aboutsummaryrefslogtreecommitdiffstats
path: root/drivers/net
diff options
context:
space:
mode:
authorJeremy Fitzhardinge <jeremy@xensource.com>2007-07-17 21:37:06 -0400
committerJeremy Fitzhardinge <jeremy@goop.org>2007-07-18 11:47:45 -0400
commit0d160211965b79de989cf2d170985abeb8da5ec6 (patch)
tree968bc01a0e7fe75b6e274d8958b04fc1d9afa3a7 /drivers/net
parent9f27ee595038653ddf8bca871200d39247d6f4fc (diff)
xen: add virtual network device driver
The network device frontend driver allows the kernel to access network devices exported exported by a virtual machine containing a physical network device driver. Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com> Signed-off-by: Chris Wright <chrisw@sous-sol.org> Acked-by: Jeff Garzik <jeff@garzik.org> Cc: Ian Pratt <ian.pratt@xensource.com> Cc: Christian Limpach <Christian.Limpach@cl.cam.ac.uk> Cc: Stephen Hemminger <shemminger@linux-foundation.org> Cc: Christoph Hellwig <hch@infradead.org> Cc: Rusty Russell <rusty@rustcorp.com.au> Cc: Herbert Xu <herbert@gondor.apana.org.au> Cc: Keir Fraser <Keir.Fraser@cl.cam.ac.uk> Cc: netdev@vger.kernel.org
Diffstat (limited to 'drivers/net')
-rw-r--r--drivers/net/Kconfig12
-rw-r--r--drivers/net/Makefile2
-rw-r--r--drivers/net/xen-netfront.c1863
3 files changed, 1877 insertions, 0 deletions
diff --git a/drivers/net/Kconfig b/drivers/net/Kconfig
index 43d03178064d..5fb659f8b20e 100644
--- a/drivers/net/Kconfig
+++ b/drivers/net/Kconfig
@@ -2486,6 +2486,18 @@ source "drivers/atm/Kconfig"
2486 2486
2487source "drivers/s390/net/Kconfig" 2487source "drivers/s390/net/Kconfig"
2488 2488
2489config XEN_NETDEV_FRONTEND
2490 tristate "Xen network device frontend driver"
2491 depends on XEN
2492 default y
2493 help
2494 The network device frontend driver allows the kernel to
2495 access network devices exported exported by a virtual
2496 machine containing a physical network device driver. The
2497 frontend driver is intended for unprivileged guest domains;
2498 if you are compiling a kernel for a Xen guest, you almost
2499 certainly want to enable this.
2500
2489config ISERIES_VETH 2501config ISERIES_VETH
2490 tristate "iSeries Virtual Ethernet driver support" 2502 tristate "iSeries Virtual Ethernet driver support"
2491 depends on PPC_ISERIES 2503 depends on PPC_ISERIES
diff --git a/drivers/net/Makefile b/drivers/net/Makefile
index eb4167622a6a..0e286ab8855a 100644
--- a/drivers/net/Makefile
+++ b/drivers/net/Makefile
@@ -127,6 +127,8 @@ obj-$(CONFIG_PPPOL2TP) += pppox.o pppol2tp.o
127obj-$(CONFIG_SLIP) += slip.o 127obj-$(CONFIG_SLIP) += slip.o
128obj-$(CONFIG_SLHC) += slhc.o 128obj-$(CONFIG_SLHC) += slhc.o
129 129
130obj-$(CONFIG_XEN_NETDEV_FRONTEND) += xen-netfront.o
131
130obj-$(CONFIG_DUMMY) += dummy.o 132obj-$(CONFIG_DUMMY) += dummy.o
131obj-$(CONFIG_IFB) += ifb.o 133obj-$(CONFIG_IFB) += ifb.o
132obj-$(CONFIG_MACVLAN) += macvlan.o 134obj-$(CONFIG_MACVLAN) += macvlan.o
diff --git a/drivers/net/xen-netfront.c b/drivers/net/xen-netfront.c
new file mode 100644
index 000000000000..489f69c5d6ca
--- /dev/null
+++ b/drivers/net/xen-netfront.c
@@ -0,0 +1,1863 @@
1/*
2 * Virtual network driver for conversing with remote driver backends.
3 *
4 * Copyright (c) 2002-2005, K A Fraser
5 * Copyright (c) 2005, XenSource Ltd
6 *
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:
12 *
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:
19 *
20 * The above copyright notice and this permission notice shall be included in
21 * all copies or substantial portions of the Software.
22 *
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
29 * IN THE SOFTWARE.
30 */
31
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>
39#include <linux/tcp.h>
40#include <linux/udp.h>
41#include <linux/moduleparam.h>
42#include <linux/mm.h>
43#include <net/ip.h>
44
45#include <xen/xenbus.h>
46#include <xen/events.h>
47#include <xen/page.h>
48#include <xen/grant_table.h>
49
50#include <xen/interface/io/netif.h>
51#include <xen/interface/memory.h>
52#include <xen/interface/grant_table.h>
53
54static struct ethtool_ops xennet_ethtool_ops;
55
56struct netfront_cb {
57 struct page *page;
58 unsigned offset;
59};
60
61#define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb))
62
63#define RX_COPY_THRESHOLD 256
64
65#define GRANT_INVALID_REF 0
66
67#define NET_TX_RING_SIZE __RING_SIZE((struct xen_netif_tx_sring *)0, PAGE_SIZE)
68#define NET_RX_RING_SIZE __RING_SIZE((struct xen_netif_rx_sring *)0, PAGE_SIZE)
69#define TX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256)
70
71struct netfront_info {
72 struct list_head list;
73 struct net_device *netdev;
74
75 struct net_device_stats stats;
76
77 struct xen_netif_tx_front_ring tx;
78 struct xen_netif_rx_front_ring rx;
79
80 spinlock_t tx_lock;
81 spinlock_t rx_lock;
82
83 unsigned int evtchn;
84
85 /* Receive-ring batched refills. */
86#define RX_MIN_TARGET 8
87#define RX_DFL_MIN_TARGET 64
88#define RX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256)
89 unsigned rx_min_target, rx_max_target, rx_target;
90 struct sk_buff_head rx_batch;
91
92 struct timer_list rx_refill_timer;
93
94 /*
95 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
96 * are linked from tx_skb_freelist through skb_entry.link.
97 *
98 * NB. Freelist index entries are always going to be less than
99 * PAGE_OFFSET, whereas pointers to skbs will always be equal or
100 * greater than PAGE_OFFSET: we use this property to distinguish
101 * them.
102 */
103 union skb_entry {
104 struct sk_buff *skb;
105 unsigned link;
106 } tx_skbs[NET_TX_RING_SIZE];
107 grant_ref_t gref_tx_head;
108 grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
109 unsigned tx_skb_freelist;
110
111 struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
112 grant_ref_t gref_rx_head;
113 grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
114
115 struct xenbus_device *xbdev;
116 int tx_ring_ref;
117 int rx_ring_ref;
118
119 unsigned long rx_pfn_array[NET_RX_RING_SIZE];
120 struct multicall_entry rx_mcl[NET_RX_RING_SIZE+1];
121 struct mmu_update rx_mmu[NET_RX_RING_SIZE];
122};
123
124struct netfront_rx_info {
125 struct xen_netif_rx_response rx;
126 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
127};
128
129/*
130 * Access macros for acquiring freeing slots in tx_skbs[].
131 */
132
133static void add_id_to_freelist(unsigned *head, union skb_entry *list,
134 unsigned short id)
135{
136 list[id].link = *head;
137 *head = id;
138}
139
140static unsigned short get_id_from_freelist(unsigned *head,
141 union skb_entry *list)
142{
143 unsigned int id = *head;
144 *head = list[id].link;
145 return id;
146}
147
148static int xennet_rxidx(RING_IDX idx)
149{
150 return idx & (NET_RX_RING_SIZE - 1);
151}
152
153static struct sk_buff *xennet_get_rx_skb(struct netfront_info *np,
154 RING_IDX ri)
155{
156 int i = xennet_rxidx(ri);
157 struct sk_buff *skb = np->rx_skbs[i];
158 np->rx_skbs[i] = NULL;
159 return skb;
160}
161
162static grant_ref_t xennet_get_rx_ref(struct netfront_info *np,
163 RING_IDX ri)
164{
165 int i = xennet_rxidx(ri);
166 grant_ref_t ref = np->grant_rx_ref[i];
167 np->grant_rx_ref[i] = GRANT_INVALID_REF;
168 return ref;
169}
170
171#ifdef CONFIG_SYSFS
172static int xennet_sysfs_addif(struct net_device *netdev);
173static void xennet_sysfs_delif(struct net_device *netdev);
174#else /* !CONFIG_SYSFS */
175#define xennet_sysfs_addif(dev) (0)
176#define xennet_sysfs_delif(dev) do { } while (0)
177#endif
178
179static int xennet_can_sg(struct net_device *dev)
180{
181 return dev->features & NETIF_F_SG;
182}
183
184
185static void rx_refill_timeout(unsigned long data)
186{
187 struct net_device *dev = (struct net_device *)data;
188 netif_rx_schedule(dev);
189}
190
191static int netfront_tx_slot_available(struct netfront_info *np)
192{
193 return ((np->tx.req_prod_pvt - np->tx.rsp_cons) <
194 (TX_MAX_TARGET - MAX_SKB_FRAGS - 2));
195}
196
197static void xennet_maybe_wake_tx(struct net_device *dev)
198{
199 struct netfront_info *np = netdev_priv(dev);
200
201 if (unlikely(netif_queue_stopped(dev)) &&
202 netfront_tx_slot_available(np) &&
203 likely(netif_running(dev)))
204 netif_wake_queue(dev);
205}
206
207static void xennet_alloc_rx_buffers(struct net_device *dev)
208{
209 unsigned short id;
210 struct netfront_info *np = netdev_priv(dev);
211 struct sk_buff *skb;
212 struct page *page;
213 int i, batch_target, notify;
214 RING_IDX req_prod = np->rx.req_prod_pvt;
215 struct xen_memory_reservation reservation;
216 grant_ref_t ref;
217 unsigned long pfn;
218 void *vaddr;
219 int nr_flips;
220 struct xen_netif_rx_request *req;
221
222 if (unlikely(!netif_carrier_ok(dev)))
223 return;
224
225 /*
226 * Allocate skbuffs greedily, even though we batch updates to the
227 * receive ring. This creates a less bursty demand on the memory
228 * allocator, so should reduce the chance of failed allocation requests
229 * both for ourself and for other kernel subsystems.
230 */
231 batch_target = np->rx_target - (req_prod - np->rx.rsp_cons);
232 for (i = skb_queue_len(&np->rx_batch); i < batch_target; i++) {
233 skb = __netdev_alloc_skb(dev, RX_COPY_THRESHOLD,
234 GFP_ATOMIC | __GFP_NOWARN);
235 if (unlikely(!skb))
236 goto no_skb;
237
238 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
239 if (!page) {
240 kfree_skb(skb);
241no_skb:
242 /* Any skbuffs queued for refill? Force them out. */
243 if (i != 0)
244 goto refill;
245 /* Could not allocate any skbuffs. Try again later. */
246 mod_timer(&np->rx_refill_timer,
247 jiffies + (HZ/10));
248 break;
249 }
250
251 skb_shinfo(skb)->frags[0].page = page;
252 skb_shinfo(skb)->nr_frags = 1;
253 __skb_queue_tail(&np->rx_batch, skb);
254 }
255
256 /* Is the batch large enough to be worthwhile? */
257 if (i < (np->rx_target/2)) {
258 if (req_prod > np->rx.sring->req_prod)
259 goto push;
260 return;
261 }
262
263 /* Adjust our fill target if we risked running out of buffers. */
264 if (((req_prod - np->rx.sring->rsp_prod) < (np->rx_target / 4)) &&
265 ((np->rx_target *= 2) > np->rx_max_target))
266 np->rx_target = np->rx_max_target;
267
268 refill:
269 for (nr_flips = i = 0; ; i++) {
270 skb = __skb_dequeue(&np->rx_batch);
271 if (skb == NULL)
272 break;
273
274 skb->dev = dev;
275
276 id = xennet_rxidx(req_prod + i);
277
278 BUG_ON(np->rx_skbs[id]);
279 np->rx_skbs[id] = skb;
280
281 ref = gnttab_claim_grant_reference(&np->gref_rx_head);
282 BUG_ON((signed short)ref < 0);
283 np->grant_rx_ref[id] = ref;
284
285 pfn = page_to_pfn(skb_shinfo(skb)->frags[0].page);
286 vaddr = page_address(skb_shinfo(skb)->frags[0].page);
287
288 req = RING_GET_REQUEST(&np->rx, req_prod + i);
289 gnttab_grant_foreign_access_ref(ref,
290 np->xbdev->otherend_id,
291 pfn_to_mfn(pfn),
292 0);
293
294 req->id = id;
295 req->gref = ref;
296 }
297
298 if (nr_flips != 0) {
299 reservation.extent_start = np->rx_pfn_array;
300 reservation.nr_extents = nr_flips;
301 reservation.extent_order = 0;
302 reservation.address_bits = 0;
303 reservation.domid = DOMID_SELF;
304
305 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
306 /* After all PTEs have been zapped, flush the TLB. */
307 np->rx_mcl[i-1].args[MULTI_UVMFLAGS_INDEX] =
308 UVMF_TLB_FLUSH|UVMF_ALL;
309
310 /* Give away a batch of pages. */
311 np->rx_mcl[i].op = __HYPERVISOR_memory_op;
312 np->rx_mcl[i].args[0] = XENMEM_decrease_reservation;
313 np->rx_mcl[i].args[1] = (unsigned long)&reservation;
314
315 /* Zap PTEs and give away pages in one big
316 * multicall. */
317 (void)HYPERVISOR_multicall(np->rx_mcl, i+1);
318
319 /* Check return status of HYPERVISOR_memory_op(). */
320 if (unlikely(np->rx_mcl[i].result != i))
321 panic("Unable to reduce memory reservation\n");
322 } else {
323 if (HYPERVISOR_memory_op(XENMEM_decrease_reservation,
324 &reservation) != i)
325 panic("Unable to reduce memory reservation\n");
326 }
327 } else {
328 wmb(); /* barrier so backend seens requests */
329 }
330
331 /* Above is a suitable barrier to ensure backend will see requests. */
332 np->rx.req_prod_pvt = req_prod + i;
333 push:
334 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->rx, notify);
335 if (notify)
336 notify_remote_via_irq(np->netdev->irq);
337}
338
339static int xennet_open(struct net_device *dev)
340{
341 struct netfront_info *np = netdev_priv(dev);
342
343 memset(&np->stats, 0, sizeof(np->stats));
344
345 spin_lock_bh(&np->rx_lock);
346 if (netif_carrier_ok(dev)) {
347 xennet_alloc_rx_buffers(dev);
348 np->rx.sring->rsp_event = np->rx.rsp_cons + 1;
349 if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx))
350 netif_rx_schedule(dev);
351 }
352 spin_unlock_bh(&np->rx_lock);
353
354 xennet_maybe_wake_tx(dev);
355
356 return 0;
357}
358
359static void xennet_tx_buf_gc(struct net_device *dev)
360{
361 RING_IDX cons, prod;
362 unsigned short id;
363 struct netfront_info *np = netdev_priv(dev);
364 struct sk_buff *skb;
365
366 BUG_ON(!netif_carrier_ok(dev));
367
368 do {
369 prod = np->tx.sring->rsp_prod;
370 rmb(); /* Ensure we see responses up to 'rp'. */
371
372 for (cons = np->tx.rsp_cons; cons != prod; cons++) {
373 struct xen_netif_tx_response *txrsp;
374
375 txrsp = RING_GET_RESPONSE(&np->tx, cons);
376 if (txrsp->status == NETIF_RSP_NULL)
377 continue;
378
379 id = txrsp->id;
380 skb = np->tx_skbs[id].skb;
381 if (unlikely(gnttab_query_foreign_access(
382 np->grant_tx_ref[id]) != 0)) {
383 printk(KERN_ALERT "xennet_tx_buf_gc: warning "
384 "-- grant still in use by backend "
385 "domain.\n");
386 BUG();
387 }
388 gnttab_end_foreign_access_ref(
389 np->grant_tx_ref[id], GNTMAP_readonly);
390 gnttab_release_grant_reference(
391 &np->gref_tx_head, np->grant_tx_ref[id]);
392 np->grant_tx_ref[id] = GRANT_INVALID_REF;
393 add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, id);
394 dev_kfree_skb_irq(skb);
395 }
396
397 np->tx.rsp_cons = prod;
398
399 /*
400 * Set a new event, then check for race with update of tx_cons.
401 * Note that it is essential to schedule a callback, no matter
402 * how few buffers are pending. Even if there is space in the
403 * transmit ring, higher layers may be blocked because too much
404 * data is outstanding: in such cases notification from Xen is
405 * likely to be the only kick that we'll get.
406 */
407 np->tx.sring->rsp_event =
408 prod + ((np->tx.sring->req_prod - prod) >> 1) + 1;
409 mb(); /* update shared area */
410 } while ((cons == prod) && (prod != np->tx.sring->rsp_prod));
411
412 xennet_maybe_wake_tx(dev);
413}
414
415static void xennet_make_frags(struct sk_buff *skb, struct net_device *dev,
416 struct xen_netif_tx_request *tx)
417{
418 struct netfront_info *np = netdev_priv(dev);
419 char *data = skb->data;
420 unsigned long mfn;
421 RING_IDX prod = np->tx.req_prod_pvt;
422 int frags = skb_shinfo(skb)->nr_frags;
423 unsigned int offset = offset_in_page(data);
424 unsigned int len = skb_headlen(skb);
425 unsigned int id;
426 grant_ref_t ref;
427 int i;
428
429 /* While the header overlaps a page boundary (including being
430 larger than a page), split it it into page-sized chunks. */
431 while (len > PAGE_SIZE - offset) {
432 tx->size = PAGE_SIZE - offset;
433 tx->flags |= NETTXF_more_data;
434 len -= tx->size;
435 data += tx->size;
436 offset = 0;
437
438 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
439 np->tx_skbs[id].skb = skb_get(skb);
440 tx = RING_GET_REQUEST(&np->tx, prod++);
441 tx->id = id;
442 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
443 BUG_ON((signed short)ref < 0);
444
445 mfn = virt_to_mfn(data);
446 gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id,
447 mfn, GNTMAP_readonly);
448
449 tx->gref = np->grant_tx_ref[id] = ref;
450 tx->offset = offset;
451 tx->size = len;
452 tx->flags = 0;
453 }
454
455 /* Grant backend access to each skb fragment page. */
456 for (i = 0; i < frags; i++) {
457 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
458
459 tx->flags |= NETTXF_more_data;
460
461 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
462 np->tx_skbs[id].skb = skb_get(skb);
463 tx = RING_GET_REQUEST(&np->tx, prod++);
464 tx->id = id;
465 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
466 BUG_ON((signed short)ref < 0);
467
468 mfn = pfn_to_mfn(page_to_pfn(frag->page));
469 gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id,
470 mfn, GNTMAP_readonly);
471
472 tx->gref = np->grant_tx_ref[id] = ref;
473 tx->offset = frag->page_offset;
474 tx->size = frag->size;
475 tx->flags = 0;
476 }
477
478 np->tx.req_prod_pvt = prod;
479}
480
481static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
482{
483 unsigned short id;
484 struct netfront_info *np = netdev_priv(dev);
485 struct xen_netif_tx_request *tx;
486 struct xen_netif_extra_info *extra;
487 char *data = skb->data;
488 RING_IDX i;
489 grant_ref_t ref;
490 unsigned long mfn;
491 int notify;
492 int frags = skb_shinfo(skb)->nr_frags;
493 unsigned int offset = offset_in_page(data);
494 unsigned int len = skb_headlen(skb);
495
496 frags += (offset + len + PAGE_SIZE - 1) / PAGE_SIZE;
497 if (unlikely(frags > MAX_SKB_FRAGS + 1)) {
498 printk(KERN_ALERT "xennet: skb rides the rocket: %d frags\n",
499 frags);
500 dump_stack();
501 goto drop;
502 }
503
504 spin_lock_irq(&np->tx_lock);
505
506 if (unlikely(!netif_carrier_ok(dev) ||
507 (frags > 1 && !xennet_can_sg(dev)) ||
508 netif_needs_gso(dev, skb))) {
509 spin_unlock_irq(&np->tx_lock);
510 goto drop;
511 }
512
513 i = np->tx.req_prod_pvt;
514
515 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
516 np->tx_skbs[id].skb = skb;
517
518 tx = RING_GET_REQUEST(&np->tx, i);
519
520 tx->id = id;
521 ref = gnttab_claim_grant_reference(&np->gref_tx_head);
522 BUG_ON((signed short)ref < 0);
523 mfn = virt_to_mfn(data);
524 gnttab_grant_foreign_access_ref(
525 ref, np->xbdev->otherend_id, mfn, GNTMAP_readonly);
526 tx->gref = np->grant_tx_ref[id] = ref;
527 tx->offset = offset;
528 tx->size = len;
529 extra = NULL;
530
531 tx->flags = 0;
532 if (skb->ip_summed == CHECKSUM_PARTIAL)
533 /* local packet? */
534 tx->flags |= NETTXF_csum_blank | NETTXF_data_validated;
535 else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
536 /* remote but checksummed. */
537 tx->flags |= NETTXF_data_validated;
538
539 if (skb_shinfo(skb)->gso_size) {
540 struct xen_netif_extra_info *gso;
541
542 gso = (struct xen_netif_extra_info *)
543 RING_GET_REQUEST(&np->tx, ++i);
544
545 if (extra)
546 extra->flags |= XEN_NETIF_EXTRA_FLAG_MORE;
547 else
548 tx->flags |= NETTXF_extra_info;
549
550 gso->u.gso.size = skb_shinfo(skb)->gso_size;
551 gso->u.gso.type = XEN_NETIF_GSO_TYPE_TCPV4;
552 gso->u.gso.pad = 0;
553 gso->u.gso.features = 0;
554
555 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
556 gso->flags = 0;
557 extra = gso;
558 }
559
560 np->tx.req_prod_pvt = i + 1;
561
562 xennet_make_frags(skb, dev, tx);
563 tx->size = skb->len;
564
565 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->tx, notify);
566 if (notify)
567 notify_remote_via_irq(np->netdev->irq);
568
569 xennet_tx_buf_gc(dev);
570
571 if (!netfront_tx_slot_available(np))
572 netif_stop_queue(dev);
573
574 spin_unlock_irq(&np->tx_lock);
575
576 np->stats.tx_bytes += skb->len;
577 np->stats.tx_packets++;
578
579 return 0;
580
581 drop:
582 np->stats.tx_dropped++;
583 dev_kfree_skb(skb);
584 return 0;
585}
586
587static int xennet_close(struct net_device *dev)
588{
589 struct netfront_info *np = netdev_priv(dev);
590 netif_stop_queue(np->netdev);
591 return 0;
592}
593
594static struct net_device_stats *xennet_get_stats(struct net_device *dev)
595{
596 struct netfront_info *np = netdev_priv(dev);
597 return &np->stats;
598}
599
600static void xennet_move_rx_slot(struct netfront_info *np, struct sk_buff *skb,
601 grant_ref_t ref)
602{
603 int new = xennet_rxidx(np->rx.req_prod_pvt);
604
605 BUG_ON(np->rx_skbs[new]);
606 np->rx_skbs[new] = skb;
607 np->grant_rx_ref[new] = ref;
608 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new;
609 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref;
610 np->rx.req_prod_pvt++;
611}
612
613static int xennet_get_extras(struct netfront_info *np,
614 struct xen_netif_extra_info *extras,
615 RING_IDX rp)
616
617{
618 struct xen_netif_extra_info *extra;
619 struct device *dev = &np->netdev->dev;
620 RING_IDX cons = np->rx.rsp_cons;
621 int err = 0;
622
623 do {
624 struct sk_buff *skb;
625 grant_ref_t ref;
626
627 if (unlikely(cons + 1 == rp)) {
628 if (net_ratelimit())
629 dev_warn(dev, "Missing extra info\n");
630 err = -EBADR;
631 break;
632 }
633
634 extra = (struct xen_netif_extra_info *)
635 RING_GET_RESPONSE(&np->rx, ++cons);
636
637 if (unlikely(!extra->type ||
638 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
639 if (net_ratelimit())
640 dev_warn(dev, "Invalid extra type: %d\n",
641 extra->type);
642 err = -EINVAL;
643 } else {
644 memcpy(&extras[extra->type - 1], extra,
645 sizeof(*extra));
646 }
647
648 skb = xennet_get_rx_skb(np, cons);
649 ref = xennet_get_rx_ref(np, cons);
650 xennet_move_rx_slot(np, skb, ref);
651 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
652
653 np->rx.rsp_cons = cons;
654 return err;
655}
656
657static int xennet_get_responses(struct netfront_info *np,
658 struct netfront_rx_info *rinfo, RING_IDX rp,
659 struct sk_buff_head *list)
660{
661 struct xen_netif_rx_response *rx = &rinfo->rx;
662 struct xen_netif_extra_info *extras = rinfo->extras;
663 struct device *dev = &np->netdev->dev;
664 RING_IDX cons = np->rx.rsp_cons;
665 struct sk_buff *skb = xennet_get_rx_skb(np, cons);
666 grant_ref_t ref = xennet_get_rx_ref(np, cons);
667 int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD);
668 int frags = 1;
669 int err = 0;
670 unsigned long ret;
671
672 if (rx->flags & NETRXF_extra_info) {
673 err = xennet_get_extras(np, extras, rp);
674 cons = np->rx.rsp_cons;
675 }
676
677 for (;;) {
678 if (unlikely(rx->status < 0 ||
679 rx->offset + rx->status > PAGE_SIZE)) {
680 if (net_ratelimit())
681 dev_warn(dev, "rx->offset: %x, size: %u\n",
682 rx->offset, rx->status);
683 xennet_move_rx_slot(np, skb, ref);
684 err = -EINVAL;
685 goto next;
686 }
687
688 /*
689 * This definitely indicates a bug, either in this driver or in
690 * the backend driver. In future this should flag the bad
691 * situation to the system controller to reboot the backed.
692 */
693 if (ref == GRANT_INVALID_REF) {
694 if (net_ratelimit())
695 dev_warn(dev, "Bad rx response id %d.\n",
696 rx->id);
697 err = -EINVAL;
698 goto next;
699 }
700
701 ret = gnttab_end_foreign_access_ref(ref, 0);
702 BUG_ON(!ret);
703
704 gnttab_release_grant_reference(&np->gref_rx_head, ref);
705
706 __skb_queue_tail(list, skb);
707
708next:
709 if (!(rx->flags & NETRXF_more_data))
710 break;
711
712 if (cons + frags == rp) {
713 if (net_ratelimit())
714 dev_warn(dev, "Need more frags\n");
715 err = -ENOENT;
716 break;
717 }
718
719 rx = RING_GET_RESPONSE(&np->rx, cons + frags);
720 skb = xennet_get_rx_skb(np, cons + frags);
721 ref = xennet_get_rx_ref(np, cons + frags);
722 frags++;
723 }
724
725 if (unlikely(frags > max)) {
726 if (net_ratelimit())
727 dev_warn(dev, "Too many frags\n");
728 err = -E2BIG;
729 }
730
731 if (unlikely(err))
732 np->rx.rsp_cons = cons + frags;
733
734 return err;
735}
736
737static int xennet_set_skb_gso(struct sk_buff *skb,
738 struct xen_netif_extra_info *gso)
739{
740 if (!gso->u.gso.size) {
741 if (net_ratelimit())
742 printk(KERN_WARNING "GSO size must not be zero.\n");
743 return -EINVAL;
744 }
745
746 /* Currently only TCPv4 S.O. is supported. */
747 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4) {
748 if (net_ratelimit())
749 printk(KERN_WARNING "Bad GSO type %d.\n", gso->u.gso.type);
750 return -EINVAL;
751 }
752
753 skb_shinfo(skb)->gso_size = gso->u.gso.size;
754 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
755
756 /* Header must be checked, and gso_segs computed. */
757 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
758 skb_shinfo(skb)->gso_segs = 0;
759
760 return 0;
761}
762
763static RING_IDX xennet_fill_frags(struct netfront_info *np,
764 struct sk_buff *skb,
765 struct sk_buff_head *list)
766{
767 struct skb_shared_info *shinfo = skb_shinfo(skb);
768 int nr_frags = shinfo->nr_frags;
769 RING_IDX cons = np->rx.rsp_cons;
770 skb_frag_t *frag = shinfo->frags + nr_frags;
771 struct sk_buff *nskb;
772
773 while ((nskb = __skb_dequeue(list))) {
774 struct xen_netif_rx_response *rx =
775 RING_GET_RESPONSE(&np->rx, ++cons);
776
777 frag->page = skb_shinfo(nskb)->frags[0].page;
778 frag->page_offset = rx->offset;
779 frag->size = rx->status;
780
781 skb->data_len += rx->status;
782
783 skb_shinfo(nskb)->nr_frags = 0;
784 kfree_skb(nskb);
785
786 frag++;
787 nr_frags++;
788 }
789
790 shinfo->nr_frags = nr_frags;
791 return cons;
792}
793
794static int skb_checksum_setup(struct sk_buff *skb)
795{
796 struct iphdr *iph;
797 unsigned char *th;
798 int err = -EPROTO;
799
800 if (skb->protocol != htons(ETH_P_IP))
801 goto out;
802
803 iph = (void *)skb->data;
804 th = skb->data + 4 * iph->ihl;
805 if (th >= skb_tail_pointer(skb))
806 goto out;
807
808 skb->csum_start = th - skb->head;
809 switch (iph->protocol) {
810 case IPPROTO_TCP:
811 skb->csum_offset = offsetof(struct tcphdr, check);
812 break;
813 case IPPROTO_UDP:
814 skb->csum_offset = offsetof(struct udphdr, check);
815 break;
816 default:
817 if (net_ratelimit())
818 printk(KERN_ERR "Attempting to checksum a non-"
819 "TCP/UDP packet, dropping a protocol"
820 " %d packet", iph->protocol);
821 goto out;
822 }
823
824 if ((th + skb->csum_offset + 2) > skb_tail_pointer(skb))
825 goto out;
826
827 err = 0;
828
829out:
830 return err;
831}
832
833static int handle_incoming_queue(struct net_device *dev,
834 struct sk_buff_head *rxq)
835{
836 struct netfront_info *np = netdev_priv(dev);
837 int packets_dropped = 0;
838 struct sk_buff *skb;
839
840 while ((skb = __skb_dequeue(rxq)) != NULL) {
841 struct page *page = NETFRONT_SKB_CB(skb)->page;
842 void *vaddr = page_address(page);
843 unsigned offset = NETFRONT_SKB_CB(skb)->offset;
844
845 memcpy(skb->data, vaddr + offset,
846 skb_headlen(skb));
847
848 if (page != skb_shinfo(skb)->frags[0].page)
849 __free_page(page);
850
851 /* Ethernet work: Delayed to here as it peeks the header. */
852 skb->protocol = eth_type_trans(skb, dev);
853
854 if (skb->ip_summed == CHECKSUM_PARTIAL) {
855 if (skb_checksum_setup(skb)) {
856 kfree_skb(skb);
857 packets_dropped++;
858 np->stats.rx_errors++;
859 continue;
860 }
861 }
862
863 np->stats.rx_packets++;
864 np->stats.rx_bytes += skb->len;
865
866 /* Pass it up. */
867 netif_receive_skb(skb);
868 dev->last_rx = jiffies;
869 }
870
871 return packets_dropped;
872}
873
874static int xennet_poll(struct net_device *dev, int *pbudget)
875{
876 struct netfront_info *np = netdev_priv(dev);
877 struct sk_buff *skb;
878 struct netfront_rx_info rinfo;
879 struct xen_netif_rx_response *rx = &rinfo.rx;
880 struct xen_netif_extra_info *extras = rinfo.extras;
881 RING_IDX i, rp;
882 int work_done, budget, more_to_do = 1;
883 struct sk_buff_head rxq;
884 struct sk_buff_head errq;
885 struct sk_buff_head tmpq;
886 unsigned long flags;
887 unsigned int len;
888 int err;
889
890 spin_lock(&np->rx_lock);
891
892 if (unlikely(!netif_carrier_ok(dev))) {
893 spin_unlock(&np->rx_lock);
894 return 0;
895 }
896
897 skb_queue_head_init(&rxq);
898 skb_queue_head_init(&errq);
899 skb_queue_head_init(&tmpq);
900
901 budget = *pbudget;
902 if (budget > dev->quota)
903 budget = dev->quota;
904 rp = np->rx.sring->rsp_prod;
905 rmb(); /* Ensure we see queued responses up to 'rp'. */
906
907 i = np->rx.rsp_cons;
908 work_done = 0;
909 while ((i != rp) && (work_done < budget)) {
910 memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx));
911 memset(extras, 0, sizeof(rinfo.extras));
912
913 err = xennet_get_responses(np, &rinfo, rp, &tmpq);
914
915 if (unlikely(err)) {
916err:
917 while ((skb = __skb_dequeue(&tmpq)))
918 __skb_queue_tail(&errq, skb);
919 np->stats.rx_errors++;
920 i = np->rx.rsp_cons;
921 continue;
922 }
923
924 skb = __skb_dequeue(&tmpq);
925
926 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
927 struct xen_netif_extra_info *gso;
928 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
929
930 if (unlikely(xennet_set_skb_gso(skb, gso))) {
931 __skb_queue_head(&tmpq, skb);
932 np->rx.rsp_cons += skb_queue_len(&tmpq);
933 goto err;
934 }
935 }
936
937 NETFRONT_SKB_CB(skb)->page = skb_shinfo(skb)->frags[0].page;
938 NETFRONT_SKB_CB(skb)->offset = rx->offset;
939
940 len = rx->status;
941 if (len > RX_COPY_THRESHOLD)
942 len = RX_COPY_THRESHOLD;
943 skb_put(skb, len);
944
945 if (rx->status > len) {
946 skb_shinfo(skb)->frags[0].page_offset =
947 rx->offset + len;
948 skb_shinfo(skb)->frags[0].size = rx->status - len;
949 skb->data_len = rx->status - len;
950 } else {
951 skb_shinfo(skb)->frags[0].page = NULL;
952 skb_shinfo(skb)->nr_frags = 0;
953 }
954
955 i = xennet_fill_frags(np, skb, &tmpq);
956
957 /*
958 * Truesize approximates the size of true data plus
959 * any supervisor overheads. Adding hypervisor
960 * overheads has been shown to significantly reduce
961 * achievable bandwidth with the default receive
962 * buffer size. It is therefore not wise to account
963 * for it here.
964 *
965 * After alloc_skb(RX_COPY_THRESHOLD), truesize is set
966 * to RX_COPY_THRESHOLD + the supervisor
967 * overheads. Here, we add the size of the data pulled
968 * in xennet_fill_frags().
969 *
970 * We also adjust for any unused space in the main
971 * data area by subtracting (RX_COPY_THRESHOLD -
972 * len). This is especially important with drivers
973 * which split incoming packets into header and data,
974 * using only 66 bytes of the main data area (see the
975 * e1000 driver for example.) On such systems,
976 * without this last adjustement, our achievable
977 * receive throughout using the standard receive
978 * buffer size was cut by 25%(!!!).
979 */
980 skb->truesize += skb->data_len - (RX_COPY_THRESHOLD - len);
981 skb->len += skb->data_len;
982
983 if (rx->flags & NETRXF_csum_blank)
984 skb->ip_summed = CHECKSUM_PARTIAL;
985 else if (rx->flags & NETRXF_data_validated)
986 skb->ip_summed = CHECKSUM_UNNECESSARY;
987
988 __skb_queue_tail(&rxq, skb);
989
990 np->rx.rsp_cons = ++i;
991 work_done++;
992 }
993
994 while ((skb = __skb_dequeue(&errq)))
995 kfree_skb(skb);
996
997 work_done -= handle_incoming_queue(dev, &rxq);
998
999 /* If we get a callback with very few responses, reduce fill target. */
1000 /* NB. Note exponential increase, linear decrease. */
1001 if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) >
1002 ((3*np->rx_target) / 4)) &&
1003 (--np->rx_target < np->rx_min_target))
1004 np->rx_target = np->rx_min_target;
1005
1006 xennet_alloc_rx_buffers(dev);
1007
1008 *pbudget -= work_done;
1009 dev->quota -= work_done;
1010
1011 if (work_done < budget) {
1012 local_irq_save(flags);
1013
1014 RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, more_to_do);
1015 if (!more_to_do)
1016 __netif_rx_complete(dev);
1017
1018 local_irq_restore(flags);
1019 }
1020
1021 spin_unlock(&np->rx_lock);
1022
1023 return more_to_do;
1024}
1025
1026static int xennet_change_mtu(struct net_device *dev, int mtu)
1027{
1028 int max = xennet_can_sg(dev) ? 65535 - ETH_HLEN : ETH_DATA_LEN;
1029
1030 if (mtu > max)
1031 return -EINVAL;
1032 dev->mtu = mtu;
1033 return 0;
1034}
1035
1036static void xennet_release_tx_bufs(struct netfront_info *np)
1037{
1038 struct sk_buff *skb;
1039 int i;
1040
1041 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1042 /* Skip over entries which are actually freelist references */
1043 if ((unsigned long)np->tx_skbs[i].skb < PAGE_OFFSET)
1044 continue;
1045
1046 skb = np->tx_skbs[i].skb;
1047 gnttab_end_foreign_access_ref(np->grant_tx_ref[i],
1048 GNTMAP_readonly);
1049 gnttab_release_grant_reference(&np->gref_tx_head,
1050 np->grant_tx_ref[i]);
1051 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1052 add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, i);
1053 dev_kfree_skb_irq(skb);
1054 }
1055}
1056
1057static void xennet_release_rx_bufs(struct netfront_info *np)
1058{
1059 struct mmu_update *mmu = np->rx_mmu;
1060 struct multicall_entry *mcl = np->rx_mcl;
1061 struct sk_buff_head free_list;
1062 struct sk_buff *skb;
1063 unsigned long mfn;
1064 int xfer = 0, noxfer = 0, unused = 0;
1065 int id, ref;
1066
1067 dev_warn(&np->netdev->dev, "%s: fix me for copying receiver.\n",
1068 __func__);
1069 return;
1070
1071 skb_queue_head_init(&free_list);
1072
1073 spin_lock_bh(&np->rx_lock);
1074
1075 for (id = 0; id < NET_RX_RING_SIZE; id++) {
1076 ref = np->grant_rx_ref[id];
1077 if (ref == GRANT_INVALID_REF) {
1078 unused++;
1079 continue;
1080 }
1081
1082 skb = np->rx_skbs[id];
1083 mfn = gnttab_end_foreign_transfer_ref(ref);
1084 gnttab_release_grant_reference(&np->gref_rx_head, ref);
1085 np->grant_rx_ref[id] = GRANT_INVALID_REF;
1086
1087 if (0 == mfn) {
1088 skb_shinfo(skb)->nr_frags = 0;
1089 dev_kfree_skb(skb);
1090 noxfer++;
1091 continue;
1092 }
1093
1094 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1095 /* Remap the page. */
1096 struct page *page = skb_shinfo(skb)->frags[0].page;
1097 unsigned long pfn = page_to_pfn(page);
1098 void *vaddr = page_address(page);
1099
1100 MULTI_update_va_mapping(mcl, (unsigned long)vaddr,
1101 mfn_pte(mfn, PAGE_KERNEL),
1102 0);
1103 mcl++;
1104 mmu->ptr = ((u64)mfn << PAGE_SHIFT)
1105 | MMU_MACHPHYS_UPDATE;
1106 mmu->val = pfn;
1107 mmu++;
1108
1109 set_phys_to_machine(pfn, mfn);
1110 }
1111 __skb_queue_tail(&free_list, skb);
1112 xfer++;
1113 }
1114
1115 dev_info(&np->netdev->dev, "%s: %d xfer, %d noxfer, %d unused\n",
1116 __func__, xfer, noxfer, unused);
1117
1118 if (xfer) {
1119 if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1120 /* Do all the remapping work and M2P updates. */
1121 MULTI_mmu_update(mcl, np->rx_mmu, mmu - np->rx_mmu,
1122 0, DOMID_SELF);
1123 mcl++;
1124 HYPERVISOR_multicall(np->rx_mcl, mcl - np->rx_mcl);
1125 }
1126 }
1127
1128 while ((skb = __skb_dequeue(&free_list)) != NULL)
1129 dev_kfree_skb(skb);
1130
1131 spin_unlock_bh(&np->rx_lock);
1132}
1133
1134static void xennet_uninit(struct net_device *dev)
1135{
1136 struct netfront_info *np = netdev_priv(dev);
1137 xennet_release_tx_bufs(np);
1138 xennet_release_rx_bufs(np);
1139 gnttab_free_grant_references(np->gref_tx_head);
1140 gnttab_free_grant_references(np->gref_rx_head);
1141}
1142
1143static struct net_device * __devinit xennet_create_dev(struct xenbus_device *dev)
1144{
1145 int i, err;
1146 struct net_device *netdev;
1147 struct netfront_info *np;
1148
1149 netdev = alloc_etherdev(sizeof(struct netfront_info));
1150 if (!netdev) {
1151 printk(KERN_WARNING "%s> alloc_etherdev failed.\n",
1152 __func__);
1153 return ERR_PTR(-ENOMEM);
1154 }
1155
1156 np = netdev_priv(netdev);
1157 np->xbdev = dev;
1158
1159 spin_lock_init(&np->tx_lock);
1160 spin_lock_init(&np->rx_lock);
1161
1162 skb_queue_head_init(&np->rx_batch);
1163 np->rx_target = RX_DFL_MIN_TARGET;
1164 np->rx_min_target = RX_DFL_MIN_TARGET;
1165 np->rx_max_target = RX_MAX_TARGET;
1166
1167 init_timer(&np->rx_refill_timer);
1168 np->rx_refill_timer.data = (unsigned long)netdev;
1169 np->rx_refill_timer.function = rx_refill_timeout;
1170
1171 /* Initialise tx_skbs as a free chain containing every entry. */
1172 np->tx_skb_freelist = 0;
1173 for (i = 0; i < NET_TX_RING_SIZE; i++) {
1174 np->tx_skbs[i].link = i+1;
1175 np->grant_tx_ref[i] = GRANT_INVALID_REF;
1176 }
1177
1178 /* Clear out rx_skbs */
1179 for (i = 0; i < NET_RX_RING_SIZE; i++) {
1180 np->rx_skbs[i] = NULL;
1181 np->grant_rx_ref[i] = GRANT_INVALID_REF;
1182 }
1183
1184 /* A grant for every tx ring slot */
1185 if (gnttab_alloc_grant_references(TX_MAX_TARGET,
1186 &np->gref_tx_head) < 0) {
1187 printk(KERN_ALERT "#### netfront can't alloc tx grant refs\n");
1188 err = -ENOMEM;
1189 goto exit;
1190 }
1191 /* A grant for every rx ring slot */
1192 if (gnttab_alloc_grant_references(RX_MAX_TARGET,
1193 &np->gref_rx_head) < 0) {
1194 printk(KERN_ALERT "#### netfront can't alloc rx grant refs\n");
1195 err = -ENOMEM;
1196 goto exit_free_tx;
1197 }
1198
1199 netdev->open = xennet_open;
1200 netdev->hard_start_xmit = xennet_start_xmit;
1201 netdev->stop = xennet_close;
1202 netdev->get_stats = xennet_get_stats;
1203 netdev->poll = xennet_poll;
1204 netdev->uninit = xennet_uninit;
1205 netdev->change_mtu = xennet_change_mtu;
1206 netdev->weight = 64;
1207 netdev->features = NETIF_F_IP_CSUM;
1208
1209 SET_ETHTOOL_OPS(netdev, &xennet_ethtool_ops);
1210 SET_MODULE_OWNER(netdev);
1211 SET_NETDEV_DEV(netdev, &dev->dev);
1212
1213 np->netdev = netdev;
1214
1215 netif_carrier_off(netdev);
1216
1217 return netdev;
1218
1219 exit_free_tx:
1220 gnttab_free_grant_references(np->gref_tx_head);
1221 exit:
1222 free_netdev(netdev);
1223 return ERR_PTR(err);
1224}
1225
1226/**
1227 * Entry point to this code when a new device is created. Allocate the basic
1228 * structures and the ring buffers for communication with the backend, and
1229 * inform the backend of the appropriate details for those.
1230 */
1231static int __devinit netfront_probe(struct xenbus_device *dev,
1232 const struct xenbus_device_id *id)
1233{
1234 int err;
1235 struct net_device *netdev;
1236 struct netfront_info *info;
1237
1238 netdev = xennet_create_dev(dev);
1239 if (IS_ERR(netdev)) {
1240 err = PTR_ERR(netdev);
1241 xenbus_dev_fatal(dev, err, "creating netdev");
1242 return err;
1243 }
1244
1245 info = netdev_priv(netdev);
1246 dev->dev.driver_data = info;
1247
1248 err = register_netdev(info->netdev);
1249 if (err) {
1250 printk(KERN_WARNING "%s: register_netdev err=%d\n",
1251 __func__, err);
1252 goto fail;
1253 }
1254
1255 err = xennet_sysfs_addif(info->netdev);
1256 if (err) {
1257 unregister_netdev(info->netdev);
1258 printk(KERN_WARNING "%s: add sysfs failed err=%d\n",
1259 __func__, err);
1260 goto fail;
1261 }
1262
1263 return 0;
1264
1265 fail:
1266 free_netdev(netdev);
1267 dev->dev.driver_data = NULL;
1268 return err;
1269}
1270
1271static void xennet_end_access(int ref, void *page)
1272{
1273 /* This frees the page as a side-effect */
1274 if (ref != GRANT_INVALID_REF)
1275 gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1276}
1277
1278static void xennet_disconnect_backend(struct netfront_info *info)
1279{
1280 /* Stop old i/f to prevent errors whilst we rebuild the state. */
1281 spin_lock_bh(&info->rx_lock);
1282 spin_lock_irq(&info->tx_lock);
1283 netif_carrier_off(info->netdev);
1284 spin_unlock_irq(&info->tx_lock);
1285 spin_unlock_bh(&info->rx_lock);
1286
1287 if (info->netdev->irq)
1288 unbind_from_irqhandler(info->netdev->irq, info->netdev);
1289 info->evtchn = info->netdev->irq = 0;
1290
1291 /* End access and free the pages */
1292 xennet_end_access(info->tx_ring_ref, info->tx.sring);
1293 xennet_end_access(info->rx_ring_ref, info->rx.sring);
1294
1295 info->tx_ring_ref = GRANT_INVALID_REF;
1296 info->rx_ring_ref = GRANT_INVALID_REF;
1297 info->tx.sring = NULL;
1298 info->rx.sring = NULL;
1299}
1300
1301/**
1302 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1303 * driver restart. We tear down our netif structure and recreate it, but
1304 * leave the device-layer structures intact so that this is transparent to the
1305 * rest of the kernel.
1306 */
1307static int netfront_resume(struct xenbus_device *dev)
1308{
1309 struct netfront_info *info = dev->dev.driver_data;
1310
1311 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1312
1313 xennet_disconnect_backend(info);
1314 return 0;
1315}
1316
1317static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1318{
1319 char *s, *e, *macstr;
1320 int i;
1321
1322 macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1323 if (IS_ERR(macstr))
1324 return PTR_ERR(macstr);
1325
1326 for (i = 0; i < ETH_ALEN; i++) {
1327 mac[i] = simple_strtoul(s, &e, 16);
1328 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1329 kfree(macstr);
1330 return -ENOENT;
1331 }
1332 s = e+1;
1333 }
1334
1335 kfree(macstr);
1336 return 0;
1337}
1338
1339static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1340{
1341 struct net_device *dev = dev_id;
1342 struct netfront_info *np = netdev_priv(dev);
1343 unsigned long flags;
1344
1345 spin_lock_irqsave(&np->tx_lock, flags);
1346
1347 if (likely(netif_carrier_ok(dev))) {
1348 xennet_tx_buf_gc(dev);
1349 /* Under tx_lock: protects access to rx shared-ring indexes. */
1350 if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx))
1351 netif_rx_schedule(dev);
1352 }
1353
1354 spin_unlock_irqrestore(&np->tx_lock, flags);
1355
1356 return IRQ_HANDLED;
1357}
1358
1359static int setup_netfront(struct xenbus_device *dev, struct netfront_info *info)
1360{
1361 struct xen_netif_tx_sring *txs;
1362 struct xen_netif_rx_sring *rxs;
1363 int err;
1364 struct net_device *netdev = info->netdev;
1365
1366 info->tx_ring_ref = GRANT_INVALID_REF;
1367 info->rx_ring_ref = GRANT_INVALID_REF;
1368 info->rx.sring = NULL;
1369 info->tx.sring = NULL;
1370 netdev->irq = 0;
1371
1372 err = xen_net_read_mac(dev, netdev->dev_addr);
1373 if (err) {
1374 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1375 goto fail;
1376 }
1377
1378 txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_KERNEL);
1379 if (!txs) {
1380 err = -ENOMEM;
1381 xenbus_dev_fatal(dev, err, "allocating tx ring page");
1382 goto fail;
1383 }
1384 SHARED_RING_INIT(txs);
1385 FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE);
1386
1387 err = xenbus_grant_ring(dev, virt_to_mfn(txs));
1388 if (err < 0) {
1389 free_page((unsigned long)txs);
1390 goto fail;
1391 }
1392
1393 info->tx_ring_ref = err;
1394 rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_KERNEL);
1395 if (!rxs) {
1396 err = -ENOMEM;
1397 xenbus_dev_fatal(dev, err, "allocating rx ring page");
1398 goto fail;
1399 }
1400 SHARED_RING_INIT(rxs);
1401 FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE);
1402
1403 err = xenbus_grant_ring(dev, virt_to_mfn(rxs));
1404 if (err < 0) {
1405 free_page((unsigned long)rxs);
1406 goto fail;
1407 }
1408 info->rx_ring_ref = err;
1409
1410 err = xenbus_alloc_evtchn(dev, &info->evtchn);
1411 if (err)
1412 goto fail;
1413
1414 err = bind_evtchn_to_irqhandler(info->evtchn, xennet_interrupt,
1415 IRQF_SAMPLE_RANDOM, netdev->name,
1416 netdev);
1417 if (err < 0)
1418 goto fail;
1419 netdev->irq = err;
1420 return 0;
1421
1422 fail:
1423 return err;
1424}
1425
1426/* Common code used when first setting up, and when resuming. */
1427static int talk_to_backend(struct xenbus_device *dev,
1428 struct netfront_info *info)
1429{
1430 const char *message;
1431 struct xenbus_transaction xbt;
1432 int err;
1433
1434 /* Create shared ring, alloc event channel. */
1435 err = setup_netfront(dev, info);
1436 if (err)
1437 goto out;
1438
1439again:
1440 err = xenbus_transaction_start(&xbt);
1441 if (err) {
1442 xenbus_dev_fatal(dev, err, "starting transaction");
1443 goto destroy_ring;
1444 }
1445
1446 err = xenbus_printf(xbt, dev->nodename, "tx-ring-ref", "%u",
1447 info->tx_ring_ref);
1448 if (err) {
1449 message = "writing tx ring-ref";
1450 goto abort_transaction;
1451 }
1452 err = xenbus_printf(xbt, dev->nodename, "rx-ring-ref", "%u",
1453 info->rx_ring_ref);
1454 if (err) {
1455 message = "writing rx ring-ref";
1456 goto abort_transaction;
1457 }
1458 err = xenbus_printf(xbt, dev->nodename,
1459 "event-channel", "%u", info->evtchn);
1460 if (err) {
1461 message = "writing event-channel";
1462 goto abort_transaction;
1463 }
1464
1465 err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1466 1);
1467 if (err) {
1468 message = "writing request-rx-copy";
1469 goto abort_transaction;
1470 }
1471
1472 err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1473 if (err) {
1474 message = "writing feature-rx-notify";
1475 goto abort_transaction;
1476 }
1477
1478 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1479 if (err) {
1480 message = "writing feature-sg";
1481 goto abort_transaction;
1482 }
1483
1484 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1485 if (err) {
1486 message = "writing feature-gso-tcpv4";
1487 goto abort_transaction;
1488 }
1489
1490 err = xenbus_transaction_end(xbt, 0);
1491 if (err) {
1492 if (err == -EAGAIN)
1493 goto again;
1494 xenbus_dev_fatal(dev, err, "completing transaction");
1495 goto destroy_ring;
1496 }
1497
1498 return 0;
1499
1500 abort_transaction:
1501 xenbus_transaction_end(xbt, 1);
1502 xenbus_dev_fatal(dev, err, "%s", message);
1503 destroy_ring:
1504 xennet_disconnect_backend(info);
1505 out:
1506 return err;
1507}
1508
1509static int xennet_set_sg(struct net_device *dev, u32 data)
1510{
1511 if (data) {
1512 struct netfront_info *np = netdev_priv(dev);
1513 int val;
1514
1515 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg",
1516 "%d", &val) < 0)
1517 val = 0;
1518 if (!val)
1519 return -ENOSYS;
1520 } else if (dev->mtu > ETH_DATA_LEN)
1521 dev->mtu = ETH_DATA_LEN;
1522
1523 return ethtool_op_set_sg(dev, data);
1524}
1525
1526static int xennet_set_tso(struct net_device *dev, u32 data)
1527{
1528 if (data) {
1529 struct netfront_info *np = netdev_priv(dev);
1530 int val;
1531
1532 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1533 "feature-gso-tcpv4", "%d", &val) < 0)
1534 val = 0;
1535 if (!val)
1536 return -ENOSYS;
1537 }
1538
1539 return ethtool_op_set_tso(dev, data);
1540}
1541
1542static void xennet_set_features(struct net_device *dev)
1543{
1544 /* Turn off all GSO bits except ROBUST. */
1545 dev->features &= (1 << NETIF_F_GSO_SHIFT) - 1;
1546 dev->features |= NETIF_F_GSO_ROBUST;
1547 xennet_set_sg(dev, 0);
1548
1549 /* We need checksum offload to enable scatter/gather and TSO. */
1550 if (!(dev->features & NETIF_F_IP_CSUM))
1551 return;
1552
1553 if (!xennet_set_sg(dev, 1))
1554 xennet_set_tso(dev, 1);
1555}
1556
1557static int xennet_connect(struct net_device *dev)
1558{
1559 struct netfront_info *np = netdev_priv(dev);
1560 int i, requeue_idx, err;
1561 struct sk_buff *skb;
1562 grant_ref_t ref;
1563 struct xen_netif_rx_request *req;
1564 unsigned int feature_rx_copy;
1565
1566 err = xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1567 "feature-rx-copy", "%u", &feature_rx_copy);
1568 if (err != 1)
1569 feature_rx_copy = 0;
1570
1571 if (!feature_rx_copy) {
1572 dev_info(&dev->dev,
1573 "backend does not support copying recieve path");
1574 return -ENODEV;
1575 }
1576
1577 err = talk_to_backend(np->xbdev, np);
1578 if (err)
1579 return err;
1580
1581 xennet_set_features(dev);
1582
1583 spin_lock_bh(&np->rx_lock);
1584 spin_lock_irq(&np->tx_lock);
1585
1586 /* Step 1: Discard all pending TX packet fragments. */
1587 xennet_release_tx_bufs(np);
1588
1589 /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1590 for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) {
1591 if (!np->rx_skbs[i])
1592 continue;
1593
1594 skb = np->rx_skbs[requeue_idx] = xennet_get_rx_skb(np, i);
1595 ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i);
1596 req = RING_GET_REQUEST(&np->rx, requeue_idx);
1597
1598 gnttab_grant_foreign_access_ref(
1599 ref, np->xbdev->otherend_id,
1600 pfn_to_mfn(page_to_pfn(skb_shinfo(skb)->
1601 frags->page)),
1602 0);
1603 req->gref = ref;
1604 req->id = requeue_idx;
1605
1606 requeue_idx++;
1607 }
1608
1609 np->rx.req_prod_pvt = requeue_idx;
1610
1611 /*
1612 * Step 3: All public and private state should now be sane. Get
1613 * ready to start sending and receiving packets and give the driver
1614 * domain a kick because we've probably just requeued some
1615 * packets.
1616 */
1617 netif_carrier_on(np->netdev);
1618 notify_remote_via_irq(np->netdev->irq);
1619 xennet_tx_buf_gc(dev);
1620 xennet_alloc_rx_buffers(dev);
1621
1622 spin_unlock_irq(&np->tx_lock);
1623 spin_unlock_bh(&np->rx_lock);
1624
1625 return 0;
1626}
1627
1628/**
1629 * Callback received when the backend's state changes.
1630 */
1631static void backend_changed(struct xenbus_device *dev,
1632 enum xenbus_state backend_state)
1633{
1634 struct netfront_info *np = dev->dev.driver_data;
1635 struct net_device *netdev = np->netdev;
1636
1637 dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
1638
1639 switch (backend_state) {
1640 case XenbusStateInitialising:
1641 case XenbusStateInitialised:
1642 case XenbusStateConnected:
1643 case XenbusStateUnknown:
1644 case XenbusStateClosed:
1645 break;
1646
1647 case XenbusStateInitWait:
1648 if (dev->state != XenbusStateInitialising)
1649 break;
1650 if (xennet_connect(netdev) != 0)
1651 break;
1652 xenbus_switch_state(dev, XenbusStateConnected);
1653 break;
1654
1655 case XenbusStateClosing:
1656 xenbus_frontend_closed(dev);
1657 break;
1658 }
1659}
1660
1661static struct ethtool_ops xennet_ethtool_ops =
1662{
1663 .get_tx_csum = ethtool_op_get_tx_csum,
1664 .set_tx_csum = ethtool_op_set_tx_csum,
1665 .get_sg = ethtool_op_get_sg,
1666 .set_sg = xennet_set_sg,
1667 .get_tso = ethtool_op_get_tso,
1668 .set_tso = xennet_set_tso,
1669 .get_link = ethtool_op_get_link,
1670};
1671
1672#ifdef CONFIG_SYSFS
1673static ssize_t show_rxbuf_min(struct device *dev,
1674 struct device_attribute *attr, char *buf)
1675{
1676 struct net_device *netdev = to_net_dev(dev);
1677 struct netfront_info *info = netdev_priv(netdev);
1678
1679 return sprintf(buf, "%u\n", info->rx_min_target);
1680}
1681
1682static ssize_t store_rxbuf_min(struct device *dev,
1683 struct device_attribute *attr,
1684 const char *buf, size_t len)
1685{
1686 struct net_device *netdev = to_net_dev(dev);
1687 struct netfront_info *np = netdev_priv(netdev);
1688 char *endp;
1689 unsigned long target;
1690
1691 if (!capable(CAP_NET_ADMIN))
1692 return -EPERM;
1693
1694 target = simple_strtoul(buf, &endp, 0);
1695 if (endp == buf)
1696 return -EBADMSG;
1697
1698 if (target < RX_MIN_TARGET)
1699 target = RX_MIN_TARGET;
1700 if (target > RX_MAX_TARGET)
1701 target = RX_MAX_TARGET;
1702
1703 spin_lock_bh(&np->rx_lock);
1704 if (target > np->rx_max_target)
1705 np->rx_max_target = target;
1706 np->rx_min_target = target;
1707 if (target > np->rx_target)
1708 np->rx_target = target;
1709
1710 xennet_alloc_rx_buffers(netdev);
1711
1712 spin_unlock_bh(&np->rx_lock);
1713 return len;
1714}
1715
1716static ssize_t show_rxbuf_max(struct device *dev,
1717 struct device_attribute *attr, char *buf)
1718{
1719 struct net_device *netdev = to_net_dev(dev);
1720 struct netfront_info *info = netdev_priv(netdev);
1721
1722 return sprintf(buf, "%u\n", info->rx_max_target);
1723}
1724
1725static ssize_t store_rxbuf_max(struct device *dev,
1726 struct device_attribute *attr,
1727 const char *buf, size_t len)
1728{
1729 struct net_device *netdev = to_net_dev(dev);
1730 struct netfront_info *np = netdev_priv(netdev);
1731 char *endp;
1732 unsigned long target;
1733
1734 if (!capable(CAP_NET_ADMIN))
1735 return -EPERM;
1736
1737 target = simple_strtoul(buf, &endp, 0);
1738 if (endp == buf)
1739 return -EBADMSG;
1740
1741 if (target < RX_MIN_TARGET)
1742 target = RX_MIN_TARGET;
1743 if (target > RX_MAX_TARGET)
1744 target = RX_MAX_TARGET;
1745
1746 spin_lock_bh(&np->rx_lock);
1747 if (target < np->rx_min_target)
1748 np->rx_min_target = target;
1749 np->rx_max_target = target;
1750 if (target < np->rx_target)
1751 np->rx_target = target;
1752
1753 xennet_alloc_rx_buffers(netdev);
1754
1755 spin_unlock_bh(&np->rx_lock);
1756 return len;
1757}
1758
1759static ssize_t show_rxbuf_cur(struct device *dev,
1760 struct device_attribute *attr, char *buf)
1761{
1762 struct net_device *netdev = to_net_dev(dev);
1763 struct netfront_info *info = netdev_priv(netdev);
1764
1765 return sprintf(buf, "%u\n", info->rx_target);
1766}
1767
1768static struct device_attribute xennet_attrs[] = {
1769 __ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf_min, store_rxbuf_min),
1770 __ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf_max, store_rxbuf_max),
1771 __ATTR(rxbuf_cur, S_IRUGO, show_rxbuf_cur, NULL),
1772};
1773
1774static int xennet_sysfs_addif(struct net_device *netdev)
1775{
1776 int i;
1777 int err;
1778
1779 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) {
1780 err = device_create_file(&netdev->dev,
1781 &xennet_attrs[i]);
1782 if (err)
1783 goto fail;
1784 }
1785 return 0;
1786
1787 fail:
1788 while (--i >= 0)
1789 device_remove_file(&netdev->dev, &xennet_attrs[i]);
1790 return err;
1791}
1792
1793static void xennet_sysfs_delif(struct net_device *netdev)
1794{
1795 int i;
1796
1797 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++)
1798 device_remove_file(&netdev->dev, &xennet_attrs[i]);
1799}
1800
1801#endif /* CONFIG_SYSFS */
1802
1803static struct xenbus_device_id netfront_ids[] = {
1804 { "vif" },
1805 { "" }
1806};
1807
1808
1809static int __devexit xennet_remove(struct xenbus_device *dev)
1810{
1811 struct netfront_info *info = dev->dev.driver_data;
1812
1813 dev_dbg(&dev->dev, "%s\n", dev->nodename);
1814
1815 unregister_netdev(info->netdev);
1816
1817 xennet_disconnect_backend(info);
1818
1819 del_timer_sync(&info->rx_refill_timer);
1820
1821 xennet_sysfs_delif(info->netdev);
1822
1823 free_netdev(info->netdev);
1824
1825 return 0;
1826}
1827
1828static struct xenbus_driver netfront = {
1829 .name = "vif",
1830 .owner = THIS_MODULE,
1831 .ids = netfront_ids,
1832 .probe = netfront_probe,
1833 .remove = __devexit_p(xennet_remove),
1834 .resume = netfront_resume,
1835 .otherend_changed = backend_changed,
1836};
1837
1838static int __init netif_init(void)
1839{
1840 if (!is_running_on_xen())
1841 return -ENODEV;
1842
1843 if (is_initial_xendomain())
1844 return 0;
1845
1846 printk(KERN_INFO "Initialising Xen virtual ethernet driver.\n");
1847
1848 return xenbus_register_frontend(&netfront);
1849}
1850module_init(netif_init);
1851
1852
1853static void __exit netif_exit(void)
1854{
1855 if (is_initial_xendomain())
1856 return;
1857
1858 return xenbus_unregister_driver(&netfront);
1859}
1860module_exit(netif_exit);
1861
1862MODULE_DESCRIPTION("Xen virtual network device frontend");
1863MODULE_LICENSE("GPL");