xref: /freebsd/sys/dev/ale/if_ale.c (revision ddaf6524682b3ab9e50f7575db319814dbbd053a)
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;
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_has_pm(dev)) {
623 		if_setcapabilitiesbit(ifp, IFCAP_WOL_MAGIC | IFCAP_WOL_MCAST, 0);
624 	}
625 	if_setcapenable(ifp, if_getcapabilities(ifp));
626 
627 	/* Set up MII bus. */
628 	error = mii_attach(dev, &sc->ale_miibus, ifp, ale_mediachange,
629 	    ale_mediastatus, BMSR_DEFCAPMASK, sc->ale_phyaddr, MII_OFFSET_ANY,
630 	    MIIF_DOPAUSE);
631 	if (error != 0) {
632 		device_printf(dev, "attaching PHYs failed\n");
633 		goto fail;
634 	}
635 
636 	ether_ifattach(ifp, sc->ale_eaddr);
637 
638 	/* VLAN capability setup. */
639 	if_setcapabilitiesbit(ifp, IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING |
640 	    IFCAP_VLAN_HWCSUM | IFCAP_VLAN_HWTSO, 0);
641 	if_setcapenable(ifp, if_getcapabilities(ifp));
642 	/*
643 	 * Even though controllers supported by ale(3) have Rx checksum
644 	 * offload bug the workaround for fragmented frames seemed to
645 	 * work so far. However it seems Rx checksum offload does not
646 	 * work under certain conditions. So disable Rx checksum offload
647 	 * until I find more clue about it but allow users to override it.
648 	 */
649 	if_setcapenablebit(ifp, 0, IFCAP_RXCSUM);
650 
651 	/* Tell the upper layer(s) we support long frames. */
652 	if_setifheaderlen(ifp, sizeof(struct ether_vlan_header));
653 
654 	/* Create local taskq. */
655 	sc->ale_tq = taskqueue_create_fast("ale_taskq", M_WAITOK,
656 	    taskqueue_thread_enqueue, &sc->ale_tq);
657 	taskqueue_start_threads(&sc->ale_tq, 1, PI_NET, "%s taskq",
658 	    device_get_nameunit(sc->ale_dev));
659 
660 	if ((sc->ale_flags & ALE_FLAG_MSIX) != 0)
661 		msic = ALE_MSIX_MESSAGES;
662 	else if ((sc->ale_flags & ALE_FLAG_MSI) != 0)
663 		msic = ALE_MSI_MESSAGES;
664 	else
665 		msic = 1;
666 	for (i = 0; i < msic; i++) {
667 		error = bus_setup_intr(dev, sc->ale_irq[i],
668 		    INTR_TYPE_NET | INTR_MPSAFE, ale_intr, NULL, sc,
669 		    &sc->ale_intrhand[i]);
670 		if (error != 0)
671 			break;
672 	}
673 	if (error != 0) {
674 		device_printf(dev, "could not set up interrupt handler.\n");
675 		taskqueue_free(sc->ale_tq);
676 		sc->ale_tq = NULL;
677 		ether_ifdetach(ifp);
678 		goto fail;
679 	}
680 
681 fail:
682 	if (error != 0)
683 		ale_detach(dev);
684 
685 	return (error);
686 }
687 
688 static int
ale_detach(device_t dev)689 ale_detach(device_t dev)
690 {
691 	struct ale_softc *sc;
692 	if_t ifp;
693 	int i, msic;
694 
695 	sc = device_get_softc(dev);
696 
697 	ifp = sc->ale_ifp;
698 	if (device_is_attached(dev)) {
699 		ether_ifdetach(ifp);
700 		ALE_LOCK(sc);
701 		ale_stop(sc);
702 		ALE_UNLOCK(sc);
703 		callout_drain(&sc->ale_tick_ch);
704 		taskqueue_drain(sc->ale_tq, &sc->ale_int_task);
705 	}
706 
707 	if (sc->ale_tq != NULL) {
708 		taskqueue_drain(sc->ale_tq, &sc->ale_int_task);
709 		taskqueue_free(sc->ale_tq);
710 		sc->ale_tq = NULL;
711 	}
712 
713 	bus_generic_detach(dev);
714 	ale_dma_free(sc);
715 
716 	if (ifp != NULL) {
717 		if_free(ifp);
718 		sc->ale_ifp = NULL;
719 	}
720 
721 	if ((sc->ale_flags & ALE_FLAG_MSIX) != 0)
722 		msic = ALE_MSIX_MESSAGES;
723 	else if ((sc->ale_flags & ALE_FLAG_MSI) != 0)
724 		msic = ALE_MSI_MESSAGES;
725 	else
726 		msic = 1;
727 	for (i = 0; i < msic; i++) {
728 		if (sc->ale_intrhand[i] != NULL) {
729 			bus_teardown_intr(dev, sc->ale_irq[i],
730 			    sc->ale_intrhand[i]);
731 			sc->ale_intrhand[i] = NULL;
732 		}
733 	}
734 
735 	bus_release_resources(dev, sc->ale_irq_spec, sc->ale_irq);
736 	if ((sc->ale_flags & (ALE_FLAG_MSI | ALE_FLAG_MSIX)) != 0)
737 		pci_release_msi(dev);
738 	bus_release_resources(dev, sc->ale_res_spec, sc->ale_res);
739 	mtx_destroy(&sc->ale_mtx);
740 
741 	return (0);
742 }
743 
744 #define	ALE_SYSCTL_STAT_ADD32(c, h, n, p, d)	\
745 	    SYSCTL_ADD_UINT(c, h, OID_AUTO, n, CTLFLAG_RD, p, 0, d)
746 
747 #define	ALE_SYSCTL_STAT_ADD64(c, h, n, p, d)	\
748 	    SYSCTL_ADD_UQUAD(c, h, OID_AUTO, n, CTLFLAG_RD, p, d)
749 
750 static void
ale_sysctl_node(struct ale_softc * sc)751 ale_sysctl_node(struct ale_softc *sc)
752 {
753 	struct sysctl_ctx_list *ctx;
754 	struct sysctl_oid_list *child, *parent;
755 	struct sysctl_oid *tree;
756 	struct ale_hw_stats *stats;
757 	int error;
758 
759 	stats = &sc->ale_stats;
760 	ctx = device_get_sysctl_ctx(sc->ale_dev);
761 	child = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->ale_dev));
762 
763 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "int_rx_mod",
764 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &sc->ale_int_rx_mod,
765 	    0, sysctl_hw_ale_int_mod, "I", "ale Rx interrupt moderation");
766 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "int_tx_mod",
767 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &sc->ale_int_tx_mod,
768 	    0, sysctl_hw_ale_int_mod, "I", "ale Tx interrupt moderation");
769 	/* Pull in device tunables. */
770 	sc->ale_int_rx_mod = ALE_IM_RX_TIMER_DEFAULT;
771 	error = resource_int_value(device_get_name(sc->ale_dev),
772 	    device_get_unit(sc->ale_dev), "int_rx_mod", &sc->ale_int_rx_mod);
773 	if (error == 0) {
774 		if (sc->ale_int_rx_mod < ALE_IM_TIMER_MIN ||
775 		    sc->ale_int_rx_mod > ALE_IM_TIMER_MAX) {
776 			device_printf(sc->ale_dev, "int_rx_mod value out of "
777 			    "range; using default: %d\n",
778 			    ALE_IM_RX_TIMER_DEFAULT);
779 			sc->ale_int_rx_mod = ALE_IM_RX_TIMER_DEFAULT;
780 		}
781 	}
782 	sc->ale_int_tx_mod = ALE_IM_TX_TIMER_DEFAULT;
783 	error = resource_int_value(device_get_name(sc->ale_dev),
784 	    device_get_unit(sc->ale_dev), "int_tx_mod", &sc->ale_int_tx_mod);
785 	if (error == 0) {
786 		if (sc->ale_int_tx_mod < ALE_IM_TIMER_MIN ||
787 		    sc->ale_int_tx_mod > ALE_IM_TIMER_MAX) {
788 			device_printf(sc->ale_dev, "int_tx_mod value out of "
789 			    "range; using default: %d\n",
790 			    ALE_IM_TX_TIMER_DEFAULT);
791 			sc->ale_int_tx_mod = ALE_IM_TX_TIMER_DEFAULT;
792 		}
793 	}
794 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "process_limit",
795 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
796 	    &sc->ale_process_limit, 0, sysctl_hw_ale_proc_limit, "I",
797 	    "max number of Rx events to process");
798 	/* Pull in device tunables. */
799 	sc->ale_process_limit = ALE_PROC_DEFAULT;
800 	error = resource_int_value(device_get_name(sc->ale_dev),
801 	    device_get_unit(sc->ale_dev), "process_limit",
802 	    &sc->ale_process_limit);
803 	if (error == 0) {
804 		if (sc->ale_process_limit < ALE_PROC_MIN ||
805 		    sc->ale_process_limit > ALE_PROC_MAX) {
806 			device_printf(sc->ale_dev,
807 			    "process_limit value out of range; "
808 			    "using default: %d\n", ALE_PROC_DEFAULT);
809 			sc->ale_process_limit = ALE_PROC_DEFAULT;
810 		}
811 	}
812 
813 	/* Misc statistics. */
814 	ALE_SYSCTL_STAT_ADD32(ctx, child, "reset_brk_seq",
815 	    &stats->reset_brk_seq,
816 	    "Controller resets due to broken Rx sequnce number");
817 
818 	tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "stats",
819 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "ATE statistics");
820 	parent = SYSCTL_CHILDREN(tree);
821 
822 	/* Rx statistics. */
823 	tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "rx",
824 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Rx MAC statistics");
825 	child = SYSCTL_CHILDREN(tree);
826 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_frames",
827 	    &stats->rx_frames, "Good frames");
828 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_bcast_frames",
829 	    &stats->rx_bcast_frames, "Good broadcast frames");
830 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_mcast_frames",
831 	    &stats->rx_mcast_frames, "Good multicast frames");
832 	ALE_SYSCTL_STAT_ADD32(ctx, child, "pause_frames",
833 	    &stats->rx_pause_frames, "Pause control frames");
834 	ALE_SYSCTL_STAT_ADD32(ctx, child, "control_frames",
835 	    &stats->rx_control_frames, "Control frames");
836 	ALE_SYSCTL_STAT_ADD32(ctx, child, "crc_errs",
837 	    &stats->rx_crcerrs, "CRC errors");
838 	ALE_SYSCTL_STAT_ADD32(ctx, child, "len_errs",
839 	    &stats->rx_lenerrs, "Frames with length mismatched");
840 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_octets",
841 	    &stats->rx_bytes, "Good octets");
842 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_bcast_octets",
843 	    &stats->rx_bcast_bytes, "Good broadcast octets");
844 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_mcast_octets",
845 	    &stats->rx_mcast_bytes, "Good multicast octets");
846 	ALE_SYSCTL_STAT_ADD32(ctx, child, "runts",
847 	    &stats->rx_runts, "Too short frames");
848 	ALE_SYSCTL_STAT_ADD32(ctx, child, "fragments",
849 	    &stats->rx_fragments, "Fragmented frames");
850 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_64",
851 	    &stats->rx_pkts_64, "64 bytes frames");
852 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_65_127",
853 	    &stats->rx_pkts_65_127, "65 to 127 bytes frames");
854 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_128_255",
855 	    &stats->rx_pkts_128_255, "128 to 255 bytes frames");
856 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_256_511",
857 	    &stats->rx_pkts_256_511, "256 to 511 bytes frames");
858 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_512_1023",
859 	    &stats->rx_pkts_512_1023, "512 to 1023 bytes frames");
860 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1024_1518",
861 	    &stats->rx_pkts_1024_1518, "1024 to 1518 bytes frames");
862 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1519_max",
863 	    &stats->rx_pkts_1519_max, "1519 to max frames");
864 	ALE_SYSCTL_STAT_ADD32(ctx, child, "trunc_errs",
865 	    &stats->rx_pkts_truncated, "Truncated frames due to MTU size");
866 	ALE_SYSCTL_STAT_ADD32(ctx, child, "fifo_oflows",
867 	    &stats->rx_fifo_oflows, "FIFO overflows");
868 	ALE_SYSCTL_STAT_ADD32(ctx, child, "rrs_errs",
869 	    &stats->rx_rrs_errs, "Return status write-back errors");
870 	ALE_SYSCTL_STAT_ADD32(ctx, child, "align_errs",
871 	    &stats->rx_alignerrs, "Alignment errors");
872 	ALE_SYSCTL_STAT_ADD32(ctx, child, "filtered",
873 	    &stats->rx_pkts_filtered,
874 	    "Frames dropped due to address filtering");
875 
876 	/* Tx statistics. */
877 	tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "tx",
878 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Tx MAC statistics");
879 	child = SYSCTL_CHILDREN(tree);
880 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_frames",
881 	    &stats->tx_frames, "Good frames");
882 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_bcast_frames",
883 	    &stats->tx_bcast_frames, "Good broadcast frames");
884 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_mcast_frames",
885 	    &stats->tx_mcast_frames, "Good multicast frames");
886 	ALE_SYSCTL_STAT_ADD32(ctx, child, "pause_frames",
887 	    &stats->tx_pause_frames, "Pause control frames");
888 	ALE_SYSCTL_STAT_ADD32(ctx, child, "control_frames",
889 	    &stats->tx_control_frames, "Control frames");
890 	ALE_SYSCTL_STAT_ADD32(ctx, child, "excess_defers",
891 	    &stats->tx_excess_defer, "Frames with excessive derferrals");
892 	ALE_SYSCTL_STAT_ADD32(ctx, child, "defers",
893 	    &stats->tx_excess_defer, "Frames with derferrals");
894 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_octets",
895 	    &stats->tx_bytes, "Good octets");
896 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_bcast_octets",
897 	    &stats->tx_bcast_bytes, "Good broadcast octets");
898 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_mcast_octets",
899 	    &stats->tx_mcast_bytes, "Good multicast octets");
900 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_64",
901 	    &stats->tx_pkts_64, "64 bytes frames");
902 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_65_127",
903 	    &stats->tx_pkts_65_127, "65 to 127 bytes frames");
904 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_128_255",
905 	    &stats->tx_pkts_128_255, "128 to 255 bytes frames");
906 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_256_511",
907 	    &stats->tx_pkts_256_511, "256 to 511 bytes frames");
908 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_512_1023",
909 	    &stats->tx_pkts_512_1023, "512 to 1023 bytes frames");
910 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1024_1518",
911 	    &stats->tx_pkts_1024_1518, "1024 to 1518 bytes frames");
912 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1519_max",
913 	    &stats->tx_pkts_1519_max, "1519 to max frames");
914 	ALE_SYSCTL_STAT_ADD32(ctx, child, "single_colls",
915 	    &stats->tx_single_colls, "Single collisions");
916 	ALE_SYSCTL_STAT_ADD32(ctx, child, "multi_colls",
917 	    &stats->tx_multi_colls, "Multiple collisions");
918 	ALE_SYSCTL_STAT_ADD32(ctx, child, "late_colls",
919 	    &stats->tx_late_colls, "Late collisions");
920 	ALE_SYSCTL_STAT_ADD32(ctx, child, "excess_colls",
921 	    &stats->tx_excess_colls, "Excessive collisions");
922 	ALE_SYSCTL_STAT_ADD32(ctx, child, "underruns",
923 	    &stats->tx_underrun, "FIFO underruns");
924 	ALE_SYSCTL_STAT_ADD32(ctx, child, "desc_underruns",
925 	    &stats->tx_desc_underrun, "Descriptor write-back errors");
926 	ALE_SYSCTL_STAT_ADD32(ctx, child, "len_errs",
927 	    &stats->tx_lenerrs, "Frames with length mismatched");
928 	ALE_SYSCTL_STAT_ADD32(ctx, child, "trunc_errs",
929 	    &stats->tx_pkts_truncated, "Truncated frames due to MTU size");
930 }
931 
932 #undef ALE_SYSCTL_STAT_ADD32
933 #undef ALE_SYSCTL_STAT_ADD64
934 
935 struct ale_dmamap_arg {
936 	bus_addr_t	ale_busaddr;
937 };
938 
939 static void
ale_dmamap_cb(void * arg,bus_dma_segment_t * segs,int nsegs,int error)940 ale_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error)
941 {
942 	struct ale_dmamap_arg *ctx;
943 
944 	if (error != 0)
945 		return;
946 
947 	KASSERT(nsegs == 1, ("%s: %d segments returned!", __func__, nsegs));
948 
949 	ctx = (struct ale_dmamap_arg *)arg;
950 	ctx->ale_busaddr = segs[0].ds_addr;
951 }
952 
953 /*
954  * Tx descriptors/RXF0/CMB DMA blocks share ALE_DESC_ADDR_HI register
955  * which specifies high address region of DMA blocks. Therefore these
956  * blocks should have the same high address of given 4GB address
957  * space(i.e. crossing 4GB boundary is not allowed).
958  */
959 static int
ale_check_boundary(struct ale_softc * sc)960 ale_check_boundary(struct ale_softc *sc)
961 {
962 	bus_addr_t rx_cmb_end[ALE_RX_PAGES], tx_cmb_end;
963 	bus_addr_t rx_page_end[ALE_RX_PAGES], tx_ring_end;
964 
965 	rx_page_end[0] = sc->ale_cdata.ale_rx_page[0].page_paddr +
966 	    sc->ale_pagesize;
967 	rx_page_end[1] = sc->ale_cdata.ale_rx_page[1].page_paddr +
968 	    sc->ale_pagesize;
969 	tx_ring_end = sc->ale_cdata.ale_tx_ring_paddr + ALE_TX_RING_SZ;
970 	tx_cmb_end = sc->ale_cdata.ale_tx_cmb_paddr + ALE_TX_CMB_SZ;
971 	rx_cmb_end[0] = sc->ale_cdata.ale_rx_page[0].cmb_paddr + ALE_RX_CMB_SZ;
972 	rx_cmb_end[1] = sc->ale_cdata.ale_rx_page[1].cmb_paddr + ALE_RX_CMB_SZ;
973 
974 	if ((ALE_ADDR_HI(tx_ring_end) !=
975 	    ALE_ADDR_HI(sc->ale_cdata.ale_tx_ring_paddr)) ||
976 	    (ALE_ADDR_HI(rx_page_end[0]) !=
977 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[0].page_paddr)) ||
978 	    (ALE_ADDR_HI(rx_page_end[1]) !=
979 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[1].page_paddr)) ||
980 	    (ALE_ADDR_HI(tx_cmb_end) !=
981 	    ALE_ADDR_HI(sc->ale_cdata.ale_tx_cmb_paddr)) ||
982 	    (ALE_ADDR_HI(rx_cmb_end[0]) !=
983 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[0].cmb_paddr)) ||
984 	    (ALE_ADDR_HI(rx_cmb_end[1]) !=
985 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[1].cmb_paddr)))
986 		return (EFBIG);
987 
988 	if ((ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_page_end[0])) ||
989 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_page_end[1])) ||
990 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_cmb_end[0])) ||
991 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_cmb_end[1])) ||
992 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(tx_cmb_end)))
993 		return (EFBIG);
994 
995 	return (0);
996 }
997 
998 static int
ale_dma_alloc(struct ale_softc * sc)999 ale_dma_alloc(struct ale_softc *sc)
1000 {
1001 	struct ale_txdesc *txd;
1002 	bus_addr_t lowaddr;
1003 	struct ale_dmamap_arg ctx;
1004 	int error, guard_size, i;
1005 
1006 	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0)
1007 		guard_size = ALE_JUMBO_FRAMELEN;
1008 	else
1009 		guard_size = ALE_MAX_FRAMELEN;
1010 	sc->ale_pagesize = roundup(guard_size + ALE_RX_PAGE_SZ,
1011 	    ALE_RX_PAGE_ALIGN);
1012 	lowaddr = BUS_SPACE_MAXADDR;
1013 again:
1014 	/* Create parent DMA tag. */
1015 	error = bus_dma_tag_create(
1016 	    bus_get_dma_tag(sc->ale_dev), /* parent */
1017 	    1, 0,			/* alignment, boundary */
1018 	    lowaddr,			/* lowaddr */
1019 	    BUS_SPACE_MAXADDR,		/* highaddr */
1020 	    NULL, NULL,			/* filter, filterarg */
1021 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsize */
1022 	    0,				/* nsegments */
1023 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsegsize */
1024 	    0,				/* flags */
1025 	    NULL, NULL,			/* lockfunc, lockarg */
1026 	    &sc->ale_cdata.ale_parent_tag);
1027 	if (error != 0) {
1028 		device_printf(sc->ale_dev,
1029 		    "could not create parent DMA tag.\n");
1030 		goto fail;
1031 	}
1032 
1033 	/* Create DMA tag for Tx descriptor ring. */
1034 	error = bus_dma_tag_create(
1035 	    sc->ale_cdata.ale_parent_tag, /* parent */
1036 	    ALE_TX_RING_ALIGN, 0,	/* alignment, boundary */
1037 	    BUS_SPACE_MAXADDR,		/* lowaddr */
1038 	    BUS_SPACE_MAXADDR,		/* highaddr */
1039 	    NULL, NULL,			/* filter, filterarg */
1040 	    ALE_TX_RING_SZ,		/* maxsize */
1041 	    1,				/* nsegments */
1042 	    ALE_TX_RING_SZ,		/* maxsegsize */
1043 	    0,				/* flags */
1044 	    NULL, NULL,			/* lockfunc, lockarg */
1045 	    &sc->ale_cdata.ale_tx_ring_tag);
1046 	if (error != 0) {
1047 		device_printf(sc->ale_dev,
1048 		    "could not create Tx ring DMA tag.\n");
1049 		goto fail;
1050 	}
1051 
1052 	/* Create DMA tag for Rx pages. */
1053 	for (i = 0; i < ALE_RX_PAGES; i++) {
1054 		error = bus_dma_tag_create(
1055 		    sc->ale_cdata.ale_parent_tag, /* parent */
1056 		    ALE_RX_PAGE_ALIGN, 0,	/* alignment, boundary */
1057 		    BUS_SPACE_MAXADDR,		/* lowaddr */
1058 		    BUS_SPACE_MAXADDR,		/* highaddr */
1059 		    NULL, NULL,			/* filter, filterarg */
1060 		    sc->ale_pagesize,		/* maxsize */
1061 		    1,				/* nsegments */
1062 		    sc->ale_pagesize,		/* maxsegsize */
1063 		    0,				/* flags */
1064 		    NULL, NULL,			/* lockfunc, lockarg */
1065 		    &sc->ale_cdata.ale_rx_page[i].page_tag);
1066 		if (error != 0) {
1067 			device_printf(sc->ale_dev,
1068 			    "could not create Rx page %d DMA tag.\n", i);
1069 			goto fail;
1070 		}
1071 	}
1072 
1073 	/* Create DMA tag for Tx coalescing message block. */
1074 	error = bus_dma_tag_create(
1075 	    sc->ale_cdata.ale_parent_tag, /* parent */
1076 	    ALE_CMB_ALIGN, 0,		/* alignment, boundary */
1077 	    BUS_SPACE_MAXADDR,		/* lowaddr */
1078 	    BUS_SPACE_MAXADDR,		/* highaddr */
1079 	    NULL, NULL,			/* filter, filterarg */
1080 	    ALE_TX_CMB_SZ,		/* maxsize */
1081 	    1,				/* nsegments */
1082 	    ALE_TX_CMB_SZ,		/* maxsegsize */
1083 	    0,				/* flags */
1084 	    NULL, NULL,			/* lockfunc, lockarg */
1085 	    &sc->ale_cdata.ale_tx_cmb_tag);
1086 	if (error != 0) {
1087 		device_printf(sc->ale_dev,
1088 		    "could not create Tx CMB DMA tag.\n");
1089 		goto fail;
1090 	}
1091 
1092 	/* Create DMA tag for Rx coalescing message block. */
1093 	for (i = 0; i < ALE_RX_PAGES; i++) {
1094 		error = bus_dma_tag_create(
1095 		    sc->ale_cdata.ale_parent_tag, /* parent */
1096 		    ALE_CMB_ALIGN, 0,		/* alignment, boundary */
1097 		    BUS_SPACE_MAXADDR,		/* lowaddr */
1098 		    BUS_SPACE_MAXADDR,		/* highaddr */
1099 		    NULL, NULL,			/* filter, filterarg */
1100 		    ALE_RX_CMB_SZ,		/* maxsize */
1101 		    1,				/* nsegments */
1102 		    ALE_RX_CMB_SZ,		/* maxsegsize */
1103 		    0,				/* flags */
1104 		    NULL, NULL,			/* lockfunc, lockarg */
1105 		    &sc->ale_cdata.ale_rx_page[i].cmb_tag);
1106 		if (error != 0) {
1107 			device_printf(sc->ale_dev,
1108 			    "could not create Rx page %d CMB DMA tag.\n", i);
1109 			goto fail;
1110 		}
1111 	}
1112 
1113 	/* Allocate DMA'able memory and load the DMA map for Tx ring. */
1114 	error = bus_dmamem_alloc(sc->ale_cdata.ale_tx_ring_tag,
1115 	    (void **)&sc->ale_cdata.ale_tx_ring,
1116 	    BUS_DMA_WAITOK | BUS_DMA_ZERO | BUS_DMA_COHERENT,
1117 	    &sc->ale_cdata.ale_tx_ring_map);
1118 	if (error != 0) {
1119 		device_printf(sc->ale_dev,
1120 		    "could not allocate DMA'able memory for Tx ring.\n");
1121 		goto fail;
1122 	}
1123 	ctx.ale_busaddr = 0;
1124 	error = bus_dmamap_load(sc->ale_cdata.ale_tx_ring_tag,
1125 	    sc->ale_cdata.ale_tx_ring_map, sc->ale_cdata.ale_tx_ring,
1126 	    ALE_TX_RING_SZ, ale_dmamap_cb, &ctx, 0);
1127 	if (error != 0 || ctx.ale_busaddr == 0) {
1128 		device_printf(sc->ale_dev,
1129 		    "could not load DMA'able memory for Tx ring.\n");
1130 		goto fail;
1131 	}
1132 	sc->ale_cdata.ale_tx_ring_paddr = ctx.ale_busaddr;
1133 
1134 	/* Rx pages. */
1135 	for (i = 0; i < ALE_RX_PAGES; i++) {
1136 		error = bus_dmamem_alloc(sc->ale_cdata.ale_rx_page[i].page_tag,
1137 		    (void **)&sc->ale_cdata.ale_rx_page[i].page_addr,
1138 		    BUS_DMA_WAITOK | BUS_DMA_ZERO | BUS_DMA_COHERENT,
1139 		    &sc->ale_cdata.ale_rx_page[i].page_map);
1140 		if (error != 0) {
1141 			device_printf(sc->ale_dev,
1142 			    "could not allocate DMA'able memory for "
1143 			    "Rx page %d.\n", i);
1144 			goto fail;
1145 		}
1146 		ctx.ale_busaddr = 0;
1147 		error = bus_dmamap_load(sc->ale_cdata.ale_rx_page[i].page_tag,
1148 		    sc->ale_cdata.ale_rx_page[i].page_map,
1149 		    sc->ale_cdata.ale_rx_page[i].page_addr,
1150 		    sc->ale_pagesize, ale_dmamap_cb, &ctx, 0);
1151 		if (error != 0 || ctx.ale_busaddr == 0) {
1152 			device_printf(sc->ale_dev,
1153 			    "could not load DMA'able memory for "
1154 			    "Rx page %d.\n", i);
1155 			goto fail;
1156 		}
1157 		sc->ale_cdata.ale_rx_page[i].page_paddr = ctx.ale_busaddr;
1158 	}
1159 
1160 	/* Tx CMB. */
1161 	error = bus_dmamem_alloc(sc->ale_cdata.ale_tx_cmb_tag,
1162 	    (void **)&sc->ale_cdata.ale_tx_cmb,
1163 	    BUS_DMA_WAITOK | BUS_DMA_ZERO | BUS_DMA_COHERENT,
1164 	    &sc->ale_cdata.ale_tx_cmb_map);
1165 	if (error != 0) {
1166 		device_printf(sc->ale_dev,
1167 		    "could not allocate DMA'able memory for Tx CMB.\n");
1168 		goto fail;
1169 	}
1170 	ctx.ale_busaddr = 0;
1171 	error = bus_dmamap_load(sc->ale_cdata.ale_tx_cmb_tag,
1172 	    sc->ale_cdata.ale_tx_cmb_map, sc->ale_cdata.ale_tx_cmb,
1173 	    ALE_TX_CMB_SZ, ale_dmamap_cb, &ctx, 0);
1174 	if (error != 0 || ctx.ale_busaddr == 0) {
1175 		device_printf(sc->ale_dev,
1176 		    "could not load DMA'able memory for Tx CMB.\n");
1177 		goto fail;
1178 	}
1179 	sc->ale_cdata.ale_tx_cmb_paddr = ctx.ale_busaddr;
1180 
1181 	/* Rx CMB. */
1182 	for (i = 0; i < ALE_RX_PAGES; i++) {
1183 		error = bus_dmamem_alloc(sc->ale_cdata.ale_rx_page[i].cmb_tag,
1184 		    (void **)&sc->ale_cdata.ale_rx_page[i].cmb_addr,
1185 		    BUS_DMA_WAITOK | BUS_DMA_ZERO | BUS_DMA_COHERENT,
1186 		    &sc->ale_cdata.ale_rx_page[i].cmb_map);
1187 		if (error != 0) {
1188 			device_printf(sc->ale_dev, "could not allocate "
1189 			    "DMA'able memory for Rx page %d CMB.\n", i);
1190 			goto fail;
1191 		}
1192 		ctx.ale_busaddr = 0;
1193 		error = bus_dmamap_load(sc->ale_cdata.ale_rx_page[i].cmb_tag,
1194 		    sc->ale_cdata.ale_rx_page[i].cmb_map,
1195 		    sc->ale_cdata.ale_rx_page[i].cmb_addr,
1196 		    ALE_RX_CMB_SZ, ale_dmamap_cb, &ctx, 0);
1197 		if (error != 0 || ctx.ale_busaddr == 0) {
1198 			device_printf(sc->ale_dev, "could not load DMA'able "
1199 			    "memory for Rx page %d CMB.\n", i);
1200 			goto fail;
1201 		}
1202 		sc->ale_cdata.ale_rx_page[i].cmb_paddr = ctx.ale_busaddr;
1203 	}
1204 
1205 	/*
1206 	 * Tx descriptors/RXF0/CMB DMA blocks share the same
1207 	 * high address region of 64bit DMA address space.
1208 	 */
1209 	if (lowaddr != BUS_SPACE_MAXADDR_32BIT &&
1210 	    (error = ale_check_boundary(sc)) != 0) {
1211 		device_printf(sc->ale_dev, "4GB boundary crossed, "
1212 		    "switching to 32bit DMA addressing mode.\n");
1213 		ale_dma_free(sc);
1214 		/*
1215 		 * Limit max allowable DMA address space to 32bit
1216 		 * and try again.
1217 		 */
1218 		lowaddr = BUS_SPACE_MAXADDR_32BIT;
1219 		goto again;
1220 	}
1221 
1222 	/*
1223 	 * Create Tx buffer parent tag.
1224 	 * AR81xx allows 64bit DMA addressing of Tx buffers so it
1225 	 * needs separate parent DMA tag as parent DMA address space
1226 	 * could be restricted to be within 32bit address space by
1227 	 * 4GB boundary crossing.
1228 	 */
1229 	error = bus_dma_tag_create(
1230 	    bus_get_dma_tag(sc->ale_dev), /* parent */
1231 	    1, 0,			/* alignment, boundary */
1232 	    BUS_SPACE_MAXADDR,		/* lowaddr */
1233 	    BUS_SPACE_MAXADDR,		/* highaddr */
1234 	    NULL, NULL,			/* filter, filterarg */
1235 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsize */
1236 	    0,				/* nsegments */
1237 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsegsize */
1238 	    0,				/* flags */
1239 	    NULL, NULL,			/* lockfunc, lockarg */
1240 	    &sc->ale_cdata.ale_buffer_tag);
1241 	if (error != 0) {
1242 		device_printf(sc->ale_dev,
1243 		    "could not create parent buffer DMA tag.\n");
1244 		goto fail;
1245 	}
1246 
1247 	/* Create DMA tag for Tx buffers. */
1248 	error = bus_dma_tag_create(
1249 	    sc->ale_cdata.ale_buffer_tag, /* parent */
1250 	    1, 0,			/* alignment, boundary */
1251 	    BUS_SPACE_MAXADDR,		/* lowaddr */
1252 	    BUS_SPACE_MAXADDR,		/* highaddr */
1253 	    NULL, NULL,			/* filter, filterarg */
1254 	    ALE_TSO_MAXSIZE,		/* maxsize */
1255 	    ALE_MAXTXSEGS,		/* nsegments */
1256 	    ALE_TSO_MAXSEGSIZE,		/* maxsegsize */
1257 	    0,				/* flags */
1258 	    NULL, NULL,			/* lockfunc, lockarg */
1259 	    &sc->ale_cdata.ale_tx_tag);
1260 	if (error != 0) {
1261 		device_printf(sc->ale_dev, "could not create Tx DMA tag.\n");
1262 		goto fail;
1263 	}
1264 
1265 	/* Create DMA maps for Tx buffers. */
1266 	for (i = 0; i < ALE_TX_RING_CNT; i++) {
1267 		txd = &sc->ale_cdata.ale_txdesc[i];
1268 		txd->tx_m = NULL;
1269 		txd->tx_dmamap = NULL;
1270 		error = bus_dmamap_create(sc->ale_cdata.ale_tx_tag, 0,
1271 		    &txd->tx_dmamap);
1272 		if (error != 0) {
1273 			device_printf(sc->ale_dev,
1274 			    "could not create Tx dmamap.\n");
1275 			goto fail;
1276 		}
1277 	}
1278 
1279 fail:
1280 	return (error);
1281 }
1282 
1283 static void
ale_dma_free(struct ale_softc * sc)1284 ale_dma_free(struct ale_softc *sc)
1285 {
1286 	struct ale_txdesc *txd;
1287 	int i;
1288 
1289 	/* Tx buffers. */
1290 	if (sc->ale_cdata.ale_tx_tag != NULL) {
1291 		for (i = 0; i < ALE_TX_RING_CNT; i++) {
1292 			txd = &sc->ale_cdata.ale_txdesc[i];
1293 			if (txd->tx_dmamap != NULL) {
1294 				bus_dmamap_destroy(sc->ale_cdata.ale_tx_tag,
1295 				    txd->tx_dmamap);
1296 				txd->tx_dmamap = NULL;
1297 			}
1298 		}
1299 		bus_dma_tag_destroy(sc->ale_cdata.ale_tx_tag);
1300 		sc->ale_cdata.ale_tx_tag = NULL;
1301 	}
1302 	/* Tx descriptor ring. */
1303 	if (sc->ale_cdata.ale_tx_ring_tag != NULL) {
1304 		if (sc->ale_cdata.ale_tx_ring_paddr != 0)
1305 			bus_dmamap_unload(sc->ale_cdata.ale_tx_ring_tag,
1306 			    sc->ale_cdata.ale_tx_ring_map);
1307 		if (sc->ale_cdata.ale_tx_ring != NULL)
1308 			bus_dmamem_free(sc->ale_cdata.ale_tx_ring_tag,
1309 			    sc->ale_cdata.ale_tx_ring,
1310 			    sc->ale_cdata.ale_tx_ring_map);
1311 		sc->ale_cdata.ale_tx_ring_paddr = 0;
1312 		sc->ale_cdata.ale_tx_ring = NULL;
1313 		bus_dma_tag_destroy(sc->ale_cdata.ale_tx_ring_tag);
1314 		sc->ale_cdata.ale_tx_ring_tag = NULL;
1315 	}
1316 	/* Rx page block. */
1317 	for (i = 0; i < ALE_RX_PAGES; i++) {
1318 		if (sc->ale_cdata.ale_rx_page[i].page_tag != NULL) {
1319 			if (sc->ale_cdata.ale_rx_page[i].page_paddr != 0)
1320 				bus_dmamap_unload(
1321 				    sc->ale_cdata.ale_rx_page[i].page_tag,
1322 				    sc->ale_cdata.ale_rx_page[i].page_map);
1323 			if (sc->ale_cdata.ale_rx_page[i].page_addr != NULL)
1324 				bus_dmamem_free(
1325 				    sc->ale_cdata.ale_rx_page[i].page_tag,
1326 				    sc->ale_cdata.ale_rx_page[i].page_addr,
1327 				    sc->ale_cdata.ale_rx_page[i].page_map);
1328 			sc->ale_cdata.ale_rx_page[i].page_paddr = 0;
1329 			sc->ale_cdata.ale_rx_page[i].page_addr = NULL;
1330 			bus_dma_tag_destroy(
1331 			    sc->ale_cdata.ale_rx_page[i].page_tag);
1332 			sc->ale_cdata.ale_rx_page[i].page_tag = NULL;
1333 		}
1334 	}
1335 	/* Rx CMB. */
1336 	for (i = 0; i < ALE_RX_PAGES; i++) {
1337 		if (sc->ale_cdata.ale_rx_page[i].cmb_tag != NULL) {
1338 			if (sc->ale_cdata.ale_rx_page[i].cmb_paddr != 0)
1339 				bus_dmamap_unload(
1340 				    sc->ale_cdata.ale_rx_page[i].cmb_tag,
1341 				    sc->ale_cdata.ale_rx_page[i].cmb_map);
1342 			if (sc->ale_cdata.ale_rx_page[i].cmb_addr != NULL)
1343 				bus_dmamem_free(
1344 				    sc->ale_cdata.ale_rx_page[i].cmb_tag,
1345 				    sc->ale_cdata.ale_rx_page[i].cmb_addr,
1346 				    sc->ale_cdata.ale_rx_page[i].cmb_map);
1347 			sc->ale_cdata.ale_rx_page[i].cmb_paddr = 0;
1348 			sc->ale_cdata.ale_rx_page[i].cmb_addr = NULL;
1349 			bus_dma_tag_destroy(
1350 			    sc->ale_cdata.ale_rx_page[i].cmb_tag);
1351 			sc->ale_cdata.ale_rx_page[i].cmb_tag = NULL;
1352 		}
1353 	}
1354 	/* Tx CMB. */
1355 	if (sc->ale_cdata.ale_tx_cmb_tag != NULL) {
1356 		if (sc->ale_cdata.ale_tx_cmb_paddr != 0)
1357 			bus_dmamap_unload(sc->ale_cdata.ale_tx_cmb_tag,
1358 			    sc->ale_cdata.ale_tx_cmb_map);
1359 		if (sc->ale_cdata.ale_tx_cmb != NULL)
1360 			bus_dmamem_free(sc->ale_cdata.ale_tx_cmb_tag,
1361 			    sc->ale_cdata.ale_tx_cmb,
1362 			    sc->ale_cdata.ale_tx_cmb_map);
1363 		sc->ale_cdata.ale_tx_cmb_paddr = 0;
1364 		sc->ale_cdata.ale_tx_cmb = NULL;
1365 		bus_dma_tag_destroy(sc->ale_cdata.ale_tx_cmb_tag);
1366 		sc->ale_cdata.ale_tx_cmb_tag = NULL;
1367 	}
1368 	if (sc->ale_cdata.ale_buffer_tag != NULL) {
1369 		bus_dma_tag_destroy(sc->ale_cdata.ale_buffer_tag);
1370 		sc->ale_cdata.ale_buffer_tag = NULL;
1371 	}
1372 	if (sc->ale_cdata.ale_parent_tag != NULL) {
1373 		bus_dma_tag_destroy(sc->ale_cdata.ale_parent_tag);
1374 		sc->ale_cdata.ale_parent_tag = NULL;
1375 	}
1376 }
1377 
1378 static int
ale_shutdown(device_t dev)1379 ale_shutdown(device_t dev)
1380 {
1381 
1382 	return (ale_suspend(dev));
1383 }
1384 
1385 /*
1386  * Note, this driver resets the link speed to 10/100Mbps by
1387  * restarting auto-negotiation in suspend/shutdown phase but we
1388  * don't know whether that auto-negotiation would succeed or not
1389  * as driver has no control after powering off/suspend operation.
1390  * If the renegotiation fail WOL may not work. Running at 1Gbps
1391  * will draw more power than 375mA at 3.3V which is specified in
1392  * PCI specification and that would result in complete
1393  * shutdowning power to ethernet controller.
1394  *
1395  * TODO
1396  * Save current negotiated media speed/duplex/flow-control to
1397  * softc and restore the same link again after resuming. PHY
1398  * handling such as power down/resetting to 100Mbps may be better
1399  * handled in suspend method in phy driver.
1400  */
1401 static void
ale_setlinkspeed(struct ale_softc * sc)1402 ale_setlinkspeed(struct ale_softc *sc)
1403 {
1404 	struct mii_data *mii;
1405 	int aneg, i;
1406 
1407 	mii = device_get_softc(sc->ale_miibus);
1408 	mii_pollstat(mii);
1409 	aneg = 0;
1410 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
1411 	    (IFM_ACTIVE | IFM_AVALID)) {
1412 		switch IFM_SUBTYPE(mii->mii_media_active) {
1413 		case IFM_10_T:
1414 		case IFM_100_TX:
1415 			return;
1416 		case IFM_1000_T:
1417 			aneg++;
1418 			break;
1419 		default:
1420 			break;
1421 		}
1422 	}
1423 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr, MII_100T2CR, 0);
1424 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
1425 	    MII_ANAR, ANAR_TX_FD | ANAR_TX | ANAR_10_FD | ANAR_10 | ANAR_CSMA);
1426 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
1427 	    MII_BMCR, BMCR_RESET | BMCR_AUTOEN | BMCR_STARTNEG);
1428 	DELAY(1000);
1429 	if (aneg != 0) {
1430 		/*
1431 		 * Poll link state until ale(4) get a 10/100Mbps link.
1432 		 */
1433 		for (i = 0; i < MII_ANEGTICKS_GIGE; i++) {
1434 			mii_pollstat(mii);
1435 			if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID))
1436 			    == (IFM_ACTIVE | IFM_AVALID)) {
1437 				switch (IFM_SUBTYPE(
1438 				    mii->mii_media_active)) {
1439 				case IFM_10_T:
1440 				case IFM_100_TX:
1441 					ale_mac_config(sc);
1442 					return;
1443 				default:
1444 					break;
1445 				}
1446 			}
1447 			ALE_UNLOCK(sc);
1448 			pause("alelnk", hz);
1449 			ALE_LOCK(sc);
1450 		}
1451 		if (i == MII_ANEGTICKS_GIGE)
1452 			device_printf(sc->ale_dev,
1453 			    "establishing a link failed, WOL may not work!");
1454 	}
1455 	/*
1456 	 * No link, force MAC to have 100Mbps, full-duplex link.
1457 	 * This is the last resort and may/may not work.
1458 	 */
1459 	mii->mii_media_status = IFM_AVALID | IFM_ACTIVE;
1460 	mii->mii_media_active = IFM_ETHER | IFM_100_TX | IFM_FDX;
1461 	ale_mac_config(sc);
1462 }
1463 
1464 static void
ale_setwol(struct ale_softc * sc)1465 ale_setwol(struct ale_softc *sc)
1466 {
1467 	if_t ifp;
1468 	uint32_t reg, pmcs;
1469 
1470 	ALE_LOCK_ASSERT(sc);
1471 
1472 	if (!pci_has_pm(sc->ale_dev)) {
1473 		/* Disable WOL. */
1474 		CSR_WRITE_4(sc, ALE_WOL_CFG, 0);
1475 		reg = CSR_READ_4(sc, ALE_PCIE_PHYMISC);
1476 		reg |= PCIE_PHYMISC_FORCE_RCV_DET;
1477 		CSR_WRITE_4(sc, ALE_PCIE_PHYMISC, reg);
1478 		/* Force PHY power down. */
1479 		CSR_WRITE_2(sc, ALE_GPHY_CTRL,
1480 		    GPHY_CTRL_EXT_RESET | GPHY_CTRL_HIB_EN |
1481 		    GPHY_CTRL_HIB_PULSE | GPHY_CTRL_PHY_PLL_ON |
1482 		    GPHY_CTRL_SEL_ANA_RESET | GPHY_CTRL_PHY_IDDQ |
1483 		    GPHY_CTRL_PCLK_SEL_DIS | GPHY_CTRL_PWDOWN_HW);
1484 		return;
1485 	}
1486 
1487 	ifp = sc->ale_ifp;
1488 	if ((if_getcapenable(ifp) & IFCAP_WOL) != 0) {
1489 		if ((sc->ale_flags & ALE_FLAG_FASTETHER) == 0)
1490 			ale_setlinkspeed(sc);
1491 	}
1492 
1493 	pmcs = 0;
1494 	if ((if_getcapenable(ifp) & IFCAP_WOL_MAGIC) != 0)
1495 		pmcs |= WOL_CFG_MAGIC | WOL_CFG_MAGIC_ENB;
1496 	CSR_WRITE_4(sc, ALE_WOL_CFG, pmcs);
1497 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
1498 	reg &= ~(MAC_CFG_DBG | MAC_CFG_PROMISC | MAC_CFG_ALLMULTI |
1499 	    MAC_CFG_BCAST);
1500 	if ((if_getcapenable(ifp) & IFCAP_WOL_MCAST) != 0)
1501 		reg |= MAC_CFG_ALLMULTI | MAC_CFG_BCAST;
1502 	if ((if_getcapenable(ifp) & IFCAP_WOL) != 0)
1503 		reg |= MAC_CFG_RX_ENB;
1504 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
1505 
1506 	if ((if_getcapenable(ifp) & IFCAP_WOL) == 0) {
1507 		/* WOL disabled, PHY power down. */
1508 		reg = CSR_READ_4(sc, ALE_PCIE_PHYMISC);
1509 		reg |= PCIE_PHYMISC_FORCE_RCV_DET;
1510 		CSR_WRITE_4(sc, ALE_PCIE_PHYMISC, reg);
1511 		CSR_WRITE_2(sc, ALE_GPHY_CTRL,
1512 		    GPHY_CTRL_EXT_RESET | GPHY_CTRL_HIB_EN |
1513 		    GPHY_CTRL_HIB_PULSE | GPHY_CTRL_SEL_ANA_RESET |
1514 		    GPHY_CTRL_PHY_IDDQ | GPHY_CTRL_PCLK_SEL_DIS |
1515 		    GPHY_CTRL_PWDOWN_HW);
1516 	}
1517 	/* Request PME. */
1518 	if ((if_getcapenable(ifp) & IFCAP_WOL) != 0)
1519 		pci_enable_pme(sc->ale_dev);
1520 }
1521 
1522 static int
ale_suspend(device_t dev)1523 ale_suspend(device_t dev)
1524 {
1525 	struct ale_softc *sc;
1526 
1527 	sc = device_get_softc(dev);
1528 
1529 	ALE_LOCK(sc);
1530 	ale_stop(sc);
1531 	ale_setwol(sc);
1532 	ALE_UNLOCK(sc);
1533 
1534 	return (0);
1535 }
1536 
1537 static int
ale_resume(device_t dev)1538 ale_resume(device_t dev)
1539 {
1540 	struct ale_softc *sc;
1541 	if_t ifp;
1542 
1543 	sc = device_get_softc(dev);
1544 
1545 	/* Reset PHY. */
1546 	ALE_LOCK(sc);
1547 	ale_phy_reset(sc);
1548 	ifp = sc->ale_ifp;
1549 	if ((if_getflags(ifp) & IFF_UP) != 0) {
1550 		if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
1551 		ale_init_locked(sc);
1552 	}
1553 	ALE_UNLOCK(sc);
1554 
1555 	return (0);
1556 }
1557 
1558 static int
ale_encap(struct ale_softc * sc,struct mbuf ** m_head)1559 ale_encap(struct ale_softc *sc, struct mbuf **m_head)
1560 {
1561 	struct ale_txdesc *txd, *txd_last;
1562 	struct tx_desc *desc;
1563 	struct mbuf *m;
1564 	struct ip *ip;
1565 	struct tcphdr *tcp;
1566 	bus_dma_segment_t txsegs[ALE_MAXTXSEGS];
1567 	bus_dmamap_t map;
1568 	uint32_t cflags, hdrlen, ip_off, poff, vtag;
1569 	int error, i, nsegs, prod, si;
1570 
1571 	ALE_LOCK_ASSERT(sc);
1572 
1573 	M_ASSERTPKTHDR((*m_head));
1574 
1575 	m = *m_head;
1576 	ip = NULL;
1577 	tcp = NULL;
1578 	cflags = vtag = 0;
1579 	ip_off = poff = 0;
1580 	if ((m->m_pkthdr.csum_flags & (ALE_CSUM_FEATURES | CSUM_TSO)) != 0) {
1581 		/*
1582 		 * AR81xx requires offset of TCP/UDP payload in its Tx
1583 		 * descriptor to perform hardware Tx checksum offload.
1584 		 * Additionally, TSO requires IP/TCP header size and
1585 		 * modification of IP/TCP header in order to make TSO
1586 		 * engine work. This kind of operation takes many CPU
1587 		 * cycles on FreeBSD so fast host CPU is required to
1588 		 * get smooth TSO performance.
1589 		 */
1590 		struct ether_header *eh;
1591 
1592 		if (M_WRITABLE(m) == 0) {
1593 			/* Get a writable copy. */
1594 			m = m_dup(*m_head, M_NOWAIT);
1595 			/* Release original mbufs. */
1596 			m_freem(*m_head);
1597 			if (m == NULL) {
1598 				*m_head = NULL;
1599 				return (ENOBUFS);
1600 			}
1601 			*m_head = m;
1602 		}
1603 
1604 		/*
1605 		 * Buggy-controller requires 4 byte aligned Tx buffer
1606 		 * to make custom checksum offload work.
1607 		 */
1608 		if ((sc->ale_flags & ALE_FLAG_TXCSUM_BUG) != 0 &&
1609 		    (m->m_pkthdr.csum_flags & ALE_CSUM_FEATURES) != 0 &&
1610 		    (mtod(m, intptr_t) & 3) != 0) {
1611 			m = m_defrag(*m_head, M_NOWAIT);
1612 			if (m == NULL) {
1613 				m_freem(*m_head);
1614 				*m_head = NULL;
1615 				return (ENOBUFS);
1616 			}
1617 			*m_head = m;
1618 		}
1619 
1620 		ip_off = sizeof(struct ether_header);
1621 		m = m_pullup(m, ip_off);
1622 		if (m == NULL) {
1623 			*m_head = NULL;
1624 			return (ENOBUFS);
1625 		}
1626 		eh = mtod(m, struct ether_header *);
1627 		/*
1628 		 * Check if hardware VLAN insertion is off.
1629 		 * Additional check for LLC/SNAP frame?
1630 		 */
1631 		if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
1632 			ip_off = sizeof(struct ether_vlan_header);
1633 			m = m_pullup(m, ip_off);
1634 			if (m == NULL) {
1635 				*m_head = NULL;
1636 				return (ENOBUFS);
1637 			}
1638 		}
1639 		m = m_pullup(m, ip_off + sizeof(struct ip));
1640 		if (m == NULL) {
1641 			*m_head = NULL;
1642 			return (ENOBUFS);
1643 		}
1644 		ip = (struct ip *)(mtod(m, char *) + ip_off);
1645 		poff = ip_off + (ip->ip_hl << 2);
1646 		if ((m->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
1647 			/*
1648 			 * XXX
1649 			 * AR81xx requires the first descriptor should
1650 			 * not include any TCP playload for TSO case.
1651 			 * (i.e. ethernet header + IP + TCP header only)
1652 			 * m_pullup(9) above will ensure this too.
1653 			 * However it's not correct if the first mbuf
1654 			 * of the chain does not use cluster.
1655 			 */
1656 			m = m_pullup(m, poff + sizeof(struct tcphdr));
1657 			if (m == NULL) {
1658 				*m_head = NULL;
1659 				return (ENOBUFS);
1660 			}
1661 			ip = (struct ip *)(mtod(m, char *) + ip_off);
1662 			tcp = (struct tcphdr *)(mtod(m, char *) + poff);
1663 			m = m_pullup(m, poff + (tcp->th_off << 2));
1664 			if (m == NULL) {
1665 				*m_head = NULL;
1666 				return (ENOBUFS);
1667 			}
1668 			/*
1669 			 * AR81xx requires IP/TCP header size and offset as
1670 			 * well as TCP pseudo checksum which complicates
1671 			 * TSO configuration. I guess this comes from the
1672 			 * adherence to Microsoft NDIS Large Send
1673 			 * specification which requires insertion of
1674 			 * pseudo checksum by upper stack. The pseudo
1675 			 * checksum that NDIS refers to doesn't include
1676 			 * TCP payload length so ale(4) should recompute
1677 			 * the pseudo checksum here. Hopefully this wouldn't
1678 			 * be much burden on modern CPUs.
1679 			 * Reset IP checksum and recompute TCP pseudo
1680 			 * checksum as NDIS specification said.
1681 			 */
1682 			ip->ip_sum = 0;
1683 			tcp->th_sum = in_pseudo(ip->ip_src.s_addr,
1684 			    ip->ip_dst.s_addr, htons(IPPROTO_TCP));
1685 		}
1686 		*m_head = m;
1687 	}
1688 
1689 	si = prod = sc->ale_cdata.ale_tx_prod;
1690 	txd = &sc->ale_cdata.ale_txdesc[prod];
1691 	txd_last = txd;
1692 	map = txd->tx_dmamap;
1693 
1694 	error =  bus_dmamap_load_mbuf_sg(sc->ale_cdata.ale_tx_tag, map,
1695 	    *m_head, txsegs, &nsegs, 0);
1696 	if (error == EFBIG) {
1697 		m = m_collapse(*m_head, M_NOWAIT, ALE_MAXTXSEGS);
1698 		if (m == NULL) {
1699 			m_freem(*m_head);
1700 			*m_head = NULL;
1701 			return (ENOMEM);
1702 		}
1703 		*m_head = m;
1704 		error = bus_dmamap_load_mbuf_sg(sc->ale_cdata.ale_tx_tag, map,
1705 		    *m_head, txsegs, &nsegs, 0);
1706 		if (error != 0) {
1707 			m_freem(*m_head);
1708 			*m_head = NULL;
1709 			return (error);
1710 		}
1711 	} else if (error != 0)
1712 		return (error);
1713 	if (nsegs == 0) {
1714 		m_freem(*m_head);
1715 		*m_head = NULL;
1716 		return (EIO);
1717 	}
1718 
1719 	/* Check descriptor overrun. */
1720 	if (sc->ale_cdata.ale_tx_cnt + nsegs >= ALE_TX_RING_CNT - 3) {
1721 		bus_dmamap_unload(sc->ale_cdata.ale_tx_tag, map);
1722 		return (ENOBUFS);
1723 	}
1724 	bus_dmamap_sync(sc->ale_cdata.ale_tx_tag, map, BUS_DMASYNC_PREWRITE);
1725 
1726 	m = *m_head;
1727 	if ((m->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
1728 		/* Request TSO and set MSS. */
1729 		cflags |= ALE_TD_TSO;
1730 		cflags |= ((uint32_t)m->m_pkthdr.tso_segsz << ALE_TD_MSS_SHIFT);
1731 		/* Set IP/TCP header size. */
1732 		cflags |= ip->ip_hl << ALE_TD_IPHDR_LEN_SHIFT;
1733 		cflags |= tcp->th_off << ALE_TD_TCPHDR_LEN_SHIFT;
1734 	} else if ((m->m_pkthdr.csum_flags & ALE_CSUM_FEATURES) != 0) {
1735 		/*
1736 		 * AR81xx supports Tx custom checksum offload feature
1737 		 * that offloads single 16bit checksum computation.
1738 		 * So you can choose one among IP, TCP and UDP.
1739 		 * Normally driver sets checksum start/insertion
1740 		 * position from the information of TCP/UDP frame as
1741 		 * TCP/UDP checksum takes more time than that of IP.
1742 		 * However it seems that custom checksum offload
1743 		 * requires 4 bytes aligned Tx buffers due to hardware
1744 		 * bug.
1745 		 * AR81xx also supports explicit Tx checksum computation
1746 		 * if it is told that the size of IP header and TCP
1747 		 * header(for UDP, the header size does not matter
1748 		 * because it's fixed length). However with this scheme
1749 		 * TSO does not work so you have to choose one either
1750 		 * TSO or explicit Tx checksum offload. I chosen TSO
1751 		 * plus custom checksum offload with work-around which
1752 		 * will cover most common usage for this consumer
1753 		 * ethernet controller. The work-around takes a lot of
1754 		 * CPU cycles if Tx buffer is not aligned on 4 bytes
1755 		 * boundary, though.
1756 		 */
1757 		cflags |= ALE_TD_CXSUM;
1758 		/* Set checksum start offset. */
1759 		cflags |= (poff << ALE_TD_CSUM_PLOADOFFSET_SHIFT);
1760 		/* Set checksum insertion position of TCP/UDP. */
1761 		cflags |= ((poff + m->m_pkthdr.csum_data) <<
1762 		    ALE_TD_CSUM_XSUMOFFSET_SHIFT);
1763 	}
1764 
1765 	/* Configure VLAN hardware tag insertion. */
1766 	if ((m->m_flags & M_VLANTAG) != 0) {
1767 		vtag = ALE_TX_VLAN_TAG(m->m_pkthdr.ether_vtag);
1768 		vtag = ((vtag << ALE_TD_VLAN_SHIFT) & ALE_TD_VLAN_MASK);
1769 		cflags |= ALE_TD_INSERT_VLAN_TAG;
1770 	}
1771 
1772 	i = 0;
1773 	if ((m->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
1774 		/*
1775 		 * Make sure the first fragment contains
1776 		 * only ethernet and IP/TCP header with options.
1777 		 */
1778 		hdrlen =  poff + (tcp->th_off << 2);
1779 		desc = &sc->ale_cdata.ale_tx_ring[prod];
1780 		desc->addr = htole64(txsegs[i].ds_addr);
1781 		desc->len = htole32(ALE_TX_BYTES(hdrlen) | vtag);
1782 		desc->flags = htole32(cflags);
1783 		sc->ale_cdata.ale_tx_cnt++;
1784 		ALE_DESC_INC(prod, ALE_TX_RING_CNT);
1785 		if (m->m_len - hdrlen > 0) {
1786 			/* Handle remaining payload of the first fragment. */
1787 			desc = &sc->ale_cdata.ale_tx_ring[prod];
1788 			desc->addr = htole64(txsegs[i].ds_addr + hdrlen);
1789 			desc->len = htole32(ALE_TX_BYTES(m->m_len - hdrlen) |
1790 			    vtag);
1791 			desc->flags = htole32(cflags);
1792 			sc->ale_cdata.ale_tx_cnt++;
1793 			ALE_DESC_INC(prod, ALE_TX_RING_CNT);
1794 		}
1795 		i = 1;
1796 	}
1797 	for (; i < nsegs; i++) {
1798 		desc = &sc->ale_cdata.ale_tx_ring[prod];
1799 		desc->addr = htole64(txsegs[i].ds_addr);
1800 		desc->len = htole32(ALE_TX_BYTES(txsegs[i].ds_len) | vtag);
1801 		desc->flags = htole32(cflags);
1802 		sc->ale_cdata.ale_tx_cnt++;
1803 		ALE_DESC_INC(prod, ALE_TX_RING_CNT);
1804 	}
1805 	/* Update producer index. */
1806 	sc->ale_cdata.ale_tx_prod = prod;
1807 	/* Set TSO header on the first descriptor. */
1808 	if ((m->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
1809 		desc = &sc->ale_cdata.ale_tx_ring[si];
1810 		desc->flags |= htole32(ALE_TD_TSO_HDR);
1811 	}
1812 
1813 	/* Finally set EOP on the last descriptor. */
1814 	prod = (prod + ALE_TX_RING_CNT - 1) % ALE_TX_RING_CNT;
1815 	desc = &sc->ale_cdata.ale_tx_ring[prod];
1816 	desc->flags |= htole32(ALE_TD_EOP);
1817 
1818 	/* Swap dmamap of the first and the last. */
1819 	txd = &sc->ale_cdata.ale_txdesc[prod];
1820 	map = txd_last->tx_dmamap;
1821 	txd_last->tx_dmamap = txd->tx_dmamap;
1822 	txd->tx_dmamap = map;
1823 	txd->tx_m = m;
1824 
1825 	/* Sync descriptors. */
1826 	bus_dmamap_sync(sc->ale_cdata.ale_tx_ring_tag,
1827 	    sc->ale_cdata.ale_tx_ring_map,
1828 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1829 
1830 	return (0);
1831 }
1832 
1833 static void
ale_start(if_t ifp)1834 ale_start(if_t ifp)
1835 {
1836         struct ale_softc *sc;
1837 
1838 	sc = if_getsoftc(ifp);
1839 	ALE_LOCK(sc);
1840 	ale_start_locked(ifp);
1841 	ALE_UNLOCK(sc);
1842 }
1843 
1844 static void
ale_start_locked(if_t ifp)1845 ale_start_locked(if_t ifp)
1846 {
1847         struct ale_softc *sc;
1848         struct mbuf *m_head;
1849 	int enq;
1850 
1851 	sc = if_getsoftc(ifp);
1852 
1853 	ALE_LOCK_ASSERT(sc);
1854 
1855 	/* Reclaim transmitted frames. */
1856 	if (sc->ale_cdata.ale_tx_cnt >= ALE_TX_DESC_HIWAT)
1857 		ale_txeof(sc);
1858 
1859 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1860 	    IFF_DRV_RUNNING || (sc->ale_flags & ALE_FLAG_LINK) == 0)
1861 		return;
1862 
1863 	for (enq = 0; !if_sendq_empty(ifp); ) {
1864 		m_head = if_dequeue(ifp);
1865 		if (m_head == NULL)
1866 			break;
1867 		/*
1868 		 * Pack the data into the transmit ring. If we
1869 		 * don't have room, set the OACTIVE flag and wait
1870 		 * for the NIC to drain the ring.
1871 		 */
1872 		if (ale_encap(sc, &m_head)) {
1873 			if (m_head == NULL)
1874 				break;
1875 			if_sendq_prepend(ifp, m_head);
1876 			if_setdrvflagbits(ifp, IFF_DRV_OACTIVE, 0);
1877 			break;
1878 		}
1879 
1880 		enq++;
1881 		/*
1882 		 * If there's a BPF listener, bounce a copy of this frame
1883 		 * to him.
1884 		 */
1885 		ETHER_BPF_MTAP(ifp, m_head);
1886 	}
1887 
1888 	if (enq > 0) {
1889 		/* Kick. */
1890 		CSR_WRITE_4(sc, ALE_MBOX_TPD_PROD_IDX,
1891 		    sc->ale_cdata.ale_tx_prod);
1892 		/* Set a timeout in case the chip goes out to lunch. */
1893 		sc->ale_watchdog_timer = ALE_TX_TIMEOUT;
1894 	}
1895 }
1896 
1897 static void
ale_watchdog(struct ale_softc * sc)1898 ale_watchdog(struct ale_softc *sc)
1899 {
1900 	if_t ifp;
1901 
1902 	ALE_LOCK_ASSERT(sc);
1903 
1904 	if (sc->ale_watchdog_timer == 0 || --sc->ale_watchdog_timer)
1905 		return;
1906 
1907 	ifp = sc->ale_ifp;
1908 	if ((sc->ale_flags & ALE_FLAG_LINK) == 0) {
1909 		if_printf(sc->ale_ifp, "watchdog timeout (lost link)\n");
1910 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1911 		if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
1912 		ale_init_locked(sc);
1913 		return;
1914 	}
1915 	if_printf(sc->ale_ifp, "watchdog timeout -- resetting\n");
1916 	if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1917 	if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
1918 	ale_init_locked(sc);
1919 	if (!if_sendq_empty(ifp))
1920 		ale_start_locked(ifp);
1921 }
1922 
1923 static int
ale_ioctl(if_t ifp,u_long cmd,caddr_t data)1924 ale_ioctl(if_t ifp, u_long cmd, caddr_t data)
1925 {
1926 	struct ale_softc *sc;
1927 	struct ifreq *ifr;
1928 	struct mii_data *mii;
1929 	int error, mask;
1930 
1931 	sc = if_getsoftc(ifp);
1932 	ifr = (struct ifreq *)data;
1933 	error = 0;
1934 	switch (cmd) {
1935 	case SIOCSIFMTU:
1936 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > ALE_JUMBO_MTU ||
1937 		    ((sc->ale_flags & ALE_FLAG_JUMBO) == 0 &&
1938 		    ifr->ifr_mtu > ETHERMTU))
1939 			error = EINVAL;
1940 		else if (if_getmtu(ifp) != ifr->ifr_mtu) {
1941 			ALE_LOCK(sc);
1942 			if_setmtu(ifp, ifr->ifr_mtu);
1943 			if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0) {
1944 				if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
1945 				ale_init_locked(sc);
1946 			}
1947 			ALE_UNLOCK(sc);
1948 		}
1949 		break;
1950 	case SIOCSIFFLAGS:
1951 		ALE_LOCK(sc);
1952 		if ((if_getflags(ifp) & IFF_UP) != 0) {
1953 			if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0) {
1954 				if (((if_getflags(ifp) ^ sc->ale_if_flags)
1955 				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
1956 					ale_rxfilter(sc);
1957 			} else {
1958 				ale_init_locked(sc);
1959 			}
1960 		} else {
1961 			if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0)
1962 				ale_stop(sc);
1963 		}
1964 		sc->ale_if_flags = if_getflags(ifp);
1965 		ALE_UNLOCK(sc);
1966 		break;
1967 	case SIOCADDMULTI:
1968 	case SIOCDELMULTI:
1969 		ALE_LOCK(sc);
1970 		if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0)
1971 			ale_rxfilter(sc);
1972 		ALE_UNLOCK(sc);
1973 		break;
1974 	case SIOCSIFMEDIA:
1975 	case SIOCGIFMEDIA:
1976 		mii = device_get_softc(sc->ale_miibus);
1977 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, cmd);
1978 		break;
1979 	case SIOCSIFCAP:
1980 		ALE_LOCK(sc);
1981 		mask = ifr->ifr_reqcap ^ if_getcapenable(ifp);
1982 		if ((mask & IFCAP_TXCSUM) != 0 &&
1983 		    (if_getcapabilities(ifp) & IFCAP_TXCSUM) != 0) {
1984 			if_togglecapenable(ifp, IFCAP_TXCSUM);
1985 			if ((if_getcapenable(ifp) & IFCAP_TXCSUM) != 0)
1986 				if_sethwassistbits(ifp, ALE_CSUM_FEATURES, 0);
1987 			else
1988 				if_sethwassistbits(ifp, 0, ALE_CSUM_FEATURES);
1989 		}
1990 		if ((mask & IFCAP_RXCSUM) != 0 &&
1991 		    (if_getcapabilities(ifp) & IFCAP_RXCSUM) != 0)
1992 			if_togglecapenable(ifp, IFCAP_RXCSUM);
1993 		if ((mask & IFCAP_TSO4) != 0 &&
1994 		    (if_getcapabilities(ifp) & IFCAP_TSO4) != 0) {
1995 			if_togglecapenable(ifp, IFCAP_TSO4);
1996 			if ((if_getcapenable(ifp) & IFCAP_TSO4) != 0)
1997 				if_sethwassistbits(ifp, CSUM_TSO, 0);
1998 			else
1999 				if_sethwassistbits(ifp, 0, CSUM_TSO);
2000 		}
2001 
2002 		if ((mask & IFCAP_WOL_MCAST) != 0 &&
2003 		    (if_getcapabilities(ifp) & IFCAP_WOL_MCAST) != 0)
2004 			if_togglecapenable(ifp, IFCAP_WOL_MCAST);
2005 		if ((mask & IFCAP_WOL_MAGIC) != 0 &&
2006 		    (if_getcapabilities(ifp) & IFCAP_WOL_MAGIC) != 0)
2007 			if_togglecapenable(ifp, IFCAP_WOL_MAGIC);
2008 		if ((mask & IFCAP_VLAN_HWCSUM) != 0 &&
2009 		    (if_getcapabilities(ifp) & IFCAP_VLAN_HWCSUM) != 0)
2010 			if_togglecapenable(ifp, IFCAP_VLAN_HWCSUM);
2011 		if ((mask & IFCAP_VLAN_HWTSO) != 0 &&
2012 		    (if_getcapabilities(ifp) & IFCAP_VLAN_HWTSO) != 0)
2013 			if_togglecapenable(ifp, IFCAP_VLAN_HWTSO);
2014 		if ((mask & IFCAP_VLAN_HWTAGGING) != 0 &&
2015 		    (if_getcapabilities(ifp) & IFCAP_VLAN_HWTAGGING) != 0) {
2016 			if_togglecapenable(ifp, IFCAP_VLAN_HWTAGGING);
2017 			if ((if_getcapenable(ifp) & IFCAP_VLAN_HWTAGGING) == 0)
2018 				if_setcapenablebit(ifp, 0, IFCAP_VLAN_HWTSO);
2019 			ale_rxvlan(sc);
2020 		}
2021 		ALE_UNLOCK(sc);
2022 		VLAN_CAPABILITIES(ifp);
2023 		break;
2024 	default:
2025 		error = ether_ioctl(ifp, cmd, data);
2026 		break;
2027 	}
2028 
2029 	return (error);
2030 }
2031 
2032 static void
ale_mac_config(struct ale_softc * sc)2033 ale_mac_config(struct ale_softc *sc)
2034 {
2035 	struct mii_data *mii;
2036 	uint32_t reg;
2037 
2038 	ALE_LOCK_ASSERT(sc);
2039 
2040 	mii = device_get_softc(sc->ale_miibus);
2041 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
2042 	reg &= ~(MAC_CFG_FULL_DUPLEX | MAC_CFG_TX_FC | MAC_CFG_RX_FC |
2043 	    MAC_CFG_SPEED_MASK);
2044 	/* Reprogram MAC with resolved speed/duplex. */
2045 	switch (IFM_SUBTYPE(mii->mii_media_active)) {
2046 	case IFM_10_T:
2047 	case IFM_100_TX:
2048 		reg |= MAC_CFG_SPEED_10_100;
2049 		break;
2050 	case IFM_1000_T:
2051 		reg |= MAC_CFG_SPEED_1000;
2052 		break;
2053 	}
2054 	if ((IFM_OPTIONS(mii->mii_media_active) & IFM_FDX) != 0) {
2055 		reg |= MAC_CFG_FULL_DUPLEX;
2056 		if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_TXPAUSE) != 0)
2057 			reg |= MAC_CFG_TX_FC;
2058 		if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_RXPAUSE) != 0)
2059 			reg |= MAC_CFG_RX_FC;
2060 	}
2061 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2062 }
2063 
2064 static void
ale_stats_clear(struct ale_softc * sc)2065 ale_stats_clear(struct ale_softc *sc)
2066 {
2067 	struct smb sb;
2068 	uint32_t *reg;
2069 	int i;
2070 
2071 	for (reg = &sb.rx_frames, i = 0; reg <= &sb.rx_pkts_filtered; reg++) {
2072 		CSR_READ_4(sc, ALE_RX_MIB_BASE + i);
2073 		i += sizeof(uint32_t);
2074 	}
2075 	/* Read Tx statistics. */
2076 	for (reg = &sb.tx_frames, i = 0; reg <= &sb.tx_mcast_bytes; reg++) {
2077 		CSR_READ_4(sc, ALE_TX_MIB_BASE + i);
2078 		i += sizeof(uint32_t);
2079 	}
2080 }
2081 
2082 static void
ale_stats_update(struct ale_softc * sc)2083 ale_stats_update(struct ale_softc *sc)
2084 {
2085 	struct ale_hw_stats *stat;
2086 	struct smb sb, *smb;
2087 	if_t ifp;
2088 	uint32_t *reg;
2089 	int i;
2090 
2091 	ALE_LOCK_ASSERT(sc);
2092 
2093 	ifp = sc->ale_ifp;
2094 	stat = &sc->ale_stats;
2095 	smb = &sb;
2096 
2097 	/* Read Rx statistics. */
2098 	for (reg = &sb.rx_frames, i = 0; reg <= &sb.rx_pkts_filtered; reg++) {
2099 		*reg = CSR_READ_4(sc, ALE_RX_MIB_BASE + i);
2100 		i += sizeof(uint32_t);
2101 	}
2102 	/* Read Tx statistics. */
2103 	for (reg = &sb.tx_frames, i = 0; reg <= &sb.tx_mcast_bytes; reg++) {
2104 		*reg = CSR_READ_4(sc, ALE_TX_MIB_BASE + i);
2105 		i += sizeof(uint32_t);
2106 	}
2107 
2108 	/* Rx stats. */
2109 	stat->rx_frames += smb->rx_frames;
2110 	stat->rx_bcast_frames += smb->rx_bcast_frames;
2111 	stat->rx_mcast_frames += smb->rx_mcast_frames;
2112 	stat->rx_pause_frames += smb->rx_pause_frames;
2113 	stat->rx_control_frames += smb->rx_control_frames;
2114 	stat->rx_crcerrs += smb->rx_crcerrs;
2115 	stat->rx_lenerrs += smb->rx_lenerrs;
2116 	stat->rx_bytes += smb->rx_bytes;
2117 	stat->rx_runts += smb->rx_runts;
2118 	stat->rx_fragments += smb->rx_fragments;
2119 	stat->rx_pkts_64 += smb->rx_pkts_64;
2120 	stat->rx_pkts_65_127 += smb->rx_pkts_65_127;
2121 	stat->rx_pkts_128_255 += smb->rx_pkts_128_255;
2122 	stat->rx_pkts_256_511 += smb->rx_pkts_256_511;
2123 	stat->rx_pkts_512_1023 += smb->rx_pkts_512_1023;
2124 	stat->rx_pkts_1024_1518 += smb->rx_pkts_1024_1518;
2125 	stat->rx_pkts_1519_max += smb->rx_pkts_1519_max;
2126 	stat->rx_pkts_truncated += smb->rx_pkts_truncated;
2127 	stat->rx_fifo_oflows += smb->rx_fifo_oflows;
2128 	stat->rx_rrs_errs += smb->rx_rrs_errs;
2129 	stat->rx_alignerrs += smb->rx_alignerrs;
2130 	stat->rx_bcast_bytes += smb->rx_bcast_bytes;
2131 	stat->rx_mcast_bytes += smb->rx_mcast_bytes;
2132 	stat->rx_pkts_filtered += smb->rx_pkts_filtered;
2133 
2134 	/* Tx stats. */
2135 	stat->tx_frames += smb->tx_frames;
2136 	stat->tx_bcast_frames += smb->tx_bcast_frames;
2137 	stat->tx_mcast_frames += smb->tx_mcast_frames;
2138 	stat->tx_pause_frames += smb->tx_pause_frames;
2139 	stat->tx_excess_defer += smb->tx_excess_defer;
2140 	stat->tx_control_frames += smb->tx_control_frames;
2141 	stat->tx_deferred += smb->tx_deferred;
2142 	stat->tx_bytes += smb->tx_bytes;
2143 	stat->tx_pkts_64 += smb->tx_pkts_64;
2144 	stat->tx_pkts_65_127 += smb->tx_pkts_65_127;
2145 	stat->tx_pkts_128_255 += smb->tx_pkts_128_255;
2146 	stat->tx_pkts_256_511 += smb->tx_pkts_256_511;
2147 	stat->tx_pkts_512_1023 += smb->tx_pkts_512_1023;
2148 	stat->tx_pkts_1024_1518 += smb->tx_pkts_1024_1518;
2149 	stat->tx_pkts_1519_max += smb->tx_pkts_1519_max;
2150 	stat->tx_single_colls += smb->tx_single_colls;
2151 	stat->tx_multi_colls += smb->tx_multi_colls;
2152 	stat->tx_late_colls += smb->tx_late_colls;
2153 	stat->tx_excess_colls += smb->tx_excess_colls;
2154 	stat->tx_underrun += smb->tx_underrun;
2155 	stat->tx_desc_underrun += smb->tx_desc_underrun;
2156 	stat->tx_lenerrs += smb->tx_lenerrs;
2157 	stat->tx_pkts_truncated += smb->tx_pkts_truncated;
2158 	stat->tx_bcast_bytes += smb->tx_bcast_bytes;
2159 	stat->tx_mcast_bytes += smb->tx_mcast_bytes;
2160 
2161 	/* Update counters in ifnet. */
2162 	if_inc_counter(ifp, IFCOUNTER_OPACKETS, smb->tx_frames);
2163 
2164 	if_inc_counter(ifp, IFCOUNTER_COLLISIONS, smb->tx_single_colls +
2165 	    smb->tx_multi_colls * 2 + smb->tx_late_colls +
2166 	    smb->tx_excess_colls * HDPX_CFG_RETRY_DEFAULT);
2167 
2168 	if_inc_counter(ifp, IFCOUNTER_OERRORS, smb->tx_late_colls +
2169 	    smb->tx_excess_colls + smb->tx_underrun + smb->tx_pkts_truncated);
2170 
2171 	if_inc_counter(ifp, IFCOUNTER_IPACKETS, smb->rx_frames);
2172 
2173 	if_inc_counter(ifp, IFCOUNTER_IERRORS,
2174 	    smb->rx_crcerrs + smb->rx_lenerrs +
2175 	    smb->rx_runts + smb->rx_pkts_truncated +
2176 	    smb->rx_fifo_oflows + smb->rx_rrs_errs +
2177 	    smb->rx_alignerrs);
2178 }
2179 
2180 static int
ale_intr(void * arg)2181 ale_intr(void *arg)
2182 {
2183 	struct ale_softc *sc;
2184 	uint32_t status;
2185 
2186 	sc = (struct ale_softc *)arg;
2187 
2188 	status = CSR_READ_4(sc, ALE_INTR_STATUS);
2189 	if ((status & ALE_INTRS) == 0)
2190 		return (FILTER_STRAY);
2191 	/* Disable interrupts. */
2192 	CSR_WRITE_4(sc, ALE_INTR_STATUS, INTR_DIS_INT);
2193 	taskqueue_enqueue(sc->ale_tq, &sc->ale_int_task);
2194 
2195 	return (FILTER_HANDLED);
2196 }
2197 
2198 static void
ale_int_task(void * arg,int pending)2199 ale_int_task(void *arg, int pending)
2200 {
2201 	struct ale_softc *sc;
2202 	if_t ifp;
2203 	uint32_t status;
2204 	int more;
2205 
2206 	sc = (struct ale_softc *)arg;
2207 
2208 	status = CSR_READ_4(sc, ALE_INTR_STATUS);
2209 	ALE_LOCK(sc);
2210 	if (sc->ale_morework != 0)
2211 		status |= INTR_RX_PKT;
2212 	if ((status & ALE_INTRS) == 0)
2213 		goto done;
2214 
2215 	/* Acknowledge interrupts but still disable interrupts. */
2216 	CSR_WRITE_4(sc, ALE_INTR_STATUS, status | INTR_DIS_INT);
2217 
2218 	ifp = sc->ale_ifp;
2219 	more = 0;
2220 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0) {
2221 		more = ale_rxeof(sc, sc->ale_process_limit);
2222 		if (more == EAGAIN)
2223 			sc->ale_morework = 1;
2224 		else if (more == EIO) {
2225 			sc->ale_stats.reset_brk_seq++;
2226 			if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
2227 			ale_init_locked(sc);
2228 			ALE_UNLOCK(sc);
2229 			return;
2230 		}
2231 
2232 		if ((status & (INTR_DMA_RD_TO_RST | INTR_DMA_WR_TO_RST)) != 0) {
2233 			if ((status & INTR_DMA_RD_TO_RST) != 0)
2234 				device_printf(sc->ale_dev,
2235 				    "DMA read error! -- resetting\n");
2236 			if ((status & INTR_DMA_WR_TO_RST) != 0)
2237 				device_printf(sc->ale_dev,
2238 				    "DMA write error! -- resetting\n");
2239 			if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
2240 			ale_init_locked(sc);
2241 			ALE_UNLOCK(sc);
2242 			return;
2243 		}
2244 		if (!if_sendq_empty(ifp))
2245 			ale_start_locked(ifp);
2246 	}
2247 
2248 	if (more == EAGAIN ||
2249 	    (CSR_READ_4(sc, ALE_INTR_STATUS) & ALE_INTRS) != 0) {
2250 		ALE_UNLOCK(sc);
2251 		taskqueue_enqueue(sc->ale_tq, &sc->ale_int_task);
2252 		return;
2253 	}
2254 
2255 done:
2256 	ALE_UNLOCK(sc);
2257 
2258 	/* Re-enable interrupts. */
2259 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0x7FFFFFFF);
2260 }
2261 
2262 static void
ale_txeof(struct ale_softc * sc)2263 ale_txeof(struct ale_softc *sc)
2264 {
2265 	if_t ifp;
2266 	struct ale_txdesc *txd;
2267 	uint32_t cons, prod;
2268 	int prog;
2269 
2270 	ALE_LOCK_ASSERT(sc);
2271 
2272 	ifp = sc->ale_ifp;
2273 
2274 	if (sc->ale_cdata.ale_tx_cnt == 0)
2275 		return;
2276 
2277 	bus_dmamap_sync(sc->ale_cdata.ale_tx_ring_tag,
2278 	    sc->ale_cdata.ale_tx_ring_map,
2279 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2280 	if ((sc->ale_flags & ALE_FLAG_TXCMB_BUG) == 0) {
2281 		bus_dmamap_sync(sc->ale_cdata.ale_tx_cmb_tag,
2282 		    sc->ale_cdata.ale_tx_cmb_map,
2283 		    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2284 		prod = *sc->ale_cdata.ale_tx_cmb & TPD_CNT_MASK;
2285 	} else
2286 		prod = CSR_READ_2(sc, ALE_TPD_CONS_IDX);
2287 	cons = sc->ale_cdata.ale_tx_cons;
2288 	/*
2289 	 * Go through our Tx list and free mbufs for those
2290 	 * frames which have been transmitted.
2291 	 */
2292 	for (prog = 0; cons != prod; prog++,
2293 	    ALE_DESC_INC(cons, ALE_TX_RING_CNT)) {
2294 		if (sc->ale_cdata.ale_tx_cnt <= 0)
2295 			break;
2296 		prog++;
2297 		if_setdrvflagbits(ifp, 0, IFF_DRV_OACTIVE);
2298 		sc->ale_cdata.ale_tx_cnt--;
2299 		txd = &sc->ale_cdata.ale_txdesc[cons];
2300 		if (txd->tx_m != NULL) {
2301 			/* Reclaim transmitted mbufs. */
2302 			bus_dmamap_sync(sc->ale_cdata.ale_tx_tag,
2303 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2304 			bus_dmamap_unload(sc->ale_cdata.ale_tx_tag,
2305 			    txd->tx_dmamap);
2306 			m_freem(txd->tx_m);
2307 			txd->tx_m = NULL;
2308 		}
2309 	}
2310 
2311 	if (prog > 0) {
2312 		sc->ale_cdata.ale_tx_cons = cons;
2313 		/*
2314 		 * Unarm watchdog timer only when there is no pending
2315 		 * Tx descriptors in queue.
2316 		 */
2317 		if (sc->ale_cdata.ale_tx_cnt == 0)
2318 			sc->ale_watchdog_timer = 0;
2319 	}
2320 }
2321 
2322 static void
ale_rx_update_page(struct ale_softc * sc,struct ale_rx_page ** page,uint32_t length,uint32_t * prod)2323 ale_rx_update_page(struct ale_softc *sc, struct ale_rx_page **page,
2324     uint32_t length, uint32_t *prod)
2325 {
2326 	struct ale_rx_page *rx_page;
2327 
2328 	rx_page = *page;
2329 	/* Update consumer position. */
2330 	rx_page->cons += roundup(length + sizeof(struct rx_rs),
2331 	    ALE_RX_PAGE_ALIGN);
2332 	if (rx_page->cons >= ALE_RX_PAGE_SZ) {
2333 		/*
2334 		 * End of Rx page reached, let hardware reuse
2335 		 * this page.
2336 		 */
2337 		rx_page->cons = 0;
2338 		*rx_page->cmb_addr = 0;
2339 		bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2340 		    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2341 		CSR_WRITE_1(sc, ALE_RXF0_PAGE0 + sc->ale_cdata.ale_rx_curp,
2342 		    RXF_VALID);
2343 		/* Switch to alternate Rx page. */
2344 		sc->ale_cdata.ale_rx_curp ^= 1;
2345 		rx_page = *page =
2346 		    &sc->ale_cdata.ale_rx_page[sc->ale_cdata.ale_rx_curp];
2347 		/* Page flipped, sync CMB and Rx page. */
2348 		bus_dmamap_sync(rx_page->page_tag, rx_page->page_map,
2349 		    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2350 		bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2351 		    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2352 		/* Sync completed, cache updated producer index. */
2353 		*prod = *rx_page->cmb_addr;
2354 	}
2355 }
2356 
2357 /*
2358  * It seems that AR81xx controller can compute partial checksum.
2359  * The partial checksum value can be used to accelerate checksum
2360  * computation for fragmented TCP/UDP packets. Upper network stack
2361  * already takes advantage of the partial checksum value in IP
2362  * reassembly stage. But I'm not sure the correctness of the
2363  * partial hardware checksum assistance due to lack of data sheet.
2364  * In addition, the Rx feature of controller that requires copying
2365  * for every frames effectively nullifies one of most nice offload
2366  * capability of controller.
2367  */
2368 static void
ale_rxcsum(struct ale_softc * sc,struct mbuf * m,uint32_t status)2369 ale_rxcsum(struct ale_softc *sc, struct mbuf *m, uint32_t status)
2370 {
2371 	if_t ifp;
2372 	struct ip *ip;
2373 	char *p;
2374 
2375 	ifp = sc->ale_ifp;
2376 	m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
2377 	if ((status & ALE_RD_IPCSUM_NOK) == 0)
2378 		m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
2379 
2380 	if ((sc->ale_flags & ALE_FLAG_RXCSUM_BUG) == 0) {
2381 		if (((status & ALE_RD_IPV4_FRAG) == 0) &&
2382 		    ((status & (ALE_RD_TCP | ALE_RD_UDP)) != 0) &&
2383 		    ((status & ALE_RD_TCP_UDPCSUM_NOK) == 0)) {
2384 			m->m_pkthdr.csum_flags |=
2385 			    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2386 			m->m_pkthdr.csum_data = 0xffff;
2387 		}
2388 	} else {
2389 		if ((status & (ALE_RD_TCP | ALE_RD_UDP)) != 0 &&
2390 		    (status & ALE_RD_TCP_UDPCSUM_NOK) == 0) {
2391 			p = mtod(m, char *);
2392 			p += ETHER_HDR_LEN;
2393 			if ((status & ALE_RD_802_3) != 0)
2394 				p += LLC_SNAPFRAMELEN;
2395 			if ((if_getcapenable(ifp) & IFCAP_VLAN_HWTAGGING) == 0 &&
2396 			    (status & ALE_RD_VLAN) != 0)
2397 				p += ETHER_VLAN_ENCAP_LEN;
2398 			ip = (struct ip *)p;
2399 			if (ip->ip_off != 0 && (status & ALE_RD_IPV4_DF) == 0)
2400 				return;
2401 			m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
2402 			    CSUM_PSEUDO_HDR;
2403 			m->m_pkthdr.csum_data = 0xffff;
2404 		}
2405 	}
2406 	/*
2407 	 * Don't mark bad checksum for TCP/UDP frames
2408 	 * as fragmented frames may always have set
2409 	 * bad checksummed bit of frame status.
2410 	 */
2411 }
2412 
2413 /* Process received frames. */
2414 static int
ale_rxeof(struct ale_softc * sc,int count)2415 ale_rxeof(struct ale_softc *sc, int count)
2416 {
2417 	struct ale_rx_page *rx_page;
2418 	struct rx_rs *rs;
2419 	if_t ifp;
2420 	struct mbuf *m;
2421 	uint32_t length, prod, seqno, status, vtags;
2422 	int prog;
2423 
2424 	ifp = sc->ale_ifp;
2425 	rx_page = &sc->ale_cdata.ale_rx_page[sc->ale_cdata.ale_rx_curp];
2426 	bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2427 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2428 	bus_dmamap_sync(rx_page->page_tag, rx_page->page_map,
2429 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2430 	/*
2431 	 * Don't directly access producer index as hardware may
2432 	 * update it while Rx handler is in progress. It would
2433 	 * be even better if there is a way to let hardware
2434 	 * know how far driver processed its received frames.
2435 	 * Alternatively, hardware could provide a way to disable
2436 	 * CMB updates until driver acknowledges the end of CMB
2437 	 * access.
2438 	 */
2439 	prod = *rx_page->cmb_addr;
2440 	for (prog = 0; prog < count; prog++) {
2441 		if (rx_page->cons >= prod)
2442 			break;
2443 		rs = (struct rx_rs *)(rx_page->page_addr + rx_page->cons);
2444 		seqno = ALE_RX_SEQNO(le32toh(rs->seqno));
2445 		if (sc->ale_cdata.ale_rx_seqno != seqno) {
2446 			/*
2447 			 * Normally I believe this should not happen unless
2448 			 * severe driver bug or corrupted memory. However
2449 			 * it seems to happen under certain conditions which
2450 			 * is triggered by abrupt Rx events such as initiation
2451 			 * of bulk transfer of remote host. It's not easy to
2452 			 * reproduce this and I doubt it could be related
2453 			 * with FIFO overflow of hardware or activity of Tx
2454 			 * CMB updates. I also remember similar behaviour
2455 			 * seen on RealTek 8139 which uses resembling Rx
2456 			 * scheme.
2457 			 */
2458 			if (bootverbose)
2459 				device_printf(sc->ale_dev,
2460 				    "garbled seq: %u, expected: %u -- "
2461 				    "resetting!\n", seqno,
2462 				    sc->ale_cdata.ale_rx_seqno);
2463 			return (EIO);
2464 		}
2465 		/* Frame received. */
2466 		sc->ale_cdata.ale_rx_seqno++;
2467 		length = ALE_RX_BYTES(le32toh(rs->length));
2468 		status = le32toh(rs->flags);
2469 		if ((status & ALE_RD_ERROR) != 0) {
2470 			/*
2471 			 * We want to pass the following frames to upper
2472 			 * layer regardless of error status of Rx return
2473 			 * status.
2474 			 *
2475 			 *  o IP/TCP/UDP checksum is bad.
2476 			 *  o frame length and protocol specific length
2477 			 *     does not match.
2478 			 */
2479 			if ((status & (ALE_RD_CRC | ALE_RD_CODE |
2480 			    ALE_RD_DRIBBLE | ALE_RD_RUNT | ALE_RD_OFLOW |
2481 			    ALE_RD_TRUNC)) != 0) {
2482 				ale_rx_update_page(sc, &rx_page, length, &prod);
2483 				continue;
2484 			}
2485 		}
2486 		/*
2487 		 * m_devget(9) is major bottle-neck of ale(4)(It comes
2488 		 * from hardware limitation). For jumbo frames we could
2489 		 * get a slightly better performance if driver use
2490 		 * m_getjcl(9) with proper buffer size argument. However
2491 		 * that would make code more complicated and I don't
2492 		 * think users would expect good Rx performance numbers
2493 		 * on these low-end consumer ethernet controller.
2494 		 */
2495 		m = m_devget((char *)(rs + 1), length - ETHER_CRC_LEN,
2496 		    ETHER_ALIGN, ifp, NULL);
2497 		if (m == NULL) {
2498 			if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
2499 			ale_rx_update_page(sc, &rx_page, length, &prod);
2500 			continue;
2501 		}
2502 		if ((if_getcapenable(ifp) & IFCAP_RXCSUM) != 0 &&
2503 		    (status & ALE_RD_IPV4) != 0)
2504 			ale_rxcsum(sc, m, status);
2505 		if ((if_getcapenable(ifp) & IFCAP_VLAN_HWTAGGING) != 0 &&
2506 		    (status & ALE_RD_VLAN) != 0) {
2507 			vtags = ALE_RX_VLAN(le32toh(rs->vtags));
2508 			m->m_pkthdr.ether_vtag = ALE_RX_VLAN_TAG(vtags);
2509 			m->m_flags |= M_VLANTAG;
2510 		}
2511 
2512 		/* Pass it to upper layer. */
2513 		ALE_UNLOCK(sc);
2514 		if_input(ifp, m);
2515 		ALE_LOCK(sc);
2516 
2517 		ale_rx_update_page(sc, &rx_page, length, &prod);
2518 	}
2519 
2520 	return (count > 0 ? 0 : EAGAIN);
2521 }
2522 
2523 static void
ale_tick(void * arg)2524 ale_tick(void *arg)
2525 {
2526 	struct ale_softc *sc;
2527 	struct mii_data *mii;
2528 
2529 	sc = (struct ale_softc *)arg;
2530 
2531 	ALE_LOCK_ASSERT(sc);
2532 
2533 	mii = device_get_softc(sc->ale_miibus);
2534 	mii_tick(mii);
2535 	ale_stats_update(sc);
2536 	/*
2537 	 * Reclaim Tx buffers that have been transferred. It's not
2538 	 * needed here but it would release allocated mbuf chains
2539 	 * faster and limit the maximum delay to a hz.
2540 	 */
2541 	ale_txeof(sc);
2542 	ale_watchdog(sc);
2543 	callout_reset(&sc->ale_tick_ch, hz, ale_tick, sc);
2544 }
2545 
2546 static void
ale_reset(struct ale_softc * sc)2547 ale_reset(struct ale_softc *sc)
2548 {
2549 	uint32_t reg;
2550 	int i;
2551 
2552 	/* Initialize PCIe module. From Linux. */
2553 	CSR_WRITE_4(sc, 0x1008, CSR_READ_4(sc, 0x1008) | 0x8000);
2554 
2555 	CSR_WRITE_4(sc, ALE_MASTER_CFG, MASTER_RESET);
2556 	for (i = ALE_RESET_TIMEOUT; i > 0; i--) {
2557 		DELAY(10);
2558 		if ((CSR_READ_4(sc, ALE_MASTER_CFG) & MASTER_RESET) == 0)
2559 			break;
2560 	}
2561 	if (i == 0)
2562 		device_printf(sc->ale_dev, "master reset timeout!\n");
2563 
2564 	for (i = ALE_RESET_TIMEOUT; i > 0; i--) {
2565 		if ((reg = CSR_READ_4(sc, ALE_IDLE_STATUS)) == 0)
2566 			break;
2567 		DELAY(10);
2568 	}
2569 
2570 	if (i == 0)
2571 		device_printf(sc->ale_dev, "reset timeout(0x%08x)!\n", reg);
2572 }
2573 
2574 static void
ale_init(void * xsc)2575 ale_init(void *xsc)
2576 {
2577 	struct ale_softc *sc;
2578 
2579 	sc = (struct ale_softc *)xsc;
2580 	ALE_LOCK(sc);
2581 	ale_init_locked(sc);
2582 	ALE_UNLOCK(sc);
2583 }
2584 
2585 static void
ale_init_locked(struct ale_softc * sc)2586 ale_init_locked(struct ale_softc *sc)
2587 {
2588 	if_t ifp;
2589 	struct mii_data *mii;
2590 	uint8_t eaddr[ETHER_ADDR_LEN];
2591 	bus_addr_t paddr;
2592 	uint32_t reg, rxf_hi, rxf_lo;
2593 
2594 	ALE_LOCK_ASSERT(sc);
2595 
2596 	ifp = sc->ale_ifp;
2597 	mii = device_get_softc(sc->ale_miibus);
2598 
2599 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0)
2600 		return;
2601 	/*
2602 	 * Cancel any pending I/O.
2603 	 */
2604 	ale_stop(sc);
2605 	/*
2606 	 * Reset the chip to a known state.
2607 	 */
2608 	ale_reset(sc);
2609 	/* Initialize Tx descriptors, DMA memory blocks. */
2610 	ale_init_rx_pages(sc);
2611 	ale_init_tx_ring(sc);
2612 
2613 	/* Reprogram the station address. */
2614 	bcopy(if_getlladdr(ifp), eaddr, ETHER_ADDR_LEN);
2615 	CSR_WRITE_4(sc, ALE_PAR0,
2616 	    eaddr[2] << 24 | eaddr[3] << 16 | eaddr[4] << 8 | eaddr[5]);
2617 	CSR_WRITE_4(sc, ALE_PAR1, eaddr[0] << 8 | eaddr[1]);
2618 	/*
2619 	 * Clear WOL status and disable all WOL feature as WOL
2620 	 * would interfere Rx operation under normal environments.
2621 	 */
2622 	CSR_READ_4(sc, ALE_WOL_CFG);
2623 	CSR_WRITE_4(sc, ALE_WOL_CFG, 0);
2624 	/*
2625 	 * Set Tx descriptor/RXF0/CMB base addresses. They share
2626 	 * the same high address part of DMAable region.
2627 	 */
2628 	paddr = sc->ale_cdata.ale_tx_ring_paddr;
2629 	CSR_WRITE_4(sc, ALE_TPD_ADDR_HI, ALE_ADDR_HI(paddr));
2630 	CSR_WRITE_4(sc, ALE_TPD_ADDR_LO, ALE_ADDR_LO(paddr));
2631 	CSR_WRITE_4(sc, ALE_TPD_CNT,
2632 	    (ALE_TX_RING_CNT << TPD_CNT_SHIFT) & TPD_CNT_MASK);
2633 	/* Set Rx page base address, note we use single queue. */
2634 	paddr = sc->ale_cdata.ale_rx_page[0].page_paddr;
2635 	CSR_WRITE_4(sc, ALE_RXF0_PAGE0_ADDR_LO, ALE_ADDR_LO(paddr));
2636 	paddr = sc->ale_cdata.ale_rx_page[1].page_paddr;
2637 	CSR_WRITE_4(sc, ALE_RXF0_PAGE1_ADDR_LO, ALE_ADDR_LO(paddr));
2638 	/* Set Tx/Rx CMB addresses. */
2639 	paddr = sc->ale_cdata.ale_tx_cmb_paddr;
2640 	CSR_WRITE_4(sc, ALE_TX_CMB_ADDR_LO, ALE_ADDR_LO(paddr));
2641 	paddr = sc->ale_cdata.ale_rx_page[0].cmb_paddr;
2642 	CSR_WRITE_4(sc, ALE_RXF0_CMB0_ADDR_LO, ALE_ADDR_LO(paddr));
2643 	paddr = sc->ale_cdata.ale_rx_page[1].cmb_paddr;
2644 	CSR_WRITE_4(sc, ALE_RXF0_CMB1_ADDR_LO, ALE_ADDR_LO(paddr));
2645 	/* Mark RXF0 is valid. */
2646 	CSR_WRITE_1(sc, ALE_RXF0_PAGE0, RXF_VALID);
2647 	CSR_WRITE_1(sc, ALE_RXF0_PAGE1, RXF_VALID);
2648 	/*
2649 	 * No need to initialize RFX1/RXF2/RXF3. We don't use
2650 	 * multi-queue yet.
2651 	 */
2652 
2653 	/* Set Rx page size, excluding guard frame size. */
2654 	CSR_WRITE_4(sc, ALE_RXF_PAGE_SIZE, ALE_RX_PAGE_SZ);
2655 	/* Tell hardware that we're ready to load DMA blocks. */
2656 	CSR_WRITE_4(sc, ALE_DMA_BLOCK, DMA_BLOCK_LOAD);
2657 
2658 	/* Set Rx/Tx interrupt trigger threshold. */
2659 	CSR_WRITE_4(sc, ALE_INT_TRIG_THRESH, (1 << INT_TRIG_RX_THRESH_SHIFT) |
2660 	    (4 << INT_TRIG_TX_THRESH_SHIFT));
2661 	/*
2662 	 * XXX
2663 	 * Set interrupt trigger timer, its purpose and relation
2664 	 * with interrupt moderation mechanism is not clear yet.
2665 	 */
2666 	CSR_WRITE_4(sc, ALE_INT_TRIG_TIMER,
2667 	    ((ALE_USECS(10) << INT_TRIG_RX_TIMER_SHIFT) |
2668 	    (ALE_USECS(1000) << INT_TRIG_TX_TIMER_SHIFT)));
2669 
2670 	/* Configure interrupt moderation timer. */
2671 	reg = ALE_USECS(sc->ale_int_rx_mod) << IM_TIMER_RX_SHIFT;
2672 	reg |= ALE_USECS(sc->ale_int_tx_mod) << IM_TIMER_TX_SHIFT;
2673 	CSR_WRITE_4(sc, ALE_IM_TIMER, reg);
2674 	reg = CSR_READ_4(sc, ALE_MASTER_CFG);
2675 	reg &= ~(MASTER_CHIP_REV_MASK | MASTER_CHIP_ID_MASK);
2676 	reg &= ~(MASTER_IM_RX_TIMER_ENB | MASTER_IM_TX_TIMER_ENB);
2677 	if (ALE_USECS(sc->ale_int_rx_mod) != 0)
2678 		reg |= MASTER_IM_RX_TIMER_ENB;
2679 	if (ALE_USECS(sc->ale_int_tx_mod) != 0)
2680 		reg |= MASTER_IM_TX_TIMER_ENB;
2681 	CSR_WRITE_4(sc, ALE_MASTER_CFG, reg);
2682 	CSR_WRITE_2(sc, ALE_INTR_CLR_TIMER, ALE_USECS(1000));
2683 
2684 	/* Set Maximum frame size of controller. */
2685 	if (if_getmtu(ifp) < ETHERMTU)
2686 		sc->ale_max_frame_size = ETHERMTU;
2687 	else
2688 		sc->ale_max_frame_size = if_getmtu(ifp);
2689 	sc->ale_max_frame_size += ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN +
2690 	    ETHER_CRC_LEN;
2691 	CSR_WRITE_4(sc, ALE_FRAME_SIZE, sc->ale_max_frame_size);
2692 	/* Configure IPG/IFG parameters. */
2693 	CSR_WRITE_4(sc, ALE_IPG_IFG_CFG,
2694 	    ((IPG_IFG_IPGT_DEFAULT << IPG_IFG_IPGT_SHIFT) & IPG_IFG_IPGT_MASK) |
2695 	    ((IPG_IFG_MIFG_DEFAULT << IPG_IFG_MIFG_SHIFT) & IPG_IFG_MIFG_MASK) |
2696 	    ((IPG_IFG_IPG1_DEFAULT << IPG_IFG_IPG1_SHIFT) & IPG_IFG_IPG1_MASK) |
2697 	    ((IPG_IFG_IPG2_DEFAULT << IPG_IFG_IPG2_SHIFT) & IPG_IFG_IPG2_MASK));
2698 	/* Set parameters for half-duplex media. */
2699 	CSR_WRITE_4(sc, ALE_HDPX_CFG,
2700 	    ((HDPX_CFG_LCOL_DEFAULT << HDPX_CFG_LCOL_SHIFT) &
2701 	    HDPX_CFG_LCOL_MASK) |
2702 	    ((HDPX_CFG_RETRY_DEFAULT << HDPX_CFG_RETRY_SHIFT) &
2703 	    HDPX_CFG_RETRY_MASK) | HDPX_CFG_EXC_DEF_EN |
2704 	    ((HDPX_CFG_ABEBT_DEFAULT << HDPX_CFG_ABEBT_SHIFT) &
2705 	    HDPX_CFG_ABEBT_MASK) |
2706 	    ((HDPX_CFG_JAMIPG_DEFAULT << HDPX_CFG_JAMIPG_SHIFT) &
2707 	    HDPX_CFG_JAMIPG_MASK));
2708 
2709 	/* Configure Tx jumbo frame parameters. */
2710 	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0) {
2711 		if (if_getmtu(ifp) < ETHERMTU)
2712 			reg = sc->ale_max_frame_size;
2713 		else if (if_getmtu(ifp) < 6 * 1024)
2714 			reg = (sc->ale_max_frame_size * 2) / 3;
2715 		else
2716 			reg = sc->ale_max_frame_size / 2;
2717 		CSR_WRITE_4(sc, ALE_TX_JUMBO_THRESH,
2718 		    roundup(reg, TX_JUMBO_THRESH_UNIT) >>
2719 		    TX_JUMBO_THRESH_UNIT_SHIFT);
2720 	}
2721 	/* Configure TxQ. */
2722 	reg = (128 << (sc->ale_dma_rd_burst >> DMA_CFG_RD_BURST_SHIFT))
2723 	    << TXQ_CFG_TX_FIFO_BURST_SHIFT;
2724 	reg |= (TXQ_CFG_TPD_BURST_DEFAULT << TXQ_CFG_TPD_BURST_SHIFT) &
2725 	    TXQ_CFG_TPD_BURST_MASK;
2726 	CSR_WRITE_4(sc, ALE_TXQ_CFG, reg | TXQ_CFG_ENHANCED_MODE | TXQ_CFG_ENB);
2727 
2728 	/* Configure Rx jumbo frame & flow control parameters. */
2729 	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0) {
2730 		reg = roundup(sc->ale_max_frame_size, RX_JUMBO_THRESH_UNIT);
2731 		CSR_WRITE_4(sc, ALE_RX_JUMBO_THRESH,
2732 		    (((reg >> RX_JUMBO_THRESH_UNIT_SHIFT) <<
2733 		    RX_JUMBO_THRESH_MASK_SHIFT) & RX_JUMBO_THRESH_MASK) |
2734 		    ((RX_JUMBO_LKAH_DEFAULT << RX_JUMBO_LKAH_SHIFT) &
2735 		    RX_JUMBO_LKAH_MASK));
2736 		reg = CSR_READ_4(sc, ALE_SRAM_RX_FIFO_LEN);
2737 		rxf_hi = (reg * 7) / 10;
2738 		rxf_lo = (reg * 3)/ 10;
2739 		CSR_WRITE_4(sc, ALE_RX_FIFO_PAUSE_THRESH,
2740 		    ((rxf_lo << RX_FIFO_PAUSE_THRESH_LO_SHIFT) &
2741 		    RX_FIFO_PAUSE_THRESH_LO_MASK) |
2742 		    ((rxf_hi << RX_FIFO_PAUSE_THRESH_HI_SHIFT) &
2743 		    RX_FIFO_PAUSE_THRESH_HI_MASK));
2744 	}
2745 
2746 	/* Disable RSS. */
2747 	CSR_WRITE_4(sc, ALE_RSS_IDT_TABLE0, 0);
2748 	CSR_WRITE_4(sc, ALE_RSS_CPU, 0);
2749 
2750 	/* Configure RxQ. */
2751 	CSR_WRITE_4(sc, ALE_RXQ_CFG,
2752 	    RXQ_CFG_ALIGN_32 | RXQ_CFG_CUT_THROUGH_ENB | RXQ_CFG_ENB);
2753 
2754 	/* Configure DMA parameters. */
2755 	reg = 0;
2756 	if ((sc->ale_flags & ALE_FLAG_TXCMB_BUG) == 0)
2757 		reg |= DMA_CFG_TXCMB_ENB;
2758 	CSR_WRITE_4(sc, ALE_DMA_CFG,
2759 	    DMA_CFG_OUT_ORDER | DMA_CFG_RD_REQ_PRI | DMA_CFG_RCB_64 |
2760 	    sc->ale_dma_rd_burst | reg |
2761 	    sc->ale_dma_wr_burst | DMA_CFG_RXCMB_ENB |
2762 	    ((DMA_CFG_RD_DELAY_CNT_DEFAULT << DMA_CFG_RD_DELAY_CNT_SHIFT) &
2763 	    DMA_CFG_RD_DELAY_CNT_MASK) |
2764 	    ((DMA_CFG_WR_DELAY_CNT_DEFAULT << DMA_CFG_WR_DELAY_CNT_SHIFT) &
2765 	    DMA_CFG_WR_DELAY_CNT_MASK));
2766 
2767 	/*
2768 	 * Hardware can be configured to issue SMB interrupt based
2769 	 * on programmed interval. Since there is a callout that is
2770 	 * invoked for every hz in driver we use that instead of
2771 	 * relying on periodic SMB interrupt.
2772 	 */
2773 	CSR_WRITE_4(sc, ALE_SMB_STAT_TIMER, ALE_USECS(0));
2774 	/* Clear MAC statistics. */
2775 	ale_stats_clear(sc);
2776 
2777 	/*
2778 	 * Configure Tx/Rx MACs.
2779 	 *  - Auto-padding for short frames.
2780 	 *  - Enable CRC generation.
2781 	 *  Actual reconfiguration of MAC for resolved speed/duplex
2782 	 *  is followed after detection of link establishment.
2783 	 *  AR81xx always does checksum computation regardless of
2784 	 *  MAC_CFG_RXCSUM_ENB bit. In fact, setting the bit will
2785 	 *  cause Rx handling issue for fragmented IP datagrams due
2786 	 *  to silicon bug.
2787 	 */
2788 	reg = MAC_CFG_TX_CRC_ENB | MAC_CFG_TX_AUTO_PAD | MAC_CFG_FULL_DUPLEX |
2789 	    ((MAC_CFG_PREAMBLE_DEFAULT << MAC_CFG_PREAMBLE_SHIFT) &
2790 	    MAC_CFG_PREAMBLE_MASK);
2791 	if ((sc->ale_flags & ALE_FLAG_FASTETHER) != 0)
2792 		reg |= MAC_CFG_SPEED_10_100;
2793 	else
2794 		reg |= MAC_CFG_SPEED_1000;
2795 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2796 
2797 	/* Set up the receive filter. */
2798 	ale_rxfilter(sc);
2799 	ale_rxvlan(sc);
2800 
2801 	/* Acknowledge all pending interrupts and clear it. */
2802 	CSR_WRITE_4(sc, ALE_INTR_MASK, ALE_INTRS);
2803 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0xFFFFFFFF);
2804 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0);
2805 
2806 	if_setdrvflagbits(ifp, IFF_DRV_RUNNING, 0);
2807 	if_setdrvflagbits(ifp, 0, IFF_DRV_OACTIVE);
2808 
2809 	sc->ale_flags &= ~ALE_FLAG_LINK;
2810 	/* Switch to the current media. */
2811 	mii_mediachg(mii);
2812 
2813 	callout_reset(&sc->ale_tick_ch, hz, ale_tick, sc);
2814 }
2815 
2816 static void
ale_stop(struct ale_softc * sc)2817 ale_stop(struct ale_softc *sc)
2818 {
2819 	if_t ifp;
2820 	struct ale_txdesc *txd;
2821 	uint32_t reg;
2822 	int i;
2823 
2824 	ALE_LOCK_ASSERT(sc);
2825 	/*
2826 	 * Mark the interface down and cancel the watchdog timer.
2827 	 */
2828 	ifp = sc->ale_ifp;
2829 	if_setdrvflagbits(ifp, 0, (IFF_DRV_RUNNING | IFF_DRV_OACTIVE));
2830 	sc->ale_flags &= ~ALE_FLAG_LINK;
2831 	callout_stop(&sc->ale_tick_ch);
2832 	sc->ale_watchdog_timer = 0;
2833 	ale_stats_update(sc);
2834 	/* Disable interrupts. */
2835 	CSR_WRITE_4(sc, ALE_INTR_MASK, 0);
2836 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0xFFFFFFFF);
2837 	/* Disable queue processing and DMA. */
2838 	reg = CSR_READ_4(sc, ALE_TXQ_CFG);
2839 	reg &= ~TXQ_CFG_ENB;
2840 	CSR_WRITE_4(sc, ALE_TXQ_CFG, reg);
2841 	reg = CSR_READ_4(sc, ALE_RXQ_CFG);
2842 	reg &= ~RXQ_CFG_ENB;
2843 	CSR_WRITE_4(sc, ALE_RXQ_CFG, reg);
2844 	reg = CSR_READ_4(sc, ALE_DMA_CFG);
2845 	reg &= ~(DMA_CFG_TXCMB_ENB | DMA_CFG_RXCMB_ENB);
2846 	CSR_WRITE_4(sc, ALE_DMA_CFG, reg);
2847 	DELAY(1000);
2848 	/* Stop Rx/Tx MACs. */
2849 	ale_stop_mac(sc);
2850 	/* Disable interrupts which might be touched in taskq handler. */
2851 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0xFFFFFFFF);
2852 
2853 	/*
2854 	 * Free TX mbufs still in the queues.
2855 	 */
2856 	for (i = 0; i < ALE_TX_RING_CNT; i++) {
2857 		txd = &sc->ale_cdata.ale_txdesc[i];
2858 		if (txd->tx_m != NULL) {
2859 			bus_dmamap_sync(sc->ale_cdata.ale_tx_tag,
2860 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2861 			bus_dmamap_unload(sc->ale_cdata.ale_tx_tag,
2862 			    txd->tx_dmamap);
2863 			m_freem(txd->tx_m);
2864 			txd->tx_m = NULL;
2865 		}
2866         }
2867 }
2868 
2869 static void
ale_stop_mac(struct ale_softc * sc)2870 ale_stop_mac(struct ale_softc *sc)
2871 {
2872 	uint32_t reg;
2873 	int i;
2874 
2875 	ALE_LOCK_ASSERT(sc);
2876 
2877 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
2878 	if ((reg & (MAC_CFG_TX_ENB | MAC_CFG_RX_ENB)) != 0) {
2879 		reg &= ~(MAC_CFG_TX_ENB | MAC_CFG_RX_ENB);
2880 		CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2881 	}
2882 
2883 	for (i = ALE_TIMEOUT; i > 0; i--) {
2884 		reg = CSR_READ_4(sc, ALE_IDLE_STATUS);
2885 		if (reg == 0)
2886 			break;
2887 		DELAY(10);
2888 	}
2889 	if (i == 0)
2890 		device_printf(sc->ale_dev,
2891 		    "could not disable Tx/Rx MAC(0x%08x)!\n", reg);
2892 }
2893 
2894 static void
ale_init_tx_ring(struct ale_softc * sc)2895 ale_init_tx_ring(struct ale_softc *sc)
2896 {
2897 	struct ale_txdesc *txd;
2898 	int i;
2899 
2900 	ALE_LOCK_ASSERT(sc);
2901 
2902 	sc->ale_cdata.ale_tx_prod = 0;
2903 	sc->ale_cdata.ale_tx_cons = 0;
2904 	sc->ale_cdata.ale_tx_cnt = 0;
2905 
2906 	bzero(sc->ale_cdata.ale_tx_ring, ALE_TX_RING_SZ);
2907 	bzero(sc->ale_cdata.ale_tx_cmb, ALE_TX_CMB_SZ);
2908 	for (i = 0; i < ALE_TX_RING_CNT; i++) {
2909 		txd = &sc->ale_cdata.ale_txdesc[i];
2910 		txd->tx_m = NULL;
2911 	}
2912 	*sc->ale_cdata.ale_tx_cmb = 0;
2913 	bus_dmamap_sync(sc->ale_cdata.ale_tx_cmb_tag,
2914 	    sc->ale_cdata.ale_tx_cmb_map,
2915 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2916 	bus_dmamap_sync(sc->ale_cdata.ale_tx_ring_tag,
2917 	    sc->ale_cdata.ale_tx_ring_map,
2918 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2919 }
2920 
2921 static void
ale_init_rx_pages(struct ale_softc * sc)2922 ale_init_rx_pages(struct ale_softc *sc)
2923 {
2924 	struct ale_rx_page *rx_page;
2925 	int i;
2926 
2927 	ALE_LOCK_ASSERT(sc);
2928 
2929 	sc->ale_morework = 0;
2930 	sc->ale_cdata.ale_rx_seqno = 0;
2931 	sc->ale_cdata.ale_rx_curp = 0;
2932 
2933 	for (i = 0; i < ALE_RX_PAGES; i++) {
2934 		rx_page = &sc->ale_cdata.ale_rx_page[i];
2935 		bzero(rx_page->page_addr, sc->ale_pagesize);
2936 		bzero(rx_page->cmb_addr, ALE_RX_CMB_SZ);
2937 		rx_page->cons = 0;
2938 		*rx_page->cmb_addr = 0;
2939 		bus_dmamap_sync(rx_page->page_tag, rx_page->page_map,
2940 		    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2941 		bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2942 		    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2943 	}
2944 }
2945 
2946 static void
ale_rxvlan(struct ale_softc * sc)2947 ale_rxvlan(struct ale_softc *sc)
2948 {
2949 	if_t ifp;
2950 	uint32_t reg;
2951 
2952 	ALE_LOCK_ASSERT(sc);
2953 
2954 	ifp = sc->ale_ifp;
2955 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
2956 	reg &= ~MAC_CFG_VLAN_TAG_STRIP;
2957 	if ((if_getcapenable(ifp) & IFCAP_VLAN_HWTAGGING) != 0)
2958 		reg |= MAC_CFG_VLAN_TAG_STRIP;
2959 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2960 }
2961 
2962 static u_int
ale_hash_maddr(void * arg,struct sockaddr_dl * sdl,u_int cnt)2963 ale_hash_maddr(void *arg, struct sockaddr_dl *sdl, u_int cnt)
2964 {
2965 	uint32_t crc, *mchash = arg;
2966 
2967 	crc = ether_crc32_be(LLADDR(sdl), ETHER_ADDR_LEN);
2968 	mchash[crc >> 31] |= 1 << ((crc >> 26) & 0x1f);
2969 
2970 	return (1);
2971 }
2972 
2973 static void
ale_rxfilter(struct ale_softc * sc)2974 ale_rxfilter(struct ale_softc *sc)
2975 {
2976 	if_t ifp;
2977 	uint32_t mchash[2];
2978 	uint32_t rxcfg;
2979 
2980 	ALE_LOCK_ASSERT(sc);
2981 
2982 	ifp = sc->ale_ifp;
2983 
2984 	rxcfg = CSR_READ_4(sc, ALE_MAC_CFG);
2985 	rxcfg &= ~(MAC_CFG_ALLMULTI | MAC_CFG_BCAST | MAC_CFG_PROMISC);
2986 	if ((if_getflags(ifp) & IFF_BROADCAST) != 0)
2987 		rxcfg |= MAC_CFG_BCAST;
2988 	if ((if_getflags(ifp) & (IFF_PROMISC | IFF_ALLMULTI)) != 0) {
2989 		if ((if_getflags(ifp) & IFF_PROMISC) != 0)
2990 			rxcfg |= MAC_CFG_PROMISC;
2991 		if ((if_getflags(ifp) & IFF_ALLMULTI) != 0)
2992 			rxcfg |= MAC_CFG_ALLMULTI;
2993 		CSR_WRITE_4(sc, ALE_MAR0, 0xFFFFFFFF);
2994 		CSR_WRITE_4(sc, ALE_MAR1, 0xFFFFFFFF);
2995 		CSR_WRITE_4(sc, ALE_MAC_CFG, rxcfg);
2996 		return;
2997 	}
2998 
2999 	/* Program new filter. */
3000 	bzero(mchash, sizeof(mchash));
3001 	if_foreach_llmaddr(ifp, ale_hash_maddr, &mchash);
3002 
3003 	CSR_WRITE_4(sc, ALE_MAR0, mchash[0]);
3004 	CSR_WRITE_4(sc, ALE_MAR1, mchash[1]);
3005 	CSR_WRITE_4(sc, ALE_MAC_CFG, rxcfg);
3006 }
3007 
3008 static int
sysctl_int_range(SYSCTL_HANDLER_ARGS,int low,int high)3009 sysctl_int_range(SYSCTL_HANDLER_ARGS, int low, int high)
3010 {
3011 	int error, value;
3012 
3013 	if (arg1 == NULL)
3014 		return (EINVAL);
3015 	value = *(int *)arg1;
3016 	error = sysctl_handle_int(oidp, &value, 0, req);
3017 	if (error || req->newptr == NULL)
3018 		return (error);
3019 	if (value < low || value > high)
3020 		return (EINVAL);
3021         *(int *)arg1 = value;
3022 
3023         return (0);
3024 }
3025 
3026 static int
sysctl_hw_ale_proc_limit(SYSCTL_HANDLER_ARGS)3027 sysctl_hw_ale_proc_limit(SYSCTL_HANDLER_ARGS)
3028 {
3029 	return (sysctl_int_range(oidp, arg1, arg2, req,
3030 	    ALE_PROC_MIN, ALE_PROC_MAX));
3031 }
3032 
3033 static int
sysctl_hw_ale_int_mod(SYSCTL_HANDLER_ARGS)3034 sysctl_hw_ale_int_mod(SYSCTL_HANDLER_ARGS)
3035 {
3036 
3037 	return (sysctl_int_range(oidp, arg1, arg2, req,
3038 	    ALE_IM_TIMER_MIN, ALE_IM_TIMER_MAX));
3039 }
3040