Home | History | Annotate | Line # | Download | only in ic
rtl8169.c revision 1.125
      1 /*	$NetBSD: rtl8169.c,v 1.125 2009/09/02 15:11:13 tsutsui 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 1.125 2009/09/02 15:11:13 tsutsui 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 
    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 <sys/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 int re_enable(struct rtk_softc *);
    166 static void re_disable(struct rtk_softc *);
    167 
    168 static int re_gmii_readreg(device_t, int, int);
    169 static void re_gmii_writereg(device_t, int, int, int);
    170 
    171 static int re_miibus_readreg(device_t, int, int);
    172 static void re_miibus_writereg(device_t, int, int, int);
    173 static void re_miibus_statchg(device_t);
    174 
    175 static void re_reset(struct rtk_softc *);
    176 
    177 static inline void
    178 re_set_bufaddr(struct re_desc *d, bus_addr_t addr)
    179 {
    180 
    181 	d->re_bufaddr_lo = htole32((uint32_t)addr);
    182 	if (sizeof(bus_addr_t) == sizeof(uint64_t))
    183 		d->re_bufaddr_hi = htole32((uint64_t)addr >> 32);
    184 	else
    185 		d->re_bufaddr_hi = 0;
    186 }
    187 
    188 static int
    189 re_gmii_readreg(device_t dev, int phy, int reg)
    190 {
    191 	struct rtk_softc *sc = device_private(dev);
    192 	uint32_t rval;
    193 	int i;
    194 
    195 	if (phy != 7)
    196 		return 0;
    197 
    198 	/* Let the rgephy driver read the GMEDIASTAT register */
    199 
    200 	if (reg == RTK_GMEDIASTAT) {
    201 		rval = CSR_READ_1(sc, RTK_GMEDIASTAT);
    202 		return rval;
    203 	}
    204 
    205 	CSR_WRITE_4(sc, RTK_PHYAR, reg << 16);
    206 	DELAY(1000);
    207 
    208 	for (i = 0; i < RTK_TIMEOUT; i++) {
    209 		rval = CSR_READ_4(sc, RTK_PHYAR);
    210 		if (rval & RTK_PHYAR_BUSY)
    211 			break;
    212 		DELAY(100);
    213 	}
    214 
    215 	if (i == RTK_TIMEOUT) {
    216 		printf("%s: PHY read failed\n", device_xname(sc->sc_dev));
    217 		return 0;
    218 	}
    219 
    220 	return rval & RTK_PHYAR_PHYDATA;
    221 }
    222 
    223 static void
    224 re_gmii_writereg(device_t dev, int phy, int reg, int data)
    225 {
    226 	struct rtk_softc *sc = device_private(dev);
    227 	uint32_t rval;
    228 	int i;
    229 
    230 	CSR_WRITE_4(sc, RTK_PHYAR, (reg << 16) |
    231 	    (data & RTK_PHYAR_PHYDATA) | RTK_PHYAR_BUSY);
    232 	DELAY(1000);
    233 
    234 	for (i = 0; i < RTK_TIMEOUT; i++) {
    235 		rval = CSR_READ_4(sc, RTK_PHYAR);
    236 		if (!(rval & RTK_PHYAR_BUSY))
    237 			break;
    238 		DELAY(100);
    239 	}
    240 
    241 	if (i == RTK_TIMEOUT) {
    242 		printf("%s: PHY write reg %x <- %x failed\n",
    243 		    device_xname(sc->sc_dev), reg, data);
    244 	}
    245 }
    246 
    247 static int
    248 re_miibus_readreg(device_t dev, int phy, int reg)
    249 {
    250 	struct rtk_softc *sc = device_private(dev);
    251 	uint16_t rval = 0;
    252 	uint16_t re8139_reg = 0;
    253 	int s;
    254 
    255 	s = splnet();
    256 
    257 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0) {
    258 		rval = re_gmii_readreg(dev, phy, reg);
    259 		splx(s);
    260 		return rval;
    261 	}
    262 
    263 	/* Pretend the internal PHY is only at address 0 */
    264 	if (phy) {
    265 		splx(s);
    266 		return 0;
    267 	}
    268 	switch (reg) {
    269 	case MII_BMCR:
    270 		re8139_reg = RTK_BMCR;
    271 		break;
    272 	case MII_BMSR:
    273 		re8139_reg = RTK_BMSR;
    274 		break;
    275 	case MII_ANAR:
    276 		re8139_reg = RTK_ANAR;
    277 		break;
    278 	case MII_ANER:
    279 		re8139_reg = RTK_ANER;
    280 		break;
    281 	case MII_ANLPAR:
    282 		re8139_reg = RTK_LPAR;
    283 		break;
    284 	case MII_PHYIDR1:
    285 	case MII_PHYIDR2:
    286 		splx(s);
    287 		return 0;
    288 	/*
    289 	 * Allow the rlphy driver to read the media status
    290 	 * register. If we have a link partner which does not
    291 	 * support NWAY, this is the register which will tell
    292 	 * us the results of parallel detection.
    293 	 */
    294 	case RTK_MEDIASTAT:
    295 		rval = CSR_READ_1(sc, RTK_MEDIASTAT);
    296 		splx(s);
    297 		return rval;
    298 	default:
    299 		printf("%s: bad phy register\n", device_xname(sc->sc_dev));
    300 		splx(s);
    301 		return 0;
    302 	}
    303 	rval = CSR_READ_2(sc, re8139_reg);
    304 	if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0 && re8139_reg == RTK_BMCR) {
    305 		/* 8139C+ has different bit layout. */
    306 		rval &= ~(BMCR_LOOP | BMCR_ISO);
    307 	}
    308 	splx(s);
    309 	return rval;
    310 }
    311 
    312 static void
    313 re_miibus_writereg(device_t dev, int phy, int reg, int data)
    314 {
    315 	struct rtk_softc *sc = device_private(dev);
    316 	uint16_t re8139_reg = 0;
    317 	int s;
    318 
    319 	s = splnet();
    320 
    321 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0) {
    322 		re_gmii_writereg(dev, phy, reg, data);
    323 		splx(s);
    324 		return;
    325 	}
    326 
    327 	/* Pretend the internal PHY is only at address 0 */
    328 	if (phy) {
    329 		splx(s);
    330 		return;
    331 	}
    332 	switch (reg) {
    333 	case MII_BMCR:
    334 		re8139_reg = RTK_BMCR;
    335 		if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0) {
    336 			/* 8139C+ has different bit layout. */
    337 			data &= ~(BMCR_LOOP | BMCR_ISO);
    338 		}
    339 		break;
    340 	case MII_BMSR:
    341 		re8139_reg = RTK_BMSR;
    342 		break;
    343 	case MII_ANAR:
    344 		re8139_reg = RTK_ANAR;
    345 		break;
    346 	case MII_ANER:
    347 		re8139_reg = RTK_ANER;
    348 		break;
    349 	case MII_ANLPAR:
    350 		re8139_reg = RTK_LPAR;
    351 		break;
    352 	case MII_PHYIDR1:
    353 	case MII_PHYIDR2:
    354 		splx(s);
    355 		return;
    356 		break;
    357 	default:
    358 		printf("%s: bad phy register\n", device_xname(sc->sc_dev));
    359 		splx(s);
    360 		return;
    361 	}
    362 	CSR_WRITE_2(sc, re8139_reg, data);
    363 	splx(s);
    364 	return;
    365 }
    366 
    367 static void
    368 re_miibus_statchg(device_t dev)
    369 {
    370 
    371 	return;
    372 }
    373 
    374 static void
    375 re_reset(struct rtk_softc *sc)
    376 {
    377 	int i;
    378 
    379 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_RESET);
    380 
    381 	for (i = 0; i < RTK_TIMEOUT; i++) {
    382 		DELAY(10);
    383 		if ((CSR_READ_1(sc, RTK_COMMAND) & RTK_CMD_RESET) == 0)
    384 			break;
    385 	}
    386 	if (i == RTK_TIMEOUT)
    387 		printf("%s: reset never completed!\n",
    388 		    device_xname(sc->sc_dev));
    389 
    390 	/*
    391 	 * NB: Realtek-supplied FreeBSD driver does this only for MACFG_3,
    392 	 *     but also says "Rtl8169s sigle chip detected".
    393 	 */
    394 	if ((sc->sc_quirk & RTKQ_MACLDPS) != 0)
    395 		CSR_WRITE_1(sc, RTK_LDPS, 1);
    396 
    397 }
    398 
    399 /*
    400  * The following routine is designed to test for a defect on some
    401  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
    402  * lines connected to the bus, however for a 32-bit only card, they
    403  * should be pulled high. The result of this defect is that the
    404  * NIC will not work right if you plug it into a 64-bit slot: DMA
    405  * operations will be done with 64-bit transfers, which will fail
    406  * because the 64-bit data lines aren't connected.
    407  *
    408  * There's no way to work around this (short of talking a soldering
    409  * iron to the board), however we can detect it. The method we use
    410  * here is to put the NIC into digital loopback mode, set the receiver
    411  * to promiscuous mode, and then try to send a frame. We then compare
    412  * the frame data we sent to what was received. If the data matches,
    413  * then the NIC is working correctly, otherwise we know the user has
    414  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
    415  * slot. In the latter case, there's no way the NIC can work correctly,
    416  * so we print out a message on the console and abort the device attach.
    417  */
    418 
    419 int
    420 re_diag(struct rtk_softc *sc)
    421 {
    422 	struct ifnet *ifp = &sc->ethercom.ec_if;
    423 	struct mbuf *m0;
    424 	struct ether_header *eh;
    425 	struct re_rxsoft *rxs;
    426 	struct re_desc *cur_rx;
    427 	bus_dmamap_t dmamap;
    428 	uint16_t status;
    429 	uint32_t rxstat;
    430 	int total_len, i, s, error = 0;
    431 	static const uint8_t dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
    432 	static const uint8_t src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
    433 
    434 	/* Allocate a single mbuf */
    435 
    436 	MGETHDR(m0, M_DONTWAIT, MT_DATA);
    437 	if (m0 == NULL)
    438 		return ENOBUFS;
    439 
    440 	/*
    441 	 * Initialize the NIC in test mode. This sets the chip up
    442 	 * so that it can send and receive frames, but performs the
    443 	 * following special functions:
    444 	 * - Puts receiver in promiscuous mode
    445 	 * - Enables digital loopback mode
    446 	 * - Leaves interrupts turned off
    447 	 */
    448 
    449 	ifp->if_flags |= IFF_PROMISC;
    450 	sc->re_testmode = 1;
    451 	re_init(ifp);
    452 	re_stop(ifp, 0);
    453 	DELAY(100000);
    454 	re_init(ifp);
    455 
    456 	/* Put some data in the mbuf */
    457 
    458 	eh = mtod(m0, struct ether_header *);
    459 	memcpy(eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN);
    460 	memcpy(eh->ether_shost, (char *)&src, ETHER_ADDR_LEN);
    461 	eh->ether_type = htons(ETHERTYPE_IP);
    462 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
    463 
    464 	/*
    465 	 * Queue the packet, start transmission.
    466 	 */
    467 
    468 	CSR_WRITE_2(sc, RTK_ISR, 0xFFFF);
    469 	s = splnet();
    470 	IF_ENQUEUE(&ifp->if_snd, m0);
    471 	re_start(ifp);
    472 	splx(s);
    473 	m0 = NULL;
    474 
    475 	/* Wait for it to propagate through the chip */
    476 
    477 	DELAY(100000);
    478 	for (i = 0; i < RTK_TIMEOUT; i++) {
    479 		status = CSR_READ_2(sc, RTK_ISR);
    480 		if ((status & (RTK_ISR_TIMEOUT_EXPIRED | RTK_ISR_RX_OK)) ==
    481 		    (RTK_ISR_TIMEOUT_EXPIRED | RTK_ISR_RX_OK))
    482 			break;
    483 		DELAY(10);
    484 	}
    485 	if (i == RTK_TIMEOUT) {
    486 		aprint_error_dev(sc->sc_dev,
    487 		    "diagnostic failed, failed to receive packet "
    488 		    "in loopback mode\n");
    489 		error = EIO;
    490 		goto done;
    491 	}
    492 
    493 	/*
    494 	 * The packet should have been dumped into the first
    495 	 * entry in the RX DMA ring. Grab it from there.
    496 	 */
    497 
    498 	rxs = &sc->re_ldata.re_rxsoft[0];
    499 	dmamap = rxs->rxs_dmamap;
    500 	bus_dmamap_sync(sc->sc_dmat, dmamap, 0, dmamap->dm_mapsize,
    501 	    BUS_DMASYNC_POSTREAD);
    502 	bus_dmamap_unload(sc->sc_dmat, dmamap);
    503 
    504 	m0 = rxs->rxs_mbuf;
    505 	rxs->rxs_mbuf = NULL;
    506 	eh = mtod(m0, struct ether_header *);
    507 
    508 	RE_RXDESCSYNC(sc, 0, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
    509 	cur_rx = &sc->re_ldata.re_rx_list[0];
    510 	rxstat = le32toh(cur_rx->re_cmdstat);
    511 	total_len = rxstat & sc->re_rxlenmask;
    512 
    513 	if (total_len != ETHER_MIN_LEN) {
    514 		aprint_error_dev(sc->sc_dev,
    515 		    "diagnostic failed, received short packet\n");
    516 		error = EIO;
    517 		goto done;
    518 	}
    519 
    520 	/* Test that the received packet data matches what we sent. */
    521 
    522 	if (memcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
    523 	    memcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
    524 	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
    525 		aprint_error_dev(sc->sc_dev, "WARNING, DMA FAILURE!\n"
    526 		    "expected TX data: %s/%s/0x%x\n"
    527 		    "received RX data: %s/%s/0x%x\n"
    528 		    "You may have a defective 32-bit NIC plugged "
    529 		    "into a 64-bit PCI slot.\n"
    530 		    "Please re-install the NIC in a 32-bit slot "
    531 		    "for proper operation.\n"
    532 		    "Read the re(4) man page for more details.\n" ,
    533 		    ether_sprintf(dst),  ether_sprintf(src), ETHERTYPE_IP,
    534 		    ether_sprintf(eh->ether_dhost),
    535 		    ether_sprintf(eh->ether_shost), ntohs(eh->ether_type));
    536 		error = EIO;
    537 	}
    538 
    539  done:
    540 	/* Turn interface off, release resources */
    541 
    542 	sc->re_testmode = 0;
    543 	ifp->if_flags &= ~IFF_PROMISC;
    544 	re_stop(ifp, 0);
    545 	if (m0 != NULL)
    546 		m_freem(m0);
    547 
    548 	return error;
    549 }
    550 
    551 
    552 /*
    553  * Attach the interface. Allocate softc structures, do ifmedia
    554  * setup and ethernet/BPF attach.
    555  */
    556 void
    557 re_attach(struct rtk_softc *sc)
    558 {
    559 	uint8_t eaddr[ETHER_ADDR_LEN];
    560 	uint16_t val;
    561 	struct ifnet *ifp;
    562 	int error = 0, i, addr_len;
    563 
    564 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0) {
    565 		uint32_t hwrev;
    566 
    567 		/* Revision of 8169/8169S/8110s in bits 30..26, 23 */
    568 		hwrev = CSR_READ_4(sc, RTK_TXCFG) & RTK_TXCFG_HWREV;
    569 		switch (hwrev) {
    570 		case RTK_HWREV_8169:
    571 			sc->sc_quirk |= RTKQ_8169NONS;
    572 			break;
    573 		case RTK_HWREV_8169S:
    574 		case RTK_HWREV_8110S:
    575 		case RTK_HWREV_8169_8110SB:
    576 		case RTK_HWREV_8169_8110SC:
    577 			sc->sc_quirk |= RTKQ_MACLDPS;
    578 			break;
    579 		case RTK_HWREV_8168_SPIN1:
    580 		case RTK_HWREV_8168_SPIN2:
    581 		case RTK_HWREV_8168_SPIN3:
    582 			sc->sc_quirk |= RTKQ_MACSTAT;
    583 			break;
    584 		case RTK_HWREV_8168C:
    585 		case RTK_HWREV_8168C_SPIN2:
    586 		case RTK_HWREV_8168CP:
    587 		case RTK_HWREV_8168D:
    588 		case RTK_HWREV_8168DP:
    589 			sc->sc_quirk |= RTKQ_DESCV2 | RTKQ_NOEECMD |
    590 			    RTKQ_MACSTAT | RTKQ_CMDSTOP;
    591 			/*
    592 			 * From FreeBSD driver:
    593 			 *
    594 			 * These (8168/8111) controllers support jumbo frame
    595 			 * but it seems that enabling it requires touching
    596 			 * additional magic registers. Depending on MAC
    597 			 * revisions some controllers need to disable
    598 			 * checksum offload. So disable jumbo frame until
    599 			 * I have better idea what it really requires to
    600 			 * make it support.
    601 			 * RTL8168C/CP : supports up to 6KB jumbo frame.
    602 			 * RTL8111C/CP : supports up to 9KB jumbo frame.
    603 			 */
    604 			sc->sc_quirk |= RTKQ_NOJUMBO;
    605 			break;
    606 		case RTK_HWREV_8100E:
    607 		case RTK_HWREV_8100E_SPIN2:
    608 		case RTK_HWREV_8101E:
    609 			sc->sc_quirk |= RTKQ_NOJUMBO;
    610 			break;
    611 		case RTK_HWREV_8102E:
    612 		case RTK_HWREV_8102EL:
    613 		case RTK_HWREV_8103E:
    614 			sc->sc_quirk |= RTKQ_DESCV2 | RTKQ_NOEECMD |
    615 			    RTKQ_MACSTAT | RTKQ_CMDSTOP | RTKQ_NOJUMBO;
    616 			break;
    617 		default:
    618 			aprint_normal_dev(sc->sc_dev,
    619 			    "Unknown revision (0x%08x)\n", hwrev);
    620 			/* assume the latest features */
    621 			sc->sc_quirk |= RTKQ_DESCV2 | RTKQ_NOEECMD;
    622 			sc->sc_quirk |= RTKQ_NOJUMBO;
    623 		}
    624 
    625 		/* Set RX length mask */
    626 		sc->re_rxlenmask = RE_RDESC_STAT_GFRAGLEN;
    627 		sc->re_ldata.re_tx_desc_cnt = RE_TX_DESC_CNT_8169;
    628 	} else {
    629 		sc->sc_quirk |= RTKQ_NOJUMBO;
    630 
    631 		/* Set RX length mask */
    632 		sc->re_rxlenmask = RE_RDESC_STAT_FRAGLEN;
    633 		sc->re_ldata.re_tx_desc_cnt = RE_TX_DESC_CNT_8139;
    634 	}
    635 
    636 	/* Reset the adapter. */
    637 	re_reset(sc);
    638 
    639 	if ((sc->sc_quirk & RTKQ_NOEECMD) != 0) {
    640 		/*
    641 		 * Get station address from ID registers.
    642 		 */
    643 		for (i = 0; i < ETHER_ADDR_LEN; i++)
    644 			eaddr[i] = CSR_READ_1(sc, RTK_IDR0 + i);
    645 	} else {
    646 		/*
    647 		 * Get station address from the EEPROM.
    648 		 */
    649 		if (rtk_read_eeprom(sc, RTK_EE_ID, RTK_EEADDR_LEN1) == 0x8129)
    650 			addr_len = RTK_EEADDR_LEN1;
    651 		else
    652 			addr_len = RTK_EEADDR_LEN0;
    653 
    654 		/*
    655 		 * Get station address from the EEPROM.
    656 		 */
    657 		for (i = 0; i < ETHER_ADDR_LEN / 2; i++) {
    658 			val = rtk_read_eeprom(sc, RTK_EE_EADDR0 + i, addr_len);
    659 			eaddr[(i * 2) + 0] = val & 0xff;
    660 			eaddr[(i * 2) + 1] = val >> 8;
    661 		}
    662 	}
    663 
    664 	aprint_normal_dev(sc->sc_dev, "Ethernet address %s\n",
    665 	    ether_sprintf(eaddr));
    666 
    667 	if (sc->re_ldata.re_tx_desc_cnt >
    668 	    PAGE_SIZE / sizeof(struct re_desc)) {
    669 		sc->re_ldata.re_tx_desc_cnt =
    670 		    PAGE_SIZE / sizeof(struct re_desc);
    671 	}
    672 
    673 	aprint_verbose_dev(sc->sc_dev, "using %d tx descriptors\n",
    674 	    sc->re_ldata.re_tx_desc_cnt);
    675 	KASSERT(RE_NEXT_TX_DESC(sc, RE_TX_DESC_CNT(sc) - 1) == 0);
    676 
    677 	/* Allocate DMA'able memory for the TX ring */
    678 	if ((error = bus_dmamem_alloc(sc->sc_dmat, RE_TX_LIST_SZ(sc),
    679 	    RE_RING_ALIGN, 0, &sc->re_ldata.re_tx_listseg, 1,
    680 	    &sc->re_ldata.re_tx_listnseg, BUS_DMA_NOWAIT)) != 0) {
    681 		aprint_error_dev(sc->sc_dev,
    682 		    "can't allocate tx listseg, error = %d\n", error);
    683 		goto fail_0;
    684 	}
    685 
    686 	/* Load the map for the TX ring. */
    687 	if ((error = bus_dmamem_map(sc->sc_dmat, &sc->re_ldata.re_tx_listseg,
    688 	    sc->re_ldata.re_tx_listnseg, RE_TX_LIST_SZ(sc),
    689 	    (void **)&sc->re_ldata.re_tx_list,
    690 	    BUS_DMA_COHERENT | BUS_DMA_NOWAIT)) != 0) {
    691 		aprint_error_dev(sc->sc_dev,
    692 		    "can't map tx list, error = %d\n", error);
    693 	  	goto fail_1;
    694 	}
    695 	memset(sc->re_ldata.re_tx_list, 0, RE_TX_LIST_SZ(sc));
    696 
    697 	if ((error = bus_dmamap_create(sc->sc_dmat, RE_TX_LIST_SZ(sc), 1,
    698 	    RE_TX_LIST_SZ(sc), 0, 0,
    699 	    &sc->re_ldata.re_tx_list_map)) != 0) {
    700 		aprint_error_dev(sc->sc_dev,
    701 		    "can't create tx list map, error = %d\n", error);
    702 		goto fail_2;
    703 	}
    704 
    705 
    706 	if ((error = bus_dmamap_load(sc->sc_dmat,
    707 	    sc->re_ldata.re_tx_list_map, sc->re_ldata.re_tx_list,
    708 	    RE_TX_LIST_SZ(sc), NULL, BUS_DMA_NOWAIT)) != 0) {
    709 		aprint_error_dev(sc->sc_dev,
    710 		    "can't load tx list, error = %d\n", error);
    711 		goto fail_3;
    712 	}
    713 
    714 	/* Create DMA maps for TX buffers */
    715 	for (i = 0; i < RE_TX_QLEN; i++) {
    716 		error = bus_dmamap_create(sc->sc_dmat,
    717 		    round_page(IP_MAXPACKET),
    718 		    RE_TX_DESC_CNT(sc), RE_TDESC_CMD_FRAGLEN,
    719 		    0, 0, &sc->re_ldata.re_txq[i].txq_dmamap);
    720 		if (error) {
    721 			aprint_error_dev(sc->sc_dev,
    722 			    "can't create DMA map for TX\n");
    723 			goto fail_4;
    724 		}
    725 	}
    726 
    727 	/* Allocate DMA'able memory for the RX ring */
    728 	/* XXX see also a comment about RE_RX_DMAMEM_SZ in rtl81x9var.h */
    729 	if ((error = bus_dmamem_alloc(sc->sc_dmat,
    730 	    RE_RX_DMAMEM_SZ, RE_RING_ALIGN, 0, &sc->re_ldata.re_rx_listseg, 1,
    731 	    &sc->re_ldata.re_rx_listnseg, BUS_DMA_NOWAIT)) != 0) {
    732 		aprint_error_dev(sc->sc_dev,
    733 		    "can't allocate rx listseg, error = %d\n", error);
    734 		goto fail_4;
    735 	}
    736 
    737 	/* Load the map for the RX ring. */
    738 	if ((error = bus_dmamem_map(sc->sc_dmat, &sc->re_ldata.re_rx_listseg,
    739 	    sc->re_ldata.re_rx_listnseg, RE_RX_DMAMEM_SZ,
    740 	    (void **)&sc->re_ldata.re_rx_list,
    741 	    BUS_DMA_COHERENT | BUS_DMA_NOWAIT)) != 0) {
    742 		aprint_error_dev(sc->sc_dev,
    743 		    "can't map rx list, error = %d\n", error);
    744 		goto fail_5;
    745 	}
    746 	memset(sc->re_ldata.re_rx_list, 0, RE_RX_DMAMEM_SZ);
    747 
    748 	if ((error = bus_dmamap_create(sc->sc_dmat,
    749 	    RE_RX_DMAMEM_SZ, 1, RE_RX_DMAMEM_SZ, 0, 0,
    750 	    &sc->re_ldata.re_rx_list_map)) != 0) {
    751 		aprint_error_dev(sc->sc_dev,
    752 		    "can't create rx list map, error = %d\n", error);
    753 		goto fail_6;
    754 	}
    755 
    756 	if ((error = bus_dmamap_load(sc->sc_dmat,
    757 	    sc->re_ldata.re_rx_list_map, sc->re_ldata.re_rx_list,
    758 	    RE_RX_DMAMEM_SZ, NULL, BUS_DMA_NOWAIT)) != 0) {
    759 		aprint_error_dev(sc->sc_dev,
    760 		    "can't load rx list, error = %d\n", error);
    761 		goto fail_7;
    762 	}
    763 
    764 	/* Create DMA maps for RX buffers */
    765 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
    766 		error = bus_dmamap_create(sc->sc_dmat, MCLBYTES, 1, MCLBYTES,
    767 		    0, 0, &sc->re_ldata.re_rxsoft[i].rxs_dmamap);
    768 		if (error) {
    769 			aprint_error_dev(sc->sc_dev,
    770 			    "can't create DMA map for RX\n");
    771 			goto fail_8;
    772 		}
    773 	}
    774 
    775 	/*
    776 	 * Record interface as attached. From here, we should not fail.
    777 	 */
    778 	sc->sc_flags |= RTK_ATTACHED;
    779 
    780 	ifp = &sc->ethercom.ec_if;
    781 	ifp->if_softc = sc;
    782 	strlcpy(ifp->if_xname, device_xname(sc->sc_dev), IFNAMSIZ);
    783 	ifp->if_mtu = ETHERMTU;
    784 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
    785 	ifp->if_ioctl = re_ioctl;
    786 	sc->ethercom.ec_capabilities |=
    787 	    ETHERCAP_VLAN_MTU | ETHERCAP_VLAN_HWTAGGING;
    788 	ifp->if_start = re_start;
    789 	ifp->if_stop = re_stop;
    790 
    791 	/*
    792 	 * IFCAP_CSUM_IPv4_Tx on re(4) is broken for small packets,
    793 	 * so we have a workaround to handle the bug by padding
    794 	 * such packets manually.
    795 	 */
    796 	ifp->if_capabilities |=
    797 	    IFCAP_CSUM_IPv4_Tx | IFCAP_CSUM_IPv4_Rx |
    798 	    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
    799 	    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx |
    800 	    IFCAP_TSOv4;
    801 
    802 	/*
    803 	 * XXX
    804 	 * Still have no idea how to make TSO work on 8168C, 8168CP,
    805 	 * 8102E, 8111C and 8111CP.
    806 	 */
    807 	if ((sc->sc_quirk & RTKQ_DESCV2) != 0)
    808 		ifp->if_capabilities &= ~IFCAP_TSOv4;
    809 
    810 	ifp->if_watchdog = re_watchdog;
    811 	ifp->if_init = re_init;
    812 	ifp->if_snd.ifq_maxlen = RE_IFQ_MAXLEN;
    813 	ifp->if_capenable = ifp->if_capabilities;
    814 	IFQ_SET_READY(&ifp->if_snd);
    815 
    816 	callout_init(&sc->rtk_tick_ch, 0);
    817 
    818 	/* Do MII setup */
    819 	sc->mii.mii_ifp = ifp;
    820 	sc->mii.mii_readreg = re_miibus_readreg;
    821 	sc->mii.mii_writereg = re_miibus_writereg;
    822 	sc->mii.mii_statchg = re_miibus_statchg;
    823 	sc->ethercom.ec_mii = &sc->mii;
    824 	ifmedia_init(&sc->mii.mii_media, IFM_IMASK, ether_mediachange,
    825 	    ether_mediastatus);
    826 	mii_attach(sc->sc_dev, &sc->mii, 0xffffffff, MII_PHY_ANY,
    827 	    MII_OFFSET_ANY, 0);
    828 	ifmedia_set(&sc->mii.mii_media, IFM_ETHER | IFM_AUTO);
    829 
    830 	/*
    831 	 * Call MI attach routine.
    832 	 */
    833 	if_attach(ifp);
    834 	ether_ifattach(ifp, eaddr);
    835 
    836 	if (pmf_device_register(sc->sc_dev, NULL, NULL))
    837 		pmf_class_network_register(sc->sc_dev, ifp);
    838 	else
    839 		aprint_error_dev(sc->sc_dev,
    840 		    "couldn't establish power handler\n");
    841 
    842 	return;
    843 
    844  fail_8:
    845 	/* Destroy DMA maps for RX buffers. */
    846 	for (i = 0; i < RE_RX_DESC_CNT; i++)
    847 		if (sc->re_ldata.re_rxsoft[i].rxs_dmamap != NULL)
    848 			bus_dmamap_destroy(sc->sc_dmat,
    849 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
    850 
    851 	/* Free DMA'able memory for the RX ring. */
    852 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
    853  fail_7:
    854 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
    855  fail_6:
    856 	bus_dmamem_unmap(sc->sc_dmat,
    857 	    (void *)sc->re_ldata.re_rx_list, RE_RX_DMAMEM_SZ);
    858  fail_5:
    859 	bus_dmamem_free(sc->sc_dmat,
    860 	    &sc->re_ldata.re_rx_listseg, sc->re_ldata.re_rx_listnseg);
    861 
    862  fail_4:
    863 	/* Destroy DMA maps for TX buffers. */
    864 	for (i = 0; i < RE_TX_QLEN; i++)
    865 		if (sc->re_ldata.re_txq[i].txq_dmamap != NULL)
    866 			bus_dmamap_destroy(sc->sc_dmat,
    867 			    sc->re_ldata.re_txq[i].txq_dmamap);
    868 
    869 	/* Free DMA'able memory for the TX ring. */
    870 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
    871  fail_3:
    872 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
    873  fail_2:
    874 	bus_dmamem_unmap(sc->sc_dmat,
    875 	    (void *)sc->re_ldata.re_tx_list, RE_TX_LIST_SZ(sc));
    876  fail_1:
    877 	bus_dmamem_free(sc->sc_dmat,
    878 	    &sc->re_ldata.re_tx_listseg, sc->re_ldata.re_tx_listnseg);
    879  fail_0:
    880 	return;
    881 }
    882 
    883 
    884 /*
    885  * re_activate:
    886  *     Handle device activation/deactivation requests.
    887  */
    888 int
    889 re_activate(device_t self, enum devact act)
    890 {
    891 	struct rtk_softc *sc = device_private(self);
    892 	int s, error = 0;
    893 
    894 	s = splnet();
    895 	switch (act) {
    896 	case DVACT_ACTIVATE:
    897 		error = EOPNOTSUPP;
    898 		break;
    899 	case DVACT_DEACTIVATE:
    900 		mii_activate(&sc->mii, act, MII_PHY_ANY, MII_OFFSET_ANY);
    901 		if_deactivate(&sc->ethercom.ec_if);
    902 		break;
    903 	}
    904 	splx(s);
    905 
    906 	return error;
    907 }
    908 
    909 /*
    910  * re_detach:
    911  *     Detach a rtk interface.
    912  */
    913 int
    914 re_detach(struct rtk_softc *sc)
    915 {
    916 	struct ifnet *ifp = &sc->ethercom.ec_if;
    917 	int i;
    918 
    919 	/*
    920 	 * Succeed now if there isn't any work to do.
    921 	 */
    922 	if ((sc->sc_flags & RTK_ATTACHED) == 0)
    923 		return 0;
    924 
    925 	/* Unhook our tick handler. */
    926 	callout_stop(&sc->rtk_tick_ch);
    927 
    928 	/* Detach all PHYs. */
    929 	mii_detach(&sc->mii, MII_PHY_ANY, MII_OFFSET_ANY);
    930 
    931 	/* Delete all remaining media. */
    932 	ifmedia_delete_instance(&sc->mii.mii_media, IFM_INST_ANY);
    933 
    934 	ether_ifdetach(ifp);
    935 	if_detach(ifp);
    936 
    937 	/* Destroy DMA maps for RX buffers. */
    938 	for (i = 0; i < RE_RX_DESC_CNT; i++)
    939 		if (sc->re_ldata.re_rxsoft[i].rxs_dmamap != NULL)
    940 			bus_dmamap_destroy(sc->sc_dmat,
    941 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
    942 
    943 	/* Free DMA'able memory for the RX ring. */
    944 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
    945 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
    946 	bus_dmamem_unmap(sc->sc_dmat,
    947 	    (void *)sc->re_ldata.re_rx_list, RE_RX_DMAMEM_SZ);
    948 	bus_dmamem_free(sc->sc_dmat,
    949 	    &sc->re_ldata.re_rx_listseg, sc->re_ldata.re_rx_listnseg);
    950 
    951 	/* Destroy DMA maps for TX buffers. */
    952 	for (i = 0; i < RE_TX_QLEN; i++)
    953 		if (sc->re_ldata.re_txq[i].txq_dmamap != NULL)
    954 			bus_dmamap_destroy(sc->sc_dmat,
    955 			    sc->re_ldata.re_txq[i].txq_dmamap);
    956 
    957 	/* Free DMA'able memory for the TX ring. */
    958 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
    959 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
    960 	bus_dmamem_unmap(sc->sc_dmat,
    961 	    (void *)sc->re_ldata.re_tx_list, RE_TX_LIST_SZ(sc));
    962 	bus_dmamem_free(sc->sc_dmat,
    963 	    &sc->re_ldata.re_tx_listseg, sc->re_ldata.re_tx_listnseg);
    964 
    965 	pmf_device_deregister(sc->sc_dev);
    966 
    967 	return 0;
    968 }
    969 
    970 /*
    971  * re_enable:
    972  *     Enable the RTL81X9 chip.
    973  */
    974 static int
    975 re_enable(struct rtk_softc *sc)
    976 {
    977 
    978 	if (RTK_IS_ENABLED(sc) == 0 && sc->sc_enable != NULL) {
    979 		if ((*sc->sc_enable)(sc) != 0) {
    980 			printf("%s: device enable failed\n",
    981 			    device_xname(sc->sc_dev));
    982 			return EIO;
    983 		}
    984 		sc->sc_flags |= RTK_ENABLED;
    985 	}
    986 	return 0;
    987 }
    988 
    989 /*
    990  * re_disable:
    991  *     Disable the RTL81X9 chip.
    992  */
    993 static void
    994 re_disable(struct rtk_softc *sc)
    995 {
    996 
    997 	if (RTK_IS_ENABLED(sc) && sc->sc_disable != NULL) {
    998 		(*sc->sc_disable)(sc);
    999 		sc->sc_flags &= ~RTK_ENABLED;
   1000 	}
   1001 }
   1002 
   1003 static int
   1004 re_newbuf(struct rtk_softc *sc, int idx, struct mbuf *m)
   1005 {
   1006 	struct mbuf *n = NULL;
   1007 	bus_dmamap_t map;
   1008 	struct re_desc *d;
   1009 	struct re_rxsoft *rxs;
   1010 	uint32_t cmdstat;
   1011 	int error;
   1012 
   1013 	if (m == NULL) {
   1014 		MGETHDR(n, M_DONTWAIT, MT_DATA);
   1015 		if (n == NULL)
   1016 			return ENOBUFS;
   1017 
   1018 		MCLGET(n, M_DONTWAIT);
   1019 		if ((n->m_flags & M_EXT) == 0) {
   1020 			m_freem(n);
   1021 			return ENOBUFS;
   1022 		}
   1023 		m = n;
   1024 	} else
   1025 		m->m_data = m->m_ext.ext_buf;
   1026 
   1027 	/*
   1028 	 * Initialize mbuf length fields and fixup
   1029 	 * alignment so that the frame payload is
   1030 	 * longword aligned.
   1031 	 */
   1032 	m->m_len = m->m_pkthdr.len = MCLBYTES - RE_ETHER_ALIGN;
   1033 	m->m_data += RE_ETHER_ALIGN;
   1034 
   1035 	rxs = &sc->re_ldata.re_rxsoft[idx];
   1036 	map = rxs->rxs_dmamap;
   1037 	error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m,
   1038 	    BUS_DMA_READ|BUS_DMA_NOWAIT);
   1039 
   1040 	if (error)
   1041 		goto out;
   1042 
   1043 	bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
   1044 	    BUS_DMASYNC_PREREAD);
   1045 
   1046 	d = &sc->re_ldata.re_rx_list[idx];
   1047 #ifdef DIAGNOSTIC
   1048 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
   1049 	cmdstat = le32toh(d->re_cmdstat);
   1050 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD);
   1051 	if (cmdstat & RE_RDESC_STAT_OWN) {
   1052 		panic("%s: tried to map busy RX descriptor",
   1053 		    device_xname(sc->sc_dev));
   1054 	}
   1055 #endif
   1056 
   1057 	rxs->rxs_mbuf = m;
   1058 
   1059 	d->re_vlanctl = 0;
   1060 	cmdstat = map->dm_segs[0].ds_len;
   1061 	if (idx == (RE_RX_DESC_CNT - 1))
   1062 		cmdstat |= RE_RDESC_CMD_EOR;
   1063 	re_set_bufaddr(d, map->dm_segs[0].ds_addr);
   1064 	d->re_cmdstat = htole32(cmdstat);
   1065 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1066 	cmdstat |= RE_RDESC_CMD_OWN;
   1067 	d->re_cmdstat = htole32(cmdstat);
   1068 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1069 
   1070 	return 0;
   1071  out:
   1072 	if (n != NULL)
   1073 		m_freem(n);
   1074 	return ENOMEM;
   1075 }
   1076 
   1077 static int
   1078 re_tx_list_init(struct rtk_softc *sc)
   1079 {
   1080 	int i;
   1081 
   1082 	memset(sc->re_ldata.re_tx_list, 0, RE_TX_LIST_SZ(sc));
   1083 	for (i = 0; i < RE_TX_QLEN; i++) {
   1084 		sc->re_ldata.re_txq[i].txq_mbuf = NULL;
   1085 	}
   1086 
   1087 	bus_dmamap_sync(sc->sc_dmat,
   1088 	    sc->re_ldata.re_tx_list_map, 0,
   1089 	    sc->re_ldata.re_tx_list_map->dm_mapsize,
   1090 	    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1091 	sc->re_ldata.re_txq_prodidx = 0;
   1092 	sc->re_ldata.re_txq_considx = 0;
   1093 	sc->re_ldata.re_txq_free = RE_TX_QLEN;
   1094 	sc->re_ldata.re_tx_free = RE_TX_DESC_CNT(sc);
   1095 	sc->re_ldata.re_tx_nextfree = 0;
   1096 
   1097 	return 0;
   1098 }
   1099 
   1100 static int
   1101 re_rx_list_init(struct rtk_softc *sc)
   1102 {
   1103 	int i;
   1104 
   1105 	memset(sc->re_ldata.re_rx_list, 0, RE_RX_LIST_SZ);
   1106 
   1107 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
   1108 		if (re_newbuf(sc, i, NULL) == ENOBUFS)
   1109 			return ENOBUFS;
   1110 	}
   1111 
   1112 	sc->re_ldata.re_rx_prodidx = 0;
   1113 	sc->re_head = sc->re_tail = NULL;
   1114 
   1115 	return 0;
   1116 }
   1117 
   1118 /*
   1119  * RX handler for C+ and 8169. For the gigE chips, we support
   1120  * the reception of jumbo frames that have been fragmented
   1121  * across multiple 2K mbuf cluster buffers.
   1122  */
   1123 static void
   1124 re_rxeof(struct rtk_softc *sc)
   1125 {
   1126 	struct mbuf *m;
   1127 	struct ifnet *ifp;
   1128 	int i, total_len;
   1129 	struct re_desc *cur_rx;
   1130 	struct re_rxsoft *rxs;
   1131 	uint32_t rxstat, rxvlan;
   1132 
   1133 	ifp = &sc->ethercom.ec_if;
   1134 
   1135 	for (i = sc->re_ldata.re_rx_prodidx;; i = RE_NEXT_RX_DESC(sc, i)) {
   1136 		cur_rx = &sc->re_ldata.re_rx_list[i];
   1137 		RE_RXDESCSYNC(sc, i,
   1138 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
   1139 		rxstat = le32toh(cur_rx->re_cmdstat);
   1140 		rxvlan = le32toh(cur_rx->re_vlanctl);
   1141 		RE_RXDESCSYNC(sc, i, BUS_DMASYNC_PREREAD);
   1142 		if ((rxstat & RE_RDESC_STAT_OWN) != 0) {
   1143 			break;
   1144 		}
   1145 		total_len = rxstat & sc->re_rxlenmask;
   1146 		rxs = &sc->re_ldata.re_rxsoft[i];
   1147 		m = rxs->rxs_mbuf;
   1148 
   1149 		/* Invalidate the RX mbuf and unload its map */
   1150 
   1151 		bus_dmamap_sync(sc->sc_dmat,
   1152 		    rxs->rxs_dmamap, 0, rxs->rxs_dmamap->dm_mapsize,
   1153 		    BUS_DMASYNC_POSTREAD);
   1154 		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
   1155 
   1156 		if ((rxstat & RE_RDESC_STAT_EOF) == 0) {
   1157 			m->m_len = MCLBYTES - RE_ETHER_ALIGN;
   1158 			if (sc->re_head == NULL)
   1159 				sc->re_head = sc->re_tail = m;
   1160 			else {
   1161 				m->m_flags &= ~M_PKTHDR;
   1162 				sc->re_tail->m_next = m;
   1163 				sc->re_tail = m;
   1164 			}
   1165 			re_newbuf(sc, i, NULL);
   1166 			continue;
   1167 		}
   1168 
   1169 		/*
   1170 		 * NOTE: for the 8139C+, the frame length field
   1171 		 * is always 12 bits in size, but for the gigE chips,
   1172 		 * it is 13 bits (since the max RX frame length is 16K).
   1173 		 * Unfortunately, all 32 bits in the status word
   1174 		 * were already used, so to make room for the extra
   1175 		 * length bit, RealTek took out the 'frame alignment
   1176 		 * error' bit and shifted the other status bits
   1177 		 * over one slot. The OWN, EOR, FS and LS bits are
   1178 		 * still in the same places. We have already extracted
   1179 		 * the frame length and checked the OWN bit, so rather
   1180 		 * than using an alternate bit mapping, we shift the
   1181 		 * status bits one space to the right so we can evaluate
   1182 		 * them using the 8169 status as though it was in the
   1183 		 * same format as that of the 8139C+.
   1184 		 */
   1185 		if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0)
   1186 			rxstat >>= 1;
   1187 
   1188 		if (__predict_false((rxstat & RE_RDESC_STAT_RXERRSUM) != 0)) {
   1189 #ifdef RE_DEBUG
   1190 			printf("%s: RX error (rxstat = 0x%08x)",
   1191 			    device_xname(sc->sc_dev), rxstat);
   1192 			if (rxstat & RE_RDESC_STAT_FRALIGN)
   1193 				printf(", frame alignment error");
   1194 			if (rxstat & RE_RDESC_STAT_BUFOFLOW)
   1195 				printf(", out of buffer space");
   1196 			if (rxstat & RE_RDESC_STAT_FIFOOFLOW)
   1197 				printf(", FIFO overrun");
   1198 			if (rxstat & RE_RDESC_STAT_GIANT)
   1199 				printf(", giant packet");
   1200 			if (rxstat & RE_RDESC_STAT_RUNT)
   1201 				printf(", runt packet");
   1202 			if (rxstat & RE_RDESC_STAT_CRCERR)
   1203 				printf(", CRC error");
   1204 			printf("\n");
   1205 #endif
   1206 			ifp->if_ierrors++;
   1207 			/*
   1208 			 * If this is part of a multi-fragment packet,
   1209 			 * discard all the pieces.
   1210 			 */
   1211 			if (sc->re_head != NULL) {
   1212 				m_freem(sc->re_head);
   1213 				sc->re_head = sc->re_tail = NULL;
   1214 			}
   1215 			re_newbuf(sc, i, m);
   1216 			continue;
   1217 		}
   1218 
   1219 		/*
   1220 		 * If allocating a replacement mbuf fails,
   1221 		 * reload the current one.
   1222 		 */
   1223 
   1224 		if (__predict_false(re_newbuf(sc, i, NULL) != 0)) {
   1225 			ifp->if_ierrors++;
   1226 			if (sc->re_head != NULL) {
   1227 				m_freem(sc->re_head);
   1228 				sc->re_head = sc->re_tail = NULL;
   1229 			}
   1230 			re_newbuf(sc, i, m);
   1231 			continue;
   1232 		}
   1233 
   1234 		if (sc->re_head != NULL) {
   1235 			m->m_len = total_len % (MCLBYTES - RE_ETHER_ALIGN);
   1236 			/*
   1237 			 * Special case: if there's 4 bytes or less
   1238 			 * in this buffer, the mbuf can be discarded:
   1239 			 * the last 4 bytes is the CRC, which we don't
   1240 			 * care about anyway.
   1241 			 */
   1242 			if (m->m_len <= ETHER_CRC_LEN) {
   1243 				sc->re_tail->m_len -=
   1244 				    (ETHER_CRC_LEN - m->m_len);
   1245 				m_freem(m);
   1246 			} else {
   1247 				m->m_len -= ETHER_CRC_LEN;
   1248 				m->m_flags &= ~M_PKTHDR;
   1249 				sc->re_tail->m_next = m;
   1250 			}
   1251 			m = sc->re_head;
   1252 			sc->re_head = sc->re_tail = NULL;
   1253 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
   1254 		} else
   1255 			m->m_pkthdr.len = m->m_len =
   1256 			    (total_len - ETHER_CRC_LEN);
   1257 
   1258 		ifp->if_ipackets++;
   1259 		m->m_pkthdr.rcvif = ifp;
   1260 
   1261 		/* Do RX checksumming */
   1262 		if ((sc->sc_quirk & RTKQ_DESCV2) == 0) {
   1263 			/* Check IP header checksum */
   1264 			if ((rxstat & RE_RDESC_STAT_PROTOID) != 0) {
   1265 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
   1266 				if (rxstat & RE_RDESC_STAT_IPSUMBAD)
   1267 					m->m_pkthdr.csum_flags |=
   1268 					    M_CSUM_IPv4_BAD;
   1269 
   1270 				/* Check TCP/UDP checksum */
   1271 				if (RE_TCPPKT(rxstat)) {
   1272 					m->m_pkthdr.csum_flags |= M_CSUM_TCPv4;
   1273 					if (rxstat & RE_RDESC_STAT_TCPSUMBAD)
   1274 						m->m_pkthdr.csum_flags |=
   1275 						    M_CSUM_TCP_UDP_BAD;
   1276 				} else if (RE_UDPPKT(rxstat)) {
   1277 					m->m_pkthdr.csum_flags |= M_CSUM_UDPv4;
   1278 					if (rxstat & RE_RDESC_STAT_UDPSUMBAD)
   1279 						m->m_pkthdr.csum_flags |=
   1280 						    M_CSUM_TCP_UDP_BAD;
   1281 				}
   1282 			}
   1283 		} else {
   1284 			/* Check IPv4 header checksum */
   1285 			if ((rxvlan & RE_RDESC_VLANCTL_IPV4) != 0) {
   1286 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
   1287 				if (rxstat & RE_RDESC_STAT_IPSUMBAD)
   1288 					m->m_pkthdr.csum_flags |=
   1289 					    M_CSUM_IPv4_BAD;
   1290 
   1291 				/* Check TCPv4/UDPv4 checksum */
   1292 				if (RE_TCPPKT(rxstat)) {
   1293 					m->m_pkthdr.csum_flags |= M_CSUM_TCPv4;
   1294 					if (rxstat & RE_RDESC_STAT_TCPSUMBAD)
   1295 						m->m_pkthdr.csum_flags |=
   1296 						    M_CSUM_TCP_UDP_BAD;
   1297 				} else if (RE_UDPPKT(rxstat)) {
   1298 					m->m_pkthdr.csum_flags |= M_CSUM_UDPv4;
   1299 					if (rxstat & RE_RDESC_STAT_UDPSUMBAD)
   1300 						m->m_pkthdr.csum_flags |=
   1301 						    M_CSUM_TCP_UDP_BAD;
   1302 				}
   1303 			}
   1304 			/* XXX Check TCPv6/UDPv6 checksum? */
   1305 		}
   1306 
   1307 		if (rxvlan & RE_RDESC_VLANCTL_TAG) {
   1308 			VLAN_INPUT_TAG(ifp, m,
   1309 			     bswap16(rxvlan & RE_RDESC_VLANCTL_DATA),
   1310 			     continue);
   1311 		}
   1312 #if NBPFILTER > 0
   1313 		if (ifp->if_bpf)
   1314 			bpf_mtap(ifp->if_bpf, m);
   1315 #endif
   1316 		(*ifp->if_input)(ifp, m);
   1317 	}
   1318 
   1319 	sc->re_ldata.re_rx_prodidx = i;
   1320 }
   1321 
   1322 static void
   1323 re_txeof(struct rtk_softc *sc)
   1324 {
   1325 	struct ifnet *ifp;
   1326 	struct re_txq *txq;
   1327 	uint32_t txstat;
   1328 	int idx, descidx;
   1329 
   1330 	ifp = &sc->ethercom.ec_if;
   1331 
   1332 	for (idx = sc->re_ldata.re_txq_considx;
   1333 	    sc->re_ldata.re_txq_free < RE_TX_QLEN;
   1334 	    idx = RE_NEXT_TXQ(sc, idx), sc->re_ldata.re_txq_free++) {
   1335 		txq = &sc->re_ldata.re_txq[idx];
   1336 		KASSERT(txq->txq_mbuf != NULL);
   1337 
   1338 		descidx = txq->txq_descidx;
   1339 		RE_TXDESCSYNC(sc, descidx,
   1340 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
   1341 		txstat =
   1342 		    le32toh(sc->re_ldata.re_tx_list[descidx].re_cmdstat);
   1343 		RE_TXDESCSYNC(sc, descidx, BUS_DMASYNC_PREREAD);
   1344 		KASSERT((txstat & RE_TDESC_CMD_EOF) != 0);
   1345 		if (txstat & RE_TDESC_CMD_OWN) {
   1346 			break;
   1347 		}
   1348 
   1349 		sc->re_ldata.re_tx_free += txq->txq_nsegs;
   1350 		KASSERT(sc->re_ldata.re_tx_free <= RE_TX_DESC_CNT(sc));
   1351 		bus_dmamap_sync(sc->sc_dmat, txq->txq_dmamap,
   1352 		    0, txq->txq_dmamap->dm_mapsize, BUS_DMASYNC_POSTWRITE);
   1353 		bus_dmamap_unload(sc->sc_dmat, txq->txq_dmamap);
   1354 		m_freem(txq->txq_mbuf);
   1355 		txq->txq_mbuf = NULL;
   1356 
   1357 		if (txstat & (RE_TDESC_STAT_EXCESSCOL | RE_TDESC_STAT_COLCNT))
   1358 			ifp->if_collisions++;
   1359 		if (txstat & RE_TDESC_STAT_TXERRSUM)
   1360 			ifp->if_oerrors++;
   1361 		else
   1362 			ifp->if_opackets++;
   1363 	}
   1364 
   1365 	sc->re_ldata.re_txq_considx = idx;
   1366 
   1367 	if (sc->re_ldata.re_txq_free > RE_NTXDESC_RSVD)
   1368 		ifp->if_flags &= ~IFF_OACTIVE;
   1369 
   1370 	/*
   1371 	 * If not all descriptors have been released reaped yet,
   1372 	 * reload the timer so that we will eventually get another
   1373 	 * interrupt that will cause us to re-enter this routine.
   1374 	 * This is done in case the transmitter has gone idle.
   1375 	 */
   1376 	if (sc->re_ldata.re_txq_free < RE_TX_QLEN) {
   1377 		CSR_WRITE_4(sc, RTK_TIMERCNT, 1);
   1378 		if ((sc->sc_quirk & RTKQ_PCIE) != 0) {
   1379 			/*
   1380 			 * Some chips will ignore a second TX request
   1381 			 * issued while an existing transmission is in
   1382 			 * progress. If the transmitter goes idle but
   1383 			 * there are still packets waiting to be sent,
   1384 			 * we need to restart the channel here to flush
   1385 			 * them out. This only seems to be required with
   1386 			 * the PCIe devices.
   1387 			 */
   1388 			CSR_WRITE_1(sc, RTK_GTXSTART, RTK_TXSTART_START);
   1389 		}
   1390 	} else
   1391 		ifp->if_timer = 0;
   1392 }
   1393 
   1394 static void
   1395 re_tick(void *arg)
   1396 {
   1397 	struct rtk_softc *sc = arg;
   1398 	int s;
   1399 
   1400 	/* XXX: just return for 8169S/8110S with rev 2 or newer phy */
   1401 	s = splnet();
   1402 
   1403 	mii_tick(&sc->mii);
   1404 	splx(s);
   1405 
   1406 	callout_reset(&sc->rtk_tick_ch, hz, re_tick, sc);
   1407 }
   1408 
   1409 int
   1410 re_intr(void *arg)
   1411 {
   1412 	struct rtk_softc *sc = arg;
   1413 	struct ifnet *ifp;
   1414 	uint16_t status;
   1415 	int handled = 0;
   1416 
   1417 	if (!device_has_power(sc->sc_dev))
   1418 		return 0;
   1419 
   1420 	ifp = &sc->ethercom.ec_if;
   1421 
   1422 	if ((ifp->if_flags & IFF_UP) == 0)
   1423 		return 0;
   1424 
   1425 	for (;;) {
   1426 
   1427 		status = CSR_READ_2(sc, RTK_ISR);
   1428 		/* If the card has gone away the read returns 0xffff. */
   1429 		if (status == 0xffff)
   1430 			break;
   1431 		if (status) {
   1432 			handled = 1;
   1433 			CSR_WRITE_2(sc, RTK_ISR, status);
   1434 		}
   1435 
   1436 		if ((status & RTK_INTRS_CPLUS) == 0)
   1437 			break;
   1438 
   1439 		if (status & (RTK_ISR_RX_OK | RTK_ISR_RX_ERR))
   1440 			re_rxeof(sc);
   1441 
   1442 		if (status & (RTK_ISR_TIMEOUT_EXPIRED | RTK_ISR_TX_ERR |
   1443 		    RTK_ISR_TX_DESC_UNAVAIL))
   1444 			re_txeof(sc);
   1445 
   1446 		if (status & RTK_ISR_SYSTEM_ERR) {
   1447 			re_init(ifp);
   1448 		}
   1449 
   1450 		if (status & RTK_ISR_LINKCHG) {
   1451 			callout_stop(&sc->rtk_tick_ch);
   1452 			re_tick(sc);
   1453 		}
   1454 	}
   1455 
   1456 	if (handled && !IFQ_IS_EMPTY(&ifp->if_snd))
   1457 		re_start(ifp);
   1458 
   1459 	return handled;
   1460 }
   1461 
   1462 
   1463 
   1464 /*
   1465  * Main transmit routine for C+ and gigE NICs.
   1466  */
   1467 
   1468 static void
   1469 re_start(struct ifnet *ifp)
   1470 {
   1471 	struct rtk_softc *sc;
   1472 	struct mbuf *m;
   1473 	bus_dmamap_t map;
   1474 	struct re_txq *txq;
   1475 	struct re_desc *d;
   1476 	struct m_tag *mtag;
   1477 	uint32_t cmdstat, re_flags, vlanctl;
   1478 	int ofree, idx, error, nsegs, seg;
   1479 	int startdesc, curdesc, lastdesc;
   1480 	bool pad;
   1481 
   1482 	sc = ifp->if_softc;
   1483 	ofree = sc->re_ldata.re_txq_free;
   1484 
   1485 	for (idx = sc->re_ldata.re_txq_prodidx;; idx = RE_NEXT_TXQ(sc, idx)) {
   1486 
   1487 		IFQ_POLL(&ifp->if_snd, m);
   1488 		if (m == NULL)
   1489 			break;
   1490 
   1491 		if (sc->re_ldata.re_txq_free == 0 ||
   1492 		    sc->re_ldata.re_tx_free == 0) {
   1493 			/* no more free slots left */
   1494 			ifp->if_flags |= IFF_OACTIVE;
   1495 			break;
   1496 		}
   1497 
   1498 		/*
   1499 		 * Set up checksum offload. Note: checksum offload bits must
   1500 		 * appear in all descriptors of a multi-descriptor transmit
   1501 		 * attempt. (This is according to testing done with an 8169
   1502 		 * chip. I'm not sure if this is a requirement or a bug.)
   1503 		 */
   1504 
   1505 		vlanctl = 0;
   1506 		if ((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) != 0) {
   1507 			uint32_t segsz = m->m_pkthdr.segsz;
   1508 
   1509 			re_flags = RE_TDESC_CMD_LGSEND |
   1510 			    (segsz << RE_TDESC_CMD_MSSVAL_SHIFT);
   1511 		} else {
   1512 			/*
   1513 			 * set RE_TDESC_CMD_IPCSUM if any checksum offloading
   1514 			 * is requested.  otherwise, RE_TDESC_CMD_TCPCSUM/
   1515 			 * RE_TDESC_CMD_UDPCSUM doesn't make effects.
   1516 			 */
   1517 			re_flags = 0;
   1518 			if ((m->m_pkthdr.csum_flags &
   1519 			    (M_CSUM_IPv4 | M_CSUM_TCPv4 | M_CSUM_UDPv4))
   1520 			    != 0) {
   1521 				if ((sc->sc_quirk & RTKQ_DESCV2) == 0) {
   1522 					re_flags |= RE_TDESC_CMD_IPCSUM;
   1523 					if (m->m_pkthdr.csum_flags &
   1524 					    M_CSUM_TCPv4) {
   1525 						re_flags |=
   1526 						    RE_TDESC_CMD_TCPCSUM;
   1527 					} else if (m->m_pkthdr.csum_flags &
   1528 					    M_CSUM_UDPv4) {
   1529 						re_flags |=
   1530 						    RE_TDESC_CMD_UDPCSUM;
   1531 					}
   1532 				} else {
   1533 					vlanctl |= RE_TDESC_VLANCTL_IPCSUM;
   1534 					if (m->m_pkthdr.csum_flags &
   1535 					    M_CSUM_TCPv4) {
   1536 						vlanctl |=
   1537 						    RE_TDESC_VLANCTL_TCPCSUM;
   1538 					} else if (m->m_pkthdr.csum_flags &
   1539 					    M_CSUM_UDPv4) {
   1540 						vlanctl |=
   1541 						    RE_TDESC_VLANCTL_UDPCSUM;
   1542 					}
   1543 				}
   1544 			}
   1545 		}
   1546 
   1547 		txq = &sc->re_ldata.re_txq[idx];
   1548 		map = txq->txq_dmamap;
   1549 		error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m,
   1550 		    BUS_DMA_WRITE|BUS_DMA_NOWAIT);
   1551 
   1552 		if (__predict_false(error)) {
   1553 			/* XXX try to defrag if EFBIG? */
   1554 			printf("%s: can't map mbuf (error %d)\n",
   1555 			    device_xname(sc->sc_dev), error);
   1556 
   1557 			IFQ_DEQUEUE(&ifp->if_snd, m);
   1558 			m_freem(m);
   1559 			ifp->if_oerrors++;
   1560 			continue;
   1561 		}
   1562 
   1563 		nsegs = map->dm_nsegs;
   1564 		pad = false;
   1565 		if (__predict_false(m->m_pkthdr.len <= RE_IP4CSUMTX_PADLEN &&
   1566 		    (re_flags & RE_TDESC_CMD_IPCSUM) != 0 &&
   1567 		    (sc->sc_quirk & RTKQ_DESCV2) == 0)) {
   1568 			pad = true;
   1569 			nsegs++;
   1570 		}
   1571 
   1572 		if (nsegs > sc->re_ldata.re_tx_free) {
   1573 			/*
   1574 			 * Not enough free descriptors to transmit this packet.
   1575 			 */
   1576 			ifp->if_flags |= IFF_OACTIVE;
   1577 			bus_dmamap_unload(sc->sc_dmat, map);
   1578 			break;
   1579 		}
   1580 
   1581 		IFQ_DEQUEUE(&ifp->if_snd, m);
   1582 
   1583 		/*
   1584 		 * Make sure that the caches are synchronized before we
   1585 		 * ask the chip to start DMA for the packet data.
   1586 		 */
   1587 		bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
   1588 		    BUS_DMASYNC_PREWRITE);
   1589 
   1590 		/*
   1591 		 * Set up hardware VLAN tagging. Note: vlan tag info must
   1592 		 * appear in all descriptors of a multi-descriptor
   1593 		 * transmission attempt.
   1594 		 */
   1595 		if ((mtag = VLAN_OUTPUT_TAG(&sc->ethercom, m)) != NULL)
   1596 			vlanctl |= bswap16(VLAN_TAG_VALUE(mtag)) |
   1597 			    RE_TDESC_VLANCTL_TAG;
   1598 
   1599 		/*
   1600 		 * Map the segment array into descriptors.
   1601 		 * Note that we set the start-of-frame and
   1602 		 * end-of-frame markers for either TX or RX,
   1603 		 * but they really only have meaning in the TX case.
   1604 		 * (In the RX case, it's the chip that tells us
   1605 		 *  where packets begin and end.)
   1606 		 * We also keep track of the end of the ring
   1607 		 * and set the end-of-ring bits as needed,
   1608 		 * and we set the ownership bits in all except
   1609 		 * the very first descriptor. (The caller will
   1610 		 * set this descriptor later when it start
   1611 		 * transmission or reception.)
   1612 		 */
   1613 		curdesc = startdesc = sc->re_ldata.re_tx_nextfree;
   1614 		lastdesc = -1;
   1615 		for (seg = 0; seg < map->dm_nsegs;
   1616 		    seg++, curdesc = RE_NEXT_TX_DESC(sc, curdesc)) {
   1617 			d = &sc->re_ldata.re_tx_list[curdesc];
   1618 #ifdef DIAGNOSTIC
   1619 			RE_TXDESCSYNC(sc, curdesc,
   1620 			    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
   1621 			cmdstat = le32toh(d->re_cmdstat);
   1622 			RE_TXDESCSYNC(sc, curdesc, BUS_DMASYNC_PREREAD);
   1623 			if (cmdstat & RE_TDESC_STAT_OWN) {
   1624 				panic("%s: tried to map busy TX descriptor",
   1625 				    device_xname(sc->sc_dev));
   1626 			}
   1627 #endif
   1628 
   1629 			d->re_vlanctl = htole32(vlanctl);
   1630 			re_set_bufaddr(d, map->dm_segs[seg].ds_addr);
   1631 			cmdstat = re_flags | map->dm_segs[seg].ds_len;
   1632 			if (seg == 0)
   1633 				cmdstat |= RE_TDESC_CMD_SOF;
   1634 			else
   1635 				cmdstat |= RE_TDESC_CMD_OWN;
   1636 			if (curdesc == (RE_TX_DESC_CNT(sc) - 1))
   1637 				cmdstat |= RE_TDESC_CMD_EOR;
   1638 			if (seg == nsegs - 1) {
   1639 				cmdstat |= RE_TDESC_CMD_EOF;
   1640 				lastdesc = curdesc;
   1641 			}
   1642 			d->re_cmdstat = htole32(cmdstat);
   1643 			RE_TXDESCSYNC(sc, curdesc,
   1644 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1645 		}
   1646 		if (__predict_false(pad)) {
   1647 			d = &sc->re_ldata.re_tx_list[curdesc];
   1648 			d->re_vlanctl = htole32(vlanctl);
   1649 			re_set_bufaddr(d, RE_TXPADDADDR(sc));
   1650 			cmdstat = re_flags |
   1651 			    RE_TDESC_CMD_OWN | RE_TDESC_CMD_EOF |
   1652 			    (RE_IP4CSUMTX_PADLEN + 1 - m->m_pkthdr.len);
   1653 			if (curdesc == (RE_TX_DESC_CNT(sc) - 1))
   1654 				cmdstat |= RE_TDESC_CMD_EOR;
   1655 			d->re_cmdstat = htole32(cmdstat);
   1656 			RE_TXDESCSYNC(sc, curdesc,
   1657 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1658 			lastdesc = curdesc;
   1659 			curdesc = RE_NEXT_TX_DESC(sc, curdesc);
   1660 		}
   1661 		KASSERT(lastdesc != -1);
   1662 
   1663 		/* Transfer ownership of packet to the chip. */
   1664 
   1665 		sc->re_ldata.re_tx_list[startdesc].re_cmdstat |=
   1666 		    htole32(RE_TDESC_CMD_OWN);
   1667 		RE_TXDESCSYNC(sc, startdesc,
   1668 		    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1669 
   1670 		/* update info of TX queue and descriptors */
   1671 		txq->txq_mbuf = m;
   1672 		txq->txq_descidx = lastdesc;
   1673 		txq->txq_nsegs = nsegs;
   1674 
   1675 		sc->re_ldata.re_txq_free--;
   1676 		sc->re_ldata.re_tx_free -= nsegs;
   1677 		sc->re_ldata.re_tx_nextfree = curdesc;
   1678 
   1679 #if NBPFILTER > 0
   1680 		/*
   1681 		 * If there's a BPF listener, bounce a copy of this frame
   1682 		 * to him.
   1683 		 */
   1684 		if (ifp->if_bpf)
   1685 			bpf_mtap(ifp->if_bpf, m);
   1686 #endif
   1687 	}
   1688 
   1689 	if (sc->re_ldata.re_txq_free < ofree) {
   1690 		/*
   1691 		 * TX packets are enqueued.
   1692 		 */
   1693 		sc->re_ldata.re_txq_prodidx = idx;
   1694 
   1695 		/*
   1696 		 * Start the transmitter to poll.
   1697 		 *
   1698 		 * RealTek put the TX poll request register in a different
   1699 		 * location on the 8169 gigE chip. I don't know why.
   1700 		 */
   1701 		if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0)
   1702 			CSR_WRITE_1(sc, RTK_TXSTART, RTK_TXSTART_START);
   1703 		else
   1704 			CSR_WRITE_1(sc, RTK_GTXSTART, RTK_TXSTART_START);
   1705 
   1706 		/*
   1707 		 * Use the countdown timer for interrupt moderation.
   1708 		 * 'TX done' interrupts are disabled. Instead, we reset the
   1709 		 * countdown timer, which will begin counting until it hits
   1710 		 * the value in the TIMERINT register, and then trigger an
   1711 		 * interrupt. Each time we write to the TIMERCNT register,
   1712 		 * the timer count is reset to 0.
   1713 		 */
   1714 		CSR_WRITE_4(sc, RTK_TIMERCNT, 1);
   1715 
   1716 		/*
   1717 		 * Set a timeout in case the chip goes out to lunch.
   1718 		 */
   1719 		ifp->if_timer = 5;
   1720 	}
   1721 }
   1722 
   1723 static int
   1724 re_init(struct ifnet *ifp)
   1725 {
   1726 	struct rtk_softc *sc = ifp->if_softc;
   1727 	const uint8_t *enaddr;
   1728 	uint32_t rxcfg = 0;
   1729 	uint32_t reg;
   1730 	uint16_t cfg;
   1731 	int error;
   1732 
   1733 	if ((error = re_enable(sc)) != 0)
   1734 		goto out;
   1735 
   1736 	/*
   1737 	 * Cancel pending I/O and free all RX/TX buffers.
   1738 	 */
   1739 	re_stop(ifp, 0);
   1740 
   1741 	re_reset(sc);
   1742 
   1743 	/*
   1744 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
   1745 	 * RX checksum offload. We must configure the C+ register
   1746 	 * before all others.
   1747 	 */
   1748 	cfg = RE_CPLUSCMD_PCI_MRW;
   1749 
   1750 	/*
   1751 	 * XXX: For old 8169 set bit 14.
   1752 	 *      For 8169S/8110S and above, do not set bit 14.
   1753 	 */
   1754 	if ((sc->sc_quirk & RTKQ_8169NONS) != 0)
   1755 		cfg |= (0x1 << 14);
   1756 
   1757 	if ((ifp->if_capenable & ETHERCAP_VLAN_HWTAGGING) != 0)
   1758 		cfg |= RE_CPLUSCMD_VLANSTRIP;
   1759 	if ((ifp->if_capenable & (IFCAP_CSUM_IPv4_Rx |
   1760 	     IFCAP_CSUM_TCPv4_Rx | IFCAP_CSUM_UDPv4_Rx)) != 0)
   1761 		cfg |= RE_CPLUSCMD_RXCSUM_ENB;
   1762 	if ((sc->sc_quirk & RTKQ_MACSTAT) != 0) {
   1763 		cfg |= RE_CPLUSCMD_MACSTAT_DIS;
   1764 		cfg |= RE_CPLUSCMD_TXENB;
   1765 	} else
   1766 		cfg |= RE_CPLUSCMD_RXENB | RE_CPLUSCMD_TXENB;
   1767 
   1768 	CSR_WRITE_2(sc, RTK_CPLUS_CMD, cfg);
   1769 
   1770 	/* XXX: from Realtek-supplied Linux driver. Wholly undocumented. */
   1771 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0)
   1772 		CSR_WRITE_2(sc, RTK_IM, 0x0000);
   1773 
   1774 	DELAY(10000);
   1775 
   1776 	/*
   1777 	 * Init our MAC address.  Even though the chipset
   1778 	 * documentation doesn't mention it, we need to enter "Config
   1779 	 * register write enable" mode to modify the ID registers.
   1780 	 */
   1781 	CSR_WRITE_1(sc, RTK_EECMD, RTK_EEMODE_WRITECFG);
   1782 	enaddr = CLLADDR(ifp->if_sadl);
   1783 	reg = enaddr[0] | (enaddr[1] << 8) |
   1784 	    (enaddr[2] << 16) | (enaddr[3] << 24);
   1785 	CSR_WRITE_4(sc, RTK_IDR0, reg);
   1786 	reg = enaddr[4] | (enaddr[5] << 8);
   1787 	CSR_WRITE_4(sc, RTK_IDR4, reg);
   1788 	CSR_WRITE_1(sc, RTK_EECMD, RTK_EEMODE_OFF);
   1789 
   1790 	/*
   1791 	 * For C+ mode, initialize the RX descriptors and mbufs.
   1792 	 */
   1793 	re_rx_list_init(sc);
   1794 	re_tx_list_init(sc);
   1795 
   1796 	/*
   1797 	 * Load the addresses of the RX and TX lists into the chip.
   1798 	 */
   1799 	CSR_WRITE_4(sc, RTK_RXLIST_ADDR_HI,
   1800 	    RE_ADDR_HI(sc->re_ldata.re_rx_list_map->dm_segs[0].ds_addr));
   1801 	CSR_WRITE_4(sc, RTK_RXLIST_ADDR_LO,
   1802 	    RE_ADDR_LO(sc->re_ldata.re_rx_list_map->dm_segs[0].ds_addr));
   1803 
   1804 	CSR_WRITE_4(sc, RTK_TXLIST_ADDR_HI,
   1805 	    RE_ADDR_HI(sc->re_ldata.re_tx_list_map->dm_segs[0].ds_addr));
   1806 	CSR_WRITE_4(sc, RTK_TXLIST_ADDR_LO,
   1807 	    RE_ADDR_LO(sc->re_ldata.re_tx_list_map->dm_segs[0].ds_addr));
   1808 
   1809 	/*
   1810 	 * Enable transmit and receive.
   1811 	 */
   1812 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_TX_ENB | RTK_CMD_RX_ENB);
   1813 
   1814 	/*
   1815 	 * Set the initial TX and RX configuration.
   1816 	 */
   1817 	if (sc->re_testmode && (sc->sc_quirk & RTKQ_8169NONS) != 0) {
   1818 		/* test mode is needed only for old 8169 */
   1819 		CSR_WRITE_4(sc, RTK_TXCFG,
   1820 		    RE_TXCFG_CONFIG | RTK_LOOPTEST_ON);
   1821 	} else
   1822 		CSR_WRITE_4(sc, RTK_TXCFG, RE_TXCFG_CONFIG);
   1823 
   1824 	CSR_WRITE_1(sc, RTK_EARLY_TX_THRESH, 16);
   1825 
   1826 	CSR_WRITE_4(sc, RTK_RXCFG, RE_RXCFG_CONFIG);
   1827 
   1828 	/* Set the individual bit to receive frames for this host only. */
   1829 	rxcfg = CSR_READ_4(sc, RTK_RXCFG);
   1830 	rxcfg |= RTK_RXCFG_RX_INDIV;
   1831 
   1832 	/* If we want promiscuous mode, set the allframes bit. */
   1833 	if (ifp->if_flags & IFF_PROMISC)
   1834 		rxcfg |= RTK_RXCFG_RX_ALLPHYS;
   1835 	else
   1836 		rxcfg &= ~RTK_RXCFG_RX_ALLPHYS;
   1837 	CSR_WRITE_4(sc, RTK_RXCFG, rxcfg);
   1838 
   1839 	/*
   1840 	 * Set capture broadcast bit to capture broadcast frames.
   1841 	 */
   1842 	if (ifp->if_flags & IFF_BROADCAST)
   1843 		rxcfg |= RTK_RXCFG_RX_BROAD;
   1844 	else
   1845 		rxcfg &= ~RTK_RXCFG_RX_BROAD;
   1846 	CSR_WRITE_4(sc, RTK_RXCFG, rxcfg);
   1847 
   1848 	/*
   1849 	 * Program the multicast filter, if necessary.
   1850 	 */
   1851 	rtk_setmulti(sc);
   1852 
   1853 	/*
   1854 	 * Enable interrupts.
   1855 	 */
   1856 	if (sc->re_testmode)
   1857 		CSR_WRITE_2(sc, RTK_IMR, 0);
   1858 	else
   1859 		CSR_WRITE_2(sc, RTK_IMR, RTK_INTRS_CPLUS);
   1860 
   1861 	/* Start RX/TX process. */
   1862 	CSR_WRITE_4(sc, RTK_MISSEDPKT, 0);
   1863 #ifdef notdef
   1864 	/* Enable receiver and transmitter. */
   1865 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_TX_ENB | RTK_CMD_RX_ENB);
   1866 #endif
   1867 
   1868 	/*
   1869 	 * Initialize the timer interrupt register so that
   1870 	 * a timer interrupt will be generated once the timer
   1871 	 * reaches a certain number of ticks. The timer is
   1872 	 * reloaded on each transmit. This gives us TX interrupt
   1873 	 * moderation, which dramatically improves TX frame rate.
   1874 	 */
   1875 
   1876 	if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0)
   1877 		CSR_WRITE_4(sc, RTK_TIMERINT, 0x400);
   1878 	else {
   1879 		CSR_WRITE_4(sc, RTK_TIMERINT_8169, 0x800);
   1880 
   1881 		/*
   1882 		 * For 8169 gigE NICs, set the max allowed RX packet
   1883 		 * size so we can receive jumbo frames.
   1884 		 */
   1885 		CSR_WRITE_2(sc, RTK_MAXRXPKTLEN, 16383);
   1886 	}
   1887 
   1888 	if (sc->re_testmode)
   1889 		return 0;
   1890 
   1891 	CSR_WRITE_1(sc, RTK_CFG1, RTK_CFG1_DRVLOAD);
   1892 
   1893 	ifp->if_flags |= IFF_RUNNING;
   1894 	ifp->if_flags &= ~IFF_OACTIVE;
   1895 
   1896 	callout_reset(&sc->rtk_tick_ch, hz, re_tick, sc);
   1897 
   1898  out:
   1899 	if (error) {
   1900 		ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
   1901 		ifp->if_timer = 0;
   1902 		printf("%s: interface not running\n",
   1903 		    device_xname(sc->sc_dev));
   1904 	}
   1905 
   1906 	return error;
   1907 }
   1908 
   1909 static int
   1910 re_ioctl(struct ifnet *ifp, u_long command, void *data)
   1911 {
   1912 	struct rtk_softc *sc = ifp->if_softc;
   1913 	struct ifreq *ifr = data;
   1914 	int s, error = 0;
   1915 
   1916 	s = splnet();
   1917 
   1918 	switch (command) {
   1919 	case SIOCSIFMTU:
   1920 		/*
   1921 		 * Disable jumbo frames if it's not supported.
   1922 		 */
   1923 		if ((sc->sc_quirk & RTKQ_NOJUMBO) != 0 &&
   1924 		    ifr->ifr_mtu > ETHERMTU) {
   1925 			error = EINVAL;
   1926 			break;
   1927 		}
   1928 
   1929 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > ETHERMTU_JUMBO)
   1930 			error = EINVAL;
   1931 		else if ((error = ifioctl_common(ifp, command, data)) ==
   1932 		    ENETRESET)
   1933 			error = 0;
   1934 		break;
   1935 	default:
   1936 		if ((error = ether_ioctl(ifp, command, data)) != ENETRESET)
   1937 			break;
   1938 
   1939 		error = 0;
   1940 
   1941 		if (command == SIOCSIFCAP)
   1942 			error = (*ifp->if_init)(ifp);
   1943 		else if (command != SIOCADDMULTI && command != SIOCDELMULTI)
   1944 			;
   1945 		else if (ifp->if_flags & IFF_RUNNING)
   1946 			rtk_setmulti(sc);
   1947 		break;
   1948 	}
   1949 
   1950 	splx(s);
   1951 
   1952 	return error;
   1953 }
   1954 
   1955 static void
   1956 re_watchdog(struct ifnet *ifp)
   1957 {
   1958 	struct rtk_softc *sc;
   1959 	int s;
   1960 
   1961 	sc = ifp->if_softc;
   1962 	s = splnet();
   1963 	printf("%s: watchdog timeout\n", device_xname(sc->sc_dev));
   1964 	ifp->if_oerrors++;
   1965 
   1966 	re_txeof(sc);
   1967 	re_rxeof(sc);
   1968 
   1969 	re_init(ifp);
   1970 
   1971 	splx(s);
   1972 }
   1973 
   1974 /*
   1975  * Stop the adapter and free any mbufs allocated to the
   1976  * RX and TX lists.
   1977  */
   1978 static void
   1979 re_stop(struct ifnet *ifp, int disable)
   1980 {
   1981 	int i;
   1982 	struct rtk_softc *sc = ifp->if_softc;
   1983 
   1984 	callout_stop(&sc->rtk_tick_ch);
   1985 
   1986 	mii_down(&sc->mii);
   1987 
   1988 	if ((sc->sc_quirk & RTKQ_CMDSTOP) != 0)
   1989 		CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_STOPREQ | RTK_CMD_TX_ENB |
   1990 		    RTK_CMD_RX_ENB);
   1991 	else
   1992 		CSR_WRITE_1(sc, RTK_COMMAND, 0x00);
   1993 	DELAY(1000);
   1994 	CSR_WRITE_2(sc, RTK_IMR, 0x0000);
   1995 	CSR_WRITE_2(sc, RTK_ISR, 0xFFFF);
   1996 
   1997 	if (sc->re_head != NULL) {
   1998 		m_freem(sc->re_head);
   1999 		sc->re_head = sc->re_tail = NULL;
   2000 	}
   2001 
   2002 	/* Free the TX list buffers. */
   2003 	for (i = 0; i < RE_TX_QLEN; i++) {
   2004 		if (sc->re_ldata.re_txq[i].txq_mbuf != NULL) {
   2005 			bus_dmamap_unload(sc->sc_dmat,
   2006 			    sc->re_ldata.re_txq[i].txq_dmamap);
   2007 			m_freem(sc->re_ldata.re_txq[i].txq_mbuf);
   2008 			sc->re_ldata.re_txq[i].txq_mbuf = NULL;
   2009 		}
   2010 	}
   2011 
   2012 	/* Free the RX list buffers. */
   2013 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
   2014 		if (sc->re_ldata.re_rxsoft[i].rxs_mbuf != NULL) {
   2015 			bus_dmamap_unload(sc->sc_dmat,
   2016 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
   2017 			m_freem(sc->re_ldata.re_rxsoft[i].rxs_mbuf);
   2018 			sc->re_ldata.re_rxsoft[i].rxs_mbuf = NULL;
   2019 		}
   2020 	}
   2021 
   2022 	if (disable)
   2023 		re_disable(sc);
   2024 
   2025 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
   2026 	ifp->if_timer = 0;
   2027 }
   2028