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 /* The module parameter tells that we have to put data
100 * for xen-netfront with the XDP_PACKET_HEADROOM offset
101 * needed for XDP processing
103 bool provides_xdp_headroom = true;
104 module_param(provides_xdp_headroom, bool, 0644);
106 static void xenvif_idx_release(struct xenvif_queue *queue, u16 pending_idx,
109 static void make_tx_response(struct xenvif_queue *queue,
110 struct xen_netif_tx_request *txp,
111 unsigned int extra_count,
113 static void push_tx_responses(struct xenvif_queue *queue);
115 static void xenvif_idx_unmap(struct xenvif_queue *queue, u16 pending_idx);
117 static inline int tx_work_todo(struct xenvif_queue *queue);
119 static inline unsigned long idx_to_pfn(struct xenvif_queue *queue,
122 return page_to_pfn(queue->mmap_pages[idx]);
125 static inline unsigned long idx_to_kaddr(struct xenvif_queue *queue,
128 return (unsigned long)pfn_to_kaddr(idx_to_pfn(queue, idx));
131 #define callback_param(vif, pending_idx) \
132 (vif->pending_tx_info[pending_idx].callback_struct)
134 /* Find the containing VIF's structure from a pointer in pending_tx_info array
136 static inline struct xenvif_queue *ubuf_to_queue(const struct ubuf_info *ubuf)
138 u16 pending_idx = ubuf->desc;
139 struct pending_tx_info *temp =
140 container_of(ubuf, struct pending_tx_info, callback_struct);
141 return container_of(temp - pending_idx,
146 static u16 frag_get_pending_idx(skb_frag_t *frag)
148 return (u16)skb_frag_off(frag);
151 static void frag_set_pending_idx(skb_frag_t *frag, u16 pending_idx)
153 skb_frag_off_set(frag, pending_idx);
156 static inline pending_ring_idx_t pending_index(unsigned i)
158 return i & (MAX_PENDING_REQS-1);
161 void xenvif_kick_thread(struct xenvif_queue *queue)
166 void xenvif_napi_schedule_or_enable_events(struct xenvif_queue *queue)
170 RING_FINAL_CHECK_FOR_REQUESTS(&queue->tx, more_to_do);
173 napi_schedule(&queue->napi);
174 else if (atomic_fetch_andnot(NETBK_TX_EOI | NETBK_COMMON_EOI,
175 &queue->eoi_pending) &
176 (NETBK_TX_EOI | NETBK_COMMON_EOI))
177 xen_irq_lateeoi(queue->tx_irq, 0);
180 static void tx_add_credit(struct xenvif_queue *queue)
182 unsigned long max_burst, max_credit;
185 * Allow a burst big enough to transmit a jumbo packet of up to 128kB.
186 * Otherwise the interface can seize up due to insufficient credit.
188 max_burst = max(131072UL, queue->credit_bytes);
190 /* Take care that adding a new chunk of credit doesn't wrap to zero. */
191 max_credit = queue->remaining_credit + queue->credit_bytes;
192 if (max_credit < queue->remaining_credit)
193 max_credit = ULONG_MAX; /* wrapped: clamp to ULONG_MAX */
195 queue->remaining_credit = min(max_credit, max_burst);
196 queue->rate_limited = false;
199 void xenvif_tx_credit_callback(struct timer_list *t)
201 struct xenvif_queue *queue = from_timer(queue, t, credit_timeout);
202 tx_add_credit(queue);
203 xenvif_napi_schedule_or_enable_events(queue);
206 static void xenvif_tx_err(struct xenvif_queue *queue,
207 struct xen_netif_tx_request *txp,
208 unsigned int extra_count, RING_IDX end)
210 RING_IDX cons = queue->tx.req_cons;
214 spin_lock_irqsave(&queue->response_lock, flags);
215 make_tx_response(queue, txp, extra_count, XEN_NETIF_RSP_ERROR);
216 push_tx_responses(queue);
217 spin_unlock_irqrestore(&queue->response_lock, flags);
220 RING_COPY_REQUEST(&queue->tx, cons++, txp);
221 extra_count = 0; /* only the first frag can have extras */
223 queue->tx.req_cons = cons;
226 static void xenvif_fatal_tx_err(struct xenvif *vif)
228 netdev_err(vif->dev, "fatal error; disabling device\n");
229 vif->disabled = true;
230 /* Disable the vif from queue 0's kthread */
232 xenvif_kick_thread(&vif->queues[0]);
235 static int xenvif_count_requests(struct xenvif_queue *queue,
236 struct xen_netif_tx_request *first,
237 unsigned int extra_count,
238 struct xen_netif_tx_request *txp,
241 RING_IDX cons = queue->tx.req_cons;
246 if (!(first->flags & XEN_NETTXF_more_data))
250 struct xen_netif_tx_request dropped_tx = { 0 };
252 if (slots >= work_to_do) {
253 netdev_err(queue->vif->dev,
254 "Asked for %d slots but exceeds this limit\n",
256 xenvif_fatal_tx_err(queue->vif);
260 /* This guest is really using too many slots and
261 * considered malicious.
263 if (unlikely(slots >= fatal_skb_slots)) {
264 netdev_err(queue->vif->dev,
265 "Malicious frontend using %d slots, threshold %u\n",
266 slots, fatal_skb_slots);
267 xenvif_fatal_tx_err(queue->vif);
271 /* Xen network protocol had implicit dependency on
272 * MAX_SKB_FRAGS. XEN_NETBK_LEGACY_SLOTS_MAX is set to
273 * the historical MAX_SKB_FRAGS value 18 to honor the
274 * same behavior as before. Any packet using more than
275 * 18 slots but less than fatal_skb_slots slots is
278 if (!drop_err && slots >= XEN_NETBK_LEGACY_SLOTS_MAX) {
280 netdev_dbg(queue->vif->dev,
281 "Too many slots (%d) exceeding limit (%d), dropping packet\n",
282 slots, XEN_NETBK_LEGACY_SLOTS_MAX);
289 RING_COPY_REQUEST(&queue->tx, cons + slots, txp);
291 /* If the guest submitted a frame >= 64 KiB then
292 * first->size overflowed and following slots will
293 * appear to be larger than the frame.
295 * This cannot be fatal error as there are buggy
296 * frontends that do this.
298 * Consume all slots and drop the packet.
300 if (!drop_err && txp->size > first->size) {
302 netdev_dbg(queue->vif->dev,
303 "Invalid tx request, slot size %u > remaining size %u\n",
304 txp->size, first->size);
308 first->size -= txp->size;
311 if (unlikely((txp->offset + txp->size) > XEN_PAGE_SIZE)) {
312 netdev_err(queue->vif->dev, "Cross page boundary, txp->offset: %u, size: %u\n",
313 txp->offset, txp->size);
314 xenvif_fatal_tx_err(queue->vif);
318 more_data = txp->flags & XEN_NETTXF_more_data;
326 xenvif_tx_err(queue, first, extra_count, cons + slots);
334 struct xenvif_tx_cb {
338 #define XENVIF_TX_CB(skb) ((struct xenvif_tx_cb *)(skb)->cb)
340 static inline void xenvif_tx_create_map_op(struct xenvif_queue *queue,
342 struct xen_netif_tx_request *txp,
343 unsigned int extra_count,
344 struct gnttab_map_grant_ref *mop)
346 queue->pages_to_map[mop-queue->tx_map_ops] = queue->mmap_pages[pending_idx];
347 gnttab_set_map_op(mop, idx_to_kaddr(queue, pending_idx),
348 GNTMAP_host_map | GNTMAP_readonly,
349 txp->gref, queue->vif->domid);
351 memcpy(&queue->pending_tx_info[pending_idx].req, txp,
353 queue->pending_tx_info[pending_idx].extra_count = extra_count;
356 static inline struct sk_buff *xenvif_alloc_skb(unsigned int size)
358 struct sk_buff *skb =
359 alloc_skb(size + NET_SKB_PAD + NET_IP_ALIGN,
360 GFP_ATOMIC | __GFP_NOWARN);
361 if (unlikely(skb == NULL))
364 /* Packets passed to netif_rx() must have some headroom. */
365 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
367 /* Initialize it here to avoid later surprises */
368 skb_shinfo(skb)->destructor_arg = NULL;
373 static struct gnttab_map_grant_ref *xenvif_get_requests(struct xenvif_queue *queue,
375 struct xen_netif_tx_request *txp,
376 struct gnttab_map_grant_ref *gop,
377 unsigned int frag_overflow,
378 struct sk_buff *nskb)
380 struct skb_shared_info *shinfo = skb_shinfo(skb);
381 skb_frag_t *frags = shinfo->frags;
382 u16 pending_idx = XENVIF_TX_CB(skb)->pending_idx;
384 pending_ring_idx_t index;
385 unsigned int nr_slots;
387 nr_slots = shinfo->nr_frags;
389 /* Skip first skb fragment if it is on same page as header fragment. */
390 start = (frag_get_pending_idx(&shinfo->frags[0]) == pending_idx);
392 for (shinfo->nr_frags = start; shinfo->nr_frags < nr_slots;
393 shinfo->nr_frags++, txp++, gop++) {
394 index = pending_index(queue->pending_cons++);
395 pending_idx = queue->pending_ring[index];
396 xenvif_tx_create_map_op(queue, pending_idx, txp, 0, gop);
397 frag_set_pending_idx(&frags[shinfo->nr_frags], pending_idx);
402 shinfo = skb_shinfo(nskb);
403 frags = shinfo->frags;
405 for (shinfo->nr_frags = 0; shinfo->nr_frags < frag_overflow;
406 shinfo->nr_frags++, txp++, gop++) {
407 index = pending_index(queue->pending_cons++);
408 pending_idx = queue->pending_ring[index];
409 xenvif_tx_create_map_op(queue, pending_idx, txp, 0,
411 frag_set_pending_idx(&frags[shinfo->nr_frags],
415 skb_shinfo(skb)->frag_list = nskb;
421 static inline void xenvif_grant_handle_set(struct xenvif_queue *queue,
423 grant_handle_t handle)
425 if (unlikely(queue->grant_tx_handle[pending_idx] !=
426 NETBACK_INVALID_HANDLE)) {
427 netdev_err(queue->vif->dev,
428 "Trying to overwrite active handle! pending_idx: 0x%x\n",
432 queue->grant_tx_handle[pending_idx] = handle;
435 static inline void xenvif_grant_handle_reset(struct xenvif_queue *queue,
438 if (unlikely(queue->grant_tx_handle[pending_idx] ==
439 NETBACK_INVALID_HANDLE)) {
440 netdev_err(queue->vif->dev,
441 "Trying to unmap invalid handle! pending_idx: 0x%x\n",
445 queue->grant_tx_handle[pending_idx] = NETBACK_INVALID_HANDLE;
448 static int xenvif_tx_check_gop(struct xenvif_queue *queue,
450 struct gnttab_map_grant_ref **gopp_map,
451 struct gnttab_copy **gopp_copy)
453 struct gnttab_map_grant_ref *gop_map = *gopp_map;
454 u16 pending_idx = XENVIF_TX_CB(skb)->pending_idx;
455 /* This always points to the shinfo of the skb being checked, which
456 * could be either the first or the one on the frag_list
458 struct skb_shared_info *shinfo = skb_shinfo(skb);
459 /* If this is non-NULL, we are currently checking the frag_list skb, and
460 * this points to the shinfo of the first one
462 struct skb_shared_info *first_shinfo = NULL;
463 int nr_frags = shinfo->nr_frags;
464 const bool sharedslot = nr_frags &&
465 frag_get_pending_idx(&shinfo->frags[0]) == pending_idx;
468 /* Check status of header. */
469 err = (*gopp_copy)->status;
472 netdev_dbg(queue->vif->dev,
473 "Grant copy of header failed! status: %d pending_idx: %u ref: %u\n",
474 (*gopp_copy)->status,
476 (*gopp_copy)->source.u.ref);
477 /* The first frag might still have this slot mapped */
479 xenvif_idx_release(queue, pending_idx,
480 XEN_NETIF_RSP_ERROR);
485 for (i = 0; i < nr_frags; i++, gop_map++) {
488 pending_idx = frag_get_pending_idx(&shinfo->frags[i]);
490 /* Check error status: if okay then remember grant handle. */
491 newerr = gop_map->status;
493 if (likely(!newerr)) {
494 xenvif_grant_handle_set(queue,
497 /* Had a previous error? Invalidate this fragment. */
499 xenvif_idx_unmap(queue, pending_idx);
500 /* If the mapping of the first frag was OK, but
501 * the header's copy failed, and they are
502 * sharing a slot, send an error
504 if (i == 0 && !first_shinfo && sharedslot)
505 xenvif_idx_release(queue, pending_idx,
506 XEN_NETIF_RSP_ERROR);
508 xenvif_idx_release(queue, pending_idx,
514 /* Error on this fragment: respond to client with an error. */
516 netdev_dbg(queue->vif->dev,
517 "Grant map of %d. frag failed! status: %d pending_idx: %u ref: %u\n",
523 xenvif_idx_release(queue, pending_idx, XEN_NETIF_RSP_ERROR);
525 /* Not the first error? Preceding frags already invalidated. */
529 /* First error: if the header haven't shared a slot with the
530 * first frag, release it as well.
533 xenvif_idx_release(queue,
534 XENVIF_TX_CB(skb)->pending_idx,
537 /* Invalidate preceding fragments of this skb. */
538 for (j = 0; j < i; j++) {
539 pending_idx = frag_get_pending_idx(&shinfo->frags[j]);
540 xenvif_idx_unmap(queue, pending_idx);
541 xenvif_idx_release(queue, pending_idx,
545 /* And if we found the error while checking the frag_list, unmap
546 * the first skb's frags
549 for (j = 0; j < first_shinfo->nr_frags; j++) {
550 pending_idx = frag_get_pending_idx(&first_shinfo->frags[j]);
551 xenvif_idx_unmap(queue, pending_idx);
552 xenvif_idx_release(queue, pending_idx,
557 /* Remember the error: invalidate all subsequent fragments. */
561 if (skb_has_frag_list(skb) && !first_shinfo) {
562 first_shinfo = shinfo;
563 shinfo = skb_shinfo(shinfo->frag_list);
564 nr_frags = shinfo->nr_frags;
573 static void xenvif_fill_frags(struct xenvif_queue *queue, struct sk_buff *skb)
575 struct skb_shared_info *shinfo = skb_shinfo(skb);
576 int nr_frags = shinfo->nr_frags;
578 u16 prev_pending_idx = INVALID_PENDING_IDX;
580 for (i = 0; i < nr_frags; i++) {
581 skb_frag_t *frag = shinfo->frags + i;
582 struct xen_netif_tx_request *txp;
586 pending_idx = frag_get_pending_idx(frag);
588 /* If this is not the first frag, chain it to the previous*/
589 if (prev_pending_idx == INVALID_PENDING_IDX)
590 skb_shinfo(skb)->destructor_arg =
591 &callback_param(queue, pending_idx);
593 callback_param(queue, prev_pending_idx).ctx =
594 &callback_param(queue, pending_idx);
596 callback_param(queue, pending_idx).ctx = NULL;
597 prev_pending_idx = pending_idx;
599 txp = &queue->pending_tx_info[pending_idx].req;
600 page = virt_to_page(idx_to_kaddr(queue, pending_idx));
601 __skb_fill_page_desc(skb, i, page, txp->offset, txp->size);
602 skb->len += txp->size;
603 skb->data_len += txp->size;
604 skb->truesize += txp->size;
606 /* Take an extra reference to offset network stack's put_page */
607 get_page(queue->mmap_pages[pending_idx]);
611 static int xenvif_get_extras(struct xenvif_queue *queue,
612 struct xen_netif_extra_info *extras,
613 unsigned int *extra_count,
616 struct xen_netif_extra_info extra;
617 RING_IDX cons = queue->tx.req_cons;
620 if (unlikely(work_to_do-- <= 0)) {
621 netdev_err(queue->vif->dev, "Missing extra info\n");
622 xenvif_fatal_tx_err(queue->vif);
626 RING_COPY_REQUEST(&queue->tx, cons, &extra);
628 queue->tx.req_cons = ++cons;
631 if (unlikely(!extra.type ||
632 extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
633 netdev_err(queue->vif->dev,
634 "Invalid extra type: %d\n", extra.type);
635 xenvif_fatal_tx_err(queue->vif);
639 memcpy(&extras[extra.type - 1], &extra, sizeof(extra));
640 } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE);
645 static int xenvif_set_skb_gso(struct xenvif *vif,
647 struct xen_netif_extra_info *gso)
649 if (!gso->u.gso.size) {
650 netdev_err(vif->dev, "GSO size must not be zero.\n");
651 xenvif_fatal_tx_err(vif);
655 switch (gso->u.gso.type) {
656 case XEN_NETIF_GSO_TYPE_TCPV4:
657 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
659 case XEN_NETIF_GSO_TYPE_TCPV6:
660 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
663 netdev_err(vif->dev, "Bad GSO type %d.\n", gso->u.gso.type);
664 xenvif_fatal_tx_err(vif);
668 skb_shinfo(skb)->gso_size = gso->u.gso.size;
669 /* gso_segs will be calculated later */
674 static int checksum_setup(struct xenvif_queue *queue, struct sk_buff *skb)
676 bool recalculate_partial_csum = false;
678 /* A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
679 * peers can fail to set NETRXF_csum_blank when sending a GSO
680 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
681 * recalculate the partial checksum.
683 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
684 queue->stats.rx_gso_checksum_fixup++;
685 skb->ip_summed = CHECKSUM_PARTIAL;
686 recalculate_partial_csum = true;
689 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
690 if (skb->ip_summed != CHECKSUM_PARTIAL)
693 return skb_checksum_setup(skb, recalculate_partial_csum);
696 static bool tx_credit_exceeded(struct xenvif_queue *queue, unsigned size)
698 u64 now = get_jiffies_64();
699 u64 next_credit = queue->credit_window_start +
700 msecs_to_jiffies(queue->credit_usec / 1000);
702 /* Timer could already be pending in rare cases. */
703 if (timer_pending(&queue->credit_timeout)) {
704 queue->rate_limited = true;
708 /* Passed the point where we can replenish credit? */
709 if (time_after_eq64(now, next_credit)) {
710 queue->credit_window_start = now;
711 tx_add_credit(queue);
714 /* Still too big to send right now? Set a callback. */
715 if (size > queue->remaining_credit) {
716 mod_timer(&queue->credit_timeout,
718 queue->credit_window_start = next_credit;
719 queue->rate_limited = true;
727 /* No locking is required in xenvif_mcast_add/del() as they are
728 * only ever invoked from NAPI poll. An RCU list is used because
729 * xenvif_mcast_match() is called asynchronously, during start_xmit.
732 static int xenvif_mcast_add(struct xenvif *vif, const u8 *addr)
734 struct xenvif_mcast_addr *mcast;
736 if (vif->fe_mcast_count == XEN_NETBK_MCAST_MAX) {
739 "Too many multicast addresses\n");
743 mcast = kzalloc(sizeof(*mcast), GFP_ATOMIC);
747 ether_addr_copy(mcast->addr, addr);
748 list_add_tail_rcu(&mcast->entry, &vif->fe_mcast_addr);
749 vif->fe_mcast_count++;
754 static void xenvif_mcast_del(struct xenvif *vif, const u8 *addr)
756 struct xenvif_mcast_addr *mcast;
758 list_for_each_entry_rcu(mcast, &vif->fe_mcast_addr, entry) {
759 if (ether_addr_equal(addr, mcast->addr)) {
760 --vif->fe_mcast_count;
761 list_del_rcu(&mcast->entry);
762 kfree_rcu(mcast, rcu);
768 bool xenvif_mcast_match(struct xenvif *vif, const u8 *addr)
770 struct xenvif_mcast_addr *mcast;
773 list_for_each_entry_rcu(mcast, &vif->fe_mcast_addr, entry) {
774 if (ether_addr_equal(addr, mcast->addr)) {
784 void xenvif_mcast_addr_list_free(struct xenvif *vif)
786 /* No need for locking or RCU here. NAPI poll and TX queue
789 while (!list_empty(&vif->fe_mcast_addr)) {
790 struct xenvif_mcast_addr *mcast;
792 mcast = list_first_entry(&vif->fe_mcast_addr,
793 struct xenvif_mcast_addr,
795 --vif->fe_mcast_count;
796 list_del(&mcast->entry);
801 static void xenvif_tx_build_gops(struct xenvif_queue *queue,
806 struct gnttab_map_grant_ref *gop = queue->tx_map_ops;
807 struct sk_buff *skb, *nskb;
809 unsigned int frag_overflow;
811 while (skb_queue_len(&queue->tx_queue) < budget) {
812 struct xen_netif_tx_request txreq;
813 struct xen_netif_tx_request txfrags[XEN_NETBK_LEGACY_SLOTS_MAX];
814 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX-1];
815 unsigned int extra_count;
819 unsigned int data_len;
820 pending_ring_idx_t index;
822 if (queue->tx.sring->req_prod - queue->tx.req_cons >
823 XEN_NETIF_TX_RING_SIZE) {
824 netdev_err(queue->vif->dev,
825 "Impossible number of requests. "
826 "req_prod %d, req_cons %d, size %ld\n",
827 queue->tx.sring->req_prod, queue->tx.req_cons,
828 XEN_NETIF_TX_RING_SIZE);
829 xenvif_fatal_tx_err(queue->vif);
833 work_to_do = XEN_RING_NR_UNCONSUMED_REQUESTS(&queue->tx);
837 idx = queue->tx.req_cons;
838 rmb(); /* Ensure that we see the request before we copy it. */
839 RING_COPY_REQUEST(&queue->tx, idx, &txreq);
841 /* Credit-based scheduling. */
842 if (txreq.size > queue->remaining_credit &&
843 tx_credit_exceeded(queue, txreq.size))
846 queue->remaining_credit -= txreq.size;
849 queue->tx.req_cons = ++idx;
851 memset(extras, 0, sizeof(extras));
853 if (txreq.flags & XEN_NETTXF_extra_info) {
854 work_to_do = xenvif_get_extras(queue, extras,
857 idx = queue->tx.req_cons;
858 if (unlikely(work_to_do < 0))
862 if (extras[XEN_NETIF_EXTRA_TYPE_MCAST_ADD - 1].type) {
863 struct xen_netif_extra_info *extra;
865 extra = &extras[XEN_NETIF_EXTRA_TYPE_MCAST_ADD - 1];
866 ret = xenvif_mcast_add(queue->vif, extra->u.mcast.addr);
868 make_tx_response(queue, &txreq, extra_count,
871 XEN_NETIF_RSP_ERROR);
872 push_tx_responses(queue);
876 if (extras[XEN_NETIF_EXTRA_TYPE_MCAST_DEL - 1].type) {
877 struct xen_netif_extra_info *extra;
879 extra = &extras[XEN_NETIF_EXTRA_TYPE_MCAST_DEL - 1];
880 xenvif_mcast_del(queue->vif, extra->u.mcast.addr);
882 make_tx_response(queue, &txreq, extra_count,
884 push_tx_responses(queue);
888 ret = xenvif_count_requests(queue, &txreq, extra_count,
889 txfrags, work_to_do);
890 if (unlikely(ret < 0))
895 if (unlikely(txreq.size < ETH_HLEN)) {
896 netdev_dbg(queue->vif->dev,
897 "Bad packet size: %d\n", txreq.size);
898 xenvif_tx_err(queue, &txreq, extra_count, idx);
902 /* No crossing a page as the payload mustn't fragment. */
903 if (unlikely((txreq.offset + txreq.size) > XEN_PAGE_SIZE)) {
904 netdev_err(queue->vif->dev,
905 "txreq.offset: %u, size: %u, end: %lu\n",
906 txreq.offset, txreq.size,
907 (unsigned long)(txreq.offset&~XEN_PAGE_MASK) + txreq.size);
908 xenvif_fatal_tx_err(queue->vif);
912 index = pending_index(queue->pending_cons);
913 pending_idx = queue->pending_ring[index];
915 data_len = (txreq.size > XEN_NETBACK_TX_COPY_LEN &&
916 ret < XEN_NETBK_LEGACY_SLOTS_MAX) ?
917 XEN_NETBACK_TX_COPY_LEN : txreq.size;
919 skb = xenvif_alloc_skb(data_len);
920 if (unlikely(skb == NULL)) {
921 netdev_dbg(queue->vif->dev,
922 "Can't allocate a skb in start_xmit.\n");
923 xenvif_tx_err(queue, &txreq, extra_count, idx);
927 skb_shinfo(skb)->nr_frags = ret;
928 if (data_len < txreq.size)
929 skb_shinfo(skb)->nr_frags++;
930 /* At this point shinfo->nr_frags is in fact the number of
931 * slots, which can be as large as XEN_NETBK_LEGACY_SLOTS_MAX.
935 if (skb_shinfo(skb)->nr_frags > MAX_SKB_FRAGS) {
936 frag_overflow = skb_shinfo(skb)->nr_frags - MAX_SKB_FRAGS;
937 BUG_ON(frag_overflow > MAX_SKB_FRAGS);
938 skb_shinfo(skb)->nr_frags = MAX_SKB_FRAGS;
939 nskb = xenvif_alloc_skb(0);
940 if (unlikely(nskb == NULL)) {
941 skb_shinfo(skb)->nr_frags = 0;
943 xenvif_tx_err(queue, &txreq, extra_count, idx);
945 netdev_err(queue->vif->dev,
946 "Can't allocate the frag_list skb.\n");
951 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
952 struct xen_netif_extra_info *gso;
953 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
955 if (xenvif_set_skb_gso(queue->vif, skb, gso)) {
956 /* Failure in xenvif_set_skb_gso is fatal. */
957 skb_shinfo(skb)->nr_frags = 0;
964 if (extras[XEN_NETIF_EXTRA_TYPE_HASH - 1].type) {
965 struct xen_netif_extra_info *extra;
966 enum pkt_hash_types type = PKT_HASH_TYPE_NONE;
968 extra = &extras[XEN_NETIF_EXTRA_TYPE_HASH - 1];
970 switch (extra->u.hash.type) {
971 case _XEN_NETIF_CTRL_HASH_TYPE_IPV4:
972 case _XEN_NETIF_CTRL_HASH_TYPE_IPV6:
973 type = PKT_HASH_TYPE_L3;
976 case _XEN_NETIF_CTRL_HASH_TYPE_IPV4_TCP:
977 case _XEN_NETIF_CTRL_HASH_TYPE_IPV6_TCP:
978 type = PKT_HASH_TYPE_L4;
985 if (type != PKT_HASH_TYPE_NONE)
987 *(u32 *)extra->u.hash.value,
991 XENVIF_TX_CB(skb)->pending_idx = pending_idx;
993 __skb_put(skb, data_len);
994 queue->tx_copy_ops[*copy_ops].source.u.ref = txreq.gref;
995 queue->tx_copy_ops[*copy_ops].source.domid = queue->vif->domid;
996 queue->tx_copy_ops[*copy_ops].source.offset = txreq.offset;
998 queue->tx_copy_ops[*copy_ops].dest.u.gmfn =
999 virt_to_gfn(skb->data);
1000 queue->tx_copy_ops[*copy_ops].dest.domid = DOMID_SELF;
1001 queue->tx_copy_ops[*copy_ops].dest.offset =
1002 offset_in_page(skb->data) & ~XEN_PAGE_MASK;
1004 queue->tx_copy_ops[*copy_ops].len = data_len;
1005 queue->tx_copy_ops[*copy_ops].flags = GNTCOPY_source_gref;
1009 if (data_len < txreq.size) {
1010 frag_set_pending_idx(&skb_shinfo(skb)->frags[0],
1012 xenvif_tx_create_map_op(queue, pending_idx, &txreq,
1016 frag_set_pending_idx(&skb_shinfo(skb)->frags[0],
1017 INVALID_PENDING_IDX);
1018 memcpy(&queue->pending_tx_info[pending_idx].req,
1019 &txreq, sizeof(txreq));
1020 queue->pending_tx_info[pending_idx].extra_count =
1024 queue->pending_cons++;
1026 gop = xenvif_get_requests(queue, skb, txfrags, gop,
1027 frag_overflow, nskb);
1029 __skb_queue_tail(&queue->tx_queue, skb);
1031 queue->tx.req_cons = idx;
1033 if (((gop-queue->tx_map_ops) >= ARRAY_SIZE(queue->tx_map_ops)) ||
1034 (*copy_ops >= ARRAY_SIZE(queue->tx_copy_ops)))
1038 (*map_ops) = gop - queue->tx_map_ops;
1042 /* Consolidate skb with a frag_list into a brand new one with local pages on
1043 * frags. Returns 0 or -ENOMEM if can't allocate new pages.
1045 static int xenvif_handle_frag_list(struct xenvif_queue *queue, struct sk_buff *skb)
1047 unsigned int offset = skb_headlen(skb);
1048 skb_frag_t frags[MAX_SKB_FRAGS];
1050 struct ubuf_info *uarg;
1051 struct sk_buff *nskb = skb_shinfo(skb)->frag_list;
1053 queue->stats.tx_zerocopy_sent += 2;
1054 queue->stats.tx_frag_overflow++;
1056 xenvif_fill_frags(queue, nskb);
1057 /* Subtract frags size, we will correct it later */
1058 skb->truesize -= skb->data_len;
1059 skb->len += nskb->len;
1060 skb->data_len += nskb->len;
1062 /* create a brand new frags array and coalesce there */
1063 for (i = 0; offset < skb->len; i++) {
1067 BUG_ON(i >= MAX_SKB_FRAGS);
1068 page = alloc_page(GFP_ATOMIC);
1071 skb->truesize += skb->data_len;
1072 for (j = 0; j < i; j++)
1073 put_page(skb_frag_page(&frags[j]));
1077 if (offset + PAGE_SIZE < skb->len)
1080 len = skb->len - offset;
1081 if (skb_copy_bits(skb, offset, page_address(page), len))
1085 __skb_frag_set_page(&frags[i], page);
1086 skb_frag_off_set(&frags[i], 0);
1087 skb_frag_size_set(&frags[i], len);
1090 /* Release all the original (foreign) frags. */
1091 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
1092 skb_frag_unref(skb, f);
1093 uarg = skb_shinfo(skb)->destructor_arg;
1094 /* increase inflight counter to offset decrement in callback */
1095 atomic_inc(&queue->inflight_packets);
1096 uarg->callback(NULL, uarg, true);
1097 skb_shinfo(skb)->destructor_arg = NULL;
1099 /* Fill the skb with the new (local) frags. */
1100 memcpy(skb_shinfo(skb)->frags, frags, i * sizeof(skb_frag_t));
1101 skb_shinfo(skb)->nr_frags = i;
1102 skb->truesize += i * PAGE_SIZE;
1107 static int xenvif_tx_submit(struct xenvif_queue *queue)
1109 struct gnttab_map_grant_ref *gop_map = queue->tx_map_ops;
1110 struct gnttab_copy *gop_copy = queue->tx_copy_ops;
1111 struct sk_buff *skb;
1114 while ((skb = __skb_dequeue(&queue->tx_queue)) != NULL) {
1115 struct xen_netif_tx_request *txp;
1119 pending_idx = XENVIF_TX_CB(skb)->pending_idx;
1120 txp = &queue->pending_tx_info[pending_idx].req;
1122 /* Check the remap error code. */
1123 if (unlikely(xenvif_tx_check_gop(queue, skb, &gop_map, &gop_copy))) {
1124 /* If there was an error, xenvif_tx_check_gop is
1125 * expected to release all the frags which were mapped,
1126 * so kfree_skb shouldn't do it again
1128 skb_shinfo(skb)->nr_frags = 0;
1129 if (skb_has_frag_list(skb)) {
1130 struct sk_buff *nskb =
1131 skb_shinfo(skb)->frag_list;
1132 skb_shinfo(nskb)->nr_frags = 0;
1138 data_len = skb->len;
1139 callback_param(queue, pending_idx).ctx = NULL;
1140 if (data_len < txp->size) {
1141 /* Append the packet payload as a fragment. */
1142 txp->offset += data_len;
1143 txp->size -= data_len;
1145 /* Schedule a response immediately. */
1146 xenvif_idx_release(queue, pending_idx,
1147 XEN_NETIF_RSP_OKAY);
1150 if (txp->flags & XEN_NETTXF_csum_blank)
1151 skb->ip_summed = CHECKSUM_PARTIAL;
1152 else if (txp->flags & XEN_NETTXF_data_validated)
1153 skb->ip_summed = CHECKSUM_UNNECESSARY;
1155 xenvif_fill_frags(queue, skb);
1157 if (unlikely(skb_has_frag_list(skb))) {
1158 struct sk_buff *nskb = skb_shinfo(skb)->frag_list;
1159 xenvif_skb_zerocopy_prepare(queue, nskb);
1160 if (xenvif_handle_frag_list(queue, skb)) {
1161 if (net_ratelimit())
1162 netdev_err(queue->vif->dev,
1163 "Not enough memory to consolidate frag_list!\n");
1164 xenvif_skb_zerocopy_prepare(queue, skb);
1168 /* Copied all the bits from the frag list -- free it. */
1169 skb_frag_list_init(skb);
1173 skb->dev = queue->vif->dev;
1174 skb->protocol = eth_type_trans(skb, skb->dev);
1175 skb_reset_network_header(skb);
1177 if (checksum_setup(queue, skb)) {
1178 netdev_dbg(queue->vif->dev,
1179 "Can't setup checksum in net_tx_action\n");
1180 /* We have to set this flag to trigger the callback */
1181 if (skb_shinfo(skb)->destructor_arg)
1182 xenvif_skb_zerocopy_prepare(queue, skb);
1187 skb_probe_transport_header(skb);
1189 /* If the packet is GSO then we will have just set up the
1190 * transport header offset in checksum_setup so it's now
1191 * straightforward to calculate gso_segs.
1193 if (skb_is_gso(skb)) {
1196 /* GSO implies having the L4 header. */
1197 WARN_ON_ONCE(!skb_transport_header_was_set(skb));
1198 if (unlikely(!skb_transport_header_was_set(skb))) {
1203 mss = skb_shinfo(skb)->gso_size;
1204 hdrlen = skb_transport_header(skb) -
1205 skb_mac_header(skb) +
1208 skb_shinfo(skb)->gso_segs =
1209 DIV_ROUND_UP(skb->len - hdrlen, mss);
1212 queue->stats.rx_bytes += skb->len;
1213 queue->stats.rx_packets++;
1217 /* Set this flag right before netif_receive_skb, otherwise
1218 * someone might think this packet already left netback, and
1219 * do a skb_copy_ubufs while we are still in control of the
1220 * skb. E.g. the __pskb_pull_tail earlier can do such thing.
1222 if (skb_shinfo(skb)->destructor_arg) {
1223 xenvif_skb_zerocopy_prepare(queue, skb);
1224 queue->stats.tx_zerocopy_sent++;
1227 netif_receive_skb(skb);
1233 void xenvif_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *ubuf,
1234 bool zerocopy_success)
1236 unsigned long flags;
1237 pending_ring_idx_t index;
1238 struct xenvif_queue *queue = ubuf_to_queue(ubuf);
1240 /* This is the only place where we grab this lock, to protect callbacks
1243 spin_lock_irqsave(&queue->callback_lock, flags);
1245 u16 pending_idx = ubuf->desc;
1246 ubuf = (struct ubuf_info *) ubuf->ctx;
1247 BUG_ON(queue->dealloc_prod - queue->dealloc_cons >=
1249 index = pending_index(queue->dealloc_prod);
1250 queue->dealloc_ring[index] = pending_idx;
1251 /* Sync with xenvif_tx_dealloc_action:
1252 * insert idx then incr producer.
1255 queue->dealloc_prod++;
1257 spin_unlock_irqrestore(&queue->callback_lock, flags);
1259 if (likely(zerocopy_success))
1260 queue->stats.tx_zerocopy_success++;
1262 queue->stats.tx_zerocopy_fail++;
1263 xenvif_skb_zerocopy_complete(queue);
1266 static inline void xenvif_tx_dealloc_action(struct xenvif_queue *queue)
1268 struct gnttab_unmap_grant_ref *gop;
1269 pending_ring_idx_t dc, dp;
1270 u16 pending_idx, pending_idx_release[MAX_PENDING_REQS];
1273 dc = queue->dealloc_cons;
1274 gop = queue->tx_unmap_ops;
1276 /* Free up any grants we have finished using */
1278 dp = queue->dealloc_prod;
1280 /* Ensure we see all indices enqueued by all
1281 * xenvif_zerocopy_callback().
1286 BUG_ON(gop - queue->tx_unmap_ops >= MAX_PENDING_REQS);
1288 queue->dealloc_ring[pending_index(dc++)];
1290 pending_idx_release[gop - queue->tx_unmap_ops] =
1292 queue->pages_to_unmap[gop - queue->tx_unmap_ops] =
1293 queue->mmap_pages[pending_idx];
1294 gnttab_set_unmap_op(gop,
1295 idx_to_kaddr(queue, pending_idx),
1297 queue->grant_tx_handle[pending_idx]);
1298 xenvif_grant_handle_reset(queue, pending_idx);
1302 } while (dp != queue->dealloc_prod);
1304 queue->dealloc_cons = dc;
1306 if (gop - queue->tx_unmap_ops > 0) {
1308 ret = gnttab_unmap_refs(queue->tx_unmap_ops,
1310 queue->pages_to_unmap,
1311 gop - queue->tx_unmap_ops);
1313 netdev_err(queue->vif->dev, "Unmap fail: nr_ops %tu ret %d\n",
1314 gop - queue->tx_unmap_ops, ret);
1315 for (i = 0; i < gop - queue->tx_unmap_ops; ++i) {
1316 if (gop[i].status != GNTST_okay)
1317 netdev_err(queue->vif->dev,
1318 " host_addr: 0x%llx handle: 0x%x status: %d\n",
1327 for (i = 0; i < gop - queue->tx_unmap_ops; ++i)
1328 xenvif_idx_release(queue, pending_idx_release[i],
1329 XEN_NETIF_RSP_OKAY);
1333 /* Called after netfront has transmitted */
1334 int xenvif_tx_action(struct xenvif_queue *queue, int budget)
1336 unsigned nr_mops, nr_cops = 0;
1339 if (unlikely(!tx_work_todo(queue)))
1342 xenvif_tx_build_gops(queue, budget, &nr_cops, &nr_mops);
1347 gnttab_batch_copy(queue->tx_copy_ops, nr_cops);
1349 ret = gnttab_map_refs(queue->tx_map_ops,
1351 queue->pages_to_map,
1356 netdev_err(queue->vif->dev, "Map fail: nr %u ret %d\n",
1358 for (i = 0; i < nr_mops; ++i)
1359 WARN_ON_ONCE(queue->tx_map_ops[i].status ==
1364 work_done = xenvif_tx_submit(queue);
1369 static void xenvif_idx_release(struct xenvif_queue *queue, u16 pending_idx,
1372 struct pending_tx_info *pending_tx_info;
1373 pending_ring_idx_t index;
1374 unsigned long flags;
1376 pending_tx_info = &queue->pending_tx_info[pending_idx];
1378 spin_lock_irqsave(&queue->response_lock, flags);
1380 make_tx_response(queue, &pending_tx_info->req,
1381 pending_tx_info->extra_count, status);
1383 /* Release the pending index before pusing the Tx response so
1384 * its available before a new Tx request is pushed by the
1387 index = pending_index(queue->pending_prod++);
1388 queue->pending_ring[index] = pending_idx;
1390 push_tx_responses(queue);
1392 spin_unlock_irqrestore(&queue->response_lock, flags);
1396 static void make_tx_response(struct xenvif_queue *queue,
1397 struct xen_netif_tx_request *txp,
1398 unsigned int extra_count,
1401 RING_IDX i = queue->tx.rsp_prod_pvt;
1402 struct xen_netif_tx_response *resp;
1404 resp = RING_GET_RESPONSE(&queue->tx, i);
1408 while (extra_count-- != 0)
1409 RING_GET_RESPONSE(&queue->tx, ++i)->status = XEN_NETIF_RSP_NULL;
1411 queue->tx.rsp_prod_pvt = ++i;
1414 static void push_tx_responses(struct xenvif_queue *queue)
1418 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&queue->tx, notify);
1420 notify_remote_via_irq(queue->tx_irq);
1423 static void xenvif_idx_unmap(struct xenvif_queue *queue, u16 pending_idx)
1426 struct gnttab_unmap_grant_ref tx_unmap_op;
1428 gnttab_set_unmap_op(&tx_unmap_op,
1429 idx_to_kaddr(queue, pending_idx),
1431 queue->grant_tx_handle[pending_idx]);
1432 xenvif_grant_handle_reset(queue, pending_idx);
1434 ret = gnttab_unmap_refs(&tx_unmap_op, NULL,
1435 &queue->mmap_pages[pending_idx], 1);
1437 netdev_err(queue->vif->dev,
1438 "Unmap fail: ret: %d pending_idx: %d host_addr: %llx handle: 0x%x status: %d\n",
1441 tx_unmap_op.host_addr,
1443 tx_unmap_op.status);
1448 static inline int tx_work_todo(struct xenvif_queue *queue)
1450 if (likely(RING_HAS_UNCONSUMED_REQUESTS(&queue->tx)))
1456 static inline bool tx_dealloc_work_todo(struct xenvif_queue *queue)
1458 return queue->dealloc_cons != queue->dealloc_prod;
1461 void xenvif_unmap_frontend_data_rings(struct xenvif_queue *queue)
1463 if (queue->tx.sring)
1464 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(queue->vif),
1466 if (queue->rx.sring)
1467 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(queue->vif),
1471 int xenvif_map_frontend_data_rings(struct xenvif_queue *queue,
1472 grant_ref_t tx_ring_ref,
1473 grant_ref_t rx_ring_ref)
1476 struct xen_netif_tx_sring *txs;
1477 struct xen_netif_rx_sring *rxs;
1478 RING_IDX rsp_prod, req_prod;
1481 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(queue->vif),
1482 &tx_ring_ref, 1, &addr);
1486 txs = (struct xen_netif_tx_sring *)addr;
1487 rsp_prod = READ_ONCE(txs->rsp_prod);
1488 req_prod = READ_ONCE(txs->req_prod);
1490 BACK_RING_ATTACH(&queue->tx, txs, rsp_prod, XEN_PAGE_SIZE);
1493 if (req_prod - rsp_prod > RING_SIZE(&queue->tx))
1496 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(queue->vif),
1497 &rx_ring_ref, 1, &addr);
1501 rxs = (struct xen_netif_rx_sring *)addr;
1502 rsp_prod = READ_ONCE(rxs->rsp_prod);
1503 req_prod = READ_ONCE(rxs->req_prod);
1505 BACK_RING_ATTACH(&queue->rx, rxs, rsp_prod, XEN_PAGE_SIZE);
1508 if (req_prod - rsp_prod > RING_SIZE(&queue->rx))
1514 xenvif_unmap_frontend_data_rings(queue);
1518 static bool xenvif_dealloc_kthread_should_stop(struct xenvif_queue *queue)
1520 /* Dealloc thread must remain running until all inflight
1523 return kthread_should_stop() &&
1524 !atomic_read(&queue->inflight_packets);
1527 int xenvif_dealloc_kthread(void *data)
1529 struct xenvif_queue *queue = data;
1532 wait_event_interruptible(queue->dealloc_wq,
1533 tx_dealloc_work_todo(queue) ||
1534 xenvif_dealloc_kthread_should_stop(queue));
1535 if (xenvif_dealloc_kthread_should_stop(queue))
1538 xenvif_tx_dealloc_action(queue);
1542 /* Unmap anything remaining*/
1543 if (tx_dealloc_work_todo(queue))
1544 xenvif_tx_dealloc_action(queue);
1549 static void make_ctrl_response(struct xenvif *vif,
1550 const struct xen_netif_ctrl_request *req,
1551 u32 status, u32 data)
1553 RING_IDX idx = vif->ctrl.rsp_prod_pvt;
1554 struct xen_netif_ctrl_response rsp = {
1561 *RING_GET_RESPONSE(&vif->ctrl, idx) = rsp;
1562 vif->ctrl.rsp_prod_pvt = ++idx;
1565 static void push_ctrl_response(struct xenvif *vif)
1569 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->ctrl, notify);
1571 notify_remote_via_irq(vif->ctrl_irq);
1574 static void process_ctrl_request(struct xenvif *vif,
1575 const struct xen_netif_ctrl_request *req)
1577 u32 status = XEN_NETIF_CTRL_STATUS_NOT_SUPPORTED;
1580 switch (req->type) {
1581 case XEN_NETIF_CTRL_TYPE_SET_HASH_ALGORITHM:
1582 status = xenvif_set_hash_alg(vif, req->data[0]);
1585 case XEN_NETIF_CTRL_TYPE_GET_HASH_FLAGS:
1586 status = xenvif_get_hash_flags(vif, &data);
1589 case XEN_NETIF_CTRL_TYPE_SET_HASH_FLAGS:
1590 status = xenvif_set_hash_flags(vif, req->data[0]);
1593 case XEN_NETIF_CTRL_TYPE_SET_HASH_KEY:
1594 status = xenvif_set_hash_key(vif, req->data[0],
1598 case XEN_NETIF_CTRL_TYPE_GET_HASH_MAPPING_SIZE:
1599 status = XEN_NETIF_CTRL_STATUS_SUCCESS;
1600 data = XEN_NETBK_MAX_HASH_MAPPING_SIZE;
1603 case XEN_NETIF_CTRL_TYPE_SET_HASH_MAPPING_SIZE:
1604 status = xenvif_set_hash_mapping_size(vif,
1608 case XEN_NETIF_CTRL_TYPE_SET_HASH_MAPPING:
1609 status = xenvif_set_hash_mapping(vif, req->data[0],
1618 make_ctrl_response(vif, req, status, data);
1619 push_ctrl_response(vif);
1622 static void xenvif_ctrl_action(struct xenvif *vif)
1625 RING_IDX req_prod, req_cons;
1627 req_prod = vif->ctrl.sring->req_prod;
1628 req_cons = vif->ctrl.req_cons;
1630 /* Make sure we can see requests before we process them. */
1633 if (req_cons == req_prod)
1636 while (req_cons != req_prod) {
1637 struct xen_netif_ctrl_request req;
1639 RING_COPY_REQUEST(&vif->ctrl, req_cons, &req);
1642 process_ctrl_request(vif, &req);
1645 vif->ctrl.req_cons = req_cons;
1646 vif->ctrl.sring->req_event = req_cons + 1;
1650 static bool xenvif_ctrl_work_todo(struct xenvif *vif)
1652 if (likely(RING_HAS_UNCONSUMED_REQUESTS(&vif->ctrl)))
1658 irqreturn_t xenvif_ctrl_irq_fn(int irq, void *data)
1660 struct xenvif *vif = data;
1661 unsigned int eoi_flag = XEN_EOI_FLAG_SPURIOUS;
1663 while (xenvif_ctrl_work_todo(vif)) {
1664 xenvif_ctrl_action(vif);
1668 xen_irq_lateeoi(irq, eoi_flag);
1673 static int __init netback_init(void)
1680 /* Allow as many queues as there are CPUs but max. 8 if user has not
1681 * specified a value.
1683 if (xenvif_max_queues == 0)
1684 xenvif_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT,
1687 if (fatal_skb_slots < XEN_NETBK_LEGACY_SLOTS_MAX) {
1688 pr_info("fatal_skb_slots too small (%d), bump it to XEN_NETBK_LEGACY_SLOTS_MAX (%d)\n",
1689 fatal_skb_slots, XEN_NETBK_LEGACY_SLOTS_MAX);
1690 fatal_skb_slots = XEN_NETBK_LEGACY_SLOTS_MAX;
1693 rc = xenvif_xenbus_init();
1697 #ifdef CONFIG_DEBUG_FS
1698 xen_netback_dbg_root = debugfs_create_dir("xen-netback", NULL);
1699 #endif /* CONFIG_DEBUG_FS */
1707 module_init(netback_init);
1709 static void __exit netback_fini(void)
1711 #ifdef CONFIG_DEBUG_FS
1712 debugfs_remove_recursive(xen_netback_dbg_root);
1713 #endif /* CONFIG_DEBUG_FS */
1714 xenvif_xenbus_fini();
1716 module_exit(netback_fini);
1718 MODULE_LICENSE("Dual BSD/GPL");
1719 MODULE_ALIAS("xen-backend:vif");