xref: /dflybsd-src/sys/dev/netif/bnx/if_bnx.c (revision 636eca08bbf84bbb648c68d757be5ec1a2b86a72)
1 /*
2  * Copyright (c) 2001 Wind River Systems
3  * Copyright (c) 1997, 1998, 1999, 2001
4  *	Bill Paul <wpaul@windriver.com>.  All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  * 3. All advertising materials mentioning features or use of this software
15  *    must display the following acknowledgement:
16  *	This product includes software developed by Bill Paul.
17  * 4. Neither the name of the author nor the names of any co-contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
25  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
28  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
29  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
30  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
31  * THE POSSIBILITY OF SUCH DAMAGE.
32  *
33  * $FreeBSD: src/sys/dev/bge/if_bge.c,v 1.3.2.39 2005/07/03 03:41:18 silby Exp $
34  */
35 
36 #include "opt_bnx.h"
37 #include "opt_ifpoll.h"
38 
39 #include <sys/param.h>
40 #include <sys/bus.h>
41 #include <sys/endian.h>
42 #include <sys/kernel.h>
43 #include <sys/interrupt.h>
44 #include <sys/mbuf.h>
45 #include <sys/malloc.h>
46 #include <sys/queue.h>
47 #include <sys/rman.h>
48 #include <sys/serialize.h>
49 #include <sys/socket.h>
50 #include <sys/sockio.h>
51 #include <sys/sysctl.h>
52 
53 #include <netinet/ip.h>
54 #include <netinet/tcp.h>
55 
56 #include <net/bpf.h>
57 #include <net/ethernet.h>
58 #include <net/if.h>
59 #include <net/if_arp.h>
60 #include <net/if_dl.h>
61 #include <net/if_media.h>
62 #include <net/if_poll.h>
63 #include <net/if_types.h>
64 #include <net/ifq_var.h>
65 #include <net/vlan/if_vlan_var.h>
66 #include <net/vlan/if_vlan_ether.h>
67 
68 #include <dev/netif/mii_layer/mii.h>
69 #include <dev/netif/mii_layer/miivar.h>
70 #include <dev/netif/mii_layer/brgphyreg.h>
71 
72 #include <bus/pci/pcidevs.h>
73 #include <bus/pci/pcireg.h>
74 #include <bus/pci/pcivar.h>
75 
76 #include <dev/netif/bge/if_bgereg.h>
77 #include <dev/netif/bnx/if_bnxvar.h>
78 
79 /* "device miibus" required.  See GENERIC if you get errors here. */
80 #include "miibus_if.h"
81 
82 #define BNX_CSUM_FEATURES	(CSUM_IP | CSUM_TCP | CSUM_UDP)
83 
84 #define BNX_INTR_CKINTVL	((10 * hz) / 1000)	/* 10ms */
85 
86 static const struct bnx_type {
87 	uint16_t		bnx_vid;
88 	uint16_t		bnx_did;
89 	char			*bnx_name;
90 } bnx_devs[] = {
91 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5717,
92 		"Broadcom BCM5717 Gigabit Ethernet" },
93 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5718,
94 		"Broadcom BCM5718 Gigabit Ethernet" },
95 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5719,
96 		"Broadcom BCM5719 Gigabit Ethernet" },
97 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5720_ALT,
98 		"Broadcom BCM5720 Gigabit Ethernet" },
99 
100 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57761,
101 		"Broadcom BCM57761 Gigabit Ethernet" },
102 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57762,
103 		"Broadcom BCM57762 Gigabit Ethernet" },
104 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57765,
105 		"Broadcom BCM57765 Gigabit Ethernet" },
106 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57766,
107 		"Broadcom BCM57766 Gigabit Ethernet" },
108 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57781,
109 		"Broadcom BCM57781 Gigabit Ethernet" },
110 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57782,
111 		"Broadcom BCM57782 Gigabit Ethernet" },
112 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57785,
113 		"Broadcom BCM57785 Gigabit Ethernet" },
114 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57786,
115 		"Broadcom BCM57786 Gigabit Ethernet" },
116 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57791,
117 		"Broadcom BCM57791 Fast Ethernet" },
118 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57795,
119 		"Broadcom BCM57795 Fast Ethernet" },
120 
121 	{ 0, 0, NULL }
122 };
123 
124 #define BNX_IS_JUMBO_CAPABLE(sc)	((sc)->bnx_flags & BNX_FLAG_JUMBO)
125 #define BNX_IS_5717_PLUS(sc)		((sc)->bnx_flags & BNX_FLAG_5717_PLUS)
126 #define BNX_IS_57765_PLUS(sc)		((sc)->bnx_flags & BNX_FLAG_57765_PLUS)
127 #define BNX_IS_57765_FAMILY(sc)	 \
128 	((sc)->bnx_flags & BNX_FLAG_57765_FAMILY)
129 
130 typedef int	(*bnx_eaddr_fcn_t)(struct bnx_softc *, uint8_t[]);
131 
132 static int	bnx_probe(device_t);
133 static int	bnx_attach(device_t);
134 static int	bnx_detach(device_t);
135 static void	bnx_shutdown(device_t);
136 static int	bnx_suspend(device_t);
137 static int	bnx_resume(device_t);
138 static int	bnx_miibus_readreg(device_t, int, int);
139 static int	bnx_miibus_writereg(device_t, int, int, int);
140 static void	bnx_miibus_statchg(device_t);
141 
142 #ifdef IFPOLL_ENABLE
143 static void	bnx_npoll(struct ifnet *, struct ifpoll_info *);
144 static void	bnx_npoll_compat(struct ifnet *, void *, int);
145 #endif
146 static void	bnx_intr_legacy(void *);
147 static void	bnx_msi(void *);
148 static void	bnx_msi_oneshot(void *);
149 static void	bnx_intr(struct bnx_softc *);
150 static void	bnx_enable_intr(struct bnx_softc *);
151 static void	bnx_disable_intr(struct bnx_softc *);
152 static void	bnx_txeof(struct bnx_softc *, uint16_t);
153 static void	bnx_rxeof(struct bnx_softc *, uint16_t, int);
154 
155 static void	bnx_start(struct ifnet *, struct ifaltq_subque *);
156 static int	bnx_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
157 static void	bnx_init(void *);
158 static void	bnx_stop(struct bnx_softc *);
159 static void	bnx_watchdog(struct ifnet *);
160 static int	bnx_ifmedia_upd(struct ifnet *);
161 static void	bnx_ifmedia_sts(struct ifnet *, struct ifmediareq *);
162 static void	bnx_tick(void *);
163 
164 static int	bnx_alloc_jumbo_mem(struct bnx_softc *);
165 static void	bnx_free_jumbo_mem(struct bnx_softc *);
166 static struct bnx_jslot
167 		*bnx_jalloc(struct bnx_softc *);
168 static void	bnx_jfree(void *);
169 static void	bnx_jref(void *);
170 static int	bnx_newbuf_std(struct bnx_softc *, int, int);
171 static int	bnx_newbuf_jumbo(struct bnx_softc *, int, int);
172 static void	bnx_setup_rxdesc_std(struct bnx_softc *, int);
173 static void	bnx_setup_rxdesc_jumbo(struct bnx_softc *, int);
174 static int	bnx_init_rx_ring_std(struct bnx_softc *);
175 static void	bnx_free_rx_ring_std(struct bnx_softc *);
176 static int	bnx_init_rx_ring_jumbo(struct bnx_softc *);
177 static void	bnx_free_rx_ring_jumbo(struct bnx_softc *);
178 static void	bnx_free_tx_ring(struct bnx_softc *);
179 static int	bnx_init_tx_ring(struct bnx_softc *);
180 static int	bnx_dma_alloc(struct bnx_softc *);
181 static void	bnx_dma_free(struct bnx_softc *);
182 static int	bnx_dma_block_alloc(struct bnx_softc *, bus_size_t,
183 		    bus_dma_tag_t *, bus_dmamap_t *, void **, bus_addr_t *);
184 static void	bnx_dma_block_free(bus_dma_tag_t, bus_dmamap_t, void *);
185 static struct mbuf *
186 		bnx_defrag_shortdma(struct mbuf *);
187 static int	bnx_encap(struct bnx_softc *, struct mbuf **,
188 			uint32_t *, int *);
189 static int	bnx_setup_tso(struct bnx_softc *, struct mbuf **,
190 		    uint16_t *, uint16_t *);
191 
192 static void	bnx_reset(struct bnx_softc *);
193 static int	bnx_chipinit(struct bnx_softc *);
194 static int	bnx_blockinit(struct bnx_softc *);
195 static void	bnx_stop_block(struct bnx_softc *, bus_size_t, uint32_t);
196 static void	bnx_enable_msi(struct bnx_softc *sc);
197 static void	bnx_setmulti(struct bnx_softc *);
198 static void	bnx_setpromisc(struct bnx_softc *);
199 static void	bnx_stats_update_regs(struct bnx_softc *);
200 static uint32_t	bnx_dma_swap_options(struct bnx_softc *);
201 
202 static uint32_t	bnx_readmem_ind(struct bnx_softc *, uint32_t);
203 static void	bnx_writemem_ind(struct bnx_softc *, uint32_t, uint32_t);
204 #ifdef notdef
205 static uint32_t	bnx_readreg_ind(struct bnx_softc *, uint32_t);
206 #endif
207 static void	bnx_writereg_ind(struct bnx_softc *, uint32_t, uint32_t);
208 static void	bnx_writemem_direct(struct bnx_softc *, uint32_t, uint32_t);
209 static void	bnx_writembx(struct bnx_softc *, int, int);
210 static uint8_t	bnx_nvram_getbyte(struct bnx_softc *, int, uint8_t *);
211 static int	bnx_read_nvram(struct bnx_softc *, caddr_t, int, int);
212 static uint8_t	bnx_eeprom_getbyte(struct bnx_softc *, uint32_t, uint8_t *);
213 static int	bnx_read_eeprom(struct bnx_softc *, caddr_t, uint32_t, size_t);
214 
215 static void	bnx_tbi_link_upd(struct bnx_softc *, uint32_t);
216 static void	bnx_copper_link_upd(struct bnx_softc *, uint32_t);
217 static void	bnx_autopoll_link_upd(struct bnx_softc *, uint32_t);
218 static void	bnx_link_poll(struct bnx_softc *);
219 
220 static int	bnx_get_eaddr_mem(struct bnx_softc *, uint8_t[]);
221 static int	bnx_get_eaddr_nvram(struct bnx_softc *, uint8_t[]);
222 static int	bnx_get_eaddr_eeprom(struct bnx_softc *, uint8_t[]);
223 static int	bnx_get_eaddr(struct bnx_softc *, uint8_t[]);
224 
225 static void	bnx_coal_change(struct bnx_softc *);
226 static int	bnx_sysctl_rx_coal_ticks(SYSCTL_HANDLER_ARGS);
227 static int	bnx_sysctl_tx_coal_ticks(SYSCTL_HANDLER_ARGS);
228 static int	bnx_sysctl_rx_coal_bds(SYSCTL_HANDLER_ARGS);
229 static int	bnx_sysctl_tx_coal_bds(SYSCTL_HANDLER_ARGS);
230 static int	bnx_sysctl_rx_coal_bds_int(SYSCTL_HANDLER_ARGS);
231 static int	bnx_sysctl_tx_coal_bds_int(SYSCTL_HANDLER_ARGS);
232 static int	bnx_sysctl_coal_chg(SYSCTL_HANDLER_ARGS, uint32_t *,
233 		    int, int, uint32_t);
234 
235 static int	bnx_msi_enable = 1;
236 TUNABLE_INT("hw.bnx.msi.enable", &bnx_msi_enable);
237 
238 static device_method_t bnx_methods[] = {
239 	/* Device interface */
240 	DEVMETHOD(device_probe,		bnx_probe),
241 	DEVMETHOD(device_attach,	bnx_attach),
242 	DEVMETHOD(device_detach,	bnx_detach),
243 	DEVMETHOD(device_shutdown,	bnx_shutdown),
244 	DEVMETHOD(device_suspend,	bnx_suspend),
245 	DEVMETHOD(device_resume,	bnx_resume),
246 
247 	/* bus interface */
248 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
249 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
250 
251 	/* MII interface */
252 	DEVMETHOD(miibus_readreg,	bnx_miibus_readreg),
253 	DEVMETHOD(miibus_writereg,	bnx_miibus_writereg),
254 	DEVMETHOD(miibus_statchg,	bnx_miibus_statchg),
255 
256 	{ 0, 0 }
257 };
258 
259 static DEFINE_CLASS_0(bnx, bnx_driver, bnx_methods, sizeof(struct bnx_softc));
260 static devclass_t bnx_devclass;
261 
262 DECLARE_DUMMY_MODULE(if_bnx);
263 DRIVER_MODULE(if_bnx, pci, bnx_driver, bnx_devclass, NULL, NULL);
264 DRIVER_MODULE(miibus, bnx, miibus_driver, miibus_devclass, NULL, NULL);
265 
266 static uint32_t
267 bnx_readmem_ind(struct bnx_softc *sc, uint32_t off)
268 {
269 	device_t dev = sc->bnx_dev;
270 	uint32_t val;
271 
272 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5906 &&
273 	    off >= BGE_STATS_BLOCK && off < BGE_SEND_RING_1_TO_4)
274 		return 0;
275 
276 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, off, 4);
277 	val = pci_read_config(dev, BGE_PCI_MEMWIN_DATA, 4);
278 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, 0, 4);
279 	return (val);
280 }
281 
282 static void
283 bnx_writemem_ind(struct bnx_softc *sc, uint32_t off, uint32_t val)
284 {
285 	device_t dev = sc->bnx_dev;
286 
287 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5906 &&
288 	    off >= BGE_STATS_BLOCK && off < BGE_SEND_RING_1_TO_4)
289 		return;
290 
291 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, off, 4);
292 	pci_write_config(dev, BGE_PCI_MEMWIN_DATA, val, 4);
293 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, 0, 4);
294 }
295 
296 #ifdef notdef
297 static uint32_t
298 bnx_readreg_ind(struct bnx_softc *sc, uin32_t off)
299 {
300 	device_t dev = sc->bnx_dev;
301 
302 	pci_write_config(dev, BGE_PCI_REG_BASEADDR, off, 4);
303 	return(pci_read_config(dev, BGE_PCI_REG_DATA, 4));
304 }
305 #endif
306 
307 static void
308 bnx_writereg_ind(struct bnx_softc *sc, uint32_t off, uint32_t val)
309 {
310 	device_t dev = sc->bnx_dev;
311 
312 	pci_write_config(dev, BGE_PCI_REG_BASEADDR, off, 4);
313 	pci_write_config(dev, BGE_PCI_REG_DATA, val, 4);
314 }
315 
316 static void
317 bnx_writemem_direct(struct bnx_softc *sc, uint32_t off, uint32_t val)
318 {
319 	CSR_WRITE_4(sc, off, val);
320 }
321 
322 static void
323 bnx_writembx(struct bnx_softc *sc, int off, int val)
324 {
325 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5906)
326 		off += BGE_LPMBX_IRQ0_HI - BGE_MBX_IRQ0_HI;
327 
328 	CSR_WRITE_4(sc, off, val);
329 }
330 
331 static uint8_t
332 bnx_nvram_getbyte(struct bnx_softc *sc, int addr, uint8_t *dest)
333 {
334 	uint32_t access, byte = 0;
335 	int i;
336 
337 	/* Lock. */
338 	CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_SET1);
339 	for (i = 0; i < 8000; i++) {
340 		if (CSR_READ_4(sc, BGE_NVRAM_SWARB) & BGE_NVRAMSWARB_GNT1)
341 			break;
342 		DELAY(20);
343 	}
344 	if (i == 8000)
345 		return (1);
346 
347 	/* Enable access. */
348 	access = CSR_READ_4(sc, BGE_NVRAM_ACCESS);
349 	CSR_WRITE_4(sc, BGE_NVRAM_ACCESS, access | BGE_NVRAMACC_ENABLE);
350 
351 	CSR_WRITE_4(sc, BGE_NVRAM_ADDR, addr & 0xfffffffc);
352 	CSR_WRITE_4(sc, BGE_NVRAM_CMD, BGE_NVRAM_READCMD);
353 	for (i = 0; i < BNX_TIMEOUT * 10; i++) {
354 		DELAY(10);
355 		if (CSR_READ_4(sc, BGE_NVRAM_CMD) & BGE_NVRAMCMD_DONE) {
356 			DELAY(10);
357 			break;
358 		}
359 	}
360 
361 	if (i == BNX_TIMEOUT * 10) {
362 		if_printf(&sc->arpcom.ac_if, "nvram read timed out\n");
363 		return (1);
364 	}
365 
366 	/* Get result. */
367 	byte = CSR_READ_4(sc, BGE_NVRAM_RDDATA);
368 
369 	*dest = (bswap32(byte) >> ((addr % 4) * 8)) & 0xFF;
370 
371 	/* Disable access. */
372 	CSR_WRITE_4(sc, BGE_NVRAM_ACCESS, access);
373 
374 	/* Unlock. */
375 	CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_CLR1);
376 	CSR_READ_4(sc, BGE_NVRAM_SWARB);
377 
378 	return (0);
379 }
380 
381 /*
382  * Read a sequence of bytes from NVRAM.
383  */
384 static int
385 bnx_read_nvram(struct bnx_softc *sc, caddr_t dest, int off, int cnt)
386 {
387 	int err = 0, i;
388 	uint8_t byte = 0;
389 
390 	if (sc->bnx_asicrev != BGE_ASICREV_BCM5906)
391 		return (1);
392 
393 	for (i = 0; i < cnt; i++) {
394 		err = bnx_nvram_getbyte(sc, off + i, &byte);
395 		if (err)
396 			break;
397 		*(dest + i) = byte;
398 	}
399 
400 	return (err ? 1 : 0);
401 }
402 
403 /*
404  * Read a byte of data stored in the EEPROM at address 'addr.' The
405  * BCM570x supports both the traditional bitbang interface and an
406  * auto access interface for reading the EEPROM. We use the auto
407  * access method.
408  */
409 static uint8_t
410 bnx_eeprom_getbyte(struct bnx_softc *sc, uint32_t addr, uint8_t *dest)
411 {
412 	int i;
413 	uint32_t byte = 0;
414 
415 	/*
416 	 * Enable use of auto EEPROM access so we can avoid
417 	 * having to use the bitbang method.
418 	 */
419 	BNX_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_AUTO_EEPROM);
420 
421 	/* Reset the EEPROM, load the clock period. */
422 	CSR_WRITE_4(sc, BGE_EE_ADDR,
423 	    BGE_EEADDR_RESET|BGE_EEHALFCLK(BGE_HALFCLK_384SCL));
424 	DELAY(20);
425 
426 	/* Issue the read EEPROM command. */
427 	CSR_WRITE_4(sc, BGE_EE_ADDR, BGE_EE_READCMD | addr);
428 
429 	/* Wait for completion */
430 	for(i = 0; i < BNX_TIMEOUT * 10; i++) {
431 		DELAY(10);
432 		if (CSR_READ_4(sc, BGE_EE_ADDR) & BGE_EEADDR_DONE)
433 			break;
434 	}
435 
436 	if (i == BNX_TIMEOUT) {
437 		if_printf(&sc->arpcom.ac_if, "eeprom read timed out\n");
438 		return(1);
439 	}
440 
441 	/* Get result. */
442 	byte = CSR_READ_4(sc, BGE_EE_DATA);
443 
444         *dest = (byte >> ((addr % 4) * 8)) & 0xFF;
445 
446 	return(0);
447 }
448 
449 /*
450  * Read a sequence of bytes from the EEPROM.
451  */
452 static int
453 bnx_read_eeprom(struct bnx_softc *sc, caddr_t dest, uint32_t off, size_t len)
454 {
455 	size_t i;
456 	int err;
457 	uint8_t byte;
458 
459 	for (byte = 0, err = 0, i = 0; i < len; i++) {
460 		err = bnx_eeprom_getbyte(sc, off + i, &byte);
461 		if (err)
462 			break;
463 		*(dest + i) = byte;
464 	}
465 
466 	return(err ? 1 : 0);
467 }
468 
469 static int
470 bnx_miibus_readreg(device_t dev, int phy, int reg)
471 {
472 	struct bnx_softc *sc = device_get_softc(dev);
473 	uint32_t val;
474 	int i;
475 
476 	KASSERT(phy == sc->bnx_phyno,
477 	    ("invalid phyno %d, should be %d", phy, sc->bnx_phyno));
478 
479 	/* Clear the autopoll bit if set, otherwise may trigger PCI errors. */
480 	if (sc->bnx_mi_mode & BGE_MIMODE_AUTOPOLL) {
481 		CSR_WRITE_4(sc, BGE_MI_MODE,
482 		    sc->bnx_mi_mode & ~BGE_MIMODE_AUTOPOLL);
483 		DELAY(80);
484 	}
485 
486 	CSR_WRITE_4(sc, BGE_MI_COMM, BGE_MICMD_READ | BGE_MICOMM_BUSY |
487 	    BGE_MIPHY(phy) | BGE_MIREG(reg));
488 
489 	/* Poll for the PHY register access to complete. */
490 	for (i = 0; i < BNX_TIMEOUT; i++) {
491 		DELAY(10);
492 		val = CSR_READ_4(sc, BGE_MI_COMM);
493 		if ((val & BGE_MICOMM_BUSY) == 0) {
494 			DELAY(5);
495 			val = CSR_READ_4(sc, BGE_MI_COMM);
496 			break;
497 		}
498 	}
499 	if (i == BNX_TIMEOUT) {
500 		if_printf(&sc->arpcom.ac_if, "PHY read timed out "
501 		    "(phy %d, reg %d, val 0x%08x)\n", phy, reg, val);
502 		val = 0;
503 	}
504 
505 	/* Restore the autopoll bit if necessary. */
506 	if (sc->bnx_mi_mode & BGE_MIMODE_AUTOPOLL) {
507 		CSR_WRITE_4(sc, BGE_MI_MODE, sc->bnx_mi_mode);
508 		DELAY(80);
509 	}
510 
511 	if (val & BGE_MICOMM_READFAIL)
512 		return 0;
513 
514 	return (val & 0xFFFF);
515 }
516 
517 static int
518 bnx_miibus_writereg(device_t dev, int phy, int reg, int val)
519 {
520 	struct bnx_softc *sc = device_get_softc(dev);
521 	int i;
522 
523 	KASSERT(phy == sc->bnx_phyno,
524 	    ("invalid phyno %d, should be %d", phy, sc->bnx_phyno));
525 
526 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5906 &&
527 	    (reg == BRGPHY_MII_1000CTL || reg == BRGPHY_MII_AUXCTL))
528 	       return 0;
529 
530 	/* Clear the autopoll bit if set, otherwise may trigger PCI errors. */
531 	if (sc->bnx_mi_mode & BGE_MIMODE_AUTOPOLL) {
532 		CSR_WRITE_4(sc, BGE_MI_MODE,
533 		    sc->bnx_mi_mode & ~BGE_MIMODE_AUTOPOLL);
534 		DELAY(80);
535 	}
536 
537 	CSR_WRITE_4(sc, BGE_MI_COMM, BGE_MICMD_WRITE | BGE_MICOMM_BUSY |
538 	    BGE_MIPHY(phy) | BGE_MIREG(reg) | val);
539 
540 	for (i = 0; i < BNX_TIMEOUT; i++) {
541 		DELAY(10);
542 		if (!(CSR_READ_4(sc, BGE_MI_COMM) & BGE_MICOMM_BUSY)) {
543 			DELAY(5);
544 			CSR_READ_4(sc, BGE_MI_COMM); /* dummy read */
545 			break;
546 		}
547 	}
548 	if (i == BNX_TIMEOUT) {
549 		if_printf(&sc->arpcom.ac_if, "PHY write timed out "
550 		    "(phy %d, reg %d, val %d)\n", phy, reg, val);
551 	}
552 
553 	/* Restore the autopoll bit if necessary. */
554 	if (sc->bnx_mi_mode & BGE_MIMODE_AUTOPOLL) {
555 		CSR_WRITE_4(sc, BGE_MI_MODE, sc->bnx_mi_mode);
556 		DELAY(80);
557 	}
558 
559 	return 0;
560 }
561 
562 static void
563 bnx_miibus_statchg(device_t dev)
564 {
565 	struct bnx_softc *sc;
566 	struct mii_data *mii;
567 
568 	sc = device_get_softc(dev);
569 	mii = device_get_softc(sc->bnx_miibus);
570 
571 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
572 	    (IFM_ACTIVE | IFM_AVALID)) {
573 		switch (IFM_SUBTYPE(mii->mii_media_active)) {
574 		case IFM_10_T:
575 		case IFM_100_TX:
576 			sc->bnx_link = 1;
577 			break;
578 		case IFM_1000_T:
579 		case IFM_1000_SX:
580 		case IFM_2500_SX:
581 			if (sc->bnx_asicrev != BGE_ASICREV_BCM5906)
582 				sc->bnx_link = 1;
583 			else
584 				sc->bnx_link = 0;
585 			break;
586 		default:
587 			sc->bnx_link = 0;
588 			break;
589 		}
590 	} else {
591 		sc->bnx_link = 0;
592 	}
593 	if (sc->bnx_link == 0)
594 		return;
595 
596 	BNX_CLRBIT(sc, BGE_MAC_MODE, BGE_MACMODE_PORTMODE);
597 	if (IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_T ||
598 	    IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_SX) {
599 		BNX_SETBIT(sc, BGE_MAC_MODE, BGE_PORTMODE_GMII);
600 	} else {
601 		BNX_SETBIT(sc, BGE_MAC_MODE, BGE_PORTMODE_MII);
602 	}
603 
604 	if ((mii->mii_media_active & IFM_GMASK) == IFM_FDX) {
605 		BNX_CLRBIT(sc, BGE_MAC_MODE, BGE_MACMODE_HALF_DUPLEX);
606 	} else {
607 		BNX_SETBIT(sc, BGE_MAC_MODE, BGE_MACMODE_HALF_DUPLEX);
608 	}
609 }
610 
611 /*
612  * Memory management for jumbo frames.
613  */
614 static int
615 bnx_alloc_jumbo_mem(struct bnx_softc *sc)
616 {
617 	struct ifnet *ifp = &sc->arpcom.ac_if;
618 	struct bnx_jslot *entry;
619 	uint8_t *ptr;
620 	bus_addr_t paddr;
621 	int i, error;
622 
623 	/*
624 	 * Create tag for jumbo mbufs.
625 	 * This is really a bit of a kludge. We allocate a special
626 	 * jumbo buffer pool which (thanks to the way our DMA
627 	 * memory allocation works) will consist of contiguous
628 	 * pages. This means that even though a jumbo buffer might
629 	 * be larger than a page size, we don't really need to
630 	 * map it into more than one DMA segment. However, the
631 	 * default mbuf tag will result in multi-segment mappings,
632 	 * so we have to create a special jumbo mbuf tag that
633 	 * lets us get away with mapping the jumbo buffers as
634 	 * a single segment. I think eventually the driver should
635 	 * be changed so that it uses ordinary mbufs and cluster
636 	 * buffers, i.e. jumbo frames can span multiple DMA
637 	 * descriptors. But that's a project for another day.
638 	 */
639 
640 	/*
641 	 * Create DMA stuffs for jumbo RX ring.
642 	 */
643 	error = bnx_dma_block_alloc(sc, BGE_JUMBO_RX_RING_SZ,
644 				    &sc->bnx_cdata.bnx_rx_jumbo_ring_tag,
645 				    &sc->bnx_cdata.bnx_rx_jumbo_ring_map,
646 				    (void *)&sc->bnx_ldata.bnx_rx_jumbo_ring,
647 				    &sc->bnx_ldata.bnx_rx_jumbo_ring_paddr);
648 	if (error) {
649 		if_printf(ifp, "could not create jumbo RX ring\n");
650 		return error;
651 	}
652 
653 	/*
654 	 * Create DMA stuffs for jumbo buffer block.
655 	 */
656 	error = bnx_dma_block_alloc(sc, BNX_JMEM,
657 				    &sc->bnx_cdata.bnx_jumbo_tag,
658 				    &sc->bnx_cdata.bnx_jumbo_map,
659 				    (void **)&sc->bnx_ldata.bnx_jumbo_buf,
660 				    &paddr);
661 	if (error) {
662 		if_printf(ifp, "could not create jumbo buffer\n");
663 		return error;
664 	}
665 
666 	SLIST_INIT(&sc->bnx_jfree_listhead);
667 
668 	/*
669 	 * Now divide it up into 9K pieces and save the addresses
670 	 * in an array. Note that we play an evil trick here by using
671 	 * the first few bytes in the buffer to hold the the address
672 	 * of the softc structure for this interface. This is because
673 	 * bnx_jfree() needs it, but it is called by the mbuf management
674 	 * code which will not pass it to us explicitly.
675 	 */
676 	for (i = 0, ptr = sc->bnx_ldata.bnx_jumbo_buf; i < BNX_JSLOTS; i++) {
677 		entry = &sc->bnx_cdata.bnx_jslots[i];
678 		entry->bnx_sc = sc;
679 		entry->bnx_buf = ptr;
680 		entry->bnx_paddr = paddr;
681 		entry->bnx_inuse = 0;
682 		entry->bnx_slot = i;
683 		SLIST_INSERT_HEAD(&sc->bnx_jfree_listhead, entry, jslot_link);
684 
685 		ptr += BNX_JLEN;
686 		paddr += BNX_JLEN;
687 	}
688 	return 0;
689 }
690 
691 static void
692 bnx_free_jumbo_mem(struct bnx_softc *sc)
693 {
694 	/* Destroy jumbo RX ring. */
695 	bnx_dma_block_free(sc->bnx_cdata.bnx_rx_jumbo_ring_tag,
696 			   sc->bnx_cdata.bnx_rx_jumbo_ring_map,
697 			   sc->bnx_ldata.bnx_rx_jumbo_ring);
698 
699 	/* Destroy jumbo buffer block. */
700 	bnx_dma_block_free(sc->bnx_cdata.bnx_jumbo_tag,
701 			   sc->bnx_cdata.bnx_jumbo_map,
702 			   sc->bnx_ldata.bnx_jumbo_buf);
703 }
704 
705 /*
706  * Allocate a jumbo buffer.
707  */
708 static struct bnx_jslot *
709 bnx_jalloc(struct bnx_softc *sc)
710 {
711 	struct bnx_jslot *entry;
712 
713 	lwkt_serialize_enter(&sc->bnx_jslot_serializer);
714 	entry = SLIST_FIRST(&sc->bnx_jfree_listhead);
715 	if (entry) {
716 		SLIST_REMOVE_HEAD(&sc->bnx_jfree_listhead, jslot_link);
717 		entry->bnx_inuse = 1;
718 	} else {
719 		if_printf(&sc->arpcom.ac_if, "no free jumbo buffers\n");
720 	}
721 	lwkt_serialize_exit(&sc->bnx_jslot_serializer);
722 	return(entry);
723 }
724 
725 /*
726  * Adjust usage count on a jumbo buffer.
727  */
728 static void
729 bnx_jref(void *arg)
730 {
731 	struct bnx_jslot *entry = (struct bnx_jslot *)arg;
732 	struct bnx_softc *sc = entry->bnx_sc;
733 
734 	if (sc == NULL)
735 		panic("bnx_jref: can't find softc pointer!");
736 
737 	if (&sc->bnx_cdata.bnx_jslots[entry->bnx_slot] != entry) {
738 		panic("bnx_jref: asked to reference buffer "
739 		    "that we don't manage!");
740 	} else if (entry->bnx_inuse == 0) {
741 		panic("bnx_jref: buffer already free!");
742 	} else {
743 		atomic_add_int(&entry->bnx_inuse, 1);
744 	}
745 }
746 
747 /*
748  * Release a jumbo buffer.
749  */
750 static void
751 bnx_jfree(void *arg)
752 {
753 	struct bnx_jslot *entry = (struct bnx_jslot *)arg;
754 	struct bnx_softc *sc = entry->bnx_sc;
755 
756 	if (sc == NULL)
757 		panic("bnx_jfree: can't find softc pointer!");
758 
759 	if (&sc->bnx_cdata.bnx_jslots[entry->bnx_slot] != entry) {
760 		panic("bnx_jfree: asked to free buffer that we don't manage!");
761 	} else if (entry->bnx_inuse == 0) {
762 		panic("bnx_jfree: buffer already free!");
763 	} else {
764 		/*
765 		 * Possible MP race to 0, use the serializer.  The atomic insn
766 		 * is still needed for races against bnx_jref().
767 		 */
768 		lwkt_serialize_enter(&sc->bnx_jslot_serializer);
769 		atomic_subtract_int(&entry->bnx_inuse, 1);
770 		if (entry->bnx_inuse == 0) {
771 			SLIST_INSERT_HEAD(&sc->bnx_jfree_listhead,
772 					  entry, jslot_link);
773 		}
774 		lwkt_serialize_exit(&sc->bnx_jslot_serializer);
775 	}
776 }
777 
778 
779 /*
780  * Intialize a standard receive ring descriptor.
781  */
782 static int
783 bnx_newbuf_std(struct bnx_softc *sc, int i, int init)
784 {
785 	struct mbuf *m_new = NULL;
786 	bus_dma_segment_t seg;
787 	bus_dmamap_t map;
788 	int error, nsegs;
789 
790 	m_new = m_getcl(init ? MB_WAIT : MB_DONTWAIT, MT_DATA, M_PKTHDR);
791 	if (m_new == NULL)
792 		return ENOBUFS;
793 	m_new->m_len = m_new->m_pkthdr.len = MCLBYTES;
794 	m_adj(m_new, ETHER_ALIGN);
795 
796 	error = bus_dmamap_load_mbuf_segment(sc->bnx_cdata.bnx_rx_mtag,
797 			sc->bnx_cdata.bnx_rx_tmpmap, m_new,
798 			&seg, 1, &nsegs, BUS_DMA_NOWAIT);
799 	if (error) {
800 		m_freem(m_new);
801 		return error;
802 	}
803 
804 	if (!init) {
805 		bus_dmamap_sync(sc->bnx_cdata.bnx_rx_mtag,
806 				sc->bnx_cdata.bnx_rx_std_dmamap[i],
807 				BUS_DMASYNC_POSTREAD);
808 		bus_dmamap_unload(sc->bnx_cdata.bnx_rx_mtag,
809 			sc->bnx_cdata.bnx_rx_std_dmamap[i]);
810 	}
811 
812 	map = sc->bnx_cdata.bnx_rx_tmpmap;
813 	sc->bnx_cdata.bnx_rx_tmpmap = sc->bnx_cdata.bnx_rx_std_dmamap[i];
814 	sc->bnx_cdata.bnx_rx_std_dmamap[i] = map;
815 
816 	sc->bnx_cdata.bnx_rx_std_chain[i].bnx_mbuf = m_new;
817 	sc->bnx_cdata.bnx_rx_std_chain[i].bnx_paddr = seg.ds_addr;
818 
819 	bnx_setup_rxdesc_std(sc, i);
820 	return 0;
821 }
822 
823 static void
824 bnx_setup_rxdesc_std(struct bnx_softc *sc, int i)
825 {
826 	struct bnx_rxchain *rc;
827 	struct bge_rx_bd *r;
828 
829 	rc = &sc->bnx_cdata.bnx_rx_std_chain[i];
830 	r = &sc->bnx_ldata.bnx_rx_std_ring[i];
831 
832 	r->bge_addr.bge_addr_lo = BGE_ADDR_LO(rc->bnx_paddr);
833 	r->bge_addr.bge_addr_hi = BGE_ADDR_HI(rc->bnx_paddr);
834 	r->bge_len = rc->bnx_mbuf->m_len;
835 	r->bge_idx = i;
836 	r->bge_flags = BGE_RXBDFLAG_END;
837 }
838 
839 /*
840  * Initialize a jumbo receive ring descriptor. This allocates
841  * a jumbo buffer from the pool managed internally by the driver.
842  */
843 static int
844 bnx_newbuf_jumbo(struct bnx_softc *sc, int i, int init)
845 {
846 	struct mbuf *m_new = NULL;
847 	struct bnx_jslot *buf;
848 	bus_addr_t paddr;
849 
850 	/* Allocate the mbuf. */
851 	MGETHDR(m_new, init ? MB_WAIT : MB_DONTWAIT, MT_DATA);
852 	if (m_new == NULL)
853 		return ENOBUFS;
854 
855 	/* Allocate the jumbo buffer */
856 	buf = bnx_jalloc(sc);
857 	if (buf == NULL) {
858 		m_freem(m_new);
859 		return ENOBUFS;
860 	}
861 
862 	/* Attach the buffer to the mbuf. */
863 	m_new->m_ext.ext_arg = buf;
864 	m_new->m_ext.ext_buf = buf->bnx_buf;
865 	m_new->m_ext.ext_free = bnx_jfree;
866 	m_new->m_ext.ext_ref = bnx_jref;
867 	m_new->m_ext.ext_size = BNX_JUMBO_FRAMELEN;
868 
869 	m_new->m_flags |= M_EXT;
870 
871 	m_new->m_data = m_new->m_ext.ext_buf;
872 	m_new->m_len = m_new->m_pkthdr.len = m_new->m_ext.ext_size;
873 
874 	paddr = buf->bnx_paddr;
875 	m_adj(m_new, ETHER_ALIGN);
876 	paddr += ETHER_ALIGN;
877 
878 	/* Save necessary information */
879 	sc->bnx_cdata.bnx_rx_jumbo_chain[i].bnx_mbuf = m_new;
880 	sc->bnx_cdata.bnx_rx_jumbo_chain[i].bnx_paddr = paddr;
881 
882 	/* Set up the descriptor. */
883 	bnx_setup_rxdesc_jumbo(sc, i);
884 	return 0;
885 }
886 
887 static void
888 bnx_setup_rxdesc_jumbo(struct bnx_softc *sc, int i)
889 {
890 	struct bge_rx_bd *r;
891 	struct bnx_rxchain *rc;
892 
893 	r = &sc->bnx_ldata.bnx_rx_jumbo_ring[i];
894 	rc = &sc->bnx_cdata.bnx_rx_jumbo_chain[i];
895 
896 	r->bge_addr.bge_addr_lo = BGE_ADDR_LO(rc->bnx_paddr);
897 	r->bge_addr.bge_addr_hi = BGE_ADDR_HI(rc->bnx_paddr);
898 	r->bge_len = rc->bnx_mbuf->m_len;
899 	r->bge_idx = i;
900 	r->bge_flags = BGE_RXBDFLAG_END|BGE_RXBDFLAG_JUMBO_RING;
901 }
902 
903 static int
904 bnx_init_rx_ring_std(struct bnx_softc *sc)
905 {
906 	int i, error;
907 
908 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
909 		error = bnx_newbuf_std(sc, i, 1);
910 		if (error)
911 			return error;
912 	}
913 
914 	sc->bnx_std = BGE_STD_RX_RING_CNT - 1;
915 	bnx_writembx(sc, BGE_MBX_RX_STD_PROD_LO, sc->bnx_std);
916 
917 	return(0);
918 }
919 
920 static void
921 bnx_free_rx_ring_std(struct bnx_softc *sc)
922 {
923 	int i;
924 
925 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
926 		struct bnx_rxchain *rc = &sc->bnx_cdata.bnx_rx_std_chain[i];
927 
928 		if (rc->bnx_mbuf != NULL) {
929 			bus_dmamap_unload(sc->bnx_cdata.bnx_rx_mtag,
930 					  sc->bnx_cdata.bnx_rx_std_dmamap[i]);
931 			m_freem(rc->bnx_mbuf);
932 			rc->bnx_mbuf = NULL;
933 		}
934 		bzero(&sc->bnx_ldata.bnx_rx_std_ring[i],
935 		    sizeof(struct bge_rx_bd));
936 	}
937 }
938 
939 static int
940 bnx_init_rx_ring_jumbo(struct bnx_softc *sc)
941 {
942 	struct bge_rcb *rcb;
943 	int i, error;
944 
945 	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
946 		error = bnx_newbuf_jumbo(sc, i, 1);
947 		if (error)
948 			return error;
949 	}
950 
951 	sc->bnx_jumbo = BGE_JUMBO_RX_RING_CNT - 1;
952 
953 	rcb = &sc->bnx_ldata.bnx_info.bnx_jumbo_rx_rcb;
954 	rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(0, 0);
955 	CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
956 
957 	bnx_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, sc->bnx_jumbo);
958 
959 	return(0);
960 }
961 
962 static void
963 bnx_free_rx_ring_jumbo(struct bnx_softc *sc)
964 {
965 	int i;
966 
967 	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
968 		struct bnx_rxchain *rc = &sc->bnx_cdata.bnx_rx_jumbo_chain[i];
969 
970 		if (rc->bnx_mbuf != NULL) {
971 			m_freem(rc->bnx_mbuf);
972 			rc->bnx_mbuf = NULL;
973 		}
974 		bzero(&sc->bnx_ldata.bnx_rx_jumbo_ring[i],
975 		    sizeof(struct bge_rx_bd));
976 	}
977 }
978 
979 static void
980 bnx_free_tx_ring(struct bnx_softc *sc)
981 {
982 	int i;
983 
984 	for (i = 0; i < BGE_TX_RING_CNT; i++) {
985 		if (sc->bnx_cdata.bnx_tx_chain[i] != NULL) {
986 			bus_dmamap_unload(sc->bnx_cdata.bnx_tx_mtag,
987 					  sc->bnx_cdata.bnx_tx_dmamap[i]);
988 			m_freem(sc->bnx_cdata.bnx_tx_chain[i]);
989 			sc->bnx_cdata.bnx_tx_chain[i] = NULL;
990 		}
991 		bzero(&sc->bnx_ldata.bnx_tx_ring[i],
992 		    sizeof(struct bge_tx_bd));
993 	}
994 }
995 
996 static int
997 bnx_init_tx_ring(struct bnx_softc *sc)
998 {
999 	sc->bnx_txcnt = 0;
1000 	sc->bnx_tx_saved_considx = 0;
1001 	sc->bnx_tx_prodidx = 0;
1002 
1003 	/* Initialize transmit producer index for host-memory send ring. */
1004 	bnx_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bnx_tx_prodidx);
1005 	bnx_writembx(sc, BGE_MBX_TX_NIC_PROD0_LO, 0);
1006 
1007 	return(0);
1008 }
1009 
1010 static void
1011 bnx_setmulti(struct bnx_softc *sc)
1012 {
1013 	struct ifnet *ifp;
1014 	struct ifmultiaddr *ifma;
1015 	uint32_t hashes[4] = { 0, 0, 0, 0 };
1016 	int h, i;
1017 
1018 	ifp = &sc->arpcom.ac_if;
1019 
1020 	if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
1021 		for (i = 0; i < 4; i++)
1022 			CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), 0xFFFFFFFF);
1023 		return;
1024 	}
1025 
1026 	/* First, zot all the existing filters. */
1027 	for (i = 0; i < 4; i++)
1028 		CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), 0);
1029 
1030 	/* Now program new ones. */
1031 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1032 		if (ifma->ifma_addr->sa_family != AF_LINK)
1033 			continue;
1034 		h = ether_crc32_le(
1035 		    LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
1036 		    ETHER_ADDR_LEN) & 0x7f;
1037 		hashes[(h & 0x60) >> 5] |= 1 << (h & 0x1F);
1038 	}
1039 
1040 	for (i = 0; i < 4; i++)
1041 		CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), hashes[i]);
1042 }
1043 
1044 /*
1045  * Do endian, PCI and DMA initialization. Also check the on-board ROM
1046  * self-test results.
1047  */
1048 static int
1049 bnx_chipinit(struct bnx_softc *sc)
1050 {
1051 	uint32_t dma_rw_ctl, mode_ctl;
1052 	int i;
1053 
1054 	/* Set endian type before we access any non-PCI registers. */
1055 	pci_write_config(sc->bnx_dev, BGE_PCI_MISC_CTL,
1056 	    BGE_INIT | BGE_PCIMISCCTL_TAGGED_STATUS, 4);
1057 
1058 	/* Clear the MAC control register */
1059 	CSR_WRITE_4(sc, BGE_MAC_MODE, 0);
1060 
1061 	/*
1062 	 * Clear the MAC statistics block in the NIC's
1063 	 * internal memory.
1064 	 */
1065 	for (i = BGE_STATS_BLOCK;
1066 	    i < BGE_STATS_BLOCK_END + 1; i += sizeof(uint32_t))
1067 		BNX_MEMWIN_WRITE(sc, i, 0);
1068 
1069 	for (i = BGE_STATUS_BLOCK;
1070 	    i < BGE_STATUS_BLOCK_END + 1; i += sizeof(uint32_t))
1071 		BNX_MEMWIN_WRITE(sc, i, 0);
1072 
1073 	if (BNX_IS_57765_FAMILY(sc)) {
1074 		uint32_t val;
1075 
1076 		if (sc->bnx_chipid == BGE_CHIPID_BCM57765_A0) {
1077 			mode_ctl = CSR_READ_4(sc, BGE_MODE_CTL);
1078 			val = mode_ctl & ~BGE_MODECTL_PCIE_PORTS;
1079 
1080 			/* Access the lower 1K of PL PCI-E block registers. */
1081 			CSR_WRITE_4(sc, BGE_MODE_CTL,
1082 			    val | BGE_MODECTL_PCIE_PL_SEL);
1083 
1084 			val = CSR_READ_4(sc, BGE_PCIE_PL_LO_PHYCTL5);
1085 			val |= BGE_PCIE_PL_LO_PHYCTL5_DIS_L2CLKREQ;
1086 			CSR_WRITE_4(sc, BGE_PCIE_PL_LO_PHYCTL5, val);
1087 
1088 			CSR_WRITE_4(sc, BGE_MODE_CTL, mode_ctl);
1089 		}
1090 		if (sc->bnx_chiprev != BGE_CHIPREV_57765_AX) {
1091 			mode_ctl = CSR_READ_4(sc, BGE_MODE_CTL);
1092 			val = mode_ctl & ~BGE_MODECTL_PCIE_PORTS;
1093 
1094 			/* Access the lower 1K of DL PCI-E block registers. */
1095 			CSR_WRITE_4(sc, BGE_MODE_CTL,
1096 			    val | BGE_MODECTL_PCIE_DL_SEL);
1097 
1098 			val = CSR_READ_4(sc, BGE_PCIE_DL_LO_FTSMAX);
1099 			val &= ~BGE_PCIE_DL_LO_FTSMAX_MASK;
1100 			val |= BGE_PCIE_DL_LO_FTSMAX_VAL;
1101 			CSR_WRITE_4(sc, BGE_PCIE_DL_LO_FTSMAX, val);
1102 
1103 			CSR_WRITE_4(sc, BGE_MODE_CTL, mode_ctl);
1104 		}
1105 
1106 		val = CSR_READ_4(sc, BGE_CPMU_LSPD_10MB_CLK);
1107 		val &= ~BGE_CPMU_LSPD_10MB_MACCLK_MASK;
1108 		val |= BGE_CPMU_LSPD_10MB_MACCLK_6_25;
1109 		CSR_WRITE_4(sc, BGE_CPMU_LSPD_10MB_CLK, val);
1110 	}
1111 
1112 	/*
1113 	 * Set up the PCI DMA control register.
1114 	 */
1115 	dma_rw_ctl = pci_read_config(sc->bnx_dev, BGE_PCI_DMA_RW_CTL, 4);
1116 	/*
1117 	 * Disable 32bytes cache alignment for DMA write to host memory
1118 	 *
1119 	 * NOTE:
1120 	 * 64bytes cache alignment for DMA write to host memory is still
1121 	 * enabled.
1122 	 */
1123 	dma_rw_ctl |= BGE_PCIDMARWCTL_DIS_CACHE_ALIGNMENT;
1124 	if (sc->bnx_chipid == BGE_CHIPID_BCM57765_A0)
1125 		dma_rw_ctl &= ~BGE_PCIDMARWCTL_CRDRDR_RDMA_MRRS_MSK;
1126 	/*
1127 	 * Enable HW workaround for controllers that misinterpret
1128 	 * a status tag update and leave interrupts permanently
1129 	 * disabled.
1130 	 */
1131 	if (sc->bnx_asicrev != BGE_ASICREV_BCM5717 &&
1132 	    !BNX_IS_57765_FAMILY(sc))
1133 		dma_rw_ctl |= BGE_PCIDMARWCTL_TAGGED_STATUS_WA;
1134 	if (bootverbose) {
1135 		if_printf(&sc->arpcom.ac_if, "DMA read/write %#x\n",
1136 		    dma_rw_ctl);
1137 	}
1138 	pci_write_config(sc->bnx_dev, BGE_PCI_DMA_RW_CTL, dma_rw_ctl, 4);
1139 
1140 	/*
1141 	 * Set up general mode register.
1142 	 */
1143 	mode_ctl = bnx_dma_swap_options(sc) | BGE_MODECTL_MAC_ATTN_INTR |
1144 	    BGE_MODECTL_HOST_SEND_BDS | BGE_MODECTL_TX_NO_PHDR_CSUM;
1145 	CSR_WRITE_4(sc, BGE_MODE_CTL, mode_ctl);
1146 
1147 	/*
1148 	 * Disable memory write invalidate.  Apparently it is not supported
1149 	 * properly by these devices.  Also ensure that INTx isn't disabled,
1150 	 * as these chips need it even when using MSI.
1151 	 */
1152 	PCI_CLRBIT(sc->bnx_dev, BGE_PCI_CMD,
1153 	    (PCIM_CMD_MWRICEN | PCIM_CMD_INTxDIS), 4);
1154 
1155 	/* Set the timer prescaler (always 66Mhz) */
1156 	CSR_WRITE_4(sc, BGE_MISC_CFG, 65 << 1/*BGE_32BITTIME_66MHZ*/);
1157 
1158 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5906) {
1159 		DELAY(40);	/* XXX */
1160 
1161 		/* Put PHY into ready state */
1162 		BNX_CLRBIT(sc, BGE_MISC_CFG, BGE_MISCCFG_EPHY_IDDQ);
1163 		CSR_READ_4(sc, BGE_MISC_CFG); /* Flush */
1164 		DELAY(40);
1165 	}
1166 
1167 	return(0);
1168 }
1169 
1170 static int
1171 bnx_blockinit(struct bnx_softc *sc)
1172 {
1173 	struct bge_rcb *rcb;
1174 	bus_size_t vrcb;
1175 	bge_hostaddr taddr;
1176 	uint32_t val;
1177 	int i, limit;
1178 
1179 	/*
1180 	 * Initialize the memory window pointer register so that
1181 	 * we can access the first 32K of internal NIC RAM. This will
1182 	 * allow us to set up the TX send ring RCBs and the RX return
1183 	 * ring RCBs, plus other things which live in NIC memory.
1184 	 */
1185 	CSR_WRITE_4(sc, BGE_PCI_MEMWIN_BASEADDR, 0);
1186 
1187 	/* Configure mbuf pool watermarks */
1188 	if (BNX_IS_57765_PLUS(sc)) {
1189 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
1190 		if (sc->arpcom.ac_if.if_mtu > ETHERMTU) {
1191 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x7e);
1192 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0xea);
1193 		} else {
1194 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x2a);
1195 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0xa0);
1196 		}
1197 	} else if (sc->bnx_asicrev == BGE_ASICREV_BCM5906) {
1198 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
1199 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x04);
1200 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x10);
1201 	} else {
1202 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
1203 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x10);
1204 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60);
1205 	}
1206 
1207 	/* Configure DMA resource watermarks */
1208 	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LOWAT, 5);
1209 	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_HIWAT, 10);
1210 
1211 	/* Enable buffer manager */
1212 	val = BGE_BMANMODE_ENABLE | BGE_BMANMODE_LOMBUF_ATTN;
1213 	/*
1214 	 * Change the arbitration algorithm of TXMBUF read request to
1215 	 * round-robin instead of priority based for BCM5719.  When
1216 	 * TXFIFO is almost empty, RDMA will hold its request until
1217 	 * TXFIFO is not almost empty.
1218 	 */
1219 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5719)
1220 		val |= BGE_BMANMODE_NO_TX_UNDERRUN;
1221 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5717 ||
1222 	    sc->bnx_chipid == BGE_CHIPID_BCM5719_A0 ||
1223 	    sc->bnx_chipid == BGE_CHIPID_BCM5720_A0)
1224 		val |= BGE_BMANMODE_LOMBUF_ATTN;
1225 	CSR_WRITE_4(sc, BGE_BMAN_MODE, val);
1226 
1227 	/* Poll for buffer manager start indication */
1228 	for (i = 0; i < BNX_TIMEOUT; i++) {
1229 		if (CSR_READ_4(sc, BGE_BMAN_MODE) & BGE_BMANMODE_ENABLE)
1230 			break;
1231 		DELAY(10);
1232 	}
1233 
1234 	if (i == BNX_TIMEOUT) {
1235 		if_printf(&sc->arpcom.ac_if,
1236 			  "buffer manager failed to start\n");
1237 		return(ENXIO);
1238 	}
1239 
1240 	/* Enable flow-through queues */
1241 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
1242 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
1243 
1244 	/* Wait until queue initialization is complete */
1245 	for (i = 0; i < BNX_TIMEOUT; i++) {
1246 		if (CSR_READ_4(sc, BGE_FTQ_RESET) == 0)
1247 			break;
1248 		DELAY(10);
1249 	}
1250 
1251 	if (i == BNX_TIMEOUT) {
1252 		if_printf(&sc->arpcom.ac_if,
1253 			  "flow-through queue init failed\n");
1254 		return(ENXIO);
1255 	}
1256 
1257 	/*
1258 	 * Summary of rings supported by the controller:
1259 	 *
1260 	 * Standard Receive Producer Ring
1261 	 * - This ring is used to feed receive buffers for "standard"
1262 	 *   sized frames (typically 1536 bytes) to the controller.
1263 	 *
1264 	 * Jumbo Receive Producer Ring
1265 	 * - This ring is used to feed receive buffers for jumbo sized
1266 	 *   frames (i.e. anything bigger than the "standard" frames)
1267 	 *   to the controller.
1268 	 *
1269 	 * Mini Receive Producer Ring
1270 	 * - This ring is used to feed receive buffers for "mini"
1271 	 *   sized frames to the controller.
1272 	 * - This feature required external memory for the controller
1273 	 *   but was never used in a production system.  Should always
1274 	 *   be disabled.
1275 	 *
1276 	 * Receive Return Ring
1277 	 * - After the controller has placed an incoming frame into a
1278 	 *   receive buffer that buffer is moved into a receive return
1279 	 *   ring.  The driver is then responsible to passing the
1280 	 *   buffer up to the stack.  Many versions of the controller
1281 	 *   support multiple RR rings.
1282 	 *
1283 	 * Send Ring
1284 	 * - This ring is used for outgoing frames.  Many versions of
1285 	 *   the controller support multiple send rings.
1286 	 */
1287 
1288 	/* Initialize the standard receive producer ring control block. */
1289 	rcb = &sc->bnx_ldata.bnx_info.bnx_std_rx_rcb;
1290 	rcb->bge_hostaddr.bge_addr_lo =
1291 	    BGE_ADDR_LO(sc->bnx_ldata.bnx_rx_std_ring_paddr);
1292 	rcb->bge_hostaddr.bge_addr_hi =
1293 	    BGE_ADDR_HI(sc->bnx_ldata.bnx_rx_std_ring_paddr);
1294 	if (BNX_IS_57765_PLUS(sc)) {
1295 		/*
1296 		 * Bits 31-16: Programmable ring size (2048, 1024, 512, .., 32)
1297 		 * Bits 15-2 : Maximum RX frame size
1298 		 * Bit 1     : 1 = Ring Disabled, 0 = Ring ENabled
1299 		 * Bit 0     : Reserved
1300 		 */
1301 		rcb->bge_maxlen_flags =
1302 		    BGE_RCB_MAXLEN_FLAGS(512, BNX_MAX_FRAMELEN << 2);
1303 	} else {
1304 		/*
1305 		 * Bits 31-16: Programmable ring size (512, 256, 128, 64, 32)
1306 		 * Bits 15-2 : Reserved (should be 0)
1307 		 * Bit 1     : 1 = Ring Disabled, 0 = Ring Enabled
1308 		 * Bit 0     : Reserved
1309 		 */
1310 		rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(512, 0);
1311 	}
1312 	if (BNX_IS_5717_PLUS(sc))
1313 		rcb->bge_nicaddr = BGE_STD_RX_RINGS_5717;
1314 	else
1315 		rcb->bge_nicaddr = BGE_STD_RX_RINGS;
1316 	/* Write the standard receive producer ring control block. */
1317 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_HI, rcb->bge_hostaddr.bge_addr_hi);
1318 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_LO, rcb->bge_hostaddr.bge_addr_lo);
1319 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
1320 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_NICADDR, rcb->bge_nicaddr);
1321 	/* Reset the standard receive producer ring producer index. */
1322 	bnx_writembx(sc, BGE_MBX_RX_STD_PROD_LO, 0);
1323 
1324 	/*
1325 	 * Initialize the jumbo RX producer ring control
1326 	 * block.  We set the 'ring disabled' bit in the
1327 	 * flags field until we're actually ready to start
1328 	 * using this ring (i.e. once we set the MTU
1329 	 * high enough to require it).
1330 	 */
1331 	if (BNX_IS_JUMBO_CAPABLE(sc)) {
1332 		rcb = &sc->bnx_ldata.bnx_info.bnx_jumbo_rx_rcb;
1333 		/* Get the jumbo receive producer ring RCB parameters. */
1334 		rcb->bge_hostaddr.bge_addr_lo =
1335 		    BGE_ADDR_LO(sc->bnx_ldata.bnx_rx_jumbo_ring_paddr);
1336 		rcb->bge_hostaddr.bge_addr_hi =
1337 		    BGE_ADDR_HI(sc->bnx_ldata.bnx_rx_jumbo_ring_paddr);
1338 		rcb->bge_maxlen_flags =
1339 		    BGE_RCB_MAXLEN_FLAGS(BNX_MAX_FRAMELEN,
1340 		    BGE_RCB_FLAG_RING_DISABLED);
1341 		if (BNX_IS_5717_PLUS(sc))
1342 			rcb->bge_nicaddr = BGE_JUMBO_RX_RINGS_5717;
1343 		else
1344 			rcb->bge_nicaddr = BGE_JUMBO_RX_RINGS;
1345 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_HI,
1346 		    rcb->bge_hostaddr.bge_addr_hi);
1347 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_LO,
1348 		    rcb->bge_hostaddr.bge_addr_lo);
1349 		/* Program the jumbo receive producer ring RCB parameters. */
1350 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS,
1351 		    rcb->bge_maxlen_flags);
1352 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_NICADDR, rcb->bge_nicaddr);
1353 		/* Reset the jumbo receive producer ring producer index. */
1354 		bnx_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, 0);
1355 	}
1356 
1357 	/* Choose de-pipeline mode for BCM5906 A0, A1 and A2. */
1358 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5906 &&
1359 	    (sc->bnx_chipid == BGE_CHIPID_BCM5906_A0 ||
1360 	     sc->bnx_chipid == BGE_CHIPID_BCM5906_A1 ||
1361 	     sc->bnx_chipid == BGE_CHIPID_BCM5906_A2)) {
1362 		CSR_WRITE_4(sc, BGE_ISO_PKT_TX,
1363 		    (CSR_READ_4(sc, BGE_ISO_PKT_TX) & ~3) | 2);
1364 	}
1365 
1366 	/*
1367 	 * The BD ring replenish thresholds control how often the
1368 	 * hardware fetches new BD's from the producer rings in host
1369 	 * memory.  Setting the value too low on a busy system can
1370 	 * starve the hardware and recue the throughpout.
1371 	 *
1372 	 * Set the BD ring replentish thresholds. The recommended
1373 	 * values are 1/8th the number of descriptors allocated to
1374 	 * each ring.
1375 	 */
1376 	val = 8;
1377 	CSR_WRITE_4(sc, BGE_RBDI_STD_REPL_THRESH, val);
1378 	if (BNX_IS_JUMBO_CAPABLE(sc)) {
1379 		CSR_WRITE_4(sc, BGE_RBDI_JUMBO_REPL_THRESH,
1380 		    BGE_JUMBO_RX_RING_CNT/8);
1381 	}
1382 	if (BNX_IS_57765_PLUS(sc)) {
1383 		CSR_WRITE_4(sc, BGE_STD_REPLENISH_LWM, 32);
1384 		CSR_WRITE_4(sc, BGE_JMB_REPLENISH_LWM, 16);
1385 	}
1386 
1387 	/*
1388 	 * Disable all send rings by setting the 'ring disabled' bit
1389 	 * in the flags field of all the TX send ring control blocks,
1390 	 * located in NIC memory.
1391 	 */
1392 	if (BNX_IS_5717_PLUS(sc))
1393 		limit = 4;
1394 	else if (BNX_IS_57765_FAMILY(sc))
1395 		limit = 2;
1396 	else
1397 		limit = 1;
1398 	vrcb = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
1399 	for (i = 0; i < limit; i++) {
1400 		RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1401 		    BGE_RCB_MAXLEN_FLAGS(0, BGE_RCB_FLAG_RING_DISABLED));
1402 		RCB_WRITE_4(sc, vrcb, bge_nicaddr, 0);
1403 		vrcb += sizeof(struct bge_rcb);
1404 	}
1405 
1406 	/* Configure send ring RCB 0 (we use only the first ring) */
1407 	vrcb = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
1408 	BGE_HOSTADDR(taddr, sc->bnx_ldata.bnx_tx_ring_paddr);
1409 	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
1410 	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
1411 	if (BNX_IS_5717_PLUS(sc)) {
1412 		RCB_WRITE_4(sc, vrcb, bge_nicaddr, BGE_SEND_RING_5717);
1413 	} else {
1414 		RCB_WRITE_4(sc, vrcb, bge_nicaddr,
1415 		    BGE_NIC_TXRING_ADDR(0, BGE_TX_RING_CNT));
1416 	}
1417 	RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1418 	    BGE_RCB_MAXLEN_FLAGS(BGE_TX_RING_CNT, 0));
1419 
1420 	/*
1421 	 * Disable all receive return rings by setting the
1422 	 * 'ring disabled' bit in the flags field of all the receive
1423 	 * return ring control blocks, located in NIC memory.
1424 	 */
1425 	if (BNX_IS_5717_PLUS(sc)) {
1426 		/* Should be 17, use 16 until we get an SRAM map. */
1427 		limit = 16;
1428 	} else if (BNX_IS_57765_FAMILY(sc)) {
1429 		limit = 4;
1430 	} else {
1431 		limit = 1;
1432 	}
1433 	/* Disable all receive return rings. */
1434 	vrcb = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
1435 	for (i = 0; i < limit; i++) {
1436 		RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, 0);
1437 		RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, 0);
1438 		RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1439 		    BGE_RCB_FLAG_RING_DISABLED);
1440 		RCB_WRITE_4(sc, vrcb, bge_nicaddr, 0);
1441 		bnx_writembx(sc, BGE_MBX_RX_CONS0_LO +
1442 		    (i * (sizeof(uint64_t))), 0);
1443 		vrcb += sizeof(struct bge_rcb);
1444 	}
1445 
1446 	/*
1447 	 * Set up receive return ring 0.  Note that the NIC address
1448 	 * for RX return rings is 0x0.  The return rings live entirely
1449 	 * within the host, so the nicaddr field in the RCB isn't used.
1450 	 */
1451 	vrcb = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
1452 	BGE_HOSTADDR(taddr, sc->bnx_ldata.bnx_rx_return_ring_paddr);
1453 	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
1454 	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
1455 	RCB_WRITE_4(sc, vrcb, bge_nicaddr, 0);
1456 	RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1457 	    BGE_RCB_MAXLEN_FLAGS(sc->bnx_return_ring_cnt, 0));
1458 
1459 	/* Set random backoff seed for TX */
1460 	CSR_WRITE_4(sc, BGE_TX_RANDOM_BACKOFF,
1461 	    sc->arpcom.ac_enaddr[0] + sc->arpcom.ac_enaddr[1] +
1462 	    sc->arpcom.ac_enaddr[2] + sc->arpcom.ac_enaddr[3] +
1463 	    sc->arpcom.ac_enaddr[4] + sc->arpcom.ac_enaddr[5] +
1464 	    BGE_TX_BACKOFF_SEED_MASK);
1465 
1466 	/* Set inter-packet gap */
1467 	val = 0x2620;
1468 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5720) {
1469 		val |= CSR_READ_4(sc, BGE_TX_LENGTHS) &
1470 		    (BGE_TXLEN_JMB_FRM_LEN_MSK | BGE_TXLEN_CNT_DN_VAL_MSK);
1471 	}
1472 	CSR_WRITE_4(sc, BGE_TX_LENGTHS, val);
1473 
1474 	/*
1475 	 * Specify which ring to use for packets that don't match
1476 	 * any RX rules.
1477 	 */
1478 	CSR_WRITE_4(sc, BGE_RX_RULES_CFG, 0x08);
1479 
1480 	/*
1481 	 * Configure number of RX lists. One interrupt distribution
1482 	 * list, sixteen active lists, one bad frames class.
1483 	 */
1484 	CSR_WRITE_4(sc, BGE_RXLP_CFG, 0x181);
1485 
1486 	/* Inialize RX list placement stats mask. */
1487 	CSR_WRITE_4(sc, BGE_RXLP_STATS_ENABLE_MASK, 0x007FFFFF);
1488 	CSR_WRITE_4(sc, BGE_RXLP_STATS_CTL, 0x1);
1489 
1490 	/* Disable host coalescing until we get it set up */
1491 	CSR_WRITE_4(sc, BGE_HCC_MODE, 0x00000000);
1492 
1493 	/* Poll to make sure it's shut down. */
1494 	for (i = 0; i < BNX_TIMEOUT; i++) {
1495 		if (!(CSR_READ_4(sc, BGE_HCC_MODE) & BGE_HCCMODE_ENABLE))
1496 			break;
1497 		DELAY(10);
1498 	}
1499 
1500 	if (i == BNX_TIMEOUT) {
1501 		if_printf(&sc->arpcom.ac_if,
1502 			  "host coalescing engine failed to idle\n");
1503 		return(ENXIO);
1504 	}
1505 
1506 	/* Set up host coalescing defaults */
1507 	CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS, sc->bnx_rx_coal_ticks);
1508 	CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS, sc->bnx_tx_coal_ticks);
1509 	CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS, sc->bnx_rx_coal_bds);
1510 	CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS, sc->bnx_tx_coal_bds);
1511 	CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT, sc->bnx_rx_coal_bds_int);
1512 	CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT, sc->bnx_tx_coal_bds_int);
1513 
1514 	/* Set up address of status block */
1515 	bzero(sc->bnx_ldata.bnx_status_block, BGE_STATUS_BLK_SZ);
1516 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_HI,
1517 	    BGE_ADDR_HI(sc->bnx_ldata.bnx_status_block_paddr));
1518 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_LO,
1519 	    BGE_ADDR_LO(sc->bnx_ldata.bnx_status_block_paddr));
1520 
1521 	/* Set up status block partail update size. */
1522 	val = BGE_STATBLKSZ_32BYTE;
1523 #if 0
1524 	/*
1525 	 * Does not seem to have visible effect in both
1526 	 * bulk data (1472B UDP datagram) and tiny data
1527 	 * (18B UDP datagram) TX tests.
1528 	 */
1529 	val |= BGE_HCCMODE_CLRTICK_TX;
1530 #endif
1531 	/* Turn on host coalescing state machine */
1532 	CSR_WRITE_4(sc, BGE_HCC_MODE, val | BGE_HCCMODE_ENABLE);
1533 
1534 	/* Turn on RX BD completion state machine and enable attentions */
1535 	CSR_WRITE_4(sc, BGE_RBDC_MODE,
1536 	    BGE_RBDCMODE_ENABLE|BGE_RBDCMODE_ATTN);
1537 
1538 	/* Turn on RX list placement state machine */
1539 	CSR_WRITE_4(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
1540 
1541 	val = BGE_MACMODE_TXDMA_ENB | BGE_MACMODE_RXDMA_ENB |
1542 	    BGE_MACMODE_RX_STATS_CLEAR | BGE_MACMODE_TX_STATS_CLEAR |
1543 	    BGE_MACMODE_RX_STATS_ENB | BGE_MACMODE_TX_STATS_ENB |
1544 	    BGE_MACMODE_FRMHDR_DMA_ENB;
1545 
1546 	if (sc->bnx_flags & BNX_FLAG_TBI)
1547 		val |= BGE_PORTMODE_TBI;
1548 	else if (sc->bnx_flags & BNX_FLAG_MII_SERDES)
1549 		val |= BGE_PORTMODE_GMII;
1550 	else
1551 		val |= BGE_PORTMODE_MII;
1552 
1553 	/* Turn on DMA, clear stats */
1554 	CSR_WRITE_4(sc, BGE_MAC_MODE, val);
1555 
1556 	/* Set misc. local control, enable interrupts on attentions */
1557 	CSR_WRITE_4(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_ONATTN);
1558 
1559 #ifdef notdef
1560 	/* Assert GPIO pins for PHY reset */
1561 	BNX_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_MISCIO_OUT0|
1562 	    BGE_MLC_MISCIO_OUT1|BGE_MLC_MISCIO_OUT2);
1563 	BNX_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_MISCIO_OUTEN0|
1564 	    BGE_MLC_MISCIO_OUTEN1|BGE_MLC_MISCIO_OUTEN2);
1565 #endif
1566 
1567 	/* Turn on write DMA state machine */
1568 	val = BGE_WDMAMODE_ENABLE|BGE_WDMAMODE_ALL_ATTNS;
1569 	/* Enable host coalescing bug fix. */
1570 	val |= BGE_WDMAMODE_STATUS_TAG_FIX;
1571 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5785) {
1572 		/* Request larger DMA burst size to get better performance. */
1573 		val |= BGE_WDMAMODE_BURST_ALL_DATA;
1574 	}
1575 	CSR_WRITE_4(sc, BGE_WDMA_MODE, val);
1576 	DELAY(40);
1577 
1578 	if (BNX_IS_57765_PLUS(sc)) {
1579 		uint32_t dmactl;
1580 
1581 		dmactl = CSR_READ_4(sc, BGE_RDMA_RSRVCTRL);
1582 		/*
1583 		 * Adjust tx margin to prevent TX data corruption and
1584 		 * fix internal FIFO overflow.
1585 		 */
1586 		if (sc->bnx_asicrev == BGE_ASICREV_BCM5719 ||
1587 		    sc->bnx_asicrev == BGE_ASICREV_BCM5720) {
1588 			dmactl &= ~(BGE_RDMA_RSRVCTRL_FIFO_LWM_MASK |
1589 			    BGE_RDMA_RSRVCTRL_FIFO_HWM_MASK |
1590 			    BGE_RDMA_RSRVCTRL_TXMRGN_MASK);
1591 			dmactl |= BGE_RDMA_RSRVCTRL_FIFO_LWM_1_5K |
1592 			    BGE_RDMA_RSRVCTRL_FIFO_HWM_1_5K |
1593 			    BGE_RDMA_RSRVCTRL_TXMRGN_320B;
1594 		}
1595 		/*
1596 		 * Enable fix for read DMA FIFO overruns.
1597 		 * The fix is to limit the number of RX BDs
1598 		 * the hardware would fetch at a fime.
1599 		 */
1600 		CSR_WRITE_4(sc, BGE_RDMA_RSRVCTRL,
1601 		    dmactl | BGE_RDMA_RSRVCTRL_FIFO_OFLW_FIX);
1602 	}
1603 
1604 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5719) {
1605 		CSR_WRITE_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL,
1606 		    CSR_READ_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL) |
1607 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_BD_4K |
1608 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_LSO_4K);
1609 	} else if (sc->bnx_asicrev == BGE_ASICREV_BCM5720) {
1610 		/*
1611 		 * Allow 4KB burst length reads for non-LSO frames.
1612 		 * Enable 512B burst length reads for buffer descriptors.
1613 		 */
1614 		CSR_WRITE_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL,
1615 		    CSR_READ_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL) |
1616 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_BD_512 |
1617 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_LSO_4K);
1618 	}
1619 
1620 	/* Turn on read DMA state machine */
1621 	val = BGE_RDMAMODE_ENABLE | BGE_RDMAMODE_ALL_ATTNS;
1622 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5717)
1623 		val |= BGE_RDMAMODE_MULT_DMA_RD_DIS;
1624         if (sc->bnx_asicrev == BGE_ASICREV_BCM5784 ||
1625             sc->bnx_asicrev == BGE_ASICREV_BCM5785 ||
1626             sc->bnx_asicrev == BGE_ASICREV_BCM57780) {
1627 		val |= BGE_RDMAMODE_BD_SBD_CRPT_ATTN |
1628 		    BGE_RDMAMODE_MBUF_RBD_CRPT_ATTN |
1629 		    BGE_RDMAMODE_MBUF_SBD_CRPT_ATTN;
1630 	}
1631 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5720) {
1632 		val |= CSR_READ_4(sc, BGE_RDMA_MODE) &
1633 		    BGE_RDMAMODE_H2BNC_VLAN_DET;
1634 		/*
1635 		 * Allow multiple outstanding read requests from
1636 		 * non-LSO read DMA engine.
1637 		 */
1638 		val &= ~BGE_RDMAMODE_MULT_DMA_RD_DIS;
1639 	}
1640 	if (sc->bnx_flags & BNX_FLAG_TSO)
1641 		val |= BGE_RDMAMODE_TSO4_ENABLE;
1642 	val |= BGE_RDMAMODE_FIFO_LONG_BURST;
1643 	CSR_WRITE_4(sc, BGE_RDMA_MODE, val);
1644 	DELAY(40);
1645 
1646 	/* Turn on RX data completion state machine */
1647 	CSR_WRITE_4(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
1648 
1649 	/* Turn on RX BD initiator state machine */
1650 	CSR_WRITE_4(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
1651 
1652 	/* Turn on RX data and RX BD initiator state machine */
1653 	CSR_WRITE_4(sc, BGE_RDBDI_MODE, BGE_RDBDIMODE_ENABLE);
1654 
1655 	/* Turn on send BD completion state machine */
1656 	CSR_WRITE_4(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
1657 
1658 	/* Turn on send data completion state machine */
1659 	val = BGE_SDCMODE_ENABLE;
1660 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5761)
1661 		val |= BGE_SDCMODE_CDELAY;
1662 	CSR_WRITE_4(sc, BGE_SDC_MODE, val);
1663 
1664 	/* Turn on send data initiator state machine */
1665 	if (sc->bnx_flags & BNX_FLAG_TSO) {
1666 		CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE |
1667 		    BGE_SDIMODE_HW_LSO_PRE_DMA);
1668 	} else {
1669 		CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
1670 	}
1671 
1672 	/* Turn on send BD initiator state machine */
1673 	CSR_WRITE_4(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
1674 
1675 	/* Turn on send BD selector state machine */
1676 	CSR_WRITE_4(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
1677 
1678 	CSR_WRITE_4(sc, BGE_SDI_STATS_ENABLE_MASK, 0x007FFFFF);
1679 	CSR_WRITE_4(sc, BGE_SDI_STATS_CTL,
1680 	    BGE_SDISTATSCTL_ENABLE|BGE_SDISTATSCTL_FASTER);
1681 
1682 	/* ack/clear link change events */
1683 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED|
1684 	    BGE_MACSTAT_CFG_CHANGED|BGE_MACSTAT_MI_COMPLETE|
1685 	    BGE_MACSTAT_LINK_CHANGED);
1686 	CSR_WRITE_4(sc, BGE_MI_STS, 0);
1687 
1688 	/*
1689 	 * Enable attention when the link has changed state for
1690 	 * devices that use auto polling.
1691 	 */
1692 	if (sc->bnx_flags & BNX_FLAG_TBI) {
1693 		CSR_WRITE_4(sc, BGE_MI_STS, BGE_MISTS_LINK);
1694  	} else {
1695 		if (sc->bnx_mi_mode & BGE_MIMODE_AUTOPOLL) {
1696 			CSR_WRITE_4(sc, BGE_MI_MODE, sc->bnx_mi_mode);
1697 			DELAY(80);
1698 		}
1699 	}
1700 
1701 	/*
1702 	 * Clear any pending link state attention.
1703 	 * Otherwise some link state change events may be lost until attention
1704 	 * is cleared by bnx_intr() -> bnx_softc.bnx_link_upd() sequence.
1705 	 * It's not necessary on newer BCM chips - perhaps enabling link
1706 	 * state change attentions implies clearing pending attention.
1707 	 */
1708 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED|
1709 	    BGE_MACSTAT_CFG_CHANGED|BGE_MACSTAT_MI_COMPLETE|
1710 	    BGE_MACSTAT_LINK_CHANGED);
1711 
1712 	/* Enable link state change attentions. */
1713 	BNX_SETBIT(sc, BGE_MAC_EVT_ENB, BGE_EVTENB_LINK_CHANGED);
1714 
1715 	return(0);
1716 }
1717 
1718 /*
1719  * Probe for a Broadcom chip. Check the PCI vendor and device IDs
1720  * against our list and return its name if we find a match. Note
1721  * that since the Broadcom controller contains VPD support, we
1722  * can get the device name string from the controller itself instead
1723  * of the compiled-in string. This is a little slow, but it guarantees
1724  * we'll always announce the right product name.
1725  */
1726 static int
1727 bnx_probe(device_t dev)
1728 {
1729 	const struct bnx_type *t;
1730 	uint16_t product, vendor;
1731 
1732 	if (!pci_is_pcie(dev))
1733 		return ENXIO;
1734 
1735 	product = pci_get_device(dev);
1736 	vendor = pci_get_vendor(dev);
1737 
1738 	for (t = bnx_devs; t->bnx_name != NULL; t++) {
1739 		if (vendor == t->bnx_vid && product == t->bnx_did)
1740 			break;
1741 	}
1742 	if (t->bnx_name == NULL)
1743 		return ENXIO;
1744 
1745 	device_set_desc(dev, t->bnx_name);
1746 	return 0;
1747 }
1748 
1749 static int
1750 bnx_attach(device_t dev)
1751 {
1752 	struct ifnet *ifp;
1753 	struct bnx_softc *sc;
1754 	uint32_t hwcfg = 0, misccfg;
1755 	int error = 0, rid, capmask;
1756 	uint8_t ether_addr[ETHER_ADDR_LEN];
1757 	uint16_t product;
1758 	driver_intr_t *intr_func;
1759 	uintptr_t mii_priv = 0;
1760 	u_int intr_flags;
1761 #ifdef BNX_TSO_DEBUG
1762 	char desc[32];
1763 	int i;
1764 #endif
1765 
1766 	sc = device_get_softc(dev);
1767 	sc->bnx_dev = dev;
1768 	callout_init_mp(&sc->bnx_stat_timer);
1769 	callout_init_mp(&sc->bnx_intr_timer);
1770 	lwkt_serialize_init(&sc->bnx_jslot_serializer);
1771 
1772 	product = pci_get_device(dev);
1773 
1774 #ifndef BURN_BRIDGES
1775 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
1776 		uint32_t irq, mem;
1777 
1778 		irq = pci_read_config(dev, PCIR_INTLINE, 4);
1779 		mem = pci_read_config(dev, BGE_PCI_BAR0, 4);
1780 
1781 		device_printf(dev, "chip is in D%d power mode "
1782 		    "-- setting to D0\n", pci_get_powerstate(dev));
1783 
1784 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
1785 
1786 		pci_write_config(dev, PCIR_INTLINE, irq, 4);
1787 		pci_write_config(dev, BGE_PCI_BAR0, mem, 4);
1788 	}
1789 #endif	/* !BURN_BRIDGE */
1790 
1791 	/*
1792 	 * Map control/status registers.
1793 	 */
1794 	pci_enable_busmaster(dev);
1795 
1796 	rid = BGE_PCI_BAR0;
1797 	sc->bnx_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid,
1798 	    RF_ACTIVE);
1799 
1800 	if (sc->bnx_res == NULL) {
1801 		device_printf(dev, "couldn't map memory\n");
1802 		return ENXIO;
1803 	}
1804 
1805 	sc->bnx_btag = rman_get_bustag(sc->bnx_res);
1806 	sc->bnx_bhandle = rman_get_bushandle(sc->bnx_res);
1807 
1808 	/* Save various chip information */
1809 	sc->bnx_chipid =
1810 	    pci_read_config(dev, BGE_PCI_MISC_CTL, 4) >>
1811 	    BGE_PCIMISCCTL_ASICREV_SHIFT;
1812 	if (BGE_ASICREV(sc->bnx_chipid) == BGE_ASICREV_USE_PRODID_REG) {
1813 		/* All chips having dedicated ASICREV register have CPMU */
1814 		sc->bnx_flags |= BNX_FLAG_CPMU;
1815 
1816 		switch (product) {
1817 		case PCI_PRODUCT_BROADCOM_BCM5717:
1818 		case PCI_PRODUCT_BROADCOM_BCM5718:
1819 		case PCI_PRODUCT_BROADCOM_BCM5719:
1820 		case PCI_PRODUCT_BROADCOM_BCM5720_ALT:
1821 			sc->bnx_chipid = pci_read_config(dev,
1822 			    BGE_PCI_GEN2_PRODID_ASICREV, 4);
1823 			break;
1824 
1825 		case PCI_PRODUCT_BROADCOM_BCM57761:
1826 		case PCI_PRODUCT_BROADCOM_BCM57762:
1827 		case PCI_PRODUCT_BROADCOM_BCM57765:
1828 		case PCI_PRODUCT_BROADCOM_BCM57766:
1829 		case PCI_PRODUCT_BROADCOM_BCM57781:
1830 		case PCI_PRODUCT_BROADCOM_BCM57782:
1831 		case PCI_PRODUCT_BROADCOM_BCM57785:
1832 		case PCI_PRODUCT_BROADCOM_BCM57786:
1833 		case PCI_PRODUCT_BROADCOM_BCM57791:
1834 		case PCI_PRODUCT_BROADCOM_BCM57795:
1835 			sc->bnx_chipid = pci_read_config(dev,
1836 			    BGE_PCI_GEN15_PRODID_ASICREV, 4);
1837 			break;
1838 
1839 		default:
1840 			sc->bnx_chipid = pci_read_config(dev,
1841 			    BGE_PCI_PRODID_ASICREV, 4);
1842 			break;
1843 		}
1844 	}
1845 	sc->bnx_asicrev = BGE_ASICREV(sc->bnx_chipid);
1846 	sc->bnx_chiprev = BGE_CHIPREV(sc->bnx_chipid);
1847 
1848 	switch (sc->bnx_asicrev) {
1849 	case BGE_ASICREV_BCM5717:
1850 	case BGE_ASICREV_BCM5719:
1851 	case BGE_ASICREV_BCM5720:
1852 		sc->bnx_flags |= BNX_FLAG_5717_PLUS | BNX_FLAG_57765_PLUS;
1853 		break;
1854 
1855 	case BGE_ASICREV_BCM57765:
1856 	case BGE_ASICREV_BCM57766:
1857 		sc->bnx_flags |= BNX_FLAG_57765_FAMILY | BNX_FLAG_57765_PLUS;
1858 		break;
1859 	}
1860 	sc->bnx_flags |= BNX_FLAG_SHORTDMA;
1861 
1862 	sc->bnx_flags |= BNX_FLAG_TSO;
1863 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5719 &&
1864 	    sc->bnx_chipid == BGE_CHIPID_BCM5719_A0)
1865 		sc->bnx_flags &= ~BNX_FLAG_TSO;
1866 
1867 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5717 ||
1868 	    BNX_IS_57765_FAMILY(sc)) {
1869 		/*
1870 		 * All BCM57785 and BCM5718 families chips have a bug that
1871 		 * under certain situation interrupt will not be enabled
1872 		 * even if status tag is written to BGE_MBX_IRQ0_LO mailbox.
1873 		 *
1874 		 * While BCM5719 and BCM5720 have a hardware workaround
1875 		 * which could fix the above bug.
1876 		 * See the comment near BGE_PCIDMARWCTL_TAGGED_STATUS_WA in
1877 		 * bnx_chipinit().
1878 		 *
1879 		 * For the rest of the chips in these two families, we will
1880 		 * have to poll the status block at high rate (10ms currently)
1881 		 * to check whether the interrupt is hosed or not.
1882 		 * See bnx_intr_check() for details.
1883 		 */
1884 		sc->bnx_flags |= BNX_FLAG_STATUSTAG_BUG;
1885 	}
1886 
1887 	misccfg = CSR_READ_4(sc, BGE_MISC_CFG) & BGE_MISCCFG_BOARD_ID_MASK;
1888 
1889 	sc->bnx_pciecap = pci_get_pciecap_ptr(sc->bnx_dev);
1890 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5719 ||
1891 	    sc->bnx_asicrev == BGE_ASICREV_BCM5720)
1892 		pcie_set_max_readrq(dev, PCIEM_DEVCTL_MAX_READRQ_2048);
1893 	else
1894 		pcie_set_max_readrq(dev, PCIEM_DEVCTL_MAX_READRQ_4096);
1895 	device_printf(dev, "CHIP ID 0x%08x; "
1896 		      "ASIC REV 0x%02x; CHIP REV 0x%02x\n",
1897 		      sc->bnx_chipid, sc->bnx_asicrev, sc->bnx_chiprev);
1898 
1899 	/*
1900 	 * Set various PHY quirk flags.
1901 	 */
1902 
1903 	capmask = MII_CAPMASK_DEFAULT;
1904 	if (product == PCI_PRODUCT_BROADCOM_BCM57791 ||
1905 	    product == PCI_PRODUCT_BROADCOM_BCM57795) {
1906 		/* 10/100 only */
1907 		capmask &= ~BMSR_EXTSTAT;
1908 	}
1909 
1910 	mii_priv |= BRGPHY_FLAG_WIRESPEED;
1911 
1912 	/*
1913 	 * Allocate interrupt
1914 	 */
1915 	sc->bnx_irq_type = pci_alloc_1intr(dev, bnx_msi_enable, &sc->bnx_irq_rid,
1916 	    &intr_flags);
1917 
1918 	sc->bnx_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &sc->bnx_irq_rid,
1919 	    intr_flags);
1920 	if (sc->bnx_irq == NULL) {
1921 		device_printf(dev, "couldn't map interrupt\n");
1922 		error = ENXIO;
1923 		goto fail;
1924 	}
1925 
1926 	if (sc->bnx_irq_type == PCI_INTR_TYPE_MSI) {
1927 		sc->bnx_flags |= BNX_FLAG_ONESHOT_MSI;
1928 		bnx_enable_msi(sc);
1929 	}
1930 
1931 	/* Initialize if_name earlier, so if_printf could be used */
1932 	ifp = &sc->arpcom.ac_if;
1933 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1934 
1935 	/* Try to reset the chip. */
1936 	bnx_reset(sc);
1937 
1938 	if (bnx_chipinit(sc)) {
1939 		device_printf(dev, "chip initialization failed\n");
1940 		error = ENXIO;
1941 		goto fail;
1942 	}
1943 
1944 	/*
1945 	 * Get station address
1946 	 */
1947 	error = bnx_get_eaddr(sc, ether_addr);
1948 	if (error) {
1949 		device_printf(dev, "failed to read station address\n");
1950 		goto fail;
1951 	}
1952 
1953 	if (BNX_IS_57765_PLUS(sc)) {
1954 		sc->bnx_return_ring_cnt = BGE_RETURN_RING_CNT;
1955 	} else {
1956 		/* 5705/5750 limits RX return ring to 512 entries. */
1957 		sc->bnx_return_ring_cnt = BGE_RETURN_RING_CNT_5705;
1958 	}
1959 
1960 	error = bnx_dma_alloc(sc);
1961 	if (error)
1962 		goto fail;
1963 
1964 	/* Set default tuneable values. */
1965 	sc->bnx_rx_coal_ticks = BNX_RX_COAL_TICKS_DEF;
1966 	sc->bnx_tx_coal_ticks = BNX_TX_COAL_TICKS_DEF;
1967 	sc->bnx_rx_coal_bds = BNX_RX_COAL_BDS_DEF;
1968 	sc->bnx_tx_coal_bds = BNX_TX_COAL_BDS_DEF;
1969 	sc->bnx_rx_coal_bds_int = BNX_RX_COAL_BDS_INT_DEF;
1970 	sc->bnx_tx_coal_bds_int = BNX_TX_COAL_BDS_INT_DEF;
1971 	sc->bnx_tx_wreg = 8;
1972 
1973 	/* Set up ifnet structure */
1974 	ifp->if_softc = sc;
1975 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1976 	ifp->if_ioctl = bnx_ioctl;
1977 	ifp->if_start = bnx_start;
1978 #ifdef IFPOLL_ENABLE
1979 	ifp->if_npoll = bnx_npoll;
1980 #endif
1981 	ifp->if_watchdog = bnx_watchdog;
1982 	ifp->if_init = bnx_init;
1983 	ifp->if_mtu = ETHERMTU;
1984 	ifp->if_capabilities = IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU;
1985 	ifq_set_maxlen(&ifp->if_snd, BGE_TX_RING_CNT - 1);
1986 	ifq_set_ready(&ifp->if_snd);
1987 
1988 	ifp->if_capabilities |= IFCAP_HWCSUM;
1989 	ifp->if_hwassist = BNX_CSUM_FEATURES;
1990 	if (sc->bnx_flags & BNX_FLAG_TSO) {
1991 		ifp->if_capabilities |= IFCAP_TSO;
1992 		ifp->if_hwassist |= CSUM_TSO;
1993 	}
1994 	ifp->if_capenable = ifp->if_capabilities;
1995 
1996 	/*
1997 	 * Figure out what sort of media we have by checking the
1998 	 * hardware config word in the first 32k of NIC internal memory,
1999 	 * or fall back to examining the EEPROM if necessary.
2000 	 * Note: on some BCM5700 cards, this value appears to be unset.
2001 	 * If that's the case, we have to rely on identifying the NIC
2002 	 * by its PCI subsystem ID, as we do below for the SysKonnect
2003 	 * SK-9D41.
2004 	 */
2005 	if (bnx_readmem_ind(sc, BGE_SOFTWARE_GENCOMM_SIG) == BGE_MAGIC_NUMBER) {
2006 		hwcfg = bnx_readmem_ind(sc, BGE_SOFTWARE_GENCOMM_NICCFG);
2007 	} else {
2008 		if (bnx_read_eeprom(sc, (caddr_t)&hwcfg, BGE_EE_HWCFG_OFFSET,
2009 				    sizeof(hwcfg))) {
2010 			device_printf(dev, "failed to read EEPROM\n");
2011 			error = ENXIO;
2012 			goto fail;
2013 		}
2014 		hwcfg = ntohl(hwcfg);
2015 	}
2016 
2017 	/* The SysKonnect SK-9D41 is a 1000baseSX card. */
2018 	if (pci_get_subvendor(dev) == PCI_PRODUCT_SCHNEIDERKOCH_SK_9D41 ||
2019 	    (hwcfg & BGE_HWCFG_MEDIA) == BGE_MEDIA_FIBER)
2020 		sc->bnx_flags |= BNX_FLAG_TBI;
2021 
2022 	/* Setup MI MODE */
2023 	if (sc->bnx_flags & BNX_FLAG_CPMU)
2024 		sc->bnx_mi_mode = BGE_MIMODE_500KHZ_CONST;
2025 	else
2026 		sc->bnx_mi_mode = BGE_MIMODE_BASE;
2027 
2028 	/* Setup link status update stuffs */
2029 	if (sc->bnx_flags & BNX_FLAG_TBI) {
2030 		sc->bnx_link_upd = bnx_tbi_link_upd;
2031 		sc->bnx_link_chg = BGE_MACSTAT_LINK_CHANGED;
2032 	} else if (sc->bnx_mi_mode & BGE_MIMODE_AUTOPOLL) {
2033 		sc->bnx_link_upd = bnx_autopoll_link_upd;
2034 		sc->bnx_link_chg = BGE_MACSTAT_LINK_CHANGED;
2035 	} else {
2036 		sc->bnx_link_upd = bnx_copper_link_upd;
2037 		sc->bnx_link_chg = BGE_MACSTAT_LINK_CHANGED;
2038 	}
2039 
2040 	/* Set default PHY address */
2041 	sc->bnx_phyno = 1;
2042 
2043 	/*
2044 	 * PHY address mapping for various devices.
2045 	 *
2046 	 *          | F0 Cu | F0 Sr | F1 Cu | F1 Sr |
2047 	 * ---------+-------+-------+-------+-------+
2048 	 * BCM57XX  |   1   |   X   |   X   |   X   |
2049 	 * BCM5704  |   1   |   X   |   1   |   X   |
2050 	 * BCM5717  |   1   |   8   |   2   |   9   |
2051 	 * BCM5719  |   1   |   8   |   2   |   9   |
2052 	 * BCM5720  |   1   |   8   |   2   |   9   |
2053 	 *
2054 	 * Other addresses may respond but they are not
2055 	 * IEEE compliant PHYs and should be ignored.
2056 	 */
2057 	if (BNX_IS_5717_PLUS(sc)) {
2058 		int f;
2059 
2060 		f = pci_get_function(dev);
2061 		if (sc->bnx_chipid == BGE_CHIPID_BCM5717_A0) {
2062 			if (CSR_READ_4(sc, BGE_SGDIG_STS) &
2063 			    BGE_SGDIGSTS_IS_SERDES)
2064 				sc->bnx_phyno = f + 8;
2065 			else
2066 				sc->bnx_phyno = f + 1;
2067 		} else {
2068 			if (CSR_READ_4(sc, BGE_CPMU_PHY_STRAP) &
2069 			    BGE_CPMU_PHY_STRAP_IS_SERDES)
2070 				sc->bnx_phyno = f + 8;
2071 			else
2072 				sc->bnx_phyno = f + 1;
2073 		}
2074 	}
2075 
2076 	if (sc->bnx_flags & BNX_FLAG_TBI) {
2077 		ifmedia_init(&sc->bnx_ifmedia, IFM_IMASK,
2078 		    bnx_ifmedia_upd, bnx_ifmedia_sts);
2079 		ifmedia_add(&sc->bnx_ifmedia, IFM_ETHER|IFM_1000_SX, 0, NULL);
2080 		ifmedia_add(&sc->bnx_ifmedia,
2081 		    IFM_ETHER|IFM_1000_SX|IFM_FDX, 0, NULL);
2082 		ifmedia_add(&sc->bnx_ifmedia, IFM_ETHER|IFM_AUTO, 0, NULL);
2083 		ifmedia_set(&sc->bnx_ifmedia, IFM_ETHER|IFM_AUTO);
2084 		sc->bnx_ifmedia.ifm_media = sc->bnx_ifmedia.ifm_cur->ifm_media;
2085 	} else {
2086 		struct mii_probe_args mii_args;
2087 
2088 		mii_probe_args_init(&mii_args, bnx_ifmedia_upd, bnx_ifmedia_sts);
2089 		mii_args.mii_probemask = 1 << sc->bnx_phyno;
2090 		mii_args.mii_capmask = capmask;
2091 		mii_args.mii_privtag = MII_PRIVTAG_BRGPHY;
2092 		mii_args.mii_priv = mii_priv;
2093 
2094 		error = mii_probe(dev, &sc->bnx_miibus, &mii_args);
2095 		if (error) {
2096 			device_printf(dev, "MII without any PHY!\n");
2097 			goto fail;
2098 		}
2099 	}
2100 
2101 	/*
2102 	 * Create sysctl nodes.
2103 	 */
2104 	sysctl_ctx_init(&sc->bnx_sysctl_ctx);
2105 	sc->bnx_sysctl_tree = SYSCTL_ADD_NODE(&sc->bnx_sysctl_ctx,
2106 					      SYSCTL_STATIC_CHILDREN(_hw),
2107 					      OID_AUTO,
2108 					      device_get_nameunit(dev),
2109 					      CTLFLAG_RD, 0, "");
2110 	if (sc->bnx_sysctl_tree == NULL) {
2111 		device_printf(dev, "can't add sysctl node\n");
2112 		error = ENXIO;
2113 		goto fail;
2114 	}
2115 
2116 	SYSCTL_ADD_PROC(&sc->bnx_sysctl_ctx,
2117 			SYSCTL_CHILDREN(sc->bnx_sysctl_tree),
2118 			OID_AUTO, "rx_coal_ticks",
2119 			CTLTYPE_INT | CTLFLAG_RW,
2120 			sc, 0, bnx_sysctl_rx_coal_ticks, "I",
2121 			"Receive coalescing ticks (usec).");
2122 	SYSCTL_ADD_PROC(&sc->bnx_sysctl_ctx,
2123 			SYSCTL_CHILDREN(sc->bnx_sysctl_tree),
2124 			OID_AUTO, "tx_coal_ticks",
2125 			CTLTYPE_INT | CTLFLAG_RW,
2126 			sc, 0, bnx_sysctl_tx_coal_ticks, "I",
2127 			"Transmit coalescing ticks (usec).");
2128 	SYSCTL_ADD_PROC(&sc->bnx_sysctl_ctx,
2129 			SYSCTL_CHILDREN(sc->bnx_sysctl_tree),
2130 			OID_AUTO, "rx_coal_bds",
2131 			CTLTYPE_INT | CTLFLAG_RW,
2132 			sc, 0, bnx_sysctl_rx_coal_bds, "I",
2133 			"Receive max coalesced BD count.");
2134 	SYSCTL_ADD_PROC(&sc->bnx_sysctl_ctx,
2135 			SYSCTL_CHILDREN(sc->bnx_sysctl_tree),
2136 			OID_AUTO, "tx_coal_bds",
2137 			CTLTYPE_INT | CTLFLAG_RW,
2138 			sc, 0, bnx_sysctl_tx_coal_bds, "I",
2139 			"Transmit max coalesced BD count.");
2140 	/*
2141 	 * A common design characteristic for many Broadcom
2142 	 * client controllers is that they only support a
2143 	 * single outstanding DMA read operation on the PCIe
2144 	 * bus. This means that it will take twice as long to
2145 	 * fetch a TX frame that is split into header and
2146 	 * payload buffers as it does to fetch a single,
2147 	 * contiguous TX frame (2 reads vs. 1 read). For these
2148 	 * controllers, coalescing buffers to reduce the number
2149 	 * of memory reads is effective way to get maximum
2150 	 * performance(about 940Mbps).  Without collapsing TX
2151 	 * buffers the maximum TCP bulk transfer performance
2152 	 * is about 850Mbps. However forcing coalescing mbufs
2153 	 * consumes a lot of CPU cycles, so leave it off by
2154 	 * default.
2155 	 */
2156 	SYSCTL_ADD_INT(&sc->bnx_sysctl_ctx,
2157 	    SYSCTL_CHILDREN(sc->bnx_sysctl_tree), OID_AUTO,
2158 	    "force_defrag", CTLFLAG_RW, &sc->bnx_force_defrag, 0,
2159 	    "Force defragment on TX path");
2160 
2161 	SYSCTL_ADD_INT(&sc->bnx_sysctl_ctx,
2162 	    SYSCTL_CHILDREN(sc->bnx_sysctl_tree), OID_AUTO,
2163 	    "tx_wreg", CTLFLAG_RW, &sc->bnx_tx_wreg, 0,
2164 	    "# of segments before writing to hardware register");
2165 
2166 	SYSCTL_ADD_PROC(&sc->bnx_sysctl_ctx,
2167 	    SYSCTL_CHILDREN(sc->bnx_sysctl_tree), OID_AUTO,
2168 	    "rx_coal_bds_int", CTLTYPE_INT | CTLFLAG_RW,
2169 	    sc, 0, bnx_sysctl_rx_coal_bds_int, "I",
2170 	    "Receive max coalesced BD count during interrupt.");
2171 	SYSCTL_ADD_PROC(&sc->bnx_sysctl_ctx,
2172 	    SYSCTL_CHILDREN(sc->bnx_sysctl_tree), OID_AUTO,
2173 	    "tx_coal_bds_int", CTLTYPE_INT | CTLFLAG_RW,
2174 	    sc, 0, bnx_sysctl_tx_coal_bds_int, "I",
2175 	    "Transmit max coalesced BD count during interrupt.");
2176 
2177 #ifdef BNX_TSO_DEBUG
2178 	for (i = 0; i < BNX_TSO_NSTATS; ++i) {
2179 		ksnprintf(desc, sizeof(desc), "tso%d", i + 1);
2180 		SYSCTL_ADD_ULONG(&sc->bnx_sysctl_ctx,
2181 		    SYSCTL_CHILDREN(sc->bnx_sysctl_tree), OID_AUTO,
2182 		    desc, CTLFLAG_RW, &sc->bnx_tsosegs[i], "");
2183 	}
2184 #endif
2185 
2186 	/*
2187 	 * Call MI attach routine.
2188 	 */
2189 	ether_ifattach(ifp, ether_addr, NULL);
2190 
2191 #ifdef IFPOLL_ENABLE
2192 	ifpoll_compat_setup(&sc->bnx_npoll,
2193 	    &sc->bnx_sysctl_ctx, sc->bnx_sysctl_tree,
2194 	    device_get_unit(dev), ifp->if_serializer);
2195 #endif
2196 
2197 	if (sc->bnx_irq_type == PCI_INTR_TYPE_MSI) {
2198 		if (sc->bnx_flags & BNX_FLAG_ONESHOT_MSI) {
2199 			intr_func = bnx_msi_oneshot;
2200 			if (bootverbose)
2201 				device_printf(dev, "oneshot MSI\n");
2202 		} else {
2203 			intr_func = bnx_msi;
2204 		}
2205 	} else {
2206 		intr_func = bnx_intr_legacy;
2207 	}
2208 	error = bus_setup_intr(dev, sc->bnx_irq, INTR_MPSAFE, intr_func, sc,
2209 	    &sc->bnx_intrhand, ifp->if_serializer);
2210 	if (error) {
2211 		ether_ifdetach(ifp);
2212 		device_printf(dev, "couldn't set up irq\n");
2213 		goto fail;
2214 	}
2215 
2216 	sc->bnx_intr_cpuid = rman_get_cpuid(sc->bnx_irq);
2217 	sc->bnx_stat_cpuid = sc->bnx_intr_cpuid;
2218 	ifq_set_cpuid(&ifp->if_snd, sc->bnx_intr_cpuid);
2219 
2220 	return(0);
2221 fail:
2222 	bnx_detach(dev);
2223 	return(error);
2224 }
2225 
2226 static int
2227 bnx_detach(device_t dev)
2228 {
2229 	struct bnx_softc *sc = device_get_softc(dev);
2230 
2231 	if (device_is_attached(dev)) {
2232 		struct ifnet *ifp = &sc->arpcom.ac_if;
2233 
2234 		lwkt_serialize_enter(ifp->if_serializer);
2235 		bnx_stop(sc);
2236 		bnx_reset(sc);
2237 		bus_teardown_intr(dev, sc->bnx_irq, sc->bnx_intrhand);
2238 		lwkt_serialize_exit(ifp->if_serializer);
2239 
2240 		ether_ifdetach(ifp);
2241 	}
2242 
2243 	if (sc->bnx_flags & BNX_FLAG_TBI)
2244 		ifmedia_removeall(&sc->bnx_ifmedia);
2245 	if (sc->bnx_miibus)
2246 		device_delete_child(dev, sc->bnx_miibus);
2247 	bus_generic_detach(dev);
2248 
2249 	if (sc->bnx_irq != NULL) {
2250 		bus_release_resource(dev, SYS_RES_IRQ, sc->bnx_irq_rid,
2251 		    sc->bnx_irq);
2252 	}
2253 	if (sc->bnx_irq_type == PCI_INTR_TYPE_MSI)
2254 		pci_release_msi(dev);
2255 
2256 	if (sc->bnx_res != NULL) {
2257 		bus_release_resource(dev, SYS_RES_MEMORY,
2258 		    BGE_PCI_BAR0, sc->bnx_res);
2259 	}
2260 
2261 	if (sc->bnx_sysctl_tree != NULL)
2262 		sysctl_ctx_free(&sc->bnx_sysctl_ctx);
2263 
2264 	bnx_dma_free(sc);
2265 
2266 	return 0;
2267 }
2268 
2269 static void
2270 bnx_reset(struct bnx_softc *sc)
2271 {
2272 	device_t dev;
2273 	uint32_t cachesize, command, pcistate, reset;
2274 	void (*write_op)(struct bnx_softc *, uint32_t, uint32_t);
2275 	int i, val = 0;
2276 	uint16_t devctl;
2277 
2278 	dev = sc->bnx_dev;
2279 
2280 	if (sc->bnx_asicrev != BGE_ASICREV_BCM5906)
2281 		write_op = bnx_writemem_direct;
2282 	else
2283 		write_op = bnx_writereg_ind;
2284 
2285 	/* Save some important PCI state. */
2286 	cachesize = pci_read_config(dev, BGE_PCI_CACHESZ, 4);
2287 	command = pci_read_config(dev, BGE_PCI_CMD, 4);
2288 	pcistate = pci_read_config(dev, BGE_PCI_PCISTATE, 4);
2289 
2290 	pci_write_config(dev, BGE_PCI_MISC_CTL,
2291 	    BGE_PCIMISCCTL_INDIRECT_ACCESS|BGE_PCIMISCCTL_MASK_PCI_INTR|
2292 	    BGE_HIF_SWAP_OPTIONS|BGE_PCIMISCCTL_PCISTATE_RW|
2293 	    BGE_PCIMISCCTL_TAGGED_STATUS, 4);
2294 
2295 	/* Disable fastboot on controllers that support it. */
2296 	if (bootverbose)
2297 		if_printf(&sc->arpcom.ac_if, "Disabling fastboot\n");
2298 	CSR_WRITE_4(sc, BGE_FASTBOOT_PC, 0x0);
2299 
2300 	/*
2301 	 * Write the magic number to SRAM at offset 0xB50.
2302 	 * When firmware finishes its initialization it will
2303 	 * write ~BGE_MAGIC_NUMBER to the same location.
2304 	 */
2305 	bnx_writemem_ind(sc, BGE_SOFTWARE_GENCOMM, BGE_MAGIC_NUMBER);
2306 
2307 	reset = BGE_MISCCFG_RESET_CORE_CLOCKS|(65<<1);
2308 
2309 	/* XXX: Broadcom Linux driver. */
2310 	/* Force PCI-E 1.0a mode */
2311 	if (!BNX_IS_57765_PLUS(sc) &&
2312 	    CSR_READ_4(sc, BGE_PCIE_PHY_TSTCTL) ==
2313 	    (BGE_PCIE_PHY_TSTCTL_PSCRAM |
2314 	     BGE_PCIE_PHY_TSTCTL_PCIE10)) {
2315 		CSR_WRITE_4(sc, BGE_PCIE_PHY_TSTCTL,
2316 		    BGE_PCIE_PHY_TSTCTL_PSCRAM);
2317 	}
2318 	if (sc->bnx_chipid != BGE_CHIPID_BCM5750_A0) {
2319 		/* Prevent PCIE link training during global reset */
2320 		CSR_WRITE_4(sc, BGE_MISC_CFG, (1<<29));
2321 		reset |= (1<<29);
2322 	}
2323 
2324 	/*
2325 	 * Set GPHY Power Down Override to leave GPHY
2326 	 * powered up in D0 uninitialized.
2327 	 */
2328 	if ((sc->bnx_flags & BNX_FLAG_CPMU) == 0)
2329 		reset |= BGE_MISCCFG_GPHY_PD_OVERRIDE;
2330 
2331 	/* Issue global reset */
2332 	write_op(sc, BGE_MISC_CFG, reset);
2333 
2334 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5906) {
2335 		uint32_t status, ctrl;
2336 
2337 		status = CSR_READ_4(sc, BGE_VCPU_STATUS);
2338 		CSR_WRITE_4(sc, BGE_VCPU_STATUS,
2339 		    status | BGE_VCPU_STATUS_DRV_RESET);
2340 		ctrl = CSR_READ_4(sc, BGE_VCPU_EXT_CTRL);
2341 		CSR_WRITE_4(sc, BGE_VCPU_EXT_CTRL,
2342 		    ctrl & ~BGE_VCPU_EXT_CTRL_HALT_CPU);
2343 	}
2344 
2345 	DELAY(1000);
2346 
2347 	/* XXX: Broadcom Linux driver. */
2348 	if (sc->bnx_chipid == BGE_CHIPID_BCM5750_A0) {
2349 		uint32_t v;
2350 
2351 		DELAY(500000); /* wait for link training to complete */
2352 		v = pci_read_config(dev, 0xc4, 4);
2353 		pci_write_config(dev, 0xc4, v | (1<<15), 4);
2354 	}
2355 
2356 	devctl = pci_read_config(dev, sc->bnx_pciecap + PCIER_DEVCTRL, 2);
2357 
2358 	/* Disable no snoop and disable relaxed ordering. */
2359 	devctl &= ~(PCIEM_DEVCTL_RELAX_ORDER | PCIEM_DEVCTL_NOSNOOP);
2360 
2361 	/* Old PCI-E chips only support 128 bytes Max PayLoad Size. */
2362 	if ((sc->bnx_flags & BNX_FLAG_CPMU) == 0) {
2363 		devctl &= ~PCIEM_DEVCTL_MAX_PAYLOAD_MASK;
2364 		devctl |= PCIEM_DEVCTL_MAX_PAYLOAD_128;
2365 	}
2366 
2367 	pci_write_config(dev, sc->bnx_pciecap + PCIER_DEVCTRL,
2368 	    devctl, 2);
2369 
2370 	/* Clear error status. */
2371 	pci_write_config(dev, sc->bnx_pciecap + PCIER_DEVSTS,
2372 	    PCIEM_DEVSTS_CORR_ERR |
2373 	    PCIEM_DEVSTS_NFATAL_ERR |
2374 	    PCIEM_DEVSTS_FATAL_ERR |
2375 	    PCIEM_DEVSTS_UNSUPP_REQ, 2);
2376 
2377 	/* Reset some of the PCI state that got zapped by reset */
2378 	pci_write_config(dev, BGE_PCI_MISC_CTL,
2379 	    BGE_PCIMISCCTL_INDIRECT_ACCESS|BGE_PCIMISCCTL_MASK_PCI_INTR|
2380 	    BGE_HIF_SWAP_OPTIONS|BGE_PCIMISCCTL_PCISTATE_RW|
2381 	    BGE_PCIMISCCTL_TAGGED_STATUS, 4);
2382 	pci_write_config(dev, BGE_PCI_CACHESZ, cachesize, 4);
2383 	pci_write_config(dev, BGE_PCI_CMD, command, 4);
2384 	write_op(sc, BGE_MISC_CFG, (65 << 1));
2385 
2386 	/* Enable memory arbiter */
2387 	CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE);
2388 
2389 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5906) {
2390 		for (i = 0; i < BNX_TIMEOUT; i++) {
2391 			val = CSR_READ_4(sc, BGE_VCPU_STATUS);
2392 			if (val & BGE_VCPU_STATUS_INIT_DONE)
2393 				break;
2394 			DELAY(100);
2395 		}
2396 		if (i == BNX_TIMEOUT) {
2397 			if_printf(&sc->arpcom.ac_if, "reset timed out\n");
2398 			return;
2399 		}
2400 	} else {
2401 		/*
2402 		 * Poll until we see the 1's complement of the magic number.
2403 		 * This indicates that the firmware initialization
2404 		 * is complete.
2405 		 */
2406 		for (i = 0; i < BNX_FIRMWARE_TIMEOUT; i++) {
2407 			val = bnx_readmem_ind(sc, BGE_SOFTWARE_GENCOMM);
2408 			if (val == ~BGE_MAGIC_NUMBER)
2409 				break;
2410 			DELAY(10);
2411 		}
2412 		if (i == BNX_FIRMWARE_TIMEOUT) {
2413 			if_printf(&sc->arpcom.ac_if, "firmware handshake "
2414 				  "timed out, found 0x%08x\n", val);
2415 		}
2416 
2417 		/* BCM57765 A0 needs additional time before accessing. */
2418 		if (sc->bnx_chipid == BGE_CHIPID_BCM57765_A0)
2419 			DELAY(10 * 1000);
2420 	}
2421 
2422 	/*
2423 	 * XXX Wait for the value of the PCISTATE register to
2424 	 * return to its original pre-reset state. This is a
2425 	 * fairly good indicator of reset completion. If we don't
2426 	 * wait for the reset to fully complete, trying to read
2427 	 * from the device's non-PCI registers may yield garbage
2428 	 * results.
2429 	 */
2430 	for (i = 0; i < BNX_TIMEOUT; i++) {
2431 		if (pci_read_config(dev, BGE_PCI_PCISTATE, 4) == pcistate)
2432 			break;
2433 		DELAY(10);
2434 	}
2435 
2436 	/* Fix up byte swapping */
2437 	CSR_WRITE_4(sc, BGE_MODE_CTL, bnx_dma_swap_options(sc));
2438 
2439 	CSR_WRITE_4(sc, BGE_MAC_MODE, 0);
2440 
2441 	/*
2442 	 * The 5704 in TBI mode apparently needs some special
2443 	 * adjustment to insure the SERDES drive level is set
2444 	 * to 1.2V.
2445 	 */
2446 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5704 &&
2447 	    (sc->bnx_flags & BNX_FLAG_TBI)) {
2448 		uint32_t serdescfg;
2449 
2450 		serdescfg = CSR_READ_4(sc, BGE_SERDES_CFG);
2451 		serdescfg = (serdescfg & ~0xFFF) | 0x880;
2452 		CSR_WRITE_4(sc, BGE_SERDES_CFG, serdescfg);
2453 	}
2454 
2455 	CSR_WRITE_4(sc, BGE_MI_MODE,
2456 	    sc->bnx_mi_mode & ~BGE_MIMODE_AUTOPOLL);
2457 	DELAY(80);
2458 
2459 	/* XXX: Broadcom Linux driver. */
2460 	if (!BNX_IS_57765_PLUS(sc)) {
2461 		uint32_t v;
2462 
2463 		/* Enable Data FIFO protection. */
2464 		v = CSR_READ_4(sc, BGE_PCIE_TLDLPL_PORT);
2465 		CSR_WRITE_4(sc, BGE_PCIE_TLDLPL_PORT, v | (1 << 25));
2466 	}
2467 
2468 	DELAY(10000);
2469 
2470 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5720) {
2471 		BNX_CLRBIT(sc, BGE_CPMU_CLCK_ORIDE,
2472 		    CPMU_CLCK_ORIDE_MAC_ORIDE_EN);
2473 	}
2474 }
2475 
2476 /*
2477  * Frame reception handling. This is called if there's a frame
2478  * on the receive return list.
2479  *
2480  * Note: we have to be able to handle two possibilities here:
2481  * 1) the frame is from the jumbo recieve ring
2482  * 2) the frame is from the standard receive ring
2483  */
2484 
2485 static void
2486 bnx_rxeof(struct bnx_softc *sc, uint16_t rx_prod, int count)
2487 {
2488 	struct ifnet *ifp;
2489 	int stdcnt = 0, jumbocnt = 0;
2490 
2491 	ifp = &sc->arpcom.ac_if;
2492 
2493 	while (sc->bnx_rx_saved_considx != rx_prod && count != 0) {
2494 		struct bge_rx_bd	*cur_rx;
2495 		uint32_t		rxidx;
2496 		struct mbuf		*m = NULL;
2497 		uint16_t		vlan_tag = 0;
2498 		int			have_tag = 0;
2499 
2500 		--count;
2501 
2502 		cur_rx =
2503 	    &sc->bnx_ldata.bnx_rx_return_ring[sc->bnx_rx_saved_considx];
2504 
2505 		rxidx = cur_rx->bge_idx;
2506 		BNX_INC(sc->bnx_rx_saved_considx, sc->bnx_return_ring_cnt);
2507 
2508 		if (cur_rx->bge_flags & BGE_RXBDFLAG_VLAN_TAG) {
2509 			have_tag = 1;
2510 			vlan_tag = cur_rx->bge_vlan_tag;
2511 		}
2512 
2513 		if (cur_rx->bge_flags & BGE_RXBDFLAG_JUMBO_RING) {
2514 			BNX_INC(sc->bnx_jumbo, BGE_JUMBO_RX_RING_CNT);
2515 			jumbocnt++;
2516 
2517 			if (rxidx != sc->bnx_jumbo) {
2518 				ifp->if_ierrors++;
2519 				if_printf(ifp, "sw jumbo index(%d) "
2520 				    "and hw jumbo index(%d) mismatch, drop!\n",
2521 				    sc->bnx_jumbo, rxidx);
2522 				bnx_setup_rxdesc_jumbo(sc, rxidx);
2523 				continue;
2524 			}
2525 
2526 			m = sc->bnx_cdata.bnx_rx_jumbo_chain[rxidx].bnx_mbuf;
2527 			if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
2528 				ifp->if_ierrors++;
2529 				bnx_setup_rxdesc_jumbo(sc, sc->bnx_jumbo);
2530 				continue;
2531 			}
2532 			if (bnx_newbuf_jumbo(sc, sc->bnx_jumbo, 0)) {
2533 				ifp->if_ierrors++;
2534 				bnx_setup_rxdesc_jumbo(sc, sc->bnx_jumbo);
2535 				continue;
2536 			}
2537 		} else {
2538 			BNX_INC(sc->bnx_std, BGE_STD_RX_RING_CNT);
2539 			stdcnt++;
2540 
2541 			if (rxidx != sc->bnx_std) {
2542 				ifp->if_ierrors++;
2543 				if_printf(ifp, "sw std index(%d) "
2544 				    "and hw std index(%d) mismatch, drop!\n",
2545 				    sc->bnx_std, rxidx);
2546 				bnx_setup_rxdesc_std(sc, rxidx);
2547 				continue;
2548 			}
2549 
2550 			m = sc->bnx_cdata.bnx_rx_std_chain[rxidx].bnx_mbuf;
2551 			if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
2552 				ifp->if_ierrors++;
2553 				bnx_setup_rxdesc_std(sc, sc->bnx_std);
2554 				continue;
2555 			}
2556 			if (bnx_newbuf_std(sc, sc->bnx_std, 0)) {
2557 				ifp->if_ierrors++;
2558 				bnx_setup_rxdesc_std(sc, sc->bnx_std);
2559 				continue;
2560 			}
2561 		}
2562 
2563 		ifp->if_ipackets++;
2564 		m->m_pkthdr.len = m->m_len = cur_rx->bge_len - ETHER_CRC_LEN;
2565 		m->m_pkthdr.rcvif = ifp;
2566 
2567 		if ((ifp->if_capenable & IFCAP_RXCSUM) &&
2568 		    (cur_rx->bge_flags & BGE_RXBDFLAG_IPV6) == 0) {
2569 			if (cur_rx->bge_flags & BGE_RXBDFLAG_IP_CSUM) {
2570 				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
2571 				if ((cur_rx->bge_error_flag &
2572 				    BGE_RXERRFLAG_IP_CSUM_NOK) == 0)
2573 					m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
2574 			}
2575 			if (cur_rx->bge_flags & BGE_RXBDFLAG_TCP_UDP_CSUM) {
2576 				m->m_pkthdr.csum_data =
2577 				    cur_rx->bge_tcp_udp_csum;
2578 				m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
2579 				    CSUM_PSEUDO_HDR;
2580 			}
2581 		}
2582 
2583 		/*
2584 		 * If we received a packet with a vlan tag, pass it
2585 		 * to vlan_input() instead of ether_input().
2586 		 */
2587 		if (have_tag) {
2588 			m->m_flags |= M_VLANTAG;
2589 			m->m_pkthdr.ether_vlantag = vlan_tag;
2590 		}
2591 		ifp->if_input(ifp, m);
2592 	}
2593 
2594 	bnx_writembx(sc, BGE_MBX_RX_CONS0_LO, sc->bnx_rx_saved_considx);
2595 	if (stdcnt)
2596 		bnx_writembx(sc, BGE_MBX_RX_STD_PROD_LO, sc->bnx_std);
2597 	if (jumbocnt)
2598 		bnx_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, sc->bnx_jumbo);
2599 }
2600 
2601 static void
2602 bnx_txeof(struct bnx_softc *sc, uint16_t tx_cons)
2603 {
2604 	struct ifnet *ifp;
2605 
2606 	ifp = &sc->arpcom.ac_if;
2607 
2608 	/*
2609 	 * Go through our tx ring and free mbufs for those
2610 	 * frames that have been sent.
2611 	 */
2612 	while (sc->bnx_tx_saved_considx != tx_cons) {
2613 		uint32_t idx = 0;
2614 
2615 		idx = sc->bnx_tx_saved_considx;
2616 		if (sc->bnx_cdata.bnx_tx_chain[idx] != NULL) {
2617 			ifp->if_opackets++;
2618 			bus_dmamap_unload(sc->bnx_cdata.bnx_tx_mtag,
2619 			    sc->bnx_cdata.bnx_tx_dmamap[idx]);
2620 			m_freem(sc->bnx_cdata.bnx_tx_chain[idx]);
2621 			sc->bnx_cdata.bnx_tx_chain[idx] = NULL;
2622 		}
2623 		sc->bnx_txcnt--;
2624 		BNX_INC(sc->bnx_tx_saved_considx, BGE_TX_RING_CNT);
2625 	}
2626 
2627 	if ((BGE_TX_RING_CNT - sc->bnx_txcnt) >=
2628 	    (BNX_NSEG_RSVD + BNX_NSEG_SPARE))
2629 		ifq_clr_oactive(&ifp->if_snd);
2630 
2631 	if (sc->bnx_txcnt == 0)
2632 		ifp->if_timer = 0;
2633 
2634 	if (!ifq_is_empty(&ifp->if_snd))
2635 		if_devstart(ifp);
2636 }
2637 
2638 #ifdef IFPOLL_ENABLE
2639 
2640 static void
2641 bnx_npoll(struct ifnet *ifp, struct ifpoll_info *info)
2642 {
2643 	struct bnx_softc *sc = ifp->if_softc;
2644 
2645 	ASSERT_SERIALIZED(ifp->if_serializer);
2646 
2647 	if (info != NULL) {
2648 		int cpuid = sc->bnx_npoll.ifpc_cpuid;
2649 
2650 		info->ifpi_rx[cpuid].poll_func = bnx_npoll_compat;
2651 		info->ifpi_rx[cpuid].arg = NULL;
2652 		info->ifpi_rx[cpuid].serializer = ifp->if_serializer;
2653 
2654 		if (ifp->if_flags & IFF_RUNNING)
2655 			bnx_disable_intr(sc);
2656 		ifq_set_cpuid(&ifp->if_snd, cpuid);
2657 	} else {
2658 		if (ifp->if_flags & IFF_RUNNING)
2659 			bnx_enable_intr(sc);
2660 		ifq_set_cpuid(&ifp->if_snd, sc->bnx_intr_cpuid);
2661 	}
2662 }
2663 
2664 static void
2665 bnx_npoll_compat(struct ifnet *ifp, void *arg __unused, int cycle)
2666 {
2667 	struct bnx_softc *sc = ifp->if_softc;
2668 	struct bge_status_block *sblk = sc->bnx_ldata.bnx_status_block;
2669 	uint16_t rx_prod, tx_cons;
2670 
2671 	ASSERT_SERIALIZED(ifp->if_serializer);
2672 
2673 	if (sc->bnx_npoll.ifpc_stcount-- == 0) {
2674 		sc->bnx_npoll.ifpc_stcount = sc->bnx_npoll.ifpc_stfrac;
2675 		/*
2676 		 * Process link state changes.
2677 		 */
2678 		bnx_link_poll(sc);
2679 	}
2680 
2681 	sc->bnx_status_tag = sblk->bge_status_tag;
2682 
2683 	/*
2684 	 * Use a load fence to ensure that status_tag is saved
2685 	 * before rx_prod and tx_cons.
2686 	 */
2687 	cpu_lfence();
2688 
2689 	rx_prod = sblk->bge_idx[0].bge_rx_prod_idx;
2690 	tx_cons = sblk->bge_idx[0].bge_tx_cons_idx;
2691 
2692 	if (sc->bnx_rx_saved_considx != rx_prod)
2693 		bnx_rxeof(sc, rx_prod, cycle);
2694 
2695 	if (sc->bnx_tx_saved_considx != tx_cons)
2696 		bnx_txeof(sc, tx_cons);
2697 
2698 	if (sc->bnx_coal_chg)
2699 		bnx_coal_change(sc);
2700 }
2701 
2702 #endif	/* IFPOLL_ENABLE */
2703 
2704 static void
2705 bnx_intr_legacy(void *xsc)
2706 {
2707 	struct bnx_softc *sc = xsc;
2708 	struct bge_status_block *sblk = sc->bnx_ldata.bnx_status_block;
2709 
2710 	if (sc->bnx_status_tag == sblk->bge_status_tag) {
2711 		uint32_t val;
2712 
2713 		val = pci_read_config(sc->bnx_dev, BGE_PCI_PCISTATE, 4);
2714 		if (val & BGE_PCISTAT_INTR_NOTACT)
2715 			return;
2716 	}
2717 
2718 	/*
2719 	 * NOTE:
2720 	 * Interrupt will have to be disabled if tagged status
2721 	 * is used, else interrupt will always be asserted on
2722 	 * certain chips (at least on BCM5750 AX/BX).
2723 	 */
2724 	bnx_writembx(sc, BGE_MBX_IRQ0_LO, 1);
2725 
2726 	bnx_intr(sc);
2727 }
2728 
2729 static void
2730 bnx_msi(void *xsc)
2731 {
2732 	struct bnx_softc *sc = xsc;
2733 
2734 	/* Disable interrupt first */
2735 	bnx_writembx(sc, BGE_MBX_IRQ0_LO, 1);
2736 	bnx_intr(sc);
2737 }
2738 
2739 static void
2740 bnx_msi_oneshot(void *xsc)
2741 {
2742 	bnx_intr(xsc);
2743 }
2744 
2745 static void
2746 bnx_intr(struct bnx_softc *sc)
2747 {
2748 	struct ifnet *ifp = &sc->arpcom.ac_if;
2749 	struct bge_status_block *sblk = sc->bnx_ldata.bnx_status_block;
2750 	uint16_t rx_prod, tx_cons;
2751 	uint32_t status;
2752 
2753 	sc->bnx_status_tag = sblk->bge_status_tag;
2754 	/*
2755 	 * Use a load fence to ensure that status_tag is saved
2756 	 * before rx_prod, tx_cons and status.
2757 	 */
2758 	cpu_lfence();
2759 
2760 	rx_prod = sblk->bge_idx[0].bge_rx_prod_idx;
2761 	tx_cons = sblk->bge_idx[0].bge_tx_cons_idx;
2762 	status = sblk->bge_status;
2763 
2764 	if ((status & BGE_STATFLAG_LINKSTATE_CHANGED) || sc->bnx_link_evt)
2765 		bnx_link_poll(sc);
2766 
2767 	if (ifp->if_flags & IFF_RUNNING) {
2768 		if (sc->bnx_rx_saved_considx != rx_prod)
2769 			bnx_rxeof(sc, rx_prod, -1);
2770 
2771 		if (sc->bnx_tx_saved_considx != tx_cons)
2772 			bnx_txeof(sc, tx_cons);
2773 	}
2774 
2775 	bnx_writembx(sc, BGE_MBX_IRQ0_LO, sc->bnx_status_tag << 24);
2776 
2777 	if (sc->bnx_coal_chg)
2778 		bnx_coal_change(sc);
2779 }
2780 
2781 static void
2782 bnx_tick(void *xsc)
2783 {
2784 	struct bnx_softc *sc = xsc;
2785 	struct ifnet *ifp = &sc->arpcom.ac_if;
2786 
2787 	lwkt_serialize_enter(ifp->if_serializer);
2788 
2789 	KKASSERT(mycpuid == sc->bnx_stat_cpuid);
2790 
2791 	bnx_stats_update_regs(sc);
2792 
2793 	if (sc->bnx_flags & BNX_FLAG_TBI) {
2794 		/*
2795 		 * Since in TBI mode auto-polling can't be used we should poll
2796 		 * link status manually. Here we register pending link event
2797 		 * and trigger interrupt.
2798 		 */
2799 		sc->bnx_link_evt++;
2800 		BNX_SETBIT(sc, BGE_HCC_MODE, BGE_HCCMODE_COAL_NOW);
2801 	} else if (!sc->bnx_link) {
2802 		mii_tick(device_get_softc(sc->bnx_miibus));
2803 	}
2804 
2805 	callout_reset(&sc->bnx_stat_timer, hz, bnx_tick, sc);
2806 
2807 	lwkt_serialize_exit(ifp->if_serializer);
2808 }
2809 
2810 static void
2811 bnx_stats_update_regs(struct bnx_softc *sc)
2812 {
2813 	struct ifnet *ifp = &sc->arpcom.ac_if;
2814 	struct bge_mac_stats_regs stats;
2815 	uint32_t *s;
2816 	int i;
2817 
2818 	s = (uint32_t *)&stats;
2819 	for (i = 0; i < sizeof(struct bge_mac_stats_regs); i += 4) {
2820 		*s = CSR_READ_4(sc, BGE_RX_STATS + i);
2821 		s++;
2822 	}
2823 
2824 	ifp->if_collisions +=
2825 	   (stats.dot3StatsSingleCollisionFrames +
2826 	   stats.dot3StatsMultipleCollisionFrames +
2827 	   stats.dot3StatsExcessiveCollisions +
2828 	   stats.dot3StatsLateCollisions) -
2829 	   ifp->if_collisions;
2830 }
2831 
2832 /*
2833  * Encapsulate an mbuf chain in the tx ring  by coupling the mbuf data
2834  * pointers to descriptors.
2835  */
2836 static int
2837 bnx_encap(struct bnx_softc *sc, struct mbuf **m_head0, uint32_t *txidx,
2838     int *segs_used)
2839 {
2840 	struct bge_tx_bd *d = NULL;
2841 	uint16_t csum_flags = 0, vlan_tag = 0, mss = 0;
2842 	bus_dma_segment_t segs[BNX_NSEG_NEW];
2843 	bus_dmamap_t map;
2844 	int error, maxsegs, nsegs, idx, i;
2845 	struct mbuf *m_head = *m_head0, *m_new;
2846 
2847 	if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
2848 #ifdef BNX_TSO_DEBUG
2849 		int tso_nsegs;
2850 #endif
2851 
2852 		error = bnx_setup_tso(sc, m_head0, &mss, &csum_flags);
2853 		if (error)
2854 			return error;
2855 		m_head = *m_head0;
2856 
2857 #ifdef BNX_TSO_DEBUG
2858 		tso_nsegs = (m_head->m_pkthdr.len /
2859 		    m_head->m_pkthdr.tso_segsz) - 1;
2860 		if (tso_nsegs > (BNX_TSO_NSTATS - 1))
2861 			tso_nsegs = BNX_TSO_NSTATS - 1;
2862 		else if (tso_nsegs < 0)
2863 			tso_nsegs = 0;
2864 		sc->bnx_tsosegs[tso_nsegs]++;
2865 #endif
2866 	} else if (m_head->m_pkthdr.csum_flags & BNX_CSUM_FEATURES) {
2867 		if (m_head->m_pkthdr.csum_flags & CSUM_IP)
2868 			csum_flags |= BGE_TXBDFLAG_IP_CSUM;
2869 		if (m_head->m_pkthdr.csum_flags & (CSUM_TCP | CSUM_UDP))
2870 			csum_flags |= BGE_TXBDFLAG_TCP_UDP_CSUM;
2871 		if (m_head->m_flags & M_LASTFRAG)
2872 			csum_flags |= BGE_TXBDFLAG_IP_FRAG_END;
2873 		else if (m_head->m_flags & M_FRAG)
2874 			csum_flags |= BGE_TXBDFLAG_IP_FRAG;
2875 	}
2876 	if (m_head->m_flags & M_VLANTAG) {
2877 		csum_flags |= BGE_TXBDFLAG_VLAN_TAG;
2878 		vlan_tag = m_head->m_pkthdr.ether_vlantag;
2879 	}
2880 
2881 	idx = *txidx;
2882 	map = sc->bnx_cdata.bnx_tx_dmamap[idx];
2883 
2884 	maxsegs = (BGE_TX_RING_CNT - sc->bnx_txcnt) - BNX_NSEG_RSVD;
2885 	KASSERT(maxsegs >= BNX_NSEG_SPARE,
2886 		("not enough segments %d", maxsegs));
2887 
2888 	if (maxsegs > BNX_NSEG_NEW)
2889 		maxsegs = BNX_NSEG_NEW;
2890 
2891 	/*
2892 	 * Pad outbound frame to BGE_MIN_FRAMELEN for an unusual reason.
2893 	 * The bge hardware will pad out Tx runts to BGE_MIN_FRAMELEN,
2894 	 * but when such padded frames employ the bge IP/TCP checksum
2895 	 * offload, the hardware checksum assist gives incorrect results
2896 	 * (possibly from incorporating its own padding into the UDP/TCP
2897 	 * checksum; who knows).  If we pad such runts with zeros, the
2898 	 * onboard checksum comes out correct.
2899 	 */
2900 	if ((csum_flags & BGE_TXBDFLAG_TCP_UDP_CSUM) &&
2901 	    m_head->m_pkthdr.len < BNX_MIN_FRAMELEN) {
2902 		error = m_devpad(m_head, BNX_MIN_FRAMELEN);
2903 		if (error)
2904 			goto back;
2905 	}
2906 
2907 	if ((sc->bnx_flags & BNX_FLAG_SHORTDMA) && m_head->m_next != NULL) {
2908 		m_new = bnx_defrag_shortdma(m_head);
2909 		if (m_new == NULL) {
2910 			error = ENOBUFS;
2911 			goto back;
2912 		}
2913 		*m_head0 = m_head = m_new;
2914 	}
2915 	if ((m_head->m_pkthdr.csum_flags & CSUM_TSO) == 0 &&
2916 	    sc->bnx_force_defrag && m_head->m_next != NULL) {
2917 		/*
2918 		 * Forcefully defragment mbuf chain to overcome hardware
2919 		 * limitation which only support a single outstanding
2920 		 * DMA read operation.  If it fails, keep moving on using
2921 		 * the original mbuf chain.
2922 		 */
2923 		m_new = m_defrag(m_head, MB_DONTWAIT);
2924 		if (m_new != NULL)
2925 			*m_head0 = m_head = m_new;
2926 	}
2927 
2928 	error = bus_dmamap_load_mbuf_defrag(sc->bnx_cdata.bnx_tx_mtag, map,
2929 			m_head0, segs, maxsegs, &nsegs, BUS_DMA_NOWAIT);
2930 	if (error)
2931 		goto back;
2932 	*segs_used += nsegs;
2933 
2934 	m_head = *m_head0;
2935 	bus_dmamap_sync(sc->bnx_cdata.bnx_tx_mtag, map, BUS_DMASYNC_PREWRITE);
2936 
2937 	for (i = 0; ; i++) {
2938 		d = &sc->bnx_ldata.bnx_tx_ring[idx];
2939 
2940 		d->bge_addr.bge_addr_lo = BGE_ADDR_LO(segs[i].ds_addr);
2941 		d->bge_addr.bge_addr_hi = BGE_ADDR_HI(segs[i].ds_addr);
2942 		d->bge_len = segs[i].ds_len;
2943 		d->bge_flags = csum_flags;
2944 		d->bge_vlan_tag = vlan_tag;
2945 		d->bge_mss = mss;
2946 
2947 		if (i == nsegs - 1)
2948 			break;
2949 		BNX_INC(idx, BGE_TX_RING_CNT);
2950 	}
2951 	/* Mark the last segment as end of packet... */
2952 	d->bge_flags |= BGE_TXBDFLAG_END;
2953 
2954 	/*
2955 	 * Insure that the map for this transmission is placed at
2956 	 * the array index of the last descriptor in this chain.
2957 	 */
2958 	sc->bnx_cdata.bnx_tx_dmamap[*txidx] = sc->bnx_cdata.bnx_tx_dmamap[idx];
2959 	sc->bnx_cdata.bnx_tx_dmamap[idx] = map;
2960 	sc->bnx_cdata.bnx_tx_chain[idx] = m_head;
2961 	sc->bnx_txcnt += nsegs;
2962 
2963 	BNX_INC(idx, BGE_TX_RING_CNT);
2964 	*txidx = idx;
2965 back:
2966 	if (error) {
2967 		m_freem(*m_head0);
2968 		*m_head0 = NULL;
2969 	}
2970 	return error;
2971 }
2972 
2973 /*
2974  * Main transmit routine. To avoid having to do mbuf copies, we put pointers
2975  * to the mbuf data regions directly in the transmit descriptors.
2976  */
2977 static void
2978 bnx_start(struct ifnet *ifp, struct ifaltq_subque *ifsq)
2979 {
2980 	struct bnx_softc *sc = ifp->if_softc;
2981 	struct mbuf *m_head = NULL;
2982 	uint32_t prodidx;
2983 	int nsegs = 0;
2984 
2985 	ASSERT_ALTQ_SQ_DEFAULT(ifp, ifsq);
2986 
2987 	if ((ifp->if_flags & IFF_RUNNING) == 0 || ifq_is_oactive(&ifp->if_snd))
2988 		return;
2989 
2990 	prodidx = sc->bnx_tx_prodidx;
2991 
2992 	while (sc->bnx_cdata.bnx_tx_chain[prodidx] == NULL) {
2993 		/*
2994 		 * Sanity check: avoid coming within BGE_NSEG_RSVD
2995 		 * descriptors of the end of the ring.  Also make
2996 		 * sure there are BGE_NSEG_SPARE descriptors for
2997 		 * jumbo buffers' or TSO segments' defragmentation.
2998 		 */
2999 		if ((BGE_TX_RING_CNT - sc->bnx_txcnt) <
3000 		    (BNX_NSEG_RSVD + BNX_NSEG_SPARE)) {
3001 			ifq_set_oactive(&ifp->if_snd);
3002 			break;
3003 		}
3004 
3005 		m_head = ifq_dequeue(&ifp->if_snd, NULL);
3006 		if (m_head == NULL)
3007 			break;
3008 
3009 		/*
3010 		 * Pack the data into the transmit ring. If we
3011 		 * don't have room, set the OACTIVE flag and wait
3012 		 * for the NIC to drain the ring.
3013 		 */
3014 		if (bnx_encap(sc, &m_head, &prodidx, &nsegs)) {
3015 			ifq_set_oactive(&ifp->if_snd);
3016 			ifp->if_oerrors++;
3017 			break;
3018 		}
3019 
3020 		if (nsegs >= sc->bnx_tx_wreg) {
3021 			/* Transmit */
3022 			bnx_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
3023 			nsegs = 0;
3024 		}
3025 
3026 		ETHER_BPF_MTAP(ifp, m_head);
3027 
3028 		/*
3029 		 * Set a timeout in case the chip goes out to lunch.
3030 		 */
3031 		ifp->if_timer = 5;
3032 	}
3033 
3034 	if (nsegs > 0) {
3035 		/* Transmit */
3036 		bnx_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
3037 	}
3038 	sc->bnx_tx_prodidx = prodidx;
3039 }
3040 
3041 static void
3042 bnx_init(void *xsc)
3043 {
3044 	struct bnx_softc *sc = xsc;
3045 	struct ifnet *ifp = &sc->arpcom.ac_if;
3046 	uint16_t *m;
3047 	uint32_t mode;
3048 
3049 	ASSERT_SERIALIZED(ifp->if_serializer);
3050 
3051 	/* Cancel pending I/O and flush buffers. */
3052 	bnx_stop(sc);
3053 	bnx_reset(sc);
3054 	bnx_chipinit(sc);
3055 
3056 	/*
3057 	 * Init the various state machines, ring
3058 	 * control blocks and firmware.
3059 	 */
3060 	if (bnx_blockinit(sc)) {
3061 		if_printf(ifp, "initialization failure\n");
3062 		bnx_stop(sc);
3063 		return;
3064 	}
3065 
3066 	/* Specify MTU. */
3067 	CSR_WRITE_4(sc, BGE_RX_MTU, ifp->if_mtu +
3068 	    ETHER_HDR_LEN + ETHER_CRC_LEN + EVL_ENCAPLEN);
3069 
3070 	/* Load our MAC address. */
3071 	m = (uint16_t *)&sc->arpcom.ac_enaddr[0];
3072 	CSR_WRITE_4(sc, BGE_MAC_ADDR1_LO, htons(m[0]));
3073 	CSR_WRITE_4(sc, BGE_MAC_ADDR1_HI, (htons(m[1]) << 16) | htons(m[2]));
3074 
3075 	/* Enable or disable promiscuous mode as needed. */
3076 	bnx_setpromisc(sc);
3077 
3078 	/* Program multicast filter. */
3079 	bnx_setmulti(sc);
3080 
3081 	/* Init RX ring. */
3082 	if (bnx_init_rx_ring_std(sc)) {
3083 		if_printf(ifp, "RX ring initialization failed\n");
3084 		bnx_stop(sc);
3085 		return;
3086 	}
3087 
3088 	/* Init jumbo RX ring. */
3089 	if (ifp->if_mtu > (ETHERMTU + ETHER_HDR_LEN + ETHER_CRC_LEN)) {
3090 		if (bnx_init_rx_ring_jumbo(sc)) {
3091 			if_printf(ifp, "Jumbo RX ring initialization failed\n");
3092 			bnx_stop(sc);
3093 			return;
3094 		}
3095 	}
3096 
3097 	/* Init our RX return ring index */
3098 	sc->bnx_rx_saved_considx = 0;
3099 
3100 	/* Init TX ring. */
3101 	bnx_init_tx_ring(sc);
3102 
3103 	/* Enable TX MAC state machine lockup fix. */
3104 	mode = CSR_READ_4(sc, BGE_TX_MODE);
3105 	mode |= BGE_TXMODE_MBUF_LOCKUP_FIX;
3106 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5720) {
3107 		mode &= ~(BGE_TXMODE_JMB_FRM_LEN | BGE_TXMODE_CNT_DN_MODE);
3108 		mode |= CSR_READ_4(sc, BGE_TX_MODE) &
3109 		    (BGE_TXMODE_JMB_FRM_LEN | BGE_TXMODE_CNT_DN_MODE);
3110 	}
3111 	/* Turn on transmitter */
3112 	CSR_WRITE_4(sc, BGE_TX_MODE, mode | BGE_TXMODE_ENABLE);
3113 
3114 	/* Turn on receiver */
3115 	BNX_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_ENABLE);
3116 
3117 	/*
3118 	 * Set the number of good frames to receive after RX MBUF
3119 	 * Low Watermark has been reached.  After the RX MAC receives
3120 	 * this number of frames, it will drop subsequent incoming
3121 	 * frames until the MBUF High Watermark is reached.
3122 	 */
3123 	if (BNX_IS_57765_FAMILY(sc))
3124 		CSR_WRITE_4(sc, BGE_MAX_RX_FRAME_LOWAT, 1);
3125 	else
3126 		CSR_WRITE_4(sc, BGE_MAX_RX_FRAME_LOWAT, 2);
3127 
3128 	if (sc->bnx_irq_type == PCI_INTR_TYPE_MSI) {
3129 		if (bootverbose) {
3130 			if_printf(ifp, "MSI_MODE: %#x\n",
3131 			    CSR_READ_4(sc, BGE_MSI_MODE));
3132 		}
3133 	}
3134 
3135 	/* Tell firmware we're alive. */
3136 	BNX_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
3137 
3138 	/* Enable host interrupts if polling(4) is not enabled. */
3139 	PCI_SETBIT(sc->bnx_dev, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_CLEAR_INTA, 4);
3140 #ifdef IFPOLL_ENABLE
3141 	if (ifp->if_flags & IFF_NPOLLING)
3142 		bnx_disable_intr(sc);
3143 	else
3144 #endif
3145 	bnx_enable_intr(sc);
3146 
3147 	bnx_ifmedia_upd(ifp);
3148 
3149 	ifp->if_flags |= IFF_RUNNING;
3150 	ifq_clr_oactive(&ifp->if_snd);
3151 
3152 	callout_reset_bycpu(&sc->bnx_stat_timer, hz, bnx_tick, sc,
3153 	    sc->bnx_stat_cpuid);
3154 }
3155 
3156 /*
3157  * Set media options.
3158  */
3159 static int
3160 bnx_ifmedia_upd(struct ifnet *ifp)
3161 {
3162 	struct bnx_softc *sc = ifp->if_softc;
3163 
3164 	/* If this is a 1000baseX NIC, enable the TBI port. */
3165 	if (sc->bnx_flags & BNX_FLAG_TBI) {
3166 		struct ifmedia *ifm = &sc->bnx_ifmedia;
3167 
3168 		if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
3169 			return(EINVAL);
3170 
3171 		switch(IFM_SUBTYPE(ifm->ifm_media)) {
3172 		case IFM_AUTO:
3173 			break;
3174 
3175 		case IFM_1000_SX:
3176 			if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX) {
3177 				BNX_CLRBIT(sc, BGE_MAC_MODE,
3178 				    BGE_MACMODE_HALF_DUPLEX);
3179 			} else {
3180 				BNX_SETBIT(sc, BGE_MAC_MODE,
3181 				    BGE_MACMODE_HALF_DUPLEX);
3182 			}
3183 			break;
3184 		default:
3185 			return(EINVAL);
3186 		}
3187 	} else {
3188 		struct mii_data *mii = device_get_softc(sc->bnx_miibus);
3189 
3190 		sc->bnx_link_evt++;
3191 		sc->bnx_link = 0;
3192 		if (mii->mii_instance) {
3193 			struct mii_softc *miisc;
3194 
3195 			LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
3196 				mii_phy_reset(miisc);
3197 		}
3198 		mii_mediachg(mii);
3199 
3200 		/*
3201 		 * Force an interrupt so that we will call bnx_link_upd
3202 		 * if needed and clear any pending link state attention.
3203 		 * Without this we are not getting any further interrupts
3204 		 * for link state changes and thus will not UP the link and
3205 		 * not be able to send in bnx_start.  The only way to get
3206 		 * things working was to receive a packet and get an RX
3207 		 * intr.
3208 		 *
3209 		 * bnx_tick should help for fiber cards and we might not
3210 		 * need to do this here if BNX_FLAG_TBI is set but as
3211 		 * we poll for fiber anyway it should not harm.
3212 		 */
3213 		BNX_SETBIT(sc, BGE_HCC_MODE, BGE_HCCMODE_COAL_NOW);
3214 	}
3215 	return(0);
3216 }
3217 
3218 /*
3219  * Report current media status.
3220  */
3221 static void
3222 bnx_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3223 {
3224 	struct bnx_softc *sc = ifp->if_softc;
3225 
3226 	if (sc->bnx_flags & BNX_FLAG_TBI) {
3227 		ifmr->ifm_status = IFM_AVALID;
3228 		ifmr->ifm_active = IFM_ETHER;
3229 		if (CSR_READ_4(sc, BGE_MAC_STS) &
3230 		    BGE_MACSTAT_TBI_PCS_SYNCHED) {
3231 			ifmr->ifm_status |= IFM_ACTIVE;
3232 		} else {
3233 			ifmr->ifm_active |= IFM_NONE;
3234 			return;
3235 		}
3236 
3237 		ifmr->ifm_active |= IFM_1000_SX;
3238 		if (CSR_READ_4(sc, BGE_MAC_MODE) & BGE_MACMODE_HALF_DUPLEX)
3239 			ifmr->ifm_active |= IFM_HDX;
3240 		else
3241 			ifmr->ifm_active |= IFM_FDX;
3242 	} else {
3243 		struct mii_data *mii = device_get_softc(sc->bnx_miibus);
3244 
3245 		mii_pollstat(mii);
3246 		ifmr->ifm_active = mii->mii_media_active;
3247 		ifmr->ifm_status = mii->mii_media_status;
3248 	}
3249 }
3250 
3251 static int
3252 bnx_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
3253 {
3254 	struct bnx_softc *sc = ifp->if_softc;
3255 	struct ifreq *ifr = (struct ifreq *)data;
3256 	int mask, error = 0;
3257 
3258 	ASSERT_SERIALIZED(ifp->if_serializer);
3259 
3260 	switch (command) {
3261 	case SIOCSIFMTU:
3262 		if ((!BNX_IS_JUMBO_CAPABLE(sc) && ifr->ifr_mtu > ETHERMTU) ||
3263 		    (BNX_IS_JUMBO_CAPABLE(sc) &&
3264 		     ifr->ifr_mtu > BNX_JUMBO_MTU)) {
3265 			error = EINVAL;
3266 		} else if (ifp->if_mtu != ifr->ifr_mtu) {
3267 			ifp->if_mtu = ifr->ifr_mtu;
3268 			if (ifp->if_flags & IFF_RUNNING)
3269 				bnx_init(sc);
3270 		}
3271 		break;
3272 	case SIOCSIFFLAGS:
3273 		if (ifp->if_flags & IFF_UP) {
3274 			if (ifp->if_flags & IFF_RUNNING) {
3275 				mask = ifp->if_flags ^ sc->bnx_if_flags;
3276 
3277 				/*
3278 				 * If only the state of the PROMISC flag
3279 				 * changed, then just use the 'set promisc
3280 				 * mode' command instead of reinitializing
3281 				 * the entire NIC. Doing a full re-init
3282 				 * means reloading the firmware and waiting
3283 				 * for it to start up, which may take a
3284 				 * second or two.  Similarly for ALLMULTI.
3285 				 */
3286 				if (mask & IFF_PROMISC)
3287 					bnx_setpromisc(sc);
3288 				if (mask & IFF_ALLMULTI)
3289 					bnx_setmulti(sc);
3290 			} else {
3291 				bnx_init(sc);
3292 			}
3293 		} else if (ifp->if_flags & IFF_RUNNING) {
3294 			bnx_stop(sc);
3295 		}
3296 		sc->bnx_if_flags = ifp->if_flags;
3297 		break;
3298 	case SIOCADDMULTI:
3299 	case SIOCDELMULTI:
3300 		if (ifp->if_flags & IFF_RUNNING)
3301 			bnx_setmulti(sc);
3302 		break;
3303 	case SIOCSIFMEDIA:
3304 	case SIOCGIFMEDIA:
3305 		if (sc->bnx_flags & BNX_FLAG_TBI) {
3306 			error = ifmedia_ioctl(ifp, ifr,
3307 			    &sc->bnx_ifmedia, command);
3308 		} else {
3309 			struct mii_data *mii;
3310 
3311 			mii = device_get_softc(sc->bnx_miibus);
3312 			error = ifmedia_ioctl(ifp, ifr,
3313 					      &mii->mii_media, command);
3314 		}
3315 		break;
3316         case SIOCSIFCAP:
3317 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
3318 		if (mask & IFCAP_HWCSUM) {
3319 			ifp->if_capenable ^= (mask & IFCAP_HWCSUM);
3320 			if (ifp->if_capenable & IFCAP_TXCSUM)
3321 				ifp->if_hwassist |= BNX_CSUM_FEATURES;
3322 			else
3323 				ifp->if_hwassist &= ~BNX_CSUM_FEATURES;
3324 		}
3325 		if (mask & IFCAP_TSO) {
3326 			ifp->if_capenable ^= (mask & IFCAP_TSO);
3327 			if (ifp->if_capenable & IFCAP_TSO)
3328 				ifp->if_hwassist |= CSUM_TSO;
3329 			else
3330 				ifp->if_hwassist &= ~CSUM_TSO;
3331 		}
3332 		break;
3333 	default:
3334 		error = ether_ioctl(ifp, command, data);
3335 		break;
3336 	}
3337 	return error;
3338 }
3339 
3340 static void
3341 bnx_watchdog(struct ifnet *ifp)
3342 {
3343 	struct bnx_softc *sc = ifp->if_softc;
3344 
3345 	if_printf(ifp, "watchdog timeout -- resetting\n");
3346 
3347 	bnx_init(sc);
3348 
3349 	ifp->if_oerrors++;
3350 
3351 	if (!ifq_is_empty(&ifp->if_snd))
3352 		if_devstart(ifp);
3353 }
3354 
3355 /*
3356  * Stop the adapter and free any mbufs allocated to the
3357  * RX and TX lists.
3358  */
3359 static void
3360 bnx_stop(struct bnx_softc *sc)
3361 {
3362 	struct ifnet *ifp = &sc->arpcom.ac_if;
3363 
3364 	ASSERT_SERIALIZED(ifp->if_serializer);
3365 
3366 	callout_stop(&sc->bnx_stat_timer);
3367 
3368 	/*
3369 	 * Disable all of the receiver blocks
3370 	 */
3371 	bnx_stop_block(sc, BGE_RX_MODE, BGE_RXMODE_ENABLE);
3372 	bnx_stop_block(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
3373 	bnx_stop_block(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
3374 	bnx_stop_block(sc, BGE_RDBDI_MODE, BGE_RBDIMODE_ENABLE);
3375 	bnx_stop_block(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
3376 	bnx_stop_block(sc, BGE_RBDC_MODE, BGE_RBDCMODE_ENABLE);
3377 
3378 	/*
3379 	 * Disable all of the transmit blocks
3380 	 */
3381 	bnx_stop_block(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
3382 	bnx_stop_block(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
3383 	bnx_stop_block(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
3384 	bnx_stop_block(sc, BGE_RDMA_MODE, BGE_RDMAMODE_ENABLE);
3385 	bnx_stop_block(sc, BGE_SDC_MODE, BGE_SDCMODE_ENABLE);
3386 	bnx_stop_block(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
3387 
3388 	/*
3389 	 * Shut down all of the memory managers and related
3390 	 * state machines.
3391 	 */
3392 	bnx_stop_block(sc, BGE_HCC_MODE, BGE_HCCMODE_ENABLE);
3393 	bnx_stop_block(sc, BGE_WDMA_MODE, BGE_WDMAMODE_ENABLE);
3394 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
3395 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
3396 
3397 	/* Disable host interrupts. */
3398 	bnx_disable_intr(sc);
3399 
3400 	/*
3401 	 * Tell firmware we're shutting down.
3402 	 */
3403 	BNX_CLRBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
3404 
3405 	/* Free the RX lists. */
3406 	bnx_free_rx_ring_std(sc);
3407 
3408 	/* Free jumbo RX list. */
3409 	if (BNX_IS_JUMBO_CAPABLE(sc))
3410 		bnx_free_rx_ring_jumbo(sc);
3411 
3412 	/* Free TX buffers. */
3413 	bnx_free_tx_ring(sc);
3414 
3415 	sc->bnx_status_tag = 0;
3416 	sc->bnx_link = 0;
3417 	sc->bnx_coal_chg = 0;
3418 
3419 	sc->bnx_tx_saved_considx = BNX_TXCONS_UNSET;
3420 
3421 	ifp->if_flags &= ~IFF_RUNNING;
3422 	ifq_clr_oactive(&ifp->if_snd);
3423 	ifp->if_timer = 0;
3424 }
3425 
3426 /*
3427  * Stop all chip I/O so that the kernel's probe routines don't
3428  * get confused by errant DMAs when rebooting.
3429  */
3430 static void
3431 bnx_shutdown(device_t dev)
3432 {
3433 	struct bnx_softc *sc = device_get_softc(dev);
3434 	struct ifnet *ifp = &sc->arpcom.ac_if;
3435 
3436 	lwkt_serialize_enter(ifp->if_serializer);
3437 	bnx_stop(sc);
3438 	bnx_reset(sc);
3439 	lwkt_serialize_exit(ifp->if_serializer);
3440 }
3441 
3442 static int
3443 bnx_suspend(device_t dev)
3444 {
3445 	struct bnx_softc *sc = device_get_softc(dev);
3446 	struct ifnet *ifp = &sc->arpcom.ac_if;
3447 
3448 	lwkt_serialize_enter(ifp->if_serializer);
3449 	bnx_stop(sc);
3450 	lwkt_serialize_exit(ifp->if_serializer);
3451 
3452 	return 0;
3453 }
3454 
3455 static int
3456 bnx_resume(device_t dev)
3457 {
3458 	struct bnx_softc *sc = device_get_softc(dev);
3459 	struct ifnet *ifp = &sc->arpcom.ac_if;
3460 
3461 	lwkt_serialize_enter(ifp->if_serializer);
3462 
3463 	if (ifp->if_flags & IFF_UP) {
3464 		bnx_init(sc);
3465 
3466 		if (!ifq_is_empty(&ifp->if_snd))
3467 			if_devstart(ifp);
3468 	}
3469 
3470 	lwkt_serialize_exit(ifp->if_serializer);
3471 
3472 	return 0;
3473 }
3474 
3475 static void
3476 bnx_setpromisc(struct bnx_softc *sc)
3477 {
3478 	struct ifnet *ifp = &sc->arpcom.ac_if;
3479 
3480 	if (ifp->if_flags & IFF_PROMISC)
3481 		BNX_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
3482 	else
3483 		BNX_CLRBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
3484 }
3485 
3486 static void
3487 bnx_dma_free(struct bnx_softc *sc)
3488 {
3489 	int i;
3490 
3491 	/* Destroy RX mbuf DMA stuffs. */
3492 	if (sc->bnx_cdata.bnx_rx_mtag != NULL) {
3493 		for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
3494 			bus_dmamap_destroy(sc->bnx_cdata.bnx_rx_mtag,
3495 			    sc->bnx_cdata.bnx_rx_std_dmamap[i]);
3496 		}
3497 		bus_dmamap_destroy(sc->bnx_cdata.bnx_rx_mtag,
3498 				   sc->bnx_cdata.bnx_rx_tmpmap);
3499 		bus_dma_tag_destroy(sc->bnx_cdata.bnx_rx_mtag);
3500 	}
3501 
3502 	/* Destroy TX mbuf DMA stuffs. */
3503 	if (sc->bnx_cdata.bnx_tx_mtag != NULL) {
3504 		for (i = 0; i < BGE_TX_RING_CNT; i++) {
3505 			bus_dmamap_destroy(sc->bnx_cdata.bnx_tx_mtag,
3506 			    sc->bnx_cdata.bnx_tx_dmamap[i]);
3507 		}
3508 		bus_dma_tag_destroy(sc->bnx_cdata.bnx_tx_mtag);
3509 	}
3510 
3511 	/* Destroy standard RX ring */
3512 	bnx_dma_block_free(sc->bnx_cdata.bnx_rx_std_ring_tag,
3513 			   sc->bnx_cdata.bnx_rx_std_ring_map,
3514 			   sc->bnx_ldata.bnx_rx_std_ring);
3515 
3516 	if (BNX_IS_JUMBO_CAPABLE(sc))
3517 		bnx_free_jumbo_mem(sc);
3518 
3519 	/* Destroy RX return ring */
3520 	bnx_dma_block_free(sc->bnx_cdata.bnx_rx_return_ring_tag,
3521 			   sc->bnx_cdata.bnx_rx_return_ring_map,
3522 			   sc->bnx_ldata.bnx_rx_return_ring);
3523 
3524 	/* Destroy TX ring */
3525 	bnx_dma_block_free(sc->bnx_cdata.bnx_tx_ring_tag,
3526 			   sc->bnx_cdata.bnx_tx_ring_map,
3527 			   sc->bnx_ldata.bnx_tx_ring);
3528 
3529 	/* Destroy status block */
3530 	bnx_dma_block_free(sc->bnx_cdata.bnx_status_tag,
3531 			   sc->bnx_cdata.bnx_status_map,
3532 			   sc->bnx_ldata.bnx_status_block);
3533 
3534 	/* Destroy the parent tag */
3535 	if (sc->bnx_cdata.bnx_parent_tag != NULL)
3536 		bus_dma_tag_destroy(sc->bnx_cdata.bnx_parent_tag);
3537 }
3538 
3539 static int
3540 bnx_dma_alloc(struct bnx_softc *sc)
3541 {
3542 	struct ifnet *ifp = &sc->arpcom.ac_if;
3543 	bus_size_t txmaxsz;
3544 	int i, error;
3545 
3546 	/*
3547 	 * Allocate the parent bus DMA tag appropriate for PCI.
3548 	 *
3549 	 * All of the NetExtreme/NetLink controllers have 4GB boundary
3550 	 * DMA bug.
3551 	 * Whenever an address crosses a multiple of the 4GB boundary
3552 	 * (including 4GB, 8Gb, 12Gb, etc.) and makes the transition
3553 	 * from 0xX_FFFF_FFFF to 0x(X+1)_0000_0000 an internal DMA
3554 	 * state machine will lockup and cause the device to hang.
3555 	 */
3556 	error = bus_dma_tag_create(NULL, 1, BGE_DMA_BOUNDARY_4G,
3557 				   BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
3558 				   NULL, NULL,
3559 				   BUS_SPACE_MAXSIZE_32BIT, 0,
3560 				   BUS_SPACE_MAXSIZE_32BIT,
3561 				   0, &sc->bnx_cdata.bnx_parent_tag);
3562 	if (error) {
3563 		if_printf(ifp, "could not allocate parent dma tag\n");
3564 		return error;
3565 	}
3566 
3567 	/*
3568 	 * Create DMA tag and maps for RX mbufs.
3569 	 */
3570 	error = bus_dma_tag_create(sc->bnx_cdata.bnx_parent_tag, 1, 0,
3571 				   BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
3572 				   NULL, NULL, MCLBYTES, 1, MCLBYTES,
3573 				   BUS_DMA_ALLOCNOW | BUS_DMA_WAITOK,
3574 				   &sc->bnx_cdata.bnx_rx_mtag);
3575 	if (error) {
3576 		if_printf(ifp, "could not allocate RX mbuf dma tag\n");
3577 		return error;
3578 	}
3579 
3580 	error = bus_dmamap_create(sc->bnx_cdata.bnx_rx_mtag,
3581 				  BUS_DMA_WAITOK, &sc->bnx_cdata.bnx_rx_tmpmap);
3582 	if (error) {
3583 		bus_dma_tag_destroy(sc->bnx_cdata.bnx_rx_mtag);
3584 		sc->bnx_cdata.bnx_rx_mtag = NULL;
3585 		return error;
3586 	}
3587 
3588 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
3589 		error = bus_dmamap_create(sc->bnx_cdata.bnx_rx_mtag,
3590 					  BUS_DMA_WAITOK,
3591 					  &sc->bnx_cdata.bnx_rx_std_dmamap[i]);
3592 		if (error) {
3593 			int j;
3594 
3595 			for (j = 0; j < i; ++j) {
3596 				bus_dmamap_destroy(sc->bnx_cdata.bnx_rx_mtag,
3597 					sc->bnx_cdata.bnx_rx_std_dmamap[j]);
3598 			}
3599 			bus_dma_tag_destroy(sc->bnx_cdata.bnx_rx_mtag);
3600 			sc->bnx_cdata.bnx_rx_mtag = NULL;
3601 
3602 			if_printf(ifp, "could not create DMA map for RX\n");
3603 			return error;
3604 		}
3605 	}
3606 
3607 	/*
3608 	 * Create DMA tag and maps for TX mbufs.
3609 	 */
3610 	if (sc->bnx_flags & BNX_FLAG_TSO)
3611 		txmaxsz = IP_MAXPACKET + sizeof(struct ether_vlan_header);
3612 	else
3613 		txmaxsz = BNX_JUMBO_FRAMELEN;
3614 	error = bus_dma_tag_create(sc->bnx_cdata.bnx_parent_tag, 1, 0,
3615 				   BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
3616 				   NULL, NULL,
3617 				   txmaxsz, BNX_NSEG_NEW, PAGE_SIZE,
3618 				   BUS_DMA_ALLOCNOW | BUS_DMA_WAITOK |
3619 				   BUS_DMA_ONEBPAGE,
3620 				   &sc->bnx_cdata.bnx_tx_mtag);
3621 	if (error) {
3622 		if_printf(ifp, "could not allocate TX mbuf dma tag\n");
3623 		return error;
3624 	}
3625 
3626 	for (i = 0; i < BGE_TX_RING_CNT; i++) {
3627 		error = bus_dmamap_create(sc->bnx_cdata.bnx_tx_mtag,
3628 					  BUS_DMA_WAITOK | BUS_DMA_ONEBPAGE,
3629 					  &sc->bnx_cdata.bnx_tx_dmamap[i]);
3630 		if (error) {
3631 			int j;
3632 
3633 			for (j = 0; j < i; ++j) {
3634 				bus_dmamap_destroy(sc->bnx_cdata.bnx_tx_mtag,
3635 					sc->bnx_cdata.bnx_tx_dmamap[j]);
3636 			}
3637 			bus_dma_tag_destroy(sc->bnx_cdata.bnx_tx_mtag);
3638 			sc->bnx_cdata.bnx_tx_mtag = NULL;
3639 
3640 			if_printf(ifp, "could not create DMA map for TX\n");
3641 			return error;
3642 		}
3643 	}
3644 
3645 	/*
3646 	 * Create DMA stuffs for standard RX ring.
3647 	 */
3648 	error = bnx_dma_block_alloc(sc, BGE_STD_RX_RING_SZ,
3649 				    &sc->bnx_cdata.bnx_rx_std_ring_tag,
3650 				    &sc->bnx_cdata.bnx_rx_std_ring_map,
3651 				    (void *)&sc->bnx_ldata.bnx_rx_std_ring,
3652 				    &sc->bnx_ldata.bnx_rx_std_ring_paddr);
3653 	if (error) {
3654 		if_printf(ifp, "could not create std RX ring\n");
3655 		return error;
3656 	}
3657 
3658 	/*
3659 	 * Create jumbo buffer pool.
3660 	 */
3661 	if (BNX_IS_JUMBO_CAPABLE(sc)) {
3662 		error = bnx_alloc_jumbo_mem(sc);
3663 		if (error) {
3664 			if_printf(ifp, "could not create jumbo buffer pool\n");
3665 			return error;
3666 		}
3667 	}
3668 
3669 	/*
3670 	 * Create DMA stuffs for RX return ring.
3671 	 */
3672 	error = bnx_dma_block_alloc(sc,
3673 	    BGE_RX_RTN_RING_SZ(sc->bnx_return_ring_cnt),
3674 	    &sc->bnx_cdata.bnx_rx_return_ring_tag,
3675 	    &sc->bnx_cdata.bnx_rx_return_ring_map,
3676 	    (void *)&sc->bnx_ldata.bnx_rx_return_ring,
3677 	    &sc->bnx_ldata.bnx_rx_return_ring_paddr);
3678 	if (error) {
3679 		if_printf(ifp, "could not create RX ret ring\n");
3680 		return error;
3681 	}
3682 
3683 	/*
3684 	 * Create DMA stuffs for TX ring.
3685 	 */
3686 	error = bnx_dma_block_alloc(sc, BGE_TX_RING_SZ,
3687 				    &sc->bnx_cdata.bnx_tx_ring_tag,
3688 				    &sc->bnx_cdata.bnx_tx_ring_map,
3689 				    (void *)&sc->bnx_ldata.bnx_tx_ring,
3690 				    &sc->bnx_ldata.bnx_tx_ring_paddr);
3691 	if (error) {
3692 		if_printf(ifp, "could not create TX ring\n");
3693 		return error;
3694 	}
3695 
3696 	/*
3697 	 * Create DMA stuffs for status block.
3698 	 */
3699 	error = bnx_dma_block_alloc(sc, BGE_STATUS_BLK_SZ,
3700 				    &sc->bnx_cdata.bnx_status_tag,
3701 				    &sc->bnx_cdata.bnx_status_map,
3702 				    (void *)&sc->bnx_ldata.bnx_status_block,
3703 				    &sc->bnx_ldata.bnx_status_block_paddr);
3704 	if (error) {
3705 		if_printf(ifp, "could not create status block\n");
3706 		return error;
3707 	}
3708 
3709 	return 0;
3710 }
3711 
3712 static int
3713 bnx_dma_block_alloc(struct bnx_softc *sc, bus_size_t size, bus_dma_tag_t *tag,
3714 		    bus_dmamap_t *map, void **addr, bus_addr_t *paddr)
3715 {
3716 	bus_dmamem_t dmem;
3717 	int error;
3718 
3719 	error = bus_dmamem_coherent(sc->bnx_cdata.bnx_parent_tag, PAGE_SIZE, 0,
3720 				    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
3721 				    size, BUS_DMA_WAITOK | BUS_DMA_ZERO, &dmem);
3722 	if (error)
3723 		return error;
3724 
3725 	*tag = dmem.dmem_tag;
3726 	*map = dmem.dmem_map;
3727 	*addr = dmem.dmem_addr;
3728 	*paddr = dmem.dmem_busaddr;
3729 
3730 	return 0;
3731 }
3732 
3733 static void
3734 bnx_dma_block_free(bus_dma_tag_t tag, bus_dmamap_t map, void *addr)
3735 {
3736 	if (tag != NULL) {
3737 		bus_dmamap_unload(tag, map);
3738 		bus_dmamem_free(tag, addr, map);
3739 		bus_dma_tag_destroy(tag);
3740 	}
3741 }
3742 
3743 static void
3744 bnx_tbi_link_upd(struct bnx_softc *sc, uint32_t status)
3745 {
3746 	struct ifnet *ifp = &sc->arpcom.ac_if;
3747 
3748 #define PCS_ENCODE_ERR	(BGE_MACSTAT_PORT_DECODE_ERROR|BGE_MACSTAT_MI_COMPLETE)
3749 
3750 	/*
3751 	 * Sometimes PCS encoding errors are detected in
3752 	 * TBI mode (on fiber NICs), and for some reason
3753 	 * the chip will signal them as link changes.
3754 	 * If we get a link change event, but the 'PCS
3755 	 * encoding error' bit in the MAC status register
3756 	 * is set, don't bother doing a link check.
3757 	 * This avoids spurious "gigabit link up" messages
3758 	 * that sometimes appear on fiber NICs during
3759 	 * periods of heavy traffic.
3760 	 */
3761 	if (status & BGE_MACSTAT_TBI_PCS_SYNCHED) {
3762 		if (!sc->bnx_link) {
3763 			sc->bnx_link++;
3764 			if (sc->bnx_asicrev == BGE_ASICREV_BCM5704) {
3765 				BNX_CLRBIT(sc, BGE_MAC_MODE,
3766 				    BGE_MACMODE_TBI_SEND_CFGS);
3767 			}
3768 			CSR_WRITE_4(sc, BGE_MAC_STS, 0xFFFFFFFF);
3769 
3770 			if (bootverbose)
3771 				if_printf(ifp, "link UP\n");
3772 
3773 			ifp->if_link_state = LINK_STATE_UP;
3774 			if_link_state_change(ifp);
3775 		}
3776 	} else if ((status & PCS_ENCODE_ERR) != PCS_ENCODE_ERR) {
3777 		if (sc->bnx_link) {
3778 			sc->bnx_link = 0;
3779 
3780 			if (bootverbose)
3781 				if_printf(ifp, "link DOWN\n");
3782 
3783 			ifp->if_link_state = LINK_STATE_DOWN;
3784 			if_link_state_change(ifp);
3785 		}
3786 	}
3787 
3788 #undef PCS_ENCODE_ERR
3789 
3790 	/* Clear the attention. */
3791 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
3792 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
3793 	    BGE_MACSTAT_LINK_CHANGED);
3794 }
3795 
3796 static void
3797 bnx_copper_link_upd(struct bnx_softc *sc, uint32_t status __unused)
3798 {
3799 	struct ifnet *ifp = &sc->arpcom.ac_if;
3800 	struct mii_data *mii = device_get_softc(sc->bnx_miibus);
3801 
3802 	mii_pollstat(mii);
3803 	bnx_miibus_statchg(sc->bnx_dev);
3804 
3805 	if (bootverbose) {
3806 		if (sc->bnx_link)
3807 			if_printf(ifp, "link UP\n");
3808 		else
3809 			if_printf(ifp, "link DOWN\n");
3810 	}
3811 
3812 	/* Clear the attention. */
3813 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
3814 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
3815 	    BGE_MACSTAT_LINK_CHANGED);
3816 }
3817 
3818 static void
3819 bnx_autopoll_link_upd(struct bnx_softc *sc, uint32_t status __unused)
3820 {
3821 	struct ifnet *ifp = &sc->arpcom.ac_if;
3822 	struct mii_data *mii = device_get_softc(sc->bnx_miibus);
3823 
3824 	mii_pollstat(mii);
3825 
3826 	if (!sc->bnx_link &&
3827 	    (mii->mii_media_status & IFM_ACTIVE) &&
3828 	    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
3829 		sc->bnx_link++;
3830 		if (bootverbose)
3831 			if_printf(ifp, "link UP\n");
3832 	} else if (sc->bnx_link &&
3833 	    (!(mii->mii_media_status & IFM_ACTIVE) ||
3834 	    IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE)) {
3835 		sc->bnx_link = 0;
3836 		if (bootverbose)
3837 			if_printf(ifp, "link DOWN\n");
3838 	}
3839 
3840 	/* Clear the attention. */
3841 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
3842 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
3843 	    BGE_MACSTAT_LINK_CHANGED);
3844 }
3845 
3846 static int
3847 bnx_sysctl_rx_coal_ticks(SYSCTL_HANDLER_ARGS)
3848 {
3849 	struct bnx_softc *sc = arg1;
3850 
3851 	return bnx_sysctl_coal_chg(oidp, arg1, arg2, req,
3852 	    &sc->bnx_rx_coal_ticks,
3853 	    BNX_RX_COAL_TICKS_MIN, BNX_RX_COAL_TICKS_MAX,
3854 	    BNX_RX_COAL_TICKS_CHG);
3855 }
3856 
3857 static int
3858 bnx_sysctl_tx_coal_ticks(SYSCTL_HANDLER_ARGS)
3859 {
3860 	struct bnx_softc *sc = arg1;
3861 
3862 	return bnx_sysctl_coal_chg(oidp, arg1, arg2, req,
3863 	    &sc->bnx_tx_coal_ticks,
3864 	    BNX_TX_COAL_TICKS_MIN, BNX_TX_COAL_TICKS_MAX,
3865 	    BNX_TX_COAL_TICKS_CHG);
3866 }
3867 
3868 static int
3869 bnx_sysctl_rx_coal_bds(SYSCTL_HANDLER_ARGS)
3870 {
3871 	struct bnx_softc *sc = arg1;
3872 
3873 	return bnx_sysctl_coal_chg(oidp, arg1, arg2, req,
3874 	    &sc->bnx_rx_coal_bds,
3875 	    BNX_RX_COAL_BDS_MIN, BNX_RX_COAL_BDS_MAX,
3876 	    BNX_RX_COAL_BDS_CHG);
3877 }
3878 
3879 static int
3880 bnx_sysctl_tx_coal_bds(SYSCTL_HANDLER_ARGS)
3881 {
3882 	struct bnx_softc *sc = arg1;
3883 
3884 	return bnx_sysctl_coal_chg(oidp, arg1, arg2, req,
3885 	    &sc->bnx_tx_coal_bds,
3886 	    BNX_TX_COAL_BDS_MIN, BNX_TX_COAL_BDS_MAX,
3887 	    BNX_TX_COAL_BDS_CHG);
3888 }
3889 
3890 static int
3891 bnx_sysctl_rx_coal_bds_int(SYSCTL_HANDLER_ARGS)
3892 {
3893 	struct bnx_softc *sc = arg1;
3894 
3895 	return bnx_sysctl_coal_chg(oidp, arg1, arg2, req,
3896 	    &sc->bnx_rx_coal_bds_int,
3897 	    BNX_RX_COAL_BDS_MIN, BNX_RX_COAL_BDS_MAX,
3898 	    BNX_RX_COAL_BDS_INT_CHG);
3899 }
3900 
3901 static int
3902 bnx_sysctl_tx_coal_bds_int(SYSCTL_HANDLER_ARGS)
3903 {
3904 	struct bnx_softc *sc = arg1;
3905 
3906 	return bnx_sysctl_coal_chg(oidp, arg1, arg2, req,
3907 	    &sc->bnx_tx_coal_bds_int,
3908 	    BNX_TX_COAL_BDS_MIN, BNX_TX_COAL_BDS_MAX,
3909 	    BNX_TX_COAL_BDS_INT_CHG);
3910 }
3911 
3912 static int
3913 bnx_sysctl_coal_chg(SYSCTL_HANDLER_ARGS, uint32_t *coal,
3914     int coal_min, int coal_max, uint32_t coal_chg_mask)
3915 {
3916 	struct bnx_softc *sc = arg1;
3917 	struct ifnet *ifp = &sc->arpcom.ac_if;
3918 	int error = 0, v;
3919 
3920 	lwkt_serialize_enter(ifp->if_serializer);
3921 
3922 	v = *coal;
3923 	error = sysctl_handle_int(oidp, &v, 0, req);
3924 	if (!error && req->newptr != NULL) {
3925 		if (v < coal_min || v > coal_max) {
3926 			error = EINVAL;
3927 		} else {
3928 			*coal = v;
3929 			sc->bnx_coal_chg |= coal_chg_mask;
3930 		}
3931 	}
3932 
3933 	lwkt_serialize_exit(ifp->if_serializer);
3934 	return error;
3935 }
3936 
3937 static void
3938 bnx_coal_change(struct bnx_softc *sc)
3939 {
3940 	struct ifnet *ifp = &sc->arpcom.ac_if;
3941 	uint32_t val;
3942 
3943 	ASSERT_SERIALIZED(ifp->if_serializer);
3944 
3945 	if (sc->bnx_coal_chg & BNX_RX_COAL_TICKS_CHG) {
3946 		CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS,
3947 			    sc->bnx_rx_coal_ticks);
3948 		DELAY(10);
3949 		val = CSR_READ_4(sc, BGE_HCC_RX_COAL_TICKS);
3950 
3951 		if (bootverbose) {
3952 			if_printf(ifp, "rx_coal_ticks -> %u\n",
3953 				  sc->bnx_rx_coal_ticks);
3954 		}
3955 	}
3956 
3957 	if (sc->bnx_coal_chg & BNX_TX_COAL_TICKS_CHG) {
3958 		CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS,
3959 			    sc->bnx_tx_coal_ticks);
3960 		DELAY(10);
3961 		val = CSR_READ_4(sc, BGE_HCC_TX_COAL_TICKS);
3962 
3963 		if (bootverbose) {
3964 			if_printf(ifp, "tx_coal_ticks -> %u\n",
3965 				  sc->bnx_tx_coal_ticks);
3966 		}
3967 	}
3968 
3969 	if (sc->bnx_coal_chg & BNX_RX_COAL_BDS_CHG) {
3970 		CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS,
3971 			    sc->bnx_rx_coal_bds);
3972 		DELAY(10);
3973 		val = CSR_READ_4(sc, BGE_HCC_RX_MAX_COAL_BDS);
3974 
3975 		if (bootverbose) {
3976 			if_printf(ifp, "rx_coal_bds -> %u\n",
3977 				  sc->bnx_rx_coal_bds);
3978 		}
3979 	}
3980 
3981 	if (sc->bnx_coal_chg & BNX_TX_COAL_BDS_CHG) {
3982 		CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS,
3983 			    sc->bnx_tx_coal_bds);
3984 		DELAY(10);
3985 		val = CSR_READ_4(sc, BGE_HCC_TX_MAX_COAL_BDS);
3986 
3987 		if (bootverbose) {
3988 			if_printf(ifp, "tx_coal_bds -> %u\n",
3989 				  sc->bnx_tx_coal_bds);
3990 		}
3991 	}
3992 
3993 	if (sc->bnx_coal_chg & BNX_RX_COAL_BDS_INT_CHG) {
3994 		CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT,
3995 		    sc->bnx_rx_coal_bds_int);
3996 		DELAY(10);
3997 		val = CSR_READ_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT);
3998 
3999 		if (bootverbose) {
4000 			if_printf(ifp, "rx_coal_bds_int -> %u\n",
4001 			    sc->bnx_rx_coal_bds_int);
4002 		}
4003 	}
4004 
4005 	if (sc->bnx_coal_chg & BNX_TX_COAL_BDS_INT_CHG) {
4006 		CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT,
4007 		    sc->bnx_tx_coal_bds_int);
4008 		DELAY(10);
4009 		val = CSR_READ_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT);
4010 
4011 		if (bootverbose) {
4012 			if_printf(ifp, "tx_coal_bds_int -> %u\n",
4013 			    sc->bnx_tx_coal_bds_int);
4014 		}
4015 	}
4016 
4017 	sc->bnx_coal_chg = 0;
4018 }
4019 
4020 static void
4021 bnx_intr_check(void *xsc)
4022 {
4023 	struct bnx_softc *sc = xsc;
4024 	struct ifnet *ifp = &sc->arpcom.ac_if;
4025 	struct bge_status_block *sblk = sc->bnx_ldata.bnx_status_block;
4026 
4027 	lwkt_serialize_enter(ifp->if_serializer);
4028 
4029 	KKASSERT(mycpuid == sc->bnx_intr_cpuid);
4030 
4031 	if ((ifp->if_flags & (IFF_RUNNING | IFF_NPOLLING)) != IFF_RUNNING) {
4032 		lwkt_serialize_exit(ifp->if_serializer);
4033 		return;
4034 	}
4035 
4036 	if (sblk->bge_idx[0].bge_rx_prod_idx != sc->bnx_rx_saved_considx ||
4037 	    sblk->bge_idx[0].bge_tx_cons_idx != sc->bnx_tx_saved_considx) {
4038 		if (sc->bnx_rx_check_considx == sc->bnx_rx_saved_considx &&
4039 		    sc->bnx_tx_check_considx == sc->bnx_tx_saved_considx) {
4040 			if (!sc->bnx_intr_maylose) {
4041 				sc->bnx_intr_maylose = TRUE;
4042 				goto done;
4043 			}
4044 			if (bootverbose)
4045 				if_printf(ifp, "lost interrupt\n");
4046 			bnx_msi(sc);
4047 		}
4048 	}
4049 	sc->bnx_intr_maylose = FALSE;
4050 	sc->bnx_rx_check_considx = sc->bnx_rx_saved_considx;
4051 	sc->bnx_tx_check_considx = sc->bnx_tx_saved_considx;
4052 
4053 done:
4054 	callout_reset(&sc->bnx_intr_timer, BNX_INTR_CKINTVL,
4055 	    bnx_intr_check, sc);
4056 	lwkt_serialize_exit(ifp->if_serializer);
4057 }
4058 
4059 static void
4060 bnx_enable_intr(struct bnx_softc *sc)
4061 {
4062 	struct ifnet *ifp = &sc->arpcom.ac_if;
4063 
4064 	lwkt_serialize_handler_enable(ifp->if_serializer);
4065 
4066 	/*
4067 	 * Enable interrupt.
4068 	 */
4069 	bnx_writembx(sc, BGE_MBX_IRQ0_LO, sc->bnx_status_tag << 24);
4070 	if (sc->bnx_flags & BNX_FLAG_ONESHOT_MSI) {
4071 		/* XXX Linux driver */
4072 		bnx_writembx(sc, BGE_MBX_IRQ0_LO, sc->bnx_status_tag << 24);
4073 	}
4074 
4075 	/*
4076 	 * Unmask the interrupt when we stop polling.
4077 	 */
4078 	PCI_CLRBIT(sc->bnx_dev, BGE_PCI_MISC_CTL,
4079 	    BGE_PCIMISCCTL_MASK_PCI_INTR, 4);
4080 
4081 	/*
4082 	 * Trigger another interrupt, since above writing
4083 	 * to interrupt mailbox0 may acknowledge pending
4084 	 * interrupt.
4085 	 */
4086 	BNX_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
4087 
4088 	if (sc->bnx_flags & BNX_FLAG_STATUSTAG_BUG) {
4089 		sc->bnx_intr_maylose = FALSE;
4090 		sc->bnx_rx_check_considx = 0;
4091 		sc->bnx_tx_check_considx = 0;
4092 
4093 		if (bootverbose)
4094 			if_printf(ifp, "status tag bug workaround\n");
4095 
4096 		/* 10ms check interval */
4097 		callout_reset_bycpu(&sc->bnx_intr_timer, BNX_INTR_CKINTVL,
4098 		    bnx_intr_check, sc, sc->bnx_intr_cpuid);
4099 	}
4100 }
4101 
4102 static void
4103 bnx_disable_intr(struct bnx_softc *sc)
4104 {
4105 	struct ifnet *ifp = &sc->arpcom.ac_if;
4106 
4107 	/*
4108 	 * Mask the interrupt when we start polling.
4109 	 */
4110 	PCI_SETBIT(sc->bnx_dev, BGE_PCI_MISC_CTL,
4111 	    BGE_PCIMISCCTL_MASK_PCI_INTR, 4);
4112 
4113 	/*
4114 	 * Acknowledge possible asserted interrupt.
4115 	 */
4116 	bnx_writembx(sc, BGE_MBX_IRQ0_LO, 1);
4117 
4118 	callout_stop(&sc->bnx_intr_timer);
4119 	sc->bnx_intr_maylose = FALSE;
4120 	sc->bnx_rx_check_considx = 0;
4121 	sc->bnx_tx_check_considx = 0;
4122 
4123 	sc->bnx_npoll.ifpc_stcount = 0;
4124 
4125 	lwkt_serialize_handler_disable(ifp->if_serializer);
4126 }
4127 
4128 static int
4129 bnx_get_eaddr_mem(struct bnx_softc *sc, uint8_t ether_addr[])
4130 {
4131 	uint32_t mac_addr;
4132 	int ret = 1;
4133 
4134 	mac_addr = bnx_readmem_ind(sc, 0x0c14);
4135 	if ((mac_addr >> 16) == 0x484b) {
4136 		ether_addr[0] = (uint8_t)(mac_addr >> 8);
4137 		ether_addr[1] = (uint8_t)mac_addr;
4138 		mac_addr = bnx_readmem_ind(sc, 0x0c18);
4139 		ether_addr[2] = (uint8_t)(mac_addr >> 24);
4140 		ether_addr[3] = (uint8_t)(mac_addr >> 16);
4141 		ether_addr[4] = (uint8_t)(mac_addr >> 8);
4142 		ether_addr[5] = (uint8_t)mac_addr;
4143 		ret = 0;
4144 	}
4145 	return ret;
4146 }
4147 
4148 static int
4149 bnx_get_eaddr_nvram(struct bnx_softc *sc, uint8_t ether_addr[])
4150 {
4151 	int mac_offset = BGE_EE_MAC_OFFSET;
4152 
4153 	if (BNX_IS_5717_PLUS(sc)) {
4154 		int f;
4155 
4156 		f = pci_get_function(sc->bnx_dev);
4157 		if (f & 1)
4158 			mac_offset = BGE_EE_MAC_OFFSET_5717;
4159 		if (f > 1)
4160 			mac_offset += BGE_EE_MAC_OFFSET_5717_OFF;
4161 	} else if (sc->bnx_asicrev == BGE_ASICREV_BCM5906) {
4162 		mac_offset = BGE_EE_MAC_OFFSET_5906;
4163 	}
4164 
4165 	return bnx_read_nvram(sc, ether_addr, mac_offset + 2, ETHER_ADDR_LEN);
4166 }
4167 
4168 static int
4169 bnx_get_eaddr_eeprom(struct bnx_softc *sc, uint8_t ether_addr[])
4170 {
4171 	if (sc->bnx_flags & BNX_FLAG_NO_EEPROM)
4172 		return 1;
4173 
4174 	return bnx_read_eeprom(sc, ether_addr, BGE_EE_MAC_OFFSET + 2,
4175 			       ETHER_ADDR_LEN);
4176 }
4177 
4178 static int
4179 bnx_get_eaddr(struct bnx_softc *sc, uint8_t eaddr[])
4180 {
4181 	static const bnx_eaddr_fcn_t bnx_eaddr_funcs[] = {
4182 		/* NOTE: Order is critical */
4183 		bnx_get_eaddr_mem,
4184 		bnx_get_eaddr_nvram,
4185 		bnx_get_eaddr_eeprom,
4186 		NULL
4187 	};
4188 	const bnx_eaddr_fcn_t *func;
4189 
4190 	for (func = bnx_eaddr_funcs; *func != NULL; ++func) {
4191 		if ((*func)(sc, eaddr) == 0)
4192 			break;
4193 	}
4194 	return (*func == NULL ? ENXIO : 0);
4195 }
4196 
4197 /*
4198  * NOTE: 'm' is not freed upon failure
4199  */
4200 struct mbuf *
4201 bnx_defrag_shortdma(struct mbuf *m)
4202 {
4203 	struct mbuf *n;
4204 	int found;
4205 
4206 	/*
4207 	 * If device receive two back-to-back send BDs with less than
4208 	 * or equal to 8 total bytes then the device may hang.  The two
4209 	 * back-to-back send BDs must in the same frame for this failure
4210 	 * to occur.  Scan mbuf chains and see whether two back-to-back
4211 	 * send BDs are there.  If this is the case, allocate new mbuf
4212 	 * and copy the frame to workaround the silicon bug.
4213 	 */
4214 	for (n = m, found = 0; n != NULL; n = n->m_next) {
4215 		if (n->m_len < 8) {
4216 			found++;
4217 			if (found > 1)
4218 				break;
4219 			continue;
4220 		}
4221 		found = 0;
4222 	}
4223 
4224 	if (found > 1)
4225 		n = m_defrag(m, MB_DONTWAIT);
4226 	else
4227 		n = m;
4228 	return n;
4229 }
4230 
4231 static void
4232 bnx_stop_block(struct bnx_softc *sc, bus_size_t reg, uint32_t bit)
4233 {
4234 	int i;
4235 
4236 	BNX_CLRBIT(sc, reg, bit);
4237 	for (i = 0; i < BNX_TIMEOUT; i++) {
4238 		if ((CSR_READ_4(sc, reg) & bit) == 0)
4239 			return;
4240 		DELAY(100);
4241 	}
4242 }
4243 
4244 static void
4245 bnx_link_poll(struct bnx_softc *sc)
4246 {
4247 	uint32_t status;
4248 
4249 	status = CSR_READ_4(sc, BGE_MAC_STS);
4250 	if ((status & sc->bnx_link_chg) || sc->bnx_link_evt) {
4251 		sc->bnx_link_evt = 0;
4252 		sc->bnx_link_upd(sc, status);
4253 	}
4254 }
4255 
4256 static void
4257 bnx_enable_msi(struct bnx_softc *sc)
4258 {
4259 	uint32_t msi_mode;
4260 
4261 	msi_mode = CSR_READ_4(sc, BGE_MSI_MODE);
4262 	msi_mode |= BGE_MSIMODE_ENABLE;
4263 	if (sc->bnx_flags & BNX_FLAG_ONESHOT_MSI) {
4264 		/*
4265 		 * NOTE:
4266 		 * 5718-PG105-R says that "one shot" mode
4267 		 * does not work if MSI is used, however,
4268 		 * it obviously works.
4269 		 */
4270 		msi_mode &= ~BGE_MSIMODE_ONESHOT_DISABLE;
4271 	}
4272 	CSR_WRITE_4(sc, BGE_MSI_MODE, msi_mode);
4273 }
4274 
4275 static uint32_t
4276 bnx_dma_swap_options(struct bnx_softc *sc)
4277 {
4278 	uint32_t dma_options;
4279 
4280 	dma_options = BGE_MODECTL_WORDSWAP_NONFRAME |
4281 	    BGE_MODECTL_BYTESWAP_DATA | BGE_MODECTL_WORDSWAP_DATA;
4282 #if BYTE_ORDER == BIG_ENDIAN
4283 	dma_options |= BGE_MODECTL_BYTESWAP_NONFRAME;
4284 #endif
4285 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5720) {
4286 		dma_options |= BGE_MODECTL_BYTESWAP_B2HRX_DATA |
4287 		    BGE_MODECTL_WORDSWAP_B2HRX_DATA | BGE_MODECTL_B2HRX_ENABLE |
4288 		    BGE_MODECTL_HTX2B_ENABLE;
4289 	}
4290 	return dma_options;
4291 }
4292 
4293 static int
4294 bnx_setup_tso(struct bnx_softc *sc, struct mbuf **mp,
4295     uint16_t *mss0, uint16_t *flags0)
4296 {
4297 	struct mbuf *m;
4298 	struct ip *ip;
4299 	struct tcphdr *th;
4300 	int thoff, iphlen, hoff, hlen;
4301 	uint16_t flags, mss;
4302 
4303 	m = *mp;
4304 	KASSERT(M_WRITABLE(m), ("TSO mbuf not writable"));
4305 
4306 	hoff = m->m_pkthdr.csum_lhlen;
4307 	iphlen = m->m_pkthdr.csum_iphlen;
4308 	thoff = m->m_pkthdr.csum_thlen;
4309 
4310 	KASSERT(hoff > 0, ("invalid ether header len"));
4311 	KASSERT(iphlen > 0, ("invalid ip header len"));
4312 	KASSERT(thoff > 0, ("invalid tcp header len"));
4313 
4314 	if (__predict_false(m->m_len < hoff + iphlen + thoff)) {
4315 		m = m_pullup(m, hoff + iphlen + thoff);
4316 		if (m == NULL) {
4317 			*mp = NULL;
4318 			return ENOBUFS;
4319 		}
4320 		*mp = m;
4321 	}
4322 	ip = mtodoff(m, struct ip *, hoff);
4323 	th = mtodoff(m, struct tcphdr *, hoff + iphlen);
4324 
4325 	mss = m->m_pkthdr.tso_segsz;
4326 	flags = BGE_TXBDFLAG_CPU_PRE_DMA | BGE_TXBDFLAG_CPU_POST_DMA;
4327 
4328 	ip->ip_len = htons(mss + iphlen + thoff);
4329 	th->th_sum = 0;
4330 
4331 	hlen = (iphlen + thoff) >> 2;
4332 	mss |= ((hlen & 0x3) << 14);
4333 	flags |= ((hlen & 0xf8) << 7) | ((hlen & 0x4) << 2);
4334 
4335 	*mss0 = mss;
4336 	*flags0 = flags;
4337 
4338 	return 0;
4339 }
4340