Home | History | Annotate | Line # | Download | only in ic
rtl8169.c revision 1.88
      1 /*	$NetBSD: rtl8169.c,v 1.88 2007/08/27 14:48:54 dyoung 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, void *);
    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 	    (void **)&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 	    (void **)&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_Tx | IFCAP_CSUM_IPv4_Rx |
    769 	    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
    770 	    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx |
    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, 0);
    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(sc->sc_dev.dv_xname,
    810 	    re_power, sc);
    811 	if (sc->sc_powerhook == NULL)
    812 		aprint_error("%s: WARNING: unable to establish power hook\n",
    813 		    sc->sc_dev.dv_xname);
    814 
    815 
    816 	return;
    817 
    818  fail_8:
    819 	/* Destroy DMA maps for RX buffers. */
    820 	for (i = 0; i < RE_RX_DESC_CNT; i++)
    821 		if (sc->re_ldata.re_rxsoft[i].rxs_dmamap != NULL)
    822 			bus_dmamap_destroy(sc->sc_dmat,
    823 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
    824 
    825 	/* Free DMA'able memory for the RX ring. */
    826 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
    827  fail_7:
    828 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
    829  fail_6:
    830 	bus_dmamem_unmap(sc->sc_dmat,
    831 	    (void *)sc->re_ldata.re_rx_list, RE_RX_DMAMEM_SZ);
    832  fail_5:
    833 	bus_dmamem_free(sc->sc_dmat,
    834 	    &sc->re_ldata.re_rx_listseg, sc->re_ldata.re_rx_listnseg);
    835 
    836  fail_4:
    837 	/* Destroy DMA maps for TX buffers. */
    838 	for (i = 0; i < RE_TX_QLEN; i++)
    839 		if (sc->re_ldata.re_txq[i].txq_dmamap != NULL)
    840 			bus_dmamap_destroy(sc->sc_dmat,
    841 			    sc->re_ldata.re_txq[i].txq_dmamap);
    842 
    843 	/* Free DMA'able memory for the TX ring. */
    844 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
    845  fail_3:
    846 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
    847  fail_2:
    848 	bus_dmamem_unmap(sc->sc_dmat,
    849 	    (void *)sc->re_ldata.re_tx_list, RE_TX_LIST_SZ(sc));
    850  fail_1:
    851 	bus_dmamem_free(sc->sc_dmat,
    852 	    &sc->re_ldata.re_tx_listseg, sc->re_ldata.re_tx_listnseg);
    853  fail_0:
    854 	return;
    855 }
    856 
    857 
    858 /*
    859  * re_activate:
    860  *     Handle device activation/deactivation requests.
    861  */
    862 int
    863 re_activate(struct device *self, enum devact act)
    864 {
    865 	struct rtk_softc *sc = (void *)self;
    866 	int s, error = 0;
    867 
    868 	s = splnet();
    869 	switch (act) {
    870 	case DVACT_ACTIVATE:
    871 		error = EOPNOTSUPP;
    872 		break;
    873 	case DVACT_DEACTIVATE:
    874 		mii_activate(&sc->mii, act, MII_PHY_ANY, MII_OFFSET_ANY);
    875 		if_deactivate(&sc->ethercom.ec_if);
    876 		break;
    877 	}
    878 	splx(s);
    879 
    880 	return error;
    881 }
    882 
    883 /*
    884  * re_detach:
    885  *     Detach a rtk interface.
    886  */
    887 int
    888 re_detach(struct rtk_softc *sc)
    889 {
    890 	struct ifnet *ifp = &sc->ethercom.ec_if;
    891 	int i;
    892 
    893 	/*
    894 	 * Succeed now if there isn't any work to do.
    895 	 */
    896 	if ((sc->sc_flags & RTK_ATTACHED) == 0)
    897 		return 0;
    898 
    899 	/* Unhook our tick handler. */
    900 	callout_stop(&sc->rtk_tick_ch);
    901 
    902 	/* Detach all PHYs. */
    903 	mii_detach(&sc->mii, MII_PHY_ANY, MII_OFFSET_ANY);
    904 
    905 	/* Delete all remaining media. */
    906 	ifmedia_delete_instance(&sc->mii.mii_media, IFM_INST_ANY);
    907 
    908 	ether_ifdetach(ifp);
    909 	if_detach(ifp);
    910 
    911 	/* Destroy DMA maps for RX buffers. */
    912 	for (i = 0; i < RE_RX_DESC_CNT; i++)
    913 		if (sc->re_ldata.re_rxsoft[i].rxs_dmamap != NULL)
    914 			bus_dmamap_destroy(sc->sc_dmat,
    915 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
    916 
    917 	/* Free DMA'able memory for the RX ring. */
    918 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
    919 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
    920 	bus_dmamem_unmap(sc->sc_dmat,
    921 	    (void *)sc->re_ldata.re_rx_list, RE_RX_DMAMEM_SZ);
    922 	bus_dmamem_free(sc->sc_dmat,
    923 	    &sc->re_ldata.re_rx_listseg, sc->re_ldata.re_rx_listnseg);
    924 
    925 	/* Destroy DMA maps for TX buffers. */
    926 	for (i = 0; i < RE_TX_QLEN; i++)
    927 		if (sc->re_ldata.re_txq[i].txq_dmamap != NULL)
    928 			bus_dmamap_destroy(sc->sc_dmat,
    929 			    sc->re_ldata.re_txq[i].txq_dmamap);
    930 
    931 	/* Free DMA'able memory for the TX ring. */
    932 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
    933 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
    934 	bus_dmamem_unmap(sc->sc_dmat,
    935 	    (void *)sc->re_ldata.re_tx_list, RE_TX_LIST_SZ(sc));
    936 	bus_dmamem_free(sc->sc_dmat,
    937 	    &sc->re_ldata.re_tx_listseg, sc->re_ldata.re_tx_listnseg);
    938 
    939 
    940 	shutdownhook_disestablish(sc->sc_sdhook);
    941 	powerhook_disestablish(sc->sc_powerhook);
    942 
    943 	return 0;
    944 }
    945 
    946 /*
    947  * re_enable:
    948  *     Enable the RTL81X9 chip.
    949  */
    950 static int
    951 re_enable(struct rtk_softc *sc)
    952 {
    953 
    954 	if (RTK_IS_ENABLED(sc) == 0 && sc->sc_enable != NULL) {
    955 		if ((*sc->sc_enable)(sc) != 0) {
    956 			aprint_error("%s: device enable failed\n",
    957 			    sc->sc_dev.dv_xname);
    958 			return EIO;
    959 		}
    960 		sc->sc_flags |= RTK_ENABLED;
    961 	}
    962 	return 0;
    963 }
    964 
    965 /*
    966  * re_disable:
    967  *     Disable the RTL81X9 chip.
    968  */
    969 static void
    970 re_disable(struct rtk_softc *sc)
    971 {
    972 
    973 	if (RTK_IS_ENABLED(sc) && sc->sc_disable != NULL) {
    974 		(*sc->sc_disable)(sc);
    975 		sc->sc_flags &= ~RTK_ENABLED;
    976 	}
    977 }
    978 
    979 /*
    980  * re_power:
    981  *     Power management (suspend/resume) hook.
    982  */
    983 void
    984 re_power(int why, void *arg)
    985 {
    986 	struct rtk_softc *sc = (void *)arg;
    987 	struct ifnet *ifp = &sc->ethercom.ec_if;
    988 	int s;
    989 
    990 	s = splnet();
    991 	switch (why) {
    992 	case PWR_SUSPEND:
    993 	case PWR_STANDBY:
    994 		re_stop(ifp, 0);
    995 		if (sc->sc_power != NULL)
    996 			(*sc->sc_power)(sc, why);
    997 		break;
    998 	case PWR_RESUME:
    999 		if (ifp->if_flags & IFF_UP) {
   1000 			if (sc->sc_power != NULL)
   1001 				(*sc->sc_power)(sc, why);
   1002 			re_init(ifp);
   1003 		}
   1004 		break;
   1005 	case PWR_SOFTSUSPEND:
   1006 	case PWR_SOFTSTANDBY:
   1007 	case PWR_SOFTRESUME:
   1008 		break;
   1009 	}
   1010 	splx(s);
   1011 }
   1012 
   1013 
   1014 static int
   1015 re_newbuf(struct rtk_softc *sc, int idx, struct mbuf *m)
   1016 {
   1017 	struct mbuf		*n = NULL;
   1018 	bus_dmamap_t		map;
   1019 	struct re_desc		*d;
   1020 	struct re_rxsoft	*rxs;
   1021 	uint32_t		cmdstat;
   1022 	int			error;
   1023 
   1024 	if (m == NULL) {
   1025 		MGETHDR(n, M_DONTWAIT, MT_DATA);
   1026 		if (n == NULL)
   1027 			return ENOBUFS;
   1028 
   1029 		MCLGET(n, M_DONTWAIT);
   1030 		if ((n->m_flags & M_EXT) == 0) {
   1031 			m_freem(n);
   1032 			return ENOBUFS;
   1033 		}
   1034 		m = n;
   1035 	} else
   1036 		m->m_data = m->m_ext.ext_buf;
   1037 
   1038 	/*
   1039 	 * Initialize mbuf length fields and fixup
   1040 	 * alignment so that the frame payload is
   1041 	 * longword aligned.
   1042 	 */
   1043 	m->m_len = m->m_pkthdr.len = MCLBYTES - RE_ETHER_ALIGN;
   1044 	m->m_data += RE_ETHER_ALIGN;
   1045 
   1046 	rxs = &sc->re_ldata.re_rxsoft[idx];
   1047 	map = rxs->rxs_dmamap;
   1048 	error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m,
   1049 	    BUS_DMA_READ|BUS_DMA_NOWAIT);
   1050 
   1051 	if (error)
   1052 		goto out;
   1053 
   1054 	bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
   1055 	    BUS_DMASYNC_PREREAD);
   1056 
   1057 	d = &sc->re_ldata.re_rx_list[idx];
   1058 #ifdef DIAGNOSTIC
   1059 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
   1060 	cmdstat = le32toh(d->re_cmdstat);
   1061 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD);
   1062 	if (cmdstat & RE_RDESC_STAT_OWN) {
   1063 		panic("%s: tried to map busy RX descriptor",
   1064 		    sc->sc_dev.dv_xname);
   1065 	}
   1066 #endif
   1067 
   1068 	rxs->rxs_mbuf = m;
   1069 
   1070 	d->re_vlanctl = 0;
   1071 	cmdstat = map->dm_segs[0].ds_len;
   1072 	if (idx == (RE_RX_DESC_CNT - 1))
   1073 		cmdstat |= RE_RDESC_CMD_EOR;
   1074 	re_set_bufaddr(d, map->dm_segs[0].ds_addr);
   1075 	d->re_cmdstat = htole32(cmdstat);
   1076 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1077 	cmdstat |= RE_RDESC_CMD_OWN;
   1078 	d->re_cmdstat = htole32(cmdstat);
   1079 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1080 
   1081 	return 0;
   1082  out:
   1083 	if (n != NULL)
   1084 		m_freem(n);
   1085 	return ENOMEM;
   1086 }
   1087 
   1088 static int
   1089 re_tx_list_init(struct rtk_softc *sc)
   1090 {
   1091 	int i;
   1092 
   1093 	memset(sc->re_ldata.re_tx_list, 0, RE_TX_LIST_SZ(sc));
   1094 	for (i = 0; i < RE_TX_QLEN; i++) {
   1095 		sc->re_ldata.re_txq[i].txq_mbuf = NULL;
   1096 	}
   1097 
   1098 	bus_dmamap_sync(sc->sc_dmat,
   1099 	    sc->re_ldata.re_tx_list_map, 0,
   1100 	    sc->re_ldata.re_tx_list_map->dm_mapsize,
   1101 	    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1102 	sc->re_ldata.re_txq_prodidx = 0;
   1103 	sc->re_ldata.re_txq_considx = 0;
   1104 	sc->re_ldata.re_txq_free = RE_TX_QLEN;
   1105 	sc->re_ldata.re_tx_free = RE_TX_DESC_CNT(sc);
   1106 	sc->re_ldata.re_tx_nextfree = 0;
   1107 
   1108 	return 0;
   1109 }
   1110 
   1111 static int
   1112 re_rx_list_init(struct rtk_softc *sc)
   1113 {
   1114 	int			i;
   1115 
   1116 	memset((char *)sc->re_ldata.re_rx_list, 0, RE_RX_LIST_SZ);
   1117 
   1118 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
   1119 		if (re_newbuf(sc, i, NULL) == ENOBUFS)
   1120 			return ENOBUFS;
   1121 	}
   1122 
   1123 	sc->re_ldata.re_rx_prodidx = 0;
   1124 	sc->re_head = sc->re_tail = NULL;
   1125 
   1126 	return 0;
   1127 }
   1128 
   1129 /*
   1130  * RX handler for C+ and 8169. For the gigE chips, we support
   1131  * the reception of jumbo frames that have been fragmented
   1132  * across multiple 2K mbuf cluster buffers.
   1133  */
   1134 static void
   1135 re_rxeof(struct rtk_softc *sc)
   1136 {
   1137 	struct mbuf		*m;
   1138 	struct ifnet		*ifp;
   1139 	int			i, total_len;
   1140 	struct re_desc		*cur_rx;
   1141 	struct re_rxsoft	*rxs;
   1142 	uint32_t		rxstat, rxvlan;
   1143 
   1144 	ifp = &sc->ethercom.ec_if;
   1145 
   1146 	for (i = sc->re_ldata.re_rx_prodidx;; i = RE_NEXT_RX_DESC(sc, i)) {
   1147 		cur_rx = &sc->re_ldata.re_rx_list[i];
   1148 		RE_RXDESCSYNC(sc, i,
   1149 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
   1150 		rxstat = le32toh(cur_rx->re_cmdstat);
   1151 		RE_RXDESCSYNC(sc, i, BUS_DMASYNC_PREREAD);
   1152 		if ((rxstat & RE_RDESC_STAT_OWN) != 0) {
   1153 			break;
   1154 		}
   1155 		total_len = rxstat & sc->re_rxlenmask;
   1156 		rxvlan = le32toh(cur_rx->re_vlanctl);
   1157 		rxs = &sc->re_ldata.re_rxsoft[i];
   1158 		m = rxs->rxs_mbuf;
   1159 
   1160 		/* Invalidate the RX mbuf and unload its map */
   1161 
   1162 		bus_dmamap_sync(sc->sc_dmat,
   1163 		    rxs->rxs_dmamap, 0, rxs->rxs_dmamap->dm_mapsize,
   1164 		    BUS_DMASYNC_POSTREAD);
   1165 		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
   1166 
   1167 		if ((rxstat & RE_RDESC_STAT_EOF) == 0) {
   1168 			m->m_len = MCLBYTES - RE_ETHER_ALIGN;
   1169 			if (sc->re_head == NULL)
   1170 				sc->re_head = sc->re_tail = m;
   1171 			else {
   1172 				m->m_flags &= ~M_PKTHDR;
   1173 				sc->re_tail->m_next = m;
   1174 				sc->re_tail = m;
   1175 			}
   1176 			re_newbuf(sc, i, NULL);
   1177 			continue;
   1178 		}
   1179 
   1180 		/*
   1181 		 * NOTE: for the 8139C+, the frame length field
   1182 		 * is always 12 bits in size, but for the gigE chips,
   1183 		 * it is 13 bits (since the max RX frame length is 16K).
   1184 		 * Unfortunately, all 32 bits in the status word
   1185 		 * were already used, so to make room for the extra
   1186 		 * length bit, RealTek took out the 'frame alignment
   1187 		 * error' bit and shifted the other status bits
   1188 		 * over one slot. The OWN, EOR, FS and LS bits are
   1189 		 * still in the same places. We have already extracted
   1190 		 * the frame length and checked the OWN bit, so rather
   1191 		 * than using an alternate bit mapping, we shift the
   1192 		 * status bits one space to the right so we can evaluate
   1193 		 * them using the 8169 status as though it was in the
   1194 		 * same format as that of the 8139C+.
   1195 		 */
   1196 		if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0)
   1197 			rxstat >>= 1;
   1198 
   1199 		if (__predict_false((rxstat & RE_RDESC_STAT_RXERRSUM) != 0)) {
   1200 #ifdef RE_DEBUG
   1201 			aprint_error("%s: RX error (rxstat = 0x%08x)",
   1202 			    sc->sc_dev.dv_xname, rxstat);
   1203 			if (rxstat & RE_RDESC_STAT_FRALIGN)
   1204 				aprint_error(", frame alignment error");
   1205 			if (rxstat & RE_RDESC_STAT_BUFOFLOW)
   1206 				aprint_error(", out of buffer space");
   1207 			if (rxstat & RE_RDESC_STAT_FIFOOFLOW)
   1208 				aprint_error(", FIFO overrun");
   1209 			if (rxstat & RE_RDESC_STAT_GIANT)
   1210 				aprint_error(", giant packet");
   1211 			if (rxstat & RE_RDESC_STAT_RUNT)
   1212 				aprint_error(", runt packet");
   1213 			if (rxstat & RE_RDESC_STAT_CRCERR)
   1214 				aprint_error(", CRC error");
   1215 			aprint_error("\n");
   1216 #endif
   1217 			ifp->if_ierrors++;
   1218 			/*
   1219 			 * If this is part of a multi-fragment packet,
   1220 			 * discard all the pieces.
   1221 			 */
   1222 			if (sc->re_head != NULL) {
   1223 				m_freem(sc->re_head);
   1224 				sc->re_head = sc->re_tail = NULL;
   1225 			}
   1226 			re_newbuf(sc, i, m);
   1227 			continue;
   1228 		}
   1229 
   1230 		/*
   1231 		 * If allocating a replacement mbuf fails,
   1232 		 * reload the current one.
   1233 		 */
   1234 
   1235 		if (__predict_false(re_newbuf(sc, i, NULL) != 0)) {
   1236 			ifp->if_ierrors++;
   1237 			if (sc->re_head != NULL) {
   1238 				m_freem(sc->re_head);
   1239 				sc->re_head = sc->re_tail = NULL;
   1240 			}
   1241 			re_newbuf(sc, i, m);
   1242 			continue;
   1243 		}
   1244 
   1245 		if (sc->re_head != NULL) {
   1246 			m->m_len = total_len % (MCLBYTES - RE_ETHER_ALIGN);
   1247 			/*
   1248 			 * Special case: if there's 4 bytes or less
   1249 			 * in this buffer, the mbuf can be discarded:
   1250 			 * the last 4 bytes is the CRC, which we don't
   1251 			 * care about anyway.
   1252 			 */
   1253 			if (m->m_len <= ETHER_CRC_LEN) {
   1254 				sc->re_tail->m_len -=
   1255 				    (ETHER_CRC_LEN - m->m_len);
   1256 				m_freem(m);
   1257 			} else {
   1258 				m->m_len -= ETHER_CRC_LEN;
   1259 				m->m_flags &= ~M_PKTHDR;
   1260 				sc->re_tail->m_next = m;
   1261 			}
   1262 			m = sc->re_head;
   1263 			sc->re_head = sc->re_tail = NULL;
   1264 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
   1265 		} else
   1266 			m->m_pkthdr.len = m->m_len =
   1267 			    (total_len - ETHER_CRC_LEN);
   1268 
   1269 		ifp->if_ipackets++;
   1270 		m->m_pkthdr.rcvif = ifp;
   1271 
   1272 		/* Do RX checksumming */
   1273 
   1274 		/* Check IP header checksum */
   1275 		if (rxstat & RE_RDESC_STAT_PROTOID) {
   1276 			m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
   1277 			if (rxstat & RE_RDESC_STAT_IPSUMBAD)
   1278 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4_BAD;
   1279 		}
   1280 
   1281 		/* Check TCP/UDP checksum */
   1282 		if (RE_TCPPKT(rxstat)) {
   1283 			m->m_pkthdr.csum_flags |= M_CSUM_TCPv4;
   1284 			if (rxstat & RE_RDESC_STAT_TCPSUMBAD)
   1285 				m->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
   1286 		} else if (RE_UDPPKT(rxstat)) {
   1287 			m->m_pkthdr.csum_flags |= M_CSUM_UDPv4;
   1288 			if (rxstat & RE_RDESC_STAT_UDPSUMBAD)
   1289 				m->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
   1290 		}
   1291 
   1292 		if (rxvlan & RE_RDESC_VLANCTL_TAG) {
   1293 			VLAN_INPUT_TAG(ifp, m,
   1294 			     bswap16(rxvlan & RE_RDESC_VLANCTL_DATA),
   1295 			     continue);
   1296 		}
   1297 #if NBPFILTER > 0
   1298 		if (ifp->if_bpf)
   1299 			bpf_mtap(ifp->if_bpf, m);
   1300 #endif
   1301 		(*ifp->if_input)(ifp, m);
   1302 	}
   1303 
   1304 	sc->re_ldata.re_rx_prodidx = i;
   1305 }
   1306 
   1307 static void
   1308 re_txeof(struct rtk_softc *sc)
   1309 {
   1310 	struct ifnet		*ifp;
   1311 	struct re_txq		*txq;
   1312 	uint32_t		txstat;
   1313 	int			idx, descidx;
   1314 
   1315 	ifp = &sc->ethercom.ec_if;
   1316 
   1317 	for (idx = sc->re_ldata.re_txq_considx;
   1318 	    sc->re_ldata.re_txq_free < RE_TX_QLEN;
   1319 	    idx = RE_NEXT_TXQ(sc, idx), sc->re_ldata.re_txq_free++) {
   1320 		txq = &sc->re_ldata.re_txq[idx];
   1321 		KASSERT(txq->txq_mbuf != NULL);
   1322 
   1323 		descidx = txq->txq_descidx;
   1324 		RE_TXDESCSYNC(sc, descidx,
   1325 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
   1326 		txstat =
   1327 		    le32toh(sc->re_ldata.re_tx_list[descidx].re_cmdstat);
   1328 		RE_TXDESCSYNC(sc, descidx, BUS_DMASYNC_PREREAD);
   1329 		KASSERT((txstat & RE_TDESC_CMD_EOF) != 0);
   1330 		if (txstat & RE_TDESC_CMD_OWN) {
   1331 			break;
   1332 		}
   1333 
   1334 		sc->re_ldata.re_tx_free += txq->txq_nsegs;
   1335 		KASSERT(sc->re_ldata.re_tx_free <= RE_TX_DESC_CNT(sc));
   1336 		bus_dmamap_sync(sc->sc_dmat, txq->txq_dmamap,
   1337 		    0, txq->txq_dmamap->dm_mapsize, BUS_DMASYNC_POSTWRITE);
   1338 		bus_dmamap_unload(sc->sc_dmat, txq->txq_dmamap);
   1339 		m_freem(txq->txq_mbuf);
   1340 		txq->txq_mbuf = NULL;
   1341 
   1342 		if (txstat & (RE_TDESC_STAT_EXCESSCOL | RE_TDESC_STAT_COLCNT))
   1343 			ifp->if_collisions++;
   1344 		if (txstat & RE_TDESC_STAT_TXERRSUM)
   1345 			ifp->if_oerrors++;
   1346 		else
   1347 			ifp->if_opackets++;
   1348 	}
   1349 
   1350 	sc->re_ldata.re_txq_considx = idx;
   1351 
   1352 	if (sc->re_ldata.re_txq_free > RE_NTXDESC_RSVD)
   1353 		ifp->if_flags &= ~IFF_OACTIVE;
   1354 
   1355 	/*
   1356 	 * If not all descriptors have been released reaped yet,
   1357 	 * reload the timer so that we will eventually get another
   1358 	 * interrupt that will cause us to re-enter this routine.
   1359 	 * This is done in case the transmitter has gone idle.
   1360 	 */
   1361 	if (sc->re_ldata.re_txq_free < RE_TX_QLEN) {
   1362 		CSR_WRITE_4(sc, RTK_TIMERCNT, 1);
   1363 		if ((sc->sc_quirk & RTKQ_PCIE) != 0) {
   1364 			/*
   1365 			 * Some chips will ignore a second TX request
   1366 			 * issued while an existing transmission is in
   1367 			 * progress. If the transmitter goes idle but
   1368 			 * there are still packets waiting to be sent,
   1369 			 * we need to restart the channel here to flush
   1370 			 * them out. This only seems to be required with
   1371 			 * the PCIe devices.
   1372 			 */
   1373 			CSR_WRITE_2(sc, RTK_GTXSTART, RTK_TXSTART_START);
   1374 		}
   1375 	} else
   1376 		ifp->if_timer = 0;
   1377 }
   1378 
   1379 /*
   1380  * Stop all chip I/O so that the kernel's probe routines don't
   1381  * get confused by errant DMAs when rebooting.
   1382  */
   1383 static void
   1384 re_shutdown(void *vsc)
   1385 
   1386 {
   1387 	struct rtk_softc	*sc = vsc;
   1388 
   1389 	re_stop(&sc->ethercom.ec_if, 0);
   1390 }
   1391 
   1392 
   1393 static void
   1394 re_tick(void *xsc)
   1395 {
   1396 	struct rtk_softc	*sc = xsc;
   1397 	int s;
   1398 
   1399 	/*XXX: just return for 8169S/8110S with rev 2 or newer phy */
   1400 	s = splnet();
   1401 
   1402 	mii_tick(&sc->mii);
   1403 	splx(s);
   1404 
   1405 	callout_reset(&sc->rtk_tick_ch, hz, re_tick, sc);
   1406 }
   1407 
   1408 #ifdef DEVICE_POLLING
   1409 static void
   1410 re_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
   1411 {
   1412 	struct rtk_softc *sc = ifp->if_softc;
   1413 
   1414 	RTK_LOCK(sc);
   1415 	if ((ifp->if_capenable & IFCAP_POLLING) == 0) {
   1416 		ether_poll_deregister(ifp);
   1417 		cmd = POLL_DEREGISTER;
   1418 	}
   1419 	if (cmd == POLL_DEREGISTER) { /* final call, enable interrupts */
   1420 		CSR_WRITE_2(sc, RTK_IMR, RTK_INTRS_CPLUS);
   1421 		goto done;
   1422 	}
   1423 
   1424 	sc->rxcycles = count;
   1425 	re_rxeof(sc);
   1426 	re_txeof(sc);
   1427 
   1428 	if (IFQ_IS_EMPTY(&ifp->if_snd) == 0)
   1429 		(*ifp->if_start)(ifp);
   1430 
   1431 	if (cmd == POLL_AND_CHECK_STATUS) { /* also check status register */
   1432 		uint16_t       status;
   1433 
   1434 		status = CSR_READ_2(sc, RTK_ISR);
   1435 		if (status == 0xffff)
   1436 			goto done;
   1437 		if (status)
   1438 			CSR_WRITE_2(sc, RTK_ISR, status);
   1439 
   1440 		/*
   1441 		 * XXX check behaviour on receiver stalls.
   1442 		 */
   1443 
   1444 		if (status & RTK_ISR_SYSTEM_ERR) {
   1445 			re_init(sc);
   1446 		}
   1447 	}
   1448  done:
   1449 	RTK_UNLOCK(sc);
   1450 }
   1451 #endif /* DEVICE_POLLING */
   1452 
   1453 int
   1454 re_intr(void *arg)
   1455 {
   1456 	struct rtk_softc	*sc = arg;
   1457 	struct ifnet		*ifp;
   1458 	uint16_t		status;
   1459 	int			handled = 0;
   1460 
   1461 	ifp = &sc->ethercom.ec_if;
   1462 
   1463 	if ((ifp->if_flags & IFF_UP) == 0)
   1464 		return 0;
   1465 
   1466 #ifdef DEVICE_POLLING
   1467 	if (ifp->if_flags & IFF_POLLING)
   1468 		goto done;
   1469 	if ((ifp->if_capenable & IFCAP_POLLING) &&
   1470 	    ether_poll_register(re_poll, ifp)) { /* ok, disable interrupts */
   1471 		CSR_WRITE_2(sc, RTK_IMR, 0x0000);
   1472 		re_poll(ifp, 0, 1);
   1473 		goto done;
   1474 	}
   1475 #endif /* DEVICE_POLLING */
   1476 
   1477 	for (;;) {
   1478 
   1479 		status = CSR_READ_2(sc, RTK_ISR);
   1480 		/* If the card has gone away the read returns 0xffff. */
   1481 		if (status == 0xffff)
   1482 			break;
   1483 		if (status) {
   1484 			handled = 1;
   1485 			CSR_WRITE_2(sc, RTK_ISR, status);
   1486 		}
   1487 
   1488 		if ((status & RTK_INTRS_CPLUS) == 0)
   1489 			break;
   1490 
   1491 		if (status & (RTK_ISR_RX_OK | RTK_ISR_RX_ERR))
   1492 			re_rxeof(sc);
   1493 
   1494 		if (status & (RTK_ISR_TIMEOUT_EXPIRED | RTK_ISR_TX_ERR |
   1495 		    RTK_ISR_TX_DESC_UNAVAIL))
   1496 			re_txeof(sc);
   1497 
   1498 		if (status & RTK_ISR_SYSTEM_ERR) {
   1499 			re_init(ifp);
   1500 		}
   1501 
   1502 		if (status & RTK_ISR_LINKCHG) {
   1503 			callout_stop(&sc->rtk_tick_ch);
   1504 			re_tick(sc);
   1505 		}
   1506 	}
   1507 
   1508 	if (handled && !IFQ_IS_EMPTY(&ifp->if_snd))
   1509 		re_start(ifp);
   1510 
   1511 #ifdef DEVICE_POLLING
   1512  done:
   1513 #endif
   1514 
   1515 	return handled;
   1516 }
   1517 
   1518 
   1519 
   1520 /*
   1521  * Main transmit routine for C+ and gigE NICs.
   1522  */
   1523 
   1524 static void
   1525 re_start(struct ifnet *ifp)
   1526 {
   1527 	struct rtk_softc	*sc;
   1528 	struct mbuf		*m;
   1529 	bus_dmamap_t		map;
   1530 	struct re_txq		*txq;
   1531 	struct re_desc		*d;
   1532 	struct m_tag		*mtag;
   1533 	uint32_t		cmdstat, re_flags;
   1534 	int			ofree, idx, error, nsegs, seg;
   1535 	int			startdesc, curdesc, lastdesc;
   1536 	bool			pad;
   1537 
   1538 	sc = ifp->if_softc;
   1539 	ofree = sc->re_ldata.re_txq_free;
   1540 
   1541 	for (idx = sc->re_ldata.re_txq_prodidx;; idx = RE_NEXT_TXQ(sc, idx)) {
   1542 
   1543 		IFQ_POLL(&ifp->if_snd, m);
   1544 		if (m == NULL)
   1545 			break;
   1546 
   1547 		if (sc->re_ldata.re_txq_free == 0 ||
   1548 		    sc->re_ldata.re_tx_free <= RE_NTXDESC_RSVD) {
   1549 			/* no more free slots left */
   1550 			ifp->if_flags |= IFF_OACTIVE;
   1551 			break;
   1552 		}
   1553 
   1554 		/*
   1555 		 * Set up checksum offload. Note: checksum offload bits must
   1556 		 * appear in all descriptors of a multi-descriptor transmit
   1557 		 * attempt. (This is according to testing done with an 8169
   1558 		 * chip. I'm not sure if this is a requirement or a bug.)
   1559 		 */
   1560 
   1561 		if ((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) != 0) {
   1562 			uint32_t segsz = m->m_pkthdr.segsz;
   1563 
   1564 			re_flags = RE_TDESC_CMD_LGSEND |
   1565 			    (segsz << RE_TDESC_CMD_MSSVAL_SHIFT);
   1566 		} else {
   1567 			/*
   1568 			 * set RE_TDESC_CMD_IPCSUM if any checksum offloading
   1569 			 * is requested.  otherwise, RE_TDESC_CMD_TCPCSUM/
   1570 			 * RE_TDESC_CMD_UDPCSUM doesn't make effects.
   1571 			 */
   1572 			re_flags = 0;
   1573 			if ((m->m_pkthdr.csum_flags &
   1574 			    (M_CSUM_IPv4 | M_CSUM_TCPv4 | M_CSUM_UDPv4))
   1575 			    != 0) {
   1576 				re_flags |= RE_TDESC_CMD_IPCSUM;
   1577 				if (m->m_pkthdr.csum_flags & M_CSUM_TCPv4) {
   1578 					re_flags |= RE_TDESC_CMD_TCPCSUM;
   1579 				} else if (m->m_pkthdr.csum_flags &
   1580 				    M_CSUM_UDPv4) {
   1581 					re_flags |= RE_TDESC_CMD_UDPCSUM;
   1582 				}
   1583 			}
   1584 		}
   1585 
   1586 		txq = &sc->re_ldata.re_txq[idx];
   1587 		map = txq->txq_dmamap;
   1588 		error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m,
   1589 		    BUS_DMA_WRITE|BUS_DMA_NOWAIT);
   1590 
   1591 		if (__predict_false(error)) {
   1592 			/* XXX try to defrag if EFBIG? */
   1593 			aprint_error("%s: can't map mbuf (error %d)\n",
   1594 			    sc->sc_dev.dv_xname, error);
   1595 
   1596 			IFQ_DEQUEUE(&ifp->if_snd, m);
   1597 			m_freem(m);
   1598 			ifp->if_oerrors++;
   1599 			continue;
   1600 		}
   1601 
   1602 		nsegs = map->dm_nsegs;
   1603 		pad = false;
   1604 		if (__predict_false(m->m_pkthdr.len <= RE_IP4CSUMTX_PADLEN &&
   1605 		    (re_flags & RE_TDESC_CMD_IPCSUM) != 0)) {
   1606 			pad = true;
   1607 			nsegs++;
   1608 		}
   1609 
   1610 		if (nsegs > sc->re_ldata.re_tx_free - RE_NTXDESC_RSVD) {
   1611 			/*
   1612 			 * Not enough free descriptors to transmit this packet.
   1613 			 */
   1614 			ifp->if_flags |= IFF_OACTIVE;
   1615 			bus_dmamap_unload(sc->sc_dmat, map);
   1616 			break;
   1617 		}
   1618 
   1619 		IFQ_DEQUEUE(&ifp->if_snd, m);
   1620 
   1621 		/*
   1622 		 * Make sure that the caches are synchronized before we
   1623 		 * ask the chip to start DMA for the packet data.
   1624 		 */
   1625 		bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
   1626 		    BUS_DMASYNC_PREWRITE);
   1627 
   1628 		/*
   1629 		 * Map the segment array into descriptors.
   1630 		 * Note that we set the start-of-frame and
   1631 		 * end-of-frame markers for either TX or RX,
   1632 		 * but they really only have meaning in the TX case.
   1633 		 * (In the RX case, it's the chip that tells us
   1634 		 *  where packets begin and end.)
   1635 		 * We also keep track of the end of the ring
   1636 		 * and set the end-of-ring bits as needed,
   1637 		 * and we set the ownership bits in all except
   1638 		 * the very first descriptor. (The caller will
   1639 		 * set this descriptor later when it start
   1640 		 * transmission or reception.)
   1641 		 */
   1642 		curdesc = startdesc = sc->re_ldata.re_tx_nextfree;
   1643 		lastdesc = -1;
   1644 		for (seg = 0; seg < map->dm_nsegs;
   1645 		    seg++, curdesc = RE_NEXT_TX_DESC(sc, curdesc)) {
   1646 			d = &sc->re_ldata.re_tx_list[curdesc];
   1647 #ifdef DIAGNOSTIC
   1648 			RE_TXDESCSYNC(sc, curdesc,
   1649 			    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
   1650 			cmdstat = le32toh(d->re_cmdstat);
   1651 			RE_TXDESCSYNC(sc, curdesc, BUS_DMASYNC_PREREAD);
   1652 			if (cmdstat & RE_TDESC_STAT_OWN) {
   1653 				panic("%s: tried to map busy TX descriptor",
   1654 				    sc->sc_dev.dv_xname);
   1655 			}
   1656 #endif
   1657 
   1658 			d->re_vlanctl = 0;
   1659 			re_set_bufaddr(d, map->dm_segs[seg].ds_addr);
   1660 			cmdstat = re_flags | map->dm_segs[seg].ds_len;
   1661 			if (seg == 0)
   1662 				cmdstat |= RE_TDESC_CMD_SOF;
   1663 			else
   1664 				cmdstat |= RE_TDESC_CMD_OWN;
   1665 			if (curdesc == (RE_TX_DESC_CNT(sc) - 1))
   1666 				cmdstat |= RE_TDESC_CMD_EOR;
   1667 			if (seg == nsegs - 1) {
   1668 				cmdstat |= RE_TDESC_CMD_EOF;
   1669 				lastdesc = curdesc;
   1670 			}
   1671 			d->re_cmdstat = htole32(cmdstat);
   1672 			RE_TXDESCSYNC(sc, curdesc,
   1673 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1674 		}
   1675 		if (__predict_false(pad)) {
   1676 			bus_addr_t paddaddr;
   1677 
   1678 			d = &sc->re_ldata.re_tx_list[curdesc];
   1679 			d->re_vlanctl = 0;
   1680 			paddaddr = RE_TXPADDADDR(sc);
   1681 			re_set_bufaddr(d, paddaddr);
   1682 			cmdstat = re_flags |
   1683 			    RE_TDESC_CMD_OWN | RE_TDESC_CMD_EOF |
   1684 			    (RE_IP4CSUMTX_PADLEN + 1 - m->m_pkthdr.len);
   1685 			if (curdesc == (RE_TX_DESC_CNT(sc) - 1))
   1686 				cmdstat |= RE_TDESC_CMD_EOR;
   1687 			d->re_cmdstat = htole32(cmdstat);
   1688 			RE_TXDESCSYNC(sc, curdesc,
   1689 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1690 			lastdesc = curdesc;
   1691 			curdesc = RE_NEXT_TX_DESC(sc, curdesc);
   1692 		}
   1693 		KASSERT(lastdesc != -1);
   1694 
   1695 		/*
   1696 		 * Set up hardware VLAN tagging. Note: vlan tag info must
   1697 		 * appear in the first descriptor of a multi-descriptor
   1698 		 * transmission attempt.
   1699 		 */
   1700 		if ((mtag = VLAN_OUTPUT_TAG(&sc->ethercom, m)) != NULL) {
   1701 			sc->re_ldata.re_tx_list[startdesc].re_vlanctl =
   1702 			    htole32(bswap16(VLAN_TAG_VALUE(mtag)) |
   1703 			    RE_TDESC_VLANCTL_TAG);
   1704 		}
   1705 
   1706 		/* Transfer ownership of packet to the chip. */
   1707 
   1708 		sc->re_ldata.re_tx_list[startdesc].re_cmdstat |=
   1709 		    htole32(RE_TDESC_CMD_OWN);
   1710 		RE_TXDESCSYNC(sc, startdesc,
   1711 		    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1712 
   1713 		/* update info of TX queue and descriptors */
   1714 		txq->txq_mbuf = m;
   1715 		txq->txq_descidx = lastdesc;
   1716 		txq->txq_nsegs = nsegs;
   1717 
   1718 		sc->re_ldata.re_txq_free--;
   1719 		sc->re_ldata.re_tx_free -= nsegs;
   1720 		sc->re_ldata.re_tx_nextfree = curdesc;
   1721 
   1722 #if NBPFILTER > 0
   1723 		/*
   1724 		 * If there's a BPF listener, bounce a copy of this frame
   1725 		 * to him.
   1726 		 */
   1727 		if (ifp->if_bpf)
   1728 			bpf_mtap(ifp->if_bpf, m);
   1729 #endif
   1730 	}
   1731 
   1732 	if (sc->re_ldata.re_txq_free < ofree) {
   1733 		/*
   1734 		 * TX packets are enqueued.
   1735 		 */
   1736 		sc->re_ldata.re_txq_prodidx = idx;
   1737 
   1738 		/*
   1739 		 * Start the transmitter to poll.
   1740 		 *
   1741 		 * RealTek put the TX poll request register in a different
   1742 		 * location on the 8169 gigE chip. I don't know why.
   1743 		 */
   1744 		if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0)
   1745 			CSR_WRITE_1(sc, RTK_TXSTART, RTK_TXSTART_START);
   1746 		else
   1747 			CSR_WRITE_2(sc, RTK_GTXSTART, RTK_TXSTART_START);
   1748 
   1749 		/*
   1750 		 * Use the countdown timer for interrupt moderation.
   1751 		 * 'TX done' interrupts are disabled. Instead, we reset the
   1752 		 * countdown timer, which will begin counting until it hits
   1753 		 * the value in the TIMERINT register, and then trigger an
   1754 		 * interrupt. Each time we write to the TIMERCNT register,
   1755 		 * the timer count is reset to 0.
   1756 		 */
   1757 		CSR_WRITE_4(sc, RTK_TIMERCNT, 1);
   1758 
   1759 		/*
   1760 		 * Set a timeout in case the chip goes out to lunch.
   1761 		 */
   1762 		ifp->if_timer = 5;
   1763 	}
   1764 }
   1765 
   1766 static int
   1767 re_init(struct ifnet *ifp)
   1768 {
   1769 	struct rtk_softc	*sc = ifp->if_softc;
   1770 	const uint8_t		*enaddr;
   1771 	uint32_t		rxcfg = 0;
   1772 	uint32_t		reg;
   1773 	int error;
   1774 
   1775 	if ((error = re_enable(sc)) != 0)
   1776 		goto out;
   1777 
   1778 	/*
   1779 	 * Cancel pending I/O and free all RX/TX buffers.
   1780 	 */
   1781 	re_stop(ifp, 0);
   1782 
   1783 	re_reset(sc);
   1784 
   1785 	/*
   1786 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
   1787 	 * RX checksum offload. We must configure the C+ register
   1788 	 * before all others.
   1789 	 */
   1790 	reg = 0;
   1791 
   1792 	/*
   1793 	 * XXX: Realtek docs say bits 0 and 1 are reserved, for 8169S/8110S.
   1794 	 * FreeBSD  drivers set these bits anyway (for 8139C+?).
   1795 	 * So far, it works.
   1796 	 */
   1797 
   1798 	/*
   1799 	 * XXX: For old 8169 set bit 14.
   1800 	 *      For 8169S/8110S and above, do not set bit 14.
   1801 	 */
   1802 	if ((sc->sc_quirk & RTKQ_8169NONS) != 0)
   1803 		reg |= (0x1 << 14) | RTK_CPLUSCMD_PCI_MRW;;
   1804 
   1805 	if (1)  {/* not for 8169S ? */
   1806 		reg |=
   1807 		    RTK_CPLUSCMD_VLANSTRIP |
   1808 		    (ifp->if_capenable &
   1809 		    (IFCAP_CSUM_IPv4_Rx | IFCAP_CSUM_TCPv4_Rx |
   1810 		     IFCAP_CSUM_UDPv4_Rx) ?
   1811 		    RTK_CPLUSCMD_RXCSUM_ENB : 0);
   1812 	}
   1813 
   1814 	CSR_WRITE_2(sc, RTK_CPLUS_CMD,
   1815 	    reg | RTK_CPLUSCMD_RXENB | RTK_CPLUSCMD_TXENB);
   1816 
   1817 	/* XXX: from Realtek-supplied Linux driver. Wholly undocumented. */
   1818 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0)
   1819 		CSR_WRITE_2(sc, RTK_IM, 0x0000);
   1820 
   1821 	DELAY(10000);
   1822 
   1823 	/*
   1824 	 * Init our MAC address.  Even though the chipset
   1825 	 * documentation doesn't mention it, we need to enter "Config
   1826 	 * register write enable" mode to modify the ID registers.
   1827 	 */
   1828 	CSR_WRITE_1(sc, RTK_EECMD, RTK_EEMODE_WRITECFG);
   1829 	enaddr = CLLADDR(ifp->if_sadl);
   1830 	reg = enaddr[0] | (enaddr[1] << 8) |
   1831 	    (enaddr[2] << 16) | (enaddr[3] << 24);
   1832 	CSR_WRITE_4(sc, RTK_IDR0, reg);
   1833 	reg = enaddr[4] | (enaddr[5] << 8);
   1834 	CSR_WRITE_4(sc, RTK_IDR4, reg);
   1835 	CSR_WRITE_1(sc, RTK_EECMD, RTK_EEMODE_OFF);
   1836 
   1837 	/*
   1838 	 * For C+ mode, initialize the RX descriptors and mbufs.
   1839 	 */
   1840 	re_rx_list_init(sc);
   1841 	re_tx_list_init(sc);
   1842 
   1843 	/*
   1844 	 * Load the addresses of the RX and TX lists into the chip.
   1845 	 */
   1846 	CSR_WRITE_4(sc, RTK_RXLIST_ADDR_HI,
   1847 	    RE_ADDR_HI(sc->re_ldata.re_rx_list_map->dm_segs[0].ds_addr));
   1848 	CSR_WRITE_4(sc, RTK_RXLIST_ADDR_LO,
   1849 	    RE_ADDR_LO(sc->re_ldata.re_rx_list_map->dm_segs[0].ds_addr));
   1850 
   1851 	CSR_WRITE_4(sc, RTK_TXLIST_ADDR_HI,
   1852 	    RE_ADDR_HI(sc->re_ldata.re_tx_list_map->dm_segs[0].ds_addr));
   1853 	CSR_WRITE_4(sc, RTK_TXLIST_ADDR_LO,
   1854 	    RE_ADDR_LO(sc->re_ldata.re_tx_list_map->dm_segs[0].ds_addr));
   1855 
   1856 	/*
   1857 	 * Enable transmit and receive.
   1858 	 */
   1859 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_TX_ENB | RTK_CMD_RX_ENB);
   1860 
   1861 	/*
   1862 	 * Set the initial TX and RX configuration.
   1863 	 */
   1864 	if (sc->re_testmode && (sc->sc_quirk & RTKQ_8169NONS) != 0) {
   1865 		/* test mode is needed only for old 8169 */
   1866 		CSR_WRITE_4(sc, RTK_TXCFG,
   1867 		    RE_TXCFG_CONFIG | RTK_LOOPTEST_ON);
   1868 	} else
   1869 		CSR_WRITE_4(sc, RTK_TXCFG, RE_TXCFG_CONFIG);
   1870 
   1871 	CSR_WRITE_1(sc, RTK_EARLY_TX_THRESH, 16);
   1872 
   1873 	CSR_WRITE_4(sc, RTK_RXCFG, RE_RXCFG_CONFIG);
   1874 
   1875 	/* Set the individual bit to receive frames for this host only. */
   1876 	rxcfg = CSR_READ_4(sc, RTK_RXCFG);
   1877 	rxcfg |= RTK_RXCFG_RX_INDIV;
   1878 
   1879 	/* If we want promiscuous mode, set the allframes bit. */
   1880 	if (ifp->if_flags & IFF_PROMISC)
   1881 		rxcfg |= RTK_RXCFG_RX_ALLPHYS;
   1882 	else
   1883 		rxcfg &= ~RTK_RXCFG_RX_ALLPHYS;
   1884 	CSR_WRITE_4(sc, RTK_RXCFG, rxcfg);
   1885 
   1886 	/*
   1887 	 * Set capture broadcast bit to capture broadcast frames.
   1888 	 */
   1889 	if (ifp->if_flags & IFF_BROADCAST)
   1890 		rxcfg |= RTK_RXCFG_RX_BROAD;
   1891 	else
   1892 		rxcfg &= ~RTK_RXCFG_RX_BROAD;
   1893 	CSR_WRITE_4(sc, RTK_RXCFG, rxcfg);
   1894 
   1895 	/*
   1896 	 * Program the multicast filter, if necessary.
   1897 	 */
   1898 	rtk_setmulti(sc);
   1899 
   1900 #ifdef DEVICE_POLLING
   1901 	/*
   1902 	 * Disable interrupts if we are polling.
   1903 	 */
   1904 	if (ifp->if_flags & IFF_POLLING)
   1905 		CSR_WRITE_2(sc, RTK_IMR, 0);
   1906 	else	/* otherwise ... */
   1907 #endif /* DEVICE_POLLING */
   1908 	/*
   1909 	 * Enable interrupts.
   1910 	 */
   1911 	if (sc->re_testmode)
   1912 		CSR_WRITE_2(sc, RTK_IMR, 0);
   1913 	else
   1914 		CSR_WRITE_2(sc, RTK_IMR, RTK_INTRS_CPLUS);
   1915 
   1916 	/* Start RX/TX process. */
   1917 	CSR_WRITE_4(sc, RTK_MISSEDPKT, 0);
   1918 #ifdef notdef
   1919 	/* Enable receiver and transmitter. */
   1920 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_TX_ENB | RTK_CMD_RX_ENB);
   1921 #endif
   1922 
   1923 	/*
   1924 	 * Initialize the timer interrupt register so that
   1925 	 * a timer interrupt will be generated once the timer
   1926 	 * reaches a certain number of ticks. The timer is
   1927 	 * reloaded on each transmit. This gives us TX interrupt
   1928 	 * moderation, which dramatically improves TX frame rate.
   1929 	 */
   1930 
   1931 	if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0)
   1932 		CSR_WRITE_4(sc, RTK_TIMERINT, 0x400);
   1933 	else {
   1934 		CSR_WRITE_4(sc, RTK_TIMERINT_8169, 0x800);
   1935 
   1936 		/*
   1937 		 * For 8169 gigE NICs, set the max allowed RX packet
   1938 		 * size so we can receive jumbo frames.
   1939 		 */
   1940 		CSR_WRITE_2(sc, RTK_MAXRXPKTLEN, 16383);
   1941 	}
   1942 
   1943 	if (sc->re_testmode)
   1944 		return 0;
   1945 
   1946 	CSR_WRITE_1(sc, RTK_CFG1, RTK_CFG1_DRVLOAD);
   1947 
   1948 	ifp->if_flags |= IFF_RUNNING;
   1949 	ifp->if_flags &= ~IFF_OACTIVE;
   1950 
   1951 	callout_reset(&sc->rtk_tick_ch, hz, re_tick, sc);
   1952 
   1953  out:
   1954 	if (error) {
   1955 		ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
   1956 		ifp->if_timer = 0;
   1957 		aprint_error("%s: interface not running\n",
   1958 		    sc->sc_dev.dv_xname);
   1959 	}
   1960 
   1961 	return error;
   1962 }
   1963 
   1964 /*
   1965  * Set media options.
   1966  */
   1967 static int
   1968 re_ifmedia_upd(struct ifnet *ifp)
   1969 {
   1970 	struct rtk_softc	*sc;
   1971 
   1972 	sc = ifp->if_softc;
   1973 
   1974 	return mii_mediachg(&sc->mii);
   1975 }
   1976 
   1977 /*
   1978  * Report current media status.
   1979  */
   1980 static void
   1981 re_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
   1982 {
   1983 	struct rtk_softc	*sc;
   1984 
   1985 	sc = ifp->if_softc;
   1986 
   1987 	mii_pollstat(&sc->mii);
   1988 	ifmr->ifm_active = sc->mii.mii_media_active;
   1989 	ifmr->ifm_status = sc->mii.mii_media_status;
   1990 }
   1991 
   1992 static int
   1993 re_ioctl(struct ifnet *ifp, u_long command, void *data)
   1994 {
   1995 	struct rtk_softc	*sc = ifp->if_softc;
   1996 	struct ifreq		*ifr = (struct ifreq *) data;
   1997 	int			s, error = 0;
   1998 
   1999 	s = splnet();
   2000 
   2001 	switch (command) {
   2002 	case SIOCSIFMTU:
   2003 		if (ifr->ifr_mtu > RE_JUMBO_MTU)
   2004 			error = EINVAL;
   2005 		ifp->if_mtu = ifr->ifr_mtu;
   2006 		break;
   2007 	case SIOCGIFMEDIA:
   2008 	case SIOCSIFMEDIA:
   2009 		error = ifmedia_ioctl(ifp, ifr, &sc->mii.mii_media, command);
   2010 		break;
   2011 	default:
   2012 		error = ether_ioctl(ifp, command, data);
   2013 		if (error == ENETRESET) {
   2014 			if (ifp->if_flags & IFF_RUNNING)
   2015 				rtk_setmulti(sc);
   2016 			error = 0;
   2017 		}
   2018 		break;
   2019 	}
   2020 
   2021 	splx(s);
   2022 
   2023 	return error;
   2024 }
   2025 
   2026 static void
   2027 re_watchdog(struct ifnet *ifp)
   2028 {
   2029 	struct rtk_softc	*sc;
   2030 	int			s;
   2031 
   2032 	sc = ifp->if_softc;
   2033 	s = splnet();
   2034 	aprint_error("%s: watchdog timeout\n", sc->sc_dev.dv_xname);
   2035 	ifp->if_oerrors++;
   2036 
   2037 	re_txeof(sc);
   2038 	re_rxeof(sc);
   2039 
   2040 	re_init(ifp);
   2041 
   2042 	splx(s);
   2043 }
   2044 
   2045 /*
   2046  * Stop the adapter and free any mbufs allocated to the
   2047  * RX and TX lists.
   2048  */
   2049 static void
   2050 re_stop(struct ifnet *ifp, int disable)
   2051 {
   2052 	int		i;
   2053 	struct rtk_softc *sc = ifp->if_softc;
   2054 
   2055 	callout_stop(&sc->rtk_tick_ch);
   2056 
   2057 #ifdef DEVICE_POLLING
   2058 	ether_poll_deregister(ifp);
   2059 #endif /* DEVICE_POLLING */
   2060 
   2061 	mii_down(&sc->mii);
   2062 
   2063 	CSR_WRITE_1(sc, RTK_COMMAND, 0x00);
   2064 	CSR_WRITE_2(sc, RTK_IMR, 0x0000);
   2065 
   2066 	if (sc->re_head != NULL) {
   2067 		m_freem(sc->re_head);
   2068 		sc->re_head = sc->re_tail = NULL;
   2069 	}
   2070 
   2071 	/* Free the TX list buffers. */
   2072 	for (i = 0; i < RE_TX_QLEN; i++) {
   2073 		if (sc->re_ldata.re_txq[i].txq_mbuf != NULL) {
   2074 			bus_dmamap_unload(sc->sc_dmat,
   2075 			    sc->re_ldata.re_txq[i].txq_dmamap);
   2076 			m_freem(sc->re_ldata.re_txq[i].txq_mbuf);
   2077 			sc->re_ldata.re_txq[i].txq_mbuf = NULL;
   2078 		}
   2079 	}
   2080 
   2081 	/* Free the RX list buffers. */
   2082 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
   2083 		if (sc->re_ldata.re_rxsoft[i].rxs_mbuf != NULL) {
   2084 			bus_dmamap_unload(sc->sc_dmat,
   2085 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
   2086 			m_freem(sc->re_ldata.re_rxsoft[i].rxs_mbuf);
   2087 			sc->re_ldata.re_rxsoft[i].rxs_mbuf = NULL;
   2088 		}
   2089 	}
   2090 
   2091 	if (disable)
   2092 		re_disable(sc);
   2093 
   2094 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
   2095 	ifp->if_timer = 0;
   2096 }
   2097