xref: /dragonfly/sys/dev/netif/ale/if_ale.c (revision cfd1aba3)
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 "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 *, struct ifaltq_subque *);
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_VERSION(if_ale, 1);
176 MODULE_DEPEND(if_ale, miibus, 1, 1, 1);
177 DRIVER_MODULE(if_ale, pci, ale_driver, ale_devclass, NULL, NULL);
178 DRIVER_MODULE(miibus, ale, miibus_driver, miibus_devclass, NULL, NULL);
179 
180 static int
181 ale_miibus_readreg(device_t dev, int phy, int reg)
182 {
183 	struct ale_softc *sc;
184 	uint32_t v;
185 	int i;
186 
187 	sc = device_get_softc(dev);
188 
189 	if (phy != sc->ale_phyaddr)
190 		return (0);
191 
192 	if (sc->ale_flags & ALE_FLAG_FASTETHER) {
193 		if (reg == MII_100T2CR || reg == MII_100T2SR ||
194 		    reg == MII_EXTSR)
195 			return (0);
196 	}
197 
198 	CSR_WRITE_4(sc, ALE_MDIO, MDIO_OP_EXECUTE | MDIO_OP_READ |
199 	    MDIO_SUP_PREAMBLE | MDIO_CLK_25_4 | MDIO_REG_ADDR(reg));
200 	for (i = ALE_PHY_TIMEOUT; i > 0; i--) {
201 		DELAY(5);
202 		v = CSR_READ_4(sc, ALE_MDIO);
203 		if ((v & (MDIO_OP_EXECUTE | MDIO_OP_BUSY)) == 0)
204 			break;
205 	}
206 
207 	if (i == 0) {
208 		device_printf(sc->ale_dev, "phy read timeout : %d\n", reg);
209 		return (0);
210 	}
211 
212 	return ((v & MDIO_DATA_MASK) >> MDIO_DATA_SHIFT);
213 }
214 
215 static int
216 ale_miibus_writereg(device_t dev, int phy, int reg, int val)
217 {
218 	struct ale_softc *sc;
219 	uint32_t v;
220 	int i;
221 
222 	sc = device_get_softc(dev);
223 
224 	if (phy != sc->ale_phyaddr)
225 		return (0);
226 
227 	if (sc->ale_flags & ALE_FLAG_FASTETHER) {
228 		if (reg == MII_100T2CR || reg == MII_100T2SR ||
229 		    reg == MII_EXTSR)
230 			return (0);
231 	}
232 
233 	CSR_WRITE_4(sc, ALE_MDIO, MDIO_OP_EXECUTE | MDIO_OP_WRITE |
234 	    (val & MDIO_DATA_MASK) << MDIO_DATA_SHIFT |
235 	    MDIO_SUP_PREAMBLE | MDIO_CLK_25_4 | MDIO_REG_ADDR(reg));
236 	for (i = ALE_PHY_TIMEOUT; i > 0; i--) {
237 		DELAY(5);
238 		v = CSR_READ_4(sc, ALE_MDIO);
239 		if ((v & (MDIO_OP_EXECUTE | MDIO_OP_BUSY)) == 0)
240 			break;
241 	}
242 
243 	if (i == 0)
244 		device_printf(sc->ale_dev, "phy write timeout : %d\n", reg);
245 
246 	return (0);
247 }
248 
249 static void
250 ale_miibus_statchg(device_t dev)
251 {
252 	struct ale_softc *sc = device_get_softc(dev);
253 	struct ifnet *ifp = &sc->arpcom.ac_if;
254 	struct mii_data *mii;
255 	uint32_t reg;
256 
257 	ASSERT_SERIALIZED(ifp->if_serializer);
258 
259 	if ((ifp->if_flags & IFF_RUNNING) == 0)
260 		return;
261 
262 	mii = device_get_softc(sc->ale_miibus);
263 
264 	sc->ale_flags &= ~ALE_FLAG_LINK;
265 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
266 	    (IFM_ACTIVE | IFM_AVALID)) {
267 		switch (IFM_SUBTYPE(mii->mii_media_active)) {
268 		case IFM_10_T:
269 		case IFM_100_TX:
270 			sc->ale_flags |= ALE_FLAG_LINK;
271 			break;
272 
273 		case IFM_1000_T:
274 			if ((sc->ale_flags & ALE_FLAG_FASTETHER) == 0)
275 				sc->ale_flags |= ALE_FLAG_LINK;
276 			break;
277 
278 		default:
279 			break;
280 		}
281 	}
282 
283 	/* Stop Rx/Tx MACs. */
284 	ale_stop_mac(sc);
285 
286 	/* Program MACs with resolved speed/duplex/flow-control. */
287 	if ((sc->ale_flags & ALE_FLAG_LINK) != 0) {
288 		ale_mac_config(sc);
289 		/* Reenable Tx/Rx MACs. */
290 		reg = CSR_READ_4(sc, ALE_MAC_CFG);
291 		reg |= MAC_CFG_TX_ENB | MAC_CFG_RX_ENB;
292 		CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
293 	}
294 }
295 
296 static void
297 ale_mediastatus(struct ifnet *ifp, struct ifmediareq *ifmr)
298 {
299 	struct ale_softc *sc = ifp->if_softc;
300 	struct mii_data *mii = device_get_softc(sc->ale_miibus);
301 
302 	ASSERT_SERIALIZED(ifp->if_serializer);
303 
304 	mii_pollstat(mii);
305 	ifmr->ifm_status = mii->mii_media_status;
306 	ifmr->ifm_active = mii->mii_media_active;
307 }
308 
309 static int
310 ale_mediachange(struct ifnet *ifp)
311 {
312 	struct ale_softc *sc = ifp->if_softc;
313 	struct mii_data *mii = device_get_softc(sc->ale_miibus);
314 	int error;
315 
316 	ASSERT_SERIALIZED(ifp->if_serializer);
317 
318 	if (mii->mii_instance != 0) {
319 		struct mii_softc *miisc;
320 
321 		LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
322 			mii_phy_reset(miisc);
323 	}
324 	error = mii_mediachg(mii);
325 
326 	return (error);
327 }
328 
329 static int
330 ale_probe(device_t dev)
331 {
332 	struct ale_dev *sp;
333 	int i;
334 	uint16_t vendor, devid;
335 
336 	vendor = pci_get_vendor(dev);
337 	devid = pci_get_device(dev);
338 	sp = ale_devs;
339 	for (i = 0; i < NELEM(ale_devs); i++) {
340 		if (vendor == sp->ale_vendorid &&
341 		    devid == sp->ale_deviceid) {
342 			device_set_desc(dev, sp->ale_name);
343 			return (0);
344 		}
345 		sp++;
346 	}
347 
348 	return (ENXIO);
349 }
350 
351 static void
352 ale_get_macaddr(struct ale_softc *sc)
353 {
354 	uint32_t ea[2], reg;
355 	int i, vpdc;
356 
357 	reg = CSR_READ_4(sc, ALE_SPI_CTRL);
358 	if ((reg & SPI_VPD_ENB) != 0) {
359 		reg &= ~SPI_VPD_ENB;
360 		CSR_WRITE_4(sc, ALE_SPI_CTRL, reg);
361 	}
362 
363 	vpdc = pci_get_vpdcap_ptr(sc->ale_dev);
364 	if (vpdc) {
365 		/*
366 		 * PCI VPD capability found, let TWSI reload EEPROM.
367 		 * This will set ethernet address of controller.
368 		 */
369 		CSR_WRITE_4(sc, ALE_TWSI_CTRL, CSR_READ_4(sc, ALE_TWSI_CTRL) |
370 		    TWSI_CTRL_SW_LD_START);
371 		for (i = 100; i > 0; i--) {
372 			DELAY(1000);
373 			reg = CSR_READ_4(sc, ALE_TWSI_CTRL);
374 			if ((reg & TWSI_CTRL_SW_LD_START) == 0)
375 				break;
376 		}
377 		if (i == 0)
378 			device_printf(sc->ale_dev,
379 			    "reloading EEPROM timeout!\n");
380 	} else {
381 		if (bootverbose)
382 			device_printf(sc->ale_dev,
383 			    "PCI VPD capability not found!\n");
384 	}
385 
386 	ea[0] = CSR_READ_4(sc, ALE_PAR0);
387 	ea[1] = CSR_READ_4(sc, ALE_PAR1);
388 	sc->ale_eaddr[0] = (ea[1] >> 8) & 0xFF;
389 	sc->ale_eaddr[1] = (ea[1] >> 0) & 0xFF;
390 	sc->ale_eaddr[2] = (ea[0] >> 24) & 0xFF;
391 	sc->ale_eaddr[3] = (ea[0] >> 16) & 0xFF;
392 	sc->ale_eaddr[4] = (ea[0] >> 8) & 0xFF;
393 	sc->ale_eaddr[5] = (ea[0] >> 0) & 0xFF;
394 }
395 
396 static void
397 ale_phy_reset(struct ale_softc *sc)
398 {
399 	/* Reset magic from Linux. */
400 	CSR_WRITE_2(sc, ALE_GPHY_CTRL,
401 	    GPHY_CTRL_HIB_EN | GPHY_CTRL_HIB_PULSE | GPHY_CTRL_SEL_ANA_RESET |
402 	    GPHY_CTRL_PHY_PLL_ON);
403 	DELAY(1000);
404 	CSR_WRITE_2(sc, ALE_GPHY_CTRL,
405 	    GPHY_CTRL_EXT_RESET | GPHY_CTRL_HIB_EN | GPHY_CTRL_HIB_PULSE |
406 	    GPHY_CTRL_SEL_ANA_RESET | GPHY_CTRL_PHY_PLL_ON);
407 	DELAY(1000);
408 
409 #define	ATPHY_DBG_ADDR		0x1D
410 #define	ATPHY_DBG_DATA		0x1E
411 
412 	/* Enable hibernation mode. */
413 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
414 	    ATPHY_DBG_ADDR, 0x0B);
415 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
416 	    ATPHY_DBG_DATA, 0xBC00);
417 	/* Set Class A/B for all modes. */
418 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
419 	    ATPHY_DBG_ADDR, 0x00);
420 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
421 	    ATPHY_DBG_DATA, 0x02EF);
422 	/* Enable 10BT power saving. */
423 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
424 	    ATPHY_DBG_ADDR, 0x12);
425 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
426 	    ATPHY_DBG_DATA, 0x4C04);
427 	/* Adjust 1000T power. */
428 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
429 	    ATPHY_DBG_ADDR, 0x04);
430 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
431 	    ATPHY_DBG_ADDR, 0x8BBB);
432 	/* 10BT center tap voltage. */
433 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
434 	    ATPHY_DBG_ADDR, 0x05);
435 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
436 	    ATPHY_DBG_ADDR, 0x2C46);
437 
438 #undef	ATPHY_DBG_ADDR
439 #undef	ATPHY_DBG_DATA
440 	DELAY(1000);
441 }
442 
443 static int
444 ale_attach(device_t dev)
445 {
446 	struct ale_softc *sc = device_get_softc(dev);
447 	struct ifnet *ifp = &sc->arpcom.ac_if;
448 	int error = 0;
449 	uint32_t rxf_len, txf_len;
450 	uint8_t pcie_ptr;
451 
452 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
453 	sc->ale_dev = dev;
454 
455 	callout_init(&sc->ale_tick_ch);
456 
457 #ifndef BURN_BRIDGES
458 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
459 		uint32_t irq, mem;
460 
461 		irq = pci_read_config(dev, PCIR_INTLINE, 4);
462 		mem = pci_read_config(dev, ALE_PCIR_BAR, 4);
463 
464 		device_printf(dev, "chip is in D%d power mode "
465 		    "-- setting to D0\n", pci_get_powerstate(dev));
466 
467 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
468 
469 		pci_write_config(dev, PCIR_INTLINE, irq, 4);
470 		pci_write_config(dev, ALE_PCIR_BAR, mem, 4);
471 	}
472 #endif	/* !BURN_BRIDGE */
473 
474 	/* Enable bus mastering */
475 	pci_enable_busmaster(dev);
476 
477 	/*
478 	 * Allocate memory mapped IO
479 	 */
480 	sc->ale_mem_rid = ALE_PCIR_BAR;
481 	sc->ale_mem_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
482 						 &sc->ale_mem_rid, RF_ACTIVE);
483 	if (sc->ale_mem_res == NULL) {
484 		device_printf(dev, "can't allocate IO memory\n");
485 		return ENXIO;
486 	}
487 	sc->ale_mem_bt = rman_get_bustag(sc->ale_mem_res);
488 	sc->ale_mem_bh = rman_get_bushandle(sc->ale_mem_res);
489 
490 	/*
491 	 * Allocate IRQ
492 	 */
493 	sc->ale_irq_rid = 0;
494 	sc->ale_irq_res = bus_alloc_resource_any(dev, SYS_RES_IRQ,
495 						 &sc->ale_irq_rid,
496 						 RF_SHAREABLE | RF_ACTIVE);
497 	if (sc->ale_irq_res == NULL) {
498 		device_printf(dev, "can't allocate irq\n");
499 		error = ENXIO;
500 		goto fail;
501 	}
502 
503 	/* Set PHY address. */
504 	sc->ale_phyaddr = ALE_PHY_ADDR;
505 
506 	/* Reset PHY. */
507 	ale_phy_reset(sc);
508 
509 	/* Reset the ethernet controller. */
510 	ale_reset(sc);
511 
512 	/* Get PCI and chip id/revision. */
513 	sc->ale_rev = pci_get_revid(dev);
514 	if (sc->ale_rev >= 0xF0) {
515 		/* L2E Rev. B. AR8114 */
516 		sc->ale_flags |= ALE_FLAG_FASTETHER;
517 	} else {
518 		if ((CSR_READ_4(sc, ALE_PHY_STATUS) & PHY_STATUS_100M) != 0) {
519 			/* L1E AR8121 */
520 			sc->ale_flags |= ALE_FLAG_JUMBO;
521 		} else {
522 			/* L2E Rev. A. AR8113 */
523 			sc->ale_flags |= ALE_FLAG_FASTETHER;
524 		}
525 	}
526 
527 	/*
528 	 * All known controllers seems to require 4 bytes alignment
529 	 * of Tx buffers to make Tx checksum offload with custom
530 	 * checksum generation method work.
531 	 */
532 	sc->ale_flags |= ALE_FLAG_TXCSUM_BUG;
533 
534 	/*
535 	 * All known controllers seems to have issues on Rx checksum
536 	 * offload for fragmented IP datagrams.
537 	 */
538 	sc->ale_flags |= ALE_FLAG_RXCSUM_BUG;
539 
540 	/*
541 	 * Don't use Tx CMB. It is known to cause RRS update failure
542 	 * under certain circumstances. Typical phenomenon of the
543 	 * issue would be unexpected sequence number encountered in
544 	 * Rx handler.
545 	 */
546 	sc->ale_flags |= ALE_FLAG_TXCMB_BUG;
547 	sc->ale_chip_rev = CSR_READ_4(sc, ALE_MASTER_CFG) >>
548 	    MASTER_CHIP_REV_SHIFT;
549 	if (bootverbose) {
550 		device_printf(dev, "PCI device revision : 0x%04x\n",
551 		    sc->ale_rev);
552 		device_printf(dev, "Chip id/revision : 0x%04x\n",
553 		    sc->ale_chip_rev);
554 	}
555 
556 	/*
557 	 * Uninitialized hardware returns an invalid chip id/revision
558 	 * as well as 0xFFFFFFFF for Tx/Rx fifo length.
559 	 */
560 	txf_len = CSR_READ_4(sc, ALE_SRAM_TX_FIFO_LEN);
561 	rxf_len = CSR_READ_4(sc, ALE_SRAM_RX_FIFO_LEN);
562 	if (sc->ale_chip_rev == 0xFFFF || txf_len == 0xFFFFFFFF ||
563 	    rxf_len == 0xFFFFFFF) {
564 		device_printf(dev,"chip revision : 0x%04x, %u Tx FIFO "
565 		    "%u Rx FIFO -- not initialized?\n", sc->ale_chip_rev,
566 		    txf_len, rxf_len);
567 		error = ENXIO;
568 		goto fail;
569 	}
570 	device_printf(dev, "%u Tx FIFO, %u Rx FIFO\n", txf_len, rxf_len);
571 
572 	/* Get DMA parameters from PCIe device control register. */
573 	pcie_ptr = pci_get_pciecap_ptr(dev);
574 	if (pcie_ptr) {
575 		uint16_t devctl;
576 
577 		sc->ale_flags |= ALE_FLAG_PCIE;
578 		devctl = pci_read_config(dev, pcie_ptr + PCIER_DEVCTRL, 2);
579 		/* Max read request size. */
580 		sc->ale_dma_rd_burst = ((devctl >> 12) & 0x07) <<
581 		    DMA_CFG_RD_BURST_SHIFT;
582 		/* Max payload size. */
583 		sc->ale_dma_wr_burst = ((devctl >> 5) & 0x07) <<
584 		    DMA_CFG_WR_BURST_SHIFT;
585 		if (bootverbose) {
586 			device_printf(dev, "Read request size : %d bytes.\n",
587 			    128 << ((devctl >> 12) & 0x07));
588 			device_printf(dev, "TLP payload size : %d bytes.\n",
589 			    128 << ((devctl >> 5) & 0x07));
590 		}
591 	} else {
592 		sc->ale_dma_rd_burst = DMA_CFG_RD_BURST_128;
593 		sc->ale_dma_wr_burst = DMA_CFG_WR_BURST_128;
594 	}
595 
596 	/* Create device sysctl node. */
597 	ale_sysctl_node(sc);
598 
599 	if ((error = ale_dma_alloc(sc) != 0))
600 		goto fail;
601 
602 	/* Load station address. */
603 	ale_get_macaddr(sc);
604 
605 	ifp->if_softc = sc;
606 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
607 	ifp->if_ioctl = ale_ioctl;
608 	ifp->if_start = ale_start;
609 	ifp->if_init = ale_init;
610 	ifp->if_watchdog = ale_watchdog;
611 	ifq_set_maxlen(&ifp->if_snd, ALE_TX_RING_CNT - 1);
612 	ifq_set_ready(&ifp->if_snd);
613 
614 	ifp->if_capabilities = IFCAP_RXCSUM |
615 			       IFCAP_VLAN_MTU |
616 			       IFCAP_VLAN_HWTAGGING;
617 #ifdef notyet
618 	ifp->if_capabilities |= IFCAP_TXCSUM;
619 	ifp->if_hwassist = ALE_CSUM_FEATURES;
620 #endif
621 	ifp->if_capenable = ifp->if_capabilities;
622 
623 	/* Set up MII bus. */
624 	if ((error = mii_phy_probe(dev, &sc->ale_miibus, ale_mediachange,
625 	    ale_mediastatus)) != 0) {
626 		device_printf(dev, "no PHY found!\n");
627 		goto fail;
628 	}
629 
630 	ether_ifattach(ifp, sc->ale_eaddr, NULL);
631 
632 	/* Tell the upper layer(s) we support long frames. */
633 	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
634 
635 	ifq_set_cpuid(&ifp->if_snd, rman_get_cpuid(sc->ale_irq_res));
636 
637 	error = bus_setup_intr(dev, sc->ale_irq_res, INTR_MPSAFE, ale_intr, sc,
638 			       &sc->ale_irq_handle, ifp->if_serializer);
639 	if (error) {
640 		device_printf(dev, "could not set up interrupt handler.\n");
641 		ether_ifdetach(ifp);
642 		goto fail;
643 	}
644 
645 	return 0;
646 fail:
647 	ale_detach(dev);
648 	return (error);
649 }
650 
651 static int
652 ale_detach(device_t dev)
653 {
654 	struct ale_softc *sc = device_get_softc(dev);
655 
656 	if (device_is_attached(dev)) {
657 		struct ifnet *ifp = &sc->arpcom.ac_if;
658 
659 		lwkt_serialize_enter(ifp->if_serializer);
660 		sc->ale_flags |= ALE_FLAG_DETACH;
661 		ale_stop(sc);
662 		bus_teardown_intr(dev, sc->ale_irq_res, sc->ale_irq_handle);
663 		lwkt_serialize_exit(ifp->if_serializer);
664 
665 		ether_ifdetach(ifp);
666 	}
667 
668 	if (sc->ale_sysctl_tree != NULL)
669 		sysctl_ctx_free(&sc->ale_sysctl_ctx);
670 
671 	if (sc->ale_miibus != NULL)
672 		device_delete_child(dev, sc->ale_miibus);
673 	bus_generic_detach(dev);
674 
675 	if (sc->ale_irq_res != NULL) {
676 		bus_release_resource(dev, SYS_RES_IRQ, sc->ale_irq_rid,
677 				     sc->ale_irq_res);
678 	}
679 	if (sc->ale_mem_res != NULL) {
680 		bus_release_resource(dev, SYS_RES_MEMORY, sc->ale_mem_rid,
681 				     sc->ale_mem_res);
682 	}
683 
684 	ale_dma_free(sc);
685 
686 	return (0);
687 }
688 
689 #define	ALE_SYSCTL_STAT_ADD32(c, h, n, p, d)	\
690 	    SYSCTL_ADD_UINT(c, h, OID_AUTO, n, CTLFLAG_RD, p, 0, d)
691 #define	ALE_SYSCTL_STAT_ADD64(c, h, n, p, d)	\
692 	    SYSCTL_ADD_QUAD(c, h, OID_AUTO, n, CTLFLAG_RD, p, 0, d)
693 
694 static void
695 ale_sysctl_node(struct ale_softc *sc)
696 {
697 	struct sysctl_ctx_list *ctx;
698 	struct sysctl_oid_list *child, *parent;
699 	struct sysctl_oid *tree;
700 	struct ale_hw_stats *stats;
701 	int error;
702 
703 	sysctl_ctx_init(&sc->ale_sysctl_ctx);
704 	sc->ale_sysctl_tree = SYSCTL_ADD_NODE(&sc->ale_sysctl_ctx,
705 				SYSCTL_STATIC_CHILDREN(_hw), OID_AUTO,
706 				device_get_nameunit(sc->ale_dev),
707 				CTLFLAG_RD, 0, "");
708 	if (sc->ale_sysctl_tree == NULL) {
709 		device_printf(sc->ale_dev, "can't add sysctl node\n");
710 		return;
711 	}
712 
713 	stats = &sc->ale_stats;
714 	ctx = &sc->ale_sysctl_ctx;
715 	child = SYSCTL_CHILDREN(sc->ale_sysctl_tree);
716 
717 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "int_rx_mod",
718 	    CTLTYPE_INT | CTLFLAG_RW, &sc->ale_int_rx_mod, 0,
719 	    sysctl_hw_ale_int_mod, "I", "ale Rx interrupt moderation");
720 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "int_tx_mod",
721 	    CTLTYPE_INT | CTLFLAG_RW, &sc->ale_int_tx_mod, 0,
722 	    sysctl_hw_ale_int_mod, "I", "ale Tx interrupt moderation");
723 
724 	/*
725 	 * Pull in device tunables.
726 	 */
727 	sc->ale_int_rx_mod = ALE_IM_RX_TIMER_DEFAULT;
728 	error = resource_int_value(device_get_name(sc->ale_dev),
729 	    device_get_unit(sc->ale_dev), "int_rx_mod", &sc->ale_int_rx_mod);
730 	if (error == 0) {
731 		if (sc->ale_int_rx_mod < ALE_IM_TIMER_MIN ||
732 		    sc->ale_int_rx_mod > ALE_IM_TIMER_MAX) {
733 			device_printf(sc->ale_dev, "int_rx_mod value out of "
734 			    "range; using default: %d\n",
735 			    ALE_IM_RX_TIMER_DEFAULT);
736 			sc->ale_int_rx_mod = ALE_IM_RX_TIMER_DEFAULT;
737 		}
738 	}
739 
740 	sc->ale_int_tx_mod = ALE_IM_TX_TIMER_DEFAULT;
741 	error = resource_int_value(device_get_name(sc->ale_dev),
742 	    device_get_unit(sc->ale_dev), "int_tx_mod", &sc->ale_int_tx_mod);
743 	if (error == 0) {
744 		if (sc->ale_int_tx_mod < ALE_IM_TIMER_MIN ||
745 		    sc->ale_int_tx_mod > ALE_IM_TIMER_MAX) {
746 			device_printf(sc->ale_dev, "int_tx_mod value out of "
747 			    "range; using default: %d\n",
748 			    ALE_IM_TX_TIMER_DEFAULT);
749 			sc->ale_int_tx_mod = ALE_IM_TX_TIMER_DEFAULT;
750 		}
751 	}
752 
753 	/* Misc statistics. */
754 	ALE_SYSCTL_STAT_ADD32(ctx, child, "reset_brk_seq",
755 	    &stats->reset_brk_seq,
756 	    "Controller resets due to broken Rx sequnce number");
757 
758 	tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "stats", CTLFLAG_RD,
759 	    NULL, "ATE statistics");
760 	parent = SYSCTL_CHILDREN(tree);
761 
762 	/* Rx statistics. */
763 	tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "rx", CTLFLAG_RD,
764 	    NULL, "Rx MAC statistics");
765 	child = SYSCTL_CHILDREN(tree);
766 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_frames",
767 	    &stats->rx_frames, "Good frames");
768 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_bcast_frames",
769 	    &stats->rx_bcast_frames, "Good broadcast frames");
770 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_mcast_frames",
771 	    &stats->rx_mcast_frames, "Good multicast frames");
772 	ALE_SYSCTL_STAT_ADD32(ctx, child, "pause_frames",
773 	    &stats->rx_pause_frames, "Pause control frames");
774 	ALE_SYSCTL_STAT_ADD32(ctx, child, "control_frames",
775 	    &stats->rx_control_frames, "Control frames");
776 	ALE_SYSCTL_STAT_ADD32(ctx, child, "crc_errs",
777 	    &stats->rx_crcerrs, "CRC errors");
778 	ALE_SYSCTL_STAT_ADD32(ctx, child, "len_errs",
779 	    &stats->rx_lenerrs, "Frames with length mismatched");
780 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_octets",
781 	    &stats->rx_bytes, "Good octets");
782 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_bcast_octets",
783 	    &stats->rx_bcast_bytes, "Good broadcast octets");
784 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_mcast_octets",
785 	    &stats->rx_mcast_bytes, "Good multicast octets");
786 	ALE_SYSCTL_STAT_ADD32(ctx, child, "runts",
787 	    &stats->rx_runts, "Too short frames");
788 	ALE_SYSCTL_STAT_ADD32(ctx, child, "fragments",
789 	    &stats->rx_fragments, "Fragmented frames");
790 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_64",
791 	    &stats->rx_pkts_64, "64 bytes frames");
792 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_65_127",
793 	    &stats->rx_pkts_65_127, "65 to 127 bytes frames");
794 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_128_255",
795 	    &stats->rx_pkts_128_255, "128 to 255 bytes frames");
796 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_256_511",
797 	    &stats->rx_pkts_256_511, "256 to 511 bytes frames");
798 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_512_1023",
799 	    &stats->rx_pkts_512_1023, "512 to 1023 bytes frames");
800 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1024_1518",
801 	    &stats->rx_pkts_1024_1518, "1024 to 1518 bytes frames");
802 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1519_max",
803 	    &stats->rx_pkts_1519_max, "1519 to max frames");
804 	ALE_SYSCTL_STAT_ADD32(ctx, child, "trunc_errs",
805 	    &stats->rx_pkts_truncated, "Truncated frames due to MTU size");
806 	ALE_SYSCTL_STAT_ADD32(ctx, child, "fifo_oflows",
807 	    &stats->rx_fifo_oflows, "FIFO overflows");
808 	ALE_SYSCTL_STAT_ADD32(ctx, child, "rrs_errs",
809 	    &stats->rx_rrs_errs, "Return status write-back errors");
810 	ALE_SYSCTL_STAT_ADD32(ctx, child, "align_errs",
811 	    &stats->rx_alignerrs, "Alignment errors");
812 	ALE_SYSCTL_STAT_ADD32(ctx, child, "filtered",
813 	    &stats->rx_pkts_filtered,
814 	    "Frames dropped due to address filtering");
815 
816 	/* Tx statistics. */
817 	tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "tx", CTLFLAG_RD,
818 	    NULL, "Tx MAC statistics");
819 	child = SYSCTL_CHILDREN(tree);
820 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_frames",
821 	    &stats->tx_frames, "Good frames");
822 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_bcast_frames",
823 	    &stats->tx_bcast_frames, "Good broadcast frames");
824 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_mcast_frames",
825 	    &stats->tx_mcast_frames, "Good multicast frames");
826 	ALE_SYSCTL_STAT_ADD32(ctx, child, "pause_frames",
827 	    &stats->tx_pause_frames, "Pause control frames");
828 	ALE_SYSCTL_STAT_ADD32(ctx, child, "control_frames",
829 	    &stats->tx_control_frames, "Control frames");
830 	ALE_SYSCTL_STAT_ADD32(ctx, child, "excess_defers",
831 	    &stats->tx_excess_defer, "Frames with excessive derferrals");
832 	ALE_SYSCTL_STAT_ADD32(ctx, child, "defers",
833 	    &stats->tx_excess_defer, "Frames with derferrals");
834 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_octets",
835 	    &stats->tx_bytes, "Good octets");
836 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_bcast_octets",
837 	    &stats->tx_bcast_bytes, "Good broadcast octets");
838 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_mcast_octets",
839 	    &stats->tx_mcast_bytes, "Good multicast octets");
840 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_64",
841 	    &stats->tx_pkts_64, "64 bytes frames");
842 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_65_127",
843 	    &stats->tx_pkts_65_127, "65 to 127 bytes frames");
844 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_128_255",
845 	    &stats->tx_pkts_128_255, "128 to 255 bytes frames");
846 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_256_511",
847 	    &stats->tx_pkts_256_511, "256 to 511 bytes frames");
848 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_512_1023",
849 	    &stats->tx_pkts_512_1023, "512 to 1023 bytes frames");
850 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1024_1518",
851 	    &stats->tx_pkts_1024_1518, "1024 to 1518 bytes frames");
852 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1519_max",
853 	    &stats->tx_pkts_1519_max, "1519 to max frames");
854 	ALE_SYSCTL_STAT_ADD32(ctx, child, "single_colls",
855 	    &stats->tx_single_colls, "Single collisions");
856 	ALE_SYSCTL_STAT_ADD32(ctx, child, "multi_colls",
857 	    &stats->tx_multi_colls, "Multiple collisions");
858 	ALE_SYSCTL_STAT_ADD32(ctx, child, "late_colls",
859 	    &stats->tx_late_colls, "Late collisions");
860 	ALE_SYSCTL_STAT_ADD32(ctx, child, "excess_colls",
861 	    &stats->tx_excess_colls, "Excessive collisions");
862 	ALE_SYSCTL_STAT_ADD32(ctx, child, "abort",
863 	    &stats->tx_abort, "Aborted frames due to Excessive collisions");
864 	ALE_SYSCTL_STAT_ADD32(ctx, child, "underruns",
865 	    &stats->tx_underrun, "FIFO underruns");
866 	ALE_SYSCTL_STAT_ADD32(ctx, child, "desc_underruns",
867 	    &stats->tx_desc_underrun, "Descriptor write-back errors");
868 	ALE_SYSCTL_STAT_ADD32(ctx, child, "len_errs",
869 	    &stats->tx_lenerrs, "Frames with length mismatched");
870 	ALE_SYSCTL_STAT_ADD32(ctx, child, "trunc_errs",
871 	    &stats->tx_pkts_truncated, "Truncated frames due to MTU size");
872 }
873 
874 #undef ALE_SYSCTL_STAT_ADD32
875 #undef ALE_SYSCTL_STAT_ADD64
876 
877 struct ale_dmamap_arg {
878 	bus_addr_t	ale_busaddr;
879 };
880 
881 static void
882 ale_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error)
883 {
884 	struct ale_dmamap_arg *ctx;
885 
886 	if (error != 0)
887 		return;
888 
889 	KASSERT(nsegs == 1, ("%s: %d segments returned!", __func__, nsegs));
890 
891 	ctx = (struct ale_dmamap_arg *)arg;
892 	ctx->ale_busaddr = segs[0].ds_addr;
893 }
894 
895 /*
896  * Tx descriptors/RXF0/CMB DMA blocks share ALE_DESC_ADDR_HI register
897  * which specifies high address region of DMA blocks. Therefore these
898  * blocks should have the same high address of given 4GB address
899  * space(i.e. crossing 4GB boundary is not allowed).
900  */
901 static int
902 ale_check_boundary(struct ale_softc *sc)
903 {
904 	bus_addr_t rx_cmb_end[ALE_RX_PAGES], tx_cmb_end;
905 	bus_addr_t rx_page_end[ALE_RX_PAGES], tx_ring_end;
906 
907 	rx_page_end[0] = sc->ale_cdata.ale_rx_page[0].page_paddr +
908 	    sc->ale_pagesize;
909 	rx_page_end[1] = sc->ale_cdata.ale_rx_page[1].page_paddr +
910 	    sc->ale_pagesize;
911 	tx_ring_end = sc->ale_cdata.ale_tx_ring_paddr + ALE_TX_RING_SZ;
912 	tx_cmb_end = sc->ale_cdata.ale_tx_cmb_paddr + ALE_TX_CMB_SZ;
913 	rx_cmb_end[0] = sc->ale_cdata.ale_rx_page[0].cmb_paddr + ALE_RX_CMB_SZ;
914 	rx_cmb_end[1] = sc->ale_cdata.ale_rx_page[1].cmb_paddr + ALE_RX_CMB_SZ;
915 
916 	if ((ALE_ADDR_HI(tx_ring_end) !=
917 	    ALE_ADDR_HI(sc->ale_cdata.ale_tx_ring_paddr)) ||
918 	    (ALE_ADDR_HI(rx_page_end[0]) !=
919 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[0].page_paddr)) ||
920 	    (ALE_ADDR_HI(rx_page_end[1]) !=
921 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[1].page_paddr)) ||
922 	    (ALE_ADDR_HI(tx_cmb_end) !=
923 	    ALE_ADDR_HI(sc->ale_cdata.ale_tx_cmb_paddr)) ||
924 	    (ALE_ADDR_HI(rx_cmb_end[0]) !=
925 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[0].cmb_paddr)) ||
926 	    (ALE_ADDR_HI(rx_cmb_end[1]) !=
927 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[1].cmb_paddr)))
928 		return (EFBIG);
929 
930 	if ((ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_page_end[0])) ||
931 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_page_end[1])) ||
932 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_cmb_end[0])) ||
933 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_cmb_end[1])) ||
934 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(tx_cmb_end)))
935 		return (EFBIG);
936 
937 	return (0);
938 }
939 
940 static int
941 ale_dma_alloc(struct ale_softc *sc)
942 {
943 	struct ale_txdesc *txd;
944 	bus_addr_t lowaddr;
945 	struct ale_dmamap_arg ctx;
946 	int error, guard_size, i;
947 
948 	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0)
949 		guard_size = ALE_JUMBO_FRAMELEN;
950 	else
951 		guard_size = ALE_MAX_FRAMELEN;
952 	sc->ale_pagesize = roundup(guard_size + ALE_RX_PAGE_SZ,
953 	    ALE_RX_PAGE_ALIGN);
954 	lowaddr = BUS_SPACE_MAXADDR;
955 again:
956 	/* Create parent DMA tag. */
957 	error = bus_dma_tag_create(
958 	    NULL,			/* parent */
959 	    1, 0,			/* alignment, boundary */
960 	    lowaddr,			/* lowaddr */
961 	    BUS_SPACE_MAXADDR,		/* highaddr */
962 	    NULL, NULL,			/* filter, filterarg */
963 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsize */
964 	    0,				/* nsegments */
965 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsegsize */
966 	    0,				/* flags */
967 	    &sc->ale_cdata.ale_parent_tag);
968 	if (error != 0) {
969 		device_printf(sc->ale_dev,
970 		    "could not create parent DMA tag.\n");
971 		goto fail;
972 	}
973 
974 	/* Create DMA tag for Tx descriptor ring. */
975 	error = bus_dma_tag_create(
976 	    sc->ale_cdata.ale_parent_tag, /* parent */
977 	    ALE_TX_RING_ALIGN, 0,	/* alignment, boundary */
978 	    BUS_SPACE_MAXADDR,		/* lowaddr */
979 	    BUS_SPACE_MAXADDR,		/* highaddr */
980 	    NULL, NULL,			/* filter, filterarg */
981 	    ALE_TX_RING_SZ,		/* maxsize */
982 	    1,				/* nsegments */
983 	    ALE_TX_RING_SZ,		/* maxsegsize */
984 	    0,				/* flags */
985 	    &sc->ale_cdata.ale_tx_ring_tag);
986 	if (error != 0) {
987 		device_printf(sc->ale_dev,
988 		    "could not create Tx ring DMA tag.\n");
989 		goto fail;
990 	}
991 
992 	/* Create DMA tag for Rx pages. */
993 	for (i = 0; i < ALE_RX_PAGES; i++) {
994 		error = bus_dma_tag_create(
995 		    sc->ale_cdata.ale_parent_tag, /* parent */
996 		    ALE_RX_PAGE_ALIGN, 0,	/* alignment, boundary */
997 		    BUS_SPACE_MAXADDR,		/* lowaddr */
998 		    BUS_SPACE_MAXADDR,		/* highaddr */
999 		    NULL, NULL,			/* filter, filterarg */
1000 		    sc->ale_pagesize,		/* maxsize */
1001 		    1,				/* nsegments */
1002 		    sc->ale_pagesize,		/* maxsegsize */
1003 		    0,				/* flags */
1004 		    &sc->ale_cdata.ale_rx_page[i].page_tag);
1005 		if (error != 0) {
1006 			device_printf(sc->ale_dev,
1007 			    "could not create Rx page %d DMA tag.\n", i);
1008 			goto fail;
1009 		}
1010 	}
1011 
1012 	/* Create DMA tag for Tx coalescing message block. */
1013 	error = bus_dma_tag_create(
1014 	    sc->ale_cdata.ale_parent_tag, /* parent */
1015 	    ALE_CMB_ALIGN, 0,		/* alignment, boundary */
1016 	    BUS_SPACE_MAXADDR,		/* lowaddr */
1017 	    BUS_SPACE_MAXADDR,		/* highaddr */
1018 	    NULL, NULL,			/* filter, filterarg */
1019 	    ALE_TX_CMB_SZ,		/* maxsize */
1020 	    1,				/* nsegments */
1021 	    ALE_TX_CMB_SZ,		/* maxsegsize */
1022 	    0,				/* flags */
1023 	    &sc->ale_cdata.ale_tx_cmb_tag);
1024 	if (error != 0) {
1025 		device_printf(sc->ale_dev,
1026 		    "could not create Tx CMB DMA tag.\n");
1027 		goto fail;
1028 	}
1029 
1030 	/* Create DMA tag for Rx coalescing message block. */
1031 	for (i = 0; i < ALE_RX_PAGES; i++) {
1032 		error = bus_dma_tag_create(
1033 		    sc->ale_cdata.ale_parent_tag, /* parent */
1034 		    ALE_CMB_ALIGN, 0,		/* alignment, boundary */
1035 		    BUS_SPACE_MAXADDR,		/* lowaddr */
1036 		    BUS_SPACE_MAXADDR,		/* highaddr */
1037 		    NULL, NULL,			/* filter, filterarg */
1038 		    ALE_RX_CMB_SZ,		/* maxsize */
1039 		    1,				/* nsegments */
1040 		    ALE_RX_CMB_SZ,		/* maxsegsize */
1041 		    0,				/* flags */
1042 		    &sc->ale_cdata.ale_rx_page[i].cmb_tag);
1043 		if (error != 0) {
1044 			device_printf(sc->ale_dev,
1045 			    "could not create Rx page %d CMB DMA tag.\n", i);
1046 			goto fail;
1047 		}
1048 	}
1049 
1050 	/* Allocate DMA'able memory and load the DMA map for Tx ring. */
1051 	error = bus_dmamem_alloc(sc->ale_cdata.ale_tx_ring_tag,
1052 	    (void **)&sc->ale_cdata.ale_tx_ring,
1053 	    BUS_DMA_WAITOK | BUS_DMA_ZERO,
1054 	    &sc->ale_cdata.ale_tx_ring_map);
1055 	if (error != 0) {
1056 		device_printf(sc->ale_dev,
1057 		    "could not allocate DMA'able memory for Tx ring.\n");
1058 		goto fail;
1059 	}
1060 	ctx.ale_busaddr = 0;
1061 	error = bus_dmamap_load(sc->ale_cdata.ale_tx_ring_tag,
1062 	    sc->ale_cdata.ale_tx_ring_map, sc->ale_cdata.ale_tx_ring,
1063 	    ALE_TX_RING_SZ, ale_dmamap_cb, &ctx, 0);
1064 	if (error != 0 || ctx.ale_busaddr == 0) {
1065 		device_printf(sc->ale_dev,
1066 		    "could not load DMA'able memory for Tx ring.\n");
1067 		goto fail;
1068 	}
1069 	sc->ale_cdata.ale_tx_ring_paddr = ctx.ale_busaddr;
1070 
1071 	/* Rx pages. */
1072 	for (i = 0; i < ALE_RX_PAGES; i++) {
1073 		error = bus_dmamem_alloc(sc->ale_cdata.ale_rx_page[i].page_tag,
1074 		    (void **)&sc->ale_cdata.ale_rx_page[i].page_addr,
1075 		    BUS_DMA_WAITOK | BUS_DMA_ZERO,
1076 		    &sc->ale_cdata.ale_rx_page[i].page_map);
1077 		if (error != 0) {
1078 			device_printf(sc->ale_dev,
1079 			    "could not allocate DMA'able memory for "
1080 			    "Rx page %d.\n", i);
1081 			goto fail;
1082 		}
1083 		ctx.ale_busaddr = 0;
1084 		error = bus_dmamap_load(sc->ale_cdata.ale_rx_page[i].page_tag,
1085 		    sc->ale_cdata.ale_rx_page[i].page_map,
1086 		    sc->ale_cdata.ale_rx_page[i].page_addr,
1087 		    sc->ale_pagesize, ale_dmamap_cb, &ctx, 0);
1088 		if (error != 0 || ctx.ale_busaddr == 0) {
1089 			device_printf(sc->ale_dev,
1090 			    "could not load DMA'able memory for "
1091 			    "Rx page %d.\n", i);
1092 			goto fail;
1093 		}
1094 		sc->ale_cdata.ale_rx_page[i].page_paddr = ctx.ale_busaddr;
1095 	}
1096 
1097 	/* Tx CMB. */
1098 	error = bus_dmamem_alloc(sc->ale_cdata.ale_tx_cmb_tag,
1099 	    (void **)&sc->ale_cdata.ale_tx_cmb,
1100 	    BUS_DMA_WAITOK | BUS_DMA_ZERO,
1101 	    &sc->ale_cdata.ale_tx_cmb_map);
1102 	if (error != 0) {
1103 		device_printf(sc->ale_dev,
1104 		    "could not allocate DMA'able memory for Tx CMB.\n");
1105 		goto fail;
1106 	}
1107 	ctx.ale_busaddr = 0;
1108 	error = bus_dmamap_load(sc->ale_cdata.ale_tx_cmb_tag,
1109 	    sc->ale_cdata.ale_tx_cmb_map, sc->ale_cdata.ale_tx_cmb,
1110 	    ALE_TX_CMB_SZ, ale_dmamap_cb, &ctx, 0);
1111 	if (error != 0 || ctx.ale_busaddr == 0) {
1112 		device_printf(sc->ale_dev,
1113 		    "could not load DMA'able memory for Tx CMB.\n");
1114 		goto fail;
1115 	}
1116 	sc->ale_cdata.ale_tx_cmb_paddr = ctx.ale_busaddr;
1117 
1118 	/* Rx CMB. */
1119 	for (i = 0; i < ALE_RX_PAGES; i++) {
1120 		error = bus_dmamem_alloc(sc->ale_cdata.ale_rx_page[i].cmb_tag,
1121 		    (void **)&sc->ale_cdata.ale_rx_page[i].cmb_addr,
1122 		    BUS_DMA_WAITOK | BUS_DMA_ZERO,
1123 		    &sc->ale_cdata.ale_rx_page[i].cmb_map);
1124 		if (error != 0) {
1125 			device_printf(sc->ale_dev, "could not allocate "
1126 			    "DMA'able memory for Rx page %d CMB.\n", i);
1127 			goto fail;
1128 		}
1129 		ctx.ale_busaddr = 0;
1130 		error = bus_dmamap_load(sc->ale_cdata.ale_rx_page[i].cmb_tag,
1131 		    sc->ale_cdata.ale_rx_page[i].cmb_map,
1132 		    sc->ale_cdata.ale_rx_page[i].cmb_addr,
1133 		    ALE_RX_CMB_SZ, ale_dmamap_cb, &ctx, 0);
1134 		if (error != 0 || ctx.ale_busaddr == 0) {
1135 			device_printf(sc->ale_dev, "could not load DMA'able "
1136 			    "memory for Rx page %d CMB.\n", i);
1137 			goto fail;
1138 		}
1139 		sc->ale_cdata.ale_rx_page[i].cmb_paddr = ctx.ale_busaddr;
1140 	}
1141 
1142 	/*
1143 	 * Tx descriptors/RXF0/CMB DMA blocks share the same
1144 	 * high address region of 64bit DMA address space.
1145 	 */
1146 	if (lowaddr != BUS_SPACE_MAXADDR_32BIT &&
1147 	    (error = ale_check_boundary(sc)) != 0) {
1148 		device_printf(sc->ale_dev, "4GB boundary crossed, "
1149 		    "switching to 32bit DMA addressing mode.\n");
1150 		ale_dma_free(sc);
1151 		/*
1152 		 * Limit max allowable DMA address space to 32bit
1153 		 * and try again.
1154 		 */
1155 		lowaddr = BUS_SPACE_MAXADDR_32BIT;
1156 		goto again;
1157 	}
1158 
1159 	/*
1160 	 * Create Tx buffer parent tag.
1161 	 * AR81xx allows 64bit DMA addressing of Tx buffers so it
1162 	 * needs separate parent DMA tag as parent DMA address space
1163 	 * could be restricted to be within 32bit address space by
1164 	 * 4GB boundary crossing.
1165 	 */
1166 	error = bus_dma_tag_create(
1167 	    NULL,			/* parent */
1168 	    1, 0,			/* alignment, boundary */
1169 	    BUS_SPACE_MAXADDR,		/* lowaddr */
1170 	    BUS_SPACE_MAXADDR,		/* highaddr */
1171 	    NULL, NULL,			/* filter, filterarg */
1172 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsize */
1173 	    0,				/* nsegments */
1174 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsegsize */
1175 	    0,				/* flags */
1176 	    &sc->ale_cdata.ale_buffer_tag);
1177 	if (error != 0) {
1178 		device_printf(sc->ale_dev,
1179 		    "could not create parent buffer DMA tag.\n");
1180 		goto fail;
1181 	}
1182 
1183 	/* Create DMA tag for Tx buffers. */
1184 	error = bus_dma_tag_create(
1185 	    sc->ale_cdata.ale_buffer_tag, /* parent */
1186 	    1, 0,			/* alignment, boundary */
1187 	    BUS_SPACE_MAXADDR,		/* lowaddr */
1188 	    BUS_SPACE_MAXADDR,		/* highaddr */
1189 	    NULL, NULL,			/* filter, filterarg */
1190 	    ALE_TSO_MAXSIZE,		/* maxsize */
1191 	    ALE_MAXTXSEGS,		/* nsegments */
1192 	    ALE_TSO_MAXSEGSIZE,		/* maxsegsize */
1193 	    0,				/* flags */
1194 	    &sc->ale_cdata.ale_tx_tag);
1195 	if (error != 0) {
1196 		device_printf(sc->ale_dev, "could not create Tx DMA tag.\n");
1197 		goto fail;
1198 	}
1199 
1200 	/* Create DMA maps for Tx buffers. */
1201 	for (i = 0; i < ALE_TX_RING_CNT; i++) {
1202 		txd = &sc->ale_cdata.ale_txdesc[i];
1203 		txd->tx_m = NULL;
1204 		txd->tx_dmamap = NULL;
1205 		error = bus_dmamap_create(sc->ale_cdata.ale_tx_tag, 0,
1206 		    &txd->tx_dmamap);
1207 		if (error != 0) {
1208 			device_printf(sc->ale_dev,
1209 			    "could not create Tx dmamap.\n");
1210 			goto fail;
1211 		}
1212 	}
1213 fail:
1214 	return (error);
1215 }
1216 
1217 static void
1218 ale_dma_free(struct ale_softc *sc)
1219 {
1220 	struct ale_txdesc *txd;
1221 	int i;
1222 
1223 	/* Tx buffers. */
1224 	if (sc->ale_cdata.ale_tx_tag != NULL) {
1225 		for (i = 0; i < ALE_TX_RING_CNT; i++) {
1226 			txd = &sc->ale_cdata.ale_txdesc[i];
1227 			if (txd->tx_dmamap != NULL) {
1228 				bus_dmamap_destroy(sc->ale_cdata.ale_tx_tag,
1229 				    txd->tx_dmamap);
1230 				txd->tx_dmamap = NULL;
1231 			}
1232 		}
1233 		bus_dma_tag_destroy(sc->ale_cdata.ale_tx_tag);
1234 		sc->ale_cdata.ale_tx_tag = NULL;
1235 	}
1236 	/* Tx descriptor ring. */
1237 	if (sc->ale_cdata.ale_tx_ring_tag != NULL) {
1238 		if (sc->ale_cdata.ale_tx_ring_map != NULL)
1239 			bus_dmamap_unload(sc->ale_cdata.ale_tx_ring_tag,
1240 			    sc->ale_cdata.ale_tx_ring_map);
1241 		if (sc->ale_cdata.ale_tx_ring_map != NULL &&
1242 		    sc->ale_cdata.ale_tx_ring != NULL)
1243 			bus_dmamem_free(sc->ale_cdata.ale_tx_ring_tag,
1244 			    sc->ale_cdata.ale_tx_ring,
1245 			    sc->ale_cdata.ale_tx_ring_map);
1246 		sc->ale_cdata.ale_tx_ring = NULL;
1247 		sc->ale_cdata.ale_tx_ring_map = NULL;
1248 		bus_dma_tag_destroy(sc->ale_cdata.ale_tx_ring_tag);
1249 		sc->ale_cdata.ale_tx_ring_tag = NULL;
1250 	}
1251 	/* Rx page block. */
1252 	for (i = 0; i < ALE_RX_PAGES; i++) {
1253 		if (sc->ale_cdata.ale_rx_page[i].page_tag != NULL) {
1254 			if (sc->ale_cdata.ale_rx_page[i].page_map != NULL)
1255 				bus_dmamap_unload(
1256 				    sc->ale_cdata.ale_rx_page[i].page_tag,
1257 				    sc->ale_cdata.ale_rx_page[i].page_map);
1258 			if (sc->ale_cdata.ale_rx_page[i].page_map != NULL &&
1259 			    sc->ale_cdata.ale_rx_page[i].page_addr != NULL)
1260 				bus_dmamem_free(
1261 				    sc->ale_cdata.ale_rx_page[i].page_tag,
1262 				    sc->ale_cdata.ale_rx_page[i].page_addr,
1263 				    sc->ale_cdata.ale_rx_page[i].page_map);
1264 			sc->ale_cdata.ale_rx_page[i].page_addr = NULL;
1265 			sc->ale_cdata.ale_rx_page[i].page_map = NULL;
1266 			bus_dma_tag_destroy(
1267 			    sc->ale_cdata.ale_rx_page[i].page_tag);
1268 			sc->ale_cdata.ale_rx_page[i].page_tag = NULL;
1269 		}
1270 	}
1271 	/* Rx CMB. */
1272 	for (i = 0; i < ALE_RX_PAGES; i++) {
1273 		if (sc->ale_cdata.ale_rx_page[i].cmb_tag != NULL) {
1274 			if (sc->ale_cdata.ale_rx_page[i].cmb_map != NULL)
1275 				bus_dmamap_unload(
1276 				    sc->ale_cdata.ale_rx_page[i].cmb_tag,
1277 				    sc->ale_cdata.ale_rx_page[i].cmb_map);
1278 			if (sc->ale_cdata.ale_rx_page[i].cmb_map != NULL &&
1279 			    sc->ale_cdata.ale_rx_page[i].cmb_addr != NULL)
1280 				bus_dmamem_free(
1281 				    sc->ale_cdata.ale_rx_page[i].cmb_tag,
1282 				    sc->ale_cdata.ale_rx_page[i].cmb_addr,
1283 				    sc->ale_cdata.ale_rx_page[i].cmb_map);
1284 			sc->ale_cdata.ale_rx_page[i].cmb_addr = NULL;
1285 			sc->ale_cdata.ale_rx_page[i].cmb_map = NULL;
1286 			bus_dma_tag_destroy(
1287 			    sc->ale_cdata.ale_rx_page[i].cmb_tag);
1288 			sc->ale_cdata.ale_rx_page[i].cmb_tag = NULL;
1289 		}
1290 	}
1291 	/* Tx CMB. */
1292 	if (sc->ale_cdata.ale_tx_cmb_tag != NULL) {
1293 		if (sc->ale_cdata.ale_tx_cmb_map != NULL)
1294 			bus_dmamap_unload(sc->ale_cdata.ale_tx_cmb_tag,
1295 			    sc->ale_cdata.ale_tx_cmb_map);
1296 		if (sc->ale_cdata.ale_tx_cmb_map != NULL &&
1297 		    sc->ale_cdata.ale_tx_cmb != NULL)
1298 			bus_dmamem_free(sc->ale_cdata.ale_tx_cmb_tag,
1299 			    sc->ale_cdata.ale_tx_cmb,
1300 			    sc->ale_cdata.ale_tx_cmb_map);
1301 		sc->ale_cdata.ale_tx_cmb = NULL;
1302 		sc->ale_cdata.ale_tx_cmb_map = NULL;
1303 		bus_dma_tag_destroy(sc->ale_cdata.ale_tx_cmb_tag);
1304 		sc->ale_cdata.ale_tx_cmb_tag = NULL;
1305 	}
1306 	if (sc->ale_cdata.ale_buffer_tag != NULL) {
1307 		bus_dma_tag_destroy(sc->ale_cdata.ale_buffer_tag);
1308 		sc->ale_cdata.ale_buffer_tag = NULL;
1309 	}
1310 	if (sc->ale_cdata.ale_parent_tag != NULL) {
1311 		bus_dma_tag_destroy(sc->ale_cdata.ale_parent_tag);
1312 		sc->ale_cdata.ale_parent_tag = NULL;
1313 	}
1314 }
1315 
1316 static int
1317 ale_shutdown(device_t dev)
1318 {
1319 	return (ale_suspend(dev));
1320 }
1321 
1322 #ifdef notyet
1323 
1324 /*
1325  * Note, this driver resets the link speed to 10/100Mbps by
1326  * restarting auto-negotiation in suspend/shutdown phase but we
1327  * don't know whether that auto-negotiation would succeed or not
1328  * as driver has no control after powering off/suspend operation.
1329  * If the renegotiation fail WOL may not work. Running at 1Gbps
1330  * will draw more power than 375mA at 3.3V which is specified in
1331  * PCI specification and that would result in complete
1332  * shutdowning power to ethernet controller.
1333  *
1334  * TODO
1335  * Save current negotiated media speed/duplex/flow-control to
1336  * softc and restore the same link again after resuming. PHY
1337  * handling such as power down/resetting to 100Mbps may be better
1338  * handled in suspend method in phy driver.
1339  */
1340 static void
1341 ale_setlinkspeed(struct ale_softc *sc)
1342 {
1343 	struct mii_data *mii;
1344 	int aneg, i;
1345 
1346 	mii = device_get_softc(sc->ale_miibus);
1347 	mii_pollstat(mii);
1348 	aneg = 0;
1349 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
1350 	    (IFM_ACTIVE | IFM_AVALID)) {
1351 		switch IFM_SUBTYPE(mii->mii_media_active) {
1352 		case IFM_10_T:
1353 		case IFM_100_TX:
1354 			return;
1355 		case IFM_1000_T:
1356 			aneg++;
1357 			break;
1358 		default:
1359 			break;
1360 		}
1361 	}
1362 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr, MII_100T2CR, 0);
1363 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
1364 	    MII_ANAR, ANAR_TX_FD | ANAR_TX | ANAR_10_FD | ANAR_10 | ANAR_CSMA);
1365 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
1366 	    MII_BMCR, BMCR_RESET | BMCR_AUTOEN | BMCR_STARTNEG);
1367 	DELAY(1000);
1368 	if (aneg != 0) {
1369 		/*
1370 		 * Poll link state until ale(4) get a 10/100Mbps link.
1371 		 */
1372 		for (i = 0; i < MII_ANEGTICKS_GIGE; i++) {
1373 			mii_pollstat(mii);
1374 			if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID))
1375 			    == (IFM_ACTIVE | IFM_AVALID)) {
1376 				switch (IFM_SUBTYPE(
1377 				    mii->mii_media_active)) {
1378 				case IFM_10_T:
1379 				case IFM_100_TX:
1380 					ale_mac_config(sc);
1381 					return;
1382 				default:
1383 					break;
1384 				}
1385 			}
1386 			ALE_UNLOCK(sc);
1387 			pause("alelnk", hz);
1388 			ALE_LOCK(sc);
1389 		}
1390 		if (i == MII_ANEGTICKS_GIGE)
1391 			device_printf(sc->ale_dev,
1392 			    "establishing a link failed, WOL may not work!");
1393 	}
1394 	/*
1395 	 * No link, force MAC to have 100Mbps, full-duplex link.
1396 	 * This is the last resort and may/may not work.
1397 	 */
1398 	mii->mii_media_status = IFM_AVALID | IFM_ACTIVE;
1399 	mii->mii_media_active = IFM_ETHER | IFM_100_TX | IFM_FDX;
1400 	ale_mac_config(sc);
1401 }
1402 
1403 static void
1404 ale_setwol(struct ale_softc *sc)
1405 {
1406 	struct ifnet *ifp;
1407 	uint32_t reg, pmcs;
1408 	uint16_t pmstat;
1409 	int pmc;
1410 
1411 	ALE_LOCK_ASSERT(sc);
1412 
1413 	if (pci_find_extcap(sc->ale_dev, PCIY_PMG, &pmc) != 0) {
1414 		/* Disable WOL. */
1415 		CSR_WRITE_4(sc, ALE_WOL_CFG, 0);
1416 		reg = CSR_READ_4(sc, ALE_PCIE_PHYMISC);
1417 		reg |= PCIE_PHYMISC_FORCE_RCV_DET;
1418 		CSR_WRITE_4(sc, ALE_PCIE_PHYMISC, reg);
1419 		/* Force PHY power down. */
1420 		CSR_WRITE_2(sc, ALE_GPHY_CTRL,
1421 		    GPHY_CTRL_EXT_RESET | GPHY_CTRL_HIB_EN |
1422 		    GPHY_CTRL_HIB_PULSE | GPHY_CTRL_PHY_PLL_ON |
1423 		    GPHY_CTRL_SEL_ANA_RESET | GPHY_CTRL_PHY_IDDQ |
1424 		    GPHY_CTRL_PCLK_SEL_DIS | GPHY_CTRL_PWDOWN_HW);
1425 		return;
1426 	}
1427 
1428 	ifp = sc->ale_ifp;
1429 	if ((ifp->if_capenable & IFCAP_WOL) != 0) {
1430 		if ((sc->ale_flags & ALE_FLAG_FASTETHER) == 0)
1431 			ale_setlinkspeed(sc);
1432 	}
1433 
1434 	pmcs = 0;
1435 	if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
1436 		pmcs |= WOL_CFG_MAGIC | WOL_CFG_MAGIC_ENB;
1437 	CSR_WRITE_4(sc, ALE_WOL_CFG, pmcs);
1438 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
1439 	reg &= ~(MAC_CFG_DBG | MAC_CFG_PROMISC | MAC_CFG_ALLMULTI |
1440 	    MAC_CFG_BCAST);
1441 	if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
1442 		reg |= MAC_CFG_ALLMULTI | MAC_CFG_BCAST;
1443 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
1444 		reg |= MAC_CFG_RX_ENB;
1445 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
1446 
1447 	if ((ifp->if_capenable & IFCAP_WOL) == 0) {
1448 		/* WOL disabled, PHY power down. */
1449 		reg = CSR_READ_4(sc, ALE_PCIE_PHYMISC);
1450 		reg |= PCIE_PHYMISC_FORCE_RCV_DET;
1451 		CSR_WRITE_4(sc, ALE_PCIE_PHYMISC, reg);
1452 		CSR_WRITE_2(sc, ALE_GPHY_CTRL,
1453 		    GPHY_CTRL_EXT_RESET | GPHY_CTRL_HIB_EN |
1454 		    GPHY_CTRL_HIB_PULSE | GPHY_CTRL_SEL_ANA_RESET |
1455 		    GPHY_CTRL_PHY_IDDQ | GPHY_CTRL_PCLK_SEL_DIS |
1456 		    GPHY_CTRL_PWDOWN_HW);
1457 	}
1458 	/* Request PME. */
1459 	pmstat = pci_read_config(sc->ale_dev, pmc + PCIR_POWER_STATUS, 2);
1460 	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
1461 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
1462 		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
1463 	pci_write_config(sc->ale_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
1464 }
1465 
1466 #endif	/* notyet */
1467 
1468 static int
1469 ale_suspend(device_t dev)
1470 {
1471 	struct ale_softc *sc = device_get_softc(dev);
1472 	struct ifnet *ifp = &sc->arpcom.ac_if;
1473 
1474 	lwkt_serialize_enter(ifp->if_serializer);
1475 	ale_stop(sc);
1476 #ifdef notyet
1477 	ale_setwol(sc);
1478 #endif
1479 	lwkt_serialize_exit(ifp->if_serializer);
1480 	return (0);
1481 }
1482 
1483 static int
1484 ale_resume(device_t dev)
1485 {
1486 	struct ale_softc *sc = device_get_softc(dev);
1487 	struct ifnet *ifp = &sc->arpcom.ac_if;
1488 	uint16_t cmd;
1489 
1490 	lwkt_serialize_enter(ifp->if_serializer);
1491 
1492 	/*
1493 	 * Clear INTx emulation disable for hardwares that
1494 	 * is set in resume event. From Linux.
1495 	 */
1496 	cmd = pci_read_config(sc->ale_dev, PCIR_COMMAND, 2);
1497 	if ((cmd & 0x0400) != 0) {
1498 		cmd &= ~0x0400;
1499 		pci_write_config(sc->ale_dev, PCIR_COMMAND, cmd, 2);
1500 	}
1501 
1502 #ifdef notyet
1503 	if (pci_find_extcap(sc->ale_dev, PCIY_PMG, &pmc) == 0) {
1504 		uint16_t pmstat;
1505 		int pmc;
1506 
1507 		/* Disable PME and clear PME status. */
1508 		pmstat = pci_read_config(sc->ale_dev,
1509 		    pmc + PCIR_POWER_STATUS, 2);
1510 		if ((pmstat & PCIM_PSTAT_PMEENABLE) != 0) {
1511 			pmstat &= ~PCIM_PSTAT_PMEENABLE;
1512 			pci_write_config(sc->ale_dev,
1513 			    pmc + PCIR_POWER_STATUS, pmstat, 2);
1514 		}
1515 	}
1516 #endif
1517 
1518 	/* Reset PHY. */
1519 	ale_phy_reset(sc);
1520 	if ((ifp->if_flags & IFF_UP) != 0)
1521 		ale_init(sc);
1522 
1523 	lwkt_serialize_exit(ifp->if_serializer);
1524 	return (0);
1525 }
1526 
1527 static int
1528 ale_encap(struct ale_softc *sc, struct mbuf **m_head)
1529 {
1530 	struct ale_txdesc *txd, *txd_last;
1531 	struct tx_desc *desc;
1532 	struct mbuf *m;
1533 	bus_dma_segment_t txsegs[ALE_MAXTXSEGS];
1534 	struct ale_dmamap_ctx ctx;
1535 	bus_dmamap_t map;
1536 	uint32_t cflags, poff, vtag;
1537 	int error, i, nsegs, prod;
1538 
1539 	M_ASSERTPKTHDR((*m_head));
1540 
1541 	m = *m_head;
1542 	cflags = vtag = 0;
1543 	poff = 0;
1544 
1545 	prod = sc->ale_cdata.ale_tx_prod;
1546 	txd = &sc->ale_cdata.ale_txdesc[prod];
1547 	txd_last = txd;
1548 	map = txd->tx_dmamap;
1549 
1550 	ctx.nsegs = ALE_MAXTXSEGS;
1551 	ctx.segs = txsegs;
1552 	error =  bus_dmamap_load_mbuf(sc->ale_cdata.ale_tx_tag, map,
1553 				      *m_head, ale_dmamap_buf_cb, &ctx,
1554 				      BUS_DMA_NOWAIT);
1555 	if (error == EFBIG) {
1556 		m = m_defrag(*m_head, MB_DONTWAIT);
1557 		if (m == NULL) {
1558 			m_freem(*m_head);
1559 			*m_head = NULL;
1560 			return (ENOMEM);
1561 		}
1562 		*m_head = m;
1563 
1564 		ctx.nsegs = ALE_MAXTXSEGS;
1565 		ctx.segs = txsegs;
1566 		error =  bus_dmamap_load_mbuf(sc->ale_cdata.ale_tx_tag, map,
1567 					      *m_head, ale_dmamap_buf_cb, &ctx,
1568 					      BUS_DMA_NOWAIT);
1569 		if (error != 0) {
1570 			m_freem(*m_head);
1571 			*m_head = NULL;
1572 			return (error);
1573 		}
1574 	} else if (error != 0) {
1575 		return (error);
1576 	}
1577 	nsegs = ctx.nsegs;
1578 
1579 	if (nsegs == 0) {
1580 		m_freem(*m_head);
1581 		*m_head = NULL;
1582 		return (EIO);
1583 	}
1584 
1585 	/* Check descriptor overrun. */
1586 	if (sc->ale_cdata.ale_tx_cnt + nsegs >= ALE_TX_RING_CNT - 2) {
1587 		bus_dmamap_unload(sc->ale_cdata.ale_tx_tag, map);
1588 		return (ENOBUFS);
1589 	}
1590 	bus_dmamap_sync(sc->ale_cdata.ale_tx_tag, map, BUS_DMASYNC_PREWRITE);
1591 
1592 	m = *m_head;
1593 	/* Configure Tx checksum offload. */
1594 	if ((m->m_pkthdr.csum_flags & ALE_CSUM_FEATURES) != 0) {
1595 		/*
1596 		 * AR81xx supports Tx custom checksum offload feature
1597 		 * that offloads single 16bit checksum computation.
1598 		 * So you can choose one among IP, TCP and UDP.
1599 		 * Normally driver sets checksum start/insertion
1600 		 * position from the information of TCP/UDP frame as
1601 		 * TCP/UDP checksum takes more time than that of IP.
1602 		 * However it seems that custom checksum offload
1603 		 * requires 4 bytes aligned Tx buffers due to hardware
1604 		 * bug.
1605 		 * AR81xx also supports explicit Tx checksum computation
1606 		 * if it is told that the size of IP header and TCP
1607 		 * header(for UDP, the header size does not matter
1608 		 * because it's fixed length). However with this scheme
1609 		 * TSO does not work so you have to choose one either
1610 		 * TSO or explicit Tx checksum offload. I chosen TSO
1611 		 * plus custom checksum offload with work-around which
1612 		 * will cover most common usage for this consumer
1613 		 * ethernet controller. The work-around takes a lot of
1614 		 * CPU cycles if Tx buffer is not aligned on 4 bytes
1615 		 * boundary, though.
1616 		 */
1617 		cflags |= ALE_TD_CXSUM;
1618 		/* Set checksum start offset. */
1619 		cflags |= (poff << ALE_TD_CSUM_PLOADOFFSET_SHIFT);
1620 		/* Set checksum insertion position of TCP/UDP. */
1621 		cflags |= ((poff + m->m_pkthdr.csum_data) <<
1622 		    ALE_TD_CSUM_XSUMOFFSET_SHIFT);
1623 	}
1624 
1625 	/* Configure VLAN hardware tag insertion. */
1626 	if ((m->m_flags & M_VLANTAG) != 0) {
1627 		vtag = ALE_TX_VLAN_TAG(m->m_pkthdr.ether_vlantag);
1628 		vtag = ((vtag << ALE_TD_VLAN_SHIFT) & ALE_TD_VLAN_MASK);
1629 		cflags |= ALE_TD_INSERT_VLAN_TAG;
1630 	}
1631 
1632 	desc = NULL;
1633 	for (i = 0; i < nsegs; i++) {
1634 		desc = &sc->ale_cdata.ale_tx_ring[prod];
1635 		desc->addr = htole64(txsegs[i].ds_addr);
1636 		desc->len = htole32(ALE_TX_BYTES(txsegs[i].ds_len) | vtag);
1637 		desc->flags = htole32(cflags);
1638 		sc->ale_cdata.ale_tx_cnt++;
1639 		ALE_DESC_INC(prod, ALE_TX_RING_CNT);
1640 	}
1641 	/* Update producer index. */
1642 	sc->ale_cdata.ale_tx_prod = prod;
1643 
1644 	/* Finally set EOP on the last descriptor. */
1645 	prod = (prod + ALE_TX_RING_CNT - 1) % ALE_TX_RING_CNT;
1646 	desc = &sc->ale_cdata.ale_tx_ring[prod];
1647 	desc->flags |= htole32(ALE_TD_EOP);
1648 
1649 	/* Swap dmamap of the first and the last. */
1650 	txd = &sc->ale_cdata.ale_txdesc[prod];
1651 	map = txd_last->tx_dmamap;
1652 	txd_last->tx_dmamap = txd->tx_dmamap;
1653 	txd->tx_dmamap = map;
1654 	txd->tx_m = m;
1655 
1656 	/* Sync descriptors. */
1657 	bus_dmamap_sync(sc->ale_cdata.ale_tx_ring_tag,
1658 	    sc->ale_cdata.ale_tx_ring_map, BUS_DMASYNC_PREWRITE);
1659 
1660 	return (0);
1661 }
1662 
1663 static void
1664 ale_start(struct ifnet *ifp, struct ifaltq_subque *ifsq)
1665 {
1666         struct ale_softc *sc = ifp->if_softc;
1667 	struct mbuf *m_head;
1668 	int enq;
1669 
1670 	ASSERT_ALTQ_SQ_DEFAULT(ifp, ifsq);
1671 	ASSERT_SERIALIZED(ifp->if_serializer);
1672 
1673 	if ((sc->ale_flags & ALE_FLAG_LINK) == 0) {
1674 		ifq_purge(&ifp->if_snd);
1675 		return;
1676 	}
1677 
1678 	if ((ifp->if_flags & IFF_RUNNING) == 0 || ifq_is_oactive(&ifp->if_snd))
1679 		return;
1680 
1681 	/* Reclaim transmitted frames. */
1682 	if (sc->ale_cdata.ale_tx_cnt >= ALE_TX_DESC_HIWAT)
1683 		ale_txeof(sc);
1684 
1685 	enq = 0;
1686 	while (!ifq_is_empty(&ifp->if_snd)) {
1687 		m_head = ifq_dequeue(&ifp->if_snd);
1688 		if (m_head == NULL)
1689 			break;
1690 
1691 		/*
1692 		 * Pack the data into the transmit ring. If we
1693 		 * don't have room, set the OACTIVE flag and wait
1694 		 * for the NIC to drain the ring.
1695 		 */
1696 		if (ale_encap(sc, &m_head)) {
1697 			if (m_head == NULL)
1698 				break;
1699 			ifq_prepend(&ifp->if_snd, m_head);
1700 			ifq_set_oactive(&ifp->if_snd);
1701 			break;
1702 		}
1703 		enq = 1;
1704 
1705 		/*
1706 		 * If there's a BPF listener, bounce a copy of this frame
1707 		 * to him.
1708 		 */
1709 		ETHER_BPF_MTAP(ifp, m_head);
1710 	}
1711 
1712 	if (enq) {
1713 		/* Kick. */
1714 		CSR_WRITE_4(sc, ALE_MBOX_TPD_PROD_IDX,
1715 		    sc->ale_cdata.ale_tx_prod);
1716 
1717 		/* Set a timeout in case the chip goes out to lunch. */
1718 		ifp->if_timer = ALE_TX_TIMEOUT;
1719 	}
1720 }
1721 
1722 static void
1723 ale_watchdog(struct ifnet *ifp)
1724 {
1725 	struct ale_softc *sc = ifp->if_softc;
1726 
1727 	ASSERT_SERIALIZED(ifp->if_serializer);
1728 
1729 	if ((sc->ale_flags & ALE_FLAG_LINK) == 0) {
1730 		if_printf(ifp, "watchdog timeout (lost link)\n");
1731 		IFNET_STAT_INC(ifp, oerrors, 1);
1732 		ale_init(sc);
1733 		return;
1734 	}
1735 
1736 	if_printf(ifp, "watchdog timeout -- resetting\n");
1737 	IFNET_STAT_INC(ifp, oerrors, 1);
1738 	ale_init(sc);
1739 
1740 	if (!ifq_is_empty(&ifp->if_snd))
1741 		if_devstart(ifp);
1742 }
1743 
1744 static int
1745 ale_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data, struct ucred *cr)
1746 {
1747 	struct ale_softc *sc;
1748 	struct ifreq *ifr;
1749 	struct mii_data *mii;
1750 	int error, mask;
1751 
1752 	ASSERT_SERIALIZED(ifp->if_serializer);
1753 
1754 	sc = ifp->if_softc;
1755 	ifr = (struct ifreq *)data;
1756 	error = 0;
1757 
1758 	switch (cmd) {
1759 	case SIOCSIFMTU:
1760 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > ALE_JUMBO_MTU ||
1761 		    ((sc->ale_flags & ALE_FLAG_JUMBO) == 0 &&
1762 		    ifr->ifr_mtu > ETHERMTU))
1763 			error = EINVAL;
1764 		else if (ifp->if_mtu != ifr->ifr_mtu) {
1765 			ifp->if_mtu = ifr->ifr_mtu;
1766 			if ((ifp->if_flags & IFF_RUNNING) != 0)
1767 				ale_init(sc);
1768 		}
1769 		break;
1770 
1771 	case SIOCSIFFLAGS:
1772 		if ((ifp->if_flags & IFF_UP) != 0) {
1773 			if ((ifp->if_flags & IFF_RUNNING) != 0) {
1774 				if (((ifp->if_flags ^ sc->ale_if_flags)
1775 				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
1776 					ale_rxfilter(sc);
1777 			} else {
1778 				if ((sc->ale_flags & ALE_FLAG_DETACH) == 0)
1779 					ale_init(sc);
1780 			}
1781 		} else {
1782 			if ((ifp->if_flags & IFF_RUNNING) != 0)
1783 				ale_stop(sc);
1784 		}
1785 		sc->ale_if_flags = ifp->if_flags;
1786 		break;
1787 
1788 	case SIOCADDMULTI:
1789 	case SIOCDELMULTI:
1790 		if ((ifp->if_flags & IFF_RUNNING) != 0)
1791 			ale_rxfilter(sc);
1792 		break;
1793 
1794 	case SIOCSIFMEDIA:
1795 	case SIOCGIFMEDIA:
1796 		mii = device_get_softc(sc->ale_miibus);
1797 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, cmd);
1798 		break;
1799 
1800 	case SIOCSIFCAP:
1801 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1802 		if ((mask & IFCAP_TXCSUM) != 0 &&
1803 		    (ifp->if_capabilities & IFCAP_TXCSUM) != 0) {
1804 			ifp->if_capenable ^= IFCAP_TXCSUM;
1805 			if ((ifp->if_capenable & IFCAP_TXCSUM) != 0)
1806 				ifp->if_hwassist |= ALE_CSUM_FEATURES;
1807 			else
1808 				ifp->if_hwassist &= ~ALE_CSUM_FEATURES;
1809 		}
1810 		if ((mask & IFCAP_RXCSUM) != 0 &&
1811 		    (ifp->if_capabilities & IFCAP_RXCSUM) != 0)
1812 			ifp->if_capenable ^= IFCAP_RXCSUM;
1813 
1814 		if ((mask & IFCAP_VLAN_HWTAGGING) != 0 &&
1815 		    (ifp->if_capabilities & IFCAP_VLAN_HWTAGGING) != 0) {
1816 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1817 			ale_rxvlan(sc);
1818 		}
1819 		break;
1820 
1821 	default:
1822 		error = ether_ioctl(ifp, cmd, data);
1823 		break;
1824 	}
1825 	return (error);
1826 }
1827 
1828 static void
1829 ale_mac_config(struct ale_softc *sc)
1830 {
1831 	struct mii_data *mii;
1832 	uint32_t reg;
1833 
1834 	mii = device_get_softc(sc->ale_miibus);
1835 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
1836 	reg &= ~(MAC_CFG_FULL_DUPLEX | MAC_CFG_TX_FC | MAC_CFG_RX_FC |
1837 	    MAC_CFG_SPEED_MASK);
1838 	/* Reprogram MAC with resolved speed/duplex. */
1839 	switch (IFM_SUBTYPE(mii->mii_media_active)) {
1840 	case IFM_10_T:
1841 	case IFM_100_TX:
1842 		reg |= MAC_CFG_SPEED_10_100;
1843 		break;
1844 	case IFM_1000_T:
1845 		reg |= MAC_CFG_SPEED_1000;
1846 		break;
1847 	}
1848 	if ((IFM_OPTIONS(mii->mii_media_active) & IFM_FDX) != 0) {
1849 		reg |= MAC_CFG_FULL_DUPLEX;
1850 #ifdef notyet
1851 		if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_TXPAUSE) != 0)
1852 			reg |= MAC_CFG_TX_FC;
1853 		if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_RXPAUSE) != 0)
1854 			reg |= MAC_CFG_RX_FC;
1855 #endif
1856 	}
1857 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
1858 }
1859 
1860 static void
1861 ale_stats_clear(struct ale_softc *sc)
1862 {
1863 	struct smb sb;
1864 	uint32_t *reg;
1865 	int i;
1866 
1867 	for (reg = &sb.rx_frames, i = 0; reg <= &sb.rx_pkts_filtered; reg++) {
1868 		CSR_READ_4(sc, ALE_RX_MIB_BASE + i);
1869 		i += sizeof(uint32_t);
1870 	}
1871 	/* Read Tx statistics. */
1872 	for (reg = &sb.tx_frames, i = 0; reg <= &sb.tx_mcast_bytes; reg++) {
1873 		CSR_READ_4(sc, ALE_TX_MIB_BASE + i);
1874 		i += sizeof(uint32_t);
1875 	}
1876 }
1877 
1878 static void
1879 ale_stats_update(struct ale_softc *sc)
1880 {
1881 	struct ale_hw_stats *stat;
1882 	struct smb sb, *smb;
1883 	struct ifnet *ifp;
1884 	uint32_t *reg;
1885 	int i;
1886 
1887 	ifp = &sc->arpcom.ac_if;
1888 	stat = &sc->ale_stats;
1889 	smb = &sb;
1890 
1891 	/* Read Rx statistics. */
1892 	for (reg = &sb.rx_frames, i = 0; reg <= &sb.rx_pkts_filtered; reg++) {
1893 		*reg = CSR_READ_4(sc, ALE_RX_MIB_BASE + i);
1894 		i += sizeof(uint32_t);
1895 	}
1896 	/* Read Tx statistics. */
1897 	for (reg = &sb.tx_frames, i = 0; reg <= &sb.tx_mcast_bytes; reg++) {
1898 		*reg = CSR_READ_4(sc, ALE_TX_MIB_BASE + i);
1899 		i += sizeof(uint32_t);
1900 	}
1901 
1902 	/* Rx stats. */
1903 	stat->rx_frames += smb->rx_frames;
1904 	stat->rx_bcast_frames += smb->rx_bcast_frames;
1905 	stat->rx_mcast_frames += smb->rx_mcast_frames;
1906 	stat->rx_pause_frames += smb->rx_pause_frames;
1907 	stat->rx_control_frames += smb->rx_control_frames;
1908 	stat->rx_crcerrs += smb->rx_crcerrs;
1909 	stat->rx_lenerrs += smb->rx_lenerrs;
1910 	stat->rx_bytes += smb->rx_bytes;
1911 	stat->rx_runts += smb->rx_runts;
1912 	stat->rx_fragments += smb->rx_fragments;
1913 	stat->rx_pkts_64 += smb->rx_pkts_64;
1914 	stat->rx_pkts_65_127 += smb->rx_pkts_65_127;
1915 	stat->rx_pkts_128_255 += smb->rx_pkts_128_255;
1916 	stat->rx_pkts_256_511 += smb->rx_pkts_256_511;
1917 	stat->rx_pkts_512_1023 += smb->rx_pkts_512_1023;
1918 	stat->rx_pkts_1024_1518 += smb->rx_pkts_1024_1518;
1919 	stat->rx_pkts_1519_max += smb->rx_pkts_1519_max;
1920 	stat->rx_pkts_truncated += smb->rx_pkts_truncated;
1921 	stat->rx_fifo_oflows += smb->rx_fifo_oflows;
1922 	stat->rx_rrs_errs += smb->rx_rrs_errs;
1923 	stat->rx_alignerrs += smb->rx_alignerrs;
1924 	stat->rx_bcast_bytes += smb->rx_bcast_bytes;
1925 	stat->rx_mcast_bytes += smb->rx_mcast_bytes;
1926 	stat->rx_pkts_filtered += smb->rx_pkts_filtered;
1927 
1928 	/* Tx stats. */
1929 	stat->tx_frames += smb->tx_frames;
1930 	stat->tx_bcast_frames += smb->tx_bcast_frames;
1931 	stat->tx_mcast_frames += smb->tx_mcast_frames;
1932 	stat->tx_pause_frames += smb->tx_pause_frames;
1933 	stat->tx_excess_defer += smb->tx_excess_defer;
1934 	stat->tx_control_frames += smb->tx_control_frames;
1935 	stat->tx_deferred += smb->tx_deferred;
1936 	stat->tx_bytes += smb->tx_bytes;
1937 	stat->tx_pkts_64 += smb->tx_pkts_64;
1938 	stat->tx_pkts_65_127 += smb->tx_pkts_65_127;
1939 	stat->tx_pkts_128_255 += smb->tx_pkts_128_255;
1940 	stat->tx_pkts_256_511 += smb->tx_pkts_256_511;
1941 	stat->tx_pkts_512_1023 += smb->tx_pkts_512_1023;
1942 	stat->tx_pkts_1024_1518 += smb->tx_pkts_1024_1518;
1943 	stat->tx_pkts_1519_max += smb->tx_pkts_1519_max;
1944 	stat->tx_single_colls += smb->tx_single_colls;
1945 	stat->tx_multi_colls += smb->tx_multi_colls;
1946 	stat->tx_late_colls += smb->tx_late_colls;
1947 	stat->tx_excess_colls += smb->tx_excess_colls;
1948 	stat->tx_abort += smb->tx_abort;
1949 	stat->tx_underrun += smb->tx_underrun;
1950 	stat->tx_desc_underrun += smb->tx_desc_underrun;
1951 	stat->tx_lenerrs += smb->tx_lenerrs;
1952 	stat->tx_pkts_truncated += smb->tx_pkts_truncated;
1953 	stat->tx_bcast_bytes += smb->tx_bcast_bytes;
1954 	stat->tx_mcast_bytes += smb->tx_mcast_bytes;
1955 
1956 	/* Update counters in ifnet. */
1957 	IFNET_STAT_INC(ifp, opackets, smb->tx_frames);
1958 
1959 	IFNET_STAT_INC(ifp, collisions, smb->tx_single_colls +
1960 	    smb->tx_multi_colls * 2 + smb->tx_late_colls +
1961 	    smb->tx_abort * HDPX_CFG_RETRY_DEFAULT);
1962 
1963 	/*
1964 	 * XXX
1965 	 * tx_pkts_truncated counter looks suspicious. It constantly
1966 	 * increments with no sign of Tx errors. This may indicate
1967 	 * the counter name is not correct one so I've removed the
1968 	 * counter in output errors.
1969 	 */
1970 	IFNET_STAT_INC(ifp, oerrors, smb->tx_abort + smb->tx_late_colls +
1971 	    smb->tx_underrun);
1972 
1973 	IFNET_STAT_INC(ifp, ipackets, smb->rx_frames);
1974 
1975 	IFNET_STAT_INC(ifp, ierrors, smb->rx_crcerrs + smb->rx_lenerrs +
1976 	    smb->rx_runts + smb->rx_pkts_truncated +
1977 	    smb->rx_fifo_oflows + smb->rx_rrs_errs +
1978 	    smb->rx_alignerrs);
1979 }
1980 
1981 static void
1982 ale_intr(void *xsc)
1983 {
1984 	struct ale_softc *sc = xsc;
1985 	struct ifnet *ifp = &sc->arpcom.ac_if;
1986 	uint32_t status;
1987 
1988 	ASSERT_SERIALIZED(ifp->if_serializer);
1989 
1990 	status = CSR_READ_4(sc, ALE_INTR_STATUS);
1991 	if ((status & ALE_INTRS) == 0)
1992 		return;
1993 
1994 	/* Acknowledge and disable interrupts. */
1995 	CSR_WRITE_4(sc, ALE_INTR_STATUS, status | INTR_DIS_INT);
1996 
1997 	if ((ifp->if_flags & IFF_RUNNING) != 0) {
1998 		int error;
1999 
2000 		error = ale_rxeof(sc);
2001 		if (error) {
2002 			sc->ale_stats.reset_brk_seq++;
2003 			ale_init(sc);
2004 			return;
2005 		}
2006 
2007 		if ((status & (INTR_DMA_RD_TO_RST | INTR_DMA_WR_TO_RST)) != 0) {
2008 			if ((status & INTR_DMA_RD_TO_RST) != 0)
2009 				device_printf(sc->ale_dev,
2010 				    "DMA read error! -- resetting\n");
2011 			if ((status & INTR_DMA_WR_TO_RST) != 0)
2012 				device_printf(sc->ale_dev,
2013 				    "DMA write error! -- resetting\n");
2014 			ale_init(sc);
2015 			return;
2016 		}
2017 
2018 		ale_txeof(sc);
2019 		if (!ifq_is_empty(&ifp->if_snd))
2020 			if_devstart(ifp);
2021 	}
2022 
2023 	/* Re-enable interrupts. */
2024 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0x7FFFFFFF);
2025 }
2026 
2027 static void
2028 ale_txeof(struct ale_softc *sc)
2029 {
2030 	struct ifnet *ifp = &sc->arpcom.ac_if;
2031 	struct ale_txdesc *txd;
2032 	uint32_t cons, prod;
2033 	int prog;
2034 
2035 	if (sc->ale_cdata.ale_tx_cnt == 0)
2036 		return;
2037 
2038 	bus_dmamap_sync(sc->ale_cdata.ale_tx_ring_tag,
2039 	    sc->ale_cdata.ale_tx_ring_map, BUS_DMASYNC_POSTREAD);
2040 	if ((sc->ale_flags & ALE_FLAG_TXCMB_BUG) == 0) {
2041 		bus_dmamap_sync(sc->ale_cdata.ale_tx_cmb_tag,
2042 		    sc->ale_cdata.ale_tx_cmb_map, BUS_DMASYNC_POSTREAD);
2043 		prod = *sc->ale_cdata.ale_tx_cmb & TPD_CNT_MASK;
2044 	} else
2045 		prod = CSR_READ_2(sc, ALE_TPD_CONS_IDX);
2046 	cons = sc->ale_cdata.ale_tx_cons;
2047 	/*
2048 	 * Go through our Tx list and free mbufs for those
2049 	 * frames which have been transmitted.
2050 	 */
2051 	for (prog = 0; cons != prod; prog++,
2052 	     ALE_DESC_INC(cons, ALE_TX_RING_CNT)) {
2053 		if (sc->ale_cdata.ale_tx_cnt <= 0)
2054 			break;
2055 		prog++;
2056 		ifq_clr_oactive(&ifp->if_snd);
2057 		sc->ale_cdata.ale_tx_cnt--;
2058 		txd = &sc->ale_cdata.ale_txdesc[cons];
2059 		if (txd->tx_m != NULL) {
2060 			/* Reclaim transmitted mbufs. */
2061 			bus_dmamap_unload(sc->ale_cdata.ale_tx_tag,
2062 			    txd->tx_dmamap);
2063 			m_freem(txd->tx_m);
2064 			txd->tx_m = NULL;
2065 		}
2066 	}
2067 
2068 	if (prog > 0) {
2069 		sc->ale_cdata.ale_tx_cons = cons;
2070 		/*
2071 		 * Unarm watchdog timer only when there is no pending
2072 		 * Tx descriptors in queue.
2073 		 */
2074 		if (sc->ale_cdata.ale_tx_cnt == 0)
2075 			ifp->if_timer = 0;
2076 	}
2077 }
2078 
2079 static void
2080 ale_rx_update_page(struct ale_softc *sc, struct ale_rx_page **page,
2081     uint32_t length, uint32_t *prod)
2082 {
2083 	struct ale_rx_page *rx_page;
2084 
2085 	rx_page = *page;
2086 	/* Update consumer position. */
2087 	rx_page->cons += roundup(length + sizeof(struct rx_rs),
2088 	    ALE_RX_PAGE_ALIGN);
2089 	if (rx_page->cons >= ALE_RX_PAGE_SZ) {
2090 		/*
2091 		 * End of Rx page reached, let hardware reuse
2092 		 * this page.
2093 		 */
2094 		rx_page->cons = 0;
2095 		*rx_page->cmb_addr = 0;
2096 		bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2097 				BUS_DMASYNC_PREWRITE);
2098 		CSR_WRITE_1(sc, ALE_RXF0_PAGE0 + sc->ale_cdata.ale_rx_curp,
2099 		    RXF_VALID);
2100 		/* Switch to alternate Rx page. */
2101 		sc->ale_cdata.ale_rx_curp ^= 1;
2102 		rx_page = *page =
2103 		    &sc->ale_cdata.ale_rx_page[sc->ale_cdata.ale_rx_curp];
2104 		/* Page flipped, sync CMB and Rx page. */
2105 		bus_dmamap_sync(rx_page->page_tag, rx_page->page_map,
2106 		    BUS_DMASYNC_POSTREAD);
2107 		bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2108 		    BUS_DMASYNC_POSTREAD);
2109 		/* Sync completed, cache updated producer index. */
2110 		*prod = *rx_page->cmb_addr;
2111 	}
2112 }
2113 
2114 
2115 /*
2116  * It seems that AR81xx controller can compute partial checksum.
2117  * The partial checksum value can be used to accelerate checksum
2118  * computation for fragmented TCP/UDP packets. Upper network stack
2119  * already takes advantage of the partial checksum value in IP
2120  * reassembly stage. But I'm not sure the correctness of the
2121  * partial hardware checksum assistance due to lack of data sheet.
2122  * In addition, the Rx feature of controller that requires copying
2123  * for every frames effectively nullifies one of most nice offload
2124  * capability of controller.
2125  */
2126 static void
2127 ale_rxcsum(struct ale_softc *sc, struct mbuf *m, uint32_t status)
2128 {
2129 	struct ifnet *ifp = &sc->arpcom.ac_if;
2130 	struct ip *ip;
2131 	char *p;
2132 
2133 	m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
2134 	if ((status & ALE_RD_IPCSUM_NOK) == 0)
2135 		m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
2136 
2137 	if ((sc->ale_flags & ALE_FLAG_RXCSUM_BUG) == 0) {
2138 		if (((status & ALE_RD_IPV4_FRAG) == 0) &&
2139 		    ((status & (ALE_RD_TCP | ALE_RD_UDP)) != 0) &&
2140 		    ((status & ALE_RD_TCP_UDPCSUM_NOK) == 0)) {
2141 			m->m_pkthdr.csum_flags |=
2142 			    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2143 			m->m_pkthdr.csum_data = 0xffff;
2144 		}
2145 	} else {
2146 		if ((status & (ALE_RD_TCP | ALE_RD_UDP)) != 0 &&
2147 		    (status & ALE_RD_TCP_UDPCSUM_NOK) == 0) {
2148 			p = mtod(m, char *);
2149 			p += ETHER_HDR_LEN;
2150 			if ((status & ALE_RD_802_3) != 0)
2151 				p += LLC_SNAPFRAMELEN;
2152 			if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0 &&
2153 			    (status & ALE_RD_VLAN) != 0)
2154 				p += EVL_ENCAPLEN;
2155 			ip = (struct ip *)p;
2156 			if (ip->ip_off != 0 && (status & ALE_RD_IPV4_DF) == 0)
2157 				return;
2158 			m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
2159 			    CSUM_PSEUDO_HDR;
2160 			m->m_pkthdr.csum_data = 0xffff;
2161 		}
2162 	}
2163 	/*
2164 	 * Don't mark bad checksum for TCP/UDP frames
2165 	 * as fragmented frames may always have set
2166 	 * bad checksummed bit of frame status.
2167 	 */
2168 }
2169 
2170 /* Process received frames. */
2171 static int
2172 ale_rxeof(struct ale_softc *sc)
2173 {
2174 	struct ifnet *ifp = &sc->arpcom.ac_if;
2175 	struct ale_rx_page *rx_page;
2176 	struct rx_rs *rs;
2177 	struct mbuf *m;
2178 	uint32_t length, prod, seqno, status, vtags;
2179 	int prog;
2180 
2181 	rx_page = &sc->ale_cdata.ale_rx_page[sc->ale_cdata.ale_rx_curp];
2182 	bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2183 			BUS_DMASYNC_POSTREAD);
2184 	bus_dmamap_sync(rx_page->page_tag, rx_page->page_map,
2185 			BUS_DMASYNC_POSTREAD);
2186 	/*
2187 	 * Don't directly access producer index as hardware may
2188 	 * update it while Rx handler is in progress. It would
2189 	 * be even better if there is a way to let hardware
2190 	 * know how far driver processed its received frames.
2191 	 * Alternatively, hardware could provide a way to disable
2192 	 * CMB updates until driver acknowledges the end of CMB
2193 	 * access.
2194 	 */
2195 	prod = *rx_page->cmb_addr;
2196 	for (prog = 0; ; prog++) {
2197 		if (rx_page->cons >= prod)
2198 			break;
2199 		rs = (struct rx_rs *)(rx_page->page_addr + rx_page->cons);
2200 		seqno = ALE_RX_SEQNO(le32toh(rs->seqno));
2201 		if (sc->ale_cdata.ale_rx_seqno != seqno) {
2202 			/*
2203 			 * Normally I believe this should not happen unless
2204 			 * severe driver bug or corrupted memory. However
2205 			 * it seems to happen under certain conditions which
2206 			 * is triggered by abrupt Rx events such as initiation
2207 			 * of bulk transfer of remote host. It's not easy to
2208 			 * reproduce this and I doubt it could be related
2209 			 * with FIFO overflow of hardware or activity of Tx
2210 			 * CMB updates. I also remember similar behaviour
2211 			 * seen on RealTek 8139 which uses resembling Rx
2212 			 * scheme.
2213 			 */
2214 			if (bootverbose)
2215 				device_printf(sc->ale_dev,
2216 				    "garbled seq: %u, expected: %u -- "
2217 				    "resetting!\n", seqno,
2218 				    sc->ale_cdata.ale_rx_seqno);
2219 			return (EIO);
2220 		}
2221 		/* Frame received. */
2222 		sc->ale_cdata.ale_rx_seqno++;
2223 		length = ALE_RX_BYTES(le32toh(rs->length));
2224 		status = le32toh(rs->flags);
2225 		if ((status & ALE_RD_ERROR) != 0) {
2226 			/*
2227 			 * We want to pass the following frames to upper
2228 			 * layer regardless of error status of Rx return
2229 			 * status.
2230 			 *
2231 			 *  o IP/TCP/UDP checksum is bad.
2232 			 *  o frame length and protocol specific length
2233 			 *     does not match.
2234 			 */
2235 			if ((status & (ALE_RD_CRC | ALE_RD_CODE |
2236 			    ALE_RD_DRIBBLE | ALE_RD_RUNT | ALE_RD_OFLOW |
2237 			    ALE_RD_TRUNC)) != 0) {
2238 				ale_rx_update_page(sc, &rx_page, length, &prod);
2239 				continue;
2240 			}
2241 		}
2242 		/*
2243 		 * m_devget(9) is major bottle-neck of ale(4)(It comes
2244 		 * from hardware limitation). For jumbo frames we could
2245 		 * get a slightly better performance if driver use
2246 		 * m_getjcl(9) with proper buffer size argument. However
2247 		 * that would make code more complicated and I don't
2248 		 * think users would expect good Rx performance numbers
2249 		 * on these low-end consumer ethernet controller.
2250 		 */
2251 		m = m_devget((char *)(rs + 1), length - ETHER_CRC_LEN,
2252 		    ETHER_ALIGN, ifp, NULL);
2253 		if (m == NULL) {
2254 			IFNET_STAT_INC(ifp, iqdrops, 1);
2255 			ale_rx_update_page(sc, &rx_page, length, &prod);
2256 			continue;
2257 		}
2258 		if ((ifp->if_capenable & IFCAP_RXCSUM) != 0 &&
2259 		    (status & ALE_RD_IPV4) != 0)
2260 			ale_rxcsum(sc, m, status);
2261 		if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0 &&
2262 		    (status & ALE_RD_VLAN) != 0) {
2263 			vtags = ALE_RX_VLAN(le32toh(rs->vtags));
2264 			m->m_pkthdr.ether_vlantag = ALE_RX_VLAN_TAG(vtags);
2265 			m->m_flags |= M_VLANTAG;
2266 		}
2267 
2268 		/* Pass it to upper layer. */
2269 		ifp->if_input(ifp, m);
2270 
2271 		ale_rx_update_page(sc, &rx_page, length, &prod);
2272 	}
2273 	return 0;
2274 }
2275 
2276 static void
2277 ale_tick(void *xsc)
2278 {
2279 	struct ale_softc *sc = xsc;
2280 	struct ifnet *ifp = &sc->arpcom.ac_if;
2281 	struct mii_data *mii;
2282 
2283 	lwkt_serialize_enter(ifp->if_serializer);
2284 
2285 	mii = device_get_softc(sc->ale_miibus);
2286 	mii_tick(mii);
2287 	ale_stats_update(sc);
2288 
2289 	callout_reset(&sc->ale_tick_ch, hz, ale_tick, sc);
2290 
2291 	lwkt_serialize_exit(ifp->if_serializer);
2292 }
2293 
2294 static void
2295 ale_reset(struct ale_softc *sc)
2296 {
2297 	uint32_t reg;
2298 	int i;
2299 
2300 	/* Initialize PCIe module. From Linux. */
2301 	CSR_WRITE_4(sc, 0x1008, CSR_READ_4(sc, 0x1008) | 0x8000);
2302 
2303 	CSR_WRITE_4(sc, ALE_MASTER_CFG, MASTER_RESET);
2304 	for (i = ALE_RESET_TIMEOUT; i > 0; i--) {
2305 		DELAY(10);
2306 		if ((CSR_READ_4(sc, ALE_MASTER_CFG) & MASTER_RESET) == 0)
2307 			break;
2308 	}
2309 	if (i == 0)
2310 		device_printf(sc->ale_dev, "master reset timeout!\n");
2311 
2312 	for (i = ALE_RESET_TIMEOUT; i > 0; i--) {
2313 		if ((reg = CSR_READ_4(sc, ALE_IDLE_STATUS)) == 0)
2314 			break;
2315 		DELAY(10);
2316 	}
2317 
2318 	if (i == 0)
2319 		device_printf(sc->ale_dev, "reset timeout(0x%08x)!\n", reg);
2320 }
2321 
2322 static void
2323 ale_init(void *xsc)
2324 {
2325 	struct ale_softc *sc = xsc;
2326 	struct ifnet *ifp = &sc->arpcom.ac_if;
2327 	struct mii_data *mii;
2328 	uint8_t eaddr[ETHER_ADDR_LEN];
2329 	bus_addr_t paddr;
2330 	uint32_t reg, rxf_hi, rxf_lo;
2331 
2332 	ASSERT_SERIALIZED(ifp->if_serializer);
2333 
2334 	mii = device_get_softc(sc->ale_miibus);
2335 
2336 	/*
2337 	 * Cancel any pending I/O.
2338 	 */
2339 	ale_stop(sc);
2340 
2341 	/*
2342 	 * Reset the chip to a known state.
2343 	 */
2344 	ale_reset(sc);
2345 
2346 	/* Initialize Tx descriptors, DMA memory blocks. */
2347 	ale_init_rx_pages(sc);
2348 	ale_init_tx_ring(sc);
2349 
2350 	/* Reprogram the station address. */
2351 	bcopy(IF_LLADDR(ifp), eaddr, ETHER_ADDR_LEN);
2352 	CSR_WRITE_4(sc, ALE_PAR0,
2353 	    eaddr[2] << 24 | eaddr[3] << 16 | eaddr[4] << 8 | eaddr[5]);
2354 	CSR_WRITE_4(sc, ALE_PAR1, eaddr[0] << 8 | eaddr[1]);
2355 
2356 	/*
2357 	 * Clear WOL status and disable all WOL feature as WOL
2358 	 * would interfere Rx operation under normal environments.
2359 	 */
2360 	CSR_READ_4(sc, ALE_WOL_CFG);
2361 	CSR_WRITE_4(sc, ALE_WOL_CFG, 0);
2362 
2363 	/*
2364 	 * Set Tx descriptor/RXF0/CMB base addresses. They share
2365 	 * the same high address part of DMAable region.
2366 	 */
2367 	paddr = sc->ale_cdata.ale_tx_ring_paddr;
2368 	CSR_WRITE_4(sc, ALE_TPD_ADDR_HI, ALE_ADDR_HI(paddr));
2369 	CSR_WRITE_4(sc, ALE_TPD_ADDR_LO, ALE_ADDR_LO(paddr));
2370 	CSR_WRITE_4(sc, ALE_TPD_CNT,
2371 	    (ALE_TX_RING_CNT << TPD_CNT_SHIFT) & TPD_CNT_MASK);
2372 
2373 	/* Set Rx page base address, note we use single queue. */
2374 	paddr = sc->ale_cdata.ale_rx_page[0].page_paddr;
2375 	CSR_WRITE_4(sc, ALE_RXF0_PAGE0_ADDR_LO, ALE_ADDR_LO(paddr));
2376 	paddr = sc->ale_cdata.ale_rx_page[1].page_paddr;
2377 	CSR_WRITE_4(sc, ALE_RXF0_PAGE1_ADDR_LO, ALE_ADDR_LO(paddr));
2378 
2379 	/* Set Tx/Rx CMB addresses. */
2380 	paddr = sc->ale_cdata.ale_tx_cmb_paddr;
2381 	CSR_WRITE_4(sc, ALE_TX_CMB_ADDR_LO, ALE_ADDR_LO(paddr));
2382 	paddr = sc->ale_cdata.ale_rx_page[0].cmb_paddr;
2383 	CSR_WRITE_4(sc, ALE_RXF0_CMB0_ADDR_LO, ALE_ADDR_LO(paddr));
2384 	paddr = sc->ale_cdata.ale_rx_page[1].cmb_paddr;
2385 	CSR_WRITE_4(sc, ALE_RXF0_CMB1_ADDR_LO, ALE_ADDR_LO(paddr));
2386 
2387 	/* Mark RXF0 is valid. */
2388 	CSR_WRITE_1(sc, ALE_RXF0_PAGE0, RXF_VALID);
2389 	CSR_WRITE_1(sc, ALE_RXF0_PAGE1, RXF_VALID);
2390 	/*
2391 	 * No need to initialize RFX1/RXF2/RXF3. We don't use
2392 	 * multi-queue yet.
2393 	 */
2394 
2395 	/* Set Rx page size, excluding guard frame size. */
2396 	CSR_WRITE_4(sc, ALE_RXF_PAGE_SIZE, ALE_RX_PAGE_SZ);
2397 
2398 	/* Tell hardware that we're ready to load DMA blocks. */
2399 	CSR_WRITE_4(sc, ALE_DMA_BLOCK, DMA_BLOCK_LOAD);
2400 
2401 	/* Set Rx/Tx interrupt trigger threshold. */
2402 	CSR_WRITE_4(sc, ALE_INT_TRIG_THRESH, (1 << INT_TRIG_RX_THRESH_SHIFT) |
2403 	    (4 << INT_TRIG_TX_THRESH_SHIFT));
2404 	/*
2405 	 * XXX
2406 	 * Set interrupt trigger timer, its purpose and relation
2407 	 * with interrupt moderation mechanism is not clear yet.
2408 	 */
2409 	CSR_WRITE_4(sc, ALE_INT_TRIG_TIMER,
2410 	    ((ALE_USECS(10) << INT_TRIG_RX_TIMER_SHIFT) |
2411 	    (ALE_USECS(1000) << INT_TRIG_TX_TIMER_SHIFT)));
2412 
2413 	/* Configure interrupt moderation timer. */
2414 	reg = ALE_USECS(sc->ale_int_rx_mod) << IM_TIMER_RX_SHIFT;
2415 	reg |= ALE_USECS(sc->ale_int_tx_mod) << IM_TIMER_TX_SHIFT;
2416 	CSR_WRITE_4(sc, ALE_IM_TIMER, reg);
2417 	reg = CSR_READ_4(sc, ALE_MASTER_CFG);
2418 	reg &= ~(MASTER_CHIP_REV_MASK | MASTER_CHIP_ID_MASK);
2419 	reg &= ~(MASTER_IM_RX_TIMER_ENB | MASTER_IM_TX_TIMER_ENB);
2420 	if (ALE_USECS(sc->ale_int_rx_mod) != 0)
2421 		reg |= MASTER_IM_RX_TIMER_ENB;
2422 	if (ALE_USECS(sc->ale_int_tx_mod) != 0)
2423 		reg |= MASTER_IM_TX_TIMER_ENB;
2424 	CSR_WRITE_4(sc, ALE_MASTER_CFG, reg);
2425 	CSR_WRITE_2(sc, ALE_INTR_CLR_TIMER, ALE_USECS(1000));
2426 
2427 	/* Set Maximum frame size of controller. */
2428 	if (ifp->if_mtu < ETHERMTU)
2429 		sc->ale_max_frame_size = ETHERMTU;
2430 	else
2431 		sc->ale_max_frame_size = ifp->if_mtu;
2432 	sc->ale_max_frame_size += ETHER_HDR_LEN + EVL_ENCAPLEN + ETHER_CRC_LEN;
2433 	CSR_WRITE_4(sc, ALE_FRAME_SIZE, sc->ale_max_frame_size);
2434 
2435 	/* Configure IPG/IFG parameters. */
2436 	CSR_WRITE_4(sc, ALE_IPG_IFG_CFG,
2437 	    ((IPG_IFG_IPGT_DEFAULT << IPG_IFG_IPGT_SHIFT) & IPG_IFG_IPGT_MASK) |
2438 	    ((IPG_IFG_MIFG_DEFAULT << IPG_IFG_MIFG_SHIFT) & IPG_IFG_MIFG_MASK) |
2439 	    ((IPG_IFG_IPG1_DEFAULT << IPG_IFG_IPG1_SHIFT) & IPG_IFG_IPG1_MASK) |
2440 	    ((IPG_IFG_IPG2_DEFAULT << IPG_IFG_IPG2_SHIFT) & IPG_IFG_IPG2_MASK));
2441 
2442 	/* Set parameters for half-duplex media. */
2443 	CSR_WRITE_4(sc, ALE_HDPX_CFG,
2444 	    ((HDPX_CFG_LCOL_DEFAULT << HDPX_CFG_LCOL_SHIFT) &
2445 	    HDPX_CFG_LCOL_MASK) |
2446 	    ((HDPX_CFG_RETRY_DEFAULT << HDPX_CFG_RETRY_SHIFT) &
2447 	    HDPX_CFG_RETRY_MASK) | HDPX_CFG_EXC_DEF_EN |
2448 	    ((HDPX_CFG_ABEBT_DEFAULT << HDPX_CFG_ABEBT_SHIFT) &
2449 	    HDPX_CFG_ABEBT_MASK) |
2450 	    ((HDPX_CFG_JAMIPG_DEFAULT << HDPX_CFG_JAMIPG_SHIFT) &
2451 	    HDPX_CFG_JAMIPG_MASK));
2452 
2453 	/* Configure Tx jumbo frame parameters. */
2454 	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0) {
2455 		if (ifp->if_mtu < ETHERMTU)
2456 			reg = sc->ale_max_frame_size;
2457 		else if (ifp->if_mtu < 6 * 1024)
2458 			reg = (sc->ale_max_frame_size * 2) / 3;
2459 		else
2460 			reg = sc->ale_max_frame_size / 2;
2461 		CSR_WRITE_4(sc, ALE_TX_JUMBO_THRESH,
2462 		    roundup(reg, TX_JUMBO_THRESH_UNIT) >>
2463 		    TX_JUMBO_THRESH_UNIT_SHIFT);
2464 	}
2465 
2466 	/* Configure TxQ. */
2467 	reg = (128 << (sc->ale_dma_rd_burst >> DMA_CFG_RD_BURST_SHIFT))
2468 	    << TXQ_CFG_TX_FIFO_BURST_SHIFT;
2469 	reg |= (TXQ_CFG_TPD_BURST_DEFAULT << TXQ_CFG_TPD_BURST_SHIFT) &
2470 	    TXQ_CFG_TPD_BURST_MASK;
2471 	CSR_WRITE_4(sc, ALE_TXQ_CFG, reg | TXQ_CFG_ENHANCED_MODE | TXQ_CFG_ENB);
2472 
2473 	/* Configure Rx jumbo frame & flow control parameters. */
2474 	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0) {
2475 		reg = roundup(sc->ale_max_frame_size, RX_JUMBO_THRESH_UNIT);
2476 		CSR_WRITE_4(sc, ALE_RX_JUMBO_THRESH,
2477 		    (((reg >> RX_JUMBO_THRESH_UNIT_SHIFT) <<
2478 		    RX_JUMBO_THRESH_MASK_SHIFT) & RX_JUMBO_THRESH_MASK) |
2479 		    ((RX_JUMBO_LKAH_DEFAULT << RX_JUMBO_LKAH_SHIFT) &
2480 		    RX_JUMBO_LKAH_MASK));
2481 		reg = CSR_READ_4(sc, ALE_SRAM_RX_FIFO_LEN);
2482 		rxf_hi = (reg * 7) / 10;
2483 		rxf_lo = (reg * 3)/ 10;
2484 		CSR_WRITE_4(sc, ALE_RX_FIFO_PAUSE_THRESH,
2485 		    ((rxf_lo << RX_FIFO_PAUSE_THRESH_LO_SHIFT) &
2486 		    RX_FIFO_PAUSE_THRESH_LO_MASK) |
2487 		    ((rxf_hi << RX_FIFO_PAUSE_THRESH_HI_SHIFT) &
2488 		     RX_FIFO_PAUSE_THRESH_HI_MASK));
2489 	}
2490 
2491 	/* Disable RSS. */
2492 	CSR_WRITE_4(sc, ALE_RSS_IDT_TABLE0, 0);
2493 	CSR_WRITE_4(sc, ALE_RSS_CPU, 0);
2494 
2495 	/* Configure RxQ. */
2496 	CSR_WRITE_4(sc, ALE_RXQ_CFG,
2497 	    RXQ_CFG_ALIGN_32 | RXQ_CFG_CUT_THROUGH_ENB | RXQ_CFG_ENB);
2498 
2499 	/* Configure DMA parameters. */
2500 	reg = 0;
2501 	if ((sc->ale_flags & ALE_FLAG_TXCMB_BUG) == 0)
2502 		reg |= DMA_CFG_TXCMB_ENB;
2503 	CSR_WRITE_4(sc, ALE_DMA_CFG,
2504 	    DMA_CFG_OUT_ORDER | DMA_CFG_RD_REQ_PRI | DMA_CFG_RCB_64 |
2505 	    sc->ale_dma_rd_burst | reg |
2506 	    sc->ale_dma_wr_burst | DMA_CFG_RXCMB_ENB |
2507 	    ((DMA_CFG_RD_DELAY_CNT_DEFAULT << DMA_CFG_RD_DELAY_CNT_SHIFT) &
2508 	    DMA_CFG_RD_DELAY_CNT_MASK) |
2509 	    ((DMA_CFG_WR_DELAY_CNT_DEFAULT << DMA_CFG_WR_DELAY_CNT_SHIFT) &
2510 	    DMA_CFG_WR_DELAY_CNT_MASK));
2511 
2512 	/*
2513 	 * Hardware can be configured to issue SMB interrupt based
2514 	 * on programmed interval. Since there is a callout that is
2515 	 * invoked for every hz in driver we use that instead of
2516 	 * relying on periodic SMB interrupt.
2517 	 */
2518 	CSR_WRITE_4(sc, ALE_SMB_STAT_TIMER, ALE_USECS(0));
2519 
2520 	/* Clear MAC statistics. */
2521 	ale_stats_clear(sc);
2522 
2523 	/*
2524 	 * Configure Tx/Rx MACs.
2525 	 *  - Auto-padding for short frames.
2526 	 *  - Enable CRC generation.
2527 	 *  Actual reconfiguration of MAC for resolved speed/duplex
2528 	 *  is followed after detection of link establishment.
2529 	 *  AR81xx always does checksum computation regardless of
2530 	 *  MAC_CFG_RXCSUM_ENB bit. In fact, setting the bit will
2531 	 *  cause Rx handling issue for fragmented IP datagrams due
2532 	 *  to silicon bug.
2533 	 */
2534 	reg = MAC_CFG_TX_CRC_ENB | MAC_CFG_TX_AUTO_PAD | MAC_CFG_FULL_DUPLEX |
2535 	    ((MAC_CFG_PREAMBLE_DEFAULT << MAC_CFG_PREAMBLE_SHIFT) &
2536 	    MAC_CFG_PREAMBLE_MASK);
2537 	if ((sc->ale_flags & ALE_FLAG_FASTETHER) != 0)
2538 		reg |= MAC_CFG_SPEED_10_100;
2539 	else
2540 		reg |= MAC_CFG_SPEED_1000;
2541 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2542 
2543 	/* Set up the receive filter. */
2544 	ale_rxfilter(sc);
2545 	ale_rxvlan(sc);
2546 
2547 	/* Acknowledge all pending interrupts and clear it. */
2548 	CSR_WRITE_4(sc, ALE_INTR_MASK, ALE_INTRS);
2549 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0xFFFFFFFF);
2550 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0);
2551 
2552 	sc->ale_flags &= ~ALE_FLAG_LINK;
2553 
2554 	/* Switch to the current media. */
2555 	mii_mediachg(mii);
2556 
2557 	callout_reset(&sc->ale_tick_ch, hz, ale_tick, sc);
2558 
2559 	ifp->if_flags |= IFF_RUNNING;
2560 	ifq_clr_oactive(&ifp->if_snd);
2561 }
2562 
2563 static void
2564 ale_stop(struct ale_softc *sc)
2565 {
2566 	struct ifnet *ifp = &sc->arpcom.ac_if;
2567 	struct ale_txdesc *txd;
2568 	uint32_t reg;
2569 	int i;
2570 
2571 	ASSERT_SERIALIZED(ifp->if_serializer);
2572 
2573 	/*
2574 	 * Mark the interface down and cancel the watchdog timer.
2575 	 */
2576 	ifp->if_flags &= ~IFF_RUNNING;
2577 	ifq_clr_oactive(&ifp->if_snd);
2578 	ifp->if_timer = 0;
2579 
2580 	callout_stop(&sc->ale_tick_ch);
2581 	sc->ale_flags &= ~ALE_FLAG_LINK;
2582 
2583 	ale_stats_update(sc);
2584 
2585 	/* Disable interrupts. */
2586 	CSR_WRITE_4(sc, ALE_INTR_MASK, 0);
2587 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0xFFFFFFFF);
2588 
2589 	/* Disable queue processing and DMA. */
2590 	reg = CSR_READ_4(sc, ALE_TXQ_CFG);
2591 	reg &= ~TXQ_CFG_ENB;
2592 	CSR_WRITE_4(sc, ALE_TXQ_CFG, reg);
2593 	reg = CSR_READ_4(sc, ALE_RXQ_CFG);
2594 	reg &= ~RXQ_CFG_ENB;
2595 	CSR_WRITE_4(sc, ALE_RXQ_CFG, reg);
2596 	reg = CSR_READ_4(sc, ALE_DMA_CFG);
2597 	reg &= ~(DMA_CFG_TXCMB_ENB | DMA_CFG_RXCMB_ENB);
2598 	CSR_WRITE_4(sc, ALE_DMA_CFG, reg);
2599 	DELAY(1000);
2600 
2601 	/* Stop Rx/Tx MACs. */
2602 	ale_stop_mac(sc);
2603 
2604 	/* Disable interrupts again? XXX */
2605 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0xFFFFFFFF);
2606 
2607 	/*
2608 	 * Free TX mbufs still in the queues.
2609 	 */
2610 	for (i = 0; i < ALE_TX_RING_CNT; i++) {
2611 		txd = &sc->ale_cdata.ale_txdesc[i];
2612 		if (txd->tx_m != NULL) {
2613 			bus_dmamap_unload(sc->ale_cdata.ale_tx_tag,
2614 			    txd->tx_dmamap);
2615 			m_freem(txd->tx_m);
2616 			txd->tx_m = NULL;
2617 		}
2618         }
2619 }
2620 
2621 static void
2622 ale_stop_mac(struct ale_softc *sc)
2623 {
2624 	uint32_t reg;
2625 	int i;
2626 
2627 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
2628 	if ((reg & (MAC_CFG_TX_ENB | MAC_CFG_RX_ENB)) != 0) {
2629 		reg &= ~MAC_CFG_TX_ENB | MAC_CFG_RX_ENB;
2630 		CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2631 	}
2632 
2633 	for (i = ALE_TIMEOUT; i > 0; i--) {
2634 		reg = CSR_READ_4(sc, ALE_IDLE_STATUS);
2635 		if (reg == 0)
2636 			break;
2637 		DELAY(10);
2638 	}
2639 	if (i == 0)
2640 		device_printf(sc->ale_dev,
2641 		    "could not disable Tx/Rx MAC(0x%08x)!\n", reg);
2642 }
2643 
2644 static void
2645 ale_init_tx_ring(struct ale_softc *sc)
2646 {
2647 	struct ale_txdesc *txd;
2648 	int i;
2649 
2650 	sc->ale_cdata.ale_tx_prod = 0;
2651 	sc->ale_cdata.ale_tx_cons = 0;
2652 	sc->ale_cdata.ale_tx_cnt = 0;
2653 
2654 	bzero(sc->ale_cdata.ale_tx_ring, ALE_TX_RING_SZ);
2655 	bzero(sc->ale_cdata.ale_tx_cmb, ALE_TX_CMB_SZ);
2656 	for (i = 0; i < ALE_TX_RING_CNT; i++) {
2657 		txd = &sc->ale_cdata.ale_txdesc[i];
2658 		txd->tx_m = NULL;
2659 	}
2660 	*sc->ale_cdata.ale_tx_cmb = 0;
2661 	bus_dmamap_sync(sc->ale_cdata.ale_tx_cmb_tag,
2662 	    sc->ale_cdata.ale_tx_cmb_map,
2663 	    BUS_DMASYNC_PREWRITE);
2664 	bus_dmamap_sync(sc->ale_cdata.ale_tx_ring_tag,
2665 	    sc->ale_cdata.ale_tx_ring_map,
2666 	    BUS_DMASYNC_PREWRITE);
2667 }
2668 
2669 static void
2670 ale_init_rx_pages(struct ale_softc *sc)
2671 {
2672 	struct ale_rx_page *rx_page;
2673 	int i;
2674 
2675 	sc->ale_cdata.ale_rx_seqno = 0;
2676 	sc->ale_cdata.ale_rx_curp = 0;
2677 
2678 	for (i = 0; i < ALE_RX_PAGES; i++) {
2679 		rx_page = &sc->ale_cdata.ale_rx_page[i];
2680 		bzero(rx_page->page_addr, sc->ale_pagesize);
2681 		bzero(rx_page->cmb_addr, ALE_RX_CMB_SZ);
2682 		rx_page->cons = 0;
2683 		*rx_page->cmb_addr = 0;
2684 		bus_dmamap_sync(rx_page->page_tag, rx_page->page_map,
2685 				BUS_DMASYNC_PREWRITE);
2686 		bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2687 				BUS_DMASYNC_PREWRITE);
2688 	}
2689 }
2690 
2691 static void
2692 ale_rxvlan(struct ale_softc *sc)
2693 {
2694 	struct ifnet *ifp;
2695 	uint32_t reg;
2696 
2697 	ifp = &sc->arpcom.ac_if;
2698 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
2699 	reg &= ~MAC_CFG_VLAN_TAG_STRIP;
2700 	if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
2701 		reg |= MAC_CFG_VLAN_TAG_STRIP;
2702 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2703 }
2704 
2705 static void
2706 ale_rxfilter(struct ale_softc *sc)
2707 {
2708 	struct ifnet *ifp;
2709 	struct ifmultiaddr *ifma;
2710 	uint32_t crc;
2711 	uint32_t mchash[2];
2712 	uint32_t rxcfg;
2713 
2714 	ifp = &sc->arpcom.ac_if;
2715 
2716 	rxcfg = CSR_READ_4(sc, ALE_MAC_CFG);
2717 	rxcfg &= ~(MAC_CFG_ALLMULTI | MAC_CFG_BCAST | MAC_CFG_PROMISC);
2718 	if ((ifp->if_flags & IFF_BROADCAST) != 0)
2719 		rxcfg |= MAC_CFG_BCAST;
2720 	if ((ifp->if_flags & (IFF_PROMISC | IFF_ALLMULTI)) != 0) {
2721 		if ((ifp->if_flags & IFF_PROMISC) != 0)
2722 			rxcfg |= MAC_CFG_PROMISC;
2723 		if ((ifp->if_flags & IFF_ALLMULTI) != 0)
2724 			rxcfg |= MAC_CFG_ALLMULTI;
2725 		CSR_WRITE_4(sc, ALE_MAR0, 0xFFFFFFFF);
2726 		CSR_WRITE_4(sc, ALE_MAR1, 0xFFFFFFFF);
2727 		CSR_WRITE_4(sc, ALE_MAC_CFG, rxcfg);
2728 		return;
2729 	}
2730 
2731 	/* Program new filter. */
2732 	bzero(mchash, sizeof(mchash));
2733 
2734 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2735 		if (ifma->ifma_addr->sa_family != AF_LINK)
2736 			continue;
2737 		crc = ether_crc32_le(LLADDR((struct sockaddr_dl *)
2738 		    ifma->ifma_addr), ETHER_ADDR_LEN);
2739 		mchash[crc >> 31] |= 1 << ((crc >> 26) & 0x1f);
2740 	}
2741 
2742 	CSR_WRITE_4(sc, ALE_MAR0, mchash[0]);
2743 	CSR_WRITE_4(sc, ALE_MAR1, mchash[1]);
2744 	CSR_WRITE_4(sc, ALE_MAC_CFG, rxcfg);
2745 }
2746 
2747 static int
2748 sysctl_hw_ale_int_mod(SYSCTL_HANDLER_ARGS)
2749 {
2750 	return (sysctl_int_range(oidp, arg1, arg2, req,
2751 	    ALE_IM_TIMER_MIN, ALE_IM_TIMER_MAX));
2752 }
2753 
2754 static void
2755 ale_dmamap_buf_cb(void *xctx, bus_dma_segment_t *segs, int nsegs,
2756 		  bus_size_t mapsz __unused, int error)
2757 {
2758 	struct ale_dmamap_ctx *ctx = xctx;
2759 	int i;
2760 
2761 	if (error)
2762 		return;
2763 
2764 	if (nsegs > ctx->nsegs) {
2765 		ctx->nsegs = 0;
2766 		return;
2767 	}
2768 
2769 	ctx->nsegs = nsegs;
2770 	for (i = 0; i < nsegs; ++i)
2771 		ctx->segs[i] = segs[i];
2772 }
2773