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>
45 #include <xen/xenbus.h>
46 #include <xen/events.h>
48 #include <xen/grant_table.h>
50 #include <xen/interface/io/netif.h>
51 #include <xen/interface/memory.h>
52 #include <xen/interface/grant_table.h>
54 static struct ethtool_ops xennet_ethtool_ops
;
61 #define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb))
63 #define RX_COPY_THRESHOLD 256
65 #define GRANT_INVALID_REF 0
67 #define NET_TX_RING_SIZE __RING_SIZE((struct xen_netif_tx_sring *)0, PAGE_SIZE)
68 #define NET_RX_RING_SIZE __RING_SIZE((struct xen_netif_rx_sring *)0, PAGE_SIZE)
69 #define TX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256)
71 struct netfront_info
{
72 struct list_head list
;
73 struct net_device
*netdev
;
75 struct napi_struct napi
;
78 struct xenbus_device
*xbdev
;
81 struct xen_netif_tx_front_ring tx
;
85 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
86 * are linked from tx_skb_freelist through skb_entry.link.
88 * NB. Freelist index entries are always going to be less than
89 * PAGE_OFFSET, whereas pointers to skbs will always be equal or
90 * greater than PAGE_OFFSET: we use this property to distinguish
96 } tx_skbs
[NET_TX_RING_SIZE
];
97 grant_ref_t gref_tx_head
;
98 grant_ref_t grant_tx_ref
[NET_TX_RING_SIZE
];
99 unsigned tx_skb_freelist
;
101 spinlock_t rx_lock ____cacheline_aligned_in_smp
;
102 struct xen_netif_rx_front_ring rx
;
105 /* Receive-ring batched refills. */
106 #define RX_MIN_TARGET 8
107 #define RX_DFL_MIN_TARGET 64
108 #define RX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256)
109 unsigned rx_min_target
, rx_max_target
, rx_target
;
110 struct sk_buff_head rx_batch
;
112 struct timer_list rx_refill_timer
;
114 struct sk_buff
*rx_skbs
[NET_RX_RING_SIZE
];
115 grant_ref_t gref_rx_head
;
116 grant_ref_t grant_rx_ref
[NET_RX_RING_SIZE
];
118 unsigned long rx_pfn_array
[NET_RX_RING_SIZE
];
119 struct multicall_entry rx_mcl
[NET_RX_RING_SIZE
+1];
120 struct mmu_update rx_mmu
[NET_RX_RING_SIZE
];
123 struct netfront_rx_info
{
124 struct xen_netif_rx_response rx
;
125 struct xen_netif_extra_info extras
[XEN_NETIF_EXTRA_TYPE_MAX
- 1];
128 static void skb_entry_set_link(union skb_entry
*list
, unsigned short id
)
133 static int skb_entry_is_link(const union skb_entry
*list
)
135 BUILD_BUG_ON(sizeof(list
->skb
) != sizeof(list
->link
));
136 return ((unsigned long)list
->skb
< PAGE_OFFSET
);
140 * Access macros for acquiring freeing slots in tx_skbs[].
143 static void add_id_to_freelist(unsigned *head
, union skb_entry
*list
,
146 skb_entry_set_link(&list
[id
], *head
);
150 static unsigned short get_id_from_freelist(unsigned *head
,
151 union skb_entry
*list
)
153 unsigned int id
= *head
;
154 *head
= list
[id
].link
;
158 static int xennet_rxidx(RING_IDX idx
)
160 return idx
& (NET_RX_RING_SIZE
- 1);
163 static struct sk_buff
*xennet_get_rx_skb(struct netfront_info
*np
,
166 int i
= xennet_rxidx(ri
);
167 struct sk_buff
*skb
= np
->rx_skbs
[i
];
168 np
->rx_skbs
[i
] = NULL
;
172 static grant_ref_t
xennet_get_rx_ref(struct netfront_info
*np
,
175 int i
= xennet_rxidx(ri
);
176 grant_ref_t ref
= np
->grant_rx_ref
[i
];
177 np
->grant_rx_ref
[i
] = GRANT_INVALID_REF
;
182 static int xennet_sysfs_addif(struct net_device
*netdev
);
183 static void xennet_sysfs_delif(struct net_device
*netdev
);
184 #else /* !CONFIG_SYSFS */
185 #define xennet_sysfs_addif(dev) (0)
186 #define xennet_sysfs_delif(dev) do { } while (0)
189 static int xennet_can_sg(struct net_device
*dev
)
191 return dev
->features
& NETIF_F_SG
;
195 static void rx_refill_timeout(unsigned long data
)
197 struct net_device
*dev
= (struct net_device
*)data
;
198 struct netfront_info
*np
= netdev_priv(dev
);
199 netif_rx_schedule(dev
, &np
->napi
);
202 static int netfront_tx_slot_available(struct netfront_info
*np
)
204 return ((np
->tx
.req_prod_pvt
- np
->tx
.rsp_cons
) <
205 (TX_MAX_TARGET
- MAX_SKB_FRAGS
- 2));
208 static void xennet_maybe_wake_tx(struct net_device
*dev
)
210 struct netfront_info
*np
= netdev_priv(dev
);
212 if (unlikely(netif_queue_stopped(dev
)) &&
213 netfront_tx_slot_available(np
) &&
214 likely(netif_running(dev
)))
215 netif_wake_queue(dev
);
218 static void xennet_alloc_rx_buffers(struct net_device
*dev
)
221 struct netfront_info
*np
= netdev_priv(dev
);
224 int i
, batch_target
, notify
;
225 RING_IDX req_prod
= np
->rx
.req_prod_pvt
;
229 struct xen_netif_rx_request
*req
;
231 if (unlikely(!netif_carrier_ok(dev
)))
235 * Allocate skbuffs greedily, even though we batch updates to the
236 * receive ring. This creates a less bursty demand on the memory
237 * allocator, so should reduce the chance of failed allocation requests
238 * both for ourself and for other kernel subsystems.
240 batch_target
= np
->rx_target
- (req_prod
- np
->rx
.rsp_cons
);
241 for (i
= skb_queue_len(&np
->rx_batch
); i
< batch_target
; i
++) {
242 skb
= __netdev_alloc_skb(dev
, RX_COPY_THRESHOLD
,
243 GFP_ATOMIC
| __GFP_NOWARN
);
247 page
= alloc_page(GFP_ATOMIC
| __GFP_NOWARN
);
251 /* Any skbuffs queued for refill? Force them out. */
254 /* Could not allocate any skbuffs. Try again later. */
255 mod_timer(&np
->rx_refill_timer
,
260 skb_shinfo(skb
)->frags
[0].page
= page
;
261 skb_shinfo(skb
)->nr_frags
= 1;
262 __skb_queue_tail(&np
->rx_batch
, skb
);
265 /* Is the batch large enough to be worthwhile? */
266 if (i
< (np
->rx_target
/2)) {
267 if (req_prod
> np
->rx
.sring
->req_prod
)
272 /* Adjust our fill target if we risked running out of buffers. */
273 if (((req_prod
- np
->rx
.sring
->rsp_prod
) < (np
->rx_target
/ 4)) &&
274 ((np
->rx_target
*= 2) > np
->rx_max_target
))
275 np
->rx_target
= np
->rx_max_target
;
279 skb
= __skb_dequeue(&np
->rx_batch
);
285 id
= xennet_rxidx(req_prod
+ i
);
287 BUG_ON(np
->rx_skbs
[id
]);
288 np
->rx_skbs
[id
] = skb
;
290 ref
= gnttab_claim_grant_reference(&np
->gref_rx_head
);
291 BUG_ON((signed short)ref
< 0);
292 np
->grant_rx_ref
[id
] = ref
;
294 pfn
= page_to_pfn(skb_shinfo(skb
)->frags
[0].page
);
295 vaddr
= page_address(skb_shinfo(skb
)->frags
[0].page
);
297 req
= RING_GET_REQUEST(&np
->rx
, req_prod
+ i
);
298 gnttab_grant_foreign_access_ref(ref
,
299 np
->xbdev
->otherend_id
,
307 wmb(); /* barrier so backend seens requests */
309 /* Above is a suitable barrier to ensure backend will see requests. */
310 np
->rx
.req_prod_pvt
= req_prod
+ i
;
312 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np
->rx
, notify
);
314 notify_remote_via_irq(np
->netdev
->irq
);
317 static int xennet_open(struct net_device
*dev
)
319 struct netfront_info
*np
= netdev_priv(dev
);
321 napi_enable(&np
->napi
);
323 spin_lock_bh(&np
->rx_lock
);
324 if (netif_carrier_ok(dev
)) {
325 xennet_alloc_rx_buffers(dev
);
326 np
->rx
.sring
->rsp_event
= np
->rx
.rsp_cons
+ 1;
327 if (RING_HAS_UNCONSUMED_RESPONSES(&np
->rx
))
328 netif_rx_schedule(dev
, &np
->napi
);
330 spin_unlock_bh(&np
->rx_lock
);
332 xennet_maybe_wake_tx(dev
);
337 static void xennet_tx_buf_gc(struct net_device
*dev
)
341 struct netfront_info
*np
= netdev_priv(dev
);
344 BUG_ON(!netif_carrier_ok(dev
));
347 prod
= np
->tx
.sring
->rsp_prod
;
348 rmb(); /* Ensure we see responses up to 'rp'. */
350 for (cons
= np
->tx
.rsp_cons
; cons
!= prod
; cons
++) {
351 struct xen_netif_tx_response
*txrsp
;
353 txrsp
= RING_GET_RESPONSE(&np
->tx
, cons
);
354 if (txrsp
->status
== NETIF_RSP_NULL
)
358 skb
= np
->tx_skbs
[id
].skb
;
359 if (unlikely(gnttab_query_foreign_access(
360 np
->grant_tx_ref
[id
]) != 0)) {
361 printk(KERN_ALERT
"xennet_tx_buf_gc: warning "
362 "-- grant still in use by backend "
366 gnttab_end_foreign_access_ref(
367 np
->grant_tx_ref
[id
], GNTMAP_readonly
);
368 gnttab_release_grant_reference(
369 &np
->gref_tx_head
, np
->grant_tx_ref
[id
]);
370 np
->grant_tx_ref
[id
] = GRANT_INVALID_REF
;
371 add_id_to_freelist(&np
->tx_skb_freelist
, np
->tx_skbs
, id
);
372 dev_kfree_skb_irq(skb
);
375 np
->tx
.rsp_cons
= prod
;
378 * Set a new event, then check for race with update of tx_cons.
379 * Note that it is essential to schedule a callback, no matter
380 * how few buffers are pending. Even if there is space in the
381 * transmit ring, higher layers may be blocked because too much
382 * data is outstanding: in such cases notification from Xen is
383 * likely to be the only kick that we'll get.
385 np
->tx
.sring
->rsp_event
=
386 prod
+ ((np
->tx
.sring
->req_prod
- prod
) >> 1) + 1;
387 mb(); /* update shared area */
388 } while ((cons
== prod
) && (prod
!= np
->tx
.sring
->rsp_prod
));
390 xennet_maybe_wake_tx(dev
);
393 static void xennet_make_frags(struct sk_buff
*skb
, struct net_device
*dev
,
394 struct xen_netif_tx_request
*tx
)
396 struct netfront_info
*np
= netdev_priv(dev
);
397 char *data
= skb
->data
;
399 RING_IDX prod
= np
->tx
.req_prod_pvt
;
400 int frags
= skb_shinfo(skb
)->nr_frags
;
401 unsigned int offset
= offset_in_page(data
);
402 unsigned int len
= skb_headlen(skb
);
407 /* While the header overlaps a page boundary (including being
408 larger than a page), split it it into page-sized chunks. */
409 while (len
> PAGE_SIZE
- offset
) {
410 tx
->size
= PAGE_SIZE
- offset
;
411 tx
->flags
|= NETTXF_more_data
;
416 id
= get_id_from_freelist(&np
->tx_skb_freelist
, np
->tx_skbs
);
417 np
->tx_skbs
[id
].skb
= skb_get(skb
);
418 tx
= RING_GET_REQUEST(&np
->tx
, prod
++);
420 ref
= gnttab_claim_grant_reference(&np
->gref_tx_head
);
421 BUG_ON((signed short)ref
< 0);
423 mfn
= virt_to_mfn(data
);
424 gnttab_grant_foreign_access_ref(ref
, np
->xbdev
->otherend_id
,
425 mfn
, GNTMAP_readonly
);
427 tx
->gref
= np
->grant_tx_ref
[id
] = ref
;
433 /* Grant backend access to each skb fragment page. */
434 for (i
= 0; i
< frags
; i
++) {
435 skb_frag_t
*frag
= skb_shinfo(skb
)->frags
+ i
;
437 tx
->flags
|= NETTXF_more_data
;
439 id
= get_id_from_freelist(&np
->tx_skb_freelist
, np
->tx_skbs
);
440 np
->tx_skbs
[id
].skb
= skb_get(skb
);
441 tx
= RING_GET_REQUEST(&np
->tx
, prod
++);
443 ref
= gnttab_claim_grant_reference(&np
->gref_tx_head
);
444 BUG_ON((signed short)ref
< 0);
446 mfn
= pfn_to_mfn(page_to_pfn(frag
->page
));
447 gnttab_grant_foreign_access_ref(ref
, np
->xbdev
->otherend_id
,
448 mfn
, GNTMAP_readonly
);
450 tx
->gref
= np
->grant_tx_ref
[id
] = ref
;
451 tx
->offset
= frag
->page_offset
;
452 tx
->size
= frag
->size
;
456 np
->tx
.req_prod_pvt
= prod
;
459 static int xennet_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
462 struct netfront_info
*np
= netdev_priv(dev
);
463 struct xen_netif_tx_request
*tx
;
464 struct xen_netif_extra_info
*extra
;
465 char *data
= skb
->data
;
470 int frags
= skb_shinfo(skb
)->nr_frags
;
471 unsigned int offset
= offset_in_page(data
);
472 unsigned int len
= skb_headlen(skb
);
474 frags
+= (offset
+ len
+ PAGE_SIZE
- 1) / PAGE_SIZE
;
475 if (unlikely(frags
> MAX_SKB_FRAGS
+ 1)) {
476 printk(KERN_ALERT
"xennet: skb rides the rocket: %d frags\n",
482 spin_lock_irq(&np
->tx_lock
);
484 if (unlikely(!netif_carrier_ok(dev
) ||
485 (frags
> 1 && !xennet_can_sg(dev
)) ||
486 netif_needs_gso(dev
, skb
))) {
487 spin_unlock_irq(&np
->tx_lock
);
491 i
= np
->tx
.req_prod_pvt
;
493 id
= get_id_from_freelist(&np
->tx_skb_freelist
, np
->tx_skbs
);
494 np
->tx_skbs
[id
].skb
= skb
;
496 tx
= RING_GET_REQUEST(&np
->tx
, i
);
499 ref
= gnttab_claim_grant_reference(&np
->gref_tx_head
);
500 BUG_ON((signed short)ref
< 0);
501 mfn
= virt_to_mfn(data
);
502 gnttab_grant_foreign_access_ref(
503 ref
, np
->xbdev
->otherend_id
, mfn
, GNTMAP_readonly
);
504 tx
->gref
= np
->grant_tx_ref
[id
] = ref
;
510 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
512 tx
->flags
|= NETTXF_csum_blank
| NETTXF_data_validated
;
513 else if (skb
->ip_summed
== CHECKSUM_UNNECESSARY
)
514 /* remote but checksummed. */
515 tx
->flags
|= NETTXF_data_validated
;
517 if (skb_shinfo(skb
)->gso_size
) {
518 struct xen_netif_extra_info
*gso
;
520 gso
= (struct xen_netif_extra_info
*)
521 RING_GET_REQUEST(&np
->tx
, ++i
);
524 extra
->flags
|= XEN_NETIF_EXTRA_FLAG_MORE
;
526 tx
->flags
|= NETTXF_extra_info
;
528 gso
->u
.gso
.size
= skb_shinfo(skb
)->gso_size
;
529 gso
->u
.gso
.type
= XEN_NETIF_GSO_TYPE_TCPV4
;
531 gso
->u
.gso
.features
= 0;
533 gso
->type
= XEN_NETIF_EXTRA_TYPE_GSO
;
538 np
->tx
.req_prod_pvt
= i
+ 1;
540 xennet_make_frags(skb
, dev
, tx
);
543 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np
->tx
, notify
);
545 notify_remote_via_irq(np
->netdev
->irq
);
547 dev
->stats
.tx_bytes
+= skb
->len
;
548 dev
->stats
.tx_packets
++;
550 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
551 xennet_tx_buf_gc(dev
);
553 if (!netfront_tx_slot_available(np
))
554 netif_stop_queue(dev
);
556 spin_unlock_irq(&np
->tx_lock
);
561 dev
->stats
.tx_dropped
++;
566 static int xennet_close(struct net_device
*dev
)
568 struct netfront_info
*np
= netdev_priv(dev
);
569 netif_stop_queue(np
->netdev
);
570 napi_disable(&np
->napi
);
574 static void xennet_move_rx_slot(struct netfront_info
*np
, struct sk_buff
*skb
,
577 int new = xennet_rxidx(np
->rx
.req_prod_pvt
);
579 BUG_ON(np
->rx_skbs
[new]);
580 np
->rx_skbs
[new] = skb
;
581 np
->grant_rx_ref
[new] = ref
;
582 RING_GET_REQUEST(&np
->rx
, np
->rx
.req_prod_pvt
)->id
= new;
583 RING_GET_REQUEST(&np
->rx
, np
->rx
.req_prod_pvt
)->gref
= ref
;
584 np
->rx
.req_prod_pvt
++;
587 static int xennet_get_extras(struct netfront_info
*np
,
588 struct xen_netif_extra_info
*extras
,
592 struct xen_netif_extra_info
*extra
;
593 struct device
*dev
= &np
->netdev
->dev
;
594 RING_IDX cons
= np
->rx
.rsp_cons
;
601 if (unlikely(cons
+ 1 == rp
)) {
603 dev_warn(dev
, "Missing extra info\n");
608 extra
= (struct xen_netif_extra_info
*)
609 RING_GET_RESPONSE(&np
->rx
, ++cons
);
611 if (unlikely(!extra
->type
||
612 extra
->type
>= XEN_NETIF_EXTRA_TYPE_MAX
)) {
614 dev_warn(dev
, "Invalid extra type: %d\n",
618 memcpy(&extras
[extra
->type
- 1], extra
,
622 skb
= xennet_get_rx_skb(np
, cons
);
623 ref
= xennet_get_rx_ref(np
, cons
);
624 xennet_move_rx_slot(np
, skb
, ref
);
625 } while (extra
->flags
& XEN_NETIF_EXTRA_FLAG_MORE
);
627 np
->rx
.rsp_cons
= cons
;
631 static int xennet_get_responses(struct netfront_info
*np
,
632 struct netfront_rx_info
*rinfo
, RING_IDX rp
,
633 struct sk_buff_head
*list
)
635 struct xen_netif_rx_response
*rx
= &rinfo
->rx
;
636 struct xen_netif_extra_info
*extras
= rinfo
->extras
;
637 struct device
*dev
= &np
->netdev
->dev
;
638 RING_IDX cons
= np
->rx
.rsp_cons
;
639 struct sk_buff
*skb
= xennet_get_rx_skb(np
, cons
);
640 grant_ref_t ref
= xennet_get_rx_ref(np
, cons
);
641 int max
= MAX_SKB_FRAGS
+ (rx
->status
<= RX_COPY_THRESHOLD
);
646 if (rx
->flags
& NETRXF_extra_info
) {
647 err
= xennet_get_extras(np
, extras
, rp
);
648 cons
= np
->rx
.rsp_cons
;
652 if (unlikely(rx
->status
< 0 ||
653 rx
->offset
+ rx
->status
> PAGE_SIZE
)) {
655 dev_warn(dev
, "rx->offset: %x, size: %u\n",
656 rx
->offset
, rx
->status
);
657 xennet_move_rx_slot(np
, skb
, ref
);
663 * This definitely indicates a bug, either in this driver or in
664 * the backend driver. In future this should flag the bad
665 * situation to the system controller to reboot the backed.
667 if (ref
== GRANT_INVALID_REF
) {
669 dev_warn(dev
, "Bad rx response id %d.\n",
675 ret
= gnttab_end_foreign_access_ref(ref
, 0);
678 gnttab_release_grant_reference(&np
->gref_rx_head
, ref
);
680 __skb_queue_tail(list
, skb
);
683 if (!(rx
->flags
& NETRXF_more_data
))
686 if (cons
+ frags
== rp
) {
688 dev_warn(dev
, "Need more frags\n");
693 rx
= RING_GET_RESPONSE(&np
->rx
, cons
+ frags
);
694 skb
= xennet_get_rx_skb(np
, cons
+ frags
);
695 ref
= xennet_get_rx_ref(np
, cons
+ frags
);
699 if (unlikely(frags
> max
)) {
701 dev_warn(dev
, "Too many frags\n");
706 np
->rx
.rsp_cons
= cons
+ frags
;
711 static int xennet_set_skb_gso(struct sk_buff
*skb
,
712 struct xen_netif_extra_info
*gso
)
714 if (!gso
->u
.gso
.size
) {
716 printk(KERN_WARNING
"GSO size must not be zero.\n");
720 /* Currently only TCPv4 S.O. is supported. */
721 if (gso
->u
.gso
.type
!= XEN_NETIF_GSO_TYPE_TCPV4
) {
723 printk(KERN_WARNING
"Bad GSO type %d.\n", gso
->u
.gso
.type
);
727 skb_shinfo(skb
)->gso_size
= gso
->u
.gso
.size
;
728 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV4
;
730 /* Header must be checked, and gso_segs computed. */
731 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
732 skb_shinfo(skb
)->gso_segs
= 0;
737 static RING_IDX
xennet_fill_frags(struct netfront_info
*np
,
739 struct sk_buff_head
*list
)
741 struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
742 int nr_frags
= shinfo
->nr_frags
;
743 RING_IDX cons
= np
->rx
.rsp_cons
;
744 skb_frag_t
*frag
= shinfo
->frags
+ nr_frags
;
745 struct sk_buff
*nskb
;
747 while ((nskb
= __skb_dequeue(list
))) {
748 struct xen_netif_rx_response
*rx
=
749 RING_GET_RESPONSE(&np
->rx
, ++cons
);
751 frag
->page
= skb_shinfo(nskb
)->frags
[0].page
;
752 frag
->page_offset
= rx
->offset
;
753 frag
->size
= rx
->status
;
755 skb
->data_len
+= rx
->status
;
757 skb_shinfo(nskb
)->nr_frags
= 0;
764 shinfo
->nr_frags
= nr_frags
;
768 static int skb_checksum_setup(struct sk_buff
*skb
)
774 if (skb
->protocol
!= htons(ETH_P_IP
))
777 iph
= (void *)skb
->data
;
778 th
= skb
->data
+ 4 * iph
->ihl
;
779 if (th
>= skb_tail_pointer(skb
))
782 skb
->csum_start
= th
- skb
->head
;
783 switch (iph
->protocol
) {
785 skb
->csum_offset
= offsetof(struct tcphdr
, check
);
788 skb
->csum_offset
= offsetof(struct udphdr
, check
);
792 printk(KERN_ERR
"Attempting to checksum a non-"
793 "TCP/UDP packet, dropping a protocol"
794 " %d packet", iph
->protocol
);
798 if ((th
+ skb
->csum_offset
+ 2) > skb_tail_pointer(skb
))
807 static int handle_incoming_queue(struct net_device
*dev
,
808 struct sk_buff_head
*rxq
)
810 int packets_dropped
= 0;
813 while ((skb
= __skb_dequeue(rxq
)) != NULL
) {
814 struct page
*page
= NETFRONT_SKB_CB(skb
)->page
;
815 void *vaddr
= page_address(page
);
816 unsigned offset
= NETFRONT_SKB_CB(skb
)->offset
;
818 memcpy(skb
->data
, vaddr
+ offset
,
821 if (page
!= skb_shinfo(skb
)->frags
[0].page
)
824 /* Ethernet work: Delayed to here as it peeks the header. */
825 skb
->protocol
= eth_type_trans(skb
, dev
);
827 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
828 if (skb_checksum_setup(skb
)) {
831 dev
->stats
.rx_errors
++;
836 dev
->stats
.rx_packets
++;
837 dev
->stats
.rx_bytes
+= skb
->len
;
840 netif_receive_skb(skb
);
841 dev
->last_rx
= jiffies
;
844 return packets_dropped
;
847 static int xennet_poll(struct napi_struct
*napi
, int budget
)
849 struct netfront_info
*np
= container_of(napi
, struct netfront_info
, napi
);
850 struct net_device
*dev
= np
->netdev
;
852 struct netfront_rx_info rinfo
;
853 struct xen_netif_rx_response
*rx
= &rinfo
.rx
;
854 struct xen_netif_extra_info
*extras
= rinfo
.extras
;
857 struct sk_buff_head rxq
;
858 struct sk_buff_head errq
;
859 struct sk_buff_head tmpq
;
864 spin_lock(&np
->rx_lock
);
866 skb_queue_head_init(&rxq
);
867 skb_queue_head_init(&errq
);
868 skb_queue_head_init(&tmpq
);
870 rp
= np
->rx
.sring
->rsp_prod
;
871 rmb(); /* Ensure we see queued responses up to 'rp'. */
875 while ((i
!= rp
) && (work_done
< budget
)) {
876 memcpy(rx
, RING_GET_RESPONSE(&np
->rx
, i
), sizeof(*rx
));
877 memset(extras
, 0, sizeof(rinfo
.extras
));
879 err
= xennet_get_responses(np
, &rinfo
, rp
, &tmpq
);
883 while ((skb
= __skb_dequeue(&tmpq
)))
884 __skb_queue_tail(&errq
, skb
);
885 dev
->stats
.rx_errors
++;
890 skb
= __skb_dequeue(&tmpq
);
892 if (extras
[XEN_NETIF_EXTRA_TYPE_GSO
- 1].type
) {
893 struct xen_netif_extra_info
*gso
;
894 gso
= &extras
[XEN_NETIF_EXTRA_TYPE_GSO
- 1];
896 if (unlikely(xennet_set_skb_gso(skb
, gso
))) {
897 __skb_queue_head(&tmpq
, skb
);
898 np
->rx
.rsp_cons
+= skb_queue_len(&tmpq
);
903 NETFRONT_SKB_CB(skb
)->page
= skb_shinfo(skb
)->frags
[0].page
;
904 NETFRONT_SKB_CB(skb
)->offset
= rx
->offset
;
907 if (len
> RX_COPY_THRESHOLD
)
908 len
= RX_COPY_THRESHOLD
;
911 if (rx
->status
> len
) {
912 skb_shinfo(skb
)->frags
[0].page_offset
=
914 skb_shinfo(skb
)->frags
[0].size
= rx
->status
- len
;
915 skb
->data_len
= rx
->status
- len
;
917 skb_shinfo(skb
)->frags
[0].page
= NULL
;
918 skb_shinfo(skb
)->nr_frags
= 0;
921 i
= xennet_fill_frags(np
, skb
, &tmpq
);
924 * Truesize approximates the size of true data plus
925 * any supervisor overheads. Adding hypervisor
926 * overheads has been shown to significantly reduce
927 * achievable bandwidth with the default receive
928 * buffer size. It is therefore not wise to account
931 * After alloc_skb(RX_COPY_THRESHOLD), truesize is set
932 * to RX_COPY_THRESHOLD + the supervisor
933 * overheads. Here, we add the size of the data pulled
934 * in xennet_fill_frags().
936 * We also adjust for any unused space in the main
937 * data area by subtracting (RX_COPY_THRESHOLD -
938 * len). This is especially important with drivers
939 * which split incoming packets into header and data,
940 * using only 66 bytes of the main data area (see the
941 * e1000 driver for example.) On such systems,
942 * without this last adjustement, our achievable
943 * receive throughout using the standard receive
944 * buffer size was cut by 25%(!!!).
946 skb
->truesize
+= skb
->data_len
- (RX_COPY_THRESHOLD
- len
);
947 skb
->len
+= skb
->data_len
;
949 if (rx
->flags
& NETRXF_csum_blank
)
950 skb
->ip_summed
= CHECKSUM_PARTIAL
;
951 else if (rx
->flags
& NETRXF_data_validated
)
952 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
954 __skb_queue_tail(&rxq
, skb
);
956 np
->rx
.rsp_cons
= ++i
;
960 __skb_queue_purge(&errq
);
962 work_done
-= handle_incoming_queue(dev
, &rxq
);
964 /* If we get a callback with very few responses, reduce fill target. */
965 /* NB. Note exponential increase, linear decrease. */
966 if (((np
->rx
.req_prod_pvt
- np
->rx
.sring
->rsp_prod
) >
967 ((3*np
->rx_target
) / 4)) &&
968 (--np
->rx_target
< np
->rx_min_target
))
969 np
->rx_target
= np
->rx_min_target
;
971 xennet_alloc_rx_buffers(dev
);
973 if (work_done
< budget
) {
976 local_irq_save(flags
);
978 RING_FINAL_CHECK_FOR_RESPONSES(&np
->rx
, more_to_do
);
980 __netif_rx_complete(dev
, napi
);
982 local_irq_restore(flags
);
985 spin_unlock(&np
->rx_lock
);
990 static int xennet_change_mtu(struct net_device
*dev
, int mtu
)
992 int max
= xennet_can_sg(dev
) ? 65535 - ETH_HLEN
: ETH_DATA_LEN
;
1000 static void xennet_release_tx_bufs(struct netfront_info
*np
)
1002 struct sk_buff
*skb
;
1005 for (i
= 0; i
< NET_TX_RING_SIZE
; i
++) {
1006 /* Skip over entries which are actually freelist references */
1007 if (skb_entry_is_link(&np
->tx_skbs
[i
]))
1010 skb
= np
->tx_skbs
[i
].skb
;
1011 gnttab_end_foreign_access_ref(np
->grant_tx_ref
[i
],
1013 gnttab_release_grant_reference(&np
->gref_tx_head
,
1014 np
->grant_tx_ref
[i
]);
1015 np
->grant_tx_ref
[i
] = GRANT_INVALID_REF
;
1016 add_id_to_freelist(&np
->tx_skb_freelist
, np
->tx_skbs
, i
);
1017 dev_kfree_skb_irq(skb
);
1021 static void xennet_release_rx_bufs(struct netfront_info
*np
)
1023 struct mmu_update
*mmu
= np
->rx_mmu
;
1024 struct multicall_entry
*mcl
= np
->rx_mcl
;
1025 struct sk_buff_head free_list
;
1026 struct sk_buff
*skb
;
1028 int xfer
= 0, noxfer
= 0, unused
= 0;
1031 dev_warn(&np
->netdev
->dev
, "%s: fix me for copying receiver.\n",
1035 skb_queue_head_init(&free_list
);
1037 spin_lock_bh(&np
->rx_lock
);
1039 for (id
= 0; id
< NET_RX_RING_SIZE
; id
++) {
1040 ref
= np
->grant_rx_ref
[id
];
1041 if (ref
== GRANT_INVALID_REF
) {
1046 skb
= np
->rx_skbs
[id
];
1047 mfn
= gnttab_end_foreign_transfer_ref(ref
);
1048 gnttab_release_grant_reference(&np
->gref_rx_head
, ref
);
1049 np
->grant_rx_ref
[id
] = GRANT_INVALID_REF
;
1052 skb_shinfo(skb
)->nr_frags
= 0;
1058 if (!xen_feature(XENFEAT_auto_translated_physmap
)) {
1059 /* Remap the page. */
1060 struct page
*page
= skb_shinfo(skb
)->frags
[0].page
;
1061 unsigned long pfn
= page_to_pfn(page
);
1062 void *vaddr
= page_address(page
);
1064 MULTI_update_va_mapping(mcl
, (unsigned long)vaddr
,
1065 mfn_pte(mfn
, PAGE_KERNEL
),
1068 mmu
->ptr
= ((u64
)mfn
<< PAGE_SHIFT
)
1069 | MMU_MACHPHYS_UPDATE
;
1073 set_phys_to_machine(pfn
, mfn
);
1075 __skb_queue_tail(&free_list
, skb
);
1079 dev_info(&np
->netdev
->dev
, "%s: %d xfer, %d noxfer, %d unused\n",
1080 __func__
, xfer
, noxfer
, unused
);
1083 if (!xen_feature(XENFEAT_auto_translated_physmap
)) {
1084 /* Do all the remapping work and M2P updates. */
1085 MULTI_mmu_update(mcl
, np
->rx_mmu
, mmu
- np
->rx_mmu
,
1088 HYPERVISOR_multicall(np
->rx_mcl
, mcl
- np
->rx_mcl
);
1092 __skb_queue_purge(&free_list
);
1094 spin_unlock_bh(&np
->rx_lock
);
1097 static void xennet_uninit(struct net_device
*dev
)
1099 struct netfront_info
*np
= netdev_priv(dev
);
1100 xennet_release_tx_bufs(np
);
1101 xennet_release_rx_bufs(np
);
1102 gnttab_free_grant_references(np
->gref_tx_head
);
1103 gnttab_free_grant_references(np
->gref_rx_head
);
1106 static struct net_device
* __devinit
xennet_create_dev(struct xenbus_device
*dev
)
1109 struct net_device
*netdev
;
1110 struct netfront_info
*np
;
1112 netdev
= alloc_etherdev(sizeof(struct netfront_info
));
1114 printk(KERN_WARNING
"%s> alloc_etherdev failed.\n",
1116 return ERR_PTR(-ENOMEM
);
1119 np
= netdev_priv(netdev
);
1122 spin_lock_init(&np
->tx_lock
);
1123 spin_lock_init(&np
->rx_lock
);
1125 skb_queue_head_init(&np
->rx_batch
);
1126 np
->rx_target
= RX_DFL_MIN_TARGET
;
1127 np
->rx_min_target
= RX_DFL_MIN_TARGET
;
1128 np
->rx_max_target
= RX_MAX_TARGET
;
1130 init_timer(&np
->rx_refill_timer
);
1131 np
->rx_refill_timer
.data
= (unsigned long)netdev
;
1132 np
->rx_refill_timer
.function
= rx_refill_timeout
;
1134 /* Initialise tx_skbs as a free chain containing every entry. */
1135 np
->tx_skb_freelist
= 0;
1136 for (i
= 0; i
< NET_TX_RING_SIZE
; i
++) {
1137 skb_entry_set_link(&np
->tx_skbs
[i
], i
+1);
1138 np
->grant_tx_ref
[i
] = GRANT_INVALID_REF
;
1141 /* Clear out rx_skbs */
1142 for (i
= 0; i
< NET_RX_RING_SIZE
; i
++) {
1143 np
->rx_skbs
[i
] = NULL
;
1144 np
->grant_rx_ref
[i
] = GRANT_INVALID_REF
;
1147 /* A grant for every tx ring slot */
1148 if (gnttab_alloc_grant_references(TX_MAX_TARGET
,
1149 &np
->gref_tx_head
) < 0) {
1150 printk(KERN_ALERT
"#### netfront can't alloc tx grant refs\n");
1154 /* A grant for every rx ring slot */
1155 if (gnttab_alloc_grant_references(RX_MAX_TARGET
,
1156 &np
->gref_rx_head
) < 0) {
1157 printk(KERN_ALERT
"#### netfront can't alloc rx grant refs\n");
1162 netdev
->open
= xennet_open
;
1163 netdev
->hard_start_xmit
= xennet_start_xmit
;
1164 netdev
->stop
= xennet_close
;
1165 netif_napi_add(netdev
, &np
->napi
, xennet_poll
, 64);
1166 netdev
->uninit
= xennet_uninit
;
1167 netdev
->change_mtu
= xennet_change_mtu
;
1168 netdev
->features
= NETIF_F_IP_CSUM
;
1170 SET_ETHTOOL_OPS(netdev
, &xennet_ethtool_ops
);
1171 SET_NETDEV_DEV(netdev
, &dev
->dev
);
1173 np
->netdev
= netdev
;
1175 netif_carrier_off(netdev
);
1180 gnttab_free_grant_references(np
->gref_tx_head
);
1182 free_netdev(netdev
);
1183 return ERR_PTR(err
);
1187 * Entry point to this code when a new device is created. Allocate the basic
1188 * structures and the ring buffers for communication with the backend, and
1189 * inform the backend of the appropriate details for those.
1191 static int __devinit
netfront_probe(struct xenbus_device
*dev
,
1192 const struct xenbus_device_id
*id
)
1195 struct net_device
*netdev
;
1196 struct netfront_info
*info
;
1198 netdev
= xennet_create_dev(dev
);
1199 if (IS_ERR(netdev
)) {
1200 err
= PTR_ERR(netdev
);
1201 xenbus_dev_fatal(dev
, err
, "creating netdev");
1205 info
= netdev_priv(netdev
);
1206 dev
->dev
.driver_data
= info
;
1208 err
= register_netdev(info
->netdev
);
1210 printk(KERN_WARNING
"%s: register_netdev err=%d\n",
1215 err
= xennet_sysfs_addif(info
->netdev
);
1217 unregister_netdev(info
->netdev
);
1218 printk(KERN_WARNING
"%s: add sysfs failed err=%d\n",
1226 free_netdev(netdev
);
1227 dev
->dev
.driver_data
= NULL
;
1231 static void xennet_end_access(int ref
, void *page
)
1233 /* This frees the page as a side-effect */
1234 if (ref
!= GRANT_INVALID_REF
)
1235 gnttab_end_foreign_access(ref
, 0, (unsigned long)page
);
1238 static void xennet_disconnect_backend(struct netfront_info
*info
)
1240 /* Stop old i/f to prevent errors whilst we rebuild the state. */
1241 spin_lock_bh(&info
->rx_lock
);
1242 spin_lock_irq(&info
->tx_lock
);
1243 netif_carrier_off(info
->netdev
);
1244 spin_unlock_irq(&info
->tx_lock
);
1245 spin_unlock_bh(&info
->rx_lock
);
1247 if (info
->netdev
->irq
)
1248 unbind_from_irqhandler(info
->netdev
->irq
, info
->netdev
);
1249 info
->evtchn
= info
->netdev
->irq
= 0;
1251 /* End access and free the pages */
1252 xennet_end_access(info
->tx_ring_ref
, info
->tx
.sring
);
1253 xennet_end_access(info
->rx_ring_ref
, info
->rx
.sring
);
1255 info
->tx_ring_ref
= GRANT_INVALID_REF
;
1256 info
->rx_ring_ref
= GRANT_INVALID_REF
;
1257 info
->tx
.sring
= NULL
;
1258 info
->rx
.sring
= NULL
;
1262 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1263 * driver restart. We tear down our netif structure and recreate it, but
1264 * leave the device-layer structures intact so that this is transparent to the
1265 * rest of the kernel.
1267 static int netfront_resume(struct xenbus_device
*dev
)
1269 struct netfront_info
*info
= dev
->dev
.driver_data
;
1271 dev_dbg(&dev
->dev
, "%s\n", dev
->nodename
);
1273 xennet_disconnect_backend(info
);
1277 static int xen_net_read_mac(struct xenbus_device
*dev
, u8 mac
[])
1279 char *s
, *e
, *macstr
;
1282 macstr
= s
= xenbus_read(XBT_NIL
, dev
->nodename
, "mac", NULL
);
1284 return PTR_ERR(macstr
);
1286 for (i
= 0; i
< ETH_ALEN
; i
++) {
1287 mac
[i
] = simple_strtoul(s
, &e
, 16);
1288 if ((s
== e
) || (*e
!= ((i
== ETH_ALEN
-1) ? '\0' : ':'))) {
1299 static irqreturn_t
xennet_interrupt(int irq
, void *dev_id
)
1301 struct net_device
*dev
= dev_id
;
1302 struct netfront_info
*np
= netdev_priv(dev
);
1303 unsigned long flags
;
1305 spin_lock_irqsave(&np
->tx_lock
, flags
);
1307 if (likely(netif_carrier_ok(dev
))) {
1308 xennet_tx_buf_gc(dev
);
1309 /* Under tx_lock: protects access to rx shared-ring indexes. */
1310 if (RING_HAS_UNCONSUMED_RESPONSES(&np
->rx
))
1311 netif_rx_schedule(dev
, &np
->napi
);
1314 spin_unlock_irqrestore(&np
->tx_lock
, flags
);
1319 static int setup_netfront(struct xenbus_device
*dev
, struct netfront_info
*info
)
1321 struct xen_netif_tx_sring
*txs
;
1322 struct xen_netif_rx_sring
*rxs
;
1324 struct net_device
*netdev
= info
->netdev
;
1326 info
->tx_ring_ref
= GRANT_INVALID_REF
;
1327 info
->rx_ring_ref
= GRANT_INVALID_REF
;
1328 info
->rx
.sring
= NULL
;
1329 info
->tx
.sring
= NULL
;
1332 err
= xen_net_read_mac(dev
, netdev
->dev_addr
);
1334 xenbus_dev_fatal(dev
, err
, "parsing %s/mac", dev
->nodename
);
1338 txs
= (struct xen_netif_tx_sring
*)get_zeroed_page(GFP_NOIO
| __GFP_HIGH
);
1341 xenbus_dev_fatal(dev
, err
, "allocating tx ring page");
1344 SHARED_RING_INIT(txs
);
1345 FRONT_RING_INIT(&info
->tx
, txs
, PAGE_SIZE
);
1347 err
= xenbus_grant_ring(dev
, virt_to_mfn(txs
));
1349 free_page((unsigned long)txs
);
1353 info
->tx_ring_ref
= err
;
1354 rxs
= (struct xen_netif_rx_sring
*)get_zeroed_page(GFP_NOIO
| __GFP_HIGH
);
1357 xenbus_dev_fatal(dev
, err
, "allocating rx ring page");
1360 SHARED_RING_INIT(rxs
);
1361 FRONT_RING_INIT(&info
->rx
, rxs
, PAGE_SIZE
);
1363 err
= xenbus_grant_ring(dev
, virt_to_mfn(rxs
));
1365 free_page((unsigned long)rxs
);
1368 info
->rx_ring_ref
= err
;
1370 err
= xenbus_alloc_evtchn(dev
, &info
->evtchn
);
1374 err
= bind_evtchn_to_irqhandler(info
->evtchn
, xennet_interrupt
,
1375 IRQF_SAMPLE_RANDOM
, netdev
->name
,
1386 /* Common code used when first setting up, and when resuming. */
1387 static int talk_to_backend(struct xenbus_device
*dev
,
1388 struct netfront_info
*info
)
1390 const char *message
;
1391 struct xenbus_transaction xbt
;
1394 /* Create shared ring, alloc event channel. */
1395 err
= setup_netfront(dev
, info
);
1400 err
= xenbus_transaction_start(&xbt
);
1402 xenbus_dev_fatal(dev
, err
, "starting transaction");
1406 err
= xenbus_printf(xbt
, dev
->nodename
, "tx-ring-ref", "%u",
1409 message
= "writing tx ring-ref";
1410 goto abort_transaction
;
1412 err
= xenbus_printf(xbt
, dev
->nodename
, "rx-ring-ref", "%u",
1415 message
= "writing rx ring-ref";
1416 goto abort_transaction
;
1418 err
= xenbus_printf(xbt
, dev
->nodename
,
1419 "event-channel", "%u", info
->evtchn
);
1421 message
= "writing event-channel";
1422 goto abort_transaction
;
1425 err
= xenbus_printf(xbt
, dev
->nodename
, "request-rx-copy", "%u",
1428 message
= "writing request-rx-copy";
1429 goto abort_transaction
;
1432 err
= xenbus_printf(xbt
, dev
->nodename
, "feature-rx-notify", "%d", 1);
1434 message
= "writing feature-rx-notify";
1435 goto abort_transaction
;
1438 err
= xenbus_printf(xbt
, dev
->nodename
, "feature-sg", "%d", 1);
1440 message
= "writing feature-sg";
1441 goto abort_transaction
;
1444 err
= xenbus_printf(xbt
, dev
->nodename
, "feature-gso-tcpv4", "%d", 1);
1446 message
= "writing feature-gso-tcpv4";
1447 goto abort_transaction
;
1450 err
= xenbus_transaction_end(xbt
, 0);
1454 xenbus_dev_fatal(dev
, err
, "completing transaction");
1461 xenbus_transaction_end(xbt
, 1);
1462 xenbus_dev_fatal(dev
, err
, "%s", message
);
1464 xennet_disconnect_backend(info
);
1469 static int xennet_set_sg(struct net_device
*dev
, u32 data
)
1472 struct netfront_info
*np
= netdev_priv(dev
);
1475 if (xenbus_scanf(XBT_NIL
, np
->xbdev
->otherend
, "feature-sg",
1480 } else if (dev
->mtu
> ETH_DATA_LEN
)
1481 dev
->mtu
= ETH_DATA_LEN
;
1483 return ethtool_op_set_sg(dev
, data
);
1486 static int xennet_set_tso(struct net_device
*dev
, u32 data
)
1489 struct netfront_info
*np
= netdev_priv(dev
);
1492 if (xenbus_scanf(XBT_NIL
, np
->xbdev
->otherend
,
1493 "feature-gso-tcpv4", "%d", &val
) < 0)
1499 return ethtool_op_set_tso(dev
, data
);
1502 static void xennet_set_features(struct net_device
*dev
)
1504 /* Turn off all GSO bits except ROBUST. */
1505 dev
->features
&= (1 << NETIF_F_GSO_SHIFT
) - 1;
1506 dev
->features
|= NETIF_F_GSO_ROBUST
;
1507 xennet_set_sg(dev
, 0);
1509 /* We need checksum offload to enable scatter/gather and TSO. */
1510 if (!(dev
->features
& NETIF_F_IP_CSUM
))
1513 if (!xennet_set_sg(dev
, 1))
1514 xennet_set_tso(dev
, 1);
1517 static int xennet_connect(struct net_device
*dev
)
1519 struct netfront_info
*np
= netdev_priv(dev
);
1520 int i
, requeue_idx
, err
;
1521 struct sk_buff
*skb
;
1523 struct xen_netif_rx_request
*req
;
1524 unsigned int feature_rx_copy
;
1526 err
= xenbus_scanf(XBT_NIL
, np
->xbdev
->otherend
,
1527 "feature-rx-copy", "%u", &feature_rx_copy
);
1529 feature_rx_copy
= 0;
1531 if (!feature_rx_copy
) {
1533 "backend does not support copying receive path\n");
1537 err
= talk_to_backend(np
->xbdev
, np
);
1541 xennet_set_features(dev
);
1543 spin_lock_bh(&np
->rx_lock
);
1544 spin_lock_irq(&np
->tx_lock
);
1546 /* Step 1: Discard all pending TX packet fragments. */
1547 xennet_release_tx_bufs(np
);
1549 /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1550 for (requeue_idx
= 0, i
= 0; i
< NET_RX_RING_SIZE
; i
++) {
1551 if (!np
->rx_skbs
[i
])
1554 skb
= np
->rx_skbs
[requeue_idx
] = xennet_get_rx_skb(np
, i
);
1555 ref
= np
->grant_rx_ref
[requeue_idx
] = xennet_get_rx_ref(np
, i
);
1556 req
= RING_GET_REQUEST(&np
->rx
, requeue_idx
);
1558 gnttab_grant_foreign_access_ref(
1559 ref
, np
->xbdev
->otherend_id
,
1560 pfn_to_mfn(page_to_pfn(skb_shinfo(skb
)->
1564 req
->id
= requeue_idx
;
1569 np
->rx
.req_prod_pvt
= requeue_idx
;
1572 * Step 3: All public and private state should now be sane. Get
1573 * ready to start sending and receiving packets and give the driver
1574 * domain a kick because we've probably just requeued some
1577 netif_carrier_on(np
->netdev
);
1578 notify_remote_via_irq(np
->netdev
->irq
);
1579 xennet_tx_buf_gc(dev
);
1580 xennet_alloc_rx_buffers(dev
);
1582 spin_unlock_irq(&np
->tx_lock
);
1583 spin_unlock_bh(&np
->rx_lock
);
1589 * Callback received when the backend's state changes.
1591 static void backend_changed(struct xenbus_device
*dev
,
1592 enum xenbus_state backend_state
)
1594 struct netfront_info
*np
= dev
->dev
.driver_data
;
1595 struct net_device
*netdev
= np
->netdev
;
1597 dev_dbg(&dev
->dev
, "%s\n", xenbus_strstate(backend_state
));
1599 switch (backend_state
) {
1600 case XenbusStateInitialising
:
1601 case XenbusStateInitialised
:
1602 case XenbusStateConnected
:
1603 case XenbusStateUnknown
:
1604 case XenbusStateClosed
:
1607 case XenbusStateInitWait
:
1608 if (dev
->state
!= XenbusStateInitialising
)
1610 if (xennet_connect(netdev
) != 0)
1612 xenbus_switch_state(dev
, XenbusStateConnected
);
1615 case XenbusStateClosing
:
1616 xenbus_frontend_closed(dev
);
1621 static struct ethtool_ops xennet_ethtool_ops
=
1623 .set_tx_csum
= ethtool_op_set_tx_csum
,
1624 .set_sg
= xennet_set_sg
,
1625 .set_tso
= xennet_set_tso
,
1626 .get_link
= ethtool_op_get_link
,
1630 static ssize_t
show_rxbuf_min(struct device
*dev
,
1631 struct device_attribute
*attr
, char *buf
)
1633 struct net_device
*netdev
= to_net_dev(dev
);
1634 struct netfront_info
*info
= netdev_priv(netdev
);
1636 return sprintf(buf
, "%u\n", info
->rx_min_target
);
1639 static ssize_t
store_rxbuf_min(struct device
*dev
,
1640 struct device_attribute
*attr
,
1641 const char *buf
, size_t len
)
1643 struct net_device
*netdev
= to_net_dev(dev
);
1644 struct netfront_info
*np
= netdev_priv(netdev
);
1646 unsigned long target
;
1648 if (!capable(CAP_NET_ADMIN
))
1651 target
= simple_strtoul(buf
, &endp
, 0);
1655 if (target
< RX_MIN_TARGET
)
1656 target
= RX_MIN_TARGET
;
1657 if (target
> RX_MAX_TARGET
)
1658 target
= RX_MAX_TARGET
;
1660 spin_lock_bh(&np
->rx_lock
);
1661 if (target
> np
->rx_max_target
)
1662 np
->rx_max_target
= target
;
1663 np
->rx_min_target
= target
;
1664 if (target
> np
->rx_target
)
1665 np
->rx_target
= target
;
1667 xennet_alloc_rx_buffers(netdev
);
1669 spin_unlock_bh(&np
->rx_lock
);
1673 static ssize_t
show_rxbuf_max(struct device
*dev
,
1674 struct device_attribute
*attr
, char *buf
)
1676 struct net_device
*netdev
= to_net_dev(dev
);
1677 struct netfront_info
*info
= netdev_priv(netdev
);
1679 return sprintf(buf
, "%u\n", info
->rx_max_target
);
1682 static ssize_t
store_rxbuf_max(struct device
*dev
,
1683 struct device_attribute
*attr
,
1684 const char *buf
, size_t len
)
1686 struct net_device
*netdev
= to_net_dev(dev
);
1687 struct netfront_info
*np
= netdev_priv(netdev
);
1689 unsigned long target
;
1691 if (!capable(CAP_NET_ADMIN
))
1694 target
= simple_strtoul(buf
, &endp
, 0);
1698 if (target
< RX_MIN_TARGET
)
1699 target
= RX_MIN_TARGET
;
1700 if (target
> RX_MAX_TARGET
)
1701 target
= RX_MAX_TARGET
;
1703 spin_lock_bh(&np
->rx_lock
);
1704 if (target
< np
->rx_min_target
)
1705 np
->rx_min_target
= target
;
1706 np
->rx_max_target
= target
;
1707 if (target
< np
->rx_target
)
1708 np
->rx_target
= target
;
1710 xennet_alloc_rx_buffers(netdev
);
1712 spin_unlock_bh(&np
->rx_lock
);
1716 static ssize_t
show_rxbuf_cur(struct device
*dev
,
1717 struct device_attribute
*attr
, char *buf
)
1719 struct net_device
*netdev
= to_net_dev(dev
);
1720 struct netfront_info
*info
= netdev_priv(netdev
);
1722 return sprintf(buf
, "%u\n", info
->rx_target
);
1725 static struct device_attribute xennet_attrs
[] = {
1726 __ATTR(rxbuf_min
, S_IRUGO
|S_IWUSR
, show_rxbuf_min
, store_rxbuf_min
),
1727 __ATTR(rxbuf_max
, S_IRUGO
|S_IWUSR
, show_rxbuf_max
, store_rxbuf_max
),
1728 __ATTR(rxbuf_cur
, S_IRUGO
, show_rxbuf_cur
, NULL
),
1731 static int xennet_sysfs_addif(struct net_device
*netdev
)
1736 for (i
= 0; i
< ARRAY_SIZE(xennet_attrs
); i
++) {
1737 err
= device_create_file(&netdev
->dev
,
1746 device_remove_file(&netdev
->dev
, &xennet_attrs
[i
]);
1750 static void xennet_sysfs_delif(struct net_device
*netdev
)
1754 for (i
= 0; i
< ARRAY_SIZE(xennet_attrs
); i
++)
1755 device_remove_file(&netdev
->dev
, &xennet_attrs
[i
]);
1758 #endif /* CONFIG_SYSFS */
1760 static struct xenbus_device_id netfront_ids
[] = {
1766 static int __devexit
xennet_remove(struct xenbus_device
*dev
)
1768 struct netfront_info
*info
= dev
->dev
.driver_data
;
1770 dev_dbg(&dev
->dev
, "%s\n", dev
->nodename
);
1772 unregister_netdev(info
->netdev
);
1774 xennet_disconnect_backend(info
);
1776 del_timer_sync(&info
->rx_refill_timer
);
1778 xennet_sysfs_delif(info
->netdev
);
1780 free_netdev(info
->netdev
);
1785 static struct xenbus_driver netfront
= {
1787 .owner
= THIS_MODULE
,
1788 .ids
= netfront_ids
,
1789 .probe
= netfront_probe
,
1790 .remove
= __devexit_p(xennet_remove
),
1791 .resume
= netfront_resume
,
1792 .otherend_changed
= backend_changed
,
1795 static int __init
netif_init(void)
1797 if (!is_running_on_xen())
1800 if (is_initial_xendomain())
1803 printk(KERN_INFO
"Initialising Xen virtual ethernet driver.\n");
1805 return xenbus_register_frontend(&netfront
);
1807 module_init(netif_init
);
1810 static void __exit
netif_exit(void)
1812 if (is_initial_xendomain())
1815 xenbus_unregister_driver(&netfront
);
1817 module_exit(netif_exit
);
1819 MODULE_DESCRIPTION("Xen virtual network device frontend");
1820 MODULE_LICENSE("GPL");
1821 MODULE_ALIAS("xen:vif");
1822 MODULE_ALIAS("xennet");