xref: /dragonfly/sys/dev/netif/bnx/if_bnx.c (revision 89a89091)
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_polling.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_types.h>
63 #include <net/ifq_var.h>
64 #include <net/vlan/if_vlan_var.h>
65 #include <net/vlan/if_vlan_ether.h>
66 
67 #include <dev/netif/mii_layer/mii.h>
68 #include <dev/netif/mii_layer/miivar.h>
69 #include <dev/netif/mii_layer/brgphyreg.h>
70 
71 #include <bus/pci/pcidevs.h>
72 #include <bus/pci/pcireg.h>
73 #include <bus/pci/pcivar.h>
74 
75 #include <dev/netif/bge/if_bgereg.h>
76 #include <dev/netif/bnx/if_bnxvar.h>
77 
78 /* "device miibus" required.  See GENERIC if you get errors here. */
79 #include "miibus_if.h"
80 
81 #define BNX_CSUM_FEATURES	(CSUM_IP | CSUM_TCP | CSUM_UDP)
82 
83 #define BNX_INTR_CKINTVL	((10 * hz) / 1000)	/* 10ms */
84 
85 static const struct bnx_type {
86 	uint16_t		bnx_vid;
87 	uint16_t		bnx_did;
88 	char			*bnx_name;
89 } bnx_devs[] = {
90 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5717,
91 		"Broadcom BCM5717 Gigabit Ethernet" },
92 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5718,
93 		"Broadcom BCM5718 Gigabit Ethernet" },
94 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5719,
95 		"Broadcom BCM5719 Gigabit Ethernet" },
96 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5720_ALT,
97 		"Broadcom BCM5720 Gigabit Ethernet" },
98 
99 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57761,
100 		"Broadcom BCM57761 Gigabit Ethernet" },
101 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57762,
102 		"Broadcom BCM57762 Gigabit Ethernet" },
103 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57765,
104 		"Broadcom BCM57765 Gigabit Ethernet" },
105 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57766,
106 		"Broadcom BCM57766 Gigabit Ethernet" },
107 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57781,
108 		"Broadcom BCM57781 Gigabit Ethernet" },
109 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57782,
110 		"Broadcom BCM57782 Gigabit Ethernet" },
111 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57785,
112 		"Broadcom BCM57785 Gigabit Ethernet" },
113 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57786,
114 		"Broadcom BCM57786 Gigabit Ethernet" },
115 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57791,
116 		"Broadcom BCM57791 Fast Ethernet" },
117 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57795,
118 		"Broadcom BCM57795 Fast Ethernet" },
119 
120 	{ 0, 0, NULL }
121 };
122 
123 #define BNX_IS_JUMBO_CAPABLE(sc)	((sc)->bnx_flags & BNX_FLAG_JUMBO)
124 #define BNX_IS_5717_PLUS(sc)		((sc)->bnx_flags & BNX_FLAG_5717_PLUS)
125 #define BNX_IS_57765_PLUS(sc)		((sc)->bnx_flags & BNX_FLAG_57765_PLUS)
126 #define BNX_IS_57765_FAMILY(sc)	 \
127 	((sc)->bnx_flags & BNX_FLAG_57765_FAMILY)
128 
129 typedef int	(*bnx_eaddr_fcn_t)(struct bnx_softc *, uint8_t[]);
130 
131 static int	bnx_probe(device_t);
132 static int	bnx_attach(device_t);
133 static int	bnx_detach(device_t);
134 static void	bnx_shutdown(device_t);
135 static int	bnx_suspend(device_t);
136 static int	bnx_resume(device_t);
137 static int	bnx_miibus_readreg(device_t, int, int);
138 static int	bnx_miibus_writereg(device_t, int, int, int);
139 static void	bnx_miibus_statchg(device_t);
140 
141 #ifdef DEVICE_POLLING
142 static void	bnx_poll(struct ifnet *ifp, enum poll_cmd cmd, int count);
143 #endif
144 static void	bnx_intr_legacy(void *);
145 static void	bnx_msi(void *);
146 static void	bnx_msi_oneshot(void *);
147 static void	bnx_intr(struct bnx_softc *);
148 static void	bnx_enable_intr(struct bnx_softc *);
149 static void	bnx_disable_intr(struct bnx_softc *);
150 static void	bnx_txeof(struct bnx_softc *, uint16_t);
151 static void	bnx_rxeof(struct bnx_softc *, uint16_t);
152 
153 static void	bnx_start(struct ifnet *);
154 static int	bnx_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
155 static void	bnx_init(void *);
156 static void	bnx_stop(struct bnx_softc *);
157 static void	bnx_watchdog(struct ifnet *);
158 static int	bnx_ifmedia_upd(struct ifnet *);
159 static void	bnx_ifmedia_sts(struct ifnet *, struct ifmediareq *);
160 static void	bnx_tick(void *);
161 
162 static int	bnx_alloc_jumbo_mem(struct bnx_softc *);
163 static void	bnx_free_jumbo_mem(struct bnx_softc *);
164 static struct bnx_jslot
165 		*bnx_jalloc(struct bnx_softc *);
166 static void	bnx_jfree(void *);
167 static void	bnx_jref(void *);
168 static int	bnx_newbuf_std(struct bnx_softc *, int, int);
169 static int	bnx_newbuf_jumbo(struct bnx_softc *, int, int);
170 static void	bnx_setup_rxdesc_std(struct bnx_softc *, int);
171 static void	bnx_setup_rxdesc_jumbo(struct bnx_softc *, int);
172 static int	bnx_init_rx_ring_std(struct bnx_softc *);
173 static void	bnx_free_rx_ring_std(struct bnx_softc *);
174 static int	bnx_init_rx_ring_jumbo(struct bnx_softc *);
175 static void	bnx_free_rx_ring_jumbo(struct bnx_softc *);
176 static void	bnx_free_tx_ring(struct bnx_softc *);
177 static int	bnx_init_tx_ring(struct bnx_softc *);
178 static int	bnx_dma_alloc(struct bnx_softc *);
179 static void	bnx_dma_free(struct bnx_softc *);
180 static int	bnx_dma_block_alloc(struct bnx_softc *, bus_size_t,
181 		    bus_dma_tag_t *, bus_dmamap_t *, void **, bus_addr_t *);
182 static void	bnx_dma_block_free(bus_dma_tag_t, bus_dmamap_t, void *);
183 static struct mbuf *
184 		bnx_defrag_shortdma(struct mbuf *);
185 static int	bnx_encap(struct bnx_softc *, struct mbuf **, uint32_t *);
186 static int	bnx_setup_tso(struct bnx_softc *, struct mbuf **,
187 		    uint16_t *, uint16_t *);
188 
189 static void	bnx_reset(struct bnx_softc *);
190 static int	bnx_chipinit(struct bnx_softc *);
191 static int	bnx_blockinit(struct bnx_softc *);
192 static void	bnx_stop_block(struct bnx_softc *, bus_size_t, uint32_t);
193 static void	bnx_enable_msi(struct bnx_softc *sc);
194 static void	bnx_setmulti(struct bnx_softc *);
195 static void	bnx_setpromisc(struct bnx_softc *);
196 static void	bnx_stats_update_regs(struct bnx_softc *);
197 static uint32_t	bnx_dma_swap_options(struct bnx_softc *);
198 
199 static uint32_t	bnx_readmem_ind(struct bnx_softc *, uint32_t);
200 static void	bnx_writemem_ind(struct bnx_softc *, uint32_t, uint32_t);
201 #ifdef notdef
202 static uint32_t	bnx_readreg_ind(struct bnx_softc *, uint32_t);
203 #endif
204 static void	bnx_writereg_ind(struct bnx_softc *, uint32_t, uint32_t);
205 static void	bnx_writemem_direct(struct bnx_softc *, uint32_t, uint32_t);
206 static void	bnx_writembx(struct bnx_softc *, int, int);
207 static uint8_t	bnx_nvram_getbyte(struct bnx_softc *, int, uint8_t *);
208 static int	bnx_read_nvram(struct bnx_softc *, caddr_t, int, int);
209 static uint8_t	bnx_eeprom_getbyte(struct bnx_softc *, uint32_t, uint8_t *);
210 static int	bnx_read_eeprom(struct bnx_softc *, caddr_t, uint32_t, size_t);
211 
212 static void	bnx_tbi_link_upd(struct bnx_softc *, uint32_t);
213 static void	bnx_copper_link_upd(struct bnx_softc *, uint32_t);
214 static void	bnx_autopoll_link_upd(struct bnx_softc *, uint32_t);
215 static void	bnx_link_poll(struct bnx_softc *);
216 
217 static int	bnx_get_eaddr_mem(struct bnx_softc *, uint8_t[]);
218 static int	bnx_get_eaddr_nvram(struct bnx_softc *, uint8_t[]);
219 static int	bnx_get_eaddr_eeprom(struct bnx_softc *, uint8_t[]);
220 static int	bnx_get_eaddr(struct bnx_softc *, uint8_t[]);
221 
222 static void	bnx_coal_change(struct bnx_softc *);
223 static int	bnx_sysctl_rx_coal_ticks(SYSCTL_HANDLER_ARGS);
224 static int	bnx_sysctl_tx_coal_ticks(SYSCTL_HANDLER_ARGS);
225 static int	bnx_sysctl_rx_coal_bds(SYSCTL_HANDLER_ARGS);
226 static int	bnx_sysctl_tx_coal_bds(SYSCTL_HANDLER_ARGS);
227 static int	bnx_sysctl_rx_coal_bds_int(SYSCTL_HANDLER_ARGS);
228 static int	bnx_sysctl_tx_coal_bds_int(SYSCTL_HANDLER_ARGS);
229 static int	bnx_sysctl_coal_chg(SYSCTL_HANDLER_ARGS, uint32_t *,
230 		    int, int, uint32_t);
231 
232 static int	bnx_msi_enable = 1;
233 TUNABLE_INT("hw.bnx.msi.enable", &bnx_msi_enable);
234 
235 static device_method_t bnx_methods[] = {
236 	/* Device interface */
237 	DEVMETHOD(device_probe,		bnx_probe),
238 	DEVMETHOD(device_attach,	bnx_attach),
239 	DEVMETHOD(device_detach,	bnx_detach),
240 	DEVMETHOD(device_shutdown,	bnx_shutdown),
241 	DEVMETHOD(device_suspend,	bnx_suspend),
242 	DEVMETHOD(device_resume,	bnx_resume),
243 
244 	/* bus interface */
245 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
246 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
247 
248 	/* MII interface */
249 	DEVMETHOD(miibus_readreg,	bnx_miibus_readreg),
250 	DEVMETHOD(miibus_writereg,	bnx_miibus_writereg),
251 	DEVMETHOD(miibus_statchg,	bnx_miibus_statchg),
252 
253 	{ 0, 0 }
254 };
255 
256 static DEFINE_CLASS_0(bnx, bnx_driver, bnx_methods, sizeof(struct bnx_softc));
257 static devclass_t bnx_devclass;
258 
259 DECLARE_DUMMY_MODULE(if_bnx);
260 DRIVER_MODULE(if_bnx, pci, bnx_driver, bnx_devclass, NULL, NULL);
261 DRIVER_MODULE(miibus, bnx, miibus_driver, miibus_devclass, NULL, NULL);
262 
263 static uint32_t
264 bnx_readmem_ind(struct bnx_softc *sc, uint32_t off)
265 {
266 	device_t dev = sc->bnx_dev;
267 	uint32_t val;
268 
269 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5906 &&
270 	    off >= BGE_STATS_BLOCK && off < BGE_SEND_RING_1_TO_4)
271 		return 0;
272 
273 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, off, 4);
274 	val = pci_read_config(dev, BGE_PCI_MEMWIN_DATA, 4);
275 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, 0, 4);
276 	return (val);
277 }
278 
279 static void
280 bnx_writemem_ind(struct bnx_softc *sc, uint32_t off, uint32_t val)
281 {
282 	device_t dev = sc->bnx_dev;
283 
284 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5906 &&
285 	    off >= BGE_STATS_BLOCK && off < BGE_SEND_RING_1_TO_4)
286 		return;
287 
288 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, off, 4);
289 	pci_write_config(dev, BGE_PCI_MEMWIN_DATA, val, 4);
290 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, 0, 4);
291 }
292 
293 #ifdef notdef
294 static uint32_t
295 bnx_readreg_ind(struct bnx_softc *sc, uin32_t off)
296 {
297 	device_t dev = sc->bnx_dev;
298 
299 	pci_write_config(dev, BGE_PCI_REG_BASEADDR, off, 4);
300 	return(pci_read_config(dev, BGE_PCI_REG_DATA, 4));
301 }
302 #endif
303 
304 static void
305 bnx_writereg_ind(struct bnx_softc *sc, uint32_t off, uint32_t val)
306 {
307 	device_t dev = sc->bnx_dev;
308 
309 	pci_write_config(dev, BGE_PCI_REG_BASEADDR, off, 4);
310 	pci_write_config(dev, BGE_PCI_REG_DATA, val, 4);
311 }
312 
313 static void
314 bnx_writemem_direct(struct bnx_softc *sc, uint32_t off, uint32_t val)
315 {
316 	CSR_WRITE_4(sc, off, val);
317 }
318 
319 static void
320 bnx_writembx(struct bnx_softc *sc, int off, int val)
321 {
322 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5906)
323 		off += BGE_LPMBX_IRQ0_HI - BGE_MBX_IRQ0_HI;
324 
325 	CSR_WRITE_4(sc, off, val);
326 }
327 
328 static uint8_t
329 bnx_nvram_getbyte(struct bnx_softc *sc, int addr, uint8_t *dest)
330 {
331 	uint32_t access, byte = 0;
332 	int i;
333 
334 	/* Lock. */
335 	CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_SET1);
336 	for (i = 0; i < 8000; i++) {
337 		if (CSR_READ_4(sc, BGE_NVRAM_SWARB) & BGE_NVRAMSWARB_GNT1)
338 			break;
339 		DELAY(20);
340 	}
341 	if (i == 8000)
342 		return (1);
343 
344 	/* Enable access. */
345 	access = CSR_READ_4(sc, BGE_NVRAM_ACCESS);
346 	CSR_WRITE_4(sc, BGE_NVRAM_ACCESS, access | BGE_NVRAMACC_ENABLE);
347 
348 	CSR_WRITE_4(sc, BGE_NVRAM_ADDR, addr & 0xfffffffc);
349 	CSR_WRITE_4(sc, BGE_NVRAM_CMD, BGE_NVRAM_READCMD);
350 	for (i = 0; i < BNX_TIMEOUT * 10; i++) {
351 		DELAY(10);
352 		if (CSR_READ_4(sc, BGE_NVRAM_CMD) & BGE_NVRAMCMD_DONE) {
353 			DELAY(10);
354 			break;
355 		}
356 	}
357 
358 	if (i == BNX_TIMEOUT * 10) {
359 		if_printf(&sc->arpcom.ac_if, "nvram read timed out\n");
360 		return (1);
361 	}
362 
363 	/* Get result. */
364 	byte = CSR_READ_4(sc, BGE_NVRAM_RDDATA);
365 
366 	*dest = (bswap32(byte) >> ((addr % 4) * 8)) & 0xFF;
367 
368 	/* Disable access. */
369 	CSR_WRITE_4(sc, BGE_NVRAM_ACCESS, access);
370 
371 	/* Unlock. */
372 	CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_CLR1);
373 	CSR_READ_4(sc, BGE_NVRAM_SWARB);
374 
375 	return (0);
376 }
377 
378 /*
379  * Read a sequence of bytes from NVRAM.
380  */
381 static int
382 bnx_read_nvram(struct bnx_softc *sc, caddr_t dest, int off, int cnt)
383 {
384 	int err = 0, i;
385 	uint8_t byte = 0;
386 
387 	if (sc->bnx_asicrev != BGE_ASICREV_BCM5906)
388 		return (1);
389 
390 	for (i = 0; i < cnt; i++) {
391 		err = bnx_nvram_getbyte(sc, off + i, &byte);
392 		if (err)
393 			break;
394 		*(dest + i) = byte;
395 	}
396 
397 	return (err ? 1 : 0);
398 }
399 
400 /*
401  * Read a byte of data stored in the EEPROM at address 'addr.' The
402  * BCM570x supports both the traditional bitbang interface and an
403  * auto access interface for reading the EEPROM. We use the auto
404  * access method.
405  */
406 static uint8_t
407 bnx_eeprom_getbyte(struct bnx_softc *sc, uint32_t addr, uint8_t *dest)
408 {
409 	int i;
410 	uint32_t byte = 0;
411 
412 	/*
413 	 * Enable use of auto EEPROM access so we can avoid
414 	 * having to use the bitbang method.
415 	 */
416 	BNX_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_AUTO_EEPROM);
417 
418 	/* Reset the EEPROM, load the clock period. */
419 	CSR_WRITE_4(sc, BGE_EE_ADDR,
420 	    BGE_EEADDR_RESET|BGE_EEHALFCLK(BGE_HALFCLK_384SCL));
421 	DELAY(20);
422 
423 	/* Issue the read EEPROM command. */
424 	CSR_WRITE_4(sc, BGE_EE_ADDR, BGE_EE_READCMD | addr);
425 
426 	/* Wait for completion */
427 	for(i = 0; i < BNX_TIMEOUT * 10; i++) {
428 		DELAY(10);
429 		if (CSR_READ_4(sc, BGE_EE_ADDR) & BGE_EEADDR_DONE)
430 			break;
431 	}
432 
433 	if (i == BNX_TIMEOUT) {
434 		if_printf(&sc->arpcom.ac_if, "eeprom read timed out\n");
435 		return(1);
436 	}
437 
438 	/* Get result. */
439 	byte = CSR_READ_4(sc, BGE_EE_DATA);
440 
441         *dest = (byte >> ((addr % 4) * 8)) & 0xFF;
442 
443 	return(0);
444 }
445 
446 /*
447  * Read a sequence of bytes from the EEPROM.
448  */
449 static int
450 bnx_read_eeprom(struct bnx_softc *sc, caddr_t dest, uint32_t off, size_t len)
451 {
452 	size_t i;
453 	int err;
454 	uint8_t byte;
455 
456 	for (byte = 0, err = 0, i = 0; i < len; i++) {
457 		err = bnx_eeprom_getbyte(sc, off + i, &byte);
458 		if (err)
459 			break;
460 		*(dest + i) = byte;
461 	}
462 
463 	return(err ? 1 : 0);
464 }
465 
466 static int
467 bnx_miibus_readreg(device_t dev, int phy, int reg)
468 {
469 	struct bnx_softc *sc = device_get_softc(dev);
470 	uint32_t val;
471 	int i;
472 
473 	KASSERT(phy == sc->bnx_phyno,
474 	    ("invalid phyno %d, should be %d", phy, sc->bnx_phyno));
475 
476 	/* Clear the autopoll bit if set, otherwise may trigger PCI errors. */
477 	if (sc->bnx_mi_mode & BGE_MIMODE_AUTOPOLL) {
478 		CSR_WRITE_4(sc, BGE_MI_MODE,
479 		    sc->bnx_mi_mode & ~BGE_MIMODE_AUTOPOLL);
480 		DELAY(80);
481 	}
482 
483 	CSR_WRITE_4(sc, BGE_MI_COMM, BGE_MICMD_READ | BGE_MICOMM_BUSY |
484 	    BGE_MIPHY(phy) | BGE_MIREG(reg));
485 
486 	/* Poll for the PHY register access to complete. */
487 	for (i = 0; i < BNX_TIMEOUT; i++) {
488 		DELAY(10);
489 		val = CSR_READ_4(sc, BGE_MI_COMM);
490 		if ((val & BGE_MICOMM_BUSY) == 0) {
491 			DELAY(5);
492 			val = CSR_READ_4(sc, BGE_MI_COMM);
493 			break;
494 		}
495 	}
496 	if (i == BNX_TIMEOUT) {
497 		if_printf(&sc->arpcom.ac_if, "PHY read timed out "
498 		    "(phy %d, reg %d, val 0x%08x)\n", phy, reg, val);
499 		val = 0;
500 	}
501 
502 	/* Restore the autopoll bit if necessary. */
503 	if (sc->bnx_mi_mode & BGE_MIMODE_AUTOPOLL) {
504 		CSR_WRITE_4(sc, BGE_MI_MODE, sc->bnx_mi_mode);
505 		DELAY(80);
506 	}
507 
508 	if (val & BGE_MICOMM_READFAIL)
509 		return 0;
510 
511 	return (val & 0xFFFF);
512 }
513 
514 static int
515 bnx_miibus_writereg(device_t dev, int phy, int reg, int val)
516 {
517 	struct bnx_softc *sc = device_get_softc(dev);
518 	int i;
519 
520 	KASSERT(phy == sc->bnx_phyno,
521 	    ("invalid phyno %d, should be %d", phy, sc->bnx_phyno));
522 
523 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5906 &&
524 	    (reg == BRGPHY_MII_1000CTL || reg == BRGPHY_MII_AUXCTL))
525 	       return 0;
526 
527 	/* Clear the autopoll bit if set, otherwise may trigger PCI errors. */
528 	if (sc->bnx_mi_mode & BGE_MIMODE_AUTOPOLL) {
529 		CSR_WRITE_4(sc, BGE_MI_MODE,
530 		    sc->bnx_mi_mode & ~BGE_MIMODE_AUTOPOLL);
531 		DELAY(80);
532 	}
533 
534 	CSR_WRITE_4(sc, BGE_MI_COMM, BGE_MICMD_WRITE | BGE_MICOMM_BUSY |
535 	    BGE_MIPHY(phy) | BGE_MIREG(reg) | val);
536 
537 	for (i = 0; i < BNX_TIMEOUT; i++) {
538 		DELAY(10);
539 		if (!(CSR_READ_4(sc, BGE_MI_COMM) & BGE_MICOMM_BUSY)) {
540 			DELAY(5);
541 			CSR_READ_4(sc, BGE_MI_COMM); /* dummy read */
542 			break;
543 		}
544 	}
545 	if (i == BNX_TIMEOUT) {
546 		if_printf(&sc->arpcom.ac_if, "PHY write timed out "
547 		    "(phy %d, reg %d, val %d)\n", phy, reg, val);
548 	}
549 
550 	/* Restore the autopoll bit if necessary. */
551 	if (sc->bnx_mi_mode & BGE_MIMODE_AUTOPOLL) {
552 		CSR_WRITE_4(sc, BGE_MI_MODE, sc->bnx_mi_mode);
553 		DELAY(80);
554 	}
555 
556 	return 0;
557 }
558 
559 static void
560 bnx_miibus_statchg(device_t dev)
561 {
562 	struct bnx_softc *sc;
563 	struct mii_data *mii;
564 
565 	sc = device_get_softc(dev);
566 	mii = device_get_softc(sc->bnx_miibus);
567 
568 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
569 	    (IFM_ACTIVE | IFM_AVALID)) {
570 		switch (IFM_SUBTYPE(mii->mii_media_active)) {
571 		case IFM_10_T:
572 		case IFM_100_TX:
573 			sc->bnx_link = 1;
574 			break;
575 		case IFM_1000_T:
576 		case IFM_1000_SX:
577 		case IFM_2500_SX:
578 			if (sc->bnx_asicrev != BGE_ASICREV_BCM5906)
579 				sc->bnx_link = 1;
580 			else
581 				sc->bnx_link = 0;
582 			break;
583 		default:
584 			sc->bnx_link = 0;
585 			break;
586 		}
587 	} else {
588 		sc->bnx_link = 0;
589 	}
590 	if (sc->bnx_link == 0)
591 		return;
592 
593 	BNX_CLRBIT(sc, BGE_MAC_MODE, BGE_MACMODE_PORTMODE);
594 	if (IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_T ||
595 	    IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_SX) {
596 		BNX_SETBIT(sc, BGE_MAC_MODE, BGE_PORTMODE_GMII);
597 	} else {
598 		BNX_SETBIT(sc, BGE_MAC_MODE, BGE_PORTMODE_MII);
599 	}
600 
601 	if ((mii->mii_media_active & IFM_GMASK) == IFM_FDX) {
602 		BNX_CLRBIT(sc, BGE_MAC_MODE, BGE_MACMODE_HALF_DUPLEX);
603 	} else {
604 		BNX_SETBIT(sc, BGE_MAC_MODE, BGE_MACMODE_HALF_DUPLEX);
605 	}
606 }
607 
608 /*
609  * Memory management for jumbo frames.
610  */
611 static int
612 bnx_alloc_jumbo_mem(struct bnx_softc *sc)
613 {
614 	struct ifnet *ifp = &sc->arpcom.ac_if;
615 	struct bnx_jslot *entry;
616 	uint8_t *ptr;
617 	bus_addr_t paddr;
618 	int i, error;
619 
620 	/*
621 	 * Create tag for jumbo mbufs.
622 	 * This is really a bit of a kludge. We allocate a special
623 	 * jumbo buffer pool which (thanks to the way our DMA
624 	 * memory allocation works) will consist of contiguous
625 	 * pages. This means that even though a jumbo buffer might
626 	 * be larger than a page size, we don't really need to
627 	 * map it into more than one DMA segment. However, the
628 	 * default mbuf tag will result in multi-segment mappings,
629 	 * so we have to create a special jumbo mbuf tag that
630 	 * lets us get away with mapping the jumbo buffers as
631 	 * a single segment. I think eventually the driver should
632 	 * be changed so that it uses ordinary mbufs and cluster
633 	 * buffers, i.e. jumbo frames can span multiple DMA
634 	 * descriptors. But that's a project for another day.
635 	 */
636 
637 	/*
638 	 * Create DMA stuffs for jumbo RX ring.
639 	 */
640 	error = bnx_dma_block_alloc(sc, BGE_JUMBO_RX_RING_SZ,
641 				    &sc->bnx_cdata.bnx_rx_jumbo_ring_tag,
642 				    &sc->bnx_cdata.bnx_rx_jumbo_ring_map,
643 				    (void *)&sc->bnx_ldata.bnx_rx_jumbo_ring,
644 				    &sc->bnx_ldata.bnx_rx_jumbo_ring_paddr);
645 	if (error) {
646 		if_printf(ifp, "could not create jumbo RX ring\n");
647 		return error;
648 	}
649 
650 	/*
651 	 * Create DMA stuffs for jumbo buffer block.
652 	 */
653 	error = bnx_dma_block_alloc(sc, BNX_JMEM,
654 				    &sc->bnx_cdata.bnx_jumbo_tag,
655 				    &sc->bnx_cdata.bnx_jumbo_map,
656 				    (void **)&sc->bnx_ldata.bnx_jumbo_buf,
657 				    &paddr);
658 	if (error) {
659 		if_printf(ifp, "could not create jumbo buffer\n");
660 		return error;
661 	}
662 
663 	SLIST_INIT(&sc->bnx_jfree_listhead);
664 
665 	/*
666 	 * Now divide it up into 9K pieces and save the addresses
667 	 * in an array. Note that we play an evil trick here by using
668 	 * the first few bytes in the buffer to hold the the address
669 	 * of the softc structure for this interface. This is because
670 	 * bnx_jfree() needs it, but it is called by the mbuf management
671 	 * code which will not pass it to us explicitly.
672 	 */
673 	for (i = 0, ptr = sc->bnx_ldata.bnx_jumbo_buf; i < BNX_JSLOTS; i++) {
674 		entry = &sc->bnx_cdata.bnx_jslots[i];
675 		entry->bnx_sc = sc;
676 		entry->bnx_buf = ptr;
677 		entry->bnx_paddr = paddr;
678 		entry->bnx_inuse = 0;
679 		entry->bnx_slot = i;
680 		SLIST_INSERT_HEAD(&sc->bnx_jfree_listhead, entry, jslot_link);
681 
682 		ptr += BNX_JLEN;
683 		paddr += BNX_JLEN;
684 	}
685 	return 0;
686 }
687 
688 static void
689 bnx_free_jumbo_mem(struct bnx_softc *sc)
690 {
691 	/* Destroy jumbo RX ring. */
692 	bnx_dma_block_free(sc->bnx_cdata.bnx_rx_jumbo_ring_tag,
693 			   sc->bnx_cdata.bnx_rx_jumbo_ring_map,
694 			   sc->bnx_ldata.bnx_rx_jumbo_ring);
695 
696 	/* Destroy jumbo buffer block. */
697 	bnx_dma_block_free(sc->bnx_cdata.bnx_jumbo_tag,
698 			   sc->bnx_cdata.bnx_jumbo_map,
699 			   sc->bnx_ldata.bnx_jumbo_buf);
700 }
701 
702 /*
703  * Allocate a jumbo buffer.
704  */
705 static struct bnx_jslot *
706 bnx_jalloc(struct bnx_softc *sc)
707 {
708 	struct bnx_jslot *entry;
709 
710 	lwkt_serialize_enter(&sc->bnx_jslot_serializer);
711 	entry = SLIST_FIRST(&sc->bnx_jfree_listhead);
712 	if (entry) {
713 		SLIST_REMOVE_HEAD(&sc->bnx_jfree_listhead, jslot_link);
714 		entry->bnx_inuse = 1;
715 	} else {
716 		if_printf(&sc->arpcom.ac_if, "no free jumbo buffers\n");
717 	}
718 	lwkt_serialize_exit(&sc->bnx_jslot_serializer);
719 	return(entry);
720 }
721 
722 /*
723  * Adjust usage count on a jumbo buffer.
724  */
725 static void
726 bnx_jref(void *arg)
727 {
728 	struct bnx_jslot *entry = (struct bnx_jslot *)arg;
729 	struct bnx_softc *sc = entry->bnx_sc;
730 
731 	if (sc == NULL)
732 		panic("bnx_jref: can't find softc pointer!");
733 
734 	if (&sc->bnx_cdata.bnx_jslots[entry->bnx_slot] != entry) {
735 		panic("bnx_jref: asked to reference buffer "
736 		    "that we don't manage!");
737 	} else if (entry->bnx_inuse == 0) {
738 		panic("bnx_jref: buffer already free!");
739 	} else {
740 		atomic_add_int(&entry->bnx_inuse, 1);
741 	}
742 }
743 
744 /*
745  * Release a jumbo buffer.
746  */
747 static void
748 bnx_jfree(void *arg)
749 {
750 	struct bnx_jslot *entry = (struct bnx_jslot *)arg;
751 	struct bnx_softc *sc = entry->bnx_sc;
752 
753 	if (sc == NULL)
754 		panic("bnx_jfree: can't find softc pointer!");
755 
756 	if (&sc->bnx_cdata.bnx_jslots[entry->bnx_slot] != entry) {
757 		panic("bnx_jfree: asked to free buffer that we don't manage!");
758 	} else if (entry->bnx_inuse == 0) {
759 		panic("bnx_jfree: buffer already free!");
760 	} else {
761 		/*
762 		 * Possible MP race to 0, use the serializer.  The atomic insn
763 		 * is still needed for races against bnx_jref().
764 		 */
765 		lwkt_serialize_enter(&sc->bnx_jslot_serializer);
766 		atomic_subtract_int(&entry->bnx_inuse, 1);
767 		if (entry->bnx_inuse == 0) {
768 			SLIST_INSERT_HEAD(&sc->bnx_jfree_listhead,
769 					  entry, jslot_link);
770 		}
771 		lwkt_serialize_exit(&sc->bnx_jslot_serializer);
772 	}
773 }
774 
775 
776 /*
777  * Intialize a standard receive ring descriptor.
778  */
779 static int
780 bnx_newbuf_std(struct bnx_softc *sc, int i, int init)
781 {
782 	struct mbuf *m_new = NULL;
783 	bus_dma_segment_t seg;
784 	bus_dmamap_t map;
785 	int error, nsegs;
786 
787 	m_new = m_getcl(init ? MB_WAIT : MB_DONTWAIT, MT_DATA, M_PKTHDR);
788 	if (m_new == NULL)
789 		return ENOBUFS;
790 	m_new->m_len = m_new->m_pkthdr.len = MCLBYTES;
791 	m_adj(m_new, ETHER_ALIGN);
792 
793 	error = bus_dmamap_load_mbuf_segment(sc->bnx_cdata.bnx_rx_mtag,
794 			sc->bnx_cdata.bnx_rx_tmpmap, m_new,
795 			&seg, 1, &nsegs, BUS_DMA_NOWAIT);
796 	if (error) {
797 		m_freem(m_new);
798 		return error;
799 	}
800 
801 	if (!init) {
802 		bus_dmamap_sync(sc->bnx_cdata.bnx_rx_mtag,
803 				sc->bnx_cdata.bnx_rx_std_dmamap[i],
804 				BUS_DMASYNC_POSTREAD);
805 		bus_dmamap_unload(sc->bnx_cdata.bnx_rx_mtag,
806 			sc->bnx_cdata.bnx_rx_std_dmamap[i]);
807 	}
808 
809 	map = sc->bnx_cdata.bnx_rx_tmpmap;
810 	sc->bnx_cdata.bnx_rx_tmpmap = sc->bnx_cdata.bnx_rx_std_dmamap[i];
811 	sc->bnx_cdata.bnx_rx_std_dmamap[i] = map;
812 
813 	sc->bnx_cdata.bnx_rx_std_chain[i].bnx_mbuf = m_new;
814 	sc->bnx_cdata.bnx_rx_std_chain[i].bnx_paddr = seg.ds_addr;
815 
816 	bnx_setup_rxdesc_std(sc, i);
817 	return 0;
818 }
819 
820 static void
821 bnx_setup_rxdesc_std(struct bnx_softc *sc, int i)
822 {
823 	struct bnx_rxchain *rc;
824 	struct bge_rx_bd *r;
825 
826 	rc = &sc->bnx_cdata.bnx_rx_std_chain[i];
827 	r = &sc->bnx_ldata.bnx_rx_std_ring[i];
828 
829 	r->bge_addr.bge_addr_lo = BGE_ADDR_LO(rc->bnx_paddr);
830 	r->bge_addr.bge_addr_hi = BGE_ADDR_HI(rc->bnx_paddr);
831 	r->bge_len = rc->bnx_mbuf->m_len;
832 	r->bge_idx = i;
833 	r->bge_flags = BGE_RXBDFLAG_END;
834 }
835 
836 /*
837  * Initialize a jumbo receive ring descriptor. This allocates
838  * a jumbo buffer from the pool managed internally by the driver.
839  */
840 static int
841 bnx_newbuf_jumbo(struct bnx_softc *sc, int i, int init)
842 {
843 	struct mbuf *m_new = NULL;
844 	struct bnx_jslot *buf;
845 	bus_addr_t paddr;
846 
847 	/* Allocate the mbuf. */
848 	MGETHDR(m_new, init ? MB_WAIT : MB_DONTWAIT, MT_DATA);
849 	if (m_new == NULL)
850 		return ENOBUFS;
851 
852 	/* Allocate the jumbo buffer */
853 	buf = bnx_jalloc(sc);
854 	if (buf == NULL) {
855 		m_freem(m_new);
856 		return ENOBUFS;
857 	}
858 
859 	/* Attach the buffer to the mbuf. */
860 	m_new->m_ext.ext_arg = buf;
861 	m_new->m_ext.ext_buf = buf->bnx_buf;
862 	m_new->m_ext.ext_free = bnx_jfree;
863 	m_new->m_ext.ext_ref = bnx_jref;
864 	m_new->m_ext.ext_size = BNX_JUMBO_FRAMELEN;
865 
866 	m_new->m_flags |= M_EXT;
867 
868 	m_new->m_data = m_new->m_ext.ext_buf;
869 	m_new->m_len = m_new->m_pkthdr.len = m_new->m_ext.ext_size;
870 
871 	paddr = buf->bnx_paddr;
872 	m_adj(m_new, ETHER_ALIGN);
873 	paddr += ETHER_ALIGN;
874 
875 	/* Save necessary information */
876 	sc->bnx_cdata.bnx_rx_jumbo_chain[i].bnx_mbuf = m_new;
877 	sc->bnx_cdata.bnx_rx_jumbo_chain[i].bnx_paddr = paddr;
878 
879 	/* Set up the descriptor. */
880 	bnx_setup_rxdesc_jumbo(sc, i);
881 	return 0;
882 }
883 
884 static void
885 bnx_setup_rxdesc_jumbo(struct bnx_softc *sc, int i)
886 {
887 	struct bge_rx_bd *r;
888 	struct bnx_rxchain *rc;
889 
890 	r = &sc->bnx_ldata.bnx_rx_jumbo_ring[i];
891 	rc = &sc->bnx_cdata.bnx_rx_jumbo_chain[i];
892 
893 	r->bge_addr.bge_addr_lo = BGE_ADDR_LO(rc->bnx_paddr);
894 	r->bge_addr.bge_addr_hi = BGE_ADDR_HI(rc->bnx_paddr);
895 	r->bge_len = rc->bnx_mbuf->m_len;
896 	r->bge_idx = i;
897 	r->bge_flags = BGE_RXBDFLAG_END|BGE_RXBDFLAG_JUMBO_RING;
898 }
899 
900 static int
901 bnx_init_rx_ring_std(struct bnx_softc *sc)
902 {
903 	int i, error;
904 
905 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
906 		error = bnx_newbuf_std(sc, i, 1);
907 		if (error)
908 			return error;
909 	};
910 
911 	sc->bnx_std = BGE_STD_RX_RING_CNT - 1;
912 	bnx_writembx(sc, BGE_MBX_RX_STD_PROD_LO, sc->bnx_std);
913 
914 	return(0);
915 }
916 
917 static void
918 bnx_free_rx_ring_std(struct bnx_softc *sc)
919 {
920 	int i;
921 
922 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
923 		struct bnx_rxchain *rc = &sc->bnx_cdata.bnx_rx_std_chain[i];
924 
925 		if (rc->bnx_mbuf != NULL) {
926 			bus_dmamap_unload(sc->bnx_cdata.bnx_rx_mtag,
927 					  sc->bnx_cdata.bnx_rx_std_dmamap[i]);
928 			m_freem(rc->bnx_mbuf);
929 			rc->bnx_mbuf = NULL;
930 		}
931 		bzero(&sc->bnx_ldata.bnx_rx_std_ring[i],
932 		    sizeof(struct bge_rx_bd));
933 	}
934 }
935 
936 static int
937 bnx_init_rx_ring_jumbo(struct bnx_softc *sc)
938 {
939 	struct bge_rcb *rcb;
940 	int i, error;
941 
942 	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
943 		error = bnx_newbuf_jumbo(sc, i, 1);
944 		if (error)
945 			return error;
946 	};
947 
948 	sc->bnx_jumbo = BGE_JUMBO_RX_RING_CNT - 1;
949 
950 	rcb = &sc->bnx_ldata.bnx_info.bnx_jumbo_rx_rcb;
951 	rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(0, 0);
952 	CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
953 
954 	bnx_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, sc->bnx_jumbo);
955 
956 	return(0);
957 }
958 
959 static void
960 bnx_free_rx_ring_jumbo(struct bnx_softc *sc)
961 {
962 	int i;
963 
964 	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
965 		struct bnx_rxchain *rc = &sc->bnx_cdata.bnx_rx_jumbo_chain[i];
966 
967 		if (rc->bnx_mbuf != NULL) {
968 			m_freem(rc->bnx_mbuf);
969 			rc->bnx_mbuf = NULL;
970 		}
971 		bzero(&sc->bnx_ldata.bnx_rx_jumbo_ring[i],
972 		    sizeof(struct bge_rx_bd));
973 	}
974 }
975 
976 static void
977 bnx_free_tx_ring(struct bnx_softc *sc)
978 {
979 	int i;
980 
981 	for (i = 0; i < BGE_TX_RING_CNT; i++) {
982 		if (sc->bnx_cdata.bnx_tx_chain[i] != NULL) {
983 			bus_dmamap_unload(sc->bnx_cdata.bnx_tx_mtag,
984 					  sc->bnx_cdata.bnx_tx_dmamap[i]);
985 			m_freem(sc->bnx_cdata.bnx_tx_chain[i]);
986 			sc->bnx_cdata.bnx_tx_chain[i] = NULL;
987 		}
988 		bzero(&sc->bnx_ldata.bnx_tx_ring[i],
989 		    sizeof(struct bge_tx_bd));
990 	}
991 }
992 
993 static int
994 bnx_init_tx_ring(struct bnx_softc *sc)
995 {
996 	sc->bnx_txcnt = 0;
997 	sc->bnx_tx_saved_considx = 0;
998 	sc->bnx_tx_prodidx = 0;
999 
1000 	/* Initialize transmit producer index for host-memory send ring. */
1001 	bnx_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, sc->bnx_tx_prodidx);
1002 	bnx_writembx(sc, BGE_MBX_TX_NIC_PROD0_LO, 0);
1003 
1004 	return(0);
1005 }
1006 
1007 static void
1008 bnx_setmulti(struct bnx_softc *sc)
1009 {
1010 	struct ifnet *ifp;
1011 	struct ifmultiaddr *ifma;
1012 	uint32_t hashes[4] = { 0, 0, 0, 0 };
1013 	int h, i;
1014 
1015 	ifp = &sc->arpcom.ac_if;
1016 
1017 	if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
1018 		for (i = 0; i < 4; i++)
1019 			CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), 0xFFFFFFFF);
1020 		return;
1021 	}
1022 
1023 	/* First, zot all the existing filters. */
1024 	for (i = 0; i < 4; i++)
1025 		CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), 0);
1026 
1027 	/* Now program new ones. */
1028 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1029 		if (ifma->ifma_addr->sa_family != AF_LINK)
1030 			continue;
1031 		h = ether_crc32_le(
1032 		    LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
1033 		    ETHER_ADDR_LEN) & 0x7f;
1034 		hashes[(h & 0x60) >> 5] |= 1 << (h & 0x1F);
1035 	}
1036 
1037 	for (i = 0; i < 4; i++)
1038 		CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), hashes[i]);
1039 }
1040 
1041 /*
1042  * Do endian, PCI and DMA initialization. Also check the on-board ROM
1043  * self-test results.
1044  */
1045 static int
1046 bnx_chipinit(struct bnx_softc *sc)
1047 {
1048 	uint32_t dma_rw_ctl, mode_ctl;
1049 	int i;
1050 
1051 	/* Set endian type before we access any non-PCI registers. */
1052 	pci_write_config(sc->bnx_dev, BGE_PCI_MISC_CTL,
1053 	    BGE_INIT | BGE_PCIMISCCTL_TAGGED_STATUS, 4);
1054 
1055 	/* Clear the MAC control register */
1056 	CSR_WRITE_4(sc, BGE_MAC_MODE, 0);
1057 
1058 	/*
1059 	 * Clear the MAC statistics block in the NIC's
1060 	 * internal memory.
1061 	 */
1062 	for (i = BGE_STATS_BLOCK;
1063 	    i < BGE_STATS_BLOCK_END + 1; i += sizeof(uint32_t))
1064 		BNX_MEMWIN_WRITE(sc, i, 0);
1065 
1066 	for (i = BGE_STATUS_BLOCK;
1067 	    i < BGE_STATUS_BLOCK_END + 1; i += sizeof(uint32_t))
1068 		BNX_MEMWIN_WRITE(sc, i, 0);
1069 
1070 	if (BNX_IS_57765_FAMILY(sc)) {
1071 		uint32_t val;
1072 
1073 		if (sc->bnx_chipid == BGE_CHIPID_BCM57765_A0) {
1074 			mode_ctl = CSR_READ_4(sc, BGE_MODE_CTL);
1075 			val = mode_ctl & ~BGE_MODECTL_PCIE_PORTS;
1076 
1077 			/* Access the lower 1K of PL PCI-E block registers. */
1078 			CSR_WRITE_4(sc, BGE_MODE_CTL,
1079 			    val | BGE_MODECTL_PCIE_PL_SEL);
1080 
1081 			val = CSR_READ_4(sc, BGE_PCIE_PL_LO_PHYCTL5);
1082 			val |= BGE_PCIE_PL_LO_PHYCTL5_DIS_L2CLKREQ;
1083 			CSR_WRITE_4(sc, BGE_PCIE_PL_LO_PHYCTL5, val);
1084 
1085 			CSR_WRITE_4(sc, BGE_MODE_CTL, mode_ctl);
1086 		}
1087 		if (sc->bnx_chiprev != BGE_CHIPREV_57765_AX) {
1088 			mode_ctl = CSR_READ_4(sc, BGE_MODE_CTL);
1089 			val = mode_ctl & ~BGE_MODECTL_PCIE_PORTS;
1090 
1091 			/* Access the lower 1K of DL PCI-E block registers. */
1092 			CSR_WRITE_4(sc, BGE_MODE_CTL,
1093 			    val | BGE_MODECTL_PCIE_DL_SEL);
1094 
1095 			val = CSR_READ_4(sc, BGE_PCIE_DL_LO_FTSMAX);
1096 			val &= ~BGE_PCIE_DL_LO_FTSMAX_MASK;
1097 			val |= BGE_PCIE_DL_LO_FTSMAX_VAL;
1098 			CSR_WRITE_4(sc, BGE_PCIE_DL_LO_FTSMAX, val);
1099 
1100 			CSR_WRITE_4(sc, BGE_MODE_CTL, mode_ctl);
1101 		}
1102 
1103 		val = CSR_READ_4(sc, BGE_CPMU_LSPD_10MB_CLK);
1104 		val &= ~BGE_CPMU_LSPD_10MB_MACCLK_MASK;
1105 		val |= BGE_CPMU_LSPD_10MB_MACCLK_6_25;
1106 		CSR_WRITE_4(sc, BGE_CPMU_LSPD_10MB_CLK, val);
1107 	}
1108 
1109 	/*
1110 	 * Set up the PCI DMA control register.
1111 	 */
1112 	dma_rw_ctl = pci_read_config(sc->bnx_dev, BGE_PCI_DMA_RW_CTL, 4);
1113 	/*
1114 	 * Disable 32bytes cache alignment for DMA write to host memory
1115 	 *
1116 	 * NOTE:
1117 	 * 64bytes cache alignment for DMA write to host memory is still
1118 	 * enabled.
1119 	 */
1120 	dma_rw_ctl |= BGE_PCIDMARWCTL_DIS_CACHE_ALIGNMENT;
1121 	if (sc->bnx_chipid == BGE_CHIPID_BCM57765_A0)
1122 		dma_rw_ctl &= ~BGE_PCIDMARWCTL_CRDRDR_RDMA_MRRS_MSK;
1123 	/*
1124 	 * Enable HW workaround for controllers that misinterpret
1125 	 * a status tag update and leave interrupts permanently
1126 	 * disabled.
1127 	 */
1128 	if (sc->bnx_asicrev != BGE_ASICREV_BCM5717 &&
1129 	    !BNX_IS_57765_FAMILY(sc))
1130 		dma_rw_ctl |= BGE_PCIDMARWCTL_TAGGED_STATUS_WA;
1131 	if (bootverbose) {
1132 		if_printf(&sc->arpcom.ac_if, "DMA read/write %#x\n",
1133 		    dma_rw_ctl);
1134 	}
1135 	pci_write_config(sc->bnx_dev, BGE_PCI_DMA_RW_CTL, dma_rw_ctl, 4);
1136 
1137 	/*
1138 	 * Set up general mode register.
1139 	 */
1140 	mode_ctl = bnx_dma_swap_options(sc) | BGE_MODECTL_MAC_ATTN_INTR |
1141 	    BGE_MODECTL_HOST_SEND_BDS | BGE_MODECTL_TX_NO_PHDR_CSUM;
1142 	CSR_WRITE_4(sc, BGE_MODE_CTL, mode_ctl);
1143 
1144 	/*
1145 	 * Disable memory write invalidate.  Apparently it is not supported
1146 	 * properly by these devices.  Also ensure that INTx isn't disabled,
1147 	 * as these chips need it even when using MSI.
1148 	 */
1149 	PCI_CLRBIT(sc->bnx_dev, BGE_PCI_CMD,
1150 	    (PCIM_CMD_MWRICEN | PCIM_CMD_INTxDIS), 4);
1151 
1152 	/* Set the timer prescaler (always 66Mhz) */
1153 	CSR_WRITE_4(sc, BGE_MISC_CFG, 65 << 1/*BGE_32BITTIME_66MHZ*/);
1154 
1155 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5906) {
1156 		DELAY(40);	/* XXX */
1157 
1158 		/* Put PHY into ready state */
1159 		BNX_CLRBIT(sc, BGE_MISC_CFG, BGE_MISCCFG_EPHY_IDDQ);
1160 		CSR_READ_4(sc, BGE_MISC_CFG); /* Flush */
1161 		DELAY(40);
1162 	}
1163 
1164 	return(0);
1165 }
1166 
1167 static int
1168 bnx_blockinit(struct bnx_softc *sc)
1169 {
1170 	struct bge_rcb *rcb;
1171 	bus_size_t vrcb;
1172 	bge_hostaddr taddr;
1173 	uint32_t val;
1174 	int i, limit;
1175 
1176 	/*
1177 	 * Initialize the memory window pointer register so that
1178 	 * we can access the first 32K of internal NIC RAM. This will
1179 	 * allow us to set up the TX send ring RCBs and the RX return
1180 	 * ring RCBs, plus other things which live in NIC memory.
1181 	 */
1182 	CSR_WRITE_4(sc, BGE_PCI_MEMWIN_BASEADDR, 0);
1183 
1184 	/* Configure mbuf pool watermarks */
1185 	if (BNX_IS_57765_PLUS(sc)) {
1186 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
1187 		if (sc->arpcom.ac_if.if_mtu > ETHERMTU) {
1188 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x7e);
1189 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0xea);
1190 		} else {
1191 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x2a);
1192 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0xa0);
1193 		}
1194 	} else if (sc->bnx_asicrev == BGE_ASICREV_BCM5906) {
1195 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
1196 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x04);
1197 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x10);
1198 	} else {
1199 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
1200 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x10);
1201 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60);
1202 	}
1203 
1204 	/* Configure DMA resource watermarks */
1205 	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LOWAT, 5);
1206 	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_HIWAT, 10);
1207 
1208 	/* Enable buffer manager */
1209 	val = BGE_BMANMODE_ENABLE | BGE_BMANMODE_LOMBUF_ATTN;
1210 	/*
1211 	 * Change the arbitration algorithm of TXMBUF read request to
1212 	 * round-robin instead of priority based for BCM5719.  When
1213 	 * TXFIFO is almost empty, RDMA will hold its request until
1214 	 * TXFIFO is not almost empty.
1215 	 */
1216 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5719)
1217 		val |= BGE_BMANMODE_NO_TX_UNDERRUN;
1218 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5717 ||
1219 	    sc->bnx_chipid == BGE_CHIPID_BCM5719_A0 ||
1220 	    sc->bnx_chipid == BGE_CHIPID_BCM5720_A0)
1221 		val |= BGE_BMANMODE_LOMBUF_ATTN;
1222 	CSR_WRITE_4(sc, BGE_BMAN_MODE, val);
1223 
1224 	/* Poll for buffer manager start indication */
1225 	for (i = 0; i < BNX_TIMEOUT; i++) {
1226 		if (CSR_READ_4(sc, BGE_BMAN_MODE) & BGE_BMANMODE_ENABLE)
1227 			break;
1228 		DELAY(10);
1229 	}
1230 
1231 	if (i == BNX_TIMEOUT) {
1232 		if_printf(&sc->arpcom.ac_if,
1233 			  "buffer manager failed to start\n");
1234 		return(ENXIO);
1235 	}
1236 
1237 	/* Enable flow-through queues */
1238 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
1239 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
1240 
1241 	/* Wait until queue initialization is complete */
1242 	for (i = 0; i < BNX_TIMEOUT; i++) {
1243 		if (CSR_READ_4(sc, BGE_FTQ_RESET) == 0)
1244 			break;
1245 		DELAY(10);
1246 	}
1247 
1248 	if (i == BNX_TIMEOUT) {
1249 		if_printf(&sc->arpcom.ac_if,
1250 			  "flow-through queue init failed\n");
1251 		return(ENXIO);
1252 	}
1253 
1254 	/*
1255 	 * Summary of rings supported by the controller:
1256 	 *
1257 	 * Standard Receive Producer Ring
1258 	 * - This ring is used to feed receive buffers for "standard"
1259 	 *   sized frames (typically 1536 bytes) to the controller.
1260 	 *
1261 	 * Jumbo Receive Producer Ring
1262 	 * - This ring is used to feed receive buffers for jumbo sized
1263 	 *   frames (i.e. anything bigger than the "standard" frames)
1264 	 *   to the controller.
1265 	 *
1266 	 * Mini Receive Producer Ring
1267 	 * - This ring is used to feed receive buffers for "mini"
1268 	 *   sized frames to the controller.
1269 	 * - This feature required external memory for the controller
1270 	 *   but was never used in a production system.  Should always
1271 	 *   be disabled.
1272 	 *
1273 	 * Receive Return Ring
1274 	 * - After the controller has placed an incoming frame into a
1275 	 *   receive buffer that buffer is moved into a receive return
1276 	 *   ring.  The driver is then responsible to passing the
1277 	 *   buffer up to the stack.  Many versions of the controller
1278 	 *   support multiple RR rings.
1279 	 *
1280 	 * Send Ring
1281 	 * - This ring is used for outgoing frames.  Many versions of
1282 	 *   the controller support multiple send rings.
1283 	 */
1284 
1285 	/* Initialize the standard receive producer ring control block. */
1286 	rcb = &sc->bnx_ldata.bnx_info.bnx_std_rx_rcb;
1287 	rcb->bge_hostaddr.bge_addr_lo =
1288 	    BGE_ADDR_LO(sc->bnx_ldata.bnx_rx_std_ring_paddr);
1289 	rcb->bge_hostaddr.bge_addr_hi =
1290 	    BGE_ADDR_HI(sc->bnx_ldata.bnx_rx_std_ring_paddr);
1291 	if (BNX_IS_57765_PLUS(sc)) {
1292 		/*
1293 		 * Bits 31-16: Programmable ring size (2048, 1024, 512, .., 32)
1294 		 * Bits 15-2 : Maximum RX frame size
1295 		 * Bit 1     : 1 = Ring Disabled, 0 = Ring ENabled
1296 		 * Bit 0     : Reserved
1297 		 */
1298 		rcb->bge_maxlen_flags =
1299 		    BGE_RCB_MAXLEN_FLAGS(512, BNX_MAX_FRAMELEN << 2);
1300 	} else {
1301 		/*
1302 		 * Bits 31-16: Programmable ring size (512, 256, 128, 64, 32)
1303 		 * Bits 15-2 : Reserved (should be 0)
1304 		 * Bit 1     : 1 = Ring Disabled, 0 = Ring Enabled
1305 		 * Bit 0     : Reserved
1306 		 */
1307 		rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(512, 0);
1308 	}
1309 	if (BNX_IS_5717_PLUS(sc))
1310 		rcb->bge_nicaddr = BGE_STD_RX_RINGS_5717;
1311 	else
1312 		rcb->bge_nicaddr = BGE_STD_RX_RINGS;
1313 	/* Write the standard receive producer ring control block. */
1314 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_HI, rcb->bge_hostaddr.bge_addr_hi);
1315 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_LO, rcb->bge_hostaddr.bge_addr_lo);
1316 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
1317 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_NICADDR, rcb->bge_nicaddr);
1318 	/* Reset the standard receive producer ring producer index. */
1319 	bnx_writembx(sc, BGE_MBX_RX_STD_PROD_LO, 0);
1320 
1321 	/*
1322 	 * Initialize the jumbo RX producer ring control
1323 	 * block.  We set the 'ring disabled' bit in the
1324 	 * flags field until we're actually ready to start
1325 	 * using this ring (i.e. once we set the MTU
1326 	 * high enough to require it).
1327 	 */
1328 	if (BNX_IS_JUMBO_CAPABLE(sc)) {
1329 		rcb = &sc->bnx_ldata.bnx_info.bnx_jumbo_rx_rcb;
1330 		/* Get the jumbo receive producer ring RCB parameters. */
1331 		rcb->bge_hostaddr.bge_addr_lo =
1332 		    BGE_ADDR_LO(sc->bnx_ldata.bnx_rx_jumbo_ring_paddr);
1333 		rcb->bge_hostaddr.bge_addr_hi =
1334 		    BGE_ADDR_HI(sc->bnx_ldata.bnx_rx_jumbo_ring_paddr);
1335 		rcb->bge_maxlen_flags =
1336 		    BGE_RCB_MAXLEN_FLAGS(BNX_MAX_FRAMELEN,
1337 		    BGE_RCB_FLAG_RING_DISABLED);
1338 		if (BNX_IS_5717_PLUS(sc))
1339 			rcb->bge_nicaddr = BGE_JUMBO_RX_RINGS_5717;
1340 		else
1341 			rcb->bge_nicaddr = BGE_JUMBO_RX_RINGS;
1342 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_HI,
1343 		    rcb->bge_hostaddr.bge_addr_hi);
1344 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_LO,
1345 		    rcb->bge_hostaddr.bge_addr_lo);
1346 		/* Program the jumbo receive producer ring RCB parameters. */
1347 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS,
1348 		    rcb->bge_maxlen_flags);
1349 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_NICADDR, rcb->bge_nicaddr);
1350 		/* Reset the jumbo receive producer ring producer index. */
1351 		bnx_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, 0);
1352 	}
1353 
1354 	/* Choose de-pipeline mode for BCM5906 A0, A1 and A2. */
1355 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5906 &&
1356 	    (sc->bnx_chipid == BGE_CHIPID_BCM5906_A0 ||
1357 	     sc->bnx_chipid == BGE_CHIPID_BCM5906_A1 ||
1358 	     sc->bnx_chipid == BGE_CHIPID_BCM5906_A2)) {
1359 		CSR_WRITE_4(sc, BGE_ISO_PKT_TX,
1360 		    (CSR_READ_4(sc, BGE_ISO_PKT_TX) & ~3) | 2);
1361 	}
1362 
1363 	/*
1364 	 * The BD ring replenish thresholds control how often the
1365 	 * hardware fetches new BD's from the producer rings in host
1366 	 * memory.  Setting the value too low on a busy system can
1367 	 * starve the hardware and recue the throughpout.
1368 	 *
1369 	 * Set the BD ring replentish thresholds. The recommended
1370 	 * values are 1/8th the number of descriptors allocated to
1371 	 * each ring.
1372 	 */
1373 	val = 8;
1374 	CSR_WRITE_4(sc, BGE_RBDI_STD_REPL_THRESH, val);
1375 	if (BNX_IS_JUMBO_CAPABLE(sc)) {
1376 		CSR_WRITE_4(sc, BGE_RBDI_JUMBO_REPL_THRESH,
1377 		    BGE_JUMBO_RX_RING_CNT/8);
1378 	}
1379 	if (BNX_IS_57765_PLUS(sc)) {
1380 		CSR_WRITE_4(sc, BGE_STD_REPLENISH_LWM, 32);
1381 		CSR_WRITE_4(sc, BGE_JMB_REPLENISH_LWM, 16);
1382 	}
1383 
1384 	/*
1385 	 * Disable all send rings by setting the 'ring disabled' bit
1386 	 * in the flags field of all the TX send ring control blocks,
1387 	 * located in NIC memory.
1388 	 */
1389 	if (BNX_IS_5717_PLUS(sc))
1390 		limit = 4;
1391 	else if (BNX_IS_57765_FAMILY(sc))
1392 		limit = 2;
1393 	else
1394 		limit = 1;
1395 	vrcb = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
1396 	for (i = 0; i < limit; i++) {
1397 		RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1398 		    BGE_RCB_MAXLEN_FLAGS(0, BGE_RCB_FLAG_RING_DISABLED));
1399 		RCB_WRITE_4(sc, vrcb, bge_nicaddr, 0);
1400 		vrcb += sizeof(struct bge_rcb);
1401 	}
1402 
1403 	/* Configure send ring RCB 0 (we use only the first ring) */
1404 	vrcb = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
1405 	BGE_HOSTADDR(taddr, sc->bnx_ldata.bnx_tx_ring_paddr);
1406 	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
1407 	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
1408 	if (BNX_IS_5717_PLUS(sc)) {
1409 		RCB_WRITE_4(sc, vrcb, bge_nicaddr, BGE_SEND_RING_5717);
1410 	} else {
1411 		RCB_WRITE_4(sc, vrcb, bge_nicaddr,
1412 		    BGE_NIC_TXRING_ADDR(0, BGE_TX_RING_CNT));
1413 	}
1414 	RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1415 	    BGE_RCB_MAXLEN_FLAGS(BGE_TX_RING_CNT, 0));
1416 
1417 	/*
1418 	 * Disable all receive return rings by setting the
1419 	 * 'ring disabled' bit in the flags field of all the receive
1420 	 * return ring control blocks, located in NIC memory.
1421 	 */
1422 	if (BNX_IS_5717_PLUS(sc)) {
1423 		/* Should be 17, use 16 until we get an SRAM map. */
1424 		limit = 16;
1425 	} else if (BNX_IS_57765_FAMILY(sc)) {
1426 		limit = 4;
1427 	} else {
1428 		limit = 1;
1429 	}
1430 	/* Disable all receive return rings. */
1431 	vrcb = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
1432 	for (i = 0; i < limit; i++) {
1433 		RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, 0);
1434 		RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, 0);
1435 		RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1436 		    BGE_RCB_FLAG_RING_DISABLED);
1437 		RCB_WRITE_4(sc, vrcb, bge_nicaddr, 0);
1438 		bnx_writembx(sc, BGE_MBX_RX_CONS0_LO +
1439 		    (i * (sizeof(uint64_t))), 0);
1440 		vrcb += sizeof(struct bge_rcb);
1441 	}
1442 
1443 	/*
1444 	 * Set up receive return ring 0.  Note that the NIC address
1445 	 * for RX return rings is 0x0.  The return rings live entirely
1446 	 * within the host, so the nicaddr field in the RCB isn't used.
1447 	 */
1448 	vrcb = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
1449 	BGE_HOSTADDR(taddr, sc->bnx_ldata.bnx_rx_return_ring_paddr);
1450 	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, taddr.bge_addr_hi);
1451 	RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, taddr.bge_addr_lo);
1452 	RCB_WRITE_4(sc, vrcb, bge_nicaddr, 0);
1453 	RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1454 	    BGE_RCB_MAXLEN_FLAGS(sc->bnx_return_ring_cnt, 0));
1455 
1456 	/* Set random backoff seed for TX */
1457 	CSR_WRITE_4(sc, BGE_TX_RANDOM_BACKOFF,
1458 	    sc->arpcom.ac_enaddr[0] + sc->arpcom.ac_enaddr[1] +
1459 	    sc->arpcom.ac_enaddr[2] + sc->arpcom.ac_enaddr[3] +
1460 	    sc->arpcom.ac_enaddr[4] + sc->arpcom.ac_enaddr[5] +
1461 	    BGE_TX_BACKOFF_SEED_MASK);
1462 
1463 	/* Set inter-packet gap */
1464 	val = 0x2620;
1465 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5720) {
1466 		val |= CSR_READ_4(sc, BGE_TX_LENGTHS) &
1467 		    (BGE_TXLEN_JMB_FRM_LEN_MSK | BGE_TXLEN_CNT_DN_VAL_MSK);
1468 	}
1469 	CSR_WRITE_4(sc, BGE_TX_LENGTHS, val);
1470 
1471 	/*
1472 	 * Specify which ring to use for packets that don't match
1473 	 * any RX rules.
1474 	 */
1475 	CSR_WRITE_4(sc, BGE_RX_RULES_CFG, 0x08);
1476 
1477 	/*
1478 	 * Configure number of RX lists. One interrupt distribution
1479 	 * list, sixteen active lists, one bad frames class.
1480 	 */
1481 	CSR_WRITE_4(sc, BGE_RXLP_CFG, 0x181);
1482 
1483 	/* Inialize RX list placement stats mask. */
1484 	CSR_WRITE_4(sc, BGE_RXLP_STATS_ENABLE_MASK, 0x007FFFFF);
1485 	CSR_WRITE_4(sc, BGE_RXLP_STATS_CTL, 0x1);
1486 
1487 	/* Disable host coalescing until we get it set up */
1488 	CSR_WRITE_4(sc, BGE_HCC_MODE, 0x00000000);
1489 
1490 	/* Poll to make sure it's shut down. */
1491 	for (i = 0; i < BNX_TIMEOUT; i++) {
1492 		if (!(CSR_READ_4(sc, BGE_HCC_MODE) & BGE_HCCMODE_ENABLE))
1493 			break;
1494 		DELAY(10);
1495 	}
1496 
1497 	if (i == BNX_TIMEOUT) {
1498 		if_printf(&sc->arpcom.ac_if,
1499 			  "host coalescing engine failed to idle\n");
1500 		return(ENXIO);
1501 	}
1502 
1503 	/* Set up host coalescing defaults */
1504 	CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS, sc->bnx_rx_coal_ticks);
1505 	CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS, sc->bnx_tx_coal_ticks);
1506 	CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS, sc->bnx_rx_coal_bds);
1507 	CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS, sc->bnx_tx_coal_bds);
1508 	CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT, sc->bnx_rx_coal_bds_int);
1509 	CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT, sc->bnx_tx_coal_bds_int);
1510 
1511 	/* Set up address of status block */
1512 	bzero(sc->bnx_ldata.bnx_status_block, BGE_STATUS_BLK_SZ);
1513 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_HI,
1514 	    BGE_ADDR_HI(sc->bnx_ldata.bnx_status_block_paddr));
1515 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_LO,
1516 	    BGE_ADDR_LO(sc->bnx_ldata.bnx_status_block_paddr));
1517 
1518 	/* Set up status block partail update size. */
1519 	val = BGE_STATBLKSZ_32BYTE;
1520 #if 0
1521 	/*
1522 	 * Does not seem to have visible effect in both
1523 	 * bulk data (1472B UDP datagram) and tiny data
1524 	 * (18B UDP datagram) TX tests.
1525 	 */
1526 	val |= BGE_HCCMODE_CLRTICK_TX;
1527 #endif
1528 	/* Turn on host coalescing state machine */
1529 	CSR_WRITE_4(sc, BGE_HCC_MODE, val | BGE_HCCMODE_ENABLE);
1530 
1531 	/* Turn on RX BD completion state machine and enable attentions */
1532 	CSR_WRITE_4(sc, BGE_RBDC_MODE,
1533 	    BGE_RBDCMODE_ENABLE|BGE_RBDCMODE_ATTN);
1534 
1535 	/* Turn on RX list placement state machine */
1536 	CSR_WRITE_4(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
1537 
1538 	val = BGE_MACMODE_TXDMA_ENB | BGE_MACMODE_RXDMA_ENB |
1539 	    BGE_MACMODE_RX_STATS_CLEAR | BGE_MACMODE_TX_STATS_CLEAR |
1540 	    BGE_MACMODE_RX_STATS_ENB | BGE_MACMODE_TX_STATS_ENB |
1541 	    BGE_MACMODE_FRMHDR_DMA_ENB;
1542 
1543 	if (sc->bnx_flags & BNX_FLAG_TBI)
1544 		val |= BGE_PORTMODE_TBI;
1545 	else if (sc->bnx_flags & BNX_FLAG_MII_SERDES)
1546 		val |= BGE_PORTMODE_GMII;
1547 	else
1548 		val |= BGE_PORTMODE_MII;
1549 
1550 	/* Turn on DMA, clear stats */
1551 	CSR_WRITE_4(sc, BGE_MAC_MODE, val);
1552 
1553 	/* Set misc. local control, enable interrupts on attentions */
1554 	CSR_WRITE_4(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_ONATTN);
1555 
1556 #ifdef notdef
1557 	/* Assert GPIO pins for PHY reset */
1558 	BNX_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_MISCIO_OUT0|
1559 	    BGE_MLC_MISCIO_OUT1|BGE_MLC_MISCIO_OUT2);
1560 	BNX_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_MISCIO_OUTEN0|
1561 	    BGE_MLC_MISCIO_OUTEN1|BGE_MLC_MISCIO_OUTEN2);
1562 #endif
1563 
1564 	/* Turn on write DMA state machine */
1565 	val = BGE_WDMAMODE_ENABLE|BGE_WDMAMODE_ALL_ATTNS;
1566 	/* Enable host coalescing bug fix. */
1567 	val |= BGE_WDMAMODE_STATUS_TAG_FIX;
1568 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5785) {
1569 		/* Request larger DMA burst size to get better performance. */
1570 		val |= BGE_WDMAMODE_BURST_ALL_DATA;
1571 	}
1572 	CSR_WRITE_4(sc, BGE_WDMA_MODE, val);
1573 	DELAY(40);
1574 
1575 	if (BNX_IS_57765_PLUS(sc)) {
1576 		uint32_t dmactl;
1577 
1578 		dmactl = CSR_READ_4(sc, BGE_RDMA_RSRVCTRL);
1579 		/*
1580 		 * Adjust tx margin to prevent TX data corruption and
1581 		 * fix internal FIFO overflow.
1582 		 */
1583 		if (sc->bnx_asicrev == BGE_ASICREV_BCM5719 ||
1584 		    sc->bnx_asicrev == BGE_ASICREV_BCM5720) {
1585 			dmactl &= ~(BGE_RDMA_RSRVCTRL_FIFO_LWM_MASK |
1586 			    BGE_RDMA_RSRVCTRL_FIFO_HWM_MASK |
1587 			    BGE_RDMA_RSRVCTRL_TXMRGN_MASK);
1588 			dmactl |= BGE_RDMA_RSRVCTRL_FIFO_LWM_1_5K |
1589 			    BGE_RDMA_RSRVCTRL_FIFO_HWM_1_5K |
1590 			    BGE_RDMA_RSRVCTRL_TXMRGN_320B;
1591 		}
1592 		/*
1593 		 * Enable fix for read DMA FIFO overruns.
1594 		 * The fix is to limit the number of RX BDs
1595 		 * the hardware would fetch at a fime.
1596 		 */
1597 		CSR_WRITE_4(sc, BGE_RDMA_RSRVCTRL,
1598 		    dmactl | BGE_RDMA_RSRVCTRL_FIFO_OFLW_FIX);
1599 	}
1600 
1601 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5719) {
1602 		CSR_WRITE_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL,
1603 		    CSR_READ_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL) |
1604 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_BD_4K |
1605 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_LSO_4K);
1606 	} else if (sc->bnx_asicrev == BGE_ASICREV_BCM5720) {
1607 		/*
1608 		 * Allow 4KB burst length reads for non-LSO frames.
1609 		 * Enable 512B burst length reads for buffer descriptors.
1610 		 */
1611 		CSR_WRITE_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL,
1612 		    CSR_READ_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL) |
1613 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_BD_512 |
1614 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_LSO_4K);
1615 	}
1616 
1617 	/* Turn on read DMA state machine */
1618 	val = BGE_RDMAMODE_ENABLE | BGE_RDMAMODE_ALL_ATTNS;
1619 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5717)
1620 		val |= BGE_RDMAMODE_MULT_DMA_RD_DIS;
1621         if (sc->bnx_asicrev == BGE_ASICREV_BCM5784 ||
1622             sc->bnx_asicrev == BGE_ASICREV_BCM5785 ||
1623             sc->bnx_asicrev == BGE_ASICREV_BCM57780) {
1624 		val |= BGE_RDMAMODE_BD_SBD_CRPT_ATTN |
1625 		    BGE_RDMAMODE_MBUF_RBD_CRPT_ATTN |
1626 		    BGE_RDMAMODE_MBUF_SBD_CRPT_ATTN;
1627 	}
1628 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5720) {
1629 		val |= CSR_READ_4(sc, BGE_RDMA_MODE) &
1630 		    BGE_RDMAMODE_H2BNC_VLAN_DET;
1631 		/*
1632 		 * Allow multiple outstanding read requests from
1633 		 * non-LSO read DMA engine.
1634 		 */
1635 		val &= ~BGE_RDMAMODE_MULT_DMA_RD_DIS;
1636 	}
1637 	if (sc->bnx_flags & BNX_FLAG_TSO)
1638 		val |= BGE_RDMAMODE_TSO4_ENABLE;
1639 	val |= BGE_RDMAMODE_FIFO_LONG_BURST;
1640 	CSR_WRITE_4(sc, BGE_RDMA_MODE, val);
1641 	DELAY(40);
1642 
1643 	/* Turn on RX data completion state machine */
1644 	CSR_WRITE_4(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
1645 
1646 	/* Turn on RX BD initiator state machine */
1647 	CSR_WRITE_4(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
1648 
1649 	/* Turn on RX data and RX BD initiator state machine */
1650 	CSR_WRITE_4(sc, BGE_RDBDI_MODE, BGE_RDBDIMODE_ENABLE);
1651 
1652 	/* Turn on send BD completion state machine */
1653 	CSR_WRITE_4(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
1654 
1655 	/* Turn on send data completion state machine */
1656 	val = BGE_SDCMODE_ENABLE;
1657 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5761)
1658 		val |= BGE_SDCMODE_CDELAY;
1659 	CSR_WRITE_4(sc, BGE_SDC_MODE, val);
1660 
1661 	/* Turn on send data initiator state machine */
1662 	if (sc->bnx_flags & BNX_FLAG_TSO) {
1663 		CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE |
1664 		    BGE_SDIMODE_HW_LSO_PRE_DMA);
1665 	} else {
1666 		CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
1667 	}
1668 
1669 	/* Turn on send BD initiator state machine */
1670 	CSR_WRITE_4(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
1671 
1672 	/* Turn on send BD selector state machine */
1673 	CSR_WRITE_4(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
1674 
1675 	CSR_WRITE_4(sc, BGE_SDI_STATS_ENABLE_MASK, 0x007FFFFF);
1676 	CSR_WRITE_4(sc, BGE_SDI_STATS_CTL,
1677 	    BGE_SDISTATSCTL_ENABLE|BGE_SDISTATSCTL_FASTER);
1678 
1679 	/* ack/clear link change events */
1680 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED|
1681 	    BGE_MACSTAT_CFG_CHANGED|BGE_MACSTAT_MI_COMPLETE|
1682 	    BGE_MACSTAT_LINK_CHANGED);
1683 	CSR_WRITE_4(sc, BGE_MI_STS, 0);
1684 
1685 	/*
1686 	 * Enable attention when the link has changed state for
1687 	 * devices that use auto polling.
1688 	 */
1689 	if (sc->bnx_flags & BNX_FLAG_TBI) {
1690 		CSR_WRITE_4(sc, BGE_MI_STS, BGE_MISTS_LINK);
1691  	} else {
1692 		if (sc->bnx_mi_mode & BGE_MIMODE_AUTOPOLL) {
1693 			CSR_WRITE_4(sc, BGE_MI_MODE, sc->bnx_mi_mode);
1694 			DELAY(80);
1695 		}
1696 	}
1697 
1698 	/*
1699 	 * Clear any pending link state attention.
1700 	 * Otherwise some link state change events may be lost until attention
1701 	 * is cleared by bnx_intr() -> bnx_softc.bnx_link_upd() sequence.
1702 	 * It's not necessary on newer BCM chips - perhaps enabling link
1703 	 * state change attentions implies clearing pending attention.
1704 	 */
1705 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED|
1706 	    BGE_MACSTAT_CFG_CHANGED|BGE_MACSTAT_MI_COMPLETE|
1707 	    BGE_MACSTAT_LINK_CHANGED);
1708 
1709 	/* Enable link state change attentions. */
1710 	BNX_SETBIT(sc, BGE_MAC_EVT_ENB, BGE_EVTENB_LINK_CHANGED);
1711 
1712 	return(0);
1713 }
1714 
1715 /*
1716  * Probe for a Broadcom chip. Check the PCI vendor and device IDs
1717  * against our list and return its name if we find a match. Note
1718  * that since the Broadcom controller contains VPD support, we
1719  * can get the device name string from the controller itself instead
1720  * of the compiled-in string. This is a little slow, but it guarantees
1721  * we'll always announce the right product name.
1722  */
1723 static int
1724 bnx_probe(device_t dev)
1725 {
1726 	const struct bnx_type *t;
1727 	uint16_t product, vendor;
1728 
1729 	if (!pci_is_pcie(dev))
1730 		return ENXIO;
1731 
1732 	product = pci_get_device(dev);
1733 	vendor = pci_get_vendor(dev);
1734 
1735 	for (t = bnx_devs; t->bnx_name != NULL; t++) {
1736 		if (vendor == t->bnx_vid && product == t->bnx_did)
1737 			break;
1738 	}
1739 	if (t->bnx_name == NULL)
1740 		return ENXIO;
1741 
1742 	device_set_desc(dev, t->bnx_name);
1743 	return 0;
1744 }
1745 
1746 static int
1747 bnx_attach(device_t dev)
1748 {
1749 	struct ifnet *ifp;
1750 	struct bnx_softc *sc;
1751 	uint32_t hwcfg = 0, misccfg;
1752 	int error = 0, rid, capmask;
1753 	uint8_t ether_addr[ETHER_ADDR_LEN];
1754 	uint16_t product, vendor;
1755 	driver_intr_t *intr_func;
1756 	uintptr_t mii_priv = 0;
1757 	u_int intr_flags;
1758 #ifdef BNX_TSO_DEBUG
1759 	char desc[32];
1760 	int i;
1761 #endif
1762 
1763 	sc = device_get_softc(dev);
1764 	sc->bnx_dev = dev;
1765 	callout_init_mp(&sc->bnx_stat_timer);
1766 	callout_init_mp(&sc->bnx_intr_timer);
1767 	lwkt_serialize_init(&sc->bnx_jslot_serializer);
1768 
1769 	product = pci_get_device(dev);
1770 	vendor = pci_get_vendor(dev);
1771 
1772 #ifndef BURN_BRIDGES
1773 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
1774 		uint32_t irq, mem;
1775 
1776 		irq = pci_read_config(dev, PCIR_INTLINE, 4);
1777 		mem = pci_read_config(dev, BGE_PCI_BAR0, 4);
1778 
1779 		device_printf(dev, "chip is in D%d power mode "
1780 		    "-- setting to D0\n", pci_get_powerstate(dev));
1781 
1782 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
1783 
1784 		pci_write_config(dev, PCIR_INTLINE, irq, 4);
1785 		pci_write_config(dev, BGE_PCI_BAR0, mem, 4);
1786 	}
1787 #endif	/* !BURN_BRIDGE */
1788 
1789 	/*
1790 	 * Map control/status registers.
1791 	 */
1792 	pci_enable_busmaster(dev);
1793 
1794 	rid = BGE_PCI_BAR0;
1795 	sc->bnx_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid,
1796 	    RF_ACTIVE);
1797 
1798 	if (sc->bnx_res == NULL) {
1799 		device_printf(dev, "couldn't map memory\n");
1800 		return ENXIO;
1801 	}
1802 
1803 	sc->bnx_btag = rman_get_bustag(sc->bnx_res);
1804 	sc->bnx_bhandle = rman_get_bushandle(sc->bnx_res);
1805 
1806 	/* Save various chip information */
1807 	sc->bnx_chipid =
1808 	    pci_read_config(dev, BGE_PCI_MISC_CTL, 4) >>
1809 	    BGE_PCIMISCCTL_ASICREV_SHIFT;
1810 	if (BGE_ASICREV(sc->bnx_chipid) == BGE_ASICREV_USE_PRODID_REG) {
1811 		/* All chips having dedicated ASICREV register have CPMU */
1812 		sc->bnx_flags |= BNX_FLAG_CPMU;
1813 
1814 		switch (product) {
1815 		case PCI_PRODUCT_BROADCOM_BCM5717:
1816 		case PCI_PRODUCT_BROADCOM_BCM5718:
1817 		case PCI_PRODUCT_BROADCOM_BCM5719:
1818 		case PCI_PRODUCT_BROADCOM_BCM5720_ALT:
1819 			sc->bnx_chipid = pci_read_config(dev,
1820 			    BGE_PCI_GEN2_PRODID_ASICREV, 4);
1821 			break;
1822 
1823 		case PCI_PRODUCT_BROADCOM_BCM57761:
1824 		case PCI_PRODUCT_BROADCOM_BCM57762:
1825 		case PCI_PRODUCT_BROADCOM_BCM57765:
1826 		case PCI_PRODUCT_BROADCOM_BCM57766:
1827 		case PCI_PRODUCT_BROADCOM_BCM57781:
1828 		case PCI_PRODUCT_BROADCOM_BCM57782:
1829 		case PCI_PRODUCT_BROADCOM_BCM57785:
1830 		case PCI_PRODUCT_BROADCOM_BCM57786:
1831 		case PCI_PRODUCT_BROADCOM_BCM57791:
1832 		case PCI_PRODUCT_BROADCOM_BCM57795:
1833 			sc->bnx_chipid = pci_read_config(dev,
1834 			    BGE_PCI_GEN15_PRODID_ASICREV, 4);
1835 			break;
1836 
1837 		default:
1838 			sc->bnx_chipid = pci_read_config(dev,
1839 			    BGE_PCI_PRODID_ASICREV, 4);
1840 			break;
1841 		}
1842 	}
1843 	sc->bnx_asicrev = BGE_ASICREV(sc->bnx_chipid);
1844 	sc->bnx_chiprev = BGE_CHIPREV(sc->bnx_chipid);
1845 
1846 	switch (sc->bnx_asicrev) {
1847 	case BGE_ASICREV_BCM5717:
1848 	case BGE_ASICREV_BCM5719:
1849 	case BGE_ASICREV_BCM5720:
1850 		sc->bnx_flags |= BNX_FLAG_5717_PLUS | BNX_FLAG_57765_PLUS;
1851 		break;
1852 
1853 	case BGE_ASICREV_BCM57765:
1854 	case BGE_ASICREV_BCM57766:
1855 		sc->bnx_flags |= BNX_FLAG_57765_FAMILY | BNX_FLAG_57765_PLUS;
1856 		break;
1857 	}
1858 	sc->bnx_flags |= BNX_FLAG_SHORTDMA;
1859 
1860 	sc->bnx_flags |= BNX_FLAG_TSO;
1861 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5719 &&
1862 	    sc->bnx_chipid == BGE_CHIPID_BCM5719_A0)
1863 		sc->bnx_flags &= ~BNX_FLAG_TSO;
1864 
1865 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5717 ||
1866 	    BNX_IS_57765_FAMILY(sc)) {
1867 		/*
1868 		 * All BCM57785 and BCM5718 families chips have a bug that
1869 		 * under certain situation interrupt will not be enabled
1870 		 * even if status tag is written to BGE_MBX_IRQ0_LO mailbox.
1871 		 *
1872 		 * While BCM5719 and BCM5720 have a hardware workaround
1873 		 * which could fix the above bug.
1874 		 * See the comment near BGE_PCIDMARWCTL_TAGGED_STATUS_WA in
1875 		 * bnx_chipinit().
1876 		 *
1877 		 * For the rest of the chips in these two families, we will
1878 		 * have to poll the status block at high rate (10ms currently)
1879 		 * to check whether the interrupt is hosed or not.
1880 		 * See bnx_intr_check() for details.
1881 		 */
1882 		sc->bnx_flags |= BNX_FLAG_STATUSTAG_BUG;
1883 	}
1884 
1885 	misccfg = CSR_READ_4(sc, BGE_MISC_CFG) & BGE_MISCCFG_BOARD_ID_MASK;
1886 
1887 	sc->bnx_pciecap = pci_get_pciecap_ptr(sc->bnx_dev);
1888 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5719 ||
1889 	    sc->bnx_asicrev == BGE_ASICREV_BCM5720)
1890 		pcie_set_max_readrq(dev, PCIEM_DEVCTL_MAX_READRQ_2048);
1891 	else
1892 		pcie_set_max_readrq(dev, PCIEM_DEVCTL_MAX_READRQ_4096);
1893 	device_printf(dev, "CHIP ID 0x%08x; "
1894 		      "ASIC REV 0x%02x; CHIP REV 0x%02x\n",
1895 		      sc->bnx_chipid, sc->bnx_asicrev, sc->bnx_chiprev);
1896 
1897 	/*
1898 	 * Set various PHY quirk flags.
1899 	 */
1900 
1901 	capmask = MII_CAPMASK_DEFAULT;
1902 	if (product == PCI_PRODUCT_BROADCOM_BCM57791 ||
1903 	    product == PCI_PRODUCT_BROADCOM_BCM57795) {
1904 		/* 10/100 only */
1905 		capmask &= ~BMSR_EXTSTAT;
1906 	}
1907 
1908 	mii_priv |= BRGPHY_FLAG_WIRESPEED;
1909 
1910 	/*
1911 	 * Allocate interrupt
1912 	 */
1913 	sc->bnx_irq_type = pci_alloc_1intr(dev, bnx_msi_enable, &sc->bnx_irq_rid,
1914 	    &intr_flags);
1915 
1916 	sc->bnx_irq = bus_alloc_resource_any(dev, SYS_RES_IRQ, &sc->bnx_irq_rid,
1917 	    intr_flags);
1918 	if (sc->bnx_irq == NULL) {
1919 		device_printf(dev, "couldn't map interrupt\n");
1920 		error = ENXIO;
1921 		goto fail;
1922 	}
1923 
1924 	if (sc->bnx_irq_type == PCI_INTR_TYPE_MSI) {
1925 		sc->bnx_flags |= BNX_FLAG_ONESHOT_MSI;
1926 		bnx_enable_msi(sc);
1927 	}
1928 
1929 	/* Initialize if_name earlier, so if_printf could be used */
1930 	ifp = &sc->arpcom.ac_if;
1931 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
1932 
1933 	/* Try to reset the chip. */
1934 	bnx_reset(sc);
1935 
1936 	if (bnx_chipinit(sc)) {
1937 		device_printf(dev, "chip initialization failed\n");
1938 		error = ENXIO;
1939 		goto fail;
1940 	}
1941 
1942 	/*
1943 	 * Get station address
1944 	 */
1945 	error = bnx_get_eaddr(sc, ether_addr);
1946 	if (error) {
1947 		device_printf(dev, "failed to read station address\n");
1948 		goto fail;
1949 	}
1950 
1951 	if (BNX_IS_57765_PLUS(sc)) {
1952 		sc->bnx_return_ring_cnt = BGE_RETURN_RING_CNT;
1953 	} else {
1954 		/* 5705/5750 limits RX return ring to 512 entries. */
1955 		sc->bnx_return_ring_cnt = BGE_RETURN_RING_CNT_5705;
1956 	}
1957 
1958 	error = bnx_dma_alloc(sc);
1959 	if (error)
1960 		goto fail;
1961 
1962 	/* Set default tuneable values. */
1963 	sc->bnx_rx_coal_ticks = BNX_RX_COAL_TICKS_DEF;
1964 	sc->bnx_tx_coal_ticks = BNX_TX_COAL_TICKS_DEF;
1965 	sc->bnx_rx_coal_bds = BNX_RX_COAL_BDS_DEF;
1966 	sc->bnx_tx_coal_bds = BNX_TX_COAL_BDS_DEF;
1967 	sc->bnx_rx_coal_bds_int = BNX_RX_COAL_BDS_DEF;
1968 	sc->bnx_tx_coal_bds_int = BNX_TX_COAL_BDS_DEF;
1969 
1970 	/* Set up ifnet structure */
1971 	ifp->if_softc = sc;
1972 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
1973 	ifp->if_ioctl = bnx_ioctl;
1974 	ifp->if_start = bnx_start;
1975 #ifdef DEVICE_POLLING
1976 	ifp->if_poll = bnx_poll;
1977 #endif
1978 	ifp->if_watchdog = bnx_watchdog;
1979 	ifp->if_init = bnx_init;
1980 	ifp->if_mtu = ETHERMTU;
1981 	ifp->if_capabilities = IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU;
1982 	ifq_set_maxlen(&ifp->if_snd, BGE_TX_RING_CNT - 1);
1983 	ifq_set_ready(&ifp->if_snd);
1984 
1985 	ifp->if_capabilities |= IFCAP_HWCSUM;
1986 	ifp->if_hwassist = BNX_CSUM_FEATURES;
1987 	if (sc->bnx_flags & BNX_FLAG_TSO) {
1988 		ifp->if_capabilities |= IFCAP_TSO;
1989 		ifp->if_hwassist |= CSUM_TSO;
1990 	}
1991 	ifp->if_capenable = ifp->if_capabilities;
1992 
1993 	/*
1994 	 * Figure out what sort of media we have by checking the
1995 	 * hardware config word in the first 32k of NIC internal memory,
1996 	 * or fall back to examining the EEPROM if necessary.
1997 	 * Note: on some BCM5700 cards, this value appears to be unset.
1998 	 * If that's the case, we have to rely on identifying the NIC
1999 	 * by its PCI subsystem ID, as we do below for the SysKonnect
2000 	 * SK-9D41.
2001 	 */
2002 	if (bnx_readmem_ind(sc, BGE_SOFTWARE_GENCOMM_SIG) == BGE_MAGIC_NUMBER) {
2003 		hwcfg = bnx_readmem_ind(sc, BGE_SOFTWARE_GENCOMM_NICCFG);
2004 	} else {
2005 		if (bnx_read_eeprom(sc, (caddr_t)&hwcfg, BGE_EE_HWCFG_OFFSET,
2006 				    sizeof(hwcfg))) {
2007 			device_printf(dev, "failed to read EEPROM\n");
2008 			error = ENXIO;
2009 			goto fail;
2010 		}
2011 		hwcfg = ntohl(hwcfg);
2012 	}
2013 
2014 	/* The SysKonnect SK-9D41 is a 1000baseSX card. */
2015 	if (pci_get_subvendor(dev) == PCI_PRODUCT_SCHNEIDERKOCH_SK_9D41 ||
2016 	    (hwcfg & BGE_HWCFG_MEDIA) == BGE_MEDIA_FIBER)
2017 		sc->bnx_flags |= BNX_FLAG_TBI;
2018 
2019 	/* Setup MI MODE */
2020 	if (sc->bnx_flags & BNX_FLAG_CPMU)
2021 		sc->bnx_mi_mode = BGE_MIMODE_500KHZ_CONST;
2022 	else
2023 		sc->bnx_mi_mode = BGE_MIMODE_BASE;
2024 
2025 	/* Setup link status update stuffs */
2026 	if (sc->bnx_flags & BNX_FLAG_TBI) {
2027 		sc->bnx_link_upd = bnx_tbi_link_upd;
2028 		sc->bnx_link_chg = BGE_MACSTAT_LINK_CHANGED;
2029 	} else if (sc->bnx_mi_mode & BGE_MIMODE_AUTOPOLL) {
2030 		sc->bnx_link_upd = bnx_autopoll_link_upd;
2031 		sc->bnx_link_chg = BGE_MACSTAT_LINK_CHANGED;
2032 	} else {
2033 		sc->bnx_link_upd = bnx_copper_link_upd;
2034 		sc->bnx_link_chg = BGE_MACSTAT_LINK_CHANGED;
2035 	}
2036 
2037 	/* Set default PHY address */
2038 	sc->bnx_phyno = 1;
2039 
2040 	/*
2041 	 * PHY address mapping for various devices.
2042 	 *
2043 	 *          | F0 Cu | F0 Sr | F1 Cu | F1 Sr |
2044 	 * ---------+-------+-------+-------+-------+
2045 	 * BCM57XX  |   1   |   X   |   X   |   X   |
2046 	 * BCM5704  |   1   |   X   |   1   |   X   |
2047 	 * BCM5717  |   1   |   8   |   2   |   9   |
2048 	 * BCM5719  |   1   |   8   |   2   |   9   |
2049 	 * BCM5720  |   1   |   8   |   2   |   9   |
2050 	 *
2051 	 * Other addresses may respond but they are not
2052 	 * IEEE compliant PHYs and should be ignored.
2053 	 */
2054 	if (BNX_IS_5717_PLUS(sc)) {
2055 		int f;
2056 
2057 		f = pci_get_function(dev);
2058 		if (sc->bnx_chipid == BGE_CHIPID_BCM5717_A0) {
2059 			if (CSR_READ_4(sc, BGE_SGDIG_STS) &
2060 			    BGE_SGDIGSTS_IS_SERDES)
2061 				sc->bnx_phyno = f + 8;
2062 			else
2063 				sc->bnx_phyno = f + 1;
2064 		} else {
2065 			if (CSR_READ_4(sc, BGE_CPMU_PHY_STRAP) &
2066 			    BGE_CPMU_PHY_STRAP_IS_SERDES)
2067 				sc->bnx_phyno = f + 8;
2068 			else
2069 				sc->bnx_phyno = f + 1;
2070 		}
2071 	}
2072 
2073 	if (sc->bnx_flags & BNX_FLAG_TBI) {
2074 		ifmedia_init(&sc->bnx_ifmedia, IFM_IMASK,
2075 		    bnx_ifmedia_upd, bnx_ifmedia_sts);
2076 		ifmedia_add(&sc->bnx_ifmedia, IFM_ETHER|IFM_1000_SX, 0, NULL);
2077 		ifmedia_add(&sc->bnx_ifmedia,
2078 		    IFM_ETHER|IFM_1000_SX|IFM_FDX, 0, NULL);
2079 		ifmedia_add(&sc->bnx_ifmedia, IFM_ETHER|IFM_AUTO, 0, NULL);
2080 		ifmedia_set(&sc->bnx_ifmedia, IFM_ETHER|IFM_AUTO);
2081 		sc->bnx_ifmedia.ifm_media = sc->bnx_ifmedia.ifm_cur->ifm_media;
2082 	} else {
2083 		struct mii_probe_args mii_args;
2084 
2085 		mii_probe_args_init(&mii_args, bnx_ifmedia_upd, bnx_ifmedia_sts);
2086 		mii_args.mii_probemask = 1 << sc->bnx_phyno;
2087 		mii_args.mii_capmask = capmask;
2088 		mii_args.mii_privtag = MII_PRIVTAG_BRGPHY;
2089 		mii_args.mii_priv = mii_priv;
2090 
2091 		error = mii_probe(dev, &sc->bnx_miibus, &mii_args);
2092 		if (error) {
2093 			device_printf(dev, "MII without any PHY!\n");
2094 			goto fail;
2095 		}
2096 	}
2097 
2098 	/*
2099 	 * Create sysctl nodes.
2100 	 */
2101 	sysctl_ctx_init(&sc->bnx_sysctl_ctx);
2102 	sc->bnx_sysctl_tree = SYSCTL_ADD_NODE(&sc->bnx_sysctl_ctx,
2103 					      SYSCTL_STATIC_CHILDREN(_hw),
2104 					      OID_AUTO,
2105 					      device_get_nameunit(dev),
2106 					      CTLFLAG_RD, 0, "");
2107 	if (sc->bnx_sysctl_tree == NULL) {
2108 		device_printf(dev, "can't add sysctl node\n");
2109 		error = ENXIO;
2110 		goto fail;
2111 	}
2112 
2113 	SYSCTL_ADD_PROC(&sc->bnx_sysctl_ctx,
2114 			SYSCTL_CHILDREN(sc->bnx_sysctl_tree),
2115 			OID_AUTO, "rx_coal_ticks",
2116 			CTLTYPE_INT | CTLFLAG_RW,
2117 			sc, 0, bnx_sysctl_rx_coal_ticks, "I",
2118 			"Receive coalescing ticks (usec).");
2119 	SYSCTL_ADD_PROC(&sc->bnx_sysctl_ctx,
2120 			SYSCTL_CHILDREN(sc->bnx_sysctl_tree),
2121 			OID_AUTO, "tx_coal_ticks",
2122 			CTLTYPE_INT | CTLFLAG_RW,
2123 			sc, 0, bnx_sysctl_tx_coal_ticks, "I",
2124 			"Transmit coalescing ticks (usec).");
2125 	SYSCTL_ADD_PROC(&sc->bnx_sysctl_ctx,
2126 			SYSCTL_CHILDREN(sc->bnx_sysctl_tree),
2127 			OID_AUTO, "rx_coal_bds",
2128 			CTLTYPE_INT | CTLFLAG_RW,
2129 			sc, 0, bnx_sysctl_rx_coal_bds, "I",
2130 			"Receive max coalesced BD count.");
2131 	SYSCTL_ADD_PROC(&sc->bnx_sysctl_ctx,
2132 			SYSCTL_CHILDREN(sc->bnx_sysctl_tree),
2133 			OID_AUTO, "tx_coal_bds",
2134 			CTLTYPE_INT | CTLFLAG_RW,
2135 			sc, 0, bnx_sysctl_tx_coal_bds, "I",
2136 			"Transmit max coalesced BD count.");
2137 	/*
2138 	 * A common design characteristic for many Broadcom
2139 	 * client controllers is that they only support a
2140 	 * single outstanding DMA read operation on the PCIe
2141 	 * bus. This means that it will take twice as long to
2142 	 * fetch a TX frame that is split into header and
2143 	 * payload buffers as it does to fetch a single,
2144 	 * contiguous TX frame (2 reads vs. 1 read). For these
2145 	 * controllers, coalescing buffers to reduce the number
2146 	 * of memory reads is effective way to get maximum
2147 	 * performance(about 940Mbps).  Without collapsing TX
2148 	 * buffers the maximum TCP bulk transfer performance
2149 	 * is about 850Mbps. However forcing coalescing mbufs
2150 	 * consumes a lot of CPU cycles, so leave it off by
2151 	 * default.
2152 	 */
2153 	SYSCTL_ADD_INT(&sc->bnx_sysctl_ctx,
2154 	    SYSCTL_CHILDREN(sc->bnx_sysctl_tree), OID_AUTO,
2155 	    "force_defrag", CTLFLAG_RW, &sc->bnx_force_defrag, 0,
2156 	    "Force defragment on TX path");
2157 
2158 	SYSCTL_ADD_PROC(&sc->bnx_sysctl_ctx,
2159 	    SYSCTL_CHILDREN(sc->bnx_sysctl_tree), OID_AUTO,
2160 	    "rx_coal_bds_int", CTLTYPE_INT | CTLFLAG_RW,
2161 	    sc, 0, bnx_sysctl_rx_coal_bds_int, "I",
2162 	    "Receive max coalesced BD count during interrupt.");
2163 	SYSCTL_ADD_PROC(&sc->bnx_sysctl_ctx,
2164 	    SYSCTL_CHILDREN(sc->bnx_sysctl_tree), OID_AUTO,
2165 	    "tx_coal_bds_int", CTLTYPE_INT | CTLFLAG_RW,
2166 	    sc, 0, bnx_sysctl_tx_coal_bds_int, "I",
2167 	    "Transmit max coalesced BD count during interrupt.");
2168 
2169 #ifdef BNX_TSO_DEBUG
2170 	for (i = 0; i < BNX_TSO_NSTATS; ++i) {
2171 		ksnprintf(desc, sizeof(desc), "tso%d", i + 1);
2172 		SYSCTL_ADD_ULONG(&sc->bnx_sysctl_ctx,
2173 		    SYSCTL_CHILDREN(sc->bnx_sysctl_tree), OID_AUTO,
2174 		    desc, CTLFLAG_RW, &sc->bnx_tsosegs[i], "");
2175 	}
2176 #endif
2177 
2178 	/*
2179 	 * Call MI attach routine.
2180 	 */
2181 	ether_ifattach(ifp, ether_addr, NULL);
2182 
2183 	if (sc->bnx_irq_type == PCI_INTR_TYPE_MSI) {
2184 		if (sc->bnx_flags & BNX_FLAG_ONESHOT_MSI) {
2185 			intr_func = bnx_msi_oneshot;
2186 			if (bootverbose)
2187 				device_printf(dev, "oneshot MSI\n");
2188 		} else {
2189 			intr_func = bnx_msi;
2190 		}
2191 	} else {
2192 		intr_func = bnx_intr_legacy;
2193 	}
2194 	error = bus_setup_intr(dev, sc->bnx_irq, INTR_MPSAFE, intr_func, sc,
2195 	    &sc->bnx_intrhand, ifp->if_serializer);
2196 	if (error) {
2197 		ether_ifdetach(ifp);
2198 		device_printf(dev, "couldn't set up irq\n");
2199 		goto fail;
2200 	}
2201 
2202 	ifp->if_cpuid = rman_get_cpuid(sc->bnx_irq);
2203 	KKASSERT(ifp->if_cpuid >= 0 && ifp->if_cpuid < ncpus);
2204 
2205 	sc->bnx_stat_cpuid = ifp->if_cpuid;
2206 	sc->bnx_intr_cpuid = ifp->if_cpuid;
2207 
2208 	return(0);
2209 fail:
2210 	bnx_detach(dev);
2211 	return(error);
2212 }
2213 
2214 static int
2215 bnx_detach(device_t dev)
2216 {
2217 	struct bnx_softc *sc = device_get_softc(dev);
2218 
2219 	if (device_is_attached(dev)) {
2220 		struct ifnet *ifp = &sc->arpcom.ac_if;
2221 
2222 		lwkt_serialize_enter(ifp->if_serializer);
2223 		bnx_stop(sc);
2224 		bnx_reset(sc);
2225 		bus_teardown_intr(dev, sc->bnx_irq, sc->bnx_intrhand);
2226 		lwkt_serialize_exit(ifp->if_serializer);
2227 
2228 		ether_ifdetach(ifp);
2229 	}
2230 
2231 	if (sc->bnx_flags & BNX_FLAG_TBI)
2232 		ifmedia_removeall(&sc->bnx_ifmedia);
2233 	if (sc->bnx_miibus)
2234 		device_delete_child(dev, sc->bnx_miibus);
2235 	bus_generic_detach(dev);
2236 
2237 	if (sc->bnx_irq != NULL) {
2238 		bus_release_resource(dev, SYS_RES_IRQ, sc->bnx_irq_rid,
2239 		    sc->bnx_irq);
2240 	}
2241 	if (sc->bnx_irq_type == PCI_INTR_TYPE_MSI)
2242 		pci_release_msi(dev);
2243 
2244 	if (sc->bnx_res != NULL) {
2245 		bus_release_resource(dev, SYS_RES_MEMORY,
2246 		    BGE_PCI_BAR0, sc->bnx_res);
2247 	}
2248 
2249 	if (sc->bnx_sysctl_tree != NULL)
2250 		sysctl_ctx_free(&sc->bnx_sysctl_ctx);
2251 
2252 	bnx_dma_free(sc);
2253 
2254 	return 0;
2255 }
2256 
2257 static void
2258 bnx_reset(struct bnx_softc *sc)
2259 {
2260 	device_t dev;
2261 	uint32_t cachesize, command, pcistate, reset;
2262 	void (*write_op)(struct bnx_softc *, uint32_t, uint32_t);
2263 	int i, val = 0;
2264 	uint16_t devctl;
2265 
2266 	dev = sc->bnx_dev;
2267 
2268 	if (sc->bnx_asicrev != BGE_ASICREV_BCM5906)
2269 		write_op = bnx_writemem_direct;
2270 	else
2271 		write_op = bnx_writereg_ind;
2272 
2273 	/* Save some important PCI state. */
2274 	cachesize = pci_read_config(dev, BGE_PCI_CACHESZ, 4);
2275 	command = pci_read_config(dev, BGE_PCI_CMD, 4);
2276 	pcistate = pci_read_config(dev, BGE_PCI_PCISTATE, 4);
2277 
2278 	pci_write_config(dev, BGE_PCI_MISC_CTL,
2279 	    BGE_PCIMISCCTL_INDIRECT_ACCESS|BGE_PCIMISCCTL_MASK_PCI_INTR|
2280 	    BGE_HIF_SWAP_OPTIONS|BGE_PCIMISCCTL_PCISTATE_RW|
2281 	    BGE_PCIMISCCTL_TAGGED_STATUS, 4);
2282 
2283 	/* Disable fastboot on controllers that support it. */
2284 	if (bootverbose)
2285 		if_printf(&sc->arpcom.ac_if, "Disabling fastboot\n");
2286 	CSR_WRITE_4(sc, BGE_FASTBOOT_PC, 0x0);
2287 
2288 	/*
2289 	 * Write the magic number to SRAM at offset 0xB50.
2290 	 * When firmware finishes its initialization it will
2291 	 * write ~BGE_MAGIC_NUMBER to the same location.
2292 	 */
2293 	bnx_writemem_ind(sc, BGE_SOFTWARE_GENCOMM, BGE_MAGIC_NUMBER);
2294 
2295 	reset = BGE_MISCCFG_RESET_CORE_CLOCKS|(65<<1);
2296 
2297 	/* XXX: Broadcom Linux driver. */
2298 	/* Force PCI-E 1.0a mode */
2299 	if (!BNX_IS_57765_PLUS(sc) &&
2300 	    CSR_READ_4(sc, BGE_PCIE_PHY_TSTCTL) ==
2301 	    (BGE_PCIE_PHY_TSTCTL_PSCRAM |
2302 	     BGE_PCIE_PHY_TSTCTL_PCIE10)) {
2303 		CSR_WRITE_4(sc, BGE_PCIE_PHY_TSTCTL,
2304 		    BGE_PCIE_PHY_TSTCTL_PSCRAM);
2305 	}
2306 	if (sc->bnx_chipid != BGE_CHIPID_BCM5750_A0) {
2307 		/* Prevent PCIE link training during global reset */
2308 		CSR_WRITE_4(sc, BGE_MISC_CFG, (1<<29));
2309 		reset |= (1<<29);
2310 	}
2311 
2312 	/*
2313 	 * Set GPHY Power Down Override to leave GPHY
2314 	 * powered up in D0 uninitialized.
2315 	 */
2316 	if ((sc->bnx_flags & BNX_FLAG_CPMU) == 0)
2317 		reset |= BGE_MISCCFG_GPHY_PD_OVERRIDE;
2318 
2319 	/* Issue global reset */
2320 	write_op(sc, BGE_MISC_CFG, reset);
2321 
2322 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5906) {
2323 		uint32_t status, ctrl;
2324 
2325 		status = CSR_READ_4(sc, BGE_VCPU_STATUS);
2326 		CSR_WRITE_4(sc, BGE_VCPU_STATUS,
2327 		    status | BGE_VCPU_STATUS_DRV_RESET);
2328 		ctrl = CSR_READ_4(sc, BGE_VCPU_EXT_CTRL);
2329 		CSR_WRITE_4(sc, BGE_VCPU_EXT_CTRL,
2330 		    ctrl & ~BGE_VCPU_EXT_CTRL_HALT_CPU);
2331 	}
2332 
2333 	DELAY(1000);
2334 
2335 	/* XXX: Broadcom Linux driver. */
2336 	if (sc->bnx_chipid == BGE_CHIPID_BCM5750_A0) {
2337 		uint32_t v;
2338 
2339 		DELAY(500000); /* wait for link training to complete */
2340 		v = pci_read_config(dev, 0xc4, 4);
2341 		pci_write_config(dev, 0xc4, v | (1<<15), 4);
2342 	}
2343 
2344 	devctl = pci_read_config(dev, sc->bnx_pciecap + PCIER_DEVCTRL, 2);
2345 
2346 	/* Disable no snoop and disable relaxed ordering. */
2347 	devctl &= ~(PCIEM_DEVCTL_RELAX_ORDER | PCIEM_DEVCTL_NOSNOOP);
2348 
2349 	/* Old PCI-E chips only support 128 bytes Max PayLoad Size. */
2350 	if ((sc->bnx_flags & BNX_FLAG_CPMU) == 0) {
2351 		devctl &= ~PCIEM_DEVCTL_MAX_PAYLOAD_MASK;
2352 		devctl |= PCIEM_DEVCTL_MAX_PAYLOAD_128;
2353 	}
2354 
2355 	pci_write_config(dev, sc->bnx_pciecap + PCIER_DEVCTRL,
2356 	    devctl, 2);
2357 
2358 	/* Clear error status. */
2359 	pci_write_config(dev, sc->bnx_pciecap + PCIER_DEVSTS,
2360 	    PCIEM_DEVSTS_CORR_ERR |
2361 	    PCIEM_DEVSTS_NFATAL_ERR |
2362 	    PCIEM_DEVSTS_FATAL_ERR |
2363 	    PCIEM_DEVSTS_UNSUPP_REQ, 2);
2364 
2365 	/* Reset some of the PCI state that got zapped by reset */
2366 	pci_write_config(dev, BGE_PCI_MISC_CTL,
2367 	    BGE_PCIMISCCTL_INDIRECT_ACCESS|BGE_PCIMISCCTL_MASK_PCI_INTR|
2368 	    BGE_HIF_SWAP_OPTIONS|BGE_PCIMISCCTL_PCISTATE_RW|
2369 	    BGE_PCIMISCCTL_TAGGED_STATUS, 4);
2370 	pci_write_config(dev, BGE_PCI_CACHESZ, cachesize, 4);
2371 	pci_write_config(dev, BGE_PCI_CMD, command, 4);
2372 	write_op(sc, BGE_MISC_CFG, (65 << 1));
2373 
2374 	/* Enable memory arbiter */
2375 	CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE);
2376 
2377 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5906) {
2378 		for (i = 0; i < BNX_TIMEOUT; i++) {
2379 			val = CSR_READ_4(sc, BGE_VCPU_STATUS);
2380 			if (val & BGE_VCPU_STATUS_INIT_DONE)
2381 				break;
2382 			DELAY(100);
2383 		}
2384 		if (i == BNX_TIMEOUT) {
2385 			if_printf(&sc->arpcom.ac_if, "reset timed out\n");
2386 			return;
2387 		}
2388 	} else {
2389 		/*
2390 		 * Poll until we see the 1's complement of the magic number.
2391 		 * This indicates that the firmware initialization
2392 		 * is complete.
2393 		 */
2394 		for (i = 0; i < BNX_FIRMWARE_TIMEOUT; i++) {
2395 			val = bnx_readmem_ind(sc, BGE_SOFTWARE_GENCOMM);
2396 			if (val == ~BGE_MAGIC_NUMBER)
2397 				break;
2398 			DELAY(10);
2399 		}
2400 		if (i == BNX_FIRMWARE_TIMEOUT) {
2401 			if_printf(&sc->arpcom.ac_if, "firmware handshake "
2402 				  "timed out, found 0x%08x\n", val);
2403 		}
2404 
2405 		/* BCM57765 A0 needs additional time before accessing. */
2406 		if (sc->bnx_chipid == BGE_CHIPID_BCM57765_A0)
2407 			DELAY(10 * 1000);
2408 	}
2409 
2410 	/*
2411 	 * XXX Wait for the value of the PCISTATE register to
2412 	 * return to its original pre-reset state. This is a
2413 	 * fairly good indicator of reset completion. If we don't
2414 	 * wait for the reset to fully complete, trying to read
2415 	 * from the device's non-PCI registers may yield garbage
2416 	 * results.
2417 	 */
2418 	for (i = 0; i < BNX_TIMEOUT; i++) {
2419 		if (pci_read_config(dev, BGE_PCI_PCISTATE, 4) == pcistate)
2420 			break;
2421 		DELAY(10);
2422 	}
2423 
2424 	/* Fix up byte swapping */
2425 	CSR_WRITE_4(sc, BGE_MODE_CTL, bnx_dma_swap_options(sc));
2426 
2427 	CSR_WRITE_4(sc, BGE_MAC_MODE, 0);
2428 
2429 	/*
2430 	 * The 5704 in TBI mode apparently needs some special
2431 	 * adjustment to insure the SERDES drive level is set
2432 	 * to 1.2V.
2433 	 */
2434 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5704 &&
2435 	    (sc->bnx_flags & BNX_FLAG_TBI)) {
2436 		uint32_t serdescfg;
2437 
2438 		serdescfg = CSR_READ_4(sc, BGE_SERDES_CFG);
2439 		serdescfg = (serdescfg & ~0xFFF) | 0x880;
2440 		CSR_WRITE_4(sc, BGE_SERDES_CFG, serdescfg);
2441 	}
2442 
2443 	/* XXX: Broadcom Linux driver. */
2444 	if (!BNX_IS_57765_PLUS(sc)) {
2445 		uint32_t v;
2446 
2447 		/* Enable Data FIFO protection. */
2448 		v = CSR_READ_4(sc, BGE_PCIE_TLDLPL_PORT);
2449 		CSR_WRITE_4(sc, BGE_PCIE_TLDLPL_PORT, v | (1 << 25));
2450 	}
2451 
2452 	DELAY(10000);
2453 
2454 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5720) {
2455 		BNX_CLRBIT(sc, BGE_CPMU_CLCK_ORIDE,
2456 		    CPMU_CLCK_ORIDE_MAC_ORIDE_EN);
2457 	}
2458 }
2459 
2460 /*
2461  * Frame reception handling. This is called if there's a frame
2462  * on the receive return list.
2463  *
2464  * Note: we have to be able to handle two possibilities here:
2465  * 1) the frame is from the jumbo recieve ring
2466  * 2) the frame is from the standard receive ring
2467  */
2468 
2469 static void
2470 bnx_rxeof(struct bnx_softc *sc, uint16_t rx_prod)
2471 {
2472 	struct ifnet *ifp;
2473 	int stdcnt = 0, jumbocnt = 0;
2474 
2475 	ifp = &sc->arpcom.ac_if;
2476 
2477 	while (sc->bnx_rx_saved_considx != rx_prod) {
2478 		struct bge_rx_bd	*cur_rx;
2479 		uint32_t		rxidx;
2480 		struct mbuf		*m = NULL;
2481 		uint16_t		vlan_tag = 0;
2482 		int			have_tag = 0;
2483 
2484 		cur_rx =
2485 	    &sc->bnx_ldata.bnx_rx_return_ring[sc->bnx_rx_saved_considx];
2486 
2487 		rxidx = cur_rx->bge_idx;
2488 		BNX_INC(sc->bnx_rx_saved_considx, sc->bnx_return_ring_cnt);
2489 
2490 		if (cur_rx->bge_flags & BGE_RXBDFLAG_VLAN_TAG) {
2491 			have_tag = 1;
2492 			vlan_tag = cur_rx->bge_vlan_tag;
2493 		}
2494 
2495 		if (cur_rx->bge_flags & BGE_RXBDFLAG_JUMBO_RING) {
2496 			BNX_INC(sc->bnx_jumbo, BGE_JUMBO_RX_RING_CNT);
2497 			jumbocnt++;
2498 
2499 			if (rxidx != sc->bnx_jumbo) {
2500 				ifp->if_ierrors++;
2501 				if_printf(ifp, "sw jumbo index(%d) "
2502 				    "and hw jumbo index(%d) mismatch, drop!\n",
2503 				    sc->bnx_jumbo, rxidx);
2504 				bnx_setup_rxdesc_jumbo(sc, rxidx);
2505 				continue;
2506 			}
2507 
2508 			m = sc->bnx_cdata.bnx_rx_jumbo_chain[rxidx].bnx_mbuf;
2509 			if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
2510 				ifp->if_ierrors++;
2511 				bnx_setup_rxdesc_jumbo(sc, sc->bnx_jumbo);
2512 				continue;
2513 			}
2514 			if (bnx_newbuf_jumbo(sc, sc->bnx_jumbo, 0)) {
2515 				ifp->if_ierrors++;
2516 				bnx_setup_rxdesc_jumbo(sc, sc->bnx_jumbo);
2517 				continue;
2518 			}
2519 		} else {
2520 			BNX_INC(sc->bnx_std, BGE_STD_RX_RING_CNT);
2521 			stdcnt++;
2522 
2523 			if (rxidx != sc->bnx_std) {
2524 				ifp->if_ierrors++;
2525 				if_printf(ifp, "sw std index(%d) "
2526 				    "and hw std index(%d) mismatch, drop!\n",
2527 				    sc->bnx_std, rxidx);
2528 				bnx_setup_rxdesc_std(sc, rxidx);
2529 				continue;
2530 			}
2531 
2532 			m = sc->bnx_cdata.bnx_rx_std_chain[rxidx].bnx_mbuf;
2533 			if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
2534 				ifp->if_ierrors++;
2535 				bnx_setup_rxdesc_std(sc, sc->bnx_std);
2536 				continue;
2537 			}
2538 			if (bnx_newbuf_std(sc, sc->bnx_std, 0)) {
2539 				ifp->if_ierrors++;
2540 				bnx_setup_rxdesc_std(sc, sc->bnx_std);
2541 				continue;
2542 			}
2543 		}
2544 
2545 		ifp->if_ipackets++;
2546 		m->m_pkthdr.len = m->m_len = cur_rx->bge_len - ETHER_CRC_LEN;
2547 		m->m_pkthdr.rcvif = ifp;
2548 
2549 		if ((ifp->if_capenable & IFCAP_RXCSUM) &&
2550 		    (cur_rx->bge_flags & BGE_RXBDFLAG_IPV6) == 0) {
2551 			if (cur_rx->bge_flags & BGE_RXBDFLAG_IP_CSUM) {
2552 				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
2553 				if ((cur_rx->bge_error_flag &
2554 				    BGE_RXERRFLAG_IP_CSUM_NOK) == 0)
2555 					m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
2556 			}
2557 			if (cur_rx->bge_flags & BGE_RXBDFLAG_TCP_UDP_CSUM) {
2558 				m->m_pkthdr.csum_data =
2559 				    cur_rx->bge_tcp_udp_csum;
2560 				m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
2561 				    CSUM_PSEUDO_HDR;
2562 			}
2563 		}
2564 
2565 		/*
2566 		 * If we received a packet with a vlan tag, pass it
2567 		 * to vlan_input() instead of ether_input().
2568 		 */
2569 		if (have_tag) {
2570 			m->m_flags |= M_VLANTAG;
2571 			m->m_pkthdr.ether_vlantag = vlan_tag;
2572 			have_tag = vlan_tag = 0;
2573 		}
2574 		ifp->if_input(ifp, m);
2575 	}
2576 
2577 	bnx_writembx(sc, BGE_MBX_RX_CONS0_LO, sc->bnx_rx_saved_considx);
2578 	if (stdcnt)
2579 		bnx_writembx(sc, BGE_MBX_RX_STD_PROD_LO, sc->bnx_std);
2580 	if (jumbocnt)
2581 		bnx_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, sc->bnx_jumbo);
2582 }
2583 
2584 static void
2585 bnx_txeof(struct bnx_softc *sc, uint16_t tx_cons)
2586 {
2587 	struct bge_tx_bd *cur_tx = NULL;
2588 	struct ifnet *ifp;
2589 
2590 	ifp = &sc->arpcom.ac_if;
2591 
2592 	/*
2593 	 * Go through our tx ring and free mbufs for those
2594 	 * frames that have been sent.
2595 	 */
2596 	while (sc->bnx_tx_saved_considx != tx_cons) {
2597 		uint32_t idx = 0;
2598 
2599 		idx = sc->bnx_tx_saved_considx;
2600 		cur_tx = &sc->bnx_ldata.bnx_tx_ring[idx];
2601 		if (cur_tx->bge_flags & BGE_TXBDFLAG_END)
2602 			ifp->if_opackets++;
2603 		if (sc->bnx_cdata.bnx_tx_chain[idx] != NULL) {
2604 			bus_dmamap_unload(sc->bnx_cdata.bnx_tx_mtag,
2605 			    sc->bnx_cdata.bnx_tx_dmamap[idx]);
2606 			m_freem(sc->bnx_cdata.bnx_tx_chain[idx]);
2607 			sc->bnx_cdata.bnx_tx_chain[idx] = NULL;
2608 		}
2609 		sc->bnx_txcnt--;
2610 		BNX_INC(sc->bnx_tx_saved_considx, BGE_TX_RING_CNT);
2611 	}
2612 
2613 	if (cur_tx != NULL &&
2614 	    (BGE_TX_RING_CNT - sc->bnx_txcnt) >=
2615 	    (BNX_NSEG_RSVD + BNX_NSEG_SPARE))
2616 		ifp->if_flags &= ~IFF_OACTIVE;
2617 
2618 	if (sc->bnx_txcnt == 0)
2619 		ifp->if_timer = 0;
2620 
2621 	if (!ifq_is_empty(&ifp->if_snd))
2622 		if_devstart(ifp);
2623 }
2624 
2625 #ifdef DEVICE_POLLING
2626 
2627 static void
2628 bnx_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
2629 {
2630 	struct bnx_softc *sc = ifp->if_softc;
2631 	struct bge_status_block *sblk = sc->bnx_ldata.bnx_status_block;
2632 	uint16_t rx_prod, tx_cons;
2633 
2634 	switch(cmd) {
2635 	case POLL_REGISTER:
2636 		bnx_disable_intr(sc);
2637 		break;
2638 	case POLL_DEREGISTER:
2639 		bnx_enable_intr(sc);
2640 		break;
2641 	case POLL_AND_CHECK_STATUS:
2642 		/*
2643 		 * Process link state changes.
2644 		 */
2645 		bnx_link_poll(sc);
2646 		/* Fall through */
2647 	case POLL_ONLY:
2648 		sc->bnx_status_tag = sblk->bge_status_tag;
2649 		/*
2650 		 * Use a load fence to ensure that status_tag
2651 		 * is saved  before rx_prod and tx_cons.
2652 		 */
2653 		cpu_lfence();
2654 
2655 		rx_prod = sblk->bge_idx[0].bge_rx_prod_idx;
2656 		tx_cons = sblk->bge_idx[0].bge_tx_cons_idx;
2657 		if (ifp->if_flags & IFF_RUNNING) {
2658 			rx_prod = sblk->bge_idx[0].bge_rx_prod_idx;
2659 			if (sc->bnx_rx_saved_considx != rx_prod)
2660 				bnx_rxeof(sc, rx_prod);
2661 
2662 			tx_cons = sblk->bge_idx[0].bge_tx_cons_idx;
2663 			if (sc->bnx_tx_saved_considx != tx_cons)
2664 				bnx_txeof(sc, tx_cons);
2665 		}
2666 		break;
2667 	}
2668 }
2669 
2670 #endif
2671 
2672 static void
2673 bnx_intr_legacy(void *xsc)
2674 {
2675 	struct bnx_softc *sc = xsc;
2676 	struct bge_status_block *sblk = sc->bnx_ldata.bnx_status_block;
2677 
2678 	if (sc->bnx_status_tag == sblk->bge_status_tag) {
2679 		uint32_t val;
2680 
2681 		val = pci_read_config(sc->bnx_dev, BGE_PCI_PCISTATE, 4);
2682 		if (val & BGE_PCISTAT_INTR_NOTACT)
2683 			return;
2684 	}
2685 
2686 	/*
2687 	 * NOTE:
2688 	 * Interrupt will have to be disabled if tagged status
2689 	 * is used, else interrupt will always be asserted on
2690 	 * certain chips (at least on BCM5750 AX/BX).
2691 	 */
2692 	bnx_writembx(sc, BGE_MBX_IRQ0_LO, 1);
2693 
2694 	bnx_intr(sc);
2695 }
2696 
2697 static void
2698 bnx_msi(void *xsc)
2699 {
2700 	struct bnx_softc *sc = xsc;
2701 
2702 	/* Disable interrupt first */
2703 	bnx_writembx(sc, BGE_MBX_IRQ0_LO, 1);
2704 	bnx_intr(sc);
2705 }
2706 
2707 static void
2708 bnx_msi_oneshot(void *xsc)
2709 {
2710 	bnx_intr(xsc);
2711 }
2712 
2713 static void
2714 bnx_intr(struct bnx_softc *sc)
2715 {
2716 	struct ifnet *ifp = &sc->arpcom.ac_if;
2717 	struct bge_status_block *sblk = sc->bnx_ldata.bnx_status_block;
2718 	uint16_t rx_prod, tx_cons;
2719 	uint32_t status;
2720 
2721 	sc->bnx_status_tag = sblk->bge_status_tag;
2722 	/*
2723 	 * Use a load fence to ensure that status_tag is saved
2724 	 * before rx_prod, tx_cons and status.
2725 	 */
2726 	cpu_lfence();
2727 
2728 	rx_prod = sblk->bge_idx[0].bge_rx_prod_idx;
2729 	tx_cons = sblk->bge_idx[0].bge_tx_cons_idx;
2730 	status = sblk->bge_status;
2731 
2732 	if ((status & BGE_STATFLAG_LINKSTATE_CHANGED) || sc->bnx_link_evt)
2733 		bnx_link_poll(sc);
2734 
2735 	if (ifp->if_flags & IFF_RUNNING) {
2736 		if (sc->bnx_rx_saved_considx != rx_prod)
2737 			bnx_rxeof(sc, rx_prod);
2738 
2739 		if (sc->bnx_tx_saved_considx != tx_cons)
2740 			bnx_txeof(sc, tx_cons);
2741 	}
2742 
2743 	bnx_writembx(sc, BGE_MBX_IRQ0_LO, sc->bnx_status_tag << 24);
2744 
2745 	if (sc->bnx_coal_chg)
2746 		bnx_coal_change(sc);
2747 }
2748 
2749 static void
2750 bnx_tick(void *xsc)
2751 {
2752 	struct bnx_softc *sc = xsc;
2753 	struct ifnet *ifp = &sc->arpcom.ac_if;
2754 
2755 	lwkt_serialize_enter(ifp->if_serializer);
2756 
2757 	KKASSERT(mycpuid == sc->bnx_stat_cpuid);
2758 
2759 	bnx_stats_update_regs(sc);
2760 
2761 	if (sc->bnx_flags & BNX_FLAG_TBI) {
2762 		/*
2763 		 * Since in TBI mode auto-polling can't be used we should poll
2764 		 * link status manually. Here we register pending link event
2765 		 * and trigger interrupt.
2766 		 */
2767 		sc->bnx_link_evt++;
2768 		BNX_SETBIT(sc, BGE_HCC_MODE, BGE_HCCMODE_COAL_NOW);
2769 	} else if (!sc->bnx_link) {
2770 		mii_tick(device_get_softc(sc->bnx_miibus));
2771 	}
2772 
2773 	callout_reset(&sc->bnx_stat_timer, hz, bnx_tick, sc);
2774 
2775 	lwkt_serialize_exit(ifp->if_serializer);
2776 }
2777 
2778 static void
2779 bnx_stats_update_regs(struct bnx_softc *sc)
2780 {
2781 	struct ifnet *ifp = &sc->arpcom.ac_if;
2782 	struct bge_mac_stats_regs stats;
2783 	uint32_t *s;
2784 	int i;
2785 
2786 	s = (uint32_t *)&stats;
2787 	for (i = 0; i < sizeof(struct bge_mac_stats_regs); i += 4) {
2788 		*s = CSR_READ_4(sc, BGE_RX_STATS + i);
2789 		s++;
2790 	}
2791 
2792 	ifp->if_collisions +=
2793 	   (stats.dot3StatsSingleCollisionFrames +
2794 	   stats.dot3StatsMultipleCollisionFrames +
2795 	   stats.dot3StatsExcessiveCollisions +
2796 	   stats.dot3StatsLateCollisions) -
2797 	   ifp->if_collisions;
2798 }
2799 
2800 /*
2801  * Encapsulate an mbuf chain in the tx ring  by coupling the mbuf data
2802  * pointers to descriptors.
2803  */
2804 static int
2805 bnx_encap(struct bnx_softc *sc, struct mbuf **m_head0, uint32_t *txidx)
2806 {
2807 	struct bge_tx_bd *d = NULL;
2808 	uint16_t csum_flags = 0, vlan_tag = 0, mss = 0;
2809 	bus_dma_segment_t segs[BNX_NSEG_NEW];
2810 	bus_dmamap_t map;
2811 	int error, maxsegs, nsegs, idx, i;
2812 	struct mbuf *m_head = *m_head0, *m_new;
2813 
2814 	if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
2815 #ifdef BNX_TSO_DEBUG
2816 		int tso_nsegs;
2817 #endif
2818 
2819 		error = bnx_setup_tso(sc, m_head0, &mss, &csum_flags);
2820 		if (error)
2821 			return error;
2822 		m_head = *m_head0;
2823 
2824 #ifdef BNX_TSO_DEBUG
2825 		tso_nsegs = (m_head->m_pkthdr.len /
2826 		    m_head->m_pkthdr.tso_segsz) - 1;
2827 		if (tso_nsegs > (BNX_TSO_NSTATS - 1))
2828 			tso_nsegs = BNX_TSO_NSTATS - 1;
2829 		else if (tso_nsegs < 0)
2830 			tso_nsegs = 0;
2831 		sc->bnx_tsosegs[tso_nsegs]++;
2832 #endif
2833 	} else if (m_head->m_pkthdr.csum_flags & BNX_CSUM_FEATURES) {
2834 		if (m_head->m_pkthdr.csum_flags & CSUM_IP)
2835 			csum_flags |= BGE_TXBDFLAG_IP_CSUM;
2836 		if (m_head->m_pkthdr.csum_flags & (CSUM_TCP | CSUM_UDP))
2837 			csum_flags |= BGE_TXBDFLAG_TCP_UDP_CSUM;
2838 		if (m_head->m_flags & M_LASTFRAG)
2839 			csum_flags |= BGE_TXBDFLAG_IP_FRAG_END;
2840 		else if (m_head->m_flags & M_FRAG)
2841 			csum_flags |= BGE_TXBDFLAG_IP_FRAG;
2842 	}
2843 	if (m_head->m_flags & M_VLANTAG) {
2844 		csum_flags |= BGE_TXBDFLAG_VLAN_TAG;
2845 		vlan_tag = m_head->m_pkthdr.ether_vlantag;
2846 	}
2847 
2848 	idx = *txidx;
2849 	map = sc->bnx_cdata.bnx_tx_dmamap[idx];
2850 
2851 	maxsegs = (BGE_TX_RING_CNT - sc->bnx_txcnt) - BNX_NSEG_RSVD;
2852 	KASSERT(maxsegs >= BNX_NSEG_SPARE,
2853 		("not enough segments %d", maxsegs));
2854 
2855 	if (maxsegs > BNX_NSEG_NEW)
2856 		maxsegs = BNX_NSEG_NEW;
2857 
2858 	/*
2859 	 * Pad outbound frame to BGE_MIN_FRAMELEN for an unusual reason.
2860 	 * The bge hardware will pad out Tx runts to BGE_MIN_FRAMELEN,
2861 	 * but when such padded frames employ the bge IP/TCP checksum
2862 	 * offload, the hardware checksum assist gives incorrect results
2863 	 * (possibly from incorporating its own padding into the UDP/TCP
2864 	 * checksum; who knows).  If we pad such runts with zeros, the
2865 	 * onboard checksum comes out correct.
2866 	 */
2867 	if ((csum_flags & BGE_TXBDFLAG_TCP_UDP_CSUM) &&
2868 	    m_head->m_pkthdr.len < BNX_MIN_FRAMELEN) {
2869 		error = m_devpad(m_head, BNX_MIN_FRAMELEN);
2870 		if (error)
2871 			goto back;
2872 	}
2873 
2874 	if ((sc->bnx_flags & BNX_FLAG_SHORTDMA) && m_head->m_next != NULL) {
2875 		m_new = bnx_defrag_shortdma(m_head);
2876 		if (m_new == NULL) {
2877 			error = ENOBUFS;
2878 			goto back;
2879 		}
2880 		*m_head0 = m_head = m_new;
2881 	}
2882 	if ((m_head->m_pkthdr.csum_flags & CSUM_TSO) == 0 &&
2883 	    sc->bnx_force_defrag && m_head->m_next != NULL) {
2884 		/*
2885 		 * Forcefully defragment mbuf chain to overcome hardware
2886 		 * limitation which only support a single outstanding
2887 		 * DMA read operation.  If it fails, keep moving on using
2888 		 * the original mbuf chain.
2889 		 */
2890 		m_new = m_defrag(m_head, MB_DONTWAIT);
2891 		if (m_new != NULL)
2892 			*m_head0 = m_head = m_new;
2893 	}
2894 
2895 	error = bus_dmamap_load_mbuf_defrag(sc->bnx_cdata.bnx_tx_mtag, map,
2896 			m_head0, segs, maxsegs, &nsegs, BUS_DMA_NOWAIT);
2897 	if (error)
2898 		goto back;
2899 
2900 	m_head = *m_head0;
2901 	bus_dmamap_sync(sc->bnx_cdata.bnx_tx_mtag, map, BUS_DMASYNC_PREWRITE);
2902 
2903 	for (i = 0; ; i++) {
2904 		d = &sc->bnx_ldata.bnx_tx_ring[idx];
2905 
2906 		d->bge_addr.bge_addr_lo = BGE_ADDR_LO(segs[i].ds_addr);
2907 		d->bge_addr.bge_addr_hi = BGE_ADDR_HI(segs[i].ds_addr);
2908 		d->bge_len = segs[i].ds_len;
2909 		d->bge_flags = csum_flags;
2910 		d->bge_vlan_tag = vlan_tag;
2911 		d->bge_mss = mss;
2912 
2913 		if (i == nsegs - 1)
2914 			break;
2915 		BNX_INC(idx, BGE_TX_RING_CNT);
2916 	}
2917 	/* Mark the last segment as end of packet... */
2918 	d->bge_flags |= BGE_TXBDFLAG_END;
2919 
2920 	/*
2921 	 * Insure that the map for this transmission is placed at
2922 	 * the array index of the last descriptor in this chain.
2923 	 */
2924 	sc->bnx_cdata.bnx_tx_dmamap[*txidx] = sc->bnx_cdata.bnx_tx_dmamap[idx];
2925 	sc->bnx_cdata.bnx_tx_dmamap[idx] = map;
2926 	sc->bnx_cdata.bnx_tx_chain[idx] = m_head;
2927 	sc->bnx_txcnt += nsegs;
2928 
2929 	BNX_INC(idx, BGE_TX_RING_CNT);
2930 	*txidx = idx;
2931 back:
2932 	if (error) {
2933 		m_freem(*m_head0);
2934 		*m_head0 = NULL;
2935 	}
2936 	return error;
2937 }
2938 
2939 /*
2940  * Main transmit routine. To avoid having to do mbuf copies, we put pointers
2941  * to the mbuf data regions directly in the transmit descriptors.
2942  */
2943 static void
2944 bnx_start(struct ifnet *ifp)
2945 {
2946 	struct bnx_softc *sc = ifp->if_softc;
2947 	struct mbuf *m_head = NULL;
2948 	uint32_t prodidx;
2949 	int need_trans;
2950 
2951 	if ((ifp->if_flags & (IFF_RUNNING | IFF_OACTIVE)) != IFF_RUNNING)
2952 		return;
2953 
2954 	prodidx = sc->bnx_tx_prodidx;
2955 
2956 	need_trans = 0;
2957 	while (sc->bnx_cdata.bnx_tx_chain[prodidx] == NULL) {
2958 		m_head = ifq_dequeue(&ifp->if_snd, NULL);
2959 		if (m_head == NULL)
2960 			break;
2961 
2962 		/*
2963 		 * XXX
2964 		 * The code inside the if() block is never reached since we
2965 		 * must mark CSUM_IP_FRAGS in our if_hwassist to start getting
2966 		 * requests to checksum TCP/UDP in a fragmented packet.
2967 		 *
2968 		 * XXX
2969 		 * safety overkill.  If this is a fragmented packet chain
2970 		 * with delayed TCP/UDP checksums, then only encapsulate
2971 		 * it if we have enough descriptors to handle the entire
2972 		 * chain at once.
2973 		 * (paranoia -- may not actually be needed)
2974 		 */
2975 		if ((m_head->m_flags & M_FIRSTFRAG) &&
2976 		    (m_head->m_pkthdr.csum_flags & CSUM_DELAY_DATA)) {
2977 			if ((BGE_TX_RING_CNT - sc->bnx_txcnt) <
2978 			    m_head->m_pkthdr.csum_data + BNX_NSEG_RSVD) {
2979 				ifp->if_flags |= IFF_OACTIVE;
2980 				ifq_prepend(&ifp->if_snd, m_head);
2981 				break;
2982 			}
2983 		}
2984 
2985 		/*
2986 		 * Sanity check: avoid coming within BGE_NSEG_RSVD
2987 		 * descriptors of the end of the ring.  Also make
2988 		 * sure there are BGE_NSEG_SPARE descriptors for
2989 		 * jumbo buffers' defragmentation.
2990 		 */
2991 		if ((BGE_TX_RING_CNT - sc->bnx_txcnt) <
2992 		    (BNX_NSEG_RSVD + BNX_NSEG_SPARE)) {
2993 			ifp->if_flags |= IFF_OACTIVE;
2994 			ifq_prepend(&ifp->if_snd, m_head);
2995 			break;
2996 		}
2997 
2998 		/*
2999 		 * Pack the data into the transmit ring. If we
3000 		 * don't have room, set the OACTIVE flag and wait
3001 		 * for the NIC to drain the ring.
3002 		 */
3003 		if (bnx_encap(sc, &m_head, &prodidx)) {
3004 			ifp->if_flags |= IFF_OACTIVE;
3005 			ifp->if_oerrors++;
3006 			break;
3007 		}
3008 		need_trans = 1;
3009 
3010 		ETHER_BPF_MTAP(ifp, m_head);
3011 	}
3012 
3013 	if (!need_trans)
3014 		return;
3015 
3016 	/* Transmit */
3017 	bnx_writembx(sc, BGE_MBX_TX_HOST_PROD0_LO, prodidx);
3018 
3019 	sc->bnx_tx_prodidx = prodidx;
3020 
3021 	/*
3022 	 * Set a timeout in case the chip goes out to lunch.
3023 	 */
3024 	ifp->if_timer = 5;
3025 }
3026 
3027 static void
3028 bnx_init(void *xsc)
3029 {
3030 	struct bnx_softc *sc = xsc;
3031 	struct ifnet *ifp = &sc->arpcom.ac_if;
3032 	uint16_t *m;
3033 	uint32_t mode;
3034 
3035 	ASSERT_SERIALIZED(ifp->if_serializer);
3036 
3037 	/* Cancel pending I/O and flush buffers. */
3038 	bnx_stop(sc);
3039 	bnx_reset(sc);
3040 	bnx_chipinit(sc);
3041 
3042 	/*
3043 	 * Init the various state machines, ring
3044 	 * control blocks and firmware.
3045 	 */
3046 	if (bnx_blockinit(sc)) {
3047 		if_printf(ifp, "initialization failure\n");
3048 		bnx_stop(sc);
3049 		return;
3050 	}
3051 
3052 	/* Specify MTU. */
3053 	CSR_WRITE_4(sc, BGE_RX_MTU, ifp->if_mtu +
3054 	    ETHER_HDR_LEN + ETHER_CRC_LEN + EVL_ENCAPLEN);
3055 
3056 	/* Load our MAC address. */
3057 	m = (uint16_t *)&sc->arpcom.ac_enaddr[0];
3058 	CSR_WRITE_4(sc, BGE_MAC_ADDR1_LO, htons(m[0]));
3059 	CSR_WRITE_4(sc, BGE_MAC_ADDR1_HI, (htons(m[1]) << 16) | htons(m[2]));
3060 
3061 	/* Enable or disable promiscuous mode as needed. */
3062 	bnx_setpromisc(sc);
3063 
3064 	/* Program multicast filter. */
3065 	bnx_setmulti(sc);
3066 
3067 	/* Init RX ring. */
3068 	if (bnx_init_rx_ring_std(sc)) {
3069 		if_printf(ifp, "RX ring initialization failed\n");
3070 		bnx_stop(sc);
3071 		return;
3072 	}
3073 
3074 	/* Init jumbo RX ring. */
3075 	if (ifp->if_mtu > (ETHERMTU + ETHER_HDR_LEN + ETHER_CRC_LEN)) {
3076 		if (bnx_init_rx_ring_jumbo(sc)) {
3077 			if_printf(ifp, "Jumbo RX ring initialization failed\n");
3078 			bnx_stop(sc);
3079 			return;
3080 		}
3081 	}
3082 
3083 	/* Init our RX return ring index */
3084 	sc->bnx_rx_saved_considx = 0;
3085 
3086 	/* Init TX ring. */
3087 	bnx_init_tx_ring(sc);
3088 
3089 	/* Enable TX MAC state machine lockup fix. */
3090 	mode = CSR_READ_4(sc, BGE_TX_MODE);
3091 	mode |= BGE_TXMODE_MBUF_LOCKUP_FIX;
3092 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5720) {
3093 		mode &= ~(BGE_TXMODE_JMB_FRM_LEN | BGE_TXMODE_CNT_DN_MODE);
3094 		mode |= CSR_READ_4(sc, BGE_TX_MODE) &
3095 		    (BGE_TXMODE_JMB_FRM_LEN | BGE_TXMODE_CNT_DN_MODE);
3096 	}
3097 	/* Turn on transmitter */
3098 	CSR_WRITE_4(sc, BGE_TX_MODE, mode | BGE_TXMODE_ENABLE);
3099 
3100 	/* Turn on receiver */
3101 	BNX_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_ENABLE);
3102 
3103 	/*
3104 	 * Set the number of good frames to receive after RX MBUF
3105 	 * Low Watermark has been reached.  After the RX MAC receives
3106 	 * this number of frames, it will drop subsequent incoming
3107 	 * frames until the MBUF High Watermark is reached.
3108 	 */
3109 	if (BNX_IS_57765_FAMILY(sc))
3110 		CSR_WRITE_4(sc, BGE_MAX_RX_FRAME_LOWAT, 1);
3111 	else
3112 		CSR_WRITE_4(sc, BGE_MAX_RX_FRAME_LOWAT, 2);
3113 
3114 	if (sc->bnx_irq_type == PCI_INTR_TYPE_MSI) {
3115 		if (bootverbose) {
3116 			if_printf(ifp, "MSI_MODE: %#x\n",
3117 			    CSR_READ_4(sc, BGE_MSI_MODE));
3118 		}
3119 	}
3120 
3121 	/* Tell firmware we're alive. */
3122 	BNX_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
3123 
3124 	/* Enable host interrupts if polling(4) is not enabled. */
3125 	PCI_SETBIT(sc->bnx_dev, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_CLEAR_INTA, 4);
3126 #ifdef DEVICE_POLLING
3127 	if (ifp->if_flags & IFF_POLLING)
3128 		bnx_disable_intr(sc);
3129 	else
3130 #endif
3131 	bnx_enable_intr(sc);
3132 
3133 	bnx_ifmedia_upd(ifp);
3134 
3135 	ifp->if_flags |= IFF_RUNNING;
3136 	ifp->if_flags &= ~IFF_OACTIVE;
3137 
3138 	callout_reset_bycpu(&sc->bnx_stat_timer, hz, bnx_tick, sc,
3139 	    sc->bnx_stat_cpuid);
3140 }
3141 
3142 /*
3143  * Set media options.
3144  */
3145 static int
3146 bnx_ifmedia_upd(struct ifnet *ifp)
3147 {
3148 	struct bnx_softc *sc = ifp->if_softc;
3149 
3150 	/* If this is a 1000baseX NIC, enable the TBI port. */
3151 	if (sc->bnx_flags & BNX_FLAG_TBI) {
3152 		struct ifmedia *ifm = &sc->bnx_ifmedia;
3153 
3154 		if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
3155 			return(EINVAL);
3156 
3157 		switch(IFM_SUBTYPE(ifm->ifm_media)) {
3158 		case IFM_AUTO:
3159 			break;
3160 
3161 		case IFM_1000_SX:
3162 			if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX) {
3163 				BNX_CLRBIT(sc, BGE_MAC_MODE,
3164 				    BGE_MACMODE_HALF_DUPLEX);
3165 			} else {
3166 				BNX_SETBIT(sc, BGE_MAC_MODE,
3167 				    BGE_MACMODE_HALF_DUPLEX);
3168 			}
3169 			break;
3170 		default:
3171 			return(EINVAL);
3172 		}
3173 	} else {
3174 		struct mii_data *mii = device_get_softc(sc->bnx_miibus);
3175 
3176 		sc->bnx_link_evt++;
3177 		sc->bnx_link = 0;
3178 		if (mii->mii_instance) {
3179 			struct mii_softc *miisc;
3180 
3181 			LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
3182 				mii_phy_reset(miisc);
3183 		}
3184 		mii_mediachg(mii);
3185 
3186 		/*
3187 		 * Force an interrupt so that we will call bnx_link_upd
3188 		 * if needed and clear any pending link state attention.
3189 		 * Without this we are not getting any further interrupts
3190 		 * for link state changes and thus will not UP the link and
3191 		 * not be able to send in bnx_start.  The only way to get
3192 		 * things working was to receive a packet and get an RX
3193 		 * intr.
3194 		 *
3195 		 * bnx_tick should help for fiber cards and we might not
3196 		 * need to do this here if BNX_FLAG_TBI is set but as
3197 		 * we poll for fiber anyway it should not harm.
3198 		 */
3199 		BNX_SETBIT(sc, BGE_HCC_MODE, BGE_HCCMODE_COAL_NOW);
3200 	}
3201 	return(0);
3202 }
3203 
3204 /*
3205  * Report current media status.
3206  */
3207 static void
3208 bnx_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3209 {
3210 	struct bnx_softc *sc = ifp->if_softc;
3211 
3212 	if (sc->bnx_flags & BNX_FLAG_TBI) {
3213 		ifmr->ifm_status = IFM_AVALID;
3214 		ifmr->ifm_active = IFM_ETHER;
3215 		if (CSR_READ_4(sc, BGE_MAC_STS) &
3216 		    BGE_MACSTAT_TBI_PCS_SYNCHED) {
3217 			ifmr->ifm_status |= IFM_ACTIVE;
3218 		} else {
3219 			ifmr->ifm_active |= IFM_NONE;
3220 			return;
3221 		}
3222 
3223 		ifmr->ifm_active |= IFM_1000_SX;
3224 		if (CSR_READ_4(sc, BGE_MAC_MODE) & BGE_MACMODE_HALF_DUPLEX)
3225 			ifmr->ifm_active |= IFM_HDX;
3226 		else
3227 			ifmr->ifm_active |= IFM_FDX;
3228 	} else {
3229 		struct mii_data *mii = device_get_softc(sc->bnx_miibus);
3230 
3231 		mii_pollstat(mii);
3232 		ifmr->ifm_active = mii->mii_media_active;
3233 		ifmr->ifm_status = mii->mii_media_status;
3234 	}
3235 }
3236 
3237 static int
3238 bnx_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
3239 {
3240 	struct bnx_softc *sc = ifp->if_softc;
3241 	struct ifreq *ifr = (struct ifreq *)data;
3242 	int mask, error = 0;
3243 
3244 	ASSERT_SERIALIZED(ifp->if_serializer);
3245 
3246 	switch (command) {
3247 	case SIOCSIFMTU:
3248 		if ((!BNX_IS_JUMBO_CAPABLE(sc) && ifr->ifr_mtu > ETHERMTU) ||
3249 		    (BNX_IS_JUMBO_CAPABLE(sc) &&
3250 		     ifr->ifr_mtu > BNX_JUMBO_MTU)) {
3251 			error = EINVAL;
3252 		} else if (ifp->if_mtu != ifr->ifr_mtu) {
3253 			ifp->if_mtu = ifr->ifr_mtu;
3254 			if (ifp->if_flags & IFF_RUNNING)
3255 				bnx_init(sc);
3256 		}
3257 		break;
3258 	case SIOCSIFFLAGS:
3259 		if (ifp->if_flags & IFF_UP) {
3260 			if (ifp->if_flags & IFF_RUNNING) {
3261 				mask = ifp->if_flags ^ sc->bnx_if_flags;
3262 
3263 				/*
3264 				 * If only the state of the PROMISC flag
3265 				 * changed, then just use the 'set promisc
3266 				 * mode' command instead of reinitializing
3267 				 * the entire NIC. Doing a full re-init
3268 				 * means reloading the firmware and waiting
3269 				 * for it to start up, which may take a
3270 				 * second or two.  Similarly for ALLMULTI.
3271 				 */
3272 				if (mask & IFF_PROMISC)
3273 					bnx_setpromisc(sc);
3274 				if (mask & IFF_ALLMULTI)
3275 					bnx_setmulti(sc);
3276 			} else {
3277 				bnx_init(sc);
3278 			}
3279 		} else if (ifp->if_flags & IFF_RUNNING) {
3280 			bnx_stop(sc);
3281 		}
3282 		sc->bnx_if_flags = ifp->if_flags;
3283 		break;
3284 	case SIOCADDMULTI:
3285 	case SIOCDELMULTI:
3286 		if (ifp->if_flags & IFF_RUNNING)
3287 			bnx_setmulti(sc);
3288 		break;
3289 	case SIOCSIFMEDIA:
3290 	case SIOCGIFMEDIA:
3291 		if (sc->bnx_flags & BNX_FLAG_TBI) {
3292 			error = ifmedia_ioctl(ifp, ifr,
3293 			    &sc->bnx_ifmedia, command);
3294 		} else {
3295 			struct mii_data *mii;
3296 
3297 			mii = device_get_softc(sc->bnx_miibus);
3298 			error = ifmedia_ioctl(ifp, ifr,
3299 					      &mii->mii_media, command);
3300 		}
3301 		break;
3302         case SIOCSIFCAP:
3303 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
3304 		if (mask & IFCAP_HWCSUM) {
3305 			ifp->if_capenable ^= (mask & IFCAP_HWCSUM);
3306 			if (ifp->if_capenable & IFCAP_TXCSUM)
3307 				ifp->if_hwassist |= BNX_CSUM_FEATURES;
3308 			else
3309 				ifp->if_hwassist &= ~BNX_CSUM_FEATURES;
3310 		}
3311 		if (mask & IFCAP_TSO) {
3312 			ifp->if_capenable ^= (mask & IFCAP_TSO);
3313 			if (ifp->if_capenable & IFCAP_TSO)
3314 				ifp->if_hwassist |= CSUM_TSO;
3315 			else
3316 				ifp->if_hwassist &= ~CSUM_TSO;
3317 		}
3318 		break;
3319 	default:
3320 		error = ether_ioctl(ifp, command, data);
3321 		break;
3322 	}
3323 	return error;
3324 }
3325 
3326 static void
3327 bnx_watchdog(struct ifnet *ifp)
3328 {
3329 	struct bnx_softc *sc = ifp->if_softc;
3330 
3331 	if_printf(ifp, "watchdog timeout -- resetting\n");
3332 
3333 	bnx_init(sc);
3334 
3335 	ifp->if_oerrors++;
3336 
3337 	if (!ifq_is_empty(&ifp->if_snd))
3338 		if_devstart(ifp);
3339 }
3340 
3341 /*
3342  * Stop the adapter and free any mbufs allocated to the
3343  * RX and TX lists.
3344  */
3345 static void
3346 bnx_stop(struct bnx_softc *sc)
3347 {
3348 	struct ifnet *ifp = &sc->arpcom.ac_if;
3349 
3350 	ASSERT_SERIALIZED(ifp->if_serializer);
3351 
3352 	callout_stop(&sc->bnx_stat_timer);
3353 
3354 	/*
3355 	 * Disable all of the receiver blocks
3356 	 */
3357 	bnx_stop_block(sc, BGE_RX_MODE, BGE_RXMODE_ENABLE);
3358 	bnx_stop_block(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
3359 	bnx_stop_block(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
3360 	bnx_stop_block(sc, BGE_RDBDI_MODE, BGE_RBDIMODE_ENABLE);
3361 	bnx_stop_block(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
3362 	bnx_stop_block(sc, BGE_RBDC_MODE, BGE_RBDCMODE_ENABLE);
3363 
3364 	/*
3365 	 * Disable all of the transmit blocks
3366 	 */
3367 	bnx_stop_block(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
3368 	bnx_stop_block(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
3369 	bnx_stop_block(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
3370 	bnx_stop_block(sc, BGE_RDMA_MODE, BGE_RDMAMODE_ENABLE);
3371 	bnx_stop_block(sc, BGE_SDC_MODE, BGE_SDCMODE_ENABLE);
3372 	bnx_stop_block(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
3373 
3374 	/*
3375 	 * Shut down all of the memory managers and related
3376 	 * state machines.
3377 	 */
3378 	bnx_stop_block(sc, BGE_HCC_MODE, BGE_HCCMODE_ENABLE);
3379 	bnx_stop_block(sc, BGE_WDMA_MODE, BGE_WDMAMODE_ENABLE);
3380 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
3381 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
3382 
3383 	/* Disable host interrupts. */
3384 	bnx_disable_intr(sc);
3385 
3386 	/*
3387 	 * Tell firmware we're shutting down.
3388 	 */
3389 	BNX_CLRBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
3390 
3391 	/* Free the RX lists. */
3392 	bnx_free_rx_ring_std(sc);
3393 
3394 	/* Free jumbo RX list. */
3395 	if (BNX_IS_JUMBO_CAPABLE(sc))
3396 		bnx_free_rx_ring_jumbo(sc);
3397 
3398 	/* Free TX buffers. */
3399 	bnx_free_tx_ring(sc);
3400 
3401 	sc->bnx_status_tag = 0;
3402 	sc->bnx_link = 0;
3403 	sc->bnx_coal_chg = 0;
3404 
3405 	sc->bnx_tx_saved_considx = BNX_TXCONS_UNSET;
3406 
3407 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
3408 	ifp->if_timer = 0;
3409 }
3410 
3411 /*
3412  * Stop all chip I/O so that the kernel's probe routines don't
3413  * get confused by errant DMAs when rebooting.
3414  */
3415 static void
3416 bnx_shutdown(device_t dev)
3417 {
3418 	struct bnx_softc *sc = device_get_softc(dev);
3419 	struct ifnet *ifp = &sc->arpcom.ac_if;
3420 
3421 	lwkt_serialize_enter(ifp->if_serializer);
3422 	bnx_stop(sc);
3423 	bnx_reset(sc);
3424 	lwkt_serialize_exit(ifp->if_serializer);
3425 }
3426 
3427 static int
3428 bnx_suspend(device_t dev)
3429 {
3430 	struct bnx_softc *sc = device_get_softc(dev);
3431 	struct ifnet *ifp = &sc->arpcom.ac_if;
3432 
3433 	lwkt_serialize_enter(ifp->if_serializer);
3434 	bnx_stop(sc);
3435 	lwkt_serialize_exit(ifp->if_serializer);
3436 
3437 	return 0;
3438 }
3439 
3440 static int
3441 bnx_resume(device_t dev)
3442 {
3443 	struct bnx_softc *sc = device_get_softc(dev);
3444 	struct ifnet *ifp = &sc->arpcom.ac_if;
3445 
3446 	lwkt_serialize_enter(ifp->if_serializer);
3447 
3448 	if (ifp->if_flags & IFF_UP) {
3449 		bnx_init(sc);
3450 
3451 		if (!ifq_is_empty(&ifp->if_snd))
3452 			if_devstart(ifp);
3453 	}
3454 
3455 	lwkt_serialize_exit(ifp->if_serializer);
3456 
3457 	return 0;
3458 }
3459 
3460 static void
3461 bnx_setpromisc(struct bnx_softc *sc)
3462 {
3463 	struct ifnet *ifp = &sc->arpcom.ac_if;
3464 
3465 	if (ifp->if_flags & IFF_PROMISC)
3466 		BNX_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
3467 	else
3468 		BNX_CLRBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
3469 }
3470 
3471 static void
3472 bnx_dma_free(struct bnx_softc *sc)
3473 {
3474 	int i;
3475 
3476 	/* Destroy RX mbuf DMA stuffs. */
3477 	if (sc->bnx_cdata.bnx_rx_mtag != NULL) {
3478 		for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
3479 			bus_dmamap_destroy(sc->bnx_cdata.bnx_rx_mtag,
3480 			    sc->bnx_cdata.bnx_rx_std_dmamap[i]);
3481 		}
3482 		bus_dmamap_destroy(sc->bnx_cdata.bnx_rx_mtag,
3483 				   sc->bnx_cdata.bnx_rx_tmpmap);
3484 		bus_dma_tag_destroy(sc->bnx_cdata.bnx_rx_mtag);
3485 	}
3486 
3487 	/* Destroy TX mbuf DMA stuffs. */
3488 	if (sc->bnx_cdata.bnx_tx_mtag != NULL) {
3489 		for (i = 0; i < BGE_TX_RING_CNT; i++) {
3490 			bus_dmamap_destroy(sc->bnx_cdata.bnx_tx_mtag,
3491 			    sc->bnx_cdata.bnx_tx_dmamap[i]);
3492 		}
3493 		bus_dma_tag_destroy(sc->bnx_cdata.bnx_tx_mtag);
3494 	}
3495 
3496 	/* Destroy standard RX ring */
3497 	bnx_dma_block_free(sc->bnx_cdata.bnx_rx_std_ring_tag,
3498 			   sc->bnx_cdata.bnx_rx_std_ring_map,
3499 			   sc->bnx_ldata.bnx_rx_std_ring);
3500 
3501 	if (BNX_IS_JUMBO_CAPABLE(sc))
3502 		bnx_free_jumbo_mem(sc);
3503 
3504 	/* Destroy RX return ring */
3505 	bnx_dma_block_free(sc->bnx_cdata.bnx_rx_return_ring_tag,
3506 			   sc->bnx_cdata.bnx_rx_return_ring_map,
3507 			   sc->bnx_ldata.bnx_rx_return_ring);
3508 
3509 	/* Destroy TX ring */
3510 	bnx_dma_block_free(sc->bnx_cdata.bnx_tx_ring_tag,
3511 			   sc->bnx_cdata.bnx_tx_ring_map,
3512 			   sc->bnx_ldata.bnx_tx_ring);
3513 
3514 	/* Destroy status block */
3515 	bnx_dma_block_free(sc->bnx_cdata.bnx_status_tag,
3516 			   sc->bnx_cdata.bnx_status_map,
3517 			   sc->bnx_ldata.bnx_status_block);
3518 
3519 	/* Destroy the parent tag */
3520 	if (sc->bnx_cdata.bnx_parent_tag != NULL)
3521 		bus_dma_tag_destroy(sc->bnx_cdata.bnx_parent_tag);
3522 }
3523 
3524 static int
3525 bnx_dma_alloc(struct bnx_softc *sc)
3526 {
3527 	struct ifnet *ifp = &sc->arpcom.ac_if;
3528 	bus_size_t txmaxsz;
3529 	int i, error;
3530 
3531 	/*
3532 	 * Allocate the parent bus DMA tag appropriate for PCI.
3533 	 *
3534 	 * All of the NetExtreme/NetLink controllers have 4GB boundary
3535 	 * DMA bug.
3536 	 * Whenever an address crosses a multiple of the 4GB boundary
3537 	 * (including 4GB, 8Gb, 12Gb, etc.) and makes the transition
3538 	 * from 0xX_FFFF_FFFF to 0x(X+1)_0000_0000 an internal DMA
3539 	 * state machine will lockup and cause the device to hang.
3540 	 */
3541 	error = bus_dma_tag_create(NULL, 1, BGE_DMA_BOUNDARY_4G,
3542 				   BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
3543 				   NULL, NULL,
3544 				   BUS_SPACE_MAXSIZE_32BIT, 0,
3545 				   BUS_SPACE_MAXSIZE_32BIT,
3546 				   0, &sc->bnx_cdata.bnx_parent_tag);
3547 	if (error) {
3548 		if_printf(ifp, "could not allocate parent dma tag\n");
3549 		return error;
3550 	}
3551 
3552 	/*
3553 	 * Create DMA tag and maps for RX mbufs.
3554 	 */
3555 	error = bus_dma_tag_create(sc->bnx_cdata.bnx_parent_tag, 1, 0,
3556 				   BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
3557 				   NULL, NULL, MCLBYTES, 1, MCLBYTES,
3558 				   BUS_DMA_ALLOCNOW | BUS_DMA_WAITOK,
3559 				   &sc->bnx_cdata.bnx_rx_mtag);
3560 	if (error) {
3561 		if_printf(ifp, "could not allocate RX mbuf dma tag\n");
3562 		return error;
3563 	}
3564 
3565 	error = bus_dmamap_create(sc->bnx_cdata.bnx_rx_mtag,
3566 				  BUS_DMA_WAITOK, &sc->bnx_cdata.bnx_rx_tmpmap);
3567 	if (error) {
3568 		bus_dma_tag_destroy(sc->bnx_cdata.bnx_rx_mtag);
3569 		sc->bnx_cdata.bnx_rx_mtag = NULL;
3570 		return error;
3571 	}
3572 
3573 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
3574 		error = bus_dmamap_create(sc->bnx_cdata.bnx_rx_mtag,
3575 					  BUS_DMA_WAITOK,
3576 					  &sc->bnx_cdata.bnx_rx_std_dmamap[i]);
3577 		if (error) {
3578 			int j;
3579 
3580 			for (j = 0; j < i; ++j) {
3581 				bus_dmamap_destroy(sc->bnx_cdata.bnx_rx_mtag,
3582 					sc->bnx_cdata.bnx_rx_std_dmamap[j]);
3583 			}
3584 			bus_dma_tag_destroy(sc->bnx_cdata.bnx_rx_mtag);
3585 			sc->bnx_cdata.bnx_rx_mtag = NULL;
3586 
3587 			if_printf(ifp, "could not create DMA map for RX\n");
3588 			return error;
3589 		}
3590 	}
3591 
3592 	/*
3593 	 * Create DMA tag and maps for TX mbufs.
3594 	 */
3595 	if (sc->bnx_flags & BNX_FLAG_TSO)
3596 		txmaxsz = IP_MAXPACKET + sizeof(struct ether_vlan_header);
3597 	else
3598 		txmaxsz = BNX_JUMBO_FRAMELEN;
3599 	error = bus_dma_tag_create(sc->bnx_cdata.bnx_parent_tag, 1, 0,
3600 				   BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
3601 				   NULL, NULL,
3602 				   txmaxsz, BNX_NSEG_NEW, PAGE_SIZE,
3603 				   BUS_DMA_ALLOCNOW | BUS_DMA_WAITOK |
3604 				   BUS_DMA_ONEBPAGE,
3605 				   &sc->bnx_cdata.bnx_tx_mtag);
3606 	if (error) {
3607 		if_printf(ifp, "could not allocate TX mbuf dma tag\n");
3608 		return error;
3609 	}
3610 
3611 	for (i = 0; i < BGE_TX_RING_CNT; i++) {
3612 		error = bus_dmamap_create(sc->bnx_cdata.bnx_tx_mtag,
3613 					  BUS_DMA_WAITOK | BUS_DMA_ONEBPAGE,
3614 					  &sc->bnx_cdata.bnx_tx_dmamap[i]);
3615 		if (error) {
3616 			int j;
3617 
3618 			for (j = 0; j < i; ++j) {
3619 				bus_dmamap_destroy(sc->bnx_cdata.bnx_tx_mtag,
3620 					sc->bnx_cdata.bnx_tx_dmamap[j]);
3621 			}
3622 			bus_dma_tag_destroy(sc->bnx_cdata.bnx_tx_mtag);
3623 			sc->bnx_cdata.bnx_tx_mtag = NULL;
3624 
3625 			if_printf(ifp, "could not create DMA map for TX\n");
3626 			return error;
3627 		}
3628 	}
3629 
3630 	/*
3631 	 * Create DMA stuffs for standard RX ring.
3632 	 */
3633 	error = bnx_dma_block_alloc(sc, BGE_STD_RX_RING_SZ,
3634 				    &sc->bnx_cdata.bnx_rx_std_ring_tag,
3635 				    &sc->bnx_cdata.bnx_rx_std_ring_map,
3636 				    (void *)&sc->bnx_ldata.bnx_rx_std_ring,
3637 				    &sc->bnx_ldata.bnx_rx_std_ring_paddr);
3638 	if (error) {
3639 		if_printf(ifp, "could not create std RX ring\n");
3640 		return error;
3641 	}
3642 
3643 	/*
3644 	 * Create jumbo buffer pool.
3645 	 */
3646 	if (BNX_IS_JUMBO_CAPABLE(sc)) {
3647 		error = bnx_alloc_jumbo_mem(sc);
3648 		if (error) {
3649 			if_printf(ifp, "could not create jumbo buffer pool\n");
3650 			return error;
3651 		}
3652 	}
3653 
3654 	/*
3655 	 * Create DMA stuffs for RX return ring.
3656 	 */
3657 	error = bnx_dma_block_alloc(sc,
3658 	    BGE_RX_RTN_RING_SZ(sc->bnx_return_ring_cnt),
3659 	    &sc->bnx_cdata.bnx_rx_return_ring_tag,
3660 	    &sc->bnx_cdata.bnx_rx_return_ring_map,
3661 	    (void *)&sc->bnx_ldata.bnx_rx_return_ring,
3662 	    &sc->bnx_ldata.bnx_rx_return_ring_paddr);
3663 	if (error) {
3664 		if_printf(ifp, "could not create RX ret ring\n");
3665 		return error;
3666 	}
3667 
3668 	/*
3669 	 * Create DMA stuffs for TX ring.
3670 	 */
3671 	error = bnx_dma_block_alloc(sc, BGE_TX_RING_SZ,
3672 				    &sc->bnx_cdata.bnx_tx_ring_tag,
3673 				    &sc->bnx_cdata.bnx_tx_ring_map,
3674 				    (void *)&sc->bnx_ldata.bnx_tx_ring,
3675 				    &sc->bnx_ldata.bnx_tx_ring_paddr);
3676 	if (error) {
3677 		if_printf(ifp, "could not create TX ring\n");
3678 		return error;
3679 	}
3680 
3681 	/*
3682 	 * Create DMA stuffs for status block.
3683 	 */
3684 	error = bnx_dma_block_alloc(sc, BGE_STATUS_BLK_SZ,
3685 				    &sc->bnx_cdata.bnx_status_tag,
3686 				    &sc->bnx_cdata.bnx_status_map,
3687 				    (void *)&sc->bnx_ldata.bnx_status_block,
3688 				    &sc->bnx_ldata.bnx_status_block_paddr);
3689 	if (error) {
3690 		if_printf(ifp, "could not create status block\n");
3691 		return error;
3692 	}
3693 
3694 	return 0;
3695 }
3696 
3697 static int
3698 bnx_dma_block_alloc(struct bnx_softc *sc, bus_size_t size, bus_dma_tag_t *tag,
3699 		    bus_dmamap_t *map, void **addr, bus_addr_t *paddr)
3700 {
3701 	bus_dmamem_t dmem;
3702 	int error;
3703 
3704 	error = bus_dmamem_coherent(sc->bnx_cdata.bnx_parent_tag, PAGE_SIZE, 0,
3705 				    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
3706 				    size, BUS_DMA_WAITOK | BUS_DMA_ZERO, &dmem);
3707 	if (error)
3708 		return error;
3709 
3710 	*tag = dmem.dmem_tag;
3711 	*map = dmem.dmem_map;
3712 	*addr = dmem.dmem_addr;
3713 	*paddr = dmem.dmem_busaddr;
3714 
3715 	return 0;
3716 }
3717 
3718 static void
3719 bnx_dma_block_free(bus_dma_tag_t tag, bus_dmamap_t map, void *addr)
3720 {
3721 	if (tag != NULL) {
3722 		bus_dmamap_unload(tag, map);
3723 		bus_dmamem_free(tag, addr, map);
3724 		bus_dma_tag_destroy(tag);
3725 	}
3726 }
3727 
3728 static void
3729 bnx_tbi_link_upd(struct bnx_softc *sc, uint32_t status)
3730 {
3731 	struct ifnet *ifp = &sc->arpcom.ac_if;
3732 
3733 #define PCS_ENCODE_ERR	(BGE_MACSTAT_PORT_DECODE_ERROR|BGE_MACSTAT_MI_COMPLETE)
3734 
3735 	/*
3736 	 * Sometimes PCS encoding errors are detected in
3737 	 * TBI mode (on fiber NICs), and for some reason
3738 	 * the chip will signal them as link changes.
3739 	 * If we get a link change event, but the 'PCS
3740 	 * encoding error' bit in the MAC status register
3741 	 * is set, don't bother doing a link check.
3742 	 * This avoids spurious "gigabit link up" messages
3743 	 * that sometimes appear on fiber NICs during
3744 	 * periods of heavy traffic.
3745 	 */
3746 	if (status & BGE_MACSTAT_TBI_PCS_SYNCHED) {
3747 		if (!sc->bnx_link) {
3748 			sc->bnx_link++;
3749 			if (sc->bnx_asicrev == BGE_ASICREV_BCM5704) {
3750 				BNX_CLRBIT(sc, BGE_MAC_MODE,
3751 				    BGE_MACMODE_TBI_SEND_CFGS);
3752 			}
3753 			CSR_WRITE_4(sc, BGE_MAC_STS, 0xFFFFFFFF);
3754 
3755 			if (bootverbose)
3756 				if_printf(ifp, "link UP\n");
3757 
3758 			ifp->if_link_state = LINK_STATE_UP;
3759 			if_link_state_change(ifp);
3760 		}
3761 	} else if ((status & PCS_ENCODE_ERR) != PCS_ENCODE_ERR) {
3762 		if (sc->bnx_link) {
3763 			sc->bnx_link = 0;
3764 
3765 			if (bootverbose)
3766 				if_printf(ifp, "link DOWN\n");
3767 
3768 			ifp->if_link_state = LINK_STATE_DOWN;
3769 			if_link_state_change(ifp);
3770 		}
3771 	}
3772 
3773 #undef PCS_ENCODE_ERR
3774 
3775 	/* Clear the attention. */
3776 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
3777 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
3778 	    BGE_MACSTAT_LINK_CHANGED);
3779 }
3780 
3781 static void
3782 bnx_copper_link_upd(struct bnx_softc *sc, uint32_t status __unused)
3783 {
3784 	struct ifnet *ifp = &sc->arpcom.ac_if;
3785 	struct mii_data *mii = device_get_softc(sc->bnx_miibus);
3786 
3787 	mii_pollstat(mii);
3788 	bnx_miibus_statchg(sc->bnx_dev);
3789 
3790 	if (bootverbose) {
3791 		if (sc->bnx_link)
3792 			if_printf(ifp, "link UP\n");
3793 		else
3794 			if_printf(ifp, "link DOWN\n");
3795 	}
3796 
3797 	/* Clear the attention. */
3798 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
3799 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
3800 	    BGE_MACSTAT_LINK_CHANGED);
3801 }
3802 
3803 static void
3804 bnx_autopoll_link_upd(struct bnx_softc *sc, uint32_t status __unused)
3805 {
3806 	struct ifnet *ifp = &sc->arpcom.ac_if;
3807 	struct mii_data *mii = device_get_softc(sc->bnx_miibus);
3808 
3809 	mii_pollstat(mii);
3810 
3811 	if (!sc->bnx_link &&
3812 	    (mii->mii_media_status & IFM_ACTIVE) &&
3813 	    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
3814 		sc->bnx_link++;
3815 		if (bootverbose)
3816 			if_printf(ifp, "link UP\n");
3817 	} else if (sc->bnx_link &&
3818 	    (!(mii->mii_media_status & IFM_ACTIVE) ||
3819 	    IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE)) {
3820 		sc->bnx_link = 0;
3821 		if (bootverbose)
3822 			if_printf(ifp, "link DOWN\n");
3823 	}
3824 
3825 	/* Clear the attention. */
3826 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
3827 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
3828 	    BGE_MACSTAT_LINK_CHANGED);
3829 }
3830 
3831 static int
3832 bnx_sysctl_rx_coal_ticks(SYSCTL_HANDLER_ARGS)
3833 {
3834 	struct bnx_softc *sc = arg1;
3835 
3836 	return bnx_sysctl_coal_chg(oidp, arg1, arg2, req,
3837 	    &sc->bnx_rx_coal_ticks,
3838 	    BNX_RX_COAL_TICKS_MIN, BNX_RX_COAL_TICKS_MAX,
3839 	    BNX_RX_COAL_TICKS_CHG);
3840 }
3841 
3842 static int
3843 bnx_sysctl_tx_coal_ticks(SYSCTL_HANDLER_ARGS)
3844 {
3845 	struct bnx_softc *sc = arg1;
3846 
3847 	return bnx_sysctl_coal_chg(oidp, arg1, arg2, req,
3848 	    &sc->bnx_tx_coal_ticks,
3849 	    BNX_TX_COAL_TICKS_MIN, BNX_TX_COAL_TICKS_MAX,
3850 	    BNX_TX_COAL_TICKS_CHG);
3851 }
3852 
3853 static int
3854 bnx_sysctl_rx_coal_bds(SYSCTL_HANDLER_ARGS)
3855 {
3856 	struct bnx_softc *sc = arg1;
3857 
3858 	return bnx_sysctl_coal_chg(oidp, arg1, arg2, req,
3859 	    &sc->bnx_rx_coal_bds,
3860 	    BNX_RX_COAL_BDS_MIN, BNX_RX_COAL_BDS_MAX,
3861 	    BNX_RX_COAL_BDS_CHG);
3862 }
3863 
3864 static int
3865 bnx_sysctl_tx_coal_bds(SYSCTL_HANDLER_ARGS)
3866 {
3867 	struct bnx_softc *sc = arg1;
3868 
3869 	return bnx_sysctl_coal_chg(oidp, arg1, arg2, req,
3870 	    &sc->bnx_tx_coal_bds,
3871 	    BNX_TX_COAL_BDS_MIN, BNX_TX_COAL_BDS_MAX,
3872 	    BNX_TX_COAL_BDS_CHG);
3873 }
3874 
3875 static int
3876 bnx_sysctl_rx_coal_bds_int(SYSCTL_HANDLER_ARGS)
3877 {
3878 	struct bnx_softc *sc = arg1;
3879 
3880 	return bnx_sysctl_coal_chg(oidp, arg1, arg2, req,
3881 	    &sc->bnx_rx_coal_bds_int,
3882 	    BNX_RX_COAL_BDS_MIN, BNX_RX_COAL_BDS_MAX,
3883 	    BNX_RX_COAL_BDS_INT_CHG);
3884 }
3885 
3886 static int
3887 bnx_sysctl_tx_coal_bds_int(SYSCTL_HANDLER_ARGS)
3888 {
3889 	struct bnx_softc *sc = arg1;
3890 
3891 	return bnx_sysctl_coal_chg(oidp, arg1, arg2, req,
3892 	    &sc->bnx_tx_coal_bds_int,
3893 	    BNX_TX_COAL_BDS_MIN, BNX_TX_COAL_BDS_MAX,
3894 	    BNX_TX_COAL_BDS_INT_CHG);
3895 }
3896 
3897 static int
3898 bnx_sysctl_coal_chg(SYSCTL_HANDLER_ARGS, uint32_t *coal,
3899     int coal_min, int coal_max, uint32_t coal_chg_mask)
3900 {
3901 	struct bnx_softc *sc = arg1;
3902 	struct ifnet *ifp = &sc->arpcom.ac_if;
3903 	int error = 0, v;
3904 
3905 	lwkt_serialize_enter(ifp->if_serializer);
3906 
3907 	v = *coal;
3908 	error = sysctl_handle_int(oidp, &v, 0, req);
3909 	if (!error && req->newptr != NULL) {
3910 		if (v < coal_min || v > coal_max) {
3911 			error = EINVAL;
3912 		} else {
3913 			*coal = v;
3914 			sc->bnx_coal_chg |= coal_chg_mask;
3915 		}
3916 	}
3917 
3918 	lwkt_serialize_exit(ifp->if_serializer);
3919 	return error;
3920 }
3921 
3922 static void
3923 bnx_coal_change(struct bnx_softc *sc)
3924 {
3925 	struct ifnet *ifp = &sc->arpcom.ac_if;
3926 	uint32_t val;
3927 
3928 	ASSERT_SERIALIZED(ifp->if_serializer);
3929 
3930 	if (sc->bnx_coal_chg & BNX_RX_COAL_TICKS_CHG) {
3931 		CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS,
3932 			    sc->bnx_rx_coal_ticks);
3933 		DELAY(10);
3934 		val = CSR_READ_4(sc, BGE_HCC_RX_COAL_TICKS);
3935 
3936 		if (bootverbose) {
3937 			if_printf(ifp, "rx_coal_ticks -> %u\n",
3938 				  sc->bnx_rx_coal_ticks);
3939 		}
3940 	}
3941 
3942 	if (sc->bnx_coal_chg & BNX_TX_COAL_TICKS_CHG) {
3943 		CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS,
3944 			    sc->bnx_tx_coal_ticks);
3945 		DELAY(10);
3946 		val = CSR_READ_4(sc, BGE_HCC_TX_COAL_TICKS);
3947 
3948 		if (bootverbose) {
3949 			if_printf(ifp, "tx_coal_ticks -> %u\n",
3950 				  sc->bnx_tx_coal_ticks);
3951 		}
3952 	}
3953 
3954 	if (sc->bnx_coal_chg & BNX_RX_COAL_BDS_CHG) {
3955 		CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS,
3956 			    sc->bnx_rx_coal_bds);
3957 		DELAY(10);
3958 		val = CSR_READ_4(sc, BGE_HCC_RX_MAX_COAL_BDS);
3959 
3960 		if (bootverbose) {
3961 			if_printf(ifp, "rx_coal_bds -> %u\n",
3962 				  sc->bnx_rx_coal_bds);
3963 		}
3964 	}
3965 
3966 	if (sc->bnx_coal_chg & BNX_TX_COAL_BDS_CHG) {
3967 		CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS,
3968 			    sc->bnx_tx_coal_bds);
3969 		DELAY(10);
3970 		val = CSR_READ_4(sc, BGE_HCC_TX_MAX_COAL_BDS);
3971 
3972 		if (bootverbose) {
3973 			if_printf(ifp, "tx_max_coal_bds -> %u\n",
3974 				  sc->bnx_tx_coal_bds);
3975 		}
3976 	}
3977 
3978 	if (sc->bnx_coal_chg & BNX_RX_COAL_BDS_INT_CHG) {
3979 		CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT,
3980 		    sc->bnx_rx_coal_bds_int);
3981 		DELAY(10);
3982 		val = CSR_READ_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT);
3983 
3984 		if (bootverbose) {
3985 			if_printf(ifp, "rx_coal_bds_int -> %u\n",
3986 			    sc->bnx_rx_coal_bds_int);
3987 		}
3988 	}
3989 
3990 	if (sc->bnx_coal_chg & BNX_TX_COAL_BDS_INT_CHG) {
3991 		CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT,
3992 		    sc->bnx_tx_coal_bds_int);
3993 		DELAY(10);
3994 		val = CSR_READ_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT);
3995 
3996 		if (bootverbose) {
3997 			if_printf(ifp, "tx_coal_bds_int -> %u\n",
3998 			    sc->bnx_tx_coal_bds_int);
3999 		}
4000 	}
4001 
4002 	sc->bnx_coal_chg = 0;
4003 }
4004 
4005 static void
4006 bnx_intr_check(void *xsc)
4007 {
4008 	struct bnx_softc *sc = xsc;
4009 	struct ifnet *ifp = &sc->arpcom.ac_if;
4010 	struct bge_status_block *sblk = sc->bnx_ldata.bnx_status_block;
4011 
4012 	lwkt_serialize_enter(ifp->if_serializer);
4013 
4014 	KKASSERT(mycpuid == sc->bnx_intr_cpuid);
4015 
4016 	if ((ifp->if_flags & (IFF_RUNNING | IFF_POLLING)) != IFF_RUNNING) {
4017 		lwkt_serialize_exit(ifp->if_serializer);
4018 		return;
4019 	}
4020 
4021 	if (sblk->bge_idx[0].bge_rx_prod_idx != sc->bnx_rx_saved_considx ||
4022 	    sblk->bge_idx[0].bge_tx_cons_idx != sc->bnx_tx_saved_considx) {
4023 		if (sc->bnx_rx_check_considx == sc->bnx_rx_saved_considx &&
4024 		    sc->bnx_tx_check_considx == sc->bnx_tx_saved_considx) {
4025 			if (!sc->bnx_intr_maylose) {
4026 				sc->bnx_intr_maylose = TRUE;
4027 				goto done;
4028 			}
4029 			if (bootverbose)
4030 				if_printf(ifp, "lost interrupt\n");
4031 			bnx_msi(sc);
4032 		}
4033 	}
4034 	sc->bnx_intr_maylose = FALSE;
4035 	sc->bnx_rx_check_considx = sc->bnx_rx_saved_considx;
4036 	sc->bnx_tx_check_considx = sc->bnx_tx_saved_considx;
4037 
4038 done:
4039 	callout_reset(&sc->bnx_intr_timer, BNX_INTR_CKINTVL,
4040 	    bnx_intr_check, sc);
4041 	lwkt_serialize_exit(ifp->if_serializer);
4042 }
4043 
4044 static void
4045 bnx_enable_intr(struct bnx_softc *sc)
4046 {
4047 	struct ifnet *ifp = &sc->arpcom.ac_if;
4048 
4049 	lwkt_serialize_handler_enable(ifp->if_serializer);
4050 
4051 	/*
4052 	 * Enable interrupt.
4053 	 */
4054 	bnx_writembx(sc, BGE_MBX_IRQ0_LO, sc->bnx_status_tag << 24);
4055 	if (sc->bnx_flags & BNX_FLAG_ONESHOT_MSI) {
4056 		/* XXX Linux driver */
4057 		bnx_writembx(sc, BGE_MBX_IRQ0_LO, sc->bnx_status_tag << 24);
4058 	}
4059 
4060 	/*
4061 	 * Unmask the interrupt when we stop polling.
4062 	 */
4063 	PCI_CLRBIT(sc->bnx_dev, BGE_PCI_MISC_CTL,
4064 	    BGE_PCIMISCCTL_MASK_PCI_INTR, 4);
4065 
4066 	/*
4067 	 * Trigger another interrupt, since above writing
4068 	 * to interrupt mailbox0 may acknowledge pending
4069 	 * interrupt.
4070 	 */
4071 	BNX_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
4072 
4073 	if (sc->bnx_flags & BNX_FLAG_STATUSTAG_BUG) {
4074 		sc->bnx_intr_maylose = FALSE;
4075 		sc->bnx_rx_check_considx = 0;
4076 		sc->bnx_tx_check_considx = 0;
4077 
4078 		if (bootverbose)
4079 			if_printf(ifp, "status tag bug workaround\n");
4080 
4081 		/* 10ms check interval */
4082 		callout_reset_bycpu(&sc->bnx_intr_timer, BNX_INTR_CKINTVL,
4083 		    bnx_intr_check, sc, sc->bnx_intr_cpuid);
4084 	}
4085 }
4086 
4087 static void
4088 bnx_disable_intr(struct bnx_softc *sc)
4089 {
4090 	struct ifnet *ifp = &sc->arpcom.ac_if;
4091 
4092 	/*
4093 	 * Mask the interrupt when we start polling.
4094 	 */
4095 	PCI_SETBIT(sc->bnx_dev, BGE_PCI_MISC_CTL,
4096 	    BGE_PCIMISCCTL_MASK_PCI_INTR, 4);
4097 
4098 	/*
4099 	 * Acknowledge possible asserted interrupt.
4100 	 */
4101 	bnx_writembx(sc, BGE_MBX_IRQ0_LO, 1);
4102 
4103 	callout_stop(&sc->bnx_intr_timer);
4104 	sc->bnx_intr_maylose = FALSE;
4105 	sc->bnx_rx_check_considx = 0;
4106 	sc->bnx_tx_check_considx = 0;
4107 
4108 	lwkt_serialize_handler_disable(ifp->if_serializer);
4109 }
4110 
4111 static int
4112 bnx_get_eaddr_mem(struct bnx_softc *sc, uint8_t ether_addr[])
4113 {
4114 	uint32_t mac_addr;
4115 	int ret = 1;
4116 
4117 	mac_addr = bnx_readmem_ind(sc, 0x0c14);
4118 	if ((mac_addr >> 16) == 0x484b) {
4119 		ether_addr[0] = (uint8_t)(mac_addr >> 8);
4120 		ether_addr[1] = (uint8_t)mac_addr;
4121 		mac_addr = bnx_readmem_ind(sc, 0x0c18);
4122 		ether_addr[2] = (uint8_t)(mac_addr >> 24);
4123 		ether_addr[3] = (uint8_t)(mac_addr >> 16);
4124 		ether_addr[4] = (uint8_t)(mac_addr >> 8);
4125 		ether_addr[5] = (uint8_t)mac_addr;
4126 		ret = 0;
4127 	}
4128 	return ret;
4129 }
4130 
4131 static int
4132 bnx_get_eaddr_nvram(struct bnx_softc *sc, uint8_t ether_addr[])
4133 {
4134 	int mac_offset = BGE_EE_MAC_OFFSET;
4135 
4136 	if (BNX_IS_5717_PLUS(sc)) {
4137 		int f;
4138 
4139 		f = pci_get_function(sc->bnx_dev);
4140 		if (f & 1)
4141 			mac_offset = BGE_EE_MAC_OFFSET_5717;
4142 		if (f > 1)
4143 			mac_offset += BGE_EE_MAC_OFFSET_5717_OFF;
4144 	} else if (sc->bnx_asicrev == BGE_ASICREV_BCM5906) {
4145 		mac_offset = BGE_EE_MAC_OFFSET_5906;
4146 	}
4147 
4148 	return bnx_read_nvram(sc, ether_addr, mac_offset + 2, ETHER_ADDR_LEN);
4149 }
4150 
4151 static int
4152 bnx_get_eaddr_eeprom(struct bnx_softc *sc, uint8_t ether_addr[])
4153 {
4154 	if (sc->bnx_flags & BNX_FLAG_NO_EEPROM)
4155 		return 1;
4156 
4157 	return bnx_read_eeprom(sc, ether_addr, BGE_EE_MAC_OFFSET + 2,
4158 			       ETHER_ADDR_LEN);
4159 }
4160 
4161 static int
4162 bnx_get_eaddr(struct bnx_softc *sc, uint8_t eaddr[])
4163 {
4164 	static const bnx_eaddr_fcn_t bnx_eaddr_funcs[] = {
4165 		/* NOTE: Order is critical */
4166 		bnx_get_eaddr_mem,
4167 		bnx_get_eaddr_nvram,
4168 		bnx_get_eaddr_eeprom,
4169 		NULL
4170 	};
4171 	const bnx_eaddr_fcn_t *func;
4172 
4173 	for (func = bnx_eaddr_funcs; *func != NULL; ++func) {
4174 		if ((*func)(sc, eaddr) == 0)
4175 			break;
4176 	}
4177 	return (*func == NULL ? ENXIO : 0);
4178 }
4179 
4180 /*
4181  * NOTE: 'm' is not freed upon failure
4182  */
4183 struct mbuf *
4184 bnx_defrag_shortdma(struct mbuf *m)
4185 {
4186 	struct mbuf *n;
4187 	int found;
4188 
4189 	/*
4190 	 * If device receive two back-to-back send BDs with less than
4191 	 * or equal to 8 total bytes then the device may hang.  The two
4192 	 * back-to-back send BDs must in the same frame for this failure
4193 	 * to occur.  Scan mbuf chains and see whether two back-to-back
4194 	 * send BDs are there.  If this is the case, allocate new mbuf
4195 	 * and copy the frame to workaround the silicon bug.
4196 	 */
4197 	for (n = m, found = 0; n != NULL; n = n->m_next) {
4198 		if (n->m_len < 8) {
4199 			found++;
4200 			if (found > 1)
4201 				break;
4202 			continue;
4203 		}
4204 		found = 0;
4205 	}
4206 
4207 	if (found > 1)
4208 		n = m_defrag(m, MB_DONTWAIT);
4209 	else
4210 		n = m;
4211 	return n;
4212 }
4213 
4214 static void
4215 bnx_stop_block(struct bnx_softc *sc, bus_size_t reg, uint32_t bit)
4216 {
4217 	int i;
4218 
4219 	BNX_CLRBIT(sc, reg, bit);
4220 	for (i = 0; i < BNX_TIMEOUT; i++) {
4221 		if ((CSR_READ_4(sc, reg) & bit) == 0)
4222 			return;
4223 		DELAY(100);
4224 	}
4225 }
4226 
4227 static void
4228 bnx_link_poll(struct bnx_softc *sc)
4229 {
4230 	uint32_t status;
4231 
4232 	status = CSR_READ_4(sc, BGE_MAC_STS);
4233 	if ((status & sc->bnx_link_chg) || sc->bnx_link_evt) {
4234 		sc->bnx_link_evt = 0;
4235 		sc->bnx_link_upd(sc, status);
4236 	}
4237 }
4238 
4239 static void
4240 bnx_enable_msi(struct bnx_softc *sc)
4241 {
4242 	uint32_t msi_mode;
4243 
4244 	msi_mode = CSR_READ_4(sc, BGE_MSI_MODE);
4245 	msi_mode |= BGE_MSIMODE_ENABLE;
4246 	if (sc->bnx_flags & BNX_FLAG_ONESHOT_MSI) {
4247 		/*
4248 		 * NOTE:
4249 		 * 5718-PG105-R says that "one shot" mode
4250 		 * does not work if MSI is used, however,
4251 		 * it obviously works.
4252 		 */
4253 		msi_mode &= ~BGE_MSIMODE_ONESHOT_DISABLE;
4254 	}
4255 	CSR_WRITE_4(sc, BGE_MSI_MODE, msi_mode);
4256 }
4257 
4258 static uint32_t
4259 bnx_dma_swap_options(struct bnx_softc *sc)
4260 {
4261 	uint32_t dma_options;
4262 
4263 	dma_options = BGE_MODECTL_WORDSWAP_NONFRAME |
4264 	    BGE_MODECTL_BYTESWAP_DATA | BGE_MODECTL_WORDSWAP_DATA;
4265 #if BYTE_ORDER == BIG_ENDIAN
4266 	dma_options |= BGE_MODECTL_BYTESWAP_NONFRAME;
4267 #endif
4268 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5720) {
4269 		dma_options |= BGE_MODECTL_BYTESWAP_B2HRX_DATA |
4270 		    BGE_MODECTL_WORDSWAP_B2HRX_DATA | BGE_MODECTL_B2HRX_ENABLE |
4271 		    BGE_MODECTL_HTX2B_ENABLE;
4272 	}
4273 	return dma_options;
4274 }
4275 
4276 static int
4277 bnx_setup_tso(struct bnx_softc *sc, struct mbuf **mp,
4278     uint16_t *mss0, uint16_t *flags0)
4279 {
4280 	struct mbuf *m;
4281 	struct ip *ip;
4282 	struct tcphdr *th;
4283 	int thoff, iphlen, hoff, hlen;
4284 	uint16_t flags, mss;
4285 
4286 	m = *mp;
4287 	KASSERT(M_WRITABLE(m), ("TSO mbuf not writable"));
4288 
4289 	hoff = m->m_pkthdr.csum_lhlen;
4290 	iphlen = m->m_pkthdr.csum_iphlen;
4291 	thoff = m->m_pkthdr.csum_thlen;
4292 
4293 	KASSERT(hoff > 0, ("invalid ether header len"));
4294 	KASSERT(iphlen > 0, ("invalid ip header len"));
4295 	KASSERT(thoff > 0, ("invalid tcp header len"));
4296 
4297 	if (__predict_false(m->m_len < hoff + iphlen + thoff)) {
4298 		m = m_pullup(m, hoff + iphlen + thoff);
4299 		if (m == NULL) {
4300 			*mp = NULL;
4301 			return ENOBUFS;
4302 		}
4303 		*mp = m;
4304 	}
4305 	ip = mtodoff(m, struct ip *, hoff);
4306 	th = mtodoff(m, struct tcphdr *, hoff + iphlen);
4307 
4308 	mss = m->m_pkthdr.tso_segsz;
4309 	flags = BGE_TXBDFLAG_CPU_PRE_DMA | BGE_TXBDFLAG_CPU_POST_DMA;
4310 
4311 	ip->ip_len = htons(mss + iphlen + thoff);
4312 	th->th_sum = 0;
4313 
4314 	hlen = (iphlen + thoff) >> 2;
4315 	mss |= ((hlen & 0x3) << 14);
4316 	flags |= ((hlen & 0xf8) << 7) | ((hlen & 0x4) << 2);
4317 
4318 	*mss0 = mss;
4319 	*flags0 = flags;
4320 
4321 	return 0;
4322 }
4323