if_ale.c revision 217542
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
28/* Driver for Atheros AR8121/AR8113/AR8114 PCIe Ethernet. */
29
30#include <sys/cdefs.h>
31__FBSDID("$FreeBSD: head/sys/dev/ale/if_ale.c 217542 2011-01-18 16:27:40Z jhb $");
32
33#include <sys/param.h>
34#include <sys/systm.h>
35#include <sys/bus.h>
36#include <sys/endian.h>
37#include <sys/kernel.h>
38#include <sys/malloc.h>
39#include <sys/mbuf.h>
40#include <sys/module.h>
41#include <sys/rman.h>
42#include <sys/queue.h>
43#include <sys/socket.h>
44#include <sys/sockio.h>
45#include <sys/sysctl.h>
46#include <sys/taskqueue.h>
47
48#include <net/bpf.h>
49#include <net/if.h>
50#include <net/if_arp.h>
51#include <net/ethernet.h>
52#include <net/if_dl.h>
53#include <net/if_llc.h>
54#include <net/if_media.h>
55#include <net/if_types.h>
56#include <net/if_vlan_var.h>
57
58#include <netinet/in.h>
59#include <netinet/in_systm.h>
60#include <netinet/ip.h>
61#include <netinet/tcp.h>
62
63#include <dev/mii/mii.h>
64#include <dev/mii/miivar.h>
65
66#include <dev/pci/pcireg.h>
67#include <dev/pci/pcivar.h>
68
69#include <machine/bus.h>
70#include <machine/in_cksum.h>
71
72#include <dev/ale/if_alereg.h>
73#include <dev/ale/if_alevar.h>
74
75/* "device miibus" required.  See GENERIC if you get errors here. */
76#include "miibus_if.h"
77
78/* For more information about Tx checksum offload issues see ale_encap(). */
79#define	ALE_CSUM_FEATURES	(CSUM_TCP | CSUM_UDP)
80
81MODULE_DEPEND(ale, pci, 1, 1, 1);
82MODULE_DEPEND(ale, ether, 1, 1, 1);
83MODULE_DEPEND(ale, miibus, 1, 1, 1);
84
85/* Tunables. */
86static int msi_disable = 0;
87static int msix_disable = 0;
88TUNABLE_INT("hw.ale.msi_disable", &msi_disable);
89TUNABLE_INT("hw.ale.msix_disable", &msix_disable);
90
91/*
92 * Devices supported by this driver.
93 */
94static struct ale_dev {
95	uint16_t	ale_vendorid;
96	uint16_t	ale_deviceid;
97	const char	*ale_name;
98} ale_devs[] = {
99    { VENDORID_ATHEROS, DEVICEID_ATHEROS_AR81XX,
100    "Atheros AR8121/AR8113/AR8114 PCIe Ethernet" },
101};
102
103static int	ale_attach(device_t);
104static int	ale_check_boundary(struct ale_softc *);
105static int	ale_detach(device_t);
106static int	ale_dma_alloc(struct ale_softc *);
107static void	ale_dma_free(struct ale_softc *);
108static void	ale_dmamap_cb(void *, bus_dma_segment_t *, int, int);
109static int	ale_encap(struct ale_softc *, struct mbuf **);
110static void	ale_get_macaddr(struct ale_softc *);
111static void	ale_init(void *);
112static void	ale_init_locked(struct ale_softc *);
113static void	ale_init_rx_pages(struct ale_softc *);
114static void	ale_init_tx_ring(struct ale_softc *);
115static void	ale_int_task(void *, int);
116static int	ale_intr(void *);
117static int	ale_ioctl(struct ifnet *, u_long, caddr_t);
118static void	ale_link_task(void *, int);
119static void	ale_mac_config(struct ale_softc *);
120static int	ale_miibus_readreg(device_t, int, int);
121static void	ale_miibus_statchg(device_t);
122static int	ale_miibus_writereg(device_t, int, int, int);
123static int	ale_mediachange(struct ifnet *);
124static void	ale_mediastatus(struct ifnet *, struct ifmediareq *);
125static void	ale_phy_reset(struct ale_softc *);
126static int	ale_probe(device_t);
127static void	ale_reset(struct ale_softc *);
128static int	ale_resume(device_t);
129static void	ale_rx_update_page(struct ale_softc *, struct ale_rx_page **,
130    uint32_t, uint32_t *);
131static void	ale_rxcsum(struct ale_softc *, struct mbuf *, uint32_t);
132static int	ale_rxeof(struct ale_softc *sc, int);
133static void	ale_rxfilter(struct ale_softc *);
134static void	ale_rxvlan(struct ale_softc *);
135static void	ale_setlinkspeed(struct ale_softc *);
136static void	ale_setwol(struct ale_softc *);
137static int	ale_shutdown(device_t);
138static void	ale_start(struct ifnet *);
139static void	ale_start_locked(struct ifnet *);
140static void	ale_stats_clear(struct ale_softc *);
141static void	ale_stats_update(struct ale_softc *);
142static void	ale_stop(struct ale_softc *);
143static void	ale_stop_mac(struct ale_softc *);
144static int	ale_suspend(device_t);
145static void	ale_sysctl_node(struct ale_softc *);
146static void	ale_tick(void *);
147static void	ale_txeof(struct ale_softc *);
148static void	ale_watchdog(struct ale_softc *);
149static int	sysctl_int_range(SYSCTL_HANDLER_ARGS, int, int);
150static int	sysctl_hw_ale_proc_limit(SYSCTL_HANDLER_ARGS);
151static int	sysctl_hw_ale_int_mod(SYSCTL_HANDLER_ARGS);
152
153static device_method_t ale_methods[] = {
154	/* Device interface. */
155	DEVMETHOD(device_probe,		ale_probe),
156	DEVMETHOD(device_attach,	ale_attach),
157	DEVMETHOD(device_detach,	ale_detach),
158	DEVMETHOD(device_shutdown,	ale_shutdown),
159	DEVMETHOD(device_suspend,	ale_suspend),
160	DEVMETHOD(device_resume,	ale_resume),
161
162	/* MII interface. */
163	DEVMETHOD(miibus_readreg,	ale_miibus_readreg),
164	DEVMETHOD(miibus_writereg,	ale_miibus_writereg),
165	DEVMETHOD(miibus_statchg,	ale_miibus_statchg),
166
167	{ NULL, NULL }
168};
169
170static driver_t ale_driver = {
171	"ale",
172	ale_methods,
173	sizeof(struct ale_softc)
174};
175
176static devclass_t ale_devclass;
177
178DRIVER_MODULE(ale, pci, ale_driver, ale_devclass, 0, 0);
179DRIVER_MODULE(miibus, ale, miibus_driver, miibus_devclass, 0, 0);
180
181static struct resource_spec ale_res_spec_mem[] = {
182	{ SYS_RES_MEMORY,	PCIR_BAR(0),	RF_ACTIVE },
183	{ -1,			0,		0 }
184};
185
186static struct resource_spec ale_irq_spec_legacy[] = {
187	{ SYS_RES_IRQ,		0,		RF_ACTIVE | RF_SHAREABLE },
188	{ -1,			0,		0 }
189};
190
191static struct resource_spec ale_irq_spec_msi[] = {
192	{ SYS_RES_IRQ,		1,		RF_ACTIVE },
193	{ -1,			0,		0 }
194};
195
196static struct resource_spec ale_irq_spec_msix[] = {
197	{ SYS_RES_IRQ,		1,		RF_ACTIVE },
198	{ -1,			0,		0 }
199};
200
201static int
202ale_miibus_readreg(device_t dev, int phy, int reg)
203{
204	struct ale_softc *sc;
205	uint32_t v;
206	int i;
207
208	sc = device_get_softc(dev);
209
210	CSR_WRITE_4(sc, ALE_MDIO, MDIO_OP_EXECUTE | MDIO_OP_READ |
211	    MDIO_SUP_PREAMBLE | MDIO_CLK_25_4 | MDIO_REG_ADDR(reg));
212	for (i = ALE_PHY_TIMEOUT; i > 0; i--) {
213		DELAY(5);
214		v = CSR_READ_4(sc, ALE_MDIO);
215		if ((v & (MDIO_OP_EXECUTE | MDIO_OP_BUSY)) == 0)
216			break;
217	}
218
219	if (i == 0) {
220		device_printf(sc->ale_dev, "phy read timeout : %d\n", reg);
221		return (0);
222	}
223
224	return ((v & MDIO_DATA_MASK) >> MDIO_DATA_SHIFT);
225}
226
227static int
228ale_miibus_writereg(device_t dev, int phy, int reg, int val)
229{
230	struct ale_softc *sc;
231	uint32_t v;
232	int i;
233
234	sc = device_get_softc(dev);
235
236	CSR_WRITE_4(sc, ALE_MDIO, MDIO_OP_EXECUTE | MDIO_OP_WRITE |
237	    (val & MDIO_DATA_MASK) << MDIO_DATA_SHIFT |
238	    MDIO_SUP_PREAMBLE | MDIO_CLK_25_4 | MDIO_REG_ADDR(reg));
239	for (i = ALE_PHY_TIMEOUT; i > 0; i--) {
240		DELAY(5);
241		v = CSR_READ_4(sc, ALE_MDIO);
242		if ((v & (MDIO_OP_EXECUTE | MDIO_OP_BUSY)) == 0)
243			break;
244	}
245
246	if (i == 0)
247		device_printf(sc->ale_dev, "phy write timeout : %d\n", reg);
248
249	return (0);
250}
251
252static void
253ale_miibus_statchg(device_t dev)
254{
255	struct ale_softc *sc;
256
257	sc = device_get_softc(dev);
258
259	taskqueue_enqueue(taskqueue_swi, &sc->ale_link_task);
260}
261
262static void
263ale_mediastatus(struct ifnet *ifp, struct ifmediareq *ifmr)
264{
265	struct ale_softc *sc;
266	struct mii_data *mii;
267
268	sc = ifp->if_softc;
269	ALE_LOCK(sc);
270	mii = device_get_softc(sc->ale_miibus);
271
272	mii_pollstat(mii);
273	ALE_UNLOCK(sc);
274	ifmr->ifm_status = mii->mii_media_status;
275	ifmr->ifm_active = mii->mii_media_active;
276}
277
278static int
279ale_mediachange(struct ifnet *ifp)
280{
281	struct ale_softc *sc;
282	struct mii_data *mii;
283	struct mii_softc *miisc;
284	int error;
285
286	sc = ifp->if_softc;
287	ALE_LOCK(sc);
288	mii = device_get_softc(sc->ale_miibus);
289	if (mii->mii_instance != 0) {
290		LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
291			mii_phy_reset(miisc);
292	}
293	error = mii_mediachg(mii);
294	ALE_UNLOCK(sc);
295
296	return (error);
297}
298
299static int
300ale_probe(device_t dev)
301{
302	struct ale_dev *sp;
303	int i;
304	uint16_t vendor, devid;
305
306	vendor = pci_get_vendor(dev);
307	devid = pci_get_device(dev);
308	sp = ale_devs;
309	for (i = 0; i < sizeof(ale_devs) / sizeof(ale_devs[0]); i++) {
310		if (vendor == sp->ale_vendorid &&
311		    devid == sp->ale_deviceid) {
312			device_set_desc(dev, sp->ale_name);
313			return (BUS_PROBE_DEFAULT);
314		}
315		sp++;
316	}
317
318	return (ENXIO);
319}
320
321static void
322ale_get_macaddr(struct ale_softc *sc)
323{
324	uint32_t ea[2], reg;
325	int i, vpdc;
326
327	reg = CSR_READ_4(sc, ALE_SPI_CTRL);
328	if ((reg & SPI_VPD_ENB) != 0) {
329		reg &= ~SPI_VPD_ENB;
330		CSR_WRITE_4(sc, ALE_SPI_CTRL, reg);
331	}
332
333	if (pci_find_extcap(sc->ale_dev, PCIY_VPD, &vpdc) == 0) {
334		/*
335		 * PCI VPD capability found, let TWSI reload EEPROM.
336		 * This will set ethernet address of controller.
337		 */
338		CSR_WRITE_4(sc, ALE_TWSI_CTRL, CSR_READ_4(sc, ALE_TWSI_CTRL) |
339		    TWSI_CTRL_SW_LD_START);
340		for (i = 100; i > 0; i--) {
341			DELAY(1000);
342			reg = CSR_READ_4(sc, ALE_TWSI_CTRL);
343			if ((reg & TWSI_CTRL_SW_LD_START) == 0)
344				break;
345		}
346		if (i == 0)
347			device_printf(sc->ale_dev,
348			    "reloading EEPROM timeout!\n");
349	} else {
350		if (bootverbose)
351			device_printf(sc->ale_dev,
352			    "PCI VPD capability not found!\n");
353	}
354
355	ea[0] = CSR_READ_4(sc, ALE_PAR0);
356	ea[1] = CSR_READ_4(sc, ALE_PAR1);
357	sc->ale_eaddr[0] = (ea[1] >> 8) & 0xFF;
358	sc->ale_eaddr[1] = (ea[1] >> 0) & 0xFF;
359	sc->ale_eaddr[2] = (ea[0] >> 24) & 0xFF;
360	sc->ale_eaddr[3] = (ea[0] >> 16) & 0xFF;
361	sc->ale_eaddr[4] = (ea[0] >> 8) & 0xFF;
362	sc->ale_eaddr[5] = (ea[0] >> 0) & 0xFF;
363}
364
365static void
366ale_phy_reset(struct ale_softc *sc)
367{
368
369	/* Reset magic from Linux. */
370	CSR_WRITE_2(sc, ALE_GPHY_CTRL,
371	    GPHY_CTRL_HIB_EN | GPHY_CTRL_HIB_PULSE | GPHY_CTRL_SEL_ANA_RESET |
372	    GPHY_CTRL_PHY_PLL_ON);
373	DELAY(1000);
374	CSR_WRITE_2(sc, ALE_GPHY_CTRL,
375	    GPHY_CTRL_EXT_RESET | GPHY_CTRL_HIB_EN | GPHY_CTRL_HIB_PULSE |
376	    GPHY_CTRL_SEL_ANA_RESET | GPHY_CTRL_PHY_PLL_ON);
377	DELAY(1000);
378
379#define	ATPHY_DBG_ADDR		0x1D
380#define	ATPHY_DBG_DATA		0x1E
381
382	/* Enable hibernation mode. */
383	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
384	    ATPHY_DBG_ADDR, 0x0B);
385	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
386	    ATPHY_DBG_DATA, 0xBC00);
387	/* Set Class A/B for all modes. */
388	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
389	    ATPHY_DBG_ADDR, 0x00);
390	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
391	    ATPHY_DBG_DATA, 0x02EF);
392	/* Enable 10BT power saving. */
393	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
394	    ATPHY_DBG_ADDR, 0x12);
395	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
396	    ATPHY_DBG_DATA, 0x4C04);
397	/* Adjust 1000T power. */
398	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
399	    ATPHY_DBG_ADDR, 0x04);
400	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
401	    ATPHY_DBG_ADDR, 0x8BBB);
402	/* 10BT center tap voltage. */
403	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
404	    ATPHY_DBG_ADDR, 0x05);
405	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
406	    ATPHY_DBG_ADDR, 0x2C46);
407
408#undef	ATPHY_DBG_ADDR
409#undef	ATPHY_DBG_DATA
410	DELAY(1000);
411}
412
413static int
414ale_attach(device_t dev)
415{
416	struct ale_softc *sc;
417	struct ifnet *ifp;
418	uint16_t burst;
419	int error, i, msic, msixc, pmc;
420	uint32_t rxf_len, txf_len;
421
422	error = 0;
423	sc = device_get_softc(dev);
424	sc->ale_dev = dev;
425
426	mtx_init(&sc->ale_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
427	    MTX_DEF);
428	callout_init_mtx(&sc->ale_tick_ch, &sc->ale_mtx, 0);
429	TASK_INIT(&sc->ale_int_task, 0, ale_int_task, sc);
430	TASK_INIT(&sc->ale_link_task, 0, ale_link_task, sc);
431
432	/* Map the device. */
433	pci_enable_busmaster(dev);
434	sc->ale_res_spec = ale_res_spec_mem;
435	sc->ale_irq_spec = ale_irq_spec_legacy;
436	error = bus_alloc_resources(dev, sc->ale_res_spec, sc->ale_res);
437	if (error != 0) {
438		device_printf(dev, "cannot allocate memory resources.\n");
439		goto fail;
440	}
441
442	/* Set PHY address. */
443	sc->ale_phyaddr = ALE_PHY_ADDR;
444
445	/* Reset PHY. */
446	ale_phy_reset(sc);
447
448	/* Reset the ethernet controller. */
449	ale_reset(sc);
450
451	/* Get PCI and chip id/revision. */
452	sc->ale_rev = pci_get_revid(dev);
453	if (sc->ale_rev >= 0xF0) {
454		/* L2E Rev. B. AR8114 */
455		sc->ale_flags |= ALE_FLAG_FASTETHER;
456	} else {
457		if ((CSR_READ_4(sc, ALE_PHY_STATUS) & PHY_STATUS_100M) != 0) {
458			/* L1E AR8121 */
459			sc->ale_flags |= ALE_FLAG_JUMBO;
460		} else {
461			/* L2E Rev. A. AR8113 */
462			sc->ale_flags |= ALE_FLAG_FASTETHER;
463		}
464	}
465	/*
466	 * All known controllers seems to require 4 bytes alignment
467	 * of Tx buffers to make Tx checksum offload with custom
468	 * checksum generation method work.
469	 */
470	sc->ale_flags |= ALE_FLAG_TXCSUM_BUG;
471	/*
472	 * All known controllers seems to have issues on Rx checksum
473	 * offload for fragmented IP datagrams.
474	 */
475	sc->ale_flags |= ALE_FLAG_RXCSUM_BUG;
476	/*
477	 * Don't use Tx CMB. It is known to cause RRS update failure
478	 * under certain circumstances. Typical phenomenon of the
479	 * issue would be unexpected sequence number encountered in
480	 * Rx handler.
481	 */
482	sc->ale_flags |= ALE_FLAG_TXCMB_BUG;
483	sc->ale_chip_rev = CSR_READ_4(sc, ALE_MASTER_CFG) >>
484	    MASTER_CHIP_REV_SHIFT;
485	if (bootverbose) {
486		device_printf(dev, "PCI device revision : 0x%04x\n",
487		    sc->ale_rev);
488		device_printf(dev, "Chip id/revision : 0x%04x\n",
489		    sc->ale_chip_rev);
490	}
491	txf_len = CSR_READ_4(sc, ALE_SRAM_TX_FIFO_LEN);
492	rxf_len = CSR_READ_4(sc, ALE_SRAM_RX_FIFO_LEN);
493	/*
494	 * Uninitialized hardware returns an invalid chip id/revision
495	 * as well as 0xFFFFFFFF for Tx/Rx fifo length.
496	 */
497	if (sc->ale_chip_rev == 0xFFFF || txf_len == 0xFFFFFFFF ||
498	    rxf_len == 0xFFFFFFF) {
499		device_printf(dev,"chip revision : 0x%04x, %u Tx FIFO "
500		    "%u Rx FIFO -- not initialized?\n", sc->ale_chip_rev,
501		    txf_len, rxf_len);
502		error = ENXIO;
503		goto fail;
504	}
505	device_printf(dev, "%u Tx FIFO, %u Rx FIFO\n", txf_len, rxf_len);
506
507	/* Allocate IRQ resources. */
508	msixc = pci_msix_count(dev);
509	msic = pci_msi_count(dev);
510	if (bootverbose) {
511		device_printf(dev, "MSIX count : %d\n", msixc);
512		device_printf(dev, "MSI count : %d\n", msic);
513	}
514
515	/* Prefer MSIX over MSI. */
516	if (msix_disable == 0 || msi_disable == 0) {
517		if (msix_disable == 0 && msixc == ALE_MSIX_MESSAGES &&
518		    pci_alloc_msix(dev, &msixc) == 0) {
519			if (msic == ALE_MSIX_MESSAGES) {
520				device_printf(dev, "Using %d MSIX messages.\n",
521				    msixc);
522				sc->ale_flags |= ALE_FLAG_MSIX;
523				sc->ale_irq_spec = ale_irq_spec_msix;
524			} else
525				pci_release_msi(dev);
526		}
527		if (msi_disable == 0 && (sc->ale_flags & ALE_FLAG_MSIX) == 0 &&
528		    msic == ALE_MSI_MESSAGES &&
529		    pci_alloc_msi(dev, &msic) == 0) {
530			if (msic == ALE_MSI_MESSAGES) {
531				device_printf(dev, "Using %d MSI messages.\n",
532				    msic);
533				sc->ale_flags |= ALE_FLAG_MSI;
534				sc->ale_irq_spec = ale_irq_spec_msi;
535			} else
536				pci_release_msi(dev);
537		}
538	}
539
540	error = bus_alloc_resources(dev, sc->ale_irq_spec, sc->ale_irq);
541	if (error != 0) {
542		device_printf(dev, "cannot allocate IRQ resources.\n");
543		goto fail;
544	}
545
546	/* Get DMA parameters from PCIe device control register. */
547	if (pci_find_extcap(dev, PCIY_EXPRESS, &i) == 0) {
548		sc->ale_flags |= ALE_FLAG_PCIE;
549		burst = pci_read_config(dev, i + 0x08, 2);
550		/* Max read request size. */
551		sc->ale_dma_rd_burst = ((burst >> 12) & 0x07) <<
552		    DMA_CFG_RD_BURST_SHIFT;
553		/* Max payload size. */
554		sc->ale_dma_wr_burst = ((burst >> 5) & 0x07) <<
555		    DMA_CFG_WR_BURST_SHIFT;
556		if (bootverbose) {
557			device_printf(dev, "Read request size : %d bytes.\n",
558			    128 << ((burst >> 12) & 0x07));
559			device_printf(dev, "TLP payload size : %d bytes.\n",
560			    128 << ((burst >> 5) & 0x07));
561		}
562	} else {
563		sc->ale_dma_rd_burst = DMA_CFG_RD_BURST_128;
564		sc->ale_dma_wr_burst = DMA_CFG_WR_BURST_128;
565	}
566
567	/* Create device sysctl node. */
568	ale_sysctl_node(sc);
569
570	if ((error = ale_dma_alloc(sc) != 0))
571		goto fail;
572
573	/* Load station address. */
574	ale_get_macaddr(sc);
575
576	ifp = sc->ale_ifp = if_alloc(IFT_ETHER);
577	if (ifp == NULL) {
578		device_printf(dev, "cannot allocate ifnet structure.\n");
579		error = ENXIO;
580		goto fail;
581	}
582
583	ifp->if_softc = sc;
584	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
585	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
586	ifp->if_ioctl = ale_ioctl;
587	ifp->if_start = ale_start;
588	ifp->if_init = ale_init;
589	ifp->if_snd.ifq_drv_maxlen = ALE_TX_RING_CNT - 1;
590	IFQ_SET_MAXLEN(&ifp->if_snd, ifp->if_snd.ifq_drv_maxlen);
591	IFQ_SET_READY(&ifp->if_snd);
592	ifp->if_capabilities = IFCAP_RXCSUM | IFCAP_TXCSUM | IFCAP_TSO4;
593	ifp->if_hwassist = ALE_CSUM_FEATURES | CSUM_TSO;
594	if (pci_find_extcap(dev, PCIY_PMG, &pmc) == 0) {
595		sc->ale_flags |= ALE_FLAG_PMCAP;
596		ifp->if_capabilities |= IFCAP_WOL_MAGIC | IFCAP_WOL_MCAST;
597	}
598	ifp->if_capenable = ifp->if_capabilities;
599
600	/* Set up MII bus. */
601	error = mii_attach(dev, &sc->ale_miibus, ifp, ale_mediachange,
602	    ale_mediastatus, BMSR_DEFCAPMASK, sc->ale_phyaddr, MII_OFFSET_ANY,
603	    0);
604	if (error != 0) {
605		device_printf(dev, "attaching PHYs failed\n");
606		goto fail;
607	}
608
609	ether_ifattach(ifp, sc->ale_eaddr);
610
611	/* VLAN capability setup. */
612	ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING |
613	    IFCAP_VLAN_HWCSUM | IFCAP_VLAN_HWTSO;
614	ifp->if_capenable = ifp->if_capabilities;
615	/*
616	 * Even though controllers supported by ale(3) have Rx checksum
617	 * offload bug the workaround for fragmented frames seemed to
618	 * work so far. However it seems Rx checksum offload does not
619	 * work under certain conditions. So disable Rx checksum offload
620	 * until I find more clue about it but allow users to override it.
621	 */
622	ifp->if_capenable &= ~IFCAP_RXCSUM;
623
624	/* Tell the upper layer(s) we support long frames. */
625	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
626
627	/* Create local taskq. */
628	sc->ale_tq = taskqueue_create_fast("ale_taskq", M_WAITOK,
629	    taskqueue_thread_enqueue, &sc->ale_tq);
630	if (sc->ale_tq == NULL) {
631		device_printf(dev, "could not create taskqueue.\n");
632		ether_ifdetach(ifp);
633		error = ENXIO;
634		goto fail;
635	}
636	taskqueue_start_threads(&sc->ale_tq, 1, PI_NET, "%s taskq",
637	    device_get_nameunit(sc->ale_dev));
638
639	if ((sc->ale_flags & ALE_FLAG_MSIX) != 0)
640		msic = ALE_MSIX_MESSAGES;
641	else if ((sc->ale_flags & ALE_FLAG_MSI) != 0)
642		msic = ALE_MSI_MESSAGES;
643	else
644		msic = 1;
645	for (i = 0; i < msic; i++) {
646		error = bus_setup_intr(dev, sc->ale_irq[i],
647		    INTR_TYPE_NET | INTR_MPSAFE, ale_intr, NULL, sc,
648		    &sc->ale_intrhand[i]);
649		if (error != 0)
650			break;
651	}
652	if (error != 0) {
653		device_printf(dev, "could not set up interrupt handler.\n");
654		taskqueue_free(sc->ale_tq);
655		sc->ale_tq = NULL;
656		ether_ifdetach(ifp);
657		goto fail;
658	}
659
660fail:
661	if (error != 0)
662		ale_detach(dev);
663
664	return (error);
665}
666
667static int
668ale_detach(device_t dev)
669{
670	struct ale_softc *sc;
671	struct ifnet *ifp;
672	int i, msic;
673
674	sc = device_get_softc(dev);
675
676	ifp = sc->ale_ifp;
677	if (device_is_attached(dev)) {
678		ether_ifdetach(ifp);
679		ALE_LOCK(sc);
680		ale_stop(sc);
681		ALE_UNLOCK(sc);
682		callout_drain(&sc->ale_tick_ch);
683		taskqueue_drain(sc->ale_tq, &sc->ale_int_task);
684		taskqueue_drain(taskqueue_swi, &sc->ale_link_task);
685	}
686
687	if (sc->ale_tq != NULL) {
688		taskqueue_drain(sc->ale_tq, &sc->ale_int_task);
689		taskqueue_free(sc->ale_tq);
690		sc->ale_tq = NULL;
691	}
692
693	if (sc->ale_miibus != NULL) {
694		device_delete_child(dev, sc->ale_miibus);
695		sc->ale_miibus = NULL;
696	}
697	bus_generic_detach(dev);
698	ale_dma_free(sc);
699
700	if (ifp != NULL) {
701		if_free(ifp);
702		sc->ale_ifp = NULL;
703	}
704
705	if ((sc->ale_flags & ALE_FLAG_MSIX) != 0)
706		msic = ALE_MSIX_MESSAGES;
707	else if ((sc->ale_flags & ALE_FLAG_MSI) != 0)
708		msic = ALE_MSI_MESSAGES;
709	else
710		msic = 1;
711	for (i = 0; i < msic; i++) {
712		if (sc->ale_intrhand[i] != NULL) {
713			bus_teardown_intr(dev, sc->ale_irq[i],
714			    sc->ale_intrhand[i]);
715			sc->ale_intrhand[i] = NULL;
716		}
717	}
718
719	bus_release_resources(dev, sc->ale_irq_spec, sc->ale_irq);
720	if ((sc->ale_flags & (ALE_FLAG_MSI | ALE_FLAG_MSIX)) != 0)
721		pci_release_msi(dev);
722	bus_release_resources(dev, sc->ale_res_spec, sc->ale_res);
723	mtx_destroy(&sc->ale_mtx);
724
725	return (0);
726}
727
728#define	ALE_SYSCTL_STAT_ADD32(c, h, n, p, d)	\
729	    SYSCTL_ADD_UINT(c, h, OID_AUTO, n, CTLFLAG_RD, p, 0, d)
730
731#if __FreeBSD_version >= 900030
732#define	ALE_SYSCTL_STAT_ADD64(c, h, n, p, d)	\
733	    SYSCTL_ADD_UQUAD(c, h, OID_AUTO, n, CTLFLAG_RD, p, d)
734#elif __FreeBSD_version > 800000
735#define	ALE_SYSCTL_STAT_ADD64(c, h, n, p, d)	\
736	    SYSCTL_ADD_QUAD(c, h, OID_AUTO, n, CTLFLAG_RD, p, d)
737#else
738#define	ALE_SYSCTL_STAT_ADD64(c, h, n, p, d)	\
739	    SYSCTL_ADD_ULONG(c, h, OID_AUTO, n, CTLFLAG_RD, p, d)
740#endif
741
742static void
743ale_sysctl_node(struct ale_softc *sc)
744{
745	struct sysctl_ctx_list *ctx;
746	struct sysctl_oid_list *child, *parent;
747	struct sysctl_oid *tree;
748	struct ale_hw_stats *stats;
749	int error;
750
751	stats = &sc->ale_stats;
752	ctx = device_get_sysctl_ctx(sc->ale_dev);
753	child = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->ale_dev));
754
755	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "int_rx_mod",
756	    CTLTYPE_INT | CTLFLAG_RW, &sc->ale_int_rx_mod, 0,
757	    sysctl_hw_ale_int_mod, "I", "ale Rx interrupt moderation");
758	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "int_tx_mod",
759	    CTLTYPE_INT | CTLFLAG_RW, &sc->ale_int_tx_mod, 0,
760	    sysctl_hw_ale_int_mod, "I", "ale Tx interrupt moderation");
761	/* Pull in device tunables. */
762	sc->ale_int_rx_mod = ALE_IM_RX_TIMER_DEFAULT;
763	error = resource_int_value(device_get_name(sc->ale_dev),
764	    device_get_unit(sc->ale_dev), "int_rx_mod", &sc->ale_int_rx_mod);
765	if (error == 0) {
766		if (sc->ale_int_rx_mod < ALE_IM_TIMER_MIN ||
767		    sc->ale_int_rx_mod > ALE_IM_TIMER_MAX) {
768			device_printf(sc->ale_dev, "int_rx_mod value out of "
769			    "range; using default: %d\n",
770			    ALE_IM_RX_TIMER_DEFAULT);
771			sc->ale_int_rx_mod = ALE_IM_RX_TIMER_DEFAULT;
772		}
773	}
774	sc->ale_int_tx_mod = ALE_IM_TX_TIMER_DEFAULT;
775	error = resource_int_value(device_get_name(sc->ale_dev),
776	    device_get_unit(sc->ale_dev), "int_tx_mod", &sc->ale_int_tx_mod);
777	if (error == 0) {
778		if (sc->ale_int_tx_mod < ALE_IM_TIMER_MIN ||
779		    sc->ale_int_tx_mod > ALE_IM_TIMER_MAX) {
780			device_printf(sc->ale_dev, "int_tx_mod value out of "
781			    "range; using default: %d\n",
782			    ALE_IM_TX_TIMER_DEFAULT);
783			sc->ale_int_tx_mod = ALE_IM_TX_TIMER_DEFAULT;
784		}
785	}
786	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "process_limit",
787	    CTLTYPE_INT | CTLFLAG_RW, &sc->ale_process_limit, 0,
788	    sysctl_hw_ale_proc_limit, "I",
789	    "max number of Rx events to process");
790	/* Pull in device tunables. */
791	sc->ale_process_limit = ALE_PROC_DEFAULT;
792	error = resource_int_value(device_get_name(sc->ale_dev),
793	    device_get_unit(sc->ale_dev), "process_limit",
794	    &sc->ale_process_limit);
795	if (error == 0) {
796		if (sc->ale_process_limit < ALE_PROC_MIN ||
797		    sc->ale_process_limit > ALE_PROC_MAX) {
798			device_printf(sc->ale_dev,
799			    "process_limit value out of range; "
800			    "using default: %d\n", ALE_PROC_DEFAULT);
801			sc->ale_process_limit = ALE_PROC_DEFAULT;
802		}
803	}
804
805	/* Misc statistics. */
806	ALE_SYSCTL_STAT_ADD32(ctx, child, "reset_brk_seq",
807	    &stats->reset_brk_seq,
808	    "Controller resets due to broken Rx sequnce number");
809
810	tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "stats", CTLFLAG_RD,
811	    NULL, "ATE statistics");
812	parent = SYSCTL_CHILDREN(tree);
813
814	/* Rx statistics. */
815	tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "rx", CTLFLAG_RD,
816	    NULL, "Rx MAC statistics");
817	child = SYSCTL_CHILDREN(tree);
818	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_frames",
819	    &stats->rx_frames, "Good frames");
820	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_bcast_frames",
821	    &stats->rx_bcast_frames, "Good broadcast frames");
822	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_mcast_frames",
823	    &stats->rx_mcast_frames, "Good multicast frames");
824	ALE_SYSCTL_STAT_ADD32(ctx, child, "pause_frames",
825	    &stats->rx_pause_frames, "Pause control frames");
826	ALE_SYSCTL_STAT_ADD32(ctx, child, "control_frames",
827	    &stats->rx_control_frames, "Control frames");
828	ALE_SYSCTL_STAT_ADD32(ctx, child, "crc_errs",
829	    &stats->rx_crcerrs, "CRC errors");
830	ALE_SYSCTL_STAT_ADD32(ctx, child, "len_errs",
831	    &stats->rx_lenerrs, "Frames with length mismatched");
832	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_octets",
833	    &stats->rx_bytes, "Good octets");
834	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_bcast_octets",
835	    &stats->rx_bcast_bytes, "Good broadcast octets");
836	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_mcast_octets",
837	    &stats->rx_mcast_bytes, "Good multicast octets");
838	ALE_SYSCTL_STAT_ADD32(ctx, child, "runts",
839	    &stats->rx_runts, "Too short frames");
840	ALE_SYSCTL_STAT_ADD32(ctx, child, "fragments",
841	    &stats->rx_fragments, "Fragmented frames");
842	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_64",
843	    &stats->rx_pkts_64, "64 bytes frames");
844	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_65_127",
845	    &stats->rx_pkts_65_127, "65 to 127 bytes frames");
846	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_128_255",
847	    &stats->rx_pkts_128_255, "128 to 255 bytes frames");
848	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_256_511",
849	    &stats->rx_pkts_256_511, "256 to 511 bytes frames");
850	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_512_1023",
851	    &stats->rx_pkts_512_1023, "512 to 1023 bytes frames");
852	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1024_1518",
853	    &stats->rx_pkts_1024_1518, "1024 to 1518 bytes frames");
854	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1519_max",
855	    &stats->rx_pkts_1519_max, "1519 to max frames");
856	ALE_SYSCTL_STAT_ADD32(ctx, child, "trunc_errs",
857	    &stats->rx_pkts_truncated, "Truncated frames due to MTU size");
858	ALE_SYSCTL_STAT_ADD32(ctx, child, "fifo_oflows",
859	    &stats->rx_fifo_oflows, "FIFO overflows");
860	ALE_SYSCTL_STAT_ADD32(ctx, child, "rrs_errs",
861	    &stats->rx_rrs_errs, "Return status write-back errors");
862	ALE_SYSCTL_STAT_ADD32(ctx, child, "align_errs",
863	    &stats->rx_alignerrs, "Alignment errors");
864	ALE_SYSCTL_STAT_ADD32(ctx, child, "filtered",
865	    &stats->rx_pkts_filtered,
866	    "Frames dropped due to address filtering");
867
868	/* Tx statistics. */
869	tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "tx", CTLFLAG_RD,
870	    NULL, "Tx MAC statistics");
871	child = SYSCTL_CHILDREN(tree);
872	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_frames",
873	    &stats->tx_frames, "Good frames");
874	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_bcast_frames",
875	    &stats->tx_bcast_frames, "Good broadcast frames");
876	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_mcast_frames",
877	    &stats->tx_mcast_frames, "Good multicast frames");
878	ALE_SYSCTL_STAT_ADD32(ctx, child, "pause_frames",
879	    &stats->tx_pause_frames, "Pause control frames");
880	ALE_SYSCTL_STAT_ADD32(ctx, child, "control_frames",
881	    &stats->tx_control_frames, "Control frames");
882	ALE_SYSCTL_STAT_ADD32(ctx, child, "excess_defers",
883	    &stats->tx_excess_defer, "Frames with excessive derferrals");
884	ALE_SYSCTL_STAT_ADD32(ctx, child, "defers",
885	    &stats->tx_excess_defer, "Frames with derferrals");
886	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_octets",
887	    &stats->tx_bytes, "Good octets");
888	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_bcast_octets",
889	    &stats->tx_bcast_bytes, "Good broadcast octets");
890	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_mcast_octets",
891	    &stats->tx_mcast_bytes, "Good multicast octets");
892	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_64",
893	    &stats->tx_pkts_64, "64 bytes frames");
894	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_65_127",
895	    &stats->tx_pkts_65_127, "65 to 127 bytes frames");
896	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_128_255",
897	    &stats->tx_pkts_128_255, "128 to 255 bytes frames");
898	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_256_511",
899	    &stats->tx_pkts_256_511, "256 to 511 bytes frames");
900	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_512_1023",
901	    &stats->tx_pkts_512_1023, "512 to 1023 bytes frames");
902	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1024_1518",
903	    &stats->tx_pkts_1024_1518, "1024 to 1518 bytes frames");
904	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1519_max",
905	    &stats->tx_pkts_1519_max, "1519 to max frames");
906	ALE_SYSCTL_STAT_ADD32(ctx, child, "single_colls",
907	    &stats->tx_single_colls, "Single collisions");
908	ALE_SYSCTL_STAT_ADD32(ctx, child, "multi_colls",
909	    &stats->tx_multi_colls, "Multiple collisions");
910	ALE_SYSCTL_STAT_ADD32(ctx, child, "late_colls",
911	    &stats->tx_late_colls, "Late collisions");
912	ALE_SYSCTL_STAT_ADD32(ctx, child, "excess_colls",
913	    &stats->tx_excess_colls, "Excessive collisions");
914	ALE_SYSCTL_STAT_ADD32(ctx, child, "abort",
915	    &stats->tx_abort, "Aborted frames due to Excessive collisions");
916	ALE_SYSCTL_STAT_ADD32(ctx, child, "underruns",
917	    &stats->tx_underrun, "FIFO underruns");
918	ALE_SYSCTL_STAT_ADD32(ctx, child, "desc_underruns",
919	    &stats->tx_desc_underrun, "Descriptor write-back errors");
920	ALE_SYSCTL_STAT_ADD32(ctx, child, "len_errs",
921	    &stats->tx_lenerrs, "Frames with length mismatched");
922	ALE_SYSCTL_STAT_ADD32(ctx, child, "trunc_errs",
923	    &stats->tx_pkts_truncated, "Truncated frames due to MTU size");
924}
925
926#undef ALE_SYSCTL_STAT_ADD32
927#undef ALE_SYSCTL_STAT_ADD64
928
929struct ale_dmamap_arg {
930	bus_addr_t	ale_busaddr;
931};
932
933static void
934ale_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error)
935{
936	struct ale_dmamap_arg *ctx;
937
938	if (error != 0)
939		return;
940
941	KASSERT(nsegs == 1, ("%s: %d segments returned!", __func__, nsegs));
942
943	ctx = (struct ale_dmamap_arg *)arg;
944	ctx->ale_busaddr = segs[0].ds_addr;
945}
946
947/*
948 * Tx descriptors/RXF0/CMB DMA blocks share ALE_DESC_ADDR_HI register
949 * which specifies high address region of DMA blocks. Therefore these
950 * blocks should have the same high address of given 4GB address
951 * space(i.e. crossing 4GB boundary is not allowed).
952 */
953static int
954ale_check_boundary(struct ale_softc *sc)
955{
956	bus_addr_t rx_cmb_end[ALE_RX_PAGES], tx_cmb_end;
957	bus_addr_t rx_page_end[ALE_RX_PAGES], tx_ring_end;
958
959	rx_page_end[0] = sc->ale_cdata.ale_rx_page[0].page_paddr +
960	    sc->ale_pagesize;
961	rx_page_end[1] = sc->ale_cdata.ale_rx_page[1].page_paddr +
962	    sc->ale_pagesize;
963	tx_ring_end = sc->ale_cdata.ale_tx_ring_paddr + ALE_TX_RING_SZ;
964	tx_cmb_end = sc->ale_cdata.ale_tx_cmb_paddr + ALE_TX_CMB_SZ;
965	rx_cmb_end[0] = sc->ale_cdata.ale_rx_page[0].cmb_paddr + ALE_RX_CMB_SZ;
966	rx_cmb_end[1] = sc->ale_cdata.ale_rx_page[1].cmb_paddr + ALE_RX_CMB_SZ;
967
968	if ((ALE_ADDR_HI(tx_ring_end) !=
969	    ALE_ADDR_HI(sc->ale_cdata.ale_tx_ring_paddr)) ||
970	    (ALE_ADDR_HI(rx_page_end[0]) !=
971	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[0].page_paddr)) ||
972	    (ALE_ADDR_HI(rx_page_end[1]) !=
973	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[1].page_paddr)) ||
974	    (ALE_ADDR_HI(tx_cmb_end) !=
975	    ALE_ADDR_HI(sc->ale_cdata.ale_tx_cmb_paddr)) ||
976	    (ALE_ADDR_HI(rx_cmb_end[0]) !=
977	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[0].cmb_paddr)) ||
978	    (ALE_ADDR_HI(rx_cmb_end[1]) !=
979	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[1].cmb_paddr)))
980		return (EFBIG);
981
982	if ((ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_page_end[0])) ||
983	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_page_end[1])) ||
984	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_cmb_end[0])) ||
985	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_cmb_end[1])) ||
986	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(tx_cmb_end)))
987		return (EFBIG);
988
989	return (0);
990}
991
992static int
993ale_dma_alloc(struct ale_softc *sc)
994{
995	struct ale_txdesc *txd;
996	bus_addr_t lowaddr;
997	struct ale_dmamap_arg ctx;
998	int error, guard_size, i;
999
1000	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0)
1001		guard_size = ALE_JUMBO_FRAMELEN;
1002	else
1003		guard_size = ALE_MAX_FRAMELEN;
1004	sc->ale_pagesize = roundup(guard_size + ALE_RX_PAGE_SZ,
1005	    ALE_RX_PAGE_ALIGN);
1006	lowaddr = BUS_SPACE_MAXADDR;
1007again:
1008	/* Create parent DMA tag. */
1009	error = bus_dma_tag_create(
1010	    bus_get_dma_tag(sc->ale_dev), /* parent */
1011	    1, 0,			/* alignment, boundary */
1012	    lowaddr,			/* lowaddr */
1013	    BUS_SPACE_MAXADDR,		/* highaddr */
1014	    NULL, NULL,			/* filter, filterarg */
1015	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsize */
1016	    0,				/* nsegments */
1017	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsegsize */
1018	    0,				/* flags */
1019	    NULL, NULL,			/* lockfunc, lockarg */
1020	    &sc->ale_cdata.ale_parent_tag);
1021	if (error != 0) {
1022		device_printf(sc->ale_dev,
1023		    "could not create parent DMA tag.\n");
1024		goto fail;
1025	}
1026
1027	/* Create DMA tag for Tx descriptor ring. */
1028	error = bus_dma_tag_create(
1029	    sc->ale_cdata.ale_parent_tag, /* parent */
1030	    ALE_TX_RING_ALIGN, 0,	/* alignment, boundary */
1031	    BUS_SPACE_MAXADDR,		/* lowaddr */
1032	    BUS_SPACE_MAXADDR,		/* highaddr */
1033	    NULL, NULL,			/* filter, filterarg */
1034	    ALE_TX_RING_SZ,		/* maxsize */
1035	    1,				/* nsegments */
1036	    ALE_TX_RING_SZ,		/* maxsegsize */
1037	    0,				/* flags */
1038	    NULL, NULL,			/* lockfunc, lockarg */
1039	    &sc->ale_cdata.ale_tx_ring_tag);
1040	if (error != 0) {
1041		device_printf(sc->ale_dev,
1042		    "could not create Tx ring DMA tag.\n");
1043		goto fail;
1044	}
1045
1046	/* Create DMA tag for Rx pages. */
1047	for (i = 0; i < ALE_RX_PAGES; i++) {
1048		error = bus_dma_tag_create(
1049		    sc->ale_cdata.ale_parent_tag, /* parent */
1050		    ALE_RX_PAGE_ALIGN, 0,	/* alignment, boundary */
1051		    BUS_SPACE_MAXADDR,		/* lowaddr */
1052		    BUS_SPACE_MAXADDR,		/* highaddr */
1053		    NULL, NULL,			/* filter, filterarg */
1054		    sc->ale_pagesize,		/* maxsize */
1055		    1,				/* nsegments */
1056		    sc->ale_pagesize,		/* maxsegsize */
1057		    0,				/* flags */
1058		    NULL, NULL,			/* lockfunc, lockarg */
1059		    &sc->ale_cdata.ale_rx_page[i].page_tag);
1060		if (error != 0) {
1061			device_printf(sc->ale_dev,
1062			    "could not create Rx page %d DMA tag.\n", i);
1063			goto fail;
1064		}
1065	}
1066
1067	/* Create DMA tag for Tx coalescing message block. */
1068	error = bus_dma_tag_create(
1069	    sc->ale_cdata.ale_parent_tag, /* parent */
1070	    ALE_CMB_ALIGN, 0,		/* alignment, boundary */
1071	    BUS_SPACE_MAXADDR,		/* lowaddr */
1072	    BUS_SPACE_MAXADDR,		/* highaddr */
1073	    NULL, NULL,			/* filter, filterarg */
1074	    ALE_TX_CMB_SZ,		/* maxsize */
1075	    1,				/* nsegments */
1076	    ALE_TX_CMB_SZ,		/* maxsegsize */
1077	    0,				/* flags */
1078	    NULL, NULL,			/* lockfunc, lockarg */
1079	    &sc->ale_cdata.ale_tx_cmb_tag);
1080	if (error != 0) {
1081		device_printf(sc->ale_dev,
1082		    "could not create Tx CMB DMA tag.\n");
1083		goto fail;
1084	}
1085
1086	/* Create DMA tag for Rx coalescing message block. */
1087	for (i = 0; i < ALE_RX_PAGES; i++) {
1088		error = bus_dma_tag_create(
1089		    sc->ale_cdata.ale_parent_tag, /* parent */
1090		    ALE_CMB_ALIGN, 0,		/* alignment, boundary */
1091		    BUS_SPACE_MAXADDR,		/* lowaddr */
1092		    BUS_SPACE_MAXADDR,		/* highaddr */
1093		    NULL, NULL,			/* filter, filterarg */
1094		    ALE_RX_CMB_SZ,		/* maxsize */
1095		    1,				/* nsegments */
1096		    ALE_RX_CMB_SZ,		/* maxsegsize */
1097		    0,				/* flags */
1098		    NULL, NULL,			/* lockfunc, lockarg */
1099		    &sc->ale_cdata.ale_rx_page[i].cmb_tag);
1100		if (error != 0) {
1101			device_printf(sc->ale_dev,
1102			    "could not create Rx page %d CMB DMA tag.\n", i);
1103			goto fail;
1104		}
1105	}
1106
1107	/* Allocate DMA'able memory and load the DMA map for Tx ring. */
1108	error = bus_dmamem_alloc(sc->ale_cdata.ale_tx_ring_tag,
1109	    (void **)&sc->ale_cdata.ale_tx_ring,
1110	    BUS_DMA_WAITOK | BUS_DMA_ZERO | BUS_DMA_COHERENT,
1111	    &sc->ale_cdata.ale_tx_ring_map);
1112	if (error != 0) {
1113		device_printf(sc->ale_dev,
1114		    "could not allocate DMA'able memory for Tx ring.\n");
1115		goto fail;
1116	}
1117	ctx.ale_busaddr = 0;
1118	error = bus_dmamap_load(sc->ale_cdata.ale_tx_ring_tag,
1119	    sc->ale_cdata.ale_tx_ring_map, sc->ale_cdata.ale_tx_ring,
1120	    ALE_TX_RING_SZ, ale_dmamap_cb, &ctx, 0);
1121	if (error != 0 || ctx.ale_busaddr == 0) {
1122		device_printf(sc->ale_dev,
1123		    "could not load DMA'able memory for Tx ring.\n");
1124		goto fail;
1125	}
1126	sc->ale_cdata.ale_tx_ring_paddr = ctx.ale_busaddr;
1127
1128	/* Rx pages. */
1129	for (i = 0; i < ALE_RX_PAGES; i++) {
1130		error = bus_dmamem_alloc(sc->ale_cdata.ale_rx_page[i].page_tag,
1131		    (void **)&sc->ale_cdata.ale_rx_page[i].page_addr,
1132		    BUS_DMA_WAITOK | BUS_DMA_ZERO | BUS_DMA_COHERENT,
1133		    &sc->ale_cdata.ale_rx_page[i].page_map);
1134		if (error != 0) {
1135			device_printf(sc->ale_dev,
1136			    "could not allocate DMA'able memory for "
1137			    "Rx page %d.\n", i);
1138			goto fail;
1139		}
1140		ctx.ale_busaddr = 0;
1141		error = bus_dmamap_load(sc->ale_cdata.ale_rx_page[i].page_tag,
1142		    sc->ale_cdata.ale_rx_page[i].page_map,
1143		    sc->ale_cdata.ale_rx_page[i].page_addr,
1144		    sc->ale_pagesize, ale_dmamap_cb, &ctx, 0);
1145		if (error != 0 || ctx.ale_busaddr == 0) {
1146			device_printf(sc->ale_dev,
1147			    "could not load DMA'able memory for "
1148			    "Rx page %d.\n", i);
1149			goto fail;
1150		}
1151		sc->ale_cdata.ale_rx_page[i].page_paddr = ctx.ale_busaddr;
1152	}
1153
1154	/* Tx CMB. */
1155	error = bus_dmamem_alloc(sc->ale_cdata.ale_tx_cmb_tag,
1156	    (void **)&sc->ale_cdata.ale_tx_cmb,
1157	    BUS_DMA_WAITOK | BUS_DMA_ZERO | BUS_DMA_COHERENT,
1158	    &sc->ale_cdata.ale_tx_cmb_map);
1159	if (error != 0) {
1160		device_printf(sc->ale_dev,
1161		    "could not allocate DMA'able memory for Tx CMB.\n");
1162		goto fail;
1163	}
1164	ctx.ale_busaddr = 0;
1165	error = bus_dmamap_load(sc->ale_cdata.ale_tx_cmb_tag,
1166	    sc->ale_cdata.ale_tx_cmb_map, sc->ale_cdata.ale_tx_cmb,
1167	    ALE_TX_CMB_SZ, ale_dmamap_cb, &ctx, 0);
1168	if (error != 0 || ctx.ale_busaddr == 0) {
1169		device_printf(sc->ale_dev,
1170		    "could not load DMA'able memory for Tx CMB.\n");
1171		goto fail;
1172	}
1173	sc->ale_cdata.ale_tx_cmb_paddr = ctx.ale_busaddr;
1174
1175	/* Rx CMB. */
1176	for (i = 0; i < ALE_RX_PAGES; i++) {
1177		error = bus_dmamem_alloc(sc->ale_cdata.ale_rx_page[i].cmb_tag,
1178		    (void **)&sc->ale_cdata.ale_rx_page[i].cmb_addr,
1179		    BUS_DMA_WAITOK | BUS_DMA_ZERO | BUS_DMA_COHERENT,
1180		    &sc->ale_cdata.ale_rx_page[i].cmb_map);
1181		if (error != 0) {
1182			device_printf(sc->ale_dev, "could not allocate "
1183			    "DMA'able memory for Rx page %d CMB.\n", i);
1184			goto fail;
1185		}
1186		ctx.ale_busaddr = 0;
1187		error = bus_dmamap_load(sc->ale_cdata.ale_rx_page[i].cmb_tag,
1188		    sc->ale_cdata.ale_rx_page[i].cmb_map,
1189		    sc->ale_cdata.ale_rx_page[i].cmb_addr,
1190		    ALE_RX_CMB_SZ, ale_dmamap_cb, &ctx, 0);
1191		if (error != 0 || ctx.ale_busaddr == 0) {
1192			device_printf(sc->ale_dev, "could not load DMA'able "
1193			    "memory for Rx page %d CMB.\n", i);
1194			goto fail;
1195		}
1196		sc->ale_cdata.ale_rx_page[i].cmb_paddr = ctx.ale_busaddr;
1197	}
1198
1199	/*
1200	 * Tx descriptors/RXF0/CMB DMA blocks share the same
1201	 * high address region of 64bit DMA address space.
1202	 */
1203	if (lowaddr != BUS_SPACE_MAXADDR_32BIT &&
1204	    (error = ale_check_boundary(sc)) != 0) {
1205		device_printf(sc->ale_dev, "4GB boundary crossed, "
1206		    "switching to 32bit DMA addressing mode.\n");
1207		ale_dma_free(sc);
1208		/*
1209		 * Limit max allowable DMA address space to 32bit
1210		 * and try again.
1211		 */
1212		lowaddr = BUS_SPACE_MAXADDR_32BIT;
1213		goto again;
1214	}
1215
1216	/*
1217	 * Create Tx buffer parent tag.
1218	 * AR81xx allows 64bit DMA addressing of Tx buffers so it
1219	 * needs separate parent DMA tag as parent DMA address space
1220	 * could be restricted to be within 32bit address space by
1221	 * 4GB boundary crossing.
1222	 */
1223	error = bus_dma_tag_create(
1224	    bus_get_dma_tag(sc->ale_dev), /* parent */
1225	    1, 0,			/* alignment, boundary */
1226	    BUS_SPACE_MAXADDR,		/* lowaddr */
1227	    BUS_SPACE_MAXADDR,		/* highaddr */
1228	    NULL, NULL,			/* filter, filterarg */
1229	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsize */
1230	    0,				/* nsegments */
1231	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsegsize */
1232	    0,				/* flags */
1233	    NULL, NULL,			/* lockfunc, lockarg */
1234	    &sc->ale_cdata.ale_buffer_tag);
1235	if (error != 0) {
1236		device_printf(sc->ale_dev,
1237		    "could not create parent buffer DMA tag.\n");
1238		goto fail;
1239	}
1240
1241	/* Create DMA tag for Tx buffers. */
1242	error = bus_dma_tag_create(
1243	    sc->ale_cdata.ale_buffer_tag, /* parent */
1244	    1, 0,			/* alignment, boundary */
1245	    BUS_SPACE_MAXADDR,		/* lowaddr */
1246	    BUS_SPACE_MAXADDR,		/* highaddr */
1247	    NULL, NULL,			/* filter, filterarg */
1248	    ALE_TSO_MAXSIZE,		/* maxsize */
1249	    ALE_MAXTXSEGS,		/* nsegments */
1250	    ALE_TSO_MAXSEGSIZE,		/* maxsegsize */
1251	    0,				/* flags */
1252	    NULL, NULL,			/* lockfunc, lockarg */
1253	    &sc->ale_cdata.ale_tx_tag);
1254	if (error != 0) {
1255		device_printf(sc->ale_dev, "could not create Tx DMA tag.\n");
1256		goto fail;
1257	}
1258
1259	/* Create DMA maps for Tx buffers. */
1260	for (i = 0; i < ALE_TX_RING_CNT; i++) {
1261		txd = &sc->ale_cdata.ale_txdesc[i];
1262		txd->tx_m = NULL;
1263		txd->tx_dmamap = NULL;
1264		error = bus_dmamap_create(sc->ale_cdata.ale_tx_tag, 0,
1265		    &txd->tx_dmamap);
1266		if (error != 0) {
1267			device_printf(sc->ale_dev,
1268			    "could not create Tx dmamap.\n");
1269			goto fail;
1270		}
1271	}
1272
1273fail:
1274	return (error);
1275}
1276
1277static void
1278ale_dma_free(struct ale_softc *sc)
1279{
1280	struct ale_txdesc *txd;
1281	int i;
1282
1283	/* Tx buffers. */
1284	if (sc->ale_cdata.ale_tx_tag != NULL) {
1285		for (i = 0; i < ALE_TX_RING_CNT; i++) {
1286			txd = &sc->ale_cdata.ale_txdesc[i];
1287			if (txd->tx_dmamap != NULL) {
1288				bus_dmamap_destroy(sc->ale_cdata.ale_tx_tag,
1289				    txd->tx_dmamap);
1290				txd->tx_dmamap = NULL;
1291			}
1292		}
1293		bus_dma_tag_destroy(sc->ale_cdata.ale_tx_tag);
1294		sc->ale_cdata.ale_tx_tag = NULL;
1295	}
1296	/* Tx descriptor ring. */
1297	if (sc->ale_cdata.ale_tx_ring_tag != NULL) {
1298		if (sc->ale_cdata.ale_tx_ring_map != NULL)
1299			bus_dmamap_unload(sc->ale_cdata.ale_tx_ring_tag,
1300			    sc->ale_cdata.ale_tx_ring_map);
1301		if (sc->ale_cdata.ale_tx_ring_map != NULL &&
1302		    sc->ale_cdata.ale_tx_ring != NULL)
1303			bus_dmamem_free(sc->ale_cdata.ale_tx_ring_tag,
1304			    sc->ale_cdata.ale_tx_ring,
1305			    sc->ale_cdata.ale_tx_ring_map);
1306		sc->ale_cdata.ale_tx_ring = NULL;
1307		sc->ale_cdata.ale_tx_ring_map = NULL;
1308		bus_dma_tag_destroy(sc->ale_cdata.ale_tx_ring_tag);
1309		sc->ale_cdata.ale_tx_ring_tag = NULL;
1310	}
1311	/* Rx page block. */
1312	for (i = 0; i < ALE_RX_PAGES; i++) {
1313		if (sc->ale_cdata.ale_rx_page[i].page_tag != NULL) {
1314			if (sc->ale_cdata.ale_rx_page[i].page_map != NULL)
1315				bus_dmamap_unload(
1316				    sc->ale_cdata.ale_rx_page[i].page_tag,
1317				    sc->ale_cdata.ale_rx_page[i].page_map);
1318			if (sc->ale_cdata.ale_rx_page[i].page_map != NULL &&
1319			    sc->ale_cdata.ale_rx_page[i].page_addr != NULL)
1320				bus_dmamem_free(
1321				    sc->ale_cdata.ale_rx_page[i].page_tag,
1322				    sc->ale_cdata.ale_rx_page[i].page_addr,
1323				    sc->ale_cdata.ale_rx_page[i].page_map);
1324			sc->ale_cdata.ale_rx_page[i].page_addr = NULL;
1325			sc->ale_cdata.ale_rx_page[i].page_map = NULL;
1326			bus_dma_tag_destroy(
1327			    sc->ale_cdata.ale_rx_page[i].page_tag);
1328			sc->ale_cdata.ale_rx_page[i].page_tag = NULL;
1329		}
1330	}
1331	/* Rx CMB. */
1332	for (i = 0; i < ALE_RX_PAGES; i++) {
1333		if (sc->ale_cdata.ale_rx_page[i].cmb_tag != NULL) {
1334			if (sc->ale_cdata.ale_rx_page[i].cmb_map != NULL)
1335				bus_dmamap_unload(
1336				    sc->ale_cdata.ale_rx_page[i].cmb_tag,
1337				    sc->ale_cdata.ale_rx_page[i].cmb_map);
1338			if (sc->ale_cdata.ale_rx_page[i].cmb_map != NULL &&
1339			    sc->ale_cdata.ale_rx_page[i].cmb_addr != NULL)
1340				bus_dmamem_free(
1341				    sc->ale_cdata.ale_rx_page[i].cmb_tag,
1342				    sc->ale_cdata.ale_rx_page[i].cmb_addr,
1343				    sc->ale_cdata.ale_rx_page[i].cmb_map);
1344			sc->ale_cdata.ale_rx_page[i].cmb_addr = NULL;
1345			sc->ale_cdata.ale_rx_page[i].cmb_map = NULL;
1346			bus_dma_tag_destroy(
1347			    sc->ale_cdata.ale_rx_page[i].cmb_tag);
1348			sc->ale_cdata.ale_rx_page[i].cmb_tag = NULL;
1349		}
1350	}
1351	/* Tx CMB. */
1352	if (sc->ale_cdata.ale_tx_cmb_tag != NULL) {
1353		if (sc->ale_cdata.ale_tx_cmb_map != NULL)
1354			bus_dmamap_unload(sc->ale_cdata.ale_tx_cmb_tag,
1355			    sc->ale_cdata.ale_tx_cmb_map);
1356		if (sc->ale_cdata.ale_tx_cmb_map != NULL &&
1357		    sc->ale_cdata.ale_tx_cmb != NULL)
1358			bus_dmamem_free(sc->ale_cdata.ale_tx_cmb_tag,
1359			    sc->ale_cdata.ale_tx_cmb,
1360			    sc->ale_cdata.ale_tx_cmb_map);
1361		sc->ale_cdata.ale_tx_cmb = NULL;
1362		sc->ale_cdata.ale_tx_cmb_map = NULL;
1363		bus_dma_tag_destroy(sc->ale_cdata.ale_tx_cmb_tag);
1364		sc->ale_cdata.ale_tx_cmb_tag = NULL;
1365	}
1366	if (sc->ale_cdata.ale_buffer_tag != NULL) {
1367		bus_dma_tag_destroy(sc->ale_cdata.ale_buffer_tag);
1368		sc->ale_cdata.ale_buffer_tag = NULL;
1369	}
1370	if (sc->ale_cdata.ale_parent_tag != NULL) {
1371		bus_dma_tag_destroy(sc->ale_cdata.ale_parent_tag);
1372		sc->ale_cdata.ale_parent_tag = NULL;
1373	}
1374}
1375
1376static int
1377ale_shutdown(device_t dev)
1378{
1379
1380	return (ale_suspend(dev));
1381}
1382
1383/*
1384 * Note, this driver resets the link speed to 10/100Mbps by
1385 * restarting auto-negotiation in suspend/shutdown phase but we
1386 * don't know whether that auto-negotiation would succeed or not
1387 * as driver has no control after powering off/suspend operation.
1388 * If the renegotiation fail WOL may not work. Running at 1Gbps
1389 * will draw more power than 375mA at 3.3V which is specified in
1390 * PCI specification and that would result in complete
1391 * shutdowning power to ethernet controller.
1392 *
1393 * TODO
1394 * Save current negotiated media speed/duplex/flow-control to
1395 * softc and restore the same link again after resuming. PHY
1396 * handling such as power down/resetting to 100Mbps may be better
1397 * handled in suspend method in phy driver.
1398 */
1399static void
1400ale_setlinkspeed(struct ale_softc *sc)
1401{
1402	struct mii_data *mii;
1403	int aneg, i;
1404
1405	mii = device_get_softc(sc->ale_miibus);
1406	mii_pollstat(mii);
1407	aneg = 0;
1408	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
1409	    (IFM_ACTIVE | IFM_AVALID)) {
1410		switch IFM_SUBTYPE(mii->mii_media_active) {
1411		case IFM_10_T:
1412		case IFM_100_TX:
1413			return;
1414		case IFM_1000_T:
1415			aneg++;
1416			break;
1417		default:
1418			break;
1419		}
1420	}
1421	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr, MII_100T2CR, 0);
1422	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
1423	    MII_ANAR, ANAR_TX_FD | ANAR_TX | ANAR_10_FD | ANAR_10 | ANAR_CSMA);
1424	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
1425	    MII_BMCR, BMCR_RESET | BMCR_AUTOEN | BMCR_STARTNEG);
1426	DELAY(1000);
1427	if (aneg != 0) {
1428		/*
1429		 * Poll link state until ale(4) get a 10/100Mbps link.
1430		 */
1431		for (i = 0; i < MII_ANEGTICKS_GIGE; i++) {
1432			mii_pollstat(mii);
1433			if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID))
1434			    == (IFM_ACTIVE | IFM_AVALID)) {
1435				switch (IFM_SUBTYPE(
1436				    mii->mii_media_active)) {
1437				case IFM_10_T:
1438				case IFM_100_TX:
1439					ale_mac_config(sc);
1440					return;
1441				default:
1442					break;
1443				}
1444			}
1445			ALE_UNLOCK(sc);
1446			pause("alelnk", hz);
1447			ALE_LOCK(sc);
1448		}
1449		if (i == MII_ANEGTICKS_GIGE)
1450			device_printf(sc->ale_dev,
1451			    "establishing a link failed, WOL may not work!");
1452	}
1453	/*
1454	 * No link, force MAC to have 100Mbps, full-duplex link.
1455	 * This is the last resort and may/may not work.
1456	 */
1457	mii->mii_media_status = IFM_AVALID | IFM_ACTIVE;
1458	mii->mii_media_active = IFM_ETHER | IFM_100_TX | IFM_FDX;
1459	ale_mac_config(sc);
1460}
1461
1462static void
1463ale_setwol(struct ale_softc *sc)
1464{
1465	struct ifnet *ifp;
1466	uint32_t reg, pmcs;
1467	uint16_t pmstat;
1468	int pmc;
1469
1470	ALE_LOCK_ASSERT(sc);
1471
1472	if (pci_find_extcap(sc->ale_dev, PCIY_PMG, &pmc) != 0) {
1473		/* Disable WOL. */
1474		CSR_WRITE_4(sc, ALE_WOL_CFG, 0);
1475		reg = CSR_READ_4(sc, ALE_PCIE_PHYMISC);
1476		reg |= PCIE_PHYMISC_FORCE_RCV_DET;
1477		CSR_WRITE_4(sc, ALE_PCIE_PHYMISC, reg);
1478		/* Force PHY power down. */
1479		CSR_WRITE_2(sc, ALE_GPHY_CTRL,
1480		    GPHY_CTRL_EXT_RESET | GPHY_CTRL_HIB_EN |
1481		    GPHY_CTRL_HIB_PULSE | GPHY_CTRL_PHY_PLL_ON |
1482		    GPHY_CTRL_SEL_ANA_RESET | GPHY_CTRL_PHY_IDDQ |
1483		    GPHY_CTRL_PCLK_SEL_DIS | GPHY_CTRL_PWDOWN_HW);
1484		return;
1485	}
1486
1487	ifp = sc->ale_ifp;
1488	if ((ifp->if_capenable & IFCAP_WOL) != 0) {
1489		if ((sc->ale_flags & ALE_FLAG_FASTETHER) == 0)
1490			ale_setlinkspeed(sc);
1491	}
1492
1493	pmcs = 0;
1494	if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
1495		pmcs |= WOL_CFG_MAGIC | WOL_CFG_MAGIC_ENB;
1496	CSR_WRITE_4(sc, ALE_WOL_CFG, pmcs);
1497	reg = CSR_READ_4(sc, ALE_MAC_CFG);
1498	reg &= ~(MAC_CFG_DBG | MAC_CFG_PROMISC | MAC_CFG_ALLMULTI |
1499	    MAC_CFG_BCAST);
1500	if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
1501		reg |= MAC_CFG_ALLMULTI | MAC_CFG_BCAST;
1502	if ((ifp->if_capenable & IFCAP_WOL) != 0)
1503		reg |= MAC_CFG_RX_ENB;
1504	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
1505
1506	if ((ifp->if_capenable & IFCAP_WOL) == 0) {
1507		/* WOL disabled, PHY power down. */
1508		reg = CSR_READ_4(sc, ALE_PCIE_PHYMISC);
1509		reg |= PCIE_PHYMISC_FORCE_RCV_DET;
1510		CSR_WRITE_4(sc, ALE_PCIE_PHYMISC, reg);
1511		CSR_WRITE_2(sc, ALE_GPHY_CTRL,
1512		    GPHY_CTRL_EXT_RESET | GPHY_CTRL_HIB_EN |
1513		    GPHY_CTRL_HIB_PULSE | GPHY_CTRL_SEL_ANA_RESET |
1514		    GPHY_CTRL_PHY_IDDQ | GPHY_CTRL_PCLK_SEL_DIS |
1515		    GPHY_CTRL_PWDOWN_HW);
1516	}
1517	/* Request PME. */
1518	pmstat = pci_read_config(sc->ale_dev, pmc + PCIR_POWER_STATUS, 2);
1519	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
1520	if ((ifp->if_capenable & IFCAP_WOL) != 0)
1521		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
1522	pci_write_config(sc->ale_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
1523}
1524
1525static int
1526ale_suspend(device_t dev)
1527{
1528	struct ale_softc *sc;
1529
1530	sc = device_get_softc(dev);
1531
1532	ALE_LOCK(sc);
1533	ale_stop(sc);
1534	ale_setwol(sc);
1535	ALE_UNLOCK(sc);
1536
1537	return (0);
1538}
1539
1540static int
1541ale_resume(device_t dev)
1542{
1543	struct ale_softc *sc;
1544	struct ifnet *ifp;
1545	int pmc;
1546	uint16_t pmstat;
1547
1548	sc = device_get_softc(dev);
1549
1550	ALE_LOCK(sc);
1551	if (pci_find_extcap(sc->ale_dev, PCIY_PMG, &pmc) == 0) {
1552		/* Disable PME and clear PME status. */
1553		pmstat = pci_read_config(sc->ale_dev,
1554		    pmc + PCIR_POWER_STATUS, 2);
1555		if ((pmstat & PCIM_PSTAT_PMEENABLE) != 0) {
1556			pmstat &= ~PCIM_PSTAT_PMEENABLE;
1557			pci_write_config(sc->ale_dev,
1558			    pmc + PCIR_POWER_STATUS, pmstat, 2);
1559		}
1560	}
1561	/* Reset PHY. */
1562	ale_phy_reset(sc);
1563	ifp = sc->ale_ifp;
1564	if ((ifp->if_flags & IFF_UP) != 0) {
1565		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1566		ale_init_locked(sc);
1567	}
1568	ALE_UNLOCK(sc);
1569
1570	return (0);
1571}
1572
1573static int
1574ale_encap(struct ale_softc *sc, struct mbuf **m_head)
1575{
1576	struct ale_txdesc *txd, *txd_last;
1577	struct tx_desc *desc;
1578	struct mbuf *m;
1579	struct ip *ip;
1580	struct tcphdr *tcp;
1581	bus_dma_segment_t txsegs[ALE_MAXTXSEGS];
1582	bus_dmamap_t map;
1583	uint32_t cflags, hdrlen, ip_off, poff, vtag;
1584	int error, i, nsegs, prod, si;
1585
1586	ALE_LOCK_ASSERT(sc);
1587
1588	M_ASSERTPKTHDR((*m_head));
1589
1590	m = *m_head;
1591	ip = NULL;
1592	tcp = NULL;
1593	cflags = vtag = 0;
1594	ip_off = poff = 0;
1595	if ((m->m_pkthdr.csum_flags & (ALE_CSUM_FEATURES | CSUM_TSO)) != 0) {
1596		/*
1597		 * AR81xx requires offset of TCP/UDP payload in its Tx
1598		 * descriptor to perform hardware Tx checksum offload.
1599		 * Additionally, TSO requires IP/TCP header size and
1600		 * modification of IP/TCP header in order to make TSO
1601		 * engine work. This kind of operation takes many CPU
1602		 * cycles on FreeBSD so fast host CPU is required to
1603		 * get smooth TSO performance.
1604		 */
1605		struct ether_header *eh;
1606
1607		if (M_WRITABLE(m) == 0) {
1608			/* Get a writable copy. */
1609			m = m_dup(*m_head, M_DONTWAIT);
1610			/* Release original mbufs. */
1611			m_freem(*m_head);
1612			if (m == NULL) {
1613				*m_head = NULL;
1614				return (ENOBUFS);
1615			}
1616			*m_head = m;
1617		}
1618
1619		/*
1620		 * Buggy-controller requires 4 byte aligned Tx buffer
1621		 * to make custom checksum offload work.
1622		 */
1623		if ((sc->ale_flags & ALE_FLAG_TXCSUM_BUG) != 0 &&
1624		    (m->m_pkthdr.csum_flags & ALE_CSUM_FEATURES) != 0 &&
1625		    (mtod(m, intptr_t) & 3) != 0) {
1626			m = m_defrag(*m_head, M_DONTWAIT);
1627			if (m == NULL) {
1628				*m_head = NULL;
1629				return (ENOBUFS);
1630			}
1631			*m_head = m;
1632		}
1633
1634		ip_off = sizeof(struct ether_header);
1635		m = m_pullup(m, ip_off);
1636		if (m == NULL) {
1637			*m_head = NULL;
1638			return (ENOBUFS);
1639		}
1640		eh = mtod(m, struct ether_header *);
1641		/*
1642		 * Check if hardware VLAN insertion is off.
1643		 * Additional check for LLC/SNAP frame?
1644		 */
1645		if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
1646			ip_off = sizeof(struct ether_vlan_header);
1647			m = m_pullup(m, ip_off);
1648			if (m == NULL) {
1649				*m_head = NULL;
1650				return (ENOBUFS);
1651			}
1652		}
1653		m = m_pullup(m, ip_off + sizeof(struct ip));
1654		if (m == NULL) {
1655			*m_head = NULL;
1656			return (ENOBUFS);
1657		}
1658		ip = (struct ip *)(mtod(m, char *) + ip_off);
1659		poff = ip_off + (ip->ip_hl << 2);
1660		if ((m->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
1661			/*
1662			 * XXX
1663			 * AR81xx requires the first descriptor should
1664			 * not include any TCP playload for TSO case.
1665			 * (i.e. ethernet header + IP + TCP header only)
1666			 * m_pullup(9) above will ensure this too.
1667			 * However it's not correct if the first mbuf
1668			 * of the chain does not use cluster.
1669			 */
1670			m = m_pullup(m, poff + sizeof(struct tcphdr));
1671			if (m == NULL) {
1672				*m_head = NULL;
1673				return (ENOBUFS);
1674			}
1675			ip = (struct ip *)(mtod(m, char *) + ip_off);
1676			tcp = (struct tcphdr *)(mtod(m, char *) + poff);
1677			m = m_pullup(m, poff + (tcp->th_off << 2));
1678			if (m == NULL) {
1679				*m_head = NULL;
1680				return (ENOBUFS);
1681			}
1682			/*
1683			 * AR81xx requires IP/TCP header size and offset as
1684			 * well as TCP pseudo checksum which complicates
1685			 * TSO configuration. I guess this comes from the
1686			 * adherence to Microsoft NDIS Large Send
1687			 * specification which requires insertion of
1688			 * pseudo checksum by upper stack. The pseudo
1689			 * checksum that NDIS refers to doesn't include
1690			 * TCP payload length so ale(4) should recompute
1691			 * the pseudo checksum here. Hopefully this wouldn't
1692			 * be much burden on modern CPUs.
1693			 * Reset IP checksum and recompute TCP pseudo
1694			 * checksum as NDIS specification said.
1695			 */
1696			ip->ip_sum = 0;
1697			tcp->th_sum = in_pseudo(ip->ip_src.s_addr,
1698			    ip->ip_dst.s_addr, htons(IPPROTO_TCP));
1699		}
1700		*m_head = m;
1701	}
1702
1703	si = prod = sc->ale_cdata.ale_tx_prod;
1704	txd = &sc->ale_cdata.ale_txdesc[prod];
1705	txd_last = txd;
1706	map = txd->tx_dmamap;
1707
1708	error =  bus_dmamap_load_mbuf_sg(sc->ale_cdata.ale_tx_tag, map,
1709	    *m_head, txsegs, &nsegs, 0);
1710	if (error == EFBIG) {
1711		m = m_collapse(*m_head, M_DONTWAIT, ALE_MAXTXSEGS);
1712		if (m == NULL) {
1713			m_freem(*m_head);
1714			*m_head = NULL;
1715			return (ENOMEM);
1716		}
1717		*m_head = m;
1718		error = bus_dmamap_load_mbuf_sg(sc->ale_cdata.ale_tx_tag, map,
1719		    *m_head, txsegs, &nsegs, 0);
1720		if (error != 0) {
1721			m_freem(*m_head);
1722			*m_head = NULL;
1723			return (error);
1724		}
1725	} else if (error != 0)
1726		return (error);
1727	if (nsegs == 0) {
1728		m_freem(*m_head);
1729		*m_head = NULL;
1730		return (EIO);
1731	}
1732
1733	/* Check descriptor overrun. */
1734	if (sc->ale_cdata.ale_tx_cnt + nsegs >= ALE_TX_RING_CNT - 3) {
1735		bus_dmamap_unload(sc->ale_cdata.ale_tx_tag, map);
1736		return (ENOBUFS);
1737	}
1738	bus_dmamap_sync(sc->ale_cdata.ale_tx_tag, map, BUS_DMASYNC_PREWRITE);
1739
1740	m = *m_head;
1741	if ((m->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
1742		/* Request TSO and set MSS. */
1743		cflags |= ALE_TD_TSO;
1744		cflags |= ((uint32_t)m->m_pkthdr.tso_segsz << ALE_TD_MSS_SHIFT);
1745		/* Set IP/TCP header size. */
1746		cflags |= ip->ip_hl << ALE_TD_IPHDR_LEN_SHIFT;
1747		cflags |= tcp->th_off << ALE_TD_TCPHDR_LEN_SHIFT;
1748	} else if ((m->m_pkthdr.csum_flags & ALE_CSUM_FEATURES) != 0) {
1749		/*
1750		 * AR81xx supports Tx custom checksum offload feature
1751		 * that offloads single 16bit checksum computation.
1752		 * So you can choose one among IP, TCP and UDP.
1753		 * Normally driver sets checksum start/insertion
1754		 * position from the information of TCP/UDP frame as
1755		 * TCP/UDP checksum takes more time than that of IP.
1756		 * However it seems that custom checksum offload
1757		 * requires 4 bytes aligned Tx buffers due to hardware
1758		 * bug.
1759		 * AR81xx also supports explicit Tx checksum computation
1760		 * if it is told that the size of IP header and TCP
1761		 * header(for UDP, the header size does not matter
1762		 * because it's fixed length). However with this scheme
1763		 * TSO does not work so you have to choose one either
1764		 * TSO or explicit Tx checksum offload. I chosen TSO
1765		 * plus custom checksum offload with work-around which
1766		 * will cover most common usage for this consumer
1767		 * ethernet controller. The work-around takes a lot of
1768		 * CPU cycles if Tx buffer is not aligned on 4 bytes
1769		 * boundary, though.
1770		 */
1771		cflags |= ALE_TD_CXSUM;
1772		/* Set checksum start offset. */
1773		cflags |= (poff << ALE_TD_CSUM_PLOADOFFSET_SHIFT);
1774		/* Set checksum insertion position of TCP/UDP. */
1775		cflags |= ((poff + m->m_pkthdr.csum_data) <<
1776		    ALE_TD_CSUM_XSUMOFFSET_SHIFT);
1777	}
1778
1779	/* Configure VLAN hardware tag insertion. */
1780	if ((m->m_flags & M_VLANTAG) != 0) {
1781		vtag = ALE_TX_VLAN_TAG(m->m_pkthdr.ether_vtag);
1782		vtag = ((vtag << ALE_TD_VLAN_SHIFT) & ALE_TD_VLAN_MASK);
1783		cflags |= ALE_TD_INSERT_VLAN_TAG;
1784	}
1785
1786	i = 0;
1787	if ((m->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
1788		/*
1789		 * Make sure the first fragment contains
1790		 * only ethernet and IP/TCP header with options.
1791		 */
1792		hdrlen =  poff + (tcp->th_off << 2);
1793		desc = &sc->ale_cdata.ale_tx_ring[prod];
1794		desc->addr = htole64(txsegs[i].ds_addr);
1795		desc->len = htole32(ALE_TX_BYTES(hdrlen) | vtag);
1796		desc->flags = htole32(cflags);
1797		sc->ale_cdata.ale_tx_cnt++;
1798		ALE_DESC_INC(prod, ALE_TX_RING_CNT);
1799		if (m->m_len - hdrlen > 0) {
1800			/* Handle remaining payload of the first fragment. */
1801			desc = &sc->ale_cdata.ale_tx_ring[prod];
1802			desc->addr = htole64(txsegs[i].ds_addr + hdrlen);
1803			desc->len = htole32(ALE_TX_BYTES(m->m_len - hdrlen) |
1804			    vtag);
1805			desc->flags = htole32(cflags);
1806			sc->ale_cdata.ale_tx_cnt++;
1807			ALE_DESC_INC(prod, ALE_TX_RING_CNT);
1808		}
1809		i = 1;
1810	}
1811	for (; i < nsegs; i++) {
1812		desc = &sc->ale_cdata.ale_tx_ring[prod];
1813		desc->addr = htole64(txsegs[i].ds_addr);
1814		desc->len = htole32(ALE_TX_BYTES(txsegs[i].ds_len) | vtag);
1815		desc->flags = htole32(cflags);
1816		sc->ale_cdata.ale_tx_cnt++;
1817		ALE_DESC_INC(prod, ALE_TX_RING_CNT);
1818	}
1819	/* Update producer index. */
1820	sc->ale_cdata.ale_tx_prod = prod;
1821	/* Set TSO header on the first descriptor. */
1822	if ((m->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
1823		desc = &sc->ale_cdata.ale_tx_ring[si];
1824		desc->flags |= htole32(ALE_TD_TSO_HDR);
1825	}
1826
1827	/* Finally set EOP on the last descriptor. */
1828	prod = (prod + ALE_TX_RING_CNT - 1) % ALE_TX_RING_CNT;
1829	desc = &sc->ale_cdata.ale_tx_ring[prod];
1830	desc->flags |= htole32(ALE_TD_EOP);
1831
1832	/* Swap dmamap of the first and the last. */
1833	txd = &sc->ale_cdata.ale_txdesc[prod];
1834	map = txd_last->tx_dmamap;
1835	txd_last->tx_dmamap = txd->tx_dmamap;
1836	txd->tx_dmamap = map;
1837	txd->tx_m = m;
1838
1839	/* Sync descriptors. */
1840	bus_dmamap_sync(sc->ale_cdata.ale_tx_ring_tag,
1841	    sc->ale_cdata.ale_tx_ring_map,
1842	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1843
1844	return (0);
1845}
1846
1847static void
1848ale_start(struct ifnet *ifp)
1849{
1850        struct ale_softc *sc;
1851
1852	sc = ifp->if_softc;
1853	ALE_LOCK(sc);
1854	ale_start_locked(ifp);
1855	ALE_UNLOCK(sc);
1856}
1857
1858static void
1859ale_start_locked(struct ifnet *ifp)
1860{
1861        struct ale_softc *sc;
1862        struct mbuf *m_head;
1863	int enq;
1864
1865	sc = ifp->if_softc;
1866
1867	ALE_LOCK_ASSERT(sc);
1868
1869	/* Reclaim transmitted frames. */
1870	if (sc->ale_cdata.ale_tx_cnt >= ALE_TX_DESC_HIWAT)
1871		ale_txeof(sc);
1872
1873	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1874	    IFF_DRV_RUNNING || (sc->ale_flags & ALE_FLAG_LINK) == 0)
1875		return;
1876
1877	for (enq = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd); ) {
1878		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
1879		if (m_head == NULL)
1880			break;
1881		/*
1882		 * Pack the data into the transmit ring. If we
1883		 * don't have room, set the OACTIVE flag and wait
1884		 * for the NIC to drain the ring.
1885		 */
1886		if (ale_encap(sc, &m_head)) {
1887			if (m_head == NULL)
1888				break;
1889			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1890			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
1891			break;
1892		}
1893
1894		enq++;
1895		/*
1896		 * If there's a BPF listener, bounce a copy of this frame
1897		 * to him.
1898		 */
1899		ETHER_BPF_MTAP(ifp, m_head);
1900	}
1901
1902	if (enq > 0) {
1903		/* Kick. */
1904		CSR_WRITE_4(sc, ALE_MBOX_TPD_PROD_IDX,
1905		    sc->ale_cdata.ale_tx_prod);
1906		/* Set a timeout in case the chip goes out to lunch. */
1907		sc->ale_watchdog_timer = ALE_TX_TIMEOUT;
1908	}
1909}
1910
1911static void
1912ale_watchdog(struct ale_softc *sc)
1913{
1914	struct ifnet *ifp;
1915
1916	ALE_LOCK_ASSERT(sc);
1917
1918	if (sc->ale_watchdog_timer == 0 || --sc->ale_watchdog_timer)
1919		return;
1920
1921	ifp = sc->ale_ifp;
1922	if ((sc->ale_flags & ALE_FLAG_LINK) == 0) {
1923		if_printf(sc->ale_ifp, "watchdog timeout (lost link)\n");
1924		ifp->if_oerrors++;
1925		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1926		ale_init_locked(sc);
1927		return;
1928	}
1929	if_printf(sc->ale_ifp, "watchdog timeout -- resetting\n");
1930	ifp->if_oerrors++;
1931	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1932	ale_init_locked(sc);
1933	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1934		ale_start_locked(ifp);
1935}
1936
1937static int
1938ale_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1939{
1940	struct ale_softc *sc;
1941	struct ifreq *ifr;
1942	struct mii_data *mii;
1943	int error, mask;
1944
1945	sc = ifp->if_softc;
1946	ifr = (struct ifreq *)data;
1947	error = 0;
1948	switch (cmd) {
1949	case SIOCSIFMTU:
1950		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > ALE_JUMBO_MTU ||
1951		    ((sc->ale_flags & ALE_FLAG_JUMBO) == 0 &&
1952		    ifr->ifr_mtu > ETHERMTU))
1953			error = EINVAL;
1954		else if (ifp->if_mtu != ifr->ifr_mtu) {
1955			ALE_LOCK(sc);
1956			ifp->if_mtu = ifr->ifr_mtu;
1957			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
1958				ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1959				ale_init_locked(sc);
1960			}
1961			ALE_UNLOCK(sc);
1962		}
1963		break;
1964	case SIOCSIFFLAGS:
1965		ALE_LOCK(sc);
1966		if ((ifp->if_flags & IFF_UP) != 0) {
1967			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
1968				if (((ifp->if_flags ^ sc->ale_if_flags)
1969				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
1970					ale_rxfilter(sc);
1971			} else {
1972				ale_init_locked(sc);
1973			}
1974		} else {
1975			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
1976				ale_stop(sc);
1977		}
1978		sc->ale_if_flags = ifp->if_flags;
1979		ALE_UNLOCK(sc);
1980		break;
1981	case SIOCADDMULTI:
1982	case SIOCDELMULTI:
1983		ALE_LOCK(sc);
1984		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
1985			ale_rxfilter(sc);
1986		ALE_UNLOCK(sc);
1987		break;
1988	case SIOCSIFMEDIA:
1989	case SIOCGIFMEDIA:
1990		mii = device_get_softc(sc->ale_miibus);
1991		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, cmd);
1992		break;
1993	case SIOCSIFCAP:
1994		ALE_LOCK(sc);
1995		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
1996		if ((mask & IFCAP_TXCSUM) != 0 &&
1997		    (ifp->if_capabilities & IFCAP_TXCSUM) != 0) {
1998			ifp->if_capenable ^= IFCAP_TXCSUM;
1999			if ((ifp->if_capenable & IFCAP_TXCSUM) != 0)
2000				ifp->if_hwassist |= ALE_CSUM_FEATURES;
2001			else
2002				ifp->if_hwassist &= ~ALE_CSUM_FEATURES;
2003		}
2004		if ((mask & IFCAP_RXCSUM) != 0 &&
2005		    (ifp->if_capabilities & IFCAP_RXCSUM) != 0)
2006			ifp->if_capenable ^= IFCAP_RXCSUM;
2007		if ((mask & IFCAP_TSO4) != 0 &&
2008		    (ifp->if_capabilities & IFCAP_TSO4) != 0) {
2009			ifp->if_capenable ^= IFCAP_TSO4;
2010			if ((ifp->if_capenable & IFCAP_TSO4) != 0)
2011				ifp->if_hwassist |= CSUM_TSO;
2012			else
2013				ifp->if_hwassist &= ~CSUM_TSO;
2014		}
2015
2016		if ((mask & IFCAP_WOL_MCAST) != 0 &&
2017		    (ifp->if_capabilities & IFCAP_WOL_MCAST) != 0)
2018			ifp->if_capenable ^= IFCAP_WOL_MCAST;
2019		if ((mask & IFCAP_WOL_MAGIC) != 0 &&
2020		    (ifp->if_capabilities & IFCAP_WOL_MAGIC) != 0)
2021			ifp->if_capenable ^= IFCAP_WOL_MAGIC;
2022		if ((mask & IFCAP_VLAN_HWCSUM) != 0 &&
2023		    (ifp->if_capabilities & IFCAP_VLAN_HWCSUM) != 0)
2024			ifp->if_capenable ^= IFCAP_VLAN_HWCSUM;
2025		if ((mask & IFCAP_VLAN_HWTSO) != 0 &&
2026		    (ifp->if_capabilities & IFCAP_VLAN_HWTSO) != 0)
2027			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
2028		if ((mask & IFCAP_VLAN_HWTAGGING) != 0 &&
2029		    (ifp->if_capabilities & IFCAP_VLAN_HWTAGGING) != 0) {
2030			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
2031			if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
2032				ifp->if_capenable &= ~IFCAP_VLAN_HWTSO;
2033			ale_rxvlan(sc);
2034		}
2035		ALE_UNLOCK(sc);
2036		VLAN_CAPABILITIES(ifp);
2037		break;
2038	default:
2039		error = ether_ioctl(ifp, cmd, data);
2040		break;
2041	}
2042
2043	return (error);
2044}
2045
2046static void
2047ale_mac_config(struct ale_softc *sc)
2048{
2049	struct mii_data *mii;
2050	uint32_t reg;
2051
2052	ALE_LOCK_ASSERT(sc);
2053
2054	mii = device_get_softc(sc->ale_miibus);
2055	reg = CSR_READ_4(sc, ALE_MAC_CFG);
2056	reg &= ~(MAC_CFG_FULL_DUPLEX | MAC_CFG_TX_FC | MAC_CFG_RX_FC |
2057	    MAC_CFG_SPEED_MASK);
2058	/* Reprogram MAC with resolved speed/duplex. */
2059	switch (IFM_SUBTYPE(mii->mii_media_active)) {
2060	case IFM_10_T:
2061	case IFM_100_TX:
2062		reg |= MAC_CFG_SPEED_10_100;
2063		break;
2064	case IFM_1000_T:
2065		reg |= MAC_CFG_SPEED_1000;
2066		break;
2067	}
2068	if ((IFM_OPTIONS(mii->mii_media_active) & IFM_FDX) != 0) {
2069		reg |= MAC_CFG_FULL_DUPLEX;
2070#ifdef notyet
2071		if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_TXPAUSE) != 0)
2072			reg |= MAC_CFG_TX_FC;
2073		if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_RXPAUSE) != 0)
2074			reg |= MAC_CFG_RX_FC;
2075#endif
2076	}
2077	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2078}
2079
2080static void
2081ale_link_task(void *arg, int pending)
2082{
2083	struct ale_softc *sc;
2084	struct mii_data *mii;
2085	struct ifnet *ifp;
2086	uint32_t reg;
2087
2088	sc = (struct ale_softc *)arg;
2089
2090	ALE_LOCK(sc);
2091	mii = device_get_softc(sc->ale_miibus);
2092	ifp = sc->ale_ifp;
2093	if (mii == NULL || ifp == NULL ||
2094	    (ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) {
2095		ALE_UNLOCK(sc);
2096		return;
2097	}
2098
2099	sc->ale_flags &= ~ALE_FLAG_LINK;
2100	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
2101	    (IFM_ACTIVE | IFM_AVALID)) {
2102		switch (IFM_SUBTYPE(mii->mii_media_active)) {
2103		case IFM_10_T:
2104		case IFM_100_TX:
2105			sc->ale_flags |= ALE_FLAG_LINK;
2106			break;
2107		case IFM_1000_T:
2108			if ((sc->ale_flags & ALE_FLAG_FASTETHER) == 0)
2109				sc->ale_flags |= ALE_FLAG_LINK;
2110			break;
2111		default:
2112			break;
2113		}
2114	}
2115
2116	/* Stop Rx/Tx MACs. */
2117	ale_stop_mac(sc);
2118
2119	/* Program MACs with resolved speed/duplex/flow-control. */
2120	if ((sc->ale_flags & ALE_FLAG_LINK) != 0) {
2121		ale_mac_config(sc);
2122		/* Reenable Tx/Rx MACs. */
2123		reg = CSR_READ_4(sc, ALE_MAC_CFG);
2124		reg |= MAC_CFG_TX_ENB | MAC_CFG_RX_ENB;
2125		CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2126	}
2127
2128	ALE_UNLOCK(sc);
2129}
2130
2131static void
2132ale_stats_clear(struct ale_softc *sc)
2133{
2134	struct smb sb;
2135	uint32_t *reg;
2136	int i;
2137
2138	for (reg = &sb.rx_frames, i = 0; reg <= &sb.rx_pkts_filtered; reg++) {
2139		CSR_READ_4(sc, ALE_RX_MIB_BASE + i);
2140		i += sizeof(uint32_t);
2141	}
2142	/* Read Tx statistics. */
2143	for (reg = &sb.tx_frames, i = 0; reg <= &sb.tx_mcast_bytes; reg++) {
2144		CSR_READ_4(sc, ALE_TX_MIB_BASE + i);
2145		i += sizeof(uint32_t);
2146	}
2147}
2148
2149static void
2150ale_stats_update(struct ale_softc *sc)
2151{
2152	struct ale_hw_stats *stat;
2153	struct smb sb, *smb;
2154	struct ifnet *ifp;
2155	uint32_t *reg;
2156	int i;
2157
2158	ALE_LOCK_ASSERT(sc);
2159
2160	ifp = sc->ale_ifp;
2161	stat = &sc->ale_stats;
2162	smb = &sb;
2163
2164	/* Read Rx statistics. */
2165	for (reg = &sb.rx_frames, i = 0; reg <= &sb.rx_pkts_filtered; reg++) {
2166		*reg = CSR_READ_4(sc, ALE_RX_MIB_BASE + i);
2167		i += sizeof(uint32_t);
2168	}
2169	/* Read Tx statistics. */
2170	for (reg = &sb.tx_frames, i = 0; reg <= &sb.tx_mcast_bytes; reg++) {
2171		*reg = CSR_READ_4(sc, ALE_TX_MIB_BASE + i);
2172		i += sizeof(uint32_t);
2173	}
2174
2175	/* Rx stats. */
2176	stat->rx_frames += smb->rx_frames;
2177	stat->rx_bcast_frames += smb->rx_bcast_frames;
2178	stat->rx_mcast_frames += smb->rx_mcast_frames;
2179	stat->rx_pause_frames += smb->rx_pause_frames;
2180	stat->rx_control_frames += smb->rx_control_frames;
2181	stat->rx_crcerrs += smb->rx_crcerrs;
2182	stat->rx_lenerrs += smb->rx_lenerrs;
2183	stat->rx_bytes += smb->rx_bytes;
2184	stat->rx_runts += smb->rx_runts;
2185	stat->rx_fragments += smb->rx_fragments;
2186	stat->rx_pkts_64 += smb->rx_pkts_64;
2187	stat->rx_pkts_65_127 += smb->rx_pkts_65_127;
2188	stat->rx_pkts_128_255 += smb->rx_pkts_128_255;
2189	stat->rx_pkts_256_511 += smb->rx_pkts_256_511;
2190	stat->rx_pkts_512_1023 += smb->rx_pkts_512_1023;
2191	stat->rx_pkts_1024_1518 += smb->rx_pkts_1024_1518;
2192	stat->rx_pkts_1519_max += smb->rx_pkts_1519_max;
2193	stat->rx_pkts_truncated += smb->rx_pkts_truncated;
2194	stat->rx_fifo_oflows += smb->rx_fifo_oflows;
2195	stat->rx_rrs_errs += smb->rx_rrs_errs;
2196	stat->rx_alignerrs += smb->rx_alignerrs;
2197	stat->rx_bcast_bytes += smb->rx_bcast_bytes;
2198	stat->rx_mcast_bytes += smb->rx_mcast_bytes;
2199	stat->rx_pkts_filtered += smb->rx_pkts_filtered;
2200
2201	/* Tx stats. */
2202	stat->tx_frames += smb->tx_frames;
2203	stat->tx_bcast_frames += smb->tx_bcast_frames;
2204	stat->tx_mcast_frames += smb->tx_mcast_frames;
2205	stat->tx_pause_frames += smb->tx_pause_frames;
2206	stat->tx_excess_defer += smb->tx_excess_defer;
2207	stat->tx_control_frames += smb->tx_control_frames;
2208	stat->tx_deferred += smb->tx_deferred;
2209	stat->tx_bytes += smb->tx_bytes;
2210	stat->tx_pkts_64 += smb->tx_pkts_64;
2211	stat->tx_pkts_65_127 += smb->tx_pkts_65_127;
2212	stat->tx_pkts_128_255 += smb->tx_pkts_128_255;
2213	stat->tx_pkts_256_511 += smb->tx_pkts_256_511;
2214	stat->tx_pkts_512_1023 += smb->tx_pkts_512_1023;
2215	stat->tx_pkts_1024_1518 += smb->tx_pkts_1024_1518;
2216	stat->tx_pkts_1519_max += smb->tx_pkts_1519_max;
2217	stat->tx_single_colls += smb->tx_single_colls;
2218	stat->tx_multi_colls += smb->tx_multi_colls;
2219	stat->tx_late_colls += smb->tx_late_colls;
2220	stat->tx_excess_colls += smb->tx_excess_colls;
2221	stat->tx_abort += smb->tx_abort;
2222	stat->tx_underrun += smb->tx_underrun;
2223	stat->tx_desc_underrun += smb->tx_desc_underrun;
2224	stat->tx_lenerrs += smb->tx_lenerrs;
2225	stat->tx_pkts_truncated += smb->tx_pkts_truncated;
2226	stat->tx_bcast_bytes += smb->tx_bcast_bytes;
2227	stat->tx_mcast_bytes += smb->tx_mcast_bytes;
2228
2229	/* Update counters in ifnet. */
2230	ifp->if_opackets += smb->tx_frames;
2231
2232	ifp->if_collisions += smb->tx_single_colls +
2233	    smb->tx_multi_colls * 2 + smb->tx_late_colls +
2234	    smb->tx_abort * HDPX_CFG_RETRY_DEFAULT;
2235
2236	/*
2237	 * XXX
2238	 * tx_pkts_truncated counter looks suspicious. It constantly
2239	 * increments with no sign of Tx errors. This may indicate
2240	 * the counter name is not correct one so I've removed the
2241	 * counter in output errors.
2242	 */
2243	ifp->if_oerrors += smb->tx_abort + smb->tx_late_colls +
2244	    smb->tx_underrun;
2245
2246	ifp->if_ipackets += smb->rx_frames;
2247
2248	ifp->if_ierrors += smb->rx_crcerrs + smb->rx_lenerrs +
2249	    smb->rx_runts + smb->rx_pkts_truncated +
2250	    smb->rx_fifo_oflows + smb->rx_rrs_errs +
2251	    smb->rx_alignerrs;
2252}
2253
2254static int
2255ale_intr(void *arg)
2256{
2257	struct ale_softc *sc;
2258	uint32_t status;
2259
2260	sc = (struct ale_softc *)arg;
2261
2262	status = CSR_READ_4(sc, ALE_INTR_STATUS);
2263	if ((status & ALE_INTRS) == 0)
2264		return (FILTER_STRAY);
2265	/* Disable interrupts. */
2266	CSR_WRITE_4(sc, ALE_INTR_STATUS, INTR_DIS_INT);
2267	taskqueue_enqueue(sc->ale_tq, &sc->ale_int_task);
2268
2269	return (FILTER_HANDLED);
2270}
2271
2272static void
2273ale_int_task(void *arg, int pending)
2274{
2275	struct ale_softc *sc;
2276	struct ifnet *ifp;
2277	uint32_t status;
2278	int more;
2279
2280	sc = (struct ale_softc *)arg;
2281
2282	status = CSR_READ_4(sc, ALE_INTR_STATUS);
2283	ALE_LOCK(sc);
2284	if (sc->ale_morework != 0)
2285		status |= INTR_RX_PKT;
2286	if ((status & ALE_INTRS) == 0)
2287		goto done;
2288
2289	/* Acknowledge interrupts but still disable interrupts. */
2290	CSR_WRITE_4(sc, ALE_INTR_STATUS, status | INTR_DIS_INT);
2291
2292	ifp = sc->ale_ifp;
2293	more = 0;
2294	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2295		more = ale_rxeof(sc, sc->ale_process_limit);
2296		if (more == EAGAIN)
2297			sc->ale_morework = 1;
2298		else if (more == EIO) {
2299			sc->ale_stats.reset_brk_seq++;
2300			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2301			ale_init_locked(sc);
2302			ALE_UNLOCK(sc);
2303			return;
2304		}
2305
2306		if ((status & (INTR_DMA_RD_TO_RST | INTR_DMA_WR_TO_RST)) != 0) {
2307			if ((status & INTR_DMA_RD_TO_RST) != 0)
2308				device_printf(sc->ale_dev,
2309				    "DMA read error! -- resetting\n");
2310			if ((status & INTR_DMA_WR_TO_RST) != 0)
2311				device_printf(sc->ale_dev,
2312				    "DMA write error! -- resetting\n");
2313			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2314			ale_init_locked(sc);
2315			ALE_UNLOCK(sc);
2316			return;
2317		}
2318		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2319			ale_start_locked(ifp);
2320	}
2321
2322	if (more == EAGAIN ||
2323	    (CSR_READ_4(sc, ALE_INTR_STATUS) & ALE_INTRS) != 0) {
2324		ALE_UNLOCK(sc);
2325		taskqueue_enqueue(sc->ale_tq, &sc->ale_int_task);
2326		return;
2327	}
2328
2329done:
2330	ALE_UNLOCK(sc);
2331
2332	/* Re-enable interrupts. */
2333	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0x7FFFFFFF);
2334}
2335
2336static void
2337ale_txeof(struct ale_softc *sc)
2338{
2339	struct ifnet *ifp;
2340	struct ale_txdesc *txd;
2341	uint32_t cons, prod;
2342	int prog;
2343
2344	ALE_LOCK_ASSERT(sc);
2345
2346	ifp = sc->ale_ifp;
2347
2348	if (sc->ale_cdata.ale_tx_cnt == 0)
2349		return;
2350
2351	bus_dmamap_sync(sc->ale_cdata.ale_tx_ring_tag,
2352	    sc->ale_cdata.ale_tx_ring_map,
2353	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2354	if ((sc->ale_flags & ALE_FLAG_TXCMB_BUG) == 0) {
2355		bus_dmamap_sync(sc->ale_cdata.ale_tx_cmb_tag,
2356		    sc->ale_cdata.ale_tx_cmb_map,
2357		    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2358		prod = *sc->ale_cdata.ale_tx_cmb & TPD_CNT_MASK;
2359	} else
2360		prod = CSR_READ_2(sc, ALE_TPD_CONS_IDX);
2361	cons = sc->ale_cdata.ale_tx_cons;
2362	/*
2363	 * Go through our Tx list and free mbufs for those
2364	 * frames which have been transmitted.
2365	 */
2366	for (prog = 0; cons != prod; prog++,
2367	    ALE_DESC_INC(cons, ALE_TX_RING_CNT)) {
2368		if (sc->ale_cdata.ale_tx_cnt <= 0)
2369			break;
2370		prog++;
2371		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2372		sc->ale_cdata.ale_tx_cnt--;
2373		txd = &sc->ale_cdata.ale_txdesc[cons];
2374		if (txd->tx_m != NULL) {
2375			/* Reclaim transmitted mbufs. */
2376			bus_dmamap_sync(sc->ale_cdata.ale_tx_tag,
2377			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2378			bus_dmamap_unload(sc->ale_cdata.ale_tx_tag,
2379			    txd->tx_dmamap);
2380			m_freem(txd->tx_m);
2381			txd->tx_m = NULL;
2382		}
2383	}
2384
2385	if (prog > 0) {
2386		sc->ale_cdata.ale_tx_cons = cons;
2387		/*
2388		 * Unarm watchdog timer only when there is no pending
2389		 * Tx descriptors in queue.
2390		 */
2391		if (sc->ale_cdata.ale_tx_cnt == 0)
2392			sc->ale_watchdog_timer = 0;
2393	}
2394}
2395
2396static void
2397ale_rx_update_page(struct ale_softc *sc, struct ale_rx_page **page,
2398    uint32_t length, uint32_t *prod)
2399{
2400	struct ale_rx_page *rx_page;
2401
2402	rx_page = *page;
2403	/* Update consumer position. */
2404	rx_page->cons += roundup(length + sizeof(struct rx_rs),
2405	    ALE_RX_PAGE_ALIGN);
2406	if (rx_page->cons >= ALE_RX_PAGE_SZ) {
2407		/*
2408		 * End of Rx page reached, let hardware reuse
2409		 * this page.
2410		 */
2411		rx_page->cons = 0;
2412		*rx_page->cmb_addr = 0;
2413		bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2414		    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2415		CSR_WRITE_1(sc, ALE_RXF0_PAGE0 + sc->ale_cdata.ale_rx_curp,
2416		    RXF_VALID);
2417		/* Switch to alternate Rx page. */
2418		sc->ale_cdata.ale_rx_curp ^= 1;
2419		rx_page = *page =
2420		    &sc->ale_cdata.ale_rx_page[sc->ale_cdata.ale_rx_curp];
2421		/* Page flipped, sync CMB and Rx page. */
2422		bus_dmamap_sync(rx_page->page_tag, rx_page->page_map,
2423		    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2424		bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2425		    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2426		/* Sync completed, cache updated producer index. */
2427		*prod = *rx_page->cmb_addr;
2428	}
2429}
2430
2431
2432/*
2433 * It seems that AR81xx controller can compute partial checksum.
2434 * The partial checksum value can be used to accelerate checksum
2435 * computation for fragmented TCP/UDP packets. Upper network stack
2436 * already takes advantage of the partial checksum value in IP
2437 * reassembly stage. But I'm not sure the correctness of the
2438 * partial hardware checksum assistance due to lack of data sheet.
2439 * In addition, the Rx feature of controller that requires copying
2440 * for every frames effectively nullifies one of most nice offload
2441 * capability of controller.
2442 */
2443static void
2444ale_rxcsum(struct ale_softc *sc, struct mbuf *m, uint32_t status)
2445{
2446	struct ifnet *ifp;
2447	struct ip *ip;
2448	char *p;
2449
2450	ifp = sc->ale_ifp;
2451	m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
2452	if ((status & ALE_RD_IPCSUM_NOK) == 0)
2453		m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
2454
2455	if ((sc->ale_flags & ALE_FLAG_RXCSUM_BUG) == 0) {
2456		if (((status & ALE_RD_IPV4_FRAG) == 0) &&
2457		    ((status & (ALE_RD_TCP | ALE_RD_UDP)) != 0) &&
2458		    ((status & ALE_RD_TCP_UDPCSUM_NOK) == 0)) {
2459			m->m_pkthdr.csum_flags |=
2460			    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2461			m->m_pkthdr.csum_data = 0xffff;
2462		}
2463	} else {
2464		if ((status & (ALE_RD_TCP | ALE_RD_UDP)) != 0 &&
2465		    (status & ALE_RD_TCP_UDPCSUM_NOK) == 0) {
2466			p = mtod(m, char *);
2467			p += ETHER_HDR_LEN;
2468			if ((status & ALE_RD_802_3) != 0)
2469				p += LLC_SNAPFRAMELEN;
2470			if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0 &&
2471			    (status & ALE_RD_VLAN) != 0)
2472				p += ETHER_VLAN_ENCAP_LEN;
2473			ip = (struct ip *)p;
2474			if (ip->ip_off != 0 && (status & ALE_RD_IPV4_DF) == 0)
2475				return;
2476			m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
2477			    CSUM_PSEUDO_HDR;
2478			m->m_pkthdr.csum_data = 0xffff;
2479		}
2480	}
2481	/*
2482	 * Don't mark bad checksum for TCP/UDP frames
2483	 * as fragmented frames may always have set
2484	 * bad checksummed bit of frame status.
2485	 */
2486}
2487
2488/* Process received frames. */
2489static int
2490ale_rxeof(struct ale_softc *sc, int count)
2491{
2492	struct ale_rx_page *rx_page;
2493	struct rx_rs *rs;
2494	struct ifnet *ifp;
2495	struct mbuf *m;
2496	uint32_t length, prod, seqno, status, vtags;
2497	int prog;
2498
2499	ifp = sc->ale_ifp;
2500	rx_page = &sc->ale_cdata.ale_rx_page[sc->ale_cdata.ale_rx_curp];
2501	bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2502	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2503	bus_dmamap_sync(rx_page->page_tag, rx_page->page_map,
2504	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2505	/*
2506	 * Don't directly access producer index as hardware may
2507	 * update it while Rx handler is in progress. It would
2508	 * be even better if there is a way to let hardware
2509	 * know how far driver processed its received frames.
2510	 * Alternatively, hardware could provide a way to disable
2511	 * CMB updates until driver acknowledges the end of CMB
2512	 * access.
2513	 */
2514	prod = *rx_page->cmb_addr;
2515	for (prog = 0; prog < count; prog++) {
2516		if (rx_page->cons >= prod)
2517			break;
2518		rs = (struct rx_rs *)(rx_page->page_addr + rx_page->cons);
2519		seqno = ALE_RX_SEQNO(le32toh(rs->seqno));
2520		if (sc->ale_cdata.ale_rx_seqno != seqno) {
2521			/*
2522			 * Normally I believe this should not happen unless
2523			 * severe driver bug or corrupted memory. However
2524			 * it seems to happen under certain conditions which
2525			 * is triggered by abrupt Rx events such as initiation
2526			 * of bulk transfer of remote host. It's not easy to
2527			 * reproduce this and I doubt it could be related
2528			 * with FIFO overflow of hardware or activity of Tx
2529			 * CMB updates. I also remember similar behaviour
2530			 * seen on RealTek 8139 which uses resembling Rx
2531			 * scheme.
2532			 */
2533			if (bootverbose)
2534				device_printf(sc->ale_dev,
2535				    "garbled seq: %u, expected: %u -- "
2536				    "resetting!\n", seqno,
2537				    sc->ale_cdata.ale_rx_seqno);
2538			return (EIO);
2539		}
2540		/* Frame received. */
2541		sc->ale_cdata.ale_rx_seqno++;
2542		length = ALE_RX_BYTES(le32toh(rs->length));
2543		status = le32toh(rs->flags);
2544		if ((status & ALE_RD_ERROR) != 0) {
2545			/*
2546			 * We want to pass the following frames to upper
2547			 * layer regardless of error status of Rx return
2548			 * status.
2549			 *
2550			 *  o IP/TCP/UDP checksum is bad.
2551			 *  o frame length and protocol specific length
2552			 *     does not match.
2553			 */
2554			if ((status & (ALE_RD_CRC | ALE_RD_CODE |
2555			    ALE_RD_DRIBBLE | ALE_RD_RUNT | ALE_RD_OFLOW |
2556			    ALE_RD_TRUNC)) != 0) {
2557				ale_rx_update_page(sc, &rx_page, length, &prod);
2558				continue;
2559			}
2560		}
2561		/*
2562		 * m_devget(9) is major bottle-neck of ale(4)(It comes
2563		 * from hardware limitation). For jumbo frames we could
2564		 * get a slightly better performance if driver use
2565		 * m_getjcl(9) with proper buffer size argument. However
2566		 * that would make code more complicated and I don't
2567		 * think users would expect good Rx performance numbers
2568		 * on these low-end consumer ethernet controller.
2569		 */
2570		m = m_devget((char *)(rs + 1), length - ETHER_CRC_LEN,
2571		    ETHER_ALIGN, ifp, NULL);
2572		if (m == NULL) {
2573			ifp->if_iqdrops++;
2574			ale_rx_update_page(sc, &rx_page, length, &prod);
2575			continue;
2576		}
2577		if ((ifp->if_capenable & IFCAP_RXCSUM) != 0 &&
2578		    (status & ALE_RD_IPV4) != 0)
2579			ale_rxcsum(sc, m, status);
2580		if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0 &&
2581		    (status & ALE_RD_VLAN) != 0) {
2582			vtags = ALE_RX_VLAN(le32toh(rs->vtags));
2583			m->m_pkthdr.ether_vtag = ALE_RX_VLAN_TAG(vtags);
2584			m->m_flags |= M_VLANTAG;
2585		}
2586
2587		/* Pass it to upper layer. */
2588		ALE_UNLOCK(sc);
2589		(*ifp->if_input)(ifp, m);
2590		ALE_LOCK(sc);
2591
2592		ale_rx_update_page(sc, &rx_page, length, &prod);
2593	}
2594
2595	return (count > 0 ? 0 : EAGAIN);
2596}
2597
2598static void
2599ale_tick(void *arg)
2600{
2601	struct ale_softc *sc;
2602	struct mii_data *mii;
2603
2604	sc = (struct ale_softc *)arg;
2605
2606	ALE_LOCK_ASSERT(sc);
2607
2608	mii = device_get_softc(sc->ale_miibus);
2609	mii_tick(mii);
2610	ale_stats_update(sc);
2611	/*
2612	 * Reclaim Tx buffers that have been transferred. It's not
2613	 * needed here but it would release allocated mbuf chains
2614	 * faster and limit the maximum delay to a hz.
2615	 */
2616	ale_txeof(sc);
2617	ale_watchdog(sc);
2618	callout_reset(&sc->ale_tick_ch, hz, ale_tick, sc);
2619}
2620
2621static void
2622ale_reset(struct ale_softc *sc)
2623{
2624	uint32_t reg;
2625	int i;
2626
2627	/* Initialize PCIe module. From Linux. */
2628	CSR_WRITE_4(sc, 0x1008, CSR_READ_4(sc, 0x1008) | 0x8000);
2629
2630	CSR_WRITE_4(sc, ALE_MASTER_CFG, MASTER_RESET);
2631	for (i = ALE_RESET_TIMEOUT; i > 0; i--) {
2632		DELAY(10);
2633		if ((CSR_READ_4(sc, ALE_MASTER_CFG) & MASTER_RESET) == 0)
2634			break;
2635	}
2636	if (i == 0)
2637		device_printf(sc->ale_dev, "master reset timeout!\n");
2638
2639	for (i = ALE_RESET_TIMEOUT; i > 0; i--) {
2640		if ((reg = CSR_READ_4(sc, ALE_IDLE_STATUS)) == 0)
2641			break;
2642		DELAY(10);
2643	}
2644
2645	if (i == 0)
2646		device_printf(sc->ale_dev, "reset timeout(0x%08x)!\n", reg);
2647}
2648
2649static void
2650ale_init(void *xsc)
2651{
2652	struct ale_softc *sc;
2653
2654	sc = (struct ale_softc *)xsc;
2655	ALE_LOCK(sc);
2656	ale_init_locked(sc);
2657	ALE_UNLOCK(sc);
2658}
2659
2660static void
2661ale_init_locked(struct ale_softc *sc)
2662{
2663	struct ifnet *ifp;
2664	struct mii_data *mii;
2665	uint8_t eaddr[ETHER_ADDR_LEN];
2666	bus_addr_t paddr;
2667	uint32_t reg, rxf_hi, rxf_lo;
2668
2669	ALE_LOCK_ASSERT(sc);
2670
2671	ifp = sc->ale_ifp;
2672	mii = device_get_softc(sc->ale_miibus);
2673
2674	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2675		return;
2676	/*
2677	 * Cancel any pending I/O.
2678	 */
2679	ale_stop(sc);
2680	/*
2681	 * Reset the chip to a known state.
2682	 */
2683	ale_reset(sc);
2684	/* Initialize Tx descriptors, DMA memory blocks. */
2685	ale_init_rx_pages(sc);
2686	ale_init_tx_ring(sc);
2687
2688	/* Reprogram the station address. */
2689	bcopy(IF_LLADDR(ifp), eaddr, ETHER_ADDR_LEN);
2690	CSR_WRITE_4(sc, ALE_PAR0,
2691	    eaddr[2] << 24 | eaddr[3] << 16 | eaddr[4] << 8 | eaddr[5]);
2692	CSR_WRITE_4(sc, ALE_PAR1, eaddr[0] << 8 | eaddr[1]);
2693	/*
2694	 * Clear WOL status and disable all WOL feature as WOL
2695	 * would interfere Rx operation under normal environments.
2696	 */
2697	CSR_READ_4(sc, ALE_WOL_CFG);
2698	CSR_WRITE_4(sc, ALE_WOL_CFG, 0);
2699	/*
2700	 * Set Tx descriptor/RXF0/CMB base addresses. They share
2701	 * the same high address part of DMAable region.
2702	 */
2703	paddr = sc->ale_cdata.ale_tx_ring_paddr;
2704	CSR_WRITE_4(sc, ALE_TPD_ADDR_HI, ALE_ADDR_HI(paddr));
2705	CSR_WRITE_4(sc, ALE_TPD_ADDR_LO, ALE_ADDR_LO(paddr));
2706	CSR_WRITE_4(sc, ALE_TPD_CNT,
2707	    (ALE_TX_RING_CNT << TPD_CNT_SHIFT) & TPD_CNT_MASK);
2708	/* Set Rx page base address, note we use single queue. */
2709	paddr = sc->ale_cdata.ale_rx_page[0].page_paddr;
2710	CSR_WRITE_4(sc, ALE_RXF0_PAGE0_ADDR_LO, ALE_ADDR_LO(paddr));
2711	paddr = sc->ale_cdata.ale_rx_page[1].page_paddr;
2712	CSR_WRITE_4(sc, ALE_RXF0_PAGE1_ADDR_LO, ALE_ADDR_LO(paddr));
2713	/* Set Tx/Rx CMB addresses. */
2714	paddr = sc->ale_cdata.ale_tx_cmb_paddr;
2715	CSR_WRITE_4(sc, ALE_TX_CMB_ADDR_LO, ALE_ADDR_LO(paddr));
2716	paddr = sc->ale_cdata.ale_rx_page[0].cmb_paddr;
2717	CSR_WRITE_4(sc, ALE_RXF0_CMB0_ADDR_LO, ALE_ADDR_LO(paddr));
2718	paddr = sc->ale_cdata.ale_rx_page[1].cmb_paddr;
2719	CSR_WRITE_4(sc, ALE_RXF0_CMB1_ADDR_LO, ALE_ADDR_LO(paddr));
2720	/* Mark RXF0 is valid. */
2721	CSR_WRITE_1(sc, ALE_RXF0_PAGE0, RXF_VALID);
2722	CSR_WRITE_1(sc, ALE_RXF0_PAGE1, RXF_VALID);
2723	/*
2724	 * No need to initialize RFX1/RXF2/RXF3. We don't use
2725	 * multi-queue yet.
2726	 */
2727
2728	/* Set Rx page size, excluding guard frame size. */
2729	CSR_WRITE_4(sc, ALE_RXF_PAGE_SIZE, ALE_RX_PAGE_SZ);
2730	/* Tell hardware that we're ready to load DMA blocks. */
2731	CSR_WRITE_4(sc, ALE_DMA_BLOCK, DMA_BLOCK_LOAD);
2732
2733	/* Set Rx/Tx interrupt trigger threshold. */
2734	CSR_WRITE_4(sc, ALE_INT_TRIG_THRESH, (1 << INT_TRIG_RX_THRESH_SHIFT) |
2735	    (4 << INT_TRIG_TX_THRESH_SHIFT));
2736	/*
2737	 * XXX
2738	 * Set interrupt trigger timer, its purpose and relation
2739	 * with interrupt moderation mechanism is not clear yet.
2740	 */
2741	CSR_WRITE_4(sc, ALE_INT_TRIG_TIMER,
2742	    ((ALE_USECS(10) << INT_TRIG_RX_TIMER_SHIFT) |
2743	    (ALE_USECS(1000) << INT_TRIG_TX_TIMER_SHIFT)));
2744
2745	/* Configure interrupt moderation timer. */
2746	reg = ALE_USECS(sc->ale_int_rx_mod) << IM_TIMER_RX_SHIFT;
2747	reg |= ALE_USECS(sc->ale_int_tx_mod) << IM_TIMER_TX_SHIFT;
2748	CSR_WRITE_4(sc, ALE_IM_TIMER, reg);
2749	reg = CSR_READ_4(sc, ALE_MASTER_CFG);
2750	reg &= ~(MASTER_CHIP_REV_MASK | MASTER_CHIP_ID_MASK);
2751	reg &= ~(MASTER_IM_RX_TIMER_ENB | MASTER_IM_TX_TIMER_ENB);
2752	if (ALE_USECS(sc->ale_int_rx_mod) != 0)
2753		reg |= MASTER_IM_RX_TIMER_ENB;
2754	if (ALE_USECS(sc->ale_int_tx_mod) != 0)
2755		reg |= MASTER_IM_TX_TIMER_ENB;
2756	CSR_WRITE_4(sc, ALE_MASTER_CFG, reg);
2757	CSR_WRITE_2(sc, ALE_INTR_CLR_TIMER, ALE_USECS(1000));
2758
2759	/* Set Maximum frame size of controller. */
2760	if (ifp->if_mtu < ETHERMTU)
2761		sc->ale_max_frame_size = ETHERMTU;
2762	else
2763		sc->ale_max_frame_size = ifp->if_mtu;
2764	sc->ale_max_frame_size += ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN +
2765	    ETHER_CRC_LEN;
2766	CSR_WRITE_4(sc, ALE_FRAME_SIZE, sc->ale_max_frame_size);
2767	/* Configure IPG/IFG parameters. */
2768	CSR_WRITE_4(sc, ALE_IPG_IFG_CFG,
2769	    ((IPG_IFG_IPGT_DEFAULT << IPG_IFG_IPGT_SHIFT) & IPG_IFG_IPGT_MASK) |
2770	    ((IPG_IFG_MIFG_DEFAULT << IPG_IFG_MIFG_SHIFT) & IPG_IFG_MIFG_MASK) |
2771	    ((IPG_IFG_IPG1_DEFAULT << IPG_IFG_IPG1_SHIFT) & IPG_IFG_IPG1_MASK) |
2772	    ((IPG_IFG_IPG2_DEFAULT << IPG_IFG_IPG2_SHIFT) & IPG_IFG_IPG2_MASK));
2773	/* Set parameters for half-duplex media. */
2774	CSR_WRITE_4(sc, ALE_HDPX_CFG,
2775	    ((HDPX_CFG_LCOL_DEFAULT << HDPX_CFG_LCOL_SHIFT) &
2776	    HDPX_CFG_LCOL_MASK) |
2777	    ((HDPX_CFG_RETRY_DEFAULT << HDPX_CFG_RETRY_SHIFT) &
2778	    HDPX_CFG_RETRY_MASK) | HDPX_CFG_EXC_DEF_EN |
2779	    ((HDPX_CFG_ABEBT_DEFAULT << HDPX_CFG_ABEBT_SHIFT) &
2780	    HDPX_CFG_ABEBT_MASK) |
2781	    ((HDPX_CFG_JAMIPG_DEFAULT << HDPX_CFG_JAMIPG_SHIFT) &
2782	    HDPX_CFG_JAMIPG_MASK));
2783
2784	/* Configure Tx jumbo frame parameters. */
2785	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0) {
2786		if (ifp->if_mtu < ETHERMTU)
2787			reg = sc->ale_max_frame_size;
2788		else if (ifp->if_mtu < 6 * 1024)
2789			reg = (sc->ale_max_frame_size * 2) / 3;
2790		else
2791			reg = sc->ale_max_frame_size / 2;
2792		CSR_WRITE_4(sc, ALE_TX_JUMBO_THRESH,
2793		    roundup(reg, TX_JUMBO_THRESH_UNIT) >>
2794		    TX_JUMBO_THRESH_UNIT_SHIFT);
2795	}
2796	/* Configure TxQ. */
2797	reg = (128 << (sc->ale_dma_rd_burst >> DMA_CFG_RD_BURST_SHIFT))
2798	    << TXQ_CFG_TX_FIFO_BURST_SHIFT;
2799	reg |= (TXQ_CFG_TPD_BURST_DEFAULT << TXQ_CFG_TPD_BURST_SHIFT) &
2800	    TXQ_CFG_TPD_BURST_MASK;
2801	CSR_WRITE_4(sc, ALE_TXQ_CFG, reg | TXQ_CFG_ENHANCED_MODE | TXQ_CFG_ENB);
2802
2803	/* Configure Rx jumbo frame & flow control parameters. */
2804	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0) {
2805		reg = roundup(sc->ale_max_frame_size, RX_JUMBO_THRESH_UNIT);
2806		CSR_WRITE_4(sc, ALE_RX_JUMBO_THRESH,
2807		    (((reg >> RX_JUMBO_THRESH_UNIT_SHIFT) <<
2808		    RX_JUMBO_THRESH_MASK_SHIFT) & RX_JUMBO_THRESH_MASK) |
2809		    ((RX_JUMBO_LKAH_DEFAULT << RX_JUMBO_LKAH_SHIFT) &
2810		    RX_JUMBO_LKAH_MASK));
2811		reg = CSR_READ_4(sc, ALE_SRAM_RX_FIFO_LEN);
2812		rxf_hi = (reg * 7) / 10;
2813		rxf_lo = (reg * 3)/ 10;
2814		CSR_WRITE_4(sc, ALE_RX_FIFO_PAUSE_THRESH,
2815		    ((rxf_lo << RX_FIFO_PAUSE_THRESH_LO_SHIFT) &
2816		    RX_FIFO_PAUSE_THRESH_LO_MASK) |
2817		    ((rxf_hi << RX_FIFO_PAUSE_THRESH_HI_SHIFT) &
2818		     RX_FIFO_PAUSE_THRESH_HI_MASK));
2819	}
2820
2821	/* Disable RSS. */
2822	CSR_WRITE_4(sc, ALE_RSS_IDT_TABLE0, 0);
2823	CSR_WRITE_4(sc, ALE_RSS_CPU, 0);
2824
2825	/* Configure RxQ. */
2826	CSR_WRITE_4(sc, ALE_RXQ_CFG,
2827	    RXQ_CFG_ALIGN_32 | RXQ_CFG_CUT_THROUGH_ENB | RXQ_CFG_ENB);
2828
2829	/* Configure DMA parameters. */
2830	reg = 0;
2831	if ((sc->ale_flags & ALE_FLAG_TXCMB_BUG) == 0)
2832		reg |= DMA_CFG_TXCMB_ENB;
2833	CSR_WRITE_4(sc, ALE_DMA_CFG,
2834	    DMA_CFG_OUT_ORDER | DMA_CFG_RD_REQ_PRI | DMA_CFG_RCB_64 |
2835	    sc->ale_dma_rd_burst | reg |
2836	    sc->ale_dma_wr_burst | DMA_CFG_RXCMB_ENB |
2837	    ((DMA_CFG_RD_DELAY_CNT_DEFAULT << DMA_CFG_RD_DELAY_CNT_SHIFT) &
2838	    DMA_CFG_RD_DELAY_CNT_MASK) |
2839	    ((DMA_CFG_WR_DELAY_CNT_DEFAULT << DMA_CFG_WR_DELAY_CNT_SHIFT) &
2840	    DMA_CFG_WR_DELAY_CNT_MASK));
2841
2842	/*
2843	 * Hardware can be configured to issue SMB interrupt based
2844	 * on programmed interval. Since there is a callout that is
2845	 * invoked for every hz in driver we use that instead of
2846	 * relying on periodic SMB interrupt.
2847	 */
2848	CSR_WRITE_4(sc, ALE_SMB_STAT_TIMER, ALE_USECS(0));
2849	/* Clear MAC statistics. */
2850	ale_stats_clear(sc);
2851
2852	/*
2853	 * Configure Tx/Rx MACs.
2854	 *  - Auto-padding for short frames.
2855	 *  - Enable CRC generation.
2856	 *  Actual reconfiguration of MAC for resolved speed/duplex
2857	 *  is followed after detection of link establishment.
2858	 *  AR81xx always does checksum computation regardless of
2859	 *  MAC_CFG_RXCSUM_ENB bit. In fact, setting the bit will
2860	 *  cause Rx handling issue for fragmented IP datagrams due
2861	 *  to silicon bug.
2862	 */
2863	reg = MAC_CFG_TX_CRC_ENB | MAC_CFG_TX_AUTO_PAD | MAC_CFG_FULL_DUPLEX |
2864	    ((MAC_CFG_PREAMBLE_DEFAULT << MAC_CFG_PREAMBLE_SHIFT) &
2865	    MAC_CFG_PREAMBLE_MASK);
2866	if ((sc->ale_flags & ALE_FLAG_FASTETHER) != 0)
2867		reg |= MAC_CFG_SPEED_10_100;
2868	else
2869		reg |= MAC_CFG_SPEED_1000;
2870	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2871
2872	/* Set up the receive filter. */
2873	ale_rxfilter(sc);
2874	ale_rxvlan(sc);
2875
2876	/* Acknowledge all pending interrupts and clear it. */
2877	CSR_WRITE_4(sc, ALE_INTR_MASK, ALE_INTRS);
2878	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0xFFFFFFFF);
2879	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0);
2880
2881	sc->ale_flags &= ~ALE_FLAG_LINK;
2882	/* Switch to the current media. */
2883	mii_mediachg(mii);
2884
2885	callout_reset(&sc->ale_tick_ch, hz, ale_tick, sc);
2886
2887	ifp->if_drv_flags |= IFF_DRV_RUNNING;
2888	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2889}
2890
2891static void
2892ale_stop(struct ale_softc *sc)
2893{
2894	struct ifnet *ifp;
2895	struct ale_txdesc *txd;
2896	uint32_t reg;
2897	int i;
2898
2899	ALE_LOCK_ASSERT(sc);
2900	/*
2901	 * Mark the interface down and cancel the watchdog timer.
2902	 */
2903	ifp = sc->ale_ifp;
2904	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
2905	sc->ale_flags &= ~ALE_FLAG_LINK;
2906	callout_stop(&sc->ale_tick_ch);
2907	sc->ale_watchdog_timer = 0;
2908	ale_stats_update(sc);
2909	/* Disable interrupts. */
2910	CSR_WRITE_4(sc, ALE_INTR_MASK, 0);
2911	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0xFFFFFFFF);
2912	/* Disable queue processing and DMA. */
2913	reg = CSR_READ_4(sc, ALE_TXQ_CFG);
2914	reg &= ~TXQ_CFG_ENB;
2915	CSR_WRITE_4(sc, ALE_TXQ_CFG, reg);
2916	reg = CSR_READ_4(sc, ALE_RXQ_CFG);
2917	reg &= ~RXQ_CFG_ENB;
2918	CSR_WRITE_4(sc, ALE_RXQ_CFG, reg);
2919	reg = CSR_READ_4(sc, ALE_DMA_CFG);
2920	reg &= ~(DMA_CFG_TXCMB_ENB | DMA_CFG_RXCMB_ENB);
2921	CSR_WRITE_4(sc, ALE_DMA_CFG, reg);
2922	DELAY(1000);
2923	/* Stop Rx/Tx MACs. */
2924	ale_stop_mac(sc);
2925	/* Disable interrupts which might be touched in taskq handler. */
2926	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0xFFFFFFFF);
2927
2928	/*
2929	 * Free TX mbufs still in the queues.
2930	 */
2931	for (i = 0; i < ALE_TX_RING_CNT; i++) {
2932		txd = &sc->ale_cdata.ale_txdesc[i];
2933		if (txd->tx_m != NULL) {
2934			bus_dmamap_sync(sc->ale_cdata.ale_tx_tag,
2935			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2936			bus_dmamap_unload(sc->ale_cdata.ale_tx_tag,
2937			    txd->tx_dmamap);
2938			m_freem(txd->tx_m);
2939			txd->tx_m = NULL;
2940		}
2941        }
2942}
2943
2944static void
2945ale_stop_mac(struct ale_softc *sc)
2946{
2947	uint32_t reg;
2948	int i;
2949
2950	ALE_LOCK_ASSERT(sc);
2951
2952	reg = CSR_READ_4(sc, ALE_MAC_CFG);
2953	if ((reg & (MAC_CFG_TX_ENB | MAC_CFG_RX_ENB)) != 0) {
2954		reg &= ~MAC_CFG_TX_ENB | MAC_CFG_RX_ENB;
2955		CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2956	}
2957
2958	for (i = ALE_TIMEOUT; i > 0; i--) {
2959		reg = CSR_READ_4(sc, ALE_IDLE_STATUS);
2960		if (reg == 0)
2961			break;
2962		DELAY(10);
2963	}
2964	if (i == 0)
2965		device_printf(sc->ale_dev,
2966		    "could not disable Tx/Rx MAC(0x%08x)!\n", reg);
2967}
2968
2969static void
2970ale_init_tx_ring(struct ale_softc *sc)
2971{
2972	struct ale_txdesc *txd;
2973	int i;
2974
2975	ALE_LOCK_ASSERT(sc);
2976
2977	sc->ale_cdata.ale_tx_prod = 0;
2978	sc->ale_cdata.ale_tx_cons = 0;
2979	sc->ale_cdata.ale_tx_cnt = 0;
2980
2981	bzero(sc->ale_cdata.ale_tx_ring, ALE_TX_RING_SZ);
2982	bzero(sc->ale_cdata.ale_tx_cmb, ALE_TX_CMB_SZ);
2983	for (i = 0; i < ALE_TX_RING_CNT; i++) {
2984		txd = &sc->ale_cdata.ale_txdesc[i];
2985		txd->tx_m = NULL;
2986	}
2987	*sc->ale_cdata.ale_tx_cmb = 0;
2988	bus_dmamap_sync(sc->ale_cdata.ale_tx_cmb_tag,
2989	    sc->ale_cdata.ale_tx_cmb_map,
2990	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2991	bus_dmamap_sync(sc->ale_cdata.ale_tx_ring_tag,
2992	    sc->ale_cdata.ale_tx_ring_map,
2993	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2994}
2995
2996static void
2997ale_init_rx_pages(struct ale_softc *sc)
2998{
2999	struct ale_rx_page *rx_page;
3000	int i;
3001
3002	ALE_LOCK_ASSERT(sc);
3003
3004	sc->ale_morework = 0;
3005	sc->ale_cdata.ale_rx_seqno = 0;
3006	sc->ale_cdata.ale_rx_curp = 0;
3007
3008	for (i = 0; i < ALE_RX_PAGES; i++) {
3009		rx_page = &sc->ale_cdata.ale_rx_page[i];
3010		bzero(rx_page->page_addr, sc->ale_pagesize);
3011		bzero(rx_page->cmb_addr, ALE_RX_CMB_SZ);
3012		rx_page->cons = 0;
3013		*rx_page->cmb_addr = 0;
3014		bus_dmamap_sync(rx_page->page_tag, rx_page->page_map,
3015		    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
3016		bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
3017		    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
3018	}
3019}
3020
3021static void
3022ale_rxvlan(struct ale_softc *sc)
3023{
3024	struct ifnet *ifp;
3025	uint32_t reg;
3026
3027	ALE_LOCK_ASSERT(sc);
3028
3029	ifp = sc->ale_ifp;
3030	reg = CSR_READ_4(sc, ALE_MAC_CFG);
3031	reg &= ~MAC_CFG_VLAN_TAG_STRIP;
3032	if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
3033		reg |= MAC_CFG_VLAN_TAG_STRIP;
3034	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
3035}
3036
3037static void
3038ale_rxfilter(struct ale_softc *sc)
3039{
3040	struct ifnet *ifp;
3041	struct ifmultiaddr *ifma;
3042	uint32_t crc;
3043	uint32_t mchash[2];
3044	uint32_t rxcfg;
3045
3046	ALE_LOCK_ASSERT(sc);
3047
3048	ifp = sc->ale_ifp;
3049
3050	rxcfg = CSR_READ_4(sc, ALE_MAC_CFG);
3051	rxcfg &= ~(MAC_CFG_ALLMULTI | MAC_CFG_BCAST | MAC_CFG_PROMISC);
3052	if ((ifp->if_flags & IFF_BROADCAST) != 0)
3053		rxcfg |= MAC_CFG_BCAST;
3054	if ((ifp->if_flags & (IFF_PROMISC | IFF_ALLMULTI)) != 0) {
3055		if ((ifp->if_flags & IFF_PROMISC) != 0)
3056			rxcfg |= MAC_CFG_PROMISC;
3057		if ((ifp->if_flags & IFF_ALLMULTI) != 0)
3058			rxcfg |= MAC_CFG_ALLMULTI;
3059		CSR_WRITE_4(sc, ALE_MAR0, 0xFFFFFFFF);
3060		CSR_WRITE_4(sc, ALE_MAR1, 0xFFFFFFFF);
3061		CSR_WRITE_4(sc, ALE_MAC_CFG, rxcfg);
3062		return;
3063	}
3064
3065	/* Program new filter. */
3066	bzero(mchash, sizeof(mchash));
3067
3068	if_maddr_rlock(ifp);
3069	TAILQ_FOREACH(ifma, &sc->ale_ifp->if_multiaddrs, ifma_link) {
3070		if (ifma->ifma_addr->sa_family != AF_LINK)
3071			continue;
3072		crc = ether_crc32_be(LLADDR((struct sockaddr_dl *)
3073		    ifma->ifma_addr), ETHER_ADDR_LEN);
3074		mchash[crc >> 31] |= 1 << ((crc >> 26) & 0x1f);
3075	}
3076	if_maddr_runlock(ifp);
3077
3078	CSR_WRITE_4(sc, ALE_MAR0, mchash[0]);
3079	CSR_WRITE_4(sc, ALE_MAR1, mchash[1]);
3080	CSR_WRITE_4(sc, ALE_MAC_CFG, rxcfg);
3081}
3082
3083static int
3084sysctl_int_range(SYSCTL_HANDLER_ARGS, int low, int high)
3085{
3086	int error, value;
3087
3088	if (arg1 == NULL)
3089		return (EINVAL);
3090	value = *(int *)arg1;
3091	error = sysctl_handle_int(oidp, &value, 0, req);
3092	if (error || req->newptr == NULL)
3093		return (error);
3094	if (value < low || value > high)
3095		return (EINVAL);
3096        *(int *)arg1 = value;
3097
3098        return (0);
3099}
3100
3101static int
3102sysctl_hw_ale_proc_limit(SYSCTL_HANDLER_ARGS)
3103{
3104	return (sysctl_int_range(oidp, arg1, arg2, req,
3105	    ALE_PROC_MIN, ALE_PROC_MAX));
3106}
3107
3108static int
3109sysctl_hw_ale_int_mod(SYSCTL_HANDLER_ARGS)
3110{
3111
3112	return (sysctl_int_range(oidp, arg1, arg2, req,
3113	    ALE_IM_TIMER_MIN, ALE_IM_TIMER_MAX));
3114}
3115