xref: /dragonfly/sys/dev/netif/nfe/if_nfe.c (revision 28c26f7e)
1 /*	$OpenBSD: if_nfe.c,v 1.63 2006/06/17 18:00:43 brad Exp $	*/
2 /*	$DragonFly: src/sys/dev/netif/nfe/if_nfe.c,v 1.46 2008/10/28 07:30:49 sephe Exp $	*/
3 
4 /*
5  * Copyright (c) 2006 The DragonFly Project.  All rights reserved.
6  *
7  * This code is derived from software contributed to The DragonFly Project
8  * by Sepherosa Ziehau <sepherosa@gmail.com> and
9  * Matthew Dillon <dillon@apollo.backplane.com>
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  *
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in
19  *    the documentation and/or other materials provided with the
20  *    distribution.
21  * 3. Neither the name of The DragonFly Project nor the names of its
22  *    contributors may be used to endorse or promote products derived
23  *    from this software without specific, prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
26  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
27  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
28  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
29  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
30  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
31  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
32  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
33  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
34  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
35  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  */
38 
39 /*
40  * Copyright (c) 2006 Damien Bergamini <damien.bergamini@free.fr>
41  * Copyright (c) 2005, 2006 Jonathan Gray <jsg@openbsd.org>
42  *
43  * Permission to use, copy, modify, and distribute this software for any
44  * purpose with or without fee is hereby granted, provided that the above
45  * copyright notice and this permission notice appear in all copies.
46  *
47  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
48  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
49  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
50  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
51  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
52  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
53  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
54  */
55 
56 /* Driver for NVIDIA nForce MCP Fast Ethernet and Gigabit Ethernet */
57 
58 #include "opt_polling.h"
59 
60 #include <sys/param.h>
61 #include <sys/endian.h>
62 #include <sys/kernel.h>
63 #include <sys/bus.h>
64 #include <sys/interrupt.h>
65 #include <sys/proc.h>
66 #include <sys/rman.h>
67 #include <sys/serialize.h>
68 #include <sys/socket.h>
69 #include <sys/sockio.h>
70 #include <sys/sysctl.h>
71 
72 #include <net/ethernet.h>
73 #include <net/if.h>
74 #include <net/bpf.h>
75 #include <net/if_arp.h>
76 #include <net/if_dl.h>
77 #include <net/if_media.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 <bus/pci/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 DEVICE_POLLING
111 static void	nfe_poll(struct ifnet *, enum poll_cmd, int);
112 #endif
113 static void	nfe_intr(void *);
114 static int	nfe_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
115 static int	nfe_rxeof(struct nfe_softc *);
116 static int	nfe_txeof(struct nfe_softc *, int);
117 static int	nfe_encap(struct nfe_softc *, struct nfe_tx_ring *,
118 			  struct mbuf *);
119 static void	nfe_start(struct ifnet *);
120 static void	nfe_watchdog(struct ifnet *);
121 static void	nfe_init(void *);
122 static void	nfe_stop(struct nfe_softc *);
123 static struct nfe_jbuf *nfe_jalloc(struct nfe_softc *);
124 static void	nfe_jfree(void *);
125 static void	nfe_jref(void *);
126 static int	nfe_jpool_alloc(struct nfe_softc *, struct nfe_rx_ring *);
127 static void	nfe_jpool_free(struct nfe_softc *, struct nfe_rx_ring *);
128 static int	nfe_alloc_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
129 static void	nfe_reset_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
130 static int	nfe_init_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
131 static void	nfe_free_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
132 static int	nfe_alloc_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
133 static void	nfe_reset_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
134 static int	nfe_init_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
135 static void	nfe_free_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
136 static int	nfe_ifmedia_upd(struct ifnet *);
137 static void	nfe_ifmedia_sts(struct ifnet *, struct ifmediareq *);
138 static void	nfe_setmulti(struct nfe_softc *);
139 static void	nfe_get_macaddr(struct nfe_softc *, uint8_t *);
140 static void	nfe_set_macaddr(struct nfe_softc *, const uint8_t *);
141 static void	nfe_powerup(device_t);
142 static void	nfe_mac_reset(struct nfe_softc *);
143 static void	nfe_tick(void *);
144 static void	nfe_set_paddr_rxdesc(struct nfe_softc *, struct nfe_rx_ring *,
145 				     int, bus_addr_t);
146 static void	nfe_set_ready_rxdesc(struct nfe_softc *, struct nfe_rx_ring *,
147 				     int);
148 static int	nfe_newbuf_std(struct nfe_softc *, struct nfe_rx_ring *, int,
149 			       int);
150 static int	nfe_newbuf_jumbo(struct nfe_softc *, struct nfe_rx_ring *, int,
151 				 int);
152 static void	nfe_enable_intrs(struct nfe_softc *);
153 static void	nfe_disable_intrs(struct nfe_softc *);
154 
155 static int	nfe_sysctl_imtime(SYSCTL_HANDLER_ARGS);
156 
157 #define NFE_DEBUG
158 #ifdef NFE_DEBUG
159 
160 static int	nfe_debug = 0;
161 static int	nfe_rx_ring_count = NFE_RX_RING_DEF_COUNT;
162 static int	nfe_tx_ring_count = NFE_TX_RING_DEF_COUNT;
163 /*
164  * hw timer simulated interrupt moderation @4000Hz.  Negative values
165  * disable the timer when the discrete interrupt rate falls below
166  * the moderation rate.
167  *
168  * XXX 8000Hz might be better but if the interrupt is shared it can
169  *     blow out the cpu.
170  */
171 static int	nfe_imtime = -250;	/* uS */
172 
173 TUNABLE_INT("hw.nfe.rx_ring_count", &nfe_rx_ring_count);
174 TUNABLE_INT("hw.nfe.tx_ring_count", &nfe_tx_ring_count);
175 TUNABLE_INT("hw.nfe.imtimer", &nfe_imtime);
176 TUNABLE_INT("hw.nfe.debug", &nfe_debug);
177 
178 #define DPRINTF(sc, fmt, ...) do {		\
179 	if ((sc)->sc_debug) {			\
180 		if_printf(&(sc)->arpcom.ac_if,	\
181 			  fmt, __VA_ARGS__);	\
182 	}					\
183 } while (0)
184 
185 #define DPRINTFN(sc, lv, fmt, ...) do {		\
186 	if ((sc)->sc_debug >= (lv)) {		\
187 		if_printf(&(sc)->arpcom.ac_if,	\
188 			  fmt, __VA_ARGS__);	\
189 	}					\
190 } while (0)
191 
192 #else	/* !NFE_DEBUG */
193 
194 #define DPRINTF(sc, fmt, ...)
195 #define DPRINTFN(sc, lv, fmt, ...)
196 
197 #endif	/* NFE_DEBUG */
198 
199 static const struct nfe_dev {
200 	uint16_t	vid;
201 	uint16_t	did;
202 	const char	*desc;
203 } nfe_devices[] = {
204 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE_LAN,
205 	  "NVIDIA nForce Fast Ethernet" },
206 
207 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE2_LAN,
208 	  "NVIDIA nForce2 Fast Ethernet" },
209 
210 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN1,
211 	  "NVIDIA nForce3 Gigabit Ethernet" },
212 
213 	/* XXX TGEN the next chip can also be found in the nForce2 Ultra 400Gb
214 	   chipset, and possibly also the 400R; it might be both nForce2- and
215 	   nForce3-based boards can use the same MCPs (= southbridges) */
216 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN2,
217 	  "NVIDIA nForce3 Gigabit Ethernet" },
218 
219 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN3,
220 	  "NVIDIA nForce3 Gigabit Ethernet" },
221 
222 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN4,
223 	  "NVIDIA nForce3 Gigabit Ethernet" },
224 
225 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN5,
226 	  "NVIDIA nForce3 Gigabit Ethernet" },
227 
228 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_CK804_LAN1,
229 	  "NVIDIA CK804 Gigabit Ethernet" },
230 
231 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_CK804_LAN2,
232 	  "NVIDIA CK804 Gigabit Ethernet" },
233 
234 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP04_LAN1,
235 	  "NVIDIA MCP04 Gigabit Ethernet" },
236 
237 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP04_LAN2,
238 	  "NVIDIA MCP04 Gigabit Ethernet" },
239 
240 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP51_LAN1,
241 	  "NVIDIA MCP51 Gigabit Ethernet" },
242 
243 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP51_LAN2,
244 	  "NVIDIA MCP51 Gigabit Ethernet" },
245 
246 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP55_LAN1,
247 	  "NVIDIA MCP55 Gigabit Ethernet" },
248 
249 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP55_LAN2,
250 	  "NVIDIA MCP55 Gigabit Ethernet" },
251 
252 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN1,
253 	  "NVIDIA MCP61 Gigabit Ethernet" },
254 
255 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN2,
256 	  "NVIDIA MCP61 Gigabit Ethernet" },
257 
258 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN3,
259 	  "NVIDIA MCP61 Gigabit Ethernet" },
260 
261 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN4,
262 	  "NVIDIA MCP61 Gigabit Ethernet" },
263 
264 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN1,
265 	  "NVIDIA MCP65 Gigabit Ethernet" },
266 
267 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN2,
268 	  "NVIDIA MCP65 Gigabit Ethernet" },
269 
270 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN3,
271 	  "NVIDIA MCP65 Gigabit Ethernet" },
272 
273 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN4,
274 	  "NVIDIA MCP65 Gigabit Ethernet" },
275 
276 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN1,
277 	  "NVIDIA MCP67 Gigabit Ethernet" },
278 
279 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN2,
280 	  "NVIDIA MCP67 Gigabit Ethernet" },
281 
282 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN3,
283 	  "NVIDIA MCP67 Gigabit Ethernet" },
284 
285 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN4,
286 	  "NVIDIA MCP67 Gigabit Ethernet" },
287 
288 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP73_LAN1,
289 	  "NVIDIA MCP73 Gigabit Ethernet" },
290 
291 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP73_LAN2,
292 	  "NVIDIA MCP73 Gigabit Ethernet" },
293 
294 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP73_LAN3,
295 	  "NVIDIA MCP73 Gigabit Ethernet" },
296 
297 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP73_LAN4,
298 	  "NVIDIA MCP73 Gigabit Ethernet" },
299 
300 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP77_LAN1,
301 	  "NVIDIA MCP77 Gigabit Ethernet" },
302 
303 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP77_LAN2,
304 	  "NVIDIA MCP77 Gigabit Ethernet" },
305 
306 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP77_LAN3,
307 	  "NVIDIA MCP77 Gigabit Ethernet" },
308 
309 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP77_LAN4,
310 	  "NVIDIA MCP77 Gigabit Ethernet" },
311 
312 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP79_LAN1,
313 	  "NVIDIA MCP79 Gigabit Ethernet" },
314 
315 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP79_LAN2,
316 	  "NVIDIA MCP79 Gigabit Ethernet" },
317 
318 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP79_LAN3,
319 	  "NVIDIA MCP79 Gigabit Ethernet" },
320 
321 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP79_LAN4,
322 	  "NVIDIA MCP79 Gigabit Ethernet" },
323 
324 	{ 0, 0, NULL }
325 };
326 
327 static device_method_t nfe_methods[] = {
328 	/* Device interface */
329 	DEVMETHOD(device_probe,		nfe_probe),
330 	DEVMETHOD(device_attach,	nfe_attach),
331 	DEVMETHOD(device_detach,	nfe_detach),
332 	DEVMETHOD(device_suspend,	nfe_suspend),
333 	DEVMETHOD(device_resume,	nfe_resume),
334 	DEVMETHOD(device_shutdown,	nfe_shutdown),
335 
336 	/* Bus interface */
337 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
338 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
339 
340 	/* MII interface */
341 	DEVMETHOD(miibus_readreg,	nfe_miibus_readreg),
342 	DEVMETHOD(miibus_writereg,	nfe_miibus_writereg),
343 	DEVMETHOD(miibus_statchg,	nfe_miibus_statchg),
344 
345 	{ 0, 0 }
346 };
347 
348 static driver_t nfe_driver = {
349 	"nfe",
350 	nfe_methods,
351 	sizeof(struct nfe_softc)
352 };
353 
354 static devclass_t	nfe_devclass;
355 
356 DECLARE_DUMMY_MODULE(if_nfe);
357 MODULE_DEPEND(if_nfe, miibus, 1, 1, 1);
358 DRIVER_MODULE(if_nfe, pci, nfe_driver, nfe_devclass, 0, 0);
359 DRIVER_MODULE(miibus, nfe, miibus_driver, miibus_devclass, 0, 0);
360 
361 static int
362 nfe_probe(device_t dev)
363 {
364 	const struct nfe_dev *n;
365 	uint16_t vid, did;
366 
367 	vid = pci_get_vendor(dev);
368 	did = pci_get_device(dev);
369 	for (n = nfe_devices; n->desc != NULL; ++n) {
370 		if (vid == n->vid && did == n->did) {
371 			struct nfe_softc *sc = device_get_softc(dev);
372 
373 			switch (did) {
374 			case PCI_PRODUCT_NVIDIA_NFORCE_LAN:
375 			case PCI_PRODUCT_NVIDIA_NFORCE2_LAN:
376 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN1:
377 				sc->sc_caps = NFE_NO_PWRCTL |
378 					      NFE_FIX_EADDR;
379 				break;
380 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN2:
381 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN3:
382 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN4:
383 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN5:
384 				sc->sc_caps = NFE_JUMBO_SUP |
385 					      NFE_HW_CSUM |
386 					      NFE_NO_PWRCTL |
387 					      NFE_FIX_EADDR;
388 				break;
389 			case PCI_PRODUCT_NVIDIA_MCP51_LAN1:
390 			case PCI_PRODUCT_NVIDIA_MCP51_LAN2:
391 				sc->sc_caps = NFE_FIX_EADDR;
392 				/* FALL THROUGH */
393 			case PCI_PRODUCT_NVIDIA_MCP61_LAN1:
394 			case PCI_PRODUCT_NVIDIA_MCP61_LAN2:
395 			case PCI_PRODUCT_NVIDIA_MCP61_LAN3:
396 			case PCI_PRODUCT_NVIDIA_MCP61_LAN4:
397 			case PCI_PRODUCT_NVIDIA_MCP67_LAN1:
398 			case PCI_PRODUCT_NVIDIA_MCP67_LAN2:
399 			case PCI_PRODUCT_NVIDIA_MCP67_LAN3:
400 			case PCI_PRODUCT_NVIDIA_MCP67_LAN4:
401 			case PCI_PRODUCT_NVIDIA_MCP73_LAN1:
402 			case PCI_PRODUCT_NVIDIA_MCP73_LAN2:
403 			case PCI_PRODUCT_NVIDIA_MCP73_LAN3:
404 			case PCI_PRODUCT_NVIDIA_MCP73_LAN4:
405 				sc->sc_caps |= NFE_40BIT_ADDR;
406 				break;
407 			case PCI_PRODUCT_NVIDIA_CK804_LAN1:
408 			case PCI_PRODUCT_NVIDIA_CK804_LAN2:
409 			case PCI_PRODUCT_NVIDIA_MCP04_LAN1:
410 			case PCI_PRODUCT_NVIDIA_MCP04_LAN2:
411 				sc->sc_caps = NFE_JUMBO_SUP |
412 					      NFE_40BIT_ADDR |
413 					      NFE_HW_CSUM |
414 					      NFE_NO_PWRCTL |
415 					      NFE_FIX_EADDR;
416 				break;
417 			case PCI_PRODUCT_NVIDIA_MCP65_LAN1:
418 			case PCI_PRODUCT_NVIDIA_MCP65_LAN2:
419 			case PCI_PRODUCT_NVIDIA_MCP65_LAN3:
420 			case PCI_PRODUCT_NVIDIA_MCP65_LAN4:
421 				sc->sc_caps = NFE_JUMBO_SUP |
422 					      NFE_40BIT_ADDR;
423 				break;
424 			case PCI_PRODUCT_NVIDIA_MCP55_LAN1:
425 			case PCI_PRODUCT_NVIDIA_MCP55_LAN2:
426 				sc->sc_caps = NFE_JUMBO_SUP |
427 					      NFE_40BIT_ADDR |
428 					      NFE_HW_CSUM |
429 					      NFE_HW_VLAN |
430 					      NFE_FIX_EADDR;
431 				break;
432 			case PCI_PRODUCT_NVIDIA_MCP77_LAN1:
433 			case PCI_PRODUCT_NVIDIA_MCP77_LAN2:
434 			case PCI_PRODUCT_NVIDIA_MCP77_LAN3:
435 			case PCI_PRODUCT_NVIDIA_MCP77_LAN4:
436 			case PCI_PRODUCT_NVIDIA_MCP79_LAN1:
437 			case PCI_PRODUCT_NVIDIA_MCP79_LAN2:
438 			case PCI_PRODUCT_NVIDIA_MCP79_LAN3:
439 			case PCI_PRODUCT_NVIDIA_MCP79_LAN4:
440 				sc->sc_caps = NFE_40BIT_ADDR |
441 					      NFE_HW_CSUM;
442 				break;
443 			}
444 
445 			device_set_desc(dev, n->desc);
446 			device_set_async_attach(dev, TRUE);
447 			return 0;
448 		}
449 	}
450 	return ENXIO;
451 }
452 
453 static int
454 nfe_attach(device_t dev)
455 {
456 	struct nfe_softc *sc = device_get_softc(dev);
457 	struct ifnet *ifp = &sc->arpcom.ac_if;
458 	uint8_t eaddr[ETHER_ADDR_LEN];
459 	bus_addr_t lowaddr;
460 	int error;
461 
462 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
463 	lwkt_serialize_init(&sc->sc_jbuf_serializer);
464 
465 	/*
466 	 * Initialize sysctl variables
467 	 */
468 	sc->sc_rx_ring_count = nfe_rx_ring_count;
469 	sc->sc_tx_ring_count = nfe_tx_ring_count;
470 	sc->sc_debug = nfe_debug;
471 	if (nfe_imtime < 0) {
472 		sc->sc_flags |= NFE_F_DYN_IM;
473 		sc->sc_imtime = -nfe_imtime;
474 	} else {
475 		sc->sc_imtime = nfe_imtime;
476 	}
477 	sc->sc_irq_enable = NFE_IRQ_ENABLE(sc);
478 
479 	sc->sc_mem_rid = PCIR_BAR(0);
480 
481 	if (sc->sc_caps & NFE_40BIT_ADDR)
482 		sc->rxtxctl_desc = NFE_RXTX_DESC_V3;
483 	else if (sc->sc_caps & NFE_JUMBO_SUP)
484 		sc->rxtxctl_desc = NFE_RXTX_DESC_V2;
485 
486 #ifndef BURN_BRIDGES
487 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
488 		uint32_t mem, irq;
489 
490 		mem = pci_read_config(dev, sc->sc_mem_rid, 4);
491 		irq = pci_read_config(dev, PCIR_INTLINE, 4);
492 
493 		device_printf(dev, "chip is in D%d power mode "
494 		    "-- setting to D0\n", pci_get_powerstate(dev));
495 
496 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
497 
498 		pci_write_config(dev, sc->sc_mem_rid, mem, 4);
499 		pci_write_config(dev, PCIR_INTLINE, irq, 4);
500 	}
501 #endif	/* !BURN_BRIDGE */
502 
503 	/* Enable bus mastering */
504 	pci_enable_busmaster(dev);
505 
506 	/* Allocate IO memory */
507 	sc->sc_mem_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
508 						&sc->sc_mem_rid, RF_ACTIVE);
509 	if (sc->sc_mem_res == NULL) {
510 		device_printf(dev, "could not allocate io memory\n");
511 		return ENXIO;
512 	}
513 	sc->sc_memh = rman_get_bushandle(sc->sc_mem_res);
514 	sc->sc_memt = rman_get_bustag(sc->sc_mem_res);
515 
516 	/* Allocate IRQ */
517 	sc->sc_irq_rid = 0;
518 	sc->sc_irq_res = bus_alloc_resource_any(dev, SYS_RES_IRQ,
519 						&sc->sc_irq_rid,
520 						RF_SHAREABLE | RF_ACTIVE);
521 	if (sc->sc_irq_res == NULL) {
522 		device_printf(dev, "could not allocate irq\n");
523 		error = ENXIO;
524 		goto fail;
525 	}
526 
527 	/* Disable WOL */
528 	NFE_WRITE(sc, NFE_WOL_CTL, 0);
529 
530 	if ((sc->sc_caps & NFE_NO_PWRCTL) == 0)
531 		nfe_powerup(dev);
532 
533 	nfe_get_macaddr(sc, eaddr);
534 
535 	/*
536 	 * Allocate top level DMA tag
537 	 */
538 	if (sc->sc_caps & NFE_40BIT_ADDR)
539 		lowaddr = NFE_BUS_SPACE_MAXADDR;
540 	else
541 		lowaddr = BUS_SPACE_MAXADDR_32BIT;
542 	error = bus_dma_tag_create(NULL,	/* parent */
543 			1, 0,			/* alignment, boundary */
544 			lowaddr,		/* lowaddr */
545 			BUS_SPACE_MAXADDR,	/* highaddr */
546 			NULL, NULL,		/* filter, filterarg */
547 			BUS_SPACE_MAXSIZE_32BIT,/* maxsize */
548 			0,			/* nsegments */
549 			BUS_SPACE_MAXSIZE_32BIT,/* maxsegsize */
550 			0,			/* flags */
551 			&sc->sc_dtag);
552 	if (error) {
553 		device_printf(dev, "could not allocate parent dma tag\n");
554 		goto fail;
555 	}
556 
557 	/*
558 	 * Allocate Tx and Rx rings.
559 	 */
560 	error = nfe_alloc_tx_ring(sc, &sc->txq);
561 	if (error) {
562 		device_printf(dev, "could not allocate Tx ring\n");
563 		goto fail;
564 	}
565 
566 	error = nfe_alloc_rx_ring(sc, &sc->rxq);
567 	if (error) {
568 		device_printf(dev, "could not allocate Rx ring\n");
569 		goto fail;
570 	}
571 
572 	/*
573 	 * Create sysctl tree
574 	 */
575 	sysctl_ctx_init(&sc->sc_sysctl_ctx);
576 	sc->sc_sysctl_tree = SYSCTL_ADD_NODE(&sc->sc_sysctl_ctx,
577 					     SYSCTL_STATIC_CHILDREN(_hw),
578 					     OID_AUTO,
579 					     device_get_nameunit(dev),
580 					     CTLFLAG_RD, 0, "");
581 	if (sc->sc_sysctl_tree == NULL) {
582 		device_printf(dev, "can't add sysctl node\n");
583 		error = ENXIO;
584 		goto fail;
585 	}
586 	SYSCTL_ADD_PROC(&sc->sc_sysctl_ctx,
587 			SYSCTL_CHILDREN(sc->sc_sysctl_tree),
588 			OID_AUTO, "imtimer", CTLTYPE_INT | CTLFLAG_RW,
589 			sc, 0, nfe_sysctl_imtime, "I",
590 			"Interrupt moderation time (usec).  "
591 			"0 to disable interrupt moderation.");
592 	SYSCTL_ADD_INT(&sc->sc_sysctl_ctx,
593 		       SYSCTL_CHILDREN(sc->sc_sysctl_tree), OID_AUTO,
594 		       "rx_ring_count", CTLFLAG_RD, &sc->sc_rx_ring_count,
595 		       0, "RX ring count");
596 	SYSCTL_ADD_INT(&sc->sc_sysctl_ctx,
597 		       SYSCTL_CHILDREN(sc->sc_sysctl_tree), OID_AUTO,
598 		       "tx_ring_count", CTLFLAG_RD, &sc->sc_tx_ring_count,
599 		       0, "TX ring count");
600 	SYSCTL_ADD_INT(&sc->sc_sysctl_ctx,
601 		       SYSCTL_CHILDREN(sc->sc_sysctl_tree), OID_AUTO,
602 		       "debug", CTLFLAG_RW, &sc->sc_debug,
603 		       0, "control debugging printfs");
604 
605 	error = mii_phy_probe(dev, &sc->sc_miibus, nfe_ifmedia_upd,
606 			      nfe_ifmedia_sts);
607 	if (error) {
608 		device_printf(dev, "MII without any phy\n");
609 		goto fail;
610 	}
611 
612 	ifp->if_softc = sc;
613 	ifp->if_mtu = ETHERMTU;
614 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
615 	ifp->if_ioctl = nfe_ioctl;
616 	ifp->if_start = nfe_start;
617 #ifdef DEVICE_POLLING
618 	ifp->if_poll = nfe_poll;
619 #endif
620 	ifp->if_watchdog = nfe_watchdog;
621 	ifp->if_init = nfe_init;
622 	ifq_set_maxlen(&ifp->if_snd, sc->sc_tx_ring_count);
623 	ifq_set_ready(&ifp->if_snd);
624 
625 	ifp->if_capabilities = IFCAP_VLAN_MTU;
626 
627 	if (sc->sc_caps & NFE_HW_VLAN)
628 		ifp->if_capabilities |= IFCAP_VLAN_HWTAGGING;
629 
630 #ifdef NFE_CSUM
631 	if (sc->sc_caps & NFE_HW_CSUM) {
632 		ifp->if_capabilities |= IFCAP_HWCSUM;
633 		ifp->if_hwassist = NFE_CSUM_FEATURES;
634 	}
635 #else
636 	sc->sc_caps &= ~NFE_HW_CSUM;
637 #endif
638 	ifp->if_capenable = ifp->if_capabilities;
639 
640 	callout_init(&sc->sc_tick_ch);
641 
642 	ether_ifattach(ifp, eaddr, NULL);
643 
644 	error = bus_setup_intr(dev, sc->sc_irq_res, INTR_MPSAFE, nfe_intr, sc,
645 			       &sc->sc_ih, ifp->if_serializer);
646 	if (error) {
647 		device_printf(dev, "could not setup intr\n");
648 		ether_ifdetach(ifp);
649 		goto fail;
650 	}
651 
652 	ifp->if_cpuid = ithread_cpuid(rman_get_start(sc->sc_irq_res));
653 	KKASSERT(ifp->if_cpuid >= 0 && ifp->if_cpuid < ncpus);
654 
655 	return 0;
656 fail:
657 	nfe_detach(dev);
658 	return error;
659 }
660 
661 static int
662 nfe_detach(device_t dev)
663 {
664 	struct nfe_softc *sc = device_get_softc(dev);
665 
666 	if (device_is_attached(dev)) {
667 		struct ifnet *ifp = &sc->arpcom.ac_if;
668 
669 		lwkt_serialize_enter(ifp->if_serializer);
670 		nfe_stop(sc);
671 		bus_teardown_intr(dev, sc->sc_irq_res, sc->sc_ih);
672 		lwkt_serialize_exit(ifp->if_serializer);
673 
674 		ether_ifdetach(ifp);
675 	}
676 
677 	if (sc->sc_miibus != NULL)
678 		device_delete_child(dev, sc->sc_miibus);
679 	bus_generic_detach(dev);
680 
681 	if (sc->sc_sysctl_tree != NULL)
682 		sysctl_ctx_free(&sc->sc_sysctl_ctx);
683 
684 	if (sc->sc_irq_res != NULL) {
685 		bus_release_resource(dev, SYS_RES_IRQ, sc->sc_irq_rid,
686 				     sc->sc_irq_res);
687 	}
688 
689 	if (sc->sc_mem_res != NULL) {
690 		bus_release_resource(dev, SYS_RES_MEMORY, sc->sc_mem_rid,
691 				     sc->sc_mem_res);
692 	}
693 
694 	nfe_free_tx_ring(sc, &sc->txq);
695 	nfe_free_rx_ring(sc, &sc->rxq);
696 	if (sc->sc_dtag != NULL)
697 		bus_dma_tag_destroy(sc->sc_dtag);
698 
699 	return 0;
700 }
701 
702 static void
703 nfe_shutdown(device_t dev)
704 {
705 	struct nfe_softc *sc = device_get_softc(dev);
706 	struct ifnet *ifp = &sc->arpcom.ac_if;
707 
708 	lwkt_serialize_enter(ifp->if_serializer);
709 	nfe_stop(sc);
710 	lwkt_serialize_exit(ifp->if_serializer);
711 }
712 
713 static int
714 nfe_suspend(device_t dev)
715 {
716 	struct nfe_softc *sc = device_get_softc(dev);
717 	struct ifnet *ifp = &sc->arpcom.ac_if;
718 
719 	lwkt_serialize_enter(ifp->if_serializer);
720 	nfe_stop(sc);
721 	lwkt_serialize_exit(ifp->if_serializer);
722 
723 	return 0;
724 }
725 
726 static int
727 nfe_resume(device_t dev)
728 {
729 	struct nfe_softc *sc = device_get_softc(dev);
730 	struct ifnet *ifp = &sc->arpcom.ac_if;
731 
732 	lwkt_serialize_enter(ifp->if_serializer);
733 	if (ifp->if_flags & IFF_UP)
734 		nfe_init(sc);
735 	lwkt_serialize_exit(ifp->if_serializer);
736 
737 	return 0;
738 }
739 
740 static void
741 nfe_miibus_statchg(device_t dev)
742 {
743 	struct nfe_softc *sc = device_get_softc(dev);
744 	struct mii_data *mii = device_get_softc(sc->sc_miibus);
745 	uint32_t phy, seed, misc = NFE_MISC1_MAGIC, link = NFE_MEDIA_SET;
746 
747 	ASSERT_SERIALIZED(sc->arpcom.ac_if.if_serializer);
748 
749 	phy = NFE_READ(sc, NFE_PHY_IFACE);
750 	phy &= ~(NFE_PHY_HDX | NFE_PHY_100TX | NFE_PHY_1000T);
751 
752 	seed = NFE_READ(sc, NFE_RNDSEED);
753 	seed &= ~NFE_SEED_MASK;
754 
755 	if ((mii->mii_media_active & IFM_GMASK) == IFM_HDX) {
756 		phy  |= NFE_PHY_HDX;	/* half-duplex */
757 		misc |= NFE_MISC1_HDX;
758 	}
759 
760 	switch (IFM_SUBTYPE(mii->mii_media_active)) {
761 	case IFM_1000_T:	/* full-duplex only */
762 		link |= NFE_MEDIA_1000T;
763 		seed |= NFE_SEED_1000T;
764 		phy  |= NFE_PHY_1000T;
765 		break;
766 	case IFM_100_TX:
767 		link |= NFE_MEDIA_100TX;
768 		seed |= NFE_SEED_100TX;
769 		phy  |= NFE_PHY_100TX;
770 		break;
771 	case IFM_10_T:
772 		link |= NFE_MEDIA_10T;
773 		seed |= NFE_SEED_10T;
774 		break;
775 	}
776 
777 	NFE_WRITE(sc, NFE_RNDSEED, seed);	/* XXX: gigabit NICs only? */
778 
779 	NFE_WRITE(sc, NFE_PHY_IFACE, phy);
780 	NFE_WRITE(sc, NFE_MISC1, misc);
781 	NFE_WRITE(sc, NFE_LINKSPEED, link);
782 }
783 
784 static int
785 nfe_miibus_readreg(device_t dev, int phy, int reg)
786 {
787 	struct nfe_softc *sc = device_get_softc(dev);
788 	uint32_t val;
789 	int ntries;
790 
791 	NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
792 
793 	if (NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY) {
794 		NFE_WRITE(sc, NFE_PHY_CTL, NFE_PHY_BUSY);
795 		DELAY(100);
796 	}
797 
798 	NFE_WRITE(sc, NFE_PHY_CTL, (phy << NFE_PHYADD_SHIFT) | reg);
799 
800 	for (ntries = 0; ntries < 1000; ntries++) {
801 		DELAY(100);
802 		if (!(NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY))
803 			break;
804 	}
805 	if (ntries == 1000) {
806 		DPRINTFN(sc, 2, "timeout waiting for PHY %s\n", "");
807 		return 0;
808 	}
809 
810 	if (NFE_READ(sc, NFE_PHY_STATUS) & NFE_PHY_ERROR) {
811 		DPRINTFN(sc, 2, "could not read PHY %s\n", "");
812 		return 0;
813 	}
814 
815 	val = NFE_READ(sc, NFE_PHY_DATA);
816 	if (val != 0xffffffff && val != 0)
817 		sc->mii_phyaddr = phy;
818 
819 	DPRINTFN(sc, 2, "mii read phy %d reg 0x%x ret 0x%x\n", phy, reg, val);
820 
821 	return val;
822 }
823 
824 static void
825 nfe_miibus_writereg(device_t dev, int phy, int reg, int val)
826 {
827 	struct nfe_softc *sc = device_get_softc(dev);
828 	uint32_t ctl;
829 	int ntries;
830 
831 	NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
832 
833 	if (NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY) {
834 		NFE_WRITE(sc, NFE_PHY_CTL, NFE_PHY_BUSY);
835 		DELAY(100);
836 	}
837 
838 	NFE_WRITE(sc, NFE_PHY_DATA, val);
839 	ctl = NFE_PHY_WRITE | (phy << NFE_PHYADD_SHIFT) | reg;
840 	NFE_WRITE(sc, NFE_PHY_CTL, ctl);
841 
842 	for (ntries = 0; ntries < 1000; ntries++) {
843 		DELAY(100);
844 		if (!(NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY))
845 			break;
846 	}
847 
848 #ifdef NFE_DEBUG
849 	if (ntries == 1000)
850 		DPRINTFN(sc, 2, "could not write to PHY %s\n", "");
851 #endif
852 }
853 
854 #ifdef DEVICE_POLLING
855 
856 static void
857 nfe_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
858 {
859 	struct nfe_softc *sc = ifp->if_softc;
860 
861 	ASSERT_SERIALIZED(ifp->if_serializer);
862 
863 	switch(cmd) {
864 	case POLL_REGISTER:
865 		nfe_disable_intrs(sc);
866 		break;
867 
868 	case POLL_DEREGISTER:
869 		nfe_enable_intrs(sc);
870 		break;
871 
872 	case POLL_AND_CHECK_STATUS:
873 		/* fall through */
874 	case POLL_ONLY:
875 		if (ifp->if_flags & IFF_RUNNING) {
876 			nfe_rxeof(sc);
877 			nfe_txeof(sc, 1);
878 		}
879 		break;
880 	}
881 }
882 
883 #endif
884 
885 static void
886 nfe_intr(void *arg)
887 {
888 	struct nfe_softc *sc = arg;
889 	struct ifnet *ifp = &sc->arpcom.ac_if;
890 	uint32_t r;
891 
892 	r = NFE_READ(sc, NFE_IRQ_STATUS);
893 	if (r == 0)
894 		return;	/* not for us */
895 	NFE_WRITE(sc, NFE_IRQ_STATUS, r);
896 
897 	if (sc->sc_rate_second != time_second) {
898 		/*
899 		 * Calculate sc_rate_avg - interrupts per second.
900 		 */
901 		sc->sc_rate_second = time_second;
902 		if (sc->sc_rate_avg < sc->sc_rate_acc)
903 			sc->sc_rate_avg = sc->sc_rate_acc;
904 		else
905 			sc->sc_rate_avg = (sc->sc_rate_avg * 3 +
906 					   sc->sc_rate_acc) / 4;
907 		sc->sc_rate_acc = 0;
908 	} else if (sc->sc_rate_avg < sc->sc_rate_acc) {
909 		/*
910 		 * Don't wait for a tick to roll over if we are taking
911 		 * a lot of interrupts.
912 		 */
913 		sc->sc_rate_avg = sc->sc_rate_acc;
914 	}
915 
916 	DPRINTFN(sc, 5, "%s: interrupt register %x\n", __func__, r);
917 
918 	if (r & NFE_IRQ_LINK) {
919 		NFE_READ(sc, NFE_PHY_STATUS);
920 		NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
921 		DPRINTF(sc, "link state changed %s\n", "");
922 	}
923 
924 	if (ifp->if_flags & IFF_RUNNING) {
925 		int ret;
926 		int rate;
927 
928 		/* check Rx ring */
929 		ret = nfe_rxeof(sc);
930 
931 		/* check Tx ring */
932 		ret |= nfe_txeof(sc, 1);
933 
934 		/* update the rate accumulator */
935 		if (ret)
936 			++sc->sc_rate_acc;
937 
938 		if (sc->sc_flags & NFE_F_DYN_IM) {
939 			rate = 1000000 / sc->sc_imtime;
940 			if ((sc->sc_flags & NFE_F_IRQ_TIMER) == 0 &&
941 			    sc->sc_rate_avg > rate) {
942 				/*
943 				 * Use the hardware timer to reduce the
944 				 * interrupt rate if the discrete interrupt
945 				 * rate has exceeded our threshold.
946 				 */
947 				NFE_WRITE(sc, NFE_IRQ_MASK, NFE_IRQ_IMTIMER);
948 				sc->sc_flags |= NFE_F_IRQ_TIMER;
949 			} else if ((sc->sc_flags & NFE_F_IRQ_TIMER) &&
950 				   sc->sc_rate_avg <= rate) {
951 				/*
952 				 * Use discrete TX/RX interrupts if the rate
953 				 * has fallen below our threshold.
954 				 */
955 				NFE_WRITE(sc, NFE_IRQ_MASK, NFE_IRQ_NOIMTIMER);
956 				sc->sc_flags &= ~NFE_F_IRQ_TIMER;
957 
958 				/*
959 				 * Recollect, mainly to avoid the possible race
960 				 * introduced by changing interrupt masks.
961 				 */
962 				nfe_rxeof(sc);
963 				nfe_txeof(sc, 1);
964 			}
965 		}
966 	}
967 }
968 
969 static int
970 nfe_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data, struct ucred *cr)
971 {
972 	struct nfe_softc *sc = ifp->if_softc;
973 	struct ifreq *ifr = (struct ifreq *)data;
974 	struct mii_data *mii;
975 	int error = 0, mask, jumbo_cap;
976 
977 	ASSERT_SERIALIZED(ifp->if_serializer);
978 
979 	switch (cmd) {
980 	case SIOCSIFMTU:
981 		if ((sc->sc_caps & NFE_JUMBO_SUP) && sc->rxq.jbuf != NULL)
982 			jumbo_cap = 1;
983 		else
984 			jumbo_cap = 0;
985 
986 		if ((jumbo_cap && ifr->ifr_mtu > NFE_JUMBO_MTU) ||
987 		    (!jumbo_cap && ifr->ifr_mtu > ETHERMTU)) {
988 			return EINVAL;
989 		} else if (ifp->if_mtu != ifr->ifr_mtu) {
990 			ifp->if_mtu = ifr->ifr_mtu;
991 			if (ifp->if_flags & IFF_RUNNING)
992 				nfe_init(sc);
993 		}
994 		break;
995 	case SIOCSIFFLAGS:
996 		if (ifp->if_flags & IFF_UP) {
997 			/*
998 			 * If only the PROMISC or ALLMULTI flag changes, then
999 			 * don't do a full re-init of the chip, just update
1000 			 * the Rx filter.
1001 			 */
1002 			if ((ifp->if_flags & IFF_RUNNING) &&
1003 			    ((ifp->if_flags ^ sc->sc_if_flags) &
1004 			     (IFF_ALLMULTI | IFF_PROMISC)) != 0) {
1005 				nfe_setmulti(sc);
1006 			} else {
1007 				if (!(ifp->if_flags & IFF_RUNNING))
1008 					nfe_init(sc);
1009 			}
1010 		} else {
1011 			if (ifp->if_flags & IFF_RUNNING)
1012 				nfe_stop(sc);
1013 		}
1014 		sc->sc_if_flags = ifp->if_flags;
1015 		break;
1016 	case SIOCADDMULTI:
1017 	case SIOCDELMULTI:
1018 		if (ifp->if_flags & IFF_RUNNING)
1019 			nfe_setmulti(sc);
1020 		break;
1021 	case SIOCSIFMEDIA:
1022 	case SIOCGIFMEDIA:
1023 		mii = device_get_softc(sc->sc_miibus);
1024 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, cmd);
1025 		break;
1026         case SIOCSIFCAP:
1027 		mask = (ifr->ifr_reqcap ^ ifp->if_capenable) & IFCAP_HWCSUM;
1028 		if (mask && (ifp->if_capabilities & IFCAP_HWCSUM)) {
1029 			ifp->if_capenable ^= mask;
1030 			if (IFCAP_TXCSUM & ifp->if_capenable)
1031 				ifp->if_hwassist = NFE_CSUM_FEATURES;
1032 			else
1033 				ifp->if_hwassist = 0;
1034 
1035 			if (ifp->if_flags & IFF_RUNNING)
1036 				nfe_init(sc);
1037 		}
1038 		break;
1039 	default:
1040 		error = ether_ioctl(ifp, cmd, data);
1041 		break;
1042 	}
1043 	return error;
1044 }
1045 
1046 static int
1047 nfe_rxeof(struct nfe_softc *sc)
1048 {
1049 	struct ifnet *ifp = &sc->arpcom.ac_if;
1050 	struct nfe_rx_ring *ring = &sc->rxq;
1051 	int reap;
1052 	struct mbuf_chain chain[MAXCPU];
1053 
1054 	reap = 0;
1055 	ether_input_chain_init(chain);
1056 
1057 	for (;;) {
1058 		struct nfe_rx_data *data = &ring->data[ring->cur];
1059 		struct mbuf *m;
1060 		uint16_t flags;
1061 		int len, error;
1062 
1063 		if (sc->sc_caps & NFE_40BIT_ADDR) {
1064 			struct nfe_desc64 *desc64 = &ring->desc64[ring->cur];
1065 
1066 			flags = le16toh(desc64->flags);
1067 			len = le16toh(desc64->length) & 0x3fff;
1068 		} else {
1069 			struct nfe_desc32 *desc32 = &ring->desc32[ring->cur];
1070 
1071 			flags = le16toh(desc32->flags);
1072 			len = le16toh(desc32->length) & 0x3fff;
1073 		}
1074 
1075 		if (flags & NFE_RX_READY)
1076 			break;
1077 
1078 		reap = 1;
1079 
1080 		if ((sc->sc_caps & (NFE_JUMBO_SUP | NFE_40BIT_ADDR)) == 0) {
1081 			if (!(flags & NFE_RX_VALID_V1))
1082 				goto skip;
1083 
1084 			if ((flags & NFE_RX_FIXME_V1) == NFE_RX_FIXME_V1) {
1085 				flags &= ~NFE_RX_ERROR;
1086 				len--;	/* fix buffer length */
1087 			}
1088 		} else {
1089 			if (!(flags & NFE_RX_VALID_V2))
1090 				goto skip;
1091 
1092 			if ((flags & NFE_RX_FIXME_V2) == NFE_RX_FIXME_V2) {
1093 				flags &= ~NFE_RX_ERROR;
1094 				len--;	/* fix buffer length */
1095 			}
1096 		}
1097 
1098 		if (flags & NFE_RX_ERROR) {
1099 			ifp->if_ierrors++;
1100 			goto skip;
1101 		}
1102 
1103 		m = data->m;
1104 
1105 		if (sc->sc_flags & NFE_F_USE_JUMBO)
1106 			error = nfe_newbuf_jumbo(sc, ring, ring->cur, 0);
1107 		else
1108 			error = nfe_newbuf_std(sc, ring, ring->cur, 0);
1109 		if (error) {
1110 			ifp->if_ierrors++;
1111 			goto skip;
1112 		}
1113 
1114 		/* finalize mbuf */
1115 		m->m_pkthdr.len = m->m_len = len;
1116 		m->m_pkthdr.rcvif = ifp;
1117 
1118 		if ((ifp->if_capenable & IFCAP_RXCSUM) &&
1119 		    (flags & NFE_RX_CSUMOK)) {
1120 			if (flags & NFE_RX_IP_CSUMOK_V2) {
1121 				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED |
1122 							  CSUM_IP_VALID;
1123 			}
1124 
1125 			if (flags &
1126 			    (NFE_RX_UDP_CSUMOK_V2 | NFE_RX_TCP_CSUMOK_V2)) {
1127 				m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
1128 							  CSUM_PSEUDO_HDR |
1129 							  CSUM_FRAG_NOT_CHECKED;
1130 				m->m_pkthdr.csum_data = 0xffff;
1131 			}
1132 		}
1133 
1134 		ifp->if_ipackets++;
1135 		ether_input_chain(ifp, m, NULL, chain);
1136 skip:
1137 		nfe_set_ready_rxdesc(sc, ring, ring->cur);
1138 		sc->rxq.cur = (sc->rxq.cur + 1) % sc->sc_rx_ring_count;
1139 	}
1140 
1141 	if (reap)
1142 		ether_input_dispatch(chain);
1143 	return reap;
1144 }
1145 
1146 static int
1147 nfe_txeof(struct nfe_softc *sc, int start)
1148 {
1149 	struct ifnet *ifp = &sc->arpcom.ac_if;
1150 	struct nfe_tx_ring *ring = &sc->txq;
1151 	struct nfe_tx_data *data = NULL;
1152 
1153 	while (ring->next != ring->cur) {
1154 		uint16_t flags;
1155 
1156 		if (sc->sc_caps & NFE_40BIT_ADDR)
1157 			flags = le16toh(ring->desc64[ring->next].flags);
1158 		else
1159 			flags = le16toh(ring->desc32[ring->next].flags);
1160 
1161 		if (flags & NFE_TX_VALID)
1162 			break;
1163 
1164 		data = &ring->data[ring->next];
1165 
1166 		if ((sc->sc_caps & (NFE_JUMBO_SUP | NFE_40BIT_ADDR)) == 0) {
1167 			if (!(flags & NFE_TX_LASTFRAG_V1) && data->m == NULL)
1168 				goto skip;
1169 
1170 			if ((flags & NFE_TX_ERROR_V1) != 0) {
1171 				if_printf(ifp, "tx v1 error 0x%4b\n", flags,
1172 					  NFE_V1_TXERR);
1173 				ifp->if_oerrors++;
1174 			} else {
1175 				ifp->if_opackets++;
1176 			}
1177 		} else {
1178 			if (!(flags & NFE_TX_LASTFRAG_V2) && data->m == NULL)
1179 				goto skip;
1180 
1181 			if ((flags & NFE_TX_ERROR_V2) != 0) {
1182 				if_printf(ifp, "tx v2 error 0x%4b\n", flags,
1183 					  NFE_V2_TXERR);
1184 				ifp->if_oerrors++;
1185 			} else {
1186 				ifp->if_opackets++;
1187 			}
1188 		}
1189 
1190 		if (data->m == NULL) {	/* should not get there */
1191 			if_printf(ifp,
1192 				  "last fragment bit w/o associated mbuf!\n");
1193 			goto skip;
1194 		}
1195 
1196 		/* last fragment of the mbuf chain transmitted */
1197 		bus_dmamap_unload(ring->data_tag, data->map);
1198 		m_freem(data->m);
1199 		data->m = NULL;
1200 skip:
1201 		ring->queued--;
1202 		KKASSERT(ring->queued >= 0);
1203 		ring->next = (ring->next + 1) % sc->sc_tx_ring_count;
1204 	}
1205 
1206 	if (sc->sc_tx_ring_count - ring->queued >=
1207 	    sc->sc_tx_spare + NFE_NSEG_RSVD)
1208 		ifp->if_flags &= ~IFF_OACTIVE;
1209 
1210 	if (ring->queued == 0)
1211 		ifp->if_timer = 0;
1212 
1213 	if (start && !ifq_is_empty(&ifp->if_snd))
1214 		if_devstart(ifp);
1215 
1216 	if (data != NULL)
1217 		return 1;
1218 	else
1219 		return 0;
1220 }
1221 
1222 static int
1223 nfe_encap(struct nfe_softc *sc, struct nfe_tx_ring *ring, struct mbuf *m0)
1224 {
1225 	bus_dma_segment_t segs[NFE_MAX_SCATTER];
1226 	struct nfe_tx_data *data, *data_map;
1227 	bus_dmamap_t map;
1228 	struct nfe_desc64 *desc64 = NULL;
1229 	struct nfe_desc32 *desc32 = NULL;
1230 	uint16_t flags = 0;
1231 	uint32_t vtag = 0;
1232 	int error, i, j, maxsegs, nsegs;
1233 
1234 	data = &ring->data[ring->cur];
1235 	map = data->map;
1236 	data_map = data;	/* Remember who owns the DMA map */
1237 
1238 	maxsegs = (sc->sc_tx_ring_count - ring->queued) - NFE_NSEG_RSVD;
1239 	if (maxsegs > NFE_MAX_SCATTER)
1240 		maxsegs = NFE_MAX_SCATTER;
1241 	KASSERT(maxsegs >= sc->sc_tx_spare,
1242 		("no enough segments %d,%d\n", maxsegs, sc->sc_tx_spare));
1243 
1244 	error = bus_dmamap_load_mbuf_defrag(ring->data_tag, map, &m0,
1245 			segs, maxsegs, &nsegs, BUS_DMA_NOWAIT);
1246 	if (error)
1247 		goto back;
1248 	bus_dmamap_sync(ring->data_tag, map, BUS_DMASYNC_PREWRITE);
1249 
1250 	error = 0;
1251 
1252 	/* setup h/w VLAN tagging */
1253 	if (m0->m_flags & M_VLANTAG)
1254 		vtag = m0->m_pkthdr.ether_vlantag;
1255 
1256 	if (sc->arpcom.ac_if.if_capenable & IFCAP_TXCSUM) {
1257 		if (m0->m_pkthdr.csum_flags & CSUM_IP)
1258 			flags |= NFE_TX_IP_CSUM;
1259 		if (m0->m_pkthdr.csum_flags & (CSUM_TCP | CSUM_UDP))
1260 			flags |= NFE_TX_TCP_CSUM;
1261 	}
1262 
1263 	/*
1264 	 * XXX urm. somebody is unaware of how hardware works.  You
1265 	 * absolutely CANNOT set NFE_TX_VALID on the next descriptor in
1266 	 * the ring until the entire chain is actually *VALID*.  Otherwise
1267 	 * the hardware may encounter a partially initialized chain that
1268 	 * is marked as being ready to go when it in fact is not ready to
1269 	 * go.
1270 	 */
1271 
1272 	for (i = 0; i < nsegs; i++) {
1273 		j = (ring->cur + i) % sc->sc_tx_ring_count;
1274 		data = &ring->data[j];
1275 
1276 		if (sc->sc_caps & NFE_40BIT_ADDR) {
1277 			desc64 = &ring->desc64[j];
1278 			desc64->physaddr[0] =
1279 			    htole32(NFE_ADDR_HI(segs[i].ds_addr));
1280 			desc64->physaddr[1] =
1281 			    htole32(NFE_ADDR_LO(segs[i].ds_addr));
1282 			desc64->length = htole16(segs[i].ds_len - 1);
1283 			desc64->vtag = htole32(vtag);
1284 			desc64->flags = htole16(flags);
1285 		} else {
1286 			desc32 = &ring->desc32[j];
1287 			desc32->physaddr = htole32(segs[i].ds_addr);
1288 			desc32->length = htole16(segs[i].ds_len - 1);
1289 			desc32->flags = htole16(flags);
1290 		}
1291 
1292 		/* csum flags and vtag belong to the first fragment only */
1293 		flags &= ~(NFE_TX_IP_CSUM | NFE_TX_TCP_CSUM);
1294 		vtag = 0;
1295 
1296 		ring->queued++;
1297 		KKASSERT(ring->queued <= sc->sc_tx_ring_count);
1298 	}
1299 
1300 	/* the whole mbuf chain has been DMA mapped, fix last descriptor */
1301 	if (sc->sc_caps & NFE_40BIT_ADDR) {
1302 		desc64->flags |= htole16(NFE_TX_LASTFRAG_V2);
1303 	} else {
1304 		if (sc->sc_caps & NFE_JUMBO_SUP)
1305 			flags = NFE_TX_LASTFRAG_V2;
1306 		else
1307 			flags = NFE_TX_LASTFRAG_V1;
1308 		desc32->flags |= htole16(flags);
1309 	}
1310 
1311 	/*
1312 	 * Set NFE_TX_VALID backwards so the hardware doesn't see the
1313 	 * whole mess until the first descriptor in the map is flagged.
1314 	 */
1315 	for (i = nsegs - 1; i >= 0; --i) {
1316 		j = (ring->cur + i) % sc->sc_tx_ring_count;
1317 		if (sc->sc_caps & NFE_40BIT_ADDR) {
1318 			desc64 = &ring->desc64[j];
1319 			desc64->flags |= htole16(NFE_TX_VALID);
1320 		} else {
1321 			desc32 = &ring->desc32[j];
1322 			desc32->flags |= htole16(NFE_TX_VALID);
1323 		}
1324 	}
1325 	ring->cur = (ring->cur + nsegs) % sc->sc_tx_ring_count;
1326 
1327 	/* Exchange DMA map */
1328 	data_map->map = data->map;
1329 	data->map = map;
1330 	data->m = m0;
1331 back:
1332 	if (error)
1333 		m_freem(m0);
1334 	return error;
1335 }
1336 
1337 static void
1338 nfe_start(struct ifnet *ifp)
1339 {
1340 	struct nfe_softc *sc = ifp->if_softc;
1341 	struct nfe_tx_ring *ring = &sc->txq;
1342 	int count = 0, oactive = 0;
1343 	struct mbuf *m0;
1344 
1345 	ASSERT_SERIALIZED(ifp->if_serializer);
1346 
1347 	if ((ifp->if_flags & (IFF_OACTIVE | IFF_RUNNING)) != IFF_RUNNING)
1348 		return;
1349 
1350 	for (;;) {
1351 		int error;
1352 
1353 		if (sc->sc_tx_ring_count - ring->queued <
1354 		    sc->sc_tx_spare + NFE_NSEG_RSVD) {
1355 			if (oactive) {
1356 				ifp->if_flags |= IFF_OACTIVE;
1357 				break;
1358 			}
1359 
1360 			nfe_txeof(sc, 0);
1361 			oactive = 1;
1362 			continue;
1363 		}
1364 
1365 		m0 = ifq_dequeue(&ifp->if_snd, NULL);
1366 		if (m0 == NULL)
1367 			break;
1368 
1369 		ETHER_BPF_MTAP(ifp, m0);
1370 
1371 		error = nfe_encap(sc, ring, m0);
1372 		if (error) {
1373 			ifp->if_oerrors++;
1374 			if (error == EFBIG) {
1375 				if (oactive) {
1376 					ifp->if_flags |= IFF_OACTIVE;
1377 					break;
1378 				}
1379 				nfe_txeof(sc, 0);
1380 				oactive = 1;
1381 			}
1382 			continue;
1383 		} else {
1384 			oactive = 0;
1385 		}
1386 		++count;
1387 
1388 		/*
1389 		 * NOTE:
1390 		 * `m0' may be freed in nfe_encap(), so
1391 		 * it should not be touched any more.
1392 		 */
1393 	}
1394 
1395 	if (count == 0)	/* nothing sent */
1396 		return;
1397 
1398 	/* Kick Tx */
1399 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_KICKTX | sc->rxtxctl);
1400 
1401 	/*
1402 	 * Set a timeout in case the chip goes out to lunch.
1403 	 */
1404 	ifp->if_timer = 5;
1405 }
1406 
1407 static void
1408 nfe_watchdog(struct ifnet *ifp)
1409 {
1410 	struct nfe_softc *sc = ifp->if_softc;
1411 
1412 	ASSERT_SERIALIZED(ifp->if_serializer);
1413 
1414 	if (ifp->if_flags & IFF_RUNNING) {
1415 		if_printf(ifp, "watchdog timeout - lost interrupt recovered\n");
1416 		nfe_txeof(sc, 1);
1417 		return;
1418 	}
1419 
1420 	if_printf(ifp, "watchdog timeout\n");
1421 
1422 	nfe_init(ifp->if_softc);
1423 
1424 	ifp->if_oerrors++;
1425 }
1426 
1427 static void
1428 nfe_init(void *xsc)
1429 {
1430 	struct nfe_softc *sc = xsc;
1431 	struct ifnet *ifp = &sc->arpcom.ac_if;
1432 	uint32_t tmp;
1433 	int error;
1434 
1435 	ASSERT_SERIALIZED(ifp->if_serializer);
1436 
1437 	nfe_stop(sc);
1438 
1439 	if ((sc->sc_caps & NFE_NO_PWRCTL) == 0)
1440 		nfe_mac_reset(sc);
1441 
1442 	/*
1443 	 * NOTE:
1444 	 * Switching between jumbo frames and normal frames should
1445 	 * be done _after_ nfe_stop() but _before_ nfe_init_rx_ring().
1446 	 */
1447 	if (ifp->if_mtu > ETHERMTU) {
1448 		sc->sc_flags |= NFE_F_USE_JUMBO;
1449 		sc->rxq.bufsz = NFE_JBYTES;
1450 		sc->sc_tx_spare = NFE_NSEG_SPARE_JUMBO;
1451 		if (bootverbose)
1452 			if_printf(ifp, "use jumbo frames\n");
1453 	} else {
1454 		sc->sc_flags &= ~NFE_F_USE_JUMBO;
1455 		sc->rxq.bufsz = MCLBYTES;
1456 		sc->sc_tx_spare = NFE_NSEG_SPARE;
1457 		if (bootverbose)
1458 			if_printf(ifp, "use non-jumbo frames\n");
1459 	}
1460 
1461 	error = nfe_init_tx_ring(sc, &sc->txq);
1462 	if (error) {
1463 		nfe_stop(sc);
1464 		return;
1465 	}
1466 
1467 	error = nfe_init_rx_ring(sc, &sc->rxq);
1468 	if (error) {
1469 		nfe_stop(sc);
1470 		return;
1471 	}
1472 
1473 	NFE_WRITE(sc, NFE_TX_POLL, 0);
1474 	NFE_WRITE(sc, NFE_STATUS, 0);
1475 
1476 	sc->rxtxctl = NFE_RXTX_BIT2 | sc->rxtxctl_desc;
1477 
1478 	if (ifp->if_capenable & IFCAP_RXCSUM)
1479 		sc->rxtxctl |= NFE_RXTX_RXCSUM;
1480 
1481 	/*
1482 	 * Although the adapter is capable of stripping VLAN tags from received
1483 	 * frames (NFE_RXTX_VTAG_STRIP), we do not enable this functionality on
1484 	 * purpose.  This will be done in software by our network stack.
1485 	 */
1486 	if (sc->sc_caps & NFE_HW_VLAN)
1487 		sc->rxtxctl |= NFE_RXTX_VTAG_INSERT;
1488 
1489 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_RESET | sc->rxtxctl);
1490 	DELAY(10);
1491 	NFE_WRITE(sc, NFE_RXTX_CTL, sc->rxtxctl);
1492 
1493 	if (sc->sc_caps & NFE_HW_VLAN)
1494 		NFE_WRITE(sc, NFE_VTAG_CTL, NFE_VTAG_ENABLE);
1495 
1496 	NFE_WRITE(sc, NFE_SETUP_R6, 0);
1497 
1498 	/* set MAC address */
1499 	nfe_set_macaddr(sc, sc->arpcom.ac_enaddr);
1500 
1501 	/* tell MAC where rings are in memory */
1502 	if (sc->sc_caps & NFE_40BIT_ADDR) {
1503 		NFE_WRITE(sc, NFE_RX_RING_ADDR_HI,
1504 			  NFE_ADDR_HI(sc->rxq.physaddr));
1505 	}
1506 	NFE_WRITE(sc, NFE_RX_RING_ADDR_LO, NFE_ADDR_LO(sc->rxq.physaddr));
1507 
1508 	if (sc->sc_caps & NFE_40BIT_ADDR) {
1509 		NFE_WRITE(sc, NFE_TX_RING_ADDR_HI,
1510 			  NFE_ADDR_HI(sc->txq.physaddr));
1511 	}
1512 	NFE_WRITE(sc, NFE_TX_RING_ADDR_LO, NFE_ADDR_LO(sc->txq.physaddr));
1513 
1514 	NFE_WRITE(sc, NFE_RING_SIZE,
1515 	    (sc->sc_rx_ring_count - 1) << 16 |
1516 	    (sc->sc_tx_ring_count - 1));
1517 
1518 	NFE_WRITE(sc, NFE_RXBUFSZ, sc->rxq.bufsz);
1519 
1520 	/* force MAC to wakeup */
1521 	tmp = NFE_READ(sc, NFE_PWR_STATE);
1522 	NFE_WRITE(sc, NFE_PWR_STATE, tmp | NFE_PWR_WAKEUP);
1523 	DELAY(10);
1524 	tmp = NFE_READ(sc, NFE_PWR_STATE);
1525 	NFE_WRITE(sc, NFE_PWR_STATE, tmp | NFE_PWR_VALID);
1526 
1527 	NFE_WRITE(sc, NFE_SETUP_R1, NFE_R1_MAGIC);
1528 	NFE_WRITE(sc, NFE_SETUP_R2, NFE_R2_MAGIC);
1529 	NFE_WRITE(sc, NFE_SETUP_R6, NFE_R6_MAGIC);
1530 
1531 	/* update MAC knowledge of PHY; generates a NFE_IRQ_LINK interrupt */
1532 	NFE_WRITE(sc, NFE_STATUS, sc->mii_phyaddr << 24 | NFE_STATUS_MAGIC);
1533 
1534 	NFE_WRITE(sc, NFE_SETUP_R4, NFE_R4_MAGIC);
1535 
1536 	sc->rxtxctl &= ~NFE_RXTX_BIT2;
1537 	NFE_WRITE(sc, NFE_RXTX_CTL, sc->rxtxctl);
1538 	DELAY(10);
1539 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_BIT1 | sc->rxtxctl);
1540 
1541 	/* set Rx filter */
1542 	nfe_setmulti(sc);
1543 
1544 	nfe_ifmedia_upd(ifp);
1545 
1546 	/* enable Rx */
1547 	NFE_WRITE(sc, NFE_RX_CTL, NFE_RX_START);
1548 
1549 	/* enable Tx */
1550 	NFE_WRITE(sc, NFE_TX_CTL, NFE_TX_START);
1551 
1552 	NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
1553 
1554 #ifdef DEVICE_POLLING
1555 	if ((ifp->if_flags & IFF_POLLING))
1556 		nfe_disable_intrs(sc);
1557 	else
1558 #endif
1559 	nfe_enable_intrs(sc);
1560 
1561 	callout_reset(&sc->sc_tick_ch, hz, nfe_tick, sc);
1562 
1563 	ifp->if_flags |= IFF_RUNNING;
1564 	ifp->if_flags &= ~IFF_OACTIVE;
1565 
1566 	/*
1567 	 * If we had stuff in the tx ring before its all cleaned out now
1568 	 * so we are not going to get an interrupt, jump-start any pending
1569 	 * output.
1570 	 */
1571 	if (!ifq_is_empty(&ifp->if_snd))
1572 		if_devstart(ifp);
1573 }
1574 
1575 static void
1576 nfe_stop(struct nfe_softc *sc)
1577 {
1578 	struct ifnet *ifp = &sc->arpcom.ac_if;
1579 	uint32_t rxtxctl = sc->rxtxctl_desc | NFE_RXTX_BIT2;
1580 	int i;
1581 
1582 	ASSERT_SERIALIZED(ifp->if_serializer);
1583 
1584 	callout_stop(&sc->sc_tick_ch);
1585 
1586 	ifp->if_timer = 0;
1587 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
1588 	sc->sc_flags &= ~NFE_F_IRQ_TIMER;
1589 
1590 #define WAITMAX	50000
1591 
1592 	/*
1593 	 * Abort Tx
1594 	 */
1595 	NFE_WRITE(sc, NFE_TX_CTL, 0);
1596 	for (i = 0; i < WAITMAX; ++i) {
1597 		DELAY(100);
1598 		if ((NFE_READ(sc, NFE_TX_STATUS) & NFE_TX_STATUS_BUSY) == 0)
1599 			break;
1600 	}
1601 	if (i == WAITMAX)
1602 		if_printf(ifp, "can't stop TX\n");
1603 	DELAY(100);
1604 
1605 	/*
1606 	 * Disable Rx
1607 	 */
1608 	NFE_WRITE(sc, NFE_RX_CTL, 0);
1609 	for (i = 0; i < WAITMAX; ++i) {
1610 		DELAY(100);
1611 		if ((NFE_READ(sc, NFE_RX_STATUS) & NFE_RX_STATUS_BUSY) == 0)
1612 			break;
1613 	}
1614 	if (i == WAITMAX)
1615 		if_printf(ifp, "can't stop RX\n");
1616 	DELAY(100);
1617 
1618 #undef WAITMAX
1619 
1620 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_RESET | rxtxctl);
1621 	DELAY(10);
1622 	NFE_WRITE(sc, NFE_RXTX_CTL, rxtxctl);
1623 
1624 	/* Disable interrupts */
1625 	NFE_WRITE(sc, NFE_IRQ_MASK, 0);
1626 
1627 	/* Reset Tx and Rx rings */
1628 	nfe_reset_tx_ring(sc, &sc->txq);
1629 	nfe_reset_rx_ring(sc, &sc->rxq);
1630 }
1631 
1632 static int
1633 nfe_alloc_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1634 {
1635 	int i, j, error, descsize;
1636 	bus_dmamem_t dmem;
1637 	void **desc;
1638 
1639 	if (sc->sc_caps & NFE_40BIT_ADDR) {
1640 		desc = (void **)&ring->desc64;
1641 		descsize = sizeof(struct nfe_desc64);
1642 	} else {
1643 		desc = (void **)&ring->desc32;
1644 		descsize = sizeof(struct nfe_desc32);
1645 	}
1646 
1647 	ring->bufsz = MCLBYTES;
1648 	ring->cur = ring->next = 0;
1649 
1650 	error = bus_dmamem_coherent(sc->sc_dtag, PAGE_SIZE, 0,
1651 				    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
1652 				    sc->sc_rx_ring_count * descsize,
1653 				    BUS_DMA_WAITOK | BUS_DMA_ZERO, &dmem);
1654 	if (error) {
1655 		if_printf(&sc->arpcom.ac_if,
1656 			  "could not create RX desc ring\n");
1657 		return error;
1658 	}
1659 	ring->tag = dmem.dmem_tag;
1660 	ring->map = dmem.dmem_map;
1661 	*desc = dmem.dmem_addr;
1662 	ring->physaddr = dmem.dmem_busaddr;
1663 
1664 	if (sc->sc_caps & NFE_JUMBO_SUP) {
1665 		ring->jbuf =
1666 		kmalloc(sizeof(struct nfe_jbuf) * NFE_JPOOL_COUNT(sc),
1667 			M_DEVBUF, M_WAITOK | M_ZERO);
1668 
1669 		error = nfe_jpool_alloc(sc, ring);
1670 		if (error) {
1671 			if_printf(&sc->arpcom.ac_if,
1672 				  "could not allocate jumbo frames\n");
1673 			kfree(ring->jbuf, M_DEVBUF);
1674 			ring->jbuf = NULL;
1675 			/* Allow jumbo frame allocation to fail */
1676 		}
1677 	}
1678 
1679 	ring->data = kmalloc(sizeof(struct nfe_rx_data) * sc->sc_rx_ring_count,
1680 			     M_DEVBUF, M_WAITOK | M_ZERO);
1681 
1682 	error = bus_dma_tag_create(sc->sc_dtag, 1, 0,
1683 				   BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
1684 				   NULL, NULL,
1685 				   MCLBYTES, 1, MCLBYTES,
1686 				   BUS_DMA_ALLOCNOW | BUS_DMA_WAITOK,
1687 				   &ring->data_tag);
1688 	if (error) {
1689 		if_printf(&sc->arpcom.ac_if,
1690 			  "could not create RX mbuf DMA tag\n");
1691 		return error;
1692 	}
1693 
1694 	/* Create a spare RX mbuf DMA map */
1695 	error = bus_dmamap_create(ring->data_tag, BUS_DMA_WAITOK,
1696 				  &ring->data_tmpmap);
1697 	if (error) {
1698 		if_printf(&sc->arpcom.ac_if,
1699 			  "could not create spare RX mbuf DMA map\n");
1700 		bus_dma_tag_destroy(ring->data_tag);
1701 		ring->data_tag = NULL;
1702 		return error;
1703 	}
1704 
1705 	for (i = 0; i < sc->sc_rx_ring_count; i++) {
1706 		error = bus_dmamap_create(ring->data_tag, BUS_DMA_WAITOK,
1707 					  &ring->data[i].map);
1708 		if (error) {
1709 			if_printf(&sc->arpcom.ac_if,
1710 				  "could not create %dth RX mbuf DMA mapn", i);
1711 			goto fail;
1712 		}
1713 	}
1714 	return 0;
1715 fail:
1716 	for (j = 0; j < i; ++j)
1717 		bus_dmamap_destroy(ring->data_tag, ring->data[i].map);
1718 	bus_dmamap_destroy(ring->data_tag, ring->data_tmpmap);
1719 	bus_dma_tag_destroy(ring->data_tag);
1720 	ring->data_tag = NULL;
1721 	return error;
1722 }
1723 
1724 static void
1725 nfe_reset_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1726 {
1727 	int i;
1728 
1729 	for (i = 0; i < sc->sc_rx_ring_count; i++) {
1730 		struct nfe_rx_data *data = &ring->data[i];
1731 
1732 		if (data->m != NULL) {
1733 			if ((sc->sc_flags & NFE_F_USE_JUMBO) == 0)
1734 				bus_dmamap_unload(ring->data_tag, data->map);
1735 			m_freem(data->m);
1736 			data->m = NULL;
1737 		}
1738 	}
1739 
1740 	ring->cur = ring->next = 0;
1741 }
1742 
1743 static int
1744 nfe_init_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1745 {
1746 	int i;
1747 
1748 	for (i = 0; i < sc->sc_rx_ring_count; ++i) {
1749 		int error;
1750 
1751 		/* XXX should use a function pointer */
1752 		if (sc->sc_flags & NFE_F_USE_JUMBO)
1753 			error = nfe_newbuf_jumbo(sc, ring, i, 1);
1754 		else
1755 			error = nfe_newbuf_std(sc, ring, i, 1);
1756 		if (error) {
1757 			if_printf(&sc->arpcom.ac_if,
1758 				  "could not allocate RX buffer\n");
1759 			return error;
1760 		}
1761 		nfe_set_ready_rxdesc(sc, ring, i);
1762 	}
1763 	return 0;
1764 }
1765 
1766 static void
1767 nfe_free_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1768 {
1769 	if (ring->data_tag != NULL) {
1770 		struct nfe_rx_data *data;
1771 		int i;
1772 
1773 		for (i = 0; i < sc->sc_rx_ring_count; i++) {
1774 			data = &ring->data[i];
1775 
1776 			if (data->m != NULL) {
1777 				bus_dmamap_unload(ring->data_tag, data->map);
1778 				m_freem(data->m);
1779 			}
1780 			bus_dmamap_destroy(ring->data_tag, data->map);
1781 		}
1782 		bus_dmamap_destroy(ring->data_tag, ring->data_tmpmap);
1783 		bus_dma_tag_destroy(ring->data_tag);
1784 	}
1785 
1786 	nfe_jpool_free(sc, ring);
1787 
1788 	if (ring->jbuf != NULL)
1789 		kfree(ring->jbuf, M_DEVBUF);
1790 	if (ring->data != NULL)
1791 		kfree(ring->data, M_DEVBUF);
1792 
1793 	if (ring->tag != NULL) {
1794 		void *desc;
1795 
1796 		if (sc->sc_caps & NFE_40BIT_ADDR)
1797 			desc = ring->desc64;
1798 		else
1799 			desc = ring->desc32;
1800 
1801 		bus_dmamap_unload(ring->tag, ring->map);
1802 		bus_dmamem_free(ring->tag, desc, ring->map);
1803 		bus_dma_tag_destroy(ring->tag);
1804 	}
1805 }
1806 
1807 static struct nfe_jbuf *
1808 nfe_jalloc(struct nfe_softc *sc)
1809 {
1810 	struct ifnet *ifp = &sc->arpcom.ac_if;
1811 	struct nfe_jbuf *jbuf;
1812 
1813 	lwkt_serialize_enter(&sc->sc_jbuf_serializer);
1814 
1815 	jbuf = SLIST_FIRST(&sc->rxq.jfreelist);
1816 	if (jbuf != NULL) {
1817 		SLIST_REMOVE_HEAD(&sc->rxq.jfreelist, jnext);
1818 		jbuf->inuse = 1;
1819 	} else {
1820 		if_printf(ifp, "no free jumbo buffer\n");
1821 	}
1822 
1823 	lwkt_serialize_exit(&sc->sc_jbuf_serializer);
1824 
1825 	return jbuf;
1826 }
1827 
1828 static void
1829 nfe_jfree(void *arg)
1830 {
1831 	struct nfe_jbuf *jbuf = arg;
1832 	struct nfe_softc *sc = jbuf->sc;
1833 	struct nfe_rx_ring *ring = jbuf->ring;
1834 
1835 	if (&ring->jbuf[jbuf->slot] != jbuf)
1836 		panic("%s: free wrong jumbo buffer\n", __func__);
1837 	else if (jbuf->inuse == 0)
1838 		panic("%s: jumbo buffer already freed\n", __func__);
1839 
1840 	lwkt_serialize_enter(&sc->sc_jbuf_serializer);
1841 	atomic_subtract_int(&jbuf->inuse, 1);
1842 	if (jbuf->inuse == 0)
1843 		SLIST_INSERT_HEAD(&ring->jfreelist, jbuf, jnext);
1844 	lwkt_serialize_exit(&sc->sc_jbuf_serializer);
1845 }
1846 
1847 static void
1848 nfe_jref(void *arg)
1849 {
1850 	struct nfe_jbuf *jbuf = arg;
1851 	struct nfe_rx_ring *ring = jbuf->ring;
1852 
1853 	if (&ring->jbuf[jbuf->slot] != jbuf)
1854 		panic("%s: ref wrong jumbo buffer\n", __func__);
1855 	else if (jbuf->inuse == 0)
1856 		panic("%s: jumbo buffer already freed\n", __func__);
1857 
1858 	atomic_add_int(&jbuf->inuse, 1);
1859 }
1860 
1861 static int
1862 nfe_jpool_alloc(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1863 {
1864 	struct nfe_jbuf *jbuf;
1865 	bus_dmamem_t dmem;
1866 	bus_addr_t physaddr;
1867 	caddr_t buf;
1868 	int i, error;
1869 
1870 	/*
1871 	 * Allocate a big chunk of DMA'able memory.
1872 	 */
1873 	error = bus_dmamem_coherent(sc->sc_dtag, PAGE_SIZE, 0,
1874 				    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
1875 				    NFE_JPOOL_SIZE(sc),
1876 				    BUS_DMA_WAITOK, &dmem);
1877 	if (error) {
1878 		if_printf(&sc->arpcom.ac_if,
1879 			  "could not create jumbo buffer\n");
1880 		return error;
1881 	}
1882 	ring->jtag = dmem.dmem_tag;
1883 	ring->jmap = dmem.dmem_map;
1884 	ring->jpool = dmem.dmem_addr;
1885 	physaddr = dmem.dmem_busaddr;
1886 
1887 	/* ..and split it into 9KB chunks */
1888 	SLIST_INIT(&ring->jfreelist);
1889 
1890 	buf = ring->jpool;
1891 	for (i = 0; i < NFE_JPOOL_COUNT(sc); i++) {
1892 		jbuf = &ring->jbuf[i];
1893 
1894 		jbuf->sc = sc;
1895 		jbuf->ring = ring;
1896 		jbuf->inuse = 0;
1897 		jbuf->slot = i;
1898 		jbuf->buf = buf;
1899 		jbuf->physaddr = physaddr;
1900 
1901 		SLIST_INSERT_HEAD(&ring->jfreelist, jbuf, jnext);
1902 
1903 		buf += NFE_JBYTES;
1904 		physaddr += NFE_JBYTES;
1905 	}
1906 
1907 	return 0;
1908 }
1909 
1910 static void
1911 nfe_jpool_free(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1912 {
1913 	if (ring->jtag != NULL) {
1914 		bus_dmamap_unload(ring->jtag, ring->jmap);
1915 		bus_dmamem_free(ring->jtag, ring->jpool, ring->jmap);
1916 		bus_dma_tag_destroy(ring->jtag);
1917 	}
1918 }
1919 
1920 static int
1921 nfe_alloc_tx_ring(struct nfe_softc *sc, struct nfe_tx_ring *ring)
1922 {
1923 	int i, j, error, descsize;
1924 	bus_dmamem_t dmem;
1925 	void **desc;
1926 
1927 	if (sc->sc_caps & NFE_40BIT_ADDR) {
1928 		desc = (void **)&ring->desc64;
1929 		descsize = sizeof(struct nfe_desc64);
1930 	} else {
1931 		desc = (void **)&ring->desc32;
1932 		descsize = sizeof(struct nfe_desc32);
1933 	}
1934 
1935 	ring->queued = 0;
1936 	ring->cur = ring->next = 0;
1937 
1938 	error = bus_dmamem_coherent(sc->sc_dtag, PAGE_SIZE, 0,
1939 				    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
1940 				    sc->sc_tx_ring_count * descsize,
1941 				    BUS_DMA_WAITOK | BUS_DMA_ZERO, &dmem);
1942 	if (error) {
1943 		if_printf(&sc->arpcom.ac_if,
1944 			  "could not create TX desc ring\n");
1945 		return error;
1946 	}
1947 	ring->tag = dmem.dmem_tag;
1948 	ring->map = dmem.dmem_map;
1949 	*desc = dmem.dmem_addr;
1950 	ring->physaddr = dmem.dmem_busaddr;
1951 
1952 	ring->data = kmalloc(sizeof(struct nfe_tx_data) * sc->sc_tx_ring_count,
1953 			     M_DEVBUF, M_WAITOK | M_ZERO);
1954 
1955 	error = bus_dma_tag_create(sc->sc_dtag, 1, 0,
1956 			BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
1957 			NULL, NULL,
1958 			NFE_JBYTES, NFE_MAX_SCATTER, MCLBYTES,
1959 			BUS_DMA_ALLOCNOW | BUS_DMA_WAITOK | BUS_DMA_ONEBPAGE,
1960 			&ring->data_tag);
1961 	if (error) {
1962 		if_printf(&sc->arpcom.ac_if,
1963 			  "could not create TX buf DMA tag\n");
1964 		return error;
1965 	}
1966 
1967 	for (i = 0; i < sc->sc_tx_ring_count; i++) {
1968 		error = bus_dmamap_create(ring->data_tag,
1969 				BUS_DMA_WAITOK | BUS_DMA_ONEBPAGE,
1970 				&ring->data[i].map);
1971 		if (error) {
1972 			if_printf(&sc->arpcom.ac_if,
1973 				  "could not create %dth TX buf DMA map\n", i);
1974 			goto fail;
1975 		}
1976 	}
1977 
1978 	return 0;
1979 fail:
1980 	for (j = 0; j < i; ++j)
1981 		bus_dmamap_destroy(ring->data_tag, ring->data[i].map);
1982 	bus_dma_tag_destroy(ring->data_tag);
1983 	ring->data_tag = NULL;
1984 	return error;
1985 }
1986 
1987 static void
1988 nfe_reset_tx_ring(struct nfe_softc *sc, struct nfe_tx_ring *ring)
1989 {
1990 	int i;
1991 
1992 	for (i = 0; i < sc->sc_tx_ring_count; i++) {
1993 		struct nfe_tx_data *data = &ring->data[i];
1994 
1995 		if (sc->sc_caps & NFE_40BIT_ADDR)
1996 			ring->desc64[i].flags = 0;
1997 		else
1998 			ring->desc32[i].flags = 0;
1999 
2000 		if (data->m != NULL) {
2001 			bus_dmamap_unload(ring->data_tag, data->map);
2002 			m_freem(data->m);
2003 			data->m = NULL;
2004 		}
2005 	}
2006 
2007 	ring->queued = 0;
2008 	ring->cur = ring->next = 0;
2009 }
2010 
2011 static int
2012 nfe_init_tx_ring(struct nfe_softc *sc __unused,
2013 		 struct nfe_tx_ring *ring __unused)
2014 {
2015 	return 0;
2016 }
2017 
2018 static void
2019 nfe_free_tx_ring(struct nfe_softc *sc, struct nfe_tx_ring *ring)
2020 {
2021 	if (ring->data_tag != NULL) {
2022 		struct nfe_tx_data *data;
2023 		int i;
2024 
2025 		for (i = 0; i < sc->sc_tx_ring_count; ++i) {
2026 			data = &ring->data[i];
2027 
2028 			if (data->m != NULL) {
2029 				bus_dmamap_unload(ring->data_tag, data->map);
2030 				m_freem(data->m);
2031 			}
2032 			bus_dmamap_destroy(ring->data_tag, data->map);
2033 		}
2034 
2035 		bus_dma_tag_destroy(ring->data_tag);
2036 	}
2037 
2038 	if (ring->data != NULL)
2039 		kfree(ring->data, M_DEVBUF);
2040 
2041 	if (ring->tag != NULL) {
2042 		void *desc;
2043 
2044 		if (sc->sc_caps & NFE_40BIT_ADDR)
2045 			desc = ring->desc64;
2046 		else
2047 			desc = ring->desc32;
2048 
2049 		bus_dmamap_unload(ring->tag, ring->map);
2050 		bus_dmamem_free(ring->tag, desc, ring->map);
2051 		bus_dma_tag_destroy(ring->tag);
2052 	}
2053 }
2054 
2055 static int
2056 nfe_ifmedia_upd(struct ifnet *ifp)
2057 {
2058 	struct nfe_softc *sc = ifp->if_softc;
2059 	struct mii_data *mii = device_get_softc(sc->sc_miibus);
2060 
2061 	ASSERT_SERIALIZED(ifp->if_serializer);
2062 
2063 	if (mii->mii_instance != 0) {
2064 		struct mii_softc *miisc;
2065 
2066 		LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
2067 			mii_phy_reset(miisc);
2068 	}
2069 	mii_mediachg(mii);
2070 
2071 	return 0;
2072 }
2073 
2074 static void
2075 nfe_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2076 {
2077 	struct nfe_softc *sc = ifp->if_softc;
2078 	struct mii_data *mii = device_get_softc(sc->sc_miibus);
2079 
2080 	ASSERT_SERIALIZED(ifp->if_serializer);
2081 
2082 	mii_pollstat(mii);
2083 	ifmr->ifm_status = mii->mii_media_status;
2084 	ifmr->ifm_active = mii->mii_media_active;
2085 }
2086 
2087 static void
2088 nfe_setmulti(struct nfe_softc *sc)
2089 {
2090 	struct ifnet *ifp = &sc->arpcom.ac_if;
2091 	struct ifmultiaddr *ifma;
2092 	uint8_t addr[ETHER_ADDR_LEN], mask[ETHER_ADDR_LEN];
2093 	uint32_t filter = NFE_RXFILTER_MAGIC;
2094 	int i;
2095 
2096 	if ((ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) != 0) {
2097 		bzero(addr, ETHER_ADDR_LEN);
2098 		bzero(mask, ETHER_ADDR_LEN);
2099 		goto done;
2100 	}
2101 
2102 	bcopy(etherbroadcastaddr, addr, ETHER_ADDR_LEN);
2103 	bcopy(etherbroadcastaddr, mask, ETHER_ADDR_LEN);
2104 
2105 	LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2106 		caddr_t maddr;
2107 
2108 		if (ifma->ifma_addr->sa_family != AF_LINK)
2109 			continue;
2110 
2111 		maddr = LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
2112 		for (i = 0; i < ETHER_ADDR_LEN; i++) {
2113 			addr[i] &= maddr[i];
2114 			mask[i] &= ~maddr[i];
2115 		}
2116 	}
2117 
2118 	for (i = 0; i < ETHER_ADDR_LEN; i++)
2119 		mask[i] |= addr[i];
2120 
2121 done:
2122 	addr[0] |= 0x01;	/* make sure multicast bit is set */
2123 
2124 	NFE_WRITE(sc, NFE_MULTIADDR_HI,
2125 	    addr[3] << 24 | addr[2] << 16 | addr[1] << 8 | addr[0]);
2126 	NFE_WRITE(sc, NFE_MULTIADDR_LO,
2127 	    addr[5] <<  8 | addr[4]);
2128 	NFE_WRITE(sc, NFE_MULTIMASK_HI,
2129 	    mask[3] << 24 | mask[2] << 16 | mask[1] << 8 | mask[0]);
2130 	NFE_WRITE(sc, NFE_MULTIMASK_LO,
2131 	    mask[5] <<  8 | mask[4]);
2132 
2133 	filter |= (ifp->if_flags & IFF_PROMISC) ? NFE_PROMISC : NFE_U2M;
2134 	NFE_WRITE(sc, NFE_RXFILTER, filter);
2135 }
2136 
2137 static void
2138 nfe_get_macaddr(struct nfe_softc *sc, uint8_t *addr)
2139 {
2140 	uint32_t lo, hi;
2141 
2142 	lo = NFE_READ(sc, NFE_MACADDR_LO);
2143 	hi = NFE_READ(sc, NFE_MACADDR_HI);
2144 	if (sc->sc_caps & NFE_FIX_EADDR) {
2145 		addr[0] = (lo >> 8) & 0xff;
2146 		addr[1] = (lo & 0xff);
2147 
2148 		addr[2] = (hi >> 24) & 0xff;
2149 		addr[3] = (hi >> 16) & 0xff;
2150 		addr[4] = (hi >>  8) & 0xff;
2151 		addr[5] = (hi & 0xff);
2152 	} else {
2153 		addr[0] = (hi & 0xff);
2154 		addr[1] = (hi >>  8) & 0xff;
2155 		addr[2] = (hi >> 16) & 0xff;
2156 		addr[3] = (hi >> 24) & 0xff;
2157 
2158 		addr[4] = (lo & 0xff);
2159 		addr[5] = (lo >>  8) & 0xff;
2160 	}
2161 }
2162 
2163 static void
2164 nfe_set_macaddr(struct nfe_softc *sc, const uint8_t *addr)
2165 {
2166 	NFE_WRITE(sc, NFE_MACADDR_LO,
2167 	    addr[5] <<  8 | addr[4]);
2168 	NFE_WRITE(sc, NFE_MACADDR_HI,
2169 	    addr[3] << 24 | addr[2] << 16 | addr[1] << 8 | addr[0]);
2170 }
2171 
2172 static void
2173 nfe_tick(void *arg)
2174 {
2175 	struct nfe_softc *sc = arg;
2176 	struct ifnet *ifp = &sc->arpcom.ac_if;
2177 	struct mii_data *mii = device_get_softc(sc->sc_miibus);
2178 
2179 	lwkt_serialize_enter(ifp->if_serializer);
2180 
2181 	mii_tick(mii);
2182 	callout_reset(&sc->sc_tick_ch, hz, nfe_tick, sc);
2183 
2184 	lwkt_serialize_exit(ifp->if_serializer);
2185 }
2186 
2187 static int
2188 nfe_newbuf_std(struct nfe_softc *sc, struct nfe_rx_ring *ring, int idx,
2189 	       int wait)
2190 {
2191 	struct nfe_rx_data *data = &ring->data[idx];
2192 	bus_dma_segment_t seg;
2193 	bus_dmamap_t map;
2194 	struct mbuf *m;
2195 	int nsegs, error;
2196 
2197 	m = m_getcl(wait ? MB_WAIT : MB_DONTWAIT, MT_DATA, M_PKTHDR);
2198 	if (m == NULL)
2199 		return ENOBUFS;
2200 	m->m_len = m->m_pkthdr.len = MCLBYTES;
2201 
2202 	error = bus_dmamap_load_mbuf_segment(ring->data_tag, ring->data_tmpmap,
2203 			m, &seg, 1, &nsegs, BUS_DMA_NOWAIT);
2204 	if (error) {
2205 		m_freem(m);
2206 		if (wait) {
2207 			if_printf(&sc->arpcom.ac_if,
2208 				  "could map RX mbuf %d\n", error);
2209 		}
2210 		return error;
2211 	}
2212 
2213 	if (data->m != NULL) {
2214 		/* Sync and unload originally mapped mbuf */
2215 		bus_dmamap_sync(ring->data_tag, data->map,
2216 				BUS_DMASYNC_POSTREAD);
2217 		bus_dmamap_unload(ring->data_tag, data->map);
2218 	}
2219 
2220 	/* Swap this DMA map with tmp DMA map */
2221 	map = data->map;
2222 	data->map = ring->data_tmpmap;
2223 	ring->data_tmpmap = map;
2224 
2225 	/* Caller is assumed to have collected the old mbuf */
2226 	data->m = m;
2227 
2228 	nfe_set_paddr_rxdesc(sc, ring, idx, seg.ds_addr);
2229 	return 0;
2230 }
2231 
2232 static int
2233 nfe_newbuf_jumbo(struct nfe_softc *sc, struct nfe_rx_ring *ring, int idx,
2234 		 int wait)
2235 {
2236 	struct nfe_rx_data *data = &ring->data[idx];
2237 	struct nfe_jbuf *jbuf;
2238 	struct mbuf *m;
2239 
2240 	MGETHDR(m, wait ? MB_WAIT : MB_DONTWAIT, MT_DATA);
2241 	if (m == NULL)
2242 		return ENOBUFS;
2243 
2244 	jbuf = nfe_jalloc(sc);
2245 	if (jbuf == NULL) {
2246 		m_freem(m);
2247 		if_printf(&sc->arpcom.ac_if, "jumbo allocation failed "
2248 		    "-- packet dropped!\n");
2249 		return ENOBUFS;
2250 	}
2251 
2252 	m->m_ext.ext_arg = jbuf;
2253 	m->m_ext.ext_buf = jbuf->buf;
2254 	m->m_ext.ext_free = nfe_jfree;
2255 	m->m_ext.ext_ref = nfe_jref;
2256 	m->m_ext.ext_size = NFE_JBYTES;
2257 
2258 	m->m_data = m->m_ext.ext_buf;
2259 	m->m_flags |= M_EXT;
2260 	m->m_len = m->m_pkthdr.len = m->m_ext.ext_size;
2261 
2262 	/* Caller is assumed to have collected the old mbuf */
2263 	data->m = m;
2264 
2265 	nfe_set_paddr_rxdesc(sc, ring, idx, jbuf->physaddr);
2266 	return 0;
2267 }
2268 
2269 static void
2270 nfe_set_paddr_rxdesc(struct nfe_softc *sc, struct nfe_rx_ring *ring, int idx,
2271 		     bus_addr_t physaddr)
2272 {
2273 	if (sc->sc_caps & NFE_40BIT_ADDR) {
2274 		struct nfe_desc64 *desc64 = &ring->desc64[idx];
2275 
2276 		desc64->physaddr[0] = htole32(NFE_ADDR_HI(physaddr));
2277 		desc64->physaddr[1] = htole32(NFE_ADDR_LO(physaddr));
2278 	} else {
2279 		struct nfe_desc32 *desc32 = &ring->desc32[idx];
2280 
2281 		desc32->physaddr = htole32(physaddr);
2282 	}
2283 }
2284 
2285 static void
2286 nfe_set_ready_rxdesc(struct nfe_softc *sc, struct nfe_rx_ring *ring, int idx)
2287 {
2288 	if (sc->sc_caps & NFE_40BIT_ADDR) {
2289 		struct nfe_desc64 *desc64 = &ring->desc64[idx];
2290 
2291 		desc64->length = htole16(ring->bufsz);
2292 		desc64->flags = htole16(NFE_RX_READY);
2293 	} else {
2294 		struct nfe_desc32 *desc32 = &ring->desc32[idx];
2295 
2296 		desc32->length = htole16(ring->bufsz);
2297 		desc32->flags = htole16(NFE_RX_READY);
2298 	}
2299 }
2300 
2301 static int
2302 nfe_sysctl_imtime(SYSCTL_HANDLER_ARGS)
2303 {
2304 	struct nfe_softc *sc = arg1;
2305 	struct ifnet *ifp = &sc->arpcom.ac_if;
2306 	uint32_t flags;
2307 	int error, v;
2308 
2309 	lwkt_serialize_enter(ifp->if_serializer);
2310 
2311 	flags = sc->sc_flags & ~NFE_F_DYN_IM;
2312 	v = sc->sc_imtime;
2313 	if (sc->sc_flags & NFE_F_DYN_IM)
2314 		v = -v;
2315 
2316 	error = sysctl_handle_int(oidp, &v, 0, req);
2317 	if (error || req->newptr == NULL)
2318 		goto back;
2319 
2320 	if (v < 0) {
2321 		flags |= NFE_F_DYN_IM;
2322 		v = -v;
2323 	}
2324 
2325 	if (v != sc->sc_imtime || (flags ^ sc->sc_flags)) {
2326 		if (NFE_IMTIME(v) == 0)
2327 			v = 0;
2328 		sc->sc_imtime = v;
2329 		sc->sc_flags = flags;
2330 		sc->sc_irq_enable = NFE_IRQ_ENABLE(sc);
2331 
2332 		if ((ifp->if_flags & (IFF_POLLING | IFF_RUNNING))
2333 		    == IFF_RUNNING) {
2334 			nfe_enable_intrs(sc);
2335 		}
2336 	}
2337 back:
2338 	lwkt_serialize_exit(ifp->if_serializer);
2339 	return error;
2340 }
2341 
2342 static void
2343 nfe_powerup(device_t dev)
2344 {
2345 	struct nfe_softc *sc = device_get_softc(dev);
2346 	uint32_t pwr_state;
2347 	uint16_t did;
2348 
2349 	/*
2350 	 * Bring MAC and PHY out of low power state
2351 	 */
2352 
2353 	pwr_state = NFE_READ(sc, NFE_PWR_STATE2) & ~NFE_PWRUP_MASK;
2354 
2355 	did = pci_get_device(dev);
2356 	if ((did == PCI_PRODUCT_NVIDIA_MCP51_LAN1 ||
2357 	     did == PCI_PRODUCT_NVIDIA_MCP51_LAN2) &&
2358 	    pci_get_revid(dev) >= 0xa3)
2359 		pwr_state |= NFE_PWRUP_REV_A3;
2360 
2361 	NFE_WRITE(sc, NFE_PWR_STATE2, pwr_state);
2362 }
2363 
2364 static void
2365 nfe_mac_reset(struct nfe_softc *sc)
2366 {
2367 	uint32_t rxtxctl = sc->rxtxctl_desc | NFE_RXTX_BIT2;
2368 	uint32_t macaddr_hi, macaddr_lo, tx_poll;
2369 
2370 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_RESET | rxtxctl);
2371 
2372 	/* Save several registers for later restoration */
2373 	macaddr_hi = NFE_READ(sc, NFE_MACADDR_HI);
2374 	macaddr_lo = NFE_READ(sc, NFE_MACADDR_LO);
2375 	tx_poll = NFE_READ(sc, NFE_TX_POLL);
2376 
2377 	NFE_WRITE(sc, NFE_MAC_RESET, NFE_RESET_ASSERT);
2378 	DELAY(100);
2379 
2380 	NFE_WRITE(sc, NFE_MAC_RESET, 0);
2381 	DELAY(100);
2382 
2383 	/* Restore saved registers */
2384 	NFE_WRITE(sc, NFE_MACADDR_HI, macaddr_hi);
2385 	NFE_WRITE(sc, NFE_MACADDR_LO, macaddr_lo);
2386 	NFE_WRITE(sc, NFE_TX_POLL, tx_poll);
2387 
2388 	NFE_WRITE(sc, NFE_RXTX_CTL, rxtxctl);
2389 }
2390 
2391 static void
2392 nfe_enable_intrs(struct nfe_softc *sc)
2393 {
2394 	/*
2395 	 * NFE_IMTIMER generates a periodic interrupt via NFE_IRQ_TIMER.
2396 	 * It is unclear how wide the timer is.  Base programming does
2397 	 * not seem to effect NFE_IRQ_TX_DONE or NFE_IRQ_RX_DONE so
2398 	 * we don't get any interrupt moderation.  TX moderation is
2399 	 * possible by using the timer interrupt instead of TX_DONE.
2400 	 *
2401 	 * It is unclear whether there are other bits that can be
2402 	 * set to make the NFE device actually do interrupt moderation
2403 	 * on the RX side.
2404 	 *
2405 	 * For now set a 128uS interval as a placemark, but don't use
2406 	 * the timer.
2407 	 */
2408 	if (sc->sc_imtime == 0)
2409 		NFE_WRITE(sc, NFE_IMTIMER, NFE_IMTIME_DEFAULT);
2410 	else
2411 		NFE_WRITE(sc, NFE_IMTIMER, NFE_IMTIME(sc->sc_imtime));
2412 
2413 	/* Enable interrupts */
2414 	NFE_WRITE(sc, NFE_IRQ_MASK, sc->sc_irq_enable);
2415 
2416 	if (sc->sc_irq_enable & NFE_IRQ_TIMER)
2417 		sc->sc_flags |= NFE_F_IRQ_TIMER;
2418 	else
2419 		sc->sc_flags &= ~NFE_F_IRQ_TIMER;
2420 }
2421 
2422 static void
2423 nfe_disable_intrs(struct nfe_softc *sc)
2424 {
2425 	/* Disable interrupts */
2426 	NFE_WRITE(sc, NFE_IRQ_MASK, 0);
2427 	sc->sc_flags &= ~NFE_F_IRQ_TIMER;
2428 }
2429