xref: /dflybsd-src/sys/dev/netif/ale/if_ale.c (revision 858d36ec0da1c070488660b7c0121e6f70100ae9)
1 /*-
2  * Copyright (c) 2008, Pyun YongHyeon <yongari@FreeBSD.org>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice unmodified, this list of conditions, and the following
10  *    disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  *
27  * $FreeBSD: src/sys/dev/ale/if_ale.c,v 1.3 2008/12/03 09:01:12 yongari Exp $
28  */
29 
30 /* Driver for Atheros AR8121/AR8113/AR8114 PCIe Ethernet. */
31 
32 #include <sys/param.h>
33 #include <sys/endian.h>
34 #include <sys/kernel.h>
35 #include <sys/bus.h>
36 #include <sys/interrupt.h>
37 #include <sys/malloc.h>
38 #include <sys/proc.h>
39 #include <sys/rman.h>
40 #include <sys/serialize.h>
41 #include <sys/socket.h>
42 #include <sys/sockio.h>
43 #include <sys/sysctl.h>
44 
45 #include <net/ethernet.h>
46 #include <net/if.h>
47 #include <net/bpf.h>
48 #include <net/if_arp.h>
49 #include <net/if_dl.h>
50 #include <net/if_llc.h>
51 #include <net/if_media.h>
52 #include <net/ifq_var.h>
53 #include <net/vlan/if_vlan_var.h>
54 #include <net/vlan/if_vlan_ether.h>
55 
56 #include <netinet/ip.h>
57 
58 #include <dev/netif/mii_layer/mii.h>
59 #include <dev/netif/mii_layer/miivar.h>
60 
61 #include <bus/pci/pcireg.h>
62 #include <bus/pci/pcivar.h>
63 #include <bus/pci/pcidevs.h>
64 
65 #include <dev/netif/ale/if_alereg.h>
66 #include <dev/netif/ale/if_alevar.h>
67 
68 /* "device miibus" required.  See GENERIC if you get errors here. */
69 #include "miibus_if.h"
70 
71 /* For more information about Tx checksum offload issues see ale_encap(). */
72 #define	ALE_CSUM_FEATURES	(CSUM_TCP | CSUM_UDP)
73 
74 struct ale_dmamap_ctx {
75 	int			nsegs;
76 	bus_dma_segment_t	*segs;
77 };
78 
79 static int	ale_probe(device_t);
80 static int	ale_attach(device_t);
81 static int	ale_detach(device_t);
82 static int	ale_shutdown(device_t);
83 static int	ale_suspend(device_t);
84 static int	ale_resume(device_t);
85 
86 static int	ale_miibus_readreg(device_t, int, int);
87 static int	ale_miibus_writereg(device_t, int, int, int);
88 static void	ale_miibus_statchg(device_t);
89 
90 static void	ale_init(void *);
91 static void	ale_start(struct ifnet *, 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 	error = bus_setup_intr(dev, sc->ale_irq_res, INTR_MPSAFE, ale_intr, sc,
636 			       &sc->ale_irq_handle, ifp->if_serializer);
637 	if (error) {
638 		device_printf(dev, "could not set up interrupt handler.\n");
639 		ether_ifdetach(ifp);
640 		goto fail;
641 	}
642 
643 	ifq_set_cpuid(&ifp->if_snd, rman_get_cpuid(sc->ale_irq_res));
644 	return 0;
645 fail:
646 	ale_detach(dev);
647 	return (error);
648 }
649 
650 static int
651 ale_detach(device_t dev)
652 {
653 	struct ale_softc *sc = device_get_softc(dev);
654 
655 	if (device_is_attached(dev)) {
656 		struct ifnet *ifp = &sc->arpcom.ac_if;
657 
658 		lwkt_serialize_enter(ifp->if_serializer);
659 		sc->ale_flags |= ALE_FLAG_DETACH;
660 		ale_stop(sc);
661 		bus_teardown_intr(dev, sc->ale_irq_res, sc->ale_irq_handle);
662 		lwkt_serialize_exit(ifp->if_serializer);
663 
664 		ether_ifdetach(ifp);
665 	}
666 
667 	if (sc->ale_sysctl_tree != NULL)
668 		sysctl_ctx_free(&sc->ale_sysctl_ctx);
669 
670 	if (sc->ale_miibus != NULL)
671 		device_delete_child(dev, sc->ale_miibus);
672 	bus_generic_detach(dev);
673 
674 	if (sc->ale_irq_res != NULL) {
675 		bus_release_resource(dev, SYS_RES_IRQ, sc->ale_irq_rid,
676 				     sc->ale_irq_res);
677 	}
678 	if (sc->ale_mem_res != NULL) {
679 		bus_release_resource(dev, SYS_RES_MEMORY, sc->ale_mem_rid,
680 				     sc->ale_mem_res);
681 	}
682 
683 	ale_dma_free(sc);
684 
685 	return (0);
686 }
687 
688 #define	ALE_SYSCTL_STAT_ADD32(c, h, n, p, d)	\
689 	    SYSCTL_ADD_UINT(c, h, OID_AUTO, n, CTLFLAG_RD, p, 0, d)
690 #define	ALE_SYSCTL_STAT_ADD64(c, h, n, p, d)	\
691 	    SYSCTL_ADD_QUAD(c, h, OID_AUTO, n, CTLFLAG_RD, p, 0, d)
692 
693 static void
694 ale_sysctl_node(struct ale_softc *sc)
695 {
696 	struct sysctl_ctx_list *ctx;
697 	struct sysctl_oid_list *child, *parent;
698 	struct sysctl_oid *tree;
699 	struct ale_hw_stats *stats;
700 	int error;
701 
702 	sysctl_ctx_init(&sc->ale_sysctl_ctx);
703 	sc->ale_sysctl_tree = SYSCTL_ADD_NODE(&sc->ale_sysctl_ctx,
704 				SYSCTL_STATIC_CHILDREN(_hw), OID_AUTO,
705 				device_get_nameunit(sc->ale_dev),
706 				CTLFLAG_RD, 0, "");
707 	if (sc->ale_sysctl_tree == NULL) {
708 		device_printf(sc->ale_dev, "can't add sysctl node\n");
709 		return;
710 	}
711 
712 	stats = &sc->ale_stats;
713 	ctx = &sc->ale_sysctl_ctx;
714 	child = SYSCTL_CHILDREN(sc->ale_sysctl_tree);
715 
716 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "int_rx_mod",
717 	    CTLTYPE_INT | CTLFLAG_RW, &sc->ale_int_rx_mod, 0,
718 	    sysctl_hw_ale_int_mod, "I", "ale Rx interrupt moderation");
719 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "int_tx_mod",
720 	    CTLTYPE_INT | CTLFLAG_RW, &sc->ale_int_tx_mod, 0,
721 	    sysctl_hw_ale_int_mod, "I", "ale Tx interrupt moderation");
722 
723 	/*
724 	 * Pull in device tunables.
725 	 */
726 	sc->ale_int_rx_mod = ALE_IM_RX_TIMER_DEFAULT;
727 	error = resource_int_value(device_get_name(sc->ale_dev),
728 	    device_get_unit(sc->ale_dev), "int_rx_mod", &sc->ale_int_rx_mod);
729 	if (error == 0) {
730 		if (sc->ale_int_rx_mod < ALE_IM_TIMER_MIN ||
731 		    sc->ale_int_rx_mod > ALE_IM_TIMER_MAX) {
732 			device_printf(sc->ale_dev, "int_rx_mod value out of "
733 			    "range; using default: %d\n",
734 			    ALE_IM_RX_TIMER_DEFAULT);
735 			sc->ale_int_rx_mod = ALE_IM_RX_TIMER_DEFAULT;
736 		}
737 	}
738 
739 	sc->ale_int_tx_mod = ALE_IM_TX_TIMER_DEFAULT;
740 	error = resource_int_value(device_get_name(sc->ale_dev),
741 	    device_get_unit(sc->ale_dev), "int_tx_mod", &sc->ale_int_tx_mod);
742 	if (error == 0) {
743 		if (sc->ale_int_tx_mod < ALE_IM_TIMER_MIN ||
744 		    sc->ale_int_tx_mod > ALE_IM_TIMER_MAX) {
745 			device_printf(sc->ale_dev, "int_tx_mod value out of "
746 			    "range; using default: %d\n",
747 			    ALE_IM_TX_TIMER_DEFAULT);
748 			sc->ale_int_tx_mod = ALE_IM_TX_TIMER_DEFAULT;
749 		}
750 	}
751 
752 	/* Misc statistics. */
753 	ALE_SYSCTL_STAT_ADD32(ctx, child, "reset_brk_seq",
754 	    &stats->reset_brk_seq,
755 	    "Controller resets due to broken Rx sequnce number");
756 
757 	tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "stats", CTLFLAG_RD,
758 	    NULL, "ATE statistics");
759 	parent = SYSCTL_CHILDREN(tree);
760 
761 	/* Rx statistics. */
762 	tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "rx", CTLFLAG_RD,
763 	    NULL, "Rx MAC statistics");
764 	child = SYSCTL_CHILDREN(tree);
765 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_frames",
766 	    &stats->rx_frames, "Good frames");
767 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_bcast_frames",
768 	    &stats->rx_bcast_frames, "Good broadcast frames");
769 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_mcast_frames",
770 	    &stats->rx_mcast_frames, "Good multicast frames");
771 	ALE_SYSCTL_STAT_ADD32(ctx, child, "pause_frames",
772 	    &stats->rx_pause_frames, "Pause control frames");
773 	ALE_SYSCTL_STAT_ADD32(ctx, child, "control_frames",
774 	    &stats->rx_control_frames, "Control frames");
775 	ALE_SYSCTL_STAT_ADD32(ctx, child, "crc_errs",
776 	    &stats->rx_crcerrs, "CRC errors");
777 	ALE_SYSCTL_STAT_ADD32(ctx, child, "len_errs",
778 	    &stats->rx_lenerrs, "Frames with length mismatched");
779 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_octets",
780 	    &stats->rx_bytes, "Good octets");
781 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_bcast_octets",
782 	    &stats->rx_bcast_bytes, "Good broadcast octets");
783 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_mcast_octets",
784 	    &stats->rx_mcast_bytes, "Good multicast octets");
785 	ALE_SYSCTL_STAT_ADD32(ctx, child, "runts",
786 	    &stats->rx_runts, "Too short frames");
787 	ALE_SYSCTL_STAT_ADD32(ctx, child, "fragments",
788 	    &stats->rx_fragments, "Fragmented frames");
789 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_64",
790 	    &stats->rx_pkts_64, "64 bytes frames");
791 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_65_127",
792 	    &stats->rx_pkts_65_127, "65 to 127 bytes frames");
793 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_128_255",
794 	    &stats->rx_pkts_128_255, "128 to 255 bytes frames");
795 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_256_511",
796 	    &stats->rx_pkts_256_511, "256 to 511 bytes frames");
797 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_512_1023",
798 	    &stats->rx_pkts_512_1023, "512 to 1023 bytes frames");
799 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1024_1518",
800 	    &stats->rx_pkts_1024_1518, "1024 to 1518 bytes frames");
801 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1519_max",
802 	    &stats->rx_pkts_1519_max, "1519 to max frames");
803 	ALE_SYSCTL_STAT_ADD32(ctx, child, "trunc_errs",
804 	    &stats->rx_pkts_truncated, "Truncated frames due to MTU size");
805 	ALE_SYSCTL_STAT_ADD32(ctx, child, "fifo_oflows",
806 	    &stats->rx_fifo_oflows, "FIFO overflows");
807 	ALE_SYSCTL_STAT_ADD32(ctx, child, "rrs_errs",
808 	    &stats->rx_rrs_errs, "Return status write-back errors");
809 	ALE_SYSCTL_STAT_ADD32(ctx, child, "align_errs",
810 	    &stats->rx_alignerrs, "Alignment errors");
811 	ALE_SYSCTL_STAT_ADD32(ctx, child, "filtered",
812 	    &stats->rx_pkts_filtered,
813 	    "Frames dropped due to address filtering");
814 
815 	/* Tx statistics. */
816 	tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "tx", CTLFLAG_RD,
817 	    NULL, "Tx MAC statistics");
818 	child = SYSCTL_CHILDREN(tree);
819 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_frames",
820 	    &stats->tx_frames, "Good frames");
821 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_bcast_frames",
822 	    &stats->tx_bcast_frames, "Good broadcast frames");
823 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_mcast_frames",
824 	    &stats->tx_mcast_frames, "Good multicast frames");
825 	ALE_SYSCTL_STAT_ADD32(ctx, child, "pause_frames",
826 	    &stats->tx_pause_frames, "Pause control frames");
827 	ALE_SYSCTL_STAT_ADD32(ctx, child, "control_frames",
828 	    &stats->tx_control_frames, "Control frames");
829 	ALE_SYSCTL_STAT_ADD32(ctx, child, "excess_defers",
830 	    &stats->tx_excess_defer, "Frames with excessive derferrals");
831 	ALE_SYSCTL_STAT_ADD32(ctx, child, "defers",
832 	    &stats->tx_excess_defer, "Frames with derferrals");
833 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_octets",
834 	    &stats->tx_bytes, "Good octets");
835 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_bcast_octets",
836 	    &stats->tx_bcast_bytes, "Good broadcast octets");
837 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_mcast_octets",
838 	    &stats->tx_mcast_bytes, "Good multicast octets");
839 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_64",
840 	    &stats->tx_pkts_64, "64 bytes frames");
841 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_65_127",
842 	    &stats->tx_pkts_65_127, "65 to 127 bytes frames");
843 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_128_255",
844 	    &stats->tx_pkts_128_255, "128 to 255 bytes frames");
845 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_256_511",
846 	    &stats->tx_pkts_256_511, "256 to 511 bytes frames");
847 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_512_1023",
848 	    &stats->tx_pkts_512_1023, "512 to 1023 bytes frames");
849 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1024_1518",
850 	    &stats->tx_pkts_1024_1518, "1024 to 1518 bytes frames");
851 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1519_max",
852 	    &stats->tx_pkts_1519_max, "1519 to max frames");
853 	ALE_SYSCTL_STAT_ADD32(ctx, child, "single_colls",
854 	    &stats->tx_single_colls, "Single collisions");
855 	ALE_SYSCTL_STAT_ADD32(ctx, child, "multi_colls",
856 	    &stats->tx_multi_colls, "Multiple collisions");
857 	ALE_SYSCTL_STAT_ADD32(ctx, child, "late_colls",
858 	    &stats->tx_late_colls, "Late collisions");
859 	ALE_SYSCTL_STAT_ADD32(ctx, child, "excess_colls",
860 	    &stats->tx_excess_colls, "Excessive collisions");
861 	ALE_SYSCTL_STAT_ADD32(ctx, child, "abort",
862 	    &stats->tx_abort, "Aborted frames due to Excessive collisions");
863 	ALE_SYSCTL_STAT_ADD32(ctx, child, "underruns",
864 	    &stats->tx_underrun, "FIFO underruns");
865 	ALE_SYSCTL_STAT_ADD32(ctx, child, "desc_underruns",
866 	    &stats->tx_desc_underrun, "Descriptor write-back errors");
867 	ALE_SYSCTL_STAT_ADD32(ctx, child, "len_errs",
868 	    &stats->tx_lenerrs, "Frames with length mismatched");
869 	ALE_SYSCTL_STAT_ADD32(ctx, child, "trunc_errs",
870 	    &stats->tx_pkts_truncated, "Truncated frames due to MTU size");
871 }
872 
873 #undef ALE_SYSCTL_STAT_ADD32
874 #undef ALE_SYSCTL_STAT_ADD64
875 
876 struct ale_dmamap_arg {
877 	bus_addr_t	ale_busaddr;
878 };
879 
880 static void
881 ale_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error)
882 {
883 	struct ale_dmamap_arg *ctx;
884 
885 	if (error != 0)
886 		return;
887 
888 	KASSERT(nsegs == 1, ("%s: %d segments returned!", __func__, nsegs));
889 
890 	ctx = (struct ale_dmamap_arg *)arg;
891 	ctx->ale_busaddr = segs[0].ds_addr;
892 }
893 
894 /*
895  * Tx descriptors/RXF0/CMB DMA blocks share ALE_DESC_ADDR_HI register
896  * which specifies high address region of DMA blocks. Therefore these
897  * blocks should have the same high address of given 4GB address
898  * space(i.e. crossing 4GB boundary is not allowed).
899  */
900 static int
901 ale_check_boundary(struct ale_softc *sc)
902 {
903 	bus_addr_t rx_cmb_end[ALE_RX_PAGES], tx_cmb_end;
904 	bus_addr_t rx_page_end[ALE_RX_PAGES], tx_ring_end;
905 
906 	rx_page_end[0] = sc->ale_cdata.ale_rx_page[0].page_paddr +
907 	    sc->ale_pagesize;
908 	rx_page_end[1] = sc->ale_cdata.ale_rx_page[1].page_paddr +
909 	    sc->ale_pagesize;
910 	tx_ring_end = sc->ale_cdata.ale_tx_ring_paddr + ALE_TX_RING_SZ;
911 	tx_cmb_end = sc->ale_cdata.ale_tx_cmb_paddr + ALE_TX_CMB_SZ;
912 	rx_cmb_end[0] = sc->ale_cdata.ale_rx_page[0].cmb_paddr + ALE_RX_CMB_SZ;
913 	rx_cmb_end[1] = sc->ale_cdata.ale_rx_page[1].cmb_paddr + ALE_RX_CMB_SZ;
914 
915 	if ((ALE_ADDR_HI(tx_ring_end) !=
916 	    ALE_ADDR_HI(sc->ale_cdata.ale_tx_ring_paddr)) ||
917 	    (ALE_ADDR_HI(rx_page_end[0]) !=
918 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[0].page_paddr)) ||
919 	    (ALE_ADDR_HI(rx_page_end[1]) !=
920 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[1].page_paddr)) ||
921 	    (ALE_ADDR_HI(tx_cmb_end) !=
922 	    ALE_ADDR_HI(sc->ale_cdata.ale_tx_cmb_paddr)) ||
923 	    (ALE_ADDR_HI(rx_cmb_end[0]) !=
924 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[0].cmb_paddr)) ||
925 	    (ALE_ADDR_HI(rx_cmb_end[1]) !=
926 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[1].cmb_paddr)))
927 		return (EFBIG);
928 
929 	if ((ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_page_end[0])) ||
930 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_page_end[1])) ||
931 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_cmb_end[0])) ||
932 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_cmb_end[1])) ||
933 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(tx_cmb_end)))
934 		return (EFBIG);
935 
936 	return (0);
937 }
938 
939 static int
940 ale_dma_alloc(struct ale_softc *sc)
941 {
942 	struct ale_txdesc *txd;
943 	bus_addr_t lowaddr;
944 	struct ale_dmamap_arg ctx;
945 	int error, guard_size, i;
946 
947 	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0)
948 		guard_size = ALE_JUMBO_FRAMELEN;
949 	else
950 		guard_size = ALE_MAX_FRAMELEN;
951 	sc->ale_pagesize = roundup(guard_size + ALE_RX_PAGE_SZ,
952 	    ALE_RX_PAGE_ALIGN);
953 	lowaddr = BUS_SPACE_MAXADDR;
954 again:
955 	/* Create parent DMA tag. */
956 	error = bus_dma_tag_create(
957 	    NULL,			/* parent */
958 	    1, 0,			/* alignment, boundary */
959 	    lowaddr,			/* lowaddr */
960 	    BUS_SPACE_MAXADDR,		/* highaddr */
961 	    NULL, NULL,			/* filter, filterarg */
962 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsize */
963 	    0,				/* nsegments */
964 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsegsize */
965 	    0,				/* flags */
966 	    &sc->ale_cdata.ale_parent_tag);
967 	if (error != 0) {
968 		device_printf(sc->ale_dev,
969 		    "could not create parent DMA tag.\n");
970 		goto fail;
971 	}
972 
973 	/* Create DMA tag for Tx descriptor ring. */
974 	error = bus_dma_tag_create(
975 	    sc->ale_cdata.ale_parent_tag, /* parent */
976 	    ALE_TX_RING_ALIGN, 0,	/* alignment, boundary */
977 	    BUS_SPACE_MAXADDR,		/* lowaddr */
978 	    BUS_SPACE_MAXADDR,		/* highaddr */
979 	    NULL, NULL,			/* filter, filterarg */
980 	    ALE_TX_RING_SZ,		/* maxsize */
981 	    1,				/* nsegments */
982 	    ALE_TX_RING_SZ,		/* maxsegsize */
983 	    0,				/* flags */
984 	    &sc->ale_cdata.ale_tx_ring_tag);
985 	if (error != 0) {
986 		device_printf(sc->ale_dev,
987 		    "could not create Tx ring DMA tag.\n");
988 		goto fail;
989 	}
990 
991 	/* Create DMA tag for Rx pages. */
992 	for (i = 0; i < ALE_RX_PAGES; i++) {
993 		error = bus_dma_tag_create(
994 		    sc->ale_cdata.ale_parent_tag, /* parent */
995 		    ALE_RX_PAGE_ALIGN, 0,	/* alignment, boundary */
996 		    BUS_SPACE_MAXADDR,		/* lowaddr */
997 		    BUS_SPACE_MAXADDR,		/* highaddr */
998 		    NULL, NULL,			/* filter, filterarg */
999 		    sc->ale_pagesize,		/* maxsize */
1000 		    1,				/* nsegments */
1001 		    sc->ale_pagesize,		/* maxsegsize */
1002 		    0,				/* flags */
1003 		    &sc->ale_cdata.ale_rx_page[i].page_tag);
1004 		if (error != 0) {
1005 			device_printf(sc->ale_dev,
1006 			    "could not create Rx page %d DMA tag.\n", i);
1007 			goto fail;
1008 		}
1009 	}
1010 
1011 	/* Create DMA tag for Tx coalescing message block. */
1012 	error = bus_dma_tag_create(
1013 	    sc->ale_cdata.ale_parent_tag, /* parent */
1014 	    ALE_CMB_ALIGN, 0,		/* alignment, boundary */
1015 	    BUS_SPACE_MAXADDR,		/* lowaddr */
1016 	    BUS_SPACE_MAXADDR,		/* highaddr */
1017 	    NULL, NULL,			/* filter, filterarg */
1018 	    ALE_TX_CMB_SZ,		/* maxsize */
1019 	    1,				/* nsegments */
1020 	    ALE_TX_CMB_SZ,		/* maxsegsize */
1021 	    0,				/* flags */
1022 	    &sc->ale_cdata.ale_tx_cmb_tag);
1023 	if (error != 0) {
1024 		device_printf(sc->ale_dev,
1025 		    "could not create Tx CMB DMA tag.\n");
1026 		goto fail;
1027 	}
1028 
1029 	/* Create DMA tag for Rx coalescing message block. */
1030 	for (i = 0; i < ALE_RX_PAGES; i++) {
1031 		error = bus_dma_tag_create(
1032 		    sc->ale_cdata.ale_parent_tag, /* parent */
1033 		    ALE_CMB_ALIGN, 0,		/* alignment, boundary */
1034 		    BUS_SPACE_MAXADDR,		/* lowaddr */
1035 		    BUS_SPACE_MAXADDR,		/* highaddr */
1036 		    NULL, NULL,			/* filter, filterarg */
1037 		    ALE_RX_CMB_SZ,		/* maxsize */
1038 		    1,				/* nsegments */
1039 		    ALE_RX_CMB_SZ,		/* maxsegsize */
1040 		    0,				/* flags */
1041 		    &sc->ale_cdata.ale_rx_page[i].cmb_tag);
1042 		if (error != 0) {
1043 			device_printf(sc->ale_dev,
1044 			    "could not create Rx page %d CMB DMA tag.\n", i);
1045 			goto fail;
1046 		}
1047 	}
1048 
1049 	/* Allocate DMA'able memory and load the DMA map for Tx ring. */
1050 	error = bus_dmamem_alloc(sc->ale_cdata.ale_tx_ring_tag,
1051 	    (void **)&sc->ale_cdata.ale_tx_ring,
1052 	    BUS_DMA_WAITOK | BUS_DMA_ZERO,
1053 	    &sc->ale_cdata.ale_tx_ring_map);
1054 	if (error != 0) {
1055 		device_printf(sc->ale_dev,
1056 		    "could not allocate DMA'able memory for Tx ring.\n");
1057 		goto fail;
1058 	}
1059 	ctx.ale_busaddr = 0;
1060 	error = bus_dmamap_load(sc->ale_cdata.ale_tx_ring_tag,
1061 	    sc->ale_cdata.ale_tx_ring_map, sc->ale_cdata.ale_tx_ring,
1062 	    ALE_TX_RING_SZ, ale_dmamap_cb, &ctx, 0);
1063 	if (error != 0 || ctx.ale_busaddr == 0) {
1064 		device_printf(sc->ale_dev,
1065 		    "could not load DMA'able memory for Tx ring.\n");
1066 		goto fail;
1067 	}
1068 	sc->ale_cdata.ale_tx_ring_paddr = ctx.ale_busaddr;
1069 
1070 	/* Rx pages. */
1071 	for (i = 0; i < ALE_RX_PAGES; i++) {
1072 		error = bus_dmamem_alloc(sc->ale_cdata.ale_rx_page[i].page_tag,
1073 		    (void **)&sc->ale_cdata.ale_rx_page[i].page_addr,
1074 		    BUS_DMA_WAITOK | BUS_DMA_ZERO,
1075 		    &sc->ale_cdata.ale_rx_page[i].page_map);
1076 		if (error != 0) {
1077 			device_printf(sc->ale_dev,
1078 			    "could not allocate DMA'able memory for "
1079 			    "Rx page %d.\n", i);
1080 			goto fail;
1081 		}
1082 		ctx.ale_busaddr = 0;
1083 		error = bus_dmamap_load(sc->ale_cdata.ale_rx_page[i].page_tag,
1084 		    sc->ale_cdata.ale_rx_page[i].page_map,
1085 		    sc->ale_cdata.ale_rx_page[i].page_addr,
1086 		    sc->ale_pagesize, ale_dmamap_cb, &ctx, 0);
1087 		if (error != 0 || ctx.ale_busaddr == 0) {
1088 			device_printf(sc->ale_dev,
1089 			    "could not load DMA'able memory for "
1090 			    "Rx page %d.\n", i);
1091 			goto fail;
1092 		}
1093 		sc->ale_cdata.ale_rx_page[i].page_paddr = ctx.ale_busaddr;
1094 	}
1095 
1096 	/* Tx CMB. */
1097 	error = bus_dmamem_alloc(sc->ale_cdata.ale_tx_cmb_tag,
1098 	    (void **)&sc->ale_cdata.ale_tx_cmb,
1099 	    BUS_DMA_WAITOK | BUS_DMA_ZERO,
1100 	    &sc->ale_cdata.ale_tx_cmb_map);
1101 	if (error != 0) {
1102 		device_printf(sc->ale_dev,
1103 		    "could not allocate DMA'able memory for Tx CMB.\n");
1104 		goto fail;
1105 	}
1106 	ctx.ale_busaddr = 0;
1107 	error = bus_dmamap_load(sc->ale_cdata.ale_tx_cmb_tag,
1108 	    sc->ale_cdata.ale_tx_cmb_map, sc->ale_cdata.ale_tx_cmb,
1109 	    ALE_TX_CMB_SZ, ale_dmamap_cb, &ctx, 0);
1110 	if (error != 0 || ctx.ale_busaddr == 0) {
1111 		device_printf(sc->ale_dev,
1112 		    "could not load DMA'able memory for Tx CMB.\n");
1113 		goto fail;
1114 	}
1115 	sc->ale_cdata.ale_tx_cmb_paddr = ctx.ale_busaddr;
1116 
1117 	/* Rx CMB. */
1118 	for (i = 0; i < ALE_RX_PAGES; i++) {
1119 		error = bus_dmamem_alloc(sc->ale_cdata.ale_rx_page[i].cmb_tag,
1120 		    (void **)&sc->ale_cdata.ale_rx_page[i].cmb_addr,
1121 		    BUS_DMA_WAITOK | BUS_DMA_ZERO,
1122 		    &sc->ale_cdata.ale_rx_page[i].cmb_map);
1123 		if (error != 0) {
1124 			device_printf(sc->ale_dev, "could not allocate "
1125 			    "DMA'able memory for Rx page %d CMB.\n", i);
1126 			goto fail;
1127 		}
1128 		ctx.ale_busaddr = 0;
1129 		error = bus_dmamap_load(sc->ale_cdata.ale_rx_page[i].cmb_tag,
1130 		    sc->ale_cdata.ale_rx_page[i].cmb_map,
1131 		    sc->ale_cdata.ale_rx_page[i].cmb_addr,
1132 		    ALE_RX_CMB_SZ, ale_dmamap_cb, &ctx, 0);
1133 		if (error != 0 || ctx.ale_busaddr == 0) {
1134 			device_printf(sc->ale_dev, "could not load DMA'able "
1135 			    "memory for Rx page %d CMB.\n", i);
1136 			goto fail;
1137 		}
1138 		sc->ale_cdata.ale_rx_page[i].cmb_paddr = ctx.ale_busaddr;
1139 	}
1140 
1141 	/*
1142 	 * Tx descriptors/RXF0/CMB DMA blocks share the same
1143 	 * high address region of 64bit DMA address space.
1144 	 */
1145 	if (lowaddr != BUS_SPACE_MAXADDR_32BIT &&
1146 	    (error = ale_check_boundary(sc)) != 0) {
1147 		device_printf(sc->ale_dev, "4GB boundary crossed, "
1148 		    "switching to 32bit DMA addressing mode.\n");
1149 		ale_dma_free(sc);
1150 		/*
1151 		 * Limit max allowable DMA address space to 32bit
1152 		 * and try again.
1153 		 */
1154 		lowaddr = BUS_SPACE_MAXADDR_32BIT;
1155 		goto again;
1156 	}
1157 
1158 	/*
1159 	 * Create Tx buffer parent tag.
1160 	 * AR81xx allows 64bit DMA addressing of Tx buffers so it
1161 	 * needs separate parent DMA tag as parent DMA address space
1162 	 * could be restricted to be within 32bit address space by
1163 	 * 4GB boundary crossing.
1164 	 */
1165 	error = bus_dma_tag_create(
1166 	    NULL,			/* parent */
1167 	    1, 0,			/* alignment, boundary */
1168 	    BUS_SPACE_MAXADDR,		/* lowaddr */
1169 	    BUS_SPACE_MAXADDR,		/* highaddr */
1170 	    NULL, NULL,			/* filter, filterarg */
1171 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsize */
1172 	    0,				/* nsegments */
1173 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsegsize */
1174 	    0,				/* flags */
1175 	    &sc->ale_cdata.ale_buffer_tag);
1176 	if (error != 0) {
1177 		device_printf(sc->ale_dev,
1178 		    "could not create parent buffer DMA tag.\n");
1179 		goto fail;
1180 	}
1181 
1182 	/* Create DMA tag for Tx buffers. */
1183 	error = bus_dma_tag_create(
1184 	    sc->ale_cdata.ale_buffer_tag, /* parent */
1185 	    1, 0,			/* alignment, boundary */
1186 	    BUS_SPACE_MAXADDR,		/* lowaddr */
1187 	    BUS_SPACE_MAXADDR,		/* highaddr */
1188 	    NULL, NULL,			/* filter, filterarg */
1189 	    ALE_TSO_MAXSIZE,		/* maxsize */
1190 	    ALE_MAXTXSEGS,		/* nsegments */
1191 	    ALE_TSO_MAXSEGSIZE,		/* maxsegsize */
1192 	    0,				/* flags */
1193 	    &sc->ale_cdata.ale_tx_tag);
1194 	if (error != 0) {
1195 		device_printf(sc->ale_dev, "could not create Tx DMA tag.\n");
1196 		goto fail;
1197 	}
1198 
1199 	/* Create DMA maps for Tx buffers. */
1200 	for (i = 0; i < ALE_TX_RING_CNT; i++) {
1201 		txd = &sc->ale_cdata.ale_txdesc[i];
1202 		txd->tx_m = NULL;
1203 		txd->tx_dmamap = NULL;
1204 		error = bus_dmamap_create(sc->ale_cdata.ale_tx_tag, 0,
1205 		    &txd->tx_dmamap);
1206 		if (error != 0) {
1207 			device_printf(sc->ale_dev,
1208 			    "could not create Tx dmamap.\n");
1209 			goto fail;
1210 		}
1211 	}
1212 fail:
1213 	return (error);
1214 }
1215 
1216 static void
1217 ale_dma_free(struct ale_softc *sc)
1218 {
1219 	struct ale_txdesc *txd;
1220 	int i;
1221 
1222 	/* Tx buffers. */
1223 	if (sc->ale_cdata.ale_tx_tag != NULL) {
1224 		for (i = 0; i < ALE_TX_RING_CNT; i++) {
1225 			txd = &sc->ale_cdata.ale_txdesc[i];
1226 			if (txd->tx_dmamap != NULL) {
1227 				bus_dmamap_destroy(sc->ale_cdata.ale_tx_tag,
1228 				    txd->tx_dmamap);
1229 				txd->tx_dmamap = NULL;
1230 			}
1231 		}
1232 		bus_dma_tag_destroy(sc->ale_cdata.ale_tx_tag);
1233 		sc->ale_cdata.ale_tx_tag = NULL;
1234 	}
1235 	/* Tx descriptor ring. */
1236 	if (sc->ale_cdata.ale_tx_ring_tag != NULL) {
1237 		if (sc->ale_cdata.ale_tx_ring_map != NULL)
1238 			bus_dmamap_unload(sc->ale_cdata.ale_tx_ring_tag,
1239 			    sc->ale_cdata.ale_tx_ring_map);
1240 		if (sc->ale_cdata.ale_tx_ring_map != NULL &&
1241 		    sc->ale_cdata.ale_tx_ring != NULL)
1242 			bus_dmamem_free(sc->ale_cdata.ale_tx_ring_tag,
1243 			    sc->ale_cdata.ale_tx_ring,
1244 			    sc->ale_cdata.ale_tx_ring_map);
1245 		sc->ale_cdata.ale_tx_ring = NULL;
1246 		sc->ale_cdata.ale_tx_ring_map = NULL;
1247 		bus_dma_tag_destroy(sc->ale_cdata.ale_tx_ring_tag);
1248 		sc->ale_cdata.ale_tx_ring_tag = NULL;
1249 	}
1250 	/* Rx page block. */
1251 	for (i = 0; i < ALE_RX_PAGES; i++) {
1252 		if (sc->ale_cdata.ale_rx_page[i].page_tag != NULL) {
1253 			if (sc->ale_cdata.ale_rx_page[i].page_map != NULL)
1254 				bus_dmamap_unload(
1255 				    sc->ale_cdata.ale_rx_page[i].page_tag,
1256 				    sc->ale_cdata.ale_rx_page[i].page_map);
1257 			if (sc->ale_cdata.ale_rx_page[i].page_map != NULL &&
1258 			    sc->ale_cdata.ale_rx_page[i].page_addr != NULL)
1259 				bus_dmamem_free(
1260 				    sc->ale_cdata.ale_rx_page[i].page_tag,
1261 				    sc->ale_cdata.ale_rx_page[i].page_addr,
1262 				    sc->ale_cdata.ale_rx_page[i].page_map);
1263 			sc->ale_cdata.ale_rx_page[i].page_addr = NULL;
1264 			sc->ale_cdata.ale_rx_page[i].page_map = NULL;
1265 			bus_dma_tag_destroy(
1266 			    sc->ale_cdata.ale_rx_page[i].page_tag);
1267 			sc->ale_cdata.ale_rx_page[i].page_tag = NULL;
1268 		}
1269 	}
1270 	/* Rx CMB. */
1271 	for (i = 0; i < ALE_RX_PAGES; i++) {
1272 		if (sc->ale_cdata.ale_rx_page[i].cmb_tag != NULL) {
1273 			if (sc->ale_cdata.ale_rx_page[i].cmb_map != NULL)
1274 				bus_dmamap_unload(
1275 				    sc->ale_cdata.ale_rx_page[i].cmb_tag,
1276 				    sc->ale_cdata.ale_rx_page[i].cmb_map);
1277 			if (sc->ale_cdata.ale_rx_page[i].cmb_map != NULL &&
1278 			    sc->ale_cdata.ale_rx_page[i].cmb_addr != NULL)
1279 				bus_dmamem_free(
1280 				    sc->ale_cdata.ale_rx_page[i].cmb_tag,
1281 				    sc->ale_cdata.ale_rx_page[i].cmb_addr,
1282 				    sc->ale_cdata.ale_rx_page[i].cmb_map);
1283 			sc->ale_cdata.ale_rx_page[i].cmb_addr = NULL;
1284 			sc->ale_cdata.ale_rx_page[i].cmb_map = NULL;
1285 			bus_dma_tag_destroy(
1286 			    sc->ale_cdata.ale_rx_page[i].cmb_tag);
1287 			sc->ale_cdata.ale_rx_page[i].cmb_tag = NULL;
1288 		}
1289 	}
1290 	/* Tx CMB. */
1291 	if (sc->ale_cdata.ale_tx_cmb_tag != NULL) {
1292 		if (sc->ale_cdata.ale_tx_cmb_map != NULL)
1293 			bus_dmamap_unload(sc->ale_cdata.ale_tx_cmb_tag,
1294 			    sc->ale_cdata.ale_tx_cmb_map);
1295 		if (sc->ale_cdata.ale_tx_cmb_map != NULL &&
1296 		    sc->ale_cdata.ale_tx_cmb != NULL)
1297 			bus_dmamem_free(sc->ale_cdata.ale_tx_cmb_tag,
1298 			    sc->ale_cdata.ale_tx_cmb,
1299 			    sc->ale_cdata.ale_tx_cmb_map);
1300 		sc->ale_cdata.ale_tx_cmb = NULL;
1301 		sc->ale_cdata.ale_tx_cmb_map = NULL;
1302 		bus_dma_tag_destroy(sc->ale_cdata.ale_tx_cmb_tag);
1303 		sc->ale_cdata.ale_tx_cmb_tag = NULL;
1304 	}
1305 	if (sc->ale_cdata.ale_buffer_tag != NULL) {
1306 		bus_dma_tag_destroy(sc->ale_cdata.ale_buffer_tag);
1307 		sc->ale_cdata.ale_buffer_tag = NULL;
1308 	}
1309 	if (sc->ale_cdata.ale_parent_tag != NULL) {
1310 		bus_dma_tag_destroy(sc->ale_cdata.ale_parent_tag);
1311 		sc->ale_cdata.ale_parent_tag = NULL;
1312 	}
1313 }
1314 
1315 static int
1316 ale_shutdown(device_t dev)
1317 {
1318 	return (ale_suspend(dev));
1319 }
1320 
1321 #ifdef notyet
1322 
1323 /*
1324  * Note, this driver resets the link speed to 10/100Mbps by
1325  * restarting auto-negotiation in suspend/shutdown phase but we
1326  * don't know whether that auto-negotiation would succeed or not
1327  * as driver has no control after powering off/suspend operation.
1328  * If the renegotiation fail WOL may not work. Running at 1Gbps
1329  * will draw more power than 375mA at 3.3V which is specified in
1330  * PCI specification and that would result in complete
1331  * shutdowning power to ethernet controller.
1332  *
1333  * TODO
1334  * Save current negotiated media speed/duplex/flow-control to
1335  * softc and restore the same link again after resuming. PHY
1336  * handling such as power down/resetting to 100Mbps may be better
1337  * handled in suspend method in phy driver.
1338  */
1339 static void
1340 ale_setlinkspeed(struct ale_softc *sc)
1341 {
1342 	struct mii_data *mii;
1343 	int aneg, i;
1344 
1345 	mii = device_get_softc(sc->ale_miibus);
1346 	mii_pollstat(mii);
1347 	aneg = 0;
1348 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
1349 	    (IFM_ACTIVE | IFM_AVALID)) {
1350 		switch IFM_SUBTYPE(mii->mii_media_active) {
1351 		case IFM_10_T:
1352 		case IFM_100_TX:
1353 			return;
1354 		case IFM_1000_T:
1355 			aneg++;
1356 			break;
1357 		default:
1358 			break;
1359 		}
1360 	}
1361 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr, MII_100T2CR, 0);
1362 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
1363 	    MII_ANAR, ANAR_TX_FD | ANAR_TX | ANAR_10_FD | ANAR_10 | ANAR_CSMA);
1364 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
1365 	    MII_BMCR, BMCR_RESET | BMCR_AUTOEN | BMCR_STARTNEG);
1366 	DELAY(1000);
1367 	if (aneg != 0) {
1368 		/*
1369 		 * Poll link state until ale(4) get a 10/100Mbps link.
1370 		 */
1371 		for (i = 0; i < MII_ANEGTICKS_GIGE; i++) {
1372 			mii_pollstat(mii);
1373 			if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID))
1374 			    == (IFM_ACTIVE | IFM_AVALID)) {
1375 				switch (IFM_SUBTYPE(
1376 				    mii->mii_media_active)) {
1377 				case IFM_10_T:
1378 				case IFM_100_TX:
1379 					ale_mac_config(sc);
1380 					return;
1381 				default:
1382 					break;
1383 				}
1384 			}
1385 			ALE_UNLOCK(sc);
1386 			pause("alelnk", hz);
1387 			ALE_LOCK(sc);
1388 		}
1389 		if (i == MII_ANEGTICKS_GIGE)
1390 			device_printf(sc->ale_dev,
1391 			    "establishing a link failed, WOL may not work!");
1392 	}
1393 	/*
1394 	 * No link, force MAC to have 100Mbps, full-duplex link.
1395 	 * This is the last resort and may/may not work.
1396 	 */
1397 	mii->mii_media_status = IFM_AVALID | IFM_ACTIVE;
1398 	mii->mii_media_active = IFM_ETHER | IFM_100_TX | IFM_FDX;
1399 	ale_mac_config(sc);
1400 }
1401 
1402 static void
1403 ale_setwol(struct ale_softc *sc)
1404 {
1405 	struct ifnet *ifp;
1406 	uint32_t reg, pmcs;
1407 	uint16_t pmstat;
1408 	int pmc;
1409 
1410 	ALE_LOCK_ASSERT(sc);
1411 
1412 	if (pci_find_extcap(sc->ale_dev, PCIY_PMG, &pmc) != 0) {
1413 		/* Disable WOL. */
1414 		CSR_WRITE_4(sc, ALE_WOL_CFG, 0);
1415 		reg = CSR_READ_4(sc, ALE_PCIE_PHYMISC);
1416 		reg |= PCIE_PHYMISC_FORCE_RCV_DET;
1417 		CSR_WRITE_4(sc, ALE_PCIE_PHYMISC, reg);
1418 		/* Force PHY power down. */
1419 		CSR_WRITE_2(sc, ALE_GPHY_CTRL,
1420 		    GPHY_CTRL_EXT_RESET | GPHY_CTRL_HIB_EN |
1421 		    GPHY_CTRL_HIB_PULSE | GPHY_CTRL_PHY_PLL_ON |
1422 		    GPHY_CTRL_SEL_ANA_RESET | GPHY_CTRL_PHY_IDDQ |
1423 		    GPHY_CTRL_PCLK_SEL_DIS | GPHY_CTRL_PWDOWN_HW);
1424 		return;
1425 	}
1426 
1427 	ifp = sc->ale_ifp;
1428 	if ((ifp->if_capenable & IFCAP_WOL) != 0) {
1429 		if ((sc->ale_flags & ALE_FLAG_FASTETHER) == 0)
1430 			ale_setlinkspeed(sc);
1431 	}
1432 
1433 	pmcs = 0;
1434 	if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
1435 		pmcs |= WOL_CFG_MAGIC | WOL_CFG_MAGIC_ENB;
1436 	CSR_WRITE_4(sc, ALE_WOL_CFG, pmcs);
1437 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
1438 	reg &= ~(MAC_CFG_DBG | MAC_CFG_PROMISC | MAC_CFG_ALLMULTI |
1439 	    MAC_CFG_BCAST);
1440 	if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
1441 		reg |= MAC_CFG_ALLMULTI | MAC_CFG_BCAST;
1442 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
1443 		reg |= MAC_CFG_RX_ENB;
1444 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
1445 
1446 	if ((ifp->if_capenable & IFCAP_WOL) == 0) {
1447 		/* WOL disabled, PHY power down. */
1448 		reg = CSR_READ_4(sc, ALE_PCIE_PHYMISC);
1449 		reg |= PCIE_PHYMISC_FORCE_RCV_DET;
1450 		CSR_WRITE_4(sc, ALE_PCIE_PHYMISC, reg);
1451 		CSR_WRITE_2(sc, ALE_GPHY_CTRL,
1452 		    GPHY_CTRL_EXT_RESET | GPHY_CTRL_HIB_EN |
1453 		    GPHY_CTRL_HIB_PULSE | GPHY_CTRL_SEL_ANA_RESET |
1454 		    GPHY_CTRL_PHY_IDDQ | GPHY_CTRL_PCLK_SEL_DIS |
1455 		    GPHY_CTRL_PWDOWN_HW);
1456 	}
1457 	/* Request PME. */
1458 	pmstat = pci_read_config(sc->ale_dev, pmc + PCIR_POWER_STATUS, 2);
1459 	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
1460 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
1461 		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
1462 	pci_write_config(sc->ale_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
1463 }
1464 
1465 #endif	/* notyet */
1466 
1467 static int
1468 ale_suspend(device_t dev)
1469 {
1470 	struct ale_softc *sc = device_get_softc(dev);
1471 	struct ifnet *ifp = &sc->arpcom.ac_if;
1472 
1473 	lwkt_serialize_enter(ifp->if_serializer);
1474 	ale_stop(sc);
1475 #ifdef notyet
1476 	ale_setwol(sc);
1477 #endif
1478 	lwkt_serialize_exit(ifp->if_serializer);
1479 	return (0);
1480 }
1481 
1482 static int
1483 ale_resume(device_t dev)
1484 {
1485 	struct ale_softc *sc = device_get_softc(dev);
1486 	struct ifnet *ifp = &sc->arpcom.ac_if;
1487 	uint16_t cmd;
1488 
1489 	lwkt_serialize_enter(ifp->if_serializer);
1490 
1491 	/*
1492 	 * Clear INTx emulation disable for hardwares that
1493 	 * is set in resume event. From Linux.
1494 	 */
1495 	cmd = pci_read_config(sc->ale_dev, PCIR_COMMAND, 2);
1496 	if ((cmd & 0x0400) != 0) {
1497 		cmd &= ~0x0400;
1498 		pci_write_config(sc->ale_dev, PCIR_COMMAND, cmd, 2);
1499 	}
1500 
1501 #ifdef notyet
1502 	if (pci_find_extcap(sc->ale_dev, PCIY_PMG, &pmc) == 0) {
1503 		uint16_t pmstat;
1504 		int pmc;
1505 
1506 		/* Disable PME and clear PME status. */
1507 		pmstat = pci_read_config(sc->ale_dev,
1508 		    pmc + PCIR_POWER_STATUS, 2);
1509 		if ((pmstat & PCIM_PSTAT_PMEENABLE) != 0) {
1510 			pmstat &= ~PCIM_PSTAT_PMEENABLE;
1511 			pci_write_config(sc->ale_dev,
1512 			    pmc + PCIR_POWER_STATUS, pmstat, 2);
1513 		}
1514 	}
1515 #endif
1516 
1517 	/* Reset PHY. */
1518 	ale_phy_reset(sc);
1519 	if ((ifp->if_flags & IFF_UP) != 0)
1520 		ale_init(sc);
1521 
1522 	lwkt_serialize_exit(ifp->if_serializer);
1523 	return (0);
1524 }
1525 
1526 static int
1527 ale_encap(struct ale_softc *sc, struct mbuf **m_head)
1528 {
1529 	struct ale_txdesc *txd, *txd_last;
1530 	struct tx_desc *desc;
1531 	struct mbuf *m;
1532 	bus_dma_segment_t txsegs[ALE_MAXTXSEGS];
1533 	struct ale_dmamap_ctx ctx;
1534 	bus_dmamap_t map;
1535 	uint32_t cflags, poff, vtag;
1536 	int error, i, nsegs, prod;
1537 
1538 	M_ASSERTPKTHDR((*m_head));
1539 
1540 	m = *m_head;
1541 	cflags = vtag = 0;
1542 	poff = 0;
1543 
1544 	prod = sc->ale_cdata.ale_tx_prod;
1545 	txd = &sc->ale_cdata.ale_txdesc[prod];
1546 	txd_last = txd;
1547 	map = txd->tx_dmamap;
1548 
1549 	ctx.nsegs = ALE_MAXTXSEGS;
1550 	ctx.segs = txsegs;
1551 	error =  bus_dmamap_load_mbuf(sc->ale_cdata.ale_tx_tag, map,
1552 				      *m_head, ale_dmamap_buf_cb, &ctx,
1553 				      BUS_DMA_NOWAIT);
1554 	if (error == EFBIG) {
1555 		m = m_defrag(*m_head, MB_DONTWAIT);
1556 		if (m == NULL) {
1557 			m_freem(*m_head);
1558 			*m_head = NULL;
1559 			return (ENOMEM);
1560 		}
1561 		*m_head = m;
1562 
1563 		ctx.nsegs = ALE_MAXTXSEGS;
1564 		ctx.segs = txsegs;
1565 		error =  bus_dmamap_load_mbuf(sc->ale_cdata.ale_tx_tag, map,
1566 					      *m_head, ale_dmamap_buf_cb, &ctx,
1567 					      BUS_DMA_NOWAIT);
1568 		if (error != 0) {
1569 			m_freem(*m_head);
1570 			*m_head = NULL;
1571 			return (error);
1572 		}
1573 	} else if (error != 0) {
1574 		return (error);
1575 	}
1576 	nsegs = ctx.nsegs;
1577 
1578 	if (nsegs == 0) {
1579 		m_freem(*m_head);
1580 		*m_head = NULL;
1581 		return (EIO);
1582 	}
1583 
1584 	/* Check descriptor overrun. */
1585 	if (sc->ale_cdata.ale_tx_cnt + nsegs >= ALE_TX_RING_CNT - 2) {
1586 		bus_dmamap_unload(sc->ale_cdata.ale_tx_tag, map);
1587 		return (ENOBUFS);
1588 	}
1589 	bus_dmamap_sync(sc->ale_cdata.ale_tx_tag, map, BUS_DMASYNC_PREWRITE);
1590 
1591 	m = *m_head;
1592 	/* Configure Tx checksum offload. */
1593 	if ((m->m_pkthdr.csum_flags & ALE_CSUM_FEATURES) != 0) {
1594 		/*
1595 		 * AR81xx supports Tx custom checksum offload feature
1596 		 * that offloads single 16bit checksum computation.
1597 		 * So you can choose one among IP, TCP and UDP.
1598 		 * Normally driver sets checksum start/insertion
1599 		 * position from the information of TCP/UDP frame as
1600 		 * TCP/UDP checksum takes more time than that of IP.
1601 		 * However it seems that custom checksum offload
1602 		 * requires 4 bytes aligned Tx buffers due to hardware
1603 		 * bug.
1604 		 * AR81xx also supports explicit Tx checksum computation
1605 		 * if it is told that the size of IP header and TCP
1606 		 * header(for UDP, the header size does not matter
1607 		 * because it's fixed length). However with this scheme
1608 		 * TSO does not work so you have to choose one either
1609 		 * TSO or explicit Tx checksum offload. I chosen TSO
1610 		 * plus custom checksum offload with work-around which
1611 		 * will cover most common usage for this consumer
1612 		 * ethernet controller. The work-around takes a lot of
1613 		 * CPU cycles if Tx buffer is not aligned on 4 bytes
1614 		 * boundary, though.
1615 		 */
1616 		cflags |= ALE_TD_CXSUM;
1617 		/* Set checksum start offset. */
1618 		cflags |= (poff << ALE_TD_CSUM_PLOADOFFSET_SHIFT);
1619 		/* Set checksum insertion position of TCP/UDP. */
1620 		cflags |= ((poff + m->m_pkthdr.csum_data) <<
1621 		    ALE_TD_CSUM_XSUMOFFSET_SHIFT);
1622 	}
1623 
1624 	/* Configure VLAN hardware tag insertion. */
1625 	if ((m->m_flags & M_VLANTAG) != 0) {
1626 		vtag = ALE_TX_VLAN_TAG(m->m_pkthdr.ether_vlantag);
1627 		vtag = ((vtag << ALE_TD_VLAN_SHIFT) & ALE_TD_VLAN_MASK);
1628 		cflags |= ALE_TD_INSERT_VLAN_TAG;
1629 	}
1630 
1631 	desc = NULL;
1632 	for (i = 0; i < nsegs; i++) {
1633 		desc = &sc->ale_cdata.ale_tx_ring[prod];
1634 		desc->addr = htole64(txsegs[i].ds_addr);
1635 		desc->len = htole32(ALE_TX_BYTES(txsegs[i].ds_len) | vtag);
1636 		desc->flags = htole32(cflags);
1637 		sc->ale_cdata.ale_tx_cnt++;
1638 		ALE_DESC_INC(prod, ALE_TX_RING_CNT);
1639 	}
1640 	/* Update producer index. */
1641 	sc->ale_cdata.ale_tx_prod = prod;
1642 
1643 	/* Finally set EOP on the last descriptor. */
1644 	prod = (prod + ALE_TX_RING_CNT - 1) % ALE_TX_RING_CNT;
1645 	desc = &sc->ale_cdata.ale_tx_ring[prod];
1646 	desc->flags |= htole32(ALE_TD_EOP);
1647 
1648 	/* Swap dmamap of the first and the last. */
1649 	txd = &sc->ale_cdata.ale_txdesc[prod];
1650 	map = txd_last->tx_dmamap;
1651 	txd_last->tx_dmamap = txd->tx_dmamap;
1652 	txd->tx_dmamap = map;
1653 	txd->tx_m = m;
1654 
1655 	/* Sync descriptors. */
1656 	bus_dmamap_sync(sc->ale_cdata.ale_tx_ring_tag,
1657 	    sc->ale_cdata.ale_tx_ring_map, BUS_DMASYNC_PREWRITE);
1658 
1659 	return (0);
1660 }
1661 
1662 static void
1663 ale_start(struct ifnet *ifp, struct ifaltq_subque *ifsq)
1664 {
1665         struct ale_softc *sc = ifp->if_softc;
1666 	struct mbuf *m_head;
1667 	int enq;
1668 
1669 	ASSERT_ALTQ_SQ_DEFAULT(ifp, ifsq);
1670 	ASSERT_SERIALIZED(ifp->if_serializer);
1671 
1672 	if ((sc->ale_flags & ALE_FLAG_LINK) == 0) {
1673 		ifq_purge(&ifp->if_snd);
1674 		return;
1675 	}
1676 
1677 	if ((ifp->if_flags & IFF_RUNNING) == 0 || ifq_is_oactive(&ifp->if_snd))
1678 		return;
1679 
1680 	/* Reclaim transmitted frames. */
1681 	if (sc->ale_cdata.ale_tx_cnt >= ALE_TX_DESC_HIWAT)
1682 		ale_txeof(sc);
1683 
1684 	enq = 0;
1685 	while (!ifq_is_empty(&ifp->if_snd)) {
1686 		m_head = ifq_dequeue(&ifp->if_snd, NULL);
1687 		if (m_head == NULL)
1688 			break;
1689 
1690 		/*
1691 		 * Pack the data into the transmit ring. If we
1692 		 * don't have room, set the OACTIVE flag and wait
1693 		 * for the NIC to drain the ring.
1694 		 */
1695 		if (ale_encap(sc, &m_head)) {
1696 			if (m_head == NULL)
1697 				break;
1698 			ifq_prepend(&ifp->if_snd, m_head);
1699 			ifq_set_oactive(&ifp->if_snd);
1700 			break;
1701 		}
1702 		enq = 1;
1703 
1704 		/*
1705 		 * If there's a BPF listener, bounce a copy of this frame
1706 		 * to him.
1707 		 */
1708 		ETHER_BPF_MTAP(ifp, m_head);
1709 	}
1710 
1711 	if (enq) {
1712 		/* Kick. */
1713 		CSR_WRITE_4(sc, ALE_MBOX_TPD_PROD_IDX,
1714 		    sc->ale_cdata.ale_tx_prod);
1715 
1716 		/* Set a timeout in case the chip goes out to lunch. */
1717 		ifp->if_timer = ALE_TX_TIMEOUT;
1718 	}
1719 }
1720 
1721 static void
1722 ale_watchdog(struct ifnet *ifp)
1723 {
1724 	struct ale_softc *sc = ifp->if_softc;
1725 
1726 	ASSERT_SERIALIZED(ifp->if_serializer);
1727 
1728 	if ((sc->ale_flags & ALE_FLAG_LINK) == 0) {
1729 		if_printf(ifp, "watchdog timeout (lost link)\n");
1730 		IFNET_STAT_INC(ifp, oerrors, 1);
1731 		ale_init(sc);
1732 		return;
1733 	}
1734 
1735 	if_printf(ifp, "watchdog timeout -- resetting\n");
1736 	IFNET_STAT_INC(ifp, oerrors, 1);
1737 	ale_init(sc);
1738 
1739 	if (!ifq_is_empty(&ifp->if_snd))
1740 		if_devstart(ifp);
1741 }
1742 
1743 static int
1744 ale_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data, struct ucred *cr)
1745 {
1746 	struct ale_softc *sc;
1747 	struct ifreq *ifr;
1748 	struct mii_data *mii;
1749 	int error, mask;
1750 
1751 	ASSERT_SERIALIZED(ifp->if_serializer);
1752 
1753 	sc = ifp->if_softc;
1754 	ifr = (struct ifreq *)data;
1755 	error = 0;
1756 
1757 	switch (cmd) {
1758 	case SIOCSIFMTU:
1759 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > ALE_JUMBO_MTU ||
1760 		    ((sc->ale_flags & ALE_FLAG_JUMBO) == 0 &&
1761 		    ifr->ifr_mtu > ETHERMTU))
1762 			error = EINVAL;
1763 		else if (ifp->if_mtu != ifr->ifr_mtu) {
1764 			ifp->if_mtu = ifr->ifr_mtu;
1765 			if ((ifp->if_flags & IFF_RUNNING) != 0)
1766 				ale_init(sc);
1767 		}
1768 		break;
1769 
1770 	case SIOCSIFFLAGS:
1771 		if ((ifp->if_flags & IFF_UP) != 0) {
1772 			if ((ifp->if_flags & IFF_RUNNING) != 0) {
1773 				if (((ifp->if_flags ^ sc->ale_if_flags)
1774 				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
1775 					ale_rxfilter(sc);
1776 			} else {
1777 				if ((sc->ale_flags & ALE_FLAG_DETACH) == 0)
1778 					ale_init(sc);
1779 			}
1780 		} else {
1781 			if ((ifp->if_flags & IFF_RUNNING) != 0)
1782 				ale_stop(sc);
1783 		}
1784 		sc->ale_if_flags = ifp->if_flags;
1785 		break;
1786 
1787 	case SIOCADDMULTI:
1788 	case SIOCDELMULTI:
1789 		if ((ifp->if_flags & IFF_RUNNING) != 0)
1790 			ale_rxfilter(sc);
1791 		break;
1792 
1793 	case SIOCSIFMEDIA:
1794 	case SIOCGIFMEDIA:
1795 		mii = device_get_softc(sc->ale_miibus);
1796 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, cmd);
1797 		break;
1798 
1799 	case SIOCSIFCAP:
1800 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1801 		if ((mask & IFCAP_TXCSUM) != 0 &&
1802 		    (ifp->if_capabilities & IFCAP_TXCSUM) != 0) {
1803 			ifp->if_capenable ^= IFCAP_TXCSUM;
1804 			if ((ifp->if_capenable & IFCAP_TXCSUM) != 0)
1805 				ifp->if_hwassist |= ALE_CSUM_FEATURES;
1806 			else
1807 				ifp->if_hwassist &= ~ALE_CSUM_FEATURES;
1808 		}
1809 		if ((mask & IFCAP_RXCSUM) != 0 &&
1810 		    (ifp->if_capabilities & IFCAP_RXCSUM) != 0)
1811 			ifp->if_capenable ^= IFCAP_RXCSUM;
1812 
1813 		if ((mask & IFCAP_VLAN_HWTAGGING) != 0 &&
1814 		    (ifp->if_capabilities & IFCAP_VLAN_HWTAGGING) != 0) {
1815 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1816 			ale_rxvlan(sc);
1817 		}
1818 		break;
1819 
1820 	default:
1821 		error = ether_ioctl(ifp, cmd, data);
1822 		break;
1823 	}
1824 	return (error);
1825 }
1826 
1827 static void
1828 ale_mac_config(struct ale_softc *sc)
1829 {
1830 	struct mii_data *mii;
1831 	uint32_t reg;
1832 
1833 	mii = device_get_softc(sc->ale_miibus);
1834 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
1835 	reg &= ~(MAC_CFG_FULL_DUPLEX | MAC_CFG_TX_FC | MAC_CFG_RX_FC |
1836 	    MAC_CFG_SPEED_MASK);
1837 	/* Reprogram MAC with resolved speed/duplex. */
1838 	switch (IFM_SUBTYPE(mii->mii_media_active)) {
1839 	case IFM_10_T:
1840 	case IFM_100_TX:
1841 		reg |= MAC_CFG_SPEED_10_100;
1842 		break;
1843 	case IFM_1000_T:
1844 		reg |= MAC_CFG_SPEED_1000;
1845 		break;
1846 	}
1847 	if ((IFM_OPTIONS(mii->mii_media_active) & IFM_FDX) != 0) {
1848 		reg |= MAC_CFG_FULL_DUPLEX;
1849 #ifdef notyet
1850 		if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_TXPAUSE) != 0)
1851 			reg |= MAC_CFG_TX_FC;
1852 		if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_RXPAUSE) != 0)
1853 			reg |= MAC_CFG_RX_FC;
1854 #endif
1855 	}
1856 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
1857 }
1858 
1859 static void
1860 ale_stats_clear(struct ale_softc *sc)
1861 {
1862 	struct smb sb;
1863 	uint32_t *reg;
1864 	int i;
1865 
1866 	for (reg = &sb.rx_frames, i = 0; reg <= &sb.rx_pkts_filtered; reg++) {
1867 		CSR_READ_4(sc, ALE_RX_MIB_BASE + i);
1868 		i += sizeof(uint32_t);
1869 	}
1870 	/* Read Tx statistics. */
1871 	for (reg = &sb.tx_frames, i = 0; reg <= &sb.tx_mcast_bytes; reg++) {
1872 		CSR_READ_4(sc, ALE_TX_MIB_BASE + i);
1873 		i += sizeof(uint32_t);
1874 	}
1875 }
1876 
1877 static void
1878 ale_stats_update(struct ale_softc *sc)
1879 {
1880 	struct ale_hw_stats *stat;
1881 	struct smb sb, *smb;
1882 	struct ifnet *ifp;
1883 	uint32_t *reg;
1884 	int i;
1885 
1886 	ifp = &sc->arpcom.ac_if;
1887 	stat = &sc->ale_stats;
1888 	smb = &sb;
1889 
1890 	/* Read Rx statistics. */
1891 	for (reg = &sb.rx_frames, i = 0; reg <= &sb.rx_pkts_filtered; reg++) {
1892 		*reg = CSR_READ_4(sc, ALE_RX_MIB_BASE + i);
1893 		i += sizeof(uint32_t);
1894 	}
1895 	/* Read Tx statistics. */
1896 	for (reg = &sb.tx_frames, i = 0; reg <= &sb.tx_mcast_bytes; reg++) {
1897 		*reg = CSR_READ_4(sc, ALE_TX_MIB_BASE + i);
1898 		i += sizeof(uint32_t);
1899 	}
1900 
1901 	/* Rx stats. */
1902 	stat->rx_frames += smb->rx_frames;
1903 	stat->rx_bcast_frames += smb->rx_bcast_frames;
1904 	stat->rx_mcast_frames += smb->rx_mcast_frames;
1905 	stat->rx_pause_frames += smb->rx_pause_frames;
1906 	stat->rx_control_frames += smb->rx_control_frames;
1907 	stat->rx_crcerrs += smb->rx_crcerrs;
1908 	stat->rx_lenerrs += smb->rx_lenerrs;
1909 	stat->rx_bytes += smb->rx_bytes;
1910 	stat->rx_runts += smb->rx_runts;
1911 	stat->rx_fragments += smb->rx_fragments;
1912 	stat->rx_pkts_64 += smb->rx_pkts_64;
1913 	stat->rx_pkts_65_127 += smb->rx_pkts_65_127;
1914 	stat->rx_pkts_128_255 += smb->rx_pkts_128_255;
1915 	stat->rx_pkts_256_511 += smb->rx_pkts_256_511;
1916 	stat->rx_pkts_512_1023 += smb->rx_pkts_512_1023;
1917 	stat->rx_pkts_1024_1518 += smb->rx_pkts_1024_1518;
1918 	stat->rx_pkts_1519_max += smb->rx_pkts_1519_max;
1919 	stat->rx_pkts_truncated += smb->rx_pkts_truncated;
1920 	stat->rx_fifo_oflows += smb->rx_fifo_oflows;
1921 	stat->rx_rrs_errs += smb->rx_rrs_errs;
1922 	stat->rx_alignerrs += smb->rx_alignerrs;
1923 	stat->rx_bcast_bytes += smb->rx_bcast_bytes;
1924 	stat->rx_mcast_bytes += smb->rx_mcast_bytes;
1925 	stat->rx_pkts_filtered += smb->rx_pkts_filtered;
1926 
1927 	/* Tx stats. */
1928 	stat->tx_frames += smb->tx_frames;
1929 	stat->tx_bcast_frames += smb->tx_bcast_frames;
1930 	stat->tx_mcast_frames += smb->tx_mcast_frames;
1931 	stat->tx_pause_frames += smb->tx_pause_frames;
1932 	stat->tx_excess_defer += smb->tx_excess_defer;
1933 	stat->tx_control_frames += smb->tx_control_frames;
1934 	stat->tx_deferred += smb->tx_deferred;
1935 	stat->tx_bytes += smb->tx_bytes;
1936 	stat->tx_pkts_64 += smb->tx_pkts_64;
1937 	stat->tx_pkts_65_127 += smb->tx_pkts_65_127;
1938 	stat->tx_pkts_128_255 += smb->tx_pkts_128_255;
1939 	stat->tx_pkts_256_511 += smb->tx_pkts_256_511;
1940 	stat->tx_pkts_512_1023 += smb->tx_pkts_512_1023;
1941 	stat->tx_pkts_1024_1518 += smb->tx_pkts_1024_1518;
1942 	stat->tx_pkts_1519_max += smb->tx_pkts_1519_max;
1943 	stat->tx_single_colls += smb->tx_single_colls;
1944 	stat->tx_multi_colls += smb->tx_multi_colls;
1945 	stat->tx_late_colls += smb->tx_late_colls;
1946 	stat->tx_excess_colls += smb->tx_excess_colls;
1947 	stat->tx_abort += smb->tx_abort;
1948 	stat->tx_underrun += smb->tx_underrun;
1949 	stat->tx_desc_underrun += smb->tx_desc_underrun;
1950 	stat->tx_lenerrs += smb->tx_lenerrs;
1951 	stat->tx_pkts_truncated += smb->tx_pkts_truncated;
1952 	stat->tx_bcast_bytes += smb->tx_bcast_bytes;
1953 	stat->tx_mcast_bytes += smb->tx_mcast_bytes;
1954 
1955 	/* Update counters in ifnet. */
1956 	IFNET_STAT_INC(ifp, opackets, smb->tx_frames);
1957 
1958 	IFNET_STAT_INC(ifp, collisions, smb->tx_single_colls +
1959 	    smb->tx_multi_colls * 2 + smb->tx_late_colls +
1960 	    smb->tx_abort * HDPX_CFG_RETRY_DEFAULT);
1961 
1962 	/*
1963 	 * XXX
1964 	 * tx_pkts_truncated counter looks suspicious. It constantly
1965 	 * increments with no sign of Tx errors. This may indicate
1966 	 * the counter name is not correct one so I've removed the
1967 	 * counter in output errors.
1968 	 */
1969 	IFNET_STAT_INC(ifp, oerrors, smb->tx_abort + smb->tx_late_colls +
1970 	    smb->tx_underrun);
1971 
1972 	IFNET_STAT_INC(ifp, ipackets, smb->rx_frames);
1973 
1974 	IFNET_STAT_INC(ifp, ierrors, smb->rx_crcerrs + smb->rx_lenerrs +
1975 	    smb->rx_runts + smb->rx_pkts_truncated +
1976 	    smb->rx_fifo_oflows + smb->rx_rrs_errs +
1977 	    smb->rx_alignerrs);
1978 }
1979 
1980 static void
1981 ale_intr(void *xsc)
1982 {
1983 	struct ale_softc *sc = xsc;
1984 	struct ifnet *ifp = &sc->arpcom.ac_if;
1985 	uint32_t status;
1986 
1987 	ASSERT_SERIALIZED(ifp->if_serializer);
1988 
1989 	status = CSR_READ_4(sc, ALE_INTR_STATUS);
1990 	if ((status & ALE_INTRS) == 0)
1991 		return;
1992 
1993 	/* Acknowledge and disable interrupts. */
1994 	CSR_WRITE_4(sc, ALE_INTR_STATUS, status | INTR_DIS_INT);
1995 
1996 	if ((ifp->if_flags & IFF_RUNNING) != 0) {
1997 		int error;
1998 
1999 		error = ale_rxeof(sc);
2000 		if (error) {
2001 			sc->ale_stats.reset_brk_seq++;
2002 			ale_init(sc);
2003 			return;
2004 		}
2005 
2006 		if ((status & (INTR_DMA_RD_TO_RST | INTR_DMA_WR_TO_RST)) != 0) {
2007 			if ((status & INTR_DMA_RD_TO_RST) != 0)
2008 				device_printf(sc->ale_dev,
2009 				    "DMA read error! -- resetting\n");
2010 			if ((status & INTR_DMA_WR_TO_RST) != 0)
2011 				device_printf(sc->ale_dev,
2012 				    "DMA write error! -- resetting\n");
2013 			ale_init(sc);
2014 			return;
2015 		}
2016 
2017 		ale_txeof(sc);
2018 		if (!ifq_is_empty(&ifp->if_snd))
2019 			if_devstart(ifp);
2020 	}
2021 
2022 	/* Re-enable interrupts. */
2023 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0x7FFFFFFF);
2024 }
2025 
2026 static void
2027 ale_txeof(struct ale_softc *sc)
2028 {
2029 	struct ifnet *ifp = &sc->arpcom.ac_if;
2030 	struct ale_txdesc *txd;
2031 	uint32_t cons, prod;
2032 	int prog;
2033 
2034 	if (sc->ale_cdata.ale_tx_cnt == 0)
2035 		return;
2036 
2037 	bus_dmamap_sync(sc->ale_cdata.ale_tx_ring_tag,
2038 	    sc->ale_cdata.ale_tx_ring_map, BUS_DMASYNC_POSTREAD);
2039 	if ((sc->ale_flags & ALE_FLAG_TXCMB_BUG) == 0) {
2040 		bus_dmamap_sync(sc->ale_cdata.ale_tx_cmb_tag,
2041 		    sc->ale_cdata.ale_tx_cmb_map, BUS_DMASYNC_POSTREAD);
2042 		prod = *sc->ale_cdata.ale_tx_cmb & TPD_CNT_MASK;
2043 	} else
2044 		prod = CSR_READ_2(sc, ALE_TPD_CONS_IDX);
2045 	cons = sc->ale_cdata.ale_tx_cons;
2046 	/*
2047 	 * Go through our Tx list and free mbufs for those
2048 	 * frames which have been transmitted.
2049 	 */
2050 	for (prog = 0; cons != prod; prog++,
2051 	     ALE_DESC_INC(cons, ALE_TX_RING_CNT)) {
2052 		if (sc->ale_cdata.ale_tx_cnt <= 0)
2053 			break;
2054 		prog++;
2055 		ifq_clr_oactive(&ifp->if_snd);
2056 		sc->ale_cdata.ale_tx_cnt--;
2057 		txd = &sc->ale_cdata.ale_txdesc[cons];
2058 		if (txd->tx_m != NULL) {
2059 			/* Reclaim transmitted mbufs. */
2060 			bus_dmamap_unload(sc->ale_cdata.ale_tx_tag,
2061 			    txd->tx_dmamap);
2062 			m_freem(txd->tx_m);
2063 			txd->tx_m = NULL;
2064 		}
2065 	}
2066 
2067 	if (prog > 0) {
2068 		sc->ale_cdata.ale_tx_cons = cons;
2069 		/*
2070 		 * Unarm watchdog timer only when there is no pending
2071 		 * Tx descriptors in queue.
2072 		 */
2073 		if (sc->ale_cdata.ale_tx_cnt == 0)
2074 			ifp->if_timer = 0;
2075 	}
2076 }
2077 
2078 static void
2079 ale_rx_update_page(struct ale_softc *sc, struct ale_rx_page **page,
2080     uint32_t length, uint32_t *prod)
2081 {
2082 	struct ale_rx_page *rx_page;
2083 
2084 	rx_page = *page;
2085 	/* Update consumer position. */
2086 	rx_page->cons += roundup(length + sizeof(struct rx_rs),
2087 	    ALE_RX_PAGE_ALIGN);
2088 	if (rx_page->cons >= ALE_RX_PAGE_SZ) {
2089 		/*
2090 		 * End of Rx page reached, let hardware reuse
2091 		 * this page.
2092 		 */
2093 		rx_page->cons = 0;
2094 		*rx_page->cmb_addr = 0;
2095 		bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2096 				BUS_DMASYNC_PREWRITE);
2097 		CSR_WRITE_1(sc, ALE_RXF0_PAGE0 + sc->ale_cdata.ale_rx_curp,
2098 		    RXF_VALID);
2099 		/* Switch to alternate Rx page. */
2100 		sc->ale_cdata.ale_rx_curp ^= 1;
2101 		rx_page = *page =
2102 		    &sc->ale_cdata.ale_rx_page[sc->ale_cdata.ale_rx_curp];
2103 		/* Page flipped, sync CMB and Rx page. */
2104 		bus_dmamap_sync(rx_page->page_tag, rx_page->page_map,
2105 		    BUS_DMASYNC_POSTREAD);
2106 		bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2107 		    BUS_DMASYNC_POSTREAD);
2108 		/* Sync completed, cache updated producer index. */
2109 		*prod = *rx_page->cmb_addr;
2110 	}
2111 }
2112 
2113 
2114 /*
2115  * It seems that AR81xx controller can compute partial checksum.
2116  * The partial checksum value can be used to accelerate checksum
2117  * computation for fragmented TCP/UDP packets. Upper network stack
2118  * already takes advantage of the partial checksum value in IP
2119  * reassembly stage. But I'm not sure the correctness of the
2120  * partial hardware checksum assistance due to lack of data sheet.
2121  * In addition, the Rx feature of controller that requires copying
2122  * for every frames effectively nullifies one of most nice offload
2123  * capability of controller.
2124  */
2125 static void
2126 ale_rxcsum(struct ale_softc *sc, struct mbuf *m, uint32_t status)
2127 {
2128 	struct ifnet *ifp = &sc->arpcom.ac_if;
2129 	struct ip *ip;
2130 	char *p;
2131 
2132 	m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
2133 	if ((status & ALE_RD_IPCSUM_NOK) == 0)
2134 		m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
2135 
2136 	if ((sc->ale_flags & ALE_FLAG_RXCSUM_BUG) == 0) {
2137 		if (((status & ALE_RD_IPV4_FRAG) == 0) &&
2138 		    ((status & (ALE_RD_TCP | ALE_RD_UDP)) != 0) &&
2139 		    ((status & ALE_RD_TCP_UDPCSUM_NOK) == 0)) {
2140 			m->m_pkthdr.csum_flags |=
2141 			    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2142 			m->m_pkthdr.csum_data = 0xffff;
2143 		}
2144 	} else {
2145 		if ((status & (ALE_RD_TCP | ALE_RD_UDP)) != 0 &&
2146 		    (status & ALE_RD_TCP_UDPCSUM_NOK) == 0) {
2147 			p = mtod(m, char *);
2148 			p += ETHER_HDR_LEN;
2149 			if ((status & ALE_RD_802_3) != 0)
2150 				p += LLC_SNAPFRAMELEN;
2151 			if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0 &&
2152 			    (status & ALE_RD_VLAN) != 0)
2153 				p += EVL_ENCAPLEN;
2154 			ip = (struct ip *)p;
2155 			if (ip->ip_off != 0 && (status & ALE_RD_IPV4_DF) == 0)
2156 				return;
2157 			m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
2158 			    CSUM_PSEUDO_HDR;
2159 			m->m_pkthdr.csum_data = 0xffff;
2160 		}
2161 	}
2162 	/*
2163 	 * Don't mark bad checksum for TCP/UDP frames
2164 	 * as fragmented frames may always have set
2165 	 * bad checksummed bit of frame status.
2166 	 */
2167 }
2168 
2169 /* Process received frames. */
2170 static int
2171 ale_rxeof(struct ale_softc *sc)
2172 {
2173 	struct ifnet *ifp = &sc->arpcom.ac_if;
2174 	struct ale_rx_page *rx_page;
2175 	struct rx_rs *rs;
2176 	struct mbuf *m;
2177 	uint32_t length, prod, seqno, status, vtags;
2178 	int prog;
2179 
2180 	rx_page = &sc->ale_cdata.ale_rx_page[sc->ale_cdata.ale_rx_curp];
2181 	bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2182 			BUS_DMASYNC_POSTREAD);
2183 	bus_dmamap_sync(rx_page->page_tag, rx_page->page_map,
2184 			BUS_DMASYNC_POSTREAD);
2185 	/*
2186 	 * Don't directly access producer index as hardware may
2187 	 * update it while Rx handler is in progress. It would
2188 	 * be even better if there is a way to let hardware
2189 	 * know how far driver processed its received frames.
2190 	 * Alternatively, hardware could provide a way to disable
2191 	 * CMB updates until driver acknowledges the end of CMB
2192 	 * access.
2193 	 */
2194 	prod = *rx_page->cmb_addr;
2195 	for (prog = 0; ; prog++) {
2196 		if (rx_page->cons >= prod)
2197 			break;
2198 		rs = (struct rx_rs *)(rx_page->page_addr + rx_page->cons);
2199 		seqno = ALE_RX_SEQNO(le32toh(rs->seqno));
2200 		if (sc->ale_cdata.ale_rx_seqno != seqno) {
2201 			/*
2202 			 * Normally I believe this should not happen unless
2203 			 * severe driver bug or corrupted memory. However
2204 			 * it seems to happen under certain conditions which
2205 			 * is triggered by abrupt Rx events such as initiation
2206 			 * of bulk transfer of remote host. It's not easy to
2207 			 * reproduce this and I doubt it could be related
2208 			 * with FIFO overflow of hardware or activity of Tx
2209 			 * CMB updates. I also remember similar behaviour
2210 			 * seen on RealTek 8139 which uses resembling Rx
2211 			 * scheme.
2212 			 */
2213 			if (bootverbose)
2214 				device_printf(sc->ale_dev,
2215 				    "garbled seq: %u, expected: %u -- "
2216 				    "resetting!\n", seqno,
2217 				    sc->ale_cdata.ale_rx_seqno);
2218 			return (EIO);
2219 		}
2220 		/* Frame received. */
2221 		sc->ale_cdata.ale_rx_seqno++;
2222 		length = ALE_RX_BYTES(le32toh(rs->length));
2223 		status = le32toh(rs->flags);
2224 		if ((status & ALE_RD_ERROR) != 0) {
2225 			/*
2226 			 * We want to pass the following frames to upper
2227 			 * layer regardless of error status of Rx return
2228 			 * status.
2229 			 *
2230 			 *  o IP/TCP/UDP checksum is bad.
2231 			 *  o frame length and protocol specific length
2232 			 *     does not match.
2233 			 */
2234 			if ((status & (ALE_RD_CRC | ALE_RD_CODE |
2235 			    ALE_RD_DRIBBLE | ALE_RD_RUNT | ALE_RD_OFLOW |
2236 			    ALE_RD_TRUNC)) != 0) {
2237 				ale_rx_update_page(sc, &rx_page, length, &prod);
2238 				continue;
2239 			}
2240 		}
2241 		/*
2242 		 * m_devget(9) is major bottle-neck of ale(4)(It comes
2243 		 * from hardware limitation). For jumbo frames we could
2244 		 * get a slightly better performance if driver use
2245 		 * m_getjcl(9) with proper buffer size argument. However
2246 		 * that would make code more complicated and I don't
2247 		 * think users would expect good Rx performance numbers
2248 		 * on these low-end consumer ethernet controller.
2249 		 */
2250 		m = m_devget((char *)(rs + 1), length - ETHER_CRC_LEN,
2251 		    ETHER_ALIGN, ifp, NULL);
2252 		if (m == NULL) {
2253 			IFNET_STAT_INC(ifp, iqdrops, 1);
2254 			ale_rx_update_page(sc, &rx_page, length, &prod);
2255 			continue;
2256 		}
2257 		if ((ifp->if_capenable & IFCAP_RXCSUM) != 0 &&
2258 		    (status & ALE_RD_IPV4) != 0)
2259 			ale_rxcsum(sc, m, status);
2260 		if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0 &&
2261 		    (status & ALE_RD_VLAN) != 0) {
2262 			vtags = ALE_RX_VLAN(le32toh(rs->vtags));
2263 			m->m_pkthdr.ether_vlantag = ALE_RX_VLAN_TAG(vtags);
2264 			m->m_flags |= M_VLANTAG;
2265 		}
2266 
2267 		/* Pass it to upper layer. */
2268 		ifp->if_input(ifp, m);
2269 
2270 		ale_rx_update_page(sc, &rx_page, length, &prod);
2271 	}
2272 	return 0;
2273 }
2274 
2275 static void
2276 ale_tick(void *xsc)
2277 {
2278 	struct ale_softc *sc = xsc;
2279 	struct ifnet *ifp = &sc->arpcom.ac_if;
2280 	struct mii_data *mii;
2281 
2282 	lwkt_serialize_enter(ifp->if_serializer);
2283 
2284 	mii = device_get_softc(sc->ale_miibus);
2285 	mii_tick(mii);
2286 	ale_stats_update(sc);
2287 
2288 	callout_reset(&sc->ale_tick_ch, hz, ale_tick, sc);
2289 
2290 	lwkt_serialize_exit(ifp->if_serializer);
2291 }
2292 
2293 static void
2294 ale_reset(struct ale_softc *sc)
2295 {
2296 	uint32_t reg;
2297 	int i;
2298 
2299 	/* Initialize PCIe module. From Linux. */
2300 	CSR_WRITE_4(sc, 0x1008, CSR_READ_4(sc, 0x1008) | 0x8000);
2301 
2302 	CSR_WRITE_4(sc, ALE_MASTER_CFG, MASTER_RESET);
2303 	for (i = ALE_RESET_TIMEOUT; i > 0; i--) {
2304 		DELAY(10);
2305 		if ((CSR_READ_4(sc, ALE_MASTER_CFG) & MASTER_RESET) == 0)
2306 			break;
2307 	}
2308 	if (i == 0)
2309 		device_printf(sc->ale_dev, "master reset timeout!\n");
2310 
2311 	for (i = ALE_RESET_TIMEOUT; i > 0; i--) {
2312 		if ((reg = CSR_READ_4(sc, ALE_IDLE_STATUS)) == 0)
2313 			break;
2314 		DELAY(10);
2315 	}
2316 
2317 	if (i == 0)
2318 		device_printf(sc->ale_dev, "reset timeout(0x%08x)!\n", reg);
2319 }
2320 
2321 static void
2322 ale_init(void *xsc)
2323 {
2324 	struct ale_softc *sc = xsc;
2325 	struct ifnet *ifp = &sc->arpcom.ac_if;
2326 	struct mii_data *mii;
2327 	uint8_t eaddr[ETHER_ADDR_LEN];
2328 	bus_addr_t paddr;
2329 	uint32_t reg, rxf_hi, rxf_lo;
2330 
2331 	ASSERT_SERIALIZED(ifp->if_serializer);
2332 
2333 	mii = device_get_softc(sc->ale_miibus);
2334 
2335 	/*
2336 	 * Cancel any pending I/O.
2337 	 */
2338 	ale_stop(sc);
2339 
2340 	/*
2341 	 * Reset the chip to a known state.
2342 	 */
2343 	ale_reset(sc);
2344 
2345 	/* Initialize Tx descriptors, DMA memory blocks. */
2346 	ale_init_rx_pages(sc);
2347 	ale_init_tx_ring(sc);
2348 
2349 	/* Reprogram the station address. */
2350 	bcopy(IF_LLADDR(ifp), eaddr, ETHER_ADDR_LEN);
2351 	CSR_WRITE_4(sc, ALE_PAR0,
2352 	    eaddr[2] << 24 | eaddr[3] << 16 | eaddr[4] << 8 | eaddr[5]);
2353 	CSR_WRITE_4(sc, ALE_PAR1, eaddr[0] << 8 | eaddr[1]);
2354 
2355 	/*
2356 	 * Clear WOL status and disable all WOL feature as WOL
2357 	 * would interfere Rx operation under normal environments.
2358 	 */
2359 	CSR_READ_4(sc, ALE_WOL_CFG);
2360 	CSR_WRITE_4(sc, ALE_WOL_CFG, 0);
2361 
2362 	/*
2363 	 * Set Tx descriptor/RXF0/CMB base addresses. They share
2364 	 * the same high address part of DMAable region.
2365 	 */
2366 	paddr = sc->ale_cdata.ale_tx_ring_paddr;
2367 	CSR_WRITE_4(sc, ALE_TPD_ADDR_HI, ALE_ADDR_HI(paddr));
2368 	CSR_WRITE_4(sc, ALE_TPD_ADDR_LO, ALE_ADDR_LO(paddr));
2369 	CSR_WRITE_4(sc, ALE_TPD_CNT,
2370 	    (ALE_TX_RING_CNT << TPD_CNT_SHIFT) & TPD_CNT_MASK);
2371 
2372 	/* Set Rx page base address, note we use single queue. */
2373 	paddr = sc->ale_cdata.ale_rx_page[0].page_paddr;
2374 	CSR_WRITE_4(sc, ALE_RXF0_PAGE0_ADDR_LO, ALE_ADDR_LO(paddr));
2375 	paddr = sc->ale_cdata.ale_rx_page[1].page_paddr;
2376 	CSR_WRITE_4(sc, ALE_RXF0_PAGE1_ADDR_LO, ALE_ADDR_LO(paddr));
2377 
2378 	/* Set Tx/Rx CMB addresses. */
2379 	paddr = sc->ale_cdata.ale_tx_cmb_paddr;
2380 	CSR_WRITE_4(sc, ALE_TX_CMB_ADDR_LO, ALE_ADDR_LO(paddr));
2381 	paddr = sc->ale_cdata.ale_rx_page[0].cmb_paddr;
2382 	CSR_WRITE_4(sc, ALE_RXF0_CMB0_ADDR_LO, ALE_ADDR_LO(paddr));
2383 	paddr = sc->ale_cdata.ale_rx_page[1].cmb_paddr;
2384 	CSR_WRITE_4(sc, ALE_RXF0_CMB1_ADDR_LO, ALE_ADDR_LO(paddr));
2385 
2386 	/* Mark RXF0 is valid. */
2387 	CSR_WRITE_1(sc, ALE_RXF0_PAGE0, RXF_VALID);
2388 	CSR_WRITE_1(sc, ALE_RXF0_PAGE1, RXF_VALID);
2389 	/*
2390 	 * No need to initialize RFX1/RXF2/RXF3. We don't use
2391 	 * multi-queue yet.
2392 	 */
2393 
2394 	/* Set Rx page size, excluding guard frame size. */
2395 	CSR_WRITE_4(sc, ALE_RXF_PAGE_SIZE, ALE_RX_PAGE_SZ);
2396 
2397 	/* Tell hardware that we're ready to load DMA blocks. */
2398 	CSR_WRITE_4(sc, ALE_DMA_BLOCK, DMA_BLOCK_LOAD);
2399 
2400 	/* Set Rx/Tx interrupt trigger threshold. */
2401 	CSR_WRITE_4(sc, ALE_INT_TRIG_THRESH, (1 << INT_TRIG_RX_THRESH_SHIFT) |
2402 	    (4 << INT_TRIG_TX_THRESH_SHIFT));
2403 	/*
2404 	 * XXX
2405 	 * Set interrupt trigger timer, its purpose and relation
2406 	 * with interrupt moderation mechanism is not clear yet.
2407 	 */
2408 	CSR_WRITE_4(sc, ALE_INT_TRIG_TIMER,
2409 	    ((ALE_USECS(10) << INT_TRIG_RX_TIMER_SHIFT) |
2410 	    (ALE_USECS(1000) << INT_TRIG_TX_TIMER_SHIFT)));
2411 
2412 	/* Configure interrupt moderation timer. */
2413 	reg = ALE_USECS(sc->ale_int_rx_mod) << IM_TIMER_RX_SHIFT;
2414 	reg |= ALE_USECS(sc->ale_int_tx_mod) << IM_TIMER_TX_SHIFT;
2415 	CSR_WRITE_4(sc, ALE_IM_TIMER, reg);
2416 	reg = CSR_READ_4(sc, ALE_MASTER_CFG);
2417 	reg &= ~(MASTER_CHIP_REV_MASK | MASTER_CHIP_ID_MASK);
2418 	reg &= ~(MASTER_IM_RX_TIMER_ENB | MASTER_IM_TX_TIMER_ENB);
2419 	if (ALE_USECS(sc->ale_int_rx_mod) != 0)
2420 		reg |= MASTER_IM_RX_TIMER_ENB;
2421 	if (ALE_USECS(sc->ale_int_tx_mod) != 0)
2422 		reg |= MASTER_IM_TX_TIMER_ENB;
2423 	CSR_WRITE_4(sc, ALE_MASTER_CFG, reg);
2424 	CSR_WRITE_2(sc, ALE_INTR_CLR_TIMER, ALE_USECS(1000));
2425 
2426 	/* Set Maximum frame size of controller. */
2427 	if (ifp->if_mtu < ETHERMTU)
2428 		sc->ale_max_frame_size = ETHERMTU;
2429 	else
2430 		sc->ale_max_frame_size = ifp->if_mtu;
2431 	sc->ale_max_frame_size += ETHER_HDR_LEN + EVL_ENCAPLEN + ETHER_CRC_LEN;
2432 	CSR_WRITE_4(sc, ALE_FRAME_SIZE, sc->ale_max_frame_size);
2433 
2434 	/* Configure IPG/IFG parameters. */
2435 	CSR_WRITE_4(sc, ALE_IPG_IFG_CFG,
2436 	    ((IPG_IFG_IPGT_DEFAULT << IPG_IFG_IPGT_SHIFT) & IPG_IFG_IPGT_MASK) |
2437 	    ((IPG_IFG_MIFG_DEFAULT << IPG_IFG_MIFG_SHIFT) & IPG_IFG_MIFG_MASK) |
2438 	    ((IPG_IFG_IPG1_DEFAULT << IPG_IFG_IPG1_SHIFT) & IPG_IFG_IPG1_MASK) |
2439 	    ((IPG_IFG_IPG2_DEFAULT << IPG_IFG_IPG2_SHIFT) & IPG_IFG_IPG2_MASK));
2440 
2441 	/* Set parameters for half-duplex media. */
2442 	CSR_WRITE_4(sc, ALE_HDPX_CFG,
2443 	    ((HDPX_CFG_LCOL_DEFAULT << HDPX_CFG_LCOL_SHIFT) &
2444 	    HDPX_CFG_LCOL_MASK) |
2445 	    ((HDPX_CFG_RETRY_DEFAULT << HDPX_CFG_RETRY_SHIFT) &
2446 	    HDPX_CFG_RETRY_MASK) | HDPX_CFG_EXC_DEF_EN |
2447 	    ((HDPX_CFG_ABEBT_DEFAULT << HDPX_CFG_ABEBT_SHIFT) &
2448 	    HDPX_CFG_ABEBT_MASK) |
2449 	    ((HDPX_CFG_JAMIPG_DEFAULT << HDPX_CFG_JAMIPG_SHIFT) &
2450 	    HDPX_CFG_JAMIPG_MASK));
2451 
2452 	/* Configure Tx jumbo frame parameters. */
2453 	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0) {
2454 		if (ifp->if_mtu < ETHERMTU)
2455 			reg = sc->ale_max_frame_size;
2456 		else if (ifp->if_mtu < 6 * 1024)
2457 			reg = (sc->ale_max_frame_size * 2) / 3;
2458 		else
2459 			reg = sc->ale_max_frame_size / 2;
2460 		CSR_WRITE_4(sc, ALE_TX_JUMBO_THRESH,
2461 		    roundup(reg, TX_JUMBO_THRESH_UNIT) >>
2462 		    TX_JUMBO_THRESH_UNIT_SHIFT);
2463 	}
2464 
2465 	/* Configure TxQ. */
2466 	reg = (128 << (sc->ale_dma_rd_burst >> DMA_CFG_RD_BURST_SHIFT))
2467 	    << TXQ_CFG_TX_FIFO_BURST_SHIFT;
2468 	reg |= (TXQ_CFG_TPD_BURST_DEFAULT << TXQ_CFG_TPD_BURST_SHIFT) &
2469 	    TXQ_CFG_TPD_BURST_MASK;
2470 	CSR_WRITE_4(sc, ALE_TXQ_CFG, reg | TXQ_CFG_ENHANCED_MODE | TXQ_CFG_ENB);
2471 
2472 	/* Configure Rx jumbo frame & flow control parameters. */
2473 	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0) {
2474 		reg = roundup(sc->ale_max_frame_size, RX_JUMBO_THRESH_UNIT);
2475 		CSR_WRITE_4(sc, ALE_RX_JUMBO_THRESH,
2476 		    (((reg >> RX_JUMBO_THRESH_UNIT_SHIFT) <<
2477 		    RX_JUMBO_THRESH_MASK_SHIFT) & RX_JUMBO_THRESH_MASK) |
2478 		    ((RX_JUMBO_LKAH_DEFAULT << RX_JUMBO_LKAH_SHIFT) &
2479 		    RX_JUMBO_LKAH_MASK));
2480 		reg = CSR_READ_4(sc, ALE_SRAM_RX_FIFO_LEN);
2481 		rxf_hi = (reg * 7) / 10;
2482 		rxf_lo = (reg * 3)/ 10;
2483 		CSR_WRITE_4(sc, ALE_RX_FIFO_PAUSE_THRESH,
2484 		    ((rxf_lo << RX_FIFO_PAUSE_THRESH_LO_SHIFT) &
2485 		    RX_FIFO_PAUSE_THRESH_LO_MASK) |
2486 		    ((rxf_hi << RX_FIFO_PAUSE_THRESH_HI_SHIFT) &
2487 		     RX_FIFO_PAUSE_THRESH_HI_MASK));
2488 	}
2489 
2490 	/* Disable RSS. */
2491 	CSR_WRITE_4(sc, ALE_RSS_IDT_TABLE0, 0);
2492 	CSR_WRITE_4(sc, ALE_RSS_CPU, 0);
2493 
2494 	/* Configure RxQ. */
2495 	CSR_WRITE_4(sc, ALE_RXQ_CFG,
2496 	    RXQ_CFG_ALIGN_32 | RXQ_CFG_CUT_THROUGH_ENB | RXQ_CFG_ENB);
2497 
2498 	/* Configure DMA parameters. */
2499 	reg = 0;
2500 	if ((sc->ale_flags & ALE_FLAG_TXCMB_BUG) == 0)
2501 		reg |= DMA_CFG_TXCMB_ENB;
2502 	CSR_WRITE_4(sc, ALE_DMA_CFG,
2503 	    DMA_CFG_OUT_ORDER | DMA_CFG_RD_REQ_PRI | DMA_CFG_RCB_64 |
2504 	    sc->ale_dma_rd_burst | reg |
2505 	    sc->ale_dma_wr_burst | DMA_CFG_RXCMB_ENB |
2506 	    ((DMA_CFG_RD_DELAY_CNT_DEFAULT << DMA_CFG_RD_DELAY_CNT_SHIFT) &
2507 	    DMA_CFG_RD_DELAY_CNT_MASK) |
2508 	    ((DMA_CFG_WR_DELAY_CNT_DEFAULT << DMA_CFG_WR_DELAY_CNT_SHIFT) &
2509 	    DMA_CFG_WR_DELAY_CNT_MASK));
2510 
2511 	/*
2512 	 * Hardware can be configured to issue SMB interrupt based
2513 	 * on programmed interval. Since there is a callout that is
2514 	 * invoked for every hz in driver we use that instead of
2515 	 * relying on periodic SMB interrupt.
2516 	 */
2517 	CSR_WRITE_4(sc, ALE_SMB_STAT_TIMER, ALE_USECS(0));
2518 
2519 	/* Clear MAC statistics. */
2520 	ale_stats_clear(sc);
2521 
2522 	/*
2523 	 * Configure Tx/Rx MACs.
2524 	 *  - Auto-padding for short frames.
2525 	 *  - Enable CRC generation.
2526 	 *  Actual reconfiguration of MAC for resolved speed/duplex
2527 	 *  is followed after detection of link establishment.
2528 	 *  AR81xx always does checksum computation regardless of
2529 	 *  MAC_CFG_RXCSUM_ENB bit. In fact, setting the bit will
2530 	 *  cause Rx handling issue for fragmented IP datagrams due
2531 	 *  to silicon bug.
2532 	 */
2533 	reg = MAC_CFG_TX_CRC_ENB | MAC_CFG_TX_AUTO_PAD | MAC_CFG_FULL_DUPLEX |
2534 	    ((MAC_CFG_PREAMBLE_DEFAULT << MAC_CFG_PREAMBLE_SHIFT) &
2535 	    MAC_CFG_PREAMBLE_MASK);
2536 	if ((sc->ale_flags & ALE_FLAG_FASTETHER) != 0)
2537 		reg |= MAC_CFG_SPEED_10_100;
2538 	else
2539 		reg |= MAC_CFG_SPEED_1000;
2540 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2541 
2542 	/* Set up the receive filter. */
2543 	ale_rxfilter(sc);
2544 	ale_rxvlan(sc);
2545 
2546 	/* Acknowledge all pending interrupts and clear it. */
2547 	CSR_WRITE_4(sc, ALE_INTR_MASK, ALE_INTRS);
2548 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0xFFFFFFFF);
2549 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0);
2550 
2551 	sc->ale_flags &= ~ALE_FLAG_LINK;
2552 
2553 	/* Switch to the current media. */
2554 	mii_mediachg(mii);
2555 
2556 	callout_reset(&sc->ale_tick_ch, hz, ale_tick, sc);
2557 
2558 	ifp->if_flags |= IFF_RUNNING;
2559 	ifq_clr_oactive(&ifp->if_snd);
2560 }
2561 
2562 static void
2563 ale_stop(struct ale_softc *sc)
2564 {
2565 	struct ifnet *ifp = &sc->arpcom.ac_if;
2566 	struct ale_txdesc *txd;
2567 	uint32_t reg;
2568 	int i;
2569 
2570 	ASSERT_SERIALIZED(ifp->if_serializer);
2571 
2572 	/*
2573 	 * Mark the interface down and cancel the watchdog timer.
2574 	 */
2575 	ifp->if_flags &= ~IFF_RUNNING;
2576 	ifq_clr_oactive(&ifp->if_snd);
2577 	ifp->if_timer = 0;
2578 
2579 	callout_stop(&sc->ale_tick_ch);
2580 	sc->ale_flags &= ~ALE_FLAG_LINK;
2581 
2582 	ale_stats_update(sc);
2583 
2584 	/* Disable interrupts. */
2585 	CSR_WRITE_4(sc, ALE_INTR_MASK, 0);
2586 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0xFFFFFFFF);
2587 
2588 	/* Disable queue processing and DMA. */
2589 	reg = CSR_READ_4(sc, ALE_TXQ_CFG);
2590 	reg &= ~TXQ_CFG_ENB;
2591 	CSR_WRITE_4(sc, ALE_TXQ_CFG, reg);
2592 	reg = CSR_READ_4(sc, ALE_RXQ_CFG);
2593 	reg &= ~RXQ_CFG_ENB;
2594 	CSR_WRITE_4(sc, ALE_RXQ_CFG, reg);
2595 	reg = CSR_READ_4(sc, ALE_DMA_CFG);
2596 	reg &= ~(DMA_CFG_TXCMB_ENB | DMA_CFG_RXCMB_ENB);
2597 	CSR_WRITE_4(sc, ALE_DMA_CFG, reg);
2598 	DELAY(1000);
2599 
2600 	/* Stop Rx/Tx MACs. */
2601 	ale_stop_mac(sc);
2602 
2603 	/* Disable interrupts again? XXX */
2604 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0xFFFFFFFF);
2605 
2606 	/*
2607 	 * Free TX mbufs still in the queues.
2608 	 */
2609 	for (i = 0; i < ALE_TX_RING_CNT; i++) {
2610 		txd = &sc->ale_cdata.ale_txdesc[i];
2611 		if (txd->tx_m != NULL) {
2612 			bus_dmamap_unload(sc->ale_cdata.ale_tx_tag,
2613 			    txd->tx_dmamap);
2614 			m_freem(txd->tx_m);
2615 			txd->tx_m = NULL;
2616 		}
2617         }
2618 }
2619 
2620 static void
2621 ale_stop_mac(struct ale_softc *sc)
2622 {
2623 	uint32_t reg;
2624 	int i;
2625 
2626 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
2627 	if ((reg & (MAC_CFG_TX_ENB | MAC_CFG_RX_ENB)) != 0) {
2628 		reg &= ~MAC_CFG_TX_ENB | MAC_CFG_RX_ENB;
2629 		CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2630 	}
2631 
2632 	for (i = ALE_TIMEOUT; i > 0; i--) {
2633 		reg = CSR_READ_4(sc, ALE_IDLE_STATUS);
2634 		if (reg == 0)
2635 			break;
2636 		DELAY(10);
2637 	}
2638 	if (i == 0)
2639 		device_printf(sc->ale_dev,
2640 		    "could not disable Tx/Rx MAC(0x%08x)!\n", reg);
2641 }
2642 
2643 static void
2644 ale_init_tx_ring(struct ale_softc *sc)
2645 {
2646 	struct ale_txdesc *txd;
2647 	int i;
2648 
2649 	sc->ale_cdata.ale_tx_prod = 0;
2650 	sc->ale_cdata.ale_tx_cons = 0;
2651 	sc->ale_cdata.ale_tx_cnt = 0;
2652 
2653 	bzero(sc->ale_cdata.ale_tx_ring, ALE_TX_RING_SZ);
2654 	bzero(sc->ale_cdata.ale_tx_cmb, ALE_TX_CMB_SZ);
2655 	for (i = 0; i < ALE_TX_RING_CNT; i++) {
2656 		txd = &sc->ale_cdata.ale_txdesc[i];
2657 		txd->tx_m = NULL;
2658 	}
2659 	*sc->ale_cdata.ale_tx_cmb = 0;
2660 	bus_dmamap_sync(sc->ale_cdata.ale_tx_cmb_tag,
2661 	    sc->ale_cdata.ale_tx_cmb_map,
2662 	    BUS_DMASYNC_PREWRITE);
2663 	bus_dmamap_sync(sc->ale_cdata.ale_tx_ring_tag,
2664 	    sc->ale_cdata.ale_tx_ring_map,
2665 	    BUS_DMASYNC_PREWRITE);
2666 }
2667 
2668 static void
2669 ale_init_rx_pages(struct ale_softc *sc)
2670 {
2671 	struct ale_rx_page *rx_page;
2672 	int i;
2673 
2674 	sc->ale_cdata.ale_rx_seqno = 0;
2675 	sc->ale_cdata.ale_rx_curp = 0;
2676 
2677 	for (i = 0; i < ALE_RX_PAGES; i++) {
2678 		rx_page = &sc->ale_cdata.ale_rx_page[i];
2679 		bzero(rx_page->page_addr, sc->ale_pagesize);
2680 		bzero(rx_page->cmb_addr, ALE_RX_CMB_SZ);
2681 		rx_page->cons = 0;
2682 		*rx_page->cmb_addr = 0;
2683 		bus_dmamap_sync(rx_page->page_tag, rx_page->page_map,
2684 				BUS_DMASYNC_PREWRITE);
2685 		bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2686 				BUS_DMASYNC_PREWRITE);
2687 	}
2688 }
2689 
2690 static void
2691 ale_rxvlan(struct ale_softc *sc)
2692 {
2693 	struct ifnet *ifp;
2694 	uint32_t reg;
2695 
2696 	ifp = &sc->arpcom.ac_if;
2697 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
2698 	reg &= ~MAC_CFG_VLAN_TAG_STRIP;
2699 	if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
2700 		reg |= MAC_CFG_VLAN_TAG_STRIP;
2701 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2702 }
2703 
2704 static void
2705 ale_rxfilter(struct ale_softc *sc)
2706 {
2707 	struct ifnet *ifp;
2708 	struct ifmultiaddr *ifma;
2709 	uint32_t crc;
2710 	uint32_t mchash[2];
2711 	uint32_t rxcfg;
2712 
2713 	ifp = &sc->arpcom.ac_if;
2714 
2715 	rxcfg = CSR_READ_4(sc, ALE_MAC_CFG);
2716 	rxcfg &= ~(MAC_CFG_ALLMULTI | MAC_CFG_BCAST | MAC_CFG_PROMISC);
2717 	if ((ifp->if_flags & IFF_BROADCAST) != 0)
2718 		rxcfg |= MAC_CFG_BCAST;
2719 	if ((ifp->if_flags & (IFF_PROMISC | IFF_ALLMULTI)) != 0) {
2720 		if ((ifp->if_flags & IFF_PROMISC) != 0)
2721 			rxcfg |= MAC_CFG_PROMISC;
2722 		if ((ifp->if_flags & IFF_ALLMULTI) != 0)
2723 			rxcfg |= MAC_CFG_ALLMULTI;
2724 		CSR_WRITE_4(sc, ALE_MAR0, 0xFFFFFFFF);
2725 		CSR_WRITE_4(sc, ALE_MAR1, 0xFFFFFFFF);
2726 		CSR_WRITE_4(sc, ALE_MAC_CFG, rxcfg);
2727 		return;
2728 	}
2729 
2730 	/* Program new filter. */
2731 	bzero(mchash, sizeof(mchash));
2732 
2733 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2734 		if (ifma->ifma_addr->sa_family != AF_LINK)
2735 			continue;
2736 		crc = ether_crc32_le(LLADDR((struct sockaddr_dl *)
2737 		    ifma->ifma_addr), ETHER_ADDR_LEN);
2738 		mchash[crc >> 31] |= 1 << ((crc >> 26) & 0x1f);
2739 	}
2740 
2741 	CSR_WRITE_4(sc, ALE_MAR0, mchash[0]);
2742 	CSR_WRITE_4(sc, ALE_MAR1, mchash[1]);
2743 	CSR_WRITE_4(sc, ALE_MAC_CFG, rxcfg);
2744 }
2745 
2746 static int
2747 sysctl_hw_ale_int_mod(SYSCTL_HANDLER_ARGS)
2748 {
2749 	return (sysctl_int_range(oidp, arg1, arg2, req,
2750 	    ALE_IM_TIMER_MIN, ALE_IM_TIMER_MAX));
2751 }
2752 
2753 static void
2754 ale_dmamap_buf_cb(void *xctx, bus_dma_segment_t *segs, int nsegs,
2755 		  bus_size_t mapsz __unused, int error)
2756 {
2757 	struct ale_dmamap_ctx *ctx = xctx;
2758 	int i;
2759 
2760 	if (error)
2761 		return;
2762 
2763 	if (nsegs > ctx->nsegs) {
2764 		ctx->nsegs = 0;
2765 		return;
2766 	}
2767 
2768 	ctx->nsegs = nsegs;
2769 	for (i = 0; i < nsegs; ++i)
2770 		ctx->segs[i] = segs[i];
2771 }
2772