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 const struct xen_netif_tx_request *txp,
111 unsigned int extra_count,
114 static void xenvif_idx_unmap(struct xenvif_queue *queue, u16 pending_idx);
116 static inline int tx_work_todo(struct xenvif_queue *queue);
118 static inline unsigned long idx_to_pfn(struct xenvif_queue *queue,
121 return page_to_pfn(queue->mmap_pages[idx]);
124 static inline unsigned long idx_to_kaddr(struct xenvif_queue *queue,
127 return (unsigned long)pfn_to_kaddr(idx_to_pfn(queue, idx));
130 #define callback_param(vif, pending_idx) \
131 (vif->pending_tx_info[pending_idx].callback_struct)
133 /* Find the containing VIF's structure from a pointer in pending_tx_info array
135 static inline struct xenvif_queue *ubuf_to_queue(const struct ubuf_info_msgzc *ubuf)
137 u16 pending_idx = ubuf->desc;
138 struct pending_tx_info *temp =
139 container_of(ubuf, struct pending_tx_info, callback_struct);
140 return container_of(temp - pending_idx,
145 static u16 frag_get_pending_idx(skb_frag_t *frag)
147 return (u16)skb_frag_off(frag);
150 static void frag_set_pending_idx(skb_frag_t *frag, u16 pending_idx)
152 skb_frag_off_set(frag, pending_idx);
155 static inline pending_ring_idx_t pending_index(unsigned i)
157 return i & (MAX_PENDING_REQS-1);
160 void xenvif_kick_thread(struct xenvif_queue *queue)
165 void xenvif_napi_schedule_or_enable_events(struct xenvif_queue *queue)
169 RING_FINAL_CHECK_FOR_REQUESTS(&queue->tx, more_to_do);
172 napi_schedule(&queue->napi);
173 else if (atomic_fetch_andnot(NETBK_TX_EOI | NETBK_COMMON_EOI,
174 &queue->eoi_pending) &
175 (NETBK_TX_EOI | NETBK_COMMON_EOI))
176 xen_irq_lateeoi(queue->tx_irq, 0);
179 static void tx_add_credit(struct xenvif_queue *queue)
181 unsigned long max_burst, max_credit;
184 * Allow a burst big enough to transmit a jumbo packet of up to 128kB.
185 * Otherwise the interface can seize up due to insufficient credit.
187 max_burst = max(131072UL, queue->credit_bytes);
189 /* Take care that adding a new chunk of credit doesn't wrap to zero. */
190 max_credit = queue->remaining_credit + queue->credit_bytes;
191 if (max_credit < queue->remaining_credit)
192 max_credit = ULONG_MAX; /* wrapped: clamp to ULONG_MAX */
194 queue->remaining_credit = min(max_credit, max_burst);
195 queue->rate_limited = false;
198 void xenvif_tx_credit_callback(struct timer_list *t)
200 struct xenvif_queue *queue = from_timer(queue, t, credit_timeout);
201 tx_add_credit(queue);
202 xenvif_napi_schedule_or_enable_events(queue);
205 static void xenvif_tx_err(struct xenvif_queue *queue,
206 struct xen_netif_tx_request *txp,
207 unsigned int extra_count, RING_IDX end)
209 RING_IDX cons = queue->tx.req_cons;
212 make_tx_response(queue, txp, extra_count, XEN_NETIF_RSP_ERROR);
215 RING_COPY_REQUEST(&queue->tx, cons++, txp);
216 extra_count = 0; /* only the first frag can have extras */
218 queue->tx.req_cons = cons;
221 static void xenvif_fatal_tx_err(struct xenvif *vif)
223 netdev_err(vif->dev, "fatal error; disabling device\n");
224 vif->disabled = true;
225 /* Disable the vif from queue 0's kthread */
227 xenvif_kick_thread(&vif->queues[0]);
230 static int xenvif_count_requests(struct xenvif_queue *queue,
231 struct xen_netif_tx_request *first,
232 unsigned int extra_count,
233 struct xen_netif_tx_request *txp,
236 RING_IDX cons = queue->tx.req_cons;
241 if (!(first->flags & XEN_NETTXF_more_data))
245 struct xen_netif_tx_request dropped_tx = { 0 };
247 if (slots >= work_to_do) {
248 netdev_err(queue->vif->dev,
249 "Asked for %d slots but exceeds this limit\n",
251 xenvif_fatal_tx_err(queue->vif);
255 /* This guest is really using too many slots and
256 * considered malicious.
258 if (unlikely(slots >= fatal_skb_slots)) {
259 netdev_err(queue->vif->dev,
260 "Malicious frontend using %d slots, threshold %u\n",
261 slots, fatal_skb_slots);
262 xenvif_fatal_tx_err(queue->vif);
266 /* Xen network protocol had implicit dependency on
267 * MAX_SKB_FRAGS. XEN_NETBK_LEGACY_SLOTS_MAX is set to
268 * the historical MAX_SKB_FRAGS value 18 to honor the
269 * same behavior as before. Any packet using more than
270 * 18 slots but less than fatal_skb_slots slots is
273 if (!drop_err && slots >= XEN_NETBK_LEGACY_SLOTS_MAX) {
275 netdev_dbg(queue->vif->dev,
276 "Too many slots (%d) exceeding limit (%d), dropping packet\n",
277 slots, XEN_NETBK_LEGACY_SLOTS_MAX);
284 RING_COPY_REQUEST(&queue->tx, cons + slots, txp);
286 /* If the guest submitted a frame >= 64 KiB then
287 * first->size overflowed and following slots will
288 * appear to be larger than the frame.
290 * This cannot be fatal error as there are buggy
291 * frontends that do this.
293 * Consume all slots and drop the packet.
295 if (!drop_err && txp->size > first->size) {
297 netdev_dbg(queue->vif->dev,
298 "Invalid tx request, slot size %u > remaining size %u\n",
299 txp->size, first->size);
303 first->size -= txp->size;
306 if (unlikely((txp->offset + txp->size) > XEN_PAGE_SIZE)) {
307 netdev_err(queue->vif->dev, "Cross page boundary, txp->offset: %u, size: %u\n",
308 txp->offset, txp->size);
309 xenvif_fatal_tx_err(queue->vif);
313 more_data = txp->flags & XEN_NETTXF_more_data;
321 xenvif_tx_err(queue, first, extra_count, cons + slots);
329 struct xenvif_tx_cb {
330 u16 copy_pending_idx[XEN_NETBK_LEGACY_SLOTS_MAX + 1];
335 #define XENVIF_TX_CB(skb) ((struct xenvif_tx_cb *)(skb)->cb)
336 #define copy_pending_idx(skb, i) (XENVIF_TX_CB(skb)->copy_pending_idx[i])
337 #define copy_count(skb) (XENVIF_TX_CB(skb)->copy_count)
339 static inline void xenvif_tx_create_map_op(struct xenvif_queue *queue,
341 struct xen_netif_tx_request *txp,
342 unsigned int extra_count,
343 struct gnttab_map_grant_ref *mop)
345 queue->pages_to_map[mop-queue->tx_map_ops] = queue->mmap_pages[pending_idx];
346 gnttab_set_map_op(mop, idx_to_kaddr(queue, pending_idx),
347 GNTMAP_host_map | GNTMAP_readonly,
348 txp->gref, queue->vif->domid);
350 memcpy(&queue->pending_tx_info[pending_idx].req, txp,
352 queue->pending_tx_info[pending_idx].extra_count = extra_count;
355 static inline struct sk_buff *xenvif_alloc_skb(unsigned int size)
357 struct sk_buff *skb =
358 alloc_skb(size + NET_SKB_PAD + NET_IP_ALIGN,
359 GFP_ATOMIC | __GFP_NOWARN);
361 BUILD_BUG_ON(sizeof(*XENVIF_TX_CB(skb)) > sizeof(skb->cb));
362 if (unlikely(skb == NULL))
365 /* Packets passed to netif_rx() must have some headroom. */
366 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
368 /* Initialize it here to avoid later surprises */
369 skb_shinfo(skb)->destructor_arg = NULL;
374 static void xenvif_get_requests(struct xenvif_queue *queue,
376 struct xen_netif_tx_request *first,
377 struct xen_netif_tx_request *txfrags,
380 unsigned int frag_overflow,
381 struct sk_buff *nskb,
382 unsigned int extra_count,
383 unsigned int data_len)
385 struct skb_shared_info *shinfo = skb_shinfo(skb);
386 skb_frag_t *frags = shinfo->frags;
388 pending_ring_idx_t index;
389 unsigned int nr_slots;
390 struct gnttab_copy *cop = queue->tx_copy_ops + *copy_ops;
391 struct gnttab_map_grant_ref *gop = queue->tx_map_ops + *map_ops;
392 struct xen_netif_tx_request *txp = first;
394 nr_slots = shinfo->nr_frags + frag_overflow + 1;
397 XENVIF_TX_CB(skb)->split_mask = 0;
399 /* Create copy ops for exactly data_len bytes into the skb head. */
400 __skb_put(skb, data_len);
401 while (data_len > 0) {
402 int amount = data_len > txp->size ? txp->size : data_len;
405 cop->source.u.ref = txp->gref;
406 cop->source.domid = queue->vif->domid;
407 cop->source.offset = txp->offset;
409 cop->dest.domid = DOMID_SELF;
410 cop->dest.offset = (offset_in_page(skb->data +
412 data_len)) & ~XEN_PAGE_MASK;
413 cop->dest.u.gmfn = virt_to_gfn(skb->data + skb_headlen(skb)
416 /* Don't cross local page boundary! */
417 if (cop->dest.offset + amount > XEN_PAGE_SIZE) {
418 amount = XEN_PAGE_SIZE - cop->dest.offset;
419 XENVIF_TX_CB(skb)->split_mask |= 1U << copy_count(skb);
424 cop->flags = GNTCOPY_source_gref;
426 index = pending_index(queue->pending_cons);
427 pending_idx = queue->pending_ring[index];
428 callback_param(queue, pending_idx).ctx = NULL;
429 copy_pending_idx(skb, copy_count(skb)) = pending_idx;
436 if (amount == txp->size) {
437 /* The copy op covered the full tx_request */
439 memcpy(&queue->pending_tx_info[pending_idx].req,
441 queue->pending_tx_info[pending_idx].extra_count =
442 (txp == first) ? extra_count : 0;
448 queue->pending_cons++;
451 /* The copy op partially covered the tx_request.
452 * The remainder will be mapped or copied in the next
455 txp->offset += amount;
460 for (shinfo->nr_frags = 0; nr_slots > 0 && shinfo->nr_frags < MAX_SKB_FRAGS;
462 if (unlikely(!txp->size)) {
463 make_tx_response(queue, txp, 0, XEN_NETIF_RSP_OKAY);
468 index = pending_index(queue->pending_cons++);
469 pending_idx = queue->pending_ring[index];
470 xenvif_tx_create_map_op(queue, pending_idx, txp,
471 txp == first ? extra_count : 0, gop);
472 frag_set_pending_idx(&frags[shinfo->nr_frags], pending_idx);
484 shinfo = skb_shinfo(nskb);
485 frags = shinfo->frags;
487 for (shinfo->nr_frags = 0; shinfo->nr_frags < nr_slots; ++txp) {
488 if (unlikely(!txp->size)) {
489 make_tx_response(queue, txp, 0,
494 index = pending_index(queue->pending_cons++);
495 pending_idx = queue->pending_ring[index];
496 xenvif_tx_create_map_op(queue, pending_idx, txp, 0,
498 frag_set_pending_idx(&frags[shinfo->nr_frags],
504 if (shinfo->nr_frags) {
505 skb_shinfo(skb)->frag_list = nskb;
511 /* A frag_list skb was allocated but it is no longer needed
512 * because enough slots were converted to copy ops above or some
518 (*copy_ops) = cop - queue->tx_copy_ops;
519 (*map_ops) = gop - queue->tx_map_ops;
522 static inline void xenvif_grant_handle_set(struct xenvif_queue *queue,
524 grant_handle_t handle)
526 if (unlikely(queue->grant_tx_handle[pending_idx] !=
527 NETBACK_INVALID_HANDLE)) {
528 netdev_err(queue->vif->dev,
529 "Trying to overwrite active handle! pending_idx: 0x%x\n",
533 queue->grant_tx_handle[pending_idx] = handle;
536 static inline void xenvif_grant_handle_reset(struct xenvif_queue *queue,
539 if (unlikely(queue->grant_tx_handle[pending_idx] ==
540 NETBACK_INVALID_HANDLE)) {
541 netdev_err(queue->vif->dev,
542 "Trying to unmap invalid handle! pending_idx: 0x%x\n",
546 queue->grant_tx_handle[pending_idx] = NETBACK_INVALID_HANDLE;
549 static int xenvif_tx_check_gop(struct xenvif_queue *queue,
551 struct gnttab_map_grant_ref **gopp_map,
552 struct gnttab_copy **gopp_copy)
554 struct gnttab_map_grant_ref *gop_map = *gopp_map;
556 /* This always points to the shinfo of the skb being checked, which
557 * could be either the first or the one on the frag_list
559 struct skb_shared_info *shinfo = skb_shinfo(skb);
560 /* If this is non-NULL, we are currently checking the frag_list skb, and
561 * this points to the shinfo of the first one
563 struct skb_shared_info *first_shinfo = NULL;
564 int nr_frags = shinfo->nr_frags;
565 const bool sharedslot = nr_frags &&
566 frag_get_pending_idx(&shinfo->frags[0]) ==
567 copy_pending_idx(skb, copy_count(skb) - 1);
570 for (i = 0; i < copy_count(skb); i++) {
573 /* Check status of header. */
574 pending_idx = copy_pending_idx(skb, i);
576 newerr = (*gopp_copy)->status;
578 /* Split copies need to be handled together. */
579 if (XENVIF_TX_CB(skb)->split_mask & (1U << i)) {
582 newerr = (*gopp_copy)->status;
584 if (likely(!newerr)) {
585 /* The first frag might still have this slot mapped */
586 if (i < copy_count(skb) - 1 || !sharedslot)
587 xenvif_idx_release(queue, pending_idx,
592 netdev_dbg(queue->vif->dev,
593 "Grant copy of header failed! status: %d pending_idx: %u ref: %u\n",
594 (*gopp_copy)->status,
596 (*gopp_copy)->source.u.ref);
597 /* The first frag might still have this slot mapped */
598 if (i < copy_count(skb) - 1 || !sharedslot)
599 xenvif_idx_release(queue, pending_idx,
600 XEN_NETIF_RSP_ERROR);
606 for (i = 0; i < nr_frags; i++, gop_map++) {
609 pending_idx = frag_get_pending_idx(&shinfo->frags[i]);
611 /* Check error status: if okay then remember grant handle. */
612 newerr = gop_map->status;
614 if (likely(!newerr)) {
615 xenvif_grant_handle_set(queue,
618 /* Had a previous error? Invalidate this fragment. */
620 xenvif_idx_unmap(queue, pending_idx);
621 /* If the mapping of the first frag was OK, but
622 * the header's copy failed, and they are
623 * sharing a slot, send an error
625 if (i == 0 && !first_shinfo && sharedslot)
626 xenvif_idx_release(queue, pending_idx,
627 XEN_NETIF_RSP_ERROR);
629 xenvif_idx_release(queue, pending_idx,
635 /* Error on this fragment: respond to client with an error. */
637 netdev_dbg(queue->vif->dev,
638 "Grant map of %d. frag failed! status: %d pending_idx: %u ref: %u\n",
644 xenvif_idx_release(queue, pending_idx, XEN_NETIF_RSP_ERROR);
646 /* Not the first error? Preceding frags already invalidated. */
650 /* Invalidate preceding fragments of this skb. */
651 for (j = 0; j < i; j++) {
652 pending_idx = frag_get_pending_idx(&shinfo->frags[j]);
653 xenvif_idx_unmap(queue, pending_idx);
654 xenvif_idx_release(queue, pending_idx,
658 /* And if we found the error while checking the frag_list, unmap
659 * the first skb's frags
662 for (j = 0; j < first_shinfo->nr_frags; j++) {
663 pending_idx = frag_get_pending_idx(&first_shinfo->frags[j]);
664 xenvif_idx_unmap(queue, pending_idx);
665 xenvif_idx_release(queue, pending_idx,
670 /* Remember the error: invalidate all subsequent fragments. */
674 if (skb_has_frag_list(skb) && !first_shinfo) {
675 first_shinfo = shinfo;
676 shinfo = skb_shinfo(shinfo->frag_list);
677 nr_frags = shinfo->nr_frags;
686 static void xenvif_fill_frags(struct xenvif_queue *queue, struct sk_buff *skb)
688 struct skb_shared_info *shinfo = skb_shinfo(skb);
689 int nr_frags = shinfo->nr_frags;
691 u16 prev_pending_idx = INVALID_PENDING_IDX;
693 for (i = 0; i < nr_frags; i++) {
694 skb_frag_t *frag = shinfo->frags + i;
695 struct xen_netif_tx_request *txp;
699 pending_idx = frag_get_pending_idx(frag);
701 /* If this is not the first frag, chain it to the previous*/
702 if (prev_pending_idx == INVALID_PENDING_IDX)
703 skb_shinfo(skb)->destructor_arg =
704 &callback_param(queue, pending_idx);
706 callback_param(queue, prev_pending_idx).ctx =
707 &callback_param(queue, pending_idx);
709 callback_param(queue, pending_idx).ctx = NULL;
710 prev_pending_idx = pending_idx;
712 txp = &queue->pending_tx_info[pending_idx].req;
713 page = virt_to_page((void *)idx_to_kaddr(queue, pending_idx));
714 __skb_fill_page_desc(skb, i, page, txp->offset, txp->size);
715 skb->len += txp->size;
716 skb->data_len += txp->size;
717 skb->truesize += txp->size;
719 /* Take an extra reference to offset network stack's put_page */
720 get_page(queue->mmap_pages[pending_idx]);
724 static int xenvif_get_extras(struct xenvif_queue *queue,
725 struct xen_netif_extra_info *extras,
726 unsigned int *extra_count,
729 struct xen_netif_extra_info extra;
730 RING_IDX cons = queue->tx.req_cons;
733 if (unlikely(work_to_do-- <= 0)) {
734 netdev_err(queue->vif->dev, "Missing extra info\n");
735 xenvif_fatal_tx_err(queue->vif);
739 RING_COPY_REQUEST(&queue->tx, cons, &extra);
741 queue->tx.req_cons = ++cons;
744 if (unlikely(!extra.type ||
745 extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
746 netdev_err(queue->vif->dev,
747 "Invalid extra type: %d\n", extra.type);
748 xenvif_fatal_tx_err(queue->vif);
752 memcpy(&extras[extra.type - 1], &extra, sizeof(extra));
753 } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE);
758 static int xenvif_set_skb_gso(struct xenvif *vif,
760 struct xen_netif_extra_info *gso)
762 if (!gso->u.gso.size) {
763 netdev_err(vif->dev, "GSO size must not be zero.\n");
764 xenvif_fatal_tx_err(vif);
768 switch (gso->u.gso.type) {
769 case XEN_NETIF_GSO_TYPE_TCPV4:
770 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
772 case XEN_NETIF_GSO_TYPE_TCPV6:
773 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
776 netdev_err(vif->dev, "Bad GSO type %d.\n", gso->u.gso.type);
777 xenvif_fatal_tx_err(vif);
781 skb_shinfo(skb)->gso_size = gso->u.gso.size;
782 /* gso_segs will be calculated later */
787 static int checksum_setup(struct xenvif_queue *queue, struct sk_buff *skb)
789 bool recalculate_partial_csum = false;
791 /* A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
792 * peers can fail to set NETRXF_csum_blank when sending a GSO
793 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
794 * recalculate the partial checksum.
796 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
797 queue->stats.rx_gso_checksum_fixup++;
798 skb->ip_summed = CHECKSUM_PARTIAL;
799 recalculate_partial_csum = true;
802 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
803 if (skb->ip_summed != CHECKSUM_PARTIAL)
806 return skb_checksum_setup(skb, recalculate_partial_csum);
809 static bool tx_credit_exceeded(struct xenvif_queue *queue, unsigned size)
811 u64 now = get_jiffies_64();
812 u64 next_credit = queue->credit_window_start +
813 msecs_to_jiffies(queue->credit_usec / 1000);
815 /* Timer could already be pending in rare cases. */
816 if (timer_pending(&queue->credit_timeout)) {
817 queue->rate_limited = true;
821 /* Passed the point where we can replenish credit? */
822 if (time_after_eq64(now, next_credit)) {
823 queue->credit_window_start = now;
824 tx_add_credit(queue);
827 /* Still too big to send right now? Set a callback. */
828 if (size > queue->remaining_credit) {
829 mod_timer(&queue->credit_timeout,
831 queue->credit_window_start = next_credit;
832 queue->rate_limited = true;
840 /* No locking is required in xenvif_mcast_add/del() as they are
841 * only ever invoked from NAPI poll. An RCU list is used because
842 * xenvif_mcast_match() is called asynchronously, during start_xmit.
845 static int xenvif_mcast_add(struct xenvif *vif, const u8 *addr)
847 struct xenvif_mcast_addr *mcast;
849 if (vif->fe_mcast_count == XEN_NETBK_MCAST_MAX) {
852 "Too many multicast addresses\n");
856 mcast = kzalloc(sizeof(*mcast), GFP_ATOMIC);
860 ether_addr_copy(mcast->addr, addr);
861 list_add_tail_rcu(&mcast->entry, &vif->fe_mcast_addr);
862 vif->fe_mcast_count++;
867 static void xenvif_mcast_del(struct xenvif *vif, const u8 *addr)
869 struct xenvif_mcast_addr *mcast;
871 list_for_each_entry_rcu(mcast, &vif->fe_mcast_addr, entry) {
872 if (ether_addr_equal(addr, mcast->addr)) {
873 --vif->fe_mcast_count;
874 list_del_rcu(&mcast->entry);
875 kfree_rcu(mcast, rcu);
881 bool xenvif_mcast_match(struct xenvif *vif, const u8 *addr)
883 struct xenvif_mcast_addr *mcast;
886 list_for_each_entry_rcu(mcast, &vif->fe_mcast_addr, entry) {
887 if (ether_addr_equal(addr, mcast->addr)) {
897 void xenvif_mcast_addr_list_free(struct xenvif *vif)
899 /* No need for locking or RCU here. NAPI poll and TX queue
902 while (!list_empty(&vif->fe_mcast_addr)) {
903 struct xenvif_mcast_addr *mcast;
905 mcast = list_first_entry(&vif->fe_mcast_addr,
906 struct xenvif_mcast_addr,
908 --vif->fe_mcast_count;
909 list_del(&mcast->entry);
914 static void xenvif_tx_build_gops(struct xenvif_queue *queue,
919 struct sk_buff *skb, *nskb;
921 unsigned int frag_overflow;
923 while (skb_queue_len(&queue->tx_queue) < budget) {
924 struct xen_netif_tx_request txreq;
925 struct xen_netif_tx_request txfrags[XEN_NETBK_LEGACY_SLOTS_MAX];
926 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX-1];
927 unsigned int extra_count;
930 unsigned int data_len;
932 if (queue->tx.sring->req_prod - queue->tx.req_cons >
933 XEN_NETIF_TX_RING_SIZE) {
934 netdev_err(queue->vif->dev,
935 "Impossible number of requests. "
936 "req_prod %d, req_cons %d, size %ld\n",
937 queue->tx.sring->req_prod, queue->tx.req_cons,
938 XEN_NETIF_TX_RING_SIZE);
939 xenvif_fatal_tx_err(queue->vif);
943 work_to_do = XEN_RING_NR_UNCONSUMED_REQUESTS(&queue->tx);
947 idx = queue->tx.req_cons;
948 rmb(); /* Ensure that we see the request before we copy it. */
949 RING_COPY_REQUEST(&queue->tx, idx, &txreq);
951 /* Credit-based scheduling. */
952 if (txreq.size > queue->remaining_credit &&
953 tx_credit_exceeded(queue, txreq.size))
956 queue->remaining_credit -= txreq.size;
959 queue->tx.req_cons = ++idx;
961 memset(extras, 0, sizeof(extras));
963 if (txreq.flags & XEN_NETTXF_extra_info) {
964 work_to_do = xenvif_get_extras(queue, extras,
967 idx = queue->tx.req_cons;
968 if (unlikely(work_to_do < 0))
972 if (extras[XEN_NETIF_EXTRA_TYPE_MCAST_ADD - 1].type) {
973 struct xen_netif_extra_info *extra;
975 extra = &extras[XEN_NETIF_EXTRA_TYPE_MCAST_ADD - 1];
976 ret = xenvif_mcast_add(queue->vif, extra->u.mcast.addr);
978 make_tx_response(queue, &txreq, extra_count,
981 XEN_NETIF_RSP_ERROR);
985 if (extras[XEN_NETIF_EXTRA_TYPE_MCAST_DEL - 1].type) {
986 struct xen_netif_extra_info *extra;
988 extra = &extras[XEN_NETIF_EXTRA_TYPE_MCAST_DEL - 1];
989 xenvif_mcast_del(queue->vif, extra->u.mcast.addr);
991 make_tx_response(queue, &txreq, extra_count,
996 data_len = (txreq.size > XEN_NETBACK_TX_COPY_LEN) ?
997 XEN_NETBACK_TX_COPY_LEN : txreq.size;
999 ret = xenvif_count_requests(queue, &txreq, extra_count,
1000 txfrags, work_to_do);
1002 if (unlikely(ret < 0))
1007 if (unlikely(txreq.size < ETH_HLEN)) {
1008 netdev_dbg(queue->vif->dev,
1009 "Bad packet size: %d\n", txreq.size);
1010 xenvif_tx_err(queue, &txreq, extra_count, idx);
1014 /* No crossing a page as the payload mustn't fragment. */
1015 if (unlikely((txreq.offset + txreq.size) > XEN_PAGE_SIZE)) {
1016 netdev_err(queue->vif->dev, "Cross page boundary, txreq.offset: %u, size: %u\n",
1017 txreq.offset, txreq.size);
1018 xenvif_fatal_tx_err(queue->vif);
1022 if (ret >= XEN_NETBK_LEGACY_SLOTS_MAX - 1 && data_len < txreq.size)
1023 data_len = txreq.size;
1025 skb = xenvif_alloc_skb(data_len);
1026 if (unlikely(skb == NULL)) {
1027 netdev_dbg(queue->vif->dev,
1028 "Can't allocate a skb in start_xmit.\n");
1029 xenvif_tx_err(queue, &txreq, extra_count, idx);
1033 skb_shinfo(skb)->nr_frags = ret;
1034 /* At this point shinfo->nr_frags is in fact the number of
1035 * slots, which can be as large as XEN_NETBK_LEGACY_SLOTS_MAX.
1039 if (skb_shinfo(skb)->nr_frags > MAX_SKB_FRAGS) {
1040 frag_overflow = skb_shinfo(skb)->nr_frags - MAX_SKB_FRAGS;
1041 BUG_ON(frag_overflow > MAX_SKB_FRAGS);
1042 skb_shinfo(skb)->nr_frags = MAX_SKB_FRAGS;
1043 nskb = xenvif_alloc_skb(0);
1044 if (unlikely(nskb == NULL)) {
1045 skb_shinfo(skb)->nr_frags = 0;
1047 xenvif_tx_err(queue, &txreq, extra_count, idx);
1048 if (net_ratelimit())
1049 netdev_err(queue->vif->dev,
1050 "Can't allocate the frag_list skb.\n");
1055 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1056 struct xen_netif_extra_info *gso;
1057 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1059 if (xenvif_set_skb_gso(queue->vif, skb, gso)) {
1060 /* Failure in xenvif_set_skb_gso is fatal. */
1061 skb_shinfo(skb)->nr_frags = 0;
1068 if (extras[XEN_NETIF_EXTRA_TYPE_HASH - 1].type) {
1069 struct xen_netif_extra_info *extra;
1070 enum pkt_hash_types type = PKT_HASH_TYPE_NONE;
1072 extra = &extras[XEN_NETIF_EXTRA_TYPE_HASH - 1];
1074 switch (extra->u.hash.type) {
1075 case _XEN_NETIF_CTRL_HASH_TYPE_IPV4:
1076 case _XEN_NETIF_CTRL_HASH_TYPE_IPV6:
1077 type = PKT_HASH_TYPE_L3;
1080 case _XEN_NETIF_CTRL_HASH_TYPE_IPV4_TCP:
1081 case _XEN_NETIF_CTRL_HASH_TYPE_IPV6_TCP:
1082 type = PKT_HASH_TYPE_L4;
1089 if (type != PKT_HASH_TYPE_NONE)
1091 *(u32 *)extra->u.hash.value,
1095 xenvif_get_requests(queue, skb, &txreq, txfrags, copy_ops,
1096 map_ops, frag_overflow, nskb, extra_count,
1099 __skb_queue_tail(&queue->tx_queue, skb);
1101 queue->tx.req_cons = idx;
1107 /* Consolidate skb with a frag_list into a brand new one with local pages on
1108 * frags. Returns 0 or -ENOMEM if can't allocate new pages.
1110 static int xenvif_handle_frag_list(struct xenvif_queue *queue, struct sk_buff *skb)
1112 unsigned int offset = skb_headlen(skb);
1113 skb_frag_t frags[MAX_SKB_FRAGS];
1115 struct ubuf_info *uarg;
1116 struct sk_buff *nskb = skb_shinfo(skb)->frag_list;
1118 queue->stats.tx_zerocopy_sent += 2;
1119 queue->stats.tx_frag_overflow++;
1121 xenvif_fill_frags(queue, nskb);
1122 /* Subtract frags size, we will correct it later */
1123 skb->truesize -= skb->data_len;
1124 skb->len += nskb->len;
1125 skb->data_len += nskb->len;
1127 /* create a brand new frags array and coalesce there */
1128 for (i = 0; offset < skb->len; i++) {
1132 BUG_ON(i >= MAX_SKB_FRAGS);
1133 page = alloc_page(GFP_ATOMIC);
1136 skb->truesize += skb->data_len;
1137 for (j = 0; j < i; j++)
1138 put_page(skb_frag_page(&frags[j]));
1142 if (offset + PAGE_SIZE < skb->len)
1145 len = skb->len - offset;
1146 if (skb_copy_bits(skb, offset, page_address(page), len))
1150 skb_frag_fill_page_desc(&frags[i], page, 0, len);
1153 /* Release all the original (foreign) frags. */
1154 for (f = 0; f < skb_shinfo(skb)->nr_frags; f++)
1155 skb_frag_unref(skb, f);
1156 uarg = skb_shinfo(skb)->destructor_arg;
1157 /* increase inflight counter to offset decrement in callback */
1158 atomic_inc(&queue->inflight_packets);
1159 uarg->callback(NULL, uarg, true);
1160 skb_shinfo(skb)->destructor_arg = NULL;
1162 /* Fill the skb with the new (local) frags. */
1163 memcpy(skb_shinfo(skb)->frags, frags, i * sizeof(skb_frag_t));
1164 skb_shinfo(skb)->nr_frags = i;
1165 skb->truesize += i * PAGE_SIZE;
1170 static int xenvif_tx_submit(struct xenvif_queue *queue)
1172 struct gnttab_map_grant_ref *gop_map = queue->tx_map_ops;
1173 struct gnttab_copy *gop_copy = queue->tx_copy_ops;
1174 struct sk_buff *skb;
1177 while ((skb = __skb_dequeue(&queue->tx_queue)) != NULL) {
1178 struct xen_netif_tx_request *txp;
1181 pending_idx = copy_pending_idx(skb, 0);
1182 txp = &queue->pending_tx_info[pending_idx].req;
1184 /* Check the remap error code. */
1185 if (unlikely(xenvif_tx_check_gop(queue, skb, &gop_map, &gop_copy))) {
1186 /* If there was an error, xenvif_tx_check_gop is
1187 * expected to release all the frags which were mapped,
1188 * so kfree_skb shouldn't do it again
1190 skb_shinfo(skb)->nr_frags = 0;
1191 if (skb_has_frag_list(skb)) {
1192 struct sk_buff *nskb =
1193 skb_shinfo(skb)->frag_list;
1194 skb_shinfo(nskb)->nr_frags = 0;
1200 if (txp->flags & XEN_NETTXF_csum_blank)
1201 skb->ip_summed = CHECKSUM_PARTIAL;
1202 else if (txp->flags & XEN_NETTXF_data_validated)
1203 skb->ip_summed = CHECKSUM_UNNECESSARY;
1205 xenvif_fill_frags(queue, skb);
1207 if (unlikely(skb_has_frag_list(skb))) {
1208 struct sk_buff *nskb = skb_shinfo(skb)->frag_list;
1209 xenvif_skb_zerocopy_prepare(queue, nskb);
1210 if (xenvif_handle_frag_list(queue, skb)) {
1211 if (net_ratelimit())
1212 netdev_err(queue->vif->dev,
1213 "Not enough memory to consolidate frag_list!\n");
1214 xenvif_skb_zerocopy_prepare(queue, skb);
1218 /* Copied all the bits from the frag list -- free it. */
1219 skb_frag_list_init(skb);
1223 skb->dev = queue->vif->dev;
1224 skb->protocol = eth_type_trans(skb, skb->dev);
1225 skb_reset_network_header(skb);
1227 if (checksum_setup(queue, skb)) {
1228 netdev_dbg(queue->vif->dev,
1229 "Can't setup checksum in net_tx_action\n");
1230 /* We have to set this flag to trigger the callback */
1231 if (skb_shinfo(skb)->destructor_arg)
1232 xenvif_skb_zerocopy_prepare(queue, skb);
1237 skb_probe_transport_header(skb);
1239 /* If the packet is GSO then we will have just set up the
1240 * transport header offset in checksum_setup so it's now
1241 * straightforward to calculate gso_segs.
1243 if (skb_is_gso(skb)) {
1246 /* GSO implies having the L4 header. */
1247 WARN_ON_ONCE(!skb_transport_header_was_set(skb));
1248 if (unlikely(!skb_transport_header_was_set(skb))) {
1253 mss = skb_shinfo(skb)->gso_size;
1254 hdrlen = skb_tcp_all_headers(skb);
1256 skb_shinfo(skb)->gso_segs =
1257 DIV_ROUND_UP(skb->len - hdrlen, mss);
1260 queue->stats.rx_bytes += skb->len;
1261 queue->stats.rx_packets++;
1265 /* Set this flag right before netif_receive_skb, otherwise
1266 * someone might think this packet already left netback, and
1267 * do a skb_copy_ubufs while we are still in control of the
1268 * skb. E.g. the __pskb_pull_tail earlier can do such thing.
1270 if (skb_shinfo(skb)->destructor_arg) {
1271 xenvif_skb_zerocopy_prepare(queue, skb);
1272 queue->stats.tx_zerocopy_sent++;
1275 netif_receive_skb(skb);
1281 void xenvif_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *ubuf_base,
1282 bool zerocopy_success)
1284 unsigned long flags;
1285 pending_ring_idx_t index;
1286 struct ubuf_info_msgzc *ubuf = uarg_to_msgzc(ubuf_base);
1287 struct xenvif_queue *queue = ubuf_to_queue(ubuf);
1289 /* This is the only place where we grab this lock, to protect callbacks
1292 spin_lock_irqsave(&queue->callback_lock, flags);
1294 u16 pending_idx = ubuf->desc;
1295 ubuf = (struct ubuf_info_msgzc *) ubuf->ctx;
1296 BUG_ON(queue->dealloc_prod - queue->dealloc_cons >=
1298 index = pending_index(queue->dealloc_prod);
1299 queue->dealloc_ring[index] = pending_idx;
1300 /* Sync with xenvif_tx_dealloc_action:
1301 * insert idx then incr producer.
1304 queue->dealloc_prod++;
1306 spin_unlock_irqrestore(&queue->callback_lock, flags);
1308 if (likely(zerocopy_success))
1309 queue->stats.tx_zerocopy_success++;
1311 queue->stats.tx_zerocopy_fail++;
1312 xenvif_skb_zerocopy_complete(queue);
1315 static inline void xenvif_tx_dealloc_action(struct xenvif_queue *queue)
1317 struct gnttab_unmap_grant_ref *gop;
1318 pending_ring_idx_t dc, dp;
1319 u16 pending_idx, pending_idx_release[MAX_PENDING_REQS];
1322 dc = queue->dealloc_cons;
1323 gop = queue->tx_unmap_ops;
1325 /* Free up any grants we have finished using */
1327 dp = queue->dealloc_prod;
1329 /* Ensure we see all indices enqueued by all
1330 * xenvif_zerocopy_callback().
1335 BUG_ON(gop - queue->tx_unmap_ops >= MAX_PENDING_REQS);
1337 queue->dealloc_ring[pending_index(dc++)];
1339 pending_idx_release[gop - queue->tx_unmap_ops] =
1341 queue->pages_to_unmap[gop - queue->tx_unmap_ops] =
1342 queue->mmap_pages[pending_idx];
1343 gnttab_set_unmap_op(gop,
1344 idx_to_kaddr(queue, pending_idx),
1346 queue->grant_tx_handle[pending_idx]);
1347 xenvif_grant_handle_reset(queue, pending_idx);
1351 } while (dp != queue->dealloc_prod);
1353 queue->dealloc_cons = dc;
1355 if (gop - queue->tx_unmap_ops > 0) {
1357 ret = gnttab_unmap_refs(queue->tx_unmap_ops,
1359 queue->pages_to_unmap,
1360 gop - queue->tx_unmap_ops);
1362 netdev_err(queue->vif->dev, "Unmap fail: nr_ops %tu ret %d\n",
1363 gop - queue->tx_unmap_ops, ret);
1364 for (i = 0; i < gop - queue->tx_unmap_ops; ++i) {
1365 if (gop[i].status != GNTST_okay)
1366 netdev_err(queue->vif->dev,
1367 " host_addr: 0x%llx handle: 0x%x status: %d\n",
1376 for (i = 0; i < gop - queue->tx_unmap_ops; ++i)
1377 xenvif_idx_release(queue, pending_idx_release[i],
1378 XEN_NETIF_RSP_OKAY);
1382 /* Called after netfront has transmitted */
1383 int xenvif_tx_action(struct xenvif_queue *queue, int budget)
1385 unsigned nr_mops = 0, nr_cops = 0;
1388 if (unlikely(!tx_work_todo(queue)))
1391 xenvif_tx_build_gops(queue, budget, &nr_cops, &nr_mops);
1396 gnttab_batch_copy(queue->tx_copy_ops, nr_cops);
1398 ret = gnttab_map_refs(queue->tx_map_ops,
1400 queue->pages_to_map,
1405 netdev_err(queue->vif->dev, "Map fail: nr %u ret %d\n",
1407 for (i = 0; i < nr_mops; ++i)
1408 WARN_ON_ONCE(queue->tx_map_ops[i].status ==
1413 work_done = xenvif_tx_submit(queue);
1418 static void _make_tx_response(struct xenvif_queue *queue,
1419 const struct xen_netif_tx_request *txp,
1420 unsigned int extra_count,
1423 RING_IDX i = queue->tx.rsp_prod_pvt;
1424 struct xen_netif_tx_response *resp;
1426 resp = RING_GET_RESPONSE(&queue->tx, i);
1428 resp->status = status;
1430 while (extra_count-- != 0)
1431 RING_GET_RESPONSE(&queue->tx, ++i)->status = XEN_NETIF_RSP_NULL;
1433 queue->tx.rsp_prod_pvt = ++i;
1436 static void push_tx_responses(struct xenvif_queue *queue)
1440 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&queue->tx, notify);
1442 notify_remote_via_irq(queue->tx_irq);
1445 static void xenvif_idx_release(struct xenvif_queue *queue, u16 pending_idx,
1448 struct pending_tx_info *pending_tx_info;
1449 pending_ring_idx_t index;
1450 unsigned long flags;
1452 pending_tx_info = &queue->pending_tx_info[pending_idx];
1454 spin_lock_irqsave(&queue->response_lock, flags);
1456 _make_tx_response(queue, &pending_tx_info->req,
1457 pending_tx_info->extra_count, status);
1459 /* Release the pending index before pusing the Tx response so
1460 * its available before a new Tx request is pushed by the
1463 index = pending_index(queue->pending_prod++);
1464 queue->pending_ring[index] = pending_idx;
1466 push_tx_responses(queue);
1468 spin_unlock_irqrestore(&queue->response_lock, flags);
1471 static void make_tx_response(struct xenvif_queue *queue,
1472 const struct xen_netif_tx_request *txp,
1473 unsigned int extra_count,
1476 unsigned long flags;
1478 spin_lock_irqsave(&queue->response_lock, flags);
1480 _make_tx_response(queue, txp, extra_count, status);
1481 push_tx_responses(queue);
1483 spin_unlock_irqrestore(&queue->response_lock, flags);
1486 static void xenvif_idx_unmap(struct xenvif_queue *queue, u16 pending_idx)
1489 struct gnttab_unmap_grant_ref tx_unmap_op;
1491 gnttab_set_unmap_op(&tx_unmap_op,
1492 idx_to_kaddr(queue, pending_idx),
1494 queue->grant_tx_handle[pending_idx]);
1495 xenvif_grant_handle_reset(queue, pending_idx);
1497 ret = gnttab_unmap_refs(&tx_unmap_op, NULL,
1498 &queue->mmap_pages[pending_idx], 1);
1500 netdev_err(queue->vif->dev,
1501 "Unmap fail: ret: %d pending_idx: %d host_addr: %llx handle: 0x%x status: %d\n",
1504 tx_unmap_op.host_addr,
1506 tx_unmap_op.status);
1511 static inline int tx_work_todo(struct xenvif_queue *queue)
1513 if (likely(RING_HAS_UNCONSUMED_REQUESTS(&queue->tx)))
1519 static inline bool tx_dealloc_work_todo(struct xenvif_queue *queue)
1521 return queue->dealloc_cons != queue->dealloc_prod;
1524 void xenvif_unmap_frontend_data_rings(struct xenvif_queue *queue)
1526 if (queue->tx.sring)
1527 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(queue->vif),
1529 if (queue->rx.sring)
1530 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(queue->vif),
1534 int xenvif_map_frontend_data_rings(struct xenvif_queue *queue,
1535 grant_ref_t tx_ring_ref,
1536 grant_ref_t rx_ring_ref)
1539 struct xen_netif_tx_sring *txs;
1540 struct xen_netif_rx_sring *rxs;
1541 RING_IDX rsp_prod, req_prod;
1544 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(queue->vif),
1545 &tx_ring_ref, 1, &addr);
1549 txs = (struct xen_netif_tx_sring *)addr;
1550 rsp_prod = READ_ONCE(txs->rsp_prod);
1551 req_prod = READ_ONCE(txs->req_prod);
1553 BACK_RING_ATTACH(&queue->tx, txs, rsp_prod, XEN_PAGE_SIZE);
1556 if (req_prod - rsp_prod > RING_SIZE(&queue->tx))
1559 err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(queue->vif),
1560 &rx_ring_ref, 1, &addr);
1564 rxs = (struct xen_netif_rx_sring *)addr;
1565 rsp_prod = READ_ONCE(rxs->rsp_prod);
1566 req_prod = READ_ONCE(rxs->req_prod);
1568 BACK_RING_ATTACH(&queue->rx, rxs, rsp_prod, XEN_PAGE_SIZE);
1571 if (req_prod - rsp_prod > RING_SIZE(&queue->rx))
1577 xenvif_unmap_frontend_data_rings(queue);
1581 static bool xenvif_dealloc_kthread_should_stop(struct xenvif_queue *queue)
1583 /* Dealloc thread must remain running until all inflight
1586 return kthread_should_stop() &&
1587 !atomic_read(&queue->inflight_packets);
1590 int xenvif_dealloc_kthread(void *data)
1592 struct xenvif_queue *queue = data;
1595 wait_event_interruptible(queue->dealloc_wq,
1596 tx_dealloc_work_todo(queue) ||
1597 xenvif_dealloc_kthread_should_stop(queue));
1598 if (xenvif_dealloc_kthread_should_stop(queue))
1601 xenvif_tx_dealloc_action(queue);
1605 /* Unmap anything remaining*/
1606 if (tx_dealloc_work_todo(queue))
1607 xenvif_tx_dealloc_action(queue);
1612 static void make_ctrl_response(struct xenvif *vif,
1613 const struct xen_netif_ctrl_request *req,
1614 u32 status, u32 data)
1616 RING_IDX idx = vif->ctrl.rsp_prod_pvt;
1617 struct xen_netif_ctrl_response rsp = {
1624 *RING_GET_RESPONSE(&vif->ctrl, idx) = rsp;
1625 vif->ctrl.rsp_prod_pvt = ++idx;
1628 static void push_ctrl_response(struct xenvif *vif)
1632 RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&vif->ctrl, notify);
1634 notify_remote_via_irq(vif->ctrl_irq);
1637 static void process_ctrl_request(struct xenvif *vif,
1638 const struct xen_netif_ctrl_request *req)
1640 u32 status = XEN_NETIF_CTRL_STATUS_NOT_SUPPORTED;
1643 switch (req->type) {
1644 case XEN_NETIF_CTRL_TYPE_SET_HASH_ALGORITHM:
1645 status = xenvif_set_hash_alg(vif, req->data[0]);
1648 case XEN_NETIF_CTRL_TYPE_GET_HASH_FLAGS:
1649 status = xenvif_get_hash_flags(vif, &data);
1652 case XEN_NETIF_CTRL_TYPE_SET_HASH_FLAGS:
1653 status = xenvif_set_hash_flags(vif, req->data[0]);
1656 case XEN_NETIF_CTRL_TYPE_SET_HASH_KEY:
1657 status = xenvif_set_hash_key(vif, req->data[0],
1661 case XEN_NETIF_CTRL_TYPE_GET_HASH_MAPPING_SIZE:
1662 status = XEN_NETIF_CTRL_STATUS_SUCCESS;
1663 data = XEN_NETBK_MAX_HASH_MAPPING_SIZE;
1666 case XEN_NETIF_CTRL_TYPE_SET_HASH_MAPPING_SIZE:
1667 status = xenvif_set_hash_mapping_size(vif,
1671 case XEN_NETIF_CTRL_TYPE_SET_HASH_MAPPING:
1672 status = xenvif_set_hash_mapping(vif, req->data[0],
1681 make_ctrl_response(vif, req, status, data);
1682 push_ctrl_response(vif);
1685 static void xenvif_ctrl_action(struct xenvif *vif)
1688 RING_IDX req_prod, req_cons;
1690 req_prod = vif->ctrl.sring->req_prod;
1691 req_cons = vif->ctrl.req_cons;
1693 /* Make sure we can see requests before we process them. */
1696 if (req_cons == req_prod)
1699 while (req_cons != req_prod) {
1700 struct xen_netif_ctrl_request req;
1702 RING_COPY_REQUEST(&vif->ctrl, req_cons, &req);
1705 process_ctrl_request(vif, &req);
1708 vif->ctrl.req_cons = req_cons;
1709 vif->ctrl.sring->req_event = req_cons + 1;
1713 static bool xenvif_ctrl_work_todo(struct xenvif *vif)
1715 if (likely(RING_HAS_UNCONSUMED_REQUESTS(&vif->ctrl)))
1721 irqreturn_t xenvif_ctrl_irq_fn(int irq, void *data)
1723 struct xenvif *vif = data;
1724 unsigned int eoi_flag = XEN_EOI_FLAG_SPURIOUS;
1726 while (xenvif_ctrl_work_todo(vif)) {
1727 xenvif_ctrl_action(vif);
1731 xen_irq_lateeoi(irq, eoi_flag);
1736 static int __init netback_init(void)
1743 /* Allow as many queues as there are CPUs but max. 8 if user has not
1744 * specified a value.
1746 if (xenvif_max_queues == 0)
1747 xenvif_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT,
1750 if (fatal_skb_slots < XEN_NETBK_LEGACY_SLOTS_MAX) {
1751 pr_info("fatal_skb_slots too small (%d), bump it to XEN_NETBK_LEGACY_SLOTS_MAX (%d)\n",
1752 fatal_skb_slots, XEN_NETBK_LEGACY_SLOTS_MAX);
1753 fatal_skb_slots = XEN_NETBK_LEGACY_SLOTS_MAX;
1756 rc = xenvif_xenbus_init();
1760 #ifdef CONFIG_DEBUG_FS
1761 xen_netback_dbg_root = debugfs_create_dir("xen-netback", NULL);
1762 #endif /* CONFIG_DEBUG_FS */
1770 module_init(netback_init);
1772 static void __exit netback_fini(void)
1774 #ifdef CONFIG_DEBUG_FS
1775 debugfs_remove_recursive(xen_netback_dbg_root);
1776 #endif /* CONFIG_DEBUG_FS */
1777 xenvif_xenbus_fini();
1779 module_exit(netback_fini);
1781 MODULE_DESCRIPTION("Xen backend network device module");
1782 MODULE_LICENSE("Dual BSD/GPL");
1783 MODULE_ALIAS("xen-backend:vif");