Home | History | Annotate | Line # | Download | only in ic
rtl8169.c revision 1.136
      1 /*	$NetBSD: rtl8169.c,v 1.136 2012/07/22 14:32:57 matt 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 1.136 2012/07/22 14:32:57 matt 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 
    115 #include <sys/param.h>
    116 #include <sys/endian.h>
    117 #include <sys/systm.h>
    118 #include <sys/sockio.h>
    119 #include <sys/mbuf.h>
    120 #include <sys/malloc.h>
    121 #include <sys/kernel.h>
    122 #include <sys/socket.h>
    123 #include <sys/device.h>
    124 
    125 #include <net/if.h>
    126 #include <net/if_arp.h>
    127 #include <net/if_dl.h>
    128 #include <net/if_ether.h>
    129 #include <net/if_media.h>
    130 #include <net/if_vlanvar.h>
    131 
    132 #include <netinet/in_systm.h>	/* XXX for IP_MAXPACKET */
    133 #include <netinet/in.h>		/* XXX for IP_MAXPACKET */
    134 #include <netinet/ip.h>		/* XXX for IP_MAXPACKET */
    135 
    136 #include <net/bpf.h>
    137 
    138 #include <sys/bus.h>
    139 
    140 #include <dev/mii/mii.h>
    141 #include <dev/mii/miivar.h>
    142 
    143 #include <dev/ic/rtl81x9reg.h>
    144 #include <dev/ic/rtl81x9var.h>
    145 
    146 #include <dev/ic/rtl8169var.h>
    147 
    148 static inline void re_set_bufaddr(struct re_desc *, bus_addr_t);
    149 
    150 static int re_newbuf(struct rtk_softc *, int, struct mbuf *);
    151 static int re_rx_list_init(struct rtk_softc *);
    152 static int re_tx_list_init(struct rtk_softc *);
    153 static void re_rxeof(struct rtk_softc *);
    154 static void re_txeof(struct rtk_softc *);
    155 static void re_tick(void *);
    156 static void re_start(struct ifnet *);
    157 static int re_ioctl(struct ifnet *, u_long, void *);
    158 static int re_init(struct ifnet *);
    159 static void re_stop(struct ifnet *, int);
    160 static void re_watchdog(struct ifnet *);
    161 
    162 static int re_enable(struct rtk_softc *);
    163 static void re_disable(struct rtk_softc *);
    164 
    165 static int re_gmii_readreg(device_t, int, int);
    166 static void re_gmii_writereg(device_t, int, int, int);
    167 
    168 static int re_miibus_readreg(device_t, int, int);
    169 static void re_miibus_writereg(device_t, int, int, int);
    170 static void re_miibus_statchg(struct ifnet *);
    171 
    172 static void re_reset(struct rtk_softc *);
    173 
    174 static inline void
    175 re_set_bufaddr(struct re_desc *d, bus_addr_t addr)
    176 {
    177 
    178 	d->re_bufaddr_lo = htole32((uint32_t)addr);
    179 	if (sizeof(bus_addr_t) == sizeof(uint64_t))
    180 		d->re_bufaddr_hi = htole32((uint64_t)addr >> 32);
    181 	else
    182 		d->re_bufaddr_hi = 0;
    183 }
    184 
    185 static int
    186 re_gmii_readreg(device_t dev, int phy, int reg)
    187 {
    188 	struct rtk_softc *sc = device_private(dev);
    189 	uint32_t rval;
    190 	int i;
    191 
    192 	if (phy != 7)
    193 		return 0;
    194 
    195 	/* Let the rgephy driver read the GMEDIASTAT register */
    196 
    197 	if (reg == RTK_GMEDIASTAT) {
    198 		rval = CSR_READ_1(sc, RTK_GMEDIASTAT);
    199 		return rval;
    200 	}
    201 
    202 	CSR_WRITE_4(sc, RTK_PHYAR, reg << 16);
    203 	DELAY(1000);
    204 
    205 	for (i = 0; i < RTK_TIMEOUT; i++) {
    206 		rval = CSR_READ_4(sc, RTK_PHYAR);
    207 		if (rval & RTK_PHYAR_BUSY)
    208 			break;
    209 		DELAY(100);
    210 	}
    211 
    212 	if (i == RTK_TIMEOUT) {
    213 		printf("%s: PHY read failed\n", device_xname(sc->sc_dev));
    214 		return 0;
    215 	}
    216 
    217 	return rval & RTK_PHYAR_PHYDATA;
    218 }
    219 
    220 static void
    221 re_gmii_writereg(device_t dev, int phy, int reg, int data)
    222 {
    223 	struct rtk_softc *sc = device_private(dev);
    224 	uint32_t rval;
    225 	int i;
    226 
    227 	CSR_WRITE_4(sc, RTK_PHYAR, (reg << 16) |
    228 	    (data & RTK_PHYAR_PHYDATA) | RTK_PHYAR_BUSY);
    229 	DELAY(1000);
    230 
    231 	for (i = 0; i < RTK_TIMEOUT; i++) {
    232 		rval = CSR_READ_4(sc, RTK_PHYAR);
    233 		if (!(rval & RTK_PHYAR_BUSY))
    234 			break;
    235 		DELAY(100);
    236 	}
    237 
    238 	if (i == RTK_TIMEOUT) {
    239 		printf("%s: PHY write reg %x <- %x failed\n",
    240 		    device_xname(sc->sc_dev), reg, data);
    241 	}
    242 }
    243 
    244 static int
    245 re_miibus_readreg(device_t dev, int phy, int reg)
    246 {
    247 	struct rtk_softc *sc = device_private(dev);
    248 	uint16_t rval = 0;
    249 	uint16_t re8139_reg = 0;
    250 	int s;
    251 
    252 	s = splnet();
    253 
    254 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0) {
    255 		rval = re_gmii_readreg(dev, phy, reg);
    256 		splx(s);
    257 		return rval;
    258 	}
    259 
    260 	/* Pretend the internal PHY is only at address 0 */
    261 	if (phy) {
    262 		splx(s);
    263 		return 0;
    264 	}
    265 	switch (reg) {
    266 	case MII_BMCR:
    267 		re8139_reg = RTK_BMCR;
    268 		break;
    269 	case MII_BMSR:
    270 		re8139_reg = RTK_BMSR;
    271 		break;
    272 	case MII_ANAR:
    273 		re8139_reg = RTK_ANAR;
    274 		break;
    275 	case MII_ANER:
    276 		re8139_reg = RTK_ANER;
    277 		break;
    278 	case MII_ANLPAR:
    279 		re8139_reg = RTK_LPAR;
    280 		break;
    281 	case MII_PHYIDR1:
    282 	case MII_PHYIDR2:
    283 		splx(s);
    284 		return 0;
    285 	/*
    286 	 * Allow the rlphy driver to read the media status
    287 	 * register. If we have a link partner which does not
    288 	 * support NWAY, this is the register which will tell
    289 	 * us the results of parallel detection.
    290 	 */
    291 	case RTK_MEDIASTAT:
    292 		rval = CSR_READ_1(sc, RTK_MEDIASTAT);
    293 		splx(s);
    294 		return rval;
    295 	default:
    296 		printf("%s: bad phy register\n", device_xname(sc->sc_dev));
    297 		splx(s);
    298 		return 0;
    299 	}
    300 	rval = CSR_READ_2(sc, re8139_reg);
    301 	if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0 && re8139_reg == RTK_BMCR) {
    302 		/* 8139C+ has different bit layout. */
    303 		rval &= ~(BMCR_LOOP | BMCR_ISO);
    304 	}
    305 	splx(s);
    306 	return rval;
    307 }
    308 
    309 static void
    310 re_miibus_writereg(device_t dev, int phy, int reg, int data)
    311 {
    312 	struct rtk_softc *sc = device_private(dev);
    313 	uint16_t re8139_reg = 0;
    314 	int s;
    315 
    316 	s = splnet();
    317 
    318 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0) {
    319 		re_gmii_writereg(dev, phy, reg, data);
    320 		splx(s);
    321 		return;
    322 	}
    323 
    324 	/* Pretend the internal PHY is only at address 0 */
    325 	if (phy) {
    326 		splx(s);
    327 		return;
    328 	}
    329 	switch (reg) {
    330 	case MII_BMCR:
    331 		re8139_reg = RTK_BMCR;
    332 		if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0) {
    333 			/* 8139C+ has different bit layout. */
    334 			data &= ~(BMCR_LOOP | BMCR_ISO);
    335 		}
    336 		break;
    337 	case MII_BMSR:
    338 		re8139_reg = RTK_BMSR;
    339 		break;
    340 	case MII_ANAR:
    341 		re8139_reg = RTK_ANAR;
    342 		break;
    343 	case MII_ANER:
    344 		re8139_reg = RTK_ANER;
    345 		break;
    346 	case MII_ANLPAR:
    347 		re8139_reg = RTK_LPAR;
    348 		break;
    349 	case MII_PHYIDR1:
    350 	case MII_PHYIDR2:
    351 		splx(s);
    352 		return;
    353 		break;
    354 	default:
    355 		printf("%s: bad phy register\n", device_xname(sc->sc_dev));
    356 		splx(s);
    357 		return;
    358 	}
    359 	CSR_WRITE_2(sc, re8139_reg, data);
    360 	splx(s);
    361 	return;
    362 }
    363 
    364 static void
    365 re_miibus_statchg(struct ifnet *ifp)
    366 {
    367 
    368 	return;
    369 }
    370 
    371 static void
    372 re_reset(struct rtk_softc *sc)
    373 {
    374 	int i;
    375 
    376 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_RESET);
    377 
    378 	for (i = 0; i < RTK_TIMEOUT; i++) {
    379 		DELAY(10);
    380 		if ((CSR_READ_1(sc, RTK_COMMAND) & RTK_CMD_RESET) == 0)
    381 			break;
    382 	}
    383 	if (i == RTK_TIMEOUT)
    384 		printf("%s: reset never completed!\n",
    385 		    device_xname(sc->sc_dev));
    386 
    387 	/*
    388 	 * NB: Realtek-supplied FreeBSD driver does this only for MACFG_3,
    389 	 *     but also says "Rtl8169s sigle chip detected".
    390 	 */
    391 	if ((sc->sc_quirk & RTKQ_MACLDPS) != 0)
    392 		CSR_WRITE_1(sc, RTK_LDPS, 1);
    393 
    394 }
    395 
    396 /*
    397  * The following routine is designed to test for a defect on some
    398  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
    399  * lines connected to the bus, however for a 32-bit only card, they
    400  * should be pulled high. The result of this defect is that the
    401  * NIC will not work right if you plug it into a 64-bit slot: DMA
    402  * operations will be done with 64-bit transfers, which will fail
    403  * because the 64-bit data lines aren't connected.
    404  *
    405  * There's no way to work around this (short of talking a soldering
    406  * iron to the board), however we can detect it. The method we use
    407  * here is to put the NIC into digital loopback mode, set the receiver
    408  * to promiscuous mode, and then try to send a frame. We then compare
    409  * the frame data we sent to what was received. If the data matches,
    410  * then the NIC is working correctly, otherwise we know the user has
    411  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
    412  * slot. In the latter case, there's no way the NIC can work correctly,
    413  * so we print out a message on the console and abort the device attach.
    414  */
    415 
    416 int
    417 re_diag(struct rtk_softc *sc)
    418 {
    419 	struct ifnet *ifp = &sc->ethercom.ec_if;
    420 	struct mbuf *m0;
    421 	struct ether_header *eh;
    422 	struct re_rxsoft *rxs;
    423 	struct re_desc *cur_rx;
    424 	bus_dmamap_t dmamap;
    425 	uint16_t status;
    426 	uint32_t rxstat;
    427 	int total_len, i, s, error = 0;
    428 	static const uint8_t dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
    429 	static const uint8_t src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
    430 
    431 	/* Allocate a single mbuf */
    432 
    433 	MGETHDR(m0, M_DONTWAIT, MT_DATA);
    434 	if (m0 == NULL)
    435 		return ENOBUFS;
    436 
    437 	/*
    438 	 * Initialize the NIC in test mode. This sets the chip up
    439 	 * so that it can send and receive frames, but performs the
    440 	 * following special functions:
    441 	 * - Puts receiver in promiscuous mode
    442 	 * - Enables digital loopback mode
    443 	 * - Leaves interrupts turned off
    444 	 */
    445 
    446 	ifp->if_flags |= IFF_PROMISC;
    447 	sc->re_testmode = 1;
    448 	re_init(ifp);
    449 	re_stop(ifp, 0);
    450 	DELAY(100000);
    451 	re_init(ifp);
    452 
    453 	/* Put some data in the mbuf */
    454 
    455 	eh = mtod(m0, struct ether_header *);
    456 	memcpy(eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN);
    457 	memcpy(eh->ether_shost, (char *)&src, ETHER_ADDR_LEN);
    458 	eh->ether_type = htons(ETHERTYPE_IP);
    459 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
    460 
    461 	/*
    462 	 * Queue the packet, start transmission.
    463 	 */
    464 
    465 	CSR_WRITE_2(sc, RTK_ISR, 0xFFFF);
    466 	s = splnet();
    467 	IF_ENQUEUE(&ifp->if_snd, m0);
    468 	re_start(ifp);
    469 	splx(s);
    470 	m0 = NULL;
    471 
    472 	/* Wait for it to propagate through the chip */
    473 
    474 	DELAY(100000);
    475 	for (i = 0; i < RTK_TIMEOUT; i++) {
    476 		status = CSR_READ_2(sc, RTK_ISR);
    477 		if ((status & (RTK_ISR_TIMEOUT_EXPIRED | RTK_ISR_RX_OK)) ==
    478 		    (RTK_ISR_TIMEOUT_EXPIRED | RTK_ISR_RX_OK))
    479 			break;
    480 		DELAY(10);
    481 	}
    482 	if (i == RTK_TIMEOUT) {
    483 		aprint_error_dev(sc->sc_dev,
    484 		    "diagnostic failed, failed to receive packet "
    485 		    "in loopback mode\n");
    486 		error = EIO;
    487 		goto done;
    488 	}
    489 
    490 	/*
    491 	 * The packet should have been dumped into the first
    492 	 * entry in the RX DMA ring. Grab it from there.
    493 	 */
    494 
    495 	rxs = &sc->re_ldata.re_rxsoft[0];
    496 	dmamap = rxs->rxs_dmamap;
    497 	bus_dmamap_sync(sc->sc_dmat, dmamap, 0, dmamap->dm_mapsize,
    498 	    BUS_DMASYNC_POSTREAD);
    499 	bus_dmamap_unload(sc->sc_dmat, dmamap);
    500 
    501 	m0 = rxs->rxs_mbuf;
    502 	rxs->rxs_mbuf = NULL;
    503 	eh = mtod(m0, struct ether_header *);
    504 
    505 	RE_RXDESCSYNC(sc, 0, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
    506 	cur_rx = &sc->re_ldata.re_rx_list[0];
    507 	rxstat = le32toh(cur_rx->re_cmdstat);
    508 	total_len = rxstat & sc->re_rxlenmask;
    509 
    510 	if (total_len != ETHER_MIN_LEN) {
    511 		aprint_error_dev(sc->sc_dev,
    512 		    "diagnostic failed, received short packet\n");
    513 		error = EIO;
    514 		goto done;
    515 	}
    516 
    517 	/* Test that the received packet data matches what we sent. */
    518 
    519 	if (memcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
    520 	    memcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
    521 	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
    522 		aprint_error_dev(sc->sc_dev, "WARNING, DMA FAILURE!\n"
    523 		    "expected TX data: %s/%s/0x%x\n"
    524 		    "received RX data: %s/%s/0x%x\n"
    525 		    "You may have a defective 32-bit NIC plugged "
    526 		    "into a 64-bit PCI slot.\n"
    527 		    "Please re-install the NIC in a 32-bit slot "
    528 		    "for proper operation.\n"
    529 		    "Read the re(4) man page for more details.\n" ,
    530 		    ether_sprintf(dst),  ether_sprintf(src), ETHERTYPE_IP,
    531 		    ether_sprintf(eh->ether_dhost),
    532 		    ether_sprintf(eh->ether_shost), ntohs(eh->ether_type));
    533 		error = EIO;
    534 	}
    535 
    536  done:
    537 	/* Turn interface off, release resources */
    538 
    539 	sc->re_testmode = 0;
    540 	ifp->if_flags &= ~IFF_PROMISC;
    541 	re_stop(ifp, 0);
    542 	if (m0 != NULL)
    543 		m_freem(m0);
    544 
    545 	return error;
    546 }
    547 
    548 
    549 /*
    550  * Attach the interface. Allocate softc structures, do ifmedia
    551  * setup and ethernet/BPF attach.
    552  */
    553 void
    554 re_attach(struct rtk_softc *sc)
    555 {
    556 	uint8_t eaddr[ETHER_ADDR_LEN];
    557 	uint16_t val;
    558 	struct ifnet *ifp;
    559 	int error = 0, i, addr_len;
    560 
    561 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0) {
    562 		uint32_t hwrev;
    563 
    564 		/* Revision of 8169/8169S/8110s in bits 30..26, 23 */
    565 		hwrev = CSR_READ_4(sc, RTK_TXCFG) & RTK_TXCFG_HWREV;
    566 		switch (hwrev) {
    567 		case RTK_HWREV_8169:
    568 			sc->sc_quirk |= RTKQ_8169NONS;
    569 			break;
    570 		case RTK_HWREV_8169S:
    571 		case RTK_HWREV_8110S:
    572 		case RTK_HWREV_8169_8110SB:
    573 		case RTK_HWREV_8169_8110SBL:
    574 		case RTK_HWREV_8169_8110SC:
    575 			sc->sc_quirk |= RTKQ_MACLDPS;
    576 			break;
    577 		case RTK_HWREV_8168_SPIN1:
    578 		case RTK_HWREV_8168_SPIN2:
    579 		case RTK_HWREV_8168_SPIN3:
    580 			sc->sc_quirk |= RTKQ_MACSTAT;
    581 			break;
    582 		case RTK_HWREV_8168C:
    583 		case RTK_HWREV_8168C_SPIN2:
    584 		case RTK_HWREV_8168CP:
    585 		case RTK_HWREV_8168D:
    586 		case RTK_HWREV_8168DP:
    587 			sc->sc_quirk |= RTKQ_DESCV2 | RTKQ_NOEECMD |
    588 			    RTKQ_MACSTAT | RTKQ_CMDSTOP;
    589 			/*
    590 			 * From FreeBSD driver:
    591 			 *
    592 			 * These (8168/8111) controllers support jumbo frame
    593 			 * but it seems that enabling it requires touching
    594 			 * additional magic registers. Depending on MAC
    595 			 * revisions some controllers need to disable
    596 			 * checksum offload. So disable jumbo frame until
    597 			 * I have better idea what it really requires to
    598 			 * make it support.
    599 			 * RTL8168C/CP : supports up to 6KB jumbo frame.
    600 			 * RTL8111C/CP : supports up to 9KB jumbo frame.
    601 			 */
    602 			sc->sc_quirk |= RTKQ_NOJUMBO;
    603 			break;
    604 		case RTK_HWREV_8168E:
    605 			sc->sc_quirk |= RTKQ_DESCV2 | RTKQ_NOEECMD |
    606 			    RTKQ_MACSTAT | RTKQ_CMDSTOP | RTKQ_PHYWAKE_PM |
    607 			    RTKQ_NOJUMBO;
    608 			break;
    609 		case RTK_HWREV_8168E_VL:
    610 			sc->sc_quirk |= RTKQ_DESCV2 | RTKQ_NOEECMD |
    611 			    RTKQ_MACSTAT | RTKQ_CMDSTOP | RTKQ_NOJUMBO;
    612 			break;
    613 		case RTK_HWREV_8100E:
    614 		case RTK_HWREV_8100E_SPIN2:
    615 		case RTK_HWREV_8101E:
    616 			sc->sc_quirk |= RTKQ_NOJUMBO;
    617 			break;
    618 		case RTK_HWREV_8102E:
    619 		case RTK_HWREV_8102EL:
    620 		case RTK_HWREV_8103E:
    621 			sc->sc_quirk |= RTKQ_DESCV2 | RTKQ_NOEECMD |
    622 			    RTKQ_MACSTAT | RTKQ_CMDSTOP | RTKQ_NOJUMBO;
    623 			break;
    624 		default:
    625 			aprint_normal_dev(sc->sc_dev,
    626 			    "Unknown revision (0x%08x)\n", hwrev);
    627 			/* assume the latest features */
    628 			sc->sc_quirk |= RTKQ_DESCV2 | RTKQ_NOEECMD;
    629 			sc->sc_quirk |= RTKQ_NOJUMBO;
    630 		}
    631 
    632 		/* Set RX length mask */
    633 		sc->re_rxlenmask = RE_RDESC_STAT_GFRAGLEN;
    634 		sc->re_ldata.re_tx_desc_cnt = RE_TX_DESC_CNT_8169;
    635 	} else {
    636 		sc->sc_quirk |= RTKQ_NOJUMBO;
    637 
    638 		/* Set RX length mask */
    639 		sc->re_rxlenmask = RE_RDESC_STAT_FRAGLEN;
    640 		sc->re_ldata.re_tx_desc_cnt = RE_TX_DESC_CNT_8139;
    641 	}
    642 
    643 	/* Reset the adapter. */
    644 	re_reset(sc);
    645 
    646 	if ((sc->sc_quirk & RTKQ_NOEECMD) != 0) {
    647 		/*
    648 		 * Get station address from ID registers.
    649 		 */
    650 		for (i = 0; i < ETHER_ADDR_LEN; i++)
    651 			eaddr[i] = CSR_READ_1(sc, RTK_IDR0 + i);
    652 	} else {
    653 		/*
    654 		 * Get station address from the EEPROM.
    655 		 */
    656 		if (rtk_read_eeprom(sc, RTK_EE_ID, RTK_EEADDR_LEN1) == 0x8129)
    657 			addr_len = RTK_EEADDR_LEN1;
    658 		else
    659 			addr_len = RTK_EEADDR_LEN0;
    660 
    661 		/*
    662 		 * Get station address from the EEPROM.
    663 		 */
    664 		for (i = 0; i < ETHER_ADDR_LEN / 2; i++) {
    665 			val = rtk_read_eeprom(sc, RTK_EE_EADDR0 + i, addr_len);
    666 			eaddr[(i * 2) + 0] = val & 0xff;
    667 			eaddr[(i * 2) + 1] = val >> 8;
    668 		}
    669 	}
    670 
    671 	/* Take PHY out of power down mode. */
    672 	if ((sc->sc_quirk & RTKQ_PHYWAKE_PM) != 0)
    673 		CSR_WRITE_1(sc, RTK_PMCH, CSR_READ_1(sc, RTK_PMCH) | 0x80);
    674 
    675 	aprint_normal_dev(sc->sc_dev, "Ethernet address %s\n",
    676 	    ether_sprintf(eaddr));
    677 
    678 	if (sc->re_ldata.re_tx_desc_cnt >
    679 	    PAGE_SIZE / sizeof(struct re_desc)) {
    680 		sc->re_ldata.re_tx_desc_cnt =
    681 		    PAGE_SIZE / sizeof(struct re_desc);
    682 	}
    683 
    684 	aprint_verbose_dev(sc->sc_dev, "using %d tx descriptors\n",
    685 	    sc->re_ldata.re_tx_desc_cnt);
    686 	KASSERT(RE_NEXT_TX_DESC(sc, RE_TX_DESC_CNT(sc) - 1) == 0);
    687 
    688 	/* Allocate DMA'able memory for the TX ring */
    689 	if ((error = bus_dmamem_alloc(sc->sc_dmat, RE_TX_LIST_SZ(sc),
    690 	    RE_RING_ALIGN, 0, &sc->re_ldata.re_tx_listseg, 1,
    691 	    &sc->re_ldata.re_tx_listnseg, BUS_DMA_NOWAIT)) != 0) {
    692 		aprint_error_dev(sc->sc_dev,
    693 		    "can't allocate tx listseg, error = %d\n", error);
    694 		goto fail_0;
    695 	}
    696 
    697 	/* Load the map for the TX ring. */
    698 	if ((error = bus_dmamem_map(sc->sc_dmat, &sc->re_ldata.re_tx_listseg,
    699 	    sc->re_ldata.re_tx_listnseg, RE_TX_LIST_SZ(sc),
    700 	    (void **)&sc->re_ldata.re_tx_list,
    701 	    BUS_DMA_COHERENT | BUS_DMA_NOWAIT)) != 0) {
    702 		aprint_error_dev(sc->sc_dev,
    703 		    "can't map tx list, error = %d\n", error);
    704 		goto fail_1;
    705 	}
    706 	memset(sc->re_ldata.re_tx_list, 0, RE_TX_LIST_SZ(sc));
    707 
    708 	if ((error = bus_dmamap_create(sc->sc_dmat, RE_TX_LIST_SZ(sc), 1,
    709 	    RE_TX_LIST_SZ(sc), 0, 0,
    710 	    &sc->re_ldata.re_tx_list_map)) != 0) {
    711 		aprint_error_dev(sc->sc_dev,
    712 		    "can't create tx list map, error = %d\n", error);
    713 		goto fail_2;
    714 	}
    715 
    716 
    717 	if ((error = bus_dmamap_load(sc->sc_dmat,
    718 	    sc->re_ldata.re_tx_list_map, sc->re_ldata.re_tx_list,
    719 	    RE_TX_LIST_SZ(sc), NULL, BUS_DMA_NOWAIT)) != 0) {
    720 		aprint_error_dev(sc->sc_dev,
    721 		    "can't load tx list, error = %d\n", error);
    722 		goto fail_3;
    723 	}
    724 
    725 	/* Create DMA maps for TX buffers */
    726 	for (i = 0; i < RE_TX_QLEN; i++) {
    727 		error = bus_dmamap_create(sc->sc_dmat,
    728 		    round_page(IP_MAXPACKET),
    729 		    RE_TX_DESC_CNT(sc), RE_TDESC_CMD_FRAGLEN,
    730 		    0, 0, &sc->re_ldata.re_txq[i].txq_dmamap);
    731 		if (error) {
    732 			aprint_error_dev(sc->sc_dev,
    733 			    "can't create DMA map for TX\n");
    734 			goto fail_4;
    735 		}
    736 	}
    737 
    738 	/* Allocate DMA'able memory for the RX ring */
    739 	/* XXX see also a comment about RE_RX_DMAMEM_SZ in rtl81x9var.h */
    740 	if ((error = bus_dmamem_alloc(sc->sc_dmat,
    741 	    RE_RX_DMAMEM_SZ, RE_RING_ALIGN, 0, &sc->re_ldata.re_rx_listseg, 1,
    742 	    &sc->re_ldata.re_rx_listnseg, BUS_DMA_NOWAIT)) != 0) {
    743 		aprint_error_dev(sc->sc_dev,
    744 		    "can't allocate rx listseg, error = %d\n", error);
    745 		goto fail_4;
    746 	}
    747 
    748 	/* Load the map for the RX ring. */
    749 	if ((error = bus_dmamem_map(sc->sc_dmat, &sc->re_ldata.re_rx_listseg,
    750 	    sc->re_ldata.re_rx_listnseg, RE_RX_DMAMEM_SZ,
    751 	    (void **)&sc->re_ldata.re_rx_list,
    752 	    BUS_DMA_COHERENT | BUS_DMA_NOWAIT)) != 0) {
    753 		aprint_error_dev(sc->sc_dev,
    754 		    "can't map rx list, error = %d\n", error);
    755 		goto fail_5;
    756 	}
    757 	memset(sc->re_ldata.re_rx_list, 0, RE_RX_DMAMEM_SZ);
    758 
    759 	if ((error = bus_dmamap_create(sc->sc_dmat,
    760 	    RE_RX_DMAMEM_SZ, 1, RE_RX_DMAMEM_SZ, 0, 0,
    761 	    &sc->re_ldata.re_rx_list_map)) != 0) {
    762 		aprint_error_dev(sc->sc_dev,
    763 		    "can't create rx list map, error = %d\n", error);
    764 		goto fail_6;
    765 	}
    766 
    767 	if ((error = bus_dmamap_load(sc->sc_dmat,
    768 	    sc->re_ldata.re_rx_list_map, sc->re_ldata.re_rx_list,
    769 	    RE_RX_DMAMEM_SZ, NULL, BUS_DMA_NOWAIT)) != 0) {
    770 		aprint_error_dev(sc->sc_dev,
    771 		    "can't load rx list, error = %d\n", error);
    772 		goto fail_7;
    773 	}
    774 
    775 	/* Create DMA maps for RX buffers */
    776 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
    777 		error = bus_dmamap_create(sc->sc_dmat, MCLBYTES, 1, MCLBYTES,
    778 		    0, 0, &sc->re_ldata.re_rxsoft[i].rxs_dmamap);
    779 		if (error) {
    780 			aprint_error_dev(sc->sc_dev,
    781 			    "can't create DMA map for RX\n");
    782 			goto fail_8;
    783 		}
    784 	}
    785 
    786 	/*
    787 	 * Record interface as attached. From here, we should not fail.
    788 	 */
    789 	sc->sc_flags |= RTK_ATTACHED;
    790 
    791 	ifp = &sc->ethercom.ec_if;
    792 	ifp->if_softc = sc;
    793 	strlcpy(ifp->if_xname, device_xname(sc->sc_dev), IFNAMSIZ);
    794 	ifp->if_mtu = ETHERMTU;
    795 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
    796 	ifp->if_ioctl = re_ioctl;
    797 	sc->ethercom.ec_capabilities |=
    798 	    ETHERCAP_VLAN_MTU | ETHERCAP_VLAN_HWTAGGING;
    799 	ifp->if_start = re_start;
    800 	ifp->if_stop = re_stop;
    801 
    802 	/*
    803 	 * IFCAP_CSUM_IPv4_Tx on re(4) is broken for small packets,
    804 	 * so we have a workaround to handle the bug by padding
    805 	 * such packets manually.
    806 	 */
    807 	ifp->if_capabilities |=
    808 	    IFCAP_CSUM_IPv4_Tx | IFCAP_CSUM_IPv4_Rx |
    809 	    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
    810 	    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx |
    811 	    IFCAP_TSOv4;
    812 
    813 	/*
    814 	 * XXX
    815 	 * Still have no idea how to make TSO work on 8168C, 8168CP,
    816 	 * 8102E, 8111C and 8111CP.
    817 	 */
    818 	if ((sc->sc_quirk & RTKQ_DESCV2) != 0)
    819 		ifp->if_capabilities &= ~IFCAP_TSOv4;
    820 
    821 	ifp->if_watchdog = re_watchdog;
    822 	ifp->if_init = re_init;
    823 	ifp->if_snd.ifq_maxlen = RE_IFQ_MAXLEN;
    824 	ifp->if_capenable = ifp->if_capabilities;
    825 	IFQ_SET_READY(&ifp->if_snd);
    826 
    827 	callout_init(&sc->rtk_tick_ch, 0);
    828 
    829 	/* Do MII setup */
    830 	sc->mii.mii_ifp = ifp;
    831 	sc->mii.mii_readreg = re_miibus_readreg;
    832 	sc->mii.mii_writereg = re_miibus_writereg;
    833 	sc->mii.mii_statchg = re_miibus_statchg;
    834 	sc->ethercom.ec_mii = &sc->mii;
    835 	ifmedia_init(&sc->mii.mii_media, IFM_IMASK, ether_mediachange,
    836 	    ether_mediastatus);
    837 	mii_attach(sc->sc_dev, &sc->mii, 0xffffffff, MII_PHY_ANY,
    838 	    MII_OFFSET_ANY, 0);
    839 	ifmedia_set(&sc->mii.mii_media, IFM_ETHER | IFM_AUTO);
    840 
    841 	/*
    842 	 * Call MI attach routine.
    843 	 */
    844 	if_attach(ifp);
    845 	ether_ifattach(ifp, eaddr);
    846 
    847 	if (pmf_device_register(sc->sc_dev, NULL, NULL))
    848 		pmf_class_network_register(sc->sc_dev, ifp);
    849 	else
    850 		aprint_error_dev(sc->sc_dev,
    851 		    "couldn't establish power handler\n");
    852 
    853 	return;
    854 
    855  fail_8:
    856 	/* Destroy DMA maps for RX buffers. */
    857 	for (i = 0; i < RE_RX_DESC_CNT; i++)
    858 		if (sc->re_ldata.re_rxsoft[i].rxs_dmamap != NULL)
    859 			bus_dmamap_destroy(sc->sc_dmat,
    860 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
    861 
    862 	/* Free DMA'able memory for the RX ring. */
    863 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
    864  fail_7:
    865 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
    866  fail_6:
    867 	bus_dmamem_unmap(sc->sc_dmat,
    868 	    (void *)sc->re_ldata.re_rx_list, RE_RX_DMAMEM_SZ);
    869  fail_5:
    870 	bus_dmamem_free(sc->sc_dmat,
    871 	    &sc->re_ldata.re_rx_listseg, sc->re_ldata.re_rx_listnseg);
    872 
    873  fail_4:
    874 	/* Destroy DMA maps for TX buffers. */
    875 	for (i = 0; i < RE_TX_QLEN; i++)
    876 		if (sc->re_ldata.re_txq[i].txq_dmamap != NULL)
    877 			bus_dmamap_destroy(sc->sc_dmat,
    878 			    sc->re_ldata.re_txq[i].txq_dmamap);
    879 
    880 	/* Free DMA'able memory for the TX ring. */
    881 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
    882  fail_3:
    883 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
    884  fail_2:
    885 	bus_dmamem_unmap(sc->sc_dmat,
    886 	    (void *)sc->re_ldata.re_tx_list, RE_TX_LIST_SZ(sc));
    887  fail_1:
    888 	bus_dmamem_free(sc->sc_dmat,
    889 	    &sc->re_ldata.re_tx_listseg, sc->re_ldata.re_tx_listnseg);
    890  fail_0:
    891 	return;
    892 }
    893 
    894 
    895 /*
    896  * re_activate:
    897  *     Handle device activation/deactivation requests.
    898  */
    899 int
    900 re_activate(device_t self, enum devact act)
    901 {
    902 	struct rtk_softc *sc = device_private(self);
    903 
    904 	switch (act) {
    905 	case DVACT_DEACTIVATE:
    906 		if_deactivate(&sc->ethercom.ec_if);
    907 		return 0;
    908 	default:
    909 		return EOPNOTSUPP;
    910 	}
    911 }
    912 
    913 /*
    914  * re_detach:
    915  *     Detach a rtk interface.
    916  */
    917 int
    918 re_detach(struct rtk_softc *sc)
    919 {
    920 	struct ifnet *ifp = &sc->ethercom.ec_if;
    921 	int i;
    922 
    923 	/*
    924 	 * Succeed now if there isn't any work to do.
    925 	 */
    926 	if ((sc->sc_flags & RTK_ATTACHED) == 0)
    927 		return 0;
    928 
    929 	/* Unhook our tick handler. */
    930 	callout_stop(&sc->rtk_tick_ch);
    931 
    932 	/* Detach all PHYs. */
    933 	mii_detach(&sc->mii, MII_PHY_ANY, MII_OFFSET_ANY);
    934 
    935 	/* Delete all remaining media. */
    936 	ifmedia_delete_instance(&sc->mii.mii_media, IFM_INST_ANY);
    937 
    938 	ether_ifdetach(ifp);
    939 	if_detach(ifp);
    940 
    941 	/* Destroy DMA maps for RX buffers. */
    942 	for (i = 0; i < RE_RX_DESC_CNT; i++)
    943 		if (sc->re_ldata.re_rxsoft[i].rxs_dmamap != NULL)
    944 			bus_dmamap_destroy(sc->sc_dmat,
    945 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
    946 
    947 	/* Free DMA'able memory for the RX ring. */
    948 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
    949 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
    950 	bus_dmamem_unmap(sc->sc_dmat,
    951 	    (void *)sc->re_ldata.re_rx_list, RE_RX_DMAMEM_SZ);
    952 	bus_dmamem_free(sc->sc_dmat,
    953 	    &sc->re_ldata.re_rx_listseg, sc->re_ldata.re_rx_listnseg);
    954 
    955 	/* Destroy DMA maps for TX buffers. */
    956 	for (i = 0; i < RE_TX_QLEN; i++)
    957 		if (sc->re_ldata.re_txq[i].txq_dmamap != NULL)
    958 			bus_dmamap_destroy(sc->sc_dmat,
    959 			    sc->re_ldata.re_txq[i].txq_dmamap);
    960 
    961 	/* Free DMA'able memory for the TX ring. */
    962 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
    963 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
    964 	bus_dmamem_unmap(sc->sc_dmat,
    965 	    (void *)sc->re_ldata.re_tx_list, RE_TX_LIST_SZ(sc));
    966 	bus_dmamem_free(sc->sc_dmat,
    967 	    &sc->re_ldata.re_tx_listseg, sc->re_ldata.re_tx_listnseg);
    968 
    969 	pmf_device_deregister(sc->sc_dev);
    970 
    971 	/* we don't want to run again */
    972 	sc->sc_flags &= ~RTK_ATTACHED;
    973 
    974 	return 0;
    975 }
    976 
    977 /*
    978  * re_enable:
    979  *     Enable the RTL81X9 chip.
    980  */
    981 static int
    982 re_enable(struct rtk_softc *sc)
    983 {
    984 
    985 	if (RTK_IS_ENABLED(sc) == 0 && sc->sc_enable != NULL) {
    986 		if ((*sc->sc_enable)(sc) != 0) {
    987 			printf("%s: device enable failed\n",
    988 			    device_xname(sc->sc_dev));
    989 			return EIO;
    990 		}
    991 		sc->sc_flags |= RTK_ENABLED;
    992 	}
    993 	return 0;
    994 }
    995 
    996 /*
    997  * re_disable:
    998  *     Disable the RTL81X9 chip.
    999  */
   1000 static void
   1001 re_disable(struct rtk_softc *sc)
   1002 {
   1003 
   1004 	if (RTK_IS_ENABLED(sc) && sc->sc_disable != NULL) {
   1005 		(*sc->sc_disable)(sc);
   1006 		sc->sc_flags &= ~RTK_ENABLED;
   1007 	}
   1008 }
   1009 
   1010 static int
   1011 re_newbuf(struct rtk_softc *sc, int idx, struct mbuf *m)
   1012 {
   1013 	struct mbuf *n = NULL;
   1014 	bus_dmamap_t map;
   1015 	struct re_desc *d;
   1016 	struct re_rxsoft *rxs;
   1017 	uint32_t cmdstat;
   1018 	int error;
   1019 
   1020 	if (m == NULL) {
   1021 		MGETHDR(n, M_DONTWAIT, MT_DATA);
   1022 		if (n == NULL)
   1023 			return ENOBUFS;
   1024 
   1025 		MCLGET(n, M_DONTWAIT);
   1026 		if ((n->m_flags & M_EXT) == 0) {
   1027 			m_freem(n);
   1028 			return ENOBUFS;
   1029 		}
   1030 		m = n;
   1031 	} else
   1032 		m->m_data = m->m_ext.ext_buf;
   1033 
   1034 	/*
   1035 	 * Initialize mbuf length fields and fixup
   1036 	 * alignment so that the frame payload is
   1037 	 * longword aligned.
   1038 	 */
   1039 	m->m_len = m->m_pkthdr.len = MCLBYTES - RE_ETHER_ALIGN;
   1040 	m->m_data += RE_ETHER_ALIGN;
   1041 
   1042 	rxs = &sc->re_ldata.re_rxsoft[idx];
   1043 	map = rxs->rxs_dmamap;
   1044 	error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m,
   1045 	    BUS_DMA_READ|BUS_DMA_NOWAIT);
   1046 
   1047 	if (error)
   1048 		goto out;
   1049 
   1050 	bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
   1051 	    BUS_DMASYNC_PREREAD);
   1052 
   1053 	d = &sc->re_ldata.re_rx_list[idx];
   1054 #ifdef DIAGNOSTIC
   1055 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
   1056 	cmdstat = le32toh(d->re_cmdstat);
   1057 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD);
   1058 	if (cmdstat & RE_RDESC_STAT_OWN) {
   1059 		panic("%s: tried to map busy RX descriptor",
   1060 		    device_xname(sc->sc_dev));
   1061 	}
   1062 #endif
   1063 
   1064 	rxs->rxs_mbuf = m;
   1065 
   1066 	d->re_vlanctl = 0;
   1067 	cmdstat = map->dm_segs[0].ds_len;
   1068 	if (idx == (RE_RX_DESC_CNT - 1))
   1069 		cmdstat |= RE_RDESC_CMD_EOR;
   1070 	re_set_bufaddr(d, map->dm_segs[0].ds_addr);
   1071 	d->re_cmdstat = htole32(cmdstat);
   1072 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1073 	cmdstat |= RE_RDESC_CMD_OWN;
   1074 	d->re_cmdstat = htole32(cmdstat);
   1075 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1076 
   1077 	return 0;
   1078  out:
   1079 	if (n != NULL)
   1080 		m_freem(n);
   1081 	return ENOMEM;
   1082 }
   1083 
   1084 static int
   1085 re_tx_list_init(struct rtk_softc *sc)
   1086 {
   1087 	int i;
   1088 
   1089 	memset(sc->re_ldata.re_tx_list, 0, RE_TX_LIST_SZ(sc));
   1090 	for (i = 0; i < RE_TX_QLEN; i++) {
   1091 		sc->re_ldata.re_txq[i].txq_mbuf = NULL;
   1092 	}
   1093 
   1094 	bus_dmamap_sync(sc->sc_dmat,
   1095 	    sc->re_ldata.re_tx_list_map, 0,
   1096 	    sc->re_ldata.re_tx_list_map->dm_mapsize,
   1097 	    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1098 	sc->re_ldata.re_txq_prodidx = 0;
   1099 	sc->re_ldata.re_txq_considx = 0;
   1100 	sc->re_ldata.re_txq_free = RE_TX_QLEN;
   1101 	sc->re_ldata.re_tx_free = RE_TX_DESC_CNT(sc);
   1102 	sc->re_ldata.re_tx_nextfree = 0;
   1103 
   1104 	return 0;
   1105 }
   1106 
   1107 static int
   1108 re_rx_list_init(struct rtk_softc *sc)
   1109 {
   1110 	int i;
   1111 
   1112 	memset(sc->re_ldata.re_rx_list, 0, RE_RX_LIST_SZ);
   1113 
   1114 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
   1115 		if (re_newbuf(sc, i, NULL) == ENOBUFS)
   1116 			return ENOBUFS;
   1117 	}
   1118 
   1119 	sc->re_ldata.re_rx_prodidx = 0;
   1120 	sc->re_head = sc->re_tail = NULL;
   1121 
   1122 	return 0;
   1123 }
   1124 
   1125 /*
   1126  * RX handler for C+ and 8169. For the gigE chips, we support
   1127  * the reception of jumbo frames that have been fragmented
   1128  * across multiple 2K mbuf cluster buffers.
   1129  */
   1130 static void
   1131 re_rxeof(struct rtk_softc *sc)
   1132 {
   1133 	struct mbuf *m;
   1134 	struct ifnet *ifp;
   1135 	int i, total_len;
   1136 	struct re_desc *cur_rx;
   1137 	struct re_rxsoft *rxs;
   1138 	uint32_t rxstat, rxvlan;
   1139 
   1140 	ifp = &sc->ethercom.ec_if;
   1141 
   1142 	for (i = sc->re_ldata.re_rx_prodidx;; i = RE_NEXT_RX_DESC(sc, i)) {
   1143 		cur_rx = &sc->re_ldata.re_rx_list[i];
   1144 		RE_RXDESCSYNC(sc, i,
   1145 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
   1146 		rxstat = le32toh(cur_rx->re_cmdstat);
   1147 		rxvlan = le32toh(cur_rx->re_vlanctl);
   1148 		RE_RXDESCSYNC(sc, i, BUS_DMASYNC_PREREAD);
   1149 		if ((rxstat & RE_RDESC_STAT_OWN) != 0) {
   1150 			break;
   1151 		}
   1152 		total_len = rxstat & sc->re_rxlenmask;
   1153 		rxs = &sc->re_ldata.re_rxsoft[i];
   1154 		m = rxs->rxs_mbuf;
   1155 
   1156 		/* Invalidate the RX mbuf and unload its map */
   1157 
   1158 		bus_dmamap_sync(sc->sc_dmat,
   1159 		    rxs->rxs_dmamap, 0, rxs->rxs_dmamap->dm_mapsize,
   1160 		    BUS_DMASYNC_POSTREAD);
   1161 		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
   1162 
   1163 		if ((rxstat & RE_RDESC_STAT_EOF) == 0) {
   1164 			m->m_len = MCLBYTES - RE_ETHER_ALIGN;
   1165 			if (sc->re_head == NULL)
   1166 				sc->re_head = sc->re_tail = m;
   1167 			else {
   1168 				m->m_flags &= ~M_PKTHDR;
   1169 				sc->re_tail->m_next = m;
   1170 				sc->re_tail = m;
   1171 			}
   1172 			re_newbuf(sc, i, NULL);
   1173 			continue;
   1174 		}
   1175 
   1176 		/*
   1177 		 * NOTE: for the 8139C+, the frame length field
   1178 		 * is always 12 bits in size, but for the gigE chips,
   1179 		 * it is 13 bits (since the max RX frame length is 16K).
   1180 		 * Unfortunately, all 32 bits in the status word
   1181 		 * were already used, so to make room for the extra
   1182 		 * length bit, RealTek took out the 'frame alignment
   1183 		 * error' bit and shifted the other status bits
   1184 		 * over one slot. The OWN, EOR, FS and LS bits are
   1185 		 * still in the same places. We have already extracted
   1186 		 * the frame length and checked the OWN bit, so rather
   1187 		 * than using an alternate bit mapping, we shift the
   1188 		 * status bits one space to the right so we can evaluate
   1189 		 * them using the 8169 status as though it was in the
   1190 		 * same format as that of the 8139C+.
   1191 		 */
   1192 		if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0)
   1193 			rxstat >>= 1;
   1194 
   1195 		if (__predict_false((rxstat & RE_RDESC_STAT_RXERRSUM) != 0)) {
   1196 #ifdef RE_DEBUG
   1197 			printf("%s: RX error (rxstat = 0x%08x)",
   1198 			    device_xname(sc->sc_dev), rxstat);
   1199 			if (rxstat & RE_RDESC_STAT_FRALIGN)
   1200 				printf(", frame alignment error");
   1201 			if (rxstat & RE_RDESC_STAT_BUFOFLOW)
   1202 				printf(", out of buffer space");
   1203 			if (rxstat & RE_RDESC_STAT_FIFOOFLOW)
   1204 				printf(", FIFO overrun");
   1205 			if (rxstat & RE_RDESC_STAT_GIANT)
   1206 				printf(", giant packet");
   1207 			if (rxstat & RE_RDESC_STAT_RUNT)
   1208 				printf(", runt packet");
   1209 			if (rxstat & RE_RDESC_STAT_CRCERR)
   1210 				printf(", CRC error");
   1211 			printf("\n");
   1212 #endif
   1213 			ifp->if_ierrors++;
   1214 			/*
   1215 			 * If this is part of a multi-fragment packet,
   1216 			 * discard all the pieces.
   1217 			 */
   1218 			if (sc->re_head != NULL) {
   1219 				m_freem(sc->re_head);
   1220 				sc->re_head = sc->re_tail = NULL;
   1221 			}
   1222 			re_newbuf(sc, i, m);
   1223 			continue;
   1224 		}
   1225 
   1226 		/*
   1227 		 * If allocating a replacement mbuf fails,
   1228 		 * reload the current one.
   1229 		 */
   1230 
   1231 		if (__predict_false(re_newbuf(sc, i, NULL) != 0)) {
   1232 			ifp->if_ierrors++;
   1233 			if (sc->re_head != NULL) {
   1234 				m_freem(sc->re_head);
   1235 				sc->re_head = sc->re_tail = NULL;
   1236 			}
   1237 			re_newbuf(sc, i, m);
   1238 			continue;
   1239 		}
   1240 
   1241 		if (sc->re_head != NULL) {
   1242 			m->m_len = total_len % (MCLBYTES - RE_ETHER_ALIGN);
   1243 			/*
   1244 			 * Special case: if there's 4 bytes or less
   1245 			 * in this buffer, the mbuf can be discarded:
   1246 			 * the last 4 bytes is the CRC, which we don't
   1247 			 * care about anyway.
   1248 			 */
   1249 			if (m->m_len <= ETHER_CRC_LEN) {
   1250 				sc->re_tail->m_len -=
   1251 				    (ETHER_CRC_LEN - m->m_len);
   1252 				m_freem(m);
   1253 			} else {
   1254 				m->m_len -= ETHER_CRC_LEN;
   1255 				m->m_flags &= ~M_PKTHDR;
   1256 				sc->re_tail->m_next = m;
   1257 			}
   1258 			m = sc->re_head;
   1259 			sc->re_head = sc->re_tail = NULL;
   1260 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
   1261 		} else
   1262 			m->m_pkthdr.len = m->m_len =
   1263 			    (total_len - ETHER_CRC_LEN);
   1264 
   1265 		ifp->if_ipackets++;
   1266 		m->m_pkthdr.rcvif = ifp;
   1267 
   1268 		/* Do RX checksumming */
   1269 		if ((sc->sc_quirk & RTKQ_DESCV2) == 0) {
   1270 			/* Check IP header checksum */
   1271 			if ((rxstat & RE_RDESC_STAT_PROTOID) != 0) {
   1272 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
   1273 				if (rxstat & RE_RDESC_STAT_IPSUMBAD)
   1274 					m->m_pkthdr.csum_flags |=
   1275 					    M_CSUM_IPv4_BAD;
   1276 
   1277 				/* Check TCP/UDP checksum */
   1278 				if (RE_TCPPKT(rxstat)) {
   1279 					m->m_pkthdr.csum_flags |= M_CSUM_TCPv4;
   1280 					if (rxstat & RE_RDESC_STAT_TCPSUMBAD)
   1281 						m->m_pkthdr.csum_flags |=
   1282 						    M_CSUM_TCP_UDP_BAD;
   1283 				} else if (RE_UDPPKT(rxstat)) {
   1284 					m->m_pkthdr.csum_flags |= M_CSUM_UDPv4;
   1285 					if (rxstat & RE_RDESC_STAT_UDPSUMBAD)
   1286 						m->m_pkthdr.csum_flags |=
   1287 						    M_CSUM_TCP_UDP_BAD;
   1288 				}
   1289 			}
   1290 		} else {
   1291 			/* Check IPv4 header checksum */
   1292 			if ((rxvlan & RE_RDESC_VLANCTL_IPV4) != 0) {
   1293 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
   1294 				if (rxstat & RE_RDESC_STAT_IPSUMBAD)
   1295 					m->m_pkthdr.csum_flags |=
   1296 					    M_CSUM_IPv4_BAD;
   1297 
   1298 				/* Check TCPv4/UDPv4 checksum */
   1299 				if (RE_TCPPKT(rxstat)) {
   1300 					m->m_pkthdr.csum_flags |= M_CSUM_TCPv4;
   1301 					if (rxstat & RE_RDESC_STAT_TCPSUMBAD)
   1302 						m->m_pkthdr.csum_flags |=
   1303 						    M_CSUM_TCP_UDP_BAD;
   1304 				} else if (RE_UDPPKT(rxstat)) {
   1305 					m->m_pkthdr.csum_flags |= M_CSUM_UDPv4;
   1306 					if (rxstat & RE_RDESC_STAT_UDPSUMBAD)
   1307 						m->m_pkthdr.csum_flags |=
   1308 						    M_CSUM_TCP_UDP_BAD;
   1309 				}
   1310 			}
   1311 			/* XXX Check TCPv6/UDPv6 checksum? */
   1312 		}
   1313 
   1314 		if (rxvlan & RE_RDESC_VLANCTL_TAG) {
   1315 			VLAN_INPUT_TAG(ifp, m,
   1316 			     bswap16(rxvlan & RE_RDESC_VLANCTL_DATA),
   1317 			     continue);
   1318 		}
   1319 		bpf_mtap(ifp, m);
   1320 		(*ifp->if_input)(ifp, m);
   1321 	}
   1322 
   1323 	sc->re_ldata.re_rx_prodidx = i;
   1324 }
   1325 
   1326 static void
   1327 re_txeof(struct rtk_softc *sc)
   1328 {
   1329 	struct ifnet *ifp;
   1330 	struct re_txq *txq;
   1331 	uint32_t txstat;
   1332 	int idx, descidx;
   1333 
   1334 	ifp = &sc->ethercom.ec_if;
   1335 
   1336 	for (idx = sc->re_ldata.re_txq_considx;
   1337 	    sc->re_ldata.re_txq_free < RE_TX_QLEN;
   1338 	    idx = RE_NEXT_TXQ(sc, idx), sc->re_ldata.re_txq_free++) {
   1339 		txq = &sc->re_ldata.re_txq[idx];
   1340 		KASSERT(txq->txq_mbuf != NULL);
   1341 
   1342 		descidx = txq->txq_descidx;
   1343 		RE_TXDESCSYNC(sc, descidx,
   1344 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
   1345 		txstat =
   1346 		    le32toh(sc->re_ldata.re_tx_list[descidx].re_cmdstat);
   1347 		RE_TXDESCSYNC(sc, descidx, BUS_DMASYNC_PREREAD);
   1348 		KASSERT((txstat & RE_TDESC_CMD_EOF) != 0);
   1349 		if (txstat & RE_TDESC_CMD_OWN) {
   1350 			break;
   1351 		}
   1352 
   1353 		sc->re_ldata.re_tx_free += txq->txq_nsegs;
   1354 		KASSERT(sc->re_ldata.re_tx_free <= RE_TX_DESC_CNT(sc));
   1355 		bus_dmamap_sync(sc->sc_dmat, txq->txq_dmamap,
   1356 		    0, txq->txq_dmamap->dm_mapsize, BUS_DMASYNC_POSTWRITE);
   1357 		bus_dmamap_unload(sc->sc_dmat, txq->txq_dmamap);
   1358 		m_freem(txq->txq_mbuf);
   1359 		txq->txq_mbuf = NULL;
   1360 
   1361 		if (txstat & (RE_TDESC_STAT_EXCESSCOL | RE_TDESC_STAT_COLCNT))
   1362 			ifp->if_collisions++;
   1363 		if (txstat & RE_TDESC_STAT_TXERRSUM)
   1364 			ifp->if_oerrors++;
   1365 		else
   1366 			ifp->if_opackets++;
   1367 	}
   1368 
   1369 	sc->re_ldata.re_txq_considx = idx;
   1370 
   1371 	if (sc->re_ldata.re_txq_free > RE_NTXDESC_RSVD)
   1372 		ifp->if_flags &= ~IFF_OACTIVE;
   1373 
   1374 	/*
   1375 	 * If not all descriptors have been released reaped yet,
   1376 	 * reload the timer so that we will eventually get another
   1377 	 * interrupt that will cause us to re-enter this routine.
   1378 	 * This is done in case the transmitter has gone idle.
   1379 	 */
   1380 	if (sc->re_ldata.re_txq_free < RE_TX_QLEN) {
   1381 		CSR_WRITE_4(sc, RTK_TIMERCNT, 1);
   1382 		if ((sc->sc_quirk & RTKQ_PCIE) != 0) {
   1383 			/*
   1384 			 * Some chips will ignore a second TX request
   1385 			 * issued while an existing transmission is in
   1386 			 * progress. If the transmitter goes idle but
   1387 			 * there are still packets waiting to be sent,
   1388 			 * we need to restart the channel here to flush
   1389 			 * them out. This only seems to be required with
   1390 			 * the PCIe devices.
   1391 			 */
   1392 			CSR_WRITE_1(sc, RTK_GTXSTART, RTK_TXSTART_START);
   1393 		}
   1394 	} else
   1395 		ifp->if_timer = 0;
   1396 }
   1397 
   1398 static void
   1399 re_tick(void *arg)
   1400 {
   1401 	struct rtk_softc *sc = arg;
   1402 	int s;
   1403 
   1404 	/* XXX: just return for 8169S/8110S with rev 2 or newer phy */
   1405 	s = splnet();
   1406 
   1407 	mii_tick(&sc->mii);
   1408 	splx(s);
   1409 
   1410 	callout_reset(&sc->rtk_tick_ch, hz, re_tick, sc);
   1411 }
   1412 
   1413 int
   1414 re_intr(void *arg)
   1415 {
   1416 	struct rtk_softc *sc = arg;
   1417 	struct ifnet *ifp;
   1418 	uint16_t status;
   1419 	int handled = 0;
   1420 
   1421 	if (!device_has_power(sc->sc_dev))
   1422 		return 0;
   1423 
   1424 	ifp = &sc->ethercom.ec_if;
   1425 
   1426 	if ((ifp->if_flags & IFF_UP) == 0)
   1427 		return 0;
   1428 
   1429 	for (;;) {
   1430 
   1431 		status = CSR_READ_2(sc, RTK_ISR);
   1432 		/* If the card has gone away the read returns 0xffff. */
   1433 		if (status == 0xffff)
   1434 			break;
   1435 		if (status) {
   1436 			handled = 1;
   1437 			CSR_WRITE_2(sc, RTK_ISR, status);
   1438 		}
   1439 
   1440 		if ((status & RTK_INTRS_CPLUS) == 0)
   1441 			break;
   1442 
   1443 		if (status & (RTK_ISR_RX_OK | RTK_ISR_RX_ERR))
   1444 			re_rxeof(sc);
   1445 
   1446 		if (status & (RTK_ISR_TIMEOUT_EXPIRED | RTK_ISR_TX_ERR |
   1447 		    RTK_ISR_TX_DESC_UNAVAIL))
   1448 			re_txeof(sc);
   1449 
   1450 		if (status & RTK_ISR_SYSTEM_ERR) {
   1451 			re_init(ifp);
   1452 		}
   1453 
   1454 		if (status & RTK_ISR_LINKCHG) {
   1455 			callout_stop(&sc->rtk_tick_ch);
   1456 			re_tick(sc);
   1457 		}
   1458 	}
   1459 
   1460 	if (handled && !IFQ_IS_EMPTY(&ifp->if_snd))
   1461 		re_start(ifp);
   1462 
   1463 	return handled;
   1464 }
   1465 
   1466 
   1467 
   1468 /*
   1469  * Main transmit routine for C+ and gigE NICs.
   1470  */
   1471 
   1472 static void
   1473 re_start(struct ifnet *ifp)
   1474 {
   1475 	struct rtk_softc *sc;
   1476 	struct mbuf *m;
   1477 	bus_dmamap_t map;
   1478 	struct re_txq *txq;
   1479 	struct re_desc *d;
   1480 	struct m_tag *mtag;
   1481 	uint32_t cmdstat, re_flags, vlanctl;
   1482 	int ofree, idx, error, nsegs, seg;
   1483 	int startdesc, curdesc, lastdesc;
   1484 	bool pad;
   1485 
   1486 	sc = ifp->if_softc;
   1487 	ofree = sc->re_ldata.re_txq_free;
   1488 
   1489 	for (idx = sc->re_ldata.re_txq_prodidx;; idx = RE_NEXT_TXQ(sc, idx)) {
   1490 
   1491 		IFQ_POLL(&ifp->if_snd, m);
   1492 		if (m == NULL)
   1493 			break;
   1494 
   1495 		if (sc->re_ldata.re_txq_free == 0 ||
   1496 		    sc->re_ldata.re_tx_free == 0) {
   1497 			/* no more free slots left */
   1498 			ifp->if_flags |= IFF_OACTIVE;
   1499 			break;
   1500 		}
   1501 
   1502 		/*
   1503 		 * Set up checksum offload. Note: checksum offload bits must
   1504 		 * appear in all descriptors of a multi-descriptor transmit
   1505 		 * attempt. (This is according to testing done with an 8169
   1506 		 * chip. I'm not sure if this is a requirement or a bug.)
   1507 		 */
   1508 
   1509 		vlanctl = 0;
   1510 		if ((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) != 0) {
   1511 			uint32_t segsz = m->m_pkthdr.segsz;
   1512 
   1513 			re_flags = RE_TDESC_CMD_LGSEND |
   1514 			    (segsz << RE_TDESC_CMD_MSSVAL_SHIFT);
   1515 		} else {
   1516 			/*
   1517 			 * set RE_TDESC_CMD_IPCSUM if any checksum offloading
   1518 			 * is requested.  otherwise, RE_TDESC_CMD_TCPCSUM/
   1519 			 * RE_TDESC_CMD_UDPCSUM doesn't make effects.
   1520 			 */
   1521 			re_flags = 0;
   1522 			if ((m->m_pkthdr.csum_flags &
   1523 			    (M_CSUM_IPv4 | M_CSUM_TCPv4 | M_CSUM_UDPv4))
   1524 			    != 0) {
   1525 				if ((sc->sc_quirk & RTKQ_DESCV2) == 0) {
   1526 					re_flags |= RE_TDESC_CMD_IPCSUM;
   1527 					if (m->m_pkthdr.csum_flags &
   1528 					    M_CSUM_TCPv4) {
   1529 						re_flags |=
   1530 						    RE_TDESC_CMD_TCPCSUM;
   1531 					} else if (m->m_pkthdr.csum_flags &
   1532 					    M_CSUM_UDPv4) {
   1533 						re_flags |=
   1534 						    RE_TDESC_CMD_UDPCSUM;
   1535 					}
   1536 				} else {
   1537 					vlanctl |= RE_TDESC_VLANCTL_IPCSUM;
   1538 					if (m->m_pkthdr.csum_flags &
   1539 					    M_CSUM_TCPv4) {
   1540 						vlanctl |=
   1541 						    RE_TDESC_VLANCTL_TCPCSUM;
   1542 					} else if (m->m_pkthdr.csum_flags &
   1543 					    M_CSUM_UDPv4) {
   1544 						vlanctl |=
   1545 						    RE_TDESC_VLANCTL_UDPCSUM;
   1546 					}
   1547 				}
   1548 			}
   1549 		}
   1550 
   1551 		txq = &sc->re_ldata.re_txq[idx];
   1552 		map = txq->txq_dmamap;
   1553 		error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m,
   1554 		    BUS_DMA_WRITE|BUS_DMA_NOWAIT);
   1555 
   1556 		if (__predict_false(error)) {
   1557 			/* XXX try to defrag if EFBIG? */
   1558 			printf("%s: can't map mbuf (error %d)\n",
   1559 			    device_xname(sc->sc_dev), error);
   1560 
   1561 			IFQ_DEQUEUE(&ifp->if_snd, m);
   1562 			m_freem(m);
   1563 			ifp->if_oerrors++;
   1564 			continue;
   1565 		}
   1566 
   1567 		nsegs = map->dm_nsegs;
   1568 		pad = false;
   1569 		if (__predict_false(m->m_pkthdr.len <= RE_IP4CSUMTX_PADLEN &&
   1570 		    (re_flags & RE_TDESC_CMD_IPCSUM) != 0 &&
   1571 		    (sc->sc_quirk & RTKQ_DESCV2) == 0)) {
   1572 			pad = true;
   1573 			nsegs++;
   1574 		}
   1575 
   1576 		if (nsegs > sc->re_ldata.re_tx_free) {
   1577 			/*
   1578 			 * Not enough free descriptors to transmit this packet.
   1579 			 */
   1580 			ifp->if_flags |= IFF_OACTIVE;
   1581 			bus_dmamap_unload(sc->sc_dmat, map);
   1582 			break;
   1583 		}
   1584 
   1585 		IFQ_DEQUEUE(&ifp->if_snd, m);
   1586 
   1587 		/*
   1588 		 * Make sure that the caches are synchronized before we
   1589 		 * ask the chip to start DMA for the packet data.
   1590 		 */
   1591 		bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
   1592 		    BUS_DMASYNC_PREWRITE);
   1593 
   1594 		/*
   1595 		 * Set up hardware VLAN tagging. Note: vlan tag info must
   1596 		 * appear in all descriptors of a multi-descriptor
   1597 		 * transmission attempt.
   1598 		 */
   1599 		if ((mtag = VLAN_OUTPUT_TAG(&sc->ethercom, m)) != NULL)
   1600 			vlanctl |= bswap16(VLAN_TAG_VALUE(mtag)) |
   1601 			    RE_TDESC_VLANCTL_TAG;
   1602 
   1603 		/*
   1604 		 * Map the segment array into descriptors.
   1605 		 * Note that we set the start-of-frame and
   1606 		 * end-of-frame markers for either TX or RX,
   1607 		 * but they really only have meaning in the TX case.
   1608 		 * (In the RX case, it's the chip that tells us
   1609 		 *  where packets begin and end.)
   1610 		 * We also keep track of the end of the ring
   1611 		 * and set the end-of-ring bits as needed,
   1612 		 * and we set the ownership bits in all except
   1613 		 * the very first descriptor. (The caller will
   1614 		 * set this descriptor later when it start
   1615 		 * transmission or reception.)
   1616 		 */
   1617 		curdesc = startdesc = sc->re_ldata.re_tx_nextfree;
   1618 		lastdesc = -1;
   1619 		for (seg = 0; seg < map->dm_nsegs;
   1620 		    seg++, curdesc = RE_NEXT_TX_DESC(sc, curdesc)) {
   1621 			d = &sc->re_ldata.re_tx_list[curdesc];
   1622 #ifdef DIAGNOSTIC
   1623 			RE_TXDESCSYNC(sc, curdesc,
   1624 			    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
   1625 			cmdstat = le32toh(d->re_cmdstat);
   1626 			RE_TXDESCSYNC(sc, curdesc, BUS_DMASYNC_PREREAD);
   1627 			if (cmdstat & RE_TDESC_STAT_OWN) {
   1628 				panic("%s: tried to map busy TX descriptor",
   1629 				    device_xname(sc->sc_dev));
   1630 			}
   1631 #endif
   1632 
   1633 			d->re_vlanctl = htole32(vlanctl);
   1634 			re_set_bufaddr(d, map->dm_segs[seg].ds_addr);
   1635 			cmdstat = re_flags | map->dm_segs[seg].ds_len;
   1636 			if (seg == 0)
   1637 				cmdstat |= RE_TDESC_CMD_SOF;
   1638 			else
   1639 				cmdstat |= RE_TDESC_CMD_OWN;
   1640 			if (curdesc == (RE_TX_DESC_CNT(sc) - 1))
   1641 				cmdstat |= RE_TDESC_CMD_EOR;
   1642 			if (seg == nsegs - 1) {
   1643 				cmdstat |= RE_TDESC_CMD_EOF;
   1644 				lastdesc = curdesc;
   1645 			}
   1646 			d->re_cmdstat = htole32(cmdstat);
   1647 			RE_TXDESCSYNC(sc, curdesc,
   1648 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1649 		}
   1650 		if (__predict_false(pad)) {
   1651 			d = &sc->re_ldata.re_tx_list[curdesc];
   1652 			d->re_vlanctl = htole32(vlanctl);
   1653 			re_set_bufaddr(d, RE_TXPADDADDR(sc));
   1654 			cmdstat = re_flags |
   1655 			    RE_TDESC_CMD_OWN | RE_TDESC_CMD_EOF |
   1656 			    (RE_IP4CSUMTX_PADLEN + 1 - m->m_pkthdr.len);
   1657 			if (curdesc == (RE_TX_DESC_CNT(sc) - 1))
   1658 				cmdstat |= RE_TDESC_CMD_EOR;
   1659 			d->re_cmdstat = htole32(cmdstat);
   1660 			RE_TXDESCSYNC(sc, curdesc,
   1661 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1662 			lastdesc = curdesc;
   1663 			curdesc = RE_NEXT_TX_DESC(sc, curdesc);
   1664 		}
   1665 		KASSERT(lastdesc != -1);
   1666 
   1667 		/* Transfer ownership of packet to the chip. */
   1668 
   1669 		sc->re_ldata.re_tx_list[startdesc].re_cmdstat |=
   1670 		    htole32(RE_TDESC_CMD_OWN);
   1671 		RE_TXDESCSYNC(sc, startdesc,
   1672 		    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1673 
   1674 		/* update info of TX queue and descriptors */
   1675 		txq->txq_mbuf = m;
   1676 		txq->txq_descidx = lastdesc;
   1677 		txq->txq_nsegs = nsegs;
   1678 
   1679 		sc->re_ldata.re_txq_free--;
   1680 		sc->re_ldata.re_tx_free -= nsegs;
   1681 		sc->re_ldata.re_tx_nextfree = curdesc;
   1682 
   1683 		/*
   1684 		 * If there's a BPF listener, bounce a copy of this frame
   1685 		 * to him.
   1686 		 */
   1687 		bpf_mtap(ifp, m);
   1688 	}
   1689 
   1690 	if (sc->re_ldata.re_txq_free < ofree) {
   1691 		/*
   1692 		 * TX packets are enqueued.
   1693 		 */
   1694 		sc->re_ldata.re_txq_prodidx = idx;
   1695 
   1696 		/*
   1697 		 * Start the transmitter to poll.
   1698 		 *
   1699 		 * RealTek put the TX poll request register in a different
   1700 		 * location on the 8169 gigE chip. I don't know why.
   1701 		 */
   1702 		if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0)
   1703 			CSR_WRITE_1(sc, RTK_TXSTART, RTK_TXSTART_START);
   1704 		else
   1705 			CSR_WRITE_1(sc, RTK_GTXSTART, RTK_TXSTART_START);
   1706 
   1707 		/*
   1708 		 * Use the countdown timer for interrupt moderation.
   1709 		 * 'TX done' interrupts are disabled. Instead, we reset the
   1710 		 * countdown timer, which will begin counting until it hits
   1711 		 * the value in the TIMERINT register, and then trigger an
   1712 		 * interrupt. Each time we write to the TIMERCNT register,
   1713 		 * the timer count is reset to 0.
   1714 		 */
   1715 		CSR_WRITE_4(sc, RTK_TIMERCNT, 1);
   1716 
   1717 		/*
   1718 		 * Set a timeout in case the chip goes out to lunch.
   1719 		 */
   1720 		ifp->if_timer = 5;
   1721 	}
   1722 }
   1723 
   1724 static int
   1725 re_init(struct ifnet *ifp)
   1726 {
   1727 	struct rtk_softc *sc = ifp->if_softc;
   1728 	const uint8_t *enaddr;
   1729 	uint32_t rxcfg = 0;
   1730 	uint32_t reg;
   1731 	uint16_t cfg;
   1732 	int error;
   1733 
   1734 	if ((error = re_enable(sc)) != 0)
   1735 		goto out;
   1736 
   1737 	/*
   1738 	 * Cancel pending I/O and free all RX/TX buffers.
   1739 	 */
   1740 	re_stop(ifp, 0);
   1741 
   1742 	re_reset(sc);
   1743 
   1744 	/*
   1745 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
   1746 	 * RX checksum offload. We must configure the C+ register
   1747 	 * before all others.
   1748 	 */
   1749 	cfg = RE_CPLUSCMD_PCI_MRW;
   1750 
   1751 	/*
   1752 	 * XXX: For old 8169 set bit 14.
   1753 	 *      For 8169S/8110S and above, do not set bit 14.
   1754 	 */
   1755 	if ((sc->sc_quirk & RTKQ_8169NONS) != 0)
   1756 		cfg |= (0x1 << 14);
   1757 
   1758 	if ((sc->ethercom.ec_capenable & ETHERCAP_VLAN_HWTAGGING) != 0)
   1759 		cfg |= RE_CPLUSCMD_VLANSTRIP;
   1760 	if ((ifp->if_capenable & (IFCAP_CSUM_IPv4_Rx |
   1761 	     IFCAP_CSUM_TCPv4_Rx | IFCAP_CSUM_UDPv4_Rx)) != 0)
   1762 		cfg |= RE_CPLUSCMD_RXCSUM_ENB;
   1763 	if ((sc->sc_quirk & RTKQ_MACSTAT) != 0) {
   1764 		cfg |= RE_CPLUSCMD_MACSTAT_DIS;
   1765 		cfg |= RE_CPLUSCMD_TXENB;
   1766 	} else
   1767 		cfg |= RE_CPLUSCMD_RXENB | RE_CPLUSCMD_TXENB;
   1768 
   1769 	CSR_WRITE_2(sc, RTK_CPLUS_CMD, cfg);
   1770 
   1771 	/* XXX: from Realtek-supplied Linux driver. Wholly undocumented. */
   1772 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0)
   1773 		CSR_WRITE_2(sc, RTK_IM, 0x0000);
   1774 
   1775 	DELAY(10000);
   1776 
   1777 	/*
   1778 	 * Init our MAC address.  Even though the chipset
   1779 	 * documentation doesn't mention it, we need to enter "Config
   1780 	 * register write enable" mode to modify the ID registers.
   1781 	 */
   1782 	CSR_WRITE_1(sc, RTK_EECMD, RTK_EEMODE_WRITECFG);
   1783 	enaddr = CLLADDR(ifp->if_sadl);
   1784 	reg = enaddr[0] | (enaddr[1] << 8) |
   1785 	    (enaddr[2] << 16) | (enaddr[3] << 24);
   1786 	CSR_WRITE_4(sc, RTK_IDR0, reg);
   1787 	reg = enaddr[4] | (enaddr[5] << 8);
   1788 	CSR_WRITE_4(sc, RTK_IDR4, reg);
   1789 	CSR_WRITE_1(sc, RTK_EECMD, RTK_EEMODE_OFF);
   1790 
   1791 	/*
   1792 	 * For C+ mode, initialize the RX descriptors and mbufs.
   1793 	 */
   1794 	re_rx_list_init(sc);
   1795 	re_tx_list_init(sc);
   1796 
   1797 	/*
   1798 	 * Load the addresses of the RX and TX lists into the chip.
   1799 	 */
   1800 	CSR_WRITE_4(sc, RTK_RXLIST_ADDR_HI,
   1801 	    RE_ADDR_HI(sc->re_ldata.re_rx_list_map->dm_segs[0].ds_addr));
   1802 	CSR_WRITE_4(sc, RTK_RXLIST_ADDR_LO,
   1803 	    RE_ADDR_LO(sc->re_ldata.re_rx_list_map->dm_segs[0].ds_addr));
   1804 
   1805 	CSR_WRITE_4(sc, RTK_TXLIST_ADDR_HI,
   1806 	    RE_ADDR_HI(sc->re_ldata.re_tx_list_map->dm_segs[0].ds_addr));
   1807 	CSR_WRITE_4(sc, RTK_TXLIST_ADDR_LO,
   1808 	    RE_ADDR_LO(sc->re_ldata.re_tx_list_map->dm_segs[0].ds_addr));
   1809 
   1810 	/*
   1811 	 * Enable transmit and receive.
   1812 	 */
   1813 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_TX_ENB | RTK_CMD_RX_ENB);
   1814 
   1815 	/*
   1816 	 * Set the initial TX and RX configuration.
   1817 	 */
   1818 	if (sc->re_testmode && (sc->sc_quirk & RTKQ_8169NONS) != 0) {
   1819 		/* test mode is needed only for old 8169 */
   1820 		CSR_WRITE_4(sc, RTK_TXCFG,
   1821 		    RE_TXCFG_CONFIG | RTK_LOOPTEST_ON);
   1822 	} else
   1823 		CSR_WRITE_4(sc, RTK_TXCFG, RE_TXCFG_CONFIG);
   1824 
   1825 	CSR_WRITE_1(sc, RTK_EARLY_TX_THRESH, 16);
   1826 
   1827 	CSR_WRITE_4(sc, RTK_RXCFG, RE_RXCFG_CONFIG);
   1828 
   1829 	/* Set the individual bit to receive frames for this host only. */
   1830 	rxcfg = CSR_READ_4(sc, RTK_RXCFG);
   1831 	rxcfg |= RTK_RXCFG_RX_INDIV;
   1832 
   1833 	/* If we want promiscuous mode, set the allframes bit. */
   1834 	if (ifp->if_flags & IFF_PROMISC)
   1835 		rxcfg |= RTK_RXCFG_RX_ALLPHYS;
   1836 	else
   1837 		rxcfg &= ~RTK_RXCFG_RX_ALLPHYS;
   1838 	CSR_WRITE_4(sc, RTK_RXCFG, rxcfg);
   1839 
   1840 	/*
   1841 	 * Set capture broadcast bit to capture broadcast frames.
   1842 	 */
   1843 	if (ifp->if_flags & IFF_BROADCAST)
   1844 		rxcfg |= RTK_RXCFG_RX_BROAD;
   1845 	else
   1846 		rxcfg &= ~RTK_RXCFG_RX_BROAD;
   1847 	CSR_WRITE_4(sc, RTK_RXCFG, rxcfg);
   1848 
   1849 	/*
   1850 	 * Program the multicast filter, if necessary.
   1851 	 */
   1852 	rtk_setmulti(sc);
   1853 
   1854 	/*
   1855 	 * Enable interrupts.
   1856 	 */
   1857 	if (sc->re_testmode)
   1858 		CSR_WRITE_2(sc, RTK_IMR, 0);
   1859 	else
   1860 		CSR_WRITE_2(sc, RTK_IMR, RTK_INTRS_CPLUS);
   1861 
   1862 	/* Start RX/TX process. */
   1863 	CSR_WRITE_4(sc, RTK_MISSEDPKT, 0);
   1864 #ifdef notdef
   1865 	/* Enable receiver and transmitter. */
   1866 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_TX_ENB | RTK_CMD_RX_ENB);
   1867 #endif
   1868 
   1869 	/*
   1870 	 * Initialize the timer interrupt register so that
   1871 	 * a timer interrupt will be generated once the timer
   1872 	 * reaches a certain number of ticks. The timer is
   1873 	 * reloaded on each transmit. This gives us TX interrupt
   1874 	 * moderation, which dramatically improves TX frame rate.
   1875 	 */
   1876 
   1877 	if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0)
   1878 		CSR_WRITE_4(sc, RTK_TIMERINT, 0x400);
   1879 	else {
   1880 		CSR_WRITE_4(sc, RTK_TIMERINT_8169, 0x800);
   1881 
   1882 		/*
   1883 		 * For 8169 gigE NICs, set the max allowed RX packet
   1884 		 * size so we can receive jumbo frames.
   1885 		 */
   1886 		CSR_WRITE_2(sc, RTK_MAXRXPKTLEN, 16383);
   1887 	}
   1888 
   1889 	if (sc->re_testmode)
   1890 		return 0;
   1891 
   1892 	CSR_WRITE_1(sc, RTK_CFG1, RTK_CFG1_DRVLOAD);
   1893 
   1894 	ifp->if_flags |= IFF_RUNNING;
   1895 	ifp->if_flags &= ~IFF_OACTIVE;
   1896 
   1897 	callout_reset(&sc->rtk_tick_ch, hz, re_tick, sc);
   1898 
   1899  out:
   1900 	if (error) {
   1901 		ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
   1902 		ifp->if_timer = 0;
   1903 		printf("%s: interface not running\n",
   1904 		    device_xname(sc->sc_dev));
   1905 	}
   1906 
   1907 	return error;
   1908 }
   1909 
   1910 static int
   1911 re_ioctl(struct ifnet *ifp, u_long command, void *data)
   1912 {
   1913 	struct rtk_softc *sc = ifp->if_softc;
   1914 	struct ifreq *ifr = data;
   1915 	int s, error = 0;
   1916 
   1917 	s = splnet();
   1918 
   1919 	switch (command) {
   1920 	case SIOCSIFMTU:
   1921 		/*
   1922 		 * Disable jumbo frames if it's not supported.
   1923 		 */
   1924 		if ((sc->sc_quirk & RTKQ_NOJUMBO) != 0 &&
   1925 		    ifr->ifr_mtu > ETHERMTU) {
   1926 			error = EINVAL;
   1927 			break;
   1928 		}
   1929 
   1930 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > ETHERMTU_JUMBO)
   1931 			error = EINVAL;
   1932 		else if ((error = ifioctl_common(ifp, command, data)) ==
   1933 		    ENETRESET)
   1934 			error = 0;
   1935 		break;
   1936 	default:
   1937 		if ((error = ether_ioctl(ifp, command, data)) != ENETRESET)
   1938 			break;
   1939 
   1940 		error = 0;
   1941 
   1942 		if (command == SIOCSIFCAP)
   1943 			error = (*ifp->if_init)(ifp);
   1944 		else if (command != SIOCADDMULTI && command != SIOCDELMULTI)
   1945 			;
   1946 		else if (ifp->if_flags & IFF_RUNNING)
   1947 			rtk_setmulti(sc);
   1948 		break;
   1949 	}
   1950 
   1951 	splx(s);
   1952 
   1953 	return error;
   1954 }
   1955 
   1956 static void
   1957 re_watchdog(struct ifnet *ifp)
   1958 {
   1959 	struct rtk_softc *sc;
   1960 	int s;
   1961 
   1962 	sc = ifp->if_softc;
   1963 	s = splnet();
   1964 	printf("%s: watchdog timeout\n", device_xname(sc->sc_dev));
   1965 	ifp->if_oerrors++;
   1966 
   1967 	re_txeof(sc);
   1968 	re_rxeof(sc);
   1969 
   1970 	re_init(ifp);
   1971 
   1972 	splx(s);
   1973 }
   1974 
   1975 /*
   1976  * Stop the adapter and free any mbufs allocated to the
   1977  * RX and TX lists.
   1978  */
   1979 static void
   1980 re_stop(struct ifnet *ifp, int disable)
   1981 {
   1982 	int i;
   1983 	struct rtk_softc *sc = ifp->if_softc;
   1984 
   1985 	callout_stop(&sc->rtk_tick_ch);
   1986 
   1987 	mii_down(&sc->mii);
   1988 
   1989 	if ((sc->sc_quirk & RTKQ_CMDSTOP) != 0)
   1990 		CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_STOPREQ | RTK_CMD_TX_ENB |
   1991 		    RTK_CMD_RX_ENB);
   1992 	else
   1993 		CSR_WRITE_1(sc, RTK_COMMAND, 0x00);
   1994 	DELAY(1000);
   1995 	CSR_WRITE_2(sc, RTK_IMR, 0x0000);
   1996 	CSR_WRITE_2(sc, RTK_ISR, 0xFFFF);
   1997 
   1998 	if (sc->re_head != NULL) {
   1999 		m_freem(sc->re_head);
   2000 		sc->re_head = sc->re_tail = NULL;
   2001 	}
   2002 
   2003 	/* Free the TX list buffers. */
   2004 	for (i = 0; i < RE_TX_QLEN; i++) {
   2005 		if (sc->re_ldata.re_txq[i].txq_mbuf != NULL) {
   2006 			bus_dmamap_unload(sc->sc_dmat,
   2007 			    sc->re_ldata.re_txq[i].txq_dmamap);
   2008 			m_freem(sc->re_ldata.re_txq[i].txq_mbuf);
   2009 			sc->re_ldata.re_txq[i].txq_mbuf = NULL;
   2010 		}
   2011 	}
   2012 
   2013 	/* Free the RX list buffers. */
   2014 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
   2015 		if (sc->re_ldata.re_rxsoft[i].rxs_mbuf != NULL) {
   2016 			bus_dmamap_unload(sc->sc_dmat,
   2017 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
   2018 			m_freem(sc->re_ldata.re_rxsoft[i].rxs_mbuf);
   2019 			sc->re_ldata.re_rxsoft[i].rxs_mbuf = NULL;
   2020 		}
   2021 	}
   2022 
   2023 	if (disable)
   2024 		re_disable(sc);
   2025 
   2026 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
   2027 	ifp->if_timer = 0;
   2028 }
   2029