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>
49 #include <xen/xenbus.h>
50 #include <xen/events.h>
52 #include <xen/platform_pci.h>
53 #include <xen/grant_table.h>
55 #include <xen/interface/io/netif.h>
56 #include <xen/interface/memory.h>
57 #include <xen/interface/grant_table.h>
59 /* Module parameters */
60 #define MAX_QUEUES_DEFAULT 8
61 static unsigned int xennet_max_queues
;
62 module_param_named(max_queues
, xennet_max_queues
, uint
, 0644);
63 MODULE_PARM_DESC(max_queues
,
64 "Maximum number of queues per virtual interface");
66 static const struct ethtool_ops xennet_ethtool_ops
;
72 #define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb))
74 #define RX_COPY_THRESHOLD 256
76 #define GRANT_INVALID_REF 0
78 #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, XEN_PAGE_SIZE)
79 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, XEN_PAGE_SIZE)
81 /* Minimum number of Rx slots (includes slot for GSO metadata). */
82 #define NET_RX_SLOTS_MIN (XEN_NETIF_NR_SLOTS_MIN + 1)
84 /* Queue name is interface name with "-qNNN" appended */
85 #define QUEUE_NAME_SIZE (IFNAMSIZ + 6)
87 /* IRQ name is queue name with "-tx" or "-rx" appended */
88 #define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3)
90 static DECLARE_WAIT_QUEUE_HEAD(module_unload_q
);
92 struct netfront_stats
{
95 struct u64_stats_sync syncp
;
100 struct netfront_queue
{
101 unsigned int id
; /* Queue ID, 0-based */
102 char name
[QUEUE_NAME_SIZE
]; /* DEVNAME-qN */
103 struct netfront_info
*info
;
105 struct napi_struct napi
;
107 /* Split event channels support, tx_* == rx_* when using
108 * single event channel.
110 unsigned int tx_evtchn
, rx_evtchn
;
111 unsigned int tx_irq
, rx_irq
;
112 /* Only used when split event channels support is enabled */
113 char tx_irq_name
[IRQ_NAME_SIZE
]; /* DEVNAME-qN-tx */
114 char rx_irq_name
[IRQ_NAME_SIZE
]; /* DEVNAME-qN-rx */
117 struct xen_netif_tx_front_ring tx
;
121 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
122 * are linked from tx_skb_freelist through skb_entry.link.
124 * NB. Freelist index entries are always going to be less than
125 * PAGE_OFFSET, whereas pointers to skbs will always be equal or
126 * greater than PAGE_OFFSET: we use this property to distinguish
132 } tx_skbs
[NET_TX_RING_SIZE
];
133 grant_ref_t gref_tx_head
;
134 grant_ref_t grant_tx_ref
[NET_TX_RING_SIZE
];
135 struct page
*grant_tx_page
[NET_TX_RING_SIZE
];
136 unsigned tx_skb_freelist
;
138 spinlock_t rx_lock ____cacheline_aligned_in_smp
;
139 struct xen_netif_rx_front_ring rx
;
142 struct timer_list rx_refill_timer
;
144 struct sk_buff
*rx_skbs
[NET_RX_RING_SIZE
];
145 grant_ref_t gref_rx_head
;
146 grant_ref_t grant_rx_ref
[NET_RX_RING_SIZE
];
149 struct netfront_info
{
150 struct list_head list
;
151 struct net_device
*netdev
;
153 struct xenbus_device
*xbdev
;
155 /* Multi-queue support */
156 struct netfront_queue
*queues
;
159 struct netfront_stats __percpu
*rx_stats
;
160 struct netfront_stats __percpu
*tx_stats
;
162 atomic_t rx_gso_checksum_fixup
;
165 struct netfront_rx_info
{
166 struct xen_netif_rx_response rx
;
167 struct xen_netif_extra_info extras
[XEN_NETIF_EXTRA_TYPE_MAX
- 1];
170 static void skb_entry_set_link(union skb_entry
*list
, unsigned short id
)
175 static int skb_entry_is_link(const union skb_entry
*list
)
177 BUILD_BUG_ON(sizeof(list
->skb
) != sizeof(list
->link
));
178 return (unsigned long)list
->skb
< PAGE_OFFSET
;
182 * Access macros for acquiring freeing slots in tx_skbs[].
185 static void add_id_to_freelist(unsigned *head
, union skb_entry
*list
,
188 skb_entry_set_link(&list
[id
], *head
);
192 static unsigned short get_id_from_freelist(unsigned *head
,
193 union skb_entry
*list
)
195 unsigned int id
= *head
;
196 *head
= list
[id
].link
;
200 static int xennet_rxidx(RING_IDX idx
)
202 return idx
& (NET_RX_RING_SIZE
- 1);
205 static struct sk_buff
*xennet_get_rx_skb(struct netfront_queue
*queue
,
208 int i
= xennet_rxidx(ri
);
209 struct sk_buff
*skb
= queue
->rx_skbs
[i
];
210 queue
->rx_skbs
[i
] = NULL
;
214 static grant_ref_t
xennet_get_rx_ref(struct netfront_queue
*queue
,
217 int i
= xennet_rxidx(ri
);
218 grant_ref_t ref
= queue
->grant_rx_ref
[i
];
219 queue
->grant_rx_ref
[i
] = GRANT_INVALID_REF
;
224 static const struct attribute_group xennet_dev_group
;
227 static bool xennet_can_sg(struct net_device
*dev
)
229 return dev
->features
& NETIF_F_SG
;
233 static void rx_refill_timeout(struct timer_list
*t
)
235 struct netfront_queue
*queue
= from_timer(queue
, t
, rx_refill_timer
);
236 napi_schedule(&queue
->napi
);
239 static int netfront_tx_slot_available(struct netfront_queue
*queue
)
241 return (queue
->tx
.req_prod_pvt
- queue
->tx
.rsp_cons
) <
242 (NET_TX_RING_SIZE
- MAX_SKB_FRAGS
- 2);
245 static void xennet_maybe_wake_tx(struct netfront_queue
*queue
)
247 struct net_device
*dev
= queue
->info
->netdev
;
248 struct netdev_queue
*dev_queue
= netdev_get_tx_queue(dev
, queue
->id
);
250 if (unlikely(netif_tx_queue_stopped(dev_queue
)) &&
251 netfront_tx_slot_available(queue
) &&
252 likely(netif_running(dev
)))
253 netif_tx_wake_queue(netdev_get_tx_queue(dev
, queue
->id
));
257 static struct sk_buff
*xennet_alloc_one_rx_buffer(struct netfront_queue
*queue
)
262 skb
= __netdev_alloc_skb(queue
->info
->netdev
,
263 RX_COPY_THRESHOLD
+ NET_IP_ALIGN
,
264 GFP_ATOMIC
| __GFP_NOWARN
);
268 page
= alloc_page(GFP_ATOMIC
| __GFP_NOWARN
);
273 skb_add_rx_frag(skb
, 0, page
, 0, 0, PAGE_SIZE
);
275 /* Align ip header to a 16 bytes boundary */
276 skb_reserve(skb
, NET_IP_ALIGN
);
277 skb
->dev
= queue
->info
->netdev
;
283 static void xennet_alloc_rx_buffers(struct netfront_queue
*queue
)
285 RING_IDX req_prod
= queue
->rx
.req_prod_pvt
;
289 if (unlikely(!netif_carrier_ok(queue
->info
->netdev
)))
292 for (req_prod
= queue
->rx
.req_prod_pvt
;
293 req_prod
- queue
->rx
.rsp_cons
< NET_RX_RING_SIZE
;
299 struct xen_netif_rx_request
*req
;
301 skb
= xennet_alloc_one_rx_buffer(queue
);
307 id
= xennet_rxidx(req_prod
);
309 BUG_ON(queue
->rx_skbs
[id
]);
310 queue
->rx_skbs
[id
] = skb
;
312 ref
= gnttab_claim_grant_reference(&queue
->gref_rx_head
);
313 WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref
));
314 queue
->grant_rx_ref
[id
] = ref
;
316 page
= skb_frag_page(&skb_shinfo(skb
)->frags
[0]);
318 req
= RING_GET_REQUEST(&queue
->rx
, req_prod
);
319 gnttab_page_grant_foreign_access_ref_one(ref
,
320 queue
->info
->xbdev
->otherend_id
,
327 queue
->rx
.req_prod_pvt
= req_prod
;
329 /* Try again later if there are not enough requests or skb allocation
331 * Enough requests is quantified as the sum of newly created slots and
332 * the unconsumed slots at the backend.
334 if (req_prod
- queue
->rx
.rsp_cons
< NET_RX_SLOTS_MIN
||
336 mod_timer(&queue
->rx_refill_timer
, jiffies
+ (HZ
/10));
340 wmb(); /* barrier so backend seens requests */
342 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue
->rx
, notify
);
344 notify_remote_via_irq(queue
->rx_irq
);
347 static int xennet_open(struct net_device
*dev
)
349 struct netfront_info
*np
= netdev_priv(dev
);
350 unsigned int num_queues
= dev
->real_num_tx_queues
;
352 struct netfront_queue
*queue
= NULL
;
354 for (i
= 0; i
< num_queues
; ++i
) {
355 queue
= &np
->queues
[i
];
356 napi_enable(&queue
->napi
);
358 spin_lock_bh(&queue
->rx_lock
);
359 if (netif_carrier_ok(dev
)) {
360 xennet_alloc_rx_buffers(queue
);
361 queue
->rx
.sring
->rsp_event
= queue
->rx
.rsp_cons
+ 1;
362 if (RING_HAS_UNCONSUMED_RESPONSES(&queue
->rx
))
363 napi_schedule(&queue
->napi
);
365 spin_unlock_bh(&queue
->rx_lock
);
368 netif_tx_start_all_queues(dev
);
373 static void xennet_tx_buf_gc(struct netfront_queue
*queue
)
380 BUG_ON(!netif_carrier_ok(queue
->info
->netdev
));
383 prod
= queue
->tx
.sring
->rsp_prod
;
384 rmb(); /* Ensure we see responses up to 'rp'. */
386 for (cons
= queue
->tx
.rsp_cons
; cons
!= prod
; cons
++) {
387 struct xen_netif_tx_response
*txrsp
;
389 txrsp
= RING_GET_RESPONSE(&queue
->tx
, cons
);
390 if (txrsp
->status
== XEN_NETIF_RSP_NULL
)
394 skb
= queue
->tx_skbs
[id
].skb
;
395 if (unlikely(gnttab_query_foreign_access(
396 queue
->grant_tx_ref
[id
]) != 0)) {
397 pr_alert("%s: warning -- grant still in use by backend domain\n",
401 gnttab_end_foreign_access_ref(
402 queue
->grant_tx_ref
[id
], GNTMAP_readonly
);
403 gnttab_release_grant_reference(
404 &queue
->gref_tx_head
, queue
->grant_tx_ref
[id
]);
405 queue
->grant_tx_ref
[id
] = GRANT_INVALID_REF
;
406 queue
->grant_tx_page
[id
] = NULL
;
407 add_id_to_freelist(&queue
->tx_skb_freelist
, queue
->tx_skbs
, id
);
408 dev_kfree_skb_irq(skb
);
411 queue
->tx
.rsp_cons
= prod
;
413 RING_FINAL_CHECK_FOR_RESPONSES(&queue
->tx
, more_to_do
);
414 } while (more_to_do
);
416 xennet_maybe_wake_tx(queue
);
419 struct xennet_gnttab_make_txreq
{
420 struct netfront_queue
*queue
;
423 struct xen_netif_tx_request
*tx
; /* Last request */
427 static void xennet_tx_setup_grant(unsigned long gfn
, unsigned int offset
,
428 unsigned int len
, void *data
)
430 struct xennet_gnttab_make_txreq
*info
= data
;
432 struct xen_netif_tx_request
*tx
;
434 /* convenient aliases */
435 struct page
*page
= info
->page
;
436 struct netfront_queue
*queue
= info
->queue
;
437 struct sk_buff
*skb
= info
->skb
;
439 id
= get_id_from_freelist(&queue
->tx_skb_freelist
, queue
->tx_skbs
);
440 tx
= RING_GET_REQUEST(&queue
->tx
, queue
->tx
.req_prod_pvt
++);
441 ref
= gnttab_claim_grant_reference(&queue
->gref_tx_head
);
442 WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref
));
444 gnttab_grant_foreign_access_ref(ref
, queue
->info
->xbdev
->otherend_id
,
445 gfn
, GNTMAP_readonly
);
447 queue
->tx_skbs
[id
].skb
= skb
;
448 queue
->grant_tx_page
[id
] = page
;
449 queue
->grant_tx_ref
[id
] = ref
;
458 info
->size
+= tx
->size
;
461 static struct xen_netif_tx_request
*xennet_make_first_txreq(
462 struct netfront_queue
*queue
, struct sk_buff
*skb
,
463 struct page
*page
, unsigned int offset
, unsigned int len
)
465 struct xennet_gnttab_make_txreq info
= {
472 gnttab_for_one_grant(page
, offset
, len
, xennet_tx_setup_grant
, &info
);
477 static void xennet_make_one_txreq(unsigned long gfn
, unsigned int offset
,
478 unsigned int len
, void *data
)
480 struct xennet_gnttab_make_txreq
*info
= data
;
482 info
->tx
->flags
|= XEN_NETTXF_more_data
;
484 xennet_tx_setup_grant(gfn
, offset
, len
, data
);
487 static struct xen_netif_tx_request
*xennet_make_txreqs(
488 struct netfront_queue
*queue
, struct xen_netif_tx_request
*tx
,
489 struct sk_buff
*skb
, struct page
*page
,
490 unsigned int offset
, unsigned int len
)
492 struct xennet_gnttab_make_txreq info
= {
498 /* Skip unused frames from start of page */
499 page
+= offset
>> PAGE_SHIFT
;
500 offset
&= ~PAGE_MASK
;
506 gnttab_foreach_grant_in_range(page
, offset
, len
,
507 xennet_make_one_txreq
,
519 * Count how many ring slots are required to send this skb. Each frag
520 * might be a compound page.
522 static int xennet_count_skb_slots(struct sk_buff
*skb
)
524 int i
, frags
= skb_shinfo(skb
)->nr_frags
;
527 slots
= gnttab_count_grant(offset_in_page(skb
->data
),
530 for (i
= 0; i
< frags
; i
++) {
531 skb_frag_t
*frag
= skb_shinfo(skb
)->frags
+ i
;
532 unsigned long size
= skb_frag_size(frag
);
533 unsigned long offset
= frag
->page_offset
;
535 /* Skip unused frames from start of page */
536 offset
&= ~PAGE_MASK
;
538 slots
+= gnttab_count_grant(offset
, size
);
544 static u16
xennet_select_queue(struct net_device
*dev
, struct sk_buff
*skb
,
545 void *accel_priv
, select_queue_fallback_t fallback
)
547 unsigned int num_queues
= dev
->real_num_tx_queues
;
551 /* First, check if there is only one queue */
552 if (num_queues
== 1) {
555 hash
= skb_get_hash(skb
);
556 queue_idx
= hash
% num_queues
;
562 #define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1)
564 static int xennet_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
566 struct netfront_info
*np
= netdev_priv(dev
);
567 struct netfront_stats
*tx_stats
= this_cpu_ptr(np
->tx_stats
);
568 struct xen_netif_tx_request
*tx
, *first_tx
;
576 struct netfront_queue
*queue
= NULL
;
577 unsigned int num_queues
= dev
->real_num_tx_queues
;
579 struct sk_buff
*nskb
;
581 /* Drop the packet if no queues are set up */
584 /* Determine which queue to transmit this SKB on */
585 queue_index
= skb_get_queue_mapping(skb
);
586 queue
= &np
->queues
[queue_index
];
588 /* If skb->len is too big for wire format, drop skb and alert
589 * user about misconfiguration.
591 if (unlikely(skb
->len
> XEN_NETIF_MAX_TX_SIZE
)) {
592 net_alert_ratelimited(
593 "xennet: skb->len = %u, too big for wire format\n",
598 slots
= xennet_count_skb_slots(skb
);
599 if (unlikely(slots
> MAX_XEN_SKB_FRAGS
+ 1)) {
600 net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n",
602 if (skb_linearize(skb
))
606 page
= virt_to_page(skb
->data
);
607 offset
= offset_in_page(skb
->data
);
609 /* The first req should be at least ETH_HLEN size or the packet will be
610 * dropped by netback.
612 if (unlikely(PAGE_SIZE
- offset
< ETH_HLEN
)) {
613 nskb
= skb_copy(skb
, GFP_ATOMIC
);
616 dev_consume_skb_any(skb
);
618 page
= virt_to_page(skb
->data
);
619 offset
= offset_in_page(skb
->data
);
622 len
= skb_headlen(skb
);
624 spin_lock_irqsave(&queue
->tx_lock
, flags
);
626 if (unlikely(!netif_carrier_ok(dev
) ||
627 (slots
> 1 && !xennet_can_sg(dev
)) ||
628 netif_needs_gso(skb
, netif_skb_features(skb
)))) {
629 spin_unlock_irqrestore(&queue
->tx_lock
, flags
);
633 /* First request for the linear area. */
634 first_tx
= tx
= xennet_make_first_txreq(queue
, skb
,
637 if (offset
== PAGE_SIZE
) {
643 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
645 tx
->flags
|= XEN_NETTXF_csum_blank
| XEN_NETTXF_data_validated
;
646 else if (skb
->ip_summed
== CHECKSUM_UNNECESSARY
)
647 /* remote but checksummed. */
648 tx
->flags
|= XEN_NETTXF_data_validated
;
650 /* Optional extra info after the first request. */
651 if (skb_shinfo(skb
)->gso_size
) {
652 struct xen_netif_extra_info
*gso
;
654 gso
= (struct xen_netif_extra_info
*)
655 RING_GET_REQUEST(&queue
->tx
, queue
->tx
.req_prod_pvt
++);
657 tx
->flags
|= XEN_NETTXF_extra_info
;
659 gso
->u
.gso
.size
= skb_shinfo(skb
)->gso_size
;
660 gso
->u
.gso
.type
= (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCPV6
) ?
661 XEN_NETIF_GSO_TYPE_TCPV6
:
662 XEN_NETIF_GSO_TYPE_TCPV4
;
664 gso
->u
.gso
.features
= 0;
666 gso
->type
= XEN_NETIF_EXTRA_TYPE_GSO
;
670 /* Requests for the rest of the linear area. */
671 tx
= xennet_make_txreqs(queue
, tx
, skb
, page
, offset
, len
);
673 /* Requests for all the frags. */
674 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
675 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
676 tx
= xennet_make_txreqs(queue
, tx
, skb
,
677 skb_frag_page(frag
), frag
->page_offset
,
678 skb_frag_size(frag
));
681 /* First request has the packet length. */
682 first_tx
->size
= skb
->len
;
684 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue
->tx
, notify
);
686 notify_remote_via_irq(queue
->tx_irq
);
688 u64_stats_update_begin(&tx_stats
->syncp
);
689 tx_stats
->bytes
+= skb
->len
;
691 u64_stats_update_end(&tx_stats
->syncp
);
693 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
694 xennet_tx_buf_gc(queue
);
696 if (!netfront_tx_slot_available(queue
))
697 netif_tx_stop_queue(netdev_get_tx_queue(dev
, queue
->id
));
699 spin_unlock_irqrestore(&queue
->tx_lock
, flags
);
704 dev
->stats
.tx_dropped
++;
705 dev_kfree_skb_any(skb
);
709 static int xennet_close(struct net_device
*dev
)
711 struct netfront_info
*np
= netdev_priv(dev
);
712 unsigned int num_queues
= dev
->real_num_tx_queues
;
714 struct netfront_queue
*queue
;
715 netif_tx_stop_all_queues(np
->netdev
);
716 for (i
= 0; i
< num_queues
; ++i
) {
717 queue
= &np
->queues
[i
];
718 napi_disable(&queue
->napi
);
723 static void xennet_move_rx_slot(struct netfront_queue
*queue
, struct sk_buff
*skb
,
726 int new = xennet_rxidx(queue
->rx
.req_prod_pvt
);
728 BUG_ON(queue
->rx_skbs
[new]);
729 queue
->rx_skbs
[new] = skb
;
730 queue
->grant_rx_ref
[new] = ref
;
731 RING_GET_REQUEST(&queue
->rx
, queue
->rx
.req_prod_pvt
)->id
= new;
732 RING_GET_REQUEST(&queue
->rx
, queue
->rx
.req_prod_pvt
)->gref
= ref
;
733 queue
->rx
.req_prod_pvt
++;
736 static int xennet_get_extras(struct netfront_queue
*queue
,
737 struct xen_netif_extra_info
*extras
,
741 struct xen_netif_extra_info
*extra
;
742 struct device
*dev
= &queue
->info
->netdev
->dev
;
743 RING_IDX cons
= queue
->rx
.rsp_cons
;
750 if (unlikely(cons
+ 1 == rp
)) {
752 dev_warn(dev
, "Missing extra info\n");
757 extra
= (struct xen_netif_extra_info
*)
758 RING_GET_RESPONSE(&queue
->rx
, ++cons
);
760 if (unlikely(!extra
->type
||
761 extra
->type
>= XEN_NETIF_EXTRA_TYPE_MAX
)) {
763 dev_warn(dev
, "Invalid extra type: %d\n",
767 memcpy(&extras
[extra
->type
- 1], extra
,
771 skb
= xennet_get_rx_skb(queue
, cons
);
772 ref
= xennet_get_rx_ref(queue
, cons
);
773 xennet_move_rx_slot(queue
, skb
, ref
);
774 } while (extra
->flags
& XEN_NETIF_EXTRA_FLAG_MORE
);
776 queue
->rx
.rsp_cons
= cons
;
780 static int xennet_get_responses(struct netfront_queue
*queue
,
781 struct netfront_rx_info
*rinfo
, RING_IDX rp
,
782 struct sk_buff_head
*list
)
784 struct xen_netif_rx_response
*rx
= &rinfo
->rx
;
785 struct xen_netif_extra_info
*extras
= rinfo
->extras
;
786 struct device
*dev
= &queue
->info
->netdev
->dev
;
787 RING_IDX cons
= queue
->rx
.rsp_cons
;
788 struct sk_buff
*skb
= xennet_get_rx_skb(queue
, cons
);
789 grant_ref_t ref
= xennet_get_rx_ref(queue
, cons
);
790 int max
= MAX_SKB_FRAGS
+ (rx
->status
<= RX_COPY_THRESHOLD
);
795 if (rx
->flags
& XEN_NETRXF_extra_info
) {
796 err
= xennet_get_extras(queue
, extras
, rp
);
797 cons
= queue
->rx
.rsp_cons
;
801 if (unlikely(rx
->status
< 0 ||
802 rx
->offset
+ rx
->status
> XEN_PAGE_SIZE
)) {
804 dev_warn(dev
, "rx->offset: %u, size: %d\n",
805 rx
->offset
, rx
->status
);
806 xennet_move_rx_slot(queue
, skb
, ref
);
812 * This definitely indicates a bug, either in this driver or in
813 * the backend driver. In future this should flag the bad
814 * situation to the system controller to reboot the backend.
816 if (ref
== GRANT_INVALID_REF
) {
818 dev_warn(dev
, "Bad rx response id %d.\n",
824 ret
= gnttab_end_foreign_access_ref(ref
, 0);
827 gnttab_release_grant_reference(&queue
->gref_rx_head
, ref
);
829 __skb_queue_tail(list
, skb
);
832 if (!(rx
->flags
& XEN_NETRXF_more_data
))
835 if (cons
+ slots
== rp
) {
837 dev_warn(dev
, "Need more slots\n");
842 rx
= RING_GET_RESPONSE(&queue
->rx
, cons
+ slots
);
843 skb
= xennet_get_rx_skb(queue
, cons
+ slots
);
844 ref
= xennet_get_rx_ref(queue
, cons
+ slots
);
848 if (unlikely(slots
> max
)) {
850 dev_warn(dev
, "Too many slots\n");
855 queue
->rx
.rsp_cons
= cons
+ slots
;
860 static int xennet_set_skb_gso(struct sk_buff
*skb
,
861 struct xen_netif_extra_info
*gso
)
863 if (!gso
->u
.gso
.size
) {
865 pr_warn("GSO size must not be zero\n");
869 if (gso
->u
.gso
.type
!= XEN_NETIF_GSO_TYPE_TCPV4
&&
870 gso
->u
.gso
.type
!= XEN_NETIF_GSO_TYPE_TCPV6
) {
872 pr_warn("Bad GSO type %d\n", gso
->u
.gso
.type
);
876 skb_shinfo(skb
)->gso_size
= gso
->u
.gso
.size
;
877 skb_shinfo(skb
)->gso_type
=
878 (gso
->u
.gso
.type
== XEN_NETIF_GSO_TYPE_TCPV4
) ?
882 /* Header must be checked, and gso_segs computed. */
883 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
884 skb_shinfo(skb
)->gso_segs
= 0;
889 static RING_IDX
xennet_fill_frags(struct netfront_queue
*queue
,
891 struct sk_buff_head
*list
)
893 struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
894 RING_IDX cons
= queue
->rx
.rsp_cons
;
895 struct sk_buff
*nskb
;
897 while ((nskb
= __skb_dequeue(list
))) {
898 struct xen_netif_rx_response
*rx
=
899 RING_GET_RESPONSE(&queue
->rx
, ++cons
);
900 skb_frag_t
*nfrag
= &skb_shinfo(nskb
)->frags
[0];
902 if (shinfo
->nr_frags
== MAX_SKB_FRAGS
) {
903 unsigned int pull_to
= NETFRONT_SKB_CB(skb
)->pull_to
;
905 BUG_ON(pull_to
<= skb_headlen(skb
));
906 __pskb_pull_tail(skb
, pull_to
- skb_headlen(skb
));
908 BUG_ON(shinfo
->nr_frags
>= MAX_SKB_FRAGS
);
910 skb_add_rx_frag(skb
, shinfo
->nr_frags
, skb_frag_page(nfrag
),
911 rx
->offset
, rx
->status
, PAGE_SIZE
);
913 skb_shinfo(nskb
)->nr_frags
= 0;
920 static int checksum_setup(struct net_device
*dev
, struct sk_buff
*skb
)
922 bool recalculate_partial_csum
= false;
925 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
926 * peers can fail to set NETRXF_csum_blank when sending a GSO
927 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
928 * recalculate the partial checksum.
930 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
&& skb_is_gso(skb
)) {
931 struct netfront_info
*np
= netdev_priv(dev
);
932 atomic_inc(&np
->rx_gso_checksum_fixup
);
933 skb
->ip_summed
= CHECKSUM_PARTIAL
;
934 recalculate_partial_csum
= true;
937 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
938 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
)
941 return skb_checksum_setup(skb
, recalculate_partial_csum
);
944 static int handle_incoming_queue(struct netfront_queue
*queue
,
945 struct sk_buff_head
*rxq
)
947 struct netfront_stats
*rx_stats
= this_cpu_ptr(queue
->info
->rx_stats
);
948 int packets_dropped
= 0;
951 while ((skb
= __skb_dequeue(rxq
)) != NULL
) {
952 int pull_to
= NETFRONT_SKB_CB(skb
)->pull_to
;
954 if (pull_to
> skb_headlen(skb
))
955 __pskb_pull_tail(skb
, pull_to
- skb_headlen(skb
));
957 /* Ethernet work: Delayed to here as it peeks the header. */
958 skb
->protocol
= eth_type_trans(skb
, queue
->info
->netdev
);
959 skb_reset_network_header(skb
);
961 if (checksum_setup(queue
->info
->netdev
, skb
)) {
964 queue
->info
->netdev
->stats
.rx_errors
++;
968 u64_stats_update_begin(&rx_stats
->syncp
);
970 rx_stats
->bytes
+= skb
->len
;
971 u64_stats_update_end(&rx_stats
->syncp
);
974 napi_gro_receive(&queue
->napi
, skb
);
977 return packets_dropped
;
980 static int xennet_poll(struct napi_struct
*napi
, int budget
)
982 struct netfront_queue
*queue
= container_of(napi
, struct netfront_queue
, napi
);
983 struct net_device
*dev
= queue
->info
->netdev
;
985 struct netfront_rx_info rinfo
;
986 struct xen_netif_rx_response
*rx
= &rinfo
.rx
;
987 struct xen_netif_extra_info
*extras
= rinfo
.extras
;
990 struct sk_buff_head rxq
;
991 struct sk_buff_head errq
;
992 struct sk_buff_head tmpq
;
995 spin_lock(&queue
->rx_lock
);
997 skb_queue_head_init(&rxq
);
998 skb_queue_head_init(&errq
);
999 skb_queue_head_init(&tmpq
);
1001 rp
= queue
->rx
.sring
->rsp_prod
;
1002 rmb(); /* Ensure we see queued responses up to 'rp'. */
1004 i
= queue
->rx
.rsp_cons
;
1006 while ((i
!= rp
) && (work_done
< budget
)) {
1007 memcpy(rx
, RING_GET_RESPONSE(&queue
->rx
, i
), sizeof(*rx
));
1008 memset(extras
, 0, sizeof(rinfo
.extras
));
1010 err
= xennet_get_responses(queue
, &rinfo
, rp
, &tmpq
);
1012 if (unlikely(err
)) {
1014 while ((skb
= __skb_dequeue(&tmpq
)))
1015 __skb_queue_tail(&errq
, skb
);
1016 dev
->stats
.rx_errors
++;
1017 i
= queue
->rx
.rsp_cons
;
1021 skb
= __skb_dequeue(&tmpq
);
1023 if (extras
[XEN_NETIF_EXTRA_TYPE_GSO
- 1].type
) {
1024 struct xen_netif_extra_info
*gso
;
1025 gso
= &extras
[XEN_NETIF_EXTRA_TYPE_GSO
- 1];
1027 if (unlikely(xennet_set_skb_gso(skb
, gso
))) {
1028 __skb_queue_head(&tmpq
, skb
);
1029 queue
->rx
.rsp_cons
+= skb_queue_len(&tmpq
);
1034 NETFRONT_SKB_CB(skb
)->pull_to
= rx
->status
;
1035 if (NETFRONT_SKB_CB(skb
)->pull_to
> RX_COPY_THRESHOLD
)
1036 NETFRONT_SKB_CB(skb
)->pull_to
= RX_COPY_THRESHOLD
;
1038 skb_shinfo(skb
)->frags
[0].page_offset
= rx
->offset
;
1039 skb_frag_size_set(&skb_shinfo(skb
)->frags
[0], rx
->status
);
1040 skb
->data_len
= rx
->status
;
1041 skb
->len
+= rx
->status
;
1043 i
= xennet_fill_frags(queue
, skb
, &tmpq
);
1045 if (rx
->flags
& XEN_NETRXF_csum_blank
)
1046 skb
->ip_summed
= CHECKSUM_PARTIAL
;
1047 else if (rx
->flags
& XEN_NETRXF_data_validated
)
1048 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1050 __skb_queue_tail(&rxq
, skb
);
1052 queue
->rx
.rsp_cons
= ++i
;
1056 __skb_queue_purge(&errq
);
1058 work_done
-= handle_incoming_queue(queue
, &rxq
);
1060 xennet_alloc_rx_buffers(queue
);
1062 if (work_done
< budget
) {
1065 napi_complete_done(napi
, work_done
);
1067 RING_FINAL_CHECK_FOR_RESPONSES(&queue
->rx
, more_to_do
);
1069 napi_schedule(napi
);
1072 spin_unlock(&queue
->rx_lock
);
1077 static int xennet_change_mtu(struct net_device
*dev
, int mtu
)
1079 int max
= xennet_can_sg(dev
) ? XEN_NETIF_MAX_TX_SIZE
: ETH_DATA_LEN
;
1087 static void xennet_get_stats64(struct net_device
*dev
,
1088 struct rtnl_link_stats64
*tot
)
1090 struct netfront_info
*np
= netdev_priv(dev
);
1093 for_each_possible_cpu(cpu
) {
1094 struct netfront_stats
*rx_stats
= per_cpu_ptr(np
->rx_stats
, cpu
);
1095 struct netfront_stats
*tx_stats
= per_cpu_ptr(np
->tx_stats
, cpu
);
1096 u64 rx_packets
, rx_bytes
, tx_packets
, tx_bytes
;
1100 start
= u64_stats_fetch_begin_irq(&tx_stats
->syncp
);
1101 tx_packets
= tx_stats
->packets
;
1102 tx_bytes
= tx_stats
->bytes
;
1103 } while (u64_stats_fetch_retry_irq(&tx_stats
->syncp
, start
));
1106 start
= u64_stats_fetch_begin_irq(&rx_stats
->syncp
);
1107 rx_packets
= rx_stats
->packets
;
1108 rx_bytes
= rx_stats
->bytes
;
1109 } while (u64_stats_fetch_retry_irq(&rx_stats
->syncp
, start
));
1111 tot
->rx_packets
+= rx_packets
;
1112 tot
->tx_packets
+= tx_packets
;
1113 tot
->rx_bytes
+= rx_bytes
;
1114 tot
->tx_bytes
+= tx_bytes
;
1117 tot
->rx_errors
= dev
->stats
.rx_errors
;
1118 tot
->tx_dropped
= dev
->stats
.tx_dropped
;
1121 static void xennet_release_tx_bufs(struct netfront_queue
*queue
)
1123 struct sk_buff
*skb
;
1126 for (i
= 0; i
< NET_TX_RING_SIZE
; i
++) {
1127 /* Skip over entries which are actually freelist references */
1128 if (skb_entry_is_link(&queue
->tx_skbs
[i
]))
1131 skb
= queue
->tx_skbs
[i
].skb
;
1132 get_page(queue
->grant_tx_page
[i
]);
1133 gnttab_end_foreign_access(queue
->grant_tx_ref
[i
],
1135 (unsigned long)page_address(queue
->grant_tx_page
[i
]));
1136 queue
->grant_tx_page
[i
] = NULL
;
1137 queue
->grant_tx_ref
[i
] = GRANT_INVALID_REF
;
1138 add_id_to_freelist(&queue
->tx_skb_freelist
, queue
->tx_skbs
, i
);
1139 dev_kfree_skb_irq(skb
);
1143 static void xennet_release_rx_bufs(struct netfront_queue
*queue
)
1147 spin_lock_bh(&queue
->rx_lock
);
1149 for (id
= 0; id
< NET_RX_RING_SIZE
; id
++) {
1150 struct sk_buff
*skb
;
1153 skb
= queue
->rx_skbs
[id
];
1157 ref
= queue
->grant_rx_ref
[id
];
1158 if (ref
== GRANT_INVALID_REF
)
1161 page
= skb_frag_page(&skb_shinfo(skb
)->frags
[0]);
1163 /* gnttab_end_foreign_access() needs a page ref until
1164 * foreign access is ended (which may be deferred).
1167 gnttab_end_foreign_access(ref
, 0,
1168 (unsigned long)page_address(page
));
1169 queue
->grant_rx_ref
[id
] = GRANT_INVALID_REF
;
1174 spin_unlock_bh(&queue
->rx_lock
);
1177 static netdev_features_t
xennet_fix_features(struct net_device
*dev
,
1178 netdev_features_t features
)
1180 struct netfront_info
*np
= netdev_priv(dev
);
1182 if (features
& NETIF_F_SG
&&
1183 !xenbus_read_unsigned(np
->xbdev
->otherend
, "feature-sg", 0))
1184 features
&= ~NETIF_F_SG
;
1186 if (features
& NETIF_F_IPV6_CSUM
&&
1187 !xenbus_read_unsigned(np
->xbdev
->otherend
,
1188 "feature-ipv6-csum-offload", 0))
1189 features
&= ~NETIF_F_IPV6_CSUM
;
1191 if (features
& NETIF_F_TSO
&&
1192 !xenbus_read_unsigned(np
->xbdev
->otherend
, "feature-gso-tcpv4", 0))
1193 features
&= ~NETIF_F_TSO
;
1195 if (features
& NETIF_F_TSO6
&&
1196 !xenbus_read_unsigned(np
->xbdev
->otherend
, "feature-gso-tcpv6", 0))
1197 features
&= ~NETIF_F_TSO6
;
1202 static int xennet_set_features(struct net_device
*dev
,
1203 netdev_features_t features
)
1205 if (!(features
& NETIF_F_SG
) && dev
->mtu
> ETH_DATA_LEN
) {
1206 netdev_info(dev
, "Reducing MTU because no SG offload");
1207 dev
->mtu
= ETH_DATA_LEN
;
1213 static irqreturn_t
xennet_tx_interrupt(int irq
, void *dev_id
)
1215 struct netfront_queue
*queue
= dev_id
;
1216 unsigned long flags
;
1218 spin_lock_irqsave(&queue
->tx_lock
, flags
);
1219 xennet_tx_buf_gc(queue
);
1220 spin_unlock_irqrestore(&queue
->tx_lock
, flags
);
1225 static irqreturn_t
xennet_rx_interrupt(int irq
, void *dev_id
)
1227 struct netfront_queue
*queue
= dev_id
;
1228 struct net_device
*dev
= queue
->info
->netdev
;
1230 if (likely(netif_carrier_ok(dev
) &&
1231 RING_HAS_UNCONSUMED_RESPONSES(&queue
->rx
)))
1232 napi_schedule(&queue
->napi
);
1237 static irqreturn_t
xennet_interrupt(int irq
, void *dev_id
)
1239 xennet_tx_interrupt(irq
, dev_id
);
1240 xennet_rx_interrupt(irq
, dev_id
);
1244 #ifdef CONFIG_NET_POLL_CONTROLLER
1245 static void xennet_poll_controller(struct net_device
*dev
)
1247 /* Poll each queue */
1248 struct netfront_info
*info
= netdev_priv(dev
);
1249 unsigned int num_queues
= dev
->real_num_tx_queues
;
1251 for (i
= 0; i
< num_queues
; ++i
)
1252 xennet_interrupt(0, &info
->queues
[i
]);
1256 static const struct net_device_ops xennet_netdev_ops
= {
1257 .ndo_open
= xennet_open
,
1258 .ndo_stop
= xennet_close
,
1259 .ndo_start_xmit
= xennet_start_xmit
,
1260 .ndo_change_mtu
= xennet_change_mtu
,
1261 .ndo_get_stats64
= xennet_get_stats64
,
1262 .ndo_set_mac_address
= eth_mac_addr
,
1263 .ndo_validate_addr
= eth_validate_addr
,
1264 .ndo_fix_features
= xennet_fix_features
,
1265 .ndo_set_features
= xennet_set_features
,
1266 .ndo_select_queue
= xennet_select_queue
,
1267 #ifdef CONFIG_NET_POLL_CONTROLLER
1268 .ndo_poll_controller
= xennet_poll_controller
,
1272 static void xennet_free_netdev(struct net_device
*netdev
)
1274 struct netfront_info
*np
= netdev_priv(netdev
);
1276 free_percpu(np
->rx_stats
);
1277 free_percpu(np
->tx_stats
);
1278 free_netdev(netdev
);
1281 static struct net_device
*xennet_create_dev(struct xenbus_device
*dev
)
1284 struct net_device
*netdev
;
1285 struct netfront_info
*np
;
1287 netdev
= alloc_etherdev_mq(sizeof(struct netfront_info
), xennet_max_queues
);
1289 return ERR_PTR(-ENOMEM
);
1291 np
= netdev_priv(netdev
);
1297 np
->rx_stats
= netdev_alloc_pcpu_stats(struct netfront_stats
);
1298 if (np
->rx_stats
== NULL
)
1300 np
->tx_stats
= netdev_alloc_pcpu_stats(struct netfront_stats
);
1301 if (np
->tx_stats
== NULL
)
1304 netdev
->netdev_ops
= &xennet_netdev_ops
;
1306 netdev
->features
= NETIF_F_IP_CSUM
| NETIF_F_RXCSUM
|
1308 netdev
->hw_features
= NETIF_F_SG
|
1310 NETIF_F_TSO
| NETIF_F_TSO6
;
1313 * Assume that all hw features are available for now. This set
1314 * will be adjusted by the call to netdev_update_features() in
1315 * xennet_connect() which is the earliest point where we can
1316 * negotiate with the backend regarding supported features.
1318 netdev
->features
|= netdev
->hw_features
;
1320 netdev
->ethtool_ops
= &xennet_ethtool_ops
;
1321 netdev
->min_mtu
= ETH_MIN_MTU
;
1322 netdev
->max_mtu
= XEN_NETIF_MAX_TX_SIZE
;
1323 SET_NETDEV_DEV(netdev
, &dev
->dev
);
1325 np
->netdev
= netdev
;
1327 netif_carrier_off(netdev
);
1329 xenbus_switch_state(dev
, XenbusStateInitialising
);
1333 xennet_free_netdev(netdev
);
1334 return ERR_PTR(err
);
1338 * Entry point to this code when a new device is created. Allocate the basic
1339 * structures and the ring buffers for communication with the backend, and
1340 * inform the backend of the appropriate details for those.
1342 static int netfront_probe(struct xenbus_device
*dev
,
1343 const struct xenbus_device_id
*id
)
1346 struct net_device
*netdev
;
1347 struct netfront_info
*info
;
1349 netdev
= xennet_create_dev(dev
);
1350 if (IS_ERR(netdev
)) {
1351 err
= PTR_ERR(netdev
);
1352 xenbus_dev_fatal(dev
, err
, "creating netdev");
1356 info
= netdev_priv(netdev
);
1357 dev_set_drvdata(&dev
->dev
, info
);
1359 info
->netdev
->sysfs_groups
[0] = &xennet_dev_group
;
1361 err
= register_netdev(info
->netdev
);
1363 pr_warn("%s: register_netdev err=%d\n", __func__
, err
);
1370 xennet_free_netdev(netdev
);
1371 dev_set_drvdata(&dev
->dev
, NULL
);
1375 static void xennet_end_access(int ref
, void *page
)
1377 /* This frees the page as a side-effect */
1378 if (ref
!= GRANT_INVALID_REF
)
1379 gnttab_end_foreign_access(ref
, 0, (unsigned long)page
);
1382 static void xennet_disconnect_backend(struct netfront_info
*info
)
1385 unsigned int num_queues
= info
->netdev
->real_num_tx_queues
;
1387 netif_carrier_off(info
->netdev
);
1389 for (i
= 0; i
< num_queues
&& info
->queues
; ++i
) {
1390 struct netfront_queue
*queue
= &info
->queues
[i
];
1392 del_timer_sync(&queue
->rx_refill_timer
);
1394 if (queue
->tx_irq
&& (queue
->tx_irq
== queue
->rx_irq
))
1395 unbind_from_irqhandler(queue
->tx_irq
, queue
);
1396 if (queue
->tx_irq
&& (queue
->tx_irq
!= queue
->rx_irq
)) {
1397 unbind_from_irqhandler(queue
->tx_irq
, queue
);
1398 unbind_from_irqhandler(queue
->rx_irq
, queue
);
1400 queue
->tx_evtchn
= queue
->rx_evtchn
= 0;
1401 queue
->tx_irq
= queue
->rx_irq
= 0;
1403 if (netif_running(info
->netdev
))
1404 napi_synchronize(&queue
->napi
);
1406 xennet_release_tx_bufs(queue
);
1407 xennet_release_rx_bufs(queue
);
1408 gnttab_free_grant_references(queue
->gref_tx_head
);
1409 gnttab_free_grant_references(queue
->gref_rx_head
);
1411 /* End access and free the pages */
1412 xennet_end_access(queue
->tx_ring_ref
, queue
->tx
.sring
);
1413 xennet_end_access(queue
->rx_ring_ref
, queue
->rx
.sring
);
1415 queue
->tx_ring_ref
= GRANT_INVALID_REF
;
1416 queue
->rx_ring_ref
= GRANT_INVALID_REF
;
1417 queue
->tx
.sring
= NULL
;
1418 queue
->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_single(struct netfront_queue
*queue
)
1464 err
= xenbus_alloc_evtchn(queue
->info
->xbdev
, &queue
->tx_evtchn
);
1468 err
= bind_evtchn_to_irqhandler(queue
->tx_evtchn
,
1470 0, queue
->info
->netdev
->name
, queue
);
1473 queue
->rx_evtchn
= queue
->tx_evtchn
;
1474 queue
->rx_irq
= queue
->tx_irq
= err
;
1479 xenbus_free_evtchn(queue
->info
->xbdev
, queue
->tx_evtchn
);
1480 queue
->tx_evtchn
= 0;
1485 static int setup_netfront_split(struct netfront_queue
*queue
)
1489 err
= xenbus_alloc_evtchn(queue
->info
->xbdev
, &queue
->tx_evtchn
);
1492 err
= xenbus_alloc_evtchn(queue
->info
->xbdev
, &queue
->rx_evtchn
);
1494 goto alloc_rx_evtchn_fail
;
1496 snprintf(queue
->tx_irq_name
, sizeof(queue
->tx_irq_name
),
1497 "%s-tx", queue
->name
);
1498 err
= bind_evtchn_to_irqhandler(queue
->tx_evtchn
,
1499 xennet_tx_interrupt
,
1500 0, queue
->tx_irq_name
, queue
);
1503 queue
->tx_irq
= err
;
1505 snprintf(queue
->rx_irq_name
, sizeof(queue
->rx_irq_name
),
1506 "%s-rx", queue
->name
);
1507 err
= bind_evtchn_to_irqhandler(queue
->rx_evtchn
,
1508 xennet_rx_interrupt
,
1509 0, queue
->rx_irq_name
, queue
);
1512 queue
->rx_irq
= err
;
1517 unbind_from_irqhandler(queue
->tx_irq
, queue
);
1520 xenbus_free_evtchn(queue
->info
->xbdev
, queue
->rx_evtchn
);
1521 queue
->rx_evtchn
= 0;
1522 alloc_rx_evtchn_fail
:
1523 xenbus_free_evtchn(queue
->info
->xbdev
, queue
->tx_evtchn
);
1524 queue
->tx_evtchn
= 0;
1529 static int setup_netfront(struct xenbus_device
*dev
,
1530 struct netfront_queue
*queue
, unsigned int feature_split_evtchn
)
1532 struct xen_netif_tx_sring
*txs
;
1533 struct xen_netif_rx_sring
*rxs
;
1537 queue
->tx_ring_ref
= GRANT_INVALID_REF
;
1538 queue
->rx_ring_ref
= GRANT_INVALID_REF
;
1539 queue
->rx
.sring
= NULL
;
1540 queue
->tx
.sring
= NULL
;
1542 txs
= (struct xen_netif_tx_sring
*)get_zeroed_page(GFP_NOIO
| __GFP_HIGH
);
1545 xenbus_dev_fatal(dev
, err
, "allocating tx ring page");
1548 SHARED_RING_INIT(txs
);
1549 FRONT_RING_INIT(&queue
->tx
, txs
, XEN_PAGE_SIZE
);
1551 err
= xenbus_grant_ring(dev
, txs
, 1, &gref
);
1553 goto grant_tx_ring_fail
;
1554 queue
->tx_ring_ref
= gref
;
1556 rxs
= (struct xen_netif_rx_sring
*)get_zeroed_page(GFP_NOIO
| __GFP_HIGH
);
1559 xenbus_dev_fatal(dev
, err
, "allocating rx ring page");
1560 goto alloc_rx_ring_fail
;
1562 SHARED_RING_INIT(rxs
);
1563 FRONT_RING_INIT(&queue
->rx
, rxs
, XEN_PAGE_SIZE
);
1565 err
= xenbus_grant_ring(dev
, rxs
, 1, &gref
);
1567 goto grant_rx_ring_fail
;
1568 queue
->rx_ring_ref
= gref
;
1570 if (feature_split_evtchn
)
1571 err
= setup_netfront_split(queue
);
1572 /* setup single event channel if
1573 * a) feature-split-event-channels == 0
1574 * b) feature-split-event-channels == 1 but failed to setup
1576 if (!feature_split_evtchn
|| (feature_split_evtchn
&& err
))
1577 err
= setup_netfront_single(queue
);
1580 goto alloc_evtchn_fail
;
1584 /* If we fail to setup netfront, it is safe to just revoke access to
1585 * granted pages because backend is not accessing it at this point.
1588 gnttab_end_foreign_access_ref(queue
->rx_ring_ref
, 0);
1590 free_page((unsigned long)rxs
);
1592 gnttab_end_foreign_access_ref(queue
->tx_ring_ref
, 0);
1594 free_page((unsigned long)txs
);
1599 /* Queue-specific initialisation
1600 * This used to be done in xennet_create_dev() but must now
1603 static int xennet_init_queue(struct netfront_queue
*queue
)
1608 spin_lock_init(&queue
->tx_lock
);
1609 spin_lock_init(&queue
->rx_lock
);
1611 timer_setup(&queue
->rx_refill_timer
, rx_refill_timeout
, 0);
1613 snprintf(queue
->name
, sizeof(queue
->name
), "%s-q%u",
1614 queue
->info
->netdev
->name
, queue
->id
);
1616 /* Initialise tx_skbs as a free chain containing every entry. */
1617 queue
->tx_skb_freelist
= 0;
1618 for (i
= 0; i
< NET_TX_RING_SIZE
; i
++) {
1619 skb_entry_set_link(&queue
->tx_skbs
[i
], i
+1);
1620 queue
->grant_tx_ref
[i
] = GRANT_INVALID_REF
;
1621 queue
->grant_tx_page
[i
] = NULL
;
1624 /* Clear out rx_skbs */
1625 for (i
= 0; i
< NET_RX_RING_SIZE
; i
++) {
1626 queue
->rx_skbs
[i
] = NULL
;
1627 queue
->grant_rx_ref
[i
] = GRANT_INVALID_REF
;
1630 /* A grant for every tx ring slot */
1631 if (gnttab_alloc_grant_references(NET_TX_RING_SIZE
,
1632 &queue
->gref_tx_head
) < 0) {
1633 pr_alert("can't alloc tx grant refs\n");
1638 /* A grant for every rx ring slot */
1639 if (gnttab_alloc_grant_references(NET_RX_RING_SIZE
,
1640 &queue
->gref_rx_head
) < 0) {
1641 pr_alert("can't alloc rx grant refs\n");
1649 gnttab_free_grant_references(queue
->gref_tx_head
);
1654 static int write_queue_xenstore_keys(struct netfront_queue
*queue
,
1655 struct xenbus_transaction
*xbt
, int write_hierarchical
)
1657 /* Write the queue-specific keys into XenStore in the traditional
1658 * way for a single queue, or in a queue subkeys for multiple
1661 struct xenbus_device
*dev
= queue
->info
->xbdev
;
1663 const char *message
;
1667 /* Choose the correct place to write the keys */
1668 if (write_hierarchical
) {
1669 pathsize
= strlen(dev
->nodename
) + 10;
1670 path
= kzalloc(pathsize
, GFP_KERNEL
);
1673 message
= "out of memory while writing ring references";
1676 snprintf(path
, pathsize
, "%s/queue-%u",
1677 dev
->nodename
, queue
->id
);
1679 path
= (char *)dev
->nodename
;
1682 /* Write ring references */
1683 err
= xenbus_printf(*xbt
, path
, "tx-ring-ref", "%u",
1684 queue
->tx_ring_ref
);
1686 message
= "writing tx-ring-ref";
1690 err
= xenbus_printf(*xbt
, path
, "rx-ring-ref", "%u",
1691 queue
->rx_ring_ref
);
1693 message
= "writing rx-ring-ref";
1697 /* Write event channels; taking into account both shared
1698 * and split event channel scenarios.
1700 if (queue
->tx_evtchn
== queue
->rx_evtchn
) {
1701 /* Shared event channel */
1702 err
= xenbus_printf(*xbt
, path
,
1703 "event-channel", "%u", queue
->tx_evtchn
);
1705 message
= "writing event-channel";
1709 /* Split event channels */
1710 err
= xenbus_printf(*xbt
, path
,
1711 "event-channel-tx", "%u", queue
->tx_evtchn
);
1713 message
= "writing event-channel-tx";
1717 err
= xenbus_printf(*xbt
, path
,
1718 "event-channel-rx", "%u", queue
->rx_evtchn
);
1720 message
= "writing event-channel-rx";
1725 if (write_hierarchical
)
1730 if (write_hierarchical
)
1732 xenbus_dev_fatal(dev
, err
, "%s", message
);
1736 static void xennet_destroy_queues(struct netfront_info
*info
)
1742 for (i
= 0; i
< info
->netdev
->real_num_tx_queues
; i
++) {
1743 struct netfront_queue
*queue
= &info
->queues
[i
];
1745 if (netif_running(info
->netdev
))
1746 napi_disable(&queue
->napi
);
1747 netif_napi_del(&queue
->napi
);
1752 kfree(info
->queues
);
1753 info
->queues
= NULL
;
1756 static int xennet_create_queues(struct netfront_info
*info
,
1757 unsigned int *num_queues
)
1762 info
->queues
= kcalloc(*num_queues
, sizeof(struct netfront_queue
),
1769 for (i
= 0; i
< *num_queues
; i
++) {
1770 struct netfront_queue
*queue
= &info
->queues
[i
];
1775 ret
= xennet_init_queue(queue
);
1777 dev_warn(&info
->netdev
->dev
,
1778 "only created %d queues\n", i
);
1783 netif_napi_add(queue
->info
->netdev
, &queue
->napi
,
1785 if (netif_running(info
->netdev
))
1786 napi_enable(&queue
->napi
);
1789 netif_set_real_num_tx_queues(info
->netdev
, *num_queues
);
1793 if (*num_queues
== 0) {
1794 dev_err(&info
->netdev
->dev
, "no queues\n");
1800 /* Common code used when first setting up, and when resuming. */
1801 static int talk_to_netback(struct xenbus_device
*dev
,
1802 struct netfront_info
*info
)
1804 const char *message
;
1805 struct xenbus_transaction xbt
;
1807 unsigned int feature_split_evtchn
;
1809 unsigned int max_queues
= 0;
1810 struct netfront_queue
*queue
= NULL
;
1811 unsigned int num_queues
= 1;
1813 info
->netdev
->irq
= 0;
1815 /* Check if backend supports multiple queues */
1816 max_queues
= xenbus_read_unsigned(info
->xbdev
->otherend
,
1817 "multi-queue-max-queues", 1);
1818 num_queues
= min(max_queues
, xennet_max_queues
);
1820 /* Check feature-split-event-channels */
1821 feature_split_evtchn
= xenbus_read_unsigned(info
->xbdev
->otherend
,
1822 "feature-split-event-channels", 0);
1824 /* Read mac addr. */
1825 err
= xen_net_read_mac(dev
, info
->netdev
->dev_addr
);
1827 xenbus_dev_fatal(dev
, err
, "parsing %s/mac", dev
->nodename
);
1832 xennet_destroy_queues(info
);
1834 err
= xennet_create_queues(info
, &num_queues
);
1836 xenbus_dev_fatal(dev
, err
, "creating queues");
1837 kfree(info
->queues
);
1838 info
->queues
= NULL
;
1842 /* Create shared ring, alloc event channel -- for each queue */
1843 for (i
= 0; i
< num_queues
; ++i
) {
1844 queue
= &info
->queues
[i
];
1845 err
= setup_netfront(dev
, queue
, feature_split_evtchn
);
1851 err
= xenbus_transaction_start(&xbt
);
1853 xenbus_dev_fatal(dev
, err
, "starting transaction");
1857 if (xenbus_exists(XBT_NIL
,
1858 info
->xbdev
->otherend
, "multi-queue-max-queues")) {
1859 /* Write the number of queues */
1860 err
= xenbus_printf(xbt
, dev
->nodename
,
1861 "multi-queue-num-queues", "%u", num_queues
);
1863 message
= "writing multi-queue-num-queues";
1864 goto abort_transaction_no_dev_fatal
;
1868 if (num_queues
== 1) {
1869 err
= write_queue_xenstore_keys(&info
->queues
[0], &xbt
, 0); /* flat */
1871 goto abort_transaction_no_dev_fatal
;
1873 /* Write the keys for each queue */
1874 for (i
= 0; i
< num_queues
; ++i
) {
1875 queue
= &info
->queues
[i
];
1876 err
= write_queue_xenstore_keys(queue
, &xbt
, 1); /* hierarchical */
1878 goto abort_transaction_no_dev_fatal
;
1882 /* The remaining keys are not queue-specific */
1883 err
= xenbus_printf(xbt
, dev
->nodename
, "request-rx-copy", "%u",
1886 message
= "writing request-rx-copy";
1887 goto abort_transaction
;
1890 err
= xenbus_printf(xbt
, dev
->nodename
, "feature-rx-notify", "%d", 1);
1892 message
= "writing feature-rx-notify";
1893 goto abort_transaction
;
1896 err
= xenbus_printf(xbt
, dev
->nodename
, "feature-sg", "%d", 1);
1898 message
= "writing feature-sg";
1899 goto abort_transaction
;
1902 err
= xenbus_printf(xbt
, dev
->nodename
, "feature-gso-tcpv4", "%d", 1);
1904 message
= "writing feature-gso-tcpv4";
1905 goto abort_transaction
;
1908 err
= xenbus_write(xbt
, dev
->nodename
, "feature-gso-tcpv6", "1");
1910 message
= "writing feature-gso-tcpv6";
1911 goto abort_transaction
;
1914 err
= xenbus_write(xbt
, dev
->nodename
, "feature-ipv6-csum-offload",
1917 message
= "writing feature-ipv6-csum-offload";
1918 goto abort_transaction
;
1921 err
= xenbus_transaction_end(xbt
, 0);
1925 xenbus_dev_fatal(dev
, err
, "completing transaction");
1932 xenbus_dev_fatal(dev
, err
, "%s", message
);
1933 abort_transaction_no_dev_fatal
:
1934 xenbus_transaction_end(xbt
, 1);
1936 xennet_disconnect_backend(info
);
1937 xennet_destroy_queues(info
);
1939 device_unregister(&dev
->dev
);
1943 static int xennet_connect(struct net_device
*dev
)
1945 struct netfront_info
*np
= netdev_priv(dev
);
1946 unsigned int num_queues
= 0;
1949 struct netfront_queue
*queue
= NULL
;
1951 if (!xenbus_read_unsigned(np
->xbdev
->otherend
, "feature-rx-copy", 0)) {
1953 "backend does not support copying receive path\n");
1957 err
= talk_to_netback(np
->xbdev
, np
);
1961 /* talk_to_netback() sets the correct number of queues */
1962 num_queues
= dev
->real_num_tx_queues
;
1965 netdev_update_features(dev
);
1969 * All public and private state should now be sane. Get
1970 * ready to start sending and receiving packets and give the driver
1971 * domain a kick because we've probably just requeued some
1974 netif_carrier_on(np
->netdev
);
1975 for (j
= 0; j
< num_queues
; ++j
) {
1976 queue
= &np
->queues
[j
];
1978 notify_remote_via_irq(queue
->tx_irq
);
1979 if (queue
->tx_irq
!= queue
->rx_irq
)
1980 notify_remote_via_irq(queue
->rx_irq
);
1982 spin_lock_irq(&queue
->tx_lock
);
1983 xennet_tx_buf_gc(queue
);
1984 spin_unlock_irq(&queue
->tx_lock
);
1986 spin_lock_bh(&queue
->rx_lock
);
1987 xennet_alloc_rx_buffers(queue
);
1988 spin_unlock_bh(&queue
->rx_lock
);
1995 * Callback received when the backend's state changes.
1997 static void netback_changed(struct xenbus_device
*dev
,
1998 enum xenbus_state backend_state
)
2000 struct netfront_info
*np
= dev_get_drvdata(&dev
->dev
);
2001 struct net_device
*netdev
= np
->netdev
;
2003 dev_dbg(&dev
->dev
, "%s\n", xenbus_strstate(backend_state
));
2005 switch (backend_state
) {
2006 case XenbusStateInitialising
:
2007 case XenbusStateInitialised
:
2008 case XenbusStateReconfiguring
:
2009 case XenbusStateReconfigured
:
2010 case XenbusStateUnknown
:
2013 case XenbusStateInitWait
:
2014 if (dev
->state
!= XenbusStateInitialising
)
2016 if (xennet_connect(netdev
) != 0)
2018 xenbus_switch_state(dev
, XenbusStateConnected
);
2021 case XenbusStateConnected
:
2022 netdev_notify_peers(netdev
);
2025 case XenbusStateClosed
:
2026 wake_up_all(&module_unload_q
);
2027 if (dev
->state
== XenbusStateClosed
)
2029 /* Missed the backend's CLOSING state -- fallthrough */
2030 case XenbusStateClosing
:
2031 wake_up_all(&module_unload_q
);
2032 xenbus_frontend_closed(dev
);
2037 static const struct xennet_stat
{
2038 char name
[ETH_GSTRING_LEN
];
2040 } xennet_stats
[] = {
2042 "rx_gso_checksum_fixup",
2043 offsetof(struct netfront_info
, rx_gso_checksum_fixup
)
2047 static int xennet_get_sset_count(struct net_device
*dev
, int string_set
)
2049 switch (string_set
) {
2051 return ARRAY_SIZE(xennet_stats
);
2057 static void xennet_get_ethtool_stats(struct net_device
*dev
,
2058 struct ethtool_stats
*stats
, u64
* data
)
2060 void *np
= netdev_priv(dev
);
2063 for (i
= 0; i
< ARRAY_SIZE(xennet_stats
); i
++)
2064 data
[i
] = atomic_read((atomic_t
*)(np
+ xennet_stats
[i
].offset
));
2067 static void xennet_get_strings(struct net_device
*dev
, u32 stringset
, u8
* data
)
2071 switch (stringset
) {
2073 for (i
= 0; i
< ARRAY_SIZE(xennet_stats
); i
++)
2074 memcpy(data
+ i
* ETH_GSTRING_LEN
,
2075 xennet_stats
[i
].name
, ETH_GSTRING_LEN
);
2080 static const struct ethtool_ops xennet_ethtool_ops
=
2082 .get_link
= ethtool_op_get_link
,
2084 .get_sset_count
= xennet_get_sset_count
,
2085 .get_ethtool_stats
= xennet_get_ethtool_stats
,
2086 .get_strings
= xennet_get_strings
,
2090 static ssize_t
show_rxbuf(struct device
*dev
,
2091 struct device_attribute
*attr
, char *buf
)
2093 return sprintf(buf
, "%lu\n", NET_RX_RING_SIZE
);
2096 static ssize_t
store_rxbuf(struct device
*dev
,
2097 struct device_attribute
*attr
,
2098 const char *buf
, size_t len
)
2101 unsigned long target
;
2103 if (!capable(CAP_NET_ADMIN
))
2106 target
= simple_strtoul(buf
, &endp
, 0);
2110 /* rxbuf_min and rxbuf_max are no longer configurable. */
2115 static DEVICE_ATTR(rxbuf_min
, S_IRUGO
|S_IWUSR
, show_rxbuf
, store_rxbuf
);
2116 static DEVICE_ATTR(rxbuf_max
, S_IRUGO
|S_IWUSR
, show_rxbuf
, store_rxbuf
);
2117 static DEVICE_ATTR(rxbuf_cur
, S_IRUGO
, show_rxbuf
, NULL
);
2119 static struct attribute
*xennet_dev_attrs
[] = {
2120 &dev_attr_rxbuf_min
.attr
,
2121 &dev_attr_rxbuf_max
.attr
,
2122 &dev_attr_rxbuf_cur
.attr
,
2126 static const struct attribute_group xennet_dev_group
= {
2127 .attrs
= xennet_dev_attrs
2129 #endif /* CONFIG_SYSFS */
2131 static int xennet_remove(struct xenbus_device
*dev
)
2133 struct netfront_info
*info
= dev_get_drvdata(&dev
->dev
);
2135 dev_dbg(&dev
->dev
, "%s\n", dev
->nodename
);
2137 if (xenbus_read_driver_state(dev
->otherend
) != XenbusStateClosed
) {
2138 xenbus_switch_state(dev
, XenbusStateClosing
);
2139 wait_event(module_unload_q
,
2140 xenbus_read_driver_state(dev
->otherend
) ==
2141 XenbusStateClosing
);
2143 xenbus_switch_state(dev
, XenbusStateClosed
);
2144 wait_event(module_unload_q
,
2145 xenbus_read_driver_state(dev
->otherend
) ==
2146 XenbusStateClosed
||
2147 xenbus_read_driver_state(dev
->otherend
) ==
2148 XenbusStateUnknown
);
2151 xennet_disconnect_backend(info
);
2153 unregister_netdev(info
->netdev
);
2156 xennet_destroy_queues(info
);
2157 xennet_free_netdev(info
->netdev
);
2162 static const struct xenbus_device_id netfront_ids
[] = {
2167 static struct xenbus_driver netfront_driver
= {
2168 .ids
= netfront_ids
,
2169 .probe
= netfront_probe
,
2170 .remove
= xennet_remove
,
2171 .resume
= netfront_resume
,
2172 .otherend_changed
= netback_changed
,
2175 static int __init
netif_init(void)
2180 if (!xen_has_pv_nic_devices())
2183 pr_info("Initialising Xen virtual ethernet driver\n");
2185 /* Allow as many queues as there are CPUs inut max. 8 if user has not
2186 * specified a value.
2188 if (xennet_max_queues
== 0)
2189 xennet_max_queues
= min_t(unsigned int, MAX_QUEUES_DEFAULT
,
2192 return xenbus_register_frontend(&netfront_driver
);
2194 module_init(netif_init
);
2197 static void __exit
netif_exit(void)
2199 xenbus_unregister_driver(&netfront_driver
);
2201 module_exit(netif_exit
);
2203 MODULE_DESCRIPTION("Xen virtual network device frontend");
2204 MODULE_LICENSE("GPL");
2205 MODULE_ALIAS("xen:vif");
2206 MODULE_ALIAS("xennet");