xref: /dragonfly/sys/dev/netif/ale/if_ale.c (revision 65cc0652)
1 /*-
2  * Copyright (c) 2008, Pyun YongHyeon <yongari@FreeBSD.org>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice unmodified, this list of conditions, and the following
10  *    disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  *
27  * $FreeBSD: src/sys/dev/ale/if_ale.c,v 1.3 2008/12/03 09:01:12 yongari Exp $
28  */
29 
30 /* Driver for Atheros AR8121/AR8113/AR8114 PCIe Ethernet. */
31 
32 #include <sys/param.h>
33 #include <sys/endian.h>
34 #include <sys/kernel.h>
35 #include <sys/bus.h>
36 #include <sys/interrupt.h>
37 #include <sys/malloc.h>
38 #include <sys/proc.h>
39 #include <sys/rman.h>
40 #include <sys/serialize.h>
41 #include <sys/socket.h>
42 #include <sys/sockio.h>
43 #include <sys/sysctl.h>
44 
45 #include <net/ethernet.h>
46 #include <net/if.h>
47 #include <net/bpf.h>
48 #include <net/if_arp.h>
49 #include <net/if_dl.h>
50 #include <net/if_llc.h>
51 #include <net/if_media.h>
52 #include <net/ifq_var.h>
53 #include <net/vlan/if_vlan_var.h>
54 #include <net/vlan/if_vlan_ether.h>
55 
56 #include <netinet/ip.h>
57 
58 #include <dev/netif/mii_layer/mii.h>
59 #include <dev/netif/mii_layer/miivar.h>
60 
61 #include <bus/pci/pcireg.h>
62 #include <bus/pci/pcivar.h>
63 #include "pcidevs.h"
64 
65 #include <dev/netif/ale/if_alereg.h>
66 #include <dev/netif/ale/if_alevar.h>
67 
68 /* "device miibus" required.  See GENERIC if you get errors here. */
69 #include "miibus_if.h"
70 
71 /* For more information about Tx checksum offload issues see ale_encap(). */
72 #define	ALE_CSUM_FEATURES	(CSUM_TCP | CSUM_UDP)
73 
74 struct ale_dmamap_ctx {
75 	int			nsegs;
76 	bus_dma_segment_t	*segs;
77 };
78 
79 static int	ale_probe(device_t);
80 static int	ale_attach(device_t);
81 static int	ale_detach(device_t);
82 static int	ale_shutdown(device_t);
83 static int	ale_suspend(device_t);
84 static int	ale_resume(device_t);
85 
86 static int	ale_miibus_readreg(device_t, int, int);
87 static int	ale_miibus_writereg(device_t, int, int, int);
88 static void	ale_miibus_statchg(device_t);
89 
90 static void	ale_init(void *);
91 static void	ale_start(struct ifnet *, struct ifaltq_subque *);
92 static int	ale_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
93 static void	ale_watchdog(struct ifnet *);
94 static int	ale_mediachange(struct ifnet *);
95 static void	ale_mediastatus(struct ifnet *, struct ifmediareq *);
96 
97 static void	ale_intr(void *);
98 static int	ale_rxeof(struct ale_softc *sc);
99 static void	ale_rx_update_page(struct ale_softc *, struct ale_rx_page **,
100 		    uint32_t, uint32_t *);
101 static void	ale_rxcsum(struct ale_softc *, struct mbuf *, uint32_t);
102 static void	ale_txeof(struct ale_softc *);
103 
104 static int	ale_dma_alloc(struct ale_softc *);
105 static void	ale_dma_free(struct ale_softc *);
106 static int	ale_check_boundary(struct ale_softc *);
107 static void	ale_dmamap_cb(void *, bus_dma_segment_t *, int, int);
108 static void	ale_dmamap_buf_cb(void *, bus_dma_segment_t *, int,
109 		    bus_size_t, int);
110 static int	ale_encap(struct ale_softc *, struct mbuf **);
111 static void	ale_init_rx_pages(struct ale_softc *);
112 static void	ale_init_tx_ring(struct ale_softc *);
113 
114 static void	ale_stop(struct ale_softc *);
115 static void	ale_tick(void *);
116 static void	ale_get_macaddr(struct ale_softc *);
117 static void	ale_mac_config(struct ale_softc *);
118 static void	ale_phy_reset(struct ale_softc *);
119 static void	ale_reset(struct ale_softc *);
120 static void	ale_rxfilter(struct ale_softc *);
121 static void	ale_rxvlan(struct ale_softc *);
122 static void	ale_stats_clear(struct ale_softc *);
123 static void	ale_stats_update(struct ale_softc *);
124 static void	ale_stop_mac(struct ale_softc *);
125 #ifdef notyet
126 static void	ale_setlinkspeed(struct ale_softc *);
127 static void	ale_setwol(struct ale_softc *);
128 #endif
129 
130 static void	ale_sysctl_node(struct ale_softc *);
131 static int	sysctl_hw_ale_int_mod(SYSCTL_HANDLER_ARGS);
132 
133 /*
134  * Devices supported by this driver.
135  */
136 static struct ale_dev {
137 	uint16_t	ale_vendorid;
138 	uint16_t	ale_deviceid;
139 	const char	*ale_name;
140 } ale_devs[] = {
141     { VENDORID_ATHEROS, DEVICEID_ATHEROS_AR81XX,
142     "Atheros AR8121/AR8113/AR8114 PCIe Ethernet" },
143 };
144 
145 static device_method_t ale_methods[] = {
146 	/* Device interface. */
147 	DEVMETHOD(device_probe,		ale_probe),
148 	DEVMETHOD(device_attach,	ale_attach),
149 	DEVMETHOD(device_detach,	ale_detach),
150 	DEVMETHOD(device_shutdown,	ale_shutdown),
151 	DEVMETHOD(device_suspend,	ale_suspend),
152 	DEVMETHOD(device_resume,	ale_resume),
153 
154 	/* Bus interface. */
155 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
156 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
157 
158 	/* MII interface. */
159 	DEVMETHOD(miibus_readreg,	ale_miibus_readreg),
160 	DEVMETHOD(miibus_writereg,	ale_miibus_writereg),
161 	DEVMETHOD(miibus_statchg,	ale_miibus_statchg),
162 
163 	{ NULL, NULL }
164 };
165 
166 static driver_t ale_driver = {
167 	"ale",
168 	ale_methods,
169 	sizeof(struct ale_softc)
170 };
171 
172 static devclass_t ale_devclass;
173 
174 DECLARE_DUMMY_MODULE(if_ale);
175 MODULE_VERSION(if_ale, 1);
176 MODULE_DEPEND(if_ale, miibus, 1, 1, 1);
177 DRIVER_MODULE(if_ale, pci, ale_driver, ale_devclass, NULL, NULL);
178 DRIVER_MODULE(miibus, ale, miibus_driver, miibus_devclass, NULL, NULL);
179 
180 static int
181 ale_miibus_readreg(device_t dev, int phy, int reg)
182 {
183 	struct ale_softc *sc;
184 	uint32_t v;
185 	int i;
186 
187 	sc = device_get_softc(dev);
188 
189 	if (phy != sc->ale_phyaddr)
190 		return (0);
191 
192 	if (sc->ale_flags & ALE_FLAG_FASTETHER) {
193 		if (reg == MII_100T2CR || reg == MII_100T2SR ||
194 		    reg == MII_EXTSR)
195 			return (0);
196 	}
197 
198 	CSR_WRITE_4(sc, ALE_MDIO, MDIO_OP_EXECUTE | MDIO_OP_READ |
199 	    MDIO_SUP_PREAMBLE | MDIO_CLK_25_4 | MDIO_REG_ADDR(reg));
200 	for (i = ALE_PHY_TIMEOUT; i > 0; i--) {
201 		DELAY(5);
202 		v = CSR_READ_4(sc, ALE_MDIO);
203 		if ((v & (MDIO_OP_EXECUTE | MDIO_OP_BUSY)) == 0)
204 			break;
205 	}
206 
207 	if (i == 0) {
208 		device_printf(sc->ale_dev, "phy read timeout : %d\n", reg);
209 		return (0);
210 	}
211 
212 	return ((v & MDIO_DATA_MASK) >> MDIO_DATA_SHIFT);
213 }
214 
215 static int
216 ale_miibus_writereg(device_t dev, int phy, int reg, int val)
217 {
218 	struct ale_softc *sc;
219 	uint32_t v;
220 	int i;
221 
222 	sc = device_get_softc(dev);
223 
224 	if (phy != sc->ale_phyaddr)
225 		return (0);
226 
227 	if (sc->ale_flags & ALE_FLAG_FASTETHER) {
228 		if (reg == MII_100T2CR || reg == MII_100T2SR ||
229 		    reg == MII_EXTSR)
230 			return (0);
231 	}
232 
233 	CSR_WRITE_4(sc, ALE_MDIO, MDIO_OP_EXECUTE | MDIO_OP_WRITE |
234 	    (val & MDIO_DATA_MASK) << MDIO_DATA_SHIFT |
235 	    MDIO_SUP_PREAMBLE | MDIO_CLK_25_4 | MDIO_REG_ADDR(reg));
236 	for (i = ALE_PHY_TIMEOUT; i > 0; i--) {
237 		DELAY(5);
238 		v = CSR_READ_4(sc, ALE_MDIO);
239 		if ((v & (MDIO_OP_EXECUTE | MDIO_OP_BUSY)) == 0)
240 			break;
241 	}
242 
243 	if (i == 0)
244 		device_printf(sc->ale_dev, "phy write timeout : %d\n", reg);
245 
246 	return (0);
247 }
248 
249 static void
250 ale_miibus_statchg(device_t dev)
251 {
252 	struct ale_softc *sc = device_get_softc(dev);
253 	struct ifnet *ifp = &sc->arpcom.ac_if;
254 	struct mii_data *mii;
255 	uint32_t reg;
256 
257 	ASSERT_SERIALIZED(ifp->if_serializer);
258 
259 	if ((ifp->if_flags & IFF_RUNNING) == 0)
260 		return;
261 
262 	mii = device_get_softc(sc->ale_miibus);
263 
264 	sc->ale_flags &= ~ALE_FLAG_LINK;
265 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
266 	    (IFM_ACTIVE | IFM_AVALID)) {
267 		switch (IFM_SUBTYPE(mii->mii_media_active)) {
268 		case IFM_10_T:
269 		case IFM_100_TX:
270 			sc->ale_flags |= ALE_FLAG_LINK;
271 			break;
272 
273 		case IFM_1000_T:
274 			if ((sc->ale_flags & ALE_FLAG_FASTETHER) == 0)
275 				sc->ale_flags |= ALE_FLAG_LINK;
276 			break;
277 
278 		default:
279 			break;
280 		}
281 	}
282 
283 	/* Stop Rx/Tx MACs. */
284 	ale_stop_mac(sc);
285 
286 	/* Program MACs with resolved speed/duplex/flow-control. */
287 	if ((sc->ale_flags & ALE_FLAG_LINK) != 0) {
288 		ale_mac_config(sc);
289 		/* Reenable Tx/Rx MACs. */
290 		reg = CSR_READ_4(sc, ALE_MAC_CFG);
291 		reg |= MAC_CFG_TX_ENB | MAC_CFG_RX_ENB;
292 		CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
293 	}
294 }
295 
296 static void
297 ale_mediastatus(struct ifnet *ifp, struct ifmediareq *ifmr)
298 {
299 	struct ale_softc *sc = ifp->if_softc;
300 	struct mii_data *mii = device_get_softc(sc->ale_miibus);
301 
302 	ASSERT_SERIALIZED(ifp->if_serializer);
303 
304 	mii_pollstat(mii);
305 	ifmr->ifm_status = mii->mii_media_status;
306 	ifmr->ifm_active = mii->mii_media_active;
307 }
308 
309 static int
310 ale_mediachange(struct ifnet *ifp)
311 {
312 	struct ale_softc *sc = ifp->if_softc;
313 	struct mii_data *mii = device_get_softc(sc->ale_miibus);
314 	int error;
315 
316 	ASSERT_SERIALIZED(ifp->if_serializer);
317 
318 	if (mii->mii_instance != 0) {
319 		struct mii_softc *miisc;
320 
321 		LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
322 			mii_phy_reset(miisc);
323 	}
324 	error = mii_mediachg(mii);
325 
326 	return (error);
327 }
328 
329 static int
330 ale_probe(device_t dev)
331 {
332 	struct ale_dev *sp;
333 	int i;
334 	uint16_t vendor, devid;
335 
336 	vendor = pci_get_vendor(dev);
337 	devid = pci_get_device(dev);
338 	sp = ale_devs;
339 	for (i = 0; i < NELEM(ale_devs); i++) {
340 		if (vendor == sp->ale_vendorid &&
341 		    devid == sp->ale_deviceid) {
342 			device_set_desc(dev, sp->ale_name);
343 			return (0);
344 		}
345 		sp++;
346 	}
347 
348 	return (ENXIO);
349 }
350 
351 static void
352 ale_get_macaddr(struct ale_softc *sc)
353 {
354 	uint32_t ea[2], reg;
355 	int i, vpdc;
356 
357 	reg = CSR_READ_4(sc, ALE_SPI_CTRL);
358 	if ((reg & SPI_VPD_ENB) != 0) {
359 		reg &= ~SPI_VPD_ENB;
360 		CSR_WRITE_4(sc, ALE_SPI_CTRL, reg);
361 	}
362 
363 	vpdc = pci_get_vpdcap_ptr(sc->ale_dev);
364 	if (vpdc) {
365 		/*
366 		 * PCI VPD capability found, let TWSI reload EEPROM.
367 		 * This will set ethernet address of controller.
368 		 */
369 		CSR_WRITE_4(sc, ALE_TWSI_CTRL, CSR_READ_4(sc, ALE_TWSI_CTRL) |
370 		    TWSI_CTRL_SW_LD_START);
371 		for (i = 100; i > 0; i--) {
372 			DELAY(1000);
373 			reg = CSR_READ_4(sc, ALE_TWSI_CTRL);
374 			if ((reg & TWSI_CTRL_SW_LD_START) == 0)
375 				break;
376 		}
377 		if (i == 0)
378 			device_printf(sc->ale_dev,
379 			    "reloading EEPROM timeout!\n");
380 	} else {
381 		if (bootverbose)
382 			device_printf(sc->ale_dev,
383 			    "PCI VPD capability not found!\n");
384 	}
385 
386 	ea[0] = CSR_READ_4(sc, ALE_PAR0);
387 	ea[1] = CSR_READ_4(sc, ALE_PAR1);
388 	sc->ale_eaddr[0] = (ea[1] >> 8) & 0xFF;
389 	sc->ale_eaddr[1] = (ea[1] >> 0) & 0xFF;
390 	sc->ale_eaddr[2] = (ea[0] >> 24) & 0xFF;
391 	sc->ale_eaddr[3] = (ea[0] >> 16) & 0xFF;
392 	sc->ale_eaddr[4] = (ea[0] >> 8) & 0xFF;
393 	sc->ale_eaddr[5] = (ea[0] >> 0) & 0xFF;
394 }
395 
396 static void
397 ale_phy_reset(struct ale_softc *sc)
398 {
399 	/* Reset magic from Linux. */
400 	CSR_WRITE_2(sc, ALE_GPHY_CTRL,
401 	    GPHY_CTRL_HIB_EN | GPHY_CTRL_HIB_PULSE | GPHY_CTRL_SEL_ANA_RESET |
402 	    GPHY_CTRL_PHY_PLL_ON);
403 	DELAY(1000);
404 	CSR_WRITE_2(sc, ALE_GPHY_CTRL,
405 	    GPHY_CTRL_EXT_RESET | GPHY_CTRL_HIB_EN | GPHY_CTRL_HIB_PULSE |
406 	    GPHY_CTRL_SEL_ANA_RESET | GPHY_CTRL_PHY_PLL_ON);
407 	DELAY(1000);
408 
409 #define	ATPHY_DBG_ADDR		0x1D
410 #define	ATPHY_DBG_DATA		0x1E
411 
412 	/* Enable hibernation mode. */
413 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
414 	    ATPHY_DBG_ADDR, 0x0B);
415 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
416 	    ATPHY_DBG_DATA, 0xBC00);
417 	/* Set Class A/B for all modes. */
418 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
419 	    ATPHY_DBG_ADDR, 0x00);
420 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
421 	    ATPHY_DBG_DATA, 0x02EF);
422 	/* Enable 10BT power saving. */
423 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
424 	    ATPHY_DBG_ADDR, 0x12);
425 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
426 	    ATPHY_DBG_DATA, 0x4C04);
427 	/* Adjust 1000T power. */
428 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
429 	    ATPHY_DBG_ADDR, 0x04);
430 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
431 	    ATPHY_DBG_ADDR, 0x8BBB);
432 	/* 10BT center tap voltage. */
433 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
434 	    ATPHY_DBG_ADDR, 0x05);
435 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
436 	    ATPHY_DBG_ADDR, 0x2C46);
437 
438 #undef	ATPHY_DBG_ADDR
439 #undef	ATPHY_DBG_DATA
440 	DELAY(1000);
441 }
442 
443 static int
444 ale_attach(device_t dev)
445 {
446 	struct ale_softc *sc = device_get_softc(dev);
447 	struct ifnet *ifp = &sc->arpcom.ac_if;
448 	int error = 0;
449 	uint32_t rxf_len, txf_len;
450 	uint8_t pcie_ptr;
451 
452 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
453 	sc->ale_dev = dev;
454 
455 	callout_init(&sc->ale_tick_ch);
456 
457 #ifndef BURN_BRIDGES
458 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
459 		uint32_t irq, mem;
460 
461 		irq = pci_read_config(dev, PCIR_INTLINE, 4);
462 		mem = pci_read_config(dev, ALE_PCIR_BAR, 4);
463 
464 		device_printf(dev, "chip is in D%d power mode "
465 		    "-- setting to D0\n", pci_get_powerstate(dev));
466 
467 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
468 
469 		pci_write_config(dev, PCIR_INTLINE, irq, 4);
470 		pci_write_config(dev, ALE_PCIR_BAR, mem, 4);
471 	}
472 #endif	/* !BURN_BRIDGE */
473 
474 	/* Enable bus mastering */
475 	pci_enable_busmaster(dev);
476 
477 	/*
478 	 * Allocate memory mapped IO
479 	 */
480 	sc->ale_mem_rid = ALE_PCIR_BAR;
481 	sc->ale_mem_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
482 						 &sc->ale_mem_rid, RF_ACTIVE);
483 	if (sc->ale_mem_res == NULL) {
484 		device_printf(dev, "can't allocate IO memory\n");
485 		return ENXIO;
486 	}
487 	sc->ale_mem_bt = rman_get_bustag(sc->ale_mem_res);
488 	sc->ale_mem_bh = rman_get_bushandle(sc->ale_mem_res);
489 
490 	/*
491 	 * Allocate IRQ
492 	 */
493 	sc->ale_irq_rid = 0;
494 	sc->ale_irq_res = bus_alloc_resource_any(dev, SYS_RES_IRQ,
495 						 &sc->ale_irq_rid,
496 						 RF_SHAREABLE | RF_ACTIVE);
497 	if (sc->ale_irq_res == NULL) {
498 		device_printf(dev, "can't allocate irq\n");
499 		error = ENXIO;
500 		goto fail;
501 	}
502 
503 	/* Set PHY address. */
504 	sc->ale_phyaddr = ALE_PHY_ADDR;
505 
506 	/* Reset PHY. */
507 	ale_phy_reset(sc);
508 
509 	/* Reset the ethernet controller. */
510 	ale_reset(sc);
511 
512 	/* Get PCI and chip id/revision. */
513 	sc->ale_rev = pci_get_revid(dev);
514 	if (sc->ale_rev >= 0xF0) {
515 		/* L2E Rev. B. AR8114 */
516 		sc->ale_flags |= ALE_FLAG_FASTETHER;
517 	} else {
518 		if ((CSR_READ_4(sc, ALE_PHY_STATUS) & PHY_STATUS_100M) != 0) {
519 			/* L1E AR8121 */
520 			sc->ale_flags |= ALE_FLAG_JUMBO;
521 		} else {
522 			/* L2E Rev. A. AR8113 */
523 			sc->ale_flags |= ALE_FLAG_FASTETHER;
524 		}
525 	}
526 
527 	/*
528 	 * All known controllers seems to require 4 bytes alignment
529 	 * of Tx buffers to make Tx checksum offload with custom
530 	 * checksum generation method work.
531 	 */
532 	sc->ale_flags |= ALE_FLAG_TXCSUM_BUG;
533 
534 	/*
535 	 * All known controllers seems to have issues on Rx checksum
536 	 * offload for fragmented IP datagrams.
537 	 */
538 	sc->ale_flags |= ALE_FLAG_RXCSUM_BUG;
539 
540 	/*
541 	 * Don't use Tx CMB. It is known to cause RRS update failure
542 	 * under certain circumstances. Typical phenomenon of the
543 	 * issue would be unexpected sequence number encountered in
544 	 * Rx handler.
545 	 */
546 	sc->ale_flags |= ALE_FLAG_TXCMB_BUG;
547 	sc->ale_chip_rev = CSR_READ_4(sc, ALE_MASTER_CFG) >>
548 	    MASTER_CHIP_REV_SHIFT;
549 	if (bootverbose) {
550 		device_printf(dev, "PCI device revision : 0x%04x\n",
551 		    sc->ale_rev);
552 		device_printf(dev, "Chip id/revision : 0x%04x\n",
553 		    sc->ale_chip_rev);
554 	}
555 
556 	/*
557 	 * Uninitialized hardware returns an invalid chip id/revision
558 	 * as well as 0xFFFFFFFF for Tx/Rx fifo length.
559 	 */
560 	txf_len = CSR_READ_4(sc, ALE_SRAM_TX_FIFO_LEN);
561 	rxf_len = CSR_READ_4(sc, ALE_SRAM_RX_FIFO_LEN);
562 	if (sc->ale_chip_rev == 0xFFFF || txf_len == 0xFFFFFFFF ||
563 	    rxf_len == 0xFFFFFFF) {
564 		device_printf(dev,"chip revision : 0x%04x, %u Tx FIFO "
565 		    "%u Rx FIFO -- not initialized?\n", sc->ale_chip_rev,
566 		    txf_len, rxf_len);
567 		error = ENXIO;
568 		goto fail;
569 	}
570 	device_printf(dev, "%u Tx FIFO, %u Rx FIFO\n", txf_len, rxf_len);
571 
572 	/* Get DMA parameters from PCIe device control register. */
573 	pcie_ptr = pci_get_pciecap_ptr(dev);
574 	if (pcie_ptr) {
575 		uint16_t devctl;
576 
577 		sc->ale_flags |= ALE_FLAG_PCIE;
578 		devctl = pci_read_config(dev, pcie_ptr + PCIER_DEVCTRL, 2);
579 		/* Max read request size. */
580 		sc->ale_dma_rd_burst = ((devctl >> 12) & 0x07) <<
581 		    DMA_CFG_RD_BURST_SHIFT;
582 		/* Max payload size. */
583 		sc->ale_dma_wr_burst = ((devctl >> 5) & 0x07) <<
584 		    DMA_CFG_WR_BURST_SHIFT;
585 		if (bootverbose) {
586 			device_printf(dev, "Read request size : %d bytes.\n",
587 			    128 << ((devctl >> 12) & 0x07));
588 			device_printf(dev, "TLP payload size : %d bytes.\n",
589 			    128 << ((devctl >> 5) & 0x07));
590 		}
591 	} else {
592 		sc->ale_dma_rd_burst = DMA_CFG_RD_BURST_128;
593 		sc->ale_dma_wr_burst = DMA_CFG_WR_BURST_128;
594 	}
595 
596 	/* Create device sysctl node. */
597 	ale_sysctl_node(sc);
598 
599 	if ((error = ale_dma_alloc(sc)) != 0)
600 		goto fail;
601 
602 	/* Load station address. */
603 	ale_get_macaddr(sc);
604 
605 	ifp->if_softc = sc;
606 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
607 	ifp->if_ioctl = ale_ioctl;
608 	ifp->if_start = ale_start;
609 	ifp->if_init = ale_init;
610 	ifp->if_watchdog = ale_watchdog;
611 	ifq_set_maxlen(&ifp->if_snd, ALE_TX_RING_CNT - 1);
612 	ifq_set_ready(&ifp->if_snd);
613 
614 	ifp->if_capabilities = IFCAP_RXCSUM |
615 			       IFCAP_VLAN_MTU |
616 			       IFCAP_VLAN_HWTAGGING;
617 #ifdef notyet
618 	ifp->if_capabilities |= IFCAP_TXCSUM;
619 	ifp->if_hwassist = ALE_CSUM_FEATURES;
620 #endif
621 	ifp->if_capenable = ifp->if_capabilities;
622 
623 	/* Set up MII bus. */
624 	if ((error = mii_phy_probe(dev, &sc->ale_miibus, ale_mediachange,
625 	    ale_mediastatus)) != 0) {
626 		device_printf(dev, "no PHY found!\n");
627 		goto fail;
628 	}
629 
630 	ether_ifattach(ifp, sc->ale_eaddr, NULL);
631 
632 	/* Tell the upper layer(s) we support long frames. */
633 	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
634 
635 	ifq_set_cpuid(&ifp->if_snd, rman_get_cpuid(sc->ale_irq_res));
636 
637 	error = bus_setup_intr(dev, sc->ale_irq_res, INTR_MPSAFE, ale_intr, sc,
638 			       &sc->ale_irq_handle, ifp->if_serializer);
639 	if (error) {
640 		device_printf(dev, "could not set up interrupt handler.\n");
641 		ether_ifdetach(ifp);
642 		goto fail;
643 	}
644 
645 	return 0;
646 fail:
647 	ale_detach(dev);
648 	return (error);
649 }
650 
651 static int
652 ale_detach(device_t dev)
653 {
654 	struct ale_softc *sc = device_get_softc(dev);
655 
656 	if (device_is_attached(dev)) {
657 		struct ifnet *ifp = &sc->arpcom.ac_if;
658 
659 		lwkt_serialize_enter(ifp->if_serializer);
660 		sc->ale_flags |= ALE_FLAG_DETACH;
661 		ale_stop(sc);
662 		bus_teardown_intr(dev, sc->ale_irq_res, sc->ale_irq_handle);
663 		lwkt_serialize_exit(ifp->if_serializer);
664 
665 		ether_ifdetach(ifp);
666 	}
667 
668 	if (sc->ale_miibus != NULL)
669 		device_delete_child(dev, sc->ale_miibus);
670 	bus_generic_detach(dev);
671 
672 	if (sc->ale_irq_res != NULL) {
673 		bus_release_resource(dev, SYS_RES_IRQ, sc->ale_irq_rid,
674 				     sc->ale_irq_res);
675 	}
676 	if (sc->ale_mem_res != NULL) {
677 		bus_release_resource(dev, SYS_RES_MEMORY, sc->ale_mem_rid,
678 				     sc->ale_mem_res);
679 	}
680 
681 	ale_dma_free(sc);
682 
683 	return (0);
684 }
685 
686 #define	ALE_SYSCTL_STAT_ADD32(c, h, n, p, d)	\
687 	    SYSCTL_ADD_UINT(c, h, OID_AUTO, n, CTLFLAG_RD, p, 0, d)
688 #define	ALE_SYSCTL_STAT_ADD64(c, h, n, p, d)	\
689 	    SYSCTL_ADD_QUAD(c, h, OID_AUTO, n, CTLFLAG_RD, p, 0, d)
690 
691 static void
692 ale_sysctl_node(struct ale_softc *sc)
693 {
694 	struct sysctl_ctx_list *ctx;
695 	struct sysctl_oid_list *child, *parent;
696 	struct sysctl_oid *tree;
697 	struct ale_hw_stats *stats;
698 	int error;
699 
700 	stats = &sc->ale_stats;
701 	ctx = device_get_sysctl_ctx(sc->ale_dev);
702 	child = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->ale_dev));
703 
704 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "int_rx_mod",
705 	    CTLTYPE_INT | CTLFLAG_RW, &sc->ale_int_rx_mod, 0,
706 	    sysctl_hw_ale_int_mod, "I", "ale Rx interrupt moderation");
707 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "int_tx_mod",
708 	    CTLTYPE_INT | CTLFLAG_RW, &sc->ale_int_tx_mod, 0,
709 	    sysctl_hw_ale_int_mod, "I", "ale Tx interrupt moderation");
710 
711 	/*
712 	 * Pull in device tunables.
713 	 */
714 	sc->ale_int_rx_mod = ALE_IM_RX_TIMER_DEFAULT;
715 	error = resource_int_value(device_get_name(sc->ale_dev),
716 	    device_get_unit(sc->ale_dev), "int_rx_mod", &sc->ale_int_rx_mod);
717 	if (error == 0) {
718 		if (sc->ale_int_rx_mod < ALE_IM_TIMER_MIN ||
719 		    sc->ale_int_rx_mod > ALE_IM_TIMER_MAX) {
720 			device_printf(sc->ale_dev, "int_rx_mod value out of "
721 			    "range; using default: %d\n",
722 			    ALE_IM_RX_TIMER_DEFAULT);
723 			sc->ale_int_rx_mod = ALE_IM_RX_TIMER_DEFAULT;
724 		}
725 	}
726 
727 	sc->ale_int_tx_mod = ALE_IM_TX_TIMER_DEFAULT;
728 	error = resource_int_value(device_get_name(sc->ale_dev),
729 	    device_get_unit(sc->ale_dev), "int_tx_mod", &sc->ale_int_tx_mod);
730 	if (error == 0) {
731 		if (sc->ale_int_tx_mod < ALE_IM_TIMER_MIN ||
732 		    sc->ale_int_tx_mod > ALE_IM_TIMER_MAX) {
733 			device_printf(sc->ale_dev, "int_tx_mod value out of "
734 			    "range; using default: %d\n",
735 			    ALE_IM_TX_TIMER_DEFAULT);
736 			sc->ale_int_tx_mod = ALE_IM_TX_TIMER_DEFAULT;
737 		}
738 	}
739 
740 	/* Misc statistics. */
741 	ALE_SYSCTL_STAT_ADD32(ctx, child, "reset_brk_seq",
742 	    &stats->reset_brk_seq,
743 	    "Controller resets due to broken Rx sequnce number");
744 
745 	tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "stats", CTLFLAG_RD,
746 	    NULL, "ATE statistics");
747 	parent = SYSCTL_CHILDREN(tree);
748 
749 	/* Rx statistics. */
750 	tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "rx", CTLFLAG_RD,
751 	    NULL, "Rx MAC statistics");
752 	child = SYSCTL_CHILDREN(tree);
753 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_frames",
754 	    &stats->rx_frames, "Good frames");
755 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_bcast_frames",
756 	    &stats->rx_bcast_frames, "Good broadcast frames");
757 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_mcast_frames",
758 	    &stats->rx_mcast_frames, "Good multicast frames");
759 	ALE_SYSCTL_STAT_ADD32(ctx, child, "pause_frames",
760 	    &stats->rx_pause_frames, "Pause control frames");
761 	ALE_SYSCTL_STAT_ADD32(ctx, child, "control_frames",
762 	    &stats->rx_control_frames, "Control frames");
763 	ALE_SYSCTL_STAT_ADD32(ctx, child, "crc_errs",
764 	    &stats->rx_crcerrs, "CRC errors");
765 	ALE_SYSCTL_STAT_ADD32(ctx, child, "len_errs",
766 	    &stats->rx_lenerrs, "Frames with length mismatched");
767 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_octets",
768 	    &stats->rx_bytes, "Good octets");
769 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_bcast_octets",
770 	    &stats->rx_bcast_bytes, "Good broadcast octets");
771 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_mcast_octets",
772 	    &stats->rx_mcast_bytes, "Good multicast octets");
773 	ALE_SYSCTL_STAT_ADD32(ctx, child, "runts",
774 	    &stats->rx_runts, "Too short frames");
775 	ALE_SYSCTL_STAT_ADD32(ctx, child, "fragments",
776 	    &stats->rx_fragments, "Fragmented frames");
777 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_64",
778 	    &stats->rx_pkts_64, "64 bytes frames");
779 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_65_127",
780 	    &stats->rx_pkts_65_127, "65 to 127 bytes frames");
781 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_128_255",
782 	    &stats->rx_pkts_128_255, "128 to 255 bytes frames");
783 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_256_511",
784 	    &stats->rx_pkts_256_511, "256 to 511 bytes frames");
785 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_512_1023",
786 	    &stats->rx_pkts_512_1023, "512 to 1023 bytes frames");
787 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1024_1518",
788 	    &stats->rx_pkts_1024_1518, "1024 to 1518 bytes frames");
789 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1519_max",
790 	    &stats->rx_pkts_1519_max, "1519 to max frames");
791 	ALE_SYSCTL_STAT_ADD32(ctx, child, "trunc_errs",
792 	    &stats->rx_pkts_truncated, "Truncated frames due to MTU size");
793 	ALE_SYSCTL_STAT_ADD32(ctx, child, "fifo_oflows",
794 	    &stats->rx_fifo_oflows, "FIFO overflows");
795 	ALE_SYSCTL_STAT_ADD32(ctx, child, "rrs_errs",
796 	    &stats->rx_rrs_errs, "Return status write-back errors");
797 	ALE_SYSCTL_STAT_ADD32(ctx, child, "align_errs",
798 	    &stats->rx_alignerrs, "Alignment errors");
799 	ALE_SYSCTL_STAT_ADD32(ctx, child, "filtered",
800 	    &stats->rx_pkts_filtered,
801 	    "Frames dropped due to address filtering");
802 
803 	/* Tx statistics. */
804 	tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "tx", CTLFLAG_RD,
805 	    NULL, "Tx MAC statistics");
806 	child = SYSCTL_CHILDREN(tree);
807 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_frames",
808 	    &stats->tx_frames, "Good frames");
809 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_bcast_frames",
810 	    &stats->tx_bcast_frames, "Good broadcast frames");
811 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_mcast_frames",
812 	    &stats->tx_mcast_frames, "Good multicast frames");
813 	ALE_SYSCTL_STAT_ADD32(ctx, child, "pause_frames",
814 	    &stats->tx_pause_frames, "Pause control frames");
815 	ALE_SYSCTL_STAT_ADD32(ctx, child, "control_frames",
816 	    &stats->tx_control_frames, "Control frames");
817 	ALE_SYSCTL_STAT_ADD32(ctx, child, "excess_defers",
818 	    &stats->tx_excess_defer, "Frames with excessive derferrals");
819 	ALE_SYSCTL_STAT_ADD32(ctx, child, "defers",
820 	    &stats->tx_excess_defer, "Frames with derferrals");
821 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_octets",
822 	    &stats->tx_bytes, "Good octets");
823 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_bcast_octets",
824 	    &stats->tx_bcast_bytes, "Good broadcast octets");
825 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_mcast_octets",
826 	    &stats->tx_mcast_bytes, "Good multicast octets");
827 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_64",
828 	    &stats->tx_pkts_64, "64 bytes frames");
829 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_65_127",
830 	    &stats->tx_pkts_65_127, "65 to 127 bytes frames");
831 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_128_255",
832 	    &stats->tx_pkts_128_255, "128 to 255 bytes frames");
833 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_256_511",
834 	    &stats->tx_pkts_256_511, "256 to 511 bytes frames");
835 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_512_1023",
836 	    &stats->tx_pkts_512_1023, "512 to 1023 bytes frames");
837 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1024_1518",
838 	    &stats->tx_pkts_1024_1518, "1024 to 1518 bytes frames");
839 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1519_max",
840 	    &stats->tx_pkts_1519_max, "1519 to max frames");
841 	ALE_SYSCTL_STAT_ADD32(ctx, child, "single_colls",
842 	    &stats->tx_single_colls, "Single collisions");
843 	ALE_SYSCTL_STAT_ADD32(ctx, child, "multi_colls",
844 	    &stats->tx_multi_colls, "Multiple collisions");
845 	ALE_SYSCTL_STAT_ADD32(ctx, child, "late_colls",
846 	    &stats->tx_late_colls, "Late collisions");
847 	ALE_SYSCTL_STAT_ADD32(ctx, child, "excess_colls",
848 	    &stats->tx_excess_colls, "Excessive collisions");
849 	ALE_SYSCTL_STAT_ADD32(ctx, child, "abort",
850 	    &stats->tx_abort, "Aborted frames due to Excessive collisions");
851 	ALE_SYSCTL_STAT_ADD32(ctx, child, "underruns",
852 	    &stats->tx_underrun, "FIFO underruns");
853 	ALE_SYSCTL_STAT_ADD32(ctx, child, "desc_underruns",
854 	    &stats->tx_desc_underrun, "Descriptor write-back errors");
855 	ALE_SYSCTL_STAT_ADD32(ctx, child, "len_errs",
856 	    &stats->tx_lenerrs, "Frames with length mismatched");
857 	ALE_SYSCTL_STAT_ADD32(ctx, child, "trunc_errs",
858 	    &stats->tx_pkts_truncated, "Truncated frames due to MTU size");
859 }
860 
861 #undef ALE_SYSCTL_STAT_ADD32
862 #undef ALE_SYSCTL_STAT_ADD64
863 
864 struct ale_dmamap_arg {
865 	bus_addr_t	ale_busaddr;
866 };
867 
868 static void
869 ale_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error)
870 {
871 	struct ale_dmamap_arg *ctx;
872 
873 	if (error != 0)
874 		return;
875 
876 	KASSERT(nsegs == 1, ("%s: %d segments returned!", __func__, nsegs));
877 
878 	ctx = (struct ale_dmamap_arg *)arg;
879 	ctx->ale_busaddr = segs[0].ds_addr;
880 }
881 
882 /*
883  * Tx descriptors/RXF0/CMB DMA blocks share ALE_DESC_ADDR_HI register
884  * which specifies high address region of DMA blocks. Therefore these
885  * blocks should have the same high address of given 4GB address
886  * space(i.e. crossing 4GB boundary is not allowed).
887  */
888 static int
889 ale_check_boundary(struct ale_softc *sc)
890 {
891 	bus_addr_t rx_cmb_end[ALE_RX_PAGES], tx_cmb_end;
892 	bus_addr_t rx_page_end[ALE_RX_PAGES], tx_ring_end;
893 
894 	rx_page_end[0] = sc->ale_cdata.ale_rx_page[0].page_paddr +
895 	    sc->ale_pagesize;
896 	rx_page_end[1] = sc->ale_cdata.ale_rx_page[1].page_paddr +
897 	    sc->ale_pagesize;
898 	tx_ring_end = sc->ale_cdata.ale_tx_ring_paddr + ALE_TX_RING_SZ;
899 	tx_cmb_end = sc->ale_cdata.ale_tx_cmb_paddr + ALE_TX_CMB_SZ;
900 	rx_cmb_end[0] = sc->ale_cdata.ale_rx_page[0].cmb_paddr + ALE_RX_CMB_SZ;
901 	rx_cmb_end[1] = sc->ale_cdata.ale_rx_page[1].cmb_paddr + ALE_RX_CMB_SZ;
902 
903 	if ((ALE_ADDR_HI(tx_ring_end) !=
904 	    ALE_ADDR_HI(sc->ale_cdata.ale_tx_ring_paddr)) ||
905 	    (ALE_ADDR_HI(rx_page_end[0]) !=
906 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[0].page_paddr)) ||
907 	    (ALE_ADDR_HI(rx_page_end[1]) !=
908 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[1].page_paddr)) ||
909 	    (ALE_ADDR_HI(tx_cmb_end) !=
910 	    ALE_ADDR_HI(sc->ale_cdata.ale_tx_cmb_paddr)) ||
911 	    (ALE_ADDR_HI(rx_cmb_end[0]) !=
912 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[0].cmb_paddr)) ||
913 	    (ALE_ADDR_HI(rx_cmb_end[1]) !=
914 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[1].cmb_paddr)))
915 		return (EFBIG);
916 
917 	if ((ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_page_end[0])) ||
918 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_page_end[1])) ||
919 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_cmb_end[0])) ||
920 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_cmb_end[1])) ||
921 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(tx_cmb_end)))
922 		return (EFBIG);
923 
924 	return (0);
925 }
926 
927 static int
928 ale_dma_alloc(struct ale_softc *sc)
929 {
930 	struct ale_txdesc *txd;
931 	bus_addr_t lowaddr;
932 	struct ale_dmamap_arg ctx;
933 	int error, guard_size, i;
934 
935 	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0)
936 		guard_size = ALE_JUMBO_FRAMELEN;
937 	else
938 		guard_size = ALE_MAX_FRAMELEN;
939 	sc->ale_pagesize = roundup(guard_size + ALE_RX_PAGE_SZ,
940 	    ALE_RX_PAGE_ALIGN);
941 	lowaddr = BUS_SPACE_MAXADDR;
942 again:
943 	/* Create parent DMA tag. */
944 	error = bus_dma_tag_create(
945 	    NULL,			/* parent */
946 	    1, 0,			/* alignment, boundary */
947 	    lowaddr,			/* lowaddr */
948 	    BUS_SPACE_MAXADDR,		/* highaddr */
949 	    NULL, NULL,			/* filter, filterarg */
950 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsize */
951 	    0,				/* nsegments */
952 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsegsize */
953 	    0,				/* flags */
954 	    &sc->ale_cdata.ale_parent_tag);
955 	if (error != 0) {
956 		device_printf(sc->ale_dev,
957 		    "could not create parent DMA tag.\n");
958 		goto fail;
959 	}
960 
961 	/* Create DMA tag for Tx descriptor ring. */
962 	error = bus_dma_tag_create(
963 	    sc->ale_cdata.ale_parent_tag, /* parent */
964 	    ALE_TX_RING_ALIGN, 0,	/* alignment, boundary */
965 	    BUS_SPACE_MAXADDR,		/* lowaddr */
966 	    BUS_SPACE_MAXADDR,		/* highaddr */
967 	    NULL, NULL,			/* filter, filterarg */
968 	    ALE_TX_RING_SZ,		/* maxsize */
969 	    1,				/* nsegments */
970 	    ALE_TX_RING_SZ,		/* maxsegsize */
971 	    0,				/* flags */
972 	    &sc->ale_cdata.ale_tx_ring_tag);
973 	if (error != 0) {
974 		device_printf(sc->ale_dev,
975 		    "could not create Tx ring DMA tag.\n");
976 		goto fail;
977 	}
978 
979 	/* Create DMA tag for Rx pages. */
980 	for (i = 0; i < ALE_RX_PAGES; i++) {
981 		error = bus_dma_tag_create(
982 		    sc->ale_cdata.ale_parent_tag, /* parent */
983 		    ALE_RX_PAGE_ALIGN, 0,	/* alignment, boundary */
984 		    BUS_SPACE_MAXADDR,		/* lowaddr */
985 		    BUS_SPACE_MAXADDR,		/* highaddr */
986 		    NULL, NULL,			/* filter, filterarg */
987 		    sc->ale_pagesize,		/* maxsize */
988 		    1,				/* nsegments */
989 		    sc->ale_pagesize,		/* maxsegsize */
990 		    0,				/* flags */
991 		    &sc->ale_cdata.ale_rx_page[i].page_tag);
992 		if (error != 0) {
993 			device_printf(sc->ale_dev,
994 			    "could not create Rx page %d DMA tag.\n", i);
995 			goto fail;
996 		}
997 	}
998 
999 	/* Create DMA tag for Tx coalescing message block. */
1000 	error = bus_dma_tag_create(
1001 	    sc->ale_cdata.ale_parent_tag, /* parent */
1002 	    ALE_CMB_ALIGN, 0,		/* alignment, boundary */
1003 	    BUS_SPACE_MAXADDR,		/* lowaddr */
1004 	    BUS_SPACE_MAXADDR,		/* highaddr */
1005 	    NULL, NULL,			/* filter, filterarg */
1006 	    ALE_TX_CMB_SZ,		/* maxsize */
1007 	    1,				/* nsegments */
1008 	    ALE_TX_CMB_SZ,		/* maxsegsize */
1009 	    0,				/* flags */
1010 	    &sc->ale_cdata.ale_tx_cmb_tag);
1011 	if (error != 0) {
1012 		device_printf(sc->ale_dev,
1013 		    "could not create Tx CMB DMA tag.\n");
1014 		goto fail;
1015 	}
1016 
1017 	/* Create DMA tag for Rx coalescing message block. */
1018 	for (i = 0; i < ALE_RX_PAGES; i++) {
1019 		error = bus_dma_tag_create(
1020 		    sc->ale_cdata.ale_parent_tag, /* parent */
1021 		    ALE_CMB_ALIGN, 0,		/* alignment, boundary */
1022 		    BUS_SPACE_MAXADDR,		/* lowaddr */
1023 		    BUS_SPACE_MAXADDR,		/* highaddr */
1024 		    NULL, NULL,			/* filter, filterarg */
1025 		    ALE_RX_CMB_SZ,		/* maxsize */
1026 		    1,				/* nsegments */
1027 		    ALE_RX_CMB_SZ,		/* maxsegsize */
1028 		    0,				/* flags */
1029 		    &sc->ale_cdata.ale_rx_page[i].cmb_tag);
1030 		if (error != 0) {
1031 			device_printf(sc->ale_dev,
1032 			    "could not create Rx page %d CMB DMA tag.\n", i);
1033 			goto fail;
1034 		}
1035 	}
1036 
1037 	/* Allocate DMA'able memory and load the DMA map for Tx ring. */
1038 	error = bus_dmamem_alloc(sc->ale_cdata.ale_tx_ring_tag,
1039 	    (void **)&sc->ale_cdata.ale_tx_ring,
1040 	    BUS_DMA_WAITOK | BUS_DMA_ZERO,
1041 	    &sc->ale_cdata.ale_tx_ring_map);
1042 	if (error != 0) {
1043 		device_printf(sc->ale_dev,
1044 		    "could not allocate DMA'able memory for Tx ring.\n");
1045 		goto fail;
1046 	}
1047 	ctx.ale_busaddr = 0;
1048 	error = bus_dmamap_load(sc->ale_cdata.ale_tx_ring_tag,
1049 	    sc->ale_cdata.ale_tx_ring_map, sc->ale_cdata.ale_tx_ring,
1050 	    ALE_TX_RING_SZ, ale_dmamap_cb, &ctx, 0);
1051 	if (error != 0 || ctx.ale_busaddr == 0) {
1052 		device_printf(sc->ale_dev,
1053 		    "could not load DMA'able memory for Tx ring.\n");
1054 		goto fail;
1055 	}
1056 	sc->ale_cdata.ale_tx_ring_paddr = ctx.ale_busaddr;
1057 
1058 	/* Rx pages. */
1059 	for (i = 0; i < ALE_RX_PAGES; i++) {
1060 		error = bus_dmamem_alloc(sc->ale_cdata.ale_rx_page[i].page_tag,
1061 		    (void **)&sc->ale_cdata.ale_rx_page[i].page_addr,
1062 		    BUS_DMA_WAITOK | BUS_DMA_ZERO,
1063 		    &sc->ale_cdata.ale_rx_page[i].page_map);
1064 		if (error != 0) {
1065 			device_printf(sc->ale_dev,
1066 			    "could not allocate DMA'able memory for "
1067 			    "Rx page %d.\n", i);
1068 			goto fail;
1069 		}
1070 		ctx.ale_busaddr = 0;
1071 		error = bus_dmamap_load(sc->ale_cdata.ale_rx_page[i].page_tag,
1072 		    sc->ale_cdata.ale_rx_page[i].page_map,
1073 		    sc->ale_cdata.ale_rx_page[i].page_addr,
1074 		    sc->ale_pagesize, ale_dmamap_cb, &ctx, 0);
1075 		if (error != 0 || ctx.ale_busaddr == 0) {
1076 			device_printf(sc->ale_dev,
1077 			    "could not load DMA'able memory for "
1078 			    "Rx page %d.\n", i);
1079 			goto fail;
1080 		}
1081 		sc->ale_cdata.ale_rx_page[i].page_paddr = ctx.ale_busaddr;
1082 	}
1083 
1084 	/* Tx CMB. */
1085 	error = bus_dmamem_alloc(sc->ale_cdata.ale_tx_cmb_tag,
1086 	    (void **)&sc->ale_cdata.ale_tx_cmb,
1087 	    BUS_DMA_WAITOK | BUS_DMA_ZERO,
1088 	    &sc->ale_cdata.ale_tx_cmb_map);
1089 	if (error != 0) {
1090 		device_printf(sc->ale_dev,
1091 		    "could not allocate DMA'able memory for Tx CMB.\n");
1092 		goto fail;
1093 	}
1094 	ctx.ale_busaddr = 0;
1095 	error = bus_dmamap_load(sc->ale_cdata.ale_tx_cmb_tag,
1096 	    sc->ale_cdata.ale_tx_cmb_map, sc->ale_cdata.ale_tx_cmb,
1097 	    ALE_TX_CMB_SZ, ale_dmamap_cb, &ctx, 0);
1098 	if (error != 0 || ctx.ale_busaddr == 0) {
1099 		device_printf(sc->ale_dev,
1100 		    "could not load DMA'able memory for Tx CMB.\n");
1101 		goto fail;
1102 	}
1103 	sc->ale_cdata.ale_tx_cmb_paddr = ctx.ale_busaddr;
1104 
1105 	/* Rx CMB. */
1106 	for (i = 0; i < ALE_RX_PAGES; i++) {
1107 		error = bus_dmamem_alloc(sc->ale_cdata.ale_rx_page[i].cmb_tag,
1108 		    (void **)&sc->ale_cdata.ale_rx_page[i].cmb_addr,
1109 		    BUS_DMA_WAITOK | BUS_DMA_ZERO,
1110 		    &sc->ale_cdata.ale_rx_page[i].cmb_map);
1111 		if (error != 0) {
1112 			device_printf(sc->ale_dev, "could not allocate "
1113 			    "DMA'able memory for Rx page %d CMB.\n", i);
1114 			goto fail;
1115 		}
1116 		ctx.ale_busaddr = 0;
1117 		error = bus_dmamap_load(sc->ale_cdata.ale_rx_page[i].cmb_tag,
1118 		    sc->ale_cdata.ale_rx_page[i].cmb_map,
1119 		    sc->ale_cdata.ale_rx_page[i].cmb_addr,
1120 		    ALE_RX_CMB_SZ, ale_dmamap_cb, &ctx, 0);
1121 		if (error != 0 || ctx.ale_busaddr == 0) {
1122 			device_printf(sc->ale_dev, "could not load DMA'able "
1123 			    "memory for Rx page %d CMB.\n", i);
1124 			goto fail;
1125 		}
1126 		sc->ale_cdata.ale_rx_page[i].cmb_paddr = ctx.ale_busaddr;
1127 	}
1128 
1129 	/*
1130 	 * Tx descriptors/RXF0/CMB DMA blocks share the same
1131 	 * high address region of 64bit DMA address space.
1132 	 */
1133 	if (lowaddr != BUS_SPACE_MAXADDR_32BIT &&
1134 	    (error = ale_check_boundary(sc)) != 0) {
1135 		device_printf(sc->ale_dev, "4GB boundary crossed, "
1136 		    "switching to 32bit DMA addressing mode.\n");
1137 		ale_dma_free(sc);
1138 		/*
1139 		 * Limit max allowable DMA address space to 32bit
1140 		 * and try again.
1141 		 */
1142 		lowaddr = BUS_SPACE_MAXADDR_32BIT;
1143 		goto again;
1144 	}
1145 
1146 	/*
1147 	 * Create Tx buffer parent tag.
1148 	 * AR81xx allows 64bit DMA addressing of Tx buffers so it
1149 	 * needs separate parent DMA tag as parent DMA address space
1150 	 * could be restricted to be within 32bit address space by
1151 	 * 4GB boundary crossing.
1152 	 */
1153 	error = bus_dma_tag_create(
1154 	    NULL,			/* parent */
1155 	    1, 0,			/* alignment, boundary */
1156 	    BUS_SPACE_MAXADDR,		/* lowaddr */
1157 	    BUS_SPACE_MAXADDR,		/* highaddr */
1158 	    NULL, NULL,			/* filter, filterarg */
1159 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsize */
1160 	    0,				/* nsegments */
1161 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsegsize */
1162 	    0,				/* flags */
1163 	    &sc->ale_cdata.ale_buffer_tag);
1164 	if (error != 0) {
1165 		device_printf(sc->ale_dev,
1166 		    "could not create parent buffer DMA tag.\n");
1167 		goto fail;
1168 	}
1169 
1170 	/* Create DMA tag for Tx buffers. */
1171 	error = bus_dma_tag_create(
1172 	    sc->ale_cdata.ale_buffer_tag, /* parent */
1173 	    1, 0,			/* alignment, boundary */
1174 	    BUS_SPACE_MAXADDR,		/* lowaddr */
1175 	    BUS_SPACE_MAXADDR,		/* highaddr */
1176 	    NULL, NULL,			/* filter, filterarg */
1177 	    ALE_TSO_MAXSIZE,		/* maxsize */
1178 	    ALE_MAXTXSEGS,		/* nsegments */
1179 	    ALE_TSO_MAXSEGSIZE,		/* maxsegsize */
1180 	    0,				/* flags */
1181 	    &sc->ale_cdata.ale_tx_tag);
1182 	if (error != 0) {
1183 		device_printf(sc->ale_dev, "could not create Tx DMA tag.\n");
1184 		goto fail;
1185 	}
1186 
1187 	/* Create DMA maps for Tx buffers. */
1188 	for (i = 0; i < ALE_TX_RING_CNT; i++) {
1189 		txd = &sc->ale_cdata.ale_txdesc[i];
1190 		txd->tx_m = NULL;
1191 		txd->tx_dmamap = NULL;
1192 		error = bus_dmamap_create(sc->ale_cdata.ale_tx_tag, 0,
1193 		    &txd->tx_dmamap);
1194 		if (error != 0) {
1195 			device_printf(sc->ale_dev,
1196 			    "could not create Tx dmamap.\n");
1197 			goto fail;
1198 		}
1199 	}
1200 fail:
1201 	return (error);
1202 }
1203 
1204 static void
1205 ale_dma_free(struct ale_softc *sc)
1206 {
1207 	struct ale_txdesc *txd;
1208 	int i;
1209 
1210 	/* Tx buffers. */
1211 	if (sc->ale_cdata.ale_tx_tag != NULL) {
1212 		for (i = 0; i < ALE_TX_RING_CNT; i++) {
1213 			txd = &sc->ale_cdata.ale_txdesc[i];
1214 			if (txd->tx_dmamap != NULL) {
1215 				bus_dmamap_destroy(sc->ale_cdata.ale_tx_tag,
1216 				    txd->tx_dmamap);
1217 				txd->tx_dmamap = NULL;
1218 			}
1219 		}
1220 		bus_dma_tag_destroy(sc->ale_cdata.ale_tx_tag);
1221 		sc->ale_cdata.ale_tx_tag = NULL;
1222 	}
1223 	/* Tx descriptor ring. */
1224 	if (sc->ale_cdata.ale_tx_ring_tag != NULL) {
1225 		if (sc->ale_cdata.ale_tx_ring_map != NULL)
1226 			bus_dmamap_unload(sc->ale_cdata.ale_tx_ring_tag,
1227 			    sc->ale_cdata.ale_tx_ring_map);
1228 		if (sc->ale_cdata.ale_tx_ring_map != NULL &&
1229 		    sc->ale_cdata.ale_tx_ring != NULL)
1230 			bus_dmamem_free(sc->ale_cdata.ale_tx_ring_tag,
1231 			    sc->ale_cdata.ale_tx_ring,
1232 			    sc->ale_cdata.ale_tx_ring_map);
1233 		sc->ale_cdata.ale_tx_ring = NULL;
1234 		sc->ale_cdata.ale_tx_ring_map = NULL;
1235 		bus_dma_tag_destroy(sc->ale_cdata.ale_tx_ring_tag);
1236 		sc->ale_cdata.ale_tx_ring_tag = NULL;
1237 	}
1238 	/* Rx page block. */
1239 	for (i = 0; i < ALE_RX_PAGES; i++) {
1240 		if (sc->ale_cdata.ale_rx_page[i].page_tag != NULL) {
1241 			if (sc->ale_cdata.ale_rx_page[i].page_map != NULL)
1242 				bus_dmamap_unload(
1243 				    sc->ale_cdata.ale_rx_page[i].page_tag,
1244 				    sc->ale_cdata.ale_rx_page[i].page_map);
1245 			if (sc->ale_cdata.ale_rx_page[i].page_map != NULL &&
1246 			    sc->ale_cdata.ale_rx_page[i].page_addr != NULL)
1247 				bus_dmamem_free(
1248 				    sc->ale_cdata.ale_rx_page[i].page_tag,
1249 				    sc->ale_cdata.ale_rx_page[i].page_addr,
1250 				    sc->ale_cdata.ale_rx_page[i].page_map);
1251 			sc->ale_cdata.ale_rx_page[i].page_addr = NULL;
1252 			sc->ale_cdata.ale_rx_page[i].page_map = NULL;
1253 			bus_dma_tag_destroy(
1254 			    sc->ale_cdata.ale_rx_page[i].page_tag);
1255 			sc->ale_cdata.ale_rx_page[i].page_tag = NULL;
1256 		}
1257 	}
1258 	/* Rx CMB. */
1259 	for (i = 0; i < ALE_RX_PAGES; i++) {
1260 		if (sc->ale_cdata.ale_rx_page[i].cmb_tag != NULL) {
1261 			if (sc->ale_cdata.ale_rx_page[i].cmb_map != NULL)
1262 				bus_dmamap_unload(
1263 				    sc->ale_cdata.ale_rx_page[i].cmb_tag,
1264 				    sc->ale_cdata.ale_rx_page[i].cmb_map);
1265 			if (sc->ale_cdata.ale_rx_page[i].cmb_map != NULL &&
1266 			    sc->ale_cdata.ale_rx_page[i].cmb_addr != NULL)
1267 				bus_dmamem_free(
1268 				    sc->ale_cdata.ale_rx_page[i].cmb_tag,
1269 				    sc->ale_cdata.ale_rx_page[i].cmb_addr,
1270 				    sc->ale_cdata.ale_rx_page[i].cmb_map);
1271 			sc->ale_cdata.ale_rx_page[i].cmb_addr = NULL;
1272 			sc->ale_cdata.ale_rx_page[i].cmb_map = NULL;
1273 			bus_dma_tag_destroy(
1274 			    sc->ale_cdata.ale_rx_page[i].cmb_tag);
1275 			sc->ale_cdata.ale_rx_page[i].cmb_tag = NULL;
1276 		}
1277 	}
1278 	/* Tx CMB. */
1279 	if (sc->ale_cdata.ale_tx_cmb_tag != NULL) {
1280 		if (sc->ale_cdata.ale_tx_cmb_map != NULL)
1281 			bus_dmamap_unload(sc->ale_cdata.ale_tx_cmb_tag,
1282 			    sc->ale_cdata.ale_tx_cmb_map);
1283 		if (sc->ale_cdata.ale_tx_cmb_map != NULL &&
1284 		    sc->ale_cdata.ale_tx_cmb != NULL)
1285 			bus_dmamem_free(sc->ale_cdata.ale_tx_cmb_tag,
1286 			    sc->ale_cdata.ale_tx_cmb,
1287 			    sc->ale_cdata.ale_tx_cmb_map);
1288 		sc->ale_cdata.ale_tx_cmb = NULL;
1289 		sc->ale_cdata.ale_tx_cmb_map = NULL;
1290 		bus_dma_tag_destroy(sc->ale_cdata.ale_tx_cmb_tag);
1291 		sc->ale_cdata.ale_tx_cmb_tag = NULL;
1292 	}
1293 	if (sc->ale_cdata.ale_buffer_tag != NULL) {
1294 		bus_dma_tag_destroy(sc->ale_cdata.ale_buffer_tag);
1295 		sc->ale_cdata.ale_buffer_tag = NULL;
1296 	}
1297 	if (sc->ale_cdata.ale_parent_tag != NULL) {
1298 		bus_dma_tag_destroy(sc->ale_cdata.ale_parent_tag);
1299 		sc->ale_cdata.ale_parent_tag = NULL;
1300 	}
1301 }
1302 
1303 static int
1304 ale_shutdown(device_t dev)
1305 {
1306 	return (ale_suspend(dev));
1307 }
1308 
1309 #ifdef notyet
1310 
1311 /*
1312  * Note, this driver resets the link speed to 10/100Mbps by
1313  * restarting auto-negotiation in suspend/shutdown phase but we
1314  * don't know whether that auto-negotiation would succeed or not
1315  * as driver has no control after powering off/suspend operation.
1316  * If the renegotiation fail WOL may not work. Running at 1Gbps
1317  * will draw more power than 375mA at 3.3V which is specified in
1318  * PCI specification and that would result in complete
1319  * shutdowning power to ethernet controller.
1320  *
1321  * TODO
1322  * Save current negotiated media speed/duplex/flow-control to
1323  * softc and restore the same link again after resuming. PHY
1324  * handling such as power down/resetting to 100Mbps may be better
1325  * handled in suspend method in phy driver.
1326  */
1327 static void
1328 ale_setlinkspeed(struct ale_softc *sc)
1329 {
1330 	struct mii_data *mii;
1331 	int aneg, i;
1332 
1333 	mii = device_get_softc(sc->ale_miibus);
1334 	mii_pollstat(mii);
1335 	aneg = 0;
1336 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
1337 	    (IFM_ACTIVE | IFM_AVALID)) {
1338 		switch IFM_SUBTYPE(mii->mii_media_active) {
1339 		case IFM_10_T:
1340 		case IFM_100_TX:
1341 			return;
1342 		case IFM_1000_T:
1343 			aneg++;
1344 			break;
1345 		default:
1346 			break;
1347 		}
1348 	}
1349 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr, MII_100T2CR, 0);
1350 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
1351 	    MII_ANAR, ANAR_TX_FD | ANAR_TX | ANAR_10_FD | ANAR_10 | ANAR_CSMA);
1352 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
1353 	    MII_BMCR, BMCR_RESET | BMCR_AUTOEN | BMCR_STARTNEG);
1354 	DELAY(1000);
1355 	if (aneg != 0) {
1356 		/*
1357 		 * Poll link state until ale(4) get a 10/100Mbps link.
1358 		 */
1359 		for (i = 0; i < MII_ANEGTICKS_GIGE; i++) {
1360 			mii_pollstat(mii);
1361 			if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID))
1362 			    == (IFM_ACTIVE | IFM_AVALID)) {
1363 				switch (IFM_SUBTYPE(
1364 				    mii->mii_media_active)) {
1365 				case IFM_10_T:
1366 				case IFM_100_TX:
1367 					ale_mac_config(sc);
1368 					return;
1369 				default:
1370 					break;
1371 				}
1372 			}
1373 			ALE_UNLOCK(sc);
1374 			pause("alelnk", hz);
1375 			ALE_LOCK(sc);
1376 		}
1377 		if (i == MII_ANEGTICKS_GIGE)
1378 			device_printf(sc->ale_dev,
1379 			    "establishing a link failed, WOL may not work!");
1380 	}
1381 	/*
1382 	 * No link, force MAC to have 100Mbps, full-duplex link.
1383 	 * This is the last resort and may/may not work.
1384 	 */
1385 	mii->mii_media_status = IFM_AVALID | IFM_ACTIVE;
1386 	mii->mii_media_active = IFM_ETHER | IFM_100_TX | IFM_FDX;
1387 	ale_mac_config(sc);
1388 }
1389 
1390 static void
1391 ale_setwol(struct ale_softc *sc)
1392 {
1393 	struct ifnet *ifp;
1394 	uint32_t reg, pmcs;
1395 	uint16_t pmstat;
1396 	int pmc;
1397 
1398 	ALE_LOCK_ASSERT(sc);
1399 
1400 	if (pci_find_extcap(sc->ale_dev, PCIY_PMG, &pmc) != 0) {
1401 		/* Disable WOL. */
1402 		CSR_WRITE_4(sc, ALE_WOL_CFG, 0);
1403 		reg = CSR_READ_4(sc, ALE_PCIE_PHYMISC);
1404 		reg |= PCIE_PHYMISC_FORCE_RCV_DET;
1405 		CSR_WRITE_4(sc, ALE_PCIE_PHYMISC, reg);
1406 		/* Force PHY power down. */
1407 		CSR_WRITE_2(sc, ALE_GPHY_CTRL,
1408 		    GPHY_CTRL_EXT_RESET | GPHY_CTRL_HIB_EN |
1409 		    GPHY_CTRL_HIB_PULSE | GPHY_CTRL_PHY_PLL_ON |
1410 		    GPHY_CTRL_SEL_ANA_RESET | GPHY_CTRL_PHY_IDDQ |
1411 		    GPHY_CTRL_PCLK_SEL_DIS | GPHY_CTRL_PWDOWN_HW);
1412 		return;
1413 	}
1414 
1415 	ifp = sc->ale_ifp;
1416 	if ((ifp->if_capenable & IFCAP_WOL) != 0) {
1417 		if ((sc->ale_flags & ALE_FLAG_FASTETHER) == 0)
1418 			ale_setlinkspeed(sc);
1419 	}
1420 
1421 	pmcs = 0;
1422 	if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
1423 		pmcs |= WOL_CFG_MAGIC | WOL_CFG_MAGIC_ENB;
1424 	CSR_WRITE_4(sc, ALE_WOL_CFG, pmcs);
1425 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
1426 	reg &= ~(MAC_CFG_DBG | MAC_CFG_PROMISC | MAC_CFG_ALLMULTI |
1427 	    MAC_CFG_BCAST);
1428 	if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
1429 		reg |= MAC_CFG_ALLMULTI | MAC_CFG_BCAST;
1430 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
1431 		reg |= MAC_CFG_RX_ENB;
1432 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
1433 
1434 	if ((ifp->if_capenable & IFCAP_WOL) == 0) {
1435 		/* WOL disabled, PHY power down. */
1436 		reg = CSR_READ_4(sc, ALE_PCIE_PHYMISC);
1437 		reg |= PCIE_PHYMISC_FORCE_RCV_DET;
1438 		CSR_WRITE_4(sc, ALE_PCIE_PHYMISC, reg);
1439 		CSR_WRITE_2(sc, ALE_GPHY_CTRL,
1440 		    GPHY_CTRL_EXT_RESET | GPHY_CTRL_HIB_EN |
1441 		    GPHY_CTRL_HIB_PULSE | GPHY_CTRL_SEL_ANA_RESET |
1442 		    GPHY_CTRL_PHY_IDDQ | GPHY_CTRL_PCLK_SEL_DIS |
1443 		    GPHY_CTRL_PWDOWN_HW);
1444 	}
1445 	/* Request PME. */
1446 	pmstat = pci_read_config(sc->ale_dev, pmc + PCIR_POWER_STATUS, 2);
1447 	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
1448 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
1449 		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
1450 	pci_write_config(sc->ale_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
1451 }
1452 
1453 #endif	/* notyet */
1454 
1455 static int
1456 ale_suspend(device_t dev)
1457 {
1458 	struct ale_softc *sc = device_get_softc(dev);
1459 	struct ifnet *ifp = &sc->arpcom.ac_if;
1460 
1461 	lwkt_serialize_enter(ifp->if_serializer);
1462 	ale_stop(sc);
1463 #ifdef notyet
1464 	ale_setwol(sc);
1465 #endif
1466 	lwkt_serialize_exit(ifp->if_serializer);
1467 	return (0);
1468 }
1469 
1470 static int
1471 ale_resume(device_t dev)
1472 {
1473 	struct ale_softc *sc = device_get_softc(dev);
1474 	struct ifnet *ifp = &sc->arpcom.ac_if;
1475 	uint16_t cmd;
1476 
1477 	lwkt_serialize_enter(ifp->if_serializer);
1478 
1479 	/*
1480 	 * Clear INTx emulation disable for hardwares that
1481 	 * is set in resume event. From Linux.
1482 	 */
1483 	cmd = pci_read_config(sc->ale_dev, PCIR_COMMAND, 2);
1484 	if ((cmd & 0x0400) != 0) {
1485 		cmd &= ~0x0400;
1486 		pci_write_config(sc->ale_dev, PCIR_COMMAND, cmd, 2);
1487 	}
1488 
1489 #ifdef notyet
1490 	if (pci_find_extcap(sc->ale_dev, PCIY_PMG, &pmc) == 0) {
1491 		uint16_t pmstat;
1492 		int pmc;
1493 
1494 		/* Disable PME and clear PME status. */
1495 		pmstat = pci_read_config(sc->ale_dev,
1496 		    pmc + PCIR_POWER_STATUS, 2);
1497 		if ((pmstat & PCIM_PSTAT_PMEENABLE) != 0) {
1498 			pmstat &= ~PCIM_PSTAT_PMEENABLE;
1499 			pci_write_config(sc->ale_dev,
1500 			    pmc + PCIR_POWER_STATUS, pmstat, 2);
1501 		}
1502 	}
1503 #endif
1504 
1505 	/* Reset PHY. */
1506 	ale_phy_reset(sc);
1507 	if ((ifp->if_flags & IFF_UP) != 0)
1508 		ale_init(sc);
1509 
1510 	lwkt_serialize_exit(ifp->if_serializer);
1511 	return (0);
1512 }
1513 
1514 static int
1515 ale_encap(struct ale_softc *sc, struct mbuf **m_head)
1516 {
1517 	struct ale_txdesc *txd, *txd_last;
1518 	struct tx_desc *desc;
1519 	struct mbuf *m;
1520 	bus_dma_segment_t txsegs[ALE_MAXTXSEGS];
1521 	struct ale_dmamap_ctx ctx;
1522 	bus_dmamap_t map;
1523 	uint32_t cflags, poff, vtag;
1524 	int error, i, nsegs, prod;
1525 
1526 	M_ASSERTPKTHDR((*m_head));
1527 
1528 	m = *m_head;
1529 	cflags = vtag = 0;
1530 	poff = 0;
1531 
1532 	prod = sc->ale_cdata.ale_tx_prod;
1533 	txd = &sc->ale_cdata.ale_txdesc[prod];
1534 	txd_last = txd;
1535 	map = txd->tx_dmamap;
1536 
1537 	ctx.nsegs = ALE_MAXTXSEGS;
1538 	ctx.segs = txsegs;
1539 	error =  bus_dmamap_load_mbuf(sc->ale_cdata.ale_tx_tag, map,
1540 				      *m_head, ale_dmamap_buf_cb, &ctx,
1541 				      BUS_DMA_NOWAIT);
1542 	if (error == EFBIG) {
1543 		m = m_defrag(*m_head, M_NOWAIT);
1544 		if (m == NULL) {
1545 			m_freem(*m_head);
1546 			*m_head = NULL;
1547 			return (ENOMEM);
1548 		}
1549 		*m_head = m;
1550 
1551 		ctx.nsegs = ALE_MAXTXSEGS;
1552 		ctx.segs = txsegs;
1553 		error =  bus_dmamap_load_mbuf(sc->ale_cdata.ale_tx_tag, map,
1554 					      *m_head, ale_dmamap_buf_cb, &ctx,
1555 					      BUS_DMA_NOWAIT);
1556 		if (error != 0) {
1557 			m_freem(*m_head);
1558 			*m_head = NULL;
1559 			return (error);
1560 		}
1561 	} else if (error != 0) {
1562 		return (error);
1563 	}
1564 	nsegs = ctx.nsegs;
1565 
1566 	if (nsegs == 0) {
1567 		m_freem(*m_head);
1568 		*m_head = NULL;
1569 		return (EIO);
1570 	}
1571 
1572 	/* Check descriptor overrun. */
1573 	if (sc->ale_cdata.ale_tx_cnt + nsegs >= ALE_TX_RING_CNT - 2) {
1574 		bus_dmamap_unload(sc->ale_cdata.ale_tx_tag, map);
1575 		return (ENOBUFS);
1576 	}
1577 	bus_dmamap_sync(sc->ale_cdata.ale_tx_tag, map, BUS_DMASYNC_PREWRITE);
1578 
1579 	m = *m_head;
1580 	/* Configure Tx checksum offload. */
1581 	if ((m->m_pkthdr.csum_flags & ALE_CSUM_FEATURES) != 0) {
1582 		/*
1583 		 * AR81xx supports Tx custom checksum offload feature
1584 		 * that offloads single 16bit checksum computation.
1585 		 * So you can choose one among IP, TCP and UDP.
1586 		 * Normally driver sets checksum start/insertion
1587 		 * position from the information of TCP/UDP frame as
1588 		 * TCP/UDP checksum takes more time than that of IP.
1589 		 * However it seems that custom checksum offload
1590 		 * requires 4 bytes aligned Tx buffers due to hardware
1591 		 * bug.
1592 		 * AR81xx also supports explicit Tx checksum computation
1593 		 * if it is told that the size of IP header and TCP
1594 		 * header(for UDP, the header size does not matter
1595 		 * because it's fixed length). However with this scheme
1596 		 * TSO does not work so you have to choose one either
1597 		 * TSO or explicit Tx checksum offload. I chosen TSO
1598 		 * plus custom checksum offload with work-around which
1599 		 * will cover most common usage for this consumer
1600 		 * ethernet controller. The work-around takes a lot of
1601 		 * CPU cycles if Tx buffer is not aligned on 4 bytes
1602 		 * boundary, though.
1603 		 */
1604 		cflags |= ALE_TD_CXSUM;
1605 		/* Set checksum start offset. */
1606 		cflags |= (poff << ALE_TD_CSUM_PLOADOFFSET_SHIFT);
1607 		/* Set checksum insertion position of TCP/UDP. */
1608 		cflags |= ((poff + m->m_pkthdr.csum_data) <<
1609 		    ALE_TD_CSUM_XSUMOFFSET_SHIFT);
1610 	}
1611 
1612 	/* Configure VLAN hardware tag insertion. */
1613 	if ((m->m_flags & M_VLANTAG) != 0) {
1614 		vtag = ALE_TX_VLAN_TAG(m->m_pkthdr.ether_vlantag);
1615 		vtag = ((vtag << ALE_TD_VLAN_SHIFT) & ALE_TD_VLAN_MASK);
1616 		cflags |= ALE_TD_INSERT_VLAN_TAG;
1617 	}
1618 
1619 	desc = NULL;
1620 	for (i = 0; i < nsegs; i++) {
1621 		desc = &sc->ale_cdata.ale_tx_ring[prod];
1622 		desc->addr = htole64(txsegs[i].ds_addr);
1623 		desc->len = htole32(ALE_TX_BYTES(txsegs[i].ds_len) | vtag);
1624 		desc->flags = htole32(cflags);
1625 		sc->ale_cdata.ale_tx_cnt++;
1626 		ALE_DESC_INC(prod, ALE_TX_RING_CNT);
1627 	}
1628 	/* Update producer index. */
1629 	sc->ale_cdata.ale_tx_prod = prod;
1630 
1631 	/* Finally set EOP on the last descriptor. */
1632 	prod = (prod + ALE_TX_RING_CNT - 1) % ALE_TX_RING_CNT;
1633 	desc = &sc->ale_cdata.ale_tx_ring[prod];
1634 	desc->flags |= htole32(ALE_TD_EOP);
1635 
1636 	/* Swap dmamap of the first and the last. */
1637 	txd = &sc->ale_cdata.ale_txdesc[prod];
1638 	map = txd_last->tx_dmamap;
1639 	txd_last->tx_dmamap = txd->tx_dmamap;
1640 	txd->tx_dmamap = map;
1641 	txd->tx_m = m;
1642 
1643 	/* Sync descriptors. */
1644 	bus_dmamap_sync(sc->ale_cdata.ale_tx_ring_tag,
1645 	    sc->ale_cdata.ale_tx_ring_map, BUS_DMASYNC_PREWRITE);
1646 
1647 	return (0);
1648 }
1649 
1650 static void
1651 ale_start(struct ifnet *ifp, struct ifaltq_subque *ifsq)
1652 {
1653         struct ale_softc *sc = ifp->if_softc;
1654 	struct mbuf *m_head;
1655 	int enq;
1656 
1657 	ASSERT_ALTQ_SQ_DEFAULT(ifp, ifsq);
1658 	ASSERT_SERIALIZED(ifp->if_serializer);
1659 
1660 	if ((sc->ale_flags & ALE_FLAG_LINK) == 0) {
1661 		ifq_purge(&ifp->if_snd);
1662 		return;
1663 	}
1664 
1665 	if ((ifp->if_flags & IFF_RUNNING) == 0 || ifq_is_oactive(&ifp->if_snd))
1666 		return;
1667 
1668 	/* Reclaim transmitted frames. */
1669 	if (sc->ale_cdata.ale_tx_cnt >= ALE_TX_DESC_HIWAT)
1670 		ale_txeof(sc);
1671 
1672 	enq = 0;
1673 	while (!ifq_is_empty(&ifp->if_snd)) {
1674 		m_head = ifq_dequeue(&ifp->if_snd);
1675 		if (m_head == NULL)
1676 			break;
1677 
1678 		/*
1679 		 * Pack the data into the transmit ring. If we
1680 		 * don't have room, set the OACTIVE flag and wait
1681 		 * for the NIC to drain the ring.
1682 		 */
1683 		if (ale_encap(sc, &m_head)) {
1684 			if (m_head == NULL)
1685 				break;
1686 			ifq_prepend(&ifp->if_snd, m_head);
1687 			ifq_set_oactive(&ifp->if_snd);
1688 			break;
1689 		}
1690 		enq = 1;
1691 
1692 		/*
1693 		 * If there's a BPF listener, bounce a copy of this frame
1694 		 * to him.
1695 		 */
1696 		ETHER_BPF_MTAP(ifp, m_head);
1697 	}
1698 
1699 	if (enq) {
1700 		/* Kick. */
1701 		CSR_WRITE_4(sc, ALE_MBOX_TPD_PROD_IDX,
1702 		    sc->ale_cdata.ale_tx_prod);
1703 
1704 		/* Set a timeout in case the chip goes out to lunch. */
1705 		ifp->if_timer = ALE_TX_TIMEOUT;
1706 	}
1707 }
1708 
1709 static void
1710 ale_watchdog(struct ifnet *ifp)
1711 {
1712 	struct ale_softc *sc = ifp->if_softc;
1713 
1714 	ASSERT_SERIALIZED(ifp->if_serializer);
1715 
1716 	if ((sc->ale_flags & ALE_FLAG_LINK) == 0) {
1717 		if_printf(ifp, "watchdog timeout (lost link)\n");
1718 		IFNET_STAT_INC(ifp, oerrors, 1);
1719 		ale_init(sc);
1720 		return;
1721 	}
1722 
1723 	if_printf(ifp, "watchdog timeout -- resetting\n");
1724 	IFNET_STAT_INC(ifp, oerrors, 1);
1725 	ale_init(sc);
1726 
1727 	if (!ifq_is_empty(&ifp->if_snd))
1728 		if_devstart(ifp);
1729 }
1730 
1731 static int
1732 ale_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data, struct ucred *cr)
1733 {
1734 	struct ale_softc *sc;
1735 	struct ifreq *ifr;
1736 	struct mii_data *mii;
1737 	int error, mask;
1738 
1739 	ASSERT_SERIALIZED(ifp->if_serializer);
1740 
1741 	sc = ifp->if_softc;
1742 	ifr = (struct ifreq *)data;
1743 	error = 0;
1744 
1745 	switch (cmd) {
1746 	case SIOCSIFMTU:
1747 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > ALE_JUMBO_MTU ||
1748 		    ((sc->ale_flags & ALE_FLAG_JUMBO) == 0 &&
1749 		    ifr->ifr_mtu > ETHERMTU))
1750 			error = EINVAL;
1751 		else if (ifp->if_mtu != ifr->ifr_mtu) {
1752 			ifp->if_mtu = ifr->ifr_mtu;
1753 			if ((ifp->if_flags & IFF_RUNNING) != 0)
1754 				ale_init(sc);
1755 		}
1756 		break;
1757 
1758 	case SIOCSIFFLAGS:
1759 		if ((ifp->if_flags & IFF_UP) != 0) {
1760 			if ((ifp->if_flags & IFF_RUNNING) != 0) {
1761 				if (((ifp->if_flags ^ sc->ale_if_flags)
1762 				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
1763 					ale_rxfilter(sc);
1764 			} else {
1765 				if ((sc->ale_flags & ALE_FLAG_DETACH) == 0)
1766 					ale_init(sc);
1767 			}
1768 		} else {
1769 			if ((ifp->if_flags & IFF_RUNNING) != 0)
1770 				ale_stop(sc);
1771 		}
1772 		sc->ale_if_flags = ifp->if_flags;
1773 		break;
1774 
1775 	case SIOCADDMULTI:
1776 	case SIOCDELMULTI:
1777 		if ((ifp->if_flags & IFF_RUNNING) != 0)
1778 			ale_rxfilter(sc);
1779 		break;
1780 
1781 	case SIOCSIFMEDIA:
1782 	case SIOCGIFMEDIA:
1783 		mii = device_get_softc(sc->ale_miibus);
1784 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, cmd);
1785 		break;
1786 
1787 	case SIOCSIFCAP:
1788 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1789 		if ((mask & IFCAP_TXCSUM) != 0 &&
1790 		    (ifp->if_capabilities & IFCAP_TXCSUM) != 0) {
1791 			ifp->if_capenable ^= IFCAP_TXCSUM;
1792 			if ((ifp->if_capenable & IFCAP_TXCSUM) != 0)
1793 				ifp->if_hwassist |= ALE_CSUM_FEATURES;
1794 			else
1795 				ifp->if_hwassist &= ~ALE_CSUM_FEATURES;
1796 		}
1797 		if ((mask & IFCAP_RXCSUM) != 0 &&
1798 		    (ifp->if_capabilities & IFCAP_RXCSUM) != 0)
1799 			ifp->if_capenable ^= IFCAP_RXCSUM;
1800 
1801 		if ((mask & IFCAP_VLAN_HWTAGGING) != 0 &&
1802 		    (ifp->if_capabilities & IFCAP_VLAN_HWTAGGING) != 0) {
1803 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
1804 			ale_rxvlan(sc);
1805 		}
1806 		break;
1807 
1808 	default:
1809 		error = ether_ioctl(ifp, cmd, data);
1810 		break;
1811 	}
1812 	return (error);
1813 }
1814 
1815 static void
1816 ale_mac_config(struct ale_softc *sc)
1817 {
1818 	struct mii_data *mii;
1819 	uint32_t reg;
1820 
1821 	mii = device_get_softc(sc->ale_miibus);
1822 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
1823 	reg &= ~(MAC_CFG_FULL_DUPLEX | MAC_CFG_TX_FC | MAC_CFG_RX_FC |
1824 	    MAC_CFG_SPEED_MASK);
1825 	/* Reprogram MAC with resolved speed/duplex. */
1826 	switch (IFM_SUBTYPE(mii->mii_media_active)) {
1827 	case IFM_10_T:
1828 	case IFM_100_TX:
1829 		reg |= MAC_CFG_SPEED_10_100;
1830 		break;
1831 	case IFM_1000_T:
1832 		reg |= MAC_CFG_SPEED_1000;
1833 		break;
1834 	}
1835 	if ((IFM_OPTIONS(mii->mii_media_active) & IFM_FDX) != 0) {
1836 		reg |= MAC_CFG_FULL_DUPLEX;
1837 #ifdef notyet
1838 		if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_TXPAUSE) != 0)
1839 			reg |= MAC_CFG_TX_FC;
1840 		if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_RXPAUSE) != 0)
1841 			reg |= MAC_CFG_RX_FC;
1842 #endif
1843 	}
1844 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
1845 }
1846 
1847 static void
1848 ale_stats_clear(struct ale_softc *sc)
1849 {
1850 	struct smb sb;
1851 	uint32_t *reg;
1852 	int i;
1853 
1854 	for (reg = &sb.rx_frames, i = 0; reg <= &sb.rx_pkts_filtered; reg++) {
1855 		CSR_READ_4(sc, ALE_RX_MIB_BASE + i);
1856 		i += sizeof(uint32_t);
1857 	}
1858 	/* Read Tx statistics. */
1859 	for (reg = &sb.tx_frames, i = 0; reg <= &sb.tx_mcast_bytes; reg++) {
1860 		CSR_READ_4(sc, ALE_TX_MIB_BASE + i);
1861 		i += sizeof(uint32_t);
1862 	}
1863 }
1864 
1865 static void
1866 ale_stats_update(struct ale_softc *sc)
1867 {
1868 	struct ale_hw_stats *stat;
1869 	struct smb sb, *smb;
1870 	struct ifnet *ifp;
1871 	uint32_t *reg;
1872 	int i;
1873 
1874 	ifp = &sc->arpcom.ac_if;
1875 	stat = &sc->ale_stats;
1876 	smb = &sb;
1877 
1878 	/* Read Rx statistics. */
1879 	for (reg = &sb.rx_frames, i = 0; reg <= &sb.rx_pkts_filtered; reg++) {
1880 		*reg = CSR_READ_4(sc, ALE_RX_MIB_BASE + i);
1881 		i += sizeof(uint32_t);
1882 	}
1883 	/* Read Tx statistics. */
1884 	for (reg = &sb.tx_frames, i = 0; reg <= &sb.tx_mcast_bytes; reg++) {
1885 		*reg = CSR_READ_4(sc, ALE_TX_MIB_BASE + i);
1886 		i += sizeof(uint32_t);
1887 	}
1888 
1889 	/* Rx stats. */
1890 	stat->rx_frames += smb->rx_frames;
1891 	stat->rx_bcast_frames += smb->rx_bcast_frames;
1892 	stat->rx_mcast_frames += smb->rx_mcast_frames;
1893 	stat->rx_pause_frames += smb->rx_pause_frames;
1894 	stat->rx_control_frames += smb->rx_control_frames;
1895 	stat->rx_crcerrs += smb->rx_crcerrs;
1896 	stat->rx_lenerrs += smb->rx_lenerrs;
1897 	stat->rx_bytes += smb->rx_bytes;
1898 	stat->rx_runts += smb->rx_runts;
1899 	stat->rx_fragments += smb->rx_fragments;
1900 	stat->rx_pkts_64 += smb->rx_pkts_64;
1901 	stat->rx_pkts_65_127 += smb->rx_pkts_65_127;
1902 	stat->rx_pkts_128_255 += smb->rx_pkts_128_255;
1903 	stat->rx_pkts_256_511 += smb->rx_pkts_256_511;
1904 	stat->rx_pkts_512_1023 += smb->rx_pkts_512_1023;
1905 	stat->rx_pkts_1024_1518 += smb->rx_pkts_1024_1518;
1906 	stat->rx_pkts_1519_max += smb->rx_pkts_1519_max;
1907 	stat->rx_pkts_truncated += smb->rx_pkts_truncated;
1908 	stat->rx_fifo_oflows += smb->rx_fifo_oflows;
1909 	stat->rx_rrs_errs += smb->rx_rrs_errs;
1910 	stat->rx_alignerrs += smb->rx_alignerrs;
1911 	stat->rx_bcast_bytes += smb->rx_bcast_bytes;
1912 	stat->rx_mcast_bytes += smb->rx_mcast_bytes;
1913 	stat->rx_pkts_filtered += smb->rx_pkts_filtered;
1914 
1915 	/* Tx stats. */
1916 	stat->tx_frames += smb->tx_frames;
1917 	stat->tx_bcast_frames += smb->tx_bcast_frames;
1918 	stat->tx_mcast_frames += smb->tx_mcast_frames;
1919 	stat->tx_pause_frames += smb->tx_pause_frames;
1920 	stat->tx_excess_defer += smb->tx_excess_defer;
1921 	stat->tx_control_frames += smb->tx_control_frames;
1922 	stat->tx_deferred += smb->tx_deferred;
1923 	stat->tx_bytes += smb->tx_bytes;
1924 	stat->tx_pkts_64 += smb->tx_pkts_64;
1925 	stat->tx_pkts_65_127 += smb->tx_pkts_65_127;
1926 	stat->tx_pkts_128_255 += smb->tx_pkts_128_255;
1927 	stat->tx_pkts_256_511 += smb->tx_pkts_256_511;
1928 	stat->tx_pkts_512_1023 += smb->tx_pkts_512_1023;
1929 	stat->tx_pkts_1024_1518 += smb->tx_pkts_1024_1518;
1930 	stat->tx_pkts_1519_max += smb->tx_pkts_1519_max;
1931 	stat->tx_single_colls += smb->tx_single_colls;
1932 	stat->tx_multi_colls += smb->tx_multi_colls;
1933 	stat->tx_late_colls += smb->tx_late_colls;
1934 	stat->tx_excess_colls += smb->tx_excess_colls;
1935 	stat->tx_abort += smb->tx_abort;
1936 	stat->tx_underrun += smb->tx_underrun;
1937 	stat->tx_desc_underrun += smb->tx_desc_underrun;
1938 	stat->tx_lenerrs += smb->tx_lenerrs;
1939 	stat->tx_pkts_truncated += smb->tx_pkts_truncated;
1940 	stat->tx_bcast_bytes += smb->tx_bcast_bytes;
1941 	stat->tx_mcast_bytes += smb->tx_mcast_bytes;
1942 
1943 	/* Update counters in ifnet. */
1944 	IFNET_STAT_INC(ifp, opackets, smb->tx_frames);
1945 
1946 	IFNET_STAT_INC(ifp, collisions, smb->tx_single_colls +
1947 	    smb->tx_multi_colls * 2 + smb->tx_late_colls +
1948 	    smb->tx_abort * HDPX_CFG_RETRY_DEFAULT);
1949 
1950 	/*
1951 	 * XXX
1952 	 * tx_pkts_truncated counter looks suspicious. It constantly
1953 	 * increments with no sign of Tx errors. This may indicate
1954 	 * the counter name is not correct one so I've removed the
1955 	 * counter in output errors.
1956 	 */
1957 	IFNET_STAT_INC(ifp, oerrors, smb->tx_abort + smb->tx_late_colls +
1958 	    smb->tx_underrun);
1959 
1960 	IFNET_STAT_INC(ifp, ipackets, smb->rx_frames);
1961 
1962 	IFNET_STAT_INC(ifp, ierrors, smb->rx_crcerrs + smb->rx_lenerrs +
1963 	    smb->rx_runts + smb->rx_pkts_truncated +
1964 	    smb->rx_fifo_oflows + smb->rx_rrs_errs +
1965 	    smb->rx_alignerrs);
1966 }
1967 
1968 static void
1969 ale_intr(void *xsc)
1970 {
1971 	struct ale_softc *sc = xsc;
1972 	struct ifnet *ifp = &sc->arpcom.ac_if;
1973 	uint32_t status;
1974 
1975 	ASSERT_SERIALIZED(ifp->if_serializer);
1976 
1977 	status = CSR_READ_4(sc, ALE_INTR_STATUS);
1978 	if ((status & ALE_INTRS) == 0)
1979 		return;
1980 
1981 	/* Acknowledge and disable interrupts. */
1982 	CSR_WRITE_4(sc, ALE_INTR_STATUS, status | INTR_DIS_INT);
1983 
1984 	if ((ifp->if_flags & IFF_RUNNING) != 0) {
1985 		int error;
1986 
1987 		error = ale_rxeof(sc);
1988 		if (error) {
1989 			sc->ale_stats.reset_brk_seq++;
1990 			ale_init(sc);
1991 			return;
1992 		}
1993 
1994 		if ((status & (INTR_DMA_RD_TO_RST | INTR_DMA_WR_TO_RST)) != 0) {
1995 			if ((status & INTR_DMA_RD_TO_RST) != 0)
1996 				device_printf(sc->ale_dev,
1997 				    "DMA read error! -- resetting\n");
1998 			if ((status & INTR_DMA_WR_TO_RST) != 0)
1999 				device_printf(sc->ale_dev,
2000 				    "DMA write error! -- resetting\n");
2001 			ale_init(sc);
2002 			return;
2003 		}
2004 
2005 		ale_txeof(sc);
2006 		if (!ifq_is_empty(&ifp->if_snd))
2007 			if_devstart(ifp);
2008 	}
2009 
2010 	/* Re-enable interrupts. */
2011 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0x7FFFFFFF);
2012 }
2013 
2014 static void
2015 ale_txeof(struct ale_softc *sc)
2016 {
2017 	struct ifnet *ifp = &sc->arpcom.ac_if;
2018 	struct ale_txdesc *txd;
2019 	uint32_t cons, prod;
2020 	int prog;
2021 
2022 	if (sc->ale_cdata.ale_tx_cnt == 0)
2023 		return;
2024 
2025 	bus_dmamap_sync(sc->ale_cdata.ale_tx_ring_tag,
2026 	    sc->ale_cdata.ale_tx_ring_map, BUS_DMASYNC_POSTREAD);
2027 	if ((sc->ale_flags & ALE_FLAG_TXCMB_BUG) == 0) {
2028 		bus_dmamap_sync(sc->ale_cdata.ale_tx_cmb_tag,
2029 		    sc->ale_cdata.ale_tx_cmb_map, BUS_DMASYNC_POSTREAD);
2030 		prod = *sc->ale_cdata.ale_tx_cmb & TPD_CNT_MASK;
2031 	} else
2032 		prod = CSR_READ_2(sc, ALE_TPD_CONS_IDX);
2033 	cons = sc->ale_cdata.ale_tx_cons;
2034 	/*
2035 	 * Go through our Tx list and free mbufs for those
2036 	 * frames which have been transmitted.
2037 	 */
2038 	for (prog = 0; cons != prod; prog++,
2039 	     ALE_DESC_INC(cons, ALE_TX_RING_CNT)) {
2040 		if (sc->ale_cdata.ale_tx_cnt <= 0)
2041 			break;
2042 		prog++;
2043 		ifq_clr_oactive(&ifp->if_snd);
2044 		sc->ale_cdata.ale_tx_cnt--;
2045 		txd = &sc->ale_cdata.ale_txdesc[cons];
2046 		if (txd->tx_m != NULL) {
2047 			/* Reclaim transmitted mbufs. */
2048 			bus_dmamap_unload(sc->ale_cdata.ale_tx_tag,
2049 			    txd->tx_dmamap);
2050 			m_freem(txd->tx_m);
2051 			txd->tx_m = NULL;
2052 		}
2053 	}
2054 
2055 	if (prog > 0) {
2056 		sc->ale_cdata.ale_tx_cons = cons;
2057 		/*
2058 		 * Unarm watchdog timer only when there is no pending
2059 		 * Tx descriptors in queue.
2060 		 */
2061 		if (sc->ale_cdata.ale_tx_cnt == 0)
2062 			ifp->if_timer = 0;
2063 	}
2064 }
2065 
2066 static void
2067 ale_rx_update_page(struct ale_softc *sc, struct ale_rx_page **page,
2068     uint32_t length, uint32_t *prod)
2069 {
2070 	struct ale_rx_page *rx_page;
2071 
2072 	rx_page = *page;
2073 	/* Update consumer position. */
2074 	rx_page->cons += roundup(length + sizeof(struct rx_rs),
2075 	    ALE_RX_PAGE_ALIGN);
2076 	if (rx_page->cons >= ALE_RX_PAGE_SZ) {
2077 		/*
2078 		 * End of Rx page reached, let hardware reuse
2079 		 * this page.
2080 		 */
2081 		rx_page->cons = 0;
2082 		*rx_page->cmb_addr = 0;
2083 		bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2084 				BUS_DMASYNC_PREWRITE);
2085 		CSR_WRITE_1(sc, ALE_RXF0_PAGE0 + sc->ale_cdata.ale_rx_curp,
2086 		    RXF_VALID);
2087 		/* Switch to alternate Rx page. */
2088 		sc->ale_cdata.ale_rx_curp ^= 1;
2089 		rx_page = *page =
2090 		    &sc->ale_cdata.ale_rx_page[sc->ale_cdata.ale_rx_curp];
2091 		/* Page flipped, sync CMB and Rx page. */
2092 		bus_dmamap_sync(rx_page->page_tag, rx_page->page_map,
2093 		    BUS_DMASYNC_POSTREAD);
2094 		bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2095 		    BUS_DMASYNC_POSTREAD);
2096 		/* Sync completed, cache updated producer index. */
2097 		*prod = *rx_page->cmb_addr;
2098 	}
2099 }
2100 
2101 
2102 /*
2103  * It seems that AR81xx controller can compute partial checksum.
2104  * The partial checksum value can be used to accelerate checksum
2105  * computation for fragmented TCP/UDP packets. Upper network stack
2106  * already takes advantage of the partial checksum value in IP
2107  * reassembly stage. But I'm not sure the correctness of the
2108  * partial hardware checksum assistance due to lack of data sheet.
2109  * In addition, the Rx feature of controller that requires copying
2110  * for every frames effectively nullifies one of most nice offload
2111  * capability of controller.
2112  */
2113 static void
2114 ale_rxcsum(struct ale_softc *sc, struct mbuf *m, uint32_t status)
2115 {
2116 	struct ifnet *ifp = &sc->arpcom.ac_if;
2117 	struct ip *ip;
2118 	char *p;
2119 
2120 	m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
2121 	if ((status & ALE_RD_IPCSUM_NOK) == 0)
2122 		m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
2123 
2124 	if ((sc->ale_flags & ALE_FLAG_RXCSUM_BUG) == 0) {
2125 		if (((status & ALE_RD_IPV4_FRAG) == 0) &&
2126 		    ((status & (ALE_RD_TCP | ALE_RD_UDP)) != 0) &&
2127 		    ((status & ALE_RD_TCP_UDPCSUM_NOK) == 0)) {
2128 			m->m_pkthdr.csum_flags |=
2129 			    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2130 			m->m_pkthdr.csum_data = 0xffff;
2131 		}
2132 	} else {
2133 		if ((status & (ALE_RD_TCP | ALE_RD_UDP)) != 0 &&
2134 		    (status & ALE_RD_TCP_UDPCSUM_NOK) == 0) {
2135 			p = mtod(m, char *);
2136 			p += ETHER_HDR_LEN;
2137 			if ((status & ALE_RD_802_3) != 0)
2138 				p += LLC_SNAPFRAMELEN;
2139 			if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0 &&
2140 			    (status & ALE_RD_VLAN) != 0)
2141 				p += EVL_ENCAPLEN;
2142 			ip = (struct ip *)p;
2143 			if (ip->ip_off != 0 && (status & ALE_RD_IPV4_DF) == 0)
2144 				return;
2145 			m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
2146 			    CSUM_PSEUDO_HDR;
2147 			m->m_pkthdr.csum_data = 0xffff;
2148 		}
2149 	}
2150 	/*
2151 	 * Don't mark bad checksum for TCP/UDP frames
2152 	 * as fragmented frames may always have set
2153 	 * bad checksummed bit of frame status.
2154 	 */
2155 }
2156 
2157 /* Process received frames. */
2158 static int
2159 ale_rxeof(struct ale_softc *sc)
2160 {
2161 	struct ifnet *ifp = &sc->arpcom.ac_if;
2162 	struct ale_rx_page *rx_page;
2163 	struct rx_rs *rs;
2164 	struct mbuf *m;
2165 	uint32_t length, prod, seqno, status, vtags;
2166 	int prog;
2167 
2168 	rx_page = &sc->ale_cdata.ale_rx_page[sc->ale_cdata.ale_rx_curp];
2169 	bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2170 			BUS_DMASYNC_POSTREAD);
2171 	bus_dmamap_sync(rx_page->page_tag, rx_page->page_map,
2172 			BUS_DMASYNC_POSTREAD);
2173 	/*
2174 	 * Don't directly access producer index as hardware may
2175 	 * update it while Rx handler is in progress. It would
2176 	 * be even better if there is a way to let hardware
2177 	 * know how far driver processed its received frames.
2178 	 * Alternatively, hardware could provide a way to disable
2179 	 * CMB updates until driver acknowledges the end of CMB
2180 	 * access.
2181 	 */
2182 	prod = *rx_page->cmb_addr;
2183 	for (prog = 0; ; prog++) {
2184 		if (rx_page->cons >= prod)
2185 			break;
2186 		rs = (struct rx_rs *)(rx_page->page_addr + rx_page->cons);
2187 		seqno = ALE_RX_SEQNO(le32toh(rs->seqno));
2188 		if (sc->ale_cdata.ale_rx_seqno != seqno) {
2189 			/*
2190 			 * Normally I believe this should not happen unless
2191 			 * severe driver bug or corrupted memory. However
2192 			 * it seems to happen under certain conditions which
2193 			 * is triggered by abrupt Rx events such as initiation
2194 			 * of bulk transfer of remote host. It's not easy to
2195 			 * reproduce this and I doubt it could be related
2196 			 * with FIFO overflow of hardware or activity of Tx
2197 			 * CMB updates. I also remember similar behaviour
2198 			 * seen on RealTek 8139 which uses resembling Rx
2199 			 * scheme.
2200 			 */
2201 			if (bootverbose)
2202 				device_printf(sc->ale_dev,
2203 				    "garbled seq: %u, expected: %u -- "
2204 				    "resetting!\n", seqno,
2205 				    sc->ale_cdata.ale_rx_seqno);
2206 			return (EIO);
2207 		}
2208 		/* Frame received. */
2209 		sc->ale_cdata.ale_rx_seqno++;
2210 		length = ALE_RX_BYTES(le32toh(rs->length));
2211 		status = le32toh(rs->flags);
2212 		if ((status & ALE_RD_ERROR) != 0) {
2213 			/*
2214 			 * We want to pass the following frames to upper
2215 			 * layer regardless of error status of Rx return
2216 			 * status.
2217 			 *
2218 			 *  o IP/TCP/UDP checksum is bad.
2219 			 *  o frame length and protocol specific length
2220 			 *     does not match.
2221 			 */
2222 			if ((status & (ALE_RD_CRC | ALE_RD_CODE |
2223 			    ALE_RD_DRIBBLE | ALE_RD_RUNT | ALE_RD_OFLOW |
2224 			    ALE_RD_TRUNC)) != 0) {
2225 				ale_rx_update_page(sc, &rx_page, length, &prod);
2226 				continue;
2227 			}
2228 		}
2229 		/*
2230 		 * m_devget(9) is major bottle-neck of ale(4) (It comes
2231 		 * from hardware limitation). For jumbo frames we could
2232 		 * get a slightly better performance if driver use
2233 		 * m_getjcl(9) with proper buffer size argument. However
2234 		 * that would make code more complicated and I don't
2235 		 * think users would expect good Rx performance numbers
2236 		 * on these low-end consumer ethernet controller.
2237 		 */
2238 		m = m_devget((char *)(rs + 1), length - ETHER_CRC_LEN,
2239 			     ETHER_ALIGN, ifp);
2240 		if (m == NULL) {
2241 			IFNET_STAT_INC(ifp, iqdrops, 1);
2242 			ale_rx_update_page(sc, &rx_page, length, &prod);
2243 			continue;
2244 		}
2245 		if ((ifp->if_capenable & IFCAP_RXCSUM) != 0 &&
2246 		    (status & ALE_RD_IPV4) != 0)
2247 			ale_rxcsum(sc, m, status);
2248 		if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0 &&
2249 		    (status & ALE_RD_VLAN) != 0) {
2250 			vtags = ALE_RX_VLAN(le32toh(rs->vtags));
2251 			m->m_pkthdr.ether_vlantag = ALE_RX_VLAN_TAG(vtags);
2252 			m->m_flags |= M_VLANTAG;
2253 		}
2254 
2255 		/* Pass it to upper layer. */
2256 		ifp->if_input(ifp, m, NULL, -1);
2257 
2258 		ale_rx_update_page(sc, &rx_page, length, &prod);
2259 	}
2260 	return 0;
2261 }
2262 
2263 static void
2264 ale_tick(void *xsc)
2265 {
2266 	struct ale_softc *sc = xsc;
2267 	struct ifnet *ifp = &sc->arpcom.ac_if;
2268 	struct mii_data *mii;
2269 
2270 	lwkt_serialize_enter(ifp->if_serializer);
2271 
2272 	mii = device_get_softc(sc->ale_miibus);
2273 	mii_tick(mii);
2274 	ale_stats_update(sc);
2275 
2276 	callout_reset(&sc->ale_tick_ch, hz, ale_tick, sc);
2277 
2278 	lwkt_serialize_exit(ifp->if_serializer);
2279 }
2280 
2281 static void
2282 ale_reset(struct ale_softc *sc)
2283 {
2284 	uint32_t reg;
2285 	int i;
2286 
2287 	/* Initialize PCIe module. From Linux. */
2288 	CSR_WRITE_4(sc, 0x1008, CSR_READ_4(sc, 0x1008) | 0x8000);
2289 
2290 	CSR_WRITE_4(sc, ALE_MASTER_CFG, MASTER_RESET);
2291 	for (i = ALE_RESET_TIMEOUT; i > 0; i--) {
2292 		DELAY(10);
2293 		if ((CSR_READ_4(sc, ALE_MASTER_CFG) & MASTER_RESET) == 0)
2294 			break;
2295 	}
2296 	if (i == 0)
2297 		device_printf(sc->ale_dev, "master reset timeout!\n");
2298 
2299 	for (i = ALE_RESET_TIMEOUT; i > 0; i--) {
2300 		if ((reg = CSR_READ_4(sc, ALE_IDLE_STATUS)) == 0)
2301 			break;
2302 		DELAY(10);
2303 	}
2304 
2305 	if (i == 0)
2306 		device_printf(sc->ale_dev, "reset timeout(0x%08x)!\n", reg);
2307 }
2308 
2309 static void
2310 ale_init(void *xsc)
2311 {
2312 	struct ale_softc *sc = xsc;
2313 	struct ifnet *ifp = &sc->arpcom.ac_if;
2314 	struct mii_data *mii;
2315 	uint8_t eaddr[ETHER_ADDR_LEN];
2316 	bus_addr_t paddr;
2317 	uint32_t reg, rxf_hi, rxf_lo;
2318 
2319 	ASSERT_SERIALIZED(ifp->if_serializer);
2320 
2321 	mii = device_get_softc(sc->ale_miibus);
2322 
2323 	/*
2324 	 * Cancel any pending I/O.
2325 	 */
2326 	ale_stop(sc);
2327 
2328 	/*
2329 	 * Reset the chip to a known state.
2330 	 */
2331 	ale_reset(sc);
2332 
2333 	/* Initialize Tx descriptors, DMA memory blocks. */
2334 	ale_init_rx_pages(sc);
2335 	ale_init_tx_ring(sc);
2336 
2337 	/* Reprogram the station address. */
2338 	bcopy(IF_LLADDR(ifp), eaddr, ETHER_ADDR_LEN);
2339 	CSR_WRITE_4(sc, ALE_PAR0,
2340 	    eaddr[2] << 24 | eaddr[3] << 16 | eaddr[4] << 8 | eaddr[5]);
2341 	CSR_WRITE_4(sc, ALE_PAR1, eaddr[0] << 8 | eaddr[1]);
2342 
2343 	/*
2344 	 * Clear WOL status and disable all WOL feature as WOL
2345 	 * would interfere Rx operation under normal environments.
2346 	 */
2347 	CSR_READ_4(sc, ALE_WOL_CFG);
2348 	CSR_WRITE_4(sc, ALE_WOL_CFG, 0);
2349 
2350 	/*
2351 	 * Set Tx descriptor/RXF0/CMB base addresses. They share
2352 	 * the same high address part of DMAable region.
2353 	 */
2354 	paddr = sc->ale_cdata.ale_tx_ring_paddr;
2355 	CSR_WRITE_4(sc, ALE_TPD_ADDR_HI, ALE_ADDR_HI(paddr));
2356 	CSR_WRITE_4(sc, ALE_TPD_ADDR_LO, ALE_ADDR_LO(paddr));
2357 	CSR_WRITE_4(sc, ALE_TPD_CNT,
2358 	    (ALE_TX_RING_CNT << TPD_CNT_SHIFT) & TPD_CNT_MASK);
2359 
2360 	/* Set Rx page base address, note we use single queue. */
2361 	paddr = sc->ale_cdata.ale_rx_page[0].page_paddr;
2362 	CSR_WRITE_4(sc, ALE_RXF0_PAGE0_ADDR_LO, ALE_ADDR_LO(paddr));
2363 	paddr = sc->ale_cdata.ale_rx_page[1].page_paddr;
2364 	CSR_WRITE_4(sc, ALE_RXF0_PAGE1_ADDR_LO, ALE_ADDR_LO(paddr));
2365 
2366 	/* Set Tx/Rx CMB addresses. */
2367 	paddr = sc->ale_cdata.ale_tx_cmb_paddr;
2368 	CSR_WRITE_4(sc, ALE_TX_CMB_ADDR_LO, ALE_ADDR_LO(paddr));
2369 	paddr = sc->ale_cdata.ale_rx_page[0].cmb_paddr;
2370 	CSR_WRITE_4(sc, ALE_RXF0_CMB0_ADDR_LO, ALE_ADDR_LO(paddr));
2371 	paddr = sc->ale_cdata.ale_rx_page[1].cmb_paddr;
2372 	CSR_WRITE_4(sc, ALE_RXF0_CMB1_ADDR_LO, ALE_ADDR_LO(paddr));
2373 
2374 	/* Mark RXF0 is valid. */
2375 	CSR_WRITE_1(sc, ALE_RXF0_PAGE0, RXF_VALID);
2376 	CSR_WRITE_1(sc, ALE_RXF0_PAGE1, RXF_VALID);
2377 	/*
2378 	 * No need to initialize RFX1/RXF2/RXF3. We don't use
2379 	 * multi-queue yet.
2380 	 */
2381 
2382 	/* Set Rx page size, excluding guard frame size. */
2383 	CSR_WRITE_4(sc, ALE_RXF_PAGE_SIZE, ALE_RX_PAGE_SZ);
2384 
2385 	/* Tell hardware that we're ready to load DMA blocks. */
2386 	CSR_WRITE_4(sc, ALE_DMA_BLOCK, DMA_BLOCK_LOAD);
2387 
2388 	/* Set Rx/Tx interrupt trigger threshold. */
2389 	CSR_WRITE_4(sc, ALE_INT_TRIG_THRESH, (1 << INT_TRIG_RX_THRESH_SHIFT) |
2390 	    (4 << INT_TRIG_TX_THRESH_SHIFT));
2391 	/*
2392 	 * XXX
2393 	 * Set interrupt trigger timer, its purpose and relation
2394 	 * with interrupt moderation mechanism is not clear yet.
2395 	 */
2396 	CSR_WRITE_4(sc, ALE_INT_TRIG_TIMER,
2397 	    ((ALE_USECS(10) << INT_TRIG_RX_TIMER_SHIFT) |
2398 	    (ALE_USECS(1000) << INT_TRIG_TX_TIMER_SHIFT)));
2399 
2400 	/* Configure interrupt moderation timer. */
2401 	reg = ALE_USECS(sc->ale_int_rx_mod) << IM_TIMER_RX_SHIFT;
2402 	reg |= ALE_USECS(sc->ale_int_tx_mod) << IM_TIMER_TX_SHIFT;
2403 	CSR_WRITE_4(sc, ALE_IM_TIMER, reg);
2404 	reg = CSR_READ_4(sc, ALE_MASTER_CFG);
2405 	reg &= ~(MASTER_CHIP_REV_MASK | MASTER_CHIP_ID_MASK);
2406 	reg &= ~(MASTER_IM_RX_TIMER_ENB | MASTER_IM_TX_TIMER_ENB);
2407 	if (ALE_USECS(sc->ale_int_rx_mod) != 0)
2408 		reg |= MASTER_IM_RX_TIMER_ENB;
2409 	if (ALE_USECS(sc->ale_int_tx_mod) != 0)
2410 		reg |= MASTER_IM_TX_TIMER_ENB;
2411 	CSR_WRITE_4(sc, ALE_MASTER_CFG, reg);
2412 	CSR_WRITE_2(sc, ALE_INTR_CLR_TIMER, ALE_USECS(1000));
2413 
2414 	/* Set Maximum frame size of controller. */
2415 	if (ifp->if_mtu < ETHERMTU)
2416 		sc->ale_max_frame_size = ETHERMTU;
2417 	else
2418 		sc->ale_max_frame_size = ifp->if_mtu;
2419 	sc->ale_max_frame_size += ETHER_HDR_LEN + EVL_ENCAPLEN + ETHER_CRC_LEN;
2420 	CSR_WRITE_4(sc, ALE_FRAME_SIZE, sc->ale_max_frame_size);
2421 
2422 	/* Configure IPG/IFG parameters. */
2423 	CSR_WRITE_4(sc, ALE_IPG_IFG_CFG,
2424 	    ((IPG_IFG_IPGT_DEFAULT << IPG_IFG_IPGT_SHIFT) & IPG_IFG_IPGT_MASK) |
2425 	    ((IPG_IFG_MIFG_DEFAULT << IPG_IFG_MIFG_SHIFT) & IPG_IFG_MIFG_MASK) |
2426 	    ((IPG_IFG_IPG1_DEFAULT << IPG_IFG_IPG1_SHIFT) & IPG_IFG_IPG1_MASK) |
2427 	    ((IPG_IFG_IPG2_DEFAULT << IPG_IFG_IPG2_SHIFT) & IPG_IFG_IPG2_MASK));
2428 
2429 	/* Set parameters for half-duplex media. */
2430 	CSR_WRITE_4(sc, ALE_HDPX_CFG,
2431 	    ((HDPX_CFG_LCOL_DEFAULT << HDPX_CFG_LCOL_SHIFT) &
2432 	    HDPX_CFG_LCOL_MASK) |
2433 	    ((HDPX_CFG_RETRY_DEFAULT << HDPX_CFG_RETRY_SHIFT) &
2434 	    HDPX_CFG_RETRY_MASK) | HDPX_CFG_EXC_DEF_EN |
2435 	    ((HDPX_CFG_ABEBT_DEFAULT << HDPX_CFG_ABEBT_SHIFT) &
2436 	    HDPX_CFG_ABEBT_MASK) |
2437 	    ((HDPX_CFG_JAMIPG_DEFAULT << HDPX_CFG_JAMIPG_SHIFT) &
2438 	    HDPX_CFG_JAMIPG_MASK));
2439 
2440 	/* Configure Tx jumbo frame parameters. */
2441 	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0) {
2442 		if (ifp->if_mtu < ETHERMTU)
2443 			reg = sc->ale_max_frame_size;
2444 		else if (ifp->if_mtu < 6 * 1024)
2445 			reg = (sc->ale_max_frame_size * 2) / 3;
2446 		else
2447 			reg = sc->ale_max_frame_size / 2;
2448 		CSR_WRITE_4(sc, ALE_TX_JUMBO_THRESH,
2449 		    roundup(reg, TX_JUMBO_THRESH_UNIT) >>
2450 		    TX_JUMBO_THRESH_UNIT_SHIFT);
2451 	}
2452 
2453 	/* Configure TxQ. */
2454 	reg = (128 << (sc->ale_dma_rd_burst >> DMA_CFG_RD_BURST_SHIFT))
2455 	    << TXQ_CFG_TX_FIFO_BURST_SHIFT;
2456 	reg |= (TXQ_CFG_TPD_BURST_DEFAULT << TXQ_CFG_TPD_BURST_SHIFT) &
2457 	    TXQ_CFG_TPD_BURST_MASK;
2458 	CSR_WRITE_4(sc, ALE_TXQ_CFG, reg | TXQ_CFG_ENHANCED_MODE | TXQ_CFG_ENB);
2459 
2460 	/* Configure Rx jumbo frame & flow control parameters. */
2461 	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0) {
2462 		reg = roundup(sc->ale_max_frame_size, RX_JUMBO_THRESH_UNIT);
2463 		CSR_WRITE_4(sc, ALE_RX_JUMBO_THRESH,
2464 		    (((reg >> RX_JUMBO_THRESH_UNIT_SHIFT) <<
2465 		    RX_JUMBO_THRESH_MASK_SHIFT) & RX_JUMBO_THRESH_MASK) |
2466 		    ((RX_JUMBO_LKAH_DEFAULT << RX_JUMBO_LKAH_SHIFT) &
2467 		    RX_JUMBO_LKAH_MASK));
2468 		reg = CSR_READ_4(sc, ALE_SRAM_RX_FIFO_LEN);
2469 		rxf_hi = (reg * 7) / 10;
2470 		rxf_lo = (reg * 3)/ 10;
2471 		CSR_WRITE_4(sc, ALE_RX_FIFO_PAUSE_THRESH,
2472 		    ((rxf_lo << RX_FIFO_PAUSE_THRESH_LO_SHIFT) &
2473 		    RX_FIFO_PAUSE_THRESH_LO_MASK) |
2474 		    ((rxf_hi << RX_FIFO_PAUSE_THRESH_HI_SHIFT) &
2475 		     RX_FIFO_PAUSE_THRESH_HI_MASK));
2476 	}
2477 
2478 	/* Disable RSS. */
2479 	CSR_WRITE_4(sc, ALE_RSS_IDT_TABLE0, 0);
2480 	CSR_WRITE_4(sc, ALE_RSS_CPU, 0);
2481 
2482 	/* Configure RxQ. */
2483 	CSR_WRITE_4(sc, ALE_RXQ_CFG,
2484 	    RXQ_CFG_ALIGN_32 | RXQ_CFG_CUT_THROUGH_ENB | RXQ_CFG_ENB);
2485 
2486 	/* Configure DMA parameters. */
2487 	reg = 0;
2488 	if ((sc->ale_flags & ALE_FLAG_TXCMB_BUG) == 0)
2489 		reg |= DMA_CFG_TXCMB_ENB;
2490 	CSR_WRITE_4(sc, ALE_DMA_CFG,
2491 	    DMA_CFG_OUT_ORDER | DMA_CFG_RD_REQ_PRI | DMA_CFG_RCB_64 |
2492 	    sc->ale_dma_rd_burst | reg |
2493 	    sc->ale_dma_wr_burst | DMA_CFG_RXCMB_ENB |
2494 	    ((DMA_CFG_RD_DELAY_CNT_DEFAULT << DMA_CFG_RD_DELAY_CNT_SHIFT) &
2495 	    DMA_CFG_RD_DELAY_CNT_MASK) |
2496 	    ((DMA_CFG_WR_DELAY_CNT_DEFAULT << DMA_CFG_WR_DELAY_CNT_SHIFT) &
2497 	    DMA_CFG_WR_DELAY_CNT_MASK));
2498 
2499 	/*
2500 	 * Hardware can be configured to issue SMB interrupt based
2501 	 * on programmed interval. Since there is a callout that is
2502 	 * invoked for every hz in driver we use that instead of
2503 	 * relying on periodic SMB interrupt.
2504 	 */
2505 	CSR_WRITE_4(sc, ALE_SMB_STAT_TIMER, ALE_USECS(0));
2506 
2507 	/* Clear MAC statistics. */
2508 	ale_stats_clear(sc);
2509 
2510 	/*
2511 	 * Configure Tx/Rx MACs.
2512 	 *  - Auto-padding for short frames.
2513 	 *  - Enable CRC generation.
2514 	 *  Actual reconfiguration of MAC for resolved speed/duplex
2515 	 *  is followed after detection of link establishment.
2516 	 *  AR81xx always does checksum computation regardless of
2517 	 *  MAC_CFG_RXCSUM_ENB bit. In fact, setting the bit will
2518 	 *  cause Rx handling issue for fragmented IP datagrams due
2519 	 *  to silicon bug.
2520 	 */
2521 	reg = MAC_CFG_TX_CRC_ENB | MAC_CFG_TX_AUTO_PAD | MAC_CFG_FULL_DUPLEX |
2522 	    ((MAC_CFG_PREAMBLE_DEFAULT << MAC_CFG_PREAMBLE_SHIFT) &
2523 	    MAC_CFG_PREAMBLE_MASK);
2524 	if ((sc->ale_flags & ALE_FLAG_FASTETHER) != 0)
2525 		reg |= MAC_CFG_SPEED_10_100;
2526 	else
2527 		reg |= MAC_CFG_SPEED_1000;
2528 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2529 
2530 	/* Set up the receive filter. */
2531 	ale_rxfilter(sc);
2532 	ale_rxvlan(sc);
2533 
2534 	/* Acknowledge all pending interrupts and clear it. */
2535 	CSR_WRITE_4(sc, ALE_INTR_MASK, ALE_INTRS);
2536 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0xFFFFFFFF);
2537 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0);
2538 
2539 	sc->ale_flags &= ~ALE_FLAG_LINK;
2540 
2541 	/* Switch to the current media. */
2542 	mii_mediachg(mii);
2543 
2544 	callout_reset(&sc->ale_tick_ch, hz, ale_tick, sc);
2545 
2546 	ifp->if_flags |= IFF_RUNNING;
2547 	ifq_clr_oactive(&ifp->if_snd);
2548 }
2549 
2550 static void
2551 ale_stop(struct ale_softc *sc)
2552 {
2553 	struct ifnet *ifp = &sc->arpcom.ac_if;
2554 	struct ale_txdesc *txd;
2555 	uint32_t reg;
2556 	int i;
2557 
2558 	ASSERT_SERIALIZED(ifp->if_serializer);
2559 
2560 	/*
2561 	 * Mark the interface down and cancel the watchdog timer.
2562 	 */
2563 	ifp->if_flags &= ~IFF_RUNNING;
2564 	ifq_clr_oactive(&ifp->if_snd);
2565 	ifp->if_timer = 0;
2566 
2567 	callout_stop(&sc->ale_tick_ch);
2568 	sc->ale_flags &= ~ALE_FLAG_LINK;
2569 
2570 	ale_stats_update(sc);
2571 
2572 	/* Disable interrupts. */
2573 	CSR_WRITE_4(sc, ALE_INTR_MASK, 0);
2574 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0xFFFFFFFF);
2575 
2576 	/* Disable queue processing and DMA. */
2577 	reg = CSR_READ_4(sc, ALE_TXQ_CFG);
2578 	reg &= ~TXQ_CFG_ENB;
2579 	CSR_WRITE_4(sc, ALE_TXQ_CFG, reg);
2580 	reg = CSR_READ_4(sc, ALE_RXQ_CFG);
2581 	reg &= ~RXQ_CFG_ENB;
2582 	CSR_WRITE_4(sc, ALE_RXQ_CFG, reg);
2583 	reg = CSR_READ_4(sc, ALE_DMA_CFG);
2584 	reg &= ~(DMA_CFG_TXCMB_ENB | DMA_CFG_RXCMB_ENB);
2585 	CSR_WRITE_4(sc, ALE_DMA_CFG, reg);
2586 	DELAY(1000);
2587 
2588 	/* Stop Rx/Tx MACs. */
2589 	ale_stop_mac(sc);
2590 
2591 	/* Disable interrupts again? XXX */
2592 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0xFFFFFFFF);
2593 
2594 	/*
2595 	 * Free TX mbufs still in the queues.
2596 	 */
2597 	for (i = 0; i < ALE_TX_RING_CNT; i++) {
2598 		txd = &sc->ale_cdata.ale_txdesc[i];
2599 		if (txd->tx_m != NULL) {
2600 			bus_dmamap_unload(sc->ale_cdata.ale_tx_tag,
2601 			    txd->tx_dmamap);
2602 			m_freem(txd->tx_m);
2603 			txd->tx_m = NULL;
2604 		}
2605         }
2606 }
2607 
2608 static void
2609 ale_stop_mac(struct ale_softc *sc)
2610 {
2611 	uint32_t reg;
2612 	int i;
2613 
2614 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
2615 	if ((reg & (MAC_CFG_TX_ENB | MAC_CFG_RX_ENB)) != 0) {
2616 		reg &= ~MAC_CFG_TX_ENB | MAC_CFG_RX_ENB;
2617 		CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2618 	}
2619 
2620 	for (i = ALE_TIMEOUT; i > 0; i--) {
2621 		reg = CSR_READ_4(sc, ALE_IDLE_STATUS);
2622 		if (reg == 0)
2623 			break;
2624 		DELAY(10);
2625 	}
2626 	if (i == 0)
2627 		device_printf(sc->ale_dev,
2628 		    "could not disable Tx/Rx MAC(0x%08x)!\n", reg);
2629 }
2630 
2631 static void
2632 ale_init_tx_ring(struct ale_softc *sc)
2633 {
2634 	struct ale_txdesc *txd;
2635 	int i;
2636 
2637 	sc->ale_cdata.ale_tx_prod = 0;
2638 	sc->ale_cdata.ale_tx_cons = 0;
2639 	sc->ale_cdata.ale_tx_cnt = 0;
2640 
2641 	bzero(sc->ale_cdata.ale_tx_ring, ALE_TX_RING_SZ);
2642 	bzero(sc->ale_cdata.ale_tx_cmb, ALE_TX_CMB_SZ);
2643 	for (i = 0; i < ALE_TX_RING_CNT; i++) {
2644 		txd = &sc->ale_cdata.ale_txdesc[i];
2645 		txd->tx_m = NULL;
2646 	}
2647 	*sc->ale_cdata.ale_tx_cmb = 0;
2648 	bus_dmamap_sync(sc->ale_cdata.ale_tx_cmb_tag,
2649 	    sc->ale_cdata.ale_tx_cmb_map,
2650 	    BUS_DMASYNC_PREWRITE);
2651 	bus_dmamap_sync(sc->ale_cdata.ale_tx_ring_tag,
2652 	    sc->ale_cdata.ale_tx_ring_map,
2653 	    BUS_DMASYNC_PREWRITE);
2654 }
2655 
2656 static void
2657 ale_init_rx_pages(struct ale_softc *sc)
2658 {
2659 	struct ale_rx_page *rx_page;
2660 	int i;
2661 
2662 	sc->ale_cdata.ale_rx_seqno = 0;
2663 	sc->ale_cdata.ale_rx_curp = 0;
2664 
2665 	for (i = 0; i < ALE_RX_PAGES; i++) {
2666 		rx_page = &sc->ale_cdata.ale_rx_page[i];
2667 		bzero(rx_page->page_addr, sc->ale_pagesize);
2668 		bzero(rx_page->cmb_addr, ALE_RX_CMB_SZ);
2669 		rx_page->cons = 0;
2670 		*rx_page->cmb_addr = 0;
2671 		bus_dmamap_sync(rx_page->page_tag, rx_page->page_map,
2672 				BUS_DMASYNC_PREWRITE);
2673 		bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2674 				BUS_DMASYNC_PREWRITE);
2675 	}
2676 }
2677 
2678 static void
2679 ale_rxvlan(struct ale_softc *sc)
2680 {
2681 	struct ifnet *ifp;
2682 	uint32_t reg;
2683 
2684 	ifp = &sc->arpcom.ac_if;
2685 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
2686 	reg &= ~MAC_CFG_VLAN_TAG_STRIP;
2687 	if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
2688 		reg |= MAC_CFG_VLAN_TAG_STRIP;
2689 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2690 }
2691 
2692 static void
2693 ale_rxfilter(struct ale_softc *sc)
2694 {
2695 	struct ifnet *ifp;
2696 	struct ifmultiaddr *ifma;
2697 	uint32_t crc;
2698 	uint32_t mchash[2];
2699 	uint32_t rxcfg;
2700 
2701 	ifp = &sc->arpcom.ac_if;
2702 
2703 	rxcfg = CSR_READ_4(sc, ALE_MAC_CFG);
2704 	rxcfg &= ~(MAC_CFG_ALLMULTI | MAC_CFG_BCAST | MAC_CFG_PROMISC);
2705 	if ((ifp->if_flags & IFF_BROADCAST) != 0)
2706 		rxcfg |= MAC_CFG_BCAST;
2707 	if ((ifp->if_flags & (IFF_PROMISC | IFF_ALLMULTI)) != 0) {
2708 		if ((ifp->if_flags & IFF_PROMISC) != 0)
2709 			rxcfg |= MAC_CFG_PROMISC;
2710 		if ((ifp->if_flags & IFF_ALLMULTI) != 0)
2711 			rxcfg |= MAC_CFG_ALLMULTI;
2712 		CSR_WRITE_4(sc, ALE_MAR0, 0xFFFFFFFF);
2713 		CSR_WRITE_4(sc, ALE_MAR1, 0xFFFFFFFF);
2714 		CSR_WRITE_4(sc, ALE_MAC_CFG, rxcfg);
2715 		return;
2716 	}
2717 
2718 	/* Program new filter. */
2719 	bzero(mchash, sizeof(mchash));
2720 
2721 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2722 		if (ifma->ifma_addr->sa_family != AF_LINK)
2723 			continue;
2724 		crc = ether_crc32_le(LLADDR((struct sockaddr_dl *)
2725 		    ifma->ifma_addr), ETHER_ADDR_LEN);
2726 		mchash[crc >> 31] |= 1 << ((crc >> 26) & 0x1f);
2727 	}
2728 
2729 	CSR_WRITE_4(sc, ALE_MAR0, mchash[0]);
2730 	CSR_WRITE_4(sc, ALE_MAR1, mchash[1]);
2731 	CSR_WRITE_4(sc, ALE_MAC_CFG, rxcfg);
2732 }
2733 
2734 static int
2735 sysctl_hw_ale_int_mod(SYSCTL_HANDLER_ARGS)
2736 {
2737 	return (sysctl_int_range(oidp, arg1, arg2, req,
2738 	    ALE_IM_TIMER_MIN, ALE_IM_TIMER_MAX));
2739 }
2740 
2741 static void
2742 ale_dmamap_buf_cb(void *xctx, bus_dma_segment_t *segs, int nsegs,
2743 		  bus_size_t mapsz __unused, int error)
2744 {
2745 	struct ale_dmamap_ctx *ctx = xctx;
2746 	int i;
2747 
2748 	if (error)
2749 		return;
2750 
2751 	if (nsegs > ctx->nsegs) {
2752 		ctx->nsegs = 0;
2753 		return;
2754 	}
2755 
2756 	ctx->nsegs = nsegs;
2757 	for (i = 0; i < nsegs; ++i)
2758 		ctx->segs[i] = segs[i];
2759 }
2760