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 static unsigned int xennet_max_queues
;
61 module_param_named(max_queues
, xennet_max_queues
, uint
, 0644);
62 MODULE_PARM_DESC(max_queues
,
63 "Maximum number of queues per virtual interface");
65 static const struct ethtool_ops xennet_ethtool_ops
;
71 #define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb))
73 #define RX_COPY_THRESHOLD 256
75 #define GRANT_INVALID_REF 0
77 #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, PAGE_SIZE)
78 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, PAGE_SIZE)
80 /* Minimum number of Rx slots (includes slot for GSO metadata). */
81 #define NET_RX_SLOTS_MIN (XEN_NETIF_NR_SLOTS_MIN + 1)
83 /* Queue name is interface name with "-qNNN" appended */
84 #define QUEUE_NAME_SIZE (IFNAMSIZ + 6)
86 /* IRQ name is queue name with "-tx" or "-rx" appended */
87 #define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3)
89 struct netfront_stats
{
92 struct u64_stats_sync syncp
;
97 struct netfront_queue
{
98 unsigned int id
; /* Queue ID, 0-based */
99 char name
[QUEUE_NAME_SIZE
]; /* DEVNAME-qN */
100 struct netfront_info
*info
;
102 struct napi_struct napi
;
104 /* Split event channels support, tx_* == rx_* when using
105 * single event channel.
107 unsigned int tx_evtchn
, rx_evtchn
;
108 unsigned int tx_irq
, rx_irq
;
109 /* Only used when split event channels support is enabled */
110 char tx_irq_name
[IRQ_NAME_SIZE
]; /* DEVNAME-qN-tx */
111 char rx_irq_name
[IRQ_NAME_SIZE
]; /* DEVNAME-qN-rx */
114 struct xen_netif_tx_front_ring tx
;
118 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
119 * are linked from tx_skb_freelist through skb_entry.link.
121 * NB. Freelist index entries are always going to be less than
122 * PAGE_OFFSET, whereas pointers to skbs will always be equal or
123 * greater than PAGE_OFFSET: we use this property to distinguish
129 } tx_skbs
[NET_TX_RING_SIZE
];
130 grant_ref_t gref_tx_head
;
131 grant_ref_t grant_tx_ref
[NET_TX_RING_SIZE
];
132 struct page
*grant_tx_page
[NET_TX_RING_SIZE
];
133 unsigned tx_skb_freelist
;
135 spinlock_t rx_lock ____cacheline_aligned_in_smp
;
136 struct xen_netif_rx_front_ring rx
;
139 struct timer_list rx_refill_timer
;
141 struct sk_buff
*rx_skbs
[NET_RX_RING_SIZE
];
142 grant_ref_t gref_rx_head
;
143 grant_ref_t grant_rx_ref
[NET_RX_RING_SIZE
];
146 struct netfront_info
{
147 struct list_head list
;
148 struct net_device
*netdev
;
150 struct xenbus_device
*xbdev
;
152 /* Multi-queue support */
153 struct netfront_queue
*queues
;
156 struct netfront_stats __percpu
*rx_stats
;
157 struct netfront_stats __percpu
*tx_stats
;
159 atomic_t rx_gso_checksum_fixup
;
162 struct netfront_rx_info
{
163 struct xen_netif_rx_response rx
;
164 struct xen_netif_extra_info extras
[XEN_NETIF_EXTRA_TYPE_MAX
- 1];
167 static void skb_entry_set_link(union skb_entry
*list
, unsigned short id
)
172 static int skb_entry_is_link(const union skb_entry
*list
)
174 BUILD_BUG_ON(sizeof(list
->skb
) != sizeof(list
->link
));
175 return (unsigned long)list
->skb
< PAGE_OFFSET
;
179 * Access macros for acquiring freeing slots in tx_skbs[].
182 static void add_id_to_freelist(unsigned *head
, union skb_entry
*list
,
185 skb_entry_set_link(&list
[id
], *head
);
189 static unsigned short get_id_from_freelist(unsigned *head
,
190 union skb_entry
*list
)
192 unsigned int id
= *head
;
193 *head
= list
[id
].link
;
197 static int xennet_rxidx(RING_IDX idx
)
199 return idx
& (NET_RX_RING_SIZE
- 1);
202 static struct sk_buff
*xennet_get_rx_skb(struct netfront_queue
*queue
,
205 int i
= xennet_rxidx(ri
);
206 struct sk_buff
*skb
= queue
->rx_skbs
[i
];
207 queue
->rx_skbs
[i
] = NULL
;
211 static grant_ref_t
xennet_get_rx_ref(struct netfront_queue
*queue
,
214 int i
= xennet_rxidx(ri
);
215 grant_ref_t ref
= queue
->grant_rx_ref
[i
];
216 queue
->grant_rx_ref
[i
] = GRANT_INVALID_REF
;
221 static const struct attribute_group xennet_dev_group
;
224 static bool xennet_can_sg(struct net_device
*dev
)
226 return dev
->features
& NETIF_F_SG
;
230 static void rx_refill_timeout(unsigned long data
)
232 struct netfront_queue
*queue
= (struct netfront_queue
*)data
;
233 napi_schedule(&queue
->napi
);
236 static int netfront_tx_slot_available(struct netfront_queue
*queue
)
238 return (queue
->tx
.req_prod_pvt
- queue
->tx
.rsp_cons
) <
239 (NET_TX_RING_SIZE
- MAX_SKB_FRAGS
- 2);
242 static void xennet_maybe_wake_tx(struct netfront_queue
*queue
)
244 struct net_device
*dev
= queue
->info
->netdev
;
245 struct netdev_queue
*dev_queue
= netdev_get_tx_queue(dev
, queue
->id
);
247 if (unlikely(netif_tx_queue_stopped(dev_queue
)) &&
248 netfront_tx_slot_available(queue
) &&
249 likely(netif_running(dev
)))
250 netif_tx_wake_queue(netdev_get_tx_queue(dev
, queue
->id
));
254 static struct sk_buff
*xennet_alloc_one_rx_buffer(struct netfront_queue
*queue
)
259 skb
= __netdev_alloc_skb(queue
->info
->netdev
,
260 RX_COPY_THRESHOLD
+ NET_IP_ALIGN
,
261 GFP_ATOMIC
| __GFP_NOWARN
);
265 page
= alloc_page(GFP_ATOMIC
| __GFP_NOWARN
);
270 skb_add_rx_frag(skb
, 0, page
, 0, 0, PAGE_SIZE
);
272 /* Align ip header to a 16 bytes boundary */
273 skb_reserve(skb
, NET_IP_ALIGN
);
274 skb
->dev
= queue
->info
->netdev
;
280 static void xennet_alloc_rx_buffers(struct netfront_queue
*queue
)
282 RING_IDX req_prod
= queue
->rx
.req_prod_pvt
;
285 if (unlikely(!netif_carrier_ok(queue
->info
->netdev
)))
288 for (req_prod
= queue
->rx
.req_prod_pvt
;
289 req_prod
- queue
->rx
.rsp_cons
< NET_RX_RING_SIZE
;
295 struct xen_netif_rx_request
*req
;
297 skb
= xennet_alloc_one_rx_buffer(queue
);
301 id
= xennet_rxidx(req_prod
);
303 BUG_ON(queue
->rx_skbs
[id
]);
304 queue
->rx_skbs
[id
] = skb
;
306 ref
= gnttab_claim_grant_reference(&queue
->gref_rx_head
);
307 BUG_ON((signed short)ref
< 0);
308 queue
->grant_rx_ref
[id
] = ref
;
310 gfn
= xen_page_to_gfn(skb_frag_page(&skb_shinfo(skb
)->frags
[0]));
312 req
= RING_GET_REQUEST(&queue
->rx
, req_prod
);
313 gnttab_grant_foreign_access_ref(ref
,
314 queue
->info
->xbdev
->otherend_id
,
322 queue
->rx
.req_prod_pvt
= req_prod
;
324 /* Not enough requests? Try again later. */
325 if (req_prod
- queue
->rx
.rsp_cons
< NET_RX_SLOTS_MIN
) {
326 mod_timer(&queue
->rx_refill_timer
, jiffies
+ (HZ
/10));
330 wmb(); /* barrier so backend seens requests */
332 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue
->rx
, notify
);
334 notify_remote_via_irq(queue
->rx_irq
);
337 static int xennet_open(struct net_device
*dev
)
339 struct netfront_info
*np
= netdev_priv(dev
);
340 unsigned int num_queues
= dev
->real_num_tx_queues
;
342 struct netfront_queue
*queue
= NULL
;
344 for (i
= 0; i
< num_queues
; ++i
) {
345 queue
= &np
->queues
[i
];
346 napi_enable(&queue
->napi
);
348 spin_lock_bh(&queue
->rx_lock
);
349 if (netif_carrier_ok(dev
)) {
350 xennet_alloc_rx_buffers(queue
);
351 queue
->rx
.sring
->rsp_event
= queue
->rx
.rsp_cons
+ 1;
352 if (RING_HAS_UNCONSUMED_RESPONSES(&queue
->rx
))
353 napi_schedule(&queue
->napi
);
355 spin_unlock_bh(&queue
->rx_lock
);
358 netif_tx_start_all_queues(dev
);
363 static void xennet_tx_buf_gc(struct netfront_queue
*queue
)
369 BUG_ON(!netif_carrier_ok(queue
->info
->netdev
));
372 prod
= queue
->tx
.sring
->rsp_prod
;
373 rmb(); /* Ensure we see responses up to 'rp'. */
375 for (cons
= queue
->tx
.rsp_cons
; cons
!= prod
; cons
++) {
376 struct xen_netif_tx_response
*txrsp
;
378 txrsp
= RING_GET_RESPONSE(&queue
->tx
, cons
);
379 if (txrsp
->status
== XEN_NETIF_RSP_NULL
)
383 skb
= queue
->tx_skbs
[id
].skb
;
384 if (unlikely(gnttab_query_foreign_access(
385 queue
->grant_tx_ref
[id
]) != 0)) {
386 pr_alert("%s: warning -- grant still in use by backend domain\n",
390 gnttab_end_foreign_access_ref(
391 queue
->grant_tx_ref
[id
], GNTMAP_readonly
);
392 gnttab_release_grant_reference(
393 &queue
->gref_tx_head
, queue
->grant_tx_ref
[id
]);
394 queue
->grant_tx_ref
[id
] = GRANT_INVALID_REF
;
395 queue
->grant_tx_page
[id
] = NULL
;
396 add_id_to_freelist(&queue
->tx_skb_freelist
, queue
->tx_skbs
, id
);
397 dev_kfree_skb_irq(skb
);
400 queue
->tx
.rsp_cons
= prod
;
403 * Set a new event, then check for race with update of tx_cons.
404 * Note that it is essential to schedule a callback, no matter
405 * how few buffers are pending. Even if there is space in the
406 * transmit ring, higher layers may be blocked because too much
407 * data is outstanding: in such cases notification from Xen is
408 * likely to be the only kick that we'll get.
410 queue
->tx
.sring
->rsp_event
=
411 prod
+ ((queue
->tx
.sring
->req_prod
- prod
) >> 1) + 1;
412 mb(); /* update shared area */
413 } while ((cons
== prod
) && (prod
!= queue
->tx
.sring
->rsp_prod
));
415 xennet_maybe_wake_tx(queue
);
418 static struct xen_netif_tx_request
*xennet_make_one_txreq(
419 struct netfront_queue
*queue
, struct sk_buff
*skb
,
420 struct page
*page
, unsigned int offset
, unsigned int len
)
423 struct xen_netif_tx_request
*tx
;
426 len
= min_t(unsigned int, PAGE_SIZE
- offset
, len
);
428 id
= get_id_from_freelist(&queue
->tx_skb_freelist
, queue
->tx_skbs
);
429 tx
= RING_GET_REQUEST(&queue
->tx
, queue
->tx
.req_prod_pvt
++);
430 ref
= gnttab_claim_grant_reference(&queue
->gref_tx_head
);
431 BUG_ON((signed short)ref
< 0);
433 gnttab_grant_foreign_access_ref(ref
,
434 queue
->info
->xbdev
->otherend_id
,
435 xen_page_to_gfn(page
),
438 queue
->tx_skbs
[id
].skb
= skb
;
439 queue
->grant_tx_page
[id
] = page
;
440 queue
->grant_tx_ref
[id
] = ref
;
451 static struct xen_netif_tx_request
*xennet_make_txreqs(
452 struct netfront_queue
*queue
, struct xen_netif_tx_request
*tx
,
453 struct sk_buff
*skb
, struct page
*page
,
454 unsigned int offset
, unsigned int len
)
456 /* Skip unused frames from start of page */
457 page
+= offset
>> PAGE_SHIFT
;
458 offset
&= ~PAGE_MASK
;
461 tx
->flags
|= XEN_NETTXF_more_data
;
462 tx
= xennet_make_one_txreq(queue
, skb_get(skb
),
473 * Count how many ring slots are required to send this skb. Each frag
474 * might be a compound page.
476 static int xennet_count_skb_slots(struct sk_buff
*skb
)
478 int i
, frags
= skb_shinfo(skb
)->nr_frags
;
481 pages
= PFN_UP(offset_in_page(skb
->data
) + skb_headlen(skb
));
483 for (i
= 0; i
< frags
; i
++) {
484 skb_frag_t
*frag
= skb_shinfo(skb
)->frags
+ i
;
485 unsigned long size
= skb_frag_size(frag
);
486 unsigned long offset
= frag
->page_offset
;
488 /* Skip unused frames from start of page */
489 offset
&= ~PAGE_MASK
;
491 pages
+= PFN_UP(offset
+ size
);
497 static u16
xennet_select_queue(struct net_device
*dev
, struct sk_buff
*skb
,
498 void *accel_priv
, select_queue_fallback_t fallback
)
500 unsigned int num_queues
= dev
->real_num_tx_queues
;
504 /* First, check if there is only one queue */
505 if (num_queues
== 1) {
508 hash
= skb_get_hash(skb
);
509 queue_idx
= hash
% num_queues
;
515 static int xennet_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
517 struct netfront_info
*np
= netdev_priv(dev
);
518 struct netfront_stats
*tx_stats
= this_cpu_ptr(np
->tx_stats
);
519 struct xen_netif_tx_request
*tx
, *first_tx
;
527 struct netfront_queue
*queue
= NULL
;
528 unsigned int num_queues
= dev
->real_num_tx_queues
;
531 /* Drop the packet if no queues are set up */
534 /* Determine which queue to transmit this SKB on */
535 queue_index
= skb_get_queue_mapping(skb
);
536 queue
= &np
->queues
[queue_index
];
538 /* If skb->len is too big for wire format, drop skb and alert
539 * user about misconfiguration.
541 if (unlikely(skb
->len
> XEN_NETIF_MAX_TX_SIZE
)) {
542 net_alert_ratelimited(
543 "xennet: skb->len = %u, too big for wire format\n",
548 slots
= xennet_count_skb_slots(skb
);
549 if (unlikely(slots
> MAX_SKB_FRAGS
+ 1)) {
550 net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n",
552 if (skb_linearize(skb
))
556 page
= virt_to_page(skb
->data
);
557 offset
= offset_in_page(skb
->data
);
558 len
= skb_headlen(skb
);
560 spin_lock_irqsave(&queue
->tx_lock
, flags
);
562 if (unlikely(!netif_carrier_ok(dev
) ||
563 (slots
> 1 && !xennet_can_sg(dev
)) ||
564 netif_needs_gso(skb
, netif_skb_features(skb
)))) {
565 spin_unlock_irqrestore(&queue
->tx_lock
, flags
);
569 /* First request for the linear area. */
570 first_tx
= tx
= xennet_make_one_txreq(queue
, skb
,
576 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
578 tx
->flags
|= XEN_NETTXF_csum_blank
| XEN_NETTXF_data_validated
;
579 else if (skb
->ip_summed
== CHECKSUM_UNNECESSARY
)
580 /* remote but checksummed. */
581 tx
->flags
|= XEN_NETTXF_data_validated
;
583 /* Optional extra info after the first request. */
584 if (skb_shinfo(skb
)->gso_size
) {
585 struct xen_netif_extra_info
*gso
;
587 gso
= (struct xen_netif_extra_info
*)
588 RING_GET_REQUEST(&queue
->tx
, queue
->tx
.req_prod_pvt
++);
590 tx
->flags
|= XEN_NETTXF_extra_info
;
592 gso
->u
.gso
.size
= skb_shinfo(skb
)->gso_size
;
593 gso
->u
.gso
.type
= (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCPV6
) ?
594 XEN_NETIF_GSO_TYPE_TCPV6
:
595 XEN_NETIF_GSO_TYPE_TCPV4
;
597 gso
->u
.gso
.features
= 0;
599 gso
->type
= XEN_NETIF_EXTRA_TYPE_GSO
;
603 /* Requests for the rest of the linear area. */
604 tx
= xennet_make_txreqs(queue
, tx
, skb
, page
, offset
, len
);
606 /* Requests for all the frags. */
607 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
608 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
609 tx
= xennet_make_txreqs(queue
, tx
, skb
,
610 skb_frag_page(frag
), frag
->page_offset
,
611 skb_frag_size(frag
));
614 /* First request has the packet length. */
615 first_tx
->size
= skb
->len
;
617 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue
->tx
, notify
);
619 notify_remote_via_irq(queue
->tx_irq
);
621 u64_stats_update_begin(&tx_stats
->syncp
);
622 tx_stats
->bytes
+= skb
->len
;
624 u64_stats_update_end(&tx_stats
->syncp
);
626 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
627 xennet_tx_buf_gc(queue
);
629 if (!netfront_tx_slot_available(queue
))
630 netif_tx_stop_queue(netdev_get_tx_queue(dev
, queue
->id
));
632 spin_unlock_irqrestore(&queue
->tx_lock
, flags
);
637 dev
->stats
.tx_dropped
++;
638 dev_kfree_skb_any(skb
);
642 static int xennet_close(struct net_device
*dev
)
644 struct netfront_info
*np
= netdev_priv(dev
);
645 unsigned int num_queues
= dev
->real_num_tx_queues
;
647 struct netfront_queue
*queue
;
648 netif_tx_stop_all_queues(np
->netdev
);
649 for (i
= 0; i
< num_queues
; ++i
) {
650 queue
= &np
->queues
[i
];
651 napi_disable(&queue
->napi
);
656 static void xennet_move_rx_slot(struct netfront_queue
*queue
, struct sk_buff
*skb
,
659 int new = xennet_rxidx(queue
->rx
.req_prod_pvt
);
661 BUG_ON(queue
->rx_skbs
[new]);
662 queue
->rx_skbs
[new] = skb
;
663 queue
->grant_rx_ref
[new] = ref
;
664 RING_GET_REQUEST(&queue
->rx
, queue
->rx
.req_prod_pvt
)->id
= new;
665 RING_GET_REQUEST(&queue
->rx
, queue
->rx
.req_prod_pvt
)->gref
= ref
;
666 queue
->rx
.req_prod_pvt
++;
669 static int xennet_get_extras(struct netfront_queue
*queue
,
670 struct xen_netif_extra_info
*extras
,
674 struct xen_netif_extra_info
*extra
;
675 struct device
*dev
= &queue
->info
->netdev
->dev
;
676 RING_IDX cons
= queue
->rx
.rsp_cons
;
683 if (unlikely(cons
+ 1 == rp
)) {
685 dev_warn(dev
, "Missing extra info\n");
690 extra
= (struct xen_netif_extra_info
*)
691 RING_GET_RESPONSE(&queue
->rx
, ++cons
);
693 if (unlikely(!extra
->type
||
694 extra
->type
>= XEN_NETIF_EXTRA_TYPE_MAX
)) {
696 dev_warn(dev
, "Invalid extra type: %d\n",
700 memcpy(&extras
[extra
->type
- 1], extra
,
704 skb
= xennet_get_rx_skb(queue
, cons
);
705 ref
= xennet_get_rx_ref(queue
, cons
);
706 xennet_move_rx_slot(queue
, skb
, ref
);
707 } while (extra
->flags
& XEN_NETIF_EXTRA_FLAG_MORE
);
709 queue
->rx
.rsp_cons
= cons
;
713 static int xennet_get_responses(struct netfront_queue
*queue
,
714 struct netfront_rx_info
*rinfo
, RING_IDX rp
,
715 struct sk_buff_head
*list
)
717 struct xen_netif_rx_response
*rx
= &rinfo
->rx
;
718 struct xen_netif_extra_info
*extras
= rinfo
->extras
;
719 struct device
*dev
= &queue
->info
->netdev
->dev
;
720 RING_IDX cons
= queue
->rx
.rsp_cons
;
721 struct sk_buff
*skb
= xennet_get_rx_skb(queue
, cons
);
722 grant_ref_t ref
= xennet_get_rx_ref(queue
, cons
);
723 int max
= MAX_SKB_FRAGS
+ (rx
->status
<= RX_COPY_THRESHOLD
);
728 if (rx
->flags
& XEN_NETRXF_extra_info
) {
729 err
= xennet_get_extras(queue
, extras
, rp
);
730 cons
= queue
->rx
.rsp_cons
;
734 if (unlikely(rx
->status
< 0 ||
735 rx
->offset
+ rx
->status
> PAGE_SIZE
)) {
737 dev_warn(dev
, "rx->offset: %u, size: %d\n",
738 rx
->offset
, rx
->status
);
739 xennet_move_rx_slot(queue
, skb
, ref
);
745 * This definitely indicates a bug, either in this driver or in
746 * the backend driver. In future this should flag the bad
747 * situation to the system controller to reboot the backend.
749 if (ref
== GRANT_INVALID_REF
) {
751 dev_warn(dev
, "Bad rx response id %d.\n",
757 ret
= gnttab_end_foreign_access_ref(ref
, 0);
760 gnttab_release_grant_reference(&queue
->gref_rx_head
, ref
);
762 __skb_queue_tail(list
, skb
);
765 if (!(rx
->flags
& XEN_NETRXF_more_data
))
768 if (cons
+ slots
== rp
) {
770 dev_warn(dev
, "Need more slots\n");
775 rx
= RING_GET_RESPONSE(&queue
->rx
, cons
+ slots
);
776 skb
= xennet_get_rx_skb(queue
, cons
+ slots
);
777 ref
= xennet_get_rx_ref(queue
, cons
+ slots
);
781 if (unlikely(slots
> max
)) {
783 dev_warn(dev
, "Too many slots\n");
788 queue
->rx
.rsp_cons
= cons
+ slots
;
793 static int xennet_set_skb_gso(struct sk_buff
*skb
,
794 struct xen_netif_extra_info
*gso
)
796 if (!gso
->u
.gso
.size
) {
798 pr_warn("GSO size must not be zero\n");
802 if (gso
->u
.gso
.type
!= XEN_NETIF_GSO_TYPE_TCPV4
&&
803 gso
->u
.gso
.type
!= XEN_NETIF_GSO_TYPE_TCPV6
) {
805 pr_warn("Bad GSO type %d\n", gso
->u
.gso
.type
);
809 skb_shinfo(skb
)->gso_size
= gso
->u
.gso
.size
;
810 skb_shinfo(skb
)->gso_type
=
811 (gso
->u
.gso
.type
== XEN_NETIF_GSO_TYPE_TCPV4
) ?
815 /* Header must be checked, and gso_segs computed. */
816 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
817 skb_shinfo(skb
)->gso_segs
= 0;
822 static RING_IDX
xennet_fill_frags(struct netfront_queue
*queue
,
824 struct sk_buff_head
*list
)
826 struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
827 RING_IDX cons
= queue
->rx
.rsp_cons
;
828 struct sk_buff
*nskb
;
830 while ((nskb
= __skb_dequeue(list
))) {
831 struct xen_netif_rx_response
*rx
=
832 RING_GET_RESPONSE(&queue
->rx
, ++cons
);
833 skb_frag_t
*nfrag
= &skb_shinfo(nskb
)->frags
[0];
835 if (shinfo
->nr_frags
== MAX_SKB_FRAGS
) {
836 unsigned int pull_to
= NETFRONT_SKB_CB(skb
)->pull_to
;
838 BUG_ON(pull_to
<= skb_headlen(skb
));
839 __pskb_pull_tail(skb
, pull_to
- skb_headlen(skb
));
841 BUG_ON(shinfo
->nr_frags
>= MAX_SKB_FRAGS
);
843 skb_add_rx_frag(skb
, shinfo
->nr_frags
, skb_frag_page(nfrag
),
844 rx
->offset
, rx
->status
, PAGE_SIZE
);
846 skb_shinfo(nskb
)->nr_frags
= 0;
853 static int checksum_setup(struct net_device
*dev
, struct sk_buff
*skb
)
855 bool recalculate_partial_csum
= false;
858 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
859 * peers can fail to set NETRXF_csum_blank when sending a GSO
860 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
861 * recalculate the partial checksum.
863 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
&& skb_is_gso(skb
)) {
864 struct netfront_info
*np
= netdev_priv(dev
);
865 atomic_inc(&np
->rx_gso_checksum_fixup
);
866 skb
->ip_summed
= CHECKSUM_PARTIAL
;
867 recalculate_partial_csum
= true;
870 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
871 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
)
874 return skb_checksum_setup(skb
, recalculate_partial_csum
);
877 static int handle_incoming_queue(struct netfront_queue
*queue
,
878 struct sk_buff_head
*rxq
)
880 struct netfront_stats
*rx_stats
= this_cpu_ptr(queue
->info
->rx_stats
);
881 int packets_dropped
= 0;
884 while ((skb
= __skb_dequeue(rxq
)) != NULL
) {
885 int pull_to
= NETFRONT_SKB_CB(skb
)->pull_to
;
887 if (pull_to
> skb_headlen(skb
))
888 __pskb_pull_tail(skb
, pull_to
- skb_headlen(skb
));
890 /* Ethernet work: Delayed to here as it peeks the header. */
891 skb
->protocol
= eth_type_trans(skb
, queue
->info
->netdev
);
892 skb_reset_network_header(skb
);
894 if (checksum_setup(queue
->info
->netdev
, skb
)) {
897 queue
->info
->netdev
->stats
.rx_errors
++;
901 u64_stats_update_begin(&rx_stats
->syncp
);
903 rx_stats
->bytes
+= skb
->len
;
904 u64_stats_update_end(&rx_stats
->syncp
);
907 napi_gro_receive(&queue
->napi
, skb
);
910 return packets_dropped
;
913 static int xennet_poll(struct napi_struct
*napi
, int budget
)
915 struct netfront_queue
*queue
= container_of(napi
, struct netfront_queue
, napi
);
916 struct net_device
*dev
= queue
->info
->netdev
;
918 struct netfront_rx_info rinfo
;
919 struct xen_netif_rx_response
*rx
= &rinfo
.rx
;
920 struct xen_netif_extra_info
*extras
= rinfo
.extras
;
923 struct sk_buff_head rxq
;
924 struct sk_buff_head errq
;
925 struct sk_buff_head tmpq
;
928 spin_lock(&queue
->rx_lock
);
930 skb_queue_head_init(&rxq
);
931 skb_queue_head_init(&errq
);
932 skb_queue_head_init(&tmpq
);
934 rp
= queue
->rx
.sring
->rsp_prod
;
935 rmb(); /* Ensure we see queued responses up to 'rp'. */
937 i
= queue
->rx
.rsp_cons
;
939 while ((i
!= rp
) && (work_done
< budget
)) {
940 memcpy(rx
, RING_GET_RESPONSE(&queue
->rx
, i
), sizeof(*rx
));
941 memset(extras
, 0, sizeof(rinfo
.extras
));
943 err
= xennet_get_responses(queue
, &rinfo
, rp
, &tmpq
);
947 while ((skb
= __skb_dequeue(&tmpq
)))
948 __skb_queue_tail(&errq
, skb
);
949 dev
->stats
.rx_errors
++;
950 i
= queue
->rx
.rsp_cons
;
954 skb
= __skb_dequeue(&tmpq
);
956 if (extras
[XEN_NETIF_EXTRA_TYPE_GSO
- 1].type
) {
957 struct xen_netif_extra_info
*gso
;
958 gso
= &extras
[XEN_NETIF_EXTRA_TYPE_GSO
- 1];
960 if (unlikely(xennet_set_skb_gso(skb
, gso
))) {
961 __skb_queue_head(&tmpq
, skb
);
962 queue
->rx
.rsp_cons
+= skb_queue_len(&tmpq
);
967 NETFRONT_SKB_CB(skb
)->pull_to
= rx
->status
;
968 if (NETFRONT_SKB_CB(skb
)->pull_to
> RX_COPY_THRESHOLD
)
969 NETFRONT_SKB_CB(skb
)->pull_to
= RX_COPY_THRESHOLD
;
971 skb_shinfo(skb
)->frags
[0].page_offset
= rx
->offset
;
972 skb_frag_size_set(&skb_shinfo(skb
)->frags
[0], rx
->status
);
973 skb
->data_len
= rx
->status
;
974 skb
->len
+= rx
->status
;
976 i
= xennet_fill_frags(queue
, skb
, &tmpq
);
978 if (rx
->flags
& XEN_NETRXF_csum_blank
)
979 skb
->ip_summed
= CHECKSUM_PARTIAL
;
980 else if (rx
->flags
& XEN_NETRXF_data_validated
)
981 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
983 __skb_queue_tail(&rxq
, skb
);
985 queue
->rx
.rsp_cons
= ++i
;
989 __skb_queue_purge(&errq
);
991 work_done
-= handle_incoming_queue(queue
, &rxq
);
993 xennet_alloc_rx_buffers(queue
);
995 if (work_done
< budget
) {
1000 RING_FINAL_CHECK_FOR_RESPONSES(&queue
->rx
, more_to_do
);
1002 napi_schedule(napi
);
1005 spin_unlock(&queue
->rx_lock
);
1010 static int xennet_change_mtu(struct net_device
*dev
, int mtu
)
1012 int max
= xennet_can_sg(dev
) ? XEN_NETIF_MAX_TX_SIZE
: ETH_DATA_LEN
;
1020 static struct rtnl_link_stats64
*xennet_get_stats64(struct net_device
*dev
,
1021 struct rtnl_link_stats64
*tot
)
1023 struct netfront_info
*np
= netdev_priv(dev
);
1026 for_each_possible_cpu(cpu
) {
1027 struct netfront_stats
*rx_stats
= per_cpu_ptr(np
->rx_stats
, cpu
);
1028 struct netfront_stats
*tx_stats
= per_cpu_ptr(np
->tx_stats
, cpu
);
1029 u64 rx_packets
, rx_bytes
, tx_packets
, tx_bytes
;
1033 start
= u64_stats_fetch_begin_irq(&tx_stats
->syncp
);
1034 tx_packets
= tx_stats
->packets
;
1035 tx_bytes
= tx_stats
->bytes
;
1036 } while (u64_stats_fetch_retry_irq(&tx_stats
->syncp
, start
));
1039 start
= u64_stats_fetch_begin_irq(&rx_stats
->syncp
);
1040 rx_packets
= rx_stats
->packets
;
1041 rx_bytes
= rx_stats
->bytes
;
1042 } while (u64_stats_fetch_retry_irq(&rx_stats
->syncp
, start
));
1044 tot
->rx_packets
+= rx_packets
;
1045 tot
->tx_packets
+= tx_packets
;
1046 tot
->rx_bytes
+= rx_bytes
;
1047 tot
->tx_bytes
+= tx_bytes
;
1050 tot
->rx_errors
= dev
->stats
.rx_errors
;
1051 tot
->tx_dropped
= dev
->stats
.tx_dropped
;
1056 static void xennet_release_tx_bufs(struct netfront_queue
*queue
)
1058 struct sk_buff
*skb
;
1061 for (i
= 0; i
< NET_TX_RING_SIZE
; i
++) {
1062 /* Skip over entries which are actually freelist references */
1063 if (skb_entry_is_link(&queue
->tx_skbs
[i
]))
1066 skb
= queue
->tx_skbs
[i
].skb
;
1067 get_page(queue
->grant_tx_page
[i
]);
1068 gnttab_end_foreign_access(queue
->grant_tx_ref
[i
],
1070 (unsigned long)page_address(queue
->grant_tx_page
[i
]));
1071 queue
->grant_tx_page
[i
] = NULL
;
1072 queue
->grant_tx_ref
[i
] = GRANT_INVALID_REF
;
1073 add_id_to_freelist(&queue
->tx_skb_freelist
, queue
->tx_skbs
, i
);
1074 dev_kfree_skb_irq(skb
);
1078 static void xennet_release_rx_bufs(struct netfront_queue
*queue
)
1082 spin_lock_bh(&queue
->rx_lock
);
1084 for (id
= 0; id
< NET_RX_RING_SIZE
; id
++) {
1085 struct sk_buff
*skb
;
1088 skb
= queue
->rx_skbs
[id
];
1092 ref
= queue
->grant_rx_ref
[id
];
1093 if (ref
== GRANT_INVALID_REF
)
1096 page
= skb_frag_page(&skb_shinfo(skb
)->frags
[0]);
1098 /* gnttab_end_foreign_access() needs a page ref until
1099 * foreign access is ended (which may be deferred).
1102 gnttab_end_foreign_access(ref
, 0,
1103 (unsigned long)page_address(page
));
1104 queue
->grant_rx_ref
[id
] = GRANT_INVALID_REF
;
1109 spin_unlock_bh(&queue
->rx_lock
);
1112 static netdev_features_t
xennet_fix_features(struct net_device
*dev
,
1113 netdev_features_t features
)
1115 struct netfront_info
*np
= netdev_priv(dev
);
1118 if (features
& NETIF_F_SG
) {
1119 if (xenbus_scanf(XBT_NIL
, np
->xbdev
->otherend
, "feature-sg",
1124 features
&= ~NETIF_F_SG
;
1127 if (features
& NETIF_F_IPV6_CSUM
) {
1128 if (xenbus_scanf(XBT_NIL
, np
->xbdev
->otherend
,
1129 "feature-ipv6-csum-offload", "%d", &val
) < 0)
1133 features
&= ~NETIF_F_IPV6_CSUM
;
1136 if (features
& NETIF_F_TSO
) {
1137 if (xenbus_scanf(XBT_NIL
, np
->xbdev
->otherend
,
1138 "feature-gso-tcpv4", "%d", &val
) < 0)
1142 features
&= ~NETIF_F_TSO
;
1145 if (features
& NETIF_F_TSO6
) {
1146 if (xenbus_scanf(XBT_NIL
, np
->xbdev
->otherend
,
1147 "feature-gso-tcpv6", "%d", &val
) < 0)
1151 features
&= ~NETIF_F_TSO6
;
1157 static int xennet_set_features(struct net_device
*dev
,
1158 netdev_features_t features
)
1160 if (!(features
& NETIF_F_SG
) && dev
->mtu
> ETH_DATA_LEN
) {
1161 netdev_info(dev
, "Reducing MTU because no SG offload");
1162 dev
->mtu
= ETH_DATA_LEN
;
1168 static irqreturn_t
xennet_tx_interrupt(int irq
, void *dev_id
)
1170 struct netfront_queue
*queue
= dev_id
;
1171 unsigned long flags
;
1173 spin_lock_irqsave(&queue
->tx_lock
, flags
);
1174 xennet_tx_buf_gc(queue
);
1175 spin_unlock_irqrestore(&queue
->tx_lock
, flags
);
1180 static irqreturn_t
xennet_rx_interrupt(int irq
, void *dev_id
)
1182 struct netfront_queue
*queue
= dev_id
;
1183 struct net_device
*dev
= queue
->info
->netdev
;
1185 if (likely(netif_carrier_ok(dev
) &&
1186 RING_HAS_UNCONSUMED_RESPONSES(&queue
->rx
)))
1187 napi_schedule(&queue
->napi
);
1192 static irqreturn_t
xennet_interrupt(int irq
, void *dev_id
)
1194 xennet_tx_interrupt(irq
, dev_id
);
1195 xennet_rx_interrupt(irq
, dev_id
);
1199 #ifdef CONFIG_NET_POLL_CONTROLLER
1200 static void xennet_poll_controller(struct net_device
*dev
)
1202 /* Poll each queue */
1203 struct netfront_info
*info
= netdev_priv(dev
);
1204 unsigned int num_queues
= dev
->real_num_tx_queues
;
1206 for (i
= 0; i
< num_queues
; ++i
)
1207 xennet_interrupt(0, &info
->queues
[i
]);
1211 static const struct net_device_ops xennet_netdev_ops
= {
1212 .ndo_open
= xennet_open
,
1213 .ndo_stop
= xennet_close
,
1214 .ndo_start_xmit
= xennet_start_xmit
,
1215 .ndo_change_mtu
= xennet_change_mtu
,
1216 .ndo_get_stats64
= xennet_get_stats64
,
1217 .ndo_set_mac_address
= eth_mac_addr
,
1218 .ndo_validate_addr
= eth_validate_addr
,
1219 .ndo_fix_features
= xennet_fix_features
,
1220 .ndo_set_features
= xennet_set_features
,
1221 .ndo_select_queue
= xennet_select_queue
,
1222 #ifdef CONFIG_NET_POLL_CONTROLLER
1223 .ndo_poll_controller
= xennet_poll_controller
,
1227 static void xennet_free_netdev(struct net_device
*netdev
)
1229 struct netfront_info
*np
= netdev_priv(netdev
);
1231 free_percpu(np
->rx_stats
);
1232 free_percpu(np
->tx_stats
);
1233 free_netdev(netdev
);
1236 static struct net_device
*xennet_create_dev(struct xenbus_device
*dev
)
1239 struct net_device
*netdev
;
1240 struct netfront_info
*np
;
1242 netdev
= alloc_etherdev_mq(sizeof(struct netfront_info
), xennet_max_queues
);
1244 return ERR_PTR(-ENOMEM
);
1246 np
= netdev_priv(netdev
);
1252 np
->rx_stats
= netdev_alloc_pcpu_stats(struct netfront_stats
);
1253 if (np
->rx_stats
== NULL
)
1255 np
->tx_stats
= netdev_alloc_pcpu_stats(struct netfront_stats
);
1256 if (np
->tx_stats
== NULL
)
1259 netdev
->netdev_ops
= &xennet_netdev_ops
;
1261 netdev
->features
= NETIF_F_IP_CSUM
| NETIF_F_RXCSUM
|
1263 netdev
->hw_features
= NETIF_F_SG
|
1265 NETIF_F_TSO
| NETIF_F_TSO6
;
1268 * Assume that all hw features are available for now. This set
1269 * will be adjusted by the call to netdev_update_features() in
1270 * xennet_connect() which is the earliest point where we can
1271 * negotiate with the backend regarding supported features.
1273 netdev
->features
|= netdev
->hw_features
;
1275 netdev
->ethtool_ops
= &xennet_ethtool_ops
;
1276 SET_NETDEV_DEV(netdev
, &dev
->dev
);
1278 np
->netdev
= netdev
;
1280 netif_carrier_off(netdev
);
1285 xennet_free_netdev(netdev
);
1286 return ERR_PTR(err
);
1290 * Entry point to this code when a new device is created. Allocate the basic
1291 * structures and the ring buffers for communication with the backend, and
1292 * inform the backend of the appropriate details for those.
1294 static int netfront_probe(struct xenbus_device
*dev
,
1295 const struct xenbus_device_id
*id
)
1298 struct net_device
*netdev
;
1299 struct netfront_info
*info
;
1301 netdev
= xennet_create_dev(dev
);
1302 if (IS_ERR(netdev
)) {
1303 err
= PTR_ERR(netdev
);
1304 xenbus_dev_fatal(dev
, err
, "creating netdev");
1308 info
= netdev_priv(netdev
);
1309 dev_set_drvdata(&dev
->dev
, info
);
1311 info
->netdev
->sysfs_groups
[0] = &xennet_dev_group
;
1313 err
= register_netdev(info
->netdev
);
1315 pr_warn("%s: register_netdev err=%d\n", __func__
, err
);
1322 xennet_free_netdev(netdev
);
1323 dev_set_drvdata(&dev
->dev
, NULL
);
1327 static void xennet_end_access(int ref
, void *page
)
1329 /* This frees the page as a side-effect */
1330 if (ref
!= GRANT_INVALID_REF
)
1331 gnttab_end_foreign_access(ref
, 0, (unsigned long)page
);
1334 static void xennet_disconnect_backend(struct netfront_info
*info
)
1337 unsigned int num_queues
= info
->netdev
->real_num_tx_queues
;
1339 netif_carrier_off(info
->netdev
);
1341 for (i
= 0; i
< num_queues
&& info
->queues
; ++i
) {
1342 struct netfront_queue
*queue
= &info
->queues
[i
];
1344 if (queue
->tx_irq
&& (queue
->tx_irq
== queue
->rx_irq
))
1345 unbind_from_irqhandler(queue
->tx_irq
, queue
);
1346 if (queue
->tx_irq
&& (queue
->tx_irq
!= queue
->rx_irq
)) {
1347 unbind_from_irqhandler(queue
->tx_irq
, queue
);
1348 unbind_from_irqhandler(queue
->rx_irq
, queue
);
1350 queue
->tx_evtchn
= queue
->rx_evtchn
= 0;
1351 queue
->tx_irq
= queue
->rx_irq
= 0;
1353 if (netif_running(info
->netdev
))
1354 napi_synchronize(&queue
->napi
);
1356 xennet_release_tx_bufs(queue
);
1357 xennet_release_rx_bufs(queue
);
1358 gnttab_free_grant_references(queue
->gref_tx_head
);
1359 gnttab_free_grant_references(queue
->gref_rx_head
);
1361 /* End access and free the pages */
1362 xennet_end_access(queue
->tx_ring_ref
, queue
->tx
.sring
);
1363 xennet_end_access(queue
->rx_ring_ref
, queue
->rx
.sring
);
1365 queue
->tx_ring_ref
= GRANT_INVALID_REF
;
1366 queue
->rx_ring_ref
= GRANT_INVALID_REF
;
1367 queue
->tx
.sring
= NULL
;
1368 queue
->rx
.sring
= NULL
;
1373 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1374 * driver restart. We tear down our netif structure and recreate it, but
1375 * leave the device-layer structures intact so that this is transparent to the
1376 * rest of the kernel.
1378 static int netfront_resume(struct xenbus_device
*dev
)
1380 struct netfront_info
*info
= dev_get_drvdata(&dev
->dev
);
1382 dev_dbg(&dev
->dev
, "%s\n", dev
->nodename
);
1384 xennet_disconnect_backend(info
);
1388 static int xen_net_read_mac(struct xenbus_device
*dev
, u8 mac
[])
1390 char *s
, *e
, *macstr
;
1393 macstr
= s
= xenbus_read(XBT_NIL
, dev
->nodename
, "mac", NULL
);
1395 return PTR_ERR(macstr
);
1397 for (i
= 0; i
< ETH_ALEN
; i
++) {
1398 mac
[i
] = simple_strtoul(s
, &e
, 16);
1399 if ((s
== e
) || (*e
!= ((i
== ETH_ALEN
-1) ? '\0' : ':'))) {
1410 static int setup_netfront_single(struct netfront_queue
*queue
)
1414 err
= xenbus_alloc_evtchn(queue
->info
->xbdev
, &queue
->tx_evtchn
);
1418 err
= bind_evtchn_to_irqhandler(queue
->tx_evtchn
,
1420 0, queue
->info
->netdev
->name
, queue
);
1423 queue
->rx_evtchn
= queue
->tx_evtchn
;
1424 queue
->rx_irq
= queue
->tx_irq
= err
;
1429 xenbus_free_evtchn(queue
->info
->xbdev
, queue
->tx_evtchn
);
1430 queue
->tx_evtchn
= 0;
1435 static int setup_netfront_split(struct netfront_queue
*queue
)
1439 err
= xenbus_alloc_evtchn(queue
->info
->xbdev
, &queue
->tx_evtchn
);
1442 err
= xenbus_alloc_evtchn(queue
->info
->xbdev
, &queue
->rx_evtchn
);
1444 goto alloc_rx_evtchn_fail
;
1446 snprintf(queue
->tx_irq_name
, sizeof(queue
->tx_irq_name
),
1447 "%s-tx", queue
->name
);
1448 err
= bind_evtchn_to_irqhandler(queue
->tx_evtchn
,
1449 xennet_tx_interrupt
,
1450 0, queue
->tx_irq_name
, queue
);
1453 queue
->tx_irq
= err
;
1455 snprintf(queue
->rx_irq_name
, sizeof(queue
->rx_irq_name
),
1456 "%s-rx", queue
->name
);
1457 err
= bind_evtchn_to_irqhandler(queue
->rx_evtchn
,
1458 xennet_rx_interrupt
,
1459 0, queue
->rx_irq_name
, queue
);
1462 queue
->rx_irq
= err
;
1467 unbind_from_irqhandler(queue
->tx_irq
, queue
);
1470 xenbus_free_evtchn(queue
->info
->xbdev
, queue
->rx_evtchn
);
1471 queue
->rx_evtchn
= 0;
1472 alloc_rx_evtchn_fail
:
1473 xenbus_free_evtchn(queue
->info
->xbdev
, queue
->tx_evtchn
);
1474 queue
->tx_evtchn
= 0;
1479 static int setup_netfront(struct xenbus_device
*dev
,
1480 struct netfront_queue
*queue
, unsigned int feature_split_evtchn
)
1482 struct xen_netif_tx_sring
*txs
;
1483 struct xen_netif_rx_sring
*rxs
;
1487 queue
->tx_ring_ref
= GRANT_INVALID_REF
;
1488 queue
->rx_ring_ref
= GRANT_INVALID_REF
;
1489 queue
->rx
.sring
= NULL
;
1490 queue
->tx
.sring
= NULL
;
1492 txs
= (struct xen_netif_tx_sring
*)get_zeroed_page(GFP_NOIO
| __GFP_HIGH
);
1495 xenbus_dev_fatal(dev
, err
, "allocating tx ring page");
1498 SHARED_RING_INIT(txs
);
1499 FRONT_RING_INIT(&queue
->tx
, txs
, PAGE_SIZE
);
1501 err
= xenbus_grant_ring(dev
, txs
, 1, &gref
);
1503 goto grant_tx_ring_fail
;
1504 queue
->tx_ring_ref
= gref
;
1506 rxs
= (struct xen_netif_rx_sring
*)get_zeroed_page(GFP_NOIO
| __GFP_HIGH
);
1509 xenbus_dev_fatal(dev
, err
, "allocating rx ring page");
1510 goto alloc_rx_ring_fail
;
1512 SHARED_RING_INIT(rxs
);
1513 FRONT_RING_INIT(&queue
->rx
, rxs
, PAGE_SIZE
);
1515 err
= xenbus_grant_ring(dev
, rxs
, 1, &gref
);
1517 goto grant_rx_ring_fail
;
1518 queue
->rx_ring_ref
= gref
;
1520 if (feature_split_evtchn
)
1521 err
= setup_netfront_split(queue
);
1522 /* setup single event channel if
1523 * a) feature-split-event-channels == 0
1524 * b) feature-split-event-channels == 1 but failed to setup
1526 if (!feature_split_evtchn
|| (feature_split_evtchn
&& err
))
1527 err
= setup_netfront_single(queue
);
1530 goto alloc_evtchn_fail
;
1534 /* If we fail to setup netfront, it is safe to just revoke access to
1535 * granted pages because backend is not accessing it at this point.
1538 gnttab_end_foreign_access_ref(queue
->rx_ring_ref
, 0);
1540 free_page((unsigned long)rxs
);
1542 gnttab_end_foreign_access_ref(queue
->tx_ring_ref
, 0);
1544 free_page((unsigned long)txs
);
1549 /* Queue-specific initialisation
1550 * This used to be done in xennet_create_dev() but must now
1553 static int xennet_init_queue(struct netfront_queue
*queue
)
1558 spin_lock_init(&queue
->tx_lock
);
1559 spin_lock_init(&queue
->rx_lock
);
1561 setup_timer(&queue
->rx_refill_timer
, rx_refill_timeout
,
1562 (unsigned long)queue
);
1564 snprintf(queue
->name
, sizeof(queue
->name
), "%s-q%u",
1565 queue
->info
->netdev
->name
, queue
->id
);
1567 /* Initialise tx_skbs as a free chain containing every entry. */
1568 queue
->tx_skb_freelist
= 0;
1569 for (i
= 0; i
< NET_TX_RING_SIZE
; i
++) {
1570 skb_entry_set_link(&queue
->tx_skbs
[i
], i
+1);
1571 queue
->grant_tx_ref
[i
] = GRANT_INVALID_REF
;
1572 queue
->grant_tx_page
[i
] = NULL
;
1575 /* Clear out rx_skbs */
1576 for (i
= 0; i
< NET_RX_RING_SIZE
; i
++) {
1577 queue
->rx_skbs
[i
] = NULL
;
1578 queue
->grant_rx_ref
[i
] = GRANT_INVALID_REF
;
1581 /* A grant for every tx ring slot */
1582 if (gnttab_alloc_grant_references(NET_TX_RING_SIZE
,
1583 &queue
->gref_tx_head
) < 0) {
1584 pr_alert("can't alloc tx grant refs\n");
1589 /* A grant for every rx ring slot */
1590 if (gnttab_alloc_grant_references(NET_RX_RING_SIZE
,
1591 &queue
->gref_rx_head
) < 0) {
1592 pr_alert("can't alloc rx grant refs\n");
1600 gnttab_free_grant_references(queue
->gref_tx_head
);
1605 static int write_queue_xenstore_keys(struct netfront_queue
*queue
,
1606 struct xenbus_transaction
*xbt
, int write_hierarchical
)
1608 /* Write the queue-specific keys into XenStore in the traditional
1609 * way for a single queue, or in a queue subkeys for multiple
1612 struct xenbus_device
*dev
= queue
->info
->xbdev
;
1614 const char *message
;
1618 /* Choose the correct place to write the keys */
1619 if (write_hierarchical
) {
1620 pathsize
= strlen(dev
->nodename
) + 10;
1621 path
= kzalloc(pathsize
, GFP_KERNEL
);
1624 message
= "out of memory while writing ring references";
1627 snprintf(path
, pathsize
, "%s/queue-%u",
1628 dev
->nodename
, queue
->id
);
1630 path
= (char *)dev
->nodename
;
1633 /* Write ring references */
1634 err
= xenbus_printf(*xbt
, path
, "tx-ring-ref", "%u",
1635 queue
->tx_ring_ref
);
1637 message
= "writing tx-ring-ref";
1641 err
= xenbus_printf(*xbt
, path
, "rx-ring-ref", "%u",
1642 queue
->rx_ring_ref
);
1644 message
= "writing rx-ring-ref";
1648 /* Write event channels; taking into account both shared
1649 * and split event channel scenarios.
1651 if (queue
->tx_evtchn
== queue
->rx_evtchn
) {
1652 /* Shared event channel */
1653 err
= xenbus_printf(*xbt
, path
,
1654 "event-channel", "%u", queue
->tx_evtchn
);
1656 message
= "writing event-channel";
1660 /* Split event channels */
1661 err
= xenbus_printf(*xbt
, path
,
1662 "event-channel-tx", "%u", queue
->tx_evtchn
);
1664 message
= "writing event-channel-tx";
1668 err
= xenbus_printf(*xbt
, path
,
1669 "event-channel-rx", "%u", queue
->rx_evtchn
);
1671 message
= "writing event-channel-rx";
1676 if (write_hierarchical
)
1681 if (write_hierarchical
)
1683 xenbus_dev_fatal(dev
, err
, "%s", message
);
1687 static void xennet_destroy_queues(struct netfront_info
*info
)
1693 for (i
= 0; i
< info
->netdev
->real_num_tx_queues
; i
++) {
1694 struct netfront_queue
*queue
= &info
->queues
[i
];
1696 if (netif_running(info
->netdev
))
1697 napi_disable(&queue
->napi
);
1698 del_timer_sync(&queue
->rx_refill_timer
);
1699 netif_napi_del(&queue
->napi
);
1704 kfree(info
->queues
);
1705 info
->queues
= NULL
;
1708 static int xennet_create_queues(struct netfront_info
*info
,
1709 unsigned int *num_queues
)
1714 info
->queues
= kcalloc(*num_queues
, sizeof(struct netfront_queue
),
1721 for (i
= 0; i
< *num_queues
; i
++) {
1722 struct netfront_queue
*queue
= &info
->queues
[i
];
1727 ret
= xennet_init_queue(queue
);
1729 dev_warn(&info
->netdev
->dev
,
1730 "only created %d queues\n", i
);
1735 netif_napi_add(queue
->info
->netdev
, &queue
->napi
,
1737 if (netif_running(info
->netdev
))
1738 napi_enable(&queue
->napi
);
1741 netif_set_real_num_tx_queues(info
->netdev
, *num_queues
);
1745 if (*num_queues
== 0) {
1746 dev_err(&info
->netdev
->dev
, "no queues\n");
1752 /* Common code used when first setting up, and when resuming. */
1753 static int talk_to_netback(struct xenbus_device
*dev
,
1754 struct netfront_info
*info
)
1756 const char *message
;
1757 struct xenbus_transaction xbt
;
1759 unsigned int feature_split_evtchn
;
1761 unsigned int max_queues
= 0;
1762 struct netfront_queue
*queue
= NULL
;
1763 unsigned int num_queues
= 1;
1765 info
->netdev
->irq
= 0;
1767 /* Check if backend supports multiple queues */
1768 err
= xenbus_scanf(XBT_NIL
, info
->xbdev
->otherend
,
1769 "multi-queue-max-queues", "%u", &max_queues
);
1772 num_queues
= min(max_queues
, xennet_max_queues
);
1774 /* Check feature-split-event-channels */
1775 err
= xenbus_scanf(XBT_NIL
, info
->xbdev
->otherend
,
1776 "feature-split-event-channels", "%u",
1777 &feature_split_evtchn
);
1779 feature_split_evtchn
= 0;
1781 /* Read mac addr. */
1782 err
= xen_net_read_mac(dev
, info
->netdev
->dev_addr
);
1784 xenbus_dev_fatal(dev
, err
, "parsing %s/mac", dev
->nodename
);
1789 xennet_destroy_queues(info
);
1791 err
= xennet_create_queues(info
, &num_queues
);
1795 /* Create shared ring, alloc event channel -- for each queue */
1796 for (i
= 0; i
< num_queues
; ++i
) {
1797 queue
= &info
->queues
[i
];
1798 err
= setup_netfront(dev
, queue
, feature_split_evtchn
);
1800 /* setup_netfront() will tidy up the current
1801 * queue on error, but we need to clean up
1802 * those already allocated.
1806 netif_set_real_num_tx_queues(info
->netdev
, i
);
1816 err
= xenbus_transaction_start(&xbt
);
1818 xenbus_dev_fatal(dev
, err
, "starting transaction");
1822 if (num_queues
== 1) {
1823 err
= write_queue_xenstore_keys(&info
->queues
[0], &xbt
, 0); /* flat */
1825 goto abort_transaction_no_dev_fatal
;
1827 /* Write the number of queues */
1828 err
= xenbus_printf(xbt
, dev
->nodename
, "multi-queue-num-queues",
1831 message
= "writing multi-queue-num-queues";
1832 goto abort_transaction_no_dev_fatal
;
1835 /* Write the keys for each queue */
1836 for (i
= 0; i
< num_queues
; ++i
) {
1837 queue
= &info
->queues
[i
];
1838 err
= write_queue_xenstore_keys(queue
, &xbt
, 1); /* hierarchical */
1840 goto abort_transaction_no_dev_fatal
;
1844 /* The remaining keys are not queue-specific */
1845 err
= xenbus_printf(xbt
, dev
->nodename
, "request-rx-copy", "%u",
1848 message
= "writing request-rx-copy";
1849 goto abort_transaction
;
1852 err
= xenbus_printf(xbt
, dev
->nodename
, "feature-rx-notify", "%d", 1);
1854 message
= "writing feature-rx-notify";
1855 goto abort_transaction
;
1858 err
= xenbus_printf(xbt
, dev
->nodename
, "feature-sg", "%d", 1);
1860 message
= "writing feature-sg";
1861 goto abort_transaction
;
1864 err
= xenbus_printf(xbt
, dev
->nodename
, "feature-gso-tcpv4", "%d", 1);
1866 message
= "writing feature-gso-tcpv4";
1867 goto abort_transaction
;
1870 err
= xenbus_write(xbt
, dev
->nodename
, "feature-gso-tcpv6", "1");
1872 message
= "writing feature-gso-tcpv6";
1873 goto abort_transaction
;
1876 err
= xenbus_write(xbt
, dev
->nodename
, "feature-ipv6-csum-offload",
1879 message
= "writing feature-ipv6-csum-offload";
1880 goto abort_transaction
;
1883 err
= xenbus_transaction_end(xbt
, 0);
1887 xenbus_dev_fatal(dev
, err
, "completing transaction");
1894 xenbus_dev_fatal(dev
, err
, "%s", message
);
1895 abort_transaction_no_dev_fatal
:
1896 xenbus_transaction_end(xbt
, 1);
1898 xennet_disconnect_backend(info
);
1899 kfree(info
->queues
);
1900 info
->queues
= NULL
;
1905 static int xennet_connect(struct net_device
*dev
)
1907 struct netfront_info
*np
= netdev_priv(dev
);
1908 unsigned int num_queues
= 0;
1910 unsigned int feature_rx_copy
;
1912 struct netfront_queue
*queue
= NULL
;
1914 err
= xenbus_scanf(XBT_NIL
, np
->xbdev
->otherend
,
1915 "feature-rx-copy", "%u", &feature_rx_copy
);
1917 feature_rx_copy
= 0;
1919 if (!feature_rx_copy
) {
1921 "backend does not support copying receive path\n");
1925 err
= talk_to_netback(np
->xbdev
, np
);
1929 /* talk_to_netback() sets the correct number of queues */
1930 num_queues
= dev
->real_num_tx_queues
;
1933 netdev_update_features(dev
);
1937 * All public and private state should now be sane. Get
1938 * ready to start sending and receiving packets and give the driver
1939 * domain a kick because we've probably just requeued some
1942 netif_carrier_on(np
->netdev
);
1943 for (j
= 0; j
< num_queues
; ++j
) {
1944 queue
= &np
->queues
[j
];
1946 notify_remote_via_irq(queue
->tx_irq
);
1947 if (queue
->tx_irq
!= queue
->rx_irq
)
1948 notify_remote_via_irq(queue
->rx_irq
);
1950 spin_lock_irq(&queue
->tx_lock
);
1951 xennet_tx_buf_gc(queue
);
1952 spin_unlock_irq(&queue
->tx_lock
);
1954 spin_lock_bh(&queue
->rx_lock
);
1955 xennet_alloc_rx_buffers(queue
);
1956 spin_unlock_bh(&queue
->rx_lock
);
1963 * Callback received when the backend's state changes.
1965 static void netback_changed(struct xenbus_device
*dev
,
1966 enum xenbus_state backend_state
)
1968 struct netfront_info
*np
= dev_get_drvdata(&dev
->dev
);
1969 struct net_device
*netdev
= np
->netdev
;
1971 dev_dbg(&dev
->dev
, "%s\n", xenbus_strstate(backend_state
));
1973 switch (backend_state
) {
1974 case XenbusStateInitialising
:
1975 case XenbusStateInitialised
:
1976 case XenbusStateReconfiguring
:
1977 case XenbusStateReconfigured
:
1978 case XenbusStateUnknown
:
1981 case XenbusStateInitWait
:
1982 if (dev
->state
!= XenbusStateInitialising
)
1984 if (xennet_connect(netdev
) != 0)
1986 xenbus_switch_state(dev
, XenbusStateConnected
);
1989 case XenbusStateConnected
:
1990 netdev_notify_peers(netdev
);
1993 case XenbusStateClosed
:
1994 if (dev
->state
== XenbusStateClosed
)
1996 /* Missed the backend's CLOSING state -- fallthrough */
1997 case XenbusStateClosing
:
1998 xenbus_frontend_closed(dev
);
2003 static const struct xennet_stat
{
2004 char name
[ETH_GSTRING_LEN
];
2006 } xennet_stats
[] = {
2008 "rx_gso_checksum_fixup",
2009 offsetof(struct netfront_info
, rx_gso_checksum_fixup
)
2013 static int xennet_get_sset_count(struct net_device
*dev
, int string_set
)
2015 switch (string_set
) {
2017 return ARRAY_SIZE(xennet_stats
);
2023 static void xennet_get_ethtool_stats(struct net_device
*dev
,
2024 struct ethtool_stats
*stats
, u64
* data
)
2026 void *np
= netdev_priv(dev
);
2029 for (i
= 0; i
< ARRAY_SIZE(xennet_stats
); i
++)
2030 data
[i
] = atomic_read((atomic_t
*)(np
+ xennet_stats
[i
].offset
));
2033 static void xennet_get_strings(struct net_device
*dev
, u32 stringset
, u8
* data
)
2037 switch (stringset
) {
2039 for (i
= 0; i
< ARRAY_SIZE(xennet_stats
); i
++)
2040 memcpy(data
+ i
* ETH_GSTRING_LEN
,
2041 xennet_stats
[i
].name
, ETH_GSTRING_LEN
);
2046 static const struct ethtool_ops xennet_ethtool_ops
=
2048 .get_link
= ethtool_op_get_link
,
2050 .get_sset_count
= xennet_get_sset_count
,
2051 .get_ethtool_stats
= xennet_get_ethtool_stats
,
2052 .get_strings
= xennet_get_strings
,
2056 static ssize_t
show_rxbuf(struct device
*dev
,
2057 struct device_attribute
*attr
, char *buf
)
2059 return sprintf(buf
, "%lu\n", NET_RX_RING_SIZE
);
2062 static ssize_t
store_rxbuf(struct device
*dev
,
2063 struct device_attribute
*attr
,
2064 const char *buf
, size_t len
)
2067 unsigned long target
;
2069 if (!capable(CAP_NET_ADMIN
))
2072 target
= simple_strtoul(buf
, &endp
, 0);
2076 /* rxbuf_min and rxbuf_max are no longer configurable. */
2081 static DEVICE_ATTR(rxbuf_min
, S_IRUGO
|S_IWUSR
, show_rxbuf
, store_rxbuf
);
2082 static DEVICE_ATTR(rxbuf_max
, S_IRUGO
|S_IWUSR
, show_rxbuf
, store_rxbuf
);
2083 static DEVICE_ATTR(rxbuf_cur
, S_IRUGO
, show_rxbuf
, NULL
);
2085 static struct attribute
*xennet_dev_attrs
[] = {
2086 &dev_attr_rxbuf_min
.attr
,
2087 &dev_attr_rxbuf_max
.attr
,
2088 &dev_attr_rxbuf_cur
.attr
,
2092 static const struct attribute_group xennet_dev_group
= {
2093 .attrs
= xennet_dev_attrs
2095 #endif /* CONFIG_SYSFS */
2097 static int xennet_remove(struct xenbus_device
*dev
)
2099 struct netfront_info
*info
= dev_get_drvdata(&dev
->dev
);
2101 dev_dbg(&dev
->dev
, "%s\n", dev
->nodename
);
2103 xennet_disconnect_backend(info
);
2105 unregister_netdev(info
->netdev
);
2108 xennet_destroy_queues(info
);
2109 xennet_free_netdev(info
->netdev
);
2114 static const struct xenbus_device_id netfront_ids
[] = {
2119 static struct xenbus_driver netfront_driver
= {
2120 .ids
= netfront_ids
,
2121 .probe
= netfront_probe
,
2122 .remove
= xennet_remove
,
2123 .resume
= netfront_resume
,
2124 .otherend_changed
= netback_changed
,
2127 static int __init
netif_init(void)
2132 if (!xen_has_pv_nic_devices())
2135 pr_info("Initialising Xen virtual ethernet driver\n");
2137 /* Allow as many queues as there are CPUs if user has not
2138 * specified a value.
2140 if (xennet_max_queues
== 0)
2141 xennet_max_queues
= num_online_cpus();
2143 return xenbus_register_frontend(&netfront_driver
);
2145 module_init(netif_init
);
2148 static void __exit
netif_exit(void)
2150 xenbus_unregister_driver(&netfront_driver
);
2152 module_exit(netif_exit
);
2154 MODULE_DESCRIPTION("Xen virtual network device frontend");
2155 MODULE_LICENSE("GPL");
2156 MODULE_ALIAS("xen:vif");
2157 MODULE_ALIAS("xennet");