Index: head/sys/dev/hyperv/netvsc/hv_net_vsc.c
===================================================================
--- head/sys/dev/hyperv/netvsc/hv_net_vsc.c	(revision 307989)
+++ head/sys/dev/hyperv/netvsc/hv_net_vsc.c	(revision 307990)
@@ -1,672 +1,687 @@
 /*-
  * Copyright (c) 2009-2012,2016 Microsoft Corp.
  * Copyright (c) 2010-2012 Citrix Inc.
  * Copyright (c) 2012 NetApp Inc.
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions
  * are met:
  * 1. Redistributions of source code must retain the above copyright
  *    notice unmodified, this list of conditions, and the following
  *    disclaimer.
  * 2. Redistributions in binary form must reproduce the above copyright
  *    notice, this list of conditions and the following disclaimer in the
  *    documentation and/or other materials provided with the distribution.
  *
  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  *
  * $FreeBSD$
  */
 
 /**
  * HyperV vmbus network VSC (virtual services client) module
  *
  */
 
 
 #include <sys/param.h>
 #include <sys/kernel.h>
 #include <sys/socket.h>
 #include <sys/limits.h>
 #include <sys/lock.h>
+#include <sys/taskqueue.h>
+
 #include <net/if.h>
 #include <net/if_var.h>
-#include <net/if_arp.h>
-#include <machine/bus.h>
-#include <machine/atomic.h>
+#include <net/if_media.h>
 
+#include <netinet/in.h>
+#include <netinet/tcp_lro.h>
+
 #include <dev/hyperv/include/hyperv.h>
+#include <dev/hyperv/include/hyperv_busdma.h>
+#include <dev/hyperv/include/vmbus.h>
 #include <dev/hyperv/include/vmbus_xact.h>
-#include <dev/hyperv/netvsc/hv_net_vsc.h>
-#include <dev/hyperv/netvsc/hv_rndis_filter.h>
+
+#include <dev/hyperv/netvsc/ndis.h>
 #include <dev/hyperv/netvsc/if_hnreg.h>
 #include <dev/hyperv/netvsc/if_hnvar.h>
+#include <dev/hyperv/netvsc/hv_net_vsc.h>
 
 /*
  * Forward declarations
  */
 static int  hn_nvs_conn_chim(struct hn_softc *sc);
 static int  hn_nvs_conn_rxbuf(struct hn_softc *);
 static int  hn_nvs_disconn_chim(struct hn_softc *sc);
 static int  hn_nvs_disconn_rxbuf(struct hn_softc *sc);
-static void hn_nvs_sent_none(struct hn_send_ctx *sndc,
+static void hn_nvs_sent_none(struct hn_nvs_sendctx *sndc,
     struct hn_softc *, struct vmbus_channel *chan,
     const void *, int);
 
-struct hn_send_ctx	hn_send_ctx_none =
-    HN_SEND_CTX_INITIALIZER(hn_nvs_sent_none, NULL);
+struct hn_nvs_sendctx	hn_nvs_sendctx_none =
+    HN_NVS_SENDCTX_INITIALIZER(hn_nvs_sent_none, NULL);
 
 static const uint32_t		hn_nvs_version[] = {
 	HN_NVS_VERSION_5,
 	HN_NVS_VERSION_4,
 	HN_NVS_VERSION_2,
 	HN_NVS_VERSION_1
 };
 
 static const void *
 hn_nvs_xact_execute(struct hn_softc *sc, struct vmbus_xact *xact,
     void *req, int reqlen, size_t *resplen0, uint32_t type)
 {
-	struct hn_send_ctx sndc;
+	struct hn_nvs_sendctx sndc;
 	size_t resplen, min_resplen = *resplen0;
 	const struct hn_nvs_hdr *hdr;
 	int error;
 
 	KASSERT(min_resplen >= sizeof(*hdr),
 	    ("invalid minimum response len %zu", min_resplen));
 
 	/*
 	 * Execute the xact setup by the caller.
 	 */
-	hn_send_ctx_init(&sndc, hn_nvs_sent_xact, xact);
+	hn_nvs_sendctx_init(&sndc, hn_nvs_sent_xact, xact);
 
 	vmbus_xact_activate(xact);
 	error = hn_nvs_send(sc->hn_prichan, VMBUS_CHANPKT_FLAG_RC,
 	    req, reqlen, &sndc);
 	if (error) {
 		vmbus_xact_deactivate(xact);
 		return (NULL);
 	}
 	hdr = vmbus_xact_wait(xact, &resplen);
 
 	/*
 	 * Check this NVS response message.
 	 */
 	if (resplen < min_resplen) {
 		if_printf(sc->hn_ifp, "invalid NVS resp len %zu\n", resplen);
 		return (NULL);
 	}
 	if (hdr->nvs_type != type) {
 		if_printf(sc->hn_ifp, "unexpected NVS resp 0x%08x, "
 		    "expect 0x%08x\n", hdr->nvs_type, type);
 		return (NULL);
 	}
 	/* All pass! */
 	*resplen0 = resplen;
 	return (hdr);
 }
 
 static __inline int
 hn_nvs_req_send(struct hn_softc *sc, void *req, int reqlen)
 {
 
 	return (hn_nvs_send(sc->hn_prichan, VMBUS_CHANPKT_FLAG_NONE,
-	    req, reqlen, &hn_send_ctx_none));
+	    req, reqlen, &hn_nvs_sendctx_none));
 }
 
 static int 
 hn_nvs_conn_rxbuf(struct hn_softc *sc)
 {
 	struct vmbus_xact *xact = NULL;
 	struct hn_nvs_rxbuf_conn *conn;
 	const struct hn_nvs_rxbuf_connresp *resp;
 	size_t resp_len;
 	uint32_t status;
 	int error, rxbuf_size;
 
 	/*
 	 * Limit RXBUF size for old NVS.
 	 */
 	if (sc->hn_nvs_ver <= HN_NVS_VERSION_2)
 		rxbuf_size = HN_RXBUF_SIZE_COMPAT;
 	else
 		rxbuf_size = HN_RXBUF_SIZE;
 
 	/*
 	 * Connect the RXBUF GPADL to the primary channel.
 	 *
 	 * NOTE:
 	 * Only primary channel has RXBUF connected to it.  Sub-channels
 	 * just share this RXBUF.
 	 */
 	error = vmbus_chan_gpadl_connect(sc->hn_prichan,
 	    sc->hn_rxbuf_dma.hv_paddr, rxbuf_size, &sc->hn_rxbuf_gpadl);
 	if (error) {
 		if_printf(sc->hn_ifp, "rxbuf gpadl conn failed: %d\n",
 		    error);
 		goto cleanup;
 	}
 
 	/*
 	 * Connect RXBUF to NVS.
 	 */
 
 	xact = vmbus_xact_get(sc->hn_xact, sizeof(*conn));
 	if (xact == NULL) {
 		if_printf(sc->hn_ifp, "no xact for nvs rxbuf conn\n");
 		error = ENXIO;
 		goto cleanup;
 	}
 	conn = vmbus_xact_req_data(xact);
 	conn->nvs_type = HN_NVS_TYPE_RXBUF_CONN;
 	conn->nvs_gpadl = sc->hn_rxbuf_gpadl;
 	conn->nvs_sig = HN_NVS_RXBUF_SIG;
 
 	resp_len = sizeof(*resp);
 	resp = hn_nvs_xact_execute(sc, xact, conn, sizeof(*conn), &resp_len,
 	    HN_NVS_TYPE_RXBUF_CONNRESP);
 	if (resp == NULL) {
 		if_printf(sc->hn_ifp, "exec nvs rxbuf conn failed\n");
 		error = EIO;
 		goto cleanup;
 	}
 
 	status = resp->nvs_status;
 	vmbus_xact_put(xact);
 	xact = NULL;
 
 	if (status != HN_NVS_STATUS_OK) {
 		if_printf(sc->hn_ifp, "nvs rxbuf conn failed: %x\n", status);
 		error = EIO;
 		goto cleanup;
 	}
 	sc->hn_flags |= HN_FLAG_RXBUF_CONNECTED;
 
 	return (0);
 
 cleanup:
 	if (xact != NULL)
 		vmbus_xact_put(xact);
 	hn_nvs_disconn_rxbuf(sc);
 	return (error);
 }
 
 static int 
 hn_nvs_conn_chim(struct hn_softc *sc)
 {
 	struct vmbus_xact *xact = NULL;
 	struct hn_nvs_chim_conn *chim;
 	const struct hn_nvs_chim_connresp *resp;
 	size_t resp_len;
 	uint32_t status, sectsz;
 	int error;
 
 	/*
 	 * Connect chimney sending buffer GPADL to the primary channel.
 	 *
 	 * NOTE:
 	 * Only primary channel has chimney sending buffer connected to it.
 	 * Sub-channels just share this chimney sending buffer.
 	 */
 	error = vmbus_chan_gpadl_connect(sc->hn_prichan,
   	    sc->hn_chim_dma.hv_paddr, HN_CHIM_SIZE, &sc->hn_chim_gpadl);
 	if (error) {
 		if_printf(sc->hn_ifp, "chim gpadl conn failed: %d\n", error);
 		goto cleanup;
 	}
 
 	/*
 	 * Connect chimney sending buffer to NVS
 	 */
 
 	xact = vmbus_xact_get(sc->hn_xact, sizeof(*chim));
 	if (xact == NULL) {
 		if_printf(sc->hn_ifp, "no xact for nvs chim conn\n");
 		error = ENXIO;
 		goto cleanup;
 	}
 	chim = vmbus_xact_req_data(xact);
 	chim->nvs_type = HN_NVS_TYPE_CHIM_CONN;
 	chim->nvs_gpadl = sc->hn_chim_gpadl;
 	chim->nvs_sig = HN_NVS_CHIM_SIG;
 
 	resp_len = sizeof(*resp);
 	resp = hn_nvs_xact_execute(sc, xact, chim, sizeof(*chim), &resp_len,
 	    HN_NVS_TYPE_CHIM_CONNRESP);
 	if (resp == NULL) {
 		if_printf(sc->hn_ifp, "exec nvs chim conn failed\n");
 		error = EIO;
 		goto cleanup;
 	}
 
 	status = resp->nvs_status;
 	sectsz = resp->nvs_sectsz;
 	vmbus_xact_put(xact);
 	xact = NULL;
 
 	if (status != HN_NVS_STATUS_OK) {
 		if_printf(sc->hn_ifp, "nvs chim conn failed: %x\n", status);
 		error = EIO;
 		goto cleanup;
 	}
 	if (sectsz == 0) {
 		if_printf(sc->hn_ifp, "zero chimney sending buffer "
 		    "section size\n");
 		return (0);
 	}
 
 	sc->hn_chim_szmax = sectsz;
 	sc->hn_chim_cnt = HN_CHIM_SIZE / sc->hn_chim_szmax;
 	if (HN_CHIM_SIZE % sc->hn_chim_szmax != 0) {
 		if_printf(sc->hn_ifp, "chimney sending sections are "
 		    "not properly aligned\n");
 	}
 	if (sc->hn_chim_cnt % LONG_BIT != 0) {
 		if_printf(sc->hn_ifp, "discard %d chimney sending sections\n",
 		    sc->hn_chim_cnt % LONG_BIT);
 	}
 
 	sc->hn_chim_bmap_cnt = sc->hn_chim_cnt / LONG_BIT;
 	sc->hn_chim_bmap = malloc(sc->hn_chim_bmap_cnt * sizeof(u_long),
 	    M_DEVBUF, M_WAITOK | M_ZERO);
 
 	/* Done! */
 	sc->hn_flags |= HN_FLAG_CHIM_CONNECTED;
 	if (bootverbose) {
 		if_printf(sc->hn_ifp, "chimney sending buffer %d/%d\n",
 		    sc->hn_chim_szmax, sc->hn_chim_cnt);
 	}
 	return (0);
 
 cleanup:
 	if (xact != NULL)
 		vmbus_xact_put(xact);
 	hn_nvs_disconn_chim(sc);
 	return (error);
 }
 
 static int
 hn_nvs_disconn_rxbuf(struct hn_softc *sc)
 {
 	int error;
 
 	if (sc->hn_flags & HN_FLAG_RXBUF_CONNECTED) {
 		struct hn_nvs_rxbuf_disconn disconn;
 
 		/*
 		 * Disconnect RXBUF from NVS.
 		 */
 		memset(&disconn, 0, sizeof(disconn));
 		disconn.nvs_type = HN_NVS_TYPE_RXBUF_DISCONN;
 		disconn.nvs_sig = HN_NVS_RXBUF_SIG;
 
 		/* NOTE: No response. */
 		error = hn_nvs_req_send(sc, &disconn, sizeof(disconn));
 		if (error) {
 			if_printf(sc->hn_ifp,
 			    "send nvs rxbuf disconn failed: %d\n", error);
 			return (error);
 		}
 		sc->hn_flags &= ~HN_FLAG_RXBUF_CONNECTED;
 
 		/*
 		 * Wait for the hypervisor to receive this NVS request.
 		 */
 		while (!vmbus_chan_tx_empty(sc->hn_prichan))
 			pause("waittx", 1);
 		/*
 		 * Linger long enough for NVS to disconnect RXBUF.
 		 */
 		pause("lingtx", (200 * hz) / 1000);
 	}
 
 	if (sc->hn_rxbuf_gpadl != 0) {
 		/*
 		 * Disconnect RXBUF from primary channel.
 		 */
 		error = vmbus_chan_gpadl_disconnect(sc->hn_prichan,
 		    sc->hn_rxbuf_gpadl);
 		if (error) {
 			if_printf(sc->hn_ifp,
 			    "rxbuf gpadl disconn failed: %d\n", error);
 			return (error);
 		}
 		sc->hn_rxbuf_gpadl = 0;
 	}
 	return (0);
 }
 
 static int
 hn_nvs_disconn_chim(struct hn_softc *sc)
 {
 	int error;
 
 	if (sc->hn_flags & HN_FLAG_CHIM_CONNECTED) {
 		struct hn_nvs_chim_disconn disconn;
 
 		/*
 		 * Disconnect chimney sending buffer from NVS.
 		 */
 		memset(&disconn, 0, sizeof(disconn));
 		disconn.nvs_type = HN_NVS_TYPE_CHIM_DISCONN;
 		disconn.nvs_sig = HN_NVS_CHIM_SIG;
 
 		/* NOTE: No response. */
 		error = hn_nvs_req_send(sc, &disconn, sizeof(disconn));
 		if (error) {
 			if_printf(sc->hn_ifp,
 			    "send nvs chim disconn failed: %d\n", error);
 			return (error);
 		}
 		sc->hn_flags &= ~HN_FLAG_CHIM_CONNECTED;
 
 		/*
 		 * Wait for the hypervisor to receive this NVS request.
 		 */
 		while (!vmbus_chan_tx_empty(sc->hn_prichan))
 			pause("waittx", 1);
 		/*
 		 * Linger long enough for NVS to disconnect chimney
 		 * sending buffer.
 		 */
 		pause("lingtx", (200 * hz) / 1000);
 	}
 
 	if (sc->hn_chim_gpadl != 0) {
 		/*
 		 * Disconnect chimney sending buffer from primary channel.
 		 */
 		error = vmbus_chan_gpadl_disconnect(sc->hn_prichan,
 		    sc->hn_chim_gpadl);
 		if (error) {
 			if_printf(sc->hn_ifp,
 			    "chim gpadl disconn failed: %d\n", error);
 			return (error);
 		}
 		sc->hn_chim_gpadl = 0;
 	}
 
 	if (sc->hn_chim_bmap != NULL) {
 		free(sc->hn_chim_bmap, M_DEVBUF);
 		sc->hn_chim_bmap = NULL;
 	}
 	return (0);
 }
 
 static int
 hn_nvs_doinit(struct hn_softc *sc, uint32_t nvs_ver)
 {
 	struct vmbus_xact *xact;
 	struct hn_nvs_init *init;
 	const struct hn_nvs_init_resp *resp;
 	size_t resp_len;
 	uint32_t status;
 
 	xact = vmbus_xact_get(sc->hn_xact, sizeof(*init));
 	if (xact == NULL) {
 		if_printf(sc->hn_ifp, "no xact for nvs init\n");
 		return (ENXIO);
 	}
 	init = vmbus_xact_req_data(xact);
 	init->nvs_type = HN_NVS_TYPE_INIT;
 	init->nvs_ver_min = nvs_ver;
 	init->nvs_ver_max = nvs_ver;
 
 	resp_len = sizeof(*resp);
 	resp = hn_nvs_xact_execute(sc, xact, init, sizeof(*init), &resp_len,
 	    HN_NVS_TYPE_INIT_RESP);
 	if (resp == NULL) {
 		if_printf(sc->hn_ifp, "exec init failed\n");
 		vmbus_xact_put(xact);
 		return (EIO);
 	}
 
 	status = resp->nvs_status;
 	vmbus_xact_put(xact);
 
 	if (status != HN_NVS_STATUS_OK) {
 		if (bootverbose) {
 			/*
 			 * Caller may try another NVS version, and will log
 			 * error if there are no more NVS versions to try,
 			 * so don't bark out loud here.
 			 */
 			if_printf(sc->hn_ifp, "nvs init failed for ver 0x%x\n",
 			    nvs_ver);
 		}
 		return (EINVAL);
 	}
 	return (0);
 }
 
 /*
  * Configure MTU and enable VLAN.
  */
 static int
 hn_nvs_conf_ndis(struct hn_softc *sc, int mtu)
 {
 	struct hn_nvs_ndis_conf conf;
 	int error;
 
 	memset(&conf, 0, sizeof(conf));
 	conf.nvs_type = HN_NVS_TYPE_NDIS_CONF;
 	conf.nvs_mtu = mtu;
 	conf.nvs_caps = HN_NVS_NDIS_CONF_VLAN;
 
 	/* NOTE: No response. */
 	error = hn_nvs_req_send(sc, &conf, sizeof(conf));
 	if (error) {
 		if_printf(sc->hn_ifp, "send nvs ndis conf failed: %d\n", error);
 		return (error);
 	}
 
 	if (bootverbose)
 		if_printf(sc->hn_ifp, "nvs ndis conf done\n");
 	sc->hn_caps |= HN_CAP_MTU | HN_CAP_VLAN;
 	return (0);
 }
 
 static int
 hn_nvs_init_ndis(struct hn_softc *sc)
 {
 	struct hn_nvs_ndis_init ndis;
 	int error;
 
 	memset(&ndis, 0, sizeof(ndis));
 	ndis.nvs_type = HN_NVS_TYPE_NDIS_INIT;
 	ndis.nvs_ndis_major = HN_NDIS_VERSION_MAJOR(sc->hn_ndis_ver);
 	ndis.nvs_ndis_minor = HN_NDIS_VERSION_MINOR(sc->hn_ndis_ver);
 
 	/* NOTE: No response. */
 	error = hn_nvs_req_send(sc, &ndis, sizeof(ndis));
 	if (error)
 		if_printf(sc->hn_ifp, "send nvs ndis init failed: %d\n", error);
 	return (error);
 }
 
 static int
 hn_nvs_init(struct hn_softc *sc)
 {
 	int i, error;
 
 	if (device_is_attached(sc->hn_dev)) {
 		/*
 		 * NVS version and NDIS version MUST NOT be changed.
 		 */
 		if (bootverbose) {
 			if_printf(sc->hn_ifp, "reinit NVS version 0x%x, "
 			    "NDIS version %u.%u\n", sc->hn_nvs_ver,
 			    HN_NDIS_VERSION_MAJOR(sc->hn_ndis_ver),
 			    HN_NDIS_VERSION_MINOR(sc->hn_ndis_ver));
 		}
 
 		error = hn_nvs_doinit(sc, sc->hn_nvs_ver);
 		if (error) {
 			if_printf(sc->hn_ifp, "reinit NVS version 0x%x "
 			    "failed: %d\n", sc->hn_nvs_ver, error);
 			return (error);
 		}
 		goto done;
 	}
 
 	/*
 	 * Find the supported NVS version and set NDIS version accordingly.
 	 */
 	for (i = 0; i < nitems(hn_nvs_version); ++i) {
 		error = hn_nvs_doinit(sc, hn_nvs_version[i]);
 		if (!error) {
 			sc->hn_nvs_ver = hn_nvs_version[i];
 
 			/* Set NDIS version according to NVS version. */
 			sc->hn_ndis_ver = HN_NDIS_VERSION_6_30;
 			if (sc->hn_nvs_ver <= HN_NVS_VERSION_4)
 				sc->hn_ndis_ver = HN_NDIS_VERSION_6_1;
 
 			if (bootverbose) {
 				if_printf(sc->hn_ifp, "NVS version 0x%x, "
 				    "NDIS version %u.%u\n", sc->hn_nvs_ver,
 				    HN_NDIS_VERSION_MAJOR(sc->hn_ndis_ver),
 				    HN_NDIS_VERSION_MINOR(sc->hn_ndis_ver));
 			}
 			goto done;
 		}
 	}
 	if_printf(sc->hn_ifp, "no NVS available\n");
 	return (ENXIO);
 
 done:
 	if (sc->hn_nvs_ver >= HN_NVS_VERSION_5)
 		sc->hn_caps |= HN_CAP_HASHVAL;
 	return (0);
 }
 
 int
 hn_nvs_attach(struct hn_softc *sc, int mtu)
 {
 	int error;
 
 	/*
 	 * Initialize NVS.
 	 */
 	error = hn_nvs_init(sc);
 	if (error)
 		return (error);
 
 	if (sc->hn_nvs_ver >= HN_NVS_VERSION_2) {
 		/*
 		 * Configure NDIS before initializing it.
 		 */
 		error = hn_nvs_conf_ndis(sc, mtu);
 		if (error)
 			return (error);
 	}
 
 	/*
 	 * Initialize NDIS.
 	 */
 	error = hn_nvs_init_ndis(sc);
 	if (error)
 		return (error);
 
 	/*
 	 * Connect RXBUF.
 	 */
 	error = hn_nvs_conn_rxbuf(sc);
 	if (error)
 		return (error);
 
 	/*
 	 * Connect chimney sending buffer.
 	 */
 	error = hn_nvs_conn_chim(sc);
 	if (error)
 		return (error);
 	return (0);
 }
 
 void
 hn_nvs_detach(struct hn_softc *sc)
 {
 
 	/* NOTE: there are no requests to stop the NVS. */
 	hn_nvs_disconn_rxbuf(sc);
 	hn_nvs_disconn_chim(sc);
 }
 
 void
-hn_nvs_sent_xact(struct hn_send_ctx *sndc,
+hn_nvs_sent_xact(struct hn_nvs_sendctx *sndc,
     struct hn_softc *sc __unused, struct vmbus_channel *chan __unused,
     const void *data, int dlen)
 {
 
 	vmbus_xact_wakeup(sndc->hn_cbarg, data, dlen);
 }
 
 static void
-hn_nvs_sent_none(struct hn_send_ctx *sndc __unused,
+hn_nvs_sent_none(struct hn_nvs_sendctx *sndc __unused,
     struct hn_softc *sc __unused, struct vmbus_channel *chan __unused,
     const void *data __unused, int dlen __unused)
 {
 	/* EMPTY */
 }
 
 int
 hn_nvs_alloc_subchans(struct hn_softc *sc, int *nsubch0)
 {
 	struct vmbus_xact *xact;
 	struct hn_nvs_subch_req *req;
 	const struct hn_nvs_subch_resp *resp;
 	int error, nsubch_req;
 	uint32_t nsubch;
 	size_t resp_len;
 
 	nsubch_req = *nsubch0;
 	KASSERT(nsubch_req > 0, ("invalid # of sub-channels %d", nsubch_req));
 
 	xact = vmbus_xact_get(sc->hn_xact, sizeof(*req));
 	if (xact == NULL) {
 		if_printf(sc->hn_ifp, "no xact for nvs subch alloc\n");
 		return (ENXIO);
 	}
 	req = vmbus_xact_req_data(xact);
 	req->nvs_type = HN_NVS_TYPE_SUBCH_REQ;
 	req->nvs_op = HN_NVS_SUBCH_OP_ALLOC;
 	req->nvs_nsubch = nsubch_req;
 
 	resp_len = sizeof(*resp);
 	resp = hn_nvs_xact_execute(sc, xact, req, sizeof(*req), &resp_len,
 	    HN_NVS_TYPE_SUBCH_RESP);
 	if (resp == NULL) {
 		if_printf(sc->hn_ifp, "exec nvs subch alloc failed\n");
 		error = EIO;
 		goto done;
 	}
 	if (resp->nvs_status != HN_NVS_STATUS_OK) {
 		if_printf(sc->hn_ifp, "nvs subch alloc failed: %x\n",
 		    resp->nvs_status);
 		error = EIO;
 		goto done;
 	}
 
 	nsubch = resp->nvs_nsubch;
 	if (nsubch > nsubch_req) {
 		if_printf(sc->hn_ifp, "%u subchans are allocated, "
 		    "requested %d\n", nsubch, nsubch_req);
 		nsubch = nsubch_req;
 	}
 	*nsubch0 = nsubch;
 	error = 0;
 done:
 	vmbus_xact_put(xact);
 	return (error);
+}
+
+int
+hn_nvs_send_rndis_ctrl(struct vmbus_channel *chan,
+    struct hn_nvs_sendctx *sndc, struct vmbus_gpa *gpa, int gpa_cnt)
+{
+
+	return hn_nvs_send_rndis_sglist(chan, HN_NVS_RNDIS_MTYPE_CTRL,
+	    sndc, gpa, gpa_cnt);
 }
Index: head/sys/dev/hyperv/netvsc/hv_net_vsc.h
===================================================================
--- head/sys/dev/hyperv/netvsc/hv_net_vsc.h	(revision 307989)
+++ head/sys/dev/hyperv/netvsc/hv_net_vsc.h	(revision 307990)
@@ -1,61 +1,106 @@
 /*-
  * Copyright (c) 2009-2012,2016 Microsoft Corp.
  * Copyright (c) 2010-2012 Citrix Inc.
  * Copyright (c) 2012 NetApp Inc.
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions
  * are met:
  * 1. Redistributions of source code must retain the above copyright
  *    notice unmodified, this list of conditions, and the following
  *    disclaimer.
  * 2. Redistributions in binary form must reproduce the above copyright
  *    notice, this list of conditions and the following disclaimer in the
  *    documentation and/or other materials provided with the distribution.
  *
  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  *
  * $FreeBSD$
  */
 
 #ifndef __HV_NET_VSC_H__
 #define __HV_NET_VSC_H__
 
-#include <sys/param.h>
-#include <sys/kernel.h>
-#include <sys/lock.h>
-#include <sys/malloc.h>
-#include <sys/queue.h>
-#include <sys/taskqueue.h>
-#include <sys/sema.h>
-#include <sys/sx.h>
+struct hn_nvs_sendctx;
+struct vmbus_channel;
+struct hn_softc;
 
-#include <machine/bus.h>
-#include <sys/bus.h>
-#include <sys/bus_dma.h>
+typedef void		(*hn_nvs_sent_t)
+			(struct hn_nvs_sendctx *, struct hn_softc *,
+			 struct vmbus_channel *, const void *, int);
 
-#include <netinet/in.h>
-#include <netinet/tcp_lro.h>
+struct hn_nvs_sendctx {
+	hn_nvs_sent_t	hn_cb;
+	void		*hn_cbarg;
+};
 
-#include <net/ethernet.h>
-#include <net/if.h>
-#include <net/if_media.h>
+#define HN_NVS_SENDCTX_INITIALIZER(cb, cbarg)	\
+{						\
+	.hn_cb		= cb,			\
+	.hn_cbarg	= cbarg			\
+}
 
-#include <dev/hyperv/include/hyperv.h>
-#include <dev/hyperv/include/hyperv_busdma.h>
-#include <dev/hyperv/include/vmbus.h>
+static __inline void
+hn_nvs_sendctx_init(struct hn_nvs_sendctx *sndc, hn_nvs_sent_t cb, void *cbarg)
+{
 
-#include <dev/hyperv/netvsc/ndis.h>
+	sndc->hn_cb = cb;
+	sndc->hn_cbarg = cbarg;
+}
 
-#endif  /* __HV_NET_VSC_H__ */
+static __inline int
+hn_nvs_send(struct vmbus_channel *chan, uint16_t flags,
+    void *nvs_msg, int nvs_msglen, struct hn_nvs_sendctx *sndc)
+{
 
+	return (vmbus_chan_send(chan, VMBUS_CHANPKT_TYPE_INBAND, flags,
+	    nvs_msg, nvs_msglen, (uint64_t)(uintptr_t)sndc));
+}
+
+static __inline int
+hn_nvs_send_sglist(struct vmbus_channel *chan, struct vmbus_gpa sg[], int sglen,
+    void *nvs_msg, int nvs_msglen, struct hn_nvs_sendctx *sndc)
+{
+
+	return (vmbus_chan_send_sglist(chan, sg, sglen, nvs_msg, nvs_msglen,
+	    (uint64_t)(uintptr_t)sndc));
+}
+
+static __inline int
+hn_nvs_send_rndis_sglist(struct vmbus_channel *chan, uint32_t rndis_mtype,
+    struct hn_nvs_sendctx *sndc, struct vmbus_gpa *gpa, int gpa_cnt)
+{
+	struct hn_nvs_rndis rndis;
+
+	rndis.nvs_type = HN_NVS_TYPE_RNDIS;
+	rndis.nvs_rndis_mtype = rndis_mtype;
+	rndis.nvs_chim_idx = HN_NVS_CHIM_IDX_INVALID;
+	rndis.nvs_chim_sz = 0;
+
+	return (hn_nvs_send_sglist(chan, gpa, gpa_cnt,
+	    &rndis, sizeof(rndis), sndc));
+}
+
+int		hn_nvs_attach(struct hn_softc *sc, int mtu);
+void		hn_nvs_detach(struct hn_softc *sc);
+int		hn_nvs_alloc_subchans(struct hn_softc *sc, int *nsubch);
+void		hn_nvs_sent_xact(struct hn_nvs_sendctx *sndc,
+		    struct hn_softc *sc, struct vmbus_channel *chan,
+		    const void *data, int dlen);
+int		hn_nvs_send_rndis_ctrl(struct vmbus_channel *chan,
+		    struct hn_nvs_sendctx *sndc, struct vmbus_gpa *gpa,
+		    int gpa_cnt);
+
+extern struct hn_nvs_sendctx	hn_nvs_sendctx_none;
+
+#endif  /* __HV_NET_VSC_H__ */
Index: head/sys/dev/hyperv/netvsc/hv_netvsc_drv_freebsd.c
===================================================================
--- head/sys/dev/hyperv/netvsc/hv_netvsc_drv_freebsd.c	(revision 307989)
+++ head/sys/dev/hyperv/netvsc/hv_netvsc_drv_freebsd.c	(revision 307990)
@@ -1,4296 +1,4272 @@
 /*-
  * Copyright (c) 2010-2012 Citrix Inc.
  * Copyright (c) 2009-2012,2016 Microsoft Corp.
  * Copyright (c) 2012 NetApp Inc.
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions
  * are met:
  * 1. Redistributions of source code must retain the above copyright
  *    notice unmodified, this list of conditions, and the following
  *    disclaimer.
  * 2. Redistributions in binary form must reproduce the above copyright
  *    notice, this list of conditions and the following disclaimer in the
  *    documentation and/or other materials provided with the distribution.
  *
  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */
 
 /*-
  * Copyright (c) 2004-2006 Kip Macy
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions
  * are met:
  * 1. Redistributions of source code must retain the above copyright
  *    notice, this list of conditions and the following disclaimer.
  * 2. Redistributions in binary form must reproduce the above copyright
  *    notice, this list of conditions and the following disclaimer in the
  *    documentation and/or other materials provided with the distribution.
  *
  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  * SUCH DAMAGE.
  */
 
 #include <sys/cdefs.h>
 __FBSDID("$FreeBSD$");
 
 #include "opt_inet6.h"
 #include "opt_inet.h"
 
 #include <sys/param.h>
 #include <sys/systm.h>
 #include <sys/sockio.h>
 #include <sys/limits.h>
 #include <sys/mbuf.h>
 #include <sys/malloc.h>
 #include <sys/module.h>
 #include <sys/kernel.h>
 #include <sys/socket.h>
 #include <sys/queue.h>
 #include <sys/lock.h>
 #include <sys/sx.h>
 #include <sys/smp.h>
 #include <sys/sysctl.h>
 #include <sys/buf_ring.h>
 #include <sys/taskqueue.h>
 
 #include <net/if.h>
 #include <net/if_arp.h>
 #include <net/ethernet.h>
 #include <net/if_dl.h>
 #include <net/if_media.h>
 #include <net/rndis.h>
 #include <net/bpf.h>
 
 #include <net/if_var.h>
 #include <net/if_types.h>
 #include <net/if_vlan_var.h>
 
 #include <netinet/in_systm.h>
 #include <netinet/in.h>
 #include <netinet/ip.h>
 #include <netinet/if_ether.h>
 #include <netinet/tcp.h>
 #include <netinet/tcp_lro.h>
 #include <netinet/udp.h>
 #include <netinet/ip6.h>
 
 #include <vm/vm.h>
 #include <vm/vm_param.h>
 #include <vm/vm_kern.h>
 #include <vm/pmap.h>
 
 #include <machine/bus.h>
 #include <machine/resource.h>
 #include <machine/frame.h>
 
 #include <sys/bus.h>
 #include <sys/rman.h>
 #include <sys/mutex.h>
 #include <sys/errno.h>
 #include <sys/types.h>
 #include <machine/atomic.h>
 
 #include <machine/intr_machdep.h>
 
 #include <machine/in_cksum.h>
 
 #include <dev/hyperv/include/hyperv.h>
 #include <dev/hyperv/include/hyperv_busdma.h>
 #include <dev/hyperv/include/vmbus.h>
 #include <dev/hyperv/include/vmbus_xact.h>
 
 #include <dev/hyperv/netvsc/ndis.h>
 #include <dev/hyperv/netvsc/if_hnreg.h>
 #include <dev/hyperv/netvsc/if_hnvar.h>
 #include <dev/hyperv/netvsc/hv_net_vsc.h>
 #include <dev/hyperv/netvsc/hv_rndis_filter.h>
 
 #include "vmbus_if.h"
 
 /* Short for Hyper-V network interface */
 #define NETVSC_DEVNAME    "hn"
 
 /*
  * It looks like offset 0 of buf is reserved to hold the softc pointer.
  * The sc pointer evidently not needed, and is not presently populated.
  * The packet offset is where the netvsc_packet starts in the buffer.
  */
 #define HV_NV_SC_PTR_OFFSET_IN_BUF         0
 #define HV_NV_PACKET_OFFSET_IN_BUF         16
 
 /* YYY should get it from the underlying channel */
 #define HN_TX_DESC_CNT			512
 
 #define HN_LROENT_CNT_DEF		128
 
 #define HN_RING_CNT_DEF_MAX		8
 
 #define HN_RNDIS_PKT_LEN					\
 	(sizeof(struct rndis_packet_msg) +			\
 	 HN_RNDIS_PKTINFO_SIZE(HN_NDIS_HASH_VALUE_SIZE) +	\
 	 HN_RNDIS_PKTINFO_SIZE(NDIS_VLAN_INFO_SIZE) +		\
 	 HN_RNDIS_PKTINFO_SIZE(NDIS_LSO2_INFO_SIZE) +		\
 	 HN_RNDIS_PKTINFO_SIZE(NDIS_TXCSUM_INFO_SIZE))
 #define HN_RNDIS_PKT_BOUNDARY		PAGE_SIZE
 #define HN_RNDIS_PKT_ALIGN		CACHE_LINE_SIZE
 
 #define HN_TX_DATA_BOUNDARY		PAGE_SIZE
 #define HN_TX_DATA_MAXSIZE		IP_MAXPACKET
 #define HN_TX_DATA_SEGSIZE		PAGE_SIZE
 /* -1 for RNDIS packet message */
 #define HN_TX_DATA_SEGCNT_MAX		(HN_GPACNT_MAX - 1)
 
 #define HN_DIRECT_TX_SIZE_DEF		128
 
 #define HN_EARLY_TXEOF_THRESH		8
 
 struct hn_txdesc {
 #ifndef HN_USE_TXDESC_BUFRING
 	SLIST_ENTRY(hn_txdesc) link;
 #endif
 	struct mbuf	*m;
 	struct hn_tx_ring *txr;
 	int		refs;
 	uint32_t	flags;		/* HN_TXD_FLAG_ */
-	struct hn_send_ctx send_ctx;
+	struct hn_nvs_sendctx send_ctx;
 	uint32_t	chim_index;
 	int		chim_size;
 
 	bus_dmamap_t	data_dmap;
 
 	bus_addr_t	rndis_pkt_paddr;
 	struct rndis_packet_msg *rndis_pkt;
 	bus_dmamap_t	rndis_pkt_dmap;
 };
 
 #define HN_TXD_FLAG_ONLIST	0x1
 #define HN_TXD_FLAG_DMAMAP	0x2
 
 #define HN_LRO_LENLIM_MULTIRX_DEF	(12 * ETHERMTU)
 #define HN_LRO_LENLIM_DEF		(25 * ETHERMTU)
 /* YYY 2*MTU is a bit rough, but should be good enough. */
 #define HN_LRO_LENLIM_MIN(ifp)		(2 * (ifp)->if_mtu)
 
 #define HN_LRO_ACKCNT_DEF		1
 
 #define HN_LOCK_INIT(sc)		\
 	sx_init(&(sc)->hn_lock, device_get_nameunit((sc)->hn_dev))
 #define HN_LOCK_ASSERT(sc)		sx_assert(&(sc)->hn_lock, SA_XLOCKED)
 #define HN_LOCK_DESTROY(sc)		sx_destroy(&(sc)->hn_lock)
 #define HN_LOCK(sc)			sx_xlock(&(sc)->hn_lock)
 #define HN_UNLOCK(sc)			sx_xunlock(&(sc)->hn_lock)
 
 #define HN_CSUM_IP_MASK			(CSUM_IP | CSUM_IP_TCP | CSUM_IP_UDP)
 #define HN_CSUM_IP6_MASK		(CSUM_IP6_TCP | CSUM_IP6_UDP)
 #define HN_CSUM_IP_HWASSIST(sc)		\
 	((sc)->hn_tx_ring[0].hn_csum_assist & HN_CSUM_IP_MASK)
 #define HN_CSUM_IP6_HWASSIST(sc)	\
 	((sc)->hn_tx_ring[0].hn_csum_assist & HN_CSUM_IP6_MASK)
 
 /*
  * Globals
  */
 
 SYSCTL_NODE(_hw, OID_AUTO, hn, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL,
     "Hyper-V network interface");
 
 /* Trust tcp segements verification on host side. */
 static int hn_trust_hosttcp = 1;
 SYSCTL_INT(_hw_hn, OID_AUTO, trust_hosttcp, CTLFLAG_RDTUN,
     &hn_trust_hosttcp, 0,
     "Trust tcp segement verification on host side, "
     "when csum info is missing (global setting)");
 
 /* Trust udp datagrams verification on host side. */
 static int hn_trust_hostudp = 1;
 SYSCTL_INT(_hw_hn, OID_AUTO, trust_hostudp, CTLFLAG_RDTUN,
     &hn_trust_hostudp, 0,
     "Trust udp datagram verification on host side, "
     "when csum info is missing (global setting)");
 
 /* Trust ip packets verification on host side. */
 static int hn_trust_hostip = 1;
 SYSCTL_INT(_hw_hn, OID_AUTO, trust_hostip, CTLFLAG_RDTUN,
     &hn_trust_hostip, 0,
     "Trust ip packet verification on host side, "
     "when csum info is missing (global setting)");
 
 /* Limit TSO burst size */
 static int hn_tso_maxlen = IP_MAXPACKET;
 SYSCTL_INT(_hw_hn, OID_AUTO, tso_maxlen, CTLFLAG_RDTUN,
     &hn_tso_maxlen, 0, "TSO burst limit");
 
 /* Limit chimney send size */
 static int hn_tx_chimney_size = 0;
 SYSCTL_INT(_hw_hn, OID_AUTO, tx_chimney_size, CTLFLAG_RDTUN,
     &hn_tx_chimney_size, 0, "Chimney send packet size limit");
 
 /* Limit the size of packet for direct transmission */
 static int hn_direct_tx_size = HN_DIRECT_TX_SIZE_DEF;
 SYSCTL_INT(_hw_hn, OID_AUTO, direct_tx_size, CTLFLAG_RDTUN,
     &hn_direct_tx_size, 0, "Size of the packet for direct transmission");
 
 #if defined(INET) || defined(INET6)
 #if __FreeBSD_version >= 1100095
 static int hn_lro_entry_count = HN_LROENT_CNT_DEF;
 SYSCTL_INT(_hw_hn, OID_AUTO, lro_entry_count, CTLFLAG_RDTUN,
     &hn_lro_entry_count, 0, "LRO entry count");
 #endif
 #endif
 
 static int hn_share_tx_taskq = 0;
 SYSCTL_INT(_hw_hn, OID_AUTO, share_tx_taskq, CTLFLAG_RDTUN,
     &hn_share_tx_taskq, 0, "Enable shared TX taskqueue");
 
 static struct taskqueue	*hn_tx_taskq;
 
 #ifndef HN_USE_TXDESC_BUFRING
 static int hn_use_txdesc_bufring = 0;
 #else
 static int hn_use_txdesc_bufring = 1;
 #endif
 SYSCTL_INT(_hw_hn, OID_AUTO, use_txdesc_bufring, CTLFLAG_RD,
     &hn_use_txdesc_bufring, 0, "Use buf_ring for TX descriptors");
 
 static int hn_bind_tx_taskq = -1;
 SYSCTL_INT(_hw_hn, OID_AUTO, bind_tx_taskq, CTLFLAG_RDTUN,
     &hn_bind_tx_taskq, 0, "Bind TX taskqueue to the specified cpu");
 
 static int hn_use_if_start = 0;
 SYSCTL_INT(_hw_hn, OID_AUTO, use_if_start, CTLFLAG_RDTUN,
     &hn_use_if_start, 0, "Use if_start TX method");
 
 static int hn_chan_cnt = 0;
 SYSCTL_INT(_hw_hn, OID_AUTO, chan_cnt, CTLFLAG_RDTUN,
     &hn_chan_cnt, 0,
     "# of channels to use; each channel has one RX ring and one TX ring");
 
 static int hn_tx_ring_cnt = 0;
 SYSCTL_INT(_hw_hn, OID_AUTO, tx_ring_cnt, CTLFLAG_RDTUN,
     &hn_tx_ring_cnt, 0, "# of TX rings to use");
 
 static int hn_tx_swq_depth = 0;
 SYSCTL_INT(_hw_hn, OID_AUTO, tx_swq_depth, CTLFLAG_RDTUN,
     &hn_tx_swq_depth, 0, "Depth of IFQ or BUFRING");
 
 #if __FreeBSD_version >= 1100095
 static u_int hn_lro_mbufq_depth = 0;
 SYSCTL_UINT(_hw_hn, OID_AUTO, lro_mbufq_depth, CTLFLAG_RDTUN,
     &hn_lro_mbufq_depth, 0, "Depth of LRO mbuf queue");
 #endif
 
 static u_int hn_cpu_index;
 
 /*
  * Forward declarations
  */
 static void hn_stop(struct hn_softc *sc);
 static void hn_init_locked(struct hn_softc *sc);
 static void hn_init(void *xsc);
 static int  hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data);
 static int hn_start_locked(struct hn_tx_ring *txr, int len);
 static void hn_start(struct ifnet *ifp);
 static void hn_start_txeof(struct hn_tx_ring *);
 static int hn_ifmedia_upd(struct ifnet *ifp);
 static void hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr);
 #if __FreeBSD_version >= 1100099
 static int hn_lro_lenlim_sysctl(SYSCTL_HANDLER_ARGS);
 static int hn_lro_ackcnt_sysctl(SYSCTL_HANDLER_ARGS);
 #endif
 static int hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS);
 static int hn_chim_size_sysctl(SYSCTL_HANDLER_ARGS);
 #if __FreeBSD_version < 1100095
 static int hn_rx_stat_int_sysctl(SYSCTL_HANDLER_ARGS);
 #else
 static int hn_rx_stat_u64_sysctl(SYSCTL_HANDLER_ARGS);
 #endif
 static int hn_rx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS);
 static int hn_tx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS);
 static int hn_tx_conf_int_sysctl(SYSCTL_HANDLER_ARGS);
 static int hn_ndis_version_sysctl(SYSCTL_HANDLER_ARGS);
 static int hn_caps_sysctl(SYSCTL_HANDLER_ARGS);
 static int hn_hwassist_sysctl(SYSCTL_HANDLER_ARGS);
 static int hn_rxfilter_sysctl(SYSCTL_HANDLER_ARGS);
 static int hn_rss_key_sysctl(SYSCTL_HANDLER_ARGS);
 static int hn_rss_ind_sysctl(SYSCTL_HANDLER_ARGS);
 static int hn_rss_hash_sysctl(SYSCTL_HANDLER_ARGS);
 static int hn_check_iplen(const struct mbuf *, int);
 static int hn_create_tx_ring(struct hn_softc *, int);
 static void hn_destroy_tx_ring(struct hn_tx_ring *);
 static int hn_create_tx_data(struct hn_softc *, int);
 static void hn_fixup_tx_data(struct hn_softc *);
 static void hn_destroy_tx_data(struct hn_softc *);
 static void hn_start_taskfunc(void *, int);
 static void hn_start_txeof_taskfunc(void *, int);
 static void hn_link_taskfunc(void *, int);
 static void hn_netchg_init_taskfunc(void *, int);
 static void hn_netchg_status_taskfunc(void *, int);
 static void hn_suspend_mgmt_taskfunc(void *, int);
 static int hn_encap(struct hn_tx_ring *, struct hn_txdesc *, struct mbuf **);
 static int hn_create_rx_data(struct hn_softc *sc, int);
 static void hn_destroy_rx_data(struct hn_softc *sc);
 static void hn_set_chim_size(struct hn_softc *, int);
 static void hn_set_tso_maxsize(struct hn_softc *, int, int);
 static int hn_chan_attach(struct hn_softc *, struct vmbus_channel *);
 static void hn_chan_detach(struct hn_softc *, struct vmbus_channel *);
 static int hn_attach_subchans(struct hn_softc *);
 static void hn_detach_allchans(struct hn_softc *);
 static void hn_chan_callback(struct vmbus_channel *chan, void *xrxr);
 static void hn_set_ring_inuse(struct hn_softc *, int);
 static int hn_synth_attach(struct hn_softc *, int);
 static void hn_synth_detach(struct hn_softc *);
 static bool hn_tx_ring_pending(struct hn_tx_ring *);
 static void hn_suspend(struct hn_softc *);
 static void hn_suspend_data(struct hn_softc *);
 static void hn_suspend_mgmt(struct hn_softc *);
 static void hn_resume(struct hn_softc *);
 static void hn_resume_data(struct hn_softc *);
 static void hn_resume_mgmt(struct hn_softc *);
 static void hn_rx_drain(struct vmbus_channel *);
 static void hn_tx_resume(struct hn_softc *, int);
 static void hn_tx_ring_qflush(struct hn_tx_ring *);
 static int netvsc_detach(device_t dev);
 static void hn_link_status(struct hn_softc *);
 static int hn_sendpkt_rndis_sglist(struct hn_tx_ring *, struct hn_txdesc *);
 static int hn_sendpkt_rndis_chim(struct hn_tx_ring *, struct hn_txdesc *);
 static int hn_set_rxfilter(struct hn_softc *);
 
 static void hn_nvs_handle_notify(struct hn_softc *sc,
 		const struct vmbus_chanpkt_hdr *pkt);
 static void hn_nvs_handle_comp(struct hn_softc *sc, struct vmbus_channel *chan,
 		const struct vmbus_chanpkt_hdr *pkt);
 static void hn_nvs_handle_rxbuf(struct hn_softc *sc, struct hn_rx_ring *rxr,
 		struct vmbus_channel *chan,
 		const struct vmbus_chanpkt_hdr *pkthdr);
 static void hn_nvs_ack_rxbuf(struct vmbus_channel *chan, uint64_t tid);
 
 static int hn_transmit(struct ifnet *, struct mbuf *);
 static void hn_xmit_qflush(struct ifnet *);
 static int hn_xmit(struct hn_tx_ring *, int);
 static void hn_xmit_txeof(struct hn_tx_ring *);
 static void hn_xmit_taskfunc(void *, int);
 static void hn_xmit_txeof_taskfunc(void *, int);
 
 static const uint8_t	hn_rss_key_default[NDIS_HASH_KEYSIZE_TOEPLITZ] = {
 	0x6d, 0x5a, 0x56, 0xda, 0x25, 0x5b, 0x0e, 0xc2,
 	0x41, 0x67, 0x25, 0x3d, 0x43, 0xa3, 0x8f, 0xb0,
 	0xd0, 0xca, 0x2b, 0xcb, 0xae, 0x7b, 0x30, 0xb4,
 	0x77, 0xcb, 0x2d, 0xa3, 0x80, 0x30, 0xf2, 0x0c,
 	0x6a, 0x42, 0xb7, 0x3b, 0xbe, 0xac, 0x01, 0xfa
 };
 
 #if __FreeBSD_version >= 1100099
 static void
 hn_set_lro_lenlim(struct hn_softc *sc, int lenlim)
 {
 	int i;
 
 	for (i = 0; i < sc->hn_rx_ring_inuse; ++i)
 		sc->hn_rx_ring[i].hn_lro.lro_length_lim = lenlim;
 }
 #endif
 
-static __inline int
-hn_nvs_send_rndis_sglist1(struct vmbus_channel *chan, uint32_t rndis_mtype,
-    struct hn_send_ctx *sndc, struct vmbus_gpa *gpa, int gpa_cnt)
-{
-	struct hn_nvs_rndis rndis;
-
-	rndis.nvs_type = HN_NVS_TYPE_RNDIS;
-	rndis.nvs_rndis_mtype = rndis_mtype;
-	rndis.nvs_chim_idx = HN_NVS_CHIM_IDX_INVALID;
-	rndis.nvs_chim_sz = 0;
-
-	return (hn_nvs_send_sglist(chan, gpa, gpa_cnt,
-	    &rndis, sizeof(rndis), sndc));
-}
-
-int
-hn_nvs_send_rndis_ctrl(struct vmbus_channel *chan,
-    struct hn_send_ctx *sndc, struct vmbus_gpa *gpa, int gpa_cnt)
-{
-
-	return hn_nvs_send_rndis_sglist1(chan, HN_NVS_RNDIS_MTYPE_CTRL,
-	    sndc, gpa, gpa_cnt);
-}
-
 static int
 hn_sendpkt_rndis_sglist(struct hn_tx_ring *txr, struct hn_txdesc *txd)
 {
 
 	KASSERT(txd->chim_index == HN_NVS_CHIM_IDX_INVALID &&
 	    txd->chim_size == 0, ("invalid rndis sglist txd"));
-	return (hn_nvs_send_rndis_sglist1(txr->hn_chan, HN_NVS_RNDIS_MTYPE_DATA,
+	return (hn_nvs_send_rndis_sglist(txr->hn_chan, HN_NVS_RNDIS_MTYPE_DATA,
 	    &txd->send_ctx, txr->hn_gpa, txr->hn_gpa_cnt));
 }
 
 static int
 hn_sendpkt_rndis_chim(struct hn_tx_ring *txr, struct hn_txdesc *txd)
 {
 	struct hn_nvs_rndis rndis;
 
 	KASSERT(txd->chim_index != HN_NVS_CHIM_IDX_INVALID &&
 	    txd->chim_size > 0, ("invalid rndis chim txd"));
 
 	rndis.nvs_type = HN_NVS_TYPE_RNDIS;
 	rndis.nvs_rndis_mtype = HN_NVS_RNDIS_MTYPE_DATA;
 	rndis.nvs_chim_idx = txd->chim_index;
 	rndis.nvs_chim_sz = txd->chim_size;
 
 	return (hn_nvs_send(txr->hn_chan, VMBUS_CHANPKT_FLAG_RC,
 	    &rndis, sizeof(rndis), &txd->send_ctx));
 }
 
 static __inline uint32_t
 hn_chim_alloc(struct hn_softc *sc)
 {
 	int i, bmap_cnt = sc->hn_chim_bmap_cnt;
 	u_long *bmap = sc->hn_chim_bmap;
 	uint32_t ret = HN_NVS_CHIM_IDX_INVALID;
 
 	for (i = 0; i < bmap_cnt; ++i) {
 		int idx;
 
 		idx = ffsl(~bmap[i]);
 		if (idx == 0)
 			continue;
 
 		--idx; /* ffsl is 1-based */
 		KASSERT(i * LONG_BIT + idx < sc->hn_chim_cnt,
 		    ("invalid i %d and idx %d", i, idx));
 
 		if (atomic_testandset_long(&bmap[i], idx))
 			continue;
 
 		ret = i * LONG_BIT + idx;
 		break;
 	}
 	return (ret);
 }
 
 static __inline void
 hn_chim_free(struct hn_softc *sc, uint32_t chim_idx)
 {
 	u_long mask;
 	uint32_t idx;
 
 	idx = chim_idx / LONG_BIT;
 	KASSERT(idx < sc->hn_chim_bmap_cnt,
 	    ("invalid chimney index 0x%x", chim_idx));
 
 	mask = 1UL << (chim_idx % LONG_BIT);
 	KASSERT(sc->hn_chim_bmap[idx] & mask,
 	    ("index bitmap 0x%lx, chimney index %u, "
 	     "bitmap idx %d, bitmask 0x%lx",
 	     sc->hn_chim_bmap[idx], chim_idx, idx, mask));
 
 	atomic_clear_long(&sc->hn_chim_bmap[idx], mask);
 }
 
 static int
 hn_set_rxfilter(struct hn_softc *sc)
 {
 	struct ifnet *ifp = sc->hn_ifp;
 	uint32_t filter;
 	int error = 0;
 
 	HN_LOCK_ASSERT(sc);
 
 	if (ifp->if_flags & IFF_PROMISC) {
 		filter = NDIS_PACKET_TYPE_PROMISCUOUS;
 	} else {
 		filter = NDIS_PACKET_TYPE_DIRECTED;
 		if (ifp->if_flags & IFF_BROADCAST)
 			filter |= NDIS_PACKET_TYPE_BROADCAST;
 #ifdef notyet
 		/*
 		 * See the comment in SIOCADDMULTI/SIOCDELMULTI.
 		 */
 		/* TODO: support multicast list */
 		if ((ifp->if_flags & IFF_ALLMULTI) ||
 		    !TAILQ_EMPTY(&ifp->if_multiaddrs))
 			filter |= NDIS_PACKET_TYPE_ALL_MULTICAST;
 #else
 		/* Always enable ALLMULTI */
 		filter |= NDIS_PACKET_TYPE_ALL_MULTICAST;
 #endif
 	}
 
 	if (sc->hn_rx_filter != filter) {
 		error = hn_rndis_set_rxfilter(sc, filter);
 		if (!error)
 			sc->hn_rx_filter = filter;
 	}
 	return (error);
 }
 
 static int
 hn_get_txswq_depth(const struct hn_tx_ring *txr)
 {
 
 	KASSERT(txr->hn_txdesc_cnt > 0, ("tx ring is not setup yet"));
 	if (hn_tx_swq_depth < txr->hn_txdesc_cnt)
 		return txr->hn_txdesc_cnt;
 	return hn_tx_swq_depth;
 }
 
 static int
 hn_rss_reconfig(struct hn_softc *sc)
 {
 	int error;
 
 	HN_LOCK_ASSERT(sc);
 
 	if ((sc->hn_flags & HN_FLAG_SYNTH_ATTACHED) == 0)
 		return (ENXIO);
 
 	/*
 	 * Disable RSS first.
 	 *
 	 * NOTE:
 	 * Direct reconfiguration by setting the UNCHG flags does
 	 * _not_ work properly.
 	 */
 	if (bootverbose)
 		if_printf(sc->hn_ifp, "disable RSS\n");
 	error = hn_rndis_conf_rss(sc, NDIS_RSS_FLAG_DISABLE);
 	if (error) {
 		if_printf(sc->hn_ifp, "RSS disable failed\n");
 		return (error);
 	}
 
 	/*
 	 * Reenable the RSS w/ the updated RSS key or indirect
 	 * table.
 	 */
 	if (bootverbose)
 		if_printf(sc->hn_ifp, "reconfig RSS\n");
 	error = hn_rndis_conf_rss(sc, NDIS_RSS_FLAG_NONE);
 	if (error) {
 		if_printf(sc->hn_ifp, "RSS reconfig failed\n");
 		return (error);
 	}
 	return (0);
 }
 
 static void
 hn_rss_ind_fixup(struct hn_softc *sc, int nchan)
 {
 	struct ndis_rssprm_toeplitz *rss = &sc->hn_rss;
 	int i;
 
 	KASSERT(nchan > 1, ("invalid # of channels %d", nchan));
 
 	/*
 	 * Check indirect table to make sure that all channels in it
 	 * can be used.
 	 */
 	for (i = 0; i < NDIS_HASH_INDCNT; ++i) {
 		if (rss->rss_ind[i] >= nchan) {
 			if_printf(sc->hn_ifp,
 			    "RSS indirect table %d fixup: %u -> %d\n",
 			    i, rss->rss_ind[i], nchan - 1);
 			rss->rss_ind[i] = nchan - 1;
 		}
 	}
 }
 
 static int
 hn_ifmedia_upd(struct ifnet *ifp __unused)
 {
 
 	return EOPNOTSUPP;
 }
 
 static void
 hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
 {
 	struct hn_softc *sc = ifp->if_softc;
 
 	ifmr->ifm_status = IFM_AVALID;
 	ifmr->ifm_active = IFM_ETHER;
 
 	if ((sc->hn_link_flags & HN_LINK_FLAG_LINKUP) == 0) {
 		ifmr->ifm_active |= IFM_NONE;
 		return;
 	}
 	ifmr->ifm_status |= IFM_ACTIVE;
 	ifmr->ifm_active |= IFM_10G_T | IFM_FDX;
 }
 
 /* {F8615163-DF3E-46c5-913F-F2D2F965ED0E} */
 static const struct hyperv_guid g_net_vsc_device_type = {
 	.hv_guid = {0x63, 0x51, 0x61, 0xF8, 0x3E, 0xDF, 0xc5, 0x46,
 		0x91, 0x3F, 0xF2, 0xD2, 0xF9, 0x65, 0xED, 0x0E}
 };
 
 /*
  * Standard probe entry point.
  *
  */
 static int
 netvsc_probe(device_t dev)
 {
 	if (VMBUS_PROBE_GUID(device_get_parent(dev), dev,
 	    &g_net_vsc_device_type) == 0) {
 		device_set_desc(dev, "Hyper-V Network Interface");
 		return BUS_PROBE_DEFAULT;
 	}
 	return ENXIO;
 }
 
 /*
  * Standard attach entry point.
  *
  * Called when the driver is loaded.  It allocates needed resources,
  * and initializes the "hardware" and software.
  */
 static int
 netvsc_attach(device_t dev)
 {
 	struct hn_softc *sc = device_get_softc(dev);
 	struct sysctl_oid_list *child;
 	struct sysctl_ctx_list *ctx;
 	uint8_t eaddr[ETHER_ADDR_LEN];
 	struct ifnet *ifp = NULL;
 	int error, ring_cnt, tx_ring_cnt;
 
 	sc->hn_dev = dev;
 	sc->hn_prichan = vmbus_get_channel(dev);
 	HN_LOCK_INIT(sc);
 
 	/*
 	 * Setup taskqueue for transmission.
 	 */
 	if (hn_tx_taskq == NULL) {
 		sc->hn_tx_taskq = taskqueue_create("hn_tx", M_WAITOK,
 		    taskqueue_thread_enqueue, &sc->hn_tx_taskq);
 		if (hn_bind_tx_taskq >= 0) {
 			int cpu = hn_bind_tx_taskq;
 			cpuset_t cpu_set;
 
 			if (cpu > mp_ncpus - 1)
 				cpu = mp_ncpus - 1;
 			CPU_SETOF(cpu, &cpu_set);
 			taskqueue_start_threads_cpuset(&sc->hn_tx_taskq, 1,
 			    PI_NET, &cpu_set, "%s tx",
 			    device_get_nameunit(dev));
 		} else {
 			taskqueue_start_threads(&sc->hn_tx_taskq, 1, PI_NET,
 			    "%s tx", device_get_nameunit(dev));
 		}
 	} else {
 		sc->hn_tx_taskq = hn_tx_taskq;
 	}
 
 	/*
 	 * Setup taskqueue for mangement tasks, e.g. link status.
 	 */
 	sc->hn_mgmt_taskq0 = taskqueue_create("hn_mgmt", M_WAITOK,
 	    taskqueue_thread_enqueue, &sc->hn_mgmt_taskq0);
 	taskqueue_start_threads(&sc->hn_mgmt_taskq0, 1, PI_NET, "%s mgmt",
 	    device_get_nameunit(dev));
 	TASK_INIT(&sc->hn_link_task, 0, hn_link_taskfunc, sc);
 	TASK_INIT(&sc->hn_netchg_init, 0, hn_netchg_init_taskfunc, sc);
 	TIMEOUT_TASK_INIT(sc->hn_mgmt_taskq0, &sc->hn_netchg_status, 0,
 	    hn_netchg_status_taskfunc, sc);
 
 	/*
 	 * Allocate ifnet and setup its name earlier, so that if_printf
 	 * can be used by functions, which will be called after
 	 * ether_ifattach().
 	 */
 	ifp = sc->hn_ifp = if_alloc(IFT_ETHER);
 	ifp->if_softc = sc;
 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
 
 	/*
 	 * Initialize ifmedia earlier so that it can be unconditionally
 	 * destroyed, if error happened later on.
 	 */
 	ifmedia_init(&sc->hn_media, 0, hn_ifmedia_upd, hn_ifmedia_sts);
 
 	/*
 	 * Figure out the # of RX rings (ring_cnt) and the # of TX rings
 	 * to use (tx_ring_cnt).
 	 *
 	 * NOTE:
 	 * The # of RX rings to use is same as the # of channels to use.
 	 */
 	ring_cnt = hn_chan_cnt;
 	if (ring_cnt <= 0) {
 		/* Default */
 		ring_cnt = mp_ncpus;
 		if (ring_cnt > HN_RING_CNT_DEF_MAX)
 			ring_cnt = HN_RING_CNT_DEF_MAX;
 	} else if (ring_cnt > mp_ncpus) {
 		ring_cnt = mp_ncpus;
 	}
 
 	tx_ring_cnt = hn_tx_ring_cnt;
 	if (tx_ring_cnt <= 0 || tx_ring_cnt > ring_cnt)
 		tx_ring_cnt = ring_cnt;
 	if (hn_use_if_start) {
 		/* ifnet.if_start only needs one TX ring. */
 		tx_ring_cnt = 1;
 	}
 
 	/*
 	 * Set the leader CPU for channels.
 	 */
 	sc->hn_cpu = atomic_fetchadd_int(&hn_cpu_index, ring_cnt) % mp_ncpus;
 
 	/*
 	 * Create enough TX/RX rings, even if only limited number of
 	 * channels can be allocated.
 	 */
 	error = hn_create_tx_data(sc, tx_ring_cnt);
 	if (error)
 		goto failed;
 	error = hn_create_rx_data(sc, ring_cnt);
 	if (error)
 		goto failed;
 
 	/*
 	 * Create transaction context for NVS and RNDIS transactions.
 	 */
 	sc->hn_xact = vmbus_xact_ctx_create(bus_get_dma_tag(dev),
 	    HN_XACT_REQ_SIZE, HN_XACT_RESP_SIZE, 0);
 	if (sc->hn_xact == NULL)
 		goto failed;
 
 	/*
 	 * Attach the synthetic parts, i.e. NVS and RNDIS.
 	 */
 	error = hn_synth_attach(sc, ETHERMTU);
 	if (error)
 		goto failed;
 
 	error = hn_rndis_get_eaddr(sc, eaddr);
 	if (error)
 		goto failed;
 
 #if __FreeBSD_version >= 1100099
 	if (sc->hn_rx_ring_inuse > 1) {
 		/*
 		 * Reduce TCP segment aggregation limit for multiple
 		 * RX rings to increase ACK timeliness.
 		 */
 		hn_set_lro_lenlim(sc, HN_LRO_LENLIM_MULTIRX_DEF);
 	}
 #endif
 
 	/*
 	 * Fixup TX stuffs after synthetic parts are attached.
 	 */
 	hn_fixup_tx_data(sc);
 
 	ctx = device_get_sysctl_ctx(dev);
 	child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev));
 	SYSCTL_ADD_UINT(ctx, child, OID_AUTO, "nvs_version", CTLFLAG_RD,
 	    &sc->hn_nvs_ver, 0, "NVS version");
 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "ndis_version",
 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
 	    hn_ndis_version_sysctl, "A", "NDIS version");
 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "caps",
 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
 	    hn_caps_sysctl, "A", "capabilities");
 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "hwassist",
 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
 	    hn_hwassist_sysctl, "A", "hwassist");
 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "rxfilter",
 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
 	    hn_rxfilter_sysctl, "A", "rxfilter");
 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "rss_hash",
 	    CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, sc, 0,
 	    hn_rss_hash_sysctl, "A", "RSS hash");
 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "rss_ind_size",
 	    CTLFLAG_RD, &sc->hn_rss_ind_size, 0, "RSS indirect entry count");
 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "rss_key",
 	    CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
 	    hn_rss_key_sysctl, "IU", "RSS key");
 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "rss_ind",
 	    CTLTYPE_OPAQUE | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
 	    hn_rss_ind_sysctl, "IU", "RSS indirect table");
 
 	/*
 	 * Setup the ifmedia, which has been initialized earlier.
 	 */
 	ifmedia_add(&sc->hn_media, IFM_ETHER | IFM_AUTO, 0, NULL);
 	ifmedia_set(&sc->hn_media, IFM_ETHER | IFM_AUTO);
 	/* XXX ifmedia_set really should do this for us */
 	sc->hn_media.ifm_media = sc->hn_media.ifm_cur->ifm_media;
 
 	/*
 	 * Setup the ifnet for this interface.
 	 */
 
 	ifp->if_baudrate = IF_Gbps(10);
 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
 	ifp->if_ioctl = hn_ioctl;
 	ifp->if_init = hn_init;
 	if (hn_use_if_start) {
 		int qdepth = hn_get_txswq_depth(&sc->hn_tx_ring[0]);
 
 		ifp->if_start = hn_start;
 		IFQ_SET_MAXLEN(&ifp->if_snd, qdepth);
 		ifp->if_snd.ifq_drv_maxlen = qdepth - 1;
 		IFQ_SET_READY(&ifp->if_snd);
 	} else {
 		ifp->if_transmit = hn_transmit;
 		ifp->if_qflush = hn_xmit_qflush;
 	}
 
 	ifp->if_capabilities |= IFCAP_RXCSUM | IFCAP_LRO;
 #ifdef foo
 	/* We can't diff IPv6 packets from IPv4 packets on RX path. */
 	ifp->if_capabilities |= IFCAP_RXCSUM_IPV6;
 #endif
 	if (sc->hn_caps & HN_CAP_VLAN) {
 		/* XXX not sure about VLAN_MTU. */
 		ifp->if_capabilities |= IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU;
 	}
 
 	ifp->if_hwassist = sc->hn_tx_ring[0].hn_csum_assist;
 	if (ifp->if_hwassist & HN_CSUM_IP_MASK)
 		ifp->if_capabilities |= IFCAP_TXCSUM;
 	if (ifp->if_hwassist & HN_CSUM_IP6_MASK)
 		ifp->if_capabilities |= IFCAP_TXCSUM_IPV6;
 	if (sc->hn_caps & HN_CAP_TSO4) {
 		ifp->if_capabilities |= IFCAP_TSO4;
 		ifp->if_hwassist |= CSUM_IP_TSO;
 	}
 	if (sc->hn_caps & HN_CAP_TSO6) {
 		ifp->if_capabilities |= IFCAP_TSO6;
 		ifp->if_hwassist |= CSUM_IP6_TSO;
 	}
 
 	/* Enable all available capabilities by default. */
 	ifp->if_capenable = ifp->if_capabilities;
 
 	if (ifp->if_capabilities & (IFCAP_TSO6 | IFCAP_TSO4)) {
 		hn_set_tso_maxsize(sc, hn_tso_maxlen, ETHERMTU);
 		ifp->if_hw_tsomaxsegcount = HN_TX_DATA_SEGCNT_MAX;
 		ifp->if_hw_tsomaxsegsize = PAGE_SIZE;
 	}
 
 	ether_ifattach(ifp, eaddr);
 
 	if ((ifp->if_capabilities & (IFCAP_TSO6 | IFCAP_TSO4)) && bootverbose) {
 		if_printf(ifp, "TSO segcnt %u segsz %u\n",
 		    ifp->if_hw_tsomaxsegcount, ifp->if_hw_tsomaxsegsize);
 	}
 
 	/* Inform the upper layer about the long frame support. */
 	ifp->if_hdrlen = sizeof(struct ether_vlan_header);
 
 	/*
 	 * Kick off link status check.
 	 */
 	sc->hn_mgmt_taskq = sc->hn_mgmt_taskq0;
 	hn_link_status_update(sc);
 
 	return (0);
 failed:
 	if (sc->hn_flags & HN_FLAG_SYNTH_ATTACHED)
 		hn_synth_detach(sc);
 	netvsc_detach(dev);
 	return (error);
 }
 
 static int
 netvsc_detach(device_t dev)
 {
 	struct hn_softc *sc = device_get_softc(dev);
 	struct ifnet *ifp = sc->hn_ifp;
 
 	if (device_is_attached(dev)) {
 		HN_LOCK(sc);
 		if (sc->hn_flags & HN_FLAG_SYNTH_ATTACHED) {
 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
 				hn_stop(sc);
 			/*
 			 * NOTE:
 			 * hn_stop() only suspends data, so managment
 			 * stuffs have to be suspended manually here.
 			 */
 			hn_suspend_mgmt(sc);
 			hn_synth_detach(sc);
 		}
 		HN_UNLOCK(sc);
 		ether_ifdetach(ifp);
 	}
 
 	ifmedia_removeall(&sc->hn_media);
 	hn_destroy_rx_data(sc);
 	hn_destroy_tx_data(sc);
 
 	if (sc->hn_tx_taskq != hn_tx_taskq)
 		taskqueue_free(sc->hn_tx_taskq);
 	taskqueue_free(sc->hn_mgmt_taskq0);
 
 	if (sc->hn_xact != NULL)
 		vmbus_xact_ctx_destroy(sc->hn_xact);
 
 	if_free(ifp);
 
 	HN_LOCK_DESTROY(sc);
 	return (0);
 }
 
 /*
  * Standard shutdown entry point
  */
 static int
 netvsc_shutdown(device_t dev)
 {
 	return (0);
 }
 
 static void
 hn_link_status(struct hn_softc *sc)
 {
 	uint32_t link_status;
 	int error;
 
 	error = hn_rndis_get_linkstatus(sc, &link_status);
 	if (error) {
 		/* XXX what to do? */
 		return;
 	}
 
 	if (link_status == NDIS_MEDIA_STATE_CONNECTED)
 		sc->hn_link_flags |= HN_LINK_FLAG_LINKUP;
 	else
 		sc->hn_link_flags &= ~HN_LINK_FLAG_LINKUP;
 	if_link_state_change(sc->hn_ifp,
 	    (sc->hn_link_flags & HN_LINK_FLAG_LINKUP) ?
 	    LINK_STATE_UP : LINK_STATE_DOWN);
 }
 
 static void
 hn_link_taskfunc(void *xsc, int pending __unused)
 {
 	struct hn_softc *sc = xsc;
 
 	if (sc->hn_link_flags & HN_LINK_FLAG_NETCHG)
 		return;
 	hn_link_status(sc);
 }
 
 static void
 hn_netchg_init_taskfunc(void *xsc, int pending __unused)
 {
 	struct hn_softc *sc = xsc;
 
 	/* Prevent any link status checks from running. */
 	sc->hn_link_flags |= HN_LINK_FLAG_NETCHG;
 
 	/*
 	 * Fake up a [link down --> link up] state change; 5 seconds
 	 * delay is used, which closely simulates miibus reaction
 	 * upon link down event.
 	 */
 	sc->hn_link_flags &= ~HN_LINK_FLAG_LINKUP;
 	if_link_state_change(sc->hn_ifp, LINK_STATE_DOWN);
 	taskqueue_enqueue_timeout(sc->hn_mgmt_taskq0,
 	    &sc->hn_netchg_status, 5 * hz);
 }
 
 static void
 hn_netchg_status_taskfunc(void *xsc, int pending __unused)
 {
 	struct hn_softc *sc = xsc;
 
 	/* Re-allow link status checks. */
 	sc->hn_link_flags &= ~HN_LINK_FLAG_NETCHG;
 	hn_link_status(sc);
 }
 
 void
 hn_link_status_update(struct hn_softc *sc)
 {
 
 	if (sc->hn_mgmt_taskq != NULL)
 		taskqueue_enqueue(sc->hn_mgmt_taskq, &sc->hn_link_task);
 }
 
 void
 hn_network_change(struct hn_softc *sc)
 {
 
 	if (sc->hn_mgmt_taskq != NULL)
 		taskqueue_enqueue(sc->hn_mgmt_taskq, &sc->hn_netchg_init);
 }
 
 static __inline int
 hn_txdesc_dmamap_load(struct hn_tx_ring *txr, struct hn_txdesc *txd,
     struct mbuf **m_head, bus_dma_segment_t *segs, int *nsegs)
 {
 	struct mbuf *m = *m_head;
 	int error;
 
 	KASSERT(txd->chim_index == HN_NVS_CHIM_IDX_INVALID, ("txd uses chim"));
 
 	error = bus_dmamap_load_mbuf_sg(txr->hn_tx_data_dtag, txd->data_dmap,
 	    m, segs, nsegs, BUS_DMA_NOWAIT);
 	if (error == EFBIG) {
 		struct mbuf *m_new;
 
 		m_new = m_collapse(m, M_NOWAIT, HN_TX_DATA_SEGCNT_MAX);
 		if (m_new == NULL)
 			return ENOBUFS;
 		else
 			*m_head = m = m_new;
 		txr->hn_tx_collapsed++;
 
 		error = bus_dmamap_load_mbuf_sg(txr->hn_tx_data_dtag,
 		    txd->data_dmap, m, segs, nsegs, BUS_DMA_NOWAIT);
 	}
 	if (!error) {
 		bus_dmamap_sync(txr->hn_tx_data_dtag, txd->data_dmap,
 		    BUS_DMASYNC_PREWRITE);
 		txd->flags |= HN_TXD_FLAG_DMAMAP;
 	}
 	return error;
 }
 
 static __inline int
 hn_txdesc_put(struct hn_tx_ring *txr, struct hn_txdesc *txd)
 {
 
 	KASSERT((txd->flags & HN_TXD_FLAG_ONLIST) == 0,
 	    ("put an onlist txd %#x", txd->flags));
 
 	KASSERT(txd->refs > 0, ("invalid txd refs %d", txd->refs));
 	if (atomic_fetchadd_int(&txd->refs, -1) != 1)
 		return 0;
 
 	if (txd->chim_index != HN_NVS_CHIM_IDX_INVALID) {
 		KASSERT((txd->flags & HN_TXD_FLAG_DMAMAP) == 0,
 		    ("chim txd uses dmamap"));
 		hn_chim_free(txr->hn_sc, txd->chim_index);
 		txd->chim_index = HN_NVS_CHIM_IDX_INVALID;
 	} else if (txd->flags & HN_TXD_FLAG_DMAMAP) {
 		bus_dmamap_sync(txr->hn_tx_data_dtag,
 		    txd->data_dmap, BUS_DMASYNC_POSTWRITE);
 		bus_dmamap_unload(txr->hn_tx_data_dtag,
 		    txd->data_dmap);
 		txd->flags &= ~HN_TXD_FLAG_DMAMAP;
 	}
 
 	if (txd->m != NULL) {
 		m_freem(txd->m);
 		txd->m = NULL;
 	}
 
 	txd->flags |= HN_TXD_FLAG_ONLIST;
 #ifndef HN_USE_TXDESC_BUFRING
 	mtx_lock_spin(&txr->hn_txlist_spin);
 	KASSERT(txr->hn_txdesc_avail >= 0 &&
 	    txr->hn_txdesc_avail < txr->hn_txdesc_cnt,
 	    ("txdesc_put: invalid txd avail %d", txr->hn_txdesc_avail));
 	txr->hn_txdesc_avail++;
 	SLIST_INSERT_HEAD(&txr->hn_txlist, txd, link);
 	mtx_unlock_spin(&txr->hn_txlist_spin);
 #else
 	atomic_add_int(&txr->hn_txdesc_avail, 1);
 	buf_ring_enqueue(txr->hn_txdesc_br, txd);
 #endif
 
 	return 1;
 }
 
 static __inline struct hn_txdesc *
 hn_txdesc_get(struct hn_tx_ring *txr)
 {
 	struct hn_txdesc *txd;
 
 #ifndef HN_USE_TXDESC_BUFRING
 	mtx_lock_spin(&txr->hn_txlist_spin);
 	txd = SLIST_FIRST(&txr->hn_txlist);
 	if (txd != NULL) {
 		KASSERT(txr->hn_txdesc_avail > 0,
 		    ("txdesc_get: invalid txd avail %d", txr->hn_txdesc_avail));
 		txr->hn_txdesc_avail--;
 		SLIST_REMOVE_HEAD(&txr->hn_txlist, link);
 	}
 	mtx_unlock_spin(&txr->hn_txlist_spin);
 #else
 	txd = buf_ring_dequeue_sc(txr->hn_txdesc_br);
 #endif
 
 	if (txd != NULL) {
 #ifdef HN_USE_TXDESC_BUFRING
 		atomic_subtract_int(&txr->hn_txdesc_avail, 1);
 #endif
 		KASSERT(txd->m == NULL && txd->refs == 0 &&
 		    txd->chim_index == HN_NVS_CHIM_IDX_INVALID &&
 		    (txd->flags & HN_TXD_FLAG_ONLIST) &&
 		    (txd->flags & HN_TXD_FLAG_DMAMAP) == 0, ("invalid txd"));
 		txd->flags &= ~HN_TXD_FLAG_ONLIST;
 		txd->refs = 1;
 	}
 	return txd;
 }
 
 static __inline void
 hn_txdesc_hold(struct hn_txdesc *txd)
 {
 
 	/* 0->1 transition will never work */
 	KASSERT(txd->refs > 0, ("invalid refs %d", txd->refs));
 	atomic_add_int(&txd->refs, 1);
 }
 
 static bool
 hn_tx_ring_pending(struct hn_tx_ring *txr)
 {
 	bool pending = false;
 
 #ifndef HN_USE_TXDESC_BUFRING
 	mtx_lock_spin(&txr->hn_txlist_spin);
 	if (txr->hn_txdesc_avail != txr->hn_txdesc_cnt)
 		pending = true;
 	mtx_unlock_spin(&txr->hn_txlist_spin);
 #else
 	if (!buf_ring_full(txr->hn_txdesc_br))
 		pending = true;
 #endif
 	return (pending);
 }
 
 static __inline void
 hn_txeof(struct hn_tx_ring *txr)
 {
 	txr->hn_has_txeof = 0;
 	txr->hn_txeof(txr);
 }
 
 static void
-hn_tx_done(struct hn_send_ctx *sndc, struct hn_softc *sc,
+hn_tx_done(struct hn_nvs_sendctx *sndc, struct hn_softc *sc,
     struct vmbus_channel *chan, const void *data __unused, int dlen __unused)
 {
 	struct hn_txdesc *txd = sndc->hn_cbarg;
 	struct hn_tx_ring *txr;
 
 	txr = txd->txr;
 	KASSERT(txr->hn_chan == chan,
 	    ("channel mismatch, on chan%u, should be chan%u",
 	     vmbus_chan_subidx(chan), vmbus_chan_subidx(txr->hn_chan)));
 
 	txr->hn_has_txeof = 1;
 	hn_txdesc_put(txr, txd);
 
 	++txr->hn_txdone_cnt;
 	if (txr->hn_txdone_cnt >= HN_EARLY_TXEOF_THRESH) {
 		txr->hn_txdone_cnt = 0;
 		if (txr->hn_oactive)
 			hn_txeof(txr);
 	}
 }
 
 void
 hn_chan_rollup(struct hn_rx_ring *rxr, struct hn_tx_ring *txr)
 {
 #if defined(INET) || defined(INET6)
 	tcp_lro_flush_all(&rxr->hn_lro);
 #endif
 
 	/*
 	 * NOTE:
 	 * 'txr' could be NULL, if multiple channels and
 	 * ifnet.if_start method are enabled.
 	 */
 	if (txr == NULL || !txr->hn_has_txeof)
 		return;
 
 	txr->hn_txdone_cnt = 0;
 	hn_txeof(txr);
 }
 
 static __inline uint32_t
 hn_rndis_pktmsg_offset(uint32_t ofs)
 {
 
 	KASSERT(ofs >= sizeof(struct rndis_packet_msg),
 	    ("invalid RNDIS packet msg offset %u", ofs));
 	return (ofs - __offsetof(struct rndis_packet_msg, rm_dataoffset));
 }
 
 /*
  * NOTE:
  * If this function fails, then both txd and m_head0 will be freed.
  */
 static int
 hn_encap(struct hn_tx_ring *txr, struct hn_txdesc *txd, struct mbuf **m_head0)
 {
 	bus_dma_segment_t segs[HN_TX_DATA_SEGCNT_MAX];
 	int error, nsegs, i;
 	struct mbuf *m_head = *m_head0;
 	struct rndis_packet_msg *pkt;
 	uint32_t *pi_data;
 	int pktlen;
 
 	/*
 	 * extension points to the area reserved for the
 	 * rndis_filter_packet, which is placed just after
 	 * the netvsc_packet (and rppi struct, if present;
 	 * length is updated later).
 	 */
 	pkt = txd->rndis_pkt;
 	pkt->rm_type = REMOTE_NDIS_PACKET_MSG;
 	pkt->rm_len = sizeof(*pkt) + m_head->m_pkthdr.len;
 	pkt->rm_dataoffset = sizeof(*pkt);
 	pkt->rm_datalen = m_head->m_pkthdr.len;
 	pkt->rm_pktinfooffset = sizeof(*pkt);
 	pkt->rm_pktinfolen = 0;
 
 	if (txr->hn_tx_flags & HN_TX_FLAG_HASHVAL) {
 		/*
 		 * Set the hash value for this packet, so that the host could
 		 * dispatch the TX done event for this packet back to this TX
 		 * ring's channel.
 		 */
 		pi_data = hn_rndis_pktinfo_append(pkt, HN_RNDIS_PKT_LEN,
 		    HN_NDIS_HASH_VALUE_SIZE, HN_NDIS_PKTINFO_TYPE_HASHVAL);
 		*pi_data = txr->hn_tx_idx;
 	}
 
 	if (m_head->m_flags & M_VLANTAG) {
 		pi_data = hn_rndis_pktinfo_append(pkt, HN_RNDIS_PKT_LEN,
 		    NDIS_VLAN_INFO_SIZE, NDIS_PKTINFO_TYPE_VLAN);
 		*pi_data = NDIS_VLAN_INFO_MAKE(
 		    EVL_VLANOFTAG(m_head->m_pkthdr.ether_vtag),
 		    EVL_PRIOFTAG(m_head->m_pkthdr.ether_vtag),
 		    EVL_CFIOFTAG(m_head->m_pkthdr.ether_vtag));
 	}
 
 	if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
 #if defined(INET6) || defined(INET)
 		struct ether_vlan_header *eh;
 		int ether_len;
 
 		/*
 		 * XXX need m_pullup and use mtodo
 		 */
 		eh = mtod(m_head, struct ether_vlan_header*);
 		if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN))
 			ether_len = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
 		else
 			ether_len = ETHER_HDR_LEN;
 
 		pi_data = hn_rndis_pktinfo_append(pkt, HN_RNDIS_PKT_LEN,
 		    NDIS_LSO2_INFO_SIZE, NDIS_PKTINFO_TYPE_LSO);
 #ifdef INET
 		if (m_head->m_pkthdr.csum_flags & CSUM_IP_TSO) {
 			struct ip *ip =
 			    (struct ip *)(m_head->m_data + ether_len);
 			unsigned long iph_len = ip->ip_hl << 2;
 			struct tcphdr *th =
 			    (struct tcphdr *)((caddr_t)ip + iph_len);
 
 			ip->ip_len = 0;
 			ip->ip_sum = 0;
 			th->th_sum = in_pseudo(ip->ip_src.s_addr,
 			    ip->ip_dst.s_addr, htons(IPPROTO_TCP));
 			*pi_data = NDIS_LSO2_INFO_MAKEIPV4(0,
 			    m_head->m_pkthdr.tso_segsz);
 		}
 #endif
 #if defined(INET6) && defined(INET)
 		else
 #endif
 #ifdef INET6
 		{
 			struct ip6_hdr *ip6 = (struct ip6_hdr *)
 			    (m_head->m_data + ether_len);
 			struct tcphdr *th = (struct tcphdr *)(ip6 + 1);
 
 			ip6->ip6_plen = 0;
 			th->th_sum = in6_cksum_pseudo(ip6, 0, IPPROTO_TCP, 0);
 			*pi_data = NDIS_LSO2_INFO_MAKEIPV6(0,
 			    m_head->m_pkthdr.tso_segsz);
 		}
 #endif
 #endif	/* INET6 || INET */
 	} else if (m_head->m_pkthdr.csum_flags & txr->hn_csum_assist) {
 		pi_data = hn_rndis_pktinfo_append(pkt, HN_RNDIS_PKT_LEN,
 		    NDIS_TXCSUM_INFO_SIZE, NDIS_PKTINFO_TYPE_CSUM);
 		if (m_head->m_pkthdr.csum_flags &
 		    (CSUM_IP6_TCP | CSUM_IP6_UDP)) {
 			*pi_data = NDIS_TXCSUM_INFO_IPV6;
 		} else {
 			*pi_data = NDIS_TXCSUM_INFO_IPV4;
 			if (m_head->m_pkthdr.csum_flags & CSUM_IP)
 				*pi_data |= NDIS_TXCSUM_INFO_IPCS;
 		}
 
 		if (m_head->m_pkthdr.csum_flags & (CSUM_IP_TCP | CSUM_IP6_TCP))
 			*pi_data |= NDIS_TXCSUM_INFO_TCPCS;
 		else if (m_head->m_pkthdr.csum_flags &
 		    (CSUM_IP_UDP | CSUM_IP6_UDP))
 			*pi_data |= NDIS_TXCSUM_INFO_UDPCS;
 	}
 
 	pktlen = pkt->rm_pktinfooffset + pkt->rm_pktinfolen;
 	/* Convert RNDIS packet message offsets */
 	pkt->rm_dataoffset = hn_rndis_pktmsg_offset(pkt->rm_dataoffset);
 	pkt->rm_pktinfooffset = hn_rndis_pktmsg_offset(pkt->rm_pktinfooffset);
 
 	/*
 	 * Chimney send, if the packet could fit into one chimney buffer.
 	 */
 	if (pkt->rm_len < txr->hn_chim_size) {
 		txr->hn_tx_chimney_tried++;
 		txd->chim_index = hn_chim_alloc(txr->hn_sc);
 		if (txd->chim_index != HN_NVS_CHIM_IDX_INVALID) {
 			uint8_t *dest = txr->hn_sc->hn_chim +
 			    (txd->chim_index * txr->hn_sc->hn_chim_szmax);
 
 			memcpy(dest, pkt, pktlen);
 			dest += pktlen;
 			m_copydata(m_head, 0, m_head->m_pkthdr.len, dest);
 
 			txd->chim_size = pkt->rm_len;
 			txr->hn_gpa_cnt = 0;
 			txr->hn_tx_chimney++;
 			txr->hn_sendpkt = hn_sendpkt_rndis_chim;
 			goto done;
 		}
 	}
 
 	error = hn_txdesc_dmamap_load(txr, txd, &m_head, segs, &nsegs);
 	if (error) {
 		int freed;
 
 		/*
 		 * This mbuf is not linked w/ the txd yet, so free it now.
 		 */
 		m_freem(m_head);
 		*m_head0 = NULL;
 
 		freed = hn_txdesc_put(txr, txd);
 		KASSERT(freed != 0,
 		    ("fail to free txd upon txdma error"));
 
 		txr->hn_txdma_failed++;
 		if_inc_counter(txr->hn_sc->hn_ifp, IFCOUNTER_OERRORS, 1);
 		return error;
 	}
 	*m_head0 = m_head;
 
 	/* +1 RNDIS packet message */
 	txr->hn_gpa_cnt = nsegs + 1;
 
 	/* send packet with page buffer */
 	txr->hn_gpa[0].gpa_page = atop(txd->rndis_pkt_paddr);
 	txr->hn_gpa[0].gpa_ofs = txd->rndis_pkt_paddr & PAGE_MASK;
 	txr->hn_gpa[0].gpa_len = pktlen;
 
 	/*
 	 * Fill the page buffers with mbuf info after the page
 	 * buffer for RNDIS packet message.
 	 */
 	for (i = 0; i < nsegs; ++i) {
 		struct vmbus_gpa *gpa = &txr->hn_gpa[i + 1];
 
 		gpa->gpa_page = atop(segs[i].ds_addr);
 		gpa->gpa_ofs = segs[i].ds_addr & PAGE_MASK;
 		gpa->gpa_len = segs[i].ds_len;
 	}
 
 	txd->chim_index = HN_NVS_CHIM_IDX_INVALID;
 	txd->chim_size = 0;
 	txr->hn_sendpkt = hn_sendpkt_rndis_sglist;
 done:
 	txd->m = m_head;
 
 	/* Set the completion routine */
-	hn_send_ctx_init(&txd->send_ctx, hn_tx_done, txd);
+	hn_nvs_sendctx_init(&txd->send_ctx, hn_tx_done, txd);
 
 	return 0;
 }
 
 /*
  * NOTE:
  * If this function fails, then txd will be freed, but the mbuf
  * associated w/ the txd will _not_ be freed.
  */
 static int
 hn_send_pkt(struct ifnet *ifp, struct hn_tx_ring *txr, struct hn_txdesc *txd)
 {
 	int error, send_failed = 0;
 
 again:
 	/*
 	 * Make sure that txd is not freed before ETHER_BPF_MTAP.
 	 */
 	hn_txdesc_hold(txd);
 	error = txr->hn_sendpkt(txr, txd);
 	if (!error) {
 		ETHER_BPF_MTAP(ifp, txd->m);
 		if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
 		if (!hn_use_if_start) {
 			if_inc_counter(ifp, IFCOUNTER_OBYTES,
 			    txd->m->m_pkthdr.len);
 			if (txd->m->m_flags & M_MCAST)
 				if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
 		}
 		txr->hn_pkts++;
 	}
 	hn_txdesc_put(txr, txd);
 
 	if (__predict_false(error)) {
 		int freed;
 
 		/*
 		 * This should "really rarely" happen.
 		 *
 		 * XXX Too many RX to be acked or too many sideband
 		 * commands to run?  Ask netvsc_channel_rollup()
 		 * to kick start later.
 		 */
 		txr->hn_has_txeof = 1;
 		if (!send_failed) {
 			txr->hn_send_failed++;
 			send_failed = 1;
 			/*
 			 * Try sending again after set hn_has_txeof;
 			 * in case that we missed the last
 			 * netvsc_channel_rollup().
 			 */
 			goto again;
 		}
 		if_printf(ifp, "send failed\n");
 
 		/*
 		 * Caller will perform further processing on the
 		 * associated mbuf, so don't free it in hn_txdesc_put();
 		 * only unload it from the DMA map in hn_txdesc_put(),
 		 * if it was loaded.
 		 */
 		txd->m = NULL;
 		freed = hn_txdesc_put(txr, txd);
 		KASSERT(freed != 0,
 		    ("fail to free txd upon send error"));
 
 		txr->hn_send_failed++;
 	}
 	return error;
 }
 
 /*
  * Start a transmit of one or more packets
  */
 static int
 hn_start_locked(struct hn_tx_ring *txr, int len)
 {
 	struct hn_softc *sc = txr->hn_sc;
 	struct ifnet *ifp = sc->hn_ifp;
 
 	KASSERT(hn_use_if_start,
 	    ("hn_start_locked is called, when if_start is disabled"));
 	KASSERT(txr == &sc->hn_tx_ring[0], ("not the first TX ring"));
 	mtx_assert(&txr->hn_tx_lock, MA_OWNED);
 
 	if (__predict_false(txr->hn_suspended))
 		return 0;
 
 	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
 	    IFF_DRV_RUNNING)
 		return 0;
 
 	while (!IFQ_DRV_IS_EMPTY(&ifp->if_snd)) {
 		struct hn_txdesc *txd;
 		struct mbuf *m_head;
 		int error;
 
 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
 		if (m_head == NULL)
 			break;
 
 		if (len > 0 && m_head->m_pkthdr.len > len) {
 			/*
 			 * This sending could be time consuming; let callers
 			 * dispatch this packet sending (and sending of any
 			 * following up packets) to tx taskqueue.
 			 */
 			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
 			return 1;
 		}
 
 		txd = hn_txdesc_get(txr);
 		if (txd == NULL) {
 			txr->hn_no_txdescs++;
 			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
 			atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
 			break;
 		}
 
 		error = hn_encap(txr, txd, &m_head);
 		if (error) {
 			/* Both txd and m_head are freed */
 			continue;
 		}
 
 		error = hn_send_pkt(ifp, txr, txd);
 		if (__predict_false(error)) {
 			/* txd is freed, but m_head is not */
 			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
 			atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
 			break;
 		}
 	}
 	return 0;
 }
 
 /*
  * Append the specified data to the indicated mbuf chain,
  * Extend the mbuf chain if the new data does not fit in
  * existing space.
  *
  * This is a minor rewrite of m_append() from sys/kern/uipc_mbuf.c.
  * There should be an equivalent in the kernel mbuf code,
  * but there does not appear to be one yet.
  *
  * Differs from m_append() in that additional mbufs are
  * allocated with cluster size MJUMPAGESIZE, and filled
  * accordingly.
  *
  * Return 1 if able to complete the job; otherwise 0.
  */
 static int
 hv_m_append(struct mbuf *m0, int len, c_caddr_t cp)
 {
 	struct mbuf *m, *n;
 	int remainder, space;
 
 	for (m = m0; m->m_next != NULL; m = m->m_next)
 		;
 	remainder = len;
 	space = M_TRAILINGSPACE(m);
 	if (space > 0) {
 		/*
 		 * Copy into available space.
 		 */
 		if (space > remainder)
 			space = remainder;
 		bcopy(cp, mtod(m, caddr_t) + m->m_len, space);
 		m->m_len += space;
 		cp += space;
 		remainder -= space;
 	}
 	while (remainder > 0) {
 		/*
 		 * Allocate a new mbuf; could check space
 		 * and allocate a cluster instead.
 		 */
 		n = m_getjcl(M_NOWAIT, m->m_type, 0, MJUMPAGESIZE);
 		if (n == NULL)
 			break;
 		n->m_len = min(MJUMPAGESIZE, remainder);
 		bcopy(cp, mtod(n, caddr_t), n->m_len);
 		cp += n->m_len;
 		remainder -= n->m_len;
 		m->m_next = n;
 		m = n;
 	}
 	if (m0->m_flags & M_PKTHDR)
 		m0->m_pkthdr.len += len - remainder;
 
 	return (remainder == 0);
 }
 
 #if defined(INET) || defined(INET6)
 static __inline int
 hn_lro_rx(struct lro_ctrl *lc, struct mbuf *m)
 {
 #if __FreeBSD_version >= 1100095
 	if (hn_lro_mbufq_depth) {
 		tcp_lro_queue_mbuf(lc, m);
 		return 0;
 	}
 #endif
 	return tcp_lro_rx(lc, m, 0);
 }
 #endif
 
 /*
  * Called when we receive a data packet from the "wire" on the
  * specified device
  *
  * Note:  This is no longer used as a callback
  */
 int
 hn_rxpkt(struct hn_rx_ring *rxr, const void *data, int dlen,
     const struct hn_recvinfo *info)
 {
 	struct ifnet *ifp = rxr->hn_ifp;
 	struct mbuf *m_new;
 	int size, do_lro = 0, do_csum = 1;
 	int hash_type;
 
 	if (!(ifp->if_drv_flags & IFF_DRV_RUNNING))
 		return (0);
 
 	/*
 	 * Bail out if packet contains more data than configured MTU.
 	 */
 	if (dlen > (ifp->if_mtu + ETHER_HDR_LEN)) {
 		return (0);
 	} else if (dlen <= MHLEN) {
 		m_new = m_gethdr(M_NOWAIT, MT_DATA);
 		if (m_new == NULL) {
 			if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
 			return (0);
 		}
 		memcpy(mtod(m_new, void *), data, dlen);
 		m_new->m_pkthdr.len = m_new->m_len = dlen;
 		rxr->hn_small_pkts++;
 	} else {
 		/*
 		 * Get an mbuf with a cluster.  For packets 2K or less,
 		 * get a standard 2K cluster.  For anything larger, get a
 		 * 4K cluster.  Any buffers larger than 4K can cause problems
 		 * if looped around to the Hyper-V TX channel, so avoid them.
 		 */
 		size = MCLBYTES;
 		if (dlen > MCLBYTES) {
 			/* 4096 */
 			size = MJUMPAGESIZE;
 		}
 
 		m_new = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, size);
 		if (m_new == NULL) {
 			if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
 			return (0);
 		}
 
 		hv_m_append(m_new, dlen, data);
 	}
 	m_new->m_pkthdr.rcvif = ifp;
 
 	if (__predict_false((ifp->if_capenable & IFCAP_RXCSUM) == 0))
 		do_csum = 0;
 
 	/* receive side checksum offload */
 	if (info->csum_info != HN_NDIS_RXCSUM_INFO_INVALID) {
 		/* IP csum offload */
 		if ((info->csum_info & NDIS_RXCSUM_INFO_IPCS_OK) && do_csum) {
 			m_new->m_pkthdr.csum_flags |=
 			    (CSUM_IP_CHECKED | CSUM_IP_VALID);
 			rxr->hn_csum_ip++;
 		}
 
 		/* TCP/UDP csum offload */
 		if ((info->csum_info & (NDIS_RXCSUM_INFO_UDPCS_OK |
 		     NDIS_RXCSUM_INFO_TCPCS_OK)) && do_csum) {
 			m_new->m_pkthdr.csum_flags |=
 			    (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
 			m_new->m_pkthdr.csum_data = 0xffff;
 			if (info->csum_info & NDIS_RXCSUM_INFO_TCPCS_OK)
 				rxr->hn_csum_tcp++;
 			else
 				rxr->hn_csum_udp++;
 		}
 
 		if ((info->csum_info &
 		     (NDIS_RXCSUM_INFO_TCPCS_OK | NDIS_RXCSUM_INFO_IPCS_OK)) ==
 		    (NDIS_RXCSUM_INFO_TCPCS_OK | NDIS_RXCSUM_INFO_IPCS_OK))
 			do_lro = 1;
 	} else {
 		const struct ether_header *eh;
 		uint16_t etype;
 		int hoff;
 
 		hoff = sizeof(*eh);
 		if (m_new->m_len < hoff)
 			goto skip;
 		eh = mtod(m_new, struct ether_header *);
 		etype = ntohs(eh->ether_type);
 		if (etype == ETHERTYPE_VLAN) {
 			const struct ether_vlan_header *evl;
 
 			hoff = sizeof(*evl);
 			if (m_new->m_len < hoff)
 				goto skip;
 			evl = mtod(m_new, struct ether_vlan_header *);
 			etype = ntohs(evl->evl_proto);
 		}
 
 		if (etype == ETHERTYPE_IP) {
 			int pr;
 
 			pr = hn_check_iplen(m_new, hoff);
 			if (pr == IPPROTO_TCP) {
 				if (do_csum &&
 				    (rxr->hn_trust_hcsum &
 				     HN_TRUST_HCSUM_TCP)) {
 					rxr->hn_csum_trusted++;
 					m_new->m_pkthdr.csum_flags |=
 					   (CSUM_IP_CHECKED | CSUM_IP_VALID |
 					    CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
 					m_new->m_pkthdr.csum_data = 0xffff;
 				}
 				do_lro = 1;
 			} else if (pr == IPPROTO_UDP) {
 				if (do_csum &&
 				    (rxr->hn_trust_hcsum &
 				     HN_TRUST_HCSUM_UDP)) {
 					rxr->hn_csum_trusted++;
 					m_new->m_pkthdr.csum_flags |=
 					   (CSUM_IP_CHECKED | CSUM_IP_VALID |
 					    CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
 					m_new->m_pkthdr.csum_data = 0xffff;
 				}
 			} else if (pr != IPPROTO_DONE && do_csum &&
 			    (rxr->hn_trust_hcsum & HN_TRUST_HCSUM_IP)) {
 				rxr->hn_csum_trusted++;
 				m_new->m_pkthdr.csum_flags |=
 				    (CSUM_IP_CHECKED | CSUM_IP_VALID);
 			}
 		}
 	}
 skip:
 	if (info->vlan_info != HN_NDIS_VLAN_INFO_INVALID) {
 		m_new->m_pkthdr.ether_vtag = EVL_MAKETAG(
 		    NDIS_VLAN_INFO_ID(info->vlan_info),
 		    NDIS_VLAN_INFO_PRI(info->vlan_info),
 		    NDIS_VLAN_INFO_CFI(info->vlan_info));
 		m_new->m_flags |= M_VLANTAG;
 	}
 
 	if (info->hash_info != HN_NDIS_HASH_INFO_INVALID) {
 		rxr->hn_rss_pkts++;
 		m_new->m_pkthdr.flowid = info->hash_value;
 		hash_type = M_HASHTYPE_OPAQUE_HASH;
 		if ((info->hash_info & NDIS_HASH_FUNCTION_MASK) ==
 		    NDIS_HASH_FUNCTION_TOEPLITZ) {
 			uint32_t type = (info->hash_info & NDIS_HASH_TYPE_MASK);
 
 			switch (type) {
 			case NDIS_HASH_IPV4:
 				hash_type = M_HASHTYPE_RSS_IPV4;
 				break;
 
 			case NDIS_HASH_TCP_IPV4:
 				hash_type = M_HASHTYPE_RSS_TCP_IPV4;
 				break;
 
 			case NDIS_HASH_IPV6:
 				hash_type = M_HASHTYPE_RSS_IPV6;
 				break;
 
 			case NDIS_HASH_IPV6_EX:
 				hash_type = M_HASHTYPE_RSS_IPV6_EX;
 				break;
 
 			case NDIS_HASH_TCP_IPV6:
 				hash_type = M_HASHTYPE_RSS_TCP_IPV6;
 				break;
 
 			case NDIS_HASH_TCP_IPV6_EX:
 				hash_type = M_HASHTYPE_RSS_TCP_IPV6_EX;
 				break;
 			}
 		}
 	} else {
 		m_new->m_pkthdr.flowid = rxr->hn_rx_idx;
 		hash_type = M_HASHTYPE_OPAQUE;
 	}
 	M_HASHTYPE_SET(m_new, hash_type);
 
 	/*
 	 * Note:  Moved RX completion back to hv_nv_on_receive() so all
 	 * messages (not just data messages) will trigger a response.
 	 */
 
 	if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
 	rxr->hn_pkts++;
 
 	if ((ifp->if_capenable & IFCAP_LRO) && do_lro) {
 #if defined(INET) || defined(INET6)
 		struct lro_ctrl *lro = &rxr->hn_lro;
 
 		if (lro->lro_cnt) {
 			rxr->hn_lro_tried++;
 			if (hn_lro_rx(lro, m_new) == 0) {
 				/* DONE! */
 				return 0;
 			}
 		}
 #endif
 	}
 
 	/* We're not holding the lock here, so don't release it */
 	(*ifp->if_input)(ifp, m_new);
 
 	return (0);
 }
 
 static int
 hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
 {
 	struct hn_softc *sc = ifp->if_softc;
 	struct ifreq *ifr = (struct ifreq *)data;
 	int mask, error = 0;
 
 	switch (cmd) {
 	case SIOCSIFMTU:
 		if (ifr->ifr_mtu > HN_MTU_MAX) {
 			error = EINVAL;
 			break;
 		}
 
 		HN_LOCK(sc);
 
 		if ((sc->hn_flags & HN_FLAG_SYNTH_ATTACHED) == 0) {
 			HN_UNLOCK(sc);
 			break;
 		}
 
 		if ((sc->hn_caps & HN_CAP_MTU) == 0) {
 			/* Can't change MTU */
 			HN_UNLOCK(sc);
 			error = EOPNOTSUPP;
 			break;
 		}
 
 		if (ifp->if_mtu == ifr->ifr_mtu) {
 			HN_UNLOCK(sc);
 			break;
 		}
 
 		/*
 		 * Suspend this interface before the synthetic parts
 		 * are ripped.
 		 */
 		hn_suspend(sc);
 
 		/*
 		 * Detach the synthetics parts, i.e. NVS and RNDIS.
 		 */
 		hn_synth_detach(sc);
 
 		/*
 		 * Reattach the synthetic parts, i.e. NVS and RNDIS,
 		 * with the new MTU setting.
 		 */
 		error = hn_synth_attach(sc, ifr->ifr_mtu);
 		if (error) {
 			HN_UNLOCK(sc);
 			break;
 		}
 
 		/*
 		 * Commit the requested MTU, after the synthetic parts
 		 * have been successfully attached.
 		 */
 		ifp->if_mtu = ifr->ifr_mtu;
 
 		/*
 		 * Make sure that various parameters based on MTU are
 		 * still valid, after the MTU change.
 		 */
 		if (sc->hn_tx_ring[0].hn_chim_size > sc->hn_chim_szmax)
 			hn_set_chim_size(sc, sc->hn_chim_szmax);
 		hn_set_tso_maxsize(sc, hn_tso_maxlen, ifp->if_mtu);
 #if __FreeBSD_version >= 1100099
 		if (sc->hn_rx_ring[0].hn_lro.lro_length_lim <
 		    HN_LRO_LENLIM_MIN(ifp))
 			hn_set_lro_lenlim(sc, HN_LRO_LENLIM_MIN(ifp));
 #endif
 
 		/*
 		 * All done!  Resume the interface now.
 		 */
 		hn_resume(sc);
 
 		HN_UNLOCK(sc);
 		break;
 
 	case SIOCSIFFLAGS:
 		HN_LOCK(sc);
 
 		if ((sc->hn_flags & HN_FLAG_SYNTH_ATTACHED) == 0) {
 			HN_UNLOCK(sc);
 			break;
 		}
 
 		if (ifp->if_flags & IFF_UP) {
 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
 				hn_set_rxfilter(sc);
 			else
 				hn_init_locked(sc);
 		} else {
 			if (ifp->if_drv_flags & IFF_DRV_RUNNING)
 				hn_stop(sc);
 		}
 		sc->hn_if_flags = ifp->if_flags;
 
 		HN_UNLOCK(sc);
 		break;
 
 	case SIOCSIFCAP:
 		HN_LOCK(sc);
 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
 
 		if (mask & IFCAP_TXCSUM) {
 			ifp->if_capenable ^= IFCAP_TXCSUM;
 			if (ifp->if_capenable & IFCAP_TXCSUM)
 				ifp->if_hwassist |= HN_CSUM_IP_HWASSIST(sc);
 			else
 				ifp->if_hwassist &= ~HN_CSUM_IP_HWASSIST(sc);
 		}
 		if (mask & IFCAP_TXCSUM_IPV6) {
 			ifp->if_capenable ^= IFCAP_TXCSUM_IPV6;
 			if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
 				ifp->if_hwassist |= HN_CSUM_IP6_HWASSIST(sc);
 			else
 				ifp->if_hwassist &= ~HN_CSUM_IP6_HWASSIST(sc);
 		}
 
 		/* TODO: flip RNDIS offload parameters for RXCSUM. */
 		if (mask & IFCAP_RXCSUM)
 			ifp->if_capenable ^= IFCAP_RXCSUM;
 #ifdef foo
 		/* We can't diff IPv6 packets from IPv4 packets on RX path. */
 		if (mask & IFCAP_RXCSUM_IPV6)
 			ifp->if_capenable ^= IFCAP_RXCSUM_IPV6;
 #endif
 
 		if (mask & IFCAP_LRO)
 			ifp->if_capenable ^= IFCAP_LRO;
 
 		if (mask & IFCAP_TSO4) {
 			ifp->if_capenable ^= IFCAP_TSO4;
 			if (ifp->if_capenable & IFCAP_TSO4)
 				ifp->if_hwassist |= CSUM_IP_TSO;
 			else
 				ifp->if_hwassist &= ~CSUM_IP_TSO;
 		}
 		if (mask & IFCAP_TSO6) {
 			ifp->if_capenable ^= IFCAP_TSO6;
 			if (ifp->if_capenable & IFCAP_TSO6)
 				ifp->if_hwassist |= CSUM_IP6_TSO;
 			else
 				ifp->if_hwassist &= ~CSUM_IP6_TSO;
 		}
 
 		HN_UNLOCK(sc);
 		break;
 
 	case SIOCADDMULTI:
 	case SIOCDELMULTI:
 #ifdef notyet
 		/*
 		 * XXX
 		 * Multicast uses mutex, while RNDIS RX filter setting
 		 * sleeps.  We workaround this by always enabling
 		 * ALLMULTI.  ALLMULTI would actually always be on, even
 		 * if we supported the SIOCADDMULTI/SIOCDELMULTI, since
 		 * we don't support multicast address list configuration
 		 * for this driver.
 		 */
 		HN_LOCK(sc);
 
 		if ((sc->hn_flags & HN_FLAG_SYNTH_ATTACHED) == 0) {
 			HN_UNLOCK(sc);
 			break;
 		}
 		if (ifp->if_drv_flags & IFF_DRV_RUNNING)
 			hn_set_rxfilter(sc);
 
 		HN_UNLOCK(sc);
 #endif
 		break;
 
 	case SIOCSIFMEDIA:
 	case SIOCGIFMEDIA:
 		error = ifmedia_ioctl(ifp, ifr, &sc->hn_media, cmd);
 		break;
 
 	default:
 		error = ether_ioctl(ifp, cmd, data);
 		break;
 	}
 	return (error);
 }
 
 static void
 hn_stop(struct hn_softc *sc)
 {
 	struct ifnet *ifp = sc->hn_ifp;
 	int i;
 
 	HN_LOCK_ASSERT(sc);
 
 	KASSERT(sc->hn_flags & HN_FLAG_SYNTH_ATTACHED,
 	    ("synthetic parts were not attached"));
 
 	/* Clear RUNNING bit _before_ hn_suspend_data() */
 	atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_RUNNING);
 	hn_suspend_data(sc);
 
 	/* Clear OACTIVE bit. */
 	atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
 	for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
 		sc->hn_tx_ring[i].hn_oactive = 0;
 }
 
 /*
  * FreeBSD transmit entry point
  */
 static void
 hn_start(struct ifnet *ifp)
 {
 	struct hn_softc *sc = ifp->if_softc;
 	struct hn_tx_ring *txr = &sc->hn_tx_ring[0];
 
 	if (txr->hn_sched_tx)
 		goto do_sched;
 
 	if (mtx_trylock(&txr->hn_tx_lock)) {
 		int sched;
 
 		sched = hn_start_locked(txr, txr->hn_direct_tx_size);
 		mtx_unlock(&txr->hn_tx_lock);
 		if (!sched)
 			return;
 	}
 do_sched:
 	taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_tx_task);
 }
 
 static void
 hn_start_txeof(struct hn_tx_ring *txr)
 {
 	struct hn_softc *sc = txr->hn_sc;
 	struct ifnet *ifp = sc->hn_ifp;
 
 	KASSERT(txr == &sc->hn_tx_ring[0], ("not the first TX ring"));
 
 	if (txr->hn_sched_tx)
 		goto do_sched;
 
 	if (mtx_trylock(&txr->hn_tx_lock)) {
 		int sched;
 
 		atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
 		sched = hn_start_locked(txr, txr->hn_direct_tx_size);
 		mtx_unlock(&txr->hn_tx_lock);
 		if (sched) {
 			taskqueue_enqueue(txr->hn_tx_taskq,
 			    &txr->hn_tx_task);
 		}
 	} else {
 do_sched:
 		/*
 		 * Release the OACTIVE earlier, with the hope, that
 		 * others could catch up.  The task will clear the
 		 * flag again with the hn_tx_lock to avoid possible
 		 * races.
 		 */
 		atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
 		taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_txeof_task);
 	}
 }
 
 static void
 hn_init_locked(struct hn_softc *sc)
 {
 	struct ifnet *ifp = sc->hn_ifp;
 	int i;
 
 	HN_LOCK_ASSERT(sc);
 
 	if ((sc->hn_flags & HN_FLAG_SYNTH_ATTACHED) == 0)
 		return;
 
 	if (ifp->if_drv_flags & IFF_DRV_RUNNING)
 		return;
 
 	/* Configure RX filter */
 	hn_set_rxfilter(sc);
 
 	/* Clear OACTIVE bit. */
 	atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE);
 	for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
 		sc->hn_tx_ring[i].hn_oactive = 0;
 
 	/* Clear TX 'suspended' bit. */
 	hn_tx_resume(sc, sc->hn_tx_ring_inuse);
 
 	/* Everything is ready; unleash! */
 	atomic_set_int(&ifp->if_drv_flags, IFF_DRV_RUNNING);
 }
 
 static void
 hn_init(void *xsc)
 {
 	struct hn_softc *sc = xsc;
 
 	HN_LOCK(sc);
 	hn_init_locked(sc);
 	HN_UNLOCK(sc);
 }
 
 #ifdef LATER
 /*
  *
  */
 static void
 hn_watchdog(struct ifnet *ifp)
 {
 
 	if_printf(ifp, "watchdog timeout -- resetting\n");
 	hn_init(ifp->if_softc);    /* XXX */
 	if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
 }
 #endif
 
 #if __FreeBSD_version >= 1100099
 
 static int
 hn_lro_lenlim_sysctl(SYSCTL_HANDLER_ARGS)
 {
 	struct hn_softc *sc = arg1;
 	unsigned int lenlim;
 	int error;
 
 	lenlim = sc->hn_rx_ring[0].hn_lro.lro_length_lim;
 	error = sysctl_handle_int(oidp, &lenlim, 0, req);
 	if (error || req->newptr == NULL)
 		return error;
 
 	HN_LOCK(sc);
 	if (lenlim < HN_LRO_LENLIM_MIN(sc->hn_ifp) ||
 	    lenlim > TCP_LRO_LENGTH_MAX) {
 		HN_UNLOCK(sc);
 		return EINVAL;
 	}
 	hn_set_lro_lenlim(sc, lenlim);
 	HN_UNLOCK(sc);
 
 	return 0;
 }
 
 static int
 hn_lro_ackcnt_sysctl(SYSCTL_HANDLER_ARGS)
 {
 	struct hn_softc *sc = arg1;
 	int ackcnt, error, i;
 
 	/*
 	 * lro_ackcnt_lim is append count limit,
 	 * +1 to turn it into aggregation limit.
 	 */
 	ackcnt = sc->hn_rx_ring[0].hn_lro.lro_ackcnt_lim + 1;
 	error = sysctl_handle_int(oidp, &ackcnt, 0, req);
 	if (error || req->newptr == NULL)
 		return error;
 
 	if (ackcnt < 2 || ackcnt > (TCP_LRO_ACKCNT_MAX + 1))
 		return EINVAL;
 
 	/*
 	 * Convert aggregation limit back to append
 	 * count limit.
 	 */
 	--ackcnt;
 	HN_LOCK(sc);
 	for (i = 0; i < sc->hn_rx_ring_inuse; ++i)
 		sc->hn_rx_ring[i].hn_lro.lro_ackcnt_lim = ackcnt;
 	HN_UNLOCK(sc);
 	return 0;
 }
 
 #endif
 
 static int
 hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS)
 {
 	struct hn_softc *sc = arg1;
 	int hcsum = arg2;
 	int on, error, i;
 
 	on = 0;
 	if (sc->hn_rx_ring[0].hn_trust_hcsum & hcsum)
 		on = 1;
 
 	error = sysctl_handle_int(oidp, &on, 0, req);
 	if (error || req->newptr == NULL)
 		return error;
 
 	HN_LOCK(sc);
 	for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
 		struct hn_rx_ring *rxr = &sc->hn_rx_ring[i];
 
 		if (on)
 			rxr->hn_trust_hcsum |= hcsum;
 		else
 			rxr->hn_trust_hcsum &= ~hcsum;
 	}
 	HN_UNLOCK(sc);
 	return 0;
 }
 
 static int
 hn_chim_size_sysctl(SYSCTL_HANDLER_ARGS)
 {
 	struct hn_softc *sc = arg1;
 	int chim_size, error;
 
 	chim_size = sc->hn_tx_ring[0].hn_chim_size;
 	error = sysctl_handle_int(oidp, &chim_size, 0, req);
 	if (error || req->newptr == NULL)
 		return error;
 
 	if (chim_size > sc->hn_chim_szmax || chim_size <= 0)
 		return EINVAL;
 
 	HN_LOCK(sc);
 	hn_set_chim_size(sc, chim_size);
 	HN_UNLOCK(sc);
 	return 0;
 }
 
 #if __FreeBSD_version < 1100095
 static int
 hn_rx_stat_int_sysctl(SYSCTL_HANDLER_ARGS)
 {
 	struct hn_softc *sc = arg1;
 	int ofs = arg2, i, error;
 	struct hn_rx_ring *rxr;
 	uint64_t stat;
 
 	stat = 0;
 	for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
 		rxr = &sc->hn_rx_ring[i];
 		stat += *((int *)((uint8_t *)rxr + ofs));
 	}
 
 	error = sysctl_handle_64(oidp, &stat, 0, req);
 	if (error || req->newptr == NULL)
 		return error;
 
 	/* Zero out this stat. */
 	for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
 		rxr = &sc->hn_rx_ring[i];
 		*((int *)((uint8_t *)rxr + ofs)) = 0;
 	}
 	return 0;
 }
 #else
 static int
 hn_rx_stat_u64_sysctl(SYSCTL_HANDLER_ARGS)
 {
 	struct hn_softc *sc = arg1;
 	int ofs = arg2, i, error;
 	struct hn_rx_ring *rxr;
 	uint64_t stat;
 
 	stat = 0;
 	for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
 		rxr = &sc->hn_rx_ring[i];
 		stat += *((uint64_t *)((uint8_t *)rxr + ofs));
 	}
 
 	error = sysctl_handle_64(oidp, &stat, 0, req);
 	if (error || req->newptr == NULL)
 		return error;
 
 	/* Zero out this stat. */
 	for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
 		rxr = &sc->hn_rx_ring[i];
 		*((uint64_t *)((uint8_t *)rxr + ofs)) = 0;
 	}
 	return 0;
 }
 
 #endif
 
 static int
 hn_rx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS)
 {
 	struct hn_softc *sc = arg1;
 	int ofs = arg2, i, error;
 	struct hn_rx_ring *rxr;
 	u_long stat;
 
 	stat = 0;
 	for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
 		rxr = &sc->hn_rx_ring[i];
 		stat += *((u_long *)((uint8_t *)rxr + ofs));
 	}
 
 	error = sysctl_handle_long(oidp, &stat, 0, req);
 	if (error || req->newptr == NULL)
 		return error;
 
 	/* Zero out this stat. */
 	for (i = 0; i < sc->hn_rx_ring_inuse; ++i) {
 		rxr = &sc->hn_rx_ring[i];
 		*((u_long *)((uint8_t *)rxr + ofs)) = 0;
 	}
 	return 0;
 }
 
 static int
 hn_tx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS)
 {
 	struct hn_softc *sc = arg1;
 	int ofs = arg2, i, error;
 	struct hn_tx_ring *txr;
 	u_long stat;
 
 	stat = 0;
 	for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
 		txr = &sc->hn_tx_ring[i];
 		stat += *((u_long *)((uint8_t *)txr + ofs));
 	}
 
 	error = sysctl_handle_long(oidp, &stat, 0, req);
 	if (error || req->newptr == NULL)
 		return error;
 
 	/* Zero out this stat. */
 	for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
 		txr = &sc->hn_tx_ring[i];
 		*((u_long *)((uint8_t *)txr + ofs)) = 0;
 	}
 	return 0;
 }
 
 static int
 hn_tx_conf_int_sysctl(SYSCTL_HANDLER_ARGS)
 {
 	struct hn_softc *sc = arg1;
 	int ofs = arg2, i, error, conf;
 	struct hn_tx_ring *txr;
 
 	txr = &sc->hn_tx_ring[0];
 	conf = *((int *)((uint8_t *)txr + ofs));
 
 	error = sysctl_handle_int(oidp, &conf, 0, req);
 	if (error || req->newptr == NULL)
 		return error;
 
 	HN_LOCK(sc);
 	for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
 		txr = &sc->hn_tx_ring[i];
 		*((int *)((uint8_t *)txr + ofs)) = conf;
 	}
 	HN_UNLOCK(sc);
 
 	return 0;
 }
 
 static int
 hn_ndis_version_sysctl(SYSCTL_HANDLER_ARGS)
 {
 	struct hn_softc *sc = arg1;
 	char verstr[16];
 
 	snprintf(verstr, sizeof(verstr), "%u.%u",
 	    HN_NDIS_VERSION_MAJOR(sc->hn_ndis_ver),
 	    HN_NDIS_VERSION_MINOR(sc->hn_ndis_ver));
 	return sysctl_handle_string(oidp, verstr, sizeof(verstr), req);
 }
 
 static int
 hn_caps_sysctl(SYSCTL_HANDLER_ARGS)
 {
 	struct hn_softc *sc = arg1;
 	char caps_str[128];
 	uint32_t caps;
 
 	HN_LOCK(sc);
 	caps = sc->hn_caps;
 	HN_UNLOCK(sc);
 	snprintf(caps_str, sizeof(caps_str), "%b", caps, HN_CAP_BITS);
 	return sysctl_handle_string(oidp, caps_str, sizeof(caps_str), req);
 }
 
 static int
 hn_hwassist_sysctl(SYSCTL_HANDLER_ARGS)
 {
 	struct hn_softc *sc = arg1;
 	char assist_str[128];
 	uint32_t hwassist;
 
 	HN_LOCK(sc);
 	hwassist = sc->hn_ifp->if_hwassist;
 	HN_UNLOCK(sc);
 	snprintf(assist_str, sizeof(assist_str), "%b", hwassist, CSUM_BITS);
 	return sysctl_handle_string(oidp, assist_str, sizeof(assist_str), req);
 }
 
 static int
 hn_rxfilter_sysctl(SYSCTL_HANDLER_ARGS)
 {
 	struct hn_softc *sc = arg1;
 	char filter_str[128];
 	uint32_t filter;
 
 	HN_LOCK(sc);
 	filter = sc->hn_rx_filter;
 	HN_UNLOCK(sc);
 	snprintf(filter_str, sizeof(filter_str), "%b", filter,
 	    NDIS_PACKET_TYPES);
 	return sysctl_handle_string(oidp, filter_str, sizeof(filter_str), req);
 }
 
 static int
 hn_rss_key_sysctl(SYSCTL_HANDLER_ARGS)
 {
 	struct hn_softc *sc = arg1;
 	int error;
 
 	HN_LOCK(sc);
 
 	error = SYSCTL_OUT(req, sc->hn_rss.rss_key, sizeof(sc->hn_rss.rss_key));
 	if (error || req->newptr == NULL)
 		goto back;
 
 	error = SYSCTL_IN(req, sc->hn_rss.rss_key, sizeof(sc->hn_rss.rss_key));
 	if (error)
 		goto back;
 	sc->hn_flags |= HN_FLAG_HAS_RSSKEY;
 
 	if (sc->hn_rx_ring_inuse > 1) {
 		error = hn_rss_reconfig(sc);
 	} else {
 		/* Not RSS capable, at least for now; just save the RSS key. */
 		error = 0;
 	}
 back:
 	HN_UNLOCK(sc);
 	return (error);
 }
 
 static int
 hn_rss_ind_sysctl(SYSCTL_HANDLER_ARGS)
 {
 	struct hn_softc *sc = arg1;
 	int error;
 
 	HN_LOCK(sc);
 
 	error = SYSCTL_OUT(req, sc->hn_rss.rss_ind, sizeof(sc->hn_rss.rss_ind));
 	if (error || req->newptr == NULL)
 		goto back;
 
 	/*
 	 * Don't allow RSS indirect table change, if this interface is not
 	 * RSS capable currently.
 	 */
 	if (sc->hn_rx_ring_inuse == 1) {
 		error = EOPNOTSUPP;
 		goto back;
 	}
 
 	error = SYSCTL_IN(req, sc->hn_rss.rss_ind, sizeof(sc->hn_rss.rss_ind));
 	if (error)
 		goto back;
 	sc->hn_flags |= HN_FLAG_HAS_RSSIND;
 
 	hn_rss_ind_fixup(sc, sc->hn_rx_ring_inuse);
 	error = hn_rss_reconfig(sc);
 back:
 	HN_UNLOCK(sc);
 	return (error);
 }
 
 static int
 hn_rss_hash_sysctl(SYSCTL_HANDLER_ARGS)
 {
 	struct hn_softc *sc = arg1;
 	char hash_str[128];
 	uint32_t hash;
 
 	HN_LOCK(sc);
 	hash = sc->hn_rss_hash;
 	HN_UNLOCK(sc);
 	snprintf(hash_str, sizeof(hash_str), "%b", hash, NDIS_HASH_BITS);
 	return sysctl_handle_string(oidp, hash_str, sizeof(hash_str), req);
 }
 
 static int
 hn_check_iplen(const struct mbuf *m, int hoff)
 {
 	const struct ip *ip;
 	int len, iphlen, iplen;
 	const struct tcphdr *th;
 	int thoff;				/* TCP data offset */
 
 	len = hoff + sizeof(struct ip);
 
 	/* The packet must be at least the size of an IP header. */
 	if (m->m_pkthdr.len < len)
 		return IPPROTO_DONE;
 
 	/* The fixed IP header must reside completely in the first mbuf. */
 	if (m->m_len < len)
 		return IPPROTO_DONE;
 
 	ip = mtodo(m, hoff);
 
 	/* Bound check the packet's stated IP header length. */
 	iphlen = ip->ip_hl << 2;
 	if (iphlen < sizeof(struct ip))		/* minimum header length */
 		return IPPROTO_DONE;
 
 	/* The full IP header must reside completely in the one mbuf. */
 	if (m->m_len < hoff + iphlen)
 		return IPPROTO_DONE;
 
 	iplen = ntohs(ip->ip_len);
 
 	/*
 	 * Check that the amount of data in the buffers is as
 	 * at least much as the IP header would have us expect.
 	 */
 	if (m->m_pkthdr.len < hoff + iplen)
 		return IPPROTO_DONE;
 
 	/*
 	 * Ignore IP fragments.
 	 */
 	if (ntohs(ip->ip_off) & (IP_OFFMASK | IP_MF))
 		return IPPROTO_DONE;
 
 	/*
 	 * The TCP/IP or UDP/IP header must be entirely contained within
 	 * the first fragment of a packet.
 	 */
 	switch (ip->ip_p) {
 	case IPPROTO_TCP:
 		if (iplen < iphlen + sizeof(struct tcphdr))
 			return IPPROTO_DONE;
 		if (m->m_len < hoff + iphlen + sizeof(struct tcphdr))
 			return IPPROTO_DONE;
 		th = (const struct tcphdr *)((const uint8_t *)ip + iphlen);
 		thoff = th->th_off << 2;
 		if (thoff < sizeof(struct tcphdr) || thoff + iphlen > iplen)
 			return IPPROTO_DONE;
 		if (m->m_len < hoff + iphlen + thoff)
 			return IPPROTO_DONE;
 		break;
 	case IPPROTO_UDP:
 		if (iplen < iphlen + sizeof(struct udphdr))
 			return IPPROTO_DONE;
 		if (m->m_len < hoff + iphlen + sizeof(struct udphdr))
 			return IPPROTO_DONE;
 		break;
 	default:
 		if (iplen < iphlen)
 			return IPPROTO_DONE;
 		break;
 	}
 	return ip->ip_p;
 }
 
 static int
 hn_create_rx_data(struct hn_softc *sc, int ring_cnt)
 {
 	struct sysctl_oid_list *child;
 	struct sysctl_ctx_list *ctx;
 	device_t dev = sc->hn_dev;
 #if defined(INET) || defined(INET6)
 #if __FreeBSD_version >= 1100095
 	int lroent_cnt;
 #endif
 #endif
 	int i;
 
 	/*
 	 * Create RXBUF for reception.
 	 *
 	 * NOTE:
 	 * - It is shared by all channels.
 	 * - A large enough buffer is allocated, certain version of NVSes
 	 *   may further limit the usable space.
 	 */
 	sc->hn_rxbuf = hyperv_dmamem_alloc(bus_get_dma_tag(dev),
 	    PAGE_SIZE, 0, HN_RXBUF_SIZE, &sc->hn_rxbuf_dma,
 	    BUS_DMA_WAITOK | BUS_DMA_ZERO);
 	if (sc->hn_rxbuf == NULL) {
 		device_printf(sc->hn_dev, "allocate rxbuf failed\n");
 		return (ENOMEM);
 	}
 
 	sc->hn_rx_ring_cnt = ring_cnt;
 	sc->hn_rx_ring_inuse = sc->hn_rx_ring_cnt;
 
 	sc->hn_rx_ring = malloc(sizeof(struct hn_rx_ring) * sc->hn_rx_ring_cnt,
 	    M_DEVBUF, M_WAITOK | M_ZERO);
 
 #if defined(INET) || defined(INET6)
 #if __FreeBSD_version >= 1100095
 	lroent_cnt = hn_lro_entry_count;
 	if (lroent_cnt < TCP_LRO_ENTRIES)
 		lroent_cnt = TCP_LRO_ENTRIES;
 	if (bootverbose)
 		device_printf(dev, "LRO: entry count %d\n", lroent_cnt);
 #endif
 #endif	/* INET || INET6 */
 
 	ctx = device_get_sysctl_ctx(dev);
 	child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev));
 
 	/* Create dev.hn.UNIT.rx sysctl tree */
 	sc->hn_rx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "rx",
 	    CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
 
 	for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
 		struct hn_rx_ring *rxr = &sc->hn_rx_ring[i];
 
 		rxr->hn_br = hyperv_dmamem_alloc(bus_get_dma_tag(dev),
 		    PAGE_SIZE, 0, HN_TXBR_SIZE + HN_RXBR_SIZE,
 		    &rxr->hn_br_dma, BUS_DMA_WAITOK);
 		if (rxr->hn_br == NULL) {
 			device_printf(dev, "allocate bufring failed\n");
 			return (ENOMEM);
 		}
 
 		if (hn_trust_hosttcp)
 			rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_TCP;
 		if (hn_trust_hostudp)
 			rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_UDP;
 		if (hn_trust_hostip)
 			rxr->hn_trust_hcsum |= HN_TRUST_HCSUM_IP;
 		rxr->hn_ifp = sc->hn_ifp;
 		if (i < sc->hn_tx_ring_cnt)
 			rxr->hn_txr = &sc->hn_tx_ring[i];
 		rxr->hn_pktbuf = malloc(HN_PKTBUF_LEN, M_DEVBUF, M_WAITOK);
 		rxr->hn_rx_idx = i;
 		rxr->hn_rxbuf = sc->hn_rxbuf;
 
 		/*
 		 * Initialize LRO.
 		 */
 #if defined(INET) || defined(INET6)
 #if __FreeBSD_version >= 1100095
 		tcp_lro_init_args(&rxr->hn_lro, sc->hn_ifp, lroent_cnt,
 		    hn_lro_mbufq_depth);
 #else
 		tcp_lro_init(&rxr->hn_lro);
 		rxr->hn_lro.ifp = sc->hn_ifp;
 #endif
 #if __FreeBSD_version >= 1100099
 		rxr->hn_lro.lro_length_lim = HN_LRO_LENLIM_DEF;
 		rxr->hn_lro.lro_ackcnt_lim = HN_LRO_ACKCNT_DEF;
 #endif
 #endif	/* INET || INET6 */
 
 		if (sc->hn_rx_sysctl_tree != NULL) {
 			char name[16];
 
 			/*
 			 * Create per RX ring sysctl tree:
 			 * dev.hn.UNIT.rx.RINGID
 			 */
 			snprintf(name, sizeof(name), "%d", i);
 			rxr->hn_rx_sysctl_tree = SYSCTL_ADD_NODE(ctx,
 			    SYSCTL_CHILDREN(sc->hn_rx_sysctl_tree),
 			    OID_AUTO, name, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
 
 			if (rxr->hn_rx_sysctl_tree != NULL) {
 				SYSCTL_ADD_ULONG(ctx,
 				    SYSCTL_CHILDREN(rxr->hn_rx_sysctl_tree),
 				    OID_AUTO, "packets", CTLFLAG_RW,
 				    &rxr->hn_pkts, "# of packets received");
 				SYSCTL_ADD_ULONG(ctx,
 				    SYSCTL_CHILDREN(rxr->hn_rx_sysctl_tree),
 				    OID_AUTO, "rss_pkts", CTLFLAG_RW,
 				    &rxr->hn_rss_pkts,
 				    "# of packets w/ RSS info received");
 			}
 		}
 	}
 
 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_queued",
 	    CTLTYPE_U64 | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
 	    __offsetof(struct hn_rx_ring, hn_lro.lro_queued),
 #if __FreeBSD_version < 1100095
 	    hn_rx_stat_int_sysctl,
 #else
 	    hn_rx_stat_u64_sysctl,
 #endif
 	    "LU", "LRO queued");
 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_flushed",
 	    CTLTYPE_U64 | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
 	    __offsetof(struct hn_rx_ring, hn_lro.lro_flushed),
 #if __FreeBSD_version < 1100095
 	    hn_rx_stat_int_sysctl,
 #else
 	    hn_rx_stat_u64_sysctl,
 #endif
 	    "LU", "LRO flushed");
 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_tried",
 	    CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
 	    __offsetof(struct hn_rx_ring, hn_lro_tried),
 	    hn_rx_stat_ulong_sysctl, "LU", "# of LRO tries");
 #if __FreeBSD_version >= 1100099
 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_length_lim",
 	    CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
 	    hn_lro_lenlim_sysctl, "IU",
 	    "Max # of data bytes to be aggregated by LRO");
 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_ackcnt_lim",
 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
 	    hn_lro_ackcnt_sysctl, "I",
 	    "Max # of ACKs to be aggregated by LRO");
 #endif
 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hosttcp",
 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, HN_TRUST_HCSUM_TCP,
 	    hn_trust_hcsum_sysctl, "I",
 	    "Trust tcp segement verification on host side, "
 	    "when csum info is missing");
 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostudp",
 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, HN_TRUST_HCSUM_UDP,
 	    hn_trust_hcsum_sysctl, "I",
 	    "Trust udp datagram verification on host side, "
 	    "when csum info is missing");
 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostip",
 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, HN_TRUST_HCSUM_IP,
 	    hn_trust_hcsum_sysctl, "I",
 	    "Trust ip packet verification on host side, "
 	    "when csum info is missing");
 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_ip",
 	    CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
 	    __offsetof(struct hn_rx_ring, hn_csum_ip),
 	    hn_rx_stat_ulong_sysctl, "LU", "RXCSUM IP");
 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_tcp",
 	    CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
 	    __offsetof(struct hn_rx_ring, hn_csum_tcp),
 	    hn_rx_stat_ulong_sysctl, "LU", "RXCSUM TCP");
 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_udp",
 	    CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
 	    __offsetof(struct hn_rx_ring, hn_csum_udp),
 	    hn_rx_stat_ulong_sysctl, "LU", "RXCSUM UDP");
 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "csum_trusted",
 	    CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
 	    __offsetof(struct hn_rx_ring, hn_csum_trusted),
 	    hn_rx_stat_ulong_sysctl, "LU",
 	    "# of packets that we trust host's csum verification");
 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "small_pkts",
 	    CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
 	    __offsetof(struct hn_rx_ring, hn_small_pkts),
 	    hn_rx_stat_ulong_sysctl, "LU", "# of small packets received");
 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "rx_ring_cnt",
 	    CTLFLAG_RD, &sc->hn_rx_ring_cnt, 0, "# created RX rings");
 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "rx_ring_inuse",
 	    CTLFLAG_RD, &sc->hn_rx_ring_inuse, 0, "# used RX rings");
 
 	return (0);
 }
 
 static void
 hn_destroy_rx_data(struct hn_softc *sc)
 {
 	int i;
 
 	if (sc->hn_rxbuf != NULL) {
 		hyperv_dmamem_free(&sc->hn_rxbuf_dma, sc->hn_rxbuf);
 		sc->hn_rxbuf = NULL;
 	}
 
 	if (sc->hn_rx_ring_cnt == 0)
 		return;
 
 	for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
 		struct hn_rx_ring *rxr = &sc->hn_rx_ring[i];
 
 		if (rxr->hn_br == NULL)
 			continue;
 		hyperv_dmamem_free(&rxr->hn_br_dma, rxr->hn_br);
 		rxr->hn_br = NULL;
 
 #if defined(INET) || defined(INET6)
 		tcp_lro_free(&rxr->hn_lro);
 #endif
 		free(rxr->hn_pktbuf, M_DEVBUF);
 	}
 	free(sc->hn_rx_ring, M_DEVBUF);
 	sc->hn_rx_ring = NULL;
 
 	sc->hn_rx_ring_cnt = 0;
 	sc->hn_rx_ring_inuse = 0;
 }
 
 static int
 hn_create_tx_ring(struct hn_softc *sc, int id)
 {
 	struct hn_tx_ring *txr = &sc->hn_tx_ring[id];
 	device_t dev = sc->hn_dev;
 	bus_dma_tag_t parent_dtag;
 	int error, i;
 
 	txr->hn_sc = sc;
 	txr->hn_tx_idx = id;
 
 #ifndef HN_USE_TXDESC_BUFRING
 	mtx_init(&txr->hn_txlist_spin, "hn txlist", NULL, MTX_SPIN);
 #endif
 	mtx_init(&txr->hn_tx_lock, "hn tx", NULL, MTX_DEF);
 
 	txr->hn_txdesc_cnt = HN_TX_DESC_CNT;
 	txr->hn_txdesc = malloc(sizeof(struct hn_txdesc) * txr->hn_txdesc_cnt,
 	    M_DEVBUF, M_WAITOK | M_ZERO);
 #ifndef HN_USE_TXDESC_BUFRING
 	SLIST_INIT(&txr->hn_txlist);
 #else
 	txr->hn_txdesc_br = buf_ring_alloc(txr->hn_txdesc_cnt, M_DEVBUF,
 	    M_WAITOK, &txr->hn_tx_lock);
 #endif
 
 	txr->hn_tx_taskq = sc->hn_tx_taskq;
 
 	if (hn_use_if_start) {
 		txr->hn_txeof = hn_start_txeof;
 		TASK_INIT(&txr->hn_tx_task, 0, hn_start_taskfunc, txr);
 		TASK_INIT(&txr->hn_txeof_task, 0, hn_start_txeof_taskfunc, txr);
 	} else {
 		int br_depth;
 
 		txr->hn_txeof = hn_xmit_txeof;
 		TASK_INIT(&txr->hn_tx_task, 0, hn_xmit_taskfunc, txr);
 		TASK_INIT(&txr->hn_txeof_task, 0, hn_xmit_txeof_taskfunc, txr);
 
 		br_depth = hn_get_txswq_depth(txr);
 		txr->hn_mbuf_br = buf_ring_alloc(br_depth, M_DEVBUF,
 		    M_WAITOK, &txr->hn_tx_lock);
 	}
 
 	txr->hn_direct_tx_size = hn_direct_tx_size;
 
 	/*
 	 * Always schedule transmission instead of trying to do direct
 	 * transmission.  This one gives the best performance so far.
 	 */
 	txr->hn_sched_tx = 1;
 
 	parent_dtag = bus_get_dma_tag(dev);
 
 	/* DMA tag for RNDIS packet messages. */
 	error = bus_dma_tag_create(parent_dtag, /* parent */
 	    HN_RNDIS_PKT_ALIGN,		/* alignment */
 	    HN_RNDIS_PKT_BOUNDARY,	/* boundary */
 	    BUS_SPACE_MAXADDR,		/* lowaddr */
 	    BUS_SPACE_MAXADDR,		/* highaddr */
 	    NULL, NULL,			/* filter, filterarg */
 	    HN_RNDIS_PKT_LEN,		/* maxsize */
 	    1,				/* nsegments */
 	    HN_RNDIS_PKT_LEN,		/* maxsegsize */
 	    0,				/* flags */
 	    NULL,			/* lockfunc */
 	    NULL,			/* lockfuncarg */
 	    &txr->hn_tx_rndis_dtag);
 	if (error) {
 		device_printf(dev, "failed to create rndis dmatag\n");
 		return error;
 	}
 
 	/* DMA tag for data. */
 	error = bus_dma_tag_create(parent_dtag, /* parent */
 	    1,				/* alignment */
 	    HN_TX_DATA_BOUNDARY,	/* boundary */
 	    BUS_SPACE_MAXADDR,		/* lowaddr */
 	    BUS_SPACE_MAXADDR,		/* highaddr */
 	    NULL, NULL,			/* filter, filterarg */
 	    HN_TX_DATA_MAXSIZE,		/* maxsize */
 	    HN_TX_DATA_SEGCNT_MAX,	/* nsegments */
 	    HN_TX_DATA_SEGSIZE,		/* maxsegsize */
 	    0,				/* flags */
 	    NULL,			/* lockfunc */
 	    NULL,			/* lockfuncarg */
 	    &txr->hn_tx_data_dtag);
 	if (error) {
 		device_printf(dev, "failed to create data dmatag\n");
 		return error;
 	}
 
 	for (i = 0; i < txr->hn_txdesc_cnt; ++i) {
 		struct hn_txdesc *txd = &txr->hn_txdesc[i];
 
 		txd->txr = txr;
 		txd->chim_index = HN_NVS_CHIM_IDX_INVALID;
 
 		/*
 		 * Allocate and load RNDIS packet message.
 		 */
         	error = bus_dmamem_alloc(txr->hn_tx_rndis_dtag,
 		    (void **)&txd->rndis_pkt,
 		    BUS_DMA_WAITOK | BUS_DMA_COHERENT | BUS_DMA_ZERO,
 		    &txd->rndis_pkt_dmap);
 		if (error) {
 			device_printf(dev,
 			    "failed to allocate rndis_packet_msg, %d\n", i);
 			return error;
 		}
 
 		error = bus_dmamap_load(txr->hn_tx_rndis_dtag,
 		    txd->rndis_pkt_dmap,
 		    txd->rndis_pkt, HN_RNDIS_PKT_LEN,
 		    hyperv_dma_map_paddr, &txd->rndis_pkt_paddr,
 		    BUS_DMA_NOWAIT);
 		if (error) {
 			device_printf(dev,
 			    "failed to load rndis_packet_msg, %d\n", i);
 			bus_dmamem_free(txr->hn_tx_rndis_dtag,
 			    txd->rndis_pkt, txd->rndis_pkt_dmap);
 			return error;
 		}
 
 		/* DMA map for TX data. */
 		error = bus_dmamap_create(txr->hn_tx_data_dtag, 0,
 		    &txd->data_dmap);
 		if (error) {
 			device_printf(dev,
 			    "failed to allocate tx data dmamap\n");
 			bus_dmamap_unload(txr->hn_tx_rndis_dtag,
 			    txd->rndis_pkt_dmap);
 			bus_dmamem_free(txr->hn_tx_rndis_dtag,
 			    txd->rndis_pkt, txd->rndis_pkt_dmap);
 			return error;
 		}
 
 		/* All set, put it to list */
 		txd->flags |= HN_TXD_FLAG_ONLIST;
 #ifndef HN_USE_TXDESC_BUFRING
 		SLIST_INSERT_HEAD(&txr->hn_txlist, txd, link);
 #else
 		buf_ring_enqueue(txr->hn_txdesc_br, txd);
 #endif
 	}
 	txr->hn_txdesc_avail = txr->hn_txdesc_cnt;
 
 	if (sc->hn_tx_sysctl_tree != NULL) {
 		struct sysctl_oid_list *child;
 		struct sysctl_ctx_list *ctx;
 		char name[16];
 
 		/*
 		 * Create per TX ring sysctl tree:
 		 * dev.hn.UNIT.tx.RINGID
 		 */
 		ctx = device_get_sysctl_ctx(dev);
 		child = SYSCTL_CHILDREN(sc->hn_tx_sysctl_tree);
 
 		snprintf(name, sizeof(name), "%d", id);
 		txr->hn_tx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO,
 		    name, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
 
 		if (txr->hn_tx_sysctl_tree != NULL) {
 			child = SYSCTL_CHILDREN(txr->hn_tx_sysctl_tree);
 
 			SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_avail",
 			    CTLFLAG_RD, &txr->hn_txdesc_avail, 0,
 			    "# of available TX descs");
 			if (!hn_use_if_start) {
 				SYSCTL_ADD_INT(ctx, child, OID_AUTO, "oactive",
 				    CTLFLAG_RD, &txr->hn_oactive, 0,
 				    "over active");
 			}
 			SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "packets",
 			    CTLFLAG_RW, &txr->hn_pkts,
 			    "# of packets transmitted");
 		}
 	}
 
 	return 0;
 }
 
 static void
 hn_txdesc_dmamap_destroy(struct hn_txdesc *txd)
 {
 	struct hn_tx_ring *txr = txd->txr;
 
 	KASSERT(txd->m == NULL, ("still has mbuf installed"));
 	KASSERT((txd->flags & HN_TXD_FLAG_DMAMAP) == 0, ("still dma mapped"));
 
 	bus_dmamap_unload(txr->hn_tx_rndis_dtag, txd->rndis_pkt_dmap);
 	bus_dmamem_free(txr->hn_tx_rndis_dtag, txd->rndis_pkt,
 	    txd->rndis_pkt_dmap);
 	bus_dmamap_destroy(txr->hn_tx_data_dtag, txd->data_dmap);
 }
 
 static void
 hn_destroy_tx_ring(struct hn_tx_ring *txr)
 {
 	struct hn_txdesc *txd;
 
 	if (txr->hn_txdesc == NULL)
 		return;
 
 #ifndef HN_USE_TXDESC_BUFRING
 	while ((txd = SLIST_FIRST(&txr->hn_txlist)) != NULL) {
 		SLIST_REMOVE_HEAD(&txr->hn_txlist, link);
 		hn_txdesc_dmamap_destroy(txd);
 	}
 #else
 	mtx_lock(&txr->hn_tx_lock);
 	while ((txd = buf_ring_dequeue_sc(txr->hn_txdesc_br)) != NULL)
 		hn_txdesc_dmamap_destroy(txd);
 	mtx_unlock(&txr->hn_tx_lock);
 #endif
 
 	if (txr->hn_tx_data_dtag != NULL)
 		bus_dma_tag_destroy(txr->hn_tx_data_dtag);
 	if (txr->hn_tx_rndis_dtag != NULL)
 		bus_dma_tag_destroy(txr->hn_tx_rndis_dtag);
 
 #ifdef HN_USE_TXDESC_BUFRING
 	buf_ring_free(txr->hn_txdesc_br, M_DEVBUF);
 #endif
 
 	free(txr->hn_txdesc, M_DEVBUF);
 	txr->hn_txdesc = NULL;
 
 	if (txr->hn_mbuf_br != NULL)
 		buf_ring_free(txr->hn_mbuf_br, M_DEVBUF);
 
 #ifndef HN_USE_TXDESC_BUFRING
 	mtx_destroy(&txr->hn_txlist_spin);
 #endif
 	mtx_destroy(&txr->hn_tx_lock);
 }
 
 static int
 hn_create_tx_data(struct hn_softc *sc, int ring_cnt)
 {
 	struct sysctl_oid_list *child;
 	struct sysctl_ctx_list *ctx;
 	int i;
 
 	/*
 	 * Create TXBUF for chimney sending.
 	 *
 	 * NOTE: It is shared by all channels.
 	 */
 	sc->hn_chim = hyperv_dmamem_alloc(bus_get_dma_tag(sc->hn_dev),
 	    PAGE_SIZE, 0, HN_CHIM_SIZE, &sc->hn_chim_dma,
 	    BUS_DMA_WAITOK | BUS_DMA_ZERO);
 	if (sc->hn_chim == NULL) {
 		device_printf(sc->hn_dev, "allocate txbuf failed\n");
 		return (ENOMEM);
 	}
 
 	sc->hn_tx_ring_cnt = ring_cnt;
 	sc->hn_tx_ring_inuse = sc->hn_tx_ring_cnt;
 
 	sc->hn_tx_ring = malloc(sizeof(struct hn_tx_ring) * sc->hn_tx_ring_cnt,
 	    M_DEVBUF, M_WAITOK | M_ZERO);
 
 	ctx = device_get_sysctl_ctx(sc->hn_dev);
 	child = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->hn_dev));
 
 	/* Create dev.hn.UNIT.tx sysctl tree */
 	sc->hn_tx_sysctl_tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "tx",
 	    CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "");
 
 	for (i = 0; i < sc->hn_tx_ring_cnt; ++i) {
 		int error;
 
 		error = hn_create_tx_ring(sc, i);
 		if (error)
 			return error;
 	}
 
 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "no_txdescs",
 	    CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
 	    __offsetof(struct hn_tx_ring, hn_no_txdescs),
 	    hn_tx_stat_ulong_sysctl, "LU", "# of times short of TX descs");
 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "send_failed",
 	    CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
 	    __offsetof(struct hn_tx_ring, hn_send_failed),
 	    hn_tx_stat_ulong_sysctl, "LU", "# of hyper-v sending failure");
 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "txdma_failed",
 	    CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
 	    __offsetof(struct hn_tx_ring, hn_txdma_failed),
 	    hn_tx_stat_ulong_sysctl, "LU", "# of TX DMA failure");
 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_collapsed",
 	    CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
 	    __offsetof(struct hn_tx_ring, hn_tx_collapsed),
 	    hn_tx_stat_ulong_sysctl, "LU", "# of TX mbuf collapsed");
 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney",
 	    CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
 	    __offsetof(struct hn_tx_ring, hn_tx_chimney),
 	    hn_tx_stat_ulong_sysctl, "LU", "# of chimney send");
 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney_tried",
 	    CTLTYPE_ULONG | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
 	    __offsetof(struct hn_tx_ring, hn_tx_chimney_tried),
 	    hn_tx_stat_ulong_sysctl, "LU", "# of chimney send tries");
 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_cnt",
 	    CTLFLAG_RD, &sc->hn_tx_ring[0].hn_txdesc_cnt, 0,
 	    "# of total TX descs");
 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_chimney_max",
 	    CTLFLAG_RD, &sc->hn_chim_szmax, 0,
 	    "Chimney send packet size upper boundary");
 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney_size",
 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc, 0,
 	    hn_chim_size_sysctl, "I", "Chimney send packet size limit");
 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "direct_tx_size",
 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
 	    __offsetof(struct hn_tx_ring, hn_direct_tx_size),
 	    hn_tx_conf_int_sysctl, "I",
 	    "Size of the packet for direct transmission");
 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "sched_tx",
 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, sc,
 	    __offsetof(struct hn_tx_ring, hn_sched_tx),
 	    hn_tx_conf_int_sysctl, "I",
 	    "Always schedule transmission "
 	    "instead of doing direct transmission");
 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_ring_cnt",
 	    CTLFLAG_RD, &sc->hn_tx_ring_cnt, 0, "# created TX rings");
 	SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_ring_inuse",
 	    CTLFLAG_RD, &sc->hn_tx_ring_inuse, 0, "# used TX rings");
 
 	return 0;
 }
 
 static void
 hn_set_chim_size(struct hn_softc *sc, int chim_size)
 {
 	int i;
 
 	for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
 		sc->hn_tx_ring[i].hn_chim_size = chim_size;
 }
 
 static void
 hn_set_tso_maxsize(struct hn_softc *sc, int tso_maxlen, int mtu)
 {
 	struct ifnet *ifp = sc->hn_ifp;
 	int tso_minlen;
 
 	if ((ifp->if_capabilities & (IFCAP_TSO4 | IFCAP_TSO6)) == 0)
 		return;
 
 	KASSERT(sc->hn_ndis_tso_sgmin >= 2,
 	    ("invalid NDIS tso sgmin %d", sc->hn_ndis_tso_sgmin));
 	tso_minlen = sc->hn_ndis_tso_sgmin * mtu;
 
 	KASSERT(sc->hn_ndis_tso_szmax >= tso_minlen &&
 	    sc->hn_ndis_tso_szmax <= IP_MAXPACKET,
 	    ("invalid NDIS tso szmax %d", sc->hn_ndis_tso_szmax));
 
 	if (tso_maxlen < tso_minlen)
 		tso_maxlen = tso_minlen;
 	else if (tso_maxlen > IP_MAXPACKET)
 		tso_maxlen = IP_MAXPACKET;
 	if (tso_maxlen > sc->hn_ndis_tso_szmax)
 		tso_maxlen = sc->hn_ndis_tso_szmax;
 	ifp->if_hw_tsomax = tso_maxlen - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
 	if (bootverbose)
 		if_printf(ifp, "TSO size max %u\n", ifp->if_hw_tsomax);
 }
 
 static void
 hn_fixup_tx_data(struct hn_softc *sc)
 {
 	uint64_t csum_assist;
 	int i;
 
 	hn_set_chim_size(sc, sc->hn_chim_szmax);
 	if (hn_tx_chimney_size > 0 &&
 	    hn_tx_chimney_size < sc->hn_chim_szmax)
 		hn_set_chim_size(sc, hn_tx_chimney_size);
 
 	csum_assist = 0;
 	if (sc->hn_caps & HN_CAP_IPCS)
 		csum_assist |= CSUM_IP;
 	if (sc->hn_caps & HN_CAP_TCP4CS)
 		csum_assist |= CSUM_IP_TCP;
 	if (sc->hn_caps & HN_CAP_UDP4CS)
 		csum_assist |= CSUM_IP_UDP;
 #ifdef notyet
 	if (sc->hn_caps & HN_CAP_TCP6CS)
 		csum_assist |= CSUM_IP6_TCP;
 	if (sc->hn_caps & HN_CAP_UDP6CS)
 		csum_assist |= CSUM_IP6_UDP;
 #endif
 	for (i = 0; i < sc->hn_tx_ring_cnt; ++i)
 		sc->hn_tx_ring[i].hn_csum_assist = csum_assist;
 
 	if (sc->hn_caps & HN_CAP_HASHVAL) {
 		/*
 		 * Support HASHVAL pktinfo on TX path.
 		 */
 		if (bootverbose)
 			if_printf(sc->hn_ifp, "support HASHVAL pktinfo\n");
 		for (i = 0; i < sc->hn_tx_ring_cnt; ++i)
 			sc->hn_tx_ring[i].hn_tx_flags |= HN_TX_FLAG_HASHVAL;
 	}
 }
 
 static void
 hn_destroy_tx_data(struct hn_softc *sc)
 {
 	int i;
 
 	if (sc->hn_chim != NULL) {
 		hyperv_dmamem_free(&sc->hn_chim_dma, sc->hn_chim);
 		sc->hn_chim = NULL;
 	}
 
 	if (sc->hn_tx_ring_cnt == 0)
 		return;
 
 	for (i = 0; i < sc->hn_tx_ring_cnt; ++i)
 		hn_destroy_tx_ring(&sc->hn_tx_ring[i]);
 
 	free(sc->hn_tx_ring, M_DEVBUF);
 	sc->hn_tx_ring = NULL;
 
 	sc->hn_tx_ring_cnt = 0;
 	sc->hn_tx_ring_inuse = 0;
 }
 
 static void
 hn_start_taskfunc(void *xtxr, int pending __unused)
 {
 	struct hn_tx_ring *txr = xtxr;
 
 	mtx_lock(&txr->hn_tx_lock);
 	hn_start_locked(txr, 0);
 	mtx_unlock(&txr->hn_tx_lock);
 }
 
 static void
 hn_start_txeof_taskfunc(void *xtxr, int pending __unused)
 {
 	struct hn_tx_ring *txr = xtxr;
 
 	mtx_lock(&txr->hn_tx_lock);
 	atomic_clear_int(&txr->hn_sc->hn_ifp->if_drv_flags, IFF_DRV_OACTIVE);
 	hn_start_locked(txr, 0);
 	mtx_unlock(&txr->hn_tx_lock);
 }
 
 static int
 hn_xmit(struct hn_tx_ring *txr, int len)
 {
 	struct hn_softc *sc = txr->hn_sc;
 	struct ifnet *ifp = sc->hn_ifp;
 	struct mbuf *m_head;
 
 	mtx_assert(&txr->hn_tx_lock, MA_OWNED);
 	KASSERT(hn_use_if_start == 0,
 	    ("hn_xmit is called, when if_start is enabled"));
 
 	if (__predict_false(txr->hn_suspended))
 		return 0;
 
 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 || txr->hn_oactive)
 		return 0;
 
 	while ((m_head = drbr_peek(ifp, txr->hn_mbuf_br)) != NULL) {
 		struct hn_txdesc *txd;
 		int error;
 
 		if (len > 0 && m_head->m_pkthdr.len > len) {
 			/*
 			 * This sending could be time consuming; let callers
 			 * dispatch this packet sending (and sending of any
 			 * following up packets) to tx taskqueue.
 			 */
 			drbr_putback(ifp, txr->hn_mbuf_br, m_head);
 			return 1;
 		}
 
 		txd = hn_txdesc_get(txr);
 		if (txd == NULL) {
 			txr->hn_no_txdescs++;
 			drbr_putback(ifp, txr->hn_mbuf_br, m_head);
 			txr->hn_oactive = 1;
 			break;
 		}
 
 		error = hn_encap(txr, txd, &m_head);
 		if (error) {
 			/* Both txd and m_head are freed; discard */
 			drbr_advance(ifp, txr->hn_mbuf_br);
 			continue;
 		}
 
 		error = hn_send_pkt(ifp, txr, txd);
 		if (__predict_false(error)) {
 			/* txd is freed, but m_head is not */
 			drbr_putback(ifp, txr->hn_mbuf_br, m_head);
 			txr->hn_oactive = 1;
 			break;
 		}
 
 		/* Sent */
 		drbr_advance(ifp, txr->hn_mbuf_br);
 	}
 	return 0;
 }
 
 static int
 hn_transmit(struct ifnet *ifp, struct mbuf *m)
 {
 	struct hn_softc *sc = ifp->if_softc;
 	struct hn_tx_ring *txr;
 	int error, idx = 0;
 
 	/*
 	 * Select the TX ring based on flowid
 	 */
 	if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
 		idx = m->m_pkthdr.flowid % sc->hn_tx_ring_inuse;
 	txr = &sc->hn_tx_ring[idx];
 
 	error = drbr_enqueue(ifp, txr->hn_mbuf_br, m);
 	if (error) {
 		if_inc_counter(ifp, IFCOUNTER_OQDROPS, 1);
 		return error;
 	}
 
 	if (txr->hn_oactive)
 		return 0;
 
 	if (txr->hn_sched_tx)
 		goto do_sched;
 
 	if (mtx_trylock(&txr->hn_tx_lock)) {
 		int sched;
 
 		sched = hn_xmit(txr, txr->hn_direct_tx_size);
 		mtx_unlock(&txr->hn_tx_lock);
 		if (!sched)
 			return 0;
 	}
 do_sched:
 	taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_tx_task);
 	return 0;
 }
 
 static void
 hn_tx_ring_qflush(struct hn_tx_ring *txr)
 {
 	struct mbuf *m;
 
 	mtx_lock(&txr->hn_tx_lock);
 	while ((m = buf_ring_dequeue_sc(txr->hn_mbuf_br)) != NULL)
 		m_freem(m);
 	mtx_unlock(&txr->hn_tx_lock);
 }
 
 static void
 hn_xmit_qflush(struct ifnet *ifp)
 {
 	struct hn_softc *sc = ifp->if_softc;
 	int i;
 
 	for (i = 0; i < sc->hn_tx_ring_inuse; ++i)
 		hn_tx_ring_qflush(&sc->hn_tx_ring[i]);
 	if_qflush(ifp);
 }
 
 static void
 hn_xmit_txeof(struct hn_tx_ring *txr)
 {
 
 	if (txr->hn_sched_tx)
 		goto do_sched;
 
 	if (mtx_trylock(&txr->hn_tx_lock)) {
 		int sched;
 
 		txr->hn_oactive = 0;
 		sched = hn_xmit(txr, txr->hn_direct_tx_size);
 		mtx_unlock(&txr->hn_tx_lock);
 		if (sched) {
 			taskqueue_enqueue(txr->hn_tx_taskq,
 			    &txr->hn_tx_task);
 		}
 	} else {
 do_sched:
 		/*
 		 * Release the oactive earlier, with the hope, that
 		 * others could catch up.  The task will clear the
 		 * oactive again with the hn_tx_lock to avoid possible
 		 * races.
 		 */
 		txr->hn_oactive = 0;
 		taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_txeof_task);
 	}
 }
 
 static void
 hn_xmit_taskfunc(void *xtxr, int pending __unused)
 {
 	struct hn_tx_ring *txr = xtxr;
 
 	mtx_lock(&txr->hn_tx_lock);
 	hn_xmit(txr, 0);
 	mtx_unlock(&txr->hn_tx_lock);
 }
 
 static void
 hn_xmit_txeof_taskfunc(void *xtxr, int pending __unused)
 {
 	struct hn_tx_ring *txr = xtxr;
 
 	mtx_lock(&txr->hn_tx_lock);
 	txr->hn_oactive = 0;
 	hn_xmit(txr, 0);
 	mtx_unlock(&txr->hn_tx_lock);
 }
 
 static int
 hn_chan_attach(struct hn_softc *sc, struct vmbus_channel *chan)
 {
 	struct vmbus_chan_br cbr;
 	struct hn_rx_ring *rxr;
 	struct hn_tx_ring *txr = NULL;
 	int idx, error;
 
 	idx = vmbus_chan_subidx(chan);
 
 	/*
 	 * Link this channel to RX/TX ring.
 	 */
 	KASSERT(idx >= 0 && idx < sc->hn_rx_ring_inuse,
 	    ("invalid channel index %d, should > 0 && < %d",
 	     idx, sc->hn_rx_ring_inuse));
 	rxr = &sc->hn_rx_ring[idx];
 	KASSERT((rxr->hn_rx_flags & HN_RX_FLAG_ATTACHED) == 0,
 	    ("RX ring %d already attached", idx));
 	rxr->hn_rx_flags |= HN_RX_FLAG_ATTACHED;
 
 	if (bootverbose) {
 		if_printf(sc->hn_ifp, "link RX ring %d to chan%u\n",
 		    idx, vmbus_chan_id(chan));
 	}
 
 	if (idx < sc->hn_tx_ring_inuse) {
 		txr = &sc->hn_tx_ring[idx];
 		KASSERT((txr->hn_tx_flags & HN_TX_FLAG_ATTACHED) == 0,
 		    ("TX ring %d already attached", idx));
 		txr->hn_tx_flags |= HN_TX_FLAG_ATTACHED;
 
 		txr->hn_chan = chan;
 		if (bootverbose) {
 			if_printf(sc->hn_ifp, "link TX ring %d to chan%u\n",
 			    idx, vmbus_chan_id(chan));
 		}
 	}
 
 	/* Bind this channel to a proper CPU. */
 	vmbus_chan_cpu_set(chan, (sc->hn_cpu + idx) % mp_ncpus);
 
 	/*
 	 * Open this channel
 	 */
 	cbr.cbr = rxr->hn_br;
 	cbr.cbr_paddr = rxr->hn_br_dma.hv_paddr;
 	cbr.cbr_txsz = HN_TXBR_SIZE;
 	cbr.cbr_rxsz = HN_RXBR_SIZE;
 	error = vmbus_chan_open_br(chan, &cbr, NULL, 0, hn_chan_callback, rxr);
 	if (error) {
 		if_printf(sc->hn_ifp, "open chan%u failed: %d\n",
 		    vmbus_chan_id(chan), error);
 		rxr->hn_rx_flags &= ~HN_RX_FLAG_ATTACHED;
 		if (txr != NULL)
 			txr->hn_tx_flags &= ~HN_TX_FLAG_ATTACHED;
 	}
 	return (error);
 }
 
 static void
 hn_chan_detach(struct hn_softc *sc, struct vmbus_channel *chan)
 {
 	struct hn_rx_ring *rxr;
 	int idx;
 
 	idx = vmbus_chan_subidx(chan);
 
 	/*
 	 * Link this channel to RX/TX ring.
 	 */
 	KASSERT(idx >= 0 && idx < sc->hn_rx_ring_inuse,
 	    ("invalid channel index %d, should > 0 && < %d",
 	     idx, sc->hn_rx_ring_inuse));
 	rxr = &sc->hn_rx_ring[idx];
 	KASSERT((rxr->hn_rx_flags & HN_RX_FLAG_ATTACHED),
 	    ("RX ring %d is not attached", idx));
 	rxr->hn_rx_flags &= ~HN_RX_FLAG_ATTACHED;
 
 	if (idx < sc->hn_tx_ring_inuse) {
 		struct hn_tx_ring *txr = &sc->hn_tx_ring[idx];
 
 		KASSERT((txr->hn_tx_flags & HN_TX_FLAG_ATTACHED),
 		    ("TX ring %d is not attached attached", idx));
 		txr->hn_tx_flags &= ~HN_TX_FLAG_ATTACHED;
 	}
 
 	/*
 	 * Close this channel.
 	 *
 	 * NOTE:
 	 * Channel closing does _not_ destroy the target channel.
 	 */
 	vmbus_chan_close(chan);
 }
 
 static int
 hn_attach_subchans(struct hn_softc *sc)
 {
 	struct vmbus_channel **subchans;
 	int subchan_cnt = sc->hn_rx_ring_inuse - 1;
 	int i, error = 0;
 
 	if (subchan_cnt == 0)
 		return (0);
 
 	/* Attach the sub-channels. */
 	subchans = vmbus_subchan_get(sc->hn_prichan, subchan_cnt);
 	for (i = 0; i < subchan_cnt; ++i) {
 		error = hn_chan_attach(sc, subchans[i]);
 		if (error)
 			break;
 	}
 	vmbus_subchan_rel(subchans, subchan_cnt);
 
 	if (error) {
 		if_printf(sc->hn_ifp, "sub-channels attach failed: %d\n", error);
 	} else {
 		if (bootverbose) {
 			if_printf(sc->hn_ifp, "%d sub-channels attached\n",
 			    subchan_cnt);
 		}
 	}
 	return (error);
 }
 
 static void
 hn_detach_allchans(struct hn_softc *sc)
 {
 	struct vmbus_channel **subchans;
 	int subchan_cnt = sc->hn_rx_ring_inuse - 1;
 	int i;
 
 	if (subchan_cnt == 0)
 		goto back;
 
 	/* Detach the sub-channels. */
 	subchans = vmbus_subchan_get(sc->hn_prichan, subchan_cnt);
 	for (i = 0; i < subchan_cnt; ++i)
 		hn_chan_detach(sc, subchans[i]);
 	vmbus_subchan_rel(subchans, subchan_cnt);
 
 back:
 	/*
 	 * Detach the primary channel, _after_ all sub-channels
 	 * are detached.
 	 */
 	hn_chan_detach(sc, sc->hn_prichan);
 
 	/* Wait for sub-channels to be destroyed, if any. */
 	vmbus_subchan_drain(sc->hn_prichan);
 
 #ifdef INVARIANTS
 	for (i = 0; i < sc->hn_rx_ring_cnt; ++i) {
 		KASSERT((sc->hn_rx_ring[i].hn_rx_flags &
 		    HN_RX_FLAG_ATTACHED) == 0,
 		    ("%dth RX ring is still attached", i));
 	}
 	for (i = 0; i < sc->hn_tx_ring_cnt; ++i) {
 		KASSERT((sc->hn_tx_ring[i].hn_tx_flags &
 		    HN_TX_FLAG_ATTACHED) == 0,
 		    ("%dth TX ring is still attached", i));
 	}
 #endif
 }
 
 static int
 hn_synth_alloc_subchans(struct hn_softc *sc, int *nsubch)
 {
 	struct vmbus_channel **subchans;
 	int nchan, rxr_cnt, error;
 
 	nchan = *nsubch + 1;
 	if (nchan == 1) {
 		/*
 		 * Multiple RX/TX rings are not requested.
 		 */
 		*nsubch = 0;
 		return (0);
 	}
 
 	/*
 	 * Query RSS capabilities, e.g. # of RX rings, and # of indirect
 	 * table entries.
 	 */
 	error = hn_rndis_query_rsscaps(sc, &rxr_cnt);
 	if (error) {
 		/* No RSS; this is benign. */
 		*nsubch = 0;
 		return (0);
 	}
 	if (bootverbose) {
 		if_printf(sc->hn_ifp, "RX rings offered %u, requested %d\n",
 		    rxr_cnt, nchan);
 	}
 
 	if (nchan > rxr_cnt)
 		nchan = rxr_cnt;
 	if (nchan == 1) {
 		if_printf(sc->hn_ifp, "only 1 channel is supported, no vRSS\n");
 		*nsubch = 0;
 		return (0);
 	}
 
 	/*
 	 * Allocate sub-channels from NVS.
 	 */
 	*nsubch = nchan - 1;
 	error = hn_nvs_alloc_subchans(sc, nsubch);
 	if (error || *nsubch == 0) {
 		/* Failed to allocate sub-channels. */
 		*nsubch = 0;
 		return (0);
 	}
 
 	/*
 	 * Wait for all sub-channels to become ready before moving on.
 	 */
 	subchans = vmbus_subchan_get(sc->hn_prichan, *nsubch);
 	vmbus_subchan_rel(subchans, *nsubch);
 	return (0);
 }
 
 static int
 hn_synth_attach(struct hn_softc *sc, int mtu)
 {
 	struct ndis_rssprm_toeplitz *rss = &sc->hn_rss;
 	int error, nsubch, nchan, i;
 	uint32_t old_caps;
 
 	KASSERT((sc->hn_flags & HN_FLAG_SYNTH_ATTACHED) == 0,
 	    ("synthetic parts were attached"));
 
 	/* Save capabilities for later verification. */
 	old_caps = sc->hn_caps;
 	sc->hn_caps = 0;
 
 	/* Clear RSS stuffs. */
 	sc->hn_rss_ind_size = 0;
 	sc->hn_rss_hash = 0;
 
 	/*
 	 * Attach the primary channel _before_ attaching NVS and RNDIS.
 	 */
 	error = hn_chan_attach(sc, sc->hn_prichan);
 	if (error)
 		return (error);
 
 	/*
 	 * Attach NVS.
 	 */
 	error = hn_nvs_attach(sc, mtu);
 	if (error)
 		return (error);
 
 	/*
 	 * Attach RNDIS _after_ NVS is attached.
 	 */
 	error = hn_rndis_attach(sc, mtu);
 	if (error)
 		return (error);
 
 	/*
 	 * Make sure capabilities are not changed.
 	 */
 	if (device_is_attached(sc->hn_dev) && old_caps != sc->hn_caps) {
 		if_printf(sc->hn_ifp, "caps mismatch old 0x%08x, new 0x%08x\n",
 		    old_caps, sc->hn_caps);
 		/* Restore old capabilities and abort. */
 		sc->hn_caps = old_caps;
 		return ENXIO;
 	}
 
 	/*
 	 * Allocate sub-channels for multi-TX/RX rings.
 	 *
 	 * NOTE:
 	 * The # of RX rings that can be used is equivalent to the # of
 	 * channels to be requested.
 	 */
 	nsubch = sc->hn_rx_ring_cnt - 1;
 	error = hn_synth_alloc_subchans(sc, &nsubch);
 	if (error)
 		return (error);
 
 	nchan = nsubch + 1;
 	if (nchan == 1) {
 		/* Only the primary channel can be used; done */
 		goto back;
 	}
 
 	/*
 	 * Configure RSS key and indirect table _after_ all sub-channels
 	 * are allocated.
 	 */
 
 	if ((sc->hn_flags & HN_FLAG_HAS_RSSKEY) == 0) {
 		/*
 		 * RSS key is not set yet; set it to the default RSS key.
 		 */
 		if (bootverbose)
 			if_printf(sc->hn_ifp, "setup default RSS key\n");
 		memcpy(rss->rss_key, hn_rss_key_default, sizeof(rss->rss_key));
 		sc->hn_flags |= HN_FLAG_HAS_RSSKEY;
 	}
 
 	if ((sc->hn_flags & HN_FLAG_HAS_RSSIND) == 0) {
 		/*
 		 * RSS indirect table is not set yet; set it up in round-
 		 * robin fashion.
 		 */
 		if (bootverbose) {
 			if_printf(sc->hn_ifp, "setup default RSS indirect "
 			    "table\n");
 		}
 		for (i = 0; i < NDIS_HASH_INDCNT; ++i)
 			rss->rss_ind[i] = i % nchan;
 		sc->hn_flags |= HN_FLAG_HAS_RSSIND;
 	} else {
 		/*
 		 * # of usable channels may be changed, so we have to
 		 * make sure that all entries in RSS indirect table
 		 * are valid.
 		 */
 		hn_rss_ind_fixup(sc, nchan);
 	}
 
 	error = hn_rndis_conf_rss(sc, NDIS_RSS_FLAG_NONE);
 	if (error) {
 		/*
 		 * Failed to configure RSS key or indirect table; only
 		 * the primary channel can be used.
 		 */
 		nchan = 1;
 	}
 back:
 	/*
 	 * Set the # of TX/RX rings that could be used according to
 	 * the # of channels that NVS offered.
 	 */
 	hn_set_ring_inuse(sc, nchan);
 
 	/*
 	 * Attach the sub-channels, if any.
 	 */
 	error = hn_attach_subchans(sc);
 	if (error)
 		return (error);
 
 	sc->hn_flags |= HN_FLAG_SYNTH_ATTACHED;
 	return (0);
 }
 
 /*
  * NOTE:
  * The interface must have been suspended though hn_suspend(), before
  * this function get called.
  */
 static void
 hn_synth_detach(struct hn_softc *sc)
 {
 	HN_LOCK_ASSERT(sc);
 
 	KASSERT(sc->hn_flags & HN_FLAG_SYNTH_ATTACHED,
 	    ("synthetic parts were not attached"));
 
 	/* Detach the RNDIS first. */
 	hn_rndis_detach(sc);
 
 	/* Detach NVS. */
 	hn_nvs_detach(sc);
 
 	/* Detach all of the channels. */
 	hn_detach_allchans(sc);
 
 	sc->hn_flags &= ~HN_FLAG_SYNTH_ATTACHED;
 }
 
 static void
 hn_set_ring_inuse(struct hn_softc *sc, int ring_cnt)
 {
 	KASSERT(ring_cnt > 0 && ring_cnt <= sc->hn_rx_ring_cnt,
 	    ("invalid ring count %d", ring_cnt));
 
 	if (sc->hn_tx_ring_cnt > ring_cnt)
 		sc->hn_tx_ring_inuse = ring_cnt;
 	else
 		sc->hn_tx_ring_inuse = sc->hn_tx_ring_cnt;
 	sc->hn_rx_ring_inuse = ring_cnt;
 
 	if (bootverbose) {
 		if_printf(sc->hn_ifp, "%d TX ring, %d RX ring\n",
 		    sc->hn_tx_ring_inuse, sc->hn_rx_ring_inuse);
 	}
 }
 
 static void
 hn_rx_drain(struct vmbus_channel *chan)
 {
 
 	while (!vmbus_chan_rx_empty(chan) || !vmbus_chan_tx_empty(chan))
 		pause("waitch", 1);
 	vmbus_chan_intr_drain(chan);
 }
 
 static void
 hn_suspend_data(struct hn_softc *sc)
 {
 	struct vmbus_channel **subch = NULL;
 	int i, nsubch;
 
 	HN_LOCK_ASSERT(sc);
 
 	/*
 	 * Suspend TX.
 	 */
 	for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
 		struct hn_tx_ring *txr = &sc->hn_tx_ring[i];
 
 		mtx_lock(&txr->hn_tx_lock);
 		txr->hn_suspended = 1;
 		mtx_unlock(&txr->hn_tx_lock);
 		/* No one is able send more packets now. */
 
 		/* Wait for all pending sends to finish. */
 		while (hn_tx_ring_pending(txr))
 			pause("hnwtx", 1 /* 1 tick */);
 
 		taskqueue_drain(txr->hn_tx_taskq, &txr->hn_tx_task);
 		taskqueue_drain(txr->hn_tx_taskq, &txr->hn_txeof_task);
 	}
 
 	/*
 	 * Disable RX by clearing RX filter.
 	 */
 	sc->hn_rx_filter = NDIS_PACKET_TYPE_NONE;
 	hn_rndis_set_rxfilter(sc, sc->hn_rx_filter);
 
 	/*
 	 * Give RNDIS enough time to flush all pending data packets.
 	 */
 	pause("waitrx", (200 * hz) / 1000);
 
 	/*
 	 * Drain RX/TX bufrings and interrupts.
 	 */
 	nsubch = sc->hn_rx_ring_inuse - 1;
 	if (nsubch > 0)
 		subch = vmbus_subchan_get(sc->hn_prichan, nsubch);
 
 	if (subch != NULL) {
 		for (i = 0; i < nsubch; ++i)
 			hn_rx_drain(subch[i]);
 	}
 	hn_rx_drain(sc->hn_prichan);
 
 	if (subch != NULL)
 		vmbus_subchan_rel(subch, nsubch);
 }
 
 static void
 hn_suspend_mgmt_taskfunc(void *xsc, int pending __unused)
 {
 
 	((struct hn_softc *)xsc)->hn_mgmt_taskq = NULL;
 }
 
 static void
 hn_suspend_mgmt(struct hn_softc *sc)
 {
 	struct task task;
 
 	HN_LOCK_ASSERT(sc);
 
 	/*
 	 * Make sure that hn_mgmt_taskq0 can nolonger be accessed
 	 * through hn_mgmt_taskq.
 	 */
 	TASK_INIT(&task, 0, hn_suspend_mgmt_taskfunc, sc);
 	vmbus_chan_run_task(sc->hn_prichan, &task);
 
 	/*
 	 * Make sure that all pending management tasks are completed.
 	 */
 	taskqueue_drain(sc->hn_mgmt_taskq0, &sc->hn_netchg_init);
 	taskqueue_drain_timeout(sc->hn_mgmt_taskq0, &sc->hn_netchg_status);
 	taskqueue_drain_all(sc->hn_mgmt_taskq0);
 }
 
 static void
 hn_suspend(struct hn_softc *sc)
 {
 
 	if (sc->hn_ifp->if_drv_flags & IFF_DRV_RUNNING)
 		hn_suspend_data(sc);
 	hn_suspend_mgmt(sc);
 }
 
 static void
 hn_tx_resume(struct hn_softc *sc, int tx_ring_cnt)
 {
 	int i;
 
 	KASSERT(tx_ring_cnt <= sc->hn_tx_ring_cnt,
 	    ("invalid TX ring count %d", tx_ring_cnt));
 
 	for (i = 0; i < tx_ring_cnt; ++i) {
 		struct hn_tx_ring *txr = &sc->hn_tx_ring[i];
 
 		mtx_lock(&txr->hn_tx_lock);
 		txr->hn_suspended = 0;
 		mtx_unlock(&txr->hn_tx_lock);
 	}
 }
 
 static void
 hn_resume_data(struct hn_softc *sc)
 {
 	int i;
 
 	HN_LOCK_ASSERT(sc);
 
 	/*
 	 * Re-enable RX.
 	 */
 	hn_set_rxfilter(sc);
 
 	/*
 	 * Make sure to clear suspend status on "all" TX rings,
 	 * since hn_tx_ring_inuse can be changed after
 	 * hn_suspend_data().
 	 */
 	hn_tx_resume(sc, sc->hn_tx_ring_cnt);
 
 	if (!hn_use_if_start) {
 		/*
 		 * Flush unused drbrs, since hn_tx_ring_inuse may be
 		 * reduced.
 		 */
 		for (i = sc->hn_tx_ring_inuse; i < sc->hn_tx_ring_cnt; ++i)
 			hn_tx_ring_qflush(&sc->hn_tx_ring[i]);
 	}
 
 	/*
 	 * Kick start TX.
 	 */
 	for (i = 0; i < sc->hn_tx_ring_inuse; ++i) {
 		struct hn_tx_ring *txr = &sc->hn_tx_ring[i];
 
 		/*
 		 * Use txeof task, so that any pending oactive can be
 		 * cleared properly.
 		 */
 		taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_txeof_task);
 	}
 }
 
 static void
 hn_resume_mgmt(struct hn_softc *sc)
 {
 
 	sc->hn_mgmt_taskq = sc->hn_mgmt_taskq0;
 
 	/*
 	 * Kick off network change detection, if it was pending.
 	 * If no network change was pending, start link status
 	 * checks, which is more lightweight than network change
 	 * detection.
 	 */
 	if (sc->hn_link_flags & HN_LINK_FLAG_NETCHG)
 		hn_network_change(sc);
 	else
 		hn_link_status_update(sc);
 }
 
 static void
 hn_resume(struct hn_softc *sc)
 {
 
 	if (sc->hn_ifp->if_drv_flags & IFF_DRV_RUNNING)
 		hn_resume_data(sc);
 	hn_resume_mgmt(sc);
 }
 
 static void
 hn_nvs_handle_notify(struct hn_softc *sc, const struct vmbus_chanpkt_hdr *pkt)
 {
 	const struct hn_nvs_hdr *hdr;
 
 	if (VMBUS_CHANPKT_DATALEN(pkt) < sizeof(*hdr)) {
 		if_printf(sc->hn_ifp, "invalid nvs notify\n");
 		return;
 	}
 	hdr = VMBUS_CHANPKT_CONST_DATA(pkt);
 
 	if (hdr->nvs_type == HN_NVS_TYPE_TXTBL_NOTE) {
 		/* Useless; ignore */
 		return;
 	}
 	if_printf(sc->hn_ifp, "got notify, nvs type %u\n", hdr->nvs_type);
 }
 
 static void
 hn_nvs_handle_comp(struct hn_softc *sc, struct vmbus_channel *chan,
     const struct vmbus_chanpkt_hdr *pkt)
 {
-	struct hn_send_ctx *sndc;
+	struct hn_nvs_sendctx *sndc;
 
-	sndc = (struct hn_send_ctx *)(uintptr_t)pkt->cph_xactid;
+	sndc = (struct hn_nvs_sendctx *)(uintptr_t)pkt->cph_xactid;
 	sndc->hn_cb(sndc, sc, chan, VMBUS_CHANPKT_CONST_DATA(pkt),
 	    VMBUS_CHANPKT_DATALEN(pkt));
 	/*
 	 * NOTE:
 	 * 'sndc' CAN NOT be accessed anymore, since it can be freed by
 	 * its callback.
 	 */
 }
 
 static void
 hn_nvs_handle_rxbuf(struct hn_softc *sc, struct hn_rx_ring *rxr,
     struct vmbus_channel *chan, const struct vmbus_chanpkt_hdr *pkthdr)
 {
 	const struct vmbus_chanpkt_rxbuf *pkt;
 	const struct hn_nvs_hdr *nvs_hdr;
 	int count, i, hlen;
 
 	if (__predict_false(VMBUS_CHANPKT_DATALEN(pkthdr) < sizeof(*nvs_hdr))) {
 		if_printf(rxr->hn_ifp, "invalid nvs RNDIS\n");
 		return;
 	}
 	nvs_hdr = VMBUS_CHANPKT_CONST_DATA(pkthdr);
 
 	/* Make sure that this is a RNDIS message. */
 	if (__predict_false(nvs_hdr->nvs_type != HN_NVS_TYPE_RNDIS)) {
 		if_printf(rxr->hn_ifp, "nvs type %u, not RNDIS\n",
 		    nvs_hdr->nvs_type);
 		return;
 	}
 
 	hlen = VMBUS_CHANPKT_GETLEN(pkthdr->cph_hlen);
 	if (__predict_false(hlen < sizeof(*pkt))) {
 		if_printf(rxr->hn_ifp, "invalid rxbuf chanpkt\n");
 		return;
 	}
 	pkt = (const struct vmbus_chanpkt_rxbuf *)pkthdr;
 
 	if (__predict_false(pkt->cp_rxbuf_id != HN_NVS_RXBUF_SIG)) {
 		if_printf(rxr->hn_ifp, "invalid rxbuf_id 0x%08x\n",
 		    pkt->cp_rxbuf_id);
 		return;
 	}
 
 	count = pkt->cp_rxbuf_cnt;
 	if (__predict_false(hlen <
 	    __offsetof(struct vmbus_chanpkt_rxbuf, cp_rxbuf[count]))) {
 		if_printf(rxr->hn_ifp, "invalid rxbuf_cnt %d\n", count);
 		return;
 	}
 
 	/* Each range represents 1 RNDIS pkt that contains 1 Ethernet frame */
 	for (i = 0; i < count; ++i) {
 		int ofs, len;
 
 		ofs = pkt->cp_rxbuf[i].rb_ofs;
 		len = pkt->cp_rxbuf[i].rb_len;
 		if (__predict_false(ofs + len > HN_RXBUF_SIZE)) {
 			if_printf(rxr->hn_ifp, "%dth RNDIS msg overflow rxbuf, "
 			    "ofs %d, len %d\n", i, ofs, len);
 			continue;
 		}
 		hv_rf_on_receive(sc, rxr, rxr->hn_rxbuf + ofs, len);
 	}
 	
 	/*
 	 * Moved completion call back here so that all received 
 	 * messages (not just data messages) will trigger a response
 	 * message back to the host.
 	 */
 	hn_nvs_ack_rxbuf(chan, pkt->cp_hdr.cph_xactid);
 }
 
 /*
  * Net VSC on receive completion
  *
  * Send a receive completion packet to RNDIS device (ie NetVsp)
  */
 static void
 hn_nvs_ack_rxbuf(struct vmbus_channel *chan, uint64_t tid)
 {
 	struct hn_nvs_rndis_ack ack;
 	int retries = 0;
 	int ret = 0;
 	
 	ack.nvs_type = HN_NVS_TYPE_RNDIS_ACK;
 	ack.nvs_status = HN_NVS_STATUS_OK;
 
 retry_send_cmplt:
 	/* Send the completion */
 	ret = vmbus_chan_send(chan, VMBUS_CHANPKT_TYPE_COMP,
 	    VMBUS_CHANPKT_FLAG_NONE, &ack, sizeof(ack), tid);
 	if (ret == 0) {
 		/* success */
 		/* no-op */
 	} else if (ret == EAGAIN) {
 		/* no more room... wait a bit and attempt to retry 3 times */
 		retries++;
 
 		if (retries < 4) {
 			DELAY(100);
 			goto retry_send_cmplt;
 		}
 	}
 }
 
 static void
 hn_chan_callback(struct vmbus_channel *chan, void *xrxr)
 {
 	struct hn_rx_ring *rxr = xrxr;
 	struct hn_softc *sc = rxr->hn_ifp->if_softc;
 	void *buffer;
 	int bufferlen = HN_PKTBUF_LEN;
 
 	buffer = rxr->hn_pktbuf;
 	do {
 		struct vmbus_chanpkt_hdr *pkt = buffer;
 		uint32_t bytes_rxed;
 		int ret;
 
 		bytes_rxed = bufferlen;
 		ret = vmbus_chan_recv_pkt(chan, pkt, &bytes_rxed);
 		if (ret == 0) {
 			switch (pkt->cph_type) {
 			case VMBUS_CHANPKT_TYPE_COMP:
 				hn_nvs_handle_comp(sc, chan, pkt);
 				break;
 			case VMBUS_CHANPKT_TYPE_RXBUF:
 				hn_nvs_handle_rxbuf(sc, rxr, chan, pkt);
 				break;
 			case VMBUS_CHANPKT_TYPE_INBAND:
 				hn_nvs_handle_notify(sc, pkt);
 				break;
 			default:
 				if_printf(rxr->hn_ifp,
 				    "unknown chan pkt %u\n",
 				    pkt->cph_type);
 				break;
 			}
 		} else if (ret == ENOBUFS) {
 			/* Handle large packet */
 			if (bufferlen > HN_PKTBUF_LEN) {
 				free(buffer, M_DEVBUF);
 				buffer = NULL;
 			}
 
 			/* alloc new buffer */
 			buffer = malloc(bytes_rxed, M_DEVBUF, M_NOWAIT);
 			if (buffer == NULL) {
 				if_printf(rxr->hn_ifp,
 				    "hv_cb malloc buffer failed, len=%u\n",
 				    bytes_rxed);
 				bufferlen = 0;
 				break;
 			}
 			bufferlen = bytes_rxed;
 		} else {
 			/* No more packets */
 			break;
 		}
 	} while (1);
 
 	if (bufferlen > HN_PKTBUF_LEN)
 		free(buffer, M_DEVBUF);
 
 	hv_rf_channel_rollup(rxr, rxr->hn_txr);
 }
 
 static void
 hn_tx_taskq_create(void *arg __unused)
 {
 	if (!hn_share_tx_taskq)
 		return;
 
 	hn_tx_taskq = taskqueue_create("hn_tx", M_WAITOK,
 	    taskqueue_thread_enqueue, &hn_tx_taskq);
 	if (hn_bind_tx_taskq >= 0) {
 		int cpu = hn_bind_tx_taskq;
 		cpuset_t cpu_set;
 
 		if (cpu > mp_ncpus - 1)
 			cpu = mp_ncpus - 1;
 		CPU_SETOF(cpu, &cpu_set);
 		taskqueue_start_threads_cpuset(&hn_tx_taskq, 1, PI_NET,
 		    &cpu_set, "hn tx");
 	} else {
 		taskqueue_start_threads(&hn_tx_taskq, 1, PI_NET, "hn tx");
 	}
 }
 SYSINIT(hn_txtq_create, SI_SUB_DRIVERS, SI_ORDER_FIRST,
     hn_tx_taskq_create, NULL);
 
 static void
 hn_tx_taskq_destroy(void *arg __unused)
 {
 	if (hn_tx_taskq != NULL)
 		taskqueue_free(hn_tx_taskq);
 }
 SYSUNINIT(hn_txtq_destroy, SI_SUB_DRIVERS, SI_ORDER_FIRST,
     hn_tx_taskq_destroy, NULL);
 
 static device_method_t netvsc_methods[] = {
         /* Device interface */
         DEVMETHOD(device_probe,         netvsc_probe),
         DEVMETHOD(device_attach,        netvsc_attach),
         DEVMETHOD(device_detach,        netvsc_detach),
         DEVMETHOD(device_shutdown,      netvsc_shutdown),
 
         { 0, 0 }
 };
 
 static driver_t netvsc_driver = {
         NETVSC_DEVNAME,
         netvsc_methods,
         sizeof(struct hn_softc)
 };
 
 static devclass_t netvsc_devclass;
 
 DRIVER_MODULE(hn, vmbus, netvsc_driver, netvsc_devclass, 0, 0);
 MODULE_VERSION(hn, 1);
 MODULE_DEPEND(hn, vmbus, 1, 1, 1);
Index: head/sys/dev/hyperv/netvsc/hv_rndis_filter.c
===================================================================
--- head/sys/dev/hyperv/netvsc/hv_rndis_filter.c	(revision 307989)
+++ head/sys/dev/hyperv/netvsc/hv_rndis_filter.c	(revision 307990)
@@ -1,1357 +1,1361 @@
 /*-
  * Copyright (c) 2009-2012,2016 Microsoft Corp.
  * Copyright (c) 2010-2012 Citrix Inc.
  * Copyright (c) 2012 NetApp Inc.
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions
  * are met:
  * 1. Redistributions of source code must retain the above copyright
  *    notice unmodified, this list of conditions, and the following
  *    disclaimer.
  * 2. Redistributions in binary form must reproduce the above copyright
  *    notice, this list of conditions and the following disclaimer in the
  *    documentation and/or other materials provided with the distribution.
  *
  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */
 
 #include <sys/cdefs.h>
 __FBSDID("$FreeBSD$");
 
 #include <sys/param.h>
 #include <sys/mbuf.h>
 #include <sys/socket.h>
 #include <sys/lock.h>
 #include <sys/mutex.h>
+#include <sys/taskqueue.h>
+
+#include <machine/atomic.h>
+
+#include <net/ethernet.h>
 #include <net/if.h>
-#include <net/if_arp.h>
 #include <net/if_var.h>
-#include <net/ethernet.h>
+#include <net/if_media.h>
 #include <net/rndis.h>
+
 #include <netinet/in.h>
 #include <netinet/ip.h>
-#include <sys/types.h>
-#include <machine/atomic.h>
-#include <sys/sema.h>
-#include <vm/vm.h>
-#include <vm/vm_param.h>
-#include <vm/pmap.h>
+#include <netinet/tcp_lro.h>
 
 #include <dev/hyperv/include/hyperv.h>
+#include <dev/hyperv/include/hyperv_busdma.h>
+#include <dev/hyperv/include/vmbus.h>
 #include <dev/hyperv/include/vmbus_xact.h>
+
+#include <dev/hyperv/netvsc/ndis.h>
+#include <dev/hyperv/netvsc/if_hnreg.h>
+#include <dev/hyperv/netvsc/if_hnvar.h>
 #include <dev/hyperv/netvsc/hv_net_vsc.h>
 #include <dev/hyperv/netvsc/hv_rndis_filter.h>
-#include <dev/hyperv/netvsc/if_hnreg.h>
-#include <dev/hyperv/netvsc/ndis.h>
 
 #define HV_RF_RECVINFO_VLAN	0x1
 #define HV_RF_RECVINFO_CSUM	0x2
 #define HV_RF_RECVINFO_HASHINF	0x4
 #define HV_RF_RECVINFO_HASHVAL	0x8
 #define HV_RF_RECVINFO_ALL		\
 	(HV_RF_RECVINFO_VLAN |		\
 	 HV_RF_RECVINFO_CSUM |		\
 	 HV_RF_RECVINFO_HASHINF |	\
 	 HV_RF_RECVINFO_HASHVAL)
 
 #define HN_RNDIS_RID_COMPAT_MASK	0xffff
 #define HN_RNDIS_RID_COMPAT_MAX		HN_RNDIS_RID_COMPAT_MASK
 
 #define HN_RNDIS_XFER_SIZE		2048
 
 #define HN_NDIS_TXCSUM_CAP_IP4		\
 	(NDIS_TXCSUM_CAP_IP4 | NDIS_TXCSUM_CAP_IP4OPT)
 #define HN_NDIS_TXCSUM_CAP_TCP4		\
 	(NDIS_TXCSUM_CAP_TCP4 | NDIS_TXCSUM_CAP_TCP4OPT)
 #define HN_NDIS_TXCSUM_CAP_TCP6		\
 	(NDIS_TXCSUM_CAP_TCP6 | NDIS_TXCSUM_CAP_TCP6OPT | \
 	 NDIS_TXCSUM_CAP_IP6EXT)
 #define HN_NDIS_TXCSUM_CAP_UDP6		\
 	(NDIS_TXCSUM_CAP_UDP6 | NDIS_TXCSUM_CAP_IP6EXT)
 #define HN_NDIS_LSOV2_CAP_IP6		\
 	(NDIS_LSOV2_CAP_IP6EXT | NDIS_LSOV2_CAP_TCP6OPT)
 
 /*
  * Forward declarations
  */
 static void hv_rf_receive_indicate_status(struct hn_softc *sc,
     const void *data, int dlen);
 static void hv_rf_receive_data(struct hn_rx_ring *rxr,
     const void *data, int dlen);
 
 static int hn_rndis_query(struct hn_softc *sc, uint32_t oid,
     const void *idata, size_t idlen, void *odata, size_t *odlen0);
 static int hn_rndis_query2(struct hn_softc *sc, uint32_t oid,
     const void *idata, size_t idlen, void *odata, size_t *odlen0,
     size_t min_odlen);
 static int hn_rndis_set(struct hn_softc *sc, uint32_t oid, const void *data,
     size_t dlen);
 static int hn_rndis_conf_offload(struct hn_softc *sc, int mtu);
 static int hn_rndis_query_hwcaps(struct hn_softc *sc,
     struct ndis_offload *caps);
 
 static __inline uint32_t
 hn_rndis_rid(struct hn_softc *sc)
 {
 	uint32_t rid;
 
 again:
 	rid = atomic_fetchadd_int(&sc->hn_rndis_rid, 1);
 	if (rid == 0)
 		goto again;
 
 	/* Use upper 16 bits for non-compat RNDIS messages. */
 	return ((rid & 0xffff) << 16);
 }
 
 void *
 hn_rndis_pktinfo_append(struct rndis_packet_msg *pkt, size_t pktsize,
     size_t pi_dlen, uint32_t pi_type)
 {
 	const size_t pi_size = HN_RNDIS_PKTINFO_SIZE(pi_dlen);
 	struct rndis_pktinfo *pi;
 
 	KASSERT((pi_size & RNDIS_PACKET_MSG_OFFSET_ALIGNMASK) == 0,
 	    ("unaligned pktinfo size %zu, pktinfo dlen %zu", pi_size, pi_dlen));
 
 	/*
 	 * Per-packet-info does not move; it only grows.
 	 *
 	 * NOTE:
 	 * rm_pktinfooffset in this phase counts from the beginning
 	 * of rndis_packet_msg.
 	 */
 	KASSERT(pkt->rm_pktinfooffset + pkt->rm_pktinfolen + pi_size <= pktsize,
 	    ("%u pktinfo overflows RNDIS packet msg", pi_type));
 	pi = (struct rndis_pktinfo *)((uint8_t *)pkt + pkt->rm_pktinfooffset +
 	    pkt->rm_pktinfolen);
 	pkt->rm_pktinfolen += pi_size;
 
 	pi->rm_size = pi_size;
 	pi->rm_type = pi_type;
 	pi->rm_pktinfooffset = RNDIS_PKTINFO_OFFSET;
 
 	/* Data immediately follow per-packet-info. */
 	pkt->rm_dataoffset += pi_size;
 
 	/* Update RNDIS packet msg length */
 	pkt->rm_len += pi_size;
 
 	return (pi->rm_data);
 }
 
 /*
  * RNDIS filter receive indicate status
  */
 static void 
 hv_rf_receive_indicate_status(struct hn_softc *sc, const void *data, int dlen)
 {
 	const struct rndis_status_msg *msg;
 	int ofs;
 
 	if (dlen < sizeof(*msg)) {
 		if_printf(sc->hn_ifp, "invalid RNDIS status\n");
 		return;
 	}
 	msg = data;
 
 	switch (msg->rm_status) {
 	case RNDIS_STATUS_MEDIA_CONNECT:
 	case RNDIS_STATUS_MEDIA_DISCONNECT:
 		hn_link_status_update(sc);
 		break;
 
 	case RNDIS_STATUS_TASK_OFFLOAD_CURRENT_CONFIG:
 		/* Not really useful; ignore. */
 		break;
 
 	case RNDIS_STATUS_NETWORK_CHANGE:
 		ofs = RNDIS_STBUFOFFSET_ABS(msg->rm_stbufoffset);
 		if (dlen < ofs + msg->rm_stbuflen ||
 		    msg->rm_stbuflen < sizeof(uint32_t)) {
 			if_printf(sc->hn_ifp, "network changed\n");
 		} else {
 			uint32_t change;
 
 			memcpy(&change, ((const uint8_t *)msg) + ofs,
 			    sizeof(change));
 			if_printf(sc->hn_ifp, "network changed, change %u\n",
 			    change);
 		}
 		hn_network_change(sc);
 		break;
 
 	default:
 		/* TODO: */
 		if_printf(sc->hn_ifp, "unknown RNDIS status 0x%08x\n",
 		    msg->rm_status);
 		break;
 	}
 }
 
 static int
 hn_rndis_rxinfo(const void *info_data, int info_dlen, struct hn_recvinfo *info)
 {
 	const struct rndis_pktinfo *pi = info_data;
 	uint32_t mask = 0;
 
 	while (info_dlen != 0) {
 		const void *data;
 		uint32_t dlen;
 
 		if (__predict_false(info_dlen < sizeof(*pi)))
 			return (EINVAL);
 		if (__predict_false(info_dlen < pi->rm_size))
 			return (EINVAL);
 		info_dlen -= pi->rm_size;
 
 		if (__predict_false(pi->rm_size & RNDIS_PKTINFO_SIZE_ALIGNMASK))
 			return (EINVAL);
 		if (__predict_false(pi->rm_size < pi->rm_pktinfooffset))
 			return (EINVAL);
 		dlen = pi->rm_size - pi->rm_pktinfooffset;
 		data = pi->rm_data;
 
 		switch (pi->rm_type) {
 		case NDIS_PKTINFO_TYPE_VLAN:
 			if (__predict_false(dlen < NDIS_VLAN_INFO_SIZE))
 				return (EINVAL);
 			info->vlan_info = *((const uint32_t *)data);
 			mask |= HV_RF_RECVINFO_VLAN;
 			break;
 
 		case NDIS_PKTINFO_TYPE_CSUM:
 			if (__predict_false(dlen < NDIS_RXCSUM_INFO_SIZE))
 				return (EINVAL);
 			info->csum_info = *((const uint32_t *)data);
 			mask |= HV_RF_RECVINFO_CSUM;
 			break;
 
 		case HN_NDIS_PKTINFO_TYPE_HASHVAL:
 			if (__predict_false(dlen < HN_NDIS_HASH_VALUE_SIZE))
 				return (EINVAL);
 			info->hash_value = *((const uint32_t *)data);
 			mask |= HV_RF_RECVINFO_HASHVAL;
 			break;
 
 		case HN_NDIS_PKTINFO_TYPE_HASHINF:
 			if (__predict_false(dlen < HN_NDIS_HASH_INFO_SIZE))
 				return (EINVAL);
 			info->hash_info = *((const uint32_t *)data);
 			mask |= HV_RF_RECVINFO_HASHINF;
 			break;
 
 		default:
 			goto next;
 		}
 
 		if (mask == HV_RF_RECVINFO_ALL) {
 			/* All found; done */
 			break;
 		}
 next:
 		pi = (const struct rndis_pktinfo *)
 		    ((const uint8_t *)pi + pi->rm_size);
 	}
 
 	/*
 	 * Final fixup.
 	 * - If there is no hash value, invalidate the hash info.
 	 */
 	if ((mask & HV_RF_RECVINFO_HASHVAL) == 0)
 		info->hash_info = HN_NDIS_HASH_INFO_INVALID;
 	return (0);
 }
 
 static __inline bool
 hn_rndis_check_overlap(int off, int len, int check_off, int check_len)
 {
 
 	if (off < check_off) {
 		if (__predict_true(off + len <= check_off))
 			return (false);
 	} else if (off > check_off) {
 		if (__predict_true(check_off + check_len <= off))
 			return (false);
 	}
 	return (true);
 }
 
 /*
  * RNDIS filter receive data
  */
 static void
 hv_rf_receive_data(struct hn_rx_ring *rxr, const void *data, int dlen)
 {
 	const struct rndis_packet_msg *pkt;
 	struct hn_recvinfo info;
 	int data_off, pktinfo_off, data_len, pktinfo_len;
 
 	/*
 	 * Check length.
 	 */
 	if (__predict_false(dlen < sizeof(*pkt))) {
 		if_printf(rxr->hn_ifp, "invalid RNDIS packet msg\n");
 		return;
 	}
 	pkt = data;
 
 	if (__predict_false(dlen < pkt->rm_len)) {
 		if_printf(rxr->hn_ifp, "truncated RNDIS packet msg, "
 		    "dlen %d, msglen %u\n", dlen, pkt->rm_len);
 		return;
 	}
 	if (__predict_false(pkt->rm_len <
 	    pkt->rm_datalen + pkt->rm_oobdatalen + pkt->rm_pktinfolen)) {
 		if_printf(rxr->hn_ifp, "invalid RNDIS packet msglen, "
 		    "msglen %u, data %u, oob %u, pktinfo %u\n",
 		    pkt->rm_len, pkt->rm_datalen, pkt->rm_oobdatalen,
 		    pkt->rm_pktinfolen);
 		return;
 	}
 	if (__predict_false(pkt->rm_datalen == 0)) {
 		if_printf(rxr->hn_ifp, "invalid RNDIS packet msg, no data\n");
 		return;
 	}
 
 	/*
 	 * Check offests.
 	 */
 #define IS_OFFSET_INVALID(ofs)			\
 	((ofs) < RNDIS_PACKET_MSG_OFFSET_MIN ||	\
 	 ((ofs) & RNDIS_PACKET_MSG_OFFSET_ALIGNMASK))
 
 	/* XXX Hyper-V does not meet data offset alignment requirement */
 	if (__predict_false(pkt->rm_dataoffset < RNDIS_PACKET_MSG_OFFSET_MIN)) {
 		if_printf(rxr->hn_ifp, "invalid RNDIS packet msg, "
 		    "data offset %u\n", pkt->rm_dataoffset);
 		return;
 	}
 	if (__predict_false(pkt->rm_oobdataoffset > 0 &&
 	    IS_OFFSET_INVALID(pkt->rm_oobdataoffset))) {
 		if_printf(rxr->hn_ifp, "invalid RNDIS packet msg, "
 		    "oob offset %u\n", pkt->rm_oobdataoffset);
 		return;
 	}
 	if (__predict_true(pkt->rm_pktinfooffset > 0) &&
 	    __predict_false(IS_OFFSET_INVALID(pkt->rm_pktinfooffset))) {
 		if_printf(rxr->hn_ifp, "invalid RNDIS packet msg, "
 		    "pktinfo offset %u\n", pkt->rm_pktinfooffset);
 		return;
 	}
 
 #undef IS_OFFSET_INVALID
 
 	data_off = RNDIS_PACKET_MSG_OFFSET_ABS(pkt->rm_dataoffset);
 	data_len = pkt->rm_datalen;
 	pktinfo_off = RNDIS_PACKET_MSG_OFFSET_ABS(pkt->rm_pktinfooffset);
 	pktinfo_len = pkt->rm_pktinfolen;
 
 	/*
 	 * Check OOB coverage.
 	 */
 	if (__predict_false(pkt->rm_oobdatalen != 0)) {
 		int oob_off, oob_len;
 
 		if_printf(rxr->hn_ifp, "got oobdata\n");
 		oob_off = RNDIS_PACKET_MSG_OFFSET_ABS(pkt->rm_oobdataoffset);
 		oob_len = pkt->rm_oobdatalen;
 
 		if (__predict_false(oob_off + oob_len > pkt->rm_len)) {
 			if_printf(rxr->hn_ifp, "invalid RNDIS packet msg, "
 			    "oob overflow, msglen %u, oob abs %d len %d\n",
 			    pkt->rm_len, oob_off, oob_len);
 			return;
 		}
 
 		/*
 		 * Check against data.
 		 */
 		if (hn_rndis_check_overlap(oob_off, oob_len,
 		    data_off, data_len)) {
 			if_printf(rxr->hn_ifp, "invalid RNDIS packet msg, "
 			    "oob overlaps data, oob abs %d len %d, "
 			    "data abs %d len %d\n",
 			    oob_off, oob_len, data_off, data_len);
 			return;
 		}
 
 		/*
 		 * Check against pktinfo.
 		 */
 		if (pktinfo_len != 0 &&
 		    hn_rndis_check_overlap(oob_off, oob_len,
 		    pktinfo_off, pktinfo_len)) {
 			if_printf(rxr->hn_ifp, "invalid RNDIS packet msg, "
 			    "oob overlaps pktinfo, oob abs %d len %d, "
 			    "pktinfo abs %d len %d\n",
 			    oob_off, oob_len, pktinfo_off, pktinfo_len);
 			return;
 		}
 	}
 
 	/*
 	 * Check per-packet-info coverage and find useful per-packet-info.
 	 */
 	info.vlan_info = HN_NDIS_VLAN_INFO_INVALID;
 	info.csum_info = HN_NDIS_RXCSUM_INFO_INVALID;
 	info.hash_info = HN_NDIS_HASH_INFO_INVALID;
 	if (__predict_true(pktinfo_len != 0)) {
 		bool overlap;
 		int error;
 
 		if (__predict_false(pktinfo_off + pktinfo_len > pkt->rm_len)) {
 			if_printf(rxr->hn_ifp, "invalid RNDIS packet msg, "
 			    "pktinfo overflow, msglen %u, "
 			    "pktinfo abs %d len %d\n",
 			    pkt->rm_len, pktinfo_off, pktinfo_len);
 			return;
 		}
 
 		/*
 		 * Check packet info coverage.
 		 */
 		overlap = hn_rndis_check_overlap(pktinfo_off, pktinfo_len,
 		    data_off, data_len);
 		if (__predict_false(overlap)) {
 			if_printf(rxr->hn_ifp, "invalid RNDIS packet msg, "
 			    "pktinfo overlap data, pktinfo abs %d len %d, "
 			    "data abs %d len %d\n",
 			    pktinfo_off, pktinfo_len, data_off, data_len);
 			return;
 		}
 
 		/*
 		 * Find useful per-packet-info.
 		 */
 		error = hn_rndis_rxinfo(((const uint8_t *)pkt) + pktinfo_off,
 		    pktinfo_len, &info);
 		if (__predict_false(error)) {
 			if_printf(rxr->hn_ifp, "invalid RNDIS packet msg "
 			    "pktinfo\n");
 			return;
 		}
 	}
 
 	if (__predict_false(data_off + data_len > pkt->rm_len)) {
 		if_printf(rxr->hn_ifp, "invalid RNDIS packet msg, "
 		    "data overflow, msglen %u, data abs %d len %d\n",
 		    pkt->rm_len, data_off, data_len);
 		return;
 	}
 	hn_rxpkt(rxr, ((const uint8_t *)pkt) + data_off, data_len, &info);
 }
 
 /*
  * RNDIS filter on receive
  */
 void
 hv_rf_on_receive(struct hn_softc *sc, struct hn_rx_ring *rxr,
     const void *data, int dlen)
 {
 	const struct rndis_comp_hdr *comp;
 	const struct rndis_msghdr *hdr;
 
 	if (__predict_false(dlen < sizeof(*hdr))) {
 		if_printf(rxr->hn_ifp, "invalid RNDIS msg\n");
 		return;
 	}
 	hdr = data;
 
 	switch (hdr->rm_type) {
 	case REMOTE_NDIS_PACKET_MSG:
 		hv_rf_receive_data(rxr, data, dlen);
 		break;
 
 	case REMOTE_NDIS_INITIALIZE_CMPLT:
 	case REMOTE_NDIS_QUERY_CMPLT:
 	case REMOTE_NDIS_SET_CMPLT:
 	case REMOTE_NDIS_KEEPALIVE_CMPLT:	/* unused */
 		if (dlen < sizeof(*comp)) {
 			if_printf(rxr->hn_ifp, "invalid RNDIS cmplt\n");
 			return;
 		}
 		comp = data;
 
 		KASSERT(comp->rm_rid > HN_RNDIS_RID_COMPAT_MAX,
 		    ("invalid RNDIS rid 0x%08x\n", comp->rm_rid));
 		vmbus_xact_ctx_wakeup(sc->hn_xact, comp, dlen);
 		break;
 
 	case REMOTE_NDIS_INDICATE_STATUS_MSG:
 		hv_rf_receive_indicate_status(sc, data, dlen);
 		break;
 
 	case REMOTE_NDIS_RESET_CMPLT:
 		/*
 		 * Reset completed, no rid.
 		 *
 		 * NOTE:
 		 * RESET is not issued by hn(4), so this message should
 		 * _not_ be observed.
 		 */
 		if_printf(rxr->hn_ifp, "RESET cmplt received\n");
 		break;
 
 	default:
 		if_printf(rxr->hn_ifp, "unknown RNDIS msg 0x%x\n",
 		    hdr->rm_type);
 		break;
 	}
 }
 
 int
 hn_rndis_get_eaddr(struct hn_softc *sc, uint8_t *eaddr)
 {
 	size_t eaddr_len;
 	int error;
 
 	eaddr_len = ETHER_ADDR_LEN;
 	error = hn_rndis_query(sc, OID_802_3_PERMANENT_ADDRESS, NULL, 0,
 	    eaddr, &eaddr_len);
 	if (error)
 		return (error);
 	if (eaddr_len != ETHER_ADDR_LEN) {
 		if_printf(sc->hn_ifp, "invalid eaddr len %zu\n", eaddr_len);
 		return (EINVAL);
 	}
 	return (0);
 }
 
 int
 hn_rndis_get_linkstatus(struct hn_softc *sc, uint32_t *link_status)
 {
 	size_t size;
 	int error;
 
 	size = sizeof(*link_status);
 	error = hn_rndis_query(sc, OID_GEN_MEDIA_CONNECT_STATUS, NULL, 0,
 	    link_status, &size);
 	if (error)
 		return (error);
 	if (size != sizeof(uint32_t)) {
 		if_printf(sc->hn_ifp, "invalid link status len %zu\n", size);
 		return (EINVAL);
 	}
 	return (0);
 }
 
 static const void *
 hn_rndis_xact_exec1(struct hn_softc *sc, struct vmbus_xact *xact, size_t reqlen,
-    struct hn_send_ctx *sndc, size_t *comp_len)
+    struct hn_nvs_sendctx *sndc, size_t *comp_len)
 {
 	struct vmbus_gpa gpa[HN_XACT_REQ_PGCNT];
 	int gpa_cnt, error;
 	bus_addr_t paddr;
 
 	KASSERT(reqlen <= HN_XACT_REQ_SIZE && reqlen > 0,
 	    ("invalid request length %zu", reqlen));
 
 	/*
 	 * Setup the SG list.
 	 */
 	paddr = vmbus_xact_req_paddr(xact);
 	KASSERT((paddr & PAGE_MASK) == 0,
 	    ("vmbus xact request is not page aligned 0x%jx", (uintmax_t)paddr));
 	for (gpa_cnt = 0; gpa_cnt < HN_XACT_REQ_PGCNT; ++gpa_cnt) {
 		int len = PAGE_SIZE;
 
 		if (reqlen == 0)
 			break;
 		if (reqlen < len)
 			len = reqlen;
 
 		gpa[gpa_cnt].gpa_page = atop(paddr) + gpa_cnt;
 		gpa[gpa_cnt].gpa_len = len;
 		gpa[gpa_cnt].gpa_ofs = 0;
 
 		reqlen -= len;
 	}
 	KASSERT(reqlen == 0, ("still have %zu request data left", reqlen));
 
 	/*
 	 * Send this RNDIS control message and wait for its completion
 	 * message.
 	 */
 	vmbus_xact_activate(xact);
 	error = hn_nvs_send_rndis_ctrl(sc->hn_prichan, sndc, gpa, gpa_cnt);
 	if (error) {
 		vmbus_xact_deactivate(xact);
 		if_printf(sc->hn_ifp, "RNDIS ctrl send failed: %d\n", error);
 		return (NULL);
 	}
 	return (vmbus_xact_wait(xact, comp_len));
 }
 
 static const void *
 hn_rndis_xact_execute(struct hn_softc *sc, struct vmbus_xact *xact, uint32_t rid,
     size_t reqlen, size_t *comp_len0, uint32_t comp_type)
 {
 	const struct rndis_comp_hdr *comp;
 	size_t comp_len, min_complen = *comp_len0;
 
 	KASSERT(rid > HN_RNDIS_RID_COMPAT_MAX, ("invalid rid %u\n", rid));
 	KASSERT(min_complen >= sizeof(*comp),
 	    ("invalid minimum complete len %zu", min_complen));
 
 	/*
 	 * Execute the xact setup by the caller.
 	 */
-	comp = hn_rndis_xact_exec1(sc, xact, reqlen, &hn_send_ctx_none,
+	comp = hn_rndis_xact_exec1(sc, xact, reqlen, &hn_nvs_sendctx_none,
 	    &comp_len);
 	if (comp == NULL)
 		return (NULL);
 
 	/*
 	 * Check this RNDIS complete message.
 	 */
 	if (comp_len < min_complen) {
 		if (comp_len >= sizeof(*comp)) {
 			/* rm_status field is valid */
 			if_printf(sc->hn_ifp, "invalid RNDIS comp len %zu, "
 			    "status 0x%08x\n", comp_len, comp->rm_status);
 		} else {
 			if_printf(sc->hn_ifp, "invalid RNDIS comp len %zu\n",
 			    comp_len);
 		}
 		return (NULL);
 	}
 	if (comp->rm_len < min_complen) {
 		if_printf(sc->hn_ifp, "invalid RNDIS comp msglen %u\n",
 		    comp->rm_len);
 		return (NULL);
 	}
 	if (comp->rm_type != comp_type) {
 		if_printf(sc->hn_ifp, "unexpected RNDIS comp 0x%08x, "
 		    "expect 0x%08x\n", comp->rm_type, comp_type);
 		return (NULL);
 	}
 	if (comp->rm_rid != rid) {
 		if_printf(sc->hn_ifp, "RNDIS comp rid mismatch %u, "
 		    "expect %u\n", comp->rm_rid, rid);
 		return (NULL);
 	}
 	/* All pass! */
 	*comp_len0 = comp_len;
 	return (comp);
 }
 
 static int
 hn_rndis_query(struct hn_softc *sc, uint32_t oid,
     const void *idata, size_t idlen, void *odata, size_t *odlen0)
 {
 
 	return (hn_rndis_query2(sc, oid, idata, idlen, odata, odlen0, *odlen0));
 }
 
 static int
 hn_rndis_query2(struct hn_softc *sc, uint32_t oid,
     const void *idata, size_t idlen, void *odata, size_t *odlen0,
     size_t min_odlen)
 {
 	struct rndis_query_req *req;
 	const struct rndis_query_comp *comp;
 	struct vmbus_xact *xact;
 	size_t reqlen, odlen = *odlen0, comp_len;
 	int error, ofs;
 	uint32_t rid;
 
 	reqlen = sizeof(*req) + idlen;
 	xact = vmbus_xact_get(sc->hn_xact, reqlen);
 	if (xact == NULL) {
 		if_printf(sc->hn_ifp, "no xact for RNDIS query 0x%08x\n", oid);
 		return (ENXIO);
 	}
 	rid = hn_rndis_rid(sc);
 	req = vmbus_xact_req_data(xact);
 	req->rm_type = REMOTE_NDIS_QUERY_MSG;
 	req->rm_len = reqlen;
 	req->rm_rid = rid;
 	req->rm_oid = oid;
 	/*
 	 * XXX
 	 * This is _not_ RNDIS Spec conforming:
 	 * "This MUST be set to 0 when there is no input data
 	 *  associated with the OID."
 	 *
 	 * If this field was set to 0 according to the RNDIS Spec,
 	 * Hyper-V would set non-SUCCESS status in the query
 	 * completion.
 	 */
 	req->rm_infobufoffset = RNDIS_QUERY_REQ_INFOBUFOFFSET;
 
 	if (idlen > 0) {
 		req->rm_infobuflen = idlen;
 		/* Input data immediately follows RNDIS query. */
 		memcpy(req + 1, idata, idlen);
 	}
 
 	comp_len = sizeof(*comp) + min_odlen;
 	comp = hn_rndis_xact_execute(sc, xact, rid, reqlen, &comp_len,
 	    REMOTE_NDIS_QUERY_CMPLT);
 	if (comp == NULL) {
 		if_printf(sc->hn_ifp, "exec RNDIS query 0x%08x failed\n", oid);
 		error = EIO;
 		goto done;
 	}
 
 	if (comp->rm_status != RNDIS_STATUS_SUCCESS) {
 		if_printf(sc->hn_ifp, "RNDIS query 0x%08x failed: "
 		    "status 0x%08x\n", oid, comp->rm_status);
 		error = EIO;
 		goto done;
 	}
 	if (comp->rm_infobuflen == 0 || comp->rm_infobufoffset == 0) {
 		/* No output data! */
 		if_printf(sc->hn_ifp, "RNDIS query 0x%08x, no data\n", oid);
 		*odlen0 = 0;
 		error = 0;
 		goto done;
 	}
 
 	/*
 	 * Check output data length and offset.
 	 */
 	/* ofs is the offset from the beginning of comp. */
 	ofs = RNDIS_QUERY_COMP_INFOBUFOFFSET_ABS(comp->rm_infobufoffset);
 	if (ofs < sizeof(*comp) || ofs + comp->rm_infobuflen > comp_len) {
 		if_printf(sc->hn_ifp, "RNDIS query invalid comp ib off/len, "
 		    "%u/%u\n", comp->rm_infobufoffset, comp->rm_infobuflen);
 		error = EINVAL;
 		goto done;
 	}
 
 	/*
 	 * Save output data.
 	 */
 	if (comp->rm_infobuflen < odlen)
 		odlen = comp->rm_infobuflen;
 	memcpy(odata, ((const uint8_t *)comp) + ofs, odlen);
 	*odlen0 = odlen;
 
 	error = 0;
 done:
 	vmbus_xact_put(xact);
 	return (error);
 }
 
 int
 hn_rndis_query_rsscaps(struct hn_softc *sc, int *rxr_cnt0)
 {
 	struct ndis_rss_caps in, caps;
 	size_t caps_len;
 	int error, indsz, rxr_cnt, hash_fnidx;
 	uint32_t hash_func = 0, hash_types = 0;
 
 	*rxr_cnt0 = 0;
 
 	if (sc->hn_ndis_ver < HN_NDIS_VERSION_6_20)
 		return (EOPNOTSUPP);
 
 	memset(&in, 0, sizeof(in));
 	in.ndis_hdr.ndis_type = NDIS_OBJTYPE_RSS_CAPS;
 	in.ndis_hdr.ndis_rev = NDIS_RSS_CAPS_REV_2;
 	in.ndis_hdr.ndis_size = NDIS_RSS_CAPS_SIZE;
 
 	caps_len = NDIS_RSS_CAPS_SIZE;
 	error = hn_rndis_query2(sc, OID_GEN_RECEIVE_SCALE_CAPABILITIES,
 	    &in, NDIS_RSS_CAPS_SIZE, &caps, &caps_len, NDIS_RSS_CAPS_SIZE_6_0);
 	if (error)
 		return (error);
 
 	/*
 	 * Preliminary verification.
 	 */
 	if (caps.ndis_hdr.ndis_type != NDIS_OBJTYPE_RSS_CAPS) {
 		if_printf(sc->hn_ifp, "invalid NDIS objtype 0x%02x\n",
 		    caps.ndis_hdr.ndis_type);
 		return (EINVAL);
 	}
 	if (caps.ndis_hdr.ndis_rev < NDIS_RSS_CAPS_REV_1) {
 		if_printf(sc->hn_ifp, "invalid NDIS objrev 0x%02x\n",
 		    caps.ndis_hdr.ndis_rev);
 		return (EINVAL);
 	}
 	if (caps.ndis_hdr.ndis_size > caps_len) {
 		if_printf(sc->hn_ifp, "invalid NDIS objsize %u, "
 		    "data size %zu\n", caps.ndis_hdr.ndis_size, caps_len);
 		return (EINVAL);
 	} else if (caps.ndis_hdr.ndis_size < NDIS_RSS_CAPS_SIZE_6_0) {
 		if_printf(sc->hn_ifp, "invalid NDIS objsize %u\n",
 		    caps.ndis_hdr.ndis_size);
 		return (EINVAL);
 	}
 
 	/*
 	 * Save information for later RSS configuration.
 	 */
 	if (caps.ndis_nrxr == 0) {
 		if_printf(sc->hn_ifp, "0 RX rings!?\n");
 		return (EINVAL);
 	}
 	if (bootverbose)
 		if_printf(sc->hn_ifp, "%u RX rings\n", caps.ndis_nrxr);
 	rxr_cnt = caps.ndis_nrxr;
 
 	if (caps.ndis_hdr.ndis_size == NDIS_RSS_CAPS_SIZE &&
 	    caps.ndis_hdr.ndis_rev >= NDIS_RSS_CAPS_REV_2) {
 		if (caps.ndis_nind > NDIS_HASH_INDCNT) {
 			if_printf(sc->hn_ifp,
 			    "too many RSS indirect table entries %u\n",
 			    caps.ndis_nind);
 			return (EOPNOTSUPP);
 		}
 		if (!powerof2(caps.ndis_nind)) {
 			if_printf(sc->hn_ifp, "RSS indirect table size is not "
 			    "power-of-2 %u\n", caps.ndis_nind);
 		}
 
 		if (bootverbose) {
 			if_printf(sc->hn_ifp, "RSS indirect table size %u\n",
 			    caps.ndis_nind);
 		}
 		indsz = caps.ndis_nind;
 	} else {
 		indsz = NDIS_HASH_INDCNT;
 	}
 	if (indsz < rxr_cnt) {
 		if_printf(sc->hn_ifp, "# of RX rings (%d) > "
 		    "RSS indirect table size %d\n", rxr_cnt, indsz);
 		rxr_cnt = indsz;
 	}
 
 	/*
 	 * NOTE:
 	 * Toeplitz is at the lowest bit, and it is prefered; so ffs(),
 	 * instead of fls(), is used here.
 	 */
 	hash_fnidx = ffs(caps.ndis_caps & NDIS_RSS_CAP_HASHFUNC_MASK);
 	if (hash_fnidx == 0) {
 		if_printf(sc->hn_ifp, "no hash functions, caps 0x%08x\n",
 		    caps.ndis_caps);
 		return (EOPNOTSUPP);
 	}
 	hash_func = 1 << (hash_fnidx - 1); /* ffs is 1-based */
 
 	if (caps.ndis_caps & NDIS_RSS_CAP_IPV4)
 		hash_types |= NDIS_HASH_IPV4 | NDIS_HASH_TCP_IPV4;
 	if (caps.ndis_caps & NDIS_RSS_CAP_IPV6)
 		hash_types |= NDIS_HASH_IPV6 | NDIS_HASH_TCP_IPV6;
 	if (caps.ndis_caps & NDIS_RSS_CAP_IPV6_EX)
 		hash_types |= NDIS_HASH_IPV6_EX | NDIS_HASH_TCP_IPV6_EX;
 	if (hash_types == 0) {
 		if_printf(sc->hn_ifp, "no hash types, caps 0x%08x\n",
 		    caps.ndis_caps);
 		return (EOPNOTSUPP);
 	}
 
 	/* Commit! */
 	sc->hn_rss_ind_size = indsz;
 	sc->hn_rss_hash = hash_func | hash_types;
 	*rxr_cnt0 = rxr_cnt;
 	return (0);
 }
 
 static int
 hn_rndis_set(struct hn_softc *sc, uint32_t oid, const void *data, size_t dlen)
 {
 	struct rndis_set_req *req;
 	const struct rndis_set_comp *comp;
 	struct vmbus_xact *xact;
 	size_t reqlen, comp_len;
 	uint32_t rid;
 	int error;
 
 	KASSERT(dlen > 0, ("invalid dlen %zu", dlen));
 
 	reqlen = sizeof(*req) + dlen;
 	xact = vmbus_xact_get(sc->hn_xact, reqlen);
 	if (xact == NULL) {
 		if_printf(sc->hn_ifp, "no xact for RNDIS set 0x%08x\n", oid);
 		return (ENXIO);
 	}
 	rid = hn_rndis_rid(sc);
 	req = vmbus_xact_req_data(xact);
 	req->rm_type = REMOTE_NDIS_SET_MSG;
 	req->rm_len = reqlen;
 	req->rm_rid = rid;
 	req->rm_oid = oid;
 	req->rm_infobuflen = dlen;
 	req->rm_infobufoffset = RNDIS_SET_REQ_INFOBUFOFFSET;
 	/* Data immediately follows RNDIS set. */
 	memcpy(req + 1, data, dlen);
 
 	comp_len = sizeof(*comp);
 	comp = hn_rndis_xact_execute(sc, xact, rid, reqlen, &comp_len,
 	    REMOTE_NDIS_SET_CMPLT);
 	if (comp == NULL) {
 		if_printf(sc->hn_ifp, "exec RNDIS set 0x%08x failed\n", oid);
 		error = EIO;
 		goto done;
 	}
 
 	if (comp->rm_status != RNDIS_STATUS_SUCCESS) {
 		if_printf(sc->hn_ifp, "RNDIS set 0x%08x failed: "
 		    "status 0x%08x\n", oid, comp->rm_status);
 		error = EIO;
 		goto done;
 	}
 	error = 0;
 done:
 	vmbus_xact_put(xact);
 	return (error);
 }
 
 static int
 hn_rndis_conf_offload(struct hn_softc *sc, int mtu)
 {
 	struct ndis_offload hwcaps;
 	struct ndis_offload_params params;
 	uint32_t caps = 0;
 	size_t paramsz;
 	int error, tso_maxsz, tso_minsg;
 
 	error = hn_rndis_query_hwcaps(sc, &hwcaps);
 	if (error) {
 		if_printf(sc->hn_ifp, "hwcaps query failed: %d\n", error);
 		return (error);
 	}
 
 	/* NOTE: 0 means "no change" */
 	memset(&params, 0, sizeof(params));
 
 	params.ndis_hdr.ndis_type = NDIS_OBJTYPE_DEFAULT;
 	if (sc->hn_ndis_ver < HN_NDIS_VERSION_6_30) {
 		params.ndis_hdr.ndis_rev = NDIS_OFFLOAD_PARAMS_REV_2;
 		paramsz = NDIS_OFFLOAD_PARAMS_SIZE_6_1;
 	} else {
 		params.ndis_hdr.ndis_rev = NDIS_OFFLOAD_PARAMS_REV_3;
 		paramsz = NDIS_OFFLOAD_PARAMS_SIZE;
 	}
 	params.ndis_hdr.ndis_size = paramsz;
 
 	/*
 	 * TSO4/TSO6 setup.
 	 */
 	tso_maxsz = IP_MAXPACKET;
 	tso_minsg = 2;
 	if (hwcaps.ndis_lsov2.ndis_ip4_encap & NDIS_OFFLOAD_ENCAP_8023) {
 		caps |= HN_CAP_TSO4;
 		params.ndis_lsov2_ip4 = NDIS_OFFLOAD_LSOV2_ON;
 
 		if (hwcaps.ndis_lsov2.ndis_ip4_maxsz < tso_maxsz)
 			tso_maxsz = hwcaps.ndis_lsov2.ndis_ip4_maxsz;
 		if (hwcaps.ndis_lsov2.ndis_ip4_minsg > tso_minsg)
 			tso_minsg = hwcaps.ndis_lsov2.ndis_ip4_minsg;
 	}
 	if ((hwcaps.ndis_lsov2.ndis_ip6_encap & NDIS_OFFLOAD_ENCAP_8023) &&
 	    (hwcaps.ndis_lsov2.ndis_ip6_opts & HN_NDIS_LSOV2_CAP_IP6) ==
 	    HN_NDIS_LSOV2_CAP_IP6) {
 #ifdef notyet
 		caps |= HN_CAP_TSO6;
 		params.ndis_lsov2_ip6 = NDIS_OFFLOAD_LSOV2_ON;
 
 		if (hwcaps.ndis_lsov2.ndis_ip6_maxsz < tso_maxsz)
 			tso_maxsz = hwcaps.ndis_lsov2.ndis_ip6_maxsz;
 		if (hwcaps.ndis_lsov2.ndis_ip6_minsg > tso_minsg)
 			tso_minsg = hwcaps.ndis_lsov2.ndis_ip6_minsg;
 #endif
 	}
 	sc->hn_ndis_tso_szmax = 0;
 	sc->hn_ndis_tso_sgmin = 0;
 	if (caps & (HN_CAP_TSO4 | HN_CAP_TSO6)) {
 		KASSERT(tso_maxsz <= IP_MAXPACKET,
 		    ("invalid NDIS TSO maxsz %d", tso_maxsz));
 		KASSERT(tso_minsg >= 2,
 		    ("invalid NDIS TSO minsg %d", tso_minsg));
 		if (tso_maxsz < tso_minsg * mtu) {
 			if_printf(sc->hn_ifp, "invalid NDIS TSO config: "
 			    "maxsz %d, minsg %d, mtu %d; "
 			    "disable TSO4 and TSO6\n",
 			    tso_maxsz, tso_minsg, mtu);
 			caps &= ~(HN_CAP_TSO4 | HN_CAP_TSO6);
 			params.ndis_lsov2_ip4 = NDIS_OFFLOAD_LSOV2_OFF;
 			params.ndis_lsov2_ip6 = NDIS_OFFLOAD_LSOV2_OFF;
 		} else {
 			sc->hn_ndis_tso_szmax = tso_maxsz;
 			sc->hn_ndis_tso_sgmin = tso_minsg;
 			if (bootverbose) {
 				if_printf(sc->hn_ifp, "NDIS TSO "
 				    "szmax %d sgmin %d\n",
 				    sc->hn_ndis_tso_szmax,
 				    sc->hn_ndis_tso_sgmin);
 			}
 		}
 	}
 
 	/* IPv4 checksum */
 	if ((hwcaps.ndis_csum.ndis_ip4_txcsum & HN_NDIS_TXCSUM_CAP_IP4) ==
 	    HN_NDIS_TXCSUM_CAP_IP4) {
 		caps |= HN_CAP_IPCS;
 		params.ndis_ip4csum = NDIS_OFFLOAD_PARAM_TX;
 	}
 	if (hwcaps.ndis_csum.ndis_ip4_rxcsum & NDIS_RXCSUM_CAP_IP4) {
 		if (params.ndis_ip4csum == NDIS_OFFLOAD_PARAM_TX)
 			params.ndis_ip4csum = NDIS_OFFLOAD_PARAM_TXRX;
 		else
 			params.ndis_ip4csum = NDIS_OFFLOAD_PARAM_RX;
 	}
 
 	/* TCP4 checksum */
 	if ((hwcaps.ndis_csum.ndis_ip4_txcsum & HN_NDIS_TXCSUM_CAP_TCP4) ==
 	    HN_NDIS_TXCSUM_CAP_TCP4) {
 		caps |= HN_CAP_TCP4CS;
 		params.ndis_tcp4csum = NDIS_OFFLOAD_PARAM_TX;
 	}
 	if (hwcaps.ndis_csum.ndis_ip4_rxcsum & NDIS_RXCSUM_CAP_TCP4) {
 		if (params.ndis_tcp4csum == NDIS_OFFLOAD_PARAM_TX)
 			params.ndis_tcp4csum = NDIS_OFFLOAD_PARAM_TXRX;
 		else
 			params.ndis_tcp4csum = NDIS_OFFLOAD_PARAM_RX;
 	}
 
 	/* UDP4 checksum */
 	if (hwcaps.ndis_csum.ndis_ip4_txcsum & NDIS_TXCSUM_CAP_UDP4) {
 		caps |= HN_CAP_UDP4CS;
 		params.ndis_udp4csum = NDIS_OFFLOAD_PARAM_TX;
 	}
 	if (hwcaps.ndis_csum.ndis_ip4_rxcsum & NDIS_RXCSUM_CAP_UDP4) {
 		if (params.ndis_udp4csum == NDIS_OFFLOAD_PARAM_TX)
 			params.ndis_udp4csum = NDIS_OFFLOAD_PARAM_TXRX;
 		else
 			params.ndis_udp4csum = NDIS_OFFLOAD_PARAM_RX;
 	}
 
 	/* TCP6 checksum */
 	if ((hwcaps.ndis_csum.ndis_ip6_txcsum & HN_NDIS_TXCSUM_CAP_TCP6) ==
 	    HN_NDIS_TXCSUM_CAP_TCP6) {
 		caps |= HN_CAP_TCP6CS;
 		params.ndis_tcp6csum = NDIS_OFFLOAD_PARAM_TX;
 	}
 	if (hwcaps.ndis_csum.ndis_ip6_rxcsum & NDIS_RXCSUM_CAP_TCP6) {
 		if (params.ndis_tcp6csum == NDIS_OFFLOAD_PARAM_TX)
 			params.ndis_tcp6csum = NDIS_OFFLOAD_PARAM_TXRX;
 		else
 			params.ndis_tcp6csum = NDIS_OFFLOAD_PARAM_RX;
 	}
 
 	/* UDP6 checksum */
 	if ((hwcaps.ndis_csum.ndis_ip6_txcsum & HN_NDIS_TXCSUM_CAP_UDP6) ==
 	    HN_NDIS_TXCSUM_CAP_UDP6) {
 		caps |= HN_CAP_UDP6CS;
 		params.ndis_udp6csum = NDIS_OFFLOAD_PARAM_TX;
 	}
 	if (hwcaps.ndis_csum.ndis_ip6_rxcsum & NDIS_RXCSUM_CAP_UDP6) {
 		if (params.ndis_udp6csum == NDIS_OFFLOAD_PARAM_TX)
 			params.ndis_udp6csum = NDIS_OFFLOAD_PARAM_TXRX;
 		else
 			params.ndis_udp6csum = NDIS_OFFLOAD_PARAM_RX;
 	}
 
 	if (bootverbose) {
 		if_printf(sc->hn_ifp, "offload csum: "
 		    "ip4 %u, tcp4 %u, udp4 %u, tcp6 %u, udp6 %u\n",
 		    params.ndis_ip4csum,
 		    params.ndis_tcp4csum,
 		    params.ndis_udp4csum,
 		    params.ndis_tcp6csum,
 		    params.ndis_udp6csum);
 		if_printf(sc->hn_ifp, "offload lsov2: ip4 %u, ip6 %u\n",
 		    params.ndis_lsov2_ip4,
 		    params.ndis_lsov2_ip6);
 	}
 
 	error = hn_rndis_set(sc, OID_TCP_OFFLOAD_PARAMETERS, &params, paramsz);
 	if (error) {
 		if_printf(sc->hn_ifp, "offload config failed: %d\n", error);
 		return (error);
 	}
 
 	if (bootverbose)
 		if_printf(sc->hn_ifp, "offload config done\n");
 	sc->hn_caps |= caps;
 	return (0);
 }
 
 int
 hn_rndis_conf_rss(struct hn_softc *sc, uint16_t flags)
 {
 	struct ndis_rssprm_toeplitz *rss = &sc->hn_rss;
 	struct ndis_rss_params *prm = &rss->rss_params;
 	int error, rss_size;
 
 	/*
 	 * Only NDIS 6.20+ is supported:
 	 * We only support 4bytes element in indirect table, which has been
 	 * adopted since NDIS 6.20.
 	 */
 	KASSERT(sc->hn_ndis_ver >= HN_NDIS_VERSION_6_20,
 	    ("NDIS 6.20+ is required, NDIS version 0x%08x", sc->hn_ndis_ver));
 
 	/* XXX only one can be specified through, popcnt? */
 	KASSERT((sc->hn_rss_hash & NDIS_HASH_FUNCTION_MASK), ("no hash func"));
 	KASSERT((sc->hn_rss_hash & NDIS_HASH_TYPE_MASK), ("no hash types"));
 	KASSERT(sc->hn_rss_ind_size > 0, ("no indirect table size"));
 
 	if (bootverbose) {
 		if_printf(sc->hn_ifp, "RSS indirect table size %d, "
 		    "hash 0x%08x\n", sc->hn_rss_ind_size, sc->hn_rss_hash);
 	}
 
 	/*
 	 * NOTE:
 	 * DO NOT whack rss_key and rss_ind, which are setup by the caller.
 	 */
 	memset(prm, 0, sizeof(*prm));
 	rss_size = NDIS_RSSPRM_TOEPLITZ_SIZE(sc->hn_rss_ind_size);
 
 	prm->ndis_hdr.ndis_type = NDIS_OBJTYPE_RSS_PARAMS;
 	prm->ndis_hdr.ndis_rev = NDIS_RSS_PARAMS_REV_2;
 	prm->ndis_hdr.ndis_size = rss_size;
 	prm->ndis_flags = flags;
 	prm->ndis_hash = sc->hn_rss_hash;
 	prm->ndis_indsize = sizeof(rss->rss_ind[0]) * sc->hn_rss_ind_size;
 	prm->ndis_indoffset =
 	    __offsetof(struct ndis_rssprm_toeplitz, rss_ind[0]);
 	prm->ndis_keysize = sizeof(rss->rss_key);
 	prm->ndis_keyoffset =
 	    __offsetof(struct ndis_rssprm_toeplitz, rss_key[0]);
 
 	error = hn_rndis_set(sc, OID_GEN_RECEIVE_SCALE_PARAMETERS,
 	    rss, rss_size);
 	if (error) {
 		if_printf(sc->hn_ifp, "RSS config failed: %d\n", error);
 	} else {
 		if (bootverbose)
 			if_printf(sc->hn_ifp, "RSS config done\n");
 	}
 	return (error);
 }
 
 int
 hn_rndis_set_rxfilter(struct hn_softc *sc, uint32_t filter)
 {
 	int error;
 
 	error = hn_rndis_set(sc, OID_GEN_CURRENT_PACKET_FILTER,
 	    &filter, sizeof(filter));
 	if (error) {
 		if_printf(sc->hn_ifp, "set RX filter 0x%08x failed: %d\n",
 		    filter, error);
 	} else {
 		if (bootverbose) {
 			if_printf(sc->hn_ifp, "set RX filter 0x%08x done\n",
 			    filter);
 		}
 	}
 	return (error);
 }
 
 static int
 hn_rndis_init(struct hn_softc *sc)
 {
 	struct rndis_init_req *req;
 	const struct rndis_init_comp *comp;
 	struct vmbus_xact *xact;
 	size_t comp_len;
 	uint32_t rid;
 	int error;
 
 	xact = vmbus_xact_get(sc->hn_xact, sizeof(*req));
 	if (xact == NULL) {
 		if_printf(sc->hn_ifp, "no xact for RNDIS init\n");
 		return (ENXIO);
 	}
 	rid = hn_rndis_rid(sc);
 	req = vmbus_xact_req_data(xact);
 	req->rm_type = REMOTE_NDIS_INITIALIZE_MSG;
 	req->rm_len = sizeof(*req);
 	req->rm_rid = rid;
 	req->rm_ver_major = RNDIS_VERSION_MAJOR;
 	req->rm_ver_minor = RNDIS_VERSION_MINOR;
 	req->rm_max_xfersz = HN_RNDIS_XFER_SIZE;
 
 	comp_len = RNDIS_INIT_COMP_SIZE_MIN;
 	comp = hn_rndis_xact_execute(sc, xact, rid, sizeof(*req), &comp_len,
 	    REMOTE_NDIS_INITIALIZE_CMPLT);
 	if (comp == NULL) {
 		if_printf(sc->hn_ifp, "exec RNDIS init failed\n");
 		error = EIO;
 		goto done;
 	}
 
 	if (comp->rm_status != RNDIS_STATUS_SUCCESS) {
 		if_printf(sc->hn_ifp, "RNDIS init failed: status 0x%08x\n",
 		    comp->rm_status);
 		error = EIO;
 		goto done;
 	}
 	if (bootverbose) {
 		if_printf(sc->hn_ifp, "RNDIS ver %u.%u, pktsz %u, pktcnt %u, "
 		    "align %u\n", comp->rm_ver_major, comp->rm_ver_minor,
 		    comp->rm_pktmaxsz, comp->rm_pktmaxcnt,
 		    1U << comp->rm_align);
 	}
 	error = 0;
 done:
 	vmbus_xact_put(xact);
 	return (error);
 }
 
 static int
 hn_rndis_halt(struct hn_softc *sc)
 {
 	struct vmbus_xact *xact;
 	struct rndis_halt_req *halt;
-	struct hn_send_ctx sndc;
+	struct hn_nvs_sendctx sndc;
 	size_t comp_len;
 
 	xact = vmbus_xact_get(sc->hn_xact, sizeof(*halt));
 	if (xact == NULL) {
 		if_printf(sc->hn_ifp, "no xact for RNDIS halt\n");
 		return (ENXIO);
 	}
 	halt = vmbus_xact_req_data(xact);
 	halt->rm_type = REMOTE_NDIS_HALT_MSG;
 	halt->rm_len = sizeof(*halt);
 	halt->rm_rid = hn_rndis_rid(sc);
 
 	/* No RNDIS completion; rely on NVS message send completion */
-	hn_send_ctx_init(&sndc, hn_nvs_sent_xact, xact);
+	hn_nvs_sendctx_init(&sndc, hn_nvs_sent_xact, xact);
 	hn_rndis_xact_exec1(sc, xact, sizeof(*halt), &sndc, &comp_len);
 
 	vmbus_xact_put(xact);
 	if (bootverbose)
 		if_printf(sc->hn_ifp, "RNDIS halt done\n");
 	return (0);
 }
 
 static int
 hn_rndis_query_hwcaps(struct hn_softc *sc, struct ndis_offload *caps)
 {
 	struct ndis_offload in;
 	size_t caps_len, size;
 	int error;
 
 	memset(&in, 0, sizeof(in));
 	in.ndis_hdr.ndis_type = NDIS_OBJTYPE_OFFLOAD;
 	if (sc->hn_ndis_ver >= HN_NDIS_VERSION_6_30) {
 		in.ndis_hdr.ndis_rev = NDIS_OFFLOAD_REV_3;
 		size = NDIS_OFFLOAD_SIZE;
 	} else if (sc->hn_ndis_ver >= HN_NDIS_VERSION_6_1) {
 		in.ndis_hdr.ndis_rev = NDIS_OFFLOAD_REV_2;
 		size = NDIS_OFFLOAD_SIZE_6_1;
 	} else {
 		in.ndis_hdr.ndis_rev = NDIS_OFFLOAD_REV_1;
 		size = NDIS_OFFLOAD_SIZE_6_0;
 	}
 	in.ndis_hdr.ndis_size = size;
 
 	caps_len = NDIS_OFFLOAD_SIZE;
 	error = hn_rndis_query2(sc, OID_TCP_OFFLOAD_HARDWARE_CAPABILITIES,
 	    &in, size, caps, &caps_len, NDIS_OFFLOAD_SIZE_6_0);
 	if (error)
 		return (error);
 
 	/*
 	 * Preliminary verification.
 	 */
 	if (caps->ndis_hdr.ndis_type != NDIS_OBJTYPE_OFFLOAD) {
 		if_printf(sc->hn_ifp, "invalid NDIS objtype 0x%02x\n",
 		    caps->ndis_hdr.ndis_type);
 		return (EINVAL);
 	}
 	if (caps->ndis_hdr.ndis_rev < NDIS_OFFLOAD_REV_1) {
 		if_printf(sc->hn_ifp, "invalid NDIS objrev 0x%02x\n",
 		    caps->ndis_hdr.ndis_rev);
 		return (EINVAL);
 	}
 	if (caps->ndis_hdr.ndis_size > caps_len) {
 		if_printf(sc->hn_ifp, "invalid NDIS objsize %u, "
 		    "data size %zu\n", caps->ndis_hdr.ndis_size, caps_len);
 		return (EINVAL);
 	} else if (caps->ndis_hdr.ndis_size < NDIS_OFFLOAD_SIZE_6_0) {
 		if_printf(sc->hn_ifp, "invalid NDIS objsize %u\n",
 		    caps->ndis_hdr.ndis_size);
 		return (EINVAL);
 	}
 
 	if (bootverbose) {
 		/*
 		 * NOTE:
 		 * caps->ndis_hdr.ndis_size MUST be checked before accessing
 		 * NDIS 6.1+ specific fields.
 		 */
 		if_printf(sc->hn_ifp, "hwcaps rev %u\n",
 		    caps->ndis_hdr.ndis_rev);
 
 		if_printf(sc->hn_ifp, "hwcaps csum: "
 		    "ip4 tx 0x%x/0x%x rx 0x%x/0x%x, "
 		    "ip6 tx 0x%x/0x%x rx 0x%x/0x%x\n",
 		    caps->ndis_csum.ndis_ip4_txcsum,
 		    caps->ndis_csum.ndis_ip4_txenc,
 		    caps->ndis_csum.ndis_ip4_rxcsum,
 		    caps->ndis_csum.ndis_ip4_rxenc,
 		    caps->ndis_csum.ndis_ip6_txcsum,
 		    caps->ndis_csum.ndis_ip6_txenc,
 		    caps->ndis_csum.ndis_ip6_rxcsum,
 		    caps->ndis_csum.ndis_ip6_rxenc);
 		if_printf(sc->hn_ifp, "hwcaps lsov2: "
 		    "ip4 maxsz %u minsg %u encap 0x%x, "
 		    "ip6 maxsz %u minsg %u encap 0x%x opts 0x%x\n",
 		    caps->ndis_lsov2.ndis_ip4_maxsz,
 		    caps->ndis_lsov2.ndis_ip4_minsg,
 		    caps->ndis_lsov2.ndis_ip4_encap,
 		    caps->ndis_lsov2.ndis_ip6_maxsz,
 		    caps->ndis_lsov2.ndis_ip6_minsg,
 		    caps->ndis_lsov2.ndis_ip6_encap,
 		    caps->ndis_lsov2.ndis_ip6_opts);
 	}
 	return (0);
 }
 
 int
 hn_rndis_attach(struct hn_softc *sc, int mtu)
 {
 	int error;
 
 	/*
 	 * Initialize RNDIS.
 	 */
 	error = hn_rndis_init(sc);
 	if (error)
 		return (error);
 
 	/*
 	 * Configure NDIS offload settings.
 	 * XXX no offloading, if error happened?
 	 */
 	hn_rndis_conf_offload(sc, mtu);
 	return (0);
 }
 
 void
 hn_rndis_detach(struct hn_softc *sc)
 {
 
 	/* Halt the RNDIS. */
 	hn_rndis_halt(sc);
 }
 
 void
 hv_rf_channel_rollup(struct hn_rx_ring *rxr, struct hn_tx_ring *txr)
 {
 
 	hn_chan_rollup(rxr, txr);
 }
Index: head/sys/dev/hyperv/netvsc/if_hnvar.h
===================================================================
--- head/sys/dev/hyperv/netvsc/if_hnvar.h	(revision 307989)
+++ head/sys/dev/hyperv/netvsc/if_hnvar.h	(revision 307990)
@@ -1,322 +1,265 @@
 /*-
  * Copyright (c) 2016 Microsoft Corp.
  * All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions
  * are met:
  * 1. Redistributions of source code must retain the above copyright
  *    notice unmodified, this list of conditions, and the following
  *    disclaimer.
  * 2. Redistributions in binary form must reproduce the above copyright
  *    notice, this list of conditions and the following disclaimer in the
  *    documentation and/or other materials provided with the distribution.
  *
  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  *
  * $FreeBSD$
  */
 
 #ifndef _IF_HNVAR_H_
 #define _IF_HNVAR_H_
 
 #define HN_USE_TXDESC_BUFRING
 
 #define HN_CHIM_SIZE			(15 * 1024 * 1024)
 
 #define HN_RXBUF_SIZE			(16 * 1024 * 1024)
 #define HN_RXBUF_SIZE_COMPAT		(15 * 1024 * 1024)
 
 /* Claimed to be 12232B */
 #define HN_MTU_MAX			(9 * 1024)
 
 #define HN_PKTBUF_LEN			4096
 
 #define HN_TXBR_SIZE			(128 * PAGE_SIZE)
 #define HN_RXBR_SIZE			(128 * PAGE_SIZE)
 
 #define HN_XACT_REQ_PGCNT		2
 #define HN_XACT_RESP_PGCNT		2
 #define HN_XACT_REQ_SIZE		(HN_XACT_REQ_PGCNT * PAGE_SIZE)
 #define HN_XACT_RESP_SIZE		(HN_XACT_RESP_PGCNT * PAGE_SIZE)
 
 #define HN_GPACNT_MAX			32
 
-struct vmbus_channel;
-struct hn_softc;
-struct hn_send_ctx;
-
-typedef void		(*hn_sent_callback_t)
-			(struct hn_send_ctx *, struct hn_softc *,
-			 struct vmbus_channel *, const void *, int);
-
-struct hn_send_ctx {
-	hn_sent_callback_t	hn_cb;
-	void			*hn_cbarg;
-};
-
-#define HN_SEND_CTX_INITIALIZER(cb, cbarg)	\
-{						\
-	.hn_cb		= cb,			\
-	.hn_cbarg	= cbarg			\
-}
-
 #define HN_NDIS_VLAN_INFO_INVALID	0xffffffff
 #define HN_NDIS_RXCSUM_INFO_INVALID	0
 #define HN_NDIS_HASH_INFO_INVALID	0
 
 struct hn_recvinfo {
 	uint32_t			vlan_info;
 	uint32_t			csum_info;
 	uint32_t			hash_info;
 	uint32_t			hash_value;
 };
 
 struct hn_txdesc;
 #ifndef HN_USE_TXDESC_BUFRING
 SLIST_HEAD(hn_txdesc_list, hn_txdesc);
 #else
 struct buf_ring;
 #endif
 struct hn_tx_ring;
 
 struct hn_rx_ring {
 	struct ifnet	*hn_ifp;
 	struct hn_tx_ring *hn_txr;
 	void		*hn_pktbuf;
 	uint8_t		*hn_rxbuf;	/* shadow sc->hn_rxbuf */
 	int		hn_rx_idx;
 
 	/* Trust csum verification on host side */
 	int		hn_trust_hcsum;	/* HN_TRUST_HCSUM_ */
 	struct lro_ctrl	hn_lro;
 
 	u_long		hn_csum_ip;
 	u_long		hn_csum_tcp;
 	u_long		hn_csum_udp;
 	u_long		hn_csum_trusted;
 	u_long		hn_lro_tried;
 	u_long		hn_small_pkts;
 	u_long		hn_pkts;
 	u_long		hn_rss_pkts;
 
 	/* Rarely used stuffs */
 	struct sysctl_oid *hn_rx_sysctl_tree;
 	int		hn_rx_flags;
 
 	void		*hn_br;		/* TX/RX bufring */
 	struct hyperv_dma hn_br_dma;
 } __aligned(CACHE_LINE_SIZE);
 
 #define HN_TRUST_HCSUM_IP	0x0001
 #define HN_TRUST_HCSUM_TCP	0x0002
 #define HN_TRUST_HCSUM_UDP	0x0004
 
 #define HN_RX_FLAG_ATTACHED	0x1
 
 struct hn_tx_ring {
 #ifndef HN_USE_TXDESC_BUFRING
 	struct mtx	hn_txlist_spin;
 	struct hn_txdesc_list hn_txlist;
 #else
 	struct buf_ring	*hn_txdesc_br;
 #endif
 	int		hn_txdesc_cnt;
 	int		hn_txdesc_avail;
 	u_short		hn_has_txeof;
 	u_short		hn_txdone_cnt;
 
 	int		hn_sched_tx;
 	void		(*hn_txeof)(struct hn_tx_ring *);
 	struct taskqueue *hn_tx_taskq;
 	struct task	hn_tx_task;
 	struct task	hn_txeof_task;
 
 	struct buf_ring	*hn_mbuf_br;
 	int		hn_oactive;
 	int		hn_tx_idx;
 	int		hn_tx_flags;
 
 	struct mtx	hn_tx_lock;
 	struct hn_softc	*hn_sc;
 	struct vmbus_channel *hn_chan;
 
 	int		hn_direct_tx_size;
 	int		hn_chim_size;
 	bus_dma_tag_t	hn_tx_data_dtag;
 	uint64_t	hn_csum_assist;
 
 	int		(*hn_sendpkt)(struct hn_tx_ring *, struct hn_txdesc *);
 	int		hn_suspended;
 	int		hn_gpa_cnt;
 	struct vmbus_gpa hn_gpa[HN_GPACNT_MAX];
 
 	u_long		hn_no_txdescs;
 	u_long		hn_send_failed;
 	u_long		hn_txdma_failed;
 	u_long		hn_tx_collapsed;
 	u_long		hn_tx_chimney_tried;
 	u_long		hn_tx_chimney;
 	u_long		hn_pkts;
 
 	/* Rarely used stuffs */
 	struct hn_txdesc *hn_txdesc;
 	bus_dma_tag_t	hn_tx_rndis_dtag;
 	struct sysctl_oid *hn_tx_sysctl_tree;
 } __aligned(CACHE_LINE_SIZE);
 
 #define HN_TX_FLAG_ATTACHED	0x1
 #define HN_TX_FLAG_HASHVAL	0x2	/* support HASHVAL pktinfo */
 
 /*
  * Device-specific softc structure
  */
 struct hn_softc {
 	struct ifnet    *hn_ifp;
 	struct ifmedia	hn_media;
 	device_t        hn_dev;
 	int             hn_if_flags;
 	struct sx	hn_lock;
 	struct vmbus_channel *hn_prichan;
 
 	int		hn_rx_ring_cnt;
 	int		hn_rx_ring_inuse;
 	struct hn_rx_ring *hn_rx_ring;
 
 	int		hn_tx_ring_cnt;
 	int		hn_tx_ring_inuse;
 	struct hn_tx_ring *hn_tx_ring;
 
 	uint8_t		*hn_chim;
 	u_long		*hn_chim_bmap;
 	int		hn_chim_bmap_cnt;
 	int		hn_chim_cnt;
 	int		hn_chim_szmax;
 
 	int		hn_cpu;
 	struct taskqueue *hn_tx_taskq;
 	struct sysctl_oid *hn_tx_sysctl_tree;
 	struct sysctl_oid *hn_rx_sysctl_tree;
 	struct vmbus_xact_ctx *hn_xact;
 	uint32_t	hn_nvs_ver;
 	uint32_t	hn_rx_filter;
 
 	struct taskqueue	*hn_mgmt_taskq;
 	struct taskqueue	*hn_mgmt_taskq0;
 	struct task		hn_link_task;
 	struct task		hn_netchg_init;
 	struct timeout_task	hn_netchg_status;
 	uint32_t		hn_link_flags;	/* HN_LINK_FLAG_ */
 
 	uint32_t		hn_caps;	/* HN_CAP_ */
 	uint32_t		hn_flags;	/* HN_FLAG_ */
 	void			*hn_rxbuf;
 	uint32_t		hn_rxbuf_gpadl;
 	struct hyperv_dma	hn_rxbuf_dma;
 
 	uint32_t		hn_chim_gpadl;
 	struct hyperv_dma	hn_chim_dma;
 
 	uint32_t		hn_rndis_rid;
 	uint32_t		hn_ndis_ver;
 	int			hn_ndis_tso_szmax;
 	int			hn_ndis_tso_sgmin;
 
 	int			hn_rss_ind_size;
 	uint32_t		hn_rss_hash;	/* NDIS_HASH_ */
 	struct ndis_rssprm_toeplitz hn_rss;
 };
 
 #define HN_FLAG_RXBUF_CONNECTED		0x0001
 #define HN_FLAG_CHIM_CONNECTED		0x0002
 #define HN_FLAG_HAS_RSSKEY		0x0004
 #define HN_FLAG_HAS_RSSIND		0x0008
 #define HN_FLAG_SYNTH_ATTACHED		0x0010
 
 #define HN_CAP_VLAN			0x0001
 #define HN_CAP_MTU			0x0002
 #define HN_CAP_IPCS			0x0004
 #define HN_CAP_TCP4CS			0x0008
 #define HN_CAP_TCP6CS			0x0010
 #define HN_CAP_UDP4CS			0x0020
 #define HN_CAP_UDP6CS			0x0040
 #define HN_CAP_TSO4			0x0080
 #define HN_CAP_TSO6			0x0100
 #define HN_CAP_HASHVAL			0x0200
 
 /* Capability description for use with printf(9) %b identifier. */
 #define HN_CAP_BITS				\
 	"\020\1VLAN\2MTU\3IPCS\4TCP4CS\5TCP6CS"	\
 	"\6UDP4CS\7UDP6CS\10TSO4\11TSO6\12HASHVAL"
 
 #define HN_LINK_FLAG_LINKUP		0x0001
 #define HN_LINK_FLAG_NETCHG		0x0002
 
-static __inline void
-hn_send_ctx_init(struct hn_send_ctx *sndc, hn_sent_callback_t cb, void *cbarg)
-{
-
-	sndc->hn_cb = cb;
-	sndc->hn_cbarg = cbarg;
-}
-
-static __inline int
-hn_nvs_send(struct vmbus_channel *chan, uint16_t flags,
-    void *nvs_msg, int nvs_msglen, struct hn_send_ctx *sndc)
-{
-
-	return (vmbus_chan_send(chan, VMBUS_CHANPKT_TYPE_INBAND, flags,
-	    nvs_msg, nvs_msglen, (uint64_t)(uintptr_t)sndc));
-}
-
-static __inline int
-hn_nvs_send_sglist(struct vmbus_channel *chan, struct vmbus_gpa sg[], int sglen,
-    void *nvs_msg, int nvs_msglen, struct hn_send_ctx *sndc)
-{
-
-	return (vmbus_chan_send_sglist(chan, sg, sglen, nvs_msg, nvs_msglen,
-	    (uint64_t)(uintptr_t)sndc));
-}
-
-struct vmbus_xact;
 struct rndis_packet_msg;
 
 int		hn_rndis_attach(struct hn_softc *sc, int mtu);
 void		hn_rndis_detach(struct hn_softc *sc);
 int		hn_rndis_conf_rss(struct hn_softc *sc, uint16_t flags);
 void		*hn_rndis_pktinfo_append(struct rndis_packet_msg *,
 		    size_t pktsize, size_t pi_dlen, uint32_t pi_type);
 int		hn_rndis_query_rsscaps(struct hn_softc *sc, int *rxr_cnt);
 int		hn_rndis_get_eaddr(struct hn_softc *sc, uint8_t *eaddr);
 int		hn_rndis_get_linkstatus(struct hn_softc *sc,
 		    uint32_t *link_status);
 /* filter: NDIS_PACKET_TYPE_. */
 int		hn_rndis_set_rxfilter(struct hn_softc *sc, uint32_t filter);
 
-int		hn_nvs_attach(struct hn_softc *sc, int mtu);
-void		hn_nvs_detach(struct hn_softc *sc);
-int		hn_nvs_alloc_subchans(struct hn_softc *sc, int *nsubch);
-void		hn_nvs_sent_xact(struct hn_send_ctx *sndc, struct hn_softc *sc,
-		    struct vmbus_channel *chan, const void *data, int dlen);
-int		hn_nvs_send_rndis_ctrl(struct vmbus_channel *chan,
-		    struct hn_send_ctx *sndc, struct vmbus_gpa *gpa,
-		    int gpa_cnt);
-
 int		hn_rxpkt(struct hn_rx_ring *rxr, const void *data, int dlen,
 		    const struct hn_recvinfo *info);
 void		hn_chan_rollup(struct hn_rx_ring *rxr, struct hn_tx_ring *txr);
 void		hn_link_status_update(struct hn_softc *sc);
 void		hn_network_change(struct hn_softc *sc);
-
-extern struct hn_send_ctx	hn_send_ctx_none;
 
 #endif	/* !_IF_HNVAR_H_ */