xref: /netbsd-src/sys/dev/pci/if_nfe.c (revision e89934bbf778a6d6d6894877c4da59d0c7835b0f)
1 /*	$NetBSD: if_nfe.c,v 1.63 2016/12/15 09:28:05 ozaki-r Exp $	*/
2 /*	$OpenBSD: if_nfe.c,v 1.77 2008/02/05 16:52:50 brad Exp $	*/
3 
4 /*-
5  * Copyright (c) 2006, 2007 Damien Bergamini <damien.bergamini@free.fr>
6  * Copyright (c) 2005, 2006 Jonathan Gray <jsg@openbsd.org>
7  *
8  * Permission to use, copy, modify, and distribute this software for any
9  * purpose with or without fee is hereby granted, provided that the above
10  * copyright notice and this permission notice appear in all copies.
11  *
12  * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
13  * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
14  * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
15  * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
16  * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
17  * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
18  * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
19  */
20 
21 /* Driver for NVIDIA nForce MCP Fast Ethernet and Gigabit Ethernet */
22 
23 #include <sys/cdefs.h>
24 __KERNEL_RCSID(0, "$NetBSD: if_nfe.c,v 1.63 2016/12/15 09:28:05 ozaki-r Exp $");
25 
26 #include "opt_inet.h"
27 #include "vlan.h"
28 
29 #include <sys/param.h>
30 #include <sys/endian.h>
31 #include <sys/systm.h>
32 #include <sys/types.h>
33 #include <sys/sockio.h>
34 #include <sys/mbuf.h>
35 #include <sys/mutex.h>
36 #include <sys/queue.h>
37 #include <sys/kernel.h>
38 #include <sys/device.h>
39 #include <sys/callout.h>
40 #include <sys/socket.h>
41 
42 #include <sys/bus.h>
43 
44 #include <net/if.h>
45 #include <net/if_dl.h>
46 #include <net/if_media.h>
47 #include <net/if_ether.h>
48 #include <net/if_arp.h>
49 
50 #ifdef INET
51 #include <netinet/in.h>
52 #include <netinet/in_systm.h>
53 #include <netinet/in_var.h>
54 #include <netinet/ip.h>
55 #include <netinet/if_inarp.h>
56 #endif
57 
58 #if NVLAN > 0
59 #include <net/if_types.h>
60 #endif
61 
62 #include <net/bpf.h>
63 
64 #include <dev/mii/mii.h>
65 #include <dev/mii/miivar.h>
66 
67 #include <dev/pci/pcireg.h>
68 #include <dev/pci/pcivar.h>
69 #include <dev/pci/pcidevs.h>
70 
71 #include <dev/pci/if_nfereg.h>
72 #include <dev/pci/if_nfevar.h>
73 
74 static int nfe_ifflags_cb(struct ethercom *);
75 
76 int	nfe_match(device_t, cfdata_t, void *);
77 void	nfe_attach(device_t, device_t, void *);
78 int	nfe_detach(device_t, int);
79 void	nfe_power(int, void *);
80 void	nfe_miibus_statchg(struct ifnet *);
81 int	nfe_miibus_readreg(device_t, int, int);
82 void	nfe_miibus_writereg(device_t, int, int, int);
83 int	nfe_intr(void *);
84 int	nfe_ioctl(struct ifnet *, u_long, void *);
85 void	nfe_txdesc32_sync(struct nfe_softc *, struct nfe_desc32 *, int);
86 void	nfe_txdesc64_sync(struct nfe_softc *, struct nfe_desc64 *, int);
87 void	nfe_txdesc32_rsync(struct nfe_softc *, int, int, int);
88 void	nfe_txdesc64_rsync(struct nfe_softc *, int, int, int);
89 void	nfe_rxdesc32_sync(struct nfe_softc *, struct nfe_desc32 *, int);
90 void	nfe_rxdesc64_sync(struct nfe_softc *, struct nfe_desc64 *, int);
91 void	nfe_rxeof(struct nfe_softc *);
92 void	nfe_txeof(struct nfe_softc *);
93 int	nfe_encap(struct nfe_softc *, struct mbuf *);
94 void	nfe_start(struct ifnet *);
95 void	nfe_watchdog(struct ifnet *);
96 int	nfe_init(struct ifnet *);
97 void	nfe_stop(struct ifnet *, int);
98 struct	nfe_jbuf *nfe_jalloc(struct nfe_softc *, int);
99 void	nfe_jfree(struct mbuf *, void *, size_t, void *);
100 int	nfe_jpool_alloc(struct nfe_softc *);
101 void	nfe_jpool_free(struct nfe_softc *);
102 int	nfe_alloc_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
103 void	nfe_reset_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
104 void	nfe_free_rx_ring(struct nfe_softc *, struct nfe_rx_ring *);
105 int	nfe_alloc_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
106 void	nfe_reset_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
107 void	nfe_free_tx_ring(struct nfe_softc *, struct nfe_tx_ring *);
108 void	nfe_setmulti(struct nfe_softc *);
109 void	nfe_get_macaddr(struct nfe_softc *, uint8_t *);
110 void	nfe_set_macaddr(struct nfe_softc *, const uint8_t *);
111 void	nfe_tick(void *);
112 void	nfe_poweron(device_t);
113 bool	nfe_resume(device_t, const pmf_qual_t *);
114 
115 CFATTACH_DECL_NEW(nfe, sizeof(struct nfe_softc),
116     nfe_match, nfe_attach, nfe_detach, NULL);
117 
118 /* #define NFE_NO_JUMBO */
119 
120 #ifdef NFE_DEBUG
121 int nfedebug = 0;
122 #define DPRINTF(x)	do { if (nfedebug) printf x; } while (0)
123 #define DPRINTFN(n,x)	do { if (nfedebug >= (n)) printf x; } while (0)
124 #else
125 #define DPRINTF(x)
126 #define DPRINTFN(n,x)
127 #endif
128 
129 /* deal with naming differences */
130 
131 #define	PCI_PRODUCT_NVIDIA_NFORCE3_LAN2 \
132 	PCI_PRODUCT_NVIDIA_NFORCE2_400_LAN1
133 #define	PCI_PRODUCT_NVIDIA_NFORCE3_LAN3 \
134 	PCI_PRODUCT_NVIDIA_NFORCE2_400_LAN2
135 #define	PCI_PRODUCT_NVIDIA_NFORCE3_LAN5 \
136 	PCI_PRODUCT_NVIDIA_NFORCE3_250_LAN
137 
138 #define	PCI_PRODUCT_NVIDIA_CK804_LAN1 \
139 	PCI_PRODUCT_NVIDIA_NFORCE4_LAN1
140 #define	PCI_PRODUCT_NVIDIA_CK804_LAN2 \
141 	PCI_PRODUCT_NVIDIA_NFORCE4_LAN2
142 
143 #define	PCI_PRODUCT_NVIDIA_MCP51_LAN1 \
144 	PCI_PRODUCT_NVIDIA_NFORCE430_LAN1
145 #define	PCI_PRODUCT_NVIDIA_MCP51_LAN2 \
146 	PCI_PRODUCT_NVIDIA_NFORCE430_LAN2
147 
148 #ifdef	_LP64
149 #define	__LP64__ 1
150 #endif
151 
152 const struct nfe_product {
153 	pci_vendor_id_t		vendor;
154 	pci_product_id_t	product;
155 } nfe_devices[] = {
156 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE_LAN },
157 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE2_LAN },
158 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN1 },
159 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN2 },
160 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN3 },
161 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN4 },
162 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE3_LAN5 },
163 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_CK804_LAN1 },
164 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_CK804_LAN2 },
165 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP04_LAN1 },
166 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP04_LAN2 },
167 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP51_LAN1 },
168 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP51_LAN2 },
169 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP55_LAN1 },
170 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP55_LAN2 },
171 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN1 },
172 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN2 },
173 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN3 },
174 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP61_LAN4 },
175 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN1 },
176 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN2 },
177 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN3 },
178 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP65_LAN4 },
179 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN1 },
180 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN2 },
181 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN3 },
182 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP67_LAN4 },
183 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP73_LAN1 },
184 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP73_LAN2 },
185 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP73_LAN3 },
186 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP73_LAN4 },
187 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP77_LAN1 },
188 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP77_LAN2 },
189 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP77_LAN3 },
190 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP77_LAN4 },
191 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP79_LAN1 },
192 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP79_LAN2 },
193 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP79_LAN3 },
194 	{ PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_MCP79_LAN4 }
195 };
196 
197 int
198 nfe_match(device_t dev, cfdata_t match, void *aux)
199 {
200 	struct pci_attach_args *pa = aux;
201 	const struct nfe_product *np;
202 	int i;
203 
204 	for (i = 0; i < __arraycount(nfe_devices); i++) {
205 		np = &nfe_devices[i];
206 		if (PCI_VENDOR(pa->pa_id) == np->vendor &&
207 		    PCI_PRODUCT(pa->pa_id) == np->product)
208 			return 1;
209 	}
210 	return 0;
211 }
212 
213 void
214 nfe_attach(device_t parent, device_t self, void *aux)
215 {
216 	struct nfe_softc *sc = device_private(self);
217 	struct pci_attach_args *pa = aux;
218 	pci_chipset_tag_t pc = pa->pa_pc;
219 	pci_intr_handle_t ih;
220 	const char *intrstr;
221 	struct ifnet *ifp;
222 	pcireg_t memtype, csr;
223 	int mii_flags = 0;
224 	char intrbuf[PCI_INTRSTR_LEN];
225 
226 	sc->sc_dev = self;
227 	sc->sc_pc = pa->pa_pc;
228 	pci_aprint_devinfo(pa, NULL);
229 
230 	memtype = pci_mapreg_type(pa->pa_pc, pa->pa_tag, NFE_PCI_BA);
231 	switch (memtype) {
232 	case PCI_MAPREG_TYPE_MEM | PCI_MAPREG_MEM_TYPE_32BIT:
233 	case PCI_MAPREG_TYPE_MEM | PCI_MAPREG_MEM_TYPE_64BIT:
234 		if (pci_mapreg_map(pa, NFE_PCI_BA, memtype, 0, &sc->sc_memt,
235 		    &sc->sc_memh, NULL, &sc->sc_mems) == 0)
236 			break;
237 		/* FALLTHROUGH */
238 	default:
239 		aprint_error_dev(self, "could not map mem space\n");
240 		return;
241 	}
242 
243 	if (pci_intr_map(pa, &ih) != 0) {
244 		aprint_error_dev(self, "could not map interrupt\n");
245 		goto fail;
246 	}
247 
248 	intrstr = pci_intr_string(pc, ih, intrbuf, sizeof(intrbuf));
249 	sc->sc_ih = pci_intr_establish(pc, ih, IPL_NET, nfe_intr, sc);
250 	if (sc->sc_ih == NULL) {
251 		aprint_error_dev(self, "could not establish interrupt");
252 		if (intrstr != NULL)
253 			aprint_error(" at %s", intrstr);
254 		aprint_error("\n");
255 		goto fail;
256 	}
257 	aprint_normal_dev(self, "interrupting at %s\n", intrstr);
258 
259 	csr = pci_conf_read(pa->pa_pc, pa->pa_tag, PCI_COMMAND_STATUS_REG);
260 	csr |= PCI_COMMAND_MASTER_ENABLE;
261 	pci_conf_write(pa->pa_pc, pa->pa_tag, PCI_COMMAND_STATUS_REG, csr);
262 
263 	sc->sc_flags = 0;
264 
265 	switch (PCI_PRODUCT(pa->pa_id)) {
266 	case PCI_PRODUCT_NVIDIA_NFORCE3_LAN2:
267 	case PCI_PRODUCT_NVIDIA_NFORCE3_LAN3:
268 	case PCI_PRODUCT_NVIDIA_NFORCE3_LAN4:
269 	case PCI_PRODUCT_NVIDIA_NFORCE3_LAN5:
270 		sc->sc_flags |= NFE_JUMBO_SUP | NFE_HW_CSUM;
271 		break;
272 	case PCI_PRODUCT_NVIDIA_MCP51_LAN1:
273 	case PCI_PRODUCT_NVIDIA_MCP51_LAN2:
274 		sc->sc_flags |= NFE_40BIT_ADDR | NFE_PWR_MGMT;
275 		break;
276 	case PCI_PRODUCT_NVIDIA_MCP61_LAN1:
277 	case PCI_PRODUCT_NVIDIA_MCP61_LAN2:
278 	case PCI_PRODUCT_NVIDIA_MCP61_LAN3:
279 	case PCI_PRODUCT_NVIDIA_MCP61_LAN4:
280 	case PCI_PRODUCT_NVIDIA_MCP67_LAN1:
281 	case PCI_PRODUCT_NVIDIA_MCP67_LAN2:
282 	case PCI_PRODUCT_NVIDIA_MCP67_LAN3:
283 	case PCI_PRODUCT_NVIDIA_MCP67_LAN4:
284 	case PCI_PRODUCT_NVIDIA_MCP73_LAN1:
285 	case PCI_PRODUCT_NVIDIA_MCP73_LAN2:
286 	case PCI_PRODUCT_NVIDIA_MCP73_LAN3:
287 	case PCI_PRODUCT_NVIDIA_MCP73_LAN4:
288 		sc->sc_flags |= NFE_40BIT_ADDR | NFE_CORRECT_MACADDR |
289 		    NFE_PWR_MGMT;
290 		break;
291 	case PCI_PRODUCT_NVIDIA_MCP77_LAN1:
292 	case PCI_PRODUCT_NVIDIA_MCP77_LAN2:
293 	case PCI_PRODUCT_NVIDIA_MCP77_LAN3:
294 	case PCI_PRODUCT_NVIDIA_MCP77_LAN4:
295 		sc->sc_flags |= NFE_40BIT_ADDR | NFE_HW_CSUM |
296 		    NFE_CORRECT_MACADDR | NFE_PWR_MGMT;
297 		break;
298 	case PCI_PRODUCT_NVIDIA_MCP79_LAN1:
299 	case PCI_PRODUCT_NVIDIA_MCP79_LAN2:
300 	case PCI_PRODUCT_NVIDIA_MCP79_LAN3:
301 	case PCI_PRODUCT_NVIDIA_MCP79_LAN4:
302 		sc->sc_flags |= NFE_JUMBO_SUP | NFE_40BIT_ADDR | NFE_HW_CSUM |
303 		    NFE_CORRECT_MACADDR | NFE_PWR_MGMT;
304 		break;
305 	case PCI_PRODUCT_NVIDIA_CK804_LAN1:
306 	case PCI_PRODUCT_NVIDIA_CK804_LAN2:
307 	case PCI_PRODUCT_NVIDIA_MCP04_LAN1:
308 	case PCI_PRODUCT_NVIDIA_MCP04_LAN2:
309 		sc->sc_flags |= NFE_JUMBO_SUP | NFE_40BIT_ADDR | NFE_HW_CSUM;
310 		break;
311 	case PCI_PRODUCT_NVIDIA_MCP65_LAN1:
312 	case PCI_PRODUCT_NVIDIA_MCP65_LAN2:
313 	case PCI_PRODUCT_NVIDIA_MCP65_LAN3:
314 	case PCI_PRODUCT_NVIDIA_MCP65_LAN4:
315 		sc->sc_flags |= NFE_JUMBO_SUP | NFE_40BIT_ADDR |
316 		    NFE_CORRECT_MACADDR | NFE_PWR_MGMT;
317 		mii_flags = MIIF_DOPAUSE;
318 		break;
319 	case PCI_PRODUCT_NVIDIA_MCP55_LAN1:
320 	case PCI_PRODUCT_NVIDIA_MCP55_LAN2:
321 		sc->sc_flags |= NFE_JUMBO_SUP | NFE_40BIT_ADDR | NFE_HW_CSUM |
322 		    NFE_HW_VLAN | NFE_PWR_MGMT;
323 		break;
324 	}
325 
326 	if (pci_dma64_available(pa) && (sc->sc_flags & NFE_40BIT_ADDR) != 0)
327 		sc->sc_dmat = pa->pa_dmat64;
328 	else
329 		sc->sc_dmat = pa->pa_dmat;
330 
331 	nfe_poweron(self);
332 
333 #ifndef NFE_NO_JUMBO
334 	/* enable jumbo frames for adapters that support it */
335 	if (sc->sc_flags & NFE_JUMBO_SUP)
336 		sc->sc_flags |= NFE_USE_JUMBO;
337 #endif
338 
339 	/* Check for reversed ethernet address */
340 	if ((NFE_READ(sc, NFE_TX_UNK) & NFE_MAC_ADDR_INORDER) != 0)
341 		sc->sc_flags |= NFE_CORRECT_MACADDR;
342 
343 	nfe_get_macaddr(sc, sc->sc_enaddr);
344 	aprint_normal_dev(self, "Ethernet address %s\n",
345 	    ether_sprintf(sc->sc_enaddr));
346 
347 	/*
348 	 * Allocate Tx and Rx rings.
349 	 */
350 	if (nfe_alloc_tx_ring(sc, &sc->txq) != 0) {
351 		aprint_error_dev(self, "could not allocate Tx ring\n");
352 		goto fail;
353 	}
354 
355 	mutex_init(&sc->rxq.mtx, MUTEX_DEFAULT, IPL_NET);
356 
357 	if (nfe_alloc_rx_ring(sc, &sc->rxq) != 0) {
358 		aprint_error_dev(self, "could not allocate Rx ring\n");
359 		nfe_free_tx_ring(sc, &sc->txq);
360 		goto fail;
361 	}
362 
363 	ifp = &sc->sc_ethercom.ec_if;
364 	ifp->if_softc = sc;
365 	ifp->if_mtu = ETHERMTU;
366 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
367 	ifp->if_ioctl = nfe_ioctl;
368 	ifp->if_start = nfe_start;
369 	ifp->if_stop = nfe_stop;
370 	ifp->if_watchdog = nfe_watchdog;
371 	ifp->if_init = nfe_init;
372 	ifp->if_baudrate = IF_Gbps(1);
373 	IFQ_SET_MAXLEN(&ifp->if_snd, NFE_IFQ_MAXLEN);
374 	IFQ_SET_READY(&ifp->if_snd);
375 	strlcpy(ifp->if_xname, device_xname(self), IFNAMSIZ);
376 
377 	if (sc->sc_flags & NFE_USE_JUMBO)
378 		sc->sc_ethercom.ec_capabilities |= ETHERCAP_JUMBO_MTU;
379 
380 #if NVLAN > 0
381 	if (sc->sc_flags & NFE_HW_VLAN)
382 		sc->sc_ethercom.ec_capabilities |=
383 			ETHERCAP_VLAN_HWTAGGING | ETHERCAP_VLAN_MTU;
384 #endif
385 	if (sc->sc_flags & NFE_HW_CSUM) {
386 		ifp->if_capabilities |=
387 		    IFCAP_CSUM_IPv4_Tx | IFCAP_CSUM_IPv4_Rx |
388 		    IFCAP_CSUM_TCPv4_Tx | IFCAP_CSUM_TCPv4_Rx |
389 		    IFCAP_CSUM_UDPv4_Tx | IFCAP_CSUM_UDPv4_Rx;
390 	}
391 
392 	sc->sc_mii.mii_ifp = ifp;
393 	sc->sc_mii.mii_readreg = nfe_miibus_readreg;
394 	sc->sc_mii.mii_writereg = nfe_miibus_writereg;
395 	sc->sc_mii.mii_statchg = nfe_miibus_statchg;
396 
397 	sc->sc_ethercom.ec_mii = &sc->sc_mii;
398 	ifmedia_init(&sc->sc_mii.mii_media, 0, ether_mediachange,
399 	    ether_mediastatus);
400 
401 	mii_attach(self, &sc->sc_mii, 0xffffffff, MII_PHY_ANY, 0, mii_flags);
402 
403 	if (LIST_FIRST(&sc->sc_mii.mii_phys) == NULL) {
404 		aprint_error_dev(self, "no PHY found!\n");
405 		ifmedia_add(&sc->sc_mii.mii_media, IFM_ETHER | IFM_MANUAL,
406 		    0, NULL);
407 		ifmedia_set(&sc->sc_mii.mii_media, IFM_ETHER | IFM_MANUAL);
408 	} else
409 		ifmedia_set(&sc->sc_mii.mii_media, IFM_ETHER | IFM_AUTO);
410 
411 	if_attach(ifp);
412 	if_deferred_start_init(ifp, NULL);
413 	ether_ifattach(ifp, sc->sc_enaddr);
414 	ether_set_ifflags_cb(&sc->sc_ethercom, nfe_ifflags_cb);
415 
416 	callout_init(&sc->sc_tick_ch, 0);
417 	callout_setfunc(&sc->sc_tick_ch, nfe_tick, sc);
418 
419 	if (pmf_device_register(self, NULL, nfe_resume))
420 		pmf_class_network_register(self, ifp);
421 	else
422 		aprint_error_dev(self, "couldn't establish power handler\n");
423 
424 	return;
425 
426 fail:
427 	if (sc->sc_ih != NULL) {
428 		pci_intr_disestablish(pc, sc->sc_ih);
429 		sc->sc_ih = NULL;
430 	}
431 	if (sc->sc_mems != 0) {
432 		bus_space_unmap(sc->sc_memt, sc->sc_memh, sc->sc_mems);
433 		sc->sc_mems = 0;
434 	}
435 }
436 
437 int
438 nfe_detach(device_t self, int flags)
439 {
440 	struct nfe_softc *sc = device_private(self);
441 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
442 	int s;
443 
444 	s = splnet();
445 
446 	nfe_stop(ifp, 1);
447 
448 	pmf_device_deregister(self);
449 	callout_destroy(&sc->sc_tick_ch);
450 	ether_ifdetach(ifp);
451 	if_detach(ifp);
452 	mii_detach(&sc->sc_mii, MII_PHY_ANY, MII_OFFSET_ANY);
453 
454 	nfe_free_rx_ring(sc, &sc->rxq);
455 	mutex_destroy(&sc->rxq.mtx);
456 	nfe_free_tx_ring(sc, &sc->txq);
457 
458 	if (sc->sc_ih != NULL) {
459 		pci_intr_disestablish(sc->sc_pc, sc->sc_ih);
460 		sc->sc_ih = NULL;
461 	}
462 
463 	if ((sc->sc_flags & NFE_CORRECT_MACADDR) != 0) {
464 		nfe_set_macaddr(sc, sc->sc_enaddr);
465 	} else {
466 		NFE_WRITE(sc, NFE_MACADDR_LO,
467 		    sc->sc_enaddr[0] <<  8 | sc->sc_enaddr[1]);
468 		NFE_WRITE(sc, NFE_MACADDR_HI,
469 		    sc->sc_enaddr[2] << 24 | sc->sc_enaddr[3] << 16 |
470 		    sc->sc_enaddr[4] <<  8 | sc->sc_enaddr[5]);
471 	}
472 
473 	if (sc->sc_mems != 0) {
474 		bus_space_unmap(sc->sc_memt, sc->sc_memh, sc->sc_mems);
475 		sc->sc_mems = 0;
476 	}
477 
478 	splx(s);
479 
480 	return 0;
481 }
482 
483 void
484 nfe_miibus_statchg(struct ifnet *ifp)
485 {
486 	struct nfe_softc *sc = ifp->if_softc;
487 	struct mii_data *mii = &sc->sc_mii;
488 	uint32_t phy, seed, misc = NFE_MISC1_MAGIC, link = NFE_MEDIA_SET;
489 
490 	phy = NFE_READ(sc, NFE_PHY_IFACE);
491 	phy &= ~(NFE_PHY_HDX | NFE_PHY_100TX | NFE_PHY_1000T);
492 
493 	seed = NFE_READ(sc, NFE_RNDSEED);
494 	seed &= ~NFE_SEED_MASK;
495 
496 	if ((mii->mii_media_active & IFM_GMASK) == IFM_HDX) {
497 		phy  |= NFE_PHY_HDX;	/* half-duplex */
498 		misc |= NFE_MISC1_HDX;
499 	}
500 
501 	switch (IFM_SUBTYPE(mii->mii_media_active)) {
502 	case IFM_1000_T:	/* full-duplex only */
503 		link |= NFE_MEDIA_1000T;
504 		seed |= NFE_SEED_1000T;
505 		phy  |= NFE_PHY_1000T;
506 		break;
507 	case IFM_100_TX:
508 		link |= NFE_MEDIA_100TX;
509 		seed |= NFE_SEED_100TX;
510 		phy  |= NFE_PHY_100TX;
511 		break;
512 	case IFM_10_T:
513 		link |= NFE_MEDIA_10T;
514 		seed |= NFE_SEED_10T;
515 		break;
516 	}
517 
518 	NFE_WRITE(sc, NFE_RNDSEED, seed);	/* XXX: gigabit NICs only? */
519 
520 	NFE_WRITE(sc, NFE_PHY_IFACE, phy);
521 	NFE_WRITE(sc, NFE_MISC1, misc);
522 	NFE_WRITE(sc, NFE_LINKSPEED, link);
523 }
524 
525 int
526 nfe_miibus_readreg(device_t dev, int phy, int reg)
527 {
528 	struct nfe_softc *sc = device_private(dev);
529 	uint32_t val;
530 	int ntries;
531 
532 	NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
533 
534 	if (NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY) {
535 		NFE_WRITE(sc, NFE_PHY_CTL, NFE_PHY_BUSY);
536 		DELAY(100);
537 	}
538 
539 	NFE_WRITE(sc, NFE_PHY_CTL, (phy << NFE_PHYADD_SHIFT) | reg);
540 
541 	for (ntries = 0; ntries < 1000; ntries++) {
542 		DELAY(100);
543 		if (!(NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY))
544 			break;
545 	}
546 	if (ntries == 1000) {
547 		DPRINTFN(2, ("%s: timeout waiting for PHY\n",
548 		    device_xname(sc->sc_dev)));
549 		return 0;
550 	}
551 
552 	if (NFE_READ(sc, NFE_PHY_STATUS) & NFE_PHY_ERROR) {
553 		DPRINTFN(2, ("%s: could not read PHY\n",
554 		    device_xname(sc->sc_dev)));
555 		return 0;
556 	}
557 
558 	val = NFE_READ(sc, NFE_PHY_DATA);
559 	if (val != 0xffffffff && val != 0)
560 		sc->mii_phyaddr = phy;
561 
562 	DPRINTFN(2, ("%s: mii read phy %d reg 0x%x ret 0x%x\n",
563 	    device_xname(sc->sc_dev), phy, reg, val));
564 
565 	return val;
566 }
567 
568 void
569 nfe_miibus_writereg(device_t dev, int phy, int reg, int val)
570 {
571 	struct nfe_softc *sc = device_private(dev);
572 	uint32_t ctl;
573 	int ntries;
574 
575 	NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
576 
577 	if (NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY) {
578 		NFE_WRITE(sc, NFE_PHY_CTL, NFE_PHY_BUSY);
579 		DELAY(100);
580 	}
581 
582 	NFE_WRITE(sc, NFE_PHY_DATA, val);
583 	ctl = NFE_PHY_WRITE | (phy << NFE_PHYADD_SHIFT) | reg;
584 	NFE_WRITE(sc, NFE_PHY_CTL, ctl);
585 
586 	for (ntries = 0; ntries < 1000; ntries++) {
587 		DELAY(100);
588 		if (!(NFE_READ(sc, NFE_PHY_CTL) & NFE_PHY_BUSY))
589 			break;
590 	}
591 #ifdef NFE_DEBUG
592 	if (nfedebug >= 2 && ntries == 1000)
593 		printf("could not write to PHY\n");
594 #endif
595 }
596 
597 int
598 nfe_intr(void *arg)
599 {
600 	struct nfe_softc *sc = arg;
601 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
602 	uint32_t r;
603 	int handled;
604 
605 	if ((ifp->if_flags & IFF_UP) == 0)
606 		return 0;
607 
608 	handled = 0;
609 
610 	for (;;) {
611 		r = NFE_READ(sc, NFE_IRQ_STATUS);
612 		if ((r & NFE_IRQ_WANTED) == 0)
613 			break;
614 
615 		NFE_WRITE(sc, NFE_IRQ_STATUS, r);
616 		handled = 1;
617 		DPRINTFN(5, ("nfe_intr: interrupt register %x\n", r));
618 
619 		if ((r & (NFE_IRQ_RXERR|NFE_IRQ_RX_NOBUF|NFE_IRQ_RX)) != 0) {
620 			/* check Rx ring */
621 			nfe_rxeof(sc);
622 		}
623 		if ((r & (NFE_IRQ_TXERR|NFE_IRQ_TXERR2|NFE_IRQ_TX_DONE)) != 0) {
624 			/* check Tx ring */
625 			nfe_txeof(sc);
626 		}
627 		if ((r & NFE_IRQ_LINK) != 0) {
628 			NFE_READ(sc, NFE_PHY_STATUS);
629 			NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
630 			DPRINTF(("%s: link state changed\n",
631 			    device_xname(sc->sc_dev)));
632 		}
633 	}
634 
635 	if (handled)
636 		if_schedule_deferred_start(ifp);
637 
638 	return handled;
639 }
640 
641 static int
642 nfe_ifflags_cb(struct ethercom *ec)
643 {
644 	struct ifnet *ifp = &ec->ec_if;
645 	struct nfe_softc *sc = ifp->if_softc;
646 	int change = ifp->if_flags ^ sc->sc_if_flags;
647 
648 	/*
649 	 * If only the PROMISC flag changes, then
650 	 * don't do a full re-init of the chip, just update
651 	 * the Rx filter.
652 	 */
653 	if ((change & ~(IFF_CANTCHANGE|IFF_DEBUG)) != 0)
654 		return ENETRESET;
655 	else if ((change & IFF_PROMISC) != 0)
656 		nfe_setmulti(sc);
657 
658 	return 0;
659 }
660 
661 int
662 nfe_ioctl(struct ifnet *ifp, u_long cmd, void *data)
663 {
664 	struct nfe_softc *sc = ifp->if_softc;
665 	struct ifaddr *ifa = (struct ifaddr *)data;
666 	int s, error = 0;
667 
668 	s = splnet();
669 
670 	switch (cmd) {
671 	case SIOCINITIFADDR:
672 		ifp->if_flags |= IFF_UP;
673 		nfe_init(ifp);
674 		switch (ifa->ifa_addr->sa_family) {
675 #ifdef INET
676 		case AF_INET:
677 			arp_ifinit(ifp, ifa);
678 			break;
679 #endif
680 		default:
681 			break;
682 		}
683 		break;
684 	default:
685 		if ((error = ether_ioctl(ifp, cmd, data)) != ENETRESET)
686 			break;
687 
688 		error = 0;
689 
690 		if (cmd != SIOCADDMULTI && cmd != SIOCDELMULTI)
691 			;
692 		else if (ifp->if_flags & IFF_RUNNING)
693 			nfe_setmulti(sc);
694 		break;
695 	}
696 	sc->sc_if_flags = ifp->if_flags;
697 
698 	splx(s);
699 
700 	return error;
701 }
702 
703 void
704 nfe_txdesc32_sync(struct nfe_softc *sc, struct nfe_desc32 *desc32, int ops)
705 {
706 	bus_dmamap_sync(sc->sc_dmat, sc->txq.map,
707 	    (char *)desc32 - (char *)sc->txq.desc32,
708 	    sizeof (struct nfe_desc32), ops);
709 }
710 
711 void
712 nfe_txdesc64_sync(struct nfe_softc *sc, struct nfe_desc64 *desc64, int ops)
713 {
714 	bus_dmamap_sync(sc->sc_dmat, sc->txq.map,
715 	    (char *)desc64 - (char *)sc->txq.desc64,
716 	    sizeof (struct nfe_desc64), ops);
717 }
718 
719 void
720 nfe_txdesc32_rsync(struct nfe_softc *sc, int start, int end, int ops)
721 {
722 	if (end > start) {
723 		bus_dmamap_sync(sc->sc_dmat, sc->txq.map,
724 		    (char *)&sc->txq.desc32[start] - (char *)sc->txq.desc32,
725 		    (char *)&sc->txq.desc32[end] -
726 		    (char *)&sc->txq.desc32[start], ops);
727 		return;
728 	}
729 	/* sync from 'start' to end of ring */
730 	bus_dmamap_sync(sc->sc_dmat, sc->txq.map,
731 	    (char *)&sc->txq.desc32[start] - (char *)sc->txq.desc32,
732 	    (char *)&sc->txq.desc32[NFE_TX_RING_COUNT] -
733 	    (char *)&sc->txq.desc32[start], ops);
734 
735 	/* sync from start of ring to 'end' */
736 	bus_dmamap_sync(sc->sc_dmat, sc->txq.map, 0,
737 	    (char *)&sc->txq.desc32[end] - (char *)sc->txq.desc32, ops);
738 }
739 
740 void
741 nfe_txdesc64_rsync(struct nfe_softc *sc, int start, int end, int ops)
742 {
743 	if (end > start) {
744 		bus_dmamap_sync(sc->sc_dmat, sc->txq.map,
745 		    (char *)&sc->txq.desc64[start] - (char *)sc->txq.desc64,
746 		    (char *)&sc->txq.desc64[end] -
747 		    (char *)&sc->txq.desc64[start], ops);
748 		return;
749 	}
750 	/* sync from 'start' to end of ring */
751 	bus_dmamap_sync(sc->sc_dmat, sc->txq.map,
752 	    (char *)&sc->txq.desc64[start] - (char *)sc->txq.desc64,
753 	    (char *)&sc->txq.desc64[NFE_TX_RING_COUNT] -
754 	    (char *)&sc->txq.desc64[start], ops);
755 
756 	/* sync from start of ring to 'end' */
757 	bus_dmamap_sync(sc->sc_dmat, sc->txq.map, 0,
758 	    (char *)&sc->txq.desc64[end] - (char *)sc->txq.desc64, ops);
759 }
760 
761 void
762 nfe_rxdesc32_sync(struct nfe_softc *sc, struct nfe_desc32 *desc32, int ops)
763 {
764 	bus_dmamap_sync(sc->sc_dmat, sc->rxq.map,
765 	    (char *)desc32 - (char *)sc->rxq.desc32,
766 	    sizeof (struct nfe_desc32), ops);
767 }
768 
769 void
770 nfe_rxdesc64_sync(struct nfe_softc *sc, struct nfe_desc64 *desc64, int ops)
771 {
772 	bus_dmamap_sync(sc->sc_dmat, sc->rxq.map,
773 	    (char *)desc64 - (char *)sc->rxq.desc64,
774 	    sizeof (struct nfe_desc64), ops);
775 }
776 
777 void
778 nfe_rxeof(struct nfe_softc *sc)
779 {
780 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
781 	struct nfe_desc32 *desc32;
782 	struct nfe_desc64 *desc64;
783 	struct nfe_rx_data *data;
784 	struct nfe_jbuf *jbuf;
785 	struct mbuf *m, *mnew;
786 	bus_addr_t physaddr;
787 	uint16_t flags;
788 	int error, len, i;
789 
790 	desc32 = NULL;
791 	desc64 = NULL;
792 	for (i = sc->rxq.cur;; i = NFE_RX_NEXTDESC(i)) {
793 		data = &sc->rxq.data[i];
794 
795 		if (sc->sc_flags & NFE_40BIT_ADDR) {
796 			desc64 = &sc->rxq.desc64[i];
797 			nfe_rxdesc64_sync(sc, desc64,
798 			    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
799 
800 			flags = le16toh(desc64->flags);
801 			len = le16toh(desc64->length) & 0x3fff;
802 		} else {
803 			desc32 = &sc->rxq.desc32[i];
804 			nfe_rxdesc32_sync(sc, desc32,
805 			    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
806 
807 			flags = le16toh(desc32->flags);
808 			len = le16toh(desc32->length) & 0x3fff;
809 		}
810 
811 		if ((flags & NFE_RX_READY) != 0)
812 			break;
813 
814 		if ((sc->sc_flags & (NFE_JUMBO_SUP | NFE_40BIT_ADDR)) == 0) {
815 			if ((flags & NFE_RX_VALID_V1) == 0)
816 				goto skip;
817 
818 			if ((flags & NFE_RX_FIXME_V1) == NFE_RX_FIXME_V1) {
819 				flags &= ~NFE_RX_ERROR;
820 				len--;	/* fix buffer length */
821 			}
822 		} else {
823 			if ((flags & NFE_RX_VALID_V2) == 0)
824 				goto skip;
825 
826 			if ((flags & NFE_RX_FIXME_V2) == NFE_RX_FIXME_V2) {
827 				flags &= ~NFE_RX_ERROR;
828 				len--;	/* fix buffer length */
829 			}
830 		}
831 
832 		if (flags & NFE_RX_ERROR) {
833 			ifp->if_ierrors++;
834 			goto skip;
835 		}
836 
837 		/*
838 		 * Try to allocate a new mbuf for this ring element and load
839 		 * it before processing the current mbuf. If the ring element
840 		 * cannot be loaded, drop the received packet and reuse the
841 		 * old mbuf. In the unlikely case that the old mbuf can't be
842 		 * reloaded either, explicitly panic.
843 		 */
844 		MGETHDR(mnew, M_DONTWAIT, MT_DATA);
845 		if (mnew == NULL) {
846 			ifp->if_ierrors++;
847 			goto skip;
848 		}
849 
850 		if (sc->sc_flags & NFE_USE_JUMBO) {
851 			physaddr =
852 			    sc->rxq.jbuf[sc->rxq.jbufmap[i]].physaddr;
853 			if ((jbuf = nfe_jalloc(sc, i)) == NULL) {
854 				if (len > MCLBYTES) {
855 					m_freem(mnew);
856 					ifp->if_ierrors++;
857 					goto skip1;
858 				}
859 				MCLGET(mnew, M_DONTWAIT);
860 				if ((mnew->m_flags & M_EXT) == 0) {
861 					m_freem(mnew);
862 					ifp->if_ierrors++;
863 					goto skip1;
864 				}
865 
866 				(void)memcpy(mtod(mnew, void *),
867 				    mtod(data->m, const void *), len);
868 				m = mnew;
869 				goto mbufcopied;
870 			} else {
871 				MEXTADD(mnew, jbuf->buf, NFE_JBYTES, 0, nfe_jfree, sc);
872 				bus_dmamap_sync(sc->sc_dmat, sc->rxq.jmap,
873 				    mtod(data->m, char *) - (char *)sc->rxq.jpool,
874 				    NFE_JBYTES, BUS_DMASYNC_POSTREAD);
875 
876 				physaddr = jbuf->physaddr;
877 			}
878 		} else {
879 			MCLGET(mnew, M_DONTWAIT);
880 			if ((mnew->m_flags & M_EXT) == 0) {
881 				m_freem(mnew);
882 				ifp->if_ierrors++;
883 				goto skip;
884 			}
885 
886 			bus_dmamap_sync(sc->sc_dmat, data->map, 0,
887 			    data->map->dm_mapsize, BUS_DMASYNC_POSTREAD);
888 			bus_dmamap_unload(sc->sc_dmat, data->map);
889 
890 			error = bus_dmamap_load(sc->sc_dmat, data->map,
891 			    mtod(mnew, void *), MCLBYTES, NULL,
892 			    BUS_DMA_READ | BUS_DMA_NOWAIT);
893 			if (error != 0) {
894 				m_freem(mnew);
895 
896 				/* try to reload the old mbuf */
897 				error = bus_dmamap_load(sc->sc_dmat, data->map,
898 				    mtod(data->m, void *), MCLBYTES, NULL,
899 				    BUS_DMA_READ | BUS_DMA_NOWAIT);
900 				if (error != 0) {
901 					/* very unlikely that it will fail.. */
902 					panic("%s: could not load old rx mbuf",
903 					    device_xname(sc->sc_dev));
904 				}
905 				ifp->if_ierrors++;
906 				goto skip;
907 			}
908 			physaddr = data->map->dm_segs[0].ds_addr;
909 		}
910 
911 		/*
912 		 * New mbuf successfully loaded, update Rx ring and continue
913 		 * processing.
914 		 */
915 		m = data->m;
916 		data->m = mnew;
917 
918 mbufcopied:
919 		/* finalize mbuf */
920 		m->m_pkthdr.len = m->m_len = len;
921 		m_set_rcvif(m, ifp);
922 
923 		if ((sc->sc_flags & NFE_HW_CSUM) != 0) {
924 			/*
925 			 * XXX
926 			 * no way to check M_CSUM_IPv4_BAD or non-IPv4 packets?
927 			 */
928 			if (flags & NFE_RX_IP_CSUMOK) {
929 				m->m_pkthdr.csum_flags |= M_CSUM_IPv4;
930 				DPRINTFN(3, ("%s: ip4csum-rx ok\n",
931 				    device_xname(sc->sc_dev)));
932 			}
933 			/*
934 			 * XXX
935 			 * no way to check M_CSUM_TCP_UDP_BAD or
936 			 * other protocols?
937 			 */
938 			if (flags & NFE_RX_UDP_CSUMOK) {
939 				m->m_pkthdr.csum_flags |= M_CSUM_UDPv4;
940 				DPRINTFN(3, ("%s: udp4csum-rx ok\n",
941 				    device_xname(sc->sc_dev)));
942 			} else if (flags & NFE_RX_TCP_CSUMOK) {
943 				m->m_pkthdr.csum_flags |= M_CSUM_TCPv4;
944 				DPRINTFN(3, ("%s: tcp4csum-rx ok\n",
945 				    device_xname(sc->sc_dev)));
946 			}
947 		}
948 		if_percpuq_enqueue(ifp->if_percpuq, m);
949 
950 skip1:
951 		/* update mapping address in h/w descriptor */
952 		if (sc->sc_flags & NFE_40BIT_ADDR) {
953 #if defined(__LP64__)
954 			desc64->physaddr[0] = htole32(physaddr >> 32);
955 #endif
956 			desc64->physaddr[1] = htole32(physaddr & 0xffffffff);
957 		} else {
958 			desc32->physaddr = htole32(physaddr);
959 		}
960 
961 skip:
962 		if (sc->sc_flags & NFE_40BIT_ADDR) {
963 			desc64->length = htole16(sc->rxq.bufsz);
964 			desc64->flags = htole16(NFE_RX_READY);
965 
966 			nfe_rxdesc64_sync(sc, desc64,
967 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
968 		} else {
969 			desc32->length = htole16(sc->rxq.bufsz);
970 			desc32->flags = htole16(NFE_RX_READY);
971 
972 			nfe_rxdesc32_sync(sc, desc32,
973 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
974 		}
975 	}
976 	/* update current RX pointer */
977 	sc->rxq.cur = i;
978 }
979 
980 void
981 nfe_txeof(struct nfe_softc *sc)
982 {
983 	struct ifnet *ifp = &sc->sc_ethercom.ec_if;
984 	struct nfe_desc32 *desc32;
985 	struct nfe_desc64 *desc64;
986 	struct nfe_tx_data *data = NULL;
987 	int i;
988 	uint16_t flags;
989 	char buf[128];
990 
991 	for (i = sc->txq.next;
992 	    sc->txq.queued > 0;
993 	    i = NFE_TX_NEXTDESC(i), sc->txq.queued--) {
994 		if (sc->sc_flags & NFE_40BIT_ADDR) {
995 			desc64 = &sc->txq.desc64[i];
996 			nfe_txdesc64_sync(sc, desc64,
997 			    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
998 
999 			flags = le16toh(desc64->flags);
1000 		} else {
1001 			desc32 = &sc->txq.desc32[i];
1002 			nfe_txdesc32_sync(sc, desc32,
1003 			    BUS_DMASYNC_POSTREAD|BUS_DMASYNC_POSTWRITE);
1004 
1005 			flags = le16toh(desc32->flags);
1006 		}
1007 
1008 		if ((flags & NFE_TX_VALID) != 0)
1009 			break;
1010 
1011 		data = &sc->txq.data[i];
1012 
1013 		if ((sc->sc_flags & (NFE_JUMBO_SUP | NFE_40BIT_ADDR)) == 0) {
1014 			if ((flags & NFE_TX_LASTFRAG_V1) == 0 &&
1015 			    data->m == NULL)
1016 				continue;
1017 
1018 			if ((flags & NFE_TX_ERROR_V1) != 0) {
1019 				snprintb(buf, sizeof(buf), NFE_V1_TXERR, flags);
1020 				aprint_error_dev(sc->sc_dev, "tx v1 error %s\n",
1021 				    buf);
1022 				ifp->if_oerrors++;
1023 			} else
1024 				ifp->if_opackets++;
1025 		} else {
1026 			if ((flags & NFE_TX_LASTFRAG_V2) == 0 &&
1027 			    data->m == NULL)
1028 				continue;
1029 
1030 			if ((flags & NFE_TX_ERROR_V2) != 0) {
1031 				snprintb(buf, sizeof(buf), NFE_V2_TXERR, flags);
1032 				aprint_error_dev(sc->sc_dev, "tx v2 error %s\n",
1033 				    buf);
1034 				ifp->if_oerrors++;
1035 			} else
1036 				ifp->if_opackets++;
1037 		}
1038 
1039 		if (data->m == NULL) {	/* should not get there */
1040 			aprint_error_dev(sc->sc_dev,
1041 			    "last fragment bit w/o associated mbuf!\n");
1042 			continue;
1043 		}
1044 
1045 		/* last fragment of the mbuf chain transmitted */
1046 		bus_dmamap_sync(sc->sc_dmat, data->active, 0,
1047 		    data->active->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1048 		bus_dmamap_unload(sc->sc_dmat, data->active);
1049 		m_freem(data->m);
1050 		data->m = NULL;
1051 	}
1052 
1053 	sc->txq.next = i;
1054 
1055 	if (sc->txq.queued < NFE_TX_RING_COUNT) {
1056 		/* at least one slot freed */
1057 		ifp->if_flags &= ~IFF_OACTIVE;
1058 	}
1059 
1060 	if (sc->txq.queued == 0) {
1061 		/* all queued packets are sent */
1062 		ifp->if_timer = 0;
1063 	}
1064 }
1065 
1066 int
1067 nfe_encap(struct nfe_softc *sc, struct mbuf *m0)
1068 {
1069 	struct nfe_desc32 *desc32;
1070 	struct nfe_desc64 *desc64;
1071 	struct nfe_tx_data *data;
1072 	bus_dmamap_t map;
1073 	uint16_t flags, csumflags;
1074 #if NVLAN > 0
1075 	struct m_tag *mtag;
1076 	uint32_t vtag = 0;
1077 #endif
1078 	int error, i, first;
1079 
1080 	desc32 = NULL;
1081 	desc64 = NULL;
1082 	data = NULL;
1083 
1084 	flags = 0;
1085 	csumflags = 0;
1086 	first = sc->txq.cur;
1087 
1088 	map = sc->txq.data[first].map;
1089 
1090 	error = bus_dmamap_load_mbuf(sc->sc_dmat, map, m0, BUS_DMA_NOWAIT);
1091 	if (error != 0) {
1092 		aprint_error_dev(sc->sc_dev, "could not map mbuf (error %d)\n",
1093 		    error);
1094 		return error;
1095 	}
1096 
1097 	if (sc->txq.queued + map->dm_nsegs >= NFE_TX_RING_COUNT - 1) {
1098 		bus_dmamap_unload(sc->sc_dmat, map);
1099 		return ENOBUFS;
1100 	}
1101 
1102 #if NVLAN > 0
1103 	/* setup h/w VLAN tagging */
1104 	if ((mtag = VLAN_OUTPUT_TAG(&sc->sc_ethercom, m0)) != NULL)
1105 		vtag = NFE_TX_VTAG | VLAN_TAG_VALUE(mtag);
1106 #endif
1107 	if ((sc->sc_flags & NFE_HW_CSUM) != 0) {
1108 		if (m0->m_pkthdr.csum_flags & M_CSUM_IPv4)
1109 			csumflags |= NFE_TX_IP_CSUM;
1110 		if (m0->m_pkthdr.csum_flags & (M_CSUM_TCPv4 | M_CSUM_UDPv4))
1111 			csumflags |= NFE_TX_TCP_UDP_CSUM;
1112 	}
1113 
1114 	for (i = 0; i < map->dm_nsegs; i++) {
1115 		data = &sc->txq.data[sc->txq.cur];
1116 
1117 		if (sc->sc_flags & NFE_40BIT_ADDR) {
1118 			desc64 = &sc->txq.desc64[sc->txq.cur];
1119 #if defined(__LP64__)
1120 			desc64->physaddr[0] =
1121 			    htole32(map->dm_segs[i].ds_addr >> 32);
1122 #endif
1123 			desc64->physaddr[1] =
1124 			    htole32(map->dm_segs[i].ds_addr & 0xffffffff);
1125 			desc64->length = htole16(map->dm_segs[i].ds_len - 1);
1126 			desc64->flags = htole16(flags);
1127 			desc64->vtag = 0;
1128 		} else {
1129 			desc32 = &sc->txq.desc32[sc->txq.cur];
1130 
1131 			desc32->physaddr = htole32(map->dm_segs[i].ds_addr);
1132 			desc32->length = htole16(map->dm_segs[i].ds_len - 1);
1133 			desc32->flags = htole16(flags);
1134 		}
1135 
1136 		/*
1137 		 * Setting of the valid bit in the first descriptor is
1138 		 * deferred until the whole chain is fully setup.
1139 		 */
1140 		flags |= NFE_TX_VALID;
1141 
1142 		sc->txq.queued++;
1143 		sc->txq.cur = NFE_TX_NEXTDESC(sc->txq.cur);
1144 	}
1145 
1146 	/* the whole mbuf chain has been setup */
1147 	if (sc->sc_flags & NFE_40BIT_ADDR) {
1148 		/* fix last descriptor */
1149 		flags |= NFE_TX_LASTFRAG_V2;
1150 		desc64->flags = htole16(flags);
1151 
1152 		/* Checksum flags and vtag belong to the first fragment only. */
1153 #if NVLAN > 0
1154 		sc->txq.desc64[first].vtag = htole32(vtag);
1155 #endif
1156 		sc->txq.desc64[first].flags |= htole16(csumflags);
1157 
1158 		/* finally, set the valid bit in the first descriptor */
1159 		sc->txq.desc64[first].flags |= htole16(NFE_TX_VALID);
1160 	} else {
1161 		/* fix last descriptor */
1162 		if (sc->sc_flags & NFE_JUMBO_SUP)
1163 			flags |= NFE_TX_LASTFRAG_V2;
1164 		else
1165 			flags |= NFE_TX_LASTFRAG_V1;
1166 		desc32->flags = htole16(flags);
1167 
1168 		/* Checksum flags belong to the first fragment only. */
1169 		sc->txq.desc32[first].flags |= htole16(csumflags);
1170 
1171 		/* finally, set the valid bit in the first descriptor */
1172 		sc->txq.desc32[first].flags |= htole16(NFE_TX_VALID);
1173 	}
1174 
1175 	data->m = m0;
1176 	data->active = map;
1177 
1178 	bus_dmamap_sync(sc->sc_dmat, map, 0, map->dm_mapsize,
1179 	    BUS_DMASYNC_PREWRITE);
1180 
1181 	return 0;
1182 }
1183 
1184 void
1185 nfe_start(struct ifnet *ifp)
1186 {
1187 	struct nfe_softc *sc = ifp->if_softc;
1188 	int old = sc->txq.queued;
1189 	struct mbuf *m0;
1190 
1191 	if ((ifp->if_flags & (IFF_RUNNING | IFF_OACTIVE)) != IFF_RUNNING)
1192 		return;
1193 
1194 	for (;;) {
1195 		IFQ_POLL(&ifp->if_snd, m0);
1196 		if (m0 == NULL)
1197 			break;
1198 
1199 		if (nfe_encap(sc, m0) != 0) {
1200 			ifp->if_flags |= IFF_OACTIVE;
1201 			break;
1202 		}
1203 
1204 		/* packet put in h/w queue, remove from s/w queue */
1205 		IFQ_DEQUEUE(&ifp->if_snd, m0);
1206 
1207 		bpf_mtap(ifp, m0);
1208 	}
1209 
1210 	if (sc->txq.queued != old) {
1211 		/* packets are queued */
1212 		if (sc->sc_flags & NFE_40BIT_ADDR)
1213 			nfe_txdesc64_rsync(sc, old, sc->txq.cur,
1214 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1215 		else
1216 			nfe_txdesc32_rsync(sc, old, sc->txq.cur,
1217 			    BUS_DMASYNC_PREREAD|BUS_DMASYNC_PREWRITE);
1218 		/* kick Tx */
1219 		NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_KICKTX | sc->rxtxctl);
1220 
1221 		/*
1222 		 * Set a timeout in case the chip goes out to lunch.
1223 		 */
1224 		ifp->if_timer = 5;
1225 	}
1226 }
1227 
1228 void
1229 nfe_watchdog(struct ifnet *ifp)
1230 {
1231 	struct nfe_softc *sc = ifp->if_softc;
1232 
1233 	aprint_error_dev(sc->sc_dev, "watchdog timeout\n");
1234 
1235 	ifp->if_flags &= ~IFF_RUNNING;
1236 	nfe_init(ifp);
1237 
1238 	ifp->if_oerrors++;
1239 }
1240 
1241 int
1242 nfe_init(struct ifnet *ifp)
1243 {
1244 	struct nfe_softc *sc = ifp->if_softc;
1245 	uint32_t tmp;
1246 	int rc = 0, s;
1247 
1248 	if (ifp->if_flags & IFF_RUNNING)
1249 		return 0;
1250 
1251 	nfe_stop(ifp, 0);
1252 
1253 	NFE_WRITE(sc, NFE_TX_UNK, 0);
1254 	NFE_WRITE(sc, NFE_STATUS, 0);
1255 
1256 	sc->rxtxctl = NFE_RXTX_BIT2;
1257 	if (sc->sc_flags & NFE_40BIT_ADDR)
1258 		sc->rxtxctl |= NFE_RXTX_V3MAGIC;
1259 	else if (sc->sc_flags & NFE_JUMBO_SUP)
1260 		sc->rxtxctl |= NFE_RXTX_V2MAGIC;
1261 	if (sc->sc_flags & NFE_HW_CSUM)
1262 		sc->rxtxctl |= NFE_RXTX_RXCSUM;
1263 #if NVLAN > 0
1264 	/*
1265 	 * Although the adapter is capable of stripping VLAN tags from received
1266 	 * frames (NFE_RXTX_VTAG_STRIP), we do not enable this functionality on
1267 	 * purpose.  This will be done in software by our network stack.
1268 	 */
1269 	if (sc->sc_flags & NFE_HW_VLAN)
1270 		sc->rxtxctl |= NFE_RXTX_VTAG_INSERT;
1271 #endif
1272 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_RESET | sc->rxtxctl);
1273 	DELAY(10);
1274 	NFE_WRITE(sc, NFE_RXTX_CTL, sc->rxtxctl);
1275 
1276 #if NVLAN
1277 	if (sc->sc_flags & NFE_HW_VLAN)
1278 		NFE_WRITE(sc, NFE_VTAG_CTL, NFE_VTAG_ENABLE);
1279 #endif
1280 
1281 	NFE_WRITE(sc, NFE_SETUP_R6, 0);
1282 
1283 	/* set MAC address */
1284 	nfe_set_macaddr(sc, sc->sc_enaddr);
1285 
1286 	/* tell MAC where rings are in memory */
1287 #ifdef __LP64__
1288 	NFE_WRITE(sc, NFE_RX_RING_ADDR_HI, sc->rxq.physaddr >> 32);
1289 #endif
1290 	NFE_WRITE(sc, NFE_RX_RING_ADDR_LO, sc->rxq.physaddr & 0xffffffff);
1291 #ifdef __LP64__
1292 	NFE_WRITE(sc, NFE_TX_RING_ADDR_HI, sc->txq.physaddr >> 32);
1293 #endif
1294 	NFE_WRITE(sc, NFE_TX_RING_ADDR_LO, sc->txq.physaddr & 0xffffffff);
1295 
1296 	NFE_WRITE(sc, NFE_RING_SIZE,
1297 	    (NFE_RX_RING_COUNT - 1) << 16 |
1298 	    (NFE_TX_RING_COUNT - 1));
1299 
1300 	NFE_WRITE(sc, NFE_RXBUFSZ, sc->rxq.bufsz);
1301 
1302 	/* force MAC to wakeup */
1303 	tmp = NFE_READ(sc, NFE_PWR_STATE);
1304 	NFE_WRITE(sc, NFE_PWR_STATE, tmp | NFE_PWR_WAKEUP);
1305 	DELAY(10);
1306 	tmp = NFE_READ(sc, NFE_PWR_STATE);
1307 	NFE_WRITE(sc, NFE_PWR_STATE, tmp | NFE_PWR_VALID);
1308 
1309 	s = splnet();
1310 	NFE_WRITE(sc, NFE_IRQ_MASK, 0);
1311 	nfe_intr(sc); /* XXX clear IRQ status registers */
1312 	NFE_WRITE(sc, NFE_IRQ_MASK, NFE_IRQ_WANTED);
1313 	splx(s);
1314 
1315 #if 1
1316 	/* configure interrupts coalescing/mitigation */
1317 	NFE_WRITE(sc, NFE_IMTIMER, NFE_IM_DEFAULT);
1318 #else
1319 	/* no interrupt mitigation: one interrupt per packet */
1320 	NFE_WRITE(sc, NFE_IMTIMER, 970);
1321 #endif
1322 
1323 	NFE_WRITE(sc, NFE_SETUP_R1, NFE_R1_MAGIC);
1324 	NFE_WRITE(sc, NFE_SETUP_R2, NFE_R2_MAGIC);
1325 	NFE_WRITE(sc, NFE_SETUP_R6, NFE_R6_MAGIC);
1326 
1327 	/* update MAC knowledge of PHY; generates a NFE_IRQ_LINK interrupt */
1328 	NFE_WRITE(sc, NFE_STATUS, sc->mii_phyaddr << 24 | NFE_STATUS_MAGIC);
1329 
1330 	NFE_WRITE(sc, NFE_SETUP_R4, NFE_R4_MAGIC);
1331 	NFE_WRITE(sc, NFE_WOL_CTL, NFE_WOL_ENABLE);
1332 
1333 	sc->rxtxctl &= ~NFE_RXTX_BIT2;
1334 	NFE_WRITE(sc, NFE_RXTX_CTL, sc->rxtxctl);
1335 	DELAY(10);
1336 	NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_BIT1 | sc->rxtxctl);
1337 
1338 	/* set Rx filter */
1339 	nfe_setmulti(sc);
1340 
1341 	if ((rc = ether_mediachange(ifp)) != 0)
1342 		goto out;
1343 
1344 	nfe_tick(sc);
1345 
1346 	/* enable Rx */
1347 	NFE_WRITE(sc, NFE_RX_CTL, NFE_RX_START);
1348 
1349 	/* enable Tx */
1350 	NFE_WRITE(sc, NFE_TX_CTL, NFE_TX_START);
1351 
1352 	NFE_WRITE(sc, NFE_PHY_STATUS, 0xf);
1353 
1354 	/* enable interrupts */
1355 	NFE_WRITE(sc, NFE_IRQ_MASK, NFE_IRQ_WANTED);
1356 
1357 	callout_schedule(&sc->sc_tick_ch, hz);
1358 
1359 	ifp->if_flags |= IFF_RUNNING;
1360 	ifp->if_flags &= ~IFF_OACTIVE;
1361 
1362 out:
1363 	return rc;
1364 }
1365 
1366 void
1367 nfe_stop(struct ifnet *ifp, int disable)
1368 {
1369 	struct nfe_softc *sc = ifp->if_softc;
1370 
1371 	callout_stop(&sc->sc_tick_ch);
1372 
1373 	ifp->if_timer = 0;
1374 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
1375 
1376 	mii_down(&sc->sc_mii);
1377 
1378 	/* abort Tx */
1379 	NFE_WRITE(sc, NFE_TX_CTL, 0);
1380 
1381 	/* disable Rx */
1382 	NFE_WRITE(sc, NFE_RX_CTL, 0);
1383 
1384 	/* disable interrupts */
1385 	NFE_WRITE(sc, NFE_IRQ_MASK, 0);
1386 
1387 	/* reset Tx and Rx rings */
1388 	nfe_reset_tx_ring(sc, &sc->txq);
1389 	nfe_reset_rx_ring(sc, &sc->rxq);
1390 }
1391 
1392 int
1393 nfe_alloc_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1394 {
1395 	struct nfe_desc32 *desc32;
1396 	struct nfe_desc64 *desc64;
1397 	struct nfe_rx_data *data;
1398 	struct nfe_jbuf *jbuf;
1399 	void **desc;
1400 	bus_addr_t physaddr;
1401 	int i, nsegs, error, descsize;
1402 
1403 	if (sc->sc_flags & NFE_40BIT_ADDR) {
1404 		desc = (void **)&ring->desc64;
1405 		descsize = sizeof (struct nfe_desc64);
1406 	} else {
1407 		desc = (void **)&ring->desc32;
1408 		descsize = sizeof (struct nfe_desc32);
1409 	}
1410 
1411 	ring->cur = ring->next = 0;
1412 	ring->bufsz = MCLBYTES;
1413 
1414 	error = bus_dmamap_create(sc->sc_dmat, NFE_RX_RING_COUNT * descsize, 1,
1415 	    NFE_RX_RING_COUNT * descsize, 0, BUS_DMA_NOWAIT, &ring->map);
1416 	if (error != 0) {
1417 		aprint_error_dev(sc->sc_dev,
1418 		    "could not create desc DMA map\n");
1419 		ring->map = NULL;
1420 		goto fail;
1421 	}
1422 
1423 	error = bus_dmamem_alloc(sc->sc_dmat, NFE_RX_RING_COUNT * descsize,
1424 	    PAGE_SIZE, 0, &ring->seg, 1, &nsegs, BUS_DMA_NOWAIT);
1425 	if (error != 0) {
1426 		aprint_error_dev(sc->sc_dev,
1427 		    "could not allocate DMA memory\n");
1428 		goto fail;
1429 	}
1430 
1431 	error = bus_dmamem_map(sc->sc_dmat, &ring->seg, nsegs,
1432 	    NFE_RX_RING_COUNT * descsize, (void **)desc, BUS_DMA_NOWAIT);
1433 	if (error != 0) {
1434 		aprint_error_dev(sc->sc_dev,
1435 		    "could not map desc DMA memory\n");
1436 		goto fail;
1437 	}
1438 
1439 	error = bus_dmamap_load(sc->sc_dmat, ring->map, *desc,
1440 	    NFE_RX_RING_COUNT * descsize, NULL, BUS_DMA_NOWAIT);
1441 	if (error != 0) {
1442 		aprint_error_dev(sc->sc_dev, "could not load desc DMA map\n");
1443 		goto fail;
1444 	}
1445 
1446 	memset(*desc, 0, NFE_RX_RING_COUNT * descsize);
1447 	ring->physaddr = ring->map->dm_segs[0].ds_addr;
1448 
1449 	if (sc->sc_flags & NFE_USE_JUMBO) {
1450 		ring->bufsz = NFE_JBYTES;
1451 		if ((error = nfe_jpool_alloc(sc)) != 0) {
1452 			aprint_error_dev(sc->sc_dev,
1453 			    "could not allocate jumbo frames\n");
1454 			goto fail;
1455 		}
1456 	}
1457 
1458 	/*
1459 	 * Pre-allocate Rx buffers and populate Rx ring.
1460 	 */
1461 	for (i = 0; i < NFE_RX_RING_COUNT; i++) {
1462 		data = &sc->rxq.data[i];
1463 
1464 		MGETHDR(data->m, M_DONTWAIT, MT_DATA);
1465 		if (data->m == NULL) {
1466 			aprint_error_dev(sc->sc_dev,
1467 			    "could not allocate rx mbuf\n");
1468 			error = ENOMEM;
1469 			goto fail;
1470 		}
1471 
1472 		if (sc->sc_flags & NFE_USE_JUMBO) {
1473 			if ((jbuf = nfe_jalloc(sc, i)) == NULL) {
1474 				aprint_error_dev(sc->sc_dev,
1475 				    "could not allocate jumbo buffer\n");
1476 				goto fail;
1477 			}
1478 			MEXTADD(data->m, jbuf->buf, NFE_JBYTES, 0, nfe_jfree,
1479 			    sc);
1480 
1481 			physaddr = jbuf->physaddr;
1482 		} else {
1483 			error = bus_dmamap_create(sc->sc_dmat, MCLBYTES, 1,
1484 			    MCLBYTES, 0, BUS_DMA_NOWAIT, &data->map);
1485 			if (error != 0) {
1486 				aprint_error_dev(sc->sc_dev,
1487 				    "could not create DMA map\n");
1488 				data->map = NULL;
1489 				goto fail;
1490 			}
1491 			MCLGET(data->m, M_DONTWAIT);
1492 			if (!(data->m->m_flags & M_EXT)) {
1493 				aprint_error_dev(sc->sc_dev,
1494 				    "could not allocate mbuf cluster\n");
1495 				error = ENOMEM;
1496 				goto fail;
1497 			}
1498 
1499 			error = bus_dmamap_load(sc->sc_dmat, data->map,
1500 			    mtod(data->m, void *), MCLBYTES, NULL,
1501 			    BUS_DMA_READ | BUS_DMA_NOWAIT);
1502 			if (error != 0) {
1503 				aprint_error_dev(sc->sc_dev,
1504 				    "could not load rx buf DMA map");
1505 				goto fail;
1506 			}
1507 			physaddr = data->map->dm_segs[0].ds_addr;
1508 		}
1509 
1510 		if (sc->sc_flags & NFE_40BIT_ADDR) {
1511 			desc64 = &sc->rxq.desc64[i];
1512 #if defined(__LP64__)
1513 			desc64->physaddr[0] = htole32(physaddr >> 32);
1514 #endif
1515 			desc64->physaddr[1] = htole32(physaddr & 0xffffffff);
1516 			desc64->length = htole16(sc->rxq.bufsz);
1517 			desc64->flags = htole16(NFE_RX_READY);
1518 		} else {
1519 			desc32 = &sc->rxq.desc32[i];
1520 			desc32->physaddr = htole32(physaddr);
1521 			desc32->length = htole16(sc->rxq.bufsz);
1522 			desc32->flags = htole16(NFE_RX_READY);
1523 		}
1524 	}
1525 
1526 	bus_dmamap_sync(sc->sc_dmat, ring->map, 0, ring->map->dm_mapsize,
1527 	    BUS_DMASYNC_PREWRITE);
1528 
1529 	return 0;
1530 
1531 fail:	nfe_free_rx_ring(sc, ring);
1532 	return error;
1533 }
1534 
1535 void
1536 nfe_reset_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1537 {
1538 	int i;
1539 
1540 	for (i = 0; i < NFE_RX_RING_COUNT; i++) {
1541 		if (sc->sc_flags & NFE_40BIT_ADDR) {
1542 			ring->desc64[i].length = htole16(ring->bufsz);
1543 			ring->desc64[i].flags = htole16(NFE_RX_READY);
1544 		} else {
1545 			ring->desc32[i].length = htole16(ring->bufsz);
1546 			ring->desc32[i].flags = htole16(NFE_RX_READY);
1547 		}
1548 	}
1549 
1550 	bus_dmamap_sync(sc->sc_dmat, ring->map, 0, ring->map->dm_mapsize,
1551 	    BUS_DMASYNC_PREWRITE);
1552 
1553 	ring->cur = ring->next = 0;
1554 }
1555 
1556 void
1557 nfe_free_rx_ring(struct nfe_softc *sc, struct nfe_rx_ring *ring)
1558 {
1559 	struct nfe_rx_data *data;
1560 	void *desc;
1561 	int i, descsize;
1562 
1563 	if (sc->sc_flags & NFE_40BIT_ADDR) {
1564 		desc = ring->desc64;
1565 		descsize = sizeof (struct nfe_desc64);
1566 	} else {
1567 		desc = ring->desc32;
1568 		descsize = sizeof (struct nfe_desc32);
1569 	}
1570 
1571 	if (desc != NULL) {
1572 		bus_dmamap_sync(sc->sc_dmat, ring->map, 0,
1573 		    ring->map->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1574 		bus_dmamap_unload(sc->sc_dmat, ring->map);
1575 		bus_dmamem_unmap(sc->sc_dmat, (void *)desc,
1576 		    NFE_RX_RING_COUNT * descsize);
1577 		bus_dmamem_free(sc->sc_dmat, &ring->seg, 1);
1578 	}
1579 
1580 	for (i = 0; i < NFE_RX_RING_COUNT; i++) {
1581 		data = &ring->data[i];
1582 
1583 		if (data->map != NULL) {
1584 			bus_dmamap_sync(sc->sc_dmat, data->map, 0,
1585 			    data->map->dm_mapsize, BUS_DMASYNC_POSTREAD);
1586 			bus_dmamap_unload(sc->sc_dmat, data->map);
1587 			bus_dmamap_destroy(sc->sc_dmat, data->map);
1588 		}
1589 		if (data->m != NULL)
1590 			m_freem(data->m);
1591 	}
1592 
1593 	nfe_jpool_free(sc);
1594 }
1595 
1596 struct nfe_jbuf *
1597 nfe_jalloc(struct nfe_softc *sc, int i)
1598 {
1599 	struct nfe_jbuf *jbuf;
1600 
1601 	mutex_enter(&sc->rxq.mtx);
1602 	jbuf = SLIST_FIRST(&sc->rxq.jfreelist);
1603 	if (jbuf != NULL)
1604 		SLIST_REMOVE_HEAD(&sc->rxq.jfreelist, jnext);
1605 	mutex_exit(&sc->rxq.mtx);
1606 	if (jbuf == NULL)
1607 		return NULL;
1608 	sc->rxq.jbufmap[i] =
1609 	    ((char *)jbuf->buf - (char *)sc->rxq.jpool) / NFE_JBYTES;
1610 	return jbuf;
1611 }
1612 
1613 /*
1614  * This is called automatically by the network stack when the mbuf is freed.
1615  * Caution must be taken that the NIC might be reset by the time the mbuf is
1616  * freed.
1617  */
1618 void
1619 nfe_jfree(struct mbuf *m, void *buf, size_t size, void *arg)
1620 {
1621 	struct nfe_softc *sc = arg;
1622 	struct nfe_jbuf *jbuf;
1623 	int i;
1624 
1625 	/* find the jbuf from the base pointer */
1626 	i = ((char *)buf - (char *)sc->rxq.jpool) / NFE_JBYTES;
1627 	if (i < 0 || i >= NFE_JPOOL_COUNT) {
1628 		aprint_error_dev(sc->sc_dev,
1629 		    "request to free a buffer (%p) not managed by us\n", buf);
1630 		return;
1631 	}
1632 	jbuf = &sc->rxq.jbuf[i];
1633 
1634 	/* ..and put it back in the free list */
1635 	mutex_enter(&sc->rxq.mtx);
1636 	SLIST_INSERT_HEAD(&sc->rxq.jfreelist, jbuf, jnext);
1637 	mutex_exit(&sc->rxq.mtx);
1638 
1639 	if (m != NULL)
1640 		pool_cache_put(mb_cache, m);
1641 }
1642 
1643 int
1644 nfe_jpool_alloc(struct nfe_softc *sc)
1645 {
1646 	struct nfe_rx_ring *ring = &sc->rxq;
1647 	struct nfe_jbuf *jbuf;
1648 	bus_addr_t physaddr;
1649 	char *buf;
1650 	int i, nsegs, error;
1651 
1652 	/*
1653 	 * Allocate a big chunk of DMA'able memory.
1654 	 */
1655 	error = bus_dmamap_create(sc->sc_dmat, NFE_JPOOL_SIZE, 1,
1656 	    NFE_JPOOL_SIZE, 0, BUS_DMA_NOWAIT, &ring->jmap);
1657 	if (error != 0) {
1658 		aprint_error_dev(sc->sc_dev,
1659 		    "could not create jumbo DMA map\n");
1660 		ring->jmap = NULL;
1661 		goto fail;
1662 	}
1663 
1664 	error = bus_dmamem_alloc(sc->sc_dmat, NFE_JPOOL_SIZE, PAGE_SIZE, 0,
1665 	    &ring->jseg, 1, &nsegs, BUS_DMA_NOWAIT);
1666 	if (error != 0) {
1667 		aprint_error_dev(sc->sc_dev,
1668 		    "could not allocate jumbo DMA memory\n");
1669 		goto fail;
1670 	}
1671 
1672 	error = bus_dmamem_map(sc->sc_dmat, &ring->jseg, nsegs, NFE_JPOOL_SIZE,
1673 	    &ring->jpool, BUS_DMA_NOWAIT);
1674 	if (error != 0) {
1675 		aprint_error_dev(sc->sc_dev,
1676 		    "could not map jumbo DMA memory\n");
1677 		goto fail;
1678 	}
1679 
1680 	error = bus_dmamap_load(sc->sc_dmat, ring->jmap, ring->jpool,
1681 	    NFE_JPOOL_SIZE, NULL, BUS_DMA_READ | BUS_DMA_NOWAIT);
1682 	if (error != 0) {
1683 		aprint_error_dev(sc->sc_dev,
1684 		    "could not load jumbo DMA map\n");
1685 		goto fail;
1686 	}
1687 
1688 	/* ..and split it into 9KB chunks */
1689 	SLIST_INIT(&ring->jfreelist);
1690 
1691 	buf = ring->jpool;
1692 	physaddr = ring->jmap->dm_segs[0].ds_addr;
1693 	for (i = 0; i < NFE_JPOOL_COUNT; i++) {
1694 		jbuf = &ring->jbuf[i];
1695 
1696 		jbuf->buf = buf;
1697 		jbuf->physaddr = physaddr;
1698 
1699 		SLIST_INSERT_HEAD(&ring->jfreelist, jbuf, jnext);
1700 
1701 		buf += NFE_JBYTES;
1702 		physaddr += NFE_JBYTES;
1703 	}
1704 
1705 	return 0;
1706 
1707 fail:	nfe_jpool_free(sc);
1708 	return error;
1709 }
1710 
1711 void
1712 nfe_jpool_free(struct nfe_softc *sc)
1713 {
1714 	struct nfe_rx_ring *ring = &sc->rxq;
1715 
1716 	if (ring->jmap != NULL) {
1717 		bus_dmamap_sync(sc->sc_dmat, ring->jmap, 0,
1718 		    ring->jmap->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1719 		bus_dmamap_unload(sc->sc_dmat, ring->jmap);
1720 		bus_dmamap_destroy(sc->sc_dmat, ring->jmap);
1721 		ring->jmap = NULL;
1722 	}
1723 	if (ring->jpool != NULL) {
1724 		bus_dmamem_unmap(sc->sc_dmat, ring->jpool, NFE_JPOOL_SIZE);
1725 		bus_dmamem_free(sc->sc_dmat, &ring->jseg, 1);
1726 		ring->jpool = NULL;
1727 	}
1728 }
1729 
1730 int
1731 nfe_alloc_tx_ring(struct nfe_softc *sc, struct nfe_tx_ring *ring)
1732 {
1733 	int i, nsegs, error;
1734 	void **desc;
1735 	int descsize;
1736 
1737 	if (sc->sc_flags & NFE_40BIT_ADDR) {
1738 		desc = (void **)&ring->desc64;
1739 		descsize = sizeof (struct nfe_desc64);
1740 	} else {
1741 		desc = (void **)&ring->desc32;
1742 		descsize = sizeof (struct nfe_desc32);
1743 	}
1744 
1745 	ring->queued = 0;
1746 	ring->cur = ring->next = 0;
1747 
1748 	error = bus_dmamap_create(sc->sc_dmat, NFE_TX_RING_COUNT * descsize, 1,
1749 	    NFE_TX_RING_COUNT * descsize, 0, BUS_DMA_NOWAIT, &ring->map);
1750 
1751 	if (error != 0) {
1752 		aprint_error_dev(sc->sc_dev,
1753 		    "could not create desc DMA map\n");
1754 		ring->map = NULL;
1755 		goto fail;
1756 	}
1757 
1758 	error = bus_dmamem_alloc(sc->sc_dmat, NFE_TX_RING_COUNT * descsize,
1759 	    PAGE_SIZE, 0, &ring->seg, 1, &nsegs, BUS_DMA_NOWAIT);
1760 	if (error != 0) {
1761 		aprint_error_dev(sc->sc_dev,
1762 		    "could not allocate DMA memory\n");
1763 		goto fail;
1764 	}
1765 
1766 	error = bus_dmamem_map(sc->sc_dmat, &ring->seg, nsegs,
1767 	    NFE_TX_RING_COUNT * descsize, (void **)desc, BUS_DMA_NOWAIT);
1768 	if (error != 0) {
1769 		aprint_error_dev(sc->sc_dev,
1770 		    "could not map desc DMA memory\n");
1771 		goto fail;
1772 	}
1773 
1774 	error = bus_dmamap_load(sc->sc_dmat, ring->map, *desc,
1775 	    NFE_TX_RING_COUNT * descsize, NULL, BUS_DMA_NOWAIT);
1776 	if (error != 0) {
1777 		aprint_error_dev(sc->sc_dev, "could not load desc DMA map\n");
1778 		goto fail;
1779 	}
1780 
1781 	memset(*desc, 0, NFE_TX_RING_COUNT * descsize);
1782 	ring->physaddr = ring->map->dm_segs[0].ds_addr;
1783 
1784 	for (i = 0; i < NFE_TX_RING_COUNT; i++) {
1785 		error = bus_dmamap_create(sc->sc_dmat, NFE_JBYTES,
1786 		    NFE_MAX_SCATTER, NFE_JBYTES, 0, BUS_DMA_NOWAIT,
1787 		    &ring->data[i].map);
1788 		if (error != 0) {
1789 			aprint_error_dev(sc->sc_dev,
1790 			    "could not create DMA map\n");
1791 			ring->data[i].map = NULL;
1792 			goto fail;
1793 		}
1794 	}
1795 
1796 	return 0;
1797 
1798 fail:	nfe_free_tx_ring(sc, ring);
1799 	return error;
1800 }
1801 
1802 void
1803 nfe_reset_tx_ring(struct nfe_softc *sc, struct nfe_tx_ring *ring)
1804 {
1805 	struct nfe_tx_data *data;
1806 	int i;
1807 
1808 	for (i = 0; i < NFE_TX_RING_COUNT; i++) {
1809 		if (sc->sc_flags & NFE_40BIT_ADDR)
1810 			ring->desc64[i].flags = 0;
1811 		else
1812 			ring->desc32[i].flags = 0;
1813 
1814 		data = &ring->data[i];
1815 
1816 		if (data->m != NULL) {
1817 			bus_dmamap_sync(sc->sc_dmat, data->active, 0,
1818 			    data->active->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1819 			bus_dmamap_unload(sc->sc_dmat, data->active);
1820 			m_freem(data->m);
1821 			data->m = NULL;
1822 		}
1823 	}
1824 
1825 	bus_dmamap_sync(sc->sc_dmat, ring->map, 0, ring->map->dm_mapsize,
1826 	    BUS_DMASYNC_PREWRITE);
1827 
1828 	ring->queued = 0;
1829 	ring->cur = ring->next = 0;
1830 }
1831 
1832 void
1833 nfe_free_tx_ring(struct nfe_softc *sc, struct nfe_tx_ring *ring)
1834 {
1835 	struct nfe_tx_data *data;
1836 	void *desc;
1837 	int i, descsize;
1838 
1839 	if (sc->sc_flags & NFE_40BIT_ADDR) {
1840 		desc = ring->desc64;
1841 		descsize = sizeof (struct nfe_desc64);
1842 	} else {
1843 		desc = ring->desc32;
1844 		descsize = sizeof (struct nfe_desc32);
1845 	}
1846 
1847 	if (desc != NULL) {
1848 		bus_dmamap_sync(sc->sc_dmat, ring->map, 0,
1849 		    ring->map->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1850 		bus_dmamap_unload(sc->sc_dmat, ring->map);
1851 		bus_dmamem_unmap(sc->sc_dmat, (void *)desc,
1852 		    NFE_TX_RING_COUNT * descsize);
1853 		bus_dmamem_free(sc->sc_dmat, &ring->seg, 1);
1854 	}
1855 
1856 	for (i = 0; i < NFE_TX_RING_COUNT; i++) {
1857 		data = &ring->data[i];
1858 
1859 		if (data->m != NULL) {
1860 			bus_dmamap_sync(sc->sc_dmat, data->active, 0,
1861 			    data->active->dm_mapsize, BUS_DMASYNC_POSTWRITE);
1862 			bus_dmamap_unload(sc->sc_dmat, data->active);
1863 			m_freem(data->m);
1864 		}
1865 	}
1866 
1867 	/* ..and now actually destroy the DMA mappings */
1868 	for (i = 0; i < NFE_TX_RING_COUNT; i++) {
1869 		data = &ring->data[i];
1870 		if (data->map == NULL)
1871 			continue;
1872 		bus_dmamap_destroy(sc->sc_dmat, data->map);
1873 	}
1874 }
1875 
1876 void
1877 nfe_setmulti(struct nfe_softc *sc)
1878 {
1879 	struct ethercom *ec = &sc->sc_ethercom;
1880 	struct ifnet *ifp = &ec->ec_if;
1881 	struct ether_multi *enm;
1882 	struct ether_multistep step;
1883 	uint8_t addr[ETHER_ADDR_LEN], mask[ETHER_ADDR_LEN];
1884 	uint32_t filter = NFE_RXFILTER_MAGIC;
1885 	int i;
1886 
1887 	if ((ifp->if_flags & (IFF_ALLMULTI | IFF_PROMISC)) != 0) {
1888 		memset(addr, 0, ETHER_ADDR_LEN);
1889 		memset(mask, 0, ETHER_ADDR_LEN);
1890 		goto done;
1891 	}
1892 
1893 	memcpy(addr, etherbroadcastaddr, ETHER_ADDR_LEN);
1894 	memcpy(mask, etherbroadcastaddr, ETHER_ADDR_LEN);
1895 
1896 	ETHER_FIRST_MULTI(step, ec, enm);
1897 	while (enm != NULL) {
1898 		if (memcmp(enm->enm_addrlo, enm->enm_addrhi, ETHER_ADDR_LEN)) {
1899 			ifp->if_flags |= IFF_ALLMULTI;
1900 			memset(addr, 0, ETHER_ADDR_LEN);
1901 			memset(mask, 0, ETHER_ADDR_LEN);
1902 			goto done;
1903 		}
1904 		for (i = 0; i < ETHER_ADDR_LEN; i++) {
1905 			addr[i] &=  enm->enm_addrlo[i];
1906 			mask[i] &= ~enm->enm_addrlo[i];
1907 		}
1908 		ETHER_NEXT_MULTI(step, enm);
1909 	}
1910 	for (i = 0; i < ETHER_ADDR_LEN; i++)
1911 		mask[i] |= addr[i];
1912 
1913 done:
1914 	addr[0] |= 0x01;	/* make sure multicast bit is set */
1915 
1916 	NFE_WRITE(sc, NFE_MULTIADDR_HI,
1917 	    addr[3] << 24 | addr[2] << 16 | addr[1] << 8 | addr[0]);
1918 	NFE_WRITE(sc, NFE_MULTIADDR_LO,
1919 	    addr[5] <<  8 | addr[4]);
1920 	NFE_WRITE(sc, NFE_MULTIMASK_HI,
1921 	    mask[3] << 24 | mask[2] << 16 | mask[1] << 8 | mask[0]);
1922 	NFE_WRITE(sc, NFE_MULTIMASK_LO,
1923 	    mask[5] <<  8 | mask[4]);
1924 
1925 	filter |= (ifp->if_flags & IFF_PROMISC) ? NFE_PROMISC : NFE_U2M;
1926 	NFE_WRITE(sc, NFE_RXFILTER, filter);
1927 }
1928 
1929 void
1930 nfe_get_macaddr(struct nfe_softc *sc, uint8_t *addr)
1931 {
1932 	uint32_t tmp;
1933 
1934 	if ((sc->sc_flags & NFE_CORRECT_MACADDR) != 0) {
1935 		tmp = NFE_READ(sc, NFE_MACADDR_HI);
1936 		addr[0] = (tmp & 0xff);
1937 		addr[1] = (tmp >>  8) & 0xff;
1938 		addr[2] = (tmp >> 16) & 0xff;
1939 		addr[3] = (tmp >> 24) & 0xff;
1940 
1941 		tmp = NFE_READ(sc, NFE_MACADDR_LO);
1942 		addr[4] = (tmp & 0xff);
1943 		addr[5] = (tmp >> 8) & 0xff;
1944 
1945 	} else {
1946 		tmp = NFE_READ(sc, NFE_MACADDR_LO);
1947 		addr[0] = (tmp >> 8) & 0xff;
1948 		addr[1] = (tmp & 0xff);
1949 
1950 		tmp = NFE_READ(sc, NFE_MACADDR_HI);
1951 		addr[2] = (tmp >> 24) & 0xff;
1952 		addr[3] = (tmp >> 16) & 0xff;
1953 		addr[4] = (tmp >>  8) & 0xff;
1954 		addr[5] = (tmp & 0xff);
1955 	}
1956 }
1957 
1958 void
1959 nfe_set_macaddr(struct nfe_softc *sc, const uint8_t *addr)
1960 {
1961 	NFE_WRITE(sc, NFE_MACADDR_LO,
1962 	    addr[5] <<  8 | addr[4]);
1963 	NFE_WRITE(sc, NFE_MACADDR_HI,
1964 	    addr[3] << 24 | addr[2] << 16 | addr[1] << 8 | addr[0]);
1965 }
1966 
1967 void
1968 nfe_tick(void *arg)
1969 {
1970 	struct nfe_softc *sc = arg;
1971 	int s;
1972 
1973 	s = splnet();
1974 	mii_tick(&sc->sc_mii);
1975 	splx(s);
1976 
1977 	callout_schedule(&sc->sc_tick_ch, hz);
1978 }
1979 
1980 void
1981 nfe_poweron(device_t self)
1982 {
1983 	struct nfe_softc *sc = device_private(self);
1984 
1985 	if ((sc->sc_flags & NFE_PWR_MGMT) != 0) {
1986 		NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_RESET | NFE_RXTX_BIT2);
1987 		NFE_WRITE(sc, NFE_MAC_RESET, NFE_MAC_RESET_MAGIC);
1988 		DELAY(100);
1989 		NFE_WRITE(sc, NFE_MAC_RESET, 0);
1990 		DELAY(100);
1991 		NFE_WRITE(sc, NFE_RXTX_CTL, NFE_RXTX_BIT2);
1992 		NFE_WRITE(sc, NFE_PWR2_CTL,
1993 		    NFE_READ(sc, NFE_PWR2_CTL) & ~NFE_PWR2_WAKEUP_MASK);
1994 	}
1995 }
1996 
1997 bool
1998 nfe_resume(device_t dv, const pmf_qual_t *qual)
1999 {
2000 	nfe_poweron(dv);
2001 
2002 	return true;
2003 }
2004