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 void xenvif_idx_unmap(struct xenvif_queue *queue, u16 pending_idx);
110 static inline int tx_work_todo(struct xenvif_queue *queue);
112 static inline unsigned long idx_to_pfn(struct xenvif_queue *queue,
115 return page_to_pfn(queue->mmap_pages[idx]);
118 static inline unsigned long idx_to_kaddr(struct xenvif_queue *queue,
121 return (unsigned long)pfn_to_kaddr(idx_to_pfn(queue, idx));
124 #define callback_param(vif, pending_idx) \
125 (vif->pending_tx_info[pending_idx].callback_struct)
127 /* Find the containing VIF's structure from a pointer in pending_tx_info array
129 static inline struct xenvif_queue *ubuf_to_queue(const struct ubuf_info *ubuf)
131 u16 pending_idx = ubuf->desc;
132 struct pending_tx_info *temp =
133 container_of(ubuf, struct pending_tx_info, callback_struct);
134 return container_of(temp - pending_idx,
139 static u16 frag_get_pending_idx(skb_frag_t *frag)
141 return (u16)frag->page_offset;
144 static void frag_set_pending_idx(skb_frag_t *frag, u16 pending_idx)
146 frag->page_offset = pending_idx;
149 static inline pending_ring_idx_t pending_index(unsigned i)
151 return i & (MAX_PENDING_REQS-1);
154 void xenvif_kick_thread(struct xenvif_queue *queue)
159 void xenvif_napi_schedule_or_enable_events(struct xenvif_queue *queue)
163 RING_FINAL_CHECK_FOR_REQUESTS(&queue->tx, more_to_do);
166 napi_schedule(&queue->napi);
167 else if (atomic_fetch_andnot(NETBK_TX_EOI | NETBK_COMMON_EOI,
168 &queue->eoi_pending) &
169 (NETBK_TX_EOI | NETBK_COMMON_EOI))
170 xen_irq_lateeoi(queue->tx_irq, 0);
173 static void tx_add_credit(struct xenvif_queue *queue)
175 unsigned long max_burst, max_credit;
178 * Allow a burst big enough to transmit a jumbo packet of up to 128kB.
179 * Otherwise the interface can seize up due to insufficient credit.
181 max_burst = max(131072UL, queue->credit_bytes);
183 /* Take care that adding a new chunk of credit doesn't wrap to zero. */
184 max_credit = queue->remaining_credit + queue->credit_bytes;
185 if (max_credit < queue->remaining_credit)
186 max_credit = ULONG_MAX; /* wrapped: clamp to ULONG_MAX */
188 queue->remaining_credit = min(max_credit, max_burst);
189 queue->rate_limited = false;
192 void xenvif_tx_credit_callback(struct timer_list *t)
194 struct xenvif_queue *queue = from_timer(queue, t, credit_timeout);
195 tx_add_credit(queue);
196 xenvif_napi_schedule_or_enable_events(queue);
199 static void xenvif_tx_err(struct xenvif_queue *queue,
200 struct xen_netif_tx_request *txp,
201 unsigned int extra_count, RING_IDX end)
203 RING_IDX cons = queue->tx.req_cons;
207 spin_lock_irqsave(&queue->response_lock, flags);
208 make_tx_response(queue, txp, extra_count, XEN_NETIF_RSP_ERROR);
209 push_tx_responses(queue);
210 spin_unlock_irqrestore(&queue->response_lock, flags);
213 RING_COPY_REQUEST(&queue->tx, cons++, txp);
214 extra_count = 0; /* only the first frag can have extras */
216 queue->tx.req_cons = cons;
219 static void xenvif_fatal_tx_err(struct xenvif *vif)
221 netdev_err(vif->dev, "fatal error; disabling device\n");
222 vif->disabled = true;
223 /* Disable the vif from queue 0's kthread */
225 xenvif_kick_thread(&vif->queues[0]);
228 static int xenvif_count_requests(struct xenvif_queue *queue,
229 struct xen_netif_tx_request *first,
230 unsigned int extra_count,
231 struct xen_netif_tx_request *txp,
234 RING_IDX cons = queue->tx.req_cons;
239 if (!(first->flags & XEN_NETTXF_more_data))
243 struct xen_netif_tx_request dropped_tx = { 0 };
245 if (slots >= work_to_do) {
246 netdev_err(queue->vif->dev,
247 "Asked for %d slots but exceeds this limit\n",
249 xenvif_fatal_tx_err(queue->vif);
253 /* This guest is really using too many slots and
254 * considered malicious.
256 if (unlikely(slots >= fatal_skb_slots)) {
257 netdev_err(queue->vif->dev,
258 "Malicious frontend using %d slots, threshold %u\n",
259 slots, fatal_skb_slots);
260 xenvif_fatal_tx_err(queue->vif);
264 /* Xen network protocol had implicit dependency on
265 * MAX_SKB_FRAGS. XEN_NETBK_LEGACY_SLOTS_MAX is set to
266 * the historical MAX_SKB_FRAGS value 18 to honor the
267 * same behavior as before. Any packet using more than
268 * 18 slots but less than fatal_skb_slots slots is
271 if (!drop_err && slots >= XEN_NETBK_LEGACY_SLOTS_MAX) {
273 netdev_dbg(queue->vif->dev,
274 "Too many slots (%d) exceeding limit (%d), dropping packet\n",
275 slots, XEN_NETBK_LEGACY_SLOTS_MAX);
282 RING_COPY_REQUEST(&queue->tx, cons + slots, txp);
284 /* If the guest submitted a frame >= 64 KiB then
285 * first->size overflowed and following slots will
286 * appear to be larger than the frame.
288 * This cannot be fatal error as there are buggy
289 * frontends that do this.
291 * Consume all slots and drop the packet.
293 if (!drop_err && txp->size > first->size) {
295 netdev_dbg(queue->vif->dev,
296 "Invalid tx request, slot size %u > remaining size %u\n",
297 txp->size, first->size);
301 first->size -= txp->size;
304 if (unlikely((txp->offset + txp->size) > XEN_PAGE_SIZE)) {
305 netdev_err(queue->vif->dev, "Cross page boundary, txp->offset: %u, size: %u\n",
306 txp->offset, txp->size);
307 xenvif_fatal_tx_err(queue->vif);
311 more_data = txp->flags & XEN_NETTXF_more_data;
319 xenvif_tx_err(queue, first, extra_count, cons + slots);
327 struct xenvif_tx_cb {
328 u16 copy_pending_idx[XEN_NETBK_LEGACY_SLOTS_MAX + 1];
333 #define XENVIF_TX_CB(skb) ((struct xenvif_tx_cb *)(skb)->cb)
334 #define copy_pending_idx(skb, i) (XENVIF_TX_CB(skb)->copy_pending_idx[i])
335 #define copy_count(skb) (XENVIF_TX_CB(skb)->copy_count)
337 static inline void xenvif_tx_create_map_op(struct xenvif_queue *queue,
339 struct xen_netif_tx_request *txp,
340 unsigned int extra_count,
341 struct gnttab_map_grant_ref *mop)
343 queue->pages_to_map[mop-queue->tx_map_ops] = queue->mmap_pages[pending_idx];
344 gnttab_set_map_op(mop, idx_to_kaddr(queue, pending_idx),
345 GNTMAP_host_map | GNTMAP_readonly,
346 txp->gref, queue->vif->domid);
348 memcpy(&queue->pending_tx_info[pending_idx].req, txp,
350 queue->pending_tx_info[pending_idx].extra_count = extra_count;
353 static inline struct sk_buff *xenvif_alloc_skb(unsigned int size)
355 struct sk_buff *skb =
356 alloc_skb(size + NET_SKB_PAD + NET_IP_ALIGN,
357 GFP_ATOMIC | __GFP_NOWARN);
359 BUILD_BUG_ON(sizeof(*XENVIF_TX_CB(skb)) > sizeof(skb->cb));
360 if (unlikely(skb == NULL))
363 /* Packets passed to netif_rx() must have some headroom. */
364 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
366 /* Initialize it here to avoid later surprises */
367 skb_shinfo(skb)->destructor_arg = NULL;
372 static void xenvif_get_requests(struct xenvif_queue *queue,
374 struct xen_netif_tx_request *first,
375 struct xen_netif_tx_request *txfrags,
378 unsigned int frag_overflow,
379 struct sk_buff *nskb,
380 unsigned int extra_count,
381 unsigned int data_len)
383 struct skb_shared_info *shinfo = skb_shinfo(skb);
384 skb_frag_t *frags = shinfo->frags;
386 pending_ring_idx_t index;
387 unsigned int nr_slots;
388 struct gnttab_copy *cop = queue->tx_copy_ops + *copy_ops;
389 struct gnttab_map_grant_ref *gop = queue->tx_map_ops + *map_ops;
390 struct xen_netif_tx_request *txp = first;
392 nr_slots = shinfo->nr_frags + 1;
395 XENVIF_TX_CB(skb)->split_mask = 0;
397 /* Create copy ops for exactly data_len bytes into the skb head. */
398 __skb_put(skb, data_len);
399 while (data_len > 0) {
400 int amount = data_len > txp->size ? txp->size : data_len;
403 cop->source.u.ref = txp->gref;
404 cop->source.domid = queue->vif->domid;
405 cop->source.offset = txp->offset;
407 cop->dest.domid = DOMID_SELF;
408 cop->dest.offset = (offset_in_page(skb->data +
410 data_len)) & ~XEN_PAGE_MASK;
411 cop->dest.u.gmfn = virt_to_gfn(skb->data + skb_headlen(skb)
414 /* Don't cross local page boundary! */
415 if (cop->dest.offset + amount > XEN_PAGE_SIZE) {
416 amount = XEN_PAGE_SIZE - cop->dest.offset;
417 XENVIF_TX_CB(skb)->split_mask |= 1U << copy_count(skb);
422 cop->flags = GNTCOPY_source_gref;
424 index = pending_index(queue->pending_cons);
425 pending_idx = queue->pending_ring[index];
426 callback_param(queue, pending_idx).ctx = NULL;
427 copy_pending_idx(skb, copy_count(skb)) = pending_idx;
434 if (amount == txp->size) {
435 /* The copy op covered the full tx_request */
437 memcpy(&queue->pending_tx_info[pending_idx].req,
439 queue->pending_tx_info[pending_idx].extra_count =
440 (txp == first) ? extra_count : 0;
446 queue->pending_cons++;
449 /* The copy op partially covered the tx_request.
450 * The remainder will be mapped or copied in the next
453 txp->offset += amount;
458 for (shinfo->nr_frags = 0; shinfo->nr_frags < nr_slots;
459 shinfo->nr_frags++, gop++) {
460 index = pending_index(queue->pending_cons++);
461 pending_idx = queue->pending_ring[index];
462 xenvif_tx_create_map_op(queue, pending_idx, txp,
463 txp == first ? extra_count : 0, gop);
464 frag_set_pending_idx(&frags[shinfo->nr_frags], pending_idx);
474 shinfo = skb_shinfo(nskb);
475 frags = shinfo->frags;
477 for (shinfo->nr_frags = 0; shinfo->nr_frags < frag_overflow;
478 shinfo->nr_frags++, txp++, gop++) {
479 index = pending_index(queue->pending_cons++);
480 pending_idx = queue->pending_ring[index];
481 xenvif_tx_create_map_op(queue, pending_idx, txp, 0,
483 frag_set_pending_idx(&frags[shinfo->nr_frags],
487 skb_shinfo(skb)->frag_list = nskb;
490 (*copy_ops) = cop - queue->tx_copy_ops;
491 (*map_ops) = gop - queue->tx_map_ops;
494 static inline void xenvif_grant_handle_set(struct xenvif_queue *queue,
496 grant_handle_t handle)
498 if (unlikely(queue->grant_tx_handle[pending_idx] !=
499 NETBACK_INVALID_HANDLE)) {
500 netdev_err(queue->vif->dev,
501 "Trying to overwrite active handle! pending_idx: 0x%x\n",
505 queue->grant_tx_handle[pending_idx] = handle;
508 static inline void xenvif_grant_handle_reset(struct xenvif_queue *queue,
511 if (unlikely(queue->grant_tx_handle[pending_idx] ==
512 NETBACK_INVALID_HANDLE)) {
513 netdev_err(queue->vif->dev,
514 "Trying to unmap invalid handle! pending_idx: 0x%x\n",
518 queue->grant_tx_handle[pending_idx] = NETBACK_INVALID_HANDLE;
521 static int xenvif_tx_check_gop(struct xenvif_queue *queue,
523 struct gnttab_map_grant_ref **gopp_map,
524 struct gnttab_copy **gopp_copy)
526 struct gnttab_map_grant_ref *gop_map = *gopp_map;
528 /* This always points to the shinfo of the skb being checked, which
529 * could be either the first or the one on the frag_list
531 struct skb_shared_info *shinfo = skb_shinfo(skb);
532 /* If this is non-NULL, we are currently checking the frag_list skb, and
533 * this points to the shinfo of the first one
535 struct skb_shared_info *first_shinfo = NULL;
536 int nr_frags = shinfo->nr_frags;
537 const bool sharedslot = nr_frags &&
538 frag_get_pending_idx(&shinfo->frags[0]) ==
539 copy_pending_idx(skb, copy_count(skb) - 1);
542 for (i = 0; i < copy_count(skb); i++) {
545 /* Check status of header. */
546 pending_idx = copy_pending_idx(skb, i);
548 newerr = (*gopp_copy)->status;
550 /* Split copies need to be handled together. */
551 if (XENVIF_TX_CB(skb)->split_mask & (1U << i)) {
554 newerr = (*gopp_copy)->status;
556 if (likely(!newerr)) {
557 /* The first frag might still have this slot mapped */
558 if (i < copy_count(skb) - 1 || !sharedslot)
559 xenvif_idx_release(queue, pending_idx,
564 netdev_dbg(queue->vif->dev,
565 "Grant copy of header failed! status: %d pending_idx: %u ref: %u\n",
566 (*gopp_copy)->status,
568 (*gopp_copy)->source.u.ref);
569 /* The first frag might still have this slot mapped */
570 if (i < copy_count(skb) - 1 || !sharedslot)
571 xenvif_idx_release(queue, pending_idx,
572 XEN_NETIF_RSP_ERROR);
578 for (i = 0; i < nr_frags; i++, gop_map++) {
581 pending_idx = frag_get_pending_idx(&shinfo->frags[i]);
583 /* Check error status: if okay then remember grant handle. */
584 newerr = gop_map->status;
586 if (likely(!newerr)) {
587 xenvif_grant_handle_set(queue,
590 /* Had a previous error? Invalidate this fragment. */
592 xenvif_idx_unmap(queue, pending_idx);
593 /* If the mapping of the first frag was OK, but
594 * the header's copy failed, and they are
595 * sharing a slot, send an error
597 if (i == 0 && !first_shinfo && sharedslot)
598 xenvif_idx_release(queue, pending_idx,
599 XEN_NETIF_RSP_ERROR);
601 xenvif_idx_release(queue, pending_idx,
607 /* Error on this fragment: respond to client with an error. */
609 netdev_dbg(queue->vif->dev,
610 "Grant map of %d. frag failed! status: %d pending_idx: %u ref: %u\n",
616 xenvif_idx_release(queue, pending_idx, XEN_NETIF_RSP_ERROR);
618 /* Not the first error? Preceding frags already invalidated. */
622 /* Invalidate preceding fragments of this skb. */
623 for (j = 0; j < i; j++) {
624 pending_idx = frag_get_pending_idx(&shinfo->frags[j]);
625 xenvif_idx_unmap(queue, pending_idx);
626 xenvif_idx_release(queue, pending_idx,
630 /* And if we found the error while checking the frag_list, unmap
631 * the first skb's frags
634 for (j = 0; j < first_shinfo->nr_frags; j++) {
635 pending_idx = frag_get_pending_idx(&first_shinfo->frags[j]);
636 xenvif_idx_unmap(queue, pending_idx);
637 xenvif_idx_release(queue, pending_idx,
642 /* Remember the error: invalidate all subsequent fragments. */
646 if (skb_has_frag_list(skb) && !first_shinfo) {
647 first_shinfo = skb_shinfo(skb);
648 shinfo = skb_shinfo(skb_shinfo(skb)->frag_list);
649 nr_frags = shinfo->nr_frags;
658 static void xenvif_fill_frags(struct xenvif_queue *queue, struct sk_buff *skb)
660 struct skb_shared_info *shinfo = skb_shinfo(skb);
661 int nr_frags = shinfo->nr_frags;
663 u16 prev_pending_idx = INVALID_PENDING_IDX;
665 for (i = 0; i < nr_frags; i++) {
666 skb_frag_t *frag = shinfo->frags + i;
667 struct xen_netif_tx_request *txp;
671 pending_idx = frag_get_pending_idx(frag);
673 /* If this is not the first frag, chain it to the previous*/
674 if (prev_pending_idx == INVALID_PENDING_IDX)
675 skb_shinfo(skb)->destructor_arg =
676 &callback_param(queue, pending_idx);
678 callback_param(queue, prev_pending_idx).ctx =
679 &callback_param(queue, pending_idx);
681 callback_param(queue, pending_idx).ctx = NULL;
682 prev_pending_idx = pending_idx;
684 txp = &queue->pending_tx_info[pending_idx].req;
685 page = virt_to_page(idx_to_kaddr(queue, pending_idx));
686 __skb_fill_page_desc(skb, i, page, txp->offset, txp->size);
687 skb->len += txp->size;
688 skb->data_len += txp->size;
689 skb->truesize += txp->size;
691 /* Take an extra reference to offset network stack's put_page */
692 get_page(queue->mmap_pages[pending_idx]);
696 static int xenvif_get_extras(struct xenvif_queue *queue,
697 struct xen_netif_extra_info *extras,
698 unsigned int *extra_count,
701 struct xen_netif_extra_info extra;
702 RING_IDX cons = queue->tx.req_cons;
705 if (unlikely(work_to_do-- <= 0)) {
706 netdev_err(queue->vif->dev, "Missing extra info\n");
707 xenvif_fatal_tx_err(queue->vif);
711 RING_COPY_REQUEST(&queue->tx, cons, &extra);
713 queue->tx.req_cons = ++cons;
716 if (unlikely(!extra.type ||
717 extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
718 netdev_err(queue->vif->dev,
719 "Invalid extra type: %d\n", extra.type);
720 xenvif_fatal_tx_err(queue->vif);
724 memcpy(&extras[extra.type - 1], &extra, sizeof(extra));
725 } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE);
730 static int xenvif_set_skb_gso(struct xenvif *vif,
732 struct xen_netif_extra_info *gso)
734 if (!gso->u.gso.size) {
735 netdev_err(vif->dev, "GSO size must not be zero.\n");
736 xenvif_fatal_tx_err(vif);
740 switch (gso->u.gso.type) {
741 case XEN_NETIF_GSO_TYPE_TCPV4:
742 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
744 case XEN_NETIF_GSO_TYPE_TCPV6:
745 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
748 netdev_err(vif->dev, "Bad GSO type %d.\n", gso->u.gso.type);
749 xenvif_fatal_tx_err(vif);
753 skb_shinfo(skb)->gso_size = gso->u.gso.size;
754 /* gso_segs will be calculated later */
759 static int checksum_setup(struct xenvif_queue *queue, struct sk_buff *skb)
761 bool recalculate_partial_csum = false;
763 /* A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
764 * peers can fail to set NETRXF_csum_blank when sending a GSO
765 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
766 * recalculate the partial checksum.
768 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
769 queue->stats.rx_gso_checksum_fixup++;
770 skb->ip_summed = CHECKSUM_PARTIAL;
771 recalculate_partial_csum = true;
774 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
775 if (skb->ip_summed != CHECKSUM_PARTIAL)
778 return skb_checksum_setup(skb, recalculate_partial_csum);
781 static bool tx_credit_exceeded(struct xenvif_queue *queue, unsigned size)
783 u64 now = get_jiffies_64();
784 u64 next_credit = queue->credit_window_start +
785 msecs_to_jiffies(queue->credit_usec / 1000);
787 /* Timer could already be pending in rare cases. */
788 if (timer_pending(&queue->credit_timeout)) {
789 queue->rate_limited = true;
793 /* Passed the point where we can replenish credit? */
794 if (time_after_eq64(now, next_credit)) {
795 queue->credit_window_start = now;
796 tx_add_credit(queue);
799 /* Still too big to send right now? Set a callback. */
800 if (size > queue->remaining_credit) {
801 mod_timer(&queue->credit_timeout,
803 queue->credit_window_start = next_credit;
804 queue->rate_limited = true;
812 /* No locking is required in xenvif_mcast_add/del() as they are
813 * only ever invoked from NAPI poll. An RCU list is used because
814 * xenvif_mcast_match() is called asynchronously, during start_xmit.
817 static int xenvif_mcast_add(struct xenvif *vif, const u8 *addr)
819 struct xenvif_mcast_addr *mcast;
821 if (vif->fe_mcast_count == XEN_NETBK_MCAST_MAX) {
824 "Too many multicast addresses\n");
828 mcast = kzalloc(sizeof(*mcast), GFP_ATOMIC);
832 ether_addr_copy(mcast->addr, addr);
833 list_add_tail_rcu(&mcast->entry, &vif->fe_mcast_addr);
834 vif->fe_mcast_count++;
839 static void xenvif_mcast_del(struct xenvif *vif, const u8 *addr)
841 struct xenvif_mcast_addr *mcast;
843 list_for_each_entry_rcu(mcast, &vif->fe_mcast_addr, entry) {
844 if (ether_addr_equal(addr, mcast->addr)) {
845 --vif->fe_mcast_count;
846 list_del_rcu(&mcast->entry);
847 kfree_rcu(mcast, rcu);
853 bool xenvif_mcast_match(struct xenvif *vif, const u8 *addr)
855 struct xenvif_mcast_addr *mcast;
858 list_for_each_entry_rcu(mcast, &vif->fe_mcast_addr, entry) {
859 if (ether_addr_equal(addr, mcast->addr)) {
869 void xenvif_mcast_addr_list_free(struct xenvif *vif)
871 /* No need for locking or RCU here. NAPI poll and TX queue
874 while (!list_empty(&vif->fe_mcast_addr)) {
875 struct xenvif_mcast_addr *mcast;
877 mcast = list_first_entry(&vif->fe_mcast_addr,
878 struct xenvif_mcast_addr,
880 --vif->fe_mcast_count;
881 list_del(&mcast->entry);
886 static void xenvif_tx_build_gops(struct xenvif_queue *queue,
891 struct sk_buff *skb, *nskb;
893 unsigned int frag_overflow;
895 while (skb_queue_len(&queue->tx_queue) < budget) {
896 struct xen_netif_tx_request txreq;
897 struct xen_netif_tx_request txfrags[XEN_NETBK_LEGACY_SLOTS_MAX];
898 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX-1];
899 unsigned int extra_count;
903 unsigned int data_len;
904 pending_ring_idx_t index;
906 if (queue->tx.sring->req_prod - queue->tx.req_cons >
907 XEN_NETIF_TX_RING_SIZE) {
908 netdev_err(queue->vif->dev,
909 "Impossible number of requests. "
910 "req_prod %d, req_cons %d, size %ld\n",
911 queue->tx.sring->req_prod, queue->tx.req_cons,
912 XEN_NETIF_TX_RING_SIZE);
913 xenvif_fatal_tx_err(queue->vif);
917 work_to_do = RING_HAS_UNCONSUMED_REQUESTS(&queue->tx);
921 idx = queue->tx.req_cons;
922 rmb(); /* Ensure that we see the request before we copy it. */
923 RING_COPY_REQUEST(&queue->tx, idx, &txreq);
925 /* Credit-based scheduling. */
926 if (txreq.size > queue->remaining_credit &&
927 tx_credit_exceeded(queue, txreq.size))
930 queue->remaining_credit -= txreq.size;
933 queue->tx.req_cons = ++idx;
935 memset(extras, 0, sizeof(extras));
937 if (txreq.flags & XEN_NETTXF_extra_info) {
938 work_to_do = xenvif_get_extras(queue, extras,
941 idx = queue->tx.req_cons;
942 if (unlikely(work_to_do < 0))
946 if (extras[XEN_NETIF_EXTRA_TYPE_MCAST_ADD - 1].type) {
947 struct xen_netif_extra_info *extra;
949 extra = &extras[XEN_NETIF_EXTRA_TYPE_MCAST_ADD - 1];
950 ret = xenvif_mcast_add(queue->vif, extra->u.mcast.addr);
952 make_tx_response(queue, &txreq, extra_count,
955 XEN_NETIF_RSP_ERROR);
956 push_tx_responses(queue);
960 if (extras[XEN_NETIF_EXTRA_TYPE_MCAST_DEL - 1].type) {
961 struct xen_netif_extra_info *extra;
963 extra = &extras[XEN_NETIF_EXTRA_TYPE_MCAST_DEL - 1];
964 xenvif_mcast_del(queue->vif, extra->u.mcast.addr);
966 make_tx_response(queue, &txreq, extra_count,
968 push_tx_responses(queue);
972 data_len = (txreq.size > XEN_NETBACK_TX_COPY_LEN) ?
973 XEN_NETBACK_TX_COPY_LEN : txreq.size;
975 ret = xenvif_count_requests(queue, &txreq, extra_count,
976 txfrags, work_to_do);
978 if (unlikely(ret < 0))
983 if (unlikely(txreq.size < ETH_HLEN)) {
984 netdev_dbg(queue->vif->dev,
985 "Bad packet size: %d\n", txreq.size);
986 xenvif_tx_err(queue, &txreq, extra_count, idx);
990 /* No crossing a page as the payload mustn't fragment. */
991 if (unlikely((txreq.offset + txreq.size) > XEN_PAGE_SIZE)) {
992 netdev_err(queue->vif->dev, "Cross page boundary, txreq.offset: %u, size: %u\n",
993 txreq.offset, txreq.size);
994 xenvif_fatal_tx_err(queue->vif);
998 index = pending_index(queue->pending_cons);
999 pending_idx = queue->pending_ring[index];
1001 if (ret >= XEN_NETBK_LEGACY_SLOTS_MAX - 1 && data_len < txreq.size)
1002 data_len = txreq.size;
1004 skb = xenvif_alloc_skb(data_len);
1005 if (unlikely(skb == NULL)) {
1006 netdev_dbg(queue->vif->dev,
1007 "Can't allocate a skb in start_xmit.\n");
1008 xenvif_tx_err(queue, &txreq, extra_count, idx);
1012 skb_shinfo(skb)->nr_frags = ret;
1013 /* At this point shinfo->nr_frags is in fact the number of
1014 * slots, which can be as large as XEN_NETBK_LEGACY_SLOTS_MAX.
1018 if (skb_shinfo(skb)->nr_frags > MAX_SKB_FRAGS) {
1019 frag_overflow = skb_shinfo(skb)->nr_frags - MAX_SKB_FRAGS;
1020 BUG_ON(frag_overflow > MAX_SKB_FRAGS);
1021 skb_shinfo(skb)->nr_frags = MAX_SKB_FRAGS;
1022 nskb = xenvif_alloc_skb(0);
1023 if (unlikely(nskb == NULL)) {
1024 skb_shinfo(skb)->nr_frags = 0;
1026 xenvif_tx_err(queue, &txreq, extra_count, idx);
1027 if (net_ratelimit())
1028 netdev_err(queue->vif->dev,
1029 "Can't allocate the frag_list skb.\n");
1034 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1035 struct xen_netif_extra_info *gso;
1036 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1038 if (xenvif_set_skb_gso(queue->vif, skb, gso)) {
1039 /* Failure in xenvif_set_skb_gso is fatal. */
1040 skb_shinfo(skb)->nr_frags = 0;
1047 if (extras[XEN_NETIF_EXTRA_TYPE_HASH - 1].type) {
1048 struct xen_netif_extra_info *extra;
1049 enum pkt_hash_types type = PKT_HASH_TYPE_NONE;
1051 extra = &extras[XEN_NETIF_EXTRA_TYPE_HASH - 1];
1053 switch (extra->u.hash.type) {
1054 case _XEN_NETIF_CTRL_HASH_TYPE_IPV4:
1055 case _XEN_NETIF_CTRL_HASH_TYPE_IPV6:
1056 type = PKT_HASH_TYPE_L3;
1059 case _XEN_NETIF_CTRL_HASH_TYPE_IPV4_TCP:
1060 case _XEN_NETIF_CTRL_HASH_TYPE_IPV6_TCP:
1061 type = PKT_HASH_TYPE_L4;
1068 if (type != PKT_HASH_TYPE_NONE)
1070 *(u32 *)extra->u.hash.value,
1074 xenvif_get_requests(queue, skb, &txreq, txfrags, copy_ops,
1075 map_ops, frag_overflow, nskb, extra_count,
1078 __skb_queue_tail(&queue->tx_queue, skb);
1080 queue->tx.req_cons = idx;
1082 if ((*map_ops >= ARRAY_SIZE(queue->tx_map_ops)) ||
1083 (*copy_ops >= ARRAY_SIZE(queue->tx_copy_ops)))
1090 /* Consolidate skb with a frag_list into a brand new one with local pages on
1091 * frags. Returns 0 or -ENOMEM if can't allocate new pages.
1093 static int xenvif_handle_frag_list(struct xenvif_queue *queue, struct sk_buff *skb)
1095 unsigned int offset = skb_headlen(skb);
1096 skb_frag_t frags[MAX_SKB_FRAGS];
1098 struct ubuf_info *uarg;
1099 struct sk_buff *nskb = skb_shinfo(skb)->frag_list;
1101 queue->stats.tx_zerocopy_sent += 2;
1102 queue->stats.tx_frag_overflow++;
1104 xenvif_fill_frags(queue, nskb);
1105 /* Subtract frags size, we will correct it later */
1106 skb->truesize -= skb->data_len;
1107 skb->len += nskb->len;
1108 skb->data_len += nskb->len;
1110 /* create a brand new frags array and coalesce there */
1111 for (i = 0; offset < skb->len; i++) {
1115 BUG_ON(i >= MAX_SKB_FRAGS);
1116 page = alloc_page(GFP_ATOMIC);
1119 skb->truesize += skb->data_len;
1120 for (j = 0; j < i; j++)
1121 put_page(frags[j].page.p);
1125 if (offset + PAGE_SIZE < skb->len)
1128 len = skb->len - offset;
1129 if (skb_copy_bits(skb, offset, page_address(page), len))
1133 frags[i].page.p = page;
1134 frags[i].page_offset = 0;
1135 skb_frag_size_set(&frags[i], len);
1138 /* Release all the original (foreign) frags. */
1139 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
1140 skb_frag_unref(skb, f);
1141 uarg = skb_shinfo(skb)->destructor_arg;
1142 /* increase inflight counter to offset decrement in callback */
1143 atomic_inc(&queue->inflight_packets);
1144 uarg->callback(uarg, true);
1145 skb_shinfo(skb)->destructor_arg = NULL;
1147 /* Fill the skb with the new (local) frags. */
1148 memcpy(skb_shinfo(skb)->frags, frags, i * sizeof(skb_frag_t));
1149 skb_shinfo(skb)->nr_frags = i;
1150 skb->truesize += i * PAGE_SIZE;
1155 static int xenvif_tx_submit(struct xenvif_queue *queue)
1157 struct gnttab_map_grant_ref *gop_map = queue->tx_map_ops;
1158 struct gnttab_copy *gop_copy = queue->tx_copy_ops;
1159 struct sk_buff *skb;
1162 while ((skb = __skb_dequeue(&queue->tx_queue)) != NULL) {
1163 struct xen_netif_tx_request *txp;
1166 pending_idx = copy_pending_idx(skb, 0);
1167 txp = &queue->pending_tx_info[pending_idx].req;
1169 /* Check the remap error code. */
1170 if (unlikely(xenvif_tx_check_gop(queue, skb, &gop_map, &gop_copy))) {
1171 /* If there was an error, xenvif_tx_check_gop is
1172 * expected to release all the frags which were mapped,
1173 * so kfree_skb shouldn't do it again
1175 skb_shinfo(skb)->nr_frags = 0;
1176 if (skb_has_frag_list(skb)) {
1177 struct sk_buff *nskb =
1178 skb_shinfo(skb)->frag_list;
1179 skb_shinfo(nskb)->nr_frags = 0;
1185 if (txp->flags & XEN_NETTXF_csum_blank)
1186 skb->ip_summed = CHECKSUM_PARTIAL;
1187 else if (txp->flags & XEN_NETTXF_data_validated)
1188 skb->ip_summed = CHECKSUM_UNNECESSARY;
1190 xenvif_fill_frags(queue, skb);
1192 if (unlikely(skb_has_frag_list(skb))) {
1193 struct sk_buff *nskb = skb_shinfo(skb)->frag_list;
1194 xenvif_skb_zerocopy_prepare(queue, nskb);
1195 if (xenvif_handle_frag_list(queue, skb)) {
1196 if (net_ratelimit())
1197 netdev_err(queue->vif->dev,
1198 "Not enough memory to consolidate frag_list!\n");
1199 xenvif_skb_zerocopy_prepare(queue, skb);
1203 /* Copied all the bits from the frag list -- free it. */
1204 skb_frag_list_init(skb);
1208 skb->dev = queue->vif->dev;
1209 skb->protocol = eth_type_trans(skb, skb->dev);
1210 skb_reset_network_header(skb);
1212 if (checksum_setup(queue, skb)) {
1213 netdev_dbg(queue->vif->dev,
1214 "Can't setup checksum in net_tx_action\n");
1215 /* We have to set this flag to trigger the callback */
1216 if (skb_shinfo(skb)->destructor_arg)
1217 xenvif_skb_zerocopy_prepare(queue, skb);
1222 skb_probe_transport_header(skb, 0);
1224 /* If the packet is GSO then we will have just set up the
1225 * transport header offset in checksum_setup so it's now
1226 * straightforward to calculate gso_segs.
1228 if (skb_is_gso(skb)) {
1229 int mss = skb_shinfo(skb)->gso_size;
1230 int hdrlen = skb_transport_header(skb) -
1231 skb_mac_header(skb) +
1234 skb_shinfo(skb)->gso_segs =
1235 DIV_ROUND_UP(skb->len - hdrlen, mss);
1238 queue->stats.rx_bytes += skb->len;
1239 queue->stats.rx_packets++;
1243 /* Set this flag right before netif_receive_skb, otherwise
1244 * someone might think this packet already left netback, and
1245 * do a skb_copy_ubufs while we are still in control of the
1246 * skb. E.g. the __pskb_pull_tail earlier can do such thing.
1248 if (skb_shinfo(skb)->destructor_arg) {
1249 xenvif_skb_zerocopy_prepare(queue, skb);
1250 queue->stats.tx_zerocopy_sent++;
1253 netif_receive_skb(skb);
1259 void xenvif_zerocopy_callback(struct ubuf_info *ubuf, bool zerocopy_success)
1261 unsigned long flags;
1262 pending_ring_idx_t index;
1263 struct xenvif_queue *queue = ubuf_to_queue(ubuf);
1265 /* This is the only place where we grab this lock, to protect callbacks
1268 spin_lock_irqsave(&queue->callback_lock, flags);
1270 u16 pending_idx = ubuf->desc;
1271 ubuf = (struct ubuf_info *) ubuf->ctx;
1272 BUG_ON(queue->dealloc_prod - queue->dealloc_cons >=
1274 index = pending_index(queue->dealloc_prod);
1275 queue->dealloc_ring[index] = pending_idx;
1276 /* Sync with xenvif_tx_dealloc_action:
1277 * insert idx then incr producer.
1280 queue->dealloc_prod++;
1282 spin_unlock_irqrestore(&queue->callback_lock, flags);
1284 if (likely(zerocopy_success))
1285 queue->stats.tx_zerocopy_success++;
1287 queue->stats.tx_zerocopy_fail++;
1288 xenvif_skb_zerocopy_complete(queue);
1291 static inline void xenvif_tx_dealloc_action(struct xenvif_queue *queue)
1293 struct gnttab_unmap_grant_ref *gop;
1294 pending_ring_idx_t dc, dp;
1295 u16 pending_idx, pending_idx_release[MAX_PENDING_REQS];
1298 dc = queue->dealloc_cons;
1299 gop = queue->tx_unmap_ops;
1301 /* Free up any grants we have finished using */
1303 dp = queue->dealloc_prod;
1305 /* Ensure we see all indices enqueued by all
1306 * xenvif_zerocopy_callback().
1311 BUG_ON(gop - queue->tx_unmap_ops >= MAX_PENDING_REQS);
1313 queue->dealloc_ring[pending_index(dc++)];
1315 pending_idx_release[gop - queue->tx_unmap_ops] =
1317 queue->pages_to_unmap[gop - queue->tx_unmap_ops] =
1318 queue->mmap_pages[pending_idx];
1319 gnttab_set_unmap_op(gop,
1320 idx_to_kaddr(queue, pending_idx),
1322 queue->grant_tx_handle[pending_idx]);
1323 xenvif_grant_handle_reset(queue, pending_idx);
1327 } while (dp != queue->dealloc_prod);
1329 queue->dealloc_cons = dc;
1331 if (gop - queue->tx_unmap_ops > 0) {
1333 ret = gnttab_unmap_refs(queue->tx_unmap_ops,
1335 queue->pages_to_unmap,
1336 gop - queue->tx_unmap_ops);
1338 netdev_err(queue->vif->dev, "Unmap fail: nr_ops %tu ret %d\n",
1339 gop - queue->tx_unmap_ops, ret);
1340 for (i = 0; i < gop - queue->tx_unmap_ops; ++i) {
1341 if (gop[i].status != GNTST_okay)
1342 netdev_err(queue->vif->dev,
1343 " host_addr: 0x%llx handle: 0x%x status: %d\n",
1352 for (i = 0; i < gop - queue->tx_unmap_ops; ++i)
1353 xenvif_idx_release(queue, pending_idx_release[i],
1354 XEN_NETIF_RSP_OKAY);
1358 /* Called after netfront has transmitted */
1359 int xenvif_tx_action(struct xenvif_queue *queue, int budget)
1361 unsigned nr_mops = 0, nr_cops = 0;
1364 if (unlikely(!tx_work_todo(queue)))
1367 xenvif_tx_build_gops(queue, budget, &nr_cops, &nr_mops);
1372 gnttab_batch_copy(queue->tx_copy_ops, nr_cops);
1374 ret = gnttab_map_refs(queue->tx_map_ops,
1376 queue->pages_to_map,
1381 netdev_err(queue->vif->dev, "Map fail: nr %u ret %d\n",
1383 for (i = 0; i < nr_mops; ++i)
1384 WARN_ON_ONCE(queue->tx_map_ops[i].status ==
1389 work_done = xenvif_tx_submit(queue);
1394 static void xenvif_idx_release(struct xenvif_queue *queue, u16 pending_idx,
1397 struct pending_tx_info *pending_tx_info;
1398 pending_ring_idx_t index;
1399 unsigned long flags;
1401 pending_tx_info = &queue->pending_tx_info[pending_idx];
1403 spin_lock_irqsave(&queue->response_lock, flags);
1405 make_tx_response(queue, &pending_tx_info->req,
1406 pending_tx_info->extra_count, status);
1408 /* Release the pending index before pusing the Tx response so
1409 * its available before a new Tx request is pushed by the
1412 index = pending_index(queue->pending_prod++);
1413 queue->pending_ring[index] = pending_idx;
1415 push_tx_responses(queue);
1417 spin_unlock_irqrestore(&queue->response_lock, flags);
1421 static void make_tx_response(struct xenvif_queue *queue,
1422 struct xen_netif_tx_request *txp,
1423 unsigned int extra_count,
1426 RING_IDX i = queue->tx.rsp_prod_pvt;
1427 struct xen_netif_tx_response *resp;
1429 resp = RING_GET_RESPONSE(&queue->tx, i);
1433 while (extra_count-- != 0)
1434 RING_GET_RESPONSE(&queue->tx, ++i)->status = XEN_NETIF_RSP_NULL;
1436 queue->tx.rsp_prod_pvt = ++i;
1439 static void push_tx_responses(struct xenvif_queue *queue)
1443 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&queue->tx, notify);
1445 notify_remote_via_irq(queue->tx_irq);
1448 static void xenvif_idx_unmap(struct xenvif_queue *queue, u16 pending_idx)
1451 struct gnttab_unmap_grant_ref tx_unmap_op;
1453 gnttab_set_unmap_op(&tx_unmap_op,
1454 idx_to_kaddr(queue, pending_idx),
1456 queue->grant_tx_handle[pending_idx]);
1457 xenvif_grant_handle_reset(queue, pending_idx);
1459 ret = gnttab_unmap_refs(&tx_unmap_op, NULL,
1460 &queue->mmap_pages[pending_idx], 1);
1462 netdev_err(queue->vif->dev,
1463 "Unmap fail: ret: %d pending_idx: %d host_addr: %llx handle: 0x%x status: %d\n",
1466 tx_unmap_op.host_addr,
1468 tx_unmap_op.status);
1473 static inline int tx_work_todo(struct xenvif_queue *queue)
1475 if (likely(RING_HAS_UNCONSUMED_REQUESTS(&queue->tx)))
1481 static inline bool tx_dealloc_work_todo(struct xenvif_queue *queue)
1483 return queue->dealloc_cons != queue->dealloc_prod;
1486 void xenvif_unmap_frontend_data_rings(struct xenvif_queue *queue)
1488 if (queue->tx.sring)
1489 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(queue->vif),
1491 if (queue->rx.sring)
1492 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(queue->vif),
1496 int xenvif_map_frontend_data_rings(struct xenvif_queue *queue,
1497 grant_ref_t tx_ring_ref,
1498 grant_ref_t rx_ring_ref)
1501 struct xen_netif_tx_sring *txs;
1502 struct xen_netif_rx_sring *rxs;
1506 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(queue->vif),
1507 &tx_ring_ref, 1, &addr);
1511 txs = (struct xen_netif_tx_sring *)addr;
1512 BACK_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE);
1514 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(queue->vif),
1515 &rx_ring_ref, 1, &addr);
1519 rxs = (struct xen_netif_rx_sring *)addr;
1520 BACK_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE);
1525 xenvif_unmap_frontend_data_rings(queue);
1529 static bool xenvif_dealloc_kthread_should_stop(struct xenvif_queue *queue)
1531 /* Dealloc thread must remain running until all inflight
1534 return kthread_should_stop() &&
1535 !atomic_read(&queue->inflight_packets);
1538 int xenvif_dealloc_kthread(void *data)
1540 struct xenvif_queue *queue = data;
1543 wait_event_interruptible(queue->dealloc_wq,
1544 tx_dealloc_work_todo(queue) ||
1545 xenvif_dealloc_kthread_should_stop(queue));
1546 if (xenvif_dealloc_kthread_should_stop(queue))
1549 xenvif_tx_dealloc_action(queue);
1553 /* Unmap anything remaining*/
1554 if (tx_dealloc_work_todo(queue))
1555 xenvif_tx_dealloc_action(queue);
1560 static void make_ctrl_response(struct xenvif *vif,
1561 const struct xen_netif_ctrl_request *req,
1562 u32 status, u32 data)
1564 RING_IDX idx = vif->ctrl.rsp_prod_pvt;
1565 struct xen_netif_ctrl_response rsp = {
1572 *RING_GET_RESPONSE(&vif->ctrl, idx) = rsp;
1573 vif->ctrl.rsp_prod_pvt = ++idx;
1576 static void push_ctrl_response(struct xenvif *vif)
1580 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->ctrl, notify);
1582 notify_remote_via_irq(vif->ctrl_irq);
1585 static void process_ctrl_request(struct xenvif *vif,
1586 const struct xen_netif_ctrl_request *req)
1588 u32 status = XEN_NETIF_CTRL_STATUS_NOT_SUPPORTED;
1591 switch (req->type) {
1592 case XEN_NETIF_CTRL_TYPE_SET_HASH_ALGORITHM:
1593 status = xenvif_set_hash_alg(vif, req->data[0]);
1596 case XEN_NETIF_CTRL_TYPE_GET_HASH_FLAGS:
1597 status = xenvif_get_hash_flags(vif, &data);
1600 case XEN_NETIF_CTRL_TYPE_SET_HASH_FLAGS:
1601 status = xenvif_set_hash_flags(vif, req->data[0]);
1604 case XEN_NETIF_CTRL_TYPE_SET_HASH_KEY:
1605 status = xenvif_set_hash_key(vif, req->data[0],
1609 case XEN_NETIF_CTRL_TYPE_GET_HASH_MAPPING_SIZE:
1610 status = XEN_NETIF_CTRL_STATUS_SUCCESS;
1611 data = XEN_NETBK_MAX_HASH_MAPPING_SIZE;
1614 case XEN_NETIF_CTRL_TYPE_SET_HASH_MAPPING_SIZE:
1615 status = xenvif_set_hash_mapping_size(vif,
1619 case XEN_NETIF_CTRL_TYPE_SET_HASH_MAPPING:
1620 status = xenvif_set_hash_mapping(vif, req->data[0],
1629 make_ctrl_response(vif, req, status, data);
1630 push_ctrl_response(vif);
1633 static void xenvif_ctrl_action(struct xenvif *vif)
1636 RING_IDX req_prod, req_cons;
1638 req_prod = vif->ctrl.sring->req_prod;
1639 req_cons = vif->ctrl.req_cons;
1641 /* Make sure we can see requests before we process them. */
1644 if (req_cons == req_prod)
1647 while (req_cons != req_prod) {
1648 struct xen_netif_ctrl_request req;
1650 RING_COPY_REQUEST(&vif->ctrl, req_cons, &req);
1653 process_ctrl_request(vif, &req);
1656 vif->ctrl.req_cons = req_cons;
1657 vif->ctrl.sring->req_event = req_cons + 1;
1661 static bool xenvif_ctrl_work_todo(struct xenvif *vif)
1663 if (likely(RING_HAS_UNCONSUMED_REQUESTS(&vif->ctrl)))
1669 irqreturn_t xenvif_ctrl_irq_fn(int irq, void *data)
1671 struct xenvif *vif = data;
1672 unsigned int eoi_flag = XEN_EOI_FLAG_SPURIOUS;
1674 while (xenvif_ctrl_work_todo(vif)) {
1675 xenvif_ctrl_action(vif);
1679 xen_irq_lateeoi(irq, eoi_flag);
1684 static int __init netback_init(void)
1691 /* Allow as many queues as there are CPUs but max. 8 if user has not
1692 * specified a value.
1694 if (xenvif_max_queues == 0)
1695 xenvif_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT,
1698 if (fatal_skb_slots < XEN_NETBK_LEGACY_SLOTS_MAX) {
1699 pr_info("fatal_skb_slots too small (%d), bump it to XEN_NETBK_LEGACY_SLOTS_MAX (%d)\n",
1700 fatal_skb_slots, XEN_NETBK_LEGACY_SLOTS_MAX);
1701 fatal_skb_slots = XEN_NETBK_LEGACY_SLOTS_MAX;
1704 rc = xenvif_xenbus_init();
1708 #ifdef CONFIG_DEBUG_FS
1709 xen_netback_dbg_root = debugfs_create_dir("xen-netback", NULL);
1710 if (IS_ERR_OR_NULL(xen_netback_dbg_root))
1711 pr_warn("Init of debugfs returned %ld!\n",
1712 PTR_ERR(xen_netback_dbg_root));
1713 #endif /* CONFIG_DEBUG_FS */
1721 module_init(netback_init);
1723 static void __exit netback_fini(void)
1725 #ifdef CONFIG_DEBUG_FS
1726 if (!IS_ERR_OR_NULL(xen_netback_dbg_root))
1727 debugfs_remove_recursive(xen_netback_dbg_root);
1728 #endif /* CONFIG_DEBUG_FS */
1729 xenvif_xenbus_fini();
1731 module_exit(netback_fini);
1733 MODULE_LICENSE("Dual BSD/GPL");
1734 MODULE_ALIAS("xen-backend:vif");