xref: /dragonfly/sys/dev/netif/nfe/if_nfe.c (revision a563ca70)
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_polling.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/ifq_var.h>
78 #include <net/if_types.h>
79 #include <net/if_var.h>
80 #include <net/vlan/if_vlan_var.h>
81 #include <net/vlan/if_vlan_ether.h>
82 
83 #include <bus/pci/pcireg.h>
84 #include <bus/pci/pcivar.h>
85 #include <bus/pci/pcidevs.h>
86 
87 #include <dev/netif/mii_layer/mii.h>
88 #include <dev/netif/mii_layer/miivar.h>
89 
90 #include "miibus_if.h"
91 
92 #include <dev/netif/nfe/if_nfereg.h>
93 #include <dev/netif/nfe/if_nfevar.h>
94 
95 #define NFE_CSUM
96 #define NFE_CSUM_FEATURES	(CSUM_IP | CSUM_TCP | CSUM_UDP)
97 
98 static int	nfe_probe(device_t);
99 static int	nfe_attach(device_t);
100 static int	nfe_detach(device_t);
101 static void	nfe_shutdown(device_t);
102 static int	nfe_resume(device_t);
103 static int	nfe_suspend(device_t);
104 
105 static int	nfe_miibus_readreg(device_t, int, int);
106 static void	nfe_miibus_writereg(device_t, int, int, int);
107 static void	nfe_miibus_statchg(device_t);
108 
109 #ifdef DEVICE_POLLING
110 static void	nfe_poll(struct ifnet *, enum poll_cmd, int);
111 static void	nfe_disable_intrs(struct nfe_softc *);
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 
154 static int	nfe_sysctl_imtime(SYSCTL_HANDLER_ARGS);
155 
156 #define NFE_DEBUG
157 #ifdef NFE_DEBUG
158 
159 static int	nfe_debug = 0;
160 static int	nfe_rx_ring_count = NFE_RX_RING_DEF_COUNT;
161 static int	nfe_tx_ring_count = NFE_TX_RING_DEF_COUNT;
162 /*
163  * hw timer simulated interrupt moderation @4000Hz.  Negative values
164  * disable the timer when the discrete interrupt rate falls below
165  * the moderation rate.
166  *
167  * XXX 8000Hz might be better but if the interrupt is shared it can
168  *     blow out the cpu.
169  */
170 static int	nfe_imtime = -250;	/* uS */
171 
172 TUNABLE_INT("hw.nfe.rx_ring_count", &nfe_rx_ring_count);
173 TUNABLE_INT("hw.nfe.tx_ring_count", &nfe_tx_ring_count);
174 TUNABLE_INT("hw.nfe.imtimer", &nfe_imtime);
175 TUNABLE_INT("hw.nfe.debug", &nfe_debug);
176 
177 #define DPRINTF(sc, fmt, ...) do {		\
178 	if ((sc)->sc_debug) {			\
179 		if_printf(&(sc)->arpcom.ac_if,	\
180 			  fmt, __VA_ARGS__);	\
181 	}					\
182 } while (0)
183 
184 #define DPRINTFN(sc, lv, fmt, ...) do {		\
185 	if ((sc)->sc_debug >= (lv)) {		\
186 		if_printf(&(sc)->arpcom.ac_if,	\
187 			  fmt, __VA_ARGS__);	\
188 	}					\
189 } while (0)
190 
191 #else	/* !NFE_DEBUG */
192 
193 #define DPRINTF(sc, fmt, ...)
194 #define DPRINTFN(sc, lv, fmt, ...)
195 
196 #endif	/* NFE_DEBUG */
197 
198 static const struct nfe_dev {
199 	uint16_t	vid;
200 	uint16_t	did;
201 	const char	*desc;
202 } nfe_devices[] = {
203 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE_LAN,
204 	  "NVIDIA nForce Fast Ethernet" },
205 
206 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE2_LAN,
207 	  "NVIDIA nForce2 Fast Ethernet" },
208 
209 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN1,
210 	  "NVIDIA nForce3 Gigabit Ethernet" },
211 
212 	/* XXX TGEN the next chip can also be found in the nForce2 Ultra 400Gb
213 	   chipset, and possibly also the 400R; it might be both nForce2- and
214 	   nForce3-based boards can use the same MCPs (= southbridges) */
215 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN2,
216 	  "NVIDIA nForce3 Gigabit Ethernet" },
217 
218 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN3,
219 	  "NVIDIA nForce3 Gigabit Ethernet" },
220 
221 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN4,
222 	  "NVIDIA nForce3 Gigabit Ethernet" },
223 
224 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN5,
225 	  "NVIDIA nForce3 Gigabit Ethernet" },
226 
227 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_CK804_LAN1,
228 	  "NVIDIA CK804 Gigabit Ethernet" },
229 
230 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_CK804_LAN2,
231 	  "NVIDIA CK804 Gigabit Ethernet" },
232 
233 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP04_LAN1,
234 	  "NVIDIA MCP04 Gigabit Ethernet" },
235 
236 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP04_LAN2,
237 	  "NVIDIA MCP04 Gigabit Ethernet" },
238 
239 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP51_LAN1,
240 	  "NVIDIA MCP51 Gigabit Ethernet" },
241 
242 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP51_LAN2,
243 	  "NVIDIA MCP51 Gigabit Ethernet" },
244 
245 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP55_LAN1,
246 	  "NVIDIA MCP55 Gigabit Ethernet" },
247 
248 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP55_LAN2,
249 	  "NVIDIA MCP55 Gigabit Ethernet" },
250 
251 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN1,
252 	  "NVIDIA MCP61 Gigabit Ethernet" },
253 
254 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN2,
255 	  "NVIDIA MCP61 Gigabit Ethernet" },
256 
257 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN3,
258 	  "NVIDIA MCP61 Gigabit Ethernet" },
259 
260 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN4,
261 	  "NVIDIA MCP61 Gigabit Ethernet" },
262 
263 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN1,
264 	  "NVIDIA MCP65 Gigabit Ethernet" },
265 
266 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN2,
267 	  "NVIDIA MCP65 Gigabit Ethernet" },
268 
269 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN3,
270 	  "NVIDIA MCP65 Gigabit Ethernet" },
271 
272 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN4,
273 	  "NVIDIA MCP65 Gigabit Ethernet" },
274 
275 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN1,
276 	  "NVIDIA MCP67 Gigabit Ethernet" },
277 
278 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN2,
279 	  "NVIDIA MCP67 Gigabit Ethernet" },
280 
281 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN3,
282 	  "NVIDIA MCP67 Gigabit Ethernet" },
283 
284 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN4,
285 	  "NVIDIA MCP67 Gigabit Ethernet" },
286 
287 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP73_LAN1,
288 	  "NVIDIA MCP73 Gigabit Ethernet" },
289 
290 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP73_LAN2,
291 	  "NVIDIA MCP73 Gigabit Ethernet" },
292 
293 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP73_LAN3,
294 	  "NVIDIA MCP73 Gigabit Ethernet" },
295 
296 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP73_LAN4,
297 	  "NVIDIA MCP73 Gigabit Ethernet" },
298 
299 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP77_LAN1,
300 	  "NVIDIA MCP77 Gigabit Ethernet" },
301 
302 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP77_LAN2,
303 	  "NVIDIA MCP77 Gigabit Ethernet" },
304 
305 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP77_LAN3,
306 	  "NVIDIA MCP77 Gigabit Ethernet" },
307 
308 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP77_LAN4,
309 	  "NVIDIA MCP77 Gigabit Ethernet" },
310 
311 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP79_LAN1,
312 	  "NVIDIA MCP79 Gigabit Ethernet" },
313 
314 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP79_LAN2,
315 	  "NVIDIA MCP79 Gigabit Ethernet" },
316 
317 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP79_LAN3,
318 	  "NVIDIA MCP79 Gigabit Ethernet" },
319 
320 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP79_LAN4,
321 	  "NVIDIA MCP79 Gigabit Ethernet" },
322 
323 	{ 0, 0, NULL }
324 };
325 
326 static device_method_t nfe_methods[] = {
327 	/* Device interface */
328 	DEVMETHOD(device_probe,		nfe_probe),
329 	DEVMETHOD(device_attach,	nfe_attach),
330 	DEVMETHOD(device_detach,	nfe_detach),
331 	DEVMETHOD(device_suspend,	nfe_suspend),
332 	DEVMETHOD(device_resume,	nfe_resume),
333 	DEVMETHOD(device_shutdown,	nfe_shutdown),
334 
335 	/* Bus interface */
336 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
337 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
338 
339 	/* MII interface */
340 	DEVMETHOD(miibus_readreg,	nfe_miibus_readreg),
341 	DEVMETHOD(miibus_writereg,	nfe_miibus_writereg),
342 	DEVMETHOD(miibus_statchg,	nfe_miibus_statchg),
343 
344 	{ 0, 0 }
345 };
346 
347 static driver_t nfe_driver = {
348 	"nfe",
349 	nfe_methods,
350 	sizeof(struct nfe_softc)
351 };
352 
353 static devclass_t	nfe_devclass;
354 
355 DECLARE_DUMMY_MODULE(if_nfe);
356 MODULE_DEPEND(if_nfe, miibus, 1, 1, 1);
357 DRIVER_MODULE(if_nfe, pci, nfe_driver, nfe_devclass, NULL, NULL);
358 DRIVER_MODULE(miibus, nfe, miibus_driver, miibus_devclass, NULL, NULL);
359 
360 /*
361  * NOTE: NFE_WORDALIGN support is guesswork right now.
362  */
363 static int
364 nfe_probe(device_t dev)
365 {
366 	const struct nfe_dev *n;
367 	uint16_t vid, did;
368 
369 	vid = pci_get_vendor(dev);
370 	did = pci_get_device(dev);
371 	for (n = nfe_devices; n->desc != NULL; ++n) {
372 		if (vid == n->vid && did == n->did) {
373 			struct nfe_softc *sc = device_get_softc(dev);
374 
375 			switch (did) {
376 			case PCI_PRODUCT_NVIDIA_NFORCE_LAN:
377 			case PCI_PRODUCT_NVIDIA_NFORCE2_LAN:
378 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN1:
379 				sc->sc_caps = NFE_NO_PWRCTL |
380 					      NFE_FIX_EADDR;
381 				break;
382 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN2:
383 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN3:
384 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN4:
385 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN5:
386 				sc->sc_caps = NFE_JUMBO_SUP |
387 					      NFE_HW_CSUM |
388 					      NFE_NO_PWRCTL |
389 					      NFE_FIX_EADDR;
390 				break;
391 			case PCI_PRODUCT_NVIDIA_MCP51_LAN1:
392 			case PCI_PRODUCT_NVIDIA_MCP51_LAN2:
393 				sc->sc_caps = NFE_FIX_EADDR;
394 				/* FALL THROUGH */
395 			case PCI_PRODUCT_NVIDIA_MCP61_LAN1:
396 			case PCI_PRODUCT_NVIDIA_MCP61_LAN2:
397 			case PCI_PRODUCT_NVIDIA_MCP61_LAN3:
398 			case PCI_PRODUCT_NVIDIA_MCP61_LAN4:
399 			case PCI_PRODUCT_NVIDIA_MCP67_LAN1:
400 			case PCI_PRODUCT_NVIDIA_MCP67_LAN2:
401 			case PCI_PRODUCT_NVIDIA_MCP67_LAN3:
402 			case PCI_PRODUCT_NVIDIA_MCP67_LAN4:
403 			case PCI_PRODUCT_NVIDIA_MCP73_LAN1:
404 			case PCI_PRODUCT_NVIDIA_MCP73_LAN2:
405 			case PCI_PRODUCT_NVIDIA_MCP73_LAN3:
406 			case PCI_PRODUCT_NVIDIA_MCP73_LAN4:
407 				sc->sc_caps |= NFE_40BIT_ADDR;
408 				break;
409 			case PCI_PRODUCT_NVIDIA_CK804_LAN1:
410 			case PCI_PRODUCT_NVIDIA_CK804_LAN2:
411 			case PCI_PRODUCT_NVIDIA_MCP04_LAN1:
412 			case PCI_PRODUCT_NVIDIA_MCP04_LAN2:
413 				sc->sc_caps = NFE_JUMBO_SUP |
414 					      NFE_40BIT_ADDR |
415 					      NFE_HW_CSUM |
416 					      NFE_NO_PWRCTL |
417 					      NFE_FIX_EADDR;
418 				break;
419 			case PCI_PRODUCT_NVIDIA_MCP65_LAN1:
420 			case PCI_PRODUCT_NVIDIA_MCP65_LAN2:
421 			case PCI_PRODUCT_NVIDIA_MCP65_LAN3:
422 			case PCI_PRODUCT_NVIDIA_MCP65_LAN4:
423 				sc->sc_caps = NFE_JUMBO_SUP |
424 					      NFE_40BIT_ADDR;
425 				break;
426 			case PCI_PRODUCT_NVIDIA_MCP55_LAN1:
427 			case PCI_PRODUCT_NVIDIA_MCP55_LAN2:
428 				sc->sc_caps = NFE_JUMBO_SUP |
429 					      NFE_40BIT_ADDR |
430 					      NFE_HW_CSUM |
431 					      NFE_HW_VLAN |
432 					      NFE_FIX_EADDR;
433 				break;
434 			case PCI_PRODUCT_NVIDIA_MCP77_LAN1:
435 			case PCI_PRODUCT_NVIDIA_MCP77_LAN2:
436 			case PCI_PRODUCT_NVIDIA_MCP77_LAN3:
437 			case PCI_PRODUCT_NVIDIA_MCP77_LAN4:
438 			case PCI_PRODUCT_NVIDIA_MCP79_LAN1:
439 			case PCI_PRODUCT_NVIDIA_MCP79_LAN2:
440 			case PCI_PRODUCT_NVIDIA_MCP79_LAN3:
441 			case PCI_PRODUCT_NVIDIA_MCP79_LAN4:
442 				sc->sc_caps = NFE_40BIT_ADDR |
443 					      NFE_HW_CSUM |
444 					      NFE_WORDALIGN;
445 				break;
446 			}
447 
448 			device_set_desc(dev, n->desc);
449 			device_set_async_attach(dev, TRUE);
450 			return 0;
451 		}
452 	}
453 	return ENXIO;
454 }
455 
456 static int
457 nfe_attach(device_t dev)
458 {
459 	struct nfe_softc *sc = device_get_softc(dev);
460 	struct ifnet *ifp = &sc->arpcom.ac_if;
461 	uint8_t eaddr[ETHER_ADDR_LEN];
462 	bus_addr_t lowaddr;
463 	int error;
464 
465 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
466 	lwkt_serialize_init(&sc->sc_jbuf_serializer);
467 
468 	/*
469 	 * Initialize sysctl variables
470 	 */
471 	sc->sc_rx_ring_count = nfe_rx_ring_count;
472 	sc->sc_tx_ring_count = nfe_tx_ring_count;
473 	sc->sc_debug = nfe_debug;
474 	if (nfe_imtime < 0) {
475 		sc->sc_flags |= NFE_F_DYN_IM;
476 		sc->sc_imtime = -nfe_imtime;
477 	} else {
478 		sc->sc_imtime = nfe_imtime;
479 	}
480 	sc->sc_irq_enable = NFE_IRQ_ENABLE(sc);
481 
482 	sc->sc_mem_rid = PCIR_BAR(0);
483 
484 	if (sc->sc_caps & NFE_40BIT_ADDR)
485 		sc->rxtxctl_desc = NFE_RXTX_DESC_V3;
486 	else if (sc->sc_caps & NFE_JUMBO_SUP)
487 		sc->rxtxctl_desc = NFE_RXTX_DESC_V2;
488 
489 #ifndef BURN_BRIDGES
490 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
491 		uint32_t mem, irq;
492 
493 		mem = pci_read_config(dev, sc->sc_mem_rid, 4);
494 		irq = pci_read_config(dev, PCIR_INTLINE, 4);
495 
496 		device_printf(dev, "chip is in D%d power mode "
497 		    "-- setting to D0\n", pci_get_powerstate(dev));
498 
499 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
500 
501 		pci_write_config(dev, sc->sc_mem_rid, mem, 4);
502 		pci_write_config(dev, PCIR_INTLINE, irq, 4);
503 	}
504 #endif	/* !BURN_BRIDGE */
505 
506 	/* Enable bus mastering */
507 	pci_enable_busmaster(dev);
508 
509 	/* Allocate IO memory */
510 	sc->sc_mem_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
511 						&sc->sc_mem_rid, RF_ACTIVE);
512 	if (sc->sc_mem_res == NULL) {
513 		device_printf(dev, "could not allocate io memory\n");
514 		return ENXIO;
515 	}
516 	sc->sc_memh = rman_get_bushandle(sc->sc_mem_res);
517 	sc->sc_memt = rman_get_bustag(sc->sc_mem_res);
518 
519 	/* Allocate IRQ */
520 	sc->sc_irq_rid = 0;
521 	sc->sc_irq_res = bus_alloc_resource_any(dev, SYS_RES_IRQ,
522 						&sc->sc_irq_rid,
523 						RF_SHAREABLE | RF_ACTIVE);
524 	if (sc->sc_irq_res == NULL) {
525 		device_printf(dev, "could not allocate irq\n");
526 		error = ENXIO;
527 		goto fail;
528 	}
529 
530 	/* Disable WOL */
531 	NFE_WRITE(sc, NFE_WOL_CTL, 0);
532 
533 	if ((sc->sc_caps & NFE_NO_PWRCTL) == 0)
534 		nfe_powerup(dev);
535 
536 	nfe_get_macaddr(sc, eaddr);
537 
538 	/*
539 	 * Allocate top level DMA tag
540 	 */
541 	if (sc->sc_caps & NFE_40BIT_ADDR)
542 		lowaddr = NFE_BUS_SPACE_MAXADDR;
543 	else
544 		lowaddr = BUS_SPACE_MAXADDR_32BIT;
545 	error = bus_dma_tag_create(NULL,	/* parent */
546 			1, 0,			/* alignment, boundary */
547 			lowaddr,		/* lowaddr */
548 			BUS_SPACE_MAXADDR,	/* highaddr */
549 			NULL, NULL,		/* filter, filterarg */
550 			BUS_SPACE_MAXSIZE_32BIT,/* maxsize */
551 			0,			/* nsegments */
552 			BUS_SPACE_MAXSIZE_32BIT,/* maxsegsize */
553 			0,			/* flags */
554 			&sc->sc_dtag);
555 	if (error) {
556 		device_printf(dev, "could not allocate parent dma tag\n");
557 		goto fail;
558 	}
559 
560 	/*
561 	 * Allocate Tx and Rx rings.
562 	 */
563 	error = nfe_alloc_tx_ring(sc, &sc->txq);
564 	if (error) {
565 		device_printf(dev, "could not allocate Tx ring\n");
566 		goto fail;
567 	}
568 
569 	error = nfe_alloc_rx_ring(sc, &sc->rxq);
570 	if (error) {
571 		device_printf(dev, "could not allocate Rx ring\n");
572 		goto fail;
573 	}
574 
575 	/*
576 	 * Create sysctl tree
577 	 */
578 	sysctl_ctx_init(&sc->sc_sysctl_ctx);
579 	sc->sc_sysctl_tree = SYSCTL_ADD_NODE(&sc->sc_sysctl_ctx,
580 					     SYSCTL_STATIC_CHILDREN(_hw),
581 					     OID_AUTO,
582 					     device_get_nameunit(dev),
583 					     CTLFLAG_RD, 0, "");
584 	if (sc->sc_sysctl_tree == NULL) {
585 		device_printf(dev, "can't add sysctl node\n");
586 		error = ENXIO;
587 		goto fail;
588 	}
589 	SYSCTL_ADD_PROC(&sc->sc_sysctl_ctx,
590 			SYSCTL_CHILDREN(sc->sc_sysctl_tree),
591 			OID_AUTO, "imtimer", CTLTYPE_INT | CTLFLAG_RW,
592 			sc, 0, nfe_sysctl_imtime, "I",
593 			"Interrupt moderation time (usec).  "
594 			"0 to disable interrupt moderation.");
595 	SYSCTL_ADD_INT(&sc->sc_sysctl_ctx,
596 		       SYSCTL_CHILDREN(sc->sc_sysctl_tree), OID_AUTO,
597 		       "rx_ring_count", CTLFLAG_RD, &sc->sc_rx_ring_count,
598 		       0, "RX ring count");
599 	SYSCTL_ADD_INT(&sc->sc_sysctl_ctx,
600 		       SYSCTL_CHILDREN(sc->sc_sysctl_tree), OID_AUTO,
601 		       "tx_ring_count", CTLFLAG_RD, &sc->sc_tx_ring_count,
602 		       0, "TX ring count");
603 	SYSCTL_ADD_INT(&sc->sc_sysctl_ctx,
604 		       SYSCTL_CHILDREN(sc->sc_sysctl_tree), OID_AUTO,
605 		       "debug", CTLFLAG_RW, &sc->sc_debug,
606 		       0, "control debugging printfs");
607 
608 	error = mii_phy_probe(dev, &sc->sc_miibus, nfe_ifmedia_upd,
609 			      nfe_ifmedia_sts);
610 	if (error) {
611 		device_printf(dev, "MII without any phy\n");
612 		goto fail;
613 	}
614 
615 	ifp->if_softc = sc;
616 	ifp->if_mtu = ETHERMTU;
617 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
618 	ifp->if_ioctl = nfe_ioctl;
619 	ifp->if_start = nfe_start;
620 #ifdef DEVICE_POLLING
621 	ifp->if_poll = nfe_poll;
622 #endif
623 	ifp->if_watchdog = nfe_watchdog;
624 	ifp->if_init = nfe_init;
625 	ifq_set_maxlen(&ifp->if_snd, sc->sc_tx_ring_count);
626 	ifq_set_ready(&ifp->if_snd);
627 
628 	ifp->if_capabilities = IFCAP_VLAN_MTU;
629 
630 	if (sc->sc_caps & NFE_HW_VLAN)
631 		ifp->if_capabilities |= IFCAP_VLAN_HWTAGGING;
632 
633 #ifdef NFE_CSUM
634 	if (sc->sc_caps & NFE_HW_CSUM) {
635 		ifp->if_capabilities |= IFCAP_HWCSUM;
636 		ifp->if_hwassist = NFE_CSUM_FEATURES;
637 	}
638 #else
639 	sc->sc_caps &= ~NFE_HW_CSUM;
640 #endif
641 	ifp->if_capenable = ifp->if_capabilities;
642 
643 	callout_init(&sc->sc_tick_ch);
644 
645 	ether_ifattach(ifp, eaddr, NULL);
646 
647 	error = bus_setup_intr(dev, sc->sc_irq_res, INTR_MPSAFE, nfe_intr, sc,
648 			       &sc->sc_ih, ifp->if_serializer);
649 	if (error) {
650 		device_printf(dev, "could not setup intr\n");
651 		ether_ifdetach(ifp);
652 		goto fail;
653 	}
654 
655 	ifp->if_cpuid = ithread_cpuid(rman_get_start(sc->sc_irq_res));
656 	KKASSERT(ifp->if_cpuid >= 0 && ifp->if_cpuid < ncpus);
657 
658 	return 0;
659 fail:
660 	nfe_detach(dev);
661 	return error;
662 }
663 
664 static int
665 nfe_detach(device_t dev)
666 {
667 	struct nfe_softc *sc = device_get_softc(dev);
668 
669 	if (device_is_attached(dev)) {
670 		struct ifnet *ifp = &sc->arpcom.ac_if;
671 
672 		lwkt_serialize_enter(ifp->if_serializer);
673 		nfe_stop(sc);
674 		bus_teardown_intr(dev, sc->sc_irq_res, sc->sc_ih);
675 		lwkt_serialize_exit(ifp->if_serializer);
676 
677 		ether_ifdetach(ifp);
678 	}
679 
680 	if (sc->sc_miibus != NULL)
681 		device_delete_child(dev, sc->sc_miibus);
682 	bus_generic_detach(dev);
683 
684 	if (sc->sc_sysctl_tree != NULL)
685 		sysctl_ctx_free(&sc->sc_sysctl_ctx);
686 
687 	if (sc->sc_irq_res != NULL) {
688 		bus_release_resource(dev, SYS_RES_IRQ, sc->sc_irq_rid,
689 				     sc->sc_irq_res);
690 	}
691 
692 	if (sc->sc_mem_res != NULL) {
693 		bus_release_resource(dev, SYS_RES_MEMORY, sc->sc_mem_rid,
694 				     sc->sc_mem_res);
695 	}
696 
697 	nfe_free_tx_ring(sc, &sc->txq);
698 	nfe_free_rx_ring(sc, &sc->rxq);
699 	if (sc->sc_dtag != NULL)
700 		bus_dma_tag_destroy(sc->sc_dtag);
701 
702 	return 0;
703 }
704 
705 static void
706 nfe_shutdown(device_t dev)
707 {
708 	struct nfe_softc *sc = device_get_softc(dev);
709 	struct ifnet *ifp = &sc->arpcom.ac_if;
710 
711 	lwkt_serialize_enter(ifp->if_serializer);
712 	nfe_stop(sc);
713 	lwkt_serialize_exit(ifp->if_serializer);
714 }
715 
716 static int
717 nfe_suspend(device_t dev)
718 {
719 	struct nfe_softc *sc = device_get_softc(dev);
720 	struct ifnet *ifp = &sc->arpcom.ac_if;
721 
722 	lwkt_serialize_enter(ifp->if_serializer);
723 	nfe_stop(sc);
724 	lwkt_serialize_exit(ifp->if_serializer);
725 
726 	return 0;
727 }
728 
729 static int
730 nfe_resume(device_t dev)
731 {
732 	struct nfe_softc *sc = device_get_softc(dev);
733 	struct ifnet *ifp = &sc->arpcom.ac_if;
734 
735 	lwkt_serialize_enter(ifp->if_serializer);
736 	if (ifp->if_flags & IFF_UP)
737 		nfe_init(sc);
738 	lwkt_serialize_exit(ifp->if_serializer);
739 
740 	return 0;
741 }
742 
743 static void
744 nfe_miibus_statchg(device_t dev)
745 {
746 	struct nfe_softc *sc = device_get_softc(dev);
747 	struct mii_data *mii = device_get_softc(sc->sc_miibus);
748 	uint32_t phy, seed, misc = NFE_MISC1_MAGIC, link = NFE_MEDIA_SET;
749 
750 	ASSERT_SERIALIZED(sc->arpcom.ac_if.if_serializer);
751 
752 	phy = NFE_READ(sc, NFE_PHY_IFACE);
753 	phy &= ~(NFE_PHY_HDX | NFE_PHY_100TX | NFE_PHY_1000T);
754 
755 	seed = NFE_READ(sc, NFE_RNDSEED);
756 	seed &= ~NFE_SEED_MASK;
757 
758 	if ((mii->mii_media_active & IFM_GMASK) == IFM_HDX) {
759 		phy  |= NFE_PHY_HDX;	/* half-duplex */
760 		misc |= NFE_MISC1_HDX;
761 	}
762 
763 	switch (IFM_SUBTYPE(mii->mii_media_active)) {
764 	case IFM_1000_T:	/* full-duplex only */
765 		link |= NFE_MEDIA_1000T;
766 		seed |= NFE_SEED_1000T;
767 		phy  |= NFE_PHY_1000T;
768 		break;
769 	case IFM_100_TX:
770 		link |= NFE_MEDIA_100TX;
771 		seed |= NFE_SEED_100TX;
772 		phy  |= NFE_PHY_100TX;
773 		break;
774 	case IFM_10_T:
775 		link |= NFE_MEDIA_10T;
776 		seed |= NFE_SEED_10T;
777 		break;
778 	}
779 
780 	NFE_WRITE(sc, NFE_RNDSEED, seed);	/* XXX: gigabit NICs only? */
781 
782 	NFE_WRITE(sc, NFE_PHY_IFACE, phy);
783 	NFE_WRITE(sc, NFE_MISC1, misc);
784 	NFE_WRITE(sc, NFE_LINKSPEED, link);
785 }
786 
787 static int
788 nfe_miibus_readreg(device_t dev, int phy, int reg)
789 {
790 	struct nfe_softc *sc = device_get_softc(dev);
791 	uint32_t val;
792 	int ntries;
793 
794 	NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
795 
796 	if (NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY) {
797 		NFE_WRITE(sc, NFE_PHY_CTL, NFE_PHY_BUSY);
798 		DELAY(100);
799 	}
800 
801 	NFE_WRITE(sc, NFE_PHY_CTL, (phy << NFE_PHYADD_SHIFT) | reg);
802 
803 	for (ntries = 0; ntries < 1000; ntries++) {
804 		DELAY(100);
805 		if (!(NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY))
806 			break;
807 	}
808 	if (ntries == 1000) {
809 		DPRINTFN(sc, 2, "timeout waiting for PHY %s\n", "");
810 		return 0;
811 	}
812 
813 	if (NFE_READ(sc, NFE_PHY_STATUS) & NFE_PHY_ERROR) {
814 		DPRINTFN(sc, 2, "could not read PHY %s\n", "");
815 		return 0;
816 	}
817 
818 	val = NFE_READ(sc, NFE_PHY_DATA);
819 	if (val != 0xffffffff && val != 0)
820 		sc->mii_phyaddr = phy;
821 
822 	DPRINTFN(sc, 2, "mii read phy %d reg 0x%x ret 0x%x\n", phy, reg, val);
823 
824 	return val;
825 }
826 
827 static void
828 nfe_miibus_writereg(device_t dev, int phy, int reg, int val)
829 {
830 	struct nfe_softc *sc = device_get_softc(dev);
831 	uint32_t ctl;
832 	int ntries;
833 
834 	NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
835 
836 	if (NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY) {
837 		NFE_WRITE(sc, NFE_PHY_CTL, NFE_PHY_BUSY);
838 		DELAY(100);
839 	}
840 
841 	NFE_WRITE(sc, NFE_PHY_DATA, val);
842 	ctl = NFE_PHY_WRITE | (phy << NFE_PHYADD_SHIFT) | reg;
843 	NFE_WRITE(sc, NFE_PHY_CTL, ctl);
844 
845 	for (ntries = 0; ntries < 1000; ntries++) {
846 		DELAY(100);
847 		if (!(NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY))
848 			break;
849 	}
850 
851 #ifdef NFE_DEBUG
852 	if (ntries == 1000)
853 		DPRINTFN(sc, 2, "could not write to PHY %s\n", "");
854 #endif
855 }
856 
857 #ifdef DEVICE_POLLING
858 
859 static void
860 nfe_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
861 {
862 	struct nfe_softc *sc = ifp->if_softc;
863 
864 	ASSERT_SERIALIZED(ifp->if_serializer);
865 
866 	switch(cmd) {
867 	case POLL_REGISTER:
868 		nfe_disable_intrs(sc);
869 		break;
870 
871 	case POLL_DEREGISTER:
872 		nfe_enable_intrs(sc);
873 		break;
874 
875 	case POLL_AND_CHECK_STATUS:
876 		/* fall through */
877 	case POLL_ONLY:
878 		if (ifp->if_flags & IFF_RUNNING) {
879 			nfe_rxeof(sc);
880 			nfe_txeof(sc, 1);
881 		}
882 		break;
883 	}
884 }
885 
886 #endif
887 
888 static void
889 nfe_intr(void *arg)
890 {
891 	struct nfe_softc *sc = arg;
892 	struct ifnet *ifp = &sc->arpcom.ac_if;
893 	uint32_t r;
894 
895 	r = NFE_READ(sc, NFE_IRQ_STATUS);
896 	if (r == 0)
897 		return;	/* not for us */
898 	NFE_WRITE(sc, NFE_IRQ_STATUS, r);
899 
900 	if (sc->sc_rate_second != time_second) {
901 		/*
902 		 * Calculate sc_rate_avg - interrupts per second.
903 		 */
904 		sc->sc_rate_second = time_second;
905 		if (sc->sc_rate_avg < sc->sc_rate_acc)
906 			sc->sc_rate_avg = sc->sc_rate_acc;
907 		else
908 			sc->sc_rate_avg = (sc->sc_rate_avg * 3 +
909 					   sc->sc_rate_acc) / 4;
910 		sc->sc_rate_acc = 0;
911 	} else if (sc->sc_rate_avg < sc->sc_rate_acc) {
912 		/*
913 		 * Don't wait for a tick to roll over if we are taking
914 		 * a lot of interrupts.
915 		 */
916 		sc->sc_rate_avg = sc->sc_rate_acc;
917 	}
918 
919 	DPRINTFN(sc, 5, "%s: interrupt register %x\n", __func__, r);
920 
921 	if (r & NFE_IRQ_LINK) {
922 		NFE_READ(sc, NFE_PHY_STATUS);
923 		NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
924 		DPRINTF(sc, "link state changed %s\n", "");
925 	}
926 
927 	if (ifp->if_flags & IFF_RUNNING) {
928 		int ret;
929 		int rate;
930 
931 		/* check Rx ring */
932 		ret = nfe_rxeof(sc);
933 
934 		/* check Tx ring */
935 		ret |= nfe_txeof(sc, 1);
936 
937 		/* update the rate accumulator */
938 		if (ret)
939 			++sc->sc_rate_acc;
940 
941 		if (sc->sc_flags & NFE_F_DYN_IM) {
942 			rate = 1000000 / sc->sc_imtime;
943 			if ((sc->sc_flags & NFE_F_IRQ_TIMER) == 0 &&
944 			    sc->sc_rate_avg > rate) {
945 				/*
946 				 * Use the hardware timer to reduce the
947 				 * interrupt rate if the discrete interrupt
948 				 * rate has exceeded our threshold.
949 				 */
950 				NFE_WRITE(sc, NFE_IRQ_MASK, NFE_IRQ_IMTIMER);
951 				sc->sc_flags |= NFE_F_IRQ_TIMER;
952 			} else if ((sc->sc_flags & NFE_F_IRQ_TIMER) &&
953 				   sc->sc_rate_avg <= rate) {
954 				/*
955 				 * Use discrete TX/RX interrupts if the rate
956 				 * has fallen below our threshold.
957 				 */
958 				NFE_WRITE(sc, NFE_IRQ_MASK, NFE_IRQ_NOIMTIMER);
959 				sc->sc_flags &= ~NFE_F_IRQ_TIMER;
960 
961 				/*
962 				 * Recollect, mainly to avoid the possible race
963 				 * introduced by changing interrupt masks.
964 				 */
965 				nfe_rxeof(sc);
966 				nfe_txeof(sc, 1);
967 			}
968 		}
969 	}
970 }
971 
972 static int
973 nfe_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data, struct ucred *cr)
974 {
975 	struct nfe_softc *sc = ifp->if_softc;
976 	struct ifreq *ifr = (struct ifreq *)data;
977 	struct mii_data *mii;
978 	int error = 0, mask, jumbo_cap;
979 
980 	ASSERT_SERIALIZED(ifp->if_serializer);
981 
982 	switch (cmd) {
983 	case SIOCSIFMTU:
984 		if ((sc->sc_caps & NFE_JUMBO_SUP) && sc->rxq.jbuf != NULL)
985 			jumbo_cap = 1;
986 		else
987 			jumbo_cap = 0;
988 
989 		if ((jumbo_cap && ifr->ifr_mtu > NFE_JUMBO_MTU) ||
990 		    (!jumbo_cap && ifr->ifr_mtu > ETHERMTU)) {
991 			return EINVAL;
992 		} else if (ifp->if_mtu != ifr->ifr_mtu) {
993 			ifp->if_mtu = ifr->ifr_mtu;
994 			if (ifp->if_flags & IFF_RUNNING)
995 				nfe_init(sc);
996 		}
997 		break;
998 	case SIOCSIFFLAGS:
999 		if (ifp->if_flags & IFF_UP) {
1000 			/*
1001 			 * If only the PROMISC or ALLMULTI flag changes, then
1002 			 * don't do a full re-init of the chip, just update
1003 			 * the Rx filter.
1004 			 */
1005 			if ((ifp->if_flags & IFF_RUNNING) &&
1006 			    ((ifp->if_flags ^ sc->sc_if_flags) &
1007 			     (IFF_ALLMULTI | IFF_PROMISC)) != 0) {
1008 				nfe_setmulti(sc);
1009 			} else {
1010 				if (!(ifp->if_flags & IFF_RUNNING))
1011 					nfe_init(sc);
1012 			}
1013 		} else {
1014 			if (ifp->if_flags & IFF_RUNNING)
1015 				nfe_stop(sc);
1016 		}
1017 		sc->sc_if_flags = ifp->if_flags;
1018 		break;
1019 	case SIOCADDMULTI:
1020 	case SIOCDELMULTI:
1021 		if (ifp->if_flags & IFF_RUNNING)
1022 			nfe_setmulti(sc);
1023 		break;
1024 	case SIOCSIFMEDIA:
1025 	case SIOCGIFMEDIA:
1026 		mii = device_get_softc(sc->sc_miibus);
1027 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, cmd);
1028 		break;
1029         case SIOCSIFCAP:
1030 		mask = (ifr->ifr_reqcap ^ ifp->if_capenable) & IFCAP_HWCSUM;
1031 		if (mask && (ifp->if_capabilities & IFCAP_HWCSUM)) {
1032 			ifp->if_capenable ^= mask;
1033 			if (IFCAP_TXCSUM & ifp->if_capenable)
1034 				ifp->if_hwassist = NFE_CSUM_FEATURES;
1035 			else
1036 				ifp->if_hwassist = 0;
1037 
1038 			if (ifp->if_flags & IFF_RUNNING)
1039 				nfe_init(sc);
1040 		}
1041 		break;
1042 	default:
1043 		error = ether_ioctl(ifp, cmd, data);
1044 		break;
1045 	}
1046 	return error;
1047 }
1048 
1049 static int
1050 nfe_rxeof(struct nfe_softc *sc)
1051 {
1052 	struct ifnet *ifp = &sc->arpcom.ac_if;
1053 	struct nfe_rx_ring *ring = &sc->rxq;
1054 	int reap;
1055 	struct mbuf_chain chain[MAXCPU];
1056 
1057 	reap = 0;
1058 	ether_input_chain_init(chain);
1059 
1060 	for (;;) {
1061 		struct nfe_rx_data *data = &ring->data[ring->cur];
1062 		struct mbuf *m;
1063 		uint16_t flags;
1064 		int len, error;
1065 
1066 		if (sc->sc_caps & NFE_40BIT_ADDR) {
1067 			struct nfe_desc64 *desc64 = &ring->desc64[ring->cur];
1068 
1069 			flags = le16toh(desc64->flags);
1070 			len = le16toh(desc64->length) & 0x3fff;
1071 		} else {
1072 			struct nfe_desc32 *desc32 = &ring->desc32[ring->cur];
1073 
1074 			flags = le16toh(desc32->flags);
1075 			len = le16toh(desc32->length) & 0x3fff;
1076 		}
1077 
1078 		if (flags & NFE_RX_READY)
1079 			break;
1080 
1081 		reap = 1;
1082 
1083 		if ((sc->sc_caps & (NFE_JUMBO_SUP | NFE_40BIT_ADDR)) == 0) {
1084 			if (!(flags & NFE_RX_VALID_V1))
1085 				goto skip;
1086 
1087 			if ((flags & NFE_RX_FIXME_V1) == NFE_RX_FIXME_V1) {
1088 				flags &= ~NFE_RX_ERROR;
1089 				len--;	/* fix buffer length */
1090 			}
1091 		} else {
1092 			if (!(flags & NFE_RX_VALID_V2))
1093 				goto skip;
1094 
1095 			if ((flags & NFE_RX_FIXME_V2) == NFE_RX_FIXME_V2) {
1096 				flags &= ~NFE_RX_ERROR;
1097 				len--;	/* fix buffer length */
1098 			}
1099 		}
1100 
1101 		if (flags & NFE_RX_ERROR) {
1102 			ifp->if_ierrors++;
1103 			goto skip;
1104 		}
1105 
1106 		m = data->m;
1107 
1108 		if (sc->sc_flags & NFE_F_USE_JUMBO)
1109 			error = nfe_newbuf_jumbo(sc, ring, ring->cur, 0);
1110 		else
1111 			error = nfe_newbuf_std(sc, ring, ring->cur, 0);
1112 		if (error) {
1113 			ifp->if_ierrors++;
1114 			goto skip;
1115 		}
1116 
1117 		/* finalize mbuf */
1118 		m->m_pkthdr.len = m->m_len = len;
1119 		m->m_pkthdr.rcvif = ifp;
1120 
1121 		if ((ifp->if_capenable & IFCAP_RXCSUM) &&
1122 		    (flags & NFE_RX_CSUMOK)) {
1123 			if (flags & NFE_RX_IP_CSUMOK_V2) {
1124 				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED |
1125 							  CSUM_IP_VALID;
1126 			}
1127 
1128 			if (flags &
1129 			    (NFE_RX_UDP_CSUMOK_V2 | NFE_RX_TCP_CSUMOK_V2)) {
1130 				m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
1131 							  CSUM_PSEUDO_HDR |
1132 							  CSUM_FRAG_NOT_CHECKED;
1133 				m->m_pkthdr.csum_data = 0xffff;
1134 			}
1135 		}
1136 
1137 		ifp->if_ipackets++;
1138 		ether_input_chain(ifp, m, NULL, chain);
1139 skip:
1140 		nfe_set_ready_rxdesc(sc, ring, ring->cur);
1141 		sc->rxq.cur = (sc->rxq.cur + 1) % sc->sc_rx_ring_count;
1142 	}
1143 
1144 	if (reap)
1145 		ether_input_dispatch(chain);
1146 	return reap;
1147 }
1148 
1149 static int
1150 nfe_txeof(struct nfe_softc *sc, int start)
1151 {
1152 	struct ifnet *ifp = &sc->arpcom.ac_if;
1153 	struct nfe_tx_ring *ring = &sc->txq;
1154 	struct nfe_tx_data *data = NULL;
1155 
1156 	while (ring->next != ring->cur) {
1157 		uint16_t flags;
1158 
1159 		if (sc->sc_caps & NFE_40BIT_ADDR)
1160 			flags = le16toh(ring->desc64[ring->next].flags);
1161 		else
1162 			flags = le16toh(ring->desc32[ring->next].flags);
1163 
1164 		if (flags & NFE_TX_VALID)
1165 			break;
1166 
1167 		data = &ring->data[ring->next];
1168 
1169 		if ((sc->sc_caps & (NFE_JUMBO_SUP | NFE_40BIT_ADDR)) == 0) {
1170 			if (!(flags & NFE_TX_LASTFRAG_V1) && data->m == NULL)
1171 				goto skip;
1172 
1173 			if ((flags & NFE_TX_ERROR_V1) != 0) {
1174 				if_printf(ifp, "tx v1 error 0x%4b\n", flags,
1175 					  NFE_V1_TXERR);
1176 				ifp->if_oerrors++;
1177 			} else {
1178 				ifp->if_opackets++;
1179 			}
1180 		} else {
1181 			if (!(flags & NFE_TX_LASTFRAG_V2) && data->m == NULL)
1182 				goto skip;
1183 
1184 			if ((flags & NFE_TX_ERROR_V2) != 0) {
1185 				if_printf(ifp, "tx v2 error 0x%4b\n", flags,
1186 					  NFE_V2_TXERR);
1187 				ifp->if_oerrors++;
1188 			} else {
1189 				ifp->if_opackets++;
1190 			}
1191 		}
1192 
1193 		if (data->m == NULL) {	/* should not get there */
1194 			if_printf(ifp,
1195 				  "last fragment bit w/o associated mbuf!\n");
1196 			goto skip;
1197 		}
1198 
1199 		/* last fragment of the mbuf chain transmitted */
1200 		bus_dmamap_unload(ring->data_tag, data->map);
1201 		m_freem(data->m);
1202 		data->m = NULL;
1203 skip:
1204 		ring->queued--;
1205 		KKASSERT(ring->queued >= 0);
1206 		ring->next = (ring->next + 1) % sc->sc_tx_ring_count;
1207 	}
1208 
1209 	if (sc->sc_tx_ring_count - ring->queued >=
1210 	    sc->sc_tx_spare + NFE_NSEG_RSVD)
1211 		ifp->if_flags &= ~IFF_OACTIVE;
1212 
1213 	if (ring->queued == 0)
1214 		ifp->if_timer = 0;
1215 
1216 	if (start && !ifq_is_empty(&ifp->if_snd))
1217 		if_devstart(ifp);
1218 
1219 	if (data != NULL)
1220 		return 1;
1221 	else
1222 		return 0;
1223 }
1224 
1225 static int
1226 nfe_encap(struct nfe_softc *sc, struct nfe_tx_ring *ring, struct mbuf *m0)
1227 {
1228 	bus_dma_segment_t segs[NFE_MAX_SCATTER];
1229 	struct nfe_tx_data *data, *data_map;
1230 	bus_dmamap_t map;
1231 	struct nfe_desc64 *desc64 = NULL;
1232 	struct nfe_desc32 *desc32 = NULL;
1233 	uint16_t flags = 0;
1234 	uint32_t vtag = 0;
1235 	int error, i, j, maxsegs, nsegs;
1236 
1237 	data = &ring->data[ring->cur];
1238 	map = data->map;
1239 	data_map = data;	/* Remember who owns the DMA map */
1240 
1241 	maxsegs = (sc->sc_tx_ring_count - ring->queued) - NFE_NSEG_RSVD;
1242 	if (maxsegs > NFE_MAX_SCATTER)
1243 		maxsegs = NFE_MAX_SCATTER;
1244 	KASSERT(maxsegs >= sc->sc_tx_spare,
1245 		("no enough segments %d,%d\n", maxsegs, sc->sc_tx_spare));
1246 
1247 	error = bus_dmamap_load_mbuf_defrag(ring->data_tag, map, &m0,
1248 			segs, maxsegs, &nsegs, BUS_DMA_NOWAIT);
1249 	if (error)
1250 		goto back;
1251 	bus_dmamap_sync(ring->data_tag, map, BUS_DMASYNC_PREWRITE);
1252 
1253 	error = 0;
1254 
1255 	/* setup h/w VLAN tagging */
1256 	if (m0->m_flags & M_VLANTAG)
1257 		vtag = m0->m_pkthdr.ether_vlantag;
1258 
1259 	if (sc->arpcom.ac_if.if_capenable & IFCAP_TXCSUM) {
1260 		if (m0->m_pkthdr.csum_flags & CSUM_IP)
1261 			flags |= NFE_TX_IP_CSUM;
1262 		if (m0->m_pkthdr.csum_flags & (CSUM_TCP | CSUM_UDP))
1263 			flags |= NFE_TX_TCP_CSUM;
1264 	}
1265 
1266 	/*
1267 	 * XXX urm. somebody is unaware of how hardware works.  You
1268 	 * absolutely CANNOT set NFE_TX_VALID on the next descriptor in
1269 	 * the ring until the entire chain is actually *VALID*.  Otherwise
1270 	 * the hardware may encounter a partially initialized chain that
1271 	 * is marked as being ready to go when it in fact is not ready to
1272 	 * go.
1273 	 */
1274 
1275 	for (i = 0; i < nsegs; i++) {
1276 		j = (ring->cur + i) % sc->sc_tx_ring_count;
1277 		data = &ring->data[j];
1278 
1279 		if (sc->sc_caps & NFE_40BIT_ADDR) {
1280 			desc64 = &ring->desc64[j];
1281 			desc64->physaddr[0] =
1282 			    htole32(NFE_ADDR_HI(segs[i].ds_addr));
1283 			desc64->physaddr[1] =
1284 			    htole32(NFE_ADDR_LO(segs[i].ds_addr));
1285 			desc64->length = htole16(segs[i].ds_len - 1);
1286 			desc64->vtag = htole32(vtag);
1287 			desc64->flags = htole16(flags);
1288 		} else {
1289 			desc32 = &ring->desc32[j];
1290 			desc32->physaddr = htole32(segs[i].ds_addr);
1291 			desc32->length = htole16(segs[i].ds_len - 1);
1292 			desc32->flags = htole16(flags);
1293 		}
1294 
1295 		/* csum flags and vtag belong to the first fragment only */
1296 		flags &= ~(NFE_TX_IP_CSUM | NFE_TX_TCP_CSUM);
1297 		vtag = 0;
1298 
1299 		ring->queued++;
1300 		KKASSERT(ring->queued <= sc->sc_tx_ring_count);
1301 	}
1302 
1303 	/* the whole mbuf chain has been DMA mapped, fix last descriptor */
1304 	if (sc->sc_caps & NFE_40BIT_ADDR) {
1305 		desc64->flags |= htole16(NFE_TX_LASTFRAG_V2);
1306 	} else {
1307 		if (sc->sc_caps & NFE_JUMBO_SUP)
1308 			flags = NFE_TX_LASTFRAG_V2;
1309 		else
1310 			flags = NFE_TX_LASTFRAG_V1;
1311 		desc32->flags |= htole16(flags);
1312 	}
1313 
1314 	/*
1315 	 * Set NFE_TX_VALID backwards so the hardware doesn't see the
1316 	 * whole mess until the first descriptor in the map is flagged.
1317 	 */
1318 	for (i = nsegs - 1; i >= 0; --i) {
1319 		j = (ring->cur + i) % sc->sc_tx_ring_count;
1320 		if (sc->sc_caps & NFE_40BIT_ADDR) {
1321 			desc64 = &ring->desc64[j];
1322 			desc64->flags |= htole16(NFE_TX_VALID);
1323 		} else {
1324 			desc32 = &ring->desc32[j];
1325 			desc32->flags |= htole16(NFE_TX_VALID);
1326 		}
1327 	}
1328 	ring->cur = (ring->cur + nsegs) % sc->sc_tx_ring_count;
1329 
1330 	/* Exchange DMA map */
1331 	data_map->map = data->map;
1332 	data->map = map;
1333 	data->m = m0;
1334 back:
1335 	if (error)
1336 		m_freem(m0);
1337 	return error;
1338 }
1339 
1340 static void
1341 nfe_start(struct ifnet *ifp)
1342 {
1343 	struct nfe_softc *sc = ifp->if_softc;
1344 	struct nfe_tx_ring *ring = &sc->txq;
1345 	int count = 0, oactive = 0;
1346 	struct mbuf *m0;
1347 
1348 	ASSERT_SERIALIZED(ifp->if_serializer);
1349 
1350 	if ((ifp->if_flags & (IFF_OACTIVE | IFF_RUNNING)) != IFF_RUNNING)
1351 		return;
1352 
1353 	for (;;) {
1354 		int error;
1355 
1356 		if (sc->sc_tx_ring_count - ring->queued <
1357 		    sc->sc_tx_spare + NFE_NSEG_RSVD) {
1358 			if (oactive) {
1359 				ifp->if_flags |= IFF_OACTIVE;
1360 				break;
1361 			}
1362 
1363 			nfe_txeof(sc, 0);
1364 			oactive = 1;
1365 			continue;
1366 		}
1367 
1368 		m0 = ifq_dequeue(&ifp->if_snd, NULL);
1369 		if (m0 == NULL)
1370 			break;
1371 
1372 		ETHER_BPF_MTAP(ifp, m0);
1373 
1374 		error = nfe_encap(sc, ring, m0);
1375 		if (error) {
1376 			ifp->if_oerrors++;
1377 			if (error == EFBIG) {
1378 				if (oactive) {
1379 					ifp->if_flags |= IFF_OACTIVE;
1380 					break;
1381 				}
1382 				nfe_txeof(sc, 0);
1383 				oactive = 1;
1384 			}
1385 			continue;
1386 		} else {
1387 			oactive = 0;
1388 		}
1389 		++count;
1390 
1391 		/*
1392 		 * NOTE:
1393 		 * `m0' may be freed in nfe_encap(), so
1394 		 * it should not be touched any more.
1395 		 */
1396 	}
1397 
1398 	if (count == 0)	/* nothing sent */
1399 		return;
1400 
1401 	/* Kick Tx */
1402 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_KICKTX | sc->rxtxctl);
1403 
1404 	/*
1405 	 * Set a timeout in case the chip goes out to lunch.
1406 	 */
1407 	ifp->if_timer = 5;
1408 }
1409 
1410 static void
1411 nfe_watchdog(struct ifnet *ifp)
1412 {
1413 	struct nfe_softc *sc = ifp->if_softc;
1414 
1415 	ASSERT_SERIALIZED(ifp->if_serializer);
1416 
1417 	if (ifp->if_flags & IFF_RUNNING) {
1418 		if_printf(ifp, "watchdog timeout - lost interrupt recovered\n");
1419 		nfe_txeof(sc, 1);
1420 		return;
1421 	}
1422 
1423 	if_printf(ifp, "watchdog timeout\n");
1424 
1425 	nfe_init(ifp->if_softc);
1426 
1427 	ifp->if_oerrors++;
1428 }
1429 
1430 static void
1431 nfe_init(void *xsc)
1432 {
1433 	struct nfe_softc *sc = xsc;
1434 	struct ifnet *ifp = &sc->arpcom.ac_if;
1435 	uint32_t tmp;
1436 	int error;
1437 
1438 	ASSERT_SERIALIZED(ifp->if_serializer);
1439 
1440 	nfe_stop(sc);
1441 
1442 	if ((sc->sc_caps & NFE_NO_PWRCTL) == 0)
1443 		nfe_mac_reset(sc);
1444 
1445 	/*
1446 	 * NOTE:
1447 	 * Switching between jumbo frames and normal frames should
1448 	 * be done _after_ nfe_stop() but _before_ nfe_init_rx_ring().
1449 	 */
1450 	if (ifp->if_mtu > ETHERMTU) {
1451 		sc->sc_flags |= NFE_F_USE_JUMBO;
1452 		sc->rxq.bufsz = NFE_JBYTES;
1453 		sc->sc_tx_spare = NFE_NSEG_SPARE_JUMBO;
1454 		if (bootverbose)
1455 			if_printf(ifp, "use jumbo frames\n");
1456 	} else {
1457 		sc->sc_flags &= ~NFE_F_USE_JUMBO;
1458 		sc->rxq.bufsz = MCLBYTES;
1459 		sc->sc_tx_spare = NFE_NSEG_SPARE;
1460 		if (bootverbose)
1461 			if_printf(ifp, "use non-jumbo frames\n");
1462 	}
1463 
1464 	error = nfe_init_tx_ring(sc, &sc->txq);
1465 	if (error) {
1466 		nfe_stop(sc);
1467 		return;
1468 	}
1469 
1470 	error = nfe_init_rx_ring(sc, &sc->rxq);
1471 	if (error) {
1472 		nfe_stop(sc);
1473 		return;
1474 	}
1475 
1476 	NFE_WRITE(sc, NFE_TX_POLL, 0);
1477 	NFE_WRITE(sc, NFE_STATUS, 0);
1478 
1479 	sc->rxtxctl = NFE_RXTX_BIT2 | sc->rxtxctl_desc;
1480 
1481 	if (ifp->if_capenable & IFCAP_RXCSUM)
1482 		sc->rxtxctl |= NFE_RXTX_RXCSUM;
1483 
1484 	/*
1485 	 * Although the adapter is capable of stripping VLAN tags from received
1486 	 * frames (NFE_RXTX_VTAG_STRIP), we do not enable this functionality on
1487 	 * purpose.  This will be done in software by our network stack.
1488 	 */
1489 	if (sc->sc_caps & NFE_HW_VLAN)
1490 		sc->rxtxctl |= NFE_RXTX_VTAG_INSERT;
1491 
1492 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_RESET | sc->rxtxctl);
1493 	DELAY(10);
1494 	NFE_WRITE(sc, NFE_RXTX_CTL, sc->rxtxctl);
1495 
1496 	if (sc->sc_caps & NFE_HW_VLAN)
1497 		NFE_WRITE(sc, NFE_VTAG_CTL, NFE_VTAG_ENABLE);
1498 
1499 	NFE_WRITE(sc, NFE_SETUP_R6, 0);
1500 
1501 	/* set MAC address */
1502 	nfe_set_macaddr(sc, sc->arpcom.ac_enaddr);
1503 
1504 	/* tell MAC where rings are in memory */
1505 	if (sc->sc_caps & NFE_40BIT_ADDR) {
1506 		NFE_WRITE(sc, NFE_RX_RING_ADDR_HI,
1507 			  NFE_ADDR_HI(sc->rxq.physaddr));
1508 	}
1509 	NFE_WRITE(sc, NFE_RX_RING_ADDR_LO, NFE_ADDR_LO(sc->rxq.physaddr));
1510 
1511 	if (sc->sc_caps & NFE_40BIT_ADDR) {
1512 		NFE_WRITE(sc, NFE_TX_RING_ADDR_HI,
1513 			  NFE_ADDR_HI(sc->txq.physaddr));
1514 	}
1515 	NFE_WRITE(sc, NFE_TX_RING_ADDR_LO, NFE_ADDR_LO(sc->txq.physaddr));
1516 
1517 	NFE_WRITE(sc, NFE_RING_SIZE,
1518 	    (sc->sc_rx_ring_count - 1) << 16 |
1519 	    (sc->sc_tx_ring_count - 1));
1520 
1521 	NFE_WRITE(sc, NFE_RXBUFSZ, sc->rxq.bufsz);
1522 
1523 	/* force MAC to wakeup */
1524 	tmp = NFE_READ(sc, NFE_PWR_STATE);
1525 	NFE_WRITE(sc, NFE_PWR_STATE, tmp | NFE_PWR_WAKEUP);
1526 	DELAY(10);
1527 	tmp = NFE_READ(sc, NFE_PWR_STATE);
1528 	NFE_WRITE(sc, NFE_PWR_STATE, tmp | NFE_PWR_VALID);
1529 
1530 	NFE_WRITE(sc, NFE_SETUP_R1, NFE_R1_MAGIC);
1531 	NFE_WRITE(sc, NFE_SETUP_R2, NFE_R2_MAGIC);
1532 	NFE_WRITE(sc, NFE_SETUP_R6, NFE_R6_MAGIC);
1533 
1534 	/* update MAC knowledge of PHY; generates a NFE_IRQ_LINK interrupt */
1535 	NFE_WRITE(sc, NFE_STATUS, sc->mii_phyaddr << 24 | NFE_STATUS_MAGIC);
1536 
1537 	NFE_WRITE(sc, NFE_SETUP_R4, NFE_R4_MAGIC);
1538 
1539 	sc->rxtxctl &= ~NFE_RXTX_BIT2;
1540 	NFE_WRITE(sc, NFE_RXTX_CTL, sc->rxtxctl);
1541 	DELAY(10);
1542 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_BIT1 | sc->rxtxctl);
1543 
1544 	/* set Rx filter */
1545 	nfe_setmulti(sc);
1546 
1547 	nfe_ifmedia_upd(ifp);
1548 
1549 	/* enable Rx */
1550 	NFE_WRITE(sc, NFE_RX_CTL, NFE_RX_START);
1551 
1552 	/* enable Tx */
1553 	NFE_WRITE(sc, NFE_TX_CTL, NFE_TX_START);
1554 
1555 	NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
1556 
1557 #ifdef DEVICE_POLLING
1558 	if ((ifp->if_flags & IFF_POLLING))
1559 		nfe_disable_intrs(sc);
1560 	else
1561 #endif
1562 	nfe_enable_intrs(sc);
1563 
1564 	callout_reset(&sc->sc_tick_ch, hz, nfe_tick, sc);
1565 
1566 	ifp->if_flags |= IFF_RUNNING;
1567 	ifp->if_flags &= ~IFF_OACTIVE;
1568 
1569 	/*
1570 	 * If we had stuff in the tx ring before its all cleaned out now
1571 	 * so we are not going to get an interrupt, jump-start any pending
1572 	 * output.
1573 	 */
1574 	if (!ifq_is_empty(&ifp->if_snd))
1575 		if_devstart(ifp);
1576 }
1577 
1578 static void
1579 nfe_stop(struct nfe_softc *sc)
1580 {
1581 	struct ifnet *ifp = &sc->arpcom.ac_if;
1582 	uint32_t rxtxctl = sc->rxtxctl_desc | NFE_RXTX_BIT2;
1583 	int i;
1584 
1585 	ASSERT_SERIALIZED(ifp->if_serializer);
1586 
1587 	callout_stop(&sc->sc_tick_ch);
1588 
1589 	ifp->if_timer = 0;
1590 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
1591 	sc->sc_flags &= ~NFE_F_IRQ_TIMER;
1592 
1593 #define WAITMAX	50000
1594 
1595 	/*
1596 	 * Abort Tx
1597 	 */
1598 	NFE_WRITE(sc, NFE_TX_CTL, 0);
1599 	for (i = 0; i < WAITMAX; ++i) {
1600 		DELAY(100);
1601 		if ((NFE_READ(sc, NFE_TX_STATUS) & NFE_TX_STATUS_BUSY) == 0)
1602 			break;
1603 	}
1604 	if (i == WAITMAX)
1605 		if_printf(ifp, "can't stop TX\n");
1606 	DELAY(100);
1607 
1608 	/*
1609 	 * Disable Rx
1610 	 */
1611 	NFE_WRITE(sc, NFE_RX_CTL, 0);
1612 	for (i = 0; i < WAITMAX; ++i) {
1613 		DELAY(100);
1614 		if ((NFE_READ(sc, NFE_RX_STATUS) & NFE_RX_STATUS_BUSY) == 0)
1615 			break;
1616 	}
1617 	if (i == WAITMAX)
1618 		if_printf(ifp, "can't stop RX\n");
1619 	DELAY(100);
1620 
1621 #undef WAITMAX
1622 
1623 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_RESET | rxtxctl);
1624 	DELAY(10);
1625 	NFE_WRITE(sc, NFE_RXTX_CTL, rxtxctl);
1626 
1627 	/* Disable interrupts */
1628 	NFE_WRITE(sc, NFE_IRQ_MASK, 0);
1629 
1630 	/* Reset Tx and Rx rings */
1631 	nfe_reset_tx_ring(sc, &sc->txq);
1632 	nfe_reset_rx_ring(sc, &sc->rxq);
1633 }
1634 
1635 static int
1636 nfe_alloc_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1637 {
1638 	int i, j, error, descsize;
1639 	bus_dmamem_t dmem;
1640 	void **desc;
1641 
1642 	if (sc->sc_caps & NFE_40BIT_ADDR) {
1643 		desc = (void *)&ring->desc64;
1644 		descsize = sizeof(struct nfe_desc64);
1645 	} else {
1646 		desc = (void *)&ring->desc32;
1647 		descsize = sizeof(struct nfe_desc32);
1648 	}
1649 
1650 	ring->bufsz = MCLBYTES;
1651 	ring->cur = ring->next = 0;
1652 
1653 	error = bus_dmamem_coherent(sc->sc_dtag, PAGE_SIZE, 0,
1654 				    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
1655 				    sc->sc_rx_ring_count * descsize,
1656 				    BUS_DMA_WAITOK | BUS_DMA_ZERO, &dmem);
1657 	if (error) {
1658 		if_printf(&sc->arpcom.ac_if,
1659 			  "could not create RX desc ring\n");
1660 		return error;
1661 	}
1662 	ring->tag = dmem.dmem_tag;
1663 	ring->map = dmem.dmem_map;
1664 	*desc = dmem.dmem_addr;
1665 	ring->physaddr = dmem.dmem_busaddr;
1666 
1667 	if (sc->sc_caps & NFE_JUMBO_SUP) {
1668 		ring->jbuf =
1669 		kmalloc(sizeof(struct nfe_jbuf) * NFE_JPOOL_COUNT(sc),
1670 			M_DEVBUF, M_WAITOK | M_ZERO);
1671 
1672 		error = nfe_jpool_alloc(sc, ring);
1673 		if (error) {
1674 			if_printf(&sc->arpcom.ac_if,
1675 				  "could not allocate jumbo frames\n");
1676 			kfree(ring->jbuf, M_DEVBUF);
1677 			ring->jbuf = NULL;
1678 			/* Allow jumbo frame allocation to fail */
1679 		}
1680 	}
1681 
1682 	ring->data = kmalloc(sizeof(struct nfe_rx_data) * sc->sc_rx_ring_count,
1683 			     M_DEVBUF, M_WAITOK | M_ZERO);
1684 
1685 	error = bus_dma_tag_create(sc->sc_dtag, 1, 0,
1686 				   BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
1687 				   NULL, NULL,
1688 				   MCLBYTES, 1, MCLBYTES,
1689 				   BUS_DMA_ALLOCNOW | BUS_DMA_WAITOK,
1690 				   &ring->data_tag);
1691 	if (error) {
1692 		if_printf(&sc->arpcom.ac_if,
1693 			  "could not create RX mbuf DMA tag\n");
1694 		return error;
1695 	}
1696 
1697 	/* Create a spare RX mbuf DMA map */
1698 	error = bus_dmamap_create(ring->data_tag, BUS_DMA_WAITOK,
1699 				  &ring->data_tmpmap);
1700 	if (error) {
1701 		if_printf(&sc->arpcom.ac_if,
1702 			  "could not create spare RX mbuf DMA map\n");
1703 		bus_dma_tag_destroy(ring->data_tag);
1704 		ring->data_tag = NULL;
1705 		return error;
1706 	}
1707 
1708 	for (i = 0; i < sc->sc_rx_ring_count; i++) {
1709 		error = bus_dmamap_create(ring->data_tag, BUS_DMA_WAITOK,
1710 					  &ring->data[i].map);
1711 		if (error) {
1712 			if_printf(&sc->arpcom.ac_if,
1713 				  "could not create %dth RX mbuf DMA mapn", i);
1714 			goto fail;
1715 		}
1716 	}
1717 	return 0;
1718 fail:
1719 	for (j = 0; j < i; ++j)
1720 		bus_dmamap_destroy(ring->data_tag, ring->data[i].map);
1721 	bus_dmamap_destroy(ring->data_tag, ring->data_tmpmap);
1722 	bus_dma_tag_destroy(ring->data_tag);
1723 	ring->data_tag = NULL;
1724 	return error;
1725 }
1726 
1727 static void
1728 nfe_reset_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1729 {
1730 	int i;
1731 
1732 	for (i = 0; i < sc->sc_rx_ring_count; i++) {
1733 		struct nfe_rx_data *data = &ring->data[i];
1734 
1735 		if (data->m != NULL) {
1736 			if ((sc->sc_flags & NFE_F_USE_JUMBO) == 0)
1737 				bus_dmamap_unload(ring->data_tag, data->map);
1738 			m_freem(data->m);
1739 			data->m = NULL;
1740 		}
1741 	}
1742 
1743 	ring->cur = ring->next = 0;
1744 }
1745 
1746 static int
1747 nfe_init_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1748 {
1749 	int i;
1750 
1751 	for (i = 0; i < sc->sc_rx_ring_count; ++i) {
1752 		int error;
1753 
1754 		/* XXX should use a function pointer */
1755 		if (sc->sc_flags & NFE_F_USE_JUMBO)
1756 			error = nfe_newbuf_jumbo(sc, ring, i, 1);
1757 		else
1758 			error = nfe_newbuf_std(sc, ring, i, 1);
1759 		if (error) {
1760 			if_printf(&sc->arpcom.ac_if,
1761 				  "could not allocate RX buffer\n");
1762 			return error;
1763 		}
1764 		nfe_set_ready_rxdesc(sc, ring, i);
1765 	}
1766 	return 0;
1767 }
1768 
1769 static void
1770 nfe_free_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1771 {
1772 	if (ring->data_tag != NULL) {
1773 		struct nfe_rx_data *data;
1774 		int i;
1775 
1776 		for (i = 0; i < sc->sc_rx_ring_count; i++) {
1777 			data = &ring->data[i];
1778 
1779 			if (data->m != NULL) {
1780 				bus_dmamap_unload(ring->data_tag, data->map);
1781 				m_freem(data->m);
1782 			}
1783 			bus_dmamap_destroy(ring->data_tag, data->map);
1784 		}
1785 		bus_dmamap_destroy(ring->data_tag, ring->data_tmpmap);
1786 		bus_dma_tag_destroy(ring->data_tag);
1787 	}
1788 
1789 	nfe_jpool_free(sc, ring);
1790 
1791 	if (ring->jbuf != NULL)
1792 		kfree(ring->jbuf, M_DEVBUF);
1793 	if (ring->data != NULL)
1794 		kfree(ring->data, M_DEVBUF);
1795 
1796 	if (ring->tag != NULL) {
1797 		void *desc;
1798 
1799 		if (sc->sc_caps & NFE_40BIT_ADDR)
1800 			desc = ring->desc64;
1801 		else
1802 			desc = ring->desc32;
1803 
1804 		bus_dmamap_unload(ring->tag, ring->map);
1805 		bus_dmamem_free(ring->tag, desc, ring->map);
1806 		bus_dma_tag_destroy(ring->tag);
1807 	}
1808 }
1809 
1810 static struct nfe_jbuf *
1811 nfe_jalloc(struct nfe_softc *sc)
1812 {
1813 	struct ifnet *ifp = &sc->arpcom.ac_if;
1814 	struct nfe_jbuf *jbuf;
1815 
1816 	lwkt_serialize_enter(&sc->sc_jbuf_serializer);
1817 
1818 	jbuf = SLIST_FIRST(&sc->rxq.jfreelist);
1819 	if (jbuf != NULL) {
1820 		SLIST_REMOVE_HEAD(&sc->rxq.jfreelist, jnext);
1821 		jbuf->inuse = 1;
1822 	} else {
1823 		if_printf(ifp, "no free jumbo buffer\n");
1824 	}
1825 
1826 	lwkt_serialize_exit(&sc->sc_jbuf_serializer);
1827 
1828 	return jbuf;
1829 }
1830 
1831 static void
1832 nfe_jfree(void *arg)
1833 {
1834 	struct nfe_jbuf *jbuf = arg;
1835 	struct nfe_softc *sc = jbuf->sc;
1836 	struct nfe_rx_ring *ring = jbuf->ring;
1837 
1838 	if (&ring->jbuf[jbuf->slot] != jbuf)
1839 		panic("%s: free wrong jumbo buffer\n", __func__);
1840 	else if (jbuf->inuse == 0)
1841 		panic("%s: jumbo buffer already freed\n", __func__);
1842 
1843 	lwkt_serialize_enter(&sc->sc_jbuf_serializer);
1844 	atomic_subtract_int(&jbuf->inuse, 1);
1845 	if (jbuf->inuse == 0)
1846 		SLIST_INSERT_HEAD(&ring->jfreelist, jbuf, jnext);
1847 	lwkt_serialize_exit(&sc->sc_jbuf_serializer);
1848 }
1849 
1850 static void
1851 nfe_jref(void *arg)
1852 {
1853 	struct nfe_jbuf *jbuf = arg;
1854 	struct nfe_rx_ring *ring = jbuf->ring;
1855 
1856 	if (&ring->jbuf[jbuf->slot] != jbuf)
1857 		panic("%s: ref wrong jumbo buffer\n", __func__);
1858 	else if (jbuf->inuse == 0)
1859 		panic("%s: jumbo buffer already freed\n", __func__);
1860 
1861 	atomic_add_int(&jbuf->inuse, 1);
1862 }
1863 
1864 static int
1865 nfe_jpool_alloc(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1866 {
1867 	struct nfe_jbuf *jbuf;
1868 	bus_dmamem_t dmem;
1869 	bus_addr_t physaddr;
1870 	caddr_t buf;
1871 	int i, error;
1872 
1873 	/*
1874 	 * Allocate a big chunk of DMA'able memory.
1875 	 */
1876 	error = bus_dmamem_coherent(sc->sc_dtag, PAGE_SIZE, 0,
1877 				    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
1878 				    NFE_JPOOL_SIZE(sc),
1879 				    BUS_DMA_WAITOK, &dmem);
1880 	if (error) {
1881 		if_printf(&sc->arpcom.ac_if,
1882 			  "could not create jumbo buffer\n");
1883 		return error;
1884 	}
1885 	ring->jtag = dmem.dmem_tag;
1886 	ring->jmap = dmem.dmem_map;
1887 	ring->jpool = dmem.dmem_addr;
1888 	physaddr = dmem.dmem_busaddr;
1889 
1890 	/* ..and split it into 9KB chunks */
1891 	SLIST_INIT(&ring->jfreelist);
1892 
1893 	buf = ring->jpool;
1894 	for (i = 0; i < NFE_JPOOL_COUNT(sc); i++) {
1895 		jbuf = &ring->jbuf[i];
1896 
1897 		jbuf->sc = sc;
1898 		jbuf->ring = ring;
1899 		jbuf->inuse = 0;
1900 		jbuf->slot = i;
1901 		jbuf->buf = buf;
1902 		jbuf->physaddr = physaddr;
1903 
1904 		SLIST_INSERT_HEAD(&ring->jfreelist, jbuf, jnext);
1905 
1906 		buf += NFE_JBYTES;
1907 		physaddr += NFE_JBYTES;
1908 	}
1909 
1910 	return 0;
1911 }
1912 
1913 static void
1914 nfe_jpool_free(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1915 {
1916 	if (ring->jtag != NULL) {
1917 		bus_dmamap_unload(ring->jtag, ring->jmap);
1918 		bus_dmamem_free(ring->jtag, ring->jpool, ring->jmap);
1919 		bus_dma_tag_destroy(ring->jtag);
1920 	}
1921 }
1922 
1923 static int
1924 nfe_alloc_tx_ring(struct nfe_softc *sc, struct nfe_tx_ring *ring)
1925 {
1926 	int i, j, error, descsize;
1927 	bus_dmamem_t dmem;
1928 	void **desc;
1929 
1930 	if (sc->sc_caps & NFE_40BIT_ADDR) {
1931 		desc = (void *)&ring->desc64;
1932 		descsize = sizeof(struct nfe_desc64);
1933 	} else {
1934 		desc = (void *)&ring->desc32;
1935 		descsize = sizeof(struct nfe_desc32);
1936 	}
1937 
1938 	ring->queued = 0;
1939 	ring->cur = ring->next = 0;
1940 
1941 	error = bus_dmamem_coherent(sc->sc_dtag, PAGE_SIZE, 0,
1942 				    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
1943 				    sc->sc_tx_ring_count * descsize,
1944 				    BUS_DMA_WAITOK | BUS_DMA_ZERO, &dmem);
1945 	if (error) {
1946 		if_printf(&sc->arpcom.ac_if,
1947 			  "could not create TX desc ring\n");
1948 		return error;
1949 	}
1950 	ring->tag = dmem.dmem_tag;
1951 	ring->map = dmem.dmem_map;
1952 	*desc = dmem.dmem_addr;
1953 	ring->physaddr = dmem.dmem_busaddr;
1954 
1955 	ring->data = kmalloc(sizeof(struct nfe_tx_data) * sc->sc_tx_ring_count,
1956 			     M_DEVBUF, M_WAITOK | M_ZERO);
1957 
1958 	error = bus_dma_tag_create(sc->sc_dtag, 1, 0,
1959 			BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
1960 			NULL, NULL,
1961 			NFE_JBYTES, NFE_MAX_SCATTER, MCLBYTES,
1962 			BUS_DMA_ALLOCNOW | BUS_DMA_WAITOK | BUS_DMA_ONEBPAGE,
1963 			&ring->data_tag);
1964 	if (error) {
1965 		if_printf(&sc->arpcom.ac_if,
1966 			  "could not create TX buf DMA tag\n");
1967 		return error;
1968 	}
1969 
1970 	for (i = 0; i < sc->sc_tx_ring_count; i++) {
1971 		error = bus_dmamap_create(ring->data_tag,
1972 				BUS_DMA_WAITOK | BUS_DMA_ONEBPAGE,
1973 				&ring->data[i].map);
1974 		if (error) {
1975 			if_printf(&sc->arpcom.ac_if,
1976 				  "could not create %dth TX buf DMA map\n", i);
1977 			goto fail;
1978 		}
1979 	}
1980 
1981 	return 0;
1982 fail:
1983 	for (j = 0; j < i; ++j)
1984 		bus_dmamap_destroy(ring->data_tag, ring->data[i].map);
1985 	bus_dma_tag_destroy(ring->data_tag);
1986 	ring->data_tag = NULL;
1987 	return error;
1988 }
1989 
1990 static void
1991 nfe_reset_tx_ring(struct nfe_softc *sc, struct nfe_tx_ring *ring)
1992 {
1993 	int i;
1994 
1995 	for (i = 0; i < sc->sc_tx_ring_count; i++) {
1996 		struct nfe_tx_data *data = &ring->data[i];
1997 
1998 		if (sc->sc_caps & NFE_40BIT_ADDR)
1999 			ring->desc64[i].flags = 0;
2000 		else
2001 			ring->desc32[i].flags = 0;
2002 
2003 		if (data->m != NULL) {
2004 			bus_dmamap_unload(ring->data_tag, data->map);
2005 			m_freem(data->m);
2006 			data->m = NULL;
2007 		}
2008 	}
2009 
2010 	ring->queued = 0;
2011 	ring->cur = ring->next = 0;
2012 }
2013 
2014 static int
2015 nfe_init_tx_ring(struct nfe_softc *sc __unused,
2016 		 struct nfe_tx_ring *ring __unused)
2017 {
2018 	return 0;
2019 }
2020 
2021 static void
2022 nfe_free_tx_ring(struct nfe_softc *sc, struct nfe_tx_ring *ring)
2023 {
2024 	if (ring->data_tag != NULL) {
2025 		struct nfe_tx_data *data;
2026 		int i;
2027 
2028 		for (i = 0; i < sc->sc_tx_ring_count; ++i) {
2029 			data = &ring->data[i];
2030 
2031 			if (data->m != NULL) {
2032 				bus_dmamap_unload(ring->data_tag, data->map);
2033 				m_freem(data->m);
2034 			}
2035 			bus_dmamap_destroy(ring->data_tag, data->map);
2036 		}
2037 
2038 		bus_dma_tag_destroy(ring->data_tag);
2039 	}
2040 
2041 	if (ring->data != NULL)
2042 		kfree(ring->data, M_DEVBUF);
2043 
2044 	if (ring->tag != NULL) {
2045 		void *desc;
2046 
2047 		if (sc->sc_caps & NFE_40BIT_ADDR)
2048 			desc = ring->desc64;
2049 		else
2050 			desc = ring->desc32;
2051 
2052 		bus_dmamap_unload(ring->tag, ring->map);
2053 		bus_dmamem_free(ring->tag, desc, ring->map);
2054 		bus_dma_tag_destroy(ring->tag);
2055 	}
2056 }
2057 
2058 static int
2059 nfe_ifmedia_upd(struct ifnet *ifp)
2060 {
2061 	struct nfe_softc *sc = ifp->if_softc;
2062 	struct mii_data *mii = device_get_softc(sc->sc_miibus);
2063 
2064 	ASSERT_SERIALIZED(ifp->if_serializer);
2065 
2066 	if (mii->mii_instance != 0) {
2067 		struct mii_softc *miisc;
2068 
2069 		LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
2070 			mii_phy_reset(miisc);
2071 	}
2072 	mii_mediachg(mii);
2073 
2074 	return 0;
2075 }
2076 
2077 static void
2078 nfe_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2079 {
2080 	struct nfe_softc *sc = ifp->if_softc;
2081 	struct mii_data *mii = device_get_softc(sc->sc_miibus);
2082 
2083 	ASSERT_SERIALIZED(ifp->if_serializer);
2084 
2085 	mii_pollstat(mii);
2086 	ifmr->ifm_status = mii->mii_media_status;
2087 	ifmr->ifm_active = mii->mii_media_active;
2088 }
2089 
2090 static void
2091 nfe_setmulti(struct nfe_softc *sc)
2092 {
2093 	struct ifnet *ifp = &sc->arpcom.ac_if;
2094 	struct ifmultiaddr *ifma;
2095 	uint8_t addr[ETHER_ADDR_LEN], mask[ETHER_ADDR_LEN];
2096 	uint32_t filter = NFE_RXFILTER_MAGIC;
2097 	int i;
2098 
2099 	if ((ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) != 0) {
2100 		bzero(addr, ETHER_ADDR_LEN);
2101 		bzero(mask, ETHER_ADDR_LEN);
2102 		goto done;
2103 	}
2104 
2105 	bcopy(etherbroadcastaddr, addr, ETHER_ADDR_LEN);
2106 	bcopy(etherbroadcastaddr, mask, ETHER_ADDR_LEN);
2107 
2108 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2109 		caddr_t maddr;
2110 
2111 		if (ifma->ifma_addr->sa_family != AF_LINK)
2112 			continue;
2113 
2114 		maddr = LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
2115 		for (i = 0; i < ETHER_ADDR_LEN; i++) {
2116 			addr[i] &= maddr[i];
2117 			mask[i] &= ~maddr[i];
2118 		}
2119 	}
2120 
2121 	for (i = 0; i < ETHER_ADDR_LEN; i++)
2122 		mask[i] |= addr[i];
2123 
2124 done:
2125 	addr[0] |= 0x01;	/* make sure multicast bit is set */
2126 
2127 	NFE_WRITE(sc, NFE_MULTIADDR_HI,
2128 	    addr[3] << 24 | addr[2] << 16 | addr[1] << 8 | addr[0]);
2129 	NFE_WRITE(sc, NFE_MULTIADDR_LO,
2130 	    addr[5] <<  8 | addr[4]);
2131 	NFE_WRITE(sc, NFE_MULTIMASK_HI,
2132 	    mask[3] << 24 | mask[2] << 16 | mask[1] << 8 | mask[0]);
2133 	NFE_WRITE(sc, NFE_MULTIMASK_LO,
2134 	    mask[5] <<  8 | mask[4]);
2135 
2136 	filter |= (ifp->if_flags & IFF_PROMISC) ? NFE_PROMISC : NFE_U2M;
2137 	NFE_WRITE(sc, NFE_RXFILTER, filter);
2138 }
2139 
2140 static void
2141 nfe_get_macaddr(struct nfe_softc *sc, uint8_t *addr)
2142 {
2143 	uint32_t lo, hi;
2144 
2145 	lo = NFE_READ(sc, NFE_MACADDR_LO);
2146 	hi = NFE_READ(sc, NFE_MACADDR_HI);
2147 	if (sc->sc_caps & NFE_FIX_EADDR) {
2148 		addr[0] = (lo >> 8) & 0xff;
2149 		addr[1] = (lo & 0xff);
2150 
2151 		addr[2] = (hi >> 24) & 0xff;
2152 		addr[3] = (hi >> 16) & 0xff;
2153 		addr[4] = (hi >>  8) & 0xff;
2154 		addr[5] = (hi & 0xff);
2155 	} else {
2156 		addr[0] = (hi & 0xff);
2157 		addr[1] = (hi >>  8) & 0xff;
2158 		addr[2] = (hi >> 16) & 0xff;
2159 		addr[3] = (hi >> 24) & 0xff;
2160 
2161 		addr[4] = (lo & 0xff);
2162 		addr[5] = (lo >>  8) & 0xff;
2163 	}
2164 }
2165 
2166 static void
2167 nfe_set_macaddr(struct nfe_softc *sc, const uint8_t *addr)
2168 {
2169 	NFE_WRITE(sc, NFE_MACADDR_LO,
2170 	    addr[5] <<  8 | addr[4]);
2171 	NFE_WRITE(sc, NFE_MACADDR_HI,
2172 	    addr[3] << 24 | addr[2] << 16 | addr[1] << 8 | addr[0]);
2173 }
2174 
2175 static void
2176 nfe_tick(void *arg)
2177 {
2178 	struct nfe_softc *sc = arg;
2179 	struct ifnet *ifp = &sc->arpcom.ac_if;
2180 	struct mii_data *mii = device_get_softc(sc->sc_miibus);
2181 
2182 	lwkt_serialize_enter(ifp->if_serializer);
2183 
2184 	mii_tick(mii);
2185 	callout_reset(&sc->sc_tick_ch, hz, nfe_tick, sc);
2186 
2187 	lwkt_serialize_exit(ifp->if_serializer);
2188 }
2189 
2190 static int
2191 nfe_newbuf_std(struct nfe_softc *sc, struct nfe_rx_ring *ring, int idx,
2192 	       int wait)
2193 {
2194 	struct nfe_rx_data *data = &ring->data[idx];
2195 	bus_dma_segment_t seg;
2196 	bus_dmamap_t map;
2197 	struct mbuf *m;
2198 	int nsegs, error;
2199 
2200 	m = m_getcl(wait ? MB_WAIT : MB_DONTWAIT, MT_DATA, M_PKTHDR);
2201 	if (m == NULL)
2202 		return ENOBUFS;
2203 	m->m_len = m->m_pkthdr.len = MCLBYTES;
2204 
2205 	/*
2206 	 * Aligning the payload improves access times.
2207 	 */
2208 	if (sc->sc_caps & NFE_WORDALIGN)
2209 		m_adj(m, ETHER_ALIGN);
2210 
2211 	error = bus_dmamap_load_mbuf_segment(ring->data_tag, ring->data_tmpmap,
2212 			m, &seg, 1, &nsegs, BUS_DMA_NOWAIT);
2213 	if (error) {
2214 		m_freem(m);
2215 		if (wait) {
2216 			if_printf(&sc->arpcom.ac_if,
2217 				  "could map RX mbuf %d\n", error);
2218 		}
2219 		return error;
2220 	}
2221 
2222 	if (data->m != NULL) {
2223 		/* Sync and unload originally mapped mbuf */
2224 		bus_dmamap_sync(ring->data_tag, data->map,
2225 				BUS_DMASYNC_POSTREAD);
2226 		bus_dmamap_unload(ring->data_tag, data->map);
2227 	}
2228 
2229 	/* Swap this DMA map with tmp DMA map */
2230 	map = data->map;
2231 	data->map = ring->data_tmpmap;
2232 	ring->data_tmpmap = map;
2233 
2234 	/* Caller is assumed to have collected the old mbuf */
2235 	data->m = m;
2236 
2237 	nfe_set_paddr_rxdesc(sc, ring, idx, seg.ds_addr);
2238 	return 0;
2239 }
2240 
2241 static int
2242 nfe_newbuf_jumbo(struct nfe_softc *sc, struct nfe_rx_ring *ring, int idx,
2243 		 int wait)
2244 {
2245 	struct nfe_rx_data *data = &ring->data[idx];
2246 	struct nfe_jbuf *jbuf;
2247 	struct mbuf *m;
2248 
2249 	MGETHDR(m, wait ? MB_WAIT : MB_DONTWAIT, MT_DATA);
2250 	if (m == NULL)
2251 		return ENOBUFS;
2252 
2253 	jbuf = nfe_jalloc(sc);
2254 	if (jbuf == NULL) {
2255 		m_freem(m);
2256 		if_printf(&sc->arpcom.ac_if, "jumbo allocation failed "
2257 		    "-- packet dropped!\n");
2258 		return ENOBUFS;
2259 	}
2260 
2261 	m->m_ext.ext_arg = jbuf;
2262 	m->m_ext.ext_buf = jbuf->buf;
2263 	m->m_ext.ext_free = nfe_jfree;
2264 	m->m_ext.ext_ref = nfe_jref;
2265 	m->m_ext.ext_size = NFE_JBYTES;
2266 
2267 	m->m_data = m->m_ext.ext_buf;
2268 	m->m_flags |= M_EXT;
2269 	m->m_len = m->m_pkthdr.len = m->m_ext.ext_size;
2270 
2271 	/*
2272 	 * Aligning the payload improves access times.
2273 	 */
2274 	if (sc->sc_caps & NFE_WORDALIGN)
2275 		m_adj(m, ETHER_ALIGN);
2276 
2277 	/* Caller is assumed to have collected the old mbuf */
2278 	data->m = m;
2279 
2280 	nfe_set_paddr_rxdesc(sc, ring, idx, jbuf->physaddr);
2281 	return 0;
2282 }
2283 
2284 static void
2285 nfe_set_paddr_rxdesc(struct nfe_softc *sc, struct nfe_rx_ring *ring, int idx,
2286 		     bus_addr_t physaddr)
2287 {
2288 	if (sc->sc_caps & NFE_40BIT_ADDR) {
2289 		struct nfe_desc64 *desc64 = &ring->desc64[idx];
2290 
2291 		desc64->physaddr[0] = htole32(NFE_ADDR_HI(physaddr));
2292 		desc64->physaddr[1] = htole32(NFE_ADDR_LO(physaddr));
2293 	} else {
2294 		struct nfe_desc32 *desc32 = &ring->desc32[idx];
2295 
2296 		desc32->physaddr = htole32(physaddr);
2297 	}
2298 }
2299 
2300 static void
2301 nfe_set_ready_rxdesc(struct nfe_softc *sc, struct nfe_rx_ring *ring, int idx)
2302 {
2303 	if (sc->sc_caps & NFE_40BIT_ADDR) {
2304 		struct nfe_desc64 *desc64 = &ring->desc64[idx];
2305 
2306 		desc64->length = htole16(ring->bufsz);
2307 		desc64->flags = htole16(NFE_RX_READY);
2308 	} else {
2309 		struct nfe_desc32 *desc32 = &ring->desc32[idx];
2310 
2311 		desc32->length = htole16(ring->bufsz);
2312 		desc32->flags = htole16(NFE_RX_READY);
2313 	}
2314 }
2315 
2316 static int
2317 nfe_sysctl_imtime(SYSCTL_HANDLER_ARGS)
2318 {
2319 	struct nfe_softc *sc = arg1;
2320 	struct ifnet *ifp = &sc->arpcom.ac_if;
2321 	uint32_t flags;
2322 	int error, v;
2323 
2324 	lwkt_serialize_enter(ifp->if_serializer);
2325 
2326 	flags = sc->sc_flags & ~NFE_F_DYN_IM;
2327 	v = sc->sc_imtime;
2328 	if (sc->sc_flags & NFE_F_DYN_IM)
2329 		v = -v;
2330 
2331 	error = sysctl_handle_int(oidp, &v, 0, req);
2332 	if (error || req->newptr == NULL)
2333 		goto back;
2334 
2335 	if (v < 0) {
2336 		flags |= NFE_F_DYN_IM;
2337 		v = -v;
2338 	}
2339 
2340 	if (v != sc->sc_imtime || (flags ^ sc->sc_flags)) {
2341 		if (NFE_IMTIME(v) == 0)
2342 			v = 0;
2343 		sc->sc_imtime = v;
2344 		sc->sc_flags = flags;
2345 		sc->sc_irq_enable = NFE_IRQ_ENABLE(sc);
2346 
2347 		if ((ifp->if_flags & (IFF_POLLING | IFF_RUNNING))
2348 		    == IFF_RUNNING) {
2349 			nfe_enable_intrs(sc);
2350 		}
2351 	}
2352 back:
2353 	lwkt_serialize_exit(ifp->if_serializer);
2354 	return error;
2355 }
2356 
2357 static void
2358 nfe_powerup(device_t dev)
2359 {
2360 	struct nfe_softc *sc = device_get_softc(dev);
2361 	uint32_t pwr_state;
2362 	uint16_t did;
2363 
2364 	/*
2365 	 * Bring MAC and PHY out of low power state
2366 	 */
2367 
2368 	pwr_state = NFE_READ(sc, NFE_PWR_STATE2) & ~NFE_PWRUP_MASK;
2369 
2370 	did = pci_get_device(dev);
2371 	if ((did == PCI_PRODUCT_NVIDIA_MCP51_LAN1 ||
2372 	     did == PCI_PRODUCT_NVIDIA_MCP51_LAN2) &&
2373 	    pci_get_revid(dev) >= 0xa3)
2374 		pwr_state |= NFE_PWRUP_REV_A3;
2375 
2376 	NFE_WRITE(sc, NFE_PWR_STATE2, pwr_state);
2377 }
2378 
2379 static void
2380 nfe_mac_reset(struct nfe_softc *sc)
2381 {
2382 	uint32_t rxtxctl = sc->rxtxctl_desc | NFE_RXTX_BIT2;
2383 	uint32_t macaddr_hi, macaddr_lo, tx_poll;
2384 
2385 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_RESET | rxtxctl);
2386 
2387 	/* Save several registers for later restoration */
2388 	macaddr_hi = NFE_READ(sc, NFE_MACADDR_HI);
2389 	macaddr_lo = NFE_READ(sc, NFE_MACADDR_LO);
2390 	tx_poll = NFE_READ(sc, NFE_TX_POLL);
2391 
2392 	NFE_WRITE(sc, NFE_MAC_RESET, NFE_RESET_ASSERT);
2393 	DELAY(100);
2394 
2395 	NFE_WRITE(sc, NFE_MAC_RESET, 0);
2396 	DELAY(100);
2397 
2398 	/* Restore saved registers */
2399 	NFE_WRITE(sc, NFE_MACADDR_HI, macaddr_hi);
2400 	NFE_WRITE(sc, NFE_MACADDR_LO, macaddr_lo);
2401 	NFE_WRITE(sc, NFE_TX_POLL, tx_poll);
2402 
2403 	NFE_WRITE(sc, NFE_RXTX_CTL, rxtxctl);
2404 }
2405 
2406 static void
2407 nfe_enable_intrs(struct nfe_softc *sc)
2408 {
2409 	/*
2410 	 * NFE_IMTIMER generates a periodic interrupt via NFE_IRQ_TIMER.
2411 	 * It is unclear how wide the timer is.  Base programming does
2412 	 * not seem to effect NFE_IRQ_TX_DONE or NFE_IRQ_RX_DONE so
2413 	 * we don't get any interrupt moderation.  TX moderation is
2414 	 * possible by using the timer interrupt instead of TX_DONE.
2415 	 *
2416 	 * It is unclear whether there are other bits that can be
2417 	 * set to make the NFE device actually do interrupt moderation
2418 	 * on the RX side.
2419 	 *
2420 	 * For now set a 128uS interval as a placemark, but don't use
2421 	 * the timer.
2422 	 */
2423 	if (sc->sc_imtime == 0)
2424 		NFE_WRITE(sc, NFE_IMTIMER, NFE_IMTIME_DEFAULT);
2425 	else
2426 		NFE_WRITE(sc, NFE_IMTIMER, NFE_IMTIME(sc->sc_imtime));
2427 
2428 	/* Enable interrupts */
2429 	NFE_WRITE(sc, NFE_IRQ_MASK, sc->sc_irq_enable);
2430 
2431 	if (sc->sc_irq_enable & NFE_IRQ_TIMER)
2432 		sc->sc_flags |= NFE_F_IRQ_TIMER;
2433 	else
2434 		sc->sc_flags &= ~NFE_F_IRQ_TIMER;
2435 }
2436 
2437 #ifdef DEVICE_POLLING
2438 static void
2439 nfe_disable_intrs(struct nfe_softc *sc)
2440 {
2441 	/* Disable interrupts */
2442 	NFE_WRITE(sc, NFE_IRQ_MASK, 0);
2443 	sc->sc_flags &= ~NFE_F_IRQ_TIMER;
2444 }
2445 #endif
2446