2 * Virtual network driver for conversing with remote driver backends.
4 * Copyright (c) 2002-2005, K A Fraser
5 * Copyright (c) 2005, XenSource Ltd
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public License version 2
9 * as published by the Free Software Foundation; or, when distributed
10 * separately from the Linux kernel or incorporated into other
11 * software packages, subject to the following license:
13 * Permission is hereby granted, free of charge, to any person obtaining a copy
14 * of this source file (the "Software"), to deal in the Software without
15 * restriction, including without limitation the rights to use, copy, modify,
16 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
17 * and to permit persons to whom the Software is furnished to do so, subject to
18 * the following conditions:
20 * The above copyright notice and this permission notice shall be included in
21 * all copies or substantial portions of the Software.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
24 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
25 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
26 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
27 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
28 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
32 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
34 #include <linux/module.h>
35 #include <linux/kernel.h>
36 #include <linux/netdevice.h>
37 #include <linux/etherdevice.h>
38 #include <linux/skbuff.h>
39 #include <linux/ethtool.h>
40 #include <linux/if_ether.h>
42 #include <linux/udp.h>
43 #include <linux/moduleparam.h>
45 #include <linux/slab.h>
48 #include <asm/xen/page.h>
50 #include <xen/xenbus.h>
51 #include <xen/events.h>
53 #include <xen/platform_pci.h>
54 #include <xen/grant_table.h>
56 #include <xen/interface/io/netif.h>
57 #include <xen/interface/memory.h>
58 #include <xen/interface/grant_table.h>
60 static const struct ethtool_ops xennet_ethtool_ops
;
66 #define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb))
68 #define RX_COPY_THRESHOLD 256
70 #define GRANT_INVALID_REF 0
72 #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, PAGE_SIZE)
73 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, PAGE_SIZE)
74 #define TX_MAX_TARGET min_t(int, NET_TX_RING_SIZE, 256)
76 struct netfront_stats
{
81 struct u64_stats_sync syncp
;
84 struct netfront_info
{
85 struct list_head list
;
86 struct net_device
*netdev
;
88 struct napi_struct napi
;
90 /* Split event channels support, tx_* == rx_* when using
91 * single event channel.
93 unsigned int tx_evtchn
, rx_evtchn
;
94 unsigned int tx_irq
, rx_irq
;
95 /* Only used when split event channels support is enabled */
96 char tx_irq_name
[IFNAMSIZ
+4]; /* DEVNAME-tx */
97 char rx_irq_name
[IFNAMSIZ
+4]; /* DEVNAME-rx */
99 struct xenbus_device
*xbdev
;
102 struct xen_netif_tx_front_ring tx
;
106 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
107 * are linked from tx_skb_freelist through skb_entry.link.
109 * NB. Freelist index entries are always going to be less than
110 * PAGE_OFFSET, whereas pointers to skbs will always be equal or
111 * greater than PAGE_OFFSET: we use this property to distinguish
117 } tx_skbs
[NET_TX_RING_SIZE
];
118 grant_ref_t gref_tx_head
;
119 grant_ref_t grant_tx_ref
[NET_TX_RING_SIZE
];
120 unsigned tx_skb_freelist
;
122 spinlock_t rx_lock ____cacheline_aligned_in_smp
;
123 struct xen_netif_rx_front_ring rx
;
126 /* Receive-ring batched refills. */
127 #define RX_MIN_TARGET 8
128 #define RX_DFL_MIN_TARGET 64
129 #define RX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256)
130 unsigned rx_min_target
, rx_max_target
, rx_target
;
131 struct sk_buff_head rx_batch
;
133 struct timer_list rx_refill_timer
;
135 struct sk_buff
*rx_skbs
[NET_RX_RING_SIZE
];
136 grant_ref_t gref_rx_head
;
137 grant_ref_t grant_rx_ref
[NET_RX_RING_SIZE
];
139 unsigned long rx_pfn_array
[NET_RX_RING_SIZE
];
140 struct multicall_entry rx_mcl
[NET_RX_RING_SIZE
+1];
141 struct mmu_update rx_mmu
[NET_RX_RING_SIZE
];
144 struct netfront_stats __percpu
*stats
;
146 unsigned long rx_gso_checksum_fixup
;
149 struct netfront_rx_info
{
150 struct xen_netif_rx_response rx
;
151 struct xen_netif_extra_info extras
[XEN_NETIF_EXTRA_TYPE_MAX
- 1];
154 static void skb_entry_set_link(union skb_entry
*list
, unsigned short id
)
159 static int skb_entry_is_link(const union skb_entry
*list
)
161 BUILD_BUG_ON(sizeof(list
->skb
) != sizeof(list
->link
));
162 return (unsigned long)list
->skb
< PAGE_OFFSET
;
166 * Access macros for acquiring freeing slots in tx_skbs[].
169 static void add_id_to_freelist(unsigned *head
, union skb_entry
*list
,
172 skb_entry_set_link(&list
[id
], *head
);
176 static unsigned short get_id_from_freelist(unsigned *head
,
177 union skb_entry
*list
)
179 unsigned int id
= *head
;
180 *head
= list
[id
].link
;
184 static int xennet_rxidx(RING_IDX idx
)
186 return idx
& (NET_RX_RING_SIZE
- 1);
189 static struct sk_buff
*xennet_get_rx_skb(struct netfront_info
*np
,
192 int i
= xennet_rxidx(ri
);
193 struct sk_buff
*skb
= np
->rx_skbs
[i
];
194 np
->rx_skbs
[i
] = NULL
;
198 static grant_ref_t
xennet_get_rx_ref(struct netfront_info
*np
,
201 int i
= xennet_rxidx(ri
);
202 grant_ref_t ref
= np
->grant_rx_ref
[i
];
203 np
->grant_rx_ref
[i
] = GRANT_INVALID_REF
;
208 static int xennet_sysfs_addif(struct net_device
*netdev
);
209 static void xennet_sysfs_delif(struct net_device
*netdev
);
210 #else /* !CONFIG_SYSFS */
211 #define xennet_sysfs_addif(dev) (0)
212 #define xennet_sysfs_delif(dev) do { } while (0)
215 static bool xennet_can_sg(struct net_device
*dev
)
217 return dev
->features
& NETIF_F_SG
;
221 static void rx_refill_timeout(unsigned long data
)
223 struct net_device
*dev
= (struct net_device
*)data
;
224 struct netfront_info
*np
= netdev_priv(dev
);
225 napi_schedule(&np
->napi
);
228 static int netfront_tx_slot_available(struct netfront_info
*np
)
230 return (np
->tx
.req_prod_pvt
- np
->tx
.rsp_cons
) <
231 (TX_MAX_TARGET
- MAX_SKB_FRAGS
- 2);
234 static void xennet_maybe_wake_tx(struct net_device
*dev
)
236 struct netfront_info
*np
= netdev_priv(dev
);
238 if (unlikely(netif_queue_stopped(dev
)) &&
239 netfront_tx_slot_available(np
) &&
240 likely(netif_running(dev
)))
241 netif_wake_queue(dev
);
244 static void xennet_alloc_rx_buffers(struct net_device
*dev
)
247 struct netfront_info
*np
= netdev_priv(dev
);
250 int i
, batch_target
, notify
;
251 RING_IDX req_prod
= np
->rx
.req_prod_pvt
;
255 struct xen_netif_rx_request
*req
;
257 if (unlikely(!netif_carrier_ok(dev
)))
261 * Allocate skbuffs greedily, even though we batch updates to the
262 * receive ring. This creates a less bursty demand on the memory
263 * allocator, so should reduce the chance of failed allocation requests
264 * both for ourself and for other kernel subsystems.
266 batch_target
= np
->rx_target
- (req_prod
- np
->rx
.rsp_cons
);
267 for (i
= skb_queue_len(&np
->rx_batch
); i
< batch_target
; i
++) {
268 skb
= __netdev_alloc_skb(dev
, RX_COPY_THRESHOLD
+ NET_IP_ALIGN
,
269 GFP_ATOMIC
| __GFP_NOWARN
);
273 /* Align ip header to a 16 bytes boundary */
274 skb_reserve(skb
, NET_IP_ALIGN
);
276 page
= alloc_page(GFP_ATOMIC
| __GFP_NOWARN
);
280 /* Could not allocate any skbuffs. Try again later. */
281 mod_timer(&np
->rx_refill_timer
,
284 /* Any skbuffs queued for refill? Force them out. */
290 skb_add_rx_frag(skb
, 0, page
, 0, 0, PAGE_SIZE
);
291 __skb_queue_tail(&np
->rx_batch
, skb
);
294 /* Is the batch large enough to be worthwhile? */
295 if (i
< (np
->rx_target
/2)) {
296 if (req_prod
> np
->rx
.sring
->req_prod
)
301 /* Adjust our fill target if we risked running out of buffers. */
302 if (((req_prod
- np
->rx
.sring
->rsp_prod
) < (np
->rx_target
/ 4)) &&
303 ((np
->rx_target
*= 2) > np
->rx_max_target
))
304 np
->rx_target
= np
->rx_max_target
;
308 skb
= __skb_dequeue(&np
->rx_batch
);
314 id
= xennet_rxidx(req_prod
+ i
);
316 BUG_ON(np
->rx_skbs
[id
]);
317 np
->rx_skbs
[id
] = skb
;
319 ref
= gnttab_claim_grant_reference(&np
->gref_rx_head
);
320 BUG_ON((signed short)ref
< 0);
321 np
->grant_rx_ref
[id
] = ref
;
323 pfn
= page_to_pfn(skb_frag_page(&skb_shinfo(skb
)->frags
[0]));
324 vaddr
= page_address(skb_frag_page(&skb_shinfo(skb
)->frags
[0]));
326 req
= RING_GET_REQUEST(&np
->rx
, req_prod
+ i
);
327 gnttab_grant_foreign_access_ref(ref
,
328 np
->xbdev
->otherend_id
,
336 wmb(); /* barrier so backend seens requests */
338 /* Above is a suitable barrier to ensure backend will see requests. */
339 np
->rx
.req_prod_pvt
= req_prod
+ i
;
341 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np
->rx
, notify
);
343 notify_remote_via_irq(np
->rx_irq
);
346 static int xennet_open(struct net_device
*dev
)
348 struct netfront_info
*np
= netdev_priv(dev
);
350 napi_enable(&np
->napi
);
352 spin_lock_bh(&np
->rx_lock
);
353 if (netif_carrier_ok(dev
)) {
354 xennet_alloc_rx_buffers(dev
);
355 np
->rx
.sring
->rsp_event
= np
->rx
.rsp_cons
+ 1;
356 if (RING_HAS_UNCONSUMED_RESPONSES(&np
->rx
))
357 napi_schedule(&np
->napi
);
359 spin_unlock_bh(&np
->rx_lock
);
361 netif_start_queue(dev
);
366 static void xennet_tx_buf_gc(struct net_device
*dev
)
370 struct netfront_info
*np
= netdev_priv(dev
);
373 BUG_ON(!netif_carrier_ok(dev
));
376 prod
= np
->tx
.sring
->rsp_prod
;
377 rmb(); /* Ensure we see responses up to 'rp'. */
379 for (cons
= np
->tx
.rsp_cons
; cons
!= prod
; cons
++) {
380 struct xen_netif_tx_response
*txrsp
;
382 txrsp
= RING_GET_RESPONSE(&np
->tx
, cons
);
383 if (txrsp
->status
== XEN_NETIF_RSP_NULL
)
387 skb
= np
->tx_skbs
[id
].skb
;
388 if (unlikely(gnttab_query_foreign_access(
389 np
->grant_tx_ref
[id
]) != 0)) {
390 pr_alert("%s: warning -- grant still in use by backend domain\n",
394 gnttab_end_foreign_access_ref(
395 np
->grant_tx_ref
[id
], GNTMAP_readonly
);
396 gnttab_release_grant_reference(
397 &np
->gref_tx_head
, np
->grant_tx_ref
[id
]);
398 np
->grant_tx_ref
[id
] = GRANT_INVALID_REF
;
399 add_id_to_freelist(&np
->tx_skb_freelist
, np
->tx_skbs
, id
);
400 dev_kfree_skb_irq(skb
);
403 np
->tx
.rsp_cons
= prod
;
406 * Set a new event, then check for race with update of tx_cons.
407 * Note that it is essential to schedule a callback, no matter
408 * how few buffers are pending. Even if there is space in the
409 * transmit ring, higher layers may be blocked because too much
410 * data is outstanding: in such cases notification from Xen is
411 * likely to be the only kick that we'll get.
413 np
->tx
.sring
->rsp_event
=
414 prod
+ ((np
->tx
.sring
->req_prod
- prod
) >> 1) + 1;
415 mb(); /* update shared area */
416 } while ((cons
== prod
) && (prod
!= np
->tx
.sring
->rsp_prod
));
418 xennet_maybe_wake_tx(dev
);
421 static void xennet_make_frags(struct sk_buff
*skb
, struct net_device
*dev
,
422 struct xen_netif_tx_request
*tx
)
424 struct netfront_info
*np
= netdev_priv(dev
);
425 char *data
= skb
->data
;
427 RING_IDX prod
= np
->tx
.req_prod_pvt
;
428 int frags
= skb_shinfo(skb
)->nr_frags
;
429 unsigned int offset
= offset_in_page(data
);
430 unsigned int len
= skb_headlen(skb
);
435 /* While the header overlaps a page boundary (including being
436 larger than a page), split it it into page-sized chunks. */
437 while (len
> PAGE_SIZE
- offset
) {
438 tx
->size
= PAGE_SIZE
- offset
;
439 tx
->flags
|= XEN_NETTXF_more_data
;
444 id
= get_id_from_freelist(&np
->tx_skb_freelist
, np
->tx_skbs
);
445 np
->tx_skbs
[id
].skb
= skb_get(skb
);
446 tx
= RING_GET_REQUEST(&np
->tx
, prod
++);
448 ref
= gnttab_claim_grant_reference(&np
->gref_tx_head
);
449 BUG_ON((signed short)ref
< 0);
451 mfn
= virt_to_mfn(data
);
452 gnttab_grant_foreign_access_ref(ref
, np
->xbdev
->otherend_id
,
453 mfn
, GNTMAP_readonly
);
455 tx
->gref
= np
->grant_tx_ref
[id
] = ref
;
461 /* Grant backend access to each skb fragment page. */
462 for (i
= 0; i
< frags
; i
++) {
463 skb_frag_t
*frag
= skb_shinfo(skb
)->frags
+ i
;
464 struct page
*page
= skb_frag_page(frag
);
466 len
= skb_frag_size(frag
);
467 offset
= frag
->page_offset
;
469 /* Data must not cross a page boundary. */
470 BUG_ON(len
+ offset
> PAGE_SIZE
<<compound_order(page
));
472 /* Skip unused frames from start of page */
473 page
+= offset
>> PAGE_SHIFT
;
474 offset
&= ~PAGE_MASK
;
479 BUG_ON(offset
>= PAGE_SIZE
);
481 bytes
= PAGE_SIZE
- offset
;
485 tx
->flags
|= XEN_NETTXF_more_data
;
487 id
= get_id_from_freelist(&np
->tx_skb_freelist
,
489 np
->tx_skbs
[id
].skb
= skb_get(skb
);
490 tx
= RING_GET_REQUEST(&np
->tx
, prod
++);
492 ref
= gnttab_claim_grant_reference(&np
->gref_tx_head
);
493 BUG_ON((signed short)ref
< 0);
495 mfn
= pfn_to_mfn(page_to_pfn(page
));
496 gnttab_grant_foreign_access_ref(ref
,
497 np
->xbdev
->otherend_id
,
498 mfn
, GNTMAP_readonly
);
500 tx
->gref
= np
->grant_tx_ref
[id
] = ref
;
509 if (offset
== PAGE_SIZE
&& len
) {
510 BUG_ON(!PageCompound(page
));
517 np
->tx
.req_prod_pvt
= prod
;
521 * Count how many ring slots are required to send the frags of this
522 * skb. Each frag might be a compound page.
524 static int xennet_count_skb_frag_slots(struct sk_buff
*skb
)
526 int i
, frags
= skb_shinfo(skb
)->nr_frags
;
529 for (i
= 0; i
< frags
; i
++) {
530 skb_frag_t
*frag
= skb_shinfo(skb
)->frags
+ i
;
531 unsigned long size
= skb_frag_size(frag
);
532 unsigned long offset
= frag
->page_offset
;
534 /* Skip unused frames from start of page */
535 offset
&= ~PAGE_MASK
;
537 pages
+= PFN_UP(offset
+ size
);
543 static int xennet_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
546 struct netfront_info
*np
= netdev_priv(dev
);
547 struct netfront_stats
*stats
= this_cpu_ptr(np
->stats
);
548 struct xen_netif_tx_request
*tx
;
549 char *data
= skb
->data
;
555 unsigned int offset
= offset_in_page(data
);
556 unsigned int len
= skb_headlen(skb
);
559 /* If skb->len is too big for wire format, drop skb and alert
560 * user about misconfiguration.
562 if (unlikely(skb
->len
> XEN_NETIF_MAX_TX_SIZE
)) {
563 net_alert_ratelimited(
564 "xennet: skb->len = %u, too big for wire format\n",
569 slots
= DIV_ROUND_UP(offset
+ len
, PAGE_SIZE
) +
570 xennet_count_skb_frag_slots(skb
);
571 if (unlikely(slots
> MAX_SKB_FRAGS
+ 1)) {
572 net_alert_ratelimited(
573 "xennet: skb rides the rocket: %d slots\n", slots
);
577 spin_lock_irqsave(&np
->tx_lock
, flags
);
579 if (unlikely(!netif_carrier_ok(dev
) ||
580 (slots
> 1 && !xennet_can_sg(dev
)) ||
581 netif_needs_gso(skb
, netif_skb_features(skb
)))) {
582 spin_unlock_irqrestore(&np
->tx_lock
, flags
);
586 i
= np
->tx
.req_prod_pvt
;
588 id
= get_id_from_freelist(&np
->tx_skb_freelist
, np
->tx_skbs
);
589 np
->tx_skbs
[id
].skb
= skb
;
591 tx
= RING_GET_REQUEST(&np
->tx
, i
);
594 ref
= gnttab_claim_grant_reference(&np
->gref_tx_head
);
595 BUG_ON((signed short)ref
< 0);
596 mfn
= virt_to_mfn(data
);
597 gnttab_grant_foreign_access_ref(
598 ref
, np
->xbdev
->otherend_id
, mfn
, GNTMAP_readonly
);
599 tx
->gref
= np
->grant_tx_ref
[id
] = ref
;
604 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
606 tx
->flags
|= XEN_NETTXF_csum_blank
| XEN_NETTXF_data_validated
;
607 else if (skb
->ip_summed
== CHECKSUM_UNNECESSARY
)
608 /* remote but checksummed. */
609 tx
->flags
|= XEN_NETTXF_data_validated
;
611 if (skb_shinfo(skb
)->gso_size
) {
612 struct xen_netif_extra_info
*gso
;
614 gso
= (struct xen_netif_extra_info
*)
615 RING_GET_REQUEST(&np
->tx
, ++i
);
617 tx
->flags
|= XEN_NETTXF_extra_info
;
619 gso
->u
.gso
.size
= skb_shinfo(skb
)->gso_size
;
620 gso
->u
.gso
.type
= XEN_NETIF_GSO_TYPE_TCPV4
;
622 gso
->u
.gso
.features
= 0;
624 gso
->type
= XEN_NETIF_EXTRA_TYPE_GSO
;
628 np
->tx
.req_prod_pvt
= i
+ 1;
630 xennet_make_frags(skb
, dev
, tx
);
633 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np
->tx
, notify
);
635 notify_remote_via_irq(np
->tx_irq
);
637 u64_stats_update_begin(&stats
->syncp
);
638 stats
->tx_bytes
+= skb
->len
;
640 u64_stats_update_end(&stats
->syncp
);
642 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
643 xennet_tx_buf_gc(dev
);
645 if (!netfront_tx_slot_available(np
))
646 netif_stop_queue(dev
);
648 spin_unlock_irqrestore(&np
->tx_lock
, flags
);
653 dev
->stats
.tx_dropped
++;
658 static int xennet_close(struct net_device
*dev
)
660 struct netfront_info
*np
= netdev_priv(dev
);
661 netif_stop_queue(np
->netdev
);
662 napi_disable(&np
->napi
);
666 static void xennet_move_rx_slot(struct netfront_info
*np
, struct sk_buff
*skb
,
669 int new = xennet_rxidx(np
->rx
.req_prod_pvt
);
671 BUG_ON(np
->rx_skbs
[new]);
672 np
->rx_skbs
[new] = skb
;
673 np
->grant_rx_ref
[new] = ref
;
674 RING_GET_REQUEST(&np
->rx
, np
->rx
.req_prod_pvt
)->id
= new;
675 RING_GET_REQUEST(&np
->rx
, np
->rx
.req_prod_pvt
)->gref
= ref
;
676 np
->rx
.req_prod_pvt
++;
679 static int xennet_get_extras(struct netfront_info
*np
,
680 struct xen_netif_extra_info
*extras
,
684 struct xen_netif_extra_info
*extra
;
685 struct device
*dev
= &np
->netdev
->dev
;
686 RING_IDX cons
= np
->rx
.rsp_cons
;
693 if (unlikely(cons
+ 1 == rp
)) {
695 dev_warn(dev
, "Missing extra info\n");
700 extra
= (struct xen_netif_extra_info
*)
701 RING_GET_RESPONSE(&np
->rx
, ++cons
);
703 if (unlikely(!extra
->type
||
704 extra
->type
>= XEN_NETIF_EXTRA_TYPE_MAX
)) {
706 dev_warn(dev
, "Invalid extra type: %d\n",
710 memcpy(&extras
[extra
->type
- 1], extra
,
714 skb
= xennet_get_rx_skb(np
, cons
);
715 ref
= xennet_get_rx_ref(np
, cons
);
716 xennet_move_rx_slot(np
, skb
, ref
);
717 } while (extra
->flags
& XEN_NETIF_EXTRA_FLAG_MORE
);
719 np
->rx
.rsp_cons
= cons
;
723 static int xennet_get_responses(struct netfront_info
*np
,
724 struct netfront_rx_info
*rinfo
, RING_IDX rp
,
725 struct sk_buff_head
*list
)
727 struct xen_netif_rx_response
*rx
= &rinfo
->rx
;
728 struct xen_netif_extra_info
*extras
= rinfo
->extras
;
729 struct device
*dev
= &np
->netdev
->dev
;
730 RING_IDX cons
= np
->rx
.rsp_cons
;
731 struct sk_buff
*skb
= xennet_get_rx_skb(np
, cons
);
732 grant_ref_t ref
= xennet_get_rx_ref(np
, cons
);
733 int max
= MAX_SKB_FRAGS
+ (rx
->status
<= RX_COPY_THRESHOLD
);
738 if (rx
->flags
& XEN_NETRXF_extra_info
) {
739 err
= xennet_get_extras(np
, extras
, rp
);
740 cons
= np
->rx
.rsp_cons
;
744 if (unlikely(rx
->status
< 0 ||
745 rx
->offset
+ rx
->status
> PAGE_SIZE
)) {
747 dev_warn(dev
, "rx->offset: %x, size: %u\n",
748 rx
->offset
, rx
->status
);
749 xennet_move_rx_slot(np
, skb
, ref
);
755 * This definitely indicates a bug, either in this driver or in
756 * the backend driver. In future this should flag the bad
757 * situation to the system controller to reboot the backend.
759 if (ref
== GRANT_INVALID_REF
) {
761 dev_warn(dev
, "Bad rx response id %d.\n",
767 ret
= gnttab_end_foreign_access_ref(ref
, 0);
770 gnttab_release_grant_reference(&np
->gref_rx_head
, ref
);
772 __skb_queue_tail(list
, skb
);
775 if (!(rx
->flags
& XEN_NETRXF_more_data
))
778 if (cons
+ slots
== rp
) {
780 dev_warn(dev
, "Need more slots\n");
785 rx
= RING_GET_RESPONSE(&np
->rx
, cons
+ slots
);
786 skb
= xennet_get_rx_skb(np
, cons
+ slots
);
787 ref
= xennet_get_rx_ref(np
, cons
+ slots
);
791 if (unlikely(slots
> max
)) {
793 dev_warn(dev
, "Too many slots\n");
798 np
->rx
.rsp_cons
= cons
+ slots
;
803 static int xennet_set_skb_gso(struct sk_buff
*skb
,
804 struct xen_netif_extra_info
*gso
)
806 if (!gso
->u
.gso
.size
) {
808 pr_warn("GSO size must not be zero\n");
812 /* Currently only TCPv4 S.O. is supported. */
813 if (gso
->u
.gso
.type
!= XEN_NETIF_GSO_TYPE_TCPV4
) {
815 pr_warn("Bad GSO type %d\n", gso
->u
.gso
.type
);
819 skb_shinfo(skb
)->gso_size
= gso
->u
.gso
.size
;
820 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV4
;
822 /* Header must be checked, and gso_segs computed. */
823 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
824 skb_shinfo(skb
)->gso_segs
= 0;
829 static RING_IDX
xennet_fill_frags(struct netfront_info
*np
,
831 struct sk_buff_head
*list
)
833 struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
834 RING_IDX cons
= np
->rx
.rsp_cons
;
835 struct sk_buff
*nskb
;
837 while ((nskb
= __skb_dequeue(list
))) {
838 struct xen_netif_rx_response
*rx
=
839 RING_GET_RESPONSE(&np
->rx
, ++cons
);
840 skb_frag_t
*nfrag
= &skb_shinfo(nskb
)->frags
[0];
842 if (shinfo
->nr_frags
== MAX_SKB_FRAGS
) {
843 unsigned int pull_to
= NETFRONT_SKB_CB(skb
)->pull_to
;
845 BUG_ON(pull_to
<= skb_headlen(skb
));
846 __pskb_pull_tail(skb
, pull_to
- skb_headlen(skb
));
848 BUG_ON(shinfo
->nr_frags
>= MAX_SKB_FRAGS
);
850 skb_add_rx_frag(skb
, shinfo
->nr_frags
, skb_frag_page(nfrag
),
851 rx
->offset
, rx
->status
, PAGE_SIZE
);
853 skb_shinfo(nskb
)->nr_frags
= 0;
860 static int checksum_setup(struct net_device
*dev
, struct sk_buff
*skb
)
864 int recalculate_partial_csum
= 0;
867 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
868 * peers can fail to set NETRXF_csum_blank when sending a GSO
869 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
870 * recalculate the partial checksum.
872 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
&& skb_is_gso(skb
)) {
873 struct netfront_info
*np
= netdev_priv(dev
);
874 np
->rx_gso_checksum_fixup
++;
875 skb
->ip_summed
= CHECKSUM_PARTIAL
;
876 recalculate_partial_csum
= 1;
879 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
880 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
)
883 if (skb
->protocol
!= htons(ETH_P_IP
))
886 iph
= (void *)skb
->data
;
888 switch (iph
->protocol
) {
890 if (!skb_partial_csum_set(skb
, 4 * iph
->ihl
,
891 offsetof(struct tcphdr
, check
)))
894 if (recalculate_partial_csum
) {
895 struct tcphdr
*tcph
= tcp_hdr(skb
);
896 tcph
->check
= ~csum_tcpudp_magic(iph
->saddr
, iph
->daddr
,
897 skb
->len
- iph
->ihl
*4,
902 if (!skb_partial_csum_set(skb
, 4 * iph
->ihl
,
903 offsetof(struct udphdr
, check
)))
906 if (recalculate_partial_csum
) {
907 struct udphdr
*udph
= udp_hdr(skb
);
908 udph
->check
= ~csum_tcpudp_magic(iph
->saddr
, iph
->daddr
,
909 skb
->len
- iph
->ihl
*4,
915 pr_err("Attempting to checksum a non-TCP/UDP packet, dropping a protocol %d packet\n",
926 static int handle_incoming_queue(struct net_device
*dev
,
927 struct sk_buff_head
*rxq
)
929 struct netfront_info
*np
= netdev_priv(dev
);
930 struct netfront_stats
*stats
= this_cpu_ptr(np
->stats
);
931 int packets_dropped
= 0;
934 while ((skb
= __skb_dequeue(rxq
)) != NULL
) {
935 int pull_to
= NETFRONT_SKB_CB(skb
)->pull_to
;
937 if (pull_to
> skb_headlen(skb
))
938 __pskb_pull_tail(skb
, pull_to
- skb_headlen(skb
));
940 /* Ethernet work: Delayed to here as it peeks the header. */
941 skb
->protocol
= eth_type_trans(skb
, dev
);
943 if (checksum_setup(dev
, skb
)) {
946 dev
->stats
.rx_errors
++;
950 u64_stats_update_begin(&stats
->syncp
);
952 stats
->rx_bytes
+= skb
->len
;
953 u64_stats_update_end(&stats
->syncp
);
956 napi_gro_receive(&np
->napi
, skb
);
959 return packets_dropped
;
962 static int xennet_poll(struct napi_struct
*napi
, int budget
)
964 struct netfront_info
*np
= container_of(napi
, struct netfront_info
, napi
);
965 struct net_device
*dev
= np
->netdev
;
967 struct netfront_rx_info rinfo
;
968 struct xen_netif_rx_response
*rx
= &rinfo
.rx
;
969 struct xen_netif_extra_info
*extras
= rinfo
.extras
;
972 struct sk_buff_head rxq
;
973 struct sk_buff_head errq
;
974 struct sk_buff_head tmpq
;
978 spin_lock(&np
->rx_lock
);
980 skb_queue_head_init(&rxq
);
981 skb_queue_head_init(&errq
);
982 skb_queue_head_init(&tmpq
);
984 rp
= np
->rx
.sring
->rsp_prod
;
985 rmb(); /* Ensure we see queued responses up to 'rp'. */
989 while ((i
!= rp
) && (work_done
< budget
)) {
990 memcpy(rx
, RING_GET_RESPONSE(&np
->rx
, i
), sizeof(*rx
));
991 memset(extras
, 0, sizeof(rinfo
.extras
));
993 err
= xennet_get_responses(np
, &rinfo
, rp
, &tmpq
);
997 while ((skb
= __skb_dequeue(&tmpq
)))
998 __skb_queue_tail(&errq
, skb
);
999 dev
->stats
.rx_errors
++;
1000 i
= np
->rx
.rsp_cons
;
1004 skb
= __skb_dequeue(&tmpq
);
1006 if (extras
[XEN_NETIF_EXTRA_TYPE_GSO
- 1].type
) {
1007 struct xen_netif_extra_info
*gso
;
1008 gso
= &extras
[XEN_NETIF_EXTRA_TYPE_GSO
- 1];
1010 if (unlikely(xennet_set_skb_gso(skb
, gso
))) {
1011 __skb_queue_head(&tmpq
, skb
);
1012 np
->rx
.rsp_cons
+= skb_queue_len(&tmpq
);
1017 NETFRONT_SKB_CB(skb
)->pull_to
= rx
->status
;
1018 if (NETFRONT_SKB_CB(skb
)->pull_to
> RX_COPY_THRESHOLD
)
1019 NETFRONT_SKB_CB(skb
)->pull_to
= RX_COPY_THRESHOLD
;
1021 skb_shinfo(skb
)->frags
[0].page_offset
= rx
->offset
;
1022 skb_frag_size_set(&skb_shinfo(skb
)->frags
[0], rx
->status
);
1023 skb
->data_len
= rx
->status
;
1024 skb
->len
+= rx
->status
;
1026 i
= xennet_fill_frags(np
, skb
, &tmpq
);
1028 if (rx
->flags
& XEN_NETRXF_csum_blank
)
1029 skb
->ip_summed
= CHECKSUM_PARTIAL
;
1030 else if (rx
->flags
& XEN_NETRXF_data_validated
)
1031 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1033 __skb_queue_tail(&rxq
, skb
);
1035 np
->rx
.rsp_cons
= ++i
;
1039 __skb_queue_purge(&errq
);
1041 work_done
-= handle_incoming_queue(dev
, &rxq
);
1043 /* If we get a callback with very few responses, reduce fill target. */
1044 /* NB. Note exponential increase, linear decrease. */
1045 if (((np
->rx
.req_prod_pvt
- np
->rx
.sring
->rsp_prod
) >
1046 ((3*np
->rx_target
) / 4)) &&
1047 (--np
->rx_target
< np
->rx_min_target
))
1048 np
->rx_target
= np
->rx_min_target
;
1050 xennet_alloc_rx_buffers(dev
);
1052 if (work_done
< budget
) {
1055 napi_gro_flush(napi
, false);
1057 local_irq_save(flags
);
1059 RING_FINAL_CHECK_FOR_RESPONSES(&np
->rx
, more_to_do
);
1061 __napi_complete(napi
);
1063 local_irq_restore(flags
);
1066 spin_unlock(&np
->rx_lock
);
1071 static int xennet_change_mtu(struct net_device
*dev
, int mtu
)
1073 int max
= xennet_can_sg(dev
) ?
1074 XEN_NETIF_MAX_TX_SIZE
- MAX_TCP_HEADER
: ETH_DATA_LEN
;
1082 static struct rtnl_link_stats64
*xennet_get_stats64(struct net_device
*dev
,
1083 struct rtnl_link_stats64
*tot
)
1085 struct netfront_info
*np
= netdev_priv(dev
);
1088 for_each_possible_cpu(cpu
) {
1089 struct netfront_stats
*stats
= per_cpu_ptr(np
->stats
, cpu
);
1090 u64 rx_packets
, rx_bytes
, tx_packets
, tx_bytes
;
1094 start
= u64_stats_fetch_begin_bh(&stats
->syncp
);
1096 rx_packets
= stats
->rx_packets
;
1097 tx_packets
= stats
->tx_packets
;
1098 rx_bytes
= stats
->rx_bytes
;
1099 tx_bytes
= stats
->tx_bytes
;
1100 } while (u64_stats_fetch_retry_bh(&stats
->syncp
, start
));
1102 tot
->rx_packets
+= rx_packets
;
1103 tot
->tx_packets
+= tx_packets
;
1104 tot
->rx_bytes
+= rx_bytes
;
1105 tot
->tx_bytes
+= tx_bytes
;
1108 tot
->rx_errors
= dev
->stats
.rx_errors
;
1109 tot
->tx_dropped
= dev
->stats
.tx_dropped
;
1114 static void xennet_release_tx_bufs(struct netfront_info
*np
)
1116 struct sk_buff
*skb
;
1119 for (i
= 0; i
< NET_TX_RING_SIZE
; i
++) {
1120 /* Skip over entries which are actually freelist references */
1121 if (skb_entry_is_link(&np
->tx_skbs
[i
]))
1124 skb
= np
->tx_skbs
[i
].skb
;
1125 gnttab_end_foreign_access_ref(np
->grant_tx_ref
[i
],
1127 gnttab_release_grant_reference(&np
->gref_tx_head
,
1128 np
->grant_tx_ref
[i
]);
1129 np
->grant_tx_ref
[i
] = GRANT_INVALID_REF
;
1130 add_id_to_freelist(&np
->tx_skb_freelist
, np
->tx_skbs
, i
);
1131 dev_kfree_skb_irq(skb
);
1135 static void xennet_release_rx_bufs(struct netfront_info
*np
)
1137 struct mmu_update
*mmu
= np
->rx_mmu
;
1138 struct multicall_entry
*mcl
= np
->rx_mcl
;
1139 struct sk_buff_head free_list
;
1140 struct sk_buff
*skb
;
1142 int xfer
= 0, noxfer
= 0, unused
= 0;
1145 dev_warn(&np
->netdev
->dev
, "%s: fix me for copying receiver.\n",
1149 skb_queue_head_init(&free_list
);
1151 spin_lock_bh(&np
->rx_lock
);
1153 for (id
= 0; id
< NET_RX_RING_SIZE
; id
++) {
1154 ref
= np
->grant_rx_ref
[id
];
1155 if (ref
== GRANT_INVALID_REF
) {
1160 skb
= np
->rx_skbs
[id
];
1161 mfn
= gnttab_end_foreign_transfer_ref(ref
);
1162 gnttab_release_grant_reference(&np
->gref_rx_head
, ref
);
1163 np
->grant_rx_ref
[id
] = GRANT_INVALID_REF
;
1166 skb_shinfo(skb
)->nr_frags
= 0;
1172 if (!xen_feature(XENFEAT_auto_translated_physmap
)) {
1173 /* Remap the page. */
1174 const struct page
*page
=
1175 skb_frag_page(&skb_shinfo(skb
)->frags
[0]);
1176 unsigned long pfn
= page_to_pfn(page
);
1177 void *vaddr
= page_address(page
);
1179 MULTI_update_va_mapping(mcl
, (unsigned long)vaddr
,
1180 mfn_pte(mfn
, PAGE_KERNEL
),
1183 mmu
->ptr
= ((u64
)mfn
<< PAGE_SHIFT
)
1184 | MMU_MACHPHYS_UPDATE
;
1188 set_phys_to_machine(pfn
, mfn
);
1190 __skb_queue_tail(&free_list
, skb
);
1194 dev_info(&np
->netdev
->dev
, "%s: %d xfer, %d noxfer, %d unused\n",
1195 __func__
, xfer
, noxfer
, unused
);
1198 if (!xen_feature(XENFEAT_auto_translated_physmap
)) {
1199 /* Do all the remapping work and M2P updates. */
1200 MULTI_mmu_update(mcl
, np
->rx_mmu
, mmu
- np
->rx_mmu
,
1203 HYPERVISOR_multicall(np
->rx_mcl
, mcl
- np
->rx_mcl
);
1207 __skb_queue_purge(&free_list
);
1209 spin_unlock_bh(&np
->rx_lock
);
1212 static void xennet_uninit(struct net_device
*dev
)
1214 struct netfront_info
*np
= netdev_priv(dev
);
1215 xennet_release_tx_bufs(np
);
1216 xennet_release_rx_bufs(np
);
1217 gnttab_free_grant_references(np
->gref_tx_head
);
1218 gnttab_free_grant_references(np
->gref_rx_head
);
1221 static netdev_features_t
xennet_fix_features(struct net_device
*dev
,
1222 netdev_features_t features
)
1224 struct netfront_info
*np
= netdev_priv(dev
);
1227 if (features
& NETIF_F_SG
) {
1228 if (xenbus_scanf(XBT_NIL
, np
->xbdev
->otherend
, "feature-sg",
1233 features
&= ~NETIF_F_SG
;
1236 if (features
& NETIF_F_TSO
) {
1237 if (xenbus_scanf(XBT_NIL
, np
->xbdev
->otherend
,
1238 "feature-gso-tcpv4", "%d", &val
) < 0)
1242 features
&= ~NETIF_F_TSO
;
1248 static int xennet_set_features(struct net_device
*dev
,
1249 netdev_features_t features
)
1251 if (!(features
& NETIF_F_SG
) && dev
->mtu
> ETH_DATA_LEN
) {
1252 netdev_info(dev
, "Reducing MTU because no SG offload");
1253 dev
->mtu
= ETH_DATA_LEN
;
1259 static irqreturn_t
xennet_tx_interrupt(int irq
, void *dev_id
)
1261 struct netfront_info
*np
= dev_id
;
1262 struct net_device
*dev
= np
->netdev
;
1263 unsigned long flags
;
1265 spin_lock_irqsave(&np
->tx_lock
, flags
);
1266 xennet_tx_buf_gc(dev
);
1267 spin_unlock_irqrestore(&np
->tx_lock
, flags
);
1272 static irqreturn_t
xennet_rx_interrupt(int irq
, void *dev_id
)
1274 struct netfront_info
*np
= dev_id
;
1275 struct net_device
*dev
= np
->netdev
;
1277 if (likely(netif_carrier_ok(dev
) &&
1278 RING_HAS_UNCONSUMED_RESPONSES(&np
->rx
)))
1279 napi_schedule(&np
->napi
);
1284 static irqreturn_t
xennet_interrupt(int irq
, void *dev_id
)
1286 xennet_tx_interrupt(irq
, dev_id
);
1287 xennet_rx_interrupt(irq
, dev_id
);
1291 #ifdef CONFIG_NET_POLL_CONTROLLER
1292 static void xennet_poll_controller(struct net_device
*dev
)
1294 xennet_interrupt(0, dev
);
1298 static const struct net_device_ops xennet_netdev_ops
= {
1299 .ndo_open
= xennet_open
,
1300 .ndo_uninit
= xennet_uninit
,
1301 .ndo_stop
= xennet_close
,
1302 .ndo_start_xmit
= xennet_start_xmit
,
1303 .ndo_change_mtu
= xennet_change_mtu
,
1304 .ndo_get_stats64
= xennet_get_stats64
,
1305 .ndo_set_mac_address
= eth_mac_addr
,
1306 .ndo_validate_addr
= eth_validate_addr
,
1307 .ndo_fix_features
= xennet_fix_features
,
1308 .ndo_set_features
= xennet_set_features
,
1309 #ifdef CONFIG_NET_POLL_CONTROLLER
1310 .ndo_poll_controller
= xennet_poll_controller
,
1314 static struct net_device
*xennet_create_dev(struct xenbus_device
*dev
)
1317 struct net_device
*netdev
;
1318 struct netfront_info
*np
;
1320 netdev
= alloc_etherdev(sizeof(struct netfront_info
));
1322 return ERR_PTR(-ENOMEM
);
1324 np
= netdev_priv(netdev
);
1327 spin_lock_init(&np
->tx_lock
);
1328 spin_lock_init(&np
->rx_lock
);
1330 skb_queue_head_init(&np
->rx_batch
);
1331 np
->rx_target
= RX_DFL_MIN_TARGET
;
1332 np
->rx_min_target
= RX_DFL_MIN_TARGET
;
1333 np
->rx_max_target
= RX_MAX_TARGET
;
1335 init_timer(&np
->rx_refill_timer
);
1336 np
->rx_refill_timer
.data
= (unsigned long)netdev
;
1337 np
->rx_refill_timer
.function
= rx_refill_timeout
;
1340 np
->stats
= alloc_percpu(struct netfront_stats
);
1341 if (np
->stats
== NULL
)
1344 for_each_possible_cpu(i
) {
1345 struct netfront_stats
*xen_nf_stats
;
1346 xen_nf_stats
= per_cpu_ptr(np
->stats
, i
);
1347 u64_stats_init(&xen_nf_stats
->syncp
);
1350 /* Initialise tx_skbs as a free chain containing every entry. */
1351 np
->tx_skb_freelist
= 0;
1352 for (i
= 0; i
< NET_TX_RING_SIZE
; i
++) {
1353 skb_entry_set_link(&np
->tx_skbs
[i
], i
+1);
1354 np
->grant_tx_ref
[i
] = GRANT_INVALID_REF
;
1357 /* Clear out rx_skbs */
1358 for (i
= 0; i
< NET_RX_RING_SIZE
; i
++) {
1359 np
->rx_skbs
[i
] = NULL
;
1360 np
->grant_rx_ref
[i
] = GRANT_INVALID_REF
;
1363 /* A grant for every tx ring slot */
1364 if (gnttab_alloc_grant_references(TX_MAX_TARGET
,
1365 &np
->gref_tx_head
) < 0) {
1366 pr_alert("can't alloc tx grant refs\n");
1368 goto exit_free_stats
;
1370 /* A grant for every rx ring slot */
1371 if (gnttab_alloc_grant_references(RX_MAX_TARGET
,
1372 &np
->gref_rx_head
) < 0) {
1373 pr_alert("can't alloc rx grant refs\n");
1378 netdev
->netdev_ops
= &xennet_netdev_ops
;
1380 netif_napi_add(netdev
, &np
->napi
, xennet_poll
, 64);
1381 netdev
->features
= NETIF_F_IP_CSUM
| NETIF_F_RXCSUM
|
1383 netdev
->hw_features
= NETIF_F_IP_CSUM
| NETIF_F_SG
| NETIF_F_TSO
;
1386 * Assume that all hw features are available for now. This set
1387 * will be adjusted by the call to netdev_update_features() in
1388 * xennet_connect() which is the earliest point where we can
1389 * negotiate with the backend regarding supported features.
1391 netdev
->features
|= netdev
->hw_features
;
1393 SET_ETHTOOL_OPS(netdev
, &xennet_ethtool_ops
);
1394 SET_NETDEV_DEV(netdev
, &dev
->dev
);
1396 netif_set_gso_max_size(netdev
, XEN_NETIF_MAX_TX_SIZE
- MAX_TCP_HEADER
);
1398 np
->netdev
= netdev
;
1400 netif_carrier_off(netdev
);
1405 gnttab_free_grant_references(np
->gref_tx_head
);
1407 free_percpu(np
->stats
);
1409 free_netdev(netdev
);
1410 return ERR_PTR(err
);
1414 * Entry point to this code when a new device is created. Allocate the basic
1415 * structures and the ring buffers for communication with the backend, and
1416 * inform the backend of the appropriate details for those.
1418 static int netfront_probe(struct xenbus_device
*dev
,
1419 const struct xenbus_device_id
*id
)
1422 struct net_device
*netdev
;
1423 struct netfront_info
*info
;
1425 netdev
= xennet_create_dev(dev
);
1426 if (IS_ERR(netdev
)) {
1427 err
= PTR_ERR(netdev
);
1428 xenbus_dev_fatal(dev
, err
, "creating netdev");
1432 info
= netdev_priv(netdev
);
1433 dev_set_drvdata(&dev
->dev
, info
);
1435 err
= register_netdev(info
->netdev
);
1437 pr_warn("%s: register_netdev err=%d\n", __func__
, err
);
1441 err
= xennet_sysfs_addif(info
->netdev
);
1443 unregister_netdev(info
->netdev
);
1444 pr_warn("%s: add sysfs failed err=%d\n", __func__
, err
);
1451 free_netdev(netdev
);
1452 dev_set_drvdata(&dev
->dev
, NULL
);
1456 static void xennet_end_access(int ref
, void *page
)
1458 /* This frees the page as a side-effect */
1459 if (ref
!= GRANT_INVALID_REF
)
1460 gnttab_end_foreign_access(ref
, 0, (unsigned long)page
);
1463 static void xennet_disconnect_backend(struct netfront_info
*info
)
1465 /* Stop old i/f to prevent errors whilst we rebuild the state. */
1466 spin_lock_bh(&info
->rx_lock
);
1467 spin_lock_irq(&info
->tx_lock
);
1468 netif_carrier_off(info
->netdev
);
1469 spin_unlock_irq(&info
->tx_lock
);
1470 spin_unlock_bh(&info
->rx_lock
);
1472 if (info
->tx_irq
&& (info
->tx_irq
== info
->rx_irq
))
1473 unbind_from_irqhandler(info
->tx_irq
, info
);
1474 if (info
->tx_irq
&& (info
->tx_irq
!= info
->rx_irq
)) {
1475 unbind_from_irqhandler(info
->tx_irq
, info
);
1476 unbind_from_irqhandler(info
->rx_irq
, info
);
1478 info
->tx_evtchn
= info
->rx_evtchn
= 0;
1479 info
->tx_irq
= info
->rx_irq
= 0;
1481 /* End access and free the pages */
1482 xennet_end_access(info
->tx_ring_ref
, info
->tx
.sring
);
1483 xennet_end_access(info
->rx_ring_ref
, info
->rx
.sring
);
1485 info
->tx_ring_ref
= GRANT_INVALID_REF
;
1486 info
->rx_ring_ref
= GRANT_INVALID_REF
;
1487 info
->tx
.sring
= NULL
;
1488 info
->rx
.sring
= NULL
;
1492 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1493 * driver restart. We tear down our netif structure and recreate it, but
1494 * leave the device-layer structures intact so that this is transparent to the
1495 * rest of the kernel.
1497 static int netfront_resume(struct xenbus_device
*dev
)
1499 struct netfront_info
*info
= dev_get_drvdata(&dev
->dev
);
1501 dev_dbg(&dev
->dev
, "%s\n", dev
->nodename
);
1503 xennet_disconnect_backend(info
);
1507 static int xen_net_read_mac(struct xenbus_device
*dev
, u8 mac
[])
1509 char *s
, *e
, *macstr
;
1512 macstr
= s
= xenbus_read(XBT_NIL
, dev
->nodename
, "mac", NULL
);
1514 return PTR_ERR(macstr
);
1516 for (i
= 0; i
< ETH_ALEN
; i
++) {
1517 mac
[i
] = simple_strtoul(s
, &e
, 16);
1518 if ((s
== e
) || (*e
!= ((i
== ETH_ALEN
-1) ? '\0' : ':'))) {
1529 static int setup_netfront_single(struct netfront_info
*info
)
1533 err
= xenbus_alloc_evtchn(info
->xbdev
, &info
->tx_evtchn
);
1537 err
= bind_evtchn_to_irqhandler(info
->tx_evtchn
,
1539 0, info
->netdev
->name
, info
);
1542 info
->rx_evtchn
= info
->tx_evtchn
;
1543 info
->rx_irq
= info
->tx_irq
= err
;
1548 xenbus_free_evtchn(info
->xbdev
, info
->tx_evtchn
);
1549 info
->tx_evtchn
= 0;
1554 static int setup_netfront_split(struct netfront_info
*info
)
1558 err
= xenbus_alloc_evtchn(info
->xbdev
, &info
->tx_evtchn
);
1561 err
= xenbus_alloc_evtchn(info
->xbdev
, &info
->rx_evtchn
);
1563 goto alloc_rx_evtchn_fail
;
1565 snprintf(info
->tx_irq_name
, sizeof(info
->tx_irq_name
),
1566 "%s-tx", info
->netdev
->name
);
1567 err
= bind_evtchn_to_irqhandler(info
->tx_evtchn
,
1568 xennet_tx_interrupt
,
1569 0, info
->tx_irq_name
, info
);
1574 snprintf(info
->rx_irq_name
, sizeof(info
->rx_irq_name
),
1575 "%s-rx", info
->netdev
->name
);
1576 err
= bind_evtchn_to_irqhandler(info
->rx_evtchn
,
1577 xennet_rx_interrupt
,
1578 0, info
->rx_irq_name
, info
);
1586 unbind_from_irqhandler(info
->tx_irq
, info
);
1589 xenbus_free_evtchn(info
->xbdev
, info
->rx_evtchn
);
1590 info
->rx_evtchn
= 0;
1591 alloc_rx_evtchn_fail
:
1592 xenbus_free_evtchn(info
->xbdev
, info
->tx_evtchn
);
1593 info
->tx_evtchn
= 0;
1598 static int setup_netfront(struct xenbus_device
*dev
, struct netfront_info
*info
)
1600 struct xen_netif_tx_sring
*txs
;
1601 struct xen_netif_rx_sring
*rxs
;
1603 struct net_device
*netdev
= info
->netdev
;
1604 unsigned int feature_split_evtchn
;
1606 info
->tx_ring_ref
= GRANT_INVALID_REF
;
1607 info
->rx_ring_ref
= GRANT_INVALID_REF
;
1608 info
->rx
.sring
= NULL
;
1609 info
->tx
.sring
= NULL
;
1612 err
= xenbus_scanf(XBT_NIL
, info
->xbdev
->otherend
,
1613 "feature-split-event-channels", "%u",
1614 &feature_split_evtchn
);
1616 feature_split_evtchn
= 0;
1618 err
= xen_net_read_mac(dev
, netdev
->dev_addr
);
1620 xenbus_dev_fatal(dev
, err
, "parsing %s/mac", dev
->nodename
);
1624 txs
= (struct xen_netif_tx_sring
*)get_zeroed_page(GFP_NOIO
| __GFP_HIGH
);
1627 xenbus_dev_fatal(dev
, err
, "allocating tx ring page");
1630 SHARED_RING_INIT(txs
);
1631 FRONT_RING_INIT(&info
->tx
, txs
, PAGE_SIZE
);
1633 err
= xenbus_grant_ring(dev
, virt_to_mfn(txs
));
1635 goto grant_tx_ring_fail
;
1637 info
->tx_ring_ref
= err
;
1638 rxs
= (struct xen_netif_rx_sring
*)get_zeroed_page(GFP_NOIO
| __GFP_HIGH
);
1641 xenbus_dev_fatal(dev
, err
, "allocating rx ring page");
1642 goto alloc_rx_ring_fail
;
1644 SHARED_RING_INIT(rxs
);
1645 FRONT_RING_INIT(&info
->rx
, rxs
, PAGE_SIZE
);
1647 err
= xenbus_grant_ring(dev
, virt_to_mfn(rxs
));
1649 goto grant_rx_ring_fail
;
1650 info
->rx_ring_ref
= err
;
1652 if (feature_split_evtchn
)
1653 err
= setup_netfront_split(info
);
1654 /* setup single event channel if
1655 * a) feature-split-event-channels == 0
1656 * b) feature-split-event-channels == 1 but failed to setup
1658 if (!feature_split_evtchn
|| (feature_split_evtchn
&& err
))
1659 err
= setup_netfront_single(info
);
1662 goto alloc_evtchn_fail
;
1666 /* If we fail to setup netfront, it is safe to just revoke access to
1667 * granted pages because backend is not accessing it at this point.
1670 gnttab_end_foreign_access_ref(info
->rx_ring_ref
, 0);
1672 free_page((unsigned long)rxs
);
1674 gnttab_end_foreign_access_ref(info
->tx_ring_ref
, 0);
1676 free_page((unsigned long)txs
);
1681 /* Common code used when first setting up, and when resuming. */
1682 static int talk_to_netback(struct xenbus_device
*dev
,
1683 struct netfront_info
*info
)
1685 const char *message
;
1686 struct xenbus_transaction xbt
;
1689 /* Create shared ring, alloc event channel. */
1690 err
= setup_netfront(dev
, info
);
1695 err
= xenbus_transaction_start(&xbt
);
1697 xenbus_dev_fatal(dev
, err
, "starting transaction");
1701 err
= xenbus_printf(xbt
, dev
->nodename
, "tx-ring-ref", "%u",
1704 message
= "writing tx ring-ref";
1705 goto abort_transaction
;
1707 err
= xenbus_printf(xbt
, dev
->nodename
, "rx-ring-ref", "%u",
1710 message
= "writing rx ring-ref";
1711 goto abort_transaction
;
1714 if (info
->tx_evtchn
== info
->rx_evtchn
) {
1715 err
= xenbus_printf(xbt
, dev
->nodename
,
1716 "event-channel", "%u", info
->tx_evtchn
);
1718 message
= "writing event-channel";
1719 goto abort_transaction
;
1722 err
= xenbus_printf(xbt
, dev
->nodename
,
1723 "event-channel-tx", "%u", info
->tx_evtchn
);
1725 message
= "writing event-channel-tx";
1726 goto abort_transaction
;
1728 err
= xenbus_printf(xbt
, dev
->nodename
,
1729 "event-channel-rx", "%u", info
->rx_evtchn
);
1731 message
= "writing event-channel-rx";
1732 goto abort_transaction
;
1736 err
= xenbus_printf(xbt
, dev
->nodename
, "request-rx-copy", "%u",
1739 message
= "writing request-rx-copy";
1740 goto abort_transaction
;
1743 err
= xenbus_printf(xbt
, dev
->nodename
, "feature-rx-notify", "%d", 1);
1745 message
= "writing feature-rx-notify";
1746 goto abort_transaction
;
1749 err
= xenbus_printf(xbt
, dev
->nodename
, "feature-sg", "%d", 1);
1751 message
= "writing feature-sg";
1752 goto abort_transaction
;
1755 err
= xenbus_printf(xbt
, dev
->nodename
, "feature-gso-tcpv4", "%d", 1);
1757 message
= "writing feature-gso-tcpv4";
1758 goto abort_transaction
;
1761 err
= xenbus_transaction_end(xbt
, 0);
1765 xenbus_dev_fatal(dev
, err
, "completing transaction");
1772 xenbus_transaction_end(xbt
, 1);
1773 xenbus_dev_fatal(dev
, err
, "%s", message
);
1775 xennet_disconnect_backend(info
);
1780 static int xennet_connect(struct net_device
*dev
)
1782 struct netfront_info
*np
= netdev_priv(dev
);
1783 int i
, requeue_idx
, err
;
1784 struct sk_buff
*skb
;
1786 struct xen_netif_rx_request
*req
;
1787 unsigned int feature_rx_copy
;
1789 err
= xenbus_scanf(XBT_NIL
, np
->xbdev
->otherend
,
1790 "feature-rx-copy", "%u", &feature_rx_copy
);
1792 feature_rx_copy
= 0;
1794 if (!feature_rx_copy
) {
1796 "backend does not support copying receive path\n");
1800 err
= talk_to_netback(np
->xbdev
, np
);
1805 netdev_update_features(dev
);
1808 spin_lock_bh(&np
->rx_lock
);
1809 spin_lock_irq(&np
->tx_lock
);
1811 /* Step 1: Discard all pending TX packet fragments. */
1812 xennet_release_tx_bufs(np
);
1814 /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1815 for (requeue_idx
= 0, i
= 0; i
< NET_RX_RING_SIZE
; i
++) {
1817 const struct page
*page
;
1818 if (!np
->rx_skbs
[i
])
1821 skb
= np
->rx_skbs
[requeue_idx
] = xennet_get_rx_skb(np
, i
);
1822 ref
= np
->grant_rx_ref
[requeue_idx
] = xennet_get_rx_ref(np
, i
);
1823 req
= RING_GET_REQUEST(&np
->rx
, requeue_idx
);
1825 frag
= &skb_shinfo(skb
)->frags
[0];
1826 page
= skb_frag_page(frag
);
1827 gnttab_grant_foreign_access_ref(
1828 ref
, np
->xbdev
->otherend_id
,
1829 pfn_to_mfn(page_to_pfn(page
)),
1832 req
->id
= requeue_idx
;
1837 np
->rx
.req_prod_pvt
= requeue_idx
;
1840 * Step 3: All public and private state should now be sane. Get
1841 * ready to start sending and receiving packets and give the driver
1842 * domain a kick because we've probably just requeued some
1845 netif_carrier_on(np
->netdev
);
1846 notify_remote_via_irq(np
->tx_irq
);
1847 if (np
->tx_irq
!= np
->rx_irq
)
1848 notify_remote_via_irq(np
->rx_irq
);
1849 xennet_tx_buf_gc(dev
);
1850 xennet_alloc_rx_buffers(dev
);
1852 spin_unlock_irq(&np
->tx_lock
);
1853 spin_unlock_bh(&np
->rx_lock
);
1859 * Callback received when the backend's state changes.
1861 static void netback_changed(struct xenbus_device
*dev
,
1862 enum xenbus_state backend_state
)
1864 struct netfront_info
*np
= dev_get_drvdata(&dev
->dev
);
1865 struct net_device
*netdev
= np
->netdev
;
1867 dev_dbg(&dev
->dev
, "%s\n", xenbus_strstate(backend_state
));
1869 switch (backend_state
) {
1870 case XenbusStateInitialising
:
1871 case XenbusStateInitialised
:
1872 case XenbusStateReconfiguring
:
1873 case XenbusStateReconfigured
:
1874 case XenbusStateUnknown
:
1875 case XenbusStateClosed
:
1878 case XenbusStateInitWait
:
1879 if (dev
->state
!= XenbusStateInitialising
)
1881 if (xennet_connect(netdev
) != 0)
1883 xenbus_switch_state(dev
, XenbusStateConnected
);
1886 case XenbusStateConnected
:
1887 netdev_notify_peers(netdev
);
1890 case XenbusStateClosing
:
1891 xenbus_frontend_closed(dev
);
1896 static const struct xennet_stat
{
1897 char name
[ETH_GSTRING_LEN
];
1899 } xennet_stats
[] = {
1901 "rx_gso_checksum_fixup",
1902 offsetof(struct netfront_info
, rx_gso_checksum_fixup
)
1906 static int xennet_get_sset_count(struct net_device
*dev
, int string_set
)
1908 switch (string_set
) {
1910 return ARRAY_SIZE(xennet_stats
);
1916 static void xennet_get_ethtool_stats(struct net_device
*dev
,
1917 struct ethtool_stats
*stats
, u64
* data
)
1919 void *np
= netdev_priv(dev
);
1922 for (i
= 0; i
< ARRAY_SIZE(xennet_stats
); i
++)
1923 data
[i
] = *(unsigned long *)(np
+ xennet_stats
[i
].offset
);
1926 static void xennet_get_strings(struct net_device
*dev
, u32 stringset
, u8
* data
)
1930 switch (stringset
) {
1932 for (i
= 0; i
< ARRAY_SIZE(xennet_stats
); i
++)
1933 memcpy(data
+ i
* ETH_GSTRING_LEN
,
1934 xennet_stats
[i
].name
, ETH_GSTRING_LEN
);
1939 static const struct ethtool_ops xennet_ethtool_ops
=
1941 .get_link
= ethtool_op_get_link
,
1943 .get_sset_count
= xennet_get_sset_count
,
1944 .get_ethtool_stats
= xennet_get_ethtool_stats
,
1945 .get_strings
= xennet_get_strings
,
1949 static ssize_t
show_rxbuf_min(struct device
*dev
,
1950 struct device_attribute
*attr
, char *buf
)
1952 struct net_device
*netdev
= to_net_dev(dev
);
1953 struct netfront_info
*info
= netdev_priv(netdev
);
1955 return sprintf(buf
, "%u\n", info
->rx_min_target
);
1958 static ssize_t
store_rxbuf_min(struct device
*dev
,
1959 struct device_attribute
*attr
,
1960 const char *buf
, size_t len
)
1962 struct net_device
*netdev
= to_net_dev(dev
);
1963 struct netfront_info
*np
= netdev_priv(netdev
);
1965 unsigned long target
;
1967 if (!capable(CAP_NET_ADMIN
))
1970 target
= simple_strtoul(buf
, &endp
, 0);
1974 if (target
< RX_MIN_TARGET
)
1975 target
= RX_MIN_TARGET
;
1976 if (target
> RX_MAX_TARGET
)
1977 target
= RX_MAX_TARGET
;
1979 spin_lock_bh(&np
->rx_lock
);
1980 if (target
> np
->rx_max_target
)
1981 np
->rx_max_target
= target
;
1982 np
->rx_min_target
= target
;
1983 if (target
> np
->rx_target
)
1984 np
->rx_target
= target
;
1986 xennet_alloc_rx_buffers(netdev
);
1988 spin_unlock_bh(&np
->rx_lock
);
1992 static ssize_t
show_rxbuf_max(struct device
*dev
,
1993 struct device_attribute
*attr
, char *buf
)
1995 struct net_device
*netdev
= to_net_dev(dev
);
1996 struct netfront_info
*info
= netdev_priv(netdev
);
1998 return sprintf(buf
, "%u\n", info
->rx_max_target
);
2001 static ssize_t
store_rxbuf_max(struct device
*dev
,
2002 struct device_attribute
*attr
,
2003 const char *buf
, size_t len
)
2005 struct net_device
*netdev
= to_net_dev(dev
);
2006 struct netfront_info
*np
= netdev_priv(netdev
);
2008 unsigned long target
;
2010 if (!capable(CAP_NET_ADMIN
))
2013 target
= simple_strtoul(buf
, &endp
, 0);
2017 if (target
< RX_MIN_TARGET
)
2018 target
= RX_MIN_TARGET
;
2019 if (target
> RX_MAX_TARGET
)
2020 target
= RX_MAX_TARGET
;
2022 spin_lock_bh(&np
->rx_lock
);
2023 if (target
< np
->rx_min_target
)
2024 np
->rx_min_target
= target
;
2025 np
->rx_max_target
= target
;
2026 if (target
< np
->rx_target
)
2027 np
->rx_target
= target
;
2029 xennet_alloc_rx_buffers(netdev
);
2031 spin_unlock_bh(&np
->rx_lock
);
2035 static ssize_t
show_rxbuf_cur(struct device
*dev
,
2036 struct device_attribute
*attr
, char *buf
)
2038 struct net_device
*netdev
= to_net_dev(dev
);
2039 struct netfront_info
*info
= netdev_priv(netdev
);
2041 return sprintf(buf
, "%u\n", info
->rx_target
);
2044 static struct device_attribute xennet_attrs
[] = {
2045 __ATTR(rxbuf_min
, S_IRUGO
|S_IWUSR
, show_rxbuf_min
, store_rxbuf_min
),
2046 __ATTR(rxbuf_max
, S_IRUGO
|S_IWUSR
, show_rxbuf_max
, store_rxbuf_max
),
2047 __ATTR(rxbuf_cur
, S_IRUGO
, show_rxbuf_cur
, NULL
),
2050 static int xennet_sysfs_addif(struct net_device
*netdev
)
2055 for (i
= 0; i
< ARRAY_SIZE(xennet_attrs
); i
++) {
2056 err
= device_create_file(&netdev
->dev
,
2065 device_remove_file(&netdev
->dev
, &xennet_attrs
[i
]);
2069 static void xennet_sysfs_delif(struct net_device
*netdev
)
2073 for (i
= 0; i
< ARRAY_SIZE(xennet_attrs
); i
++)
2074 device_remove_file(&netdev
->dev
, &xennet_attrs
[i
]);
2077 #endif /* CONFIG_SYSFS */
2079 static const struct xenbus_device_id netfront_ids
[] = {
2085 static int xennet_remove(struct xenbus_device
*dev
)
2087 struct netfront_info
*info
= dev_get_drvdata(&dev
->dev
);
2089 dev_dbg(&dev
->dev
, "%s\n", dev
->nodename
);
2091 xennet_disconnect_backend(info
);
2093 xennet_sysfs_delif(info
->netdev
);
2095 unregister_netdev(info
->netdev
);
2097 del_timer_sync(&info
->rx_refill_timer
);
2099 free_percpu(info
->stats
);
2101 free_netdev(info
->netdev
);
2106 static DEFINE_XENBUS_DRIVER(netfront
, ,
2107 .probe
= netfront_probe
,
2108 .remove
= xennet_remove
,
2109 .resume
= netfront_resume
,
2110 .otherend_changed
= netback_changed
,
2113 static int __init
netif_init(void)
2118 if (xen_hvm_domain() && !xen_platform_pci_unplug
)
2121 pr_info("Initialising Xen virtual ethernet driver\n");
2123 return xenbus_register_frontend(&netfront_driver
);
2125 module_init(netif_init
);
2128 static void __exit
netif_exit(void)
2130 xenbus_unregister_driver(&netfront_driver
);
2132 module_exit(netif_exit
);
2134 MODULE_DESCRIPTION("Xen virtual network device frontend");
2135 MODULE_LICENSE("GPL");
2136 MODULE_ALIAS("xen:vif");
2137 MODULE_ALIAS("xennet");