2 * Back-end of the driver for virtual network devices. This portion of the
3 * driver exports a 'unified' network-device interface that can be accessed
4 * by any operating system that implements a compatible front end. A
5 * reference front-end implementation can be found in:
6 * drivers/net/xen-netfront.c
8 * Copyright (c) 2002-2005, K A Fraser
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License version 2
12 * as published by the Free Software Foundation; or, when distributed
13 * separately from the Linux kernel or incorporated into other
14 * software packages, subject to the following license:
16 * Permission is hereby granted, free of charge, to any person obtaining a copy
17 * of this source file (the "Software"), to deal in the Software without
18 * restriction, including without limitation the rights to use, copy, modify,
19 * merge, publish, distribute, sublicense, and/or sell copies of the Software,
20 * and to permit persons to whom the Software is furnished to do so, subject to
21 * the following conditions:
23 * The above copyright notice and this permission notice shall be included in
24 * all copies or substantial portions of the Software.
26 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
27 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
28 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
29 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
30 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
31 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
37 #include <linux/kthread.h>
38 #include <linux/if_vlan.h>
39 #include <linux/udp.h>
40 #include <linux/highmem.h>
45 #include <xen/events.h>
46 #include <xen/interface/memory.h>
49 #include <asm/xen/hypercall.h>
51 /* Provide an option to disable split event channels at load time as
52 * event channels are limited resource. Split event channels are
55 bool separate_tx_rx_irq = true;
56 module_param(separate_tx_rx_irq, bool, 0644);
58 /* The time that packets can stay on the guest Rx internal queue
59 * before they are dropped.
61 unsigned int rx_drain_timeout_msecs = 10000;
62 module_param(rx_drain_timeout_msecs, uint, 0444);
64 /* The length of time before the frontend is considered unresponsive
65 * because it isn't providing Rx slots.
67 unsigned int rx_stall_timeout_msecs = 60000;
68 module_param(rx_stall_timeout_msecs, uint, 0444);
70 #define MAX_QUEUES_DEFAULT 8
71 unsigned int xenvif_max_queues;
72 module_param_named(max_queues, xenvif_max_queues, uint, 0644);
73 MODULE_PARM_DESC(max_queues,
74 "Maximum number of queues per virtual interface");
77 * This is the maximum slots a skb can have. If a guest sends a skb
78 * which exceeds this limit it is considered malicious.
80 #define FATAL_SKB_SLOTS_DEFAULT 20
81 static unsigned int fatal_skb_slots = FATAL_SKB_SLOTS_DEFAULT;
82 module_param(fatal_skb_slots, uint, 0444);
84 /* The amount to copy out of the first guest Tx slot into the skb's
85 * linear area. If the first slot has more data, it will be mapped
86 * and put into the first frag.
88 * This is sized to avoid pulling headers from the frags for most
91 #define XEN_NETBACK_TX_COPY_LEN 128
93 /* This is the maximum number of flows in the hash cache. */
94 #define XENVIF_HASH_CACHE_SIZE_DEFAULT 64
95 unsigned int xenvif_hash_cache_size = XENVIF_HASH_CACHE_SIZE_DEFAULT;
96 module_param_named(hash_cache_size, xenvif_hash_cache_size, uint, 0644);
97 MODULE_PARM_DESC(hash_cache_size, "Number of flows in the hash cache");
99 static void xenvif_idx_release(struct xenvif_queue *queue, u16 pending_idx,
102 static void make_tx_response(struct xenvif_queue *queue,
103 struct xen_netif_tx_request *txp,
104 unsigned int extra_count,
106 static void push_tx_responses(struct xenvif_queue *queue);
108 static inline int tx_work_todo(struct xenvif_queue *queue);
110 static inline unsigned long idx_to_pfn(struct xenvif_queue *queue,
113 return page_to_pfn(queue->mmap_pages[idx]);
116 static inline unsigned long idx_to_kaddr(struct xenvif_queue *queue,
119 return (unsigned long)pfn_to_kaddr(idx_to_pfn(queue, idx));
122 #define callback_param(vif, pending_idx) \
123 (vif->pending_tx_info[pending_idx].callback_struct)
125 /* Find the containing VIF's structure from a pointer in pending_tx_info array
127 static inline struct xenvif_queue *ubuf_to_queue(const struct ubuf_info *ubuf)
129 u16 pending_idx = ubuf->desc;
130 struct pending_tx_info *temp =
131 container_of(ubuf, struct pending_tx_info, callback_struct);
132 return container_of(temp - pending_idx,
137 static u16 frag_get_pending_idx(skb_frag_t *frag)
139 return (u16)frag->page_offset;
142 static void frag_set_pending_idx(skb_frag_t *frag, u16 pending_idx)
144 frag->page_offset = pending_idx;
147 static inline pending_ring_idx_t pending_index(unsigned i)
149 return i & (MAX_PENDING_REQS-1);
152 void xenvif_kick_thread(struct xenvif_queue *queue)
157 void xenvif_napi_schedule_or_enable_events(struct xenvif_queue *queue)
161 RING_FINAL_CHECK_FOR_REQUESTS(&queue->tx, more_to_do);
164 napi_schedule(&queue->napi);
165 else if (atomic_fetch_andnot(NETBK_TX_EOI | NETBK_COMMON_EOI,
166 &queue->eoi_pending) &
167 (NETBK_TX_EOI | NETBK_COMMON_EOI))
168 xen_irq_lateeoi(queue->tx_irq, 0);
171 static void tx_add_credit(struct xenvif_queue *queue)
173 unsigned long max_burst, max_credit;
176 * Allow a burst big enough to transmit a jumbo packet of up to 128kB.
177 * Otherwise the interface can seize up due to insufficient credit.
179 max_burst = max(131072UL, queue->credit_bytes);
181 /* Take care that adding a new chunk of credit doesn't wrap to zero. */
182 max_credit = queue->remaining_credit + queue->credit_bytes;
183 if (max_credit < queue->remaining_credit)
184 max_credit = ULONG_MAX; /* wrapped: clamp to ULONG_MAX */
186 queue->remaining_credit = min(max_credit, max_burst);
187 queue->rate_limited = false;
190 void xenvif_tx_credit_callback(unsigned long data)
192 struct xenvif_queue *queue = (struct xenvif_queue *)data;
193 tx_add_credit(queue);
194 xenvif_napi_schedule_or_enable_events(queue);
197 static void xenvif_tx_err(struct xenvif_queue *queue,
198 struct xen_netif_tx_request *txp,
199 unsigned int extra_count, RING_IDX end)
201 RING_IDX cons = queue->tx.req_cons;
205 spin_lock_irqsave(&queue->response_lock, flags);
206 make_tx_response(queue, txp, extra_count, XEN_NETIF_RSP_ERROR);
207 push_tx_responses(queue);
208 spin_unlock_irqrestore(&queue->response_lock, flags);
211 RING_COPY_REQUEST(&queue->tx, cons++, txp);
212 extra_count = 0; /* only the first frag can have extras */
214 queue->tx.req_cons = cons;
217 static void xenvif_fatal_tx_err(struct xenvif *vif)
219 netdev_err(vif->dev, "fatal error; disabling device\n");
220 vif->disabled = true;
221 /* Disable the vif from queue 0's kthread */
223 xenvif_kick_thread(&vif->queues[0]);
226 static int xenvif_count_requests(struct xenvif_queue *queue,
227 struct xen_netif_tx_request *first,
228 unsigned int extra_count,
229 struct xen_netif_tx_request *txp,
232 RING_IDX cons = queue->tx.req_cons;
237 if (!(first->flags & XEN_NETTXF_more_data))
241 struct xen_netif_tx_request dropped_tx = { 0 };
243 if (slots >= work_to_do) {
244 netdev_err(queue->vif->dev,
245 "Asked for %d slots but exceeds this limit\n",
247 xenvif_fatal_tx_err(queue->vif);
251 /* This guest is really using too many slots and
252 * considered malicious.
254 if (unlikely(slots >= fatal_skb_slots)) {
255 netdev_err(queue->vif->dev,
256 "Malicious frontend using %d slots, threshold %u\n",
257 slots, fatal_skb_slots);
258 xenvif_fatal_tx_err(queue->vif);
262 /* Xen network protocol had implicit dependency on
263 * MAX_SKB_FRAGS. XEN_NETBK_LEGACY_SLOTS_MAX is set to
264 * the historical MAX_SKB_FRAGS value 18 to honor the
265 * same behavior as before. Any packet using more than
266 * 18 slots but less than fatal_skb_slots slots is
269 if (!drop_err && slots >= XEN_NETBK_LEGACY_SLOTS_MAX) {
271 netdev_dbg(queue->vif->dev,
272 "Too many slots (%d) exceeding limit (%d), dropping packet\n",
273 slots, XEN_NETBK_LEGACY_SLOTS_MAX);
280 RING_COPY_REQUEST(&queue->tx, cons + slots, txp);
282 /* If the guest submitted a frame >= 64 KiB then
283 * first->size overflowed and following slots will
284 * appear to be larger than the frame.
286 * This cannot be fatal error as there are buggy
287 * frontends that do this.
289 * Consume all slots and drop the packet.
291 if (!drop_err && txp->size > first->size) {
293 netdev_dbg(queue->vif->dev,
294 "Invalid tx request, slot size %u > remaining size %u\n",
295 txp->size, first->size);
299 first->size -= txp->size;
302 if (unlikely((txp->offset + txp->size) > XEN_PAGE_SIZE)) {
303 netdev_err(queue->vif->dev, "Cross page boundary, txp->offset: %u, size: %u\n",
304 txp->offset, txp->size);
305 xenvif_fatal_tx_err(queue->vif);
309 more_data = txp->flags & XEN_NETTXF_more_data;
317 xenvif_tx_err(queue, first, extra_count, cons + slots);
325 struct xenvif_tx_cb {
329 #define XENVIF_TX_CB(skb) ((struct xenvif_tx_cb *)(skb)->cb)
331 static inline void xenvif_tx_create_map_op(struct xenvif_queue *queue,
333 struct xen_netif_tx_request *txp,
334 unsigned int extra_count,
335 struct gnttab_map_grant_ref *mop)
337 queue->pages_to_map[mop-queue->tx_map_ops] = queue->mmap_pages[pending_idx];
338 gnttab_set_map_op(mop, idx_to_kaddr(queue, pending_idx),
339 GNTMAP_host_map | GNTMAP_readonly,
340 txp->gref, queue->vif->domid);
342 memcpy(&queue->pending_tx_info[pending_idx].req, txp,
344 queue->pending_tx_info[pending_idx].extra_count = extra_count;
347 static inline struct sk_buff *xenvif_alloc_skb(unsigned int size)
349 struct sk_buff *skb =
350 alloc_skb(size + NET_SKB_PAD + NET_IP_ALIGN,
351 GFP_ATOMIC | __GFP_NOWARN);
352 if (unlikely(skb == NULL))
355 /* Packets passed to netif_rx() must have some headroom. */
356 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
358 /* Initialize it here to avoid later surprises */
359 skb_shinfo(skb)->destructor_arg = NULL;
364 static struct gnttab_map_grant_ref *xenvif_get_requests(struct xenvif_queue *queue,
366 struct xen_netif_tx_request *txp,
367 struct gnttab_map_grant_ref *gop,
368 unsigned int frag_overflow,
369 struct sk_buff *nskb)
371 struct skb_shared_info *shinfo = skb_shinfo(skb);
372 skb_frag_t *frags = shinfo->frags;
373 u16 pending_idx = XENVIF_TX_CB(skb)->pending_idx;
375 pending_ring_idx_t index;
376 unsigned int nr_slots;
378 nr_slots = shinfo->nr_frags;
380 /* Skip first skb fragment if it is on same page as header fragment. */
381 start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx);
383 for (shinfo->nr_frags = start; shinfo->nr_frags < nr_slots;
384 shinfo->nr_frags++, txp++, gop++) {
385 index = pending_index(queue->pending_cons++);
386 pending_idx = queue->pending_ring[index];
387 xenvif_tx_create_map_op(queue, pending_idx, txp, 0, gop);
388 frag_set_pending_idx(&frags[shinfo->nr_frags], pending_idx);
393 shinfo = skb_shinfo(nskb);
394 frags = shinfo->frags;
396 for (shinfo->nr_frags = 0; shinfo->nr_frags < frag_overflow;
397 shinfo->nr_frags++, txp++, gop++) {
398 index = pending_index(queue->pending_cons++);
399 pending_idx = queue->pending_ring[index];
400 xenvif_tx_create_map_op(queue, pending_idx, txp, 0,
402 frag_set_pending_idx(&frags[shinfo->nr_frags],
406 skb_shinfo(skb)->frag_list = nskb;
412 static inline void xenvif_grant_handle_set(struct xenvif_queue *queue,
414 grant_handle_t handle)
416 if (unlikely(queue->grant_tx_handle[pending_idx] !=
417 NETBACK_INVALID_HANDLE)) {
418 netdev_err(queue->vif->dev,
419 "Trying to overwrite active handle! pending_idx: 0x%x\n",
423 queue->grant_tx_handle[pending_idx] = handle;
426 static inline void xenvif_grant_handle_reset(struct xenvif_queue *queue,
429 if (unlikely(queue->grant_tx_handle[pending_idx] ==
430 NETBACK_INVALID_HANDLE)) {
431 netdev_err(queue->vif->dev,
432 "Trying to unmap invalid handle! pending_idx: 0x%x\n",
436 queue->grant_tx_handle[pending_idx] = NETBACK_INVALID_HANDLE;
439 static int xenvif_tx_check_gop(struct xenvif_queue *queue,
441 struct gnttab_map_grant_ref **gopp_map,
442 struct gnttab_copy **gopp_copy)
444 struct gnttab_map_grant_ref *gop_map = *gopp_map;
445 u16 pending_idx = XENVIF_TX_CB(skb)->pending_idx;
446 /* This always points to the shinfo of the skb being checked, which
447 * could be either the first or the one on the frag_list
449 struct skb_shared_info *shinfo = skb_shinfo(skb);
450 /* If this is non-NULL, we are currently checking the frag_list skb, and
451 * this points to the shinfo of the first one
453 struct skb_shared_info *first_shinfo = NULL;
454 int nr_frags = shinfo->nr_frags;
455 const bool sharedslot = nr_frags &&
456 frag_get_pending_idx(&shinfo->frags[0]) == pending_idx;
459 /* Check status of header. */
460 err = (*gopp_copy)->status;
463 netdev_dbg(queue->vif->dev,
464 "Grant copy of header failed! status: %d pending_idx: %u ref: %u\n",
465 (*gopp_copy)->status,
467 (*gopp_copy)->source.u.ref);
468 /* The first frag might still have this slot mapped */
470 xenvif_idx_release(queue, pending_idx,
471 XEN_NETIF_RSP_ERROR);
476 for (i = 0; i < nr_frags; i++, gop_map++) {
479 pending_idx = frag_get_pending_idx(&shinfo->frags[i]);
481 /* Check error status: if okay then remember grant handle. */
482 newerr = gop_map->status;
484 if (likely(!newerr)) {
485 xenvif_grant_handle_set(queue,
488 /* Had a previous error? Invalidate this fragment. */
490 xenvif_idx_unmap(queue, pending_idx);
491 /* If the mapping of the first frag was OK, but
492 * the header's copy failed, and they are
493 * sharing a slot, send an error
495 if (i == 0 && !first_shinfo && sharedslot)
496 xenvif_idx_release(queue, pending_idx,
497 XEN_NETIF_RSP_ERROR);
499 xenvif_idx_release(queue, pending_idx,
505 /* Error on this fragment: respond to client with an error. */
507 netdev_dbg(queue->vif->dev,
508 "Grant map of %d. frag failed! status: %d pending_idx: %u ref: %u\n",
514 xenvif_idx_release(queue, pending_idx, XEN_NETIF_RSP_ERROR);
516 /* Not the first error? Preceding frags already invalidated. */
520 /* First error: if the header haven't shared a slot with the
521 * first frag, release it as well.
524 xenvif_idx_release(queue,
525 XENVIF_TX_CB(skb)->pending_idx,
528 /* Invalidate preceding fragments of this skb. */
529 for (j = 0; j < i; j++) {
530 pending_idx = frag_get_pending_idx(&shinfo->frags[j]);
531 xenvif_idx_unmap(queue, pending_idx);
532 xenvif_idx_release(queue, pending_idx,
536 /* And if we found the error while checking the frag_list, unmap
537 * the first skb's frags
540 for (j = 0; j < first_shinfo->nr_frags; j++) {
541 pending_idx = frag_get_pending_idx(&first_shinfo->frags[j]);
542 xenvif_idx_unmap(queue, pending_idx);
543 xenvif_idx_release(queue, pending_idx,
548 /* Remember the error: invalidate all subsequent fragments. */
552 if (skb_has_frag_list(skb) && !first_shinfo) {
553 first_shinfo = skb_shinfo(skb);
554 shinfo = skb_shinfo(skb_shinfo(skb)->frag_list);
555 nr_frags = shinfo->nr_frags;
564 static void xenvif_fill_frags(struct xenvif_queue *queue, struct sk_buff *skb)
566 struct skb_shared_info *shinfo = skb_shinfo(skb);
567 int nr_frags = shinfo->nr_frags;
569 u16 prev_pending_idx = INVALID_PENDING_IDX;
571 for (i = 0; i < nr_frags; i++) {
572 skb_frag_t *frag = shinfo->frags + i;
573 struct xen_netif_tx_request *txp;
577 pending_idx = frag_get_pending_idx(frag);
579 /* If this is not the first frag, chain it to the previous*/
580 if (prev_pending_idx == INVALID_PENDING_IDX)
581 skb_shinfo(skb)->destructor_arg =
582 &callback_param(queue, pending_idx);
584 callback_param(queue, prev_pending_idx).ctx =
585 &callback_param(queue, pending_idx);
587 callback_param(queue, pending_idx).ctx = NULL;
588 prev_pending_idx = pending_idx;
590 txp = &queue->pending_tx_info[pending_idx].req;
591 page = virt_to_page(idx_to_kaddr(queue, pending_idx));
592 __skb_fill_page_desc(skb, i, page, txp->offset, txp->size);
593 skb->len += txp->size;
594 skb->data_len += txp->size;
595 skb->truesize += txp->size;
597 /* Take an extra reference to offset network stack's put_page */
598 get_page(queue->mmap_pages[pending_idx]);
602 static int xenvif_get_extras(struct xenvif_queue *queue,
603 struct xen_netif_extra_info *extras,
604 unsigned int *extra_count,
607 struct xen_netif_extra_info extra;
608 RING_IDX cons = queue->tx.req_cons;
611 if (unlikely(work_to_do-- <= 0)) {
612 netdev_err(queue->vif->dev, "Missing extra info\n");
613 xenvif_fatal_tx_err(queue->vif);
617 RING_COPY_REQUEST(&queue->tx, cons, &extra);
619 queue->tx.req_cons = ++cons;
622 if (unlikely(!extra.type ||
623 extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
624 netdev_err(queue->vif->dev,
625 "Invalid extra type: %d\n", extra.type);
626 xenvif_fatal_tx_err(queue->vif);
630 memcpy(&extras[extra.type - 1], &extra, sizeof(extra));
631 } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE);
636 static int xenvif_set_skb_gso(struct xenvif *vif,
638 struct xen_netif_extra_info *gso)
640 if (!gso->u.gso.size) {
641 netdev_err(vif->dev, "GSO size must not be zero.\n");
642 xenvif_fatal_tx_err(vif);
646 switch (gso->u.gso.type) {
647 case XEN_NETIF_GSO_TYPE_TCPV4:
648 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
650 case XEN_NETIF_GSO_TYPE_TCPV6:
651 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
654 netdev_err(vif->dev, "Bad GSO type %d.\n", gso->u.gso.type);
655 xenvif_fatal_tx_err(vif);
659 skb_shinfo(skb)->gso_size = gso->u.gso.size;
660 /* gso_segs will be calculated later */
665 static int checksum_setup(struct xenvif_queue *queue, struct sk_buff *skb)
667 bool recalculate_partial_csum = false;
669 /* A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
670 * peers can fail to set NETRXF_csum_blank when sending a GSO
671 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
672 * recalculate the partial checksum.
674 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
675 queue->stats.rx_gso_checksum_fixup++;
676 skb->ip_summed = CHECKSUM_PARTIAL;
677 recalculate_partial_csum = true;
680 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
681 if (skb->ip_summed != CHECKSUM_PARTIAL)
684 return skb_checksum_setup(skb, recalculate_partial_csum);
687 static bool tx_credit_exceeded(struct xenvif_queue *queue, unsigned size)
689 u64 now = get_jiffies_64();
690 u64 next_credit = queue->credit_window_start +
691 msecs_to_jiffies(queue->credit_usec / 1000);
693 /* Timer could already be pending in rare cases. */
694 if (timer_pending(&queue->credit_timeout)) {
695 queue->rate_limited = true;
699 /* Passed the point where we can replenish credit? */
700 if (time_after_eq64(now, next_credit)) {
701 queue->credit_window_start = now;
702 tx_add_credit(queue);
705 /* Still too big to send right now? Set a callback. */
706 if (size > queue->remaining_credit) {
707 queue->credit_timeout.data =
708 (unsigned long)queue;
709 mod_timer(&queue->credit_timeout,
711 queue->credit_window_start = next_credit;
712 queue->rate_limited = true;
720 /* No locking is required in xenvif_mcast_add/del() as they are
721 * only ever invoked from NAPI poll. An RCU list is used because
722 * xenvif_mcast_match() is called asynchronously, during start_xmit.
725 static int xenvif_mcast_add(struct xenvif *vif, const u8 *addr)
727 struct xenvif_mcast_addr *mcast;
729 if (vif->fe_mcast_count == XEN_NETBK_MCAST_MAX) {
732 "Too many multicast addresses\n");
736 mcast = kzalloc(sizeof(*mcast), GFP_ATOMIC);
740 ether_addr_copy(mcast->addr, addr);
741 list_add_tail_rcu(&mcast->entry, &vif->fe_mcast_addr);
742 vif->fe_mcast_count++;
747 static void xenvif_mcast_del(struct xenvif *vif, const u8 *addr)
749 struct xenvif_mcast_addr *mcast;
751 list_for_each_entry_rcu(mcast, &vif->fe_mcast_addr, entry) {
752 if (ether_addr_equal(addr, mcast->addr)) {
753 --vif->fe_mcast_count;
754 list_del_rcu(&mcast->entry);
755 kfree_rcu(mcast, rcu);
761 bool xenvif_mcast_match(struct xenvif *vif, const u8 *addr)
763 struct xenvif_mcast_addr *mcast;
766 list_for_each_entry_rcu(mcast, &vif->fe_mcast_addr, entry) {
767 if (ether_addr_equal(addr, mcast->addr)) {
777 void xenvif_mcast_addr_list_free(struct xenvif *vif)
779 /* No need for locking or RCU here. NAPI poll and TX queue
782 while (!list_empty(&vif->fe_mcast_addr)) {
783 struct xenvif_mcast_addr *mcast;
785 mcast = list_first_entry(&vif->fe_mcast_addr,
786 struct xenvif_mcast_addr,
788 --vif->fe_mcast_count;
789 list_del(&mcast->entry);
794 static void xenvif_tx_build_gops(struct xenvif_queue *queue,
799 struct gnttab_map_grant_ref *gop = queue->tx_map_ops;
800 struct sk_buff *skb, *nskb;
802 unsigned int frag_overflow;
804 while (skb_queue_len(&queue->tx_queue) < budget) {
805 struct xen_netif_tx_request txreq;
806 struct xen_netif_tx_request txfrags[XEN_NETBK_LEGACY_SLOTS_MAX];
807 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX-1];
808 unsigned int extra_count;
812 unsigned int data_len;
813 pending_ring_idx_t index;
815 if (queue->tx.sring->req_prod - queue->tx.req_cons >
816 XEN_NETIF_TX_RING_SIZE) {
817 netdev_err(queue->vif->dev,
818 "Impossible number of requests. "
819 "req_prod %d, req_cons %d, size %ld\n",
820 queue->tx.sring->req_prod, queue->tx.req_cons,
821 XEN_NETIF_TX_RING_SIZE);
822 xenvif_fatal_tx_err(queue->vif);
826 work_to_do = RING_HAS_UNCONSUMED_REQUESTS(&queue->tx);
830 idx = queue->tx.req_cons;
831 rmb(); /* Ensure that we see the request before we copy it. */
832 RING_COPY_REQUEST(&queue->tx, idx, &txreq);
834 /* Credit-based scheduling. */
835 if (txreq.size > queue->remaining_credit &&
836 tx_credit_exceeded(queue, txreq.size))
839 queue->remaining_credit -= txreq.size;
842 queue->tx.req_cons = ++idx;
844 memset(extras, 0, sizeof(extras));
846 if (txreq.flags & XEN_NETTXF_extra_info) {
847 work_to_do = xenvif_get_extras(queue, extras,
850 idx = queue->tx.req_cons;
851 if (unlikely(work_to_do < 0))
855 if (extras[XEN_NETIF_EXTRA_TYPE_MCAST_ADD - 1].type) {
856 struct xen_netif_extra_info *extra;
858 extra = &extras[XEN_NETIF_EXTRA_TYPE_MCAST_ADD - 1];
859 ret = xenvif_mcast_add(queue->vif, extra->u.mcast.addr);
861 make_tx_response(queue, &txreq, extra_count,
864 XEN_NETIF_RSP_ERROR);
865 push_tx_responses(queue);
869 if (extras[XEN_NETIF_EXTRA_TYPE_MCAST_DEL - 1].type) {
870 struct xen_netif_extra_info *extra;
872 extra = &extras[XEN_NETIF_EXTRA_TYPE_MCAST_DEL - 1];
873 xenvif_mcast_del(queue->vif, extra->u.mcast.addr);
875 make_tx_response(queue, &txreq, extra_count,
877 push_tx_responses(queue);
881 ret = xenvif_count_requests(queue, &txreq, extra_count,
882 txfrags, work_to_do);
883 if (unlikely(ret < 0))
888 if (unlikely(txreq.size < ETH_HLEN)) {
889 netdev_dbg(queue->vif->dev,
890 "Bad packet size: %d\n", txreq.size);
891 xenvif_tx_err(queue, &txreq, extra_count, idx);
895 /* No crossing a page as the payload mustn't fragment. */
896 if (unlikely((txreq.offset + txreq.size) > XEN_PAGE_SIZE)) {
897 netdev_err(queue->vif->dev,
898 "txreq.offset: %u, size: %u, end: %lu\n",
899 txreq.offset, txreq.size,
900 (unsigned long)(txreq.offset&~XEN_PAGE_MASK) + txreq.size);
901 xenvif_fatal_tx_err(queue->vif);
905 index = pending_index(queue->pending_cons);
906 pending_idx = queue->pending_ring[index];
908 data_len = (txreq.size > XEN_NETBACK_TX_COPY_LEN &&
909 ret < XEN_NETBK_LEGACY_SLOTS_MAX) ?
910 XEN_NETBACK_TX_COPY_LEN : txreq.size;
912 skb = xenvif_alloc_skb(data_len);
913 if (unlikely(skb == NULL)) {
914 netdev_dbg(queue->vif->dev,
915 "Can't allocate a skb in start_xmit.\n");
916 xenvif_tx_err(queue, &txreq, extra_count, idx);
920 skb_shinfo(skb)->nr_frags = ret;
921 if (data_len < txreq.size)
922 skb_shinfo(skb)->nr_frags++;
923 /* At this point shinfo->nr_frags is in fact the number of
924 * slots, which can be as large as XEN_NETBK_LEGACY_SLOTS_MAX.
928 if (skb_shinfo(skb)->nr_frags > MAX_SKB_FRAGS) {
929 frag_overflow = skb_shinfo(skb)->nr_frags - MAX_SKB_FRAGS;
930 BUG_ON(frag_overflow > MAX_SKB_FRAGS);
931 skb_shinfo(skb)->nr_frags = MAX_SKB_FRAGS;
932 nskb = xenvif_alloc_skb(0);
933 if (unlikely(nskb == NULL)) {
934 skb_shinfo(skb)->nr_frags = 0;
936 xenvif_tx_err(queue, &txreq, extra_count, idx);
938 netdev_err(queue->vif->dev,
939 "Can't allocate the frag_list skb.\n");
944 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
945 struct xen_netif_extra_info *gso;
946 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
948 if (xenvif_set_skb_gso(queue->vif, skb, gso)) {
949 /* Failure in xenvif_set_skb_gso is fatal. */
950 skb_shinfo(skb)->nr_frags = 0;
957 if (extras[XEN_NETIF_EXTRA_TYPE_HASH - 1].type) {
958 struct xen_netif_extra_info *extra;
959 enum pkt_hash_types type = PKT_HASH_TYPE_NONE;
961 extra = &extras[XEN_NETIF_EXTRA_TYPE_HASH - 1];
963 switch (extra->u.hash.type) {
964 case _XEN_NETIF_CTRL_HASH_TYPE_IPV4:
965 case _XEN_NETIF_CTRL_HASH_TYPE_IPV6:
966 type = PKT_HASH_TYPE_L3;
969 case _XEN_NETIF_CTRL_HASH_TYPE_IPV4_TCP:
970 case _XEN_NETIF_CTRL_HASH_TYPE_IPV6_TCP:
971 type = PKT_HASH_TYPE_L4;
978 if (type != PKT_HASH_TYPE_NONE)
980 *(u32 *)extra->u.hash.value,
984 XENVIF_TX_CB(skb)->pending_idx = pending_idx;
986 __skb_put(skb, data_len);
987 queue->tx_copy_ops[*copy_ops].source.u.ref = txreq.gref;
988 queue->tx_copy_ops[*copy_ops].source.domid = queue->vif->domid;
989 queue->tx_copy_ops[*copy_ops].source.offset = txreq.offset;
991 queue->tx_copy_ops[*copy_ops].dest.u.gmfn =
992 virt_to_gfn(skb->data);
993 queue->tx_copy_ops[*copy_ops].dest.domid = DOMID_SELF;
994 queue->tx_copy_ops[*copy_ops].dest.offset =
995 offset_in_page(skb->data) & ~XEN_PAGE_MASK;
997 queue->tx_copy_ops[*copy_ops].len = data_len;
998 queue->tx_copy_ops[*copy_ops].flags = GNTCOPY_source_gref;
1002 if (data_len < txreq.size) {
1003 frag_set_pending_idx(&skb_shinfo(skb)->frags[0],
1005 xenvif_tx_create_map_op(queue, pending_idx, &txreq,
1009 frag_set_pending_idx(&skb_shinfo(skb)->frags[0],
1010 INVALID_PENDING_IDX);
1011 memcpy(&queue->pending_tx_info[pending_idx].req,
1012 &txreq, sizeof(txreq));
1013 queue->pending_tx_info[pending_idx].extra_count =
1017 queue->pending_cons++;
1019 gop = xenvif_get_requests(queue, skb, txfrags, gop,
1020 frag_overflow, nskb);
1022 __skb_queue_tail(&queue->tx_queue, skb);
1024 queue->tx.req_cons = idx;
1026 if (((gop-queue->tx_map_ops) >= ARRAY_SIZE(queue->tx_map_ops)) ||
1027 (*copy_ops >= ARRAY_SIZE(queue->tx_copy_ops)))
1031 (*map_ops) = gop - queue->tx_map_ops;
1035 /* Consolidate skb with a frag_list into a brand new one with local pages on
1036 * frags. Returns 0 or -ENOMEM if can't allocate new pages.
1038 static int xenvif_handle_frag_list(struct xenvif_queue *queue, struct sk_buff *skb)
1040 unsigned int offset = skb_headlen(skb);
1041 skb_frag_t frags[MAX_SKB_FRAGS];
1043 struct ubuf_info *uarg;
1044 struct sk_buff *nskb = skb_shinfo(skb)->frag_list;
1046 queue->stats.tx_zerocopy_sent += 2;
1047 queue->stats.tx_frag_overflow++;
1049 xenvif_fill_frags(queue, nskb);
1050 /* Subtract frags size, we will correct it later */
1051 skb->truesize -= skb->data_len;
1052 skb->len += nskb->len;
1053 skb->data_len += nskb->len;
1055 /* create a brand new frags array and coalesce there */
1056 for (i = 0; offset < skb->len; i++) {
1060 BUG_ON(i >= MAX_SKB_FRAGS);
1061 page = alloc_page(GFP_ATOMIC);
1064 skb->truesize += skb->data_len;
1065 for (j = 0; j < i; j++)
1066 put_page(frags[j].page.p);
1070 if (offset + PAGE_SIZE < skb->len)
1073 len = skb->len - offset;
1074 if (skb_copy_bits(skb, offset, page_address(page), len))
1078 frags[i].page.p = page;
1079 frags[i].page_offset = 0;
1080 skb_frag_size_set(&frags[i], len);
1083 /* Release all the original (foreign) frags. */
1084 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
1085 skb_frag_unref(skb, f);
1086 uarg = skb_shinfo(skb)->destructor_arg;
1087 /* increase inflight counter to offset decrement in callback */
1088 atomic_inc(&queue->inflight_packets);
1089 uarg->callback(uarg, true);
1090 skb_shinfo(skb)->destructor_arg = NULL;
1092 /* Fill the skb with the new (local) frags. */
1093 memcpy(skb_shinfo(skb)->frags, frags, i * sizeof(skb_frag_t));
1094 skb_shinfo(skb)->nr_frags = i;
1095 skb->truesize += i * PAGE_SIZE;
1100 static int xenvif_tx_submit(struct xenvif_queue *queue)
1102 struct gnttab_map_grant_ref *gop_map = queue->tx_map_ops;
1103 struct gnttab_copy *gop_copy = queue->tx_copy_ops;
1104 struct sk_buff *skb;
1107 while ((skb = __skb_dequeue(&queue->tx_queue)) != NULL) {
1108 struct xen_netif_tx_request *txp;
1112 pending_idx = XENVIF_TX_CB(skb)->pending_idx;
1113 txp = &queue->pending_tx_info[pending_idx].req;
1115 /* Check the remap error code. */
1116 if (unlikely(xenvif_tx_check_gop(queue, skb, &gop_map, &gop_copy))) {
1117 /* If there was an error, xenvif_tx_check_gop is
1118 * expected to release all the frags which were mapped,
1119 * so kfree_skb shouldn't do it again
1121 skb_shinfo(skb)->nr_frags = 0;
1122 if (skb_has_frag_list(skb)) {
1123 struct sk_buff *nskb =
1124 skb_shinfo(skb)->frag_list;
1125 skb_shinfo(nskb)->nr_frags = 0;
1131 data_len = skb->len;
1132 callback_param(queue, pending_idx).ctx = NULL;
1133 if (data_len < txp->size) {
1134 /* Append the packet payload as a fragment. */
1135 txp->offset += data_len;
1136 txp->size -= data_len;
1138 /* Schedule a response immediately. */
1139 xenvif_idx_release(queue, pending_idx,
1140 XEN_NETIF_RSP_OKAY);
1143 if (txp->flags & XEN_NETTXF_csum_blank)
1144 skb->ip_summed = CHECKSUM_PARTIAL;
1145 else if (txp->flags & XEN_NETTXF_data_validated)
1146 skb->ip_summed = CHECKSUM_UNNECESSARY;
1148 xenvif_fill_frags(queue, skb);
1150 if (unlikely(skb_has_frag_list(skb))) {
1151 struct sk_buff *nskb = skb_shinfo(skb)->frag_list;
1152 xenvif_skb_zerocopy_prepare(queue, nskb);
1153 if (xenvif_handle_frag_list(queue, skb)) {
1154 if (net_ratelimit())
1155 netdev_err(queue->vif->dev,
1156 "Not enough memory to consolidate frag_list!\n");
1157 xenvif_skb_zerocopy_prepare(queue, skb);
1161 /* Copied all the bits from the frag list -- free it. */
1162 skb_frag_list_init(skb);
1166 skb->dev = queue->vif->dev;
1167 skb->protocol = eth_type_trans(skb, skb->dev);
1168 skb_reset_network_header(skb);
1170 if (checksum_setup(queue, skb)) {
1171 netdev_dbg(queue->vif->dev,
1172 "Can't setup checksum in net_tx_action\n");
1173 /* We have to set this flag to trigger the callback */
1174 if (skb_shinfo(skb)->destructor_arg)
1175 xenvif_skb_zerocopy_prepare(queue, skb);
1180 skb_probe_transport_header(skb, 0);
1182 /* If the packet is GSO then we will have just set up the
1183 * transport header offset in checksum_setup so it's now
1184 * straightforward to calculate gso_segs.
1186 if (skb_is_gso(skb)) {
1187 int mss = skb_shinfo(skb)->gso_size;
1188 int hdrlen = skb_transport_header(skb) -
1189 skb_mac_header(skb) +
1192 skb_shinfo(skb)->gso_segs =
1193 DIV_ROUND_UP(skb->len - hdrlen, mss);
1196 queue->stats.rx_bytes += skb->len;
1197 queue->stats.rx_packets++;
1201 /* Set this flag right before netif_receive_skb, otherwise
1202 * someone might think this packet already left netback, and
1203 * do a skb_copy_ubufs while we are still in control of the
1204 * skb. E.g. the __pskb_pull_tail earlier can do such thing.
1206 if (skb_shinfo(skb)->destructor_arg) {
1207 xenvif_skb_zerocopy_prepare(queue, skb);
1208 queue->stats.tx_zerocopy_sent++;
1211 netif_receive_skb(skb);
1217 void xenvif_zerocopy_callback(struct ubuf_info *ubuf, bool zerocopy_success)
1219 unsigned long flags;
1220 pending_ring_idx_t index;
1221 struct xenvif_queue *queue = ubuf_to_queue(ubuf);
1223 /* This is the only place where we grab this lock, to protect callbacks
1226 spin_lock_irqsave(&queue->callback_lock, flags);
1228 u16 pending_idx = ubuf->desc;
1229 ubuf = (struct ubuf_info *) ubuf->ctx;
1230 BUG_ON(queue->dealloc_prod - queue->dealloc_cons >=
1232 index = pending_index(queue->dealloc_prod);
1233 queue->dealloc_ring[index] = pending_idx;
1234 /* Sync with xenvif_tx_dealloc_action:
1235 * insert idx then incr producer.
1238 queue->dealloc_prod++;
1240 spin_unlock_irqrestore(&queue->callback_lock, flags);
1242 if (likely(zerocopy_success))
1243 queue->stats.tx_zerocopy_success++;
1245 queue->stats.tx_zerocopy_fail++;
1246 xenvif_skb_zerocopy_complete(queue);
1249 static inline void xenvif_tx_dealloc_action(struct xenvif_queue *queue)
1251 struct gnttab_unmap_grant_ref *gop;
1252 pending_ring_idx_t dc, dp;
1253 u16 pending_idx, pending_idx_release[MAX_PENDING_REQS];
1256 dc = queue->dealloc_cons;
1257 gop = queue->tx_unmap_ops;
1259 /* Free up any grants we have finished using */
1261 dp = queue->dealloc_prod;
1263 /* Ensure we see all indices enqueued by all
1264 * xenvif_zerocopy_callback().
1269 BUG_ON(gop - queue->tx_unmap_ops >= MAX_PENDING_REQS);
1271 queue->dealloc_ring[pending_index(dc++)];
1273 pending_idx_release[gop - queue->tx_unmap_ops] =
1275 queue->pages_to_unmap[gop - queue->tx_unmap_ops] =
1276 queue->mmap_pages[pending_idx];
1277 gnttab_set_unmap_op(gop,
1278 idx_to_kaddr(queue, pending_idx),
1280 queue->grant_tx_handle[pending_idx]);
1281 xenvif_grant_handle_reset(queue, pending_idx);
1285 } while (dp != queue->dealloc_prod);
1287 queue->dealloc_cons = dc;
1289 if (gop - queue->tx_unmap_ops > 0) {
1291 ret = gnttab_unmap_refs(queue->tx_unmap_ops,
1293 queue->pages_to_unmap,
1294 gop - queue->tx_unmap_ops);
1296 netdev_err(queue->vif->dev, "Unmap fail: nr_ops %tu ret %d\n",
1297 gop - queue->tx_unmap_ops, ret);
1298 for (i = 0; i < gop - queue->tx_unmap_ops; ++i) {
1299 if (gop[i].status != GNTST_okay)
1300 netdev_err(queue->vif->dev,
1301 " host_addr: 0x%llx handle: 0x%x status: %d\n",
1310 for (i = 0; i < gop - queue->tx_unmap_ops; ++i)
1311 xenvif_idx_release(queue, pending_idx_release[i],
1312 XEN_NETIF_RSP_OKAY);
1316 /* Called after netfront has transmitted */
1317 int xenvif_tx_action(struct xenvif_queue *queue, int budget)
1319 unsigned nr_mops, nr_cops = 0;
1322 if (unlikely(!tx_work_todo(queue)))
1325 xenvif_tx_build_gops(queue, budget, &nr_cops, &nr_mops);
1330 gnttab_batch_copy(queue->tx_copy_ops, nr_cops);
1332 ret = gnttab_map_refs(queue->tx_map_ops,
1334 queue->pages_to_map,
1339 netdev_err(queue->vif->dev, "Map fail: nr %u ret %d\n",
1341 for (i = 0; i < nr_mops; ++i)
1342 WARN_ON_ONCE(queue->tx_map_ops[i].status ==
1347 work_done = xenvif_tx_submit(queue);
1352 static void xenvif_idx_release(struct xenvif_queue *queue, u16 pending_idx,
1355 struct pending_tx_info *pending_tx_info;
1356 pending_ring_idx_t index;
1357 unsigned long flags;
1359 pending_tx_info = &queue->pending_tx_info[pending_idx];
1361 spin_lock_irqsave(&queue->response_lock, flags);
1363 make_tx_response(queue, &pending_tx_info->req,
1364 pending_tx_info->extra_count, status);
1366 /* Release the pending index before pusing the Tx response so
1367 * its available before a new Tx request is pushed by the
1370 index = pending_index(queue->pending_prod++);
1371 queue->pending_ring[index] = pending_idx;
1373 push_tx_responses(queue);
1375 spin_unlock_irqrestore(&queue->response_lock, flags);
1379 static void make_tx_response(struct xenvif_queue *queue,
1380 struct xen_netif_tx_request *txp,
1381 unsigned int extra_count,
1384 RING_IDX i = queue->tx.rsp_prod_pvt;
1385 struct xen_netif_tx_response *resp;
1387 resp = RING_GET_RESPONSE(&queue->tx, i);
1391 while (extra_count-- != 0)
1392 RING_GET_RESPONSE(&queue->tx, ++i)->status = XEN_NETIF_RSP_NULL;
1394 queue->tx.rsp_prod_pvt = ++i;
1397 static void push_tx_responses(struct xenvif_queue *queue)
1401 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&queue->tx, notify);
1403 notify_remote_via_irq(queue->tx_irq);
1406 void xenvif_idx_unmap(struct xenvif_queue *queue, u16 pending_idx)
1409 struct gnttab_unmap_grant_ref tx_unmap_op;
1411 gnttab_set_unmap_op(&tx_unmap_op,
1412 idx_to_kaddr(queue, pending_idx),
1414 queue->grant_tx_handle[pending_idx]);
1415 xenvif_grant_handle_reset(queue, pending_idx);
1417 ret = gnttab_unmap_refs(&tx_unmap_op, NULL,
1418 &queue->mmap_pages[pending_idx], 1);
1420 netdev_err(queue->vif->dev,
1421 "Unmap fail: ret: %d pending_idx: %d host_addr: %llx handle: 0x%x status: %d\n",
1424 tx_unmap_op.host_addr,
1426 tx_unmap_op.status);
1431 static inline int tx_work_todo(struct xenvif_queue *queue)
1433 if (likely(RING_HAS_UNCONSUMED_REQUESTS(&queue->tx)))
1439 static inline bool tx_dealloc_work_todo(struct xenvif_queue *queue)
1441 return queue->dealloc_cons != queue->dealloc_prod;
1444 void xenvif_unmap_frontend_data_rings(struct xenvif_queue *queue)
1446 if (queue->tx.sring)
1447 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(queue->vif),
1449 if (queue->rx.sring)
1450 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(queue->vif),
1454 int xenvif_map_frontend_data_rings(struct xenvif_queue *queue,
1455 grant_ref_t tx_ring_ref,
1456 grant_ref_t rx_ring_ref)
1459 struct xen_netif_tx_sring *txs;
1460 struct xen_netif_rx_sring *rxs;
1464 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(queue->vif),
1465 &tx_ring_ref, 1, &addr);
1469 txs = (struct xen_netif_tx_sring *)addr;
1470 BACK_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE);
1472 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(queue->vif),
1473 &rx_ring_ref, 1, &addr);
1477 rxs = (struct xen_netif_rx_sring *)addr;
1478 BACK_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE);
1483 xenvif_unmap_frontend_data_rings(queue);
1487 static bool xenvif_dealloc_kthread_should_stop(struct xenvif_queue *queue)
1489 /* Dealloc thread must remain running until all inflight
1492 return kthread_should_stop() &&
1493 !atomic_read(&queue->inflight_packets);
1496 int xenvif_dealloc_kthread(void *data)
1498 struct xenvif_queue *queue = data;
1501 wait_event_interruptible(queue->dealloc_wq,
1502 tx_dealloc_work_todo(queue) ||
1503 xenvif_dealloc_kthread_should_stop(queue));
1504 if (xenvif_dealloc_kthread_should_stop(queue))
1507 xenvif_tx_dealloc_action(queue);
1511 /* Unmap anything remaining*/
1512 if (tx_dealloc_work_todo(queue))
1513 xenvif_tx_dealloc_action(queue);
1518 static void make_ctrl_response(struct xenvif *vif,
1519 const struct xen_netif_ctrl_request *req,
1520 u32 status, u32 data)
1522 RING_IDX idx = vif->ctrl.rsp_prod_pvt;
1523 struct xen_netif_ctrl_response rsp = {
1530 *RING_GET_RESPONSE(&vif->ctrl, idx) = rsp;
1531 vif->ctrl.rsp_prod_pvt = ++idx;
1534 static void push_ctrl_response(struct xenvif *vif)
1538 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->ctrl, notify);
1540 notify_remote_via_irq(vif->ctrl_irq);
1543 static void process_ctrl_request(struct xenvif *vif,
1544 const struct xen_netif_ctrl_request *req)
1546 u32 status = XEN_NETIF_CTRL_STATUS_NOT_SUPPORTED;
1549 switch (req->type) {
1550 case XEN_NETIF_CTRL_TYPE_SET_HASH_ALGORITHM:
1551 status = xenvif_set_hash_alg(vif, req->data[0]);
1554 case XEN_NETIF_CTRL_TYPE_GET_HASH_FLAGS:
1555 status = xenvif_get_hash_flags(vif, &data);
1558 case XEN_NETIF_CTRL_TYPE_SET_HASH_FLAGS:
1559 status = xenvif_set_hash_flags(vif, req->data[0]);
1562 case XEN_NETIF_CTRL_TYPE_SET_HASH_KEY:
1563 status = xenvif_set_hash_key(vif, req->data[0],
1567 case XEN_NETIF_CTRL_TYPE_GET_HASH_MAPPING_SIZE:
1568 status = XEN_NETIF_CTRL_STATUS_SUCCESS;
1569 data = XEN_NETBK_MAX_HASH_MAPPING_SIZE;
1572 case XEN_NETIF_CTRL_TYPE_SET_HASH_MAPPING_SIZE:
1573 status = xenvif_set_hash_mapping_size(vif,
1577 case XEN_NETIF_CTRL_TYPE_SET_HASH_MAPPING:
1578 status = xenvif_set_hash_mapping(vif, req->data[0],
1587 make_ctrl_response(vif, req, status, data);
1588 push_ctrl_response(vif);
1591 static void xenvif_ctrl_action(struct xenvif *vif)
1594 RING_IDX req_prod, req_cons;
1596 req_prod = vif->ctrl.sring->req_prod;
1597 req_cons = vif->ctrl.req_cons;
1599 /* Make sure we can see requests before we process them. */
1602 if (req_cons == req_prod)
1605 while (req_cons != req_prod) {
1606 struct xen_netif_ctrl_request req;
1608 RING_COPY_REQUEST(&vif->ctrl, req_cons, &req);
1611 process_ctrl_request(vif, &req);
1614 vif->ctrl.req_cons = req_cons;
1615 vif->ctrl.sring->req_event = req_cons + 1;
1619 static bool xenvif_ctrl_work_todo(struct xenvif *vif)
1621 if (likely(RING_HAS_UNCONSUMED_REQUESTS(&vif->ctrl)))
1627 irqreturn_t xenvif_ctrl_irq_fn(int irq, void *data)
1629 struct xenvif *vif = data;
1630 unsigned int eoi_flag = XEN_EOI_FLAG_SPURIOUS;
1632 while (xenvif_ctrl_work_todo(vif)) {
1633 xenvif_ctrl_action(vif);
1637 xen_irq_lateeoi(irq, eoi_flag);
1642 static int __init netback_init(void)
1649 /* Allow as many queues as there are CPUs but max. 8 if user has not
1650 * specified a value.
1652 if (xenvif_max_queues == 0)
1653 xenvif_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT,
1656 if (fatal_skb_slots < XEN_NETBK_LEGACY_SLOTS_MAX) {
1657 pr_info("fatal_skb_slots too small (%d), bump it to XEN_NETBK_LEGACY_SLOTS_MAX (%d)\n",
1658 fatal_skb_slots, XEN_NETBK_LEGACY_SLOTS_MAX);
1659 fatal_skb_slots = XEN_NETBK_LEGACY_SLOTS_MAX;
1662 rc = xenvif_xenbus_init();
1666 #ifdef CONFIG_DEBUG_FS
1667 xen_netback_dbg_root = debugfs_create_dir("xen-netback", NULL);
1668 if (IS_ERR_OR_NULL(xen_netback_dbg_root))
1669 pr_warn("Init of debugfs returned %ld!\n",
1670 PTR_ERR(xen_netback_dbg_root));
1671 #endif /* CONFIG_DEBUG_FS */
1679 module_init(netback_init);
1681 static void __exit netback_fini(void)
1683 #ifdef CONFIG_DEBUG_FS
1684 if (!IS_ERR_OR_NULL(xen_netback_dbg_root))
1685 debugfs_remove_recursive(xen_netback_dbg_root);
1686 #endif /* CONFIG_DEBUG_FS */
1687 xenvif_xenbus_fini();
1689 module_exit(netback_fini);
1691 MODULE_LICENSE("Dual BSD/GPL");
1692 MODULE_ALIAS("xen-backend:vif");