xref: /dragonfly/sys/dev/netif/ale/if_ale.c (revision ad7a2457)
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/miivar.h>
59 #include <dev/netif/mii_layer/jmphyreg.h>
60 
61 #include <bus/pci/pcireg.h>
62 #include <bus/pci/pcivar.h>
63 #include <bus/pci/pcidevs.h>
64 
65 #include <dev/netif/ale/if_alereg.h>
66 #include <dev/netif/ale/if_alevar.h>
67 
68 /* "device miibus" required.  See GENERIC if you get errors here. */
69 #include "miibus_if.h"
70 
71 /* For more information about Tx checksum offload issues see ale_encap(). */
72 #define	ALE_CSUM_FEATURES	(CSUM_TCP | CSUM_UDP)
73 
74 struct ale_dmamap_ctx {
75 	int			nsegs;
76 	bus_dma_segment_t	*segs;
77 };
78 
79 static int	ale_probe(device_t);
80 static int	ale_attach(device_t);
81 static int	ale_detach(device_t);
82 static int	ale_shutdown(device_t);
83 static int	ale_suspend(device_t);
84 static int	ale_resume(device_t);
85 
86 static int	ale_miibus_readreg(device_t, int, int);
87 static int	ale_miibus_writereg(device_t, int, int, int);
88 static void	ale_miibus_statchg(device_t);
89 
90 static void	ale_init(void *);
91 static void	ale_start(struct ifnet *);
92 static int	ale_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
93 static void	ale_watchdog(struct ifnet *);
94 static int	ale_mediachange(struct ifnet *);
95 static void	ale_mediastatus(struct ifnet *, struct ifmediareq *);
96 
97 static void	ale_intr(void *);
98 static int	ale_rxeof(struct ale_softc *sc);
99 static void	ale_rx_update_page(struct ale_softc *, struct ale_rx_page **,
100 		    uint32_t, uint32_t *);
101 static void	ale_rxcsum(struct ale_softc *, struct mbuf *, uint32_t);
102 static void	ale_txeof(struct ale_softc *);
103 
104 static int	ale_dma_alloc(struct ale_softc *);
105 static void	ale_dma_free(struct ale_softc *);
106 static int	ale_check_boundary(struct ale_softc *);
107 static void	ale_dmamap_cb(void *, bus_dma_segment_t *, int, int);
108 static void	ale_dmamap_buf_cb(void *, bus_dma_segment_t *, int,
109 		    bus_size_t, int);
110 static int	ale_encap(struct ale_softc *, struct mbuf **);
111 static void	ale_init_rx_pages(struct ale_softc *);
112 static void	ale_init_tx_ring(struct ale_softc *);
113 
114 static void	ale_stop(struct ale_softc *);
115 static void	ale_tick(void *);
116 static void	ale_get_macaddr(struct ale_softc *);
117 static void	ale_mac_config(struct ale_softc *);
118 static void	ale_phy_reset(struct ale_softc *);
119 static void	ale_reset(struct ale_softc *);
120 static void	ale_rxfilter(struct ale_softc *);
121 static void	ale_rxvlan(struct ale_softc *);
122 static void	ale_stats_clear(struct ale_softc *);
123 static void	ale_stats_update(struct ale_softc *);
124 static void	ale_stop_mac(struct ale_softc *);
125 #ifdef notyet
126 static void	ale_setlinkspeed(struct ale_softc *);
127 static void	ale_setwol(struct ale_softc *);
128 #endif
129 
130 static void	ale_sysctl_node(struct ale_softc *);
131 static int	sysctl_hw_ale_int_mod(SYSCTL_HANDLER_ARGS);
132 
133 /*
134  * Devices supported by this driver.
135  */
136 static struct ale_dev {
137 	uint16_t	ale_vendorid;
138 	uint16_t	ale_deviceid;
139 	const char	*ale_name;
140 } ale_devs[] = {
141     { VENDORID_ATHEROS, DEVICEID_ATHEROS_AR81XX,
142     "Atheros AR8121/AR8113/AR8114 PCIe Ethernet" },
143 };
144 
145 static device_method_t ale_methods[] = {
146 	/* Device interface. */
147 	DEVMETHOD(device_probe,		ale_probe),
148 	DEVMETHOD(device_attach,	ale_attach),
149 	DEVMETHOD(device_detach,	ale_detach),
150 	DEVMETHOD(device_shutdown,	ale_shutdown),
151 	DEVMETHOD(device_suspend,	ale_suspend),
152 	DEVMETHOD(device_resume,	ale_resume),
153 
154 	/* Bus interface. */
155 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
156 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
157 
158 	/* MII interface. */
159 	DEVMETHOD(miibus_readreg,	ale_miibus_readreg),
160 	DEVMETHOD(miibus_writereg,	ale_miibus_writereg),
161 	DEVMETHOD(miibus_statchg,	ale_miibus_statchg),
162 
163 	{ NULL, NULL }
164 };
165 
166 static driver_t ale_driver = {
167 	"ale",
168 	ale_methods,
169 	sizeof(struct ale_softc)
170 };
171 
172 static devclass_t ale_devclass;
173 
174 DECLARE_DUMMY_MODULE(if_ale);
175 MODULE_DEPEND(if_ale, miibus, 1, 1, 1);
176 DRIVER_MODULE(if_ale, pci, ale_driver, ale_devclass, 0, 0);
177 DRIVER_MODULE(miibus, ale, miibus_driver, miibus_devclass, 0, 0);
178 
179 static int
180 ale_miibus_readreg(device_t dev, int phy, int reg)
181 {
182 	struct ale_softc *sc;
183 	uint32_t v;
184 	int i;
185 
186 	sc = device_get_softc(dev);
187 
188 	if (phy != sc->ale_phyaddr)
189 		return (0);
190 
191 	CSR_WRITE_4(sc, ALE_MDIO, MDIO_OP_EXECUTE | MDIO_OP_READ |
192 	    MDIO_SUP_PREAMBLE | MDIO_CLK_25_4 | MDIO_REG_ADDR(reg));
193 	for (i = ALE_PHY_TIMEOUT; i > 0; i--) {
194 		DELAY(5);
195 		v = CSR_READ_4(sc, ALE_MDIO);
196 		if ((v & (MDIO_OP_EXECUTE | MDIO_OP_BUSY)) == 0)
197 			break;
198 	}
199 
200 	if (i == 0) {
201 		device_printf(sc->ale_dev, "phy read timeout : %d\n", reg);
202 		return (0);
203 	}
204 
205 	return ((v & MDIO_DATA_MASK) >> MDIO_DATA_SHIFT);
206 }
207 
208 static int
209 ale_miibus_writereg(device_t dev, int phy, int reg, int val)
210 {
211 	struct ale_softc *sc;
212 	uint32_t v;
213 	int i;
214 
215 	sc = device_get_softc(dev);
216 
217 	if (phy != sc->ale_phyaddr)
218 		return (0);
219 
220 	CSR_WRITE_4(sc, ALE_MDIO, MDIO_OP_EXECUTE | MDIO_OP_WRITE |
221 	    (val & MDIO_DATA_MASK) << MDIO_DATA_SHIFT |
222 	    MDIO_SUP_PREAMBLE | MDIO_CLK_25_4 | MDIO_REG_ADDR(reg));
223 	for (i = ALE_PHY_TIMEOUT; i > 0; i--) {
224 		DELAY(5);
225 		v = CSR_READ_4(sc, ALE_MDIO);
226 		if ((v & (MDIO_OP_EXECUTE | MDIO_OP_BUSY)) == 0)
227 			break;
228 	}
229 
230 	if (i == 0)
231 		device_printf(sc->ale_dev, "phy write timeout : %d\n", reg);
232 
233 	return (0);
234 }
235 
236 static void
237 ale_miibus_statchg(device_t dev)
238 {
239 	struct ale_softc *sc = device_get_softc(dev);
240 	struct ifnet *ifp = &sc->arpcom.ac_if;
241 	struct mii_data *mii;
242 	uint32_t reg;
243 
244 	ASSERT_SERIALIZED(ifp->if_serializer);
245 
246 	if ((ifp->if_flags & IFF_RUNNING) == 0)
247 		return;
248 
249 	mii = device_get_softc(sc->ale_miibus);
250 
251 	sc->ale_flags &= ~ALE_FLAG_LINK;
252 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
253 	    (IFM_ACTIVE | IFM_AVALID)) {
254 		switch (IFM_SUBTYPE(mii->mii_media_active)) {
255 		case IFM_10_T:
256 		case IFM_100_TX:
257 			sc->ale_flags |= ALE_FLAG_LINK;
258 			break;
259 
260 		case IFM_1000_T:
261 			if ((sc->ale_flags & ALE_FLAG_FASTETHER) == 0)
262 				sc->ale_flags |= ALE_FLAG_LINK;
263 			break;
264 
265 		default:
266 			break;
267 		}
268 	}
269 
270 	/* Stop Rx/Tx MACs. */
271 	ale_stop_mac(sc);
272 
273 	/* Program MACs with resolved speed/duplex/flow-control. */
274 	if ((sc->ale_flags & ALE_FLAG_LINK) != 0) {
275 		ale_mac_config(sc);
276 		/* Reenable Tx/Rx MACs. */
277 		reg = CSR_READ_4(sc, ALE_MAC_CFG);
278 		reg |= MAC_CFG_TX_ENB | MAC_CFG_RX_ENB;
279 		CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
280 	}
281 }
282 
283 static void
284 ale_mediastatus(struct ifnet *ifp, struct ifmediareq *ifmr)
285 {
286 	struct ale_softc *sc = ifp->if_softc;
287 	struct mii_data *mii = device_get_softc(sc->ale_miibus);
288 
289 	ASSERT_SERIALIZED(ifp->if_serializer);
290 
291 	mii_pollstat(mii);
292 	ifmr->ifm_status = mii->mii_media_status;
293 	ifmr->ifm_active = mii->mii_media_active;
294 }
295 
296 static int
297 ale_mediachange(struct ifnet *ifp)
298 {
299 	struct ale_softc *sc = ifp->if_softc;
300 	struct mii_data *mii = device_get_softc(sc->ale_miibus);
301 	int error;
302 
303 	ASSERT_SERIALIZED(ifp->if_serializer);
304 
305 	if (mii->mii_instance != 0) {
306 		struct mii_softc *miisc;
307 
308 		LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
309 			mii_phy_reset(miisc);
310 	}
311 	error = mii_mediachg(mii);
312 
313 	return (error);
314 }
315 
316 static int
317 ale_probe(device_t dev)
318 {
319 	struct ale_dev *sp;
320 	int i;
321 	uint16_t vendor, devid;
322 
323 	vendor = pci_get_vendor(dev);
324 	devid = pci_get_device(dev);
325 	sp = ale_devs;
326 	for (i = 0; i < sizeof(ale_devs) / sizeof(ale_devs[0]); i++) {
327 		if (vendor == sp->ale_vendorid &&
328 		    devid == sp->ale_deviceid) {
329 			device_set_desc(dev, sp->ale_name);
330 			return (0);
331 		}
332 		sp++;
333 	}
334 
335 	return (ENXIO);
336 }
337 
338 static void
339 ale_get_macaddr(struct ale_softc *sc)
340 {
341 	uint32_t ea[2], reg;
342 	int i, vpdc;
343 
344 	reg = CSR_READ_4(sc, ALE_SPI_CTRL);
345 	if ((reg & SPI_VPD_ENB) != 0) {
346 		reg &= ~SPI_VPD_ENB;
347 		CSR_WRITE_4(sc, ALE_SPI_CTRL, reg);
348 	}
349 
350 	vpdc = pci_get_vpdcap_ptr(sc->ale_dev);
351 	if (vpdc) {
352 		/*
353 		 * PCI VPD capability found, let TWSI reload EEPROM.
354 		 * This will set ethernet address of controller.
355 		 */
356 		CSR_WRITE_4(sc, ALE_TWSI_CTRL, CSR_READ_4(sc, ALE_TWSI_CTRL) |
357 		    TWSI_CTRL_SW_LD_START);
358 		for (i = 100; i > 0; i--) {
359 			DELAY(1000);
360 			reg = CSR_READ_4(sc, ALE_TWSI_CTRL);
361 			if ((reg & TWSI_CTRL_SW_LD_START) == 0)
362 				break;
363 		}
364 		if (i == 0)
365 			device_printf(sc->ale_dev,
366 			    "reloading EEPROM timeout!\n");
367 	} else {
368 		if (bootverbose)
369 			device_printf(sc->ale_dev,
370 			    "PCI VPD capability not found!\n");
371 	}
372 
373 	ea[0] = CSR_READ_4(sc, ALE_PAR0);
374 	ea[1] = CSR_READ_4(sc, ALE_PAR1);
375 	sc->ale_eaddr[0] = (ea[1] >> 8) & 0xFF;
376 	sc->ale_eaddr[1] = (ea[1] >> 0) & 0xFF;
377 	sc->ale_eaddr[2] = (ea[0] >> 24) & 0xFF;
378 	sc->ale_eaddr[3] = (ea[0] >> 16) & 0xFF;
379 	sc->ale_eaddr[4] = (ea[0] >> 8) & 0xFF;
380 	sc->ale_eaddr[5] = (ea[0] >> 0) & 0xFF;
381 }
382 
383 static void
384 ale_phy_reset(struct ale_softc *sc)
385 {
386 	/* Reset magic from Linux. */
387 	CSR_WRITE_2(sc, ALE_GPHY_CTRL,
388 	    GPHY_CTRL_HIB_EN | GPHY_CTRL_HIB_PULSE | GPHY_CTRL_SEL_ANA_RESET |
389 	    GPHY_CTRL_PHY_PLL_ON);
390 	DELAY(1000);
391 	CSR_WRITE_2(sc, ALE_GPHY_CTRL,
392 	    GPHY_CTRL_EXT_RESET | GPHY_CTRL_HIB_EN | GPHY_CTRL_HIB_PULSE |
393 	    GPHY_CTRL_SEL_ANA_RESET | GPHY_CTRL_PHY_PLL_ON);
394 	DELAY(1000);
395 
396 #define	ATPHY_DBG_ADDR		0x1D
397 #define	ATPHY_DBG_DATA		0x1E
398 
399 	/* Enable hibernation mode. */
400 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
401 	    ATPHY_DBG_ADDR, 0x0B);
402 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
403 	    ATPHY_DBG_DATA, 0xBC00);
404 	/* Set Class A/B for all modes. */
405 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
406 	    ATPHY_DBG_ADDR, 0x00);
407 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
408 	    ATPHY_DBG_DATA, 0x02EF);
409 	/* Enable 10BT power saving. */
410 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
411 	    ATPHY_DBG_ADDR, 0x12);
412 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
413 	    ATPHY_DBG_DATA, 0x4C04);
414 	/* Adjust 1000T power. */
415 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
416 	    ATPHY_DBG_ADDR, 0x04);
417 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
418 	    ATPHY_DBG_ADDR, 0x8BBB);
419 	/* 10BT center tap voltage. */
420 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
421 	    ATPHY_DBG_ADDR, 0x05);
422 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
423 	    ATPHY_DBG_ADDR, 0x2C46);
424 
425 #undef	ATPHY_DBG_ADDR
426 #undef	ATPHY_DBG_DATA
427 	DELAY(1000);
428 }
429 
430 static int
431 ale_attach(device_t dev)
432 {
433 	struct ale_softc *sc = device_get_softc(dev);
434 	struct ifnet *ifp = &sc->arpcom.ac_if;
435 	int error = 0;
436 	uint32_t rxf_len, txf_len;
437 	uint8_t pcie_ptr;
438 
439 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
440 	sc->ale_dev = dev;
441 
442 	callout_init(&sc->ale_tick_ch);
443 
444 #ifndef BURN_BRIDGES
445 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
446 		uint32_t irq, mem;
447 
448 		irq = pci_read_config(dev, PCIR_INTLINE, 4);
449 		mem = pci_read_config(dev, ALE_PCIR_BAR, 4);
450 
451 		device_printf(dev, "chip is in D%d power mode "
452 		    "-- setting to D0\n", pci_get_powerstate(dev));
453 
454 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
455 
456 		pci_write_config(dev, PCIR_INTLINE, irq, 4);
457 		pci_write_config(dev, ALE_PCIR_BAR, mem, 4);
458 	}
459 #endif	/* !BURN_BRIDGE */
460 
461 	/* Enable bus mastering */
462 	pci_enable_busmaster(dev);
463 
464 	/*
465 	 * Allocate memory mapped IO
466 	 */
467 	sc->ale_mem_rid = ALE_PCIR_BAR;
468 	sc->ale_mem_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
469 						 &sc->ale_mem_rid, RF_ACTIVE);
470 	if (sc->ale_mem_res == NULL) {
471 		device_printf(dev, "can't allocate IO memory\n");
472 		return ENXIO;
473 	}
474 	sc->ale_mem_bt = rman_get_bustag(sc->ale_mem_res);
475 	sc->ale_mem_bh = rman_get_bushandle(sc->ale_mem_res);
476 
477 	/*
478 	 * Allocate IRQ
479 	 */
480 	sc->ale_irq_rid = 0;
481 	sc->ale_irq_res = bus_alloc_resource_any(dev, SYS_RES_IRQ,
482 						 &sc->ale_irq_rid,
483 						 RF_SHAREABLE | RF_ACTIVE);
484 	if (sc->ale_irq_res == NULL) {
485 		device_printf(dev, "can't allocate irq\n");
486 		error = ENXIO;
487 		goto fail;
488 	}
489 
490 	/* Set PHY address. */
491 	sc->ale_phyaddr = ALE_PHY_ADDR;
492 
493 	/* Reset PHY. */
494 	ale_phy_reset(sc);
495 
496 	/* Reset the ethernet controller. */
497 	ale_reset(sc);
498 
499 	/* Get PCI and chip id/revision. */
500 	sc->ale_rev = pci_get_revid(dev);
501 	if (sc->ale_rev >= 0xF0) {
502 		/* L2E Rev. B. AR8114 */
503 		sc->ale_flags |= ALE_FLAG_FASTETHER;
504 	} else {
505 		if ((CSR_READ_4(sc, ALE_PHY_STATUS) & PHY_STATUS_100M) != 0) {
506 			/* L1E AR8121 */
507 			sc->ale_flags |= ALE_FLAG_JUMBO;
508 		} else {
509 			/* L2E Rev. A. AR8113 */
510 			sc->ale_flags |= ALE_FLAG_FASTETHER;
511 		}
512 	}
513 
514 	/*
515 	 * All known controllers seems to require 4 bytes alignment
516 	 * of Tx buffers to make Tx checksum offload with custom
517 	 * checksum generation method work.
518 	 */
519 	sc->ale_flags |= ALE_FLAG_TXCSUM_BUG;
520 
521 	/*
522 	 * All known controllers seems to have issues on Rx checksum
523 	 * offload for fragmented IP datagrams.
524 	 */
525 	sc->ale_flags |= ALE_FLAG_RXCSUM_BUG;
526 
527 	/*
528 	 * Don't use Tx CMB. It is known to cause RRS update failure
529 	 * under certain circumstances. Typical phenomenon of the
530 	 * issue would be unexpected sequence number encountered in
531 	 * Rx handler.
532 	 */
533 	sc->ale_flags |= ALE_FLAG_TXCMB_BUG;
534 	sc->ale_chip_rev = CSR_READ_4(sc, ALE_MASTER_CFG) >>
535 	    MASTER_CHIP_REV_SHIFT;
536 	if (bootverbose) {
537 		device_printf(dev, "PCI device revision : 0x%04x\n",
538 		    sc->ale_rev);
539 		device_printf(dev, "Chip id/revision : 0x%04x\n",
540 		    sc->ale_chip_rev);
541 	}
542 
543 	/*
544 	 * Uninitialized hardware returns an invalid chip id/revision
545 	 * as well as 0xFFFFFFFF for Tx/Rx fifo length.
546 	 */
547 	txf_len = CSR_READ_4(sc, ALE_SRAM_TX_FIFO_LEN);
548 	rxf_len = CSR_READ_4(sc, ALE_SRAM_RX_FIFO_LEN);
549 	if (sc->ale_chip_rev == 0xFFFF || txf_len == 0xFFFFFFFF ||
550 	    rxf_len == 0xFFFFFFF) {
551 		device_printf(dev,"chip revision : 0x%04x, %u Tx FIFO "
552 		    "%u Rx FIFO -- not initialized?\n", sc->ale_chip_rev,
553 		    txf_len, rxf_len);
554 		error = ENXIO;
555 		goto fail;
556 	}
557 	device_printf(dev, "%u Tx FIFO, %u Rx FIFO\n", txf_len, rxf_len);
558 
559 	/* Get DMA parameters from PCIe device control register. */
560 	pcie_ptr = pci_get_pciecap_ptr(dev);
561 	if (pcie_ptr) {
562 		uint16_t devctl;
563 
564 		sc->ale_flags |= ALE_FLAG_PCIE;
565 		devctl = pci_read_config(dev, pcie_ptr + PCIER_DEVCTRL, 2);
566 		/* Max read request size. */
567 		sc->ale_dma_rd_burst = ((devctl >> 12) & 0x07) <<
568 		    DMA_CFG_RD_BURST_SHIFT;
569 		/* Max payload size. */
570 		sc->ale_dma_wr_burst = ((devctl >> 5) & 0x07) <<
571 		    DMA_CFG_WR_BURST_SHIFT;
572 		if (bootverbose) {
573 			device_printf(dev, "Read request size : %d bytes.\n",
574 			    128 << ((devctl >> 12) & 0x07));
575 			device_printf(dev, "TLP payload size : %d bytes.\n",
576 			    128 << ((devctl >> 5) & 0x07));
577 		}
578 	} else {
579 		sc->ale_dma_rd_burst = DMA_CFG_RD_BURST_128;
580 		sc->ale_dma_wr_burst = DMA_CFG_WR_BURST_128;
581 	}
582 
583 	/* Create device sysctl node. */
584 	ale_sysctl_node(sc);
585 
586 	if ((error = ale_dma_alloc(sc) != 0))
587 		goto fail;
588 
589 	/* Load station address. */
590 	ale_get_macaddr(sc);
591 
592 	ifp->if_softc = sc;
593 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
594 	ifp->if_ioctl = ale_ioctl;
595 	ifp->if_start = ale_start;
596 	ifp->if_init = ale_init;
597 	ifp->if_watchdog = ale_watchdog;
598 	ifq_set_maxlen(&ifp->if_snd, ALE_TX_RING_CNT - 1);
599 	ifq_set_ready(&ifp->if_snd);
600 
601 	ifp->if_capabilities = IFCAP_HWCSUM |
602 			       IFCAP_VLAN_MTU |
603 			       IFCAP_VLAN_HWTAGGING;
604 	ifp->if_hwassist = ALE_CSUM_FEATURES;
605 	ifp->if_capenable = ifp->if_capabilities;
606 
607 	/* Set up MII bus. */
608 	if ((error = mii_phy_probe(dev, &sc->ale_miibus, ale_mediachange,
609 	    ale_mediastatus)) != 0) {
610 		device_printf(dev, "no PHY found!\n");
611 		goto fail;
612 	}
613 
614 	ether_ifattach(ifp, sc->ale_eaddr, NULL);
615 
616 	/* Tell the upper layer(s) we support long frames. */
617 	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
618 
619 	error = bus_setup_intr(dev, sc->ale_irq_res, INTR_MPSAFE, ale_intr, sc,
620 			       &sc->ale_irq_handle, ifp->if_serializer);
621 	if (error) {
622 		device_printf(dev, "could not set up interrupt handler.\n");
623 		ether_ifdetach(ifp);
624 		goto fail;
625 	}
626 
627 	ifp->if_cpuid = ithread_cpuid(rman_get_start(sc->ale_irq_res));
628 	KKASSERT(ifp->if_cpuid >= 0 && ifp->if_cpuid < ncpus);
629 	return 0;
630 fail:
631 	ale_detach(dev);
632 	return (error);
633 }
634 
635 static int
636 ale_detach(device_t dev)
637 {
638 	struct ale_softc *sc = device_get_softc(dev);
639 
640 	if (device_is_attached(dev)) {
641 		struct ifnet *ifp = &sc->arpcom.ac_if;
642 
643 		lwkt_serialize_enter(ifp->if_serializer);
644 		sc->ale_flags |= ALE_FLAG_DETACH;
645 		ale_stop(sc);
646 		bus_teardown_intr(dev, sc->ale_irq_res, sc->ale_irq_handle);
647 		lwkt_serialize_exit(ifp->if_serializer);
648 
649 		ether_ifdetach(ifp);
650 	}
651 
652 	if (sc->ale_sysctl_tree != NULL)
653 		sysctl_ctx_free(&sc->ale_sysctl_ctx);
654 
655 	if (sc->ale_miibus != NULL)
656 		device_delete_child(dev, sc->ale_miibus);
657 	bus_generic_detach(dev);
658 
659 	if (sc->ale_irq_res != NULL) {
660 		bus_release_resource(dev, SYS_RES_IRQ, sc->ale_irq_rid,
661 				     sc->ale_irq_res);
662 	}
663 	if (sc->ale_mem_res != NULL) {
664 		bus_release_resource(dev, SYS_RES_MEMORY, sc->ale_mem_rid,
665 				     sc->ale_mem_res);
666 	}
667 
668 	ale_dma_free(sc);
669 
670 	return (0);
671 }
672 
673 #define	ALE_SYSCTL_STAT_ADD32(c, h, n, p, d)	\
674 	    SYSCTL_ADD_UINT(c, h, OID_AUTO, n, CTLFLAG_RD, p, 0, d)
675 #define	ALE_SYSCTL_STAT_ADD64(c, h, n, p, d)	\
676 	    SYSCTL_ADD_QUAD(c, h, OID_AUTO, n, CTLFLAG_RD, p, 0, d)
677 
678 static void
679 ale_sysctl_node(struct ale_softc *sc)
680 {
681 	struct sysctl_ctx_list *ctx;
682 	struct sysctl_oid_list *child, *parent;
683 	struct sysctl_oid *tree;
684 	struct ale_hw_stats *stats;
685 	int error;
686 
687 	sysctl_ctx_init(&sc->ale_sysctl_ctx);
688 	sc->ale_sysctl_tree = SYSCTL_ADD_NODE(&sc->ale_sysctl_ctx,
689 				SYSCTL_STATIC_CHILDREN(_hw), OID_AUTO,
690 				device_get_nameunit(sc->ale_dev),
691 				CTLFLAG_RD, 0, "");
692 	if (sc->ale_sysctl_tree == NULL) {
693 		device_printf(sc->ale_dev, "can't add sysctl node\n");
694 		return;
695 	}
696 
697 	stats = &sc->ale_stats;
698 	ctx = &sc->ale_sysctl_ctx;
699 	child = SYSCTL_CHILDREN(sc->ale_sysctl_tree);
700 
701 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "int_rx_mod",
702 	    CTLTYPE_INT | CTLFLAG_RW, &sc->ale_int_rx_mod, 0,
703 	    sysctl_hw_ale_int_mod, "I", "ale Rx interrupt moderation");
704 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "int_tx_mod",
705 	    CTLTYPE_INT | CTLFLAG_RW, &sc->ale_int_tx_mod, 0,
706 	    sysctl_hw_ale_int_mod, "I", "ale Tx interrupt moderation");
707 
708 	/*
709 	 * Pull in device tunables.
710 	 */
711 	sc->ale_int_rx_mod = ALE_IM_RX_TIMER_DEFAULT;
712 	error = resource_int_value(device_get_name(sc->ale_dev),
713 	    device_get_unit(sc->ale_dev), "int_rx_mod", &sc->ale_int_rx_mod);
714 	if (error == 0) {
715 		if (sc->ale_int_rx_mod < ALE_IM_TIMER_MIN ||
716 		    sc->ale_int_rx_mod > ALE_IM_TIMER_MAX) {
717 			device_printf(sc->ale_dev, "int_rx_mod value out of "
718 			    "range; using default: %d\n",
719 			    ALE_IM_RX_TIMER_DEFAULT);
720 			sc->ale_int_rx_mod = ALE_IM_RX_TIMER_DEFAULT;
721 		}
722 	}
723 
724 	sc->ale_int_tx_mod = ALE_IM_TX_TIMER_DEFAULT;
725 	error = resource_int_value(device_get_name(sc->ale_dev),
726 	    device_get_unit(sc->ale_dev), "int_tx_mod", &sc->ale_int_tx_mod);
727 	if (error == 0) {
728 		if (sc->ale_int_tx_mod < ALE_IM_TIMER_MIN ||
729 		    sc->ale_int_tx_mod > ALE_IM_TIMER_MAX) {
730 			device_printf(sc->ale_dev, "int_tx_mod value out of "
731 			    "range; using default: %d\n",
732 			    ALE_IM_TX_TIMER_DEFAULT);
733 			sc->ale_int_tx_mod = ALE_IM_TX_TIMER_DEFAULT;
734 		}
735 	}
736 
737 	/* Misc statistics. */
738 	ALE_SYSCTL_STAT_ADD32(ctx, child, "reset_brk_seq",
739 	    &stats->reset_brk_seq,
740 	    "Controller resets due to broken Rx sequnce number");
741 
742 	tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "stats", CTLFLAG_RD,
743 	    NULL, "ATE statistics");
744 	parent = SYSCTL_CHILDREN(tree);
745 
746 	/* Rx statistics. */
747 	tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "rx", CTLFLAG_RD,
748 	    NULL, "Rx MAC statistics");
749 	child = SYSCTL_CHILDREN(tree);
750 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_frames",
751 	    &stats->rx_frames, "Good frames");
752 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_bcast_frames",
753 	    &stats->rx_bcast_frames, "Good broadcast frames");
754 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_mcast_frames",
755 	    &stats->rx_mcast_frames, "Good multicast frames");
756 	ALE_SYSCTL_STAT_ADD32(ctx, child, "pause_frames",
757 	    &stats->rx_pause_frames, "Pause control frames");
758 	ALE_SYSCTL_STAT_ADD32(ctx, child, "control_frames",
759 	    &stats->rx_control_frames, "Control frames");
760 	ALE_SYSCTL_STAT_ADD32(ctx, child, "crc_errs",
761 	    &stats->rx_crcerrs, "CRC errors");
762 	ALE_SYSCTL_STAT_ADD32(ctx, child, "len_errs",
763 	    &stats->rx_lenerrs, "Frames with length mismatched");
764 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_octets",
765 	    &stats->rx_bytes, "Good octets");
766 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_bcast_octets",
767 	    &stats->rx_bcast_bytes, "Good broadcast octets");
768 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_mcast_octets",
769 	    &stats->rx_mcast_bytes, "Good multicast octets");
770 	ALE_SYSCTL_STAT_ADD32(ctx, child, "runts",
771 	    &stats->rx_runts, "Too short frames");
772 	ALE_SYSCTL_STAT_ADD32(ctx, child, "fragments",
773 	    &stats->rx_fragments, "Fragmented frames");
774 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_64",
775 	    &stats->rx_pkts_64, "64 bytes frames");
776 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_65_127",
777 	    &stats->rx_pkts_65_127, "65 to 127 bytes frames");
778 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_128_255",
779 	    &stats->rx_pkts_128_255, "128 to 255 bytes frames");
780 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_256_511",
781 	    &stats->rx_pkts_256_511, "256 to 511 bytes frames");
782 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_512_1023",
783 	    &stats->rx_pkts_512_1023, "512 to 1023 bytes frames");
784 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1024_1518",
785 	    &stats->rx_pkts_1024_1518, "1024 to 1518 bytes frames");
786 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1519_max",
787 	    &stats->rx_pkts_1519_max, "1519 to max frames");
788 	ALE_SYSCTL_STAT_ADD32(ctx, child, "trunc_errs",
789 	    &stats->rx_pkts_truncated, "Truncated frames due to MTU size");
790 	ALE_SYSCTL_STAT_ADD32(ctx, child, "fifo_oflows",
791 	    &stats->rx_fifo_oflows, "FIFO overflows");
792 	ALE_SYSCTL_STAT_ADD32(ctx, child, "rrs_errs",
793 	    &stats->rx_rrs_errs, "Return status write-back errors");
794 	ALE_SYSCTL_STAT_ADD32(ctx, child, "align_errs",
795 	    &stats->rx_alignerrs, "Alignment errors");
796 	ALE_SYSCTL_STAT_ADD32(ctx, child, "filtered",
797 	    &stats->rx_pkts_filtered,
798 	    "Frames dropped due to address filtering");
799 
800 	/* Tx statistics. */
801 	tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "tx", CTLFLAG_RD,
802 	    NULL, "Tx MAC statistics");
803 	child = SYSCTL_CHILDREN(tree);
804 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_frames",
805 	    &stats->tx_frames, "Good frames");
806 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_bcast_frames",
807 	    &stats->tx_bcast_frames, "Good broadcast frames");
808 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_mcast_frames",
809 	    &stats->tx_mcast_frames, "Good multicast frames");
810 	ALE_SYSCTL_STAT_ADD32(ctx, child, "pause_frames",
811 	    &stats->tx_pause_frames, "Pause control frames");
812 	ALE_SYSCTL_STAT_ADD32(ctx, child, "control_frames",
813 	    &stats->tx_control_frames, "Control frames");
814 	ALE_SYSCTL_STAT_ADD32(ctx, child, "excess_defers",
815 	    &stats->tx_excess_defer, "Frames with excessive derferrals");
816 	ALE_SYSCTL_STAT_ADD32(ctx, child, "defers",
817 	    &stats->tx_excess_defer, "Frames with derferrals");
818 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_octets",
819 	    &stats->tx_bytes, "Good octets");
820 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_bcast_octets",
821 	    &stats->tx_bcast_bytes, "Good broadcast octets");
822 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_mcast_octets",
823 	    &stats->tx_mcast_bytes, "Good multicast octets");
824 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_64",
825 	    &stats->tx_pkts_64, "64 bytes frames");
826 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_65_127",
827 	    &stats->tx_pkts_65_127, "65 to 127 bytes frames");
828 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_128_255",
829 	    &stats->tx_pkts_128_255, "128 to 255 bytes frames");
830 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_256_511",
831 	    &stats->tx_pkts_256_511, "256 to 511 bytes frames");
832 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_512_1023",
833 	    &stats->tx_pkts_512_1023, "512 to 1023 bytes frames");
834 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1024_1518",
835 	    &stats->tx_pkts_1024_1518, "1024 to 1518 bytes frames");
836 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1519_max",
837 	    &stats->tx_pkts_1519_max, "1519 to max frames");
838 	ALE_SYSCTL_STAT_ADD32(ctx, child, "single_colls",
839 	    &stats->tx_single_colls, "Single collisions");
840 	ALE_SYSCTL_STAT_ADD32(ctx, child, "multi_colls",
841 	    &stats->tx_multi_colls, "Multiple collisions");
842 	ALE_SYSCTL_STAT_ADD32(ctx, child, "late_colls",
843 	    &stats->tx_late_colls, "Late collisions");
844 	ALE_SYSCTL_STAT_ADD32(ctx, child, "excess_colls",
845 	    &stats->tx_excess_colls, "Excessive collisions");
846 	ALE_SYSCTL_STAT_ADD32(ctx, child, "abort",
847 	    &stats->tx_abort, "Aborted frames due to Excessive collisions");
848 	ALE_SYSCTL_STAT_ADD32(ctx, child, "underruns",
849 	    &stats->tx_underrun, "FIFO underruns");
850 	ALE_SYSCTL_STAT_ADD32(ctx, child, "desc_underruns",
851 	    &stats->tx_desc_underrun, "Descriptor write-back errors");
852 	ALE_SYSCTL_STAT_ADD32(ctx, child, "len_errs",
853 	    &stats->tx_lenerrs, "Frames with length mismatched");
854 	ALE_SYSCTL_STAT_ADD32(ctx, child, "trunc_errs",
855 	    &stats->tx_pkts_truncated, "Truncated frames due to MTU size");
856 }
857 
858 #undef ALE_SYSCTL_STAT_ADD32
859 #undef ALE_SYSCTL_STAT_ADD64
860 
861 struct ale_dmamap_arg {
862 	bus_addr_t	ale_busaddr;
863 };
864 
865 static void
866 ale_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error)
867 {
868 	struct ale_dmamap_arg *ctx;
869 
870 	if (error != 0)
871 		return;
872 
873 	KASSERT(nsegs == 1, ("%s: %d segments returned!", __func__, nsegs));
874 
875 	ctx = (struct ale_dmamap_arg *)arg;
876 	ctx->ale_busaddr = segs[0].ds_addr;
877 }
878 
879 /*
880  * Tx descriptors/RXF0/CMB DMA blocks share ALE_DESC_ADDR_HI register
881  * which specifies high address region of DMA blocks. Therefore these
882  * blocks should have the same high address of given 4GB address
883  * space(i.e. crossing 4GB boundary is not allowed).
884  */
885 static int
886 ale_check_boundary(struct ale_softc *sc)
887 {
888 	bus_addr_t rx_cmb_end[ALE_RX_PAGES], tx_cmb_end;
889 	bus_addr_t rx_page_end[ALE_RX_PAGES], tx_ring_end;
890 
891 	rx_page_end[0] = sc->ale_cdata.ale_rx_page[0].page_paddr +
892 	    sc->ale_pagesize;
893 	rx_page_end[1] = sc->ale_cdata.ale_rx_page[1].page_paddr +
894 	    sc->ale_pagesize;
895 	tx_ring_end = sc->ale_cdata.ale_tx_ring_paddr + ALE_TX_RING_SZ;
896 	tx_cmb_end = sc->ale_cdata.ale_tx_cmb_paddr + ALE_TX_CMB_SZ;
897 	rx_cmb_end[0] = sc->ale_cdata.ale_rx_page[0].cmb_paddr + ALE_RX_CMB_SZ;
898 	rx_cmb_end[1] = sc->ale_cdata.ale_rx_page[1].cmb_paddr + ALE_RX_CMB_SZ;
899 
900 	if ((ALE_ADDR_HI(tx_ring_end) !=
901 	    ALE_ADDR_HI(sc->ale_cdata.ale_tx_ring_paddr)) ||
902 	    (ALE_ADDR_HI(rx_page_end[0]) !=
903 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[0].page_paddr)) ||
904 	    (ALE_ADDR_HI(rx_page_end[1]) !=
905 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[1].page_paddr)) ||
906 	    (ALE_ADDR_HI(tx_cmb_end) !=
907 	    ALE_ADDR_HI(sc->ale_cdata.ale_tx_cmb_paddr)) ||
908 	    (ALE_ADDR_HI(rx_cmb_end[0]) !=
909 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[0].cmb_paddr)) ||
910 	    (ALE_ADDR_HI(rx_cmb_end[1]) !=
911 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[1].cmb_paddr)))
912 		return (EFBIG);
913 
914 	if ((ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_page_end[0])) ||
915 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_page_end[1])) ||
916 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_cmb_end[0])) ||
917 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_cmb_end[1])) ||
918 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(tx_cmb_end)))
919 		return (EFBIG);
920 
921 	return (0);
922 }
923 
924 static int
925 ale_dma_alloc(struct ale_softc *sc)
926 {
927 	struct ale_txdesc *txd;
928 	bus_addr_t lowaddr;
929 	struct ale_dmamap_arg ctx;
930 	int error, guard_size, i;
931 
932 	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0)
933 		guard_size = ALE_JUMBO_FRAMELEN;
934 	else
935 		guard_size = ALE_MAX_FRAMELEN;
936 	sc->ale_pagesize = roundup(guard_size + ALE_RX_PAGE_SZ,
937 	    ALE_RX_PAGE_ALIGN);
938 	lowaddr = BUS_SPACE_MAXADDR;
939 again:
940 	/* Create parent DMA tag. */
941 	error = bus_dma_tag_create(
942 	    NULL,			/* parent */
943 	    1, 0,			/* alignment, boundary */
944 	    lowaddr,			/* lowaddr */
945 	    BUS_SPACE_MAXADDR,		/* highaddr */
946 	    NULL, NULL,			/* filter, filterarg */
947 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsize */
948 	    0,				/* nsegments */
949 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsegsize */
950 	    0,				/* flags */
951 	    &sc->ale_cdata.ale_parent_tag);
952 	if (error != 0) {
953 		device_printf(sc->ale_dev,
954 		    "could not create parent DMA tag.\n");
955 		goto fail;
956 	}
957 
958 	/* Create DMA tag for Tx descriptor ring. */
959 	error = bus_dma_tag_create(
960 	    sc->ale_cdata.ale_parent_tag, /* parent */
961 	    ALE_TX_RING_ALIGN, 0,	/* alignment, boundary */
962 	    BUS_SPACE_MAXADDR,		/* lowaddr */
963 	    BUS_SPACE_MAXADDR,		/* highaddr */
964 	    NULL, NULL,			/* filter, filterarg */
965 	    ALE_TX_RING_SZ,		/* maxsize */
966 	    1,				/* nsegments */
967 	    ALE_TX_RING_SZ,		/* maxsegsize */
968 	    0,				/* flags */
969 	    &sc->ale_cdata.ale_tx_ring_tag);
970 	if (error != 0) {
971 		device_printf(sc->ale_dev,
972 		    "could not create Tx ring DMA tag.\n");
973 		goto fail;
974 	}
975 
976 	/* Create DMA tag for Rx pages. */
977 	for (i = 0; i < ALE_RX_PAGES; i++) {
978 		error = bus_dma_tag_create(
979 		    sc->ale_cdata.ale_parent_tag, /* parent */
980 		    ALE_RX_PAGE_ALIGN, 0,	/* alignment, boundary */
981 		    BUS_SPACE_MAXADDR,		/* lowaddr */
982 		    BUS_SPACE_MAXADDR,		/* highaddr */
983 		    NULL, NULL,			/* filter, filterarg */
984 		    sc->ale_pagesize,		/* maxsize */
985 		    1,				/* nsegments */
986 		    sc->ale_pagesize,		/* maxsegsize */
987 		    0,				/* flags */
988 		    &sc->ale_cdata.ale_rx_page[i].page_tag);
989 		if (error != 0) {
990 			device_printf(sc->ale_dev,
991 			    "could not create Rx page %d DMA tag.\n", i);
992 			goto fail;
993 		}
994 	}
995 
996 	/* Create DMA tag for Tx coalescing message block. */
997 	error = bus_dma_tag_create(
998 	    sc->ale_cdata.ale_parent_tag, /* parent */
999 	    ALE_CMB_ALIGN, 0,		/* alignment, boundary */
1000 	    BUS_SPACE_MAXADDR,		/* lowaddr */
1001 	    BUS_SPACE_MAXADDR,		/* highaddr */
1002 	    NULL, NULL,			/* filter, filterarg */
1003 	    ALE_TX_CMB_SZ,		/* maxsize */
1004 	    1,				/* nsegments */
1005 	    ALE_TX_CMB_SZ,		/* maxsegsize */
1006 	    0,				/* flags */
1007 	    &sc->ale_cdata.ale_tx_cmb_tag);
1008 	if (error != 0) {
1009 		device_printf(sc->ale_dev,
1010 		    "could not create Tx CMB DMA tag.\n");
1011 		goto fail;
1012 	}
1013 
1014 	/* Create DMA tag for Rx coalescing message block. */
1015 	for (i = 0; i < ALE_RX_PAGES; i++) {
1016 		error = bus_dma_tag_create(
1017 		    sc->ale_cdata.ale_parent_tag, /* parent */
1018 		    ALE_CMB_ALIGN, 0,		/* alignment, boundary */
1019 		    BUS_SPACE_MAXADDR,		/* lowaddr */
1020 		    BUS_SPACE_MAXADDR,		/* highaddr */
1021 		    NULL, NULL,			/* filter, filterarg */
1022 		    ALE_RX_CMB_SZ,		/* maxsize */
1023 		    1,				/* nsegments */
1024 		    ALE_RX_CMB_SZ,		/* maxsegsize */
1025 		    0,				/* flags */
1026 		    &sc->ale_cdata.ale_rx_page[i].cmb_tag);
1027 		if (error != 0) {
1028 			device_printf(sc->ale_dev,
1029 			    "could not create Rx page %d CMB DMA tag.\n", i);
1030 			goto fail;
1031 		}
1032 	}
1033 
1034 	/* Allocate DMA'able memory and load the DMA map for Tx ring. */
1035 	error = bus_dmamem_alloc(sc->ale_cdata.ale_tx_ring_tag,
1036 	    (void **)&sc->ale_cdata.ale_tx_ring,
1037 	    BUS_DMA_WAITOK | BUS_DMA_ZERO,
1038 	    &sc->ale_cdata.ale_tx_ring_map);
1039 	if (error != 0) {
1040 		device_printf(sc->ale_dev,
1041 		    "could not allocate DMA'able memory for Tx ring.\n");
1042 		goto fail;
1043 	}
1044 	ctx.ale_busaddr = 0;
1045 	error = bus_dmamap_load(sc->ale_cdata.ale_tx_ring_tag,
1046 	    sc->ale_cdata.ale_tx_ring_map, sc->ale_cdata.ale_tx_ring,
1047 	    ALE_TX_RING_SZ, ale_dmamap_cb, &ctx, 0);
1048 	if (error != 0 || ctx.ale_busaddr == 0) {
1049 		device_printf(sc->ale_dev,
1050 		    "could not load DMA'able memory for Tx ring.\n");
1051 		goto fail;
1052 	}
1053 	sc->ale_cdata.ale_tx_ring_paddr = ctx.ale_busaddr;
1054 
1055 	/* Rx pages. */
1056 	for (i = 0; i < ALE_RX_PAGES; i++) {
1057 		error = bus_dmamem_alloc(sc->ale_cdata.ale_rx_page[i].page_tag,
1058 		    (void **)&sc->ale_cdata.ale_rx_page[i].page_addr,
1059 		    BUS_DMA_WAITOK | BUS_DMA_ZERO,
1060 		    &sc->ale_cdata.ale_rx_page[i].page_map);
1061 		if (error != 0) {
1062 			device_printf(sc->ale_dev,
1063 			    "could not allocate DMA'able memory for "
1064 			    "Rx page %d.\n", i);
1065 			goto fail;
1066 		}
1067 		ctx.ale_busaddr = 0;
1068 		error = bus_dmamap_load(sc->ale_cdata.ale_rx_page[i].page_tag,
1069 		    sc->ale_cdata.ale_rx_page[i].page_map,
1070 		    sc->ale_cdata.ale_rx_page[i].page_addr,
1071 		    sc->ale_pagesize, ale_dmamap_cb, &ctx, 0);
1072 		if (error != 0 || ctx.ale_busaddr == 0) {
1073 			device_printf(sc->ale_dev,
1074 			    "could not load DMA'able memory for "
1075 			    "Rx page %d.\n", i);
1076 			goto fail;
1077 		}
1078 		sc->ale_cdata.ale_rx_page[i].page_paddr = ctx.ale_busaddr;
1079 	}
1080 
1081 	/* Tx CMB. */
1082 	error = bus_dmamem_alloc(sc->ale_cdata.ale_tx_cmb_tag,
1083 	    (void **)&sc->ale_cdata.ale_tx_cmb,
1084 	    BUS_DMA_WAITOK | BUS_DMA_ZERO,
1085 	    &sc->ale_cdata.ale_tx_cmb_map);
1086 	if (error != 0) {
1087 		device_printf(sc->ale_dev,
1088 		    "could not allocate DMA'able memory for Tx CMB.\n");
1089 		goto fail;
1090 	}
1091 	ctx.ale_busaddr = 0;
1092 	error = bus_dmamap_load(sc->ale_cdata.ale_tx_cmb_tag,
1093 	    sc->ale_cdata.ale_tx_cmb_map, sc->ale_cdata.ale_tx_cmb,
1094 	    ALE_TX_CMB_SZ, ale_dmamap_cb, &ctx, 0);
1095 	if (error != 0 || ctx.ale_busaddr == 0) {
1096 		device_printf(sc->ale_dev,
1097 		    "could not load DMA'able memory for Tx CMB.\n");
1098 		goto fail;
1099 	}
1100 	sc->ale_cdata.ale_tx_cmb_paddr = ctx.ale_busaddr;
1101 
1102 	/* Rx CMB. */
1103 	for (i = 0; i < ALE_RX_PAGES; i++) {
1104 		error = bus_dmamem_alloc(sc->ale_cdata.ale_rx_page[i].cmb_tag,
1105 		    (void **)&sc->ale_cdata.ale_rx_page[i].cmb_addr,
1106 		    BUS_DMA_WAITOK | BUS_DMA_ZERO,
1107 		    &sc->ale_cdata.ale_rx_page[i].cmb_map);
1108 		if (error != 0) {
1109 			device_printf(sc->ale_dev, "could not allocate "
1110 			    "DMA'able memory for Rx page %d CMB.\n", i);
1111 			goto fail;
1112 		}
1113 		ctx.ale_busaddr = 0;
1114 		error = bus_dmamap_load(sc->ale_cdata.ale_rx_page[i].cmb_tag,
1115 		    sc->ale_cdata.ale_rx_page[i].cmb_map,
1116 		    sc->ale_cdata.ale_rx_page[i].cmb_addr,
1117 		    ALE_RX_CMB_SZ, ale_dmamap_cb, &ctx, 0);
1118 		if (error != 0 || ctx.ale_busaddr == 0) {
1119 			device_printf(sc->ale_dev, "could not load DMA'able "
1120 			    "memory for Rx page %d CMB.\n", i);
1121 			goto fail;
1122 		}
1123 		sc->ale_cdata.ale_rx_page[i].cmb_paddr = ctx.ale_busaddr;
1124 	}
1125 
1126 	/*
1127 	 * Tx descriptors/RXF0/CMB DMA blocks share the same
1128 	 * high address region of 64bit DMA address space.
1129 	 */
1130 	if (lowaddr != BUS_SPACE_MAXADDR_32BIT &&
1131 	    (error = ale_check_boundary(sc)) != 0) {
1132 		device_printf(sc->ale_dev, "4GB boundary crossed, "
1133 		    "switching to 32bit DMA addressing mode.\n");
1134 		ale_dma_free(sc);
1135 		/*
1136 		 * Limit max allowable DMA address space to 32bit
1137 		 * and try again.
1138 		 */
1139 		lowaddr = BUS_SPACE_MAXADDR_32BIT;
1140 		goto again;
1141 	}
1142 
1143 	/*
1144 	 * Create Tx buffer parent tag.
1145 	 * AR81xx allows 64bit DMA addressing of Tx buffers so it
1146 	 * needs separate parent DMA tag as parent DMA address space
1147 	 * could be restricted to be within 32bit address space by
1148 	 * 4GB boundary crossing.
1149 	 */
1150 	error = bus_dma_tag_create(
1151 	    NULL,			/* parent */
1152 	    1, 0,			/* alignment, boundary */
1153 	    BUS_SPACE_MAXADDR,		/* lowaddr */
1154 	    BUS_SPACE_MAXADDR,		/* highaddr */
1155 	    NULL, NULL,			/* filter, filterarg */
1156 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsize */
1157 	    0,				/* nsegments */
1158 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsegsize */
1159 	    0,				/* flags */
1160 	    &sc->ale_cdata.ale_buffer_tag);
1161 	if (error != 0) {
1162 		device_printf(sc->ale_dev,
1163 		    "could not create parent buffer DMA tag.\n");
1164 		goto fail;
1165 	}
1166 
1167 	/* Create DMA tag for Tx buffers. */
1168 	error = bus_dma_tag_create(
1169 	    sc->ale_cdata.ale_buffer_tag, /* parent */
1170 	    1, 0,			/* alignment, boundary */
1171 	    BUS_SPACE_MAXADDR,		/* lowaddr */
1172 	    BUS_SPACE_MAXADDR,		/* highaddr */
1173 	    NULL, NULL,			/* filter, filterarg */
1174 	    ALE_TSO_MAXSIZE,		/* maxsize */
1175 	    ALE_MAXTXSEGS,		/* nsegments */
1176 	    ALE_TSO_MAXSEGSIZE,		/* maxsegsize */
1177 	    0,				/* flags */
1178 	    &sc->ale_cdata.ale_tx_tag);
1179 	if (error != 0) {
1180 		device_printf(sc->ale_dev, "could not create Tx DMA tag.\n");
1181 		goto fail;
1182 	}
1183 
1184 	/* Create DMA maps for Tx buffers. */
1185 	for (i = 0; i < ALE_TX_RING_CNT; i++) {
1186 		txd = &sc->ale_cdata.ale_txdesc[i];
1187 		txd->tx_m = NULL;
1188 		txd->tx_dmamap = NULL;
1189 		error = bus_dmamap_create(sc->ale_cdata.ale_tx_tag, 0,
1190 		    &txd->tx_dmamap);
1191 		if (error != 0) {
1192 			device_printf(sc->ale_dev,
1193 			    "could not create Tx dmamap.\n");
1194 			goto fail;
1195 		}
1196 	}
1197 fail:
1198 	return (error);
1199 }
1200 
1201 static void
1202 ale_dma_free(struct ale_softc *sc)
1203 {
1204 	struct ale_txdesc *txd;
1205 	int i;
1206 
1207 	/* Tx buffers. */
1208 	if (sc->ale_cdata.ale_tx_tag != NULL) {
1209 		for (i = 0; i < ALE_TX_RING_CNT; i++) {
1210 			txd = &sc->ale_cdata.ale_txdesc[i];
1211 			if (txd->tx_dmamap != NULL) {
1212 				bus_dmamap_destroy(sc->ale_cdata.ale_tx_tag,
1213 				    txd->tx_dmamap);
1214 				txd->tx_dmamap = NULL;
1215 			}
1216 		}
1217 		bus_dma_tag_destroy(sc->ale_cdata.ale_tx_tag);
1218 		sc->ale_cdata.ale_tx_tag = NULL;
1219 	}
1220 	/* Tx descriptor ring. */
1221 	if (sc->ale_cdata.ale_tx_ring_tag != NULL) {
1222 		if (sc->ale_cdata.ale_tx_ring_map != NULL)
1223 			bus_dmamap_unload(sc->ale_cdata.ale_tx_ring_tag,
1224 			    sc->ale_cdata.ale_tx_ring_map);
1225 		if (sc->ale_cdata.ale_tx_ring_map != NULL &&
1226 		    sc->ale_cdata.ale_tx_ring != NULL)
1227 			bus_dmamem_free(sc->ale_cdata.ale_tx_ring_tag,
1228 			    sc->ale_cdata.ale_tx_ring,
1229 			    sc->ale_cdata.ale_tx_ring_map);
1230 		sc->ale_cdata.ale_tx_ring = NULL;
1231 		sc->ale_cdata.ale_tx_ring_map = NULL;
1232 		bus_dma_tag_destroy(sc->ale_cdata.ale_tx_ring_tag);
1233 		sc->ale_cdata.ale_tx_ring_tag = NULL;
1234 	}
1235 	/* Rx page block. */
1236 	for (i = 0; i < ALE_RX_PAGES; i++) {
1237 		if (sc->ale_cdata.ale_rx_page[i].page_tag != NULL) {
1238 			if (sc->ale_cdata.ale_rx_page[i].page_map != NULL)
1239 				bus_dmamap_unload(
1240 				    sc->ale_cdata.ale_rx_page[i].page_tag,
1241 				    sc->ale_cdata.ale_rx_page[i].page_map);
1242 			if (sc->ale_cdata.ale_rx_page[i].page_map != NULL &&
1243 			    sc->ale_cdata.ale_rx_page[i].page_addr != NULL)
1244 				bus_dmamem_free(
1245 				    sc->ale_cdata.ale_rx_page[i].page_tag,
1246 				    sc->ale_cdata.ale_rx_page[i].page_addr,
1247 				    sc->ale_cdata.ale_rx_page[i].page_map);
1248 			sc->ale_cdata.ale_rx_page[i].page_addr = NULL;
1249 			sc->ale_cdata.ale_rx_page[i].page_map = NULL;
1250 			bus_dma_tag_destroy(
1251 			    sc->ale_cdata.ale_rx_page[i].page_tag);
1252 			sc->ale_cdata.ale_rx_page[i].page_tag = NULL;
1253 		}
1254 	}
1255 	/* Rx CMB. */
1256 	for (i = 0; i < ALE_RX_PAGES; i++) {
1257 		if (sc->ale_cdata.ale_rx_page[i].cmb_tag != NULL) {
1258 			if (sc->ale_cdata.ale_rx_page[i].cmb_map != NULL)
1259 				bus_dmamap_unload(
1260 				    sc->ale_cdata.ale_rx_page[i].cmb_tag,
1261 				    sc->ale_cdata.ale_rx_page[i].cmb_map);
1262 			if (sc->ale_cdata.ale_rx_page[i].cmb_map != NULL &&
1263 			    sc->ale_cdata.ale_rx_page[i].cmb_addr != NULL)
1264 				bus_dmamem_free(
1265 				    sc->ale_cdata.ale_rx_page[i].cmb_tag,
1266 				    sc->ale_cdata.ale_rx_page[i].cmb_addr,
1267 				    sc->ale_cdata.ale_rx_page[i].cmb_map);
1268 			sc->ale_cdata.ale_rx_page[i].cmb_addr = NULL;
1269 			sc->ale_cdata.ale_rx_page[i].cmb_map = NULL;
1270 			bus_dma_tag_destroy(
1271 			    sc->ale_cdata.ale_rx_page[i].cmb_tag);
1272 			sc->ale_cdata.ale_rx_page[i].cmb_tag = NULL;
1273 		}
1274 	}
1275 	/* Tx CMB. */
1276 	if (sc->ale_cdata.ale_tx_cmb_tag != NULL) {
1277 		if (sc->ale_cdata.ale_tx_cmb_map != NULL)
1278 			bus_dmamap_unload(sc->ale_cdata.ale_tx_cmb_tag,
1279 			    sc->ale_cdata.ale_tx_cmb_map);
1280 		if (sc->ale_cdata.ale_tx_cmb_map != NULL &&
1281 		    sc->ale_cdata.ale_tx_cmb != NULL)
1282 			bus_dmamem_free(sc->ale_cdata.ale_tx_cmb_tag,
1283 			    sc->ale_cdata.ale_tx_cmb,
1284 			    sc->ale_cdata.ale_tx_cmb_map);
1285 		sc->ale_cdata.ale_tx_cmb = NULL;
1286 		sc->ale_cdata.ale_tx_cmb_map = NULL;
1287 		bus_dma_tag_destroy(sc->ale_cdata.ale_tx_cmb_tag);
1288 		sc->ale_cdata.ale_tx_cmb_tag = NULL;
1289 	}
1290 	if (sc->ale_cdata.ale_buffer_tag != NULL) {
1291 		bus_dma_tag_destroy(sc->ale_cdata.ale_buffer_tag);
1292 		sc->ale_cdata.ale_buffer_tag = NULL;
1293 	}
1294 	if (sc->ale_cdata.ale_parent_tag != NULL) {
1295 		bus_dma_tag_destroy(sc->ale_cdata.ale_parent_tag);
1296 		sc->ale_cdata.ale_parent_tag = NULL;
1297 	}
1298 }
1299 
1300 static int
1301 ale_shutdown(device_t dev)
1302 {
1303 	return (ale_suspend(dev));
1304 }
1305 
1306 #ifdef notyet
1307 
1308 /*
1309  * Note, this driver resets the link speed to 10/100Mbps by
1310  * restarting auto-negotiation in suspend/shutdown phase but we
1311  * don't know whether that auto-negotiation would succeed or not
1312  * as driver has no control after powering off/suspend operation.
1313  * If the renegotiation fail WOL may not work. Running at 1Gbps
1314  * will draw more power than 375mA at 3.3V which is specified in
1315  * PCI specification and that would result in complete
1316  * shutdowning power to ethernet controller.
1317  *
1318  * TODO
1319  * Save current negotiated media speed/duplex/flow-control to
1320  * softc and restore the same link again after resuming. PHY
1321  * handling such as power down/resetting to 100Mbps may be better
1322  * handled in suspend method in phy driver.
1323  */
1324 static void
1325 ale_setlinkspeed(struct ale_softc *sc)
1326 {
1327 	struct mii_data *mii;
1328 	int aneg, i;
1329 
1330 	mii = device_get_softc(sc->ale_miibus);
1331 	mii_pollstat(mii);
1332 	aneg = 0;
1333 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
1334 	    (IFM_ACTIVE | IFM_AVALID)) {
1335 		switch IFM_SUBTYPE(mii->mii_media_active) {
1336 		case IFM_10_T:
1337 		case IFM_100_TX:
1338 			return;
1339 		case IFM_1000_T:
1340 			aneg++;
1341 			break;
1342 		default:
1343 			break;
1344 		}
1345 	}
1346 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr, MII_100T2CR, 0);
1347 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
1348 	    MII_ANAR, ANAR_TX_FD | ANAR_TX | ANAR_10_FD | ANAR_10 | ANAR_CSMA);
1349 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
1350 	    MII_BMCR, BMCR_RESET | BMCR_AUTOEN | BMCR_STARTNEG);
1351 	DELAY(1000);
1352 	if (aneg != 0) {
1353 		/*
1354 		 * Poll link state until ale(4) get a 10/100Mbps link.
1355 		 */
1356 		for (i = 0; i < MII_ANEGTICKS_GIGE; i++) {
1357 			mii_pollstat(mii);
1358 			if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID))
1359 			    == (IFM_ACTIVE | IFM_AVALID)) {
1360 				switch (IFM_SUBTYPE(
1361 				    mii->mii_media_active)) {
1362 				case IFM_10_T:
1363 				case IFM_100_TX:
1364 					ale_mac_config(sc);
1365 					return;
1366 				default:
1367 					break;
1368 				}
1369 			}
1370 			ALE_UNLOCK(sc);
1371 			pause("alelnk", hz);
1372 			ALE_LOCK(sc);
1373 		}
1374 		if (i == MII_ANEGTICKS_GIGE)
1375 			device_printf(sc->ale_dev,
1376 			    "establishing a link failed, WOL may not work!");
1377 	}
1378 	/*
1379 	 * No link, force MAC to have 100Mbps, full-duplex link.
1380 	 * This is the last resort and may/may not work.
1381 	 */
1382 	mii->mii_media_status = IFM_AVALID | IFM_ACTIVE;
1383 	mii->mii_media_active = IFM_ETHER | IFM_100_TX | IFM_FDX;
1384 	ale_mac_config(sc);
1385 }
1386 
1387 static void
1388 ale_setwol(struct ale_softc *sc)
1389 {
1390 	struct ifnet *ifp;
1391 	uint32_t reg, pmcs;
1392 	uint16_t pmstat;
1393 	int pmc;
1394 
1395 	ALE_LOCK_ASSERT(sc);
1396 
1397 	if (pci_find_extcap(sc->ale_dev, PCIY_PMG, &pmc) != 0) {
1398 		/* Disable WOL. */
1399 		CSR_WRITE_4(sc, ALE_WOL_CFG, 0);
1400 		reg = CSR_READ_4(sc, ALE_PCIE_PHYMISC);
1401 		reg |= PCIE_PHYMISC_FORCE_RCV_DET;
1402 		CSR_WRITE_4(sc, ALE_PCIE_PHYMISC, reg);
1403 		/* Force PHY power down. */
1404 		CSR_WRITE_2(sc, ALE_GPHY_CTRL,
1405 		    GPHY_CTRL_EXT_RESET | GPHY_CTRL_HIB_EN |
1406 		    GPHY_CTRL_HIB_PULSE | GPHY_CTRL_PHY_PLL_ON |
1407 		    GPHY_CTRL_SEL_ANA_RESET | GPHY_CTRL_PHY_IDDQ |
1408 		    GPHY_CTRL_PCLK_SEL_DIS | GPHY_CTRL_PWDOWN_HW);
1409 		return;
1410 	}
1411 
1412 	ifp = sc->ale_ifp;
1413 	if ((ifp->if_capenable & IFCAP_WOL) != 0) {
1414 		if ((sc->ale_flags & ALE_FLAG_FASTETHER) == 0)
1415 			ale_setlinkspeed(sc);
1416 	}
1417 
1418 	pmcs = 0;
1419 	if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
1420 		pmcs |= WOL_CFG_MAGIC | WOL_CFG_MAGIC_ENB;
1421 	CSR_WRITE_4(sc, ALE_WOL_CFG, pmcs);
1422 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
1423 	reg &= ~(MAC_CFG_DBG | MAC_CFG_PROMISC | MAC_CFG_ALLMULTI |
1424 	    MAC_CFG_BCAST);
1425 	if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
1426 		reg |= MAC_CFG_ALLMULTI | MAC_CFG_BCAST;
1427 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
1428 		reg |= MAC_CFG_RX_ENB;
1429 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
1430 
1431 	if ((ifp->if_capenable & IFCAP_WOL) == 0) {
1432 		/* WOL disabled, PHY power down. */
1433 		reg = CSR_READ_4(sc, ALE_PCIE_PHYMISC);
1434 		reg |= PCIE_PHYMISC_FORCE_RCV_DET;
1435 		CSR_WRITE_4(sc, ALE_PCIE_PHYMISC, reg);
1436 		CSR_WRITE_2(sc, ALE_GPHY_CTRL,
1437 		    GPHY_CTRL_EXT_RESET | GPHY_CTRL_HIB_EN |
1438 		    GPHY_CTRL_HIB_PULSE | GPHY_CTRL_SEL_ANA_RESET |
1439 		    GPHY_CTRL_PHY_IDDQ | GPHY_CTRL_PCLK_SEL_DIS |
1440 		    GPHY_CTRL_PWDOWN_HW);
1441 	}
1442 	/* Request PME. */
1443 	pmstat = pci_read_config(sc->ale_dev, pmc + PCIR_POWER_STATUS, 2);
1444 	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
1445 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
1446 		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
1447 	pci_write_config(sc->ale_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
1448 }
1449 
1450 #endif	/* notyet */
1451 
1452 static int
1453 ale_suspend(device_t dev)
1454 {
1455 	struct ale_softc *sc = device_get_softc(dev);
1456 	struct ifnet *ifp = &sc->arpcom.ac_if;
1457 
1458 	lwkt_serialize_enter(ifp->if_serializer);
1459 	ale_stop(sc);
1460 #ifdef notyet
1461 	ale_setwol(sc);
1462 #endif
1463 	lwkt_serialize_exit(ifp->if_serializer);
1464 	return (0);
1465 }
1466 
1467 static int
1468 ale_resume(device_t dev)
1469 {
1470 	struct ale_softc *sc = device_get_softc(dev);
1471 	struct ifnet *ifp = &sc->arpcom.ac_if;
1472 	uint16_t cmd;
1473 
1474 	lwkt_serialize_enter(ifp->if_serializer);
1475 
1476 	/*
1477 	 * Clear INTx emulation disable for hardwares that
1478 	 * is set in resume event. From Linux.
1479 	 */
1480 	cmd = pci_read_config(sc->ale_dev, PCIR_COMMAND, 2);
1481 	if ((cmd & 0x0400) != 0) {
1482 		cmd &= ~0x0400;
1483 		pci_write_config(sc->ale_dev, PCIR_COMMAND, cmd, 2);
1484 	}
1485 
1486 #ifdef notyet
1487 	if (pci_find_extcap(sc->ale_dev, PCIY_PMG, &pmc) == 0) {
1488 		uint16_t pmstat;
1489 		int pmc;
1490 
1491 		/* Disable PME and clear PME status. */
1492 		pmstat = pci_read_config(sc->ale_dev,
1493 		    pmc + PCIR_POWER_STATUS, 2);
1494 		if ((pmstat & PCIM_PSTAT_PMEENABLE) != 0) {
1495 			pmstat &= ~PCIM_PSTAT_PMEENABLE;
1496 			pci_write_config(sc->ale_dev,
1497 			    pmc + PCIR_POWER_STATUS, pmstat, 2);
1498 		}
1499 	}
1500 #endif
1501 
1502 	/* Reset PHY. */
1503 	ale_phy_reset(sc);
1504 	if ((ifp->if_flags & IFF_UP) != 0)
1505 		ale_init(sc);
1506 
1507 	lwkt_serialize_exit(ifp->if_serializer);
1508 	return (0);
1509 }
1510 
1511 static int
1512 ale_encap(struct ale_softc *sc, struct mbuf **m_head)
1513 {
1514 	struct ale_txdesc *txd, *txd_last;
1515 	struct tx_desc *desc;
1516 	struct mbuf *m;
1517 	bus_dma_segment_t txsegs[ALE_MAXTXSEGS];
1518 	struct ale_dmamap_ctx ctx;
1519 	bus_dmamap_t map;
1520 	uint32_t cflags, poff, vtag;
1521 	int error, i, nsegs, prod, si;
1522 
1523 	M_ASSERTPKTHDR((*m_head));
1524 
1525 	m = *m_head;
1526 	cflags = vtag = 0;
1527 	poff = 0;
1528 
1529 	si = prod = sc->ale_cdata.ale_tx_prod;
1530 	txd = &sc->ale_cdata.ale_txdesc[prod];
1531 	txd_last = txd;
1532 	map = txd->tx_dmamap;
1533 
1534 	ctx.nsegs = ALE_MAXTXSEGS;
1535 	ctx.segs = txsegs;
1536 	error =  bus_dmamap_load_mbuf(sc->ale_cdata.ale_tx_tag, map,
1537 				      *m_head, ale_dmamap_buf_cb, &ctx,
1538 				      BUS_DMA_NOWAIT);
1539 	if (error == EFBIG) {
1540 		m = m_defrag(*m_head, MB_DONTWAIT);
1541 		if (m == NULL) {
1542 			m_freem(*m_head);
1543 			*m_head = NULL;
1544 			return (ENOMEM);
1545 		}
1546 		*m_head = m;
1547 
1548 		ctx.nsegs = ALE_MAXTXSEGS;
1549 		ctx.segs = txsegs;
1550 		error =  bus_dmamap_load_mbuf(sc->ale_cdata.ale_tx_tag, map,
1551 					      *m_head, ale_dmamap_buf_cb, &ctx,
1552 					      BUS_DMA_NOWAIT);
1553 		if (error != 0) {
1554 			m_freem(*m_head);
1555 			*m_head = NULL;
1556 			return (error);
1557 		}
1558 	} else if (error != 0) {
1559 		return (error);
1560 	}
1561 	nsegs = ctx.nsegs;
1562 
1563 	if (nsegs == 0) {
1564 		m_freem(*m_head);
1565 		*m_head = NULL;
1566 		return (EIO);
1567 	}
1568 
1569 	/* Check descriptor overrun. */
1570 	if (sc->ale_cdata.ale_tx_cnt + nsegs >= ALE_TX_RING_CNT - 2) {
1571 		bus_dmamap_unload(sc->ale_cdata.ale_tx_tag, map);
1572 		return (ENOBUFS);
1573 	}
1574 	bus_dmamap_sync(sc->ale_cdata.ale_tx_tag, map, BUS_DMASYNC_PREWRITE);
1575 
1576 	m = *m_head;
1577 	/* Configure Tx checksum offload. */
1578 	if ((m->m_pkthdr.csum_flags & ALE_CSUM_FEATURES) != 0) {
1579 		/*
1580 		 * AR81xx supports Tx custom checksum offload feature
1581 		 * that offloads single 16bit checksum computation.
1582 		 * So you can choose one among IP, TCP and UDP.
1583 		 * Normally driver sets checksum start/insertion
1584 		 * position from the information of TCP/UDP frame as
1585 		 * TCP/UDP checksum takes more time than that of IP.
1586 		 * However it seems that custom checksum offload
1587 		 * requires 4 bytes aligned Tx buffers due to hardware
1588 		 * bug.
1589 		 * AR81xx also supports explicit Tx checksum computation
1590 		 * if it is told that the size of IP header and TCP
1591 		 * header(for UDP, the header size does not matter
1592 		 * because it's fixed length). However with this scheme
1593 		 * TSO does not work so you have to choose one either
1594 		 * TSO or explicit Tx checksum offload. I chosen TSO
1595 		 * plus custom checksum offload with work-around which
1596 		 * will cover most common usage for this consumer
1597 		 * ethernet controller. The work-around takes a lot of
1598 		 * CPU cycles if Tx buffer is not aligned on 4 bytes
1599 		 * boundary, though.
1600 		 */
1601 		cflags |= ALE_TD_CXSUM;
1602 		/* Set checksum start offset. */
1603 		cflags |= (poff << ALE_TD_CSUM_PLOADOFFSET_SHIFT);
1604 		/* Set checksum insertion position of TCP/UDP. */
1605 		cflags |= ((poff + m->m_pkthdr.csum_data) <<
1606 		    ALE_TD_CSUM_XSUMOFFSET_SHIFT);
1607 	}
1608 
1609 	/* Configure VLAN hardware tag insertion. */
1610 	if ((m->m_flags & M_VLANTAG) != 0) {
1611 		vtag = ALE_TX_VLAN_TAG(m->m_pkthdr.ether_vlantag);
1612 		vtag = ((vtag << ALE_TD_VLAN_SHIFT) & ALE_TD_VLAN_MASK);
1613 		cflags |= ALE_TD_INSERT_VLAN_TAG;
1614 	}
1615 
1616 	desc = NULL;
1617 	for (i = 0; i < nsegs; i++) {
1618 		desc = &sc->ale_cdata.ale_tx_ring[prod];
1619 		desc->addr = htole64(txsegs[i].ds_addr);
1620 		desc->len = htole32(ALE_TX_BYTES(txsegs[i].ds_len) | vtag);
1621 		desc->flags = htole32(cflags);
1622 		sc->ale_cdata.ale_tx_cnt++;
1623 		ALE_DESC_INC(prod, ALE_TX_RING_CNT);
1624 	}
1625 	/* Update producer index. */
1626 	sc->ale_cdata.ale_tx_prod = prod;
1627 
1628 	/* Finally set EOP on the last descriptor. */
1629 	prod = (prod + ALE_TX_RING_CNT - 1) % ALE_TX_RING_CNT;
1630 	desc = &sc->ale_cdata.ale_tx_ring[prod];
1631 	desc->flags |= htole32(ALE_TD_EOP);
1632 
1633 	/* Swap dmamap of the first and the last. */
1634 	txd = &sc->ale_cdata.ale_txdesc[prod];
1635 	map = txd_last->tx_dmamap;
1636 	txd_last->tx_dmamap = txd->tx_dmamap;
1637 	txd->tx_dmamap = map;
1638 	txd->tx_m = m;
1639 
1640 	/* Sync descriptors. */
1641 	bus_dmamap_sync(sc->ale_cdata.ale_tx_ring_tag,
1642 	    sc->ale_cdata.ale_tx_ring_map, BUS_DMASYNC_PREWRITE);
1643 
1644 	return (0);
1645 }
1646 
1647 static void
1648 ale_start(struct ifnet *ifp)
1649 {
1650         struct ale_softc *sc = ifp->if_softc;
1651 	struct mbuf *m_head;
1652 	int enq;
1653 
1654 	ASSERT_SERIALIZED(ifp->if_serializer);
1655 
1656 	if ((sc->ale_flags & ALE_FLAG_LINK) == 0) {
1657 		ifq_purge(&ifp->if_snd);
1658 		return;
1659 	}
1660 
1661 	if ((ifp->if_flags & (IFF_RUNNING | IFF_OACTIVE)) != IFF_RUNNING)
1662 		return;
1663 
1664 	/* Reclaim transmitted frames. */
1665 	if (sc->ale_cdata.ale_tx_cnt >= ALE_TX_DESC_HIWAT)
1666 		ale_txeof(sc);
1667 
1668 	enq = 0;
1669 	while (!ifq_is_empty(&ifp->if_snd)) {
1670 		m_head = ifq_dequeue(&ifp->if_snd, NULL);
1671 		if (m_head == NULL)
1672 			break;
1673 
1674 		/*
1675 		 * Pack the data into the transmit ring. If we
1676 		 * don't have room, set the OACTIVE flag and wait
1677 		 * for the NIC to drain the ring.
1678 		 */
1679 		if (ale_encap(sc, &m_head)) {
1680 			if (m_head == NULL)
1681 				break;
1682 			ifq_prepend(&ifp->if_snd, m_head);
1683 			ifp->if_flags |= IFF_OACTIVE;
1684 			break;
1685 		}
1686 		enq = 1;
1687 
1688 		/*
1689 		 * If there's a BPF listener, bounce a copy of this frame
1690 		 * to him.
1691 		 */
1692 		ETHER_BPF_MTAP(ifp, m_head);
1693 	}
1694 
1695 	if (enq) {
1696 		/* Kick. */
1697 		CSR_WRITE_4(sc, ALE_MBOX_TPD_PROD_IDX,
1698 		    sc->ale_cdata.ale_tx_prod);
1699 
1700 		/* Set a timeout in case the chip goes out to lunch. */
1701 		ifp->if_timer = ALE_TX_TIMEOUT;
1702 	}
1703 }
1704 
1705 static void
1706 ale_watchdog(struct ifnet *ifp)
1707 {
1708 	struct ale_softc *sc = ifp->if_softc;
1709 
1710 	ASSERT_SERIALIZED(ifp->if_serializer);
1711 
1712 	if ((sc->ale_flags & ALE_FLAG_LINK) == 0) {
1713 		if_printf(ifp, "watchdog timeout (lost link)\n");
1714 		ifp->if_oerrors++;
1715 		ale_init(sc);
1716 		return;
1717 	}
1718 
1719 	if_printf(ifp, "watchdog timeout -- resetting\n");
1720 	ifp->if_oerrors++;
1721 	ale_init(sc);
1722 
1723 	if (!ifq_is_empty(&ifp->if_snd))
1724 		if_devstart(ifp);
1725 }
1726 
1727 static int
1728 ale_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data, struct ucred *cr)
1729 {
1730 	struct ale_softc *sc;
1731 	struct ifreq *ifr;
1732 	struct mii_data *mii;
1733 	int error, mask;
1734 
1735 	ASSERT_SERIALIZED(ifp->if_serializer);
1736 
1737 	sc = ifp->if_softc;
1738 	ifr = (struct ifreq *)data;
1739 	error = 0;
1740 
1741 	switch (cmd) {
1742 	case SIOCSIFMTU:
1743 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > ALE_JUMBO_MTU ||
1744 		    ((sc->ale_flags & ALE_FLAG_JUMBO) == 0 &&
1745 		    ifr->ifr_mtu > ETHERMTU))
1746 			error = EINVAL;
1747 		else if (ifp->if_mtu != ifr->ifr_mtu) {
1748 			ifp->if_mtu = ifr->ifr_mtu;
1749 			if ((ifp->if_flags & IFF_RUNNING) != 0)
1750 				ale_init(sc);
1751 		}
1752 		break;
1753 
1754 	case SIOCSIFFLAGS:
1755 		if ((ifp->if_flags & IFF_UP) != 0) {
1756 			if ((ifp->if_flags & IFF_RUNNING) != 0) {
1757 				if (((ifp->if_flags ^ sc->ale_if_flags)
1758 				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
1759 					ale_rxfilter(sc);
1760 			} else {
1761 				if ((sc->ale_flags & ALE_FLAG_DETACH) == 0)
1762 					ale_init(sc);
1763 			}
1764 		} else {
1765 			if ((ifp->if_flags & IFF_RUNNING) != 0)
1766 				ale_stop(sc);
1767 		}
1768 		sc->ale_if_flags = ifp->if_flags;
1769 		break;
1770 
1771 	case SIOCADDMULTI:
1772 	case SIOCDELMULTI:
1773 		if ((ifp->if_flags & IFF_RUNNING) != 0)
1774 			ale_rxfilter(sc);
1775 		break;
1776 
1777 	case SIOCSIFMEDIA:
1778 	case SIOCGIFMEDIA:
1779 		mii = device_get_softc(sc->ale_miibus);
1780 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, cmd);
1781 		break;
1782 
1783 	case SIOCSIFCAP:
1784 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1785 		if ((mask & IFCAP_TXCSUM) != 0 &&
1786 		    (ifp->if_capabilities & IFCAP_TXCSUM) != 0) {
1787 			ifp->if_capenable ^= IFCAP_TXCSUM;
1788 			if ((ifp->if_capenable & IFCAP_TXCSUM) != 0)
1789 				ifp->if_hwassist |= ALE_CSUM_FEATURES;
1790 			else
1791 				ifp->if_hwassist &= ~ALE_CSUM_FEATURES;
1792 		}
1793 		if ((mask & IFCAP_RXCSUM) != 0 &&
1794 		    (ifp->if_capabilities & IFCAP_RXCSUM) != 0)
1795 			ifp->if_capenable ^= IFCAP_RXCSUM;
1796 
1797 		if ((mask & IFCAP_VLAN_HWTAGGING) != 0 &&
1798 		    (ifp->if_capabilities & IFCAP_VLAN_HWTAGGING) != 0) {
1799 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1800 			ale_rxvlan(sc);
1801 		}
1802 		break;
1803 
1804 	default:
1805 		error = ether_ioctl(ifp, cmd, data);
1806 		break;
1807 	}
1808 	return (error);
1809 }
1810 
1811 static void
1812 ale_mac_config(struct ale_softc *sc)
1813 {
1814 	struct mii_data *mii;
1815 	uint32_t reg;
1816 
1817 	mii = device_get_softc(sc->ale_miibus);
1818 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
1819 	reg &= ~(MAC_CFG_FULL_DUPLEX | MAC_CFG_TX_FC | MAC_CFG_RX_FC |
1820 	    MAC_CFG_SPEED_MASK);
1821 	/* Reprogram MAC with resolved speed/duplex. */
1822 	switch (IFM_SUBTYPE(mii->mii_media_active)) {
1823 	case IFM_10_T:
1824 	case IFM_100_TX:
1825 		reg |= MAC_CFG_SPEED_10_100;
1826 		break;
1827 	case IFM_1000_T:
1828 		reg |= MAC_CFG_SPEED_1000;
1829 		break;
1830 	}
1831 	if ((IFM_OPTIONS(mii->mii_media_active) & IFM_FDX) != 0) {
1832 		reg |= MAC_CFG_FULL_DUPLEX;
1833 #ifdef notyet
1834 		if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_TXPAUSE) != 0)
1835 			reg |= MAC_CFG_TX_FC;
1836 		if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_RXPAUSE) != 0)
1837 			reg |= MAC_CFG_RX_FC;
1838 #endif
1839 	}
1840 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
1841 }
1842 
1843 static void
1844 ale_stats_clear(struct ale_softc *sc)
1845 {
1846 	struct smb sb;
1847 	uint32_t *reg;
1848 	int i;
1849 
1850 	for (reg = &sb.rx_frames, i = 0; reg <= &sb.rx_pkts_filtered; reg++) {
1851 		CSR_READ_4(sc, ALE_RX_MIB_BASE + i);
1852 		i += sizeof(uint32_t);
1853 	}
1854 	/* Read Tx statistics. */
1855 	for (reg = &sb.tx_frames, i = 0; reg <= &sb.tx_mcast_bytes; reg++) {
1856 		CSR_READ_4(sc, ALE_TX_MIB_BASE + i);
1857 		i += sizeof(uint32_t);
1858 	}
1859 }
1860 
1861 static void
1862 ale_stats_update(struct ale_softc *sc)
1863 {
1864 	struct ale_hw_stats *stat;
1865 	struct smb sb, *smb;
1866 	struct ifnet *ifp;
1867 	uint32_t *reg;
1868 	int i;
1869 
1870 	ifp = &sc->arpcom.ac_if;
1871 	stat = &sc->ale_stats;
1872 	smb = &sb;
1873 
1874 	/* Read Rx statistics. */
1875 	for (reg = &sb.rx_frames, i = 0; reg <= &sb.rx_pkts_filtered; reg++) {
1876 		*reg = CSR_READ_4(sc, ALE_RX_MIB_BASE + i);
1877 		i += sizeof(uint32_t);
1878 	}
1879 	/* Read Tx statistics. */
1880 	for (reg = &sb.tx_frames, i = 0; reg <= &sb.tx_mcast_bytes; reg++) {
1881 		*reg = CSR_READ_4(sc, ALE_TX_MIB_BASE + i);
1882 		i += sizeof(uint32_t);
1883 	}
1884 
1885 	/* Rx stats. */
1886 	stat->rx_frames += smb->rx_frames;
1887 	stat->rx_bcast_frames += smb->rx_bcast_frames;
1888 	stat->rx_mcast_frames += smb->rx_mcast_frames;
1889 	stat->rx_pause_frames += smb->rx_pause_frames;
1890 	stat->rx_control_frames += smb->rx_control_frames;
1891 	stat->rx_crcerrs += smb->rx_crcerrs;
1892 	stat->rx_lenerrs += smb->rx_lenerrs;
1893 	stat->rx_bytes += smb->rx_bytes;
1894 	stat->rx_runts += smb->rx_runts;
1895 	stat->rx_fragments += smb->rx_fragments;
1896 	stat->rx_pkts_64 += smb->rx_pkts_64;
1897 	stat->rx_pkts_65_127 += smb->rx_pkts_65_127;
1898 	stat->rx_pkts_128_255 += smb->rx_pkts_128_255;
1899 	stat->rx_pkts_256_511 += smb->rx_pkts_256_511;
1900 	stat->rx_pkts_512_1023 += smb->rx_pkts_512_1023;
1901 	stat->rx_pkts_1024_1518 += smb->rx_pkts_1024_1518;
1902 	stat->rx_pkts_1519_max += smb->rx_pkts_1519_max;
1903 	stat->rx_pkts_truncated += smb->rx_pkts_truncated;
1904 	stat->rx_fifo_oflows += smb->rx_fifo_oflows;
1905 	stat->rx_rrs_errs += smb->rx_rrs_errs;
1906 	stat->rx_alignerrs += smb->rx_alignerrs;
1907 	stat->rx_bcast_bytes += smb->rx_bcast_bytes;
1908 	stat->rx_mcast_bytes += smb->rx_mcast_bytes;
1909 	stat->rx_pkts_filtered += smb->rx_pkts_filtered;
1910 
1911 	/* Tx stats. */
1912 	stat->tx_frames += smb->tx_frames;
1913 	stat->tx_bcast_frames += smb->tx_bcast_frames;
1914 	stat->tx_mcast_frames += smb->tx_mcast_frames;
1915 	stat->tx_pause_frames += smb->tx_pause_frames;
1916 	stat->tx_excess_defer += smb->tx_excess_defer;
1917 	stat->tx_control_frames += smb->tx_control_frames;
1918 	stat->tx_deferred += smb->tx_deferred;
1919 	stat->tx_bytes += smb->tx_bytes;
1920 	stat->tx_pkts_64 += smb->tx_pkts_64;
1921 	stat->tx_pkts_65_127 += smb->tx_pkts_65_127;
1922 	stat->tx_pkts_128_255 += smb->tx_pkts_128_255;
1923 	stat->tx_pkts_256_511 += smb->tx_pkts_256_511;
1924 	stat->tx_pkts_512_1023 += smb->tx_pkts_512_1023;
1925 	stat->tx_pkts_1024_1518 += smb->tx_pkts_1024_1518;
1926 	stat->tx_pkts_1519_max += smb->tx_pkts_1519_max;
1927 	stat->tx_single_colls += smb->tx_single_colls;
1928 	stat->tx_multi_colls += smb->tx_multi_colls;
1929 	stat->tx_late_colls += smb->tx_late_colls;
1930 	stat->tx_excess_colls += smb->tx_excess_colls;
1931 	stat->tx_abort += smb->tx_abort;
1932 	stat->tx_underrun += smb->tx_underrun;
1933 	stat->tx_desc_underrun += smb->tx_desc_underrun;
1934 	stat->tx_lenerrs += smb->tx_lenerrs;
1935 	stat->tx_pkts_truncated += smb->tx_pkts_truncated;
1936 	stat->tx_bcast_bytes += smb->tx_bcast_bytes;
1937 	stat->tx_mcast_bytes += smb->tx_mcast_bytes;
1938 
1939 	/* Update counters in ifnet. */
1940 	ifp->if_opackets += smb->tx_frames;
1941 
1942 	ifp->if_collisions += smb->tx_single_colls +
1943 	    smb->tx_multi_colls * 2 + smb->tx_late_colls +
1944 	    smb->tx_abort * HDPX_CFG_RETRY_DEFAULT;
1945 
1946 	/*
1947 	 * XXX
1948 	 * tx_pkts_truncated counter looks suspicious. It constantly
1949 	 * increments with no sign of Tx errors. This may indicate
1950 	 * the counter name is not correct one so I've removed the
1951 	 * counter in output errors.
1952 	 */
1953 	ifp->if_oerrors += smb->tx_abort + smb->tx_late_colls +
1954 	    smb->tx_underrun;
1955 
1956 	ifp->if_ipackets += smb->rx_frames;
1957 
1958 	ifp->if_ierrors += smb->rx_crcerrs + smb->rx_lenerrs +
1959 	    smb->rx_runts + smb->rx_pkts_truncated +
1960 	    smb->rx_fifo_oflows + smb->rx_rrs_errs +
1961 	    smb->rx_alignerrs;
1962 }
1963 
1964 static void
1965 ale_intr(void *xsc)
1966 {
1967 	struct ale_softc *sc = xsc;
1968 	struct ifnet *ifp = &sc->arpcom.ac_if;
1969 	uint32_t status;
1970 
1971 	ASSERT_SERIALIZED(ifp->if_serializer);
1972 
1973 	status = CSR_READ_4(sc, ALE_INTR_STATUS);
1974 	if ((status & ALE_INTRS) == 0)
1975 		return;
1976 
1977 	/* Acknowledge and disable interrupts. */
1978 	CSR_WRITE_4(sc, ALE_INTR_STATUS, status | INTR_DIS_INT);
1979 
1980 	if ((ifp->if_flags & IFF_RUNNING) != 0) {
1981 		int error;
1982 
1983 		error = ale_rxeof(sc);
1984 		if (error) {
1985 			sc->ale_stats.reset_brk_seq++;
1986 			ale_init(sc);
1987 			return;
1988 		}
1989 
1990 		if ((status & (INTR_DMA_RD_TO_RST | INTR_DMA_WR_TO_RST)) != 0) {
1991 			if ((status & INTR_DMA_RD_TO_RST) != 0)
1992 				device_printf(sc->ale_dev,
1993 				    "DMA read error! -- resetting\n");
1994 			if ((status & INTR_DMA_WR_TO_RST) != 0)
1995 				device_printf(sc->ale_dev,
1996 				    "DMA write error! -- resetting\n");
1997 			ale_init(sc);
1998 			return;
1999 		}
2000 
2001 		ale_txeof(sc);
2002 		if (!ifq_is_empty(&ifp->if_snd))
2003 			if_devstart(ifp);
2004 	}
2005 
2006 	/* Re-enable interrupts. */
2007 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0x7FFFFFFF);
2008 }
2009 
2010 static void
2011 ale_txeof(struct ale_softc *sc)
2012 {
2013 	struct ifnet *ifp = &sc->arpcom.ac_if;
2014 	struct ale_txdesc *txd;
2015 	uint32_t cons, prod;
2016 	int prog;
2017 
2018 	if (sc->ale_cdata.ale_tx_cnt == 0)
2019 		return;
2020 
2021 	bus_dmamap_sync(sc->ale_cdata.ale_tx_ring_tag,
2022 	    sc->ale_cdata.ale_tx_ring_map, BUS_DMASYNC_POSTREAD);
2023 	if ((sc->ale_flags & ALE_FLAG_TXCMB_BUG) == 0) {
2024 		bus_dmamap_sync(sc->ale_cdata.ale_tx_cmb_tag,
2025 		    sc->ale_cdata.ale_tx_cmb_map, BUS_DMASYNC_POSTREAD);
2026 		prod = *sc->ale_cdata.ale_tx_cmb & TPD_CNT_MASK;
2027 	} else
2028 		prod = CSR_READ_2(sc, ALE_TPD_CONS_IDX);
2029 	cons = sc->ale_cdata.ale_tx_cons;
2030 	/*
2031 	 * Go through our Tx list and free mbufs for those
2032 	 * frames which have been transmitted.
2033 	 */
2034 	for (prog = 0; cons != prod; prog++,
2035 	     ALE_DESC_INC(cons, ALE_TX_RING_CNT)) {
2036 		if (sc->ale_cdata.ale_tx_cnt <= 0)
2037 			break;
2038 		prog++;
2039 		ifp->if_flags &= ~IFF_OACTIVE;
2040 		sc->ale_cdata.ale_tx_cnt--;
2041 		txd = &sc->ale_cdata.ale_txdesc[cons];
2042 		if (txd->tx_m != NULL) {
2043 			/* Reclaim transmitted mbufs. */
2044 			bus_dmamap_unload(sc->ale_cdata.ale_tx_tag,
2045 			    txd->tx_dmamap);
2046 			m_freem(txd->tx_m);
2047 			txd->tx_m = NULL;
2048 		}
2049 	}
2050 
2051 	if (prog > 0) {
2052 		sc->ale_cdata.ale_tx_cons = cons;
2053 		/*
2054 		 * Unarm watchdog timer only when there is no pending
2055 		 * Tx descriptors in queue.
2056 		 */
2057 		if (sc->ale_cdata.ale_tx_cnt == 0)
2058 			ifp->if_timer = 0;
2059 	}
2060 }
2061 
2062 static void
2063 ale_rx_update_page(struct ale_softc *sc, struct ale_rx_page **page,
2064     uint32_t length, uint32_t *prod)
2065 {
2066 	struct ale_rx_page *rx_page;
2067 
2068 	rx_page = *page;
2069 	/* Update consumer position. */
2070 	rx_page->cons += roundup(length + sizeof(struct rx_rs),
2071 	    ALE_RX_PAGE_ALIGN);
2072 	if (rx_page->cons >= ALE_RX_PAGE_SZ) {
2073 		/*
2074 		 * End of Rx page reached, let hardware reuse
2075 		 * this page.
2076 		 */
2077 		rx_page->cons = 0;
2078 		*rx_page->cmb_addr = 0;
2079 		bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2080 				BUS_DMASYNC_PREWRITE);
2081 		CSR_WRITE_1(sc, ALE_RXF0_PAGE0 + sc->ale_cdata.ale_rx_curp,
2082 		    RXF_VALID);
2083 		/* Switch to alternate Rx page. */
2084 		sc->ale_cdata.ale_rx_curp ^= 1;
2085 		rx_page = *page =
2086 		    &sc->ale_cdata.ale_rx_page[sc->ale_cdata.ale_rx_curp];
2087 		/* Page flipped, sync CMB and Rx page. */
2088 		bus_dmamap_sync(rx_page->page_tag, rx_page->page_map,
2089 		    BUS_DMASYNC_POSTREAD);
2090 		bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2091 		    BUS_DMASYNC_POSTREAD);
2092 		/* Sync completed, cache updated producer index. */
2093 		*prod = *rx_page->cmb_addr;
2094 	}
2095 }
2096 
2097 
2098 /*
2099  * It seems that AR81xx controller can compute partial checksum.
2100  * The partial checksum value can be used to accelerate checksum
2101  * computation for fragmented TCP/UDP packets. Upper network stack
2102  * already takes advantage of the partial checksum value in IP
2103  * reassembly stage. But I'm not sure the correctness of the
2104  * partial hardware checksum assistance due to lack of data sheet.
2105  * In addition, the Rx feature of controller that requires copying
2106  * for every frames effectively nullifies one of most nice offload
2107  * capability of controller.
2108  */
2109 static void
2110 ale_rxcsum(struct ale_softc *sc, struct mbuf *m, uint32_t status)
2111 {
2112 	struct ifnet *ifp = &sc->arpcom.ac_if;
2113 	struct ip *ip;
2114 	char *p;
2115 
2116 	m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
2117 	if ((status & ALE_RD_IPCSUM_NOK) == 0)
2118 		m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
2119 
2120 	if ((sc->ale_flags & ALE_FLAG_RXCSUM_BUG) == 0) {
2121 		if (((status & ALE_RD_IPV4_FRAG) == 0) &&
2122 		    ((status & (ALE_RD_TCP | ALE_RD_UDP)) != 0) &&
2123 		    ((status & ALE_RD_TCP_UDPCSUM_NOK) == 0)) {
2124 			m->m_pkthdr.csum_flags |=
2125 			    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2126 			m->m_pkthdr.csum_data = 0xffff;
2127 		}
2128 	} else {
2129 		if ((status & (ALE_RD_TCP | ALE_RD_UDP)) != 0 &&
2130 		    (status & ALE_RD_TCP_UDPCSUM_NOK) == 0) {
2131 			p = mtod(m, char *);
2132 			p += ETHER_HDR_LEN;
2133 			if ((status & ALE_RD_802_3) != 0)
2134 				p += LLC_SNAPFRAMELEN;
2135 			if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0 &&
2136 			    (status & ALE_RD_VLAN) != 0)
2137 				p += EVL_ENCAPLEN;
2138 			ip = (struct ip *)p;
2139 			if (ip->ip_off != 0 && (status & ALE_RD_IPV4_DF) == 0)
2140 				return;
2141 			m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
2142 			    CSUM_PSEUDO_HDR;
2143 			m->m_pkthdr.csum_data = 0xffff;
2144 		}
2145 	}
2146 	/*
2147 	 * Don't mark bad checksum for TCP/UDP frames
2148 	 * as fragmented frames may always have set
2149 	 * bad checksummed bit of frame status.
2150 	 */
2151 }
2152 
2153 /* Process received frames. */
2154 static int
2155 ale_rxeof(struct ale_softc *sc)
2156 {
2157 	struct ifnet *ifp = &sc->arpcom.ac_if;
2158 	struct ale_rx_page *rx_page;
2159 	struct rx_rs *rs;
2160 	struct mbuf *m;
2161 	uint32_t length, prod, seqno, status, vtags;
2162 	int prog;
2163 
2164 	rx_page = &sc->ale_cdata.ale_rx_page[sc->ale_cdata.ale_rx_curp];
2165 	bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2166 			BUS_DMASYNC_POSTREAD);
2167 	bus_dmamap_sync(rx_page->page_tag, rx_page->page_map,
2168 			BUS_DMASYNC_POSTREAD);
2169 	/*
2170 	 * Don't directly access producer index as hardware may
2171 	 * update it while Rx handler is in progress. It would
2172 	 * be even better if there is a way to let hardware
2173 	 * know how far driver processed its received frames.
2174 	 * Alternatively, hardware could provide a way to disable
2175 	 * CMB updates until driver acknowledges the end of CMB
2176 	 * access.
2177 	 */
2178 	prod = *rx_page->cmb_addr;
2179 	for (prog = 0; ; prog++) {
2180 		if (rx_page->cons >= prod)
2181 			break;
2182 		rs = (struct rx_rs *)(rx_page->page_addr + rx_page->cons);
2183 		seqno = ALE_RX_SEQNO(le32toh(rs->seqno));
2184 		if (sc->ale_cdata.ale_rx_seqno != seqno) {
2185 			/*
2186 			 * Normally I believe this should not happen unless
2187 			 * severe driver bug or corrupted memory. However
2188 			 * it seems to happen under certain conditions which
2189 			 * is triggered by abrupt Rx events such as initiation
2190 			 * of bulk transfer of remote host. It's not easy to
2191 			 * reproduce this and I doubt it could be related
2192 			 * with FIFO overflow of hardware or activity of Tx
2193 			 * CMB updates. I also remember similar behaviour
2194 			 * seen on RealTek 8139 which uses resembling Rx
2195 			 * scheme.
2196 			 */
2197 			if (bootverbose)
2198 				device_printf(sc->ale_dev,
2199 				    "garbled seq: %u, expected: %u -- "
2200 				    "resetting!\n", seqno,
2201 				    sc->ale_cdata.ale_rx_seqno);
2202 			return (EIO);
2203 		}
2204 		/* Frame received. */
2205 		sc->ale_cdata.ale_rx_seqno++;
2206 		length = ALE_RX_BYTES(le32toh(rs->length));
2207 		status = le32toh(rs->flags);
2208 		if ((status & ALE_RD_ERROR) != 0) {
2209 			/*
2210 			 * We want to pass the following frames to upper
2211 			 * layer regardless of error status of Rx return
2212 			 * status.
2213 			 *
2214 			 *  o IP/TCP/UDP checksum is bad.
2215 			 *  o frame length and protocol specific length
2216 			 *     does not match.
2217 			 */
2218 			if ((status & (ALE_RD_CRC | ALE_RD_CODE |
2219 			    ALE_RD_DRIBBLE | ALE_RD_RUNT | ALE_RD_OFLOW |
2220 			    ALE_RD_TRUNC)) != 0) {
2221 				ale_rx_update_page(sc, &rx_page, length, &prod);
2222 				continue;
2223 			}
2224 		}
2225 		/*
2226 		 * m_devget(9) is major bottle-neck of ale(4)(It comes
2227 		 * from hardware limitation). For jumbo frames we could
2228 		 * get a slightly better performance if driver use
2229 		 * m_getjcl(9) with proper buffer size argument. However
2230 		 * that would make code more complicated and I don't
2231 		 * think users would expect good Rx performance numbers
2232 		 * on these low-end consumer ethernet controller.
2233 		 */
2234 		m = m_devget((char *)(rs + 1), length - ETHER_CRC_LEN,
2235 		    ETHER_ALIGN, ifp, NULL);
2236 		if (m == NULL) {
2237 			ifp->if_iqdrops++;
2238 			ale_rx_update_page(sc, &rx_page, length, &prod);
2239 			continue;
2240 		}
2241 		if ((ifp->if_capenable & IFCAP_RXCSUM) != 0 &&
2242 		    (status & ALE_RD_IPV4) != 0)
2243 			ale_rxcsum(sc, m, status);
2244 		if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0 &&
2245 		    (status & ALE_RD_VLAN) != 0) {
2246 			vtags = ALE_RX_VLAN(le32toh(rs->vtags));
2247 			m->m_pkthdr.ether_vlantag = ALE_RX_VLAN_TAG(vtags);
2248 			m->m_flags |= M_VLANTAG;
2249 		}
2250 
2251 		/* Pass it to upper layer. */
2252 		ifp->if_input(ifp, m);
2253 
2254 		ale_rx_update_page(sc, &rx_page, length, &prod);
2255 	}
2256 	return 0;
2257 }
2258 
2259 static void
2260 ale_tick(void *xsc)
2261 {
2262 	struct ale_softc *sc = xsc;
2263 	struct ifnet *ifp = &sc->arpcom.ac_if;
2264 	struct mii_data *mii;
2265 
2266 	lwkt_serialize_enter(ifp->if_serializer);
2267 
2268 	mii = device_get_softc(sc->ale_miibus);
2269 	mii_tick(mii);
2270 	ale_stats_update(sc);
2271 
2272 	callout_reset(&sc->ale_tick_ch, hz, ale_tick, sc);
2273 
2274 	lwkt_serialize_exit(ifp->if_serializer);
2275 }
2276 
2277 static void
2278 ale_reset(struct ale_softc *sc)
2279 {
2280 	uint32_t reg;
2281 	int i;
2282 
2283 	/* Initialize PCIe module. From Linux. */
2284 	CSR_WRITE_4(sc, 0x1008, CSR_READ_4(sc, 0x1008) | 0x8000);
2285 
2286 	CSR_WRITE_4(sc, ALE_MASTER_CFG, MASTER_RESET);
2287 	for (i = ALE_RESET_TIMEOUT; i > 0; i--) {
2288 		DELAY(10);
2289 		if ((CSR_READ_4(sc, ALE_MASTER_CFG) & MASTER_RESET) == 0)
2290 			break;
2291 	}
2292 	if (i == 0)
2293 		device_printf(sc->ale_dev, "master reset timeout!\n");
2294 
2295 	for (i = ALE_RESET_TIMEOUT; i > 0; i--) {
2296 		if ((reg = CSR_READ_4(sc, ALE_IDLE_STATUS)) == 0)
2297 			break;
2298 		DELAY(10);
2299 	}
2300 
2301 	if (i == 0)
2302 		device_printf(sc->ale_dev, "reset timeout(0x%08x)!\n", reg);
2303 }
2304 
2305 static void
2306 ale_init(void *xsc)
2307 {
2308 	struct ale_softc *sc = xsc;
2309 	struct ifnet *ifp = &sc->arpcom.ac_if;
2310 	struct mii_data *mii;
2311 	uint8_t eaddr[ETHER_ADDR_LEN];
2312 	bus_addr_t paddr;
2313 	uint32_t reg, rxf_hi, rxf_lo;
2314 
2315 	ASSERT_SERIALIZED(ifp->if_serializer);
2316 
2317 	mii = device_get_softc(sc->ale_miibus);
2318 
2319 	/*
2320 	 * Cancel any pending I/O.
2321 	 */
2322 	ale_stop(sc);
2323 
2324 	/*
2325 	 * Reset the chip to a known state.
2326 	 */
2327 	ale_reset(sc);
2328 
2329 	/* Initialize Tx descriptors, DMA memory blocks. */
2330 	ale_init_rx_pages(sc);
2331 	ale_init_tx_ring(sc);
2332 
2333 	/* Reprogram the station address. */
2334 	bcopy(IF_LLADDR(ifp), eaddr, ETHER_ADDR_LEN);
2335 	CSR_WRITE_4(sc, ALE_PAR0,
2336 	    eaddr[2] << 24 | eaddr[3] << 16 | eaddr[4] << 8 | eaddr[5]);
2337 	CSR_WRITE_4(sc, ALE_PAR1, eaddr[0] << 8 | eaddr[1]);
2338 
2339 	/*
2340 	 * Clear WOL status and disable all WOL feature as WOL
2341 	 * would interfere Rx operation under normal environments.
2342 	 */
2343 	CSR_READ_4(sc, ALE_WOL_CFG);
2344 	CSR_WRITE_4(sc, ALE_WOL_CFG, 0);
2345 
2346 	/*
2347 	 * Set Tx descriptor/RXF0/CMB base addresses. They share
2348 	 * the same high address part of DMAable region.
2349 	 */
2350 	paddr = sc->ale_cdata.ale_tx_ring_paddr;
2351 	CSR_WRITE_4(sc, ALE_TPD_ADDR_HI, ALE_ADDR_HI(paddr));
2352 	CSR_WRITE_4(sc, ALE_TPD_ADDR_LO, ALE_ADDR_LO(paddr));
2353 	CSR_WRITE_4(sc, ALE_TPD_CNT,
2354 	    (ALE_TX_RING_CNT << TPD_CNT_SHIFT) & TPD_CNT_MASK);
2355 
2356 	/* Set Rx page base address, note we use single queue. */
2357 	paddr = sc->ale_cdata.ale_rx_page[0].page_paddr;
2358 	CSR_WRITE_4(sc, ALE_RXF0_PAGE0_ADDR_LO, ALE_ADDR_LO(paddr));
2359 	paddr = sc->ale_cdata.ale_rx_page[1].page_paddr;
2360 	CSR_WRITE_4(sc, ALE_RXF0_PAGE1_ADDR_LO, ALE_ADDR_LO(paddr));
2361 
2362 	/* Set Tx/Rx CMB addresses. */
2363 	paddr = sc->ale_cdata.ale_tx_cmb_paddr;
2364 	CSR_WRITE_4(sc, ALE_TX_CMB_ADDR_LO, ALE_ADDR_LO(paddr));
2365 	paddr = sc->ale_cdata.ale_rx_page[0].cmb_paddr;
2366 	CSR_WRITE_4(sc, ALE_RXF0_CMB0_ADDR_LO, ALE_ADDR_LO(paddr));
2367 	paddr = sc->ale_cdata.ale_rx_page[1].cmb_paddr;
2368 	CSR_WRITE_4(sc, ALE_RXF0_CMB1_ADDR_LO, ALE_ADDR_LO(paddr));
2369 
2370 	/* Mark RXF0 is valid. */
2371 	CSR_WRITE_1(sc, ALE_RXF0_PAGE0, RXF_VALID);
2372 	CSR_WRITE_1(sc, ALE_RXF0_PAGE1, RXF_VALID);
2373 	/*
2374 	 * No need to initialize RFX1/RXF2/RXF3. We don't use
2375 	 * multi-queue yet.
2376 	 */
2377 
2378 	/* Set Rx page size, excluding guard frame size. */
2379 	CSR_WRITE_4(sc, ALE_RXF_PAGE_SIZE, ALE_RX_PAGE_SZ);
2380 
2381 	/* Tell hardware that we're ready to load DMA blocks. */
2382 	CSR_WRITE_4(sc, ALE_DMA_BLOCK, DMA_BLOCK_LOAD);
2383 
2384 	/* Set Rx/Tx interrupt trigger threshold. */
2385 	CSR_WRITE_4(sc, ALE_INT_TRIG_THRESH, (1 << INT_TRIG_RX_THRESH_SHIFT) |
2386 	    (4 << INT_TRIG_TX_THRESH_SHIFT));
2387 	/*
2388 	 * XXX
2389 	 * Set interrupt trigger timer, its purpose and relation
2390 	 * with interrupt moderation mechanism is not clear yet.
2391 	 */
2392 	CSR_WRITE_4(sc, ALE_INT_TRIG_TIMER,
2393 	    ((ALE_USECS(10) << INT_TRIG_RX_TIMER_SHIFT) |
2394 	    (ALE_USECS(1000) << INT_TRIG_TX_TIMER_SHIFT)));
2395 
2396 	/* Configure interrupt moderation timer. */
2397 	reg = ALE_USECS(sc->ale_int_rx_mod) << IM_TIMER_RX_SHIFT;
2398 	reg |= ALE_USECS(sc->ale_int_tx_mod) << IM_TIMER_TX_SHIFT;
2399 	CSR_WRITE_4(sc, ALE_IM_TIMER, reg);
2400 	reg = CSR_READ_4(sc, ALE_MASTER_CFG);
2401 	reg &= ~(MASTER_CHIP_REV_MASK | MASTER_CHIP_ID_MASK);
2402 	reg &= ~(MASTER_IM_RX_TIMER_ENB | MASTER_IM_TX_TIMER_ENB);
2403 	if (ALE_USECS(sc->ale_int_rx_mod) != 0)
2404 		reg |= MASTER_IM_RX_TIMER_ENB;
2405 	if (ALE_USECS(sc->ale_int_tx_mod) != 0)
2406 		reg |= MASTER_IM_TX_TIMER_ENB;
2407 	CSR_WRITE_4(sc, ALE_MASTER_CFG, reg);
2408 	CSR_WRITE_2(sc, ALE_INTR_CLR_TIMER, ALE_USECS(1000));
2409 
2410 	/* Set Maximum frame size of controller. */
2411 	if (ifp->if_mtu < ETHERMTU)
2412 		sc->ale_max_frame_size = ETHERMTU;
2413 	else
2414 		sc->ale_max_frame_size = ifp->if_mtu;
2415 	sc->ale_max_frame_size += ETHER_HDR_LEN + EVL_ENCAPLEN + ETHER_CRC_LEN;
2416 	CSR_WRITE_4(sc, ALE_FRAME_SIZE, sc->ale_max_frame_size);
2417 
2418 	/* Configure IPG/IFG parameters. */
2419 	CSR_WRITE_4(sc, ALE_IPG_IFG_CFG,
2420 	    ((IPG_IFG_IPGT_DEFAULT << IPG_IFG_IPGT_SHIFT) & IPG_IFG_IPGT_MASK) |
2421 	    ((IPG_IFG_MIFG_DEFAULT << IPG_IFG_MIFG_SHIFT) & IPG_IFG_MIFG_MASK) |
2422 	    ((IPG_IFG_IPG1_DEFAULT << IPG_IFG_IPG1_SHIFT) & IPG_IFG_IPG1_MASK) |
2423 	    ((IPG_IFG_IPG2_DEFAULT << IPG_IFG_IPG2_SHIFT) & IPG_IFG_IPG2_MASK));
2424 
2425 	/* Set parameters for half-duplex media. */
2426 	CSR_WRITE_4(sc, ALE_HDPX_CFG,
2427 	    ((HDPX_CFG_LCOL_DEFAULT << HDPX_CFG_LCOL_SHIFT) &
2428 	    HDPX_CFG_LCOL_MASK) |
2429 	    ((HDPX_CFG_RETRY_DEFAULT << HDPX_CFG_RETRY_SHIFT) &
2430 	    HDPX_CFG_RETRY_MASK) | HDPX_CFG_EXC_DEF_EN |
2431 	    ((HDPX_CFG_ABEBT_DEFAULT << HDPX_CFG_ABEBT_SHIFT) &
2432 	    HDPX_CFG_ABEBT_MASK) |
2433 	    ((HDPX_CFG_JAMIPG_DEFAULT << HDPX_CFG_JAMIPG_SHIFT) &
2434 	    HDPX_CFG_JAMIPG_MASK));
2435 
2436 	/* Configure Tx jumbo frame parameters. */
2437 	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0) {
2438 		if (ifp->if_mtu < ETHERMTU)
2439 			reg = sc->ale_max_frame_size;
2440 		else if (ifp->if_mtu < 6 * 1024)
2441 			reg = (sc->ale_max_frame_size * 2) / 3;
2442 		else
2443 			reg = sc->ale_max_frame_size / 2;
2444 		CSR_WRITE_4(sc, ALE_TX_JUMBO_THRESH,
2445 		    roundup(reg, TX_JUMBO_THRESH_UNIT) >>
2446 		    TX_JUMBO_THRESH_UNIT_SHIFT);
2447 	}
2448 
2449 	/* Configure TxQ. */
2450 	reg = (128 << (sc->ale_dma_rd_burst >> DMA_CFG_RD_BURST_SHIFT))
2451 	    << TXQ_CFG_TX_FIFO_BURST_SHIFT;
2452 	reg |= (TXQ_CFG_TPD_BURST_DEFAULT << TXQ_CFG_TPD_BURST_SHIFT) &
2453 	    TXQ_CFG_TPD_BURST_MASK;
2454 	CSR_WRITE_4(sc, ALE_TXQ_CFG, reg | TXQ_CFG_ENHANCED_MODE | TXQ_CFG_ENB);
2455 
2456 	/* Configure Rx jumbo frame & flow control parameters. */
2457 	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0) {
2458 		reg = roundup(sc->ale_max_frame_size, RX_JUMBO_THRESH_UNIT);
2459 		CSR_WRITE_4(sc, ALE_RX_JUMBO_THRESH,
2460 		    (((reg >> RX_JUMBO_THRESH_UNIT_SHIFT) <<
2461 		    RX_JUMBO_THRESH_MASK_SHIFT) & RX_JUMBO_THRESH_MASK) |
2462 		    ((RX_JUMBO_LKAH_DEFAULT << RX_JUMBO_LKAH_SHIFT) &
2463 		    RX_JUMBO_LKAH_MASK));
2464 		reg = CSR_READ_4(sc, ALE_SRAM_RX_FIFO_LEN);
2465 		rxf_hi = (reg * 7) / 10;
2466 		rxf_lo = (reg * 3)/ 10;
2467 		CSR_WRITE_4(sc, ALE_RX_FIFO_PAUSE_THRESH,
2468 		    ((rxf_lo << RX_FIFO_PAUSE_THRESH_LO_SHIFT) &
2469 		    RX_FIFO_PAUSE_THRESH_LO_MASK) |
2470 		    ((rxf_hi << RX_FIFO_PAUSE_THRESH_HI_SHIFT) &
2471 		     RX_FIFO_PAUSE_THRESH_HI_MASK));
2472 	}
2473 
2474 	/* Disable RSS. */
2475 	CSR_WRITE_4(sc, ALE_RSS_IDT_TABLE0, 0);
2476 	CSR_WRITE_4(sc, ALE_RSS_CPU, 0);
2477 
2478 	/* Configure RxQ. */
2479 	CSR_WRITE_4(sc, ALE_RXQ_CFG,
2480 	    RXQ_CFG_ALIGN_32 | RXQ_CFG_CUT_THROUGH_ENB | RXQ_CFG_ENB);
2481 
2482 	/* Configure DMA parameters. */
2483 	reg = 0;
2484 	if ((sc->ale_flags & ALE_FLAG_TXCMB_BUG) == 0)
2485 		reg |= DMA_CFG_TXCMB_ENB;
2486 	CSR_WRITE_4(sc, ALE_DMA_CFG,
2487 	    DMA_CFG_OUT_ORDER | DMA_CFG_RD_REQ_PRI | DMA_CFG_RCB_64 |
2488 	    sc->ale_dma_rd_burst | reg |
2489 	    sc->ale_dma_wr_burst | DMA_CFG_RXCMB_ENB |
2490 	    ((DMA_CFG_RD_DELAY_CNT_DEFAULT << DMA_CFG_RD_DELAY_CNT_SHIFT) &
2491 	    DMA_CFG_RD_DELAY_CNT_MASK) |
2492 	    ((DMA_CFG_WR_DELAY_CNT_DEFAULT << DMA_CFG_WR_DELAY_CNT_SHIFT) &
2493 	    DMA_CFG_WR_DELAY_CNT_MASK));
2494 
2495 	/*
2496 	 * Hardware can be configured to issue SMB interrupt based
2497 	 * on programmed interval. Since there is a callout that is
2498 	 * invoked for every hz in driver we use that instead of
2499 	 * relying on periodic SMB interrupt.
2500 	 */
2501 	CSR_WRITE_4(sc, ALE_SMB_STAT_TIMER, ALE_USECS(0));
2502 
2503 	/* Clear MAC statistics. */
2504 	ale_stats_clear(sc);
2505 
2506 	/*
2507 	 * Configure Tx/Rx MACs.
2508 	 *  - Auto-padding for short frames.
2509 	 *  - Enable CRC generation.
2510 	 *  Actual reconfiguration of MAC for resolved speed/duplex
2511 	 *  is followed after detection of link establishment.
2512 	 *  AR81xx always does checksum computation regardless of
2513 	 *  MAC_CFG_RXCSUM_ENB bit. In fact, setting the bit will
2514 	 *  cause Rx handling issue for fragmented IP datagrams due
2515 	 *  to silicon bug.
2516 	 */
2517 	reg = MAC_CFG_TX_CRC_ENB | MAC_CFG_TX_AUTO_PAD | MAC_CFG_FULL_DUPLEX |
2518 	    ((MAC_CFG_PREAMBLE_DEFAULT << MAC_CFG_PREAMBLE_SHIFT) &
2519 	    MAC_CFG_PREAMBLE_MASK);
2520 	if ((sc->ale_flags & ALE_FLAG_FASTETHER) != 0)
2521 		reg |= MAC_CFG_SPEED_10_100;
2522 	else
2523 		reg |= MAC_CFG_SPEED_1000;
2524 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2525 
2526 	/* Set up the receive filter. */
2527 	ale_rxfilter(sc);
2528 	ale_rxvlan(sc);
2529 
2530 	/* Acknowledge all pending interrupts and clear it. */
2531 	CSR_WRITE_4(sc, ALE_INTR_MASK, ALE_INTRS);
2532 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0xFFFFFFFF);
2533 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0);
2534 
2535 	sc->ale_flags &= ~ALE_FLAG_LINK;
2536 
2537 	/* Switch to the current media. */
2538 	mii_mediachg(mii);
2539 
2540 	callout_reset(&sc->ale_tick_ch, hz, ale_tick, sc);
2541 
2542 	ifp->if_flags |= IFF_RUNNING;
2543 	ifp->if_flags &= ~IFF_OACTIVE;
2544 }
2545 
2546 static void
2547 ale_stop(struct ale_softc *sc)
2548 {
2549 	struct ifnet *ifp = &sc->arpcom.ac_if;
2550 	struct ale_txdesc *txd;
2551 	uint32_t reg;
2552 	int i;
2553 
2554 	ASSERT_SERIALIZED(ifp->if_serializer);
2555 
2556 	/*
2557 	 * Mark the interface down and cancel the watchdog timer.
2558 	 */
2559 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
2560 	ifp->if_timer = 0;
2561 
2562 	callout_stop(&sc->ale_tick_ch);
2563 	sc->ale_flags &= ~ALE_FLAG_LINK;
2564 
2565 	ale_stats_update(sc);
2566 
2567 	/* Disable interrupts. */
2568 	CSR_WRITE_4(sc, ALE_INTR_MASK, 0);
2569 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0xFFFFFFFF);
2570 
2571 	/* Disable queue processing and DMA. */
2572 	reg = CSR_READ_4(sc, ALE_TXQ_CFG);
2573 	reg &= ~TXQ_CFG_ENB;
2574 	CSR_WRITE_4(sc, ALE_TXQ_CFG, reg);
2575 	reg = CSR_READ_4(sc, ALE_RXQ_CFG);
2576 	reg &= ~RXQ_CFG_ENB;
2577 	CSR_WRITE_4(sc, ALE_RXQ_CFG, reg);
2578 	reg = CSR_READ_4(sc, ALE_DMA_CFG);
2579 	reg &= ~(DMA_CFG_TXCMB_ENB | DMA_CFG_RXCMB_ENB);
2580 	CSR_WRITE_4(sc, ALE_DMA_CFG, reg);
2581 	DELAY(1000);
2582 
2583 	/* Stop Rx/Tx MACs. */
2584 	ale_stop_mac(sc);
2585 
2586 	/* Disable interrupts again? XXX */
2587 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0xFFFFFFFF);
2588 
2589 	/*
2590 	 * Free TX mbufs still in the queues.
2591 	 */
2592 	for (i = 0; i < ALE_TX_RING_CNT; i++) {
2593 		txd = &sc->ale_cdata.ale_txdesc[i];
2594 		if (txd->tx_m != NULL) {
2595 			bus_dmamap_unload(sc->ale_cdata.ale_tx_tag,
2596 			    txd->tx_dmamap);
2597 			m_freem(txd->tx_m);
2598 			txd->tx_m = NULL;
2599 		}
2600         }
2601 }
2602 
2603 static void
2604 ale_stop_mac(struct ale_softc *sc)
2605 {
2606 	uint32_t reg;
2607 	int i;
2608 
2609 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
2610 	if ((reg & (MAC_CFG_TX_ENB | MAC_CFG_RX_ENB)) != 0) {
2611 		reg &= ~MAC_CFG_TX_ENB | MAC_CFG_RX_ENB;
2612 		CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2613 	}
2614 
2615 	for (i = ALE_TIMEOUT; i > 0; i--) {
2616 		reg = CSR_READ_4(sc, ALE_IDLE_STATUS);
2617 		if (reg == 0)
2618 			break;
2619 		DELAY(10);
2620 	}
2621 	if (i == 0)
2622 		device_printf(sc->ale_dev,
2623 		    "could not disable Tx/Rx MAC(0x%08x)!\n", reg);
2624 }
2625 
2626 static void
2627 ale_init_tx_ring(struct ale_softc *sc)
2628 {
2629 	struct ale_txdesc *txd;
2630 	int i;
2631 
2632 	sc->ale_cdata.ale_tx_prod = 0;
2633 	sc->ale_cdata.ale_tx_cons = 0;
2634 	sc->ale_cdata.ale_tx_cnt = 0;
2635 
2636 	bzero(sc->ale_cdata.ale_tx_ring, ALE_TX_RING_SZ);
2637 	bzero(sc->ale_cdata.ale_tx_cmb, ALE_TX_CMB_SZ);
2638 	for (i = 0; i < ALE_TX_RING_CNT; i++) {
2639 		txd = &sc->ale_cdata.ale_txdesc[i];
2640 		txd->tx_m = NULL;
2641 	}
2642 	*sc->ale_cdata.ale_tx_cmb = 0;
2643 	bus_dmamap_sync(sc->ale_cdata.ale_tx_cmb_tag,
2644 	    sc->ale_cdata.ale_tx_cmb_map,
2645 	    BUS_DMASYNC_PREWRITE);
2646 	bus_dmamap_sync(sc->ale_cdata.ale_tx_ring_tag,
2647 	    sc->ale_cdata.ale_tx_ring_map,
2648 	    BUS_DMASYNC_PREWRITE);
2649 }
2650 
2651 static void
2652 ale_init_rx_pages(struct ale_softc *sc)
2653 {
2654 	struct ale_rx_page *rx_page;
2655 	int i;
2656 
2657 	sc->ale_cdata.ale_rx_seqno = 0;
2658 	sc->ale_cdata.ale_rx_curp = 0;
2659 
2660 	for (i = 0; i < ALE_RX_PAGES; i++) {
2661 		rx_page = &sc->ale_cdata.ale_rx_page[i];
2662 		bzero(rx_page->page_addr, sc->ale_pagesize);
2663 		bzero(rx_page->cmb_addr, ALE_RX_CMB_SZ);
2664 		rx_page->cons = 0;
2665 		*rx_page->cmb_addr = 0;
2666 		bus_dmamap_sync(rx_page->page_tag, rx_page->page_map,
2667 				BUS_DMASYNC_PREWRITE);
2668 		bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2669 				BUS_DMASYNC_PREWRITE);
2670 	}
2671 }
2672 
2673 static void
2674 ale_rxvlan(struct ale_softc *sc)
2675 {
2676 	struct ifnet *ifp;
2677 	uint32_t reg;
2678 
2679 	ifp = &sc->arpcom.ac_if;
2680 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
2681 	reg &= ~MAC_CFG_VLAN_TAG_STRIP;
2682 	if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
2683 		reg |= MAC_CFG_VLAN_TAG_STRIP;
2684 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2685 }
2686 
2687 static void
2688 ale_rxfilter(struct ale_softc *sc)
2689 {
2690 	struct ifnet *ifp;
2691 	struct ifmultiaddr *ifma;
2692 	uint32_t crc;
2693 	uint32_t mchash[2];
2694 	uint32_t rxcfg;
2695 
2696 	ifp = &sc->arpcom.ac_if;
2697 
2698 	rxcfg = CSR_READ_4(sc, ALE_MAC_CFG);
2699 	rxcfg &= ~(MAC_CFG_ALLMULTI | MAC_CFG_BCAST | MAC_CFG_PROMISC);
2700 	if ((ifp->if_flags & IFF_BROADCAST) != 0)
2701 		rxcfg |= MAC_CFG_BCAST;
2702 	if ((ifp->if_flags & (IFF_PROMISC | IFF_ALLMULTI)) != 0) {
2703 		if ((ifp->if_flags & IFF_PROMISC) != 0)
2704 			rxcfg |= MAC_CFG_PROMISC;
2705 		if ((ifp->if_flags & IFF_ALLMULTI) != 0)
2706 			rxcfg |= MAC_CFG_ALLMULTI;
2707 		CSR_WRITE_4(sc, ALE_MAR0, 0xFFFFFFFF);
2708 		CSR_WRITE_4(sc, ALE_MAR1, 0xFFFFFFFF);
2709 		CSR_WRITE_4(sc, ALE_MAC_CFG, rxcfg);
2710 		return;
2711 	}
2712 
2713 	/* Program new filter. */
2714 	bzero(mchash, sizeof(mchash));
2715 
2716 	LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2717 		if (ifma->ifma_addr->sa_family != AF_LINK)
2718 			continue;
2719 		crc = ether_crc32_le(LLADDR((struct sockaddr_dl *)
2720 		    ifma->ifma_addr), ETHER_ADDR_LEN);
2721 		mchash[crc >> 31] |= 1 << ((crc >> 26) & 0x1f);
2722 	}
2723 
2724 	CSR_WRITE_4(sc, ALE_MAR0, mchash[0]);
2725 	CSR_WRITE_4(sc, ALE_MAR1, mchash[1]);
2726 	CSR_WRITE_4(sc, ALE_MAC_CFG, rxcfg);
2727 }
2728 
2729 static int
2730 sysctl_hw_ale_int_mod(SYSCTL_HANDLER_ARGS)
2731 {
2732 	return (sysctl_int_range(oidp, arg1, arg2, req,
2733 	    ALE_IM_TIMER_MIN, ALE_IM_TIMER_MAX));
2734 }
2735 
2736 static void
2737 ale_dmamap_buf_cb(void *xctx, bus_dma_segment_t *segs, int nsegs,
2738 		  bus_size_t mapsz __unused, int error)
2739 {
2740 	struct ale_dmamap_ctx *ctx = xctx;
2741 	int i;
2742 
2743 	if (error)
2744 		return;
2745 
2746 	if (nsegs > ctx->nsegs) {
2747 		ctx->nsegs = 0;
2748 		return;
2749 	}
2750 
2751 	ctx->nsegs = nsegs;
2752 	for (i = 0; i < nsegs; ++i)
2753 		ctx->segs[i] = segs[i];
2754 }
2755