2 * Virtual network driver for conversing with remote driver backends.
4 * Copyright (c) 2002-2005, K A Fraser
5 * Copyright (c) 2005, XenSource Ltd
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License version 2
9 * as published by the Free Software Foundation; or, when distributed
10 * separately from the Linux kernel or incorporated into other
11 * software packages, subject to the following license:
13 * Permission is hereby granted, free of charge, to any person obtaining a copy
14 * of this source file (the "Software"), to deal in the Software without
15 * restriction, including without limitation the rights to use, copy, modify,
16 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
17 * and to permit persons to whom the Software is furnished to do so, subject to
18 * the following conditions:
20 * The above copyright notice and this permission notice shall be included in
21 * all copies or substantial portions of the Software.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
24 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
25 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
26 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
27 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
28 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
32 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
34 #include <linux/module.h>
35 #include <linux/kernel.h>
36 #include <linux/netdevice.h>
37 #include <linux/etherdevice.h>
38 #include <linux/skbuff.h>
39 #include <linux/ethtool.h>
40 #include <linux/if_ether.h>
42 #include <linux/udp.h>
43 #include <linux/moduleparam.h>
45 #include <linux/slab.h>
48 #include <asm/xen/page.h>
50 #include <xen/xenbus.h>
51 #include <xen/events.h>
53 #include <xen/platform_pci.h>
54 #include <xen/grant_table.h>
56 #include <xen/interface/io/netif.h>
57 #include <xen/interface/memory.h>
58 #include <xen/interface/grant_table.h>
60 /* Module parameters */
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, PAGE_SIZE)
79 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, 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 struct netfront_stats
{
93 struct u64_stats_sync syncp
;
98 struct netfront_queue
{
99 unsigned int id
; /* Queue ID, 0-based */
100 char name
[QUEUE_NAME_SIZE
]; /* DEVNAME-qN */
101 struct netfront_info
*info
;
103 struct napi_struct napi
;
105 /* Split event channels support, tx_* == rx_* when using
106 * single event channel.
108 unsigned int tx_evtchn
, rx_evtchn
;
109 unsigned int tx_irq
, rx_irq
;
110 /* Only used when split event channels support is enabled */
111 char tx_irq_name
[IRQ_NAME_SIZE
]; /* DEVNAME-qN-tx */
112 char rx_irq_name
[IRQ_NAME_SIZE
]; /* DEVNAME-qN-rx */
115 struct xen_netif_tx_front_ring tx
;
119 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
120 * are linked from tx_skb_freelist through skb_entry.link.
122 * NB. Freelist index entries are always going to be less than
123 * PAGE_OFFSET, whereas pointers to skbs will always be equal or
124 * greater than PAGE_OFFSET: we use this property to distinguish
130 } tx_skbs
[NET_TX_RING_SIZE
];
131 grant_ref_t gref_tx_head
;
132 grant_ref_t grant_tx_ref
[NET_TX_RING_SIZE
];
133 struct page
*grant_tx_page
[NET_TX_RING_SIZE
];
134 unsigned tx_skb_freelist
;
136 spinlock_t rx_lock ____cacheline_aligned_in_smp
;
137 struct xen_netif_rx_front_ring rx
;
140 struct timer_list rx_refill_timer
;
142 struct sk_buff
*rx_skbs
[NET_RX_RING_SIZE
];
143 grant_ref_t gref_rx_head
;
144 grant_ref_t grant_rx_ref
[NET_RX_RING_SIZE
];
147 struct netfront_info
{
148 struct list_head list
;
149 struct net_device
*netdev
;
151 struct xenbus_device
*xbdev
;
153 /* Multi-queue support */
154 struct netfront_queue
*queues
;
157 struct netfront_stats __percpu
*rx_stats
;
158 struct netfront_stats __percpu
*tx_stats
;
160 atomic_t rx_gso_checksum_fixup
;
163 struct netfront_rx_info
{
164 struct xen_netif_rx_response rx
;
165 struct xen_netif_extra_info extras
[XEN_NETIF_EXTRA_TYPE_MAX
- 1];
168 static void skb_entry_set_link(union skb_entry
*list
, unsigned short id
)
173 static int skb_entry_is_link(const union skb_entry
*list
)
175 BUILD_BUG_ON(sizeof(list
->skb
) != sizeof(list
->link
));
176 return (unsigned long)list
->skb
< PAGE_OFFSET
;
180 * Access macros for acquiring freeing slots in tx_skbs[].
183 static void add_id_to_freelist(unsigned *head
, union skb_entry
*list
,
186 skb_entry_set_link(&list
[id
], *head
);
190 static unsigned short get_id_from_freelist(unsigned *head
,
191 union skb_entry
*list
)
193 unsigned int id
= *head
;
194 *head
= list
[id
].link
;
198 static int xennet_rxidx(RING_IDX idx
)
200 return idx
& (NET_RX_RING_SIZE
- 1);
203 static struct sk_buff
*xennet_get_rx_skb(struct netfront_queue
*queue
,
206 int i
= xennet_rxidx(ri
);
207 struct sk_buff
*skb
= queue
->rx_skbs
[i
];
208 queue
->rx_skbs
[i
] = NULL
;
212 static grant_ref_t
xennet_get_rx_ref(struct netfront_queue
*queue
,
215 int i
= xennet_rxidx(ri
);
216 grant_ref_t ref
= queue
->grant_rx_ref
[i
];
217 queue
->grant_rx_ref
[i
] = GRANT_INVALID_REF
;
222 static const struct attribute_group xennet_dev_group
;
225 static bool xennet_can_sg(struct net_device
*dev
)
227 return dev
->features
& NETIF_F_SG
;
231 static void rx_refill_timeout(unsigned long data
)
233 struct netfront_queue
*queue
= (struct netfront_queue
*)data
;
234 napi_schedule(&queue
->napi
);
237 static int netfront_tx_slot_available(struct netfront_queue
*queue
)
239 return (queue
->tx
.req_prod_pvt
- queue
->tx
.rsp_cons
) <
240 (NET_TX_RING_SIZE
- MAX_SKB_FRAGS
- 2);
243 static void xennet_maybe_wake_tx(struct netfront_queue
*queue
)
245 struct net_device
*dev
= queue
->info
->netdev
;
246 struct netdev_queue
*dev_queue
= netdev_get_tx_queue(dev
, queue
->id
);
248 if (unlikely(netif_tx_queue_stopped(dev_queue
)) &&
249 netfront_tx_slot_available(queue
) &&
250 likely(netif_running(dev
)))
251 netif_tx_wake_queue(netdev_get_tx_queue(dev
, queue
->id
));
255 static struct sk_buff
*xennet_alloc_one_rx_buffer(struct netfront_queue
*queue
)
260 skb
= __netdev_alloc_skb(queue
->info
->netdev
,
261 RX_COPY_THRESHOLD
+ NET_IP_ALIGN
,
262 GFP_ATOMIC
| __GFP_NOWARN
);
266 page
= alloc_page(GFP_ATOMIC
| __GFP_NOWARN
);
271 skb_add_rx_frag(skb
, 0, page
, 0, 0, PAGE_SIZE
);
273 /* Align ip header to a 16 bytes boundary */
274 skb_reserve(skb
, NET_IP_ALIGN
);
275 skb
->dev
= queue
->info
->netdev
;
281 static void xennet_alloc_rx_buffers(struct netfront_queue
*queue
)
283 RING_IDX req_prod
= queue
->rx
.req_prod_pvt
;
286 if (unlikely(!netif_carrier_ok(queue
->info
->netdev
)))
289 for (req_prod
= queue
->rx
.req_prod_pvt
;
290 req_prod
- queue
->rx
.rsp_cons
< NET_RX_RING_SIZE
;
296 struct xen_netif_rx_request
*req
;
298 skb
= xennet_alloc_one_rx_buffer(queue
);
302 id
= xennet_rxidx(req_prod
);
304 BUG_ON(queue
->rx_skbs
[id
]);
305 queue
->rx_skbs
[id
] = skb
;
307 ref
= gnttab_claim_grant_reference(&queue
->gref_rx_head
);
308 BUG_ON((signed short)ref
< 0);
309 queue
->grant_rx_ref
[id
] = ref
;
311 pfn
= page_to_pfn(skb_frag_page(&skb_shinfo(skb
)->frags
[0]));
313 req
= RING_GET_REQUEST(&queue
->rx
, req_prod
);
314 gnttab_grant_foreign_access_ref(ref
,
315 queue
->info
->xbdev
->otherend_id
,
323 queue
->rx
.req_prod_pvt
= req_prod
;
325 /* Not enough requests? Try again later. */
326 if (req_prod
- queue
->rx
.rsp_cons
< NET_RX_SLOTS_MIN
) {
327 mod_timer(&queue
->rx_refill_timer
, jiffies
+ (HZ
/10));
331 wmb(); /* barrier so backend seens requests */
333 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue
->rx
, notify
);
335 notify_remote_via_irq(queue
->rx_irq
);
338 static int xennet_open(struct net_device
*dev
)
340 struct netfront_info
*np
= netdev_priv(dev
);
341 unsigned int num_queues
= dev
->real_num_tx_queues
;
343 struct netfront_queue
*queue
= NULL
;
345 for (i
= 0; i
< num_queues
; ++i
) {
346 queue
= &np
->queues
[i
];
347 napi_enable(&queue
->napi
);
349 spin_lock_bh(&queue
->rx_lock
);
350 if (netif_carrier_ok(dev
)) {
351 xennet_alloc_rx_buffers(queue
);
352 queue
->rx
.sring
->rsp_event
= queue
->rx
.rsp_cons
+ 1;
353 if (RING_HAS_UNCONSUMED_RESPONSES(&queue
->rx
))
354 napi_schedule(&queue
->napi
);
356 spin_unlock_bh(&queue
->rx_lock
);
359 netif_tx_start_all_queues(dev
);
364 static void xennet_tx_buf_gc(struct netfront_queue
*queue
)
370 BUG_ON(!netif_carrier_ok(queue
->info
->netdev
));
373 prod
= queue
->tx
.sring
->rsp_prod
;
374 rmb(); /* Ensure we see responses up to 'rp'. */
376 for (cons
= queue
->tx
.rsp_cons
; cons
!= prod
; cons
++) {
377 struct xen_netif_tx_response
*txrsp
;
379 txrsp
= RING_GET_RESPONSE(&queue
->tx
, cons
);
380 if (txrsp
->status
== XEN_NETIF_RSP_NULL
)
384 skb
= queue
->tx_skbs
[id
].skb
;
385 if (unlikely(gnttab_query_foreign_access(
386 queue
->grant_tx_ref
[id
]) != 0)) {
387 pr_alert("%s: warning -- grant still in use by backend domain\n",
391 gnttab_end_foreign_access_ref(
392 queue
->grant_tx_ref
[id
], GNTMAP_readonly
);
393 gnttab_release_grant_reference(
394 &queue
->gref_tx_head
, queue
->grant_tx_ref
[id
]);
395 queue
->grant_tx_ref
[id
] = GRANT_INVALID_REF
;
396 queue
->grant_tx_page
[id
] = NULL
;
397 add_id_to_freelist(&queue
->tx_skb_freelist
, queue
->tx_skbs
, id
);
398 dev_kfree_skb_irq(skb
);
401 queue
->tx
.rsp_cons
= prod
;
404 * Set a new event, then check for race with update of tx_cons.
405 * Note that it is essential to schedule a callback, no matter
406 * how few buffers are pending. Even if there is space in the
407 * transmit ring, higher layers may be blocked because too much
408 * data is outstanding: in such cases notification from Xen is
409 * likely to be the only kick that we'll get.
411 queue
->tx
.sring
->rsp_event
=
412 prod
+ ((queue
->tx
.sring
->req_prod
- prod
) >> 1) + 1;
413 mb(); /* update shared area */
414 } while ((cons
== prod
) && (prod
!= queue
->tx
.sring
->rsp_prod
));
416 xennet_maybe_wake_tx(queue
);
419 static struct xen_netif_tx_request
*xennet_make_one_txreq(
420 struct netfront_queue
*queue
, struct sk_buff
*skb
,
421 struct page
*page
, unsigned int offset
, unsigned int len
)
424 struct xen_netif_tx_request
*tx
;
427 len
= min_t(unsigned int, PAGE_SIZE
- offset
, len
);
429 id
= get_id_from_freelist(&queue
->tx_skb_freelist
, queue
->tx_skbs
);
430 tx
= RING_GET_REQUEST(&queue
->tx
, queue
->tx
.req_prod_pvt
++);
431 ref
= gnttab_claim_grant_reference(&queue
->gref_tx_head
);
432 BUG_ON((signed short)ref
< 0);
434 gnttab_grant_foreign_access_ref(ref
, queue
->info
->xbdev
->otherend_id
,
435 page_to_mfn(page
), GNTMAP_readonly
);
437 queue
->tx_skbs
[id
].skb
= skb
;
438 queue
->grant_tx_page
[id
] = page
;
439 queue
->grant_tx_ref
[id
] = ref
;
450 static struct xen_netif_tx_request
*xennet_make_txreqs(
451 struct netfront_queue
*queue
, struct xen_netif_tx_request
*tx
,
452 struct sk_buff
*skb
, struct page
*page
,
453 unsigned int offset
, unsigned int len
)
455 /* Skip unused frames from start of page */
456 page
+= offset
>> PAGE_SHIFT
;
457 offset
&= ~PAGE_MASK
;
460 tx
->flags
|= XEN_NETTXF_more_data
;
461 tx
= xennet_make_one_txreq(queue
, skb_get(skb
),
472 * Count how many ring slots are required to send this skb. Each frag
473 * might be a compound page.
475 static int xennet_count_skb_slots(struct sk_buff
*skb
)
477 int i
, frags
= skb_shinfo(skb
)->nr_frags
;
480 pages
= PFN_UP(offset_in_page(skb
->data
) + skb_headlen(skb
));
482 for (i
= 0; i
< frags
; i
++) {
483 skb_frag_t
*frag
= skb_shinfo(skb
)->frags
+ i
;
484 unsigned long size
= skb_frag_size(frag
);
485 unsigned long offset
= frag
->page_offset
;
487 /* Skip unused frames from start of page */
488 offset
&= ~PAGE_MASK
;
490 pages
+= PFN_UP(offset
+ size
);
496 static u16
xennet_select_queue(struct net_device
*dev
, struct sk_buff
*skb
,
497 void *accel_priv
, select_queue_fallback_t fallback
)
499 unsigned int num_queues
= dev
->real_num_tx_queues
;
503 /* First, check if there is only one queue */
504 if (num_queues
== 1) {
507 hash
= skb_get_hash(skb
);
508 queue_idx
= hash
% num_queues
;
514 static int xennet_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
516 struct netfront_info
*np
= netdev_priv(dev
);
517 struct netfront_stats
*tx_stats
= this_cpu_ptr(np
->tx_stats
);
518 struct xen_netif_tx_request
*tx
, *first_tx
;
526 struct netfront_queue
*queue
= NULL
;
527 unsigned int num_queues
= dev
->real_num_tx_queues
;
530 /* Drop the packet if no queues are set up */
533 /* Determine which queue to transmit this SKB on */
534 queue_index
= skb_get_queue_mapping(skb
);
535 queue
= &np
->queues
[queue_index
];
537 /* If skb->len is too big for wire format, drop skb and alert
538 * user about misconfiguration.
540 if (unlikely(skb
->len
> XEN_NETIF_MAX_TX_SIZE
)) {
541 net_alert_ratelimited(
542 "xennet: skb->len = %u, too big for wire format\n",
547 slots
= xennet_count_skb_slots(skb
);
548 if (unlikely(slots
> MAX_SKB_FRAGS
+ 1)) {
549 net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n",
551 if (skb_linearize(skb
))
555 page
= virt_to_page(skb
->data
);
556 offset
= offset_in_page(skb
->data
);
557 len
= skb_headlen(skb
);
559 spin_lock_irqsave(&queue
->tx_lock
, flags
);
561 if (unlikely(!netif_carrier_ok(dev
) ||
562 (slots
> 1 && !xennet_can_sg(dev
)) ||
563 netif_needs_gso(skb
, netif_skb_features(skb
)))) {
564 spin_unlock_irqrestore(&queue
->tx_lock
, flags
);
568 /* First request for the linear area. */
569 first_tx
= tx
= xennet_make_one_txreq(queue
, skb
,
575 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
577 tx
->flags
|= XEN_NETTXF_csum_blank
| XEN_NETTXF_data_validated
;
578 else if (skb
->ip_summed
== CHECKSUM_UNNECESSARY
)
579 /* remote but checksummed. */
580 tx
->flags
|= XEN_NETTXF_data_validated
;
582 /* Optional extra info after the first request. */
583 if (skb_shinfo(skb
)->gso_size
) {
584 struct xen_netif_extra_info
*gso
;
586 gso
= (struct xen_netif_extra_info
*)
587 RING_GET_REQUEST(&queue
->tx
, queue
->tx
.req_prod_pvt
++);
589 tx
->flags
|= XEN_NETTXF_extra_info
;
591 gso
->u
.gso
.size
= skb_shinfo(skb
)->gso_size
;
592 gso
->u
.gso
.type
= (skb_shinfo(skb
)->gso_type
& SKB_GSO_TCPV6
) ?
593 XEN_NETIF_GSO_TYPE_TCPV6
:
594 XEN_NETIF_GSO_TYPE_TCPV4
;
596 gso
->u
.gso
.features
= 0;
598 gso
->type
= XEN_NETIF_EXTRA_TYPE_GSO
;
602 /* Requests for the rest of the linear area. */
603 tx
= xennet_make_txreqs(queue
, tx
, skb
, page
, offset
, len
);
605 /* Requests for all the frags. */
606 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
607 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
608 tx
= xennet_make_txreqs(queue
, tx
, skb
,
609 skb_frag_page(frag
), frag
->page_offset
,
610 skb_frag_size(frag
));
613 /* First request has the packet length. */
614 first_tx
->size
= skb
->len
;
616 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue
->tx
, notify
);
618 notify_remote_via_irq(queue
->tx_irq
);
620 u64_stats_update_begin(&tx_stats
->syncp
);
621 tx_stats
->bytes
+= skb
->len
;
623 u64_stats_update_end(&tx_stats
->syncp
);
625 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
626 xennet_tx_buf_gc(queue
);
628 if (!netfront_tx_slot_available(queue
))
629 netif_tx_stop_queue(netdev_get_tx_queue(dev
, queue
->id
));
631 spin_unlock_irqrestore(&queue
->tx_lock
, flags
);
636 dev
->stats
.tx_dropped
++;
637 dev_kfree_skb_any(skb
);
641 static int xennet_close(struct net_device
*dev
)
643 struct netfront_info
*np
= netdev_priv(dev
);
644 unsigned int num_queues
= dev
->real_num_tx_queues
;
646 struct netfront_queue
*queue
;
647 netif_tx_stop_all_queues(np
->netdev
);
648 for (i
= 0; i
< num_queues
; ++i
) {
649 queue
= &np
->queues
[i
];
650 napi_disable(&queue
->napi
);
655 static void xennet_move_rx_slot(struct netfront_queue
*queue
, struct sk_buff
*skb
,
658 int new = xennet_rxidx(queue
->rx
.req_prod_pvt
);
660 BUG_ON(queue
->rx_skbs
[new]);
661 queue
->rx_skbs
[new] = skb
;
662 queue
->grant_rx_ref
[new] = ref
;
663 RING_GET_REQUEST(&queue
->rx
, queue
->rx
.req_prod_pvt
)->id
= new;
664 RING_GET_REQUEST(&queue
->rx
, queue
->rx
.req_prod_pvt
)->gref
= ref
;
665 queue
->rx
.req_prod_pvt
++;
668 static int xennet_get_extras(struct netfront_queue
*queue
,
669 struct xen_netif_extra_info
*extras
,
673 struct xen_netif_extra_info
*extra
;
674 struct device
*dev
= &queue
->info
->netdev
->dev
;
675 RING_IDX cons
= queue
->rx
.rsp_cons
;
682 if (unlikely(cons
+ 1 == rp
)) {
684 dev_warn(dev
, "Missing extra info\n");
689 extra
= (struct xen_netif_extra_info
*)
690 RING_GET_RESPONSE(&queue
->rx
, ++cons
);
692 if (unlikely(!extra
->type
||
693 extra
->type
>= XEN_NETIF_EXTRA_TYPE_MAX
)) {
695 dev_warn(dev
, "Invalid extra type: %d\n",
699 memcpy(&extras
[extra
->type
- 1], extra
,
703 skb
= xennet_get_rx_skb(queue
, cons
);
704 ref
= xennet_get_rx_ref(queue
, cons
);
705 xennet_move_rx_slot(queue
, skb
, ref
);
706 } while (extra
->flags
& XEN_NETIF_EXTRA_FLAG_MORE
);
708 queue
->rx
.rsp_cons
= cons
;
712 static int xennet_get_responses(struct netfront_queue
*queue
,
713 struct netfront_rx_info
*rinfo
, RING_IDX rp
,
714 struct sk_buff_head
*list
)
716 struct xen_netif_rx_response
*rx
= &rinfo
->rx
;
717 struct xen_netif_extra_info
*extras
= rinfo
->extras
;
718 struct device
*dev
= &queue
->info
->netdev
->dev
;
719 RING_IDX cons
= queue
->rx
.rsp_cons
;
720 struct sk_buff
*skb
= xennet_get_rx_skb(queue
, cons
);
721 grant_ref_t ref
= xennet_get_rx_ref(queue
, cons
);
722 int max
= MAX_SKB_FRAGS
+ (rx
->status
<= RX_COPY_THRESHOLD
);
727 if (rx
->flags
& XEN_NETRXF_extra_info
) {
728 err
= xennet_get_extras(queue
, extras
, rp
);
729 cons
= queue
->rx
.rsp_cons
;
733 if (unlikely(rx
->status
< 0 ||
734 rx
->offset
+ rx
->status
> PAGE_SIZE
)) {
736 dev_warn(dev
, "rx->offset: %x, size: %u\n",
737 rx
->offset
, rx
->status
);
738 xennet_move_rx_slot(queue
, skb
, ref
);
744 * This definitely indicates a bug, either in this driver or in
745 * the backend driver. In future this should flag the bad
746 * situation to the system controller to reboot the backend.
748 if (ref
== GRANT_INVALID_REF
) {
750 dev_warn(dev
, "Bad rx response id %d.\n",
756 ret
= gnttab_end_foreign_access_ref(ref
, 0);
759 gnttab_release_grant_reference(&queue
->gref_rx_head
, ref
);
761 __skb_queue_tail(list
, skb
);
764 if (!(rx
->flags
& XEN_NETRXF_more_data
))
767 if (cons
+ slots
== rp
) {
769 dev_warn(dev
, "Need more slots\n");
774 rx
= RING_GET_RESPONSE(&queue
->rx
, cons
+ slots
);
775 skb
= xennet_get_rx_skb(queue
, cons
+ slots
);
776 ref
= xennet_get_rx_ref(queue
, cons
+ slots
);
780 if (unlikely(slots
> max
)) {
782 dev_warn(dev
, "Too many slots\n");
787 queue
->rx
.rsp_cons
= cons
+ slots
;
792 static int xennet_set_skb_gso(struct sk_buff
*skb
,
793 struct xen_netif_extra_info
*gso
)
795 if (!gso
->u
.gso
.size
) {
797 pr_warn("GSO size must not be zero\n");
801 if (gso
->u
.gso
.type
!= XEN_NETIF_GSO_TYPE_TCPV4
&&
802 gso
->u
.gso
.type
!= XEN_NETIF_GSO_TYPE_TCPV6
) {
804 pr_warn("Bad GSO type %d\n", gso
->u
.gso
.type
);
808 skb_shinfo(skb
)->gso_size
= gso
->u
.gso
.size
;
809 skb_shinfo(skb
)->gso_type
=
810 (gso
->u
.gso
.type
== XEN_NETIF_GSO_TYPE_TCPV4
) ?
814 /* Header must be checked, and gso_segs computed. */
815 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
816 skb_shinfo(skb
)->gso_segs
= 0;
821 static RING_IDX
xennet_fill_frags(struct netfront_queue
*queue
,
823 struct sk_buff_head
*list
)
825 struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
826 RING_IDX cons
= queue
->rx
.rsp_cons
;
827 struct sk_buff
*nskb
;
829 while ((nskb
= __skb_dequeue(list
))) {
830 struct xen_netif_rx_response
*rx
=
831 RING_GET_RESPONSE(&queue
->rx
, ++cons
);
832 skb_frag_t
*nfrag
= &skb_shinfo(nskb
)->frags
[0];
834 if (shinfo
->nr_frags
== MAX_SKB_FRAGS
) {
835 unsigned int pull_to
= NETFRONT_SKB_CB(skb
)->pull_to
;
837 BUG_ON(pull_to
<= skb_headlen(skb
));
838 __pskb_pull_tail(skb
, pull_to
- skb_headlen(skb
));
840 BUG_ON(shinfo
->nr_frags
>= MAX_SKB_FRAGS
);
842 skb_add_rx_frag(skb
, shinfo
->nr_frags
, skb_frag_page(nfrag
),
843 rx
->offset
, rx
->status
, PAGE_SIZE
);
845 skb_shinfo(nskb
)->nr_frags
= 0;
852 static int checksum_setup(struct net_device
*dev
, struct sk_buff
*skb
)
854 bool recalculate_partial_csum
= false;
857 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
858 * peers can fail to set NETRXF_csum_blank when sending a GSO
859 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
860 * recalculate the partial checksum.
862 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
&& skb_is_gso(skb
)) {
863 struct netfront_info
*np
= netdev_priv(dev
);
864 atomic_inc(&np
->rx_gso_checksum_fixup
);
865 skb
->ip_summed
= CHECKSUM_PARTIAL
;
866 recalculate_partial_csum
= true;
869 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
870 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
)
873 return skb_checksum_setup(skb
, recalculate_partial_csum
);
876 static int handle_incoming_queue(struct netfront_queue
*queue
,
877 struct sk_buff_head
*rxq
)
879 struct netfront_stats
*rx_stats
= this_cpu_ptr(queue
->info
->rx_stats
);
880 int packets_dropped
= 0;
883 while ((skb
= __skb_dequeue(rxq
)) != NULL
) {
884 int pull_to
= NETFRONT_SKB_CB(skb
)->pull_to
;
886 if (pull_to
> skb_headlen(skb
))
887 __pskb_pull_tail(skb
, pull_to
- skb_headlen(skb
));
889 /* Ethernet work: Delayed to here as it peeks the header. */
890 skb
->protocol
= eth_type_trans(skb
, queue
->info
->netdev
);
891 skb_reset_network_header(skb
);
893 if (checksum_setup(queue
->info
->netdev
, skb
)) {
896 queue
->info
->netdev
->stats
.rx_errors
++;
900 u64_stats_update_begin(&rx_stats
->syncp
);
902 rx_stats
->bytes
+= skb
->len
;
903 u64_stats_update_end(&rx_stats
->syncp
);
906 napi_gro_receive(&queue
->napi
, skb
);
909 return packets_dropped
;
912 static int xennet_poll(struct napi_struct
*napi
, int budget
)
914 struct netfront_queue
*queue
= container_of(napi
, struct netfront_queue
, napi
);
915 struct net_device
*dev
= queue
->info
->netdev
;
917 struct netfront_rx_info rinfo
;
918 struct xen_netif_rx_response
*rx
= &rinfo
.rx
;
919 struct xen_netif_extra_info
*extras
= rinfo
.extras
;
922 struct sk_buff_head rxq
;
923 struct sk_buff_head errq
;
924 struct sk_buff_head tmpq
;
927 spin_lock(&queue
->rx_lock
);
929 skb_queue_head_init(&rxq
);
930 skb_queue_head_init(&errq
);
931 skb_queue_head_init(&tmpq
);
933 rp
= queue
->rx
.sring
->rsp_prod
;
934 rmb(); /* Ensure we see queued responses up to 'rp'. */
936 i
= queue
->rx
.rsp_cons
;
938 while ((i
!= rp
) && (work_done
< budget
)) {
939 memcpy(rx
, RING_GET_RESPONSE(&queue
->rx
, i
), sizeof(*rx
));
940 memset(extras
, 0, sizeof(rinfo
.extras
));
942 err
= xennet_get_responses(queue
, &rinfo
, rp
, &tmpq
);
946 while ((skb
= __skb_dequeue(&tmpq
)))
947 __skb_queue_tail(&errq
, skb
);
948 dev
->stats
.rx_errors
++;
949 i
= queue
->rx
.rsp_cons
;
953 skb
= __skb_dequeue(&tmpq
);
955 if (extras
[XEN_NETIF_EXTRA_TYPE_GSO
- 1].type
) {
956 struct xen_netif_extra_info
*gso
;
957 gso
= &extras
[XEN_NETIF_EXTRA_TYPE_GSO
- 1];
959 if (unlikely(xennet_set_skb_gso(skb
, gso
))) {
960 __skb_queue_head(&tmpq
, skb
);
961 queue
->rx
.rsp_cons
+= skb_queue_len(&tmpq
);
966 NETFRONT_SKB_CB(skb
)->pull_to
= rx
->status
;
967 if (NETFRONT_SKB_CB(skb
)->pull_to
> RX_COPY_THRESHOLD
)
968 NETFRONT_SKB_CB(skb
)->pull_to
= RX_COPY_THRESHOLD
;
970 skb_shinfo(skb
)->frags
[0].page_offset
= rx
->offset
;
971 skb_frag_size_set(&skb_shinfo(skb
)->frags
[0], rx
->status
);
972 skb
->data_len
= rx
->status
;
973 skb
->len
+= rx
->status
;
975 i
= xennet_fill_frags(queue
, skb
, &tmpq
);
977 if (rx
->flags
& XEN_NETRXF_csum_blank
)
978 skb
->ip_summed
= CHECKSUM_PARTIAL
;
979 else if (rx
->flags
& XEN_NETRXF_data_validated
)
980 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
982 __skb_queue_tail(&rxq
, skb
);
984 queue
->rx
.rsp_cons
= ++i
;
988 __skb_queue_purge(&errq
);
990 work_done
-= handle_incoming_queue(queue
, &rxq
);
992 xennet_alloc_rx_buffers(queue
);
994 if (work_done
< budget
) {
999 RING_FINAL_CHECK_FOR_RESPONSES(&queue
->rx
, more_to_do
);
1001 napi_schedule(napi
);
1004 spin_unlock(&queue
->rx_lock
);
1009 static int xennet_change_mtu(struct net_device
*dev
, int mtu
)
1011 int max
= xennet_can_sg(dev
) ? XEN_NETIF_MAX_TX_SIZE
: ETH_DATA_LEN
;
1019 static struct rtnl_link_stats64
*xennet_get_stats64(struct net_device
*dev
,
1020 struct rtnl_link_stats64
*tot
)
1022 struct netfront_info
*np
= netdev_priv(dev
);
1025 for_each_possible_cpu(cpu
) {
1026 struct netfront_stats
*rx_stats
= per_cpu_ptr(np
->rx_stats
, cpu
);
1027 struct netfront_stats
*tx_stats
= per_cpu_ptr(np
->tx_stats
, cpu
);
1028 u64 rx_packets
, rx_bytes
, tx_packets
, tx_bytes
;
1032 start
= u64_stats_fetch_begin_irq(&tx_stats
->syncp
);
1033 tx_packets
= tx_stats
->packets
;
1034 tx_bytes
= tx_stats
->bytes
;
1035 } while (u64_stats_fetch_retry_irq(&tx_stats
->syncp
, start
));
1038 start
= u64_stats_fetch_begin_irq(&rx_stats
->syncp
);
1039 rx_packets
= rx_stats
->packets
;
1040 rx_bytes
= rx_stats
->bytes
;
1041 } while (u64_stats_fetch_retry_irq(&rx_stats
->syncp
, start
));
1043 tot
->rx_packets
+= rx_packets
;
1044 tot
->tx_packets
+= tx_packets
;
1045 tot
->rx_bytes
+= rx_bytes
;
1046 tot
->tx_bytes
+= tx_bytes
;
1049 tot
->rx_errors
= dev
->stats
.rx_errors
;
1050 tot
->tx_dropped
= dev
->stats
.tx_dropped
;
1055 static void xennet_release_tx_bufs(struct netfront_queue
*queue
)
1057 struct sk_buff
*skb
;
1060 for (i
= 0; i
< NET_TX_RING_SIZE
; i
++) {
1061 /* Skip over entries which are actually freelist references */
1062 if (skb_entry_is_link(&queue
->tx_skbs
[i
]))
1065 skb
= queue
->tx_skbs
[i
].skb
;
1066 get_page(queue
->grant_tx_page
[i
]);
1067 gnttab_end_foreign_access(queue
->grant_tx_ref
[i
],
1069 (unsigned long)page_address(queue
->grant_tx_page
[i
]));
1070 queue
->grant_tx_page
[i
] = NULL
;
1071 queue
->grant_tx_ref
[i
] = GRANT_INVALID_REF
;
1072 add_id_to_freelist(&queue
->tx_skb_freelist
, queue
->tx_skbs
, i
);
1073 dev_kfree_skb_irq(skb
);
1077 static void xennet_release_rx_bufs(struct netfront_queue
*queue
)
1081 spin_lock_bh(&queue
->rx_lock
);
1083 for (id
= 0; id
< NET_RX_RING_SIZE
; id
++) {
1084 struct sk_buff
*skb
;
1087 skb
= queue
->rx_skbs
[id
];
1091 ref
= queue
->grant_rx_ref
[id
];
1092 if (ref
== GRANT_INVALID_REF
)
1095 page
= skb_frag_page(&skb_shinfo(skb
)->frags
[0]);
1097 /* gnttab_end_foreign_access() needs a page ref until
1098 * foreign access is ended (which may be deferred).
1101 gnttab_end_foreign_access(ref
, 0,
1102 (unsigned long)page_address(page
));
1103 queue
->grant_rx_ref
[id
] = GRANT_INVALID_REF
;
1108 spin_unlock_bh(&queue
->rx_lock
);
1111 static netdev_features_t
xennet_fix_features(struct net_device
*dev
,
1112 netdev_features_t features
)
1114 struct netfront_info
*np
= netdev_priv(dev
);
1117 if (features
& NETIF_F_SG
) {
1118 if (xenbus_scanf(XBT_NIL
, np
->xbdev
->otherend
, "feature-sg",
1123 features
&= ~NETIF_F_SG
;
1126 if (features
& NETIF_F_IPV6_CSUM
) {
1127 if (xenbus_scanf(XBT_NIL
, np
->xbdev
->otherend
,
1128 "feature-ipv6-csum-offload", "%d", &val
) < 0)
1132 features
&= ~NETIF_F_IPV6_CSUM
;
1135 if (features
& NETIF_F_TSO
) {
1136 if (xenbus_scanf(XBT_NIL
, np
->xbdev
->otherend
,
1137 "feature-gso-tcpv4", "%d", &val
) < 0)
1141 features
&= ~NETIF_F_TSO
;
1144 if (features
& NETIF_F_TSO6
) {
1145 if (xenbus_scanf(XBT_NIL
, np
->xbdev
->otherend
,
1146 "feature-gso-tcpv6", "%d", &val
) < 0)
1150 features
&= ~NETIF_F_TSO6
;
1156 static int xennet_set_features(struct net_device
*dev
,
1157 netdev_features_t features
)
1159 if (!(features
& NETIF_F_SG
) && dev
->mtu
> ETH_DATA_LEN
) {
1160 netdev_info(dev
, "Reducing MTU because no SG offload");
1161 dev
->mtu
= ETH_DATA_LEN
;
1167 static irqreturn_t
xennet_tx_interrupt(int irq
, void *dev_id
)
1169 struct netfront_queue
*queue
= dev_id
;
1170 unsigned long flags
;
1172 spin_lock_irqsave(&queue
->tx_lock
, flags
);
1173 xennet_tx_buf_gc(queue
);
1174 spin_unlock_irqrestore(&queue
->tx_lock
, flags
);
1179 static irqreturn_t
xennet_rx_interrupt(int irq
, void *dev_id
)
1181 struct netfront_queue
*queue
= dev_id
;
1182 struct net_device
*dev
= queue
->info
->netdev
;
1184 if (likely(netif_carrier_ok(dev
) &&
1185 RING_HAS_UNCONSUMED_RESPONSES(&queue
->rx
)))
1186 napi_schedule(&queue
->napi
);
1191 static irqreturn_t
xennet_interrupt(int irq
, void *dev_id
)
1193 xennet_tx_interrupt(irq
, dev_id
);
1194 xennet_rx_interrupt(irq
, dev_id
);
1198 #ifdef CONFIG_NET_POLL_CONTROLLER
1199 static void xennet_poll_controller(struct net_device
*dev
)
1201 /* Poll each queue */
1202 struct netfront_info
*info
= netdev_priv(dev
);
1203 unsigned int num_queues
= dev
->real_num_tx_queues
;
1205 for (i
= 0; i
< num_queues
; ++i
)
1206 xennet_interrupt(0, &info
->queues
[i
]);
1210 static const struct net_device_ops xennet_netdev_ops
= {
1211 .ndo_open
= xennet_open
,
1212 .ndo_stop
= xennet_close
,
1213 .ndo_start_xmit
= xennet_start_xmit
,
1214 .ndo_change_mtu
= xennet_change_mtu
,
1215 .ndo_get_stats64
= xennet_get_stats64
,
1216 .ndo_set_mac_address
= eth_mac_addr
,
1217 .ndo_validate_addr
= eth_validate_addr
,
1218 .ndo_fix_features
= xennet_fix_features
,
1219 .ndo_set_features
= xennet_set_features
,
1220 .ndo_select_queue
= xennet_select_queue
,
1221 #ifdef CONFIG_NET_POLL_CONTROLLER
1222 .ndo_poll_controller
= xennet_poll_controller
,
1226 static void xennet_free_netdev(struct net_device
*netdev
)
1228 struct netfront_info
*np
= netdev_priv(netdev
);
1230 free_percpu(np
->rx_stats
);
1231 free_percpu(np
->tx_stats
);
1232 free_netdev(netdev
);
1235 static struct net_device
*xennet_create_dev(struct xenbus_device
*dev
)
1238 struct net_device
*netdev
;
1239 struct netfront_info
*np
;
1241 netdev
= alloc_etherdev_mq(sizeof(struct netfront_info
), xennet_max_queues
);
1243 return ERR_PTR(-ENOMEM
);
1245 np
= netdev_priv(netdev
);
1248 /* No need to use rtnl_lock() before the call below as it
1249 * happens before register_netdev().
1251 netif_set_real_num_tx_queues(netdev
, 0);
1255 np
->rx_stats
= netdev_alloc_pcpu_stats(struct netfront_stats
);
1256 if (np
->rx_stats
== NULL
)
1258 np
->tx_stats
= netdev_alloc_pcpu_stats(struct netfront_stats
);
1259 if (np
->tx_stats
== NULL
)
1262 netdev
->netdev_ops
= &xennet_netdev_ops
;
1264 netdev
->features
= NETIF_F_IP_CSUM
| NETIF_F_RXCSUM
|
1266 netdev
->hw_features
= NETIF_F_SG
|
1268 NETIF_F_TSO
| NETIF_F_TSO6
;
1271 * Assume that all hw features are available for now. This set
1272 * will be adjusted by the call to netdev_update_features() in
1273 * xennet_connect() which is the earliest point where we can
1274 * negotiate with the backend regarding supported features.
1276 netdev
->features
|= netdev
->hw_features
;
1278 netdev
->ethtool_ops
= &xennet_ethtool_ops
;
1279 SET_NETDEV_DEV(netdev
, &dev
->dev
);
1281 np
->netdev
= netdev
;
1283 netif_carrier_off(netdev
);
1288 xennet_free_netdev(netdev
);
1289 return ERR_PTR(err
);
1293 * Entry point to this code when a new device is created. Allocate the basic
1294 * structures and the ring buffers for communication with the backend, and
1295 * inform the backend of the appropriate details for those.
1297 static int netfront_probe(struct xenbus_device
*dev
,
1298 const struct xenbus_device_id
*id
)
1301 struct net_device
*netdev
;
1302 struct netfront_info
*info
;
1304 netdev
= xennet_create_dev(dev
);
1305 if (IS_ERR(netdev
)) {
1306 err
= PTR_ERR(netdev
);
1307 xenbus_dev_fatal(dev
, err
, "creating netdev");
1311 info
= netdev_priv(netdev
);
1312 dev_set_drvdata(&dev
->dev
, info
);
1314 info
->netdev
->sysfs_groups
[0] = &xennet_dev_group
;
1316 err
= register_netdev(info
->netdev
);
1318 pr_warn("%s: register_netdev err=%d\n", __func__
, err
);
1325 xennet_free_netdev(netdev
);
1326 dev_set_drvdata(&dev
->dev
, NULL
);
1330 static void xennet_end_access(int ref
, void *page
)
1332 /* This frees the page as a side-effect */
1333 if (ref
!= GRANT_INVALID_REF
)
1334 gnttab_end_foreign_access(ref
, 0, (unsigned long)page
);
1337 static void xennet_disconnect_backend(struct netfront_info
*info
)
1340 unsigned int num_queues
= info
->netdev
->real_num_tx_queues
;
1342 netif_carrier_off(info
->netdev
);
1344 for (i
= 0; i
< num_queues
; ++i
) {
1345 struct netfront_queue
*queue
= &info
->queues
[i
];
1347 if (queue
->tx_irq
&& (queue
->tx_irq
== queue
->rx_irq
))
1348 unbind_from_irqhandler(queue
->tx_irq
, queue
);
1349 if (queue
->tx_irq
&& (queue
->tx_irq
!= queue
->rx_irq
)) {
1350 unbind_from_irqhandler(queue
->tx_irq
, queue
);
1351 unbind_from_irqhandler(queue
->rx_irq
, queue
);
1353 queue
->tx_evtchn
= queue
->rx_evtchn
= 0;
1354 queue
->tx_irq
= queue
->rx_irq
= 0;
1356 if (netif_running(info
->netdev
))
1357 napi_synchronize(&queue
->napi
);
1359 xennet_release_tx_bufs(queue
);
1360 xennet_release_rx_bufs(queue
);
1361 gnttab_free_grant_references(queue
->gref_tx_head
);
1362 gnttab_free_grant_references(queue
->gref_rx_head
);
1364 /* End access and free the pages */
1365 xennet_end_access(queue
->tx_ring_ref
, queue
->tx
.sring
);
1366 xennet_end_access(queue
->rx_ring_ref
, queue
->rx
.sring
);
1368 queue
->tx_ring_ref
= GRANT_INVALID_REF
;
1369 queue
->rx_ring_ref
= GRANT_INVALID_REF
;
1370 queue
->tx
.sring
= NULL
;
1371 queue
->rx
.sring
= NULL
;
1376 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1377 * driver restart. We tear down our netif structure and recreate it, but
1378 * leave the device-layer structures intact so that this is transparent to the
1379 * rest of the kernel.
1381 static int netfront_resume(struct xenbus_device
*dev
)
1383 struct netfront_info
*info
= dev_get_drvdata(&dev
->dev
);
1385 dev_dbg(&dev
->dev
, "%s\n", dev
->nodename
);
1387 xennet_disconnect_backend(info
);
1391 static int xen_net_read_mac(struct xenbus_device
*dev
, u8 mac
[])
1393 char *s
, *e
, *macstr
;
1396 macstr
= s
= xenbus_read(XBT_NIL
, dev
->nodename
, "mac", NULL
);
1398 return PTR_ERR(macstr
);
1400 for (i
= 0; i
< ETH_ALEN
; i
++) {
1401 mac
[i
] = simple_strtoul(s
, &e
, 16);
1402 if ((s
== e
) || (*e
!= ((i
== ETH_ALEN
-1) ? '\0' : ':'))) {
1413 static int setup_netfront_single(struct netfront_queue
*queue
)
1417 err
= xenbus_alloc_evtchn(queue
->info
->xbdev
, &queue
->tx_evtchn
);
1421 err
= bind_evtchn_to_irqhandler(queue
->tx_evtchn
,
1423 0, queue
->info
->netdev
->name
, queue
);
1426 queue
->rx_evtchn
= queue
->tx_evtchn
;
1427 queue
->rx_irq
= queue
->tx_irq
= err
;
1432 xenbus_free_evtchn(queue
->info
->xbdev
, queue
->tx_evtchn
);
1433 queue
->tx_evtchn
= 0;
1438 static int setup_netfront_split(struct netfront_queue
*queue
)
1442 err
= xenbus_alloc_evtchn(queue
->info
->xbdev
, &queue
->tx_evtchn
);
1445 err
= xenbus_alloc_evtchn(queue
->info
->xbdev
, &queue
->rx_evtchn
);
1447 goto alloc_rx_evtchn_fail
;
1449 snprintf(queue
->tx_irq_name
, sizeof(queue
->tx_irq_name
),
1450 "%s-tx", queue
->name
);
1451 err
= bind_evtchn_to_irqhandler(queue
->tx_evtchn
,
1452 xennet_tx_interrupt
,
1453 0, queue
->tx_irq_name
, queue
);
1456 queue
->tx_irq
= err
;
1458 snprintf(queue
->rx_irq_name
, sizeof(queue
->rx_irq_name
),
1459 "%s-rx", queue
->name
);
1460 err
= bind_evtchn_to_irqhandler(queue
->rx_evtchn
,
1461 xennet_rx_interrupt
,
1462 0, queue
->rx_irq_name
, queue
);
1465 queue
->rx_irq
= err
;
1470 unbind_from_irqhandler(queue
->tx_irq
, queue
);
1473 xenbus_free_evtchn(queue
->info
->xbdev
, queue
->rx_evtchn
);
1474 queue
->rx_evtchn
= 0;
1475 alloc_rx_evtchn_fail
:
1476 xenbus_free_evtchn(queue
->info
->xbdev
, queue
->tx_evtchn
);
1477 queue
->tx_evtchn
= 0;
1482 static int setup_netfront(struct xenbus_device
*dev
,
1483 struct netfront_queue
*queue
, unsigned int feature_split_evtchn
)
1485 struct xen_netif_tx_sring
*txs
;
1486 struct xen_netif_rx_sring
*rxs
;
1490 queue
->tx_ring_ref
= GRANT_INVALID_REF
;
1491 queue
->rx_ring_ref
= GRANT_INVALID_REF
;
1492 queue
->rx
.sring
= NULL
;
1493 queue
->tx
.sring
= NULL
;
1495 txs
= (struct xen_netif_tx_sring
*)get_zeroed_page(GFP_NOIO
| __GFP_HIGH
);
1498 xenbus_dev_fatal(dev
, err
, "allocating tx ring page");
1501 SHARED_RING_INIT(txs
);
1502 FRONT_RING_INIT(&queue
->tx
, txs
, PAGE_SIZE
);
1504 err
= xenbus_grant_ring(dev
, txs
, 1, &gref
);
1506 goto grant_tx_ring_fail
;
1507 queue
->tx_ring_ref
= gref
;
1509 rxs
= (struct xen_netif_rx_sring
*)get_zeroed_page(GFP_NOIO
| __GFP_HIGH
);
1512 xenbus_dev_fatal(dev
, err
, "allocating rx ring page");
1513 goto alloc_rx_ring_fail
;
1515 SHARED_RING_INIT(rxs
);
1516 FRONT_RING_INIT(&queue
->rx
, rxs
, PAGE_SIZE
);
1518 err
= xenbus_grant_ring(dev
, rxs
, 1, &gref
);
1520 goto grant_rx_ring_fail
;
1521 queue
->rx_ring_ref
= gref
;
1523 if (feature_split_evtchn
)
1524 err
= setup_netfront_split(queue
);
1525 /* setup single event channel if
1526 * a) feature-split-event-channels == 0
1527 * b) feature-split-event-channels == 1 but failed to setup
1529 if (!feature_split_evtchn
|| (feature_split_evtchn
&& err
))
1530 err
= setup_netfront_single(queue
);
1533 goto alloc_evtchn_fail
;
1537 /* If we fail to setup netfront, it is safe to just revoke access to
1538 * granted pages because backend is not accessing it at this point.
1541 gnttab_end_foreign_access_ref(queue
->rx_ring_ref
, 0);
1543 free_page((unsigned long)rxs
);
1545 gnttab_end_foreign_access_ref(queue
->tx_ring_ref
, 0);
1547 free_page((unsigned long)txs
);
1552 /* Queue-specific initialisation
1553 * This used to be done in xennet_create_dev() but must now
1556 static int xennet_init_queue(struct netfront_queue
*queue
)
1561 spin_lock_init(&queue
->tx_lock
);
1562 spin_lock_init(&queue
->rx_lock
);
1564 init_timer(&queue
->rx_refill_timer
);
1565 queue
->rx_refill_timer
.data
= (unsigned long)queue
;
1566 queue
->rx_refill_timer
.function
= rx_refill_timeout
;
1568 snprintf(queue
->name
, sizeof(queue
->name
), "%s-q%u",
1569 queue
->info
->netdev
->name
, queue
->id
);
1571 /* Initialise tx_skbs as a free chain containing every entry. */
1572 queue
->tx_skb_freelist
= 0;
1573 for (i
= 0; i
< NET_TX_RING_SIZE
; i
++) {
1574 skb_entry_set_link(&queue
->tx_skbs
[i
], i
+1);
1575 queue
->grant_tx_ref
[i
] = GRANT_INVALID_REF
;
1576 queue
->grant_tx_page
[i
] = NULL
;
1579 /* Clear out rx_skbs */
1580 for (i
= 0; i
< NET_RX_RING_SIZE
; i
++) {
1581 queue
->rx_skbs
[i
] = NULL
;
1582 queue
->grant_rx_ref
[i
] = GRANT_INVALID_REF
;
1585 /* A grant for every tx ring slot */
1586 if (gnttab_alloc_grant_references(NET_TX_RING_SIZE
,
1587 &queue
->gref_tx_head
) < 0) {
1588 pr_alert("can't alloc tx grant refs\n");
1593 /* A grant for every rx ring slot */
1594 if (gnttab_alloc_grant_references(NET_RX_RING_SIZE
,
1595 &queue
->gref_rx_head
) < 0) {
1596 pr_alert("can't alloc rx grant refs\n");
1604 gnttab_free_grant_references(queue
->gref_tx_head
);
1609 static int write_queue_xenstore_keys(struct netfront_queue
*queue
,
1610 struct xenbus_transaction
*xbt
, int write_hierarchical
)
1612 /* Write the queue-specific keys into XenStore in the traditional
1613 * way for a single queue, or in a queue subkeys for multiple
1616 struct xenbus_device
*dev
= queue
->info
->xbdev
;
1618 const char *message
;
1622 /* Choose the correct place to write the keys */
1623 if (write_hierarchical
) {
1624 pathsize
= strlen(dev
->nodename
) + 10;
1625 path
= kzalloc(pathsize
, GFP_KERNEL
);
1628 message
= "out of memory while writing ring references";
1631 snprintf(path
, pathsize
, "%s/queue-%u",
1632 dev
->nodename
, queue
->id
);
1634 path
= (char *)dev
->nodename
;
1637 /* Write ring references */
1638 err
= xenbus_printf(*xbt
, path
, "tx-ring-ref", "%u",
1639 queue
->tx_ring_ref
);
1641 message
= "writing tx-ring-ref";
1645 err
= xenbus_printf(*xbt
, path
, "rx-ring-ref", "%u",
1646 queue
->rx_ring_ref
);
1648 message
= "writing rx-ring-ref";
1652 /* Write event channels; taking into account both shared
1653 * and split event channel scenarios.
1655 if (queue
->tx_evtchn
== queue
->rx_evtchn
) {
1656 /* Shared event channel */
1657 err
= xenbus_printf(*xbt
, path
,
1658 "event-channel", "%u", queue
->tx_evtchn
);
1660 message
= "writing event-channel";
1664 /* Split event channels */
1665 err
= xenbus_printf(*xbt
, path
,
1666 "event-channel-tx", "%u", queue
->tx_evtchn
);
1668 message
= "writing event-channel-tx";
1672 err
= xenbus_printf(*xbt
, path
,
1673 "event-channel-rx", "%u", queue
->rx_evtchn
);
1675 message
= "writing event-channel-rx";
1680 if (write_hierarchical
)
1685 if (write_hierarchical
)
1687 xenbus_dev_fatal(dev
, err
, "%s", message
);
1691 static void xennet_destroy_queues(struct netfront_info
*info
)
1697 for (i
= 0; i
< info
->netdev
->real_num_tx_queues
; i
++) {
1698 struct netfront_queue
*queue
= &info
->queues
[i
];
1700 if (netif_running(info
->netdev
))
1701 napi_disable(&queue
->napi
);
1702 del_timer_sync(&queue
->rx_refill_timer
);
1703 netif_napi_del(&queue
->napi
);
1708 kfree(info
->queues
);
1709 info
->queues
= NULL
;
1712 static int xennet_create_queues(struct netfront_info
*info
,
1713 unsigned int num_queues
)
1718 info
->queues
= kcalloc(num_queues
, sizeof(struct netfront_queue
),
1725 for (i
= 0; i
< num_queues
; i
++) {
1726 struct netfront_queue
*queue
= &info
->queues
[i
];
1731 ret
= xennet_init_queue(queue
);
1733 dev_warn(&info
->netdev
->dev
,
1734 "only created %d queues\n", i
);
1739 netif_napi_add(queue
->info
->netdev
, &queue
->napi
,
1741 if (netif_running(info
->netdev
))
1742 napi_enable(&queue
->napi
);
1745 netif_set_real_num_tx_queues(info
->netdev
, num_queues
);
1749 if (num_queues
== 0) {
1750 dev_err(&info
->netdev
->dev
, "no queues\n");
1756 /* Common code used when first setting up, and when resuming. */
1757 static int talk_to_netback(struct xenbus_device
*dev
,
1758 struct netfront_info
*info
)
1760 const char *message
;
1761 struct xenbus_transaction xbt
;
1763 unsigned int feature_split_evtchn
;
1765 unsigned int max_queues
= 0;
1766 struct netfront_queue
*queue
= NULL
;
1767 unsigned int num_queues
= 1;
1769 info
->netdev
->irq
= 0;
1771 /* Check if backend supports multiple queues */
1772 err
= xenbus_scanf(XBT_NIL
, info
->xbdev
->otherend
,
1773 "multi-queue-max-queues", "%u", &max_queues
);
1776 num_queues
= min(max_queues
, xennet_max_queues
);
1778 /* Check feature-split-event-channels */
1779 err
= xenbus_scanf(XBT_NIL
, info
->xbdev
->otherend
,
1780 "feature-split-event-channels", "%u",
1781 &feature_split_evtchn
);
1783 feature_split_evtchn
= 0;
1785 /* Read mac addr. */
1786 err
= xen_net_read_mac(dev
, info
->netdev
->dev_addr
);
1788 xenbus_dev_fatal(dev
, err
, "parsing %s/mac", dev
->nodename
);
1793 xennet_destroy_queues(info
);
1795 err
= xennet_create_queues(info
, num_queues
);
1799 /* Create shared ring, alloc event channel -- for each queue */
1800 for (i
= 0; i
< num_queues
; ++i
) {
1801 queue
= &info
->queues
[i
];
1802 err
= setup_netfront(dev
, queue
, feature_split_evtchn
);
1804 /* setup_netfront() will tidy up the current
1805 * queue on error, but we need to clean up
1806 * those already allocated.
1810 netif_set_real_num_tx_queues(info
->netdev
, i
);
1820 err
= xenbus_transaction_start(&xbt
);
1822 xenbus_dev_fatal(dev
, err
, "starting transaction");
1826 if (num_queues
== 1) {
1827 err
= write_queue_xenstore_keys(&info
->queues
[0], &xbt
, 0); /* flat */
1829 goto abort_transaction_no_dev_fatal
;
1831 /* Write the number of queues */
1832 err
= xenbus_printf(xbt
, dev
->nodename
, "multi-queue-num-queues",
1835 message
= "writing multi-queue-num-queues";
1836 goto abort_transaction_no_dev_fatal
;
1839 /* Write the keys for each queue */
1840 for (i
= 0; i
< num_queues
; ++i
) {
1841 queue
= &info
->queues
[i
];
1842 err
= write_queue_xenstore_keys(queue
, &xbt
, 1); /* hierarchical */
1844 goto abort_transaction_no_dev_fatal
;
1848 /* The remaining keys are not queue-specific */
1849 err
= xenbus_printf(xbt
, dev
->nodename
, "request-rx-copy", "%u",
1852 message
= "writing request-rx-copy";
1853 goto abort_transaction
;
1856 err
= xenbus_printf(xbt
, dev
->nodename
, "feature-rx-notify", "%d", 1);
1858 message
= "writing feature-rx-notify";
1859 goto abort_transaction
;
1862 err
= xenbus_printf(xbt
, dev
->nodename
, "feature-sg", "%d", 1);
1864 message
= "writing feature-sg";
1865 goto abort_transaction
;
1868 err
= xenbus_printf(xbt
, dev
->nodename
, "feature-gso-tcpv4", "%d", 1);
1870 message
= "writing feature-gso-tcpv4";
1871 goto abort_transaction
;
1874 err
= xenbus_write(xbt
, dev
->nodename
, "feature-gso-tcpv6", "1");
1876 message
= "writing feature-gso-tcpv6";
1877 goto abort_transaction
;
1880 err
= xenbus_write(xbt
, dev
->nodename
, "feature-ipv6-csum-offload",
1883 message
= "writing feature-ipv6-csum-offload";
1884 goto abort_transaction
;
1887 err
= xenbus_transaction_end(xbt
, 0);
1891 xenbus_dev_fatal(dev
, err
, "completing transaction");
1898 xenbus_dev_fatal(dev
, err
, "%s", message
);
1899 abort_transaction_no_dev_fatal
:
1900 xenbus_transaction_end(xbt
, 1);
1902 xennet_disconnect_backend(info
);
1903 kfree(info
->queues
);
1904 info
->queues
= NULL
;
1906 netif_set_real_num_tx_queues(info
->netdev
, 0);
1912 static int xennet_connect(struct net_device
*dev
)
1914 struct netfront_info
*np
= netdev_priv(dev
);
1915 unsigned int num_queues
= 0;
1917 unsigned int feature_rx_copy
;
1919 struct netfront_queue
*queue
= NULL
;
1921 err
= xenbus_scanf(XBT_NIL
, np
->xbdev
->otherend
,
1922 "feature-rx-copy", "%u", &feature_rx_copy
);
1924 feature_rx_copy
= 0;
1926 if (!feature_rx_copy
) {
1928 "backend does not support copying receive path\n");
1932 err
= talk_to_netback(np
->xbdev
, np
);
1936 /* talk_to_netback() sets the correct number of queues */
1937 num_queues
= dev
->real_num_tx_queues
;
1940 netdev_update_features(dev
);
1944 * All public and private state should now be sane. Get
1945 * ready to start sending and receiving packets and give the driver
1946 * domain a kick because we've probably just requeued some
1949 netif_carrier_on(np
->netdev
);
1950 for (j
= 0; j
< num_queues
; ++j
) {
1951 queue
= &np
->queues
[j
];
1953 notify_remote_via_irq(queue
->tx_irq
);
1954 if (queue
->tx_irq
!= queue
->rx_irq
)
1955 notify_remote_via_irq(queue
->rx_irq
);
1957 spin_lock_irq(&queue
->tx_lock
);
1958 xennet_tx_buf_gc(queue
);
1959 spin_unlock_irq(&queue
->tx_lock
);
1961 spin_lock_bh(&queue
->rx_lock
);
1962 xennet_alloc_rx_buffers(queue
);
1963 spin_unlock_bh(&queue
->rx_lock
);
1970 * Callback received when the backend's state changes.
1972 static void netback_changed(struct xenbus_device
*dev
,
1973 enum xenbus_state backend_state
)
1975 struct netfront_info
*np
= dev_get_drvdata(&dev
->dev
);
1976 struct net_device
*netdev
= np
->netdev
;
1978 dev_dbg(&dev
->dev
, "%s\n", xenbus_strstate(backend_state
));
1980 switch (backend_state
) {
1981 case XenbusStateInitialising
:
1982 case XenbusStateInitialised
:
1983 case XenbusStateReconfiguring
:
1984 case XenbusStateReconfigured
:
1985 case XenbusStateUnknown
:
1988 case XenbusStateInitWait
:
1989 if (dev
->state
!= XenbusStateInitialising
)
1991 if (xennet_connect(netdev
) != 0)
1993 xenbus_switch_state(dev
, XenbusStateConnected
);
1996 case XenbusStateConnected
:
1997 netdev_notify_peers(netdev
);
2000 case XenbusStateClosed
:
2001 if (dev
->state
== XenbusStateClosed
)
2003 /* Missed the backend's CLOSING state -- fallthrough */
2004 case XenbusStateClosing
:
2005 xenbus_frontend_closed(dev
);
2010 static const struct xennet_stat
{
2011 char name
[ETH_GSTRING_LEN
];
2013 } xennet_stats
[] = {
2015 "rx_gso_checksum_fixup",
2016 offsetof(struct netfront_info
, rx_gso_checksum_fixup
)
2020 static int xennet_get_sset_count(struct net_device
*dev
, int string_set
)
2022 switch (string_set
) {
2024 return ARRAY_SIZE(xennet_stats
);
2030 static void xennet_get_ethtool_stats(struct net_device
*dev
,
2031 struct ethtool_stats
*stats
, u64
* data
)
2033 void *np
= netdev_priv(dev
);
2036 for (i
= 0; i
< ARRAY_SIZE(xennet_stats
); i
++)
2037 data
[i
] = atomic_read((atomic_t
*)(np
+ xennet_stats
[i
].offset
));
2040 static void xennet_get_strings(struct net_device
*dev
, u32 stringset
, u8
* data
)
2044 switch (stringset
) {
2046 for (i
= 0; i
< ARRAY_SIZE(xennet_stats
); i
++)
2047 memcpy(data
+ i
* ETH_GSTRING_LEN
,
2048 xennet_stats
[i
].name
, ETH_GSTRING_LEN
);
2053 static const struct ethtool_ops xennet_ethtool_ops
=
2055 .get_link
= ethtool_op_get_link
,
2057 .get_sset_count
= xennet_get_sset_count
,
2058 .get_ethtool_stats
= xennet_get_ethtool_stats
,
2059 .get_strings
= xennet_get_strings
,
2063 static ssize_t
show_rxbuf(struct device
*dev
,
2064 struct device_attribute
*attr
, char *buf
)
2066 return sprintf(buf
, "%lu\n", NET_RX_RING_SIZE
);
2069 static ssize_t
store_rxbuf(struct device
*dev
,
2070 struct device_attribute
*attr
,
2071 const char *buf
, size_t len
)
2074 unsigned long target
;
2076 if (!capable(CAP_NET_ADMIN
))
2079 target
= simple_strtoul(buf
, &endp
, 0);
2083 /* rxbuf_min and rxbuf_max are no longer configurable. */
2088 static DEVICE_ATTR(rxbuf_min
, S_IRUGO
|S_IWUSR
, show_rxbuf
, store_rxbuf
);
2089 static DEVICE_ATTR(rxbuf_max
, S_IRUGO
|S_IWUSR
, show_rxbuf
, store_rxbuf
);
2090 static DEVICE_ATTR(rxbuf_cur
, S_IRUGO
, show_rxbuf
, NULL
);
2092 static struct attribute
*xennet_dev_attrs
[] = {
2093 &dev_attr_rxbuf_min
.attr
,
2094 &dev_attr_rxbuf_max
.attr
,
2095 &dev_attr_rxbuf_cur
.attr
,
2099 static const struct attribute_group xennet_dev_group
= {
2100 .attrs
= xennet_dev_attrs
2102 #endif /* CONFIG_SYSFS */
2104 static int xennet_remove(struct xenbus_device
*dev
)
2106 struct netfront_info
*info
= dev_get_drvdata(&dev
->dev
);
2108 dev_dbg(&dev
->dev
, "%s\n", dev
->nodename
);
2110 xennet_disconnect_backend(info
);
2112 unregister_netdev(info
->netdev
);
2114 xennet_destroy_queues(info
);
2115 xennet_free_netdev(info
->netdev
);
2120 static const struct xenbus_device_id netfront_ids
[] = {
2125 static struct xenbus_driver netfront_driver
= {
2126 .ids
= netfront_ids
,
2127 .probe
= netfront_probe
,
2128 .remove
= xennet_remove
,
2129 .resume
= netfront_resume
,
2130 .otherend_changed
= netback_changed
,
2133 static int __init
netif_init(void)
2138 if (!xen_has_pv_nic_devices())
2141 pr_info("Initialising Xen virtual ethernet driver\n");
2143 /* Allow as many queues as there are CPUs, by default */
2144 xennet_max_queues
= num_online_cpus();
2146 return xenbus_register_frontend(&netfront_driver
);
2148 module_init(netif_init
);
2151 static void __exit
netif_exit(void)
2153 xenbus_unregister_driver(&netfront_driver
);
2155 module_exit(netif_exit
);
2157 MODULE_DESCRIPTION("Xen virtual network device frontend");
2158 MODULE_LICENSE("GPL");
2159 MODULE_ALIAS("xen:vif");
2160 MODULE_ALIAS("xennet");