xref: /freebsd/sys/dev/ale/if_ale.c (revision f5524be39e26e3d7fb774ffe7711602c676a8b9e)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
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 
30 /* Driver for Atheros AR8121/AR8113/AR8114 PCIe Ethernet. */
31 
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/bus.h>
35 #include <sys/endian.h>
36 #include <sys/kernel.h>
37 #include <sys/malloc.h>
38 #include <sys/mbuf.h>
39 #include <sys/module.h>
40 #include <sys/rman.h>
41 #include <sys/queue.h>
42 #include <sys/socket.h>
43 #include <sys/sockio.h>
44 #include <sys/sysctl.h>
45 #include <sys/taskqueue.h>
46 
47 #include <net/bpf.h>
48 #include <net/if.h>
49 #include <net/if_var.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 
81 MODULE_DEPEND(ale, pci, 1, 1, 1);
82 MODULE_DEPEND(ale, ether, 1, 1, 1);
83 MODULE_DEPEND(ale, miibus, 1, 1, 1);
84 
85 /* Tunables. */
86 static int msi_disable = 0;
87 static int msix_disable = 0;
88 TUNABLE_INT("hw.ale.msi_disable", &msi_disable);
89 TUNABLE_INT("hw.ale.msix_disable", &msix_disable);
90 
91 /*
92  * Devices supported by this driver.
93  */
94 static const 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 
103 static int	ale_attach(device_t);
104 static int	ale_check_boundary(struct ale_softc *);
105 static int	ale_detach(device_t);
106 static int	ale_dma_alloc(struct ale_softc *);
107 static void	ale_dma_free(struct ale_softc *);
108 static void	ale_dmamap_cb(void *, bus_dma_segment_t *, int, int);
109 static int	ale_encap(struct ale_softc *, struct mbuf **);
110 static void	ale_get_macaddr(struct ale_softc *);
111 static void	ale_init(void *);
112 static void	ale_init_locked(struct ale_softc *);
113 static void	ale_init_rx_pages(struct ale_softc *);
114 static void	ale_init_tx_ring(struct ale_softc *);
115 static void	ale_int_task(void *, int);
116 static int	ale_intr(void *);
117 static int	ale_ioctl(if_t, u_long, caddr_t);
118 static void	ale_mac_config(struct ale_softc *);
119 static int	ale_miibus_readreg(device_t, int, int);
120 static void	ale_miibus_statchg(device_t);
121 static int	ale_miibus_writereg(device_t, int, int, int);
122 static int	ale_mediachange(if_t);
123 static void	ale_mediastatus(if_t, struct ifmediareq *);
124 static void	ale_phy_reset(struct ale_softc *);
125 static int	ale_probe(device_t);
126 static void	ale_reset(struct ale_softc *);
127 static int	ale_resume(device_t);
128 static void	ale_rx_update_page(struct ale_softc *, struct ale_rx_page **,
129     uint32_t, uint32_t *);
130 static void	ale_rxcsum(struct ale_softc *, struct mbuf *, uint32_t);
131 static int	ale_rxeof(struct ale_softc *sc, int);
132 static void	ale_rxfilter(struct ale_softc *);
133 static void	ale_rxvlan(struct ale_softc *);
134 static void	ale_setlinkspeed(struct ale_softc *);
135 static void	ale_setwol(struct ale_softc *);
136 static int	ale_shutdown(device_t);
137 static void	ale_start(if_t);
138 static void	ale_start_locked(if_t);
139 static void	ale_stats_clear(struct ale_softc *);
140 static void	ale_stats_update(struct ale_softc *);
141 static void	ale_stop(struct ale_softc *);
142 static void	ale_stop_mac(struct ale_softc *);
143 static int	ale_suspend(device_t);
144 static void	ale_sysctl_node(struct ale_softc *);
145 static void	ale_tick(void *);
146 static void	ale_txeof(struct ale_softc *);
147 static void	ale_watchdog(struct ale_softc *);
148 static int	sysctl_int_range(SYSCTL_HANDLER_ARGS, int, int);
149 static int	sysctl_hw_ale_proc_limit(SYSCTL_HANDLER_ARGS);
150 static int	sysctl_hw_ale_int_mod(SYSCTL_HANDLER_ARGS);
151 
152 static device_method_t ale_methods[] = {
153 	/* Device interface. */
154 	DEVMETHOD(device_probe,		ale_probe),
155 	DEVMETHOD(device_attach,	ale_attach),
156 	DEVMETHOD(device_detach,	ale_detach),
157 	DEVMETHOD(device_shutdown,	ale_shutdown),
158 	DEVMETHOD(device_suspend,	ale_suspend),
159 	DEVMETHOD(device_resume,	ale_resume),
160 
161 	/* MII interface. */
162 	DEVMETHOD(miibus_readreg,	ale_miibus_readreg),
163 	DEVMETHOD(miibus_writereg,	ale_miibus_writereg),
164 	DEVMETHOD(miibus_statchg,	ale_miibus_statchg),
165 
166 	DEVMETHOD_END
167 };
168 
169 static driver_t ale_driver = {
170 	"ale",
171 	ale_methods,
172 	sizeof(struct ale_softc)
173 };
174 
175 DRIVER_MODULE(ale, pci, ale_driver, NULL, NULL);
176 MODULE_PNP_INFO("U16:vendor;U16:device;D:#", pci, ale, ale_devs,
177     nitems(ale_devs));
178 DRIVER_MODULE(miibus, ale, miibus_driver, NULL, NULL);
179 
180 static struct resource_spec ale_res_spec_mem[] = {
181 	{ SYS_RES_MEMORY,	PCIR_BAR(0),	RF_ACTIVE },
182 	{ -1,			0,		0 }
183 };
184 
185 static struct resource_spec ale_irq_spec_legacy[] = {
186 	{ SYS_RES_IRQ,		0,		RF_ACTIVE | RF_SHAREABLE },
187 	{ -1,			0,		0 }
188 };
189 
190 static struct resource_spec ale_irq_spec_msi[] = {
191 	{ SYS_RES_IRQ,		1,		RF_ACTIVE },
192 	{ -1,			0,		0 }
193 };
194 
195 static struct resource_spec ale_irq_spec_msix[] = {
196 	{ SYS_RES_IRQ,		1,		RF_ACTIVE },
197 	{ -1,			0,		0 }
198 };
199 
200 static int
ale_miibus_readreg(device_t dev,int phy,int reg)201 ale_miibus_readreg(device_t dev, int phy, int reg)
202 {
203 	struct ale_softc *sc;
204 	uint32_t v;
205 	int i;
206 
207 	sc = device_get_softc(dev);
208 
209 	CSR_WRITE_4(sc, ALE_MDIO, MDIO_OP_EXECUTE | MDIO_OP_READ |
210 	    MDIO_SUP_PREAMBLE | MDIO_CLK_25_4 | MDIO_REG_ADDR(reg));
211 	for (i = ALE_PHY_TIMEOUT; i > 0; i--) {
212 		DELAY(5);
213 		v = CSR_READ_4(sc, ALE_MDIO);
214 		if ((v & (MDIO_OP_EXECUTE | MDIO_OP_BUSY)) == 0)
215 			break;
216 	}
217 
218 	if (i == 0) {
219 		device_printf(sc->ale_dev, "phy read timeout : %d\n", reg);
220 		return (0);
221 	}
222 
223 	return ((v & MDIO_DATA_MASK) >> MDIO_DATA_SHIFT);
224 }
225 
226 static int
ale_miibus_writereg(device_t dev,int phy,int reg,int val)227 ale_miibus_writereg(device_t dev, int phy, int reg, int val)
228 {
229 	struct ale_softc *sc;
230 	uint32_t v;
231 	int i;
232 
233 	sc = device_get_softc(dev);
234 
235 	CSR_WRITE_4(sc, ALE_MDIO, MDIO_OP_EXECUTE | MDIO_OP_WRITE |
236 	    (val & MDIO_DATA_MASK) << MDIO_DATA_SHIFT |
237 	    MDIO_SUP_PREAMBLE | MDIO_CLK_25_4 | MDIO_REG_ADDR(reg));
238 	for (i = ALE_PHY_TIMEOUT; i > 0; i--) {
239 		DELAY(5);
240 		v = CSR_READ_4(sc, ALE_MDIO);
241 		if ((v & (MDIO_OP_EXECUTE | MDIO_OP_BUSY)) == 0)
242 			break;
243 	}
244 
245 	if (i == 0)
246 		device_printf(sc->ale_dev, "phy write timeout : %d\n", reg);
247 
248 	return (0);
249 }
250 
251 static void
ale_miibus_statchg(device_t dev)252 ale_miibus_statchg(device_t dev)
253 {
254 	struct ale_softc *sc;
255 	struct mii_data *mii;
256 	if_t ifp;
257 	uint32_t reg;
258 
259 	sc = device_get_softc(dev);
260 	mii = device_get_softc(sc->ale_miibus);
261 	ifp = sc->ale_ifp;
262 	if (mii == NULL || ifp == NULL ||
263 	    (if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0)
264 		return;
265 
266 	sc->ale_flags &= ~ALE_FLAG_LINK;
267 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
268 	    (IFM_ACTIVE | IFM_AVALID)) {
269 		switch (IFM_SUBTYPE(mii->mii_media_active)) {
270 		case IFM_10_T:
271 		case IFM_100_TX:
272 			sc->ale_flags |= ALE_FLAG_LINK;
273 			break;
274 		case IFM_1000_T:
275 			if ((sc->ale_flags & ALE_FLAG_FASTETHER) == 0)
276 				sc->ale_flags |= ALE_FLAG_LINK;
277 			break;
278 		default:
279 			break;
280 		}
281 	}
282 
283 	/* Stop Rx/Tx MACs. */
284 	ale_stop_mac(sc);
285 
286 	/* Program MACs with resolved speed/duplex/flow-control. */
287 	if ((sc->ale_flags & ALE_FLAG_LINK) != 0) {
288 		ale_mac_config(sc);
289 		/* Reenable Tx/Rx MACs. */
290 		reg = CSR_READ_4(sc, ALE_MAC_CFG);
291 		reg |= MAC_CFG_TX_ENB | MAC_CFG_RX_ENB;
292 		CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
293 	}
294 }
295 
296 static void
ale_mediastatus(if_t ifp,struct ifmediareq * ifmr)297 ale_mediastatus(if_t ifp, struct ifmediareq *ifmr)
298 {
299 	struct ale_softc *sc;
300 	struct mii_data *mii;
301 
302 	sc = if_getsoftc(ifp);
303 	ALE_LOCK(sc);
304 	if ((if_getflags(ifp) & IFF_UP) == 0) {
305 		ALE_UNLOCK(sc);
306 		return;
307 	}
308 	mii = device_get_softc(sc->ale_miibus);
309 
310 	mii_pollstat(mii);
311 	ifmr->ifm_status = mii->mii_media_status;
312 	ifmr->ifm_active = mii->mii_media_active;
313 	ALE_UNLOCK(sc);
314 }
315 
316 static int
ale_mediachange(if_t ifp)317 ale_mediachange(if_t ifp)
318 {
319 	struct ale_softc *sc;
320 	struct mii_data *mii;
321 	struct mii_softc *miisc;
322 	int error;
323 
324 	sc = if_getsoftc(ifp);
325 	ALE_LOCK(sc);
326 	mii = device_get_softc(sc->ale_miibus);
327 	LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
328 		PHY_RESET(miisc);
329 	error = mii_mediachg(mii);
330 	ALE_UNLOCK(sc);
331 
332 	return (error);
333 }
334 
335 static int
ale_probe(device_t dev)336 ale_probe(device_t dev)
337 {
338 	const struct ale_dev *sp;
339 	int i;
340 	uint16_t vendor, devid;
341 
342 	vendor = pci_get_vendor(dev);
343 	devid = pci_get_device(dev);
344 	sp = ale_devs;
345 	for (i = 0; i < nitems(ale_devs); i++) {
346 		if (vendor == sp->ale_vendorid &&
347 		    devid == sp->ale_deviceid) {
348 			device_set_desc(dev, sp->ale_name);
349 			return (BUS_PROBE_DEFAULT);
350 		}
351 		sp++;
352 	}
353 
354 	return (ENXIO);
355 }
356 
357 static void
ale_get_macaddr(struct ale_softc * sc)358 ale_get_macaddr(struct ale_softc *sc)
359 {
360 	uint32_t ea[2], reg;
361 	int i, vpdc;
362 
363 	reg = CSR_READ_4(sc, ALE_SPI_CTRL);
364 	if ((reg & SPI_VPD_ENB) != 0) {
365 		reg &= ~SPI_VPD_ENB;
366 		CSR_WRITE_4(sc, ALE_SPI_CTRL, reg);
367 	}
368 
369 	if (pci_find_cap(sc->ale_dev, PCIY_VPD, &vpdc) == 0) {
370 		/*
371 		 * PCI VPD capability found, let TWSI reload EEPROM.
372 		 * This will set ethernet address of controller.
373 		 */
374 		CSR_WRITE_4(sc, ALE_TWSI_CTRL, CSR_READ_4(sc, ALE_TWSI_CTRL) |
375 		    TWSI_CTRL_SW_LD_START);
376 		for (i = 100; i > 0; i--) {
377 			DELAY(1000);
378 			reg = CSR_READ_4(sc, ALE_TWSI_CTRL);
379 			if ((reg & TWSI_CTRL_SW_LD_START) == 0)
380 				break;
381 		}
382 		if (i == 0)
383 			device_printf(sc->ale_dev,
384 			    "reloading EEPROM timeout!\n");
385 	} else {
386 		if (bootverbose)
387 			device_printf(sc->ale_dev,
388 			    "PCI VPD capability not found!\n");
389 	}
390 
391 	ea[0] = CSR_READ_4(sc, ALE_PAR0);
392 	ea[1] = CSR_READ_4(sc, ALE_PAR1);
393 	sc->ale_eaddr[0] = (ea[1] >> 8) & 0xFF;
394 	sc->ale_eaddr[1] = (ea[1] >> 0) & 0xFF;
395 	sc->ale_eaddr[2] = (ea[0] >> 24) & 0xFF;
396 	sc->ale_eaddr[3] = (ea[0] >> 16) & 0xFF;
397 	sc->ale_eaddr[4] = (ea[0] >> 8) & 0xFF;
398 	sc->ale_eaddr[5] = (ea[0] >> 0) & 0xFF;
399 }
400 
401 static void
ale_phy_reset(struct ale_softc * sc)402 ale_phy_reset(struct ale_softc *sc)
403 {
404 
405 	/* Reset magic from Linux. */
406 	CSR_WRITE_2(sc, ALE_GPHY_CTRL,
407 	    GPHY_CTRL_HIB_EN | GPHY_CTRL_HIB_PULSE | GPHY_CTRL_SEL_ANA_RESET |
408 	    GPHY_CTRL_PHY_PLL_ON);
409 	DELAY(1000);
410 	CSR_WRITE_2(sc, ALE_GPHY_CTRL,
411 	    GPHY_CTRL_EXT_RESET | GPHY_CTRL_HIB_EN | GPHY_CTRL_HIB_PULSE |
412 	    GPHY_CTRL_SEL_ANA_RESET | GPHY_CTRL_PHY_PLL_ON);
413 	DELAY(1000);
414 
415 #define	ATPHY_DBG_ADDR		0x1D
416 #define	ATPHY_DBG_DATA		0x1E
417 
418 	/* Enable hibernation mode. */
419 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
420 	    ATPHY_DBG_ADDR, 0x0B);
421 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
422 	    ATPHY_DBG_DATA, 0xBC00);
423 	/* Set Class A/B for all modes. */
424 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
425 	    ATPHY_DBG_ADDR, 0x00);
426 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
427 	    ATPHY_DBG_DATA, 0x02EF);
428 	/* Enable 10BT power saving. */
429 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
430 	    ATPHY_DBG_ADDR, 0x12);
431 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
432 	    ATPHY_DBG_DATA, 0x4C04);
433 	/* Adjust 1000T power. */
434 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
435 	    ATPHY_DBG_ADDR, 0x04);
436 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
437 	    ATPHY_DBG_ADDR, 0x8BBB);
438 	/* 10BT center tap voltage. */
439 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
440 	    ATPHY_DBG_ADDR, 0x05);
441 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
442 	    ATPHY_DBG_ADDR, 0x2C46);
443 
444 #undef	ATPHY_DBG_ADDR
445 #undef	ATPHY_DBG_DATA
446 	DELAY(1000);
447 }
448 
449 static int
ale_attach(device_t dev)450 ale_attach(device_t dev)
451 {
452 	struct ale_softc *sc;
453 	if_t ifp;
454 	uint16_t burst;
455 	int error, i, msic, msixc, pmc;
456 	uint32_t rxf_len, txf_len;
457 
458 	error = 0;
459 	sc = device_get_softc(dev);
460 	sc->ale_dev = dev;
461 
462 	mtx_init(&sc->ale_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
463 	    MTX_DEF);
464 	callout_init_mtx(&sc->ale_tick_ch, &sc->ale_mtx, 0);
465 	NET_TASK_INIT(&sc->ale_int_task, 0, ale_int_task, sc);
466 
467 	/* Map the device. */
468 	pci_enable_busmaster(dev);
469 	sc->ale_res_spec = ale_res_spec_mem;
470 	sc->ale_irq_spec = ale_irq_spec_legacy;
471 	error = bus_alloc_resources(dev, sc->ale_res_spec, sc->ale_res);
472 	if (error != 0) {
473 		device_printf(dev, "cannot allocate memory resources.\n");
474 		goto fail;
475 	}
476 
477 	/* Set PHY address. */
478 	sc->ale_phyaddr = ALE_PHY_ADDR;
479 
480 	/* Reset PHY. */
481 	ale_phy_reset(sc);
482 
483 	/* Reset the ethernet controller. */
484 	ale_reset(sc);
485 
486 	/* Get PCI and chip id/revision. */
487 	sc->ale_rev = pci_get_revid(dev);
488 	if (sc->ale_rev >= 0xF0) {
489 		/* L2E Rev. B. AR8114 */
490 		sc->ale_flags |= ALE_FLAG_FASTETHER;
491 	} else {
492 		if ((CSR_READ_4(sc, ALE_PHY_STATUS) & PHY_STATUS_100M) != 0) {
493 			/* L1E AR8121 */
494 			sc->ale_flags |= ALE_FLAG_JUMBO;
495 		} else {
496 			/* L2E Rev. A. AR8113 */
497 			sc->ale_flags |= ALE_FLAG_FASTETHER;
498 		}
499 	}
500 	/*
501 	 * All known controllers seems to require 4 bytes alignment
502 	 * of Tx buffers to make Tx checksum offload with custom
503 	 * checksum generation method work.
504 	 */
505 	sc->ale_flags |= ALE_FLAG_TXCSUM_BUG;
506 	/*
507 	 * All known controllers seems to have issues on Rx checksum
508 	 * offload for fragmented IP datagrams.
509 	 */
510 	sc->ale_flags |= ALE_FLAG_RXCSUM_BUG;
511 	/*
512 	 * Don't use Tx CMB. It is known to cause RRS update failure
513 	 * under certain circumstances. Typical phenomenon of the
514 	 * issue would be unexpected sequence number encountered in
515 	 * Rx handler.
516 	 */
517 	sc->ale_flags |= ALE_FLAG_TXCMB_BUG;
518 	sc->ale_chip_rev = CSR_READ_4(sc, ALE_MASTER_CFG) >>
519 	    MASTER_CHIP_REV_SHIFT;
520 	if (bootverbose) {
521 		device_printf(dev, "PCI device revision : 0x%04x\n",
522 		    sc->ale_rev);
523 		device_printf(dev, "Chip id/revision : 0x%04x\n",
524 		    sc->ale_chip_rev);
525 	}
526 	txf_len = CSR_READ_4(sc, ALE_SRAM_TX_FIFO_LEN);
527 	rxf_len = CSR_READ_4(sc, ALE_SRAM_RX_FIFO_LEN);
528 	/*
529 	 * Uninitialized hardware returns an invalid chip id/revision
530 	 * as well as 0xFFFFFFFF for Tx/Rx fifo length.
531 	 */
532 	if (sc->ale_chip_rev == 0xFFFF || txf_len == 0xFFFFFFFF ||
533 	    rxf_len == 0xFFFFFFF) {
534 		device_printf(dev,"chip revision : 0x%04x, %u Tx FIFO "
535 		    "%u Rx FIFO -- not initialized?\n", sc->ale_chip_rev,
536 		    txf_len, rxf_len);
537 		error = ENXIO;
538 		goto fail;
539 	}
540 	device_printf(dev, "%u Tx FIFO, %u Rx FIFO\n", txf_len, rxf_len);
541 
542 	/* Allocate IRQ resources. */
543 	msixc = pci_msix_count(dev);
544 	msic = pci_msi_count(dev);
545 	if (bootverbose) {
546 		device_printf(dev, "MSIX count : %d\n", msixc);
547 		device_printf(dev, "MSI count : %d\n", msic);
548 	}
549 
550 	/* Prefer MSIX over MSI. */
551 	if (msix_disable == 0 || msi_disable == 0) {
552 		if (msix_disable == 0 && msixc == ALE_MSIX_MESSAGES &&
553 		    pci_alloc_msix(dev, &msixc) == 0) {
554 			if (msixc == ALE_MSIX_MESSAGES) {
555 				device_printf(dev, "Using %d MSIX messages.\n",
556 				    msixc);
557 				sc->ale_flags |= ALE_FLAG_MSIX;
558 				sc->ale_irq_spec = ale_irq_spec_msix;
559 			} else
560 				pci_release_msi(dev);
561 		}
562 		if (msi_disable == 0 && (sc->ale_flags & ALE_FLAG_MSIX) == 0 &&
563 		    msic == ALE_MSI_MESSAGES &&
564 		    pci_alloc_msi(dev, &msic) == 0) {
565 			if (msic == ALE_MSI_MESSAGES) {
566 				device_printf(dev, "Using %d MSI messages.\n",
567 				    msic);
568 				sc->ale_flags |= ALE_FLAG_MSI;
569 				sc->ale_irq_spec = ale_irq_spec_msi;
570 			} else
571 				pci_release_msi(dev);
572 		}
573 	}
574 
575 	error = bus_alloc_resources(dev, sc->ale_irq_spec, sc->ale_irq);
576 	if (error != 0) {
577 		device_printf(dev, "cannot allocate IRQ resources.\n");
578 		goto fail;
579 	}
580 
581 	/* Get DMA parameters from PCIe device control register. */
582 	if (pci_find_cap(dev, PCIY_EXPRESS, &i) == 0) {
583 		sc->ale_flags |= ALE_FLAG_PCIE;
584 		burst = pci_read_config(dev, i + 0x08, 2);
585 		/* Max read request size. */
586 		sc->ale_dma_rd_burst = ((burst >> 12) & 0x07) <<
587 		    DMA_CFG_RD_BURST_SHIFT;
588 		/* Max payload size. */
589 		sc->ale_dma_wr_burst = ((burst >> 5) & 0x07) <<
590 		    DMA_CFG_WR_BURST_SHIFT;
591 		if (bootverbose) {
592 			device_printf(dev, "Read request size : %d bytes.\n",
593 			    128 << ((burst >> 12) & 0x07));
594 			device_printf(dev, "TLP payload size : %d bytes.\n",
595 			    128 << ((burst >> 5) & 0x07));
596 		}
597 	} else {
598 		sc->ale_dma_rd_burst = DMA_CFG_RD_BURST_128;
599 		sc->ale_dma_wr_burst = DMA_CFG_WR_BURST_128;
600 	}
601 
602 	/* Create device sysctl node. */
603 	ale_sysctl_node(sc);
604 
605 	if ((error = ale_dma_alloc(sc)) != 0)
606 		goto fail;
607 
608 	/* Load station address. */
609 	ale_get_macaddr(sc);
610 
611 	ifp = sc->ale_ifp = if_alloc(IFT_ETHER);
612 	if_setsoftc(ifp, sc);
613 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
614 	if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
615 	if_setioctlfn(ifp, ale_ioctl);
616 	if_setstartfn(ifp, ale_start);
617 	if_setinitfn(ifp, ale_init);
618 	if_setsendqlen(ifp, ALE_TX_RING_CNT - 1);
619 	if_setsendqready(ifp);
620 	if_setcapabilities(ifp, IFCAP_RXCSUM | IFCAP_TXCSUM | IFCAP_TSO4);
621 	if_sethwassist(ifp, ALE_CSUM_FEATURES | CSUM_TSO);
622 	if (pci_find_cap(dev, PCIY_PMG, &pmc) == 0) {
623 		sc->ale_flags |= ALE_FLAG_PMCAP;
624 		if_setcapabilitiesbit(ifp, IFCAP_WOL_MAGIC | IFCAP_WOL_MCAST, 0);
625 	}
626 	if_setcapenable(ifp, if_getcapabilities(ifp));
627 
628 	/* Set up MII bus. */
629 	error = mii_attach(dev, &sc->ale_miibus, ifp, ale_mediachange,
630 	    ale_mediastatus, BMSR_DEFCAPMASK, sc->ale_phyaddr, MII_OFFSET_ANY,
631 	    MIIF_DOPAUSE);
632 	if (error != 0) {
633 		device_printf(dev, "attaching PHYs failed\n");
634 		goto fail;
635 	}
636 
637 	ether_ifattach(ifp, sc->ale_eaddr);
638 
639 	/* VLAN capability setup. */
640 	if_setcapabilitiesbit(ifp, IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING |
641 	    IFCAP_VLAN_HWCSUM | IFCAP_VLAN_HWTSO, 0);
642 	if_setcapenable(ifp, if_getcapabilities(ifp));
643 	/*
644 	 * Even though controllers supported by ale(3) have Rx checksum
645 	 * offload bug the workaround for fragmented frames seemed to
646 	 * work so far. However it seems Rx checksum offload does not
647 	 * work under certain conditions. So disable Rx checksum offload
648 	 * until I find more clue about it but allow users to override it.
649 	 */
650 	if_setcapenablebit(ifp, 0, IFCAP_RXCSUM);
651 
652 	/* Tell the upper layer(s) we support long frames. */
653 	if_setifheaderlen(ifp, sizeof(struct ether_vlan_header));
654 
655 	/* Create local taskq. */
656 	sc->ale_tq = taskqueue_create_fast("ale_taskq", M_WAITOK,
657 	    taskqueue_thread_enqueue, &sc->ale_tq);
658 	taskqueue_start_threads(&sc->ale_tq, 1, PI_NET, "%s taskq",
659 	    device_get_nameunit(sc->ale_dev));
660 
661 	if ((sc->ale_flags & ALE_FLAG_MSIX) != 0)
662 		msic = ALE_MSIX_MESSAGES;
663 	else if ((sc->ale_flags & ALE_FLAG_MSI) != 0)
664 		msic = ALE_MSI_MESSAGES;
665 	else
666 		msic = 1;
667 	for (i = 0; i < msic; i++) {
668 		error = bus_setup_intr(dev, sc->ale_irq[i],
669 		    INTR_TYPE_NET | INTR_MPSAFE, ale_intr, NULL, sc,
670 		    &sc->ale_intrhand[i]);
671 		if (error != 0)
672 			break;
673 	}
674 	if (error != 0) {
675 		device_printf(dev, "could not set up interrupt handler.\n");
676 		taskqueue_free(sc->ale_tq);
677 		sc->ale_tq = NULL;
678 		ether_ifdetach(ifp);
679 		goto fail;
680 	}
681 
682 fail:
683 	if (error != 0)
684 		ale_detach(dev);
685 
686 	return (error);
687 }
688 
689 static int
ale_detach(device_t dev)690 ale_detach(device_t dev)
691 {
692 	struct ale_softc *sc;
693 	if_t ifp;
694 	int i, msic;
695 
696 	sc = device_get_softc(dev);
697 
698 	ifp = sc->ale_ifp;
699 	if (device_is_attached(dev)) {
700 		ether_ifdetach(ifp);
701 		ALE_LOCK(sc);
702 		ale_stop(sc);
703 		ALE_UNLOCK(sc);
704 		callout_drain(&sc->ale_tick_ch);
705 		taskqueue_drain(sc->ale_tq, &sc->ale_int_task);
706 	}
707 
708 	if (sc->ale_tq != NULL) {
709 		taskqueue_drain(sc->ale_tq, &sc->ale_int_task);
710 		taskqueue_free(sc->ale_tq);
711 		sc->ale_tq = NULL;
712 	}
713 
714 	if (sc->ale_miibus != NULL) {
715 		device_delete_child(dev, sc->ale_miibus);
716 		sc->ale_miibus = NULL;
717 	}
718 	bus_generic_detach(dev);
719 	ale_dma_free(sc);
720 
721 	if (ifp != NULL) {
722 		if_free(ifp);
723 		sc->ale_ifp = NULL;
724 	}
725 
726 	if ((sc->ale_flags & ALE_FLAG_MSIX) != 0)
727 		msic = ALE_MSIX_MESSAGES;
728 	else if ((sc->ale_flags & ALE_FLAG_MSI) != 0)
729 		msic = ALE_MSI_MESSAGES;
730 	else
731 		msic = 1;
732 	for (i = 0; i < msic; i++) {
733 		if (sc->ale_intrhand[i] != NULL) {
734 			bus_teardown_intr(dev, sc->ale_irq[i],
735 			    sc->ale_intrhand[i]);
736 			sc->ale_intrhand[i] = NULL;
737 		}
738 	}
739 
740 	bus_release_resources(dev, sc->ale_irq_spec, sc->ale_irq);
741 	if ((sc->ale_flags & (ALE_FLAG_MSI | ALE_FLAG_MSIX)) != 0)
742 		pci_release_msi(dev);
743 	bus_release_resources(dev, sc->ale_res_spec, sc->ale_res);
744 	mtx_destroy(&sc->ale_mtx);
745 
746 	return (0);
747 }
748 
749 #define	ALE_SYSCTL_STAT_ADD32(c, h, n, p, d)	\
750 	    SYSCTL_ADD_UINT(c, h, OID_AUTO, n, CTLFLAG_RD, p, 0, d)
751 
752 #define	ALE_SYSCTL_STAT_ADD64(c, h, n, p, d)	\
753 	    SYSCTL_ADD_UQUAD(c, h, OID_AUTO, n, CTLFLAG_RD, p, d)
754 
755 static void
ale_sysctl_node(struct ale_softc * sc)756 ale_sysctl_node(struct ale_softc *sc)
757 {
758 	struct sysctl_ctx_list *ctx;
759 	struct sysctl_oid_list *child, *parent;
760 	struct sysctl_oid *tree;
761 	struct ale_hw_stats *stats;
762 	int error;
763 
764 	stats = &sc->ale_stats;
765 	ctx = device_get_sysctl_ctx(sc->ale_dev);
766 	child = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->ale_dev));
767 
768 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "int_rx_mod",
769 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &sc->ale_int_rx_mod,
770 	    0, sysctl_hw_ale_int_mod, "I", "ale Rx interrupt moderation");
771 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "int_tx_mod",
772 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &sc->ale_int_tx_mod,
773 	    0, sysctl_hw_ale_int_mod, "I", "ale Tx interrupt moderation");
774 	/* Pull in device tunables. */
775 	sc->ale_int_rx_mod = ALE_IM_RX_TIMER_DEFAULT;
776 	error = resource_int_value(device_get_name(sc->ale_dev),
777 	    device_get_unit(sc->ale_dev), "int_rx_mod", &sc->ale_int_rx_mod);
778 	if (error == 0) {
779 		if (sc->ale_int_rx_mod < ALE_IM_TIMER_MIN ||
780 		    sc->ale_int_rx_mod > ALE_IM_TIMER_MAX) {
781 			device_printf(sc->ale_dev, "int_rx_mod value out of "
782 			    "range; using default: %d\n",
783 			    ALE_IM_RX_TIMER_DEFAULT);
784 			sc->ale_int_rx_mod = ALE_IM_RX_TIMER_DEFAULT;
785 		}
786 	}
787 	sc->ale_int_tx_mod = ALE_IM_TX_TIMER_DEFAULT;
788 	error = resource_int_value(device_get_name(sc->ale_dev),
789 	    device_get_unit(sc->ale_dev), "int_tx_mod", &sc->ale_int_tx_mod);
790 	if (error == 0) {
791 		if (sc->ale_int_tx_mod < ALE_IM_TIMER_MIN ||
792 		    sc->ale_int_tx_mod > ALE_IM_TIMER_MAX) {
793 			device_printf(sc->ale_dev, "int_tx_mod value out of "
794 			    "range; using default: %d\n",
795 			    ALE_IM_TX_TIMER_DEFAULT);
796 			sc->ale_int_tx_mod = ALE_IM_TX_TIMER_DEFAULT;
797 		}
798 	}
799 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "process_limit",
800 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
801 	    &sc->ale_process_limit, 0, sysctl_hw_ale_proc_limit, "I",
802 	    "max number of Rx events to process");
803 	/* Pull in device tunables. */
804 	sc->ale_process_limit = ALE_PROC_DEFAULT;
805 	error = resource_int_value(device_get_name(sc->ale_dev),
806 	    device_get_unit(sc->ale_dev), "process_limit",
807 	    &sc->ale_process_limit);
808 	if (error == 0) {
809 		if (sc->ale_process_limit < ALE_PROC_MIN ||
810 		    sc->ale_process_limit > ALE_PROC_MAX) {
811 			device_printf(sc->ale_dev,
812 			    "process_limit value out of range; "
813 			    "using default: %d\n", ALE_PROC_DEFAULT);
814 			sc->ale_process_limit = ALE_PROC_DEFAULT;
815 		}
816 	}
817 
818 	/* Misc statistics. */
819 	ALE_SYSCTL_STAT_ADD32(ctx, child, "reset_brk_seq",
820 	    &stats->reset_brk_seq,
821 	    "Controller resets due to broken Rx sequnce number");
822 
823 	tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "stats",
824 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "ATE statistics");
825 	parent = SYSCTL_CHILDREN(tree);
826 
827 	/* Rx statistics. */
828 	tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "rx",
829 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Rx MAC statistics");
830 	child = SYSCTL_CHILDREN(tree);
831 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_frames",
832 	    &stats->rx_frames, "Good frames");
833 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_bcast_frames",
834 	    &stats->rx_bcast_frames, "Good broadcast frames");
835 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_mcast_frames",
836 	    &stats->rx_mcast_frames, "Good multicast frames");
837 	ALE_SYSCTL_STAT_ADD32(ctx, child, "pause_frames",
838 	    &stats->rx_pause_frames, "Pause control frames");
839 	ALE_SYSCTL_STAT_ADD32(ctx, child, "control_frames",
840 	    &stats->rx_control_frames, "Control frames");
841 	ALE_SYSCTL_STAT_ADD32(ctx, child, "crc_errs",
842 	    &stats->rx_crcerrs, "CRC errors");
843 	ALE_SYSCTL_STAT_ADD32(ctx, child, "len_errs",
844 	    &stats->rx_lenerrs, "Frames with length mismatched");
845 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_octets",
846 	    &stats->rx_bytes, "Good octets");
847 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_bcast_octets",
848 	    &stats->rx_bcast_bytes, "Good broadcast octets");
849 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_mcast_octets",
850 	    &stats->rx_mcast_bytes, "Good multicast octets");
851 	ALE_SYSCTL_STAT_ADD32(ctx, child, "runts",
852 	    &stats->rx_runts, "Too short frames");
853 	ALE_SYSCTL_STAT_ADD32(ctx, child, "fragments",
854 	    &stats->rx_fragments, "Fragmented frames");
855 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_64",
856 	    &stats->rx_pkts_64, "64 bytes frames");
857 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_65_127",
858 	    &stats->rx_pkts_65_127, "65 to 127 bytes frames");
859 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_128_255",
860 	    &stats->rx_pkts_128_255, "128 to 255 bytes frames");
861 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_256_511",
862 	    &stats->rx_pkts_256_511, "256 to 511 bytes frames");
863 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_512_1023",
864 	    &stats->rx_pkts_512_1023, "512 to 1023 bytes frames");
865 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1024_1518",
866 	    &stats->rx_pkts_1024_1518, "1024 to 1518 bytes frames");
867 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1519_max",
868 	    &stats->rx_pkts_1519_max, "1519 to max frames");
869 	ALE_SYSCTL_STAT_ADD32(ctx, child, "trunc_errs",
870 	    &stats->rx_pkts_truncated, "Truncated frames due to MTU size");
871 	ALE_SYSCTL_STAT_ADD32(ctx, child, "fifo_oflows",
872 	    &stats->rx_fifo_oflows, "FIFO overflows");
873 	ALE_SYSCTL_STAT_ADD32(ctx, child, "rrs_errs",
874 	    &stats->rx_rrs_errs, "Return status write-back errors");
875 	ALE_SYSCTL_STAT_ADD32(ctx, child, "align_errs",
876 	    &stats->rx_alignerrs, "Alignment errors");
877 	ALE_SYSCTL_STAT_ADD32(ctx, child, "filtered",
878 	    &stats->rx_pkts_filtered,
879 	    "Frames dropped due to address filtering");
880 
881 	/* Tx statistics. */
882 	tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "tx",
883 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Tx MAC statistics");
884 	child = SYSCTL_CHILDREN(tree);
885 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_frames",
886 	    &stats->tx_frames, "Good frames");
887 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_bcast_frames",
888 	    &stats->tx_bcast_frames, "Good broadcast frames");
889 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_mcast_frames",
890 	    &stats->tx_mcast_frames, "Good multicast frames");
891 	ALE_SYSCTL_STAT_ADD32(ctx, child, "pause_frames",
892 	    &stats->tx_pause_frames, "Pause control frames");
893 	ALE_SYSCTL_STAT_ADD32(ctx, child, "control_frames",
894 	    &stats->tx_control_frames, "Control frames");
895 	ALE_SYSCTL_STAT_ADD32(ctx, child, "excess_defers",
896 	    &stats->tx_excess_defer, "Frames with excessive derferrals");
897 	ALE_SYSCTL_STAT_ADD32(ctx, child, "defers",
898 	    &stats->tx_excess_defer, "Frames with derferrals");
899 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_octets",
900 	    &stats->tx_bytes, "Good octets");
901 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_bcast_octets",
902 	    &stats->tx_bcast_bytes, "Good broadcast octets");
903 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_mcast_octets",
904 	    &stats->tx_mcast_bytes, "Good multicast octets");
905 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_64",
906 	    &stats->tx_pkts_64, "64 bytes frames");
907 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_65_127",
908 	    &stats->tx_pkts_65_127, "65 to 127 bytes frames");
909 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_128_255",
910 	    &stats->tx_pkts_128_255, "128 to 255 bytes frames");
911 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_256_511",
912 	    &stats->tx_pkts_256_511, "256 to 511 bytes frames");
913 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_512_1023",
914 	    &stats->tx_pkts_512_1023, "512 to 1023 bytes frames");
915 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1024_1518",
916 	    &stats->tx_pkts_1024_1518, "1024 to 1518 bytes frames");
917 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1519_max",
918 	    &stats->tx_pkts_1519_max, "1519 to max frames");
919 	ALE_SYSCTL_STAT_ADD32(ctx, child, "single_colls",
920 	    &stats->tx_single_colls, "Single collisions");
921 	ALE_SYSCTL_STAT_ADD32(ctx, child, "multi_colls",
922 	    &stats->tx_multi_colls, "Multiple collisions");
923 	ALE_SYSCTL_STAT_ADD32(ctx, child, "late_colls",
924 	    &stats->tx_late_colls, "Late collisions");
925 	ALE_SYSCTL_STAT_ADD32(ctx, child, "excess_colls",
926 	    &stats->tx_excess_colls, "Excessive collisions");
927 	ALE_SYSCTL_STAT_ADD32(ctx, child, "underruns",
928 	    &stats->tx_underrun, "FIFO underruns");
929 	ALE_SYSCTL_STAT_ADD32(ctx, child, "desc_underruns",
930 	    &stats->tx_desc_underrun, "Descriptor write-back errors");
931 	ALE_SYSCTL_STAT_ADD32(ctx, child, "len_errs",
932 	    &stats->tx_lenerrs, "Frames with length mismatched");
933 	ALE_SYSCTL_STAT_ADD32(ctx, child, "trunc_errs",
934 	    &stats->tx_pkts_truncated, "Truncated frames due to MTU size");
935 }
936 
937 #undef ALE_SYSCTL_STAT_ADD32
938 #undef ALE_SYSCTL_STAT_ADD64
939 
940 struct ale_dmamap_arg {
941 	bus_addr_t	ale_busaddr;
942 };
943 
944 static void
ale_dmamap_cb(void * arg,bus_dma_segment_t * segs,int nsegs,int error)945 ale_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error)
946 {
947 	struct ale_dmamap_arg *ctx;
948 
949 	if (error != 0)
950 		return;
951 
952 	KASSERT(nsegs == 1, ("%s: %d segments returned!", __func__, nsegs));
953 
954 	ctx = (struct ale_dmamap_arg *)arg;
955 	ctx->ale_busaddr = segs[0].ds_addr;
956 }
957 
958 /*
959  * Tx descriptors/RXF0/CMB DMA blocks share ALE_DESC_ADDR_HI register
960  * which specifies high address region of DMA blocks. Therefore these
961  * blocks should have the same high address of given 4GB address
962  * space(i.e. crossing 4GB boundary is not allowed).
963  */
964 static int
ale_check_boundary(struct ale_softc * sc)965 ale_check_boundary(struct ale_softc *sc)
966 {
967 	bus_addr_t rx_cmb_end[ALE_RX_PAGES], tx_cmb_end;
968 	bus_addr_t rx_page_end[ALE_RX_PAGES], tx_ring_end;
969 
970 	rx_page_end[0] = sc->ale_cdata.ale_rx_page[0].page_paddr +
971 	    sc->ale_pagesize;
972 	rx_page_end[1] = sc->ale_cdata.ale_rx_page[1].page_paddr +
973 	    sc->ale_pagesize;
974 	tx_ring_end = sc->ale_cdata.ale_tx_ring_paddr + ALE_TX_RING_SZ;
975 	tx_cmb_end = sc->ale_cdata.ale_tx_cmb_paddr + ALE_TX_CMB_SZ;
976 	rx_cmb_end[0] = sc->ale_cdata.ale_rx_page[0].cmb_paddr + ALE_RX_CMB_SZ;
977 	rx_cmb_end[1] = sc->ale_cdata.ale_rx_page[1].cmb_paddr + ALE_RX_CMB_SZ;
978 
979 	if ((ALE_ADDR_HI(tx_ring_end) !=
980 	    ALE_ADDR_HI(sc->ale_cdata.ale_tx_ring_paddr)) ||
981 	    (ALE_ADDR_HI(rx_page_end[0]) !=
982 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[0].page_paddr)) ||
983 	    (ALE_ADDR_HI(rx_page_end[1]) !=
984 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[1].page_paddr)) ||
985 	    (ALE_ADDR_HI(tx_cmb_end) !=
986 	    ALE_ADDR_HI(sc->ale_cdata.ale_tx_cmb_paddr)) ||
987 	    (ALE_ADDR_HI(rx_cmb_end[0]) !=
988 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[0].cmb_paddr)) ||
989 	    (ALE_ADDR_HI(rx_cmb_end[1]) !=
990 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[1].cmb_paddr)))
991 		return (EFBIG);
992 
993 	if ((ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_page_end[0])) ||
994 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_page_end[1])) ||
995 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_cmb_end[0])) ||
996 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_cmb_end[1])) ||
997 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(tx_cmb_end)))
998 		return (EFBIG);
999 
1000 	return (0);
1001 }
1002 
1003 static int
ale_dma_alloc(struct ale_softc * sc)1004 ale_dma_alloc(struct ale_softc *sc)
1005 {
1006 	struct ale_txdesc *txd;
1007 	bus_addr_t lowaddr;
1008 	struct ale_dmamap_arg ctx;
1009 	int error, guard_size, i;
1010 
1011 	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0)
1012 		guard_size = ALE_JUMBO_FRAMELEN;
1013 	else
1014 		guard_size = ALE_MAX_FRAMELEN;
1015 	sc->ale_pagesize = roundup(guard_size + ALE_RX_PAGE_SZ,
1016 	    ALE_RX_PAGE_ALIGN);
1017 	lowaddr = BUS_SPACE_MAXADDR;
1018 again:
1019 	/* Create parent DMA tag. */
1020 	error = bus_dma_tag_create(
1021 	    bus_get_dma_tag(sc->ale_dev), /* parent */
1022 	    1, 0,			/* alignment, boundary */
1023 	    lowaddr,			/* lowaddr */
1024 	    BUS_SPACE_MAXADDR,		/* highaddr */
1025 	    NULL, NULL,			/* filter, filterarg */
1026 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsize */
1027 	    0,				/* nsegments */
1028 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsegsize */
1029 	    0,				/* flags */
1030 	    NULL, NULL,			/* lockfunc, lockarg */
1031 	    &sc->ale_cdata.ale_parent_tag);
1032 	if (error != 0) {
1033 		device_printf(sc->ale_dev,
1034 		    "could not create parent DMA tag.\n");
1035 		goto fail;
1036 	}
1037 
1038 	/* Create DMA tag for Tx descriptor ring. */
1039 	error = bus_dma_tag_create(
1040 	    sc->ale_cdata.ale_parent_tag, /* parent */
1041 	    ALE_TX_RING_ALIGN, 0,	/* alignment, boundary */
1042 	    BUS_SPACE_MAXADDR,		/* lowaddr */
1043 	    BUS_SPACE_MAXADDR,		/* highaddr */
1044 	    NULL, NULL,			/* filter, filterarg */
1045 	    ALE_TX_RING_SZ,		/* maxsize */
1046 	    1,				/* nsegments */
1047 	    ALE_TX_RING_SZ,		/* maxsegsize */
1048 	    0,				/* flags */
1049 	    NULL, NULL,			/* lockfunc, lockarg */
1050 	    &sc->ale_cdata.ale_tx_ring_tag);
1051 	if (error != 0) {
1052 		device_printf(sc->ale_dev,
1053 		    "could not create Tx ring DMA tag.\n");
1054 		goto fail;
1055 	}
1056 
1057 	/* Create DMA tag for Rx pages. */
1058 	for (i = 0; i < ALE_RX_PAGES; i++) {
1059 		error = bus_dma_tag_create(
1060 		    sc->ale_cdata.ale_parent_tag, /* parent */
1061 		    ALE_RX_PAGE_ALIGN, 0,	/* alignment, boundary */
1062 		    BUS_SPACE_MAXADDR,		/* lowaddr */
1063 		    BUS_SPACE_MAXADDR,		/* highaddr */
1064 		    NULL, NULL,			/* filter, filterarg */
1065 		    sc->ale_pagesize,		/* maxsize */
1066 		    1,				/* nsegments */
1067 		    sc->ale_pagesize,		/* maxsegsize */
1068 		    0,				/* flags */
1069 		    NULL, NULL,			/* lockfunc, lockarg */
1070 		    &sc->ale_cdata.ale_rx_page[i].page_tag);
1071 		if (error != 0) {
1072 			device_printf(sc->ale_dev,
1073 			    "could not create Rx page %d DMA tag.\n", i);
1074 			goto fail;
1075 		}
1076 	}
1077 
1078 	/* Create DMA tag for Tx coalescing message block. */
1079 	error = bus_dma_tag_create(
1080 	    sc->ale_cdata.ale_parent_tag, /* parent */
1081 	    ALE_CMB_ALIGN, 0,		/* alignment, boundary */
1082 	    BUS_SPACE_MAXADDR,		/* lowaddr */
1083 	    BUS_SPACE_MAXADDR,		/* highaddr */
1084 	    NULL, NULL,			/* filter, filterarg */
1085 	    ALE_TX_CMB_SZ,		/* maxsize */
1086 	    1,				/* nsegments */
1087 	    ALE_TX_CMB_SZ,		/* maxsegsize */
1088 	    0,				/* flags */
1089 	    NULL, NULL,			/* lockfunc, lockarg */
1090 	    &sc->ale_cdata.ale_tx_cmb_tag);
1091 	if (error != 0) {
1092 		device_printf(sc->ale_dev,
1093 		    "could not create Tx CMB DMA tag.\n");
1094 		goto fail;
1095 	}
1096 
1097 	/* Create DMA tag for Rx coalescing message block. */
1098 	for (i = 0; i < ALE_RX_PAGES; i++) {
1099 		error = bus_dma_tag_create(
1100 		    sc->ale_cdata.ale_parent_tag, /* parent */
1101 		    ALE_CMB_ALIGN, 0,		/* alignment, boundary */
1102 		    BUS_SPACE_MAXADDR,		/* lowaddr */
1103 		    BUS_SPACE_MAXADDR,		/* highaddr */
1104 		    NULL, NULL,			/* filter, filterarg */
1105 		    ALE_RX_CMB_SZ,		/* maxsize */
1106 		    1,				/* nsegments */
1107 		    ALE_RX_CMB_SZ,		/* maxsegsize */
1108 		    0,				/* flags */
1109 		    NULL, NULL,			/* lockfunc, lockarg */
1110 		    &sc->ale_cdata.ale_rx_page[i].cmb_tag);
1111 		if (error != 0) {
1112 			device_printf(sc->ale_dev,
1113 			    "could not create Rx page %d CMB DMA tag.\n", i);
1114 			goto fail;
1115 		}
1116 	}
1117 
1118 	/* Allocate DMA'able memory and load the DMA map for Tx ring. */
1119 	error = bus_dmamem_alloc(sc->ale_cdata.ale_tx_ring_tag,
1120 	    (void **)&sc->ale_cdata.ale_tx_ring,
1121 	    BUS_DMA_WAITOK | BUS_DMA_ZERO | BUS_DMA_COHERENT,
1122 	    &sc->ale_cdata.ale_tx_ring_map);
1123 	if (error != 0) {
1124 		device_printf(sc->ale_dev,
1125 		    "could not allocate DMA'able memory for Tx ring.\n");
1126 		goto fail;
1127 	}
1128 	ctx.ale_busaddr = 0;
1129 	error = bus_dmamap_load(sc->ale_cdata.ale_tx_ring_tag,
1130 	    sc->ale_cdata.ale_tx_ring_map, sc->ale_cdata.ale_tx_ring,
1131 	    ALE_TX_RING_SZ, ale_dmamap_cb, &ctx, 0);
1132 	if (error != 0 || ctx.ale_busaddr == 0) {
1133 		device_printf(sc->ale_dev,
1134 		    "could not load DMA'able memory for Tx ring.\n");
1135 		goto fail;
1136 	}
1137 	sc->ale_cdata.ale_tx_ring_paddr = ctx.ale_busaddr;
1138 
1139 	/* Rx pages. */
1140 	for (i = 0; i < ALE_RX_PAGES; i++) {
1141 		error = bus_dmamem_alloc(sc->ale_cdata.ale_rx_page[i].page_tag,
1142 		    (void **)&sc->ale_cdata.ale_rx_page[i].page_addr,
1143 		    BUS_DMA_WAITOK | BUS_DMA_ZERO | BUS_DMA_COHERENT,
1144 		    &sc->ale_cdata.ale_rx_page[i].page_map);
1145 		if (error != 0) {
1146 			device_printf(sc->ale_dev,
1147 			    "could not allocate DMA'able memory for "
1148 			    "Rx page %d.\n", i);
1149 			goto fail;
1150 		}
1151 		ctx.ale_busaddr = 0;
1152 		error = bus_dmamap_load(sc->ale_cdata.ale_rx_page[i].page_tag,
1153 		    sc->ale_cdata.ale_rx_page[i].page_map,
1154 		    sc->ale_cdata.ale_rx_page[i].page_addr,
1155 		    sc->ale_pagesize, ale_dmamap_cb, &ctx, 0);
1156 		if (error != 0 || ctx.ale_busaddr == 0) {
1157 			device_printf(sc->ale_dev,
1158 			    "could not load DMA'able memory for "
1159 			    "Rx page %d.\n", i);
1160 			goto fail;
1161 		}
1162 		sc->ale_cdata.ale_rx_page[i].page_paddr = ctx.ale_busaddr;
1163 	}
1164 
1165 	/* Tx CMB. */
1166 	error = bus_dmamem_alloc(sc->ale_cdata.ale_tx_cmb_tag,
1167 	    (void **)&sc->ale_cdata.ale_tx_cmb,
1168 	    BUS_DMA_WAITOK | BUS_DMA_ZERO | BUS_DMA_COHERENT,
1169 	    &sc->ale_cdata.ale_tx_cmb_map);
1170 	if (error != 0) {
1171 		device_printf(sc->ale_dev,
1172 		    "could not allocate DMA'able memory for Tx CMB.\n");
1173 		goto fail;
1174 	}
1175 	ctx.ale_busaddr = 0;
1176 	error = bus_dmamap_load(sc->ale_cdata.ale_tx_cmb_tag,
1177 	    sc->ale_cdata.ale_tx_cmb_map, sc->ale_cdata.ale_tx_cmb,
1178 	    ALE_TX_CMB_SZ, ale_dmamap_cb, &ctx, 0);
1179 	if (error != 0 || ctx.ale_busaddr == 0) {
1180 		device_printf(sc->ale_dev,
1181 		    "could not load DMA'able memory for Tx CMB.\n");
1182 		goto fail;
1183 	}
1184 	sc->ale_cdata.ale_tx_cmb_paddr = ctx.ale_busaddr;
1185 
1186 	/* Rx CMB. */
1187 	for (i = 0; i < ALE_RX_PAGES; i++) {
1188 		error = bus_dmamem_alloc(sc->ale_cdata.ale_rx_page[i].cmb_tag,
1189 		    (void **)&sc->ale_cdata.ale_rx_page[i].cmb_addr,
1190 		    BUS_DMA_WAITOK | BUS_DMA_ZERO | BUS_DMA_COHERENT,
1191 		    &sc->ale_cdata.ale_rx_page[i].cmb_map);
1192 		if (error != 0) {
1193 			device_printf(sc->ale_dev, "could not allocate "
1194 			    "DMA'able memory for Rx page %d CMB.\n", i);
1195 			goto fail;
1196 		}
1197 		ctx.ale_busaddr = 0;
1198 		error = bus_dmamap_load(sc->ale_cdata.ale_rx_page[i].cmb_tag,
1199 		    sc->ale_cdata.ale_rx_page[i].cmb_map,
1200 		    sc->ale_cdata.ale_rx_page[i].cmb_addr,
1201 		    ALE_RX_CMB_SZ, ale_dmamap_cb, &ctx, 0);
1202 		if (error != 0 || ctx.ale_busaddr == 0) {
1203 			device_printf(sc->ale_dev, "could not load DMA'able "
1204 			    "memory for Rx page %d CMB.\n", i);
1205 			goto fail;
1206 		}
1207 		sc->ale_cdata.ale_rx_page[i].cmb_paddr = ctx.ale_busaddr;
1208 	}
1209 
1210 	/*
1211 	 * Tx descriptors/RXF0/CMB DMA blocks share the same
1212 	 * high address region of 64bit DMA address space.
1213 	 */
1214 	if (lowaddr != BUS_SPACE_MAXADDR_32BIT &&
1215 	    (error = ale_check_boundary(sc)) != 0) {
1216 		device_printf(sc->ale_dev, "4GB boundary crossed, "
1217 		    "switching to 32bit DMA addressing mode.\n");
1218 		ale_dma_free(sc);
1219 		/*
1220 		 * Limit max allowable DMA address space to 32bit
1221 		 * and try again.
1222 		 */
1223 		lowaddr = BUS_SPACE_MAXADDR_32BIT;
1224 		goto again;
1225 	}
1226 
1227 	/*
1228 	 * Create Tx buffer parent tag.
1229 	 * AR81xx allows 64bit DMA addressing of Tx buffers so it
1230 	 * needs separate parent DMA tag as parent DMA address space
1231 	 * could be restricted to be within 32bit address space by
1232 	 * 4GB boundary crossing.
1233 	 */
1234 	error = bus_dma_tag_create(
1235 	    bus_get_dma_tag(sc->ale_dev), /* parent */
1236 	    1, 0,			/* alignment, boundary */
1237 	    BUS_SPACE_MAXADDR,		/* lowaddr */
1238 	    BUS_SPACE_MAXADDR,		/* highaddr */
1239 	    NULL, NULL,			/* filter, filterarg */
1240 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsize */
1241 	    0,				/* nsegments */
1242 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsegsize */
1243 	    0,				/* flags */
1244 	    NULL, NULL,			/* lockfunc, lockarg */
1245 	    &sc->ale_cdata.ale_buffer_tag);
1246 	if (error != 0) {
1247 		device_printf(sc->ale_dev,
1248 		    "could not create parent buffer DMA tag.\n");
1249 		goto fail;
1250 	}
1251 
1252 	/* Create DMA tag for Tx buffers. */
1253 	error = bus_dma_tag_create(
1254 	    sc->ale_cdata.ale_buffer_tag, /* parent */
1255 	    1, 0,			/* alignment, boundary */
1256 	    BUS_SPACE_MAXADDR,		/* lowaddr */
1257 	    BUS_SPACE_MAXADDR,		/* highaddr */
1258 	    NULL, NULL,			/* filter, filterarg */
1259 	    ALE_TSO_MAXSIZE,		/* maxsize */
1260 	    ALE_MAXTXSEGS,		/* nsegments */
1261 	    ALE_TSO_MAXSEGSIZE,		/* maxsegsize */
1262 	    0,				/* flags */
1263 	    NULL, NULL,			/* lockfunc, lockarg */
1264 	    &sc->ale_cdata.ale_tx_tag);
1265 	if (error != 0) {
1266 		device_printf(sc->ale_dev, "could not create Tx DMA tag.\n");
1267 		goto fail;
1268 	}
1269 
1270 	/* Create DMA maps for Tx buffers. */
1271 	for (i = 0; i < ALE_TX_RING_CNT; i++) {
1272 		txd = &sc->ale_cdata.ale_txdesc[i];
1273 		txd->tx_m = NULL;
1274 		txd->tx_dmamap = NULL;
1275 		error = bus_dmamap_create(sc->ale_cdata.ale_tx_tag, 0,
1276 		    &txd->tx_dmamap);
1277 		if (error != 0) {
1278 			device_printf(sc->ale_dev,
1279 			    "could not create Tx dmamap.\n");
1280 			goto fail;
1281 		}
1282 	}
1283 
1284 fail:
1285 	return (error);
1286 }
1287 
1288 static void
ale_dma_free(struct ale_softc * sc)1289 ale_dma_free(struct ale_softc *sc)
1290 {
1291 	struct ale_txdesc *txd;
1292 	int i;
1293 
1294 	/* Tx buffers. */
1295 	if (sc->ale_cdata.ale_tx_tag != NULL) {
1296 		for (i = 0; i < ALE_TX_RING_CNT; i++) {
1297 			txd = &sc->ale_cdata.ale_txdesc[i];
1298 			if (txd->tx_dmamap != NULL) {
1299 				bus_dmamap_destroy(sc->ale_cdata.ale_tx_tag,
1300 				    txd->tx_dmamap);
1301 				txd->tx_dmamap = NULL;
1302 			}
1303 		}
1304 		bus_dma_tag_destroy(sc->ale_cdata.ale_tx_tag);
1305 		sc->ale_cdata.ale_tx_tag = NULL;
1306 	}
1307 	/* Tx descriptor ring. */
1308 	if (sc->ale_cdata.ale_tx_ring_tag != NULL) {
1309 		if (sc->ale_cdata.ale_tx_ring_paddr != 0)
1310 			bus_dmamap_unload(sc->ale_cdata.ale_tx_ring_tag,
1311 			    sc->ale_cdata.ale_tx_ring_map);
1312 		if (sc->ale_cdata.ale_tx_ring != NULL)
1313 			bus_dmamem_free(sc->ale_cdata.ale_tx_ring_tag,
1314 			    sc->ale_cdata.ale_tx_ring,
1315 			    sc->ale_cdata.ale_tx_ring_map);
1316 		sc->ale_cdata.ale_tx_ring_paddr = 0;
1317 		sc->ale_cdata.ale_tx_ring = NULL;
1318 		bus_dma_tag_destroy(sc->ale_cdata.ale_tx_ring_tag);
1319 		sc->ale_cdata.ale_tx_ring_tag = NULL;
1320 	}
1321 	/* Rx page block. */
1322 	for (i = 0; i < ALE_RX_PAGES; i++) {
1323 		if (sc->ale_cdata.ale_rx_page[i].page_tag != NULL) {
1324 			if (sc->ale_cdata.ale_rx_page[i].page_paddr != 0)
1325 				bus_dmamap_unload(
1326 				    sc->ale_cdata.ale_rx_page[i].page_tag,
1327 				    sc->ale_cdata.ale_rx_page[i].page_map);
1328 			if (sc->ale_cdata.ale_rx_page[i].page_addr != NULL)
1329 				bus_dmamem_free(
1330 				    sc->ale_cdata.ale_rx_page[i].page_tag,
1331 				    sc->ale_cdata.ale_rx_page[i].page_addr,
1332 				    sc->ale_cdata.ale_rx_page[i].page_map);
1333 			sc->ale_cdata.ale_rx_page[i].page_paddr = 0;
1334 			sc->ale_cdata.ale_rx_page[i].page_addr = NULL;
1335 			bus_dma_tag_destroy(
1336 			    sc->ale_cdata.ale_rx_page[i].page_tag);
1337 			sc->ale_cdata.ale_rx_page[i].page_tag = NULL;
1338 		}
1339 	}
1340 	/* Rx CMB. */
1341 	for (i = 0; i < ALE_RX_PAGES; i++) {
1342 		if (sc->ale_cdata.ale_rx_page[i].cmb_tag != NULL) {
1343 			if (sc->ale_cdata.ale_rx_page[i].cmb_paddr != 0)
1344 				bus_dmamap_unload(
1345 				    sc->ale_cdata.ale_rx_page[i].cmb_tag,
1346 				    sc->ale_cdata.ale_rx_page[i].cmb_map);
1347 			if (sc->ale_cdata.ale_rx_page[i].cmb_addr != NULL)
1348 				bus_dmamem_free(
1349 				    sc->ale_cdata.ale_rx_page[i].cmb_tag,
1350 				    sc->ale_cdata.ale_rx_page[i].cmb_addr,
1351 				    sc->ale_cdata.ale_rx_page[i].cmb_map);
1352 			sc->ale_cdata.ale_rx_page[i].cmb_paddr = 0;
1353 			sc->ale_cdata.ale_rx_page[i].cmb_addr = NULL;
1354 			bus_dma_tag_destroy(
1355 			    sc->ale_cdata.ale_rx_page[i].cmb_tag);
1356 			sc->ale_cdata.ale_rx_page[i].cmb_tag = NULL;
1357 		}
1358 	}
1359 	/* Tx CMB. */
1360 	if (sc->ale_cdata.ale_tx_cmb_tag != NULL) {
1361 		if (sc->ale_cdata.ale_tx_cmb_paddr != 0)
1362 			bus_dmamap_unload(sc->ale_cdata.ale_tx_cmb_tag,
1363 			    sc->ale_cdata.ale_tx_cmb_map);
1364 		if (sc->ale_cdata.ale_tx_cmb != NULL)
1365 			bus_dmamem_free(sc->ale_cdata.ale_tx_cmb_tag,
1366 			    sc->ale_cdata.ale_tx_cmb,
1367 			    sc->ale_cdata.ale_tx_cmb_map);
1368 		sc->ale_cdata.ale_tx_cmb_paddr = 0;
1369 		sc->ale_cdata.ale_tx_cmb = NULL;
1370 		bus_dma_tag_destroy(sc->ale_cdata.ale_tx_cmb_tag);
1371 		sc->ale_cdata.ale_tx_cmb_tag = NULL;
1372 	}
1373 	if (sc->ale_cdata.ale_buffer_tag != NULL) {
1374 		bus_dma_tag_destroy(sc->ale_cdata.ale_buffer_tag);
1375 		sc->ale_cdata.ale_buffer_tag = NULL;
1376 	}
1377 	if (sc->ale_cdata.ale_parent_tag != NULL) {
1378 		bus_dma_tag_destroy(sc->ale_cdata.ale_parent_tag);
1379 		sc->ale_cdata.ale_parent_tag = NULL;
1380 	}
1381 }
1382 
1383 static int
ale_shutdown(device_t dev)1384 ale_shutdown(device_t dev)
1385 {
1386 
1387 	return (ale_suspend(dev));
1388 }
1389 
1390 /*
1391  * Note, this driver resets the link speed to 10/100Mbps by
1392  * restarting auto-negotiation in suspend/shutdown phase but we
1393  * don't know whether that auto-negotiation would succeed or not
1394  * as driver has no control after powering off/suspend operation.
1395  * If the renegotiation fail WOL may not work. Running at 1Gbps
1396  * will draw more power than 375mA at 3.3V which is specified in
1397  * PCI specification and that would result in complete
1398  * shutdowning power to ethernet controller.
1399  *
1400  * TODO
1401  * Save current negotiated media speed/duplex/flow-control to
1402  * softc and restore the same link again after resuming. PHY
1403  * handling such as power down/resetting to 100Mbps may be better
1404  * handled in suspend method in phy driver.
1405  */
1406 static void
ale_setlinkspeed(struct ale_softc * sc)1407 ale_setlinkspeed(struct ale_softc *sc)
1408 {
1409 	struct mii_data *mii;
1410 	int aneg, i;
1411 
1412 	mii = device_get_softc(sc->ale_miibus);
1413 	mii_pollstat(mii);
1414 	aneg = 0;
1415 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
1416 	    (IFM_ACTIVE | IFM_AVALID)) {
1417 		switch IFM_SUBTYPE(mii->mii_media_active) {
1418 		case IFM_10_T:
1419 		case IFM_100_TX:
1420 			return;
1421 		case IFM_1000_T:
1422 			aneg++;
1423 			break;
1424 		default:
1425 			break;
1426 		}
1427 	}
1428 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr, MII_100T2CR, 0);
1429 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
1430 	    MII_ANAR, ANAR_TX_FD | ANAR_TX | ANAR_10_FD | ANAR_10 | ANAR_CSMA);
1431 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
1432 	    MII_BMCR, BMCR_RESET | BMCR_AUTOEN | BMCR_STARTNEG);
1433 	DELAY(1000);
1434 	if (aneg != 0) {
1435 		/*
1436 		 * Poll link state until ale(4) get a 10/100Mbps link.
1437 		 */
1438 		for (i = 0; i < MII_ANEGTICKS_GIGE; i++) {
1439 			mii_pollstat(mii);
1440 			if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID))
1441 			    == (IFM_ACTIVE | IFM_AVALID)) {
1442 				switch (IFM_SUBTYPE(
1443 				    mii->mii_media_active)) {
1444 				case IFM_10_T:
1445 				case IFM_100_TX:
1446 					ale_mac_config(sc);
1447 					return;
1448 				default:
1449 					break;
1450 				}
1451 			}
1452 			ALE_UNLOCK(sc);
1453 			pause("alelnk", hz);
1454 			ALE_LOCK(sc);
1455 		}
1456 		if (i == MII_ANEGTICKS_GIGE)
1457 			device_printf(sc->ale_dev,
1458 			    "establishing a link failed, WOL may not work!");
1459 	}
1460 	/*
1461 	 * No link, force MAC to have 100Mbps, full-duplex link.
1462 	 * This is the last resort and may/may not work.
1463 	 */
1464 	mii->mii_media_status = IFM_AVALID | IFM_ACTIVE;
1465 	mii->mii_media_active = IFM_ETHER | IFM_100_TX | IFM_FDX;
1466 	ale_mac_config(sc);
1467 }
1468 
1469 static void
ale_setwol(struct ale_softc * sc)1470 ale_setwol(struct ale_softc *sc)
1471 {
1472 	if_t ifp;
1473 	uint32_t reg, pmcs;
1474 	uint16_t pmstat;
1475 	int pmc;
1476 
1477 	ALE_LOCK_ASSERT(sc);
1478 
1479 	if (pci_find_cap(sc->ale_dev, PCIY_PMG, &pmc) != 0) {
1480 		/* Disable WOL. */
1481 		CSR_WRITE_4(sc, ALE_WOL_CFG, 0);
1482 		reg = CSR_READ_4(sc, ALE_PCIE_PHYMISC);
1483 		reg |= PCIE_PHYMISC_FORCE_RCV_DET;
1484 		CSR_WRITE_4(sc, ALE_PCIE_PHYMISC, reg);
1485 		/* Force PHY power down. */
1486 		CSR_WRITE_2(sc, ALE_GPHY_CTRL,
1487 		    GPHY_CTRL_EXT_RESET | GPHY_CTRL_HIB_EN |
1488 		    GPHY_CTRL_HIB_PULSE | GPHY_CTRL_PHY_PLL_ON |
1489 		    GPHY_CTRL_SEL_ANA_RESET | GPHY_CTRL_PHY_IDDQ |
1490 		    GPHY_CTRL_PCLK_SEL_DIS | GPHY_CTRL_PWDOWN_HW);
1491 		return;
1492 	}
1493 
1494 	ifp = sc->ale_ifp;
1495 	if ((if_getcapenable(ifp) & IFCAP_WOL) != 0) {
1496 		if ((sc->ale_flags & ALE_FLAG_FASTETHER) == 0)
1497 			ale_setlinkspeed(sc);
1498 	}
1499 
1500 	pmcs = 0;
1501 	if ((if_getcapenable(ifp) & IFCAP_WOL_MAGIC) != 0)
1502 		pmcs |= WOL_CFG_MAGIC | WOL_CFG_MAGIC_ENB;
1503 	CSR_WRITE_4(sc, ALE_WOL_CFG, pmcs);
1504 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
1505 	reg &= ~(MAC_CFG_DBG | MAC_CFG_PROMISC | MAC_CFG_ALLMULTI |
1506 	    MAC_CFG_BCAST);
1507 	if ((if_getcapenable(ifp) & IFCAP_WOL_MCAST) != 0)
1508 		reg |= MAC_CFG_ALLMULTI | MAC_CFG_BCAST;
1509 	if ((if_getcapenable(ifp) & IFCAP_WOL) != 0)
1510 		reg |= MAC_CFG_RX_ENB;
1511 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
1512 
1513 	if ((if_getcapenable(ifp) & IFCAP_WOL) == 0) {
1514 		/* WOL disabled, PHY power down. */
1515 		reg = CSR_READ_4(sc, ALE_PCIE_PHYMISC);
1516 		reg |= PCIE_PHYMISC_FORCE_RCV_DET;
1517 		CSR_WRITE_4(sc, ALE_PCIE_PHYMISC, reg);
1518 		CSR_WRITE_2(sc, ALE_GPHY_CTRL,
1519 		    GPHY_CTRL_EXT_RESET | GPHY_CTRL_HIB_EN |
1520 		    GPHY_CTRL_HIB_PULSE | GPHY_CTRL_SEL_ANA_RESET |
1521 		    GPHY_CTRL_PHY_IDDQ | GPHY_CTRL_PCLK_SEL_DIS |
1522 		    GPHY_CTRL_PWDOWN_HW);
1523 	}
1524 	/* Request PME. */
1525 	pmstat = pci_read_config(sc->ale_dev, pmc + PCIR_POWER_STATUS, 2);
1526 	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
1527 	if ((if_getcapenable(ifp) & IFCAP_WOL) != 0)
1528 		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
1529 	pci_write_config(sc->ale_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
1530 }
1531 
1532 static int
ale_suspend(device_t dev)1533 ale_suspend(device_t dev)
1534 {
1535 	struct ale_softc *sc;
1536 
1537 	sc = device_get_softc(dev);
1538 
1539 	ALE_LOCK(sc);
1540 	ale_stop(sc);
1541 	ale_setwol(sc);
1542 	ALE_UNLOCK(sc);
1543 
1544 	return (0);
1545 }
1546 
1547 static int
ale_resume(device_t dev)1548 ale_resume(device_t dev)
1549 {
1550 	struct ale_softc *sc;
1551 	if_t ifp;
1552 	int pmc;
1553 	uint16_t pmstat;
1554 
1555 	sc = device_get_softc(dev);
1556 
1557 	ALE_LOCK(sc);
1558 	if (pci_find_cap(sc->ale_dev, PCIY_PMG, &pmc) == 0) {
1559 		/* Disable PME and clear PME status. */
1560 		pmstat = pci_read_config(sc->ale_dev,
1561 		    pmc + PCIR_POWER_STATUS, 2);
1562 		if ((pmstat & PCIM_PSTAT_PMEENABLE) != 0) {
1563 			pmstat &= ~PCIM_PSTAT_PMEENABLE;
1564 			pci_write_config(sc->ale_dev,
1565 			    pmc + PCIR_POWER_STATUS, pmstat, 2);
1566 		}
1567 	}
1568 	/* Reset PHY. */
1569 	ale_phy_reset(sc);
1570 	ifp = sc->ale_ifp;
1571 	if ((if_getflags(ifp) & IFF_UP) != 0) {
1572 		if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
1573 		ale_init_locked(sc);
1574 	}
1575 	ALE_UNLOCK(sc);
1576 
1577 	return (0);
1578 }
1579 
1580 static int
ale_encap(struct ale_softc * sc,struct mbuf ** m_head)1581 ale_encap(struct ale_softc *sc, struct mbuf **m_head)
1582 {
1583 	struct ale_txdesc *txd, *txd_last;
1584 	struct tx_desc *desc;
1585 	struct mbuf *m;
1586 	struct ip *ip;
1587 	struct tcphdr *tcp;
1588 	bus_dma_segment_t txsegs[ALE_MAXTXSEGS];
1589 	bus_dmamap_t map;
1590 	uint32_t cflags, hdrlen, ip_off, poff, vtag;
1591 	int error, i, nsegs, prod, si;
1592 
1593 	ALE_LOCK_ASSERT(sc);
1594 
1595 	M_ASSERTPKTHDR((*m_head));
1596 
1597 	m = *m_head;
1598 	ip = NULL;
1599 	tcp = NULL;
1600 	cflags = vtag = 0;
1601 	ip_off = poff = 0;
1602 	if ((m->m_pkthdr.csum_flags & (ALE_CSUM_FEATURES | CSUM_TSO)) != 0) {
1603 		/*
1604 		 * AR81xx requires offset of TCP/UDP payload in its Tx
1605 		 * descriptor to perform hardware Tx checksum offload.
1606 		 * Additionally, TSO requires IP/TCP header size and
1607 		 * modification of IP/TCP header in order to make TSO
1608 		 * engine work. This kind of operation takes many CPU
1609 		 * cycles on FreeBSD so fast host CPU is required to
1610 		 * get smooth TSO performance.
1611 		 */
1612 		struct ether_header *eh;
1613 
1614 		if (M_WRITABLE(m) == 0) {
1615 			/* Get a writable copy. */
1616 			m = m_dup(*m_head, M_NOWAIT);
1617 			/* Release original mbufs. */
1618 			m_freem(*m_head);
1619 			if (m == NULL) {
1620 				*m_head = NULL;
1621 				return (ENOBUFS);
1622 			}
1623 			*m_head = m;
1624 		}
1625 
1626 		/*
1627 		 * Buggy-controller requires 4 byte aligned Tx buffer
1628 		 * to make custom checksum offload work.
1629 		 */
1630 		if ((sc->ale_flags & ALE_FLAG_TXCSUM_BUG) != 0 &&
1631 		    (m->m_pkthdr.csum_flags & ALE_CSUM_FEATURES) != 0 &&
1632 		    (mtod(m, intptr_t) & 3) != 0) {
1633 			m = m_defrag(*m_head, M_NOWAIT);
1634 			if (m == NULL) {
1635 				m_freem(*m_head);
1636 				*m_head = NULL;
1637 				return (ENOBUFS);
1638 			}
1639 			*m_head = m;
1640 		}
1641 
1642 		ip_off = sizeof(struct ether_header);
1643 		m = m_pullup(m, ip_off);
1644 		if (m == NULL) {
1645 			*m_head = NULL;
1646 			return (ENOBUFS);
1647 		}
1648 		eh = mtod(m, struct ether_header *);
1649 		/*
1650 		 * Check if hardware VLAN insertion is off.
1651 		 * Additional check for LLC/SNAP frame?
1652 		 */
1653 		if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
1654 			ip_off = sizeof(struct ether_vlan_header);
1655 			m = m_pullup(m, ip_off);
1656 			if (m == NULL) {
1657 				*m_head = NULL;
1658 				return (ENOBUFS);
1659 			}
1660 		}
1661 		m = m_pullup(m, ip_off + sizeof(struct ip));
1662 		if (m == NULL) {
1663 			*m_head = NULL;
1664 			return (ENOBUFS);
1665 		}
1666 		ip = (struct ip *)(mtod(m, char *) + ip_off);
1667 		poff = ip_off + (ip->ip_hl << 2);
1668 		if ((m->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
1669 			/*
1670 			 * XXX
1671 			 * AR81xx requires the first descriptor should
1672 			 * not include any TCP playload for TSO case.
1673 			 * (i.e. ethernet header + IP + TCP header only)
1674 			 * m_pullup(9) above will ensure this too.
1675 			 * However it's not correct if the first mbuf
1676 			 * of the chain does not use cluster.
1677 			 */
1678 			m = m_pullup(m, poff + sizeof(struct tcphdr));
1679 			if (m == NULL) {
1680 				*m_head = NULL;
1681 				return (ENOBUFS);
1682 			}
1683 			ip = (struct ip *)(mtod(m, char *) + ip_off);
1684 			tcp = (struct tcphdr *)(mtod(m, char *) + poff);
1685 			m = m_pullup(m, poff + (tcp->th_off << 2));
1686 			if (m == NULL) {
1687 				*m_head = NULL;
1688 				return (ENOBUFS);
1689 			}
1690 			/*
1691 			 * AR81xx requires IP/TCP header size and offset as
1692 			 * well as TCP pseudo checksum which complicates
1693 			 * TSO configuration. I guess this comes from the
1694 			 * adherence to Microsoft NDIS Large Send
1695 			 * specification which requires insertion of
1696 			 * pseudo checksum by upper stack. The pseudo
1697 			 * checksum that NDIS refers to doesn't include
1698 			 * TCP payload length so ale(4) should recompute
1699 			 * the pseudo checksum here. Hopefully this wouldn't
1700 			 * be much burden on modern CPUs.
1701 			 * Reset IP checksum and recompute TCP pseudo
1702 			 * checksum as NDIS specification said.
1703 			 */
1704 			ip->ip_sum = 0;
1705 			tcp->th_sum = in_pseudo(ip->ip_src.s_addr,
1706 			    ip->ip_dst.s_addr, htons(IPPROTO_TCP));
1707 		}
1708 		*m_head = m;
1709 	}
1710 
1711 	si = prod = sc->ale_cdata.ale_tx_prod;
1712 	txd = &sc->ale_cdata.ale_txdesc[prod];
1713 	txd_last = txd;
1714 	map = txd->tx_dmamap;
1715 
1716 	error =  bus_dmamap_load_mbuf_sg(sc->ale_cdata.ale_tx_tag, map,
1717 	    *m_head, txsegs, &nsegs, 0);
1718 	if (error == EFBIG) {
1719 		m = m_collapse(*m_head, M_NOWAIT, ALE_MAXTXSEGS);
1720 		if (m == NULL) {
1721 			m_freem(*m_head);
1722 			*m_head = NULL;
1723 			return (ENOMEM);
1724 		}
1725 		*m_head = m;
1726 		error = bus_dmamap_load_mbuf_sg(sc->ale_cdata.ale_tx_tag, map,
1727 		    *m_head, txsegs, &nsegs, 0);
1728 		if (error != 0) {
1729 			m_freem(*m_head);
1730 			*m_head = NULL;
1731 			return (error);
1732 		}
1733 	} else if (error != 0)
1734 		return (error);
1735 	if (nsegs == 0) {
1736 		m_freem(*m_head);
1737 		*m_head = NULL;
1738 		return (EIO);
1739 	}
1740 
1741 	/* Check descriptor overrun. */
1742 	if (sc->ale_cdata.ale_tx_cnt + nsegs >= ALE_TX_RING_CNT - 3) {
1743 		bus_dmamap_unload(sc->ale_cdata.ale_tx_tag, map);
1744 		return (ENOBUFS);
1745 	}
1746 	bus_dmamap_sync(sc->ale_cdata.ale_tx_tag, map, BUS_DMASYNC_PREWRITE);
1747 
1748 	m = *m_head;
1749 	if ((m->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
1750 		/* Request TSO and set MSS. */
1751 		cflags |= ALE_TD_TSO;
1752 		cflags |= ((uint32_t)m->m_pkthdr.tso_segsz << ALE_TD_MSS_SHIFT);
1753 		/* Set IP/TCP header size. */
1754 		cflags |= ip->ip_hl << ALE_TD_IPHDR_LEN_SHIFT;
1755 		cflags |= tcp->th_off << ALE_TD_TCPHDR_LEN_SHIFT;
1756 	} else if ((m->m_pkthdr.csum_flags & ALE_CSUM_FEATURES) != 0) {
1757 		/*
1758 		 * AR81xx supports Tx custom checksum offload feature
1759 		 * that offloads single 16bit checksum computation.
1760 		 * So you can choose one among IP, TCP and UDP.
1761 		 * Normally driver sets checksum start/insertion
1762 		 * position from the information of TCP/UDP frame as
1763 		 * TCP/UDP checksum takes more time than that of IP.
1764 		 * However it seems that custom checksum offload
1765 		 * requires 4 bytes aligned Tx buffers due to hardware
1766 		 * bug.
1767 		 * AR81xx also supports explicit Tx checksum computation
1768 		 * if it is told that the size of IP header and TCP
1769 		 * header(for UDP, the header size does not matter
1770 		 * because it's fixed length). However with this scheme
1771 		 * TSO does not work so you have to choose one either
1772 		 * TSO or explicit Tx checksum offload. I chosen TSO
1773 		 * plus custom checksum offload with work-around which
1774 		 * will cover most common usage for this consumer
1775 		 * ethernet controller. The work-around takes a lot of
1776 		 * CPU cycles if Tx buffer is not aligned on 4 bytes
1777 		 * boundary, though.
1778 		 */
1779 		cflags |= ALE_TD_CXSUM;
1780 		/* Set checksum start offset. */
1781 		cflags |= (poff << ALE_TD_CSUM_PLOADOFFSET_SHIFT);
1782 		/* Set checksum insertion position of TCP/UDP. */
1783 		cflags |= ((poff + m->m_pkthdr.csum_data) <<
1784 		    ALE_TD_CSUM_XSUMOFFSET_SHIFT);
1785 	}
1786 
1787 	/* Configure VLAN hardware tag insertion. */
1788 	if ((m->m_flags & M_VLANTAG) != 0) {
1789 		vtag = ALE_TX_VLAN_TAG(m->m_pkthdr.ether_vtag);
1790 		vtag = ((vtag << ALE_TD_VLAN_SHIFT) & ALE_TD_VLAN_MASK);
1791 		cflags |= ALE_TD_INSERT_VLAN_TAG;
1792 	}
1793 
1794 	i = 0;
1795 	if ((m->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
1796 		/*
1797 		 * Make sure the first fragment contains
1798 		 * only ethernet and IP/TCP header with options.
1799 		 */
1800 		hdrlen =  poff + (tcp->th_off << 2);
1801 		desc = &sc->ale_cdata.ale_tx_ring[prod];
1802 		desc->addr = htole64(txsegs[i].ds_addr);
1803 		desc->len = htole32(ALE_TX_BYTES(hdrlen) | vtag);
1804 		desc->flags = htole32(cflags);
1805 		sc->ale_cdata.ale_tx_cnt++;
1806 		ALE_DESC_INC(prod, ALE_TX_RING_CNT);
1807 		if (m->m_len - hdrlen > 0) {
1808 			/* Handle remaining payload of the first fragment. */
1809 			desc = &sc->ale_cdata.ale_tx_ring[prod];
1810 			desc->addr = htole64(txsegs[i].ds_addr + hdrlen);
1811 			desc->len = htole32(ALE_TX_BYTES(m->m_len - hdrlen) |
1812 			    vtag);
1813 			desc->flags = htole32(cflags);
1814 			sc->ale_cdata.ale_tx_cnt++;
1815 			ALE_DESC_INC(prod, ALE_TX_RING_CNT);
1816 		}
1817 		i = 1;
1818 	}
1819 	for (; i < nsegs; i++) {
1820 		desc = &sc->ale_cdata.ale_tx_ring[prod];
1821 		desc->addr = htole64(txsegs[i].ds_addr);
1822 		desc->len = htole32(ALE_TX_BYTES(txsegs[i].ds_len) | vtag);
1823 		desc->flags = htole32(cflags);
1824 		sc->ale_cdata.ale_tx_cnt++;
1825 		ALE_DESC_INC(prod, ALE_TX_RING_CNT);
1826 	}
1827 	/* Update producer index. */
1828 	sc->ale_cdata.ale_tx_prod = prod;
1829 	/* Set TSO header on the first descriptor. */
1830 	if ((m->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
1831 		desc = &sc->ale_cdata.ale_tx_ring[si];
1832 		desc->flags |= htole32(ALE_TD_TSO_HDR);
1833 	}
1834 
1835 	/* Finally set EOP on the last descriptor. */
1836 	prod = (prod + ALE_TX_RING_CNT - 1) % ALE_TX_RING_CNT;
1837 	desc = &sc->ale_cdata.ale_tx_ring[prod];
1838 	desc->flags |= htole32(ALE_TD_EOP);
1839 
1840 	/* Swap dmamap of the first and the last. */
1841 	txd = &sc->ale_cdata.ale_txdesc[prod];
1842 	map = txd_last->tx_dmamap;
1843 	txd_last->tx_dmamap = txd->tx_dmamap;
1844 	txd->tx_dmamap = map;
1845 	txd->tx_m = m;
1846 
1847 	/* Sync descriptors. */
1848 	bus_dmamap_sync(sc->ale_cdata.ale_tx_ring_tag,
1849 	    sc->ale_cdata.ale_tx_ring_map,
1850 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1851 
1852 	return (0);
1853 }
1854 
1855 static void
ale_start(if_t ifp)1856 ale_start(if_t ifp)
1857 {
1858         struct ale_softc *sc;
1859 
1860 	sc = if_getsoftc(ifp);
1861 	ALE_LOCK(sc);
1862 	ale_start_locked(ifp);
1863 	ALE_UNLOCK(sc);
1864 }
1865 
1866 static void
ale_start_locked(if_t ifp)1867 ale_start_locked(if_t ifp)
1868 {
1869         struct ale_softc *sc;
1870         struct mbuf *m_head;
1871 	int enq;
1872 
1873 	sc = if_getsoftc(ifp);
1874 
1875 	ALE_LOCK_ASSERT(sc);
1876 
1877 	/* Reclaim transmitted frames. */
1878 	if (sc->ale_cdata.ale_tx_cnt >= ALE_TX_DESC_HIWAT)
1879 		ale_txeof(sc);
1880 
1881 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1882 	    IFF_DRV_RUNNING || (sc->ale_flags & ALE_FLAG_LINK) == 0)
1883 		return;
1884 
1885 	for (enq = 0; !if_sendq_empty(ifp); ) {
1886 		m_head = if_dequeue(ifp);
1887 		if (m_head == NULL)
1888 			break;
1889 		/*
1890 		 * Pack the data into the transmit ring. If we
1891 		 * don't have room, set the OACTIVE flag and wait
1892 		 * for the NIC to drain the ring.
1893 		 */
1894 		if (ale_encap(sc, &m_head)) {
1895 			if (m_head == NULL)
1896 				break;
1897 			if_sendq_prepend(ifp, m_head);
1898 			if_setdrvflagbits(ifp, IFF_DRV_OACTIVE, 0);
1899 			break;
1900 		}
1901 
1902 		enq++;
1903 		/*
1904 		 * If there's a BPF listener, bounce a copy of this frame
1905 		 * to him.
1906 		 */
1907 		ETHER_BPF_MTAP(ifp, m_head);
1908 	}
1909 
1910 	if (enq > 0) {
1911 		/* Kick. */
1912 		CSR_WRITE_4(sc, ALE_MBOX_TPD_PROD_IDX,
1913 		    sc->ale_cdata.ale_tx_prod);
1914 		/* Set a timeout in case the chip goes out to lunch. */
1915 		sc->ale_watchdog_timer = ALE_TX_TIMEOUT;
1916 	}
1917 }
1918 
1919 static void
ale_watchdog(struct ale_softc * sc)1920 ale_watchdog(struct ale_softc *sc)
1921 {
1922 	if_t ifp;
1923 
1924 	ALE_LOCK_ASSERT(sc);
1925 
1926 	if (sc->ale_watchdog_timer == 0 || --sc->ale_watchdog_timer)
1927 		return;
1928 
1929 	ifp = sc->ale_ifp;
1930 	if ((sc->ale_flags & ALE_FLAG_LINK) == 0) {
1931 		if_printf(sc->ale_ifp, "watchdog timeout (lost link)\n");
1932 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1933 		if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
1934 		ale_init_locked(sc);
1935 		return;
1936 	}
1937 	if_printf(sc->ale_ifp, "watchdog timeout -- resetting\n");
1938 	if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1939 	if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
1940 	ale_init_locked(sc);
1941 	if (!if_sendq_empty(ifp))
1942 		ale_start_locked(ifp);
1943 }
1944 
1945 static int
ale_ioctl(if_t ifp,u_long cmd,caddr_t data)1946 ale_ioctl(if_t ifp, u_long cmd, caddr_t data)
1947 {
1948 	struct ale_softc *sc;
1949 	struct ifreq *ifr;
1950 	struct mii_data *mii;
1951 	int error, mask;
1952 
1953 	sc = if_getsoftc(ifp);
1954 	ifr = (struct ifreq *)data;
1955 	error = 0;
1956 	switch (cmd) {
1957 	case SIOCSIFMTU:
1958 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > ALE_JUMBO_MTU ||
1959 		    ((sc->ale_flags & ALE_FLAG_JUMBO) == 0 &&
1960 		    ifr->ifr_mtu > ETHERMTU))
1961 			error = EINVAL;
1962 		else if (if_getmtu(ifp) != ifr->ifr_mtu) {
1963 			ALE_LOCK(sc);
1964 			if_setmtu(ifp, ifr->ifr_mtu);
1965 			if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0) {
1966 				if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
1967 				ale_init_locked(sc);
1968 			}
1969 			ALE_UNLOCK(sc);
1970 		}
1971 		break;
1972 	case SIOCSIFFLAGS:
1973 		ALE_LOCK(sc);
1974 		if ((if_getflags(ifp) & IFF_UP) != 0) {
1975 			if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0) {
1976 				if (((if_getflags(ifp) ^ sc->ale_if_flags)
1977 				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
1978 					ale_rxfilter(sc);
1979 			} else {
1980 				ale_init_locked(sc);
1981 			}
1982 		} else {
1983 			if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0)
1984 				ale_stop(sc);
1985 		}
1986 		sc->ale_if_flags = if_getflags(ifp);
1987 		ALE_UNLOCK(sc);
1988 		break;
1989 	case SIOCADDMULTI:
1990 	case SIOCDELMULTI:
1991 		ALE_LOCK(sc);
1992 		if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0)
1993 			ale_rxfilter(sc);
1994 		ALE_UNLOCK(sc);
1995 		break;
1996 	case SIOCSIFMEDIA:
1997 	case SIOCGIFMEDIA:
1998 		mii = device_get_softc(sc->ale_miibus);
1999 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, cmd);
2000 		break;
2001 	case SIOCSIFCAP:
2002 		ALE_LOCK(sc);
2003 		mask = ifr->ifr_reqcap ^ if_getcapenable(ifp);
2004 		if ((mask & IFCAP_TXCSUM) != 0 &&
2005 		    (if_getcapabilities(ifp) & IFCAP_TXCSUM) != 0) {
2006 			if_togglecapenable(ifp, IFCAP_TXCSUM);
2007 			if ((if_getcapenable(ifp) & IFCAP_TXCSUM) != 0)
2008 				if_sethwassistbits(ifp, ALE_CSUM_FEATURES, 0);
2009 			else
2010 				if_sethwassistbits(ifp, 0, ALE_CSUM_FEATURES);
2011 		}
2012 		if ((mask & IFCAP_RXCSUM) != 0 &&
2013 		    (if_getcapabilities(ifp) & IFCAP_RXCSUM) != 0)
2014 			if_togglecapenable(ifp, IFCAP_RXCSUM);
2015 		if ((mask & IFCAP_TSO4) != 0 &&
2016 		    (if_getcapabilities(ifp) & IFCAP_TSO4) != 0) {
2017 			if_togglecapenable(ifp, IFCAP_TSO4);
2018 			if ((if_getcapenable(ifp) & IFCAP_TSO4) != 0)
2019 				if_sethwassistbits(ifp, CSUM_TSO, 0);
2020 			else
2021 				if_sethwassistbits(ifp, 0, CSUM_TSO);
2022 		}
2023 
2024 		if ((mask & IFCAP_WOL_MCAST) != 0 &&
2025 		    (if_getcapabilities(ifp) & IFCAP_WOL_MCAST) != 0)
2026 			if_togglecapenable(ifp, IFCAP_WOL_MCAST);
2027 		if ((mask & IFCAP_WOL_MAGIC) != 0 &&
2028 		    (if_getcapabilities(ifp) & IFCAP_WOL_MAGIC) != 0)
2029 			if_togglecapenable(ifp, IFCAP_WOL_MAGIC);
2030 		if ((mask & IFCAP_VLAN_HWCSUM) != 0 &&
2031 		    (if_getcapabilities(ifp) & IFCAP_VLAN_HWCSUM) != 0)
2032 			if_togglecapenable(ifp, IFCAP_VLAN_HWCSUM);
2033 		if ((mask & IFCAP_VLAN_HWTSO) != 0 &&
2034 		    (if_getcapabilities(ifp) & IFCAP_VLAN_HWTSO) != 0)
2035 			if_togglecapenable(ifp, IFCAP_VLAN_HWTSO);
2036 		if ((mask & IFCAP_VLAN_HWTAGGING) != 0 &&
2037 		    (if_getcapabilities(ifp) & IFCAP_VLAN_HWTAGGING) != 0) {
2038 			if_togglecapenable(ifp, IFCAP_VLAN_HWTAGGING);
2039 			if ((if_getcapenable(ifp) & IFCAP_VLAN_HWTAGGING) == 0)
2040 				if_setcapenablebit(ifp, 0, IFCAP_VLAN_HWTSO);
2041 			ale_rxvlan(sc);
2042 		}
2043 		ALE_UNLOCK(sc);
2044 		VLAN_CAPABILITIES(ifp);
2045 		break;
2046 	default:
2047 		error = ether_ioctl(ifp, cmd, data);
2048 		break;
2049 	}
2050 
2051 	return (error);
2052 }
2053 
2054 static void
ale_mac_config(struct ale_softc * sc)2055 ale_mac_config(struct ale_softc *sc)
2056 {
2057 	struct mii_data *mii;
2058 	uint32_t reg;
2059 
2060 	ALE_LOCK_ASSERT(sc);
2061 
2062 	mii = device_get_softc(sc->ale_miibus);
2063 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
2064 	reg &= ~(MAC_CFG_FULL_DUPLEX | MAC_CFG_TX_FC | MAC_CFG_RX_FC |
2065 	    MAC_CFG_SPEED_MASK);
2066 	/* Reprogram MAC with resolved speed/duplex. */
2067 	switch (IFM_SUBTYPE(mii->mii_media_active)) {
2068 	case IFM_10_T:
2069 	case IFM_100_TX:
2070 		reg |= MAC_CFG_SPEED_10_100;
2071 		break;
2072 	case IFM_1000_T:
2073 		reg |= MAC_CFG_SPEED_1000;
2074 		break;
2075 	}
2076 	if ((IFM_OPTIONS(mii->mii_media_active) & IFM_FDX) != 0) {
2077 		reg |= MAC_CFG_FULL_DUPLEX;
2078 		if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_TXPAUSE) != 0)
2079 			reg |= MAC_CFG_TX_FC;
2080 		if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_RXPAUSE) != 0)
2081 			reg |= MAC_CFG_RX_FC;
2082 	}
2083 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2084 }
2085 
2086 static void
ale_stats_clear(struct ale_softc * sc)2087 ale_stats_clear(struct ale_softc *sc)
2088 {
2089 	struct smb sb;
2090 	uint32_t *reg;
2091 	int i;
2092 
2093 	for (reg = &sb.rx_frames, i = 0; reg <= &sb.rx_pkts_filtered; reg++) {
2094 		CSR_READ_4(sc, ALE_RX_MIB_BASE + i);
2095 		i += sizeof(uint32_t);
2096 	}
2097 	/* Read Tx statistics. */
2098 	for (reg = &sb.tx_frames, i = 0; reg <= &sb.tx_mcast_bytes; reg++) {
2099 		CSR_READ_4(sc, ALE_TX_MIB_BASE + i);
2100 		i += sizeof(uint32_t);
2101 	}
2102 }
2103 
2104 static void
ale_stats_update(struct ale_softc * sc)2105 ale_stats_update(struct ale_softc *sc)
2106 {
2107 	struct ale_hw_stats *stat;
2108 	struct smb sb, *smb;
2109 	if_t ifp;
2110 	uint32_t *reg;
2111 	int i;
2112 
2113 	ALE_LOCK_ASSERT(sc);
2114 
2115 	ifp = sc->ale_ifp;
2116 	stat = &sc->ale_stats;
2117 	smb = &sb;
2118 
2119 	/* Read Rx statistics. */
2120 	for (reg = &sb.rx_frames, i = 0; reg <= &sb.rx_pkts_filtered; reg++) {
2121 		*reg = CSR_READ_4(sc, ALE_RX_MIB_BASE + i);
2122 		i += sizeof(uint32_t);
2123 	}
2124 	/* Read Tx statistics. */
2125 	for (reg = &sb.tx_frames, i = 0; reg <= &sb.tx_mcast_bytes; reg++) {
2126 		*reg = CSR_READ_4(sc, ALE_TX_MIB_BASE + i);
2127 		i += sizeof(uint32_t);
2128 	}
2129 
2130 	/* Rx stats. */
2131 	stat->rx_frames += smb->rx_frames;
2132 	stat->rx_bcast_frames += smb->rx_bcast_frames;
2133 	stat->rx_mcast_frames += smb->rx_mcast_frames;
2134 	stat->rx_pause_frames += smb->rx_pause_frames;
2135 	stat->rx_control_frames += smb->rx_control_frames;
2136 	stat->rx_crcerrs += smb->rx_crcerrs;
2137 	stat->rx_lenerrs += smb->rx_lenerrs;
2138 	stat->rx_bytes += smb->rx_bytes;
2139 	stat->rx_runts += smb->rx_runts;
2140 	stat->rx_fragments += smb->rx_fragments;
2141 	stat->rx_pkts_64 += smb->rx_pkts_64;
2142 	stat->rx_pkts_65_127 += smb->rx_pkts_65_127;
2143 	stat->rx_pkts_128_255 += smb->rx_pkts_128_255;
2144 	stat->rx_pkts_256_511 += smb->rx_pkts_256_511;
2145 	stat->rx_pkts_512_1023 += smb->rx_pkts_512_1023;
2146 	stat->rx_pkts_1024_1518 += smb->rx_pkts_1024_1518;
2147 	stat->rx_pkts_1519_max += smb->rx_pkts_1519_max;
2148 	stat->rx_pkts_truncated += smb->rx_pkts_truncated;
2149 	stat->rx_fifo_oflows += smb->rx_fifo_oflows;
2150 	stat->rx_rrs_errs += smb->rx_rrs_errs;
2151 	stat->rx_alignerrs += smb->rx_alignerrs;
2152 	stat->rx_bcast_bytes += smb->rx_bcast_bytes;
2153 	stat->rx_mcast_bytes += smb->rx_mcast_bytes;
2154 	stat->rx_pkts_filtered += smb->rx_pkts_filtered;
2155 
2156 	/* Tx stats. */
2157 	stat->tx_frames += smb->tx_frames;
2158 	stat->tx_bcast_frames += smb->tx_bcast_frames;
2159 	stat->tx_mcast_frames += smb->tx_mcast_frames;
2160 	stat->tx_pause_frames += smb->tx_pause_frames;
2161 	stat->tx_excess_defer += smb->tx_excess_defer;
2162 	stat->tx_control_frames += smb->tx_control_frames;
2163 	stat->tx_deferred += smb->tx_deferred;
2164 	stat->tx_bytes += smb->tx_bytes;
2165 	stat->tx_pkts_64 += smb->tx_pkts_64;
2166 	stat->tx_pkts_65_127 += smb->tx_pkts_65_127;
2167 	stat->tx_pkts_128_255 += smb->tx_pkts_128_255;
2168 	stat->tx_pkts_256_511 += smb->tx_pkts_256_511;
2169 	stat->tx_pkts_512_1023 += smb->tx_pkts_512_1023;
2170 	stat->tx_pkts_1024_1518 += smb->tx_pkts_1024_1518;
2171 	stat->tx_pkts_1519_max += smb->tx_pkts_1519_max;
2172 	stat->tx_single_colls += smb->tx_single_colls;
2173 	stat->tx_multi_colls += smb->tx_multi_colls;
2174 	stat->tx_late_colls += smb->tx_late_colls;
2175 	stat->tx_excess_colls += smb->tx_excess_colls;
2176 	stat->tx_underrun += smb->tx_underrun;
2177 	stat->tx_desc_underrun += smb->tx_desc_underrun;
2178 	stat->tx_lenerrs += smb->tx_lenerrs;
2179 	stat->tx_pkts_truncated += smb->tx_pkts_truncated;
2180 	stat->tx_bcast_bytes += smb->tx_bcast_bytes;
2181 	stat->tx_mcast_bytes += smb->tx_mcast_bytes;
2182 
2183 	/* Update counters in ifnet. */
2184 	if_inc_counter(ifp, IFCOUNTER_OPACKETS, smb->tx_frames);
2185 
2186 	if_inc_counter(ifp, IFCOUNTER_COLLISIONS, smb->tx_single_colls +
2187 	    smb->tx_multi_colls * 2 + smb->tx_late_colls +
2188 	    smb->tx_excess_colls * HDPX_CFG_RETRY_DEFAULT);
2189 
2190 	if_inc_counter(ifp, IFCOUNTER_OERRORS, smb->tx_late_colls +
2191 	    smb->tx_excess_colls + smb->tx_underrun + smb->tx_pkts_truncated);
2192 
2193 	if_inc_counter(ifp, IFCOUNTER_IPACKETS, smb->rx_frames);
2194 
2195 	if_inc_counter(ifp, IFCOUNTER_IERRORS,
2196 	    smb->rx_crcerrs + smb->rx_lenerrs +
2197 	    smb->rx_runts + smb->rx_pkts_truncated +
2198 	    smb->rx_fifo_oflows + smb->rx_rrs_errs +
2199 	    smb->rx_alignerrs);
2200 }
2201 
2202 static int
ale_intr(void * arg)2203 ale_intr(void *arg)
2204 {
2205 	struct ale_softc *sc;
2206 	uint32_t status;
2207 
2208 	sc = (struct ale_softc *)arg;
2209 
2210 	status = CSR_READ_4(sc, ALE_INTR_STATUS);
2211 	if ((status & ALE_INTRS) == 0)
2212 		return (FILTER_STRAY);
2213 	/* Disable interrupts. */
2214 	CSR_WRITE_4(sc, ALE_INTR_STATUS, INTR_DIS_INT);
2215 	taskqueue_enqueue(sc->ale_tq, &sc->ale_int_task);
2216 
2217 	return (FILTER_HANDLED);
2218 }
2219 
2220 static void
ale_int_task(void * arg,int pending)2221 ale_int_task(void *arg, int pending)
2222 {
2223 	struct ale_softc *sc;
2224 	if_t ifp;
2225 	uint32_t status;
2226 	int more;
2227 
2228 	sc = (struct ale_softc *)arg;
2229 
2230 	status = CSR_READ_4(sc, ALE_INTR_STATUS);
2231 	ALE_LOCK(sc);
2232 	if (sc->ale_morework != 0)
2233 		status |= INTR_RX_PKT;
2234 	if ((status & ALE_INTRS) == 0)
2235 		goto done;
2236 
2237 	/* Acknowledge interrupts but still disable interrupts. */
2238 	CSR_WRITE_4(sc, ALE_INTR_STATUS, status | INTR_DIS_INT);
2239 
2240 	ifp = sc->ale_ifp;
2241 	more = 0;
2242 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0) {
2243 		more = ale_rxeof(sc, sc->ale_process_limit);
2244 		if (more == EAGAIN)
2245 			sc->ale_morework = 1;
2246 		else if (more == EIO) {
2247 			sc->ale_stats.reset_brk_seq++;
2248 			if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
2249 			ale_init_locked(sc);
2250 			ALE_UNLOCK(sc);
2251 			return;
2252 		}
2253 
2254 		if ((status & (INTR_DMA_RD_TO_RST | INTR_DMA_WR_TO_RST)) != 0) {
2255 			if ((status & INTR_DMA_RD_TO_RST) != 0)
2256 				device_printf(sc->ale_dev,
2257 				    "DMA read error! -- resetting\n");
2258 			if ((status & INTR_DMA_WR_TO_RST) != 0)
2259 				device_printf(sc->ale_dev,
2260 				    "DMA write error! -- resetting\n");
2261 			if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
2262 			ale_init_locked(sc);
2263 			ALE_UNLOCK(sc);
2264 			return;
2265 		}
2266 		if (!if_sendq_empty(ifp))
2267 			ale_start_locked(ifp);
2268 	}
2269 
2270 	if (more == EAGAIN ||
2271 	    (CSR_READ_4(sc, ALE_INTR_STATUS) & ALE_INTRS) != 0) {
2272 		ALE_UNLOCK(sc);
2273 		taskqueue_enqueue(sc->ale_tq, &sc->ale_int_task);
2274 		return;
2275 	}
2276 
2277 done:
2278 	ALE_UNLOCK(sc);
2279 
2280 	/* Re-enable interrupts. */
2281 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0x7FFFFFFF);
2282 }
2283 
2284 static void
ale_txeof(struct ale_softc * sc)2285 ale_txeof(struct ale_softc *sc)
2286 {
2287 	if_t ifp;
2288 	struct ale_txdesc *txd;
2289 	uint32_t cons, prod;
2290 	int prog;
2291 
2292 	ALE_LOCK_ASSERT(sc);
2293 
2294 	ifp = sc->ale_ifp;
2295 
2296 	if (sc->ale_cdata.ale_tx_cnt == 0)
2297 		return;
2298 
2299 	bus_dmamap_sync(sc->ale_cdata.ale_tx_ring_tag,
2300 	    sc->ale_cdata.ale_tx_ring_map,
2301 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2302 	if ((sc->ale_flags & ALE_FLAG_TXCMB_BUG) == 0) {
2303 		bus_dmamap_sync(sc->ale_cdata.ale_tx_cmb_tag,
2304 		    sc->ale_cdata.ale_tx_cmb_map,
2305 		    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2306 		prod = *sc->ale_cdata.ale_tx_cmb & TPD_CNT_MASK;
2307 	} else
2308 		prod = CSR_READ_2(sc, ALE_TPD_CONS_IDX);
2309 	cons = sc->ale_cdata.ale_tx_cons;
2310 	/*
2311 	 * Go through our Tx list and free mbufs for those
2312 	 * frames which have been transmitted.
2313 	 */
2314 	for (prog = 0; cons != prod; prog++,
2315 	    ALE_DESC_INC(cons, ALE_TX_RING_CNT)) {
2316 		if (sc->ale_cdata.ale_tx_cnt <= 0)
2317 			break;
2318 		prog++;
2319 		if_setdrvflagbits(ifp, 0, IFF_DRV_OACTIVE);
2320 		sc->ale_cdata.ale_tx_cnt--;
2321 		txd = &sc->ale_cdata.ale_txdesc[cons];
2322 		if (txd->tx_m != NULL) {
2323 			/* Reclaim transmitted mbufs. */
2324 			bus_dmamap_sync(sc->ale_cdata.ale_tx_tag,
2325 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2326 			bus_dmamap_unload(sc->ale_cdata.ale_tx_tag,
2327 			    txd->tx_dmamap);
2328 			m_freem(txd->tx_m);
2329 			txd->tx_m = NULL;
2330 		}
2331 	}
2332 
2333 	if (prog > 0) {
2334 		sc->ale_cdata.ale_tx_cons = cons;
2335 		/*
2336 		 * Unarm watchdog timer only when there is no pending
2337 		 * Tx descriptors in queue.
2338 		 */
2339 		if (sc->ale_cdata.ale_tx_cnt == 0)
2340 			sc->ale_watchdog_timer = 0;
2341 	}
2342 }
2343 
2344 static void
ale_rx_update_page(struct ale_softc * sc,struct ale_rx_page ** page,uint32_t length,uint32_t * prod)2345 ale_rx_update_page(struct ale_softc *sc, struct ale_rx_page **page,
2346     uint32_t length, uint32_t *prod)
2347 {
2348 	struct ale_rx_page *rx_page;
2349 
2350 	rx_page = *page;
2351 	/* Update consumer position. */
2352 	rx_page->cons += roundup(length + sizeof(struct rx_rs),
2353 	    ALE_RX_PAGE_ALIGN);
2354 	if (rx_page->cons >= ALE_RX_PAGE_SZ) {
2355 		/*
2356 		 * End of Rx page reached, let hardware reuse
2357 		 * this page.
2358 		 */
2359 		rx_page->cons = 0;
2360 		*rx_page->cmb_addr = 0;
2361 		bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2362 		    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2363 		CSR_WRITE_1(sc, ALE_RXF0_PAGE0 + sc->ale_cdata.ale_rx_curp,
2364 		    RXF_VALID);
2365 		/* Switch to alternate Rx page. */
2366 		sc->ale_cdata.ale_rx_curp ^= 1;
2367 		rx_page = *page =
2368 		    &sc->ale_cdata.ale_rx_page[sc->ale_cdata.ale_rx_curp];
2369 		/* Page flipped, sync CMB and Rx page. */
2370 		bus_dmamap_sync(rx_page->page_tag, rx_page->page_map,
2371 		    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2372 		bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2373 		    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2374 		/* Sync completed, cache updated producer index. */
2375 		*prod = *rx_page->cmb_addr;
2376 	}
2377 }
2378 
2379 /*
2380  * It seems that AR81xx controller can compute partial checksum.
2381  * The partial checksum value can be used to accelerate checksum
2382  * computation for fragmented TCP/UDP packets. Upper network stack
2383  * already takes advantage of the partial checksum value in IP
2384  * reassembly stage. But I'm not sure the correctness of the
2385  * partial hardware checksum assistance due to lack of data sheet.
2386  * In addition, the Rx feature of controller that requires copying
2387  * for every frames effectively nullifies one of most nice offload
2388  * capability of controller.
2389  */
2390 static void
ale_rxcsum(struct ale_softc * sc,struct mbuf * m,uint32_t status)2391 ale_rxcsum(struct ale_softc *sc, struct mbuf *m, uint32_t status)
2392 {
2393 	if_t ifp;
2394 	struct ip *ip;
2395 	char *p;
2396 
2397 	ifp = sc->ale_ifp;
2398 	m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
2399 	if ((status & ALE_RD_IPCSUM_NOK) == 0)
2400 		m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
2401 
2402 	if ((sc->ale_flags & ALE_FLAG_RXCSUM_BUG) == 0) {
2403 		if (((status & ALE_RD_IPV4_FRAG) == 0) &&
2404 		    ((status & (ALE_RD_TCP | ALE_RD_UDP)) != 0) &&
2405 		    ((status & ALE_RD_TCP_UDPCSUM_NOK) == 0)) {
2406 			m->m_pkthdr.csum_flags |=
2407 			    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2408 			m->m_pkthdr.csum_data = 0xffff;
2409 		}
2410 	} else {
2411 		if ((status & (ALE_RD_TCP | ALE_RD_UDP)) != 0 &&
2412 		    (status & ALE_RD_TCP_UDPCSUM_NOK) == 0) {
2413 			p = mtod(m, char *);
2414 			p += ETHER_HDR_LEN;
2415 			if ((status & ALE_RD_802_3) != 0)
2416 				p += LLC_SNAPFRAMELEN;
2417 			if ((if_getcapenable(ifp) & IFCAP_VLAN_HWTAGGING) == 0 &&
2418 			    (status & ALE_RD_VLAN) != 0)
2419 				p += ETHER_VLAN_ENCAP_LEN;
2420 			ip = (struct ip *)p;
2421 			if (ip->ip_off != 0 && (status & ALE_RD_IPV4_DF) == 0)
2422 				return;
2423 			m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
2424 			    CSUM_PSEUDO_HDR;
2425 			m->m_pkthdr.csum_data = 0xffff;
2426 		}
2427 	}
2428 	/*
2429 	 * Don't mark bad checksum for TCP/UDP frames
2430 	 * as fragmented frames may always have set
2431 	 * bad checksummed bit of frame status.
2432 	 */
2433 }
2434 
2435 /* Process received frames. */
2436 static int
ale_rxeof(struct ale_softc * sc,int count)2437 ale_rxeof(struct ale_softc *sc, int count)
2438 {
2439 	struct ale_rx_page *rx_page;
2440 	struct rx_rs *rs;
2441 	if_t ifp;
2442 	struct mbuf *m;
2443 	uint32_t length, prod, seqno, status, vtags;
2444 	int prog;
2445 
2446 	ifp = sc->ale_ifp;
2447 	rx_page = &sc->ale_cdata.ale_rx_page[sc->ale_cdata.ale_rx_curp];
2448 	bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2449 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2450 	bus_dmamap_sync(rx_page->page_tag, rx_page->page_map,
2451 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2452 	/*
2453 	 * Don't directly access producer index as hardware may
2454 	 * update it while Rx handler is in progress. It would
2455 	 * be even better if there is a way to let hardware
2456 	 * know how far driver processed its received frames.
2457 	 * Alternatively, hardware could provide a way to disable
2458 	 * CMB updates until driver acknowledges the end of CMB
2459 	 * access.
2460 	 */
2461 	prod = *rx_page->cmb_addr;
2462 	for (prog = 0; prog < count; prog++) {
2463 		if (rx_page->cons >= prod)
2464 			break;
2465 		rs = (struct rx_rs *)(rx_page->page_addr + rx_page->cons);
2466 		seqno = ALE_RX_SEQNO(le32toh(rs->seqno));
2467 		if (sc->ale_cdata.ale_rx_seqno != seqno) {
2468 			/*
2469 			 * Normally I believe this should not happen unless
2470 			 * severe driver bug or corrupted memory. However
2471 			 * it seems to happen under certain conditions which
2472 			 * is triggered by abrupt Rx events such as initiation
2473 			 * of bulk transfer of remote host. It's not easy to
2474 			 * reproduce this and I doubt it could be related
2475 			 * with FIFO overflow of hardware or activity of Tx
2476 			 * CMB updates. I also remember similar behaviour
2477 			 * seen on RealTek 8139 which uses resembling Rx
2478 			 * scheme.
2479 			 */
2480 			if (bootverbose)
2481 				device_printf(sc->ale_dev,
2482 				    "garbled seq: %u, expected: %u -- "
2483 				    "resetting!\n", seqno,
2484 				    sc->ale_cdata.ale_rx_seqno);
2485 			return (EIO);
2486 		}
2487 		/* Frame received. */
2488 		sc->ale_cdata.ale_rx_seqno++;
2489 		length = ALE_RX_BYTES(le32toh(rs->length));
2490 		status = le32toh(rs->flags);
2491 		if ((status & ALE_RD_ERROR) != 0) {
2492 			/*
2493 			 * We want to pass the following frames to upper
2494 			 * layer regardless of error status of Rx return
2495 			 * status.
2496 			 *
2497 			 *  o IP/TCP/UDP checksum is bad.
2498 			 *  o frame length and protocol specific length
2499 			 *     does not match.
2500 			 */
2501 			if ((status & (ALE_RD_CRC | ALE_RD_CODE |
2502 			    ALE_RD_DRIBBLE | ALE_RD_RUNT | ALE_RD_OFLOW |
2503 			    ALE_RD_TRUNC)) != 0) {
2504 				ale_rx_update_page(sc, &rx_page, length, &prod);
2505 				continue;
2506 			}
2507 		}
2508 		/*
2509 		 * m_devget(9) is major bottle-neck of ale(4)(It comes
2510 		 * from hardware limitation). For jumbo frames we could
2511 		 * get a slightly better performance if driver use
2512 		 * m_getjcl(9) with proper buffer size argument. However
2513 		 * that would make code more complicated and I don't
2514 		 * think users would expect good Rx performance numbers
2515 		 * on these low-end consumer ethernet controller.
2516 		 */
2517 		m = m_devget((char *)(rs + 1), length - ETHER_CRC_LEN,
2518 		    ETHER_ALIGN, ifp, NULL);
2519 		if (m == NULL) {
2520 			if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
2521 			ale_rx_update_page(sc, &rx_page, length, &prod);
2522 			continue;
2523 		}
2524 		if ((if_getcapenable(ifp) & IFCAP_RXCSUM) != 0 &&
2525 		    (status & ALE_RD_IPV4) != 0)
2526 			ale_rxcsum(sc, m, status);
2527 		if ((if_getcapenable(ifp) & IFCAP_VLAN_HWTAGGING) != 0 &&
2528 		    (status & ALE_RD_VLAN) != 0) {
2529 			vtags = ALE_RX_VLAN(le32toh(rs->vtags));
2530 			m->m_pkthdr.ether_vtag = ALE_RX_VLAN_TAG(vtags);
2531 			m->m_flags |= M_VLANTAG;
2532 		}
2533 
2534 		/* Pass it to upper layer. */
2535 		ALE_UNLOCK(sc);
2536 		if_input(ifp, m);
2537 		ALE_LOCK(sc);
2538 
2539 		ale_rx_update_page(sc, &rx_page, length, &prod);
2540 	}
2541 
2542 	return (count > 0 ? 0 : EAGAIN);
2543 }
2544 
2545 static void
ale_tick(void * arg)2546 ale_tick(void *arg)
2547 {
2548 	struct ale_softc *sc;
2549 	struct mii_data *mii;
2550 
2551 	sc = (struct ale_softc *)arg;
2552 
2553 	ALE_LOCK_ASSERT(sc);
2554 
2555 	mii = device_get_softc(sc->ale_miibus);
2556 	mii_tick(mii);
2557 	ale_stats_update(sc);
2558 	/*
2559 	 * Reclaim Tx buffers that have been transferred. It's not
2560 	 * needed here but it would release allocated mbuf chains
2561 	 * faster and limit the maximum delay to a hz.
2562 	 */
2563 	ale_txeof(sc);
2564 	ale_watchdog(sc);
2565 	callout_reset(&sc->ale_tick_ch, hz, ale_tick, sc);
2566 }
2567 
2568 static void
ale_reset(struct ale_softc * sc)2569 ale_reset(struct ale_softc *sc)
2570 {
2571 	uint32_t reg;
2572 	int i;
2573 
2574 	/* Initialize PCIe module. From Linux. */
2575 	CSR_WRITE_4(sc, 0x1008, CSR_READ_4(sc, 0x1008) | 0x8000);
2576 
2577 	CSR_WRITE_4(sc, ALE_MASTER_CFG, MASTER_RESET);
2578 	for (i = ALE_RESET_TIMEOUT; i > 0; i--) {
2579 		DELAY(10);
2580 		if ((CSR_READ_4(sc, ALE_MASTER_CFG) & MASTER_RESET) == 0)
2581 			break;
2582 	}
2583 	if (i == 0)
2584 		device_printf(sc->ale_dev, "master reset timeout!\n");
2585 
2586 	for (i = ALE_RESET_TIMEOUT; i > 0; i--) {
2587 		if ((reg = CSR_READ_4(sc, ALE_IDLE_STATUS)) == 0)
2588 			break;
2589 		DELAY(10);
2590 	}
2591 
2592 	if (i == 0)
2593 		device_printf(sc->ale_dev, "reset timeout(0x%08x)!\n", reg);
2594 }
2595 
2596 static void
ale_init(void * xsc)2597 ale_init(void *xsc)
2598 {
2599 	struct ale_softc *sc;
2600 
2601 	sc = (struct ale_softc *)xsc;
2602 	ALE_LOCK(sc);
2603 	ale_init_locked(sc);
2604 	ALE_UNLOCK(sc);
2605 }
2606 
2607 static void
ale_init_locked(struct ale_softc * sc)2608 ale_init_locked(struct ale_softc *sc)
2609 {
2610 	if_t ifp;
2611 	struct mii_data *mii;
2612 	uint8_t eaddr[ETHER_ADDR_LEN];
2613 	bus_addr_t paddr;
2614 	uint32_t reg, rxf_hi, rxf_lo;
2615 
2616 	ALE_LOCK_ASSERT(sc);
2617 
2618 	ifp = sc->ale_ifp;
2619 	mii = device_get_softc(sc->ale_miibus);
2620 
2621 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0)
2622 		return;
2623 	/*
2624 	 * Cancel any pending I/O.
2625 	 */
2626 	ale_stop(sc);
2627 	/*
2628 	 * Reset the chip to a known state.
2629 	 */
2630 	ale_reset(sc);
2631 	/* Initialize Tx descriptors, DMA memory blocks. */
2632 	ale_init_rx_pages(sc);
2633 	ale_init_tx_ring(sc);
2634 
2635 	/* Reprogram the station address. */
2636 	bcopy(if_getlladdr(ifp), eaddr, ETHER_ADDR_LEN);
2637 	CSR_WRITE_4(sc, ALE_PAR0,
2638 	    eaddr[2] << 24 | eaddr[3] << 16 | eaddr[4] << 8 | eaddr[5]);
2639 	CSR_WRITE_4(sc, ALE_PAR1, eaddr[0] << 8 | eaddr[1]);
2640 	/*
2641 	 * Clear WOL status and disable all WOL feature as WOL
2642 	 * would interfere Rx operation under normal environments.
2643 	 */
2644 	CSR_READ_4(sc, ALE_WOL_CFG);
2645 	CSR_WRITE_4(sc, ALE_WOL_CFG, 0);
2646 	/*
2647 	 * Set Tx descriptor/RXF0/CMB base addresses. They share
2648 	 * the same high address part of DMAable region.
2649 	 */
2650 	paddr = sc->ale_cdata.ale_tx_ring_paddr;
2651 	CSR_WRITE_4(sc, ALE_TPD_ADDR_HI, ALE_ADDR_HI(paddr));
2652 	CSR_WRITE_4(sc, ALE_TPD_ADDR_LO, ALE_ADDR_LO(paddr));
2653 	CSR_WRITE_4(sc, ALE_TPD_CNT,
2654 	    (ALE_TX_RING_CNT << TPD_CNT_SHIFT) & TPD_CNT_MASK);
2655 	/* Set Rx page base address, note we use single queue. */
2656 	paddr = sc->ale_cdata.ale_rx_page[0].page_paddr;
2657 	CSR_WRITE_4(sc, ALE_RXF0_PAGE0_ADDR_LO, ALE_ADDR_LO(paddr));
2658 	paddr = sc->ale_cdata.ale_rx_page[1].page_paddr;
2659 	CSR_WRITE_4(sc, ALE_RXF0_PAGE1_ADDR_LO, ALE_ADDR_LO(paddr));
2660 	/* Set Tx/Rx CMB addresses. */
2661 	paddr = sc->ale_cdata.ale_tx_cmb_paddr;
2662 	CSR_WRITE_4(sc, ALE_TX_CMB_ADDR_LO, ALE_ADDR_LO(paddr));
2663 	paddr = sc->ale_cdata.ale_rx_page[0].cmb_paddr;
2664 	CSR_WRITE_4(sc, ALE_RXF0_CMB0_ADDR_LO, ALE_ADDR_LO(paddr));
2665 	paddr = sc->ale_cdata.ale_rx_page[1].cmb_paddr;
2666 	CSR_WRITE_4(sc, ALE_RXF0_CMB1_ADDR_LO, ALE_ADDR_LO(paddr));
2667 	/* Mark RXF0 is valid. */
2668 	CSR_WRITE_1(sc, ALE_RXF0_PAGE0, RXF_VALID);
2669 	CSR_WRITE_1(sc, ALE_RXF0_PAGE1, RXF_VALID);
2670 	/*
2671 	 * No need to initialize RFX1/RXF2/RXF3. We don't use
2672 	 * multi-queue yet.
2673 	 */
2674 
2675 	/* Set Rx page size, excluding guard frame size. */
2676 	CSR_WRITE_4(sc, ALE_RXF_PAGE_SIZE, ALE_RX_PAGE_SZ);
2677 	/* Tell hardware that we're ready to load DMA blocks. */
2678 	CSR_WRITE_4(sc, ALE_DMA_BLOCK, DMA_BLOCK_LOAD);
2679 
2680 	/* Set Rx/Tx interrupt trigger threshold. */
2681 	CSR_WRITE_4(sc, ALE_INT_TRIG_THRESH, (1 << INT_TRIG_RX_THRESH_SHIFT) |
2682 	    (4 << INT_TRIG_TX_THRESH_SHIFT));
2683 	/*
2684 	 * XXX
2685 	 * Set interrupt trigger timer, its purpose and relation
2686 	 * with interrupt moderation mechanism is not clear yet.
2687 	 */
2688 	CSR_WRITE_4(sc, ALE_INT_TRIG_TIMER,
2689 	    ((ALE_USECS(10) << INT_TRIG_RX_TIMER_SHIFT) |
2690 	    (ALE_USECS(1000) << INT_TRIG_TX_TIMER_SHIFT)));
2691 
2692 	/* Configure interrupt moderation timer. */
2693 	reg = ALE_USECS(sc->ale_int_rx_mod) << IM_TIMER_RX_SHIFT;
2694 	reg |= ALE_USECS(sc->ale_int_tx_mod) << IM_TIMER_TX_SHIFT;
2695 	CSR_WRITE_4(sc, ALE_IM_TIMER, reg);
2696 	reg = CSR_READ_4(sc, ALE_MASTER_CFG);
2697 	reg &= ~(MASTER_CHIP_REV_MASK | MASTER_CHIP_ID_MASK);
2698 	reg &= ~(MASTER_IM_RX_TIMER_ENB | MASTER_IM_TX_TIMER_ENB);
2699 	if (ALE_USECS(sc->ale_int_rx_mod) != 0)
2700 		reg |= MASTER_IM_RX_TIMER_ENB;
2701 	if (ALE_USECS(sc->ale_int_tx_mod) != 0)
2702 		reg |= MASTER_IM_TX_TIMER_ENB;
2703 	CSR_WRITE_4(sc, ALE_MASTER_CFG, reg);
2704 	CSR_WRITE_2(sc, ALE_INTR_CLR_TIMER, ALE_USECS(1000));
2705 
2706 	/* Set Maximum frame size of controller. */
2707 	if (if_getmtu(ifp) < ETHERMTU)
2708 		sc->ale_max_frame_size = ETHERMTU;
2709 	else
2710 		sc->ale_max_frame_size = if_getmtu(ifp);
2711 	sc->ale_max_frame_size += ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN +
2712 	    ETHER_CRC_LEN;
2713 	CSR_WRITE_4(sc, ALE_FRAME_SIZE, sc->ale_max_frame_size);
2714 	/* Configure IPG/IFG parameters. */
2715 	CSR_WRITE_4(sc, ALE_IPG_IFG_CFG,
2716 	    ((IPG_IFG_IPGT_DEFAULT << IPG_IFG_IPGT_SHIFT) & IPG_IFG_IPGT_MASK) |
2717 	    ((IPG_IFG_MIFG_DEFAULT << IPG_IFG_MIFG_SHIFT) & IPG_IFG_MIFG_MASK) |
2718 	    ((IPG_IFG_IPG1_DEFAULT << IPG_IFG_IPG1_SHIFT) & IPG_IFG_IPG1_MASK) |
2719 	    ((IPG_IFG_IPG2_DEFAULT << IPG_IFG_IPG2_SHIFT) & IPG_IFG_IPG2_MASK));
2720 	/* Set parameters for half-duplex media. */
2721 	CSR_WRITE_4(sc, ALE_HDPX_CFG,
2722 	    ((HDPX_CFG_LCOL_DEFAULT << HDPX_CFG_LCOL_SHIFT) &
2723 	    HDPX_CFG_LCOL_MASK) |
2724 	    ((HDPX_CFG_RETRY_DEFAULT << HDPX_CFG_RETRY_SHIFT) &
2725 	    HDPX_CFG_RETRY_MASK) | HDPX_CFG_EXC_DEF_EN |
2726 	    ((HDPX_CFG_ABEBT_DEFAULT << HDPX_CFG_ABEBT_SHIFT) &
2727 	    HDPX_CFG_ABEBT_MASK) |
2728 	    ((HDPX_CFG_JAMIPG_DEFAULT << HDPX_CFG_JAMIPG_SHIFT) &
2729 	    HDPX_CFG_JAMIPG_MASK));
2730 
2731 	/* Configure Tx jumbo frame parameters. */
2732 	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0) {
2733 		if (if_getmtu(ifp) < ETHERMTU)
2734 			reg = sc->ale_max_frame_size;
2735 		else if (if_getmtu(ifp) < 6 * 1024)
2736 			reg = (sc->ale_max_frame_size * 2) / 3;
2737 		else
2738 			reg = sc->ale_max_frame_size / 2;
2739 		CSR_WRITE_4(sc, ALE_TX_JUMBO_THRESH,
2740 		    roundup(reg, TX_JUMBO_THRESH_UNIT) >>
2741 		    TX_JUMBO_THRESH_UNIT_SHIFT);
2742 	}
2743 	/* Configure TxQ. */
2744 	reg = (128 << (sc->ale_dma_rd_burst >> DMA_CFG_RD_BURST_SHIFT))
2745 	    << TXQ_CFG_TX_FIFO_BURST_SHIFT;
2746 	reg |= (TXQ_CFG_TPD_BURST_DEFAULT << TXQ_CFG_TPD_BURST_SHIFT) &
2747 	    TXQ_CFG_TPD_BURST_MASK;
2748 	CSR_WRITE_4(sc, ALE_TXQ_CFG, reg | TXQ_CFG_ENHANCED_MODE | TXQ_CFG_ENB);
2749 
2750 	/* Configure Rx jumbo frame & flow control parameters. */
2751 	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0) {
2752 		reg = roundup(sc->ale_max_frame_size, RX_JUMBO_THRESH_UNIT);
2753 		CSR_WRITE_4(sc, ALE_RX_JUMBO_THRESH,
2754 		    (((reg >> RX_JUMBO_THRESH_UNIT_SHIFT) <<
2755 		    RX_JUMBO_THRESH_MASK_SHIFT) & RX_JUMBO_THRESH_MASK) |
2756 		    ((RX_JUMBO_LKAH_DEFAULT << RX_JUMBO_LKAH_SHIFT) &
2757 		    RX_JUMBO_LKAH_MASK));
2758 		reg = CSR_READ_4(sc, ALE_SRAM_RX_FIFO_LEN);
2759 		rxf_hi = (reg * 7) / 10;
2760 		rxf_lo = (reg * 3)/ 10;
2761 		CSR_WRITE_4(sc, ALE_RX_FIFO_PAUSE_THRESH,
2762 		    ((rxf_lo << RX_FIFO_PAUSE_THRESH_LO_SHIFT) &
2763 		    RX_FIFO_PAUSE_THRESH_LO_MASK) |
2764 		    ((rxf_hi << RX_FIFO_PAUSE_THRESH_HI_SHIFT) &
2765 		    RX_FIFO_PAUSE_THRESH_HI_MASK));
2766 	}
2767 
2768 	/* Disable RSS. */
2769 	CSR_WRITE_4(sc, ALE_RSS_IDT_TABLE0, 0);
2770 	CSR_WRITE_4(sc, ALE_RSS_CPU, 0);
2771 
2772 	/* Configure RxQ. */
2773 	CSR_WRITE_4(sc, ALE_RXQ_CFG,
2774 	    RXQ_CFG_ALIGN_32 | RXQ_CFG_CUT_THROUGH_ENB | RXQ_CFG_ENB);
2775 
2776 	/* Configure DMA parameters. */
2777 	reg = 0;
2778 	if ((sc->ale_flags & ALE_FLAG_TXCMB_BUG) == 0)
2779 		reg |= DMA_CFG_TXCMB_ENB;
2780 	CSR_WRITE_4(sc, ALE_DMA_CFG,
2781 	    DMA_CFG_OUT_ORDER | DMA_CFG_RD_REQ_PRI | DMA_CFG_RCB_64 |
2782 	    sc->ale_dma_rd_burst | reg |
2783 	    sc->ale_dma_wr_burst | DMA_CFG_RXCMB_ENB |
2784 	    ((DMA_CFG_RD_DELAY_CNT_DEFAULT << DMA_CFG_RD_DELAY_CNT_SHIFT) &
2785 	    DMA_CFG_RD_DELAY_CNT_MASK) |
2786 	    ((DMA_CFG_WR_DELAY_CNT_DEFAULT << DMA_CFG_WR_DELAY_CNT_SHIFT) &
2787 	    DMA_CFG_WR_DELAY_CNT_MASK));
2788 
2789 	/*
2790 	 * Hardware can be configured to issue SMB interrupt based
2791 	 * on programmed interval. Since there is a callout that is
2792 	 * invoked for every hz in driver we use that instead of
2793 	 * relying on periodic SMB interrupt.
2794 	 */
2795 	CSR_WRITE_4(sc, ALE_SMB_STAT_TIMER, ALE_USECS(0));
2796 	/* Clear MAC statistics. */
2797 	ale_stats_clear(sc);
2798 
2799 	/*
2800 	 * Configure Tx/Rx MACs.
2801 	 *  - Auto-padding for short frames.
2802 	 *  - Enable CRC generation.
2803 	 *  Actual reconfiguration of MAC for resolved speed/duplex
2804 	 *  is followed after detection of link establishment.
2805 	 *  AR81xx always does checksum computation regardless of
2806 	 *  MAC_CFG_RXCSUM_ENB bit. In fact, setting the bit will
2807 	 *  cause Rx handling issue for fragmented IP datagrams due
2808 	 *  to silicon bug.
2809 	 */
2810 	reg = MAC_CFG_TX_CRC_ENB | MAC_CFG_TX_AUTO_PAD | MAC_CFG_FULL_DUPLEX |
2811 	    ((MAC_CFG_PREAMBLE_DEFAULT << MAC_CFG_PREAMBLE_SHIFT) &
2812 	    MAC_CFG_PREAMBLE_MASK);
2813 	if ((sc->ale_flags & ALE_FLAG_FASTETHER) != 0)
2814 		reg |= MAC_CFG_SPEED_10_100;
2815 	else
2816 		reg |= MAC_CFG_SPEED_1000;
2817 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2818 
2819 	/* Set up the receive filter. */
2820 	ale_rxfilter(sc);
2821 	ale_rxvlan(sc);
2822 
2823 	/* Acknowledge all pending interrupts and clear it. */
2824 	CSR_WRITE_4(sc, ALE_INTR_MASK, ALE_INTRS);
2825 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0xFFFFFFFF);
2826 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0);
2827 
2828 	if_setdrvflagbits(ifp, IFF_DRV_RUNNING, 0);
2829 	if_setdrvflagbits(ifp, 0, IFF_DRV_OACTIVE);
2830 
2831 	sc->ale_flags &= ~ALE_FLAG_LINK;
2832 	/* Switch to the current media. */
2833 	mii_mediachg(mii);
2834 
2835 	callout_reset(&sc->ale_tick_ch, hz, ale_tick, sc);
2836 }
2837 
2838 static void
ale_stop(struct ale_softc * sc)2839 ale_stop(struct ale_softc *sc)
2840 {
2841 	if_t ifp;
2842 	struct ale_txdesc *txd;
2843 	uint32_t reg;
2844 	int i;
2845 
2846 	ALE_LOCK_ASSERT(sc);
2847 	/*
2848 	 * Mark the interface down and cancel the watchdog timer.
2849 	 */
2850 	ifp = sc->ale_ifp;
2851 	if_setdrvflagbits(ifp, 0, (IFF_DRV_RUNNING | IFF_DRV_OACTIVE));
2852 	sc->ale_flags &= ~ALE_FLAG_LINK;
2853 	callout_stop(&sc->ale_tick_ch);
2854 	sc->ale_watchdog_timer = 0;
2855 	ale_stats_update(sc);
2856 	/* Disable interrupts. */
2857 	CSR_WRITE_4(sc, ALE_INTR_MASK, 0);
2858 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0xFFFFFFFF);
2859 	/* Disable queue processing and DMA. */
2860 	reg = CSR_READ_4(sc, ALE_TXQ_CFG);
2861 	reg &= ~TXQ_CFG_ENB;
2862 	CSR_WRITE_4(sc, ALE_TXQ_CFG, reg);
2863 	reg = CSR_READ_4(sc, ALE_RXQ_CFG);
2864 	reg &= ~RXQ_CFG_ENB;
2865 	CSR_WRITE_4(sc, ALE_RXQ_CFG, reg);
2866 	reg = CSR_READ_4(sc, ALE_DMA_CFG);
2867 	reg &= ~(DMA_CFG_TXCMB_ENB | DMA_CFG_RXCMB_ENB);
2868 	CSR_WRITE_4(sc, ALE_DMA_CFG, reg);
2869 	DELAY(1000);
2870 	/* Stop Rx/Tx MACs. */
2871 	ale_stop_mac(sc);
2872 	/* Disable interrupts which might be touched in taskq handler. */
2873 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0xFFFFFFFF);
2874 
2875 	/*
2876 	 * Free TX mbufs still in the queues.
2877 	 */
2878 	for (i = 0; i < ALE_TX_RING_CNT; i++) {
2879 		txd = &sc->ale_cdata.ale_txdesc[i];
2880 		if (txd->tx_m != NULL) {
2881 			bus_dmamap_sync(sc->ale_cdata.ale_tx_tag,
2882 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2883 			bus_dmamap_unload(sc->ale_cdata.ale_tx_tag,
2884 			    txd->tx_dmamap);
2885 			m_freem(txd->tx_m);
2886 			txd->tx_m = NULL;
2887 		}
2888         }
2889 }
2890 
2891 static void
ale_stop_mac(struct ale_softc * sc)2892 ale_stop_mac(struct ale_softc *sc)
2893 {
2894 	uint32_t reg;
2895 	int i;
2896 
2897 	ALE_LOCK_ASSERT(sc);
2898 
2899 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
2900 	if ((reg & (MAC_CFG_TX_ENB | MAC_CFG_RX_ENB)) != 0) {
2901 		reg &= ~(MAC_CFG_TX_ENB | MAC_CFG_RX_ENB);
2902 		CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2903 	}
2904 
2905 	for (i = ALE_TIMEOUT; i > 0; i--) {
2906 		reg = CSR_READ_4(sc, ALE_IDLE_STATUS);
2907 		if (reg == 0)
2908 			break;
2909 		DELAY(10);
2910 	}
2911 	if (i == 0)
2912 		device_printf(sc->ale_dev,
2913 		    "could not disable Tx/Rx MAC(0x%08x)!\n", reg);
2914 }
2915 
2916 static void
ale_init_tx_ring(struct ale_softc * sc)2917 ale_init_tx_ring(struct ale_softc *sc)
2918 {
2919 	struct ale_txdesc *txd;
2920 	int i;
2921 
2922 	ALE_LOCK_ASSERT(sc);
2923 
2924 	sc->ale_cdata.ale_tx_prod = 0;
2925 	sc->ale_cdata.ale_tx_cons = 0;
2926 	sc->ale_cdata.ale_tx_cnt = 0;
2927 
2928 	bzero(sc->ale_cdata.ale_tx_ring, ALE_TX_RING_SZ);
2929 	bzero(sc->ale_cdata.ale_tx_cmb, ALE_TX_CMB_SZ);
2930 	for (i = 0; i < ALE_TX_RING_CNT; i++) {
2931 		txd = &sc->ale_cdata.ale_txdesc[i];
2932 		txd->tx_m = NULL;
2933 	}
2934 	*sc->ale_cdata.ale_tx_cmb = 0;
2935 	bus_dmamap_sync(sc->ale_cdata.ale_tx_cmb_tag,
2936 	    sc->ale_cdata.ale_tx_cmb_map,
2937 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2938 	bus_dmamap_sync(sc->ale_cdata.ale_tx_ring_tag,
2939 	    sc->ale_cdata.ale_tx_ring_map,
2940 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2941 }
2942 
2943 static void
ale_init_rx_pages(struct ale_softc * sc)2944 ale_init_rx_pages(struct ale_softc *sc)
2945 {
2946 	struct ale_rx_page *rx_page;
2947 	int i;
2948 
2949 	ALE_LOCK_ASSERT(sc);
2950 
2951 	sc->ale_morework = 0;
2952 	sc->ale_cdata.ale_rx_seqno = 0;
2953 	sc->ale_cdata.ale_rx_curp = 0;
2954 
2955 	for (i = 0; i < ALE_RX_PAGES; i++) {
2956 		rx_page = &sc->ale_cdata.ale_rx_page[i];
2957 		bzero(rx_page->page_addr, sc->ale_pagesize);
2958 		bzero(rx_page->cmb_addr, ALE_RX_CMB_SZ);
2959 		rx_page->cons = 0;
2960 		*rx_page->cmb_addr = 0;
2961 		bus_dmamap_sync(rx_page->page_tag, rx_page->page_map,
2962 		    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2963 		bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2964 		    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2965 	}
2966 }
2967 
2968 static void
ale_rxvlan(struct ale_softc * sc)2969 ale_rxvlan(struct ale_softc *sc)
2970 {
2971 	if_t ifp;
2972 	uint32_t reg;
2973 
2974 	ALE_LOCK_ASSERT(sc);
2975 
2976 	ifp = sc->ale_ifp;
2977 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
2978 	reg &= ~MAC_CFG_VLAN_TAG_STRIP;
2979 	if ((if_getcapenable(ifp) & IFCAP_VLAN_HWTAGGING) != 0)
2980 		reg |= MAC_CFG_VLAN_TAG_STRIP;
2981 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2982 }
2983 
2984 static u_int
ale_hash_maddr(void * arg,struct sockaddr_dl * sdl,u_int cnt)2985 ale_hash_maddr(void *arg, struct sockaddr_dl *sdl, u_int cnt)
2986 {
2987 	uint32_t crc, *mchash = arg;
2988 
2989 	crc = ether_crc32_be(LLADDR(sdl), ETHER_ADDR_LEN);
2990 	mchash[crc >> 31] |= 1 << ((crc >> 26) & 0x1f);
2991 
2992 	return (1);
2993 }
2994 
2995 static void
ale_rxfilter(struct ale_softc * sc)2996 ale_rxfilter(struct ale_softc *sc)
2997 {
2998 	if_t ifp;
2999 	uint32_t mchash[2];
3000 	uint32_t rxcfg;
3001 
3002 	ALE_LOCK_ASSERT(sc);
3003 
3004 	ifp = sc->ale_ifp;
3005 
3006 	rxcfg = CSR_READ_4(sc, ALE_MAC_CFG);
3007 	rxcfg &= ~(MAC_CFG_ALLMULTI | MAC_CFG_BCAST | MAC_CFG_PROMISC);
3008 	if ((if_getflags(ifp) & IFF_BROADCAST) != 0)
3009 		rxcfg |= MAC_CFG_BCAST;
3010 	if ((if_getflags(ifp) & (IFF_PROMISC | IFF_ALLMULTI)) != 0) {
3011 		if ((if_getflags(ifp) & IFF_PROMISC) != 0)
3012 			rxcfg |= MAC_CFG_PROMISC;
3013 		if ((if_getflags(ifp) & IFF_ALLMULTI) != 0)
3014 			rxcfg |= MAC_CFG_ALLMULTI;
3015 		CSR_WRITE_4(sc, ALE_MAR0, 0xFFFFFFFF);
3016 		CSR_WRITE_4(sc, ALE_MAR1, 0xFFFFFFFF);
3017 		CSR_WRITE_4(sc, ALE_MAC_CFG, rxcfg);
3018 		return;
3019 	}
3020 
3021 	/* Program new filter. */
3022 	bzero(mchash, sizeof(mchash));
3023 	if_foreach_llmaddr(ifp, ale_hash_maddr, &mchash);
3024 
3025 	CSR_WRITE_4(sc, ALE_MAR0, mchash[0]);
3026 	CSR_WRITE_4(sc, ALE_MAR1, mchash[1]);
3027 	CSR_WRITE_4(sc, ALE_MAC_CFG, rxcfg);
3028 }
3029 
3030 static int
sysctl_int_range(SYSCTL_HANDLER_ARGS,int low,int high)3031 sysctl_int_range(SYSCTL_HANDLER_ARGS, int low, int high)
3032 {
3033 	int error, value;
3034 
3035 	if (arg1 == NULL)
3036 		return (EINVAL);
3037 	value = *(int *)arg1;
3038 	error = sysctl_handle_int(oidp, &value, 0, req);
3039 	if (error || req->newptr == NULL)
3040 		return (error);
3041 	if (value < low || value > high)
3042 		return (EINVAL);
3043         *(int *)arg1 = value;
3044 
3045         return (0);
3046 }
3047 
3048 static int
sysctl_hw_ale_proc_limit(SYSCTL_HANDLER_ARGS)3049 sysctl_hw_ale_proc_limit(SYSCTL_HANDLER_ARGS)
3050 {
3051 	return (sysctl_int_range(oidp, arg1, arg2, req,
3052 	    ALE_PROC_MIN, ALE_PROC_MAX));
3053 }
3054 
3055 static int
sysctl_hw_ale_int_mod(SYSCTL_HANDLER_ARGS)3056 sysctl_hw_ale_int_mod(SYSCTL_HANDLER_ARGS)
3057 {
3058 
3059 	return (sysctl_int_range(oidp, arg1, arg2, req,
3060 	    ALE_IM_TIMER_MIN, ALE_IM_TIMER_MAX));
3061 }
3062