Home | History | Annotate | Line # | Download | only in ic
rtl8169.c revision 1.134
      1 /*	$NetBSD: rtl8169.c,v 1.134 2011/11/22 18:42:56 garbled 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 1.134 2011/11/22 18:42:56 garbled 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(device_t);
    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(device_t dev)
    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_8100E:
    610 		case RTK_HWREV_8100E_SPIN2:
    611 		case RTK_HWREV_8101E:
    612 			sc->sc_quirk |= RTKQ_NOJUMBO;
    613 			break;
    614 		case RTK_HWREV_8102E:
    615 		case RTK_HWREV_8102EL:
    616 		case RTK_HWREV_8103E:
    617 			sc->sc_quirk |= RTKQ_DESCV2 | RTKQ_NOEECMD |
    618 			    RTKQ_MACSTAT | RTKQ_CMDSTOP | RTKQ_NOJUMBO;
    619 			break;
    620 		default:
    621 			aprint_normal_dev(sc->sc_dev,
    622 			    "Unknown revision (0x%08x)\n", hwrev);
    623 			/* assume the latest features */
    624 			sc->sc_quirk |= RTKQ_DESCV2 | RTKQ_NOEECMD;
    625 			sc->sc_quirk |= RTKQ_NOJUMBO;
    626 		}
    627 
    628 		/* Set RX length mask */
    629 		sc->re_rxlenmask = RE_RDESC_STAT_GFRAGLEN;
    630 		sc->re_ldata.re_tx_desc_cnt = RE_TX_DESC_CNT_8169;
    631 	} else {
    632 		sc->sc_quirk |= RTKQ_NOJUMBO;
    633 
    634 		/* Set RX length mask */
    635 		sc->re_rxlenmask = RE_RDESC_STAT_FRAGLEN;
    636 		sc->re_ldata.re_tx_desc_cnt = RE_TX_DESC_CNT_8139;
    637 	}
    638 
    639 	/* Reset the adapter. */
    640 	re_reset(sc);
    641 
    642 	if ((sc->sc_quirk & RTKQ_NOEECMD) != 0) {
    643 		/*
    644 		 * Get station address from ID registers.
    645 		 */
    646 		for (i = 0; i < ETHER_ADDR_LEN; i++)
    647 			eaddr[i] = CSR_READ_1(sc, RTK_IDR0 + i);
    648 	} else {
    649 		/*
    650 		 * Get station address from the EEPROM.
    651 		 */
    652 		if (rtk_read_eeprom(sc, RTK_EE_ID, RTK_EEADDR_LEN1) == 0x8129)
    653 			addr_len = RTK_EEADDR_LEN1;
    654 		else
    655 			addr_len = RTK_EEADDR_LEN0;
    656 
    657 		/*
    658 		 * Get station address from the EEPROM.
    659 		 */
    660 		for (i = 0; i < ETHER_ADDR_LEN / 2; i++) {
    661 			val = rtk_read_eeprom(sc, RTK_EE_EADDR0 + i, addr_len);
    662 			eaddr[(i * 2) + 0] = val & 0xff;
    663 			eaddr[(i * 2) + 1] = val >> 8;
    664 		}
    665 	}
    666 
    667 	/* Take PHY out of power down mode. */
    668 	if ((sc->sc_quirk & RTKQ_PHYWAKE_PM) != 0)
    669 		CSR_WRITE_1(sc, RTK_PMCH, CSR_READ_1(sc, RTK_PMCH) | 0x80);
    670 
    671 	aprint_normal_dev(sc->sc_dev, "Ethernet address %s\n",
    672 	    ether_sprintf(eaddr));
    673 
    674 	if (sc->re_ldata.re_tx_desc_cnt >
    675 	    PAGE_SIZE / sizeof(struct re_desc)) {
    676 		sc->re_ldata.re_tx_desc_cnt =
    677 		    PAGE_SIZE / sizeof(struct re_desc);
    678 	}
    679 
    680 	aprint_verbose_dev(sc->sc_dev, "using %d tx descriptors\n",
    681 	    sc->re_ldata.re_tx_desc_cnt);
    682 	KASSERT(RE_NEXT_TX_DESC(sc, RE_TX_DESC_CNT(sc) - 1) == 0);
    683 
    684 	/* Allocate DMA'able memory for the TX ring */
    685 	if ((error = bus_dmamem_alloc(sc->sc_dmat, RE_TX_LIST_SZ(sc),
    686 	    RE_RING_ALIGN, 0, &sc->re_ldata.re_tx_listseg, 1,
    687 	    &sc->re_ldata.re_tx_listnseg, BUS_DMA_NOWAIT)) != 0) {
    688 		aprint_error_dev(sc->sc_dev,
    689 		    "can't allocate tx listseg, error = %d\n", error);
    690 		goto fail_0;
    691 	}
    692 
    693 	/* Load the map for the TX ring. */
    694 	if ((error = bus_dmamem_map(sc->sc_dmat, &sc->re_ldata.re_tx_listseg,
    695 	    sc->re_ldata.re_tx_listnseg, RE_TX_LIST_SZ(sc),
    696 	    (void **)&sc->re_ldata.re_tx_list,
    697 	    BUS_DMA_COHERENT | BUS_DMA_NOWAIT)) != 0) {
    698 		aprint_error_dev(sc->sc_dev,
    699 		    "can't map tx list, error = %d\n", error);
    700 		goto fail_1;
    701 	}
    702 	memset(sc->re_ldata.re_tx_list, 0, RE_TX_LIST_SZ(sc));
    703 
    704 	if ((error = bus_dmamap_create(sc->sc_dmat, RE_TX_LIST_SZ(sc), 1,
    705 	    RE_TX_LIST_SZ(sc), 0, 0,
    706 	    &sc->re_ldata.re_tx_list_map)) != 0) {
    707 		aprint_error_dev(sc->sc_dev,
    708 		    "can't create tx list map, error = %d\n", error);
    709 		goto fail_2;
    710 	}
    711 
    712 
    713 	if ((error = bus_dmamap_load(sc->sc_dmat,
    714 	    sc->re_ldata.re_tx_list_map, sc->re_ldata.re_tx_list,
    715 	    RE_TX_LIST_SZ(sc), NULL, BUS_DMA_NOWAIT)) != 0) {
    716 		aprint_error_dev(sc->sc_dev,
    717 		    "can't load tx list, error = %d\n", error);
    718 		goto fail_3;
    719 	}
    720 
    721 	/* Create DMA maps for TX buffers */
    722 	for (i = 0; i < RE_TX_QLEN; i++) {
    723 		error = bus_dmamap_create(sc->sc_dmat,
    724 		    round_page(IP_MAXPACKET),
    725 		    RE_TX_DESC_CNT(sc), RE_TDESC_CMD_FRAGLEN,
    726 		    0, 0, &sc->re_ldata.re_txq[i].txq_dmamap);
    727 		if (error) {
    728 			aprint_error_dev(sc->sc_dev,
    729 			    "can't create DMA map for TX\n");
    730 			goto fail_4;
    731 		}
    732 	}
    733 
    734 	/* Allocate DMA'able memory for the RX ring */
    735 	/* XXX see also a comment about RE_RX_DMAMEM_SZ in rtl81x9var.h */
    736 	if ((error = bus_dmamem_alloc(sc->sc_dmat,
    737 	    RE_RX_DMAMEM_SZ, RE_RING_ALIGN, 0, &sc->re_ldata.re_rx_listseg, 1,
    738 	    &sc->re_ldata.re_rx_listnseg, BUS_DMA_NOWAIT)) != 0) {
    739 		aprint_error_dev(sc->sc_dev,
    740 		    "can't allocate rx listseg, error = %d\n", error);
    741 		goto fail_4;
    742 	}
    743 
    744 	/* Load the map for the RX ring. */
    745 	if ((error = bus_dmamem_map(sc->sc_dmat, &sc->re_ldata.re_rx_listseg,
    746 	    sc->re_ldata.re_rx_listnseg, RE_RX_DMAMEM_SZ,
    747 	    (void **)&sc->re_ldata.re_rx_list,
    748 	    BUS_DMA_COHERENT | BUS_DMA_NOWAIT)) != 0) {
    749 		aprint_error_dev(sc->sc_dev,
    750 		    "can't map rx list, error = %d\n", error);
    751 		goto fail_5;
    752 	}
    753 	memset(sc->re_ldata.re_rx_list, 0, RE_RX_DMAMEM_SZ);
    754 
    755 	if ((error = bus_dmamap_create(sc->sc_dmat,
    756 	    RE_RX_DMAMEM_SZ, 1, RE_RX_DMAMEM_SZ, 0, 0,
    757 	    &sc->re_ldata.re_rx_list_map)) != 0) {
    758 		aprint_error_dev(sc->sc_dev,
    759 		    "can't create rx list map, error = %d\n", error);
    760 		goto fail_6;
    761 	}
    762 
    763 	if ((error = bus_dmamap_load(sc->sc_dmat,
    764 	    sc->re_ldata.re_rx_list_map, sc->re_ldata.re_rx_list,
    765 	    RE_RX_DMAMEM_SZ, NULL, BUS_DMA_NOWAIT)) != 0) {
    766 		aprint_error_dev(sc->sc_dev,
    767 		    "can't load rx list, error = %d\n", error);
    768 		goto fail_7;
    769 	}
    770 
    771 	/* Create DMA maps for RX buffers */
    772 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
    773 		error = bus_dmamap_create(sc->sc_dmat, MCLBYTES, 1, MCLBYTES,
    774 		    0, 0, &sc->re_ldata.re_rxsoft[i].rxs_dmamap);
    775 		if (error) {
    776 			aprint_error_dev(sc->sc_dev,
    777 			    "can't create DMA map for RX\n");
    778 			goto fail_8;
    779 		}
    780 	}
    781 
    782 	/*
    783 	 * Record interface as attached. From here, we should not fail.
    784 	 */
    785 	sc->sc_flags |= RTK_ATTACHED;
    786 
    787 	ifp = &sc->ethercom.ec_if;
    788 	ifp->if_softc = sc;
    789 	strlcpy(ifp->if_xname, device_xname(sc->sc_dev), IFNAMSIZ);
    790 	ifp->if_mtu = ETHERMTU;
    791 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
    792 	ifp->if_ioctl = re_ioctl;
    793 	sc->ethercom.ec_capabilities |=
    794 	    ETHERCAP_VLAN_MTU | ETHERCAP_VLAN_HWTAGGING;
    795 	ifp->if_start = re_start;
    796 	ifp->if_stop = re_stop;
    797 
    798 	/*
    799 	 * IFCAP_CSUM_IPv4_Tx on re(4) is broken for small packets,
    800 	 * so we have a workaround to handle the bug by padding
    801 	 * such packets manually.
    802 	 */
    803 	ifp->if_capabilities |=
    804 	    IFCAP_CSUM_IPv4_Tx | IFCAP_CSUM_IPv4_Rx |
    805 	    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
    806 	    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx |
    807 	    IFCAP_TSOv4;
    808 
    809 	/*
    810 	 * XXX
    811 	 * Still have no idea how to make TSO work on 8168C, 8168CP,
    812 	 * 8102E, 8111C and 8111CP.
    813 	 */
    814 	if ((sc->sc_quirk & RTKQ_DESCV2) != 0)
    815 		ifp->if_capabilities &= ~IFCAP_TSOv4;
    816 
    817 	ifp->if_watchdog = re_watchdog;
    818 	ifp->if_init = re_init;
    819 	ifp->if_snd.ifq_maxlen = RE_IFQ_MAXLEN;
    820 	ifp->if_capenable = ifp->if_capabilities;
    821 	IFQ_SET_READY(&ifp->if_snd);
    822 
    823 	callout_init(&sc->rtk_tick_ch, 0);
    824 
    825 	/* Do MII setup */
    826 	sc->mii.mii_ifp = ifp;
    827 	sc->mii.mii_readreg = re_miibus_readreg;
    828 	sc->mii.mii_writereg = re_miibus_writereg;
    829 	sc->mii.mii_statchg = re_miibus_statchg;
    830 	sc->ethercom.ec_mii = &sc->mii;
    831 	ifmedia_init(&sc->mii.mii_media, IFM_IMASK, ether_mediachange,
    832 	    ether_mediastatus);
    833 	mii_attach(sc->sc_dev, &sc->mii, 0xffffffff, MII_PHY_ANY,
    834 	    MII_OFFSET_ANY, 0);
    835 	ifmedia_set(&sc->mii.mii_media, IFM_ETHER | IFM_AUTO);
    836 
    837 	/*
    838 	 * Call MI attach routine.
    839 	 */
    840 	if_attach(ifp);
    841 	ether_ifattach(ifp, eaddr);
    842 
    843 	if (pmf_device_register(sc->sc_dev, NULL, NULL))
    844 		pmf_class_network_register(sc->sc_dev, ifp);
    845 	else
    846 		aprint_error_dev(sc->sc_dev,
    847 		    "couldn't establish power handler\n");
    848 
    849 	return;
    850 
    851  fail_8:
    852 	/* Destroy DMA maps for RX buffers. */
    853 	for (i = 0; i < RE_RX_DESC_CNT; i++)
    854 		if (sc->re_ldata.re_rxsoft[i].rxs_dmamap != NULL)
    855 			bus_dmamap_destroy(sc->sc_dmat,
    856 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
    857 
    858 	/* Free DMA'able memory for the RX ring. */
    859 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
    860  fail_7:
    861 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
    862  fail_6:
    863 	bus_dmamem_unmap(sc->sc_dmat,
    864 	    (void *)sc->re_ldata.re_rx_list, RE_RX_DMAMEM_SZ);
    865  fail_5:
    866 	bus_dmamem_free(sc->sc_dmat,
    867 	    &sc->re_ldata.re_rx_listseg, sc->re_ldata.re_rx_listnseg);
    868 
    869  fail_4:
    870 	/* Destroy DMA maps for TX buffers. */
    871 	for (i = 0; i < RE_TX_QLEN; i++)
    872 		if (sc->re_ldata.re_txq[i].txq_dmamap != NULL)
    873 			bus_dmamap_destroy(sc->sc_dmat,
    874 			    sc->re_ldata.re_txq[i].txq_dmamap);
    875 
    876 	/* Free DMA'able memory for the TX ring. */
    877 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
    878  fail_3:
    879 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
    880  fail_2:
    881 	bus_dmamem_unmap(sc->sc_dmat,
    882 	    (void *)sc->re_ldata.re_tx_list, RE_TX_LIST_SZ(sc));
    883  fail_1:
    884 	bus_dmamem_free(sc->sc_dmat,
    885 	    &sc->re_ldata.re_tx_listseg, sc->re_ldata.re_tx_listnseg);
    886  fail_0:
    887 	return;
    888 }
    889 
    890 
    891 /*
    892  * re_activate:
    893  *     Handle device activation/deactivation requests.
    894  */
    895 int
    896 re_activate(device_t self, enum devact act)
    897 {
    898 	struct rtk_softc *sc = device_private(self);
    899 
    900 	switch (act) {
    901 	case DVACT_DEACTIVATE:
    902 		if_deactivate(&sc->ethercom.ec_if);
    903 		return 0;
    904 	default:
    905 		return EOPNOTSUPP;
    906 	}
    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 	/* we don't want to run again */
    968 	sc->sc_flags &= ~RTK_ATTACHED;
    969 
    970 	return 0;
    971 }
    972 
    973 /*
    974  * re_enable:
    975  *     Enable the RTL81X9 chip.
    976  */
    977 static int
    978 re_enable(struct rtk_softc *sc)
    979 {
    980 
    981 	if (RTK_IS_ENABLED(sc) == 0 && sc->sc_enable != NULL) {
    982 		if ((*sc->sc_enable)(sc) != 0) {
    983 			printf("%s: device enable failed\n",
    984 			    device_xname(sc->sc_dev));
    985 			return EIO;
    986 		}
    987 		sc->sc_flags |= RTK_ENABLED;
    988 	}
    989 	return 0;
    990 }
    991 
    992 /*
    993  * re_disable:
    994  *     Disable the RTL81X9 chip.
    995  */
    996 static void
    997 re_disable(struct rtk_softc *sc)
    998 {
    999 
   1000 	if (RTK_IS_ENABLED(sc) && sc->sc_disable != NULL) {
   1001 		(*sc->sc_disable)(sc);
   1002 		sc->sc_flags &= ~RTK_ENABLED;
   1003 	}
   1004 }
   1005 
   1006 static int
   1007 re_newbuf(struct rtk_softc *sc, int idx, struct mbuf *m)
   1008 {
   1009 	struct mbuf *n = NULL;
   1010 	bus_dmamap_t map;
   1011 	struct re_desc *d;
   1012 	struct re_rxsoft *rxs;
   1013 	uint32_t cmdstat;
   1014 	int error;
   1015 
   1016 	if (m == NULL) {
   1017 		MGETHDR(n, M_DONTWAIT, MT_DATA);
   1018 		if (n == NULL)
   1019 			return ENOBUFS;
   1020 
   1021 		MCLGET(n, M_DONTWAIT);
   1022 		if ((n->m_flags & M_EXT) == 0) {
   1023 			m_freem(n);
   1024 			return ENOBUFS;
   1025 		}
   1026 		m = n;
   1027 	} else
   1028 		m->m_data = m->m_ext.ext_buf;
   1029 
   1030 	/*
   1031 	 * Initialize mbuf length fields and fixup
   1032 	 * alignment so that the frame payload is
   1033 	 * longword aligned.
   1034 	 */
   1035 	m->m_len = m->m_pkthdr.len = MCLBYTES - RE_ETHER_ALIGN;
   1036 	m->m_data += RE_ETHER_ALIGN;
   1037 
   1038 	rxs = &sc->re_ldata.re_rxsoft[idx];
   1039 	map = rxs->rxs_dmamap;
   1040 	error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m,
   1041 	    BUS_DMA_READ|BUS_DMA_NOWAIT);
   1042 
   1043 	if (error)
   1044 		goto out;
   1045 
   1046 	bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
   1047 	    BUS_DMASYNC_PREREAD);
   1048 
   1049 	d = &sc->re_ldata.re_rx_list[idx];
   1050 #ifdef DIAGNOSTIC
   1051 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
   1052 	cmdstat = le32toh(d->re_cmdstat);
   1053 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD);
   1054 	if (cmdstat & RE_RDESC_STAT_OWN) {
   1055 		panic("%s: tried to map busy RX descriptor",
   1056 		    device_xname(sc->sc_dev));
   1057 	}
   1058 #endif
   1059 
   1060 	rxs->rxs_mbuf = m;
   1061 
   1062 	d->re_vlanctl = 0;
   1063 	cmdstat = map->dm_segs[0].ds_len;
   1064 	if (idx == (RE_RX_DESC_CNT - 1))
   1065 		cmdstat |= RE_RDESC_CMD_EOR;
   1066 	re_set_bufaddr(d, map->dm_segs[0].ds_addr);
   1067 	d->re_cmdstat = htole32(cmdstat);
   1068 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1069 	cmdstat |= RE_RDESC_CMD_OWN;
   1070 	d->re_cmdstat = htole32(cmdstat);
   1071 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1072 
   1073 	return 0;
   1074  out:
   1075 	if (n != NULL)
   1076 		m_freem(n);
   1077 	return ENOMEM;
   1078 }
   1079 
   1080 static int
   1081 re_tx_list_init(struct rtk_softc *sc)
   1082 {
   1083 	int i;
   1084 
   1085 	memset(sc->re_ldata.re_tx_list, 0, RE_TX_LIST_SZ(sc));
   1086 	for (i = 0; i < RE_TX_QLEN; i++) {
   1087 		sc->re_ldata.re_txq[i].txq_mbuf = NULL;
   1088 	}
   1089 
   1090 	bus_dmamap_sync(sc->sc_dmat,
   1091 	    sc->re_ldata.re_tx_list_map, 0,
   1092 	    sc->re_ldata.re_tx_list_map->dm_mapsize,
   1093 	    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1094 	sc->re_ldata.re_txq_prodidx = 0;
   1095 	sc->re_ldata.re_txq_considx = 0;
   1096 	sc->re_ldata.re_txq_free = RE_TX_QLEN;
   1097 	sc->re_ldata.re_tx_free = RE_TX_DESC_CNT(sc);
   1098 	sc->re_ldata.re_tx_nextfree = 0;
   1099 
   1100 	return 0;
   1101 }
   1102 
   1103 static int
   1104 re_rx_list_init(struct rtk_softc *sc)
   1105 {
   1106 	int i;
   1107 
   1108 	memset(sc->re_ldata.re_rx_list, 0, RE_RX_LIST_SZ);
   1109 
   1110 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
   1111 		if (re_newbuf(sc, i, NULL) == ENOBUFS)
   1112 			return ENOBUFS;
   1113 	}
   1114 
   1115 	sc->re_ldata.re_rx_prodidx = 0;
   1116 	sc->re_head = sc->re_tail = NULL;
   1117 
   1118 	return 0;
   1119 }
   1120 
   1121 /*
   1122  * RX handler for C+ and 8169. For the gigE chips, we support
   1123  * the reception of jumbo frames that have been fragmented
   1124  * across multiple 2K mbuf cluster buffers.
   1125  */
   1126 static void
   1127 re_rxeof(struct rtk_softc *sc)
   1128 {
   1129 	struct mbuf *m;
   1130 	struct ifnet *ifp;
   1131 	int i, total_len;
   1132 	struct re_desc *cur_rx;
   1133 	struct re_rxsoft *rxs;
   1134 	uint32_t rxstat, rxvlan;
   1135 
   1136 	ifp = &sc->ethercom.ec_if;
   1137 
   1138 	for (i = sc->re_ldata.re_rx_prodidx;; i = RE_NEXT_RX_DESC(sc, i)) {
   1139 		cur_rx = &sc->re_ldata.re_rx_list[i];
   1140 		RE_RXDESCSYNC(sc, i,
   1141 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
   1142 		rxstat = le32toh(cur_rx->re_cmdstat);
   1143 		rxvlan = le32toh(cur_rx->re_vlanctl);
   1144 		RE_RXDESCSYNC(sc, i, BUS_DMASYNC_PREREAD);
   1145 		if ((rxstat & RE_RDESC_STAT_OWN) != 0) {
   1146 			break;
   1147 		}
   1148 		total_len = rxstat & sc->re_rxlenmask;
   1149 		rxs = &sc->re_ldata.re_rxsoft[i];
   1150 		m = rxs->rxs_mbuf;
   1151 
   1152 		/* Invalidate the RX mbuf and unload its map */
   1153 
   1154 		bus_dmamap_sync(sc->sc_dmat,
   1155 		    rxs->rxs_dmamap, 0, rxs->rxs_dmamap->dm_mapsize,
   1156 		    BUS_DMASYNC_POSTREAD);
   1157 		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
   1158 
   1159 		if ((rxstat & RE_RDESC_STAT_EOF) == 0) {
   1160 			m->m_len = MCLBYTES - RE_ETHER_ALIGN;
   1161 			if (sc->re_head == NULL)
   1162 				sc->re_head = sc->re_tail = m;
   1163 			else {
   1164 				m->m_flags &= ~M_PKTHDR;
   1165 				sc->re_tail->m_next = m;
   1166 				sc->re_tail = m;
   1167 			}
   1168 			re_newbuf(sc, i, NULL);
   1169 			continue;
   1170 		}
   1171 
   1172 		/*
   1173 		 * NOTE: for the 8139C+, the frame length field
   1174 		 * is always 12 bits in size, but for the gigE chips,
   1175 		 * it is 13 bits (since the max RX frame length is 16K).
   1176 		 * Unfortunately, all 32 bits in the status word
   1177 		 * were already used, so to make room for the extra
   1178 		 * length bit, RealTek took out the 'frame alignment
   1179 		 * error' bit and shifted the other status bits
   1180 		 * over one slot. The OWN, EOR, FS and LS bits are
   1181 		 * still in the same places. We have already extracted
   1182 		 * the frame length and checked the OWN bit, so rather
   1183 		 * than using an alternate bit mapping, we shift the
   1184 		 * status bits one space to the right so we can evaluate
   1185 		 * them using the 8169 status as though it was in the
   1186 		 * same format as that of the 8139C+.
   1187 		 */
   1188 		if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0)
   1189 			rxstat >>= 1;
   1190 
   1191 		if (__predict_false((rxstat & RE_RDESC_STAT_RXERRSUM) != 0)) {
   1192 #ifdef RE_DEBUG
   1193 			printf("%s: RX error (rxstat = 0x%08x)",
   1194 			    device_xname(sc->sc_dev), rxstat);
   1195 			if (rxstat & RE_RDESC_STAT_FRALIGN)
   1196 				printf(", frame alignment error");
   1197 			if (rxstat & RE_RDESC_STAT_BUFOFLOW)
   1198 				printf(", out of buffer space");
   1199 			if (rxstat & RE_RDESC_STAT_FIFOOFLOW)
   1200 				printf(", FIFO overrun");
   1201 			if (rxstat & RE_RDESC_STAT_GIANT)
   1202 				printf(", giant packet");
   1203 			if (rxstat & RE_RDESC_STAT_RUNT)
   1204 				printf(", runt packet");
   1205 			if (rxstat & RE_RDESC_STAT_CRCERR)
   1206 				printf(", CRC error");
   1207 			printf("\n");
   1208 #endif
   1209 			ifp->if_ierrors++;
   1210 			/*
   1211 			 * If this is part of a multi-fragment packet,
   1212 			 * discard all the pieces.
   1213 			 */
   1214 			if (sc->re_head != NULL) {
   1215 				m_freem(sc->re_head);
   1216 				sc->re_head = sc->re_tail = NULL;
   1217 			}
   1218 			re_newbuf(sc, i, m);
   1219 			continue;
   1220 		}
   1221 
   1222 		/*
   1223 		 * If allocating a replacement mbuf fails,
   1224 		 * reload the current one.
   1225 		 */
   1226 
   1227 		if (__predict_false(re_newbuf(sc, i, NULL) != 0)) {
   1228 			ifp->if_ierrors++;
   1229 			if (sc->re_head != NULL) {
   1230 				m_freem(sc->re_head);
   1231 				sc->re_head = sc->re_tail = NULL;
   1232 			}
   1233 			re_newbuf(sc, i, m);
   1234 			continue;
   1235 		}
   1236 
   1237 		if (sc->re_head != NULL) {
   1238 			m->m_len = total_len % (MCLBYTES - RE_ETHER_ALIGN);
   1239 			/*
   1240 			 * Special case: if there's 4 bytes or less
   1241 			 * in this buffer, the mbuf can be discarded:
   1242 			 * the last 4 bytes is the CRC, which we don't
   1243 			 * care about anyway.
   1244 			 */
   1245 			if (m->m_len <= ETHER_CRC_LEN) {
   1246 				sc->re_tail->m_len -=
   1247 				    (ETHER_CRC_LEN - m->m_len);
   1248 				m_freem(m);
   1249 			} else {
   1250 				m->m_len -= ETHER_CRC_LEN;
   1251 				m->m_flags &= ~M_PKTHDR;
   1252 				sc->re_tail->m_next = m;
   1253 			}
   1254 			m = sc->re_head;
   1255 			sc->re_head = sc->re_tail = NULL;
   1256 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
   1257 		} else
   1258 			m->m_pkthdr.len = m->m_len =
   1259 			    (total_len - ETHER_CRC_LEN);
   1260 
   1261 		ifp->if_ipackets++;
   1262 		m->m_pkthdr.rcvif = ifp;
   1263 
   1264 		/* Do RX checksumming */
   1265 		if ((sc->sc_quirk & RTKQ_DESCV2) == 0) {
   1266 			/* Check IP header checksum */
   1267 			if ((rxstat & RE_RDESC_STAT_PROTOID) != 0) {
   1268 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
   1269 				if (rxstat & RE_RDESC_STAT_IPSUMBAD)
   1270 					m->m_pkthdr.csum_flags |=
   1271 					    M_CSUM_IPv4_BAD;
   1272 
   1273 				/* Check TCP/UDP checksum */
   1274 				if (RE_TCPPKT(rxstat)) {
   1275 					m->m_pkthdr.csum_flags |= M_CSUM_TCPv4;
   1276 					if (rxstat & RE_RDESC_STAT_TCPSUMBAD)
   1277 						m->m_pkthdr.csum_flags |=
   1278 						    M_CSUM_TCP_UDP_BAD;
   1279 				} else if (RE_UDPPKT(rxstat)) {
   1280 					m->m_pkthdr.csum_flags |= M_CSUM_UDPv4;
   1281 					if (rxstat & RE_RDESC_STAT_UDPSUMBAD)
   1282 						m->m_pkthdr.csum_flags |=
   1283 						    M_CSUM_TCP_UDP_BAD;
   1284 				}
   1285 			}
   1286 		} else {
   1287 			/* Check IPv4 header checksum */
   1288 			if ((rxvlan & RE_RDESC_VLANCTL_IPV4) != 0) {
   1289 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
   1290 				if (rxstat & RE_RDESC_STAT_IPSUMBAD)
   1291 					m->m_pkthdr.csum_flags |=
   1292 					    M_CSUM_IPv4_BAD;
   1293 
   1294 				/* Check TCPv4/UDPv4 checksum */
   1295 				if (RE_TCPPKT(rxstat)) {
   1296 					m->m_pkthdr.csum_flags |= M_CSUM_TCPv4;
   1297 					if (rxstat & RE_RDESC_STAT_TCPSUMBAD)
   1298 						m->m_pkthdr.csum_flags |=
   1299 						    M_CSUM_TCP_UDP_BAD;
   1300 				} else if (RE_UDPPKT(rxstat)) {
   1301 					m->m_pkthdr.csum_flags |= M_CSUM_UDPv4;
   1302 					if (rxstat & RE_RDESC_STAT_UDPSUMBAD)
   1303 						m->m_pkthdr.csum_flags |=
   1304 						    M_CSUM_TCP_UDP_BAD;
   1305 				}
   1306 			}
   1307 			/* XXX Check TCPv6/UDPv6 checksum? */
   1308 		}
   1309 
   1310 		if (rxvlan & RE_RDESC_VLANCTL_TAG) {
   1311 			VLAN_INPUT_TAG(ifp, m,
   1312 			     bswap16(rxvlan & RE_RDESC_VLANCTL_DATA),
   1313 			     continue);
   1314 		}
   1315 		bpf_mtap(ifp, m);
   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 		/*
   1680 		 * If there's a BPF listener, bounce a copy of this frame
   1681 		 * to him.
   1682 		 */
   1683 		bpf_mtap(ifp, m);
   1684 	}
   1685 
   1686 	if (sc->re_ldata.re_txq_free < ofree) {
   1687 		/*
   1688 		 * TX packets are enqueued.
   1689 		 */
   1690 		sc->re_ldata.re_txq_prodidx = idx;
   1691 
   1692 		/*
   1693 		 * Start the transmitter to poll.
   1694 		 *
   1695 		 * RealTek put the TX poll request register in a different
   1696 		 * location on the 8169 gigE chip. I don't know why.
   1697 		 */
   1698 		if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0)
   1699 			CSR_WRITE_1(sc, RTK_TXSTART, RTK_TXSTART_START);
   1700 		else
   1701 			CSR_WRITE_1(sc, RTK_GTXSTART, RTK_TXSTART_START);
   1702 
   1703 		/*
   1704 		 * Use the countdown timer for interrupt moderation.
   1705 		 * 'TX done' interrupts are disabled. Instead, we reset the
   1706 		 * countdown timer, which will begin counting until it hits
   1707 		 * the value in the TIMERINT register, and then trigger an
   1708 		 * interrupt. Each time we write to the TIMERCNT register,
   1709 		 * the timer count is reset to 0.
   1710 		 */
   1711 		CSR_WRITE_4(sc, RTK_TIMERCNT, 1);
   1712 
   1713 		/*
   1714 		 * Set a timeout in case the chip goes out to lunch.
   1715 		 */
   1716 		ifp->if_timer = 5;
   1717 	}
   1718 }
   1719 
   1720 static int
   1721 re_init(struct ifnet *ifp)
   1722 {
   1723 	struct rtk_softc *sc = ifp->if_softc;
   1724 	const uint8_t *enaddr;
   1725 	uint32_t rxcfg = 0;
   1726 	uint32_t reg;
   1727 	uint16_t cfg;
   1728 	int error;
   1729 
   1730 	if ((error = re_enable(sc)) != 0)
   1731 		goto out;
   1732 
   1733 	/*
   1734 	 * Cancel pending I/O and free all RX/TX buffers.
   1735 	 */
   1736 	re_stop(ifp, 0);
   1737 
   1738 	re_reset(sc);
   1739 
   1740 	/*
   1741 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
   1742 	 * RX checksum offload. We must configure the C+ register
   1743 	 * before all others.
   1744 	 */
   1745 	cfg = RE_CPLUSCMD_PCI_MRW;
   1746 
   1747 	/*
   1748 	 * XXX: For old 8169 set bit 14.
   1749 	 *      For 8169S/8110S and above, do not set bit 14.
   1750 	 */
   1751 	if ((sc->sc_quirk & RTKQ_8169NONS) != 0)
   1752 		cfg |= (0x1 << 14);
   1753 
   1754 	if ((sc->ethercom.ec_capenable & ETHERCAP_VLAN_HWTAGGING) != 0)
   1755 		cfg |= RE_CPLUSCMD_VLANSTRIP;
   1756 	if ((ifp->if_capenable & (IFCAP_CSUM_IPv4_Rx |
   1757 	     IFCAP_CSUM_TCPv4_Rx | IFCAP_CSUM_UDPv4_Rx)) != 0)
   1758 		cfg |= RE_CPLUSCMD_RXCSUM_ENB;
   1759 	if ((sc->sc_quirk & RTKQ_MACSTAT) != 0) {
   1760 		cfg |= RE_CPLUSCMD_MACSTAT_DIS;
   1761 		cfg |= RE_CPLUSCMD_TXENB;
   1762 	} else
   1763 		cfg |= RE_CPLUSCMD_RXENB | RE_CPLUSCMD_TXENB;
   1764 
   1765 	CSR_WRITE_2(sc, RTK_CPLUS_CMD, cfg);
   1766 
   1767 	/* XXX: from Realtek-supplied Linux driver. Wholly undocumented. */
   1768 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0)
   1769 		CSR_WRITE_2(sc, RTK_IM, 0x0000);
   1770 
   1771 	DELAY(10000);
   1772 
   1773 	/*
   1774 	 * Init our MAC address.  Even though the chipset
   1775 	 * documentation doesn't mention it, we need to enter "Config
   1776 	 * register write enable" mode to modify the ID registers.
   1777 	 */
   1778 	CSR_WRITE_1(sc, RTK_EECMD, RTK_EEMODE_WRITECFG);
   1779 	enaddr = CLLADDR(ifp->if_sadl);
   1780 	reg = enaddr[0] | (enaddr[1] << 8) |
   1781 	    (enaddr[2] << 16) | (enaddr[3] << 24);
   1782 	CSR_WRITE_4(sc, RTK_IDR0, reg);
   1783 	reg = enaddr[4] | (enaddr[5] << 8);
   1784 	CSR_WRITE_4(sc, RTK_IDR4, reg);
   1785 	CSR_WRITE_1(sc, RTK_EECMD, RTK_EEMODE_OFF);
   1786 
   1787 	/*
   1788 	 * For C+ mode, initialize the RX descriptors and mbufs.
   1789 	 */
   1790 	re_rx_list_init(sc);
   1791 	re_tx_list_init(sc);
   1792 
   1793 	/*
   1794 	 * Load the addresses of the RX and TX lists into the chip.
   1795 	 */
   1796 	CSR_WRITE_4(sc, RTK_RXLIST_ADDR_HI,
   1797 	    RE_ADDR_HI(sc->re_ldata.re_rx_list_map->dm_segs[0].ds_addr));
   1798 	CSR_WRITE_4(sc, RTK_RXLIST_ADDR_LO,
   1799 	    RE_ADDR_LO(sc->re_ldata.re_rx_list_map->dm_segs[0].ds_addr));
   1800 
   1801 	CSR_WRITE_4(sc, RTK_TXLIST_ADDR_HI,
   1802 	    RE_ADDR_HI(sc->re_ldata.re_tx_list_map->dm_segs[0].ds_addr));
   1803 	CSR_WRITE_4(sc, RTK_TXLIST_ADDR_LO,
   1804 	    RE_ADDR_LO(sc->re_ldata.re_tx_list_map->dm_segs[0].ds_addr));
   1805 
   1806 	/*
   1807 	 * Enable transmit and receive.
   1808 	 */
   1809 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_TX_ENB | RTK_CMD_RX_ENB);
   1810 
   1811 	/*
   1812 	 * Set the initial TX and RX configuration.
   1813 	 */
   1814 	if (sc->re_testmode && (sc->sc_quirk & RTKQ_8169NONS) != 0) {
   1815 		/* test mode is needed only for old 8169 */
   1816 		CSR_WRITE_4(sc, RTK_TXCFG,
   1817 		    RE_TXCFG_CONFIG | RTK_LOOPTEST_ON);
   1818 	} else
   1819 		CSR_WRITE_4(sc, RTK_TXCFG, RE_TXCFG_CONFIG);
   1820 
   1821 	CSR_WRITE_1(sc, RTK_EARLY_TX_THRESH, 16);
   1822 
   1823 	CSR_WRITE_4(sc, RTK_RXCFG, RE_RXCFG_CONFIG);
   1824 
   1825 	/* Set the individual bit to receive frames for this host only. */
   1826 	rxcfg = CSR_READ_4(sc, RTK_RXCFG);
   1827 	rxcfg |= RTK_RXCFG_RX_INDIV;
   1828 
   1829 	/* If we want promiscuous mode, set the allframes bit. */
   1830 	if (ifp->if_flags & IFF_PROMISC)
   1831 		rxcfg |= RTK_RXCFG_RX_ALLPHYS;
   1832 	else
   1833 		rxcfg &= ~RTK_RXCFG_RX_ALLPHYS;
   1834 	CSR_WRITE_4(sc, RTK_RXCFG, rxcfg);
   1835 
   1836 	/*
   1837 	 * Set capture broadcast bit to capture broadcast frames.
   1838 	 */
   1839 	if (ifp->if_flags & IFF_BROADCAST)
   1840 		rxcfg |= RTK_RXCFG_RX_BROAD;
   1841 	else
   1842 		rxcfg &= ~RTK_RXCFG_RX_BROAD;
   1843 	CSR_WRITE_4(sc, RTK_RXCFG, rxcfg);
   1844 
   1845 	/*
   1846 	 * Program the multicast filter, if necessary.
   1847 	 */
   1848 	rtk_setmulti(sc);
   1849 
   1850 	/*
   1851 	 * Enable interrupts.
   1852 	 */
   1853 	if (sc->re_testmode)
   1854 		CSR_WRITE_2(sc, RTK_IMR, 0);
   1855 	else
   1856 		CSR_WRITE_2(sc, RTK_IMR, RTK_INTRS_CPLUS);
   1857 
   1858 	/* Start RX/TX process. */
   1859 	CSR_WRITE_4(sc, RTK_MISSEDPKT, 0);
   1860 #ifdef notdef
   1861 	/* Enable receiver and transmitter. */
   1862 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_TX_ENB | RTK_CMD_RX_ENB);
   1863 #endif
   1864 
   1865 	/*
   1866 	 * Initialize the timer interrupt register so that
   1867 	 * a timer interrupt will be generated once the timer
   1868 	 * reaches a certain number of ticks. The timer is
   1869 	 * reloaded on each transmit. This gives us TX interrupt
   1870 	 * moderation, which dramatically improves TX frame rate.
   1871 	 */
   1872 
   1873 	if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0)
   1874 		CSR_WRITE_4(sc, RTK_TIMERINT, 0x400);
   1875 	else {
   1876 		CSR_WRITE_4(sc, RTK_TIMERINT_8169, 0x800);
   1877 
   1878 		/*
   1879 		 * For 8169 gigE NICs, set the max allowed RX packet
   1880 		 * size so we can receive jumbo frames.
   1881 		 */
   1882 		CSR_WRITE_2(sc, RTK_MAXRXPKTLEN, 16383);
   1883 	}
   1884 
   1885 	if (sc->re_testmode)
   1886 		return 0;
   1887 
   1888 	CSR_WRITE_1(sc, RTK_CFG1, RTK_CFG1_DRVLOAD);
   1889 
   1890 	ifp->if_flags |= IFF_RUNNING;
   1891 	ifp->if_flags &= ~IFF_OACTIVE;
   1892 
   1893 	callout_reset(&sc->rtk_tick_ch, hz, re_tick, sc);
   1894 
   1895  out:
   1896 	if (error) {
   1897 		ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
   1898 		ifp->if_timer = 0;
   1899 		printf("%s: interface not running\n",
   1900 		    device_xname(sc->sc_dev));
   1901 	}
   1902 
   1903 	return error;
   1904 }
   1905 
   1906 static int
   1907 re_ioctl(struct ifnet *ifp, u_long command, void *data)
   1908 {
   1909 	struct rtk_softc *sc = ifp->if_softc;
   1910 	struct ifreq *ifr = data;
   1911 	int s, error = 0;
   1912 
   1913 	s = splnet();
   1914 
   1915 	switch (command) {
   1916 	case SIOCSIFMTU:
   1917 		/*
   1918 		 * Disable jumbo frames if it's not supported.
   1919 		 */
   1920 		if ((sc->sc_quirk & RTKQ_NOJUMBO) != 0 &&
   1921 		    ifr->ifr_mtu > ETHERMTU) {
   1922 			error = EINVAL;
   1923 			break;
   1924 		}
   1925 
   1926 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > ETHERMTU_JUMBO)
   1927 			error = EINVAL;
   1928 		else if ((error = ifioctl_common(ifp, command, data)) ==
   1929 		    ENETRESET)
   1930 			error = 0;
   1931 		break;
   1932 	default:
   1933 		if ((error = ether_ioctl(ifp, command, data)) != ENETRESET)
   1934 			break;
   1935 
   1936 		error = 0;
   1937 
   1938 		if (command == SIOCSIFCAP)
   1939 			error = (*ifp->if_init)(ifp);
   1940 		else if (command != SIOCADDMULTI && command != SIOCDELMULTI)
   1941 			;
   1942 		else if (ifp->if_flags & IFF_RUNNING)
   1943 			rtk_setmulti(sc);
   1944 		break;
   1945 	}
   1946 
   1947 	splx(s);
   1948 
   1949 	return error;
   1950 }
   1951 
   1952 static void
   1953 re_watchdog(struct ifnet *ifp)
   1954 {
   1955 	struct rtk_softc *sc;
   1956 	int s;
   1957 
   1958 	sc = ifp->if_softc;
   1959 	s = splnet();
   1960 	printf("%s: watchdog timeout\n", device_xname(sc->sc_dev));
   1961 	ifp->if_oerrors++;
   1962 
   1963 	re_txeof(sc);
   1964 	re_rxeof(sc);
   1965 
   1966 	re_init(ifp);
   1967 
   1968 	splx(s);
   1969 }
   1970 
   1971 /*
   1972  * Stop the adapter and free any mbufs allocated to the
   1973  * RX and TX lists.
   1974  */
   1975 static void
   1976 re_stop(struct ifnet *ifp, int disable)
   1977 {
   1978 	int i;
   1979 	struct rtk_softc *sc = ifp->if_softc;
   1980 
   1981 	callout_stop(&sc->rtk_tick_ch);
   1982 
   1983 	mii_down(&sc->mii);
   1984 
   1985 	if ((sc->sc_quirk & RTKQ_CMDSTOP) != 0)
   1986 		CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_STOPREQ | RTK_CMD_TX_ENB |
   1987 		    RTK_CMD_RX_ENB);
   1988 	else
   1989 		CSR_WRITE_1(sc, RTK_COMMAND, 0x00);
   1990 	DELAY(1000);
   1991 	CSR_WRITE_2(sc, RTK_IMR, 0x0000);
   1992 	CSR_WRITE_2(sc, RTK_ISR, 0xFFFF);
   1993 
   1994 	if (sc->re_head != NULL) {
   1995 		m_freem(sc->re_head);
   1996 		sc->re_head = sc->re_tail = NULL;
   1997 	}
   1998 
   1999 	/* Free the TX list buffers. */
   2000 	for (i = 0; i < RE_TX_QLEN; i++) {
   2001 		if (sc->re_ldata.re_txq[i].txq_mbuf != NULL) {
   2002 			bus_dmamap_unload(sc->sc_dmat,
   2003 			    sc->re_ldata.re_txq[i].txq_dmamap);
   2004 			m_freem(sc->re_ldata.re_txq[i].txq_mbuf);
   2005 			sc->re_ldata.re_txq[i].txq_mbuf = NULL;
   2006 		}
   2007 	}
   2008 
   2009 	/* Free the RX list buffers. */
   2010 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
   2011 		if (sc->re_ldata.re_rxsoft[i].rxs_mbuf != NULL) {
   2012 			bus_dmamap_unload(sc->sc_dmat,
   2013 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
   2014 			m_freem(sc->re_ldata.re_rxsoft[i].rxs_mbuf);
   2015 			sc->re_ldata.re_rxsoft[i].rxs_mbuf = NULL;
   2016 		}
   2017 	}
   2018 
   2019 	if (disable)
   2020 		re_disable(sc);
   2021 
   2022 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
   2023 	ifp->if_timer = 0;
   2024 }
   2025