initial port of a driver for the Neterion Xframe-I 10Gb Ethernet adapter.

From NetBSD

3 files changed, 1756 insertions, 1 deletions

diff --git a/sys/dev/pci/files.pci b/sys/dev/pci/files.pci
index 0bfbdd3797a..14949acd316 100644
--- a/sys/dev/pci/files.pci
+++ b/sys/dev/pci/files.pci
@@ -1,4 +1,4 @@
-#	$OpenBSD: files.pci,v 1.203 2006/04/26 15:53:08 jason Exp $
+#	$OpenBSD: files.pci,v 1.204 2006/05/01 00:34:12 brad Exp $
 #	$NetBSD: files.pci,v 1.20 1996/09/24 17:47:15 christos Exp $
 #
 # Config file and device description for machine-independent PCI code.
@@ -321,6 +321,11 @@ file	dev/pci/if_ixgb.c		ixgb
 file	dev/pci/ixgb_ee.c		ixgb
 file	dev/pci/ixgb_hw.c		ixgb
 
+# Neterion Xframe 10 Gigabit ethernet 
+device	xge: ether, ifnet, ifmedia
+attach	xge  at pci
+file	dev/pci/if_xge.c		xge
+
 # DEC/Intel 21143 and "tulip" clone ethernet
 attach	dc at pci with dc_pci
 file	dev/pci/if_dc_pci.c		dc_pci
diff --git a/sys/dev/pci/if_xge.c b/sys/dev/pci/if_xge.c
new file mode 100644
index 00000000000..6ef00961578
--- /dev/null
+++ b/sys/dev/pci/if_xge.c
@@ -0,0 +1,1323 @@
+/*	$OpenBSD: if_xge.c,v 1.1 2006/05/01 00:34:12 brad Exp $	*/
+/*	$NetBSD: if_xge.c,v 1.1 2005/09/09 10:30:27 ragge Exp $	*/
+
+/*
+ * Copyright (c) 2004, SUNET, Swedish University Computer Network.
+ * All rights reserved.
+ *
+ * Written by Anders Magnusson for SUNET, Swedish University Computer Network.
+ *
+ * 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.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *      This product includes software developed for the NetBSD Project by
+ *      SUNET, Swedish University Computer Network.
+ * 4. The name of SUNET may not be used to endorse or promote products
+ *    derived from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY SUNET ``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 SUNET
+ * 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.
+ */
+
+/*
+ * Device driver for the Neterion Xframe Ten Gigabit Ethernet controller.
+ *
+ * TODO (in no specific order):
+ *	HW VLAN support.
+ *	IPv6 HW cksum.
+ */
+
+#include <sys/cdefs.h>
+#if 0
+__KERNEL_RCSID(0, "$NetBSD: if_xge.c,v 1.1 2005/09/09 10:30:27 ragge Exp $");
+#endif
+
+#include "bpfilter.h"
+
+#include <sys/param.h>
+#include <sys/systm.h>
+#include <sys/sockio.h>
+#include <sys/mbuf.h>
+#include <sys/malloc.h>
+#include <sys/kernel.h>
+#include <sys/socket.h>
+#include <sys/device.h>
+
+#include <net/if.h>
+#include <net/if_dl.h>
+#include <net/if_media.h>
+
+#ifdef INET
+#include <netinet/in.h>
+#include <netinet/in_systm.h>
+#include <netinet/in_var.h>
+#include <netinet/ip.h>
+#include <netinet/if_ether.h>
+#include <netinet/tcp.h>
+#include <netinet/udp.h>
+#endif
+
+#if NBPFILTER > 0
+#include <net/bpf.h>
+#endif
+
+#include <machine/bus.h>
+#include <machine/intr.h>
+#include <machine/endian.h>
+
+#include <dev/mii/mii.h>
+#include <dev/mii/miivar.h>
+
+#include <dev/pci/pcivar.h>
+#include <dev/pci/pcireg.h>
+#include <dev/pci/pcidevs.h>
+
+#include <sys/lock.h>
+#include <sys/proc.h>
+
+#include <dev/pci/if_xgereg.h>
+
+/*
+ * Some tunable constants, tune with care!
+ */
+#define RX_MODE		RX_MODE_1  /* Receive mode (buffer usage, see below) */
+#define NRXDESCS	1016	   /* # of receive descriptors (requested) */
+#define NTXDESCS	8192	   /* Number of transmit descriptors */
+#define NTXFRAGS	100	   /* Max fragments per packet */
+
+/*
+ * Receive buffer modes; 1, 3 or 5 buffers.
+ */
+#define RX_MODE_1 1
+#define RX_MODE_3 3
+#define RX_MODE_5 5
+
+/*
+ * Use clever macros to avoid a bunch of #ifdef's.
+ */
+#define XCONCAT3(x,y,z) x ## y ## z
+#define CONCAT3(x,y,z) XCONCAT3(x,y,z)
+#define NDESC_BUFMODE CONCAT3(NDESC_,RX_MODE,BUFMODE)
+#define rxd_4k CONCAT3(rxd,RX_MODE,_4k)
+/* XXX */
+#if 0
+#define rxdesc ___CONCAT(rxd,RX_MODE)
+#endif
+#define rxdesc rxd1
+
+#define NEXTTX(x)	(((x)+1) % NTXDESCS)
+#define NRXFRAGS	RX_MODE /* hardware imposed frags */
+#define NRXPAGES	((NRXDESCS/NDESC_BUFMODE)+1)
+#define NRXREAL		(NRXPAGES*NDESC_BUFMODE)
+#define RXMAPSZ		(NRXPAGES*PAGE_SIZE)
+
+/*
+ * Magics to fix a bug when the mac address can't be read correctly.
+ * Comes from the Linux driver.
+ */
+static uint64_t fix_mac[] = {
+	0x0060000000000000ULL, 0x0060600000000000ULL,
+	0x0040600000000000ULL, 0x0000600000000000ULL,
+	0x0020600000000000ULL, 0x0060600000000000ULL,
+	0x0020600000000000ULL, 0x0060600000000000ULL,
+	0x0020600000000000ULL, 0x0060600000000000ULL,
+	0x0020600000000000ULL, 0x0060600000000000ULL,
+	0x0020600000000000ULL, 0x0060600000000000ULL,
+	0x0020600000000000ULL, 0x0060600000000000ULL,
+	0x0020600000000000ULL, 0x0060600000000000ULL,
+	0x0020600000000000ULL, 0x0060600000000000ULL,
+	0x0020600000000000ULL, 0x0060600000000000ULL,
+	0x0020600000000000ULL, 0x0060600000000000ULL,
+	0x0020600000000000ULL, 0x0000600000000000ULL,
+	0x0040600000000000ULL, 0x0060600000000000ULL,
+};
+
+
+struct xge_softc {
+	struct device sc_dev;
+	struct arpcom sc_arpcom;
+#define sc_if sc_arpcom.ac_if
+	bus_dma_tag_t sc_dmat;
+	bus_space_tag_t sc_st;
+	bus_space_handle_t sc_sh;
+	bus_space_tag_t sc_txt;
+	bus_space_handle_t sc_txh;
+	void *sc_ih;
+
+	struct ifmedia xena_media;
+	pcireg_t sc_pciregs[16];
+
+	/* Transmit structures */
+	struct txd *sc_txd[NTXDESCS];	/* transmit frags array */
+	bus_addr_t sc_txdp[NTXDESCS];	/* bus address of transmit frags */
+	bus_dmamap_t sc_txm[NTXDESCS];	/* transmit frags map */
+	struct mbuf *sc_txb[NTXDESCS];	/* transmit mbuf pointer */
+	int sc_nexttx, sc_lasttx;
+	bus_dmamap_t sc_txmap;		/* transmit descriptor map */
+
+	/* Receive data */
+	bus_dmamap_t sc_rxmap;		/* receive descriptor map */
+	struct rxd_4k *sc_rxd_4k[NRXPAGES]; /* receive desc pages */
+	bus_dmamap_t sc_rxm[NRXREAL];	/* receive buffer map */
+	struct mbuf *sc_rxb[NRXREAL];	/* mbufs on receive descriptors */
+	int sc_nextrx;			/* next descriptor to check */
+};
+
+int xge_match(struct device *, void *, void *);
+void xge_attach(struct device *, struct device *, void *);
+int xge_alloc_txmem(struct xge_softc *);
+int xge_alloc_rxmem(struct xge_softc *);
+void xge_start(struct ifnet *);
+void xge_stop(struct ifnet *, int);
+int xge_add_rxbuf(struct xge_softc *, int);
+void xge_mcast_filter(struct xge_softc *);
+int xge_setup_xgxs(struct xge_softc *);
+int xge_ioctl(struct ifnet *, u_long, caddr_t);
+int xge_init(struct ifnet *);
+void xge_ifmedia_status(struct ifnet *, struct ifmediareq *);
+int xge_xgmii_mediachange(struct ifnet *);
+void xge_enable(struct xge_softc *);
+int xge_intr(void  *);
+
+/*
+ * Helpers to address registers.
+ */
+#define PIF_WCSR(csr, val)	pif_wcsr(sc, csr, val)
+#define PIF_RCSR(csr)		pif_rcsr(sc, csr)
+#define TXP_WCSR(csr, val)	txp_wcsr(sc, csr, val)
+#define PIF_WKEY(csr, val)	pif_wkey(sc, csr, val)
+
+static inline void
+pif_wcsr(struct xge_softc *sc, bus_size_t csr, uint64_t val)
+{
+	uint32_t lval, hval;
+
+	lval = val&0xffffffff;
+	hval = val>>32;
+	bus_space_write_4(sc->sc_st, sc->sc_sh, csr, lval); 
+	bus_space_write_4(sc->sc_st, sc->sc_sh, csr+4, hval);
+}
+
+static inline uint64_t
+pif_rcsr(struct xge_softc *sc, bus_size_t csr)
+{
+	uint64_t val, val2;
+	val = bus_space_read_4(sc->sc_st, sc->sc_sh, csr);
+	val2 = bus_space_read_4(sc->sc_st, sc->sc_sh, csr+4);
+	val |= (val2 << 32);
+	return val;
+}
+
+static inline void
+txp_wcsr(struct xge_softc *sc, bus_size_t csr, uint64_t val)
+{
+	uint32_t lval, hval;
+
+	lval = val&0xffffffff;
+	hval = val>>32;
+	bus_space_write_4(sc->sc_txt, sc->sc_txh, csr, lval); 
+	bus_space_write_4(sc->sc_txt, sc->sc_txh, csr+4, hval);
+}
+
+
+static inline void
+pif_wkey(struct xge_softc *sc, bus_size_t csr, uint64_t val)
+{
+	uint32_t lval, hval;
+
+	lval = val&0xffffffff;
+	hval = val>>32;
+	PIF_WCSR(RMAC_CFG_KEY, RMAC_KEY_VALUE);
+	bus_space_write_4(sc->sc_st, sc->sc_sh, csr, lval); 
+	PIF_WCSR(RMAC_CFG_KEY, RMAC_KEY_VALUE);
+	bus_space_write_4(sc->sc_st, sc->sc_sh, csr+4, hval);
+}
+
+struct cfattach xge_ca = {
+	sizeof(struct xge_softc), xge_match, xge_attach
+};
+
+struct cfdriver xge_cd = {
+	0, "xge", DV_IFNET
+};
+
+#define XNAME sc->sc_dev.dv_xname
+
+#define XGE_RXSYNC(desc, what) \
+	bus_dmamap_sync(sc->sc_dmat, sc->sc_rxmap, \
+	(desc/NDESC_BUFMODE) * XGE_PAGE + sizeof(struct rxdesc) * \
+	(desc%NDESC_BUFMODE), sizeof(struct rxdesc), what)
+#define XGE_RXD(desc)	&sc->sc_rxd_4k[desc/NDESC_BUFMODE]-> \
+	r4_rxd[desc%NDESC_BUFMODE]
+
+/*
+ * Non-tunable constants.
+ */
+#define XGE_MAX_MTU		9600
+
+const struct pci_matchid xge_devices[] = {
+	{ PCI_VENDOR_NETERION, PCI_PRODUCT_NETERION_XFRAME },
+};
+
+int
+xge_match(struct device *parent, void *match, void *aux)
+{
+	return (pci_matchbyid((struct pci_attach_args *)aux, xge_devices,
+	    sizeof(xge_devices)/sizeof(xge_devices[0])));
+}
+
+void
+xge_attach(struct device *parent, struct device *self, void *aux)
+{
+	struct pci_attach_args *pa = aux;
+	struct xge_softc *sc;
+	struct ifnet *ifp;
+	pcireg_t memtype;
+	pci_intr_handle_t ih;
+	const char *intrstr = NULL;
+	pci_chipset_tag_t pc = pa->pa_pc;
+	uint8_t enaddr[ETHER_ADDR_LEN];
+	uint64_t val;
+	int i;
+
+	sc = (struct xge_softc *)self;
+
+	sc->sc_dmat = pa->pa_dmat;
+
+	/* Get BAR0 address */
+	memtype = pci_mapreg_type(pa->pa_pc, pa->pa_tag, XGE_PIF_BAR);
+	if (pci_mapreg_map(pa, XGE_PIF_BAR, memtype, 0,
+	    &sc->sc_st, &sc->sc_sh, 0, 0, 0)) {
+		printf("%s: unable to map PIF BAR registers\n", XNAME);
+		return;
+	}
+
+	memtype = pci_mapreg_type(pa->pa_pc, pa->pa_tag, XGE_TXP_BAR);
+	if (pci_mapreg_map(pa, XGE_TXP_BAR, memtype, 0,
+	    &sc->sc_txt, &sc->sc_txh, 0, 0, 0)) {
+		printf("%s: unable to map TXP BAR registers\n", XNAME);
+		return;
+	}
+
+	/* Save PCI config space */
+	for (i = 0; i < 64; i += 4)
+		sc->sc_pciregs[i/4] = pci_conf_read(pa->pa_pc, pa->pa_tag, i);
+
+#if BYTE_ORDER == LITTLE_ENDIAN
+	val = (uint64_t)0xFFFFFFFFFFFFFFFFULL;
+	val &= ~(TxF_R_SE|RxF_W_SE);
+	PIF_WCSR(SWAPPER_CTRL, val);
+	PIF_WCSR(SWAPPER_CTRL, val);
+#elif BYTE_ORDER == BIG_ENDIAN
+	/* do nothing */
+#else
+#error bad endianness!
+#endif
+
+	if ((val = PIF_RCSR(PIF_RD_SWAPPER_Fb)) != SWAPPER_MAGIC) {
+		printf("%s: failed configuring endian, %llx != %llx!\n",
+		    XNAME, (unsigned long long)val, SWAPPER_MAGIC);
+		return;
+	}
+
+	/*
+	 * The MAC addr may be all FF's, which is not good.
+	 * Resolve it by writing some magics to GPIO_CONTROL and 
+	 * force a chip reset to read in the serial eeprom again.
+	 */
+	for (i = 0; i < sizeof(fix_mac)/sizeof(fix_mac[0]); i++) {
+		PIF_WCSR(GPIO_CONTROL, fix_mac[i]);
+		PIF_RCSR(GPIO_CONTROL);
+	}
+
+	/*
+	 * Reset the chip and restore the PCI registers.
+	 */
+	PIF_WCSR(SW_RESET, 0xa5a5a50000000000ULL);
+	DELAY(500000);
+	for (i = 0; i < 64; i += 4)
+		pci_conf_write(pa->pa_pc, pa->pa_tag, i, sc->sc_pciregs[i/4]);
+
+	/*
+	 * Restore the byte order registers.
+	 */
+#if BYTE_ORDER == LITTLE_ENDIAN
+	val = (uint64_t)0xFFFFFFFFFFFFFFFFULL;
+	val &= ~(TxF_R_SE|RxF_W_SE);
+	PIF_WCSR(SWAPPER_CTRL, val);
+	PIF_WCSR(SWAPPER_CTRL, val);
+#elif BYTE_ORDER == BIG_ENDIAN
+	/* do nothing */
+#else
+#error bad endianness!
+#endif
+
+	if ((val = PIF_RCSR(PIF_RD_SWAPPER_Fb)) != SWAPPER_MAGIC) {
+		printf("%s: failed configuring endian2, %llx != %llx!\n",
+		    XNAME, (unsigned long long)val, SWAPPER_MAGIC);
+		return;
+	}
+
+	/*
+	 * XGXS initialization.
+	 */
+	/* 29, reset */
+	PIF_WCSR(SW_RESET, 0);
+	DELAY(500000);
+
+	/* 30, configure XGXS transceiver */
+	xge_setup_xgxs(sc);
+
+	/* 33, program MAC address (not needed here) */
+	/* Get ethernet address */
+	PIF_WCSR(RMAC_ADDR_CMD_MEM,
+	    RMAC_ADDR_CMD_MEM_STR|RMAC_ADDR_CMD_MEM_OFF(0));
+	while (PIF_RCSR(RMAC_ADDR_CMD_MEM) & RMAC_ADDR_CMD_MEM_STR)
+		;
+	val = PIF_RCSR(RMAC_ADDR_DATA0_MEM);
+	for (i = 0; i < ETHER_ADDR_LEN; i++)
+		enaddr[i] = (uint8_t)(val >> (56 - (8*i)));
+
+	/*
+	 * Get memory for transmit descriptor lists.
+	 */
+	if (xge_alloc_txmem(sc)) {
+		printf("%s: failed allocating txmem.\n", XNAME);
+		return;
+	}
+
+	/* 9 and 10 - set FIFO number/prio */
+	PIF_WCSR(TX_FIFO_P0, TX_FIFO_LEN0(NTXDESCS));
+	PIF_WCSR(TX_FIFO_P1, 0ULL);
+	PIF_WCSR(TX_FIFO_P2, 0ULL);
+	PIF_WCSR(TX_FIFO_P3, 0ULL);
+
+	/* 11, XXX set round-robin prio? */
+
+	/* 12, enable transmit FIFO */
+	val = PIF_RCSR(TX_FIFO_P0);
+	val |= TX_FIFO_ENABLE;
+	PIF_WCSR(TX_FIFO_P0, val);
+
+	/* 13, disable some error checks */
+	PIF_WCSR(TX_PA_CFG,
+	    TX_PA_CFG_IFR|TX_PA_CFG_ISO|TX_PA_CFG_ILC|TX_PA_CFG_ILE);
+
+	/* Create transmit DMA maps */
+	for (i = 0; i < NTXDESCS; i++) {
+		if (bus_dmamap_create(sc->sc_dmat, XGE_MAX_MTU,
+		    NTXFRAGS, MCLBYTES, 0, 0, &sc->sc_txm[i])) {
+			printf("%s: cannot create TX DMA maps\n", XNAME);
+			return;
+		}
+	}
+
+	sc->sc_lasttx = NTXDESCS-1;
+
+	/*
+	 * RxDMA initialization.
+	 * Only use one out of 8 possible receive queues.
+	 */
+	/* allocate rx descriptor memory */
+	if (xge_alloc_rxmem(sc)) {
+		printf("%s: failed allocating rxmem\n", XNAME);
+		return;
+	}
+
+	/* Create receive buffer DMA maps */
+	for (i = 0; i < NRXREAL; i++) {
+		if (bus_dmamap_create(sc->sc_dmat, XGE_MAX_MTU,
+		    NRXFRAGS, MCLBYTES, 0, 0, &sc->sc_rxm[i])) {
+			printf("%s: cannot create RX DMA maps\n", XNAME);
+			return;
+		}
+	}
+
+	/* allocate mbufs to receive descriptors */
+	for (i = 0; i < NRXREAL; i++)
+		if (xge_add_rxbuf(sc, i))
+			panic("out of mbufs too early");
+
+	/* 14, setup receive ring priority */
+	PIF_WCSR(RX_QUEUE_PRIORITY, 0ULL); /* only use one ring */
+
+	/* 15, setup receive ring round-robin calendar */
+	PIF_WCSR(RX_W_ROUND_ROBIN_0, 0ULL); /* only use one ring */
+	PIF_WCSR(RX_W_ROUND_ROBIN_1, 0ULL);
+	PIF_WCSR(RX_W_ROUND_ROBIN_2, 0ULL);
+	PIF_WCSR(RX_W_ROUND_ROBIN_3, 0ULL);
+	PIF_WCSR(RX_W_ROUND_ROBIN_4, 0ULL);
+
+	/* 16, write receive ring start address */
+	PIF_WCSR(PRC_RXD0_0, (uint64_t)sc->sc_rxmap->dm_segs[0].ds_addr);
+	/* PRC_RXD0_[1-7] are not used */
+
+	/* 17, Setup alarm registers */
+	PIF_WCSR(PRC_ALARM_ACTION, 0ULL); /* Default everything to retry */
+
+	/* 18, init receive ring controller */
+#if RX_MODE == RX_MODE_1
+	val = RING_MODE_1;
+#elif RX_MODE == RX_MODE_3
+	val = RING_MODE_3;
+#else /* RX_MODE == RX_MODE_5 */
+	val = RING_MODE_5;
+#endif
+	PIF_WCSR(PRC_CTRL_0, RC_IN_SVC|val);
+	/* leave 1-7 disabled */
+	/* XXXX snoop configuration? */
+
+	/* 19, set chip memory assigned to the queue */
+	PIF_WCSR(RX_QUEUE_CFG, MC_QUEUE(0, 64)); /* all 64M to queue 0 */
+
+	/* 20, setup RLDRAM parameters */
+	/* do not touch it for now */
+
+	/* 21, setup pause frame thresholds */
+	/* so not touch the defaults */
+	/* XXX - must 0xff be written as stated in the manual? */
+
+	/* 22, configure RED */
+	/* we do not want to drop packets, so ignore */
+
+	/* 23, initiate RLDRAM */
+	val = PIF_RCSR(MC_RLDRAM_MRS);
+	val |= MC_QUEUE_SIZE_ENABLE|MC_RLDRAM_MRS_ENABLE;
+	PIF_WCSR(MC_RLDRAM_MRS, val);
+	DELAY(1000);
+
+	/*
+	 * Setup interrupt policies.
+	 */
+	/* 40, Transmit interrupts */
+	PIF_WCSR(TTI_DATA1_MEM, TX_TIMER_VAL(0x1ff) | TX_TIMER_AC |
+	    TX_URNG_A(5) | TX_URNG_B(20) | TX_URNG_C(48));
+	PIF_WCSR(TTI_DATA2_MEM,
+	    TX_UFC_A(25) | TX_UFC_B(64) | TX_UFC_C(128) | TX_UFC_D(512));
+	PIF_WCSR(TTI_COMMAND_MEM, TTI_CMD_MEM_WE | TTI_CMD_MEM_STROBE);
+	while (PIF_RCSR(TTI_COMMAND_MEM) & TTI_CMD_MEM_STROBE)
+		;
+
+	/* 41, Receive interrupts */
+	PIF_WCSR(RTI_DATA1_MEM, RX_TIMER_VAL(0x800) | RX_TIMER_AC |
+	    RX_URNG_A(5) | RX_URNG_B(20) | RX_URNG_C(50));
+	PIF_WCSR(RTI_DATA2_MEM,
+	    RX_UFC_A(64) | RX_UFC_B(128) | RX_UFC_C(256) | RX_UFC_D(512));
+	PIF_WCSR(RTI_COMMAND_MEM, RTI_CMD_MEM_WE | RTI_CMD_MEM_STROBE);
+	while (PIF_RCSR(RTI_COMMAND_MEM) & RTI_CMD_MEM_STROBE)
+		;
+
+	/*
+	 * Setup media stuff.
+	 */
+	ifmedia_init(&sc->xena_media, IFM_IMASK, xge_xgmii_mediachange,
+	    xge_ifmedia_status);
+	ifmedia_add(&sc->xena_media, IFM_ETHER|IFM_1000_SX, 0, NULL);
+	ifmedia_set(&sc->xena_media, IFM_ETHER|IFM_1000_SX);
+
+	printf(", address %s\n", ether_sprintf(enaddr));
+
+	ifp = &sc->sc_arpcom.ac_if;
+	strlcpy(ifp->if_xname, XNAME, IFNAMSIZ);
+	strlcpy(sc->sc_arpcom.ac_enaddr, enaddr, ETHER_ADDR_LEN);
+	ifp->if_baudrate = 1000000000;
+	ifp->if_softc = sc;
+	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
+	ifp->if_ioctl = xge_ioctl;
+	ifp->if_start = xge_start;
+	IFQ_SET_MAXLEN(&ifp->if_snd, NTXDESCS - 1);
+	IFQ_SET_READY(&ifp->if_snd);
+
+	ifp->if_capabilities = IFCAP_VLAN_MTU;
+
+	/*
+	 * Attach the interface.
+	 */
+	if_attach(ifp);
+	ether_ifattach(ifp);
+
+	/*
+	 * Setup interrupt vector before initializing.
+	 */
+	if (pci_intr_map(pa, &ih)) {
+		printf("%s: unable to map interrupt\n",
+		    XNAME);
+		return;
+	}
+	intrstr = pci_intr_string(pc, ih);
+	if ((sc->sc_ih =
+	    pci_intr_establish(pc, ih, IPL_NET, xge_intr, sc, XNAME)) == NULL) {
+		printf("%s: unable to establish interrupt at %s\n",
+		    XNAME, intrstr ? intrstr : "<unknown>");
+		return;
+	    }
+	printf("%s: interrupting at %s\n", XNAME, intrstr);
+}
+
+void
+xge_ifmedia_status(struct ifnet *ifp, struct ifmediareq *ifmr)
+{
+	struct xge_softc *sc = ifp->if_softc;
+	uint64_t reg;
+
+	ifmr->ifm_status = IFM_AVALID;
+	ifmr->ifm_active = IFM_ETHER|IFM_1000_SX;
+
+	reg = PIF_RCSR(ADAPTER_STATUS);
+	if ((reg & (RMAC_REMOTE_FAULT|RMAC_LOCAL_FAULT)) == 0)	
+		ifmr->ifm_status |= IFM_ACTIVE;
+}
+
+int
+xge_xgmii_mediachange(struct ifnet *ifp)
+{
+	return 0;
+}
+
+void
+xge_enable(struct xge_softc *sc)
+{
+	uint64_t val;
+
+	/* 2, enable adapter */
+	val = PIF_RCSR(ADAPTER_CONTROL);
+	val |= ADAPTER_EN;
+	PIF_WCSR(ADAPTER_CONTROL, val);
+
+	/* 3, light the card enable led */
+	val = PIF_RCSR(ADAPTER_CONTROL);
+	val |= LED_ON;
+	PIF_WCSR(ADAPTER_CONTROL, val);
+	printf("%s: link up\n", XNAME);
+
+}
+
+int 
+xge_init(struct ifnet *ifp)
+{
+	struct xge_softc *sc = ifp->if_softc;
+	uint64_t val;
+
+	/* 31+32, setup MAC config */
+	PIF_WKEY(MAC_CFG, TMAC_EN|RMAC_EN|TMAC_APPEND_PAD|RMAC_STRIP_FCS|
+	    RMAC_BCAST_EN|RMAC_DISCARD_PFRM|RMAC_PROM_EN);
+
+	DELAY(1000);
+
+	/* 54, ensure that the adapter is 'quiescent' */
+	val = PIF_RCSR(ADAPTER_STATUS);
+	if ((val & QUIESCENT) != QUIESCENT) {
+#if 0
+		char buf[200];
+#endif
+		printf("%s: adapter not quiescent, aborting\n", XNAME);
+		val = (val & QUIESCENT) ^ QUIESCENT;
+#if 0
+		bitmask_snprintf(val, QUIESCENT_BMSK, buf, sizeof buf);
+		printf("%s: ADAPTER_STATUS missing bits %s\n", XNAME, buf);
+#endif
+		return 1;
+	}
+
+	/* 56, enable the transmit laser */
+	val = PIF_RCSR(ADAPTER_CONTROL);
+	val |= EOI_TX_ON;
+	PIF_WCSR(ADAPTER_CONTROL, val);
+
+	xge_enable(sc);
+	/*
+	 * Enable all interrupts
+	 */
+	PIF_WCSR(TX_TRAFFIC_MASK, 0);
+	PIF_WCSR(RX_TRAFFIC_MASK, 0);
+	PIF_WCSR(GENERAL_INT_MASK, 0);
+	PIF_WCSR(TXPIC_INT_MASK, 0);
+	PIF_WCSR(RXPIC_INT_MASK, 0);
+	PIF_WCSR(MAC_INT_MASK, MAC_TMAC_INT); /* only from RMAC */
+	PIF_WCSR(MAC_RMAC_ERR_MASK, ~RMAC_LINK_STATE_CHANGE_INT);
+
+	/* Done... */
+	ifp->if_flags |= IFF_RUNNING;
+	ifp->if_flags &= ~IFF_OACTIVE;
+
+	return 0;
+}
+
+void
+xge_stop(struct ifnet *ifp, int disable)
+{
+	struct xge_softc *sc = ifp->if_softc;
+	uint64_t val;
+
+	val = PIF_RCSR(ADAPTER_CONTROL);
+	val &= ~ADAPTER_EN;
+	PIF_WCSR(ADAPTER_CONTROL, val);
+
+	while ((PIF_RCSR(ADAPTER_STATUS) & QUIESCENT) != QUIESCENT)
+		;
+}
+
+int
+xge_intr(void *pv)
+{
+	struct xge_softc *sc = pv;
+	struct txd *txd;
+	struct ifnet *ifp = &sc->sc_if;
+	bus_dmamap_t dmp;
+	uint64_t val;
+	int i, lasttx, plen;
+
+	val = PIF_RCSR(GENERAL_INT_STATUS);
+	if (val == 0)
+		return 0; /* no interrupt here */
+
+	PIF_WCSR(GENERAL_INT_STATUS, val);
+
+	if ((val = PIF_RCSR(MAC_RMAC_ERR_REG)) & RMAC_LINK_STATE_CHANGE_INT) {
+		/* Wait for quiescence */
+		printf("%s: link down\n", XNAME);
+		while ((PIF_RCSR(ADAPTER_STATUS) & QUIESCENT) != QUIESCENT)
+			;
+		PIF_WCSR(MAC_RMAC_ERR_REG, RMAC_LINK_STATE_CHANGE_INT);
+			
+		val = PIF_RCSR(ADAPTER_STATUS);
+		if ((val & (RMAC_REMOTE_FAULT|RMAC_LOCAL_FAULT)) == 0)
+			xge_enable(sc); /* Only if link restored */
+	}
+
+	if ((val = PIF_RCSR(TX_TRAFFIC_INT)))
+		PIF_WCSR(TX_TRAFFIC_INT, val); /* clear interrupt bits */
+	/*
+	 * Collect sent packets.
+	 */
+	lasttx = sc->sc_lasttx;
+	while ((i = NEXTTX(sc->sc_lasttx)) != sc->sc_nexttx) {
+		txd = sc->sc_txd[i];
+		dmp = sc->sc_txm[i];
+
+		bus_dmamap_sync(sc->sc_dmat, dmp, 0,
+		    dmp->dm_mapsize,
+		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
+
+		if (txd->txd_control1 & TXD_CTL1_OWN) {
+			bus_dmamap_sync(sc->sc_dmat, dmp, 0,
+			    dmp->dm_mapsize, BUS_DMASYNC_PREREAD);
+			break;
+		}
+		bus_dmamap_unload(sc->sc_dmat, dmp);
+		m_freem(sc->sc_txb[i]);
+		ifp->if_opackets++;
+		sc->sc_lasttx = i;
+	}
+
+	if (sc->sc_lasttx != lasttx)
+		ifp->if_flags &= ~IFF_OACTIVE;
+
+	/* Try to get more packets on the wire */
+	xge_start(ifp);
+
+	/* clear interrupt bits */
+	if ((val = PIF_RCSR(RX_TRAFFIC_INT)))
+		PIF_WCSR(RX_TRAFFIC_INT, val);
+
+	for (;;) {
+		struct rxdesc *rxd;
+		struct mbuf *m;
+
+		XGE_RXSYNC(sc->sc_nextrx,
+		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
+
+		rxd = XGE_RXD(sc->sc_nextrx);
+		if (rxd->rxd_control1 & RXD_CTL1_OWN) {
+			XGE_RXSYNC(sc->sc_nextrx, BUS_DMASYNC_PREREAD);
+			break;
+		}
+
+		/* got a packet */
+		m = sc->sc_rxb[sc->sc_nextrx];
+#if RX_MODE == RX_MODE_1
+		plen = m->m_len = RXD_CTL2_BUF0SIZ(rxd->rxd_control2);
+#elif RX_MODE == RX_MODE_3
+#error Fix rxmodes in xge_intr
+#elif RX_MODE == RX_MODE_5
+		plen = m->m_len = RXD_CTL2_BUF0SIZ(rxd->rxd_control2);
+		plen += m->m_next->m_len = RXD_CTL2_BUF1SIZ(rxd->rxd_control2);
+		plen += m->m_next->m_next->m_len =
+		    RXD_CTL2_BUF2SIZ(rxd->rxd_control2);
+		plen += m->m_next->m_next->m_next->m_len =
+		    RXD_CTL3_BUF3SIZ(rxd->rxd_control3);
+		plen += m->m_next->m_next->m_next->m_next->m_len =
+		    RXD_CTL3_BUF4SIZ(rxd->rxd_control3);
+#endif
+		m->m_pkthdr.rcvif = ifp;
+		m->m_pkthdr.len = plen;
+
+		val = rxd->rxd_control1;
+
+		if (xge_add_rxbuf(sc, sc->sc_nextrx)) {
+			/* Failed, recycle this mbuf */
+#if RX_MODE == RX_MODE_1
+			rxd->rxd_control2 = RXD_MKCTL2(MCLBYTES, 0, 0);
+			rxd->rxd_control1 = RXD_CTL1_OWN;
+#elif RX_MODE == RX_MODE_3
+#elif RX_MODE == RX_MODE_5
+#endif
+			XGE_RXSYNC(sc->sc_nextrx,
+			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
+			ifp->if_ierrors++;
+			break;
+		}
+
+		ifp->if_ipackets++;
+
+#if XGE_CKSUM
+		if (RXD_CTL1_PROTOS(val) & (RXD_CTL1_P_IPv4|RXD_CTL1_P_IPv6)) {
+			m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
+			if (RXD_CTL1_L3CSUM(val) != 0xffff)
+				m->m_pkthdr.csum_flags |= M_CSUM_IPv4_BAD;
+		}
+		if (RXD_CTL1_PROTOS(val) & RXD_CTL1_P_TCP) {
+			m->m_pkthdr.csum_flags |= M_CSUM_TCPv4|M_CSUM_TCPv6;
+			if (RXD_CTL1_L4CSUM(val) != 0xffff)
+				m->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
+		}
+		if (RXD_CTL1_PROTOS(val) & RXD_CTL1_P_UDP) {
+			m->m_pkthdr.csum_flags |= M_CSUM_UDPv4|M_CSUM_UDPv6;
+			if (RXD_CTL1_L4CSUM(val) != 0xffff)
+				m->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
+		}
+#endif
+
+#if NBPFILTER > 0
+		if (ifp->if_bpf)
+			bpf_mtap(ifp->if_bpf, m, BPF_DIRECTION_IN);
+#endif /* NBPFILTER > 0 */
+
+		ether_input_mbuf(ifp, m);
+
+		if (++sc->sc_nextrx == NRXREAL)
+			sc->sc_nextrx = 0;
+
+	}
+
+	return 0;
+}
+
+int 
+xge_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
+{
+	struct xge_softc *sc = ifp->if_softc;
+	struct ifreq *ifr = (struct ifreq *) data;
+	struct ifaddr  *ifa = (struct ifaddr *)data;
+	int s, error = 0;
+
+	s = splnet();
+
+	if ((error = ether_ioctl(ifp, &sc->sc_arpcom, cmd, data)) > 0) {
+		splx(s);
+		return (error);
+	}
+
+	switch (cmd) {
+	case SIOCSIFADDR:
+		ifp->if_flags |= IFF_UP;
+		if (!(ifp->if_flags & IFF_RUNNING))
+			xge_init(ifp);
+#ifdef INET
+		if (ifa->ifa_addr->sa_family == AF_INET)
+			arp_ifinit(&sc->sc_arpcom, ifa);
+#endif /* INET */
+		break;
+	case SIOCSIFMTU:
+		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > XGE_MAX_MTU) {
+			error = EINVAL;
+		} else if (ifp->if_mtu != ifr->ifr_mtu) {
+			PIF_WCSR(RMAC_MAX_PYLD_LEN,
+			    RMAC_PYLD_LEN(ifr->ifr_mtu));
+			ifp->if_mtu = ifr->ifr_mtu;
+		}
+		break;
+	case SIOCSIFFLAGS:
+		/*
+		 * If interface is marked up and not running, then start it.
+		 * If it is marked down and running, stop it.
+		 * XXX If it's up then re-initialize it. This is so flags
+		 * such as IFF_PROMISC are handled.
+		 */
+		if (ifp->if_flags & IFF_UP) {
+			if (!(ifp->if_flags & IFF_RUNNING))
+				xge_init(ifp);
+                } else {
+			if (ifp->if_flags & IFF_RUNNING)
+				xge_stop(ifp, 0);
+		}
+		break;
+	case SIOCADDMULTI:
+	case SIOCDELMULTI:
+		error = (cmd == SIOCADDMULTI)
+			? ether_addmulti(ifr, &sc->sc_arpcom)
+			: ether_delmulti(ifr, &sc->sc_arpcom);
+
+                if (error == ENETRESET) {
+                        if (ifp->if_flags & IFF_RUNNING)
+				xge_mcast_filter(sc);
+			error = 0;
+		}
+		break;
+	case SIOCGIFMEDIA:
+	case SIOCSIFMEDIA:
+		error = ifmedia_ioctl(ifp, ifr, &sc->xena_media, cmd);
+		break;
+	default:
+		error = EINVAL;
+	}
+
+	splx(s);
+	return(error);
+}
+
+void
+xge_mcast_filter(struct xge_softc *sc)
+{
+	struct ifnet *ifp = &sc->sc_arpcom.ac_if;
+	struct arpcom *ac = &sc->sc_arpcom;
+	struct ether_multi *enm;
+	struct ether_multistep step;
+	int i, numaddr = 1; /* first slot used for card unicast address */
+	uint64_t val;
+
+	ETHER_FIRST_MULTI(step, ac, enm);
+	while (enm != NULL) {
+		if (memcmp(enm->enm_addrlo, enm->enm_addrhi, ETHER_ADDR_LEN)) {
+			/* Skip ranges */
+			goto allmulti;
+		}
+		if (numaddr == MAX_MCAST_ADDR)
+			goto allmulti;
+		for (val = 0, i = 0; i < ETHER_ADDR_LEN; i++) {
+			val <<= 8;
+			val |= enm->enm_addrlo[i];
+		}
+		PIF_WCSR(RMAC_ADDR_DATA0_MEM, val << 16);
+		PIF_WCSR(RMAC_ADDR_DATA1_MEM, 0xFFFFFFFFFFFFFFFFULL);
+		PIF_WCSR(RMAC_ADDR_CMD_MEM, RMAC_ADDR_CMD_MEM_WE|
+		    RMAC_ADDR_CMD_MEM_STR|RMAC_ADDR_CMD_MEM_OFF(numaddr));
+		while (PIF_RCSR(RMAC_ADDR_CMD_MEM) & RMAC_ADDR_CMD_MEM_STR)
+			;
+		numaddr++;
+		ETHER_NEXT_MULTI(step, enm);
+	}
+	/* set the remaining entries to the broadcast address */
+	for (i = numaddr; i < MAX_MCAST_ADDR; i++) {
+		PIF_WCSR(RMAC_ADDR_DATA0_MEM, 0xffffffffffff0000ULL);
+		PIF_WCSR(RMAC_ADDR_DATA1_MEM, 0xFFFFFFFFFFFFFFFFULL);
+		PIF_WCSR(RMAC_ADDR_CMD_MEM, RMAC_ADDR_CMD_MEM_WE|
+		    RMAC_ADDR_CMD_MEM_STR|RMAC_ADDR_CMD_MEM_OFF(i));
+		while (PIF_RCSR(RMAC_ADDR_CMD_MEM) & RMAC_ADDR_CMD_MEM_STR)
+			;
+	}
+	ifp->if_flags &= ~IFF_ALLMULTI;
+	return;
+
+allmulti:
+	/* Just receive everything with the multicast bit set */
+	ifp->if_flags |= IFF_ALLMULTI;
+	PIF_WCSR(RMAC_ADDR_DATA0_MEM, 0x8000000000000000ULL);
+	PIF_WCSR(RMAC_ADDR_DATA1_MEM, 0xF000000000000000ULL);
+	PIF_WCSR(RMAC_ADDR_CMD_MEM, RMAC_ADDR_CMD_MEM_WE|
+	    RMAC_ADDR_CMD_MEM_STR|RMAC_ADDR_CMD_MEM_OFF(1));
+	while (PIF_RCSR(RMAC_ADDR_CMD_MEM) & RMAC_ADDR_CMD_MEM_STR)
+		;
+}
+
+void 
+xge_start(struct ifnet *ifp)
+{
+	struct xge_softc *sc = ifp->if_softc;
+	struct txd *txd = NULL; /* XXX - gcc */
+	bus_dmamap_t dmp;
+	struct	mbuf *m;
+	uint64_t par, lcr;
+	int nexttx = 0, ntxd, error, i;
+
+	if ((ifp->if_flags & (IFF_RUNNING|IFF_OACTIVE)) != IFF_RUNNING)
+		return;
+
+	par = lcr = 0;
+	for (;;) {
+		IFQ_POLL(&ifp->if_snd, m);
+		if (m == NULL)
+			break;	/* out of packets */
+
+		if (sc->sc_nexttx == sc->sc_lasttx)
+			break;	/* No more space */
+
+		nexttx = sc->sc_nexttx;
+		dmp = sc->sc_txm[nexttx];
+
+		if ((error = bus_dmamap_load_mbuf(sc->sc_dmat, dmp, m,
+		    BUS_DMA_WRITE|BUS_DMA_NOWAIT)) != 0) {
+			printf("%s: bus_dmamap_load_mbuf error %d\n",
+			    XNAME, error);
+			break;
+		}
+		IFQ_DEQUEUE(&ifp->if_snd, m);
+
+		bus_dmamap_sync(sc->sc_dmat, dmp, 0, dmp->dm_mapsize,
+		    BUS_DMASYNC_PREWRITE);
+
+		txd = sc->sc_txd[nexttx];
+		sc->sc_txb[nexttx] = m;
+		for (i = 0; i < dmp->dm_nsegs; i++) {
+			if (dmp->dm_segs[i].ds_len == 0)
+				continue;
+			txd->txd_control1 = dmp->dm_segs[i].ds_len;
+			txd->txd_control2 = 0;
+			txd->txd_bufaddr = dmp->dm_segs[i].ds_addr;
+			txd++;
+		}
+		ntxd = txd - sc->sc_txd[nexttx] - 1;
+		txd = sc->sc_txd[nexttx];
+		txd->txd_control1 |= TXD_CTL1_OWN|TXD_CTL1_GCF;
+		txd->txd_control2 = TXD_CTL2_UTIL;
+
+#ifdef XGE_CKSUM
+		if (m->m_pkthdr.csum_flags & M_CSUM_IPv4)
+			txd->txd_control2 |= TXD_CTL2_CIPv4;
+		if (m->m_pkthdr.csum_flags & M_CSUM_TCPv4)
+			txd->txd_control2 |= TXD_CTL2_CTCP;
+		if (m->m_pkthdr.csum_flags & M_CSUM_UDPv4)
+			txd->txd_control2 |= TXD_CTL2_CUDP;
+#endif
+		txd[ntxd].txd_control1 |= TXD_CTL1_GCL;
+
+		bus_dmamap_sync(sc->sc_dmat, dmp, 0, dmp->dm_mapsize,
+		    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
+
+		par = sc->sc_txdp[nexttx];
+		lcr = TXDL_NUMTXD(ntxd) | TXDL_LGC_FIRST | TXDL_LGC_LAST;
+		TXP_WCSR(TXDL_PAR, par);
+		TXP_WCSR(TXDL_LCR, lcr);
+
+#if NBPFILTER > 0
+		if (ifp->if_bpf)
+			bpf_mtap(ifp->if_bpf, m, BPF_DIRECTION_OUT);
+#endif /* NBPFILTER > 0 */
+
+		sc->sc_nexttx = NEXTTX(nexttx);
+	}
+}
+
+/*
+ * Allocate DMA memory for transmit descriptor fragments.
+ * Only one map is used for all descriptors.
+ */
+int
+xge_alloc_txmem(struct xge_softc *sc)
+{
+	struct txd *txp;
+	bus_dma_segment_t seg;
+	bus_addr_t txdp;
+	caddr_t kva;
+	int i, rseg, state;
+
+#define TXMAPSZ (NTXDESCS*NTXFRAGS*sizeof(struct txd))
+	state = 0;
+	if (bus_dmamem_alloc(sc->sc_dmat, TXMAPSZ, PAGE_SIZE, 0,
+	    &seg, 1, &rseg, BUS_DMA_NOWAIT))
+		goto err;
+	state++;
+	if (bus_dmamem_map(sc->sc_dmat, &seg, rseg, TXMAPSZ, &kva,
+	    BUS_DMA_NOWAIT))
+		goto err;
+
+	state++;
+	if (bus_dmamap_create(sc->sc_dmat, TXMAPSZ, 1, TXMAPSZ, 0,
+	    BUS_DMA_NOWAIT, &sc->sc_txmap))
+		goto err;
+	state++;
+	if (bus_dmamap_load(sc->sc_dmat, sc->sc_txmap,
+	    kva, TXMAPSZ, NULL, BUS_DMA_NOWAIT))
+		goto err;
+
+	/* setup transmit array pointers */
+	txp = (struct txd *)kva;
+	txdp = seg.ds_addr;
+	for (txp = (struct txd *)kva, i = 0; i < NTXDESCS; i++) {
+		sc->sc_txd[i] = txp;
+		sc->sc_txdp[i] = txdp;
+		txp += NTXFRAGS;
+		txdp += (NTXFRAGS * sizeof(struct txd));
+	}
+
+	return 0;
+
+err:
+	if (state > 2)
+		bus_dmamap_destroy(sc->sc_dmat, sc->sc_txmap);
+	if (state > 1)
+		bus_dmamem_unmap(sc->sc_dmat, kva, TXMAPSZ);
+	if (state > 0)
+		bus_dmamem_free(sc->sc_dmat, &seg, rseg);
+	return ENOBUFS;
+}
+
+/*
+ * Allocate DMA memory for receive descriptor,
+ * only one map is used for all descriptors.
+ * link receive descriptor pages together.
+ */
+int
+xge_alloc_rxmem(struct xge_softc *sc)
+{
+	struct rxd_4k *rxpp;
+	bus_dma_segment_t seg;
+	caddr_t kva;
+	int i, rseg, state;
+
+	/* sanity check */
+	if (sizeof(struct rxd_4k) != XGE_PAGE) {
+		printf("bad compiler struct alignment, %d != %d\n",
+		    (int)sizeof(struct rxd_4k), XGE_PAGE);
+		return EINVAL;
+	}
+
+	state = 0;
+	if (bus_dmamem_alloc(sc->sc_dmat, RXMAPSZ, PAGE_SIZE, 0,
+	    &seg, 1, &rseg, BUS_DMA_NOWAIT))
+		goto err;
+	state++;
+	if (bus_dmamem_map(sc->sc_dmat, &seg, rseg, RXMAPSZ, &kva,
+	    BUS_DMA_NOWAIT))
+		goto err;
+
+	state++;
+	if (bus_dmamap_create(sc->sc_dmat, RXMAPSZ, 1, RXMAPSZ, 0,
+	    BUS_DMA_NOWAIT, &sc->sc_rxmap))
+		goto err;
+	state++;
+	if (bus_dmamap_load(sc->sc_dmat, sc->sc_rxmap,
+	    kva, RXMAPSZ, NULL, BUS_DMA_NOWAIT))
+		goto err;
+
+	/* setup receive page link pointers */
+	for (rxpp = (struct rxd_4k *)kva, i = 0; i < NRXPAGES; i++, rxpp++) {
+		sc->sc_rxd_4k[i] = rxpp;
+		rxpp->r4_next = (uint64_t)sc->sc_rxmap->dm_segs[0].ds_addr +
+		    (i*sizeof(struct rxd_4k)) + sizeof(struct rxd_4k);
+	}
+	sc->sc_rxd_4k[NRXPAGES-1]->r4_next = 
+	    (uint64_t)sc->sc_rxmap->dm_segs[0].ds_addr;
+
+	return 0;
+
+err:
+	if (state > 2)
+		bus_dmamap_destroy(sc->sc_dmat, sc->sc_txmap);
+	if (state > 1)
+		bus_dmamem_unmap(sc->sc_dmat, kva, TXMAPSZ);
+	if (state > 0)
+		bus_dmamem_free(sc->sc_dmat, &seg, rseg);
+	return ENOBUFS;
+}
+
+
+/*
+ * Add a new mbuf chain to descriptor id.
+ */
+int
+xge_add_rxbuf(struct xge_softc *sc, int id)
+{
+	struct rxdesc *rxd;
+	struct mbuf *m[5];
+	int page, desc, error;
+#if RX_MODE == RX_MODE_5
+	int i;
+#endif
+
+	page = id/NDESC_BUFMODE;
+	desc = id%NDESC_BUFMODE;
+
+	rxd = &sc->sc_rxd_4k[page]->r4_rxd[desc];
+
+	/*
+	 * Allocate mbufs.
+	 * Currently five mbufs and two clusters are used,
+	 * the hardware will put (ethernet, ip, tcp/udp) headers in
+	 * their own buffer and the clusters are only used for data.
+	 */
+#if RX_MODE == RX_MODE_1
+	MGETHDR(m[0], M_DONTWAIT, MT_DATA);
+	if (m[0] == NULL)
+		return ENOBUFS;
+	MCLGET(m[0], M_DONTWAIT);
+	if ((m[0]->m_flags & M_EXT) == 0) {
+		m_freem(m[0]);
+		return ENOBUFS;
+	}
+	m[0]->m_len = m[0]->m_pkthdr.len = m[0]->m_ext.ext_size;
+#elif RX_MODE == RX_MODE_3
+#error missing rxmode 3.
+#elif RX_MODE == RX_MODE_5
+	MGETHDR(m[0], M_DONTWAIT, MT_DATA);
+	for (i = 1; i < 5; i++) {
+		MGET(m[i], M_DONTWAIT, MT_DATA);
+	}
+	if (m[3])
+		MCLGET(m[3], M_DONTWAIT);
+	if (m[4])
+		MCLGET(m[4], M_DONTWAIT);
+	if (!m[0] || !m[1] || !m[2] || !m[3] || !m[4] || 
+	    ((m[3]->m_flags & M_EXT) == 0) || ((m[4]->m_flags & M_EXT) == 0)) {
+		/* Out of something */
+		for (i = 0; i < 5; i++)
+			if (m[i] != NULL)
+				m_free(m[i]);
+		return ENOBUFS;
+	}
+	/* Link'em together */
+	m[0]->m_next = m[1];
+	m[1]->m_next = m[2];
+	m[2]->m_next = m[3];
+	m[3]->m_next = m[4];
+#else
+#error bad mode RX_MODE
+#endif
+
+	if (sc->sc_rxb[id])
+		bus_dmamap_unload(sc->sc_dmat, sc->sc_rxm[id]);
+	sc->sc_rxb[id] = m[0];
+
+	error = bus_dmamap_load_mbuf(sc->sc_dmat, sc->sc_rxm[id], m[0],
+	    BUS_DMA_READ|BUS_DMA_NOWAIT);
+	if (error)
+		return error;
+	bus_dmamap_sync(sc->sc_dmat, sc->sc_rxm[id], 0,
+	    sc->sc_rxm[id]->dm_mapsize, BUS_DMASYNC_PREREAD);
+
+#if RX_MODE == RX_MODE_1
+	rxd->rxd_control2 = RXD_MKCTL2(m[0]->m_len, 0, 0);
+	rxd->rxd_buf0 = (uint64_t)sc->sc_rxm[id]->dm_segs[0].ds_addr;
+	rxd->rxd_control1 = RXD_CTL1_OWN;
+#elif RX_MODE == RX_MODE_3
+#elif RX_MODE == RX_MODE_5
+	rxd->rxd_control3 = RXD_MKCTL3(0, m[3]->m_len, m[4]->m_len);
+	rxd->rxd_control2 = RXD_MKCTL2(m[0]->m_len, m[1]->m_len, m[2]->m_len);
+	rxd->rxd_buf0 = (uint64_t)sc->sc_rxm[id]->dm_segs[0].ds_addr;
+	rxd->rxd_buf1 = (uint64_t)sc->sc_rxm[id]->dm_segs[1].ds_addr;
+	rxd->rxd_buf2 = (uint64_t)sc->sc_rxm[id]->dm_segs[2].ds_addr;
+	rxd->rxd_buf3 = (uint64_t)sc->sc_rxm[id]->dm_segs[3].ds_addr;
+	rxd->rxd_buf4 = (uint64_t)sc->sc_rxm[id]->dm_segs[4].ds_addr;
+	rxd->rxd_control1 = RXD_CTL1_OWN;
+#endif
+
+	XGE_RXSYNC(id, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
+	return 0;
+}
+
+/*
+ * These magics comes from the FreeBSD driver.
+ */
+int
+xge_setup_xgxs(struct xge_softc *sc)
+{
+	/* The magic numbers are described in the users guide */
+
+	/* Writing to MDIO 0x8000 (Global Config 0) */
+	PIF_WCSR(DTX_CONTROL, 0x8000051500000000ULL); DELAY(50);
+	PIF_WCSR(DTX_CONTROL, 0x80000515000000E0ULL); DELAY(50);
+	PIF_WCSR(DTX_CONTROL, 0x80000515D93500E4ULL); DELAY(50);
+
+	/* Writing to MDIO 0x8000 (Global Config 1) */
+	PIF_WCSR(DTX_CONTROL, 0x8001051500000000ULL); DELAY(50);
+	PIF_WCSR(DTX_CONTROL, 0x80010515000000e0ULL); DELAY(50);
+	PIF_WCSR(DTX_CONTROL, 0x80010515001e00e4ULL); DELAY(50);
+
+	/* Reset the Gigablaze */
+	PIF_WCSR(DTX_CONTROL, 0x8002051500000000ULL); DELAY(50);
+	PIF_WCSR(DTX_CONTROL, 0x80020515000000E0ULL); DELAY(50);
+	PIF_WCSR(DTX_CONTROL, 0x80020515F21000E4ULL); DELAY(50);
+
+	/* read the pole settings */
+	PIF_WCSR(DTX_CONTROL, 0x8000051500000000ULL); DELAY(50);
+	PIF_WCSR(DTX_CONTROL, 0x80000515000000e0ULL); DELAY(50);
+	PIF_WCSR(DTX_CONTROL, 0x80000515000000ecULL); DELAY(50);
+
+	PIF_WCSR(DTX_CONTROL, 0x8001051500000000ULL); DELAY(50);
+	PIF_WCSR(DTX_CONTROL, 0x80010515000000e0ULL); DELAY(50);
+	PIF_WCSR(DTX_CONTROL, 0x80010515000000ecULL); DELAY(50);
+
+	PIF_WCSR(DTX_CONTROL, 0x8002051500000000ULL); DELAY(50);
+	PIF_WCSR(DTX_CONTROL, 0x80020515000000e0ULL); DELAY(50);
+	PIF_WCSR(DTX_CONTROL, 0x80020515000000ecULL); DELAY(50);
+
+	/* Workaround for TX Lane XAUI initialization error.
+	   Read Xpak PHY register 24 for XAUI lane status */
+	PIF_WCSR(DTX_CONTROL, 0x0018040000000000ULL); DELAY(50);
+	PIF_WCSR(DTX_CONTROL, 0x00180400000000e0ULL); DELAY(50);
+	PIF_WCSR(DTX_CONTROL, 0x00180400000000ecULL); DELAY(50);
+
+	/* 
+	 * Reading the MDIO control with value 0x1804001c0F001c
+	 * means the TxLanes were already in sync
+	 * Reading the MDIO control with value 0x1804000c0x001c
+	 * means some TxLanes are not in sync where x is a 4-bit
+	 * value representing each lanes
+	 */
+#if 0
+	val = PIF_RCSR(MDIO_CONTROL);
+	if (val != 0x1804001c0F001cULL) {
+		printf("%s: MDIO_CONTROL: %llx != %llx\n", 
+		    XNAME, val, 0x1804001c0F001cULL);
+		return 1;
+	}
+#endif
+
+	/* Set and remove the DTE XS INTLoopBackN */
+	PIF_WCSR(DTX_CONTROL, 0x0000051500000000ULL); DELAY(50);
+	PIF_WCSR(DTX_CONTROL, 0x00000515604000e0ULL); DELAY(50);
+	PIF_WCSR(DTX_CONTROL, 0x00000515604000e4ULL); DELAY(50);
+	PIF_WCSR(DTX_CONTROL, 0x00000515204000e4ULL); DELAY(50);
+	PIF_WCSR(DTX_CONTROL, 0x00000515204000ecULL); DELAY(50);
+
+#if 0
+	/* Reading the DTX control register Should be 0x5152040001c */
+	val = PIF_RCSR(DTX_CONTROL);
+	if (val != 0x5152040001cULL) {
+		printf("%s: DTX_CONTROL: %llx != %llx\n", 
+		    XNAME, val, 0x5152040001cULL);
+		return 1;
+	}
+#endif
+
+	PIF_WCSR(MDIO_CONTROL, 0x0018040000000000ULL); DELAY(50);
+	PIF_WCSR(MDIO_CONTROL, 0x00180400000000e0ULL); DELAY(50);
+	PIF_WCSR(MDIO_CONTROL, 0x00180400000000ecULL); DELAY(50);
+
+#if 0
+	/* Reading the MIOD control should be 0x1804001c0f001c */
+	val = PIF_RCSR(MDIO_CONTROL);
+	if (val != 0x1804001c0f001cULL) {
+		printf("%s: MDIO_CONTROL2: %llx != %llx\n",
+		    XNAME, val, 0x1804001c0f001cULL);
+		return 1;
+	}
+#endif
+	return 0;
+}
diff --git a/sys/dev/pci/if_xgereg.h b/sys/dev/pci/if_xgereg.h
new file mode 100644
index 00000000000..fac200e6d1c
--- /dev/null
+++ b/sys/dev/pci/if_xgereg.h
@@ -0,0 +1,427 @@
+/*	$OpenBSD: if_xgereg.h,v 1.1 2006/05/01 00:34:12 brad Exp $	*/
+/*	$NetBSD: if_xgereg.h,v 1.1 2005/09/09 10:30:27 ragge Exp $	*/
+
+/*
+ * Copyright (c) 2004, SUNET, Swedish University Computer Network.
+ * All rights reserved.
+ *
+ * Written by Anders Magnusson for SUNET, Swedish University Computer Network.
+ *
+ * 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.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *      This product includes software developed for the NetBSD Project by
+ *      SUNET, Swedish University Computer Network.
+ * 4. The name of SUNET may not be used to endorse or promote products
+ *    derived from this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY SUNET ``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 SUNET
+ * 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.
+ */
+
+/*
+ * Defines for the Neterion Xframe adapter.
+ */
+
+/* PCI address space */
+#define	XGE_PIF_BAR	0x10
+#define	XGE_TXP_BAR	0x18
+
+/* PIF register address calculation */
+#define	DCSRB(x) (0x0000+(x))	/* 10GbE Device Control and Status Registers */
+#define	PCIXB(x) (0x0800+(x))	/* PCI-X Interface Functional Registers */
+#define	TDMAB(x) (0x1000+(x))	/* Transmit DMA Functional Registers */
+#define	RDMAB(x) (0x1800+(x))	/* Receive DMA Functional Registers */
+#define	MACRB(x) (0x2000+(x))	/* MAC functional registers */
+#define	RLDRB(x) (0x2800+(x))	/* RLDRAM memory controller */
+#define	XGXSB(x) (0x3000+(x))	/* XGXS functional Registers */
+
+/*
+ * Control and Status Registers
+ */
+#define	GENERAL_INT_STATUS	DCSRB(0x0000)
+#define	GENERAL_INT_MASK	DCSRB(0x0008)
+#define	SW_RESET		DCSRB(0x0100)
+#define	 XGXS_RESET(x)		((uint64_t)(x) << 32)
+#define	ADAPTER_STATUS		DCSRB(0x0108)
+#define	 TDMA_READY		(1ULL<<63)
+#define	 RDMA_READY		(1ULL<<62)
+#define	 PFC_READY		(1ULL<<61)
+#define	 TMAC_BUF_EMPTY		(1ULL<<60)
+#define	 PIC_QUIESCENT		(1ULL<<58)
+#define	 RMAC_REMOTE_FAULT	(1ULL<<57)
+#define	 RMAC_LOCAL_FAULT	(1ULL<<56)
+#define	 MC_DRAM_READY		(1ULL<<39)
+#define	 MC_QUEUES_READY	(1ULL<<38)
+#define	 M_PLL_LOCK		(1ULL<<33)
+#define	 P_PLL_LOCK		(1ULL<<32)
+#define	ADAPTER_CONTROL		DCSRB(0x0110)
+#define	 ADAPTER_EN		(1ULL<<56)
+#define	 EOI_TX_ON		(1ULL<<48)
+#define	 LED_ON			(1ULL<<40)
+#define	 WAIT_INT_EN		(1ULL<<15)
+#define	 ECC_ENABLE_N		(1ULL<<8)
+
+/* for debug of ADAPTER_STATUS */
+#define	QUIESCENT (TDMA_READY|RDMA_READY|PFC_READY|TMAC_BUF_EMPTY|\
+	PIC_QUIESCENT|MC_DRAM_READY|MC_QUEUES_READY|M_PLL_LOCK|P_PLL_LOCK)
+#define	QUIESCENT_BMSK	\
+	"\177\20b\x3fTDMA_READY\0b\x3eRDMA_READY\0b\x3dPFC_READY\0" \
+	"b\x3cTMAC_BUF_EMPTY\0b\x3aPIC_QUIESCENT\0\x39RMAC_REMOTE_FAULT\0" \
+	"b\x38RMAC_LOCAL_FAULT\0b\x27MC_DRAM_READY\0b\x26MC_QUEUES_READY\0" \
+	"b\x21M_PLL_LOCK\0b\x20P_PLL_LOCK"
+
+/*
+ * PCI-X registers
+ */
+/* Interrupt control registers */
+#define	PIC_INT_STATUS		PCIXB(0)
+#define	PIC_INT_MASK		PCIXB(0x008)
+#define	TXPIC_INT_MASK		PCIXB(0x018)
+#define	RXPIC_INT_MASK		PCIXB(0x030)
+#define	FLASH_INT_MASK		PCIXB(0x048)
+#define	MDIO_INT_MASK		PCIXB(0x060)
+#define	IIC_INT_MASK		PCIXB(0x078)
+#define	GPIO_INT_MASK		PCIXB(0x098)
+#define	TX_TRAFFIC_INT		PCIXB(0x0e0)
+#define	TX_TRAFFIC_MASK		PCIXB(0x0e8)
+#define	RX_TRAFFIC_INT		PCIXB(0x0f0)
+#define	RX_TRAFFIC_MASK		PCIXB(0x0f8)
+#define	PIC_CONTROL		PCIXB(0x100)
+
+/* Byte swapping for little-endian */
+#define	SWAPPER_CTRL		PCIXB(0x108)
+#define	 PIF_R_FE		(1ULL<<63)
+#define	 PIF_R_SE		(1ULL<<62)
+#define	 PIF_W_FE		(1ULL<<55)
+#define	 PIF_W_SE		(1ULL<<54)
+#define	 TxP_FE			(1ULL<<47)
+#define	 TxP_SE			(1ULL<<46)
+#define	 TxD_R_FE		(1ULL<<45)
+#define	 TxD_R_SE		(1ULL<<44)
+#define	 TxD_W_FE		(1ULL<<43)
+#define	 TxD_W_SE		(1ULL<<42)
+#define	 TxF_R_FE		(1ULL<<41)
+#define	 TxF_R_SE		(1ULL<<40)
+#define	 RxD_R_FE		(1ULL<<31)
+#define	 RxD_R_SE		(1ULL<<30)
+#define	 RxD_W_FE		(1ULL<<29)
+#define	 RxD_W_SE		(1ULL<<28)
+#define	 RxF_W_FE		(1ULL<<27)
+#define	 RxF_W_SE		(1ULL<<26)
+#define	 XMSI_FE		(1ULL<<23)
+#define	 XMSI_SE		(1ULL<<22)
+#define	 STATS_FE		(1ULL<<15)
+#define	 STATS_SE		(1ULL<<14)
+
+/* Diagnostic register to check byte-swapping conf */
+#define	PIF_RD_SWAPPER_Fb	PCIXB(0x110)
+#define	 SWAPPER_MAGIC		0x0123456789abcdefULL
+
+/* Stats registers */
+#define	STAT_CFG		PCIXB(0x1d0)
+#define	STAT_ADDR		PCIXB(0x1d8)
+
+/* DTE-XGXS Interface */
+#define	MDIO_CONTROL		PCIXB(0x1e0)
+#define	DTX_CONTROL		PCIXB(0x1e8)
+#define	I2C_CONTROL		PCIXB(0x1f0)
+#define	GPIO_CONTROL		PCIXB(0x1f8)
+
+/*
+ * Transmit DMA registers.
+ */
+#define	TXDMA_INT_MASK		TDMAB(0x008)
+#define	PFC_ERR_MASK		TDMAB(0x018)
+#define	TDA_ERR_MASK		TDMAB(0x030)
+#define	PCC_ERR_MASK		TDMAB(0x048)
+#define	TTI_ERR_MASK		TDMAB(0x060)
+#define	LSO_ERR_MASK		TDMAB(0x078)
+#define	TPA_ERR_MASK		TDMAB(0x090)
+#define	SM_ERR_MASK		TDMAB(0x0a8)
+
+/* Transmit FIFO config */
+#define	TX_FIFO_P0		TDMAB(0x0108)
+#define	TX_FIFO_P1		TDMAB(0x0110)
+#define	TX_FIFO_P2		TDMAB(0x0118)
+#define	TX_FIFO_P3		TDMAB(0x0120)
+#define	 TX_FIFO_ENABLE		(1ULL<<63)
+#define	 TX_FIFO_NUM0(x)	((uint64_t)(x) << 56)
+#define	 TX_FIFO_LEN0(x)	((uint64_t)((x)-1) << 32)	
+#define	 TX_FIFO_NUM1(x)	((uint64_t)(x) << 24)
+#define	 TX_FIFO_LEN1(x)	((uint64_t)((x)-1) << 0)	
+
+/* Transmit interrupts */
+#define	TTI_COMMAND_MEM		TDMAB(0x150)
+#define	 TTI_CMD_MEM_WE		(1ULL<<56)
+#define	 TTI_CMD_MEM_STROBE	(1ULL<<48)
+#define	TTI_DATA1_MEM		TDMAB(0x158)
+#define	 TX_TIMER_VAL(x)	((uint64_t)(x) << 32)
+#define	 TX_TIMER_AC		(1ULL<<25)
+#define	 TX_TIMER_CI		(1ULL<<24)
+#define	 TX_URNG_A(x)		((uint64_t)(x) << 16)
+#define	 TX_URNG_B(x)		((uint64_t)(x) << 8)
+#define	 TX_URNG_C(x)		((uint64_t)(x) << 0)
+#define	TTI_DATA2_MEM		TDMAB(0x160)
+#define	 TX_UFC_A(x)		((uint64_t)(x) << 48)
+#define	 TX_UFC_B(x)		((uint64_t)(x) << 32)
+#define	 TX_UFC_C(x)		((uint64_t)(x) << 16)
+#define	 TX_UFC_D(x)		((uint64_t)(x) << 0)
+
+
+/* Transmit protocol assist */
+#define	TX_PA_CFG		TDMAB(0x0168)
+#define	 TX_PA_CFG_IFR		(1ULL<<62)	/* Ignore frame error */
+#define	 TX_PA_CFG_ISO		(1ULL<<61)	/* Ignore snap OUI */
+#define	 TX_PA_CFG_ILC		(1ULL<<60)	/* Ignore LLC ctrl */
+#define	 TX_PA_CFG_ILE		(1ULL<<57)	/* Ignore L2 error */
+
+/*
+ * Transmit descriptor list (TxDL) pointer and control.
+ * There may be up to 8192 TxDL's per FIFO, but with a NIC total
+ * of 8192. The TxDL's are located in the NIC memory.
+ * Each TxDL can have up to 256 Transmit descriptors (TxD)
+ * that are located in host memory.
+ *
+ * The txdl struct fields must be written in order.
+ */
+#ifdef notdef  /* Use bus_space stuff instead */
+struct txdl {
+	uint64_t txdl_pointer;	/* address of TxD's */
+	uint64_t txdl_control;
+};
+#endif
+#define	TXDLOFF1(x)	(16*(x))	/* byte offset in txdl for list */
+#define	TXDLOFF2(x)	(16*(x)+8)	/* byte offset in txdl for list */
+#define	TXDL_NUMTXD(x)	((uint64_t)(x) << 56)	/* # of TxD's in the list */
+#define	TXDL_LGC_FIRST	(1ULL << 49)	/* First special list */
+#define	TXDL_LGC_LAST	(1ULL << 48)	/* Last special list */
+#define	TXDL_SFF	(1ULL << 40)	/* List is a special function list */
+#define	TXDL_PAR	0		/* Pointer address register */
+#define	TXDL_LCR	8		/* List control register */
+
+struct txd {
+	uint64_t txd_control1;
+	uint64_t txd_control2;
+	uint64_t txd_bufaddr;
+	uint64_t txd_hostctrl;
+};
+#define	TXD_CTL1_OWN	(1ULL << 56)	/* Owner, 0 == host, 1 == NIC */
+#define	TXD_CTL1_GCF	(1ULL << 41)	/* First frame or LSO */
+#define	TXD_CTL1_GCL	(1ULL << 40)	/* Last frame or LSO */
+#define	TXD_CTL1_LSO	(1ULL << 33)	/* LSO should be performed */
+#define	TXD_CTL1_COF	(1ULL << 32)	/* UDP Checksum over fragments */
+#define	TXD_CTL1_MSS(x)	((uint64_t)(x) << 16)
+
+#define	TXD_CTL2_INTLST	(1ULL << 16)	/* Per-list interrupt */
+#define	TXD_CTL2_UTIL	(1ULL << 17)	/* Utilization interrupt */
+#define	TXD_CTL2_CIPv4	(1ULL << 58)	/* Calculate IPv4 header checksum */
+#define	TXD_CTL2_CTCP	(1ULL << 57)	/* Calculate TCP checksum */
+#define	TXD_CTL2_CUDP	(1ULL << 56)	/* Calculate UDP checksum */
+/*
+ * Receive DMA registers
+ */
+/* Receive interrupt registers */
+#define	RXDMA_INT_MASK		RDMAB(0x008)
+#define	RDA_ERR_MASK		RDMAB(0x018)
+#define	RC_ERR_MASK		RDMAB(0x030)
+#define	PRC_PCIX_ERR_MASK	RDMAB(0x048)
+#define	RPA_ERR_MASK		RDMAB(0x060)
+#define	RTI_ERR_MASK		RDMAB(0x078)
+
+#define	RX_QUEUE_PRIORITY	RDMAB(0x100)
+#define	RX_W_ROUND_ROBIN_0	RDMAB(0x108)
+#define	RX_W_ROUND_ROBIN_1	RDMAB(0x110)
+#define	RX_W_ROUND_ROBIN_2	RDMAB(0x118)
+#define	RX_W_ROUND_ROBIN_3	RDMAB(0x120)
+#define	RX_W_ROUND_ROBIN_4	RDMAB(0x128)
+#define	PRC_RXD0_0		RDMAB(0x130)
+#define	PRC_CTRL_0		RDMAB(0x170)
+#define	 RC_IN_SVC		(1ULL << 56)
+#define	 RING_MODE_1		(0ULL << 48)
+#define	 RING_MODE_3		(1ULL << 48)
+#define	 RING_MODE_5		(2ULL << 48)
+#define	 RC_NO_SNOOP_D		(1ULL << 41)
+#define	 RC_NO_SNOOP_B		(1ULL << 40)
+#define	PRC_ALARM_ACTION	RDMAB(0x1b0)
+#define	RTI_COMMAND_MEM		RDMAB(0x1b8)
+#define	 RTI_CMD_MEM_WE		(1ULL << 56)
+#define	 RTI_CMD_MEM_STROBE	(1ULL << 48)
+#define	RTI_DATA1_MEM		RDMAB(0x1c0)
+#define	 RX_TIMER_VAL(x)	((uint64_t)(x) << 32)
+#define	 RX_TIMER_AC		(1ULL << 25)
+#define	 RX_URNG_A(x)		((uint64_t)(x) << 16)
+#define	 RX_URNG_B(x)		((uint64_t)(x) << 8)
+#define	 RX_URNG_C(x)		((uint64_t)(x) << 0)
+#define	RTI_DATA2_MEM		RDMAB(0x1c8)
+#define	 RX_UFC_A(x)		((uint64_t)(x) << 48)
+#define	 RX_UFC_B(x)		((uint64_t)(x) << 32)
+#define	 RX_UFC_C(x)		((uint64_t)(x) << 16)
+#define	 RX_UFC_D(x)		((uint64_t)(x) << 0)
+#define	RX_PA_CFG		RDMAB(0x1d0)
+/*
+ * Receive descriptor (RxD) format.
+ * There are three formats of receive descriptors, 1, 3 and 5 buffer format.
+ */
+#define	RX_MODE_1 1
+#define	RX_MODE_3 3
+#define	RX_MODE_5 5
+
+struct rxd1 {
+	uint64_t rxd_hcontrol;
+	uint64_t rxd_control1;
+	uint64_t rxd_control2;  
+	uint64_t rxd_buf0;
+};
+
+/* 4k struct for 5 buffer mode */
+#define	NDESC_1BUFMODE		127	/* # desc/page for 5-buffer mode */
+struct rxd1_4k {
+	struct rxd1 r4_rxd[NDESC_1BUFMODE];
+	uint64_t pad[3];
+	uint64_t r4_next; /* phys address of next 4k buffer */
+};
+
+struct rxd3 {
+	uint64_t rxd_hcontrol;
+	uint64_t rxd_control1;
+	uint64_t rxd_control2;
+	uint64_t rxd_buf0;
+	uint64_t rxd_buf1;      
+	uint64_t rxd_buf2;      
+};
+
+struct rxd5 {
+	uint64_t rxd_control3;
+	uint64_t rxd_control1;
+	uint64_t rxd_control2;
+	uint64_t rxd_buf0;
+	uint64_t rxd_buf1;
+	uint64_t rxd_buf2;
+	uint64_t rxd_buf3;
+	uint64_t rxd_buf4;
+};
+
+/* 4k struct for 5 buffer mode */
+#define	NDESC_5BUFMODE		63	/* # desc/page for 5-buffer mode */
+#define	XGE_PAGE		4096	/* page size used for receive */
+struct rxd5_4k {
+	struct rxd5 r4_rxd[NDESC_5BUFMODE];
+	uint64_t pad[7];
+	uint64_t r4_next; /* phys address of next 4k buffer */
+};
+
+#define	RXD_MKCTL3(h,bs3,bs4)	\
+	(((uint64_t)(h) << 32) | ((uint64_t)(bs3) << 16) | (uint64_t)(bs4))
+#define	RXD_MKCTL2(bs0,bs1,bs2)	\
+	(((uint64_t)(bs0) << 48) | ((uint64_t)(bs1) << 32) | \
+	((uint64_t)(bs2) << 16))
+
+#define	RXD_CTL2_BUF0SIZ(x)	(((x) >> 48) & 0xffff)
+#define	RXD_CTL2_BUF1SIZ(x)	(((x) >> 32) & 0xffff)
+#define	RXD_CTL2_BUF2SIZ(x)	(((x) >> 16) & 0xffff)
+#define	RXD_CTL3_BUF3SIZ(x)	(((x) >> 16) & 0xffff)
+#define	RXD_CTL3_BUF4SIZ(x)	((x) & 0xffff)
+#define	RXD_CTL1_OWN		(1ULL << 56)
+#define	RXD_CTL1_XCODE(x)	(((x) >> 48) & 0xf)	/* Status bits */
+#define	 RXD_CTL1_X_OK		0
+#define	 RXD_CTL1_X_PERR	1	/* Parity error */
+#define	 RXD_CTL1_X_ABORT	2	/* Abort during xfer */
+#define	 RXD_CTL1_X_PA		3	/* Parity error and abort */
+#define	 RXD_CTL1_X_RDA		4	/* RDA failure */
+#define	 RXD_CTL1_X_UP		5	/* Unknown protocol */
+#define	 RXD_CTL1_X_FI		6	/* Frame integrity (FCS) error */
+#define	 RXD_CTL1_X_BSZ		7	/* Buffer size error */
+#define	 RXD_CTL1_X_ECC		8	/* Internal ECC */
+#define	 RXD_CTL1_X_UNK		15	/* Unknown error */
+#define	RXD_CTL1_PROTOS(x)	(((x) >> 32) & 0xff)
+#define	 RXD_CTL1_P_VLAN	0x80	/* VLAN tagged */
+#define	 RXD_CTL1_P_MSK		0x60	/* Mask for frame type */
+#define	  RXD_CTL1_P_DIX	0x00
+#define	  RXD_CTL1_P_LLC	0x20
+#define	  RXD_CTL1_P_SNAP	0x40
+#define	  RXD_CTL1_P_IPX	0x60
+#define	 RXD_CTL1_P_IPv4	0x10
+#define	 RXD_CTL1_P_IPv6	0x08
+#define	 RXD_CTL1_P_IPFRAG	0x04
+#define	 RXD_CTL1_P_TCP		0x02
+#define	 RXD_CTL1_P_UDP		0x01
+#define	RXD_CTL1_L3CSUM(x)	(((x) >> 16) & 0xffff)
+#define	RXD_CTL1_L4CSUM(x)	((x) & 0xffff)
+#define	RXD_CTL2_VLANTAG(x)	((x) & 0xffff)
+
+/*
+ * MAC Configuration/Status
+ */
+#define	MAC_INT_STATUS		MACRB(0x000)
+#define	 MAC_TMAC_INT		(1ULL<<63)
+#define	 MAC_RMAC_INT		(1ULL<<62)
+#define	MAC_INT_MASK		MACRB(0x008)
+#define	MAC_TMAC_ERR_MASK	MACRB(0x018)
+#define	MAC_RMAC_ERR_REG	MACRB(0x028)
+#define	 RMAC_LINK_STATE_CHANGE_INT (1ULL<<32)
+#define	MAC_RMAC_ERR_MASK	MACRB(0x030)
+
+#define	MAC_CFG			MACRB(0x0100)
+#define	 TMAC_EN		(1ULL<<63)
+#define	 RMAC_EN		(1ULL<<62)
+#define	 UTILZATION_CALC_SEL	(1ULL<<61)
+#define	 TMAC_LOOPBACK		(1ULL<<60)
+#define	 TMAC_APPEND_PAD	(1ULL<<59)
+#define	 RMAC_STRIP_FCS		(1ULL<<58)
+#define	 RMAC_STRIP_PAD		(1ULL<<57)
+#define	 RMAC_PROM_EN		(1ULL<<56)
+#define	 RMAC_DISCARD_PFRM	(1ULL<<55)
+#define	 RMAC_BCAST_EN		(1ULL<<54)
+#define	 RMAC_ALL_ADDR_EN	(1ULL<<53)
+#define	RMAC_MAX_PYLD_LEN	MACRB(0x0110)
+#define	 RMAC_PYLD_LEN(x)	((uint64_t)(x) << 48)
+#define	RMAC_CFG_KEY		MACRB(0x0120)
+#define	 RMAC_KEY_VALUE		(0x4c0dULL<<48)
+#define	RMAC_ADDR_CMD_MEM	MACRB(0x0128)
+#define	 RMAC_ADDR_CMD_MEM_WE	(1ULL<<56)
+#define	 RMAC_ADDR_CMD_MEM_STR	(1ULL<<48)
+#define	 RMAC_ADDR_CMD_MEM_OFF(x) ((uint64_t)(x) << 32)
+#define	MAX_MCAST_ADDR		64	/* slots in mcast table */
+#define	RMAC_ADDR_DATA0_MEM	MACRB(0x0130)
+#define	RMAC_ADDR_DATA1_MEM	MACRB(0x0138)
+#define	RMAC_PAUSE_CFG		MACRB(0x150)
+#define	 RMAC_PAUSE_GEN_EN	(1ULL<<63)
+#define	 RMAC_PAUSE_RCV_EN	(1ULL<<62)
+
+/*
+ * RLDRAM registers.
+ */
+#define	MC_INT_MASK		RLDRB(0x008)
+#define	MC_ERR_MASK		RLDRB(0x018)
+
+#define	RX_QUEUE_CFG		RLDRB(0x100)
+#define	 MC_QUEUE(q,s)		((uint64_t)(s)<<(56-(q*8)))
+#define	MC_RLDRAM_MRS		RLDRB(0x108)
+#define	 MC_QUEUE_SIZE_ENABLE	(1ULL<<24)
+#define	 MC_RLDRAM_MRS_ENABLE	(1ULL<<16)
+
+/*
+ * XGXS registers.
+ */
+/* XGXS control/statue */
+#define	XGXS_INT_MASK		XGXSB(0x008)
+#define	XGXS_TXGXS_ERR_MASK	XGXSB(0x018)
+#define	XGXS_RXGXS_ERR_MASK	XGXSB(0x030)
+#define	XGXS_CFG		XGXSB(0x0100)