1 /*-
2  * Copyright (c) 2008, Pyun YongHyeon <yongari@FreeBSD.org>
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice unmodified, this list of conditions, and the following
10  *    disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
25  * SUCH DAMAGE.
26  */
27 
28 /* Driver for Atheros AR8121/AR8113/AR8114 PCIe Ethernet. */
29 
30 #include <sys/cdefs.h>
31 __FBSDID("$FreeBSD: stable/9/sys/dev/ale/if_ale.c 273358 2014-10-21 01:15:43Z yongari $");
32 
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/bus.h>
36 #include <sys/endian.h>
37 #include <sys/kernel.h>
38 #include <sys/malloc.h>
39 #include <sys/mbuf.h>
40 #include <sys/module.h>
41 #include <sys/rman.h>
42 #include <sys/queue.h>
43 #include <sys/socket.h>
44 #include <sys/sockio.h>
45 #include <sys/sysctl.h>
46 #include <sys/taskqueue.h>
47 
48 #include <net/bpf.h>
49 #include <net/if.h>
50 #include <net/if_arp.h>
51 #include <net/ethernet.h>
52 #include <net/if_dl.h>
53 #include <net/if_llc.h>
54 #include <net/if_media.h>
55 #include <net/if_types.h>
56 #include <net/if_vlan_var.h>
57 
58 #include <netinet/in.h>
59 #include <netinet/in_systm.h>
60 #include <netinet/ip.h>
61 #include <netinet/tcp.h>
62 
63 #include <dev/mii/mii.h>
64 #include <dev/mii/miivar.h>
65 
66 #include <dev/pci/pcireg.h>
67 #include <dev/pci/pcivar.h>
68 
69 #include <machine/bus.h>
70 #include <machine/in_cksum.h>
71 
72 #include <dev/ale/if_alereg.h>
73 #include <dev/ale/if_alevar.h>
74 
75 /* "device miibus" required.  See GENERIC if you get errors here. */
76 #include "miibus_if.h"
77 
78 /* For more information about Tx checksum offload issues see ale_encap(). */
79 #define	ALE_CSUM_FEATURES	(CSUM_TCP | CSUM_UDP)
80 
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(struct ifnet *, 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(struct ifnet *);
123 static void	ale_mediastatus(struct ifnet *, 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(struct ifnet *);
138 static void	ale_start_locked(struct ifnet *);
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 static devclass_t ale_devclass;
176 
177 DRIVER_MODULE(ale, pci, ale_driver, ale_devclass, NULL, NULL);
178 DRIVER_MODULE(miibus, ale, miibus_driver, miibus_devclass, 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 	struct ifnet *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 	    (ifp->if_drv_flags & 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(struct ifnet * ifp,struct ifmediareq * ifmr)297 ale_mediastatus(struct ifnet *ifp, struct ifmediareq *ifmr)
298 {
299 	struct ale_softc *sc;
300 	struct mii_data *mii;
301 
302 	sc = ifp->if_softc;
303 	ALE_LOCK(sc);
304 	if ((ifp->if_flags & 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(struct ifnet * ifp)317 ale_mediachange(struct ifnet *ifp)
318 {
319 	struct ale_softc *sc;
320 	struct mii_data *mii;
321 	struct mii_softc *miisc;
322 	int error;
323 
324 	sc = ifp->if_softc;
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 < sizeof(ale_devs) / sizeof(ale_devs[0]); 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 	struct ifnet *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 	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 (ifp == NULL) {
613 		device_printf(dev, "cannot allocate ifnet structure.\n");
614 		error = ENXIO;
615 		goto fail;
616 	}
617 
618 	ifp->if_softc = sc;
619 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
620 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
621 	ifp->if_ioctl = ale_ioctl;
622 	ifp->if_start = ale_start;
623 	ifp->if_init = ale_init;
624 	ifp->if_snd.ifq_drv_maxlen = ALE_TX_RING_CNT - 1;
625 	IFQ_SET_MAXLEN(&ifp->if_snd, ifp->if_snd.ifq_drv_maxlen);
626 	IFQ_SET_READY(&ifp->if_snd);
627 	ifp->if_capabilities = IFCAP_RXCSUM | IFCAP_TXCSUM | IFCAP_TSO4;
628 	ifp->if_hwassist = ALE_CSUM_FEATURES | CSUM_TSO;
629 	if (pci_find_cap(dev, PCIY_PMG, &pmc) == 0) {
630 		sc->ale_flags |= ALE_FLAG_PMCAP;
631 		ifp->if_capabilities |= IFCAP_WOL_MAGIC | IFCAP_WOL_MCAST;
632 	}
633 	ifp->if_capenable = ifp->if_capabilities;
634 
635 	/* Set up MII bus. */
636 	error = mii_attach(dev, &sc->ale_miibus, ifp, ale_mediachange,
637 	    ale_mediastatus, BMSR_DEFCAPMASK, sc->ale_phyaddr, MII_OFFSET_ANY,
638 	    MIIF_DOPAUSE);
639 	if (error != 0) {
640 		device_printf(dev, "attaching PHYs failed\n");
641 		goto fail;
642 	}
643 
644 	ether_ifattach(ifp, sc->ale_eaddr);
645 
646 	/* VLAN capability setup. */
647 	ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING |
648 	    IFCAP_VLAN_HWCSUM | IFCAP_VLAN_HWTSO;
649 	ifp->if_capenable = ifp->if_capabilities;
650 	/*
651 	 * Even though controllers supported by ale(3) have Rx checksum
652 	 * offload bug the workaround for fragmented frames seemed to
653 	 * work so far. However it seems Rx checksum offload does not
654 	 * work under certain conditions. So disable Rx checksum offload
655 	 * until I find more clue about it but allow users to override it.
656 	 */
657 	ifp->if_capenable &= ~IFCAP_RXCSUM;
658 
659 	/* Tell the upper layer(s) we support long frames. */
660 	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
661 
662 	/* Create local taskq. */
663 	sc->ale_tq = taskqueue_create_fast("ale_taskq", M_WAITOK,
664 	    taskqueue_thread_enqueue, &sc->ale_tq);
665 	if (sc->ale_tq == NULL) {
666 		device_printf(dev, "could not create taskqueue.\n");
667 		ether_ifdetach(ifp);
668 		error = ENXIO;
669 		goto fail;
670 	}
671 	taskqueue_start_threads(&sc->ale_tq, 1, PI_NET, "%s taskq",
672 	    device_get_nameunit(sc->ale_dev));
673 
674 	if ((sc->ale_flags & ALE_FLAG_MSIX) != 0)
675 		msic = ALE_MSIX_MESSAGES;
676 	else if ((sc->ale_flags & ALE_FLAG_MSI) != 0)
677 		msic = ALE_MSI_MESSAGES;
678 	else
679 		msic = 1;
680 	for (i = 0; i < msic; i++) {
681 		error = bus_setup_intr(dev, sc->ale_irq[i],
682 		    INTR_TYPE_NET | INTR_MPSAFE, ale_intr, NULL, sc,
683 		    &sc->ale_intrhand[i]);
684 		if (error != 0)
685 			break;
686 	}
687 	if (error != 0) {
688 		device_printf(dev, "could not set up interrupt handler.\n");
689 		taskqueue_free(sc->ale_tq);
690 		sc->ale_tq = NULL;
691 		ether_ifdetach(ifp);
692 		goto fail;
693 	}
694 
695 fail:
696 	if (error != 0)
697 		ale_detach(dev);
698 
699 	return (error);
700 }
701 
702 static int
ale_detach(device_t dev)703 ale_detach(device_t dev)
704 {
705 	struct ale_softc *sc;
706 	struct ifnet *ifp;
707 	int i, msic;
708 
709 	sc = device_get_softc(dev);
710 
711 	ifp = sc->ale_ifp;
712 	if (device_is_attached(dev)) {
713 		ether_ifdetach(ifp);
714 		ALE_LOCK(sc);
715 		ale_stop(sc);
716 		ALE_UNLOCK(sc);
717 		callout_drain(&sc->ale_tick_ch);
718 		taskqueue_drain(sc->ale_tq, &sc->ale_int_task);
719 	}
720 
721 	if (sc->ale_tq != NULL) {
722 		taskqueue_drain(sc->ale_tq, &sc->ale_int_task);
723 		taskqueue_free(sc->ale_tq);
724 		sc->ale_tq = NULL;
725 	}
726 
727 	if (sc->ale_miibus != NULL) {
728 		device_delete_child(dev, sc->ale_miibus);
729 		sc->ale_miibus = NULL;
730 	}
731 	bus_generic_detach(dev);
732 	ale_dma_free(sc);
733 
734 	if (ifp != NULL) {
735 		if_free(ifp);
736 		sc->ale_ifp = NULL;
737 	}
738 
739 	if ((sc->ale_flags & ALE_FLAG_MSIX) != 0)
740 		msic = ALE_MSIX_MESSAGES;
741 	else if ((sc->ale_flags & ALE_FLAG_MSI) != 0)
742 		msic = ALE_MSI_MESSAGES;
743 	else
744 		msic = 1;
745 	for (i = 0; i < msic; i++) {
746 		if (sc->ale_intrhand[i] != NULL) {
747 			bus_teardown_intr(dev, sc->ale_irq[i],
748 			    sc->ale_intrhand[i]);
749 			sc->ale_intrhand[i] = NULL;
750 		}
751 	}
752 
753 	bus_release_resources(dev, sc->ale_irq_spec, sc->ale_irq);
754 	if ((sc->ale_flags & (ALE_FLAG_MSI | ALE_FLAG_MSIX)) != 0)
755 		pci_release_msi(dev);
756 	bus_release_resources(dev, sc->ale_res_spec, sc->ale_res);
757 	mtx_destroy(&sc->ale_mtx);
758 
759 	return (0);
760 }
761 
762 #define	ALE_SYSCTL_STAT_ADD32(c, h, n, p, d)	\
763 	    SYSCTL_ADD_UINT(c, h, OID_AUTO, n, CTLFLAG_RD, p, 0, d)
764 
765 #if __FreeBSD_version >= 900030
766 #define	ALE_SYSCTL_STAT_ADD64(c, h, n, p, d)	\
767 	    SYSCTL_ADD_UQUAD(c, h, OID_AUTO, n, CTLFLAG_RD, p, d)
768 #elif __FreeBSD_version > 800000
769 #define	ALE_SYSCTL_STAT_ADD64(c, h, n, p, d)	\
770 	    SYSCTL_ADD_QUAD(c, h, OID_AUTO, n, CTLFLAG_RD, p, d)
771 #else
772 #define	ALE_SYSCTL_STAT_ADD64(c, h, n, p, d)	\
773 	    SYSCTL_ADD_ULONG(c, h, OID_AUTO, n, CTLFLAG_RD, p, d)
774 #endif
775 
776 static void
ale_sysctl_node(struct ale_softc * sc)777 ale_sysctl_node(struct ale_softc *sc)
778 {
779 	struct sysctl_ctx_list *ctx;
780 	struct sysctl_oid_list *child, *parent;
781 	struct sysctl_oid *tree;
782 	struct ale_hw_stats *stats;
783 	int error;
784 
785 	stats = &sc->ale_stats;
786 	ctx = device_get_sysctl_ctx(sc->ale_dev);
787 	child = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->ale_dev));
788 
789 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "int_rx_mod",
790 	    CTLTYPE_INT | CTLFLAG_RW, &sc->ale_int_rx_mod, 0,
791 	    sysctl_hw_ale_int_mod, "I", "ale Rx interrupt moderation");
792 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "int_tx_mod",
793 	    CTLTYPE_INT | CTLFLAG_RW, &sc->ale_int_tx_mod, 0,
794 	    sysctl_hw_ale_int_mod, "I", "ale Tx interrupt moderation");
795 	/* Pull in device tunables. */
796 	sc->ale_int_rx_mod = ALE_IM_RX_TIMER_DEFAULT;
797 	error = resource_int_value(device_get_name(sc->ale_dev),
798 	    device_get_unit(sc->ale_dev), "int_rx_mod", &sc->ale_int_rx_mod);
799 	if (error == 0) {
800 		if (sc->ale_int_rx_mod < ALE_IM_TIMER_MIN ||
801 		    sc->ale_int_rx_mod > ALE_IM_TIMER_MAX) {
802 			device_printf(sc->ale_dev, "int_rx_mod value out of "
803 			    "range; using default: %d\n",
804 			    ALE_IM_RX_TIMER_DEFAULT);
805 			sc->ale_int_rx_mod = ALE_IM_RX_TIMER_DEFAULT;
806 		}
807 	}
808 	sc->ale_int_tx_mod = ALE_IM_TX_TIMER_DEFAULT;
809 	error = resource_int_value(device_get_name(sc->ale_dev),
810 	    device_get_unit(sc->ale_dev), "int_tx_mod", &sc->ale_int_tx_mod);
811 	if (error == 0) {
812 		if (sc->ale_int_tx_mod < ALE_IM_TIMER_MIN ||
813 		    sc->ale_int_tx_mod > ALE_IM_TIMER_MAX) {
814 			device_printf(sc->ale_dev, "int_tx_mod value out of "
815 			    "range; using default: %d\n",
816 			    ALE_IM_TX_TIMER_DEFAULT);
817 			sc->ale_int_tx_mod = ALE_IM_TX_TIMER_DEFAULT;
818 		}
819 	}
820 	SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "process_limit",
821 	    CTLTYPE_INT | CTLFLAG_RW, &sc->ale_process_limit, 0,
822 	    sysctl_hw_ale_proc_limit, "I",
823 	    "max number of Rx events to process");
824 	/* Pull in device tunables. */
825 	sc->ale_process_limit = ALE_PROC_DEFAULT;
826 	error = resource_int_value(device_get_name(sc->ale_dev),
827 	    device_get_unit(sc->ale_dev), "process_limit",
828 	    &sc->ale_process_limit);
829 	if (error == 0) {
830 		if (sc->ale_process_limit < ALE_PROC_MIN ||
831 		    sc->ale_process_limit > ALE_PROC_MAX) {
832 			device_printf(sc->ale_dev,
833 			    "process_limit value out of range; "
834 			    "using default: %d\n", ALE_PROC_DEFAULT);
835 			sc->ale_process_limit = ALE_PROC_DEFAULT;
836 		}
837 	}
838 
839 	/* Misc statistics. */
840 	ALE_SYSCTL_STAT_ADD32(ctx, child, "reset_brk_seq",
841 	    &stats->reset_brk_seq,
842 	    "Controller resets due to broken Rx sequnce number");
843 
844 	tree = SYSCTL_ADD_NODE(ctx, child, OID_AUTO, "stats", CTLFLAG_RD,
845 	    NULL, "ATE statistics");
846 	parent = SYSCTL_CHILDREN(tree);
847 
848 	/* Rx statistics. */
849 	tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "rx", CTLFLAG_RD,
850 	    NULL, "Rx MAC statistics");
851 	child = SYSCTL_CHILDREN(tree);
852 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_frames",
853 	    &stats->rx_frames, "Good frames");
854 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_bcast_frames",
855 	    &stats->rx_bcast_frames, "Good broadcast frames");
856 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_mcast_frames",
857 	    &stats->rx_mcast_frames, "Good multicast frames");
858 	ALE_SYSCTL_STAT_ADD32(ctx, child, "pause_frames",
859 	    &stats->rx_pause_frames, "Pause control frames");
860 	ALE_SYSCTL_STAT_ADD32(ctx, child, "control_frames",
861 	    &stats->rx_control_frames, "Control frames");
862 	ALE_SYSCTL_STAT_ADD32(ctx, child, "crc_errs",
863 	    &stats->rx_crcerrs, "CRC errors");
864 	ALE_SYSCTL_STAT_ADD32(ctx, child, "len_errs",
865 	    &stats->rx_lenerrs, "Frames with length mismatched");
866 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_octets",
867 	    &stats->rx_bytes, "Good octets");
868 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_bcast_octets",
869 	    &stats->rx_bcast_bytes, "Good broadcast octets");
870 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_mcast_octets",
871 	    &stats->rx_mcast_bytes, "Good multicast octets");
872 	ALE_SYSCTL_STAT_ADD32(ctx, child, "runts",
873 	    &stats->rx_runts, "Too short frames");
874 	ALE_SYSCTL_STAT_ADD32(ctx, child, "fragments",
875 	    &stats->rx_fragments, "Fragmented frames");
876 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_64",
877 	    &stats->rx_pkts_64, "64 bytes frames");
878 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_65_127",
879 	    &stats->rx_pkts_65_127, "65 to 127 bytes frames");
880 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_128_255",
881 	    &stats->rx_pkts_128_255, "128 to 255 bytes frames");
882 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_256_511",
883 	    &stats->rx_pkts_256_511, "256 to 511 bytes frames");
884 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_512_1023",
885 	    &stats->rx_pkts_512_1023, "512 to 1023 bytes frames");
886 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1024_1518",
887 	    &stats->rx_pkts_1024_1518, "1024 to 1518 bytes frames");
888 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1519_max",
889 	    &stats->rx_pkts_1519_max, "1519 to max frames");
890 	ALE_SYSCTL_STAT_ADD32(ctx, child, "trunc_errs",
891 	    &stats->rx_pkts_truncated, "Truncated frames due to MTU size");
892 	ALE_SYSCTL_STAT_ADD32(ctx, child, "fifo_oflows",
893 	    &stats->rx_fifo_oflows, "FIFO overflows");
894 	ALE_SYSCTL_STAT_ADD32(ctx, child, "rrs_errs",
895 	    &stats->rx_rrs_errs, "Return status write-back errors");
896 	ALE_SYSCTL_STAT_ADD32(ctx, child, "align_errs",
897 	    &stats->rx_alignerrs, "Alignment errors");
898 	ALE_SYSCTL_STAT_ADD32(ctx, child, "filtered",
899 	    &stats->rx_pkts_filtered,
900 	    "Frames dropped due to address filtering");
901 
902 	/* Tx statistics. */
903 	tree = SYSCTL_ADD_NODE(ctx, parent, OID_AUTO, "tx", CTLFLAG_RD,
904 	    NULL, "Tx MAC statistics");
905 	child = SYSCTL_CHILDREN(tree);
906 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_frames",
907 	    &stats->tx_frames, "Good frames");
908 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_bcast_frames",
909 	    &stats->tx_bcast_frames, "Good broadcast frames");
910 	ALE_SYSCTL_STAT_ADD32(ctx, child, "good_mcast_frames",
911 	    &stats->tx_mcast_frames, "Good multicast frames");
912 	ALE_SYSCTL_STAT_ADD32(ctx, child, "pause_frames",
913 	    &stats->tx_pause_frames, "Pause control frames");
914 	ALE_SYSCTL_STAT_ADD32(ctx, child, "control_frames",
915 	    &stats->tx_control_frames, "Control frames");
916 	ALE_SYSCTL_STAT_ADD32(ctx, child, "excess_defers",
917 	    &stats->tx_excess_defer, "Frames with excessive derferrals");
918 	ALE_SYSCTL_STAT_ADD32(ctx, child, "defers",
919 	    &stats->tx_excess_defer, "Frames with derferrals");
920 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_octets",
921 	    &stats->tx_bytes, "Good octets");
922 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_bcast_octets",
923 	    &stats->tx_bcast_bytes, "Good broadcast octets");
924 	ALE_SYSCTL_STAT_ADD64(ctx, child, "good_mcast_octets",
925 	    &stats->tx_mcast_bytes, "Good multicast octets");
926 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_64",
927 	    &stats->tx_pkts_64, "64 bytes frames");
928 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_65_127",
929 	    &stats->tx_pkts_65_127, "65 to 127 bytes frames");
930 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_128_255",
931 	    &stats->tx_pkts_128_255, "128 to 255 bytes frames");
932 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_256_511",
933 	    &stats->tx_pkts_256_511, "256 to 511 bytes frames");
934 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_512_1023",
935 	    &stats->tx_pkts_512_1023, "512 to 1023 bytes frames");
936 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1024_1518",
937 	    &stats->tx_pkts_1024_1518, "1024 to 1518 bytes frames");
938 	ALE_SYSCTL_STAT_ADD32(ctx, child, "frames_1519_max",
939 	    &stats->tx_pkts_1519_max, "1519 to max frames");
940 	ALE_SYSCTL_STAT_ADD32(ctx, child, "single_colls",
941 	    &stats->tx_single_colls, "Single collisions");
942 	ALE_SYSCTL_STAT_ADD32(ctx, child, "multi_colls",
943 	    &stats->tx_multi_colls, "Multiple collisions");
944 	ALE_SYSCTL_STAT_ADD32(ctx, child, "late_colls",
945 	    &stats->tx_late_colls, "Late collisions");
946 	ALE_SYSCTL_STAT_ADD32(ctx, child, "excess_colls",
947 	    &stats->tx_excess_colls, "Excessive collisions");
948 	ALE_SYSCTL_STAT_ADD32(ctx, child, "underruns",
949 	    &stats->tx_underrun, "FIFO underruns");
950 	ALE_SYSCTL_STAT_ADD32(ctx, child, "desc_underruns",
951 	    &stats->tx_desc_underrun, "Descriptor write-back errors");
952 	ALE_SYSCTL_STAT_ADD32(ctx, child, "len_errs",
953 	    &stats->tx_lenerrs, "Frames with length mismatched");
954 	ALE_SYSCTL_STAT_ADD32(ctx, child, "trunc_errs",
955 	    &stats->tx_pkts_truncated, "Truncated frames due to MTU size");
956 }
957 
958 #undef ALE_SYSCTL_STAT_ADD32
959 #undef ALE_SYSCTL_STAT_ADD64
960 
961 struct ale_dmamap_arg {
962 	bus_addr_t	ale_busaddr;
963 };
964 
965 static void
ale_dmamap_cb(void * arg,bus_dma_segment_t * segs,int nsegs,int error)966 ale_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nsegs, int error)
967 {
968 	struct ale_dmamap_arg *ctx;
969 
970 	if (error != 0)
971 		return;
972 
973 	KASSERT(nsegs == 1, ("%s: %d segments returned!", __func__, nsegs));
974 
975 	ctx = (struct ale_dmamap_arg *)arg;
976 	ctx->ale_busaddr = segs[0].ds_addr;
977 }
978 
979 /*
980  * Tx descriptors/RXF0/CMB DMA blocks share ALE_DESC_ADDR_HI register
981  * which specifies high address region of DMA blocks. Therefore these
982  * blocks should have the same high address of given 4GB address
983  * space(i.e. crossing 4GB boundary is not allowed).
984  */
985 static int
ale_check_boundary(struct ale_softc * sc)986 ale_check_boundary(struct ale_softc *sc)
987 {
988 	bus_addr_t rx_cmb_end[ALE_RX_PAGES], tx_cmb_end;
989 	bus_addr_t rx_page_end[ALE_RX_PAGES], tx_ring_end;
990 
991 	rx_page_end[0] = sc->ale_cdata.ale_rx_page[0].page_paddr +
992 	    sc->ale_pagesize;
993 	rx_page_end[1] = sc->ale_cdata.ale_rx_page[1].page_paddr +
994 	    sc->ale_pagesize;
995 	tx_ring_end = sc->ale_cdata.ale_tx_ring_paddr + ALE_TX_RING_SZ;
996 	tx_cmb_end = sc->ale_cdata.ale_tx_cmb_paddr + ALE_TX_CMB_SZ;
997 	rx_cmb_end[0] = sc->ale_cdata.ale_rx_page[0].cmb_paddr + ALE_RX_CMB_SZ;
998 	rx_cmb_end[1] = sc->ale_cdata.ale_rx_page[1].cmb_paddr + ALE_RX_CMB_SZ;
999 
1000 	if ((ALE_ADDR_HI(tx_ring_end) !=
1001 	    ALE_ADDR_HI(sc->ale_cdata.ale_tx_ring_paddr)) ||
1002 	    (ALE_ADDR_HI(rx_page_end[0]) !=
1003 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[0].page_paddr)) ||
1004 	    (ALE_ADDR_HI(rx_page_end[1]) !=
1005 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[1].page_paddr)) ||
1006 	    (ALE_ADDR_HI(tx_cmb_end) !=
1007 	    ALE_ADDR_HI(sc->ale_cdata.ale_tx_cmb_paddr)) ||
1008 	    (ALE_ADDR_HI(rx_cmb_end[0]) !=
1009 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[0].cmb_paddr)) ||
1010 	    (ALE_ADDR_HI(rx_cmb_end[1]) !=
1011 	    ALE_ADDR_HI(sc->ale_cdata.ale_rx_page[1].cmb_paddr)))
1012 		return (EFBIG);
1013 
1014 	if ((ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_page_end[0])) ||
1015 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_page_end[1])) ||
1016 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_cmb_end[0])) ||
1017 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(rx_cmb_end[1])) ||
1018 	    (ALE_ADDR_HI(tx_ring_end) != ALE_ADDR_HI(tx_cmb_end)))
1019 		return (EFBIG);
1020 
1021 	return (0);
1022 }
1023 
1024 static int
ale_dma_alloc(struct ale_softc * sc)1025 ale_dma_alloc(struct ale_softc *sc)
1026 {
1027 	struct ale_txdesc *txd;
1028 	bus_addr_t lowaddr;
1029 	struct ale_dmamap_arg ctx;
1030 	int error, guard_size, i;
1031 
1032 	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0)
1033 		guard_size = ALE_JUMBO_FRAMELEN;
1034 	else
1035 		guard_size = ALE_MAX_FRAMELEN;
1036 	sc->ale_pagesize = roundup(guard_size + ALE_RX_PAGE_SZ,
1037 	    ALE_RX_PAGE_ALIGN);
1038 	lowaddr = BUS_SPACE_MAXADDR;
1039 again:
1040 	/* Create parent DMA tag. */
1041 	error = bus_dma_tag_create(
1042 	    bus_get_dma_tag(sc->ale_dev), /* parent */
1043 	    1, 0,			/* alignment, boundary */
1044 	    lowaddr,			/* lowaddr */
1045 	    BUS_SPACE_MAXADDR,		/* highaddr */
1046 	    NULL, NULL,			/* filter, filterarg */
1047 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsize */
1048 	    0,				/* nsegments */
1049 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsegsize */
1050 	    0,				/* flags */
1051 	    NULL, NULL,			/* lockfunc, lockarg */
1052 	    &sc->ale_cdata.ale_parent_tag);
1053 	if (error != 0) {
1054 		device_printf(sc->ale_dev,
1055 		    "could not create parent DMA tag.\n");
1056 		goto fail;
1057 	}
1058 
1059 	/* Create DMA tag for Tx descriptor ring. */
1060 	error = bus_dma_tag_create(
1061 	    sc->ale_cdata.ale_parent_tag, /* parent */
1062 	    ALE_TX_RING_ALIGN, 0,	/* alignment, boundary */
1063 	    BUS_SPACE_MAXADDR,		/* lowaddr */
1064 	    BUS_SPACE_MAXADDR,		/* highaddr */
1065 	    NULL, NULL,			/* filter, filterarg */
1066 	    ALE_TX_RING_SZ,		/* maxsize */
1067 	    1,				/* nsegments */
1068 	    ALE_TX_RING_SZ,		/* maxsegsize */
1069 	    0,				/* flags */
1070 	    NULL, NULL,			/* lockfunc, lockarg */
1071 	    &sc->ale_cdata.ale_tx_ring_tag);
1072 	if (error != 0) {
1073 		device_printf(sc->ale_dev,
1074 		    "could not create Tx ring DMA tag.\n");
1075 		goto fail;
1076 	}
1077 
1078 	/* Create DMA tag for Rx pages. */
1079 	for (i = 0; i < ALE_RX_PAGES; i++) {
1080 		error = bus_dma_tag_create(
1081 		    sc->ale_cdata.ale_parent_tag, /* parent */
1082 		    ALE_RX_PAGE_ALIGN, 0,	/* alignment, boundary */
1083 		    BUS_SPACE_MAXADDR,		/* lowaddr */
1084 		    BUS_SPACE_MAXADDR,		/* highaddr */
1085 		    NULL, NULL,			/* filter, filterarg */
1086 		    sc->ale_pagesize,		/* maxsize */
1087 		    1,				/* nsegments */
1088 		    sc->ale_pagesize,		/* maxsegsize */
1089 		    0,				/* flags */
1090 		    NULL, NULL,			/* lockfunc, lockarg */
1091 		    &sc->ale_cdata.ale_rx_page[i].page_tag);
1092 		if (error != 0) {
1093 			device_printf(sc->ale_dev,
1094 			    "could not create Rx page %d DMA tag.\n", i);
1095 			goto fail;
1096 		}
1097 	}
1098 
1099 	/* Create DMA tag for Tx coalescing message block. */
1100 	error = bus_dma_tag_create(
1101 	    sc->ale_cdata.ale_parent_tag, /* parent */
1102 	    ALE_CMB_ALIGN, 0,		/* alignment, boundary */
1103 	    BUS_SPACE_MAXADDR,		/* lowaddr */
1104 	    BUS_SPACE_MAXADDR,		/* highaddr */
1105 	    NULL, NULL,			/* filter, filterarg */
1106 	    ALE_TX_CMB_SZ,		/* maxsize */
1107 	    1,				/* nsegments */
1108 	    ALE_TX_CMB_SZ,		/* maxsegsize */
1109 	    0,				/* flags */
1110 	    NULL, NULL,			/* lockfunc, lockarg */
1111 	    &sc->ale_cdata.ale_tx_cmb_tag);
1112 	if (error != 0) {
1113 		device_printf(sc->ale_dev,
1114 		    "could not create Tx CMB DMA tag.\n");
1115 		goto fail;
1116 	}
1117 
1118 	/* Create DMA tag for Rx coalescing message block. */
1119 	for (i = 0; i < ALE_RX_PAGES; i++) {
1120 		error = bus_dma_tag_create(
1121 		    sc->ale_cdata.ale_parent_tag, /* parent */
1122 		    ALE_CMB_ALIGN, 0,		/* alignment, boundary */
1123 		    BUS_SPACE_MAXADDR,		/* lowaddr */
1124 		    BUS_SPACE_MAXADDR,		/* highaddr */
1125 		    NULL, NULL,			/* filter, filterarg */
1126 		    ALE_RX_CMB_SZ,		/* maxsize */
1127 		    1,				/* nsegments */
1128 		    ALE_RX_CMB_SZ,		/* maxsegsize */
1129 		    0,				/* flags */
1130 		    NULL, NULL,			/* lockfunc, lockarg */
1131 		    &sc->ale_cdata.ale_rx_page[i].cmb_tag);
1132 		if (error != 0) {
1133 			device_printf(sc->ale_dev,
1134 			    "could not create Rx page %d CMB DMA tag.\n", i);
1135 			goto fail;
1136 		}
1137 	}
1138 
1139 	/* Allocate DMA'able memory and load the DMA map for Tx ring. */
1140 	error = bus_dmamem_alloc(sc->ale_cdata.ale_tx_ring_tag,
1141 	    (void **)&sc->ale_cdata.ale_tx_ring,
1142 	    BUS_DMA_WAITOK | BUS_DMA_ZERO | BUS_DMA_COHERENT,
1143 	    &sc->ale_cdata.ale_tx_ring_map);
1144 	if (error != 0) {
1145 		device_printf(sc->ale_dev,
1146 		    "could not allocate DMA'able memory for Tx ring.\n");
1147 		goto fail;
1148 	}
1149 	ctx.ale_busaddr = 0;
1150 	error = bus_dmamap_load(sc->ale_cdata.ale_tx_ring_tag,
1151 	    sc->ale_cdata.ale_tx_ring_map, sc->ale_cdata.ale_tx_ring,
1152 	    ALE_TX_RING_SZ, ale_dmamap_cb, &ctx, 0);
1153 	if (error != 0 || ctx.ale_busaddr == 0) {
1154 		device_printf(sc->ale_dev,
1155 		    "could not load DMA'able memory for Tx ring.\n");
1156 		goto fail;
1157 	}
1158 	sc->ale_cdata.ale_tx_ring_paddr = ctx.ale_busaddr;
1159 
1160 	/* Rx pages. */
1161 	for (i = 0; i < ALE_RX_PAGES; i++) {
1162 		error = bus_dmamem_alloc(sc->ale_cdata.ale_rx_page[i].page_tag,
1163 		    (void **)&sc->ale_cdata.ale_rx_page[i].page_addr,
1164 		    BUS_DMA_WAITOK | BUS_DMA_ZERO | BUS_DMA_COHERENT,
1165 		    &sc->ale_cdata.ale_rx_page[i].page_map);
1166 		if (error != 0) {
1167 			device_printf(sc->ale_dev,
1168 			    "could not allocate DMA'able memory for "
1169 			    "Rx page %d.\n", i);
1170 			goto fail;
1171 		}
1172 		ctx.ale_busaddr = 0;
1173 		error = bus_dmamap_load(sc->ale_cdata.ale_rx_page[i].page_tag,
1174 		    sc->ale_cdata.ale_rx_page[i].page_map,
1175 		    sc->ale_cdata.ale_rx_page[i].page_addr,
1176 		    sc->ale_pagesize, ale_dmamap_cb, &ctx, 0);
1177 		if (error != 0 || ctx.ale_busaddr == 0) {
1178 			device_printf(sc->ale_dev,
1179 			    "could not load DMA'able memory for "
1180 			    "Rx page %d.\n", i);
1181 			goto fail;
1182 		}
1183 		sc->ale_cdata.ale_rx_page[i].page_paddr = ctx.ale_busaddr;
1184 	}
1185 
1186 	/* Tx CMB. */
1187 	error = bus_dmamem_alloc(sc->ale_cdata.ale_tx_cmb_tag,
1188 	    (void **)&sc->ale_cdata.ale_tx_cmb,
1189 	    BUS_DMA_WAITOK | BUS_DMA_ZERO | BUS_DMA_COHERENT,
1190 	    &sc->ale_cdata.ale_tx_cmb_map);
1191 	if (error != 0) {
1192 		device_printf(sc->ale_dev,
1193 		    "could not allocate DMA'able memory for Tx CMB.\n");
1194 		goto fail;
1195 	}
1196 	ctx.ale_busaddr = 0;
1197 	error = bus_dmamap_load(sc->ale_cdata.ale_tx_cmb_tag,
1198 	    sc->ale_cdata.ale_tx_cmb_map, sc->ale_cdata.ale_tx_cmb,
1199 	    ALE_TX_CMB_SZ, ale_dmamap_cb, &ctx, 0);
1200 	if (error != 0 || ctx.ale_busaddr == 0) {
1201 		device_printf(sc->ale_dev,
1202 		    "could not load DMA'able memory for Tx CMB.\n");
1203 		goto fail;
1204 	}
1205 	sc->ale_cdata.ale_tx_cmb_paddr = ctx.ale_busaddr;
1206 
1207 	/* Rx CMB. */
1208 	for (i = 0; i < ALE_RX_PAGES; i++) {
1209 		error = bus_dmamem_alloc(sc->ale_cdata.ale_rx_page[i].cmb_tag,
1210 		    (void **)&sc->ale_cdata.ale_rx_page[i].cmb_addr,
1211 		    BUS_DMA_WAITOK | BUS_DMA_ZERO | BUS_DMA_COHERENT,
1212 		    &sc->ale_cdata.ale_rx_page[i].cmb_map);
1213 		if (error != 0) {
1214 			device_printf(sc->ale_dev, "could not allocate "
1215 			    "DMA'able memory for Rx page %d CMB.\n", i);
1216 			goto fail;
1217 		}
1218 		ctx.ale_busaddr = 0;
1219 		error = bus_dmamap_load(sc->ale_cdata.ale_rx_page[i].cmb_tag,
1220 		    sc->ale_cdata.ale_rx_page[i].cmb_map,
1221 		    sc->ale_cdata.ale_rx_page[i].cmb_addr,
1222 		    ALE_RX_CMB_SZ, ale_dmamap_cb, &ctx, 0);
1223 		if (error != 0 || ctx.ale_busaddr == 0) {
1224 			device_printf(sc->ale_dev, "could not load DMA'able "
1225 			    "memory for Rx page %d CMB.\n", i);
1226 			goto fail;
1227 		}
1228 		sc->ale_cdata.ale_rx_page[i].cmb_paddr = ctx.ale_busaddr;
1229 	}
1230 
1231 	/*
1232 	 * Tx descriptors/RXF0/CMB DMA blocks share the same
1233 	 * high address region of 64bit DMA address space.
1234 	 */
1235 	if (lowaddr != BUS_SPACE_MAXADDR_32BIT &&
1236 	    (error = ale_check_boundary(sc)) != 0) {
1237 		device_printf(sc->ale_dev, "4GB boundary crossed, "
1238 		    "switching to 32bit DMA addressing mode.\n");
1239 		ale_dma_free(sc);
1240 		/*
1241 		 * Limit max allowable DMA address space to 32bit
1242 		 * and try again.
1243 		 */
1244 		lowaddr = BUS_SPACE_MAXADDR_32BIT;
1245 		goto again;
1246 	}
1247 
1248 	/*
1249 	 * Create Tx buffer parent tag.
1250 	 * AR81xx allows 64bit DMA addressing of Tx buffers so it
1251 	 * needs separate parent DMA tag as parent DMA address space
1252 	 * could be restricted to be within 32bit address space by
1253 	 * 4GB boundary crossing.
1254 	 */
1255 	error = bus_dma_tag_create(
1256 	    bus_get_dma_tag(sc->ale_dev), /* parent */
1257 	    1, 0,			/* alignment, boundary */
1258 	    BUS_SPACE_MAXADDR,		/* lowaddr */
1259 	    BUS_SPACE_MAXADDR,		/* highaddr */
1260 	    NULL, NULL,			/* filter, filterarg */
1261 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsize */
1262 	    0,				/* nsegments */
1263 	    BUS_SPACE_MAXSIZE_32BIT,	/* maxsegsize */
1264 	    0,				/* flags */
1265 	    NULL, NULL,			/* lockfunc, lockarg */
1266 	    &sc->ale_cdata.ale_buffer_tag);
1267 	if (error != 0) {
1268 		device_printf(sc->ale_dev,
1269 		    "could not create parent buffer DMA tag.\n");
1270 		goto fail;
1271 	}
1272 
1273 	/* Create DMA tag for Tx buffers. */
1274 	error = bus_dma_tag_create(
1275 	    sc->ale_cdata.ale_buffer_tag, /* parent */
1276 	    1, 0,			/* alignment, boundary */
1277 	    BUS_SPACE_MAXADDR,		/* lowaddr */
1278 	    BUS_SPACE_MAXADDR,		/* highaddr */
1279 	    NULL, NULL,			/* filter, filterarg */
1280 	    ALE_TSO_MAXSIZE,		/* maxsize */
1281 	    ALE_MAXTXSEGS,		/* nsegments */
1282 	    ALE_TSO_MAXSEGSIZE,		/* maxsegsize */
1283 	    0,				/* flags */
1284 	    NULL, NULL,			/* lockfunc, lockarg */
1285 	    &sc->ale_cdata.ale_tx_tag);
1286 	if (error != 0) {
1287 		device_printf(sc->ale_dev, "could not create Tx DMA tag.\n");
1288 		goto fail;
1289 	}
1290 
1291 	/* Create DMA maps for Tx buffers. */
1292 	for (i = 0; i < ALE_TX_RING_CNT; i++) {
1293 		txd = &sc->ale_cdata.ale_txdesc[i];
1294 		txd->tx_m = NULL;
1295 		txd->tx_dmamap = NULL;
1296 		error = bus_dmamap_create(sc->ale_cdata.ale_tx_tag, 0,
1297 		    &txd->tx_dmamap);
1298 		if (error != 0) {
1299 			device_printf(sc->ale_dev,
1300 			    "could not create Tx dmamap.\n");
1301 			goto fail;
1302 		}
1303 	}
1304 
1305 fail:
1306 	return (error);
1307 }
1308 
1309 static void
ale_dma_free(struct ale_softc * sc)1310 ale_dma_free(struct ale_softc *sc)
1311 {
1312 	struct ale_txdesc *txd;
1313 	int i;
1314 
1315 	/* Tx buffers. */
1316 	if (sc->ale_cdata.ale_tx_tag != NULL) {
1317 		for (i = 0; i < ALE_TX_RING_CNT; i++) {
1318 			txd = &sc->ale_cdata.ale_txdesc[i];
1319 			if (txd->tx_dmamap != NULL) {
1320 				bus_dmamap_destroy(sc->ale_cdata.ale_tx_tag,
1321 				    txd->tx_dmamap);
1322 				txd->tx_dmamap = NULL;
1323 			}
1324 		}
1325 		bus_dma_tag_destroy(sc->ale_cdata.ale_tx_tag);
1326 		sc->ale_cdata.ale_tx_tag = NULL;
1327 	}
1328 	/* Tx descriptor ring. */
1329 	if (sc->ale_cdata.ale_tx_ring_tag != NULL) {
1330 		if (sc->ale_cdata.ale_tx_ring_map != NULL)
1331 			bus_dmamap_unload(sc->ale_cdata.ale_tx_ring_tag,
1332 			    sc->ale_cdata.ale_tx_ring_map);
1333 		if (sc->ale_cdata.ale_tx_ring_map != NULL &&
1334 		    sc->ale_cdata.ale_tx_ring != NULL)
1335 			bus_dmamem_free(sc->ale_cdata.ale_tx_ring_tag,
1336 			    sc->ale_cdata.ale_tx_ring,
1337 			    sc->ale_cdata.ale_tx_ring_map);
1338 		sc->ale_cdata.ale_tx_ring = NULL;
1339 		sc->ale_cdata.ale_tx_ring_map = NULL;
1340 		bus_dma_tag_destroy(sc->ale_cdata.ale_tx_ring_tag);
1341 		sc->ale_cdata.ale_tx_ring_tag = NULL;
1342 	}
1343 	/* Rx page block. */
1344 	for (i = 0; i < ALE_RX_PAGES; i++) {
1345 		if (sc->ale_cdata.ale_rx_page[i].page_tag != NULL) {
1346 			if (sc->ale_cdata.ale_rx_page[i].page_map != NULL)
1347 				bus_dmamap_unload(
1348 				    sc->ale_cdata.ale_rx_page[i].page_tag,
1349 				    sc->ale_cdata.ale_rx_page[i].page_map);
1350 			if (sc->ale_cdata.ale_rx_page[i].page_map != NULL &&
1351 			    sc->ale_cdata.ale_rx_page[i].page_addr != NULL)
1352 				bus_dmamem_free(
1353 				    sc->ale_cdata.ale_rx_page[i].page_tag,
1354 				    sc->ale_cdata.ale_rx_page[i].page_addr,
1355 				    sc->ale_cdata.ale_rx_page[i].page_map);
1356 			sc->ale_cdata.ale_rx_page[i].page_addr = NULL;
1357 			sc->ale_cdata.ale_rx_page[i].page_map = NULL;
1358 			bus_dma_tag_destroy(
1359 			    sc->ale_cdata.ale_rx_page[i].page_tag);
1360 			sc->ale_cdata.ale_rx_page[i].page_tag = NULL;
1361 		}
1362 	}
1363 	/* Rx CMB. */
1364 	for (i = 0; i < ALE_RX_PAGES; i++) {
1365 		if (sc->ale_cdata.ale_rx_page[i].cmb_tag != NULL) {
1366 			if (sc->ale_cdata.ale_rx_page[i].cmb_map != NULL)
1367 				bus_dmamap_unload(
1368 				    sc->ale_cdata.ale_rx_page[i].cmb_tag,
1369 				    sc->ale_cdata.ale_rx_page[i].cmb_map);
1370 			if (sc->ale_cdata.ale_rx_page[i].cmb_map != NULL &&
1371 			    sc->ale_cdata.ale_rx_page[i].cmb_addr != NULL)
1372 				bus_dmamem_free(
1373 				    sc->ale_cdata.ale_rx_page[i].cmb_tag,
1374 				    sc->ale_cdata.ale_rx_page[i].cmb_addr,
1375 				    sc->ale_cdata.ale_rx_page[i].cmb_map);
1376 			sc->ale_cdata.ale_rx_page[i].cmb_addr = NULL;
1377 			sc->ale_cdata.ale_rx_page[i].cmb_map = NULL;
1378 			bus_dma_tag_destroy(
1379 			    sc->ale_cdata.ale_rx_page[i].cmb_tag);
1380 			sc->ale_cdata.ale_rx_page[i].cmb_tag = NULL;
1381 		}
1382 	}
1383 	/* Tx CMB. */
1384 	if (sc->ale_cdata.ale_tx_cmb_tag != NULL) {
1385 		if (sc->ale_cdata.ale_tx_cmb_map != NULL)
1386 			bus_dmamap_unload(sc->ale_cdata.ale_tx_cmb_tag,
1387 			    sc->ale_cdata.ale_tx_cmb_map);
1388 		if (sc->ale_cdata.ale_tx_cmb_map != NULL &&
1389 		    sc->ale_cdata.ale_tx_cmb != NULL)
1390 			bus_dmamem_free(sc->ale_cdata.ale_tx_cmb_tag,
1391 			    sc->ale_cdata.ale_tx_cmb,
1392 			    sc->ale_cdata.ale_tx_cmb_map);
1393 		sc->ale_cdata.ale_tx_cmb = NULL;
1394 		sc->ale_cdata.ale_tx_cmb_map = NULL;
1395 		bus_dma_tag_destroy(sc->ale_cdata.ale_tx_cmb_tag);
1396 		sc->ale_cdata.ale_tx_cmb_tag = NULL;
1397 	}
1398 	if (sc->ale_cdata.ale_buffer_tag != NULL) {
1399 		bus_dma_tag_destroy(sc->ale_cdata.ale_buffer_tag);
1400 		sc->ale_cdata.ale_buffer_tag = NULL;
1401 	}
1402 	if (sc->ale_cdata.ale_parent_tag != NULL) {
1403 		bus_dma_tag_destroy(sc->ale_cdata.ale_parent_tag);
1404 		sc->ale_cdata.ale_parent_tag = NULL;
1405 	}
1406 }
1407 
1408 static int
ale_shutdown(device_t dev)1409 ale_shutdown(device_t dev)
1410 {
1411 
1412 	return (ale_suspend(dev));
1413 }
1414 
1415 /*
1416  * Note, this driver resets the link speed to 10/100Mbps by
1417  * restarting auto-negotiation in suspend/shutdown phase but we
1418  * don't know whether that auto-negotiation would succeed or not
1419  * as driver has no control after powering off/suspend operation.
1420  * If the renegotiation fail WOL may not work. Running at 1Gbps
1421  * will draw more power than 375mA at 3.3V which is specified in
1422  * PCI specification and that would result in complete
1423  * shutdowning power to ethernet controller.
1424  *
1425  * TODO
1426  * Save current negotiated media speed/duplex/flow-control to
1427  * softc and restore the same link again after resuming. PHY
1428  * handling such as power down/resetting to 100Mbps may be better
1429  * handled in suspend method in phy driver.
1430  */
1431 static void
ale_setlinkspeed(struct ale_softc * sc)1432 ale_setlinkspeed(struct ale_softc *sc)
1433 {
1434 	struct mii_data *mii;
1435 	int aneg, i;
1436 
1437 	mii = device_get_softc(sc->ale_miibus);
1438 	mii_pollstat(mii);
1439 	aneg = 0;
1440 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
1441 	    (IFM_ACTIVE | IFM_AVALID)) {
1442 		switch IFM_SUBTYPE(mii->mii_media_active) {
1443 		case IFM_10_T:
1444 		case IFM_100_TX:
1445 			return;
1446 		case IFM_1000_T:
1447 			aneg++;
1448 			break;
1449 		default:
1450 			break;
1451 		}
1452 	}
1453 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr, MII_100T2CR, 0);
1454 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
1455 	    MII_ANAR, ANAR_TX_FD | ANAR_TX | ANAR_10_FD | ANAR_10 | ANAR_CSMA);
1456 	ale_miibus_writereg(sc->ale_dev, sc->ale_phyaddr,
1457 	    MII_BMCR, BMCR_RESET | BMCR_AUTOEN | BMCR_STARTNEG);
1458 	DELAY(1000);
1459 	if (aneg != 0) {
1460 		/*
1461 		 * Poll link state until ale(4) get a 10/100Mbps link.
1462 		 */
1463 		for (i = 0; i < MII_ANEGTICKS_GIGE; i++) {
1464 			mii_pollstat(mii);
1465 			if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID))
1466 			    == (IFM_ACTIVE | IFM_AVALID)) {
1467 				switch (IFM_SUBTYPE(
1468 				    mii->mii_media_active)) {
1469 				case IFM_10_T:
1470 				case IFM_100_TX:
1471 					ale_mac_config(sc);
1472 					return;
1473 				default:
1474 					break;
1475 				}
1476 			}
1477 			ALE_UNLOCK(sc);
1478 			pause("alelnk", hz);
1479 			ALE_LOCK(sc);
1480 		}
1481 		if (i == MII_ANEGTICKS_GIGE)
1482 			device_printf(sc->ale_dev,
1483 			    "establishing a link failed, WOL may not work!");
1484 	}
1485 	/*
1486 	 * No link, force MAC to have 100Mbps, full-duplex link.
1487 	 * This is the last resort and may/may not work.
1488 	 */
1489 	mii->mii_media_status = IFM_AVALID | IFM_ACTIVE;
1490 	mii->mii_media_active = IFM_ETHER | IFM_100_TX | IFM_FDX;
1491 	ale_mac_config(sc);
1492 }
1493 
1494 static void
ale_setwol(struct ale_softc * sc)1495 ale_setwol(struct ale_softc *sc)
1496 {
1497 	struct ifnet *ifp;
1498 	uint32_t reg, pmcs;
1499 	uint16_t pmstat;
1500 	int pmc;
1501 
1502 	ALE_LOCK_ASSERT(sc);
1503 
1504 	if (pci_find_cap(sc->ale_dev, PCIY_PMG, &pmc) != 0) {
1505 		/* Disable WOL. */
1506 		CSR_WRITE_4(sc, ALE_WOL_CFG, 0);
1507 		reg = CSR_READ_4(sc, ALE_PCIE_PHYMISC);
1508 		reg |= PCIE_PHYMISC_FORCE_RCV_DET;
1509 		CSR_WRITE_4(sc, ALE_PCIE_PHYMISC, reg);
1510 		/* Force PHY power down. */
1511 		CSR_WRITE_2(sc, ALE_GPHY_CTRL,
1512 		    GPHY_CTRL_EXT_RESET | GPHY_CTRL_HIB_EN |
1513 		    GPHY_CTRL_HIB_PULSE | GPHY_CTRL_PHY_PLL_ON |
1514 		    GPHY_CTRL_SEL_ANA_RESET | GPHY_CTRL_PHY_IDDQ |
1515 		    GPHY_CTRL_PCLK_SEL_DIS | GPHY_CTRL_PWDOWN_HW);
1516 		return;
1517 	}
1518 
1519 	ifp = sc->ale_ifp;
1520 	if ((ifp->if_capenable & IFCAP_WOL) != 0) {
1521 		if ((sc->ale_flags & ALE_FLAG_FASTETHER) == 0)
1522 			ale_setlinkspeed(sc);
1523 	}
1524 
1525 	pmcs = 0;
1526 	if ((ifp->if_capenable & IFCAP_WOL_MAGIC) != 0)
1527 		pmcs |= WOL_CFG_MAGIC | WOL_CFG_MAGIC_ENB;
1528 	CSR_WRITE_4(sc, ALE_WOL_CFG, pmcs);
1529 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
1530 	reg &= ~(MAC_CFG_DBG | MAC_CFG_PROMISC | MAC_CFG_ALLMULTI |
1531 	    MAC_CFG_BCAST);
1532 	if ((ifp->if_capenable & IFCAP_WOL_MCAST) != 0)
1533 		reg |= MAC_CFG_ALLMULTI | MAC_CFG_BCAST;
1534 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
1535 		reg |= MAC_CFG_RX_ENB;
1536 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
1537 
1538 	if ((ifp->if_capenable & IFCAP_WOL) == 0) {
1539 		/* WOL disabled, PHY power down. */
1540 		reg = CSR_READ_4(sc, ALE_PCIE_PHYMISC);
1541 		reg |= PCIE_PHYMISC_FORCE_RCV_DET;
1542 		CSR_WRITE_4(sc, ALE_PCIE_PHYMISC, reg);
1543 		CSR_WRITE_2(sc, ALE_GPHY_CTRL,
1544 		    GPHY_CTRL_EXT_RESET | GPHY_CTRL_HIB_EN |
1545 		    GPHY_CTRL_HIB_PULSE | GPHY_CTRL_SEL_ANA_RESET |
1546 		    GPHY_CTRL_PHY_IDDQ | GPHY_CTRL_PCLK_SEL_DIS |
1547 		    GPHY_CTRL_PWDOWN_HW);
1548 	}
1549 	/* Request PME. */
1550 	pmstat = pci_read_config(sc->ale_dev, pmc + PCIR_POWER_STATUS, 2);
1551 	pmstat &= ~(PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE);
1552 	if ((ifp->if_capenable & IFCAP_WOL) != 0)
1553 		pmstat |= PCIM_PSTAT_PME | PCIM_PSTAT_PMEENABLE;
1554 	pci_write_config(sc->ale_dev, pmc + PCIR_POWER_STATUS, pmstat, 2);
1555 }
1556 
1557 static int
ale_suspend(device_t dev)1558 ale_suspend(device_t dev)
1559 {
1560 	struct ale_softc *sc;
1561 
1562 	sc = device_get_softc(dev);
1563 
1564 	ALE_LOCK(sc);
1565 	ale_stop(sc);
1566 	ale_setwol(sc);
1567 	ALE_UNLOCK(sc);
1568 
1569 	return (0);
1570 }
1571 
1572 static int
ale_resume(device_t dev)1573 ale_resume(device_t dev)
1574 {
1575 	struct ale_softc *sc;
1576 	struct ifnet *ifp;
1577 	int pmc;
1578 	uint16_t pmstat;
1579 
1580 	sc = device_get_softc(dev);
1581 
1582 	ALE_LOCK(sc);
1583 	if (pci_find_cap(sc->ale_dev, PCIY_PMG, &pmc) == 0) {
1584 		/* Disable PME and clear PME status. */
1585 		pmstat = pci_read_config(sc->ale_dev,
1586 		    pmc + PCIR_POWER_STATUS, 2);
1587 		if ((pmstat & PCIM_PSTAT_PMEENABLE) != 0) {
1588 			pmstat &= ~PCIM_PSTAT_PMEENABLE;
1589 			pci_write_config(sc->ale_dev,
1590 			    pmc + PCIR_POWER_STATUS, pmstat, 2);
1591 		}
1592 	}
1593 	/* Reset PHY. */
1594 	ale_phy_reset(sc);
1595 	ifp = sc->ale_ifp;
1596 	if ((ifp->if_flags & IFF_UP) != 0) {
1597 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1598 		ale_init_locked(sc);
1599 	}
1600 	ALE_UNLOCK(sc);
1601 
1602 	return (0);
1603 }
1604 
1605 static int
ale_encap(struct ale_softc * sc,struct mbuf ** m_head)1606 ale_encap(struct ale_softc *sc, struct mbuf **m_head)
1607 {
1608 	struct ale_txdesc *txd, *txd_last;
1609 	struct tx_desc *desc;
1610 	struct mbuf *m;
1611 	struct ip *ip;
1612 	struct tcphdr *tcp;
1613 	bus_dma_segment_t txsegs[ALE_MAXTXSEGS];
1614 	bus_dmamap_t map;
1615 	uint32_t cflags, hdrlen, ip_off, poff, vtag;
1616 	int error, i, nsegs, prod, si;
1617 
1618 	ALE_LOCK_ASSERT(sc);
1619 
1620 	M_ASSERTPKTHDR((*m_head));
1621 
1622 	m = *m_head;
1623 	ip = NULL;
1624 	tcp = NULL;
1625 	cflags = vtag = 0;
1626 	ip_off = poff = 0;
1627 	if ((m->m_pkthdr.csum_flags & (ALE_CSUM_FEATURES | CSUM_TSO)) != 0) {
1628 		/*
1629 		 * AR81xx requires offset of TCP/UDP payload in its Tx
1630 		 * descriptor to perform hardware Tx checksum offload.
1631 		 * Additionally, TSO requires IP/TCP header size and
1632 		 * modification of IP/TCP header in order to make TSO
1633 		 * engine work. This kind of operation takes many CPU
1634 		 * cycles on FreeBSD so fast host CPU is required to
1635 		 * get smooth TSO performance.
1636 		 */
1637 		struct ether_header *eh;
1638 
1639 		if (M_WRITABLE(m) == 0) {
1640 			/* Get a writable copy. */
1641 			m = m_dup(*m_head, M_NOWAIT);
1642 			/* Release original mbufs. */
1643 			m_freem(*m_head);
1644 			if (m == NULL) {
1645 				*m_head = NULL;
1646 				return (ENOBUFS);
1647 			}
1648 			*m_head = m;
1649 		}
1650 
1651 		/*
1652 		 * Buggy-controller requires 4 byte aligned Tx buffer
1653 		 * to make custom checksum offload work.
1654 		 */
1655 		if ((sc->ale_flags & ALE_FLAG_TXCSUM_BUG) != 0 &&
1656 		    (m->m_pkthdr.csum_flags & ALE_CSUM_FEATURES) != 0 &&
1657 		    (mtod(m, intptr_t) & 3) != 0) {
1658 			m = m_defrag(*m_head, M_NOWAIT);
1659 			if (m == NULL) {
1660 				m_freem(*m_head);
1661 				*m_head = NULL;
1662 				return (ENOBUFS);
1663 			}
1664 			*m_head = m;
1665 		}
1666 
1667 		ip_off = sizeof(struct ether_header);
1668 		m = m_pullup(m, ip_off);
1669 		if (m == NULL) {
1670 			*m_head = NULL;
1671 			return (ENOBUFS);
1672 		}
1673 		eh = mtod(m, struct ether_header *);
1674 		/*
1675 		 * Check if hardware VLAN insertion is off.
1676 		 * Additional check for LLC/SNAP frame?
1677 		 */
1678 		if (eh->ether_type == htons(ETHERTYPE_VLAN)) {
1679 			ip_off = sizeof(struct ether_vlan_header);
1680 			m = m_pullup(m, ip_off);
1681 			if (m == NULL) {
1682 				*m_head = NULL;
1683 				return (ENOBUFS);
1684 			}
1685 		}
1686 		m = m_pullup(m, ip_off + sizeof(struct ip));
1687 		if (m == NULL) {
1688 			*m_head = NULL;
1689 			return (ENOBUFS);
1690 		}
1691 		ip = (struct ip *)(mtod(m, char *) + ip_off);
1692 		poff = ip_off + (ip->ip_hl << 2);
1693 		if ((m->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
1694 			/*
1695 			 * XXX
1696 			 * AR81xx requires the first descriptor should
1697 			 * not include any TCP playload for TSO case.
1698 			 * (i.e. ethernet header + IP + TCP header only)
1699 			 * m_pullup(9) above will ensure this too.
1700 			 * However it's not correct if the first mbuf
1701 			 * of the chain does not use cluster.
1702 			 */
1703 			m = m_pullup(m, poff + sizeof(struct tcphdr));
1704 			if (m == NULL) {
1705 				*m_head = NULL;
1706 				return (ENOBUFS);
1707 			}
1708 			ip = (struct ip *)(mtod(m, char *) + ip_off);
1709 			tcp = (struct tcphdr *)(mtod(m, char *) + poff);
1710 			m = m_pullup(m, poff + (tcp->th_off << 2));
1711 			if (m == NULL) {
1712 				*m_head = NULL;
1713 				return (ENOBUFS);
1714 			}
1715 			/*
1716 			 * AR81xx requires IP/TCP header size and offset as
1717 			 * well as TCP pseudo checksum which complicates
1718 			 * TSO configuration. I guess this comes from the
1719 			 * adherence to Microsoft NDIS Large Send
1720 			 * specification which requires insertion of
1721 			 * pseudo checksum by upper stack. The pseudo
1722 			 * checksum that NDIS refers to doesn't include
1723 			 * TCP payload length so ale(4) should recompute
1724 			 * the pseudo checksum here. Hopefully this wouldn't
1725 			 * be much burden on modern CPUs.
1726 			 * Reset IP checksum and recompute TCP pseudo
1727 			 * checksum as NDIS specification said.
1728 			 */
1729 			ip->ip_sum = 0;
1730 			tcp->th_sum = in_pseudo(ip->ip_src.s_addr,
1731 			    ip->ip_dst.s_addr, htons(IPPROTO_TCP));
1732 		}
1733 		*m_head = m;
1734 	}
1735 
1736 	si = prod = sc->ale_cdata.ale_tx_prod;
1737 	txd = &sc->ale_cdata.ale_txdesc[prod];
1738 	txd_last = txd;
1739 	map = txd->tx_dmamap;
1740 
1741 	error =  bus_dmamap_load_mbuf_sg(sc->ale_cdata.ale_tx_tag, map,
1742 	    *m_head, txsegs, &nsegs, 0);
1743 	if (error == EFBIG) {
1744 		m = m_collapse(*m_head, M_NOWAIT, ALE_MAXTXSEGS);
1745 		if (m == NULL) {
1746 			m_freem(*m_head);
1747 			*m_head = NULL;
1748 			return (ENOMEM);
1749 		}
1750 		*m_head = m;
1751 		error = bus_dmamap_load_mbuf_sg(sc->ale_cdata.ale_tx_tag, map,
1752 		    *m_head, txsegs, &nsegs, 0);
1753 		if (error != 0) {
1754 			m_freem(*m_head);
1755 			*m_head = NULL;
1756 			return (error);
1757 		}
1758 	} else if (error != 0)
1759 		return (error);
1760 	if (nsegs == 0) {
1761 		m_freem(*m_head);
1762 		*m_head = NULL;
1763 		return (EIO);
1764 	}
1765 
1766 	/* Check descriptor overrun. */
1767 	if (sc->ale_cdata.ale_tx_cnt + nsegs >= ALE_TX_RING_CNT - 3) {
1768 		bus_dmamap_unload(sc->ale_cdata.ale_tx_tag, map);
1769 		return (ENOBUFS);
1770 	}
1771 	bus_dmamap_sync(sc->ale_cdata.ale_tx_tag, map, BUS_DMASYNC_PREWRITE);
1772 
1773 	m = *m_head;
1774 	if ((m->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
1775 		/* Request TSO and set MSS. */
1776 		cflags |= ALE_TD_TSO;
1777 		cflags |= ((uint32_t)m->m_pkthdr.tso_segsz << ALE_TD_MSS_SHIFT);
1778 		/* Set IP/TCP header size. */
1779 		cflags |= ip->ip_hl << ALE_TD_IPHDR_LEN_SHIFT;
1780 		cflags |= tcp->th_off << ALE_TD_TCPHDR_LEN_SHIFT;
1781 	} else if ((m->m_pkthdr.csum_flags & ALE_CSUM_FEATURES) != 0) {
1782 		/*
1783 		 * AR81xx supports Tx custom checksum offload feature
1784 		 * that offloads single 16bit checksum computation.
1785 		 * So you can choose one among IP, TCP and UDP.
1786 		 * Normally driver sets checksum start/insertion
1787 		 * position from the information of TCP/UDP frame as
1788 		 * TCP/UDP checksum takes more time than that of IP.
1789 		 * However it seems that custom checksum offload
1790 		 * requires 4 bytes aligned Tx buffers due to hardware
1791 		 * bug.
1792 		 * AR81xx also supports explicit Tx checksum computation
1793 		 * if it is told that the size of IP header and TCP
1794 		 * header(for UDP, the header size does not matter
1795 		 * because it's fixed length). However with this scheme
1796 		 * TSO does not work so you have to choose one either
1797 		 * TSO or explicit Tx checksum offload. I chosen TSO
1798 		 * plus custom checksum offload with work-around which
1799 		 * will cover most common usage for this consumer
1800 		 * ethernet controller. The work-around takes a lot of
1801 		 * CPU cycles if Tx buffer is not aligned on 4 bytes
1802 		 * boundary, though.
1803 		 */
1804 		cflags |= ALE_TD_CXSUM;
1805 		/* Set checksum start offset. */
1806 		cflags |= (poff << ALE_TD_CSUM_PLOADOFFSET_SHIFT);
1807 		/* Set checksum insertion position of TCP/UDP. */
1808 		cflags |= ((poff + m->m_pkthdr.csum_data) <<
1809 		    ALE_TD_CSUM_XSUMOFFSET_SHIFT);
1810 	}
1811 
1812 	/* Configure VLAN hardware tag insertion. */
1813 	if ((m->m_flags & M_VLANTAG) != 0) {
1814 		vtag = ALE_TX_VLAN_TAG(m->m_pkthdr.ether_vtag);
1815 		vtag = ((vtag << ALE_TD_VLAN_SHIFT) & ALE_TD_VLAN_MASK);
1816 		cflags |= ALE_TD_INSERT_VLAN_TAG;
1817 	}
1818 
1819 	i = 0;
1820 	if ((m->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
1821 		/*
1822 		 * Make sure the first fragment contains
1823 		 * only ethernet and IP/TCP header with options.
1824 		 */
1825 		hdrlen =  poff + (tcp->th_off << 2);
1826 		desc = &sc->ale_cdata.ale_tx_ring[prod];
1827 		desc->addr = htole64(txsegs[i].ds_addr);
1828 		desc->len = htole32(ALE_TX_BYTES(hdrlen) | vtag);
1829 		desc->flags = htole32(cflags);
1830 		sc->ale_cdata.ale_tx_cnt++;
1831 		ALE_DESC_INC(prod, ALE_TX_RING_CNT);
1832 		if (m->m_len - hdrlen > 0) {
1833 			/* Handle remaining payload of the first fragment. */
1834 			desc = &sc->ale_cdata.ale_tx_ring[prod];
1835 			desc->addr = htole64(txsegs[i].ds_addr + hdrlen);
1836 			desc->len = htole32(ALE_TX_BYTES(m->m_len - hdrlen) |
1837 			    vtag);
1838 			desc->flags = htole32(cflags);
1839 			sc->ale_cdata.ale_tx_cnt++;
1840 			ALE_DESC_INC(prod, ALE_TX_RING_CNT);
1841 		}
1842 		i = 1;
1843 	}
1844 	for (; i < nsegs; i++) {
1845 		desc = &sc->ale_cdata.ale_tx_ring[prod];
1846 		desc->addr = htole64(txsegs[i].ds_addr);
1847 		desc->len = htole32(ALE_TX_BYTES(txsegs[i].ds_len) | vtag);
1848 		desc->flags = htole32(cflags);
1849 		sc->ale_cdata.ale_tx_cnt++;
1850 		ALE_DESC_INC(prod, ALE_TX_RING_CNT);
1851 	}
1852 	/* Update producer index. */
1853 	sc->ale_cdata.ale_tx_prod = prod;
1854 	/* Set TSO header on the first descriptor. */
1855 	if ((m->m_pkthdr.csum_flags & CSUM_TSO) != 0) {
1856 		desc = &sc->ale_cdata.ale_tx_ring[si];
1857 		desc->flags |= htole32(ALE_TD_TSO_HDR);
1858 	}
1859 
1860 	/* Finally set EOP on the last descriptor. */
1861 	prod = (prod + ALE_TX_RING_CNT - 1) % ALE_TX_RING_CNT;
1862 	desc = &sc->ale_cdata.ale_tx_ring[prod];
1863 	desc->flags |= htole32(ALE_TD_EOP);
1864 
1865 	/* Swap dmamap of the first and the last. */
1866 	txd = &sc->ale_cdata.ale_txdesc[prod];
1867 	map = txd_last->tx_dmamap;
1868 	txd_last->tx_dmamap = txd->tx_dmamap;
1869 	txd->tx_dmamap = map;
1870 	txd->tx_m = m;
1871 
1872 	/* Sync descriptors. */
1873 	bus_dmamap_sync(sc->ale_cdata.ale_tx_ring_tag,
1874 	    sc->ale_cdata.ale_tx_ring_map,
1875 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1876 
1877 	return (0);
1878 }
1879 
1880 static void
ale_start(struct ifnet * ifp)1881 ale_start(struct ifnet *ifp)
1882 {
1883         struct ale_softc *sc;
1884 
1885 	sc = ifp->if_softc;
1886 	ALE_LOCK(sc);
1887 	ale_start_locked(ifp);
1888 	ALE_UNLOCK(sc);
1889 }
1890 
1891 static void
ale_start_locked(struct ifnet * ifp)1892 ale_start_locked(struct ifnet *ifp)
1893 {
1894         struct ale_softc *sc;
1895         struct mbuf *m_head;
1896 	int enq;
1897 
1898 	sc = ifp->if_softc;
1899 
1900 	ALE_LOCK_ASSERT(sc);
1901 
1902 	/* Reclaim transmitted frames. */
1903 	if (sc->ale_cdata.ale_tx_cnt >= ALE_TX_DESC_HIWAT)
1904 		ale_txeof(sc);
1905 
1906 	if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1907 	    IFF_DRV_RUNNING || (sc->ale_flags & ALE_FLAG_LINK) == 0)
1908 		return;
1909 
1910 	for (enq = 0; !IFQ_DRV_IS_EMPTY(&ifp->if_snd); ) {
1911 		IFQ_DRV_DEQUEUE(&ifp->if_snd, m_head);
1912 		if (m_head == NULL)
1913 			break;
1914 		/*
1915 		 * Pack the data into the transmit ring. If we
1916 		 * don't have room, set the OACTIVE flag and wait
1917 		 * for the NIC to drain the ring.
1918 		 */
1919 		if (ale_encap(sc, &m_head)) {
1920 			if (m_head == NULL)
1921 				break;
1922 			IFQ_DRV_PREPEND(&ifp->if_snd, m_head);
1923 			ifp->if_drv_flags |= IFF_DRV_OACTIVE;
1924 			break;
1925 		}
1926 
1927 		enq++;
1928 		/*
1929 		 * If there's a BPF listener, bounce a copy of this frame
1930 		 * to him.
1931 		 */
1932 		ETHER_BPF_MTAP(ifp, m_head);
1933 	}
1934 
1935 	if (enq > 0) {
1936 		/* Kick. */
1937 		CSR_WRITE_4(sc, ALE_MBOX_TPD_PROD_IDX,
1938 		    sc->ale_cdata.ale_tx_prod);
1939 		/* Set a timeout in case the chip goes out to lunch. */
1940 		sc->ale_watchdog_timer = ALE_TX_TIMEOUT;
1941 	}
1942 }
1943 
1944 static void
ale_watchdog(struct ale_softc * sc)1945 ale_watchdog(struct ale_softc *sc)
1946 {
1947 	struct ifnet *ifp;
1948 
1949 	ALE_LOCK_ASSERT(sc);
1950 
1951 	if (sc->ale_watchdog_timer == 0 || --sc->ale_watchdog_timer)
1952 		return;
1953 
1954 	ifp = sc->ale_ifp;
1955 	if ((sc->ale_flags & ALE_FLAG_LINK) == 0) {
1956 		if_printf(sc->ale_ifp, "watchdog timeout (lost link)\n");
1957 		ifp->if_oerrors++;
1958 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1959 		ale_init_locked(sc);
1960 		return;
1961 	}
1962 	if_printf(sc->ale_ifp, "watchdog timeout -- resetting\n");
1963 	ifp->if_oerrors++;
1964 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1965 	ale_init_locked(sc);
1966 	if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
1967 		ale_start_locked(ifp);
1968 }
1969 
1970 static int
ale_ioctl(struct ifnet * ifp,u_long cmd,caddr_t data)1971 ale_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1972 {
1973 	struct ale_softc *sc;
1974 	struct ifreq *ifr;
1975 	struct mii_data *mii;
1976 	int error, mask;
1977 
1978 	sc = ifp->if_softc;
1979 	ifr = (struct ifreq *)data;
1980 	error = 0;
1981 	switch (cmd) {
1982 	case SIOCSIFMTU:
1983 		if (ifr->ifr_mtu < ETHERMIN || ifr->ifr_mtu > ALE_JUMBO_MTU ||
1984 		    ((sc->ale_flags & ALE_FLAG_JUMBO) == 0 &&
1985 		    ifr->ifr_mtu > ETHERMTU))
1986 			error = EINVAL;
1987 		else if (ifp->if_mtu != ifr->ifr_mtu) {
1988 			ALE_LOCK(sc);
1989 			ifp->if_mtu = ifr->ifr_mtu;
1990 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
1991 				ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1992 				ale_init_locked(sc);
1993 			}
1994 			ALE_UNLOCK(sc);
1995 		}
1996 		break;
1997 	case SIOCSIFFLAGS:
1998 		ALE_LOCK(sc);
1999 		if ((ifp->if_flags & IFF_UP) != 0) {
2000 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2001 				if (((ifp->if_flags ^ sc->ale_if_flags)
2002 				    & (IFF_PROMISC | IFF_ALLMULTI)) != 0)
2003 					ale_rxfilter(sc);
2004 			} else {
2005 				ale_init_locked(sc);
2006 			}
2007 		} else {
2008 			if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2009 				ale_stop(sc);
2010 		}
2011 		sc->ale_if_flags = ifp->if_flags;
2012 		ALE_UNLOCK(sc);
2013 		break;
2014 	case SIOCADDMULTI:
2015 	case SIOCDELMULTI:
2016 		ALE_LOCK(sc);
2017 		if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2018 			ale_rxfilter(sc);
2019 		ALE_UNLOCK(sc);
2020 		break;
2021 	case SIOCSIFMEDIA:
2022 	case SIOCGIFMEDIA:
2023 		mii = device_get_softc(sc->ale_miibus);
2024 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, cmd);
2025 		break;
2026 	case SIOCSIFCAP:
2027 		ALE_LOCK(sc);
2028 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
2029 		if ((mask & IFCAP_TXCSUM) != 0 &&
2030 		    (ifp->if_capabilities & IFCAP_TXCSUM) != 0) {
2031 			ifp->if_capenable ^= IFCAP_TXCSUM;
2032 			if ((ifp->if_capenable & IFCAP_TXCSUM) != 0)
2033 				ifp->if_hwassist |= ALE_CSUM_FEATURES;
2034 			else
2035 				ifp->if_hwassist &= ~ALE_CSUM_FEATURES;
2036 		}
2037 		if ((mask & IFCAP_RXCSUM) != 0 &&
2038 		    (ifp->if_capabilities & IFCAP_RXCSUM) != 0)
2039 			ifp->if_capenable ^= IFCAP_RXCSUM;
2040 		if ((mask & IFCAP_TSO4) != 0 &&
2041 		    (ifp->if_capabilities & IFCAP_TSO4) != 0) {
2042 			ifp->if_capenable ^= IFCAP_TSO4;
2043 			if ((ifp->if_capenable & IFCAP_TSO4) != 0)
2044 				ifp->if_hwassist |= CSUM_TSO;
2045 			else
2046 				ifp->if_hwassist &= ~CSUM_TSO;
2047 		}
2048 
2049 		if ((mask & IFCAP_WOL_MCAST) != 0 &&
2050 		    (ifp->if_capabilities & IFCAP_WOL_MCAST) != 0)
2051 			ifp->if_capenable ^= IFCAP_WOL_MCAST;
2052 		if ((mask & IFCAP_WOL_MAGIC) != 0 &&
2053 		    (ifp->if_capabilities & IFCAP_WOL_MAGIC) != 0)
2054 			ifp->if_capenable ^= IFCAP_WOL_MAGIC;
2055 		if ((mask & IFCAP_VLAN_HWCSUM) != 0 &&
2056 		    (ifp->if_capabilities & IFCAP_VLAN_HWCSUM) != 0)
2057 			ifp->if_capenable ^= IFCAP_VLAN_HWCSUM;
2058 		if ((mask & IFCAP_VLAN_HWTSO) != 0 &&
2059 		    (ifp->if_capabilities & IFCAP_VLAN_HWTSO) != 0)
2060 			ifp->if_capenable ^= IFCAP_VLAN_HWTSO;
2061 		if ((mask & IFCAP_VLAN_HWTAGGING) != 0 &&
2062 		    (ifp->if_capabilities & IFCAP_VLAN_HWTAGGING) != 0) {
2063 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
2064 			if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0)
2065 				ifp->if_capenable &= ~IFCAP_VLAN_HWTSO;
2066 			ale_rxvlan(sc);
2067 		}
2068 		ALE_UNLOCK(sc);
2069 		VLAN_CAPABILITIES(ifp);
2070 		break;
2071 	default:
2072 		error = ether_ioctl(ifp, cmd, data);
2073 		break;
2074 	}
2075 
2076 	return (error);
2077 }
2078 
2079 static void
ale_mac_config(struct ale_softc * sc)2080 ale_mac_config(struct ale_softc *sc)
2081 {
2082 	struct mii_data *mii;
2083 	uint32_t reg;
2084 
2085 	ALE_LOCK_ASSERT(sc);
2086 
2087 	mii = device_get_softc(sc->ale_miibus);
2088 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
2089 	reg &= ~(MAC_CFG_FULL_DUPLEX | MAC_CFG_TX_FC | MAC_CFG_RX_FC |
2090 	    MAC_CFG_SPEED_MASK);
2091 	/* Reprogram MAC with resolved speed/duplex. */
2092 	switch (IFM_SUBTYPE(mii->mii_media_active)) {
2093 	case IFM_10_T:
2094 	case IFM_100_TX:
2095 		reg |= MAC_CFG_SPEED_10_100;
2096 		break;
2097 	case IFM_1000_T:
2098 		reg |= MAC_CFG_SPEED_1000;
2099 		break;
2100 	}
2101 	if ((IFM_OPTIONS(mii->mii_media_active) & IFM_FDX) != 0) {
2102 		reg |= MAC_CFG_FULL_DUPLEX;
2103 		if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_TXPAUSE) != 0)
2104 			reg |= MAC_CFG_TX_FC;
2105 		if ((IFM_OPTIONS(mii->mii_media_active) & IFM_ETH_RXPAUSE) != 0)
2106 			reg |= MAC_CFG_RX_FC;
2107 	}
2108 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2109 }
2110 
2111 static void
ale_stats_clear(struct ale_softc * sc)2112 ale_stats_clear(struct ale_softc *sc)
2113 {
2114 	struct smb sb;
2115 	uint32_t *reg;
2116 	int i;
2117 
2118 	for (reg = &sb.rx_frames, i = 0; reg <= &sb.rx_pkts_filtered; reg++) {
2119 		CSR_READ_4(sc, ALE_RX_MIB_BASE + i);
2120 		i += sizeof(uint32_t);
2121 	}
2122 	/* Read Tx statistics. */
2123 	for (reg = &sb.tx_frames, i = 0; reg <= &sb.tx_mcast_bytes; reg++) {
2124 		CSR_READ_4(sc, ALE_TX_MIB_BASE + i);
2125 		i += sizeof(uint32_t);
2126 	}
2127 }
2128 
2129 static void
ale_stats_update(struct ale_softc * sc)2130 ale_stats_update(struct ale_softc *sc)
2131 {
2132 	struct ale_hw_stats *stat;
2133 	struct smb sb, *smb;
2134 	struct ifnet *ifp;
2135 	uint32_t *reg;
2136 	int i;
2137 
2138 	ALE_LOCK_ASSERT(sc);
2139 
2140 	ifp = sc->ale_ifp;
2141 	stat = &sc->ale_stats;
2142 	smb = &sb;
2143 
2144 	/* Read Rx statistics. */
2145 	for (reg = &sb.rx_frames, i = 0; reg <= &sb.rx_pkts_filtered; reg++) {
2146 		*reg = CSR_READ_4(sc, ALE_RX_MIB_BASE + i);
2147 		i += sizeof(uint32_t);
2148 	}
2149 	/* Read Tx statistics. */
2150 	for (reg = &sb.tx_frames, i = 0; reg <= &sb.tx_mcast_bytes; reg++) {
2151 		*reg = CSR_READ_4(sc, ALE_TX_MIB_BASE + i);
2152 		i += sizeof(uint32_t);
2153 	}
2154 
2155 	/* Rx stats. */
2156 	stat->rx_frames += smb->rx_frames;
2157 	stat->rx_bcast_frames += smb->rx_bcast_frames;
2158 	stat->rx_mcast_frames += smb->rx_mcast_frames;
2159 	stat->rx_pause_frames += smb->rx_pause_frames;
2160 	stat->rx_control_frames += smb->rx_control_frames;
2161 	stat->rx_crcerrs += smb->rx_crcerrs;
2162 	stat->rx_lenerrs += smb->rx_lenerrs;
2163 	stat->rx_bytes += smb->rx_bytes;
2164 	stat->rx_runts += smb->rx_runts;
2165 	stat->rx_fragments += smb->rx_fragments;
2166 	stat->rx_pkts_64 += smb->rx_pkts_64;
2167 	stat->rx_pkts_65_127 += smb->rx_pkts_65_127;
2168 	stat->rx_pkts_128_255 += smb->rx_pkts_128_255;
2169 	stat->rx_pkts_256_511 += smb->rx_pkts_256_511;
2170 	stat->rx_pkts_512_1023 += smb->rx_pkts_512_1023;
2171 	stat->rx_pkts_1024_1518 += smb->rx_pkts_1024_1518;
2172 	stat->rx_pkts_1519_max += smb->rx_pkts_1519_max;
2173 	stat->rx_pkts_truncated += smb->rx_pkts_truncated;
2174 	stat->rx_fifo_oflows += smb->rx_fifo_oflows;
2175 	stat->rx_rrs_errs += smb->rx_rrs_errs;
2176 	stat->rx_alignerrs += smb->rx_alignerrs;
2177 	stat->rx_bcast_bytes += smb->rx_bcast_bytes;
2178 	stat->rx_mcast_bytes += smb->rx_mcast_bytes;
2179 	stat->rx_pkts_filtered += smb->rx_pkts_filtered;
2180 
2181 	/* Tx stats. */
2182 	stat->tx_frames += smb->tx_frames;
2183 	stat->tx_bcast_frames += smb->tx_bcast_frames;
2184 	stat->tx_mcast_frames += smb->tx_mcast_frames;
2185 	stat->tx_pause_frames += smb->tx_pause_frames;
2186 	stat->tx_excess_defer += smb->tx_excess_defer;
2187 	stat->tx_control_frames += smb->tx_control_frames;
2188 	stat->tx_deferred += smb->tx_deferred;
2189 	stat->tx_bytes += smb->tx_bytes;
2190 	stat->tx_pkts_64 += smb->tx_pkts_64;
2191 	stat->tx_pkts_65_127 += smb->tx_pkts_65_127;
2192 	stat->tx_pkts_128_255 += smb->tx_pkts_128_255;
2193 	stat->tx_pkts_256_511 += smb->tx_pkts_256_511;
2194 	stat->tx_pkts_512_1023 += smb->tx_pkts_512_1023;
2195 	stat->tx_pkts_1024_1518 += smb->tx_pkts_1024_1518;
2196 	stat->tx_pkts_1519_max += smb->tx_pkts_1519_max;
2197 	stat->tx_single_colls += smb->tx_single_colls;
2198 	stat->tx_multi_colls += smb->tx_multi_colls;
2199 	stat->tx_late_colls += smb->tx_late_colls;
2200 	stat->tx_excess_colls += smb->tx_excess_colls;
2201 	stat->tx_underrun += smb->tx_underrun;
2202 	stat->tx_desc_underrun += smb->tx_desc_underrun;
2203 	stat->tx_lenerrs += smb->tx_lenerrs;
2204 	stat->tx_pkts_truncated += smb->tx_pkts_truncated;
2205 	stat->tx_bcast_bytes += smb->tx_bcast_bytes;
2206 	stat->tx_mcast_bytes += smb->tx_mcast_bytes;
2207 
2208 	/* Update counters in ifnet. */
2209 	ifp->if_opackets += smb->tx_frames;
2210 
2211 	ifp->if_collisions += smb->tx_single_colls +
2212 	    smb->tx_multi_colls * 2 + smb->tx_late_colls +
2213 	    smb->tx_excess_colls * HDPX_CFG_RETRY_DEFAULT;
2214 
2215 	ifp->if_oerrors += smb->tx_late_colls + smb->tx_excess_colls +
2216 	    smb->tx_underrun + smb->tx_pkts_truncated;
2217 
2218 	ifp->if_ipackets += smb->rx_frames;
2219 
2220 	ifp->if_ierrors += smb->rx_crcerrs + smb->rx_lenerrs +
2221 	    smb->rx_runts + smb->rx_pkts_truncated +
2222 	    smb->rx_fifo_oflows + smb->rx_rrs_errs +
2223 	    smb->rx_alignerrs;
2224 }
2225 
2226 static int
ale_intr(void * arg)2227 ale_intr(void *arg)
2228 {
2229 	struct ale_softc *sc;
2230 	uint32_t status;
2231 
2232 	sc = (struct ale_softc *)arg;
2233 
2234 	status = CSR_READ_4(sc, ALE_INTR_STATUS);
2235 	if ((status & ALE_INTRS) == 0)
2236 		return (FILTER_STRAY);
2237 	/* Disable interrupts. */
2238 	CSR_WRITE_4(sc, ALE_INTR_STATUS, INTR_DIS_INT);
2239 	taskqueue_enqueue(sc->ale_tq, &sc->ale_int_task);
2240 
2241 	return (FILTER_HANDLED);
2242 }
2243 
2244 static void
ale_int_task(void * arg,int pending)2245 ale_int_task(void *arg, int pending)
2246 {
2247 	struct ale_softc *sc;
2248 	struct ifnet *ifp;
2249 	uint32_t status;
2250 	int more;
2251 
2252 	sc = (struct ale_softc *)arg;
2253 
2254 	status = CSR_READ_4(sc, ALE_INTR_STATUS);
2255 	ALE_LOCK(sc);
2256 	if (sc->ale_morework != 0)
2257 		status |= INTR_RX_PKT;
2258 	if ((status & ALE_INTRS) == 0)
2259 		goto done;
2260 
2261 	/* Acknowledge interrupts but still disable interrupts. */
2262 	CSR_WRITE_4(sc, ALE_INTR_STATUS, status | INTR_DIS_INT);
2263 
2264 	ifp = sc->ale_ifp;
2265 	more = 0;
2266 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0) {
2267 		more = ale_rxeof(sc, sc->ale_process_limit);
2268 		if (more == EAGAIN)
2269 			sc->ale_morework = 1;
2270 		else if (more == EIO) {
2271 			sc->ale_stats.reset_brk_seq++;
2272 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2273 			ale_init_locked(sc);
2274 			ALE_UNLOCK(sc);
2275 			return;
2276 		}
2277 
2278 		if ((status & (INTR_DMA_RD_TO_RST | INTR_DMA_WR_TO_RST)) != 0) {
2279 			if ((status & INTR_DMA_RD_TO_RST) != 0)
2280 				device_printf(sc->ale_dev,
2281 				    "DMA read error! -- resetting\n");
2282 			if ((status & INTR_DMA_WR_TO_RST) != 0)
2283 				device_printf(sc->ale_dev,
2284 				    "DMA write error! -- resetting\n");
2285 			ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
2286 			ale_init_locked(sc);
2287 			ALE_UNLOCK(sc);
2288 			return;
2289 		}
2290 		if (!IFQ_DRV_IS_EMPTY(&ifp->if_snd))
2291 			ale_start_locked(ifp);
2292 	}
2293 
2294 	if (more == EAGAIN ||
2295 	    (CSR_READ_4(sc, ALE_INTR_STATUS) & ALE_INTRS) != 0) {
2296 		ALE_UNLOCK(sc);
2297 		taskqueue_enqueue(sc->ale_tq, &sc->ale_int_task);
2298 		return;
2299 	}
2300 
2301 done:
2302 	ALE_UNLOCK(sc);
2303 
2304 	/* Re-enable interrupts. */
2305 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0x7FFFFFFF);
2306 }
2307 
2308 static void
ale_txeof(struct ale_softc * sc)2309 ale_txeof(struct ale_softc *sc)
2310 {
2311 	struct ifnet *ifp;
2312 	struct ale_txdesc *txd;
2313 	uint32_t cons, prod;
2314 	int prog;
2315 
2316 	ALE_LOCK_ASSERT(sc);
2317 
2318 	ifp = sc->ale_ifp;
2319 
2320 	if (sc->ale_cdata.ale_tx_cnt == 0)
2321 		return;
2322 
2323 	bus_dmamap_sync(sc->ale_cdata.ale_tx_ring_tag,
2324 	    sc->ale_cdata.ale_tx_ring_map,
2325 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2326 	if ((sc->ale_flags & ALE_FLAG_TXCMB_BUG) == 0) {
2327 		bus_dmamap_sync(sc->ale_cdata.ale_tx_cmb_tag,
2328 		    sc->ale_cdata.ale_tx_cmb_map,
2329 		    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2330 		prod = *sc->ale_cdata.ale_tx_cmb & TPD_CNT_MASK;
2331 	} else
2332 		prod = CSR_READ_2(sc, ALE_TPD_CONS_IDX);
2333 	cons = sc->ale_cdata.ale_tx_cons;
2334 	/*
2335 	 * Go through our Tx list and free mbufs for those
2336 	 * frames which have been transmitted.
2337 	 */
2338 	for (prog = 0; cons != prod; prog++,
2339 	    ALE_DESC_INC(cons, ALE_TX_RING_CNT)) {
2340 		if (sc->ale_cdata.ale_tx_cnt <= 0)
2341 			break;
2342 		prog++;
2343 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2344 		sc->ale_cdata.ale_tx_cnt--;
2345 		txd = &sc->ale_cdata.ale_txdesc[cons];
2346 		if (txd->tx_m != NULL) {
2347 			/* Reclaim transmitted mbufs. */
2348 			bus_dmamap_sync(sc->ale_cdata.ale_tx_tag,
2349 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2350 			bus_dmamap_unload(sc->ale_cdata.ale_tx_tag,
2351 			    txd->tx_dmamap);
2352 			m_freem(txd->tx_m);
2353 			txd->tx_m = NULL;
2354 		}
2355 	}
2356 
2357 	if (prog > 0) {
2358 		sc->ale_cdata.ale_tx_cons = cons;
2359 		/*
2360 		 * Unarm watchdog timer only when there is no pending
2361 		 * Tx descriptors in queue.
2362 		 */
2363 		if (sc->ale_cdata.ale_tx_cnt == 0)
2364 			sc->ale_watchdog_timer = 0;
2365 	}
2366 }
2367 
2368 static void
ale_rx_update_page(struct ale_softc * sc,struct ale_rx_page ** page,uint32_t length,uint32_t * prod)2369 ale_rx_update_page(struct ale_softc *sc, struct ale_rx_page **page,
2370     uint32_t length, uint32_t *prod)
2371 {
2372 	struct ale_rx_page *rx_page;
2373 
2374 	rx_page = *page;
2375 	/* Update consumer position. */
2376 	rx_page->cons += roundup(length + sizeof(struct rx_rs),
2377 	    ALE_RX_PAGE_ALIGN);
2378 	if (rx_page->cons >= ALE_RX_PAGE_SZ) {
2379 		/*
2380 		 * End of Rx page reached, let hardware reuse
2381 		 * this page.
2382 		 */
2383 		rx_page->cons = 0;
2384 		*rx_page->cmb_addr = 0;
2385 		bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2386 		    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2387 		CSR_WRITE_1(sc, ALE_RXF0_PAGE0 + sc->ale_cdata.ale_rx_curp,
2388 		    RXF_VALID);
2389 		/* Switch to alternate Rx page. */
2390 		sc->ale_cdata.ale_rx_curp ^= 1;
2391 		rx_page = *page =
2392 		    &sc->ale_cdata.ale_rx_page[sc->ale_cdata.ale_rx_curp];
2393 		/* Page flipped, sync CMB and Rx page. */
2394 		bus_dmamap_sync(rx_page->page_tag, rx_page->page_map,
2395 		    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2396 		bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2397 		    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2398 		/* Sync completed, cache updated producer index. */
2399 		*prod = *rx_page->cmb_addr;
2400 	}
2401 }
2402 
2403 
2404 /*
2405  * It seems that AR81xx controller can compute partial checksum.
2406  * The partial checksum value can be used to accelerate checksum
2407  * computation for fragmented TCP/UDP packets. Upper network stack
2408  * already takes advantage of the partial checksum value in IP
2409  * reassembly stage. But I'm not sure the correctness of the
2410  * partial hardware checksum assistance due to lack of data sheet.
2411  * In addition, the Rx feature of controller that requires copying
2412  * for every frames effectively nullifies one of most nice offload
2413  * capability of controller.
2414  */
2415 static void
ale_rxcsum(struct ale_softc * sc,struct mbuf * m,uint32_t status)2416 ale_rxcsum(struct ale_softc *sc, struct mbuf *m, uint32_t status)
2417 {
2418 	struct ifnet *ifp;
2419 	struct ip *ip;
2420 	char *p;
2421 
2422 	ifp = sc->ale_ifp;
2423 	m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
2424 	if ((status & ALE_RD_IPCSUM_NOK) == 0)
2425 		m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
2426 
2427 	if ((sc->ale_flags & ALE_FLAG_RXCSUM_BUG) == 0) {
2428 		if (((status & ALE_RD_IPV4_FRAG) == 0) &&
2429 		    ((status & (ALE_RD_TCP | ALE_RD_UDP)) != 0) &&
2430 		    ((status & ALE_RD_TCP_UDPCSUM_NOK) == 0)) {
2431 			m->m_pkthdr.csum_flags |=
2432 			    CSUM_DATA_VALID | CSUM_PSEUDO_HDR;
2433 			m->m_pkthdr.csum_data = 0xffff;
2434 		}
2435 	} else {
2436 		if ((status & (ALE_RD_TCP | ALE_RD_UDP)) != 0 &&
2437 		    (status & ALE_RD_TCP_UDPCSUM_NOK) == 0) {
2438 			p = mtod(m, char *);
2439 			p += ETHER_HDR_LEN;
2440 			if ((status & ALE_RD_802_3) != 0)
2441 				p += LLC_SNAPFRAMELEN;
2442 			if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) == 0 &&
2443 			    (status & ALE_RD_VLAN) != 0)
2444 				p += ETHER_VLAN_ENCAP_LEN;
2445 			ip = (struct ip *)p;
2446 			if (ip->ip_off != 0 && (status & ALE_RD_IPV4_DF) == 0)
2447 				return;
2448 			m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
2449 			    CSUM_PSEUDO_HDR;
2450 			m->m_pkthdr.csum_data = 0xffff;
2451 		}
2452 	}
2453 	/*
2454 	 * Don't mark bad checksum for TCP/UDP frames
2455 	 * as fragmented frames may always have set
2456 	 * bad checksummed bit of frame status.
2457 	 */
2458 }
2459 
2460 /* Process received frames. */
2461 static int
ale_rxeof(struct ale_softc * sc,int count)2462 ale_rxeof(struct ale_softc *sc, int count)
2463 {
2464 	struct ale_rx_page *rx_page;
2465 	struct rx_rs *rs;
2466 	struct ifnet *ifp;
2467 	struct mbuf *m;
2468 	uint32_t length, prod, seqno, status, vtags;
2469 	int prog;
2470 
2471 	ifp = sc->ale_ifp;
2472 	rx_page = &sc->ale_cdata.ale_rx_page[sc->ale_cdata.ale_rx_curp];
2473 	bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2474 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2475 	bus_dmamap_sync(rx_page->page_tag, rx_page->page_map,
2476 	    BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE);
2477 	/*
2478 	 * Don't directly access producer index as hardware may
2479 	 * update it while Rx handler is in progress. It would
2480 	 * be even better if there is a way to let hardware
2481 	 * know how far driver processed its received frames.
2482 	 * Alternatively, hardware could provide a way to disable
2483 	 * CMB updates until driver acknowledges the end of CMB
2484 	 * access.
2485 	 */
2486 	prod = *rx_page->cmb_addr;
2487 	for (prog = 0; prog < count; prog++) {
2488 		if (rx_page->cons >= prod)
2489 			break;
2490 		rs = (struct rx_rs *)(rx_page->page_addr + rx_page->cons);
2491 		seqno = ALE_RX_SEQNO(le32toh(rs->seqno));
2492 		if (sc->ale_cdata.ale_rx_seqno != seqno) {
2493 			/*
2494 			 * Normally I believe this should not happen unless
2495 			 * severe driver bug or corrupted memory. However
2496 			 * it seems to happen under certain conditions which
2497 			 * is triggered by abrupt Rx events such as initiation
2498 			 * of bulk transfer of remote host. It's not easy to
2499 			 * reproduce this and I doubt it could be related
2500 			 * with FIFO overflow of hardware or activity of Tx
2501 			 * CMB updates. I also remember similar behaviour
2502 			 * seen on RealTek 8139 which uses resembling Rx
2503 			 * scheme.
2504 			 */
2505 			if (bootverbose)
2506 				device_printf(sc->ale_dev,
2507 				    "garbled seq: %u, expected: %u -- "
2508 				    "resetting!\n", seqno,
2509 				    sc->ale_cdata.ale_rx_seqno);
2510 			return (EIO);
2511 		}
2512 		/* Frame received. */
2513 		sc->ale_cdata.ale_rx_seqno++;
2514 		length = ALE_RX_BYTES(le32toh(rs->length));
2515 		status = le32toh(rs->flags);
2516 		if ((status & ALE_RD_ERROR) != 0) {
2517 			/*
2518 			 * We want to pass the following frames to upper
2519 			 * layer regardless of error status of Rx return
2520 			 * status.
2521 			 *
2522 			 *  o IP/TCP/UDP checksum is bad.
2523 			 *  o frame length and protocol specific length
2524 			 *     does not match.
2525 			 */
2526 			if ((status & (ALE_RD_CRC | ALE_RD_CODE |
2527 			    ALE_RD_DRIBBLE | ALE_RD_RUNT | ALE_RD_OFLOW |
2528 			    ALE_RD_TRUNC)) != 0) {
2529 				ale_rx_update_page(sc, &rx_page, length, &prod);
2530 				continue;
2531 			}
2532 		}
2533 		/*
2534 		 * m_devget(9) is major bottle-neck of ale(4)(It comes
2535 		 * from hardware limitation). For jumbo frames we could
2536 		 * get a slightly better performance if driver use
2537 		 * m_getjcl(9) with proper buffer size argument. However
2538 		 * that would make code more complicated and I don't
2539 		 * think users would expect good Rx performance numbers
2540 		 * on these low-end consumer ethernet controller.
2541 		 */
2542 		m = m_devget((char *)(rs + 1), length - ETHER_CRC_LEN,
2543 		    ETHER_ALIGN, ifp, NULL);
2544 		if (m == NULL) {
2545 			ifp->if_iqdrops++;
2546 			ale_rx_update_page(sc, &rx_page, length, &prod);
2547 			continue;
2548 		}
2549 		if ((ifp->if_capenable & IFCAP_RXCSUM) != 0 &&
2550 		    (status & ALE_RD_IPV4) != 0)
2551 			ale_rxcsum(sc, m, status);
2552 		if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0 &&
2553 		    (status & ALE_RD_VLAN) != 0) {
2554 			vtags = ALE_RX_VLAN(le32toh(rs->vtags));
2555 			m->m_pkthdr.ether_vtag = ALE_RX_VLAN_TAG(vtags);
2556 			m->m_flags |= M_VLANTAG;
2557 		}
2558 
2559 		/* Pass it to upper layer. */
2560 		ALE_UNLOCK(sc);
2561 		(*ifp->if_input)(ifp, m);
2562 		ALE_LOCK(sc);
2563 
2564 		ale_rx_update_page(sc, &rx_page, length, &prod);
2565 	}
2566 
2567 	return (count > 0 ? 0 : EAGAIN);
2568 }
2569 
2570 static void
ale_tick(void * arg)2571 ale_tick(void *arg)
2572 {
2573 	struct ale_softc *sc;
2574 	struct mii_data *mii;
2575 
2576 	sc = (struct ale_softc *)arg;
2577 
2578 	ALE_LOCK_ASSERT(sc);
2579 
2580 	mii = device_get_softc(sc->ale_miibus);
2581 	mii_tick(mii);
2582 	ale_stats_update(sc);
2583 	/*
2584 	 * Reclaim Tx buffers that have been transferred. It's not
2585 	 * needed here but it would release allocated mbuf chains
2586 	 * faster and limit the maximum delay to a hz.
2587 	 */
2588 	ale_txeof(sc);
2589 	ale_watchdog(sc);
2590 	callout_reset(&sc->ale_tick_ch, hz, ale_tick, sc);
2591 }
2592 
2593 static void
ale_reset(struct ale_softc * sc)2594 ale_reset(struct ale_softc *sc)
2595 {
2596 	uint32_t reg;
2597 	int i;
2598 
2599 	/* Initialize PCIe module. From Linux. */
2600 	CSR_WRITE_4(sc, 0x1008, CSR_READ_4(sc, 0x1008) | 0x8000);
2601 
2602 	CSR_WRITE_4(sc, ALE_MASTER_CFG, MASTER_RESET);
2603 	for (i = ALE_RESET_TIMEOUT; i > 0; i--) {
2604 		DELAY(10);
2605 		if ((CSR_READ_4(sc, ALE_MASTER_CFG) & MASTER_RESET) == 0)
2606 			break;
2607 	}
2608 	if (i == 0)
2609 		device_printf(sc->ale_dev, "master reset timeout!\n");
2610 
2611 	for (i = ALE_RESET_TIMEOUT; i > 0; i--) {
2612 		if ((reg = CSR_READ_4(sc, ALE_IDLE_STATUS)) == 0)
2613 			break;
2614 		DELAY(10);
2615 	}
2616 
2617 	if (i == 0)
2618 		device_printf(sc->ale_dev, "reset timeout(0x%08x)!\n", reg);
2619 }
2620 
2621 static void
ale_init(void * xsc)2622 ale_init(void *xsc)
2623 {
2624 	struct ale_softc *sc;
2625 
2626 	sc = (struct ale_softc *)xsc;
2627 	ALE_LOCK(sc);
2628 	ale_init_locked(sc);
2629 	ALE_UNLOCK(sc);
2630 }
2631 
2632 static void
ale_init_locked(struct ale_softc * sc)2633 ale_init_locked(struct ale_softc *sc)
2634 {
2635 	struct ifnet *ifp;
2636 	struct mii_data *mii;
2637 	uint8_t eaddr[ETHER_ADDR_LEN];
2638 	bus_addr_t paddr;
2639 	uint32_t reg, rxf_hi, rxf_lo;
2640 
2641 	ALE_LOCK_ASSERT(sc);
2642 
2643 	ifp = sc->ale_ifp;
2644 	mii = device_get_softc(sc->ale_miibus);
2645 
2646 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) != 0)
2647 		return;
2648 	/*
2649 	 * Cancel any pending I/O.
2650 	 */
2651 	ale_stop(sc);
2652 	/*
2653 	 * Reset the chip to a known state.
2654 	 */
2655 	ale_reset(sc);
2656 	/* Initialize Tx descriptors, DMA memory blocks. */
2657 	ale_init_rx_pages(sc);
2658 	ale_init_tx_ring(sc);
2659 
2660 	/* Reprogram the station address. */
2661 	bcopy(IF_LLADDR(ifp), eaddr, ETHER_ADDR_LEN);
2662 	CSR_WRITE_4(sc, ALE_PAR0,
2663 	    eaddr[2] << 24 | eaddr[3] << 16 | eaddr[4] << 8 | eaddr[5]);
2664 	CSR_WRITE_4(sc, ALE_PAR1, eaddr[0] << 8 | eaddr[1]);
2665 	/*
2666 	 * Clear WOL status and disable all WOL feature as WOL
2667 	 * would interfere Rx operation under normal environments.
2668 	 */
2669 	CSR_READ_4(sc, ALE_WOL_CFG);
2670 	CSR_WRITE_4(sc, ALE_WOL_CFG, 0);
2671 	/*
2672 	 * Set Tx descriptor/RXF0/CMB base addresses. They share
2673 	 * the same high address part of DMAable region.
2674 	 */
2675 	paddr = sc->ale_cdata.ale_tx_ring_paddr;
2676 	CSR_WRITE_4(sc, ALE_TPD_ADDR_HI, ALE_ADDR_HI(paddr));
2677 	CSR_WRITE_4(sc, ALE_TPD_ADDR_LO, ALE_ADDR_LO(paddr));
2678 	CSR_WRITE_4(sc, ALE_TPD_CNT,
2679 	    (ALE_TX_RING_CNT << TPD_CNT_SHIFT) & TPD_CNT_MASK);
2680 	/* Set Rx page base address, note we use single queue. */
2681 	paddr = sc->ale_cdata.ale_rx_page[0].page_paddr;
2682 	CSR_WRITE_4(sc, ALE_RXF0_PAGE0_ADDR_LO, ALE_ADDR_LO(paddr));
2683 	paddr = sc->ale_cdata.ale_rx_page[1].page_paddr;
2684 	CSR_WRITE_4(sc, ALE_RXF0_PAGE1_ADDR_LO, ALE_ADDR_LO(paddr));
2685 	/* Set Tx/Rx CMB addresses. */
2686 	paddr = sc->ale_cdata.ale_tx_cmb_paddr;
2687 	CSR_WRITE_4(sc, ALE_TX_CMB_ADDR_LO, ALE_ADDR_LO(paddr));
2688 	paddr = sc->ale_cdata.ale_rx_page[0].cmb_paddr;
2689 	CSR_WRITE_4(sc, ALE_RXF0_CMB0_ADDR_LO, ALE_ADDR_LO(paddr));
2690 	paddr = sc->ale_cdata.ale_rx_page[1].cmb_paddr;
2691 	CSR_WRITE_4(sc, ALE_RXF0_CMB1_ADDR_LO, ALE_ADDR_LO(paddr));
2692 	/* Mark RXF0 is valid. */
2693 	CSR_WRITE_1(sc, ALE_RXF0_PAGE0, RXF_VALID);
2694 	CSR_WRITE_1(sc, ALE_RXF0_PAGE1, RXF_VALID);
2695 	/*
2696 	 * No need to initialize RFX1/RXF2/RXF3. We don't use
2697 	 * multi-queue yet.
2698 	 */
2699 
2700 	/* Set Rx page size, excluding guard frame size. */
2701 	CSR_WRITE_4(sc, ALE_RXF_PAGE_SIZE, ALE_RX_PAGE_SZ);
2702 	/* Tell hardware that we're ready to load DMA blocks. */
2703 	CSR_WRITE_4(sc, ALE_DMA_BLOCK, DMA_BLOCK_LOAD);
2704 
2705 	/* Set Rx/Tx interrupt trigger threshold. */
2706 	CSR_WRITE_4(sc, ALE_INT_TRIG_THRESH, (1 << INT_TRIG_RX_THRESH_SHIFT) |
2707 	    (4 << INT_TRIG_TX_THRESH_SHIFT));
2708 	/*
2709 	 * XXX
2710 	 * Set interrupt trigger timer, its purpose and relation
2711 	 * with interrupt moderation mechanism is not clear yet.
2712 	 */
2713 	CSR_WRITE_4(sc, ALE_INT_TRIG_TIMER,
2714 	    ((ALE_USECS(10) << INT_TRIG_RX_TIMER_SHIFT) |
2715 	    (ALE_USECS(1000) << INT_TRIG_TX_TIMER_SHIFT)));
2716 
2717 	/* Configure interrupt moderation timer. */
2718 	reg = ALE_USECS(sc->ale_int_rx_mod) << IM_TIMER_RX_SHIFT;
2719 	reg |= ALE_USECS(sc->ale_int_tx_mod) << IM_TIMER_TX_SHIFT;
2720 	CSR_WRITE_4(sc, ALE_IM_TIMER, reg);
2721 	reg = CSR_READ_4(sc, ALE_MASTER_CFG);
2722 	reg &= ~(MASTER_CHIP_REV_MASK | MASTER_CHIP_ID_MASK);
2723 	reg &= ~(MASTER_IM_RX_TIMER_ENB | MASTER_IM_TX_TIMER_ENB);
2724 	if (ALE_USECS(sc->ale_int_rx_mod) != 0)
2725 		reg |= MASTER_IM_RX_TIMER_ENB;
2726 	if (ALE_USECS(sc->ale_int_tx_mod) != 0)
2727 		reg |= MASTER_IM_TX_TIMER_ENB;
2728 	CSR_WRITE_4(sc, ALE_MASTER_CFG, reg);
2729 	CSR_WRITE_2(sc, ALE_INTR_CLR_TIMER, ALE_USECS(1000));
2730 
2731 	/* Set Maximum frame size of controller. */
2732 	if (ifp->if_mtu < ETHERMTU)
2733 		sc->ale_max_frame_size = ETHERMTU;
2734 	else
2735 		sc->ale_max_frame_size = ifp->if_mtu;
2736 	sc->ale_max_frame_size += ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN +
2737 	    ETHER_CRC_LEN;
2738 	CSR_WRITE_4(sc, ALE_FRAME_SIZE, sc->ale_max_frame_size);
2739 	/* Configure IPG/IFG parameters. */
2740 	CSR_WRITE_4(sc, ALE_IPG_IFG_CFG,
2741 	    ((IPG_IFG_IPGT_DEFAULT << IPG_IFG_IPGT_SHIFT) & IPG_IFG_IPGT_MASK) |
2742 	    ((IPG_IFG_MIFG_DEFAULT << IPG_IFG_MIFG_SHIFT) & IPG_IFG_MIFG_MASK) |
2743 	    ((IPG_IFG_IPG1_DEFAULT << IPG_IFG_IPG1_SHIFT) & IPG_IFG_IPG1_MASK) |
2744 	    ((IPG_IFG_IPG2_DEFAULT << IPG_IFG_IPG2_SHIFT) & IPG_IFG_IPG2_MASK));
2745 	/* Set parameters for half-duplex media. */
2746 	CSR_WRITE_4(sc, ALE_HDPX_CFG,
2747 	    ((HDPX_CFG_LCOL_DEFAULT << HDPX_CFG_LCOL_SHIFT) &
2748 	    HDPX_CFG_LCOL_MASK) |
2749 	    ((HDPX_CFG_RETRY_DEFAULT << HDPX_CFG_RETRY_SHIFT) &
2750 	    HDPX_CFG_RETRY_MASK) | HDPX_CFG_EXC_DEF_EN |
2751 	    ((HDPX_CFG_ABEBT_DEFAULT << HDPX_CFG_ABEBT_SHIFT) &
2752 	    HDPX_CFG_ABEBT_MASK) |
2753 	    ((HDPX_CFG_JAMIPG_DEFAULT << HDPX_CFG_JAMIPG_SHIFT) &
2754 	    HDPX_CFG_JAMIPG_MASK));
2755 
2756 	/* Configure Tx jumbo frame parameters. */
2757 	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0) {
2758 		if (ifp->if_mtu < ETHERMTU)
2759 			reg = sc->ale_max_frame_size;
2760 		else if (ifp->if_mtu < 6 * 1024)
2761 			reg = (sc->ale_max_frame_size * 2) / 3;
2762 		else
2763 			reg = sc->ale_max_frame_size / 2;
2764 		CSR_WRITE_4(sc, ALE_TX_JUMBO_THRESH,
2765 		    roundup(reg, TX_JUMBO_THRESH_UNIT) >>
2766 		    TX_JUMBO_THRESH_UNIT_SHIFT);
2767 	}
2768 	/* Configure TxQ. */
2769 	reg = (128 << (sc->ale_dma_rd_burst >> DMA_CFG_RD_BURST_SHIFT))
2770 	    << TXQ_CFG_TX_FIFO_BURST_SHIFT;
2771 	reg |= (TXQ_CFG_TPD_BURST_DEFAULT << TXQ_CFG_TPD_BURST_SHIFT) &
2772 	    TXQ_CFG_TPD_BURST_MASK;
2773 	CSR_WRITE_4(sc, ALE_TXQ_CFG, reg | TXQ_CFG_ENHANCED_MODE | TXQ_CFG_ENB);
2774 
2775 	/* Configure Rx jumbo frame & flow control parameters. */
2776 	if ((sc->ale_flags & ALE_FLAG_JUMBO) != 0) {
2777 		reg = roundup(sc->ale_max_frame_size, RX_JUMBO_THRESH_UNIT);
2778 		CSR_WRITE_4(sc, ALE_RX_JUMBO_THRESH,
2779 		    (((reg >> RX_JUMBO_THRESH_UNIT_SHIFT) <<
2780 		    RX_JUMBO_THRESH_MASK_SHIFT) & RX_JUMBO_THRESH_MASK) |
2781 		    ((RX_JUMBO_LKAH_DEFAULT << RX_JUMBO_LKAH_SHIFT) &
2782 		    RX_JUMBO_LKAH_MASK));
2783 		reg = CSR_READ_4(sc, ALE_SRAM_RX_FIFO_LEN);
2784 		rxf_hi = (reg * 7) / 10;
2785 		rxf_lo = (reg * 3)/ 10;
2786 		CSR_WRITE_4(sc, ALE_RX_FIFO_PAUSE_THRESH,
2787 		    ((rxf_lo << RX_FIFO_PAUSE_THRESH_LO_SHIFT) &
2788 		    RX_FIFO_PAUSE_THRESH_LO_MASK) |
2789 		    ((rxf_hi << RX_FIFO_PAUSE_THRESH_HI_SHIFT) &
2790 		    RX_FIFO_PAUSE_THRESH_HI_MASK));
2791 	}
2792 
2793 	/* Disable RSS. */
2794 	CSR_WRITE_4(sc, ALE_RSS_IDT_TABLE0, 0);
2795 	CSR_WRITE_4(sc, ALE_RSS_CPU, 0);
2796 
2797 	/* Configure RxQ. */
2798 	CSR_WRITE_4(sc, ALE_RXQ_CFG,
2799 	    RXQ_CFG_ALIGN_32 | RXQ_CFG_CUT_THROUGH_ENB | RXQ_CFG_ENB);
2800 
2801 	/* Configure DMA parameters. */
2802 	reg = 0;
2803 	if ((sc->ale_flags & ALE_FLAG_TXCMB_BUG) == 0)
2804 		reg |= DMA_CFG_TXCMB_ENB;
2805 	CSR_WRITE_4(sc, ALE_DMA_CFG,
2806 	    DMA_CFG_OUT_ORDER | DMA_CFG_RD_REQ_PRI | DMA_CFG_RCB_64 |
2807 	    sc->ale_dma_rd_burst | reg |
2808 	    sc->ale_dma_wr_burst | DMA_CFG_RXCMB_ENB |
2809 	    ((DMA_CFG_RD_DELAY_CNT_DEFAULT << DMA_CFG_RD_DELAY_CNT_SHIFT) &
2810 	    DMA_CFG_RD_DELAY_CNT_MASK) |
2811 	    ((DMA_CFG_WR_DELAY_CNT_DEFAULT << DMA_CFG_WR_DELAY_CNT_SHIFT) &
2812 	    DMA_CFG_WR_DELAY_CNT_MASK));
2813 
2814 	/*
2815 	 * Hardware can be configured to issue SMB interrupt based
2816 	 * on programmed interval. Since there is a callout that is
2817 	 * invoked for every hz in driver we use that instead of
2818 	 * relying on periodic SMB interrupt.
2819 	 */
2820 	CSR_WRITE_4(sc, ALE_SMB_STAT_TIMER, ALE_USECS(0));
2821 	/* Clear MAC statistics. */
2822 	ale_stats_clear(sc);
2823 
2824 	/*
2825 	 * Configure Tx/Rx MACs.
2826 	 *  - Auto-padding for short frames.
2827 	 *  - Enable CRC generation.
2828 	 *  Actual reconfiguration of MAC for resolved speed/duplex
2829 	 *  is followed after detection of link establishment.
2830 	 *  AR81xx always does checksum computation regardless of
2831 	 *  MAC_CFG_RXCSUM_ENB bit. In fact, setting the bit will
2832 	 *  cause Rx handling issue for fragmented IP datagrams due
2833 	 *  to silicon bug.
2834 	 */
2835 	reg = MAC_CFG_TX_CRC_ENB | MAC_CFG_TX_AUTO_PAD | MAC_CFG_FULL_DUPLEX |
2836 	    ((MAC_CFG_PREAMBLE_DEFAULT << MAC_CFG_PREAMBLE_SHIFT) &
2837 	    MAC_CFG_PREAMBLE_MASK);
2838 	if ((sc->ale_flags & ALE_FLAG_FASTETHER) != 0)
2839 		reg |= MAC_CFG_SPEED_10_100;
2840 	else
2841 		reg |= MAC_CFG_SPEED_1000;
2842 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2843 
2844 	/* Set up the receive filter. */
2845 	ale_rxfilter(sc);
2846 	ale_rxvlan(sc);
2847 
2848 	/* Acknowledge all pending interrupts and clear it. */
2849 	CSR_WRITE_4(sc, ALE_INTR_MASK, ALE_INTRS);
2850 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0xFFFFFFFF);
2851 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0);
2852 
2853 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
2854 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
2855 
2856 	sc->ale_flags &= ~ALE_FLAG_LINK;
2857 	/* Switch to the current media. */
2858 	mii_mediachg(mii);
2859 
2860 	callout_reset(&sc->ale_tick_ch, hz, ale_tick, sc);
2861 }
2862 
2863 static void
ale_stop(struct ale_softc * sc)2864 ale_stop(struct ale_softc *sc)
2865 {
2866 	struct ifnet *ifp;
2867 	struct ale_txdesc *txd;
2868 	uint32_t reg;
2869 	int i;
2870 
2871 	ALE_LOCK_ASSERT(sc);
2872 	/*
2873 	 * Mark the interface down and cancel the watchdog timer.
2874 	 */
2875 	ifp = sc->ale_ifp;
2876 	ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
2877 	sc->ale_flags &= ~ALE_FLAG_LINK;
2878 	callout_stop(&sc->ale_tick_ch);
2879 	sc->ale_watchdog_timer = 0;
2880 	ale_stats_update(sc);
2881 	/* Disable interrupts. */
2882 	CSR_WRITE_4(sc, ALE_INTR_MASK, 0);
2883 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0xFFFFFFFF);
2884 	/* Disable queue processing and DMA. */
2885 	reg = CSR_READ_4(sc, ALE_TXQ_CFG);
2886 	reg &= ~TXQ_CFG_ENB;
2887 	CSR_WRITE_4(sc, ALE_TXQ_CFG, reg);
2888 	reg = CSR_READ_4(sc, ALE_RXQ_CFG);
2889 	reg &= ~RXQ_CFG_ENB;
2890 	CSR_WRITE_4(sc, ALE_RXQ_CFG, reg);
2891 	reg = CSR_READ_4(sc, ALE_DMA_CFG);
2892 	reg &= ~(DMA_CFG_TXCMB_ENB | DMA_CFG_RXCMB_ENB);
2893 	CSR_WRITE_4(sc, ALE_DMA_CFG, reg);
2894 	DELAY(1000);
2895 	/* Stop Rx/Tx MACs. */
2896 	ale_stop_mac(sc);
2897 	/* Disable interrupts which might be touched in taskq handler. */
2898 	CSR_WRITE_4(sc, ALE_INTR_STATUS, 0xFFFFFFFF);
2899 
2900 	/*
2901 	 * Free TX mbufs still in the queues.
2902 	 */
2903 	for (i = 0; i < ALE_TX_RING_CNT; i++) {
2904 		txd = &sc->ale_cdata.ale_txdesc[i];
2905 		if (txd->tx_m != NULL) {
2906 			bus_dmamap_sync(sc->ale_cdata.ale_tx_tag,
2907 			    txd->tx_dmamap, BUS_DMASYNC_POSTWRITE);
2908 			bus_dmamap_unload(sc->ale_cdata.ale_tx_tag,
2909 			    txd->tx_dmamap);
2910 			m_freem(txd->tx_m);
2911 			txd->tx_m = NULL;
2912 		}
2913         }
2914 }
2915 
2916 static void
ale_stop_mac(struct ale_softc * sc)2917 ale_stop_mac(struct ale_softc *sc)
2918 {
2919 	uint32_t reg;
2920 	int i;
2921 
2922 	ALE_LOCK_ASSERT(sc);
2923 
2924 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
2925 	if ((reg & (MAC_CFG_TX_ENB | MAC_CFG_RX_ENB)) != 0) {
2926 		reg &= ~(MAC_CFG_TX_ENB | MAC_CFG_RX_ENB);
2927 		CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
2928 	}
2929 
2930 	for (i = ALE_TIMEOUT; i > 0; i--) {
2931 		reg = CSR_READ_4(sc, ALE_IDLE_STATUS);
2932 		if (reg == 0)
2933 			break;
2934 		DELAY(10);
2935 	}
2936 	if (i == 0)
2937 		device_printf(sc->ale_dev,
2938 		    "could not disable Tx/Rx MAC(0x%08x)!\n", reg);
2939 }
2940 
2941 static void
ale_init_tx_ring(struct ale_softc * sc)2942 ale_init_tx_ring(struct ale_softc *sc)
2943 {
2944 	struct ale_txdesc *txd;
2945 	int i;
2946 
2947 	ALE_LOCK_ASSERT(sc);
2948 
2949 	sc->ale_cdata.ale_tx_prod = 0;
2950 	sc->ale_cdata.ale_tx_cons = 0;
2951 	sc->ale_cdata.ale_tx_cnt = 0;
2952 
2953 	bzero(sc->ale_cdata.ale_tx_ring, ALE_TX_RING_SZ);
2954 	bzero(sc->ale_cdata.ale_tx_cmb, ALE_TX_CMB_SZ);
2955 	for (i = 0; i < ALE_TX_RING_CNT; i++) {
2956 		txd = &sc->ale_cdata.ale_txdesc[i];
2957 		txd->tx_m = NULL;
2958 	}
2959 	*sc->ale_cdata.ale_tx_cmb = 0;
2960 	bus_dmamap_sync(sc->ale_cdata.ale_tx_cmb_tag,
2961 	    sc->ale_cdata.ale_tx_cmb_map,
2962 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2963 	bus_dmamap_sync(sc->ale_cdata.ale_tx_ring_tag,
2964 	    sc->ale_cdata.ale_tx_ring_map,
2965 	    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2966 }
2967 
2968 static void
ale_init_rx_pages(struct ale_softc * sc)2969 ale_init_rx_pages(struct ale_softc *sc)
2970 {
2971 	struct ale_rx_page *rx_page;
2972 	int i;
2973 
2974 	ALE_LOCK_ASSERT(sc);
2975 
2976 	sc->ale_morework = 0;
2977 	sc->ale_cdata.ale_rx_seqno = 0;
2978 	sc->ale_cdata.ale_rx_curp = 0;
2979 
2980 	for (i = 0; i < ALE_RX_PAGES; i++) {
2981 		rx_page = &sc->ale_cdata.ale_rx_page[i];
2982 		bzero(rx_page->page_addr, sc->ale_pagesize);
2983 		bzero(rx_page->cmb_addr, ALE_RX_CMB_SZ);
2984 		rx_page->cons = 0;
2985 		*rx_page->cmb_addr = 0;
2986 		bus_dmamap_sync(rx_page->page_tag, rx_page->page_map,
2987 		    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2988 		bus_dmamap_sync(rx_page->cmb_tag, rx_page->cmb_map,
2989 		    BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
2990 	}
2991 }
2992 
2993 static void
ale_rxvlan(struct ale_softc * sc)2994 ale_rxvlan(struct ale_softc *sc)
2995 {
2996 	struct ifnet *ifp;
2997 	uint32_t reg;
2998 
2999 	ALE_LOCK_ASSERT(sc);
3000 
3001 	ifp = sc->ale_ifp;
3002 	reg = CSR_READ_4(sc, ALE_MAC_CFG);
3003 	reg &= ~MAC_CFG_VLAN_TAG_STRIP;
3004 	if ((ifp->if_capenable & IFCAP_VLAN_HWTAGGING) != 0)
3005 		reg |= MAC_CFG_VLAN_TAG_STRIP;
3006 	CSR_WRITE_4(sc, ALE_MAC_CFG, reg);
3007 }
3008 
3009 static void
ale_rxfilter(struct ale_softc * sc)3010 ale_rxfilter(struct ale_softc *sc)
3011 {
3012 	struct ifnet *ifp;
3013 	struct ifmultiaddr *ifma;
3014 	uint32_t crc;
3015 	uint32_t mchash[2];
3016 	uint32_t rxcfg;
3017 
3018 	ALE_LOCK_ASSERT(sc);
3019 
3020 	ifp = sc->ale_ifp;
3021 
3022 	rxcfg = CSR_READ_4(sc, ALE_MAC_CFG);
3023 	rxcfg &= ~(MAC_CFG_ALLMULTI | MAC_CFG_BCAST | MAC_CFG_PROMISC);
3024 	if ((ifp->if_flags & IFF_BROADCAST) != 0)
3025 		rxcfg |= MAC_CFG_BCAST;
3026 	if ((ifp->if_flags & (IFF_PROMISC | IFF_ALLMULTI)) != 0) {
3027 		if ((ifp->if_flags & IFF_PROMISC) != 0)
3028 			rxcfg |= MAC_CFG_PROMISC;
3029 		if ((ifp->if_flags & IFF_ALLMULTI) != 0)
3030 			rxcfg |= MAC_CFG_ALLMULTI;
3031 		CSR_WRITE_4(sc, ALE_MAR0, 0xFFFFFFFF);
3032 		CSR_WRITE_4(sc, ALE_MAR1, 0xFFFFFFFF);
3033 		CSR_WRITE_4(sc, ALE_MAC_CFG, rxcfg);
3034 		return;
3035 	}
3036 
3037 	/* Program new filter. */
3038 	bzero(mchash, sizeof(mchash));
3039 
3040 	if_maddr_rlock(ifp);
3041 	TAILQ_FOREACH(ifma, &sc->ale_ifp->if_multiaddrs, ifma_link) {
3042 		if (ifma->ifma_addr->sa_family != AF_LINK)
3043 			continue;
3044 		crc = ether_crc32_be(LLADDR((struct sockaddr_dl *)
3045 		    ifma->ifma_addr), ETHER_ADDR_LEN);
3046 		mchash[crc >> 31] |= 1 << ((crc >> 26) & 0x1f);
3047 	}
3048 	if_maddr_runlock(ifp);
3049 
3050 	CSR_WRITE_4(sc, ALE_MAR0, mchash[0]);
3051 	CSR_WRITE_4(sc, ALE_MAR1, mchash[1]);
3052 	CSR_WRITE_4(sc, ALE_MAC_CFG, rxcfg);
3053 }
3054 
3055 static int
sysctl_int_range(SYSCTL_HANDLER_ARGS,int low,int high)3056 sysctl_int_range(SYSCTL_HANDLER_ARGS, int low, int high)
3057 {
3058 	int error, value;
3059 
3060 	if (arg1 == NULL)
3061 		return (EINVAL);
3062 	value = *(int *)arg1;
3063 	error = sysctl_handle_int(oidp, &value, 0, req);
3064 	if (error || req->newptr == NULL)
3065 		return (error);
3066 	if (value < low || value > high)
3067 		return (EINVAL);
3068         *(int *)arg1 = value;
3069 
3070         return (0);
3071 }
3072 
3073 static int
sysctl_hw_ale_proc_limit(SYSCTL_HANDLER_ARGS)3074 sysctl_hw_ale_proc_limit(SYSCTL_HANDLER_ARGS)
3075 {
3076 	return (sysctl_int_range(oidp, arg1, arg2, req,
3077 	    ALE_PROC_MIN, ALE_PROC_MAX));
3078 }
3079 
3080 static int
sysctl_hw_ale_int_mod(SYSCTL_HANDLER_ARGS)3081 sysctl_hw_ale_int_mod(SYSCTL_HANDLER_ARGS)
3082 {
3083 
3084 	return (sysctl_int_range(oidp, arg1, arg2, req,
3085 	    ALE_IM_TIMER_MIN, ALE_IM_TIMER_MAX));
3086 }
3087