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 #include <linux/module.h>
33 #include <linux/kernel.h>
34 #include <linux/netdevice.h>
35 #include <linux/etherdevice.h>
36 #include <linux/skbuff.h>
37 #include <linux/ethtool.h>
38 #include <linux/if_ether.h>
40 #include <linux/udp.h>
41 #include <linux/moduleparam.h>
43 #include <linux/slab.h>
46 #include <asm/xen/page.h>
48 #include <xen/xenbus.h>
49 #include <xen/events.h>
51 #include <xen/platform_pci.h>
52 #include <xen/grant_table.h>
54 #include <xen/interface/io/netif.h>
55 #include <xen/interface/memory.h>
56 #include <xen/interface/grant_table.h>
58 static const struct ethtool_ops xennet_ethtool_ops
;
64 #define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb))
66 #define RX_COPY_THRESHOLD 256
68 #define GRANT_INVALID_REF 0
70 #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, PAGE_SIZE)
71 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, PAGE_SIZE)
72 #define TX_MAX_TARGET min_t(int, NET_TX_RING_SIZE, 256)
74 struct netfront_stats
{
79 struct u64_stats_sync syncp
;
82 struct netfront_info
{
83 struct list_head list
;
84 struct net_device
*netdev
;
86 struct napi_struct napi
;
88 /* Split event channels support, tx_* == rx_* when using
89 * single event channel.
91 unsigned int tx_evtchn
, rx_evtchn
;
92 unsigned int tx_irq
, rx_irq
;
93 /* Only used when split event channels support is enabled */
94 char tx_irq_name
[IFNAMSIZ
+4]; /* DEVNAME-tx */
95 char rx_irq_name
[IFNAMSIZ
+4]; /* DEVNAME-rx */
97 struct xenbus_device
*xbdev
;
100 struct xen_netif_tx_front_ring tx
;
104 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
105 * are linked from tx_skb_freelist through skb_entry.link.
107 * NB. Freelist index entries are always going to be less than
108 * PAGE_OFFSET, whereas pointers to skbs will always be equal or
109 * greater than PAGE_OFFSET: we use this property to distinguish
115 } tx_skbs
[NET_TX_RING_SIZE
];
116 grant_ref_t gref_tx_head
;
117 grant_ref_t grant_tx_ref
[NET_TX_RING_SIZE
];
118 unsigned tx_skb_freelist
;
120 spinlock_t rx_lock ____cacheline_aligned_in_smp
;
121 struct xen_netif_rx_front_ring rx
;
124 /* Receive-ring batched refills. */
125 #define RX_MIN_TARGET 8
126 #define RX_DFL_MIN_TARGET 64
127 #define RX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256)
128 unsigned rx_min_target
, rx_max_target
, rx_target
;
129 struct sk_buff_head rx_batch
;
131 struct timer_list rx_refill_timer
;
133 struct sk_buff
*rx_skbs
[NET_RX_RING_SIZE
];
134 grant_ref_t gref_rx_head
;
135 grant_ref_t grant_rx_ref
[NET_RX_RING_SIZE
];
137 unsigned long rx_pfn_array
[NET_RX_RING_SIZE
];
138 struct multicall_entry rx_mcl
[NET_RX_RING_SIZE
+1];
139 struct mmu_update rx_mmu
[NET_RX_RING_SIZE
];
142 struct netfront_stats __percpu
*stats
;
144 unsigned long rx_gso_checksum_fixup
;
147 struct netfront_rx_info
{
148 struct xen_netif_rx_response rx
;
149 struct xen_netif_extra_info extras
[XEN_NETIF_EXTRA_TYPE_MAX
- 1];
152 static void skb_entry_set_link(union skb_entry
*list
, unsigned short id
)
157 static int skb_entry_is_link(const union skb_entry
*list
)
159 BUILD_BUG_ON(sizeof(list
->skb
) != sizeof(list
->link
));
160 return (unsigned long)list
->skb
< PAGE_OFFSET
;
164 * Access macros for acquiring freeing slots in tx_skbs[].
167 static void add_id_to_freelist(unsigned *head
, union skb_entry
*list
,
170 skb_entry_set_link(&list
[id
], *head
);
174 static unsigned short get_id_from_freelist(unsigned *head
,
175 union skb_entry
*list
)
177 unsigned int id
= *head
;
178 *head
= list
[id
].link
;
182 static int xennet_rxidx(RING_IDX idx
)
184 return idx
& (NET_RX_RING_SIZE
- 1);
187 static struct sk_buff
*xennet_get_rx_skb(struct netfront_info
*np
,
190 int i
= xennet_rxidx(ri
);
191 struct sk_buff
*skb
= np
->rx_skbs
[i
];
192 np
->rx_skbs
[i
] = NULL
;
196 static grant_ref_t
xennet_get_rx_ref(struct netfront_info
*np
,
199 int i
= xennet_rxidx(ri
);
200 grant_ref_t ref
= np
->grant_rx_ref
[i
];
201 np
->grant_rx_ref
[i
] = GRANT_INVALID_REF
;
206 static int xennet_sysfs_addif(struct net_device
*netdev
);
207 static void xennet_sysfs_delif(struct net_device
*netdev
);
208 #else /* !CONFIG_SYSFS */
209 #define xennet_sysfs_addif(dev) (0)
210 #define xennet_sysfs_delif(dev) do { } while (0)
213 static bool xennet_can_sg(struct net_device
*dev
)
215 return dev
->features
& NETIF_F_SG
;
219 static void rx_refill_timeout(unsigned long data
)
221 struct net_device
*dev
= (struct net_device
*)data
;
222 struct netfront_info
*np
= netdev_priv(dev
);
223 napi_schedule(&np
->napi
);
226 static int netfront_tx_slot_available(struct netfront_info
*np
)
228 return (np
->tx
.req_prod_pvt
- np
->tx
.rsp_cons
) <
229 (TX_MAX_TARGET
- MAX_SKB_FRAGS
- 2);
232 static void xennet_maybe_wake_tx(struct net_device
*dev
)
234 struct netfront_info
*np
= netdev_priv(dev
);
236 if (unlikely(netif_queue_stopped(dev
)) &&
237 netfront_tx_slot_available(np
) &&
238 likely(netif_running(dev
)))
239 netif_wake_queue(dev
);
242 static void xennet_alloc_rx_buffers(struct net_device
*dev
)
245 struct netfront_info
*np
= netdev_priv(dev
);
248 int i
, batch_target
, notify
;
249 RING_IDX req_prod
= np
->rx
.req_prod_pvt
;
253 struct xen_netif_rx_request
*req
;
255 if (unlikely(!netif_carrier_ok(dev
)))
259 * Allocate skbuffs greedily, even though we batch updates to the
260 * receive ring. This creates a less bursty demand on the memory
261 * allocator, so should reduce the chance of failed allocation requests
262 * both for ourself and for other kernel subsystems.
264 batch_target
= np
->rx_target
- (req_prod
- np
->rx
.rsp_cons
);
265 for (i
= skb_queue_len(&np
->rx_batch
); i
< batch_target
; i
++) {
266 skb
= __netdev_alloc_skb(dev
, RX_COPY_THRESHOLD
+ NET_IP_ALIGN
,
267 GFP_ATOMIC
| __GFP_NOWARN
);
271 /* Align ip header to a 16 bytes boundary */
272 skb_reserve(skb
, NET_IP_ALIGN
);
274 page
= alloc_page(GFP_ATOMIC
| __GFP_NOWARN
);
278 /* Any skbuffs queued for refill? Force them out. */
281 /* Could not allocate any skbuffs. Try again later. */
282 mod_timer(&np
->rx_refill_timer
,
287 __skb_fill_page_desc(skb
, 0, page
, 0, 0);
288 skb_shinfo(skb
)->nr_frags
= 1;
289 __skb_queue_tail(&np
->rx_batch
, skb
);
292 /* Is the batch large enough to be worthwhile? */
293 if (i
< (np
->rx_target
/2)) {
294 if (req_prod
> np
->rx
.sring
->req_prod
)
299 /* Adjust our fill target if we risked running out of buffers. */
300 if (((req_prod
- np
->rx
.sring
->rsp_prod
) < (np
->rx_target
/ 4)) &&
301 ((np
->rx_target
*= 2) > np
->rx_max_target
))
302 np
->rx_target
= np
->rx_max_target
;
306 skb
= __skb_dequeue(&np
->rx_batch
);
312 id
= xennet_rxidx(req_prod
+ i
);
314 BUG_ON(np
->rx_skbs
[id
]);
315 np
->rx_skbs
[id
] = skb
;
317 ref
= gnttab_claim_grant_reference(&np
->gref_rx_head
);
318 BUG_ON((signed short)ref
< 0);
319 np
->grant_rx_ref
[id
] = ref
;
321 pfn
= page_to_pfn(skb_frag_page(&skb_shinfo(skb
)->frags
[0]));
322 vaddr
= page_address(skb_frag_page(&skb_shinfo(skb
)->frags
[0]));
324 req
= RING_GET_REQUEST(&np
->rx
, req_prod
+ i
);
325 gnttab_grant_foreign_access_ref(ref
,
326 np
->xbdev
->otherend_id
,
334 wmb(); /* barrier so backend seens requests */
336 /* Above is a suitable barrier to ensure backend will see requests. */
337 np
->rx
.req_prod_pvt
= req_prod
+ i
;
339 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np
->rx
, notify
);
341 notify_remote_via_irq(np
->rx_irq
);
344 static int xennet_open(struct net_device
*dev
)
346 struct netfront_info
*np
= netdev_priv(dev
);
348 napi_enable(&np
->napi
);
350 spin_lock_bh(&np
->rx_lock
);
351 if (netif_carrier_ok(dev
)) {
352 xennet_alloc_rx_buffers(dev
);
353 np
->rx
.sring
->rsp_event
= np
->rx
.rsp_cons
+ 1;
354 if (RING_HAS_UNCONSUMED_RESPONSES(&np
->rx
))
355 napi_schedule(&np
->napi
);
357 spin_unlock_bh(&np
->rx_lock
);
359 netif_start_queue(dev
);
364 static void xennet_tx_buf_gc(struct net_device
*dev
)
368 struct netfront_info
*np
= netdev_priv(dev
);
371 BUG_ON(!netif_carrier_ok(dev
));
374 prod
= np
->tx
.sring
->rsp_prod
;
375 rmb(); /* Ensure we see responses up to 'rp'. */
377 for (cons
= np
->tx
.rsp_cons
; cons
!= prod
; cons
++) {
378 struct xen_netif_tx_response
*txrsp
;
380 txrsp
= RING_GET_RESPONSE(&np
->tx
, cons
);
381 if (txrsp
->status
== XEN_NETIF_RSP_NULL
)
385 skb
= np
->tx_skbs
[id
].skb
;
386 if (unlikely(gnttab_query_foreign_access(
387 np
->grant_tx_ref
[id
]) != 0)) {
388 printk(KERN_ALERT
"xennet_tx_buf_gc: warning "
389 "-- grant still in use by backend "
393 gnttab_end_foreign_access_ref(
394 np
->grant_tx_ref
[id
], GNTMAP_readonly
);
395 gnttab_release_grant_reference(
396 &np
->gref_tx_head
, np
->grant_tx_ref
[id
]);
397 np
->grant_tx_ref
[id
] = GRANT_INVALID_REF
;
398 add_id_to_freelist(&np
->tx_skb_freelist
, np
->tx_skbs
, id
);
399 dev_kfree_skb_irq(skb
);
402 np
->tx
.rsp_cons
= prod
;
405 * Set a new event, then check for race with update of tx_cons.
406 * Note that it is essential to schedule a callback, no matter
407 * how few buffers are pending. Even if there is space in the
408 * transmit ring, higher layers may be blocked because too much
409 * data is outstanding: in such cases notification from Xen is
410 * likely to be the only kick that we'll get.
412 np
->tx
.sring
->rsp_event
=
413 prod
+ ((np
->tx
.sring
->req_prod
- prod
) >> 1) + 1;
414 mb(); /* update shared area */
415 } while ((cons
== prod
) && (prod
!= np
->tx
.sring
->rsp_prod
));
417 xennet_maybe_wake_tx(dev
);
420 static void xennet_make_frags(struct sk_buff
*skb
, struct net_device
*dev
,
421 struct xen_netif_tx_request
*tx
)
423 struct netfront_info
*np
= netdev_priv(dev
);
424 char *data
= skb
->data
;
426 RING_IDX prod
= np
->tx
.req_prod_pvt
;
427 int frags
= skb_shinfo(skb
)->nr_frags
;
428 unsigned int offset
= offset_in_page(data
);
429 unsigned int len
= skb_headlen(skb
);
434 /* While the header overlaps a page boundary (including being
435 larger than a page), split it it into page-sized chunks. */
436 while (len
> PAGE_SIZE
- offset
) {
437 tx
->size
= PAGE_SIZE
- offset
;
438 tx
->flags
|= XEN_NETTXF_more_data
;
443 id
= get_id_from_freelist(&np
->tx_skb_freelist
, np
->tx_skbs
);
444 np
->tx_skbs
[id
].skb
= skb_get(skb
);
445 tx
= RING_GET_REQUEST(&np
->tx
, prod
++);
447 ref
= gnttab_claim_grant_reference(&np
->gref_tx_head
);
448 BUG_ON((signed short)ref
< 0);
450 mfn
= virt_to_mfn(data
);
451 gnttab_grant_foreign_access_ref(ref
, np
->xbdev
->otherend_id
,
452 mfn
, GNTMAP_readonly
);
454 tx
->gref
= np
->grant_tx_ref
[id
] = ref
;
460 /* Grant backend access to each skb fragment page. */
461 for (i
= 0; i
< frags
; i
++) {
462 skb_frag_t
*frag
= skb_shinfo(skb
)->frags
+ i
;
463 struct page
*page
= skb_frag_page(frag
);
465 len
= skb_frag_size(frag
);
466 offset
= frag
->page_offset
;
468 /* Data must not cross a page boundary. */
469 BUG_ON(len
+ offset
> PAGE_SIZE
<<compound_order(page
));
471 /* Skip unused frames from start of page */
472 page
+= offset
>> PAGE_SHIFT
;
473 offset
&= ~PAGE_MASK
;
478 BUG_ON(offset
>= PAGE_SIZE
);
480 bytes
= PAGE_SIZE
- offset
;
484 tx
->flags
|= XEN_NETTXF_more_data
;
486 id
= get_id_from_freelist(&np
->tx_skb_freelist
,
488 np
->tx_skbs
[id
].skb
= skb_get(skb
);
489 tx
= RING_GET_REQUEST(&np
->tx
, prod
++);
491 ref
= gnttab_claim_grant_reference(&np
->gref_tx_head
);
492 BUG_ON((signed short)ref
< 0);
494 mfn
= pfn_to_mfn(page_to_pfn(page
));
495 gnttab_grant_foreign_access_ref(ref
,
496 np
->xbdev
->otherend_id
,
497 mfn
, GNTMAP_readonly
);
499 tx
->gref
= np
->grant_tx_ref
[id
] = ref
;
508 if (offset
== PAGE_SIZE
&& len
) {
509 BUG_ON(!PageCompound(page
));
516 np
->tx
.req_prod_pvt
= prod
;
520 * Count how many ring slots are required to send the frags of this
521 * skb. Each frag might be a compound page.
523 static int xennet_count_skb_frag_slots(struct sk_buff
*skb
)
525 int i
, frags
= skb_shinfo(skb
)->nr_frags
;
528 for (i
= 0; i
< frags
; i
++) {
529 skb_frag_t
*frag
= skb_shinfo(skb
)->frags
+ i
;
530 unsigned long size
= skb_frag_size(frag
);
531 unsigned long offset
= frag
->page_offset
;
533 /* Skip unused frames from start of page */
534 offset
&= ~PAGE_MASK
;
536 pages
+= PFN_UP(offset
+ size
);
542 static int xennet_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
)
545 struct netfront_info
*np
= netdev_priv(dev
);
546 struct netfront_stats
*stats
= this_cpu_ptr(np
->stats
);
547 struct xen_netif_tx_request
*tx
;
548 char *data
= skb
->data
;
554 unsigned int offset
= offset_in_page(data
);
555 unsigned int len
= skb_headlen(skb
);
558 /* If skb->len is too big for wire format, drop skb and alert
559 * user about misconfiguration.
561 if (unlikely(skb
->len
> XEN_NETIF_MAX_TX_SIZE
)) {
562 net_alert_ratelimited(
563 "xennet: skb->len = %u, too big for wire format\n",
568 slots
= DIV_ROUND_UP(offset
+ len
, PAGE_SIZE
) +
569 xennet_count_skb_frag_slots(skb
);
570 if (unlikely(slots
> MAX_SKB_FRAGS
+ 1)) {
571 net_alert_ratelimited(
572 "xennet: skb rides the rocket: %d slots\n", slots
);
576 spin_lock_irqsave(&np
->tx_lock
, flags
);
578 if (unlikely(!netif_carrier_ok(dev
) ||
579 (slots
> 1 && !xennet_can_sg(dev
)) ||
580 netif_needs_gso(skb
, netif_skb_features(skb
)))) {
581 spin_unlock_irqrestore(&np
->tx_lock
, flags
);
585 i
= np
->tx
.req_prod_pvt
;
587 id
= get_id_from_freelist(&np
->tx_skb_freelist
, np
->tx_skbs
);
588 np
->tx_skbs
[id
].skb
= skb
;
590 tx
= RING_GET_REQUEST(&np
->tx
, i
);
593 ref
= gnttab_claim_grant_reference(&np
->gref_tx_head
);
594 BUG_ON((signed short)ref
< 0);
595 mfn
= virt_to_mfn(data
);
596 gnttab_grant_foreign_access_ref(
597 ref
, np
->xbdev
->otherend_id
, mfn
, GNTMAP_readonly
);
598 tx
->gref
= np
->grant_tx_ref
[id
] = ref
;
603 if (skb
->ip_summed
== CHECKSUM_PARTIAL
)
605 tx
->flags
|= XEN_NETTXF_csum_blank
| XEN_NETTXF_data_validated
;
606 else if (skb
->ip_summed
== CHECKSUM_UNNECESSARY
)
607 /* remote but checksummed. */
608 tx
->flags
|= XEN_NETTXF_data_validated
;
610 if (skb_shinfo(skb
)->gso_size
) {
611 struct xen_netif_extra_info
*gso
;
613 gso
= (struct xen_netif_extra_info
*)
614 RING_GET_REQUEST(&np
->tx
, ++i
);
616 tx
->flags
|= XEN_NETTXF_extra_info
;
618 gso
->u
.gso
.size
= skb_shinfo(skb
)->gso_size
;
619 gso
->u
.gso
.type
= XEN_NETIF_GSO_TYPE_TCPV4
;
621 gso
->u
.gso
.features
= 0;
623 gso
->type
= XEN_NETIF_EXTRA_TYPE_GSO
;
627 np
->tx
.req_prod_pvt
= i
+ 1;
629 xennet_make_frags(skb
, dev
, tx
);
632 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np
->tx
, notify
);
634 notify_remote_via_irq(np
->tx_irq
);
636 u64_stats_update_begin(&stats
->syncp
);
637 stats
->tx_bytes
+= skb
->len
;
639 u64_stats_update_end(&stats
->syncp
);
641 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
642 xennet_tx_buf_gc(dev
);
644 if (!netfront_tx_slot_available(np
))
645 netif_stop_queue(dev
);
647 spin_unlock_irqrestore(&np
->tx_lock
, flags
);
652 dev
->stats
.tx_dropped
++;
657 static int xennet_close(struct net_device
*dev
)
659 struct netfront_info
*np
= netdev_priv(dev
);
660 netif_stop_queue(np
->netdev
);
661 napi_disable(&np
->napi
);
665 static void xennet_move_rx_slot(struct netfront_info
*np
, struct sk_buff
*skb
,
668 int new = xennet_rxidx(np
->rx
.req_prod_pvt
);
670 BUG_ON(np
->rx_skbs
[new]);
671 np
->rx_skbs
[new] = skb
;
672 np
->grant_rx_ref
[new] = ref
;
673 RING_GET_REQUEST(&np
->rx
, np
->rx
.req_prod_pvt
)->id
= new;
674 RING_GET_REQUEST(&np
->rx
, np
->rx
.req_prod_pvt
)->gref
= ref
;
675 np
->rx
.req_prod_pvt
++;
678 static int xennet_get_extras(struct netfront_info
*np
,
679 struct xen_netif_extra_info
*extras
,
683 struct xen_netif_extra_info
*extra
;
684 struct device
*dev
= &np
->netdev
->dev
;
685 RING_IDX cons
= np
->rx
.rsp_cons
;
692 if (unlikely(cons
+ 1 == rp
)) {
694 dev_warn(dev
, "Missing extra info\n");
699 extra
= (struct xen_netif_extra_info
*)
700 RING_GET_RESPONSE(&np
->rx
, ++cons
);
702 if (unlikely(!extra
->type
||
703 extra
->type
>= XEN_NETIF_EXTRA_TYPE_MAX
)) {
705 dev_warn(dev
, "Invalid extra type: %d\n",
709 memcpy(&extras
[extra
->type
- 1], extra
,
713 skb
= xennet_get_rx_skb(np
, cons
);
714 ref
= xennet_get_rx_ref(np
, cons
);
715 xennet_move_rx_slot(np
, skb
, ref
);
716 } while (extra
->flags
& XEN_NETIF_EXTRA_FLAG_MORE
);
718 np
->rx
.rsp_cons
= cons
;
722 static int xennet_get_responses(struct netfront_info
*np
,
723 struct netfront_rx_info
*rinfo
, RING_IDX rp
,
724 struct sk_buff_head
*list
)
726 struct xen_netif_rx_response
*rx
= &rinfo
->rx
;
727 struct xen_netif_extra_info
*extras
= rinfo
->extras
;
728 struct device
*dev
= &np
->netdev
->dev
;
729 RING_IDX cons
= np
->rx
.rsp_cons
;
730 struct sk_buff
*skb
= xennet_get_rx_skb(np
, cons
);
731 grant_ref_t ref
= xennet_get_rx_ref(np
, cons
);
732 int max
= MAX_SKB_FRAGS
+ (rx
->status
<= RX_COPY_THRESHOLD
);
737 if (rx
->flags
& XEN_NETRXF_extra_info
) {
738 err
= xennet_get_extras(np
, extras
, rp
);
739 cons
= np
->rx
.rsp_cons
;
743 if (unlikely(rx
->status
< 0 ||
744 rx
->offset
+ rx
->status
> PAGE_SIZE
)) {
746 dev_warn(dev
, "rx->offset: %x, size: %u\n",
747 rx
->offset
, rx
->status
);
748 xennet_move_rx_slot(np
, skb
, ref
);
754 * This definitely indicates a bug, either in this driver or in
755 * the backend driver. In future this should flag the bad
756 * situation to the system controller to reboot the backend.
758 if (ref
== GRANT_INVALID_REF
) {
760 dev_warn(dev
, "Bad rx response id %d.\n",
766 ret
= gnttab_end_foreign_access_ref(ref
, 0);
769 gnttab_release_grant_reference(&np
->gref_rx_head
, ref
);
771 __skb_queue_tail(list
, skb
);
774 if (!(rx
->flags
& XEN_NETRXF_more_data
))
777 if (cons
+ slots
== rp
) {
779 dev_warn(dev
, "Need more slots\n");
784 rx
= RING_GET_RESPONSE(&np
->rx
, cons
+ slots
);
785 skb
= xennet_get_rx_skb(np
, cons
+ slots
);
786 ref
= xennet_get_rx_ref(np
, cons
+ slots
);
790 if (unlikely(slots
> max
)) {
792 dev_warn(dev
, "Too many slots\n");
797 np
->rx
.rsp_cons
= cons
+ slots
;
802 static int xennet_set_skb_gso(struct sk_buff
*skb
,
803 struct xen_netif_extra_info
*gso
)
805 if (!gso
->u
.gso
.size
) {
807 printk(KERN_WARNING
"GSO size must not be zero.\n");
811 /* Currently only TCPv4 S.O. is supported. */
812 if (gso
->u
.gso
.type
!= XEN_NETIF_GSO_TYPE_TCPV4
) {
814 printk(KERN_WARNING
"Bad GSO type %d.\n", gso
->u
.gso
.type
);
818 skb_shinfo(skb
)->gso_size
= gso
->u
.gso
.size
;
819 skb_shinfo(skb
)->gso_type
= SKB_GSO_TCPV4
;
821 /* Header must be checked, and gso_segs computed. */
822 skb_shinfo(skb
)->gso_type
|= SKB_GSO_DODGY
;
823 skb_shinfo(skb
)->gso_segs
= 0;
828 static RING_IDX
xennet_fill_frags(struct netfront_info
*np
,
830 struct sk_buff_head
*list
)
832 struct skb_shared_info
*shinfo
= skb_shinfo(skb
);
833 int nr_frags
= shinfo
->nr_frags
;
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 __skb_fill_page_desc(skb
, nr_frags
,
843 skb_frag_page(nfrag
),
844 rx
->offset
, rx
->status
);
846 skb
->data_len
+= rx
->status
;
848 skb_shinfo(nskb
)->nr_frags
= 0;
854 shinfo
->nr_frags
= nr_frags
;
858 static int checksum_setup(struct net_device
*dev
, struct sk_buff
*skb
)
862 int recalculate_partial_csum
= 0;
865 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
866 * peers can fail to set NETRXF_csum_blank when sending a GSO
867 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
868 * recalculate the partial checksum.
870 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
&& skb_is_gso(skb
)) {
871 struct netfront_info
*np
= netdev_priv(dev
);
872 np
->rx_gso_checksum_fixup
++;
873 skb
->ip_summed
= CHECKSUM_PARTIAL
;
874 recalculate_partial_csum
= 1;
877 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
878 if (skb
->ip_summed
!= CHECKSUM_PARTIAL
)
881 if (skb
->protocol
!= htons(ETH_P_IP
))
884 iph
= (void *)skb
->data
;
886 switch (iph
->protocol
) {
888 if (!skb_partial_csum_set(skb
, 4 * iph
->ihl
,
889 offsetof(struct tcphdr
, check
)))
892 if (recalculate_partial_csum
) {
893 struct tcphdr
*tcph
= tcp_hdr(skb
);
894 tcph
->check
= ~csum_tcpudp_magic(iph
->saddr
, iph
->daddr
,
895 skb
->len
- iph
->ihl
*4,
900 if (!skb_partial_csum_set(skb
, 4 * iph
->ihl
,
901 offsetof(struct udphdr
, check
)))
904 if (recalculate_partial_csum
) {
905 struct udphdr
*udph
= udp_hdr(skb
);
906 udph
->check
= ~csum_tcpudp_magic(iph
->saddr
, iph
->daddr
,
907 skb
->len
- iph
->ihl
*4,
913 printk(KERN_ERR
"Attempting to checksum a non-"
914 "TCP/UDP packet, dropping a protocol"
915 " %d packet", iph
->protocol
);
925 static int handle_incoming_queue(struct net_device
*dev
,
926 struct sk_buff_head
*rxq
)
928 struct netfront_info
*np
= netdev_priv(dev
);
929 struct netfront_stats
*stats
= this_cpu_ptr(np
->stats
);
930 int packets_dropped
= 0;
933 while ((skb
= __skb_dequeue(rxq
)) != NULL
) {
934 int pull_to
= NETFRONT_SKB_CB(skb
)->pull_to
;
936 __pskb_pull_tail(skb
, pull_to
- skb_headlen(skb
));
938 /* Ethernet work: Delayed to here as it peeks the header. */
939 skb
->protocol
= eth_type_trans(skb
, dev
);
941 if (checksum_setup(dev
, skb
)) {
944 dev
->stats
.rx_errors
++;
948 u64_stats_update_begin(&stats
->syncp
);
950 stats
->rx_bytes
+= skb
->len
;
951 u64_stats_update_end(&stats
->syncp
);
954 netif_receive_skb(skb
);
957 return packets_dropped
;
960 static int xennet_poll(struct napi_struct
*napi
, int budget
)
962 struct netfront_info
*np
= container_of(napi
, struct netfront_info
, napi
);
963 struct net_device
*dev
= np
->netdev
;
965 struct netfront_rx_info rinfo
;
966 struct xen_netif_rx_response
*rx
= &rinfo
.rx
;
967 struct xen_netif_extra_info
*extras
= rinfo
.extras
;
970 struct sk_buff_head rxq
;
971 struct sk_buff_head errq
;
972 struct sk_buff_head tmpq
;
976 spin_lock(&np
->rx_lock
);
978 skb_queue_head_init(&rxq
);
979 skb_queue_head_init(&errq
);
980 skb_queue_head_init(&tmpq
);
982 rp
= np
->rx
.sring
->rsp_prod
;
983 rmb(); /* Ensure we see queued responses up to 'rp'. */
987 while ((i
!= rp
) && (work_done
< budget
)) {
988 memcpy(rx
, RING_GET_RESPONSE(&np
->rx
, i
), sizeof(*rx
));
989 memset(extras
, 0, sizeof(rinfo
.extras
));
991 err
= xennet_get_responses(np
, &rinfo
, rp
, &tmpq
);
995 while ((skb
= __skb_dequeue(&tmpq
)))
996 __skb_queue_tail(&errq
, skb
);
997 dev
->stats
.rx_errors
++;
1002 skb
= __skb_dequeue(&tmpq
);
1004 if (extras
[XEN_NETIF_EXTRA_TYPE_GSO
- 1].type
) {
1005 struct xen_netif_extra_info
*gso
;
1006 gso
= &extras
[XEN_NETIF_EXTRA_TYPE_GSO
- 1];
1008 if (unlikely(xennet_set_skb_gso(skb
, gso
))) {
1009 __skb_queue_head(&tmpq
, skb
);
1010 np
->rx
.rsp_cons
+= skb_queue_len(&tmpq
);
1015 NETFRONT_SKB_CB(skb
)->pull_to
= rx
->status
;
1016 if (NETFRONT_SKB_CB(skb
)->pull_to
> RX_COPY_THRESHOLD
)
1017 NETFRONT_SKB_CB(skb
)->pull_to
= RX_COPY_THRESHOLD
;
1019 skb_shinfo(skb
)->frags
[0].page_offset
= rx
->offset
;
1020 skb_frag_size_set(&skb_shinfo(skb
)->frags
[0], rx
->status
);
1021 skb
->data_len
= rx
->status
;
1023 i
= xennet_fill_frags(np
, skb
, &tmpq
);
1026 * Truesize is the actual allocation size, even if the
1027 * allocation is only partially used.
1029 skb
->truesize
+= PAGE_SIZE
* skb_shinfo(skb
)->nr_frags
;
1030 skb
->len
+= skb
->data_len
;
1032 if (rx
->flags
& XEN_NETRXF_csum_blank
)
1033 skb
->ip_summed
= CHECKSUM_PARTIAL
;
1034 else if (rx
->flags
& XEN_NETRXF_data_validated
)
1035 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
1037 __skb_queue_tail(&rxq
, skb
);
1039 np
->rx
.rsp_cons
= ++i
;
1043 __skb_queue_purge(&errq
);
1045 work_done
-= handle_incoming_queue(dev
, &rxq
);
1047 /* If we get a callback with very few responses, reduce fill target. */
1048 /* NB. Note exponential increase, linear decrease. */
1049 if (((np
->rx
.req_prod_pvt
- np
->rx
.sring
->rsp_prod
) >
1050 ((3*np
->rx_target
) / 4)) &&
1051 (--np
->rx_target
< np
->rx_min_target
))
1052 np
->rx_target
= np
->rx_min_target
;
1054 xennet_alloc_rx_buffers(dev
);
1056 if (work_done
< budget
) {
1059 local_irq_save(flags
);
1061 RING_FINAL_CHECK_FOR_RESPONSES(&np
->rx
, more_to_do
);
1063 __napi_complete(napi
);
1065 local_irq_restore(flags
);
1068 spin_unlock(&np
->rx_lock
);
1073 static int xennet_change_mtu(struct net_device
*dev
, int mtu
)
1075 int max
= xennet_can_sg(dev
) ?
1076 XEN_NETIF_MAX_TX_SIZE
- MAX_TCP_HEADER
: ETH_DATA_LEN
;
1084 static struct rtnl_link_stats64
*xennet_get_stats64(struct net_device
*dev
,
1085 struct rtnl_link_stats64
*tot
)
1087 struct netfront_info
*np
= netdev_priv(dev
);
1090 for_each_possible_cpu(cpu
) {
1091 struct netfront_stats
*stats
= per_cpu_ptr(np
->stats
, cpu
);
1092 u64 rx_packets
, rx_bytes
, tx_packets
, tx_bytes
;
1096 start
= u64_stats_fetch_begin_bh(&stats
->syncp
);
1098 rx_packets
= stats
->rx_packets
;
1099 tx_packets
= stats
->tx_packets
;
1100 rx_bytes
= stats
->rx_bytes
;
1101 tx_bytes
= stats
->tx_bytes
;
1102 } while (u64_stats_fetch_retry_bh(&stats
->syncp
, start
));
1104 tot
->rx_packets
+= rx_packets
;
1105 tot
->tx_packets
+= tx_packets
;
1106 tot
->rx_bytes
+= rx_bytes
;
1107 tot
->tx_bytes
+= tx_bytes
;
1110 tot
->rx_errors
= dev
->stats
.rx_errors
;
1111 tot
->tx_dropped
= dev
->stats
.tx_dropped
;
1116 static void xennet_release_tx_bufs(struct netfront_info
*np
)
1118 struct sk_buff
*skb
;
1121 for (i
= 0; i
< NET_TX_RING_SIZE
; i
++) {
1122 /* Skip over entries which are actually freelist references */
1123 if (skb_entry_is_link(&np
->tx_skbs
[i
]))
1126 skb
= np
->tx_skbs
[i
].skb
;
1127 gnttab_end_foreign_access_ref(np
->grant_tx_ref
[i
],
1129 gnttab_release_grant_reference(&np
->gref_tx_head
,
1130 np
->grant_tx_ref
[i
]);
1131 np
->grant_tx_ref
[i
] = GRANT_INVALID_REF
;
1132 add_id_to_freelist(&np
->tx_skb_freelist
, np
->tx_skbs
, i
);
1133 dev_kfree_skb_irq(skb
);
1137 static void xennet_release_rx_bufs(struct netfront_info
*np
)
1139 struct mmu_update
*mmu
= np
->rx_mmu
;
1140 struct multicall_entry
*mcl
= np
->rx_mcl
;
1141 struct sk_buff_head free_list
;
1142 struct sk_buff
*skb
;
1144 int xfer
= 0, noxfer
= 0, unused
= 0;
1147 dev_warn(&np
->netdev
->dev
, "%s: fix me for copying receiver.\n",
1151 skb_queue_head_init(&free_list
);
1153 spin_lock_bh(&np
->rx_lock
);
1155 for (id
= 0; id
< NET_RX_RING_SIZE
; id
++) {
1156 ref
= np
->grant_rx_ref
[id
];
1157 if (ref
== GRANT_INVALID_REF
) {
1162 skb
= np
->rx_skbs
[id
];
1163 mfn
= gnttab_end_foreign_transfer_ref(ref
);
1164 gnttab_release_grant_reference(&np
->gref_rx_head
, ref
);
1165 np
->grant_rx_ref
[id
] = GRANT_INVALID_REF
;
1168 skb_shinfo(skb
)->nr_frags
= 0;
1174 if (!xen_feature(XENFEAT_auto_translated_physmap
)) {
1175 /* Remap the page. */
1176 const struct page
*page
=
1177 skb_frag_page(&skb_shinfo(skb
)->frags
[0]);
1178 unsigned long pfn
= page_to_pfn(page
);
1179 void *vaddr
= page_address(page
);
1181 MULTI_update_va_mapping(mcl
, (unsigned long)vaddr
,
1182 mfn_pte(mfn
, PAGE_KERNEL
),
1185 mmu
->ptr
= ((u64
)mfn
<< PAGE_SHIFT
)
1186 | MMU_MACHPHYS_UPDATE
;
1190 set_phys_to_machine(pfn
, mfn
);
1192 __skb_queue_tail(&free_list
, skb
);
1196 dev_info(&np
->netdev
->dev
, "%s: %d xfer, %d noxfer, %d unused\n",
1197 __func__
, xfer
, noxfer
, unused
);
1200 if (!xen_feature(XENFEAT_auto_translated_physmap
)) {
1201 /* Do all the remapping work and M2P updates. */
1202 MULTI_mmu_update(mcl
, np
->rx_mmu
, mmu
- np
->rx_mmu
,
1205 HYPERVISOR_multicall(np
->rx_mcl
, mcl
- np
->rx_mcl
);
1209 __skb_queue_purge(&free_list
);
1211 spin_unlock_bh(&np
->rx_lock
);
1214 static void xennet_uninit(struct net_device
*dev
)
1216 struct netfront_info
*np
= netdev_priv(dev
);
1217 xennet_release_tx_bufs(np
);
1218 xennet_release_rx_bufs(np
);
1219 gnttab_free_grant_references(np
->gref_tx_head
);
1220 gnttab_free_grant_references(np
->gref_rx_head
);
1223 static netdev_features_t
xennet_fix_features(struct net_device
*dev
,
1224 netdev_features_t features
)
1226 struct netfront_info
*np
= netdev_priv(dev
);
1229 if (features
& NETIF_F_SG
) {
1230 if (xenbus_scanf(XBT_NIL
, np
->xbdev
->otherend
, "feature-sg",
1235 features
&= ~NETIF_F_SG
;
1238 if (features
& NETIF_F_TSO
) {
1239 if (xenbus_scanf(XBT_NIL
, np
->xbdev
->otherend
,
1240 "feature-gso-tcpv4", "%d", &val
) < 0)
1244 features
&= ~NETIF_F_TSO
;
1250 static int xennet_set_features(struct net_device
*dev
,
1251 netdev_features_t features
)
1253 if (!(features
& NETIF_F_SG
) && dev
->mtu
> ETH_DATA_LEN
) {
1254 netdev_info(dev
, "Reducing MTU because no SG offload");
1255 dev
->mtu
= ETH_DATA_LEN
;
1261 static irqreturn_t
xennet_tx_interrupt(int irq
, void *dev_id
)
1263 struct netfront_info
*np
= dev_id
;
1264 struct net_device
*dev
= np
->netdev
;
1265 unsigned long flags
;
1267 spin_lock_irqsave(&np
->tx_lock
, flags
);
1268 xennet_tx_buf_gc(dev
);
1269 spin_unlock_irqrestore(&np
->tx_lock
, flags
);
1274 static irqreturn_t
xennet_rx_interrupt(int irq
, void *dev_id
)
1276 struct netfront_info
*np
= dev_id
;
1277 struct net_device
*dev
= np
->netdev
;
1279 if (likely(netif_carrier_ok(dev
) &&
1280 RING_HAS_UNCONSUMED_RESPONSES(&np
->rx
)))
1281 napi_schedule(&np
->napi
);
1286 static irqreturn_t
xennet_interrupt(int irq
, void *dev_id
)
1288 xennet_tx_interrupt(irq
, dev_id
);
1289 xennet_rx_interrupt(irq
, dev_id
);
1293 #ifdef CONFIG_NET_POLL_CONTROLLER
1294 static void xennet_poll_controller(struct net_device
*dev
)
1296 xennet_interrupt(0, dev
);
1300 static const struct net_device_ops xennet_netdev_ops
= {
1301 .ndo_open
= xennet_open
,
1302 .ndo_uninit
= xennet_uninit
,
1303 .ndo_stop
= xennet_close
,
1304 .ndo_start_xmit
= xennet_start_xmit
,
1305 .ndo_change_mtu
= xennet_change_mtu
,
1306 .ndo_get_stats64
= xennet_get_stats64
,
1307 .ndo_set_mac_address
= eth_mac_addr
,
1308 .ndo_validate_addr
= eth_validate_addr
,
1309 .ndo_fix_features
= xennet_fix_features
,
1310 .ndo_set_features
= xennet_set_features
,
1311 #ifdef CONFIG_NET_POLL_CONTROLLER
1312 .ndo_poll_controller
= xennet_poll_controller
,
1316 static struct net_device
*xennet_create_dev(struct xenbus_device
*dev
)
1319 struct net_device
*netdev
;
1320 struct netfront_info
*np
;
1322 netdev
= alloc_etherdev(sizeof(struct netfront_info
));
1324 return ERR_PTR(-ENOMEM
);
1326 np
= netdev_priv(netdev
);
1329 spin_lock_init(&np
->tx_lock
);
1330 spin_lock_init(&np
->rx_lock
);
1332 skb_queue_head_init(&np
->rx_batch
);
1333 np
->rx_target
= RX_DFL_MIN_TARGET
;
1334 np
->rx_min_target
= RX_DFL_MIN_TARGET
;
1335 np
->rx_max_target
= RX_MAX_TARGET
;
1337 init_timer(&np
->rx_refill_timer
);
1338 np
->rx_refill_timer
.data
= (unsigned long)netdev
;
1339 np
->rx_refill_timer
.function
= rx_refill_timeout
;
1342 np
->stats
= alloc_percpu(struct netfront_stats
);
1343 if (np
->stats
== NULL
)
1346 /* Initialise tx_skbs as a free chain containing every entry. */
1347 np
->tx_skb_freelist
= 0;
1348 for (i
= 0; i
< NET_TX_RING_SIZE
; i
++) {
1349 skb_entry_set_link(&np
->tx_skbs
[i
], i
+1);
1350 np
->grant_tx_ref
[i
] = GRANT_INVALID_REF
;
1353 /* Clear out rx_skbs */
1354 for (i
= 0; i
< NET_RX_RING_SIZE
; i
++) {
1355 np
->rx_skbs
[i
] = NULL
;
1356 np
->grant_rx_ref
[i
] = GRANT_INVALID_REF
;
1359 /* A grant for every tx ring slot */
1360 if (gnttab_alloc_grant_references(TX_MAX_TARGET
,
1361 &np
->gref_tx_head
) < 0) {
1362 printk(KERN_ALERT
"#### netfront can't alloc tx grant refs\n");
1364 goto exit_free_stats
;
1366 /* A grant for every rx ring slot */
1367 if (gnttab_alloc_grant_references(RX_MAX_TARGET
,
1368 &np
->gref_rx_head
) < 0) {
1369 printk(KERN_ALERT
"#### netfront can't alloc rx grant refs\n");
1374 netdev
->netdev_ops
= &xennet_netdev_ops
;
1376 netif_napi_add(netdev
, &np
->napi
, xennet_poll
, 64);
1377 netdev
->features
= NETIF_F_IP_CSUM
| NETIF_F_RXCSUM
|
1379 netdev
->hw_features
= NETIF_F_IP_CSUM
| NETIF_F_SG
| NETIF_F_TSO
;
1382 * Assume that all hw features are available for now. This set
1383 * will be adjusted by the call to netdev_update_features() in
1384 * xennet_connect() which is the earliest point where we can
1385 * negotiate with the backend regarding supported features.
1387 netdev
->features
|= netdev
->hw_features
;
1389 SET_ETHTOOL_OPS(netdev
, &xennet_ethtool_ops
);
1390 SET_NETDEV_DEV(netdev
, &dev
->dev
);
1392 netif_set_gso_max_size(netdev
, XEN_NETIF_MAX_TX_SIZE
- MAX_TCP_HEADER
);
1394 np
->netdev
= netdev
;
1396 netif_carrier_off(netdev
);
1401 gnttab_free_grant_references(np
->gref_tx_head
);
1403 free_percpu(np
->stats
);
1405 free_netdev(netdev
);
1406 return ERR_PTR(err
);
1410 * Entry point to this code when a new device is created. Allocate the basic
1411 * structures and the ring buffers for communication with the backend, and
1412 * inform the backend of the appropriate details for those.
1414 static int netfront_probe(struct xenbus_device
*dev
,
1415 const struct xenbus_device_id
*id
)
1418 struct net_device
*netdev
;
1419 struct netfront_info
*info
;
1421 netdev
= xennet_create_dev(dev
);
1422 if (IS_ERR(netdev
)) {
1423 err
= PTR_ERR(netdev
);
1424 xenbus_dev_fatal(dev
, err
, "creating netdev");
1428 info
= netdev_priv(netdev
);
1429 dev_set_drvdata(&dev
->dev
, info
);
1431 err
= register_netdev(info
->netdev
);
1433 printk(KERN_WARNING
"%s: register_netdev err=%d\n",
1438 err
= xennet_sysfs_addif(info
->netdev
);
1440 unregister_netdev(info
->netdev
);
1441 printk(KERN_WARNING
"%s: add sysfs failed err=%d\n",
1449 free_netdev(netdev
);
1450 dev_set_drvdata(&dev
->dev
, NULL
);
1454 static void xennet_end_access(int ref
, void *page
)
1456 /* This frees the page as a side-effect */
1457 if (ref
!= GRANT_INVALID_REF
)
1458 gnttab_end_foreign_access(ref
, 0, (unsigned long)page
);
1461 static void xennet_disconnect_backend(struct netfront_info
*info
)
1463 /* Stop old i/f to prevent errors whilst we rebuild the state. */
1464 spin_lock_bh(&info
->rx_lock
);
1465 spin_lock_irq(&info
->tx_lock
);
1466 netif_carrier_off(info
->netdev
);
1467 spin_unlock_irq(&info
->tx_lock
);
1468 spin_unlock_bh(&info
->rx_lock
);
1470 if (info
->tx_irq
&& (info
->tx_irq
== info
->rx_irq
))
1471 unbind_from_irqhandler(info
->tx_irq
, info
);
1472 if (info
->tx_irq
&& (info
->tx_irq
!= info
->rx_irq
)) {
1473 unbind_from_irqhandler(info
->tx_irq
, info
);
1474 unbind_from_irqhandler(info
->rx_irq
, info
);
1476 info
->tx_evtchn
= info
->rx_evtchn
= 0;
1477 info
->tx_irq
= info
->rx_irq
= 0;
1479 /* End access and free the pages */
1480 xennet_end_access(info
->tx_ring_ref
, info
->tx
.sring
);
1481 xennet_end_access(info
->rx_ring_ref
, info
->rx
.sring
);
1483 info
->tx_ring_ref
= GRANT_INVALID_REF
;
1484 info
->rx_ring_ref
= GRANT_INVALID_REF
;
1485 info
->tx
.sring
= NULL
;
1486 info
->rx
.sring
= NULL
;
1490 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1491 * driver restart. We tear down our netif structure and recreate it, but
1492 * leave the device-layer structures intact so that this is transparent to the
1493 * rest of the kernel.
1495 static int netfront_resume(struct xenbus_device
*dev
)
1497 struct netfront_info
*info
= dev_get_drvdata(&dev
->dev
);
1499 dev_dbg(&dev
->dev
, "%s\n", dev
->nodename
);
1501 xennet_disconnect_backend(info
);
1505 static int xen_net_read_mac(struct xenbus_device
*dev
, u8 mac
[])
1507 char *s
, *e
, *macstr
;
1510 macstr
= s
= xenbus_read(XBT_NIL
, dev
->nodename
, "mac", NULL
);
1512 return PTR_ERR(macstr
);
1514 for (i
= 0; i
< ETH_ALEN
; i
++) {
1515 mac
[i
] = simple_strtoul(s
, &e
, 16);
1516 if ((s
== e
) || (*e
!= ((i
== ETH_ALEN
-1) ? '\0' : ':'))) {
1527 static int setup_netfront_single(struct netfront_info
*info
)
1531 err
= xenbus_alloc_evtchn(info
->xbdev
, &info
->tx_evtchn
);
1535 err
= bind_evtchn_to_irqhandler(info
->tx_evtchn
,
1537 0, info
->netdev
->name
, info
);
1540 info
->rx_evtchn
= info
->tx_evtchn
;
1541 info
->rx_irq
= info
->tx_irq
= err
;
1546 xenbus_free_evtchn(info
->xbdev
, info
->tx_evtchn
);
1547 info
->tx_evtchn
= 0;
1552 static int setup_netfront_split(struct netfront_info
*info
)
1556 err
= xenbus_alloc_evtchn(info
->xbdev
, &info
->tx_evtchn
);
1559 err
= xenbus_alloc_evtchn(info
->xbdev
, &info
->rx_evtchn
);
1561 goto alloc_rx_evtchn_fail
;
1563 snprintf(info
->tx_irq_name
, sizeof(info
->tx_irq_name
),
1564 "%s-tx", info
->netdev
->name
);
1565 err
= bind_evtchn_to_irqhandler(info
->tx_evtchn
,
1566 xennet_tx_interrupt
,
1567 0, info
->tx_irq_name
, info
);
1572 snprintf(info
->rx_irq_name
, sizeof(info
->rx_irq_name
),
1573 "%s-rx", info
->netdev
->name
);
1574 err
= bind_evtchn_to_irqhandler(info
->rx_evtchn
,
1575 xennet_rx_interrupt
,
1576 0, info
->rx_irq_name
, info
);
1584 unbind_from_irqhandler(info
->tx_irq
, info
);
1587 xenbus_free_evtchn(info
->xbdev
, info
->rx_evtchn
);
1588 info
->rx_evtchn
= 0;
1589 alloc_rx_evtchn_fail
:
1590 xenbus_free_evtchn(info
->xbdev
, info
->tx_evtchn
);
1591 info
->tx_evtchn
= 0;
1596 static int setup_netfront(struct xenbus_device
*dev
, struct netfront_info
*info
)
1598 struct xen_netif_tx_sring
*txs
;
1599 struct xen_netif_rx_sring
*rxs
;
1601 struct net_device
*netdev
= info
->netdev
;
1602 unsigned int feature_split_evtchn
;
1604 info
->tx_ring_ref
= GRANT_INVALID_REF
;
1605 info
->rx_ring_ref
= GRANT_INVALID_REF
;
1606 info
->rx
.sring
= NULL
;
1607 info
->tx
.sring
= NULL
;
1610 err
= xenbus_scanf(XBT_NIL
, info
->xbdev
->otherend
,
1611 "feature-split-event-channels", "%u",
1612 &feature_split_evtchn
);
1614 feature_split_evtchn
= 0;
1616 err
= xen_net_read_mac(dev
, netdev
->dev_addr
);
1618 xenbus_dev_fatal(dev
, err
, "parsing %s/mac", dev
->nodename
);
1622 txs
= (struct xen_netif_tx_sring
*)get_zeroed_page(GFP_NOIO
| __GFP_HIGH
);
1625 xenbus_dev_fatal(dev
, err
, "allocating tx ring page");
1628 SHARED_RING_INIT(txs
);
1629 FRONT_RING_INIT(&info
->tx
, txs
, PAGE_SIZE
);
1631 err
= xenbus_grant_ring(dev
, virt_to_mfn(txs
));
1633 goto grant_tx_ring_fail
;
1635 info
->tx_ring_ref
= err
;
1636 rxs
= (struct xen_netif_rx_sring
*)get_zeroed_page(GFP_NOIO
| __GFP_HIGH
);
1639 xenbus_dev_fatal(dev
, err
, "allocating rx ring page");
1640 goto alloc_rx_ring_fail
;
1642 SHARED_RING_INIT(rxs
);
1643 FRONT_RING_INIT(&info
->rx
, rxs
, PAGE_SIZE
);
1645 err
= xenbus_grant_ring(dev
, virt_to_mfn(rxs
));
1647 goto grant_rx_ring_fail
;
1648 info
->rx_ring_ref
= err
;
1650 if (feature_split_evtchn
)
1651 err
= setup_netfront_split(info
);
1652 /* setup single event channel if
1653 * a) feature-split-event-channels == 0
1654 * b) feature-split-event-channels == 1 but failed to setup
1656 if (!feature_split_evtchn
|| (feature_split_evtchn
&& err
))
1657 err
= setup_netfront_single(info
);
1660 goto alloc_evtchn_fail
;
1664 /* If we fail to setup netfront, it is safe to just revoke access to
1665 * granted pages because backend is not accessing it at this point.
1668 gnttab_end_foreign_access_ref(info
->rx_ring_ref
, 0);
1670 free_page((unsigned long)rxs
);
1672 gnttab_end_foreign_access_ref(info
->tx_ring_ref
, 0);
1674 free_page((unsigned long)txs
);
1679 /* Common code used when first setting up, and when resuming. */
1680 static int talk_to_netback(struct xenbus_device
*dev
,
1681 struct netfront_info
*info
)
1683 const char *message
;
1684 struct xenbus_transaction xbt
;
1687 /* Create shared ring, alloc event channel. */
1688 err
= setup_netfront(dev
, info
);
1693 err
= xenbus_transaction_start(&xbt
);
1695 xenbus_dev_fatal(dev
, err
, "starting transaction");
1699 err
= xenbus_printf(xbt
, dev
->nodename
, "tx-ring-ref", "%u",
1702 message
= "writing tx ring-ref";
1703 goto abort_transaction
;
1705 err
= xenbus_printf(xbt
, dev
->nodename
, "rx-ring-ref", "%u",
1708 message
= "writing rx ring-ref";
1709 goto abort_transaction
;
1712 if (info
->tx_evtchn
== info
->rx_evtchn
) {
1713 err
= xenbus_printf(xbt
, dev
->nodename
,
1714 "event-channel", "%u", info
->tx_evtchn
);
1716 message
= "writing event-channel";
1717 goto abort_transaction
;
1720 err
= xenbus_printf(xbt
, dev
->nodename
,
1721 "event-channel-tx", "%u", info
->tx_evtchn
);
1723 message
= "writing event-channel-tx";
1724 goto abort_transaction
;
1726 err
= xenbus_printf(xbt
, dev
->nodename
,
1727 "event-channel-rx", "%u", info
->rx_evtchn
);
1729 message
= "writing event-channel-rx";
1730 goto abort_transaction
;
1734 err
= xenbus_printf(xbt
, dev
->nodename
, "request-rx-copy", "%u",
1737 message
= "writing request-rx-copy";
1738 goto abort_transaction
;
1741 err
= xenbus_printf(xbt
, dev
->nodename
, "feature-rx-notify", "%d", 1);
1743 message
= "writing feature-rx-notify";
1744 goto abort_transaction
;
1747 err
= xenbus_printf(xbt
, dev
->nodename
, "feature-sg", "%d", 1);
1749 message
= "writing feature-sg";
1750 goto abort_transaction
;
1753 err
= xenbus_printf(xbt
, dev
->nodename
, "feature-gso-tcpv4", "%d", 1);
1755 message
= "writing feature-gso-tcpv4";
1756 goto abort_transaction
;
1759 err
= xenbus_transaction_end(xbt
, 0);
1763 xenbus_dev_fatal(dev
, err
, "completing transaction");
1770 xenbus_transaction_end(xbt
, 1);
1771 xenbus_dev_fatal(dev
, err
, "%s", message
);
1773 xennet_disconnect_backend(info
);
1778 static int xennet_connect(struct net_device
*dev
)
1780 struct netfront_info
*np
= netdev_priv(dev
);
1781 int i
, requeue_idx
, err
;
1782 struct sk_buff
*skb
;
1784 struct xen_netif_rx_request
*req
;
1785 unsigned int feature_rx_copy
;
1787 err
= xenbus_scanf(XBT_NIL
, np
->xbdev
->otherend
,
1788 "feature-rx-copy", "%u", &feature_rx_copy
);
1790 feature_rx_copy
= 0;
1792 if (!feature_rx_copy
) {
1794 "backend does not support copying receive path\n");
1798 err
= talk_to_netback(np
->xbdev
, np
);
1803 netdev_update_features(dev
);
1806 spin_lock_bh(&np
->rx_lock
);
1807 spin_lock_irq(&np
->tx_lock
);
1809 /* Step 1: Discard all pending TX packet fragments. */
1810 xennet_release_tx_bufs(np
);
1812 /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1813 for (requeue_idx
= 0, i
= 0; i
< NET_RX_RING_SIZE
; i
++) {
1815 const struct page
*page
;
1816 if (!np
->rx_skbs
[i
])
1819 skb
= np
->rx_skbs
[requeue_idx
] = xennet_get_rx_skb(np
, i
);
1820 ref
= np
->grant_rx_ref
[requeue_idx
] = xennet_get_rx_ref(np
, i
);
1821 req
= RING_GET_REQUEST(&np
->rx
, requeue_idx
);
1823 frag
= &skb_shinfo(skb
)->frags
[0];
1824 page
= skb_frag_page(frag
);
1825 gnttab_grant_foreign_access_ref(
1826 ref
, np
->xbdev
->otherend_id
,
1827 pfn_to_mfn(page_to_pfn(page
)),
1830 req
->id
= requeue_idx
;
1835 np
->rx
.req_prod_pvt
= requeue_idx
;
1838 * Step 3: All public and private state should now be sane. Get
1839 * ready to start sending and receiving packets and give the driver
1840 * domain a kick because we've probably just requeued some
1843 netif_carrier_on(np
->netdev
);
1844 notify_remote_via_irq(np
->tx_irq
);
1845 if (np
->tx_irq
!= np
->rx_irq
)
1846 notify_remote_via_irq(np
->rx_irq
);
1847 xennet_tx_buf_gc(dev
);
1848 xennet_alloc_rx_buffers(dev
);
1850 spin_unlock_irq(&np
->tx_lock
);
1851 spin_unlock_bh(&np
->rx_lock
);
1857 * Callback received when the backend's state changes.
1859 static void netback_changed(struct xenbus_device
*dev
,
1860 enum xenbus_state backend_state
)
1862 struct netfront_info
*np
= dev_get_drvdata(&dev
->dev
);
1863 struct net_device
*netdev
= np
->netdev
;
1865 dev_dbg(&dev
->dev
, "%s\n", xenbus_strstate(backend_state
));
1867 switch (backend_state
) {
1868 case XenbusStateInitialising
:
1869 case XenbusStateInitialised
:
1870 case XenbusStateReconfiguring
:
1871 case XenbusStateReconfigured
:
1872 case XenbusStateUnknown
:
1873 case XenbusStateClosed
:
1876 case XenbusStateInitWait
:
1877 if (dev
->state
!= XenbusStateInitialising
)
1879 if (xennet_connect(netdev
) != 0)
1881 xenbus_switch_state(dev
, XenbusStateConnected
);
1884 case XenbusStateConnected
:
1885 netdev_notify_peers(netdev
);
1888 case XenbusStateClosing
:
1889 xenbus_frontend_closed(dev
);
1894 static const struct xennet_stat
{
1895 char name
[ETH_GSTRING_LEN
];
1897 } xennet_stats
[] = {
1899 "rx_gso_checksum_fixup",
1900 offsetof(struct netfront_info
, rx_gso_checksum_fixup
)
1904 static int xennet_get_sset_count(struct net_device
*dev
, int string_set
)
1906 switch (string_set
) {
1908 return ARRAY_SIZE(xennet_stats
);
1914 static void xennet_get_ethtool_stats(struct net_device
*dev
,
1915 struct ethtool_stats
*stats
, u64
* data
)
1917 void *np
= netdev_priv(dev
);
1920 for (i
= 0; i
< ARRAY_SIZE(xennet_stats
); i
++)
1921 data
[i
] = *(unsigned long *)(np
+ xennet_stats
[i
].offset
);
1924 static void xennet_get_strings(struct net_device
*dev
, u32 stringset
, u8
* data
)
1928 switch (stringset
) {
1930 for (i
= 0; i
< ARRAY_SIZE(xennet_stats
); i
++)
1931 memcpy(data
+ i
* ETH_GSTRING_LEN
,
1932 xennet_stats
[i
].name
, ETH_GSTRING_LEN
);
1937 static const struct ethtool_ops xennet_ethtool_ops
=
1939 .get_link
= ethtool_op_get_link
,
1941 .get_sset_count
= xennet_get_sset_count
,
1942 .get_ethtool_stats
= xennet_get_ethtool_stats
,
1943 .get_strings
= xennet_get_strings
,
1947 static ssize_t
show_rxbuf_min(struct device
*dev
,
1948 struct device_attribute
*attr
, char *buf
)
1950 struct net_device
*netdev
= to_net_dev(dev
);
1951 struct netfront_info
*info
= netdev_priv(netdev
);
1953 return sprintf(buf
, "%u\n", info
->rx_min_target
);
1956 static ssize_t
store_rxbuf_min(struct device
*dev
,
1957 struct device_attribute
*attr
,
1958 const char *buf
, size_t len
)
1960 struct net_device
*netdev
= to_net_dev(dev
);
1961 struct netfront_info
*np
= netdev_priv(netdev
);
1963 unsigned long target
;
1965 if (!capable(CAP_NET_ADMIN
))
1968 target
= simple_strtoul(buf
, &endp
, 0);
1972 if (target
< RX_MIN_TARGET
)
1973 target
= RX_MIN_TARGET
;
1974 if (target
> RX_MAX_TARGET
)
1975 target
= RX_MAX_TARGET
;
1977 spin_lock_bh(&np
->rx_lock
);
1978 if (target
> np
->rx_max_target
)
1979 np
->rx_max_target
= target
;
1980 np
->rx_min_target
= target
;
1981 if (target
> np
->rx_target
)
1982 np
->rx_target
= target
;
1984 xennet_alloc_rx_buffers(netdev
);
1986 spin_unlock_bh(&np
->rx_lock
);
1990 static ssize_t
show_rxbuf_max(struct device
*dev
,
1991 struct device_attribute
*attr
, char *buf
)
1993 struct net_device
*netdev
= to_net_dev(dev
);
1994 struct netfront_info
*info
= netdev_priv(netdev
);
1996 return sprintf(buf
, "%u\n", info
->rx_max_target
);
1999 static ssize_t
store_rxbuf_max(struct device
*dev
,
2000 struct device_attribute
*attr
,
2001 const char *buf
, size_t len
)
2003 struct net_device
*netdev
= to_net_dev(dev
);
2004 struct netfront_info
*np
= netdev_priv(netdev
);
2006 unsigned long target
;
2008 if (!capable(CAP_NET_ADMIN
))
2011 target
= simple_strtoul(buf
, &endp
, 0);
2015 if (target
< RX_MIN_TARGET
)
2016 target
= RX_MIN_TARGET
;
2017 if (target
> RX_MAX_TARGET
)
2018 target
= RX_MAX_TARGET
;
2020 spin_lock_bh(&np
->rx_lock
);
2021 if (target
< np
->rx_min_target
)
2022 np
->rx_min_target
= target
;
2023 np
->rx_max_target
= target
;
2024 if (target
< np
->rx_target
)
2025 np
->rx_target
= target
;
2027 xennet_alloc_rx_buffers(netdev
);
2029 spin_unlock_bh(&np
->rx_lock
);
2033 static ssize_t
show_rxbuf_cur(struct device
*dev
,
2034 struct device_attribute
*attr
, char *buf
)
2036 struct net_device
*netdev
= to_net_dev(dev
);
2037 struct netfront_info
*info
= netdev_priv(netdev
);
2039 return sprintf(buf
, "%u\n", info
->rx_target
);
2042 static struct device_attribute xennet_attrs
[] = {
2043 __ATTR(rxbuf_min
, S_IRUGO
|S_IWUSR
, show_rxbuf_min
, store_rxbuf_min
),
2044 __ATTR(rxbuf_max
, S_IRUGO
|S_IWUSR
, show_rxbuf_max
, store_rxbuf_max
),
2045 __ATTR(rxbuf_cur
, S_IRUGO
, show_rxbuf_cur
, NULL
),
2048 static int xennet_sysfs_addif(struct net_device
*netdev
)
2053 for (i
= 0; i
< ARRAY_SIZE(xennet_attrs
); i
++) {
2054 err
= device_create_file(&netdev
->dev
,
2063 device_remove_file(&netdev
->dev
, &xennet_attrs
[i
]);
2067 static void xennet_sysfs_delif(struct net_device
*netdev
)
2071 for (i
= 0; i
< ARRAY_SIZE(xennet_attrs
); i
++)
2072 device_remove_file(&netdev
->dev
, &xennet_attrs
[i
]);
2075 #endif /* CONFIG_SYSFS */
2077 static const struct xenbus_device_id netfront_ids
[] = {
2083 static int xennet_remove(struct xenbus_device
*dev
)
2085 struct netfront_info
*info
= dev_get_drvdata(&dev
->dev
);
2087 dev_dbg(&dev
->dev
, "%s\n", dev
->nodename
);
2089 xennet_disconnect_backend(info
);
2091 xennet_sysfs_delif(info
->netdev
);
2093 unregister_netdev(info
->netdev
);
2095 del_timer_sync(&info
->rx_refill_timer
);
2097 free_percpu(info
->stats
);
2099 free_netdev(info
->netdev
);
2104 static DEFINE_XENBUS_DRIVER(netfront
, ,
2105 .probe
= netfront_probe
,
2106 .remove
= xennet_remove
,
2107 .resume
= netfront_resume
,
2108 .otherend_changed
= netback_changed
,
2111 static int __init
netif_init(void)
2116 if (xen_hvm_domain() && !xen_platform_pci_unplug
)
2119 printk(KERN_INFO
"Initialising Xen virtual ethernet driver.\n");
2121 return xenbus_register_frontend(&netfront_driver
);
2123 module_init(netif_init
);
2126 static void __exit
netif_exit(void)
2128 xenbus_unregister_driver(&netfront_driver
);
2130 module_exit(netif_exit
);
2132 MODULE_DESCRIPTION("Xen virtual network device frontend");
2133 MODULE_LICENSE("GPL");
2134 MODULE_ALIAS("xen:vif");
2135 MODULE_ALIAS("xennet");