_REF
;
1419 info
->tx
.sring
= NULL
;
1420 info
->rx
.sring
= NULL
;
1424 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1425 * driver restart. We tear down our netif structure and recreate it, but
1426 * leave the device-layer structures intact so that this is transparent to the
1427 * rest of the kernel.
1429 static int netfront_resume(struct xenbus_device
*dev
)
1431 struct netfront_info
*info
= dev_get_drvdata(&dev
->dev
);
1433 dev_dbg(&dev
->dev
, "%s\n", dev
->nodename
);
1435 xennet_disconnect_backend(info
);
1439 static int xen_net_read_mac(struct xenbus_device
*dev
, u8 mac
[])
1441 char *s
, *e
, *macstr
;
1444 macstr
= s
= xenbus_read(XBT_NIL
, dev
->nodename
, "mac", NULL
);
1446 return PTR_ERR(macstr
);
1448 for (i
= 0; i
< ETH_ALEN
; i
++) {
1449 mac
[i
] = simple_strtoul(s
, &e
, 16);
1450 if ((s
== e
) || (*e
!= ((i
== ETH_ALEN
-1) ? '\0' : ':'))) {
1461 static int setup_netfront(struct xenbus_device
*dev
, struct netfront_info
*info
)
1463 struct xen_netif_tx_sring
*txs
;
1464 struct xen_netif_rx_sring
*rxs
;
1466 struct net_device
*netdev
= info
->netdev
;
1468 info
->tx_ring_ref
= GRANT_INVALID_REF
;
1469 info
->rx_ring_ref
= GRANT_INVALID_REF
;
1470 info
->rx
.sring
= NULL
;
1471 info
->tx
.sring
= NULL
;
1474 err
= xen_net_read_mac(dev
, netdev
->dev_addr
);
1476 xenbus_dev_fatal(dev
, err
, "parsing %s/mac", dev
->nodename
);
1480 txs
= (struct xen_netif_tx_sring
*)get_zeroed_page(GFP_NOIO
| __GFP_HIGH
);
1483 xenbus_dev_fatal(dev
, err
, "allocating tx ring page");
1486 SHARED_RING_INIT(txs
);
1487 FRONT_RING_INIT(&info
->tx
, txs
, PAGE_SIZE
);
1489 err
= xenbus_grant_ring(dev
, virt_to_mfn(txs
));
1491 free_page((unsigned long)txs
);
1495 info
->tx_ring_ref
= err
;
1496 rxs
= (struct xen_netif_rx_sring
*)get_zeroed_page(GFP_NOIO
| __GFP_HIGH
);
1499 xenbus_dev_fatal(dev
, err
, "allocating rx ring page");
1502 SHARED_RING_INIT(rxs
);
1503 FRONT_RING_INIT(&info
->rx
, rxs
, PAGE_SIZE
);
1505 err
= xenbus_grant_ring(dev
, virt_to_mfn(rxs
));
1507 free_page((unsigned long)rxs
);
1510 info
->rx_ring_ref
= err
;
1512 err
= xenbus_alloc_evtchn(dev
, &info
->evtchn
);
1516 err
= bind_evtchn_to_irqhandler(info
->evtchn
, xennet_interrupt
,
1517 0, netdev
->name
, netdev
);
1527 /* Common code used when first setting up, and when resuming. */
1528 static int talk_to_netback(struct xenbus_device
*dev
,
1529 struct netfront_info
*info
)
1531 const char *message
;
1532 struct xenbus_transaction xbt
;
1535 /* Create shared ring, alloc event channel. */
1536 err
= setup_netfront(dev
, info
);
1541 err
= xenbus_transaction_start(&xbt
);
1543 xenbus_dev_fatal(dev
, err
, "starting transaction");
1547 err
= xenbus_printf(xbt
, dev
->nodename
, "tx-ring-ref", "%u",
1550 message
= "writing tx ring-ref";
1551 goto abort_transaction
;
1553 err
= xenbus_printf(xbt
, dev
->nodename
, "rx-ring-ref", "%u",
1556 message
= "writing rx ring-ref";
1557 goto abort_transaction
;
1559 err
= xenbus_printf(xbt
, dev
->nodename
,
1560 "event-channel", "%u", info
->evtchn
);
1562 message
= "writing event-channel";
1563 goto abort_transaction
;
1566 err
= xenbus_printf(xbt
, dev
->nodename
, "request-rx-copy", "%u",
1569 message
= "writing request-rx-copy";
1570 goto abort_transaction
;
1573 err
= xenbus_printf(xbt
, dev
->nodename
, "feature-rx-notify", "%d", 1);
1575 message
= "writing feature-rx-notify";
1576 goto abort_transaction
;
1579 err
= xenbus_printf(xbt
, dev
->nodename
, "feature-sg", "%d", 1);
1581 message
= "writing feature-sg";
1582 goto abort_transaction
;
1585 err
= xenbus_printf(xbt
, dev
->nodename
, "feature-gso-tcpv4", "%d", 1);
1587 message
= "writing feature-gso-tcpv4";
1588 goto abort_transaction
;
1591 err
= xenbus_transaction_end(xbt
, 0);
1595 xenbus_dev_fatal(dev
, err
, "completing transaction");
1602 xenbus_transaction_end(xbt
, 1);
1603 xenbus_dev_fatal(dev
, err
, "%s", message
);
1605 xennet_disconnect_backend(info
);
1610 static int xennet_connect(struct net_device
*dev
)
1612 struct netfront_info
*np
= netdev_priv(dev
);
1613 int i
, requeue_idx
, err
;
1614 struct sk_buff
*skb
;
1616 struct xen_netif_rx_request
*req
;
1617 unsigned int feature_rx_copy
;
1619 err
= xenbus_scanf(XBT_NIL
, np
->xbdev
->otherend
,
1620 "feature-rx-copy", "%u", &feature_rx_copy
);
1622 feature_rx_copy
= 0;
1624 if (!feature_rx_copy
) {
1626 "backend does not support copying receive path\n");
1630 err
= talk_to_netback(np
->xbdev
, np
);
1635 netdev_update_features(dev
);
1638 spin_lock_bh(&np
->rx_lock
);
1639 spin_lock_irq(&np
->tx_lock
);
1641 /* Step 1: Discard all pending TX packet fragments. */
1642 xennet_release_tx_bufs(np
);
1644 /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1645 for (requeue_idx
= 0, i
= 0; i
< NET_RX_RING_SIZE
; i
++) {
1647 const struct page
*page
;
1648 if (!np
->rx_skbs
[i
])
1651 skb
= np
->rx_skbs
[requeue_idx
] = xennet_get_rx_skb(np
, i
);
1652 ref
= np
->grant_rx_ref
[requeue_idx
] = xennet_get_rx_ref(np
, i
);
1653 req
= RING_GET_REQUEST(&np
->rx
, requeue_idx
);
1655 frag
= &skb_shinfo(skb
)->frags
[0];
1656 page
= skb_frag_page(frag
);
1657 gnttab_grant_foreign_access_ref(
1658 ref
, np
->xbdev
->otherend_id
,
1659 pfn_to_mfn(page_to_pfn(page
)),
1662 req
->id
= requeue_idx
;
1667 np
->rx
.req_prod_pvt
= requeue_idx
;
1670 * Step 3: All public and private state should now be sane. Get
1671 * ready to start sending and receiving packets and give the driver
1672 * domain a kick because we've probably just requeued some
1675 netif_carrier_on(np
->netdev
);
1676 notify_remote_via_irq(np
->netdev
->irq
);
1677 xennet_tx_buf_gc(dev
);
1678 xennet_alloc_rx_buffers(dev
);
1680 spin_unlock_irq(&np
->tx_lock
);
1681 spin_unlock_bh(&np
->rx_lock
);
1687 * Callback received when the backend's state changes.
1689 static void netback_changed(struct xenbus_device
*dev
,
1690 enum xenbus_state backend_state
)
1692 struct netfront_info
*np
= dev_get_drvdata(&dev
->dev
);
1693 struct net_device
*netdev
= np
->netdev
;
1695 dev_dbg(&dev
->dev
, "%s\n", xenbus_strstate(backend_state
));
1697 switch (backend_state
) {
1698 case XenbusStateInitialising
:
1699 case XenbusStateInitialised
:
1700 case XenbusStateReconfiguring
:
1701 case XenbusStateReconfigured
:
1702 case XenbusStateUnknown
:
1703 case XenbusStateClosed
:
1706 case XenbusStateInitWait
:
1707 if (dev
->state
!= XenbusStateInitialising
)
1709 if (xennet_connect(netdev
) != 0)
1711 xenbus_switch_state(dev
, XenbusStateConnected
);
1714 case XenbusStateConnected
:
1715 netif_notify_peers(netdev
);
1718 case XenbusStateClosing
:
1719 xenbus_frontend_closed(dev
);
1724 static const struct xennet_stat
{
1725 char name
[ETH_GSTRING_LEN
];
1727 } xennet_stats
[] = {
1729 "rx_gso_checksum_fixup",
1730 offsetof(struct netfront_info
, rx_gso_checksum_fixup
)
1734 static int xennet_get_sset_count(struct net_device
*dev
, int string_set
)
1736 switch (string_set
) {
1738 return ARRAY_SIZE(xennet_stats
);
1744 static void xennet_get_ethtool_stats(struct net_device
*dev
,
1745 struct ethtool_stats
*stats
, u64
* data
)
1747 void *np
= netdev_priv(dev
);
1750 for (i
= 0; i
< ARRAY_SIZE(xennet_stats
); i
++)
1751 data
[i
] = *(unsigned long *)(np
+ xennet_stats
[i
].offset
);
1754 static void xennet_get_strings(struct net_device
*dev
, u32 stringset
, u8
* data
)
1758 switch (stringset
) {
1760 for (i
= 0; i
< ARRAY_SIZE(xennet_stats
); i
++)
1761 memcpy(data
+ i
* ETH_GSTRING_LEN
,
1762 xennet_stats
[i
].name
, ETH_GSTRING_LEN
);
1767 static const struct ethtool_ops xennet_ethtool_ops
=
1769 .get_link
= ethtool_op_get_link
,
1771 .get_sset_count
= xennet_get_sset_count
,
1772 .get_ethtool_stats
= xennet_get_ethtool_stats
,
1773 .get_strings
= xennet_get_strings
,
1777 static ssize_t
show_rxbuf_min(struct device
*dev
,
1778 struct device_attribute
*attr
, char *buf
)
1780 struct net_device
*netdev
= to_net_dev(dev
);
1781 struct netfront_info
*info
= netdev_priv(netdev
);
1783 return sprintf(buf
, "%u\n", info
->rx_min_target
);
1786 static ssize_t
store_rxbuf_min(struct device
*dev
,
1787 struct device_attribute
*attr
,
1788 const char *buf
, size_t len
)
1790 struct net_device
*netdev
= to_net_dev(dev
);
1791 struct netfront_info
*np
= netdev_priv(netdev
);
1793 unsigned long target
;
1795 if (!capable(CAP_NET_ADMIN
))
1798 target
= simple_strtoul(buf
, &endp
, 0);
1802 if (target
< RX_MIN_TARGET
)
1803 target
= RX_MIN_TARGET
;
1804 if (target
> RX_MAX_TARGET
)
1805 target
= RX_MAX_TARGET
;
1807 spin_lock_bh(&np
->rx_lock
);
1808 if (target
> np
->rx_max_target
)
1809 np
->rx_max_target
= target
;
1810 np
->rx_min_target
= target
;
1811 if (target
> np
->rx_target
)
1812 np
->rx_target
= target
;
1814 xennet_alloc_rx_buffers(netdev
);
1816 spin_unlock_bh(&np
->rx_lock
);
1820 static ssize_t
show_rxbuf_max(struct device
*dev
,
1821 struct device_attribute
*attr
, char *buf
)
1823 struct net_device
*netdev
= to_net_dev(dev
);
1824 struct netfront_info
*info
= netdev_priv(netdev
);
1826 return sprintf(buf
, "%u\n", info
->rx_max_target
);
1829 static ssize_t
store_rxbuf_max(struct device
*dev
,
1830 struct device_attribute
*attr
,
1831 const char *buf
, size_t len
)
1833 struct net_device
*netdev
= to_net_dev(dev
);
1834 struct netfront_info
*np
= netdev_priv(netdev
);
1836 unsigned long target
;
1838 if (!capable(CAP_NET_ADMIN
))
1841 target
= simple_strtoul(buf
, &endp
, 0);
1845 if (target
< RX_MIN_TARGET
)
1846 target
= RX_MIN_TARGET
;
1847 if (target
> RX_MAX_TARGET
)
1848 target
= RX_MAX_TARGET
;
1850 spin_lock_bh(&np
->rx_lock
);
1851 if (target
< np
->rx_min_target
)
1852 np
->rx_min_target
= target
;
1853 np
->rx_max_target
= target
;
1854 if (target
< np
->rx_target
)
1855 np
->rx_target
= target
;
1857 xennet_alloc_rx_buffers(netdev
);
1859 spin_unlock_bh(&np
->rx_lock
);
1863 static ssize_t
show_rxbuf_cur(struct device
*dev
,
1864 struct device_attribute
*attr
, char *buf
)
1866 struct net_device
*netdev
= to_net_dev(dev
);
1867 struct netfront_info
*info
= netdev_priv(netdev
);
1869 return sprintf(buf
, "%u\n", info
->rx_target
);
1872 static struct device_attribute xennet_attrs
[] = {
1873 __ATTR(rxbuf_min
, S_IRUGO
|S_IWUSR
, show_rxbuf_min
, store_rxbuf_min
),
1874 __ATTR(rxbuf_max
, S_IRUGO
|S_IWUSR
, show_rxbuf_max
, store_rxbuf_max
),
1875 __ATTR(rxbuf_cur
, S_IRUGO
, show_rxbuf_cur
, NULL
),
1878 static int xennet_sysfs_addif(struct net_device
*netdev
)
1883 for (i
= 0; i
< ARRAY_SIZE(xennet_attrs
); i
++) {
1884 err
= device_create_file(&netdev
->dev
,
1893 device_remove_file(&netdev
->dev
, &xennet_attrs
[i
]);
1897 static void xennet_sysfs_delif(struct net_device
*netdev
)
1901 for (i
= 0; i
< ARRAY_SIZE(xennet_attrs
); i
++)
1902 device_remove_file(&netdev
->dev
, &xennet_attrs
[i
]);
1905 #endif /* CONFIG_SYSFS */
1907 static const struct xenbus_device_id netfront_ids
[] = {
1913 static int __devexit
xennet_remove(struct xenbus_device
*dev
)
1915 struct netfront_info
*info
= dev_get_drvdata(&dev
->dev
);
1917 dev_dbg(&dev
->dev
, "%s\n", dev
->nodename
);
1919 xennet_disconnect_backend(info
);
1921 xennet_sysfs_delif(info
->netdev
);
1923 unregister_netdev(info
->netdev
);
1925 del_timer_sync(&info
->rx_refill_timer
);
1927 free_percpu(info
->stats
);
1929 free_netdev(info
->netdev
);
1934 static DEFINE_XENBUS_DRIVER(netfront
, ,
1935 .probe
= netfront_probe
,
1936 .remove
= __devexit_p(xennet_remove
),
1937 .resume
= netfront_resume
,
1938 .otherend_changed
= netback_changed
,
1941 static int __init
netif_init(void)
1946 if (xen_hvm_domain() && !xen_platform_pci_unplug
)
1949 printk(KERN_INFO
"Initialising Xen virtual ethernet driver.\n");
1951 return xenbus_register_frontend(&netfront_driver
);
1953 module_init(netif_init
);
1956 static void __exit
netif_exit(void)
1958 xenbus_unregister_driver(&netfront_driver
);
1960 module_exit(netif_exit
);
1962 MODULE_DESCRIPTION("Xen virtual network device frontend");
1963 MODULE_LICENSE("GPL");
1964 MODULE_ALIAS("xen:vif");
1965 MODULE_ALIAS("xennet");