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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
34 #include <linux/module.h>
35 #include <linux/kernel.h>
36 #include <linux/netdevice.h>
37 #include <linux/etherdevice.h>
38 #include <linux/skbuff.h>
39 #include <linux/ethtool.h>
40 #include <linux/if_ether.h>
42 #include <linux/udp.h>
43 #include <linux/moduleparam.h>
45 #include <linux/slab.h>
48 #include <asm/xen/page.h>
50 #include <xen/xenbus.h>
51 #include <xen/events.h>
53 #include <xen/platform_pci.h>
54 #include <xen/grant_table.h>
56 #include <xen/interface/io/netif.h>
57 #include <xen/interface/memory.h>
58 #include <xen/interface/grant_table.h>
60 static const struct ethtool_ops xennet_ethtool_ops
;
66 #define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb))
68 #define RX_COPY_THRESHOLD 256
70 #define GRANT_INVALID_REF 0
72 #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, PAGE_SIZE)
73 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, PAGE_SIZE)
74 #define TX_MAX_TARGET min_t(int, NET_TX_RING_SIZE, 256)
76 struct netfront_stats
{
81 struct u64_stats_sync syncp
;
84 struct netfront_info
{
85 struct list_head list
;
86 struct net_device
*netdev
;
88 struct napi_struct napi
;
90 /* Split event channels support, tx_* == rx_* when using
91 * single event channel.
93 unsigned int tx_evtchn
, rx_evtchn
;
94 unsigned int tx_irq
, rx_irq
;
95 /* Only used when split event channels support is enabled */
96 char tx_irq_name
[IFNAMSIZ
+4]; /* DEVNAME-tx */
97 char rx_irq_name
[IFNAMSIZ
+4]; /* DEVNAME-rx */
99 struct xenbus_device
*xbdev
;
102 struct xen_netif_tx_front_ring tx
;
106 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
107 * are linked from tx_skb_freelist through skb_entry.link.
109 * NB. Freelist index entries are always going to be less than
110 * PAGE_OFFSET, whereas pointers to skbs will always be equal or
111 * greater than PAGE_OFFSET: we use this property to distinguish
117 } tx_skbs
[NET_TX_RING_SIZE
];
118 grant_ref_t gref_tx_head
;
119 grant_ref_t grant_tx_ref
[NET_TX_RING_SIZE
];
120 struct page
*grant_tx_page
[NET_TX_RING_SIZE
];
121 unsigned tx_skb_freelist
;
123 spinlock_t rx_lock ____cacheline_aligned_in_smp
;
124 struct xen_netif_rx_front_ring rx
;
127 /* Receive-ring batched refills. */
128 #define RX_MIN_TARGET 8
129 #define RX_DFL_MIN_TARGET 64
130 #define RX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256)
131 unsigned rx_min_target
, rx_max_target
, rx_target
;
132 struct sk_buff_head rx_batch
;
134 struct timer_list rx_refill_timer
;
136 struct sk_buff
*rx_skbs
[NET_RX_RING_SIZE
];
137 grant_ref_t gref_rx_head
;
138 grant_ref_t grant_rx_ref
[NET_RX_RING_SIZE
];
140 unsigned long rx_pfn_array
[NET_RX_RING_SIZE
];
141 struct multicall_entry rx_mcl
[NET_RX_RING_SIZE
+1];
142 struct mmu_update rx_mmu
[NET_RX_RING_SIZE
];
145 struct netfront_stats __percpu
*stats
;
147 unsigned long rx_gso_checksum_fixup
;
150 struct netfront_rx_info
{
151 struct xen_netif_rx_response rx
;
152 struct xen_netif_extra_info extras
[XEN_NETIF_EXTRA_TYPE_MAX
- 1];
155 static void skb_entry_set_link(union skb_entry
*list
, unsigned short id
)
160 static int skb_entry_is_link(const union skb_entry
*list
)
162 BUILD_BUG_ON(sizeof(list
->skb
) != sizeof(list
->link
));
163 return (unsigned long)list
->skb
< PAGE_OFFSET
;
167 * Access macros for acquiring freeing slots in tx_skbs[].
170 static void add_id_to_freelist(unsigned *head
, union skb_entry
*list
,
173 skb_entry_set_link(&list
[id
], *head
);
177 static unsigned short get_id_from_freelist(unsigned *head
,
178 union skb_entry
*list
)
180 unsigned int id
= *head
;
181 *head
= list
[id
].link
;
185 static int xennet_rxidx(RING_IDX idx
)
187 return idx
& (NET_RX_RING_SIZE
- 1);
190 static struct sk_buff
*xennet_get_rx_skb(struct netfront_info
*np
,
193 int i
= xennet_rxidx(ri
);
194 struct sk_buff
*skb
= np
->rx_skbs
[i
];
195 np
->rx_skbs
[i
] = NULL
;
199 static grant_ref_t
xennet_get_rx_ref(struct netfront_info
*np
,
202 int i
= xennet_rxidx(ri
);
203 grant_ref_t ref
= np
->grant_rx_ref
[i
];
204 np
->grant_rx_ref
[i
] = GRANT_INVALID_REF
;
209 static int xennet_sysfs_addif(struct net_device
*netdev
);
210 static void xennet_sysfs_delif(struct net_device
*netdev
);
211 #else /* !CONFIG_SYSFS */
212 #define xennet_sysfs_addif(dev) (0)
213 #define xennet_sysfs_delif(dev) do { } while (0)
216 static bool xennet_can_sg(struct net_device
*dev
)
218 return dev
->features
& NETIF_F_SG
;
222 static void rx_refill_timeout(unsigned long data
)
224 struct net_device
*dev
= (struct net_device
*)data
;
225 struct netfront_info
*np
= netdev_priv(dev
);
226 napi_schedule(&np
->napi
);
229 static int netfront_tx_slot_available(struct netfront_info
*np
)
231 return (np
->tx
.req_prod_pvt
- np
->tx
.rsp_cons
) <
232 (TX_MAX_TARGET
- MAX_SKB_FRAGS
- 2);
235 static void xennet_maybe_wake_tx(struct net_device
*dev
)
237 struct netfront_info
*np
= netdev_priv(dev
);
239 if (unlikely(netif_queue_stopped(dev
)) &&
240 netfront_tx_slot_available(np
) &&
241 likely(netif_running(dev
)))
242 netif_wake_queue(dev
);
245 static void xennet_alloc_rx_buffers(struct net_device
*dev
)
248 struct netfront_info
*np
= netdev_priv(dev
);
251 int i
, batch_target
, notify
;
252 RING_IDX req_prod
= np
->rx
.req_prod_pvt
;
256 struct xen_netif_rx_request
*req
;
258 if (unlikely(!netif_carrier_ok(dev
)))
262 * Allocate skbuffs greedily, even though we batch updates to the
263 * receive ring. This creates a less bursty demand on the memory
264 * allocator, so should reduce the chance of failed allocation requests
265 * both for ourself and for other kernel subsystems.
267 batch_target
= np
->rx_target
- (req_prod
- np
->rx
.rsp_cons
);
268 for (i
= skb_queue_len(&np
->rx_batch
); i
< batch_target
; i
++) {
269 skb
= __netdev_alloc_skb(dev
, RX_COPY_THRESHOLD
+ NET_IP_ALIGN
,
270 GFP_ATOMIC
| __GFP_NOWARN
);
274 /* Align ip header to a 16 bytes boundary */
275 skb_reserve(skb
, NET_IP_ALIGN
);
277 page
= alloc_page(GFP_ATOMIC
| __GFP_NOWARN
);
281 /* Could not allocate any skbuffs. Try again later. */
282 mod_timer(&np
->rx_refill_timer
,
285 /* Any skbuffs queued for refill? Force them out. */
291 skb_add_rx_frag(skb
, 0, page
, 0, 0, PAGE_SIZE
);
292 __skb_queue_tail(&np
->rx_batch
, skb
);
295 /* Is the batch large enough to be worthwhile? */
296 if (i
< (np
->rx_target
/2)) {
297 if (req_prod
> np
->rx
.sring
->req_prod
)
302 /* Adjust our fill target if we risked running out of buffers. */
303 if (((req_prod
- np
->rx
.sring
->rsp_prod
) < (np
->rx_target
/ 4)) &&
304 ((np
->rx_target
*= 2) > np
->rx_max_target
))
305 np
->rx_target
= np
->rx_max_target
;
309 skb
= __skb_dequeue(&np
->rx_batch
);
315 id
= xennet_rxidx(req_prod
+ i
);
317 BUG_ON(np
->rx_skbs
[id
]);
318 np
->rx_skbs
[id
] = skb
;
320 ref
= gnttab_claim_grant_reference(&np
->gref_rx_head
);
321 BUG_ON((signed short)ref
< 0);
322 np
->grant_rx_ref
[id
] = ref
;
324 pfn
= page_to_pfn(skb_frag_page(&skb_shinfo(skb
)->frags
[0]));
325 vaddr
= page_address(skb_frag_page(&skb_shinfo(skb
)->frags
[0]));
327 req
= RING_GET_REQUEST(&np
->rx
, req_prod
+ i
);
328 gnttab_grant_foreign_access_ref(ref
,
329 np
->xbdev
->otherend_id
,
337 wmb(); /* barrier so backend seens requests */
339 /* Above is a suitable barrier to ensure backend will see requests. */
340 np
->rx
.req_prod_pvt
= req_prod
+ i
;
342 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np
->rx
, notify
);
344 notify_remote_via_irq(np
->rx_irq
);
347 static int xennet_open(struct net_device
*dev
)
349 struct netfront_info
*np
= netdev_priv(dev
);
351 napi_enable(&np
->napi
);
353 spin_lock_bh(&np
->rx_lock
);
354 if (netif_carrier_ok(dev
)) {
355 xennet_alloc_rx_buffers(dev
);
356 np
->rx
.sring
->rsp_event
= np
->rx
.rsp_cons
+ 1;
357 if (RING_HAS_UNCONSUMED_RESPONSES(&np
->rx
))
358 napi_schedule(&np
->napi
);
360 spin_unlock_bh(&np
->rx_lock
);
362 netif_start_queue(dev
);
367 static void xennet_tx_buf_gc(struct net_device
*dev
)
371 struct netfront_info
*np
= netdev_priv(dev
);
374 BUG_ON(!netif_carrier_ok(dev
));
377 prod
= np
->tx
.sring
->rsp_prod
;
378 rmb(); /* Ensure we see responses up to 'rp'. */
380 for (cons
= np
->tx
.rsp_cons
; cons
!= prod
; cons
++) {
381 struct xen_netif_tx_response
*txrsp
;
383 txrsp
= RING_GET_RESPONSE(&np
->tx
, cons
);
384 if (txrsp
->status
== XEN_NETIF_RSP_NULL
)
388 skb
= np
->tx_skbs
[id
].skb
;
389 if (unlikely(gnttab_query_foreign_access(
390 np
->grant_tx_ref
[id
]) != 0)) {
391 pr_alert("%s: warning -- grant still in use by backend domain\n",
395 gnttab_end_foreign_access_ref(
396 np
->grant_tx_ref
[id
], GNTMAP_readonly
);
397 gnttab_release_grant_reference(
398 &np
->gref_tx_head
, np
->grant_tx_ref
[id
]);
399 np
->grant_tx_ref
[id
] = GRANT_INVALID_REF
;
400 np
->grant_tx_page
[id
] = NULL
;
401 add_id_to_freelist(&np
->tx_skb_freelist
, np
->tx_skbs
, id
);
402 dev_kfree_skb_irq(skb
);
405 np
->tx
.rsp_cons
= prod
;
408 * Set a new event, then check for race with update of tx_cons.
409 * Note that it is essential to schedule a callback, no matter
410 * how few buffers are pending. Even if there is space in the
411 * transmit ring, higher layers may be blocked because too much
412 * data is outstanding: in such cases notification from Xen is
413 * likely to be the only kick that we'll get.
415 np
->tx
.sring
->rsp_event
=
416 prod
+ ((np
->tx
.sring
->req_prod
- prod
) >> 1) + 1;
417 mb(); /* update shared area */
418 } while ((cons
== prod
) && (prod
!= np
->tx
.sring
->rsp_prod
));
420 xennet_maybe_wake_tx(dev
);
423 static void xennet_make_frags(struct sk_buff
*skb
, struct net_device
*dev
,
424 struct xen_netif_tx_request
*tx
)
426 struct netfront_info
*np
= netdev_priv(dev
);
427 char *data
= skb
->data
;
429 RING_IDX prod
= np
->tx
.req_prod_pvt
;
430 int frags
= skb_shinfo(skb
)->nr_frags
;
431 unsigned int offset
= offset_in_page(data
);
432 unsigned int len
= skb_headlen(skb
);
437 /* While the header overlaps a page boundary (including being
438 larger than a page), split it it into page-sized chunks. */
439 while (len
> PAGE_SIZE
- offset
) {
440 tx
->size
= PAGE_SIZE
- offset
;
441 tx
->flags
|= XEN_NETTXF_more_data
;
446 id
= get_id_from_freelist(&np
->tx_skb_freelist
, np
->tx_skbs
);
447 np
->tx_skbs
[id
].skb
= skb_get(skb
);
448 tx
= RING_GET_REQUEST(&np
->tx
, prod
++);
450 ref
= gnttab_claim_grant_reference(&np
->gref_tx_head
);
451 BUG_ON((signed short)ref
< 0);
453 mfn
= virt_to_mfn(data
);
454 gnttab_grant_foreign_access_ref(ref
, np
->xbdev
->otherend_id
,
455 mfn
, GNTMAP_readonly
);
457 np
->grant_tx_page
[id
] = virt_to_page(data
);
458 tx
->gref
= np
->grant_tx_ref
[id
] = ref
;
464 /* Grant backend access to each skb fragment page. */
465 for (i
= 0; i
< frags
; i
++) {
466 skb_frag_t
*frag
= skb_shinfo(skb
)->frags
+ i
;
467 struct page
*page
= skb_frag_page(frag
);
469 len
= skb_frag_size(frag
);
470 offset
= frag
->page_offset
;
472 /* Data must not cross a page boundary. */
473 BUG_ON(len
+ offset
> PAGE_SIZE
<<compound_order(page
));
475 /* Skip unused frames from start of page */
476 page
+= offset
>> PAGE_SHIFT
;
477 offset
&= ~PAGE_MASK
;
482 BUG_ON(offset
>= PAGE_SIZE
);
484 bytes
= PAGE_SIZE
- offset
;
488 tx
->flags
|= XEN_NETTXF_more_data
;
490 id
= get_id_from_freelist(&np
->tx_skb_freelist
,
492 np
->tx_skbs
[id
].skb
= skb_get(skb
);
493 tx
= RING_GET_REQUEST(&np
->tx
, prod
++);
495 ref
= gnttab_claim_grant_reference(&np
->gref_tx_head
);
496 BUG_ON((signed short)ref
< 0);
498 mfn
= pfn_to_mfn(page_to_pfn(page
));
499 gnttab_grant_foreign_access_ref(ref
,
500 np
->xbdev
->otherend_id
,
501 mfn
, GNTMAP_readonly
);
503 np
->grant_tx_page
[id
] = page
;
504 tx
->gref
= np
->grant_tx_ref
[id
] = ref
;
513 if (offset
== PAGE_SIZE
&& len
) {
514 BUG_ON(!PageCompound(page
));
521 np
->tx
.req_prod_pvt
= prod
;
525 * Count how many ring slots are required to send the frags of this
526 * skb. Each frag might be a compound page.
528 static int xennet_count_skb_frag_slots(struct sk_buff
*skb
)
530 int i
, frags
= skb_shinfo(skb
)->nr_frags
;
533 for (i
= 0; i
< frags
; i
++) {
534 skb_frag_t
*frag
= skb_shinfo(skb
)->frags
+ i
;
535 unsigned long size
= skb_frag_size(frag
);
536 unsigned long offset
= frag
->page_offset
;
538 /* Skip unused frames from start of page */
539 offset
&= ~PAGE_MASK
;
541 pages
+= PFN_UP(offset
+ size
);
547 static int xennet_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
550 struct netfront_info
*np
= netdev_priv(dev
);
551 struct netfront_stats
*stats
= this_cpu_ptr(np
->stats
);
552 struct xen_netif_tx_request
*tx
;
553 char *data
= skb
->data
;
559 unsigned int offset
= offset_in_page(data
);
560 unsigned int len
= skb_headlen(skb
);
563 /* If skb->len is too big for wire format, drop skb and alert
564 * user about misconfiguration.
566 if (unlikely(skb
->len
> XEN_NETIF_MAX_TX_SIZE
)) {
567 net_alert_ratelimited(
568 "xennet: skb->len = %u, too big for wire format\n",
573 slots
= DIV_ROUND_UP(offset
+ len
, PAGE_SIZE
) +
574 xennet_count_skb_frag_slots(skb
);
575 if (unlikely(slots
> MAX_SKB_FRAGS
+ 1)) {
576 net_alert_ratelimited(
577 "xennet: skb rides the rocket: %d slots\n", slots
);
581 spin_lock_irqsave(&np
->tx_lock
, flags
);
583 if (unlikely(!netif_carrier_ok(dev
) ||
584 (slots
> 1 && !xennet_can_sg(dev
)) ||
585 netif_needs_gso(skb
, netif_skb_features(skb
)))) {
586 spin_unlock_irqrestore(&np
->tx_lock
, flags
);
590 i
= np
->tx
.req_prod_pvt
;
592 id
= get_id_from_freelist(&np
->tx_skb_freelist
, np
->tx_skbs
);
593 np
->tx_skbs
[id
].skb
= skb
;
595 tx
= RING_GET_REQUEST(&np
->tx
, i
);
598 ref
= gnttab_claim_grant_reference(&np
->gref_tx_head
);
599 BUG_ON((signed short)ref
< 0);
600 mfn
= virt_to_mfn(data
);
601 gnttab_grant_foreign_access_ref(
602 ref
, np
->xbdev
->otherend_id
, mfn
, GNTMAP_readonly
);
603 np
->grant_tx_page
[id
] = virt_to_page(data
);
604 tx
->gref
= np
->grant_tx_ref
[id
] = ref
;
609 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
611 tx
->flags
|= XEN_NETTXF_csum_blank
| XEN_NETTXF_data_validated
;
612 else if (skb
->ip_summed
== CHECKSUM_UNNECESSARY
)
613 /* remote but checksummed. */
614 tx
->flags
|= XEN_NETTXF_data_validated
;
616 if (skb_shinfo(skb
)->gso_size
) {
617 struct xen_netif_extra_info
*gso
;
619 gso
= (struct xen_netif_extra_info
*)
620 RING_GET_REQUEST(&np
->tx
, ++i
);
622 tx
->flags
|= XEN_NETTXF_extra_info
;
624 gso
->u
.gso
.size
= skb_shinfo(skb
)->gso_size
;
625 gso
->u
.gso
.type
= (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCPV6
) ?
626 XEN_NETIF_GSO_TYPE_TCPV6
:
627 XEN_NETIF_GSO_TYPE_TCPV4
;
629 gso
->u
.gso
.features
= 0;
631 gso
->type
= XEN_NETIF_EXTRA_TYPE_GSO
;
635 np
->tx
.req_prod_pvt
= i
+ 1;
637 xennet_make_frags(skb
, dev
, tx
);
640 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np
->tx
, notify
);
642 notify_remote_via_irq(np
->tx_irq
);
644 u64_stats_update_begin(&stats
->syncp
);
645 stats
->tx_bytes
+= skb
->len
;
647 u64_stats_update_end(&stats
->syncp
);
649 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
650 xennet_tx_buf_gc(dev
);
652 if (!netfront_tx_slot_available(np
))
653 netif_stop_queue(dev
);
655 spin_unlock_irqrestore(&np
->tx_lock
, flags
);
660 dev
->stats
.tx_dropped
++;
665 static int xennet_close(struct net_device
*dev
)
667 struct netfront_info
*np
= netdev_priv(dev
);
668 netif_stop_queue(np
->netdev
);
669 napi_disable(&np
->napi
);
673 static void xennet_move_rx_slot(struct netfront_info
*np
, struct sk_buff
*skb
,
676 int new = xennet_rxidx(np
->rx
.req_prod_pvt
);
678 BUG_ON(np
->rx_skbs
[new]);
679 np
->rx_skbs
[new] = skb
;
680 np
->grant_rx_ref
[new] = ref
;
681 RING_GET_REQUEST(&np
->rx
, np
->rx
.req_prod_pvt
)->id
= new;
682 RING_GET_REQUEST(&np
->rx
, np
->rx
.req_prod_pvt
)->gref
= ref
;
683 np
->rx
.req_prod_pvt
++;
686 static int xennet_get_extras(struct netfront_info
*np
,
687 struct xen_netif_extra_info
*extras
,
691 struct xen_netif_extra_info
*extra
;
692 struct device
*dev
= &np
->netdev
->dev
;
693 RING_IDX cons
= np
->rx
.rsp_cons
;
700 if (unlikely(cons
+ 1 == rp
)) {
702 dev_warn(dev
, "Missing extra info\n");
707 extra
= (struct xen_netif_extra_info
*)
708 RING_GET_RESPONSE(&np
->rx
, ++cons
);
710 if (unlikely(!extra
->type
||
711 extra
->type
>= XEN_NETIF_EXTRA_TYPE_MAX
)) {
713 dev_warn(dev
, "Invalid extra type: %d\n",
717 memcpy(&extras
[extra
->type
- 1], extra
,
721 skb
= xennet_get_rx_skb(np
, cons
);
722 ref
= xennet_get_rx_ref(np
, cons
);
723 xennet_move_rx_slot(np
, skb
, ref
);
724 } while (extra
->flags
& XEN_NETIF_EXTRA_FLAG_MORE
);
726 np
->rx
.rsp_cons
= cons
;
730 static int xennet_get_responses(struct netfront_info
*np
,
731 struct netfront_rx_info
*rinfo
, RING_IDX rp
,
732 struct sk_buff_head
*list
)
734 struct xen_netif_rx_response
*rx
= &rinfo
->rx
;
735 struct xen_netif_extra_info
*extras
= rinfo
->extras
;
736 struct device
*dev
= &np
->netdev
->dev
;
737 RING_IDX cons
= np
->rx
.rsp_cons
;
738 struct sk_buff
*skb
= xennet_get_rx_skb(np
, cons
);
739 grant_ref_t ref
= xennet_get_rx_ref(np
, cons
);
740 int max
= MAX_SKB_FRAGS
+ (rx
->status
<= RX_COPY_THRESHOLD
);
745 if (rx
->flags
& XEN_NETRXF_extra_info
) {
746 err
= xennet_get_extras(np
, extras
, rp
);
747 cons
= np
->rx
.rsp_cons
;
751 if (unlikely(rx
->status
< 0 ||
752 rx
->offset
+ rx
->status
> PAGE_SIZE
)) {
754 dev_warn(dev
, "rx->offset: %x, size: %u\n",
755 rx
->offset
, rx
->status
);
756 xennet_move_rx_slot(np
, skb
, ref
);
762 * This definitely indicates a bug, either in this driver or in
763 * the backend driver. In future this should flag the bad
764 * situation to the system controller to reboot the backend.
766 if (ref
== GRANT_INVALID_REF
) {
768 dev_warn(dev
, "Bad rx response id %d.\n",
774 ret
= gnttab_end_foreign_access_ref(ref
, 0);
777 gnttab_release_grant_reference(&np
->gref_rx_head
, ref
);
779 __skb_queue_tail(list
, skb
);
782 if (!(rx
->flags
& XEN_NETRXF_more_data
))
785 if (cons
+ slots
== rp
) {
787 dev_warn(dev
, "Need more slots\n");
792 rx
= RING_GET_RESPONSE(&np
->rx
, cons
+ slots
);
793 skb
= xennet_get_rx_skb(np
, cons
+ slots
);
794 ref
= xennet_get_rx_ref(np
, cons
+ slots
);
798 if (unlikely(slots
> max
)) {
800 dev_warn(dev
, "Too many slots\n");
805 np
->rx
.rsp_cons
= cons
+ slots
;
810 static int xennet_set_skb_gso(struct sk_buff
*skb
,
811 struct xen_netif_extra_info
*gso
)
813 if (!gso
->u
.gso
.size
) {
815 pr_warn("GSO size must not be zero\n");
819 if (gso
->u
.gso
.type
!= XEN_NETIF_GSO_TYPE_TCPV4
&&
820 gso
->u
.gso
.type
!= XEN_NETIF_GSO_TYPE_TCPV6
) {
822 pr_warn("Bad GSO type %d\n", gso
->u
.gso
.type
);
826 skb_shinfo(skb
)->gso_size
= gso
->u
.gso
.size
;
827 skb_shinfo(skb
)->gso_type
=
828 (gso
->u
.gso
.type
== XEN_NETIF_GSO_TYPE_TCPV4
) ?
832 /* Header must be checked, and gso_segs computed. */
833 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
834 skb_shinfo(skb
)->gso_segs
= 0;
839 static RING_IDX
xennet_fill_frags(struct netfront_info
*np
,
841 struct sk_buff_head
*list
)
843 struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
844 RING_IDX cons
= np
->rx
.rsp_cons
;
845 struct sk_buff
*nskb
;
847 while ((nskb
= __skb_dequeue(list
))) {
848 struct xen_netif_rx_response
*rx
=
849 RING_GET_RESPONSE(&np
->rx
, ++cons
);
850 skb_frag_t
*nfrag
= &skb_shinfo(nskb
)->frags
[0];
852 if (shinfo
->nr_frags
== MAX_SKB_FRAGS
) {
853 unsigned int pull_to
= NETFRONT_SKB_CB(skb
)->pull_to
;
855 BUG_ON(pull_to
<= skb_headlen(skb
));
856 __pskb_pull_tail(skb
, pull_to
- skb_headlen(skb
));
858 BUG_ON(shinfo
->nr_frags
>= MAX_SKB_FRAGS
);
860 skb_add_rx_frag(skb
, shinfo
->nr_frags
, skb_frag_page(nfrag
),
861 rx
->offset
, rx
->status
, PAGE_SIZE
);
863 skb_shinfo(nskb
)->nr_frags
= 0;
870 static int checksum_setup(struct net_device
*dev
, struct sk_buff
*skb
)
872 bool recalculate_partial_csum
= false;
875 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
876 * peers can fail to set NETRXF_csum_blank when sending a GSO
877 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
878 * recalculate the partial checksum.
880 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
&& skb_is_gso(skb
)) {
881 struct netfront_info
*np
= netdev_priv(dev
);
882 np
->rx_gso_checksum_fixup
++;
883 skb
->ip_summed
= CHECKSUM_PARTIAL
;
884 recalculate_partial_csum
= true;
887 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
888 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
)
891 return skb_checksum_setup(skb
, recalculate_partial_csum
);
894 static int handle_incoming_queue(struct net_device
*dev
,
895 struct sk_buff_head
*rxq
)
897 struct netfront_info
*np
= netdev_priv(dev
);
898 struct netfront_stats
*stats
= this_cpu_ptr(np
->stats
);
899 int packets_dropped
= 0;
902 while ((skb
= __skb_dequeue(rxq
)) != NULL
) {
903 int pull_to
= NETFRONT_SKB_CB(skb
)->pull_to
;
905 if (pull_to
> skb_headlen(skb
))
906 __pskb_pull_tail(skb
, pull_to
- skb_headlen(skb
));
908 /* Ethernet work: Delayed to here as it peeks the header. */
909 skb
->protocol
= eth_type_trans(skb
, dev
);
910 skb_reset_network_header(skb
);
912 if (checksum_setup(dev
, skb
)) {
915 dev
->stats
.rx_errors
++;
919 u64_stats_update_begin(&stats
->syncp
);
921 stats
->rx_bytes
+= skb
->len
;
922 u64_stats_update_end(&stats
->syncp
);
925 napi_gro_receive(&np
->napi
, skb
);
928 return packets_dropped
;
931 static int xennet_poll(struct napi_struct
*napi
, int budget
)
933 struct netfront_info
*np
= container_of(napi
, struct netfront_info
, napi
);
934 struct net_device
*dev
= np
->netdev
;
936 struct netfront_rx_info rinfo
;
937 struct xen_netif_rx_response
*rx
= &rinfo
.rx
;
938 struct xen_netif_extra_info
*extras
= rinfo
.extras
;
941 struct sk_buff_head rxq
;
942 struct sk_buff_head errq
;
943 struct sk_buff_head tmpq
;
947 spin_lock(&np
->rx_lock
);
949 skb_queue_head_init(&rxq
);
950 skb_queue_head_init(&errq
);
951 skb_queue_head_init(&tmpq
);
953 rp
= np
->rx
.sring
->rsp_prod
;
954 rmb(); /* Ensure we see queued responses up to 'rp'. */
958 while ((i
!= rp
) && (work_done
< budget
)) {
959 memcpy(rx
, RING_GET_RESPONSE(&np
->rx
, i
), sizeof(*rx
));
960 memset(extras
, 0, sizeof(rinfo
.extras
));
962 err
= xennet_get_responses(np
, &rinfo
, rp
, &tmpq
);
966 while ((skb
= __skb_dequeue(&tmpq
)))
967 __skb_queue_tail(&errq
, skb
);
968 dev
->stats
.rx_errors
++;
973 skb
= __skb_dequeue(&tmpq
);
975 if (extras
[XEN_NETIF_EXTRA_TYPE_GSO
- 1].type
) {
976 struct xen_netif_extra_info
*gso
;
977 gso
= &extras
[XEN_NETIF_EXTRA_TYPE_GSO
- 1];
979 if (unlikely(xennet_set_skb_gso(skb
, gso
))) {
980 __skb_queue_head(&tmpq
, skb
);
981 np
->rx
.rsp_cons
+= skb_queue_len(&tmpq
);
986 NETFRONT_SKB_CB(skb
)->pull_to
= rx
->status
;
987 if (NETFRONT_SKB_CB(skb
)->pull_to
> RX_COPY_THRESHOLD
)
988 NETFRONT_SKB_CB(skb
)->pull_to
= RX_COPY_THRESHOLD
;
990 skb_shinfo(skb
)->frags
[0].page_offset
= rx
->offset
;
991 skb_frag_size_set(&skb_shinfo(skb
)->frags
[0], rx
->status
);
992 skb
->data_len
= rx
->status
;
993 skb
->len
+= rx
->status
;
995 i
= xennet_fill_frags(np
, skb
, &tmpq
);
997 if (rx
->flags
& XEN_NETRXF_csum_blank
)
998 skb
->ip_summed
= CHECKSUM_PARTIAL
;
999 else if (rx
->flags
& XEN_NETRXF_data_validated
)
1000 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1002 __skb_queue_tail(&rxq
, skb
);
1004 np
->rx
.rsp_cons
= ++i
;
1008 __skb_queue_purge(&errq
);
1010 work_done
-= handle_incoming_queue(dev
, &rxq
);
1012 /* If we get a callback with very few responses, reduce fill target. */
1013 /* NB. Note exponential increase, linear decrease. */
1014 if (((np
->rx
.req_prod_pvt
- np
->rx
.sring
->rsp_prod
) >
1015 ((3*np
->rx_target
) / 4)) &&
1016 (--np
->rx_target
< np
->rx_min_target
))
1017 np
->rx_target
= np
->rx_min_target
;
1019 xennet_alloc_rx_buffers(dev
);
1021 if (work_done
< budget
) {
1024 napi_gro_flush(napi
, false);
1026 local_irq_save(flags
);
1028 RING_FINAL_CHECK_FOR_RESPONSES(&np
->rx
, more_to_do
);
1030 __napi_complete(napi
);
1032 local_irq_restore(flags
);
1035 spin_unlock(&np
->rx_lock
);
1040 static int xennet_change_mtu(struct net_device
*dev
, int mtu
)
1042 int max
= xennet_can_sg(dev
) ?
1043 XEN_NETIF_MAX_TX_SIZE
- MAX_TCP_HEADER
: ETH_DATA_LEN
;
1051 static struct rtnl_link_stats64
*xennet_get_stats64(struct net_device
*dev
,
1052 struct rtnl_link_stats64
*tot
)
1054 struct netfront_info
*np
= netdev_priv(dev
);
1057 for_each_possible_cpu(cpu
) {
1058 struct netfront_stats
*stats
= per_cpu_ptr(np
->stats
, cpu
);
1059 u64 rx_packets
, rx_bytes
, tx_packets
, tx_bytes
;
1063 start
= u64_stats_fetch_begin_bh(&stats
->syncp
);
1065 rx_packets
= stats
->rx_packets
;
1066 tx_packets
= stats
->tx_packets
;
1067 rx_bytes
= stats
->rx_bytes
;
1068 tx_bytes
= stats
->tx_bytes
;
1069 } while (u64_stats_fetch_retry_bh(&stats
->syncp
, start
));
1071 tot
->rx_packets
+= rx_packets
;
1072 tot
->tx_packets
+= tx_packets
;
1073 tot
->rx_bytes
+= rx_bytes
;
1074 tot
->tx_bytes
+= tx_bytes
;
1077 tot
->rx_errors
= dev
->stats
.rx_errors
;
1078 tot
->tx_dropped
= dev
->stats
.tx_dropped
;
1083 static void xennet_release_tx_bufs(struct netfront_info
*np
)
1085 struct sk_buff
*skb
;
1088 for (i
= 0; i
< NET_TX_RING_SIZE
; i
++) {
1089 /* Skip over entries which are actually freelist references */
1090 if (skb_entry_is_link(&np
->tx_skbs
[i
]))
1093 skb
= np
->tx_skbs
[i
].skb
;
1094 get_page(np
->grant_tx_page
[i
]);
1095 gnttab_end_foreign_access(np
->grant_tx_ref
[i
],
1097 (unsigned long)page_address(np
->grant_tx_page
[i
]));
1098 np
->grant_tx_page
[i
] = NULL
;
1099 np
->grant_tx_ref
[i
] = GRANT_INVALID_REF
;
1100 add_id_to_freelist(&np
->tx_skb_freelist
, np
->tx_skbs
, i
);
1101 dev_kfree_skb_irq(skb
);
1105 static void xennet_release_rx_bufs(struct netfront_info
*np
)
1109 spin_lock_bh(&np
->rx_lock
);
1111 for (id
= 0; id
< NET_RX_RING_SIZE
; id
++) {
1112 struct sk_buff
*skb
;
1115 skb
= np
->rx_skbs
[id
];
1119 ref
= np
->grant_rx_ref
[id
];
1120 if (ref
== GRANT_INVALID_REF
)
1123 page
= skb_frag_page(&skb_shinfo(skb
)->frags
[0]);
1125 /* gnttab_end_foreign_access() needs a page ref until
1126 * foreign access is ended (which may be deferred).
1129 gnttab_end_foreign_access(ref
, 0,
1130 (unsigned long)page_address(page
));
1131 np
->grant_rx_ref
[id
] = GRANT_INVALID_REF
;
1136 spin_unlock_bh(&np
->rx_lock
);
1139 static void xennet_uninit(struct net_device
*dev
)
1141 struct netfront_info
*np
= netdev_priv(dev
);
1142 xennet_release_tx_bufs(np
);
1143 xennet_release_rx_bufs(np
);
1144 gnttab_free_grant_references(np
->gref_tx_head
);
1145 gnttab_free_grant_references(np
->gref_rx_head
);
1148 static netdev_features_t
xennet_fix_features(struct net_device
*dev
,
1149 netdev_features_t features
)
1151 struct netfront_info
*np
= netdev_priv(dev
);
1154 if (features
& NETIF_F_SG
) {
1155 if (xenbus_scanf(XBT_NIL
, np
->xbdev
->otherend
, "feature-sg",
1160 features
&= ~NETIF_F_SG
;
1163 if (features
& NETIF_F_IPV6_CSUM
) {
1164 if (xenbus_scanf(XBT_NIL
, np
->xbdev
->otherend
,
1165 "feature-ipv6-csum-offload", "%d", &val
) < 0)
1169 features
&= ~NETIF_F_IPV6_CSUM
;
1172 if (features
& NETIF_F_TSO
) {
1173 if (xenbus_scanf(XBT_NIL
, np
->xbdev
->otherend
,
1174 "feature-gso-tcpv4", "%d", &val
) < 0)
1178 features
&= ~NETIF_F_TSO
;
1181 if (features
& NETIF_F_TSO6
) {
1182 if (xenbus_scanf(XBT_NIL
, np
->xbdev
->otherend
,
1183 "feature-gso-tcpv6", "%d", &val
) < 0)
1187 features
&= ~NETIF_F_TSO6
;
1193 static int xennet_set_features(struct net_device
*dev
,
1194 netdev_features_t features
)
1196 if (!(features
& NETIF_F_SG
) && dev
->mtu
> ETH_DATA_LEN
) {
1197 netdev_info(dev
, "Reducing MTU because no SG offload");
1198 dev
->mtu
= ETH_DATA_LEN
;
1204 static irqreturn_t
xennet_tx_interrupt(int irq
, void *dev_id
)
1206 struct netfront_info
*np
= dev_id
;
1207 struct net_device
*dev
= np
->netdev
;
1208 unsigned long flags
;
1210 spin_lock_irqsave(&np
->tx_lock
, flags
);
1211 xennet_tx_buf_gc(dev
);
1212 spin_unlock_irqrestore(&np
->tx_lock
, flags
);
1217 static irqreturn_t
xennet_rx_interrupt(int irq
, void *dev_id
)
1219 struct netfront_info
*np
= dev_id
;
1220 struct net_device
*dev
= np
->netdev
;
1222 if (likely(netif_carrier_ok(dev
) &&
1223 RING_HAS_UNCONSUMED_RESPONSES(&np
->rx
)))
1224 napi_schedule(&np
->napi
);
1229 static irqreturn_t
xennet_interrupt(int irq
, void *dev_id
)
1231 xennet_tx_interrupt(irq
, dev_id
);
1232 xennet_rx_interrupt(irq
, dev_id
);
1236 #ifdef CONFIG_NET_POLL_CONTROLLER
1237 static void xennet_poll_controller(struct net_device
*dev
)
1239 xennet_interrupt(0, dev
);
1243 static const struct net_device_ops xennet_netdev_ops
= {
1244 .ndo_open
= xennet_open
,
1245 .ndo_uninit
= xennet_uninit
,
1246 .ndo_stop
= xennet_close
,
1247 .ndo_start_xmit
= xennet_start_xmit
,
1248 .ndo_change_mtu
= xennet_change_mtu
,
1249 .ndo_get_stats64
= xennet_get_stats64
,
1250 .ndo_set_mac_address
= eth_mac_addr
,
1251 .ndo_validate_addr
= eth_validate_addr
,
1252 .ndo_fix_features
= xennet_fix_features
,
1253 .ndo_set_features
= xennet_set_features
,
1254 #ifdef CONFIG_NET_POLL_CONTROLLER
1255 .ndo_poll_controller
= xennet_poll_controller
,
1259 static struct net_device
*xennet_create_dev(struct xenbus_device
*dev
)
1262 struct net_device
*netdev
;
1263 struct netfront_info
*np
;
1265 netdev
= alloc_etherdev(sizeof(struct netfront_info
));
1267 return ERR_PTR(-ENOMEM
);
1269 np
= netdev_priv(netdev
);
1272 spin_lock_init(&np
->tx_lock
);
1273 spin_lock_init(&np
->rx_lock
);
1275 skb_queue_head_init(&np
->rx_batch
);
1276 np
->rx_target
= RX_DFL_MIN_TARGET
;
1277 np
->rx_min_target
= RX_DFL_MIN_TARGET
;
1278 np
->rx_max_target
= RX_MAX_TARGET
;
1280 init_timer(&np
->rx_refill_timer
);
1281 np
->rx_refill_timer
.data
= (unsigned long)netdev
;
1282 np
->rx_refill_timer
.function
= rx_refill_timeout
;
1285 np
->stats
= alloc_percpu(struct netfront_stats
);
1286 if (np
->stats
== NULL
)
1289 for_each_possible_cpu(i
) {
1290 struct netfront_stats
*xen_nf_stats
;
1291 xen_nf_stats
= per_cpu_ptr(np
->stats
, i
);
1292 u64_stats_init(&xen_nf_stats
->syncp
);
1295 /* Initialise tx_skbs as a free chain containing every entry. */
1296 np
->tx_skb_freelist
= 0;
1297 for (i
= 0; i
< NET_TX_RING_SIZE
; i
++) {
1298 skb_entry_set_link(&np
->tx_skbs
[i
], i
+1);
1299 np
->grant_tx_ref
[i
] = GRANT_INVALID_REF
;
1302 /* Clear out rx_skbs */
1303 for (i
= 0; i
< NET_RX_RING_SIZE
; i
++) {
1304 np
->rx_skbs
[i
] = NULL
;
1305 np
->grant_rx_ref
[i
] = GRANT_INVALID_REF
;
1306 np
->grant_tx_page
[i
] = NULL
;
1309 /* A grant for every tx ring slot */
1310 if (gnttab_alloc_grant_references(TX_MAX_TARGET
,
1311 &np
->gref_tx_head
) < 0) {
1312 pr_alert("can't alloc tx grant refs\n");
1314 goto exit_free_stats
;
1316 /* A grant for every rx ring slot */
1317 if (gnttab_alloc_grant_references(RX_MAX_TARGET
,
1318 &np
->gref_rx_head
) < 0) {
1319 pr_alert("can't alloc rx grant refs\n");
1324 netdev
->netdev_ops
= &xennet_netdev_ops
;
1326 netif_napi_add(netdev
, &np
->napi
, xennet_poll
, 64);
1327 netdev
->features
= NETIF_F_IP_CSUM
| NETIF_F_RXCSUM
|
1329 netdev
->hw_features
= NETIF_F_SG
|
1331 NETIF_F_TSO
| NETIF_F_TSO6
;
1334 * Assume that all hw features are available for now. This set
1335 * will be adjusted by the call to netdev_update_features() in
1336 * xennet_connect() which is the earliest point where we can
1337 * negotiate with the backend regarding supported features.
1339 netdev
->features
|= netdev
->hw_features
;
1341 SET_ETHTOOL_OPS(netdev
, &xennet_ethtool_ops
);
1342 SET_NETDEV_DEV(netdev
, &dev
->dev
);
1344 netif_set_gso_max_size(netdev
, XEN_NETIF_MAX_TX_SIZE
- MAX_TCP_HEADER
);
1346 np
->netdev
= netdev
;
1348 netif_carrier_off(netdev
);
1353 gnttab_free_grant_references(np
->gref_tx_head
);
1355 free_percpu(np
->stats
);
1357 free_netdev(netdev
);
1358 return ERR_PTR(err
);
1362 * Entry point to this code when a new device is created. Allocate the basic
1363 * structures and the ring buffers for communication with the backend, and
1364 * inform the backend of the appropriate details for those.
1366 static int netfront_probe(struct xenbus_device
*dev
,
1367 const struct xenbus_device_id
*id
)
1370 struct net_device
*netdev
;
1371 struct netfront_info
*info
;
1373 netdev
= xennet_create_dev(dev
);
1374 if (IS_ERR(netdev
)) {
1375 err
= PTR_ERR(netdev
);
1376 xenbus_dev_fatal(dev
, err
, "creating netdev");
1380 info
= netdev_priv(netdev
);
1381 dev_set_drvdata(&dev
->dev
, info
);
1383 err
= register_netdev(info
->netdev
);
1385 pr_warn("%s: register_netdev err=%d\n", __func__
, err
);
1389 err
= xennet_sysfs_addif(info
->netdev
);
1391 unregister_netdev(info
->netdev
);
1392 pr_warn("%s: add sysfs failed err=%d\n", __func__
, err
);
1399 free_netdev(netdev
);
1400 dev_set_drvdata(&dev
->dev
, NULL
);
1404 static void xennet_end_access(int ref
, void *page
)
1406 /* This frees the page as a side-effect */
1407 if (ref
!= GRANT_INVALID_REF
)
1408 gnttab_end_foreign_access(ref
, 0, (unsigned long)page
);
1411 static void xennet_disconnect_backend(struct netfront_info
*info
)
1413 /* Stop old i/f to prevent errors whilst we rebuild the state. */
1414 spin_lock_bh(&info
->rx_lock
);
1415 spin_lock_irq(&info
->tx_lock
);
1416 netif_carrier_off(info
->netdev
);
1417 spin_unlock_irq(&info
->tx_lock
);
1418 spin_unlock_bh(&info
->rx_lock
);
1420 if (info
->tx_irq
&& (info
->tx_irq
== info
->rx_irq
))
1421 unbind_from_irqhandler(info
->tx_irq
, info
);
1422 if (info
->tx_irq
&& (info
->tx_irq
!= info
->rx_irq
)) {
1423 unbind_from_irqhandler(info
->tx_irq
, info
);
1424 unbind_from_irqhandler(info
->rx_irq
, info
);
1426 info
->tx_evtchn
= info
->rx_evtchn
= 0;
1427 info
->tx_irq
= info
->rx_irq
= 0;
1429 /* End access and free the pages */
1430 xennet_end_access(info
->tx_ring_ref
, info
->tx
.sring
);
1431 xennet_end_access(info
->rx_ring_ref
, info
->rx
.sring
);
1433 info
->tx_ring_ref
= GRANT_INVALID_REF
;
1434 info
->rx_ring_ref
= GRANT_INVALID_REF
;
1435 info
->tx
.sring
= NULL
;
1436 info
->rx
.sring
= NULL
;
1440 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1441 * driver restart. We tear down our netif structure and recreate it, but
1442 * leave the device-layer structures intact so that this is transparent to the
1443 * rest of the kernel.
1445 static int netfront_resume(struct xenbus_device
*dev
)
1447 struct netfront_info
*info
= dev_get_drvdata(&dev
->dev
);
1449 dev_dbg(&dev
->dev
, "%s\n", dev
->nodename
);
1451 xennet_disconnect_backend(info
);
1455 static int xen_net_read_mac(struct xenbus_device
*dev
, u8 mac
[])
1457 char *s
, *e
, *macstr
;
1460 macstr
= s
= xenbus_read(XBT_NIL
, dev
->nodename
, "mac", NULL
);
1462 return PTR_ERR(macstr
);
1464 for (i
= 0; i
< ETH_ALEN
; i
++) {
1465 mac
[i
] = simple_strtoul(s
, &e
, 16);
1466 if ((s
== e
) || (*e
!= ((i
== ETH_ALEN
-1) ? '\0' : ':'))) {
1477 static int setup_netfront_single(struct netfront_info
*info
)
1481 err
= xenbus_alloc_evtchn(info
->xbdev
, &info
->tx_evtchn
);
1485 err
= bind_evtchn_to_irqhandler(info
->tx_evtchn
,
1487 0, info
->netdev
->name
, info
);
1490 info
->rx_evtchn
= info
->tx_evtchn
;
1491 info
->rx_irq
= info
->tx_irq
= err
;
1496 xenbus_free_evtchn(info
->xbdev
, info
->tx_evtchn
);
1497 info
->tx_evtchn
= 0;
1502 static int setup_netfront_split(struct netfront_info
*info
)
1506 err
= xenbus_alloc_evtchn(info
->xbdev
, &info
->tx_evtchn
);
1509 err
= xenbus_alloc_evtchn(info
->xbdev
, &info
->rx_evtchn
);
1511 goto alloc_rx_evtchn_fail
;
1513 snprintf(info
->tx_irq_name
, sizeof(info
->tx_irq_name
),
1514 "%s-tx", info
->netdev
->name
);
1515 err
= bind_evtchn_to_irqhandler(info
->tx_evtchn
,
1516 xennet_tx_interrupt
,
1517 0, info
->tx_irq_name
, info
);
1522 snprintf(info
->rx_irq_name
, sizeof(info
->rx_irq_name
),
1523 "%s-rx", info
->netdev
->name
);
1524 err
= bind_evtchn_to_irqhandler(info
->rx_evtchn
,
1525 xennet_rx_interrupt
,
1526 0, info
->rx_irq_name
, info
);
1534 unbind_from_irqhandler(info
->tx_irq
, info
);
1537 xenbus_free_evtchn(info
->xbdev
, info
->rx_evtchn
);
1538 info
->rx_evtchn
= 0;
1539 alloc_rx_evtchn_fail
:
1540 xenbus_free_evtchn(info
->xbdev
, info
->tx_evtchn
);
1541 info
->tx_evtchn
= 0;
1546 static int setup_netfront(struct xenbus_device
*dev
, struct netfront_info
*info
)
1548 struct xen_netif_tx_sring
*txs
;
1549 struct xen_netif_rx_sring
*rxs
;
1551 struct net_device
*netdev
= info
->netdev
;
1552 unsigned int feature_split_evtchn
;
1554 info
->tx_ring_ref
= GRANT_INVALID_REF
;
1555 info
->rx_ring_ref
= GRANT_INVALID_REF
;
1556 info
->rx
.sring
= NULL
;
1557 info
->tx
.sring
= NULL
;
1560 err
= xenbus_scanf(XBT_NIL
, info
->xbdev
->otherend
,
1561 "feature-split-event-channels", "%u",
1562 &feature_split_evtchn
);
1564 feature_split_evtchn
= 0;
1566 err
= xen_net_read_mac(dev
, netdev
->dev_addr
);
1568 xenbus_dev_fatal(dev
, err
, "parsing %s/mac", dev
->nodename
);
1572 txs
= (struct xen_netif_tx_sring
*)get_zeroed_page(GFP_NOIO
| __GFP_HIGH
);
1575 xenbus_dev_fatal(dev
, err
, "allocating tx ring page");
1578 SHARED_RING_INIT(txs
);
1579 FRONT_RING_INIT(&info
->tx
, txs
, PAGE_SIZE
);
1581 err
= xenbus_grant_ring(dev
, virt_to_mfn(txs
));
1583 goto grant_tx_ring_fail
;
1585 info
->tx_ring_ref
= err
;
1586 rxs
= (struct xen_netif_rx_sring
*)get_zeroed_page(GFP_NOIO
| __GFP_HIGH
);
1589 xenbus_dev_fatal(dev
, err
, "allocating rx ring page");
1590 goto alloc_rx_ring_fail
;
1592 SHARED_RING_INIT(rxs
);
1593 FRONT_RING_INIT(&info
->rx
, rxs
, PAGE_SIZE
);
1595 err
= xenbus_grant_ring(dev
, virt_to_mfn(rxs
));
1597 goto grant_rx_ring_fail
;
1598 info
->rx_ring_ref
= err
;
1600 if (feature_split_evtchn
)
1601 err
= setup_netfront_split(info
);
1602 /* setup single event channel if
1603 * a) feature-split-event-channels == 0
1604 * b) feature-split-event-channels == 1 but failed to setup
1606 if (!feature_split_evtchn
|| (feature_split_evtchn
&& err
))
1607 err
= setup_netfront_single(info
);
1610 goto alloc_evtchn_fail
;
1614 /* If we fail to setup netfront, it is safe to just revoke access to
1615 * granted pages because backend is not accessing it at this point.
1618 gnttab_end_foreign_access_ref(info
->rx_ring_ref
, 0);
1620 free_page((unsigned long)rxs
);
1622 gnttab_end_foreign_access_ref(info
->tx_ring_ref
, 0);
1624 free_page((unsigned long)txs
);
1629 /* Common code used when first setting up, and when resuming. */
1630 static int talk_to_netback(struct xenbus_device
*dev
,
1631 struct netfront_info
*info
)
1633 const char *message
;
1634 struct xenbus_transaction xbt
;
1637 /* Create shared ring, alloc event channel. */
1638 err
= setup_netfront(dev
, info
);
1643 err
= xenbus_transaction_start(&xbt
);
1645 xenbus_dev_fatal(dev
, err
, "starting transaction");
1649 err
= xenbus_printf(xbt
, dev
->nodename
, "tx-ring-ref", "%u",
1652 message
= "writing tx ring-ref";
1653 goto abort_transaction
;
1655 err
= xenbus_printf(xbt
, dev
->nodename
, "rx-ring-ref", "%u",
1658 message
= "writing rx ring-ref";
1659 goto abort_transaction
;
1662 if (info
->tx_evtchn
== info
->rx_evtchn
) {
1663 err
= xenbus_printf(xbt
, dev
->nodename
,
1664 "event-channel", "%u", info
->tx_evtchn
);
1666 message
= "writing event-channel";
1667 goto abort_transaction
;
1670 err
= xenbus_printf(xbt
, dev
->nodename
,
1671 "event-channel-tx", "%u", info
->tx_evtchn
);
1673 message
= "writing event-channel-tx";
1674 goto abort_transaction
;
1676 err
= xenbus_printf(xbt
, dev
->nodename
,
1677 "event-channel-rx", "%u", info
->rx_evtchn
);
1679 message
= "writing event-channel-rx";
1680 goto abort_transaction
;
1684 err
= xenbus_printf(xbt
, dev
->nodename
, "request-rx-copy", "%u",
1687 message
= "writing request-rx-copy";
1688 goto abort_transaction
;
1691 err
= xenbus_printf(xbt
, dev
->nodename
, "feature-rx-notify", "%d", 1);
1693 message
= "writing feature-rx-notify";
1694 goto abort_transaction
;
1697 err
= xenbus_printf(xbt
, dev
->nodename
, "feature-sg", "%d", 1);
1699 message
= "writing feature-sg";
1700 goto abort_transaction
;
1703 err
= xenbus_printf(xbt
, dev
->nodename
, "feature-gso-tcpv4", "%d", 1);
1705 message
= "writing feature-gso-tcpv4";
1706 goto abort_transaction
;
1709 err
= xenbus_write(xbt
, dev
->nodename
, "feature-gso-tcpv6", "1");
1711 message
= "writing feature-gso-tcpv6";
1712 goto abort_transaction
;
1715 err
= xenbus_write(xbt
, dev
->nodename
, "feature-ipv6-csum-offload",
1718 message
= "writing feature-ipv6-csum-offload";
1719 goto abort_transaction
;
1722 err
= xenbus_transaction_end(xbt
, 0);
1726 xenbus_dev_fatal(dev
, err
, "completing transaction");
1733 xenbus_transaction_end(xbt
, 1);
1734 xenbus_dev_fatal(dev
, err
, "%s", message
);
1736 xennet_disconnect_backend(info
);
1741 static int xennet_connect(struct net_device
*dev
)
1743 struct netfront_info
*np
= netdev_priv(dev
);
1744 int i
, requeue_idx
, err
;
1745 struct sk_buff
*skb
;
1747 struct xen_netif_rx_request
*req
;
1748 unsigned int feature_rx_copy
;
1750 err
= xenbus_scanf(XBT_NIL
, np
->xbdev
->otherend
,
1751 "feature-rx-copy", "%u", &feature_rx_copy
);
1753 feature_rx_copy
= 0;
1755 if (!feature_rx_copy
) {
1757 "backend does not support copying receive path\n");
1761 err
= talk_to_netback(np
->xbdev
, np
);
1766 netdev_update_features(dev
);
1769 spin_lock_bh(&np
->rx_lock
);
1770 spin_lock_irq(&np
->tx_lock
);
1772 /* Step 1: Discard all pending TX packet fragments. */
1773 xennet_release_tx_bufs(np
);
1775 /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1776 for (requeue_idx
= 0, i
= 0; i
< NET_RX_RING_SIZE
; i
++) {
1778 const struct page
*page
;
1779 if (!np
->rx_skbs
[i
])
1782 skb
= np
->rx_skbs
[requeue_idx
] = xennet_get_rx_skb(np
, i
);
1783 ref
= np
->grant_rx_ref
[requeue_idx
] = xennet_get_rx_ref(np
, i
);
1784 req
= RING_GET_REQUEST(&np
->rx
, requeue_idx
);
1786 frag
= &skb_shinfo(skb
)->frags
[0];
1787 page
= skb_frag_page(frag
);
1788 gnttab_grant_foreign_access_ref(
1789 ref
, np
->xbdev
->otherend_id
,
1790 pfn_to_mfn(page_to_pfn(page
)),
1793 req
->id
= requeue_idx
;
1798 np
->rx
.req_prod_pvt
= requeue_idx
;
1801 * Step 3: All public and private state should now be sane. Get
1802 * ready to start sending and receiving packets and give the driver
1803 * domain a kick because we've probably just requeued some
1806 netif_carrier_on(np
->netdev
);
1807 notify_remote_via_irq(np
->tx_irq
);
1808 if (np
->tx_irq
!= np
->rx_irq
)
1809 notify_remote_via_irq(np
->rx_irq
);
1810 xennet_tx_buf_gc(dev
);
1811 xennet_alloc_rx_buffers(dev
);
1813 spin_unlock_irq(&np
->tx_lock
);
1814 spin_unlock_bh(&np
->rx_lock
);
1820 * Callback received when the backend's state changes.
1822 static void netback_changed(struct xenbus_device
*dev
,
1823 enum xenbus_state backend_state
)
1825 struct netfront_info
*np
= dev_get_drvdata(&dev
->dev
);
1826 struct net_device
*netdev
= np
->netdev
;
1828 dev_dbg(&dev
->dev
, "%s\n", xenbus_strstate(backend_state
));
1830 switch (backend_state
) {
1831 case XenbusStateInitialising
:
1832 case XenbusStateInitialised
:
1833 case XenbusStateReconfiguring
:
1834 case XenbusStateReconfigured
:
1835 case XenbusStateUnknown
:
1838 case XenbusStateInitWait
:
1839 if (dev
->state
!= XenbusStateInitialising
)
1841 if (xennet_connect(netdev
) != 0)
1843 xenbus_switch_state(dev
, XenbusStateConnected
);
1846 case XenbusStateConnected
:
1847 netdev_notify_peers(netdev
);
1850 case XenbusStateClosed
:
1851 if (dev
->state
== XenbusStateClosed
)
1853 /* Missed the backend's CLOSING state -- fallthrough */
1854 case XenbusStateClosing
:
1855 xenbus_frontend_closed(dev
);
1860 static const struct xennet_stat
{
1861 char name
[ETH_GSTRING_LEN
];
1863 } xennet_stats
[] = {
1865 "rx_gso_checksum_fixup",
1866 offsetof(struct netfront_info
, rx_gso_checksum_fixup
)
1870 static int xennet_get_sset_count(struct net_device
*dev
, int string_set
)
1872 switch (string_set
) {
1874 return ARRAY_SIZE(xennet_stats
);
1880 static void xennet_get_ethtool_stats(struct net_device
*dev
,
1881 struct ethtool_stats
*stats
, u64
* data
)
1883 void *np
= netdev_priv(dev
);
1886 for (i
= 0; i
< ARRAY_SIZE(xennet_stats
); i
++)
1887 data
[i
] = *(unsigned long *)(np
+ xennet_stats
[i
].offset
);
1890 static void xennet_get_strings(struct net_device
*dev
, u32 stringset
, u8
* data
)
1894 switch (stringset
) {
1896 for (i
= 0; i
< ARRAY_SIZE(xennet_stats
); i
++)
1897 memcpy(data
+ i
* ETH_GSTRING_LEN
,
1898 xennet_stats
[i
].name
, ETH_GSTRING_LEN
);
1903 static const struct ethtool_ops xennet_ethtool_ops
=
1905 .get_link
= ethtool_op_get_link
,
1907 .get_sset_count
= xennet_get_sset_count
,
1908 .get_ethtool_stats
= xennet_get_ethtool_stats
,
1909 .get_strings
= xennet_get_strings
,
1913 static ssize_t
show_rxbuf_min(struct device
*dev
,
1914 struct device_attribute
*attr
, char *buf
)
1916 struct net_device
*netdev
= to_net_dev(dev
);
1917 struct netfront_info
*info
= netdev_priv(netdev
);
1919 return sprintf(buf
, "%u\n", info
->rx_min_target
);
1922 static ssize_t
store_rxbuf_min(struct device
*dev
,
1923 struct device_attribute
*attr
,
1924 const char *buf
, size_t len
)
1926 struct net_device
*netdev
= to_net_dev(dev
);
1927 struct netfront_info
*np
= netdev_priv(netdev
);
1929 unsigned long target
;
1931 if (!capable(CAP_NET_ADMIN
))
1934 target
= simple_strtoul(buf
, &endp
, 0);
1938 if (target
< RX_MIN_TARGET
)
1939 target
= RX_MIN_TARGET
;
1940 if (target
> RX_MAX_TARGET
)
1941 target
= RX_MAX_TARGET
;
1943 spin_lock_bh(&np
->rx_lock
);
1944 if (target
> np
->rx_max_target
)
1945 np
->rx_max_target
= target
;
1946 np
->rx_min_target
= target
;
1947 if (target
> np
->rx_target
)
1948 np
->rx_target
= target
;
1950 xennet_alloc_rx_buffers(netdev
);
1952 spin_unlock_bh(&np
->rx_lock
);
1956 static ssize_t
show_rxbuf_max(struct device
*dev
,
1957 struct device_attribute
*attr
, char *buf
)
1959 struct net_device
*netdev
= to_net_dev(dev
);
1960 struct netfront_info
*info
= netdev_priv(netdev
);
1962 return sprintf(buf
, "%u\n", info
->rx_max_target
);
1965 static ssize_t
store_rxbuf_max(struct device
*dev
,
1966 struct device_attribute
*attr
,
1967 const char *buf
, size_t len
)
1969 struct net_device
*netdev
= to_net_dev(dev
);
1970 struct netfront_info
*np
= netdev_priv(netdev
);
1972 unsigned long target
;
1974 if (!capable(CAP_NET_ADMIN
))
1977 target
= simple_strtoul(buf
, &endp
, 0);
1981 if (target
< RX_MIN_TARGET
)
1982 target
= RX_MIN_TARGET
;
1983 if (target
> RX_MAX_TARGET
)
1984 target
= RX_MAX_TARGET
;
1986 spin_lock_bh(&np
->rx_lock
);
1987 if (target
< np
->rx_min_target
)
1988 np
->rx_min_target
= target
;
1989 np
->rx_max_target
= target
;
1990 if (target
< np
->rx_target
)
1991 np
->rx_target
= target
;
1993 xennet_alloc_rx_buffers(netdev
);
1995 spin_unlock_bh(&np
->rx_lock
);
1999 static ssize_t
show_rxbuf_cur(struct device
*dev
,
2000 struct device_attribute
*attr
, char *buf
)
2002 struct net_device
*netdev
= to_net_dev(dev
);
2003 struct netfront_info
*info
= netdev_priv(netdev
);
2005 return sprintf(buf
, "%u\n", info
->rx_target
);
2008 static struct device_attribute xennet_attrs
[] = {
2009 __ATTR(rxbuf_min
, S_IRUGO
|S_IWUSR
, show_rxbuf_min
, store_rxbuf_min
),
2010 __ATTR(rxbuf_max
, S_IRUGO
|S_IWUSR
, show_rxbuf_max
, store_rxbuf_max
),
2011 __ATTR(rxbuf_cur
, S_IRUGO
, show_rxbuf_cur
, NULL
),
2014 static int xennet_sysfs_addif(struct net_device
*netdev
)
2019 for (i
= 0; i
< ARRAY_SIZE(xennet_attrs
); i
++) {
2020 err
= device_create_file(&netdev
->dev
,
2029 device_remove_file(&netdev
->dev
, &xennet_attrs
[i
]);
2033 static void xennet_sysfs_delif(struct net_device
*netdev
)
2037 for (i
= 0; i
< ARRAY_SIZE(xennet_attrs
); i
++)
2038 device_remove_file(&netdev
->dev
, &xennet_attrs
[i
]);
2041 #endif /* CONFIG_SYSFS */
2043 static const struct xenbus_device_id netfront_ids
[] = {
2049 static int xennet_remove(struct xenbus_device
*dev
)
2051 struct netfront_info
*info
= dev_get_drvdata(&dev
->dev
);
2053 dev_dbg(&dev
->dev
, "%s\n", dev
->nodename
);
2055 xennet_disconnect_backend(info
);
2057 xennet_sysfs_delif(info
->netdev
);
2059 unregister_netdev(info
->netdev
);
2061 del_timer_sync(&info
->rx_refill_timer
);
2063 free_percpu(info
->stats
);
2065 free_netdev(info
->netdev
);
2070 static DEFINE_XENBUS_DRIVER(netfront
, ,
2071 .probe
= netfront_probe
,
2072 .remove
= xennet_remove
,
2073 .resume
= netfront_resume
,
2074 .otherend_changed
= netback_changed
,
2077 static int __init
netif_init(void)
2082 if (!xen_has_pv_nic_devices())
2085 pr_info("Initialising Xen virtual ethernet driver\n");
2087 return xenbus_register_frontend(&netfront_driver
);
2089 module_init(netif_init
);
2092 static void __exit
netif_exit(void)
2094 xenbus_unregister_driver(&netfront_driver
);
2096 module_exit(netif_exit
);
2098 MODULE_DESCRIPTION("Xen virtual network device frontend");
2099 MODULE_LICENSE("GPL");
2100 MODULE_ALIAS("xen:vif");
2101 MODULE_ALIAS("xennet");