xref: /dragonfly/sys/dev/netif/nfe/if_nfe.c (revision c03f08f3)
1 /*	$OpenBSD: if_nfe.c,v 1.63 2006/06/17 18:00:43 brad Exp $	*/
2 /*	$DragonFly: src/sys/dev/netif/nfe/if_nfe.c,v 1.16 2007/08/14 13:30:35 sephe Exp $	*/
3 
4 /*
5  * Copyright (c) 2006 The DragonFly Project.  All rights reserved.
6  *
7  * This code is derived from software contributed to The DragonFly Project
8  * by Sepherosa Ziehau <sepherosa@gmail.com> and
9  * Matthew Dillon <dillon@apollo.backplane.com>
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  *
15  * 1. Redistributions of source code must retain the above copyright
16  *    notice, this list of conditions and the following disclaimer.
17  * 2. Redistributions in binary form must reproduce the above copyright
18  *    notice, this list of conditions and the following disclaimer in
19  *    the documentation and/or other materials provided with the
20  *    distribution.
21  * 3. Neither the name of The DragonFly Project nor the names of its
22  *    contributors may be used to endorse or promote products derived
23  *    from this software without specific, prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
26  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
27  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
28  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
29  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
30  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
31  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
32  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
33  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
34  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
35  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  */
38 
39 /*
40  * Copyright (c) 2006 Damien Bergamini <damien.bergamini@free.fr>
41  * Copyright (c) 2005, 2006 Jonathan Gray <jsg@openbsd.org>
42  *
43  * Permission to use, copy, modify, and distribute this software for any
44  * purpose with or without fee is hereby granted, provided that the above
45  * copyright notice and this permission notice appear in all copies.
46  *
47  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
48  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
49  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
50  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
51  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
52  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
53  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
54  */
55 
56 /* Driver for NVIDIA nForce MCP Fast Ethernet and Gigabit Ethernet */
57 
58 #include "opt_polling.h"
59 
60 #include <sys/param.h>
61 #include <sys/endian.h>
62 #include <sys/kernel.h>
63 #include <sys/bus.h>
64 #include <sys/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 
82 #include <bus/pci/pcireg.h>
83 #include <bus/pci/pcivar.h>
84 #include <bus/pci/pcidevs.h>
85 
86 #include <dev/netif/mii_layer/mii.h>
87 #include <dev/netif/mii_layer/miivar.h>
88 
89 #include "miibus_if.h"
90 
91 #include <dev/netif/nfe/if_nfereg.h>
92 #include <dev/netif/nfe/if_nfevar.h>
93 
94 #define NFE_CSUM
95 #define NFE_CSUM_FEATURES	(CSUM_IP | CSUM_TCP | CSUM_UDP)
96 
97 static int	nfe_probe(device_t);
98 static int	nfe_attach(device_t);
99 static int	nfe_detach(device_t);
100 static void	nfe_shutdown(device_t);
101 static int	nfe_resume(device_t);
102 static int	nfe_suspend(device_t);
103 
104 static int	nfe_miibus_readreg(device_t, int, int);
105 static void	nfe_miibus_writereg(device_t, int, int, int);
106 static void	nfe_miibus_statchg(device_t);
107 
108 #ifdef DEVICE_POLLING
109 static void	nfe_poll(struct ifnet *, enum poll_cmd, int);
110 #endif
111 static void	nfe_intr(void *);
112 static int	nfe_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
113 static void	nfe_rxeof(struct nfe_softc *);
114 static void	nfe_txeof(struct nfe_softc *);
115 static int	nfe_encap(struct nfe_softc *, struct nfe_tx_ring *,
116 			  struct mbuf *);
117 static void	nfe_start(struct ifnet *);
118 static void	nfe_watchdog(struct ifnet *);
119 static void	nfe_init(void *);
120 static void	nfe_stop(struct nfe_softc *);
121 static struct nfe_jbuf *nfe_jalloc(struct nfe_softc *);
122 static void	nfe_jfree(void *);
123 static void	nfe_jref(void *);
124 static int	nfe_jpool_alloc(struct nfe_softc *, struct nfe_rx_ring *);
125 static void	nfe_jpool_free(struct nfe_softc *, struct nfe_rx_ring *);
126 static int	nfe_alloc_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
127 static void	nfe_reset_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
128 static int	nfe_init_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
129 static void	nfe_free_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
130 static int	nfe_alloc_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
131 static void	nfe_reset_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
132 static int	nfe_init_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
133 static void	nfe_free_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
134 static int	nfe_ifmedia_upd(struct ifnet *);
135 static void	nfe_ifmedia_sts(struct ifnet *, struct ifmediareq *);
136 static void	nfe_setmulti(struct nfe_softc *);
137 static void	nfe_get_macaddr(struct nfe_softc *, uint8_t *);
138 static void	nfe_set_macaddr(struct nfe_softc *, const uint8_t *);
139 static void	nfe_tick(void *);
140 static void	nfe_ring_dma_addr(void *, bus_dma_segment_t *, int, int);
141 static void	nfe_buf_dma_addr(void *, bus_dma_segment_t *, int, bus_size_t,
142 				 int);
143 static void	nfe_set_paddr_rxdesc(struct nfe_softc *, struct nfe_rx_ring *,
144 				     int, bus_addr_t);
145 static void	nfe_set_ready_rxdesc(struct nfe_softc *, struct nfe_rx_ring *,
146 				     int);
147 static int	nfe_newbuf_std(struct nfe_softc *, struct nfe_rx_ring *, int,
148 			       int);
149 static int	nfe_newbuf_jumbo(struct nfe_softc *, struct nfe_rx_ring *, int,
150 				 int);
151 
152 #define NFE_DEBUG
153 #ifdef NFE_DEBUG
154 
155 static int	nfe_debug = 0;
156 static int	nfe_rx_ring_count = NFE_RX_RING_DEF_COUNT;
157 
158 TUNABLE_INT("hw.nfe.rx_ring_count", &nfe_rx_ring_count);
159 
160 SYSCTL_NODE(_hw, OID_AUTO, nfe, CTLFLAG_RD, 0, "nVidia GigE parameters");
161 SYSCTL_INT(_hw_nfe, OID_AUTO, rx_ring_count, CTLFLAG_RD, &nfe_rx_ring_count,
162 	   NFE_RX_RING_DEF_COUNT, "rx ring count");
163 SYSCTL_INT(_hw_nfe, OID_AUTO, debug, CTLFLAG_RW, &nfe_debug, 0,
164 	   "control debugging printfs");
165 
166 #define DPRINTF(sc, fmt, ...) do {		\
167 	if (nfe_debug) {			\
168 		if_printf(&(sc)->arpcom.ac_if,	\
169 			  fmt, __VA_ARGS__);	\
170 	}					\
171 } while (0)
172 
173 #define DPRINTFN(sc, lv, fmt, ...) do {		\
174 	if (nfe_debug >= (lv)) {		\
175 		if_printf(&(sc)->arpcom.ac_if,	\
176 			  fmt, __VA_ARGS__);	\
177 	}					\
178 } while (0)
179 
180 #else	/* !NFE_DEBUG */
181 
182 #define DPRINTF(sc, fmt, ...)
183 #define DPRINTFN(sc, lv, fmt, ...)
184 
185 #endif	/* NFE_DEBUG */
186 
187 struct nfe_dma_ctx {
188 	int			nsegs;
189 	bus_dma_segment_t	*segs;
190 };
191 
192 static const struct nfe_dev {
193 	uint16_t	vid;
194 	uint16_t	did;
195 	const char	*desc;
196 } nfe_devices[] = {
197 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE_LAN,
198 	  "NVIDIA nForce Fast Ethernet" },
199 
200 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE2_LAN,
201 	  "NVIDIA nForce2 Fast Ethernet" },
202 
203 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN1,
204 	  "NVIDIA nForce3 Gigabit Ethernet" },
205 
206 	/* XXX TGEN the next chip can also be found in the nForce2 Ultra 400Gb
207 	   chipset, and possibly also the 400R; it might be both nForce2- and
208 	   nForce3-based boards can use the same MCPs (= southbridges) */
209 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN2,
210 	  "NVIDIA nForce3 Gigabit Ethernet" },
211 
212 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN3,
213 	  "NVIDIA nForce3 Gigabit Ethernet" },
214 
215 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN4,
216 	  "NVIDIA nForce3 Gigabit Ethernet" },
217 
218 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN5,
219 	  "NVIDIA nForce3 Gigabit Ethernet" },
220 
221 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_CK804_LAN1,
222 	  "NVIDIA CK804 Gigabit Ethernet" },
223 
224 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_CK804_LAN2,
225 	  "NVIDIA CK804 Gigabit Ethernet" },
226 
227 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP04_LAN1,
228 	  "NVIDIA MCP04 Gigabit Ethernet" },
229 
230 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP04_LAN2,
231 	  "NVIDIA MCP04 Gigabit Ethernet" },
232 
233 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP51_LAN1,
234 	  "NVIDIA MCP51 Gigabit Ethernet" },
235 
236 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP51_LAN2,
237 	  "NVIDIA MCP51 Gigabit Ethernet" },
238 
239 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP55_LAN1,
240 	  "NVIDIA MCP55 Gigabit Ethernet" },
241 
242 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP55_LAN2,
243 	  "NVIDIA MCP55 Gigabit Ethernet" },
244 
245 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN1,
246 	  "NVIDIA MCP61 Gigabit Ethernet" },
247 
248 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN2,
249 	  "NVIDIA MCP61 Gigabit Ethernet" },
250 
251 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN3,
252 	  "NVIDIA MCP61 Gigabit Ethernet" },
253 
254 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN4,
255 	  "NVIDIA MCP61 Gigabit Ethernet" },
256 
257 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN1,
258 	  "NVIDIA MCP65 Gigabit Ethernet" },
259 
260 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN2,
261 	  "NVIDIA MCP65 Gigabit Ethernet" },
262 
263 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN3,
264 	  "NVIDIA MCP65 Gigabit Ethernet" },
265 
266 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN4,
267 	  "NVIDIA MCP65 Gigabit Ethernet" },
268 
269 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN1,
270 	  "NVIDIA MCP67 Gigabit Ethernet" },
271 
272 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN2,
273 	  "NVIDIA MCP67 Gigabit Ethernet" },
274 
275 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN3,
276 	  "NVIDIA MCP67 Gigabit Ethernet" },
277 
278 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN4,
279 	  "NVIDIA MCP67 Gigabit Ethernet" }
280 };
281 
282 static device_method_t nfe_methods[] = {
283 	/* Device interface */
284 	DEVMETHOD(device_probe,		nfe_probe),
285 	DEVMETHOD(device_attach,	nfe_attach),
286 	DEVMETHOD(device_detach,	nfe_detach),
287 	DEVMETHOD(device_suspend,	nfe_suspend),
288 	DEVMETHOD(device_resume,	nfe_resume),
289 	DEVMETHOD(device_shutdown,	nfe_shutdown),
290 
291 	/* Bus interface */
292 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
293 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
294 
295 	/* MII interface */
296 	DEVMETHOD(miibus_readreg,	nfe_miibus_readreg),
297 	DEVMETHOD(miibus_writereg,	nfe_miibus_writereg),
298 	DEVMETHOD(miibus_statchg,	nfe_miibus_statchg),
299 
300 	{ 0, 0 }
301 };
302 
303 static driver_t nfe_driver = {
304 	"nfe",
305 	nfe_methods,
306 	sizeof(struct nfe_softc)
307 };
308 
309 static devclass_t	nfe_devclass;
310 
311 DECLARE_DUMMY_MODULE(if_nfe);
312 MODULE_DEPEND(if_nfe, miibus, 1, 1, 1);
313 DRIVER_MODULE(if_nfe, pci, nfe_driver, nfe_devclass, 0, 0);
314 DRIVER_MODULE(miibus, nfe, miibus_driver, miibus_devclass, 0, 0);
315 
316 static int
317 nfe_probe(device_t dev)
318 {
319 	const struct nfe_dev *n;
320 	uint16_t vid, did;
321 
322 	vid = pci_get_vendor(dev);
323 	did = pci_get_device(dev);
324 	for (n = nfe_devices; n->desc != NULL; ++n) {
325 		if (vid == n->vid && did == n->did) {
326 			struct nfe_softc *sc = device_get_softc(dev);
327 
328 			switch (did) {
329 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN2:
330 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN3:
331 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN4:
332 			case PCI_PRODUCT_NVIDIA_NFORCE3_LAN5:
333 				sc->sc_flags = NFE_JUMBO_SUP |
334 					       NFE_HW_CSUM;
335 				break;
336 			case PCI_PRODUCT_NVIDIA_MCP51_LAN1:
337 			case PCI_PRODUCT_NVIDIA_MCP51_LAN2:
338 			case PCI_PRODUCT_NVIDIA_MCP61_LAN1:
339 			case PCI_PRODUCT_NVIDIA_MCP61_LAN2:
340 			case PCI_PRODUCT_NVIDIA_MCP61_LAN3:
341 			case PCI_PRODUCT_NVIDIA_MCP61_LAN4:
342 			case PCI_PRODUCT_NVIDIA_MCP67_LAN1:
343 			case PCI_PRODUCT_NVIDIA_MCP67_LAN2:
344 			case PCI_PRODUCT_NVIDIA_MCP67_LAN3:
345 			case PCI_PRODUCT_NVIDIA_MCP67_LAN4:
346 				sc->sc_flags = NFE_40BIT_ADDR;
347 				break;
348 			case PCI_PRODUCT_NVIDIA_CK804_LAN1:
349 			case PCI_PRODUCT_NVIDIA_CK804_LAN2:
350 			case PCI_PRODUCT_NVIDIA_MCP04_LAN1:
351 			case PCI_PRODUCT_NVIDIA_MCP04_LAN2:
352 			case PCI_PRODUCT_NVIDIA_MCP65_LAN1:
353 			case PCI_PRODUCT_NVIDIA_MCP65_LAN2:
354 			case PCI_PRODUCT_NVIDIA_MCP65_LAN3:
355 			case PCI_PRODUCT_NVIDIA_MCP65_LAN4:
356 				sc->sc_flags = NFE_JUMBO_SUP |
357 					       NFE_40BIT_ADDR |
358 					       NFE_HW_CSUM;
359 				break;
360 			case PCI_PRODUCT_NVIDIA_MCP55_LAN1:
361 			case PCI_PRODUCT_NVIDIA_MCP55_LAN2:
362 				sc->sc_flags = NFE_JUMBO_SUP |
363 					       NFE_40BIT_ADDR |
364 					       NFE_HW_CSUM |
365 					       NFE_HW_VLAN;
366 				break;
367 			}
368 
369 			device_set_desc(dev, n->desc);
370 			device_set_async_attach(dev, TRUE);
371 			return 0;
372 		}
373 	}
374 	return ENXIO;
375 }
376 
377 static int
378 nfe_attach(device_t dev)
379 {
380 	struct nfe_softc *sc = device_get_softc(dev);
381 	struct ifnet *ifp = &sc->arpcom.ac_if;
382 	uint8_t eaddr[ETHER_ADDR_LEN];
383 	int error;
384 
385 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
386 	lwkt_serialize_init(&sc->sc_jbuf_serializer);
387 
388 	sc->sc_mem_rid = PCIR_BAR(0);
389 
390 #ifndef BURN_BRIDGES
391 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
392 		uint32_t mem, irq;
393 
394 		mem = pci_read_config(dev, sc->sc_mem_rid, 4);
395 		irq = pci_read_config(dev, PCIR_INTLINE, 4);
396 
397 		device_printf(dev, "chip is in D%d power mode "
398 		    "-- setting to D0\n", pci_get_powerstate(dev));
399 
400 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
401 
402 		pci_write_config(dev, sc->sc_mem_rid, mem, 4);
403 		pci_write_config(dev, PCIR_INTLINE, irq, 4);
404 	}
405 #endif	/* !BURN_BRIDGE */
406 
407 	/* Enable bus mastering */
408 	pci_enable_busmaster(dev);
409 
410 	/* Allocate IO memory */
411 	sc->sc_mem_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
412 						&sc->sc_mem_rid, RF_ACTIVE);
413 	if (sc->sc_mem_res == NULL) {
414 		device_printf(dev, "cound not allocate io memory\n");
415 		return ENXIO;
416 	}
417 	sc->sc_memh = rman_get_bushandle(sc->sc_mem_res);
418 	sc->sc_memt = rman_get_bustag(sc->sc_mem_res);
419 
420 	/* Allocate IRQ */
421 	sc->sc_irq_rid = 0;
422 	sc->sc_irq_res = bus_alloc_resource_any(dev, SYS_RES_IRQ,
423 						&sc->sc_irq_rid,
424 						RF_SHAREABLE | RF_ACTIVE);
425 	if (sc->sc_irq_res == NULL) {
426 		device_printf(dev, "could not allocate irq\n");
427 		error = ENXIO;
428 		goto fail;
429 	}
430 
431 	nfe_get_macaddr(sc, eaddr);
432 
433 	/*
434 	 * Allocate Tx and Rx rings.
435 	 */
436 	error = nfe_alloc_tx_ring(sc, &sc->txq);
437 	if (error) {
438 		device_printf(dev, "could not allocate Tx ring\n");
439 		goto fail;
440 	}
441 
442 	error = nfe_alloc_rx_ring(sc, &sc->rxq);
443 	if (error) {
444 		device_printf(dev, "could not allocate Rx ring\n");
445 		goto fail;
446 	}
447 
448 	error = mii_phy_probe(dev, &sc->sc_miibus, nfe_ifmedia_upd,
449 			      nfe_ifmedia_sts);
450 	if (error) {
451 		device_printf(dev, "MII without any phy\n");
452 		goto fail;
453 	}
454 
455 	ifp->if_softc = sc;
456 	ifp->if_mtu = ETHERMTU;
457 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
458 	ifp->if_ioctl = nfe_ioctl;
459 	ifp->if_start = nfe_start;
460 #ifdef DEVICE_POLLING
461 	ifp->if_poll = nfe_poll;
462 #endif
463 	ifp->if_watchdog = nfe_watchdog;
464 	ifp->if_init = nfe_init;
465 	ifq_set_maxlen(&ifp->if_snd, NFE_IFQ_MAXLEN);
466 	ifq_set_ready(&ifp->if_snd);
467 
468 	ifp->if_capabilities = IFCAP_VLAN_MTU;
469 
470 	if (sc->sc_flags & NFE_HW_VLAN)
471 		ifp->if_capabilities |= IFCAP_VLAN_HWTAGGING;
472 
473 #ifdef NFE_CSUM
474 	if (sc->sc_flags & NFE_HW_CSUM) {
475 		ifp->if_capabilities |= IFCAP_HWCSUM;
476 		ifp->if_hwassist = NFE_CSUM_FEATURES;
477 	}
478 #else
479 	sc->sc_flags &= ~NFE_HW_CSUM;
480 #endif
481 	ifp->if_capenable = ifp->if_capabilities;
482 
483 	callout_init(&sc->sc_tick_ch);
484 
485 	ether_ifattach(ifp, eaddr, NULL);
486 
487 	error = bus_setup_intr(dev, sc->sc_irq_res, INTR_MPSAFE, nfe_intr, sc,
488 			       &sc->sc_ih, ifp->if_serializer);
489 	if (error) {
490 		device_printf(dev, "could not setup intr\n");
491 		ether_ifdetach(ifp);
492 		goto fail;
493 	}
494 
495 	return 0;
496 fail:
497 	nfe_detach(dev);
498 	return error;
499 }
500 
501 static int
502 nfe_detach(device_t dev)
503 {
504 	struct nfe_softc *sc = device_get_softc(dev);
505 
506 	if (device_is_attached(dev)) {
507 		struct ifnet *ifp = &sc->arpcom.ac_if;
508 
509 		lwkt_serialize_enter(ifp->if_serializer);
510 		nfe_stop(sc);
511 		bus_teardown_intr(dev, sc->sc_irq_res, sc->sc_ih);
512 		lwkt_serialize_exit(ifp->if_serializer);
513 
514 		ether_ifdetach(ifp);
515 	}
516 
517 	if (sc->sc_miibus != NULL)
518 		device_delete_child(dev, sc->sc_miibus);
519 	bus_generic_detach(dev);
520 
521 	if (sc->sc_irq_res != NULL) {
522 		bus_release_resource(dev, SYS_RES_IRQ, sc->sc_irq_rid,
523 				     sc->sc_irq_res);
524 	}
525 
526 	if (sc->sc_mem_res != NULL) {
527 		bus_release_resource(dev, SYS_RES_MEMORY, sc->sc_mem_rid,
528 				     sc->sc_mem_res);
529 	}
530 
531 	nfe_free_tx_ring(sc, &sc->txq);
532 	nfe_free_rx_ring(sc, &sc->rxq);
533 
534 	return 0;
535 }
536 
537 static void
538 nfe_shutdown(device_t dev)
539 {
540 	struct nfe_softc *sc = device_get_softc(dev);
541 	struct ifnet *ifp = &sc->arpcom.ac_if;
542 
543 	lwkt_serialize_enter(ifp->if_serializer);
544 	nfe_stop(sc);
545 	lwkt_serialize_exit(ifp->if_serializer);
546 }
547 
548 static int
549 nfe_suspend(device_t dev)
550 {
551 	struct nfe_softc *sc = device_get_softc(dev);
552 	struct ifnet *ifp = &sc->arpcom.ac_if;
553 
554 	lwkt_serialize_enter(ifp->if_serializer);
555 	nfe_stop(sc);
556 	lwkt_serialize_exit(ifp->if_serializer);
557 
558 	return 0;
559 }
560 
561 static int
562 nfe_resume(device_t dev)
563 {
564 	struct nfe_softc *sc = device_get_softc(dev);
565 	struct ifnet *ifp = &sc->arpcom.ac_if;
566 
567 	lwkt_serialize_enter(ifp->if_serializer);
568 	if (ifp->if_flags & IFF_UP)
569 		nfe_init(sc);
570 	lwkt_serialize_exit(ifp->if_serializer);
571 
572 	return 0;
573 }
574 
575 static void
576 nfe_miibus_statchg(device_t dev)
577 {
578 	struct nfe_softc *sc = device_get_softc(dev);
579 	struct mii_data *mii = device_get_softc(sc->sc_miibus);
580 	uint32_t phy, seed, misc = NFE_MISC1_MAGIC, link = NFE_MEDIA_SET;
581 
582 	phy = NFE_READ(sc, NFE_PHY_IFACE);
583 	phy &= ~(NFE_PHY_HDX | NFE_PHY_100TX | NFE_PHY_1000T);
584 
585 	seed = NFE_READ(sc, NFE_RNDSEED);
586 	seed &= ~NFE_SEED_MASK;
587 
588 	if ((mii->mii_media_active & IFM_GMASK) == IFM_HDX) {
589 		phy  |= NFE_PHY_HDX;	/* half-duplex */
590 		misc |= NFE_MISC1_HDX;
591 	}
592 
593 	switch (IFM_SUBTYPE(mii->mii_media_active)) {
594 	case IFM_1000_T:	/* full-duplex only */
595 		link |= NFE_MEDIA_1000T;
596 		seed |= NFE_SEED_1000T;
597 		phy  |= NFE_PHY_1000T;
598 		break;
599 	case IFM_100_TX:
600 		link |= NFE_MEDIA_100TX;
601 		seed |= NFE_SEED_100TX;
602 		phy  |= NFE_PHY_100TX;
603 		break;
604 	case IFM_10_T:
605 		link |= NFE_MEDIA_10T;
606 		seed |= NFE_SEED_10T;
607 		break;
608 	}
609 
610 	NFE_WRITE(sc, NFE_RNDSEED, seed);	/* XXX: gigabit NICs only? */
611 
612 	NFE_WRITE(sc, NFE_PHY_IFACE, phy);
613 	NFE_WRITE(sc, NFE_MISC1, misc);
614 	NFE_WRITE(sc, NFE_LINKSPEED, link);
615 }
616 
617 static int
618 nfe_miibus_readreg(device_t dev, int phy, int reg)
619 {
620 	struct nfe_softc *sc = device_get_softc(dev);
621 	uint32_t val;
622 	int ntries;
623 
624 	NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
625 
626 	if (NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY) {
627 		NFE_WRITE(sc, NFE_PHY_CTL, NFE_PHY_BUSY);
628 		DELAY(100);
629 	}
630 
631 	NFE_WRITE(sc, NFE_PHY_CTL, (phy << NFE_PHYADD_SHIFT) | reg);
632 
633 	for (ntries = 0; ntries < 1000; ntries++) {
634 		DELAY(100);
635 		if (!(NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY))
636 			break;
637 	}
638 	if (ntries == 1000) {
639 		DPRINTFN(sc, 2, "timeout waiting for PHY %s\n", "");
640 		return 0;
641 	}
642 
643 	if (NFE_READ(sc, NFE_PHY_STATUS) & NFE_PHY_ERROR) {
644 		DPRINTFN(sc, 2, "could not read PHY %s\n", "");
645 		return 0;
646 	}
647 
648 	val = NFE_READ(sc, NFE_PHY_DATA);
649 	if (val != 0xffffffff && val != 0)
650 		sc->mii_phyaddr = phy;
651 
652 	DPRINTFN(sc, 2, "mii read phy %d reg 0x%x ret 0x%x\n", phy, reg, val);
653 
654 	return val;
655 }
656 
657 static void
658 nfe_miibus_writereg(device_t dev, int phy, int reg, int val)
659 {
660 	struct nfe_softc *sc = device_get_softc(dev);
661 	uint32_t ctl;
662 	int ntries;
663 
664 	NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
665 
666 	if (NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY) {
667 		NFE_WRITE(sc, NFE_PHY_CTL, NFE_PHY_BUSY);
668 		DELAY(100);
669 	}
670 
671 	NFE_WRITE(sc, NFE_PHY_DATA, val);
672 	ctl = NFE_PHY_WRITE | (phy << NFE_PHYADD_SHIFT) | reg;
673 	NFE_WRITE(sc, NFE_PHY_CTL, ctl);
674 
675 	for (ntries = 0; ntries < 1000; ntries++) {
676 		DELAY(100);
677 		if (!(NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY))
678 			break;
679 	}
680 
681 #ifdef NFE_DEBUG
682 	if (ntries == 1000)
683 		DPRINTFN(sc, 2, "could not write to PHY %s\n", "");
684 #endif
685 }
686 
687 #ifdef DEVICE_POLLING
688 
689 static void
690 nfe_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
691 {
692 	struct nfe_softc *sc = ifp->if_softc;
693 
694 	switch(cmd) {
695 	case POLL_REGISTER:
696 		/* Disable interrupts */
697 		NFE_WRITE(sc, NFE_IRQ_MASK, 0);
698 		break;
699 	case POLL_DEREGISTER:
700 		/* enable interrupts */
701 		NFE_WRITE(sc, NFE_IRQ_MASK, NFE_IRQ_WANTED);
702 		break;
703 	case POLL_AND_CHECK_STATUS:
704 		/* fall through */
705 	case POLL_ONLY:
706 		if (ifp->if_flags & IFF_RUNNING) {
707 			nfe_rxeof(sc);
708 			nfe_txeof(sc);
709 		}
710 		break;
711 	}
712 }
713 
714 #endif
715 
716 static void
717 nfe_intr(void *arg)
718 {
719 	struct nfe_softc *sc = arg;
720 	struct ifnet *ifp = &sc->arpcom.ac_if;
721 	uint32_t r;
722 
723 	r = NFE_READ(sc, NFE_IRQ_STATUS);
724 	if (r == 0)
725 		return;	/* not for us */
726 	NFE_WRITE(sc, NFE_IRQ_STATUS, r);
727 
728 	DPRINTFN(sc, 5, "%s: interrupt register %x\n", __func__, r);
729 
730 	if (r & NFE_IRQ_LINK) {
731 		NFE_READ(sc, NFE_PHY_STATUS);
732 		NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
733 		DPRINTF(sc, "link state changed %s\n", "");
734 	}
735 
736 	if (ifp->if_flags & IFF_RUNNING) {
737 		/* check Rx ring */
738 		nfe_rxeof(sc);
739 
740 		/* check Tx ring */
741 		nfe_txeof(sc);
742 	}
743 }
744 
745 static int
746 nfe_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data, struct ucred *cr)
747 {
748 	struct nfe_softc *sc = ifp->if_softc;
749 	struct ifreq *ifr = (struct ifreq *)data;
750 	struct mii_data *mii;
751 	int error = 0, mask;
752 
753 	switch (cmd) {
754 	case SIOCSIFMTU:
755 		if (((sc->sc_flags & NFE_JUMBO_SUP) &&
756 		     ifr->ifr_mtu > NFE_JUMBO_MTU) ||
757 		    ((sc->sc_flags & NFE_JUMBO_SUP) == 0 &&
758 		     ifr->ifr_mtu > ETHERMTU)) {
759 			return EINVAL;
760 		} else if (ifp->if_mtu != ifr->ifr_mtu) {
761 			ifp->if_mtu = ifr->ifr_mtu;
762 			nfe_init(sc);
763 		}
764 		break;
765 	case SIOCSIFFLAGS:
766 		if (ifp->if_flags & IFF_UP) {
767 			/*
768 			 * If only the PROMISC or ALLMULTI flag changes, then
769 			 * don't do a full re-init of the chip, just update
770 			 * the Rx filter.
771 			 */
772 			if ((ifp->if_flags & IFF_RUNNING) &&
773 			    ((ifp->if_flags ^ sc->sc_if_flags) &
774 			     (IFF_ALLMULTI | IFF_PROMISC)) != 0) {
775 				nfe_setmulti(sc);
776 			} else {
777 				if (!(ifp->if_flags & IFF_RUNNING))
778 					nfe_init(sc);
779 			}
780 		} else {
781 			if (ifp->if_flags & IFF_RUNNING)
782 				nfe_stop(sc);
783 		}
784 		sc->sc_if_flags = ifp->if_flags;
785 		break;
786 	case SIOCADDMULTI:
787 	case SIOCDELMULTI:
788 		if (ifp->if_flags & IFF_RUNNING)
789 			nfe_setmulti(sc);
790 		break;
791 	case SIOCSIFMEDIA:
792 	case SIOCGIFMEDIA:
793 		mii = device_get_softc(sc->sc_miibus);
794 		error = ifmedia_ioctl(ifp, ifr, &mii->mii_media, cmd);
795 		break;
796         case SIOCSIFCAP:
797 		mask = (ifr->ifr_reqcap ^ ifp->if_capenable) & IFCAP_HWCSUM;
798 		if (mask && (ifp->if_capabilities & IFCAP_HWCSUM)) {
799 			ifp->if_capenable ^= mask;
800 			if (IFCAP_TXCSUM & ifp->if_capenable)
801 				ifp->if_hwassist = NFE_CSUM_FEATURES;
802 			else
803 				ifp->if_hwassist = 0;
804 
805 			if (ifp->if_flags & IFF_RUNNING)
806 				nfe_init(sc);
807 		}
808 		break;
809 	default:
810 		error = ether_ioctl(ifp, cmd, data);
811 		break;
812 	}
813 	return error;
814 }
815 
816 static void
817 nfe_rxeof(struct nfe_softc *sc)
818 {
819 	struct ifnet *ifp = &sc->arpcom.ac_if;
820 	struct nfe_rx_ring *ring = &sc->rxq;
821 	int reap;
822 
823 	reap = 0;
824 	bus_dmamap_sync(ring->tag, ring->map, BUS_DMASYNC_POSTREAD);
825 
826 	for (;;) {
827 		struct nfe_rx_data *data = &ring->data[ring->cur];
828 		struct mbuf *m;
829 		uint16_t flags;
830 		int len, error;
831 
832 		if (sc->sc_flags & NFE_40BIT_ADDR) {
833 			struct nfe_desc64 *desc64 = &ring->desc64[ring->cur];
834 
835 			flags = le16toh(desc64->flags);
836 			len = le16toh(desc64->length) & 0x3fff;
837 		} else {
838 			struct nfe_desc32 *desc32 = &ring->desc32[ring->cur];
839 
840 			flags = le16toh(desc32->flags);
841 			len = le16toh(desc32->length) & 0x3fff;
842 		}
843 
844 		if (flags & NFE_RX_READY)
845 			break;
846 
847 		reap = 1;
848 
849 		if ((sc->sc_flags & (NFE_JUMBO_SUP | NFE_40BIT_ADDR)) == 0) {
850 			if (!(flags & NFE_RX_VALID_V1))
851 				goto skip;
852 
853 			if ((flags & NFE_RX_FIXME_V1) == NFE_RX_FIXME_V1) {
854 				flags &= ~NFE_RX_ERROR;
855 				len--;	/* fix buffer length */
856 			}
857 		} else {
858 			if (!(flags & NFE_RX_VALID_V2))
859 				goto skip;
860 
861 			if ((flags & NFE_RX_FIXME_V2) == NFE_RX_FIXME_V2) {
862 				flags &= ~NFE_RX_ERROR;
863 				len--;	/* fix buffer length */
864 			}
865 		}
866 
867 		if (flags & NFE_RX_ERROR) {
868 			ifp->if_ierrors++;
869 			goto skip;
870 		}
871 
872 		m = data->m;
873 
874 		if (sc->sc_flags & NFE_USE_JUMBO)
875 			error = nfe_newbuf_jumbo(sc, ring, ring->cur, 0);
876 		else
877 			error = nfe_newbuf_std(sc, ring, ring->cur, 0);
878 		if (error) {
879 			ifp->if_ierrors++;
880 			goto skip;
881 		}
882 
883 		/* finalize mbuf */
884 		m->m_pkthdr.len = m->m_len = len;
885 		m->m_pkthdr.rcvif = ifp;
886 
887 		if ((ifp->if_capenable & IFCAP_RXCSUM) &&
888 		    (flags & NFE_RX_CSUMOK)) {
889 			if (flags & NFE_RX_IP_CSUMOK_V2) {
890 				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED |
891 							  CSUM_IP_VALID;
892 			}
893 
894 			if (flags &
895 			    (NFE_RX_UDP_CSUMOK_V2 | NFE_RX_TCP_CSUMOK_V2)) {
896 				m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
897 							  CSUM_PSEUDO_HDR |
898 							  CSUM_FRAG_NOT_CHECKED;
899 				m->m_pkthdr.csum_data = 0xffff;
900 			}
901 		}
902 
903 		ifp->if_ipackets++;
904 		ifp->if_input(ifp, m);
905 skip:
906 		nfe_set_ready_rxdesc(sc, ring, ring->cur);
907 		sc->rxq.cur = (sc->rxq.cur + 1) % nfe_rx_ring_count;
908 	}
909 
910 	if (reap)
911 		bus_dmamap_sync(ring->tag, ring->map, BUS_DMASYNC_PREWRITE);
912 }
913 
914 static void
915 nfe_txeof(struct nfe_softc *sc)
916 {
917 	struct ifnet *ifp = &sc->arpcom.ac_if;
918 	struct nfe_tx_ring *ring = &sc->txq;
919 	struct nfe_tx_data *data = NULL;
920 
921 	bus_dmamap_sync(ring->tag, ring->map, BUS_DMASYNC_POSTREAD);
922 	while (ring->next != ring->cur) {
923 		uint16_t flags;
924 
925 		if (sc->sc_flags & NFE_40BIT_ADDR)
926 			flags = le16toh(ring->desc64[ring->next].flags);
927 		else
928 			flags = le16toh(ring->desc32[ring->next].flags);
929 
930 		if (flags & NFE_TX_VALID)
931 			break;
932 
933 		data = &ring->data[ring->next];
934 
935 		if ((sc->sc_flags & (NFE_JUMBO_SUP | NFE_40BIT_ADDR)) == 0) {
936 			if (!(flags & NFE_TX_LASTFRAG_V1) && data->m == NULL)
937 				goto skip;
938 
939 			if ((flags & NFE_TX_ERROR_V1) != 0) {
940 				if_printf(ifp, "tx v1 error 0x%4b\n", flags,
941 					  NFE_V1_TXERR);
942 				ifp->if_oerrors++;
943 			} else {
944 				ifp->if_opackets++;
945 			}
946 		} else {
947 			if (!(flags & NFE_TX_LASTFRAG_V2) && data->m == NULL)
948 				goto skip;
949 
950 			if ((flags & NFE_TX_ERROR_V2) != 0) {
951 				if_printf(ifp, "tx v2 error 0x%4b\n", flags,
952 					  NFE_V2_TXERR);
953 				ifp->if_oerrors++;
954 			} else {
955 				ifp->if_opackets++;
956 			}
957 		}
958 
959 		if (data->m == NULL) {	/* should not get there */
960 			if_printf(ifp,
961 				  "last fragment bit w/o associated mbuf!\n");
962 			goto skip;
963 		}
964 
965 		/* last fragment of the mbuf chain transmitted */
966 		bus_dmamap_sync(ring->data_tag, data->map,
967 				BUS_DMASYNC_POSTWRITE);
968 		bus_dmamap_unload(ring->data_tag, data->map);
969 		m_freem(data->m);
970 		data->m = NULL;
971 
972 		ifp->if_timer = 0;
973 skip:
974 		ring->queued--;
975 		KKASSERT(ring->queued >= 0);
976 		ring->next = (ring->next + 1) % NFE_TX_RING_COUNT;
977 	}
978 
979 	if (data != NULL) {	/* at least one slot freed */
980 		ifp->if_flags &= ~IFF_OACTIVE;
981 		ifp->if_start(ifp);
982 	}
983 }
984 
985 static int
986 nfe_encap(struct nfe_softc *sc, struct nfe_tx_ring *ring, struct mbuf *m0)
987 {
988 	struct nfe_dma_ctx ctx;
989 	bus_dma_segment_t segs[NFE_MAX_SCATTER];
990 	struct nfe_tx_data *data, *data_map;
991 	bus_dmamap_t map;
992 	struct nfe_desc64 *desc64 = NULL;
993 	struct nfe_desc32 *desc32 = NULL;
994 	uint16_t flags = 0;
995 	uint32_t vtag = 0;
996 	int error, i, j;
997 
998 	data = &ring->data[ring->cur];
999 	map = data->map;
1000 	data_map = data;	/* Remember who owns the DMA map */
1001 
1002 	ctx.nsegs = NFE_MAX_SCATTER;
1003 	ctx.segs = segs;
1004 	error = bus_dmamap_load_mbuf(ring->data_tag, map, m0,
1005 				     nfe_buf_dma_addr, &ctx, BUS_DMA_NOWAIT);
1006 	if (error && error != EFBIG) {
1007 		if_printf(&sc->arpcom.ac_if, "could not map TX mbuf\n");
1008 		goto back;
1009 	}
1010 
1011 	if (error) {	/* error == EFBIG */
1012 		struct mbuf *m_new;
1013 
1014 		m_new = m_defrag(m0, MB_DONTWAIT);
1015 		if (m_new == NULL) {
1016 			if_printf(&sc->arpcom.ac_if,
1017 				  "could not defrag TX mbuf\n");
1018 			error = ENOBUFS;
1019 			goto back;
1020 		} else {
1021 			m0 = m_new;
1022 		}
1023 
1024 		ctx.nsegs = NFE_MAX_SCATTER;
1025 		ctx.segs = segs;
1026 		error = bus_dmamap_load_mbuf(ring->data_tag, map, m0,
1027 					     nfe_buf_dma_addr, &ctx,
1028 					     BUS_DMA_NOWAIT);
1029 		if (error) {
1030 			if_printf(&sc->arpcom.ac_if,
1031 				  "could not map defraged TX mbuf\n");
1032 			goto back;
1033 		}
1034 	}
1035 
1036 	error = 0;
1037 
1038 	if (ring->queued + ctx.nsegs >= NFE_TX_RING_COUNT - 1) {
1039 		bus_dmamap_unload(ring->data_tag, map);
1040 		error = ENOBUFS;
1041 		goto back;
1042 	}
1043 
1044 	/* setup h/w VLAN tagging */
1045 	if ((m0->m_flags & (M_PROTO1 | M_PKTHDR)) == (M_PROTO1 | M_PKTHDR) &&
1046 	    m0->m_pkthdr.rcvif != NULL &&
1047 	    m0->m_pkthdr.rcvif->if_type == IFT_L2VLAN) {
1048 		struct ifvlan *ifv = m0->m_pkthdr.rcvif->if_softc;
1049 
1050 		if (ifv != NULL)
1051 			vtag = NFE_TX_VTAG | htons(ifv->ifv_tag);
1052 	}
1053 
1054 	if (sc->arpcom.ac_if.if_capenable & IFCAP_TXCSUM) {
1055 		if (m0->m_pkthdr.csum_flags & CSUM_IP)
1056 			flags |= NFE_TX_IP_CSUM;
1057 		if (m0->m_pkthdr.csum_flags & (CSUM_TCP | CSUM_UDP))
1058 			flags |= NFE_TX_TCP_CSUM;
1059 	}
1060 
1061 	/*
1062 	 * XXX urm. somebody is unaware of how hardware works.  You
1063 	 * absolutely CANNOT set NFE_TX_VALID on the next descriptor in
1064 	 * the ring until the entire chain is actually *VALID*.  Otherwise
1065 	 * the hardware may encounter a partially initialized chain that
1066 	 * is marked as being ready to go when it in fact is not ready to
1067 	 * go.
1068 	 */
1069 
1070 	for (i = 0; i < ctx.nsegs; i++) {
1071 		j = (ring->cur + i) % NFE_TX_RING_COUNT;
1072 		data = &ring->data[j];
1073 
1074 		if (sc->sc_flags & NFE_40BIT_ADDR) {
1075 			desc64 = &ring->desc64[j];
1076 #if defined(__LP64__)
1077 			desc64->physaddr[0] =
1078 			    htole32(segs[i].ds_addr >> 32);
1079 #endif
1080 			desc64->physaddr[1] =
1081 			    htole32(segs[i].ds_addr & 0xffffffff);
1082 			desc64->length = htole16(segs[i].ds_len - 1);
1083 			desc64->vtag = htole32(vtag);
1084 			desc64->flags = htole16(flags);
1085 		} else {
1086 			desc32 = &ring->desc32[j];
1087 			desc32->physaddr = htole32(segs[i].ds_addr);
1088 			desc32->length = htole16(segs[i].ds_len - 1);
1089 			desc32->flags = htole16(flags);
1090 		}
1091 
1092 		/* csum flags and vtag belong to the first fragment only */
1093 		flags &= ~(NFE_TX_IP_CSUM | NFE_TX_TCP_CSUM);
1094 		vtag = 0;
1095 
1096 		ring->queued++;
1097 		KKASSERT(ring->queued <= NFE_TX_RING_COUNT);
1098 	}
1099 
1100 	/* the whole mbuf chain has been DMA mapped, fix last descriptor */
1101 	if (sc->sc_flags & NFE_40BIT_ADDR) {
1102 		desc64->flags |= htole16(NFE_TX_LASTFRAG_V2);
1103 	} else {
1104 		if (sc->sc_flags & NFE_JUMBO_SUP)
1105 			flags = NFE_TX_LASTFRAG_V2;
1106 		else
1107 			flags = NFE_TX_LASTFRAG_V1;
1108 		desc32->flags |= htole16(flags);
1109 	}
1110 
1111 	/*
1112 	 * Set NFE_TX_VALID backwards so the hardware doesn't see the
1113 	 * whole mess until the first descriptor in the map is flagged.
1114 	 */
1115 	for (i = ctx.nsegs - 1; i >= 0; --i) {
1116 		j = (ring->cur + i) % NFE_TX_RING_COUNT;
1117 		if (sc->sc_flags & NFE_40BIT_ADDR) {
1118 			desc64 = &ring->desc64[j];
1119 			desc64->flags |= htole16(NFE_TX_VALID);
1120 		} else {
1121 			desc32 = &ring->desc32[j];
1122 			desc32->flags |= htole16(NFE_TX_VALID);
1123 		}
1124 	}
1125 	ring->cur = (ring->cur + ctx.nsegs) % NFE_TX_RING_COUNT;
1126 
1127 	/* Exchange DMA map */
1128 	data_map->map = data->map;
1129 	data->map = map;
1130 	data->m = m0;
1131 
1132 	bus_dmamap_sync(ring->data_tag, map, BUS_DMASYNC_PREWRITE);
1133 back:
1134 	if (error)
1135 		m_freem(m0);
1136 	return error;
1137 }
1138 
1139 static void
1140 nfe_start(struct ifnet *ifp)
1141 {
1142 	struct nfe_softc *sc = ifp->if_softc;
1143 	struct nfe_tx_ring *ring = &sc->txq;
1144 	int count = 0;
1145 	struct mbuf *m0;
1146 
1147 	if (ifp->if_flags & IFF_OACTIVE)
1148 		return;
1149 
1150 	if (ifq_is_empty(&ifp->if_snd))
1151 		return;
1152 
1153 	for (;;) {
1154 		m0 = ifq_dequeue(&ifp->if_snd, NULL);
1155 		if (m0 == NULL)
1156 			break;
1157 
1158 		BPF_MTAP(ifp, m0);
1159 
1160 		if (nfe_encap(sc, ring, m0) != 0) {
1161 			ifp->if_flags |= IFF_OACTIVE;
1162 			break;
1163 		}
1164 		++count;
1165 
1166 		/*
1167 		 * NOTE:
1168 		 * `m0' may be freed in nfe_encap(), so
1169 		 * it should not be touched any more.
1170 		 */
1171 	}
1172 	if (count == 0)	/* nothing sent */
1173 		return;
1174 
1175 	/* Sync TX descriptor ring */
1176 	bus_dmamap_sync(ring->tag, ring->map, BUS_DMASYNC_PREWRITE);
1177 
1178 	/* Kick Tx */
1179 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_KICKTX | sc->rxtxctl);
1180 
1181 	/*
1182 	 * Set a timeout in case the chip goes out to lunch.
1183 	 */
1184 	ifp->if_timer = 5;
1185 }
1186 
1187 static void
1188 nfe_watchdog(struct ifnet *ifp)
1189 {
1190 	struct nfe_softc *sc = ifp->if_softc;
1191 
1192 	if (ifp->if_flags & IFF_RUNNING) {
1193 		if_printf(ifp, "watchdog timeout - lost interrupt recovered\n");
1194 		nfe_txeof(sc);
1195 		return;
1196 	}
1197 
1198 	if_printf(ifp, "watchdog timeout\n");
1199 
1200 	nfe_init(ifp->if_softc);
1201 
1202 	ifp->if_oerrors++;
1203 }
1204 
1205 static void
1206 nfe_init(void *xsc)
1207 {
1208 	struct nfe_softc *sc = xsc;
1209 	struct ifnet *ifp = &sc->arpcom.ac_if;
1210 	uint32_t tmp;
1211 	int error;
1212 
1213 	nfe_stop(sc);
1214 
1215 	/*
1216 	 * NOTE:
1217 	 * Switching between jumbo frames and normal frames should
1218 	 * be done _after_ nfe_stop() but _before_ nfe_init_rx_ring().
1219 	 */
1220 	if (ifp->if_mtu > ETHERMTU) {
1221 		sc->sc_flags |= NFE_USE_JUMBO;
1222 		sc->rxq.bufsz = NFE_JBYTES;
1223 		if (bootverbose)
1224 			if_printf(ifp, "use jumbo frames\n");
1225 	} else {
1226 		sc->sc_flags &= ~NFE_USE_JUMBO;
1227 		sc->rxq.bufsz = MCLBYTES;
1228 		if (bootverbose)
1229 			if_printf(ifp, "use non-jumbo frames\n");
1230 	}
1231 
1232 	error = nfe_init_tx_ring(sc, &sc->txq);
1233 	if (error) {
1234 		nfe_stop(sc);
1235 		return;
1236 	}
1237 
1238 	error = nfe_init_rx_ring(sc, &sc->rxq);
1239 	if (error) {
1240 		nfe_stop(sc);
1241 		return;
1242 	}
1243 
1244 	NFE_WRITE(sc, NFE_TX_UNK, 0);
1245 	NFE_WRITE(sc, NFE_STATUS, 0);
1246 
1247 	sc->rxtxctl = NFE_RXTX_BIT2;
1248 	if (sc->sc_flags & NFE_40BIT_ADDR)
1249 		sc->rxtxctl |= NFE_RXTX_V3MAGIC;
1250 	else if (sc->sc_flags & NFE_JUMBO_SUP)
1251 		sc->rxtxctl |= NFE_RXTX_V2MAGIC;
1252 
1253 	if (ifp->if_capenable & IFCAP_RXCSUM)
1254 		sc->rxtxctl |= NFE_RXTX_RXCSUM;
1255 
1256 	/*
1257 	 * Although the adapter is capable of stripping VLAN tags from received
1258 	 * frames (NFE_RXTX_VTAG_STRIP), we do not enable this functionality on
1259 	 * purpose.  This will be done in software by our network stack.
1260 	 */
1261 	if (sc->sc_flags & NFE_HW_VLAN)
1262 		sc->rxtxctl |= NFE_RXTX_VTAG_INSERT;
1263 
1264 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_RESET | sc->rxtxctl);
1265 	DELAY(10);
1266 	NFE_WRITE(sc, NFE_RXTX_CTL, sc->rxtxctl);
1267 
1268 	if (sc->sc_flags & NFE_HW_VLAN)
1269 		NFE_WRITE(sc, NFE_VTAG_CTL, NFE_VTAG_ENABLE);
1270 
1271 	NFE_WRITE(sc, NFE_SETUP_R6, 0);
1272 
1273 	/* set MAC address */
1274 	nfe_set_macaddr(sc, sc->arpcom.ac_enaddr);
1275 
1276 	/* tell MAC where rings are in memory */
1277 #ifdef __LP64__
1278 	NFE_WRITE(sc, NFE_RX_RING_ADDR_HI, sc->rxq.physaddr >> 32);
1279 #endif
1280 	NFE_WRITE(sc, NFE_RX_RING_ADDR_LO, sc->rxq.physaddr & 0xffffffff);
1281 #ifdef __LP64__
1282 	NFE_WRITE(sc, NFE_TX_RING_ADDR_HI, sc->txq.physaddr >> 32);
1283 #endif
1284 	NFE_WRITE(sc, NFE_TX_RING_ADDR_LO, sc->txq.physaddr & 0xffffffff);
1285 
1286 	NFE_WRITE(sc, NFE_RING_SIZE,
1287 	    (nfe_rx_ring_count - 1) << 16 |
1288 	    (NFE_TX_RING_COUNT - 1));
1289 
1290 	NFE_WRITE(sc, NFE_RXBUFSZ, sc->rxq.bufsz);
1291 
1292 	/* force MAC to wakeup */
1293 	tmp = NFE_READ(sc, NFE_PWR_STATE);
1294 	NFE_WRITE(sc, NFE_PWR_STATE, tmp | NFE_PWR_WAKEUP);
1295 	DELAY(10);
1296 	tmp = NFE_READ(sc, NFE_PWR_STATE);
1297 	NFE_WRITE(sc, NFE_PWR_STATE, tmp | NFE_PWR_VALID);
1298 
1299 	/*
1300 	 * NFE_IMTIMER generates a periodic interrupt via NFE_IRQ_TIMER.
1301 	 * It is unclear how wide the timer is.  Base programming does
1302 	 * not seem to effect NFE_IRQ_TX_DONE or NFE_IRQ_RX_DONE so
1303 	 * we don't get any interrupt moderation.  TX moderation is
1304 	 * possible by using the timer interrupt instead of TX_DONE.
1305 	 *
1306 	 * It is unclear whether there are other bits that can be
1307 	 * set to make the NFE device actually do interrupt moderation
1308 	 * on the RX side.
1309 	 *
1310 	 * For now set a 128uS interval as a placemark, but don't use
1311 	 * the timer.
1312 	 */
1313 	NFE_WRITE(sc, NFE_IMTIMER, NFE_IM_DEFAULT);
1314 
1315 	NFE_WRITE(sc, NFE_SETUP_R1, NFE_R1_MAGIC);
1316 	NFE_WRITE(sc, NFE_SETUP_R2, NFE_R2_MAGIC);
1317 	NFE_WRITE(sc, NFE_SETUP_R6, NFE_R6_MAGIC);
1318 
1319 	/* update MAC knowledge of PHY; generates a NFE_IRQ_LINK interrupt */
1320 	NFE_WRITE(sc, NFE_STATUS, sc->mii_phyaddr << 24 | NFE_STATUS_MAGIC);
1321 
1322 	NFE_WRITE(sc, NFE_SETUP_R4, NFE_R4_MAGIC);
1323 	NFE_WRITE(sc, NFE_WOL_CTL, NFE_WOL_MAGIC);
1324 
1325 	sc->rxtxctl &= ~NFE_RXTX_BIT2;
1326 	NFE_WRITE(sc, NFE_RXTX_CTL, sc->rxtxctl);
1327 	DELAY(10);
1328 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_BIT1 | sc->rxtxctl);
1329 
1330 	/* set Rx filter */
1331 	nfe_setmulti(sc);
1332 
1333 	nfe_ifmedia_upd(ifp);
1334 
1335 	/* enable Rx */
1336 	NFE_WRITE(sc, NFE_RX_CTL, NFE_RX_START);
1337 
1338 	/* enable Tx */
1339 	NFE_WRITE(sc, NFE_TX_CTL, NFE_TX_START);
1340 
1341 	NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
1342 
1343 #ifdef DEVICE_POLLING
1344 	if ((ifp->if_flags & IFF_POLLING) == 0)
1345 #endif
1346 	/* enable interrupts */
1347 	NFE_WRITE(sc, NFE_IRQ_MASK, NFE_IRQ_WANTED);
1348 
1349 	callout_reset(&sc->sc_tick_ch, hz, nfe_tick, sc);
1350 
1351 	ifp->if_flags |= IFF_RUNNING;
1352 	ifp->if_flags &= ~IFF_OACTIVE;
1353 
1354 	/*
1355 	 * If we had stuff in the tx ring before its all cleaned out now
1356 	 * so we are not going to get an interrupt, jump-start any pending
1357 	 * output.
1358 	 */
1359 	ifp->if_start(ifp);
1360 }
1361 
1362 static void
1363 nfe_stop(struct nfe_softc *sc)
1364 {
1365 	struct ifnet *ifp = &sc->arpcom.ac_if;
1366 
1367 	callout_stop(&sc->sc_tick_ch);
1368 
1369 	ifp->if_timer = 0;
1370 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
1371 
1372 	/*
1373 	 * Are NFE_TX_CTL and NFE_RX_CTL polled by the chip microcontroller
1374 	 * or do they directly reset/terminate the DMA hardware?  Nobody
1375 	 * knows.
1376 	 *
1377 	 * Add two delays:
1378 	 *
1379 	 * (1) Delay before zeroing out NFE_TX_CTL.  This seems to help a
1380 	 * watchdog timeout that occurs after a stop/init sequence.  I am
1381 	 * theorizing that a TX KICK occuring just prior to a reinit (e.g.
1382 	 * due to dhclient) is queueing an interrupt to the microcontroller
1383 	 * which gets delayed until after we clear the control registers
1384 	 * down below, resulting in mass confusion.  TX KICK is clearly
1385 	 * hardware aided whereas the other bits in the control register
1386 	 * are more likely to be polled by the microcontroller.
1387 	 *
1388 	 * (2) Delay after zeroing out TX and RX CTL registers, under the
1389 	 * assumption that primary DMA is initiated and terminated by
1390 	 * the microcontroller and not hardware (and anyway, one can hardly
1391 	 * expect the DMA engine to just instantly stop!).  We don't want
1392 	 * to rip the rings out from under it before it has had a chance to
1393 	 * actually stop!
1394 	 */
1395 	DELAY(1000);
1396 
1397 	/* Abort Tx */
1398 	NFE_WRITE(sc, NFE_TX_CTL, 0);
1399 
1400 	/* Disable Rx */
1401 	NFE_WRITE(sc, NFE_RX_CTL, 0);
1402 
1403 	/* Disable interrupts */
1404 	NFE_WRITE(sc, NFE_IRQ_MASK, 0);
1405 
1406 	DELAY(1000);
1407 
1408 	/* Reset Tx and Rx rings */
1409 	nfe_reset_tx_ring(sc, &sc->txq);
1410 	nfe_reset_rx_ring(sc, &sc->rxq);
1411 }
1412 
1413 static int
1414 nfe_alloc_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1415 {
1416 	int i, j, error, descsize;
1417 	void **desc;
1418 
1419 	if (sc->sc_flags & NFE_40BIT_ADDR) {
1420 		desc = (void **)&ring->desc64;
1421 		descsize = sizeof(struct nfe_desc64);
1422 	} else {
1423 		desc = (void **)&ring->desc32;
1424 		descsize = sizeof(struct nfe_desc32);
1425 	}
1426 
1427 	ring->jbuf = kmalloc(sizeof(struct nfe_jbuf) * NFE_JPOOL_COUNT,
1428 			     M_DEVBUF, M_WAITOK | M_ZERO);
1429 	ring->data = kmalloc(sizeof(struct nfe_rx_data) * nfe_rx_ring_count,
1430 			     M_DEVBUF, M_WAITOK | M_ZERO);
1431 
1432 	ring->bufsz = MCLBYTES;
1433 	ring->cur = ring->next = 0;
1434 
1435 	error = bus_dma_tag_create(NULL, PAGE_SIZE, 0,
1436 				   BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR,
1437 				   NULL, NULL,
1438 				   nfe_rx_ring_count * descsize, 1,
1439 				   nfe_rx_ring_count * descsize,
1440 				   0, &ring->tag);
1441 	if (error) {
1442 		if_printf(&sc->arpcom.ac_if,
1443 			  "could not create desc RX DMA tag\n");
1444 		return error;
1445 	}
1446 
1447 	error = bus_dmamem_alloc(ring->tag, desc, BUS_DMA_WAITOK | BUS_DMA_ZERO,
1448 				 &ring->map);
1449 	if (error) {
1450 		if_printf(&sc->arpcom.ac_if,
1451 			  "could not allocate RX desc DMA memory\n");
1452 		bus_dma_tag_destroy(ring->tag);
1453 		ring->tag = NULL;
1454 		return error;
1455 	}
1456 
1457 	error = bus_dmamap_load(ring->tag, ring->map, *desc,
1458 				nfe_rx_ring_count * descsize,
1459 				nfe_ring_dma_addr, &ring->physaddr,
1460 				BUS_DMA_WAITOK);
1461 	if (error) {
1462 		if_printf(&sc->arpcom.ac_if,
1463 			  "could not load RX desc DMA map\n");
1464 		bus_dmamem_free(ring->tag, *desc, ring->map);
1465 		bus_dma_tag_destroy(ring->tag);
1466 		ring->tag = NULL;
1467 		return error;
1468 	}
1469 
1470 	if (sc->sc_flags & NFE_JUMBO_SUP) {
1471 		error = nfe_jpool_alloc(sc, ring);
1472 		if (error) {
1473 			if_printf(&sc->arpcom.ac_if,
1474 				  "could not allocate jumbo frames\n");
1475 			return error;
1476 		}
1477 	}
1478 
1479 	error = bus_dma_tag_create(NULL, 1, 0,
1480 				   BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR,
1481 				   NULL, NULL,
1482 				   MCLBYTES, 1, MCLBYTES,
1483 				   0, &ring->data_tag);
1484 	if (error) {
1485 		if_printf(&sc->arpcom.ac_if,
1486 			  "could not create RX mbuf DMA tag\n");
1487 		return error;
1488 	}
1489 
1490 	/* Create a spare RX mbuf DMA map */
1491 	error = bus_dmamap_create(ring->data_tag, 0, &ring->data_tmpmap);
1492 	if (error) {
1493 		if_printf(&sc->arpcom.ac_if,
1494 			  "could not create spare RX mbuf DMA map\n");
1495 		bus_dma_tag_destroy(ring->data_tag);
1496 		ring->data_tag = NULL;
1497 		return error;
1498 	}
1499 
1500 	for (i = 0; i < nfe_rx_ring_count; i++) {
1501 		error = bus_dmamap_create(ring->data_tag, 0,
1502 					  &ring->data[i].map);
1503 		if (error) {
1504 			if_printf(&sc->arpcom.ac_if,
1505 				  "could not create %dth RX mbuf DMA mapn", i);
1506 			goto fail;
1507 		}
1508 	}
1509 	return 0;
1510 fail:
1511 	for (j = 0; j < i; ++j)
1512 		bus_dmamap_destroy(ring->data_tag, ring->data[i].map);
1513 	bus_dmamap_destroy(ring->data_tag, ring->data_tmpmap);
1514 	bus_dma_tag_destroy(ring->data_tag);
1515 	ring->data_tag = NULL;
1516 	return error;
1517 }
1518 
1519 static void
1520 nfe_reset_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1521 {
1522 	int i;
1523 
1524 	for (i = 0; i < nfe_rx_ring_count; i++) {
1525 		struct nfe_rx_data *data = &ring->data[i];
1526 
1527 		if (data->m != NULL) {
1528 			if ((sc->sc_flags & NFE_USE_JUMBO) == 0)
1529 				bus_dmamap_unload(ring->data_tag, data->map);
1530 			m_freem(data->m);
1531 			data->m = NULL;
1532 		}
1533 	}
1534 	bus_dmamap_sync(ring->tag, ring->map, BUS_DMASYNC_PREWRITE);
1535 
1536 	ring->cur = ring->next = 0;
1537 }
1538 
1539 static int
1540 nfe_init_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1541 {
1542 	int i;
1543 
1544 	for (i = 0; i < nfe_rx_ring_count; ++i) {
1545 		int error;
1546 
1547 		/* XXX should use a function pointer */
1548 		if (sc->sc_flags & NFE_USE_JUMBO)
1549 			error = nfe_newbuf_jumbo(sc, ring, i, 1);
1550 		else
1551 			error = nfe_newbuf_std(sc, ring, i, 1);
1552 		if (error) {
1553 			if_printf(&sc->arpcom.ac_if,
1554 				  "could not allocate RX buffer\n");
1555 			return error;
1556 		}
1557 
1558 		nfe_set_ready_rxdesc(sc, ring, i);
1559 	}
1560 	bus_dmamap_sync(ring->tag, ring->map, BUS_DMASYNC_PREWRITE);
1561 
1562 	return 0;
1563 }
1564 
1565 static void
1566 nfe_free_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1567 {
1568 	if (ring->data_tag != NULL) {
1569 		struct nfe_rx_data *data;
1570 		int i;
1571 
1572 		for (i = 0; i < nfe_rx_ring_count; i++) {
1573 			data = &ring->data[i];
1574 
1575 			if (data->m != NULL) {
1576 				bus_dmamap_unload(ring->data_tag, data->map);
1577 				m_freem(data->m);
1578 			}
1579 			bus_dmamap_destroy(ring->data_tag, data->map);
1580 		}
1581 		bus_dmamap_destroy(ring->data_tag, ring->data_tmpmap);
1582 		bus_dma_tag_destroy(ring->data_tag);
1583 	}
1584 
1585 	nfe_jpool_free(sc, ring);
1586 
1587 	if (ring->jbuf != NULL)
1588 		kfree(ring->jbuf, M_DEVBUF);
1589 	if (ring->data != NULL)
1590 		kfree(ring->data, M_DEVBUF);
1591 
1592 	if (ring->tag != NULL) {
1593 		void *desc;
1594 
1595 		if (sc->sc_flags & NFE_40BIT_ADDR)
1596 			desc = ring->desc64;
1597 		else
1598 			desc = ring->desc32;
1599 
1600 		bus_dmamap_unload(ring->tag, ring->map);
1601 		bus_dmamem_free(ring->tag, desc, ring->map);
1602 		bus_dma_tag_destroy(ring->tag);
1603 	}
1604 }
1605 
1606 static struct nfe_jbuf *
1607 nfe_jalloc(struct nfe_softc *sc)
1608 {
1609 	struct ifnet *ifp = &sc->arpcom.ac_if;
1610 	struct nfe_jbuf *jbuf;
1611 
1612 	lwkt_serialize_enter(&sc->sc_jbuf_serializer);
1613 
1614 	jbuf = SLIST_FIRST(&sc->rxq.jfreelist);
1615 	if (jbuf != NULL) {
1616 		SLIST_REMOVE_HEAD(&sc->rxq.jfreelist, jnext);
1617 		jbuf->inuse = 1;
1618 	} else {
1619 		if_printf(ifp, "no free jumbo buffer\n");
1620 	}
1621 
1622 	lwkt_serialize_exit(&sc->sc_jbuf_serializer);
1623 
1624 	return jbuf;
1625 }
1626 
1627 static void
1628 nfe_jfree(void *arg)
1629 {
1630 	struct nfe_jbuf *jbuf = arg;
1631 	struct nfe_softc *sc = jbuf->sc;
1632 	struct nfe_rx_ring *ring = jbuf->ring;
1633 
1634 	if (&ring->jbuf[jbuf->slot] != jbuf)
1635 		panic("%s: free wrong jumbo buffer\n", __func__);
1636 	else if (jbuf->inuse == 0)
1637 		panic("%s: jumbo buffer already freed\n", __func__);
1638 
1639 	lwkt_serialize_enter(&sc->sc_jbuf_serializer);
1640 	atomic_subtract_int(&jbuf->inuse, 1);
1641 	if (jbuf->inuse == 0)
1642 		SLIST_INSERT_HEAD(&ring->jfreelist, jbuf, jnext);
1643 	lwkt_serialize_exit(&sc->sc_jbuf_serializer);
1644 }
1645 
1646 static void
1647 nfe_jref(void *arg)
1648 {
1649 	struct nfe_jbuf *jbuf = arg;
1650 	struct nfe_rx_ring *ring = jbuf->ring;
1651 
1652 	if (&ring->jbuf[jbuf->slot] != jbuf)
1653 		panic("%s: ref wrong jumbo buffer\n", __func__);
1654 	else if (jbuf->inuse == 0)
1655 		panic("%s: jumbo buffer already freed\n", __func__);
1656 
1657 	atomic_add_int(&jbuf->inuse, 1);
1658 }
1659 
1660 static int
1661 nfe_jpool_alloc(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1662 {
1663 	struct nfe_jbuf *jbuf;
1664 	bus_addr_t physaddr;
1665 	caddr_t buf;
1666 	int i, error;
1667 
1668 	/*
1669 	 * Allocate a big chunk of DMA'able memory.
1670 	 */
1671 	error = bus_dma_tag_create(NULL, PAGE_SIZE, 0,
1672 				   BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR,
1673 				   NULL, NULL,
1674 				   NFE_JPOOL_SIZE, 1, NFE_JPOOL_SIZE,
1675 				   0, &ring->jtag);
1676 	if (error) {
1677 		if_printf(&sc->arpcom.ac_if,
1678 			  "could not create jumbo DMA tag\n");
1679 		return error;
1680 	}
1681 
1682 	error = bus_dmamem_alloc(ring->jtag, (void **)&ring->jpool,
1683 				 BUS_DMA_WAITOK, &ring->jmap);
1684 	if (error) {
1685 		if_printf(&sc->arpcom.ac_if,
1686 			  "could not allocate jumbo DMA memory\n");
1687 		bus_dma_tag_destroy(ring->jtag);
1688 		ring->jtag = NULL;
1689 		return error;
1690 	}
1691 
1692 	error = bus_dmamap_load(ring->jtag, ring->jmap, ring->jpool,
1693 				NFE_JPOOL_SIZE, nfe_ring_dma_addr, &physaddr,
1694 				BUS_DMA_WAITOK);
1695 	if (error) {
1696 		if_printf(&sc->arpcom.ac_if,
1697 			  "could not load jumbo DMA map\n");
1698 		bus_dmamem_free(ring->jtag, ring->jpool, ring->jmap);
1699 		bus_dma_tag_destroy(ring->jtag);
1700 		ring->jtag = NULL;
1701 		return error;
1702 	}
1703 
1704 	/* ..and split it into 9KB chunks */
1705 	SLIST_INIT(&ring->jfreelist);
1706 
1707 	buf = ring->jpool;
1708 	for (i = 0; i < NFE_JPOOL_COUNT; i++) {
1709 		jbuf = &ring->jbuf[i];
1710 
1711 		jbuf->sc = sc;
1712 		jbuf->ring = ring;
1713 		jbuf->inuse = 0;
1714 		jbuf->slot = i;
1715 		jbuf->buf = buf;
1716 		jbuf->physaddr = physaddr;
1717 
1718 		SLIST_INSERT_HEAD(&ring->jfreelist, jbuf, jnext);
1719 
1720 		buf += NFE_JBYTES;
1721 		physaddr += NFE_JBYTES;
1722 	}
1723 
1724 	return 0;
1725 }
1726 
1727 static void
1728 nfe_jpool_free(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1729 {
1730 	if (ring->jtag != NULL) {
1731 		bus_dmamap_unload(ring->jtag, ring->jmap);
1732 		bus_dmamem_free(ring->jtag, ring->jpool, ring->jmap);
1733 		bus_dma_tag_destroy(ring->jtag);
1734 	}
1735 }
1736 
1737 static int
1738 nfe_alloc_tx_ring(struct nfe_softc *sc, struct nfe_tx_ring *ring)
1739 {
1740 	int i, j, error, descsize;
1741 	void **desc;
1742 
1743 	if (sc->sc_flags & NFE_40BIT_ADDR) {
1744 		desc = (void **)&ring->desc64;
1745 		descsize = sizeof(struct nfe_desc64);
1746 	} else {
1747 		desc = (void **)&ring->desc32;
1748 		descsize = sizeof(struct nfe_desc32);
1749 	}
1750 
1751 	ring->queued = 0;
1752 	ring->cur = ring->next = 0;
1753 
1754 	error = bus_dma_tag_create(NULL, PAGE_SIZE, 0,
1755 				   BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR,
1756 				   NULL, NULL,
1757 				   NFE_TX_RING_COUNT * descsize, 1,
1758 				   NFE_TX_RING_COUNT * descsize,
1759 				   0, &ring->tag);
1760 	if (error) {
1761 		if_printf(&sc->arpcom.ac_if,
1762 			  "could not create TX desc DMA map\n");
1763 		return error;
1764 	}
1765 
1766 	error = bus_dmamem_alloc(ring->tag, desc, BUS_DMA_WAITOK | BUS_DMA_ZERO,
1767 				 &ring->map);
1768 	if (error) {
1769 		if_printf(&sc->arpcom.ac_if,
1770 			  "could not allocate TX desc DMA memory\n");
1771 		bus_dma_tag_destroy(ring->tag);
1772 		ring->tag = NULL;
1773 		return error;
1774 	}
1775 
1776 	error = bus_dmamap_load(ring->tag, ring->map, *desc,
1777 				NFE_TX_RING_COUNT * descsize,
1778 				nfe_ring_dma_addr, &ring->physaddr,
1779 				BUS_DMA_WAITOK);
1780 	if (error) {
1781 		if_printf(&sc->arpcom.ac_if,
1782 			  "could not load TX desc DMA map\n");
1783 		bus_dmamem_free(ring->tag, *desc, ring->map);
1784 		bus_dma_tag_destroy(ring->tag);
1785 		ring->tag = NULL;
1786 		return error;
1787 	}
1788 
1789 	error = bus_dma_tag_create(NULL, PAGE_SIZE, 0,
1790 				   BUS_SPACE_MAXADDR_32BIT, BUS_SPACE_MAXADDR,
1791 				   NULL, NULL,
1792 				   NFE_JBYTES * NFE_MAX_SCATTER,
1793 				   NFE_MAX_SCATTER, NFE_JBYTES,
1794 				   0, &ring->data_tag);
1795 	if (error) {
1796 		if_printf(&sc->arpcom.ac_if,
1797 			  "could not create TX buf DMA tag\n");
1798 		return error;
1799 	}
1800 
1801 	for (i = 0; i < NFE_TX_RING_COUNT; i++) {
1802 		error = bus_dmamap_create(ring->data_tag, 0,
1803 					  &ring->data[i].map);
1804 		if (error) {
1805 			if_printf(&sc->arpcom.ac_if,
1806 				  "could not create %dth TX buf DMA map\n", i);
1807 			goto fail;
1808 		}
1809 	}
1810 
1811 	return 0;
1812 fail:
1813 	for (j = 0; j < i; ++j)
1814 		bus_dmamap_destroy(ring->data_tag, ring->data[i].map);
1815 	bus_dma_tag_destroy(ring->data_tag);
1816 	ring->data_tag = NULL;
1817 	return error;
1818 }
1819 
1820 static void
1821 nfe_reset_tx_ring(struct nfe_softc *sc, struct nfe_tx_ring *ring)
1822 {
1823 	int i;
1824 
1825 	for (i = 0; i < NFE_TX_RING_COUNT; i++) {
1826 		struct nfe_tx_data *data = &ring->data[i];
1827 
1828 		if (sc->sc_flags & NFE_40BIT_ADDR)
1829 			ring->desc64[i].flags = 0;
1830 		else
1831 			ring->desc32[i].flags = 0;
1832 
1833 		if (data->m != NULL) {
1834 			bus_dmamap_sync(ring->data_tag, data->map,
1835 					BUS_DMASYNC_POSTWRITE);
1836 			bus_dmamap_unload(ring->data_tag, data->map);
1837 			m_freem(data->m);
1838 			data->m = NULL;
1839 		}
1840 	}
1841 	bus_dmamap_sync(ring->tag, ring->map, BUS_DMASYNC_PREWRITE);
1842 
1843 	ring->queued = 0;
1844 	ring->cur = ring->next = 0;
1845 }
1846 
1847 static int
1848 nfe_init_tx_ring(struct nfe_softc *sc __unused,
1849 		 struct nfe_tx_ring *ring __unused)
1850 {
1851 	return 0;
1852 }
1853 
1854 static void
1855 nfe_free_tx_ring(struct nfe_softc *sc, struct nfe_tx_ring *ring)
1856 {
1857 	if (ring->data_tag != NULL) {
1858 		struct nfe_tx_data *data;
1859 		int i;
1860 
1861 		for (i = 0; i < NFE_TX_RING_COUNT; ++i) {
1862 			data = &ring->data[i];
1863 
1864 			if (data->m != NULL) {
1865 				bus_dmamap_unload(ring->data_tag, data->map);
1866 				m_freem(data->m);
1867 			}
1868 			bus_dmamap_destroy(ring->data_tag, data->map);
1869 		}
1870 
1871 		bus_dma_tag_destroy(ring->data_tag);
1872 	}
1873 
1874 	if (ring->tag != NULL) {
1875 		void *desc;
1876 
1877 		if (sc->sc_flags & NFE_40BIT_ADDR)
1878 			desc = ring->desc64;
1879 		else
1880 			desc = ring->desc32;
1881 
1882 		bus_dmamap_unload(ring->tag, ring->map);
1883 		bus_dmamem_free(ring->tag, desc, ring->map);
1884 		bus_dma_tag_destroy(ring->tag);
1885 	}
1886 }
1887 
1888 static int
1889 nfe_ifmedia_upd(struct ifnet *ifp)
1890 {
1891 	struct nfe_softc *sc = ifp->if_softc;
1892 	struct mii_data *mii = device_get_softc(sc->sc_miibus);
1893 
1894 	if (mii->mii_instance != 0) {
1895 		struct mii_softc *miisc;
1896 
1897 		LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
1898 			mii_phy_reset(miisc);
1899 	}
1900 	mii_mediachg(mii);
1901 
1902 	return 0;
1903 }
1904 
1905 static void
1906 nfe_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
1907 {
1908 	struct nfe_softc *sc = ifp->if_softc;
1909 	struct mii_data *mii = device_get_softc(sc->sc_miibus);
1910 
1911 	mii_pollstat(mii);
1912 	ifmr->ifm_status = mii->mii_media_status;
1913 	ifmr->ifm_active = mii->mii_media_active;
1914 }
1915 
1916 static void
1917 nfe_setmulti(struct nfe_softc *sc)
1918 {
1919 	struct ifnet *ifp = &sc->arpcom.ac_if;
1920 	struct ifmultiaddr *ifma;
1921 	uint8_t addr[ETHER_ADDR_LEN], mask[ETHER_ADDR_LEN];
1922 	uint32_t filter = NFE_RXFILTER_MAGIC;
1923 	int i;
1924 
1925 	if ((ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) != 0) {
1926 		bzero(addr, ETHER_ADDR_LEN);
1927 		bzero(mask, ETHER_ADDR_LEN);
1928 		goto done;
1929 	}
1930 
1931 	bcopy(etherbroadcastaddr, addr, ETHER_ADDR_LEN);
1932 	bcopy(etherbroadcastaddr, mask, ETHER_ADDR_LEN);
1933 
1934 	LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1935 		caddr_t maddr;
1936 
1937 		if (ifma->ifma_addr->sa_family != AF_LINK)
1938 			continue;
1939 
1940 		maddr = LLADDR((struct sockaddr_dl *)ifma->ifma_addr);
1941 		for (i = 0; i < ETHER_ADDR_LEN; i++) {
1942 			addr[i] &= maddr[i];
1943 			mask[i] &= ~maddr[i];
1944 		}
1945 	}
1946 
1947 	for (i = 0; i < ETHER_ADDR_LEN; i++)
1948 		mask[i] |= addr[i];
1949 
1950 done:
1951 	addr[0] |= 0x01;	/* make sure multicast bit is set */
1952 
1953 	NFE_WRITE(sc, NFE_MULTIADDR_HI,
1954 	    addr[3] << 24 | addr[2] << 16 | addr[1] << 8 | addr[0]);
1955 	NFE_WRITE(sc, NFE_MULTIADDR_LO,
1956 	    addr[5] <<  8 | addr[4]);
1957 	NFE_WRITE(sc, NFE_MULTIMASK_HI,
1958 	    mask[3] << 24 | mask[2] << 16 | mask[1] << 8 | mask[0]);
1959 	NFE_WRITE(sc, NFE_MULTIMASK_LO,
1960 	    mask[5] <<  8 | mask[4]);
1961 
1962 	filter |= (ifp->if_flags & IFF_PROMISC) ? NFE_PROMISC : NFE_U2M;
1963 	NFE_WRITE(sc, NFE_RXFILTER, filter);
1964 }
1965 
1966 static void
1967 nfe_get_macaddr(struct nfe_softc *sc, uint8_t *addr)
1968 {
1969 	uint32_t tmp;
1970 
1971 	tmp = NFE_READ(sc, NFE_MACADDR_LO);
1972 	addr[0] = (tmp >> 8) & 0xff;
1973 	addr[1] = (tmp & 0xff);
1974 
1975 	tmp = NFE_READ(sc, NFE_MACADDR_HI);
1976 	addr[2] = (tmp >> 24) & 0xff;
1977 	addr[3] = (tmp >> 16) & 0xff;
1978 	addr[4] = (tmp >>  8) & 0xff;
1979 	addr[5] = (tmp & 0xff);
1980 }
1981 
1982 static void
1983 nfe_set_macaddr(struct nfe_softc *sc, const uint8_t *addr)
1984 {
1985 	NFE_WRITE(sc, NFE_MACADDR_LO,
1986 	    addr[5] <<  8 | addr[4]);
1987 	NFE_WRITE(sc, NFE_MACADDR_HI,
1988 	    addr[3] << 24 | addr[2] << 16 | addr[1] << 8 | addr[0]);
1989 }
1990 
1991 static void
1992 nfe_tick(void *arg)
1993 {
1994 	struct nfe_softc *sc = arg;
1995 	struct ifnet *ifp = &sc->arpcom.ac_if;
1996 	struct mii_data *mii = device_get_softc(sc->sc_miibus);
1997 
1998 	lwkt_serialize_enter(ifp->if_serializer);
1999 
2000 	mii_tick(mii);
2001 	callout_reset(&sc->sc_tick_ch, hz, nfe_tick, sc);
2002 
2003 	lwkt_serialize_exit(ifp->if_serializer);
2004 }
2005 
2006 static void
2007 nfe_ring_dma_addr(void *arg, bus_dma_segment_t *seg, int nseg, int error)
2008 {
2009 	if (error)
2010 		return;
2011 
2012 	KASSERT(nseg == 1, ("too many segments, should be 1\n"));
2013 
2014 	*((uint32_t *)arg) = seg->ds_addr;
2015 }
2016 
2017 static void
2018 nfe_buf_dma_addr(void *arg, bus_dma_segment_t *segs, int nsegs,
2019 		 bus_size_t mapsz __unused, int error)
2020 {
2021 	struct nfe_dma_ctx *ctx = arg;
2022 	int i;
2023 
2024 	if (error)
2025 		return;
2026 
2027 	KASSERT(nsegs <= ctx->nsegs,
2028 		("too many segments(%d), should be <= %d\n",
2029 		 nsegs, ctx->nsegs));
2030 
2031 	ctx->nsegs = nsegs;
2032 	for (i = 0; i < nsegs; ++i)
2033 		ctx->segs[i] = segs[i];
2034 }
2035 
2036 static int
2037 nfe_newbuf_std(struct nfe_softc *sc, struct nfe_rx_ring *ring, int idx,
2038 	       int wait)
2039 {
2040 	struct nfe_rx_data *data = &ring->data[idx];
2041 	struct nfe_dma_ctx ctx;
2042 	bus_dma_segment_t seg;
2043 	bus_dmamap_t map;
2044 	struct mbuf *m;
2045 	int error;
2046 
2047 	m = m_getcl(wait ? MB_WAIT : MB_DONTWAIT, MT_DATA, M_PKTHDR);
2048 	if (m == NULL)
2049 		return ENOBUFS;
2050 	m->m_len = m->m_pkthdr.len = MCLBYTES;
2051 
2052 	ctx.nsegs = 1;
2053 	ctx.segs = &seg;
2054 	error = bus_dmamap_load_mbuf(ring->data_tag, ring->data_tmpmap,
2055 				     m, nfe_buf_dma_addr, &ctx,
2056 				     wait ? BUS_DMA_WAITOK : BUS_DMA_NOWAIT);
2057 	if (error) {
2058 		m_freem(m);
2059 		if_printf(&sc->arpcom.ac_if, "could map RX mbuf %d\n", error);
2060 		return error;
2061 	}
2062 
2063 	/* Unload originally mapped mbuf */
2064 	bus_dmamap_unload(ring->data_tag, data->map);
2065 
2066 	/* Swap this DMA map with tmp DMA map */
2067 	map = data->map;
2068 	data->map = ring->data_tmpmap;
2069 	ring->data_tmpmap = map;
2070 
2071 	/* Caller is assumed to have collected the old mbuf */
2072 	data->m = m;
2073 
2074 	nfe_set_paddr_rxdesc(sc, ring, idx, seg.ds_addr);
2075 
2076 	bus_dmamap_sync(ring->data_tag, data->map, BUS_DMASYNC_PREREAD);
2077 	return 0;
2078 }
2079 
2080 static int
2081 nfe_newbuf_jumbo(struct nfe_softc *sc, struct nfe_rx_ring *ring, int idx,
2082 		 int wait)
2083 {
2084 	struct nfe_rx_data *data = &ring->data[idx];
2085 	struct nfe_jbuf *jbuf;
2086 	struct mbuf *m;
2087 
2088 	MGETHDR(m, wait ? MB_WAIT : MB_DONTWAIT, MT_DATA);
2089 	if (m == NULL)
2090 		return ENOBUFS;
2091 
2092 	jbuf = nfe_jalloc(sc);
2093 	if (jbuf == NULL) {
2094 		m_freem(m);
2095 		if_printf(&sc->arpcom.ac_if, "jumbo allocation failed "
2096 		    "-- packet dropped!\n");
2097 		return ENOBUFS;
2098 	}
2099 
2100 	m->m_ext.ext_arg = jbuf;
2101 	m->m_ext.ext_buf = jbuf->buf;
2102 	m->m_ext.ext_free = nfe_jfree;
2103 	m->m_ext.ext_ref = nfe_jref;
2104 	m->m_ext.ext_size = NFE_JBYTES;
2105 
2106 	m->m_data = m->m_ext.ext_buf;
2107 	m->m_flags |= M_EXT;
2108 	m->m_len = m->m_pkthdr.len = m->m_ext.ext_size;
2109 
2110 	/* Caller is assumed to have collected the old mbuf */
2111 	data->m = m;
2112 
2113 	nfe_set_paddr_rxdesc(sc, ring, idx, jbuf->physaddr);
2114 
2115 	bus_dmamap_sync(ring->jtag, ring->jmap, BUS_DMASYNC_PREREAD);
2116 	return 0;
2117 }
2118 
2119 static void
2120 nfe_set_paddr_rxdesc(struct nfe_softc *sc, struct nfe_rx_ring *ring, int idx,
2121 		     bus_addr_t physaddr)
2122 {
2123 	if (sc->sc_flags & NFE_40BIT_ADDR) {
2124 		struct nfe_desc64 *desc64 = &ring->desc64[idx];
2125 
2126 #if defined(__LP64__)
2127 		desc64->physaddr[0] = htole32(physaddr >> 32);
2128 #endif
2129 		desc64->physaddr[1] = htole32(physaddr & 0xffffffff);
2130 	} else {
2131 		struct nfe_desc32 *desc32 = &ring->desc32[idx];
2132 
2133 		desc32->physaddr = htole32(physaddr);
2134 	}
2135 }
2136 
2137 static void
2138 nfe_set_ready_rxdesc(struct nfe_softc *sc, struct nfe_rx_ring *ring, int idx)
2139 {
2140 	if (sc->sc_flags & NFE_40BIT_ADDR) {
2141 		struct nfe_desc64 *desc64 = &ring->desc64[idx];
2142 
2143 		desc64->length = htole16(ring->bufsz);
2144 		desc64->flags = htole16(NFE_RX_READY);
2145 	} else {
2146 		struct nfe_desc32 *desc32 = &ring->desc32[idx];
2147 
2148 		desc32->length = htole16(ring->bufsz);
2149 		desc32->flags = htole16(NFE_RX_READY);
2150 	}
2151 }
2152