Home | History | Annotate | Line # | Download | only in ic
rtl8169.c revision 1.14.2.7
      1 /*	$NetBSD: rtl8169.c,v 1.14.2.7 2007/10/04 18:50:22 bouyer Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 1997, 1998-2003
      5  *	Bill Paul <wpaul (at) windriver.com>.  All rights reserved.
      6  *
      7  * Redistribution and use in source and binary forms, with or without
      8  * modification, are permitted provided that the following conditions
      9  * are met:
     10  * 1. Redistributions of source code must retain the above copyright
     11  *    notice, this list of conditions and the following disclaimer.
     12  * 2. Redistributions in binary form must reproduce the above copyright
     13  *    notice, this list of conditions and the following disclaimer in the
     14  *    documentation and/or other materials provided with the distribution.
     15  * 3. All advertising materials mentioning features or use of this software
     16  *    must display the following acknowledgement:
     17  *	This product includes software developed by Bill Paul.
     18  * 4. Neither the name of the author nor the names of any co-contributors
     19  *    may be used to endorse or promote products derived from this software
     20  *    without specific prior written permission.
     21  *
     22  * THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
     23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     25  * ARE DISCLAIMED.  IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
     26  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     27  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     28  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     29  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     30  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     31  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
     32  * THE POSSIBILITY OF SUCH DAMAGE.
     33  */
     34 
     35 #include <sys/cdefs.h>
     36 /* $FreeBSD: /repoman/r/ncvs/src/sys/dev/re/if_re.c,v 1.20 2004/04/11 20:34:08 ru Exp $ */
     37 
     38 /*
     39  * RealTek 8139C+/8169/8169S/8110S PCI NIC driver
     40  *
     41  * Written by Bill Paul <wpaul (at) windriver.com>
     42  * Senior Networking Software Engineer
     43  * Wind River Systems
     44  */
     45 
     46 /*
     47  * This driver is designed to support RealTek's next generation of
     48  * 10/100 and 10/100/1000 PCI ethernet controllers. There are currently
     49  * four devices in this family: the RTL8139C+, the RTL8169, the RTL8169S
     50  * and the RTL8110S.
     51  *
     52  * The 8139C+ is a 10/100 ethernet chip. It is backwards compatible
     53  * with the older 8139 family, however it also supports a special
     54  * C+ mode of operation that provides several new performance enhancing
     55  * features. These include:
     56  *
     57  *	o Descriptor based DMA mechanism. Each descriptor represents
     58  *	  a single packet fragment. Data buffers may be aligned on
     59  *	  any byte boundary.
     60  *
     61  *	o 64-bit DMA
     62  *
     63  *	o TCP/IP checksum offload for both RX and TX
     64  *
     65  *	o High and normal priority transmit DMA rings
     66  *
     67  *	o VLAN tag insertion and extraction
     68  *
     69  *	o TCP large send (segmentation offload)
     70  *
     71  * Like the 8139, the 8139C+ also has a built-in 10/100 PHY. The C+
     72  * programming API is fairly straightforward. The RX filtering, EEPROM
     73  * access and PHY access is the same as it is on the older 8139 series
     74  * chips.
     75  *
     76  * The 8169 is a 64-bit 10/100/1000 gigabit ethernet MAC. It has almost the
     77  * same programming API and feature set as the 8139C+ with the following
     78  * differences and additions:
     79  *
     80  *	o 1000Mbps mode
     81  *
     82  *	o Jumbo frames
     83  *
     84  * 	o GMII and TBI ports/registers for interfacing with copper
     85  *	  or fiber PHYs
     86  *
     87  *      o RX and TX DMA rings can have up to 1024 descriptors
     88  *        (the 8139C+ allows a maximum of 64)
     89  *
     90  *	o Slight differences in register layout from the 8139C+
     91  *
     92  * The TX start and timer interrupt registers are at different locations
     93  * on the 8169 than they are on the 8139C+. Also, the status word in the
     94  * RX descriptor has a slightly different bit layout. The 8169 does not
     95  * have a built-in PHY. Most reference boards use a Marvell 88E1000 'Alaska'
     96  * copper gigE PHY.
     97  *
     98  * The 8169S/8110S 10/100/1000 devices have built-in copper gigE PHYs
     99  * (the 'S' stands for 'single-chip'). These devices have the same
    100  * programming API as the older 8169, but also have some vendor-specific
    101  * registers for the on-board PHY. The 8110S is a LAN-on-motherboard
    102  * part designed to be pin-compatible with the RealTek 8100 10/100 chip.
    103  *
    104  * This driver takes advantage of the RX and TX checksum offload and
    105  * VLAN tag insertion/extraction features. It also implements TX
    106  * interrupt moderation using the timer interrupt registers, which
    107  * significantly reduces TX interrupt load. There is also support
    108  * for jumbo frames, however the 8169/8169S/8110S can not transmit
    109  * jumbo frames larger than 7.5K, so the max MTU possible with this
    110  * driver is 7500 bytes.
    111  */
    112 
    113 #include "bpfilter.h"
    114 #include "vlan.h"
    115 
    116 #include <sys/param.h>
    117 #include <sys/endian.h>
    118 #include <sys/systm.h>
    119 #include <sys/sockio.h>
    120 #include <sys/mbuf.h>
    121 #include <sys/malloc.h>
    122 #include <sys/kernel.h>
    123 #include <sys/socket.h>
    124 #include <sys/device.h>
    125 
    126 #include <net/if.h>
    127 #include <net/if_arp.h>
    128 #include <net/if_dl.h>
    129 #include <net/if_ether.h>
    130 #include <net/if_media.h>
    131 #include <net/if_vlanvar.h>
    132 
    133 #include <netinet/in_systm.h>	/* XXX for IP_MAXPACKET */
    134 #include <netinet/in.h>		/* XXX for IP_MAXPACKET */
    135 #include <netinet/ip.h>		/* XXX for IP_MAXPACKET */
    136 
    137 #if NBPFILTER > 0
    138 #include <net/bpf.h>
    139 #endif
    140 
    141 #include <machine/bus.h>
    142 
    143 #include <dev/mii/mii.h>
    144 #include <dev/mii/miivar.h>
    145 
    146 #include <dev/ic/rtl81x9reg.h>
    147 #include <dev/ic/rtl81x9var.h>
    148 
    149 #include <dev/ic/rtl8169var.h>
    150 
    151 static inline void re_set_bufaddr(struct re_desc *, bus_addr_t);
    152 
    153 static int re_newbuf(struct rtk_softc *, int, struct mbuf *);
    154 static int re_rx_list_init(struct rtk_softc *);
    155 static int re_tx_list_init(struct rtk_softc *);
    156 static void re_rxeof(struct rtk_softc *);
    157 static void re_txeof(struct rtk_softc *);
    158 static void re_tick(void *);
    159 static void re_start(struct ifnet *);
    160 static int re_ioctl(struct ifnet *, u_long, caddr_t);
    161 static int re_init(struct ifnet *);
    162 static void re_stop(struct ifnet *, int);
    163 static void re_watchdog(struct ifnet *);
    164 
    165 static void re_shutdown(void *);
    166 static int re_enable(struct rtk_softc *);
    167 static void re_disable(struct rtk_softc *);
    168 static void re_power(int, void *);
    169 
    170 static int re_ifmedia_upd(struct ifnet *);
    171 static void re_ifmedia_sts(struct ifnet *, struct ifmediareq *);
    172 
    173 static int re_gmii_readreg(struct device *, int, int);
    174 static void re_gmii_writereg(struct device *, int, int, int);
    175 
    176 static int re_miibus_readreg(struct device *, int, int);
    177 static void re_miibus_writereg(struct device *, int, int, int);
    178 static void re_miibus_statchg(struct device *);
    179 
    180 static void re_reset(struct rtk_softc *);
    181 
    182 static inline void
    183 re_set_bufaddr(struct re_desc *d, bus_addr_t addr)
    184 {
    185 
    186 	d->re_bufaddr_lo = htole32((uint32_t)addr);
    187 	if (sizeof(bus_addr_t) == sizeof(uint64_t))
    188 		d->re_bufaddr_hi = htole32((uint64_t)addr >> 32);
    189 	else
    190 		d->re_bufaddr_hi = 0;
    191 }
    192 
    193 static int
    194 re_gmii_readreg(struct device *self, int phy, int reg)
    195 {
    196 	struct rtk_softc	*sc = (void *)self;
    197 	uint32_t		rval;
    198 	int			i;
    199 
    200 	if (phy != 7)
    201 		return 0;
    202 
    203 	/* Let the rgephy driver read the GMEDIASTAT register */
    204 
    205 	if (reg == RTK_GMEDIASTAT) {
    206 		rval = CSR_READ_1(sc, RTK_GMEDIASTAT);
    207 		return rval;
    208 	}
    209 
    210 	CSR_WRITE_4(sc, RTK_PHYAR, reg << 16);
    211 	DELAY(1000);
    212 
    213 	for (i = 0; i < RTK_TIMEOUT; i++) {
    214 		rval = CSR_READ_4(sc, RTK_PHYAR);
    215 		if (rval & RTK_PHYAR_BUSY)
    216 			break;
    217 		DELAY(100);
    218 	}
    219 
    220 	if (i == RTK_TIMEOUT) {
    221 		aprint_error("%s: PHY read failed\n", sc->sc_dev.dv_xname);
    222 		return 0;
    223 	}
    224 
    225 	return rval & RTK_PHYAR_PHYDATA;
    226 }
    227 
    228 static void
    229 re_gmii_writereg(struct device *dev, int phy, int reg, int data)
    230 {
    231 	struct rtk_softc	*sc = (void *)dev;
    232 	uint32_t		rval;
    233 	int			i;
    234 
    235 	CSR_WRITE_4(sc, RTK_PHYAR, (reg << 16) |
    236 	    (data & RTK_PHYAR_PHYDATA) | RTK_PHYAR_BUSY);
    237 	DELAY(1000);
    238 
    239 	for (i = 0; i < RTK_TIMEOUT; i++) {
    240 		rval = CSR_READ_4(sc, RTK_PHYAR);
    241 		if (!(rval & RTK_PHYAR_BUSY))
    242 			break;
    243 		DELAY(100);
    244 	}
    245 
    246 	if (i == RTK_TIMEOUT) {
    247 		aprint_error("%s: PHY write reg %x <- %x failed\n",
    248 		    sc->sc_dev.dv_xname, reg, data);
    249 	}
    250 }
    251 
    252 static int
    253 re_miibus_readreg(struct device *dev, int phy, int reg)
    254 {
    255 	struct rtk_softc	*sc = (void *)dev;
    256 	uint16_t		rval = 0;
    257 	uint16_t		re8139_reg = 0;
    258 	int			s;
    259 
    260 	s = splnet();
    261 
    262 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0) {
    263 		rval = re_gmii_readreg(dev, phy, reg);
    264 		splx(s);
    265 		return rval;
    266 	}
    267 
    268 	/* Pretend the internal PHY is only at address 0 */
    269 	if (phy) {
    270 		splx(s);
    271 		return 0;
    272 	}
    273 	switch (reg) {
    274 	case MII_BMCR:
    275 		re8139_reg = RTK_BMCR;
    276 		break;
    277 	case MII_BMSR:
    278 		re8139_reg = RTK_BMSR;
    279 		break;
    280 	case MII_ANAR:
    281 		re8139_reg = RTK_ANAR;
    282 		break;
    283 	case MII_ANER:
    284 		re8139_reg = RTK_ANER;
    285 		break;
    286 	case MII_ANLPAR:
    287 		re8139_reg = RTK_LPAR;
    288 		break;
    289 	case MII_PHYIDR1:
    290 	case MII_PHYIDR2:
    291 		splx(s);
    292 		return 0;
    293 	/*
    294 	 * Allow the rlphy driver to read the media status
    295 	 * register. If we have a link partner which does not
    296 	 * support NWAY, this is the register which will tell
    297 	 * us the results of parallel detection.
    298 	 */
    299 	case RTK_MEDIASTAT:
    300 		rval = CSR_READ_1(sc, RTK_MEDIASTAT);
    301 		splx(s);
    302 		return rval;
    303 	default:
    304 		aprint_error("%s: bad phy register\n", sc->sc_dev.dv_xname);
    305 		splx(s);
    306 		return 0;
    307 	}
    308 	rval = CSR_READ_2(sc, re8139_reg);
    309 	if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0 && re8139_reg == RTK_BMCR) {
    310 		/* 8139C+ has different bit layout. */
    311 		rval &= ~(BMCR_LOOP | BMCR_ISO);
    312 	}
    313 	splx(s);
    314 	return rval;
    315 }
    316 
    317 static void
    318 re_miibus_writereg(struct device *dev, int phy, int reg, int data)
    319 {
    320 	struct rtk_softc	*sc = (void *)dev;
    321 	uint16_t		re8139_reg = 0;
    322 	int			s;
    323 
    324 	s = splnet();
    325 
    326 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0) {
    327 		re_gmii_writereg(dev, phy, reg, data);
    328 		splx(s);
    329 		return;
    330 	}
    331 
    332 	/* Pretend the internal PHY is only at address 0 */
    333 	if (phy) {
    334 		splx(s);
    335 		return;
    336 	}
    337 	switch (reg) {
    338 	case MII_BMCR:
    339 		re8139_reg = RTK_BMCR;
    340 		if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0) {
    341 			/* 8139C+ has different bit layout. */
    342 			data &= ~(BMCR_LOOP | BMCR_ISO);
    343 		}
    344 		break;
    345 	case MII_BMSR:
    346 		re8139_reg = RTK_BMSR;
    347 		break;
    348 	case MII_ANAR:
    349 		re8139_reg = RTK_ANAR;
    350 		break;
    351 	case MII_ANER:
    352 		re8139_reg = RTK_ANER;
    353 		break;
    354 	case MII_ANLPAR:
    355 		re8139_reg = RTK_LPAR;
    356 		break;
    357 	case MII_PHYIDR1:
    358 	case MII_PHYIDR2:
    359 		splx(s);
    360 		return;
    361 		break;
    362 	default:
    363 		aprint_error("%s: bad phy register\n", sc->sc_dev.dv_xname);
    364 		splx(s);
    365 		return;
    366 	}
    367 	CSR_WRITE_2(sc, re8139_reg, data);
    368 	splx(s);
    369 	return;
    370 }
    371 
    372 static void
    373 re_miibus_statchg(struct device *dev)
    374 {
    375 
    376 	return;
    377 }
    378 
    379 static void
    380 re_reset(struct rtk_softc *sc)
    381 {
    382 	int		i;
    383 
    384 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_RESET);
    385 
    386 	for (i = 0; i < RTK_TIMEOUT; i++) {
    387 		DELAY(10);
    388 		if ((CSR_READ_1(sc, RTK_COMMAND) & RTK_CMD_RESET) == 0)
    389 			break;
    390 	}
    391 	if (i == RTK_TIMEOUT)
    392 		aprint_error("%s: reset never completed!\n",
    393 		    sc->sc_dev.dv_xname);
    394 
    395 	/*
    396 	 * NB: Realtek-supplied Linux driver does this only for
    397 	 * MCFG_METHOD_2, which corresponds to sc->sc_rev == 2.
    398 	 */
    399 	if (1) /* XXX check softc flag for 8169s version */
    400 		CSR_WRITE_1(sc, RTK_LDPS, 1);
    401 
    402 	return;
    403 }
    404 
    405 /*
    406  * The following routine is designed to test for a defect on some
    407  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
    408  * lines connected to the bus, however for a 32-bit only card, they
    409  * should be pulled high. The result of this defect is that the
    410  * NIC will not work right if you plug it into a 64-bit slot: DMA
    411  * operations will be done with 64-bit transfers, which will fail
    412  * because the 64-bit data lines aren't connected.
    413  *
    414  * There's no way to work around this (short of talking a soldering
    415  * iron to the board), however we can detect it. The method we use
    416  * here is to put the NIC into digital loopback mode, set the receiver
    417  * to promiscuous mode, and then try to send a frame. We then compare
    418  * the frame data we sent to what was received. If the data matches,
    419  * then the NIC is working correctly, otherwise we know the user has
    420  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
    421  * slot. In the latter case, there's no way the NIC can work correctly,
    422  * so we print out a message on the console and abort the device attach.
    423  */
    424 
    425 int
    426 re_diag(struct rtk_softc *sc)
    427 {
    428 	struct ifnet		*ifp = &sc->ethercom.ec_if;
    429 	struct mbuf		*m0;
    430 	struct ether_header	*eh;
    431 	struct re_rxsoft	*rxs;
    432 	struct re_desc		*cur_rx;
    433 	bus_dmamap_t		dmamap;
    434 	uint16_t		status;
    435 	uint32_t		rxstat;
    436 	int			total_len, i, s, error = 0;
    437 	static const uint8_t	dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
    438 	static const uint8_t	src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
    439 
    440 	/* Allocate a single mbuf */
    441 
    442 	MGETHDR(m0, M_DONTWAIT, MT_DATA);
    443 	if (m0 == NULL)
    444 		return ENOBUFS;
    445 
    446 	/*
    447 	 * Initialize the NIC in test mode. This sets the chip up
    448 	 * so that it can send and receive frames, but performs the
    449 	 * following special functions:
    450 	 * - Puts receiver in promiscuous mode
    451 	 * - Enables digital loopback mode
    452 	 * - Leaves interrupts turned off
    453 	 */
    454 
    455 	ifp->if_flags |= IFF_PROMISC;
    456 	sc->re_testmode = 1;
    457 	re_init(ifp);
    458 	re_stop(ifp, 0);
    459 	DELAY(100000);
    460 	re_init(ifp);
    461 
    462 	/* Put some data in the mbuf */
    463 
    464 	eh = mtod(m0, struct ether_header *);
    465 	memcpy(eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN);
    466 	memcpy(eh->ether_shost, (char *)&src, ETHER_ADDR_LEN);
    467 	eh->ether_type = htons(ETHERTYPE_IP);
    468 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
    469 
    470 	/*
    471 	 * Queue the packet, start transmission.
    472 	 */
    473 
    474 	CSR_WRITE_2(sc, RTK_ISR, 0xFFFF);
    475 	s = splnet();
    476 	IF_ENQUEUE(&ifp->if_snd, m0);
    477 	re_start(ifp);
    478 	splx(s);
    479 	m0 = NULL;
    480 
    481 	/* Wait for it to propagate through the chip */
    482 
    483 	DELAY(100000);
    484 	for (i = 0; i < RTK_TIMEOUT; i++) {
    485 		status = CSR_READ_2(sc, RTK_ISR);
    486 		if ((status & (RTK_ISR_TIMEOUT_EXPIRED | RTK_ISR_RX_OK)) ==
    487 		    (RTK_ISR_TIMEOUT_EXPIRED | RTK_ISR_RX_OK))
    488 			break;
    489 		DELAY(10);
    490 	}
    491 	if (i == RTK_TIMEOUT) {
    492 		aprint_error("%s: diagnostic failed, failed to receive packet "
    493 		    "in loopback mode\n", sc->sc_dev.dv_xname);
    494 		error = EIO;
    495 		goto done;
    496 	}
    497 
    498 	/*
    499 	 * The packet should have been dumped into the first
    500 	 * entry in the RX DMA ring. Grab it from there.
    501 	 */
    502 
    503 	rxs = &sc->re_ldata.re_rxsoft[0];
    504 	dmamap = rxs->rxs_dmamap;
    505 	bus_dmamap_sync(sc->sc_dmat, dmamap, 0, dmamap->dm_mapsize,
    506 	    BUS_DMASYNC_POSTREAD);
    507 	bus_dmamap_unload(sc->sc_dmat, dmamap);
    508 
    509 	m0 = rxs->rxs_mbuf;
    510 	rxs->rxs_mbuf = NULL;
    511 	eh = mtod(m0, struct ether_header *);
    512 
    513 	RE_RXDESCSYNC(sc, 0, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
    514 	cur_rx = &sc->re_ldata.re_rx_list[0];
    515 	rxstat = le32toh(cur_rx->re_cmdstat);
    516 	total_len = rxstat & sc->re_rxlenmask;
    517 
    518 	if (total_len != ETHER_MIN_LEN) {
    519 		aprint_error("%s: diagnostic failed, received short packet\n",
    520 		    sc->sc_dev.dv_xname);
    521 		error = EIO;
    522 		goto done;
    523 	}
    524 
    525 	/* Test that the received packet data matches what we sent. */
    526 
    527 	if (memcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
    528 	    memcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
    529 	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
    530 		aprint_error("%s: WARNING, DMA FAILURE!\n",
    531 		    sc->sc_dev.dv_xname);
    532 		aprint_error("%s: expected TX data: %s",
    533 		    sc->sc_dev.dv_xname, ether_sprintf(dst));
    534 		aprint_error("/%s/0x%x\n", ether_sprintf(src), ETHERTYPE_IP);
    535 		aprint_error("%s: received RX data: %s",
    536 		    sc->sc_dev.dv_xname,
    537 		    ether_sprintf(eh->ether_dhost));
    538 		aprint_error("/%s/0x%x\n", ether_sprintf(eh->ether_shost),
    539 		    ntohs(eh->ether_type));
    540 		aprint_error("%s: You may have a defective 32-bit NIC plugged "
    541 		    "into a 64-bit PCI slot.\n", sc->sc_dev.dv_xname);
    542 		aprint_error("%s: Please re-install the NIC in a 32-bit slot "
    543 		    "for proper operation.\n", sc->sc_dev.dv_xname);
    544 		aprint_error("%s: Read the re(4) man page for more details.\n",
    545 		    sc->sc_dev.dv_xname);
    546 		error = EIO;
    547 	}
    548 
    549  done:
    550 	/* Turn interface off, release resources */
    551 
    552 	sc->re_testmode = 0;
    553 	ifp->if_flags &= ~IFF_PROMISC;
    554 	re_stop(ifp, 0);
    555 	if (m0 != NULL)
    556 		m_freem(m0);
    557 
    558 	return error;
    559 }
    560 
    561 
    562 /*
    563  * Attach the interface. Allocate softc structures, do ifmedia
    564  * setup and ethernet/BPF attach.
    565  */
    566 void
    567 re_attach(struct rtk_softc *sc)
    568 {
    569 	u_char			eaddr[ETHER_ADDR_LEN];
    570 	uint16_t		val;
    571 	struct ifnet		*ifp;
    572 	int			error = 0, i, addr_len;
    573 
    574 	/* Reset the adapter. */
    575 	re_reset(sc);
    576 
    577 	if (rtk_read_eeprom(sc, RTK_EE_ID, RTK_EEADDR_LEN1) == 0x8129)
    578 		addr_len = RTK_EEADDR_LEN1;
    579 	else
    580 		addr_len = RTK_EEADDR_LEN0;
    581 
    582 	/*
    583 	 * Get station address from the EEPROM.
    584 	 */
    585 	for (i = 0; i < 3; i++) {
    586 		val = rtk_read_eeprom(sc, RTK_EE_EADDR0 + i, addr_len);
    587 		eaddr[(i * 2) + 0] = val & 0xff;
    588 		eaddr[(i * 2) + 1] = val >> 8;
    589 	}
    590 
    591 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0) {
    592 		uint32_t hwrev;
    593 
    594 		/* Revision of 8169/8169S/8110s in bits 30..26, 23 */
    595 		hwrev = CSR_READ_4(sc, RTK_TXCFG) & RTK_TXCFG_HWREV;
    596 		/* These rev numbers are taken from Realtek's driver */
    597 		if (       hwrev == RTK_HWREV_8100E_SPIN2) {
    598 			sc->sc_rev = 15;
    599 		} else if (hwrev == RTK_HWREV_8100E) {
    600 			sc->sc_rev = 14;
    601 		} else if (hwrev == RTK_HWREV_8101E) {
    602 			sc->sc_rev = 13;
    603 		} else if (hwrev == RTK_HWREV_8168_SPIN2 ||
    604 		           hwrev == RTK_HWREV_8168_SPIN3) {
    605 			sc->sc_rev = 12;
    606 		} else if (hwrev == RTK_HWREV_8168_SPIN1) {
    607 			sc->sc_rev = 11;
    608 		} else if (hwrev == RTK_HWREV_8169_8110SC) {
    609 			sc->sc_rev = 5;
    610 		} else if (hwrev == RTK_HWREV_8169_8110SB) {
    611 			sc->sc_rev = 4;
    612 		} else if (hwrev == RTK_HWREV_8169S) {
    613 			sc->sc_rev = 3;
    614 		} else if (hwrev == RTK_HWREV_8110S) {
    615 			sc->sc_rev = 2;
    616 		} else if (hwrev == RTK_HWREV_8169) {
    617 			sc->sc_rev = 1;
    618 			sc->sc_quirk |= RTKQ_8169NONS;
    619 		} else {
    620 			aprint_normal("%s: Unknown revision (0x%08x)\n",
    621 			    sc->sc_dev.dv_xname, hwrev);
    622 			/* assume the latest one */
    623 			sc->sc_rev = 15;
    624 		}
    625 
    626 		/* Set RX length mask */
    627 		sc->re_rxlenmask = RE_RDESC_STAT_GFRAGLEN;
    628 		sc->re_ldata.re_tx_desc_cnt = RE_TX_DESC_CNT_8169;
    629 	} else {
    630 		/* Set RX length mask */
    631 		sc->re_rxlenmask = RE_RDESC_STAT_FRAGLEN;
    632 		sc->re_ldata.re_tx_desc_cnt = RE_TX_DESC_CNT_8139;
    633 	}
    634 
    635 	aprint_normal("%s: Ethernet address %s\n",
    636 	    sc->sc_dev.dv_xname, ether_sprintf(eaddr));
    637 
    638 	if (sc->re_ldata.re_tx_desc_cnt >
    639 	    PAGE_SIZE / sizeof(struct re_desc)) {
    640 		sc->re_ldata.re_tx_desc_cnt =
    641 		    PAGE_SIZE / sizeof(struct re_desc);
    642 	}
    643 
    644 	aprint_verbose("%s: using %d tx descriptors\n",
    645 	    sc->sc_dev.dv_xname, sc->re_ldata.re_tx_desc_cnt);
    646 	KASSERT(RE_NEXT_TX_DESC(sc, RE_TX_DESC_CNT(sc) - 1) == 0);
    647 
    648 	/* Allocate DMA'able memory for the TX ring */
    649 	if ((error = bus_dmamem_alloc(sc->sc_dmat, RE_TX_LIST_SZ(sc),
    650 	    RE_RING_ALIGN, 0, &sc->re_ldata.re_tx_listseg, 1,
    651 	    &sc->re_ldata.re_tx_listnseg, BUS_DMA_NOWAIT)) != 0) {
    652 		aprint_error("%s: can't allocate tx listseg, error = %d\n",
    653 		    sc->sc_dev.dv_xname, error);
    654 		goto fail_0;
    655 	}
    656 
    657 	/* Load the map for the TX ring. */
    658 	if ((error = bus_dmamem_map(sc->sc_dmat, &sc->re_ldata.re_tx_listseg,
    659 	    sc->re_ldata.re_tx_listnseg, RE_TX_LIST_SZ(sc),
    660 	    (caddr_t *)&sc->re_ldata.re_tx_list,
    661 	    BUS_DMA_COHERENT | BUS_DMA_NOWAIT)) != 0) {
    662 		aprint_error("%s: can't map tx list, error = %d\n",
    663 		    sc->sc_dev.dv_xname, error);
    664 	  	goto fail_1;
    665 	}
    666 	memset(sc->re_ldata.re_tx_list, 0, RE_TX_LIST_SZ(sc));
    667 
    668 	if ((error = bus_dmamap_create(sc->sc_dmat, RE_TX_LIST_SZ(sc), 1,
    669 	    RE_TX_LIST_SZ(sc), 0, 0,
    670 	    &sc->re_ldata.re_tx_list_map)) != 0) {
    671 		aprint_error("%s: can't create tx list map, error = %d\n",
    672 		    sc->sc_dev.dv_xname, error);
    673 		goto fail_2;
    674 	}
    675 
    676 
    677 	if ((error = bus_dmamap_load(sc->sc_dmat,
    678 	    sc->re_ldata.re_tx_list_map, sc->re_ldata.re_tx_list,
    679 	    RE_TX_LIST_SZ(sc), NULL, BUS_DMA_NOWAIT)) != 0) {
    680 		aprint_error("%s: can't load tx list, error = %d\n",
    681 		    sc->sc_dev.dv_xname, error);
    682 		goto fail_3;
    683 	}
    684 
    685 	/* Create DMA maps for TX buffers */
    686 	for (i = 0; i < RE_TX_QLEN; i++) {
    687 		error = bus_dmamap_create(sc->sc_dmat,
    688 		    round_page(IP_MAXPACKET),
    689 		    RE_TX_DESC_CNT(sc) - RE_NTXDESC_RSVD, RE_TDESC_CMD_FRAGLEN,
    690 		    0, 0, &sc->re_ldata.re_txq[i].txq_dmamap);
    691 		if (error) {
    692 			aprint_error("%s: can't create DMA map for TX\n",
    693 			    sc->sc_dev.dv_xname);
    694 			goto fail_4;
    695 		}
    696 	}
    697 
    698 	/* Allocate DMA'able memory for the RX ring */
    699 	/* XXX see also a comment about RE_RX_DMAMEM_SZ in rtl81x9var.h */
    700 	if ((error = bus_dmamem_alloc(sc->sc_dmat,
    701 	    RE_RX_DMAMEM_SZ, RE_RING_ALIGN, 0, &sc->re_ldata.re_rx_listseg, 1,
    702 	    &sc->re_ldata.re_rx_listnseg, BUS_DMA_NOWAIT)) != 0) {
    703 		aprint_error("%s: can't allocate rx listseg, error = %d\n",
    704 		    sc->sc_dev.dv_xname, error);
    705 		goto fail_4;
    706 	}
    707 
    708 	/* Load the map for the RX ring. */
    709 	if ((error = bus_dmamem_map(sc->sc_dmat, &sc->re_ldata.re_rx_listseg,
    710 	    sc->re_ldata.re_rx_listnseg, RE_RX_DMAMEM_SZ,
    711 	    (caddr_t *)&sc->re_ldata.re_rx_list,
    712 	    BUS_DMA_COHERENT | BUS_DMA_NOWAIT)) != 0) {
    713 		aprint_error("%s: can't map rx list, error = %d\n",
    714 		    sc->sc_dev.dv_xname, error);
    715 		goto fail_5;
    716 	}
    717 	memset(sc->re_ldata.re_rx_list, 0, RE_RX_DMAMEM_SZ);
    718 
    719 	if ((error = bus_dmamap_create(sc->sc_dmat,
    720 	    RE_RX_DMAMEM_SZ, 1, RE_RX_DMAMEM_SZ, 0, 0,
    721 	    &sc->re_ldata.re_rx_list_map)) != 0) {
    722 		aprint_error("%s: can't create rx list map, error = %d\n",
    723 		    sc->sc_dev.dv_xname, error);
    724 		goto fail_6;
    725 	}
    726 
    727 	if ((error = bus_dmamap_load(sc->sc_dmat,
    728 	    sc->re_ldata.re_rx_list_map, sc->re_ldata.re_rx_list,
    729 	    RE_RX_DMAMEM_SZ, NULL, BUS_DMA_NOWAIT)) != 0) {
    730 		aprint_error("%s: can't load rx list, error = %d\n",
    731 		    sc->sc_dev.dv_xname, error);
    732 		goto fail_7;
    733 	}
    734 
    735 	/* Create DMA maps for RX buffers */
    736 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
    737 		error = bus_dmamap_create(sc->sc_dmat, MCLBYTES, 1, MCLBYTES,
    738 		    0, 0, &sc->re_ldata.re_rxsoft[i].rxs_dmamap);
    739 		if (error) {
    740 			aprint_error("%s: can't create DMA map for RX\n",
    741 			    sc->sc_dev.dv_xname);
    742 			goto fail_8;
    743 		}
    744 	}
    745 
    746 	/*
    747 	 * Record interface as attached. From here, we should not fail.
    748 	 */
    749 	sc->sc_flags |= RTK_ATTACHED;
    750 
    751 	ifp = &sc->ethercom.ec_if;
    752 	ifp->if_softc = sc;
    753 	strcpy(ifp->if_xname, sc->sc_dev.dv_xname);
    754 	ifp->if_mtu = ETHERMTU;
    755 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
    756 	ifp->if_ioctl = re_ioctl;
    757 	sc->ethercom.ec_capabilities |=
    758 	    ETHERCAP_VLAN_MTU | ETHERCAP_VLAN_HWTAGGING;
    759 	ifp->if_start = re_start;
    760 	ifp->if_stop = re_stop;
    761 
    762 	/*
    763 	 * IFCAP_CSUM_IPv4_Tx on re(4) is broken for small packets,
    764 	 * so we have a workaround to handle the bug by padding
    765 	 * such packets manually.
    766 	 */
    767 	ifp->if_capabilities |=
    768 	    IFCAP_CSUM_IPv4 |
    769 	    IFCAP_CSUM_TCPv4 |
    770 	    IFCAP_CSUM_UDPv4 |
    771 	    IFCAP_TSOv4;
    772 	ifp->if_watchdog = re_watchdog;
    773 	ifp->if_init = re_init;
    774 	ifp->if_snd.ifq_maxlen = RE_IFQ_MAXLEN;
    775 	ifp->if_capenable = ifp->if_capabilities;
    776 	IFQ_SET_READY(&ifp->if_snd);
    777 
    778 	callout_init(&sc->rtk_tick_ch);
    779 
    780 	/* Do MII setup */
    781 	sc->mii.mii_ifp = ifp;
    782 	sc->mii.mii_readreg = re_miibus_readreg;
    783 	sc->mii.mii_writereg = re_miibus_writereg;
    784 	sc->mii.mii_statchg = re_miibus_statchg;
    785 	ifmedia_init(&sc->mii.mii_media, IFM_IMASK, re_ifmedia_upd,
    786 	    re_ifmedia_sts);
    787 	mii_attach(&sc->sc_dev, &sc->mii, 0xffffffff, MII_PHY_ANY,
    788 	    MII_OFFSET_ANY, 0);
    789 	ifmedia_set(&sc->mii.mii_media, IFM_ETHER | IFM_AUTO);
    790 
    791 	/*
    792 	 * Call MI attach routine.
    793 	 */
    794 	if_attach(ifp);
    795 	ether_ifattach(ifp, eaddr);
    796 
    797 
    798 	/*
    799 	 * Make sure the interface is shutdown during reboot.
    800 	 */
    801 	sc->sc_sdhook = shutdownhook_establish(re_shutdown, sc);
    802 	if (sc->sc_sdhook == NULL)
    803 		aprint_error("%s: WARNING: unable to establish shutdown hook\n",
    804 		    sc->sc_dev.dv_xname);
    805 	/*
    806 	 * Add a suspend hook to make sure we come back up after a
    807 	 * resume.
    808 	 */
    809 	sc->sc_powerhook = powerhook_establish(re_power, sc);
    810 	if (sc->sc_powerhook == NULL)
    811 		aprint_error("%s: WARNING: unable to establish power hook\n",
    812 		    sc->sc_dev.dv_xname);
    813 
    814 
    815 	return;
    816 
    817  fail_8:
    818 	/* Destroy DMA maps for RX buffers. */
    819 	for (i = 0; i < RE_RX_DESC_CNT; i++)
    820 		if (sc->re_ldata.re_rxsoft[i].rxs_dmamap != NULL)
    821 			bus_dmamap_destroy(sc->sc_dmat,
    822 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
    823 
    824 	/* Free DMA'able memory for the RX ring. */
    825 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
    826  fail_7:
    827 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
    828  fail_6:
    829 	bus_dmamem_unmap(sc->sc_dmat,
    830 	    (caddr_t)sc->re_ldata.re_rx_list, RE_RX_DMAMEM_SZ);
    831  fail_5:
    832 	bus_dmamem_free(sc->sc_dmat,
    833 	    &sc->re_ldata.re_rx_listseg, sc->re_ldata.re_rx_listnseg);
    834 
    835  fail_4:
    836 	/* Destroy DMA maps for TX buffers. */
    837 	for (i = 0; i < RE_TX_QLEN; i++)
    838 		if (sc->re_ldata.re_txq[i].txq_dmamap != NULL)
    839 			bus_dmamap_destroy(sc->sc_dmat,
    840 			    sc->re_ldata.re_txq[i].txq_dmamap);
    841 
    842 	/* Free DMA'able memory for the TX ring. */
    843 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
    844  fail_3:
    845 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
    846  fail_2:
    847 	bus_dmamem_unmap(sc->sc_dmat,
    848 	    (caddr_t)sc->re_ldata.re_tx_list, RE_TX_LIST_SZ(sc));
    849  fail_1:
    850 	bus_dmamem_free(sc->sc_dmat,
    851 	    &sc->re_ldata.re_tx_listseg, sc->re_ldata.re_tx_listnseg);
    852  fail_0:
    853 	return;
    854 }
    855 
    856 
    857 /*
    858  * re_activate:
    859  *     Handle device activation/deactivation requests.
    860  */
    861 int
    862 re_activate(struct device *self, enum devact act)
    863 {
    864 	struct rtk_softc *sc = (void *)self;
    865 	int s, error = 0;
    866 
    867 	s = splnet();
    868 	switch (act) {
    869 	case DVACT_ACTIVATE:
    870 		error = EOPNOTSUPP;
    871 		break;
    872 	case DVACT_DEACTIVATE:
    873 		mii_activate(&sc->mii, act, MII_PHY_ANY, MII_OFFSET_ANY);
    874 		if_deactivate(&sc->ethercom.ec_if);
    875 		break;
    876 	}
    877 	splx(s);
    878 
    879 	return error;
    880 }
    881 
    882 /*
    883  * re_detach:
    884  *     Detach a rtk interface.
    885  */
    886 int
    887 re_detach(struct rtk_softc *sc)
    888 {
    889 	struct ifnet *ifp = &sc->ethercom.ec_if;
    890 	int i;
    891 
    892 	/*
    893 	 * Succeed now if there isn't any work to do.
    894 	 */
    895 	if ((sc->sc_flags & RTK_ATTACHED) == 0)
    896 		return 0;
    897 
    898 	/* Unhook our tick handler. */
    899 	callout_stop(&sc->rtk_tick_ch);
    900 
    901 	/* Detach all PHYs. */
    902 	mii_detach(&sc->mii, MII_PHY_ANY, MII_OFFSET_ANY);
    903 
    904 	/* Delete all remaining media. */
    905 	ifmedia_delete_instance(&sc->mii.mii_media, IFM_INST_ANY);
    906 
    907 	ether_ifdetach(ifp);
    908 	if_detach(ifp);
    909 
    910 	/* Destroy DMA maps for RX buffers. */
    911 	for (i = 0; i < RE_RX_DESC_CNT; i++)
    912 		if (sc->re_ldata.re_rxsoft[i].rxs_dmamap != NULL)
    913 			bus_dmamap_destroy(sc->sc_dmat,
    914 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
    915 
    916 	/* Free DMA'able memory for the RX ring. */
    917 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
    918 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
    919 	bus_dmamem_unmap(sc->sc_dmat,
    920 	    (caddr_t)sc->re_ldata.re_rx_list, RE_RX_DMAMEM_SZ);
    921 	bus_dmamem_free(sc->sc_dmat,
    922 	    &sc->re_ldata.re_rx_listseg, sc->re_ldata.re_rx_listnseg);
    923 
    924 	/* Destroy DMA maps for TX buffers. */
    925 	for (i = 0; i < RE_TX_QLEN; i++)
    926 		if (sc->re_ldata.re_txq[i].txq_dmamap != NULL)
    927 			bus_dmamap_destroy(sc->sc_dmat,
    928 			    sc->re_ldata.re_txq[i].txq_dmamap);
    929 
    930 	/* Free DMA'able memory for the TX ring. */
    931 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
    932 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
    933 	bus_dmamem_unmap(sc->sc_dmat,
    934 	    (caddr_t)sc->re_ldata.re_tx_list, RE_TX_LIST_SZ(sc));
    935 	bus_dmamem_free(sc->sc_dmat,
    936 	    &sc->re_ldata.re_tx_listseg, sc->re_ldata.re_tx_listnseg);
    937 
    938 
    939 	shutdownhook_disestablish(sc->sc_sdhook);
    940 	powerhook_disestablish(sc->sc_powerhook);
    941 
    942 	return 0;
    943 }
    944 
    945 /*
    946  * re_enable:
    947  *     Enable the RTL81X9 chip.
    948  */
    949 static int
    950 re_enable(struct rtk_softc *sc)
    951 {
    952 
    953 	if (RTK_IS_ENABLED(sc) == 0 && sc->sc_enable != NULL) {
    954 		if ((*sc->sc_enable)(sc) != 0) {
    955 			aprint_error("%s: device enable failed\n",
    956 			    sc->sc_dev.dv_xname);
    957 			return EIO;
    958 		}
    959 		sc->sc_flags |= RTK_ENABLED;
    960 	}
    961 	return 0;
    962 }
    963 
    964 /*
    965  * re_disable:
    966  *     Disable the RTL81X9 chip.
    967  */
    968 static void
    969 re_disable(struct rtk_softc *sc)
    970 {
    971 
    972 	if (RTK_IS_ENABLED(sc) && sc->sc_disable != NULL) {
    973 		(*sc->sc_disable)(sc);
    974 		sc->sc_flags &= ~RTK_ENABLED;
    975 	}
    976 }
    977 
    978 /*
    979  * re_power:
    980  *     Power management (suspend/resume) hook.
    981  */
    982 void
    983 re_power(int why, void *arg)
    984 {
    985 	struct rtk_softc *sc = (void *)arg;
    986 	struct ifnet *ifp = &sc->ethercom.ec_if;
    987 	int s;
    988 
    989 	s = splnet();
    990 	switch (why) {
    991 	case PWR_SUSPEND:
    992 	case PWR_STANDBY:
    993 		re_stop(ifp, 0);
    994 		if (sc->sc_power != NULL)
    995 			(*sc->sc_power)(sc, why);
    996 		break;
    997 	case PWR_RESUME:
    998 		if (ifp->if_flags & IFF_UP) {
    999 			if (sc->sc_power != NULL)
   1000 				(*sc->sc_power)(sc, why);
   1001 			re_init(ifp);
   1002 		}
   1003 		break;
   1004 	case PWR_SOFTSUSPEND:
   1005 	case PWR_SOFTSTANDBY:
   1006 	case PWR_SOFTRESUME:
   1007 		break;
   1008 	}
   1009 	splx(s);
   1010 }
   1011 
   1012 
   1013 static int
   1014 re_newbuf(struct rtk_softc *sc, int idx, struct mbuf *m)
   1015 {
   1016 	struct mbuf		*n = NULL;
   1017 	bus_dmamap_t		map;
   1018 	struct re_desc		*d;
   1019 	struct re_rxsoft	*rxs;
   1020 	uint32_t		cmdstat;
   1021 	int			error;
   1022 
   1023 	if (m == NULL) {
   1024 		MGETHDR(n, M_DONTWAIT, MT_DATA);
   1025 		if (n == NULL)
   1026 			return ENOBUFS;
   1027 
   1028 		MCLGET(n, M_DONTWAIT);
   1029 		if ((n->m_flags & M_EXT) == 0) {
   1030 			m_freem(n);
   1031 			return ENOBUFS;
   1032 		}
   1033 		m = n;
   1034 	} else
   1035 		m->m_data = m->m_ext.ext_buf;
   1036 
   1037 	/*
   1038 	 * Initialize mbuf length fields and fixup
   1039 	 * alignment so that the frame payload is
   1040 	 * longword aligned.
   1041 	 */
   1042 	m->m_len = m->m_pkthdr.len = MCLBYTES - RE_ETHER_ALIGN;
   1043 	m->m_data += RE_ETHER_ALIGN;
   1044 
   1045 	rxs = &sc->re_ldata.re_rxsoft[idx];
   1046 	map = rxs->rxs_dmamap;
   1047 	error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m,
   1048 	    BUS_DMA_READ|BUS_DMA_NOWAIT);
   1049 
   1050 	if (error)
   1051 		goto out;
   1052 
   1053 	bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
   1054 	    BUS_DMASYNC_PREREAD);
   1055 
   1056 	d = &sc->re_ldata.re_rx_list[idx];
   1057 #ifdef DIAGNOSTIC
   1058 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
   1059 	cmdstat = le32toh(d->re_cmdstat);
   1060 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD);
   1061 	if (cmdstat & RE_RDESC_STAT_OWN) {
   1062 		panic("%s: tried to map busy RX descriptor",
   1063 		    sc->sc_dev.dv_xname);
   1064 	}
   1065 #endif
   1066 
   1067 	rxs->rxs_mbuf = m;
   1068 
   1069 	d->re_vlanctl = 0;
   1070 	cmdstat = map->dm_segs[0].ds_len;
   1071 	if (idx == (RE_RX_DESC_CNT - 1))
   1072 		cmdstat |= RE_RDESC_CMD_EOR;
   1073 	re_set_bufaddr(d, map->dm_segs[0].ds_addr);
   1074 	d->re_cmdstat = htole32(cmdstat);
   1075 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1076 	cmdstat |= RE_RDESC_CMD_OWN;
   1077 	d->re_cmdstat = htole32(cmdstat);
   1078 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1079 
   1080 	return 0;
   1081  out:
   1082 	if (n != NULL)
   1083 		m_freem(n);
   1084 	return ENOMEM;
   1085 }
   1086 
   1087 static int
   1088 re_tx_list_init(struct rtk_softc *sc)
   1089 {
   1090 	int i;
   1091 
   1092 	memset(sc->re_ldata.re_tx_list, 0, RE_TX_LIST_SZ(sc));
   1093 	for (i = 0; i < RE_TX_QLEN; i++) {
   1094 		sc->re_ldata.re_txq[i].txq_mbuf = NULL;
   1095 	}
   1096 
   1097 	bus_dmamap_sync(sc->sc_dmat,
   1098 	    sc->re_ldata.re_tx_list_map, 0,
   1099 	    sc->re_ldata.re_tx_list_map->dm_mapsize,
   1100 	    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1101 	sc->re_ldata.re_txq_prodidx = 0;
   1102 	sc->re_ldata.re_txq_considx = 0;
   1103 	sc->re_ldata.re_txq_free = RE_TX_QLEN;
   1104 	sc->re_ldata.re_tx_free = RE_TX_DESC_CNT(sc);
   1105 	sc->re_ldata.re_tx_nextfree = 0;
   1106 
   1107 	return 0;
   1108 }
   1109 
   1110 static int
   1111 re_rx_list_init(struct rtk_softc *sc)
   1112 {
   1113 	int			i;
   1114 
   1115 	memset((char *)sc->re_ldata.re_rx_list, 0, RE_RX_LIST_SZ);
   1116 
   1117 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
   1118 		if (re_newbuf(sc, i, NULL) == ENOBUFS)
   1119 			return ENOBUFS;
   1120 	}
   1121 
   1122 	sc->re_ldata.re_rx_prodidx = 0;
   1123 	sc->re_head = sc->re_tail = NULL;
   1124 
   1125 	return 0;
   1126 }
   1127 
   1128 /*
   1129  * RX handler for C+ and 8169. For the gigE chips, we support
   1130  * the reception of jumbo frames that have been fragmented
   1131  * across multiple 2K mbuf cluster buffers.
   1132  */
   1133 static void
   1134 re_rxeof(struct rtk_softc *sc)
   1135 {
   1136 	struct mbuf		*m;
   1137 	struct ifnet		*ifp;
   1138 	int			i, total_len;
   1139 	struct re_desc		*cur_rx;
   1140 	struct re_rxsoft	*rxs;
   1141 	uint32_t		rxstat, rxvlan;
   1142 
   1143 	ifp = &sc->ethercom.ec_if;
   1144 
   1145 	for (i = sc->re_ldata.re_rx_prodidx;; i = RE_NEXT_RX_DESC(sc, i)) {
   1146 		cur_rx = &sc->re_ldata.re_rx_list[i];
   1147 		RE_RXDESCSYNC(sc, i,
   1148 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
   1149 		rxstat = le32toh(cur_rx->re_cmdstat);
   1150 		RE_RXDESCSYNC(sc, i, BUS_DMASYNC_PREREAD);
   1151 		if ((rxstat & RE_RDESC_STAT_OWN) != 0) {
   1152 			break;
   1153 		}
   1154 		total_len = rxstat & sc->re_rxlenmask;
   1155 		rxvlan = le32toh(cur_rx->re_vlanctl);
   1156 		rxs = &sc->re_ldata.re_rxsoft[i];
   1157 		m = rxs->rxs_mbuf;
   1158 
   1159 		/* Invalidate the RX mbuf and unload its map */
   1160 
   1161 		bus_dmamap_sync(sc->sc_dmat,
   1162 		    rxs->rxs_dmamap, 0, rxs->rxs_dmamap->dm_mapsize,
   1163 		    BUS_DMASYNC_POSTREAD);
   1164 		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
   1165 
   1166 		if ((rxstat & RE_RDESC_STAT_EOF) == 0) {
   1167 			m->m_len = MCLBYTES - RE_ETHER_ALIGN;
   1168 			if (sc->re_head == NULL)
   1169 				sc->re_head = sc->re_tail = m;
   1170 			else {
   1171 				m->m_flags &= ~M_PKTHDR;
   1172 				sc->re_tail->m_next = m;
   1173 				sc->re_tail = m;
   1174 			}
   1175 			re_newbuf(sc, i, NULL);
   1176 			continue;
   1177 		}
   1178 
   1179 		/*
   1180 		 * NOTE: for the 8139C+, the frame length field
   1181 		 * is always 12 bits in size, but for the gigE chips,
   1182 		 * it is 13 bits (since the max RX frame length is 16K).
   1183 		 * Unfortunately, all 32 bits in the status word
   1184 		 * were already used, so to make room for the extra
   1185 		 * length bit, RealTek took out the 'frame alignment
   1186 		 * error' bit and shifted the other status bits
   1187 		 * over one slot. The OWN, EOR, FS and LS bits are
   1188 		 * still in the same places. We have already extracted
   1189 		 * the frame length and checked the OWN bit, so rather
   1190 		 * than using an alternate bit mapping, we shift the
   1191 		 * status bits one space to the right so we can evaluate
   1192 		 * them using the 8169 status as though it was in the
   1193 		 * same format as that of the 8139C+.
   1194 		 */
   1195 		if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0)
   1196 			rxstat >>= 1;
   1197 
   1198 		if (__predict_false((rxstat & RE_RDESC_STAT_RXERRSUM) != 0)) {
   1199 #ifdef RE_DEBUG
   1200 			aprint_error("%s: RX error (rxstat = 0x%08x)",
   1201 			    sc->sc_dev.dv_xname, rxstat);
   1202 			if (rxstat & RE_RDESC_STAT_FRALIGN)
   1203 				aprint_error(", frame alignment error");
   1204 			if (rxstat & RE_RDESC_STAT_BUFOFLOW)
   1205 				aprint_error(", out of buffer space");
   1206 			if (rxstat & RE_RDESC_STAT_FIFOOFLOW)
   1207 				aprint_error(", FIFO overrun");
   1208 			if (rxstat & RE_RDESC_STAT_GIANT)
   1209 				aprint_error(", giant packet");
   1210 			if (rxstat & RE_RDESC_STAT_RUNT)
   1211 				aprint_error(", runt packet");
   1212 			if (rxstat & RE_RDESC_STAT_CRCERR)
   1213 				aprint_error(", CRC error");
   1214 			aprint_error("\n");
   1215 #endif
   1216 			ifp->if_ierrors++;
   1217 			/*
   1218 			 * If this is part of a multi-fragment packet,
   1219 			 * discard all the pieces.
   1220 			 */
   1221 			if (sc->re_head != NULL) {
   1222 				m_freem(sc->re_head);
   1223 				sc->re_head = sc->re_tail = NULL;
   1224 			}
   1225 			re_newbuf(sc, i, m);
   1226 			continue;
   1227 		}
   1228 
   1229 		/*
   1230 		 * If allocating a replacement mbuf fails,
   1231 		 * reload the current one.
   1232 		 */
   1233 
   1234 		if (__predict_false(re_newbuf(sc, i, NULL) != 0)) {
   1235 			ifp->if_ierrors++;
   1236 			if (sc->re_head != NULL) {
   1237 				m_freem(sc->re_head);
   1238 				sc->re_head = sc->re_tail = NULL;
   1239 			}
   1240 			re_newbuf(sc, i, m);
   1241 			continue;
   1242 		}
   1243 
   1244 		if (sc->re_head != NULL) {
   1245 			m->m_len = total_len % (MCLBYTES - RE_ETHER_ALIGN);
   1246 			/*
   1247 			 * Special case: if there's 4 bytes or less
   1248 			 * in this buffer, the mbuf can be discarded:
   1249 			 * the last 4 bytes is the CRC, which we don't
   1250 			 * care about anyway.
   1251 			 */
   1252 			if (m->m_len <= ETHER_CRC_LEN) {
   1253 				sc->re_tail->m_len -=
   1254 				    (ETHER_CRC_LEN - m->m_len);
   1255 				m_freem(m);
   1256 			} else {
   1257 				m->m_len -= ETHER_CRC_LEN;
   1258 				m->m_flags &= ~M_PKTHDR;
   1259 				sc->re_tail->m_next = m;
   1260 			}
   1261 			m = sc->re_head;
   1262 			sc->re_head = sc->re_tail = NULL;
   1263 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
   1264 		} else
   1265 			m->m_pkthdr.len = m->m_len =
   1266 			    (total_len - ETHER_CRC_LEN);
   1267 
   1268 		ifp->if_ipackets++;
   1269 		m->m_pkthdr.rcvif = ifp;
   1270 
   1271 		/* Do RX checksumming */
   1272 
   1273 		/* Check IP header checksum */
   1274 		if (rxstat & RE_RDESC_STAT_PROTOID) {
   1275 			m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
   1276 			if (rxstat & RE_RDESC_STAT_IPSUMBAD)
   1277 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4_BAD;
   1278 		}
   1279 
   1280 		/* Check TCP/UDP checksum */
   1281 		if (RE_TCPPKT(rxstat)) {
   1282 			m->m_pkthdr.csum_flags |= M_CSUM_TCPv4;
   1283 			if (rxstat & RE_RDESC_STAT_TCPSUMBAD)
   1284 				m->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
   1285 		} else if (RE_UDPPKT(rxstat)) {
   1286 			m->m_pkthdr.csum_flags |= M_CSUM_UDPv4;
   1287 			if (rxstat & RE_RDESC_STAT_UDPSUMBAD)
   1288 				m->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
   1289 		}
   1290 
   1291 		if (rxvlan & RE_RDESC_VLANCTL_TAG) {
   1292 			VLAN_INPUT_TAG(ifp, m,
   1293 			     bswap16(rxvlan & RE_RDESC_VLANCTL_DATA),
   1294 			     continue);
   1295 		}
   1296 #if NBPFILTER > 0
   1297 		if (ifp->if_bpf)
   1298 			bpf_mtap(ifp->if_bpf, m);
   1299 #endif
   1300 		(*ifp->if_input)(ifp, m);
   1301 	}
   1302 
   1303 	sc->re_ldata.re_rx_prodidx = i;
   1304 }
   1305 
   1306 static void
   1307 re_txeof(struct rtk_softc *sc)
   1308 {
   1309 	struct ifnet		*ifp;
   1310 	struct re_txq		*txq;
   1311 	uint32_t		txstat;
   1312 	int			idx, descidx;
   1313 
   1314 	ifp = &sc->ethercom.ec_if;
   1315 
   1316 	for (idx = sc->re_ldata.re_txq_considx;
   1317 	    sc->re_ldata.re_txq_free < RE_TX_QLEN;
   1318 	    idx = RE_NEXT_TXQ(sc, idx), sc->re_ldata.re_txq_free++) {
   1319 		txq = &sc->re_ldata.re_txq[idx];
   1320 		KASSERT(txq->txq_mbuf != NULL);
   1321 
   1322 		descidx = txq->txq_descidx;
   1323 		RE_TXDESCSYNC(sc, descidx,
   1324 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
   1325 		txstat =
   1326 		    le32toh(sc->re_ldata.re_tx_list[descidx].re_cmdstat);
   1327 		RE_TXDESCSYNC(sc, descidx, BUS_DMASYNC_PREREAD);
   1328 		KASSERT((txstat & RE_TDESC_CMD_EOF) != 0);
   1329 		if (txstat & RE_TDESC_CMD_OWN) {
   1330 			break;
   1331 		}
   1332 
   1333 		sc->re_ldata.re_tx_free += txq->txq_nsegs;
   1334 		KASSERT(sc->re_ldata.re_tx_free <= RE_TX_DESC_CNT(sc));
   1335 		bus_dmamap_sync(sc->sc_dmat, txq->txq_dmamap,
   1336 		    0, txq->txq_dmamap->dm_mapsize, BUS_DMASYNC_POSTWRITE);
   1337 		bus_dmamap_unload(sc->sc_dmat, txq->txq_dmamap);
   1338 		m_freem(txq->txq_mbuf);
   1339 		txq->txq_mbuf = NULL;
   1340 
   1341 		if (txstat & (RE_TDESC_STAT_EXCESSCOL | RE_TDESC_STAT_COLCNT))
   1342 			ifp->if_collisions++;
   1343 		if (txstat & RE_TDESC_STAT_TXERRSUM)
   1344 			ifp->if_oerrors++;
   1345 		else
   1346 			ifp->if_opackets++;
   1347 	}
   1348 
   1349 	sc->re_ldata.re_txq_considx = idx;
   1350 
   1351 	if (sc->re_ldata.re_txq_free > RE_NTXDESC_RSVD)
   1352 		ifp->if_flags &= ~IFF_OACTIVE;
   1353 
   1354 	/*
   1355 	 * If not all descriptors have been released reaped yet,
   1356 	 * reload the timer so that we will eventually get another
   1357 	 * interrupt that will cause us to re-enter this routine.
   1358 	 * This is done in case the transmitter has gone idle.
   1359 	 */
   1360 	if (sc->re_ldata.re_txq_free < RE_TX_QLEN) {
   1361 		CSR_WRITE_4(sc, RTK_TIMERCNT, 1);
   1362 		if ((sc->sc_quirk & RTKQ_PCIE) != 0) {
   1363 			/*
   1364 			 * Some chips will ignore a second TX request
   1365 			 * issued while an existing transmission is in
   1366 			 * progress. If the transmitter goes idle but
   1367 			 * there are still packets waiting to be sent,
   1368 			 * we need to restart the channel here to flush
   1369 			 * them out. This only seems to be required with
   1370 			 * the PCIe devices.
   1371 			 */
   1372 			CSR_WRITE_2(sc, RTK_GTXSTART, RTK_TXSTART_START);
   1373 		}
   1374 	} else
   1375 		ifp->if_timer = 0;
   1376 }
   1377 
   1378 /*
   1379  * Stop all chip I/O so that the kernel's probe routines don't
   1380  * get confused by errant DMAs when rebooting.
   1381  */
   1382 static void
   1383 re_shutdown(void *vsc)
   1384 
   1385 {
   1386 	struct rtk_softc	*sc = vsc;
   1387 
   1388 	re_stop(&sc->ethercom.ec_if, 0);
   1389 }
   1390 
   1391 
   1392 static void
   1393 re_tick(void *xsc)
   1394 {
   1395 	struct rtk_softc	*sc = xsc;
   1396 	int s;
   1397 
   1398 	/*XXX: just return for 8169S/8110S with rev 2 or newer phy */
   1399 	s = splnet();
   1400 
   1401 	mii_tick(&sc->mii);
   1402 	splx(s);
   1403 
   1404 	callout_reset(&sc->rtk_tick_ch, hz, re_tick, sc);
   1405 }
   1406 
   1407 #ifdef DEVICE_POLLING
   1408 static void
   1409 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
   1410 {
   1411 	struct rtk_softc *sc = ifp->if_softc;
   1412 
   1413 	RTK_LOCK(sc);
   1414 	if ((ifp->if_capenable & IFCAP_POLLING) == 0) {
   1415 		ether_poll_deregister(ifp);
   1416 		cmd = POLL_DEREGISTER;
   1417 	}
   1418 	if (cmd == POLL_DEREGISTER) { /* final call, enable interrupts */
   1419 		CSR_WRITE_2(sc, RTK_IMR, RTK_INTRS_CPLUS);
   1420 		goto done;
   1421 	}
   1422 
   1423 	sc->rxcycles = count;
   1424 	re_rxeof(sc);
   1425 	re_txeof(sc);
   1426 
   1427 	if (IFQ_IS_EMPTY(&ifp->if_snd) == 0)
   1428 		(*ifp->if_start)(ifp);
   1429 
   1430 	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
   1431 		uint16_t       status;
   1432 
   1433 		status = CSR_READ_2(sc, RTK_ISR);
   1434 		if (status == 0xffff)
   1435 			goto done;
   1436 		if (status)
   1437 			CSR_WRITE_2(sc, RTK_ISR, status);
   1438 
   1439 		/*
   1440 		 * XXX check behaviour on receiver stalls.
   1441 		 */
   1442 
   1443 		if (status & RTK_ISR_SYSTEM_ERR) {
   1444 			re_init(sc);
   1445 		}
   1446 	}
   1447  done:
   1448 	RTK_UNLOCK(sc);
   1449 }
   1450 #endif /* DEVICE_POLLING */
   1451 
   1452 int
   1453 re_intr(void *arg)
   1454 {
   1455 	struct rtk_softc	*sc = arg;
   1456 	struct ifnet		*ifp;
   1457 	uint16_t		status;
   1458 	int			handled = 0;
   1459 
   1460 	ifp = &sc->ethercom.ec_if;
   1461 
   1462 	if ((ifp->if_flags & IFF_UP) == 0)
   1463 		return 0;
   1464 
   1465 #ifdef DEVICE_POLLING
   1466 	if (ifp->if_flags & IFF_POLLING)
   1467 		goto done;
   1468 	if ((ifp->if_capenable & IFCAP_POLLING) &&
   1469 	    ether_poll_register(re_poll, ifp)) { /* ok, disable interrupts */
   1470 		CSR_WRITE_2(sc, RTK_IMR, 0x0000);
   1471 		re_poll(ifp, 0, 1);
   1472 		goto done;
   1473 	}
   1474 #endif /* DEVICE_POLLING */
   1475 
   1476 	for (;;) {
   1477 
   1478 		status = CSR_READ_2(sc, RTK_ISR);
   1479 		/* If the card has gone away the read returns 0xffff. */
   1480 		if (status == 0xffff)
   1481 			break;
   1482 		if (status) {
   1483 			handled = 1;
   1484 			CSR_WRITE_2(sc, RTK_ISR, status);
   1485 		}
   1486 
   1487 		if ((status & RTK_INTRS_CPLUS) == 0)
   1488 			break;
   1489 
   1490 		if (status & (RTK_ISR_RX_OK | RTK_ISR_RX_ERR))
   1491 			re_rxeof(sc);
   1492 
   1493 		if (status & (RTK_ISR_TIMEOUT_EXPIRED | RTK_ISR_TX_ERR |
   1494 		    RTK_ISR_TX_DESC_UNAVAIL))
   1495 			re_txeof(sc);
   1496 
   1497 		if (status & RTK_ISR_SYSTEM_ERR) {
   1498 			re_init(ifp);
   1499 		}
   1500 
   1501 		if (status & RTK_ISR_LINKCHG) {
   1502 			callout_stop(&sc->rtk_tick_ch);
   1503 			re_tick(sc);
   1504 		}
   1505 	}
   1506 
   1507 	if (handled && !IFQ_IS_EMPTY(&ifp->if_snd))
   1508 		re_start(ifp);
   1509 
   1510 #ifdef DEVICE_POLLING
   1511  done:
   1512 #endif
   1513 
   1514 	return handled;
   1515 }
   1516 
   1517 
   1518 
   1519 /*
   1520  * Main transmit routine for C+ and gigE NICs.
   1521  */
   1522 
   1523 static void
   1524 re_start(struct ifnet *ifp)
   1525 {
   1526 	struct rtk_softc	*sc;
   1527 	struct mbuf		*m;
   1528 	bus_dmamap_t		map;
   1529 	struct re_txq		*txq;
   1530 	struct re_desc		*d;
   1531 	struct m_tag		*mtag;
   1532 	uint32_t		cmdstat, re_flags;
   1533 	int			ofree, idx, error, nsegs, seg;
   1534 	int			startdesc, curdesc, lastdesc;
   1535 	boolean_t		pad;
   1536 
   1537 	sc = ifp->if_softc;
   1538 	ofree = sc->re_ldata.re_txq_free;
   1539 
   1540 	for (idx = sc->re_ldata.re_txq_prodidx;; idx = RE_NEXT_TXQ(sc, idx)) {
   1541 
   1542 		IFQ_POLL(&ifp->if_snd, m);
   1543 		if (m == NULL)
   1544 			break;
   1545 
   1546 		if (sc->re_ldata.re_txq_free == 0 ||
   1547 		    sc->re_ldata.re_tx_free <= RE_NTXDESC_RSVD) {
   1548 			/* no more free slots left */
   1549 			ifp->if_flags |= IFF_OACTIVE;
   1550 			break;
   1551 		}
   1552 
   1553 		/*
   1554 		 * Set up checksum offload. Note: checksum offload bits must
   1555 		 * appear in all descriptors of a multi-descriptor transmit
   1556 		 * attempt. (This is according to testing done with an 8169
   1557 		 * chip. I'm not sure if this is a requirement or a bug.)
   1558 		 */
   1559 
   1560 		if ((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) != 0) {
   1561 			uint32_t segsz = m->m_pkthdr.segsz;
   1562 
   1563 			re_flags = RE_TDESC_CMD_LGSEND |
   1564 			    (segsz << RE_TDESC_CMD_MSSVAL_SHIFT);
   1565 		} else {
   1566 			/*
   1567 			 * set RE_TDESC_CMD_IPCSUM if any checksum offloading
   1568 			 * is requested.  otherwise, RE_TDESC_CMD_TCPCSUM/
   1569 			 * RE_TDESC_CMD_UDPCSUM doesn't make effects.
   1570 			 */
   1571 			re_flags = 0;
   1572 			if ((m->m_pkthdr.csum_flags &
   1573 			    (M_CSUM_IPv4 | M_CSUM_TCPv4 | M_CSUM_UDPv4))
   1574 			    != 0) {
   1575 				re_flags |= RE_TDESC_CMD_IPCSUM;
   1576 				if (m->m_pkthdr.csum_flags & M_CSUM_TCPv4) {
   1577 					re_flags |= RE_TDESC_CMD_TCPCSUM;
   1578 				} else if (m->m_pkthdr.csum_flags &
   1579 				    M_CSUM_UDPv4) {
   1580 					re_flags |= RE_TDESC_CMD_UDPCSUM;
   1581 				}
   1582 			}
   1583 		}
   1584 
   1585 		txq = &sc->re_ldata.re_txq[idx];
   1586 		map = txq->txq_dmamap;
   1587 		error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m,
   1588 		    BUS_DMA_WRITE|BUS_DMA_NOWAIT);
   1589 
   1590 		if (__predict_false(error)) {
   1591 			/* XXX try to defrag if EFBIG? */
   1592 			aprint_error("%s: can't map mbuf (error %d)\n",
   1593 			    sc->sc_dev.dv_xname, error);
   1594 
   1595 			IFQ_DEQUEUE(&ifp->if_snd, m);
   1596 			m_freem(m);
   1597 			ifp->if_oerrors++;
   1598 			continue;
   1599 		}
   1600 
   1601 		nsegs = map->dm_nsegs;
   1602 		pad = FALSE;
   1603 		if (__predict_false(m->m_pkthdr.len <= RE_IP4CSUMTX_PADLEN &&
   1604 		    (re_flags & RE_TDESC_CMD_IPCSUM) != 0)) {
   1605 			pad = TRUE;
   1606 			nsegs++;
   1607 		}
   1608 
   1609 		if (nsegs > sc->re_ldata.re_tx_free - RE_NTXDESC_RSVD) {
   1610 			/*
   1611 			 * Not enough free descriptors to transmit this packet.
   1612 			 */
   1613 			ifp->if_flags |= IFF_OACTIVE;
   1614 			bus_dmamap_unload(sc->sc_dmat, map);
   1615 			break;
   1616 		}
   1617 
   1618 		IFQ_DEQUEUE(&ifp->if_snd, m);
   1619 
   1620 		/*
   1621 		 * Make sure that the caches are synchronized before we
   1622 		 * ask the chip to start DMA for the packet data.
   1623 		 */
   1624 		bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
   1625 		    BUS_DMASYNC_PREWRITE);
   1626 
   1627 		/*
   1628 		 * Map the segment array into descriptors.
   1629 		 * Note that we set the start-of-frame and
   1630 		 * end-of-frame markers for either TX or RX,
   1631 		 * but they really only have meaning in the TX case.
   1632 		 * (In the RX case, it's the chip that tells us
   1633 		 *  where packets begin and end.)
   1634 		 * We also keep track of the end of the ring
   1635 		 * and set the end-of-ring bits as needed,
   1636 		 * and we set the ownership bits in all except
   1637 		 * the very first descriptor. (The caller will
   1638 		 * set this descriptor later when it start
   1639 		 * transmission or reception.)
   1640 		 */
   1641 		curdesc = startdesc = sc->re_ldata.re_tx_nextfree;
   1642 		lastdesc = -1;
   1643 		for (seg = 0; seg < map->dm_nsegs;
   1644 		    seg++, curdesc = RE_NEXT_TX_DESC(sc, curdesc)) {
   1645 			d = &sc->re_ldata.re_tx_list[curdesc];
   1646 #ifdef DIAGNOSTIC
   1647 			RE_TXDESCSYNC(sc, curdesc,
   1648 			    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
   1649 			cmdstat = le32toh(d->re_cmdstat);
   1650 			RE_TXDESCSYNC(sc, curdesc, BUS_DMASYNC_PREREAD);
   1651 			if (cmdstat & RE_TDESC_STAT_OWN) {
   1652 				panic("%s: tried to map busy TX descriptor",
   1653 				    sc->sc_dev.dv_xname);
   1654 			}
   1655 #endif
   1656 
   1657 			d->re_vlanctl = 0;
   1658 			re_set_bufaddr(d, map->dm_segs[seg].ds_addr);
   1659 			cmdstat = re_flags | map->dm_segs[seg].ds_len;
   1660 			if (seg == 0)
   1661 				cmdstat |= RE_TDESC_CMD_SOF;
   1662 			else
   1663 				cmdstat |= RE_TDESC_CMD_OWN;
   1664 			if (curdesc == (RE_TX_DESC_CNT(sc) - 1))
   1665 				cmdstat |= RE_TDESC_CMD_EOR;
   1666 			if (seg == nsegs - 1) {
   1667 				cmdstat |= RE_TDESC_CMD_EOF;
   1668 				lastdesc = curdesc;
   1669 			}
   1670 			d->re_cmdstat = htole32(cmdstat);
   1671 			RE_TXDESCSYNC(sc, curdesc,
   1672 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1673 		}
   1674 		if (__predict_false(pad)) {
   1675 			bus_addr_t paddaddr;
   1676 
   1677 			d = &sc->re_ldata.re_tx_list[curdesc];
   1678 			d->re_vlanctl = 0;
   1679 			paddaddr = RE_TXPADDADDR(sc);
   1680 			re_set_bufaddr(d, paddaddr);
   1681 			cmdstat = re_flags |
   1682 			    RE_TDESC_CMD_OWN | RE_TDESC_CMD_EOF |
   1683 			    (RE_IP4CSUMTX_PADLEN + 1 - m->m_pkthdr.len);
   1684 			if (curdesc == (RE_TX_DESC_CNT(sc) - 1))
   1685 				cmdstat |= RE_TDESC_CMD_EOR;
   1686 			d->re_cmdstat = htole32(cmdstat);
   1687 			RE_TXDESCSYNC(sc, curdesc,
   1688 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1689 			lastdesc = curdesc;
   1690 			curdesc = RE_NEXT_TX_DESC(sc, curdesc);
   1691 		}
   1692 		KASSERT(lastdesc != -1);
   1693 
   1694 		/*
   1695 		 * Set up hardware VLAN tagging. Note: vlan tag info must
   1696 		 * appear in the first descriptor of a multi-descriptor
   1697 		 * transmission attempt.
   1698 		 */
   1699 		if ((mtag = VLAN_OUTPUT_TAG(&sc->ethercom, m)) != NULL) {
   1700 			sc->re_ldata.re_tx_list[startdesc].re_vlanctl =
   1701 			    htole32(bswap16(VLAN_TAG_VALUE(mtag)) |
   1702 			    RE_TDESC_VLANCTL_TAG);
   1703 		}
   1704 
   1705 		/* Transfer ownership of packet to the chip. */
   1706 
   1707 		sc->re_ldata.re_tx_list[startdesc].re_cmdstat |=
   1708 		    htole32(RE_TDESC_CMD_OWN);
   1709 		RE_TXDESCSYNC(sc, startdesc,
   1710 		    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1711 
   1712 		/* update info of TX queue and descriptors */
   1713 		txq->txq_mbuf = m;
   1714 		txq->txq_descidx = lastdesc;
   1715 		txq->txq_nsegs = nsegs;
   1716 
   1717 		sc->re_ldata.re_txq_free--;
   1718 		sc->re_ldata.re_tx_free -= nsegs;
   1719 		sc->re_ldata.re_tx_nextfree = curdesc;
   1720 
   1721 #if NBPFILTER > 0
   1722 		/*
   1723 		 * If there's a BPF listener, bounce a copy of this frame
   1724 		 * to him.
   1725 		 */
   1726 		if (ifp->if_bpf)
   1727 			bpf_mtap(ifp->if_bpf, m);
   1728 #endif
   1729 	}
   1730 
   1731 	if (sc->re_ldata.re_txq_free < ofree) {
   1732 		/*
   1733 		 * TX packets are enqueued.
   1734 		 */
   1735 		sc->re_ldata.re_txq_prodidx = idx;
   1736 
   1737 		/*
   1738 		 * Start the transmitter to poll.
   1739 		 *
   1740 		 * RealTek put the TX poll request register in a different
   1741 		 * location on the 8169 gigE chip. I don't know why.
   1742 		 */
   1743 		if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0)
   1744 			CSR_WRITE_1(sc, RTK_TXSTART, RTK_TXSTART_START);
   1745 		else
   1746 			CSR_WRITE_2(sc, RTK_GTXSTART, RTK_TXSTART_START);
   1747 
   1748 		/*
   1749 		 * Use the countdown timer for interrupt moderation.
   1750 		 * 'TX done' interrupts are disabled. Instead, we reset the
   1751 		 * countdown timer, which will begin counting until it hits
   1752 		 * the value in the TIMERINT register, and then trigger an
   1753 		 * interrupt. Each time we write to the TIMERCNT register,
   1754 		 * the timer count is reset to 0.
   1755 		 */
   1756 		CSR_WRITE_4(sc, RTK_TIMERCNT, 1);
   1757 
   1758 		/*
   1759 		 * Set a timeout in case the chip goes out to lunch.
   1760 		 */
   1761 		ifp->if_timer = 5;
   1762 	}
   1763 }
   1764 
   1765 static int
   1766 re_init(struct ifnet *ifp)
   1767 {
   1768 	struct rtk_softc	*sc = ifp->if_softc;
   1769 	uint8_t			*enaddr;
   1770 	uint32_t		rxcfg = 0;
   1771 	uint32_t		reg;
   1772 	int error;
   1773 
   1774 	if ((error = re_enable(sc)) != 0)
   1775 		goto out;
   1776 
   1777 	/*
   1778 	 * Cancel pending I/O and free all RX/TX buffers.
   1779 	 */
   1780 	re_stop(ifp, 0);
   1781 
   1782 	re_reset(sc);
   1783 
   1784 	/*
   1785 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
   1786 	 * RX checksum offload. We must configure the C+ register
   1787 	 * before all others.
   1788 	 */
   1789 	reg = 0;
   1790 
   1791 	/*
   1792 	 * XXX: Realtek docs say bits 0 and 1 are reserved, for 8169S/8110S.
   1793 	 * FreeBSD  drivers set these bits anyway (for 8139C+?).
   1794 	 * So far, it works.
   1795 	 */
   1796 
   1797 	/*
   1798 	 * XXX: For old 8169 set bit 14.
   1799 	 *      For 8169S/8110S and above, do not set bit 14.
   1800 	 */
   1801 	if ((sc->sc_quirk & RTKQ_8169NONS) != 0)
   1802 		reg |= (0x1 << 14) | RTK_CPLUSCMD_PCI_MRW;;
   1803 
   1804 	if (1)  {/* not for 8169S ? */
   1805 		reg |=
   1806 		    RTK_CPLUSCMD_VLANSTRIP |
   1807 		    (ifp->if_capenable &
   1808 		    (IFCAP_CSUM_IPv4 | IFCAP_CSUM_TCPv4 |
   1809 		     IFCAP_CSUM_UDPv4) ?
   1810 		    RTK_CPLUSCMD_RXCSUM_ENB : 0);
   1811 	}
   1812 
   1813 	CSR_WRITE_2(sc, RTK_CPLUS_CMD,
   1814 	    reg | RTK_CPLUSCMD_RXENB | RTK_CPLUSCMD_TXENB);
   1815 
   1816 	/* XXX: from Realtek-supplied Linux driver. Wholly undocumented. */
   1817 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0)
   1818 		CSR_WRITE_2(sc, RTK_IM, 0x0000);
   1819 
   1820 	DELAY(10000);
   1821 
   1822 	/*
   1823 	 * Init our MAC address.  Even though the chipset
   1824 	 * documentation doesn't mention it, we need to enter "Config
   1825 	 * register write enable" mode to modify the ID registers.
   1826 	 */
   1827 	CSR_WRITE_1(sc, RTK_EECMD, RTK_EEMODE_WRITECFG);
   1828 	enaddr = LLADDR(ifp->if_sadl);
   1829 	reg = enaddr[0] | (enaddr[1] << 8) |
   1830 	    (enaddr[2] << 16) | (enaddr[3] << 24);
   1831 	CSR_WRITE_4(sc, RTK_IDR0, reg);
   1832 	reg = enaddr[4] | (enaddr[5] << 8);
   1833 	CSR_WRITE_4(sc, RTK_IDR4, reg);
   1834 	CSR_WRITE_1(sc, RTK_EECMD, RTK_EEMODE_OFF);
   1835 
   1836 	/*
   1837 	 * For C+ mode, initialize the RX descriptors and mbufs.
   1838 	 */
   1839 	re_rx_list_init(sc);
   1840 	re_tx_list_init(sc);
   1841 
   1842 	/*
   1843 	 * Load the addresses of the RX and TX lists into the chip.
   1844 	 */
   1845 	CSR_WRITE_4(sc, RTK_RXLIST_ADDR_HI,
   1846 	    RE_ADDR_HI(sc->re_ldata.re_rx_list_map->dm_segs[0].ds_addr));
   1847 	CSR_WRITE_4(sc, RTK_RXLIST_ADDR_LO,
   1848 	    RE_ADDR_LO(sc->re_ldata.re_rx_list_map->dm_segs[0].ds_addr));
   1849 
   1850 	CSR_WRITE_4(sc, RTK_TXLIST_ADDR_HI,
   1851 	    RE_ADDR_HI(sc->re_ldata.re_tx_list_map->dm_segs[0].ds_addr));
   1852 	CSR_WRITE_4(sc, RTK_TXLIST_ADDR_LO,
   1853 	    RE_ADDR_LO(sc->re_ldata.re_tx_list_map->dm_segs[0].ds_addr));
   1854 
   1855 	/*
   1856 	 * Enable transmit and receive.
   1857 	 */
   1858 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_TX_ENB | RTK_CMD_RX_ENB);
   1859 
   1860 	/*
   1861 	 * Set the initial TX and RX configuration.
   1862 	 */
   1863 	if (sc->re_testmode && (sc->sc_quirk & RTKQ_8169NONS) != 0) {
   1864 		/* test mode is needed only for old 8169 */
   1865 		CSR_WRITE_4(sc, RTK_TXCFG,
   1866 		    RE_TXCFG_CONFIG | RTK_LOOPTEST_ON);
   1867 	} else
   1868 		CSR_WRITE_4(sc, RTK_TXCFG, RE_TXCFG_CONFIG);
   1869 
   1870 	CSR_WRITE_1(sc, RTK_EARLY_TX_THRESH, 16);
   1871 
   1872 	CSR_WRITE_4(sc, RTK_RXCFG, RE_RXCFG_CONFIG);
   1873 
   1874 	/* Set the individual bit to receive frames for this host only. */
   1875 	rxcfg = CSR_READ_4(sc, RTK_RXCFG);
   1876 	rxcfg |= RTK_RXCFG_RX_INDIV;
   1877 
   1878 	/* If we want promiscuous mode, set the allframes bit. */
   1879 	if (ifp->if_flags & IFF_PROMISC)
   1880 		rxcfg |= RTK_RXCFG_RX_ALLPHYS;
   1881 	else
   1882 		rxcfg &= ~RTK_RXCFG_RX_ALLPHYS;
   1883 	CSR_WRITE_4(sc, RTK_RXCFG, rxcfg);
   1884 
   1885 	/*
   1886 	 * Set capture broadcast bit to capture broadcast frames.
   1887 	 */
   1888 	if (ifp->if_flags & IFF_BROADCAST)
   1889 		rxcfg |= RTK_RXCFG_RX_BROAD;
   1890 	else
   1891 		rxcfg &= ~RTK_RXCFG_RX_BROAD;
   1892 	CSR_WRITE_4(sc, RTK_RXCFG, rxcfg);
   1893 
   1894 	/*
   1895 	 * Program the multicast filter, if necessary.
   1896 	 */
   1897 	rtk_setmulti(sc);
   1898 
   1899 #ifdef DEVICE_POLLING
   1900 	/*
   1901 	 * Disable interrupts if we are polling.
   1902 	 */
   1903 	if (ifp->if_flags & IFF_POLLING)
   1904 		CSR_WRITE_2(sc, RTK_IMR, 0);
   1905 	else	/* otherwise ... */
   1906 #endif /* DEVICE_POLLING */
   1907 	/*
   1908 	 * Enable interrupts.
   1909 	 */
   1910 	if (sc->re_testmode)
   1911 		CSR_WRITE_2(sc, RTK_IMR, 0);
   1912 	else
   1913 		CSR_WRITE_2(sc, RTK_IMR, RTK_INTRS_CPLUS);
   1914 
   1915 	/* Start RX/TX process. */
   1916 	CSR_WRITE_4(sc, RTK_MISSEDPKT, 0);
   1917 #ifdef notdef
   1918 	/* Enable receiver and transmitter. */
   1919 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_TX_ENB | RTK_CMD_RX_ENB);
   1920 #endif
   1921 
   1922 	/*
   1923 	 * Initialize the timer interrupt register so that
   1924 	 * a timer interrupt will be generated once the timer
   1925 	 * reaches a certain number of ticks. The timer is
   1926 	 * reloaded on each transmit. This gives us TX interrupt
   1927 	 * moderation, which dramatically improves TX frame rate.
   1928 	 */
   1929 
   1930 	if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0)
   1931 		CSR_WRITE_4(sc, RTK_TIMERINT, 0x400);
   1932 	else {
   1933 		CSR_WRITE_4(sc, RTK_TIMERINT_8169, 0x800);
   1934 
   1935 		/*
   1936 		 * For 8169 gigE NICs, set the max allowed RX packet
   1937 		 * size so we can receive jumbo frames.
   1938 		 */
   1939 		CSR_WRITE_2(sc, RTK_MAXRXPKTLEN, 16383);
   1940 	}
   1941 
   1942 	if (sc->re_testmode)
   1943 		return 0;
   1944 
   1945 	CSR_WRITE_1(sc, RTK_CFG1, RTK_CFG1_DRVLOAD | RTK_CFG1_FULLDUPLEX);
   1946 
   1947 	ifp->if_flags |= IFF_RUNNING;
   1948 	ifp->if_flags &= ~IFF_OACTIVE;
   1949 
   1950 	callout_reset(&sc->rtk_tick_ch, hz, re_tick, sc);
   1951 
   1952  out:
   1953 	if (error) {
   1954 		ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
   1955 		ifp->if_timer = 0;
   1956 		aprint_error("%s: interface not running\n",
   1957 		    sc->sc_dev.dv_xname);
   1958 	}
   1959 
   1960 	return error;
   1961 }
   1962 
   1963 /*
   1964  * Set media options.
   1965  */
   1966 static int
   1967 re_ifmedia_upd(struct ifnet *ifp)
   1968 {
   1969 	struct rtk_softc	*sc;
   1970 
   1971 	sc = ifp->if_softc;
   1972 
   1973 	return mii_mediachg(&sc->mii);
   1974 }
   1975 
   1976 /*
   1977  * Report current media status.
   1978  */
   1979 static void
   1980 re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
   1981 {
   1982 	struct rtk_softc	*sc;
   1983 
   1984 	sc = ifp->if_softc;
   1985 
   1986 	mii_pollstat(&sc->mii);
   1987 	ifmr->ifm_active = sc->mii.mii_media_active;
   1988 	ifmr->ifm_status = sc->mii.mii_media_status;
   1989 }
   1990 
   1991 static int
   1992 re_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
   1993 {
   1994 	struct rtk_softc	*sc = ifp->if_softc;
   1995 	struct ifreq		*ifr = (struct ifreq *) data;
   1996 	int			s, error = 0;
   1997 
   1998 	s = splnet();
   1999 
   2000 	switch (command) {
   2001 	case SIOCSIFMTU:
   2002 		if (ifr->ifr_mtu > RE_JUMBO_MTU)
   2003 			error = EINVAL;
   2004 		ifp->if_mtu = ifr->ifr_mtu;
   2005 		break;
   2006 	case SIOCGIFMEDIA:
   2007 	case SIOCSIFMEDIA:
   2008 		error = ifmedia_ioctl(ifp, ifr, &sc->mii.mii_media, command);
   2009 		break;
   2010 	default:
   2011 		error = ether_ioctl(ifp, command, data);
   2012 		if (error == ENETRESET) {
   2013 			if (ifp->if_flags & IFF_RUNNING)
   2014 				rtk_setmulti(sc);
   2015 			error = 0;
   2016 		}
   2017 		break;
   2018 	}
   2019 
   2020 	splx(s);
   2021 
   2022 	return error;
   2023 }
   2024 
   2025 static void
   2026 re_watchdog(struct ifnet *ifp)
   2027 {
   2028 	struct rtk_softc	*sc;
   2029 	int			s;
   2030 
   2031 	sc = ifp->if_softc;
   2032 	s = splnet();
   2033 	aprint_error("%s: watchdog timeout\n", sc->sc_dev.dv_xname);
   2034 	ifp->if_oerrors++;
   2035 
   2036 	re_txeof(sc);
   2037 	re_rxeof(sc);
   2038 
   2039 	re_init(ifp);
   2040 
   2041 	splx(s);
   2042 }
   2043 
   2044 /*
   2045  * Stop the adapter and free any mbufs allocated to the
   2046  * RX and TX lists.
   2047  */
   2048 static void
   2049 re_stop(struct ifnet *ifp, int disable)
   2050 {
   2051 	int		i;
   2052 	struct rtk_softc *sc = ifp->if_softc;
   2053 
   2054 	callout_stop(&sc->rtk_tick_ch);
   2055 
   2056 #ifdef DEVICE_POLLING
   2057 	ether_poll_deregister(ifp);
   2058 #endif /* DEVICE_POLLING */
   2059 
   2060 	mii_down(&sc->mii);
   2061 
   2062 	CSR_WRITE_1(sc, RTK_COMMAND, 0x00);
   2063 	CSR_WRITE_2(sc, RTK_IMR, 0x0000);
   2064 
   2065 	if (sc->re_head != NULL) {
   2066 		m_freem(sc->re_head);
   2067 		sc->re_head = sc->re_tail = NULL;
   2068 	}
   2069 
   2070 	/* Free the TX list buffers. */
   2071 	for (i = 0; i < RE_TX_QLEN; i++) {
   2072 		if (sc->re_ldata.re_txq[i].txq_mbuf != NULL) {
   2073 			bus_dmamap_unload(sc->sc_dmat,
   2074 			    sc->re_ldata.re_txq[i].txq_dmamap);
   2075 			m_freem(sc->re_ldata.re_txq[i].txq_mbuf);
   2076 			sc->re_ldata.re_txq[i].txq_mbuf = NULL;
   2077 		}
   2078 	}
   2079 
   2080 	/* Free the RX list buffers. */
   2081 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
   2082 		if (sc->re_ldata.re_rxsoft[i].rxs_mbuf != NULL) {
   2083 			bus_dmamap_unload(sc->sc_dmat,
   2084 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
   2085 			m_freem(sc->re_ldata.re_rxsoft[i].rxs_mbuf);
   2086 			sc->re_ldata.re_rxsoft[i].rxs_mbuf = NULL;
   2087 		}
   2088 	}
   2089 
   2090 	if (disable)
   2091 		re_disable(sc);
   2092 
   2093 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
   2094 	ifp->if_timer = 0;
   2095 }
   2096