xref: /dragonfly/sys/dev/netif/nfe/if_nfe.c (revision abf903a5)
1 /*	$OpenBSD: if_nfe.c,v 1.63 2006/06/17 18:00:43 brad Exp $	*/
2 
3 /*
4  * Copyright (c) 2006 The DragonFly Project.  All rights reserved.
5  *
6  * This code is derived from software contributed to The DragonFly Project
7  * by Sepherosa Ziehau <sepherosa@gmail.com> and
8  * Matthew Dillon <dillon@apollo.backplane.com>
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  *
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in
18  *    the documentation and/or other materials provided with the
19  *    distribution.
20  * 3. Neither the name of The DragonFly Project nor the names of its
21  *    contributors may be used to endorse or promote products derived
22  *    from this software without specific, prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
25  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
26  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
27  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
28  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
29  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
30  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
31  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
32  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
33  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
34  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  */
37 
38 /*
39  * Copyright (c) 2006 Damien Bergamini <damien.bergamini@free.fr>
40  * Copyright (c) 2005, 2006 Jonathan Gray <jsg@openbsd.org>
41  *
42  * Permission to use, copy, modify, and distribute this software for any
43  * purpose with or without fee is hereby granted, provided that the above
44  * copyright notice and this permission notice appear in all copies.
45  *
46  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
47  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
48  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
49  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
50  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
51  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
52  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
53  */
54 
55 /* Driver for NVIDIA nForce MCP Fast Ethernet and Gigabit Ethernet */
56 
57 #include "opt_ifpoll.h"
58 
59 #include <sys/param.h>
60 #include <sys/endian.h>
61 #include <sys/kernel.h>
62 #include <sys/malloc.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/if_poll.h>
79 #include <net/ifq_var.h>
80 #include <net/if_types.h>
81 #include <net/if_var.h>
82 #include <net/vlan/if_vlan_var.h>
83 #include <net/vlan/if_vlan_ether.h>
84 
85 #include <bus/pci/pcireg.h>
86 #include <bus/pci/pcivar.h>
87 #include "pcidevs.h"
88 
89 #include <dev/netif/mii_layer/mii.h>
90 #include <dev/netif/mii_layer/miivar.h>
91 
92 #include "miibus_if.h"
93 
94 #include <dev/netif/nfe/if_nfereg.h>
95 #include <dev/netif/nfe/if_nfevar.h>
96 
97 #define NFE_CSUM
98 #define NFE_CSUM_FEATURES	(CSUM_IP | CSUM_TCP | CSUM_UDP)
99 
100 static int	nfe_probe(device_t);
101 static int	nfe_attach(device_t);
102 static int	nfe_detach(device_t);
103 static void	nfe_shutdown(device_t);
104 static int	nfe_resume(device_t);
105 static int	nfe_suspend(device_t);
106 
107 static int	nfe_miibus_readreg(device_t, int, int);
108 static void	nfe_miibus_writereg(device_t, int, int, int);
109 static void	nfe_miibus_statchg(device_t);
110 
111 #ifdef IFPOLL_ENABLE
112 static void	nfe_npoll(struct ifnet *, struct ifpoll_info *);
113 static void	nfe_npoll_compat(struct ifnet *, void *, int);
114 static void	nfe_disable_intrs(struct nfe_softc *);
115 #endif
116 static void	nfe_intr(void *);
117 static int	nfe_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
118 static int	nfe_rxeof(struct nfe_softc *);
119 static int	nfe_txeof(struct nfe_softc *, int);
120 static int	nfe_encap(struct nfe_softc *, struct nfe_tx_ring *,
121 			  struct mbuf *);
122 static void	nfe_start(struct ifnet *, struct ifaltq_subque *);
123 static void	nfe_watchdog(struct ifnet *);
124 static void	nfe_init(void *);
125 static void	nfe_stop(struct nfe_softc *);
126 static struct nfe_jbuf *nfe_jalloc(struct nfe_softc *);
127 static void	nfe_jfree(void *);
128 static void	nfe_jref(void *);
129 static int	nfe_jpool_alloc(struct nfe_softc *, struct nfe_rx_ring *);
130 static void	nfe_jpool_free(struct nfe_softc *, struct nfe_rx_ring *);
131 static int	nfe_alloc_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
132 static void	nfe_reset_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
133 static int	nfe_init_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
134 static void	nfe_free_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
135 static int	nfe_alloc_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
136 static void	nfe_reset_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
137 static int	nfe_init_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
138 static void	nfe_free_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
139 static int	nfe_ifmedia_upd(struct ifnet *);
140 static void	nfe_ifmedia_sts(struct ifnet *, struct ifmediareq *);
141 static void	nfe_setmulti(struct nfe_softc *);
142 static void	nfe_get_macaddr(struct nfe_softc *, uint8_t *);
143 static void	nfe_set_macaddr(struct nfe_softc *, const uint8_t *);
144 static void	nfe_powerup(device_t);
145 static void	nfe_mac_reset(struct nfe_softc *);
146 static void	nfe_tick(void *);
147 static void	nfe_set_paddr_rxdesc(struct nfe_softc *, struct nfe_rx_ring *,
148 				     int, bus_addr_t);
149 static void	nfe_set_ready_rxdesc(struct nfe_softc *, struct nfe_rx_ring *,
150 				     int);
151 static int	nfe_newbuf_std(struct nfe_softc *, struct nfe_rx_ring *, int,
152 			       int);
153 static int	nfe_newbuf_jumbo(struct nfe_softc *, struct nfe_rx_ring *, int,
154 				 int);
155 static void	nfe_enable_intrs(struct nfe_softc *);
156 
157 static int	nfe_sysctl_imtime(SYSCTL_HANDLER_ARGS);
158 
159 #define NFE_DEBUG
160 #ifdef NFE_DEBUG
161 
162 static int	nfe_debug = 0;
163 static int	nfe_rx_ring_count = NFE_RX_RING_DEF_COUNT;
164 static int	nfe_tx_ring_count = NFE_TX_RING_DEF_COUNT;
165 /*
166  * hw timer simulated interrupt moderation @4000Hz.  Negative values
167  * disable the timer when the discrete interrupt rate falls below
168  * the moderation rate.
169  *
170  * XXX 8000Hz might be better but if the interrupt is shared it can
171  *     blow out the cpu.
172  */
173 static int	nfe_imtime = -250;	/* uS */
174 
175 TUNABLE_INT("hw.nfe.rx_ring_count", &nfe_rx_ring_count);
176 TUNABLE_INT("hw.nfe.tx_ring_count", &nfe_tx_ring_count);
177 TUNABLE_INT("hw.nfe.imtimer", &nfe_imtime);
178 TUNABLE_INT("hw.nfe.debug", &nfe_debug);
179 
180 #define DPRINTF(sc, fmt, ...) do {		\
181 	if ((sc)->sc_debug) {			\
182 		if_printf(&(sc)->arpcom.ac_if,	\
183 			  fmt, __VA_ARGS__);	\
184 	}					\
185 } while (0)
186 
187 #define DPRINTFN(sc, lv, fmt, ...) do {		\
188 	if ((sc)->sc_debug >= (lv)) {		\
189 		if_printf(&(sc)->arpcom.ac_if,	\
190 			  fmt, __VA_ARGS__);	\
191 	}					\
192 } while (0)
193 
194 #else	/* !NFE_DEBUG */
195 
196 #define DPRINTF(sc, fmt, ...)
197 #define DPRINTFN(sc, lv, fmt, ...)
198 
199 #endif	/* NFE_DEBUG */
200 
201 static const struct nfe_dev {
202 	uint16_t	vid;
203 	uint16_t	did;
204 	const char	*desc;
205 } nfe_devices[] = {
206 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE_LAN,
207 	  "NVIDIA nForce Fast Ethernet" },
208 
209 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE2_LAN,
210 	  "NVIDIA nForce2 Fast Ethernet" },
211 
212 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN1,
213 	  "NVIDIA nForce3 Gigabit Ethernet" },
214 
215 	/* XXX TGEN the next chip can also be found in the nForce2 Ultra 400Gb
216 	   chipset, and possibly also the 400R; it might be both nForce2- and
217 	   nForce3-based boards can use the same MCPs (= southbridges) */
218 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN2,
219 	  "NVIDIA nForce3 Gigabit Ethernet" },
220 
221 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN3,
222 	  "NVIDIA nForce3 Gigabit Ethernet" },
223 
224 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN4,
225 	  "NVIDIA nForce3 Gigabit Ethernet" },
226 
227 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN5,
228 	  "NVIDIA nForce3 Gigabit Ethernet" },
229 
230 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_CK804_LAN1,
231 	  "NVIDIA CK804 Gigabit Ethernet" },
232 
233 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_CK804_LAN2,
234 	  "NVIDIA CK804 Gigabit Ethernet" },
235 
236 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP04_LAN1,
237 	  "NVIDIA MCP04 Gigabit Ethernet" },
238 
239 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP04_LAN2,
240 	  "NVIDIA MCP04 Gigabit Ethernet" },
241 
242 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP51_LAN1,
243 	  "NVIDIA MCP51 Gigabit Ethernet" },
244 
245 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP51_LAN2,
246 	  "NVIDIA MCP51 Gigabit Ethernet" },
247 
248 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP55_LAN1,
249 	  "NVIDIA MCP55 Gigabit Ethernet" },
250 
251 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP55_LAN2,
252 	  "NVIDIA MCP55 Gigabit Ethernet" },
253 
254 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN1,
255 	  "NVIDIA MCP61 Gigabit Ethernet" },
256 
257 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN2,
258 	  "NVIDIA MCP61 Gigabit Ethernet" },
259 
260 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN3,
261 	  "NVIDIA MCP61 Gigabit Ethernet" },
262 
263 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN4,
264 	  "NVIDIA MCP61 Gigabit Ethernet" },
265 
266 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN1,
267 	  "NVIDIA MCP65 Gigabit Ethernet" },
268 
269 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN2,
270 	  "NVIDIA MCP65 Gigabit Ethernet" },
271 
272 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN3,
273 	  "NVIDIA MCP65 Gigabit Ethernet" },
274 
275 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN4,
276 	  "NVIDIA MCP65 Gigabit Ethernet" },
277 
278 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN1,
279 	  "NVIDIA MCP67 Gigabit Ethernet" },
280 
281 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN2,
282 	  "NVIDIA MCP67 Gigabit Ethernet" },
283 
284 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN3,
285 	  "NVIDIA MCP67 Gigabit Ethernet" },
286 
287 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN4,
288 	  "NVIDIA MCP67 Gigabit Ethernet" },
289 
290 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP73_LAN1,
291 	  "NVIDIA MCP73 Gigabit Ethernet" },
292 
293 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP73_LAN2,
294 	  "NVIDIA MCP73 Gigabit Ethernet" },
295 
296 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP73_LAN3,
297 	  "NVIDIA MCP73 Gigabit Ethernet" },
298 
299 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP73_LAN4,
300 	  "NVIDIA MCP73 Gigabit Ethernet" },
301 
302 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP77_LAN1,
303 	  "NVIDIA MCP77 Gigabit Ethernet" },
304 
305 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP77_LAN2,
306 	  "NVIDIA MCP77 Gigabit Ethernet" },
307 
308 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP77_LAN3,
309 	  "NVIDIA MCP77 Gigabit Ethernet" },
310 
311 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP77_LAN4,
312 	  "NVIDIA MCP77 Gigabit Ethernet" },
313 
314 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP79_LAN1,
315 	  "NVIDIA MCP79 Gigabit Ethernet" },
316 
317 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP79_LAN2,
318 	  "NVIDIA MCP79 Gigabit Ethernet" },
319 
320 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP79_LAN3,
321 	  "NVIDIA MCP79 Gigabit Ethernet" },
322 
323 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP79_LAN4,
324 	  "NVIDIA MCP79 Gigabit Ethernet" },
325 
326 	{ 0, 0, NULL }
327 };
328 
329 static device_method_t nfe_methods[] = {
330 	/* Device interface */
331 	DEVMETHOD(device_probe,		nfe_probe),
332 	DEVMETHOD(device_attach,	nfe_attach),
333 	DEVMETHOD(device_detach,	nfe_detach),
334 	DEVMETHOD(device_suspend,	nfe_suspend),
335 	DEVMETHOD(device_resume,	nfe_resume),
336 	DEVMETHOD(device_shutdown,	nfe_shutdown),
337 
338 	/* Bus interface */
339 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
340 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
341 
342 	/* MII interface */
343 	DEVMETHOD(miibus_readreg,	nfe_miibus_readreg),
344 	DEVMETHOD(miibus_writereg,	nfe_miibus_writereg),
345 	DEVMETHOD(miibus_statchg,	nfe_miibus_statchg),
346 
347 	DEVMETHOD_END
348 };
349 
350 static driver_t nfe_driver = {
351 	"nfe",
352 	nfe_methods,
353 	sizeof(struct nfe_softc)
354 };
355 
356 static devclass_t	nfe_devclass;
357 
358 DECLARE_DUMMY_MODULE(if_nfe);
359 MODULE_DEPEND(if_nfe, miibus, 1, 1, 1);
360 DRIVER_MODULE(if_nfe, pci, nfe_driver, nfe_devclass, NULL, NULL);
361 DRIVER_MODULE(miibus, nfe, miibus_driver, miibus_devclass, NULL, NULL);
362 
363 /*
364  * NOTE: NFE_WORDALIGN support is guesswork right now.
365  */
366 static int
367 nfe_probe(device_t dev)
368 {
369 	const struct nfe_dev *n;
370 	uint16_t vid, did;
371 
372 	vid = pci_get_vendor(dev);
373 	did = pci_get_device(dev);
374 	for (n = nfe_devices; n->desc != NULL; ++n) {
375 		if (vid == n->vid && did == n->did) {
376 			struct nfe_softc *sc = device_get_softc(dev);
377 
378 			switch (did) {
379 			case PCI_PRODUCT_NVIDIA_NFORCE_LAN:
380 			case PCI_PRODUCT_NVIDIA_NFORCE2_LAN:
381 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN1:
382 				sc->sc_caps = NFE_NO_PWRCTL |
383 					      NFE_FIX_EADDR;
384 				break;
385 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN2:
386 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN3:
387 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN4:
388 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN5:
389 				sc->sc_caps = NFE_JUMBO_SUP |
390 					      NFE_HW_CSUM |
391 					      NFE_NO_PWRCTL |
392 					      NFE_FIX_EADDR;
393 				break;
394 			case PCI_PRODUCT_NVIDIA_MCP51_LAN1:
395 			case PCI_PRODUCT_NVIDIA_MCP51_LAN2:
396 				sc->sc_caps = NFE_FIX_EADDR;
397 				/* FALL THROUGH */
398 			case PCI_PRODUCT_NVIDIA_MCP61_LAN1:
399 			case PCI_PRODUCT_NVIDIA_MCP61_LAN2:
400 			case PCI_PRODUCT_NVIDIA_MCP61_LAN3:
401 			case PCI_PRODUCT_NVIDIA_MCP61_LAN4:
402 			case PCI_PRODUCT_NVIDIA_MCP67_LAN1:
403 			case PCI_PRODUCT_NVIDIA_MCP67_LAN2:
404 			case PCI_PRODUCT_NVIDIA_MCP67_LAN3:
405 			case PCI_PRODUCT_NVIDIA_MCP67_LAN4:
406 			case PCI_PRODUCT_NVIDIA_MCP73_LAN1:
407 			case PCI_PRODUCT_NVIDIA_MCP73_LAN2:
408 			case PCI_PRODUCT_NVIDIA_MCP73_LAN3:
409 			case PCI_PRODUCT_NVIDIA_MCP73_LAN4:
410 				sc->sc_caps |= NFE_40BIT_ADDR;
411 				break;
412 			case PCI_PRODUCT_NVIDIA_CK804_LAN1:
413 			case PCI_PRODUCT_NVIDIA_CK804_LAN2:
414 			case PCI_PRODUCT_NVIDIA_MCP04_LAN1:
415 			case PCI_PRODUCT_NVIDIA_MCP04_LAN2:
416 				sc->sc_caps = NFE_JUMBO_SUP |
417 					      NFE_40BIT_ADDR |
418 					      NFE_HW_CSUM |
419 					      NFE_NO_PWRCTL |
420 					      NFE_FIX_EADDR;
421 				break;
422 			case PCI_PRODUCT_NVIDIA_MCP65_LAN1:
423 			case PCI_PRODUCT_NVIDIA_MCP65_LAN2:
424 			case PCI_PRODUCT_NVIDIA_MCP65_LAN3:
425 			case PCI_PRODUCT_NVIDIA_MCP65_LAN4:
426 				sc->sc_caps = NFE_JUMBO_SUP |
427 					      NFE_40BIT_ADDR;
428 				break;
429 			case PCI_PRODUCT_NVIDIA_MCP55_LAN1:
430 			case PCI_PRODUCT_NVIDIA_MCP55_LAN2:
431 				sc->sc_caps = NFE_JUMBO_SUP |
432 					      NFE_40BIT_ADDR |
433 					      NFE_HW_CSUM |
434 					      NFE_HW_VLAN |
435 					      NFE_FIX_EADDR;
436 				break;
437 			case PCI_PRODUCT_NVIDIA_MCP77_LAN1:
438 			case PCI_PRODUCT_NVIDIA_MCP77_LAN2:
439 			case PCI_PRODUCT_NVIDIA_MCP77_LAN3:
440 			case PCI_PRODUCT_NVIDIA_MCP77_LAN4:
441 			case PCI_PRODUCT_NVIDIA_MCP79_LAN1:
442 			case PCI_PRODUCT_NVIDIA_MCP79_LAN2:
443 			case PCI_PRODUCT_NVIDIA_MCP79_LAN3:
444 			case PCI_PRODUCT_NVIDIA_MCP79_LAN4:
445 				sc->sc_caps = NFE_40BIT_ADDR |
446 					      NFE_HW_CSUM |
447 					      NFE_WORDALIGN;
448 				break;
449 			}
450 
451 			device_set_desc(dev, n->desc);
452 			device_set_async_attach(dev, TRUE);
453 			return 0;
454 		}
455 	}
456 	return ENXIO;
457 }
458 
459 static int
460 nfe_attach(device_t dev)
461 {
462 	struct nfe_softc *sc = device_get_softc(dev);
463 	struct ifnet *ifp = &sc->arpcom.ac_if;
464 	struct sysctl_ctx_list *ctx;
465 	struct sysctl_oid *tree;
466 	uint8_t eaddr[ETHER_ADDR_LEN];
467 	bus_addr_t lowaddr;
468 	int error;
469 
470 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
471 	lwkt_serialize_init(&sc->sc_jbuf_serializer);
472 
473 	/*
474 	 * Initialize sysctl variables
475 	 */
476 	sc->sc_rx_ring_count = nfe_rx_ring_count;
477 	sc->sc_tx_ring_count = nfe_tx_ring_count;
478 	sc->sc_debug = nfe_debug;
479 	if (nfe_imtime < 0) {
480 		sc->sc_flags |= NFE_F_DYN_IM;
481 		sc->sc_imtime = -nfe_imtime;
482 	} else {
483 		sc->sc_imtime = nfe_imtime;
484 	}
485 	sc->sc_irq_enable = NFE_IRQ_ENABLE(sc);
486 
487 	sc->sc_mem_rid = PCIR_BAR(0);
488 
489 	if (sc->sc_caps & NFE_40BIT_ADDR)
490 		sc->rxtxctl_desc = NFE_RXTX_DESC_V3;
491 	else if (sc->sc_caps & NFE_JUMBO_SUP)
492 		sc->rxtxctl_desc = NFE_RXTX_DESC_V2;
493 
494 #ifndef BURN_BRIDGES
495 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
496 		uint32_t mem, irq;
497 
498 		mem = pci_read_config(dev, sc->sc_mem_rid, 4);
499 		irq = pci_read_config(dev, PCIR_INTLINE, 4);
500 
501 		device_printf(dev, "chip is in D%d power mode "
502 		    "-- setting to D0\n", pci_get_powerstate(dev));
503 
504 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
505 
506 		pci_write_config(dev, sc->sc_mem_rid, mem, 4);
507 		pci_write_config(dev, PCIR_INTLINE, irq, 4);
508 	}
509 #endif	/* !BURN_BRIDGE */
510 
511 	/* Enable bus mastering */
512 	pci_enable_busmaster(dev);
513 
514 	/* Allocate IO memory */
515 	sc->sc_mem_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
516 						&sc->sc_mem_rid, RF_ACTIVE);
517 	if (sc->sc_mem_res == NULL) {
518 		device_printf(dev, "could not allocate io memory\n");
519 		return ENXIO;
520 	}
521 	sc->sc_memh = rman_get_bushandle(sc->sc_mem_res);
522 	sc->sc_memt = rman_get_bustag(sc->sc_mem_res);
523 
524 	/* Allocate IRQ */
525 	sc->sc_irq_rid = 0;
526 	sc->sc_irq_res = bus_alloc_resource_any(dev, SYS_RES_IRQ,
527 						&sc->sc_irq_rid,
528 						RF_SHAREABLE | RF_ACTIVE);
529 	if (sc->sc_irq_res == NULL) {
530 		device_printf(dev, "could not allocate irq\n");
531 		error = ENXIO;
532 		goto fail;
533 	}
534 
535 	/* Disable WOL */
536 	NFE_WRITE(sc, NFE_WOL_CTL, 0);
537 
538 	if ((sc->sc_caps & NFE_NO_PWRCTL) == 0)
539 		nfe_powerup(dev);
540 
541 	nfe_get_macaddr(sc, eaddr);
542 
543 	/*
544 	 * Allocate top level DMA tag
545 	 */
546 	if (sc->sc_caps & NFE_40BIT_ADDR)
547 		lowaddr = NFE_BUS_SPACE_MAXADDR;
548 	else
549 		lowaddr = BUS_SPACE_MAXADDR_32BIT;
550 	error = bus_dma_tag_create(NULL,	/* parent */
551 			1, 0,			/* alignment, boundary */
552 			lowaddr,		/* lowaddr */
553 			BUS_SPACE_MAXADDR,	/* highaddr */
554 			NULL, NULL,		/* filter, filterarg */
555 			BUS_SPACE_MAXSIZE_32BIT,/* maxsize */
556 			0,			/* nsegments */
557 			BUS_SPACE_MAXSIZE_32BIT,/* maxsegsize */
558 			0,			/* flags */
559 			&sc->sc_dtag);
560 	if (error) {
561 		device_printf(dev, "could not allocate parent dma tag\n");
562 		goto fail;
563 	}
564 
565 	/*
566 	 * Allocate Tx and Rx rings.
567 	 */
568 	error = nfe_alloc_tx_ring(sc, &sc->txq);
569 	if (error) {
570 		device_printf(dev, "could not allocate Tx ring\n");
571 		goto fail;
572 	}
573 
574 	error = nfe_alloc_rx_ring(sc, &sc->rxq);
575 	if (error) {
576 		device_printf(dev, "could not allocate Rx ring\n");
577 		goto fail;
578 	}
579 
580 	/*
581 	 * Create sysctl tree
582 	 */
583 	ctx = device_get_sysctl_ctx(dev);
584 	tree = device_get_sysctl_tree(dev);
585 	SYSCTL_ADD_PROC(ctx, SYSCTL_CHILDREN(tree),
586 			OID_AUTO, "imtimer", CTLTYPE_INT | CTLFLAG_RW,
587 			sc, 0, nfe_sysctl_imtime, "I",
588 			"Interrupt moderation time (usec).  "
589 			"0 to disable interrupt moderation.");
590 	SYSCTL_ADD_INT(ctx, SYSCTL_CHILDREN(tree), OID_AUTO,
591 		       "rx_ring_count", CTLFLAG_RD, &sc->sc_rx_ring_count,
592 		       0, "RX ring count");
593 	SYSCTL_ADD_INT(ctx, SYSCTL_CHILDREN(tree), OID_AUTO,
594 		       "tx_ring_count", CTLFLAG_RD, &sc->sc_tx_ring_count,
595 		       0, "TX ring count");
596 	SYSCTL_ADD_INT(ctx, SYSCTL_CHILDREN(tree), OID_AUTO,
597 		       "debug", CTLFLAG_RW, &sc->sc_debug,
598 		       0, "control debugging printfs");
599 
600 	error = mii_phy_probe(dev, &sc->sc_miibus, nfe_ifmedia_upd,
601 			      nfe_ifmedia_sts);
602 	if (error) {
603 		device_printf(dev, "MII without any phy\n");
604 		goto fail;
605 	}
606 
607 	ifp->if_softc = sc;
608 	ifp->if_mtu = ETHERMTU;
609 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
610 	ifp->if_ioctl = nfe_ioctl;
611 	ifp->if_start = nfe_start;
612 #ifdef IFPOLL_ENABLE
613 	ifp->if_npoll = nfe_npoll;
614 #endif
615 	ifp->if_watchdog = nfe_watchdog;
616 	ifp->if_init = nfe_init;
617 	ifp->if_nmbclusters = sc->sc_rx_ring_count;
618 	ifq_set_maxlen(&ifp->if_snd, sc->sc_tx_ring_count);
619 	ifq_set_ready(&ifp->if_snd);
620 
621 	ifp->if_capabilities = IFCAP_VLAN_MTU;
622 
623 	if (sc->sc_caps & NFE_HW_VLAN)
624 		ifp->if_capabilities |= IFCAP_VLAN_HWTAGGING;
625 
626 #ifdef NFE_CSUM
627 	if (sc->sc_caps & NFE_HW_CSUM) {
628 		ifp->if_capabilities |= IFCAP_HWCSUM;
629 		ifp->if_hwassist = NFE_CSUM_FEATURES;
630 	}
631 #else
632 	sc->sc_caps &= ~NFE_HW_CSUM;
633 #endif
634 	ifp->if_capenable = ifp->if_capabilities;
635 
636 	callout_init(&sc->sc_tick_ch);
637 
638 	ether_ifattach(ifp, eaddr, NULL);
639 
640 	ifq_set_cpuid(&ifp->if_snd, rman_get_cpuid(sc->sc_irq_res));
641 
642 #ifdef IFPOLL_ENABLE
643 	ifpoll_compat_setup(&sc->sc_npoll, ctx, (struct sysctl_oid *)tree,
644 	    device_get_unit(dev), ifp->if_serializer);
645 #endif
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 	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_irq_res != NULL) {
682 		bus_release_resource(dev, SYS_RES_IRQ, sc->sc_irq_rid,
683 				     sc->sc_irq_res);
684 	}
685 
686 	if (sc->sc_mem_res != NULL) {
687 		bus_release_resource(dev, SYS_RES_MEMORY, sc->sc_mem_rid,
688 				     sc->sc_mem_res);
689 	}
690 
691 	nfe_free_tx_ring(sc, &sc->txq);
692 	nfe_free_rx_ring(sc, &sc->rxq);
693 	if (sc->sc_dtag != NULL)
694 		bus_dma_tag_destroy(sc->sc_dtag);
695 
696 	return 0;
697 }
698 
699 static void
700 nfe_shutdown(device_t dev)
701 {
702 	struct nfe_softc *sc = device_get_softc(dev);
703 	struct ifnet *ifp = &sc->arpcom.ac_if;
704 
705 	lwkt_serialize_enter(ifp->if_serializer);
706 	nfe_stop(sc);
707 	lwkt_serialize_exit(ifp->if_serializer);
708 }
709 
710 static int
711 nfe_suspend(device_t dev)
712 {
713 	struct nfe_softc *sc = device_get_softc(dev);
714 	struct ifnet *ifp = &sc->arpcom.ac_if;
715 
716 	lwkt_serialize_enter(ifp->if_serializer);
717 	nfe_stop(sc);
718 	lwkt_serialize_exit(ifp->if_serializer);
719 
720 	return 0;
721 }
722 
723 static int
724 nfe_resume(device_t dev)
725 {
726 	struct nfe_softc *sc = device_get_softc(dev);
727 	struct ifnet *ifp = &sc->arpcom.ac_if;
728 
729 	lwkt_serialize_enter(ifp->if_serializer);
730 	if (ifp->if_flags & IFF_UP)
731 		nfe_init(sc);
732 	lwkt_serialize_exit(ifp->if_serializer);
733 
734 	return 0;
735 }
736 
737 static void
738 nfe_miibus_statchg(device_t dev)
739 {
740 	struct nfe_softc *sc = device_get_softc(dev);
741 	struct mii_data *mii = device_get_softc(sc->sc_miibus);
742 	uint32_t phy, seed, misc = NFE_MISC1_MAGIC, link = NFE_MEDIA_SET;
743 
744 	ASSERT_SERIALIZED(sc->arpcom.ac_if.if_serializer);
745 
746 	phy = NFE_READ(sc, NFE_PHY_IFACE);
747 	phy &= ~(NFE_PHY_HDX | NFE_PHY_100TX | NFE_PHY_1000T);
748 
749 	seed = NFE_READ(sc, NFE_RNDSEED);
750 	seed &= ~NFE_SEED_MASK;
751 
752 	if ((mii->mii_media_active & IFM_GMASK) == IFM_HDX) {
753 		phy  |= NFE_PHY_HDX;	/* half-duplex */
754 		misc |= NFE_MISC1_HDX;
755 	}
756 
757 	switch (IFM_SUBTYPE(mii->mii_media_active)) {
758 	case IFM_1000_T:	/* full-duplex only */
759 		link |= NFE_MEDIA_1000T;
760 		seed |= NFE_SEED_1000T;
761 		phy  |= NFE_PHY_1000T;
762 		break;
763 	case IFM_100_TX:
764 		link |= NFE_MEDIA_100TX;
765 		seed |= NFE_SEED_100TX;
766 		phy  |= NFE_PHY_100TX;
767 		break;
768 	case IFM_10_T:
769 		link |= NFE_MEDIA_10T;
770 		seed |= NFE_SEED_10T;
771 		break;
772 	}
773 
774 	NFE_WRITE(sc, NFE_RNDSEED, seed);	/* XXX: gigabit NICs only? */
775 
776 	NFE_WRITE(sc, NFE_PHY_IFACE, phy);
777 	NFE_WRITE(sc, NFE_MISC1, misc);
778 	NFE_WRITE(sc, NFE_LINKSPEED, link);
779 }
780 
781 static int
782 nfe_miibus_readreg(device_t dev, int phy, int reg)
783 {
784 	struct nfe_softc *sc = device_get_softc(dev);
785 	uint32_t val;
786 	int ntries;
787 
788 	NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
789 
790 	if (NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY) {
791 		NFE_WRITE(sc, NFE_PHY_CTL, NFE_PHY_BUSY);
792 		DELAY(100);
793 	}
794 
795 	NFE_WRITE(sc, NFE_PHY_CTL, (phy << NFE_PHYADD_SHIFT) | reg);
796 
797 	for (ntries = 0; ntries < 1000; ntries++) {
798 		DELAY(100);
799 		if (!(NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY))
800 			break;
801 	}
802 	if (ntries == 1000) {
803 		DPRINTFN(sc, 2, "timeout waiting for PHY %s\n", "");
804 		return 0;
805 	}
806 
807 	if (NFE_READ(sc, NFE_PHY_STATUS) & NFE_PHY_ERROR) {
808 		DPRINTFN(sc, 2, "could not read PHY %s\n", "");
809 		return 0;
810 	}
811 
812 	val = NFE_READ(sc, NFE_PHY_DATA);
813 	if (val != 0xffffffff && val != 0)
814 		sc->mii_phyaddr = phy;
815 
816 	DPRINTFN(sc, 2, "mii read phy %d reg 0x%x ret 0x%x\n", phy, reg, val);
817 
818 	return val;
819 }
820 
821 static void
822 nfe_miibus_writereg(device_t dev, int phy, int reg, int val)
823 {
824 	struct nfe_softc *sc = device_get_softc(dev);
825 	uint32_t ctl;
826 	int ntries;
827 
828 	NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
829 
830 	if (NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY) {
831 		NFE_WRITE(sc, NFE_PHY_CTL, NFE_PHY_BUSY);
832 		DELAY(100);
833 	}
834 
835 	NFE_WRITE(sc, NFE_PHY_DATA, val);
836 	ctl = NFE_PHY_WRITE | (phy << NFE_PHYADD_SHIFT) | reg;
837 	NFE_WRITE(sc, NFE_PHY_CTL, ctl);
838 
839 	for (ntries = 0; ntries < 1000; ntries++) {
840 		DELAY(100);
841 		if (!(NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY))
842 			break;
843 	}
844 
845 #ifdef NFE_DEBUG
846 	if (ntries == 1000)
847 		DPRINTFN(sc, 2, "could not write to PHY %s\n", "");
848 #endif
849 }
850 
851 #ifdef IFPOLL_ENABLE
852 
853 static void
854 nfe_npoll_compat(struct ifnet *ifp, void *arg __unused, int count __unused)
855 {
856 	struct nfe_softc *sc = ifp->if_softc;
857 
858 	ASSERT_SERIALIZED(ifp->if_serializer);
859 
860 	nfe_rxeof(sc);
861 	nfe_txeof(sc, 1);
862 }
863 
864 static void
865 nfe_disable_intrs(struct nfe_softc *sc)
866 {
867 	/* Disable interrupts */
868 	NFE_WRITE(sc, NFE_IRQ_MASK, 0);
869 	sc->sc_flags &= ~NFE_F_IRQ_TIMER;
870 	sc->sc_npoll.ifpc_stcount = 0;
871 }
872 
873 static void
874 nfe_npoll(struct ifnet *ifp, struct ifpoll_info *info)
875 {
876 	struct nfe_softc *sc = ifp->if_softc;
877 
878 	ASSERT_SERIALIZED(ifp->if_serializer);
879 
880 	if (info != NULL) {
881 		int cpuid = sc->sc_npoll.ifpc_cpuid;
882 
883 		info->ifpi_rx[cpuid].poll_func = nfe_npoll_compat;
884 		info->ifpi_rx[cpuid].arg = NULL;
885 		info->ifpi_rx[cpuid].serializer = ifp->if_serializer;
886 
887 		if (ifp->if_flags & IFF_RUNNING)
888 			nfe_disable_intrs(sc);
889 		ifq_set_cpuid(&ifp->if_snd, cpuid);
890 	} else {
891 		if (ifp->if_flags & IFF_RUNNING)
892 			nfe_enable_intrs(sc);
893 		ifq_set_cpuid(&ifp->if_snd, rman_get_cpuid(sc->sc_irq_res));
894 	}
895 }
896 
897 #endif	/* IFPOLL_ENABLE */
898 
899 static void
900 nfe_intr(void *arg)
901 {
902 	struct nfe_softc *sc = arg;
903 	struct ifnet *ifp = &sc->arpcom.ac_if;
904 	uint32_t r;
905 
906 	r = NFE_READ(sc, NFE_IRQ_STATUS);
907 	if (r == 0)
908 		return;	/* not for us */
909 	NFE_WRITE(sc, NFE_IRQ_STATUS, r);
910 
911 	if (sc->sc_rate_second != time_uptime) {
912 		/*
913 		 * Calculate sc_rate_avg - interrupts per second.
914 		 */
915 		sc->sc_rate_second = time_uptime;
916 		if (sc->sc_rate_avg < sc->sc_rate_acc)
917 			sc->sc_rate_avg = sc->sc_rate_acc;
918 		else
919 			sc->sc_rate_avg = (sc->sc_rate_avg * 3 +
920 					   sc->sc_rate_acc) / 4;
921 		sc->sc_rate_acc = 0;
922 	} else if (sc->sc_rate_avg < sc->sc_rate_acc) {
923 		/*
924 		 * Don't wait for a tick to roll over if we are taking
925 		 * a lot of interrupts.
926 		 */
927 		sc->sc_rate_avg = sc->sc_rate_acc;
928 	}
929 
930 	DPRINTFN(sc, 5, "%s: interrupt register %x\n", __func__, r);
931 
932 	if (r & NFE_IRQ_LINK) {
933 		NFE_READ(sc, NFE_PHY_STATUS);
934 		NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
935 		DPRINTF(sc, "link state changed %s\n", "");
936 	}
937 
938 	if (ifp->if_flags & IFF_RUNNING) {
939 		int ret;
940 		int rate;
941 
942 		/* check Rx ring */
943 		ret = nfe_rxeof(sc);
944 
945 		/* check Tx ring */
946 		ret |= nfe_txeof(sc, 1);
947 
948 		/* update the rate accumulator */
949 		if (ret)
950 			++sc->sc_rate_acc;
951 
952 		if (sc->sc_flags & NFE_F_DYN_IM) {
953 			rate = 1000000 / sc->sc_imtime;
954 			if ((sc->sc_flags & NFE_F_IRQ_TIMER) == 0 &&
955 			    sc->sc_rate_avg > rate) {
956 				/*
957 				 * Use the hardware timer to reduce the
958 				 * interrupt rate if the discrete interrupt
959 				 * rate has exceeded our threshold.
960 				 */
961 				NFE_WRITE(sc, NFE_IRQ_MASK, NFE_IRQ_IMTIMER);
962 				sc->sc_flags |= NFE_F_IRQ_TIMER;
963 			} else if ((sc->sc_flags & NFE_F_IRQ_TIMER) &&
964 				   sc->sc_rate_avg <= rate) {
965 				/*
966 				 * Use discrete TX/RX interrupts if the rate
967 				 * has fallen below our threshold.
968 				 */
969 				NFE_WRITE(sc, NFE_IRQ_MASK, NFE_IRQ_NOIMTIMER);
970 				sc->sc_flags &= ~NFE_F_IRQ_TIMER;
971 
972 				/*
973 				 * Recollect, mainly to avoid the possible race
974 				 * introduced by changing interrupt masks.
975 				 */
976 				nfe_rxeof(sc);
977 				nfe_txeof(sc, 1);
978 			}
979 		}
980 	}
981 }
982 
983 static int
984 nfe_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data, struct ucred *cr)
985 {
986 	struct nfe_softc *sc = ifp->if_softc;
987 	struct ifreq *ifr = (struct ifreq *)data;
988 	struct mii_data *mii;
989 	int error = 0, mask, jumbo_cap;
990 
991 	ASSERT_SERIALIZED(ifp->if_serializer);
992 
993 	switch (cmd) {
994 	case SIOCSIFMTU:
995 		if ((sc->sc_caps & NFE_JUMBO_SUP) && sc->rxq.jbuf != NULL)
996 			jumbo_cap = 1;
997 		else
998 			jumbo_cap = 0;
999 
1000 		if ((jumbo_cap && ifr->ifr_mtu > NFE_JUMBO_MTU) ||
1001 		    (!jumbo_cap && ifr->ifr_mtu > ETHERMTU)) {
1002 			return EINVAL;
1003 		} else if (ifp->if_mtu != ifr->ifr_mtu) {
1004 			ifp->if_mtu = ifr->ifr_mtu;
1005 			if (ifp->if_flags & IFF_RUNNING)
1006 				nfe_init(sc);
1007 		}
1008 		break;
1009 	case SIOCSIFFLAGS:
1010 		if (ifp->if_flags & IFF_UP) {
1011 			/*
1012 			 * If only the PROMISC or ALLMULTI flag changes, then
1013 			 * don't do a full re-init of the chip, just update
1014 			 * the Rx filter.
1015 			 */
1016 			if ((ifp->if_flags & IFF_RUNNING) &&
1017 			    ((ifp->if_flags ^ sc->sc_if_flags) &
1018 			     (IFF_ALLMULTI | IFF_PROMISC)) != 0) {
1019 				nfe_setmulti(sc);
1020 			} else {
1021 				if (!(ifp->if_flags & IFF_RUNNING))
1022 					nfe_init(sc);
1023 			}
1024 		} else {
1025 			if (ifp->if_flags & IFF_RUNNING)
1026 				nfe_stop(sc);
1027 		}
1028 		sc->sc_if_flags = ifp->if_flags;
1029 		break;
1030 	case SIOCADDMULTI:
1031 	case SIOCDELMULTI:
1032 		if (ifp->if_flags & IFF_RUNNING)
1033 			nfe_setmulti(sc);
1034 		break;
1035 	case SIOCSIFMEDIA:
1036 	case SIOCGIFMEDIA:
1037 		mii = device_get_softc(sc->sc_miibus);
1038 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, cmd);
1039 		break;
1040         case SIOCSIFCAP:
1041 		mask = (ifr->ifr_reqcap ^ ifp->if_capenable) & IFCAP_HWCSUM;
1042 		if (mask && (ifp->if_capabilities & IFCAP_HWCSUM)) {
1043 			ifp->if_capenable ^= mask;
1044 			if (IFCAP_TXCSUM & ifp->if_capenable)
1045 				ifp->if_hwassist = NFE_CSUM_FEATURES;
1046 			else
1047 				ifp->if_hwassist = 0;
1048 
1049 			if (ifp->if_flags & IFF_RUNNING)
1050 				nfe_init(sc);
1051 		}
1052 		break;
1053 	default:
1054 		error = ether_ioctl(ifp, cmd, data);
1055 		break;
1056 	}
1057 	return error;
1058 }
1059 
1060 static int
1061 nfe_rxeof(struct nfe_softc *sc)
1062 {
1063 	struct ifnet *ifp = &sc->arpcom.ac_if;
1064 	struct nfe_rx_ring *ring = &sc->rxq;
1065 	int reap;
1066 
1067 	reap = 0;
1068 	for (;;) {
1069 		struct nfe_rx_data *data = &ring->data[ring->cur];
1070 		struct mbuf *m;
1071 		uint16_t flags;
1072 		int len, error;
1073 
1074 		if (sc->sc_caps & NFE_40BIT_ADDR) {
1075 			struct nfe_desc64 *desc64 = &ring->desc64[ring->cur];
1076 
1077 			flags = le16toh(desc64->flags);
1078 			len = le16toh(desc64->length) & 0x3fff;
1079 		} else {
1080 			struct nfe_desc32 *desc32 = &ring->desc32[ring->cur];
1081 
1082 			flags = le16toh(desc32->flags);
1083 			len = le16toh(desc32->length) & 0x3fff;
1084 		}
1085 
1086 		if (flags & NFE_RX_READY)
1087 			break;
1088 
1089 		reap = 1;
1090 
1091 		if ((sc->sc_caps & (NFE_JUMBO_SUP | NFE_40BIT_ADDR)) == 0) {
1092 			if (!(flags & NFE_RX_VALID_V1))
1093 				goto skip;
1094 
1095 			if ((flags & NFE_RX_FIXME_V1) == NFE_RX_FIXME_V1) {
1096 				flags &= ~NFE_RX_ERROR;
1097 				len--;	/* fix buffer length */
1098 			}
1099 		} else {
1100 			if (!(flags & NFE_RX_VALID_V2))
1101 				goto skip;
1102 
1103 			if ((flags & NFE_RX_FIXME_V2) == NFE_RX_FIXME_V2) {
1104 				flags &= ~NFE_RX_ERROR;
1105 				len--;	/* fix buffer length */
1106 			}
1107 		}
1108 
1109 		if (flags & NFE_RX_ERROR) {
1110 			IFNET_STAT_INC(ifp, ierrors, 1);
1111 			goto skip;
1112 		}
1113 
1114 		m = data->m;
1115 
1116 		if (sc->sc_flags & NFE_F_USE_JUMBO)
1117 			error = nfe_newbuf_jumbo(sc, ring, ring->cur, 0);
1118 		else
1119 			error = nfe_newbuf_std(sc, ring, ring->cur, 0);
1120 		if (error) {
1121 			IFNET_STAT_INC(ifp, ierrors, 1);
1122 			goto skip;
1123 		}
1124 
1125 		/* finalize mbuf */
1126 		m->m_pkthdr.len = m->m_len = len;
1127 		m->m_pkthdr.rcvif = ifp;
1128 
1129 		if ((ifp->if_capenable & IFCAP_RXCSUM) &&
1130 		    (flags & NFE_RX_CSUMOK)) {
1131 			if (flags & NFE_RX_IP_CSUMOK_V2) {
1132 				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED |
1133 							  CSUM_IP_VALID;
1134 			}
1135 
1136 			if (flags &
1137 			    (NFE_RX_UDP_CSUMOK_V2 | NFE_RX_TCP_CSUMOK_V2)) {
1138 				m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
1139 							  CSUM_PSEUDO_HDR |
1140 							  CSUM_FRAG_NOT_CHECKED;
1141 				m->m_pkthdr.csum_data = 0xffff;
1142 			}
1143 		}
1144 
1145 		IFNET_STAT_INC(ifp, ipackets, 1);
1146 		ifp->if_input(ifp, m, NULL, -1);
1147 skip:
1148 		nfe_set_ready_rxdesc(sc, ring, ring->cur);
1149 		sc->rxq.cur = (sc->rxq.cur + 1) % sc->sc_rx_ring_count;
1150 	}
1151 	return reap;
1152 }
1153 
1154 static int
1155 nfe_txeof(struct nfe_softc *sc, int start)
1156 {
1157 	struct ifnet *ifp = &sc->arpcom.ac_if;
1158 	struct nfe_tx_ring *ring = &sc->txq;
1159 	struct nfe_tx_data *data = NULL;
1160 
1161 	while (ring->next != ring->cur) {
1162 		uint16_t flags;
1163 
1164 		if (sc->sc_caps & NFE_40BIT_ADDR)
1165 			flags = le16toh(ring->desc64[ring->next].flags);
1166 		else
1167 			flags = le16toh(ring->desc32[ring->next].flags);
1168 
1169 		if (flags & NFE_TX_VALID)
1170 			break;
1171 
1172 		data = &ring->data[ring->next];
1173 
1174 		if ((sc->sc_caps & (NFE_JUMBO_SUP | NFE_40BIT_ADDR)) == 0) {
1175 			if (!(flags & NFE_TX_LASTFRAG_V1) && data->m == NULL)
1176 				goto skip;
1177 
1178 			if ((flags & NFE_TX_ERROR_V1) != 0) {
1179 				if_printf(ifp, "tx v1 error 0x%pb%i\n",
1180 					  NFE_V1_TXERR, flags);
1181 				IFNET_STAT_INC(ifp, oerrors, 1);
1182 			} else {
1183 				IFNET_STAT_INC(ifp, opackets, 1);
1184 			}
1185 		} else {
1186 			if (!(flags & NFE_TX_LASTFRAG_V2) && data->m == NULL)
1187 				goto skip;
1188 
1189 			if ((flags & NFE_TX_ERROR_V2) != 0) {
1190 				if_printf(ifp, "tx v2 error 0x%pb%i\n",
1191 					  NFE_V2_TXERR, flags);
1192 				IFNET_STAT_INC(ifp, oerrors, 1);
1193 			} else {
1194 				IFNET_STAT_INC(ifp, opackets, 1);
1195 			}
1196 		}
1197 
1198 		if (data->m == NULL) {	/* should not get there */
1199 			if_printf(ifp,
1200 				  "last fragment bit w/o associated mbuf!\n");
1201 			goto skip;
1202 		}
1203 
1204 		/* last fragment of the mbuf chain transmitted */
1205 		bus_dmamap_unload(ring->data_tag, data->map);
1206 		m_freem(data->m);
1207 		data->m = NULL;
1208 skip:
1209 		ring->queued--;
1210 		KKASSERT(ring->queued >= 0);
1211 		ring->next = (ring->next + 1) % sc->sc_tx_ring_count;
1212 	}
1213 
1214 	if (sc->sc_tx_ring_count - ring->queued >=
1215 	    sc->sc_tx_spare + NFE_NSEG_RSVD)
1216 		ifq_clr_oactive(&ifp->if_snd);
1217 
1218 	if (ring->queued == 0)
1219 		ifp->if_timer = 0;
1220 
1221 	if (start && !ifq_is_empty(&ifp->if_snd))
1222 		if_devstart(ifp);
1223 
1224 	if (data != NULL)
1225 		return 1;
1226 	else
1227 		return 0;
1228 }
1229 
1230 static int
1231 nfe_encap(struct nfe_softc *sc, struct nfe_tx_ring *ring, struct mbuf *m0)
1232 {
1233 	bus_dma_segment_t segs[NFE_MAX_SCATTER];
1234 	struct nfe_tx_data *data, *data_map;
1235 	bus_dmamap_t map;
1236 	struct nfe_desc64 *desc64 = NULL;
1237 	struct nfe_desc32 *desc32 = NULL;
1238 	uint16_t flags = 0;
1239 	uint32_t vtag = 0;
1240 	int error, i, j, maxsegs, nsegs;
1241 
1242 	data = &ring->data[ring->cur];
1243 	map = data->map;
1244 	data_map = data;	/* Remember who owns the DMA map */
1245 
1246 	maxsegs = (sc->sc_tx_ring_count - ring->queued) - NFE_NSEG_RSVD;
1247 	if (maxsegs > NFE_MAX_SCATTER)
1248 		maxsegs = NFE_MAX_SCATTER;
1249 	KASSERT(maxsegs >= sc->sc_tx_spare,
1250 		("not enough segments %d,%d", maxsegs, sc->sc_tx_spare));
1251 
1252 	error = bus_dmamap_load_mbuf_defrag(ring->data_tag, map, &m0,
1253 			segs, maxsegs, &nsegs, BUS_DMA_NOWAIT);
1254 	if (error)
1255 		goto back;
1256 	bus_dmamap_sync(ring->data_tag, map, BUS_DMASYNC_PREWRITE);
1257 
1258 	error = 0;
1259 
1260 	/* setup h/w VLAN tagging */
1261 	if (m0->m_flags & M_VLANTAG)
1262 		vtag = m0->m_pkthdr.ether_vlantag;
1263 
1264 	if (sc->arpcom.ac_if.if_capenable & IFCAP_TXCSUM) {
1265 		if (m0->m_pkthdr.csum_flags & CSUM_IP)
1266 			flags |= NFE_TX_IP_CSUM;
1267 		if (m0->m_pkthdr.csum_flags & (CSUM_TCP | CSUM_UDP))
1268 			flags |= NFE_TX_TCP_CSUM;
1269 	}
1270 
1271 	/*
1272 	 * XXX urm. somebody is unaware of how hardware works.  You
1273 	 * absolutely CANNOT set NFE_TX_VALID on the next descriptor in
1274 	 * the ring until the entire chain is actually *VALID*.  Otherwise
1275 	 * the hardware may encounter a partially initialized chain that
1276 	 * is marked as being ready to go when it in fact is not ready to
1277 	 * go.
1278 	 */
1279 
1280 	for (i = 0; i < nsegs; i++) {
1281 		j = (ring->cur + i) % sc->sc_tx_ring_count;
1282 		data = &ring->data[j];
1283 
1284 		if (sc->sc_caps & NFE_40BIT_ADDR) {
1285 			desc64 = &ring->desc64[j];
1286 			desc64->physaddr[0] =
1287 			    htole32(NFE_ADDR_HI(segs[i].ds_addr));
1288 			desc64->physaddr[1] =
1289 			    htole32(NFE_ADDR_LO(segs[i].ds_addr));
1290 			desc64->length = htole16(segs[i].ds_len - 1);
1291 			desc64->vtag = htole32(vtag);
1292 			desc64->flags = htole16(flags);
1293 		} else {
1294 			desc32 = &ring->desc32[j];
1295 			desc32->physaddr = htole32(segs[i].ds_addr);
1296 			desc32->length = htole16(segs[i].ds_len - 1);
1297 			desc32->flags = htole16(flags);
1298 		}
1299 
1300 		/* csum flags and vtag belong to the first fragment only */
1301 		flags &= ~(NFE_TX_IP_CSUM | NFE_TX_TCP_CSUM);
1302 		vtag = 0;
1303 
1304 		ring->queued++;
1305 		KKASSERT(ring->queued <= sc->sc_tx_ring_count);
1306 	}
1307 
1308 	/* the whole mbuf chain has been DMA mapped, fix last descriptor */
1309 	if (sc->sc_caps & NFE_40BIT_ADDR) {
1310 		desc64->flags |= htole16(NFE_TX_LASTFRAG_V2);
1311 	} else {
1312 		if (sc->sc_caps & NFE_JUMBO_SUP)
1313 			flags = NFE_TX_LASTFRAG_V2;
1314 		else
1315 			flags = NFE_TX_LASTFRAG_V1;
1316 		desc32->flags |= htole16(flags);
1317 	}
1318 
1319 	/*
1320 	 * Set NFE_TX_VALID backwards so the hardware doesn't see the
1321 	 * whole mess until the first descriptor in the map is flagged.
1322 	 */
1323 	for (i = nsegs - 1; i >= 0; --i) {
1324 		j = (ring->cur + i) % sc->sc_tx_ring_count;
1325 		if (sc->sc_caps & NFE_40BIT_ADDR) {
1326 			desc64 = &ring->desc64[j];
1327 			desc64->flags |= htole16(NFE_TX_VALID);
1328 		} else {
1329 			desc32 = &ring->desc32[j];
1330 			desc32->flags |= htole16(NFE_TX_VALID);
1331 		}
1332 	}
1333 	ring->cur = (ring->cur + nsegs) % sc->sc_tx_ring_count;
1334 
1335 	/* Exchange DMA map */
1336 	data_map->map = data->map;
1337 	data->map = map;
1338 	data->m = m0;
1339 back:
1340 	if (error)
1341 		m_freem(m0);
1342 	return error;
1343 }
1344 
1345 static void
1346 nfe_start(struct ifnet *ifp, struct ifaltq_subque *ifsq)
1347 {
1348 	struct nfe_softc *sc = ifp->if_softc;
1349 	struct nfe_tx_ring *ring = &sc->txq;
1350 	int count = 0, oactive = 0;
1351 	struct mbuf *m0;
1352 
1353 	ASSERT_ALTQ_SQ_DEFAULT(ifp, ifsq);
1354 	ASSERT_SERIALIZED(ifp->if_serializer);
1355 
1356 	if ((ifp->if_flags & IFF_RUNNING) == 0 || ifq_is_oactive(&ifp->if_snd))
1357 		return;
1358 
1359 	for (;;) {
1360 		int error;
1361 
1362 		if (sc->sc_tx_ring_count - ring->queued <
1363 		    sc->sc_tx_spare + NFE_NSEG_RSVD) {
1364 			if (oactive) {
1365 				ifq_set_oactive(&ifp->if_snd);
1366 				break;
1367 			}
1368 
1369 			nfe_txeof(sc, 0);
1370 			oactive = 1;
1371 			continue;
1372 		}
1373 
1374 		m0 = ifq_dequeue(&ifp->if_snd);
1375 		if (m0 == NULL)
1376 			break;
1377 
1378 		ETHER_BPF_MTAP(ifp, m0);
1379 
1380 		error = nfe_encap(sc, ring, m0);
1381 		if (error) {
1382 			IFNET_STAT_INC(ifp, oerrors, 1);
1383 			if (error == EFBIG) {
1384 				if (oactive) {
1385 					ifq_set_oactive(&ifp->if_snd);
1386 					break;
1387 				}
1388 				nfe_txeof(sc, 0);
1389 				oactive = 1;
1390 			}
1391 			continue;
1392 		} else {
1393 			oactive = 0;
1394 		}
1395 		++count;
1396 
1397 		/*
1398 		 * NOTE:
1399 		 * `m0' may be freed in nfe_encap(), so
1400 		 * it should not be touched any more.
1401 		 */
1402 	}
1403 
1404 	if (count == 0)	/* nothing sent */
1405 		return;
1406 
1407 	/* Kick Tx */
1408 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_KICKTX | sc->rxtxctl);
1409 
1410 	/*
1411 	 * Set a timeout in case the chip goes out to lunch.
1412 	 */
1413 	ifp->if_timer = 5;
1414 }
1415 
1416 static void
1417 nfe_watchdog(struct ifnet *ifp)
1418 {
1419 	struct nfe_softc *sc = ifp->if_softc;
1420 
1421 	ASSERT_SERIALIZED(ifp->if_serializer);
1422 
1423 	if (ifp->if_flags & IFF_RUNNING) {
1424 		if_printf(ifp, "watchdog timeout - lost interrupt recovered\n");
1425 		nfe_txeof(sc, 1);
1426 		return;
1427 	}
1428 
1429 	if_printf(ifp, "watchdog timeout\n");
1430 
1431 	nfe_init(ifp->if_softc);
1432 
1433 	IFNET_STAT_INC(ifp, oerrors, 1);
1434 }
1435 
1436 static void
1437 nfe_init(void *xsc)
1438 {
1439 	struct nfe_softc *sc = xsc;
1440 	struct ifnet *ifp = &sc->arpcom.ac_if;
1441 	uint32_t tmp;
1442 	int error;
1443 
1444 	ASSERT_SERIALIZED(ifp->if_serializer);
1445 
1446 	nfe_stop(sc);
1447 
1448 	if ((sc->sc_caps & NFE_NO_PWRCTL) == 0)
1449 		nfe_mac_reset(sc);
1450 
1451 	/*
1452 	 * NOTE:
1453 	 * Switching between jumbo frames and normal frames should
1454 	 * be done _after_ nfe_stop() but _before_ nfe_init_rx_ring().
1455 	 */
1456 	if (ifp->if_mtu > ETHERMTU) {
1457 		sc->sc_flags |= NFE_F_USE_JUMBO;
1458 		sc->rxq.bufsz = NFE_JBYTES;
1459 		sc->sc_tx_spare = NFE_NSEG_SPARE_JUMBO;
1460 		if (bootverbose)
1461 			if_printf(ifp, "use jumbo frames\n");
1462 	} else {
1463 		sc->sc_flags &= ~NFE_F_USE_JUMBO;
1464 		sc->rxq.bufsz = MCLBYTES;
1465 		sc->sc_tx_spare = NFE_NSEG_SPARE;
1466 		if (bootverbose)
1467 			if_printf(ifp, "use non-jumbo frames\n");
1468 	}
1469 
1470 	error = nfe_init_tx_ring(sc, &sc->txq);
1471 	if (error) {
1472 		nfe_stop(sc);
1473 		return;
1474 	}
1475 
1476 	error = nfe_init_rx_ring(sc, &sc->rxq);
1477 	if (error) {
1478 		nfe_stop(sc);
1479 		return;
1480 	}
1481 
1482 	NFE_WRITE(sc, NFE_TX_POLL, 0);
1483 	NFE_WRITE(sc, NFE_STATUS, 0);
1484 
1485 	sc->rxtxctl = NFE_RXTX_BIT2 | sc->rxtxctl_desc;
1486 
1487 	if (ifp->if_capenable & IFCAP_RXCSUM)
1488 		sc->rxtxctl |= NFE_RXTX_RXCSUM;
1489 
1490 	/*
1491 	 * Although the adapter is capable of stripping VLAN tags from received
1492 	 * frames (NFE_RXTX_VTAG_STRIP), we do not enable this functionality on
1493 	 * purpose.  This will be done in software by our network stack.
1494 	 */
1495 	if (sc->sc_caps & NFE_HW_VLAN)
1496 		sc->rxtxctl |= NFE_RXTX_VTAG_INSERT;
1497 
1498 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_RESET | sc->rxtxctl);
1499 	DELAY(10);
1500 	NFE_WRITE(sc, NFE_RXTX_CTL, sc->rxtxctl);
1501 
1502 	if (sc->sc_caps & NFE_HW_VLAN)
1503 		NFE_WRITE(sc, NFE_VTAG_CTL, NFE_VTAG_ENABLE);
1504 
1505 	NFE_WRITE(sc, NFE_SETUP_R6, 0);
1506 
1507 	/* set MAC address */
1508 	nfe_set_macaddr(sc, sc->arpcom.ac_enaddr);
1509 
1510 	/* tell MAC where rings are in memory */
1511 	if (sc->sc_caps & NFE_40BIT_ADDR) {
1512 		NFE_WRITE(sc, NFE_RX_RING_ADDR_HI,
1513 			  NFE_ADDR_HI(sc->rxq.physaddr));
1514 	}
1515 	NFE_WRITE(sc, NFE_RX_RING_ADDR_LO, NFE_ADDR_LO(sc->rxq.physaddr));
1516 
1517 	if (sc->sc_caps & NFE_40BIT_ADDR) {
1518 		NFE_WRITE(sc, NFE_TX_RING_ADDR_HI,
1519 			  NFE_ADDR_HI(sc->txq.physaddr));
1520 	}
1521 	NFE_WRITE(sc, NFE_TX_RING_ADDR_LO, NFE_ADDR_LO(sc->txq.physaddr));
1522 
1523 	NFE_WRITE(sc, NFE_RING_SIZE,
1524 	    (sc->sc_rx_ring_count - 1) << 16 |
1525 	    (sc->sc_tx_ring_count - 1));
1526 
1527 	NFE_WRITE(sc, NFE_RXBUFSZ, sc->rxq.bufsz);
1528 
1529 	/* force MAC to wakeup */
1530 	tmp = NFE_READ(sc, NFE_PWR_STATE);
1531 	NFE_WRITE(sc, NFE_PWR_STATE, tmp | NFE_PWR_WAKEUP);
1532 	DELAY(10);
1533 	tmp = NFE_READ(sc, NFE_PWR_STATE);
1534 	NFE_WRITE(sc, NFE_PWR_STATE, tmp | NFE_PWR_VALID);
1535 
1536 	NFE_WRITE(sc, NFE_SETUP_R1, NFE_R1_MAGIC);
1537 	NFE_WRITE(sc, NFE_SETUP_R2, NFE_R2_MAGIC);
1538 	NFE_WRITE(sc, NFE_SETUP_R6, NFE_R6_MAGIC);
1539 
1540 	/* update MAC knowledge of PHY; generates a NFE_IRQ_LINK interrupt */
1541 	NFE_WRITE(sc, NFE_STATUS, sc->mii_phyaddr << 24 | NFE_STATUS_MAGIC);
1542 
1543 	NFE_WRITE(sc, NFE_SETUP_R4, NFE_R4_MAGIC);
1544 
1545 	sc->rxtxctl &= ~NFE_RXTX_BIT2;
1546 	NFE_WRITE(sc, NFE_RXTX_CTL, sc->rxtxctl);
1547 	DELAY(10);
1548 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_BIT1 | sc->rxtxctl);
1549 
1550 	/* set Rx filter */
1551 	nfe_setmulti(sc);
1552 
1553 	nfe_ifmedia_upd(ifp);
1554 
1555 	/* enable Rx */
1556 	NFE_WRITE(sc, NFE_RX_CTL, NFE_RX_START);
1557 
1558 	/* enable Tx */
1559 	NFE_WRITE(sc, NFE_TX_CTL, NFE_TX_START);
1560 
1561 	NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
1562 
1563 #ifdef IFPOLL_ENABLE
1564 	if (ifp->if_flags & IFF_NPOLLING)
1565 		nfe_disable_intrs(sc);
1566 	else
1567 #endif
1568 	nfe_enable_intrs(sc);
1569 
1570 	callout_reset(&sc->sc_tick_ch, hz, nfe_tick, sc);
1571 
1572 	ifp->if_flags |= IFF_RUNNING;
1573 	ifq_clr_oactive(&ifp->if_snd);
1574 
1575 	/*
1576 	 * If we had stuff in the tx ring before its all cleaned out now
1577 	 * so we are not going to get an interrupt, jump-start any pending
1578 	 * output.
1579 	 */
1580 	if (!ifq_is_empty(&ifp->if_snd))
1581 		if_devstart(ifp);
1582 }
1583 
1584 static void
1585 nfe_stop(struct nfe_softc *sc)
1586 {
1587 	struct ifnet *ifp = &sc->arpcom.ac_if;
1588 	uint32_t rxtxctl = sc->rxtxctl_desc | NFE_RXTX_BIT2;
1589 	int i;
1590 
1591 	ASSERT_SERIALIZED(ifp->if_serializer);
1592 
1593 	callout_stop(&sc->sc_tick_ch);
1594 
1595 	ifp->if_timer = 0;
1596 	ifp->if_flags &= ~IFF_RUNNING;
1597 	ifq_clr_oactive(&ifp->if_snd);
1598 	sc->sc_flags &= ~NFE_F_IRQ_TIMER;
1599 
1600 #define WAITMAX	50000
1601 
1602 	/*
1603 	 * Abort Tx
1604 	 */
1605 	NFE_WRITE(sc, NFE_TX_CTL, 0);
1606 	for (i = 0; i < WAITMAX; ++i) {
1607 		DELAY(100);
1608 		if ((NFE_READ(sc, NFE_TX_STATUS) & NFE_TX_STATUS_BUSY) == 0)
1609 			break;
1610 	}
1611 	if (i == WAITMAX)
1612 		if_printf(ifp, "can't stop TX\n");
1613 	DELAY(100);
1614 
1615 	/*
1616 	 * Disable Rx
1617 	 */
1618 	NFE_WRITE(sc, NFE_RX_CTL, 0);
1619 	for (i = 0; i < WAITMAX; ++i) {
1620 		DELAY(100);
1621 		if ((NFE_READ(sc, NFE_RX_STATUS) & NFE_RX_STATUS_BUSY) == 0)
1622 			break;
1623 	}
1624 	if (i == WAITMAX)
1625 		if_printf(ifp, "can't stop RX\n");
1626 	DELAY(100);
1627 
1628 #undef WAITMAX
1629 
1630 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_RESET | rxtxctl);
1631 	DELAY(10);
1632 	NFE_WRITE(sc, NFE_RXTX_CTL, rxtxctl);
1633 
1634 	/* Disable interrupts */
1635 	NFE_WRITE(sc, NFE_IRQ_MASK, 0);
1636 
1637 	/* Reset Tx and Rx rings */
1638 	nfe_reset_tx_ring(sc, &sc->txq);
1639 	nfe_reset_rx_ring(sc, &sc->rxq);
1640 }
1641 
1642 static int
1643 nfe_alloc_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1644 {
1645 	int i, j, error, descsize;
1646 	bus_dmamem_t dmem;
1647 	void **desc;
1648 
1649 	if (sc->sc_caps & NFE_40BIT_ADDR) {
1650 		desc = (void *)&ring->desc64;
1651 		descsize = sizeof(struct nfe_desc64);
1652 	} else {
1653 		desc = (void *)&ring->desc32;
1654 		descsize = sizeof(struct nfe_desc32);
1655 	}
1656 
1657 	ring->bufsz = MCLBYTES;
1658 	ring->cur = ring->next = 0;
1659 
1660 	error = bus_dmamem_coherent(sc->sc_dtag, PAGE_SIZE, 0,
1661 				    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
1662 				    sc->sc_rx_ring_count * descsize,
1663 				    BUS_DMA_WAITOK | BUS_DMA_ZERO, &dmem);
1664 	if (error) {
1665 		if_printf(&sc->arpcom.ac_if,
1666 			  "could not create RX desc ring\n");
1667 		return error;
1668 	}
1669 	ring->tag = dmem.dmem_tag;
1670 	ring->map = dmem.dmem_map;
1671 	*desc = dmem.dmem_addr;
1672 	ring->physaddr = dmem.dmem_busaddr;
1673 
1674 	if (sc->sc_caps & NFE_JUMBO_SUP) {
1675 		ring->jbuf =
1676 		kmalloc(sizeof(struct nfe_jbuf) * NFE_JPOOL_COUNT(sc),
1677 			M_DEVBUF, M_WAITOK | M_ZERO);
1678 
1679 		error = nfe_jpool_alloc(sc, ring);
1680 		if (error) {
1681 			if_printf(&sc->arpcom.ac_if,
1682 				  "could not allocate jumbo frames\n");
1683 			kfree(ring->jbuf, M_DEVBUF);
1684 			ring->jbuf = NULL;
1685 			/* Allow jumbo frame allocation to fail */
1686 		}
1687 	}
1688 
1689 	ring->data = kmalloc(sizeof(struct nfe_rx_data) * sc->sc_rx_ring_count,
1690 			     M_DEVBUF, M_WAITOK | M_ZERO);
1691 
1692 	error = bus_dma_tag_create(sc->sc_dtag, 1, 0,
1693 				   BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
1694 				   NULL, NULL,
1695 				   MCLBYTES, 1, MCLBYTES,
1696 				   BUS_DMA_ALLOCNOW | BUS_DMA_WAITOK,
1697 				   &ring->data_tag);
1698 	if (error) {
1699 		if_printf(&sc->arpcom.ac_if,
1700 			  "could not create RX mbuf DMA tag\n");
1701 		return error;
1702 	}
1703 
1704 	/* Create a spare RX mbuf DMA map */
1705 	error = bus_dmamap_create(ring->data_tag, BUS_DMA_WAITOK,
1706 				  &ring->data_tmpmap);
1707 	if (error) {
1708 		if_printf(&sc->arpcom.ac_if,
1709 			  "could not create spare RX mbuf DMA map\n");
1710 		bus_dma_tag_destroy(ring->data_tag);
1711 		ring->data_tag = NULL;
1712 		return error;
1713 	}
1714 
1715 	for (i = 0; i < sc->sc_rx_ring_count; i++) {
1716 		error = bus_dmamap_create(ring->data_tag, BUS_DMA_WAITOK,
1717 					  &ring->data[i].map);
1718 		if (error) {
1719 			if_printf(&sc->arpcom.ac_if,
1720 				  "could not create %dth RX mbuf DMA mapn", i);
1721 			goto fail;
1722 		}
1723 	}
1724 	return 0;
1725 fail:
1726 	for (j = 0; j < i; ++j)
1727 		bus_dmamap_destroy(ring->data_tag, ring->data[i].map);
1728 	bus_dmamap_destroy(ring->data_tag, ring->data_tmpmap);
1729 	bus_dma_tag_destroy(ring->data_tag);
1730 	ring->data_tag = NULL;
1731 	return error;
1732 }
1733 
1734 static void
1735 nfe_reset_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1736 {
1737 	int i;
1738 
1739 	for (i = 0; i < sc->sc_rx_ring_count; i++) {
1740 		struct nfe_rx_data *data = &ring->data[i];
1741 
1742 		if (data->m != NULL) {
1743 			if ((sc->sc_flags & NFE_F_USE_JUMBO) == 0)
1744 				bus_dmamap_unload(ring->data_tag, data->map);
1745 			m_freem(data->m);
1746 			data->m = NULL;
1747 		}
1748 	}
1749 
1750 	ring->cur = ring->next = 0;
1751 }
1752 
1753 static int
1754 nfe_init_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1755 {
1756 	int i;
1757 
1758 	for (i = 0; i < sc->sc_rx_ring_count; ++i) {
1759 		int error;
1760 
1761 		/* XXX should use a function pointer */
1762 		if (sc->sc_flags & NFE_F_USE_JUMBO)
1763 			error = nfe_newbuf_jumbo(sc, ring, i, 1);
1764 		else
1765 			error = nfe_newbuf_std(sc, ring, i, 1);
1766 		if (error) {
1767 			if_printf(&sc->arpcom.ac_if,
1768 				  "could not allocate RX buffer\n");
1769 			return error;
1770 		}
1771 		nfe_set_ready_rxdesc(sc, ring, i);
1772 	}
1773 	return 0;
1774 }
1775 
1776 static void
1777 nfe_free_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1778 {
1779 	if (ring->data_tag != NULL) {
1780 		struct nfe_rx_data *data;
1781 		int i;
1782 
1783 		for (i = 0; i < sc->sc_rx_ring_count; i++) {
1784 			data = &ring->data[i];
1785 
1786 			if (data->m != NULL) {
1787 				bus_dmamap_unload(ring->data_tag, data->map);
1788 				m_freem(data->m);
1789 			}
1790 			bus_dmamap_destroy(ring->data_tag, data->map);
1791 		}
1792 		bus_dmamap_destroy(ring->data_tag, ring->data_tmpmap);
1793 		bus_dma_tag_destroy(ring->data_tag);
1794 	}
1795 
1796 	nfe_jpool_free(sc, ring);
1797 
1798 	if (ring->jbuf != NULL)
1799 		kfree(ring->jbuf, M_DEVBUF);
1800 	if (ring->data != NULL)
1801 		kfree(ring->data, M_DEVBUF);
1802 
1803 	if (ring->tag != NULL) {
1804 		void *desc;
1805 
1806 		if (sc->sc_caps & NFE_40BIT_ADDR)
1807 			desc = ring->desc64;
1808 		else
1809 			desc = ring->desc32;
1810 
1811 		bus_dmamap_unload(ring->tag, ring->map);
1812 		bus_dmamem_free(ring->tag, desc, ring->map);
1813 		bus_dma_tag_destroy(ring->tag);
1814 	}
1815 }
1816 
1817 static struct nfe_jbuf *
1818 nfe_jalloc(struct nfe_softc *sc)
1819 {
1820 	struct ifnet *ifp = &sc->arpcom.ac_if;
1821 	struct nfe_jbuf *jbuf;
1822 
1823 	lwkt_serialize_enter(&sc->sc_jbuf_serializer);
1824 
1825 	jbuf = SLIST_FIRST(&sc->rxq.jfreelist);
1826 	if (jbuf != NULL) {
1827 		SLIST_REMOVE_HEAD(&sc->rxq.jfreelist, jnext);
1828 		jbuf->inuse = 1;
1829 	} else {
1830 		if_printf(ifp, "no free jumbo buffer\n");
1831 	}
1832 
1833 	lwkt_serialize_exit(&sc->sc_jbuf_serializer);
1834 
1835 	return jbuf;
1836 }
1837 
1838 static void
1839 nfe_jfree(void *arg)
1840 {
1841 	struct nfe_jbuf *jbuf = arg;
1842 	struct nfe_softc *sc = jbuf->sc;
1843 	struct nfe_rx_ring *ring = jbuf->ring;
1844 
1845 	if (&ring->jbuf[jbuf->slot] != jbuf)
1846 		panic("%s: free wrong jumbo buffer", __func__);
1847 	else if (jbuf->inuse == 0)
1848 		panic("%s: jumbo buffer already freed", __func__);
1849 
1850 	lwkt_serialize_enter(&sc->sc_jbuf_serializer);
1851 	atomic_subtract_int(&jbuf->inuse, 1);
1852 	if (jbuf->inuse == 0)
1853 		SLIST_INSERT_HEAD(&ring->jfreelist, jbuf, jnext);
1854 	lwkt_serialize_exit(&sc->sc_jbuf_serializer);
1855 }
1856 
1857 static void
1858 nfe_jref(void *arg)
1859 {
1860 	struct nfe_jbuf *jbuf = arg;
1861 	struct nfe_rx_ring *ring = jbuf->ring;
1862 
1863 	if (&ring->jbuf[jbuf->slot] != jbuf)
1864 		panic("%s: ref wrong jumbo buffer", __func__);
1865 	else if (jbuf->inuse == 0)
1866 		panic("%s: jumbo buffer already freed", __func__);
1867 
1868 	atomic_add_int(&jbuf->inuse, 1);
1869 }
1870 
1871 static int
1872 nfe_jpool_alloc(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1873 {
1874 	struct nfe_jbuf *jbuf;
1875 	bus_dmamem_t dmem;
1876 	bus_addr_t physaddr;
1877 	caddr_t buf;
1878 	int i, error;
1879 
1880 	/*
1881 	 * Allocate a big chunk of DMA'able memory.
1882 	 */
1883 	error = bus_dmamem_coherent(sc->sc_dtag, PAGE_SIZE, 0,
1884 				    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
1885 				    NFE_JPOOL_SIZE(sc),
1886 				    BUS_DMA_WAITOK, &dmem);
1887 	if (error) {
1888 		if_printf(&sc->arpcom.ac_if,
1889 			  "could not create jumbo buffer\n");
1890 		return error;
1891 	}
1892 	ring->jtag = dmem.dmem_tag;
1893 	ring->jmap = dmem.dmem_map;
1894 	ring->jpool = dmem.dmem_addr;
1895 	physaddr = dmem.dmem_busaddr;
1896 
1897 	/* ..and split it into 9KB chunks */
1898 	SLIST_INIT(&ring->jfreelist);
1899 
1900 	buf = ring->jpool;
1901 	for (i = 0; i < NFE_JPOOL_COUNT(sc); i++) {
1902 		jbuf = &ring->jbuf[i];
1903 
1904 		jbuf->sc = sc;
1905 		jbuf->ring = ring;
1906 		jbuf->inuse = 0;
1907 		jbuf->slot = i;
1908 		jbuf->buf = buf;
1909 		jbuf->physaddr = physaddr;
1910 
1911 		SLIST_INSERT_HEAD(&ring->jfreelist, jbuf, jnext);
1912 
1913 		buf += NFE_JBYTES;
1914 		physaddr += NFE_JBYTES;
1915 	}
1916 
1917 	return 0;
1918 }
1919 
1920 static void
1921 nfe_jpool_free(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1922 {
1923 	if (ring->jtag != NULL) {
1924 		bus_dmamap_unload(ring->jtag, ring->jmap);
1925 		bus_dmamem_free(ring->jtag, ring->jpool, ring->jmap);
1926 		bus_dma_tag_destroy(ring->jtag);
1927 	}
1928 }
1929 
1930 static int
1931 nfe_alloc_tx_ring(struct nfe_softc *sc, struct nfe_tx_ring *ring)
1932 {
1933 	int i, j, error, descsize;
1934 	bus_dmamem_t dmem;
1935 	void **desc;
1936 
1937 	if (sc->sc_caps & NFE_40BIT_ADDR) {
1938 		desc = (void *)&ring->desc64;
1939 		descsize = sizeof(struct nfe_desc64);
1940 	} else {
1941 		desc = (void *)&ring->desc32;
1942 		descsize = sizeof(struct nfe_desc32);
1943 	}
1944 
1945 	ring->queued = 0;
1946 	ring->cur = ring->next = 0;
1947 
1948 	error = bus_dmamem_coherent(sc->sc_dtag, PAGE_SIZE, 0,
1949 				    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
1950 				    sc->sc_tx_ring_count * descsize,
1951 				    BUS_DMA_WAITOK | BUS_DMA_ZERO, &dmem);
1952 	if (error) {
1953 		if_printf(&sc->arpcom.ac_if,
1954 			  "could not create TX desc ring\n");
1955 		return error;
1956 	}
1957 	ring->tag = dmem.dmem_tag;
1958 	ring->map = dmem.dmem_map;
1959 	*desc = dmem.dmem_addr;
1960 	ring->physaddr = dmem.dmem_busaddr;
1961 
1962 	ring->data = kmalloc(sizeof(struct nfe_tx_data) * sc->sc_tx_ring_count,
1963 			     M_DEVBUF, M_WAITOK | M_ZERO);
1964 
1965 	error = bus_dma_tag_create(sc->sc_dtag, 1, 0,
1966 			BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
1967 			NULL, NULL,
1968 			NFE_JBYTES, NFE_MAX_SCATTER, MCLBYTES,
1969 			BUS_DMA_ALLOCNOW | BUS_DMA_WAITOK | BUS_DMA_ONEBPAGE,
1970 			&ring->data_tag);
1971 	if (error) {
1972 		if_printf(&sc->arpcom.ac_if,
1973 			  "could not create TX buf DMA tag\n");
1974 		return error;
1975 	}
1976 
1977 	for (i = 0; i < sc->sc_tx_ring_count; i++) {
1978 		error = bus_dmamap_create(ring->data_tag,
1979 				BUS_DMA_WAITOK | BUS_DMA_ONEBPAGE,
1980 				&ring->data[i].map);
1981 		if (error) {
1982 			if_printf(&sc->arpcom.ac_if,
1983 				  "could not create %dth TX buf DMA map\n", i);
1984 			goto fail;
1985 		}
1986 	}
1987 
1988 	return 0;
1989 fail:
1990 	for (j = 0; j < i; ++j)
1991 		bus_dmamap_destroy(ring->data_tag, ring->data[i].map);
1992 	bus_dma_tag_destroy(ring->data_tag);
1993 	ring->data_tag = NULL;
1994 	return error;
1995 }
1996 
1997 static void
1998 nfe_reset_tx_ring(struct nfe_softc *sc, struct nfe_tx_ring *ring)
1999 {
2000 	int i;
2001 
2002 	for (i = 0; i < sc->sc_tx_ring_count; i++) {
2003 		struct nfe_tx_data *data = &ring->data[i];
2004 
2005 		if (sc->sc_caps & NFE_40BIT_ADDR)
2006 			ring->desc64[i].flags = 0;
2007 		else
2008 			ring->desc32[i].flags = 0;
2009 
2010 		if (data->m != NULL) {
2011 			bus_dmamap_unload(ring->data_tag, data->map);
2012 			m_freem(data->m);
2013 			data->m = NULL;
2014 		}
2015 	}
2016 
2017 	ring->queued = 0;
2018 	ring->cur = ring->next = 0;
2019 }
2020 
2021 static int
2022 nfe_init_tx_ring(struct nfe_softc *sc __unused,
2023 		 struct nfe_tx_ring *ring __unused)
2024 {
2025 	return 0;
2026 }
2027 
2028 static void
2029 nfe_free_tx_ring(struct nfe_softc *sc, struct nfe_tx_ring *ring)
2030 {
2031 	if (ring->data_tag != NULL) {
2032 		struct nfe_tx_data *data;
2033 		int i;
2034 
2035 		for (i = 0; i < sc->sc_tx_ring_count; ++i) {
2036 			data = &ring->data[i];
2037 
2038 			if (data->m != NULL) {
2039 				bus_dmamap_unload(ring->data_tag, data->map);
2040 				m_freem(data->m);
2041 			}
2042 			bus_dmamap_destroy(ring->data_tag, data->map);
2043 		}
2044 
2045 		bus_dma_tag_destroy(ring->data_tag);
2046 	}
2047 
2048 	if (ring->data != NULL)
2049 		kfree(ring->data, M_DEVBUF);
2050 
2051 	if (ring->tag != NULL) {
2052 		void *desc;
2053 
2054 		if (sc->sc_caps & NFE_40BIT_ADDR)
2055 			desc = ring->desc64;
2056 		else
2057 			desc = ring->desc32;
2058 
2059 		bus_dmamap_unload(ring->tag, ring->map);
2060 		bus_dmamem_free(ring->tag, desc, ring->map);
2061 		bus_dma_tag_destroy(ring->tag);
2062 	}
2063 }
2064 
2065 static int
2066 nfe_ifmedia_upd(struct ifnet *ifp)
2067 {
2068 	struct nfe_softc *sc = ifp->if_softc;
2069 	struct mii_data *mii = device_get_softc(sc->sc_miibus);
2070 
2071 	ASSERT_SERIALIZED(ifp->if_serializer);
2072 
2073 	if (mii->mii_instance != 0) {
2074 		struct mii_softc *miisc;
2075 
2076 		LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
2077 			mii_phy_reset(miisc);
2078 	}
2079 	mii_mediachg(mii);
2080 
2081 	return 0;
2082 }
2083 
2084 static void
2085 nfe_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
2086 {
2087 	struct nfe_softc *sc = ifp->if_softc;
2088 	struct mii_data *mii = device_get_softc(sc->sc_miibus);
2089 
2090 	ASSERT_SERIALIZED(ifp->if_serializer);
2091 
2092 	mii_pollstat(mii);
2093 	ifmr->ifm_status = mii->mii_media_status;
2094 	ifmr->ifm_active = mii->mii_media_active;
2095 }
2096 
2097 static void
2098 nfe_setmulti(struct nfe_softc *sc)
2099 {
2100 	struct ifnet *ifp = &sc->arpcom.ac_if;
2101 	struct ifmultiaddr *ifma;
2102 	uint8_t addr[ETHER_ADDR_LEN], mask[ETHER_ADDR_LEN];
2103 	uint32_t filter = NFE_RXFILTER_MAGIC;
2104 	int i;
2105 
2106 	if ((ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) != 0) {
2107 		bzero(addr, ETHER_ADDR_LEN);
2108 		bzero(mask, ETHER_ADDR_LEN);
2109 		goto done;
2110 	}
2111 
2112 	bcopy(etherbroadcastaddr, addr, ETHER_ADDR_LEN);
2113 	bcopy(etherbroadcastaddr, mask, ETHER_ADDR_LEN);
2114 
2115 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
2116 		caddr_t maddr;
2117 
2118 		if (ifma->ifma_addr->sa_family != AF_LINK)
2119 			continue;
2120 
2121 		maddr = LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
2122 		for (i = 0; i < ETHER_ADDR_LEN; i++) {
2123 			addr[i] &= maddr[i];
2124 			mask[i] &= ~maddr[i];
2125 		}
2126 	}
2127 
2128 	for (i = 0; i < ETHER_ADDR_LEN; i++)
2129 		mask[i] |= addr[i];
2130 
2131 done:
2132 	addr[0] |= 0x01;	/* make sure multicast bit is set */
2133 
2134 	NFE_WRITE(sc, NFE_MULTIADDR_HI,
2135 	    addr[3] << 24 | addr[2] << 16 | addr[1] << 8 | addr[0]);
2136 	NFE_WRITE(sc, NFE_MULTIADDR_LO,
2137 	    addr[5] <<  8 | addr[4]);
2138 	NFE_WRITE(sc, NFE_MULTIMASK_HI,
2139 	    mask[3] << 24 | mask[2] << 16 | mask[1] << 8 | mask[0]);
2140 	NFE_WRITE(sc, NFE_MULTIMASK_LO,
2141 	    mask[5] <<  8 | mask[4]);
2142 
2143 	filter |= (ifp->if_flags & IFF_PROMISC) ? NFE_PROMISC : NFE_U2M;
2144 	NFE_WRITE(sc, NFE_RXFILTER, filter);
2145 }
2146 
2147 static void
2148 nfe_get_macaddr(struct nfe_softc *sc, uint8_t *addr)
2149 {
2150 	uint32_t lo, hi;
2151 
2152 	lo = NFE_READ(sc, NFE_MACADDR_LO);
2153 	hi = NFE_READ(sc, NFE_MACADDR_HI);
2154 	if (sc->sc_caps & NFE_FIX_EADDR) {
2155 		addr[0] = (lo >> 8) & 0xff;
2156 		addr[1] = (lo & 0xff);
2157 
2158 		addr[2] = (hi >> 24) & 0xff;
2159 		addr[3] = (hi >> 16) & 0xff;
2160 		addr[4] = (hi >>  8) & 0xff;
2161 		addr[5] = (hi & 0xff);
2162 	} else {
2163 		addr[0] = (hi & 0xff);
2164 		addr[1] = (hi >>  8) & 0xff;
2165 		addr[2] = (hi >> 16) & 0xff;
2166 		addr[3] = (hi >> 24) & 0xff;
2167 
2168 		addr[4] = (lo & 0xff);
2169 		addr[5] = (lo >>  8) & 0xff;
2170 	}
2171 }
2172 
2173 static void
2174 nfe_set_macaddr(struct nfe_softc *sc, const uint8_t *addr)
2175 {
2176 	NFE_WRITE(sc, NFE_MACADDR_LO,
2177 	    addr[5] <<  8 | addr[4]);
2178 	NFE_WRITE(sc, NFE_MACADDR_HI,
2179 	    addr[3] << 24 | addr[2] << 16 | addr[1] << 8 | addr[0]);
2180 }
2181 
2182 static void
2183 nfe_tick(void *arg)
2184 {
2185 	struct nfe_softc *sc = arg;
2186 	struct ifnet *ifp = &sc->arpcom.ac_if;
2187 	struct mii_data *mii = device_get_softc(sc->sc_miibus);
2188 
2189 	lwkt_serialize_enter(ifp->if_serializer);
2190 
2191 	mii_tick(mii);
2192 	callout_reset(&sc->sc_tick_ch, hz, nfe_tick, sc);
2193 
2194 	lwkt_serialize_exit(ifp->if_serializer);
2195 }
2196 
2197 static int
2198 nfe_newbuf_std(struct nfe_softc *sc, struct nfe_rx_ring *ring, int idx,
2199 	       int wait)
2200 {
2201 	struct nfe_rx_data *data = &ring->data[idx];
2202 	bus_dma_segment_t seg;
2203 	bus_dmamap_t map;
2204 	struct mbuf *m;
2205 	int nsegs, error;
2206 
2207 	m = m_getcl(wait ? M_WAITOK : M_NOWAIT, MT_DATA, M_PKTHDR);
2208 	if (m == NULL)
2209 		return ENOBUFS;
2210 	m->m_len = m->m_pkthdr.len = MCLBYTES;
2211 
2212 	/*
2213 	 * Aligning the payload improves access times.
2214 	 */
2215 	if (sc->sc_caps & NFE_WORDALIGN)
2216 		m_adj(m, ETHER_ALIGN);
2217 
2218 	error = bus_dmamap_load_mbuf_segment(ring->data_tag, ring->data_tmpmap,
2219 			m, &seg, 1, &nsegs, BUS_DMA_NOWAIT);
2220 	if (error) {
2221 		m_freem(m);
2222 		if (wait) {
2223 			if_printf(&sc->arpcom.ac_if,
2224 				  "could map RX mbuf %d\n", error);
2225 		}
2226 		return error;
2227 	}
2228 
2229 	if (data->m != NULL) {
2230 		/* Sync and unload originally mapped mbuf */
2231 		bus_dmamap_sync(ring->data_tag, data->map,
2232 				BUS_DMASYNC_POSTREAD);
2233 		bus_dmamap_unload(ring->data_tag, data->map);
2234 	}
2235 
2236 	/* Swap this DMA map with tmp DMA map */
2237 	map = data->map;
2238 	data->map = ring->data_tmpmap;
2239 	ring->data_tmpmap = map;
2240 
2241 	/* Caller is assumed to have collected the old mbuf */
2242 	data->m = m;
2243 
2244 	nfe_set_paddr_rxdesc(sc, ring, idx, seg.ds_addr);
2245 	return 0;
2246 }
2247 
2248 static int
2249 nfe_newbuf_jumbo(struct nfe_softc *sc, struct nfe_rx_ring *ring, int idx,
2250 		 int wait)
2251 {
2252 	struct nfe_rx_data *data = &ring->data[idx];
2253 	struct nfe_jbuf *jbuf;
2254 	struct mbuf *m;
2255 
2256 	MGETHDR(m, wait ? M_WAITOK : M_NOWAIT, MT_DATA);
2257 	if (m == NULL)
2258 		return ENOBUFS;
2259 
2260 	jbuf = nfe_jalloc(sc);
2261 	if (jbuf == NULL) {
2262 		m_freem(m);
2263 		if_printf(&sc->arpcom.ac_if, "jumbo allocation failed "
2264 		    "-- packet dropped!\n");
2265 		return ENOBUFS;
2266 	}
2267 
2268 	m->m_ext.ext_arg = jbuf;
2269 	m->m_ext.ext_buf = jbuf->buf;
2270 	m->m_ext.ext_free = nfe_jfree;
2271 	m->m_ext.ext_ref = nfe_jref;
2272 	m->m_ext.ext_size = NFE_JBYTES;
2273 
2274 	m->m_data = m->m_ext.ext_buf;
2275 	m->m_flags |= M_EXT;
2276 	m->m_len = m->m_pkthdr.len = m->m_ext.ext_size;
2277 
2278 	/*
2279 	 * Aligning the payload improves access times.
2280 	 */
2281 	if (sc->sc_caps & NFE_WORDALIGN)
2282 		m_adj(m, ETHER_ALIGN);
2283 
2284 	/* Caller is assumed to have collected the old mbuf */
2285 	data->m = m;
2286 
2287 	nfe_set_paddr_rxdesc(sc, ring, idx, jbuf->physaddr);
2288 	return 0;
2289 }
2290 
2291 static void
2292 nfe_set_paddr_rxdesc(struct nfe_softc *sc, struct nfe_rx_ring *ring, int idx,
2293 		     bus_addr_t physaddr)
2294 {
2295 	if (sc->sc_caps & NFE_40BIT_ADDR) {
2296 		struct nfe_desc64 *desc64 = &ring->desc64[idx];
2297 
2298 		desc64->physaddr[0] = htole32(NFE_ADDR_HI(physaddr));
2299 		desc64->physaddr[1] = htole32(NFE_ADDR_LO(physaddr));
2300 	} else {
2301 		struct nfe_desc32 *desc32 = &ring->desc32[idx];
2302 
2303 		desc32->physaddr = htole32(physaddr);
2304 	}
2305 }
2306 
2307 static void
2308 nfe_set_ready_rxdesc(struct nfe_softc *sc, struct nfe_rx_ring *ring, int idx)
2309 {
2310 	if (sc->sc_caps & NFE_40BIT_ADDR) {
2311 		struct nfe_desc64 *desc64 = &ring->desc64[idx];
2312 
2313 		desc64->length = htole16(ring->bufsz);
2314 		desc64->flags = htole16(NFE_RX_READY);
2315 	} else {
2316 		struct nfe_desc32 *desc32 = &ring->desc32[idx];
2317 
2318 		desc32->length = htole16(ring->bufsz);
2319 		desc32->flags = htole16(NFE_RX_READY);
2320 	}
2321 }
2322 
2323 static int
2324 nfe_sysctl_imtime(SYSCTL_HANDLER_ARGS)
2325 {
2326 	struct nfe_softc *sc = arg1;
2327 	struct ifnet *ifp = &sc->arpcom.ac_if;
2328 	uint32_t flags;
2329 	int error, v;
2330 
2331 	lwkt_serialize_enter(ifp->if_serializer);
2332 
2333 	flags = sc->sc_flags & ~NFE_F_DYN_IM;
2334 	v = sc->sc_imtime;
2335 	if (sc->sc_flags & NFE_F_DYN_IM)
2336 		v = -v;
2337 
2338 	error = sysctl_handle_int(oidp, &v, 0, req);
2339 	if (error || req->newptr == NULL)
2340 		goto back;
2341 
2342 	if (v < 0) {
2343 		flags |= NFE_F_DYN_IM;
2344 		v = -v;
2345 	}
2346 
2347 	if (v != sc->sc_imtime || (flags ^ sc->sc_flags)) {
2348 		if (NFE_IMTIME(v) == 0)
2349 			v = 0;
2350 		sc->sc_imtime = v;
2351 		sc->sc_flags = flags;
2352 		sc->sc_irq_enable = NFE_IRQ_ENABLE(sc);
2353 
2354 		if ((ifp->if_flags & (IFF_NPOLLING | IFF_RUNNING))
2355 		    == IFF_RUNNING) {
2356 			nfe_enable_intrs(sc);
2357 		}
2358 	}
2359 back:
2360 	lwkt_serialize_exit(ifp->if_serializer);
2361 	return error;
2362 }
2363 
2364 static void
2365 nfe_powerup(device_t dev)
2366 {
2367 	struct nfe_softc *sc = device_get_softc(dev);
2368 	uint32_t pwr_state;
2369 	uint16_t did;
2370 
2371 	/*
2372 	 * Bring MAC and PHY out of low power state
2373 	 */
2374 
2375 	pwr_state = NFE_READ(sc, NFE_PWR_STATE2) & ~NFE_PWRUP_MASK;
2376 
2377 	did = pci_get_device(dev);
2378 	if ((did == PCI_PRODUCT_NVIDIA_MCP51_LAN1 ||
2379 	     did == PCI_PRODUCT_NVIDIA_MCP51_LAN2) &&
2380 	    pci_get_revid(dev) >= 0xa3)
2381 		pwr_state |= NFE_PWRUP_REV_A3;
2382 
2383 	NFE_WRITE(sc, NFE_PWR_STATE2, pwr_state);
2384 }
2385 
2386 static void
2387 nfe_mac_reset(struct nfe_softc *sc)
2388 {
2389 	uint32_t rxtxctl = sc->rxtxctl_desc | NFE_RXTX_BIT2;
2390 	uint32_t macaddr_hi, macaddr_lo, tx_poll;
2391 
2392 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_RESET | rxtxctl);
2393 
2394 	/* Save several registers for later restoration */
2395 	macaddr_hi = NFE_READ(sc, NFE_MACADDR_HI);
2396 	macaddr_lo = NFE_READ(sc, NFE_MACADDR_LO);
2397 	tx_poll = NFE_READ(sc, NFE_TX_POLL);
2398 
2399 	NFE_WRITE(sc, NFE_MAC_RESET, NFE_RESET_ASSERT);
2400 	DELAY(100);
2401 
2402 	NFE_WRITE(sc, NFE_MAC_RESET, 0);
2403 	DELAY(100);
2404 
2405 	/* Restore saved registers */
2406 	NFE_WRITE(sc, NFE_MACADDR_HI, macaddr_hi);
2407 	NFE_WRITE(sc, NFE_MACADDR_LO, macaddr_lo);
2408 	NFE_WRITE(sc, NFE_TX_POLL, tx_poll);
2409 
2410 	NFE_WRITE(sc, NFE_RXTX_CTL, rxtxctl);
2411 }
2412 
2413 static void
2414 nfe_enable_intrs(struct nfe_softc *sc)
2415 {
2416 	/*
2417 	 * NFE_IMTIMER generates a periodic interrupt via NFE_IRQ_TIMER.
2418 	 * It is unclear how wide the timer is.  Base programming does
2419 	 * not seem to effect NFE_IRQ_TX_DONE or NFE_IRQ_RX_DONE so
2420 	 * we don't get any interrupt moderation.  TX moderation is
2421 	 * possible by using the timer interrupt instead of TX_DONE.
2422 	 *
2423 	 * It is unclear whether there are other bits that can be
2424 	 * set to make the NFE device actually do interrupt moderation
2425 	 * on the RX side.
2426 	 *
2427 	 * For now set a 128uS interval as a placemark, but don't use
2428 	 * the timer.
2429 	 */
2430 	if (sc->sc_imtime == 0)
2431 		NFE_WRITE(sc, NFE_IMTIMER, NFE_IMTIME_DEFAULT);
2432 	else
2433 		NFE_WRITE(sc, NFE_IMTIMER, NFE_IMTIME(sc->sc_imtime));
2434 
2435 	/* Enable interrupts */
2436 	NFE_WRITE(sc, NFE_IRQ_MASK, sc->sc_irq_enable);
2437 
2438 	if (sc->sc_irq_enable & NFE_IRQ_TIMER)
2439 		sc->sc_flags |= NFE_F_IRQ_TIMER;
2440 	else
2441 		sc->sc_flags &= ~NFE_F_IRQ_TIMER;
2442 }
2443