Home | History | Annotate | Line # | Download | only in ic
rtl8169.c revision 1.115
      1 /*	$NetBSD: rtl8169.c,v 1.115 2009/04/13 12:33:05 tsutsui Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 1997, 1998-2003
      5  *	Bill Paul <wpaul (at) windriver.com>.  All rights reserved.
      6  *
      7  * Redistribution and use in source and binary forms, with or without
      8  * modification, are permitted provided that the following conditions
      9  * are met:
     10  * 1. Redistributions of source code must retain the above copyright
     11  *    notice, this list of conditions and the following disclaimer.
     12  * 2. Redistributions in binary form must reproduce the above copyright
     13  *    notice, this list of conditions and the following disclaimer in the
     14  *    documentation and/or other materials provided with the distribution.
     15  * 3. All advertising materials mentioning features or use of this software
     16  *    must display the following acknowledgement:
     17  *	This product includes software developed by Bill Paul.
     18  * 4. Neither the name of the author nor the names of any co-contributors
     19  *    may be used to endorse or promote products derived from this software
     20  *    without specific prior written permission.
     21  *
     22  * THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
     23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     25  * ARE DISCLAIMED.  IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
     26  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
     27  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
     28  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
     29  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
     30  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
     31  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
     32  * THE POSSIBILITY OF SUCH DAMAGE.
     33  */
     34 
     35 #include <sys/cdefs.h>
     36 __KERNEL_RCSID(0, "$NetBSD: rtl8169.c,v 1.115 2009/04/13 12:33:05 tsutsui Exp $");
     37 /* $FreeBSD: /repoman/r/ncvs/src/sys/dev/re/if_re.c,v 1.20 2004/04/11 20:34:08 ru Exp $ */
     38 
     39 /*
     40  * RealTek 8139C+/8169/8169S/8110S PCI NIC driver
     41  *
     42  * Written by Bill Paul <wpaul (at) windriver.com>
     43  * Senior Networking Software Engineer
     44  * Wind River Systems
     45  */
     46 
     47 /*
     48  * This driver is designed to support RealTek's next generation of
     49  * 10/100 and 10/100/1000 PCI ethernet controllers. There are currently
     50  * four devices in this family: the RTL8139C+, the RTL8169, the RTL8169S
     51  * and the RTL8110S.
     52  *
     53  * The 8139C+ is a 10/100 ethernet chip. It is backwards compatible
     54  * with the older 8139 family, however it also supports a special
     55  * C+ mode of operation that provides several new performance enhancing
     56  * features. These include:
     57  *
     58  *	o Descriptor based DMA mechanism. Each descriptor represents
     59  *	  a single packet fragment. Data buffers may be aligned on
     60  *	  any byte boundary.
     61  *
     62  *	o 64-bit DMA
     63  *
     64  *	o TCP/IP checksum offload for both RX and TX
     65  *
     66  *	o High and normal priority transmit DMA rings
     67  *
     68  *	o VLAN tag insertion and extraction
     69  *
     70  *	o TCP large send (segmentation offload)
     71  *
     72  * Like the 8139, the 8139C+ also has a built-in 10/100 PHY. The C+
     73  * programming API is fairly straightforward. The RX filtering, EEPROM
     74  * access and PHY access is the same as it is on the older 8139 series
     75  * chips.
     76  *
     77  * The 8169 is a 64-bit 10/100/1000 gigabit ethernet MAC. It has almost the
     78  * same programming API and feature set as the 8139C+ with the following
     79  * differences and additions:
     80  *
     81  *	o 1000Mbps mode
     82  *
     83  *	o Jumbo frames
     84  *
     85  * 	o GMII and TBI ports/registers for interfacing with copper
     86  *	  or fiber PHYs
     87  *
     88  *      o RX and TX DMA rings can have up to 1024 descriptors
     89  *        (the 8139C+ allows a maximum of 64)
     90  *
     91  *	o Slight differences in register layout from the 8139C+
     92  *
     93  * The TX start and timer interrupt registers are at different locations
     94  * on the 8169 than they are on the 8139C+. Also, the status word in the
     95  * RX descriptor has a slightly different bit layout. The 8169 does not
     96  * have a built-in PHY. Most reference boards use a Marvell 88E1000 'Alaska'
     97  * copper gigE PHY.
     98  *
     99  * The 8169S/8110S 10/100/1000 devices have built-in copper gigE PHYs
    100  * (the 'S' stands for 'single-chip'). These devices have the same
    101  * programming API as the older 8169, but also have some vendor-specific
    102  * registers for the on-board PHY. The 8110S is a LAN-on-motherboard
    103  * part designed to be pin-compatible with the RealTek 8100 10/100 chip.
    104  *
    105  * This driver takes advantage of the RX and TX checksum offload and
    106  * VLAN tag insertion/extraction features. It also implements TX
    107  * interrupt moderation using the timer interrupt registers, which
    108  * significantly reduces TX interrupt load. There is also support
    109  * for jumbo frames, however the 8169/8169S/8110S can not transmit
    110  * jumbo frames larger than 7.5K, so the max MTU possible with this
    111  * driver is 7500 bytes.
    112  */
    113 
    114 #include "bpfilter.h"
    115 #include "vlan.h"
    116 
    117 #include <sys/param.h>
    118 #include <sys/endian.h>
    119 #include <sys/systm.h>
    120 #include <sys/sockio.h>
    121 #include <sys/mbuf.h>
    122 #include <sys/malloc.h>
    123 #include <sys/kernel.h>
    124 #include <sys/socket.h>
    125 #include <sys/device.h>
    126 
    127 #include <net/if.h>
    128 #include <net/if_arp.h>
    129 #include <net/if_dl.h>
    130 #include <net/if_ether.h>
    131 #include <net/if_media.h>
    132 #include <net/if_vlanvar.h>
    133 
    134 #include <netinet/in_systm.h>	/* XXX for IP_MAXPACKET */
    135 #include <netinet/in.h>		/* XXX for IP_MAXPACKET */
    136 #include <netinet/ip.h>		/* XXX for IP_MAXPACKET */
    137 
    138 #if NBPFILTER > 0
    139 #include <net/bpf.h>
    140 #endif
    141 
    142 #include <sys/bus.h>
    143 
    144 #include <dev/mii/mii.h>
    145 #include <dev/mii/miivar.h>
    146 
    147 #include <dev/ic/rtl81x9reg.h>
    148 #include <dev/ic/rtl81x9var.h>
    149 
    150 #include <dev/ic/rtl8169var.h>
    151 
    152 static inline void re_set_bufaddr(struct re_desc *, bus_addr_t);
    153 
    154 static int re_newbuf(struct rtk_softc *, int, struct mbuf *);
    155 static int re_rx_list_init(struct rtk_softc *);
    156 static int re_tx_list_init(struct rtk_softc *);
    157 static void re_rxeof(struct rtk_softc *);
    158 static void re_txeof(struct rtk_softc *);
    159 static void re_tick(void *);
    160 static void re_start(struct ifnet *);
    161 static int re_ioctl(struct ifnet *, u_long, void *);
    162 static int re_init(struct ifnet *);
    163 static void re_stop(struct ifnet *, int);
    164 static void re_watchdog(struct ifnet *);
    165 
    166 static int re_enable(struct rtk_softc *);
    167 static void re_disable(struct rtk_softc *);
    168 
    169 static int re_gmii_readreg(struct device *, int, int);
    170 static void re_gmii_writereg(struct device *, int, int, int);
    171 
    172 static int re_miibus_readreg(struct device *, int, int);
    173 static void re_miibus_writereg(struct device *, int, int, int);
    174 static void re_miibus_statchg(struct device *);
    175 
    176 static void re_reset(struct rtk_softc *);
    177 
    178 static inline void
    179 re_set_bufaddr(struct re_desc *d, bus_addr_t addr)
    180 {
    181 
    182 	d->re_bufaddr_lo = htole32((uint32_t)addr);
    183 	if (sizeof(bus_addr_t) == sizeof(uint64_t))
    184 		d->re_bufaddr_hi = htole32((uint64_t)addr >> 32);
    185 	else
    186 		d->re_bufaddr_hi = 0;
    187 }
    188 
    189 static int
    190 re_gmii_readreg(device_t dev, int phy, int reg)
    191 {
    192 	struct rtk_softc *sc = device_private(dev);
    193 	uint32_t rval;
    194 	int i;
    195 
    196 	if (phy != 7)
    197 		return 0;
    198 
    199 	/* Let the rgephy driver read the GMEDIASTAT register */
    200 
    201 	if (reg == RTK_GMEDIASTAT) {
    202 		rval = CSR_READ_1(sc, RTK_GMEDIASTAT);
    203 		return rval;
    204 	}
    205 
    206 	CSR_WRITE_4(sc, RTK_PHYAR, reg << 16);
    207 	DELAY(1000);
    208 
    209 	for (i = 0; i < RTK_TIMEOUT; i++) {
    210 		rval = CSR_READ_4(sc, RTK_PHYAR);
    211 		if (rval & RTK_PHYAR_BUSY)
    212 			break;
    213 		DELAY(100);
    214 	}
    215 
    216 	if (i == RTK_TIMEOUT) {
    217 		printf("%s: PHY read failed\n", device_xname(sc->sc_dev));
    218 		return 0;
    219 	}
    220 
    221 	return rval & RTK_PHYAR_PHYDATA;
    222 }
    223 
    224 static void
    225 re_gmii_writereg(device_t dev, int phy, int reg, int data)
    226 {
    227 	struct rtk_softc *sc = device_private(dev);
    228 	uint32_t rval;
    229 	int i;
    230 
    231 	CSR_WRITE_4(sc, RTK_PHYAR, (reg << 16) |
    232 	    (data & RTK_PHYAR_PHYDATA) | RTK_PHYAR_BUSY);
    233 	DELAY(1000);
    234 
    235 	for (i = 0; i < RTK_TIMEOUT; i++) {
    236 		rval = CSR_READ_4(sc, RTK_PHYAR);
    237 		if (!(rval & RTK_PHYAR_BUSY))
    238 			break;
    239 		DELAY(100);
    240 	}
    241 
    242 	if (i == RTK_TIMEOUT) {
    243 		printf("%s: PHY write reg %x <- %x failed\n",
    244 		    device_xname(sc->sc_dev), reg, data);
    245 	}
    246 }
    247 
    248 static int
    249 re_miibus_readreg(device_t dev, int phy, int reg)
    250 {
    251 	struct rtk_softc *sc = device_private(dev);
    252 	uint16_t rval = 0;
    253 	uint16_t re8139_reg = 0;
    254 	int s;
    255 
    256 	s = splnet();
    257 
    258 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0) {
    259 		rval = re_gmii_readreg(dev, phy, reg);
    260 		splx(s);
    261 		return rval;
    262 	}
    263 
    264 	/* Pretend the internal PHY is only at address 0 */
    265 	if (phy) {
    266 		splx(s);
    267 		return 0;
    268 	}
    269 	switch (reg) {
    270 	case MII_BMCR:
    271 		re8139_reg = RTK_BMCR;
    272 		break;
    273 	case MII_BMSR:
    274 		re8139_reg = RTK_BMSR;
    275 		break;
    276 	case MII_ANAR:
    277 		re8139_reg = RTK_ANAR;
    278 		break;
    279 	case MII_ANER:
    280 		re8139_reg = RTK_ANER;
    281 		break;
    282 	case MII_ANLPAR:
    283 		re8139_reg = RTK_LPAR;
    284 		break;
    285 	case MII_PHYIDR1:
    286 	case MII_PHYIDR2:
    287 		splx(s);
    288 		return 0;
    289 	/*
    290 	 * Allow the rlphy driver to read the media status
    291 	 * register. If we have a link partner which does not
    292 	 * support NWAY, this is the register which will tell
    293 	 * us the results of parallel detection.
    294 	 */
    295 	case RTK_MEDIASTAT:
    296 		rval = CSR_READ_1(sc, RTK_MEDIASTAT);
    297 		splx(s);
    298 		return rval;
    299 	default:
    300 		printf("%s: bad phy register\n", device_xname(sc->sc_dev));
    301 		splx(s);
    302 		return 0;
    303 	}
    304 	rval = CSR_READ_2(sc, re8139_reg);
    305 	if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0 && re8139_reg == RTK_BMCR) {
    306 		/* 8139C+ has different bit layout. */
    307 		rval &= ~(BMCR_LOOP | BMCR_ISO);
    308 	}
    309 	splx(s);
    310 	return rval;
    311 }
    312 
    313 static void
    314 re_miibus_writereg(device_t dev, int phy, int reg, int data)
    315 {
    316 	struct rtk_softc *sc = device_private(dev);
    317 	uint16_t re8139_reg = 0;
    318 	int s;
    319 
    320 	s = splnet();
    321 
    322 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0) {
    323 		re_gmii_writereg(dev, phy, reg, data);
    324 		splx(s);
    325 		return;
    326 	}
    327 
    328 	/* Pretend the internal PHY is only at address 0 */
    329 	if (phy) {
    330 		splx(s);
    331 		return;
    332 	}
    333 	switch (reg) {
    334 	case MII_BMCR:
    335 		re8139_reg = RTK_BMCR;
    336 		if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0) {
    337 			/* 8139C+ has different bit layout. */
    338 			data &= ~(BMCR_LOOP | BMCR_ISO);
    339 		}
    340 		break;
    341 	case MII_BMSR:
    342 		re8139_reg = RTK_BMSR;
    343 		break;
    344 	case MII_ANAR:
    345 		re8139_reg = RTK_ANAR;
    346 		break;
    347 	case MII_ANER:
    348 		re8139_reg = RTK_ANER;
    349 		break;
    350 	case MII_ANLPAR:
    351 		re8139_reg = RTK_LPAR;
    352 		break;
    353 	case MII_PHYIDR1:
    354 	case MII_PHYIDR2:
    355 		splx(s);
    356 		return;
    357 		break;
    358 	default:
    359 		printf("%s: bad phy register\n", device_xname(sc->sc_dev));
    360 		splx(s);
    361 		return;
    362 	}
    363 	CSR_WRITE_2(sc, re8139_reg, data);
    364 	splx(s);
    365 	return;
    366 }
    367 
    368 static void
    369 re_miibus_statchg(device_t dev)
    370 {
    371 
    372 	return;
    373 }
    374 
    375 static void
    376 re_reset(struct rtk_softc *sc)
    377 {
    378 	int i;
    379 
    380 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_RESET);
    381 
    382 	for (i = 0; i < RTK_TIMEOUT; i++) {
    383 		DELAY(10);
    384 		if ((CSR_READ_1(sc, RTK_COMMAND) & RTK_CMD_RESET) == 0)
    385 			break;
    386 	}
    387 	if (i == RTK_TIMEOUT)
    388 		printf("%s: reset never completed!\n",
    389 		    device_xname(sc->sc_dev));
    390 
    391 	/*
    392 	 * NB: Realtek-supplied FreeBSD driver does this only for MACFG_3,
    393 	 *     but also says "Rtl8169s sigle chip detected".
    394 	 */
    395 	if ((sc->sc_quirk & RTKQ_MACLDPS) != 0)
    396 		CSR_WRITE_1(sc, RTK_LDPS, 1);
    397 
    398 }
    399 
    400 /*
    401  * The following routine is designed to test for a defect on some
    402  * 32-bit 8169 cards. Some of these NICs have the REQ64# and ACK64#
    403  * lines connected to the bus, however for a 32-bit only card, they
    404  * should be pulled high. The result of this defect is that the
    405  * NIC will not work right if you plug it into a 64-bit slot: DMA
    406  * operations will be done with 64-bit transfers, which will fail
    407  * because the 64-bit data lines aren't connected.
    408  *
    409  * There's no way to work around this (short of talking a soldering
    410  * iron to the board), however we can detect it. The method we use
    411  * here is to put the NIC into digital loopback mode, set the receiver
    412  * to promiscuous mode, and then try to send a frame. We then compare
    413  * the frame data we sent to what was received. If the data matches,
    414  * then the NIC is working correctly, otherwise we know the user has
    415  * a defective NIC which has been mistakenly plugged into a 64-bit PCI
    416  * slot. In the latter case, there's no way the NIC can work correctly,
    417  * so we print out a message on the console and abort the device attach.
    418  */
    419 
    420 int
    421 re_diag(struct rtk_softc *sc)
    422 {
    423 	struct ifnet *ifp = &sc->ethercom.ec_if;
    424 	struct mbuf *m0;
    425 	struct ether_header *eh;
    426 	struct re_rxsoft *rxs;
    427 	struct re_desc *cur_rx;
    428 	bus_dmamap_t dmamap;
    429 	uint16_t status;
    430 	uint32_t rxstat;
    431 	int total_len, i, s, error = 0;
    432 	static const uint8_t dst[] = { 0x00, 'h', 'e', 'l', 'l', 'o' };
    433 	static const uint8_t src[] = { 0x00, 'w', 'o', 'r', 'l', 'd' };
    434 
    435 	/* Allocate a single mbuf */
    436 
    437 	MGETHDR(m0, M_DONTWAIT, MT_DATA);
    438 	if (m0 == NULL)
    439 		return ENOBUFS;
    440 
    441 	/*
    442 	 * Initialize the NIC in test mode. This sets the chip up
    443 	 * so that it can send and receive frames, but performs the
    444 	 * following special functions:
    445 	 * - Puts receiver in promiscuous mode
    446 	 * - Enables digital loopback mode
    447 	 * - Leaves interrupts turned off
    448 	 */
    449 
    450 	ifp->if_flags |= IFF_PROMISC;
    451 	sc->re_testmode = 1;
    452 	re_init(ifp);
    453 	re_stop(ifp, 0);
    454 	DELAY(100000);
    455 	re_init(ifp);
    456 
    457 	/* Put some data in the mbuf */
    458 
    459 	eh = mtod(m0, struct ether_header *);
    460 	memcpy(eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN);
    461 	memcpy(eh->ether_shost, (char *)&src, ETHER_ADDR_LEN);
    462 	eh->ether_type = htons(ETHERTYPE_IP);
    463 	m0->m_pkthdr.len = m0->m_len = ETHER_MIN_LEN - ETHER_CRC_LEN;
    464 
    465 	/*
    466 	 * Queue the packet, start transmission.
    467 	 */
    468 
    469 	CSR_WRITE_2(sc, RTK_ISR, 0xFFFF);
    470 	s = splnet();
    471 	IF_ENQUEUE(&ifp->if_snd, m0);
    472 	re_start(ifp);
    473 	splx(s);
    474 	m0 = NULL;
    475 
    476 	/* Wait for it to propagate through the chip */
    477 
    478 	DELAY(100000);
    479 	for (i = 0; i < RTK_TIMEOUT; i++) {
    480 		status = CSR_READ_2(sc, RTK_ISR);
    481 		if ((status & (RTK_ISR_TIMEOUT_EXPIRED | RTK_ISR_RX_OK)) ==
    482 		    (RTK_ISR_TIMEOUT_EXPIRED | RTK_ISR_RX_OK))
    483 			break;
    484 		DELAY(10);
    485 	}
    486 	if (i == RTK_TIMEOUT) {
    487 		aprint_error_dev(sc->sc_dev,
    488 		    "diagnostic failed, failed to receive packet "
    489 		    "in loopback mode\n");
    490 		error = EIO;
    491 		goto done;
    492 	}
    493 
    494 	/*
    495 	 * The packet should have been dumped into the first
    496 	 * entry in the RX DMA ring. Grab it from there.
    497 	 */
    498 
    499 	rxs = &sc->re_ldata.re_rxsoft[0];
    500 	dmamap = rxs->rxs_dmamap;
    501 	bus_dmamap_sync(sc->sc_dmat, dmamap, 0, dmamap->dm_mapsize,
    502 	    BUS_DMASYNC_POSTREAD);
    503 	bus_dmamap_unload(sc->sc_dmat, dmamap);
    504 
    505 	m0 = rxs->rxs_mbuf;
    506 	rxs->rxs_mbuf = NULL;
    507 	eh = mtod(m0, struct ether_header *);
    508 
    509 	RE_RXDESCSYNC(sc, 0, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
    510 	cur_rx = &sc->re_ldata.re_rx_list[0];
    511 	rxstat = le32toh(cur_rx->re_cmdstat);
    512 	total_len = rxstat & sc->re_rxlenmask;
    513 
    514 	if (total_len != ETHER_MIN_LEN) {
    515 		aprint_error_dev(sc->sc_dev,
    516 		    "diagnostic failed, received short packet\n");
    517 		error = EIO;
    518 		goto done;
    519 	}
    520 
    521 	/* Test that the received packet data matches what we sent. */
    522 
    523 	if (memcmp((char *)&eh->ether_dhost, (char *)&dst, ETHER_ADDR_LEN) ||
    524 	    memcmp((char *)&eh->ether_shost, (char *)&src, ETHER_ADDR_LEN) ||
    525 	    ntohs(eh->ether_type) != ETHERTYPE_IP) {
    526 		aprint_error_dev(sc->sc_dev, "WARNING, DMA FAILURE!\n"
    527 		    "expected TX data: %s/%s/0x%x\n"
    528 		    "received RX data: %s/%s/0x%x\n"
    529 		    "You may have a defective 32-bit NIC plugged "
    530 		    "into a 64-bit PCI slot.\n"
    531 		    "Please re-install the NIC in a 32-bit slot "
    532 		    "for proper operation.\n"
    533 		    "Read the re(4) man page for more details.\n" ,
    534 		    ether_sprintf(dst),  ether_sprintf(src), ETHERTYPE_IP,
    535 		    ether_sprintf(eh->ether_dhost),
    536 		    ether_sprintf(eh->ether_shost), ntohs(eh->ether_type));
    537 		error = EIO;
    538 	}
    539 
    540  done:
    541 	/* Turn interface off, release resources */
    542 
    543 	sc->re_testmode = 0;
    544 	ifp->if_flags &= ~IFF_PROMISC;
    545 	re_stop(ifp, 0);
    546 	if (m0 != NULL)
    547 		m_freem(m0);
    548 
    549 	return error;
    550 }
    551 
    552 
    553 /*
    554  * Attach the interface. Allocate softc structures, do ifmedia
    555  * setup and ethernet/BPF attach.
    556  */
    557 void
    558 re_attach(struct rtk_softc *sc)
    559 {
    560 	uint8_t eaddr[ETHER_ADDR_LEN];
    561 	uint16_t val;
    562 	struct ifnet *ifp;
    563 	int error = 0, i, addr_len;
    564 
    565 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0) {
    566 		uint32_t hwrev;
    567 
    568 		/* Revision of 8169/8169S/8110s in bits 30..26, 23 */
    569 		hwrev = CSR_READ_4(sc, RTK_TXCFG) & RTK_TXCFG_HWREV;
    570 		/* These rev numbers are taken from Realtek's driver */
    571 		switch (hwrev) {
    572 		case RTK_HWREV_8169:
    573 			/* XXX not in the Realtek driver */
    574 			sc->sc_rev = 1;
    575 			sc->sc_quirk |= RTKQ_8169NONS;
    576 			break;
    577 		case RTK_HWREV_8169S:
    578 		case RTK_HWREV_8110S:
    579 			sc->sc_rev = 3;
    580 			sc->sc_quirk |= RTKQ_MACLDPS;
    581 			break;
    582 		case RTK_HWREV_8169_8110SB:
    583 			sc->sc_rev = 4;
    584 			sc->sc_quirk |= RTKQ_MACLDPS;
    585 			break;
    586 		case RTK_HWREV_8169_8110SC:
    587 			sc->sc_rev = 5;
    588 			sc->sc_quirk |= RTKQ_MACLDPS;
    589 			break;
    590 		case RTK_HWREV_8101E:
    591 			sc->sc_rev = 11;
    592 			sc->sc_quirk |= RTKQ_NOJUMBO;
    593 			break;
    594 		case RTK_HWREV_8168_SPIN1:
    595 			sc->sc_rev = 21;
    596 			break;
    597 		case RTK_HWREV_8168_SPIN2:
    598 			sc->sc_rev = 22;
    599 			break;
    600 		case RTK_HWREV_8168_SPIN3:
    601 			sc->sc_rev = 23;
    602 			break;
    603 		case RTK_HWREV_8168C:
    604 		case RTK_HWREV_8168C_SPIN2:
    605 		case RTK_HWREV_8168CP:
    606 		case RTK_HWREV_8168D:
    607 			sc->sc_rev = 24;
    608 			sc->sc_quirk |= RTKQ_DESCV2 | RTKQ_NOEECMD;
    609 			/*
    610 			 * From FreeBSD driver:
    611 			 *
    612 			 * These (8168/8111) controllers support jumbo frame
    613 			 * but it seems that enabling it requires touching
    614 			 * additional magic registers. Depending on MAC
    615 			 * revisions some controllers need to disable
    616 			 * checksum offload. So disable jumbo frame until
    617 			 * I have better idea what it really requires to
    618 			 * make it support.
    619 			 * RTL8168C/CP : supports up to 6KB jumbo frame.
    620 			 * RTL8111C/CP : supports up to 9KB jumbo frame.
    621 			 */
    622 			sc->sc_quirk |= RTKQ_NOJUMBO;
    623 			break;
    624 		case RTK_HWREV_8102E:
    625 		case RTK_HWREV_8102EL:
    626 		case RTK_HWREV_8102EL_SPIN2:
    627 			sc->sc_rev = 25;
    628 			sc->sc_quirk |=
    629 			    RTKQ_DESCV2 | RTKQ_NOEECMD | RTKQ_NOJUMBO;
    630 			break;
    631 		case RTK_HWREV_8100E:
    632 		case RTK_HWREV_8100E_SPIN2:
    633 			/* XXX not in the Realtek driver */
    634 			sc->sc_rev = 0;
    635 			sc->sc_quirk |= RTKQ_NOJUMBO;
    636 			break;
    637 		default:
    638 			aprint_normal_dev(sc->sc_dev,
    639 			    "Unknown revision (0x%08x)\n", hwrev);
    640 			sc->sc_rev = 0;
    641 		}
    642 
    643 		/* Set RX length mask */
    644 		sc->re_rxlenmask = RE_RDESC_STAT_GFRAGLEN;
    645 		sc->re_ldata.re_tx_desc_cnt = RE_TX_DESC_CNT_8169;
    646 	} else {
    647 		sc->sc_quirk |= RTKQ_NOJUMBO;
    648 
    649 		/* Set RX length mask */
    650 		sc->re_rxlenmask = RE_RDESC_STAT_FRAGLEN;
    651 		sc->re_ldata.re_tx_desc_cnt = RE_TX_DESC_CNT_8139;
    652 	}
    653 
    654 	/* Reset the adapter. */
    655 	re_reset(sc);
    656 
    657 	if ((sc->sc_quirk & RTKQ_NOEECMD) != 0) {
    658 		/*
    659 		 * Get station address from ID registers.
    660 		 */
    661 		for (i = 0; i < ETHER_ADDR_LEN; i++)
    662 			eaddr[i] = CSR_READ_1(sc, RTK_IDR0 + i);
    663 	} else {
    664 		/*
    665 		 * Get station address from the EEPROM.
    666 		 */
    667 		if (rtk_read_eeprom(sc, RTK_EE_ID, RTK_EEADDR_LEN1) == 0x8129)
    668 			addr_len = RTK_EEADDR_LEN1;
    669 		else
    670 			addr_len = RTK_EEADDR_LEN0;
    671 
    672 		/*
    673 		 * Get station address from the EEPROM.
    674 		 */
    675 		for (i = 0; i < ETHER_ADDR_LEN / 2; i++) {
    676 			val = rtk_read_eeprom(sc, RTK_EE_EADDR0 + i, addr_len);
    677 			eaddr[(i * 2) + 0] = val & 0xff;
    678 			eaddr[(i * 2) + 1] = val >> 8;
    679 		}
    680 	}
    681 
    682 	aprint_normal_dev(sc->sc_dev, "Ethernet address %s\n",
    683 	    ether_sprintf(eaddr));
    684 
    685 	if (sc->re_ldata.re_tx_desc_cnt >
    686 	    PAGE_SIZE / sizeof(struct re_desc)) {
    687 		sc->re_ldata.re_tx_desc_cnt =
    688 		    PAGE_SIZE / sizeof(struct re_desc);
    689 	}
    690 
    691 	aprint_verbose_dev(sc->sc_dev, "using %d tx descriptors\n",
    692 	    sc->re_ldata.re_tx_desc_cnt);
    693 	KASSERT(RE_NEXT_TX_DESC(sc, RE_TX_DESC_CNT(sc) - 1) == 0);
    694 
    695 	/* Allocate DMA'able memory for the TX ring */
    696 	if ((error = bus_dmamem_alloc(sc->sc_dmat, RE_TX_LIST_SZ(sc),
    697 	    RE_RING_ALIGN, 0, &sc->re_ldata.re_tx_listseg, 1,
    698 	    &sc->re_ldata.re_tx_listnseg, BUS_DMA_NOWAIT)) != 0) {
    699 		aprint_error_dev(sc->sc_dev,
    700 		    "can't allocate tx listseg, error = %d\n", error);
    701 		goto fail_0;
    702 	}
    703 
    704 	/* Load the map for the TX ring. */
    705 	if ((error = bus_dmamem_map(sc->sc_dmat, &sc->re_ldata.re_tx_listseg,
    706 	    sc->re_ldata.re_tx_listnseg, RE_TX_LIST_SZ(sc),
    707 	    (void **)&sc->re_ldata.re_tx_list,
    708 	    BUS_DMA_COHERENT | BUS_DMA_NOWAIT)) != 0) {
    709 		aprint_error_dev(sc->sc_dev,
    710 		    "can't map tx list, error = %d\n", error);
    711 	  	goto fail_1;
    712 	}
    713 	memset(sc->re_ldata.re_tx_list, 0, RE_TX_LIST_SZ(sc));
    714 
    715 	if ((error = bus_dmamap_create(sc->sc_dmat, RE_TX_LIST_SZ(sc), 1,
    716 	    RE_TX_LIST_SZ(sc), 0, 0,
    717 	    &sc->re_ldata.re_tx_list_map)) != 0) {
    718 		aprint_error_dev(sc->sc_dev,
    719 		    "can't create tx list map, error = %d\n", error);
    720 		goto fail_2;
    721 	}
    722 
    723 
    724 	if ((error = bus_dmamap_load(sc->sc_dmat,
    725 	    sc->re_ldata.re_tx_list_map, sc->re_ldata.re_tx_list,
    726 	    RE_TX_LIST_SZ(sc), NULL, BUS_DMA_NOWAIT)) != 0) {
    727 		aprint_error_dev(sc->sc_dev,
    728 		    "can't load tx list, error = %d\n", error);
    729 		goto fail_3;
    730 	}
    731 
    732 	/* Create DMA maps for TX buffers */
    733 	for (i = 0; i < RE_TX_QLEN; i++) {
    734 		error = bus_dmamap_create(sc->sc_dmat,
    735 		    round_page(IP_MAXPACKET),
    736 		    RE_TX_DESC_CNT(sc), RE_TDESC_CMD_FRAGLEN,
    737 		    0, 0, &sc->re_ldata.re_txq[i].txq_dmamap);
    738 		if (error) {
    739 			aprint_error_dev(sc->sc_dev,
    740 			    "can't create DMA map for TX\n");
    741 			goto fail_4;
    742 		}
    743 	}
    744 
    745 	/* Allocate DMA'able memory for the RX ring */
    746 	/* XXX see also a comment about RE_RX_DMAMEM_SZ in rtl81x9var.h */
    747 	if ((error = bus_dmamem_alloc(sc->sc_dmat,
    748 	    RE_RX_DMAMEM_SZ, RE_RING_ALIGN, 0, &sc->re_ldata.re_rx_listseg, 1,
    749 	    &sc->re_ldata.re_rx_listnseg, BUS_DMA_NOWAIT)) != 0) {
    750 		aprint_error_dev(sc->sc_dev,
    751 		    "can't allocate rx listseg, error = %d\n", error);
    752 		goto fail_4;
    753 	}
    754 
    755 	/* Load the map for the RX ring. */
    756 	if ((error = bus_dmamem_map(sc->sc_dmat, &sc->re_ldata.re_rx_listseg,
    757 	    sc->re_ldata.re_rx_listnseg, RE_RX_DMAMEM_SZ,
    758 	    (void **)&sc->re_ldata.re_rx_list,
    759 	    BUS_DMA_COHERENT | BUS_DMA_NOWAIT)) != 0) {
    760 		aprint_error_dev(sc->sc_dev,
    761 		    "can't map rx list, error = %d\n", error);
    762 		goto fail_5;
    763 	}
    764 	memset(sc->re_ldata.re_rx_list, 0, RE_RX_DMAMEM_SZ);
    765 
    766 	if ((error = bus_dmamap_create(sc->sc_dmat,
    767 	    RE_RX_DMAMEM_SZ, 1, RE_RX_DMAMEM_SZ, 0, 0,
    768 	    &sc->re_ldata.re_rx_list_map)) != 0) {
    769 		aprint_error_dev(sc->sc_dev,
    770 		    "can't create rx list map, error = %d\n", error);
    771 		goto fail_6;
    772 	}
    773 
    774 	if ((error = bus_dmamap_load(sc->sc_dmat,
    775 	    sc->re_ldata.re_rx_list_map, sc->re_ldata.re_rx_list,
    776 	    RE_RX_DMAMEM_SZ, NULL, BUS_DMA_NOWAIT)) != 0) {
    777 		aprint_error_dev(sc->sc_dev,
    778 		    "can't load rx list, error = %d\n", error);
    779 		goto fail_7;
    780 	}
    781 
    782 	/* Create DMA maps for RX buffers */
    783 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
    784 		error = bus_dmamap_create(sc->sc_dmat, MCLBYTES, 1, MCLBYTES,
    785 		    0, 0, &sc->re_ldata.re_rxsoft[i].rxs_dmamap);
    786 		if (error) {
    787 			aprint_error_dev(sc->sc_dev,
    788 			    "can't create DMA map for RX\n");
    789 			goto fail_8;
    790 		}
    791 	}
    792 
    793 	/*
    794 	 * Record interface as attached. From here, we should not fail.
    795 	 */
    796 	sc->sc_flags |= RTK_ATTACHED;
    797 
    798 	ifp = &sc->ethercom.ec_if;
    799 	ifp->if_softc = sc;
    800 	strlcpy(ifp->if_xname, device_xname(sc->sc_dev), IFNAMSIZ);
    801 	ifp->if_mtu = ETHERMTU;
    802 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
    803 	ifp->if_ioctl = re_ioctl;
    804 	sc->ethercom.ec_capabilities |=
    805 	    ETHERCAP_VLAN_MTU | ETHERCAP_VLAN_HWTAGGING;
    806 	ifp->if_start = re_start;
    807 	ifp->if_stop = re_stop;
    808 
    809 	/*
    810 	 * IFCAP_CSUM_IPv4_Tx on re(4) is broken for small packets,
    811 	 * so we have a workaround to handle the bug by padding
    812 	 * such packets manually.
    813 	 */
    814 	ifp->if_capabilities |=
    815 	    IFCAP_CSUM_IPv4_Tx | IFCAP_CSUM_IPv4_Rx |
    816 	    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
    817 	    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx |
    818 	    IFCAP_TSOv4;
    819 
    820 	/*
    821 	 * XXX
    822 	 * Still have no idea how to make TSO work on 8168C, 8168CP,
    823 	 * 8102E, 8111C and 8111CP.
    824 	 */
    825 	if ((sc->sc_quirk & RTKQ_DESCV2) != 0)
    826 		ifp->if_capabilities &= ~IFCAP_TSOv4;
    827 
    828 	ifp->if_watchdog = re_watchdog;
    829 	ifp->if_init = re_init;
    830 	ifp->if_snd.ifq_maxlen = RE_IFQ_MAXLEN;
    831 	ifp->if_capenable = ifp->if_capabilities;
    832 	IFQ_SET_READY(&ifp->if_snd);
    833 
    834 	callout_init(&sc->rtk_tick_ch, 0);
    835 
    836 	/* Do MII setup */
    837 	sc->mii.mii_ifp = ifp;
    838 	sc->mii.mii_readreg = re_miibus_readreg;
    839 	sc->mii.mii_writereg = re_miibus_writereg;
    840 	sc->mii.mii_statchg = re_miibus_statchg;
    841 	sc->ethercom.ec_mii = &sc->mii;
    842 	ifmedia_init(&sc->mii.mii_media, IFM_IMASK, ether_mediachange,
    843 	    ether_mediastatus);
    844 	mii_attach(sc->sc_dev, &sc->mii, 0xffffffff, MII_PHY_ANY,
    845 	    MII_OFFSET_ANY, 0);
    846 	ifmedia_set(&sc->mii.mii_media, IFM_ETHER | IFM_AUTO);
    847 
    848 	/*
    849 	 * Call MI attach routine.
    850 	 */
    851 	if_attach(ifp);
    852 	ether_ifattach(ifp, eaddr);
    853 
    854 	return;
    855 
    856  fail_8:
    857 	/* Destroy DMA maps for RX buffers. */
    858 	for (i = 0; i < RE_RX_DESC_CNT; i++)
    859 		if (sc->re_ldata.re_rxsoft[i].rxs_dmamap != NULL)
    860 			bus_dmamap_destroy(sc->sc_dmat,
    861 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
    862 
    863 	/* Free DMA'able memory for the RX ring. */
    864 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
    865  fail_7:
    866 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
    867  fail_6:
    868 	bus_dmamem_unmap(sc->sc_dmat,
    869 	    (void *)sc->re_ldata.re_rx_list, RE_RX_DMAMEM_SZ);
    870  fail_5:
    871 	bus_dmamem_free(sc->sc_dmat,
    872 	    &sc->re_ldata.re_rx_listseg, sc->re_ldata.re_rx_listnseg);
    873 
    874  fail_4:
    875 	/* Destroy DMA maps for TX buffers. */
    876 	for (i = 0; i < RE_TX_QLEN; i++)
    877 		if (sc->re_ldata.re_txq[i].txq_dmamap != NULL)
    878 			bus_dmamap_destroy(sc->sc_dmat,
    879 			    sc->re_ldata.re_txq[i].txq_dmamap);
    880 
    881 	/* Free DMA'able memory for the TX ring. */
    882 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
    883  fail_3:
    884 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
    885  fail_2:
    886 	bus_dmamem_unmap(sc->sc_dmat,
    887 	    (void *)sc->re_ldata.re_tx_list, RE_TX_LIST_SZ(sc));
    888  fail_1:
    889 	bus_dmamem_free(sc->sc_dmat,
    890 	    &sc->re_ldata.re_tx_listseg, sc->re_ldata.re_tx_listnseg);
    891  fail_0:
    892 	return;
    893 }
    894 
    895 
    896 /*
    897  * re_activate:
    898  *     Handle device activation/deactivation requests.
    899  */
    900 int
    901 re_activate(device_t self, enum devact act)
    902 {
    903 	struct rtk_softc *sc = device_private(self);
    904 	int s, error = 0;
    905 
    906 	s = splnet();
    907 	switch (act) {
    908 	case DVACT_ACTIVATE:
    909 		error = EOPNOTSUPP;
    910 		break;
    911 	case DVACT_DEACTIVATE:
    912 		mii_activate(&sc->mii, act, MII_PHY_ANY, MII_OFFSET_ANY);
    913 		if_deactivate(&sc->ethercom.ec_if);
    914 		break;
    915 	}
    916 	splx(s);
    917 
    918 	return error;
    919 }
    920 
    921 /*
    922  * re_detach:
    923  *     Detach a rtk interface.
    924  */
    925 int
    926 re_detach(struct rtk_softc *sc)
    927 {
    928 	struct ifnet *ifp = &sc->ethercom.ec_if;
    929 	int i;
    930 
    931 	/*
    932 	 * Succeed now if there isn't any work to do.
    933 	 */
    934 	if ((sc->sc_flags & RTK_ATTACHED) == 0)
    935 		return 0;
    936 
    937 	/* Unhook our tick handler. */
    938 	callout_stop(&sc->rtk_tick_ch);
    939 
    940 	/* Detach all PHYs. */
    941 	mii_detach(&sc->mii, MII_PHY_ANY, MII_OFFSET_ANY);
    942 
    943 	/* Delete all remaining media. */
    944 	ifmedia_delete_instance(&sc->mii.mii_media, IFM_INST_ANY);
    945 
    946 	ether_ifdetach(ifp);
    947 	if_detach(ifp);
    948 
    949 	/* Destroy DMA maps for RX buffers. */
    950 	for (i = 0; i < RE_RX_DESC_CNT; i++)
    951 		if (sc->re_ldata.re_rxsoft[i].rxs_dmamap != NULL)
    952 			bus_dmamap_destroy(sc->sc_dmat,
    953 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
    954 
    955 	/* Free DMA'able memory for the RX ring. */
    956 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
    957 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_rx_list_map);
    958 	bus_dmamem_unmap(sc->sc_dmat,
    959 	    (void *)sc->re_ldata.re_rx_list, RE_RX_DMAMEM_SZ);
    960 	bus_dmamem_free(sc->sc_dmat,
    961 	    &sc->re_ldata.re_rx_listseg, sc->re_ldata.re_rx_listnseg);
    962 
    963 	/* Destroy DMA maps for TX buffers. */
    964 	for (i = 0; i < RE_TX_QLEN; i++)
    965 		if (sc->re_ldata.re_txq[i].txq_dmamap != NULL)
    966 			bus_dmamap_destroy(sc->sc_dmat,
    967 			    sc->re_ldata.re_txq[i].txq_dmamap);
    968 
    969 	/* Free DMA'able memory for the TX ring. */
    970 	bus_dmamap_unload(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
    971 	bus_dmamap_destroy(sc->sc_dmat, sc->re_ldata.re_tx_list_map);
    972 	bus_dmamem_unmap(sc->sc_dmat,
    973 	    (void *)sc->re_ldata.re_tx_list, RE_TX_LIST_SZ(sc));
    974 	bus_dmamem_free(sc->sc_dmat,
    975 	    &sc->re_ldata.re_tx_listseg, sc->re_ldata.re_tx_listnseg);
    976 
    977 	return 0;
    978 }
    979 
    980 /*
    981  * re_enable:
    982  *     Enable the RTL81X9 chip.
    983  */
    984 static int
    985 re_enable(struct rtk_softc *sc)
    986 {
    987 
    988 	if (RTK_IS_ENABLED(sc) == 0 && sc->sc_enable != NULL) {
    989 		if ((*sc->sc_enable)(sc) != 0) {
    990 			printf("%s: device enable failed\n",
    991 			    device_xname(sc->sc_dev));
    992 			return EIO;
    993 		}
    994 		sc->sc_flags |= RTK_ENABLED;
    995 	}
    996 	return 0;
    997 }
    998 
    999 /*
   1000  * re_disable:
   1001  *     Disable the RTL81X9 chip.
   1002  */
   1003 static void
   1004 re_disable(struct rtk_softc *sc)
   1005 {
   1006 
   1007 	if (RTK_IS_ENABLED(sc) && sc->sc_disable != NULL) {
   1008 		(*sc->sc_disable)(sc);
   1009 		sc->sc_flags &= ~RTK_ENABLED;
   1010 	}
   1011 }
   1012 
   1013 static int
   1014 re_newbuf(struct rtk_softc *sc, int idx, struct mbuf *m)
   1015 {
   1016 	struct mbuf *n = NULL;
   1017 	bus_dmamap_t map;
   1018 	struct re_desc *d;
   1019 	struct re_rxsoft *rxs;
   1020 	uint32_t cmdstat;
   1021 	int error;
   1022 
   1023 	if (m == NULL) {
   1024 		MGETHDR(n, M_DONTWAIT, MT_DATA);
   1025 		if (n == NULL)
   1026 			return ENOBUFS;
   1027 
   1028 		MCLGET(n, M_DONTWAIT);
   1029 		if ((n->m_flags & M_EXT) == 0) {
   1030 			m_freem(n);
   1031 			return ENOBUFS;
   1032 		}
   1033 		m = n;
   1034 	} else
   1035 		m->m_data = m->m_ext.ext_buf;
   1036 
   1037 	/*
   1038 	 * Initialize mbuf length fields and fixup
   1039 	 * alignment so that the frame payload is
   1040 	 * longword aligned.
   1041 	 */
   1042 	m->m_len = m->m_pkthdr.len = MCLBYTES - RE_ETHER_ALIGN;
   1043 	m->m_data += RE_ETHER_ALIGN;
   1044 
   1045 	rxs = &sc->re_ldata.re_rxsoft[idx];
   1046 	map = rxs->rxs_dmamap;
   1047 	error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m,
   1048 	    BUS_DMA_READ|BUS_DMA_NOWAIT);
   1049 
   1050 	if (error)
   1051 		goto out;
   1052 
   1053 	bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
   1054 	    BUS_DMASYNC_PREREAD);
   1055 
   1056 	d = &sc->re_ldata.re_rx_list[idx];
   1057 #ifdef DIAGNOSTIC
   1058 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
   1059 	cmdstat = le32toh(d->re_cmdstat);
   1060 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD);
   1061 	if (cmdstat & RE_RDESC_STAT_OWN) {
   1062 		panic("%s: tried to map busy RX descriptor",
   1063 		    device_xname(sc->sc_dev));
   1064 	}
   1065 #endif
   1066 
   1067 	rxs->rxs_mbuf = m;
   1068 
   1069 	d->re_vlanctl = 0;
   1070 	cmdstat = map->dm_segs[0].ds_len;
   1071 	if (idx == (RE_RX_DESC_CNT - 1))
   1072 		cmdstat |= RE_RDESC_CMD_EOR;
   1073 	re_set_bufaddr(d, map->dm_segs[0].ds_addr);
   1074 	d->re_cmdstat = htole32(cmdstat);
   1075 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1076 	cmdstat |= RE_RDESC_CMD_OWN;
   1077 	d->re_cmdstat = htole32(cmdstat);
   1078 	RE_RXDESCSYNC(sc, idx, BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1079 
   1080 	return 0;
   1081  out:
   1082 	if (n != NULL)
   1083 		m_freem(n);
   1084 	return ENOMEM;
   1085 }
   1086 
   1087 static int
   1088 re_tx_list_init(struct rtk_softc *sc)
   1089 {
   1090 	int i;
   1091 
   1092 	memset(sc->re_ldata.re_tx_list, 0, RE_TX_LIST_SZ(sc));
   1093 	for (i = 0; i < RE_TX_QLEN; i++) {
   1094 		sc->re_ldata.re_txq[i].txq_mbuf = NULL;
   1095 	}
   1096 
   1097 	bus_dmamap_sync(sc->sc_dmat,
   1098 	    sc->re_ldata.re_tx_list_map, 0,
   1099 	    sc->re_ldata.re_tx_list_map->dm_mapsize,
   1100 	    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1101 	sc->re_ldata.re_txq_prodidx = 0;
   1102 	sc->re_ldata.re_txq_considx = 0;
   1103 	sc->re_ldata.re_txq_free = RE_TX_QLEN;
   1104 	sc->re_ldata.re_tx_free = RE_TX_DESC_CNT(sc);
   1105 	sc->re_ldata.re_tx_nextfree = 0;
   1106 
   1107 	return 0;
   1108 }
   1109 
   1110 static int
   1111 re_rx_list_init(struct rtk_softc *sc)
   1112 {
   1113 	int i;
   1114 
   1115 	memset(sc->re_ldata.re_rx_list, 0, RE_RX_LIST_SZ);
   1116 
   1117 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
   1118 		if (re_newbuf(sc, i, NULL) == ENOBUFS)
   1119 			return ENOBUFS;
   1120 	}
   1121 
   1122 	sc->re_ldata.re_rx_prodidx = 0;
   1123 	sc->re_head = sc->re_tail = NULL;
   1124 
   1125 	return 0;
   1126 }
   1127 
   1128 /*
   1129  * RX handler for C+ and 8169. For the gigE chips, we support
   1130  * the reception of jumbo frames that have been fragmented
   1131  * across multiple 2K mbuf cluster buffers.
   1132  */
   1133 static void
   1134 re_rxeof(struct rtk_softc *sc)
   1135 {
   1136 	struct mbuf *m;
   1137 	struct ifnet *ifp;
   1138 	int i, total_len;
   1139 	struct re_desc *cur_rx;
   1140 	struct re_rxsoft *rxs;
   1141 	uint32_t rxstat, rxvlan;
   1142 
   1143 	ifp = &sc->ethercom.ec_if;
   1144 
   1145 	for (i = sc->re_ldata.re_rx_prodidx;; i = RE_NEXT_RX_DESC(sc, i)) {
   1146 		cur_rx = &sc->re_ldata.re_rx_list[i];
   1147 		RE_RXDESCSYNC(sc, i,
   1148 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
   1149 		rxstat = le32toh(cur_rx->re_cmdstat);
   1150 		rxvlan = le32toh(cur_rx->re_vlanctl);
   1151 		RE_RXDESCSYNC(sc, i, BUS_DMASYNC_PREREAD);
   1152 		if ((rxstat & RE_RDESC_STAT_OWN) != 0) {
   1153 			break;
   1154 		}
   1155 		total_len = rxstat & sc->re_rxlenmask;
   1156 		rxs = &sc->re_ldata.re_rxsoft[i];
   1157 		m = rxs->rxs_mbuf;
   1158 
   1159 		/* Invalidate the RX mbuf and unload its map */
   1160 
   1161 		bus_dmamap_sync(sc->sc_dmat,
   1162 		    rxs->rxs_dmamap, 0, rxs->rxs_dmamap->dm_mapsize,
   1163 		    BUS_DMASYNC_POSTREAD);
   1164 		bus_dmamap_unload(sc->sc_dmat, rxs->rxs_dmamap);
   1165 
   1166 		if ((rxstat & RE_RDESC_STAT_EOF) == 0) {
   1167 			m->m_len = MCLBYTES - RE_ETHER_ALIGN;
   1168 			if (sc->re_head == NULL)
   1169 				sc->re_head = sc->re_tail = m;
   1170 			else {
   1171 				m->m_flags &= ~M_PKTHDR;
   1172 				sc->re_tail->m_next = m;
   1173 				sc->re_tail = m;
   1174 			}
   1175 			re_newbuf(sc, i, NULL);
   1176 			continue;
   1177 		}
   1178 
   1179 		/*
   1180 		 * NOTE: for the 8139C+, the frame length field
   1181 		 * is always 12 bits in size, but for the gigE chips,
   1182 		 * it is 13 bits (since the max RX frame length is 16K).
   1183 		 * Unfortunately, all 32 bits in the status word
   1184 		 * were already used, so to make room for the extra
   1185 		 * length bit, RealTek took out the 'frame alignment
   1186 		 * error' bit and shifted the other status bits
   1187 		 * over one slot. The OWN, EOR, FS and LS bits are
   1188 		 * still in the same places. We have already extracted
   1189 		 * the frame length and checked the OWN bit, so rather
   1190 		 * than using an alternate bit mapping, we shift the
   1191 		 * status bits one space to the right so we can evaluate
   1192 		 * them using the 8169 status as though it was in the
   1193 		 * same format as that of the 8139C+.
   1194 		 */
   1195 		if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0)
   1196 			rxstat >>= 1;
   1197 
   1198 		if (__predict_false((rxstat & RE_RDESC_STAT_RXERRSUM) != 0)) {
   1199 #ifdef RE_DEBUG
   1200 			printf("%s: RX error (rxstat = 0x%08x)",
   1201 			    device_xname(sc->sc_dev), rxstat);
   1202 			if (rxstat & RE_RDESC_STAT_FRALIGN)
   1203 				printf(", frame alignment error");
   1204 			if (rxstat & RE_RDESC_STAT_BUFOFLOW)
   1205 				printf(", out of buffer space");
   1206 			if (rxstat & RE_RDESC_STAT_FIFOOFLOW)
   1207 				printf(", FIFO overrun");
   1208 			if (rxstat & RE_RDESC_STAT_GIANT)
   1209 				printf(", giant packet");
   1210 			if (rxstat & RE_RDESC_STAT_RUNT)
   1211 				printf(", runt packet");
   1212 			if (rxstat & RE_RDESC_STAT_CRCERR)
   1213 				printf(", CRC error");
   1214 			printf("\n");
   1215 #endif
   1216 			ifp->if_ierrors++;
   1217 			/*
   1218 			 * If this is part of a multi-fragment packet,
   1219 			 * discard all the pieces.
   1220 			 */
   1221 			if (sc->re_head != NULL) {
   1222 				m_freem(sc->re_head);
   1223 				sc->re_head = sc->re_tail = NULL;
   1224 			}
   1225 			re_newbuf(sc, i, m);
   1226 			continue;
   1227 		}
   1228 
   1229 		/*
   1230 		 * If allocating a replacement mbuf fails,
   1231 		 * reload the current one.
   1232 		 */
   1233 
   1234 		if (__predict_false(re_newbuf(sc, i, NULL) != 0)) {
   1235 			ifp->if_ierrors++;
   1236 			if (sc->re_head != NULL) {
   1237 				m_freem(sc->re_head);
   1238 				sc->re_head = sc->re_tail = NULL;
   1239 			}
   1240 			re_newbuf(sc, i, m);
   1241 			continue;
   1242 		}
   1243 
   1244 		if (sc->re_head != NULL) {
   1245 			m->m_len = total_len % (MCLBYTES - RE_ETHER_ALIGN);
   1246 			/*
   1247 			 * Special case: if there's 4 bytes or less
   1248 			 * in this buffer, the mbuf can be discarded:
   1249 			 * the last 4 bytes is the CRC, which we don't
   1250 			 * care about anyway.
   1251 			 */
   1252 			if (m->m_len <= ETHER_CRC_LEN) {
   1253 				sc->re_tail->m_len -=
   1254 				    (ETHER_CRC_LEN - m->m_len);
   1255 				m_freem(m);
   1256 			} else {
   1257 				m->m_len -= ETHER_CRC_LEN;
   1258 				m->m_flags &= ~M_PKTHDR;
   1259 				sc->re_tail->m_next = m;
   1260 			}
   1261 			m = sc->re_head;
   1262 			sc->re_head = sc->re_tail = NULL;
   1263 			m->m_pkthdr.len = total_len - ETHER_CRC_LEN;
   1264 		} else
   1265 			m->m_pkthdr.len = m->m_len =
   1266 			    (total_len - ETHER_CRC_LEN);
   1267 
   1268 		ifp->if_ipackets++;
   1269 		m->m_pkthdr.rcvif = ifp;
   1270 
   1271 		/* Do RX checksumming */
   1272 
   1273 		/* Check IP header checksum */
   1274 		if ((rxstat & RE_RDESC_STAT_PROTOID) != 0 &&
   1275 		    ((sc->sc_quirk & RTKQ_DESCV2) == 0 ||
   1276 		     (rxvlan & RE_RDESC_VLANCTL_IPV4) != 0)) {
   1277 			m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
   1278 			if (rxstat & RE_RDESC_STAT_IPSUMBAD)
   1279 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4_BAD;
   1280 		}
   1281 
   1282 		/* Check TCP/UDP checksum */
   1283 		if (RE_TCPPKT(rxstat)) {
   1284 			m->m_pkthdr.csum_flags |= M_CSUM_TCPv4;
   1285 			if (rxstat & RE_RDESC_STAT_TCPSUMBAD)
   1286 				m->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
   1287 		} else if (RE_UDPPKT(rxstat)) {
   1288 			m->m_pkthdr.csum_flags |= M_CSUM_UDPv4;
   1289 			if (rxstat & RE_RDESC_STAT_UDPSUMBAD)
   1290 				m->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
   1291 		}
   1292 
   1293 		if (rxvlan & RE_RDESC_VLANCTL_TAG) {
   1294 			VLAN_INPUT_TAG(ifp, m,
   1295 			     bswap16(rxvlan & RE_RDESC_VLANCTL_DATA),
   1296 			     continue);
   1297 		}
   1298 #if NBPFILTER > 0
   1299 		if (ifp->if_bpf)
   1300 			bpf_mtap(ifp->if_bpf, m);
   1301 #endif
   1302 		(*ifp->if_input)(ifp, m);
   1303 	}
   1304 
   1305 	sc->re_ldata.re_rx_prodidx = i;
   1306 }
   1307 
   1308 static void
   1309 re_txeof(struct rtk_softc *sc)
   1310 {
   1311 	struct ifnet *ifp;
   1312 	struct re_txq *txq;
   1313 	uint32_t txstat;
   1314 	int idx, descidx;
   1315 
   1316 	ifp = &sc->ethercom.ec_if;
   1317 
   1318 	for (idx = sc->re_ldata.re_txq_considx;
   1319 	    sc->re_ldata.re_txq_free < RE_TX_QLEN;
   1320 	    idx = RE_NEXT_TXQ(sc, idx), sc->re_ldata.re_txq_free++) {
   1321 		txq = &sc->re_ldata.re_txq[idx];
   1322 		KASSERT(txq->txq_mbuf != NULL);
   1323 
   1324 		descidx = txq->txq_descidx;
   1325 		RE_TXDESCSYNC(sc, descidx,
   1326 		    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
   1327 		txstat =
   1328 		    le32toh(sc->re_ldata.re_tx_list[descidx].re_cmdstat);
   1329 		RE_TXDESCSYNC(sc, descidx, BUS_DMASYNC_PREREAD);
   1330 		KASSERT((txstat & RE_TDESC_CMD_EOF) != 0);
   1331 		if (txstat & RE_TDESC_CMD_OWN) {
   1332 			break;
   1333 		}
   1334 
   1335 		sc->re_ldata.re_tx_free += txq->txq_nsegs;
   1336 		KASSERT(sc->re_ldata.re_tx_free <= RE_TX_DESC_CNT(sc));
   1337 		bus_dmamap_sync(sc->sc_dmat, txq->txq_dmamap,
   1338 		    0, txq->txq_dmamap->dm_mapsize, BUS_DMASYNC_POSTWRITE);
   1339 		bus_dmamap_unload(sc->sc_dmat, txq->txq_dmamap);
   1340 		m_freem(txq->txq_mbuf);
   1341 		txq->txq_mbuf = NULL;
   1342 
   1343 		if (txstat & (RE_TDESC_STAT_EXCESSCOL | RE_TDESC_STAT_COLCNT))
   1344 			ifp->if_collisions++;
   1345 		if (txstat & RE_TDESC_STAT_TXERRSUM)
   1346 			ifp->if_oerrors++;
   1347 		else
   1348 			ifp->if_opackets++;
   1349 	}
   1350 
   1351 	sc->re_ldata.re_txq_considx = idx;
   1352 
   1353 	if (sc->re_ldata.re_txq_free > RE_NTXDESC_RSVD)
   1354 		ifp->if_flags &= ~IFF_OACTIVE;
   1355 
   1356 	/*
   1357 	 * If not all descriptors have been released reaped yet,
   1358 	 * reload the timer so that we will eventually get another
   1359 	 * interrupt that will cause us to re-enter this routine.
   1360 	 * This is done in case the transmitter has gone idle.
   1361 	 */
   1362 	if (sc->re_ldata.re_txq_free < RE_TX_QLEN) {
   1363 		CSR_WRITE_4(sc, RTK_TIMERCNT, 1);
   1364 		if ((sc->sc_quirk & RTKQ_PCIE) != 0) {
   1365 			/*
   1366 			 * Some chips will ignore a second TX request
   1367 			 * issued while an existing transmission is in
   1368 			 * progress. If the transmitter goes idle but
   1369 			 * there are still packets waiting to be sent,
   1370 			 * we need to restart the channel here to flush
   1371 			 * them out. This only seems to be required with
   1372 			 * the PCIe devices.
   1373 			 */
   1374 			CSR_WRITE_1(sc, RTK_GTXSTART, RTK_TXSTART_START);
   1375 		}
   1376 	} else
   1377 		ifp->if_timer = 0;
   1378 }
   1379 
   1380 static void
   1381 re_tick(void *arg)
   1382 {
   1383 	struct rtk_softc *sc = arg;
   1384 	int s;
   1385 
   1386 	/*XXX: just return for 8169S/8110S with rev 2 or newer phy */
   1387 	s = splnet();
   1388 
   1389 	mii_tick(&sc->mii);
   1390 	splx(s);
   1391 
   1392 	callout_reset(&sc->rtk_tick_ch, hz, re_tick, sc);
   1393 }
   1394 
   1395 int
   1396 re_intr(void *arg)
   1397 {
   1398 	struct rtk_softc *sc = arg;
   1399 	struct ifnet *ifp;
   1400 	uint16_t status;
   1401 	int handled = 0;
   1402 
   1403 	if (!device_has_power(sc->sc_dev))
   1404 		return 0;
   1405 
   1406 	ifp = &sc->ethercom.ec_if;
   1407 
   1408 	if ((ifp->if_flags & IFF_UP) == 0)
   1409 		return 0;
   1410 
   1411 	for (;;) {
   1412 
   1413 		status = CSR_READ_2(sc, RTK_ISR);
   1414 		/* If the card has gone away the read returns 0xffff. */
   1415 		if (status == 0xffff)
   1416 			break;
   1417 		if (status) {
   1418 			handled = 1;
   1419 			CSR_WRITE_2(sc, RTK_ISR, status);
   1420 		}
   1421 
   1422 		if ((status & RTK_INTRS_CPLUS) == 0)
   1423 			break;
   1424 
   1425 		if (status & (RTK_ISR_RX_OK | RTK_ISR_RX_ERR))
   1426 			re_rxeof(sc);
   1427 
   1428 		if (status & (RTK_ISR_TIMEOUT_EXPIRED | RTK_ISR_TX_ERR |
   1429 		    RTK_ISR_TX_DESC_UNAVAIL))
   1430 			re_txeof(sc);
   1431 
   1432 		if (status & RTK_ISR_SYSTEM_ERR) {
   1433 			re_init(ifp);
   1434 		}
   1435 
   1436 		if (status & RTK_ISR_LINKCHG) {
   1437 			callout_stop(&sc->rtk_tick_ch);
   1438 			re_tick(sc);
   1439 		}
   1440 	}
   1441 
   1442 	if (handled && !IFQ_IS_EMPTY(&ifp->if_snd))
   1443 		re_start(ifp);
   1444 
   1445 	return handled;
   1446 }
   1447 
   1448 
   1449 
   1450 /*
   1451  * Main transmit routine for C+ and gigE NICs.
   1452  */
   1453 
   1454 static void
   1455 re_start(struct ifnet *ifp)
   1456 {
   1457 	struct rtk_softc *sc;
   1458 	struct mbuf *m;
   1459 	bus_dmamap_t map;
   1460 	struct re_txq *txq;
   1461 	struct re_desc *d;
   1462 	struct m_tag *mtag;
   1463 	uint32_t cmdstat, re_flags, vlanctl;
   1464 	int ofree, idx, error, nsegs, seg;
   1465 	int startdesc, curdesc, lastdesc;
   1466 	bool pad;
   1467 
   1468 	sc = ifp->if_softc;
   1469 	ofree = sc->re_ldata.re_txq_free;
   1470 
   1471 	for (idx = sc->re_ldata.re_txq_prodidx;; idx = RE_NEXT_TXQ(sc, idx)) {
   1472 
   1473 		IFQ_POLL(&ifp->if_snd, m);
   1474 		if (m == NULL)
   1475 			break;
   1476 
   1477 		if (sc->re_ldata.re_txq_free == 0 ||
   1478 		    sc->re_ldata.re_tx_free == 0) {
   1479 			/* no more free slots left */
   1480 			ifp->if_flags |= IFF_OACTIVE;
   1481 			break;
   1482 		}
   1483 
   1484 		/*
   1485 		 * Set up checksum offload. Note: checksum offload bits must
   1486 		 * appear in all descriptors of a multi-descriptor transmit
   1487 		 * attempt. (This is according to testing done with an 8169
   1488 		 * chip. I'm not sure if this is a requirement or a bug.)
   1489 		 */
   1490 
   1491 		vlanctl = 0;
   1492 		if ((m->m_pkthdr.csum_flags & M_CSUM_TSOv4) != 0) {
   1493 			uint32_t segsz = m->m_pkthdr.segsz;
   1494 
   1495 			re_flags = RE_TDESC_CMD_LGSEND |
   1496 			    (segsz << RE_TDESC_CMD_MSSVAL_SHIFT);
   1497 		} else {
   1498 			/*
   1499 			 * set RE_TDESC_CMD_IPCSUM if any checksum offloading
   1500 			 * is requested.  otherwise, RE_TDESC_CMD_TCPCSUM/
   1501 			 * RE_TDESC_CMD_UDPCSUM doesn't make effects.
   1502 			 */
   1503 			re_flags = 0;
   1504 			if ((m->m_pkthdr.csum_flags &
   1505 			    (M_CSUM_IPv4 | M_CSUM_TCPv4 | M_CSUM_UDPv4))
   1506 			    != 0) {
   1507 				if ((sc->sc_quirk & RTKQ_DESCV2) == 0) {
   1508 					re_flags |= RE_TDESC_CMD_IPCSUM;
   1509 					if (m->m_pkthdr.csum_flags &
   1510 					    M_CSUM_TCPv4) {
   1511 						re_flags |=
   1512 						    RE_TDESC_CMD_TCPCSUM;
   1513 					} else if (m->m_pkthdr.csum_flags &
   1514 					    M_CSUM_UDPv4) {
   1515 						re_flags |=
   1516 						    RE_TDESC_CMD_UDPCSUM;
   1517 					}
   1518 				} else {
   1519 					vlanctl |= RE_TDESC_VLANCTL_IPCSUM;
   1520 					if (m->m_pkthdr.csum_flags &
   1521 					    M_CSUM_TCPv4) {
   1522 						vlanctl |=
   1523 						    RE_TDESC_VLANCTL_TCPCSUM;
   1524 					} else if (m->m_pkthdr.csum_flags &
   1525 					    M_CSUM_UDPv4) {
   1526 						vlanctl |=
   1527 						    RE_TDESC_VLANCTL_UDPCSUM;
   1528 					}
   1529 				}
   1530 			}
   1531 		}
   1532 
   1533 		txq = &sc->re_ldata.re_txq[idx];
   1534 		map = txq->txq_dmamap;
   1535 		error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m,
   1536 		    BUS_DMA_WRITE|BUS_DMA_NOWAIT);
   1537 
   1538 		if (__predict_false(error)) {
   1539 			/* XXX try to defrag if EFBIG? */
   1540 			printf("%s: can't map mbuf (error %d)\n",
   1541 			    device_xname(sc->sc_dev), error);
   1542 
   1543 			IFQ_DEQUEUE(&ifp->if_snd, m);
   1544 			m_freem(m);
   1545 			ifp->if_oerrors++;
   1546 			continue;
   1547 		}
   1548 
   1549 		nsegs = map->dm_nsegs;
   1550 		pad = false;
   1551 		if (__predict_false(m->m_pkthdr.len <= RE_IP4CSUMTX_PADLEN &&
   1552 		    (re_flags & RE_TDESC_CMD_IPCSUM) != 0 &&
   1553 		    (sc->sc_quirk & RTKQ_DESCV2) == 0)) {
   1554 			pad = true;
   1555 			nsegs++;
   1556 		}
   1557 
   1558 		if (nsegs > sc->re_ldata.re_tx_free) {
   1559 			/*
   1560 			 * Not enough free descriptors to transmit this packet.
   1561 			 */
   1562 			ifp->if_flags |= IFF_OACTIVE;
   1563 			bus_dmamap_unload(sc->sc_dmat, map);
   1564 			break;
   1565 		}
   1566 
   1567 		IFQ_DEQUEUE(&ifp->if_snd, m);
   1568 
   1569 		/*
   1570 		 * Make sure that the caches are synchronized before we
   1571 		 * ask the chip to start DMA for the packet data.
   1572 		 */
   1573 		bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
   1574 		    BUS_DMASYNC_PREWRITE);
   1575 
   1576 		/*
   1577 		 * Set up hardware VLAN tagging. Note: vlan tag info must
   1578 		 * appear in all descriptors of a multi-descriptor
   1579 		 * transmission attempt.
   1580 		 */
   1581 		if ((mtag = VLAN_OUTPUT_TAG(&sc->ethercom, m)) != NULL)
   1582 			vlanctl |= bswap16(VLAN_TAG_VALUE(mtag)) |
   1583 			    RE_TDESC_VLANCTL_TAG;
   1584 
   1585 		/*
   1586 		 * Map the segment array into descriptors.
   1587 		 * Note that we set the start-of-frame and
   1588 		 * end-of-frame markers for either TX or RX,
   1589 		 * but they really only have meaning in the TX case.
   1590 		 * (In the RX case, it's the chip that tells us
   1591 		 *  where packets begin and end.)
   1592 		 * We also keep track of the end of the ring
   1593 		 * and set the end-of-ring bits as needed,
   1594 		 * and we set the ownership bits in all except
   1595 		 * the very first descriptor. (The caller will
   1596 		 * set this descriptor later when it start
   1597 		 * transmission or reception.)
   1598 		 */
   1599 		curdesc = startdesc = sc->re_ldata.re_tx_nextfree;
   1600 		lastdesc = -1;
   1601 		for (seg = 0; seg < map->dm_nsegs;
   1602 		    seg++, curdesc = RE_NEXT_TX_DESC(sc, curdesc)) {
   1603 			d = &sc->re_ldata.re_tx_list[curdesc];
   1604 #ifdef DIAGNOSTIC
   1605 			RE_TXDESCSYNC(sc, curdesc,
   1606 			    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
   1607 			cmdstat = le32toh(d->re_cmdstat);
   1608 			RE_TXDESCSYNC(sc, curdesc, BUS_DMASYNC_PREREAD);
   1609 			if (cmdstat & RE_TDESC_STAT_OWN) {
   1610 				panic("%s: tried to map busy TX descriptor",
   1611 				    device_xname(sc->sc_dev));
   1612 			}
   1613 #endif
   1614 
   1615 			d->re_vlanctl = htole32(vlanctl);
   1616 			re_set_bufaddr(d, map->dm_segs[seg].ds_addr);
   1617 			cmdstat = re_flags | map->dm_segs[seg].ds_len;
   1618 			if (seg == 0)
   1619 				cmdstat |= RE_TDESC_CMD_SOF;
   1620 			else
   1621 				cmdstat |= RE_TDESC_CMD_OWN;
   1622 			if (curdesc == (RE_TX_DESC_CNT(sc) - 1))
   1623 				cmdstat |= RE_TDESC_CMD_EOR;
   1624 			if (seg == nsegs - 1) {
   1625 				cmdstat |= RE_TDESC_CMD_EOF;
   1626 				lastdesc = curdesc;
   1627 			}
   1628 			d->re_cmdstat = htole32(cmdstat);
   1629 			RE_TXDESCSYNC(sc, curdesc,
   1630 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1631 		}
   1632 		if (__predict_false(pad)) {
   1633 			bus_addr_t paddaddr;
   1634 
   1635 			d = &sc->re_ldata.re_tx_list[curdesc];
   1636 			d->re_vlanctl = htole32(vlanctl);
   1637 			paddaddr = RE_TXPADDADDR(sc);
   1638 			re_set_bufaddr(d, paddaddr);
   1639 			cmdstat = re_flags |
   1640 			    RE_TDESC_CMD_OWN | RE_TDESC_CMD_EOF |
   1641 			    (RE_IP4CSUMTX_PADLEN + 1 - m->m_pkthdr.len);
   1642 			if (curdesc == (RE_TX_DESC_CNT(sc) - 1))
   1643 				cmdstat |= RE_TDESC_CMD_EOR;
   1644 			d->re_cmdstat = htole32(cmdstat);
   1645 			RE_TXDESCSYNC(sc, curdesc,
   1646 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1647 			lastdesc = curdesc;
   1648 			curdesc = RE_NEXT_TX_DESC(sc, curdesc);
   1649 		}
   1650 		KASSERT(lastdesc != -1);
   1651 
   1652 		/* Transfer ownership of packet to the chip. */
   1653 
   1654 		sc->re_ldata.re_tx_list[startdesc].re_cmdstat |=
   1655 		    htole32(RE_TDESC_CMD_OWN);
   1656 		RE_TXDESCSYNC(sc, startdesc,
   1657 		    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
   1658 
   1659 		/* update info of TX queue and descriptors */
   1660 		txq->txq_mbuf = m;
   1661 		txq->txq_descidx = lastdesc;
   1662 		txq->txq_nsegs = nsegs;
   1663 
   1664 		sc->re_ldata.re_txq_free--;
   1665 		sc->re_ldata.re_tx_free -= nsegs;
   1666 		sc->re_ldata.re_tx_nextfree = curdesc;
   1667 
   1668 #if NBPFILTER > 0
   1669 		/*
   1670 		 * If there's a BPF listener, bounce a copy of this frame
   1671 		 * to him.
   1672 		 */
   1673 		if (ifp->if_bpf)
   1674 			bpf_mtap(ifp->if_bpf, m);
   1675 #endif
   1676 	}
   1677 
   1678 	if (sc->re_ldata.re_txq_free < ofree) {
   1679 		/*
   1680 		 * TX packets are enqueued.
   1681 		 */
   1682 		sc->re_ldata.re_txq_prodidx = idx;
   1683 
   1684 		/*
   1685 		 * Start the transmitter to poll.
   1686 		 *
   1687 		 * RealTek put the TX poll request register in a different
   1688 		 * location on the 8169 gigE chip. I don't know why.
   1689 		 */
   1690 		if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0)
   1691 			CSR_WRITE_1(sc, RTK_TXSTART, RTK_TXSTART_START);
   1692 		else
   1693 			CSR_WRITE_1(sc, RTK_GTXSTART, RTK_TXSTART_START);
   1694 
   1695 		/*
   1696 		 * Use the countdown timer for interrupt moderation.
   1697 		 * 'TX done' interrupts are disabled. Instead, we reset the
   1698 		 * countdown timer, which will begin counting until it hits
   1699 		 * the value in the TIMERINT register, and then trigger an
   1700 		 * interrupt. Each time we write to the TIMERCNT register,
   1701 		 * the timer count is reset to 0.
   1702 		 */
   1703 		CSR_WRITE_4(sc, RTK_TIMERCNT, 1);
   1704 
   1705 		/*
   1706 		 * Set a timeout in case the chip goes out to lunch.
   1707 		 */
   1708 		ifp->if_timer = 5;
   1709 	}
   1710 }
   1711 
   1712 static int
   1713 re_init(struct ifnet *ifp)
   1714 {
   1715 	struct rtk_softc *sc = ifp->if_softc;
   1716 	const uint8_t *enaddr;
   1717 	uint32_t rxcfg = 0;
   1718 	uint32_t reg;
   1719 	int error;
   1720 
   1721 	if ((error = re_enable(sc)) != 0)
   1722 		goto out;
   1723 
   1724 	/*
   1725 	 * Cancel pending I/O and free all RX/TX buffers.
   1726 	 */
   1727 	re_stop(ifp, 0);
   1728 
   1729 	re_reset(sc);
   1730 
   1731 	/*
   1732 	 * Enable C+ RX and TX mode, as well as VLAN stripping and
   1733 	 * RX checksum offload. We must configure the C+ register
   1734 	 * before all others.
   1735 	 */
   1736 	reg = 0;
   1737 
   1738 	/*
   1739 	 * XXX: Realtek docs say bits 0 and 1 are reserved, for 8169S/8110S.
   1740 	 * FreeBSD  drivers set these bits anyway (for 8139C+?).
   1741 	 * So far, it works.
   1742 	 */
   1743 
   1744 	/*
   1745 	 * XXX: For old 8169 set bit 14.
   1746 	 *      For 8169S/8110S and above, do not set bit 14.
   1747 	 */
   1748 	if ((sc->sc_quirk & RTKQ_8169NONS) != 0)
   1749 		reg |= (0x1 << 14) | RTK_CPLUSCMD_PCI_MRW;
   1750 
   1751 	if (1)  {/* not for 8169S ? */
   1752 		reg |=
   1753 		    RTK_CPLUSCMD_VLANSTRIP |
   1754 		    (ifp->if_capenable &
   1755 		    (IFCAP_CSUM_IPv4_Rx | IFCAP_CSUM_TCPv4_Rx |
   1756 		     IFCAP_CSUM_UDPv4_Rx) ?
   1757 		    RTK_CPLUSCMD_RXCSUM_ENB : 0);
   1758 	}
   1759 
   1760 	CSR_WRITE_2(sc, RTK_CPLUS_CMD,
   1761 	    reg | RTK_CPLUSCMD_RXENB | RTK_CPLUSCMD_TXENB);
   1762 
   1763 	/* XXX: from Realtek-supplied Linux driver. Wholly undocumented. */
   1764 	if ((sc->sc_quirk & RTKQ_8139CPLUS) == 0)
   1765 		CSR_WRITE_2(sc, RTK_IM, 0x0000);
   1766 
   1767 	DELAY(10000);
   1768 
   1769 	/*
   1770 	 * Init our MAC address.  Even though the chipset
   1771 	 * documentation doesn't mention it, we need to enter "Config
   1772 	 * register write enable" mode to modify the ID registers.
   1773 	 */
   1774 	CSR_WRITE_1(sc, RTK_EECMD, RTK_EEMODE_WRITECFG);
   1775 	enaddr = CLLADDR(ifp->if_sadl);
   1776 	reg = enaddr[0] | (enaddr[1] << 8) |
   1777 	    (enaddr[2] << 16) | (enaddr[3] << 24);
   1778 	CSR_WRITE_4(sc, RTK_IDR0, reg);
   1779 	reg = enaddr[4] | (enaddr[5] << 8);
   1780 	CSR_WRITE_4(sc, RTK_IDR4, reg);
   1781 	CSR_WRITE_1(sc, RTK_EECMD, RTK_EEMODE_OFF);
   1782 
   1783 	/*
   1784 	 * For C+ mode, initialize the RX descriptors and mbufs.
   1785 	 */
   1786 	re_rx_list_init(sc);
   1787 	re_tx_list_init(sc);
   1788 
   1789 	/*
   1790 	 * Load the addresses of the RX and TX lists into the chip.
   1791 	 */
   1792 	CSR_WRITE_4(sc, RTK_RXLIST_ADDR_HI,
   1793 	    RE_ADDR_HI(sc->re_ldata.re_rx_list_map->dm_segs[0].ds_addr));
   1794 	CSR_WRITE_4(sc, RTK_RXLIST_ADDR_LO,
   1795 	    RE_ADDR_LO(sc->re_ldata.re_rx_list_map->dm_segs[0].ds_addr));
   1796 
   1797 	CSR_WRITE_4(sc, RTK_TXLIST_ADDR_HI,
   1798 	    RE_ADDR_HI(sc->re_ldata.re_tx_list_map->dm_segs[0].ds_addr));
   1799 	CSR_WRITE_4(sc, RTK_TXLIST_ADDR_LO,
   1800 	    RE_ADDR_LO(sc->re_ldata.re_tx_list_map->dm_segs[0].ds_addr));
   1801 
   1802 	/*
   1803 	 * Enable transmit and receive.
   1804 	 */
   1805 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_TX_ENB | RTK_CMD_RX_ENB);
   1806 
   1807 	/*
   1808 	 * Set the initial TX and RX configuration.
   1809 	 */
   1810 	if (sc->re_testmode && (sc->sc_quirk & RTKQ_8169NONS) != 0) {
   1811 		/* test mode is needed only for old 8169 */
   1812 		CSR_WRITE_4(sc, RTK_TXCFG,
   1813 		    RE_TXCFG_CONFIG | RTK_LOOPTEST_ON);
   1814 	} else
   1815 		CSR_WRITE_4(sc, RTK_TXCFG, RE_TXCFG_CONFIG);
   1816 
   1817 	CSR_WRITE_1(sc, RTK_EARLY_TX_THRESH, 16);
   1818 
   1819 	CSR_WRITE_4(sc, RTK_RXCFG, RE_RXCFG_CONFIG);
   1820 
   1821 	/* Set the individual bit to receive frames for this host only. */
   1822 	rxcfg = CSR_READ_4(sc, RTK_RXCFG);
   1823 	rxcfg |= RTK_RXCFG_RX_INDIV;
   1824 
   1825 	/* If we want promiscuous mode, set the allframes bit. */
   1826 	if (ifp->if_flags & IFF_PROMISC)
   1827 		rxcfg |= RTK_RXCFG_RX_ALLPHYS;
   1828 	else
   1829 		rxcfg &= ~RTK_RXCFG_RX_ALLPHYS;
   1830 	CSR_WRITE_4(sc, RTK_RXCFG, rxcfg);
   1831 
   1832 	/*
   1833 	 * Set capture broadcast bit to capture broadcast frames.
   1834 	 */
   1835 	if (ifp->if_flags & IFF_BROADCAST)
   1836 		rxcfg |= RTK_RXCFG_RX_BROAD;
   1837 	else
   1838 		rxcfg &= ~RTK_RXCFG_RX_BROAD;
   1839 	CSR_WRITE_4(sc, RTK_RXCFG, rxcfg);
   1840 
   1841 	/*
   1842 	 * Program the multicast filter, if necessary.
   1843 	 */
   1844 	rtk_setmulti(sc);
   1845 
   1846 	/*
   1847 	 * Enable interrupts.
   1848 	 */
   1849 	if (sc->re_testmode)
   1850 		CSR_WRITE_2(sc, RTK_IMR, 0);
   1851 	else
   1852 		CSR_WRITE_2(sc, RTK_IMR, RTK_INTRS_CPLUS);
   1853 
   1854 	/* Start RX/TX process. */
   1855 	CSR_WRITE_4(sc, RTK_MISSEDPKT, 0);
   1856 #ifdef notdef
   1857 	/* Enable receiver and transmitter. */
   1858 	CSR_WRITE_1(sc, RTK_COMMAND, RTK_CMD_TX_ENB | RTK_CMD_RX_ENB);
   1859 #endif
   1860 
   1861 	/*
   1862 	 * Initialize the timer interrupt register so that
   1863 	 * a timer interrupt will be generated once the timer
   1864 	 * reaches a certain number of ticks. The timer is
   1865 	 * reloaded on each transmit. This gives us TX interrupt
   1866 	 * moderation, which dramatically improves TX frame rate.
   1867 	 */
   1868 
   1869 	if ((sc->sc_quirk & RTKQ_8139CPLUS) != 0)
   1870 		CSR_WRITE_4(sc, RTK_TIMERINT, 0x400);
   1871 	else {
   1872 		CSR_WRITE_4(sc, RTK_TIMERINT_8169, 0x800);
   1873 
   1874 		/*
   1875 		 * For 8169 gigE NICs, set the max allowed RX packet
   1876 		 * size so we can receive jumbo frames.
   1877 		 */
   1878 		CSR_WRITE_2(sc, RTK_MAXRXPKTLEN, 16383);
   1879 	}
   1880 
   1881 	if (sc->re_testmode)
   1882 		return 0;
   1883 
   1884 	CSR_WRITE_1(sc, RTK_CFG1, RTK_CFG1_DRVLOAD);
   1885 
   1886 	ifp->if_flags |= IFF_RUNNING;
   1887 	ifp->if_flags &= ~IFF_OACTIVE;
   1888 
   1889 	callout_reset(&sc->rtk_tick_ch, hz, re_tick, sc);
   1890 
   1891  out:
   1892 	if (error) {
   1893 		ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
   1894 		ifp->if_timer = 0;
   1895 		printf("%s: interface not running\n",
   1896 		    device_xname(sc->sc_dev));
   1897 	}
   1898 
   1899 	return error;
   1900 }
   1901 
   1902 static int
   1903 re_ioctl(struct ifnet *ifp, u_long command, void *data)
   1904 {
   1905 	struct rtk_softc *sc = ifp->if_softc;
   1906 	struct ifreq *ifr = data;
   1907 	int s, error = 0;
   1908 
   1909 	s = splnet();
   1910 
   1911 	switch (command) {
   1912 	case SIOCSIFMTU:
   1913 		/*
   1914 		 * Disable jumbo frames if it's not supported.
   1915 		 */
   1916 		if ((sc->sc_quirk & RTKQ_NOJUMBO) != 0 &&
   1917 		    ifr->ifr_mtu > ETHERMTU) {
   1918 			error = EINVAL;
   1919 			break;
   1920 		}
   1921 
   1922 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > ETHERMTU_JUMBO)
   1923 			error = EINVAL;
   1924 		else if ((error = ifioctl_common(ifp, command, data)) ==
   1925 		    ENETRESET)
   1926 			error = 0;
   1927 		break;
   1928 	default:
   1929 		if ((error = ether_ioctl(ifp, command, data)) != ENETRESET)
   1930 			break;
   1931 
   1932 		error = 0;
   1933 
   1934 		if (command == SIOCSIFCAP)
   1935 			error = (*ifp->if_init)(ifp);
   1936 		else if (command != SIOCADDMULTI && command != SIOCDELMULTI)
   1937 			;
   1938 		else if (ifp->if_flags & IFF_RUNNING)
   1939 			rtk_setmulti(sc);
   1940 		break;
   1941 	}
   1942 
   1943 	splx(s);
   1944 
   1945 	return error;
   1946 }
   1947 
   1948 static void
   1949 re_watchdog(struct ifnet *ifp)
   1950 {
   1951 	struct rtk_softc *sc;
   1952 	int s;
   1953 
   1954 	sc = ifp->if_softc;
   1955 	s = splnet();
   1956 	printf("%s: watchdog timeout\n", device_xname(sc->sc_dev));
   1957 	ifp->if_oerrors++;
   1958 
   1959 	re_txeof(sc);
   1960 	re_rxeof(sc);
   1961 
   1962 	re_init(ifp);
   1963 
   1964 	splx(s);
   1965 }
   1966 
   1967 /*
   1968  * Stop the adapter and free any mbufs allocated to the
   1969  * RX and TX lists.
   1970  */
   1971 static void
   1972 re_stop(struct ifnet *ifp, int disable)
   1973 {
   1974 	int i;
   1975 	struct rtk_softc *sc = ifp->if_softc;
   1976 
   1977 	callout_stop(&sc->rtk_tick_ch);
   1978 
   1979 	mii_down(&sc->mii);
   1980 
   1981 	CSR_WRITE_1(sc, RTK_COMMAND, 0x00);
   1982 	CSR_WRITE_2(sc, RTK_IMR, 0x0000);
   1983 
   1984 	if (sc->re_head != NULL) {
   1985 		m_freem(sc->re_head);
   1986 		sc->re_head = sc->re_tail = NULL;
   1987 	}
   1988 
   1989 	/* Free the TX list buffers. */
   1990 	for (i = 0; i < RE_TX_QLEN; i++) {
   1991 		if (sc->re_ldata.re_txq[i].txq_mbuf != NULL) {
   1992 			bus_dmamap_unload(sc->sc_dmat,
   1993 			    sc->re_ldata.re_txq[i].txq_dmamap);
   1994 			m_freem(sc->re_ldata.re_txq[i].txq_mbuf);
   1995 			sc->re_ldata.re_txq[i].txq_mbuf = NULL;
   1996 		}
   1997 	}
   1998 
   1999 	/* Free the RX list buffers. */
   2000 	for (i = 0; i < RE_RX_DESC_CNT; i++) {
   2001 		if (sc->re_ldata.re_rxsoft[i].rxs_mbuf != NULL) {
   2002 			bus_dmamap_unload(sc->sc_dmat,
   2003 			    sc->re_ldata.re_rxsoft[i].rxs_dmamap);
   2004 			m_freem(sc->re_ldata.re_rxsoft[i].rxs_mbuf);
   2005 			sc->re_ldata.re_rxsoft[i].rxs_mbuf = NULL;
   2006 		}
   2007 	}
   2008 
   2009 	if (disable)
   2010 		re_disable(sc);
   2011 
   2012 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
   2013 	ifp->if_timer = 0;
   2014 }
   2015