xref: /dragonfly/sys/dev/netif/nfe/if_nfe.c (revision 33311965)
1 /*	$OpenBSD: if_nfe.c,v 1.63 2006/06/17 18:00:43 brad Exp $	*/
2 
3 /*
4  * Copyright (c) 2006 The DragonFly Project.  All rights reserved.
5  *
6  * This code is derived from software contributed to The DragonFly Project
7  * by Sepherosa Ziehau <sepherosa@gmail.com> and
8  * Matthew Dillon <dillon@apollo.backplane.com>
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  *
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in
18  *    the documentation and/or other materials provided with the
19  *    distribution.
20  * 3. Neither the name of The DragonFly Project nor the names of its
21  *    contributors may be used to endorse or promote products derived
22  *    from this software without specific, prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
25  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
26  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
27  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
28  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
29  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
30  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
31  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
32  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
33  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
34  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  */
37 
38 /*
39  * Copyright (c) 2006 Damien Bergamini <damien.bergamini@free.fr>
40  * Copyright (c) 2005, 2006 Jonathan Gray <jsg@openbsd.org>
41  *
42  * Permission to use, copy, modify, and distribute this software for any
43  * purpose with or without fee is hereby granted, provided that the above
44  * copyright notice and this permission notice appear in all copies.
45  *
46  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
47  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
48  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
49  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
50  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
51  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
52  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
53  */
54 
55 /* Driver for NVIDIA nForce MCP Fast Ethernet and Gigabit Ethernet */
56 
57 #include "opt_ifpoll.h"
58 
59 #include <sys/param.h>
60 #include <sys/endian.h>
61 #include <sys/kernel.h>
62 #include <sys/bus.h>
63 #include <sys/interrupt.h>
64 #include <sys/proc.h>
65 #include <sys/rman.h>
66 #include <sys/serialize.h>
67 #include <sys/socket.h>
68 #include <sys/sockio.h>
69 #include <sys/sysctl.h>
70 
71 #include <net/ethernet.h>
72 #include <net/if.h>
73 #include <net/bpf.h>
74 #include <net/if_arp.h>
75 #include <net/if_dl.h>
76 #include <net/if_media.h>
77 #include <net/if_poll.h>
78 #include <net/ifq_var.h>
79 #include <net/if_types.h>
80 #include <net/if_var.h>
81 #include <net/vlan/if_vlan_var.h>
82 #include <net/vlan/if_vlan_ether.h>
83 
84 #include <bus/pci/pcireg.h>
85 #include <bus/pci/pcivar.h>
86 #include "pcidevs.h"
87 
88 #include <dev/netif/mii_layer/mii.h>
89 #include <dev/netif/mii_layer/miivar.h>
90 
91 #include "miibus_if.h"
92 
93 #include <dev/netif/nfe/if_nfereg.h>
94 #include <dev/netif/nfe/if_nfevar.h>
95 
96 #define NFE_CSUM
97 #define NFE_CSUM_FEATURES	(CSUM_IP | CSUM_TCP | CSUM_UDP)
98 
99 static int	nfe_probe(device_t);
100 static int	nfe_attach(device_t);
101 static int	nfe_detach(device_t);
102 static void	nfe_shutdown(device_t);
103 static int	nfe_resume(device_t);
104 static int	nfe_suspend(device_t);
105 
106 static int	nfe_miibus_readreg(device_t, int, int);
107 static void	nfe_miibus_writereg(device_t, int, int, int);
108 static void	nfe_miibus_statchg(device_t);
109 
110 #ifdef IFPOLL_ENABLE
111 static void	nfe_npoll(struct ifnet *, struct ifpoll_info *);
112 static void	nfe_npoll_compat(struct ifnet *, void *, int);
113 static void	nfe_disable_intrs(struct nfe_softc *);
114 #endif
115 static void	nfe_intr(void *);
116 static int	nfe_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
117 static int	nfe_rxeof(struct nfe_softc *);
118 static int	nfe_txeof(struct nfe_softc *, int);
119 static int	nfe_encap(struct nfe_softc *, struct nfe_tx_ring *,
120 			  struct mbuf *);
121 static void	nfe_start(struct ifnet *, struct ifaltq_subque *);
122 static void	nfe_watchdog(struct ifnet *);
123 static void	nfe_init(void *);
124 static void	nfe_stop(struct nfe_softc *);
125 static struct nfe_jbuf *nfe_jalloc(struct nfe_softc *);
126 static void	nfe_jfree(void *);
127 static void	nfe_jref(void *);
128 static int	nfe_jpool_alloc(struct nfe_softc *, struct nfe_rx_ring *);
129 static void	nfe_jpool_free(struct nfe_softc *, struct nfe_rx_ring *);
130 static int	nfe_alloc_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
131 static void	nfe_reset_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
132 static int	nfe_init_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
133 static void	nfe_free_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
134 static int	nfe_alloc_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
135 static void	nfe_reset_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
136 static int	nfe_init_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
137 static void	nfe_free_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
138 static int	nfe_ifmedia_upd(struct ifnet *);
139 static void	nfe_ifmedia_sts(struct ifnet *, struct ifmediareq *);
140 static void	nfe_setmulti(struct nfe_softc *);
141 static void	nfe_get_macaddr(struct nfe_softc *, uint8_t *);
142 static void	nfe_set_macaddr(struct nfe_softc *, const uint8_t *);
143 static void	nfe_powerup(device_t);
144 static void	nfe_mac_reset(struct nfe_softc *);
145 static void	nfe_tick(void *);
146 static void	nfe_set_paddr_rxdesc(struct nfe_softc *, struct nfe_rx_ring *,
147 				     int, bus_addr_t);
148 static void	nfe_set_ready_rxdesc(struct nfe_softc *, struct nfe_rx_ring *,
149 				     int);
150 static int	nfe_newbuf_std(struct nfe_softc *, struct nfe_rx_ring *, int,
151 			       int);
152 static int	nfe_newbuf_jumbo(struct nfe_softc *, struct nfe_rx_ring *, int,
153 				 int);
154 static void	nfe_enable_intrs(struct nfe_softc *);
155 
156 static int	nfe_sysctl_imtime(SYSCTL_HANDLER_ARGS);
157 
158 #define NFE_DEBUG
159 #ifdef NFE_DEBUG
160 
161 static int	nfe_debug = 0;
162 static int	nfe_rx_ring_count = NFE_RX_RING_DEF_COUNT;
163 static int	nfe_tx_ring_count = NFE_TX_RING_DEF_COUNT;
164 /*
165  * hw timer simulated interrupt moderation @4000Hz.  Negative values
166  * disable the timer when the discrete interrupt rate falls below
167  * the moderation rate.
168  *
169  * XXX 8000Hz might be better but if the interrupt is shared it can
170  *     blow out the cpu.
171  */
172 static int	nfe_imtime = -250;	/* uS */
173 
174 TUNABLE_INT("hw.nfe.rx_ring_count", &nfe_rx_ring_count);
175 TUNABLE_INT("hw.nfe.tx_ring_count", &nfe_tx_ring_count);
176 TUNABLE_INT("hw.nfe.imtimer", &nfe_imtime);
177 TUNABLE_INT("hw.nfe.debug", &nfe_debug);
178 
179 #define DPRINTF(sc, fmt, ...) do {		\
180 	if ((sc)->sc_debug) {			\
181 		if_printf(&(sc)->arpcom.ac_if,	\
182 			  fmt, __VA_ARGS__);	\
183 	}					\
184 } while (0)
185 
186 #define DPRINTFN(sc, lv, fmt, ...) do {		\
187 	if ((sc)->sc_debug >= (lv)) {		\
188 		if_printf(&(sc)->arpcom.ac_if,	\
189 			  fmt, __VA_ARGS__);	\
190 	}					\
191 } while (0)
192 
193 #else	/* !NFE_DEBUG */
194 
195 #define DPRINTF(sc, fmt, ...)
196 #define DPRINTFN(sc, lv, fmt, ...)
197 
198 #endif	/* NFE_DEBUG */
199 
200 static const struct nfe_dev {
201 	uint16_t	vid;
202 	uint16_t	did;
203 	const char	*desc;
204 } nfe_devices[] = {
205 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE_LAN,
206 	  "NVIDIA nForce Fast Ethernet" },
207 
208 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE2_LAN,
209 	  "NVIDIA nForce2 Fast Ethernet" },
210 
211 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN1,
212 	  "NVIDIA nForce3 Gigabit Ethernet" },
213 
214 	/* XXX TGEN the next chip can also be found in the nForce2 Ultra 400Gb
215 	   chipset, and possibly also the 400R; it might be both nForce2- and
216 	   nForce3-based boards can use the same MCPs (= southbridges) */
217 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN2,
218 	  "NVIDIA nForce3 Gigabit Ethernet" },
219 
220 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN3,
221 	  "NVIDIA nForce3 Gigabit Ethernet" },
222 
223 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN4,
224 	  "NVIDIA nForce3 Gigabit Ethernet" },
225 
226 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN5,
227 	  "NVIDIA nForce3 Gigabit Ethernet" },
228 
229 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_CK804_LAN1,
230 	  "NVIDIA CK804 Gigabit Ethernet" },
231 
232 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_CK804_LAN2,
233 	  "NVIDIA CK804 Gigabit Ethernet" },
234 
235 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP04_LAN1,
236 	  "NVIDIA MCP04 Gigabit Ethernet" },
237 
238 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP04_LAN2,
239 	  "NVIDIA MCP04 Gigabit Ethernet" },
240 
241 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP51_LAN1,
242 	  "NVIDIA MCP51 Gigabit Ethernet" },
243 
244 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP51_LAN2,
245 	  "NVIDIA MCP51 Gigabit Ethernet" },
246 
247 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP55_LAN1,
248 	  "NVIDIA MCP55 Gigabit Ethernet" },
249 
250 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP55_LAN2,
251 	  "NVIDIA MCP55 Gigabit Ethernet" },
252 
253 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN1,
254 	  "NVIDIA MCP61 Gigabit Ethernet" },
255 
256 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN2,
257 	  "NVIDIA MCP61 Gigabit Ethernet" },
258 
259 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN3,
260 	  "NVIDIA MCP61 Gigabit Ethernet" },
261 
262 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN4,
263 	  "NVIDIA MCP61 Gigabit Ethernet" },
264 
265 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN1,
266 	  "NVIDIA MCP65 Gigabit Ethernet" },
267 
268 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN2,
269 	  "NVIDIA MCP65 Gigabit Ethernet" },
270 
271 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN3,
272 	  "NVIDIA MCP65 Gigabit Ethernet" },
273 
274 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN4,
275 	  "NVIDIA MCP65 Gigabit Ethernet" },
276 
277 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN1,
278 	  "NVIDIA MCP67 Gigabit Ethernet" },
279 
280 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN2,
281 	  "NVIDIA MCP67 Gigabit Ethernet" },
282 
283 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN3,
284 	  "NVIDIA MCP67 Gigabit Ethernet" },
285 
286 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN4,
287 	  "NVIDIA MCP67 Gigabit Ethernet" },
288 
289 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP73_LAN1,
290 	  "NVIDIA MCP73 Gigabit Ethernet" },
291 
292 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP73_LAN2,
293 	  "NVIDIA MCP73 Gigabit Ethernet" },
294 
295 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP73_LAN3,
296 	  "NVIDIA MCP73 Gigabit Ethernet" },
297 
298 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP73_LAN4,
299 	  "NVIDIA MCP73 Gigabit Ethernet" },
300 
301 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP77_LAN1,
302 	  "NVIDIA MCP77 Gigabit Ethernet" },
303 
304 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP77_LAN2,
305 	  "NVIDIA MCP77 Gigabit Ethernet" },
306 
307 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP77_LAN3,
308 	  "NVIDIA MCP77 Gigabit Ethernet" },
309 
310 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP77_LAN4,
311 	  "NVIDIA MCP77 Gigabit Ethernet" },
312 
313 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP79_LAN1,
314 	  "NVIDIA MCP79 Gigabit Ethernet" },
315 
316 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP79_LAN2,
317 	  "NVIDIA MCP79 Gigabit Ethernet" },
318 
319 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP79_LAN3,
320 	  "NVIDIA MCP79 Gigabit Ethernet" },
321 
322 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP79_LAN4,
323 	  "NVIDIA MCP79 Gigabit Ethernet" },
324 
325 	{ 0, 0, NULL }
326 };
327 
328 static device_method_t nfe_methods[] = {
329 	/* Device interface */
330 	DEVMETHOD(device_probe,		nfe_probe),
331 	DEVMETHOD(device_attach,	nfe_attach),
332 	DEVMETHOD(device_detach,	nfe_detach),
333 	DEVMETHOD(device_suspend,	nfe_suspend),
334 	DEVMETHOD(device_resume,	nfe_resume),
335 	DEVMETHOD(device_shutdown,	nfe_shutdown),
336 
337 	/* Bus interface */
338 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
339 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
340 
341 	/* MII interface */
342 	DEVMETHOD(miibus_readreg,	nfe_miibus_readreg),
343 	DEVMETHOD(miibus_writereg,	nfe_miibus_writereg),
344 	DEVMETHOD(miibus_statchg,	nfe_miibus_statchg),
345 
346 	DEVMETHOD_END
347 };
348 
349 static driver_t nfe_driver = {
350 	"nfe",
351 	nfe_methods,
352 	sizeof(struct nfe_softc)
353 };
354 
355 static devclass_t	nfe_devclass;
356 
357 DECLARE_DUMMY_MODULE(if_nfe);
358 MODULE_DEPEND(if_nfe, miibus, 1, 1, 1);
359 DRIVER_MODULE(if_nfe, pci, nfe_driver, nfe_devclass, NULL, NULL);
360 DRIVER_MODULE(miibus, nfe, miibus_driver, miibus_devclass, NULL, NULL);
361 
362 /*
363  * NOTE: NFE_WORDALIGN support is guesswork right now.
364  */
365 static int
366 nfe_probe(device_t dev)
367 {
368 	const struct nfe_dev *n;
369 	uint16_t vid, did;
370 
371 	vid = pci_get_vendor(dev);
372 	did = pci_get_device(dev);
373 	for (n = nfe_devices; n->desc != NULL; ++n) {
374 		if (vid == n->vid && did == n->did) {
375 			struct nfe_softc *sc = device_get_softc(dev);
376 
377 			switch (did) {
378 			case PCI_PRODUCT_NVIDIA_NFORCE_LAN:
379 			case PCI_PRODUCT_NVIDIA_NFORCE2_LAN:
380 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN1:
381 				sc->sc_caps = NFE_NO_PWRCTL |
382 					      NFE_FIX_EADDR;
383 				break;
384 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN2:
385 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN3:
386 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN4:
387 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN5:
388 				sc->sc_caps = NFE_JUMBO_SUP |
389 					      NFE_HW_CSUM |
390 					      NFE_NO_PWRCTL |
391 					      NFE_FIX_EADDR;
392 				break;
393 			case PCI_PRODUCT_NVIDIA_MCP51_LAN1:
394 			case PCI_PRODUCT_NVIDIA_MCP51_LAN2:
395 				sc->sc_caps = NFE_FIX_EADDR;
396 				/* FALL THROUGH */
397 			case PCI_PRODUCT_NVIDIA_MCP61_LAN1:
398 			case PCI_PRODUCT_NVIDIA_MCP61_LAN2:
399 			case PCI_PRODUCT_NVIDIA_MCP61_LAN3:
400 			case PCI_PRODUCT_NVIDIA_MCP61_LAN4:
401 			case PCI_PRODUCT_NVIDIA_MCP67_LAN1:
402 			case PCI_PRODUCT_NVIDIA_MCP67_LAN2:
403 			case PCI_PRODUCT_NVIDIA_MCP67_LAN3:
404 			case PCI_PRODUCT_NVIDIA_MCP67_LAN4:
405 			case PCI_PRODUCT_NVIDIA_MCP73_LAN1:
406 			case PCI_PRODUCT_NVIDIA_MCP73_LAN2:
407 			case PCI_PRODUCT_NVIDIA_MCP73_LAN3:
408 			case PCI_PRODUCT_NVIDIA_MCP73_LAN4:
409 				sc->sc_caps |= NFE_40BIT_ADDR;
410 				break;
411 			case PCI_PRODUCT_NVIDIA_CK804_LAN1:
412 			case PCI_PRODUCT_NVIDIA_CK804_LAN2:
413 			case PCI_PRODUCT_NVIDIA_MCP04_LAN1:
414 			case PCI_PRODUCT_NVIDIA_MCP04_LAN2:
415 				sc->sc_caps = NFE_JUMBO_SUP |
416 					      NFE_40BIT_ADDR |
417 					      NFE_HW_CSUM |
418 					      NFE_NO_PWRCTL |
419 					      NFE_FIX_EADDR;
420 				break;
421 			case PCI_PRODUCT_NVIDIA_MCP65_LAN1:
422 			case PCI_PRODUCT_NVIDIA_MCP65_LAN2:
423 			case PCI_PRODUCT_NVIDIA_MCP65_LAN3:
424 			case PCI_PRODUCT_NVIDIA_MCP65_LAN4:
425 				sc->sc_caps = NFE_JUMBO_SUP |
426 					      NFE_40BIT_ADDR;
427 				break;
428 			case PCI_PRODUCT_NVIDIA_MCP55_LAN1:
429 			case PCI_PRODUCT_NVIDIA_MCP55_LAN2:
430 				sc->sc_caps = NFE_JUMBO_SUP |
431 					      NFE_40BIT_ADDR |
432 					      NFE_HW_CSUM |
433 					      NFE_HW_VLAN |
434 					      NFE_FIX_EADDR;
435 				break;
436 			case PCI_PRODUCT_NVIDIA_MCP77_LAN1:
437 			case PCI_PRODUCT_NVIDIA_MCP77_LAN2:
438 			case PCI_PRODUCT_NVIDIA_MCP77_LAN3:
439 			case PCI_PRODUCT_NVIDIA_MCP77_LAN4:
440 			case PCI_PRODUCT_NVIDIA_MCP79_LAN1:
441 			case PCI_PRODUCT_NVIDIA_MCP79_LAN2:
442 			case PCI_PRODUCT_NVIDIA_MCP79_LAN3:
443 			case PCI_PRODUCT_NVIDIA_MCP79_LAN4:
444 				sc->sc_caps = NFE_40BIT_ADDR |
445 					      NFE_HW_CSUM |
446 					      NFE_WORDALIGN;
447 				break;
448 			}
449 
450 			device_set_desc(dev, n->desc);
451 			device_set_async_attach(dev, TRUE);
452 			return 0;
453 		}
454 	}
455 	return ENXIO;
456 }
457 
458 static int
459 nfe_attach(device_t dev)
460 {
461 	struct nfe_softc *sc = device_get_softc(dev);
462 	struct ifnet *ifp = &sc->arpcom.ac_if;
463 	uint8_t eaddr[ETHER_ADDR_LEN];
464 	bus_addr_t lowaddr;
465 	int error;
466 
467 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
468 	lwkt_serialize_init(&sc->sc_jbuf_serializer);
469 
470 	/*
471 	 * Initialize sysctl variables
472 	 */
473 	sc->sc_rx_ring_count = nfe_rx_ring_count;
474 	sc->sc_tx_ring_count = nfe_tx_ring_count;
475 	sc->sc_debug = nfe_debug;
476 	if (nfe_imtime < 0) {
477 		sc->sc_flags |= NFE_F_DYN_IM;
478 		sc->sc_imtime = -nfe_imtime;
479 	} else {
480 		sc->sc_imtime = nfe_imtime;
481 	}
482 	sc->sc_irq_enable = NFE_IRQ_ENABLE(sc);
483 
484 	sc->sc_mem_rid = PCIR_BAR(0);
485 
486 	if (sc->sc_caps & NFE_40BIT_ADDR)
487 		sc->rxtxctl_desc = NFE_RXTX_DESC_V3;
488 	else if (sc->sc_caps & NFE_JUMBO_SUP)
489 		sc->rxtxctl_desc = NFE_RXTX_DESC_V2;
490 
491 #ifndef BURN_BRIDGES
492 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
493 		uint32_t mem, irq;
494 
495 		mem = pci_read_config(dev, sc->sc_mem_rid, 4);
496 		irq = pci_read_config(dev, PCIR_INTLINE, 4);
497 
498 		device_printf(dev, "chip is in D%d power mode "
499 		    "-- setting to D0\n", pci_get_powerstate(dev));
500 
501 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
502 
503 		pci_write_config(dev, sc->sc_mem_rid, mem, 4);
504 		pci_write_config(dev, PCIR_INTLINE, irq, 4);
505 	}
506 #endif	/* !BURN_BRIDGE */
507 
508 	/* Enable bus mastering */
509 	pci_enable_busmaster(dev);
510 
511 	/* Allocate IO memory */
512 	sc->sc_mem_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
513 						&sc->sc_mem_rid, RF_ACTIVE);
514 	if (sc->sc_mem_res == NULL) {
515 		device_printf(dev, "could not allocate io memory\n");
516 		return ENXIO;
517 	}
518 	sc->sc_memh = rman_get_bushandle(sc->sc_mem_res);
519 	sc->sc_memt = rman_get_bustag(sc->sc_mem_res);
520 
521 	/* Allocate IRQ */
522 	sc->sc_irq_rid = 0;
523 	sc->sc_irq_res = bus_alloc_resource_any(dev, SYS_RES_IRQ,
524 						&sc->sc_irq_rid,
525 						RF_SHAREABLE | RF_ACTIVE);
526 	if (sc->sc_irq_res == NULL) {
527 		device_printf(dev, "could not allocate irq\n");
528 		error = ENXIO;
529 		goto fail;
530 	}
531 
532 	/* Disable WOL */
533 	NFE_WRITE(sc, NFE_WOL_CTL, 0);
534 
535 	if ((sc->sc_caps & NFE_NO_PWRCTL) == 0)
536 		nfe_powerup(dev);
537 
538 	nfe_get_macaddr(sc, eaddr);
539 
540 	/*
541 	 * Allocate top level DMA tag
542 	 */
543 	if (sc->sc_caps & NFE_40BIT_ADDR)
544 		lowaddr = NFE_BUS_SPACE_MAXADDR;
545 	else
546 		lowaddr = BUS_SPACE_MAXADDR_32BIT;
547 	error = bus_dma_tag_create(NULL,	/* parent */
548 			1, 0,			/* alignment, boundary */
549 			lowaddr,		/* lowaddr */
550 			BUS_SPACE_MAXADDR,	/* highaddr */
551 			NULL, NULL,		/* filter, filterarg */
552 			BUS_SPACE_MAXSIZE_32BIT,/* maxsize */
553 			0,			/* nsegments */
554 			BUS_SPACE_MAXSIZE_32BIT,/* maxsegsize */
555 			0,			/* flags */
556 			&sc->sc_dtag);
557 	if (error) {
558 		device_printf(dev, "could not allocate parent dma tag\n");
559 		goto fail;
560 	}
561 
562 	/*
563 	 * Allocate Tx and Rx rings.
564 	 */
565 	error = nfe_alloc_tx_ring(sc, &sc->txq);
566 	if (error) {
567 		device_printf(dev, "could not allocate Tx ring\n");
568 		goto fail;
569 	}
570 
571 	error = nfe_alloc_rx_ring(sc, &sc->rxq);
572 	if (error) {
573 		device_printf(dev, "could not allocate Rx ring\n");
574 		goto fail;
575 	}
576 
577 	/*
578 	 * Create sysctl tree
579 	 */
580 	sysctl_ctx_init(&sc->sc_sysctl_ctx);
581 	sc->sc_sysctl_tree = SYSCTL_ADD_NODE(&sc->sc_sysctl_ctx,
582 					     SYSCTL_STATIC_CHILDREN(_hw),
583 					     OID_AUTO,
584 					     device_get_nameunit(dev),
585 					     CTLFLAG_RD, 0, "");
586 	if (sc->sc_sysctl_tree == NULL) {
587 		device_printf(dev, "can't add sysctl node\n");
588 		error = ENXIO;
589 		goto fail;
590 	}
591 	SYSCTL_ADD_PROC(&sc->sc_sysctl_ctx,
592 			SYSCTL_CHILDREN(sc->sc_sysctl_tree),
593 			OID_AUTO, "imtimer", CTLTYPE_INT | CTLFLAG_RW,
594 			sc, 0, nfe_sysctl_imtime, "I",
595 			"Interrupt moderation time (usec).  "
596 			"0 to disable interrupt moderation.");
597 	SYSCTL_ADD_INT(&sc->sc_sysctl_ctx,
598 		       SYSCTL_CHILDREN(sc->sc_sysctl_tree), OID_AUTO,
599 		       "rx_ring_count", CTLFLAG_RD, &sc->sc_rx_ring_count,
600 		       0, "RX ring count");
601 	SYSCTL_ADD_INT(&sc->sc_sysctl_ctx,
602 		       SYSCTL_CHILDREN(sc->sc_sysctl_tree), OID_AUTO,
603 		       "tx_ring_count", CTLFLAG_RD, &sc->sc_tx_ring_count,
604 		       0, "TX ring count");
605 	SYSCTL_ADD_INT(&sc->sc_sysctl_ctx,
606 		       SYSCTL_CHILDREN(sc->sc_sysctl_tree), OID_AUTO,
607 		       "debug", CTLFLAG_RW, &sc->sc_debug,
608 		       0, "control debugging printfs");
609 
610 	error = mii_phy_probe(dev, &sc->sc_miibus, nfe_ifmedia_upd,
611 			      nfe_ifmedia_sts);
612 	if (error) {
613 		device_printf(dev, "MII without any phy\n");
614 		goto fail;
615 	}
616 
617 	ifp->if_softc = sc;
618 	ifp->if_mtu = ETHERMTU;
619 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
620 	ifp->if_ioctl = nfe_ioctl;
621 	ifp->if_start = nfe_start;
622 #ifdef IFPOLL_ENABLE
623 	ifp->if_npoll = nfe_npoll;
624 #endif
625 	ifp->if_watchdog = nfe_watchdog;
626 	ifp->if_init = nfe_init;
627 	ifq_set_maxlen(&ifp->if_snd, sc->sc_tx_ring_count);
628 	ifq_set_ready(&ifp->if_snd);
629 
630 	ifp->if_capabilities = IFCAP_VLAN_MTU;
631 
632 	if (sc->sc_caps & NFE_HW_VLAN)
633 		ifp->if_capabilities |= IFCAP_VLAN_HWTAGGING;
634 
635 #ifdef NFE_CSUM
636 	if (sc->sc_caps & NFE_HW_CSUM) {
637 		ifp->if_capabilities |= IFCAP_HWCSUM;
638 		ifp->if_hwassist = NFE_CSUM_FEATURES;
639 	}
640 #else
641 	sc->sc_caps &= ~NFE_HW_CSUM;
642 #endif
643 	ifp->if_capenable = ifp->if_capabilities;
644 
645 	callout_init(&sc->sc_tick_ch);
646 
647 	ether_ifattach(ifp, eaddr, NULL);
648 
649 	ifq_set_cpuid(&ifp->if_snd, rman_get_cpuid(sc->sc_irq_res));
650 
651 #ifdef IFPOLL_ENABLE
652 	ifpoll_compat_setup(&sc->sc_npoll,
653 	    &sc->sc_sysctl_ctx, sc->sc_sysctl_tree, device_get_unit(dev),
654 	    ifp->if_serializer);
655 #endif
656 
657 	error = bus_setup_intr(dev, sc->sc_irq_res, INTR_MPSAFE, nfe_intr, sc,
658 			       &sc->sc_ih, ifp->if_serializer);
659 	if (error) {
660 		device_printf(dev, "could not setup intr\n");
661 		ether_ifdetach(ifp);
662 		goto fail;
663 	}
664 
665 	return 0;
666 fail:
667 	nfe_detach(dev);
668 	return error;
669 }
670 
671 static int
672 nfe_detach(device_t dev)
673 {
674 	struct nfe_softc *sc = device_get_softc(dev);
675 
676 	if (device_is_attached(dev)) {
677 		struct ifnet *ifp = &sc->arpcom.ac_if;
678 
679 		lwkt_serialize_enter(ifp->if_serializer);
680 		nfe_stop(sc);
681 		bus_teardown_intr(dev, sc->sc_irq_res, sc->sc_ih);
682 		lwkt_serialize_exit(ifp->if_serializer);
683 
684 		ether_ifdetach(ifp);
685 	}
686 
687 	if (sc->sc_miibus != NULL)
688 		device_delete_child(dev, sc->sc_miibus);
689 	bus_generic_detach(dev);
690 
691 	if (sc->sc_sysctl_tree != NULL)
692 		sysctl_ctx_free(&sc->sc_sysctl_ctx);
693 
694 	if (sc->sc_irq_res != NULL) {
695 		bus_release_resource(dev, SYS_RES_IRQ, sc->sc_irq_rid,
696 				     sc->sc_irq_res);
697 	}
698 
699 	if (sc->sc_mem_res != NULL) {
700 		bus_release_resource(dev, SYS_RES_MEMORY, sc->sc_mem_rid,
701 				     sc->sc_mem_res);
702 	}
703 
704 	nfe_free_tx_ring(sc, &sc->txq);
705 	nfe_free_rx_ring(sc, &sc->rxq);
706 	if (sc->sc_dtag != NULL)
707 		bus_dma_tag_destroy(sc->sc_dtag);
708 
709 	return 0;
710 }
711 
712 static void
713 nfe_shutdown(device_t dev)
714 {
715 	struct nfe_softc *sc = device_get_softc(dev);
716 	struct ifnet *ifp = &sc->arpcom.ac_if;
717 
718 	lwkt_serialize_enter(ifp->if_serializer);
719 	nfe_stop(sc);
720 	lwkt_serialize_exit(ifp->if_serializer);
721 }
722 
723 static int
724 nfe_suspend(device_t dev)
725 {
726 	struct nfe_softc *sc = device_get_softc(dev);
727 	struct ifnet *ifp = &sc->arpcom.ac_if;
728 
729 	lwkt_serialize_enter(ifp->if_serializer);
730 	nfe_stop(sc);
731 	lwkt_serialize_exit(ifp->if_serializer);
732 
733 	return 0;
734 }
735 
736 static int
737 nfe_resume(device_t dev)
738 {
739 	struct nfe_softc *sc = device_get_softc(dev);
740 	struct ifnet *ifp = &sc->arpcom.ac_if;
741 
742 	lwkt_serialize_enter(ifp->if_serializer);
743 	if (ifp->if_flags & IFF_UP)
744 		nfe_init(sc);
745 	lwkt_serialize_exit(ifp->if_serializer);
746 
747 	return 0;
748 }
749 
750 static void
751 nfe_miibus_statchg(device_t dev)
752 {
753 	struct nfe_softc *sc = device_get_softc(dev);
754 	struct mii_data *mii = device_get_softc(sc->sc_miibus);
755 	uint32_t phy, seed, misc = NFE_MISC1_MAGIC, link = NFE_MEDIA_SET;
756 
757 	ASSERT_SERIALIZED(sc->arpcom.ac_if.if_serializer);
758 
759 	phy = NFE_READ(sc, NFE_PHY_IFACE);
760 	phy &= ~(NFE_PHY_HDX | NFE_PHY_100TX | NFE_PHY_1000T);
761 
762 	seed = NFE_READ(sc, NFE_RNDSEED);
763 	seed &= ~NFE_SEED_MASK;
764 
765 	if ((mii->mii_media_active & IFM_GMASK) == IFM_HDX) {
766 		phy  |= NFE_PHY_HDX;	/* half-duplex */
767 		misc |= NFE_MISC1_HDX;
768 	}
769 
770 	switch (IFM_SUBTYPE(mii->mii_media_active)) {
771 	case IFM_1000_T:	/* full-duplex only */
772 		link |= NFE_MEDIA_1000T;
773 		seed |= NFE_SEED_1000T;
774 		phy  |= NFE_PHY_1000T;
775 		break;
776 	case IFM_100_TX:
777 		link |= NFE_MEDIA_100TX;
778 		seed |= NFE_SEED_100TX;
779 		phy  |= NFE_PHY_100TX;
780 		break;
781 	case IFM_10_T:
782 		link |= NFE_MEDIA_10T;
783 		seed |= NFE_SEED_10T;
784 		break;
785 	}
786 
787 	NFE_WRITE(sc, NFE_RNDSEED, seed);	/* XXX: gigabit NICs only? */
788 
789 	NFE_WRITE(sc, NFE_PHY_IFACE, phy);
790 	NFE_WRITE(sc, NFE_MISC1, misc);
791 	NFE_WRITE(sc, NFE_LINKSPEED, link);
792 }
793 
794 static int
795 nfe_miibus_readreg(device_t dev, int phy, int reg)
796 {
797 	struct nfe_softc *sc = device_get_softc(dev);
798 	uint32_t val;
799 	int ntries;
800 
801 	NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
802 
803 	if (NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY) {
804 		NFE_WRITE(sc, NFE_PHY_CTL, NFE_PHY_BUSY);
805 		DELAY(100);
806 	}
807 
808 	NFE_WRITE(sc, NFE_PHY_CTL, (phy << NFE_PHYADD_SHIFT) | reg);
809 
810 	for (ntries = 0; ntries < 1000; ntries++) {
811 		DELAY(100);
812 		if (!(NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY))
813 			break;
814 	}
815 	if (ntries == 1000) {
816 		DPRINTFN(sc, 2, "timeout waiting for PHY %s\n", "");
817 		return 0;
818 	}
819 
820 	if (NFE_READ(sc, NFE_PHY_STATUS) & NFE_PHY_ERROR) {
821 		DPRINTFN(sc, 2, "could not read PHY %s\n", "");
822 		return 0;
823 	}
824 
825 	val = NFE_READ(sc, NFE_PHY_DATA);
826 	if (val != 0xffffffff && val != 0)
827 		sc->mii_phyaddr = phy;
828 
829 	DPRINTFN(sc, 2, "mii read phy %d reg 0x%x ret 0x%x\n", phy, reg, val);
830 
831 	return val;
832 }
833 
834 static void
835 nfe_miibus_writereg(device_t dev, int phy, int reg, int val)
836 {
837 	struct nfe_softc *sc = device_get_softc(dev);
838 	uint32_t ctl;
839 	int ntries;
840 
841 	NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
842 
843 	if (NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY) {
844 		NFE_WRITE(sc, NFE_PHY_CTL, NFE_PHY_BUSY);
845 		DELAY(100);
846 	}
847 
848 	NFE_WRITE(sc, NFE_PHY_DATA, val);
849 	ctl = NFE_PHY_WRITE | (phy << NFE_PHYADD_SHIFT) | reg;
850 	NFE_WRITE(sc, NFE_PHY_CTL, ctl);
851 
852 	for (ntries = 0; ntries < 1000; ntries++) {
853 		DELAY(100);
854 		if (!(NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY))
855 			break;
856 	}
857 
858 #ifdef NFE_DEBUG
859 	if (ntries == 1000)
860 		DPRINTFN(sc, 2, "could not write to PHY %s\n", "");
861 #endif
862 }
863 
864 #ifdef IFPOLL_ENABLE
865 
866 static void
867 nfe_npoll_compat(struct ifnet *ifp, void *arg __unused, int count __unused)
868 {
869 	struct nfe_softc *sc = ifp->if_softc;
870 
871 	ASSERT_SERIALIZED(ifp->if_serializer);
872 
873 	nfe_rxeof(sc);
874 	nfe_txeof(sc, 1);
875 }
876 
877 static void
878 nfe_disable_intrs(struct nfe_softc *sc)
879 {
880 	/* Disable interrupts */
881 	NFE_WRITE(sc, NFE_IRQ_MASK, 0);
882 	sc->sc_flags &= ~NFE_F_IRQ_TIMER;
883 	sc->sc_npoll.ifpc_stcount = 0;
884 }
885 
886 static void
887 nfe_npoll(struct ifnet *ifp, struct ifpoll_info *info)
888 {
889 	struct nfe_softc *sc = ifp->if_softc;
890 
891 	ASSERT_SERIALIZED(ifp->if_serializer);
892 
893 	if (info != NULL) {
894 		int cpuid = sc->sc_npoll.ifpc_cpuid;
895 
896 		info->ifpi_rx[cpuid].poll_func = nfe_npoll_compat;
897 		info->ifpi_rx[cpuid].arg = NULL;
898 		info->ifpi_rx[cpuid].serializer = ifp->if_serializer;
899 
900 		if (ifp->if_flags & IFF_RUNNING)
901 			nfe_disable_intrs(sc);
902 		ifq_set_cpuid(&ifp->if_snd, cpuid);
903 	} else {
904 		if (ifp->if_flags & IFF_RUNNING)
905 			nfe_enable_intrs(sc);
906 		ifq_set_cpuid(&ifp->if_snd, rman_get_cpuid(sc->sc_irq_res));
907 	}
908 }
909 
910 #endif	/* IFPOLL_ENABLE */
911 
912 static void
913 nfe_intr(void *arg)
914 {
915 	struct nfe_softc *sc = arg;
916 	struct ifnet *ifp = &sc->arpcom.ac_if;
917 	uint32_t r;
918 
919 	r = NFE_READ(sc, NFE_IRQ_STATUS);
920 	if (r == 0)
921 		return;	/* not for us */
922 	NFE_WRITE(sc, NFE_IRQ_STATUS, r);
923 
924 	if (sc->sc_rate_second != time_uptime) {
925 		/*
926 		 * Calculate sc_rate_avg - interrupts per second.
927 		 */
928 		sc->sc_rate_second = time_uptime;
929 		if (sc->sc_rate_avg < sc->sc_rate_acc)
930 			sc->sc_rate_avg = sc->sc_rate_acc;
931 		else
932 			sc->sc_rate_avg = (sc->sc_rate_avg * 3 +
933 					   sc->sc_rate_acc) / 4;
934 		sc->sc_rate_acc = 0;
935 	} else if (sc->sc_rate_avg < sc->sc_rate_acc) {
936 		/*
937 		 * Don't wait for a tick to roll over if we are taking
938 		 * a lot of interrupts.
939 		 */
940 		sc->sc_rate_avg = sc->sc_rate_acc;
941 	}
942 
943 	DPRINTFN(sc, 5, "%s: interrupt register %x\n", __func__, r);
944 
945 	if (r & NFE_IRQ_LINK) {
946 		NFE_READ(sc, NFE_PHY_STATUS);
947 		NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
948 		DPRINTF(sc, "link state changed %s\n", "");
949 	}
950 
951 	if (ifp->if_flags & IFF_RUNNING) {
952 		int ret;
953 		int rate;
954 
955 		/* check Rx ring */
956 		ret = nfe_rxeof(sc);
957 
958 		/* check Tx ring */
959 		ret |= nfe_txeof(sc, 1);
960 
961 		/* update the rate accumulator */
962 		if (ret)
963 			++sc->sc_rate_acc;
964 
965 		if (sc->sc_flags & NFE_F_DYN_IM) {
966 			rate = 1000000 / sc->sc_imtime;
967 			if ((sc->sc_flags & NFE_F_IRQ_TIMER) == 0 &&
968 			    sc->sc_rate_avg > rate) {
969 				/*
970 				 * Use the hardware timer to reduce the
971 				 * interrupt rate if the discrete interrupt
972 				 * rate has exceeded our threshold.
973 				 */
974 				NFE_WRITE(sc, NFE_IRQ_MASK, NFE_IRQ_IMTIMER);
975 				sc->sc_flags |= NFE_F_IRQ_TIMER;
976 			} else if ((sc->sc_flags & NFE_F_IRQ_TIMER) &&
977 				   sc->sc_rate_avg <= rate) {
978 				/*
979 				 * Use discrete TX/RX interrupts if the rate
980 				 * has fallen below our threshold.
981 				 */
982 				NFE_WRITE(sc, NFE_IRQ_MASK, NFE_IRQ_NOIMTIMER);
983 				sc->sc_flags &= ~NFE_F_IRQ_TIMER;
984 
985 				/*
986 				 * Recollect, mainly to avoid the possible race
987 				 * introduced by changing interrupt masks.
988 				 */
989 				nfe_rxeof(sc);
990 				nfe_txeof(sc, 1);
991 			}
992 		}
993 	}
994 }
995 
996 static int
997 nfe_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data, struct ucred *cr)
998 {
999 	struct nfe_softc *sc = ifp->if_softc;
1000 	struct ifreq *ifr = (struct ifreq *)data;
1001 	struct mii_data *mii;
1002 	int error = 0, mask, jumbo_cap;
1003 
1004 	ASSERT_SERIALIZED(ifp->if_serializer);
1005 
1006 	switch (cmd) {
1007 	case SIOCSIFMTU:
1008 		if ((sc->sc_caps & NFE_JUMBO_SUP) && sc->rxq.jbuf != NULL)
1009 			jumbo_cap = 1;
1010 		else
1011 			jumbo_cap = 0;
1012 
1013 		if ((jumbo_cap && ifr->ifr_mtu > NFE_JUMBO_MTU) ||
1014 		    (!jumbo_cap && ifr->ifr_mtu > ETHERMTU)) {
1015 			return EINVAL;
1016 		} else if (ifp->if_mtu != ifr->ifr_mtu) {
1017 			ifp->if_mtu = ifr->ifr_mtu;
1018 			if (ifp->if_flags & IFF_RUNNING)
1019 				nfe_init(sc);
1020 		}
1021 		break;
1022 	case SIOCSIFFLAGS:
1023 		if (ifp->if_flags & IFF_UP) {
1024 			/*
1025 			 * If only the PROMISC or ALLMULTI flag changes, then
1026 			 * don't do a full re-init of the chip, just update
1027 			 * the Rx filter.
1028 			 */
1029 			if ((ifp->if_flags & IFF_RUNNING) &&
1030 			    ((ifp->if_flags ^ sc->sc_if_flags) &
1031 			     (IFF_ALLMULTI | IFF_PROMISC)) != 0) {
1032 				nfe_setmulti(sc);
1033 			} else {
1034 				if (!(ifp->if_flags & IFF_RUNNING))
1035 					nfe_init(sc);
1036 			}
1037 		} else {
1038 			if (ifp->if_flags & IFF_RUNNING)
1039 				nfe_stop(sc);
1040 		}
1041 		sc->sc_if_flags = ifp->if_flags;
1042 		break;
1043 	case SIOCADDMULTI:
1044 	case SIOCDELMULTI:
1045 		if (ifp->if_flags & IFF_RUNNING)
1046 			nfe_setmulti(sc);
1047 		break;
1048 	case SIOCSIFMEDIA:
1049 	case SIOCGIFMEDIA:
1050 		mii = device_get_softc(sc->sc_miibus);
1051 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, cmd);
1052 		break;
1053         case SIOCSIFCAP:
1054 		mask = (ifr->ifr_reqcap ^ ifp->if_capenable) & IFCAP_HWCSUM;
1055 		if (mask && (ifp->if_capabilities & IFCAP_HWCSUM)) {
1056 			ifp->if_capenable ^= mask;
1057 			if (IFCAP_TXCSUM & ifp->if_capenable)
1058 				ifp->if_hwassist = NFE_CSUM_FEATURES;
1059 			else
1060 				ifp->if_hwassist = 0;
1061 
1062 			if (ifp->if_flags & IFF_RUNNING)
1063 				nfe_init(sc);
1064 		}
1065 		break;
1066 	default:
1067 		error = ether_ioctl(ifp, cmd, data);
1068 		break;
1069 	}
1070 	return error;
1071 }
1072 
1073 static int
1074 nfe_rxeof(struct nfe_softc *sc)
1075 {
1076 	struct ifnet *ifp = &sc->arpcom.ac_if;
1077 	struct nfe_rx_ring *ring = &sc->rxq;
1078 	int reap;
1079 
1080 	reap = 0;
1081 	for (;;) {
1082 		struct nfe_rx_data *data = &ring->data[ring->cur];
1083 		struct mbuf *m;
1084 		uint16_t flags;
1085 		int len, error;
1086 
1087 		if (sc->sc_caps & NFE_40BIT_ADDR) {
1088 			struct nfe_desc64 *desc64 = &ring->desc64[ring->cur];
1089 
1090 			flags = le16toh(desc64->flags);
1091 			len = le16toh(desc64->length) & 0x3fff;
1092 		} else {
1093 			struct nfe_desc32 *desc32 = &ring->desc32[ring->cur];
1094 
1095 			flags = le16toh(desc32->flags);
1096 			len = le16toh(desc32->length) & 0x3fff;
1097 		}
1098 
1099 		if (flags & NFE_RX_READY)
1100 			break;
1101 
1102 		reap = 1;
1103 
1104 		if ((sc->sc_caps & (NFE_JUMBO_SUP | NFE_40BIT_ADDR)) == 0) {
1105 			if (!(flags & NFE_RX_VALID_V1))
1106 				goto skip;
1107 
1108 			if ((flags & NFE_RX_FIXME_V1) == NFE_RX_FIXME_V1) {
1109 				flags &= ~NFE_RX_ERROR;
1110 				len--;	/* fix buffer length */
1111 			}
1112 		} else {
1113 			if (!(flags & NFE_RX_VALID_V2))
1114 				goto skip;
1115 
1116 			if ((flags & NFE_RX_FIXME_V2) == NFE_RX_FIXME_V2) {
1117 				flags &= ~NFE_RX_ERROR;
1118 				len--;	/* fix buffer length */
1119 			}
1120 		}
1121 
1122 		if (flags & NFE_RX_ERROR) {
1123 			IFNET_STAT_INC(ifp, ierrors, 1);
1124 			goto skip;
1125 		}
1126 
1127 		m = data->m;
1128 
1129 		if (sc->sc_flags & NFE_F_USE_JUMBO)
1130 			error = nfe_newbuf_jumbo(sc, ring, ring->cur, 0);
1131 		else
1132 			error = nfe_newbuf_std(sc, ring, ring->cur, 0);
1133 		if (error) {
1134 			IFNET_STAT_INC(ifp, ierrors, 1);
1135 			goto skip;
1136 		}
1137 
1138 		/* finalize mbuf */
1139 		m->m_pkthdr.len = m->m_len = len;
1140 		m->m_pkthdr.rcvif = ifp;
1141 
1142 		if ((ifp->if_capenable & IFCAP_RXCSUM) &&
1143 		    (flags & NFE_RX_CSUMOK)) {
1144 			if (flags & NFE_RX_IP_CSUMOK_V2) {
1145 				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED |
1146 							  CSUM_IP_VALID;
1147 			}
1148 
1149 			if (flags &
1150 			    (NFE_RX_UDP_CSUMOK_V2 | NFE_RX_TCP_CSUMOK_V2)) {
1151 				m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
1152 							  CSUM_PSEUDO_HDR |
1153 							  CSUM_FRAG_NOT_CHECKED;
1154 				m->m_pkthdr.csum_data = 0xffff;
1155 			}
1156 		}
1157 
1158 		IFNET_STAT_INC(ifp, ipackets, 1);
1159 		ifp->if_input(ifp, m);
1160 skip:
1161 		nfe_set_ready_rxdesc(sc, ring, ring->cur);
1162 		sc->rxq.cur = (sc->rxq.cur + 1) % sc->sc_rx_ring_count;
1163 	}
1164 	return reap;
1165 }
1166 
1167 static int
1168 nfe_txeof(struct nfe_softc *sc, int start)
1169 {
1170 	struct ifnet *ifp = &sc->arpcom.ac_if;
1171 	struct nfe_tx_ring *ring = &sc->txq;
1172 	struct nfe_tx_data *data = NULL;
1173 
1174 	while (ring->next != ring->cur) {
1175 		uint16_t flags;
1176 
1177 		if (sc->sc_caps & NFE_40BIT_ADDR)
1178 			flags = le16toh(ring->desc64[ring->next].flags);
1179 		else
1180 			flags = le16toh(ring->desc32[ring->next].flags);
1181 
1182 		if (flags & NFE_TX_VALID)
1183 			break;
1184 
1185 		data = &ring->data[ring->next];
1186 
1187 		if ((sc->sc_caps & (NFE_JUMBO_SUP | NFE_40BIT_ADDR)) == 0) {
1188 			if (!(flags & NFE_TX_LASTFRAG_V1) && data->m == NULL)
1189 				goto skip;
1190 
1191 			if ((flags & NFE_TX_ERROR_V1) != 0) {
1192 				if_printf(ifp, "tx v1 error 0x%4b\n", flags,
1193 					  NFE_V1_TXERR);
1194 				IFNET_STAT_INC(ifp, oerrors, 1);
1195 			} else {
1196 				IFNET_STAT_INC(ifp, opackets, 1);
1197 			}
1198 		} else {
1199 			if (!(flags & NFE_TX_LASTFRAG_V2) && data->m == NULL)
1200 				goto skip;
1201 
1202 			if ((flags & NFE_TX_ERROR_V2) != 0) {
1203 				if_printf(ifp, "tx v2 error 0x%4b\n", flags,
1204 					  NFE_V2_TXERR);
1205 				IFNET_STAT_INC(ifp, oerrors, 1);
1206 			} else {
1207 				IFNET_STAT_INC(ifp, opackets, 1);
1208 			}
1209 		}
1210 
1211 		if (data->m == NULL) {	/* should not get there */
1212 			if_printf(ifp,
1213 				  "last fragment bit w/o associated mbuf!\n");
1214 			goto skip;
1215 		}
1216 
1217 		/* last fragment of the mbuf chain transmitted */
1218 		bus_dmamap_unload(ring->data_tag, data->map);
1219 		m_freem(data->m);
1220 		data->m = NULL;
1221 skip:
1222 		ring->queued--;
1223 		KKASSERT(ring->queued >= 0);
1224 		ring->next = (ring->next + 1) % sc->sc_tx_ring_count;
1225 	}
1226 
1227 	if (sc->sc_tx_ring_count - ring->queued >=
1228 	    sc->sc_tx_spare + NFE_NSEG_RSVD)
1229 		ifq_clr_oactive(&ifp->if_snd);
1230 
1231 	if (ring->queued == 0)
1232 		ifp->if_timer = 0;
1233 
1234 	if (start && !ifq_is_empty(&ifp->if_snd))
1235 		if_devstart(ifp);
1236 
1237 	if (data != NULL)
1238 		return 1;
1239 	else
1240 		return 0;
1241 }
1242 
1243 static int
1244 nfe_encap(struct nfe_softc *sc, struct nfe_tx_ring *ring, struct mbuf *m0)
1245 {
1246 	bus_dma_segment_t segs[NFE_MAX_SCATTER];
1247 	struct nfe_tx_data *data, *data_map;
1248 	bus_dmamap_t map;
1249 	struct nfe_desc64 *desc64 = NULL;
1250 	struct nfe_desc32 *desc32 = NULL;
1251 	uint16_t flags = 0;
1252 	uint32_t vtag = 0;
1253 	int error, i, j, maxsegs, nsegs;
1254 
1255 	data = &ring->data[ring->cur];
1256 	map = data->map;
1257 	data_map = data;	/* Remember who owns the DMA map */
1258 
1259 	maxsegs = (sc->sc_tx_ring_count - ring->queued) - NFE_NSEG_RSVD;
1260 	if (maxsegs > NFE_MAX_SCATTER)
1261 		maxsegs = NFE_MAX_SCATTER;
1262 	KASSERT(maxsegs >= sc->sc_tx_spare,
1263 		("not enough segments %d,%d", maxsegs, sc->sc_tx_spare));
1264 
1265 	error = bus_dmamap_load_mbuf_defrag(ring->data_tag, map, &m0,
1266 			segs, maxsegs, &nsegs, BUS_DMA_NOWAIT);
1267 	if (error)
1268 		goto back;
1269 	bus_dmamap_sync(ring->data_tag, map, BUS_DMASYNC_PREWRITE);
1270 
1271 	error = 0;
1272 
1273 	/* setup h/w VLAN tagging */
1274 	if (m0->m_flags & M_VLANTAG)
1275 		vtag = m0->m_pkthdr.ether_vlantag;
1276 
1277 	if (sc->arpcom.ac_if.if_capenable & IFCAP_TXCSUM) {
1278 		if (m0->m_pkthdr.csum_flags & CSUM_IP)
1279 			flags |= NFE_TX_IP_CSUM;
1280 		if (m0->m_pkthdr.csum_flags & (CSUM_TCP | CSUM_UDP))
1281 			flags |= NFE_TX_TCP_CSUM;
1282 	}
1283 
1284 	/*
1285 	 * XXX urm. somebody is unaware of how hardware works.  You
1286 	 * absolutely CANNOT set NFE_TX_VALID on the next descriptor in
1287 	 * the ring until the entire chain is actually *VALID*.  Otherwise
1288 	 * the hardware may encounter a partially initialized chain that
1289 	 * is marked as being ready to go when it in fact is not ready to
1290 	 * go.
1291 	 */
1292 
1293 	for (i = 0; i < nsegs; i++) {
1294 		j = (ring->cur + i) % sc->sc_tx_ring_count;
1295 		data = &ring->data[j];
1296 
1297 		if (sc->sc_caps & NFE_40BIT_ADDR) {
1298 			desc64 = &ring->desc64[j];
1299 			desc64->physaddr[0] =
1300 			    htole32(NFE_ADDR_HI(segs[i].ds_addr));
1301 			desc64->physaddr[1] =
1302 			    htole32(NFE_ADDR_LO(segs[i].ds_addr));
1303 			desc64->length = htole16(segs[i].ds_len - 1);
1304 			desc64->vtag = htole32(vtag);
1305 			desc64->flags = htole16(flags);
1306 		} else {
1307 			desc32 = &ring->desc32[j];
1308 			desc32->physaddr = htole32(segs[i].ds_addr);
1309 			desc32->length = htole16(segs[i].ds_len - 1);
1310 			desc32->flags = htole16(flags);
1311 		}
1312 
1313 		/* csum flags and vtag belong to the first fragment only */
1314 		flags &= ~(NFE_TX_IP_CSUM | NFE_TX_TCP_CSUM);
1315 		vtag = 0;
1316 
1317 		ring->queued++;
1318 		KKASSERT(ring->queued <= sc->sc_tx_ring_count);
1319 	}
1320 
1321 	/* the whole mbuf chain has been DMA mapped, fix last descriptor */
1322 	if (sc->sc_caps & NFE_40BIT_ADDR) {
1323 		desc64->flags |= htole16(NFE_TX_LASTFRAG_V2);
1324 	} else {
1325 		if (sc->sc_caps & NFE_JUMBO_SUP)
1326 			flags = NFE_TX_LASTFRAG_V2;
1327 		else
1328 			flags = NFE_TX_LASTFRAG_V1;
1329 		desc32->flags |= htole16(flags);
1330 	}
1331 
1332 	/*
1333 	 * Set NFE_TX_VALID backwards so the hardware doesn't see the
1334 	 * whole mess until the first descriptor in the map is flagged.
1335 	 */
1336 	for (i = nsegs - 1; i >= 0; --i) {
1337 		j = (ring->cur + i) % sc->sc_tx_ring_count;
1338 		if (sc->sc_caps & NFE_40BIT_ADDR) {
1339 			desc64 = &ring->desc64[j];
1340 			desc64->flags |= htole16(NFE_TX_VALID);
1341 		} else {
1342 			desc32 = &ring->desc32[j];
1343 			desc32->flags |= htole16(NFE_TX_VALID);
1344 		}
1345 	}
1346 	ring->cur = (ring->cur + nsegs) % sc->sc_tx_ring_count;
1347 
1348 	/* Exchange DMA map */
1349 	data_map->map = data->map;
1350 	data->map = map;
1351 	data->m = m0;
1352 back:
1353 	if (error)
1354 		m_freem(m0);
1355 	return error;
1356 }
1357 
1358 static void
1359 nfe_start(struct ifnet *ifp, struct ifaltq_subque *ifsq)
1360 {
1361 	struct nfe_softc *sc = ifp->if_softc;
1362 	struct nfe_tx_ring *ring = &sc->txq;
1363 	int count = 0, oactive = 0;
1364 	struct mbuf *m0;
1365 
1366 	ASSERT_ALTQ_SQ_DEFAULT(ifp, ifsq);
1367 	ASSERT_SERIALIZED(ifp->if_serializer);
1368 
1369 	if ((ifp->if_flags & IFF_RUNNING) == 0 || ifq_is_oactive(&ifp->if_snd))
1370 		return;
1371 
1372 	for (;;) {
1373 		int error;
1374 
1375 		if (sc->sc_tx_ring_count - ring->queued <
1376 		    sc->sc_tx_spare + NFE_NSEG_RSVD) {
1377 			if (oactive) {
1378 				ifq_set_oactive(&ifp->if_snd);
1379 				break;
1380 			}
1381 
1382 			nfe_txeof(sc, 0);
1383 			oactive = 1;
1384 			continue;
1385 		}
1386 
1387 		m0 = ifq_dequeue(&ifp->if_snd);
1388 		if (m0 == NULL)
1389 			break;
1390 
1391 		ETHER_BPF_MTAP(ifp, m0);
1392 
1393 		error = nfe_encap(sc, ring, m0);
1394 		if (error) {
1395 			IFNET_STAT_INC(ifp, oerrors, 1);
1396 			if (error == EFBIG) {
1397 				if (oactive) {
1398 					ifq_set_oactive(&ifp->if_snd);
1399 					break;
1400 				}
1401 				nfe_txeof(sc, 0);
1402 				oactive = 1;
1403 			}
1404 			continue;
1405 		} else {
1406 			oactive = 0;
1407 		}
1408 		++count;
1409 
1410 		/*
1411 		 * NOTE:
1412 		 * `m0' may be freed in nfe_encap(), so
1413 		 * it should not be touched any more.
1414 		 */
1415 	}
1416 
1417 	if (count == 0)	/* nothing sent */
1418 		return;
1419 
1420 	/* Kick Tx */
1421 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_KICKTX | sc->rxtxctl);
1422 
1423 	/*
1424 	 * Set a timeout in case the chip goes out to lunch.
1425 	 */
1426 	ifp->if_timer = 5;
1427 }
1428 
1429 static void
1430 nfe_watchdog(struct ifnet *ifp)
1431 {
1432 	struct nfe_softc *sc = ifp->if_softc;
1433 
1434 	ASSERT_SERIALIZED(ifp->if_serializer);
1435 
1436 	if (ifp->if_flags & IFF_RUNNING) {
1437 		if_printf(ifp, "watchdog timeout - lost interrupt recovered\n");
1438 		nfe_txeof(sc, 1);
1439 		return;
1440 	}
1441 
1442 	if_printf(ifp, "watchdog timeout\n");
1443 
1444 	nfe_init(ifp->if_softc);
1445 
1446 	IFNET_STAT_INC(ifp, oerrors, 1);
1447 }
1448 
1449 static void
1450 nfe_init(void *xsc)
1451 {
1452 	struct nfe_softc *sc = xsc;
1453 	struct ifnet *ifp = &sc->arpcom.ac_if;
1454 	uint32_t tmp;
1455 	int error;
1456 
1457 	ASSERT_SERIALIZED(ifp->if_serializer);
1458 
1459 	nfe_stop(sc);
1460 
1461 	if ((sc->sc_caps & NFE_NO_PWRCTL) == 0)
1462 		nfe_mac_reset(sc);
1463 
1464 	/*
1465 	 * NOTE:
1466 	 * Switching between jumbo frames and normal frames should
1467 	 * be done _after_ nfe_stop() but _before_ nfe_init_rx_ring().
1468 	 */
1469 	if (ifp->if_mtu > ETHERMTU) {
1470 		sc->sc_flags |= NFE_F_USE_JUMBO;
1471 		sc->rxq.bufsz = NFE_JBYTES;
1472 		sc->sc_tx_spare = NFE_NSEG_SPARE_JUMBO;
1473 		if (bootverbose)
1474 			if_printf(ifp, "use jumbo frames\n");
1475 	} else {
1476 		sc->sc_flags &= ~NFE_F_USE_JUMBO;
1477 		sc->rxq.bufsz = MCLBYTES;
1478 		sc->sc_tx_spare = NFE_NSEG_SPARE;
1479 		if (bootverbose)
1480 			if_printf(ifp, "use non-jumbo frames\n");
1481 	}
1482 
1483 	error = nfe_init_tx_ring(sc, &sc->txq);
1484 	if (error) {
1485 		nfe_stop(sc);
1486 		return;
1487 	}
1488 
1489 	error = nfe_init_rx_ring(sc, &sc->rxq);
1490 	if (error) {
1491 		nfe_stop(sc);
1492 		return;
1493 	}
1494 
1495 	NFE_WRITE(sc, NFE_TX_POLL, 0);
1496 	NFE_WRITE(sc, NFE_STATUS, 0);
1497 
1498 	sc->rxtxctl = NFE_RXTX_BIT2 | sc->rxtxctl_desc;
1499 
1500 	if (ifp->if_capenable & IFCAP_RXCSUM)
1501 		sc->rxtxctl |= NFE_RXTX_RXCSUM;
1502 
1503 	/*
1504 	 * Although the adapter is capable of stripping VLAN tags from received
1505 	 * frames (NFE_RXTX_VTAG_STRIP), we do not enable this functionality on
1506 	 * purpose.  This will be done in software by our network stack.
1507 	 */
1508 	if (sc->sc_caps & NFE_HW_VLAN)
1509 		sc->rxtxctl |= NFE_RXTX_VTAG_INSERT;
1510 
1511 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_RESET | sc->rxtxctl);
1512 	DELAY(10);
1513 	NFE_WRITE(sc, NFE_RXTX_CTL, sc->rxtxctl);
1514 
1515 	if (sc->sc_caps & NFE_HW_VLAN)
1516 		NFE_WRITE(sc, NFE_VTAG_CTL, NFE_VTAG_ENABLE);
1517 
1518 	NFE_WRITE(sc, NFE_SETUP_R6, 0);
1519 
1520 	/* set MAC address */
1521 	nfe_set_macaddr(sc, sc->arpcom.ac_enaddr);
1522 
1523 	/* tell MAC where rings are in memory */
1524 	if (sc->sc_caps & NFE_40BIT_ADDR) {
1525 		NFE_WRITE(sc, NFE_RX_RING_ADDR_HI,
1526 			  NFE_ADDR_HI(sc->rxq.physaddr));
1527 	}
1528 	NFE_WRITE(sc, NFE_RX_RING_ADDR_LO, NFE_ADDR_LO(sc->rxq.physaddr));
1529 
1530 	if (sc->sc_caps & NFE_40BIT_ADDR) {
1531 		NFE_WRITE(sc, NFE_TX_RING_ADDR_HI,
1532 			  NFE_ADDR_HI(sc->txq.physaddr));
1533 	}
1534 	NFE_WRITE(sc, NFE_TX_RING_ADDR_LO, NFE_ADDR_LO(sc->txq.physaddr));
1535 
1536 	NFE_WRITE(sc, NFE_RING_SIZE,
1537 	    (sc->sc_rx_ring_count - 1) << 16 |
1538 	    (sc->sc_tx_ring_count - 1));
1539 
1540 	NFE_WRITE(sc, NFE_RXBUFSZ, sc->rxq.bufsz);
1541 
1542 	/* force MAC to wakeup */
1543 	tmp = NFE_READ(sc, NFE_PWR_STATE);
1544 	NFE_WRITE(sc, NFE_PWR_STATE, tmp | NFE_PWR_WAKEUP);
1545 	DELAY(10);
1546 	tmp = NFE_READ(sc, NFE_PWR_STATE);
1547 	NFE_WRITE(sc, NFE_PWR_STATE, tmp | NFE_PWR_VALID);
1548 
1549 	NFE_WRITE(sc, NFE_SETUP_R1, NFE_R1_MAGIC);
1550 	NFE_WRITE(sc, NFE_SETUP_R2, NFE_R2_MAGIC);
1551 	NFE_WRITE(sc, NFE_SETUP_R6, NFE_R6_MAGIC);
1552 
1553 	/* update MAC knowledge of PHY; generates a NFE_IRQ_LINK interrupt */
1554 	NFE_WRITE(sc, NFE_STATUS, sc->mii_phyaddr << 24 | NFE_STATUS_MAGIC);
1555 
1556 	NFE_WRITE(sc, NFE_SETUP_R4, NFE_R4_MAGIC);
1557 
1558 	sc->rxtxctl &= ~NFE_RXTX_BIT2;
1559 	NFE_WRITE(sc, NFE_RXTX_CTL, sc->rxtxctl);
1560 	DELAY(10);
1561 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_BIT1 | sc->rxtxctl);
1562 
1563 	/* set Rx filter */
1564 	nfe_setmulti(sc);
1565 
1566 	nfe_ifmedia_upd(ifp);
1567 
1568 	/* enable Rx */
1569 	NFE_WRITE(sc, NFE_RX_CTL, NFE_RX_START);
1570 
1571 	/* enable Tx */
1572 	NFE_WRITE(sc, NFE_TX_CTL, NFE_TX_START);
1573 
1574 	NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
1575 
1576 #ifdef IFPOLL_ENABLE
1577 	if (ifp->if_flags & IFF_NPOLLING)
1578 		nfe_disable_intrs(sc);
1579 	else
1580 #endif
1581 	nfe_enable_intrs(sc);
1582 
1583 	callout_reset(&sc->sc_tick_ch, hz, nfe_tick, sc);
1584 
1585 	ifp->if_flags |= IFF_RUNNING;
1586 	ifq_clr_oactive(&ifp->if_snd);
1587 
1588 	/*
1589 	 * If we had stuff in the tx ring before its all cleaned out now
1590 	 * so we are not going to get an interrupt, jump-start any pending
1591 	 * output.
1592 	 */
1593 	if (!ifq_is_empty(&ifp->if_snd))
1594 		if_devstart(ifp);
1595 }
1596 
1597 static void
1598 nfe_stop(struct nfe_softc *sc)
1599 {
1600 	struct ifnet *ifp = &sc->arpcom.ac_if;
1601 	uint32_t rxtxctl = sc->rxtxctl_desc | NFE_RXTX_BIT2;
1602 	int i;
1603 
1604 	ASSERT_SERIALIZED(ifp->if_serializer);
1605 
1606 	callout_stop(&sc->sc_tick_ch);
1607 
1608 	ifp->if_timer = 0;
1609 	ifp->if_flags &= ~IFF_RUNNING;
1610 	ifq_clr_oactive(&ifp->if_snd);
1611 	sc->sc_flags &= ~NFE_F_IRQ_TIMER;
1612 
1613 #define WAITMAX	50000
1614 
1615 	/*
1616 	 * Abort Tx
1617 	 */
1618 	NFE_WRITE(sc, NFE_TX_CTL, 0);
1619 	for (i = 0; i < WAITMAX; ++i) {
1620 		DELAY(100);
1621 		if ((NFE_READ(sc, NFE_TX_STATUS) & NFE_TX_STATUS_BUSY) == 0)
1622 			break;
1623 	}
1624 	if (i == WAITMAX)
1625 		if_printf(ifp, "can't stop TX\n");
1626 	DELAY(100);
1627 
1628 	/*
1629 	 * Disable Rx
1630 	 */
1631 	NFE_WRITE(sc, NFE_RX_CTL, 0);
1632 	for (i = 0; i < WAITMAX; ++i) {
1633 		DELAY(100);
1634 		if ((NFE_READ(sc, NFE_RX_STATUS) & NFE_RX_STATUS_BUSY) == 0)
1635 			break;
1636 	}
1637 	if (i == WAITMAX)
1638 		if_printf(ifp, "can't stop RX\n");
1639 	DELAY(100);
1640 
1641 #undef WAITMAX
1642 
1643 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_RESET | rxtxctl);
1644 	DELAY(10);
1645 	NFE_WRITE(sc, NFE_RXTX_CTL, rxtxctl);
1646 
1647 	/* Disable interrupts */
1648 	NFE_WRITE(sc, NFE_IRQ_MASK, 0);
1649 
1650 	/* Reset Tx and Rx rings */
1651 	nfe_reset_tx_ring(sc, &sc->txq);
1652 	nfe_reset_rx_ring(sc, &sc->rxq);
1653 }
1654 
1655 static int
1656 nfe_alloc_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1657 {
1658 	int i, j, error, descsize;
1659 	bus_dmamem_t dmem;
1660 	void **desc;
1661 
1662 	if (sc->sc_caps & NFE_40BIT_ADDR) {
1663 		desc = (void *)&ring->desc64;
1664 		descsize = sizeof(struct nfe_desc64);
1665 	} else {
1666 		desc = (void *)&ring->desc32;
1667 		descsize = sizeof(struct nfe_desc32);
1668 	}
1669 
1670 	ring->bufsz = MCLBYTES;
1671 	ring->cur = ring->next = 0;
1672 
1673 	error = bus_dmamem_coherent(sc->sc_dtag, PAGE_SIZE, 0,
1674 				    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
1675 				    sc->sc_rx_ring_count * descsize,
1676 				    BUS_DMA_WAITOK | BUS_DMA_ZERO, &dmem);
1677 	if (error) {
1678 		if_printf(&sc->arpcom.ac_if,
1679 			  "could not create RX desc ring\n");
1680 		return error;
1681 	}
1682 	ring->tag = dmem.dmem_tag;
1683 	ring->map = dmem.dmem_map;
1684 	*desc = dmem.dmem_addr;
1685 	ring->physaddr = dmem.dmem_busaddr;
1686 
1687 	if (sc->sc_caps & NFE_JUMBO_SUP) {
1688 		ring->jbuf =
1689 		kmalloc(sizeof(struct nfe_jbuf) * NFE_JPOOL_COUNT(sc),
1690 			M_DEVBUF, M_WAITOK | M_ZERO);
1691 
1692 		error = nfe_jpool_alloc(sc, ring);
1693 		if (error) {
1694 			if_printf(&sc->arpcom.ac_if,
1695 				  "could not allocate jumbo frames\n");
1696 			kfree(ring->jbuf, M_DEVBUF);
1697 			ring->jbuf = NULL;
1698 			/* Allow jumbo frame allocation to fail */
1699 		}
1700 	}
1701 
1702 	ring->data = kmalloc(sizeof(struct nfe_rx_data) * sc->sc_rx_ring_count,
1703 			     M_DEVBUF, M_WAITOK | M_ZERO);
1704 
1705 	error = bus_dma_tag_create(sc->sc_dtag, 1, 0,
1706 				   BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
1707 				   NULL, NULL,
1708 				   MCLBYTES, 1, MCLBYTES,
1709 				   BUS_DMA_ALLOCNOW | BUS_DMA_WAITOK,
1710 				   &ring->data_tag);
1711 	if (error) {
1712 		if_printf(&sc->arpcom.ac_if,
1713 			  "could not create RX mbuf DMA tag\n");
1714 		return error;
1715 	}
1716 
1717 	/* Create a spare RX mbuf DMA map */
1718 	error = bus_dmamap_create(ring->data_tag, BUS_DMA_WAITOK,
1719 				  &ring->data_tmpmap);
1720 	if (error) {
1721 		if_printf(&sc->arpcom.ac_if,
1722 			  "could not create spare RX mbuf DMA map\n");
1723 		bus_dma_tag_destroy(ring->data_tag);
1724 		ring->data_tag = NULL;
1725 		return error;
1726 	}
1727 
1728 	for (i = 0; i < sc->sc_rx_ring_count; i++) {
1729 		error = bus_dmamap_create(ring->data_tag, BUS_DMA_WAITOK,
1730 					  &ring->data[i].map);
1731 		if (error) {
1732 			if_printf(&sc->arpcom.ac_if,
1733 				  "could not create %dth RX mbuf DMA mapn", i);
1734 			goto fail;
1735 		}
1736 	}
1737 	return 0;
1738 fail:
1739 	for (j = 0; j < i; ++j)
1740 		bus_dmamap_destroy(ring->data_tag, ring->data[i].map);
1741 	bus_dmamap_destroy(ring->data_tag, ring->data_tmpmap);
1742 	bus_dma_tag_destroy(ring->data_tag);
1743 	ring->data_tag = NULL;
1744 	return error;
1745 }
1746 
1747 static void
1748 nfe_reset_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1749 {
1750 	int i;
1751 
1752 	for (i = 0; i < sc->sc_rx_ring_count; i++) {
1753 		struct nfe_rx_data *data = &ring->data[i];
1754 
1755 		if (data->m != NULL) {
1756 			if ((sc->sc_flags & NFE_F_USE_JUMBO) == 0)
1757 				bus_dmamap_unload(ring->data_tag, data->map);
1758 			m_freem(data->m);
1759 			data->m = NULL;
1760 		}
1761 	}
1762 
1763 	ring->cur = ring->next = 0;
1764 }
1765 
1766 static int
1767 nfe_init_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1768 {
1769 	int i;
1770 
1771 	for (i = 0; i < sc->sc_rx_ring_count; ++i) {
1772 		int error;
1773 
1774 		/* XXX should use a function pointer */
1775 		if (sc->sc_flags & NFE_F_USE_JUMBO)
1776 			error = nfe_newbuf_jumbo(sc, ring, i, 1);
1777 		else
1778 			error = nfe_newbuf_std(sc, ring, i, 1);
1779 		if (error) {
1780 			if_printf(&sc->arpcom.ac_if,
1781 				  "could not allocate RX buffer\n");
1782 			return error;
1783 		}
1784 		nfe_set_ready_rxdesc(sc, ring, i);
1785 	}
1786 	return 0;
1787 }
1788 
1789 static void
1790 nfe_free_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1791 {
1792 	if (ring->data_tag != NULL) {
1793 		struct nfe_rx_data *data;
1794 		int i;
1795 
1796 		for (i = 0; i < sc->sc_rx_ring_count; i++) {
1797 			data = &ring->data[i];
1798 
1799 			if (data->m != NULL) {
1800 				bus_dmamap_unload(ring->data_tag, data->map);
1801 				m_freem(data->m);
1802 			}
1803 			bus_dmamap_destroy(ring->data_tag, data->map);
1804 		}
1805 		bus_dmamap_destroy(ring->data_tag, ring->data_tmpmap);
1806 		bus_dma_tag_destroy(ring->data_tag);
1807 	}
1808 
1809 	nfe_jpool_free(sc, ring);
1810 
1811 	if (ring->jbuf != NULL)
1812 		kfree(ring->jbuf, M_DEVBUF);
1813 	if (ring->data != NULL)
1814 		kfree(ring->data, M_DEVBUF);
1815 
1816 	if (ring->tag != NULL) {
1817 		void *desc;
1818 
1819 		if (sc->sc_caps & NFE_40BIT_ADDR)
1820 			desc = ring->desc64;
1821 		else
1822 			desc = ring->desc32;
1823 
1824 		bus_dmamap_unload(ring->tag, ring->map);
1825 		bus_dmamem_free(ring->tag, desc, ring->map);
1826 		bus_dma_tag_destroy(ring->tag);
1827 	}
1828 }
1829 
1830 static struct nfe_jbuf *
1831 nfe_jalloc(struct nfe_softc *sc)
1832 {
1833 	struct ifnet *ifp = &sc->arpcom.ac_if;
1834 	struct nfe_jbuf *jbuf;
1835 
1836 	lwkt_serialize_enter(&sc->sc_jbuf_serializer);
1837 
1838 	jbuf = SLIST_FIRST(&sc->rxq.jfreelist);
1839 	if (jbuf != NULL) {
1840 		SLIST_REMOVE_HEAD(&sc->rxq.jfreelist, jnext);
1841 		jbuf->inuse = 1;
1842 	} else {
1843 		if_printf(ifp, "no free jumbo buffer\n");
1844 	}
1845 
1846 	lwkt_serialize_exit(&sc->sc_jbuf_serializer);
1847 
1848 	return jbuf;
1849 }
1850 
1851 static void
1852 nfe_jfree(void *arg)
1853 {
1854 	struct nfe_jbuf *jbuf = arg;
1855 	struct nfe_softc *sc = jbuf->sc;
1856 	struct nfe_rx_ring *ring = jbuf->ring;
1857 
1858 	if (&ring->jbuf[jbuf->slot] != jbuf)
1859 		panic("%s: free wrong jumbo buffer", __func__);
1860 	else if (jbuf->inuse == 0)
1861 		panic("%s: jumbo buffer already freed", __func__);
1862 
1863 	lwkt_serialize_enter(&sc->sc_jbuf_serializer);
1864 	atomic_subtract_int(&jbuf->inuse, 1);
1865 	if (jbuf->inuse == 0)
1866 		SLIST_INSERT_HEAD(&ring->jfreelist, jbuf, jnext);
1867 	lwkt_serialize_exit(&sc->sc_jbuf_serializer);
1868 }
1869 
1870 static void
1871 nfe_jref(void *arg)
1872 {
1873 	struct nfe_jbuf *jbuf = arg;
1874 	struct nfe_rx_ring *ring = jbuf->ring;
1875 
1876 	if (&ring->jbuf[jbuf->slot] != jbuf)
1877 		panic("%s: ref wrong jumbo buffer", __func__);
1878 	else if (jbuf->inuse == 0)
1879 		panic("%s: jumbo buffer already freed", __func__);
1880 
1881 	atomic_add_int(&jbuf->inuse, 1);
1882 }
1883 
1884 static int
1885 nfe_jpool_alloc(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1886 {
1887 	struct nfe_jbuf *jbuf;
1888 	bus_dmamem_t dmem;
1889 	bus_addr_t physaddr;
1890 	caddr_t buf;
1891 	int i, error;
1892 
1893 	/*
1894 	 * Allocate a big chunk of DMA'able memory.
1895 	 */
1896 	error = bus_dmamem_coherent(sc->sc_dtag, PAGE_SIZE, 0,
1897 				    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
1898 				    NFE_JPOOL_SIZE(sc),
1899 				    BUS_DMA_WAITOK, &dmem);
1900 	if (error) {
1901 		if_printf(&sc->arpcom.ac_if,
1902 			  "could not create jumbo buffer\n");
1903 		return error;
1904 	}
1905 	ring->jtag = dmem.dmem_tag;
1906 	ring->jmap = dmem.dmem_map;
1907 	ring->jpool = dmem.dmem_addr;
1908 	physaddr = dmem.dmem_busaddr;
1909 
1910 	/* ..and split it into 9KB chunks */
1911 	SLIST_INIT(&ring->jfreelist);
1912 
1913 	buf = ring->jpool;
1914 	for (i = 0; i < NFE_JPOOL_COUNT(sc); i++) {
1915 		jbuf = &ring->jbuf[i];
1916 
1917 		jbuf->sc = sc;
1918 		jbuf->ring = ring;
1919 		jbuf->inuse = 0;
1920 		jbuf->slot = i;
1921 		jbuf->buf = buf;
1922 		jbuf->physaddr = physaddr;
1923 
1924 		SLIST_INSERT_HEAD(&ring->jfreelist, jbuf, jnext);
1925 
1926 		buf += NFE_JBYTES;
1927 		physaddr += NFE_JBYTES;
1928 	}
1929 
1930 	return 0;
1931 }
1932 
1933 static void
1934 nfe_jpool_free(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1935 {
1936 	if (ring->jtag != NULL) {
1937 		bus_dmamap_unload(ring->jtag, ring->jmap);
1938 		bus_dmamem_free(ring->jtag, ring->jpool, ring->jmap);
1939 		bus_dma_tag_destroy(ring->jtag);
1940 	}
1941 }
1942 
1943 static int
1944 nfe_alloc_tx_ring(struct nfe_softc *sc, struct nfe_tx_ring *ring)
1945 {
1946 	int i, j, error, descsize;
1947 	bus_dmamem_t dmem;
1948 	void **desc;
1949 
1950 	if (sc->sc_caps & NFE_40BIT_ADDR) {
1951 		desc = (void *)&ring->desc64;
1952 		descsize = sizeof(struct nfe_desc64);
1953 	} else {
1954 		desc = (void *)&ring->desc32;
1955 		descsize = sizeof(struct nfe_desc32);
1956 	}
1957 
1958 	ring->queued = 0;
1959 	ring->cur = ring->next = 0;
1960 
1961 	error = bus_dmamem_coherent(sc->sc_dtag, PAGE_SIZE, 0,
1962 				    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
1963 				    sc->sc_tx_ring_count * descsize,
1964 				    BUS_DMA_WAITOK | BUS_DMA_ZERO, &dmem);
1965 	if (error) {
1966 		if_printf(&sc->arpcom.ac_if,
1967 			  "could not create TX desc ring\n");
1968 		return error;
1969 	}
1970 	ring->tag = dmem.dmem_tag;
1971 	ring->map = dmem.dmem_map;
1972 	*desc = dmem.dmem_addr;
1973 	ring->physaddr = dmem.dmem_busaddr;
1974 
1975 	ring->data = kmalloc(sizeof(struct nfe_tx_data) * sc->sc_tx_ring_count,
1976 			     M_DEVBUF, M_WAITOK | M_ZERO);
1977 
1978 	error = bus_dma_tag_create(sc->sc_dtag, 1, 0,
1979 			BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
1980 			NULL, NULL,
1981 			NFE_JBYTES, NFE_MAX_SCATTER, MCLBYTES,
1982 			BUS_DMA_ALLOCNOW | BUS_DMA_WAITOK | BUS_DMA_ONEBPAGE,
1983 			&ring->data_tag);
1984 	if (error) {
1985 		if_printf(&sc->arpcom.ac_if,
1986 			  "could not create TX buf DMA tag\n");
1987 		return error;
1988 	}
1989 
1990 	for (i = 0; i < sc->sc_tx_ring_count; i++) {
1991 		error = bus_dmamap_create(ring->data_tag,
1992 				BUS_DMA_WAITOK | BUS_DMA_ONEBPAGE,
1993 				&ring->data[i].map);
1994 		if (error) {
1995 			if_printf(&sc->arpcom.ac_if,
1996 				  "could not create %dth TX buf DMA map\n", i);
1997 			goto fail;
1998 		}
1999 	}
2000 
2001 	return 0;
2002 fail:
2003 	for (j = 0; j < i; ++j)
2004 		bus_dmamap_destroy(ring->data_tag, ring->data[i].map);
2005 	bus_dma_tag_destroy(ring->data_tag);
2006 	ring->data_tag = NULL;
2007 	return error;
2008 }
2009 
2010 static void
2011 nfe_reset_tx_ring(struct nfe_softc *sc, struct nfe_tx_ring *ring)
2012 {
2013 	int i;
2014 
2015 	for (i = 0; i < sc->sc_tx_ring_count; i++) {
2016 		struct nfe_tx_data *data = &ring->data[i];
2017 
2018 		if (sc->sc_caps & NFE_40BIT_ADDR)
2019 			ring->desc64[i].flags = 0;
2020 		else
2021 			ring->desc32[i].flags = 0;
2022 
2023 		if (data->m != NULL) {
2024 			bus_dmamap_unload(ring->data_tag, data->map);
2025 			m_freem(data->m);
2026 			data->m = NULL;
2027 		}
2028 	}
2029 
2030 	ring->queued = 0;
2031 	ring->cur = ring->next = 0;
2032 }
2033 
2034 static int
2035 nfe_init_tx_ring(struct nfe_softc *sc __unused,
2036 		 struct nfe_tx_ring *ring __unused)
2037 {
2038 	return 0;
2039 }
2040 
2041 static void
2042 nfe_free_tx_ring(struct nfe_softc *sc, struct nfe_tx_ring *ring)
2043 {
2044 	if (ring->data_tag != NULL) {
2045 		struct nfe_tx_data *data;
2046 		int i;
2047 
2048 		for (i = 0; i < sc->sc_tx_ring_count; ++i) {
2049 			data = &ring->data[i];
2050 
2051 			if (data->m != NULL) {
2052 				bus_dmamap_unload(ring->data_tag, data->map);
2053 				m_freem(data->m);
2054 			}
2055 			bus_dmamap_destroy(ring->data_tag, data->map);
2056 		}
2057 
2058 		bus_dma_tag_destroy(ring->data_tag);
2059 	}
2060 
2061 	if (ring->data != NULL)
2062 		kfree(ring->data, M_DEVBUF);
2063 
2064 	if (ring->tag != NULL) {
2065 		void *desc;
2066 
2067 		if (sc->sc_caps & NFE_40BIT_ADDR)
2068 			desc = ring->desc64;
2069 		else
2070 			desc = ring->desc32;
2071 
2072 		bus_dmamap_unload(ring->tag, ring->map);
2073 		bus_dmamem_free(ring->tag, desc, ring->map);
2074 		bus_dma_tag_destroy(ring->tag);
2075 	}
2076 }
2077 
2078 static int
2079 nfe_ifmedia_upd(struct ifnet *ifp)
2080 {
2081 	struct nfe_softc *sc = ifp->if_softc;
2082 	struct mii_data *mii = device_get_softc(sc->sc_miibus);
2083 
2084 	ASSERT_SERIALIZED(ifp->if_serializer);
2085 
2086 	if (mii->mii_instance != 0) {
2087 		struct mii_softc *miisc;
2088 
2089 		LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
2090 			mii_phy_reset(miisc);
2091 	}
2092 	mii_mediachg(mii);
2093 
2094 	return 0;
2095 }
2096 
2097 static void
2098 nfe_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2099 {
2100 	struct nfe_softc *sc = ifp->if_softc;
2101 	struct mii_data *mii = device_get_softc(sc->sc_miibus);
2102 
2103 	ASSERT_SERIALIZED(ifp->if_serializer);
2104 
2105 	mii_pollstat(mii);
2106 	ifmr->ifm_status = mii->mii_media_status;
2107 	ifmr->ifm_active = mii->mii_media_active;
2108 }
2109 
2110 static void
2111 nfe_setmulti(struct nfe_softc *sc)
2112 {
2113 	struct ifnet *ifp = &sc->arpcom.ac_if;
2114 	struct ifmultiaddr *ifma;
2115 	uint8_t addr[ETHER_ADDR_LEN], mask[ETHER_ADDR_LEN];
2116 	uint32_t filter = NFE_RXFILTER_MAGIC;
2117 	int i;
2118 
2119 	if ((ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) != 0) {
2120 		bzero(addr, ETHER_ADDR_LEN);
2121 		bzero(mask, ETHER_ADDR_LEN);
2122 		goto done;
2123 	}
2124 
2125 	bcopy(etherbroadcastaddr, addr, ETHER_ADDR_LEN);
2126 	bcopy(etherbroadcastaddr, mask, ETHER_ADDR_LEN);
2127 
2128 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2129 		caddr_t maddr;
2130 
2131 		if (ifma->ifma_addr->sa_family != AF_LINK)
2132 			continue;
2133 
2134 		maddr = LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
2135 		for (i = 0; i < ETHER_ADDR_LEN; i++) {
2136 			addr[i] &= maddr[i];
2137 			mask[i] &= ~maddr[i];
2138 		}
2139 	}
2140 
2141 	for (i = 0; i < ETHER_ADDR_LEN; i++)
2142 		mask[i] |= addr[i];
2143 
2144 done:
2145 	addr[0] |= 0x01;	/* make sure multicast bit is set */
2146 
2147 	NFE_WRITE(sc, NFE_MULTIADDR_HI,
2148 	    addr[3] << 24 | addr[2] << 16 | addr[1] << 8 | addr[0]);
2149 	NFE_WRITE(sc, NFE_MULTIADDR_LO,
2150 	    addr[5] <<  8 | addr[4]);
2151 	NFE_WRITE(sc, NFE_MULTIMASK_HI,
2152 	    mask[3] << 24 | mask[2] << 16 | mask[1] << 8 | mask[0]);
2153 	NFE_WRITE(sc, NFE_MULTIMASK_LO,
2154 	    mask[5] <<  8 | mask[4]);
2155 
2156 	filter |= (ifp->if_flags & IFF_PROMISC) ? NFE_PROMISC : NFE_U2M;
2157 	NFE_WRITE(sc, NFE_RXFILTER, filter);
2158 }
2159 
2160 static void
2161 nfe_get_macaddr(struct nfe_softc *sc, uint8_t *addr)
2162 {
2163 	uint32_t lo, hi;
2164 
2165 	lo = NFE_READ(sc, NFE_MACADDR_LO);
2166 	hi = NFE_READ(sc, NFE_MACADDR_HI);
2167 	if (sc->sc_caps & NFE_FIX_EADDR) {
2168 		addr[0] = (lo >> 8) & 0xff;
2169 		addr[1] = (lo & 0xff);
2170 
2171 		addr[2] = (hi >> 24) & 0xff;
2172 		addr[3] = (hi >> 16) & 0xff;
2173 		addr[4] = (hi >>  8) & 0xff;
2174 		addr[5] = (hi & 0xff);
2175 	} else {
2176 		addr[0] = (hi & 0xff);
2177 		addr[1] = (hi >>  8) & 0xff;
2178 		addr[2] = (hi >> 16) & 0xff;
2179 		addr[3] = (hi >> 24) & 0xff;
2180 
2181 		addr[4] = (lo & 0xff);
2182 		addr[5] = (lo >>  8) & 0xff;
2183 	}
2184 }
2185 
2186 static void
2187 nfe_set_macaddr(struct nfe_softc *sc, const uint8_t *addr)
2188 {
2189 	NFE_WRITE(sc, NFE_MACADDR_LO,
2190 	    addr[5] <<  8 | addr[4]);
2191 	NFE_WRITE(sc, NFE_MACADDR_HI,
2192 	    addr[3] << 24 | addr[2] << 16 | addr[1] << 8 | addr[0]);
2193 }
2194 
2195 static void
2196 nfe_tick(void *arg)
2197 {
2198 	struct nfe_softc *sc = arg;
2199 	struct ifnet *ifp = &sc->arpcom.ac_if;
2200 	struct mii_data *mii = device_get_softc(sc->sc_miibus);
2201 
2202 	lwkt_serialize_enter(ifp->if_serializer);
2203 
2204 	mii_tick(mii);
2205 	callout_reset(&sc->sc_tick_ch, hz, nfe_tick, sc);
2206 
2207 	lwkt_serialize_exit(ifp->if_serializer);
2208 }
2209 
2210 static int
2211 nfe_newbuf_std(struct nfe_softc *sc, struct nfe_rx_ring *ring, int idx,
2212 	       int wait)
2213 {
2214 	struct nfe_rx_data *data = &ring->data[idx];
2215 	bus_dma_segment_t seg;
2216 	bus_dmamap_t map;
2217 	struct mbuf *m;
2218 	int nsegs, error;
2219 
2220 	m = m_getcl(wait ? MB_WAIT : MB_DONTWAIT, MT_DATA, M_PKTHDR);
2221 	if (m == NULL)
2222 		return ENOBUFS;
2223 	m->m_len = m->m_pkthdr.len = MCLBYTES;
2224 
2225 	/*
2226 	 * Aligning the payload improves access times.
2227 	 */
2228 	if (sc->sc_caps & NFE_WORDALIGN)
2229 		m_adj(m, ETHER_ALIGN);
2230 
2231 	error = bus_dmamap_load_mbuf_segment(ring->data_tag, ring->data_tmpmap,
2232 			m, &seg, 1, &nsegs, BUS_DMA_NOWAIT);
2233 	if (error) {
2234 		m_freem(m);
2235 		if (wait) {
2236 			if_printf(&sc->arpcom.ac_if,
2237 				  "could map RX mbuf %d\n", error);
2238 		}
2239 		return error;
2240 	}
2241 
2242 	if (data->m != NULL) {
2243 		/* Sync and unload originally mapped mbuf */
2244 		bus_dmamap_sync(ring->data_tag, data->map,
2245 				BUS_DMASYNC_POSTREAD);
2246 		bus_dmamap_unload(ring->data_tag, data->map);
2247 	}
2248 
2249 	/* Swap this DMA map with tmp DMA map */
2250 	map = data->map;
2251 	data->map = ring->data_tmpmap;
2252 	ring->data_tmpmap = map;
2253 
2254 	/* Caller is assumed to have collected the old mbuf */
2255 	data->m = m;
2256 
2257 	nfe_set_paddr_rxdesc(sc, ring, idx, seg.ds_addr);
2258 	return 0;
2259 }
2260 
2261 static int
2262 nfe_newbuf_jumbo(struct nfe_softc *sc, struct nfe_rx_ring *ring, int idx,
2263 		 int wait)
2264 {
2265 	struct nfe_rx_data *data = &ring->data[idx];
2266 	struct nfe_jbuf *jbuf;
2267 	struct mbuf *m;
2268 
2269 	MGETHDR(m, wait ? MB_WAIT : MB_DONTWAIT, MT_DATA);
2270 	if (m == NULL)
2271 		return ENOBUFS;
2272 
2273 	jbuf = nfe_jalloc(sc);
2274 	if (jbuf == NULL) {
2275 		m_freem(m);
2276 		if_printf(&sc->arpcom.ac_if, "jumbo allocation failed "
2277 		    "-- packet dropped!\n");
2278 		return ENOBUFS;
2279 	}
2280 
2281 	m->m_ext.ext_arg = jbuf;
2282 	m->m_ext.ext_buf = jbuf->buf;
2283 	m->m_ext.ext_free = nfe_jfree;
2284 	m->m_ext.ext_ref = nfe_jref;
2285 	m->m_ext.ext_size = NFE_JBYTES;
2286 
2287 	m->m_data = m->m_ext.ext_buf;
2288 	m->m_flags |= M_EXT;
2289 	m->m_len = m->m_pkthdr.len = m->m_ext.ext_size;
2290 
2291 	/*
2292 	 * Aligning the payload improves access times.
2293 	 */
2294 	if (sc->sc_caps & NFE_WORDALIGN)
2295 		m_adj(m, ETHER_ALIGN);
2296 
2297 	/* Caller is assumed to have collected the old mbuf */
2298 	data->m = m;
2299 
2300 	nfe_set_paddr_rxdesc(sc, ring, idx, jbuf->physaddr);
2301 	return 0;
2302 }
2303 
2304 static void
2305 nfe_set_paddr_rxdesc(struct nfe_softc *sc, struct nfe_rx_ring *ring, int idx,
2306 		     bus_addr_t physaddr)
2307 {
2308 	if (sc->sc_caps & NFE_40BIT_ADDR) {
2309 		struct nfe_desc64 *desc64 = &ring->desc64[idx];
2310 
2311 		desc64->physaddr[0] = htole32(NFE_ADDR_HI(physaddr));
2312 		desc64->physaddr[1] = htole32(NFE_ADDR_LO(physaddr));
2313 	} else {
2314 		struct nfe_desc32 *desc32 = &ring->desc32[idx];
2315 
2316 		desc32->physaddr = htole32(physaddr);
2317 	}
2318 }
2319 
2320 static void
2321 nfe_set_ready_rxdesc(struct nfe_softc *sc, struct nfe_rx_ring *ring, int idx)
2322 {
2323 	if (sc->sc_caps & NFE_40BIT_ADDR) {
2324 		struct nfe_desc64 *desc64 = &ring->desc64[idx];
2325 
2326 		desc64->length = htole16(ring->bufsz);
2327 		desc64->flags = htole16(NFE_RX_READY);
2328 	} else {
2329 		struct nfe_desc32 *desc32 = &ring->desc32[idx];
2330 
2331 		desc32->length = htole16(ring->bufsz);
2332 		desc32->flags = htole16(NFE_RX_READY);
2333 	}
2334 }
2335 
2336 static int
2337 nfe_sysctl_imtime(SYSCTL_HANDLER_ARGS)
2338 {
2339 	struct nfe_softc *sc = arg1;
2340 	struct ifnet *ifp = &sc->arpcom.ac_if;
2341 	uint32_t flags;
2342 	int error, v;
2343 
2344 	lwkt_serialize_enter(ifp->if_serializer);
2345 
2346 	flags = sc->sc_flags & ~NFE_F_DYN_IM;
2347 	v = sc->sc_imtime;
2348 	if (sc->sc_flags & NFE_F_DYN_IM)
2349 		v = -v;
2350 
2351 	error = sysctl_handle_int(oidp, &v, 0, req);
2352 	if (error || req->newptr == NULL)
2353 		goto back;
2354 
2355 	if (v < 0) {
2356 		flags |= NFE_F_DYN_IM;
2357 		v = -v;
2358 	}
2359 
2360 	if (v != sc->sc_imtime || (flags ^ sc->sc_flags)) {
2361 		if (NFE_IMTIME(v) == 0)
2362 			v = 0;
2363 		sc->sc_imtime = v;
2364 		sc->sc_flags = flags;
2365 		sc->sc_irq_enable = NFE_IRQ_ENABLE(sc);
2366 
2367 		if ((ifp->if_flags & (IFF_NPOLLING | IFF_RUNNING))
2368 		    == IFF_RUNNING) {
2369 			nfe_enable_intrs(sc);
2370 		}
2371 	}
2372 back:
2373 	lwkt_serialize_exit(ifp->if_serializer);
2374 	return error;
2375 }
2376 
2377 static void
2378 nfe_powerup(device_t dev)
2379 {
2380 	struct nfe_softc *sc = device_get_softc(dev);
2381 	uint32_t pwr_state;
2382 	uint16_t did;
2383 
2384 	/*
2385 	 * Bring MAC and PHY out of low power state
2386 	 */
2387 
2388 	pwr_state = NFE_READ(sc, NFE_PWR_STATE2) & ~NFE_PWRUP_MASK;
2389 
2390 	did = pci_get_device(dev);
2391 	if ((did == PCI_PRODUCT_NVIDIA_MCP51_LAN1 ||
2392 	     did == PCI_PRODUCT_NVIDIA_MCP51_LAN2) &&
2393 	    pci_get_revid(dev) >= 0xa3)
2394 		pwr_state |= NFE_PWRUP_REV_A3;
2395 
2396 	NFE_WRITE(sc, NFE_PWR_STATE2, pwr_state);
2397 }
2398 
2399 static void
2400 nfe_mac_reset(struct nfe_softc *sc)
2401 {
2402 	uint32_t rxtxctl = sc->rxtxctl_desc | NFE_RXTX_BIT2;
2403 	uint32_t macaddr_hi, macaddr_lo, tx_poll;
2404 
2405 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_RESET | rxtxctl);
2406 
2407 	/* Save several registers for later restoration */
2408 	macaddr_hi = NFE_READ(sc, NFE_MACADDR_HI);
2409 	macaddr_lo = NFE_READ(sc, NFE_MACADDR_LO);
2410 	tx_poll = NFE_READ(sc, NFE_TX_POLL);
2411 
2412 	NFE_WRITE(sc, NFE_MAC_RESET, NFE_RESET_ASSERT);
2413 	DELAY(100);
2414 
2415 	NFE_WRITE(sc, NFE_MAC_RESET, 0);
2416 	DELAY(100);
2417 
2418 	/* Restore saved registers */
2419 	NFE_WRITE(sc, NFE_MACADDR_HI, macaddr_hi);
2420 	NFE_WRITE(sc, NFE_MACADDR_LO, macaddr_lo);
2421 	NFE_WRITE(sc, NFE_TX_POLL, tx_poll);
2422 
2423 	NFE_WRITE(sc, NFE_RXTX_CTL, rxtxctl);
2424 }
2425 
2426 static void
2427 nfe_enable_intrs(struct nfe_softc *sc)
2428 {
2429 	/*
2430 	 * NFE_IMTIMER generates a periodic interrupt via NFE_IRQ_TIMER.
2431 	 * It is unclear how wide the timer is.  Base programming does
2432 	 * not seem to effect NFE_IRQ_TX_DONE or NFE_IRQ_RX_DONE so
2433 	 * we don't get any interrupt moderation.  TX moderation is
2434 	 * possible by using the timer interrupt instead of TX_DONE.
2435 	 *
2436 	 * It is unclear whether there are other bits that can be
2437 	 * set to make the NFE device actually do interrupt moderation
2438 	 * on the RX side.
2439 	 *
2440 	 * For now set a 128uS interval as a placemark, but don't use
2441 	 * the timer.
2442 	 */
2443 	if (sc->sc_imtime == 0)
2444 		NFE_WRITE(sc, NFE_IMTIMER, NFE_IMTIME_DEFAULT);
2445 	else
2446 		NFE_WRITE(sc, NFE_IMTIMER, NFE_IMTIME(sc->sc_imtime));
2447 
2448 	/* Enable interrupts */
2449 	NFE_WRITE(sc, NFE_IRQ_MASK, sc->sc_irq_enable);
2450 
2451 	if (sc->sc_irq_enable & NFE_IRQ_TIMER)
2452 		sc->sc_flags |= NFE_F_IRQ_TIMER;
2453 	else
2454 		sc->sc_flags &= ~NFE_F_IRQ_TIMER;
2455 }
2456