diff -r 823a0086cd5b sys/dev/usb/if_ncm.c
--- a/sys/dev/usb/if_ncm.c	Sun Jun 28 11:38:58 2026 +0000
+++ b/sys/dev/usb/if_ncm.c	Mon Jun 29 02:09:55 2026 +0000
@@ -54,6 +54,7 @@
 
 struct ncm_softc {
 	struct usbnet	ncm_un;
+	struct ncm_ntb_parameters sc_ntb_param;
 	uint32_t	sc_tx_seq;
 };
 
@@ -67,10 +68,22 @@ static void	ncm_uno_rx_loop(struct usbne
 		    uint32_t);
 static unsigned	ncm_uno_tx_prepare(struct usbnet *, struct mbuf *,
 		    struct usbnet_chain *);
+static int	ncm_uno_tx_begin(struct usbnet *);
+static int	ncm_uno_tx_add(struct usbnet *, struct mbuf *,
+		    unsigned, unsigned);
+static void	ncm_uno_tx_commit(struct usbnet *, struct mbuf *,
+		    struct usbnet_chain *, unsigned, unsigned);
 
-static const struct usbnet_ops ncm_ops = {
+static const struct usbnet_ops_v2 ncm_ops = {
+    .v1 = {
 	.uno_tx_prepare = ncm_uno_tx_prepare,
 	.uno_rx_loop = ncm_uno_rx_loop,
+    },
+    .v2 = {
+	.uno_tx_begin = ncm_uno_tx_begin,
+	.uno_tx_add = ncm_uno_tx_add,
+	.uno_tx_commit = ncm_uno_tx_commit,
+    },
 };
 
 static int
@@ -102,8 +115,6 @@ ncm_attach(device_t parent, device_t sel
 	const usb_cdc_ethernet_descriptor_t *ue;
 	char				 eaddr_str[USB_MAX_ENCODED_STRING_LEN];
 	usb_device_request_t		 req;
-	struct ncm_ntb_parameters	 np;
-
 
 	aprint_naive("\n");
 	aprint_normal("\n");
@@ -114,7 +125,7 @@ ncm_attach(device_t parent, device_t sel
 	un->un_dev = self;
 	un->un_udev = dev;
 	un->un_sc = sc;
-	un->un_ops = &ncm_ops;
+	un->un_ops = &ncm_ops.v1;
 	un->un_rx_xfer_flags = USBD_SHORT_XFER_OK;
 	un->un_tx_xfer_flags = USBD_FORCE_SHORT_XFER;
 
@@ -226,24 +237,62 @@ ncm_attach(device_t parent, device_t sel
 	req.bRequest = NCM_GET_NTB_PARAMETERS;
 	USETW(req.wValue, 0);
 	USETW(req.wIndex, uiaa->uiaa_ifaceno);
-	USETW(req.wLength, sizeof(np));
-	if (usbd_do_request(un->un_udev, &req, &np) != USBD_NORMAL_COMPLETION ||
-	    UGETW(np.wLength) != sizeof(np)) {
+	USETW(req.wLength, sizeof(sc->sc_ntb_param));
+	if (usbd_do_request(un->un_udev, &req, &sc->sc_ntb_param) !=
+	    USBD_NORMAL_COMPLETION ||
+	    UGETW(sc->sc_ntb_param.wLength) != sizeof(sc->sc_ntb_param)) {
 		aprint_error_dev(un->un_dev,
 		    "NCM_GET_NTB_PARAMETERS failed\n");
 		return;
 	}
+	aprint_verbose_dev(un->un_dev, "formats: 0x%02x%02x\n",
+	    sc->sc_ntb_param.bmNtbFormatsSupported[1],
+	    sc->sc_ntb_param.bmNtbFormatsSupported[0]);
+	aprint_verbose_dev(un->un_dev,
+	    "in: max %u, =%u mod %u, align %u\n",
+	    UGETDW(sc->sc_ntb_param.dwNtbInMaxSize),
+	    UGETW(sc->sc_ntb_param.wNdpInPayloadRemainder),
+	    UGETW(sc->sc_ntb_param.wNdpInDivisor),
+	    UGETW(sc->sc_ntb_param.wNdpInAlignment));
+	aprint_verbose_dev(un->un_dev,
+	    "out: max %u (%u pkt), =%u mod %u, align %u\n",
+	    UGETDW(sc->sc_ntb_param.dwNtbOutMaxSize),
+	    UGETW(sc->sc_ntb_param.wNtbOutMaxDatagrams),
+	    UGETW(sc->sc_ntb_param.wNdpOutPayloadRemainder),
+	    UGETW(sc->sc_ntb_param.wNdpOutDivisor),
+	    UGETW(sc->sc_ntb_param.wNdpOutAlignment));
+
+	/* paranoia */
+	if (UGETW(sc->sc_ntb_param.wNdpInDivisor) == 0)
+		USETW(sc->sc_ntb_param.wNdpInDivisor, 1);
+	if (UGETW(sc->sc_ntb_param.wNdpOutDivisor) == 0)
+		USETW(sc->sc_ntb_param.wNdpOutDivisor, 1);
+	if (UGETW(sc->sc_ntb_param.wNdpInPayloadRemainder) >=
+	    UGETW(sc->sc_ntb_param.wNdpInDivisor)) {
+		USETW(sc->sc_ntb_param.wNdpInPayloadRemainder,
+		    UGETW(sc->sc_ntb_param.wNdpInPayloadRemainder) %
+		    UGETW(sc->sc_ntb_param.wNdpInDivisor));
+	}
+	if (UGETW(sc->sc_ntb_param.wNdpOutPayloadRemainder) >=
+	    UGETW(sc->sc_ntb_param.wNdpOutDivisor)) {
+		USETW(sc->sc_ntb_param.wNdpOutPayloadRemainder,
+		    UGETW(sc->sc_ntb_param.wNdpOutPayloadRemainder) %
+		    UGETW(sc->sc_ntb_param.wNdpOutDivisor));
+	}
+
 	un->un_rx_list_cnt = 1;
 	un->un_tx_list_cnt = 1;
-	un->un_rx_bufsz = UGETDW(np.dwNtbInMaxSize);
-	un->un_tx_bufsz = UGETDW(np.dwNtbOutMaxSize);
-	if (un->un_tx_bufsz < NCM_MIN_TX_BUFSZ) {
+	un->un_rx_bufsz = UGETDW(sc->sc_ntb_param.dwNtbInMaxSize);
+	un->un_tx_bufsz = UGETDW(sc->sc_ntb_param.dwNtbOutMaxSize);
+	if (un->un_tx_bufsz < NCM_MIN_TX_BUFSZ ||
+	    un->un_tx_bufsz - NCM_MIN_TX_BUFSZ <
+	    (unsigned)UGETW(sc->sc_ntb_param.wNdpOutDivisor)) {
 		aprint_error_dev(un->un_dev, "dwNtbOutMaxSize %u too small\n",
 		    un->un_tx_bufsz);
 		return;
 	}
 
-	usbnet_attach(un);
+	usbnet_attach_v2(un, &ncm_ops);
 	usbnet_attach_ifp(un, IFF_SIMPLEX | IFF_BROADCAST | IFF_MULTICAST,
 	    0, NULL);
 
@@ -300,7 +349,7 @@ ncm_uno_rx_loop(struct usbnet *un, struc
 		return;
 	}
 
-	const unsigned max_datagrams = (UGETW(ptr->wLength) - 
+	const unsigned max_datagrams = (UGETW(ptr->wLength) -
 	    offsetof(struct ncm_dptab16, datagram))/sizeof(ptr->datagram[0]);
 	for (i = 0; i < max_datagrams; i++) {
 		uint16_t data_start = UGETW(ptr->datagram[i].wDatagramIndex);
@@ -320,46 +369,236 @@ ncm_uno_rx_loop(struct usbnet *un, struc
 	}
 }
 
+/*
+ * The layout of a tx xfer buffer is:
+ *
+ *	+--------------+-------------+---------------+---------------+-----+
+ *	| ncm_header16 | ncm_dptab16 | ncm_pointer16 | ncm_pointer16 | ... |
+ *	+--------------+-------------+---------------+---------------+-----+
+ *	+-----+----------------+-------+-----------+------------+--------------
+ *	| pad | payload 0      | pad   | payload 1 | pad        | payload 2 ...
+ *	+-----+----------------+-------+-----------+------------+--------------
+ *	      |                        |                        |
+ *	   k*d + r                (k + 1)*d + r           (k + 2)*d + r
+ *
+ * The ncm_header16 gives the offset of the datagram pointer table,
+ * headed by ncm_dptab16.  There is one more ncm_pointer16 than packet,
+ * with zero offset/length to mark the end of the transfer.
+ *
+ * The NTB parameters require the packet payloads to start at specific
+ * offsets modulo a divisor:
+ *
+ *	offset % wNdpOutDivisor = wNdpOutPayloadRemainder
+ *
+ * In other words, each offset is k*d + r for some k, where d is
+ * wNdpOutDivisor and r is wNdpOutPayloadRemainder.
+ */
+
+/*
+ * ncm_out_align(sc, off0)
+ *
+ *	Return the smallest offset off >= off0 such that
+ *
+ *		off % wNdpOutDivisor = wNdpOutPayloadRemainder.
+ */
+static unsigned
+ncm_out_align(struct ncm_softc *sc, unsigned off0)
+{
+	unsigned off;
+
+	KASSERTMSG(off0 <= sc->ncm_un.un_tx_bufsz, "off0=%u bufsz=%u",
+	    off0, sc->ncm_un.un_tx_bufsz);
+
+	off = rounddown(off0, UGETW(sc->sc_ntb_param.wNdpOutDivisor));
+	off += UGETW(sc->sc_ntb_param.wNdpOutPayloadRemainder);
+	if (off < off0)
+		off += UGETW(sc->sc_ntb_param.wNdpOutDivisor);
+
+	KASSERTMSG(((off % UGETW(sc->sc_ntb_param.wNdpOutDivisor)) ==
+		(unsigned)UGETW(sc->sc_ntb_param.wNdpOutPayloadRemainder)),
+	    "off=%u divisor=%u remainder=%u",
+	    off,
+	    UGETW(sc->sc_ntb_param.wNdpOutDivisor),
+	    UGETW(sc->sc_ntb_param.wNdpOutPayloadRemainder));
+	return off;
+}
+
 static unsigned
 ncm_uno_tx_prepare(struct usbnet *un, struct mbuf *m, struct usbnet_chain *c)
 {
 	struct ncm_softc *sc = un->un_sc;
 	struct ncm_dptab16 *ptr;
 	struct ncm_header16 *hdr;
-	unsigned hdr_len, len;
+	unsigned hdr_len, off, len;
 
-	hdr_len = sizeof(*hdr) + sizeof(*ptr);
+	hdr_len = sizeof(*hdr) + offsetof(struct ncm_dptab16, datagram[2]);
+	off = ncm_out_align(sc, hdr_len);
 	len = m->m_pkthdr.len;
-	KASSERT(hdr_len <= un->un_tx_bufsz);
-	if (len > un->un_tx_bufsz - hdr_len)
+	KASSERTMSG(hdr_len <= un->un_tx_bufsz,
+	    "hdr_len=%u bufsz=%u", hdr_len, un->un_tx_bufsz);
+	if (off > un->un_tx_bufsz || len > un->un_tx_bufsz - off)
 		return 0;
 
 	hdr = (struct ncm_header16 *)c->unc_buf;
 	ptr = (struct ncm_dptab16 *)(hdr + 1);
 	memset(hdr, 0, sizeof(*hdr));
+	memset(ptr, 0, offsetof(struct ncm_dptab16, datagram[2]));
+
+	USETDW(hdr->dwSignature, NCM_HDR16_SIG);
+	USETW(hdr->wHeaderLength, sizeof(*hdr));
+	USETW(hdr->wSequence, sc->sc_tx_seq);
+	sc->sc_tx_seq++;
+	USETW(hdr->wBlockLength, off + len);
+	USETW(hdr->wNdpIndex, sizeof(*hdr));
+
+	USETDW(ptr->dwSignature, NCM_PTR16_SIG_NO_CRC);
+	USETW(ptr->wLength, offsetof(struct ncm_dptab16, datagram[2]));
+	USETW(ptr->wNextNdpIndex, 0);
+
+	USETW(ptr->datagram[0].wDatagramIndex, off);
+	USETW(ptr->datagram[0].wDatagramLength, len);
+
+	USETW(ptr->datagram[1].wDatagramIndex, 0);
+	USETW(ptr->datagram[1].wDatagramLength, 0);
+
+	m_copydata(m, 0, len, c->unc_buf + off);
+
+	return off + len;
+}
+
+/*
+ * ncm_uno_tx_begin(un)
+ *
+ *	Called when starting a batch of packets for tx in a single
+ *	transfer.  Returns the number of bytes of overhead in the
+ *	transfer buffer, no matter how many packets are being
+ *	transmitted.
+ */
+static int
+ncm_uno_tx_begin(struct usbnet *un)
+{
+	struct ncm_softc *sc = un->un_sc;
+
+	/*
+	 * Find how many bytes of overhead we need for:
+	 * - a transfer header
+	 * - a datagram pointer table header
+	 * - padding to align the first payload.
+	 */
+	return ncm_out_align(sc, sizeof(struct ncm_header16) +
+	    offsetof(struct ncm_dptab16, datagram[1]));
+}
+
+/*
+ * ncm_uno_tx_add(un, m, pktno, offset)
+ *
+ *	Called for each candidate packet in a batch for tx in a single
+ *	transfer.  Returns the number of bytes of overhead, on top of
+ *	m->m_pkthdr.len, to store this packet in the transfer buffer,
+ *	given that pktno packets are already included.
+ *
+ *	offset is the sum of ncm_uno_tx_begin(un) and
+ *	ncm_uno_tx_add(un, m0, ...) for each previous packet m0 already
+ *	included in the batch.
+ *
+ *	Caller guarantees the transfer buffer already has enough space
+ *	for offset + m->m_pkthdr.len bytes, i.e., that
+ *
+ *		offset + m->m_pkthdr.len <= un->un_tx_bufsz.
+ */
+static int
+ncm_uno_tx_add(struct usbnet *un, struct mbuf *m,
+    unsigned pktno, unsigned off)
+{
+	struct ncm_softc *sc = un->un_sc;
+	unsigned old_start, new_start;
+
+	KASSERTMSG(off <= un->un_tx_bufsz, "off=%u bufsz=%u",
+	    off, un->un_tx_bufsz);
+	KASSERTMSG((unsigned)m->m_pkthdr.len <= un->un_tx_bufsz - off,
+	    "off=%u m->m_pkthdr.len=%u bufsz=%u",
+	    off, (unsigned)m->m_pkthdr.len, un->un_tx_bufsz);
+
+	if (UGETW(sc->sc_ntb_param.wNtbOutMaxDatagrams) != 0 &&
+	    pktno >= (unsigned)UGETW(sc->sc_ntb_param.wNtbOutMaxDatagrams))
+		return -1;
+
+	/*
+	 * Find how many bytes we would need for header and padding to
+	 * align the first payload with the previous number of packets,
+	 * so we can subtract that from the number of bytes for the
+	 * next number of packets.
+	 *
+	 * Note that we always need one extra entry in the datagram
+	 * pointer table with zero index/length to terminate the list.
+	 */
+	old_start = ncm_out_align(sc, sizeof(struct ncm_header16) +
+	    offsetof(struct ncm_dptab16, datagram[pktno + 1]));
+	new_start = ncm_out_align(sc, sizeof(struct ncm_header16) +
+	    offsetof(struct ncm_dptab16, datagram[pktno + 2]));
+
+	return (new_start - old_start) + (ncm_out_align(sc, off) - off);
+}
+
+/*
+ * ncm_uno_tx_commit(un, m0, c, npkt, nbyte)
+ *
+ *	Called when a batch of npkt packets, totalling nbyte bytes of
+ *	payload and overhead (as returned by ncm_uno_tx_begin and
+ *	ncm_uno_tx_add), has been gathered for tx in a single transfer.
+ *	Formats the packets m0, m1 := m0->m_nextpkt, m2 :=
+ *	m1->m_nextpkt, ..., with appropriate headers into the buffer c.
+ */
+static void
+ncm_uno_tx_commit(struct usbnet *un, struct mbuf *m0, struct usbnet_chain *c,
+    unsigned npkt, unsigned nbyte)
+{
+	struct ncm_softc *sc = un->un_sc;
+	struct ncm_header16 *hdr;
+	struct ncm_dptab16 *ptr;
+	unsigned i, off, len;
+	struct mbuf *m;
+
+	KASSERTMSG(nbyte <= un->un_tx_bufsz, "nbyte=%u bufsz=%u",
+	    nbyte, un->un_tx_bufsz);
+
+	hdr = (struct ncm_header16 *)c->unc_buf;
+	ptr = (struct ncm_dptab16 *)(hdr + 1);
+	memset(hdr, 0, sizeof(*hdr));
 	memset(ptr, 0, sizeof(*ptr));
 
 	USETDW(hdr->dwSignature, NCM_HDR16_SIG);
 	USETW(hdr->wHeaderLength, sizeof(*hdr));
 	USETW(hdr->wSequence, sc->sc_tx_seq);
 	sc->sc_tx_seq++;
-
-	USETW(hdr->wBlockLength, hdr_len + len);
+	USETW(hdr->wBlockLength, nbyte);
 	USETW(hdr->wNdpIndex, sizeof(*hdr));
 
 	USETDW(ptr->dwSignature, NCM_PTR16_SIG_NO_CRC);
-	USETW(ptr->wLength, sizeof(*ptr));
+	USETW(ptr->wLength, offsetof(struct ncm_dptab16, datagram[npkt + 1]));
 	USETW(ptr->wNextNdpIndex, 0);
 
-	USETW(ptr->datagram[0].wDatagramIndex, hdr_len);
-	USETW(ptr->datagram[0].wDatagramLength, len);
+	off = (unsigned)((const uint8_t *)&ptr->datagram[npkt + 1] -
+	    c->unc_buf);
+	for (i = 0, m = m0; i < npkt; i++, m = m->m_nextpkt) {
+		KASSERTMSG(m != NULL, "i=%u npkt=%u", i, npkt);
+		KASSERTMSG(off <= nbyte, "off=%u nbyte=%u", off, nbyte);
+		off = ncm_out_align(sc, off);
+		KASSERTMSG(off <= nbyte, "off=%u nbyte=%u", off, nbyte);
+		len = (unsigned)m->m_pkthdr.len;
+		KASSERTMSG(len <= nbyte - off, "off=%u len=%u nbyte=%u",
+		    off, len, nbyte);
+		memset(&ptr->datagram[i], 0, sizeof(ptr->datagram[0]));
+		USETW(ptr->datagram[i].wDatagramIndex, off);
+		USETW(ptr->datagram[i].wDatagramLength, len);
+		m_copydata(m, 0, len, &c->unc_buf[off]);
+		off += len;
+	}
+	KASSERTMSG(i == npkt, "i=%u npkt=%u", i, npkt);
+	KASSERTMSG(off == nbyte, "off=%u nbyte=%u", off, nbyte);
 
-	USETW(ptr->datagram[1].wDatagramIndex, 0);
-	USETW(ptr->datagram[1].wDatagramLength, 0);
-
-	m_copydata(m, 0, len, c->unc_buf + hdr_len);
-
-	return len + hdr_len;
+	USETW(ptr->datagram[npkt].wDatagramIndex, 0);
+	USETW(ptr->datagram[npkt].wDatagramLength, 0);
 }
 
 #ifdef _MODULE
diff -r 823a0086cd5b sys/dev/usb/if_ncmreg.h
--- a/sys/dev/usb/if_ncmreg.h	Sun Jun 28 11:38:58 2026 +0000
+++ b/sys/dev/usb/if_ncmreg.h	Mon Jun 29 02:09:55 2026 +0000
@@ -42,7 +42,7 @@
 #define NCM_MIN_TX_BUFSZ					\
 	(ETHERMTU + ETHER_HDR_LEN				\
 	    + sizeof(struct ncm_header16)			\
-	    + sizeof(struct ncm_pointer16))
+	    + offsetof(struct ncm_dptab16, datagram[2]))
 
 struct ncm_header16 {
 #define	NCM_HDR16_SIG	0x484d434e
@@ -73,7 +73,7 @@ struct ncm_dptab16 {
 	uDWord			dwSignature;
 	uWord			wLength;
 	uWord			wNextNdpIndex;
-	struct ncm_pointer16	datagram[2];
+	struct ncm_pointer16	datagram[0];
 } UPACKED;
 
 struct ncm_pointer32 {
@@ -89,7 +89,7 @@ struct ncm_dptab32 {
 	uWord			wReserved6;
 	uWord			wNextNdpIndex;
 	uDWord			dwReserved12;
-	struct ncm_pointer32	datagram[2];
+	struct ncm_pointer32	datagram[0];
 } UPACKED;
 
 #define NCM_GET_NTB_PARAMETERS 0x80
diff -r 823a0086cd5b sys/dev/usb/if_ure.c
--- a/sys/dev/usb/if_ure.c	Sun Jun 28 11:38:58 2026 +0000
+++ b/sys/dev/usb/if_ure.c	Mon Jun 29 02:09:55 2026 +0000
@@ -95,6 +95,11 @@ static int	ure_uno_mii_write_reg(struct 
 static void	ure_uno_miibus_statchg(struct ifnet *);
 static unsigned ure_uno_tx_prepare(struct usbnet *, struct mbuf *,
 				   struct usbnet_chain *);
+static int	ure_uno_tx_begin(struct usbnet *);
+static int	ure_uno_tx_add(struct usbnet *, struct mbuf *,
+			       unsigned, unsigned);
+static void	ure_uno_tx_commit(struct usbnet *, struct mbuf *,
+				  struct usbnet_chain *, unsigned, unsigned);
 static void	ure_uno_rx_loop(struct usbnet *, struct usbnet_chain *,
 				uint32_t);
 static int	ure_uno_init(struct ifnet *);
@@ -105,7 +110,8 @@ static void	ure_attach(device_t, device_
 CFATTACH_DECL_NEW(ure, sizeof(struct usbnet), ure_match, ure_attach,
     usbnet_detach, usbnet_activate);
 
-static const struct usbnet_ops ure_ops = {
+static const struct usbnet_ops_v2 ure_ops = {
+    .v1 = {
 	.uno_stop = ure_uno_stop,
 	.uno_mcast = ure_uno_mcast,
 	.uno_read_reg = ure_uno_mii_read_reg,
@@ -114,6 +120,12 @@ static const struct usbnet_ops ure_ops =
 	.uno_tx_prepare = ure_uno_tx_prepare,
 	.uno_rx_loop = ure_uno_rx_loop,
 	.uno_init = ure_uno_init,
+    },
+    .v2 = {
+	.uno_tx_begin = ure_uno_tx_begin,
+	.uno_tx_add = ure_uno_tx_add,
+	.uno_tx_commit = ure_uno_tx_commit,
+    }
 };
 
 static int
@@ -818,7 +830,7 @@ ure_attach(device_t parent, device_t sel
 	un->un_dev = self;
 	un->un_udev = dev;
 	un->un_sc = un;
-	un->un_ops = &ure_ops;
+	un->un_ops = &ure_ops.v1;
 	un->un_rx_xfer_flags = USBD_SHORT_XFER_OK;
 	un->un_tx_xfer_flags = USBD_FORCE_SHORT_XFER;
 	un->un_rx_list_cnt = URE_RX_LIST_CNT;
@@ -864,7 +876,7 @@ ure_attach(device_t parent, device_t sel
 	}
 
 	/* Set these up now for ure_ctl().  */
-	usbnet_attach(un);
+	usbnet_attach_v2(un, &ure_ops);
 
 	un->un_phyno = 0;
 
@@ -1036,14 +1048,20 @@ ure_rxcsum(struct ifnet *ifp, struct ure
 }
 
 static unsigned
-ure_uno_tx_prepare(struct usbnet *un, struct mbuf *m, struct usbnet_chain *c)
+ure_uno_tx_prepare_offset(struct usbnet *un, struct mbuf *m,
+    struct usbnet_chain *c, unsigned offset)
 {
 	struct ure_txpkt txhdr;
 	uint32_t frm_len = 0;
-	uint8_t *buf = c->unc_buf;
+	uint8_t *buf = c->unc_buf + offset;
 
-	if ((unsigned)m->m_pkthdr.len > un->un_tx_bufsz - sizeof(txhdr))
-		return 0;
+	__CTASSERT(sizeof(txhdr) <= URE_TX_BUFSZ);
+	KASSERTMSG((sizeof(txhdr) <= un->un_tx_bufsz &&
+		offset <= un->un_tx_bufsz - sizeof(txhdr) &&
+		(unsigned)m->m_pkthdr.len <= (un->un_tx_bufsz - sizeof(txhdr) -
+		    offset)),
+	    "offset=%u sizeof(txhdr)=%zu len=%u bufsz=%u",
+	    offset, sizeof(txhdr), m->m_pkthdr.len, un->un_tx_bufsz);
 
 	/* header */
 	txhdr.ure_pktlen = htole32(m->m_pkthdr.len | URE_TXPKT_TX_FS |
@@ -1062,6 +1080,52 @@ ure_uno_tx_prepare(struct usbnet *un, st
 	return frm_len;
 }
 
+static unsigned
+ure_uno_tx_prepare(struct usbnet *un, struct mbuf *m, struct usbnet_chain *c)
+{
+
+	__CTASSERT(sizeof(struct ure_txpkt) <= URE_TX_BUFSZ);
+	KASSERTMSG(sizeof(struct ure_txpkt) <= un->un_tx_bufsz,
+	    "sizeof(struct ure_txpkt)=%zu bufsz=%u",
+	    sizeof(struct ure_txpkt), un->un_tx_bufsz);
+	if ((unsigned)m->m_pkthdr.len >
+	    un->un_tx_bufsz - sizeof(struct ure_txpkt))
+		return 0;
+
+	return ure_uno_tx_prepare_offset(un, m, c, 0);
+}
+
+static int
+ure_uno_tx_begin(struct usbnet *un)
+{
+
+	/* zero per-xfer overhead */
+	return 0;
+}
+
+static int
+ure_uno_tx_add(struct usbnet *un, struct mbuf *m,
+    unsigned pktno, unsigned offset)
+{
+
+	return sizeof(struct ure_txpkt);
+}
+
+static void
+ure_uno_tx_commit(struct usbnet *un, struct mbuf *m0, struct usbnet_chain *c,
+    unsigned npkt, unsigned nbyte)
+{
+	struct mbuf *m;
+	unsigned i, offset = 0;
+
+	for (i = 0, m = m0; i < npkt; i++, m = m->m_nextpkt) {
+		KASSERT(m != NULL);
+		offset += ure_uno_tx_prepare_offset(un, m, c, offset);
+		KASSERT(offset <= nbyte);
+	}
+	KASSERTMSG(offset == nbyte, "offset=%u nbyte=%u", offset, nbyte);
+}
+
 /*
  * We need to calculate L4 checksum in software, if the offset of
  * L4 header is larger than 0x7ff = 2047.
diff -r 823a0086cd5b sys/dev/usb/usbnet.c
--- a/sys/dev/usb/usbnet.c	Sun Jun 28 11:38:58 2026 +0000
+++ b/sys/dev/usb/usbnet.c	Mon Jun 29 02:09:55 2026 +0000
@@ -94,6 +94,11 @@ struct usbnet_private {
 	struct timeval		unp_rx_notice;
 	struct timeval		unp_tx_notice;
 	struct timeval		unp_intr_notice;
+
+	enum usbnet_version {
+		USBNET_V1 = 1,
+		USBNET_V2,
+	}			unp_version;
 };
 
 #define un_cdata(un)	(&(un)->un_pri->unp_cdata)
@@ -229,6 +234,60 @@ uno_tx_prepare(struct usbnet *un, struct
 	return (*un->un_ops->uno_tx_prepare)(un, m, c);
 }
 
+static const struct usbnet_v2_ops *
+usbnet_get_v2_ops(struct usbnet *un)
+{
+
+	KASSERT(un->un_pri->unp_version >= USBNET_V2);
+	switch (un->un_pri->unp_version) {
+	case USBNET_V2:
+		return &const_container_of(un->un_ops, struct usbnet_ops_v2,
+		    v1)->v2;
+	default:
+		__unreachable();
+	}
+}
+
+static int
+uno_tx_begin(struct usbnet *un)
+{
+	usbnet_isowned_tx(un);
+	return (*usbnet_get_v2_ops(un)->uno_tx_begin)(un);
+}
+
+static int
+uno_tx_add(struct usbnet *un, struct mbuf *m, unsigned pktno, unsigned offset)
+{
+	usbnet_isowned_tx(un);
+	return (*usbnet_get_v2_ops(un)->uno_tx_add)(un, m, pktno, offset);
+}
+
+static void
+uno_tx_abort(struct usbnet *un)
+{
+	usbnet_isowned_tx(un);
+	if (usbnet_get_v2_ops(un)->uno_tx_abort)
+		return (*usbnet_get_v2_ops(un)->uno_tx_abort)(un);
+}
+
+static void
+uno_tx_commit(struct usbnet *un, struct mbuf *m, struct usbnet_chain *c,
+    unsigned npkt, unsigned nbyte)
+{
+	usbnet_isowned_tx(un);
+	return (*usbnet_get_v2_ops(un)->uno_tx_commit)(un, m, c, npkt, nbyte);
+}
+
+static bool
+usbnet_batch_tx(struct usbnet *un)
+{
+
+	return un->un_pri->unp_version >= USBNET_V2 &&
+	    usbnet_get_v2_ops(un)->uno_tx_begin != NULL &&
+	    usbnet_get_v2_ops(un)->uno_tx_add != NULL &&
+	    usbnet_get_v2_ops(un)->uno_tx_commit != NULL;
+}
+
 static void
 uno_rx_loop(struct usbnet *un, struct usbnet_chain *c, uint32_t total_len)
 {
@@ -521,11 +580,16 @@ usbnet_start_locked(struct ifnet *ifp)
 	idx = cd->uncd_tx_prod;
 	count = 0;
 	while (cd->uncd_tx_cnt < un->un_tx_list_cnt) {
+		struct mbuf *batch = NULL, **batch_tail = &batch;
+		unsigned npkt = 0;
+
 		IFQ_DEQUEUE(&ifp->if_snd, m);
 		if (m == NULL) {
 			DPRINTF("start called, queue empty", 0, 0, 0, 0);
 			break;
 		}
+		npkt++;
+
 		if ((unsigned)m->m_pkthdr.len > un->un_tx_bufsz) {
 			DPRINTF("oversize packet, %ju > %ju",
 			    (unsigned)m->m_pkthdr.len, un->un_tx_bufsz, 0, 0);
@@ -540,6 +604,127 @@ usbnet_start_locked(struct ifnet *ifp)
 		struct usbnet_chain *const c = &cd->uncd_tx_chain[idx];
 		KASSERT(c->unc_xfer != NULL);
 
+		if (usbnet_batch_tx(un)) {
+			unsigned offset = 0;
+			int overhead;
+
+			overhead = uno_tx_begin(un);
+			KASSERTMSG((overhead == -1 ||
+				(unsigned)overhead <= un->un_tx_bufsz),
+			    "overhead=%d bufsz=%u", overhead, un->un_tx_bufsz);
+			if (overhead == -1) {
+				DPRINTF("uno_tx_begin failed", 0, 0, 0, 0);
+				if_statinc(ifp, if_oerrors);
+				m_freem(m);
+				continue;
+			}
+			if ((unsigned)overhead >
+			    (un->un_tx_bufsz - (unsigned)m->m_pkthdr.len)) {
+				DPRINTF("uno_tx_begin overhead too large",
+				    0, 0, 0, 0);
+				uno_tx_abort(un);
+				if_statinc(ifp, if_oerrors);
+				m_freem(m);
+				continue;
+			}
+			offset = (unsigned)overhead;
+
+			KASSERTMSG(((unsigned)m->m_pkthdr.len <=
+				un->un_tx_bufsz - offset),
+			    "offset=%u m->m_pkthdr.len=%d bufsz=%u",
+			    offset, m->m_pkthdr.len, un->un_tx_bufsz);
+			overhead = uno_tx_add(un, m, 0, offset);
+			KASSERTMSG((overhead == -1 ||
+				(unsigned)overhead <= un->un_tx_bufsz),
+			    "overhead=%d bufsz=%u",
+			    overhead, un->un_tx_bufsz);
+			if (overhead == -1 ||
+			    (unsigned)overhead > (un->un_tx_bufsz - offset -
+				(unsigned)m->m_pkthdr.len)) {
+				DPRINTF("uno_tx_add failed", 0, 0, 0, 0);
+				uno_tx_abort(un);
+				if_statinc(ifp, if_oerrors);
+				m_freem(m);
+				continue;
+			}
+			for (;;) {
+				/*
+				 * Loop invariant: We have a packet m,
+				 * and we know its overhead and length
+				 * both fit in the remaining buffer
+				 * space.
+				 *
+				 * Queue it up, and then examine the
+				 * next packet.
+				 */
+				KASSERT(m != NULL);
+				KASSERTMSG(((unsigned)overhead <=
+					un->un_tx_bufsz - offset),
+				    "offset=%u overhead=%d len=%d bufsz=%u",
+				    offset, overhead, m->m_pkthdr.len,
+				    un->un_tx_bufsz);
+				KASSERTMSG(((unsigned)m->m_pkthdr.len <=
+					un->un_tx_bufsz - offset -
+					(unsigned)overhead),
+				    "offset=%u overhead=%d len=%d bufsz=%u",
+				    offset, overhead, m->m_pkthdr.len,
+				    un->un_tx_bufsz);
+				offset += (unsigned)overhead;
+				offset += (unsigned)m->m_pkthdr.len;
+				KASSERTMSG(offset <= un->un_tx_bufsz,
+				    "offset=%u bufsz=%u",
+				    offset, un->un_tx_bufsz);
+				m->m_nextpkt = NULL;
+				*batch_tail = m;
+				batch_tail = &m->m_nextpkt;
+
+				IFQ_POLL(&ifp->if_snd, m);
+				if (m == NULL)
+					break;
+				KASSERTMSG(((unsigned)m->m_pkthdr.len <=
+					un->un_tx_bufsz),
+				    "m->m_pkthdr.len=%d bufsz=%u",
+				    m->m_pkthdr.len, un->un_tx_bufsz);
+				if ((unsigned)m->m_pkthdr.len >
+				    un->un_tx_bufsz - offset) {
+					/* packet payload too large */
+					m = NULL; /* leave in ifp->if_snd q */
+					break;
+				}
+				KASSERTMSG(((unsigned)m->m_pkthdr.len <=
+					un->un_tx_bufsz - offset),
+				    "offset=%u m->m_pkthdr.len=%d bufsz=%u",
+				    offset, m->m_pkthdr.len, un->un_tx_bufsz);
+				overhead = uno_tx_add(un, m, npkt, offset);
+				KASSERTMSG((overhead == -1 ||
+					(unsigned)overhead <= un->un_tx_bufsz),
+				    "overhead=%d bufsz=%u",
+				    overhead, un->un_tx_bufsz);
+				if (overhead == -1 ||
+				    (unsigned)overhead > (un->un_tx_bufsz -
+					offset - (unsigned)m->m_pkthdr.len)) {
+					/*
+					 * driver didn't like the
+					 * packet, or packet overhead
+					 * too large
+					 */
+					m = NULL; /* leave in ifp->if_snd q */
+					break;
+				}
+				IFQ_DEQUEUE(&ifp->if_snd, m);
+				npkt++;
+			}
+
+			/*
+			 * We have gathered a nonempty batch of packets
+			 * to transfer.  Ask the driver to fill up the
+			 * transfer buffer with the packets.
+			 */
+			uno_tx_commit(un, batch, c, npkt, offset);
+			length = offset;
+			goto xfer;
+		}
+
 		length = uno_tx_prepare(un, m, c);
 		if (length == 0) {
 			DPRINTF("uno_tx_prepare gave zero length", 0, 0, 0, 0);
@@ -548,6 +733,8 @@ usbnet_start_locked(struct ifnet *ifp)
 			continue;
 		}
 
+xfer:		KASSERTMSG(length <= un->un_tx_bufsz,
+		    "length=%u bufsz=%u", length, un->un_tx_bufsz);
 		usbd_setup_xfer(c->unc_xfer, c, c->unc_buf, length,
 		    un->un_tx_xfer_flags, 10000, usbnet_txeof);
 
@@ -556,18 +743,35 @@ usbnet_start_locked(struct ifnet *ifp)
 		if (err != USBD_IN_PROGRESS) {
 			DPRINTF("usbd_transfer on %#jx for %ju bytes: %jd",
 			    (uintptr_t)c->unc_buf, length, err, 0);
-			if_statinc(ifp, if_oerrors);
-			m_freem(m);
+			if_statadd(ifp, if_oerrors, npkt);
+			if (batch) {
+				while (batch != NULL) {
+					m = batch;
+					batch = m->m_nextpkt;
+					m_freem(m);
+				}
+			} else {
+				m_freem(m);
+			}
 			continue;
 		}
 		done_transmit = true;
 
 		/*
-		 * If there's a BPF listener, bounce a copy of this frame
-		 * to him.
+		 * If there's a BPF listener, bounce a copy of these
+		 * frames to them, and free them all.
 		 */
-		bpf_mtap(ifp, m, BPF_D_OUT);
-		m_freem(m);
+		if (batch) {
+			while (batch != NULL) {
+				m = batch;
+				batch = m->m_nextpkt;
+				bpf_mtap(ifp, m, BPF_D_OUT);
+				m_freem(m);
+			}
+		} else {
+			bpf_mtap(ifp, m, BPF_D_OUT);
+			m_freem(m);
+		}
 
 		idx = (idx + 1) % un->un_tx_list_cnt;
 		cd->uncd_tx_cnt++;
@@ -1439,8 +1643,8 @@ usbnet_empty_eaddr(struct usbnet * const
  * the un_eaddr[] is fully zero.
  */
 
-void
-usbnet_attach(struct usbnet *un)
+static void
+usbnet_attach_version(struct usbnet *un, enum usbnet_version ver)
 {
 	USBNETHIST_FUNC(); USBNETHIST_CALLED();
 
@@ -1480,6 +1684,23 @@ usbnet_attach(struct usbnet *un)
 	unp->unp_rxstopped = true;
 	unp->unp_txstopped = true;
 	unp->unp_attached = true;
+
+	unp->unp_version = ver;
+}
+
+void
+usbnet_attach(struct usbnet *un)
+{
+
+	usbnet_attach_version(un, USBNET_V1);
+}
+
+void
+usbnet_attach_v2(struct usbnet *un, const struct usbnet_ops_v2 *v2ops)
+{
+
+	KASSERT(un->un_ops == &v2ops->v1);
+	usbnet_attach_version(un, USBNET_V2);
 }
 
 static void
diff -r 823a0086cd5b sys/dev/usb/usbnet.h
--- a/sys/dev/usb/usbnet.h	Sun Jun 28 11:38:58 2026 +0000
+++ b/sys/dev/usb/usbnet.h	Mon Jun 29 02:09:55 2026 +0000
@@ -141,6 +141,12 @@ typedef void (*usbnet_mii_statchg_cb)(st
 /* Prepare packet to send callback, returns length. */
 typedef unsigned (*usbnet_tx_prepare_cb)(struct usbnet *, struct mbuf *,
 					 struct usbnet_chain *);
+typedef int (*usbnet_tx_begin_cb)(struct usbnet *);
+typedef int (*usbnet_tx_add_cb)(struct usbnet *, struct mbuf *,
+				unsigned, unsigned);
+typedef void (*usbnet_tx_abort_cb)(struct usbnet *);
+typedef void (*usbnet_tx_commit_cb)(struct usbnet *, struct mbuf *,
+				    struct usbnet_chain *, unsigned, unsigned);
 /* Receive some packets callback. */
 typedef void (*usbnet_rx_loop_cb)(struct usbnet *, struct usbnet_chain *,
 				  uint32_t);
@@ -183,6 +189,18 @@ struct usbnet_ops {
 	usbnet_intr_cb		uno_intr;		/* n */
 };
 
+struct usbnet_v2_ops {
+	usbnet_tx_begin_cb	uno_tx_begin;		/* T */
+	usbnet_tx_add_cb	uno_tx_add;		/* T */
+	usbnet_tx_abort_cb	uno_tx_abort;		/* T */
+	usbnet_tx_commit_cb	uno_tx_commit;		/* T */
+};
+
+struct usbnet_ops_v2 {
+	struct usbnet_ops	v1;
+	struct usbnet_v2_ops	v2;
+};
+
 /*
  * USB interrupt pipe support.  Use this if usbd_open_pipe_intr() should
  * be used for the interrupt pipe.
@@ -314,6 +332,7 @@ void	usbnet_input(struct usbnet * const,
 
 /* autoconf */
 void	usbnet_attach(struct usbnet *);
+void	usbnet_attach_v2(struct usbnet *, const struct usbnet_ops_v2 *);
 void	usbnet_attach_ifp(struct usbnet *, unsigned, unsigned,
 			  const struct usbnet_mii *);
 int	usbnet_detach(device_t, int);