xref: /freebsd/sys/dev/ale/if_ale.c (revision 4e62c3cafa4c4e41efd6f87b7fe559cf819cf3e4)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2008, Pyun YongHyeon <yongari@FreeBSD.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice unmodified, this list of conditions, and the following
12  *    disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  */
29 
30 /* Driver for Atheros AR8121/AR8113/AR8114 PCIe Ethernet. */
31 
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/bus.h>
35 #include <sys/endian.h>
36 #include <sys/kernel.h>
37 #include <sys/malloc.h>
38 #include <sys/mbuf.h>
39 #include <sys/module.h>
40 #include <sys/rman.h>
41 #include <sys/queue.h>
42 #include <sys/socket.h>
43 #include <sys/sockio.h>
44 #include <sys/sysctl.h>
45 #include <sys/taskqueue.h>
46 
47 #include <net/bpf.h>
48 #include <net/if.h>
49 #include <net/if_var.h>
50 #include <net/if_arp.h>
51 #include <net/ethernet.h>
52 #include <net/if_dl.h>
53 #include <net/if_llc.h>
54 #include <net/if_media.h>
55 #include <net/if_types.h>
56 #include <net/if_vlan_var.h>
57 
58 #include <netinet/in.h>
59 #include <netinet/in_systm.h>
60 #include <netinet/ip.h>
61 #include <netinet/tcp.h>
62 
63 #include <dev/mii/mii.h>
64 #include <dev/mii/miivar.h>
65 
66 #include <dev/pci/pcireg.h>
67 #include <dev/pci/pcivar.h>
68 
69 #include <machine/bus.h>
70 #include <machine/in_cksum.h>
71 
72 #include <dev/ale/if_alereg.h>
73 #include <dev/ale/if_alevar.h>
74 
75 /* "device miibus" required.  See GENERIC if you get errors here. */
76 #include "miibus_if.h"
77 
78 /* For more information about Tx checksum offload issues see ale_encap(). */
79 #define	ALE_CSUM_FEATURES	(CSUM_TCP | CSUM_UDP)
80 
81 MODULE_DEPEND(ale, pci, 1, 1, 1);
82 MODULE_DEPEND(ale, ether, 1, 1, 1);
83 MODULE_DEPEND(ale, miibus, 1, 1, 1);
84 
85 /* Tunables. */
86 static int msi_disable = 0;
87 static int msix_disable = 0;
88 TUNABLE_INT("hw.ale.msi_disable", &msi_disable);
89 TUNABLE_INT("hw.ale.msix_disable", &msix_disable);
90 
91 /*
92  * Devices supported by this driver.
93  */
94 static const struct ale_dev {
95 	uint16_t	ale_vendorid;
96 	uint16_t	ale_deviceid;
97 	const char	*ale_name;
98 } ale_devs[] = {
99     { VENDORID_ATHEROS, DEVICEID_ATHEROS_AR81XX,
100     "Atheros AR8121/AR8113/AR8114 PCIe Ethernet" },
101 };
102 
103 static int	ale_attach(device_t);
104 static int	ale_check_boundary(struct ale_softc *);
105 static int	ale_detach(device_t);
106 static int	ale_dma_alloc(struct ale_softc *);
107 static void	ale_dma_free(struct ale_softc *);
108 static void	ale_dmamap_cb(void *, bus_dma_segment_t *, int, int);
109 static int	ale_encap(struct ale_softc *, struct mbuf **);
110 static void	ale_get_macaddr(struct ale_softc *);
111 static void	ale_init(void *);
112 static void	ale_init_locked(struct ale_softc *);
113 static void	ale_init_rx_pages(struct ale_softc *);
114 static void	ale_init_tx_ring(struct ale_softc *);
115 static void	ale_int_task(void *, int);
116 static int	ale_intr(void *);
117 static int	ale_ioctl(if_t, u_long, caddr_t);
118 static void	ale_mac_config(struct ale_softc *);
119 static int	ale_miibus_readreg(device_t, int, int);
120 static void	ale_miibus_statchg(device_t);
121 static int	ale_miibus_writereg(device_t, int, int, int);
122 static int	ale_mediachange(if_t);
123 static void	ale_mediastatus(if_t, struct ifmediareq *);
124 static void	ale_phy_reset(struct ale_softc *);
125 static int	ale_probe(device_t);
126 static void	ale_reset(struct ale_softc *);
127 static int	ale_resume(device_t);
128 static void	ale_rx_update_page(struct ale_softc *, struct ale_rx_page **,
129     uint32_t, uint32_t *);
130 static void	ale_rxcsum(struct ale_softc *, struct mbuf *, uint32_t);
131 static int	ale_rxeof(struct ale_softc *sc, int);
132 static void	ale_rxfilter(struct ale_softc *);
133 static void	ale_rxvlan(struct ale_softc *);
134 static void	ale_setlinkspeed(struct ale_softc *);
135 static void	ale_setwol(struct ale_softc *);
136 static int	ale_shutdown(device_t);
137 static void	ale_start(if_t);
138 static void	ale_start_locked(if_t);
139 static void	ale_stats_clear(struct ale_softc *);
140 static void	ale_stats_update(struct ale_softc *);
141 static void	ale_stop(struct ale_softc *);
142 static void	ale_stop_mac(struct ale_softc *);
143 static int	ale_suspend(device_t);
144 static void	ale_sysctl_node(struct ale_softc *);
145 static void	ale_tick(void *);
146 static void	ale_txeof(struct ale_softc *);
147 static void	ale_watchdog(struct ale_softc *);
148 static int	sysctl_int_range(SYSCTL_HANDLER_ARGS, int, int);
149 static int	sysctl_hw_ale_proc_limit(SYSCTL_HANDLER_ARGS);
150 static int	sysctl_hw_ale_int_mod(SYSCTL_HANDLER_ARGS);
151 
152 static device_method_t ale_methods[] = {
153 	/* Device interface. */
154 	DEVMETHOD(device_probe,		ale_probe),
155 	DEVMETHOD(device_attach,	ale_attach),
156 	DEVMETHOD(device_detach,	ale_detach),
157 	DEVMETHOD(device_shutdown,	ale_shutdown),
158 	DEVMETHOD(device_suspend,	ale_suspend),
159 	DEVMETHOD(device_resume,	ale_resume),
160 
161 	/* MII interface. */
162 	DEVMETHOD(miibus_readreg,	ale_miibus_readreg),
163 	DEVMETHOD(miibus_writereg,	ale_miibus_writereg),
164 	DEVMETHOD(miibus_statchg,	ale_miibus_statchg),
165 
166 	DEVMETHOD_END
167 };
168 
169 static driver_t ale_driver = {
170 	"ale",
171 	ale_methods,
172 	sizeof(struct ale_softc)
173 };
174 
175 DRIVER_MODULE(ale, pci, ale_driver, NULL, NULL);
176 MODULE_PNP_INFO("U16:vendor;U16:device;D:#", pci, ale, ale_devs,
177     nitems(ale_devs));
178 DRIVER_MODULE(miibus, ale, miibus_driver, NULL, NULL);
179 
180 static struct resource_spec ale_res_spec_mem[] = {
181 	{ SYS_RES_MEMORY,	PCIR_BAR(0),	RF_ACTIVE },
182 	{ -1,			0,		0 }
183 };
184 
185 static struct resource_spec ale_irq_spec_legacy[] = {
186 	{ SYS_RES_IRQ,		0,		RF_ACTIVE | RF_SHAREABLE },
187 	{ -1,			0,		0 }
188 };
189 
190 static struct resource_spec ale_irq_spec_msi[] = {
191 	{ SYS_RES_IRQ,		1,		RF_ACTIVE },
192 	{ -1,			0,		0 }
193 };
194 
195 static struct resource_spec ale_irq_spec_msix[] = {
196 	{ SYS_RES_IRQ,		1,		RF_ACTIVE },
197 	{ -1,			0,		0 }
198 };
199 
200 static int
ale_miibus_readreg(device_t dev,int phy,int reg)201 ale_miibus_readreg(device_t dev, int phy, int reg)
202 {
203 	struct ale_softc *sc;
204 	uint32_t v;
205 	int i;
206 
207 	sc = device_get_softc(dev);
208 
209 	CSR_WRITE_4(sc, ALE_MDIO, MDIO_OP_EXECUTE | MDIO_OP_READ |
210 	    MDIO_SUP_PREAMBLE | MDIO_CLK_25_4 | MDIO_REG_ADDR(reg));
211 	for (i = ALE_PHY_TIMEOUT; i > 0; i--) {
212 		DELAY(5);
213 		v = CSR_READ_4(sc, ALE_MDIO);
214 		if ((v & (MDIO_OP_EXECUTE | MDIO_OP_BUSY)) == 0)
215 			break;
216 	}
217 
218 	if (i == 0) {
219 		device_printf(sc->ale_dev, "phy read timeout : %d\n", reg);
220 		return (0);
221 	}
222 
223 	return ((v & MDIO_DATA_MASK) >> MDIO_DATA_SHIFT);
224 }
225 
226 static int
ale_miibus_writereg(device_t dev,int phy,int reg,int val)227 ale_miibus_writereg(device_t dev, int phy, int reg, int val)
228 {
229 	struct ale_softc *sc;
230 	uint32_t v;
231 	int i;
232 
233 	sc = device_get_softc(dev);
234 
235 	CSR_WRITE_4(sc, ALE_MDIO, MDIO_OP_EXECUTE | MDIO_OP_WRITE |
236 	    (val & MDIO_DATA_MASK) << MDIO_DATA_SHIFT |
237 	    MDIO_SUP_PREAMBLE | MDIO_CLK_25_4 | MDIO_REG_ADDR(reg));
238 	for (i = ALE_PHY_TIMEOUT; i > 0; i--) {
239 		DELAY(5);
240 		v = CSR_READ_4(sc, ALE_MDIO);
241 		if ((v & (MDIO_OP_EXECUTE | MDIO_OP_BUSY)) == 0)
242 			break;
243 	}
244 
245 	if (i == 0)
246 		device_printf(sc->ale_dev, "phy write timeout : %d\n", reg);
247 
248 	return (0);
249 }
250 
251 static void
ale_miibus_statchg(device_t dev)252 ale_miibus_statchg(device_t dev)
253 {
254 	struct ale_softc *sc;
255 	struct mii_data *mii;
256 	if_t ifp;
257 	uint32_t reg;
258 
259 	sc = device_get_softc(dev);
260 	mii = device_get_softc(sc->ale_miibus);
261 	ifp = sc->ale_ifp;
262 	if (mii == NULL || ifp == NULL ||
263 	    (if_getdrvflags(ifp) & IFF_DRV_RUNNING) == 0)
264 		return;
265 
266 	sc->ale_flags &= ~ALE_FLAG_LINK;
267 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
268 	    (IFM_ACTIVE | IFM_AVALID)) {
269 		switch (IFM_SUBTYPE(mii->mii_media_active)) {
270 		case IFM_10_T:
271 		case IFM_100_TX:
272 			sc->ale_flags |= ALE_FLAG_LINK;
273 			break;
274 		case IFM_1000_T:
275 			if ((sc->ale_flags & ALE_FLAG_FASTETHER) == 0)
276 				sc->ale_flags |= ALE_FLAG_LINK;
277 			break;
278 		default:
279 			break;
280 		}
281 	}
282 
283 	/* Stop Rx/Tx MACs. */
284 	ale_stop_mac(sc);
285 
286 	/* Program MACs with resolved speed/duplex/flow-control. */
287 	if ((sc->ale_flags & ALE_FLAG_LINK) != 0) {
288 		ale_mac_config(sc);
289 		/* Reenable Tx/Rx MACs. */
290 		reg = CSR_READ_4(sc, ALE_MAC_CFG);
291 		reg |= MAC_CFG_TX_ENB | MAC_CFG_RX_ENB;
292 		CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
293 	}
294 }
295 
296 static void
ale_mediastatus(if_t ifp,struct ifmediareq * ifmr)297 ale_mediastatus(if_t ifp, struct ifmediareq *ifmr)
298 {
299 	struct ale_softc *sc;
300 	struct mii_data *mii;
301 
302 	sc = if_getsoftc(ifp);
303 	ALE_LOCK(sc);
304 	if ((if_getflags(ifp) & IFF_UP) == 0) {
305 		ALE_UNLOCK(sc);
306 		return;
307 	}
308 	mii = device_get_softc(sc->ale_miibus);
309 
310 	mii_pollstat(mii);
311 	ifmr->ifm_status = mii->mii_media_status;
312 	ifmr->ifm_active = mii->mii_media_active;
313 	ALE_UNLOCK(sc);
314 }
315 
316 static int
ale_mediachange(if_t ifp)317 ale_mediachange(if_t ifp)
318 {
319 	struct ale_softc *sc;
320 	struct mii_data *mii;
321 	struct mii_softc *miisc;
322 	int error;
323 
324 	sc = if_getsoftc(ifp);
325 	ALE_LOCK(sc);
326 	mii = device_get_softc(sc->ale_miibus);
327 	LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
328 		PHY_RESET(miisc);
329 	error = mii_mediachg(mii);
330 	ALE_UNLOCK(sc);
331 
332 	return (error);
333 }
334 
335 static int
ale_probe(device_t dev)336 ale_probe(device_t dev)
337 {
338 	const struct ale_dev *sp;
339 	int i;
340 	uint16_t vendor, devid;
341 
342 	vendor = pci_get_vendor(dev);
343 	devid = pci_get_device(dev);
344 	sp = ale_devs;
345 	for (i = 0; i < nitems(ale_devs); i++) {
346 		if (vendor == sp->ale_vendorid &&
347 		    devid == sp->ale_deviceid) {
348 			device_set_desc(dev, sp->ale_name);
349 			return (BUS_PROBE_DEFAULT);
350 		}
351 		sp++;
352 	}
353 
354 	return (ENXIO);
355 }
356 
357 static void
ale_get_macaddr(struct ale_softc * sc)358 ale_get_macaddr(struct ale_softc *sc)
359 {
360 	uint32_t ea[2], reg;
361 	int i, vpdc;
362 
363 	reg = CSR_READ_4(sc, ALE_SPI_CTRL);
364 	if ((reg & SPI_VPD_ENB) != 0) {
365 		reg &= ~SPI_VPD_ENB;
366 		CSR_WRITE_4(sc, ALE_SPI_CTRL, reg);
367 	}
368 
369 	if (pci_find_cap(sc->ale_dev, PCIY_VPD, &vpdc) == 0) {
370 		/*
371 		 * PCI VPD capability found, let TWSI reload EEPROM.
372 		 * This will set ethernet address of controller.
373 		 */
374 		CSR_WRITE_4(sc, ALE_TWSI_CTRL, CSR_READ_4(sc, ALE_TWSI_CTRL) |
375 		    TWSI_CTRL_SW_LD_START);
376 		for (i = 100; i > 0; i--) {
377 			DELAY(1000);
378 			reg = CSR_READ_4(sc, ALE_TWSI_CTRL);
379 			if ((reg & TWSI_CTRL_SW_LD_START) == 0)
380 				break;
381 		}
382 		if (i == 0)
383 			device_printf(sc->ale_dev,
384 			    "reloading EEPROM timeout!\n");
385 	} else {
386 		if (bootverbose)
387 			device_printf(sc->ale_dev,
388 			    "PCI VPD capability not found!\n");
389 	}
390 
391 	ea[0] = CSR_READ_4(sc, ALE_PAR0);
392 	ea[1] = CSR_READ_4(sc, ALE_PAR1);
393 	sc->ale_eaddr[0] = (ea[1] >> 8) & 0xFF;
394 	sc->ale_eaddr[1] = (ea[1] >> 0) & 0xFF;
395 	sc->ale_eaddr[2] = (ea[0] >> 24) & 0xFF;
396 	sc->ale_eaddr[3] = (ea[0] >> 16) & 0xFF;
397 	sc->ale_eaddr[4] = (ea[0] >> 8) & 0xFF;
398 	sc->ale_eaddr[5] = (ea[0] >> 0) & 0xFF;
399 }
400 
401 static void
ale_phy_reset(struct ale_softc * sc)402 ale_phy_reset(struct ale_softc *sc)
403 {
404 
405 	/* Reset magic from Linux. */
406 	CSR_WRITE_2(sc, ALE_GPHY_CTRL,
407 	    GPHY_CTRL_HIB_EN | GPHY_CTRL_HIB_PULSE | GPHY_CTRL_SEL_ANA_RESET |
408 	    GPHY_CTRL_PHY_PLL_ON);
409 	DELAY(1000);
410 	CSR_WRITE_2(sc, ALE_GPHY_CTRL,
411 	    GPHY_CTRL_EXT_RESET | GPHY_CTRL_HIB_EN | GPHY_CTRL_HIB_PULSE |
412 	    GPHY_CTRL_SEL_ANA_RESET | GPHY_CTRL_PHY_PLL_ON);
413 	DELAY(1000);
414 
415 #define	ATPHY_DBG_ADDR		0x1D
416 #define	ATPHY_DBG_DATA		0x1E
417 
418 	/* Enable hibernation mode. */
419 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
420 	    ATPHY_DBG_ADDR, 0x0B);
421 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
422 	    ATPHY_DBG_DATA, 0xBC00);
423 	/* Set Class A/B for all modes. */
424 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
425 	    ATPHY_DBG_ADDR, 0x00);
426 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
427 	    ATPHY_DBG_DATA, 0x02EF);
428 	/* Enable 10BT power saving. */
429 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
430 	    ATPHY_DBG_ADDR, 0x12);
431 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
432 	    ATPHY_DBG_DATA, 0x4C04);
433 	/* Adjust 1000T power. */
434 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
435 	    ATPHY_DBG_ADDR, 0x04);
436 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
437 	    ATPHY_DBG_ADDR, 0x8BBB);
438 	/* 10BT center tap voltage. */
439 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
440 	    ATPHY_DBG_ADDR, 0x05);
441 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
442 	    ATPHY_DBG_ADDR, 0x2C46);
443 
444 #undef	ATPHY_DBG_ADDR
445 #undef	ATPHY_DBG_DATA
446 	DELAY(1000);
447 }
448 
449 static int
ale_attach(device_t dev)450 ale_attach(device_t dev)
451 {
452 	struct ale_softc *sc;
453 	if_t ifp;
454 	uint16_t burst;
455 	int error, i, msic, msixc, pmc;
456 	uint32_t rxf_len, txf_len;
457 
458 	error = 0;
459 	sc = device_get_softc(dev);
460 	sc->ale_dev = dev;
461 
462 	mtx_init(&sc->ale_mtx, device_get_nameunit(dev), MTX_NETWORK_LOCK,
463 	    MTX_DEF);
464 	callout_init_mtx(&sc->ale_tick_ch, &sc->ale_mtx, 0);
465 	NET_TASK_INIT(&sc->ale_int_task, 0, ale_int_task, sc);
466 
467 	/* Map the device. */
468 	pci_enable_busmaster(dev);
469 	sc->ale_res_spec = ale_res_spec_mem;
470 	sc->ale_irq_spec = ale_irq_spec_legacy;
471 	error = bus_alloc_resources(dev, sc->ale_res_spec, sc->ale_res);
472 	if (error != 0) {
473 		device_printf(dev, "cannot allocate memory resources.\n");
474 		goto fail;
475 	}
476 
477 	/* Set PHY address. */
478 	sc->ale_phyaddr = ALE_PHY_ADDR;
479 
480 	/* Reset PHY. */
481 	ale_phy_reset(sc);
482 
483 	/* Reset the ethernet controller. */
484 	ale_reset(sc);
485 
486 	/* Get PCI and chip id/revision. */
487 	sc->ale_rev = pci_get_revid(dev);
488 	if (sc->ale_rev >= 0xF0) {
489 		/* L2E Rev. B. AR8114 */
490 		sc->ale_flags |= ALE_FLAG_FASTETHER;
491 	} else {
492 		if ((CSR_READ_4(sc, ALE_PHY_STATUS) & PHY_STATUS_100M) != 0) {
493 			/* L1E AR8121 */
494 			sc->ale_flags |= ALE_FLAG_JUMBO;
495 		} else {
496 			/* L2E Rev. A. AR8113 */
497 			sc->ale_flags |= ALE_FLAG_FASTETHER;
498 		}
499 	}
500 	/*
501 	 * All known controllers seems to require 4 bytes alignment
502 	 * of Tx buffers to make Tx checksum offload with custom
503 	 * checksum generation method work.
504 	 */
505 	sc->ale_flags |= ALE_FLAG_TXCSUM_BUG;
506 	/*
507 	 * All known controllers seems to have issues on Rx checksum
508 	 * offload for fragmented IP datagrams.
509 	 */
510 	sc->ale_flags |= ALE_FLAG_RXCSUM_BUG;
511 	/*
512 	 * Don't use Tx CMB. It is known to cause RRS update failure
513 	 * under certain circumstances. Typical phenomenon of the
514 	 * issue would be unexpected sequence number encountered in
515 	 * Rx handler.
516 	 */
517 	sc->ale_flags |= ALE_FLAG_TXCMB_BUG;
518 	sc->ale_chip_rev = CSR_READ_4(sc, ALE_MASTER_CFG) >>
519 	    MASTER_CHIP_REV_SHIFT;
520 	if (bootverbose) {
521 		device_printf(dev, "PCI device revision : 0x%04x\n",
522 		    sc->ale_rev);
523 		device_printf(dev, "Chip id/revision : 0x%04x\n",
524 		    sc->ale_chip_rev);
525 	}
526 	txf_len = CSR_READ_4(sc, ALE_SRAM_TX_FIFO_LEN);
527 	rxf_len = CSR_READ_4(sc, ALE_SRAM_RX_FIFO_LEN);
528 	/*
529 	 * Uninitialized hardware returns an invalid chip id/revision
530 	 * as well as 0xFFFFFFFF for Tx/Rx fifo length.
531 	 */
532 	if (sc->ale_chip_rev == 0xFFFF || txf_len == 0xFFFFFFFF ||
533 	    rxf_len == 0xFFFFFFF) {
534 		device_printf(dev,"chip revision : 0x%04x, %u Tx FIFO "
535 		    "%u Rx FIFO -- not initialized?\n", sc->ale_chip_rev,
536 		    txf_len, rxf_len);
537 		error = ENXIO;
538 		goto fail;
539 	}
540 	device_printf(dev, "%u Tx FIFO, %u Rx FIFO\n", txf_len, rxf_len);
541 
542 	/* Allocate IRQ resources. */
543 	msixc = pci_msix_count(dev);
544 	msic = pci_msi_count(dev);
545 	if (bootverbose) {
546 		device_printf(dev, "MSIX count : %d\n", msixc);
547 		device_printf(dev, "MSI count : %d\n", msic);
548 	}
549 
550 	/* Prefer MSIX over MSI. */
551 	if (msix_disable == 0 || msi_disable == 0) {
552 		if (msix_disable == 0 && msixc == ALE_MSIX_MESSAGES &&
553 		    pci_alloc_msix(dev, &msixc) == 0) {
554 			if (msixc == ALE_MSIX_MESSAGES) {
555 				device_printf(dev, "Using %d MSIX messages.\n",
556 				    msixc);
557 				sc->ale_flags |= ALE_FLAG_MSIX;
558 				sc->ale_irq_spec = ale_irq_spec_msix;
559 			} else
560 				pci_release_msi(dev);
561 		}
562 		if (msi_disable == 0 && (sc->ale_flags & ALE_FLAG_MSIX) == 0 &&
563 		    msic == ALE_MSI_MESSAGES &&
564 		    pci_alloc_msi(dev, &msic) == 0) {
565 			if (msic == ALE_MSI_MESSAGES) {
566 				device_printf(dev, "Using %d MSI messages.\n",
567 				    msic);
568 				sc->ale_flags |= ALE_FLAG_MSI;
569 				sc->ale_irq_spec = ale_irq_spec_msi;
570 			} else
571 				pci_release_msi(dev);
572 		}
573 	}
574 
575 	error = bus_alloc_resources(dev, sc->ale_irq_spec, sc->ale_irq);
576 	if (error != 0) {
577 		device_printf(dev, "cannot allocate IRQ resources.\n");
578 		goto fail;
579 	}
580 
581 	/* Get DMA parameters from PCIe device control register. */
582 	if (pci_find_cap(dev, PCIY_EXPRESS, &i) == 0) {
583 		sc->ale_flags |= ALE_FLAG_PCIE;
584 		burst = pci_read_config(dev, i + 0x08, 2);
585 		/* Max read request size. */
586 		sc->ale_dma_rd_burst = ((burst >> 12) & 0x07) <<
587 		    DMA_CFG_RD_BURST_SHIFT;
588 		/* Max payload size. */
589 		sc->ale_dma_wr_burst = ((burst >> 5) & 0x07) <<
590 		    DMA_CFG_WR_BURST_SHIFT;
591 		if (bootverbose) {
592 			device_printf(dev, "Read request size : %d bytes.\n",
593 			    128 << ((burst >> 12) & 0x07));
594 			device_printf(dev, "TLP payload size : %d bytes.\n",
595 			    128 << ((burst >> 5) & 0x07));
596 		}
597 	} else {
598 		sc->ale_dma_rd_burst = DMA_CFG_RD_BURST_128;
599 		sc->ale_dma_wr_burst = DMA_CFG_WR_BURST_128;
600 	}
601 
602 	/* Create device sysctl node. */
603 	ale_sysctl_node(sc);
604 
605 	if ((error = ale_dma_alloc(sc)) != 0)
606 		goto fail;
607 
608 	/* Load station address. */
609 	ale_get_macaddr(sc);
610 
611 	ifp = sc->ale_ifp = if_alloc(IFT_ETHER);
612 	if_setsoftc(ifp, sc);
613 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
614 	if_setflags(ifp, IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST);
615 	if_setioctlfn(ifp, ale_ioctl);
616 	if_setstartfn(ifp, ale_start);
617 	if_setinitfn(ifp, ale_init);
618 	if_setsendqlen(ifp, ALE_TX_RING_CNT - 1);
619 	if_setsendqready(ifp);
620 	if_setcapabilities(ifp, IFCAP_RXCSUM | IFCAP_TXCSUM | IFCAP_TSO4);
621 	if_sethwassist(ifp, ALE_CSUM_FEATURES | CSUM_TSO);
622 	if (pci_find_cap(dev, PCIY_PMG, &pmc) == 0) {
623 		sc->ale_flags |= ALE_FLAG_PMCAP;
624 		if_setcapabilitiesbit(ifp, IFCAP_WOL_MAGIC | IFCAP_WOL_MCAST, 0);
625 	}
626 	if_setcapenable(ifp, if_getcapabilities(ifp));
627 
628 	/* Set up MII bus. */
629 	error = mii_attach(dev, &sc->ale_miibus, ifp, ale_mediachange,
630 	    ale_mediastatus, BMSR_DEFCAPMASK, sc->ale_phyaddr, MII_OFFSET_ANY,
631 	    MIIF_DOPAUSE);
632 	if (error != 0) {
633 		device_printf(dev, "attaching PHYs failed\n");
634 		goto fail;
635 	}
636 
637 	ether_ifattach(ifp, sc->ale_eaddr);
638 
639 	/* VLAN capability setup. */
640 	if_setcapabilitiesbit(ifp, IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING |
641 	    IFCAP_VLAN_HWCSUM | IFCAP_VLAN_HWTSO, 0);
642 	if_setcapenable(ifp, if_getcapabilities(ifp));
643 	/*
644 	 * Even though controllers supported by ale(3) have Rx checksum
645 	 * offload bug the workaround for fragmented frames seemed to
646 	 * work so far. However it seems Rx checksum offload does not
647 	 * work under certain conditions. So disable Rx checksum offload
648 	 * until I find more clue about it but allow users to override it.
649 	 */
650 	if_setcapenablebit(ifp, 0, IFCAP_RXCSUM);
651 
652 	/* Tell the upper layer(s) we support long frames. */
653 	if_setifheaderlen(ifp, sizeof(struct ether_vlan_header));
654 
655 	/* Create local taskq. */
656 	sc->ale_tq = taskqueue_create_fast("ale_taskq", M_WAITOK,
657 	    taskqueue_thread_enqueue, &sc->ale_tq);
658 	taskqueue_start_threads(&sc->ale_tq, 1, PI_NET, "%s taskq",
659 	    device_get_nameunit(sc->ale_dev));
660 
661 	if ((sc->ale_flags & ALE_FLAG_MSIX) != 0)
662 		msic = ALE_MSIX_MESSAGES;
663 	else if ((sc->ale_flags & ALE_FLAG_MSI) != 0)
664 		msic = ALE_MSI_MESSAGES;
665 	else
666 		msic = 1;
667 	for (i = 0; i < msic; i++) {
668 		error = bus_setup_intr(dev, sc->ale_irq[i],
669 		    INTR_TYPE_NET | INTR_MPSAFE, ale_intr, NULL, sc,
670 		    &sc->ale_intrhand[i]);
671 		if (error != 0)
672 			break;
673 	}
674 	if (error != 0) {
675 		device_printf(dev, "could not set up interrupt handler.\n");
676 		taskqueue_free(sc->ale_tq);
677 		sc->ale_tq = NULL;
678 		ether_ifdetach(ifp);
679 		goto fail;
680 	}
681 
682 fail:
683 	if (error != 0)
684 		ale_detach(dev);
685 
686 	return (error);
687 }
688 
689 static int
ale_detach(device_t dev)690 ale_detach(device_t dev)
691 {
692 	struct ale_softc *sc;
693 	if_t ifp;
694 	int i, msic;
695 
696 	sc = device_get_softc(dev);
697 
698 	ifp = sc->ale_ifp;
699 	if (device_is_attached(dev)) {
700 		ether_ifdetach(ifp);
701 		ALE_LOCK(sc);
702 		ale_stop(sc);
703 		ALE_UNLOCK(sc);
704 		callout_drain(&sc->ale_tick_ch);
705 		taskqueue_drain(sc->ale_tq, &sc->ale_int_task);
706 	}
707 
708 	if (sc->ale_tq != NULL) {
709 		taskqueue_drain(sc->ale_tq, &sc->ale_int_task);
710 		taskqueue_free(sc->ale_tq);
711 		sc->ale_tq = NULL;
712 	}
713 
714 	bus_generic_detach(dev);
715 	ale_dma_free(sc);
716 
717 	if (ifp != NULL) {
718 		if_free(ifp);
719 		sc->ale_ifp = NULL;
720 	}
721 
722 	if ((sc->ale_flags & ALE_FLAG_MSIX) != 0)
723 		msic = ALE_MSIX_MESSAGES;
724 	else if ((sc->ale_flags & ALE_FLAG_MSI) != 0)
725 		msic = ALE_MSI_MESSAGES;
726 	else
727 		msic = 1;
728 	for (i = 0; i < msic; i++) {
729 		if (sc->ale_intrhand[i] != NULL) {
730 			bus_teardown_intr(dev, sc->ale_irq[i],
731 			    sc->ale_intrhand[i]);
732 			sc->ale_intrhand[i] = NULL;
733 		}
734 	}
735 
736 	bus_release_resources(dev, sc->ale_irq_spec, sc->ale_irq);
737 	if ((sc->ale_flags & (ALE_FLAG_MSI | ALE_FLAG_MSIX)) != 0)
738 		pci_release_msi(dev);
739 	bus_release_resources(dev, sc->ale_res_spec, sc->ale_res);
740 	mtx_destroy(&sc->ale_mtx);
741 
742 	return (0);
743 }
744 
745 #define	ALE_SYSCTL_STAT_ADD32(c, h, n, p, d)	\
746 	    SYSCTL_ADD_UINT(c, h, OID_AUTO, n, CTLFLAG_RD, p, 0, d)
747 
748 #define	ALE_SYSCTL_STAT_ADD64(c, h, n, p, d)	\
749 	    SYSCTL_ADD_UQUAD(c, h, OID_AUTO, n, CTLFLAG_RD, p, d)
750 
751 static void
ale_sysctl_node(struct ale_softc * sc)752 ale_sysctl_node(struct ale_softc *sc)
753 {
754 	struct sysctl_ctx_list *ctx;
755 	struct sysctl_oid_list *child, *parent;
756 	struct sysctl_oid *tree;
757 	struct ale_hw_stats *stats;
758 	int error;
759 
760 	stats = &sc->ale_stats;
761 	ctx = device_get_sysctl_ctx(sc->ale_dev);
762 	child = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->ale_dev));
763 
764 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "int_rx_mod",
765 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &sc->ale_int_rx_mod,
766 	    0, sysctl_hw_ale_int_mod, "I", "ale Rx interrupt moderation");
767 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "int_tx_mod",
768 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &sc->ale_int_tx_mod,
769 	    0, sysctl_hw_ale_int_mod, "I", "ale Tx interrupt moderation");
770 	/* Pull in device tunables. */
771 	sc->ale_int_rx_mod = ALE_IM_RX_TIMER_DEFAULT;
772 	error = resource_int_value(device_get_name(sc->ale_dev),
773 	    device_get_unit(sc->ale_dev), "int_rx_mod", &sc->ale_int_rx_mod);
774 	if (error == 0) {
775 		if (sc->ale_int_rx_mod < ALE_IM_TIMER_MIN ||
776 		    sc->ale_int_rx_mod > ALE_IM_TIMER_MAX) {
777 			device_printf(sc->ale_dev, "int_rx_mod value out of "
778 			    "range; using default: %d\n",
779 			    ALE_IM_RX_TIMER_DEFAULT);
780 			sc->ale_int_rx_mod = ALE_IM_RX_TIMER_DEFAULT;
781 		}
782 	}
783 	sc->ale_int_tx_mod = ALE_IM_TX_TIMER_DEFAULT;
784 	error = resource_int_value(device_get_name(sc->ale_dev),
785 	    device_get_unit(sc->ale_dev), "int_tx_mod", &sc->ale_int_tx_mod);
786 	if (error == 0) {
787 		if (sc->ale_int_tx_mod < ALE_IM_TIMER_MIN ||
788 		    sc->ale_int_tx_mod > ALE_IM_TIMER_MAX) {
789 			device_printf(sc->ale_dev, "int_tx_mod value out of "
790 			    "range; using default: %d\n",
791 			    ALE_IM_TX_TIMER_DEFAULT);
792 			sc->ale_int_tx_mod = ALE_IM_TX_TIMER_DEFAULT;
793 		}
794 	}
795 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "process_limit",
796 	    CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
797 	    &sc->ale_process_limit, 0, sysctl_hw_ale_proc_limit, "I",
798 	    "max number of Rx events to process");
799 	/* Pull in device tunables. */
800 	sc->ale_process_limit = ALE_PROC_DEFAULT;
801 	error = resource_int_value(device_get_name(sc->ale_dev),
802 	    device_get_unit(sc->ale_dev), "process_limit",
803 	    &sc->ale_process_limit);
804 	if (error == 0) {
805 		if (sc->ale_process_limit < ALE_PROC_MIN ||
806 		    sc->ale_process_limit > ALE_PROC_MAX) {
807 			device_printf(sc->ale_dev,
808 			    "process_limit value out of range; "
809 			    "using default: %d\n", ALE_PROC_DEFAULT);
810 			sc->ale_process_limit = ALE_PROC_DEFAULT;
811 		}
812 	}
813 
814 	/* Misc statistics. */
815 	ALE_SYSCTL_STAT_ADD32(ctx, child, "reset_brk_seq",
816 	    &stats->reset_brk_seq,
817 	    "Controller resets due to broken Rx sequnce number");
818 
819 	tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "stats",
820 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "ATE statistics");
821 	parent = SYSCTL_CHILDREN(tree);
822 
823 	/* Rx statistics. */
824 	tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "rx",
825 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Rx MAC statistics");
826 	child = SYSCTL_CHILDREN(tree);
827 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_frames",
828 	    &stats->rx_frames, "Good frames");
829 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_bcast_frames",
830 	    &stats->rx_bcast_frames, "Good broadcast frames");
831 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_mcast_frames",
832 	    &stats->rx_mcast_frames, "Good multicast frames");
833 	ALE_SYSCTL_STAT_ADD32(ctx, child, "pause_frames",
834 	    &stats->rx_pause_frames, "Pause control frames");
835 	ALE_SYSCTL_STAT_ADD32(ctx, child, "control_frames",
836 	    &stats->rx_control_frames, "Control frames");
837 	ALE_SYSCTL_STAT_ADD32(ctx, child, "crc_errs",
838 	    &stats->rx_crcerrs, "CRC errors");
839 	ALE_SYSCTL_STAT_ADD32(ctx, child, "len_errs",
840 	    &stats->rx_lenerrs, "Frames with length mismatched");
841 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_octets",
842 	    &stats->rx_bytes, "Good octets");
843 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_bcast_octets",
844 	    &stats->rx_bcast_bytes, "Good broadcast octets");
845 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_mcast_octets",
846 	    &stats->rx_mcast_bytes, "Good multicast octets");
847 	ALE_SYSCTL_STAT_ADD32(ctx, child, "runts",
848 	    &stats->rx_runts, "Too short frames");
849 	ALE_SYSCTL_STAT_ADD32(ctx, child, "fragments",
850 	    &stats->rx_fragments, "Fragmented frames");
851 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_64",
852 	    &stats->rx_pkts_64, "64 bytes frames");
853 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_65_127",
854 	    &stats->rx_pkts_65_127, "65 to 127 bytes frames");
855 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_128_255",
856 	    &stats->rx_pkts_128_255, "128 to 255 bytes frames");
857 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_256_511",
858 	    &stats->rx_pkts_256_511, "256 to 511 bytes frames");
859 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_512_1023",
860 	    &stats->rx_pkts_512_1023, "512 to 1023 bytes frames");
861 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1024_1518",
862 	    &stats->rx_pkts_1024_1518, "1024 to 1518 bytes frames");
863 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1519_max",
864 	    &stats->rx_pkts_1519_max, "1519 to max frames");
865 	ALE_SYSCTL_STAT_ADD32(ctx, child, "trunc_errs",
866 	    &stats->rx_pkts_truncated, "Truncated frames due to MTU size");
867 	ALE_SYSCTL_STAT_ADD32(ctx, child, "fifo_oflows",
868 	    &stats->rx_fifo_oflows, "FIFO overflows");
869 	ALE_SYSCTL_STAT_ADD32(ctx, child, "rrs_errs",
870 	    &stats->rx_rrs_errs, "Return status write-back errors");
871 	ALE_SYSCTL_STAT_ADD32(ctx, child, "align_errs",
872 	    &stats->rx_alignerrs, "Alignment errors");
873 	ALE_SYSCTL_STAT_ADD32(ctx, child, "filtered",
874 	    &stats->rx_pkts_filtered,
875 	    "Frames dropped due to address filtering");
876 
877 	/* Tx statistics. */
878 	tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "tx",
879 	    CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Tx MAC statistics");
880 	child = SYSCTL_CHILDREN(tree);
881 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_frames",
882 	    &stats->tx_frames, "Good frames");
883 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_bcast_frames",
884 	    &stats->tx_bcast_frames, "Good broadcast frames");
885 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_mcast_frames",
886 	    &stats->tx_mcast_frames, "Good multicast frames");
887 	ALE_SYSCTL_STAT_ADD32(ctx, child, "pause_frames",
888 	    &stats->tx_pause_frames, "Pause control frames");
889 	ALE_SYSCTL_STAT_ADD32(ctx, child, "control_frames",
890 	    &stats->tx_control_frames, "Control frames");
891 	ALE_SYSCTL_STAT_ADD32(ctx, child, "excess_defers",
892 	    &stats->tx_excess_defer, "Frames with excessive derferrals");
893 	ALE_SYSCTL_STAT_ADD32(ctx, child, "defers",
894 	    &stats->tx_excess_defer, "Frames with derferrals");
895 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_octets",
896 	    &stats->tx_bytes, "Good octets");
897 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_bcast_octets",
898 	    &stats->tx_bcast_bytes, "Good broadcast octets");
899 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_mcast_octets",
900 	    &stats->tx_mcast_bytes, "Good multicast octets");
901 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_64",
902 	    &stats->tx_pkts_64, "64 bytes frames");
903 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_65_127",
904 	    &stats->tx_pkts_65_127, "65 to 127 bytes frames");
905 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_128_255",
906 	    &stats->tx_pkts_128_255, "128 to 255 bytes frames");
907 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_256_511",
908 	    &stats->tx_pkts_256_511, "256 to 511 bytes frames");
909 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_512_1023",
910 	    &stats->tx_pkts_512_1023, "512 to 1023 bytes frames");
911 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1024_1518",
912 	    &stats->tx_pkts_1024_1518, "1024 to 1518 bytes frames");
913 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1519_max",
914 	    &stats->tx_pkts_1519_max, "1519 to max frames");
915 	ALE_SYSCTL_STAT_ADD32(ctx, child, "single_colls",
916 	    &stats->tx_single_colls, "Single collisions");
917 	ALE_SYSCTL_STAT_ADD32(ctx, child, "multi_colls",
918 	    &stats->tx_multi_colls, "Multiple collisions");
919 	ALE_SYSCTL_STAT_ADD32(ctx, child, "late_colls",
920 	    &stats->tx_late_colls, "Late collisions");
921 	ALE_SYSCTL_STAT_ADD32(ctx, child, "excess_colls",
922 	    &stats->tx_excess_colls, "Excessive collisions");
923 	ALE_SYSCTL_STAT_ADD32(ctx, child, "underruns",
924 	    &stats->tx_underrun, "FIFO underruns");
925 	ALE_SYSCTL_STAT_ADD32(ctx, child, "desc_underruns",
926 	    &stats->tx_desc_underrun, "Descriptor write-back errors");
927 	ALE_SYSCTL_STAT_ADD32(ctx, child, "len_errs",
928 	    &stats->tx_lenerrs, "Frames with length mismatched");
929 	ALE_SYSCTL_STAT_ADD32(ctx, child, "trunc_errs",
930 	    &stats->tx_pkts_truncated, "Truncated frames due to MTU size");
931 }
932 
933 #undef ALE_SYSCTL_STAT_ADD32
934 #undef ALE_SYSCTL_STAT_ADD64
935 
936 struct ale_dmamap_arg {
937 	bus_addr_t	ale_busaddr;
938 };
939 
940 static void
ale_dmamap_cb(void * arg,bus_dma_segment_t * segs,int nsegs,int error)941 ale_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error)
942 {
943 	struct ale_dmamap_arg *ctx;
944 
945 	if (error != 0)
946 		return;
947 
948 	KASSERT(nsegs == 1, ("%s: %d segments returned!", __func__, nsegs));
949 
950 	ctx = (struct ale_dmamap_arg *)arg;
951 	ctx->ale_busaddr = segs[0].ds_addr;
952 }
953 
954 /*
955  * Tx descriptors/RXF0/CMB DMA blocks share ALE_DESC_ADDR_HI register
956  * which specifies high address region of DMA blocks. Therefore these
957  * blocks should have the same high address of given 4GB address
958  * space(i.e. crossing 4GB boundary is not allowed).
959  */
960 static int
ale_check_boundary(struct ale_softc * sc)961 ale_check_boundary(struct ale_softc *sc)
962 {
963 	bus_addr_t rx_cmb_end[ALE_RX_PAGES], tx_cmb_end;
964 	bus_addr_t rx_page_end[ALE_RX_PAGES], tx_ring_end;
965 
966 	rx_page_end[0] = sc->ale_cdata.ale_rx_page[0].page_paddr +
967 	    sc->ale_pagesize;
968 	rx_page_end[1] = sc->ale_cdata.ale_rx_page[1].page_paddr +
969 	    sc->ale_pagesize;
970 	tx_ring_end = sc->ale_cdata.ale_tx_ring_paddr + ALE_TX_RING_SZ;
971 	tx_cmb_end = sc->ale_cdata.ale_tx_cmb_paddr + ALE_TX_CMB_SZ;
972 	rx_cmb_end[0] = sc->ale_cdata.ale_rx_page[0].cmb_paddr + ALE_RX_CMB_SZ;
973 	rx_cmb_end[1] = sc->ale_cdata.ale_rx_page[1].cmb_paddr + ALE_RX_CMB_SZ;
974 
975 	if ((ALE_ADDR_HI(tx_ring_end) !=
976 	    ALE_ADDR_HI(sc->ale_cdata.ale_tx_ring_paddr)) ||
977 	    (ALE_ADDR_HI(rx_page_end[0]) !=
978 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[0].page_paddr)) ||
979 	    (ALE_ADDR_HI(rx_page_end[1]) !=
980 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[1].page_paddr)) ||
981 	    (ALE_ADDR_HI(tx_cmb_end) !=
982 	    ALE_ADDR_HI(sc->ale_cdata.ale_tx_cmb_paddr)) ||
983 	    (ALE_ADDR_HI(rx_cmb_end[0]) !=
984 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[0].cmb_paddr)) ||
985 	    (ALE_ADDR_HI(rx_cmb_end[1]) !=
986 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[1].cmb_paddr)))
987 		return (EFBIG);
988 
989 	if ((ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_page_end[0])) ||
990 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_page_end[1])) ||
991 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_cmb_end[0])) ||
992 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_cmb_end[1])) ||
993 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(tx_cmb_end)))
994 		return (EFBIG);
995 
996 	return (0);
997 }
998 
999 static int
ale_dma_alloc(struct ale_softc * sc)1000 ale_dma_alloc(struct ale_softc *sc)
1001 {
1002 	struct ale_txdesc *txd;
1003 	bus_addr_t lowaddr;
1004 	struct ale_dmamap_arg ctx;
1005 	int error, guard_size, i;
1006 
1007 	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0)
1008 		guard_size = ALE_JUMBO_FRAMELEN;
1009 	else
1010 		guard_size = ALE_MAX_FRAMELEN;
1011 	sc->ale_pagesize = roundup(guard_size + ALE_RX_PAGE_SZ,
1012 	    ALE_RX_PAGE_ALIGN);
1013 	lowaddr = BUS_SPACE_MAXADDR;
1014 again:
1015 	/* Create parent DMA tag. */
1016 	error = bus_dma_tag_create(
1017 	    bus_get_dma_tag(sc->ale_dev), /* parent */
1018 	    1, 0,			/* alignment, boundary */
1019 	    lowaddr,			/* lowaddr */
1020 	    BUS_SPACE_MAXADDR,		/* highaddr */
1021 	    NULL, NULL,			/* filter, filterarg */
1022 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsize */
1023 	    0,				/* nsegments */
1024 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsegsize */
1025 	    0,				/* flags */
1026 	    NULL, NULL,			/* lockfunc, lockarg */
1027 	    &sc->ale_cdata.ale_parent_tag);
1028 	if (error != 0) {
1029 		device_printf(sc->ale_dev,
1030 		    "could not create parent DMA tag.\n");
1031 		goto fail;
1032 	}
1033 
1034 	/* Create DMA tag for Tx descriptor ring. */
1035 	error = bus_dma_tag_create(
1036 	    sc->ale_cdata.ale_parent_tag, /* parent */
1037 	    ALE_TX_RING_ALIGN, 0,	/* alignment, boundary */
1038 	    BUS_SPACE_MAXADDR,		/* lowaddr */
1039 	    BUS_SPACE_MAXADDR,		/* highaddr */
1040 	    NULL, NULL,			/* filter, filterarg */
1041 	    ALE_TX_RING_SZ,		/* maxsize */
1042 	    1,				/* nsegments */
1043 	    ALE_TX_RING_SZ,		/* maxsegsize */
1044 	    0,				/* flags */
1045 	    NULL, NULL,			/* lockfunc, lockarg */
1046 	    &sc->ale_cdata.ale_tx_ring_tag);
1047 	if (error != 0) {
1048 		device_printf(sc->ale_dev,
1049 		    "could not create Tx ring DMA tag.\n");
1050 		goto fail;
1051 	}
1052 
1053 	/* Create DMA tag for Rx pages. */
1054 	for (i = 0; i < ALE_RX_PAGES; i++) {
1055 		error = bus_dma_tag_create(
1056 		    sc->ale_cdata.ale_parent_tag, /* parent */
1057 		    ALE_RX_PAGE_ALIGN, 0,	/* alignment, boundary */
1058 		    BUS_SPACE_MAXADDR,		/* lowaddr */
1059 		    BUS_SPACE_MAXADDR,		/* highaddr */
1060 		    NULL, NULL,			/* filter, filterarg */
1061 		    sc->ale_pagesize,		/* maxsize */
1062 		    1,				/* nsegments */
1063 		    sc->ale_pagesize,		/* maxsegsize */
1064 		    0,				/* flags */
1065 		    NULL, NULL,			/* lockfunc, lockarg */
1066 		    &sc->ale_cdata.ale_rx_page[i].page_tag);
1067 		if (error != 0) {
1068 			device_printf(sc->ale_dev,
1069 			    "could not create Rx page %d DMA tag.\n", i);
1070 			goto fail;
1071 		}
1072 	}
1073 
1074 	/* Create DMA tag for Tx coalescing message block. */
1075 	error = bus_dma_tag_create(
1076 	    sc->ale_cdata.ale_parent_tag, /* parent */
1077 	    ALE_CMB_ALIGN, 0,		/* alignment, boundary */
1078 	    BUS_SPACE_MAXADDR,		/* lowaddr */
1079 	    BUS_SPACE_MAXADDR,		/* highaddr */
1080 	    NULL, NULL,			/* filter, filterarg */
1081 	    ALE_TX_CMB_SZ,		/* maxsize */
1082 	    1,				/* nsegments */
1083 	    ALE_TX_CMB_SZ,		/* maxsegsize */
1084 	    0,				/* flags */
1085 	    NULL, NULL,			/* lockfunc, lockarg */
1086 	    &sc->ale_cdata.ale_tx_cmb_tag);
1087 	if (error != 0) {
1088 		device_printf(sc->ale_dev,
1089 		    "could not create Tx CMB DMA tag.\n");
1090 		goto fail;
1091 	}
1092 
1093 	/* Create DMA tag for Rx coalescing message block. */
1094 	for (i = 0; i < ALE_RX_PAGES; i++) {
1095 		error = bus_dma_tag_create(
1096 		    sc->ale_cdata.ale_parent_tag, /* parent */
1097 		    ALE_CMB_ALIGN, 0,		/* alignment, boundary */
1098 		    BUS_SPACE_MAXADDR,		/* lowaddr */
1099 		    BUS_SPACE_MAXADDR,		/* highaddr */
1100 		    NULL, NULL,			/* filter, filterarg */
1101 		    ALE_RX_CMB_SZ,		/* maxsize */
1102 		    1,				/* nsegments */
1103 		    ALE_RX_CMB_SZ,		/* maxsegsize */
1104 		    0,				/* flags */
1105 		    NULL, NULL,			/* lockfunc, lockarg */
1106 		    &sc->ale_cdata.ale_rx_page[i].cmb_tag);
1107 		if (error != 0) {
1108 			device_printf(sc->ale_dev,
1109 			    "could not create Rx page %d CMB DMA tag.\n", i);
1110 			goto fail;
1111 		}
1112 	}
1113 
1114 	/* Allocate DMA'able memory and load the DMA map for Tx ring. */
1115 	error = bus_dmamem_alloc(sc->ale_cdata.ale_tx_ring_tag,
1116 	    (void **)&sc->ale_cdata.ale_tx_ring,
1117 	    BUS_DMA_WAITOK | BUS_DMA_ZERO | BUS_DMA_COHERENT,
1118 	    &sc->ale_cdata.ale_tx_ring_map);
1119 	if (error != 0) {
1120 		device_printf(sc->ale_dev,
1121 		    "could not allocate DMA'able memory for Tx ring.\n");
1122 		goto fail;
1123 	}
1124 	ctx.ale_busaddr = 0;
1125 	error = bus_dmamap_load(sc->ale_cdata.ale_tx_ring_tag,
1126 	    sc->ale_cdata.ale_tx_ring_map, sc->ale_cdata.ale_tx_ring,
1127 	    ALE_TX_RING_SZ, ale_dmamap_cb, &ctx, 0);
1128 	if (error != 0 || ctx.ale_busaddr == 0) {
1129 		device_printf(sc->ale_dev,
1130 		    "could not load DMA'able memory for Tx ring.\n");
1131 		goto fail;
1132 	}
1133 	sc->ale_cdata.ale_tx_ring_paddr = ctx.ale_busaddr;
1134 
1135 	/* Rx pages. */
1136 	for (i = 0; i < ALE_RX_PAGES; i++) {
1137 		error = bus_dmamem_alloc(sc->ale_cdata.ale_rx_page[i].page_tag,
1138 		    (void **)&sc->ale_cdata.ale_rx_page[i].page_addr,
1139 		    BUS_DMA_WAITOK | BUS_DMA_ZERO | BUS_DMA_COHERENT,
1140 		    &sc->ale_cdata.ale_rx_page[i].page_map);
1141 		if (error != 0) {
1142 			device_printf(sc->ale_dev,
1143 			    "could not allocate DMA'able memory for "
1144 			    "Rx page %d.\n", i);
1145 			goto fail;
1146 		}
1147 		ctx.ale_busaddr = 0;
1148 		error = bus_dmamap_load(sc->ale_cdata.ale_rx_page[i].page_tag,
1149 		    sc->ale_cdata.ale_rx_page[i].page_map,
1150 		    sc->ale_cdata.ale_rx_page[i].page_addr,
1151 		    sc->ale_pagesize, ale_dmamap_cb, &ctx, 0);
1152 		if (error != 0 || ctx.ale_busaddr == 0) {
1153 			device_printf(sc->ale_dev,
1154 			    "could not load DMA'able memory for "
1155 			    "Rx page %d.\n", i);
1156 			goto fail;
1157 		}
1158 		sc->ale_cdata.ale_rx_page[i].page_paddr = ctx.ale_busaddr;
1159 	}
1160 
1161 	/* Tx CMB. */
1162 	error = bus_dmamem_alloc(sc->ale_cdata.ale_tx_cmb_tag,
1163 	    (void **)&sc->ale_cdata.ale_tx_cmb,
1164 	    BUS_DMA_WAITOK | BUS_DMA_ZERO | BUS_DMA_COHERENT,
1165 	    &sc->ale_cdata.ale_tx_cmb_map);
1166 	if (error != 0) {
1167 		device_printf(sc->ale_dev,
1168 		    "could not allocate DMA'able memory for Tx CMB.\n");
1169 		goto fail;
1170 	}
1171 	ctx.ale_busaddr = 0;
1172 	error = bus_dmamap_load(sc->ale_cdata.ale_tx_cmb_tag,
1173 	    sc->ale_cdata.ale_tx_cmb_map, sc->ale_cdata.ale_tx_cmb,
1174 	    ALE_TX_CMB_SZ, ale_dmamap_cb, &ctx, 0);
1175 	if (error != 0 || ctx.ale_busaddr == 0) {
1176 		device_printf(sc->ale_dev,
1177 		    "could not load DMA'able memory for Tx CMB.\n");
1178 		goto fail;
1179 	}
1180 	sc->ale_cdata.ale_tx_cmb_paddr = ctx.ale_busaddr;
1181 
1182 	/* Rx CMB. */
1183 	for (i = 0; i < ALE_RX_PAGES; i++) {
1184 		error = bus_dmamem_alloc(sc->ale_cdata.ale_rx_page[i].cmb_tag,
1185 		    (void **)&sc->ale_cdata.ale_rx_page[i].cmb_addr,
1186 		    BUS_DMA_WAITOK | BUS_DMA_ZERO | BUS_DMA_COHERENT,
1187 		    &sc->ale_cdata.ale_rx_page[i].cmb_map);
1188 		if (error != 0) {
1189 			device_printf(sc->ale_dev, "could not allocate "
1190 			    "DMA'able memory for Rx page %d CMB.\n", i);
1191 			goto fail;
1192 		}
1193 		ctx.ale_busaddr = 0;
1194 		error = bus_dmamap_load(sc->ale_cdata.ale_rx_page[i].cmb_tag,
1195 		    sc->ale_cdata.ale_rx_page[i].cmb_map,
1196 		    sc->ale_cdata.ale_rx_page[i].cmb_addr,
1197 		    ALE_RX_CMB_SZ, ale_dmamap_cb, &ctx, 0);
1198 		if (error != 0 || ctx.ale_busaddr == 0) {
1199 			device_printf(sc->ale_dev, "could not load DMA'able "
1200 			    "memory for Rx page %d CMB.\n", i);
1201 			goto fail;
1202 		}
1203 		sc->ale_cdata.ale_rx_page[i].cmb_paddr = ctx.ale_busaddr;
1204 	}
1205 
1206 	/*
1207 	 * Tx descriptors/RXF0/CMB DMA blocks share the same
1208 	 * high address region of 64bit DMA address space.
1209 	 */
1210 	if (lowaddr != BUS_SPACE_MAXADDR_32BIT &&
1211 	    (error = ale_check_boundary(sc)) != 0) {
1212 		device_printf(sc->ale_dev, "4GB boundary crossed, "
1213 		    "switching to 32bit DMA addressing mode.\n");
1214 		ale_dma_free(sc);
1215 		/*
1216 		 * Limit max allowable DMA address space to 32bit
1217 		 * and try again.
1218 		 */
1219 		lowaddr = BUS_SPACE_MAXADDR_32BIT;
1220 		goto again;
1221 	}
1222 
1223 	/*
1224 	 * Create Tx buffer parent tag.
1225 	 * AR81xx allows 64bit DMA addressing of Tx buffers so it
1226 	 * needs separate parent DMA tag as parent DMA address space
1227 	 * could be restricted to be within 32bit address space by
1228 	 * 4GB boundary crossing.
1229 	 */
1230 	error = bus_dma_tag_create(
1231 	    bus_get_dma_tag(sc->ale_dev), /* parent */
1232 	    1, 0,			/* alignment, boundary */
1233 	    BUS_SPACE_MAXADDR,		/* lowaddr */
1234 	    BUS_SPACE_MAXADDR,		/* highaddr */
1235 	    NULL, NULL,			/* filter, filterarg */
1236 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsize */
1237 	    0,				/* nsegments */
1238 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsegsize */
1239 	    0,				/* flags */
1240 	    NULL, NULL,			/* lockfunc, lockarg */
1241 	    &sc->ale_cdata.ale_buffer_tag);
1242 	if (error != 0) {
1243 		device_printf(sc->ale_dev,
1244 		    "could not create parent buffer DMA tag.\n");
1245 		goto fail;
1246 	}
1247 
1248 	/* Create DMA tag for Tx buffers. */
1249 	error = bus_dma_tag_create(
1250 	    sc->ale_cdata.ale_buffer_tag, /* parent */
1251 	    1, 0,			/* alignment, boundary */
1252 	    BUS_SPACE_MAXADDR,		/* lowaddr */
1253 	    BUS_SPACE_MAXADDR,		/* highaddr */
1254 	    NULL, NULL,			/* filter, filterarg */
1255 	    ALE_TSO_MAXSIZE,		/* maxsize */
1256 	    ALE_MAXTXSEGS,		/* nsegments */
1257 	    ALE_TSO_MAXSEGSIZE,		/* maxsegsize */
1258 	    0,				/* flags */
1259 	    NULL, NULL,			/* lockfunc, lockarg */
1260 	    &sc->ale_cdata.ale_tx_tag);
1261 	if (error != 0) {
1262 		device_printf(sc->ale_dev, "could not create Tx DMA tag.\n");
1263 		goto fail;
1264 	}
1265 
1266 	/* Create DMA maps for Tx buffers. */
1267 	for (i = 0; i < ALE_TX_RING_CNT; i++) {
1268 		txd = &sc->ale_cdata.ale_txdesc[i];
1269 		txd->tx_m = NULL;
1270 		txd->tx_dmamap = NULL;
1271 		error = bus_dmamap_create(sc->ale_cdata.ale_tx_tag, 0,
1272 		    &txd->tx_dmamap);
1273 		if (error != 0) {
1274 			device_printf(sc->ale_dev,
1275 			    "could not create Tx dmamap.\n");
1276 			goto fail;
1277 		}
1278 	}
1279 
1280 fail:
1281 	return (error);
1282 }
1283 
1284 static void
ale_dma_free(struct ale_softc * sc)1285 ale_dma_free(struct ale_softc *sc)
1286 {
1287 	struct ale_txdesc *txd;
1288 	int i;
1289 
1290 	/* Tx buffers. */
1291 	if (sc->ale_cdata.ale_tx_tag != NULL) {
1292 		for (i = 0; i < ALE_TX_RING_CNT; i++) {
1293 			txd = &sc->ale_cdata.ale_txdesc[i];
1294 			if (txd->tx_dmamap != NULL) {
1295 				bus_dmamap_destroy(sc->ale_cdata.ale_tx_tag,
1296 				    txd->tx_dmamap);
1297 				txd->tx_dmamap = NULL;
1298 			}
1299 		}
1300 		bus_dma_tag_destroy(sc->ale_cdata.ale_tx_tag);
1301 		sc->ale_cdata.ale_tx_tag = NULL;
1302 	}
1303 	/* Tx descriptor ring. */
1304 	if (sc->ale_cdata.ale_tx_ring_tag != NULL) {
1305 		if (sc->ale_cdata.ale_tx_ring_paddr != 0)
1306 			bus_dmamap_unload(sc->ale_cdata.ale_tx_ring_tag,
1307 			    sc->ale_cdata.ale_tx_ring_map);
1308 		if (sc->ale_cdata.ale_tx_ring != NULL)
1309 			bus_dmamem_free(sc->ale_cdata.ale_tx_ring_tag,
1310 			    sc->ale_cdata.ale_tx_ring,
1311 			    sc->ale_cdata.ale_tx_ring_map);
1312 		sc->ale_cdata.ale_tx_ring_paddr = 0;
1313 		sc->ale_cdata.ale_tx_ring = NULL;
1314 		bus_dma_tag_destroy(sc->ale_cdata.ale_tx_ring_tag);
1315 		sc->ale_cdata.ale_tx_ring_tag = NULL;
1316 	}
1317 	/* Rx page block. */
1318 	for (i = 0; i < ALE_RX_PAGES; i++) {
1319 		if (sc->ale_cdata.ale_rx_page[i].page_tag != NULL) {
1320 			if (sc->ale_cdata.ale_rx_page[i].page_paddr != 0)
1321 				bus_dmamap_unload(
1322 				    sc->ale_cdata.ale_rx_page[i].page_tag,
1323 				    sc->ale_cdata.ale_rx_page[i].page_map);
1324 			if (sc->ale_cdata.ale_rx_page[i].page_addr != NULL)
1325 				bus_dmamem_free(
1326 				    sc->ale_cdata.ale_rx_page[i].page_tag,
1327 				    sc->ale_cdata.ale_rx_page[i].page_addr,
1328 				    sc->ale_cdata.ale_rx_page[i].page_map);
1329 			sc->ale_cdata.ale_rx_page[i].page_paddr = 0;
1330 			sc->ale_cdata.ale_rx_page[i].page_addr = NULL;
1331 			bus_dma_tag_destroy(
1332 			    sc->ale_cdata.ale_rx_page[i].page_tag);
1333 			sc->ale_cdata.ale_rx_page[i].page_tag = NULL;
1334 		}
1335 	}
1336 	/* Rx CMB. */
1337 	for (i = 0; i < ALE_RX_PAGES; i++) {
1338 		if (sc->ale_cdata.ale_rx_page[i].cmb_tag != NULL) {
1339 			if (sc->ale_cdata.ale_rx_page[i].cmb_paddr != 0)
1340 				bus_dmamap_unload(
1341 				    sc->ale_cdata.ale_rx_page[i].cmb_tag,
1342 				    sc->ale_cdata.ale_rx_page[i].cmb_map);
1343 			if (sc->ale_cdata.ale_rx_page[i].cmb_addr != NULL)
1344 				bus_dmamem_free(
1345 				    sc->ale_cdata.ale_rx_page[i].cmb_tag,
1346 				    sc->ale_cdata.ale_rx_page[i].cmb_addr,
1347 				    sc->ale_cdata.ale_rx_page[i].cmb_map);
1348 			sc->ale_cdata.ale_rx_page[i].cmb_paddr = 0;
1349 			sc->ale_cdata.ale_rx_page[i].cmb_addr = NULL;
1350 			bus_dma_tag_destroy(
1351 			    sc->ale_cdata.ale_rx_page[i].cmb_tag);
1352 			sc->ale_cdata.ale_rx_page[i].cmb_tag = NULL;
1353 		}
1354 	}
1355 	/* Tx CMB. */
1356 	if (sc->ale_cdata.ale_tx_cmb_tag != NULL) {
1357 		if (sc->ale_cdata.ale_tx_cmb_paddr != 0)
1358 			bus_dmamap_unload(sc->ale_cdata.ale_tx_cmb_tag,
1359 			    sc->ale_cdata.ale_tx_cmb_map);
1360 		if (sc->ale_cdata.ale_tx_cmb != NULL)
1361 			bus_dmamem_free(sc->ale_cdata.ale_tx_cmb_tag,
1362 			    sc->ale_cdata.ale_tx_cmb,
1363 			    sc->ale_cdata.ale_tx_cmb_map);
1364 		sc->ale_cdata.ale_tx_cmb_paddr = 0;
1365 		sc->ale_cdata.ale_tx_cmb = NULL;
1366 		bus_dma_tag_destroy(sc->ale_cdata.ale_tx_cmb_tag);
1367 		sc->ale_cdata.ale_tx_cmb_tag = NULL;
1368 	}
1369 	if (sc->ale_cdata.ale_buffer_tag != NULL) {
1370 		bus_dma_tag_destroy(sc->ale_cdata.ale_buffer_tag);
1371 		sc->ale_cdata.ale_buffer_tag = NULL;
1372 	}
1373 	if (sc->ale_cdata.ale_parent_tag != NULL) {
1374 		bus_dma_tag_destroy(sc->ale_cdata.ale_parent_tag);
1375 		sc->ale_cdata.ale_parent_tag = NULL;
1376 	}
1377 }
1378 
1379 static int
ale_shutdown(device_t dev)1380 ale_shutdown(device_t dev)
1381 {
1382 
1383 	return (ale_suspend(dev));
1384 }
1385 
1386 /*
1387  * Note, this driver resets the link speed to 10/100Mbps by
1388  * restarting auto-negotiation in suspend/shutdown phase but we
1389  * don't know whether that auto-negotiation would succeed or not
1390  * as driver has no control after powering off/suspend operation.
1391  * If the renegotiation fail WOL may not work. Running at 1Gbps
1392  * will draw more power than 375mA at 3.3V which is specified in
1393  * PCI specification and that would result in complete
1394  * shutdowning power to ethernet controller.
1395  *
1396  * TODO
1397  * Save current negotiated media speed/duplex/flow-control to
1398  * softc and restore the same link again after resuming. PHY
1399  * handling such as power down/resetting to 100Mbps may be better
1400  * handled in suspend method in phy driver.
1401  */
1402 static void
ale_setlinkspeed(struct ale_softc * sc)1403 ale_setlinkspeed(struct ale_softc *sc)
1404 {
1405 	struct mii_data *mii;
1406 	int aneg, i;
1407 
1408 	mii = device_get_softc(sc->ale_miibus);
1409 	mii_pollstat(mii);
1410 	aneg = 0;
1411 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
1412 	    (IFM_ACTIVE | IFM_AVALID)) {
1413 		switch IFM_SUBTYPE(mii->mii_media_active) {
1414 		case IFM_10_T:
1415 		case IFM_100_TX:
1416 			return;
1417 		case IFM_1000_T:
1418 			aneg++;
1419 			break;
1420 		default:
1421 			break;
1422 		}
1423 	}
1424 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr, MII_100T2CR, 0);
1425 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
1426 	    MII_ANAR, ANAR_TX_FD | ANAR_TX | ANAR_10_FD | ANAR_10 | ANAR_CSMA);
1427 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
1428 	    MII_BMCR, BMCR_RESET | BMCR_AUTOEN | BMCR_STARTNEG);
1429 	DELAY(1000);
1430 	if (aneg != 0) {
1431 		/*
1432 		 * Poll link state until ale(4) get a 10/100Mbps link.
1433 		 */
1434 		for (i = 0; i < MII_ANEGTICKS_GIGE; i++) {
1435 			mii_pollstat(mii);
1436 			if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID))
1437 			    == (IFM_ACTIVE | IFM_AVALID)) {
1438 				switch (IFM_SUBTYPE(
1439 				    mii->mii_media_active)) {
1440 				case IFM_10_T:
1441 				case IFM_100_TX:
1442 					ale_mac_config(sc);
1443 					return;
1444 				default:
1445 					break;
1446 				}
1447 			}
1448 			ALE_UNLOCK(sc);
1449 			pause("alelnk", hz);
1450 			ALE_LOCK(sc);
1451 		}
1452 		if (i == MII_ANEGTICKS_GIGE)
1453 			device_printf(sc->ale_dev,
1454 			    "establishing a link failed, WOL may not work!");
1455 	}
1456 	/*
1457 	 * No link, force MAC to have 100Mbps, full-duplex link.
1458 	 * This is the last resort and may/may not work.
1459 	 */
1460 	mii->mii_media_status = IFM_AVALID | IFM_ACTIVE;
1461 	mii->mii_media_active = IFM_ETHER | IFM_100_TX | IFM_FDX;
1462 	ale_mac_config(sc);
1463 }
1464 
1465 static void
ale_setwol(struct ale_softc * sc)1466 ale_setwol(struct ale_softc *sc)
1467 {
1468 	if_t ifp;
1469 	uint32_t reg, pmcs;
1470 	uint16_t pmstat;
1471 	int pmc;
1472 
1473 	ALE_LOCK_ASSERT(sc);
1474 
1475 	if (pci_find_cap(sc->ale_dev, PCIY_PMG, &pmc) != 0) {
1476 		/* Disable WOL. */
1477 		CSR_WRITE_4(sc, ALE_WOL_CFG, 0);
1478 		reg = CSR_READ_4(sc, ALE_PCIE_PHYMISC);
1479 		reg |= PCIE_PHYMISC_FORCE_RCV_DET;
1480 		CSR_WRITE_4(sc, ALE_PCIE_PHYMISC, reg);
1481 		/* Force PHY power down. */
1482 		CSR_WRITE_2(sc, ALE_GPHY_CTRL,
1483 		    GPHY_CTRL_EXT_RESET | GPHY_CTRL_HIB_EN |
1484 		    GPHY_CTRL_HIB_PULSE | GPHY_CTRL_PHY_PLL_ON |
1485 		    GPHY_CTRL_SEL_ANA_RESET | GPHY_CTRL_PHY_IDDQ |
1486 		    GPHY_CTRL_PCLK_SEL_DIS | GPHY_CTRL_PWDOWN_HW);
1487 		return;
1488 	}
1489 
1490 	ifp = sc->ale_ifp;
1491 	if ((if_getcapenable(ifp) & IFCAP_WOL) != 0) {
1492 		if ((sc->ale_flags & ALE_FLAG_FASTETHER) == 0)
1493 			ale_setlinkspeed(sc);
1494 	}
1495 
1496 	pmcs = 0;
1497 	if ((if_getcapenable(ifp) & IFCAP_WOL_MAGIC) != 0)
1498 		pmcs |= WOL_CFG_MAGIC | WOL_CFG_MAGIC_ENB;
1499 	CSR_WRITE_4(sc, ALE_WOL_CFG, pmcs);
1500 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
1501 	reg &= ~(MAC_CFG_DBG | MAC_CFG_PROMISC | MAC_CFG_ALLMULTI |
1502 	    MAC_CFG_BCAST);
1503 	if ((if_getcapenable(ifp) & IFCAP_WOL_MCAST) != 0)
1504 		reg |= MAC_CFG_ALLMULTI | MAC_CFG_BCAST;
1505 	if ((if_getcapenable(ifp) & IFCAP_WOL) != 0)
1506 		reg |= MAC_CFG_RX_ENB;
1507 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
1508 
1509 	if ((if_getcapenable(ifp) & IFCAP_WOL) == 0) {
1510 		/* WOL disabled, PHY power down. */
1511 		reg = CSR_READ_4(sc, ALE_PCIE_PHYMISC);
1512 		reg |= PCIE_PHYMISC_FORCE_RCV_DET;
1513 		CSR_WRITE_4(sc, ALE_PCIE_PHYMISC, reg);
1514 		CSR_WRITE_2(sc, ALE_GPHY_CTRL,
1515 		    GPHY_CTRL_EXT_RESET | GPHY_CTRL_HIB_EN |
1516 		    GPHY_CTRL_HIB_PULSE | GPHY_CTRL_SEL_ANA_RESET |
1517 		    GPHY_CTRL_PHY_IDDQ | GPHY_CTRL_PCLK_SEL_DIS |
1518 		    GPHY_CTRL_PWDOWN_HW);
1519 	}
1520 	/* Request PME. */
1521 	pmstat = pci_read_config(sc->ale_dev, pmc + PCIR_POWER_STATUS, 2);
1522 	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
1523 	if ((if_getcapenable(ifp) & IFCAP_WOL) != 0)
1524 		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
1525 	pci_write_config(sc->ale_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
1526 }
1527 
1528 static int
ale_suspend(device_t dev)1529 ale_suspend(device_t dev)
1530 {
1531 	struct ale_softc *sc;
1532 
1533 	sc = device_get_softc(dev);
1534 
1535 	ALE_LOCK(sc);
1536 	ale_stop(sc);
1537 	ale_setwol(sc);
1538 	ALE_UNLOCK(sc);
1539 
1540 	return (0);
1541 }
1542 
1543 static int
ale_resume(device_t dev)1544 ale_resume(device_t dev)
1545 {
1546 	struct ale_softc *sc;
1547 	if_t ifp;
1548 	int pmc;
1549 	uint16_t pmstat;
1550 
1551 	sc = device_get_softc(dev);
1552 
1553 	ALE_LOCK(sc);
1554 	if (pci_find_cap(sc->ale_dev, PCIY_PMG, &pmc) == 0) {
1555 		/* Disable PME and clear PME status. */
1556 		pmstat = pci_read_config(sc->ale_dev,
1557 		    pmc + PCIR_POWER_STATUS, 2);
1558 		if ((pmstat & PCIM_PSTAT_PMEENABLE) != 0) {
1559 			pmstat &= ~PCIM_PSTAT_PMEENABLE;
1560 			pci_write_config(sc->ale_dev,
1561 			    pmc + PCIR_POWER_STATUS, pmstat, 2);
1562 		}
1563 	}
1564 	/* Reset PHY. */
1565 	ale_phy_reset(sc);
1566 	ifp = sc->ale_ifp;
1567 	if ((if_getflags(ifp) & IFF_UP) != 0) {
1568 		if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
1569 		ale_init_locked(sc);
1570 	}
1571 	ALE_UNLOCK(sc);
1572 
1573 	return (0);
1574 }
1575 
1576 static int
ale_encap(struct ale_softc * sc,struct mbuf ** m_head)1577 ale_encap(struct ale_softc *sc, struct mbuf **m_head)
1578 {
1579 	struct ale_txdesc *txd, *txd_last;
1580 	struct tx_desc *desc;
1581 	struct mbuf *m;
1582 	struct ip *ip;
1583 	struct tcphdr *tcp;
1584 	bus_dma_segment_t txsegs[ALE_MAXTXSEGS];
1585 	bus_dmamap_t map;
1586 	uint32_t cflags, hdrlen, ip_off, poff, vtag;
1587 	int error, i, nsegs, prod, si;
1588 
1589 	ALE_LOCK_ASSERT(sc);
1590 
1591 	M_ASSERTPKTHDR((*m_head));
1592 
1593 	m = *m_head;
1594 	ip = NULL;
1595 	tcp = NULL;
1596 	cflags = vtag = 0;
1597 	ip_off = poff = 0;
1598 	if ((m->m_pkthdr.csum_flags & (ALE_CSUM_FEATURES | CSUM_TSO)) != 0) {
1599 		/*
1600 		 * AR81xx requires offset of TCP/UDP payload in its Tx
1601 		 * descriptor to perform hardware Tx checksum offload.
1602 		 * Additionally, TSO requires IP/TCP header size and
1603 		 * modification of IP/TCP header in order to make TSO
1604 		 * engine work. This kind of operation takes many CPU
1605 		 * cycles on FreeBSD so fast host CPU is required to
1606 		 * get smooth TSO performance.
1607 		 */
1608 		struct ether_header *eh;
1609 
1610 		if (M_WRITABLE(m) == 0) {
1611 			/* Get a writable copy. */
1612 			m = m_dup(*m_head, M_NOWAIT);
1613 			/* Release original mbufs. */
1614 			m_freem(*m_head);
1615 			if (m == NULL) {
1616 				*m_head = NULL;
1617 				return (ENOBUFS);
1618 			}
1619 			*m_head = m;
1620 		}
1621 
1622 		/*
1623 		 * Buggy-controller requires 4 byte aligned Tx buffer
1624 		 * to make custom checksum offload work.
1625 		 */
1626 		if ((sc->ale_flags & ALE_FLAG_TXCSUM_BUG) != 0 &&
1627 		    (m->m_pkthdr.csum_flags & ALE_CSUM_FEATURES) != 0 &&
1628 		    (mtod(m, intptr_t) & 3) != 0) {
1629 			m = m_defrag(*m_head, M_NOWAIT);
1630 			if (m == NULL) {
1631 				m_freem(*m_head);
1632 				*m_head = NULL;
1633 				return (ENOBUFS);
1634 			}
1635 			*m_head = m;
1636 		}
1637 
1638 		ip_off = sizeof(struct ether_header);
1639 		m = m_pullup(m, ip_off);
1640 		if (m == NULL) {
1641 			*m_head = NULL;
1642 			return (ENOBUFS);
1643 		}
1644 		eh = mtod(m, struct ether_header *);
1645 		/*
1646 		 * Check if hardware VLAN insertion is off.
1647 		 * Additional check for LLC/SNAP frame?
1648 		 */
1649 		if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
1650 			ip_off = sizeof(struct ether_vlan_header);
1651 			m = m_pullup(m, ip_off);
1652 			if (m == NULL) {
1653 				*m_head = NULL;
1654 				return (ENOBUFS);
1655 			}
1656 		}
1657 		m = m_pullup(m, ip_off + sizeof(struct ip));
1658 		if (m == NULL) {
1659 			*m_head = NULL;
1660 			return (ENOBUFS);
1661 		}
1662 		ip = (struct ip *)(mtod(m, char *) + ip_off);
1663 		poff = ip_off + (ip->ip_hl << 2);
1664 		if ((m->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
1665 			/*
1666 			 * XXX
1667 			 * AR81xx requires the first descriptor should
1668 			 * not include any TCP playload for TSO case.
1669 			 * (i.e. ethernet header + IP + TCP header only)
1670 			 * m_pullup(9) above will ensure this too.
1671 			 * However it's not correct if the first mbuf
1672 			 * of the chain does not use cluster.
1673 			 */
1674 			m = m_pullup(m, poff + sizeof(struct tcphdr));
1675 			if (m == NULL) {
1676 				*m_head = NULL;
1677 				return (ENOBUFS);
1678 			}
1679 			ip = (struct ip *)(mtod(m, char *) + ip_off);
1680 			tcp = (struct tcphdr *)(mtod(m, char *) + poff);
1681 			m = m_pullup(m, poff + (tcp->th_off << 2));
1682 			if (m == NULL) {
1683 				*m_head = NULL;
1684 				return (ENOBUFS);
1685 			}
1686 			/*
1687 			 * AR81xx requires IP/TCP header size and offset as
1688 			 * well as TCP pseudo checksum which complicates
1689 			 * TSO configuration. I guess this comes from the
1690 			 * adherence to Microsoft NDIS Large Send
1691 			 * specification which requires insertion of
1692 			 * pseudo checksum by upper stack. The pseudo
1693 			 * checksum that NDIS refers to doesn't include
1694 			 * TCP payload length so ale(4) should recompute
1695 			 * the pseudo checksum here. Hopefully this wouldn't
1696 			 * be much burden on modern CPUs.
1697 			 * Reset IP checksum and recompute TCP pseudo
1698 			 * checksum as NDIS specification said.
1699 			 */
1700 			ip->ip_sum = 0;
1701 			tcp->th_sum = in_pseudo(ip->ip_src.s_addr,
1702 			    ip->ip_dst.s_addr, htons(IPPROTO_TCP));
1703 		}
1704 		*m_head = m;
1705 	}
1706 
1707 	si = prod = sc->ale_cdata.ale_tx_prod;
1708 	txd = &sc->ale_cdata.ale_txdesc[prod];
1709 	txd_last = txd;
1710 	map = txd->tx_dmamap;
1711 
1712 	error =  bus_dmamap_load_mbuf_sg(sc->ale_cdata.ale_tx_tag, map,
1713 	    *m_head, txsegs, &nsegs, 0);
1714 	if (error == EFBIG) {
1715 		m = m_collapse(*m_head, M_NOWAIT, ALE_MAXTXSEGS);
1716 		if (m == NULL) {
1717 			m_freem(*m_head);
1718 			*m_head = NULL;
1719 			return (ENOMEM);
1720 		}
1721 		*m_head = m;
1722 		error = bus_dmamap_load_mbuf_sg(sc->ale_cdata.ale_tx_tag, map,
1723 		    *m_head, txsegs, &nsegs, 0);
1724 		if (error != 0) {
1725 			m_freem(*m_head);
1726 			*m_head = NULL;
1727 			return (error);
1728 		}
1729 	} else if (error != 0)
1730 		return (error);
1731 	if (nsegs == 0) {
1732 		m_freem(*m_head);
1733 		*m_head = NULL;
1734 		return (EIO);
1735 	}
1736 
1737 	/* Check descriptor overrun. */
1738 	if (sc->ale_cdata.ale_tx_cnt + nsegs >= ALE_TX_RING_CNT - 3) {
1739 		bus_dmamap_unload(sc->ale_cdata.ale_tx_tag, map);
1740 		return (ENOBUFS);
1741 	}
1742 	bus_dmamap_sync(sc->ale_cdata.ale_tx_tag, map, BUS_DMASYNC_PREWRITE);
1743 
1744 	m = *m_head;
1745 	if ((m->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
1746 		/* Request TSO and set MSS. */
1747 		cflags |= ALE_TD_TSO;
1748 		cflags |= ((uint32_t)m->m_pkthdr.tso_segsz << ALE_TD_MSS_SHIFT);
1749 		/* Set IP/TCP header size. */
1750 		cflags |= ip->ip_hl << ALE_TD_IPHDR_LEN_SHIFT;
1751 		cflags |= tcp->th_off << ALE_TD_TCPHDR_LEN_SHIFT;
1752 	} else if ((m->m_pkthdr.csum_flags & ALE_CSUM_FEATURES) != 0) {
1753 		/*
1754 		 * AR81xx supports Tx custom checksum offload feature
1755 		 * that offloads single 16bit checksum computation.
1756 		 * So you can choose one among IP, TCP and UDP.
1757 		 * Normally driver sets checksum start/insertion
1758 		 * position from the information of TCP/UDP frame as
1759 		 * TCP/UDP checksum takes more time than that of IP.
1760 		 * However it seems that custom checksum offload
1761 		 * requires 4 bytes aligned Tx buffers due to hardware
1762 		 * bug.
1763 		 * AR81xx also supports explicit Tx checksum computation
1764 		 * if it is told that the size of IP header and TCP
1765 		 * header(for UDP, the header size does not matter
1766 		 * because it's fixed length). However with this scheme
1767 		 * TSO does not work so you have to choose one either
1768 		 * TSO or explicit Tx checksum offload. I chosen TSO
1769 		 * plus custom checksum offload with work-around which
1770 		 * will cover most common usage for this consumer
1771 		 * ethernet controller. The work-around takes a lot of
1772 		 * CPU cycles if Tx buffer is not aligned on 4 bytes
1773 		 * boundary, though.
1774 		 */
1775 		cflags |= ALE_TD_CXSUM;
1776 		/* Set checksum start offset. */
1777 		cflags |= (poff << ALE_TD_CSUM_PLOADOFFSET_SHIFT);
1778 		/* Set checksum insertion position of TCP/UDP. */
1779 		cflags |= ((poff + m->m_pkthdr.csum_data) <<
1780 		    ALE_TD_CSUM_XSUMOFFSET_SHIFT);
1781 	}
1782 
1783 	/* Configure VLAN hardware tag insertion. */
1784 	if ((m->m_flags & M_VLANTAG) != 0) {
1785 		vtag = ALE_TX_VLAN_TAG(m->m_pkthdr.ether_vtag);
1786 		vtag = ((vtag << ALE_TD_VLAN_SHIFT) & ALE_TD_VLAN_MASK);
1787 		cflags |= ALE_TD_INSERT_VLAN_TAG;
1788 	}
1789 
1790 	i = 0;
1791 	if ((m->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
1792 		/*
1793 		 * Make sure the first fragment contains
1794 		 * only ethernet and IP/TCP header with options.
1795 		 */
1796 		hdrlen =  poff + (tcp->th_off << 2);
1797 		desc = &sc->ale_cdata.ale_tx_ring[prod];
1798 		desc->addr = htole64(txsegs[i].ds_addr);
1799 		desc->len = htole32(ALE_TX_BYTES(hdrlen) | vtag);
1800 		desc->flags = htole32(cflags);
1801 		sc->ale_cdata.ale_tx_cnt++;
1802 		ALE_DESC_INC(prod, ALE_TX_RING_CNT);
1803 		if (m->m_len - hdrlen > 0) {
1804 			/* Handle remaining payload of the first fragment. */
1805 			desc = &sc->ale_cdata.ale_tx_ring[prod];
1806 			desc->addr = htole64(txsegs[i].ds_addr + hdrlen);
1807 			desc->len = htole32(ALE_TX_BYTES(m->m_len - hdrlen) |
1808 			    vtag);
1809 			desc->flags = htole32(cflags);
1810 			sc->ale_cdata.ale_tx_cnt++;
1811 			ALE_DESC_INC(prod, ALE_TX_RING_CNT);
1812 		}
1813 		i = 1;
1814 	}
1815 	for (; i < nsegs; i++) {
1816 		desc = &sc->ale_cdata.ale_tx_ring[prod];
1817 		desc->addr = htole64(txsegs[i].ds_addr);
1818 		desc->len = htole32(ALE_TX_BYTES(txsegs[i].ds_len) | vtag);
1819 		desc->flags = htole32(cflags);
1820 		sc->ale_cdata.ale_tx_cnt++;
1821 		ALE_DESC_INC(prod, ALE_TX_RING_CNT);
1822 	}
1823 	/* Update producer index. */
1824 	sc->ale_cdata.ale_tx_prod = prod;
1825 	/* Set TSO header on the first descriptor. */
1826 	if ((m->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
1827 		desc = &sc->ale_cdata.ale_tx_ring[si];
1828 		desc->flags |= htole32(ALE_TD_TSO_HDR);
1829 	}
1830 
1831 	/* Finally set EOP on the last descriptor. */
1832 	prod = (prod + ALE_TX_RING_CNT - 1) % ALE_TX_RING_CNT;
1833 	desc = &sc->ale_cdata.ale_tx_ring[prod];
1834 	desc->flags |= htole32(ALE_TD_EOP);
1835 
1836 	/* Swap dmamap of the first and the last. */
1837 	txd = &sc->ale_cdata.ale_txdesc[prod];
1838 	map = txd_last->tx_dmamap;
1839 	txd_last->tx_dmamap = txd->tx_dmamap;
1840 	txd->tx_dmamap = map;
1841 	txd->tx_m = m;
1842 
1843 	/* Sync descriptors. */
1844 	bus_dmamap_sync(sc->ale_cdata.ale_tx_ring_tag,
1845 	    sc->ale_cdata.ale_tx_ring_map,
1846 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1847 
1848 	return (0);
1849 }
1850 
1851 static void
ale_start(if_t ifp)1852 ale_start(if_t ifp)
1853 {
1854         struct ale_softc *sc;
1855 
1856 	sc = if_getsoftc(ifp);
1857 	ALE_LOCK(sc);
1858 	ale_start_locked(ifp);
1859 	ALE_UNLOCK(sc);
1860 }
1861 
1862 static void
ale_start_locked(if_t ifp)1863 ale_start_locked(if_t ifp)
1864 {
1865         struct ale_softc *sc;
1866         struct mbuf *m_head;
1867 	int enq;
1868 
1869 	sc = if_getsoftc(ifp);
1870 
1871 	ALE_LOCK_ASSERT(sc);
1872 
1873 	/* Reclaim transmitted frames. */
1874 	if (sc->ale_cdata.ale_tx_cnt >= ALE_TX_DESC_HIWAT)
1875 		ale_txeof(sc);
1876 
1877 	if ((if_getdrvflags(ifp) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1878 	    IFF_DRV_RUNNING || (sc->ale_flags & ALE_FLAG_LINK) == 0)
1879 		return;
1880 
1881 	for (enq = 0; !if_sendq_empty(ifp); ) {
1882 		m_head = if_dequeue(ifp);
1883 		if (m_head == NULL)
1884 			break;
1885 		/*
1886 		 * Pack the data into the transmit ring. If we
1887 		 * don't have room, set the OACTIVE flag and wait
1888 		 * for the NIC to drain the ring.
1889 		 */
1890 		if (ale_encap(sc, &m_head)) {
1891 			if (m_head == NULL)
1892 				break;
1893 			if_sendq_prepend(ifp, m_head);
1894 			if_setdrvflagbits(ifp, IFF_DRV_OACTIVE, 0);
1895 			break;
1896 		}
1897 
1898 		enq++;
1899 		/*
1900 		 * If there's a BPF listener, bounce a copy of this frame
1901 		 * to him.
1902 		 */
1903 		ETHER_BPF_MTAP(ifp, m_head);
1904 	}
1905 
1906 	if (enq > 0) {
1907 		/* Kick. */
1908 		CSR_WRITE_4(sc, ALE_MBOX_TPD_PROD_IDX,
1909 		    sc->ale_cdata.ale_tx_prod);
1910 		/* Set a timeout in case the chip goes out to lunch. */
1911 		sc->ale_watchdog_timer = ALE_TX_TIMEOUT;
1912 	}
1913 }
1914 
1915 static void
ale_watchdog(struct ale_softc * sc)1916 ale_watchdog(struct ale_softc *sc)
1917 {
1918 	if_t ifp;
1919 
1920 	ALE_LOCK_ASSERT(sc);
1921 
1922 	if (sc->ale_watchdog_timer == 0 || --sc->ale_watchdog_timer)
1923 		return;
1924 
1925 	ifp = sc->ale_ifp;
1926 	if ((sc->ale_flags & ALE_FLAG_LINK) == 0) {
1927 		if_printf(sc->ale_ifp, "watchdog timeout (lost link)\n");
1928 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1929 		if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
1930 		ale_init_locked(sc);
1931 		return;
1932 	}
1933 	if_printf(sc->ale_ifp, "watchdog timeout -- resetting\n");
1934 	if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1935 	if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
1936 	ale_init_locked(sc);
1937 	if (!if_sendq_empty(ifp))
1938 		ale_start_locked(ifp);
1939 }
1940 
1941 static int
ale_ioctl(if_t ifp,u_long cmd,caddr_t data)1942 ale_ioctl(if_t ifp, u_long cmd, caddr_t data)
1943 {
1944 	struct ale_softc *sc;
1945 	struct ifreq *ifr;
1946 	struct mii_data *mii;
1947 	int error, mask;
1948 
1949 	sc = if_getsoftc(ifp);
1950 	ifr = (struct ifreq *)data;
1951 	error = 0;
1952 	switch (cmd) {
1953 	case SIOCSIFMTU:
1954 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > ALE_JUMBO_MTU ||
1955 		    ((sc->ale_flags & ALE_FLAG_JUMBO) == 0 &&
1956 		    ifr->ifr_mtu > ETHERMTU))
1957 			error = EINVAL;
1958 		else if (if_getmtu(ifp) != ifr->ifr_mtu) {
1959 			ALE_LOCK(sc);
1960 			if_setmtu(ifp, ifr->ifr_mtu);
1961 			if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0) {
1962 				if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
1963 				ale_init_locked(sc);
1964 			}
1965 			ALE_UNLOCK(sc);
1966 		}
1967 		break;
1968 	case SIOCSIFFLAGS:
1969 		ALE_LOCK(sc);
1970 		if ((if_getflags(ifp) & IFF_UP) != 0) {
1971 			if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0) {
1972 				if (((if_getflags(ifp) ^ sc->ale_if_flags)
1973 				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
1974 					ale_rxfilter(sc);
1975 			} else {
1976 				ale_init_locked(sc);
1977 			}
1978 		} else {
1979 			if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0)
1980 				ale_stop(sc);
1981 		}
1982 		sc->ale_if_flags = if_getflags(ifp);
1983 		ALE_UNLOCK(sc);
1984 		break;
1985 	case SIOCADDMULTI:
1986 	case SIOCDELMULTI:
1987 		ALE_LOCK(sc);
1988 		if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0)
1989 			ale_rxfilter(sc);
1990 		ALE_UNLOCK(sc);
1991 		break;
1992 	case SIOCSIFMEDIA:
1993 	case SIOCGIFMEDIA:
1994 		mii = device_get_softc(sc->ale_miibus);
1995 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, cmd);
1996 		break;
1997 	case SIOCSIFCAP:
1998 		ALE_LOCK(sc);
1999 		mask = ifr->ifr_reqcap ^ if_getcapenable(ifp);
2000 		if ((mask & IFCAP_TXCSUM) != 0 &&
2001 		    (if_getcapabilities(ifp) & IFCAP_TXCSUM) != 0) {
2002 			if_togglecapenable(ifp, IFCAP_TXCSUM);
2003 			if ((if_getcapenable(ifp) & IFCAP_TXCSUM) != 0)
2004 				if_sethwassistbits(ifp, ALE_CSUM_FEATURES, 0);
2005 			else
2006 				if_sethwassistbits(ifp, 0, ALE_CSUM_FEATURES);
2007 		}
2008 		if ((mask & IFCAP_RXCSUM) != 0 &&
2009 		    (if_getcapabilities(ifp) & IFCAP_RXCSUM) != 0)
2010 			if_togglecapenable(ifp, IFCAP_RXCSUM);
2011 		if ((mask & IFCAP_TSO4) != 0 &&
2012 		    (if_getcapabilities(ifp) & IFCAP_TSO4) != 0) {
2013 			if_togglecapenable(ifp, IFCAP_TSO4);
2014 			if ((if_getcapenable(ifp) & IFCAP_TSO4) != 0)
2015 				if_sethwassistbits(ifp, CSUM_TSO, 0);
2016 			else
2017 				if_sethwassistbits(ifp, 0, CSUM_TSO);
2018 		}
2019 
2020 		if ((mask & IFCAP_WOL_MCAST) != 0 &&
2021 		    (if_getcapabilities(ifp) & IFCAP_WOL_MCAST) != 0)
2022 			if_togglecapenable(ifp, IFCAP_WOL_MCAST);
2023 		if ((mask & IFCAP_WOL_MAGIC) != 0 &&
2024 		    (if_getcapabilities(ifp) & IFCAP_WOL_MAGIC) != 0)
2025 			if_togglecapenable(ifp, IFCAP_WOL_MAGIC);
2026 		if ((mask & IFCAP_VLAN_HWCSUM) != 0 &&
2027 		    (if_getcapabilities(ifp) & IFCAP_VLAN_HWCSUM) != 0)
2028 			if_togglecapenable(ifp, IFCAP_VLAN_HWCSUM);
2029 		if ((mask & IFCAP_VLAN_HWTSO) != 0 &&
2030 		    (if_getcapabilities(ifp) & IFCAP_VLAN_HWTSO) != 0)
2031 			if_togglecapenable(ifp, IFCAP_VLAN_HWTSO);
2032 		if ((mask & IFCAP_VLAN_HWTAGGING) != 0 &&
2033 		    (if_getcapabilities(ifp) & IFCAP_VLAN_HWTAGGING) != 0) {
2034 			if_togglecapenable(ifp, IFCAP_VLAN_HWTAGGING);
2035 			if ((if_getcapenable(ifp) & IFCAP_VLAN_HWTAGGING) == 0)
2036 				if_setcapenablebit(ifp, 0, IFCAP_VLAN_HWTSO);
2037 			ale_rxvlan(sc);
2038 		}
2039 		ALE_UNLOCK(sc);
2040 		VLAN_CAPABILITIES(ifp);
2041 		break;
2042 	default:
2043 		error = ether_ioctl(ifp, cmd, data);
2044 		break;
2045 	}
2046 
2047 	return (error);
2048 }
2049 
2050 static void
ale_mac_config(struct ale_softc * sc)2051 ale_mac_config(struct ale_softc *sc)
2052 {
2053 	struct mii_data *mii;
2054 	uint32_t reg;
2055 
2056 	ALE_LOCK_ASSERT(sc);
2057 
2058 	mii = device_get_softc(sc->ale_miibus);
2059 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
2060 	reg &= ~(MAC_CFG_FULL_DUPLEX | MAC_CFG_TX_FC | MAC_CFG_RX_FC |
2061 	    MAC_CFG_SPEED_MASK);
2062 	/* Reprogram MAC with resolved speed/duplex. */
2063 	switch (IFM_SUBTYPE(mii->mii_media_active)) {
2064 	case IFM_10_T:
2065 	case IFM_100_TX:
2066 		reg |= MAC_CFG_SPEED_10_100;
2067 		break;
2068 	case IFM_1000_T:
2069 		reg |= MAC_CFG_SPEED_1000;
2070 		break;
2071 	}
2072 	if ((IFM_OPTIONS(mii->mii_media_active) & IFM_FDX) != 0) {
2073 		reg |= MAC_CFG_FULL_DUPLEX;
2074 		if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_TXPAUSE) != 0)
2075 			reg |= MAC_CFG_TX_FC;
2076 		if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_RXPAUSE) != 0)
2077 			reg |= MAC_CFG_RX_FC;
2078 	}
2079 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2080 }
2081 
2082 static void
ale_stats_clear(struct ale_softc * sc)2083 ale_stats_clear(struct ale_softc *sc)
2084 {
2085 	struct smb sb;
2086 	uint32_t *reg;
2087 	int i;
2088 
2089 	for (reg = &sb.rx_frames, i = 0; reg <= &sb.rx_pkts_filtered; reg++) {
2090 		CSR_READ_4(sc, ALE_RX_MIB_BASE + i);
2091 		i += sizeof(uint32_t);
2092 	}
2093 	/* Read Tx statistics. */
2094 	for (reg = &sb.tx_frames, i = 0; reg <= &sb.tx_mcast_bytes; reg++) {
2095 		CSR_READ_4(sc, ALE_TX_MIB_BASE + i);
2096 		i += sizeof(uint32_t);
2097 	}
2098 }
2099 
2100 static void
ale_stats_update(struct ale_softc * sc)2101 ale_stats_update(struct ale_softc *sc)
2102 {
2103 	struct ale_hw_stats *stat;
2104 	struct smb sb, *smb;
2105 	if_t ifp;
2106 	uint32_t *reg;
2107 	int i;
2108 
2109 	ALE_LOCK_ASSERT(sc);
2110 
2111 	ifp = sc->ale_ifp;
2112 	stat = &sc->ale_stats;
2113 	smb = &sb;
2114 
2115 	/* Read Rx statistics. */
2116 	for (reg = &sb.rx_frames, i = 0; reg <= &sb.rx_pkts_filtered; reg++) {
2117 		*reg = CSR_READ_4(sc, ALE_RX_MIB_BASE + i);
2118 		i += sizeof(uint32_t);
2119 	}
2120 	/* Read Tx statistics. */
2121 	for (reg = &sb.tx_frames, i = 0; reg <= &sb.tx_mcast_bytes; reg++) {
2122 		*reg = CSR_READ_4(sc, ALE_TX_MIB_BASE + i);
2123 		i += sizeof(uint32_t);
2124 	}
2125 
2126 	/* Rx stats. */
2127 	stat->rx_frames += smb->rx_frames;
2128 	stat->rx_bcast_frames += smb->rx_bcast_frames;
2129 	stat->rx_mcast_frames += smb->rx_mcast_frames;
2130 	stat->rx_pause_frames += smb->rx_pause_frames;
2131 	stat->rx_control_frames += smb->rx_control_frames;
2132 	stat->rx_crcerrs += smb->rx_crcerrs;
2133 	stat->rx_lenerrs += smb->rx_lenerrs;
2134 	stat->rx_bytes += smb->rx_bytes;
2135 	stat->rx_runts += smb->rx_runts;
2136 	stat->rx_fragments += smb->rx_fragments;
2137 	stat->rx_pkts_64 += smb->rx_pkts_64;
2138 	stat->rx_pkts_65_127 += smb->rx_pkts_65_127;
2139 	stat->rx_pkts_128_255 += smb->rx_pkts_128_255;
2140 	stat->rx_pkts_256_511 += smb->rx_pkts_256_511;
2141 	stat->rx_pkts_512_1023 += smb->rx_pkts_512_1023;
2142 	stat->rx_pkts_1024_1518 += smb->rx_pkts_1024_1518;
2143 	stat->rx_pkts_1519_max += smb->rx_pkts_1519_max;
2144 	stat->rx_pkts_truncated += smb->rx_pkts_truncated;
2145 	stat->rx_fifo_oflows += smb->rx_fifo_oflows;
2146 	stat->rx_rrs_errs += smb->rx_rrs_errs;
2147 	stat->rx_alignerrs += smb->rx_alignerrs;
2148 	stat->rx_bcast_bytes += smb->rx_bcast_bytes;
2149 	stat->rx_mcast_bytes += smb->rx_mcast_bytes;
2150 	stat->rx_pkts_filtered += smb->rx_pkts_filtered;
2151 
2152 	/* Tx stats. */
2153 	stat->tx_frames += smb->tx_frames;
2154 	stat->tx_bcast_frames += smb->tx_bcast_frames;
2155 	stat->tx_mcast_frames += smb->tx_mcast_frames;
2156 	stat->tx_pause_frames += smb->tx_pause_frames;
2157 	stat->tx_excess_defer += smb->tx_excess_defer;
2158 	stat->tx_control_frames += smb->tx_control_frames;
2159 	stat->tx_deferred += smb->tx_deferred;
2160 	stat->tx_bytes += smb->tx_bytes;
2161 	stat->tx_pkts_64 += smb->tx_pkts_64;
2162 	stat->tx_pkts_65_127 += smb->tx_pkts_65_127;
2163 	stat->tx_pkts_128_255 += smb->tx_pkts_128_255;
2164 	stat->tx_pkts_256_511 += smb->tx_pkts_256_511;
2165 	stat->tx_pkts_512_1023 += smb->tx_pkts_512_1023;
2166 	stat->tx_pkts_1024_1518 += smb->tx_pkts_1024_1518;
2167 	stat->tx_pkts_1519_max += smb->tx_pkts_1519_max;
2168 	stat->tx_single_colls += smb->tx_single_colls;
2169 	stat->tx_multi_colls += smb->tx_multi_colls;
2170 	stat->tx_late_colls += smb->tx_late_colls;
2171 	stat->tx_excess_colls += smb->tx_excess_colls;
2172 	stat->tx_underrun += smb->tx_underrun;
2173 	stat->tx_desc_underrun += smb->tx_desc_underrun;
2174 	stat->tx_lenerrs += smb->tx_lenerrs;
2175 	stat->tx_pkts_truncated += smb->tx_pkts_truncated;
2176 	stat->tx_bcast_bytes += smb->tx_bcast_bytes;
2177 	stat->tx_mcast_bytes += smb->tx_mcast_bytes;
2178 
2179 	/* Update counters in ifnet. */
2180 	if_inc_counter(ifp, IFCOUNTER_OPACKETS, smb->tx_frames);
2181 
2182 	if_inc_counter(ifp, IFCOUNTER_COLLISIONS, smb->tx_single_colls +
2183 	    smb->tx_multi_colls * 2 + smb->tx_late_colls +
2184 	    smb->tx_excess_colls * HDPX_CFG_RETRY_DEFAULT);
2185 
2186 	if_inc_counter(ifp, IFCOUNTER_OERRORS, smb->tx_late_colls +
2187 	    smb->tx_excess_colls + smb->tx_underrun + smb->tx_pkts_truncated);
2188 
2189 	if_inc_counter(ifp, IFCOUNTER_IPACKETS, smb->rx_frames);
2190 
2191 	if_inc_counter(ifp, IFCOUNTER_IERRORS,
2192 	    smb->rx_crcerrs + smb->rx_lenerrs +
2193 	    smb->rx_runts + smb->rx_pkts_truncated +
2194 	    smb->rx_fifo_oflows + smb->rx_rrs_errs +
2195 	    smb->rx_alignerrs);
2196 }
2197 
2198 static int
ale_intr(void * arg)2199 ale_intr(void *arg)
2200 {
2201 	struct ale_softc *sc;
2202 	uint32_t status;
2203 
2204 	sc = (struct ale_softc *)arg;
2205 
2206 	status = CSR_READ_4(sc, ALE_INTR_STATUS);
2207 	if ((status & ALE_INTRS) == 0)
2208 		return (FILTER_STRAY);
2209 	/* Disable interrupts. */
2210 	CSR_WRITE_4(sc, ALE_INTR_STATUS, INTR_DIS_INT);
2211 	taskqueue_enqueue(sc->ale_tq, &sc->ale_int_task);
2212 
2213 	return (FILTER_HANDLED);
2214 }
2215 
2216 static void
ale_int_task(void * arg,int pending)2217 ale_int_task(void *arg, int pending)
2218 {
2219 	struct ale_softc *sc;
2220 	if_t ifp;
2221 	uint32_t status;
2222 	int more;
2223 
2224 	sc = (struct ale_softc *)arg;
2225 
2226 	status = CSR_READ_4(sc, ALE_INTR_STATUS);
2227 	ALE_LOCK(sc);
2228 	if (sc->ale_morework != 0)
2229 		status |= INTR_RX_PKT;
2230 	if ((status & ALE_INTRS) == 0)
2231 		goto done;
2232 
2233 	/* Acknowledge interrupts but still disable interrupts. */
2234 	CSR_WRITE_4(sc, ALE_INTR_STATUS, status | INTR_DIS_INT);
2235 
2236 	ifp = sc->ale_ifp;
2237 	more = 0;
2238 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0) {
2239 		more = ale_rxeof(sc, sc->ale_process_limit);
2240 		if (more == EAGAIN)
2241 			sc->ale_morework = 1;
2242 		else if (more == EIO) {
2243 			sc->ale_stats.reset_brk_seq++;
2244 			if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
2245 			ale_init_locked(sc);
2246 			ALE_UNLOCK(sc);
2247 			return;
2248 		}
2249 
2250 		if ((status & (INTR_DMA_RD_TO_RST | INTR_DMA_WR_TO_RST)) != 0) {
2251 			if ((status & INTR_DMA_RD_TO_RST) != 0)
2252 				device_printf(sc->ale_dev,
2253 				    "DMA read error! -- resetting\n");
2254 			if ((status & INTR_DMA_WR_TO_RST) != 0)
2255 				device_printf(sc->ale_dev,
2256 				    "DMA write error! -- resetting\n");
2257 			if_setdrvflagbits(ifp, 0, IFF_DRV_RUNNING);
2258 			ale_init_locked(sc);
2259 			ALE_UNLOCK(sc);
2260 			return;
2261 		}
2262 		if (!if_sendq_empty(ifp))
2263 			ale_start_locked(ifp);
2264 	}
2265 
2266 	if (more == EAGAIN ||
2267 	    (CSR_READ_4(sc, ALE_INTR_STATUS) & ALE_INTRS) != 0) {
2268 		ALE_UNLOCK(sc);
2269 		taskqueue_enqueue(sc->ale_tq, &sc->ale_int_task);
2270 		return;
2271 	}
2272 
2273 done:
2274 	ALE_UNLOCK(sc);
2275 
2276 	/* Re-enable interrupts. */
2277 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0x7FFFFFFF);
2278 }
2279 
2280 static void
ale_txeof(struct ale_softc * sc)2281 ale_txeof(struct ale_softc *sc)
2282 {
2283 	if_t ifp;
2284 	struct ale_txdesc *txd;
2285 	uint32_t cons, prod;
2286 	int prog;
2287 
2288 	ALE_LOCK_ASSERT(sc);
2289 
2290 	ifp = sc->ale_ifp;
2291 
2292 	if (sc->ale_cdata.ale_tx_cnt == 0)
2293 		return;
2294 
2295 	bus_dmamap_sync(sc->ale_cdata.ale_tx_ring_tag,
2296 	    sc->ale_cdata.ale_tx_ring_map,
2297 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2298 	if ((sc->ale_flags & ALE_FLAG_TXCMB_BUG) == 0) {
2299 		bus_dmamap_sync(sc->ale_cdata.ale_tx_cmb_tag,
2300 		    sc->ale_cdata.ale_tx_cmb_map,
2301 		    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2302 		prod = *sc->ale_cdata.ale_tx_cmb & TPD_CNT_MASK;
2303 	} else
2304 		prod = CSR_READ_2(sc, ALE_TPD_CONS_IDX);
2305 	cons = sc->ale_cdata.ale_tx_cons;
2306 	/*
2307 	 * Go through our Tx list and free mbufs for those
2308 	 * frames which have been transmitted.
2309 	 */
2310 	for (prog = 0; cons != prod; prog++,
2311 	    ALE_DESC_INC(cons, ALE_TX_RING_CNT)) {
2312 		if (sc->ale_cdata.ale_tx_cnt <= 0)
2313 			break;
2314 		prog++;
2315 		if_setdrvflagbits(ifp, 0, IFF_DRV_OACTIVE);
2316 		sc->ale_cdata.ale_tx_cnt--;
2317 		txd = &sc->ale_cdata.ale_txdesc[cons];
2318 		if (txd->tx_m != NULL) {
2319 			/* Reclaim transmitted mbufs. */
2320 			bus_dmamap_sync(sc->ale_cdata.ale_tx_tag,
2321 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2322 			bus_dmamap_unload(sc->ale_cdata.ale_tx_tag,
2323 			    txd->tx_dmamap);
2324 			m_freem(txd->tx_m);
2325 			txd->tx_m = NULL;
2326 		}
2327 	}
2328 
2329 	if (prog > 0) {
2330 		sc->ale_cdata.ale_tx_cons = cons;
2331 		/*
2332 		 * Unarm watchdog timer only when there is no pending
2333 		 * Tx descriptors in queue.
2334 		 */
2335 		if (sc->ale_cdata.ale_tx_cnt == 0)
2336 			sc->ale_watchdog_timer = 0;
2337 	}
2338 }
2339 
2340 static void
ale_rx_update_page(struct ale_softc * sc,struct ale_rx_page ** page,uint32_t length,uint32_t * prod)2341 ale_rx_update_page(struct ale_softc *sc, struct ale_rx_page **page,
2342     uint32_t length, uint32_t *prod)
2343 {
2344 	struct ale_rx_page *rx_page;
2345 
2346 	rx_page = *page;
2347 	/* Update consumer position. */
2348 	rx_page->cons += roundup(length + sizeof(struct rx_rs),
2349 	    ALE_RX_PAGE_ALIGN);
2350 	if (rx_page->cons >= ALE_RX_PAGE_SZ) {
2351 		/*
2352 		 * End of Rx page reached, let hardware reuse
2353 		 * this page.
2354 		 */
2355 		rx_page->cons = 0;
2356 		*rx_page->cmb_addr = 0;
2357 		bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2358 		    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2359 		CSR_WRITE_1(sc, ALE_RXF0_PAGE0 + sc->ale_cdata.ale_rx_curp,
2360 		    RXF_VALID);
2361 		/* Switch to alternate Rx page. */
2362 		sc->ale_cdata.ale_rx_curp ^= 1;
2363 		rx_page = *page =
2364 		    &sc->ale_cdata.ale_rx_page[sc->ale_cdata.ale_rx_curp];
2365 		/* Page flipped, sync CMB and Rx page. */
2366 		bus_dmamap_sync(rx_page->page_tag, rx_page->page_map,
2367 		    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2368 		bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2369 		    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2370 		/* Sync completed, cache updated producer index. */
2371 		*prod = *rx_page->cmb_addr;
2372 	}
2373 }
2374 
2375 /*
2376  * It seems that AR81xx controller can compute partial checksum.
2377  * The partial checksum value can be used to accelerate checksum
2378  * computation for fragmented TCP/UDP packets. Upper network stack
2379  * already takes advantage of the partial checksum value in IP
2380  * reassembly stage. But I'm not sure the correctness of the
2381  * partial hardware checksum assistance due to lack of data sheet.
2382  * In addition, the Rx feature of controller that requires copying
2383  * for every frames effectively nullifies one of most nice offload
2384  * capability of controller.
2385  */
2386 static void
ale_rxcsum(struct ale_softc * sc,struct mbuf * m,uint32_t status)2387 ale_rxcsum(struct ale_softc *sc, struct mbuf *m, uint32_t status)
2388 {
2389 	if_t ifp;
2390 	struct ip *ip;
2391 	char *p;
2392 
2393 	ifp = sc->ale_ifp;
2394 	m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
2395 	if ((status & ALE_RD_IPCSUM_NOK) == 0)
2396 		m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
2397 
2398 	if ((sc->ale_flags & ALE_FLAG_RXCSUM_BUG) == 0) {
2399 		if (((status & ALE_RD_IPV4_FRAG) == 0) &&
2400 		    ((status & (ALE_RD_TCP | ALE_RD_UDP)) != 0) &&
2401 		    ((status & ALE_RD_TCP_UDPCSUM_NOK) == 0)) {
2402 			m->m_pkthdr.csum_flags |=
2403 			    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2404 			m->m_pkthdr.csum_data = 0xffff;
2405 		}
2406 	} else {
2407 		if ((status & (ALE_RD_TCP | ALE_RD_UDP)) != 0 &&
2408 		    (status & ALE_RD_TCP_UDPCSUM_NOK) == 0) {
2409 			p = mtod(m, char *);
2410 			p += ETHER_HDR_LEN;
2411 			if ((status & ALE_RD_802_3) != 0)
2412 				p += LLC_SNAPFRAMELEN;
2413 			if ((if_getcapenable(ifp) & IFCAP_VLAN_HWTAGGING) == 0 &&
2414 			    (status & ALE_RD_VLAN) != 0)
2415 				p += ETHER_VLAN_ENCAP_LEN;
2416 			ip = (struct ip *)p;
2417 			if (ip->ip_off != 0 && (status & ALE_RD_IPV4_DF) == 0)
2418 				return;
2419 			m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
2420 			    CSUM_PSEUDO_HDR;
2421 			m->m_pkthdr.csum_data = 0xffff;
2422 		}
2423 	}
2424 	/*
2425 	 * Don't mark bad checksum for TCP/UDP frames
2426 	 * as fragmented frames may always have set
2427 	 * bad checksummed bit of frame status.
2428 	 */
2429 }
2430 
2431 /* Process received frames. */
2432 static int
ale_rxeof(struct ale_softc * sc,int count)2433 ale_rxeof(struct ale_softc *sc, int count)
2434 {
2435 	struct ale_rx_page *rx_page;
2436 	struct rx_rs *rs;
2437 	if_t ifp;
2438 	struct mbuf *m;
2439 	uint32_t length, prod, seqno, status, vtags;
2440 	int prog;
2441 
2442 	ifp = sc->ale_ifp;
2443 	rx_page = &sc->ale_cdata.ale_rx_page[sc->ale_cdata.ale_rx_curp];
2444 	bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2445 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2446 	bus_dmamap_sync(rx_page->page_tag, rx_page->page_map,
2447 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2448 	/*
2449 	 * Don't directly access producer index as hardware may
2450 	 * update it while Rx handler is in progress. It would
2451 	 * be even better if there is a way to let hardware
2452 	 * know how far driver processed its received frames.
2453 	 * Alternatively, hardware could provide a way to disable
2454 	 * CMB updates until driver acknowledges the end of CMB
2455 	 * access.
2456 	 */
2457 	prod = *rx_page->cmb_addr;
2458 	for (prog = 0; prog < count; prog++) {
2459 		if (rx_page->cons >= prod)
2460 			break;
2461 		rs = (struct rx_rs *)(rx_page->page_addr + rx_page->cons);
2462 		seqno = ALE_RX_SEQNO(le32toh(rs->seqno));
2463 		if (sc->ale_cdata.ale_rx_seqno != seqno) {
2464 			/*
2465 			 * Normally I believe this should not happen unless
2466 			 * severe driver bug or corrupted memory. However
2467 			 * it seems to happen under certain conditions which
2468 			 * is triggered by abrupt Rx events such as initiation
2469 			 * of bulk transfer of remote host. It's not easy to
2470 			 * reproduce this and I doubt it could be related
2471 			 * with FIFO overflow of hardware or activity of Tx
2472 			 * CMB updates. I also remember similar behaviour
2473 			 * seen on RealTek 8139 which uses resembling Rx
2474 			 * scheme.
2475 			 */
2476 			if (bootverbose)
2477 				device_printf(sc->ale_dev,
2478 				    "garbled seq: %u, expected: %u -- "
2479 				    "resetting!\n", seqno,
2480 				    sc->ale_cdata.ale_rx_seqno);
2481 			return (EIO);
2482 		}
2483 		/* Frame received. */
2484 		sc->ale_cdata.ale_rx_seqno++;
2485 		length = ALE_RX_BYTES(le32toh(rs->length));
2486 		status = le32toh(rs->flags);
2487 		if ((status & ALE_RD_ERROR) != 0) {
2488 			/*
2489 			 * We want to pass the following frames to upper
2490 			 * layer regardless of error status of Rx return
2491 			 * status.
2492 			 *
2493 			 *  o IP/TCP/UDP checksum is bad.
2494 			 *  o frame length and protocol specific length
2495 			 *     does not match.
2496 			 */
2497 			if ((status & (ALE_RD_CRC | ALE_RD_CODE |
2498 			    ALE_RD_DRIBBLE | ALE_RD_RUNT | ALE_RD_OFLOW |
2499 			    ALE_RD_TRUNC)) != 0) {
2500 				ale_rx_update_page(sc, &rx_page, length, &prod);
2501 				continue;
2502 			}
2503 		}
2504 		/*
2505 		 * m_devget(9) is major bottle-neck of ale(4)(It comes
2506 		 * from hardware limitation). For jumbo frames we could
2507 		 * get a slightly better performance if driver use
2508 		 * m_getjcl(9) with proper buffer size argument. However
2509 		 * that would make code more complicated and I don't
2510 		 * think users would expect good Rx performance numbers
2511 		 * on these low-end consumer ethernet controller.
2512 		 */
2513 		m = m_devget((char *)(rs + 1), length - ETHER_CRC_LEN,
2514 		    ETHER_ALIGN, ifp, NULL);
2515 		if (m == NULL) {
2516 			if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
2517 			ale_rx_update_page(sc, &rx_page, length, &prod);
2518 			continue;
2519 		}
2520 		if ((if_getcapenable(ifp) & IFCAP_RXCSUM) != 0 &&
2521 		    (status & ALE_RD_IPV4) != 0)
2522 			ale_rxcsum(sc, m, status);
2523 		if ((if_getcapenable(ifp) & IFCAP_VLAN_HWTAGGING) != 0 &&
2524 		    (status & ALE_RD_VLAN) != 0) {
2525 			vtags = ALE_RX_VLAN(le32toh(rs->vtags));
2526 			m->m_pkthdr.ether_vtag = ALE_RX_VLAN_TAG(vtags);
2527 			m->m_flags |= M_VLANTAG;
2528 		}
2529 
2530 		/* Pass it to upper layer. */
2531 		ALE_UNLOCK(sc);
2532 		if_input(ifp, m);
2533 		ALE_LOCK(sc);
2534 
2535 		ale_rx_update_page(sc, &rx_page, length, &prod);
2536 	}
2537 
2538 	return (count > 0 ? 0 : EAGAIN);
2539 }
2540 
2541 static void
ale_tick(void * arg)2542 ale_tick(void *arg)
2543 {
2544 	struct ale_softc *sc;
2545 	struct mii_data *mii;
2546 
2547 	sc = (struct ale_softc *)arg;
2548 
2549 	ALE_LOCK_ASSERT(sc);
2550 
2551 	mii = device_get_softc(sc->ale_miibus);
2552 	mii_tick(mii);
2553 	ale_stats_update(sc);
2554 	/*
2555 	 * Reclaim Tx buffers that have been transferred. It's not
2556 	 * needed here but it would release allocated mbuf chains
2557 	 * faster and limit the maximum delay to a hz.
2558 	 */
2559 	ale_txeof(sc);
2560 	ale_watchdog(sc);
2561 	callout_reset(&sc->ale_tick_ch, hz, ale_tick, sc);
2562 }
2563 
2564 static void
ale_reset(struct ale_softc * sc)2565 ale_reset(struct ale_softc *sc)
2566 {
2567 	uint32_t reg;
2568 	int i;
2569 
2570 	/* Initialize PCIe module. From Linux. */
2571 	CSR_WRITE_4(sc, 0x1008, CSR_READ_4(sc, 0x1008) | 0x8000);
2572 
2573 	CSR_WRITE_4(sc, ALE_MASTER_CFG, MASTER_RESET);
2574 	for (i = ALE_RESET_TIMEOUT; i > 0; i--) {
2575 		DELAY(10);
2576 		if ((CSR_READ_4(sc, ALE_MASTER_CFG) & MASTER_RESET) == 0)
2577 			break;
2578 	}
2579 	if (i == 0)
2580 		device_printf(sc->ale_dev, "master reset timeout!\n");
2581 
2582 	for (i = ALE_RESET_TIMEOUT; i > 0; i--) {
2583 		if ((reg = CSR_READ_4(sc, ALE_IDLE_STATUS)) == 0)
2584 			break;
2585 		DELAY(10);
2586 	}
2587 
2588 	if (i == 0)
2589 		device_printf(sc->ale_dev, "reset timeout(0x%08x)!\n", reg);
2590 }
2591 
2592 static void
ale_init(void * xsc)2593 ale_init(void *xsc)
2594 {
2595 	struct ale_softc *sc;
2596 
2597 	sc = (struct ale_softc *)xsc;
2598 	ALE_LOCK(sc);
2599 	ale_init_locked(sc);
2600 	ALE_UNLOCK(sc);
2601 }
2602 
2603 static void
ale_init_locked(struct ale_softc * sc)2604 ale_init_locked(struct ale_softc *sc)
2605 {
2606 	if_t ifp;
2607 	struct mii_data *mii;
2608 	uint8_t eaddr[ETHER_ADDR_LEN];
2609 	bus_addr_t paddr;
2610 	uint32_t reg, rxf_hi, rxf_lo;
2611 
2612 	ALE_LOCK_ASSERT(sc);
2613 
2614 	ifp = sc->ale_ifp;
2615 	mii = device_get_softc(sc->ale_miibus);
2616 
2617 	if ((if_getdrvflags(ifp) & IFF_DRV_RUNNING) != 0)
2618 		return;
2619 	/*
2620 	 * Cancel any pending I/O.
2621 	 */
2622 	ale_stop(sc);
2623 	/*
2624 	 * Reset the chip to a known state.
2625 	 */
2626 	ale_reset(sc);
2627 	/* Initialize Tx descriptors, DMA memory blocks. */
2628 	ale_init_rx_pages(sc);
2629 	ale_init_tx_ring(sc);
2630 
2631 	/* Reprogram the station address. */
2632 	bcopy(if_getlladdr(ifp), eaddr, ETHER_ADDR_LEN);
2633 	CSR_WRITE_4(sc, ALE_PAR0,
2634 	    eaddr[2] << 24 | eaddr[3] << 16 | eaddr[4] << 8 | eaddr[5]);
2635 	CSR_WRITE_4(sc, ALE_PAR1, eaddr[0] << 8 | eaddr[1]);
2636 	/*
2637 	 * Clear WOL status and disable all WOL feature as WOL
2638 	 * would interfere Rx operation under normal environments.
2639 	 */
2640 	CSR_READ_4(sc, ALE_WOL_CFG);
2641 	CSR_WRITE_4(sc, ALE_WOL_CFG, 0);
2642 	/*
2643 	 * Set Tx descriptor/RXF0/CMB base addresses. They share
2644 	 * the same high address part of DMAable region.
2645 	 */
2646 	paddr = sc->ale_cdata.ale_tx_ring_paddr;
2647 	CSR_WRITE_4(sc, ALE_TPD_ADDR_HI, ALE_ADDR_HI(paddr));
2648 	CSR_WRITE_4(sc, ALE_TPD_ADDR_LO, ALE_ADDR_LO(paddr));
2649 	CSR_WRITE_4(sc, ALE_TPD_CNT,
2650 	    (ALE_TX_RING_CNT << TPD_CNT_SHIFT) & TPD_CNT_MASK);
2651 	/* Set Rx page base address, note we use single queue. */
2652 	paddr = sc->ale_cdata.ale_rx_page[0].page_paddr;
2653 	CSR_WRITE_4(sc, ALE_RXF0_PAGE0_ADDR_LO, ALE_ADDR_LO(paddr));
2654 	paddr = sc->ale_cdata.ale_rx_page[1].page_paddr;
2655 	CSR_WRITE_4(sc, ALE_RXF0_PAGE1_ADDR_LO, ALE_ADDR_LO(paddr));
2656 	/* Set Tx/Rx CMB addresses. */
2657 	paddr = sc->ale_cdata.ale_tx_cmb_paddr;
2658 	CSR_WRITE_4(sc, ALE_TX_CMB_ADDR_LO, ALE_ADDR_LO(paddr));
2659 	paddr = sc->ale_cdata.ale_rx_page[0].cmb_paddr;
2660 	CSR_WRITE_4(sc, ALE_RXF0_CMB0_ADDR_LO, ALE_ADDR_LO(paddr));
2661 	paddr = sc->ale_cdata.ale_rx_page[1].cmb_paddr;
2662 	CSR_WRITE_4(sc, ALE_RXF0_CMB1_ADDR_LO, ALE_ADDR_LO(paddr));
2663 	/* Mark RXF0 is valid. */
2664 	CSR_WRITE_1(sc, ALE_RXF0_PAGE0, RXF_VALID);
2665 	CSR_WRITE_1(sc, ALE_RXF0_PAGE1, RXF_VALID);
2666 	/*
2667 	 * No need to initialize RFX1/RXF2/RXF3. We don't use
2668 	 * multi-queue yet.
2669 	 */
2670 
2671 	/* Set Rx page size, excluding guard frame size. */
2672 	CSR_WRITE_4(sc, ALE_RXF_PAGE_SIZE, ALE_RX_PAGE_SZ);
2673 	/* Tell hardware that we're ready to load DMA blocks. */
2674 	CSR_WRITE_4(sc, ALE_DMA_BLOCK, DMA_BLOCK_LOAD);
2675 
2676 	/* Set Rx/Tx interrupt trigger threshold. */
2677 	CSR_WRITE_4(sc, ALE_INT_TRIG_THRESH, (1 << INT_TRIG_RX_THRESH_SHIFT) |
2678 	    (4 << INT_TRIG_TX_THRESH_SHIFT));
2679 	/*
2680 	 * XXX
2681 	 * Set interrupt trigger timer, its purpose and relation
2682 	 * with interrupt moderation mechanism is not clear yet.
2683 	 */
2684 	CSR_WRITE_4(sc, ALE_INT_TRIG_TIMER,
2685 	    ((ALE_USECS(10) << INT_TRIG_RX_TIMER_SHIFT) |
2686 	    (ALE_USECS(1000) << INT_TRIG_TX_TIMER_SHIFT)));
2687 
2688 	/* Configure interrupt moderation timer. */
2689 	reg = ALE_USECS(sc->ale_int_rx_mod) << IM_TIMER_RX_SHIFT;
2690 	reg |= ALE_USECS(sc->ale_int_tx_mod) << IM_TIMER_TX_SHIFT;
2691 	CSR_WRITE_4(sc, ALE_IM_TIMER, reg);
2692 	reg = CSR_READ_4(sc, ALE_MASTER_CFG);
2693 	reg &= ~(MASTER_CHIP_REV_MASK | MASTER_CHIP_ID_MASK);
2694 	reg &= ~(MASTER_IM_RX_TIMER_ENB | MASTER_IM_TX_TIMER_ENB);
2695 	if (ALE_USECS(sc->ale_int_rx_mod) != 0)
2696 		reg |= MASTER_IM_RX_TIMER_ENB;
2697 	if (ALE_USECS(sc->ale_int_tx_mod) != 0)
2698 		reg |= MASTER_IM_TX_TIMER_ENB;
2699 	CSR_WRITE_4(sc, ALE_MASTER_CFG, reg);
2700 	CSR_WRITE_2(sc, ALE_INTR_CLR_TIMER, ALE_USECS(1000));
2701 
2702 	/* Set Maximum frame size of controller. */
2703 	if (if_getmtu(ifp) < ETHERMTU)
2704 		sc->ale_max_frame_size = ETHERMTU;
2705 	else
2706 		sc->ale_max_frame_size = if_getmtu(ifp);
2707 	sc->ale_max_frame_size += ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN +
2708 	    ETHER_CRC_LEN;
2709 	CSR_WRITE_4(sc, ALE_FRAME_SIZE, sc->ale_max_frame_size);
2710 	/* Configure IPG/IFG parameters. */
2711 	CSR_WRITE_4(sc, ALE_IPG_IFG_CFG,
2712 	    ((IPG_IFG_IPGT_DEFAULT << IPG_IFG_IPGT_SHIFT) & IPG_IFG_IPGT_MASK) |
2713 	    ((IPG_IFG_MIFG_DEFAULT << IPG_IFG_MIFG_SHIFT) & IPG_IFG_MIFG_MASK) |
2714 	    ((IPG_IFG_IPG1_DEFAULT << IPG_IFG_IPG1_SHIFT) & IPG_IFG_IPG1_MASK) |
2715 	    ((IPG_IFG_IPG2_DEFAULT << IPG_IFG_IPG2_SHIFT) & IPG_IFG_IPG2_MASK));
2716 	/* Set parameters for half-duplex media. */
2717 	CSR_WRITE_4(sc, ALE_HDPX_CFG,
2718 	    ((HDPX_CFG_LCOL_DEFAULT << HDPX_CFG_LCOL_SHIFT) &
2719 	    HDPX_CFG_LCOL_MASK) |
2720 	    ((HDPX_CFG_RETRY_DEFAULT << HDPX_CFG_RETRY_SHIFT) &
2721 	    HDPX_CFG_RETRY_MASK) | HDPX_CFG_EXC_DEF_EN |
2722 	    ((HDPX_CFG_ABEBT_DEFAULT << HDPX_CFG_ABEBT_SHIFT) &
2723 	    HDPX_CFG_ABEBT_MASK) |
2724 	    ((HDPX_CFG_JAMIPG_DEFAULT << HDPX_CFG_JAMIPG_SHIFT) &
2725 	    HDPX_CFG_JAMIPG_MASK));
2726 
2727 	/* Configure Tx jumbo frame parameters. */
2728 	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0) {
2729 		if (if_getmtu(ifp) < ETHERMTU)
2730 			reg = sc->ale_max_frame_size;
2731 		else if (if_getmtu(ifp) < 6 * 1024)
2732 			reg = (sc->ale_max_frame_size * 2) / 3;
2733 		else
2734 			reg = sc->ale_max_frame_size / 2;
2735 		CSR_WRITE_4(sc, ALE_TX_JUMBO_THRESH,
2736 		    roundup(reg, TX_JUMBO_THRESH_UNIT) >>
2737 		    TX_JUMBO_THRESH_UNIT_SHIFT);
2738 	}
2739 	/* Configure TxQ. */
2740 	reg = (128 << (sc->ale_dma_rd_burst >> DMA_CFG_RD_BURST_SHIFT))
2741 	    << TXQ_CFG_TX_FIFO_BURST_SHIFT;
2742 	reg |= (TXQ_CFG_TPD_BURST_DEFAULT << TXQ_CFG_TPD_BURST_SHIFT) &
2743 	    TXQ_CFG_TPD_BURST_MASK;
2744 	CSR_WRITE_4(sc, ALE_TXQ_CFG, reg | TXQ_CFG_ENHANCED_MODE | TXQ_CFG_ENB);
2745 
2746 	/* Configure Rx jumbo frame & flow control parameters. */
2747 	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0) {
2748 		reg = roundup(sc->ale_max_frame_size, RX_JUMBO_THRESH_UNIT);
2749 		CSR_WRITE_4(sc, ALE_RX_JUMBO_THRESH,
2750 		    (((reg >> RX_JUMBO_THRESH_UNIT_SHIFT) <<
2751 		    RX_JUMBO_THRESH_MASK_SHIFT) & RX_JUMBO_THRESH_MASK) |
2752 		    ((RX_JUMBO_LKAH_DEFAULT << RX_JUMBO_LKAH_SHIFT) &
2753 		    RX_JUMBO_LKAH_MASK));
2754 		reg = CSR_READ_4(sc, ALE_SRAM_RX_FIFO_LEN);
2755 		rxf_hi = (reg * 7) / 10;
2756 		rxf_lo = (reg * 3)/ 10;
2757 		CSR_WRITE_4(sc, ALE_RX_FIFO_PAUSE_THRESH,
2758 		    ((rxf_lo << RX_FIFO_PAUSE_THRESH_LO_SHIFT) &
2759 		    RX_FIFO_PAUSE_THRESH_LO_MASK) |
2760 		    ((rxf_hi << RX_FIFO_PAUSE_THRESH_HI_SHIFT) &
2761 		    RX_FIFO_PAUSE_THRESH_HI_MASK));
2762 	}
2763 
2764 	/* Disable RSS. */
2765 	CSR_WRITE_4(sc, ALE_RSS_IDT_TABLE0, 0);
2766 	CSR_WRITE_4(sc, ALE_RSS_CPU, 0);
2767 
2768 	/* Configure RxQ. */
2769 	CSR_WRITE_4(sc, ALE_RXQ_CFG,
2770 	    RXQ_CFG_ALIGN_32 | RXQ_CFG_CUT_THROUGH_ENB | RXQ_CFG_ENB);
2771 
2772 	/* Configure DMA parameters. */
2773 	reg = 0;
2774 	if ((sc->ale_flags & ALE_FLAG_TXCMB_BUG) == 0)
2775 		reg |= DMA_CFG_TXCMB_ENB;
2776 	CSR_WRITE_4(sc, ALE_DMA_CFG,
2777 	    DMA_CFG_OUT_ORDER | DMA_CFG_RD_REQ_PRI | DMA_CFG_RCB_64 |
2778 	    sc->ale_dma_rd_burst | reg |
2779 	    sc->ale_dma_wr_burst | DMA_CFG_RXCMB_ENB |
2780 	    ((DMA_CFG_RD_DELAY_CNT_DEFAULT << DMA_CFG_RD_DELAY_CNT_SHIFT) &
2781 	    DMA_CFG_RD_DELAY_CNT_MASK) |
2782 	    ((DMA_CFG_WR_DELAY_CNT_DEFAULT << DMA_CFG_WR_DELAY_CNT_SHIFT) &
2783 	    DMA_CFG_WR_DELAY_CNT_MASK));
2784 
2785 	/*
2786 	 * Hardware can be configured to issue SMB interrupt based
2787 	 * on programmed interval. Since there is a callout that is
2788 	 * invoked for every hz in driver we use that instead of
2789 	 * relying on periodic SMB interrupt.
2790 	 */
2791 	CSR_WRITE_4(sc, ALE_SMB_STAT_TIMER, ALE_USECS(0));
2792 	/* Clear MAC statistics. */
2793 	ale_stats_clear(sc);
2794 
2795 	/*
2796 	 * Configure Tx/Rx MACs.
2797 	 *  - Auto-padding for short frames.
2798 	 *  - Enable CRC generation.
2799 	 *  Actual reconfiguration of MAC for resolved speed/duplex
2800 	 *  is followed after detection of link establishment.
2801 	 *  AR81xx always does checksum computation regardless of
2802 	 *  MAC_CFG_RXCSUM_ENB bit. In fact, setting the bit will
2803 	 *  cause Rx handling issue for fragmented IP datagrams due
2804 	 *  to silicon bug.
2805 	 */
2806 	reg = MAC_CFG_TX_CRC_ENB | MAC_CFG_TX_AUTO_PAD | MAC_CFG_FULL_DUPLEX |
2807 	    ((MAC_CFG_PREAMBLE_DEFAULT << MAC_CFG_PREAMBLE_SHIFT) &
2808 	    MAC_CFG_PREAMBLE_MASK);
2809 	if ((sc->ale_flags & ALE_FLAG_FASTETHER) != 0)
2810 		reg |= MAC_CFG_SPEED_10_100;
2811 	else
2812 		reg |= MAC_CFG_SPEED_1000;
2813 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2814 
2815 	/* Set up the receive filter. */
2816 	ale_rxfilter(sc);
2817 	ale_rxvlan(sc);
2818 
2819 	/* Acknowledge all pending interrupts and clear it. */
2820 	CSR_WRITE_4(sc, ALE_INTR_MASK, ALE_INTRS);
2821 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0xFFFFFFFF);
2822 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0);
2823 
2824 	if_setdrvflagbits(ifp, IFF_DRV_RUNNING, 0);
2825 	if_setdrvflagbits(ifp, 0, IFF_DRV_OACTIVE);
2826 
2827 	sc->ale_flags &= ~ALE_FLAG_LINK;
2828 	/* Switch to the current media. */
2829 	mii_mediachg(mii);
2830 
2831 	callout_reset(&sc->ale_tick_ch, hz, ale_tick, sc);
2832 }
2833 
2834 static void
ale_stop(struct ale_softc * sc)2835 ale_stop(struct ale_softc *sc)
2836 {
2837 	if_t ifp;
2838 	struct ale_txdesc *txd;
2839 	uint32_t reg;
2840 	int i;
2841 
2842 	ALE_LOCK_ASSERT(sc);
2843 	/*
2844 	 * Mark the interface down and cancel the watchdog timer.
2845 	 */
2846 	ifp = sc->ale_ifp;
2847 	if_setdrvflagbits(ifp, 0, (IFF_DRV_RUNNING | IFF_DRV_OACTIVE));
2848 	sc->ale_flags &= ~ALE_FLAG_LINK;
2849 	callout_stop(&sc->ale_tick_ch);
2850 	sc->ale_watchdog_timer = 0;
2851 	ale_stats_update(sc);
2852 	/* Disable interrupts. */
2853 	CSR_WRITE_4(sc, ALE_INTR_MASK, 0);
2854 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0xFFFFFFFF);
2855 	/* Disable queue processing and DMA. */
2856 	reg = CSR_READ_4(sc, ALE_TXQ_CFG);
2857 	reg &= ~TXQ_CFG_ENB;
2858 	CSR_WRITE_4(sc, ALE_TXQ_CFG, reg);
2859 	reg = CSR_READ_4(sc, ALE_RXQ_CFG);
2860 	reg &= ~RXQ_CFG_ENB;
2861 	CSR_WRITE_4(sc, ALE_RXQ_CFG, reg);
2862 	reg = CSR_READ_4(sc, ALE_DMA_CFG);
2863 	reg &= ~(DMA_CFG_TXCMB_ENB | DMA_CFG_RXCMB_ENB);
2864 	CSR_WRITE_4(sc, ALE_DMA_CFG, reg);
2865 	DELAY(1000);
2866 	/* Stop Rx/Tx MACs. */
2867 	ale_stop_mac(sc);
2868 	/* Disable interrupts which might be touched in taskq handler. */
2869 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0xFFFFFFFF);
2870 
2871 	/*
2872 	 * Free TX mbufs still in the queues.
2873 	 */
2874 	for (i = 0; i < ALE_TX_RING_CNT; i++) {
2875 		txd = &sc->ale_cdata.ale_txdesc[i];
2876 		if (txd->tx_m != NULL) {
2877 			bus_dmamap_sync(sc->ale_cdata.ale_tx_tag,
2878 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2879 			bus_dmamap_unload(sc->ale_cdata.ale_tx_tag,
2880 			    txd->tx_dmamap);
2881 			m_freem(txd->tx_m);
2882 			txd->tx_m = NULL;
2883 		}
2884         }
2885 }
2886 
2887 static void
ale_stop_mac(struct ale_softc * sc)2888 ale_stop_mac(struct ale_softc *sc)
2889 {
2890 	uint32_t reg;
2891 	int i;
2892 
2893 	ALE_LOCK_ASSERT(sc);
2894 
2895 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
2896 	if ((reg & (MAC_CFG_TX_ENB | MAC_CFG_RX_ENB)) != 0) {
2897 		reg &= ~(MAC_CFG_TX_ENB | MAC_CFG_RX_ENB);
2898 		CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2899 	}
2900 
2901 	for (i = ALE_TIMEOUT; i > 0; i--) {
2902 		reg = CSR_READ_4(sc, ALE_IDLE_STATUS);
2903 		if (reg == 0)
2904 			break;
2905 		DELAY(10);
2906 	}
2907 	if (i == 0)
2908 		device_printf(sc->ale_dev,
2909 		    "could not disable Tx/Rx MAC(0x%08x)!\n", reg);
2910 }
2911 
2912 static void
ale_init_tx_ring(struct ale_softc * sc)2913 ale_init_tx_ring(struct ale_softc *sc)
2914 {
2915 	struct ale_txdesc *txd;
2916 	int i;
2917 
2918 	ALE_LOCK_ASSERT(sc);
2919 
2920 	sc->ale_cdata.ale_tx_prod = 0;
2921 	sc->ale_cdata.ale_tx_cons = 0;
2922 	sc->ale_cdata.ale_tx_cnt = 0;
2923 
2924 	bzero(sc->ale_cdata.ale_tx_ring, ALE_TX_RING_SZ);
2925 	bzero(sc->ale_cdata.ale_tx_cmb, ALE_TX_CMB_SZ);
2926 	for (i = 0; i < ALE_TX_RING_CNT; i++) {
2927 		txd = &sc->ale_cdata.ale_txdesc[i];
2928 		txd->tx_m = NULL;
2929 	}
2930 	*sc->ale_cdata.ale_tx_cmb = 0;
2931 	bus_dmamap_sync(sc->ale_cdata.ale_tx_cmb_tag,
2932 	    sc->ale_cdata.ale_tx_cmb_map,
2933 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2934 	bus_dmamap_sync(sc->ale_cdata.ale_tx_ring_tag,
2935 	    sc->ale_cdata.ale_tx_ring_map,
2936 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2937 }
2938 
2939 static void
ale_init_rx_pages(struct ale_softc * sc)2940 ale_init_rx_pages(struct ale_softc *sc)
2941 {
2942 	struct ale_rx_page *rx_page;
2943 	int i;
2944 
2945 	ALE_LOCK_ASSERT(sc);
2946 
2947 	sc->ale_morework = 0;
2948 	sc->ale_cdata.ale_rx_seqno = 0;
2949 	sc->ale_cdata.ale_rx_curp = 0;
2950 
2951 	for (i = 0; i < ALE_RX_PAGES; i++) {
2952 		rx_page = &sc->ale_cdata.ale_rx_page[i];
2953 		bzero(rx_page->page_addr, sc->ale_pagesize);
2954 		bzero(rx_page->cmb_addr, ALE_RX_CMB_SZ);
2955 		rx_page->cons = 0;
2956 		*rx_page->cmb_addr = 0;
2957 		bus_dmamap_sync(rx_page->page_tag, rx_page->page_map,
2958 		    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2959 		bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2960 		    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2961 	}
2962 }
2963 
2964 static void
ale_rxvlan(struct ale_softc * sc)2965 ale_rxvlan(struct ale_softc *sc)
2966 {
2967 	if_t ifp;
2968 	uint32_t reg;
2969 
2970 	ALE_LOCK_ASSERT(sc);
2971 
2972 	ifp = sc->ale_ifp;
2973 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
2974 	reg &= ~MAC_CFG_VLAN_TAG_STRIP;
2975 	if ((if_getcapenable(ifp) & IFCAP_VLAN_HWTAGGING) != 0)
2976 		reg |= MAC_CFG_VLAN_TAG_STRIP;
2977 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2978 }
2979 
2980 static u_int
ale_hash_maddr(void * arg,struct sockaddr_dl * sdl,u_int cnt)2981 ale_hash_maddr(void *arg, struct sockaddr_dl *sdl, u_int cnt)
2982 {
2983 	uint32_t crc, *mchash = arg;
2984 
2985 	crc = ether_crc32_be(LLADDR(sdl), ETHER_ADDR_LEN);
2986 	mchash[crc >> 31] |= 1 << ((crc >> 26) & 0x1f);
2987 
2988 	return (1);
2989 }
2990 
2991 static void
ale_rxfilter(struct ale_softc * sc)2992 ale_rxfilter(struct ale_softc *sc)
2993 {
2994 	if_t ifp;
2995 	uint32_t mchash[2];
2996 	uint32_t rxcfg;
2997 
2998 	ALE_LOCK_ASSERT(sc);
2999 
3000 	ifp = sc->ale_ifp;
3001 
3002 	rxcfg = CSR_READ_4(sc, ALE_MAC_CFG);
3003 	rxcfg &= ~(MAC_CFG_ALLMULTI | MAC_CFG_BCAST | MAC_CFG_PROMISC);
3004 	if ((if_getflags(ifp) & IFF_BROADCAST) != 0)
3005 		rxcfg |= MAC_CFG_BCAST;
3006 	if ((if_getflags(ifp) & (IFF_PROMISC | IFF_ALLMULTI)) != 0) {
3007 		if ((if_getflags(ifp) & IFF_PROMISC) != 0)
3008 			rxcfg |= MAC_CFG_PROMISC;
3009 		if ((if_getflags(ifp) & IFF_ALLMULTI) != 0)
3010 			rxcfg |= MAC_CFG_ALLMULTI;
3011 		CSR_WRITE_4(sc, ALE_MAR0, 0xFFFFFFFF);
3012 		CSR_WRITE_4(sc, ALE_MAR1, 0xFFFFFFFF);
3013 		CSR_WRITE_4(sc, ALE_MAC_CFG, rxcfg);
3014 		return;
3015 	}
3016 
3017 	/* Program new filter. */
3018 	bzero(mchash, sizeof(mchash));
3019 	if_foreach_llmaddr(ifp, ale_hash_maddr, &mchash);
3020 
3021 	CSR_WRITE_4(sc, ALE_MAR0, mchash[0]);
3022 	CSR_WRITE_4(sc, ALE_MAR1, mchash[1]);
3023 	CSR_WRITE_4(sc, ALE_MAC_CFG, rxcfg);
3024 }
3025 
3026 static int
sysctl_int_range(SYSCTL_HANDLER_ARGS,int low,int high)3027 sysctl_int_range(SYSCTL_HANDLER_ARGS, int low, int high)
3028 {
3029 	int error, value;
3030 
3031 	if (arg1 == NULL)
3032 		return (EINVAL);
3033 	value = *(int *)arg1;
3034 	error = sysctl_handle_int(oidp, &value, 0, req);
3035 	if (error || req->newptr == NULL)
3036 		return (error);
3037 	if (value < low || value > high)
3038 		return (EINVAL);
3039         *(int *)arg1 = value;
3040 
3041         return (0);
3042 }
3043 
3044 static int
sysctl_hw_ale_proc_limit(SYSCTL_HANDLER_ARGS)3045 sysctl_hw_ale_proc_limit(SYSCTL_HANDLER_ARGS)
3046 {
3047 	return (sysctl_int_range(oidp, arg1, arg2, req,
3048 	    ALE_PROC_MIN, ALE_PROC_MAX));
3049 }
3050 
3051 static int
sysctl_hw_ale_int_mod(SYSCTL_HANDLER_ARGS)3052 sysctl_hw_ale_int_mod(SYSCTL_HANDLER_ARGS)
3053 {
3054 
3055 	return (sysctl_int_range(oidp, arg1, arg2, req,
3056 	    ALE_IM_TIMER_MIN, ALE_IM_TIMER_MAX));
3057 }
3058