1/*	$NetBSD: if_ale.c,v 1.43 2022/08/22 15:59:42 thorpej Exp $	*/
2
3/*-
4 * Copyright (c) 2008, Pyun YongHyeon <yongari@FreeBSD.org>
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 *    notice unmodified, this list of conditions, and the following
12 *    disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 *    notice, this list of conditions and the following disclaimer in the
15 *    documentation and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 * $FreeBSD: src/sys/dev/ale/if_ale.c,v 1.3 2008/12/03 09:01:12 yongari Exp $
30 */
31
32/* Driver for Atheros AR8121/AR8113/AR8114 PCIe Ethernet. */
33
34#include <sys/cdefs.h>
35__KERNEL_RCSID(0, "$NetBSD: if_ale.c,v 1.43 2022/08/22 15:59:42 thorpej Exp $");
36
37#include "vlan.h"
38
39#include <sys/param.h>
40#include <sys/proc.h>
41#include <sys/endian.h>
42#include <sys/systm.h>
43#include <sys/types.h>
44#include <sys/sockio.h>
45#include <sys/mbuf.h>
46#include <sys/queue.h>
47#include <sys/kernel.h>
48#include <sys/device.h>
49#include <sys/callout.h>
50#include <sys/socket.h>
51
52#include <sys/bus.h>
53
54#include <net/if.h>
55#include <net/if_dl.h>
56#include <net/if_llc.h>
57#include <net/if_media.h>
58#include <net/if_ether.h>
59
60#ifdef INET
61#include <netinet/in.h>
62#include <netinet/in_systm.h>
63#include <netinet/in_var.h>
64#include <netinet/ip.h>
65#endif
66
67#include <net/if_types.h>
68#include <net/if_vlanvar.h>
69
70#include <net/bpf.h>
71
72#include <dev/mii/mii.h>
73#include <dev/mii/miivar.h>
74
75#include <dev/pci/pcireg.h>
76#include <dev/pci/pcivar.h>
77#include <dev/pci/pcidevs.h>
78
79#include <dev/pci/if_alereg.h>
80
81static int	ale_match(device_t, cfdata_t, void *);
82static void	ale_attach(device_t, device_t, void *);
83static int	ale_detach(device_t, int);
84
85static int	ale_miibus_readreg(device_t, int, int, uint16_t *);
86static int	ale_miibus_writereg(device_t, int, int, uint16_t);
87static void	ale_miibus_statchg(struct ifnet *);
88
89static int	ale_init(struct ifnet *);
90static void	ale_start(struct ifnet *);
91static int	ale_ioctl(struct ifnet *, u_long, void *);
92static void	ale_watchdog(struct ifnet *);
93static int	ale_mediachange(struct ifnet *);
94static void	ale_mediastatus(struct ifnet *, struct ifmediareq *);
95
96static int	ale_intr(void *);
97static int	ale_rxeof(struct ale_softc *sc);
98static void	ale_rx_update_page(struct ale_softc *, struct ale_rx_page **,
99		    uint32_t, uint32_t *);
100static void	ale_rxcsum(struct ale_softc *, struct mbuf *, uint32_t);
101static void	ale_txeof(struct ale_softc *);
102
103static int	ale_dma_alloc(struct ale_softc *);
104static void	ale_dma_free(struct ale_softc *);
105static int	ale_encap(struct ale_softc *, struct mbuf *);
106static void	ale_init_rx_pages(struct ale_softc *);
107static void	ale_init_tx_ring(struct ale_softc *);
108
109static void	ale_stop(struct ifnet *, int);
110static void	ale_tick(void *);
111static void	ale_get_macaddr(struct ale_softc *);
112static void	ale_mac_config(struct ale_softc *);
113static void	ale_phy_reset(struct ale_softc *);
114static void	ale_reset(struct ale_softc *);
115static void	ale_rxfilter(struct ale_softc *);
116static void	ale_rxvlan(struct ale_softc *);
117static void	ale_stats_clear(struct ale_softc *);
118static void	ale_stats_update(struct ale_softc *);
119static void	ale_stop_mac(struct ale_softc *);
120
121CFATTACH_DECL_NEW(ale, sizeof(struct ale_softc),
122	ale_match, ale_attach, ale_detach, NULL);
123
124int aledebug = 0;
125#define DPRINTF(x)	do { if (aledebug) printf x; } while (0)
126
127#define ALE_CSUM_FEATURES	(M_CSUM_TCPv4 | M_CSUM_UDPv4)
128
129static int
130ale_miibus_readreg(device_t dev, int phy, int reg, uint16_t *val)
131{
132	struct ale_softc *sc = device_private(dev);
133	uint32_t v;
134	int i;
135
136	if (phy != sc->ale_phyaddr)
137		return -1;
138
139	if (sc->ale_flags & ALE_FLAG_FASTETHER) {
140		switch (reg) {
141		case MII_100T2CR:
142		case MII_100T2SR:
143		case MII_EXTSR:
144			*val = 0;
145			return 0;
146		default:
147			break;
148		}
149	}
150
151	CSR_WRITE_4(sc, ALE_MDIO, MDIO_OP_EXECUTE | MDIO_OP_READ |
152	    MDIO_SUP_PREAMBLE | MDIO_CLK_25_4 | MDIO_REG_ADDR(reg));
153	for (i = ALE_PHY_TIMEOUT; i > 0; i--) {
154		DELAY(5);
155		v = CSR_READ_4(sc, ALE_MDIO);
156		if ((v & (MDIO_OP_EXECUTE | MDIO_OP_BUSY)) == 0)
157			break;
158	}
159
160	if (i == 0) {
161		printf("%s: phy read timeout: phy %d, reg %d\n",
162		    device_xname(sc->sc_dev), phy, reg);
163		return ETIMEDOUT;
164	}
165
166	*val = (v & MDIO_DATA_MASK) >> MDIO_DATA_SHIFT;
167	return 0;
168}
169
170static int
171ale_miibus_writereg(device_t dev, int phy, int reg, uint16_t val)
172{
173	struct ale_softc *sc = device_private(dev);
174	uint32_t v;
175	int i;
176
177	if (phy != sc->ale_phyaddr)
178		return -1;
179
180	if (sc->ale_flags & ALE_FLAG_FASTETHER) {
181		switch (reg) {
182		case MII_100T2CR:
183		case MII_100T2SR:
184		case MII_EXTSR:
185			return 0;
186		default:
187			break;
188		}
189	}
190
191	CSR_WRITE_4(sc, ALE_MDIO, MDIO_OP_EXECUTE | MDIO_OP_WRITE |
192	    (val & MDIO_DATA_MASK) << MDIO_DATA_SHIFT |
193	    MDIO_SUP_PREAMBLE | MDIO_CLK_25_4 | MDIO_REG_ADDR(reg));
194	for (i = ALE_PHY_TIMEOUT; i > 0; i--) {
195		DELAY(5);
196		v = CSR_READ_4(sc, ALE_MDIO);
197		if ((v & (MDIO_OP_EXECUTE | MDIO_OP_BUSY)) == 0)
198			break;
199	}
200
201	if (i == 0) {
202		printf("%s: phy write timeout: phy %d, reg %d\n",
203		    device_xname(sc->sc_dev), phy, reg);
204		return ETIMEDOUT;
205	}
206
207	return 0;
208}
209
210static void
211ale_miibus_statchg(struct ifnet *ifp)
212{
213	struct ale_softc *sc = ifp->if_softc;
214	struct mii_data *mii = &sc->sc_miibus;
215	uint32_t reg;
216
217	if ((ifp->if_flags & IFF_RUNNING) == 0)
218		return;
219
220	sc->ale_flags &= ~ALE_FLAG_LINK;
221	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
222	    (IFM_ACTIVE | IFM_AVALID)) {
223		switch (IFM_SUBTYPE(mii->mii_media_active)) {
224		case IFM_10_T:
225		case IFM_100_TX:
226			sc->ale_flags |= ALE_FLAG_LINK;
227			break;
228
229		case IFM_1000_T:
230			if ((sc->ale_flags & ALE_FLAG_FASTETHER) == 0)
231				sc->ale_flags |= ALE_FLAG_LINK;
232			break;
233
234		default:
235			break;
236		}
237	}
238
239	/* Stop Rx/Tx MACs. */
240	ale_stop_mac(sc);
241
242	/* Program MACs with resolved speed/duplex/flow-control. */
243	if ((sc->ale_flags & ALE_FLAG_LINK) != 0) {
244		ale_mac_config(sc);
245		/* Reenable Tx/Rx MACs. */
246		reg = CSR_READ_4(sc, ALE_MAC_CFG);
247		reg |= MAC_CFG_TX_ENB | MAC_CFG_RX_ENB;
248		CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
249	}
250}
251
252void
253ale_mediastatus(struct ifnet *ifp, struct ifmediareq *ifmr)
254{
255	struct ale_softc *sc = ifp->if_softc;
256	struct mii_data *mii = &sc->sc_miibus;
257
258	mii_pollstat(mii);
259	ifmr->ifm_status = mii->mii_media_status;
260	ifmr->ifm_active = mii->mii_media_active;
261}
262
263int
264ale_mediachange(struct ifnet *ifp)
265{
266	struct ale_softc *sc = ifp->if_softc;
267	struct mii_data *mii = &sc->sc_miibus;
268	int error;
269
270	if (mii->mii_instance != 0) {
271		struct mii_softc *miisc;
272
273		LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
274			mii_phy_reset(miisc);
275	}
276	error = mii_mediachg(mii);
277
278	return error;
279}
280
281int
282ale_match(device_t dev, cfdata_t match, void *aux)
283{
284	struct pci_attach_args *pa = aux;
285
286	return (PCI_VENDOR(pa->pa_id) == PCI_VENDOR_ATTANSIC &&
287	    PCI_PRODUCT(pa->pa_id) == PCI_PRODUCT_ATTANSIC_ETHERNET_L1E);
288}
289
290void
291ale_get_macaddr(struct ale_softc *sc)
292{
293	uint32_t ea[2], reg;
294	int i, vpdc;
295
296	reg = CSR_READ_4(sc, ALE_SPI_CTRL);
297	if ((reg & SPI_VPD_ENB) != 0) {
298		reg &= ~SPI_VPD_ENB;
299		CSR_WRITE_4(sc, ALE_SPI_CTRL, reg);
300	}
301
302	if (pci_get_capability(sc->sc_pct, sc->sc_pcitag, PCI_CAP_VPD,
303	    &vpdc, NULL)) {
304		/*
305		 * PCI VPD capability found, let TWSI reload EEPROM.
306		 * This will set ethernet address of controller.
307		 */
308		CSR_WRITE_4(sc, ALE_TWSI_CTRL, CSR_READ_4(sc, ALE_TWSI_CTRL) |
309		    TWSI_CTRL_SW_LD_START);
310		for (i = 100; i > 0; i--) {
311			DELAY(1000);
312			reg = CSR_READ_4(sc, ALE_TWSI_CTRL);
313			if ((reg & TWSI_CTRL_SW_LD_START) == 0)
314				break;
315		}
316		if (i == 0)
317			printf("%s: reloading EEPROM timeout!\n",
318			    device_xname(sc->sc_dev));
319	} else {
320		if (aledebug)
321			printf("%s: PCI VPD capability not found!\n",
322			    device_xname(sc->sc_dev));
323	}
324
325	ea[0] = CSR_READ_4(sc, ALE_PAR0);
326	ea[1] = CSR_READ_4(sc, ALE_PAR1);
327	sc->ale_eaddr[0] = (ea[1] >> 8) & 0xFF;
328	sc->ale_eaddr[1] = (ea[1] >> 0) & 0xFF;
329	sc->ale_eaddr[2] = (ea[0] >> 24) & 0xFF;
330	sc->ale_eaddr[3] = (ea[0] >> 16) & 0xFF;
331	sc->ale_eaddr[4] = (ea[0] >> 8) & 0xFF;
332	sc->ale_eaddr[5] = (ea[0] >> 0) & 0xFF;
333}
334
335void
336ale_phy_reset(struct ale_softc *sc)
337{
338	/* Reset magic from Linux. */
339	CSR_WRITE_2(sc, ALE_GPHY_CTRL,
340	    GPHY_CTRL_HIB_EN | GPHY_CTRL_HIB_PULSE | GPHY_CTRL_SEL_ANA_RESET |
341	    GPHY_CTRL_PHY_PLL_ON);
342	DELAY(1000);
343	CSR_WRITE_2(sc, ALE_GPHY_CTRL,
344	    GPHY_CTRL_EXT_RESET | GPHY_CTRL_HIB_EN | GPHY_CTRL_HIB_PULSE |
345	    GPHY_CTRL_SEL_ANA_RESET | GPHY_CTRL_PHY_PLL_ON);
346	DELAY(1000);
347
348#define	ATPHY_DBG_ADDR		0x1D
349#define	ATPHY_DBG_DATA		0x1E
350
351	/* Enable hibernation mode. */
352	ale_miibus_writereg(sc->sc_dev, sc->ale_phyaddr,
353	    ATPHY_DBG_ADDR, 0x0B);
354	ale_miibus_writereg(sc->sc_dev, sc->ale_phyaddr,
355	    ATPHY_DBG_DATA, 0xBC00);
356	/* Set Class A/B for all modes. */
357	ale_miibus_writereg(sc->sc_dev, sc->ale_phyaddr,
358	    ATPHY_DBG_ADDR, 0x00);
359	ale_miibus_writereg(sc->sc_dev, sc->ale_phyaddr,
360	    ATPHY_DBG_DATA, 0x02EF);
361	/* Enable 10BT power saving. */
362	ale_miibus_writereg(sc->sc_dev, sc->ale_phyaddr,
363	    ATPHY_DBG_ADDR, 0x12);
364	ale_miibus_writereg(sc->sc_dev, sc->ale_phyaddr,
365	    ATPHY_DBG_DATA, 0x4C04);
366	/* Adjust 1000T power. */
367	ale_miibus_writereg(sc->sc_dev, sc->ale_phyaddr,
368	    ATPHY_DBG_ADDR, 0x04);
369	ale_miibus_writereg(sc->sc_dev, sc->ale_phyaddr,
370	    ATPHY_DBG_DATA, 0x8BBB);
371	/* 10BT center tap voltage. */
372	ale_miibus_writereg(sc->sc_dev, sc->ale_phyaddr,
373	    ATPHY_DBG_ADDR, 0x05);
374	ale_miibus_writereg(sc->sc_dev, sc->ale_phyaddr,
375	    ATPHY_DBG_DATA, 0x2C46);
376
377#undef	ATPHY_DBG_ADDR
378#undef	ATPHY_DBG_DATA
379	DELAY(1000);
380}
381
382void
383ale_attach(device_t parent, device_t self, void *aux)
384{
385	struct ale_softc *sc = device_private(self);
386	struct pci_attach_args *pa = aux;
387	pci_chipset_tag_t pc = pa->pa_pc;
388	pci_intr_handle_t ih;
389	const char *intrstr;
390	struct ifnet *ifp;
391	struct mii_data * const mii = &sc->sc_miibus;
392	pcireg_t memtype;
393	int mii_flags, error = 0;
394	uint32_t rxf_len, txf_len;
395	const char *chipname;
396	char intrbuf[PCI_INTRSTR_LEN];
397
398	aprint_naive("\n");
399	aprint_normal(": Attansic/Atheros L1E Ethernet\n");
400
401	sc->sc_dev = self;
402	sc->sc_dmat = pa->pa_dmat;
403	sc->sc_pct = pa->pa_pc;
404	sc->sc_pcitag = pa->pa_tag;
405
406	/*
407	 * Allocate IO memory
408	 */
409	memtype = pci_mapreg_type(sc->sc_pct, sc->sc_pcitag, ALE_PCIR_BAR);
410	switch (memtype) {
411	case PCI_MAPREG_TYPE_MEM | PCI_MAPREG_MEM_TYPE_32BIT:
412	case PCI_MAPREG_TYPE_MEM | PCI_MAPREG_MEM_TYPE_32BIT_1M:
413	case PCI_MAPREG_TYPE_MEM | PCI_MAPREG_MEM_TYPE_64BIT:
414		break;
415	default:
416		aprint_error_dev(self, "invalid base address register\n");
417		break;
418	}
419
420	if (pci_mapreg_map(pa, ALE_PCIR_BAR, memtype, 0, &sc->sc_mem_bt,
421	    &sc->sc_mem_bh, NULL, &sc->sc_mem_size)) {
422		aprint_error_dev(self, "could not map mem space\n");
423		return;
424	}
425
426	if (pci_intr_map(pa, &ih) != 0) {
427		aprint_error_dev(self, "could not map interrupt\n");
428		goto fail;
429	}
430
431	/*
432	 * Allocate IRQ
433	 */
434	intrstr = pci_intr_string(sc->sc_pct, ih, intrbuf, sizeof(intrbuf));
435	sc->sc_irq_handle = pci_intr_establish_xname(pc, ih, IPL_NET, ale_intr,
436	    sc, device_xname(self));
437	if (sc->sc_irq_handle == NULL) {
438		aprint_error_dev(self, "could not establish interrupt");
439		if (intrstr != NULL)
440			aprint_error(" at %s", intrstr);
441		aprint_error("\n");
442		goto fail;
443	}
444
445	/* Set PHY address. */
446	sc->ale_phyaddr = ALE_PHY_ADDR;
447
448	/* Reset PHY. */
449	ale_phy_reset(sc);
450
451	/* Reset the ethernet controller. */
452	ale_reset(sc);
453
454	/* Get PCI and chip id/revision. */
455	sc->ale_rev = PCI_REVISION(pa->pa_class);
456	if (sc->ale_rev >= 0xF0) {
457		/* L2E Rev. B. AR8114 */
458		sc->ale_flags |= ALE_FLAG_FASTETHER;
459		chipname = "AR8114 (L2E RevB)";
460	} else {
461		if ((CSR_READ_4(sc, ALE_PHY_STATUS) & PHY_STATUS_100M) != 0) {
462			/* L1E AR8121 */
463			sc->ale_flags |= ALE_FLAG_JUMBO;
464			chipname = "AR8121 (L1E)";
465		} else {
466			/* L2E Rev. A. AR8113 */
467			sc->ale_flags |= ALE_FLAG_FASTETHER;
468			chipname = "AR8113 (L2E RevA)";
469		}
470	}
471	aprint_normal_dev(self, "%s, %s\n", chipname, intrstr);
472
473	/*
474	 * All known controllers seems to require 4 bytes alignment
475	 * of Tx buffers to make Tx checksum offload with custom
476	 * checksum generation method work.
477	 */
478	sc->ale_flags |= ALE_FLAG_TXCSUM_BUG;
479
480	/*
481	 * All known controllers seems to have issues on Rx checksum
482	 * offload for fragmented IP datagrams.
483	 */
484	sc->ale_flags |= ALE_FLAG_RXCSUM_BUG;
485
486	/*
487	 * Don't use Tx CMB. It is known to cause RRS update failure
488	 * under certain circumstances. Typical phenomenon of the
489	 * issue would be unexpected sequence number encountered in
490	 * Rx handler.
491	 */
492	sc->ale_flags |= ALE_FLAG_TXCMB_BUG;
493	sc->ale_chip_rev = CSR_READ_4(sc, ALE_MASTER_CFG) >>
494	    MASTER_CHIP_REV_SHIFT;
495	aprint_debug_dev(self, "PCI device revision : 0x%04x\n", sc->ale_rev);
496	aprint_debug_dev(self, "Chip id/revision : 0x%04x\n", sc->ale_chip_rev);
497
498	/*
499	 * Uninitialized hardware returns an invalid chip id/revision
500	 * as well as 0xFFFFFFFF for Tx/Rx fifo length.
501	 */
502	txf_len = CSR_READ_4(sc, ALE_SRAM_TX_FIFO_LEN);
503	rxf_len = CSR_READ_4(sc, ALE_SRAM_RX_FIFO_LEN);
504	if (sc->ale_chip_rev == 0xFFFF || txf_len == 0xFFFFFFFF ||
505	    rxf_len == 0xFFFFFFF) {
506		aprint_error_dev(self, "chip revision : 0x%04x, %u Tx FIFO "
507		    "%u Rx FIFO -- not initialized?\n",
508		    sc->ale_chip_rev, txf_len, rxf_len);
509		goto fail;
510	}
511
512	if (aledebug) {
513		printf("%s: %u Tx FIFO, %u Rx FIFO\n", device_xname(sc->sc_dev),
514		    txf_len, rxf_len);
515	}
516
517	/* Set max allowable DMA size. */
518	sc->ale_dma_rd_burst = DMA_CFG_RD_BURST_128;
519	sc->ale_dma_wr_burst = DMA_CFG_WR_BURST_128;
520
521	callout_init(&sc->sc_tick_ch, 0);
522	callout_setfunc(&sc->sc_tick_ch, ale_tick, sc);
523
524	error = ale_dma_alloc(sc);
525	if (error)
526		goto fail;
527
528	/* Load station address. */
529	ale_get_macaddr(sc);
530
531	aprint_normal_dev(self, "Ethernet address %s\n",
532	    ether_sprintf(sc->ale_eaddr));
533
534	ifp = &sc->sc_ec.ec_if;
535	ifp->if_softc = sc;
536	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
537	ifp->if_init = ale_init;
538	ifp->if_ioctl = ale_ioctl;
539	ifp->if_start = ale_start;
540	ifp->if_stop = ale_stop;
541	ifp->if_watchdog = ale_watchdog;
542	IFQ_SET_MAXLEN(&ifp->if_snd, ALE_TX_RING_CNT - 1);
543	IFQ_SET_READY(&ifp->if_snd);
544	strlcpy(ifp->if_xname, device_xname(sc->sc_dev), IFNAMSIZ);
545
546	sc->sc_ec.ec_capabilities = ETHERCAP_VLAN_MTU;
547
548#ifdef ALE_CHECKSUM
549	ifp->if_capabilities |= IFCAP_CSUM_IPv4_Tx | IFCAP_CSUM_IPv4_Rx |
550				IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
551				IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx;
552#endif
553
554#if NVLAN > 0
555	sc->sc_ec.ec_capabilities |= ETHERCAP_VLAN_HWTAGGING;
556	sc->sc_ec.ec_capenable |= ETHERCAP_VLAN_HWTAGGING;
557#endif
558
559	/* Set up MII bus. */
560	mii->mii_ifp = ifp;
561	mii->mii_readreg = ale_miibus_readreg;
562	mii->mii_writereg = ale_miibus_writereg;
563	mii->mii_statchg = ale_miibus_statchg;
564
565	sc->sc_ec.ec_mii = mii;
566	ifmedia_init(&mii->mii_media, 0, ale_mediachange, ale_mediastatus);
567	mii_flags = 0;
568	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0)
569		mii_flags |= MIIF_DOPAUSE;
570	mii_attach(self, mii, 0xffffffff, MII_PHY_ANY,
571	    MII_OFFSET_ANY, mii_flags);
572
573	if (LIST_FIRST(&mii->mii_phys) == NULL) {
574		aprint_error_dev(self, "no PHY found!\n");
575		ifmedia_add(&mii->mii_media, IFM_ETHER | IFM_MANUAL, 0, NULL);
576		ifmedia_set(&mii->mii_media, IFM_ETHER | IFM_MANUAL);
577	} else
578		ifmedia_set(&mii->mii_media, IFM_ETHER | IFM_AUTO);
579
580	if_attach(ifp);
581	if_deferred_start_init(ifp, NULL);
582	ether_ifattach(ifp, sc->ale_eaddr);
583
584	if (pmf_device_register(self, NULL, NULL))
585		pmf_class_network_register(self, ifp);
586	else
587		aprint_error_dev(self, "couldn't establish power handler\n");
588
589	return;
590fail:
591	ale_dma_free(sc);
592	if (sc->sc_irq_handle != NULL) {
593		pci_intr_disestablish(pc, sc->sc_irq_handle);
594		sc->sc_irq_handle = NULL;
595	}
596	if (sc->sc_mem_size) {
597		bus_space_unmap(sc->sc_mem_bt, sc->sc_mem_bh, sc->sc_mem_size);
598		sc->sc_mem_size = 0;
599	}
600}
601
602static int
603ale_detach(device_t self, int flags)
604{
605	struct ale_softc *sc = device_private(self);
606	struct ifnet *ifp = &sc->sc_ec.ec_if;
607	int s;
608
609	pmf_device_deregister(self);
610	s = splnet();
611	ale_stop(ifp, 0);
612	splx(s);
613
614	mii_detach(&sc->sc_miibus, MII_PHY_ANY, MII_OFFSET_ANY);
615
616	ether_ifdetach(ifp);
617	if_detach(ifp);
618	ale_dma_free(sc);
619
620	/* Delete all remaining media. */
621	ifmedia_fini(&sc->sc_miibus.mii_media);
622
623	if (sc->sc_irq_handle != NULL) {
624		pci_intr_disestablish(sc->sc_pct, sc->sc_irq_handle);
625		sc->sc_irq_handle = NULL;
626	}
627	if (sc->sc_mem_size) {
628		bus_space_unmap(sc->sc_mem_bt, sc->sc_mem_bh, sc->sc_mem_size);
629		sc->sc_mem_size = 0;
630	}
631
632	return 0;
633}
634
635
636static int
637ale_dma_alloc(struct ale_softc *sc)
638{
639	struct ale_txdesc *txd;
640	int nsegs, error, guard_size, i;
641
642	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0)
643		guard_size = ALE_JUMBO_FRAMELEN;
644	else
645		guard_size = ALE_MAX_FRAMELEN;
646	sc->ale_pagesize = roundup(guard_size + ALE_RX_PAGE_SZ,
647	    ALE_RX_PAGE_ALIGN);
648
649	/*
650	 * Create DMA stuffs for TX ring
651	 */
652	error = bus_dmamap_create(sc->sc_dmat, ALE_TX_RING_SZ, 1,
653	    ALE_TX_RING_SZ, 0, BUS_DMA_NOWAIT, &sc->ale_cdata.ale_tx_ring_map);
654	if (error) {
655		sc->ale_cdata.ale_tx_ring_map = NULL;
656		return ENOBUFS;
657	}
658
659	/* Allocate DMA'able memory for TX ring */
660	error = bus_dmamem_alloc(sc->sc_dmat, ALE_TX_RING_SZ,
661	    PAGE_SIZE, 0, &sc->ale_cdata.ale_tx_ring_seg, 1,
662	    &nsegs, BUS_DMA_WAITOK);
663	if (error) {
664		printf("%s: could not allocate DMA'able memory for Tx ring, "
665		    "error = %i\n", device_xname(sc->sc_dev), error);
666		return error;
667	}
668
669	error = bus_dmamem_map(sc->sc_dmat, &sc->ale_cdata.ale_tx_ring_seg,
670	    nsegs, ALE_TX_RING_SZ, (void **)&sc->ale_cdata.ale_tx_ring,
671	    BUS_DMA_NOWAIT);
672	if (error)
673		return ENOBUFS;
674
675	memset(sc->ale_cdata.ale_tx_ring, 0, ALE_TX_RING_SZ);
676
677	/* Load the DMA map for Tx ring. */
678	error = bus_dmamap_load(sc->sc_dmat, sc->ale_cdata.ale_tx_ring_map,
679	    sc->ale_cdata.ale_tx_ring, ALE_TX_RING_SZ, NULL, BUS_DMA_WAITOK);
680	if (error) {
681		printf("%s: could not load DMA'able memory for Tx ring.\n",
682		    device_xname(sc->sc_dev));
683		bus_dmamem_free(sc->sc_dmat,
684		    &sc->ale_cdata.ale_tx_ring_seg, 1);
685		return error;
686	}
687	sc->ale_cdata.ale_tx_ring_paddr =
688	    sc->ale_cdata.ale_tx_ring_map->dm_segs[0].ds_addr;
689
690	for (i = 0; i < ALE_RX_PAGES; i++) {
691		/*
692		 * Create DMA stuffs for RX pages
693		 */
694		error = bus_dmamap_create(sc->sc_dmat, sc->ale_pagesize, 1,
695		    sc->ale_pagesize, 0, BUS_DMA_NOWAIT,
696		    &sc->ale_cdata.ale_rx_page[i].page_map);
697		if (error) {
698			sc->ale_cdata.ale_rx_page[i].page_map = NULL;
699			return ENOBUFS;
700		}
701
702		/* Allocate DMA'able memory for RX pages */
703		error = bus_dmamem_alloc(sc->sc_dmat, sc->ale_pagesize,
704		    PAGE_SIZE, 0, &sc->ale_cdata.ale_rx_page[i].page_seg,
705		    1, &nsegs, BUS_DMA_WAITOK);
706		if (error) {
707			printf("%s: could not allocate DMA'able memory for "
708			    "Rx ring.\n", device_xname(sc->sc_dev));
709			return error;
710		}
711		error = bus_dmamem_map(sc->sc_dmat,
712		    &sc->ale_cdata.ale_rx_page[i].page_seg, nsegs,
713		    sc->ale_pagesize,
714		    (void **)&sc->ale_cdata.ale_rx_page[i].page_addr,
715		    BUS_DMA_NOWAIT);
716		if (error)
717			return ENOBUFS;
718
719		memset(sc->ale_cdata.ale_rx_page[i].page_addr, 0,
720		    sc->ale_pagesize);
721
722		/* Load the DMA map for Rx pages. */
723		error = bus_dmamap_load(sc->sc_dmat,
724		    sc->ale_cdata.ale_rx_page[i].page_map,
725		    sc->ale_cdata.ale_rx_page[i].page_addr,
726		    sc->ale_pagesize, NULL, BUS_DMA_WAITOK);
727		if (error) {
728			printf("%s: could not load DMA'able memory for "
729			    "Rx pages.\n", device_xname(sc->sc_dev));
730			bus_dmamem_free(sc->sc_dmat,
731			    &sc->ale_cdata.ale_rx_page[i].page_seg, 1);
732			return error;
733		}
734		sc->ale_cdata.ale_rx_page[i].page_paddr =
735		    sc->ale_cdata.ale_rx_page[i].page_map->dm_segs[0].ds_addr;
736	}
737
738	/*
739	 * Create DMA stuffs for Tx CMB.
740	 */
741	error = bus_dmamap_create(sc->sc_dmat, ALE_TX_CMB_SZ, 1,
742	    ALE_TX_CMB_SZ, 0, BUS_DMA_NOWAIT, &sc->ale_cdata.ale_tx_cmb_map);
743	if (error) {
744		sc->ale_cdata.ale_tx_cmb_map = NULL;
745		return ENOBUFS;
746	}
747
748	/* Allocate DMA'able memory for Tx CMB. */
749	error = bus_dmamem_alloc(sc->sc_dmat, ALE_TX_CMB_SZ, PAGE_SIZE, 0,
750	    &sc->ale_cdata.ale_tx_cmb_seg, 1, &nsegs, BUS_DMA_WAITOK);
751
752	if (error) {
753		printf("%s: could not allocate DMA'able memory for Tx CMB.\n",
754		    device_xname(sc->sc_dev));
755		return error;
756	}
757
758	error = bus_dmamem_map(sc->sc_dmat, &sc->ale_cdata.ale_tx_cmb_seg,
759	    nsegs, ALE_TX_CMB_SZ, (void **)&sc->ale_cdata.ale_tx_cmb,
760	    BUS_DMA_NOWAIT);
761	if (error)
762		return ENOBUFS;
763
764	memset(sc->ale_cdata.ale_tx_cmb, 0, ALE_TX_CMB_SZ);
765
766	/* Load the DMA map for Tx CMB. */
767	error = bus_dmamap_load(sc->sc_dmat, sc->ale_cdata.ale_tx_cmb_map,
768	    sc->ale_cdata.ale_tx_cmb, ALE_TX_CMB_SZ, NULL, BUS_DMA_WAITOK);
769	if (error) {
770		printf("%s: could not load DMA'able memory for Tx CMB.\n",
771		    device_xname(sc->sc_dev));
772		bus_dmamem_free(sc->sc_dmat,
773		    &sc->ale_cdata.ale_tx_cmb_seg, 1);
774		return error;
775	}
776
777	sc->ale_cdata.ale_tx_cmb_paddr =
778	    sc->ale_cdata.ale_tx_cmb_map->dm_segs[0].ds_addr;
779
780	for (i = 0; i < ALE_RX_PAGES; i++) {
781		/*
782		 * Create DMA stuffs for Rx CMB.
783		 */
784		error = bus_dmamap_create(sc->sc_dmat, ALE_RX_CMB_SZ, 1,
785		    ALE_RX_CMB_SZ, 0, BUS_DMA_NOWAIT,
786		    &sc->ale_cdata.ale_rx_page[i].cmb_map);
787		if (error) {
788			sc->ale_cdata.ale_rx_page[i].cmb_map = NULL;
789			return ENOBUFS;
790		}
791
792		/* Allocate DMA'able memory for Rx CMB */
793		error = bus_dmamem_alloc(sc->sc_dmat, ALE_RX_CMB_SZ,
794		    PAGE_SIZE, 0, &sc->ale_cdata.ale_rx_page[i].cmb_seg, 1,
795		    &nsegs, BUS_DMA_WAITOK);
796		if (error) {
797			printf("%s: could not allocate DMA'able memory for "
798			    "Rx CMB\n", device_xname(sc->sc_dev));
799			return error;
800		}
801		error = bus_dmamem_map(sc->sc_dmat,
802		    &sc->ale_cdata.ale_rx_page[i].cmb_seg, nsegs,
803		    ALE_RX_CMB_SZ,
804		    (void **)&sc->ale_cdata.ale_rx_page[i].cmb_addr,
805		    BUS_DMA_NOWAIT);
806		if (error)
807			return ENOBUFS;
808
809		memset(sc->ale_cdata.ale_rx_page[i].cmb_addr, 0, ALE_RX_CMB_SZ);
810
811		/* Load the DMA map for Rx CMB */
812		error = bus_dmamap_load(sc->sc_dmat,
813		    sc->ale_cdata.ale_rx_page[i].cmb_map,
814		    sc->ale_cdata.ale_rx_page[i].cmb_addr,
815		    ALE_RX_CMB_SZ, NULL, BUS_DMA_WAITOK);
816		if (error) {
817			printf("%s: could not load DMA'able memory for Rx CMB"
818			    "\n", device_xname(sc->sc_dev));
819			bus_dmamem_free(sc->sc_dmat,
820			    &sc->ale_cdata.ale_rx_page[i].cmb_seg, 1);
821			return error;
822		}
823		sc->ale_cdata.ale_rx_page[i].cmb_paddr =
824		    sc->ale_cdata.ale_rx_page[i].cmb_map->dm_segs[0].ds_addr;
825	}
826
827
828	/* Create DMA maps for Tx buffers. */
829	for (i = 0; i < ALE_TX_RING_CNT; i++) {
830		txd = &sc->ale_cdata.ale_txdesc[i];
831		txd->tx_m = NULL;
832		txd->tx_dmamap = NULL;
833		error = bus_dmamap_create(sc->sc_dmat, ALE_TSO_MAXSIZE,
834		    ALE_MAXTXSEGS, ALE_TSO_MAXSEGSIZE, 0, BUS_DMA_NOWAIT,
835		    &txd->tx_dmamap);
836		if (error) {
837			txd->tx_dmamap = NULL;
838			printf("%s: could not create Tx dmamap.\n",
839			    device_xname(sc->sc_dev));
840			return error;
841		}
842	}
843
844	return 0;
845}
846
847static void
848ale_dma_free(struct ale_softc *sc)
849{
850	struct ale_txdesc *txd;
851	int i;
852
853	/* Tx buffers. */
854	for (i = 0; i < ALE_TX_RING_CNT; i++) {
855		txd = &sc->ale_cdata.ale_txdesc[i];
856		if (txd->tx_dmamap != NULL) {
857			bus_dmamap_destroy(sc->sc_dmat, txd->tx_dmamap);
858			txd->tx_dmamap = NULL;
859		}
860	}
861
862	/* Tx descriptor ring. */
863	if (sc->ale_cdata.ale_tx_ring_map != NULL)
864		bus_dmamap_unload(sc->sc_dmat, sc->ale_cdata.ale_tx_ring_map);
865	if (sc->ale_cdata.ale_tx_ring_map != NULL &&
866	    sc->ale_cdata.ale_tx_ring != NULL)
867		bus_dmamem_free(sc->sc_dmat,
868		    &sc->ale_cdata.ale_tx_ring_seg, 1);
869	sc->ale_cdata.ale_tx_ring = NULL;
870	sc->ale_cdata.ale_tx_ring_map = NULL;
871
872	/* Rx page block. */
873	for (i = 0; i < ALE_RX_PAGES; i++) {
874		if (sc->ale_cdata.ale_rx_page[i].page_map != NULL)
875			bus_dmamap_unload(sc->sc_dmat,
876			    sc->ale_cdata.ale_rx_page[i].page_map);
877		if (sc->ale_cdata.ale_rx_page[i].page_map != NULL &&
878		    sc->ale_cdata.ale_rx_page[i].page_addr != NULL)
879			bus_dmamem_free(sc->sc_dmat,
880			    &sc->ale_cdata.ale_rx_page[i].page_seg, 1);
881		sc->ale_cdata.ale_rx_page[i].page_addr = NULL;
882		sc->ale_cdata.ale_rx_page[i].page_map = NULL;
883	}
884
885	/* Rx CMB. */
886	for (i = 0; i < ALE_RX_PAGES; i++) {
887		if (sc->ale_cdata.ale_rx_page[i].cmb_map != NULL)
888			bus_dmamap_unload(sc->sc_dmat,
889			    sc->ale_cdata.ale_rx_page[i].cmb_map);
890		if (sc->ale_cdata.ale_rx_page[i].cmb_map != NULL &&
891		    sc->ale_cdata.ale_rx_page[i].cmb_addr != NULL)
892			bus_dmamem_free(sc->sc_dmat,
893			    &sc->ale_cdata.ale_rx_page[i].cmb_seg, 1);
894		sc->ale_cdata.ale_rx_page[i].cmb_addr = NULL;
895		sc->ale_cdata.ale_rx_page[i].cmb_map = NULL;
896	}
897
898	/* Tx CMB. */
899	if (sc->ale_cdata.ale_tx_cmb_map != NULL)
900		bus_dmamap_unload(sc->sc_dmat, sc->ale_cdata.ale_tx_cmb_map);
901	if (sc->ale_cdata.ale_tx_cmb_map != NULL &&
902	    sc->ale_cdata.ale_tx_cmb != NULL)
903		bus_dmamem_free(sc->sc_dmat,
904		    &sc->ale_cdata.ale_tx_cmb_seg, 1);
905	sc->ale_cdata.ale_tx_cmb = NULL;
906	sc->ale_cdata.ale_tx_cmb_map = NULL;
907
908}
909
910static int
911ale_encap(struct ale_softc *sc, struct mbuf * const m)
912{
913	struct ale_txdesc *txd, *txd_last;
914	struct tx_desc *desc;
915	bus_dmamap_t map;
916	uint32_t cflags, poff, vtag;
917	int error, i, nsegs, prod;
918
919	cflags = vtag = 0;
920	poff = 0;
921
922	prod = sc->ale_cdata.ale_tx_prod;
923	txd = &sc->ale_cdata.ale_txdesc[prod];
924	txd_last = txd;
925	map = txd->tx_dmamap;
926
927	error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m, BUS_DMA_NOWAIT);
928	if (error == EFBIG) {
929		struct mbuf *mnew = m_defrag(m, M_NOWAIT);
930		if (mnew != NULL) {
931			KASSERT(m == mnew);
932			error = bus_dmamap_load_mbuf(sc->sc_dmat, map, mnew,
933			    BUS_DMA_NOWAIT);
934		} else {
935			/* Just drop if we can't defrag. */
936			error = EFBIG;
937		}
938		if (error) {
939			if (error == EFBIG) {
940				printf("%s: Tx packet consumes too many "
941				    "DMA segments, dropping...\n",
942				    device_xname(sc->sc_dev));
943			}
944			return error;
945		}
946	} else if (error) {
947		return error;
948	}
949
950	nsegs = map->dm_nsegs;
951	KASSERT(nsegs != 0);
952
953	/* Check descriptor overrun. */
954	if (sc->ale_cdata.ale_tx_cnt + nsegs >= ALE_TX_RING_CNT - 2) {
955		bus_dmamap_unload(sc->sc_dmat, map);
956		return ENOBUFS;
957	}
958	bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
959	    BUS_DMASYNC_PREWRITE);
960
961	/* Configure Tx checksum offload. */
962	if ((m->m_pkthdr.csum_flags & ALE_CSUM_FEATURES) != 0) {
963		/*
964		 * AR81xx supports Tx custom checksum offload feature
965		 * that offloads single 16bit checksum computation.
966		 * So you can choose one among IP, TCP and UDP.
967		 * Normally driver sets checksum start/insertion
968		 * position from the information of TCP/UDP frame as
969		 * TCP/UDP checksum takes more time than that of IP.
970		 * However it seems that custom checksum offload
971		 * requires 4 bytes aligned Tx buffers due to hardware
972		 * bug.
973		 * AR81xx also supports explicit Tx checksum computation
974		 * if it is told that the size of IP header and TCP
975		 * header(for UDP, the header size does not matter
976		 * because it's fixed length). However with this scheme
977		 * TSO does not work so you have to choose one either
978		 * TSO or explicit Tx checksum offload. I chosen TSO
979		 * plus custom checksum offload with work-around which
980		 * will cover most common usage for this consumer
981		 * ethernet controller. The work-around takes a lot of
982		 * CPU cycles if Tx buffer is not aligned on 4 bytes
983		 * boundary, though.
984		 */
985		cflags |= ALE_TD_CXSUM;
986		/* Set checksum start offset. */
987		cflags |= (poff << ALE_TD_CSUM_PLOADOFFSET_SHIFT);
988	}
989
990#if NVLAN > 0
991	/* Configure VLAN hardware tag insertion. */
992	if (vlan_has_tag(m)) {
993		vtag = ALE_TX_VLAN_TAG(htons(vlan_get_tag(m)));
994		vtag = ((vtag << ALE_TD_VLAN_SHIFT) & ALE_TD_VLAN_MASK);
995		cflags |= ALE_TD_INSERT_VLAN_TAG;
996	}
997#endif
998
999	desc = NULL;
1000	for (i = 0; i < nsegs; i++) {
1001		desc = &sc->ale_cdata.ale_tx_ring[prod];
1002		desc->addr = htole64(map->dm_segs[i].ds_addr);
1003		desc->len =
1004		    htole32(ALE_TX_BYTES(map->dm_segs[i].ds_len) | vtag);
1005		desc->flags = htole32(cflags);
1006		sc->ale_cdata.ale_tx_cnt++;
1007		ALE_DESC_INC(prod, ALE_TX_RING_CNT);
1008	}
1009	/* Update producer index. */
1010	sc->ale_cdata.ale_tx_prod = prod;
1011
1012	/* Finally set EOP on the last descriptor. */
1013	prod = (prod + ALE_TX_RING_CNT - 1) % ALE_TX_RING_CNT;
1014	desc = &sc->ale_cdata.ale_tx_ring[prod];
1015	desc->flags |= htole32(ALE_TD_EOP);
1016
1017	/* Swap dmamap of the first and the last. */
1018	txd = &sc->ale_cdata.ale_txdesc[prod];
1019	map = txd_last->tx_dmamap;
1020	txd_last->tx_dmamap = txd->tx_dmamap;
1021	txd->tx_dmamap = map;
1022	txd->tx_m = m;
1023
1024	/* Sync descriptors. */
1025	bus_dmamap_sync(sc->sc_dmat, sc->ale_cdata.ale_tx_ring_map, 0,
1026	    sc->ale_cdata.ale_tx_ring_map->dm_mapsize, BUS_DMASYNC_PREWRITE);
1027
1028	return 0;
1029}
1030
1031static void
1032ale_start(struct ifnet *ifp)
1033{
1034	struct ale_softc *sc = ifp->if_softc;
1035	struct mbuf *m_head;
1036	int enq, error;
1037
1038	if ((ifp->if_flags & (IFF_RUNNING | IFF_OACTIVE)) != IFF_RUNNING)
1039		return;
1040
1041	/* Reclaim transmitted frames. */
1042	if (sc->ale_cdata.ale_tx_cnt >= ALE_TX_DESC_HIWAT)
1043		ale_txeof(sc);
1044
1045	enq = 0;
1046	for (;;) {
1047		IFQ_POLL(&ifp->if_snd, m_head);
1048		if (m_head == NULL)
1049			break;
1050
1051		/*
1052		 * Pack the data into the transmit ring. If we
1053		 * don't have room, set the OACTIVE flag and wait
1054		 * for the NIC to drain the ring.
1055		 */
1056		if ((error = ale_encap(sc, m_head)) != 0) {
1057			if (error == EFBIG) {
1058				/* This is fatal for the packet. */
1059				IFQ_DEQUEUE(&ifp->if_snd, m_head);
1060				m_freem(m_head);
1061				if_statinc(ifp, if_oerrors);
1062				continue;
1063			}
1064			ifp->if_flags |= IFF_OACTIVE;
1065			break;
1066		}
1067		IFQ_DEQUEUE(&ifp->if_snd, m_head);
1068		enq = 1;
1069
1070		/*
1071		 * If there's a BPF listener, bounce a copy of this frame
1072		 * to him.
1073		 */
1074		bpf_mtap(ifp, m_head, BPF_D_OUT);
1075	}
1076
1077	if (enq) {
1078		/* Kick. */
1079		CSR_WRITE_4(sc, ALE_MBOX_TPD_PROD_IDX,
1080		    sc->ale_cdata.ale_tx_prod);
1081
1082		/* Set a timeout in case the chip goes out to lunch. */
1083		ifp->if_timer = ALE_TX_TIMEOUT;
1084	}
1085}
1086
1087static void
1088ale_watchdog(struct ifnet *ifp)
1089{
1090	struct ale_softc *sc = ifp->if_softc;
1091
1092	if ((sc->ale_flags & ALE_FLAG_LINK) == 0) {
1093		printf("%s: watchdog timeout (missed link)\n",
1094		    device_xname(sc->sc_dev));
1095		if_statinc(ifp, if_oerrors);
1096		ale_init(ifp);
1097		return;
1098	}
1099
1100	printf("%s: watchdog timeout\n", device_xname(sc->sc_dev));
1101	if_statinc(ifp, if_oerrors);
1102	ale_init(ifp);
1103
1104	if (!IFQ_IS_EMPTY(&ifp->if_snd))
1105		ale_start(ifp);
1106}
1107
1108static int
1109ale_ioctl(struct ifnet *ifp, u_long cmd, void *data)
1110{
1111	struct ale_softc *sc = ifp->if_softc;
1112	int s, error;
1113
1114	s = splnet();
1115
1116	error = ether_ioctl(ifp, cmd, data);
1117	if (error == ENETRESET) {
1118		if (ifp->if_flags & IFF_RUNNING)
1119			ale_rxfilter(sc);
1120		error = 0;
1121	}
1122
1123	splx(s);
1124	return error;
1125}
1126
1127static void
1128ale_mac_config(struct ale_softc *sc)
1129{
1130	struct mii_data *mii;
1131	uint32_t reg;
1132
1133	mii = &sc->sc_miibus;
1134	reg = CSR_READ_4(sc, ALE_MAC_CFG);
1135	reg &= ~(MAC_CFG_FULL_DUPLEX | MAC_CFG_TX_FC | MAC_CFG_RX_FC |
1136	    MAC_CFG_SPEED_MASK);
1137
1138	/* Reprogram MAC with resolved speed/duplex. */
1139	switch (IFM_SUBTYPE(mii->mii_media_active)) {
1140	case IFM_10_T:
1141	case IFM_100_TX:
1142		reg |= MAC_CFG_SPEED_10_100;
1143		break;
1144	case IFM_1000_T:
1145		reg |= MAC_CFG_SPEED_1000;
1146		break;
1147	}
1148	if ((IFM_OPTIONS(mii->mii_media_active) & IFM_FDX) != 0) {
1149		reg |= MAC_CFG_FULL_DUPLEX;
1150		if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_TXPAUSE) != 0)
1151			reg |= MAC_CFG_TX_FC;
1152		if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_RXPAUSE) != 0)
1153			reg |= MAC_CFG_RX_FC;
1154	}
1155	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
1156}
1157
1158static void
1159ale_stats_clear(struct ale_softc *sc)
1160{
1161	struct smb sb;
1162	uint32_t *reg;
1163	int i;
1164
1165	for (reg = &sb.rx_frames, i = 0; reg <= &sb.rx_pkts_filtered; reg++) {
1166		CSR_READ_4(sc, ALE_RX_MIB_BASE + i);
1167		i += sizeof(uint32_t);
1168	}
1169	/* Read Tx statistics. */
1170	for (reg = &sb.tx_frames, i = 0; reg <= &sb.tx_mcast_bytes; reg++) {
1171		CSR_READ_4(sc, ALE_TX_MIB_BASE + i);
1172		i += sizeof(uint32_t);
1173	}
1174}
1175
1176static void
1177ale_stats_update(struct ale_softc *sc)
1178{
1179	struct ifnet *ifp = &sc->sc_ec.ec_if;
1180	struct ale_hw_stats *stat;
1181	struct smb sb, *smb;
1182	uint32_t *reg;
1183	int i;
1184
1185	stat = &sc->ale_stats;
1186	smb = &sb;
1187
1188	/* Read Rx statistics. */
1189	for (reg = &sb.rx_frames, i = 0; reg <= &sb.rx_pkts_filtered; reg++) {
1190		*reg = CSR_READ_4(sc, ALE_RX_MIB_BASE + i);
1191		i += sizeof(uint32_t);
1192	}
1193	/* Read Tx statistics. */
1194	for (reg = &sb.tx_frames, i = 0; reg <= &sb.tx_mcast_bytes; reg++) {
1195		*reg = CSR_READ_4(sc, ALE_TX_MIB_BASE + i);
1196		i += sizeof(uint32_t);
1197	}
1198
1199	/* Rx stats. */
1200	stat->rx_frames += smb->rx_frames;
1201	stat->rx_bcast_frames += smb->rx_bcast_frames;
1202	stat->rx_mcast_frames += smb->rx_mcast_frames;
1203	stat->rx_pause_frames += smb->rx_pause_frames;
1204	stat->rx_control_frames += smb->rx_control_frames;
1205	stat->rx_crcerrs += smb->rx_crcerrs;
1206	stat->rx_lenerrs += smb->rx_lenerrs;
1207	stat->rx_bytes += smb->rx_bytes;
1208	stat->rx_runts += smb->rx_runts;
1209	stat->rx_fragments += smb->rx_fragments;
1210	stat->rx_pkts_64 += smb->rx_pkts_64;
1211	stat->rx_pkts_65_127 += smb->rx_pkts_65_127;
1212	stat->rx_pkts_128_255 += smb->rx_pkts_128_255;
1213	stat->rx_pkts_256_511 += smb->rx_pkts_256_511;
1214	stat->rx_pkts_512_1023 += smb->rx_pkts_512_1023;
1215	stat->rx_pkts_1024_1518 += smb->rx_pkts_1024_1518;
1216	stat->rx_pkts_1519_max += smb->rx_pkts_1519_max;
1217	stat->rx_pkts_truncated += smb->rx_pkts_truncated;
1218	stat->rx_fifo_oflows += smb->rx_fifo_oflows;
1219	stat->rx_rrs_errs += smb->rx_rrs_errs;
1220	stat->rx_alignerrs += smb->rx_alignerrs;
1221	stat->rx_bcast_bytes += smb->rx_bcast_bytes;
1222	stat->rx_mcast_bytes += smb->rx_mcast_bytes;
1223	stat->rx_pkts_filtered += smb->rx_pkts_filtered;
1224
1225	/* Tx stats. */
1226	stat->tx_frames += smb->tx_frames;
1227	stat->tx_bcast_frames += smb->tx_bcast_frames;
1228	stat->tx_mcast_frames += smb->tx_mcast_frames;
1229	stat->tx_pause_frames += smb->tx_pause_frames;
1230	stat->tx_excess_defer += smb->tx_excess_defer;
1231	stat->tx_control_frames += smb->tx_control_frames;
1232	stat->tx_deferred += smb->tx_deferred;
1233	stat->tx_bytes += smb->tx_bytes;
1234	stat->tx_pkts_64 += smb->tx_pkts_64;
1235	stat->tx_pkts_65_127 += smb->tx_pkts_65_127;
1236	stat->tx_pkts_128_255 += smb->tx_pkts_128_255;
1237	stat->tx_pkts_256_511 += smb->tx_pkts_256_511;
1238	stat->tx_pkts_512_1023 += smb->tx_pkts_512_1023;
1239	stat->tx_pkts_1024_1518 += smb->tx_pkts_1024_1518;
1240	stat->tx_pkts_1519_max += smb->tx_pkts_1519_max;
1241	stat->tx_single_colls += smb->tx_single_colls;
1242	stat->tx_multi_colls += smb->tx_multi_colls;
1243	stat->tx_late_colls += smb->tx_late_colls;
1244	stat->tx_excess_colls += smb->tx_excess_colls;
1245	stat->tx_abort += smb->tx_abort;
1246	stat->tx_underrun += smb->tx_underrun;
1247	stat->tx_desc_underrun += smb->tx_desc_underrun;
1248	stat->tx_lenerrs += smb->tx_lenerrs;
1249	stat->tx_pkts_truncated += smb->tx_pkts_truncated;
1250	stat->tx_bcast_bytes += smb->tx_bcast_bytes;
1251	stat->tx_mcast_bytes += smb->tx_mcast_bytes;
1252
1253	/* Update counters in ifnet. */
1254	net_stat_ref_t nsr = IF_STAT_GETREF(ifp);
1255
1256	if_statadd_ref(nsr, if_opackets, smb->tx_frames);
1257
1258	if_statadd_ref(nsr, if_collisions,
1259	    smb->tx_single_colls +
1260	    smb->tx_multi_colls * 2 + smb->tx_late_colls +
1261	    smb->tx_abort * HDPX_CFG_RETRY_DEFAULT);
1262
1263	/*
1264	 * XXX
1265	 * tx_pkts_truncated counter looks suspicious. It constantly
1266	 * increments with no sign of Tx errors. This may indicate
1267	 * the counter name is not correct one so I've removed the
1268	 * counter in output errors.
1269	 */
1270	if_statadd_ref(nsr, if_oerrors,
1271	    smb->tx_abort + smb->tx_late_colls +
1272	    smb->tx_underrun);
1273
1274	if_statadd_ref(nsr, if_ierrors,
1275	    smb->rx_crcerrs + smb->rx_lenerrs +
1276	    smb->rx_runts + smb->rx_pkts_truncated +
1277	    smb->rx_fifo_oflows + smb->rx_rrs_errs +
1278	    smb->rx_alignerrs);
1279
1280	IF_STAT_PUTREF(ifp);
1281}
1282
1283static int
1284ale_intr(void *xsc)
1285{
1286	struct ale_softc *sc = xsc;
1287	struct ifnet *ifp = &sc->sc_ec.ec_if;
1288	uint32_t status;
1289
1290	status = CSR_READ_4(sc, ALE_INTR_STATUS);
1291	if ((status & ALE_INTRS) == 0)
1292		return 0;
1293
1294	/* Acknowledge and disable interrupts. */
1295	CSR_WRITE_4(sc, ALE_INTR_STATUS, status | INTR_DIS_INT);
1296
1297	if (ifp->if_flags & IFF_RUNNING) {
1298		int error;
1299
1300		error = ale_rxeof(sc);
1301		if (error) {
1302			sc->ale_stats.reset_brk_seq++;
1303			ale_init(ifp);
1304			return 0;
1305		}
1306
1307		if (status & (INTR_DMA_RD_TO_RST | INTR_DMA_WR_TO_RST)) {
1308			if (status & INTR_DMA_RD_TO_RST)
1309				printf("%s: DMA read error! -- resetting\n",
1310				    device_xname(sc->sc_dev));
1311			if (status & INTR_DMA_WR_TO_RST)
1312				printf("%s: DMA write error! -- resetting\n",
1313				    device_xname(sc->sc_dev));
1314			ale_init(ifp);
1315			return 0;
1316		}
1317
1318		ale_txeof(sc);
1319		if_schedule_deferred_start(ifp);
1320	}
1321
1322	/* Re-enable interrupts. */
1323	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0x7FFFFFFF);
1324	return 1;
1325}
1326
1327static void
1328ale_txeof(struct ale_softc *sc)
1329{
1330	struct ifnet *ifp = &sc->sc_ec.ec_if;
1331	struct ale_txdesc *txd;
1332	uint32_t cons, prod;
1333	int prog;
1334
1335	if (sc->ale_cdata.ale_tx_cnt == 0)
1336		return;
1337
1338	bus_dmamap_sync(sc->sc_dmat, sc->ale_cdata.ale_tx_ring_map, 0,
1339	    sc->ale_cdata.ale_tx_ring_map->dm_mapsize, BUS_DMASYNC_POSTREAD);
1340	if ((sc->ale_flags & ALE_FLAG_TXCMB_BUG) == 0) {
1341		bus_dmamap_sync(sc->sc_dmat, sc->ale_cdata.ale_tx_cmb_map, 0,
1342		    sc->ale_cdata.ale_tx_cmb_map->dm_mapsize,
1343		    BUS_DMASYNC_POSTREAD);
1344		prod = *sc->ale_cdata.ale_tx_cmb & TPD_CNT_MASK;
1345	} else
1346		prod = CSR_READ_2(sc, ALE_TPD_CONS_IDX);
1347	cons = sc->ale_cdata.ale_tx_cons;
1348	/*
1349	 * Go through our Tx list and free mbufs for those
1350	 * frames which have been transmitted.
1351	 */
1352	for (prog = 0; cons != prod; prog++,
1353	     ALE_DESC_INC(cons, ALE_TX_RING_CNT)) {
1354		if (sc->ale_cdata.ale_tx_cnt <= 0)
1355			break;
1356		prog++;
1357		ifp->if_flags &= ~IFF_OACTIVE;
1358		sc->ale_cdata.ale_tx_cnt--;
1359		txd = &sc->ale_cdata.ale_txdesc[cons];
1360		if (txd->tx_m != NULL) {
1361			/* Reclaim transmitted mbufs. */
1362			bus_dmamap_unload(sc->sc_dmat, txd->tx_dmamap);
1363			m_freem(txd->tx_m);
1364			txd->tx_m = NULL;
1365		}
1366	}
1367
1368	if (prog > 0) {
1369		sc->ale_cdata.ale_tx_cons = cons;
1370		/*
1371		 * Unarm watchdog timer only when there is no pending
1372		 * Tx descriptors in queue.
1373		 */
1374		if (sc->ale_cdata.ale_tx_cnt == 0)
1375			ifp->if_timer = 0;
1376	}
1377}
1378
1379static void
1380ale_rx_update_page(struct ale_softc *sc, struct ale_rx_page **page,
1381    uint32_t length, uint32_t *prod)
1382{
1383	struct ale_rx_page *rx_page;
1384
1385	rx_page = *page;
1386	/* Update consumer position. */
1387	rx_page->cons += roundup(length + sizeof(struct rx_rs),
1388	    ALE_RX_PAGE_ALIGN);
1389	if (rx_page->cons >= ALE_RX_PAGE_SZ) {
1390		/*
1391		 * End of Rx page reached, let hardware reuse
1392		 * this page.
1393		 */
1394		rx_page->cons = 0;
1395		*rx_page->cmb_addr = 0;
1396		bus_dmamap_sync(sc->sc_dmat, rx_page->cmb_map, 0,
1397		    rx_page->cmb_map->dm_mapsize, BUS_DMASYNC_PREWRITE);
1398		CSR_WRITE_1(sc, ALE_RXF0_PAGE0 + sc->ale_cdata.ale_rx_curp,
1399		    RXF_VALID);
1400		/* Switch to alternate Rx page. */
1401		sc->ale_cdata.ale_rx_curp ^= 1;
1402		rx_page = *page =
1403		    &sc->ale_cdata.ale_rx_page[sc->ale_cdata.ale_rx_curp];
1404		/* Page flipped, sync CMB and Rx page. */
1405		bus_dmamap_sync(sc->sc_dmat, rx_page->page_map, 0,
1406		    rx_page->page_map->dm_mapsize, BUS_DMASYNC_POSTREAD);
1407		bus_dmamap_sync(sc->sc_dmat, rx_page->cmb_map, 0,
1408		    rx_page->cmb_map->dm_mapsize, BUS_DMASYNC_POSTREAD);
1409		/* Sync completed, cache updated producer index. */
1410		*prod = *rx_page->cmb_addr;
1411	}
1412}
1413
1414
1415/*
1416 * It seems that AR81xx controller can compute partial checksum.
1417 * The partial checksum value can be used to accelerate checksum
1418 * computation for fragmented TCP/UDP packets. Upper network stack
1419 * already takes advantage of the partial checksum value in IP
1420 * reassembly stage. But I'm not sure the correctness of the
1421 * partial hardware checksum assistance due to lack of data sheet.
1422 * In addition, the Rx feature of controller that requires copying
1423 * for every frames effectively nullifies one of most nice offload
1424 * capability of controller.
1425 */
1426static void
1427ale_rxcsum(struct ale_softc *sc, struct mbuf *m, uint32_t status)
1428{
1429	if (status & ALE_RD_IPCSUM_NOK)
1430		m->m_pkthdr.csum_flags |= M_CSUM_IPv4_BAD;
1431
1432	if ((sc->ale_flags & ALE_FLAG_RXCSUM_BUG) == 0) {
1433		if (((status & ALE_RD_IPV4_FRAG) == 0) &&
1434		    ((status & (ALE_RD_TCP | ALE_RD_UDP)) != 0) &&
1435		    (status & ALE_RD_TCP_UDPCSUM_NOK))
1436		{
1437			m->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
1438		}
1439	} else {
1440		if ((status & (ALE_RD_TCP | ALE_RD_UDP)) != 0) {
1441			if (status & ALE_RD_TCP_UDPCSUM_NOK) {
1442				m->m_pkthdr.csum_flags |= M_CSUM_TCP_UDP_BAD;
1443			}
1444		}
1445	}
1446	/*
1447	 * Don't mark bad checksum for TCP/UDP frames
1448	 * as fragmented frames may always have set
1449	 * bad checksummed bit of frame status.
1450	 */
1451}
1452
1453/* Process received frames. */
1454static int
1455ale_rxeof(struct ale_softc *sc)
1456{
1457	struct ifnet *ifp = &sc->sc_ec.ec_if;
1458	struct ale_rx_page *rx_page;
1459	struct rx_rs *rs;
1460	struct mbuf *m;
1461	uint32_t length, prod, seqno, status;
1462	int prog;
1463
1464	rx_page = &sc->ale_cdata.ale_rx_page[sc->ale_cdata.ale_rx_curp];
1465	bus_dmamap_sync(sc->sc_dmat, rx_page->cmb_map, 0,
1466	    rx_page->cmb_map->dm_mapsize, BUS_DMASYNC_POSTREAD);
1467	bus_dmamap_sync(sc->sc_dmat, rx_page->page_map, 0,
1468	    rx_page->page_map->dm_mapsize, BUS_DMASYNC_POSTREAD);
1469	/*
1470	 * Don't directly access producer index as hardware may
1471	 * update it while Rx handler is in progress. It would
1472	 * be even better if there is a way to let hardware
1473	 * know how far driver processed its received frames.
1474	 * Alternatively, hardware could provide a way to disable
1475	 * CMB updates until driver acknowledges the end of CMB
1476	 * access.
1477	 */
1478	prod = *rx_page->cmb_addr;
1479	for (prog = 0; ; prog++) {
1480		if (rx_page->cons >= prod)
1481			break;
1482		rs = (struct rx_rs *)(rx_page->page_addr + rx_page->cons);
1483		seqno = ALE_RX_SEQNO(le32toh(rs->seqno));
1484		if (sc->ale_cdata.ale_rx_seqno != seqno) {
1485			/*
1486			 * Normally I believe this should not happen unless
1487			 * severe driver bug or corrupted memory. However
1488			 * it seems to happen under certain conditions which
1489			 * is triggered by abrupt Rx events such as initiation
1490			 * of bulk transfer of remote host. It's not easy to
1491			 * reproduce this and I doubt it could be related
1492			 * with FIFO overflow of hardware or activity of Tx
1493			 * CMB updates. I also remember similar behaviour
1494			 * seen on RealTek 8139 which uses resembling Rx
1495			 * scheme.
1496			 */
1497			if (aledebug)
1498				printf("%s: garbled seq: %u, expected: %u -- "
1499				    "resetting!\n", device_xname(sc->sc_dev),
1500				    seqno, sc->ale_cdata.ale_rx_seqno);
1501			return EIO;
1502		}
1503		/* Frame received. */
1504		sc->ale_cdata.ale_rx_seqno++;
1505		length = ALE_RX_BYTES(le32toh(rs->length));
1506		status = le32toh(rs->flags);
1507		if (status & ALE_RD_ERROR) {
1508			/*
1509			 * We want to pass the following frames to upper
1510			 * layer regardless of error status of Rx return
1511			 * status.
1512			 *
1513			 *  o IP/TCP/UDP checksum is bad.
1514			 *  o frame length and protocol specific length
1515			 *     does not match.
1516			 */
1517			if (status & (ALE_RD_CRC | ALE_RD_CODE |
1518			    ALE_RD_DRIBBLE | ALE_RD_RUNT | ALE_RD_OFLOW |
1519			    ALE_RD_TRUNC)) {
1520				ale_rx_update_page(sc, &rx_page, length, &prod);
1521				continue;
1522			}
1523		}
1524		/*
1525		 * m_devget(9) is major bottle-neck of ale(4)(It comes
1526		 * from hardware limitation). For jumbo frames we could
1527		 * get a slightly better performance if driver use
1528		 * m_getjcl(9) with proper buffer size argument. However
1529		 * that would make code more complicated and I don't
1530		 * think users would expect good Rx performance numbers
1531		 * on these low-end consumer ethernet controller.
1532		 */
1533		m = m_devget((char *)(rs + 1), length - ETHER_CRC_LEN,
1534		    0, ifp);
1535		if (m == NULL) {
1536			if_statinc(ifp, if_iqdrops);
1537			ale_rx_update_page(sc, &rx_page, length, &prod);
1538			continue;
1539		}
1540		if (status & ALE_RD_IPV4)
1541			ale_rxcsum(sc, m, status);
1542#if NVLAN > 0
1543		if (status & ALE_RD_VLAN) {
1544			uint32_t vtags = ALE_RX_VLAN(le32toh(rs->vtags));
1545			vlan_set_tag(m, ALE_RX_VLAN_TAG(vtags));
1546		}
1547#endif
1548
1549		/* Pass it to upper layer. */
1550		if_percpuq_enqueue(ifp->if_percpuq, m);
1551
1552		ale_rx_update_page(sc, &rx_page, length, &prod);
1553	}
1554
1555	return 0;
1556}
1557
1558static void
1559ale_tick(void *xsc)
1560{
1561	struct ale_softc *sc = xsc;
1562	struct mii_data *mii = &sc->sc_miibus;
1563	int s;
1564
1565	s = splnet();
1566	mii_tick(mii);
1567	ale_stats_update(sc);
1568	splx(s);
1569
1570	callout_schedule(&sc->sc_tick_ch, hz);
1571}
1572
1573static void
1574ale_reset(struct ale_softc *sc)
1575{
1576	uint32_t reg;
1577	int i;
1578
1579	/* Initialize PCIe module. From Linux. */
1580	CSR_WRITE_4(sc, 0x1008, CSR_READ_4(sc, 0x1008) | 0x8000);
1581
1582	CSR_WRITE_4(sc, ALE_MASTER_CFG, MASTER_RESET);
1583	for (i = ALE_RESET_TIMEOUT; i > 0; i--) {
1584		DELAY(10);
1585		if ((CSR_READ_4(sc, ALE_MASTER_CFG) & MASTER_RESET) == 0)
1586			break;
1587	}
1588	if (i == 0)
1589		printf("%s: master reset timeout!\n", device_xname(sc->sc_dev));
1590
1591	for (i = ALE_RESET_TIMEOUT; i > 0; i--) {
1592		if ((reg = CSR_READ_4(sc, ALE_IDLE_STATUS)) == 0)
1593			break;
1594		DELAY(10);
1595	}
1596
1597	if (i == 0)
1598		printf("%s: reset timeout(0x%08x)!\n", device_xname(sc->sc_dev),
1599		    reg);
1600}
1601
1602static int
1603ale_init(struct ifnet *ifp)
1604{
1605	struct ale_softc *sc = ifp->if_softc;
1606	struct mii_data *mii;
1607	uint8_t eaddr[ETHER_ADDR_LEN];
1608	bus_addr_t paddr;
1609	uint32_t reg, rxf_hi, rxf_lo;
1610
1611	/*
1612	 * Cancel any pending I/O.
1613	 */
1614	ale_stop(ifp, 0);
1615
1616	/*
1617	 * Reset the chip to a known state.
1618	 */
1619	ale_reset(sc);
1620
1621	/* Initialize Tx descriptors, DMA memory blocks. */
1622	ale_init_rx_pages(sc);
1623	ale_init_tx_ring(sc);
1624
1625	/* Reprogram the station address. */
1626	memcpy(eaddr, CLLADDR(ifp->if_sadl), ETHER_ADDR_LEN);
1627	CSR_WRITE_4(sc, ALE_PAR0,
1628	    eaddr[2] << 24 | eaddr[3] << 16 | eaddr[4] << 8 | eaddr[5]);
1629	CSR_WRITE_4(sc, ALE_PAR1, eaddr[0] << 8 | eaddr[1]);
1630
1631	/*
1632	 * Clear WOL status and disable all WOL feature as WOL
1633	 * would interfere Rx operation under normal environments.
1634	 */
1635	CSR_READ_4(sc, ALE_WOL_CFG);
1636	CSR_WRITE_4(sc, ALE_WOL_CFG, 0);
1637
1638	/*
1639	 * Set Tx descriptor/RXF0/CMB base addresses. They share
1640	 * the same high address part of DMAable region.
1641	 */
1642	paddr = sc->ale_cdata.ale_tx_ring_paddr;
1643	CSR_WRITE_4(sc, ALE_TPD_ADDR_HI, ALE_ADDR_HI(paddr));
1644	CSR_WRITE_4(sc, ALE_TPD_ADDR_LO, ALE_ADDR_LO(paddr));
1645	CSR_WRITE_4(sc, ALE_TPD_CNT,
1646	    (ALE_TX_RING_CNT << TPD_CNT_SHIFT) & TPD_CNT_MASK);
1647
1648	/* Set Rx page base address, note we use single queue. */
1649	paddr = sc->ale_cdata.ale_rx_page[0].page_paddr;
1650	CSR_WRITE_4(sc, ALE_RXF0_PAGE0_ADDR_LO, ALE_ADDR_LO(paddr));
1651	paddr = sc->ale_cdata.ale_rx_page[1].page_paddr;
1652	CSR_WRITE_4(sc, ALE_RXF0_PAGE1_ADDR_LO, ALE_ADDR_LO(paddr));
1653
1654	/* Set Tx/Rx CMB addresses. */
1655	paddr = sc->ale_cdata.ale_tx_cmb_paddr;
1656	CSR_WRITE_4(sc, ALE_TX_CMB_ADDR_LO, ALE_ADDR_LO(paddr));
1657	paddr = sc->ale_cdata.ale_rx_page[0].cmb_paddr;
1658	CSR_WRITE_4(sc, ALE_RXF0_CMB0_ADDR_LO, ALE_ADDR_LO(paddr));
1659	paddr = sc->ale_cdata.ale_rx_page[1].cmb_paddr;
1660	CSR_WRITE_4(sc, ALE_RXF0_CMB1_ADDR_LO, ALE_ADDR_LO(paddr));
1661
1662	/* Mark RXF0 is valid. */
1663	CSR_WRITE_1(sc, ALE_RXF0_PAGE0, RXF_VALID);
1664	CSR_WRITE_1(sc, ALE_RXF0_PAGE1, RXF_VALID);
1665	/*
1666	 * No need to initialize RFX1/RXF2/RXF3. We don't use
1667	 * multi-queue yet.
1668	 */
1669
1670	/* Set Rx page size, excluding guard frame size. */
1671	CSR_WRITE_4(sc, ALE_RXF_PAGE_SIZE, ALE_RX_PAGE_SZ);
1672
1673	/* Tell hardware that we're ready to load DMA blocks. */
1674	CSR_WRITE_4(sc, ALE_DMA_BLOCK, DMA_BLOCK_LOAD);
1675
1676	/* Set Rx/Tx interrupt trigger threshold. */
1677	CSR_WRITE_4(sc, ALE_INT_TRIG_THRESH, (1 << INT_TRIG_RX_THRESH_SHIFT) |
1678	    (4 << INT_TRIG_TX_THRESH_SHIFT));
1679	/*
1680	 * XXX
1681	 * Set interrupt trigger timer, its purpose and relation
1682	 * with interrupt moderation mechanism is not clear yet.
1683	 */
1684	CSR_WRITE_4(sc, ALE_INT_TRIG_TIMER,
1685	    ((ALE_USECS(10) << INT_TRIG_RX_TIMER_SHIFT) |
1686	    (ALE_USECS(1000) << INT_TRIG_TX_TIMER_SHIFT)));
1687
1688	/* Configure interrupt moderation timer. */
1689	sc->ale_int_rx_mod = ALE_IM_RX_TIMER_DEFAULT;
1690	sc->ale_int_tx_mod = ALE_IM_TX_TIMER_DEFAULT;
1691	reg = ALE_USECS(sc->ale_int_rx_mod) << IM_TIMER_RX_SHIFT;
1692	reg |= ALE_USECS(sc->ale_int_tx_mod) << IM_TIMER_TX_SHIFT;
1693	CSR_WRITE_4(sc, ALE_IM_TIMER, reg);
1694	reg = CSR_READ_4(sc, ALE_MASTER_CFG);
1695	reg &= ~(MASTER_CHIP_REV_MASK | MASTER_CHIP_ID_MASK);
1696	reg &= ~(MASTER_IM_RX_TIMER_ENB | MASTER_IM_TX_TIMER_ENB);
1697	if (ALE_USECS(sc->ale_int_rx_mod) != 0)
1698		reg |= MASTER_IM_RX_TIMER_ENB;
1699	if (ALE_USECS(sc->ale_int_tx_mod) != 0)
1700		reg |= MASTER_IM_TX_TIMER_ENB;
1701	CSR_WRITE_4(sc, ALE_MASTER_CFG, reg);
1702	CSR_WRITE_2(sc, ALE_INTR_CLR_TIMER, ALE_USECS(1000));
1703
1704	/* Set Maximum frame size of controller. */
1705	if (ifp->if_mtu < ETHERMTU)
1706		sc->ale_max_frame_size = ETHERMTU;
1707	else
1708		sc->ale_max_frame_size = ifp->if_mtu;
1709	sc->ale_max_frame_size += ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN + ETHER_CRC_LEN;
1710	CSR_WRITE_4(sc, ALE_FRAME_SIZE, sc->ale_max_frame_size);
1711
1712	/* Configure IPG/IFG parameters. */
1713	CSR_WRITE_4(sc, ALE_IPG_IFG_CFG,
1714	    ((IPG_IFG_IPGT_DEFAULT << IPG_IFG_IPGT_SHIFT) & IPG_IFG_IPGT_MASK) |
1715	    ((IPG_IFG_MIFG_DEFAULT << IPG_IFG_MIFG_SHIFT) & IPG_IFG_MIFG_MASK) |
1716	    ((IPG_IFG_IPG1_DEFAULT << IPG_IFG_IPG1_SHIFT) & IPG_IFG_IPG1_MASK) |
1717	    ((IPG_IFG_IPG2_DEFAULT << IPG_IFG_IPG2_SHIFT) & IPG_IFG_IPG2_MASK));
1718
1719	/* Set parameters for half-duplex media. */
1720	CSR_WRITE_4(sc, ALE_HDPX_CFG,
1721	    ((HDPX_CFG_LCOL_DEFAULT << HDPX_CFG_LCOL_SHIFT) &
1722	    HDPX_CFG_LCOL_MASK) |
1723	    ((HDPX_CFG_RETRY_DEFAULT << HDPX_CFG_RETRY_SHIFT) &
1724	    HDPX_CFG_RETRY_MASK) | HDPX_CFG_EXC_DEF_EN |
1725	    ((HDPX_CFG_ABEBT_DEFAULT << HDPX_CFG_ABEBT_SHIFT) &
1726	    HDPX_CFG_ABEBT_MASK) |
1727	    ((HDPX_CFG_JAMIPG_DEFAULT << HDPX_CFG_JAMIPG_SHIFT) &
1728	    HDPX_CFG_JAMIPG_MASK));
1729
1730	/* Configure Tx jumbo frame parameters. */
1731	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0) {
1732		if (ifp->if_mtu < ETHERMTU)
1733			reg = sc->ale_max_frame_size;
1734		else if (ifp->if_mtu < 6 * 1024)
1735			reg = (sc->ale_max_frame_size * 2) / 3;
1736		else
1737			reg = sc->ale_max_frame_size / 2;
1738		CSR_WRITE_4(sc, ALE_TX_JUMBO_THRESH,
1739		    roundup(reg, TX_JUMBO_THRESH_UNIT) >>
1740		    TX_JUMBO_THRESH_UNIT_SHIFT);
1741	}
1742
1743	/* Configure TxQ. */
1744	reg = (128 << (sc->ale_dma_rd_burst >> DMA_CFG_RD_BURST_SHIFT))
1745	    << TXQ_CFG_TX_FIFO_BURST_SHIFT;
1746	reg |= (TXQ_CFG_TPD_BURST_DEFAULT << TXQ_CFG_TPD_BURST_SHIFT) &
1747	    TXQ_CFG_TPD_BURST_MASK;
1748	CSR_WRITE_4(sc, ALE_TXQ_CFG, reg | TXQ_CFG_ENHANCED_MODE | TXQ_CFG_ENB);
1749
1750	/* Configure Rx jumbo frame & flow control parameters. */
1751	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0) {
1752		reg = roundup(sc->ale_max_frame_size, RX_JUMBO_THRESH_UNIT);
1753		CSR_WRITE_4(sc, ALE_RX_JUMBO_THRESH,
1754		    (((reg >> RX_JUMBO_THRESH_UNIT_SHIFT) <<
1755		    RX_JUMBO_THRESH_MASK_SHIFT) & RX_JUMBO_THRESH_MASK) |
1756		    ((RX_JUMBO_LKAH_DEFAULT << RX_JUMBO_LKAH_SHIFT) &
1757		    RX_JUMBO_LKAH_MASK));
1758		reg = CSR_READ_4(sc, ALE_SRAM_RX_FIFO_LEN);
1759		rxf_hi = (reg * 7) / 10;
1760		rxf_lo = (reg * 3)/ 10;
1761		CSR_WRITE_4(sc, ALE_RX_FIFO_PAUSE_THRESH,
1762		    ((rxf_lo << RX_FIFO_PAUSE_THRESH_LO_SHIFT) &
1763		    RX_FIFO_PAUSE_THRESH_LO_MASK) |
1764		    ((rxf_hi << RX_FIFO_PAUSE_THRESH_HI_SHIFT) &
1765		     RX_FIFO_PAUSE_THRESH_HI_MASK));
1766	}
1767
1768	/* Disable RSS. */
1769	CSR_WRITE_4(sc, ALE_RSS_IDT_TABLE0, 0);
1770	CSR_WRITE_4(sc, ALE_RSS_CPU, 0);
1771
1772	/* Configure RxQ. */
1773	CSR_WRITE_4(sc, ALE_RXQ_CFG,
1774	    RXQ_CFG_ALIGN_32 | RXQ_CFG_CUT_THROUGH_ENB | RXQ_CFG_ENB);
1775
1776	/* Configure DMA parameters. */
1777	reg = 0;
1778	if ((sc->ale_flags & ALE_FLAG_TXCMB_BUG) == 0)
1779		reg |= DMA_CFG_TXCMB_ENB;
1780	CSR_WRITE_4(sc, ALE_DMA_CFG,
1781	    DMA_CFG_OUT_ORDER | DMA_CFG_RD_REQ_PRI | DMA_CFG_RCB_64 |
1782	    sc->ale_dma_rd_burst | reg |
1783	    sc->ale_dma_wr_burst | DMA_CFG_RXCMB_ENB |
1784	    ((DMA_CFG_RD_DELAY_CNT_DEFAULT << DMA_CFG_RD_DELAY_CNT_SHIFT) &
1785	    DMA_CFG_RD_DELAY_CNT_MASK) |
1786	    ((DMA_CFG_WR_DELAY_CNT_DEFAULT << DMA_CFG_WR_DELAY_CNT_SHIFT) &
1787	    DMA_CFG_WR_DELAY_CNT_MASK));
1788
1789	/*
1790	 * Hardware can be configured to issue SMB interrupt based
1791	 * on programmed interval. Since there is a callout that is
1792	 * invoked for every hz in driver we use that instead of
1793	 * relying on periodic SMB interrupt.
1794	 */
1795	CSR_WRITE_4(sc, ALE_SMB_STAT_TIMER, ALE_USECS(0));
1796
1797	/* Clear MAC statistics. */
1798	ale_stats_clear(sc);
1799
1800	/*
1801	 * Configure Tx/Rx MACs.
1802	 *  - Auto-padding for short frames.
1803	 *  - Enable CRC generation.
1804	 *  Actual reconfiguration of MAC for resolved speed/duplex
1805	 *  is followed after detection of link establishment.
1806	 *  AR81xx always does checksum computation regardless of
1807	 *  MAC_CFG_RXCSUM_ENB bit. In fact, setting the bit will
1808	 *  cause Rx handling issue for fragmented IP datagrams due
1809	 *  to silicon bug.
1810	 */
1811	reg = MAC_CFG_TX_CRC_ENB | MAC_CFG_TX_AUTO_PAD | MAC_CFG_FULL_DUPLEX |
1812	    ((MAC_CFG_PREAMBLE_DEFAULT << MAC_CFG_PREAMBLE_SHIFT) &
1813	    MAC_CFG_PREAMBLE_MASK);
1814	if ((sc->ale_flags & ALE_FLAG_FASTETHER) != 0)
1815		reg |= MAC_CFG_SPEED_10_100;
1816	else
1817		reg |= MAC_CFG_SPEED_1000;
1818	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
1819
1820	/* Set up the receive filter. */
1821	ale_rxfilter(sc);
1822	ale_rxvlan(sc);
1823
1824	/* Acknowledge all pending interrupts and clear it. */
1825	CSR_WRITE_4(sc, ALE_INTR_MASK, ALE_INTRS);
1826	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0xFFFFFFFF);
1827	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0);
1828
1829	sc->ale_flags &= ~ALE_FLAG_LINK;
1830
1831	/* Switch to the current media. */
1832	mii = &sc->sc_miibus;
1833	mii_mediachg(mii);
1834
1835	callout_schedule(&sc->sc_tick_ch, hz);
1836
1837	ifp->if_flags |= IFF_RUNNING;
1838	ifp->if_flags &= ~IFF_OACTIVE;
1839
1840	return 0;
1841}
1842
1843static void
1844ale_stop(struct ifnet *ifp, int disable)
1845{
1846	struct ale_softc *sc = ifp->if_softc;
1847	struct ale_txdesc *txd;
1848	uint32_t reg;
1849	int i;
1850
1851	callout_stop(&sc->sc_tick_ch);
1852
1853	/*
1854	 * Mark the interface down and cancel the watchdog timer.
1855	 */
1856	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
1857	ifp->if_timer = 0;
1858
1859	sc->ale_flags &= ~ALE_FLAG_LINK;
1860
1861	ale_stats_update(sc);
1862
1863	mii_down(&sc->sc_miibus);
1864
1865	/* Disable interrupts. */
1866	CSR_WRITE_4(sc, ALE_INTR_MASK, 0);
1867	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0xFFFFFFFF);
1868
1869	/* Disable queue processing and DMA. */
1870	reg = CSR_READ_4(sc, ALE_TXQ_CFG);
1871	reg &= ~TXQ_CFG_ENB;
1872	CSR_WRITE_4(sc, ALE_TXQ_CFG, reg);
1873	reg = CSR_READ_4(sc, ALE_RXQ_CFG);
1874	reg &= ~RXQ_CFG_ENB;
1875	CSR_WRITE_4(sc, ALE_RXQ_CFG, reg);
1876	reg = CSR_READ_4(sc, ALE_DMA_CFG);
1877	reg &= ~(DMA_CFG_TXCMB_ENB | DMA_CFG_RXCMB_ENB);
1878	CSR_WRITE_4(sc, ALE_DMA_CFG, reg);
1879	DELAY(1000);
1880
1881	/* Stop Rx/Tx MACs. */
1882	ale_stop_mac(sc);
1883
1884	/* Disable interrupts again? XXX */
1885	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0xFFFFFFFF);
1886
1887	/*
1888	 * Free TX mbufs still in the queues.
1889	 */
1890	for (i = 0; i < ALE_TX_RING_CNT; i++) {
1891		txd = &sc->ale_cdata.ale_txdesc[i];
1892		if (txd->tx_m != NULL) {
1893			bus_dmamap_unload(sc->sc_dmat, txd->tx_dmamap);
1894			m_freem(txd->tx_m);
1895			txd->tx_m = NULL;
1896		}
1897	}
1898}
1899
1900static void
1901ale_stop_mac(struct ale_softc *sc)
1902{
1903	uint32_t reg;
1904	int i;
1905
1906	reg = CSR_READ_4(sc, ALE_MAC_CFG);
1907	if ((reg & (MAC_CFG_TX_ENB | MAC_CFG_RX_ENB)) != 0) {
1908		reg &= ~(MAC_CFG_TX_ENB | MAC_CFG_RX_ENB);
1909		CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
1910	}
1911
1912	for (i = ALE_TIMEOUT; i > 0; i--) {
1913		reg = CSR_READ_4(sc, ALE_IDLE_STATUS);
1914		if (reg == 0)
1915			break;
1916		DELAY(10);
1917	}
1918	if (i == 0)
1919		printf("%s: could not disable Tx/Rx MAC(0x%08x)!\n",
1920		    device_xname(sc->sc_dev), reg);
1921}
1922
1923static void
1924ale_init_tx_ring(struct ale_softc *sc)
1925{
1926	struct ale_txdesc *txd;
1927	int i;
1928
1929	sc->ale_cdata.ale_tx_prod = 0;
1930	sc->ale_cdata.ale_tx_cons = 0;
1931	sc->ale_cdata.ale_tx_cnt = 0;
1932
1933	memset(sc->ale_cdata.ale_tx_ring, 0, ALE_TX_RING_SZ);
1934	memset(sc->ale_cdata.ale_tx_cmb, 0, ALE_TX_CMB_SZ);
1935	for (i = 0; i < ALE_TX_RING_CNT; i++) {
1936		txd = &sc->ale_cdata.ale_txdesc[i];
1937		txd->tx_m = NULL;
1938	}
1939	*sc->ale_cdata.ale_tx_cmb = 0;
1940	bus_dmamap_sync(sc->sc_dmat, sc->ale_cdata.ale_tx_cmb_map, 0,
1941	    sc->ale_cdata.ale_tx_cmb_map->dm_mapsize, BUS_DMASYNC_PREWRITE);
1942	bus_dmamap_sync(sc->sc_dmat, sc->ale_cdata.ale_tx_ring_map, 0,
1943	    sc->ale_cdata.ale_tx_ring_map->dm_mapsize, BUS_DMASYNC_PREWRITE);
1944}
1945
1946static void
1947ale_init_rx_pages(struct ale_softc *sc)
1948{
1949	struct ale_rx_page *rx_page;
1950	int i;
1951
1952	sc->ale_cdata.ale_rx_seqno = 0;
1953	sc->ale_cdata.ale_rx_curp = 0;
1954
1955	for (i = 0; i < ALE_RX_PAGES; i++) {
1956		rx_page = &sc->ale_cdata.ale_rx_page[i];
1957		memset(rx_page->page_addr, 0, sc->ale_pagesize);
1958		memset(rx_page->cmb_addr, 0, ALE_RX_CMB_SZ);
1959		rx_page->cons = 0;
1960		*rx_page->cmb_addr = 0;
1961		bus_dmamap_sync(sc->sc_dmat, rx_page->page_map, 0,
1962		    rx_page->page_map->dm_mapsize, BUS_DMASYNC_PREWRITE);
1963		bus_dmamap_sync(sc->sc_dmat, rx_page->cmb_map, 0,
1964		    rx_page->cmb_map->dm_mapsize, BUS_DMASYNC_PREWRITE);
1965	}
1966}
1967
1968static void
1969ale_rxvlan(struct ale_softc *sc)
1970{
1971	uint32_t reg;
1972
1973	reg = CSR_READ_4(sc, ALE_MAC_CFG);
1974	reg &= ~MAC_CFG_VLAN_TAG_STRIP;
1975	if (sc->sc_ec.ec_capenable & ETHERCAP_VLAN_HWTAGGING)
1976		reg |= MAC_CFG_VLAN_TAG_STRIP;
1977	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
1978}
1979
1980static void
1981ale_rxfilter(struct ale_softc *sc)
1982{
1983	struct ethercom *ec = &sc->sc_ec;
1984	struct ifnet *ifp = &ec->ec_if;
1985	struct ether_multi *enm;
1986	struct ether_multistep step;
1987	uint32_t crc;
1988	uint32_t mchash[2];
1989	uint32_t rxcfg;
1990
1991	rxcfg = CSR_READ_4(sc, ALE_MAC_CFG);
1992	rxcfg &= ~(MAC_CFG_ALLMULTI | MAC_CFG_BCAST | MAC_CFG_PROMISC);
1993	ifp->if_flags &= ~IFF_ALLMULTI;
1994
1995	/*
1996	 * Always accept broadcast frames.
1997	 */
1998	rxcfg |= MAC_CFG_BCAST;
1999
2000	/* Program new filter. */
2001	if ((ifp->if_flags & IFF_PROMISC) != 0)
2002		goto update;
2003
2004	memset(mchash, 0, sizeof(mchash));
2005
2006	ETHER_LOCK(ec);
2007	ETHER_FIRST_MULTI(step, ec, enm);
2008	while (enm != NULL) {
2009		if (memcmp(enm->enm_addrlo, enm->enm_addrhi, ETHER_ADDR_LEN)) {
2010			/* XXX Use ETHER_F_ALLMULTI in future. */
2011			ifp->if_flags |= IFF_ALLMULTI;
2012			ETHER_UNLOCK(ec);
2013			goto update;
2014		}
2015		crc = ether_crc32_be(enm->enm_addrlo, ETHER_ADDR_LEN);
2016		mchash[crc >> 31] |= 1U << ((crc >> 26) & 0x1f);
2017		ETHER_NEXT_MULTI(step, enm);
2018	}
2019	ETHER_UNLOCK(ec);
2020
2021update:
2022	if ((ifp->if_flags & (IFF_PROMISC | IFF_ALLMULTI)) != 0) {
2023		if (ifp->if_flags & IFF_PROMISC) {
2024			rxcfg |= MAC_CFG_PROMISC;
2025			/* XXX Use ETHER_F_ALLMULTI in future. */
2026			ifp->if_flags |= IFF_ALLMULTI;
2027		} else
2028			rxcfg |= MAC_CFG_ALLMULTI;
2029		mchash[0] = mchash[1] = 0xFFFFFFFF;
2030	}
2031	CSR_WRITE_4(sc, ALE_MAR0, mchash[0]);
2032	CSR_WRITE_4(sc, ALE_MAR1, mchash[1]);
2033	CSR_WRITE_4(sc, ALE_MAC_CFG, rxcfg);
2034}
2035