xref: /dragonfly/sys/dev/netif/ale/if_ale.c (revision a68e0df0)
1 /*-
2  * Copyright (c) 2008, Pyun YongHyeon <yongari@FreeBSD.org>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice unmodified, this list of conditions, and the following
10  *    disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  *
27  * $FreeBSD: src/sys/dev/ale/if_ale.c,v 1.3 2008/12/03 09:01:12 yongari Exp $
28  */
29 
30 /* Driver for Atheros AR8121/AR8113/AR8114 PCIe Ethernet. */
31 
32 #include <sys/param.h>
33 #include <sys/endian.h>
34 #include <sys/kernel.h>
35 #include <sys/bus.h>
36 #include <sys/interrupt.h>
37 #include <sys/malloc.h>
38 #include <sys/proc.h>
39 #include <sys/rman.h>
40 #include <sys/serialize.h>
41 #include <sys/socket.h>
42 #include <sys/sockio.h>
43 #include <sys/sysctl.h>
44 
45 #include <net/ethernet.h>
46 #include <net/if.h>
47 #include <net/bpf.h>
48 #include <net/if_arp.h>
49 #include <net/if_dl.h>
50 #include <net/if_llc.h>
51 #include <net/if_media.h>
52 #include <net/ifq_var.h>
53 #include <net/vlan/if_vlan_var.h>
54 #include <net/vlan/if_vlan_ether.h>
55 
56 #include <netinet/ip.h>
57 
58 #include <dev/netif/mii_layer/mii.h>
59 #include <dev/netif/mii_layer/miivar.h>
60 
61 #include <bus/pci/pcireg.h>
62 #include <bus/pci/pcivar.h>
63 #include <bus/pci/pcidevs.h>
64 
65 #include <dev/netif/ale/if_alereg.h>
66 #include <dev/netif/ale/if_alevar.h>
67 
68 /* "device miibus" required.  See GENERIC if you get errors here. */
69 #include "miibus_if.h"
70 
71 /* For more information about Tx checksum offload issues see ale_encap(). */
72 #define	ALE_CSUM_FEATURES	(CSUM_TCP | CSUM_UDP)
73 
74 struct ale_dmamap_ctx {
75 	int			nsegs;
76 	bus_dma_segment_t	*segs;
77 };
78 
79 static int	ale_probe(device_t);
80 static int	ale_attach(device_t);
81 static int	ale_detach(device_t);
82 static int	ale_shutdown(device_t);
83 static int	ale_suspend(device_t);
84 static int	ale_resume(device_t);
85 
86 static int	ale_miibus_readreg(device_t, int, int);
87 static int	ale_miibus_writereg(device_t, int, int, int);
88 static void	ale_miibus_statchg(device_t);
89 
90 static void	ale_init(void *);
91 static void	ale_start(struct ifnet *);
92 static int	ale_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
93 static void	ale_watchdog(struct ifnet *);
94 static int	ale_mediachange(struct ifnet *);
95 static void	ale_mediastatus(struct ifnet *, struct ifmediareq *);
96 
97 static void	ale_intr(void *);
98 static int	ale_rxeof(struct ale_softc *sc);
99 static void	ale_rx_update_page(struct ale_softc *, struct ale_rx_page **,
100 		    uint32_t, uint32_t *);
101 static void	ale_rxcsum(struct ale_softc *, struct mbuf *, uint32_t);
102 static void	ale_txeof(struct ale_softc *);
103 
104 static int	ale_dma_alloc(struct ale_softc *);
105 static void	ale_dma_free(struct ale_softc *);
106 static int	ale_check_boundary(struct ale_softc *);
107 static void	ale_dmamap_cb(void *, bus_dma_segment_t *, int, int);
108 static void	ale_dmamap_buf_cb(void *, bus_dma_segment_t *, int,
109 		    bus_size_t, int);
110 static int	ale_encap(struct ale_softc *, struct mbuf **);
111 static void	ale_init_rx_pages(struct ale_softc *);
112 static void	ale_init_tx_ring(struct ale_softc *);
113 
114 static void	ale_stop(struct ale_softc *);
115 static void	ale_tick(void *);
116 static void	ale_get_macaddr(struct ale_softc *);
117 static void	ale_mac_config(struct ale_softc *);
118 static void	ale_phy_reset(struct ale_softc *);
119 static void	ale_reset(struct ale_softc *);
120 static void	ale_rxfilter(struct ale_softc *);
121 static void	ale_rxvlan(struct ale_softc *);
122 static void	ale_stats_clear(struct ale_softc *);
123 static void	ale_stats_update(struct ale_softc *);
124 static void	ale_stop_mac(struct ale_softc *);
125 #ifdef notyet
126 static void	ale_setlinkspeed(struct ale_softc *);
127 static void	ale_setwol(struct ale_softc *);
128 #endif
129 
130 static void	ale_sysctl_node(struct ale_softc *);
131 static int	sysctl_hw_ale_int_mod(SYSCTL_HANDLER_ARGS);
132 
133 /*
134  * Devices supported by this driver.
135  */
136 static struct ale_dev {
137 	uint16_t	ale_vendorid;
138 	uint16_t	ale_deviceid;
139 	const char	*ale_name;
140 } ale_devs[] = {
141     { VENDORID_ATHEROS, DEVICEID_ATHEROS_AR81XX,
142     "Atheros AR8121/AR8113/AR8114 PCIe Ethernet" },
143 };
144 
145 static device_method_t ale_methods[] = {
146 	/* Device interface. */
147 	DEVMETHOD(device_probe,		ale_probe),
148 	DEVMETHOD(device_attach,	ale_attach),
149 	DEVMETHOD(device_detach,	ale_detach),
150 	DEVMETHOD(device_shutdown,	ale_shutdown),
151 	DEVMETHOD(device_suspend,	ale_suspend),
152 	DEVMETHOD(device_resume,	ale_resume),
153 
154 	/* Bus interface. */
155 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
156 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
157 
158 	/* MII interface. */
159 	DEVMETHOD(miibus_readreg,	ale_miibus_readreg),
160 	DEVMETHOD(miibus_writereg,	ale_miibus_writereg),
161 	DEVMETHOD(miibus_statchg,	ale_miibus_statchg),
162 
163 	{ NULL, NULL }
164 };
165 
166 static driver_t ale_driver = {
167 	"ale",
168 	ale_methods,
169 	sizeof(struct ale_softc)
170 };
171 
172 static devclass_t ale_devclass;
173 
174 DECLARE_DUMMY_MODULE(if_ale);
175 MODULE_DEPEND(if_ale, miibus, 1, 1, 1);
176 DRIVER_MODULE(if_ale, pci, ale_driver, ale_devclass, 0, 0);
177 DRIVER_MODULE(miibus, ale, miibus_driver, miibus_devclass, 0, 0);
178 
179 static int
180 ale_miibus_readreg(device_t dev, int phy, int reg)
181 {
182 	struct ale_softc *sc;
183 	uint32_t v;
184 	int i;
185 
186 	sc = device_get_softc(dev);
187 
188 	if (phy != sc->ale_phyaddr)
189 		return (0);
190 
191 	if (sc->ale_flags & ALE_FLAG_FASTETHER) {
192 		if (reg == MII_100T2CR || reg == MII_100T2SR ||
193 		    reg == MII_EXTSR)
194 			return (0);
195 	}
196 
197 	CSR_WRITE_4(sc, ALE_MDIO, MDIO_OP_EXECUTE | MDIO_OP_READ |
198 	    MDIO_SUP_PREAMBLE | MDIO_CLK_25_4 | MDIO_REG_ADDR(reg));
199 	for (i = ALE_PHY_TIMEOUT; i > 0; i--) {
200 		DELAY(5);
201 		v = CSR_READ_4(sc, ALE_MDIO);
202 		if ((v & (MDIO_OP_EXECUTE | MDIO_OP_BUSY)) == 0)
203 			break;
204 	}
205 
206 	if (i == 0) {
207 		device_printf(sc->ale_dev, "phy read timeout : %d\n", reg);
208 		return (0);
209 	}
210 
211 	return ((v & MDIO_DATA_MASK) >> MDIO_DATA_SHIFT);
212 }
213 
214 static int
215 ale_miibus_writereg(device_t dev, int phy, int reg, int val)
216 {
217 	struct ale_softc *sc;
218 	uint32_t v;
219 	int i;
220 
221 	sc = device_get_softc(dev);
222 
223 	if (phy != sc->ale_phyaddr)
224 		return (0);
225 
226 	if (sc->ale_flags & ALE_FLAG_FASTETHER) {
227 		if (reg == MII_100T2CR || reg == MII_100T2SR ||
228 		    reg == MII_EXTSR)
229 			return (0);
230 	}
231 
232 	CSR_WRITE_4(sc, ALE_MDIO, MDIO_OP_EXECUTE | MDIO_OP_WRITE |
233 	    (val & MDIO_DATA_MASK) << MDIO_DATA_SHIFT |
234 	    MDIO_SUP_PREAMBLE | MDIO_CLK_25_4 | MDIO_REG_ADDR(reg));
235 	for (i = ALE_PHY_TIMEOUT; i > 0; i--) {
236 		DELAY(5);
237 		v = CSR_READ_4(sc, ALE_MDIO);
238 		if ((v & (MDIO_OP_EXECUTE | MDIO_OP_BUSY)) == 0)
239 			break;
240 	}
241 
242 	if (i == 0)
243 		device_printf(sc->ale_dev, "phy write timeout : %d\n", reg);
244 
245 	return (0);
246 }
247 
248 static void
249 ale_miibus_statchg(device_t dev)
250 {
251 	struct ale_softc *sc = device_get_softc(dev);
252 	struct ifnet *ifp = &sc->arpcom.ac_if;
253 	struct mii_data *mii;
254 	uint32_t reg;
255 
256 	ASSERT_SERIALIZED(ifp->if_serializer);
257 
258 	if ((ifp->if_flags & IFF_RUNNING) == 0)
259 		return;
260 
261 	mii = device_get_softc(sc->ale_miibus);
262 
263 	sc->ale_flags &= ~ALE_FLAG_LINK;
264 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
265 	    (IFM_ACTIVE | IFM_AVALID)) {
266 		switch (IFM_SUBTYPE(mii->mii_media_active)) {
267 		case IFM_10_T:
268 		case IFM_100_TX:
269 			sc->ale_flags |= ALE_FLAG_LINK;
270 			break;
271 
272 		case IFM_1000_T:
273 			if ((sc->ale_flags & ALE_FLAG_FASTETHER) == 0)
274 				sc->ale_flags |= ALE_FLAG_LINK;
275 			break;
276 
277 		default:
278 			break;
279 		}
280 	}
281 
282 	/* Stop Rx/Tx MACs. */
283 	ale_stop_mac(sc);
284 
285 	/* Program MACs with resolved speed/duplex/flow-control. */
286 	if ((sc->ale_flags & ALE_FLAG_LINK) != 0) {
287 		ale_mac_config(sc);
288 		/* Reenable Tx/Rx MACs. */
289 		reg = CSR_READ_4(sc, ALE_MAC_CFG);
290 		reg |= MAC_CFG_TX_ENB | MAC_CFG_RX_ENB;
291 		CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
292 	}
293 }
294 
295 static void
296 ale_mediastatus(struct ifnet *ifp, struct ifmediareq *ifmr)
297 {
298 	struct ale_softc *sc = ifp->if_softc;
299 	struct mii_data *mii = device_get_softc(sc->ale_miibus);
300 
301 	ASSERT_SERIALIZED(ifp->if_serializer);
302 
303 	mii_pollstat(mii);
304 	ifmr->ifm_status = mii->mii_media_status;
305 	ifmr->ifm_active = mii->mii_media_active;
306 }
307 
308 static int
309 ale_mediachange(struct ifnet *ifp)
310 {
311 	struct ale_softc *sc = ifp->if_softc;
312 	struct mii_data *mii = device_get_softc(sc->ale_miibus);
313 	int error;
314 
315 	ASSERT_SERIALIZED(ifp->if_serializer);
316 
317 	if (mii->mii_instance != 0) {
318 		struct mii_softc *miisc;
319 
320 		LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
321 			mii_phy_reset(miisc);
322 	}
323 	error = mii_mediachg(mii);
324 
325 	return (error);
326 }
327 
328 static int
329 ale_probe(device_t dev)
330 {
331 	struct ale_dev *sp;
332 	int i;
333 	uint16_t vendor, devid;
334 
335 	vendor = pci_get_vendor(dev);
336 	devid = pci_get_device(dev);
337 	sp = ale_devs;
338 	for (i = 0; i < sizeof(ale_devs) / sizeof(ale_devs[0]); i++) {
339 		if (vendor == sp->ale_vendorid &&
340 		    devid == sp->ale_deviceid) {
341 			device_set_desc(dev, sp->ale_name);
342 			return (0);
343 		}
344 		sp++;
345 	}
346 
347 	return (ENXIO);
348 }
349 
350 static void
351 ale_get_macaddr(struct ale_softc *sc)
352 {
353 	uint32_t ea[2], reg;
354 	int i, vpdc;
355 
356 	reg = CSR_READ_4(sc, ALE_SPI_CTRL);
357 	if ((reg & SPI_VPD_ENB) != 0) {
358 		reg &= ~SPI_VPD_ENB;
359 		CSR_WRITE_4(sc, ALE_SPI_CTRL, reg);
360 	}
361 
362 	vpdc = pci_get_vpdcap_ptr(sc->ale_dev);
363 	if (vpdc) {
364 		/*
365 		 * PCI VPD capability found, let TWSI reload EEPROM.
366 		 * This will set ethernet address of controller.
367 		 */
368 		CSR_WRITE_4(sc, ALE_TWSI_CTRL, CSR_READ_4(sc, ALE_TWSI_CTRL) |
369 		    TWSI_CTRL_SW_LD_START);
370 		for (i = 100; i > 0; i--) {
371 			DELAY(1000);
372 			reg = CSR_READ_4(sc, ALE_TWSI_CTRL);
373 			if ((reg & TWSI_CTRL_SW_LD_START) == 0)
374 				break;
375 		}
376 		if (i == 0)
377 			device_printf(sc->ale_dev,
378 			    "reloading EEPROM timeout!\n");
379 	} else {
380 		if (bootverbose)
381 			device_printf(sc->ale_dev,
382 			    "PCI VPD capability not found!\n");
383 	}
384 
385 	ea[0] = CSR_READ_4(sc, ALE_PAR0);
386 	ea[1] = CSR_READ_4(sc, ALE_PAR1);
387 	sc->ale_eaddr[0] = (ea[1] >> 8) & 0xFF;
388 	sc->ale_eaddr[1] = (ea[1] >> 0) & 0xFF;
389 	sc->ale_eaddr[2] = (ea[0] >> 24) & 0xFF;
390 	sc->ale_eaddr[3] = (ea[0] >> 16) & 0xFF;
391 	sc->ale_eaddr[4] = (ea[0] >> 8) & 0xFF;
392 	sc->ale_eaddr[5] = (ea[0] >> 0) & 0xFF;
393 }
394 
395 static void
396 ale_phy_reset(struct ale_softc *sc)
397 {
398 	/* Reset magic from Linux. */
399 	CSR_WRITE_2(sc, ALE_GPHY_CTRL,
400 	    GPHY_CTRL_HIB_EN | GPHY_CTRL_HIB_PULSE | GPHY_CTRL_SEL_ANA_RESET |
401 	    GPHY_CTRL_PHY_PLL_ON);
402 	DELAY(1000);
403 	CSR_WRITE_2(sc, ALE_GPHY_CTRL,
404 	    GPHY_CTRL_EXT_RESET | GPHY_CTRL_HIB_EN | GPHY_CTRL_HIB_PULSE |
405 	    GPHY_CTRL_SEL_ANA_RESET | GPHY_CTRL_PHY_PLL_ON);
406 	DELAY(1000);
407 
408 #define	ATPHY_DBG_ADDR		0x1D
409 #define	ATPHY_DBG_DATA		0x1E
410 
411 	/* Enable hibernation mode. */
412 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
413 	    ATPHY_DBG_ADDR, 0x0B);
414 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
415 	    ATPHY_DBG_DATA, 0xBC00);
416 	/* Set Class A/B for all modes. */
417 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
418 	    ATPHY_DBG_ADDR, 0x00);
419 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
420 	    ATPHY_DBG_DATA, 0x02EF);
421 	/* Enable 10BT power saving. */
422 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
423 	    ATPHY_DBG_ADDR, 0x12);
424 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
425 	    ATPHY_DBG_DATA, 0x4C04);
426 	/* Adjust 1000T power. */
427 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
428 	    ATPHY_DBG_ADDR, 0x04);
429 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
430 	    ATPHY_DBG_ADDR, 0x8BBB);
431 	/* 10BT center tap voltage. */
432 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
433 	    ATPHY_DBG_ADDR, 0x05);
434 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
435 	    ATPHY_DBG_ADDR, 0x2C46);
436 
437 #undef	ATPHY_DBG_ADDR
438 #undef	ATPHY_DBG_DATA
439 	DELAY(1000);
440 }
441 
442 static int
443 ale_attach(device_t dev)
444 {
445 	struct ale_softc *sc = device_get_softc(dev);
446 	struct ifnet *ifp = &sc->arpcom.ac_if;
447 	int error = 0;
448 	uint32_t rxf_len, txf_len;
449 	uint8_t pcie_ptr;
450 
451 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
452 	sc->ale_dev = dev;
453 
454 	callout_init(&sc->ale_tick_ch);
455 
456 #ifndef BURN_BRIDGES
457 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
458 		uint32_t irq, mem;
459 
460 		irq = pci_read_config(dev, PCIR_INTLINE, 4);
461 		mem = pci_read_config(dev, ALE_PCIR_BAR, 4);
462 
463 		device_printf(dev, "chip is in D%d power mode "
464 		    "-- setting to D0\n", pci_get_powerstate(dev));
465 
466 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
467 
468 		pci_write_config(dev, PCIR_INTLINE, irq, 4);
469 		pci_write_config(dev, ALE_PCIR_BAR, mem, 4);
470 	}
471 #endif	/* !BURN_BRIDGE */
472 
473 	/* Enable bus mastering */
474 	pci_enable_busmaster(dev);
475 
476 	/*
477 	 * Allocate memory mapped IO
478 	 */
479 	sc->ale_mem_rid = ALE_PCIR_BAR;
480 	sc->ale_mem_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
481 						 &sc->ale_mem_rid, RF_ACTIVE);
482 	if (sc->ale_mem_res == NULL) {
483 		device_printf(dev, "can't allocate IO memory\n");
484 		return ENXIO;
485 	}
486 	sc->ale_mem_bt = rman_get_bustag(sc->ale_mem_res);
487 	sc->ale_mem_bh = rman_get_bushandle(sc->ale_mem_res);
488 
489 	/*
490 	 * Allocate IRQ
491 	 */
492 	sc->ale_irq_rid = 0;
493 	sc->ale_irq_res = bus_alloc_resource_any(dev, SYS_RES_IRQ,
494 						 &sc->ale_irq_rid,
495 						 RF_SHAREABLE | RF_ACTIVE);
496 	if (sc->ale_irq_res == NULL) {
497 		device_printf(dev, "can't allocate irq\n");
498 		error = ENXIO;
499 		goto fail;
500 	}
501 
502 	/* Set PHY address. */
503 	sc->ale_phyaddr = ALE_PHY_ADDR;
504 
505 	/* Reset PHY. */
506 	ale_phy_reset(sc);
507 
508 	/* Reset the ethernet controller. */
509 	ale_reset(sc);
510 
511 	/* Get PCI and chip id/revision. */
512 	sc->ale_rev = pci_get_revid(dev);
513 	if (sc->ale_rev >= 0xF0) {
514 		/* L2E Rev. B. AR8114 */
515 		sc->ale_flags |= ALE_FLAG_FASTETHER;
516 	} else {
517 		if ((CSR_READ_4(sc, ALE_PHY_STATUS) & PHY_STATUS_100M) != 0) {
518 			/* L1E AR8121 */
519 			sc->ale_flags |= ALE_FLAG_JUMBO;
520 		} else {
521 			/* L2E Rev. A. AR8113 */
522 			sc->ale_flags |= ALE_FLAG_FASTETHER;
523 		}
524 	}
525 
526 	/*
527 	 * All known controllers seems to require 4 bytes alignment
528 	 * of Tx buffers to make Tx checksum offload with custom
529 	 * checksum generation method work.
530 	 */
531 	sc->ale_flags |= ALE_FLAG_TXCSUM_BUG;
532 
533 	/*
534 	 * All known controllers seems to have issues on Rx checksum
535 	 * offload for fragmented IP datagrams.
536 	 */
537 	sc->ale_flags |= ALE_FLAG_RXCSUM_BUG;
538 
539 	/*
540 	 * Don't use Tx CMB. It is known to cause RRS update failure
541 	 * under certain circumstances. Typical phenomenon of the
542 	 * issue would be unexpected sequence number encountered in
543 	 * Rx handler.
544 	 */
545 	sc->ale_flags |= ALE_FLAG_TXCMB_BUG;
546 	sc->ale_chip_rev = CSR_READ_4(sc, ALE_MASTER_CFG) >>
547 	    MASTER_CHIP_REV_SHIFT;
548 	if (bootverbose) {
549 		device_printf(dev, "PCI device revision : 0x%04x\n",
550 		    sc->ale_rev);
551 		device_printf(dev, "Chip id/revision : 0x%04x\n",
552 		    sc->ale_chip_rev);
553 	}
554 
555 	/*
556 	 * Uninitialized hardware returns an invalid chip id/revision
557 	 * as well as 0xFFFFFFFF for Tx/Rx fifo length.
558 	 */
559 	txf_len = CSR_READ_4(sc, ALE_SRAM_TX_FIFO_LEN);
560 	rxf_len = CSR_READ_4(sc, ALE_SRAM_RX_FIFO_LEN);
561 	if (sc->ale_chip_rev == 0xFFFF || txf_len == 0xFFFFFFFF ||
562 	    rxf_len == 0xFFFFFFF) {
563 		device_printf(dev,"chip revision : 0x%04x, %u Tx FIFO "
564 		    "%u Rx FIFO -- not initialized?\n", sc->ale_chip_rev,
565 		    txf_len, rxf_len);
566 		error = ENXIO;
567 		goto fail;
568 	}
569 	device_printf(dev, "%u Tx FIFO, %u Rx FIFO\n", txf_len, rxf_len);
570 
571 	/* Get DMA parameters from PCIe device control register. */
572 	pcie_ptr = pci_get_pciecap_ptr(dev);
573 	if (pcie_ptr) {
574 		uint16_t devctl;
575 
576 		sc->ale_flags |= ALE_FLAG_PCIE;
577 		devctl = pci_read_config(dev, pcie_ptr + PCIER_DEVCTRL, 2);
578 		/* Max read request size. */
579 		sc->ale_dma_rd_burst = ((devctl >> 12) & 0x07) <<
580 		    DMA_CFG_RD_BURST_SHIFT;
581 		/* Max payload size. */
582 		sc->ale_dma_wr_burst = ((devctl >> 5) & 0x07) <<
583 		    DMA_CFG_WR_BURST_SHIFT;
584 		if (bootverbose) {
585 			device_printf(dev, "Read request size : %d bytes.\n",
586 			    128 << ((devctl >> 12) & 0x07));
587 			device_printf(dev, "TLP payload size : %d bytes.\n",
588 			    128 << ((devctl >> 5) & 0x07));
589 		}
590 	} else {
591 		sc->ale_dma_rd_burst = DMA_CFG_RD_BURST_128;
592 		sc->ale_dma_wr_burst = DMA_CFG_WR_BURST_128;
593 	}
594 
595 	/* Create device sysctl node. */
596 	ale_sysctl_node(sc);
597 
598 	if ((error = ale_dma_alloc(sc) != 0))
599 		goto fail;
600 
601 	/* Load station address. */
602 	ale_get_macaddr(sc);
603 
604 	ifp->if_softc = sc;
605 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
606 	ifp->if_ioctl = ale_ioctl;
607 	ifp->if_start = ale_start;
608 	ifp->if_init = ale_init;
609 	ifp->if_watchdog = ale_watchdog;
610 	ifq_set_maxlen(&ifp->if_snd, ALE_TX_RING_CNT - 1);
611 	ifq_set_ready(&ifp->if_snd);
612 
613 	ifp->if_capabilities = IFCAP_RXCSUM |
614 			       IFCAP_VLAN_MTU |
615 			       IFCAP_VLAN_HWTAGGING;
616 #ifdef notyet
617 	ifp->if_capabilities |= IFCAP_TXCSUM;
618 	ifp->if_hwassist = ALE_CSUM_FEATURES;
619 #endif
620 	ifp->if_capenable = ifp->if_capabilities;
621 
622 	/* Set up MII bus. */
623 	if ((error = mii_phy_probe(dev, &sc->ale_miibus, ale_mediachange,
624 	    ale_mediastatus)) != 0) {
625 		device_printf(dev, "no PHY found!\n");
626 		goto fail;
627 	}
628 
629 	ether_ifattach(ifp, sc->ale_eaddr, NULL);
630 
631 	/* Tell the upper layer(s) we support long frames. */
632 	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
633 
634 	error = bus_setup_intr(dev, sc->ale_irq_res, INTR_MPSAFE, ale_intr, sc,
635 			       &sc->ale_irq_handle, ifp->if_serializer);
636 	if (error) {
637 		device_printf(dev, "could not set up interrupt handler.\n");
638 		ether_ifdetach(ifp);
639 		goto fail;
640 	}
641 
642 	ifp->if_cpuid = ithread_cpuid(rman_get_start(sc->ale_irq_res));
643 	KKASSERT(ifp->if_cpuid >= 0 && ifp->if_cpuid < ncpus);
644 	return 0;
645 fail:
646 	ale_detach(dev);
647 	return (error);
648 }
649 
650 static int
651 ale_detach(device_t dev)
652 {
653 	struct ale_softc *sc = device_get_softc(dev);
654 
655 	if (device_is_attached(dev)) {
656 		struct ifnet *ifp = &sc->arpcom.ac_if;
657 
658 		lwkt_serialize_enter(ifp->if_serializer);
659 		sc->ale_flags |= ALE_FLAG_DETACH;
660 		ale_stop(sc);
661 		bus_teardown_intr(dev, sc->ale_irq_res, sc->ale_irq_handle);
662 		lwkt_serialize_exit(ifp->if_serializer);
663 
664 		ether_ifdetach(ifp);
665 	}
666 
667 	if (sc->ale_sysctl_tree != NULL)
668 		sysctl_ctx_free(&sc->ale_sysctl_ctx);
669 
670 	if (sc->ale_miibus != NULL)
671 		device_delete_child(dev, sc->ale_miibus);
672 	bus_generic_detach(dev);
673 
674 	if (sc->ale_irq_res != NULL) {
675 		bus_release_resource(dev, SYS_RES_IRQ, sc->ale_irq_rid,
676 				     sc->ale_irq_res);
677 	}
678 	if (sc->ale_mem_res != NULL) {
679 		bus_release_resource(dev, SYS_RES_MEMORY, sc->ale_mem_rid,
680 				     sc->ale_mem_res);
681 	}
682 
683 	ale_dma_free(sc);
684 
685 	return (0);
686 }
687 
688 #define	ALE_SYSCTL_STAT_ADD32(c, h, n, p, d)	\
689 	    SYSCTL_ADD_UINT(c, h, OID_AUTO, n, CTLFLAG_RD, p, 0, d)
690 #define	ALE_SYSCTL_STAT_ADD64(c, h, n, p, d)	\
691 	    SYSCTL_ADD_QUAD(c, h, OID_AUTO, n, CTLFLAG_RD, p, 0, d)
692 
693 static void
694 ale_sysctl_node(struct ale_softc *sc)
695 {
696 	struct sysctl_ctx_list *ctx;
697 	struct sysctl_oid_list *child, *parent;
698 	struct sysctl_oid *tree;
699 	struct ale_hw_stats *stats;
700 	int error;
701 
702 	sysctl_ctx_init(&sc->ale_sysctl_ctx);
703 	sc->ale_sysctl_tree = SYSCTL_ADD_NODE(&sc->ale_sysctl_ctx,
704 				SYSCTL_STATIC_CHILDREN(_hw), OID_AUTO,
705 				device_get_nameunit(sc->ale_dev),
706 				CTLFLAG_RD, 0, "");
707 	if (sc->ale_sysctl_tree == NULL) {
708 		device_printf(sc->ale_dev, "can't add sysctl node\n");
709 		return;
710 	}
711 
712 	stats = &sc->ale_stats;
713 	ctx = &sc->ale_sysctl_ctx;
714 	child = SYSCTL_CHILDREN(sc->ale_sysctl_tree);
715 
716 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "int_rx_mod",
717 	    CTLTYPE_INT | CTLFLAG_RW, &sc->ale_int_rx_mod, 0,
718 	    sysctl_hw_ale_int_mod, "I", "ale Rx interrupt moderation");
719 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "int_tx_mod",
720 	    CTLTYPE_INT | CTLFLAG_RW, &sc->ale_int_tx_mod, 0,
721 	    sysctl_hw_ale_int_mod, "I", "ale Tx interrupt moderation");
722 
723 	/*
724 	 * Pull in device tunables.
725 	 */
726 	sc->ale_int_rx_mod = ALE_IM_RX_TIMER_DEFAULT;
727 	error = resource_int_value(device_get_name(sc->ale_dev),
728 	    device_get_unit(sc->ale_dev), "int_rx_mod", &sc->ale_int_rx_mod);
729 	if (error == 0) {
730 		if (sc->ale_int_rx_mod < ALE_IM_TIMER_MIN ||
731 		    sc->ale_int_rx_mod > ALE_IM_TIMER_MAX) {
732 			device_printf(sc->ale_dev, "int_rx_mod value out of "
733 			    "range; using default: %d\n",
734 			    ALE_IM_RX_TIMER_DEFAULT);
735 			sc->ale_int_rx_mod = ALE_IM_RX_TIMER_DEFAULT;
736 		}
737 	}
738 
739 	sc->ale_int_tx_mod = ALE_IM_TX_TIMER_DEFAULT;
740 	error = resource_int_value(device_get_name(sc->ale_dev),
741 	    device_get_unit(sc->ale_dev), "int_tx_mod", &sc->ale_int_tx_mod);
742 	if (error == 0) {
743 		if (sc->ale_int_tx_mod < ALE_IM_TIMER_MIN ||
744 		    sc->ale_int_tx_mod > ALE_IM_TIMER_MAX) {
745 			device_printf(sc->ale_dev, "int_tx_mod value out of "
746 			    "range; using default: %d\n",
747 			    ALE_IM_TX_TIMER_DEFAULT);
748 			sc->ale_int_tx_mod = ALE_IM_TX_TIMER_DEFAULT;
749 		}
750 	}
751 
752 	/* Misc statistics. */
753 	ALE_SYSCTL_STAT_ADD32(ctx, child, "reset_brk_seq",
754 	    &stats->reset_brk_seq,
755 	    "Controller resets due to broken Rx sequnce number");
756 
757 	tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "stats", CTLFLAG_RD,
758 	    NULL, "ATE statistics");
759 	parent = SYSCTL_CHILDREN(tree);
760 
761 	/* Rx statistics. */
762 	tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "rx", CTLFLAG_RD,
763 	    NULL, "Rx MAC statistics");
764 	child = SYSCTL_CHILDREN(tree);
765 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_frames",
766 	    &stats->rx_frames, "Good frames");
767 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_bcast_frames",
768 	    &stats->rx_bcast_frames, "Good broadcast frames");
769 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_mcast_frames",
770 	    &stats->rx_mcast_frames, "Good multicast frames");
771 	ALE_SYSCTL_STAT_ADD32(ctx, child, "pause_frames",
772 	    &stats->rx_pause_frames, "Pause control frames");
773 	ALE_SYSCTL_STAT_ADD32(ctx, child, "control_frames",
774 	    &stats->rx_control_frames, "Control frames");
775 	ALE_SYSCTL_STAT_ADD32(ctx, child, "crc_errs",
776 	    &stats->rx_crcerrs, "CRC errors");
777 	ALE_SYSCTL_STAT_ADD32(ctx, child, "len_errs",
778 	    &stats->rx_lenerrs, "Frames with length mismatched");
779 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_octets",
780 	    &stats->rx_bytes, "Good octets");
781 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_bcast_octets",
782 	    &stats->rx_bcast_bytes, "Good broadcast octets");
783 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_mcast_octets",
784 	    &stats->rx_mcast_bytes, "Good multicast octets");
785 	ALE_SYSCTL_STAT_ADD32(ctx, child, "runts",
786 	    &stats->rx_runts, "Too short frames");
787 	ALE_SYSCTL_STAT_ADD32(ctx, child, "fragments",
788 	    &stats->rx_fragments, "Fragmented frames");
789 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_64",
790 	    &stats->rx_pkts_64, "64 bytes frames");
791 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_65_127",
792 	    &stats->rx_pkts_65_127, "65 to 127 bytes frames");
793 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_128_255",
794 	    &stats->rx_pkts_128_255, "128 to 255 bytes frames");
795 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_256_511",
796 	    &stats->rx_pkts_256_511, "256 to 511 bytes frames");
797 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_512_1023",
798 	    &stats->rx_pkts_512_1023, "512 to 1023 bytes frames");
799 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1024_1518",
800 	    &stats->rx_pkts_1024_1518, "1024 to 1518 bytes frames");
801 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1519_max",
802 	    &stats->rx_pkts_1519_max, "1519 to max frames");
803 	ALE_SYSCTL_STAT_ADD32(ctx, child, "trunc_errs",
804 	    &stats->rx_pkts_truncated, "Truncated frames due to MTU size");
805 	ALE_SYSCTL_STAT_ADD32(ctx, child, "fifo_oflows",
806 	    &stats->rx_fifo_oflows, "FIFO overflows");
807 	ALE_SYSCTL_STAT_ADD32(ctx, child, "rrs_errs",
808 	    &stats->rx_rrs_errs, "Return status write-back errors");
809 	ALE_SYSCTL_STAT_ADD32(ctx, child, "align_errs",
810 	    &stats->rx_alignerrs, "Alignment errors");
811 	ALE_SYSCTL_STAT_ADD32(ctx, child, "filtered",
812 	    &stats->rx_pkts_filtered,
813 	    "Frames dropped due to address filtering");
814 
815 	/* Tx statistics. */
816 	tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "tx", CTLFLAG_RD,
817 	    NULL, "Tx MAC statistics");
818 	child = SYSCTL_CHILDREN(tree);
819 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_frames",
820 	    &stats->tx_frames, "Good frames");
821 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_bcast_frames",
822 	    &stats->tx_bcast_frames, "Good broadcast frames");
823 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_mcast_frames",
824 	    &stats->tx_mcast_frames, "Good multicast frames");
825 	ALE_SYSCTL_STAT_ADD32(ctx, child, "pause_frames",
826 	    &stats->tx_pause_frames, "Pause control frames");
827 	ALE_SYSCTL_STAT_ADD32(ctx, child, "control_frames",
828 	    &stats->tx_control_frames, "Control frames");
829 	ALE_SYSCTL_STAT_ADD32(ctx, child, "excess_defers",
830 	    &stats->tx_excess_defer, "Frames with excessive derferrals");
831 	ALE_SYSCTL_STAT_ADD32(ctx, child, "defers",
832 	    &stats->tx_excess_defer, "Frames with derferrals");
833 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_octets",
834 	    &stats->tx_bytes, "Good octets");
835 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_bcast_octets",
836 	    &stats->tx_bcast_bytes, "Good broadcast octets");
837 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_mcast_octets",
838 	    &stats->tx_mcast_bytes, "Good multicast octets");
839 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_64",
840 	    &stats->tx_pkts_64, "64 bytes frames");
841 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_65_127",
842 	    &stats->tx_pkts_65_127, "65 to 127 bytes frames");
843 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_128_255",
844 	    &stats->tx_pkts_128_255, "128 to 255 bytes frames");
845 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_256_511",
846 	    &stats->tx_pkts_256_511, "256 to 511 bytes frames");
847 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_512_1023",
848 	    &stats->tx_pkts_512_1023, "512 to 1023 bytes frames");
849 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1024_1518",
850 	    &stats->tx_pkts_1024_1518, "1024 to 1518 bytes frames");
851 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1519_max",
852 	    &stats->tx_pkts_1519_max, "1519 to max frames");
853 	ALE_SYSCTL_STAT_ADD32(ctx, child, "single_colls",
854 	    &stats->tx_single_colls, "Single collisions");
855 	ALE_SYSCTL_STAT_ADD32(ctx, child, "multi_colls",
856 	    &stats->tx_multi_colls, "Multiple collisions");
857 	ALE_SYSCTL_STAT_ADD32(ctx, child, "late_colls",
858 	    &stats->tx_late_colls, "Late collisions");
859 	ALE_SYSCTL_STAT_ADD32(ctx, child, "excess_colls",
860 	    &stats->tx_excess_colls, "Excessive collisions");
861 	ALE_SYSCTL_STAT_ADD32(ctx, child, "abort",
862 	    &stats->tx_abort, "Aborted frames due to Excessive collisions");
863 	ALE_SYSCTL_STAT_ADD32(ctx, child, "underruns",
864 	    &stats->tx_underrun, "FIFO underruns");
865 	ALE_SYSCTL_STAT_ADD32(ctx, child, "desc_underruns",
866 	    &stats->tx_desc_underrun, "Descriptor write-back errors");
867 	ALE_SYSCTL_STAT_ADD32(ctx, child, "len_errs",
868 	    &stats->tx_lenerrs, "Frames with length mismatched");
869 	ALE_SYSCTL_STAT_ADD32(ctx, child, "trunc_errs",
870 	    &stats->tx_pkts_truncated, "Truncated frames due to MTU size");
871 }
872 
873 #undef ALE_SYSCTL_STAT_ADD32
874 #undef ALE_SYSCTL_STAT_ADD64
875 
876 struct ale_dmamap_arg {
877 	bus_addr_t	ale_busaddr;
878 };
879 
880 static void
881 ale_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error)
882 {
883 	struct ale_dmamap_arg *ctx;
884 
885 	if (error != 0)
886 		return;
887 
888 	KASSERT(nsegs == 1, ("%s: %d segments returned!", __func__, nsegs));
889 
890 	ctx = (struct ale_dmamap_arg *)arg;
891 	ctx->ale_busaddr = segs[0].ds_addr;
892 }
893 
894 /*
895  * Tx descriptors/RXF0/CMB DMA blocks share ALE_DESC_ADDR_HI register
896  * which specifies high address region of DMA blocks. Therefore these
897  * blocks should have the same high address of given 4GB address
898  * space(i.e. crossing 4GB boundary is not allowed).
899  */
900 static int
901 ale_check_boundary(struct ale_softc *sc)
902 {
903 	bus_addr_t rx_cmb_end[ALE_RX_PAGES], tx_cmb_end;
904 	bus_addr_t rx_page_end[ALE_RX_PAGES], tx_ring_end;
905 
906 	rx_page_end[0] = sc->ale_cdata.ale_rx_page[0].page_paddr +
907 	    sc->ale_pagesize;
908 	rx_page_end[1] = sc->ale_cdata.ale_rx_page[1].page_paddr +
909 	    sc->ale_pagesize;
910 	tx_ring_end = sc->ale_cdata.ale_tx_ring_paddr + ALE_TX_RING_SZ;
911 	tx_cmb_end = sc->ale_cdata.ale_tx_cmb_paddr + ALE_TX_CMB_SZ;
912 	rx_cmb_end[0] = sc->ale_cdata.ale_rx_page[0].cmb_paddr + ALE_RX_CMB_SZ;
913 	rx_cmb_end[1] = sc->ale_cdata.ale_rx_page[1].cmb_paddr + ALE_RX_CMB_SZ;
914 
915 	if ((ALE_ADDR_HI(tx_ring_end) !=
916 	    ALE_ADDR_HI(sc->ale_cdata.ale_tx_ring_paddr)) ||
917 	    (ALE_ADDR_HI(rx_page_end[0]) !=
918 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[0].page_paddr)) ||
919 	    (ALE_ADDR_HI(rx_page_end[1]) !=
920 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[1].page_paddr)) ||
921 	    (ALE_ADDR_HI(tx_cmb_end) !=
922 	    ALE_ADDR_HI(sc->ale_cdata.ale_tx_cmb_paddr)) ||
923 	    (ALE_ADDR_HI(rx_cmb_end[0]) !=
924 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[0].cmb_paddr)) ||
925 	    (ALE_ADDR_HI(rx_cmb_end[1]) !=
926 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[1].cmb_paddr)))
927 		return (EFBIG);
928 
929 	if ((ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_page_end[0])) ||
930 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_page_end[1])) ||
931 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_cmb_end[0])) ||
932 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_cmb_end[1])) ||
933 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(tx_cmb_end)))
934 		return (EFBIG);
935 
936 	return (0);
937 }
938 
939 static int
940 ale_dma_alloc(struct ale_softc *sc)
941 {
942 	struct ale_txdesc *txd;
943 	bus_addr_t lowaddr;
944 	struct ale_dmamap_arg ctx;
945 	int error, guard_size, i;
946 
947 	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0)
948 		guard_size = ALE_JUMBO_FRAMELEN;
949 	else
950 		guard_size = ALE_MAX_FRAMELEN;
951 	sc->ale_pagesize = roundup(guard_size + ALE_RX_PAGE_SZ,
952 	    ALE_RX_PAGE_ALIGN);
953 	lowaddr = BUS_SPACE_MAXADDR;
954 again:
955 	/* Create parent DMA tag. */
956 	error = bus_dma_tag_create(
957 	    NULL,			/* parent */
958 	    1, 0,			/* alignment, boundary */
959 	    lowaddr,			/* lowaddr */
960 	    BUS_SPACE_MAXADDR,		/* highaddr */
961 	    NULL, NULL,			/* filter, filterarg */
962 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsize */
963 	    0,				/* nsegments */
964 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsegsize */
965 	    0,				/* flags */
966 	    &sc->ale_cdata.ale_parent_tag);
967 	if (error != 0) {
968 		device_printf(sc->ale_dev,
969 		    "could not create parent DMA tag.\n");
970 		goto fail;
971 	}
972 
973 	/* Create DMA tag for Tx descriptor ring. */
974 	error = bus_dma_tag_create(
975 	    sc->ale_cdata.ale_parent_tag, /* parent */
976 	    ALE_TX_RING_ALIGN, 0,	/* alignment, boundary */
977 	    BUS_SPACE_MAXADDR,		/* lowaddr */
978 	    BUS_SPACE_MAXADDR,		/* highaddr */
979 	    NULL, NULL,			/* filter, filterarg */
980 	    ALE_TX_RING_SZ,		/* maxsize */
981 	    1,				/* nsegments */
982 	    ALE_TX_RING_SZ,		/* maxsegsize */
983 	    0,				/* flags */
984 	    &sc->ale_cdata.ale_tx_ring_tag);
985 	if (error != 0) {
986 		device_printf(sc->ale_dev,
987 		    "could not create Tx ring DMA tag.\n");
988 		goto fail;
989 	}
990 
991 	/* Create DMA tag for Rx pages. */
992 	for (i = 0; i < ALE_RX_PAGES; i++) {
993 		error = bus_dma_tag_create(
994 		    sc->ale_cdata.ale_parent_tag, /* parent */
995 		    ALE_RX_PAGE_ALIGN, 0,	/* alignment, boundary */
996 		    BUS_SPACE_MAXADDR,		/* lowaddr */
997 		    BUS_SPACE_MAXADDR,		/* highaddr */
998 		    NULL, NULL,			/* filter, filterarg */
999 		    sc->ale_pagesize,		/* maxsize */
1000 		    1,				/* nsegments */
1001 		    sc->ale_pagesize,		/* maxsegsize */
1002 		    0,				/* flags */
1003 		    &sc->ale_cdata.ale_rx_page[i].page_tag);
1004 		if (error != 0) {
1005 			device_printf(sc->ale_dev,
1006 			    "could not create Rx page %d DMA tag.\n", i);
1007 			goto fail;
1008 		}
1009 	}
1010 
1011 	/* Create DMA tag for Tx coalescing message block. */
1012 	error = bus_dma_tag_create(
1013 	    sc->ale_cdata.ale_parent_tag, /* parent */
1014 	    ALE_CMB_ALIGN, 0,		/* alignment, boundary */
1015 	    BUS_SPACE_MAXADDR,		/* lowaddr */
1016 	    BUS_SPACE_MAXADDR,		/* highaddr */
1017 	    NULL, NULL,			/* filter, filterarg */
1018 	    ALE_TX_CMB_SZ,		/* maxsize */
1019 	    1,				/* nsegments */
1020 	    ALE_TX_CMB_SZ,		/* maxsegsize */
1021 	    0,				/* flags */
1022 	    &sc->ale_cdata.ale_tx_cmb_tag);
1023 	if (error != 0) {
1024 		device_printf(sc->ale_dev,
1025 		    "could not create Tx CMB DMA tag.\n");
1026 		goto fail;
1027 	}
1028 
1029 	/* Create DMA tag for Rx coalescing message block. */
1030 	for (i = 0; i < ALE_RX_PAGES; i++) {
1031 		error = bus_dma_tag_create(
1032 		    sc->ale_cdata.ale_parent_tag, /* parent */
1033 		    ALE_CMB_ALIGN, 0,		/* alignment, boundary */
1034 		    BUS_SPACE_MAXADDR,		/* lowaddr */
1035 		    BUS_SPACE_MAXADDR,		/* highaddr */
1036 		    NULL, NULL,			/* filter, filterarg */
1037 		    ALE_RX_CMB_SZ,		/* maxsize */
1038 		    1,				/* nsegments */
1039 		    ALE_RX_CMB_SZ,		/* maxsegsize */
1040 		    0,				/* flags */
1041 		    &sc->ale_cdata.ale_rx_page[i].cmb_tag);
1042 		if (error != 0) {
1043 			device_printf(sc->ale_dev,
1044 			    "could not create Rx page %d CMB DMA tag.\n", i);
1045 			goto fail;
1046 		}
1047 	}
1048 
1049 	/* Allocate DMA'able memory and load the DMA map for Tx ring. */
1050 	error = bus_dmamem_alloc(sc->ale_cdata.ale_tx_ring_tag,
1051 	    (void **)&sc->ale_cdata.ale_tx_ring,
1052 	    BUS_DMA_WAITOK | BUS_DMA_ZERO,
1053 	    &sc->ale_cdata.ale_tx_ring_map);
1054 	if (error != 0) {
1055 		device_printf(sc->ale_dev,
1056 		    "could not allocate DMA'able memory for Tx ring.\n");
1057 		goto fail;
1058 	}
1059 	ctx.ale_busaddr = 0;
1060 	error = bus_dmamap_load(sc->ale_cdata.ale_tx_ring_tag,
1061 	    sc->ale_cdata.ale_tx_ring_map, sc->ale_cdata.ale_tx_ring,
1062 	    ALE_TX_RING_SZ, ale_dmamap_cb, &ctx, 0);
1063 	if (error != 0 || ctx.ale_busaddr == 0) {
1064 		device_printf(sc->ale_dev,
1065 		    "could not load DMA'able memory for Tx ring.\n");
1066 		goto fail;
1067 	}
1068 	sc->ale_cdata.ale_tx_ring_paddr = ctx.ale_busaddr;
1069 
1070 	/* Rx pages. */
1071 	for (i = 0; i < ALE_RX_PAGES; i++) {
1072 		error = bus_dmamem_alloc(sc->ale_cdata.ale_rx_page[i].page_tag,
1073 		    (void **)&sc->ale_cdata.ale_rx_page[i].page_addr,
1074 		    BUS_DMA_WAITOK | BUS_DMA_ZERO,
1075 		    &sc->ale_cdata.ale_rx_page[i].page_map);
1076 		if (error != 0) {
1077 			device_printf(sc->ale_dev,
1078 			    "could not allocate DMA'able memory for "
1079 			    "Rx page %d.\n", i);
1080 			goto fail;
1081 		}
1082 		ctx.ale_busaddr = 0;
1083 		error = bus_dmamap_load(sc->ale_cdata.ale_rx_page[i].page_tag,
1084 		    sc->ale_cdata.ale_rx_page[i].page_map,
1085 		    sc->ale_cdata.ale_rx_page[i].page_addr,
1086 		    sc->ale_pagesize, ale_dmamap_cb, &ctx, 0);
1087 		if (error != 0 || ctx.ale_busaddr == 0) {
1088 			device_printf(sc->ale_dev,
1089 			    "could not load DMA'able memory for "
1090 			    "Rx page %d.\n", i);
1091 			goto fail;
1092 		}
1093 		sc->ale_cdata.ale_rx_page[i].page_paddr = ctx.ale_busaddr;
1094 	}
1095 
1096 	/* Tx CMB. */
1097 	error = bus_dmamem_alloc(sc->ale_cdata.ale_tx_cmb_tag,
1098 	    (void **)&sc->ale_cdata.ale_tx_cmb,
1099 	    BUS_DMA_WAITOK | BUS_DMA_ZERO,
1100 	    &sc->ale_cdata.ale_tx_cmb_map);
1101 	if (error != 0) {
1102 		device_printf(sc->ale_dev,
1103 		    "could not allocate DMA'able memory for Tx CMB.\n");
1104 		goto fail;
1105 	}
1106 	ctx.ale_busaddr = 0;
1107 	error = bus_dmamap_load(sc->ale_cdata.ale_tx_cmb_tag,
1108 	    sc->ale_cdata.ale_tx_cmb_map, sc->ale_cdata.ale_tx_cmb,
1109 	    ALE_TX_CMB_SZ, ale_dmamap_cb, &ctx, 0);
1110 	if (error != 0 || ctx.ale_busaddr == 0) {
1111 		device_printf(sc->ale_dev,
1112 		    "could not load DMA'able memory for Tx CMB.\n");
1113 		goto fail;
1114 	}
1115 	sc->ale_cdata.ale_tx_cmb_paddr = ctx.ale_busaddr;
1116 
1117 	/* Rx CMB. */
1118 	for (i = 0; i < ALE_RX_PAGES; i++) {
1119 		error = bus_dmamem_alloc(sc->ale_cdata.ale_rx_page[i].cmb_tag,
1120 		    (void **)&sc->ale_cdata.ale_rx_page[i].cmb_addr,
1121 		    BUS_DMA_WAITOK | BUS_DMA_ZERO,
1122 		    &sc->ale_cdata.ale_rx_page[i].cmb_map);
1123 		if (error != 0) {
1124 			device_printf(sc->ale_dev, "could not allocate "
1125 			    "DMA'able memory for Rx page %d CMB.\n", i);
1126 			goto fail;
1127 		}
1128 		ctx.ale_busaddr = 0;
1129 		error = bus_dmamap_load(sc->ale_cdata.ale_rx_page[i].cmb_tag,
1130 		    sc->ale_cdata.ale_rx_page[i].cmb_map,
1131 		    sc->ale_cdata.ale_rx_page[i].cmb_addr,
1132 		    ALE_RX_CMB_SZ, ale_dmamap_cb, &ctx, 0);
1133 		if (error != 0 || ctx.ale_busaddr == 0) {
1134 			device_printf(sc->ale_dev, "could not load DMA'able "
1135 			    "memory for Rx page %d CMB.\n", i);
1136 			goto fail;
1137 		}
1138 		sc->ale_cdata.ale_rx_page[i].cmb_paddr = ctx.ale_busaddr;
1139 	}
1140 
1141 	/*
1142 	 * Tx descriptors/RXF0/CMB DMA blocks share the same
1143 	 * high address region of 64bit DMA address space.
1144 	 */
1145 	if (lowaddr != BUS_SPACE_MAXADDR_32BIT &&
1146 	    (error = ale_check_boundary(sc)) != 0) {
1147 		device_printf(sc->ale_dev, "4GB boundary crossed, "
1148 		    "switching to 32bit DMA addressing mode.\n");
1149 		ale_dma_free(sc);
1150 		/*
1151 		 * Limit max allowable DMA address space to 32bit
1152 		 * and try again.
1153 		 */
1154 		lowaddr = BUS_SPACE_MAXADDR_32BIT;
1155 		goto again;
1156 	}
1157 
1158 	/*
1159 	 * Create Tx buffer parent tag.
1160 	 * AR81xx allows 64bit DMA addressing of Tx buffers so it
1161 	 * needs separate parent DMA tag as parent DMA address space
1162 	 * could be restricted to be within 32bit address space by
1163 	 * 4GB boundary crossing.
1164 	 */
1165 	error = bus_dma_tag_create(
1166 	    NULL,			/* parent */
1167 	    1, 0,			/* alignment, boundary */
1168 	    BUS_SPACE_MAXADDR,		/* lowaddr */
1169 	    BUS_SPACE_MAXADDR,		/* highaddr */
1170 	    NULL, NULL,			/* filter, filterarg */
1171 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsize */
1172 	    0,				/* nsegments */
1173 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsegsize */
1174 	    0,				/* flags */
1175 	    &sc->ale_cdata.ale_buffer_tag);
1176 	if (error != 0) {
1177 		device_printf(sc->ale_dev,
1178 		    "could not create parent buffer DMA tag.\n");
1179 		goto fail;
1180 	}
1181 
1182 	/* Create DMA tag for Tx buffers. */
1183 	error = bus_dma_tag_create(
1184 	    sc->ale_cdata.ale_buffer_tag, /* parent */
1185 	    1, 0,			/* alignment, boundary */
1186 	    BUS_SPACE_MAXADDR,		/* lowaddr */
1187 	    BUS_SPACE_MAXADDR,		/* highaddr */
1188 	    NULL, NULL,			/* filter, filterarg */
1189 	    ALE_TSO_MAXSIZE,		/* maxsize */
1190 	    ALE_MAXTXSEGS,		/* nsegments */
1191 	    ALE_TSO_MAXSEGSIZE,		/* maxsegsize */
1192 	    0,				/* flags */
1193 	    &sc->ale_cdata.ale_tx_tag);
1194 	if (error != 0) {
1195 		device_printf(sc->ale_dev, "could not create Tx DMA tag.\n");
1196 		goto fail;
1197 	}
1198 
1199 	/* Create DMA maps for Tx buffers. */
1200 	for (i = 0; i < ALE_TX_RING_CNT; i++) {
1201 		txd = &sc->ale_cdata.ale_txdesc[i];
1202 		txd->tx_m = NULL;
1203 		txd->tx_dmamap = NULL;
1204 		error = bus_dmamap_create(sc->ale_cdata.ale_tx_tag, 0,
1205 		    &txd->tx_dmamap);
1206 		if (error != 0) {
1207 			device_printf(sc->ale_dev,
1208 			    "could not create Tx dmamap.\n");
1209 			goto fail;
1210 		}
1211 	}
1212 fail:
1213 	return (error);
1214 }
1215 
1216 static void
1217 ale_dma_free(struct ale_softc *sc)
1218 {
1219 	struct ale_txdesc *txd;
1220 	int i;
1221 
1222 	/* Tx buffers. */
1223 	if (sc->ale_cdata.ale_tx_tag != NULL) {
1224 		for (i = 0; i < ALE_TX_RING_CNT; i++) {
1225 			txd = &sc->ale_cdata.ale_txdesc[i];
1226 			if (txd->tx_dmamap != NULL) {
1227 				bus_dmamap_destroy(sc->ale_cdata.ale_tx_tag,
1228 				    txd->tx_dmamap);
1229 				txd->tx_dmamap = NULL;
1230 			}
1231 		}
1232 		bus_dma_tag_destroy(sc->ale_cdata.ale_tx_tag);
1233 		sc->ale_cdata.ale_tx_tag = NULL;
1234 	}
1235 	/* Tx descriptor ring. */
1236 	if (sc->ale_cdata.ale_tx_ring_tag != NULL) {
1237 		if (sc->ale_cdata.ale_tx_ring_map != NULL)
1238 			bus_dmamap_unload(sc->ale_cdata.ale_tx_ring_tag,
1239 			    sc->ale_cdata.ale_tx_ring_map);
1240 		if (sc->ale_cdata.ale_tx_ring_map != NULL &&
1241 		    sc->ale_cdata.ale_tx_ring != NULL)
1242 			bus_dmamem_free(sc->ale_cdata.ale_tx_ring_tag,
1243 			    sc->ale_cdata.ale_tx_ring,
1244 			    sc->ale_cdata.ale_tx_ring_map);
1245 		sc->ale_cdata.ale_tx_ring = NULL;
1246 		sc->ale_cdata.ale_tx_ring_map = NULL;
1247 		bus_dma_tag_destroy(sc->ale_cdata.ale_tx_ring_tag);
1248 		sc->ale_cdata.ale_tx_ring_tag = NULL;
1249 	}
1250 	/* Rx page block. */
1251 	for (i = 0; i < ALE_RX_PAGES; i++) {
1252 		if (sc->ale_cdata.ale_rx_page[i].page_tag != NULL) {
1253 			if (sc->ale_cdata.ale_rx_page[i].page_map != NULL)
1254 				bus_dmamap_unload(
1255 				    sc->ale_cdata.ale_rx_page[i].page_tag,
1256 				    sc->ale_cdata.ale_rx_page[i].page_map);
1257 			if (sc->ale_cdata.ale_rx_page[i].page_map != NULL &&
1258 			    sc->ale_cdata.ale_rx_page[i].page_addr != NULL)
1259 				bus_dmamem_free(
1260 				    sc->ale_cdata.ale_rx_page[i].page_tag,
1261 				    sc->ale_cdata.ale_rx_page[i].page_addr,
1262 				    sc->ale_cdata.ale_rx_page[i].page_map);
1263 			sc->ale_cdata.ale_rx_page[i].page_addr = NULL;
1264 			sc->ale_cdata.ale_rx_page[i].page_map = NULL;
1265 			bus_dma_tag_destroy(
1266 			    sc->ale_cdata.ale_rx_page[i].page_tag);
1267 			sc->ale_cdata.ale_rx_page[i].page_tag = NULL;
1268 		}
1269 	}
1270 	/* Rx CMB. */
1271 	for (i = 0; i < ALE_RX_PAGES; i++) {
1272 		if (sc->ale_cdata.ale_rx_page[i].cmb_tag != NULL) {
1273 			if (sc->ale_cdata.ale_rx_page[i].cmb_map != NULL)
1274 				bus_dmamap_unload(
1275 				    sc->ale_cdata.ale_rx_page[i].cmb_tag,
1276 				    sc->ale_cdata.ale_rx_page[i].cmb_map);
1277 			if (sc->ale_cdata.ale_rx_page[i].cmb_map != NULL &&
1278 			    sc->ale_cdata.ale_rx_page[i].cmb_addr != NULL)
1279 				bus_dmamem_free(
1280 				    sc->ale_cdata.ale_rx_page[i].cmb_tag,
1281 				    sc->ale_cdata.ale_rx_page[i].cmb_addr,
1282 				    sc->ale_cdata.ale_rx_page[i].cmb_map);
1283 			sc->ale_cdata.ale_rx_page[i].cmb_addr = NULL;
1284 			sc->ale_cdata.ale_rx_page[i].cmb_map = NULL;
1285 			bus_dma_tag_destroy(
1286 			    sc->ale_cdata.ale_rx_page[i].cmb_tag);
1287 			sc->ale_cdata.ale_rx_page[i].cmb_tag = NULL;
1288 		}
1289 	}
1290 	/* Tx CMB. */
1291 	if (sc->ale_cdata.ale_tx_cmb_tag != NULL) {
1292 		if (sc->ale_cdata.ale_tx_cmb_map != NULL)
1293 			bus_dmamap_unload(sc->ale_cdata.ale_tx_cmb_tag,
1294 			    sc->ale_cdata.ale_tx_cmb_map);
1295 		if (sc->ale_cdata.ale_tx_cmb_map != NULL &&
1296 		    sc->ale_cdata.ale_tx_cmb != NULL)
1297 			bus_dmamem_free(sc->ale_cdata.ale_tx_cmb_tag,
1298 			    sc->ale_cdata.ale_tx_cmb,
1299 			    sc->ale_cdata.ale_tx_cmb_map);
1300 		sc->ale_cdata.ale_tx_cmb = NULL;
1301 		sc->ale_cdata.ale_tx_cmb_map = NULL;
1302 		bus_dma_tag_destroy(sc->ale_cdata.ale_tx_cmb_tag);
1303 		sc->ale_cdata.ale_tx_cmb_tag = NULL;
1304 	}
1305 	if (sc->ale_cdata.ale_buffer_tag != NULL) {
1306 		bus_dma_tag_destroy(sc->ale_cdata.ale_buffer_tag);
1307 		sc->ale_cdata.ale_buffer_tag = NULL;
1308 	}
1309 	if (sc->ale_cdata.ale_parent_tag != NULL) {
1310 		bus_dma_tag_destroy(sc->ale_cdata.ale_parent_tag);
1311 		sc->ale_cdata.ale_parent_tag = NULL;
1312 	}
1313 }
1314 
1315 static int
1316 ale_shutdown(device_t dev)
1317 {
1318 	return (ale_suspend(dev));
1319 }
1320 
1321 #ifdef notyet
1322 
1323 /*
1324  * Note, this driver resets the link speed to 10/100Mbps by
1325  * restarting auto-negotiation in suspend/shutdown phase but we
1326  * don't know whether that auto-negotiation would succeed or not
1327  * as driver has no control after powering off/suspend operation.
1328  * If the renegotiation fail WOL may not work. Running at 1Gbps
1329  * will draw more power than 375mA at 3.3V which is specified in
1330  * PCI specification and that would result in complete
1331  * shutdowning power to ethernet controller.
1332  *
1333  * TODO
1334  * Save current negotiated media speed/duplex/flow-control to
1335  * softc and restore the same link again after resuming. PHY
1336  * handling such as power down/resetting to 100Mbps may be better
1337  * handled in suspend method in phy driver.
1338  */
1339 static void
1340 ale_setlinkspeed(struct ale_softc *sc)
1341 {
1342 	struct mii_data *mii;
1343 	int aneg, i;
1344 
1345 	mii = device_get_softc(sc->ale_miibus);
1346 	mii_pollstat(mii);
1347 	aneg = 0;
1348 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
1349 	    (IFM_ACTIVE | IFM_AVALID)) {
1350 		switch IFM_SUBTYPE(mii->mii_media_active) {
1351 		case IFM_10_T:
1352 		case IFM_100_TX:
1353 			return;
1354 		case IFM_1000_T:
1355 			aneg++;
1356 			break;
1357 		default:
1358 			break;
1359 		}
1360 	}
1361 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr, MII_100T2CR, 0);
1362 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
1363 	    MII_ANAR, ANAR_TX_FD | ANAR_TX | ANAR_10_FD | ANAR_10 | ANAR_CSMA);
1364 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
1365 	    MII_BMCR, BMCR_RESET | BMCR_AUTOEN | BMCR_STARTNEG);
1366 	DELAY(1000);
1367 	if (aneg != 0) {
1368 		/*
1369 		 * Poll link state until ale(4) get a 10/100Mbps link.
1370 		 */
1371 		for (i = 0; i < MII_ANEGTICKS_GIGE; i++) {
1372 			mii_pollstat(mii);
1373 			if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID))
1374 			    == (IFM_ACTIVE | IFM_AVALID)) {
1375 				switch (IFM_SUBTYPE(
1376 				    mii->mii_media_active)) {
1377 				case IFM_10_T:
1378 				case IFM_100_TX:
1379 					ale_mac_config(sc);
1380 					return;
1381 				default:
1382 					break;
1383 				}
1384 			}
1385 			ALE_UNLOCK(sc);
1386 			pause("alelnk", hz);
1387 			ALE_LOCK(sc);
1388 		}
1389 		if (i == MII_ANEGTICKS_GIGE)
1390 			device_printf(sc->ale_dev,
1391 			    "establishing a link failed, WOL may not work!");
1392 	}
1393 	/*
1394 	 * No link, force MAC to have 100Mbps, full-duplex link.
1395 	 * This is the last resort and may/may not work.
1396 	 */
1397 	mii->mii_media_status = IFM_AVALID | IFM_ACTIVE;
1398 	mii->mii_media_active = IFM_ETHER | IFM_100_TX | IFM_FDX;
1399 	ale_mac_config(sc);
1400 }
1401 
1402 static void
1403 ale_setwol(struct ale_softc *sc)
1404 {
1405 	struct ifnet *ifp;
1406 	uint32_t reg, pmcs;
1407 	uint16_t pmstat;
1408 	int pmc;
1409 
1410 	ALE_LOCK_ASSERT(sc);
1411 
1412 	if (pci_find_extcap(sc->ale_dev, PCIY_PMG, &pmc) != 0) {
1413 		/* Disable WOL. */
1414 		CSR_WRITE_4(sc, ALE_WOL_CFG, 0);
1415 		reg = CSR_READ_4(sc, ALE_PCIE_PHYMISC);
1416 		reg |= PCIE_PHYMISC_FORCE_RCV_DET;
1417 		CSR_WRITE_4(sc, ALE_PCIE_PHYMISC, reg);
1418 		/* Force PHY power down. */
1419 		CSR_WRITE_2(sc, ALE_GPHY_CTRL,
1420 		    GPHY_CTRL_EXT_RESET | GPHY_CTRL_HIB_EN |
1421 		    GPHY_CTRL_HIB_PULSE | GPHY_CTRL_PHY_PLL_ON |
1422 		    GPHY_CTRL_SEL_ANA_RESET | GPHY_CTRL_PHY_IDDQ |
1423 		    GPHY_CTRL_PCLK_SEL_DIS | GPHY_CTRL_PWDOWN_HW);
1424 		return;
1425 	}
1426 
1427 	ifp = sc->ale_ifp;
1428 	if ((ifp->if_capenable & IFCAP_WOL) != 0) {
1429 		if ((sc->ale_flags & ALE_FLAG_FASTETHER) == 0)
1430 			ale_setlinkspeed(sc);
1431 	}
1432 
1433 	pmcs = 0;
1434 	if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
1435 		pmcs |= WOL_CFG_MAGIC | WOL_CFG_MAGIC_ENB;
1436 	CSR_WRITE_4(sc, ALE_WOL_CFG, pmcs);
1437 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
1438 	reg &= ~(MAC_CFG_DBG | MAC_CFG_PROMISC | MAC_CFG_ALLMULTI |
1439 	    MAC_CFG_BCAST);
1440 	if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
1441 		reg |= MAC_CFG_ALLMULTI | MAC_CFG_BCAST;
1442 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
1443 		reg |= MAC_CFG_RX_ENB;
1444 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
1445 
1446 	if ((ifp->if_capenable & IFCAP_WOL) == 0) {
1447 		/* WOL disabled, PHY power down. */
1448 		reg = CSR_READ_4(sc, ALE_PCIE_PHYMISC);
1449 		reg |= PCIE_PHYMISC_FORCE_RCV_DET;
1450 		CSR_WRITE_4(sc, ALE_PCIE_PHYMISC, reg);
1451 		CSR_WRITE_2(sc, ALE_GPHY_CTRL,
1452 		    GPHY_CTRL_EXT_RESET | GPHY_CTRL_HIB_EN |
1453 		    GPHY_CTRL_HIB_PULSE | GPHY_CTRL_SEL_ANA_RESET |
1454 		    GPHY_CTRL_PHY_IDDQ | GPHY_CTRL_PCLK_SEL_DIS |
1455 		    GPHY_CTRL_PWDOWN_HW);
1456 	}
1457 	/* Request PME. */
1458 	pmstat = pci_read_config(sc->ale_dev, pmc + PCIR_POWER_STATUS, 2);
1459 	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
1460 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
1461 		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
1462 	pci_write_config(sc->ale_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
1463 }
1464 
1465 #endif	/* notyet */
1466 
1467 static int
1468 ale_suspend(device_t dev)
1469 {
1470 	struct ale_softc *sc = device_get_softc(dev);
1471 	struct ifnet *ifp = &sc->arpcom.ac_if;
1472 
1473 	lwkt_serialize_enter(ifp->if_serializer);
1474 	ale_stop(sc);
1475 #ifdef notyet
1476 	ale_setwol(sc);
1477 #endif
1478 	lwkt_serialize_exit(ifp->if_serializer);
1479 	return (0);
1480 }
1481 
1482 static int
1483 ale_resume(device_t dev)
1484 {
1485 	struct ale_softc *sc = device_get_softc(dev);
1486 	struct ifnet *ifp = &sc->arpcom.ac_if;
1487 	uint16_t cmd;
1488 
1489 	lwkt_serialize_enter(ifp->if_serializer);
1490 
1491 	/*
1492 	 * Clear INTx emulation disable for hardwares that
1493 	 * is set in resume event. From Linux.
1494 	 */
1495 	cmd = pci_read_config(sc->ale_dev, PCIR_COMMAND, 2);
1496 	if ((cmd & 0x0400) != 0) {
1497 		cmd &= ~0x0400;
1498 		pci_write_config(sc->ale_dev, PCIR_COMMAND, cmd, 2);
1499 	}
1500 
1501 #ifdef notyet
1502 	if (pci_find_extcap(sc->ale_dev, PCIY_PMG, &pmc) == 0) {
1503 		uint16_t pmstat;
1504 		int pmc;
1505 
1506 		/* Disable PME and clear PME status. */
1507 		pmstat = pci_read_config(sc->ale_dev,
1508 		    pmc + PCIR_POWER_STATUS, 2);
1509 		if ((pmstat & PCIM_PSTAT_PMEENABLE) != 0) {
1510 			pmstat &= ~PCIM_PSTAT_PMEENABLE;
1511 			pci_write_config(sc->ale_dev,
1512 			    pmc + PCIR_POWER_STATUS, pmstat, 2);
1513 		}
1514 	}
1515 #endif
1516 
1517 	/* Reset PHY. */
1518 	ale_phy_reset(sc);
1519 	if ((ifp->if_flags & IFF_UP) != 0)
1520 		ale_init(sc);
1521 
1522 	lwkt_serialize_exit(ifp->if_serializer);
1523 	return (0);
1524 }
1525 
1526 static int
1527 ale_encap(struct ale_softc *sc, struct mbuf **m_head)
1528 {
1529 	struct ale_txdesc *txd, *txd_last;
1530 	struct tx_desc *desc;
1531 	struct mbuf *m;
1532 	bus_dma_segment_t txsegs[ALE_MAXTXSEGS];
1533 	struct ale_dmamap_ctx ctx;
1534 	bus_dmamap_t map;
1535 	uint32_t cflags, poff, vtag;
1536 	int error, i, nsegs, prod, si;
1537 
1538 	M_ASSERTPKTHDR((*m_head));
1539 
1540 	m = *m_head;
1541 	cflags = vtag = 0;
1542 	poff = 0;
1543 
1544 	si = prod = sc->ale_cdata.ale_tx_prod;
1545 	txd = &sc->ale_cdata.ale_txdesc[prod];
1546 	txd_last = txd;
1547 	map = txd->tx_dmamap;
1548 
1549 	ctx.nsegs = ALE_MAXTXSEGS;
1550 	ctx.segs = txsegs;
1551 	error =  bus_dmamap_load_mbuf(sc->ale_cdata.ale_tx_tag, map,
1552 				      *m_head, ale_dmamap_buf_cb, &ctx,
1553 				      BUS_DMA_NOWAIT);
1554 	if (error == EFBIG) {
1555 		m = m_defrag(*m_head, MB_DONTWAIT);
1556 		if (m == NULL) {
1557 			m_freem(*m_head);
1558 			*m_head = NULL;
1559 			return (ENOMEM);
1560 		}
1561 		*m_head = m;
1562 
1563 		ctx.nsegs = ALE_MAXTXSEGS;
1564 		ctx.segs = txsegs;
1565 		error =  bus_dmamap_load_mbuf(sc->ale_cdata.ale_tx_tag, map,
1566 					      *m_head, ale_dmamap_buf_cb, &ctx,
1567 					      BUS_DMA_NOWAIT);
1568 		if (error != 0) {
1569 			m_freem(*m_head);
1570 			*m_head = NULL;
1571 			return (error);
1572 		}
1573 	} else if (error != 0) {
1574 		return (error);
1575 	}
1576 	nsegs = ctx.nsegs;
1577 
1578 	if (nsegs == 0) {
1579 		m_freem(*m_head);
1580 		*m_head = NULL;
1581 		return (EIO);
1582 	}
1583 
1584 	/* Check descriptor overrun. */
1585 	if (sc->ale_cdata.ale_tx_cnt + nsegs >= ALE_TX_RING_CNT - 2) {
1586 		bus_dmamap_unload(sc->ale_cdata.ale_tx_tag, map);
1587 		return (ENOBUFS);
1588 	}
1589 	bus_dmamap_sync(sc->ale_cdata.ale_tx_tag, map, BUS_DMASYNC_PREWRITE);
1590 
1591 	m = *m_head;
1592 	/* Configure Tx checksum offload. */
1593 	if ((m->m_pkthdr.csum_flags & ALE_CSUM_FEATURES) != 0) {
1594 		/*
1595 		 * AR81xx supports Tx custom checksum offload feature
1596 		 * that offloads single 16bit checksum computation.
1597 		 * So you can choose one among IP, TCP and UDP.
1598 		 * Normally driver sets checksum start/insertion
1599 		 * position from the information of TCP/UDP frame as
1600 		 * TCP/UDP checksum takes more time than that of IP.
1601 		 * However it seems that custom checksum offload
1602 		 * requires 4 bytes aligned Tx buffers due to hardware
1603 		 * bug.
1604 		 * AR81xx also supports explicit Tx checksum computation
1605 		 * if it is told that the size of IP header and TCP
1606 		 * header(for UDP, the header size does not matter
1607 		 * because it's fixed length). However with this scheme
1608 		 * TSO does not work so you have to choose one either
1609 		 * TSO or explicit Tx checksum offload. I chosen TSO
1610 		 * plus custom checksum offload with work-around which
1611 		 * will cover most common usage for this consumer
1612 		 * ethernet controller. The work-around takes a lot of
1613 		 * CPU cycles if Tx buffer is not aligned on 4 bytes
1614 		 * boundary, though.
1615 		 */
1616 		cflags |= ALE_TD_CXSUM;
1617 		/* Set checksum start offset. */
1618 		cflags |= (poff << ALE_TD_CSUM_PLOADOFFSET_SHIFT);
1619 		/* Set checksum insertion position of TCP/UDP. */
1620 		cflags |= ((poff + m->m_pkthdr.csum_data) <<
1621 		    ALE_TD_CSUM_XSUMOFFSET_SHIFT);
1622 	}
1623 
1624 	/* Configure VLAN hardware tag insertion. */
1625 	if ((m->m_flags & M_VLANTAG) != 0) {
1626 		vtag = ALE_TX_VLAN_TAG(m->m_pkthdr.ether_vlantag);
1627 		vtag = ((vtag << ALE_TD_VLAN_SHIFT) & ALE_TD_VLAN_MASK);
1628 		cflags |= ALE_TD_INSERT_VLAN_TAG;
1629 	}
1630 
1631 	desc = NULL;
1632 	for (i = 0; i < nsegs; i++) {
1633 		desc = &sc->ale_cdata.ale_tx_ring[prod];
1634 		desc->addr = htole64(txsegs[i].ds_addr);
1635 		desc->len = htole32(ALE_TX_BYTES(txsegs[i].ds_len) | vtag);
1636 		desc->flags = htole32(cflags);
1637 		sc->ale_cdata.ale_tx_cnt++;
1638 		ALE_DESC_INC(prod, ALE_TX_RING_CNT);
1639 	}
1640 	/* Update producer index. */
1641 	sc->ale_cdata.ale_tx_prod = prod;
1642 
1643 	/* Finally set EOP on the last descriptor. */
1644 	prod = (prod + ALE_TX_RING_CNT - 1) % ALE_TX_RING_CNT;
1645 	desc = &sc->ale_cdata.ale_tx_ring[prod];
1646 	desc->flags |= htole32(ALE_TD_EOP);
1647 
1648 	/* Swap dmamap of the first and the last. */
1649 	txd = &sc->ale_cdata.ale_txdesc[prod];
1650 	map = txd_last->tx_dmamap;
1651 	txd_last->tx_dmamap = txd->tx_dmamap;
1652 	txd->tx_dmamap = map;
1653 	txd->tx_m = m;
1654 
1655 	/* Sync descriptors. */
1656 	bus_dmamap_sync(sc->ale_cdata.ale_tx_ring_tag,
1657 	    sc->ale_cdata.ale_tx_ring_map, BUS_DMASYNC_PREWRITE);
1658 
1659 	return (0);
1660 }
1661 
1662 static void
1663 ale_start(struct ifnet *ifp)
1664 {
1665         struct ale_softc *sc = ifp->if_softc;
1666 	struct mbuf *m_head;
1667 	int enq;
1668 
1669 	ASSERT_SERIALIZED(ifp->if_serializer);
1670 
1671 	if ((sc->ale_flags & ALE_FLAG_LINK) == 0) {
1672 		ifq_purge(&ifp->if_snd);
1673 		return;
1674 	}
1675 
1676 	if ((ifp->if_flags & (IFF_RUNNING | IFF_OACTIVE)) != IFF_RUNNING)
1677 		return;
1678 
1679 	/* Reclaim transmitted frames. */
1680 	if (sc->ale_cdata.ale_tx_cnt >= ALE_TX_DESC_HIWAT)
1681 		ale_txeof(sc);
1682 
1683 	enq = 0;
1684 	while (!ifq_is_empty(&ifp->if_snd)) {
1685 		m_head = ifq_dequeue(&ifp->if_snd, NULL);
1686 		if (m_head == NULL)
1687 			break;
1688 
1689 		/*
1690 		 * Pack the data into the transmit ring. If we
1691 		 * don't have room, set the OACTIVE flag and wait
1692 		 * for the NIC to drain the ring.
1693 		 */
1694 		if (ale_encap(sc, &m_head)) {
1695 			if (m_head == NULL)
1696 				break;
1697 			ifq_prepend(&ifp->if_snd, m_head);
1698 			ifp->if_flags |= IFF_OACTIVE;
1699 			break;
1700 		}
1701 		enq = 1;
1702 
1703 		/*
1704 		 * If there's a BPF listener, bounce a copy of this frame
1705 		 * to him.
1706 		 */
1707 		ETHER_BPF_MTAP(ifp, m_head);
1708 	}
1709 
1710 	if (enq) {
1711 		/* Kick. */
1712 		CSR_WRITE_4(sc, ALE_MBOX_TPD_PROD_IDX,
1713 		    sc->ale_cdata.ale_tx_prod);
1714 
1715 		/* Set a timeout in case the chip goes out to lunch. */
1716 		ifp->if_timer = ALE_TX_TIMEOUT;
1717 	}
1718 }
1719 
1720 static void
1721 ale_watchdog(struct ifnet *ifp)
1722 {
1723 	struct ale_softc *sc = ifp->if_softc;
1724 
1725 	ASSERT_SERIALIZED(ifp->if_serializer);
1726 
1727 	if ((sc->ale_flags & ALE_FLAG_LINK) == 0) {
1728 		if_printf(ifp, "watchdog timeout (lost link)\n");
1729 		ifp->if_oerrors++;
1730 		ale_init(sc);
1731 		return;
1732 	}
1733 
1734 	if_printf(ifp, "watchdog timeout -- resetting\n");
1735 	ifp->if_oerrors++;
1736 	ale_init(sc);
1737 
1738 	if (!ifq_is_empty(&ifp->if_snd))
1739 		if_devstart(ifp);
1740 }
1741 
1742 static int
1743 ale_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data, struct ucred *cr)
1744 {
1745 	struct ale_softc *sc;
1746 	struct ifreq *ifr;
1747 	struct mii_data *mii;
1748 	int error, mask;
1749 
1750 	ASSERT_SERIALIZED(ifp->if_serializer);
1751 
1752 	sc = ifp->if_softc;
1753 	ifr = (struct ifreq *)data;
1754 	error = 0;
1755 
1756 	switch (cmd) {
1757 	case SIOCSIFMTU:
1758 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > ALE_JUMBO_MTU ||
1759 		    ((sc->ale_flags & ALE_FLAG_JUMBO) == 0 &&
1760 		    ifr->ifr_mtu > ETHERMTU))
1761 			error = EINVAL;
1762 		else if (ifp->if_mtu != ifr->ifr_mtu) {
1763 			ifp->if_mtu = ifr->ifr_mtu;
1764 			if ((ifp->if_flags & IFF_RUNNING) != 0)
1765 				ale_init(sc);
1766 		}
1767 		break;
1768 
1769 	case SIOCSIFFLAGS:
1770 		if ((ifp->if_flags & IFF_UP) != 0) {
1771 			if ((ifp->if_flags & IFF_RUNNING) != 0) {
1772 				if (((ifp->if_flags ^ sc->ale_if_flags)
1773 				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
1774 					ale_rxfilter(sc);
1775 			} else {
1776 				if ((sc->ale_flags & ALE_FLAG_DETACH) == 0)
1777 					ale_init(sc);
1778 			}
1779 		} else {
1780 			if ((ifp->if_flags & IFF_RUNNING) != 0)
1781 				ale_stop(sc);
1782 		}
1783 		sc->ale_if_flags = ifp->if_flags;
1784 		break;
1785 
1786 	case SIOCADDMULTI:
1787 	case SIOCDELMULTI:
1788 		if ((ifp->if_flags & IFF_RUNNING) != 0)
1789 			ale_rxfilter(sc);
1790 		break;
1791 
1792 	case SIOCSIFMEDIA:
1793 	case SIOCGIFMEDIA:
1794 		mii = device_get_softc(sc->ale_miibus);
1795 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, cmd);
1796 		break;
1797 
1798 	case SIOCSIFCAP:
1799 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1800 		if ((mask & IFCAP_TXCSUM) != 0 &&
1801 		    (ifp->if_capabilities & IFCAP_TXCSUM) != 0) {
1802 			ifp->if_capenable ^= IFCAP_TXCSUM;
1803 			if ((ifp->if_capenable & IFCAP_TXCSUM) != 0)
1804 				ifp->if_hwassist |= ALE_CSUM_FEATURES;
1805 			else
1806 				ifp->if_hwassist &= ~ALE_CSUM_FEATURES;
1807 		}
1808 		if ((mask & IFCAP_RXCSUM) != 0 &&
1809 		    (ifp->if_capabilities & IFCAP_RXCSUM) != 0)
1810 			ifp->if_capenable ^= IFCAP_RXCSUM;
1811 
1812 		if ((mask & IFCAP_VLAN_HWTAGGING) != 0 &&
1813 		    (ifp->if_capabilities & IFCAP_VLAN_HWTAGGING) != 0) {
1814 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1815 			ale_rxvlan(sc);
1816 		}
1817 		break;
1818 
1819 	default:
1820 		error = ether_ioctl(ifp, cmd, data);
1821 		break;
1822 	}
1823 	return (error);
1824 }
1825 
1826 static void
1827 ale_mac_config(struct ale_softc *sc)
1828 {
1829 	struct mii_data *mii;
1830 	uint32_t reg;
1831 
1832 	mii = device_get_softc(sc->ale_miibus);
1833 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
1834 	reg &= ~(MAC_CFG_FULL_DUPLEX | MAC_CFG_TX_FC | MAC_CFG_RX_FC |
1835 	    MAC_CFG_SPEED_MASK);
1836 	/* Reprogram MAC with resolved speed/duplex. */
1837 	switch (IFM_SUBTYPE(mii->mii_media_active)) {
1838 	case IFM_10_T:
1839 	case IFM_100_TX:
1840 		reg |= MAC_CFG_SPEED_10_100;
1841 		break;
1842 	case IFM_1000_T:
1843 		reg |= MAC_CFG_SPEED_1000;
1844 		break;
1845 	}
1846 	if ((IFM_OPTIONS(mii->mii_media_active) & IFM_FDX) != 0) {
1847 		reg |= MAC_CFG_FULL_DUPLEX;
1848 #ifdef notyet
1849 		if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_TXPAUSE) != 0)
1850 			reg |= MAC_CFG_TX_FC;
1851 		if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_RXPAUSE) != 0)
1852 			reg |= MAC_CFG_RX_FC;
1853 #endif
1854 	}
1855 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
1856 }
1857 
1858 static void
1859 ale_stats_clear(struct ale_softc *sc)
1860 {
1861 	struct smb sb;
1862 	uint32_t *reg;
1863 	int i;
1864 
1865 	for (reg = &sb.rx_frames, i = 0; reg <= &sb.rx_pkts_filtered; reg++) {
1866 		CSR_READ_4(sc, ALE_RX_MIB_BASE + i);
1867 		i += sizeof(uint32_t);
1868 	}
1869 	/* Read Tx statistics. */
1870 	for (reg = &sb.tx_frames, i = 0; reg <= &sb.tx_mcast_bytes; reg++) {
1871 		CSR_READ_4(sc, ALE_TX_MIB_BASE + i);
1872 		i += sizeof(uint32_t);
1873 	}
1874 }
1875 
1876 static void
1877 ale_stats_update(struct ale_softc *sc)
1878 {
1879 	struct ale_hw_stats *stat;
1880 	struct smb sb, *smb;
1881 	struct ifnet *ifp;
1882 	uint32_t *reg;
1883 	int i;
1884 
1885 	ifp = &sc->arpcom.ac_if;
1886 	stat = &sc->ale_stats;
1887 	smb = &sb;
1888 
1889 	/* Read Rx statistics. */
1890 	for (reg = &sb.rx_frames, i = 0; reg <= &sb.rx_pkts_filtered; reg++) {
1891 		*reg = CSR_READ_4(sc, ALE_RX_MIB_BASE + i);
1892 		i += sizeof(uint32_t);
1893 	}
1894 	/* Read Tx statistics. */
1895 	for (reg = &sb.tx_frames, i = 0; reg <= &sb.tx_mcast_bytes; reg++) {
1896 		*reg = CSR_READ_4(sc, ALE_TX_MIB_BASE + i);
1897 		i += sizeof(uint32_t);
1898 	}
1899 
1900 	/* Rx stats. */
1901 	stat->rx_frames += smb->rx_frames;
1902 	stat->rx_bcast_frames += smb->rx_bcast_frames;
1903 	stat->rx_mcast_frames += smb->rx_mcast_frames;
1904 	stat->rx_pause_frames += smb->rx_pause_frames;
1905 	stat->rx_control_frames += smb->rx_control_frames;
1906 	stat->rx_crcerrs += smb->rx_crcerrs;
1907 	stat->rx_lenerrs += smb->rx_lenerrs;
1908 	stat->rx_bytes += smb->rx_bytes;
1909 	stat->rx_runts += smb->rx_runts;
1910 	stat->rx_fragments += smb->rx_fragments;
1911 	stat->rx_pkts_64 += smb->rx_pkts_64;
1912 	stat->rx_pkts_65_127 += smb->rx_pkts_65_127;
1913 	stat->rx_pkts_128_255 += smb->rx_pkts_128_255;
1914 	stat->rx_pkts_256_511 += smb->rx_pkts_256_511;
1915 	stat->rx_pkts_512_1023 += smb->rx_pkts_512_1023;
1916 	stat->rx_pkts_1024_1518 += smb->rx_pkts_1024_1518;
1917 	stat->rx_pkts_1519_max += smb->rx_pkts_1519_max;
1918 	stat->rx_pkts_truncated += smb->rx_pkts_truncated;
1919 	stat->rx_fifo_oflows += smb->rx_fifo_oflows;
1920 	stat->rx_rrs_errs += smb->rx_rrs_errs;
1921 	stat->rx_alignerrs += smb->rx_alignerrs;
1922 	stat->rx_bcast_bytes += smb->rx_bcast_bytes;
1923 	stat->rx_mcast_bytes += smb->rx_mcast_bytes;
1924 	stat->rx_pkts_filtered += smb->rx_pkts_filtered;
1925 
1926 	/* Tx stats. */
1927 	stat->tx_frames += smb->tx_frames;
1928 	stat->tx_bcast_frames += smb->tx_bcast_frames;
1929 	stat->tx_mcast_frames += smb->tx_mcast_frames;
1930 	stat->tx_pause_frames += smb->tx_pause_frames;
1931 	stat->tx_excess_defer += smb->tx_excess_defer;
1932 	stat->tx_control_frames += smb->tx_control_frames;
1933 	stat->tx_deferred += smb->tx_deferred;
1934 	stat->tx_bytes += smb->tx_bytes;
1935 	stat->tx_pkts_64 += smb->tx_pkts_64;
1936 	stat->tx_pkts_65_127 += smb->tx_pkts_65_127;
1937 	stat->tx_pkts_128_255 += smb->tx_pkts_128_255;
1938 	stat->tx_pkts_256_511 += smb->tx_pkts_256_511;
1939 	stat->tx_pkts_512_1023 += smb->tx_pkts_512_1023;
1940 	stat->tx_pkts_1024_1518 += smb->tx_pkts_1024_1518;
1941 	stat->tx_pkts_1519_max += smb->tx_pkts_1519_max;
1942 	stat->tx_single_colls += smb->tx_single_colls;
1943 	stat->tx_multi_colls += smb->tx_multi_colls;
1944 	stat->tx_late_colls += smb->tx_late_colls;
1945 	stat->tx_excess_colls += smb->tx_excess_colls;
1946 	stat->tx_abort += smb->tx_abort;
1947 	stat->tx_underrun += smb->tx_underrun;
1948 	stat->tx_desc_underrun += smb->tx_desc_underrun;
1949 	stat->tx_lenerrs += smb->tx_lenerrs;
1950 	stat->tx_pkts_truncated += smb->tx_pkts_truncated;
1951 	stat->tx_bcast_bytes += smb->tx_bcast_bytes;
1952 	stat->tx_mcast_bytes += smb->tx_mcast_bytes;
1953 
1954 	/* Update counters in ifnet. */
1955 	ifp->if_opackets += smb->tx_frames;
1956 
1957 	ifp->if_collisions += smb->tx_single_colls +
1958 	    smb->tx_multi_colls * 2 + smb->tx_late_colls +
1959 	    smb->tx_abort * HDPX_CFG_RETRY_DEFAULT;
1960 
1961 	/*
1962 	 * XXX
1963 	 * tx_pkts_truncated counter looks suspicious. It constantly
1964 	 * increments with no sign of Tx errors. This may indicate
1965 	 * the counter name is not correct one so I've removed the
1966 	 * counter in output errors.
1967 	 */
1968 	ifp->if_oerrors += smb->tx_abort + smb->tx_late_colls +
1969 	    smb->tx_underrun;
1970 
1971 	ifp->if_ipackets += smb->rx_frames;
1972 
1973 	ifp->if_ierrors += smb->rx_crcerrs + smb->rx_lenerrs +
1974 	    smb->rx_runts + smb->rx_pkts_truncated +
1975 	    smb->rx_fifo_oflows + smb->rx_rrs_errs +
1976 	    smb->rx_alignerrs;
1977 }
1978 
1979 static void
1980 ale_intr(void *xsc)
1981 {
1982 	struct ale_softc *sc = xsc;
1983 	struct ifnet *ifp = &sc->arpcom.ac_if;
1984 	uint32_t status;
1985 
1986 	ASSERT_SERIALIZED(ifp->if_serializer);
1987 
1988 	status = CSR_READ_4(sc, ALE_INTR_STATUS);
1989 	if ((status & ALE_INTRS) == 0)
1990 		return;
1991 
1992 	/* Acknowledge and disable interrupts. */
1993 	CSR_WRITE_4(sc, ALE_INTR_STATUS, status | INTR_DIS_INT);
1994 
1995 	if ((ifp->if_flags & IFF_RUNNING) != 0) {
1996 		int error;
1997 
1998 		error = ale_rxeof(sc);
1999 		if (error) {
2000 			sc->ale_stats.reset_brk_seq++;
2001 			ale_init(sc);
2002 			return;
2003 		}
2004 
2005 		if ((status & (INTR_DMA_RD_TO_RST | INTR_DMA_WR_TO_RST)) != 0) {
2006 			if ((status & INTR_DMA_RD_TO_RST) != 0)
2007 				device_printf(sc->ale_dev,
2008 				    "DMA read error! -- resetting\n");
2009 			if ((status & INTR_DMA_WR_TO_RST) != 0)
2010 				device_printf(sc->ale_dev,
2011 				    "DMA write error! -- resetting\n");
2012 			ale_init(sc);
2013 			return;
2014 		}
2015 
2016 		ale_txeof(sc);
2017 		if (!ifq_is_empty(&ifp->if_snd))
2018 			if_devstart(ifp);
2019 	}
2020 
2021 	/* Re-enable interrupts. */
2022 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0x7FFFFFFF);
2023 }
2024 
2025 static void
2026 ale_txeof(struct ale_softc *sc)
2027 {
2028 	struct ifnet *ifp = &sc->arpcom.ac_if;
2029 	struct ale_txdesc *txd;
2030 	uint32_t cons, prod;
2031 	int prog;
2032 
2033 	if (sc->ale_cdata.ale_tx_cnt == 0)
2034 		return;
2035 
2036 	bus_dmamap_sync(sc->ale_cdata.ale_tx_ring_tag,
2037 	    sc->ale_cdata.ale_tx_ring_map, BUS_DMASYNC_POSTREAD);
2038 	if ((sc->ale_flags & ALE_FLAG_TXCMB_BUG) == 0) {
2039 		bus_dmamap_sync(sc->ale_cdata.ale_tx_cmb_tag,
2040 		    sc->ale_cdata.ale_tx_cmb_map, BUS_DMASYNC_POSTREAD);
2041 		prod = *sc->ale_cdata.ale_tx_cmb & TPD_CNT_MASK;
2042 	} else
2043 		prod = CSR_READ_2(sc, ALE_TPD_CONS_IDX);
2044 	cons = sc->ale_cdata.ale_tx_cons;
2045 	/*
2046 	 * Go through our Tx list and free mbufs for those
2047 	 * frames which have been transmitted.
2048 	 */
2049 	for (prog = 0; cons != prod; prog++,
2050 	     ALE_DESC_INC(cons, ALE_TX_RING_CNT)) {
2051 		if (sc->ale_cdata.ale_tx_cnt <= 0)
2052 			break;
2053 		prog++;
2054 		ifp->if_flags &= ~IFF_OACTIVE;
2055 		sc->ale_cdata.ale_tx_cnt--;
2056 		txd = &sc->ale_cdata.ale_txdesc[cons];
2057 		if (txd->tx_m != NULL) {
2058 			/* Reclaim transmitted mbufs. */
2059 			bus_dmamap_unload(sc->ale_cdata.ale_tx_tag,
2060 			    txd->tx_dmamap);
2061 			m_freem(txd->tx_m);
2062 			txd->tx_m = NULL;
2063 		}
2064 	}
2065 
2066 	if (prog > 0) {
2067 		sc->ale_cdata.ale_tx_cons = cons;
2068 		/*
2069 		 * Unarm watchdog timer only when there is no pending
2070 		 * Tx descriptors in queue.
2071 		 */
2072 		if (sc->ale_cdata.ale_tx_cnt == 0)
2073 			ifp->if_timer = 0;
2074 	}
2075 }
2076 
2077 static void
2078 ale_rx_update_page(struct ale_softc *sc, struct ale_rx_page **page,
2079     uint32_t length, uint32_t *prod)
2080 {
2081 	struct ale_rx_page *rx_page;
2082 
2083 	rx_page = *page;
2084 	/* Update consumer position. */
2085 	rx_page->cons += roundup(length + sizeof(struct rx_rs),
2086 	    ALE_RX_PAGE_ALIGN);
2087 	if (rx_page->cons >= ALE_RX_PAGE_SZ) {
2088 		/*
2089 		 * End of Rx page reached, let hardware reuse
2090 		 * this page.
2091 		 */
2092 		rx_page->cons = 0;
2093 		*rx_page->cmb_addr = 0;
2094 		bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2095 				BUS_DMASYNC_PREWRITE);
2096 		CSR_WRITE_1(sc, ALE_RXF0_PAGE0 + sc->ale_cdata.ale_rx_curp,
2097 		    RXF_VALID);
2098 		/* Switch to alternate Rx page. */
2099 		sc->ale_cdata.ale_rx_curp ^= 1;
2100 		rx_page = *page =
2101 		    &sc->ale_cdata.ale_rx_page[sc->ale_cdata.ale_rx_curp];
2102 		/* Page flipped, sync CMB and Rx page. */
2103 		bus_dmamap_sync(rx_page->page_tag, rx_page->page_map,
2104 		    BUS_DMASYNC_POSTREAD);
2105 		bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2106 		    BUS_DMASYNC_POSTREAD);
2107 		/* Sync completed, cache updated producer index. */
2108 		*prod = *rx_page->cmb_addr;
2109 	}
2110 }
2111 
2112 
2113 /*
2114  * It seems that AR81xx controller can compute partial checksum.
2115  * The partial checksum value can be used to accelerate checksum
2116  * computation for fragmented TCP/UDP packets. Upper network stack
2117  * already takes advantage of the partial checksum value in IP
2118  * reassembly stage. But I'm not sure the correctness of the
2119  * partial hardware checksum assistance due to lack of data sheet.
2120  * In addition, the Rx feature of controller that requires copying
2121  * for every frames effectively nullifies one of most nice offload
2122  * capability of controller.
2123  */
2124 static void
2125 ale_rxcsum(struct ale_softc *sc, struct mbuf *m, uint32_t status)
2126 {
2127 	struct ifnet *ifp = &sc->arpcom.ac_if;
2128 	struct ip *ip;
2129 	char *p;
2130 
2131 	m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
2132 	if ((status & ALE_RD_IPCSUM_NOK) == 0)
2133 		m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
2134 
2135 	if ((sc->ale_flags & ALE_FLAG_RXCSUM_BUG) == 0) {
2136 		if (((status & ALE_RD_IPV4_FRAG) == 0) &&
2137 		    ((status & (ALE_RD_TCP | ALE_RD_UDP)) != 0) &&
2138 		    ((status & ALE_RD_TCP_UDPCSUM_NOK) == 0)) {
2139 			m->m_pkthdr.csum_flags |=
2140 			    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2141 			m->m_pkthdr.csum_data = 0xffff;
2142 		}
2143 	} else {
2144 		if ((status & (ALE_RD_TCP | ALE_RD_UDP)) != 0 &&
2145 		    (status & ALE_RD_TCP_UDPCSUM_NOK) == 0) {
2146 			p = mtod(m, char *);
2147 			p += ETHER_HDR_LEN;
2148 			if ((status & ALE_RD_802_3) != 0)
2149 				p += LLC_SNAPFRAMELEN;
2150 			if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0 &&
2151 			    (status & ALE_RD_VLAN) != 0)
2152 				p += EVL_ENCAPLEN;
2153 			ip = (struct ip *)p;
2154 			if (ip->ip_off != 0 && (status & ALE_RD_IPV4_DF) == 0)
2155 				return;
2156 			m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
2157 			    CSUM_PSEUDO_HDR;
2158 			m->m_pkthdr.csum_data = 0xffff;
2159 		}
2160 	}
2161 	/*
2162 	 * Don't mark bad checksum for TCP/UDP frames
2163 	 * as fragmented frames may always have set
2164 	 * bad checksummed bit of frame status.
2165 	 */
2166 }
2167 
2168 /* Process received frames. */
2169 static int
2170 ale_rxeof(struct ale_softc *sc)
2171 {
2172 	struct ifnet *ifp = &sc->arpcom.ac_if;
2173 	struct ale_rx_page *rx_page;
2174 	struct rx_rs *rs;
2175 	struct mbuf *m;
2176 	uint32_t length, prod, seqno, status, vtags;
2177 	int prog;
2178 
2179 	rx_page = &sc->ale_cdata.ale_rx_page[sc->ale_cdata.ale_rx_curp];
2180 	bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2181 			BUS_DMASYNC_POSTREAD);
2182 	bus_dmamap_sync(rx_page->page_tag, rx_page->page_map,
2183 			BUS_DMASYNC_POSTREAD);
2184 	/*
2185 	 * Don't directly access producer index as hardware may
2186 	 * update it while Rx handler is in progress. It would
2187 	 * be even better if there is a way to let hardware
2188 	 * know how far driver processed its received frames.
2189 	 * Alternatively, hardware could provide a way to disable
2190 	 * CMB updates until driver acknowledges the end of CMB
2191 	 * access.
2192 	 */
2193 	prod = *rx_page->cmb_addr;
2194 	for (prog = 0; ; prog++) {
2195 		if (rx_page->cons >= prod)
2196 			break;
2197 		rs = (struct rx_rs *)(rx_page->page_addr + rx_page->cons);
2198 		seqno = ALE_RX_SEQNO(le32toh(rs->seqno));
2199 		if (sc->ale_cdata.ale_rx_seqno != seqno) {
2200 			/*
2201 			 * Normally I believe this should not happen unless
2202 			 * severe driver bug or corrupted memory. However
2203 			 * it seems to happen under certain conditions which
2204 			 * is triggered by abrupt Rx events such as initiation
2205 			 * of bulk transfer of remote host. It's not easy to
2206 			 * reproduce this and I doubt it could be related
2207 			 * with FIFO overflow of hardware or activity of Tx
2208 			 * CMB updates. I also remember similar behaviour
2209 			 * seen on RealTek 8139 which uses resembling Rx
2210 			 * scheme.
2211 			 */
2212 			if (bootverbose)
2213 				device_printf(sc->ale_dev,
2214 				    "garbled seq: %u, expected: %u -- "
2215 				    "resetting!\n", seqno,
2216 				    sc->ale_cdata.ale_rx_seqno);
2217 			return (EIO);
2218 		}
2219 		/* Frame received. */
2220 		sc->ale_cdata.ale_rx_seqno++;
2221 		length = ALE_RX_BYTES(le32toh(rs->length));
2222 		status = le32toh(rs->flags);
2223 		if ((status & ALE_RD_ERROR) != 0) {
2224 			/*
2225 			 * We want to pass the following frames to upper
2226 			 * layer regardless of error status of Rx return
2227 			 * status.
2228 			 *
2229 			 *  o IP/TCP/UDP checksum is bad.
2230 			 *  o frame length and protocol specific length
2231 			 *     does not match.
2232 			 */
2233 			if ((status & (ALE_RD_CRC | ALE_RD_CODE |
2234 			    ALE_RD_DRIBBLE | ALE_RD_RUNT | ALE_RD_OFLOW |
2235 			    ALE_RD_TRUNC)) != 0) {
2236 				ale_rx_update_page(sc, &rx_page, length, &prod);
2237 				continue;
2238 			}
2239 		}
2240 		/*
2241 		 * m_devget(9) is major bottle-neck of ale(4)(It comes
2242 		 * from hardware limitation). For jumbo frames we could
2243 		 * get a slightly better performance if driver use
2244 		 * m_getjcl(9) with proper buffer size argument. However
2245 		 * that would make code more complicated and I don't
2246 		 * think users would expect good Rx performance numbers
2247 		 * on these low-end consumer ethernet controller.
2248 		 */
2249 		m = m_devget((char *)(rs + 1), length - ETHER_CRC_LEN,
2250 		    ETHER_ALIGN, ifp, NULL);
2251 		if (m == NULL) {
2252 			ifp->if_iqdrops++;
2253 			ale_rx_update_page(sc, &rx_page, length, &prod);
2254 			continue;
2255 		}
2256 		if ((ifp->if_capenable & IFCAP_RXCSUM) != 0 &&
2257 		    (status & ALE_RD_IPV4) != 0)
2258 			ale_rxcsum(sc, m, status);
2259 		if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0 &&
2260 		    (status & ALE_RD_VLAN) != 0) {
2261 			vtags = ALE_RX_VLAN(le32toh(rs->vtags));
2262 			m->m_pkthdr.ether_vlantag = ALE_RX_VLAN_TAG(vtags);
2263 			m->m_flags |= M_VLANTAG;
2264 		}
2265 
2266 		/* Pass it to upper layer. */
2267 		ifp->if_input(ifp, m);
2268 
2269 		ale_rx_update_page(sc, &rx_page, length, &prod);
2270 	}
2271 	return 0;
2272 }
2273 
2274 static void
2275 ale_tick(void *xsc)
2276 {
2277 	struct ale_softc *sc = xsc;
2278 	struct ifnet *ifp = &sc->arpcom.ac_if;
2279 	struct mii_data *mii;
2280 
2281 	lwkt_serialize_enter(ifp->if_serializer);
2282 
2283 	mii = device_get_softc(sc->ale_miibus);
2284 	mii_tick(mii);
2285 	ale_stats_update(sc);
2286 
2287 	callout_reset(&sc->ale_tick_ch, hz, ale_tick, sc);
2288 
2289 	lwkt_serialize_exit(ifp->if_serializer);
2290 }
2291 
2292 static void
2293 ale_reset(struct ale_softc *sc)
2294 {
2295 	uint32_t reg;
2296 	int i;
2297 
2298 	/* Initialize PCIe module. From Linux. */
2299 	CSR_WRITE_4(sc, 0x1008, CSR_READ_4(sc, 0x1008) | 0x8000);
2300 
2301 	CSR_WRITE_4(sc, ALE_MASTER_CFG, MASTER_RESET);
2302 	for (i = ALE_RESET_TIMEOUT; i > 0; i--) {
2303 		DELAY(10);
2304 		if ((CSR_READ_4(sc, ALE_MASTER_CFG) & MASTER_RESET) == 0)
2305 			break;
2306 	}
2307 	if (i == 0)
2308 		device_printf(sc->ale_dev, "master reset timeout!\n");
2309 
2310 	for (i = ALE_RESET_TIMEOUT; i > 0; i--) {
2311 		if ((reg = CSR_READ_4(sc, ALE_IDLE_STATUS)) == 0)
2312 			break;
2313 		DELAY(10);
2314 	}
2315 
2316 	if (i == 0)
2317 		device_printf(sc->ale_dev, "reset timeout(0x%08x)!\n", reg);
2318 }
2319 
2320 static void
2321 ale_init(void *xsc)
2322 {
2323 	struct ale_softc *sc = xsc;
2324 	struct ifnet *ifp = &sc->arpcom.ac_if;
2325 	struct mii_data *mii;
2326 	uint8_t eaddr[ETHER_ADDR_LEN];
2327 	bus_addr_t paddr;
2328 	uint32_t reg, rxf_hi, rxf_lo;
2329 
2330 	ASSERT_SERIALIZED(ifp->if_serializer);
2331 
2332 	mii = device_get_softc(sc->ale_miibus);
2333 
2334 	/*
2335 	 * Cancel any pending I/O.
2336 	 */
2337 	ale_stop(sc);
2338 
2339 	/*
2340 	 * Reset the chip to a known state.
2341 	 */
2342 	ale_reset(sc);
2343 
2344 	/* Initialize Tx descriptors, DMA memory blocks. */
2345 	ale_init_rx_pages(sc);
2346 	ale_init_tx_ring(sc);
2347 
2348 	/* Reprogram the station address. */
2349 	bcopy(IF_LLADDR(ifp), eaddr, ETHER_ADDR_LEN);
2350 	CSR_WRITE_4(sc, ALE_PAR0,
2351 	    eaddr[2] << 24 | eaddr[3] << 16 | eaddr[4] << 8 | eaddr[5]);
2352 	CSR_WRITE_4(sc, ALE_PAR1, eaddr[0] << 8 | eaddr[1]);
2353 
2354 	/*
2355 	 * Clear WOL status and disable all WOL feature as WOL
2356 	 * would interfere Rx operation under normal environments.
2357 	 */
2358 	CSR_READ_4(sc, ALE_WOL_CFG);
2359 	CSR_WRITE_4(sc, ALE_WOL_CFG, 0);
2360 
2361 	/*
2362 	 * Set Tx descriptor/RXF0/CMB base addresses. They share
2363 	 * the same high address part of DMAable region.
2364 	 */
2365 	paddr = sc->ale_cdata.ale_tx_ring_paddr;
2366 	CSR_WRITE_4(sc, ALE_TPD_ADDR_HI, ALE_ADDR_HI(paddr));
2367 	CSR_WRITE_4(sc, ALE_TPD_ADDR_LO, ALE_ADDR_LO(paddr));
2368 	CSR_WRITE_4(sc, ALE_TPD_CNT,
2369 	    (ALE_TX_RING_CNT << TPD_CNT_SHIFT) & TPD_CNT_MASK);
2370 
2371 	/* Set Rx page base address, note we use single queue. */
2372 	paddr = sc->ale_cdata.ale_rx_page[0].page_paddr;
2373 	CSR_WRITE_4(sc, ALE_RXF0_PAGE0_ADDR_LO, ALE_ADDR_LO(paddr));
2374 	paddr = sc->ale_cdata.ale_rx_page[1].page_paddr;
2375 	CSR_WRITE_4(sc, ALE_RXF0_PAGE1_ADDR_LO, ALE_ADDR_LO(paddr));
2376 
2377 	/* Set Tx/Rx CMB addresses. */
2378 	paddr = sc->ale_cdata.ale_tx_cmb_paddr;
2379 	CSR_WRITE_4(sc, ALE_TX_CMB_ADDR_LO, ALE_ADDR_LO(paddr));
2380 	paddr = sc->ale_cdata.ale_rx_page[0].cmb_paddr;
2381 	CSR_WRITE_4(sc, ALE_RXF0_CMB0_ADDR_LO, ALE_ADDR_LO(paddr));
2382 	paddr = sc->ale_cdata.ale_rx_page[1].cmb_paddr;
2383 	CSR_WRITE_4(sc, ALE_RXF0_CMB1_ADDR_LO, ALE_ADDR_LO(paddr));
2384 
2385 	/* Mark RXF0 is valid. */
2386 	CSR_WRITE_1(sc, ALE_RXF0_PAGE0, RXF_VALID);
2387 	CSR_WRITE_1(sc, ALE_RXF0_PAGE1, RXF_VALID);
2388 	/*
2389 	 * No need to initialize RFX1/RXF2/RXF3. We don't use
2390 	 * multi-queue yet.
2391 	 */
2392 
2393 	/* Set Rx page size, excluding guard frame size. */
2394 	CSR_WRITE_4(sc, ALE_RXF_PAGE_SIZE, ALE_RX_PAGE_SZ);
2395 
2396 	/* Tell hardware that we're ready to load DMA blocks. */
2397 	CSR_WRITE_4(sc, ALE_DMA_BLOCK, DMA_BLOCK_LOAD);
2398 
2399 	/* Set Rx/Tx interrupt trigger threshold. */
2400 	CSR_WRITE_4(sc, ALE_INT_TRIG_THRESH, (1 << INT_TRIG_RX_THRESH_SHIFT) |
2401 	    (4 << INT_TRIG_TX_THRESH_SHIFT));
2402 	/*
2403 	 * XXX
2404 	 * Set interrupt trigger timer, its purpose and relation
2405 	 * with interrupt moderation mechanism is not clear yet.
2406 	 */
2407 	CSR_WRITE_4(sc, ALE_INT_TRIG_TIMER,
2408 	    ((ALE_USECS(10) << INT_TRIG_RX_TIMER_SHIFT) |
2409 	    (ALE_USECS(1000) << INT_TRIG_TX_TIMER_SHIFT)));
2410 
2411 	/* Configure interrupt moderation timer. */
2412 	reg = ALE_USECS(sc->ale_int_rx_mod) << IM_TIMER_RX_SHIFT;
2413 	reg |= ALE_USECS(sc->ale_int_tx_mod) << IM_TIMER_TX_SHIFT;
2414 	CSR_WRITE_4(sc, ALE_IM_TIMER, reg);
2415 	reg = CSR_READ_4(sc, ALE_MASTER_CFG);
2416 	reg &= ~(MASTER_CHIP_REV_MASK | MASTER_CHIP_ID_MASK);
2417 	reg &= ~(MASTER_IM_RX_TIMER_ENB | MASTER_IM_TX_TIMER_ENB);
2418 	if (ALE_USECS(sc->ale_int_rx_mod) != 0)
2419 		reg |= MASTER_IM_RX_TIMER_ENB;
2420 	if (ALE_USECS(sc->ale_int_tx_mod) != 0)
2421 		reg |= MASTER_IM_TX_TIMER_ENB;
2422 	CSR_WRITE_4(sc, ALE_MASTER_CFG, reg);
2423 	CSR_WRITE_2(sc, ALE_INTR_CLR_TIMER, ALE_USECS(1000));
2424 
2425 	/* Set Maximum frame size of controller. */
2426 	if (ifp->if_mtu < ETHERMTU)
2427 		sc->ale_max_frame_size = ETHERMTU;
2428 	else
2429 		sc->ale_max_frame_size = ifp->if_mtu;
2430 	sc->ale_max_frame_size += ETHER_HDR_LEN + EVL_ENCAPLEN + ETHER_CRC_LEN;
2431 	CSR_WRITE_4(sc, ALE_FRAME_SIZE, sc->ale_max_frame_size);
2432 
2433 	/* Configure IPG/IFG parameters. */
2434 	CSR_WRITE_4(sc, ALE_IPG_IFG_CFG,
2435 	    ((IPG_IFG_IPGT_DEFAULT << IPG_IFG_IPGT_SHIFT) & IPG_IFG_IPGT_MASK) |
2436 	    ((IPG_IFG_MIFG_DEFAULT << IPG_IFG_MIFG_SHIFT) & IPG_IFG_MIFG_MASK) |
2437 	    ((IPG_IFG_IPG1_DEFAULT << IPG_IFG_IPG1_SHIFT) & IPG_IFG_IPG1_MASK) |
2438 	    ((IPG_IFG_IPG2_DEFAULT << IPG_IFG_IPG2_SHIFT) & IPG_IFG_IPG2_MASK));
2439 
2440 	/* Set parameters for half-duplex media. */
2441 	CSR_WRITE_4(sc, ALE_HDPX_CFG,
2442 	    ((HDPX_CFG_LCOL_DEFAULT << HDPX_CFG_LCOL_SHIFT) &
2443 	    HDPX_CFG_LCOL_MASK) |
2444 	    ((HDPX_CFG_RETRY_DEFAULT << HDPX_CFG_RETRY_SHIFT) &
2445 	    HDPX_CFG_RETRY_MASK) | HDPX_CFG_EXC_DEF_EN |
2446 	    ((HDPX_CFG_ABEBT_DEFAULT << HDPX_CFG_ABEBT_SHIFT) &
2447 	    HDPX_CFG_ABEBT_MASK) |
2448 	    ((HDPX_CFG_JAMIPG_DEFAULT << HDPX_CFG_JAMIPG_SHIFT) &
2449 	    HDPX_CFG_JAMIPG_MASK));
2450 
2451 	/* Configure Tx jumbo frame parameters. */
2452 	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0) {
2453 		if (ifp->if_mtu < ETHERMTU)
2454 			reg = sc->ale_max_frame_size;
2455 		else if (ifp->if_mtu < 6 * 1024)
2456 			reg = (sc->ale_max_frame_size * 2) / 3;
2457 		else
2458 			reg = sc->ale_max_frame_size / 2;
2459 		CSR_WRITE_4(sc, ALE_TX_JUMBO_THRESH,
2460 		    roundup(reg, TX_JUMBO_THRESH_UNIT) >>
2461 		    TX_JUMBO_THRESH_UNIT_SHIFT);
2462 	}
2463 
2464 	/* Configure TxQ. */
2465 	reg = (128 << (sc->ale_dma_rd_burst >> DMA_CFG_RD_BURST_SHIFT))
2466 	    << TXQ_CFG_TX_FIFO_BURST_SHIFT;
2467 	reg |= (TXQ_CFG_TPD_BURST_DEFAULT << TXQ_CFG_TPD_BURST_SHIFT) &
2468 	    TXQ_CFG_TPD_BURST_MASK;
2469 	CSR_WRITE_4(sc, ALE_TXQ_CFG, reg | TXQ_CFG_ENHANCED_MODE | TXQ_CFG_ENB);
2470 
2471 	/* Configure Rx jumbo frame & flow control parameters. */
2472 	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0) {
2473 		reg = roundup(sc->ale_max_frame_size, RX_JUMBO_THRESH_UNIT);
2474 		CSR_WRITE_4(sc, ALE_RX_JUMBO_THRESH,
2475 		    (((reg >> RX_JUMBO_THRESH_UNIT_SHIFT) <<
2476 		    RX_JUMBO_THRESH_MASK_SHIFT) & RX_JUMBO_THRESH_MASK) |
2477 		    ((RX_JUMBO_LKAH_DEFAULT << RX_JUMBO_LKAH_SHIFT) &
2478 		    RX_JUMBO_LKAH_MASK));
2479 		reg = CSR_READ_4(sc, ALE_SRAM_RX_FIFO_LEN);
2480 		rxf_hi = (reg * 7) / 10;
2481 		rxf_lo = (reg * 3)/ 10;
2482 		CSR_WRITE_4(sc, ALE_RX_FIFO_PAUSE_THRESH,
2483 		    ((rxf_lo << RX_FIFO_PAUSE_THRESH_LO_SHIFT) &
2484 		    RX_FIFO_PAUSE_THRESH_LO_MASK) |
2485 		    ((rxf_hi << RX_FIFO_PAUSE_THRESH_HI_SHIFT) &
2486 		     RX_FIFO_PAUSE_THRESH_HI_MASK));
2487 	}
2488 
2489 	/* Disable RSS. */
2490 	CSR_WRITE_4(sc, ALE_RSS_IDT_TABLE0, 0);
2491 	CSR_WRITE_4(sc, ALE_RSS_CPU, 0);
2492 
2493 	/* Configure RxQ. */
2494 	CSR_WRITE_4(sc, ALE_RXQ_CFG,
2495 	    RXQ_CFG_ALIGN_32 | RXQ_CFG_CUT_THROUGH_ENB | RXQ_CFG_ENB);
2496 
2497 	/* Configure DMA parameters. */
2498 	reg = 0;
2499 	if ((sc->ale_flags & ALE_FLAG_TXCMB_BUG) == 0)
2500 		reg |= DMA_CFG_TXCMB_ENB;
2501 	CSR_WRITE_4(sc, ALE_DMA_CFG,
2502 	    DMA_CFG_OUT_ORDER | DMA_CFG_RD_REQ_PRI | DMA_CFG_RCB_64 |
2503 	    sc->ale_dma_rd_burst | reg |
2504 	    sc->ale_dma_wr_burst | DMA_CFG_RXCMB_ENB |
2505 	    ((DMA_CFG_RD_DELAY_CNT_DEFAULT << DMA_CFG_RD_DELAY_CNT_SHIFT) &
2506 	    DMA_CFG_RD_DELAY_CNT_MASK) |
2507 	    ((DMA_CFG_WR_DELAY_CNT_DEFAULT << DMA_CFG_WR_DELAY_CNT_SHIFT) &
2508 	    DMA_CFG_WR_DELAY_CNT_MASK));
2509 
2510 	/*
2511 	 * Hardware can be configured to issue SMB interrupt based
2512 	 * on programmed interval. Since there is a callout that is
2513 	 * invoked for every hz in driver we use that instead of
2514 	 * relying on periodic SMB interrupt.
2515 	 */
2516 	CSR_WRITE_4(sc, ALE_SMB_STAT_TIMER, ALE_USECS(0));
2517 
2518 	/* Clear MAC statistics. */
2519 	ale_stats_clear(sc);
2520 
2521 	/*
2522 	 * Configure Tx/Rx MACs.
2523 	 *  - Auto-padding for short frames.
2524 	 *  - Enable CRC generation.
2525 	 *  Actual reconfiguration of MAC for resolved speed/duplex
2526 	 *  is followed after detection of link establishment.
2527 	 *  AR81xx always does checksum computation regardless of
2528 	 *  MAC_CFG_RXCSUM_ENB bit. In fact, setting the bit will
2529 	 *  cause Rx handling issue for fragmented IP datagrams due
2530 	 *  to silicon bug.
2531 	 */
2532 	reg = MAC_CFG_TX_CRC_ENB | MAC_CFG_TX_AUTO_PAD | MAC_CFG_FULL_DUPLEX |
2533 	    ((MAC_CFG_PREAMBLE_DEFAULT << MAC_CFG_PREAMBLE_SHIFT) &
2534 	    MAC_CFG_PREAMBLE_MASK);
2535 	if ((sc->ale_flags & ALE_FLAG_FASTETHER) != 0)
2536 		reg |= MAC_CFG_SPEED_10_100;
2537 	else
2538 		reg |= MAC_CFG_SPEED_1000;
2539 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2540 
2541 	/* Set up the receive filter. */
2542 	ale_rxfilter(sc);
2543 	ale_rxvlan(sc);
2544 
2545 	/* Acknowledge all pending interrupts and clear it. */
2546 	CSR_WRITE_4(sc, ALE_INTR_MASK, ALE_INTRS);
2547 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0xFFFFFFFF);
2548 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0);
2549 
2550 	sc->ale_flags &= ~ALE_FLAG_LINK;
2551 
2552 	/* Switch to the current media. */
2553 	mii_mediachg(mii);
2554 
2555 	callout_reset(&sc->ale_tick_ch, hz, ale_tick, sc);
2556 
2557 	ifp->if_flags |= IFF_RUNNING;
2558 	ifp->if_flags &= ~IFF_OACTIVE;
2559 }
2560 
2561 static void
2562 ale_stop(struct ale_softc *sc)
2563 {
2564 	struct ifnet *ifp = &sc->arpcom.ac_if;
2565 	struct ale_txdesc *txd;
2566 	uint32_t reg;
2567 	int i;
2568 
2569 	ASSERT_SERIALIZED(ifp->if_serializer);
2570 
2571 	/*
2572 	 * Mark the interface down and cancel the watchdog timer.
2573 	 */
2574 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2575 	ifp->if_timer = 0;
2576 
2577 	callout_stop(&sc->ale_tick_ch);
2578 	sc->ale_flags &= ~ALE_FLAG_LINK;
2579 
2580 	ale_stats_update(sc);
2581 
2582 	/* Disable interrupts. */
2583 	CSR_WRITE_4(sc, ALE_INTR_MASK, 0);
2584 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0xFFFFFFFF);
2585 
2586 	/* Disable queue processing and DMA. */
2587 	reg = CSR_READ_4(sc, ALE_TXQ_CFG);
2588 	reg &= ~TXQ_CFG_ENB;
2589 	CSR_WRITE_4(sc, ALE_TXQ_CFG, reg);
2590 	reg = CSR_READ_4(sc, ALE_RXQ_CFG);
2591 	reg &= ~RXQ_CFG_ENB;
2592 	CSR_WRITE_4(sc, ALE_RXQ_CFG, reg);
2593 	reg = CSR_READ_4(sc, ALE_DMA_CFG);
2594 	reg &= ~(DMA_CFG_TXCMB_ENB | DMA_CFG_RXCMB_ENB);
2595 	CSR_WRITE_4(sc, ALE_DMA_CFG, reg);
2596 	DELAY(1000);
2597 
2598 	/* Stop Rx/Tx MACs. */
2599 	ale_stop_mac(sc);
2600 
2601 	/* Disable interrupts again? XXX */
2602 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0xFFFFFFFF);
2603 
2604 	/*
2605 	 * Free TX mbufs still in the queues.
2606 	 */
2607 	for (i = 0; i < ALE_TX_RING_CNT; i++) {
2608 		txd = &sc->ale_cdata.ale_txdesc[i];
2609 		if (txd->tx_m != NULL) {
2610 			bus_dmamap_unload(sc->ale_cdata.ale_tx_tag,
2611 			    txd->tx_dmamap);
2612 			m_freem(txd->tx_m);
2613 			txd->tx_m = NULL;
2614 		}
2615         }
2616 }
2617 
2618 static void
2619 ale_stop_mac(struct ale_softc *sc)
2620 {
2621 	uint32_t reg;
2622 	int i;
2623 
2624 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
2625 	if ((reg & (MAC_CFG_TX_ENB | MAC_CFG_RX_ENB)) != 0) {
2626 		reg &= ~MAC_CFG_TX_ENB | MAC_CFG_RX_ENB;
2627 		CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2628 	}
2629 
2630 	for (i = ALE_TIMEOUT; i > 0; i--) {
2631 		reg = CSR_READ_4(sc, ALE_IDLE_STATUS);
2632 		if (reg == 0)
2633 			break;
2634 		DELAY(10);
2635 	}
2636 	if (i == 0)
2637 		device_printf(sc->ale_dev,
2638 		    "could not disable Tx/Rx MAC(0x%08x)!\n", reg);
2639 }
2640 
2641 static void
2642 ale_init_tx_ring(struct ale_softc *sc)
2643 {
2644 	struct ale_txdesc *txd;
2645 	int i;
2646 
2647 	sc->ale_cdata.ale_tx_prod = 0;
2648 	sc->ale_cdata.ale_tx_cons = 0;
2649 	sc->ale_cdata.ale_tx_cnt = 0;
2650 
2651 	bzero(sc->ale_cdata.ale_tx_ring, ALE_TX_RING_SZ);
2652 	bzero(sc->ale_cdata.ale_tx_cmb, ALE_TX_CMB_SZ);
2653 	for (i = 0; i < ALE_TX_RING_CNT; i++) {
2654 		txd = &sc->ale_cdata.ale_txdesc[i];
2655 		txd->tx_m = NULL;
2656 	}
2657 	*sc->ale_cdata.ale_tx_cmb = 0;
2658 	bus_dmamap_sync(sc->ale_cdata.ale_tx_cmb_tag,
2659 	    sc->ale_cdata.ale_tx_cmb_map,
2660 	    BUS_DMASYNC_PREWRITE);
2661 	bus_dmamap_sync(sc->ale_cdata.ale_tx_ring_tag,
2662 	    sc->ale_cdata.ale_tx_ring_map,
2663 	    BUS_DMASYNC_PREWRITE);
2664 }
2665 
2666 static void
2667 ale_init_rx_pages(struct ale_softc *sc)
2668 {
2669 	struct ale_rx_page *rx_page;
2670 	int i;
2671 
2672 	sc->ale_cdata.ale_rx_seqno = 0;
2673 	sc->ale_cdata.ale_rx_curp = 0;
2674 
2675 	for (i = 0; i < ALE_RX_PAGES; i++) {
2676 		rx_page = &sc->ale_cdata.ale_rx_page[i];
2677 		bzero(rx_page->page_addr, sc->ale_pagesize);
2678 		bzero(rx_page->cmb_addr, ALE_RX_CMB_SZ);
2679 		rx_page->cons = 0;
2680 		*rx_page->cmb_addr = 0;
2681 		bus_dmamap_sync(rx_page->page_tag, rx_page->page_map,
2682 				BUS_DMASYNC_PREWRITE);
2683 		bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2684 				BUS_DMASYNC_PREWRITE);
2685 	}
2686 }
2687 
2688 static void
2689 ale_rxvlan(struct ale_softc *sc)
2690 {
2691 	struct ifnet *ifp;
2692 	uint32_t reg;
2693 
2694 	ifp = &sc->arpcom.ac_if;
2695 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
2696 	reg &= ~MAC_CFG_VLAN_TAG_STRIP;
2697 	if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
2698 		reg |= MAC_CFG_VLAN_TAG_STRIP;
2699 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2700 }
2701 
2702 static void
2703 ale_rxfilter(struct ale_softc *sc)
2704 {
2705 	struct ifnet *ifp;
2706 	struct ifmultiaddr *ifma;
2707 	uint32_t crc;
2708 	uint32_t mchash[2];
2709 	uint32_t rxcfg;
2710 
2711 	ifp = &sc->arpcom.ac_if;
2712 
2713 	rxcfg = CSR_READ_4(sc, ALE_MAC_CFG);
2714 	rxcfg &= ~(MAC_CFG_ALLMULTI | MAC_CFG_BCAST | MAC_CFG_PROMISC);
2715 	if ((ifp->if_flags & IFF_BROADCAST) != 0)
2716 		rxcfg |= MAC_CFG_BCAST;
2717 	if ((ifp->if_flags & (IFF_PROMISC | IFF_ALLMULTI)) != 0) {
2718 		if ((ifp->if_flags & IFF_PROMISC) != 0)
2719 			rxcfg |= MAC_CFG_PROMISC;
2720 		if ((ifp->if_flags & IFF_ALLMULTI) != 0)
2721 			rxcfg |= MAC_CFG_ALLMULTI;
2722 		CSR_WRITE_4(sc, ALE_MAR0, 0xFFFFFFFF);
2723 		CSR_WRITE_4(sc, ALE_MAR1, 0xFFFFFFFF);
2724 		CSR_WRITE_4(sc, ALE_MAC_CFG, rxcfg);
2725 		return;
2726 	}
2727 
2728 	/* Program new filter. */
2729 	bzero(mchash, sizeof(mchash));
2730 
2731 	LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2732 		if (ifma->ifma_addr->sa_family != AF_LINK)
2733 			continue;
2734 		crc = ether_crc32_le(LLADDR((struct sockaddr_dl *)
2735 		    ifma->ifma_addr), ETHER_ADDR_LEN);
2736 		mchash[crc >> 31] |= 1 << ((crc >> 26) & 0x1f);
2737 	}
2738 
2739 	CSR_WRITE_4(sc, ALE_MAR0, mchash[0]);
2740 	CSR_WRITE_4(sc, ALE_MAR1, mchash[1]);
2741 	CSR_WRITE_4(sc, ALE_MAC_CFG, rxcfg);
2742 }
2743 
2744 static int
2745 sysctl_hw_ale_int_mod(SYSCTL_HANDLER_ARGS)
2746 {
2747 	return (sysctl_int_range(oidp, arg1, arg2, req,
2748 	    ALE_IM_TIMER_MIN, ALE_IM_TIMER_MAX));
2749 }
2750 
2751 static void
2752 ale_dmamap_buf_cb(void *xctx, bus_dma_segment_t *segs, int nsegs,
2753 		  bus_size_t mapsz __unused, int error)
2754 {
2755 	struct ale_dmamap_ctx *ctx = xctx;
2756 	int i;
2757 
2758 	if (error)
2759 		return;
2760 
2761 	if (nsegs > ctx->nsegs) {
2762 		ctx->nsegs = 0;
2763 		return;
2764 	}
2765 
2766 	ctx->nsegs = nsegs;
2767 	for (i = 0; i < nsegs; ++i)
2768 		ctx->segs[i] = segs[i];
2769 }
2770