xen-netback: separate guest side rx code into separate module
authorPaul Durrant <Paul.Durrant@citrix.com>
Tue, 4 Oct 2016 09:29:12 +0000 (10:29 +0100)
committerDavid S. Miller <davem@davemloft.net>
Fri, 7 Oct 2016 00:37:35 +0000 (20:37 -0400)
The netback source module has become very large and somewhat confusing.
This patch simply moves all code related to the backend to frontend (i.e
guest side rx) data-path into a separate rx source module.

This patch contains no functional change, it is code movement and
minimal changes to avoid patch style-check issues.

Signed-off-by: Paul Durrant <paul.durrant@citrix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/xen-netback/Makefile
drivers/net/xen-netback/netback.c
drivers/net/xen-netback/rx.c [new file with mode: 0644]

index 11e02be9db1a0b1ed5e6116bb2d1c3dd3b68fda4..d49798a46b51b848ff1b4571dc390feb2c0c92fc 100644 (file)
@@ -1,3 +1,3 @@
 obj-$(CONFIG_XEN_NETDEV_BACKEND) := xen-netback.o
 
-xen-netback-y := netback.o xenbus.o interface.o hash.o
+xen-netback-y := netback.o xenbus.o interface.o hash.o rx.o
index 3d0c989384b5bdbf8814622b0d61f6ee906069cb..47b481095d773fc9aaf1bf0045bdcee5bc89e3cc 100644 (file)
@@ -106,13 +106,6 @@ static void push_tx_responses(struct xenvif_queue *queue);
 
 static inline int tx_work_todo(struct xenvif_queue *queue);
 
-static struct xen_netif_rx_response *make_rx_response(struct xenvif_queue *queue,
-                                            u16      id,
-                                            s8       st,
-                                            u16      offset,
-                                            u16      size,
-                                            u16      flags);
-
 static inline unsigned long idx_to_pfn(struct xenvif_queue *queue,
                                       u16 idx)
 {
@@ -155,571 +148,11 @@ static inline pending_ring_idx_t pending_index(unsigned i)
        return i & (MAX_PENDING_REQS-1);
 }
 
-static bool xenvif_rx_ring_slots_available(struct xenvif_queue *queue)
-{
-       RING_IDX prod, cons;
-       struct sk_buff *skb;
-       int needed;
-
-       skb = skb_peek(&queue->rx_queue);
-       if (!skb)
-               return false;
-
-       needed = DIV_ROUND_UP(skb->len, XEN_PAGE_SIZE);
-       if (skb_is_gso(skb))
-               needed++;
-       if (skb->sw_hash)
-               needed++;
-
-       do {
-               prod = queue->rx.sring->req_prod;
-               cons = queue->rx.req_cons;
-
-               if (prod - cons >= needed)
-                       return true;
-
-               queue->rx.sring->req_event = prod + 1;
-
-               /* Make sure event is visible before we check prod
-                * again.
-                */
-               mb();
-       } while (queue->rx.sring->req_prod != prod);
-
-       return false;
-}
-
-void xenvif_rx_queue_tail(struct xenvif_queue *queue, struct sk_buff *skb)
-{
-       unsigned long flags;
-
-       spin_lock_irqsave(&queue->rx_queue.lock, flags);
-
-       __skb_queue_tail(&queue->rx_queue, skb);
-
-       queue->rx_queue_len += skb->len;
-       if (queue->rx_queue_len > queue->rx_queue_max)
-               netif_tx_stop_queue(netdev_get_tx_queue(queue->vif->dev, queue->id));
-
-       spin_unlock_irqrestore(&queue->rx_queue.lock, flags);
-}
-
-static struct sk_buff *xenvif_rx_dequeue(struct xenvif_queue *queue)
-{
-       struct sk_buff *skb;
-
-       spin_lock_irq(&queue->rx_queue.lock);
-
-       skb = __skb_dequeue(&queue->rx_queue);
-       if (skb)
-               queue->rx_queue_len -= skb->len;
-
-       spin_unlock_irq(&queue->rx_queue.lock);
-
-       return skb;
-}
-
-static void xenvif_rx_queue_maybe_wake(struct xenvif_queue *queue)
-{
-       spin_lock_irq(&queue->rx_queue.lock);
-
-       if (queue->rx_queue_len < queue->rx_queue_max)
-               netif_tx_wake_queue(netdev_get_tx_queue(queue->vif->dev, queue->id));
-
-       spin_unlock_irq(&queue->rx_queue.lock);
-}
-
-
-static void xenvif_rx_queue_purge(struct xenvif_queue *queue)
-{
-       struct sk_buff *skb;
-       while ((skb = xenvif_rx_dequeue(queue)) != NULL)
-               kfree_skb(skb);
-}
-
-static void xenvif_rx_queue_drop_expired(struct xenvif_queue *queue)
-{
-       struct sk_buff *skb;
-
-       for(;;) {
-               skb = skb_peek(&queue->rx_queue);
-               if (!skb)
-                       break;
-               if (time_before(jiffies, XENVIF_RX_CB(skb)->expires))
-                       break;
-               xenvif_rx_dequeue(queue);
-               kfree_skb(skb);
-       }
-}
-
-struct netrx_pending_operations {
-       unsigned copy_prod, copy_cons;
-       unsigned meta_prod, meta_cons;
-       struct gnttab_copy *copy;
-       struct xenvif_rx_meta *meta;
-       int copy_off;
-       grant_ref_t copy_gref;
-};
-
-static struct xenvif_rx_meta *get_next_rx_buffer(struct xenvif_queue *queue,
-                                                struct netrx_pending_operations *npo)
-{
-       struct xenvif_rx_meta *meta;
-       struct xen_netif_rx_request req;
-
-       RING_COPY_REQUEST(&queue->rx, queue->rx.req_cons++, &req);
-
-       meta = npo->meta + npo->meta_prod++;
-       meta->gso_type = XEN_NETIF_GSO_TYPE_NONE;
-       meta->gso_size = 0;
-       meta->size = 0;
-       meta->id = req.id;
-
-       npo->copy_off = 0;
-       npo->copy_gref = req.gref;
-
-       return meta;
-}
-
-struct gop_frag_copy {
-       struct xenvif_queue *queue;
-       struct netrx_pending_operations *npo;
-       struct xenvif_rx_meta *meta;
-       int head;
-       int gso_type;
-       int protocol;
-       int hash_present;
-
-       struct page *page;
-};
-
-static void xenvif_setup_copy_gop(unsigned long gfn,
-                                 unsigned int offset,
-                                 unsigned int *len,
-                                 struct gop_frag_copy *info)
-{
-       struct gnttab_copy *copy_gop;
-       struct xen_page_foreign *foreign;
-       /* Convenient aliases */
-       struct xenvif_queue *queue = info->queue;
-       struct netrx_pending_operations *npo = info->npo;
-       struct page *page = info->page;
-
-       BUG_ON(npo->copy_off > MAX_BUFFER_OFFSET);
-
-       if (npo->copy_off == MAX_BUFFER_OFFSET)
-               info->meta = get_next_rx_buffer(queue, npo);
-
-       if (npo->copy_off + *len > MAX_BUFFER_OFFSET)
-               *len = MAX_BUFFER_OFFSET - npo->copy_off;
-
-       copy_gop = npo->copy + npo->copy_prod++;
-       copy_gop->flags = GNTCOPY_dest_gref;
-       copy_gop->len = *len;
-
-       foreign = xen_page_foreign(page);
-       if (foreign) {
-               copy_gop->source.domid = foreign->domid;
-               copy_gop->source.u.ref = foreign->gref;
-               copy_gop->flags |= GNTCOPY_source_gref;
-       } else {
-               copy_gop->source.domid = DOMID_SELF;
-               copy_gop->source.u.gmfn = gfn;
-       }
-       copy_gop->source.offset = offset;
-
-       copy_gop->dest.domid = queue->vif->domid;
-       copy_gop->dest.offset = npo->copy_off;
-       copy_gop->dest.u.ref = npo->copy_gref;
-
-       npo->copy_off += *len;
-       info->meta->size += *len;
-
-       if (!info->head)
-               return;
-
-       /* Leave a gap for the GSO descriptor. */
-       if ((1 << info->gso_type) & queue->vif->gso_mask)
-               queue->rx.req_cons++;
-
-       /* Leave a gap for the hash extra segment. */
-       if (info->hash_present)
-               queue->rx.req_cons++;
-
-       info->head = 0; /* There must be something in this buffer now */
-}
-
-static void xenvif_gop_frag_copy_grant(unsigned long gfn,
-                                      unsigned offset,
-                                      unsigned int len,
-                                      void *data)
-{
-       unsigned int bytes;
-
-       while (len) {
-               bytes = len;
-               xenvif_setup_copy_gop(gfn, offset, &bytes, data);
-               offset += bytes;
-               len -= bytes;
-       }
-}
-
-/*
- * Set up the grant operations for this fragment. If it's a flipping
- * interface, we also set up the unmap request from here.
- */
-static void xenvif_gop_frag_copy(struct xenvif_queue *queue, struct sk_buff *skb,
-                                struct netrx_pending_operations *npo,
-                                struct page *page, unsigned long size,
-                                unsigned long offset, int *head)
-{
-       struct gop_frag_copy info = {
-               .queue = queue,
-               .npo = npo,
-               .head = *head,
-               .gso_type = XEN_NETIF_GSO_TYPE_NONE,
-               /* xenvif_set_skb_hash() will have either set a s/w
-                * hash or cleared the hash depending on
-                * whether the the frontend wants a hash for this skb.
-                */
-               .hash_present = skb->sw_hash,
-       };
-       unsigned long bytes;
-
-       if (skb_is_gso(skb)) {
-               if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
-                       info.gso_type = XEN_NETIF_GSO_TYPE_TCPV4;
-               else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
-                       info.gso_type = XEN_NETIF_GSO_TYPE_TCPV6;
-       }
-
-       /* Data must not cross a page boundary. */
-       BUG_ON(size + offset > PAGE_SIZE<<compound_order(page));
-
-       info.meta = npo->meta + npo->meta_prod - 1;
-
-       /* Skip unused frames from start of page */
-       page += offset >> PAGE_SHIFT;
-       offset &= ~PAGE_MASK;
-
-       while (size > 0) {
-               BUG_ON(offset >= PAGE_SIZE);
-
-               bytes = PAGE_SIZE - offset;
-               if (bytes > size)
-                       bytes = size;
-
-               info.page = page;
-               gnttab_foreach_grant_in_range(page, offset, bytes,
-                                             xenvif_gop_frag_copy_grant,
-                                             &info);
-               size -= bytes;
-               offset = 0;
-
-               /* Next page */
-               if (size) {
-                       BUG_ON(!PageCompound(page));
-                       page++;
-               }
-       }
-
-       *head = info.head;
-}
-
-/*
- * Prepare an SKB to be transmitted to the frontend.
- *
- * This function is responsible for allocating grant operations, meta
- * structures, etc.
- *
- * It returns the number of meta structures consumed. The number of
- * ring slots used is always equal to the number of meta slots used
- * plus the number of GSO descriptors used. Currently, we use either
- * zero GSO descriptors (for non-GSO packets) or one descriptor (for
- * frontend-side LRO).
- */
-static int xenvif_gop_skb(struct sk_buff *skb,
-                         struct netrx_pending_operations *npo,
-                         struct xenvif_queue *queue)
-{
-       struct xenvif *vif = netdev_priv(skb->dev);
-       int nr_frags = skb_shinfo(skb)->nr_frags;
-       int i;
-       struct xen_netif_rx_request req;
-       struct xenvif_rx_meta *meta;
-       unsigned char *data;
-       int head = 1;
-       int old_meta_prod;
-       int gso_type;
-
-       old_meta_prod = npo->meta_prod;
-
-       gso_type = XEN_NETIF_GSO_TYPE_NONE;
-       if (skb_is_gso(skb)) {
-               if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
-                       gso_type = XEN_NETIF_GSO_TYPE_TCPV4;
-               else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
-                       gso_type = XEN_NETIF_GSO_TYPE_TCPV6;
-       }
-
-       /* Set up a GSO prefix descriptor, if necessary */
-       if ((1 << gso_type) & vif->gso_prefix_mask) {
-               RING_COPY_REQUEST(&queue->rx, queue->rx.req_cons++, &req);
-               meta = npo->meta + npo->meta_prod++;
-               meta->gso_type = gso_type;
-               meta->gso_size = skb_shinfo(skb)->gso_size;
-               meta->size = 0;
-               meta->id = req.id;
-       }
-
-       RING_COPY_REQUEST(&queue->rx, queue->rx.req_cons++, &req);
-       meta = npo->meta + npo->meta_prod++;
-
-       if ((1 << gso_type) & vif->gso_mask) {
-               meta->gso_type = gso_type;
-               meta->gso_size = skb_shinfo(skb)->gso_size;
-       } else {
-               meta->gso_type = XEN_NETIF_GSO_TYPE_NONE;
-               meta->gso_size = 0;
-       }
-
-       meta->size = 0;
-       meta->id = req.id;
-       npo->copy_off = 0;
-       npo->copy_gref = req.gref;
-
-       data = skb->data;
-       while (data < skb_tail_pointer(skb)) {
-               unsigned int offset = offset_in_page(data);
-               unsigned int len = PAGE_SIZE - offset;
-
-               if (data + len > skb_tail_pointer(skb))
-                       len = skb_tail_pointer(skb) - data;
-
-               xenvif_gop_frag_copy(queue, skb, npo,
-                                    virt_to_page(data), len, offset, &head);
-               data += len;
-       }
-
-       for (i = 0; i < nr_frags; i++) {
-               xenvif_gop_frag_copy(queue, skb, npo,
-                                    skb_frag_page(&skb_shinfo(skb)->frags[i]),
-                                    skb_frag_size(&skb_shinfo(skb)->frags[i]),
-                                    skb_shinfo(skb)->frags[i].page_offset,
-                                    &head);
-       }
-
-       return npo->meta_prod - old_meta_prod;
-}
-
-/*
- * This is a twin to xenvif_gop_skb.  Assume that xenvif_gop_skb was
- * used to set up the operations on the top of
- * netrx_pending_operations, which have since been done.  Check that
- * they didn't give any errors and advance over them.
- */
-static int xenvif_check_gop(struct xenvif *vif, int nr_meta_slots,
-                           struct netrx_pending_operations *npo)
-{
-       struct gnttab_copy     *copy_op;
-       int status = XEN_NETIF_RSP_OKAY;
-       int i;
-
-       for (i = 0; i < nr_meta_slots; i++) {
-               copy_op = npo->copy + npo->copy_cons++;
-               if (copy_op->status != GNTST_okay) {
-                       netdev_dbg(vif->dev,
-                                  "Bad status %d from copy to DOM%d.\n",
-                                  copy_op->status, vif->domid);
-                       status = XEN_NETIF_RSP_ERROR;
-               }
-       }
-
-       return status;
-}
-
-static void xenvif_add_frag_responses(struct xenvif_queue *queue, int status,
-                                     struct xenvif_rx_meta *meta,
-                                     int nr_meta_slots)
-{
-       int i;
-       unsigned long offset;
-
-       /* No fragments used */
-       if (nr_meta_slots <= 1)
-               return;
-
-       nr_meta_slots--;
-
-       for (i = 0; i < nr_meta_slots; i++) {
-               int flags;
-               if (i == nr_meta_slots - 1)
-                       flags = 0;
-               else
-                       flags = XEN_NETRXF_more_data;
-
-               offset = 0;
-               make_rx_response(queue, meta[i].id, status, offset,
-                                meta[i].size, flags);
-       }
-}
-
 void xenvif_kick_thread(struct xenvif_queue *queue)
 {
        wake_up(&queue->wq);
 }
 
-static void xenvif_rx_action(struct xenvif_queue *queue)
-{
-       struct xenvif *vif = queue->vif;
-       s8 status;
-       u16 flags;
-       struct xen_netif_rx_response *resp;
-       struct sk_buff_head rxq;
-       struct sk_buff *skb;
-       LIST_HEAD(notify);
-       int ret;
-       unsigned long offset;
-       bool need_to_notify = false;
-
-       struct netrx_pending_operations npo = {
-               .copy  = queue->grant_copy_op,
-               .meta  = queue->meta,
-       };
-
-       skb_queue_head_init(&rxq);
-
-       while (xenvif_rx_ring_slots_available(queue)
-              && (skb = xenvif_rx_dequeue(queue)) != NULL) {
-               queue->last_rx_time = jiffies;
-
-               XENVIF_RX_CB(skb)->meta_slots_used = xenvif_gop_skb(skb, &npo, queue);
-
-               __skb_queue_tail(&rxq, skb);
-       }
-
-       BUG_ON(npo.meta_prod > ARRAY_SIZE(queue->meta));
-
-       if (!npo.copy_prod)
-               goto done;
-
-       BUG_ON(npo.copy_prod > MAX_GRANT_COPY_OPS);
-       gnttab_batch_copy(queue->grant_copy_op, npo.copy_prod);
-
-       while ((skb = __skb_dequeue(&rxq)) != NULL) {
-               struct xen_netif_extra_info *extra = NULL;
-
-               if ((1 << queue->meta[npo.meta_cons].gso_type) &
-                   vif->gso_prefix_mask) {
-                       resp = RING_GET_RESPONSE(&queue->rx,
-                                                queue->rx.rsp_prod_pvt++);
-
-                       resp->flags = XEN_NETRXF_gso_prefix | XEN_NETRXF_more_data;
-
-                       resp->offset = queue->meta[npo.meta_cons].gso_size;
-                       resp->id = queue->meta[npo.meta_cons].id;
-                       resp->status = XENVIF_RX_CB(skb)->meta_slots_used;
-
-                       npo.meta_cons++;
-                       XENVIF_RX_CB(skb)->meta_slots_used--;
-               }
-
-
-               queue->stats.tx_bytes += skb->len;
-               queue->stats.tx_packets++;
-
-               status = xenvif_check_gop(vif,
-                                         XENVIF_RX_CB(skb)->meta_slots_used,
-                                         &npo);
-
-               if (XENVIF_RX_CB(skb)->meta_slots_used == 1)
-                       flags = 0;
-               else
-                       flags = XEN_NETRXF_more_data;
-
-               if (skb->ip_summed == CHECKSUM_PARTIAL) /* local packet? */
-                       flags |= XEN_NETRXF_csum_blank | XEN_NETRXF_data_validated;
-               else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
-                       /* remote but checksummed. */
-                       flags |= XEN_NETRXF_data_validated;
-
-               offset = 0;
-               resp = make_rx_response(queue, queue->meta[npo.meta_cons].id,
-                                       status, offset,
-                                       queue->meta[npo.meta_cons].size,
-                                       flags);
-
-               if ((1 << queue->meta[npo.meta_cons].gso_type) &
-                   vif->gso_mask) {
-                       extra = (struct xen_netif_extra_info *)
-                               RING_GET_RESPONSE(&queue->rx,
-                                                 queue->rx.rsp_prod_pvt++);
-
-                       resp->flags |= XEN_NETRXF_extra_info;
-
-                       extra->u.gso.type = queue->meta[npo.meta_cons].gso_type;
-                       extra->u.gso.size = queue->meta[npo.meta_cons].gso_size;
-                       extra->u.gso.pad = 0;
-                       extra->u.gso.features = 0;
-
-                       extra->type = XEN_NETIF_EXTRA_TYPE_GSO;
-                       extra->flags = 0;
-               }
-
-               if (skb->sw_hash) {
-                       /* Since the skb got here via xenvif_select_queue()
-                        * we know that the hash has been re-calculated
-                        * according to a configuration set by the frontend
-                        * and therefore we know that it is legitimate to
-                        * pass it to the frontend.
-                        */
-                       if (resp->flags & XEN_NETRXF_extra_info)
-                               extra->flags |= XEN_NETIF_EXTRA_FLAG_MORE;
-                       else
-                               resp->flags |= XEN_NETRXF_extra_info;
-
-                       extra = (struct xen_netif_extra_info *)
-                               RING_GET_RESPONSE(&queue->rx,
-                                                 queue->rx.rsp_prod_pvt++);
-
-                       extra->u.hash.algorithm =
-                               XEN_NETIF_CTRL_HASH_ALGORITHM_TOEPLITZ;
-
-                       if (skb->l4_hash)
-                               extra->u.hash.type =
-                                       skb->protocol == htons(ETH_P_IP) ?
-                                       _XEN_NETIF_CTRL_HASH_TYPE_IPV4_TCP :
-                                       _XEN_NETIF_CTRL_HASH_TYPE_IPV6_TCP;
-                       else
-                               extra->u.hash.type =
-                                       skb->protocol == htons(ETH_P_IP) ?
-                                       _XEN_NETIF_CTRL_HASH_TYPE_IPV4 :
-                                       _XEN_NETIF_CTRL_HASH_TYPE_IPV6;
-
-                       *(uint32_t *)extra->u.hash.value =
-                               skb_get_hash_raw(skb);
-
-                       extra->type = XEN_NETIF_EXTRA_TYPE_HASH;
-                       extra->flags = 0;
-               }
-
-               xenvif_add_frag_responses(queue, status,
-                                         queue->meta + npo.meta_cons + 1,
-                                         XENVIF_RX_CB(skb)->meta_slots_used);
-
-               RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&queue->rx, ret);
-
-               need_to_notify |= !!ret;
-
-               npo.meta_cons += XENVIF_RX_CB(skb)->meta_slots_used;
-               dev_kfree_skb(skb);
-       }
-
-done:
-       if (need_to_notify)
-               notify_remote_via_irq(queue->rx_irq);
-}
-
 void xenvif_napi_schedule_or_enable_events(struct xenvif_queue *queue)
 {
        int more_to_do;
@@ -1951,29 +1384,6 @@ static void push_tx_responses(struct xenvif_queue *queue)
                notify_remote_via_irq(queue->tx_irq);
 }
 
-static struct xen_netif_rx_response *make_rx_response(struct xenvif_queue *queue,
-                                            u16      id,
-                                            s8       st,
-                                            u16      offset,
-                                            u16      size,
-                                            u16      flags)
-{
-       RING_IDX i = queue->rx.rsp_prod_pvt;
-       struct xen_netif_rx_response *resp;
-
-       resp = RING_GET_RESPONSE(&queue->rx, i);
-       resp->offset     = offset;
-       resp->flags      = flags;
-       resp->id         = id;
-       resp->status     = (s16)size;
-       if (st < 0)
-               resp->status = (s16)st;
-
-       queue->rx.rsp_prod_pvt = ++i;
-
-       return resp;
-}
-
 void xenvif_idx_unmap(struct xenvif_queue *queue, u16 pending_idx)
 {
        int ret;
@@ -2055,170 +1465,6 @@ err:
        return err;
 }
 
-static void xenvif_queue_carrier_off(struct xenvif_queue *queue)
-{
-       struct xenvif *vif = queue->vif;
-
-       queue->stalled = true;
-
-       /* At least one queue has stalled? Disable the carrier. */
-       spin_lock(&vif->lock);
-       if (vif->stalled_queues++ == 0) {
-               netdev_info(vif->dev, "Guest Rx stalled");
-               netif_carrier_off(vif->dev);
-       }
-       spin_unlock(&vif->lock);
-}
-
-static void xenvif_queue_carrier_on(struct xenvif_queue *queue)
-{
-       struct xenvif *vif = queue->vif;
-
-       queue->last_rx_time = jiffies; /* Reset Rx stall detection. */
-       queue->stalled = false;
-
-       /* All queues are ready? Enable the carrier. */
-       spin_lock(&vif->lock);
-       if (--vif->stalled_queues == 0) {
-               netdev_info(vif->dev, "Guest Rx ready");
-               netif_carrier_on(vif->dev);
-       }
-       spin_unlock(&vif->lock);
-}
-
-static bool xenvif_rx_queue_stalled(struct xenvif_queue *queue)
-{
-       RING_IDX prod, cons;
-
-       prod = queue->rx.sring->req_prod;
-       cons = queue->rx.req_cons;
-
-       return !queue->stalled && prod - cons < 1
-               && time_after(jiffies,
-                             queue->last_rx_time + queue->vif->stall_timeout);
-}
-
-static bool xenvif_rx_queue_ready(struct xenvif_queue *queue)
-{
-       RING_IDX prod, cons;
-
-       prod = queue->rx.sring->req_prod;
-       cons = queue->rx.req_cons;
-
-       return queue->stalled && prod - cons >= 1;
-}
-
-static bool xenvif_have_rx_work(struct xenvif_queue *queue)
-{
-       return xenvif_rx_ring_slots_available(queue)
-               || (queue->vif->stall_timeout &&
-                   (xenvif_rx_queue_stalled(queue)
-                    || xenvif_rx_queue_ready(queue)))
-               || kthread_should_stop()
-               || queue->vif->disabled;
-}
-
-static long xenvif_rx_queue_timeout(struct xenvif_queue *queue)
-{
-       struct sk_buff *skb;
-       long timeout;
-
-       skb = skb_peek(&queue->rx_queue);
-       if (!skb)
-               return MAX_SCHEDULE_TIMEOUT;
-
-       timeout = XENVIF_RX_CB(skb)->expires - jiffies;
-       return timeout < 0 ? 0 : timeout;
-}
-
-/* Wait until the guest Rx thread has work.
- *
- * The timeout needs to be adjusted based on the current head of the
- * queue (and not just the head at the beginning).  In particular, if
- * the queue is initially empty an infinite timeout is used and this
- * needs to be reduced when a skb is queued.
- *
- * This cannot be done with wait_event_timeout() because it only
- * calculates the timeout once.
- */
-static void xenvif_wait_for_rx_work(struct xenvif_queue *queue)
-{
-       DEFINE_WAIT(wait);
-
-       if (xenvif_have_rx_work(queue))
-               return;
-
-       for (;;) {
-               long ret;
-
-               prepare_to_wait(&queue->wq, &wait, TASK_INTERRUPTIBLE);
-               if (xenvif_have_rx_work(queue))
-                       break;
-               ret = schedule_timeout(xenvif_rx_queue_timeout(queue));
-               if (!ret)
-                       break;
-       }
-       finish_wait(&queue->wq, &wait);
-}
-
-int xenvif_kthread_guest_rx(void *data)
-{
-       struct xenvif_queue *queue = data;
-       struct xenvif *vif = queue->vif;
-
-       if (!vif->stall_timeout)
-               xenvif_queue_carrier_on(queue);
-
-       for (;;) {
-               xenvif_wait_for_rx_work(queue);
-
-               if (kthread_should_stop())
-                       break;
-
-               /* This frontend is found to be rogue, disable it in
-                * kthread context. Currently this is only set when
-                * netback finds out frontend sends malformed packet,
-                * but we cannot disable the interface in softirq
-                * context so we defer it here, if this thread is
-                * associated with queue 0.
-                */
-               if (unlikely(vif->disabled && queue->id == 0)) {
-                       xenvif_carrier_off(vif);
-                       break;
-               }
-
-               if (!skb_queue_empty(&queue->rx_queue))
-                       xenvif_rx_action(queue);
-
-               /* If the guest hasn't provided any Rx slots for a
-                * while it's probably not responsive, drop the
-                * carrier so packets are dropped earlier.
-                */
-               if (vif->stall_timeout) {
-                       if (xenvif_rx_queue_stalled(queue))
-                               xenvif_queue_carrier_off(queue);
-                       else if (xenvif_rx_queue_ready(queue))
-                               xenvif_queue_carrier_on(queue);
-               }
-
-               /* Queued packets may have foreign pages from other
-                * domains.  These cannot be queued indefinitely as
-                * this would starve guests of grant refs and transmit
-                * slots.
-                */
-               xenvif_rx_queue_drop_expired(queue);
-
-               xenvif_rx_queue_maybe_wake(queue);
-
-               cond_resched();
-       }
-
-       /* Bin any remaining skbs */
-       xenvif_rx_queue_purge(queue);
-
-       return 0;
-}
-
 static bool xenvif_dealloc_kthread_should_stop(struct xenvif_queue *queue)
 {
        /* Dealloc thread must remain running until all inflight
diff --git a/drivers/net/xen-netback/rx.c b/drivers/net/xen-netback/rx.c
new file mode 100644 (file)
index 0000000..03836aa
--- /dev/null
@@ -0,0 +1,789 @@
+/*
+ * Copyright (c) 2016 Citrix Systems Inc.
+ * Copyright (c) 2002-2005, K A Fraser
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License version 2
+ * as published by the Free Software Foundation; or, when distributed
+ * separately from the Linux kernel or incorporated into other
+ * software packages, subject to the following license:
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this source file (the "Software"), to deal in the Software without
+ * restriction, including without limitation the rights to use, copy, modify,
+ * merge, publish, distribute, sublicense, and/or sell copies of the Software,
+ * and to permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
+ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
+ * IN THE SOFTWARE.
+ */
+
+#include "common.h"
+
+#include <linux/kthread.h>
+
+#include <xen/xen.h>
+#include <xen/events.h>
+
+static bool xenvif_rx_ring_slots_available(struct xenvif_queue *queue)
+{
+       RING_IDX prod, cons;
+       struct sk_buff *skb;
+       int needed;
+
+       skb = skb_peek(&queue->rx_queue);
+       if (!skb)
+               return false;
+
+       needed = DIV_ROUND_UP(skb->len, XEN_PAGE_SIZE);
+       if (skb_is_gso(skb))
+               needed++;
+       if (skb->sw_hash)
+               needed++;
+
+       do {
+               prod = queue->rx.sring->req_prod;
+               cons = queue->rx.req_cons;
+
+               if (prod - cons >= needed)
+                       return true;
+
+               queue->rx.sring->req_event = prod + 1;
+
+               /* Make sure event is visible before we check prod
+                * again.
+                */
+               mb();
+       } while (queue->rx.sring->req_prod != prod);
+
+       return false;
+}
+
+void xenvif_rx_queue_tail(struct xenvif_queue *queue, struct sk_buff *skb)
+{
+       unsigned long flags;
+
+       spin_lock_irqsave(&queue->rx_queue.lock, flags);
+
+       __skb_queue_tail(&queue->rx_queue, skb);
+
+       queue->rx_queue_len += skb->len;
+       if (queue->rx_queue_len > queue->rx_queue_max) {
+               struct net_device *dev = queue->vif->dev;
+
+               netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
+       }
+
+       spin_unlock_irqrestore(&queue->rx_queue.lock, flags);
+}
+
+static struct sk_buff *xenvif_rx_dequeue(struct xenvif_queue *queue)
+{
+       struct sk_buff *skb;
+
+       spin_lock_irq(&queue->rx_queue.lock);
+
+       skb = __skb_dequeue(&queue->rx_queue);
+       if (skb)
+               queue->rx_queue_len -= skb->len;
+
+       spin_unlock_irq(&queue->rx_queue.lock);
+
+       return skb;
+}
+
+static void xenvif_rx_queue_maybe_wake(struct xenvif_queue *queue)
+{
+       spin_lock_irq(&queue->rx_queue.lock);
+
+       if (queue->rx_queue_len < queue->rx_queue_max) {
+               struct net_device *dev = queue->vif->dev;
+
+               netif_tx_wake_queue(netdev_get_tx_queue(dev, queue->id));
+       }
+
+       spin_unlock_irq(&queue->rx_queue.lock);
+}
+
+static void xenvif_rx_queue_purge(struct xenvif_queue *queue)
+{
+       struct sk_buff *skb;
+
+       while ((skb = xenvif_rx_dequeue(queue)) != NULL)
+               kfree_skb(skb);
+}
+
+static void xenvif_rx_queue_drop_expired(struct xenvif_queue *queue)
+{
+       struct sk_buff *skb;
+
+       for (;;) {
+               skb = skb_peek(&queue->rx_queue);
+               if (!skb)
+                       break;
+               if (time_before(jiffies, XENVIF_RX_CB(skb)->expires))
+                       break;
+               xenvif_rx_dequeue(queue);
+               kfree_skb(skb);
+       }
+}
+
+struct netrx_pending_operations {
+       unsigned int copy_prod, copy_cons;
+       unsigned int meta_prod, meta_cons;
+       struct gnttab_copy *copy;
+       struct xenvif_rx_meta *meta;
+       int copy_off;
+       grant_ref_t copy_gref;
+};
+
+static struct xenvif_rx_meta *get_next_rx_buffer(
+       struct xenvif_queue *queue,
+       struct netrx_pending_operations *npo)
+{
+       struct xenvif_rx_meta *meta;
+       struct xen_netif_rx_request req;
+
+       RING_COPY_REQUEST(&queue->rx, queue->rx.req_cons++, &req);
+
+       meta = npo->meta + npo->meta_prod++;
+       meta->gso_type = XEN_NETIF_GSO_TYPE_NONE;
+       meta->gso_size = 0;
+       meta->size = 0;
+       meta->id = req.id;
+
+       npo->copy_off = 0;
+       npo->copy_gref = req.gref;
+
+       return meta;
+}
+
+struct gop_frag_copy {
+       struct xenvif_queue *queue;
+       struct netrx_pending_operations *npo;
+       struct xenvif_rx_meta *meta;
+       int head;
+       int gso_type;
+       int protocol;
+       int hash_present;
+
+       struct page *page;
+};
+
+static void xenvif_setup_copy_gop(unsigned long gfn,
+                                 unsigned int offset,
+                                 unsigned int *len,
+                                 struct gop_frag_copy *info)
+{
+       struct gnttab_copy *copy_gop;
+       struct xen_page_foreign *foreign;
+       /* Convenient aliases */
+       struct xenvif_queue *queue = info->queue;
+       struct netrx_pending_operations *npo = info->npo;
+       struct page *page = info->page;
+
+       WARN_ON(npo->copy_off > MAX_BUFFER_OFFSET);
+
+       if (npo->copy_off == MAX_BUFFER_OFFSET)
+               info->meta = get_next_rx_buffer(queue, npo);
+
+       if (npo->copy_off + *len > MAX_BUFFER_OFFSET)
+               *len = MAX_BUFFER_OFFSET - npo->copy_off;
+
+       copy_gop = npo->copy + npo->copy_prod++;
+       copy_gop->flags = GNTCOPY_dest_gref;
+       copy_gop->len = *len;
+
+       foreign = xen_page_foreign(page);
+       if (foreign) {
+               copy_gop->source.domid = foreign->domid;
+               copy_gop->source.u.ref = foreign->gref;
+               copy_gop->flags |= GNTCOPY_source_gref;
+       } else {
+               copy_gop->source.domid = DOMID_SELF;
+               copy_gop->source.u.gmfn = gfn;
+       }
+       copy_gop->source.offset = offset;
+
+       copy_gop->dest.domid = queue->vif->domid;
+       copy_gop->dest.offset = npo->copy_off;
+       copy_gop->dest.u.ref = npo->copy_gref;
+
+       npo->copy_off += *len;
+       info->meta->size += *len;
+
+       if (!info->head)
+               return;
+
+       /* Leave a gap for the GSO descriptor. */
+       if ((1 << info->gso_type) & queue->vif->gso_mask)
+               queue->rx.req_cons++;
+
+       /* Leave a gap for the hash extra segment. */
+       if (info->hash_present)
+               queue->rx.req_cons++;
+
+       info->head = 0; /* There must be something in this buffer now */
+}
+
+static void xenvif_gop_frag_copy_grant(unsigned long gfn,
+                                      unsigned int offset,
+                                      unsigned int len,
+                                      void *data)
+{
+       unsigned int bytes;
+
+       while (len) {
+               bytes = len;
+               xenvif_setup_copy_gop(gfn, offset, &bytes, data);
+               offset += bytes;
+               len -= bytes;
+       }
+}
+
+/* Set up the grant operations for this fragment. If it's a flipping
+ * interface, we also set up the unmap request from here.
+ */
+static void xenvif_gop_frag_copy(struct xenvif_queue *queue,
+                                struct sk_buff *skb,
+                                struct netrx_pending_operations *npo,
+                                struct page *page, unsigned long size,
+                                unsigned long offset, int *head)
+{
+       struct gop_frag_copy info = {
+               .queue = queue,
+               .npo = npo,
+               .head = *head,
+               .gso_type = XEN_NETIF_GSO_TYPE_NONE,
+               /* xenvif_set_skb_hash() will have either set a s/w
+                * hash or cleared the hash depending on
+                * whether the the frontend wants a hash for this skb.
+                */
+               .hash_present = skb->sw_hash,
+       };
+       unsigned long bytes;
+
+       if (skb_is_gso(skb)) {
+               if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
+                       info.gso_type = XEN_NETIF_GSO_TYPE_TCPV4;
+               else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
+                       info.gso_type = XEN_NETIF_GSO_TYPE_TCPV6;
+       }
+
+       /* Data must not cross a page boundary. */
+       WARN_ON(size + offset > (PAGE_SIZE << compound_order(page)));
+
+       info.meta = npo->meta + npo->meta_prod - 1;
+
+       /* Skip unused frames from start of page */
+       page += offset >> PAGE_SHIFT;
+       offset &= ~PAGE_MASK;
+
+       while (size > 0) {
+               WARN_ON(offset >= PAGE_SIZE);
+
+               bytes = PAGE_SIZE - offset;
+               if (bytes > size)
+                       bytes = size;
+
+               info.page = page;
+               gnttab_foreach_grant_in_range(page, offset, bytes,
+                                             xenvif_gop_frag_copy_grant,
+                                             &info);
+               size -= bytes;
+               offset = 0;
+
+               /* Next page */
+               if (size) {
+                       WARN_ON(!PageCompound(page));
+                       page++;
+               }
+       }
+
+       *head = info.head;
+}
+
+/* Prepare an SKB to be transmitted to the frontend.
+ *
+ * This function is responsible for allocating grant operations, meta
+ * structures, etc.
+ *
+ * It returns the number of meta structures consumed. The number of
+ * ring slots used is always equal to the number of meta slots used
+ * plus the number of GSO descriptors used. Currently, we use either
+ * zero GSO descriptors (for non-GSO packets) or one descriptor (for
+ * frontend-side LRO).
+ */
+static int xenvif_gop_skb(struct sk_buff *skb,
+                         struct netrx_pending_operations *npo,
+                         struct xenvif_queue *queue)
+{
+       struct xenvif *vif = netdev_priv(skb->dev);
+       int nr_frags = skb_shinfo(skb)->nr_frags;
+       int i;
+       struct xen_netif_rx_request req;
+       struct xenvif_rx_meta *meta;
+       unsigned char *data;
+       int head = 1;
+       int old_meta_prod;
+       int gso_type;
+
+       old_meta_prod = npo->meta_prod;
+
+       gso_type = XEN_NETIF_GSO_TYPE_NONE;
+       if (skb_is_gso(skb)) {
+               if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
+                       gso_type = XEN_NETIF_GSO_TYPE_TCPV4;
+               else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
+                       gso_type = XEN_NETIF_GSO_TYPE_TCPV6;
+       }
+
+       /* Set up a GSO prefix descriptor, if necessary */
+       if ((1 << gso_type) & vif->gso_prefix_mask) {
+               RING_COPY_REQUEST(&queue->rx, queue->rx.req_cons++, &req);
+               meta = npo->meta + npo->meta_prod++;
+               meta->gso_type = gso_type;
+               meta->gso_size = skb_shinfo(skb)->gso_size;
+               meta->size = 0;
+               meta->id = req.id;
+       }
+
+       RING_COPY_REQUEST(&queue->rx, queue->rx.req_cons++, &req);
+       meta = npo->meta + npo->meta_prod++;
+
+       if ((1 << gso_type) & vif->gso_mask) {
+               meta->gso_type = gso_type;
+               meta->gso_size = skb_shinfo(skb)->gso_size;
+       } else {
+               meta->gso_type = XEN_NETIF_GSO_TYPE_NONE;
+               meta->gso_size = 0;
+       }
+
+       meta->size = 0;
+       meta->id = req.id;
+       npo->copy_off = 0;
+       npo->copy_gref = req.gref;
+
+       data = skb->data;
+       while (data < skb_tail_pointer(skb)) {
+               unsigned int offset = offset_in_page(data);
+               unsigned int len = PAGE_SIZE - offset;
+
+               if (data + len > skb_tail_pointer(skb))
+                       len = skb_tail_pointer(skb) - data;
+
+               xenvif_gop_frag_copy(queue, skb, npo,
+                                    virt_to_page(data), len, offset, &head);
+               data += len;
+       }
+
+       for (i = 0; i < nr_frags; i++) {
+               xenvif_gop_frag_copy(queue, skb, npo,
+                                    skb_frag_page(&skb_shinfo(skb)->frags[i]),
+                                    skb_frag_size(&skb_shinfo(skb)->frags[i]),
+                                    skb_shinfo(skb)->frags[i].page_offset,
+                                    &head);
+       }
+
+       return npo->meta_prod - old_meta_prod;
+}
+
+/* This is a twin to xenvif_gop_skb.  Assume that xenvif_gop_skb was
+ * used to set up the operations on the top of
+ * netrx_pending_operations, which have since been done.  Check that
+ * they didn't give any errors and advance over them.
+ */
+static int xenvif_check_gop(struct xenvif *vif, int nr_meta_slots,
+                           struct netrx_pending_operations *npo)
+{
+       struct gnttab_copy     *copy_op;
+       int status = XEN_NETIF_RSP_OKAY;
+       int i;
+
+       for (i = 0; i < nr_meta_slots; i++) {
+               copy_op = npo->copy + npo->copy_cons++;
+               if (copy_op->status != GNTST_okay) {
+                       netdev_dbg(vif->dev,
+                                  "Bad status %d from copy to DOM%d.\n",
+                                  copy_op->status, vif->domid);
+                       status = XEN_NETIF_RSP_ERROR;
+               }
+       }
+
+       return status;
+}
+
+static struct xen_netif_rx_response *make_rx_response(
+       struct xenvif_queue *queue, u16 id, s8 st, u16 offset, u16 size,
+       u16 flags)
+{
+       RING_IDX i = queue->rx.rsp_prod_pvt;
+       struct xen_netif_rx_response *resp;
+
+       resp = RING_GET_RESPONSE(&queue->rx, i);
+       resp->offset     = offset;
+       resp->flags      = flags;
+       resp->id         = id;
+       resp->status     = (s16)size;
+       if (st < 0)
+               resp->status = (s16)st;
+
+       queue->rx.rsp_prod_pvt = ++i;
+
+       return resp;
+}
+
+static void xenvif_add_frag_responses(struct xenvif_queue *queue,
+                                     int status,
+                                     struct xenvif_rx_meta *meta,
+                                     int nr_meta_slots)
+{
+       int i;
+       unsigned long offset;
+
+       /* No fragments used */
+       if (nr_meta_slots <= 1)
+               return;
+
+       nr_meta_slots--;
+
+       for (i = 0; i < nr_meta_slots; i++) {
+               int flags;
+
+               if (i == nr_meta_slots - 1)
+                       flags = 0;
+               else
+                       flags = XEN_NETRXF_more_data;
+
+               offset = 0;
+               make_rx_response(queue, meta[i].id, status, offset,
+                                meta[i].size, flags);
+       }
+}
+
+static void xenvif_rx_action(struct xenvif_queue *queue)
+{
+       struct xenvif *vif = queue->vif;
+       s8 status;
+       u16 flags;
+       struct xen_netif_rx_response *resp;
+       struct sk_buff_head rxq;
+       struct sk_buff *skb;
+       LIST_HEAD(notify);
+       int ret;
+       unsigned long offset;
+       bool need_to_notify = false;
+
+       struct netrx_pending_operations npo = {
+               .copy  = queue->grant_copy_op,
+               .meta  = queue->meta,
+       };
+
+       skb_queue_head_init(&rxq);
+
+       while (xenvif_rx_ring_slots_available(queue) &&
+              (skb = xenvif_rx_dequeue(queue)) != NULL) {
+               queue->last_rx_time = jiffies;
+
+               XENVIF_RX_CB(skb)->meta_slots_used =
+                       xenvif_gop_skb(skb, &npo, queue);
+
+               __skb_queue_tail(&rxq, skb);
+       }
+
+       WARN_ON(npo.meta_prod > ARRAY_SIZE(queue->meta));
+
+       if (!npo.copy_prod)
+               goto done;
+
+       WARN_ON(npo.copy_prod > MAX_GRANT_COPY_OPS);
+       gnttab_batch_copy(queue->grant_copy_op, npo.copy_prod);
+
+       while ((skb = __skb_dequeue(&rxq)) != NULL) {
+               struct xen_netif_extra_info *extra = NULL;
+
+               if ((1 << queue->meta[npo.meta_cons].gso_type) &
+                   vif->gso_prefix_mask) {
+                       resp = RING_GET_RESPONSE(&queue->rx,
+                                                queue->rx.rsp_prod_pvt++);
+
+                       resp->flags = XEN_NETRXF_gso_prefix |
+                                     XEN_NETRXF_more_data;
+
+                       resp->offset = queue->meta[npo.meta_cons].gso_size;
+                       resp->id = queue->meta[npo.meta_cons].id;
+                       resp->status = XENVIF_RX_CB(skb)->meta_slots_used;
+
+                       npo.meta_cons++;
+                       XENVIF_RX_CB(skb)->meta_slots_used--;
+               }
+
+               queue->stats.tx_bytes += skb->len;
+               queue->stats.tx_packets++;
+
+               status = xenvif_check_gop(vif,
+                                         XENVIF_RX_CB(skb)->meta_slots_used,
+                                         &npo);
+
+               if (XENVIF_RX_CB(skb)->meta_slots_used == 1)
+                       flags = 0;
+               else
+                       flags = XEN_NETRXF_more_data;
+
+               if (skb->ip_summed == CHECKSUM_PARTIAL) /* local packet? */
+                       flags |= XEN_NETRXF_csum_blank |
+                                XEN_NETRXF_data_validated;
+               else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
+                       /* remote but checksummed. */
+                       flags |= XEN_NETRXF_data_validated;
+
+               offset = 0;
+               resp = make_rx_response(queue, queue->meta[npo.meta_cons].id,
+                                       status, offset,
+                                       queue->meta[npo.meta_cons].size,
+                                       flags);
+
+               if ((1 << queue->meta[npo.meta_cons].gso_type) &
+                   vif->gso_mask) {
+                       extra = (struct xen_netif_extra_info *)
+                               RING_GET_RESPONSE(&queue->rx,
+                                                 queue->rx.rsp_prod_pvt++);
+
+                       resp->flags |= XEN_NETRXF_extra_info;
+
+                       extra->u.gso.type = queue->meta[npo.meta_cons].gso_type;
+                       extra->u.gso.size = queue->meta[npo.meta_cons].gso_size;
+                       extra->u.gso.pad = 0;
+                       extra->u.gso.features = 0;
+
+                       extra->type = XEN_NETIF_EXTRA_TYPE_GSO;
+                       extra->flags = 0;
+               }
+
+               if (skb->sw_hash) {
+                       /* Since the skb got here via xenvif_select_queue()
+                        * we know that the hash has been re-calculated
+                        * according to a configuration set by the frontend
+                        * and therefore we know that it is legitimate to
+                        * pass it to the frontend.
+                        */
+                       if (resp->flags & XEN_NETRXF_extra_info)
+                               extra->flags |= XEN_NETIF_EXTRA_FLAG_MORE;
+                       else
+                               resp->flags |= XEN_NETRXF_extra_info;
+
+                       extra = (struct xen_netif_extra_info *)
+                               RING_GET_RESPONSE(&queue->rx,
+                                                 queue->rx.rsp_prod_pvt++);
+
+                       extra->u.hash.algorithm =
+                               XEN_NETIF_CTRL_HASH_ALGORITHM_TOEPLITZ;
+
+                       if (skb->l4_hash)
+                               extra->u.hash.type =
+                                       skb->protocol == htons(ETH_P_IP) ?
+                                       _XEN_NETIF_CTRL_HASH_TYPE_IPV4_TCP :
+                                       _XEN_NETIF_CTRL_HASH_TYPE_IPV6_TCP;
+                       else
+                               extra->u.hash.type =
+                                       skb->protocol == htons(ETH_P_IP) ?
+                                       _XEN_NETIF_CTRL_HASH_TYPE_IPV4 :
+                                       _XEN_NETIF_CTRL_HASH_TYPE_IPV6;
+
+                       *(uint32_t *)extra->u.hash.value =
+                               skb_get_hash_raw(skb);
+
+                       extra->type = XEN_NETIF_EXTRA_TYPE_HASH;
+                       extra->flags = 0;
+               }
+
+               xenvif_add_frag_responses(queue, status,
+                                         queue->meta + npo.meta_cons + 1,
+                                         XENVIF_RX_CB(skb)->meta_slots_used);
+
+               RING_PUSH_RESPONSES_AND_CHECK_NOTIFY(&queue->rx, ret);
+
+               need_to_notify |= !!ret;
+
+               npo.meta_cons += XENVIF_RX_CB(skb)->meta_slots_used;
+               dev_kfree_skb(skb);
+       }
+
+done:
+       if (need_to_notify)
+               notify_remote_via_irq(queue->rx_irq);
+}
+
+static bool xenvif_rx_queue_stalled(struct xenvif_queue *queue)
+{
+       RING_IDX prod, cons;
+
+       prod = queue->rx.sring->req_prod;
+       cons = queue->rx.req_cons;
+
+       return !queue->stalled &&
+               prod - cons < 1 &&
+               time_after(jiffies,
+                          queue->last_rx_time + queue->vif->stall_timeout);
+}
+
+static bool xenvif_rx_queue_ready(struct xenvif_queue *queue)
+{
+       RING_IDX prod, cons;
+
+       prod = queue->rx.sring->req_prod;
+       cons = queue->rx.req_cons;
+
+       return queue->stalled && prod - cons >= 1;
+}
+
+static bool xenvif_have_rx_work(struct xenvif_queue *queue)
+{
+       return xenvif_rx_ring_slots_available(queue) ||
+               (queue->vif->stall_timeout &&
+                (xenvif_rx_queue_stalled(queue) ||
+                 xenvif_rx_queue_ready(queue))) ||
+               kthread_should_stop() ||
+               queue->vif->disabled;
+}
+
+static long xenvif_rx_queue_timeout(struct xenvif_queue *queue)
+{
+       struct sk_buff *skb;
+       long timeout;
+
+       skb = skb_peek(&queue->rx_queue);
+       if (!skb)
+               return MAX_SCHEDULE_TIMEOUT;
+
+       timeout = XENVIF_RX_CB(skb)->expires - jiffies;
+       return timeout < 0 ? 0 : timeout;
+}
+
+/* Wait until the guest Rx thread has work.
+ *
+ * The timeout needs to be adjusted based on the current head of the
+ * queue (and not just the head at the beginning).  In particular, if
+ * the queue is initially empty an infinite timeout is used and this
+ * needs to be reduced when a skb is queued.
+ *
+ * This cannot be done with wait_event_timeout() because it only
+ * calculates the timeout once.
+ */
+static void xenvif_wait_for_rx_work(struct xenvif_queue *queue)
+{
+       DEFINE_WAIT(wait);
+
+       if (xenvif_have_rx_work(queue))
+               return;
+
+       for (;;) {
+               long ret;
+
+               prepare_to_wait(&queue->wq, &wait, TASK_INTERRUPTIBLE);
+               if (xenvif_have_rx_work(queue))
+                       break;
+               ret = schedule_timeout(xenvif_rx_queue_timeout(queue));
+               if (!ret)
+                       break;
+       }
+       finish_wait(&queue->wq, &wait);
+}
+
+static void xenvif_queue_carrier_off(struct xenvif_queue *queue)
+{
+       struct xenvif *vif = queue->vif;
+
+       queue->stalled = true;
+
+       /* At least one queue has stalled? Disable the carrier. */
+       spin_lock(&vif->lock);
+       if (vif->stalled_queues++ == 0) {
+               netdev_info(vif->dev, "Guest Rx stalled");
+               netif_carrier_off(vif->dev);
+       }
+       spin_unlock(&vif->lock);
+}
+
+static void xenvif_queue_carrier_on(struct xenvif_queue *queue)
+{
+       struct xenvif *vif = queue->vif;
+
+       queue->last_rx_time = jiffies; /* Reset Rx stall detection. */
+       queue->stalled = false;
+
+       /* All queues are ready? Enable the carrier. */
+       spin_lock(&vif->lock);
+       if (--vif->stalled_queues == 0) {
+               netdev_info(vif->dev, "Guest Rx ready");
+               netif_carrier_on(vif->dev);
+       }
+       spin_unlock(&vif->lock);
+}
+
+int xenvif_kthread_guest_rx(void *data)
+{
+       struct xenvif_queue *queue = data;
+       struct xenvif *vif = queue->vif;
+
+       if (!vif->stall_timeout)
+               xenvif_queue_carrier_on(queue);
+
+       for (;;) {
+               xenvif_wait_for_rx_work(queue);
+
+               if (kthread_should_stop())
+                       break;
+
+               /* This frontend is found to be rogue, disable it in
+                * kthread context. Currently this is only set when
+                * netback finds out frontend sends malformed packet,
+                * but we cannot disable the interface in softirq
+                * context so we defer it here, if this thread is
+                * associated with queue 0.
+                */
+               if (unlikely(vif->disabled && queue->id == 0)) {
+                       xenvif_carrier_off(vif);
+                       break;
+               }
+
+               if (!skb_queue_empty(&queue->rx_queue))
+                       xenvif_rx_action(queue);
+
+               /* If the guest hasn't provided any Rx slots for a
+                * while it's probably not responsive, drop the
+                * carrier so packets are dropped earlier.
+                */
+               if (vif->stall_timeout) {
+                       if (xenvif_rx_queue_stalled(queue))
+                               xenvif_queue_carrier_off(queue);
+                       else if (xenvif_rx_queue_ready(queue))
+                               xenvif_queue_carrier_on(queue);
+               }
+
+               /* Queued packets may have foreign pages from other
+                * domains.  These cannot be queued indefinitely as
+                * this would starve guests of grant refs and transmit
+                * slots.
+                */
+               xenvif_rx_queue_drop_expired(queue);
+
+               xenvif_rx_queue_maybe_wake(queue);
+
+               cond_resched();
+       }
+
+       /* Bin any remaining skbs */
+       xenvif_rx_queue_purge(queue);
+
+       return 0;
+}