2 * Virtual network driver for conversing with remote driver backends.
4 * Copyright (c) 2002-2005, K A Fraser
5 * Copyright (c) 2005, XenSource Ltd
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License version 2
9 * as published by the Free Software Foundation; or, when distributed
10 * separately from the Linux kernel or incorporated into other
11 * software packages, subject to the following license:
13 * Permission is hereby granted, free of charge, to any person obtaining a copy
14 * of this source file (the "Software"), to deal in the Software without
15 * restriction, including without limitation the rights to use, copy, modify,
16 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
17 * and to permit persons to whom the Software is furnished to do so, subject to
18 * the following conditions:
20 * The above copyright notice and this permission notice shall be included in
21 * all copies or substantial portions of the Software.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
24 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
25 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
26 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
27 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
28 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
32 #include <linux/module.h>
33 #include <linux/kernel.h>
34 #include <linux/netdevice.h>
35 #include <linux/etherdevice.h>
36 #include <linux/skbuff.h>
37 #include <linux/ethtool.h>
38 #include <linux/if_ether.h>
40 #include <linux/udp.h>
41 #include <linux/moduleparam.h>
43 #include <linux/slab.h>
46 #include <asm/xen/page.h>
48 #include <xen/xenbus.h>
49 #include <xen/events.h>
51 #include <xen/platform_pci.h>
52 #include <xen/grant_table.h>
54 #include <xen/interface/io/netif.h>
55 #include <xen/interface/memory.h>
56 #include <xen/interface/grant_table.h>
58 static const struct ethtool_ops xennet_ethtool_ops
;
64 #define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb))
66 #define RX_COPY_THRESHOLD 256
68 #define GRANT_INVALID_REF 0
70 #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, PAGE_SIZE)
71 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, PAGE_SIZE)
72 #define TX_MAX_TARGET min_t(int, NET_TX_RING_SIZE, 256)
74 struct netfront_stats
{
79 struct u64_stats_sync syncp
;
82 struct netfront_info
{
83 struct list_head list
;
84 struct net_device
*netdev
;
86 struct napi_struct napi
;
89 struct xenbus_device
*xbdev
;
92 struct xen_netif_tx_front_ring tx
;
96 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
97 * are linked from tx_skb_freelist through skb_entry.link.
99 * NB. Freelist index entries are always going to be less than
100 * PAGE_OFFSET, whereas pointers to skbs will always be equal or
101 * greater than PAGE_OFFSET: we use this property to distinguish
107 } tx_skbs
[NET_TX_RING_SIZE
];
108 grant_ref_t gref_tx_head
;
109 grant_ref_t grant_tx_ref
[NET_TX_RING_SIZE
];
110 struct page
*grant_tx_page
[NET_TX_RING_SIZE
];
111 unsigned tx_skb_freelist
;
113 spinlock_t rx_lock ____cacheline_aligned_in_smp
;
114 struct xen_netif_rx_front_ring rx
;
117 /* Receive-ring batched refills. */
118 #define RX_MIN_TARGET 8
119 #define RX_DFL_MIN_TARGET 64
120 #define RX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256)
121 unsigned rx_min_target
, rx_max_target
, rx_target
;
122 struct sk_buff_head rx_batch
;
124 struct timer_list rx_refill_timer
;
126 struct sk_buff
*rx_skbs
[NET_RX_RING_SIZE
];
127 grant_ref_t gref_rx_head
;
128 grant_ref_t grant_rx_ref
[NET_RX_RING_SIZE
];
130 unsigned long rx_pfn_array
[NET_RX_RING_SIZE
];
131 struct multicall_entry rx_mcl
[NET_RX_RING_SIZE
+1];
132 struct mmu_update rx_mmu
[NET_RX_RING_SIZE
];
135 struct netfront_stats __percpu
*stats
;
137 unsigned long rx_gso_checksum_fixup
;
140 struct netfront_rx_info
{
141 struct xen_netif_rx_response rx
;
142 struct xen_netif_extra_info extras
[XEN_NETIF_EXTRA_TYPE_MAX
- 1];
145 static void skb_entry_set_link(union skb_entry
*list
, unsigned short id
)
150 static int skb_entry_is_link(const union skb_entry
*list
)
152 BUILD_BUG_ON(sizeof(list
->skb
) != sizeof(list
->link
));
153 return (unsigned long)list
->skb
< PAGE_OFFSET
;
157 * Access macros for acquiring freeing slots in tx_skbs[].
160 static void add_id_to_freelist(unsigned *head
, union skb_entry
*list
,
163 skb_entry_set_link(&list
[id
], *head
);
167 static unsigned short get_id_from_freelist(unsigned *head
,
168 union skb_entry
*list
)
170 unsigned int id
= *head
;
171 *head
= list
[id
].link
;
175 static int xennet_rxidx(RING_IDX idx
)
177 return idx
& (NET_RX_RING_SIZE
- 1);
180 static struct sk_buff
*xennet_get_rx_skb(struct netfront_info
*np
,
183 int i
= xennet_rxidx(ri
);
184 struct sk_buff
*skb
= np
->rx_skbs
[i
];
185 np
->rx_skbs
[i
] = NULL
;
189 static grant_ref_t
xennet_get_rx_ref(struct netfront_info
*np
,
192 int i
= xennet_rxidx(ri
);
193 grant_ref_t ref
= np
->grant_rx_ref
[i
];
194 np
->grant_rx_ref
[i
] = GRANT_INVALID_REF
;
199 static int xennet_sysfs_addif(struct net_device
*netdev
);
200 static void xennet_sysfs_delif(struct net_device
*netdev
);
201 #else /* !CONFIG_SYSFS */
202 #define xennet_sysfs_addif(dev) (0)
203 #define xennet_sysfs_delif(dev) do { } while (0)
206 static bool xennet_can_sg(struct net_device
*dev
)
208 return dev
->features
& NETIF_F_SG
;
212 static void rx_refill_timeout(unsigned long data
)
214 struct net_device
*dev
= (struct net_device
*)data
;
215 struct netfront_info
*np
= netdev_priv(dev
);
216 napi_schedule(&np
->napi
);
219 static int netfront_tx_slot_available(struct netfront_info
*np
)
221 return (np
->tx
.req_prod_pvt
- np
->tx
.rsp_cons
) <
222 (TX_MAX_TARGET
- MAX_SKB_FRAGS
- 2);
225 static void xennet_maybe_wake_tx(struct net_device
*dev
)
227 struct netfront_info
*np
= netdev_priv(dev
);
229 if (unlikely(netif_queue_stopped(dev
)) &&
230 netfront_tx_slot_available(np
) &&
231 likely(netif_running(dev
)))
232 netif_wake_queue(dev
);
235 static void xennet_alloc_rx_buffers(struct net_device
*dev
)
238 struct netfront_info
*np
= netdev_priv(dev
);
241 int i
, batch_target
, notify
;
242 RING_IDX req_prod
= np
->rx
.req_prod_pvt
;
246 struct xen_netif_rx_request
*req
;
248 if (unlikely(!netif_carrier_ok(dev
)))
252 * Allocate skbuffs greedily, even though we batch updates to the
253 * receive ring. This creates a less bursty demand on the memory
254 * allocator, so should reduce the chance of failed allocation requests
255 * both for ourself and for other kernel subsystems.
257 batch_target
= np
->rx_target
- (req_prod
- np
->rx
.rsp_cons
);
258 for (i
= skb_queue_len(&np
->rx_batch
); i
< batch_target
; i
++) {
259 skb
= __netdev_alloc_skb(dev
, RX_COPY_THRESHOLD
+ NET_IP_ALIGN
,
260 GFP_ATOMIC
| __GFP_NOWARN
);
264 /* Align ip header to a 16 bytes boundary */
265 skb_reserve(skb
, NET_IP_ALIGN
);
267 page
= alloc_page(GFP_ATOMIC
| __GFP_NOWARN
);
271 /* Any skbuffs queued for refill? Force them out. */
274 /* Could not allocate any skbuffs. Try again later. */
275 mod_timer(&np
->rx_refill_timer
,
280 skb_add_rx_frag(skb
, 0, page
, 0, 0, PAGE_SIZE
);
281 __skb_queue_tail(&np
->rx_batch
, skb
);
284 /* Is the batch large enough to be worthwhile? */
285 if (i
< (np
->rx_target
/2)) {
286 if (req_prod
> np
->rx
.sring
->req_prod
)
291 /* Adjust our fill target if we risked running out of buffers. */
292 if (((req_prod
- np
->rx
.sring
->rsp_prod
) < (np
->rx_target
/ 4)) &&
293 ((np
->rx_target
*= 2) > np
->rx_max_target
))
294 np
->rx_target
= np
->rx_max_target
;
298 skb
= __skb_dequeue(&np
->rx_batch
);
304 id
= xennet_rxidx(req_prod
+ i
);
306 BUG_ON(np
->rx_skbs
[id
]);
307 np
->rx_skbs
[id
] = skb
;
309 ref
= gnttab_claim_grant_reference(&np
->gref_rx_head
);
310 BUG_ON((signed short)ref
< 0);
311 np
->grant_rx_ref
[id
] = ref
;
313 pfn
= page_to_pfn(skb_frag_page(&skb_shinfo(skb
)->frags
[0]));
314 vaddr
= page_address(skb_frag_page(&skb_shinfo(skb
)->frags
[0]));
316 req
= RING_GET_REQUEST(&np
->rx
, req_prod
+ i
);
317 gnttab_grant_foreign_access_ref(ref
,
318 np
->xbdev
->otherend_id
,
326 wmb(); /* barrier so backend seens requests */
328 /* Above is a suitable barrier to ensure backend will see requests. */
329 np
->rx
.req_prod_pvt
= req_prod
+ i
;
331 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np
->rx
, notify
);
333 notify_remote_via_irq(np
->netdev
->irq
);
336 static int xennet_open(struct net_device
*dev
)
338 struct netfront_info
*np
= netdev_priv(dev
);
340 napi_enable(&np
->napi
);
342 spin_lock_bh(&np
->rx_lock
);
343 if (netif_carrier_ok(dev
)) {
344 xennet_alloc_rx_buffers(dev
);
345 np
->rx
.sring
->rsp_event
= np
->rx
.rsp_cons
+ 1;
346 if (RING_HAS_UNCONSUMED_RESPONSES(&np
->rx
))
347 napi_schedule(&np
->napi
);
349 spin_unlock_bh(&np
->rx_lock
);
351 netif_start_queue(dev
);
356 static void xennet_tx_buf_gc(struct net_device
*dev
)
360 struct netfront_info
*np
= netdev_priv(dev
);
363 BUG_ON(!netif_carrier_ok(dev
));
366 prod
= np
->tx
.sring
->rsp_prod
;
367 rmb(); /* Ensure we see responses up to 'rp'. */
369 for (cons
= np
->tx
.rsp_cons
; cons
!= prod
; cons
++) {
370 struct xen_netif_tx_response
*txrsp
;
372 txrsp
= RING_GET_RESPONSE(&np
->tx
, cons
);
373 if (txrsp
->status
== XEN_NETIF_RSP_NULL
)
377 skb
= np
->tx_skbs
[id
].skb
;
378 if (unlikely(gnttab_query_foreign_access(
379 np
->grant_tx_ref
[id
]) != 0)) {
380 printk(KERN_ALERT
"xennet_tx_buf_gc: warning "
381 "-- grant still in use by backend "
385 gnttab_end_foreign_access_ref(
386 np
->grant_tx_ref
[id
], GNTMAP_readonly
);
387 gnttab_release_grant_reference(
388 &np
->gref_tx_head
, np
->grant_tx_ref
[id
]);
389 np
->grant_tx_ref
[id
] = GRANT_INVALID_REF
;
390 np
->grant_tx_page
[id
] = NULL
;
391 add_id_to_freelist(&np
->tx_skb_freelist
, np
->tx_skbs
, id
);
392 dev_kfree_skb_irq(skb
);
395 np
->tx
.rsp_cons
= prod
;
398 * Set a new event, then check for race with update of tx_cons.
399 * Note that it is essential to schedule a callback, no matter
400 * how few buffers are pending. Even if there is space in the
401 * transmit ring, higher layers may be blocked because too much
402 * data is outstanding: in such cases notification from Xen is
403 * likely to be the only kick that we'll get.
405 np
->tx
.sring
->rsp_event
=
406 prod
+ ((np
->tx
.sring
->req_prod
- prod
) >> 1) + 1;
407 mb(); /* update shared area */
408 } while ((cons
== prod
) && (prod
!= np
->tx
.sring
->rsp_prod
));
410 xennet_maybe_wake_tx(dev
);
413 static void xennet_make_frags(struct sk_buff
*skb
, struct net_device
*dev
,
414 struct xen_netif_tx_request
*tx
)
416 struct netfront_info
*np
= netdev_priv(dev
);
417 char *data
= skb
->data
;
419 RING_IDX prod
= np
->tx
.req_prod_pvt
;
420 int frags
= skb_shinfo(skb
)->nr_frags
;
421 unsigned int offset
= offset_in_page(data
);
422 unsigned int len
= skb_headlen(skb
);
427 /* While the header overlaps a page boundary (including being
428 larger than a page), split it it into page-sized chunks. */
429 while (len
> PAGE_SIZE
- offset
) {
430 tx
->size
= PAGE_SIZE
- offset
;
431 tx
->flags
|= XEN_NETTXF_more_data
;
436 id
= get_id_from_freelist(&np
->tx_skb_freelist
, np
->tx_skbs
);
437 np
->tx_skbs
[id
].skb
= skb_get(skb
);
438 tx
= RING_GET_REQUEST(&np
->tx
, prod
++);
440 ref
= gnttab_claim_grant_reference(&np
->gref_tx_head
);
441 BUG_ON((signed short)ref
< 0);
443 mfn
= virt_to_mfn(data
);
444 gnttab_grant_foreign_access_ref(ref
, np
->xbdev
->otherend_id
,
445 mfn
, GNTMAP_readonly
);
447 np
->grant_tx_page
[id
] = virt_to_page(data
);
448 tx
->gref
= np
->grant_tx_ref
[id
] = ref
;
454 /* Grant backend access to each skb fragment page. */
455 for (i
= 0; i
< frags
; i
++) {
456 skb_frag_t
*frag
= skb_shinfo(skb
)->frags
+ i
;
457 struct page
*page
= skb_frag_page(frag
);
459 len
= skb_frag_size(frag
);
460 offset
= frag
->page_offset
;
462 /* Skip unused frames from start of page */
463 page
+= offset
>> PAGE_SHIFT
;
464 offset
&= ~PAGE_MASK
;
469 bytes
= PAGE_SIZE
- offset
;
473 tx
->flags
|= XEN_NETTXF_more_data
;
475 id
= get_id_from_freelist(&np
->tx_skb_freelist
,
477 np
->tx_skbs
[id
].skb
= skb_get(skb
);
478 tx
= RING_GET_REQUEST(&np
->tx
, prod
++);
480 ref
= gnttab_claim_grant_reference(&np
->gref_tx_head
);
481 BUG_ON((signed short)ref
< 0);
483 mfn
= pfn_to_mfn(page_to_pfn(page
));
484 gnttab_grant_foreign_access_ref(ref
,
485 np
->xbdev
->otherend_id
,
486 mfn
, GNTMAP_readonly
);
488 np
->grant_tx_page
[id
] = page
;
489 tx
->gref
= np
->grant_tx_ref
[id
] = ref
;
498 if (offset
== PAGE_SIZE
&& len
) {
499 BUG_ON(!PageCompound(page
));
506 np
->tx
.req_prod_pvt
= prod
;
510 * Count how many ring slots are required to send the frags of this
511 * skb. Each frag might be a compound page.
513 static int xennet_count_skb_frag_slots(struct sk_buff
*skb
)
515 int i
, frags
= skb_shinfo(skb
)->nr_frags
;
518 for (i
= 0; i
< frags
; i
++) {
519 skb_frag_t
*frag
= skb_shinfo(skb
)->frags
+ i
;
520 unsigned long size
= skb_frag_size(frag
);
521 unsigned long offset
= frag
->page_offset
;
523 /* Skip unused frames from start of page */
524 offset
&= ~PAGE_MASK
;
526 pages
+= PFN_UP(offset
+ size
);
532 static int xennet_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
535 struct netfront_info
*np
= netdev_priv(dev
);
536 struct netfront_stats
*stats
= this_cpu_ptr(np
->stats
);
537 struct xen_netif_tx_request
*tx
;
538 char *data
= skb
->data
;
544 unsigned int offset
= offset_in_page(data
);
545 unsigned int len
= skb_headlen(skb
);
548 /* If skb->len is too big for wire format, drop skb and alert
549 * user about misconfiguration.
551 if (unlikely(skb
->len
> XEN_NETIF_MAX_TX_SIZE
)) {
552 net_alert_ratelimited(
553 "xennet: skb->len = %u, too big for wire format\n",
558 slots
= DIV_ROUND_UP(offset
+ len
, PAGE_SIZE
) +
559 xennet_count_skb_frag_slots(skb
);
560 if (unlikely(slots
> MAX_SKB_FRAGS
+ 1)) {
561 net_alert_ratelimited(
562 "xennet: skb rides the rocket: %d slots\n", slots
);
566 spin_lock_irqsave(&np
->tx_lock
, flags
);
568 if (unlikely(!netif_carrier_ok(dev
) ||
569 (slots
> 1 && !xennet_can_sg(dev
)) ||
570 netif_needs_gso(skb
, netif_skb_features(skb
)))) {
571 spin_unlock_irqrestore(&np
->tx_lock
, flags
);
575 i
= np
->tx
.req_prod_pvt
;
577 id
= get_id_from_freelist(&np
->tx_skb_freelist
, np
->tx_skbs
);
578 np
->tx_skbs
[id
].skb
= skb
;
580 tx
= RING_GET_REQUEST(&np
->tx
, i
);
583 ref
= gnttab_claim_grant_reference(&np
->gref_tx_head
);
584 BUG_ON((signed short)ref
< 0);
585 mfn
= virt_to_mfn(data
);
586 gnttab_grant_foreign_access_ref(
587 ref
, np
->xbdev
->otherend_id
, mfn
, GNTMAP_readonly
);
588 np
->grant_tx_page
[id
] = virt_to_page(data
);
589 tx
->gref
= np
->grant_tx_ref
[id
] = ref
;
594 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
596 tx
->flags
|= XEN_NETTXF_csum_blank
| XEN_NETTXF_data_validated
;
597 else if (skb
->ip_summed
== CHECKSUM_UNNECESSARY
)
598 /* remote but checksummed. */
599 tx
->flags
|= XEN_NETTXF_data_validated
;
601 if (skb_shinfo(skb
)->gso_size
) {
602 struct xen_netif_extra_info
*gso
;
604 gso
= (struct xen_netif_extra_info
*)
605 RING_GET_REQUEST(&np
->tx
, ++i
);
607 tx
->flags
|= XEN_NETTXF_extra_info
;
609 gso
->u
.gso
.size
= skb_shinfo(skb
)->gso_size
;
610 gso
->u
.gso
.type
= XEN_NETIF_GSO_TYPE_TCPV4
;
612 gso
->u
.gso
.features
= 0;
614 gso
->type
= XEN_NETIF_EXTRA_TYPE_GSO
;
618 np
->tx
.req_prod_pvt
= i
+ 1;
620 xennet_make_frags(skb
, dev
, tx
);
623 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np
->tx
, notify
);
625 notify_remote_via_irq(np
->netdev
->irq
);
627 u64_stats_update_begin(&stats
->syncp
);
628 stats
->tx_bytes
+= skb
->len
;
630 u64_stats_update_end(&stats
->syncp
);
632 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
633 xennet_tx_buf_gc(dev
);
635 if (!netfront_tx_slot_available(np
))
636 netif_stop_queue(dev
);
638 spin_unlock_irqrestore(&np
->tx_lock
, flags
);
643 dev
->stats
.tx_dropped
++;
648 static int xennet_close(struct net_device
*dev
)
650 struct netfront_info
*np
= netdev_priv(dev
);
651 netif_stop_queue(np
->netdev
);
652 napi_disable(&np
->napi
);
656 static void xennet_move_rx_slot(struct netfront_info
*np
, struct sk_buff
*skb
,
659 int new = xennet_rxidx(np
->rx
.req_prod_pvt
);
661 BUG_ON(np
->rx_skbs
[new]);
662 np
->rx_skbs
[new] = skb
;
663 np
->grant_rx_ref
[new] = ref
;
664 RING_GET_REQUEST(&np
->rx
, np
->rx
.req_prod_pvt
)->id
= new;
665 RING_GET_REQUEST(&np
->rx
, np
->rx
.req_prod_pvt
)->gref
= ref
;
666 np
->rx
.req_prod_pvt
++;
669 static int xennet_get_extras(struct netfront_info
*np
,
670 struct xen_netif_extra_info
*extras
,
674 struct xen_netif_extra_info
*extra
;
675 struct device
*dev
= &np
->netdev
->dev
;
676 RING_IDX cons
= np
->rx
.rsp_cons
;
683 if (unlikely(cons
+ 1 == rp
)) {
685 dev_warn(dev
, "Missing extra info\n");
690 extra
= (struct xen_netif_extra_info
*)
691 RING_GET_RESPONSE(&np
->rx
, ++cons
);
693 if (unlikely(!extra
->type
||
694 extra
->type
>= XEN_NETIF_EXTRA_TYPE_MAX
)) {
696 dev_warn(dev
, "Invalid extra type: %d\n",
700 memcpy(&extras
[extra
->type
- 1], extra
,
704 skb
= xennet_get_rx_skb(np
, cons
);
705 ref
= xennet_get_rx_ref(np
, cons
);
706 xennet_move_rx_slot(np
, skb
, ref
);
707 } while (extra
->flags
& XEN_NETIF_EXTRA_FLAG_MORE
);
709 np
->rx
.rsp_cons
= cons
;
713 static int xennet_get_responses(struct netfront_info
*np
,
714 struct netfront_rx_info
*rinfo
, RING_IDX rp
,
715 struct sk_buff_head
*list
)
717 struct xen_netif_rx_response
*rx
= &rinfo
->rx
;
718 struct xen_netif_extra_info
*extras
= rinfo
->extras
;
719 struct device
*dev
= &np
->netdev
->dev
;
720 RING_IDX cons
= np
->rx
.rsp_cons
;
721 struct sk_buff
*skb
= xennet_get_rx_skb(np
, cons
);
722 grant_ref_t ref
= xennet_get_rx_ref(np
, cons
);
723 int max
= MAX_SKB_FRAGS
+ (rx
->status
<= RX_COPY_THRESHOLD
);
728 if (rx
->flags
& XEN_NETRXF_extra_info
) {
729 err
= xennet_get_extras(np
, extras
, rp
);
730 cons
= np
->rx
.rsp_cons
;
734 if (unlikely(rx
->status
< 0 ||
735 rx
->offset
+ rx
->status
> PAGE_SIZE
)) {
737 dev_warn(dev
, "rx->offset: %x, size: %u\n",
738 rx
->offset
, rx
->status
);
739 xennet_move_rx_slot(np
, skb
, ref
);
745 * This definitely indicates a bug, either in this driver or in
746 * the backend driver. In future this should flag the bad
747 * situation to the system controller to reboot the backend.
749 if (ref
== GRANT_INVALID_REF
) {
751 dev_warn(dev
, "Bad rx response id %d.\n",
757 ret
= gnttab_end_foreign_access_ref(ref
, 0);
760 gnttab_release_grant_reference(&np
->gref_rx_head
, ref
);
762 __skb_queue_tail(list
, skb
);
765 if (!(rx
->flags
& XEN_NETRXF_more_data
))
768 if (cons
+ slots
== rp
) {
770 dev_warn(dev
, "Need more slots\n");
775 rx
= RING_GET_RESPONSE(&np
->rx
, cons
+ slots
);
776 skb
= xennet_get_rx_skb(np
, cons
+ slots
);
777 ref
= xennet_get_rx_ref(np
, cons
+ slots
);
781 if (unlikely(slots
> max
)) {
783 dev_warn(dev
, "Too many slots\n");
788 np
->rx
.rsp_cons
= cons
+ slots
;
793 static int xennet_set_skb_gso(struct sk_buff
*skb
,
794 struct xen_netif_extra_info
*gso
)
796 if (!gso
->u
.gso
.size
) {
798 printk(KERN_WARNING
"GSO size must not be zero.\n");
802 /* Currently only TCPv4 S.O. is supported. */
803 if (gso
->u
.gso
.type
!= XEN_NETIF_GSO_TYPE_TCPV4
) {
805 printk(KERN_WARNING
"Bad GSO type %d.\n", gso
->u
.gso
.type
);
809 skb_shinfo(skb
)->gso_size
= gso
->u
.gso
.size
;
810 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV4
;
812 /* Header must be checked, and gso_segs computed. */
813 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
814 skb_shinfo(skb
)->gso_segs
= 0;
819 static RING_IDX
xennet_fill_frags(struct netfront_info
*np
,
821 struct sk_buff_head
*list
)
823 struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
824 RING_IDX cons
= np
->rx
.rsp_cons
;
825 struct sk_buff
*nskb
;
827 while ((nskb
= __skb_dequeue(list
))) {
828 struct xen_netif_rx_response
*rx
=
829 RING_GET_RESPONSE(&np
->rx
, ++cons
);
830 skb_frag_t
*nfrag
= &skb_shinfo(nskb
)->frags
[0];
832 if (shinfo
->nr_frags
== MAX_SKB_FRAGS
) {
833 unsigned int pull_to
= NETFRONT_SKB_CB(skb
)->pull_to
;
835 BUG_ON(pull_to
<= skb_headlen(skb
));
836 __pskb_pull_tail(skb
, pull_to
- skb_headlen(skb
));
838 BUG_ON(shinfo
->nr_frags
>= MAX_SKB_FRAGS
);
840 skb_add_rx_frag(skb
, shinfo
->nr_frags
, skb_frag_page(nfrag
),
841 rx
->offset
, rx
->status
, PAGE_SIZE
);
843 skb_shinfo(nskb
)->nr_frags
= 0;
850 static int checksum_setup(struct net_device
*dev
, struct sk_buff
*skb
)
855 int recalculate_partial_csum
= 0;
858 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
859 * peers can fail to set NETRXF_csum_blank when sending a GSO
860 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
861 * recalculate the partial checksum.
863 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
&& skb_is_gso(skb
)) {
864 struct netfront_info
*np
= netdev_priv(dev
);
865 np
->rx_gso_checksum_fixup
++;
866 skb
->ip_summed
= CHECKSUM_PARTIAL
;
867 recalculate_partial_csum
= 1;
870 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
871 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
)
874 if (skb
->protocol
!= htons(ETH_P_IP
))
877 iph
= (void *)skb
->data
;
878 th
= skb
->data
+ 4 * iph
->ihl
;
879 if (th
>= skb_tail_pointer(skb
))
882 skb
->csum_start
= th
- skb
->head
;
883 switch (iph
->protocol
) {
885 skb
->csum_offset
= offsetof(struct tcphdr
, check
);
887 if (recalculate_partial_csum
) {
888 struct tcphdr
*tcph
= (struct tcphdr
*)th
;
889 tcph
->check
= ~csum_tcpudp_magic(iph
->saddr
, iph
->daddr
,
890 skb
->len
- iph
->ihl
*4,
895 skb
->csum_offset
= offsetof(struct udphdr
, check
);
897 if (recalculate_partial_csum
) {
898 struct udphdr
*udph
= (struct udphdr
*)th
;
899 udph
->check
= ~csum_tcpudp_magic(iph
->saddr
, iph
->daddr
,
900 skb
->len
- iph
->ihl
*4,
906 printk(KERN_ERR
"Attempting to checksum a non-"
907 "TCP/UDP packet, dropping a protocol"
908 " %d packet", iph
->protocol
);
912 if ((th
+ skb
->csum_offset
+ 2) > skb_tail_pointer(skb
))
921 static int handle_incoming_queue(struct net_device
*dev
,
922 struct sk_buff_head
*rxq
)
924 struct netfront_info
*np
= netdev_priv(dev
);
925 struct netfront_stats
*stats
= this_cpu_ptr(np
->stats
);
926 int packets_dropped
= 0;
929 while ((skb
= __skb_dequeue(rxq
)) != NULL
) {
930 int pull_to
= NETFRONT_SKB_CB(skb
)->pull_to
;
932 if (pull_to
> skb_headlen(skb
))
933 __pskb_pull_tail(skb
, pull_to
- skb_headlen(skb
));
935 /* Ethernet work: Delayed to here as it peeks the header. */
936 skb
->protocol
= eth_type_trans(skb
, dev
);
938 if (checksum_setup(dev
, skb
)) {
941 dev
->stats
.rx_errors
++;
945 u64_stats_update_begin(&stats
->syncp
);
947 stats
->rx_bytes
+= skb
->len
;
948 u64_stats_update_end(&stats
->syncp
);
951 netif_receive_skb(skb
);
954 return packets_dropped
;
957 static int xennet_poll(struct napi_struct
*napi
, int budget
)
959 struct netfront_info
*np
= container_of(napi
, struct netfront_info
, napi
);
960 struct net_device
*dev
= np
->netdev
;
962 struct netfront_rx_info rinfo
;
963 struct xen_netif_rx_response
*rx
= &rinfo
.rx
;
964 struct xen_netif_extra_info
*extras
= rinfo
.extras
;
967 struct sk_buff_head rxq
;
968 struct sk_buff_head errq
;
969 struct sk_buff_head tmpq
;
973 spin_lock(&np
->rx_lock
);
975 skb_queue_head_init(&rxq
);
976 skb_queue_head_init(&errq
);
977 skb_queue_head_init(&tmpq
);
979 rp
= np
->rx
.sring
->rsp_prod
;
980 rmb(); /* Ensure we see queued responses up to 'rp'. */
984 while ((i
!= rp
) && (work_done
< budget
)) {
985 memcpy(rx
, RING_GET_RESPONSE(&np
->rx
, i
), sizeof(*rx
));
986 memset(extras
, 0, sizeof(rinfo
.extras
));
988 err
= xennet_get_responses(np
, &rinfo
, rp
, &tmpq
);
992 while ((skb
= __skb_dequeue(&tmpq
)))
993 __skb_queue_tail(&errq
, skb
);
994 dev
->stats
.rx_errors
++;
999 skb
= __skb_dequeue(&tmpq
);
1001 if (extras
[XEN_NETIF_EXTRA_TYPE_GSO
- 1].type
) {
1002 struct xen_netif_extra_info
*gso
;
1003 gso
= &extras
[XEN_NETIF_EXTRA_TYPE_GSO
- 1];
1005 if (unlikely(xennet_set_skb_gso(skb
, gso
))) {
1006 __skb_queue_head(&tmpq
, skb
);
1007 np
->rx
.rsp_cons
+= skb_queue_len(&tmpq
);
1012 NETFRONT_SKB_CB(skb
)->pull_to
= rx
->status
;
1013 if (NETFRONT_SKB_CB(skb
)->pull_to
> RX_COPY_THRESHOLD
)
1014 NETFRONT_SKB_CB(skb
)->pull_to
= RX_COPY_THRESHOLD
;
1016 skb_shinfo(skb
)->frags
[0].page_offset
= rx
->offset
;
1017 skb_frag_size_set(&skb_shinfo(skb
)->frags
[0], rx
->status
);
1018 skb
->data_len
= rx
->status
;
1019 skb
->len
+= rx
->status
;
1021 i
= xennet_fill_frags(np
, skb
, &tmpq
);
1023 if (rx
->flags
& XEN_NETRXF_csum_blank
)
1024 skb
->ip_summed
= CHECKSUM_PARTIAL
;
1025 else if (rx
->flags
& XEN_NETRXF_data_validated
)
1026 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1028 __skb_queue_tail(&rxq
, skb
);
1030 np
->rx
.rsp_cons
= ++i
;
1034 __skb_queue_purge(&errq
);
1036 work_done
-= handle_incoming_queue(dev
, &rxq
);
1038 /* If we get a callback with very few responses, reduce fill target. */
1039 /* NB. Note exponential increase, linear decrease. */
1040 if (((np
->rx
.req_prod_pvt
- np
->rx
.sring
->rsp_prod
) >
1041 ((3*np
->rx_target
) / 4)) &&
1042 (--np
->rx_target
< np
->rx_min_target
))
1043 np
->rx_target
= np
->rx_min_target
;
1045 xennet_alloc_rx_buffers(dev
);
1047 if (work_done
< budget
) {
1050 local_irq_save(flags
);
1052 RING_FINAL_CHECK_FOR_RESPONSES(&np
->rx
, more_to_do
);
1054 __napi_complete(napi
);
1056 local_irq_restore(flags
);
1059 spin_unlock(&np
->rx_lock
);
1064 static int xennet_change_mtu(struct net_device
*dev
, int mtu
)
1066 int max
= xennet_can_sg(dev
) ?
1067 XEN_NETIF_MAX_TX_SIZE
- MAX_TCP_HEADER
: ETH_DATA_LEN
;
1075 static struct rtnl_link_stats64
*xennet_get_stats64(struct net_device
*dev
,
1076 struct rtnl_link_stats64
*tot
)
1078 struct netfront_info
*np
= netdev_priv(dev
);
1081 for_each_possible_cpu(cpu
) {
1082 struct netfront_stats
*stats
= per_cpu_ptr(np
->stats
, cpu
);
1083 u64 rx_packets
, rx_bytes
, tx_packets
, tx_bytes
;
1087 start
= u64_stats_fetch_begin_bh(&stats
->syncp
);
1089 rx_packets
= stats
->rx_packets
;
1090 tx_packets
= stats
->tx_packets
;
1091 rx_bytes
= stats
->rx_bytes
;
1092 tx_bytes
= stats
->tx_bytes
;
1093 } while (u64_stats_fetch_retry_bh(&stats
->syncp
, start
));
1095 tot
->rx_packets
+= rx_packets
;
1096 tot
->tx_packets
+= tx_packets
;
1097 tot
->rx_bytes
+= rx_bytes
;
1098 tot
->tx_bytes
+= tx_bytes
;
1101 tot
->rx_errors
= dev
->stats
.rx_errors
;
1102 tot
->tx_dropped
= dev
->stats
.tx_dropped
;
1107 static void xennet_release_tx_bufs(struct netfront_info
*np
)
1109 struct sk_buff
*skb
;
1112 for (i
= 0; i
< NET_TX_RING_SIZE
; i
++) {
1113 /* Skip over entries which are actually freelist references */
1114 if (skb_entry_is_link(&np
->tx_skbs
[i
]))
1117 skb
= np
->tx_skbs
[i
].skb
;
1118 get_page(np
->grant_tx_page
[i
]);
1119 gnttab_end_foreign_access(np
->grant_tx_ref
[i
],
1121 (unsigned long)page_address(np
->grant_tx_page
[i
]));
1122 np
->grant_tx_page
[i
] = NULL
;
1123 np
->grant_tx_ref
[i
] = GRANT_INVALID_REF
;
1124 add_id_to_freelist(&np
->tx_skb_freelist
, np
->tx_skbs
, i
);
1125 dev_kfree_skb_irq(skb
);
1129 static void xennet_release_rx_bufs(struct netfront_info
*np
)
1133 spin_lock_bh(&np
->rx_lock
);
1135 for (id
= 0; id
< NET_RX_RING_SIZE
; id
++) {
1136 struct sk_buff
*skb
;
1139 skb
= np
->rx_skbs
[id
];
1143 ref
= np
->grant_rx_ref
[id
];
1144 if (ref
== GRANT_INVALID_REF
)
1147 page
= skb_frag_page(&skb_shinfo(skb
)->frags
[0]);
1149 /* gnttab_end_foreign_access() needs a page ref until
1150 * foreign access is ended (which may be deferred).
1153 gnttab_end_foreign_access(ref
, 0,
1154 (unsigned long)page_address(page
));
1155 np
->grant_rx_ref
[id
] = GRANT_INVALID_REF
;
1160 spin_unlock_bh(&np
->rx_lock
);
1163 static void xennet_uninit(struct net_device
*dev
)
1165 struct netfront_info
*np
= netdev_priv(dev
);
1166 xennet_release_tx_bufs(np
);
1167 xennet_release_rx_bufs(np
);
1168 gnttab_free_grant_references(np
->gref_tx_head
);
1169 gnttab_free_grant_references(np
->gref_rx_head
);
1172 static netdev_features_t
xennet_fix_features(struct net_device
*dev
,
1173 netdev_features_t features
)
1175 struct netfront_info
*np
= netdev_priv(dev
);
1178 if (features
& NETIF_F_SG
) {
1179 if (xenbus_scanf(XBT_NIL
, np
->xbdev
->otherend
, "feature-sg",
1184 features
&= ~NETIF_F_SG
;
1187 if (features
& NETIF_F_TSO
) {
1188 if (xenbus_scanf(XBT_NIL
, np
->xbdev
->otherend
,
1189 "feature-gso-tcpv4", "%d", &val
) < 0)
1193 features
&= ~NETIF_F_TSO
;
1199 static int xennet_set_features(struct net_device
*dev
,
1200 netdev_features_t features
)
1202 if (!(features
& NETIF_F_SG
) && dev
->mtu
> ETH_DATA_LEN
) {
1203 netdev_info(dev
, "Reducing MTU because no SG offload");
1204 dev
->mtu
= ETH_DATA_LEN
;
1210 static irqreturn_t
xennet_interrupt(int irq
, void *dev_id
)
1212 struct net_device
*dev
= dev_id
;
1213 struct netfront_info
*np
= netdev_priv(dev
);
1214 unsigned long flags
;
1216 spin_lock_irqsave(&np
->tx_lock
, flags
);
1218 if (likely(netif_carrier_ok(dev
))) {
1219 xennet_tx_buf_gc(dev
);
1220 /* Under tx_lock: protects access to rx shared-ring indexes. */
1221 if (RING_HAS_UNCONSUMED_RESPONSES(&np
->rx
))
1222 napi_schedule(&np
->napi
);
1225 spin_unlock_irqrestore(&np
->tx_lock
, flags
);
1230 #ifdef CONFIG_NET_POLL_CONTROLLER
1231 static void xennet_poll_controller(struct net_device
*dev
)
1233 xennet_interrupt(0, dev
);
1237 static const struct net_device_ops xennet_netdev_ops
= {
1238 .ndo_open
= xennet_open
,
1239 .ndo_uninit
= xennet_uninit
,
1240 .ndo_stop
= xennet_close
,
1241 .ndo_start_xmit
= xennet_start_xmit
,
1242 .ndo_change_mtu
= xennet_change_mtu
,
1243 .ndo_get_stats64
= xennet_get_stats64
,
1244 .ndo_set_mac_address
= eth_mac_addr
,
1245 .ndo_validate_addr
= eth_validate_addr
,
1246 .ndo_fix_features
= xennet_fix_features
,
1247 .ndo_set_features
= xennet_set_features
,
1248 #ifdef CONFIG_NET_POLL_CONTROLLER
1249 .ndo_poll_controller
= xennet_poll_controller
,
1253 static struct net_device
*xennet_create_dev(struct xenbus_device
*dev
)
1256 struct net_device
*netdev
;
1257 struct netfront_info
*np
;
1259 netdev
= alloc_etherdev(sizeof(struct netfront_info
));
1261 return ERR_PTR(-ENOMEM
);
1263 np
= netdev_priv(netdev
);
1266 spin_lock_init(&np
->tx_lock
);
1267 spin_lock_init(&np
->rx_lock
);
1269 skb_queue_head_init(&np
->rx_batch
);
1270 np
->rx_target
= RX_DFL_MIN_TARGET
;
1271 np
->rx_min_target
= RX_DFL_MIN_TARGET
;
1272 np
->rx_max_target
= RX_MAX_TARGET
;
1274 init_timer(&np
->rx_refill_timer
);
1275 np
->rx_refill_timer
.data
= (unsigned long)netdev
;
1276 np
->rx_refill_timer
.function
= rx_refill_timeout
;
1279 np
->stats
= alloc_percpu(struct netfront_stats
);
1280 if (np
->stats
== NULL
)
1283 /* Initialise tx_skbs as a free chain containing every entry. */
1284 np
->tx_skb_freelist
= 0;
1285 for (i
= 0; i
< NET_TX_RING_SIZE
; i
++) {
1286 skb_entry_set_link(&np
->tx_skbs
[i
], i
+1);
1287 np
->grant_tx_ref
[i
] = GRANT_INVALID_REF
;
1290 /* Clear out rx_skbs */
1291 for (i
= 0; i
< NET_RX_RING_SIZE
; i
++) {
1292 np
->rx_skbs
[i
] = NULL
;
1293 np
->grant_rx_ref
[i
] = GRANT_INVALID_REF
;
1294 np
->grant_tx_page
[i
] = NULL
;
1297 /* A grant for every tx ring slot */
1298 if (gnttab_alloc_grant_references(TX_MAX_TARGET
,
1299 &np
->gref_tx_head
) < 0) {
1300 printk(KERN_ALERT
"#### netfront can't alloc tx grant refs\n");
1302 goto exit_free_stats
;
1304 /* A grant for every rx ring slot */
1305 if (gnttab_alloc_grant_references(RX_MAX_TARGET
,
1306 &np
->gref_rx_head
) < 0) {
1307 printk(KERN_ALERT
"#### netfront can't alloc rx grant refs\n");
1312 netdev
->netdev_ops
= &xennet_netdev_ops
;
1314 netif_napi_add(netdev
, &np
->napi
, xennet_poll
, 64);
1315 netdev
->features
= NETIF_F_IP_CSUM
| NETIF_F_RXCSUM
|
1317 netdev
->hw_features
= NETIF_F_IP_CSUM
| NETIF_F_SG
| NETIF_F_TSO
;
1320 * Assume that all hw features are available for now. This set
1321 * will be adjusted by the call to netdev_update_features() in
1322 * xennet_connect() which is the earliest point where we can
1323 * negotiate with the backend regarding supported features.
1325 netdev
->features
|= netdev
->hw_features
;
1327 SET_ETHTOOL_OPS(netdev
, &xennet_ethtool_ops
);
1328 SET_NETDEV_DEV(netdev
, &dev
->dev
);
1330 netif_set_gso_max_size(netdev
, XEN_NETIF_MAX_TX_SIZE
- MAX_TCP_HEADER
);
1332 np
->netdev
= netdev
;
1334 netif_carrier_off(netdev
);
1339 gnttab_free_grant_references(np
->gref_tx_head
);
1341 free_percpu(np
->stats
);
1343 free_netdev(netdev
);
1344 return ERR_PTR(err
);
1348 * Entry point to this code when a new device is created. Allocate the basic
1349 * structures and the ring buffers for communication with the backend, and
1350 * inform the backend of the appropriate details for those.
1352 static int netfront_probe(struct xenbus_device
*dev
,
1353 const struct xenbus_device_id
*id
)
1356 struct net_device
*netdev
;
1357 struct netfront_info
*info
;
1359 netdev
= xennet_create_dev(dev
);
1360 if (IS_ERR(netdev
)) {
1361 err
= PTR_ERR(netdev
);
1362 xenbus_dev_fatal(dev
, err
, "creating netdev");
1366 info
= netdev_priv(netdev
);
1367 dev_set_drvdata(&dev
->dev
, info
);
1369 err
= register_netdev(info
->netdev
);
1371 printk(KERN_WARNING
"%s: register_netdev err=%d\n",
1376 err
= xennet_sysfs_addif(info
->netdev
);
1378 unregister_netdev(info
->netdev
);
1379 printk(KERN_WARNING
"%s: add sysfs failed err=%d\n",
1387 free_netdev(netdev
);
1388 dev_set_drvdata(&dev
->dev
, NULL
);
1392 static void xennet_end_access(int ref
, void *page
)
1394 /* This frees the page as a side-effect */
1395 if (ref
!= GRANT_INVALID_REF
)
1396 gnttab_end_foreign_access(ref
, 0, (unsigned long)page
);
1399 static void xennet_disconnect_backend(struct netfront_info
*info
)
1401 /* Stop old i/f to prevent errors whilst we rebuild the state. */
1402 spin_lock_bh(&info
->rx_lock
);
1403 spin_lock_irq(&info
->tx_lock
);
1404 netif_carrier_off(info
->netdev
);
1405 spin_unlock_irq(&info
->tx_lock
);
1406 spin_unlock_bh(&info
->rx_lock
);
1408 if (info
->netdev
->irq
)
1409 unbind_from_irqhandler(info
->netdev
->irq
, info
->netdev
);
1410 info
->evtchn
= info
->netdev
->irq
= 0;
1412 /* End access and free the pages */
1413 xennet_end_access(info
->tx_ring_ref
, info
->tx
.sring
);
1414 xennet_end_access(info
->rx_ring_ref
, info
->rx
.sring
);
1416 info
->tx_ring_ref
= GRANT_INVALID_REF
;
1417 info
->rx_ring_ref
= GRANT_INVALID_REF
;
1418 info
->tx
.sring
= NULL
;
1419 info
->rx
.sring
= NULL
;
1423 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1424 * driver restart. We tear down our netif structure and recreate it, but
1425 * leave the device-layer structures intact so that this is transparent to the
1426 * rest of the kernel.
1428 static int netfront_resume(struct xenbus_device
*dev
)
1430 struct netfront_info
*info
= dev_get_drvdata(&dev
->dev
);
1432 dev_dbg(&dev
->dev
, "%s\n", dev
->nodename
);
1434 xennet_disconnect_backend(info
);
1438 static int xen_net_read_mac(struct xenbus_device
*dev
, u8 mac
[])
1440 char *s
, *e
, *macstr
;
1443 macstr
= s
= xenbus_read(XBT_NIL
, dev
->nodename
, "mac", NULL
);
1445 return PTR_ERR(macstr
);
1447 for (i
= 0; i
< ETH_ALEN
; i
++) {
1448 mac
[i
] = simple_strtoul(s
, &e
, 16);
1449 if ((s
== e
) || (*e
!= ((i
== ETH_ALEN
-1) ? '\0' : ':'))) {
1460 static int setup_netfront(struct xenbus_device
*dev
, struct netfront_info
*info
)
1462 struct xen_netif_tx_sring
*txs
;
1463 struct xen_netif_rx_sring
*rxs
;
1465 struct net_device
*netdev
= info
->netdev
;
1467 info
->tx_ring_ref
= GRANT_INVALID_REF
;
1468 info
->rx_ring_ref
= GRANT_INVALID_REF
;
1469 info
->rx
.sring
= NULL
;
1470 info
->tx
.sring
= NULL
;
1473 err
= xen_net_read_mac(dev
, netdev
->dev_addr
);
1475 xenbus_dev_fatal(dev
, err
, "parsing %s/mac", dev
->nodename
);
1479 txs
= (struct xen_netif_tx_sring
*)get_zeroed_page(GFP_NOIO
| __GFP_HIGH
);
1482 xenbus_dev_fatal(dev
, err
, "allocating tx ring page");
1485 SHARED_RING_INIT(txs
);
1486 FRONT_RING_INIT(&info
->tx
, txs
, PAGE_SIZE
);
1488 err
= xenbus_grant_ring(dev
, virt_to_mfn(txs
));
1490 free_page((unsigned long)txs
);
1494 info
->tx_ring_ref
= err
;
1495 rxs
= (struct xen_netif_rx_sring
*)get_zeroed_page(GFP_NOIO
| __GFP_HIGH
);
1498 xenbus_dev_fatal(dev
, err
, "allocating rx ring page");
1501 SHARED_RING_INIT(rxs
);
1502 FRONT_RING_INIT(&info
->rx
, rxs
, PAGE_SIZE
);
1504 err
= xenbus_grant_ring(dev
, virt_to_mfn(rxs
));
1506 free_page((unsigned long)rxs
);
1509 info
->rx_ring_ref
= err
;
1511 err
= xenbus_alloc_evtchn(dev
, &info
->evtchn
);
1515 err
= bind_evtchn_to_irqhandler(info
->evtchn
, xennet_interrupt
,
1516 0, netdev
->name
, netdev
);
1526 /* Common code used when first setting up, and when resuming. */
1527 static int talk_to_netback(struct xenbus_device
*dev
,
1528 struct netfront_info
*info
)
1530 const char *message
;
1531 struct xenbus_transaction xbt
;
1534 /* Create shared ring, alloc event channel. */
1535 err
= setup_netfront(dev
, info
);
1540 err
= xenbus_transaction_start(&xbt
);
1542 xenbus_dev_fatal(dev
, err
, "starting transaction");
1546 err
= xenbus_printf(xbt
, dev
->nodename
, "tx-ring-ref", "%u",
1549 message
= "writing tx ring-ref";
1550 goto abort_transaction
;
1552 err
= xenbus_printf(xbt
, dev
->nodename
, "rx-ring-ref", "%u",
1555 message
= "writing rx ring-ref";
1556 goto abort_transaction
;
1558 err
= xenbus_printf(xbt
, dev
->nodename
,
1559 "event-channel", "%u", info
->evtchn
);
1561 message
= "writing event-channel";
1562 goto abort_transaction
;
1565 err
= xenbus_printf(xbt
, dev
->nodename
, "request-rx-copy", "%u",
1568 message
= "writing request-rx-copy";
1569 goto abort_transaction
;
1572 err
= xenbus_printf(xbt
, dev
->nodename
, "feature-rx-notify", "%d", 1);
1574 message
= "writing feature-rx-notify";
1575 goto abort_transaction
;
1578 err
= xenbus_printf(xbt
, dev
->nodename
, "feature-sg", "%d", 1);
1580 message
= "writing feature-sg";
1581 goto abort_transaction
;
1584 err
= xenbus_printf(xbt
, dev
->nodename
, "feature-gso-tcpv4", "%d", 1);
1586 message
= "writing feature-gso-tcpv4";
1587 goto abort_transaction
;
1590 err
= xenbus_transaction_end(xbt
, 0);
1594 xenbus_dev_fatal(dev
, err
, "completing transaction");
1601 xenbus_transaction_end(xbt
, 1);
1602 xenbus_dev_fatal(dev
, err
, "%s", message
);
1604 xennet_disconnect_backend(info
);
1609 static int xennet_connect(struct net_device
*dev
)
1611 struct netfront_info
*np
= netdev_priv(dev
);
1612 int i
, requeue_idx
, err
;
1613 struct sk_buff
*skb
;
1615 struct xen_netif_rx_request
*req
;
1616 unsigned int feature_rx_copy
;
1618 err
= xenbus_scanf(XBT_NIL
, np
->xbdev
->otherend
,
1619 "feature-rx-copy", "%u", &feature_rx_copy
);
1621 feature_rx_copy
= 0;
1623 if (!feature_rx_copy
) {
1625 "backend does not support copying receive path\n");
1629 err
= talk_to_netback(np
->xbdev
, np
);
1634 netdev_update_features(dev
);
1637 spin_lock_bh(&np
->rx_lock
);
1638 spin_lock_irq(&np
->tx_lock
);
1640 /* Step 1: Discard all pending TX packet fragments. */
1641 xennet_release_tx_bufs(np
);
1643 /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1644 for (requeue_idx
= 0, i
= 0; i
< NET_RX_RING_SIZE
; i
++) {
1646 const struct page
*page
;
1647 if (!np
->rx_skbs
[i
])
1650 skb
= np
->rx_skbs
[requeue_idx
] = xennet_get_rx_skb(np
, i
);
1651 ref
= np
->grant_rx_ref
[requeue_idx
] = xennet_get_rx_ref(np
, i
);
1652 req
= RING_GET_REQUEST(&np
->rx
, requeue_idx
);
1654 frag
= &skb_shinfo(skb
)->frags
[0];
1655 page
= skb_frag_page(frag
);
1656 gnttab_grant_foreign_access_ref(
1657 ref
, np
->xbdev
->otherend_id
,
1658 pfn_to_mfn(page_to_pfn(page
)),
1661 req
->id
= requeue_idx
;
1666 np
->rx
.req_prod_pvt
= requeue_idx
;
1669 * Step 3: All public and private state should now be sane. Get
1670 * ready to start sending and receiving packets and give the driver
1671 * domain a kick because we've probably just requeued some
1674 netif_carrier_on(np
->netdev
);
1675 notify_remote_via_irq(np
->netdev
->irq
);
1676 xennet_tx_buf_gc(dev
);
1677 xennet_alloc_rx_buffers(dev
);
1679 spin_unlock_irq(&np
->tx_lock
);
1680 spin_unlock_bh(&np
->rx_lock
);
1686 * Callback received when the backend's state changes.
1688 static void netback_changed(struct xenbus_device
*dev
,
1689 enum xenbus_state backend_state
)
1691 struct netfront_info
*np
= dev_get_drvdata(&dev
->dev
);
1692 struct net_device
*netdev
= np
->netdev
;
1694 dev_dbg(&dev
->dev
, "%s\n", xenbus_strstate(backend_state
));
1696 switch (backend_state
) {
1697 case XenbusStateInitialising
:
1698 case XenbusStateInitialised
:
1699 case XenbusStateReconfiguring
:
1700 case XenbusStateReconfigured
:
1701 case XenbusStateUnknown
:
1702 case XenbusStateClosed
:
1705 case XenbusStateInitWait
:
1706 if (dev
->state
!= XenbusStateInitialising
)
1708 if (xennet_connect(netdev
) != 0)
1710 xenbus_switch_state(dev
, XenbusStateConnected
);
1713 case XenbusStateConnected
:
1714 netdev_notify_peers(netdev
);
1717 case XenbusStateClosing
:
1718 xenbus_frontend_closed(dev
);
1723 static const struct xennet_stat
{
1724 char name
[ETH_GSTRING_LEN
];
1726 } xennet_stats
[] = {
1728 "rx_gso_checksum_fixup",
1729 offsetof(struct netfront_info
, rx_gso_checksum_fixup
)
1733 static int xennet_get_sset_count(struct net_device
*dev
, int string_set
)
1735 switch (string_set
) {
1737 return ARRAY_SIZE(xennet_stats
);
1743 static void xennet_get_ethtool_stats(struct net_device
*dev
,
1744 struct ethtool_stats
*stats
, u64
* data
)
1746 void *np
= netdev_priv(dev
);
1749 for (i
= 0; i
< ARRAY_SIZE(xennet_stats
); i
++)
1750 data
[i
] = *(unsigned long *)(np
+ xennet_stats
[i
].offset
);
1753 static void xennet_get_strings(struct net_device
*dev
, u32 stringset
, u8
* data
)
1757 switch (stringset
) {
1759 for (i
= 0; i
< ARRAY_SIZE(xennet_stats
); i
++)
1760 memcpy(data
+ i
* ETH_GSTRING_LEN
,
1761 xennet_stats
[i
].name
, ETH_GSTRING_LEN
);
1766 static const struct ethtool_ops xennet_ethtool_ops
=
1768 .get_link
= ethtool_op_get_link
,
1770 .get_sset_count
= xennet_get_sset_count
,
1771 .get_ethtool_stats
= xennet_get_ethtool_stats
,
1772 .get_strings
= xennet_get_strings
,
1776 static ssize_t
show_rxbuf_min(struct device
*dev
,
1777 struct device_attribute
*attr
, char *buf
)
1779 struct net_device
*netdev
= to_net_dev(dev
);
1780 struct netfront_info
*info
= netdev_priv(netdev
);
1782 return sprintf(buf
, "%u\n", info
->rx_min_target
);
1785 static ssize_t
store_rxbuf_min(struct device
*dev
,
1786 struct device_attribute
*attr
,
1787 const char *buf
, size_t len
)
1789 struct net_device
*netdev
= to_net_dev(dev
);
1790 struct netfront_info
*np
= netdev_priv(netdev
);
1792 unsigned long target
;
1794 if (!capable(CAP_NET_ADMIN
))
1797 target
= simple_strtoul(buf
, &endp
, 0);
1801 if (target
< RX_MIN_TARGET
)
1802 target
= RX_MIN_TARGET
;
1803 if (target
> RX_MAX_TARGET
)
1804 target
= RX_MAX_TARGET
;
1806 spin_lock_bh(&np
->rx_lock
);
1807 if (target
> np
->rx_max_target
)
1808 np
->rx_max_target
= target
;
1809 np
->rx_min_target
= target
;
1810 if (target
> np
->rx_target
)
1811 np
->rx_target
= target
;
1813 xennet_alloc_rx_buffers(netdev
);
1815 spin_unlock_bh(&np
->rx_lock
);
1819 static ssize_t
show_rxbuf_max(struct device
*dev
,
1820 struct device_attribute
*attr
, char *buf
)
1822 struct net_device
*netdev
= to_net_dev(dev
);
1823 struct netfront_info
*info
= netdev_priv(netdev
);
1825 return sprintf(buf
, "%u\n", info
->rx_max_target
);
1828 static ssize_t
store_rxbuf_max(struct device
*dev
,
1829 struct device_attribute
*attr
,
1830 const char *buf
, size_t len
)
1832 struct net_device
*netdev
= to_net_dev(dev
);
1833 struct netfront_info
*np
= netdev_priv(netdev
);
1835 unsigned long target
;
1837 if (!capable(CAP_NET_ADMIN
))
1840 target
= simple_strtoul(buf
, &endp
, 0);
1844 if (target
< RX_MIN_TARGET
)
1845 target
= RX_MIN_TARGET
;
1846 if (target
> RX_MAX_TARGET
)
1847 target
= RX_MAX_TARGET
;
1849 spin_lock_bh(&np
->rx_lock
);
1850 if (target
< np
->rx_min_target
)
1851 np
->rx_min_target
= target
;
1852 np
->rx_max_target
= target
;
1853 if (target
< np
->rx_target
)
1854 np
->rx_target
= target
;
1856 xennet_alloc_rx_buffers(netdev
);
1858 spin_unlock_bh(&np
->rx_lock
);
1862 static ssize_t
show_rxbuf_cur(struct device
*dev
,
1863 struct device_attribute
*attr
, char *buf
)
1865 struct net_device
*netdev
= to_net_dev(dev
);
1866 struct netfront_info
*info
= netdev_priv(netdev
);
1868 return sprintf(buf
, "%u\n", info
->rx_target
);
1871 static struct device_attribute xennet_attrs
[] = {
1872 __ATTR(rxbuf_min
, S_IRUGO
|S_IWUSR
, show_rxbuf_min
, store_rxbuf_min
),
1873 __ATTR(rxbuf_max
, S_IRUGO
|S_IWUSR
, show_rxbuf_max
, store_rxbuf_max
),
1874 __ATTR(rxbuf_cur
, S_IRUGO
, show_rxbuf_cur
, NULL
),
1877 static int xennet_sysfs_addif(struct net_device
*netdev
)
1882 for (i
= 0; i
< ARRAY_SIZE(xennet_attrs
); i
++) {
1883 err
= device_create_file(&netdev
->dev
,
1892 device_remove_file(&netdev
->dev
, &xennet_attrs
[i
]);
1896 static void xennet_sysfs_delif(struct net_device
*netdev
)
1900 for (i
= 0; i
< ARRAY_SIZE(xennet_attrs
); i
++)
1901 device_remove_file(&netdev
->dev
, &xennet_attrs
[i
]);
1904 #endif /* CONFIG_SYSFS */
1906 static const struct xenbus_device_id netfront_ids
[] = {
1912 static int xennet_remove(struct xenbus_device
*dev
)
1914 struct netfront_info
*info
= dev_get_drvdata(&dev
->dev
);
1916 dev_dbg(&dev
->dev
, "%s\n", dev
->nodename
);
1918 xennet_disconnect_backend(info
);
1920 xennet_sysfs_delif(info
->netdev
);
1922 unregister_netdev(info
->netdev
);
1924 del_timer_sync(&info
->rx_refill_timer
);
1926 free_percpu(info
->stats
);
1928 free_netdev(info
->netdev
);
1933 static DEFINE_XENBUS_DRIVER(netfront
, ,
1934 .probe
= netfront_probe
,
1935 .remove
= xennet_remove
,
1936 .resume
= netfront_resume
,
1937 .otherend_changed
= netback_changed
,
1940 static int __init
netif_init(void)
1945 if (xen_hvm_domain() && !xen_platform_pci_unplug
)
1948 printk(KERN_INFO
"Initialising Xen virtual ethernet driver.\n");
1950 return xenbus_register_frontend(&netfront_driver
);
1952 module_init(netif_init
);
1955 static void __exit
netif_exit(void)
1957 xenbus_unregister_driver(&netfront_driver
);
1959 module_exit(netif_exit
);
1961 MODULE_DESCRIPTION("Xen virtual network device frontend");
1962 MODULE_LICENSE("GPL");
1963 MODULE_ALIAS("xen:vif");
1964 MODULE_ALIAS("xennet");