xref: /src/sys/dev/ic/i82596.c

/* $NetBSD: i82596.c,v 1.9 2005/06/02 14:41:26 he Exp $ */

/*
 * Copyright (c) 2003 Jochen Kunz.
 * 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.
 * 3. The name of Jochen Kunz may not be used to endorse or promote
 *    products derived from this software without specific prior
 *    written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY JOCHEN KUNZ
 * ``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 JOCHEN KUNZ
 * 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.
 */

/*
 * Driver for the Intel i82596 10MBit/s Ethernet chip.
 * It operates the i82596 in 32-Bit Linear Mode, opposed to the old i82586
 * ie(4) driver (src/sys/dev/ic/i82586.c), that degrades the i82596 to
 * i82586 compatibility mode.
 * Documentation about this chip can be found on http://www.openpa.net/
 * file names 29021806.pdf and 29021906.pdf
 */

#include <sys/cdefs.h>
__KERNEL_RCSID(0, "$NetBSD: i82596.c,v 1.9 2005/06/02 14:41:26 he Exp $");

/* autoconfig and device stuff */
#include <sys/param.h>
#include <sys/device.h>
#include <sys/conf.h>
#include "locators.h"
#include "ioconf.h"

/* bus_space / bus_dma etc. */
#include <machine/bus.h>
#include <machine/intr.h>

/* general system data and functions */
#include <sys/systm.h>
#include <sys/ioctl.h>

/* tsleep / sleep / wakeup */
#include <sys/proc.h>
/* hz for above */
#include <sys/kernel.h>

/* network stuff */
#include <net/if.h>
#include <net/if_dl.h>
#include <net/if_media.h>
#include <net/if_ether.h>
#include <sys/socket.h>
#include <sys/mbuf.h>

#include "bpfilter.h"
#if NBPFILTER > 0
#include <net/bpf.h>
#endif

#include <dev/ic/i82596reg.h>
#include <dev/ic/i82596var.h>



/* Supported chip variants */
const char *i82596_typenames[] = { "unknowen", "DX/SX", "CA" };



/* media change and status callback */
static int iee_mediachange(struct ifnet *);
static void iee_mediastatus(struct ifnet *, struct ifmediareq *);

/* interface routines to upper protocols */
static void iee_start(struct ifnet *);			/* initiate output */
static int iee_ioctl(struct ifnet *, u_long, caddr_t);	/* ioctl routine */
static int iee_init(struct ifnet *);			/* init routine */
static void iee_stop(struct ifnet *, int);		/* stop routine */
static void iee_watchdog(struct ifnet *);		/* timer routine */
static void iee_drain(struct ifnet *);			/* release resources */

/* internal helper functions */
static void iee_cb_setup(struct iee_softc *, uint32_t);

/*
Things a MD frontend has to provide:

The functions via function pointers in the softc:
        int (*sc_iee_cmd)(struct iee_softc *sc, uint32_t cmd);
        int (*sc_iee_reset)(struct iee_softc *sc);
        void (*sc_mediastatus)(struct ifnet *, struct ifmediareq *);
        int (*sc_mediachange)(struct ifnet *);

sc_iee_cmd(): send a command to the i82596 by writing the cmd parameter
	to the SCP cmd word and issuing a Channel Attention.
sc_iee_reset(): initiate a reset, supply the address of the SCP to the
	chip, wait for the chip to initialize and ACK interrupts that
	this may have caused by caling (sc->sc_iee_cmd)(sc, IEE_SCB_ACK);
This functions must carefully bus_dmamap_sync() all data they have touched!

sc_mediastatus() and  sc_mediachange() are just MD hooks to the according
MI functions. The MD frontend may set this pointers to NULL when they
are not needed.

sc->sc_type has to be set to I82596_UNKNOWN or I82596_DX or I82596_CA.
This is for printing out the correct chip type at attach time only. The
MI backend doesn't distinguish different chip types when programming
the chip.

sc->sc_flags has to be set to 0 on litle endian hardware and to
IEE_NEED_SWAP on big endian hardware, when endianes conversion is not
done by the bus attachment. Usually you need to set IEE_NEED_SWAP
when IEE_SYSBUS_BE is set in the sysbus byte.

sc->sc_cl_align bust be set to 1 or to the cache line size. When set to
1 no special alignment of DMA descriptors is done. If sc->sc_cl_align != 1
it forces alignment of the data structres in the shared memory to a multiple
of sc->sc_cl_align. This is needed on archs like hp700 that have non DMA
I/O coherent caches and are unable to map the shared memory uncachable.
(At least pre PA7100LC CPUs are unable to map memory uncachable.)

sc->sc_cl_align MUST BE INITIALIZED BEFORE THE FOLOWING MACROS ARE USED:
SC_* IEE_*_SZ IEE_*_OFF IEE_SHMEM_MAX (shell style glob(3) pattern)

The MD frontend has to allocate a piece of DMA memory at least of
IEE_SHMEM_MAX bytes size. All communication with the chip is done via
this shared memory. If possible map this memory non-cachable on
archs with non DMA I/O coherent caches. The base of the memory needs
to be aligend to an even address if sc->sc_cl_align == 1 and aligend
to a cache line if sc->sc_cl_align != 1.

An interrupt with iee_intr() as handler must be established.

Call void iee_attach(struct iee_softc *sc, uint8_t *ether_address,
int *media, int nmedia, int defmedia); when everything is set up. First
parameter is a pointer to the MI softc, ether_address is an array that
contains the ethernet address. media is an array of the media types
provided by the hardware. The members of this array are supplied to
ifmedia_add() in sequence. nmedia is the count of elements in media.
defmedia is the default media that is set via ifmedia_set().
nmedia and defmedia are ignored when media == NULL.

The MD backend may call iee_detach() to detach the device.

See sys/arch/hp700/gsc/if_iee.c for an example.
*/


/*
How frame reception is done:
Each Recieve Frame Descriptor has one associated Recieve Buffer Descriptor.
Each RBD points to the data area of a mbuf cluster. The RFDs are linked
together in a circular list. sc->sc_rx_done is the count of RFDs in the
list already processed / the number of the RFD that has to be checked for
a new frame first at the next RX interrupt. Upon successful reception of
a frame the mbuf cluster is handled to upper protocol layers, a new mbuf
cluster is allocated and the RFD / RBD are reinitialized accordingly.

When a RFD list overrun occured the whole RFD and RBD lists are reinitialized
and frame reception is started again.
*/
int
iee_intr(void *intarg)
{
	struct iee_softc *sc = intarg;
	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
	struct iee_rfd *rfd;
	struct iee_rbd *rbd;
	bus_dmamap_t rx_map;
	struct mbuf *rx_mbuf;
	struct mbuf *new_mbuf;
	int scb_status;
	int scb_cmd;
	int n, col;

	if ((ifp->if_flags & IFF_RUNNING) == 0) {
		(sc->sc_iee_cmd)(sc, IEE_SCB_ACK);
		return(1);
	}
	bus_dmamap_sync(sc->sc_dmat, sc->sc_shmem_map, 0, IEE_SHMEM_MAX,
	    BUS_DMASYNC_POSTREAD);
	scb_status = SC_SCB->scb_status;
	scb_cmd = SC_SCB->scb_cmd;
	rfd = SC_RFD(sc->sc_rx_done);
	while ((rfd->rfd_status & IEE_RFD_C) != 0) {
		/* At least one packet was received. */
		rbd = SC_RBD(sc->sc_rx_done);
		rx_map = sc->sc_rx_map[sc->sc_rx_done];
		rx_mbuf = sc->sc_rx_mbuf[sc->sc_rx_done];
		SC_RBD((sc->sc_rx_done + IEE_NRFD - 1) % IEE_NRFD)->rbd_size
		    &= ~IEE_RBD_EL;
		if ((rfd->rfd_status & IEE_RFD_OK) == 0
		    || (rbd->rbd_count & IEE_RBD_EOF) == 0
		    || (rbd->rbd_count & IEE_RBD_F) == 0){
			/* Receive error, skip frame and reuse buffer. */
			rfd->rfd_status = 0;
			rbd->rbd_count = 0;
			rbd->rbd_size = IEE_RBD_EL | rx_map->dm_segs[0].ds_len;
			printf("%s: iee_intr: receive error %d, rfd_status="
			    "0x%.4x, rfd_count=0x%.4x\n", sc->sc_dev.dv_xname,
			    ++sc->sc_rx_err, rfd->rfd_status, rbd->rbd_count);
			sc->sc_rx_done = (sc->sc_rx_done + 1) % IEE_NRFD;
			continue;
		}
		rfd->rfd_status = 0;
		bus_dmamap_sync(sc->sc_dmat, rx_map, 0, rx_mbuf->m_ext.ext_size,
		    BUS_DMASYNC_POSTREAD);
		rx_mbuf->m_pkthdr.len = rx_mbuf->m_len =
		    rbd->rbd_count & IEE_RBD_COUNT;
		rx_mbuf->m_pkthdr.rcvif = ifp;
		MGETHDR(new_mbuf, M_DONTWAIT, MT_DATA);
		if (new_mbuf == NULL) {
			printf("%s: iee_intr: can't allocate mbuf\n",
			    sc->sc_dev.dv_xname);
			break;
		}
		MCLAIM(new_mbuf, &sc->sc_ethercom.ec_rx_mowner);
		MCLGET(new_mbuf, M_DONTWAIT);
		if ((new_mbuf->m_flags & M_EXT) == 0) {
			printf("%s: iee_intr: can't alloc mbuf cluster\n",
			    sc->sc_dev.dv_xname);
			m_freem(new_mbuf);
			break;
		}
		bus_dmamap_unload(sc->sc_dmat, rx_map);
		if (bus_dmamap_load(sc->sc_dmat, rx_map,
		    new_mbuf->m_ext.ext_buf, new_mbuf->m_ext.ext_size,
		    NULL, BUS_DMA_READ | BUS_DMA_NOWAIT) != 0)
			panic("%s: iee_intr: can't load RX DMA map\n",
			    sc->sc_dev.dv_xname);
		bus_dmamap_sync(sc->sc_dmat, rx_map, 0,
		    new_mbuf->m_ext.ext_size, BUS_DMASYNC_PREREAD);
#if NBPFILTER > 0
		if (ifp->if_bpf != 0)
			bpf_mtap(ifp->if_bpf, rx_mbuf);
#endif /* NBPFILTER > 0 */
		(*ifp->if_input)(ifp, rx_mbuf);
		ifp->if_ipackets++;
		sc->sc_rx_mbuf[sc->sc_rx_done] = new_mbuf;
		rbd->rbd_count = 0;
		rbd->rbd_size = IEE_RBD_EL | rx_map->dm_segs[0].ds_len;
		rbd->rbd_rb_addr = rx_map->dm_segs[0].ds_addr;
		sc->sc_rx_done = (sc->sc_rx_done + 1) % IEE_NRFD;
		rfd = SC_RFD(sc->sc_rx_done);
	}
	if ((scb_status & IEE_SCB_RUS) == IEE_SCB_RUS_NR1
	    || (scb_status & IEE_SCB_RUS) == IEE_SCB_RUS_NR2
	    || (scb_status & IEE_SCB_RUS) == IEE_SCB_RUS_NR3) {
		/* Receive Overrun, reinit receive ring buffer. */
		for (n = 0 ; n < IEE_NRFD ; n++) {
			SC_RFD(n)->rfd_cmd = IEE_RFD_SF;
			SC_RFD(n)->rfd_link_addr = IEE_PHYS_SHMEM(IEE_RFD_OFF
			    + IEE_RFD_SZ * ((n + 1) % IEE_NRFD));
			SC_RBD(n)->rbd_next_rbd = IEE_PHYS_SHMEM(IEE_RBD_OFF
			    + IEE_RBD_SZ * ((n + 1) % IEE_NRFD));
			SC_RBD(n)->rbd_size = IEE_RBD_EL |
			    sc->sc_rx_map[n]->dm_segs[0].ds_len;
			SC_RBD(n)->rbd_rb_addr =
			    sc->sc_rx_map[n]->dm_segs[0].ds_addr;
		}
		SC_RFD(0)->rfd_rbd_addr = IEE_PHYS_SHMEM(IEE_RBD_OFF);
		sc->sc_rx_done = 0;
		bus_dmamap_sync(sc->sc_dmat, sc->sc_shmem_map, IEE_RFD_OFF,
		    IEE_RFD_LIST_SZ + IEE_RBD_LIST_SZ, BUS_DMASYNC_PREWRITE);
		(sc->sc_iee_cmd)(sc, IEE_SCB_RUC_ST);
		printf("%s: iee_intr: receive ring buffer overrun\n",
		    sc->sc_dev.dv_xname);
	}

	if (sc->sc_next_cb != 0
	    && (SC_CB(sc->sc_next_cb - 1)->cb_status & IEE_CB_C) != 0) {
		/* CMD list finished */
		ifp->if_timer = 0;
		if (sc->sc_next_tbd != 0) {
			/* A TX CMD list finished, clenup */
			for (n = 0 ; n < sc->sc_next_cb ; n++) {
				m_freem(sc->sc_tx_mbuf[n]);
				sc->sc_tx_mbuf[n] = NULL;
				bus_dmamap_unload(sc->sc_dmat,sc->sc_tx_map[n]);
				if ((SC_CB(n)->cb_status & IEE_CB_COL) != 0 &&
				    (SC_CB(n)->cb_status & IEE_CB_MAXCOL) == 0)
					col = 16;
				else
					col = SC_CB(n)->cb_status
					    & IEE_CB_MAXCOL;
				sc->sc_tx_col += col;
				if ((SC_CB(n)->cb_status & IEE_CB_OK) != 0) {
					ifp->if_opackets++;
					ifp->if_collisions += col;
				}
			}
			sc->sc_next_tbd = 0;
			ifp->if_flags &= ~IFF_OACTIVE;
		}
		for (n = 0 ; n < sc->sc_next_cb ; n++) {
			/* Check if a CMD failed, but ignore TX errors. */
			if ((SC_CB(n)->cb_cmd & IEE_CB_CMD) != IEE_CB_CMD_TR
			    && ((SC_CB(n)->cb_status & IEE_CB_OK) == 0))
				printf("%s: iee_intr: scb_status=0x%x "
				    "scb_cmd=0x%x failed command %d: "
				    "cb_status[%d]=0x%.4x cb_cmd[%d]=0x%.4x\n",
				    sc->sc_dev.dv_xname, scb_status, scb_cmd,
				    ++sc->sc_cmd_err, n, SC_CB(n)->cb_status,
				    n, SC_CB(n)->cb_cmd);
		}
		sc->sc_next_cb = 0;
		if ((sc->sc_flags & IEE_WANT_MCAST) != 0) {
			iee_cb_setup(sc, IEE_CB_CMD_MCS | IEE_CB_S | IEE_CB_EL
			    | IEE_CB_I);
			(sc->sc_iee_cmd)(sc, IEE_SCB_CUC_EXE);
		} else
			/* Try to get defered packets going. */
			iee_start(ifp);
	}
	if (IEE_SWAP(SC_SCB->scb_crc_err) != sc->sc_crc_err) {
		sc->sc_crc_err = IEE_SWAP(SC_SCB->scb_crc_err);
		printf("%s: iee_intr: crc_err=%d\n", sc->sc_dev.dv_xname,
		    sc->sc_crc_err);
	}
	if (IEE_SWAP(SC_SCB->scb_align_err) != sc->sc_align_err) {
		sc->sc_align_err = IEE_SWAP(SC_SCB->scb_align_err);
		printf("%s: iee_intr: align_err=%d\n", sc->sc_dev.dv_xname,
		    sc->sc_align_err);
	}
	if (IEE_SWAP(SC_SCB->scb_resource_err) != sc->sc_resource_err) {
		sc->sc_resource_err = IEE_SWAP(SC_SCB->scb_resource_err);
		printf("%s: iee_intr: resource_err=%d\n", sc->sc_dev.dv_xname,
		    sc->sc_resource_err);
	}
	if (IEE_SWAP(SC_SCB->scb_overrun_err) != sc->sc_overrun_err) {
		sc->sc_overrun_err = IEE_SWAP(SC_SCB->scb_overrun_err);
		printf("%s: iee_intr: overrun_err=%d\n", sc->sc_dev.dv_xname,
		    sc->sc_overrun_err);
	}
	if (IEE_SWAP(SC_SCB->scb_rcvcdt_err) != sc->sc_rcvcdt_err) {
		sc->sc_rcvcdt_err = IEE_SWAP(SC_SCB->scb_rcvcdt_err);
		printf("%s: iee_intr: rcvcdt_err=%d\n", sc->sc_dev.dv_xname,
		    sc->sc_rcvcdt_err);
	}
	if (IEE_SWAP(SC_SCB->scb_short_fr_err) != sc->sc_short_fr_err) {
		sc->sc_short_fr_err = IEE_SWAP(SC_SCB->scb_short_fr_err);
		printf("%s: iee_intr: short_fr_err=%d\n", sc->sc_dev.dv_xname,
		    sc->sc_short_fr_err);
	}
	bus_dmamap_sync(sc->sc_dmat, sc->sc_shmem_map, 0, IEE_SHMEM_MAX,
	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
	(sc->sc_iee_cmd)(sc, IEE_SCB_ACK);
	return(1);
}



/*
How Command Block List Processing is done.

A runing CBL is never manipulated. If there is a CBL already runing,
further CMDs are deferd until the current list is done. A new list is
setup when the old has finished.
This eases programming. To manipulate a runing CBL it is neccesary to
suspend the Command Unit to avoid race conditions. After a suspend
is sent we have to wait for an interrupt that ACKs the suspend. Then
we can manipulate the CBL and resume operation. I am not sure that this
is more effective then the current, much simpler approach. => KISS
See i82596CA data sheet page 26.

A CBL is runing or on the way to be set up when (sc->sc_next_cb != 0).

A CBL may consist of TX CMDs, and _only_ TX CMDs.
A TX CBL is runing or on the way to be set up when
((sc->sc_next_cb != 0) && (sc->sc_next_tbd != 0)).

A CBL may consist of other non-TX CMDs like IAS or CONF, and _only_
non-TX CMDs.

This comes mostly through the way how an Ethernet driver works and
because runing CBLs are not manipulated when they are on the way. If
if_start() is called there will be TX CMDs enqueued so we have a runing
CBL and other CMDs from e.g. if_ioctl() will be deferd and vice versa.

The Multicast Setup Command is special. A MCS needs more space then
a single CB has. Actual space requiement depends on the length of the
multicast list. So we allways defer MCS until other CBLs are finished,
then we setup a CONF CMD in the first CB. The CONF CMD is needed to
turn ALLMULTI on the hardware on or off. The MCS is the 2nd CB and may
use all the remaining space in the CBL and the Transmit Buffer Descriptor
List. (Therefore CBL and TBDL must be continious in pysical and virtual
memory. This is guaranteed through the definitions of the list offsets
in i82596reg.h and because it is only a single DMA segment used for all
lists.) When ALLMULTI is enabled via the CONF CMD, the MCS is run with
a multicast list length of 0, thus disabling the multicast filter.
A defered MCS is signaled via ((sc->sc_flags & IEE_WANT_MCAST) != 0)
*/
void
iee_cb_setup(struct iee_softc *sc, uint32_t cmd)
{
	struct iee_cb *cb = SC_CB(sc->sc_next_cb);
	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
	struct ether_multistep step;
	struct ether_multi *enm;

	memset(cb, 0, IEE_CB_SZ);
	cb->cb_cmd = cmd;
	switch(cmd & IEE_CB_CMD) {
	case IEE_CB_CMD_NOP:	/* NOP CMD */
		break;
	case IEE_CB_CMD_IAS:	/* Individual Address Setup */
		memcpy(__UNVOLATILE(cb->cb_ind_addr), LLADDR(ifp->if_sadl),
		    ETHER_ADDR_LEN);
		break;
	case IEE_CB_CMD_CONF:	/* Configure */
		memcpy(__UNVOLATILE(cb->cb_cf), sc->sc_cf, sc->sc_cf[0]
		    & IEE_CF_0_CNT_M);
		break;
	case IEE_CB_CMD_MCS:	/* Multicast Setup */
		if (sc->sc_next_cb != 0) {
			sc->sc_flags |= IEE_WANT_MCAST;
			return;
		}
		sc->sc_flags &= ~IEE_WANT_MCAST;
		if ((sc->sc_cf[8] & IEE_CF_8_PRM) != 0) {
			/* Need no multicast filter in promisc mode. */
			iee_cb_setup(sc, IEE_CB_CMD_CONF | IEE_CB_S | IEE_CB_EL
			    | IEE_CB_I);
			return;
		}
		/* Leave room for a CONF CMD to en/dis-able ALLMULTI mode */
		cb = SC_CB(sc->sc_next_cb + 1);
		cb->cb_cmd = cmd;
		cb->cb_mcast.mc_size = 0;
		ETHER_FIRST_MULTI(step, &sc->sc_ethercom, enm);
		while (enm != NULL) {
			if (memcmp(enm->enm_addrlo, enm->enm_addrhi,
			    ETHER_ADDR_LEN) != 0 || cb->cb_mcast.mc_size
			    * ETHER_ADDR_LEN + 2 * IEE_CB_SZ
			    > IEE_CB_LIST_SZ + IEE_TBD_LIST_SZ) {
				cb->cb_mcast.mc_size = 0;
				break;
			}
			memcpy(__UNVOLATILE(&cb->cb_mcast.mc_addrs[
			    cb->cb_mcast.mc_size * ETHER_ADDR_LEN]),
			    enm->enm_addrlo, ETHER_ADDR_LEN);
			ETHER_NEXT_MULTI(step, enm);
			cb->cb_mcast.mc_size++;
		}
		if (cb->cb_mcast.mc_size == 0) {
			/* Can't do exact mcast filtering, do ALLMULTI mode. */
			ifp->if_flags |= IFF_ALLMULTI;
			sc->sc_cf[11] &= ~IEE_CF_11_MCALL;
		} else {
			/* disable ALLMULTI and load mcast list */
			ifp->if_flags &= ~IFF_ALLMULTI;
			sc->sc_cf[11] |= IEE_CF_11_MCALL;
			/* Mcast setup may need more then IEE_CB_SZ bytes. */
			bus_dmamap_sync(sc->sc_dmat, sc->sc_shmem_map,
			    IEE_CB_OFF, IEE_CB_LIST_SZ + IEE_TBD_LIST_SZ,
			    BUS_DMASYNC_PREWRITE);
		}
		iee_cb_setup(sc, IEE_CB_CMD_CONF);
		break;
	case IEE_CB_CMD_TR:	/* Transmit */
		cb->cb_transmit.tx_tbd_addr = IEE_PHYS_SHMEM(IEE_TBD_OFF
		    + IEE_TBD_SZ * sc->sc_next_tbd);
		cb->cb_cmd |= IEE_CB_SF; /* Allways use Flexible Mode. */
		break;
	case IEE_CB_CMD_TDR:	/* Time Domain Reflectometry */
		break;
	case IEE_CB_CMD_DUMP:	/* Dump */
		break;
	case IEE_CB_CMD_DIAG:	/* Diagnose */
		break;
	default:
		/* can't happen */
		break;
	}
	cb->cb_link_addr = IEE_PHYS_SHMEM(IEE_CB_OFF + IEE_CB_SZ *
	    (sc->sc_next_cb + 1));
	bus_dmamap_sync(sc->sc_dmat, sc->sc_shmem_map, IEE_CB_OFF
	    + IEE_CB_SZ * sc->sc_next_cb, IEE_CB_SZ, BUS_DMASYNC_PREWRITE);
	sc->sc_next_cb++;
	ifp->if_timer = 5;
	return;
}



void
iee_attach(struct iee_softc *sc, uint8_t *eth_addr, int *media, int nmedia,
    int defmedia)
{
	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
	int n;

	/* Set pointer to Intermediate System Configuration Pointer. */
	/* Phys. addr. in big endian order. (Big endian as defined by Intel.) */
	SC_SCP->scp_iscp_addr = IEE_SWAP(IEE_PHYS_SHMEM(IEE_ISCP_OFF));
	/* Set pointer to System Control Block. */
	/* Phys. addr. in big endian order. (Big endian as defined by Intel.) */
	SC_ISCP->iscp_scb_addr = IEE_SWAP(IEE_PHYS_SHMEM(IEE_SCB_OFF));
	/* Set pointer to Receive Frame Area. (physical address) */
	SC_SCB->scb_rfa_addr = IEE_PHYS_SHMEM(IEE_RFD_OFF);
	/* Set pointer to Command Block. (physical address) */
	SC_SCB->scb_cmd_blk_addr = IEE_PHYS_SHMEM(IEE_CB_OFF);

	ifmedia_init(&sc->sc_ifmedia, 0, iee_mediachange, iee_mediastatus);
	if (media != NULL) {
		for (n = 0 ; n < nmedia ; n++)
			ifmedia_add(&sc->sc_ifmedia, media[n], 0, NULL);
		ifmedia_set(&sc->sc_ifmedia, defmedia);
	} else {
		ifmedia_add(&sc->sc_ifmedia, IFM_ETHER | IFM_NONE, 0, NULL);
		ifmedia_set(&sc->sc_ifmedia, IFM_ETHER | IFM_NONE);
	}

	ifp->if_softc = sc;
	strcpy(ifp->if_xname, sc->sc_dev.dv_xname);
	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
	ifp->if_start = iee_start;	/* initiate output routine */
	ifp->if_ioctl = iee_ioctl;	/* ioctl routine */
	ifp->if_init = iee_init;	/* init routine */
	ifp->if_stop = iee_stop;	/* stop routine */
	ifp->if_watchdog = iee_watchdog;	/* timer routine */
	ifp->if_drain = iee_drain;	/* routine to release resources */
	IFQ_SET_READY(&ifp->if_snd);
	/* iee supports IEEE 802.1Q Virtual LANs, see vlan(4). */
	sc->sc_ethercom.ec_capabilities |= ETHERCAP_VLAN_MTU;

	if_attach(ifp);
	ether_ifattach(ifp, eth_addr);

	aprint_normal(": Intel 82596%s address %s\n",
	    i82596_typenames[ sc->sc_type], ether_sprintf(eth_addr));

	for (n = 0 ; n < IEE_NCB ; n++)
		sc->sc_tx_map[n] = NULL;
	for (n = 0 ; n < IEE_NRFD ; n++) {
		sc->sc_rx_mbuf[n] = NULL;
		sc->sc_rx_map[n] = NULL;
	}
	sc->sc_tx_timeout = 0;
	sc->sc_setup_timeout = 0;
	(sc->sc_iee_reset)(sc);
	return;
}



void
iee_detach(struct iee_softc *sc, int flags)
{
	struct ifnet *ifp = &sc->sc_ethercom.ec_if;

	if ((ifp->if_flags & IFF_RUNNING) != 0)
		iee_stop(ifp, 1);
	ether_ifdetach(ifp);
	if_detach(ifp);
	return;
}



/* media change and status callback */
int
iee_mediachange(struct ifnet *ifp)
{
	struct iee_softc *sc = ifp->if_softc;

	if (sc->sc_mediachange != NULL)
		return ((sc->sc_mediachange)(ifp));
	return(0);
}



void
iee_mediastatus(struct ifnet *ifp, struct ifmediareq *ifmreq)
{
	struct iee_softc *sc = ifp->if_softc;

	if (sc->sc_mediastatus != NULL)
		return ((sc->sc_mediastatus)(ifp, ifmreq));
	return;
}



/* initiate output routine */
void
iee_start(struct ifnet *ifp)
{
	struct iee_softc *sc = ifp->if_softc;
	struct mbuf *m = NULL;
	int t;
	int n;

	if (sc->sc_next_cb != 0)
		/* There is already a CMD runing. Defer packet enqueueing. */
		return;
	for (t = 0 ; t < IEE_NCB ; t++) {
		IFQ_DEQUEUE(&ifp->if_snd, sc->sc_tx_mbuf[t]);
		if (sc->sc_tx_mbuf[t] == NULL)
			break;
		if (bus_dmamap_load_mbuf(sc->sc_dmat, sc->sc_tx_map[t],
		    sc->sc_tx_mbuf[t], BUS_DMA_WRITE | BUS_DMA_NOWAIT) != 0) {
			/*
			 * The packet needs more TBD then we support.
			 * Copy the packet into a mbuf cluster to get it out.
			 */
			printf("%s: iee_start: failed to load DMA map\n",
			    sc->sc_dev.dv_xname);
			MGETHDR(m, M_DONTWAIT, MT_DATA);
			if (m == NULL) {
				printf("%s: iee_start: can't allocate mbuf\n",
				    sc->sc_dev.dv_xname);
				m_freem(sc->sc_tx_mbuf[t]);
				t--;
				continue;
			}
			MCLAIM(m, &sc->sc_ethercom.ec_rx_mowner);
			MCLGET(m, M_DONTWAIT);
			if ((m->m_flags & M_EXT) == 0) {
				printf("%s: iee_start: can't allocate mbuf "
				    "cluster\n", sc->sc_dev.dv_xname);
				m_freem(sc->sc_tx_mbuf[t]);
				m_freem(m);
				t--;
				continue;
			}
			m_copydata(sc->sc_tx_mbuf[t], 0,
			    sc->sc_tx_mbuf[t]->m_pkthdr.len, mtod(m, caddr_t));
			m->m_pkthdr.len = sc->sc_tx_mbuf[t]->m_pkthdr.len;
			m->m_len = sc->sc_tx_mbuf[t]->m_pkthdr.len;
			m_freem(sc->sc_tx_mbuf[t]);
			sc->sc_tx_mbuf[t] = m;
			if(bus_dmamap_load_mbuf(sc->sc_dmat, sc->sc_tx_map[t],
		    	    m, BUS_DMA_WRITE | BUS_DMA_NOWAIT) != 0) {
				printf("%s: iee_start: can't load TX DMA map\n",
				    sc->sc_dev.dv_xname);
				m_freem(sc->sc_tx_mbuf[t]);
				t--;
				continue;
			}
		}
		for (n = 0 ; n < sc->sc_tx_map[t]->dm_nsegs ; n++) {
			SC_TBD(sc->sc_next_tbd + n)->tbd_tb_addr =
			    sc->sc_tx_map[t]->dm_segs[n].ds_addr;
			SC_TBD(sc->sc_next_tbd + n)->tbd_size =
			    sc->sc_tx_map[t]->dm_segs[n].ds_len;
			SC_TBD(sc->sc_next_tbd + n)->tbd_link_addr =
			    IEE_PHYS_SHMEM(IEE_TBD_OFF + IEE_TBD_SZ
			    * (sc->sc_next_tbd + n + 1));
		}
		SC_TBD(sc->sc_next_tbd + n - 1)->tbd_size |= IEE_CB_EL;
		bus_dmamap_sync(sc->sc_dmat, sc->sc_tx_map[t], 0,
		    sc->sc_tx_map[t]->dm_mapsize, BUS_DMASYNC_PREWRITE);
		IFQ_POLL(&ifp->if_snd, m);
		if (m == NULL)
			iee_cb_setup(sc, IEE_CB_CMD_TR | IEE_CB_S | IEE_CB_EL
			    | IEE_CB_I);
		else
			iee_cb_setup(sc, IEE_CB_CMD_TR);
		sc->sc_next_tbd += n;
#if NBPFILTER > 0
		/* Pass packet to bpf if someone listens. */
		if (ifp->if_bpf)
			bpf_mtap(ifp->if_bpf, sc->sc_tx_mbuf[t]);
#endif
	}
	if (t == 0)
		/* No packets got set up for TX. */
		return;
	if (t == IEE_NCB)
		ifp->if_flags |= IFF_OACTIVE;
	bus_dmamap_sync(sc->sc_dmat, sc->sc_shmem_map, IEE_CB_SZ,
	    IEE_CB_LIST_SZ + IEE_TBD_LIST_SZ, BUS_DMASYNC_PREWRITE);
	(sc->sc_iee_cmd)(sc, IEE_SCB_CUC_EXE);
	return;
}



/* ioctl routine */
int
iee_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
{
	struct iee_softc *sc = ifp->if_softc;
	int s;
	int err;

	s = splnet();
	switch (cmd) {
	case SIOCSIFMEDIA:
	case SIOCGIFMEDIA:
		err = ifmedia_ioctl(ifp, (struct ifreq *) data,
		    &sc->sc_ifmedia, cmd);
		break;

	default:
		err = ether_ioctl(ifp, cmd, data);
		if (err == ENETRESET) {
			/*
			 * Multicast list as changed; set the hardware filter
			 * accordingly.
			 */
			if (ifp->if_flags & IFF_RUNNING) {
				iee_cb_setup(sc, IEE_CB_CMD_MCS | IEE_CB_S |
				    IEE_CB_EL | IEE_CB_I);
				if ((sc->sc_flags & IEE_WANT_MCAST) == 0)
					(*sc->sc_iee_cmd)(sc, IEE_SCB_CUC_EXE);
			}
			err = 0;
		}
		break;
	}
	splx(s);
	return(err);
}



/* init routine */
int
iee_init(struct ifnet *ifp)
{
	struct iee_softc *sc = ifp->if_softc;
	int r;
	int t;
	int n;
	int err;

	sc->sc_next_cb = 0;
	sc->sc_next_tbd = 0;
	sc->sc_flags &= ~IEE_WANT_MCAST;
	sc->sc_rx_done = 0;
	SC_SCB->scb_crc_err = 0;
	SC_SCB->scb_align_err = 0;
	SC_SCB->scb_resource_err = 0;
	SC_SCB->scb_overrun_err = 0;
	SC_SCB->scb_rcvcdt_err = 0;
	SC_SCB->scb_short_fr_err = 0;
	sc->sc_crc_err = 0;
	sc->sc_align_err = 0;
	sc->sc_resource_err = 0;
	sc->sc_overrun_err = 0;
	sc->sc_rcvcdt_err = 0;
	sc->sc_short_fr_err = 0;
	sc->sc_tx_col = 0;
	sc->sc_rx_err = 0;
	sc->sc_cmd_err = 0;
	/* Create Transmit DMA maps. */
	for (t = 0 ; t < IEE_NCB ; t++) {
		if (sc->sc_tx_map[t] == NULL && bus_dmamap_create(sc->sc_dmat,
		    MCLBYTES, IEE_NTBD, MCLBYTES, 0, BUS_DMA_NOWAIT,
		    &sc->sc_tx_map[t]) != 0) {
			printf("%s: iee_init: can't create TX DMA map\n",
			    sc->sc_dev.dv_xname);
			for (n = 0 ; n < t ; n++)
				bus_dmamap_destroy(sc->sc_dmat,
				    sc->sc_tx_map[n]);
			return(ENOBUFS);
		}
	}
	/* Initialize Receive Frame and Receive Buffer Descriptors */
	err = 0;
	memset(SC_RFD(0), 0, IEE_RFD_LIST_SZ);
	memset(SC_RBD(0), 0, IEE_RBD_LIST_SZ);
	for (r = 0 ; r < IEE_NRFD ; r++) {
		SC_RFD(r)->rfd_cmd = IEE_RFD_SF;
		SC_RFD(r)->rfd_link_addr = IEE_PHYS_SHMEM(IEE_RFD_OFF
		    + IEE_RFD_SZ * ((r + 1) % IEE_NRFD));

		SC_RBD(r)->rbd_next_rbd = IEE_PHYS_SHMEM(IEE_RBD_OFF
		    + IEE_RBD_SZ * ((r + 1) % IEE_NRFD));
		if (sc->sc_rx_mbuf[r] == NULL) {
			MGETHDR(sc->sc_rx_mbuf[r], M_DONTWAIT, MT_DATA);
			if (sc->sc_rx_mbuf[r] == NULL) {
				printf("%s: iee_init: can't allocate mbuf\n",
				    sc->sc_dev.dv_xname);
				err = 1;
				break;
			}
			MCLAIM(sc->sc_rx_mbuf[r],&sc->sc_ethercom.ec_rx_mowner);
			MCLGET(sc->sc_rx_mbuf[r], M_DONTWAIT);
			if ((sc->sc_rx_mbuf[r]->m_flags & M_EXT) == 0) {
				printf("%s: iee_init: can't allocate mbuf"
				    " cluster\n", sc->sc_dev.dv_xname);
				m_freem(sc->sc_rx_mbuf[r]);
				err = 1;
				break;
			}
		}
		if (sc->sc_rx_map[r] == NULL && bus_dmamap_create(sc->sc_dmat,
		    MCLBYTES, 1, MCLBYTES , 0, BUS_DMA_NOWAIT,
		    &sc->sc_rx_map[r]) != 0) {
				printf("%s: iee_init: can't create RX "
				    "DMA map\n", sc->sc_dev.dv_xname);
				m_freem(sc->sc_rx_mbuf[r]);
				err = 1;
				break;
			}
		if (bus_dmamap_load(sc->sc_dmat, sc->sc_rx_map[r],
		    sc->sc_rx_mbuf[r]->m_ext.ext_buf,
		    sc->sc_rx_mbuf[r]->m_ext.ext_size, NULL,
		    BUS_DMA_READ | BUS_DMA_NOWAIT) != 0) {
			printf("%s: iee_init: can't load RX DMA map\n",
			    sc->sc_dev.dv_xname);
			bus_dmamap_destroy(sc->sc_dmat, sc->sc_rx_map[r]);
			m_freem(sc->sc_rx_mbuf[r]);
			err = 1;
			break;
		}
		bus_dmamap_sync(sc->sc_dmat, sc->sc_rx_map[r], 0,
		    sc->sc_rx_mbuf[r]->m_ext.ext_size, BUS_DMASYNC_PREREAD);
		SC_RBD(r)->rbd_size = sc->sc_rx_map[r]->dm_segs[0].ds_len;
		SC_RBD(r)->rbd_rb_addr= sc->sc_rx_map[r]->dm_segs[0].ds_addr;
	}
	SC_RFD(0)->rfd_rbd_addr = IEE_PHYS_SHMEM(IEE_RBD_OFF);
	if (err != 0) {
		for (n = 0 ; n < r; n++) {
			m_freem(sc->sc_rx_mbuf[n]);
			sc->sc_rx_mbuf[n] = NULL;
			bus_dmamap_unload(sc->sc_dmat, sc->sc_rx_map[n]);
			bus_dmamap_destroy(sc->sc_dmat, sc->sc_rx_map[n]);
			sc->sc_rx_map[n] = NULL;
		}
		for (n = 0 ; n < t ; n++) {
			bus_dmamap_destroy(sc->sc_dmat, sc->sc_tx_map[n]);
			sc->sc_tx_map[n] = NULL;
		}
		return(ENOBUFS);
	}

	(sc->sc_iee_reset)(sc);
	iee_cb_setup(sc, IEE_CB_CMD_IAS);
	sc->sc_cf[0] = IEE_CF_0_DEF | IEE_CF_0_PREF;
	sc->sc_cf[1] = IEE_CF_1_DEF;
	sc->sc_cf[2] = IEE_CF_2_DEF;
	sc->sc_cf[3] = IEE_CF_3_ADDRLEN_DEF | IEE_CF_3_NSAI
	    | IEE_CF_3_PREAMLEN_DEF;
	sc->sc_cf[4] = IEE_CF_4_DEF;
	sc->sc_cf[5] = IEE_CF_5_DEF;
	sc->sc_cf[6] = IEE_CF_6_DEF;
	sc->sc_cf[7] = IEE_CF_7_DEF;
	sc->sc_cf[8] = IEE_CF_8_DEF;
	sc->sc_cf[9] = IEE_CF_9_DEF;
	sc->sc_cf[10] = IEE_CF_10_DEF;
	sc->sc_cf[11] = IEE_CF_11_DEF & ~IEE_CF_11_LNGFLD;
	sc->sc_cf[12] = IEE_CF_12_DEF;
	sc->sc_cf[13] = IEE_CF_13_DEF;
	iee_cb_setup(sc, IEE_CB_CMD_CONF | IEE_CB_S | IEE_CB_EL);
	SC_SCB->scb_rfa_addr = IEE_PHYS_SHMEM(IEE_RFD_OFF);
	bus_dmamap_sync(sc->sc_dmat, sc->sc_shmem_map, 0, IEE_SHMEM_MAX,
	    BUS_DMASYNC_PREWRITE);
	(sc->sc_iee_cmd)(sc, IEE_SCB_CUC_EXE | IEE_SCB_RUC_ST);
	/* Issue a Channel Attention to ACK interrupts we may have caused. */
	(sc->sc_iee_cmd)(sc, IEE_SCB_ACK);

	/* Mark the interface as running and ready to RX/TX packets. */
	ifp->if_flags |= IFF_RUNNING;
	ifp->if_flags &= ~IFF_OACTIVE;
	return(0);
}



/* stop routine */
void
iee_stop(struct ifnet *ifp, int disable)
{
	struct iee_softc *sc = ifp->if_softc;
	int n;

	ifp->if_flags &= ~IFF_RUNNING;
	ifp->if_flags |= IFF_OACTIVE;
	ifp->if_timer = 0;
	/* Reset the chip to get it quiet. */
	(sc->sc_iee_reset)(ifp->if_softc);
	/* Issue a Channel Attention to ACK interrupts we may have caused. */
	(sc->sc_iee_cmd)(ifp->if_softc, IEE_SCB_ACK);
	/* Release any dynamically allocated ressources. */
	for (n = 0 ; n < IEE_NCB ; n++) {
		if (sc->sc_tx_map[n] != NULL)
			bus_dmamap_destroy(sc->sc_dmat, sc->sc_tx_map[n]);
		sc->sc_tx_map[n] = NULL;
	}
	for (n = 0 ; n < IEE_NRFD ; n++) {
		if (sc->sc_rx_mbuf[n] != NULL)
			m_freem(sc->sc_rx_mbuf[n]);
		sc->sc_rx_mbuf[n] = NULL;
		if (sc->sc_rx_map[n] != NULL) {
			bus_dmamap_unload(sc->sc_dmat, sc->sc_rx_map[n]);
			bus_dmamap_destroy(sc->sc_dmat, sc->sc_rx_map[n]);
		}
		sc->sc_rx_map[n] = NULL;
	}
	return;
}



/* timer routine */
void
iee_watchdog(struct ifnet *ifp)
{
	struct iee_softc *sc = ifp->if_softc;

	(sc->sc_iee_reset)(sc);
	if (sc->sc_next_tbd != 0)
		printf("%s: iee_watchdog: transmit timeout %d\n",
		    sc->sc_dev.dv_xname, ++sc->sc_tx_timeout);
	else
		printf("%s: iee_watchdog: setup timeout %d\n",
		    sc->sc_dev.dv_xname, ++sc->sc_setup_timeout);
	iee_init(ifp);
	return;
}



/* routine to release res. */
void
iee_drain(struct ifnet *ifp)
{
	iee_stop(ifp, 0);
	return;
}