Home | History | Annotate | Line # | Download | only in ic
rtl8169.c revision 1.89.2.1
      1 /*	$NetBSD: rtl8169.c,v 1.89.2.1 2008/02/18 21:05:41 mjf 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 __KERNEL_RCSID(0, "$NetBSD: rtl8169.c,v 1.89.2.1 2008/02/18 21:05:41 mjf Exp $");
     37 /* $FreeBSD: /repoman/r/ncvs/src/sys/dev/re/if_re.c,v 1.20 2004/04/11 20:34:08 ru Exp $ */
     38 
     39 /*
     40  * RealTek 8139C+/8169/8169S/8110S PCI NIC driver
     41  *
     42  * Written by Bill Paul <wpaul (at) windriver.com>
     43  * Senior Networking Software Engineer
     44  * Wind River Systems
     45  */
     46 
     47 /*
     48  * This driver is designed to support RealTek's next generation of
     49  * 10/100 and 10/100/1000 PCI ethernet controllers. There are currently
     50  * four devices in this family: the RTL8139C+, the RTL8169, the RTL8169S
     51  * and the RTL8110S.
     52  *
     53  * The 8139C+ is a 10/100 ethernet chip. It is backwards compatible
     54  * with the older 8139 family, however it also supports a special
     55  * C+ mode of operation that provides several new performance enhancing
     56  * features. These include:
     57  *
     58  *	o Descriptor based DMA mechanism. Each descriptor represents
     59  *	  a single packet fragment. Data buffers may be aligned on
     60  *	  any byte boundary.
     61  *
     62  *	o 64-bit DMA
     63  *
     64  *	o TCP/IP checksum offload for both RX and TX
     65  *
     66  *	o High and normal priority transmit DMA rings
     67  *
     68  *	o VLAN tag insertion and extraction
     69  *
     70  *	o TCP large send (segmentation offload)
     71  *
     72  * Like the 8139, the 8139C+ also has a built-in 10/100 PHY. The C+
     73  * programming API is fairly straightforward. The RX filtering, EEPROM
     74  * access and PHY access is the same as it is on the older 8139 series
     75  * chips.
     76  *
     77  * The 8169 is a 64-bit 10/100/1000 gigabit ethernet MAC. It has almost the
     78  * same programming API and feature set as the 8139C+ with the following
     79  * differences and additions:
     80  *
     81  *	o 1000Mbps mode
     82  *
     83  *	o Jumbo frames
     84  *
     85  * 	o GMII and TBI ports/registers for interfacing with copper
     86  *	  or fiber PHYs
     87  *
     88  *      o RX and TX DMA rings can have up to 1024 descriptors
     89  *        (the 8139C+ allows a maximum of 64)
     90  *
     91  *	o Slight differences in register layout from the 8139C+
     92  *
     93  * The TX start and timer interrupt registers are at different locations
     94  * on the 8169 than they are on the 8139C+. Also, the status word in the
     95  * RX descriptor has a slightly different bit layout. The 8169 does not
     96  * have a built-in PHY. Most reference boards use a Marvell 88E1000 'Alaska'
     97  * copper gigE PHY.
     98  *
     99  * The 8169S/8110S 10/100/1000 devices have built-in copper gigE PHYs
    100  * (the 'S' stands for 'single-chip'). These devices have the same
    101  * programming API as the older 8169, but also have some vendor-specific
    102  * registers for the on-board PHY. The 8110S is a LAN-on-motherboard
    103  * part designed to be pin-compatible with the RealTek 8100 10/100 chip.
    104  *
    105  * This driver takes advantage of the RX and TX checksum offload and
    106  * VLAN tag insertion/extraction features. It also implements TX
    107  * interrupt moderation using the timer interrupt registers, which
    108  * significantly reduces TX interrupt load. There is also support
    109  * for jumbo frames, however the 8169/8169S/8110S can not transmit
    110  * jumbo frames larger than 7.5K, so the max MTU possible with this
    111  * driver is 7500 bytes.
    112  */
    113 
    114 #include "bpfilter.h"
    115 #include "vlan.h"
    116 
    117 #include <sys/param.h>
    118 #include <sys/endian.h>
    119 #include <sys/systm.h>
    120 #include <sys/sockio.h>
    121 #include <sys/mbuf.h>
    122 #include <sys/malloc.h>
    123 #include <sys/kernel.h>
    124 #include <sys/socket.h>
    125 #include <sys/device.h>
    126 
    127 #include <net/if.h>
    128 #include <net/if_arp.h>
    129 #include <net/if_dl.h>
    130 #include <net/if_ether.h>
    131 #include <net/if_media.h>
    132 #include <net/if_vlanvar.h>
    133 
    134 #include <netinet/in_systm.h>	/* XXX for IP_MAXPACKET */
    135 #include <netinet/in.h>		/* XXX for IP_MAXPACKET */
    136 #include <netinet/ip.h>		/* XXX for IP_MAXPACKET */
    137 
    138 #if NBPFILTER > 0
    139 #include <net/bpf.h>
    140 #endif
    141 
    142 #include <sys/bus.h>
    143 
    144 #include <dev/mii/mii.h>
    145 #include <dev/mii/miivar.h>
    146 
    147 #include <dev/ic/rtl81x9reg.h>
    148 #include <dev/ic/rtl81x9var.h>
    149 
    150 #include <dev/ic/rtl8169var.h>
    151 
    152 static inline void re_set_bufaddr(struct re_desc *, bus_addr_t);
    153 
    154 static int re_newbuf(struct rtk_softc *, int, struct mbuf *);
    155 static int re_rx_list_init(struct rtk_softc *);
    156 static int re_tx_list_init(struct rtk_softc *);
    157 static void re_rxeof(struct rtk_softc *);
    158 static void re_txeof(struct rtk_softc *);
    159 static void re_tick(void *);
    160 static void re_start(struct ifnet *);
    161 static int re_ioctl(struct ifnet *, u_long, void *);
    162 static int re_init(struct ifnet *);
    163 static void re_stop(struct ifnet *, int);
    164 static void re_watchdog(struct ifnet *);
    165 
    166 static void re_shutdown(void *);
    167 static int re_enable(struct rtk_softc *);
    168 static void re_disable(struct rtk_softc *);
    169 static void re_power(int, void *);
    170 
    171 static int re_gmii_readreg(struct device *, int, int);
    172 static void re_gmii_writereg(struct device *, int, int, int);
    173 
    174 static int re_miibus_readreg(struct device *, int, int);
    175 static void re_miibus_writereg(struct device *, int, int, int);
    176 static void re_miibus_statchg(struct device *);
    177 
    178 static void re_reset(struct rtk_softc *);
    179 
    180 static inline void
    181 re_set_bufaddr(struct re_desc *d, bus_addr_t addr)
    182 {
    183 
    184 	d->re_bufaddr_lo = htole32((uint32_t)addr);
    185 	if (sizeof(bus_addr_t) == sizeof(uint64_t))
    186 		d->re_bufaddr_hi = htole32((uint64_t)addr >> 32);
    187 	else
    188 		d->re_bufaddr_hi = 0;
    189 }
    190 
    191 static int
    192 re_gmii_readreg(struct device *self, int phy, int reg)
    193 {
    194 	struct rtk_softc	*sc = (void *)self;
    195 	uint32_t		rval;
    196 	int			i;
    197 
    198 	if (phy != 7)
    199 		return 0;
    200 
    201 	/* Let the rgephy driver read the GMEDIASTAT register */
    202 
    203 	if (reg == RTK_GMEDIASTAT) {
    204 		rval = CSR_READ_1(sc, RTK_GMEDIASTAT);
    205 		return rval;
    206 	}
    207 
    208 	CSR_WRITE_4(sc, RTK_PHYAR, reg << 16);
    209 	DELAY(1000);
    210 
    211 	for (i = 0; i < RTK_TIMEOUT; i++) {
    212 		rval = CSR_READ_4(sc, RTK_PHYAR);
    213 		if (rval & RTK_PHYAR_BUSY)
    214 			break;
    215 		DELAY(100);
    216 	}
    217 
    218 	if (i == RTK_TIMEOUT) {
    219 		aprint_error("%s: PHY read failed\n", sc->sc_dev.dv_xname);
    220 		return 0;
    221 	}
    222 
    223 	return rval & RTK_PHYAR_PHYDATA;
    224 }
    225 
    226 static void
    227 re_gmii_writereg(struct device *dev, int phy, int reg, int data)
    228 {
    229 	struct rtk_softc	*sc = (void *)dev;
    230 	uint32_t		rval;
    231 	int			i;
    232 
    233 	CSR_WRITE_4(sc, RTK_PHYAR, (reg << 16) |
    234 	    (data & RTK_PHYAR_PHYDATA) | RTK_PHYAR_BUSY);
    235 	DELAY(1000);
    236 
    237 	for (i = 0; i < RTK_TIMEOUT; i++) {
    238 		rval = CSR_READ_4(sc, RTK_PHYAR);
    239 		if (!(rval & RTK_PHYAR_BUSY))
    240 			break;
    241 		DELAY(100);
    242 	}
    243 
    244 	if (i == RTK_TIMEOUT) {
    245 		aprint_error("%s: PHY write reg %x <- %x failed\n",
    246 		    sc->sc_dev.dv_xname, reg, data);
    247 	}
    248 }
    249 
    250 static int
    251 re_miibus_readreg(struct device *dev, int phy, int reg)
    252 {
    253 	struct rtk_softc	*sc = (void *)dev;
    254 	uint16_t		rval = 0;
    255 	uint16_t		re8139_reg = 0;
    256 	int			s;
    257 
    258 	s = splnet();
    259 
    260 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0) {
    261 		rval = re_gmii_readreg(dev, phy, reg);
    262 		splx(s);
    263 		return rval;
    264 	}
    265 
    266 	/* Pretend the internal PHY is only at address 0 */
    267 	if (phy) {
    268 		splx(s);
    269 		return 0;
    270 	}
    271 	switch (reg) {
    272 	case MII_BMCR:
    273 		re8139_reg = RTK_BMCR;
    274 		break;
    275 	case MII_BMSR:
    276 		re8139_reg = RTK_BMSR;
    277 		break;
    278 	case MII_ANAR:
    279 		re8139_reg = RTK_ANAR;
    280 		break;
    281 	case MII_ANER:
    282 		re8139_reg = RTK_ANER;
    283 		break;
    284 	case MII_ANLPAR:
    285 		re8139_reg = RTK_LPAR;
    286 		break;
    287 	case MII_PHYIDR1:
    288 	case MII_PHYIDR2:
    289 		splx(s);
    290 		return 0;
    291 	/*
    292 	 * Allow the rlphy driver to read the media status
    293 	 * register. If we have a link partner which does not
    294 	 * support NWAY, this is the register which will tell
    295 	 * us the results of parallel detection.
    296 	 */
    297 	case RTK_MEDIASTAT:
    298 		rval = CSR_READ_1(sc, RTK_MEDIASTAT);
    299 		splx(s);
    300 		return rval;
    301 	default:
    302 		aprint_error("%s: bad phy register\n", sc->sc_dev.dv_xname);
    303 		splx(s);
    304 		return 0;
    305 	}
    306 	rval = CSR_READ_2(sc, re8139_reg);
    307 	if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0 && re8139_reg == RTK_BMCR) {
    308 		/* 8139C+ has different bit layout. */
    309 		rval &= ~(BMCR_LOOP | BMCR_ISO);
    310 	}
    311 	splx(s);
    312 	return rval;
    313 }
    314 
    315 static void
    316 re_miibus_writereg(struct device *dev, int phy, int reg, int data)
    317 {
    318 	struct rtk_softc	*sc = (void *)dev;
    319 	uint16_t		re8139_reg = 0;
    320 	int			s;
    321 
    322 	s = splnet();
    323 
    324 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0) {
    325 		re_gmii_writereg(dev, phy, reg, data);
    326 		splx(s);
    327 		return;
    328 	}
    329 
    330 	/* Pretend the internal PHY is only at address 0 */
    331 	if (phy) {
    332 		splx(s);
    333 		return;
    334 	}
    335 	switch (reg) {
    336 	case MII_BMCR:
    337 		re8139_reg = RTK_BMCR;
    338 		if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0) {
    339 			/* 8139C+ has different bit layout. */
    340 			data &= ~(BMCR_LOOP | BMCR_ISO);
    341 		}
    342 		break;
    343 	case MII_BMSR:
    344 		re8139_reg = RTK_BMSR;
    345 		break;
    346 	case MII_ANAR:
    347 		re8139_reg = RTK_ANAR;
    348 		break;
    349 	case MII_ANER:
    350 		re8139_reg = RTK_ANER;
    351 		break;
    352 	case MII_ANLPAR:
    353 		re8139_reg = RTK_LPAR;
    354 		break;
    355 	case MII_PHYIDR1:
    356 	case MII_PHYIDR2:
    357 		splx(s);
    358 		return;
    359 		break;
    360 	default:
    361 		aprint_error("%s: bad phy register\n", sc->sc_dev.dv_xname);
    362 		splx(s);
    363 		return;
    364 	}
    365 	CSR_WRITE_2(sc, re8139_reg, data);
    366 	splx(s);
    367 	return;
    368 }
    369 
    370 static void
    371 re_miibus_statchg(struct device *dev)
    372 {
    373 
    374 	return;
    375 }
    376 
    377 static void
    378 re_reset(struct rtk_softc *sc)
    379 {
    380 	int		i;
    381 
    382 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_RESET);
    383 
    384 	for (i = 0; i < RTK_TIMEOUT; i++) {
    385 		DELAY(10);
    386 		if ((CSR_READ_1(sc, RTK_COMMAND) & RTK_CMD_RESET) == 0)
    387 			break;
    388 	}
    389 	if (i == RTK_TIMEOUT)
    390 		aprint_error("%s: reset never completed!\n",
    391 		    sc->sc_dev.dv_xname);
    392 
    393 	/*
    394 	 * NB: Realtek-supplied Linux driver does this only for
    395 	 * MCFG_METHOD_2, which corresponds to sc->sc_rev == 2.
    396 	 */
    397 	if (1) /* XXX check softc flag for 8169s version */
    398 		CSR_WRITE_1(sc, RTK_LDPS, 1);
    399 
    400 	return;
    401 }
    402 
    403 /*
    404  * The following routine is designed to test for a defect on some
    405  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
    406  * lines connected to the bus, however for a 32-bit only card, they
    407  * should be pulled high. The result of this defect is that the
    408  * NIC will not work right if you plug it into a 64-bit slot: DMA
    409  * operations will be done with 64-bit transfers, which will fail
    410  * because the 64-bit data lines aren't connected.
    411  *
    412  * There's no way to work around this (short of talking a soldering
    413  * iron to the board), however we can detect it. The method we use
    414  * here is to put the NIC into digital loopback mode, set the receiver
    415  * to promiscuous mode, and then try to send a frame. We then compare
    416  * the frame data we sent to what was received. If the data matches,
    417  * then the NIC is working correctly, otherwise we know the user has
    418  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
    419  * slot. In the latter case, there's no way the NIC can work correctly,
    420  * so we print out a message on the console and abort the device attach.
    421  */
    422 
    423 int
    424 re_diag(struct rtk_softc *sc)
    425 {
    426 	struct ifnet		*ifp = &sc->ethercom.ec_if;
    427 	struct mbuf		*m0;
    428 	struct ether_header	*eh;
    429 	struct re_rxsoft	*rxs;
    430 	struct re_desc		*cur_rx;
    431 	bus_dmamap_t		dmamap;
    432 	uint16_t		status;
    433 	uint32_t		rxstat;
    434 	int			total_len, i, s, error = 0;
    435 	static const uint8_t	dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
    436 	static const uint8_t	src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
    437 
    438 	/* Allocate a single mbuf */
    439 
    440 	MGETHDR(m0, M_DONTWAIT, MT_DATA);
    441 	if (m0 == NULL)
    442 		return ENOBUFS;
    443 
    444 	/*
    445 	 * Initialize the NIC in test mode. This sets the chip up
    446 	 * so that it can send and receive frames, but performs the
    447 	 * following special functions:
    448 	 * - Puts receiver in promiscuous mode
    449 	 * - Enables digital loopback mode
    450 	 * - Leaves interrupts turned off
    451 	 */
    452 
    453 	ifp->if_flags |= IFF_PROMISC;
    454 	sc->re_testmode = 1;
    455 	re_init(ifp);
    456 	re_stop(ifp, 0);
    457 	DELAY(100000);
    458 	re_init(ifp);
    459 
    460 	/* Put some data in the mbuf */
    461 
    462 	eh = mtod(m0, struct ether_header *);
    463 	memcpy(eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN);
    464 	memcpy(eh->ether_shost, (char *)&src, ETHER_ADDR_LEN);
    465 	eh->ether_type = htons(ETHERTYPE_IP);
    466 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
    467 
    468 	/*
    469 	 * Queue the packet, start transmission.
    470 	 */
    471 
    472 	CSR_WRITE_2(sc, RTK_ISR, 0xFFFF);
    473 	s = splnet();
    474 	IF_ENQUEUE(&ifp->if_snd, m0);
    475 	re_start(ifp);
    476 	splx(s);
    477 	m0 = NULL;
    478 
    479 	/* Wait for it to propagate through the chip */
    480 
    481 	DELAY(100000);
    482 	for (i = 0; i < RTK_TIMEOUT; i++) {
    483 		status = CSR_READ_2(sc, RTK_ISR);
    484 		if ((status & (RTK_ISR_TIMEOUT_EXPIRED | RTK_ISR_RX_OK)) ==
    485 		    (RTK_ISR_TIMEOUT_EXPIRED | RTK_ISR_RX_OK))
    486 			break;
    487 		DELAY(10);
    488 	}
    489 	if (i == RTK_TIMEOUT) {
    490 		aprint_error("%s: diagnostic failed, failed to receive packet "
    491 		    "in loopback mode\n", sc->sc_dev.dv_xname);
    492 		error = EIO;
    493 		goto done;
    494 	}
    495 
    496 	/*
    497 	 * The packet should have been dumped into the first
    498 	 * entry in the RX DMA ring. Grab it from there.
    499 	 */
    500 
    501 	rxs = &sc->re_ldata.re_rxsoft[0];
    502 	dmamap = rxs->rxs_dmamap;
    503 	bus_dmamap_sync(sc->sc_dmat, dmamap, 0, dmamap->dm_mapsize,
    504 	    BUS_DMASYNC_POSTREAD);
    505 	bus_dmamap_unload(sc->sc_dmat, dmamap);
    506 
    507 	m0 = rxs->rxs_mbuf;
    508 	rxs->rxs_mbuf = NULL;
    509 	eh = mtod(m0, struct ether_header *);
    510 
    511 	RE_RXDESCSYNC(sc, 0, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
    512 	cur_rx = &sc->re_ldata.re_rx_list[0];
    513 	rxstat = le32toh(cur_rx->re_cmdstat);
    514 	total_len = rxstat & sc->re_rxlenmask;
    515 
    516 	if (total_len != ETHER_MIN_LEN) {
    517 		aprint_error("%s: diagnostic failed, received short packet\n",
    518 		    sc->sc_dev.dv_xname);
    519 		error = EIO;
    520 		goto done;
    521 	}
    522 
    523 	/* Test that the received packet data matches what we sent. */
    524 
    525 	if (memcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
    526 	    memcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
    527 	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
    528 		aprint_error("%s: WARNING, DMA FAILURE!\n",
    529 		    sc->sc_dev.dv_xname);
    530 		aprint_error("%s: expected TX data: %s",
    531 		    sc->sc_dev.dv_xname, ether_sprintf(dst));
    532 		aprint_error("/%s/0x%x\n", ether_sprintf(src), ETHERTYPE_IP);
    533 		aprint_error("%s: received RX data: %s",
    534 		    sc->sc_dev.dv_xname,
    535 		    ether_sprintf(eh->ether_dhost));
    536 		aprint_error("/%s/0x%x\n", ether_sprintf(eh->ether_shost),
    537 		    ntohs(eh->ether_type));
    538 		aprint_error("%s: You may have a defective 32-bit NIC plugged "
    539 		    "into a 64-bit PCI slot.\n", sc->sc_dev.dv_xname);
    540 		aprint_error("%s: Please re-install the NIC in a 32-bit slot "
    541 		    "for proper operation.\n", sc->sc_dev.dv_xname);
    542 		aprint_error("%s: Read the re(4) man page for more details.\n",
    543 		    sc->sc_dev.dv_xname);
    544 		error = EIO;
    545 	}
    546 
    547  done:
    548 	/* Turn interface off, release resources */
    549 
    550 	sc->re_testmode = 0;
    551 	ifp->if_flags &= ~IFF_PROMISC;
    552 	re_stop(ifp, 0);
    553 	if (m0 != NULL)
    554 		m_freem(m0);
    555 
    556 	return error;
    557 }
    558 
    559 
    560 /*
    561  * Attach the interface. Allocate softc structures, do ifmedia
    562  * setup and ethernet/BPF attach.
    563  */
    564 void
    565 re_attach(struct rtk_softc *sc)
    566 {
    567 	u_char			eaddr[ETHER_ADDR_LEN];
    568 	uint16_t		val;
    569 	struct ifnet		*ifp;
    570 	int			error = 0, i, addr_len;
    571 
    572 	/* Reset the adapter. */
    573 	re_reset(sc);
    574 
    575 	if (rtk_read_eeprom(sc, RTK_EE_ID, RTK_EEADDR_LEN1) == 0x8129)
    576 		addr_len = RTK_EEADDR_LEN1;
    577 	else
    578 		addr_len = RTK_EEADDR_LEN0;
    579 
    580 	/*
    581 	 * Get station address from the EEPROM.
    582 	 */
    583 	for (i = 0; i < 3; i++) {
    584 		val = rtk_read_eeprom(sc, RTK_EE_EADDR0 + i, addr_len);
    585 		eaddr[(i * 2) + 0] = val & 0xff;
    586 		eaddr[(i * 2) + 1] = val >> 8;
    587 	}
    588 
    589 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0) {
    590 		uint32_t hwrev;
    591 
    592 		/* Revision of 8169/8169S/8110s in bits 30..26, 23 */
    593 		hwrev = CSR_READ_4(sc, RTK_TXCFG) & RTK_TXCFG_HWREV;
    594 		/* These rev numbers are taken from Realtek's driver */
    595 		if (       hwrev == RTK_HWREV_8100E_SPIN2) {
    596 			sc->sc_rev = 15;
    597 		} else if (hwrev == RTK_HWREV_8100E) {
    598 			sc->sc_rev = 14;
    599 		} else if (hwrev == RTK_HWREV_8101E) {
    600 			sc->sc_rev = 13;
    601 		} else if (hwrev == RTK_HWREV_8168_SPIN2 ||
    602 		           hwrev == RTK_HWREV_8168_SPIN3) {
    603 			sc->sc_rev = 12;
    604 		} else if (hwrev == RTK_HWREV_8168_SPIN1) {
    605 			sc->sc_rev = 11;
    606 		} else if (hwrev == RTK_HWREV_8169_8110SC) {
    607 			sc->sc_rev = 5;
    608 		} else if (hwrev == RTK_HWREV_8169_8110SB) {
    609 			sc->sc_rev = 4;
    610 		} else if (hwrev == RTK_HWREV_8169S) {
    611 			sc->sc_rev = 3;
    612 		} else if (hwrev == RTK_HWREV_8110S) {
    613 			sc->sc_rev = 2;
    614 		} else if (hwrev == RTK_HWREV_8169) {
    615 			sc->sc_rev = 1;
    616 			sc->sc_quirk |= RTKQ_8169NONS;
    617 		} else {
    618 			aprint_normal("%s: Unknown revision (0x%08x)\n",
    619 			    sc->sc_dev.dv_xname, hwrev);
    620 			/* assume the latest one */
    621 			sc->sc_rev = 15;
    622 		}
    623 
    624 		/* Set RX length mask */
    625 		sc->re_rxlenmask = RE_RDESC_STAT_GFRAGLEN;
    626 		sc->re_ldata.re_tx_desc_cnt = RE_TX_DESC_CNT_8169;
    627 	} else {
    628 		/* Set RX length mask */
    629 		sc->re_rxlenmask = RE_RDESC_STAT_FRAGLEN;
    630 		sc->re_ldata.re_tx_desc_cnt = RE_TX_DESC_CNT_8139;
    631 	}
    632 
    633 	aprint_normal("%s: Ethernet address %s\n",
    634 	    sc->sc_dev.dv_xname, ether_sprintf(eaddr));
    635 
    636 	if (sc->re_ldata.re_tx_desc_cnt >
    637 	    PAGE_SIZE / sizeof(struct re_desc)) {
    638 		sc->re_ldata.re_tx_desc_cnt =
    639 		    PAGE_SIZE / sizeof(struct re_desc);
    640 	}
    641 
    642 	aprint_verbose("%s: using %d tx descriptors\n",
    643 	    sc->sc_dev.dv_xname, sc->re_ldata.re_tx_desc_cnt);
    644 	KASSERT(RE_NEXT_TX_DESC(sc, RE_TX_DESC_CNT(sc) - 1) == 0);
    645 
    646 	/* Allocate DMA'able memory for the TX ring */
    647 	if ((error = bus_dmamem_alloc(sc->sc_dmat, RE_TX_LIST_SZ(sc),
    648 	    RE_RING_ALIGN, 0, &sc->re_ldata.re_tx_listseg, 1,
    649 	    &sc->re_ldata.re_tx_listnseg, BUS_DMA_NOWAIT)) != 0) {
    650 		aprint_error("%s: can't allocate tx listseg, error = %d\n",
    651 		    sc->sc_dev.dv_xname, error);
    652 		goto fail_0;
    653 	}
    654 
    655 	/* Load the map for the TX ring. */
    656 	if ((error = bus_dmamem_map(sc->sc_dmat, &sc->re_ldata.re_tx_listseg,
    657 	    sc->re_ldata.re_tx_listnseg, RE_TX_LIST_SZ(sc),
    658 	    (void **)&sc->re_ldata.re_tx_list,
    659 	    BUS_DMA_COHERENT | BUS_DMA_NOWAIT)) != 0) {
    660 		aprint_error("%s: can't map tx list, error = %d\n",
    661 		    sc->sc_dev.dv_xname, error);
    662 	  	goto fail_1;
    663 	}
    664 	memset(sc->re_ldata.re_tx_list, 0, RE_TX_LIST_SZ(sc));
    665 
    666 	if ((error = bus_dmamap_create(sc->sc_dmat, RE_TX_LIST_SZ(sc), 1,
    667 	    RE_TX_LIST_SZ(sc), 0, 0,
    668 	    &sc->re_ldata.re_tx_list_map)) != 0) {
    669 		aprint_error("%s: can't create tx list map, error = %d\n",
    670 		    sc->sc_dev.dv_xname, error);
    671 		goto fail_2;
    672 	}
    673 
    674 
    675 	if ((error = bus_dmamap_load(sc->sc_dmat,
    676 	    sc->re_ldata.re_tx_list_map, sc->re_ldata.re_tx_list,
    677 	    RE_TX_LIST_SZ(sc), NULL, BUS_DMA_NOWAIT)) != 0) {
    678 		aprint_error("%s: can't load tx list, error = %d\n",
    679 		    sc->sc_dev.dv_xname, error);
    680 		goto fail_3;
    681 	}
    682 
    683 	/* Create DMA maps for TX buffers */
    684 	for (i = 0; i < RE_TX_QLEN; i++) {
    685 		error = bus_dmamap_create(sc->sc_dmat,
    686 		    round_page(IP_MAXPACKET),
    687 		    RE_TX_DESC_CNT(sc), RE_TDESC_CMD_FRAGLEN,
    688 		    0, 0, &sc->re_ldata.re_txq[i].txq_dmamap);
    689 		if (error) {
    690 			aprint_error("%s: can't create DMA map for TX\n",
    691 			    sc->sc_dev.dv_xname);
    692 			goto fail_4;
    693 		}
    694 	}
    695 
    696 	/* Allocate DMA'able memory for the RX ring */
    697 	/* XXX see also a comment about RE_RX_DMAMEM_SZ in rtl81x9var.h */
    698 	if ((error = bus_dmamem_alloc(sc->sc_dmat,
    699 	    RE_RX_DMAMEM_SZ, RE_RING_ALIGN, 0, &sc->re_ldata.re_rx_listseg, 1,
    700 	    &sc->re_ldata.re_rx_listnseg, BUS_DMA_NOWAIT)) != 0) {
    701 		aprint_error("%s: can't allocate rx listseg, error = %d\n",
    702 		    sc->sc_dev.dv_xname, error);
    703 		goto fail_4;
    704 	}
    705 
    706 	/* Load the map for the RX ring. */
    707 	if ((error = bus_dmamem_map(sc->sc_dmat, &sc->re_ldata.re_rx_listseg,
    708 	    sc->re_ldata.re_rx_listnseg, RE_RX_DMAMEM_SZ,
    709 	    (void **)&sc->re_ldata.re_rx_list,
    710 	    BUS_DMA_COHERENT | BUS_DMA_NOWAIT)) != 0) {
    711 		aprint_error("%s: can't map rx list, error = %d\n",
    712 		    sc->sc_dev.dv_xname, error);
    713 		goto fail_5;
    714 	}
    715 	memset(sc->re_ldata.re_rx_list, 0, RE_RX_DMAMEM_SZ);
    716 
    717 	if ((error = bus_dmamap_create(sc->sc_dmat,
    718 	    RE_RX_DMAMEM_SZ, 1, RE_RX_DMAMEM_SZ, 0, 0,
    719 	    &sc->re_ldata.re_rx_list_map)) != 0) {
    720 		aprint_error("%s: can't create rx list map, error = %d\n",
    721 		    sc->sc_dev.dv_xname, error);
    722 		goto fail_6;
    723 	}
    724 
    725 	if ((error = bus_dmamap_load(sc->sc_dmat,
    726 	    sc->re_ldata.re_rx_list_map, sc->re_ldata.re_rx_list,
    727 	    RE_RX_DMAMEM_SZ, NULL, BUS_DMA_NOWAIT)) != 0) {
    728 		aprint_error("%s: can't load rx list, error = %d\n",
    729 		    sc->sc_dev.dv_xname, error);
    730 		goto fail_7;
    731 	}
    732 
    733 	/* Create DMA maps for RX buffers */
    734 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
    735 		error = bus_dmamap_create(sc->sc_dmat, MCLBYTES, 1, MCLBYTES,
    736 		    0, 0, &sc->re_ldata.re_rxsoft[i].rxs_dmamap);
    737 		if (error) {
    738 			aprint_error("%s: can't create DMA map for RX\n",
    739 			    sc->sc_dev.dv_xname);
    740 			goto fail_8;
    741 		}
    742 	}
    743 
    744 	/*
    745 	 * Record interface as attached. From here, we should not fail.
    746 	 */
    747 	sc->sc_flags |= RTK_ATTACHED;
    748 
    749 	ifp = &sc->ethercom.ec_if;
    750 	ifp->if_softc = sc;
    751 	strcpy(ifp->if_xname, sc->sc_dev.dv_xname);
    752 	ifp->if_mtu = ETHERMTU;
    753 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
    754 	ifp->if_ioctl = re_ioctl;
    755 	sc->ethercom.ec_capabilities |=
    756 	    ETHERCAP_VLAN_MTU | ETHERCAP_VLAN_HWTAGGING;
    757 	ifp->if_start = re_start;
    758 	ifp->if_stop = re_stop;
    759 
    760 	/*
    761 	 * IFCAP_CSUM_IPv4_Tx on re(4) is broken for small packets,
    762 	 * so we have a workaround to handle the bug by padding
    763 	 * such packets manually.
    764 	 */
    765 	ifp->if_capabilities |=
    766 	    IFCAP_CSUM_IPv4_Tx | IFCAP_CSUM_IPv4_Rx |
    767 	    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
    768 	    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx |
    769 	    IFCAP_TSOv4;
    770 	ifp->if_watchdog = re_watchdog;
    771 	ifp->if_init = re_init;
    772 	ifp->if_snd.ifq_maxlen = RE_IFQ_MAXLEN;
    773 	ifp->if_capenable = ifp->if_capabilities;
    774 	IFQ_SET_READY(&ifp->if_snd);
    775 
    776 	callout_init(&sc->rtk_tick_ch, 0);
    777 
    778 	/* Do MII setup */
    779 	sc->mii.mii_ifp = ifp;
    780 	sc->mii.mii_readreg = re_miibus_readreg;
    781 	sc->mii.mii_writereg = re_miibus_writereg;
    782 	sc->mii.mii_statchg = re_miibus_statchg;
    783 	sc->ethercom.ec_mii = &sc->mii;
    784 	ifmedia_init(&sc->mii.mii_media, IFM_IMASK, ether_mediachange,
    785 	    ether_mediastatus);
    786 	mii_attach(&sc->sc_dev, &sc->mii, 0xffffffff, MII_PHY_ANY,
    787 	    MII_OFFSET_ANY, 0);
    788 	ifmedia_set(&sc->mii.mii_media, IFM_ETHER | IFM_AUTO);
    789 
    790 	/*
    791 	 * Call MI attach routine.
    792 	 */
    793 	if_attach(ifp);
    794 	ether_ifattach(ifp, eaddr);
    795 
    796 
    797 	/*
    798 	 * Make sure the interface is shutdown during reboot.
    799 	 */
    800 	sc->sc_sdhook = shutdownhook_establish(re_shutdown, sc);
    801 	if (sc->sc_sdhook == NULL)
    802 		aprint_error("%s: WARNING: unable to establish shutdown hook\n",
    803 		    sc->sc_dev.dv_xname);
    804 	/*
    805 	 * Add a suspend hook to make sure we come back up after a
    806 	 * resume.
    807 	 */
    808 	sc->sc_powerhook = powerhook_establish(sc->sc_dev.dv_xname,
    809 	    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 	    (void *)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 	    (void *)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 	    (void *)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 	    (void *)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_1(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 	if (!device_has_power(&sc->sc_dev))
   1461 		return 0;
   1462 
   1463 	ifp = &sc->ethercom.ec_if;
   1464 
   1465 	if ((ifp->if_flags & IFF_UP) == 0)
   1466 		return 0;
   1467 
   1468 #ifdef DEVICE_POLLING
   1469 	if (ifp->if_flags & IFF_POLLING)
   1470 		goto done;
   1471 	if ((ifp->if_capenable & IFCAP_POLLING) &&
   1472 	    ether_poll_register(re_poll, ifp)) { /* ok, disable interrupts */
   1473 		CSR_WRITE_2(sc, RTK_IMR, 0x0000);
   1474 		re_poll(ifp, 0, 1);
   1475 		goto done;
   1476 	}
   1477 #endif /* DEVICE_POLLING */
   1478 
   1479 	for (;;) {
   1480 
   1481 		status = CSR_READ_2(sc, RTK_ISR);
   1482 		/* If the card has gone away the read returns 0xffff. */
   1483 		if (status == 0xffff)
   1484 			break;
   1485 		if (status) {
   1486 			handled = 1;
   1487 			CSR_WRITE_2(sc, RTK_ISR, status);
   1488 		}
   1489 
   1490 		if ((status & RTK_INTRS_CPLUS) == 0)
   1491 			break;
   1492 
   1493 		if (status & (RTK_ISR_RX_OK | RTK_ISR_RX_ERR))
   1494 			re_rxeof(sc);
   1495 
   1496 		if (status & (RTK_ISR_TIMEOUT_EXPIRED | RTK_ISR_TX_ERR |
   1497 		    RTK_ISR_TX_DESC_UNAVAIL))
   1498 			re_txeof(sc);
   1499 
   1500 		if (status & RTK_ISR_SYSTEM_ERR) {
   1501 			re_init(ifp);
   1502 		}
   1503 
   1504 		if (status & RTK_ISR_LINKCHG) {
   1505 			callout_stop(&sc->rtk_tick_ch);
   1506 			re_tick(sc);
   1507 		}
   1508 	}
   1509 
   1510 	if (handled && !IFQ_IS_EMPTY(&ifp->if_snd))
   1511 		re_start(ifp);
   1512 
   1513 #ifdef DEVICE_POLLING
   1514  done:
   1515 #endif
   1516 
   1517 	return handled;
   1518 }
   1519 
   1520 
   1521 
   1522 /*
   1523  * Main transmit routine for C+ and gigE NICs.
   1524  */
   1525 
   1526 static void
   1527 re_start(struct ifnet *ifp)
   1528 {
   1529 	struct rtk_softc	*sc;
   1530 	struct mbuf		*m;
   1531 	bus_dmamap_t		map;
   1532 	struct re_txq		*txq;
   1533 	struct re_desc		*d;
   1534 	struct m_tag		*mtag;
   1535 	uint32_t		cmdstat, re_flags;
   1536 	int			ofree, idx, error, nsegs, seg;
   1537 	int			startdesc, curdesc, lastdesc;
   1538 	bool			pad;
   1539 
   1540 	sc = ifp->if_softc;
   1541 	ofree = sc->re_ldata.re_txq_free;
   1542 
   1543 	for (idx = sc->re_ldata.re_txq_prodidx;; idx = RE_NEXT_TXQ(sc, idx)) {
   1544 
   1545 		IFQ_POLL(&ifp->if_snd, m);
   1546 		if (m == NULL)
   1547 			break;
   1548 
   1549 		if (sc->re_ldata.re_txq_free == 0 ||
   1550 		    sc->re_ldata.re_tx_free == 0) {
   1551 			/* no more free slots left */
   1552 			ifp->if_flags |= IFF_OACTIVE;
   1553 			break;
   1554 		}
   1555 
   1556 		/*
   1557 		 * Set up checksum offload. Note: checksum offload bits must
   1558 		 * appear in all descriptors of a multi-descriptor transmit
   1559 		 * attempt. (This is according to testing done with an 8169
   1560 		 * chip. I'm not sure if this is a requirement or a bug.)
   1561 		 */
   1562 
   1563 		if ((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) != 0) {
   1564 			uint32_t segsz = m->m_pkthdr.segsz;
   1565 
   1566 			re_flags = RE_TDESC_CMD_LGSEND |
   1567 			    (segsz << RE_TDESC_CMD_MSSVAL_SHIFT);
   1568 		} else {
   1569 			/*
   1570 			 * set RE_TDESC_CMD_IPCSUM if any checksum offloading
   1571 			 * is requested.  otherwise, RE_TDESC_CMD_TCPCSUM/
   1572 			 * RE_TDESC_CMD_UDPCSUM doesn't make effects.
   1573 			 */
   1574 			re_flags = 0;
   1575 			if ((m->m_pkthdr.csum_flags &
   1576 			    (M_CSUM_IPv4 | M_CSUM_TCPv4 | M_CSUM_UDPv4))
   1577 			    != 0) {
   1578 				re_flags |= RE_TDESC_CMD_IPCSUM;
   1579 				if (m->m_pkthdr.csum_flags & M_CSUM_TCPv4) {
   1580 					re_flags |= RE_TDESC_CMD_TCPCSUM;
   1581 				} else if (m->m_pkthdr.csum_flags &
   1582 				    M_CSUM_UDPv4) {
   1583 					re_flags |= RE_TDESC_CMD_UDPCSUM;
   1584 				}
   1585 			}
   1586 		}
   1587 
   1588 		txq = &sc->re_ldata.re_txq[idx];
   1589 		map = txq->txq_dmamap;
   1590 		error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m,
   1591 		    BUS_DMA_WRITE|BUS_DMA_NOWAIT);
   1592 
   1593 		if (__predict_false(error)) {
   1594 			/* XXX try to defrag if EFBIG? */
   1595 			aprint_error("%s: can't map mbuf (error %d)\n",
   1596 			    sc->sc_dev.dv_xname, error);
   1597 
   1598 			IFQ_DEQUEUE(&ifp->if_snd, m);
   1599 			m_freem(m);
   1600 			ifp->if_oerrors++;
   1601 			continue;
   1602 		}
   1603 
   1604 		nsegs = map->dm_nsegs;
   1605 		pad = false;
   1606 		if (__predict_false(m->m_pkthdr.len <= RE_IP4CSUMTX_PADLEN &&
   1607 		    (re_flags & RE_TDESC_CMD_IPCSUM) != 0)) {
   1608 			pad = true;
   1609 			nsegs++;
   1610 		}
   1611 
   1612 		if (nsegs > sc->re_ldata.re_tx_free) {
   1613 			/*
   1614 			 * Not enough free descriptors to transmit this packet.
   1615 			 */
   1616 			ifp->if_flags |= IFF_OACTIVE;
   1617 			bus_dmamap_unload(sc->sc_dmat, map);
   1618 			break;
   1619 		}
   1620 
   1621 		IFQ_DEQUEUE(&ifp->if_snd, m);
   1622 
   1623 		/*
   1624 		 * Make sure that the caches are synchronized before we
   1625 		 * ask the chip to start DMA for the packet data.
   1626 		 */
   1627 		bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
   1628 		    BUS_DMASYNC_PREWRITE);
   1629 
   1630 		/*
   1631 		 * Map the segment array into descriptors.
   1632 		 * Note that we set the start-of-frame and
   1633 		 * end-of-frame markers for either TX or RX,
   1634 		 * but they really only have meaning in the TX case.
   1635 		 * (In the RX case, it's the chip that tells us
   1636 		 *  where packets begin and end.)
   1637 		 * We also keep track of the end of the ring
   1638 		 * and set the end-of-ring bits as needed,
   1639 		 * and we set the ownership bits in all except
   1640 		 * the very first descriptor. (The caller will
   1641 		 * set this descriptor later when it start
   1642 		 * transmission or reception.)
   1643 		 */
   1644 		curdesc = startdesc = sc->re_ldata.re_tx_nextfree;
   1645 		lastdesc = -1;
   1646 		for (seg = 0; seg < map->dm_nsegs;
   1647 		    seg++, curdesc = RE_NEXT_TX_DESC(sc, curdesc)) {
   1648 			d = &sc->re_ldata.re_tx_list[curdesc];
   1649 #ifdef DIAGNOSTIC
   1650 			RE_TXDESCSYNC(sc, curdesc,
   1651 			    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
   1652 			cmdstat = le32toh(d->re_cmdstat);
   1653 			RE_TXDESCSYNC(sc, curdesc, BUS_DMASYNC_PREREAD);
   1654 			if (cmdstat & RE_TDESC_STAT_OWN) {
   1655 				panic("%s: tried to map busy TX descriptor",
   1656 				    sc->sc_dev.dv_xname);
   1657 			}
   1658 #endif
   1659 
   1660 			d->re_vlanctl = 0;
   1661 			re_set_bufaddr(d, map->dm_segs[seg].ds_addr);
   1662 			cmdstat = re_flags | map->dm_segs[seg].ds_len;
   1663 			if (seg == 0)
   1664 				cmdstat |= RE_TDESC_CMD_SOF;
   1665 			else
   1666 				cmdstat |= RE_TDESC_CMD_OWN;
   1667 			if (curdesc == (RE_TX_DESC_CNT(sc) - 1))
   1668 				cmdstat |= RE_TDESC_CMD_EOR;
   1669 			if (seg == nsegs - 1) {
   1670 				cmdstat |= RE_TDESC_CMD_EOF;
   1671 				lastdesc = curdesc;
   1672 			}
   1673 			d->re_cmdstat = htole32(cmdstat);
   1674 			RE_TXDESCSYNC(sc, curdesc,
   1675 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1676 		}
   1677 		if (__predict_false(pad)) {
   1678 			bus_addr_t paddaddr;
   1679 
   1680 			d = &sc->re_ldata.re_tx_list[curdesc];
   1681 			d->re_vlanctl = 0;
   1682 			paddaddr = RE_TXPADDADDR(sc);
   1683 			re_set_bufaddr(d, paddaddr);
   1684 			cmdstat = re_flags |
   1685 			    RE_TDESC_CMD_OWN | RE_TDESC_CMD_EOF |
   1686 			    (RE_IP4CSUMTX_PADLEN + 1 - m->m_pkthdr.len);
   1687 			if (curdesc == (RE_TX_DESC_CNT(sc) - 1))
   1688 				cmdstat |= RE_TDESC_CMD_EOR;
   1689 			d->re_cmdstat = htole32(cmdstat);
   1690 			RE_TXDESCSYNC(sc, curdesc,
   1691 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1692 			lastdesc = curdesc;
   1693 			curdesc = RE_NEXT_TX_DESC(sc, curdesc);
   1694 		}
   1695 		KASSERT(lastdesc != -1);
   1696 
   1697 		/*
   1698 		 * Set up hardware VLAN tagging. Note: vlan tag info must
   1699 		 * appear in the first descriptor of a multi-descriptor
   1700 		 * transmission attempt.
   1701 		 */
   1702 		if ((mtag = VLAN_OUTPUT_TAG(&sc->ethercom, m)) != NULL) {
   1703 			sc->re_ldata.re_tx_list[startdesc].re_vlanctl =
   1704 			    htole32(bswap16(VLAN_TAG_VALUE(mtag)) |
   1705 			    RE_TDESC_VLANCTL_TAG);
   1706 		}
   1707 
   1708 		/* Transfer ownership of packet to the chip. */
   1709 
   1710 		sc->re_ldata.re_tx_list[startdesc].re_cmdstat |=
   1711 		    htole32(RE_TDESC_CMD_OWN);
   1712 		RE_TXDESCSYNC(sc, startdesc,
   1713 		    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1714 
   1715 		/* update info of TX queue and descriptors */
   1716 		txq->txq_mbuf = m;
   1717 		txq->txq_descidx = lastdesc;
   1718 		txq->txq_nsegs = nsegs;
   1719 
   1720 		sc->re_ldata.re_txq_free--;
   1721 		sc->re_ldata.re_tx_free -= nsegs;
   1722 		sc->re_ldata.re_tx_nextfree = curdesc;
   1723 
   1724 #if NBPFILTER > 0
   1725 		/*
   1726 		 * If there's a BPF listener, bounce a copy of this frame
   1727 		 * to him.
   1728 		 */
   1729 		if (ifp->if_bpf)
   1730 			bpf_mtap(ifp->if_bpf, m);
   1731 #endif
   1732 	}
   1733 
   1734 	if (sc->re_ldata.re_txq_free < ofree) {
   1735 		/*
   1736 		 * TX packets are enqueued.
   1737 		 */
   1738 		sc->re_ldata.re_txq_prodidx = idx;
   1739 
   1740 		/*
   1741 		 * Start the transmitter to poll.
   1742 		 *
   1743 		 * RealTek put the TX poll request register in a different
   1744 		 * location on the 8169 gigE chip. I don't know why.
   1745 		 */
   1746 		if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0)
   1747 			CSR_WRITE_1(sc, RTK_TXSTART, RTK_TXSTART_START);
   1748 		else
   1749 			CSR_WRITE_1(sc, RTK_GTXSTART, RTK_TXSTART_START);
   1750 
   1751 		/*
   1752 		 * Use the countdown timer for interrupt moderation.
   1753 		 * 'TX done' interrupts are disabled. Instead, we reset the
   1754 		 * countdown timer, which will begin counting until it hits
   1755 		 * the value in the TIMERINT register, and then trigger an
   1756 		 * interrupt. Each time we write to the TIMERCNT register,
   1757 		 * the timer count is reset to 0.
   1758 		 */
   1759 		CSR_WRITE_4(sc, RTK_TIMERCNT, 1);
   1760 
   1761 		/*
   1762 		 * Set a timeout in case the chip goes out to lunch.
   1763 		 */
   1764 		ifp->if_timer = 5;
   1765 	}
   1766 }
   1767 
   1768 static int
   1769 re_init(struct ifnet *ifp)
   1770 {
   1771 	struct rtk_softc	*sc = ifp->if_softc;
   1772 	const uint8_t		*enaddr;
   1773 	uint32_t		rxcfg = 0;
   1774 	uint32_t		reg;
   1775 	int error;
   1776 
   1777 	if ((error = re_enable(sc)) != 0)
   1778 		goto out;
   1779 
   1780 	/*
   1781 	 * Cancel pending I/O and free all RX/TX buffers.
   1782 	 */
   1783 	re_stop(ifp, 0);
   1784 
   1785 	re_reset(sc);
   1786 
   1787 	/*
   1788 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
   1789 	 * RX checksum offload. We must configure the C+ register
   1790 	 * before all others.
   1791 	 */
   1792 	reg = 0;
   1793 
   1794 	/*
   1795 	 * XXX: Realtek docs say bits 0 and 1 are reserved, for 8169S/8110S.
   1796 	 * FreeBSD  drivers set these bits anyway (for 8139C+?).
   1797 	 * So far, it works.
   1798 	 */
   1799 
   1800 	/*
   1801 	 * XXX: For old 8169 set bit 14.
   1802 	 *      For 8169S/8110S and above, do not set bit 14.
   1803 	 */
   1804 	if ((sc->sc_quirk & RTKQ_8169NONS) != 0)
   1805 		reg |= (0x1 << 14) | RTK_CPLUSCMD_PCI_MRW;;
   1806 
   1807 	if (1)  {/* not for 8169S ? */
   1808 		reg |=
   1809 		    RTK_CPLUSCMD_VLANSTRIP |
   1810 		    (ifp->if_capenable &
   1811 		    (IFCAP_CSUM_IPv4_Rx | IFCAP_CSUM_TCPv4_Rx |
   1812 		     IFCAP_CSUM_UDPv4_Rx) ?
   1813 		    RTK_CPLUSCMD_RXCSUM_ENB : 0);
   1814 	}
   1815 
   1816 	CSR_WRITE_2(sc, RTK_CPLUS_CMD,
   1817 	    reg | RTK_CPLUSCMD_RXENB | RTK_CPLUSCMD_TXENB);
   1818 
   1819 	/* XXX: from Realtek-supplied Linux driver. Wholly undocumented. */
   1820 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0)
   1821 		CSR_WRITE_2(sc, RTK_IM, 0x0000);
   1822 
   1823 	DELAY(10000);
   1824 
   1825 	/*
   1826 	 * Init our MAC address.  Even though the chipset
   1827 	 * documentation doesn't mention it, we need to enter "Config
   1828 	 * register write enable" mode to modify the ID registers.
   1829 	 */
   1830 	CSR_WRITE_1(sc, RTK_EECMD, RTK_EEMODE_WRITECFG);
   1831 	enaddr = CLLADDR(ifp->if_sadl);
   1832 	reg = enaddr[0] | (enaddr[1] << 8) |
   1833 	    (enaddr[2] << 16) | (enaddr[3] << 24);
   1834 	CSR_WRITE_4(sc, RTK_IDR0, reg);
   1835 	reg = enaddr[4] | (enaddr[5] << 8);
   1836 	CSR_WRITE_4(sc, RTK_IDR4, reg);
   1837 	CSR_WRITE_1(sc, RTK_EECMD, RTK_EEMODE_OFF);
   1838 
   1839 	/*
   1840 	 * For C+ mode, initialize the RX descriptors and mbufs.
   1841 	 */
   1842 	re_rx_list_init(sc);
   1843 	re_tx_list_init(sc);
   1844 
   1845 	/*
   1846 	 * Load the addresses of the RX and TX lists into the chip.
   1847 	 */
   1848 	CSR_WRITE_4(sc, RTK_RXLIST_ADDR_HI,
   1849 	    RE_ADDR_HI(sc->re_ldata.re_rx_list_map->dm_segs[0].ds_addr));
   1850 	CSR_WRITE_4(sc, RTK_RXLIST_ADDR_LO,
   1851 	    RE_ADDR_LO(sc->re_ldata.re_rx_list_map->dm_segs[0].ds_addr));
   1852 
   1853 	CSR_WRITE_4(sc, RTK_TXLIST_ADDR_HI,
   1854 	    RE_ADDR_HI(sc->re_ldata.re_tx_list_map->dm_segs[0].ds_addr));
   1855 	CSR_WRITE_4(sc, RTK_TXLIST_ADDR_LO,
   1856 	    RE_ADDR_LO(sc->re_ldata.re_tx_list_map->dm_segs[0].ds_addr));
   1857 
   1858 	/*
   1859 	 * Enable transmit and receive.
   1860 	 */
   1861 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_TX_ENB | RTK_CMD_RX_ENB);
   1862 
   1863 	/*
   1864 	 * Set the initial TX and RX configuration.
   1865 	 */
   1866 	if (sc->re_testmode && (sc->sc_quirk & RTKQ_8169NONS) != 0) {
   1867 		/* test mode is needed only for old 8169 */
   1868 		CSR_WRITE_4(sc, RTK_TXCFG,
   1869 		    RE_TXCFG_CONFIG | RTK_LOOPTEST_ON);
   1870 	} else
   1871 		CSR_WRITE_4(sc, RTK_TXCFG, RE_TXCFG_CONFIG);
   1872 
   1873 	CSR_WRITE_1(sc, RTK_EARLY_TX_THRESH, 16);
   1874 
   1875 	CSR_WRITE_4(sc, RTK_RXCFG, RE_RXCFG_CONFIG);
   1876 
   1877 	/* Set the individual bit to receive frames for this host only. */
   1878 	rxcfg = CSR_READ_4(sc, RTK_RXCFG);
   1879 	rxcfg |= RTK_RXCFG_RX_INDIV;
   1880 
   1881 	/* If we want promiscuous mode, set the allframes bit. */
   1882 	if (ifp->if_flags & IFF_PROMISC)
   1883 		rxcfg |= RTK_RXCFG_RX_ALLPHYS;
   1884 	else
   1885 		rxcfg &= ~RTK_RXCFG_RX_ALLPHYS;
   1886 	CSR_WRITE_4(sc, RTK_RXCFG, rxcfg);
   1887 
   1888 	/*
   1889 	 * Set capture broadcast bit to capture broadcast frames.
   1890 	 */
   1891 	if (ifp->if_flags & IFF_BROADCAST)
   1892 		rxcfg |= RTK_RXCFG_RX_BROAD;
   1893 	else
   1894 		rxcfg &= ~RTK_RXCFG_RX_BROAD;
   1895 	CSR_WRITE_4(sc, RTK_RXCFG, rxcfg);
   1896 
   1897 	/*
   1898 	 * Program the multicast filter, if necessary.
   1899 	 */
   1900 	rtk_setmulti(sc);
   1901 
   1902 #ifdef DEVICE_POLLING
   1903 	/*
   1904 	 * Disable interrupts if we are polling.
   1905 	 */
   1906 	if (ifp->if_flags & IFF_POLLING)
   1907 		CSR_WRITE_2(sc, RTK_IMR, 0);
   1908 	else	/* otherwise ... */
   1909 #endif /* DEVICE_POLLING */
   1910 	/*
   1911 	 * Enable interrupts.
   1912 	 */
   1913 	if (sc->re_testmode)
   1914 		CSR_WRITE_2(sc, RTK_IMR, 0);
   1915 	else
   1916 		CSR_WRITE_2(sc, RTK_IMR, RTK_INTRS_CPLUS);
   1917 
   1918 	/* Start RX/TX process. */
   1919 	CSR_WRITE_4(sc, RTK_MISSEDPKT, 0);
   1920 #ifdef notdef
   1921 	/* Enable receiver and transmitter. */
   1922 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_TX_ENB | RTK_CMD_RX_ENB);
   1923 #endif
   1924 
   1925 	/*
   1926 	 * Initialize the timer interrupt register so that
   1927 	 * a timer interrupt will be generated once the timer
   1928 	 * reaches a certain number of ticks. The timer is
   1929 	 * reloaded on each transmit. This gives us TX interrupt
   1930 	 * moderation, which dramatically improves TX frame rate.
   1931 	 */
   1932 
   1933 	if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0)
   1934 		CSR_WRITE_4(sc, RTK_TIMERINT, 0x400);
   1935 	else {
   1936 		CSR_WRITE_4(sc, RTK_TIMERINT_8169, 0x800);
   1937 
   1938 		/*
   1939 		 * For 8169 gigE NICs, set the max allowed RX packet
   1940 		 * size so we can receive jumbo frames.
   1941 		 */
   1942 		CSR_WRITE_2(sc, RTK_MAXRXPKTLEN, 16383);
   1943 	}
   1944 
   1945 	if (sc->re_testmode)
   1946 		return 0;
   1947 
   1948 	CSR_WRITE_1(sc, RTK_CFG1, RTK_CFG1_DRVLOAD);
   1949 
   1950 	ifp->if_flags |= IFF_RUNNING;
   1951 	ifp->if_flags &= ~IFF_OACTIVE;
   1952 
   1953 	callout_reset(&sc->rtk_tick_ch, hz, re_tick, sc);
   1954 
   1955  out:
   1956 	if (error) {
   1957 		ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
   1958 		ifp->if_timer = 0;
   1959 		aprint_error("%s: interface not running\n",
   1960 		    sc->sc_dev.dv_xname);
   1961 	}
   1962 
   1963 	return error;
   1964 }
   1965 
   1966 static int
   1967 re_ioctl(struct ifnet *ifp, u_long command, void *data)
   1968 {
   1969 	struct rtk_softc	*sc = ifp->if_softc;
   1970 	struct ifreq		*ifr = (struct ifreq *) data;
   1971 	int			s, error = 0;
   1972 
   1973 	s = splnet();
   1974 
   1975 	switch (command) {
   1976 	case SIOCSIFMTU:
   1977 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > ETHERMTU_JUMBO)
   1978 			error = EINVAL;
   1979 		else if ((error = ifioctl_common(ifp, command, data)) == ENETRESET)
   1980 			error = 0;
   1981 		break;
   1982 	default:
   1983 		if ((error = ether_ioctl(ifp, command, data)) != ENETRESET)
   1984 			break;
   1985 
   1986 		error = 0;
   1987 
   1988 		if (command == SIOCSIFCAP)
   1989 			error = (*ifp->if_init)(ifp);
   1990 		else if (command != SIOCADDMULTI && command != SIOCDELMULTI)
   1991 			;
   1992 		else if (ifp->if_flags & IFF_RUNNING)
   1993 			rtk_setmulti(sc);
   1994 		break;
   1995 	}
   1996 
   1997 	splx(s);
   1998 
   1999 	return error;
   2000 }
   2001 
   2002 static void
   2003 re_watchdog(struct ifnet *ifp)
   2004 {
   2005 	struct rtk_softc	*sc;
   2006 	int			s;
   2007 
   2008 	sc = ifp->if_softc;
   2009 	s = splnet();
   2010 	aprint_error("%s: watchdog timeout\n", sc->sc_dev.dv_xname);
   2011 	ifp->if_oerrors++;
   2012 
   2013 	re_txeof(sc);
   2014 	re_rxeof(sc);
   2015 
   2016 	re_init(ifp);
   2017 
   2018 	splx(s);
   2019 }
   2020 
   2021 /*
   2022  * Stop the adapter and free any mbufs allocated to the
   2023  * RX and TX lists.
   2024  */
   2025 static void
   2026 re_stop(struct ifnet *ifp, int disable)
   2027 {
   2028 	int		i;
   2029 	struct rtk_softc *sc = ifp->if_softc;
   2030 
   2031 	callout_stop(&sc->rtk_tick_ch);
   2032 
   2033 #ifdef DEVICE_POLLING
   2034 	ether_poll_deregister(ifp);
   2035 #endif /* DEVICE_POLLING */
   2036 
   2037 	mii_down(&sc->mii);
   2038 
   2039 	CSR_WRITE_1(sc, RTK_COMMAND, 0x00);
   2040 	CSR_WRITE_2(sc, RTK_IMR, 0x0000);
   2041 
   2042 	if (sc->re_head != NULL) {
   2043 		m_freem(sc->re_head);
   2044 		sc->re_head = sc->re_tail = NULL;
   2045 	}
   2046 
   2047 	/* Free the TX list buffers. */
   2048 	for (i = 0; i < RE_TX_QLEN; i++) {
   2049 		if (sc->re_ldata.re_txq[i].txq_mbuf != NULL) {
   2050 			bus_dmamap_unload(sc->sc_dmat,
   2051 			    sc->re_ldata.re_txq[i].txq_dmamap);
   2052 			m_freem(sc->re_ldata.re_txq[i].txq_mbuf);
   2053 			sc->re_ldata.re_txq[i].txq_mbuf = NULL;
   2054 		}
   2055 	}
   2056 
   2057 	/* Free the RX list buffers. */
   2058 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
   2059 		if (sc->re_ldata.re_rxsoft[i].rxs_mbuf != NULL) {
   2060 			bus_dmamap_unload(sc->sc_dmat,
   2061 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
   2062 			m_freem(sc->re_ldata.re_rxsoft[i].rxs_mbuf);
   2063 			sc->re_ldata.re_rxsoft[i].rxs_mbuf = NULL;
   2064 		}
   2065 	}
   2066 
   2067 	if (disable)
   2068 		re_disable(sc);
   2069 
   2070 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
   2071 	ifp->if_timer = 0;
   2072 }
   2073