xref: /dragonfly/sys/dev/netif/bnx/if_bnx.c (revision f26349bc)
1 /*
2  * Copyright (c) 2001 Wind River Systems
3  * Copyright (c) 1997, 1998, 1999, 2001
4  *	Bill Paul <wpaul@windriver.com>.  All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  * 3. All advertising materials mentioning features or use of this software
15  *    must display the following acknowledgement:
16  *	This product includes software developed by Bill Paul.
17  * 4. Neither the name of the author nor the names of any co-contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY Bill Paul AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL Bill Paul OR THE VOICES IN HIS HEAD
25  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
28  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
29  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
30  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
31  * THE POSSIBILITY OF SUCH DAMAGE.
32  *
33  * $FreeBSD: src/sys/dev/bge/if_bge.c,v 1.3.2.39 2005/07/03 03:41:18 silby Exp $
34  */
35 
36 #include "opt_bnx.h"
37 #include "opt_ifpoll.h"
38 
39 #include <sys/param.h>
40 #include <sys/bus.h>
41 #include <sys/endian.h>
42 #include <sys/kernel.h>
43 #include <sys/interrupt.h>
44 #include <sys/mbuf.h>
45 #include <sys/malloc.h>
46 #include <sys/queue.h>
47 #include <sys/rman.h>
48 #include <sys/serialize.h>
49 #include <sys/socket.h>
50 #include <sys/sockio.h>
51 #include <sys/sysctl.h>
52 
53 #include <netinet/ip.h>
54 #include <netinet/tcp.h>
55 
56 #include <net/bpf.h>
57 #include <net/ethernet.h>
58 #include <net/if.h>
59 #include <net/if_arp.h>
60 #include <net/if_dl.h>
61 #include <net/if_media.h>
62 #include <net/if_poll.h>
63 #include <net/if_types.h>
64 #include <net/ifq_var.h>
65 #include <net/toeplitz.h>
66 #include <net/toeplitz2.h>
67 #include <net/vlan/if_vlan_var.h>
68 #include <net/vlan/if_vlan_ether.h>
69 
70 #include <dev/netif/mii_layer/mii.h>
71 #include <dev/netif/mii_layer/miivar.h>
72 #include <dev/netif/mii_layer/brgphyreg.h>
73 
74 #include "pcidevs.h"
75 #include <bus/pci/pcireg.h>
76 #include <bus/pci/pcivar.h>
77 
78 #include <dev/netif/bge/if_bgereg.h>
79 #include <dev/netif/bnx/if_bnxvar.h>
80 
81 /* "device miibus" required.  See GENERIC if you get errors here. */
82 #include "miibus_if.h"
83 
84 #define BNX_CSUM_FEATURES	(CSUM_IP | CSUM_TCP | CSUM_UDP)
85 
86 #define	BNX_RESET_SHUTDOWN	0
87 #define	BNX_RESET_START		1
88 #define	BNX_RESET_SUSPEND	2
89 
90 #define BNX_INTR_CKINTVL	((10 * hz) / 1000)	/* 10ms */
91 
92 #ifdef BNX_RSS_DEBUG
93 #define BNX_RSS_DPRINTF(sc, lvl, fmt, ...) \
94 do { \
95 	if (sc->bnx_rss_debug >= lvl) \
96 		if_printf(&sc->arpcom.ac_if, fmt, __VA_ARGS__); \
97 } while (0)
98 #else	/* !BNX_RSS_DEBUG */
99 #define BNX_RSS_DPRINTF(sc, lvl, fmt, ...)	((void)0)
100 #endif	/* BNX_RSS_DEBUG */
101 
102 static const struct bnx_type {
103 	uint16_t		bnx_vid;
104 	uint16_t		bnx_did;
105 	char			*bnx_name;
106 } bnx_devs[] = {
107 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5717,
108 		"Broadcom BCM5717 Gigabit Ethernet" },
109 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5717C,
110 		"Broadcom BCM5717C Gigabit Ethernet" },
111 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5718,
112 		"Broadcom BCM5718 Gigabit Ethernet" },
113 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5719,
114 		"Broadcom BCM5719 Gigabit Ethernet" },
115 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5720_ALT,
116 		"Broadcom BCM5720 Gigabit Ethernet" },
117 
118 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5725,
119 		"Broadcom BCM5725 Gigabit Ethernet" },
120 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5727,
121 		"Broadcom BCM5727 Gigabit Ethernet" },
122 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM5762,
123 		"Broadcom BCM5762 Gigabit Ethernet" },
124 
125 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57761,
126 		"Broadcom BCM57761 Gigabit Ethernet" },
127 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57762,
128 		"Broadcom BCM57762 Gigabit Ethernet" },
129 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57765,
130 		"Broadcom BCM57765 Gigabit Ethernet" },
131 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57766,
132 		"Broadcom BCM57766 Gigabit Ethernet" },
133 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57781,
134 		"Broadcom BCM57781 Gigabit Ethernet" },
135 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57782,
136 		"Broadcom BCM57782 Gigabit Ethernet" },
137 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57785,
138 		"Broadcom BCM57785 Gigabit Ethernet" },
139 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57786,
140 		"Broadcom BCM57786 Gigabit Ethernet" },
141 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57791,
142 		"Broadcom BCM57791 Fast Ethernet" },
143 	{ PCI_VENDOR_BROADCOM, PCI_PRODUCT_BROADCOM_BCM57795,
144 		"Broadcom BCM57795 Fast Ethernet" },
145 
146 	{ 0, 0, NULL }
147 };
148 
149 static const int bnx_tx_mailbox[BNX_TX_RING_MAX] = {
150 	BGE_MBX_TX_HOST_PROD0_LO,
151 	BGE_MBX_TX_HOST_PROD0_HI,
152 	BGE_MBX_TX_HOST_PROD1_LO,
153 	BGE_MBX_TX_HOST_PROD1_HI
154 };
155 
156 #define BNX_IS_JUMBO_CAPABLE(sc)	((sc)->bnx_flags & BNX_FLAG_JUMBO)
157 #define BNX_IS_5717_PLUS(sc)		((sc)->bnx_flags & BNX_FLAG_5717_PLUS)
158 #define BNX_IS_57765_PLUS(sc)		((sc)->bnx_flags & BNX_FLAG_57765_PLUS)
159 #define BNX_IS_57765_FAMILY(sc)	 \
160 	((sc)->bnx_flags & BNX_FLAG_57765_FAMILY)
161 
162 typedef int	(*bnx_eaddr_fcn_t)(struct bnx_softc *, uint8_t[]);
163 
164 static int	bnx_probe(device_t);
165 static int	bnx_attach(device_t);
166 static int	bnx_detach(device_t);
167 static void	bnx_shutdown(device_t);
168 static int	bnx_suspend(device_t);
169 static int	bnx_resume(device_t);
170 static int	bnx_miibus_readreg(device_t, int, int);
171 static int	bnx_miibus_writereg(device_t, int, int, int);
172 static void	bnx_miibus_statchg(device_t);
173 
174 static int	bnx_handle_status(struct bnx_softc *);
175 #ifdef IFPOLL_ENABLE
176 static void	bnx_npoll(struct ifnet *, struct ifpoll_info *);
177 static void	bnx_npoll_rx(struct ifnet *, void *, int);
178 static void	bnx_npoll_tx(struct ifnet *, void *, int);
179 static void	bnx_npoll_tx_notag(struct ifnet *, void *, int);
180 static void	bnx_npoll_status(struct ifnet *);
181 static void	bnx_npoll_status_notag(struct ifnet *);
182 #endif
183 static void	bnx_intr_legacy(void *);
184 static void	bnx_msi(void *);
185 static void	bnx_intr(struct bnx_softc *);
186 static void	bnx_msix_status(void *);
187 static void	bnx_msix_tx_status(void *);
188 static void	bnx_msix_rx(void *);
189 static void	bnx_msix_rxtx(void *);
190 static void	bnx_enable_intr(struct bnx_softc *);
191 static void	bnx_disable_intr(struct bnx_softc *);
192 static void	bnx_txeof(struct bnx_tx_ring *, uint16_t);
193 static void	bnx_rxeof(struct bnx_rx_ret_ring *, uint16_t, int);
194 static int	bnx_alloc_intr(struct bnx_softc *);
195 static int	bnx_setup_intr(struct bnx_softc *);
196 static void	bnx_free_intr(struct bnx_softc *);
197 static void	bnx_teardown_intr(struct bnx_softc *, int);
198 static int	bnx_alloc_msix(struct bnx_softc *);
199 static void	bnx_free_msix(struct bnx_softc *, boolean_t);
200 static void	bnx_check_intr_rxtx(void *);
201 static void	bnx_check_intr_rx(void *);
202 static void	bnx_check_intr_tx(void *);
203 static void	bnx_rx_std_refill_ithread(void *);
204 static void	bnx_rx_std_refill(void *, void *);
205 static void	bnx_rx_std_refill_sched_ipi(void *);
206 static void	bnx_rx_std_refill_stop(void *);
207 static void	bnx_rx_std_refill_sched(struct bnx_rx_ret_ring *,
208 		    struct bnx_rx_std_ring *);
209 
210 static void	bnx_start(struct ifnet *, struct ifaltq_subque *);
211 static int	bnx_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
212 static void	bnx_init(void *);
213 static void	bnx_stop(struct bnx_softc *);
214 static void	bnx_watchdog(struct ifaltq_subque *);
215 static int	bnx_ifmedia_upd(struct ifnet *);
216 static void	bnx_ifmedia_sts(struct ifnet *, struct ifmediareq *);
217 static void	bnx_tick(void *);
218 static void	bnx_serialize(struct ifnet *, enum ifnet_serialize);
219 static void	bnx_deserialize(struct ifnet *, enum ifnet_serialize);
220 static int	bnx_tryserialize(struct ifnet *, enum ifnet_serialize);
221 #ifdef INVARIANTS
222 static void	bnx_serialize_assert(struct ifnet *, enum ifnet_serialize,
223 		    boolean_t);
224 #endif
225 static void	bnx_serialize_skipmain(struct bnx_softc *);
226 static void	bnx_deserialize_skipmain(struct bnx_softc *sc);
227 
228 static int	bnx_alloc_jumbo_mem(struct bnx_softc *);
229 static void	bnx_free_jumbo_mem(struct bnx_softc *);
230 static struct bnx_jslot
231 		*bnx_jalloc(struct bnx_softc *);
232 static void	bnx_jfree(void *);
233 static void	bnx_jref(void *);
234 static int	bnx_newbuf_std(struct bnx_rx_ret_ring *, int, int);
235 static int	bnx_newbuf_jumbo(struct bnx_softc *, int, int);
236 static void	bnx_setup_rxdesc_std(struct bnx_rx_std_ring *, int);
237 static void	bnx_setup_rxdesc_jumbo(struct bnx_softc *, int);
238 static int	bnx_init_rx_ring_std(struct bnx_rx_std_ring *);
239 static void	bnx_free_rx_ring_std(struct bnx_rx_std_ring *);
240 static int	bnx_init_rx_ring_jumbo(struct bnx_softc *);
241 static void	bnx_free_rx_ring_jumbo(struct bnx_softc *);
242 static void	bnx_free_tx_ring(struct bnx_tx_ring *);
243 static int	bnx_init_tx_ring(struct bnx_tx_ring *);
244 static int	bnx_create_tx_ring(struct bnx_tx_ring *);
245 static void	bnx_destroy_tx_ring(struct bnx_tx_ring *);
246 static int	bnx_create_rx_ret_ring(struct bnx_rx_ret_ring *);
247 static void	bnx_destroy_rx_ret_ring(struct bnx_rx_ret_ring *);
248 static int	bnx_dma_alloc(device_t);
249 static void	bnx_dma_free(struct bnx_softc *);
250 static int	bnx_dma_block_alloc(struct bnx_softc *, bus_size_t,
251 		    bus_dma_tag_t *, bus_dmamap_t *, void **, bus_addr_t *);
252 static void	bnx_dma_block_free(bus_dma_tag_t, bus_dmamap_t, void *);
253 static struct mbuf *
254 		bnx_defrag_shortdma(struct mbuf *);
255 static int	bnx_encap(struct bnx_tx_ring *, struct mbuf **,
256 		    uint32_t *, int *);
257 static int	bnx_setup_tso(struct bnx_tx_ring *, struct mbuf **,
258 		    uint16_t *, uint16_t *);
259 static void	bnx_setup_serialize(struct bnx_softc *);
260 static void	bnx_set_tick_cpuid(struct bnx_softc *, boolean_t);
261 static void	bnx_setup_ring_cnt(struct bnx_softc *);
262 
263 static struct pktinfo *bnx_rss_info(struct pktinfo *,
264 		    const struct bge_rx_bd *);
265 static void	bnx_init_rss(struct bnx_softc *);
266 static void	bnx_reset(struct bnx_softc *);
267 static int	bnx_chipinit(struct bnx_softc *);
268 static int	bnx_blockinit(struct bnx_softc *);
269 static void	bnx_stop_block(struct bnx_softc *, bus_size_t, uint32_t);
270 static void	bnx_enable_msi(struct bnx_softc *, boolean_t);
271 static void	bnx_setmulti(struct bnx_softc *);
272 static void	bnx_setpromisc(struct bnx_softc *);
273 static void	bnx_stats_update_regs(struct bnx_softc *);
274 static uint32_t	bnx_dma_swap_options(struct bnx_softc *);
275 
276 static uint32_t	bnx_readmem_ind(struct bnx_softc *, uint32_t);
277 static void	bnx_writemem_ind(struct bnx_softc *, uint32_t, uint32_t);
278 #ifdef notdef
279 static uint32_t	bnx_readreg_ind(struct bnx_softc *, uint32_t);
280 #endif
281 static void	bnx_writemem_direct(struct bnx_softc *, uint32_t, uint32_t);
282 static void	bnx_writembx(struct bnx_softc *, int, int);
283 static int	bnx_read_nvram(struct bnx_softc *, caddr_t, int, int);
284 static uint8_t	bnx_eeprom_getbyte(struct bnx_softc *, uint32_t, uint8_t *);
285 static int	bnx_read_eeprom(struct bnx_softc *, caddr_t, uint32_t, size_t);
286 
287 static void	bnx_tbi_link_upd(struct bnx_softc *, uint32_t);
288 static void	bnx_copper_link_upd(struct bnx_softc *, uint32_t);
289 static void	bnx_autopoll_link_upd(struct bnx_softc *, uint32_t);
290 static void	bnx_link_poll(struct bnx_softc *);
291 
292 static int	bnx_get_eaddr_mem(struct bnx_softc *, uint8_t[]);
293 static int	bnx_get_eaddr_nvram(struct bnx_softc *, uint8_t[]);
294 static int	bnx_get_eaddr_eeprom(struct bnx_softc *, uint8_t[]);
295 static int	bnx_get_eaddr(struct bnx_softc *, uint8_t[]);
296 
297 static void	bnx_coal_change(struct bnx_softc *);
298 static int	bnx_sysctl_force_defrag(SYSCTL_HANDLER_ARGS);
299 static int	bnx_sysctl_tx_wreg(SYSCTL_HANDLER_ARGS);
300 static int	bnx_sysctl_rx_coal_ticks(SYSCTL_HANDLER_ARGS);
301 static int	bnx_sysctl_tx_coal_ticks(SYSCTL_HANDLER_ARGS);
302 static int	bnx_sysctl_rx_coal_bds(SYSCTL_HANDLER_ARGS);
303 static int	bnx_sysctl_rx_coal_bds_poll(SYSCTL_HANDLER_ARGS);
304 static int	bnx_sysctl_tx_coal_bds(SYSCTL_HANDLER_ARGS);
305 static int	bnx_sysctl_tx_coal_bds_poll(SYSCTL_HANDLER_ARGS);
306 static int	bnx_sysctl_rx_coal_bds_int(SYSCTL_HANDLER_ARGS);
307 static int	bnx_sysctl_tx_coal_bds_int(SYSCTL_HANDLER_ARGS);
308 static int	bnx_sysctl_coal_chg(SYSCTL_HANDLER_ARGS, uint32_t *,
309 		    int, int, uint32_t);
310 #ifdef IFPOLL_ENABLE
311 static int	bnx_sysctl_npoll_offset(SYSCTL_HANDLER_ARGS);
312 static int	bnx_sysctl_npoll_rxoff(SYSCTL_HANDLER_ARGS);
313 static int	bnx_sysctl_npoll_txoff(SYSCTL_HANDLER_ARGS);
314 #endif
315 static int	bnx_sysctl_std_refill(SYSCTL_HANDLER_ARGS);
316 
317 static void	bnx_sig_post_reset(struct bnx_softc *, int);
318 static void	bnx_sig_pre_reset(struct bnx_softc *, int);
319 static void	bnx_ape_lock_init(struct bnx_softc *);
320 static void	bnx_ape_read_fw_ver(struct bnx_softc *);
321 static int	bnx_ape_lock(struct bnx_softc *, int);
322 static void	bnx_ape_unlock(struct bnx_softc *, int);
323 static void	bnx_ape_send_event(struct bnx_softc *, uint32_t);
324 static void	bnx_ape_driver_state_change(struct bnx_softc *, int);
325 
326 static int	bnx_msi_enable = 1;
327 static int	bnx_msix_enable = 1;
328 
329 static int	bnx_rx_rings = 0; /* auto */
330 static int	bnx_tx_rings = 0; /* auto */
331 
332 TUNABLE_INT("hw.bnx.msi.enable", &bnx_msi_enable);
333 TUNABLE_INT("hw.bnx.msix.enable", &bnx_msix_enable);
334 TUNABLE_INT("hw.bnx.rx_rings", &bnx_rx_rings);
335 TUNABLE_INT("hw.bnx.tx_rings", &bnx_tx_rings);
336 
337 static device_method_t bnx_methods[] = {
338 	/* Device interface */
339 	DEVMETHOD(device_probe,		bnx_probe),
340 	DEVMETHOD(device_attach,	bnx_attach),
341 	DEVMETHOD(device_detach,	bnx_detach),
342 	DEVMETHOD(device_shutdown,	bnx_shutdown),
343 	DEVMETHOD(device_suspend,	bnx_suspend),
344 	DEVMETHOD(device_resume,	bnx_resume),
345 
346 	/* bus interface */
347 	DEVMETHOD(bus_print_child,	bus_generic_print_child),
348 	DEVMETHOD(bus_driver_added,	bus_generic_driver_added),
349 
350 	/* MII interface */
351 	DEVMETHOD(miibus_readreg,	bnx_miibus_readreg),
352 	DEVMETHOD(miibus_writereg,	bnx_miibus_writereg),
353 	DEVMETHOD(miibus_statchg,	bnx_miibus_statchg),
354 
355 	DEVMETHOD_END
356 };
357 
358 static DEFINE_CLASS_0(bnx, bnx_driver, bnx_methods, sizeof(struct bnx_softc));
359 static devclass_t bnx_devclass;
360 
361 DECLARE_DUMMY_MODULE(if_bnx);
362 DRIVER_MODULE(if_bnx, pci, bnx_driver, bnx_devclass, NULL, NULL);
363 DRIVER_MODULE(miibus, bnx, miibus_driver, miibus_devclass, NULL, NULL);
364 
365 static uint32_t
366 bnx_readmem_ind(struct bnx_softc *sc, uint32_t off)
367 {
368 	device_t dev = sc->bnx_dev;
369 	uint32_t val;
370 
371 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, off, 4);
372 	val = pci_read_config(dev, BGE_PCI_MEMWIN_DATA, 4);
373 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, 0, 4);
374 	return (val);
375 }
376 
377 static void
378 bnx_writemem_ind(struct bnx_softc *sc, uint32_t off, uint32_t val)
379 {
380 	device_t dev = sc->bnx_dev;
381 
382 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, off, 4);
383 	pci_write_config(dev, BGE_PCI_MEMWIN_DATA, val, 4);
384 	pci_write_config(dev, BGE_PCI_MEMWIN_BASEADDR, 0, 4);
385 }
386 
387 static void
388 bnx_writemem_direct(struct bnx_softc *sc, uint32_t off, uint32_t val)
389 {
390 	CSR_WRITE_4(sc, off, val);
391 }
392 
393 static void
394 bnx_writembx(struct bnx_softc *sc, int off, int val)
395 {
396 	CSR_WRITE_4(sc, off, val);
397 }
398 
399 /*
400  * Read a sequence of bytes from NVRAM.
401  */
402 static int
403 bnx_read_nvram(struct bnx_softc *sc, caddr_t dest, int off, int cnt)
404 {
405 	return (1);
406 }
407 
408 /*
409  * Read a byte of data stored in the EEPROM at address 'addr.' The
410  * BCM570x supports both the traditional bitbang interface and an
411  * auto access interface for reading the EEPROM. We use the auto
412  * access method.
413  */
414 static uint8_t
415 bnx_eeprom_getbyte(struct bnx_softc *sc, uint32_t addr, uint8_t *dest)
416 {
417 	int i;
418 	uint32_t byte = 0;
419 
420 	/*
421 	 * Enable use of auto EEPROM access so we can avoid
422 	 * having to use the bitbang method.
423 	 */
424 	BNX_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_AUTO_EEPROM);
425 
426 	/* Reset the EEPROM, load the clock period. */
427 	CSR_WRITE_4(sc, BGE_EE_ADDR,
428 	    BGE_EEADDR_RESET|BGE_EEHALFCLK(BGE_HALFCLK_384SCL));
429 	DELAY(20);
430 
431 	/* Issue the read EEPROM command. */
432 	CSR_WRITE_4(sc, BGE_EE_ADDR, BGE_EE_READCMD | addr);
433 
434 	/* Wait for completion */
435 	for(i = 0; i < BNX_TIMEOUT * 10; i++) {
436 		DELAY(10);
437 		if (CSR_READ_4(sc, BGE_EE_ADDR) & BGE_EEADDR_DONE)
438 			break;
439 	}
440 
441 	if (i == BNX_TIMEOUT) {
442 		if_printf(&sc->arpcom.ac_if, "eeprom read timed out\n");
443 		return(1);
444 	}
445 
446 	/* Get result. */
447 	byte = CSR_READ_4(sc, BGE_EE_DATA);
448 
449         *dest = (byte >> ((addr % 4) * 8)) & 0xFF;
450 
451 	return(0);
452 }
453 
454 /*
455  * Read a sequence of bytes from the EEPROM.
456  */
457 static int
458 bnx_read_eeprom(struct bnx_softc *sc, caddr_t dest, uint32_t off, size_t len)
459 {
460 	size_t i;
461 	int err;
462 	uint8_t byte;
463 
464 	for (byte = 0, err = 0, i = 0; i < len; i++) {
465 		err = bnx_eeprom_getbyte(sc, off + i, &byte);
466 		if (err)
467 			break;
468 		*(dest + i) = byte;
469 	}
470 
471 	return(err ? 1 : 0);
472 }
473 
474 static int
475 bnx_miibus_readreg(device_t dev, int phy, int reg)
476 {
477 	struct bnx_softc *sc = device_get_softc(dev);
478 	uint32_t val;
479 	int i;
480 
481 	KASSERT(phy == sc->bnx_phyno,
482 	    ("invalid phyno %d, should be %d", phy, sc->bnx_phyno));
483 
484 	if (bnx_ape_lock(sc, sc->bnx_phy_ape_lock) != 0)
485 		return 0;
486 
487 	/* Clear the autopoll bit if set, otherwise may trigger PCI errors. */
488 	if (sc->bnx_mi_mode & BGE_MIMODE_AUTOPOLL) {
489 		CSR_WRITE_4(sc, BGE_MI_MODE,
490 		    sc->bnx_mi_mode & ~BGE_MIMODE_AUTOPOLL);
491 		DELAY(80);
492 	}
493 
494 	CSR_WRITE_4(sc, BGE_MI_COMM, BGE_MICMD_READ | BGE_MICOMM_BUSY |
495 	    BGE_MIPHY(phy) | BGE_MIREG(reg));
496 
497 	/* Poll for the PHY register access to complete. */
498 	for (i = 0; i < BNX_TIMEOUT; i++) {
499 		DELAY(10);
500 		val = CSR_READ_4(sc, BGE_MI_COMM);
501 		if ((val & BGE_MICOMM_BUSY) == 0) {
502 			DELAY(5);
503 			val = CSR_READ_4(sc, BGE_MI_COMM);
504 			break;
505 		}
506 	}
507 	if (i == BNX_TIMEOUT) {
508 		if_printf(&sc->arpcom.ac_if, "PHY read timed out "
509 		    "(phy %d, reg %d, val 0x%08x)\n", phy, reg, val);
510 		val = 0;
511 	}
512 
513 	/* Restore the autopoll bit if necessary. */
514 	if (sc->bnx_mi_mode & BGE_MIMODE_AUTOPOLL) {
515 		CSR_WRITE_4(sc, BGE_MI_MODE, sc->bnx_mi_mode);
516 		DELAY(80);
517 	}
518 
519 	bnx_ape_unlock(sc, sc->bnx_phy_ape_lock);
520 
521 	if (val & BGE_MICOMM_READFAIL)
522 		return 0;
523 
524 	return (val & 0xFFFF);
525 }
526 
527 static int
528 bnx_miibus_writereg(device_t dev, int phy, int reg, int val)
529 {
530 	struct bnx_softc *sc = device_get_softc(dev);
531 	int i;
532 
533 	KASSERT(phy == sc->bnx_phyno,
534 	    ("invalid phyno %d, should be %d", phy, sc->bnx_phyno));
535 
536 	if (bnx_ape_lock(sc, sc->bnx_phy_ape_lock) != 0)
537 		return 0;
538 
539 	/* Clear the autopoll bit if set, otherwise may trigger PCI errors. */
540 	if (sc->bnx_mi_mode & BGE_MIMODE_AUTOPOLL) {
541 		CSR_WRITE_4(sc, BGE_MI_MODE,
542 		    sc->bnx_mi_mode & ~BGE_MIMODE_AUTOPOLL);
543 		DELAY(80);
544 	}
545 
546 	CSR_WRITE_4(sc, BGE_MI_COMM, BGE_MICMD_WRITE | BGE_MICOMM_BUSY |
547 	    BGE_MIPHY(phy) | BGE_MIREG(reg) | val);
548 
549 	for (i = 0; i < BNX_TIMEOUT; i++) {
550 		DELAY(10);
551 		if (!(CSR_READ_4(sc, BGE_MI_COMM) & BGE_MICOMM_BUSY)) {
552 			DELAY(5);
553 			CSR_READ_4(sc, BGE_MI_COMM); /* dummy read */
554 			break;
555 		}
556 	}
557 	if (i == BNX_TIMEOUT) {
558 		if_printf(&sc->arpcom.ac_if, "PHY write timed out "
559 		    "(phy %d, reg %d, val %d)\n", phy, reg, val);
560 	}
561 
562 	/* Restore the autopoll bit if necessary. */
563 	if (sc->bnx_mi_mode & BGE_MIMODE_AUTOPOLL) {
564 		CSR_WRITE_4(sc, BGE_MI_MODE, sc->bnx_mi_mode);
565 		DELAY(80);
566 	}
567 
568 	bnx_ape_unlock(sc, sc->bnx_phy_ape_lock);
569 
570 	return 0;
571 }
572 
573 static void
574 bnx_miibus_statchg(device_t dev)
575 {
576 	struct bnx_softc *sc;
577 	struct mii_data *mii;
578 	uint32_t mac_mode;
579 
580 	sc = device_get_softc(dev);
581 	if ((sc->arpcom.ac_if.if_flags & IFF_RUNNING) == 0)
582 		return;
583 
584 	mii = device_get_softc(sc->bnx_miibus);
585 
586 	if ((mii->mii_media_status & (IFM_ACTIVE | IFM_AVALID)) ==
587 	    (IFM_ACTIVE | IFM_AVALID)) {
588 		switch (IFM_SUBTYPE(mii->mii_media_active)) {
589 		case IFM_10_T:
590 		case IFM_100_TX:
591 			sc->bnx_link = 1;
592 			break;
593 		case IFM_1000_T:
594 		case IFM_1000_SX:
595 		case IFM_2500_SX:
596 			sc->bnx_link = 1;
597 			break;
598 		default:
599 			sc->bnx_link = 0;
600 			break;
601 		}
602 	} else {
603 		sc->bnx_link = 0;
604 	}
605 	if (sc->bnx_link == 0)
606 		return;
607 
608 	/*
609 	 * APE firmware touches these registers to keep the MAC
610 	 * connected to the outside world.  Try to keep the
611 	 * accesses atomic.
612 	 */
613 
614 	mac_mode = CSR_READ_4(sc, BGE_MAC_MODE) &
615 	    ~(BGE_MACMODE_PORTMODE | BGE_MACMODE_HALF_DUPLEX);
616 
617 	if (IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_T ||
618 	    IFM_SUBTYPE(mii->mii_media_active) == IFM_1000_SX)
619 		mac_mode |= BGE_PORTMODE_GMII;
620 	else
621 		mac_mode |= BGE_PORTMODE_MII;
622 
623 	if ((mii->mii_media_active & IFM_GMASK) != IFM_FDX)
624 		mac_mode |= BGE_MACMODE_HALF_DUPLEX;
625 
626 	CSR_WRITE_4(sc, BGE_MAC_MODE, mac_mode);
627 	DELAY(40);
628 }
629 
630 /*
631  * Memory management for jumbo frames.
632  */
633 static int
634 bnx_alloc_jumbo_mem(struct bnx_softc *sc)
635 {
636 	struct ifnet *ifp = &sc->arpcom.ac_if;
637 	struct bnx_jslot *entry;
638 	uint8_t *ptr;
639 	bus_addr_t paddr;
640 	int i, error;
641 
642 	/*
643 	 * Create tag for jumbo mbufs.
644 	 * This is really a bit of a kludge. We allocate a special
645 	 * jumbo buffer pool which (thanks to the way our DMA
646 	 * memory allocation works) will consist of contiguous
647 	 * pages. This means that even though a jumbo buffer might
648 	 * be larger than a page size, we don't really need to
649 	 * map it into more than one DMA segment. However, the
650 	 * default mbuf tag will result in multi-segment mappings,
651 	 * so we have to create a special jumbo mbuf tag that
652 	 * lets us get away with mapping the jumbo buffers as
653 	 * a single segment. I think eventually the driver should
654 	 * be changed so that it uses ordinary mbufs and cluster
655 	 * buffers, i.e. jumbo frames can span multiple DMA
656 	 * descriptors. But that's a project for another day.
657 	 */
658 
659 	/*
660 	 * Create DMA stuffs for jumbo RX ring.
661 	 */
662 	error = bnx_dma_block_alloc(sc, BGE_JUMBO_RX_RING_SZ,
663 				    &sc->bnx_cdata.bnx_rx_jumbo_ring_tag,
664 				    &sc->bnx_cdata.bnx_rx_jumbo_ring_map,
665 				    (void *)&sc->bnx_ldata.bnx_rx_jumbo_ring,
666 				    &sc->bnx_ldata.bnx_rx_jumbo_ring_paddr);
667 	if (error) {
668 		if_printf(ifp, "could not create jumbo RX ring\n");
669 		return error;
670 	}
671 
672 	/*
673 	 * Create DMA stuffs for jumbo buffer block.
674 	 */
675 	error = bnx_dma_block_alloc(sc, BNX_JMEM,
676 				    &sc->bnx_cdata.bnx_jumbo_tag,
677 				    &sc->bnx_cdata.bnx_jumbo_map,
678 				    (void **)&sc->bnx_ldata.bnx_jumbo_buf,
679 				    &paddr);
680 	if (error) {
681 		if_printf(ifp, "could not create jumbo buffer\n");
682 		return error;
683 	}
684 
685 	SLIST_INIT(&sc->bnx_jfree_listhead);
686 
687 	/*
688 	 * Now divide it up into 9K pieces and save the addresses
689 	 * in an array. Note that we play an evil trick here by using
690 	 * the first few bytes in the buffer to hold the the address
691 	 * of the softc structure for this interface. This is because
692 	 * bnx_jfree() needs it, but it is called by the mbuf management
693 	 * code which will not pass it to us explicitly.
694 	 */
695 	for (i = 0, ptr = sc->bnx_ldata.bnx_jumbo_buf; i < BNX_JSLOTS; i++) {
696 		entry = &sc->bnx_cdata.bnx_jslots[i];
697 		entry->bnx_sc = sc;
698 		entry->bnx_buf = ptr;
699 		entry->bnx_paddr = paddr;
700 		entry->bnx_inuse = 0;
701 		entry->bnx_slot = i;
702 		SLIST_INSERT_HEAD(&sc->bnx_jfree_listhead, entry, jslot_link);
703 
704 		ptr += BNX_JLEN;
705 		paddr += BNX_JLEN;
706 	}
707 	return 0;
708 }
709 
710 static void
711 bnx_free_jumbo_mem(struct bnx_softc *sc)
712 {
713 	/* Destroy jumbo RX ring. */
714 	bnx_dma_block_free(sc->bnx_cdata.bnx_rx_jumbo_ring_tag,
715 			   sc->bnx_cdata.bnx_rx_jumbo_ring_map,
716 			   sc->bnx_ldata.bnx_rx_jumbo_ring);
717 
718 	/* Destroy jumbo buffer block. */
719 	bnx_dma_block_free(sc->bnx_cdata.bnx_jumbo_tag,
720 			   sc->bnx_cdata.bnx_jumbo_map,
721 			   sc->bnx_ldata.bnx_jumbo_buf);
722 }
723 
724 /*
725  * Allocate a jumbo buffer.
726  */
727 static struct bnx_jslot *
728 bnx_jalloc(struct bnx_softc *sc)
729 {
730 	struct bnx_jslot *entry;
731 
732 	lwkt_serialize_enter(&sc->bnx_jslot_serializer);
733 	entry = SLIST_FIRST(&sc->bnx_jfree_listhead);
734 	if (entry) {
735 		SLIST_REMOVE_HEAD(&sc->bnx_jfree_listhead, jslot_link);
736 		entry->bnx_inuse = 1;
737 	} else {
738 		if_printf(&sc->arpcom.ac_if, "no free jumbo buffers\n");
739 	}
740 	lwkt_serialize_exit(&sc->bnx_jslot_serializer);
741 	return(entry);
742 }
743 
744 /*
745  * Adjust usage count on a jumbo buffer.
746  */
747 static void
748 bnx_jref(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_jref: can't find softc pointer!");
755 
756 	if (&sc->bnx_cdata.bnx_jslots[entry->bnx_slot] != entry) {
757 		panic("bnx_jref: asked to reference buffer "
758 		    "that we don't manage!");
759 	} else if (entry->bnx_inuse == 0) {
760 		panic("bnx_jref: buffer already free!");
761 	} else {
762 		atomic_add_int(&entry->bnx_inuse, 1);
763 	}
764 }
765 
766 /*
767  * Release a jumbo buffer.
768  */
769 static void
770 bnx_jfree(void *arg)
771 {
772 	struct bnx_jslot *entry = (struct bnx_jslot *)arg;
773 	struct bnx_softc *sc = entry->bnx_sc;
774 
775 	if (sc == NULL)
776 		panic("bnx_jfree: can't find softc pointer!");
777 
778 	if (&sc->bnx_cdata.bnx_jslots[entry->bnx_slot] != entry) {
779 		panic("bnx_jfree: asked to free buffer that we don't manage!");
780 	} else if (entry->bnx_inuse == 0) {
781 		panic("bnx_jfree: buffer already free!");
782 	} else {
783 		/*
784 		 * Possible MP race to 0, use the serializer.  The atomic insn
785 		 * is still needed for races against bnx_jref().
786 		 */
787 		lwkt_serialize_enter(&sc->bnx_jslot_serializer);
788 		atomic_subtract_int(&entry->bnx_inuse, 1);
789 		if (entry->bnx_inuse == 0) {
790 			SLIST_INSERT_HEAD(&sc->bnx_jfree_listhead,
791 					  entry, jslot_link);
792 		}
793 		lwkt_serialize_exit(&sc->bnx_jslot_serializer);
794 	}
795 }
796 
797 
798 /*
799  * Intialize a standard receive ring descriptor.
800  */
801 static int
802 bnx_newbuf_std(struct bnx_rx_ret_ring *ret, int i, int init)
803 {
804 	struct mbuf *m_new = NULL;
805 	bus_dma_segment_t seg;
806 	bus_dmamap_t map;
807 	int error, nsegs;
808 	struct bnx_rx_buf *rb;
809 
810 	rb = &ret->bnx_std->bnx_rx_std_buf[i];
811 	KASSERT(!rb->bnx_rx_refilled, ("RX buf %dth has been refilled", i));
812 
813 	m_new = m_getcl(init ? MB_WAIT : MB_DONTWAIT, MT_DATA, M_PKTHDR);
814 	if (m_new == NULL) {
815 		error = ENOBUFS;
816 		goto back;
817 	}
818 	m_new->m_len = m_new->m_pkthdr.len = MCLBYTES;
819 	m_adj(m_new, ETHER_ALIGN);
820 
821 	error = bus_dmamap_load_mbuf_segment(ret->bnx_rx_mtag,
822 	    ret->bnx_rx_tmpmap, m_new, &seg, 1, &nsegs, BUS_DMA_NOWAIT);
823 	if (error) {
824 		m_freem(m_new);
825 		goto back;
826 	}
827 
828 	if (!init) {
829 		bus_dmamap_sync(ret->bnx_rx_mtag, rb->bnx_rx_dmamap,
830 		    BUS_DMASYNC_POSTREAD);
831 		bus_dmamap_unload(ret->bnx_rx_mtag, rb->bnx_rx_dmamap);
832 	}
833 
834 	map = ret->bnx_rx_tmpmap;
835 	ret->bnx_rx_tmpmap = rb->bnx_rx_dmamap;
836 
837 	rb->bnx_rx_dmamap = map;
838 	rb->bnx_rx_mbuf = m_new;
839 	rb->bnx_rx_paddr = seg.ds_addr;
840 	rb->bnx_rx_len = m_new->m_len;
841 back:
842 	cpu_sfence();
843 	rb->bnx_rx_refilled = 1;
844 	return error;
845 }
846 
847 static void
848 bnx_setup_rxdesc_std(struct bnx_rx_std_ring *std, int i)
849 {
850 	struct bnx_rx_buf *rb;
851 	struct bge_rx_bd *r;
852 	bus_addr_t paddr;
853 	int len;
854 
855 	rb = &std->bnx_rx_std_buf[i];
856 	KASSERT(rb->bnx_rx_refilled, ("RX buf %dth is not refilled", i));
857 
858 	paddr = rb->bnx_rx_paddr;
859 	len = rb->bnx_rx_len;
860 
861 	cpu_mfence();
862 
863 	rb->bnx_rx_refilled = 0;
864 
865 	r = &std->bnx_rx_std_ring[i];
866 	r->bge_addr.bge_addr_lo = BGE_ADDR_LO(paddr);
867 	r->bge_addr.bge_addr_hi = BGE_ADDR_HI(paddr);
868 	r->bge_len = len;
869 	r->bge_idx = i;
870 	r->bge_flags = BGE_RXBDFLAG_END;
871 }
872 
873 /*
874  * Initialize a jumbo receive ring descriptor. This allocates
875  * a jumbo buffer from the pool managed internally by the driver.
876  */
877 static int
878 bnx_newbuf_jumbo(struct bnx_softc *sc, int i, int init)
879 {
880 	struct mbuf *m_new = NULL;
881 	struct bnx_jslot *buf;
882 	bus_addr_t paddr;
883 
884 	/* Allocate the mbuf. */
885 	MGETHDR(m_new, init ? MB_WAIT : MB_DONTWAIT, MT_DATA);
886 	if (m_new == NULL)
887 		return ENOBUFS;
888 
889 	/* Allocate the jumbo buffer */
890 	buf = bnx_jalloc(sc);
891 	if (buf == NULL) {
892 		m_freem(m_new);
893 		return ENOBUFS;
894 	}
895 
896 	/* Attach the buffer to the mbuf. */
897 	m_new->m_ext.ext_arg = buf;
898 	m_new->m_ext.ext_buf = buf->bnx_buf;
899 	m_new->m_ext.ext_free = bnx_jfree;
900 	m_new->m_ext.ext_ref = bnx_jref;
901 	m_new->m_ext.ext_size = BNX_JUMBO_FRAMELEN;
902 
903 	m_new->m_flags |= M_EXT;
904 
905 	m_new->m_data = m_new->m_ext.ext_buf;
906 	m_new->m_len = m_new->m_pkthdr.len = m_new->m_ext.ext_size;
907 
908 	paddr = buf->bnx_paddr;
909 	m_adj(m_new, ETHER_ALIGN);
910 	paddr += ETHER_ALIGN;
911 
912 	/* Save necessary information */
913 	sc->bnx_cdata.bnx_rx_jumbo_chain[i].bnx_rx_mbuf = m_new;
914 	sc->bnx_cdata.bnx_rx_jumbo_chain[i].bnx_rx_paddr = paddr;
915 
916 	/* Set up the descriptor. */
917 	bnx_setup_rxdesc_jumbo(sc, i);
918 	return 0;
919 }
920 
921 static void
922 bnx_setup_rxdesc_jumbo(struct bnx_softc *sc, int i)
923 {
924 	struct bge_rx_bd *r;
925 	struct bnx_rx_buf *rc;
926 
927 	r = &sc->bnx_ldata.bnx_rx_jumbo_ring[i];
928 	rc = &sc->bnx_cdata.bnx_rx_jumbo_chain[i];
929 
930 	r->bge_addr.bge_addr_lo = BGE_ADDR_LO(rc->bnx_rx_paddr);
931 	r->bge_addr.bge_addr_hi = BGE_ADDR_HI(rc->bnx_rx_paddr);
932 	r->bge_len = rc->bnx_rx_mbuf->m_len;
933 	r->bge_idx = i;
934 	r->bge_flags = BGE_RXBDFLAG_END|BGE_RXBDFLAG_JUMBO_RING;
935 }
936 
937 static int
938 bnx_init_rx_ring_std(struct bnx_rx_std_ring *std)
939 {
940 	int i, error;
941 
942 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
943 		/* Use the first RX return ring's tmp RX mbuf DMA map */
944 		error = bnx_newbuf_std(&std->bnx_sc->bnx_rx_ret_ring[0], i, 1);
945 		if (error)
946 			return error;
947 		bnx_setup_rxdesc_std(std, i);
948 	}
949 
950 	std->bnx_rx_std_used = 0;
951 	std->bnx_rx_std_refill = 0;
952 	std->bnx_rx_std_running = 0;
953 	cpu_sfence();
954 	lwkt_serialize_handler_enable(&std->bnx_rx_std_serialize);
955 
956 	std->bnx_rx_std = BGE_STD_RX_RING_CNT - 1;
957 	bnx_writembx(std->bnx_sc, BGE_MBX_RX_STD_PROD_LO, std->bnx_rx_std);
958 
959 	return(0);
960 }
961 
962 static void
963 bnx_free_rx_ring_std(struct bnx_rx_std_ring *std)
964 {
965 	int i;
966 
967 	lwkt_serialize_handler_disable(&std->bnx_rx_std_serialize);
968 
969 	for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
970 		struct bnx_rx_buf *rb = &std->bnx_rx_std_buf[i];
971 
972 		rb->bnx_rx_refilled = 0;
973 		if (rb->bnx_rx_mbuf != NULL) {
974 			bus_dmamap_unload(std->bnx_rx_mtag, rb->bnx_rx_dmamap);
975 			m_freem(rb->bnx_rx_mbuf);
976 			rb->bnx_rx_mbuf = NULL;
977 		}
978 		bzero(&std->bnx_rx_std_ring[i], sizeof(struct bge_rx_bd));
979 	}
980 }
981 
982 static int
983 bnx_init_rx_ring_jumbo(struct bnx_softc *sc)
984 {
985 	struct bge_rcb *rcb;
986 	int i, error;
987 
988 	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
989 		error = bnx_newbuf_jumbo(sc, i, 1);
990 		if (error)
991 			return error;
992 	}
993 
994 	sc->bnx_jumbo = BGE_JUMBO_RX_RING_CNT - 1;
995 
996 	rcb = &sc->bnx_ldata.bnx_info.bnx_jumbo_rx_rcb;
997 	rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(0, 0);
998 	CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
999 
1000 	bnx_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, sc->bnx_jumbo);
1001 
1002 	return(0);
1003 }
1004 
1005 static void
1006 bnx_free_rx_ring_jumbo(struct bnx_softc *sc)
1007 {
1008 	int i;
1009 
1010 	for (i = 0; i < BGE_JUMBO_RX_RING_CNT; i++) {
1011 		struct bnx_rx_buf *rc = &sc->bnx_cdata.bnx_rx_jumbo_chain[i];
1012 
1013 		if (rc->bnx_rx_mbuf != NULL) {
1014 			m_freem(rc->bnx_rx_mbuf);
1015 			rc->bnx_rx_mbuf = NULL;
1016 		}
1017 		bzero(&sc->bnx_ldata.bnx_rx_jumbo_ring[i],
1018 		    sizeof(struct bge_rx_bd));
1019 	}
1020 }
1021 
1022 static void
1023 bnx_free_tx_ring(struct bnx_tx_ring *txr)
1024 {
1025 	int i;
1026 
1027 	for (i = 0; i < BGE_TX_RING_CNT; i++) {
1028 		struct bnx_tx_buf *buf = &txr->bnx_tx_buf[i];
1029 
1030 		if (buf->bnx_tx_mbuf != NULL) {
1031 			bus_dmamap_unload(txr->bnx_tx_mtag,
1032 			    buf->bnx_tx_dmamap);
1033 			m_freem(buf->bnx_tx_mbuf);
1034 			buf->bnx_tx_mbuf = NULL;
1035 		}
1036 		bzero(&txr->bnx_tx_ring[i], sizeof(struct bge_tx_bd));
1037 	}
1038 	txr->bnx_tx_saved_considx = BNX_TXCONS_UNSET;
1039 }
1040 
1041 static int
1042 bnx_init_tx_ring(struct bnx_tx_ring *txr)
1043 {
1044 	txr->bnx_tx_cnt = 0;
1045 	txr->bnx_tx_saved_considx = 0;
1046 	txr->bnx_tx_prodidx = 0;
1047 
1048 	/* Initialize transmit producer index for host-memory send ring. */
1049 	bnx_writembx(txr->bnx_sc, txr->bnx_tx_mbx, txr->bnx_tx_prodidx);
1050 
1051 	return(0);
1052 }
1053 
1054 static void
1055 bnx_setmulti(struct bnx_softc *sc)
1056 {
1057 	struct ifnet *ifp;
1058 	struct ifmultiaddr *ifma;
1059 	uint32_t hashes[4] = { 0, 0, 0, 0 };
1060 	int h, i;
1061 
1062 	ifp = &sc->arpcom.ac_if;
1063 
1064 	if (ifp->if_flags & IFF_ALLMULTI || ifp->if_flags & IFF_PROMISC) {
1065 		for (i = 0; i < 4; i++)
1066 			CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), 0xFFFFFFFF);
1067 		return;
1068 	}
1069 
1070 	/* First, zot all the existing filters. */
1071 	for (i = 0; i < 4; i++)
1072 		CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), 0);
1073 
1074 	/* Now program new ones. */
1075 	TAILQ_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1076 		if (ifma->ifma_addr->sa_family != AF_LINK)
1077 			continue;
1078 		h = ether_crc32_le(
1079 		    LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
1080 		    ETHER_ADDR_LEN) & 0x7f;
1081 		hashes[(h & 0x60) >> 5] |= 1 << (h & 0x1F);
1082 	}
1083 
1084 	for (i = 0; i < 4; i++)
1085 		CSR_WRITE_4(sc, BGE_MAR0 + (i * 4), hashes[i]);
1086 }
1087 
1088 /*
1089  * Do endian, PCI and DMA initialization. Also check the on-board ROM
1090  * self-test results.
1091  */
1092 static int
1093 bnx_chipinit(struct bnx_softc *sc)
1094 {
1095 	uint32_t dma_rw_ctl, mode_ctl;
1096 	int i;
1097 
1098 	/* Set endian type before we access any non-PCI registers. */
1099 	pci_write_config(sc->bnx_dev, BGE_PCI_MISC_CTL,
1100 	    BGE_INIT | BGE_PCIMISCCTL_TAGGED_STATUS, 4);
1101 
1102 	/*
1103 	 * Clear the MAC statistics block in the NIC's
1104 	 * internal memory.
1105 	 */
1106 	for (i = BGE_STATS_BLOCK;
1107 	    i < BGE_STATS_BLOCK_END + 1; i += sizeof(uint32_t))
1108 		BNX_MEMWIN_WRITE(sc, i, 0);
1109 
1110 	for (i = BGE_STATUS_BLOCK;
1111 	    i < BGE_STATUS_BLOCK_END + 1; i += sizeof(uint32_t))
1112 		BNX_MEMWIN_WRITE(sc, i, 0);
1113 
1114 	if (BNX_IS_57765_FAMILY(sc)) {
1115 		uint32_t val;
1116 
1117 		if (sc->bnx_chipid == BGE_CHIPID_BCM57765_A0) {
1118 			mode_ctl = CSR_READ_4(sc, BGE_MODE_CTL);
1119 			val = mode_ctl & ~BGE_MODECTL_PCIE_PORTS;
1120 
1121 			/* Access the lower 1K of PL PCI-E block registers. */
1122 			CSR_WRITE_4(sc, BGE_MODE_CTL,
1123 			    val | BGE_MODECTL_PCIE_PL_SEL);
1124 
1125 			val = CSR_READ_4(sc, BGE_PCIE_PL_LO_PHYCTL5);
1126 			val |= BGE_PCIE_PL_LO_PHYCTL5_DIS_L2CLKREQ;
1127 			CSR_WRITE_4(sc, BGE_PCIE_PL_LO_PHYCTL5, val);
1128 
1129 			CSR_WRITE_4(sc, BGE_MODE_CTL, mode_ctl);
1130 		}
1131 		if (sc->bnx_chiprev != BGE_CHIPREV_57765_AX) {
1132 			/* Fix transmit hangs */
1133 			val = CSR_READ_4(sc, BGE_CPMU_PADRNG_CTL);
1134 			val |= BGE_CPMU_PADRNG_CTL_RDIV2;
1135 			CSR_WRITE_4(sc, BGE_CPMU_PADRNG_CTL, val);
1136 
1137 			mode_ctl = CSR_READ_4(sc, BGE_MODE_CTL);
1138 			val = mode_ctl & ~BGE_MODECTL_PCIE_PORTS;
1139 
1140 			/* Access the lower 1K of DL PCI-E block registers. */
1141 			CSR_WRITE_4(sc, BGE_MODE_CTL,
1142 			    val | BGE_MODECTL_PCIE_DL_SEL);
1143 
1144 			val = CSR_READ_4(sc, BGE_PCIE_DL_LO_FTSMAX);
1145 			val &= ~BGE_PCIE_DL_LO_FTSMAX_MASK;
1146 			val |= BGE_PCIE_DL_LO_FTSMAX_VAL;
1147 			CSR_WRITE_4(sc, BGE_PCIE_DL_LO_FTSMAX, val);
1148 
1149 			CSR_WRITE_4(sc, BGE_MODE_CTL, mode_ctl);
1150 		}
1151 
1152 		val = CSR_READ_4(sc, BGE_CPMU_LSPD_10MB_CLK);
1153 		val &= ~BGE_CPMU_LSPD_10MB_MACCLK_MASK;
1154 		val |= BGE_CPMU_LSPD_10MB_MACCLK_6_25;
1155 		CSR_WRITE_4(sc, BGE_CPMU_LSPD_10MB_CLK, val);
1156 	}
1157 
1158 	/*
1159 	 * Set up the PCI DMA control register.
1160 	 */
1161 	dma_rw_ctl = pci_read_config(sc->bnx_dev, BGE_PCI_DMA_RW_CTL, 4);
1162 	/*
1163 	 * Disable 32bytes cache alignment for DMA write to host memory
1164 	 *
1165 	 * NOTE:
1166 	 * 64bytes cache alignment for DMA write to host memory is still
1167 	 * enabled.
1168 	 */
1169 	dma_rw_ctl |= BGE_PCIDMARWCTL_DIS_CACHE_ALIGNMENT;
1170 	if (sc->bnx_chipid == BGE_CHIPID_BCM57765_A0)
1171 		dma_rw_ctl &= ~BGE_PCIDMARWCTL_CRDRDR_RDMA_MRRS_MSK;
1172 	/*
1173 	 * Enable HW workaround for controllers that misinterpret
1174 	 * a status tag update and leave interrupts permanently
1175 	 * disabled.
1176 	 */
1177 	if (sc->bnx_asicrev != BGE_ASICREV_BCM5717 &&
1178 	    sc->bnx_asicrev != BGE_ASICREV_BCM5762 &&
1179 	    !BNX_IS_57765_FAMILY(sc))
1180 		dma_rw_ctl |= BGE_PCIDMARWCTL_TAGGED_STATUS_WA;
1181 	if (bootverbose) {
1182 		if_printf(&sc->arpcom.ac_if, "DMA read/write %#x\n",
1183 		    dma_rw_ctl);
1184 	}
1185 	pci_write_config(sc->bnx_dev, BGE_PCI_DMA_RW_CTL, dma_rw_ctl, 4);
1186 
1187 	/*
1188 	 * Set up general mode register.
1189 	 */
1190 	mode_ctl = bnx_dma_swap_options(sc);
1191 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5720 ||
1192 	    sc->bnx_asicrev == BGE_ASICREV_BCM5762) {
1193 		/* Retain Host-2-BMC settings written by APE firmware. */
1194 		mode_ctl |= CSR_READ_4(sc, BGE_MODE_CTL) &
1195 		    (BGE_MODECTL_BYTESWAP_B2HRX_DATA |
1196 		    BGE_MODECTL_WORDSWAP_B2HRX_DATA |
1197 		    BGE_MODECTL_B2HRX_ENABLE | BGE_MODECTL_HTX2B_ENABLE);
1198 	}
1199 	mode_ctl |= BGE_MODECTL_MAC_ATTN_INTR |
1200 	    BGE_MODECTL_HOST_SEND_BDS | BGE_MODECTL_TX_NO_PHDR_CSUM;
1201 	CSR_WRITE_4(sc, BGE_MODE_CTL, mode_ctl);
1202 
1203 	/*
1204 	 * Disable memory write invalidate.  Apparently it is not supported
1205 	 * properly by these devices.  Also ensure that INTx isn't disabled,
1206 	 * as these chips need it even when using MSI.
1207 	 */
1208 	PCI_CLRBIT(sc->bnx_dev, BGE_PCI_CMD,
1209 	    (PCIM_CMD_MWRICEN | PCIM_CMD_INTxDIS), 4);
1210 
1211 	/* Set the timer prescaler (always 66Mhz) */
1212 	CSR_WRITE_4(sc, BGE_MISC_CFG, 65 << 1/*BGE_32BITTIME_66MHZ*/);
1213 
1214 	return(0);
1215 }
1216 
1217 static int
1218 bnx_blockinit(struct bnx_softc *sc)
1219 {
1220 	struct bnx_intr_data *intr;
1221 	struct bge_rcb *rcb;
1222 	bus_size_t vrcb;
1223 	bge_hostaddr taddr;
1224 	uint32_t val;
1225 	int i, limit;
1226 
1227 	/*
1228 	 * Initialize the memory window pointer register so that
1229 	 * we can access the first 32K of internal NIC RAM. This will
1230 	 * allow us to set up the TX send ring RCBs and the RX return
1231 	 * ring RCBs, plus other things which live in NIC memory.
1232 	 */
1233 	CSR_WRITE_4(sc, BGE_PCI_MEMWIN_BASEADDR, 0);
1234 
1235 	/* Configure mbuf pool watermarks */
1236 	if (BNX_IS_57765_PLUS(sc)) {
1237 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
1238 		if (sc->arpcom.ac_if.if_mtu > ETHERMTU) {
1239 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x7e);
1240 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0xea);
1241 		} else {
1242 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x2a);
1243 			CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0xa0);
1244 		}
1245 	} else {
1246 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_READDMA_LOWAT, 0x0);
1247 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_MACRX_LOWAT, 0x10);
1248 		CSR_WRITE_4(sc, BGE_BMAN_MBUFPOOL_HIWAT, 0x60);
1249 	}
1250 
1251 	/* Configure DMA resource watermarks */
1252 	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_LOWAT, 5);
1253 	CSR_WRITE_4(sc, BGE_BMAN_DMA_DESCPOOL_HIWAT, 10);
1254 
1255 	/* Enable buffer manager */
1256 	val = BGE_BMANMODE_ENABLE | BGE_BMANMODE_LOMBUF_ATTN;
1257 	/*
1258 	 * Change the arbitration algorithm of TXMBUF read request to
1259 	 * round-robin instead of priority based for BCM5719.  When
1260 	 * TXFIFO is almost empty, RDMA will hold its request until
1261 	 * TXFIFO is not almost empty.
1262 	 */
1263 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5719)
1264 		val |= BGE_BMANMODE_NO_TX_UNDERRUN;
1265 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5717 ||
1266 	    sc->bnx_chipid == BGE_CHIPID_BCM5719_A0 ||
1267 	    sc->bnx_chipid == BGE_CHIPID_BCM5720_A0)
1268 		val |= BGE_BMANMODE_LOMBUF_ATTN;
1269 	CSR_WRITE_4(sc, BGE_BMAN_MODE, val);
1270 
1271 	/* Poll for buffer manager start indication */
1272 	for (i = 0; i < BNX_TIMEOUT; i++) {
1273 		if (CSR_READ_4(sc, BGE_BMAN_MODE) & BGE_BMANMODE_ENABLE)
1274 			break;
1275 		DELAY(10);
1276 	}
1277 
1278 	if (i == BNX_TIMEOUT) {
1279 		if_printf(&sc->arpcom.ac_if,
1280 			  "buffer manager failed to start\n");
1281 		return(ENXIO);
1282 	}
1283 
1284 	/* Enable flow-through queues */
1285 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
1286 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
1287 
1288 	/* Wait until queue initialization is complete */
1289 	for (i = 0; i < BNX_TIMEOUT; i++) {
1290 		if (CSR_READ_4(sc, BGE_FTQ_RESET) == 0)
1291 			break;
1292 		DELAY(10);
1293 	}
1294 
1295 	if (i == BNX_TIMEOUT) {
1296 		if_printf(&sc->arpcom.ac_if,
1297 			  "flow-through queue init failed\n");
1298 		return(ENXIO);
1299 	}
1300 
1301 	/*
1302 	 * Summary of rings supported by the controller:
1303 	 *
1304 	 * Standard Receive Producer Ring
1305 	 * - This ring is used to feed receive buffers for "standard"
1306 	 *   sized frames (typically 1536 bytes) to the controller.
1307 	 *
1308 	 * Jumbo Receive Producer Ring
1309 	 * - This ring is used to feed receive buffers for jumbo sized
1310 	 *   frames (i.e. anything bigger than the "standard" frames)
1311 	 *   to the controller.
1312 	 *
1313 	 * Mini Receive Producer Ring
1314 	 * - This ring is used to feed receive buffers for "mini"
1315 	 *   sized frames to the controller.
1316 	 * - This feature required external memory for the controller
1317 	 *   but was never used in a production system.  Should always
1318 	 *   be disabled.
1319 	 *
1320 	 * Receive Return Ring
1321 	 * - After the controller has placed an incoming frame into a
1322 	 *   receive buffer that buffer is moved into a receive return
1323 	 *   ring.  The driver is then responsible to passing the
1324 	 *   buffer up to the stack.  BCM5718/BCM57785 families support
1325 	 *   multiple receive return rings.
1326 	 *
1327 	 * Send Ring
1328 	 * - This ring is used for outgoing frames.  BCM5719/BCM5720
1329 	 *   support multiple send rings.
1330 	 */
1331 
1332 	/* Initialize the standard receive producer ring control block. */
1333 	rcb = &sc->bnx_ldata.bnx_info.bnx_std_rx_rcb;
1334 	rcb->bge_hostaddr.bge_addr_lo =
1335 	    BGE_ADDR_LO(sc->bnx_rx_std_ring.bnx_rx_std_ring_paddr);
1336 	rcb->bge_hostaddr.bge_addr_hi =
1337 	    BGE_ADDR_HI(sc->bnx_rx_std_ring.bnx_rx_std_ring_paddr);
1338 	if (BNX_IS_57765_PLUS(sc)) {
1339 		/*
1340 		 * Bits 31-16: Programmable ring size (2048, 1024, 512, .., 32)
1341 		 * Bits 15-2 : Maximum RX frame size
1342 		 * Bit 1     : 1 = Ring Disabled, 0 = Ring ENabled
1343 		 * Bit 0     : Reserved
1344 		 */
1345 		rcb->bge_maxlen_flags =
1346 		    BGE_RCB_MAXLEN_FLAGS(512, BNX_MAX_FRAMELEN << 2);
1347 	} else {
1348 		/*
1349 		 * Bits 31-16: Programmable ring size (512, 256, 128, 64, 32)
1350 		 * Bits 15-2 : Reserved (should be 0)
1351 		 * Bit 1     : 1 = Ring Disabled, 0 = Ring Enabled
1352 		 * Bit 0     : Reserved
1353 		 */
1354 		rcb->bge_maxlen_flags = BGE_RCB_MAXLEN_FLAGS(512, 0);
1355 	}
1356 	if (BNX_IS_5717_PLUS(sc))
1357 		rcb->bge_nicaddr = BGE_STD_RX_RINGS_5717;
1358 	else
1359 		rcb->bge_nicaddr = BGE_STD_RX_RINGS;
1360 	/* Write the standard receive producer ring control block. */
1361 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_HI, rcb->bge_hostaddr.bge_addr_hi);
1362 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_HADDR_LO, rcb->bge_hostaddr.bge_addr_lo);
1363 	CSR_WRITE_4(sc, BGE_RX_STD_RCB_MAXLEN_FLAGS, rcb->bge_maxlen_flags);
1364 	if (!BNX_IS_5717_PLUS(sc))
1365 		CSR_WRITE_4(sc, BGE_RX_STD_RCB_NICADDR, rcb->bge_nicaddr);
1366 	/* Reset the standard receive producer ring producer index. */
1367 	bnx_writembx(sc, BGE_MBX_RX_STD_PROD_LO, 0);
1368 
1369 	/*
1370 	 * Initialize the jumbo RX producer ring control
1371 	 * block.  We set the 'ring disabled' bit in the
1372 	 * flags field until we're actually ready to start
1373 	 * using this ring (i.e. once we set the MTU
1374 	 * high enough to require it).
1375 	 */
1376 	if (BNX_IS_JUMBO_CAPABLE(sc)) {
1377 		rcb = &sc->bnx_ldata.bnx_info.bnx_jumbo_rx_rcb;
1378 		/* Get the jumbo receive producer ring RCB parameters. */
1379 		rcb->bge_hostaddr.bge_addr_lo =
1380 		    BGE_ADDR_LO(sc->bnx_ldata.bnx_rx_jumbo_ring_paddr);
1381 		rcb->bge_hostaddr.bge_addr_hi =
1382 		    BGE_ADDR_HI(sc->bnx_ldata.bnx_rx_jumbo_ring_paddr);
1383 		rcb->bge_maxlen_flags =
1384 		    BGE_RCB_MAXLEN_FLAGS(BNX_MAX_FRAMELEN,
1385 		    BGE_RCB_FLAG_RING_DISABLED);
1386 		if (BNX_IS_5717_PLUS(sc))
1387 			rcb->bge_nicaddr = BGE_JUMBO_RX_RINGS_5717;
1388 		else
1389 			rcb->bge_nicaddr = BGE_JUMBO_RX_RINGS;
1390 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_HI,
1391 		    rcb->bge_hostaddr.bge_addr_hi);
1392 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_HADDR_LO,
1393 		    rcb->bge_hostaddr.bge_addr_lo);
1394 		/* Program the jumbo receive producer ring RCB parameters. */
1395 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_MAXLEN_FLAGS,
1396 		    rcb->bge_maxlen_flags);
1397 		CSR_WRITE_4(sc, BGE_RX_JUMBO_RCB_NICADDR, rcb->bge_nicaddr);
1398 		/* Reset the jumbo receive producer ring producer index. */
1399 		bnx_writembx(sc, BGE_MBX_RX_JUMBO_PROD_LO, 0);
1400 	}
1401 
1402 	/*
1403 	 * The BD ring replenish thresholds control how often the
1404 	 * hardware fetches new BD's from the producer rings in host
1405 	 * memory.  Setting the value too low on a busy system can
1406 	 * starve the hardware and recue the throughpout.
1407 	 *
1408 	 * Set the BD ring replentish thresholds. The recommended
1409 	 * values are 1/8th the number of descriptors allocated to
1410 	 * each ring.
1411 	 */
1412 	val = 8;
1413 	CSR_WRITE_4(sc, BGE_RBDI_STD_REPL_THRESH, val);
1414 	if (BNX_IS_JUMBO_CAPABLE(sc)) {
1415 		CSR_WRITE_4(sc, BGE_RBDI_JUMBO_REPL_THRESH,
1416 		    BGE_JUMBO_RX_RING_CNT/8);
1417 	}
1418 	if (BNX_IS_57765_PLUS(sc)) {
1419 		CSR_WRITE_4(sc, BGE_STD_REPLENISH_LWM, 32);
1420 		CSR_WRITE_4(sc, BGE_JMB_REPLENISH_LWM, 16);
1421 	}
1422 
1423 	/*
1424 	 * Disable all send rings by setting the 'ring disabled' bit
1425 	 * in the flags field of all the TX send ring control blocks,
1426 	 * located in NIC memory.
1427 	 */
1428 	if (BNX_IS_5717_PLUS(sc))
1429 		limit = 4;
1430 	else if (BNX_IS_57765_FAMILY(sc) ||
1431 	    sc->bnx_asicrev == BGE_ASICREV_BCM5762)
1432 		limit = 2;
1433 	else
1434 		limit = 1;
1435 	vrcb = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
1436 	for (i = 0; i < limit; i++) {
1437 		RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1438 		    BGE_RCB_MAXLEN_FLAGS(0, BGE_RCB_FLAG_RING_DISABLED));
1439 		vrcb += sizeof(struct bge_rcb);
1440 	}
1441 
1442 	/*
1443 	 * Configure send ring RCBs
1444 	 */
1445 	vrcb = BGE_MEMWIN_START + BGE_SEND_RING_RCB;
1446 	for (i = 0; i < sc->bnx_tx_ringcnt; ++i) {
1447 		struct bnx_tx_ring *txr = &sc->bnx_tx_ring[i];
1448 
1449 		BGE_HOSTADDR(taddr, txr->bnx_tx_ring_paddr);
1450 		RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi,
1451 		    taddr.bge_addr_hi);
1452 		RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo,
1453 		    taddr.bge_addr_lo);
1454 		RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1455 		    BGE_RCB_MAXLEN_FLAGS(BGE_TX_RING_CNT, 0));
1456 		vrcb += sizeof(struct bge_rcb);
1457 	}
1458 
1459 	/*
1460 	 * Disable all receive return rings by setting the
1461 	 * 'ring disabled' bit in the flags field of all the receive
1462 	 * return ring control blocks, located in NIC memory.
1463 	 */
1464 	if (BNX_IS_5717_PLUS(sc)) {
1465 		/* Should be 17, use 16 until we get an SRAM map. */
1466 		limit = 16;
1467 	} else if (BNX_IS_57765_FAMILY(sc) ||
1468 	    sc->bnx_asicrev == BGE_ASICREV_BCM5762) {
1469 		limit = 4;
1470 	} else {
1471 		limit = 1;
1472 	}
1473 	/* Disable all receive return rings. */
1474 	vrcb = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
1475 	for (i = 0; i < limit; i++) {
1476 		RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi, 0);
1477 		RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo, 0);
1478 		RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1479 		    BGE_RCB_FLAG_RING_DISABLED);
1480 		bnx_writembx(sc, BGE_MBX_RX_CONS0_LO +
1481 		    (i * (sizeof(uint64_t))), 0);
1482 		vrcb += sizeof(struct bge_rcb);
1483 	}
1484 
1485 	/*
1486 	 * Set up receive return rings.
1487 	 */
1488 	vrcb = BGE_MEMWIN_START + BGE_RX_RETURN_RING_RCB;
1489 	for (i = 0; i < sc->bnx_rx_retcnt; ++i) {
1490 		struct bnx_rx_ret_ring *ret = &sc->bnx_rx_ret_ring[i];
1491 
1492 		BGE_HOSTADDR(taddr, ret->bnx_rx_ret_ring_paddr);
1493 		RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_hi,
1494 		    taddr.bge_addr_hi);
1495 		RCB_WRITE_4(sc, vrcb, bge_hostaddr.bge_addr_lo,
1496 		    taddr.bge_addr_lo);
1497 		RCB_WRITE_4(sc, vrcb, bge_maxlen_flags,
1498 		    BGE_RCB_MAXLEN_FLAGS(BNX_RETURN_RING_CNT, 0));
1499 		vrcb += sizeof(struct bge_rcb);
1500 	}
1501 
1502 	/* Set random backoff seed for TX */
1503 	CSR_WRITE_4(sc, BGE_TX_RANDOM_BACKOFF,
1504 	    (sc->arpcom.ac_enaddr[0] + sc->arpcom.ac_enaddr[1] +
1505 	     sc->arpcom.ac_enaddr[2] + sc->arpcom.ac_enaddr[3] +
1506 	     sc->arpcom.ac_enaddr[4] + sc->arpcom.ac_enaddr[5]) &
1507 	    BGE_TX_BACKOFF_SEED_MASK);
1508 
1509 	/* Set inter-packet gap */
1510 	val = 0x2620;
1511 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5720 ||
1512 	    sc->bnx_asicrev == BGE_ASICREV_BCM5762) {
1513 		val |= CSR_READ_4(sc, BGE_TX_LENGTHS) &
1514 		    (BGE_TXLEN_JMB_FRM_LEN_MSK | BGE_TXLEN_CNT_DN_VAL_MSK);
1515 	}
1516 	CSR_WRITE_4(sc, BGE_TX_LENGTHS, val);
1517 
1518 	/*
1519 	 * Specify which ring to use for packets that don't match
1520 	 * any RX rules.
1521 	 */
1522 	CSR_WRITE_4(sc, BGE_RX_RULES_CFG, 0x08);
1523 
1524 	/*
1525 	 * Configure number of RX lists. One interrupt distribution
1526 	 * list, sixteen active lists, one bad frames class.
1527 	 */
1528 	CSR_WRITE_4(sc, BGE_RXLP_CFG, 0x181);
1529 
1530 	/* Inialize RX list placement stats mask. */
1531 	CSR_WRITE_4(sc, BGE_RXLP_STATS_ENABLE_MASK, 0x007FFFFF);
1532 	CSR_WRITE_4(sc, BGE_RXLP_STATS_CTL, 0x1);
1533 
1534 	/* Disable host coalescing until we get it set up */
1535 	CSR_WRITE_4(sc, BGE_HCC_MODE, 0x00000000);
1536 
1537 	/* Poll to make sure it's shut down. */
1538 	for (i = 0; i < BNX_TIMEOUT; i++) {
1539 		if (!(CSR_READ_4(sc, BGE_HCC_MODE) & BGE_HCCMODE_ENABLE))
1540 			break;
1541 		DELAY(10);
1542 	}
1543 
1544 	if (i == BNX_TIMEOUT) {
1545 		if_printf(&sc->arpcom.ac_if,
1546 			  "host coalescing engine failed to idle\n");
1547 		return(ENXIO);
1548 	}
1549 
1550 	/* Set up host coalescing defaults */
1551 	sc->bnx_coal_chg = BNX_RX_COAL_TICKS_CHG |
1552 	    BNX_TX_COAL_TICKS_CHG |
1553 	    BNX_RX_COAL_BDS_CHG |
1554 	    BNX_TX_COAL_BDS_CHG |
1555 	    BNX_RX_COAL_BDS_INT_CHG |
1556 	    BNX_TX_COAL_BDS_INT_CHG;
1557 	bnx_coal_change(sc);
1558 
1559 	/*
1560 	 * Set up addresses of status blocks
1561 	 */
1562 	intr = &sc->bnx_intr_data[0];
1563 	bzero(intr->bnx_status_block, BGE_STATUS_BLK_SZ);
1564 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_HI,
1565 	    BGE_ADDR_HI(intr->bnx_status_block_paddr));
1566 	CSR_WRITE_4(sc, BGE_HCC_STATUSBLK_ADDR_LO,
1567 	    BGE_ADDR_LO(intr->bnx_status_block_paddr));
1568 	for (i = 1; i < sc->bnx_intr_cnt; ++i) {
1569 		intr = &sc->bnx_intr_data[i];
1570 		bzero(intr->bnx_status_block, BGE_STATUS_BLK_SZ);
1571 		CSR_WRITE_4(sc, BGE_VEC1_STATUSBLK_ADDR_HI + ((i - 1) * 8),
1572 		    BGE_ADDR_HI(intr->bnx_status_block_paddr));
1573 		CSR_WRITE_4(sc, BGE_VEC1_STATUSBLK_ADDR_LO + ((i - 1) * 8),
1574 		    BGE_ADDR_LO(intr->bnx_status_block_paddr));
1575 	}
1576 
1577 	/* Set up status block partail update size. */
1578 	val = BGE_STATBLKSZ_32BYTE;
1579 #if 0
1580 	/*
1581 	 * Does not seem to have visible effect in both
1582 	 * bulk data (1472B UDP datagram) and tiny data
1583 	 * (18B UDP datagram) TX tests.
1584 	 */
1585 	val |= BGE_HCCMODE_CLRTICK_TX;
1586 #endif
1587 	/* Turn on host coalescing state machine */
1588 	CSR_WRITE_4(sc, BGE_HCC_MODE, val | BGE_HCCMODE_ENABLE);
1589 
1590 	/* Turn on RX BD completion state machine and enable attentions */
1591 	CSR_WRITE_4(sc, BGE_RBDC_MODE,
1592 	    BGE_RBDCMODE_ENABLE|BGE_RBDCMODE_ATTN);
1593 
1594 	/* Turn on RX list placement state machine */
1595 	CSR_WRITE_4(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
1596 
1597 	val = BGE_MACMODE_TXDMA_ENB | BGE_MACMODE_RXDMA_ENB |
1598 	    BGE_MACMODE_RX_STATS_CLEAR | BGE_MACMODE_TX_STATS_CLEAR |
1599 	    BGE_MACMODE_RX_STATS_ENB | BGE_MACMODE_TX_STATS_ENB |
1600 	    BGE_MACMODE_FRMHDR_DMA_ENB;
1601 
1602 	if (sc->bnx_flags & BNX_FLAG_TBI)
1603 		val |= BGE_PORTMODE_TBI;
1604 	else if (sc->bnx_flags & BNX_FLAG_MII_SERDES)
1605 		val |= BGE_PORTMODE_GMII;
1606 	else
1607 		val |= BGE_PORTMODE_MII;
1608 
1609 	/* Allow APE to send/receive frames. */
1610 	if (sc->bnx_mfw_flags & BNX_MFW_ON_APE)
1611 		val |= BGE_MACMODE_APE_RX_EN | BGE_MACMODE_APE_TX_EN;
1612 
1613 	/* Turn on DMA, clear stats */
1614 	CSR_WRITE_4(sc, BGE_MAC_MODE, val);
1615 	DELAY(40);
1616 
1617 	/* Set misc. local control, enable interrupts on attentions */
1618 	BNX_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_ONATTN);
1619 
1620 #ifdef notdef
1621 	/* Assert GPIO pins for PHY reset */
1622 	BNX_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_MISCIO_OUT0|
1623 	    BGE_MLC_MISCIO_OUT1|BGE_MLC_MISCIO_OUT2);
1624 	BNX_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_MISCIO_OUTEN0|
1625 	    BGE_MLC_MISCIO_OUTEN1|BGE_MLC_MISCIO_OUTEN2);
1626 #endif
1627 
1628 	if (sc->bnx_intr_type == PCI_INTR_TYPE_MSIX)
1629 		bnx_enable_msi(sc, TRUE);
1630 
1631 	/* Turn on write DMA state machine */
1632 	val = BGE_WDMAMODE_ENABLE|BGE_WDMAMODE_ALL_ATTNS;
1633 	/* Enable host coalescing bug fix. */
1634 	val |= BGE_WDMAMODE_STATUS_TAG_FIX;
1635 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5785) {
1636 		/* Request larger DMA burst size to get better performance. */
1637 		val |= BGE_WDMAMODE_BURST_ALL_DATA;
1638 	}
1639 	CSR_WRITE_4(sc, BGE_WDMA_MODE, val);
1640 	DELAY(40);
1641 
1642 	if (BNX_IS_57765_PLUS(sc)) {
1643 		uint32_t dmactl, dmactl_reg;
1644 
1645 		if (sc->bnx_asicrev == BGE_ASICREV_BCM5762)
1646 			dmactl_reg = BGE_RDMA_RSRVCTRL2;
1647 		else
1648 			dmactl_reg = BGE_RDMA_RSRVCTRL;
1649 
1650 		dmactl = CSR_READ_4(sc, dmactl_reg);
1651 		/*
1652 		 * Adjust tx margin to prevent TX data corruption and
1653 		 * fix internal FIFO overflow.
1654 		 */
1655 		if (sc->bnx_asicrev == BGE_ASICREV_BCM5719 ||
1656 		    sc->bnx_asicrev == BGE_ASICREV_BCM5720 ||
1657 		    sc->bnx_asicrev == BGE_ASICREV_BCM5762) {
1658 			dmactl &= ~(BGE_RDMA_RSRVCTRL_FIFO_LWM_MASK |
1659 			    BGE_RDMA_RSRVCTRL_FIFO_HWM_MASK |
1660 			    BGE_RDMA_RSRVCTRL_TXMRGN_MASK);
1661 			dmactl |= BGE_RDMA_RSRVCTRL_FIFO_LWM_1_5K |
1662 			    BGE_RDMA_RSRVCTRL_FIFO_HWM_1_5K |
1663 			    BGE_RDMA_RSRVCTRL_TXMRGN_320B;
1664 		}
1665 		/*
1666 		 * Enable fix for read DMA FIFO overruns.
1667 		 * The fix is to limit the number of RX BDs
1668 		 * the hardware would fetch at a fime.
1669 		 */
1670 		CSR_WRITE_4(sc, dmactl_reg,
1671 		    dmactl | BGE_RDMA_RSRVCTRL_FIFO_OFLW_FIX);
1672 	}
1673 
1674 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5719) {
1675 		CSR_WRITE_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL,
1676 		    CSR_READ_4(sc, BGE_RDMA_LSO_CRPTEN_CTRL) |
1677 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_BD_4K |
1678 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_LSO_4K);
1679 	} else if (sc->bnx_asicrev == BGE_ASICREV_BCM5720 ||
1680 	    sc->bnx_asicrev == BGE_ASICREV_BCM5762) {
1681 		uint32_t ctrl_reg;
1682 
1683 		if (sc->bnx_asicrev == BGE_ASICREV_BCM5762)
1684 			ctrl_reg = BGE_RDMA_LSO_CRPTEN_CTRL2;
1685 		else
1686 			ctrl_reg = BGE_RDMA_LSO_CRPTEN_CTRL;
1687 
1688 		/*
1689 		 * Allow 4KB burst length reads for non-LSO frames.
1690 		 * Enable 512B burst length reads for buffer descriptors.
1691 		 */
1692 		CSR_WRITE_4(sc, ctrl_reg,
1693 		    CSR_READ_4(sc, ctrl_reg) |
1694 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_BD_512 |
1695 		    BGE_RDMA_LSO_CRPTEN_CTRL_BLEN_LSO_4K);
1696 	}
1697 
1698 	/* Turn on read DMA state machine */
1699 	val = BGE_RDMAMODE_ENABLE | BGE_RDMAMODE_ALL_ATTNS;
1700 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5717)
1701 		val |= BGE_RDMAMODE_MULT_DMA_RD_DIS;
1702         if (sc->bnx_asicrev == BGE_ASICREV_BCM5784 ||
1703             sc->bnx_asicrev == BGE_ASICREV_BCM5785 ||
1704             sc->bnx_asicrev == BGE_ASICREV_BCM57780) {
1705 		val |= BGE_RDMAMODE_BD_SBD_CRPT_ATTN |
1706 		    BGE_RDMAMODE_MBUF_RBD_CRPT_ATTN |
1707 		    BGE_RDMAMODE_MBUF_SBD_CRPT_ATTN;
1708 	}
1709 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5720 ||
1710 	    sc->bnx_asicrev == BGE_ASICREV_BCM5762) {
1711 		val |= CSR_READ_4(sc, BGE_RDMA_MODE) &
1712 		    BGE_RDMAMODE_H2BNC_VLAN_DET;
1713 		/*
1714 		 * Allow multiple outstanding read requests from
1715 		 * non-LSO read DMA engine.
1716 		 */
1717 		val &= ~BGE_RDMAMODE_MULT_DMA_RD_DIS;
1718 	}
1719 	if (sc->bnx_asicrev == BGE_ASICREV_BCM57766)
1720 		val |= BGE_RDMAMODE_JMB_2K_MMRR;
1721 	if (sc->bnx_flags & BNX_FLAG_TSO)
1722 		val |= BGE_RDMAMODE_TSO4_ENABLE;
1723 	val |= BGE_RDMAMODE_FIFO_LONG_BURST;
1724 	CSR_WRITE_4(sc, BGE_RDMA_MODE, val);
1725 	DELAY(40);
1726 
1727 	/* Turn on RX data completion state machine */
1728 	CSR_WRITE_4(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
1729 
1730 	/* Turn on RX BD initiator state machine */
1731 	CSR_WRITE_4(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
1732 
1733 	/* Turn on RX data and RX BD initiator state machine */
1734 	CSR_WRITE_4(sc, BGE_RDBDI_MODE, BGE_RDBDIMODE_ENABLE);
1735 
1736 	/* Turn on send BD completion state machine */
1737 	CSR_WRITE_4(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
1738 
1739 	/* Turn on send data completion state machine */
1740 	val = BGE_SDCMODE_ENABLE;
1741 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5761)
1742 		val |= BGE_SDCMODE_CDELAY;
1743 	CSR_WRITE_4(sc, BGE_SDC_MODE, val);
1744 
1745 	/* Turn on send data initiator state machine */
1746 	if (sc->bnx_flags & BNX_FLAG_TSO) {
1747 		CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE |
1748 		    BGE_SDIMODE_HW_LSO_PRE_DMA);
1749 	} else {
1750 		CSR_WRITE_4(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
1751 	}
1752 
1753 	/* Turn on send BD initiator state machine */
1754 	val = BGE_SBDIMODE_ENABLE;
1755 	if (sc->bnx_tx_ringcnt > 1)
1756 		val |= BGE_SBDIMODE_MULTI_TXR;
1757 	CSR_WRITE_4(sc, BGE_SBDI_MODE, val);
1758 
1759 	/* Turn on send BD selector state machine */
1760 	CSR_WRITE_4(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
1761 
1762 	CSR_WRITE_4(sc, BGE_SDI_STATS_ENABLE_MASK, 0x007FFFFF);
1763 	CSR_WRITE_4(sc, BGE_SDI_STATS_CTL,
1764 	    BGE_SDISTATSCTL_ENABLE|BGE_SDISTATSCTL_FASTER);
1765 
1766 	/* ack/clear link change events */
1767 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED|
1768 	    BGE_MACSTAT_CFG_CHANGED|BGE_MACSTAT_MI_COMPLETE|
1769 	    BGE_MACSTAT_LINK_CHANGED);
1770 	CSR_WRITE_4(sc, BGE_MI_STS, 0);
1771 
1772 	/*
1773 	 * Enable attention when the link has changed state for
1774 	 * devices that use auto polling.
1775 	 */
1776 	if (sc->bnx_flags & BNX_FLAG_TBI) {
1777 		CSR_WRITE_4(sc, BGE_MI_STS, BGE_MISTS_LINK);
1778  	} else {
1779 		if (sc->bnx_mi_mode & BGE_MIMODE_AUTOPOLL) {
1780 			CSR_WRITE_4(sc, BGE_MI_MODE, sc->bnx_mi_mode);
1781 			DELAY(80);
1782 		}
1783 	}
1784 
1785 	/*
1786 	 * Clear any pending link state attention.
1787 	 * Otherwise some link state change events may be lost until attention
1788 	 * is cleared by bnx_intr() -> bnx_softc.bnx_link_upd() sequence.
1789 	 * It's not necessary on newer BCM chips - perhaps enabling link
1790 	 * state change attentions implies clearing pending attention.
1791 	 */
1792 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED|
1793 	    BGE_MACSTAT_CFG_CHANGED|BGE_MACSTAT_MI_COMPLETE|
1794 	    BGE_MACSTAT_LINK_CHANGED);
1795 
1796 	/* Enable link state change attentions. */
1797 	BNX_SETBIT(sc, BGE_MAC_EVT_ENB, BGE_EVTENB_LINK_CHANGED);
1798 
1799 	return(0);
1800 }
1801 
1802 /*
1803  * Probe for a Broadcom chip. Check the PCI vendor and device IDs
1804  * against our list and return its name if we find a match. Note
1805  * that since the Broadcom controller contains VPD support, we
1806  * can get the device name string from the controller itself instead
1807  * of the compiled-in string. This is a little slow, but it guarantees
1808  * we'll always announce the right product name.
1809  */
1810 static int
1811 bnx_probe(device_t dev)
1812 {
1813 	const struct bnx_type *t;
1814 	uint16_t product, vendor;
1815 
1816 	if (!pci_is_pcie(dev))
1817 		return ENXIO;
1818 
1819 	product = pci_get_device(dev);
1820 	vendor = pci_get_vendor(dev);
1821 
1822 	for (t = bnx_devs; t->bnx_name != NULL; t++) {
1823 		if (vendor == t->bnx_vid && product == t->bnx_did)
1824 			break;
1825 	}
1826 	if (t->bnx_name == NULL)
1827 		return ENXIO;
1828 
1829 	device_set_desc(dev, t->bnx_name);
1830 	return 0;
1831 }
1832 
1833 static int
1834 bnx_attach(device_t dev)
1835 {
1836 	struct ifnet *ifp;
1837 	struct bnx_softc *sc;
1838 	struct bnx_rx_std_ring *std;
1839 	struct sysctl_ctx_list *ctx;
1840 	struct sysctl_oid_list *tree;
1841 	uint32_t hwcfg = 0;
1842 	int error = 0, rid, capmask, i, std_cpuid, std_cpuid_def;
1843 	uint8_t ether_addr[ETHER_ADDR_LEN];
1844 	uint16_t product;
1845 	uintptr_t mii_priv = 0;
1846 #if defined(BNX_TSO_DEBUG) || defined(BNX_RSS_DEBUG) || defined(BNX_TSS_DEBUG)
1847 	char desc[32];
1848 #endif
1849 #ifdef IFPOLL_ENABLE
1850 	int offset, offset_def;
1851 #endif
1852 
1853 	sc = device_get_softc(dev);
1854 	sc->bnx_dev = dev;
1855 	callout_init_mp(&sc->bnx_tick_timer);
1856 	lwkt_serialize_init(&sc->bnx_jslot_serializer);
1857 	lwkt_serialize_init(&sc->bnx_main_serialize);
1858 
1859 	/* Always setup interrupt mailboxes */
1860 	for (i = 0; i < BNX_INTR_MAX; ++i) {
1861 		callout_init_mp(&sc->bnx_intr_data[i].bnx_intr_timer);
1862 		sc->bnx_intr_data[i].bnx_sc = sc;
1863 		sc->bnx_intr_data[i].bnx_intr_mbx = BGE_MBX_IRQ0_LO + (i * 8);
1864 		sc->bnx_intr_data[i].bnx_intr_rid = -1;
1865 		sc->bnx_intr_data[i].bnx_intr_cpuid = -1;
1866 	}
1867 
1868 	sc->bnx_func_addr = pci_get_function(dev);
1869 	product = pci_get_device(dev);
1870 
1871 #ifndef BURN_BRIDGES
1872 	if (pci_get_powerstate(dev) != PCI_POWERSTATE_D0) {
1873 		uint32_t irq, mem;
1874 
1875 		irq = pci_read_config(dev, PCIR_INTLINE, 4);
1876 		mem = pci_read_config(dev, BGE_PCI_BAR0, 4);
1877 
1878 		device_printf(dev, "chip is in D%d power mode "
1879 		    "-- setting to D0\n", pci_get_powerstate(dev));
1880 
1881 		pci_set_powerstate(dev, PCI_POWERSTATE_D0);
1882 
1883 		pci_write_config(dev, PCIR_INTLINE, irq, 4);
1884 		pci_write_config(dev, BGE_PCI_BAR0, mem, 4);
1885 	}
1886 #endif	/* !BURN_BRIDGE */
1887 
1888 	/*
1889 	 * Map control/status registers.
1890 	 */
1891 	pci_enable_busmaster(dev);
1892 
1893 	rid = BGE_PCI_BAR0;
1894 	sc->bnx_res = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid,
1895 	    RF_ACTIVE);
1896 
1897 	if (sc->bnx_res == NULL) {
1898 		device_printf(dev, "couldn't map memory\n");
1899 		return ENXIO;
1900 	}
1901 
1902 	sc->bnx_btag = rman_get_bustag(sc->bnx_res);
1903 	sc->bnx_bhandle = rman_get_bushandle(sc->bnx_res);
1904 
1905 	/* Save various chip information */
1906 	sc->bnx_chipid =
1907 	    pci_read_config(dev, BGE_PCI_MISC_CTL, 4) >>
1908 	    BGE_PCIMISCCTL_ASICREV_SHIFT;
1909 	if (BGE_ASICREV(sc->bnx_chipid) == BGE_ASICREV_USE_PRODID_REG) {
1910 		/* All chips having dedicated ASICREV register have CPMU */
1911 		sc->bnx_flags |= BNX_FLAG_CPMU;
1912 
1913 		switch (product) {
1914 		case PCI_PRODUCT_BROADCOM_BCM5717:
1915 		case PCI_PRODUCT_BROADCOM_BCM5717C:
1916 		case PCI_PRODUCT_BROADCOM_BCM5718:
1917 		case PCI_PRODUCT_BROADCOM_BCM5719:
1918 		case PCI_PRODUCT_BROADCOM_BCM5720_ALT:
1919 		case PCI_PRODUCT_BROADCOM_BCM5725:
1920 		case PCI_PRODUCT_BROADCOM_BCM5727:
1921 		case PCI_PRODUCT_BROADCOM_BCM5762:
1922 			sc->bnx_chipid = pci_read_config(dev,
1923 			    BGE_PCI_GEN2_PRODID_ASICREV, 4);
1924 			break;
1925 
1926 		case PCI_PRODUCT_BROADCOM_BCM57761:
1927 		case PCI_PRODUCT_BROADCOM_BCM57762:
1928 		case PCI_PRODUCT_BROADCOM_BCM57765:
1929 		case PCI_PRODUCT_BROADCOM_BCM57766:
1930 		case PCI_PRODUCT_BROADCOM_BCM57781:
1931 		case PCI_PRODUCT_BROADCOM_BCM57782:
1932 		case PCI_PRODUCT_BROADCOM_BCM57785:
1933 		case PCI_PRODUCT_BROADCOM_BCM57786:
1934 		case PCI_PRODUCT_BROADCOM_BCM57791:
1935 		case PCI_PRODUCT_BROADCOM_BCM57795:
1936 			sc->bnx_chipid = pci_read_config(dev,
1937 			    BGE_PCI_GEN15_PRODID_ASICREV, 4);
1938 			break;
1939 
1940 		default:
1941 			sc->bnx_chipid = pci_read_config(dev,
1942 			    BGE_PCI_PRODID_ASICREV, 4);
1943 			break;
1944 		}
1945 	}
1946 	if (sc->bnx_chipid == BGE_CHIPID_BCM5717_C0)
1947 		sc->bnx_chipid = BGE_CHIPID_BCM5720_A0;
1948 
1949 	sc->bnx_asicrev = BGE_ASICREV(sc->bnx_chipid);
1950 	sc->bnx_chiprev = BGE_CHIPREV(sc->bnx_chipid);
1951 
1952 	switch (sc->bnx_asicrev) {
1953 	case BGE_ASICREV_BCM5717:
1954 	case BGE_ASICREV_BCM5719:
1955 	case BGE_ASICREV_BCM5720:
1956 		sc->bnx_flags |= BNX_FLAG_5717_PLUS | BNX_FLAG_57765_PLUS;
1957 		break;
1958 
1959 	case BGE_ASICREV_BCM5762:
1960 		sc->bnx_flags |= BNX_FLAG_57765_PLUS;
1961 		break;
1962 
1963 	case BGE_ASICREV_BCM57765:
1964 	case BGE_ASICREV_BCM57766:
1965 		sc->bnx_flags |= BNX_FLAG_57765_FAMILY | BNX_FLAG_57765_PLUS;
1966 		break;
1967 	}
1968 
1969 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5717 ||
1970 	    sc->bnx_asicrev == BGE_ASICREV_BCM5719 ||
1971 	    sc->bnx_asicrev == BGE_ASICREV_BCM5720 ||
1972 	    sc->bnx_asicrev == BGE_ASICREV_BCM5762)
1973 		sc->bnx_flags |= BNX_FLAG_APE;
1974 
1975 	sc->bnx_flags |= BNX_FLAG_TSO;
1976 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5719 &&
1977 	    sc->bnx_chipid == BGE_CHIPID_BCM5719_A0)
1978 		sc->bnx_flags &= ~BNX_FLAG_TSO;
1979 
1980 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5717 ||
1981 	    BNX_IS_57765_FAMILY(sc)) {
1982 		/*
1983 		 * All BCM57785 and BCM5718 families chips have a bug that
1984 		 * under certain situation interrupt will not be enabled
1985 		 * even if status tag is written to interrupt mailbox.
1986 		 *
1987 		 * While BCM5719 and BCM5720 have a hardware workaround
1988 		 * which could fix the above bug.
1989 		 * See the comment near BGE_PCIDMARWCTL_TAGGED_STATUS_WA in
1990 		 * bnx_chipinit().
1991 		 *
1992 		 * For the rest of the chips in these two families, we will
1993 		 * have to poll the status block at high rate (10ms currently)
1994 		 * to check whether the interrupt is hosed or not.
1995 		 * See bnx_check_intr_*() for details.
1996 		 */
1997 		sc->bnx_flags |= BNX_FLAG_STATUSTAG_BUG;
1998 	}
1999 
2000 	sc->bnx_pciecap = pci_get_pciecap_ptr(sc->bnx_dev);
2001 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5719 ||
2002 	    sc->bnx_asicrev == BGE_ASICREV_BCM5720)
2003 		pcie_set_max_readrq(dev, PCIEM_DEVCTL_MAX_READRQ_2048);
2004 	else
2005 		pcie_set_max_readrq(dev, PCIEM_DEVCTL_MAX_READRQ_4096);
2006 	device_printf(dev, "CHIP ID 0x%08x; "
2007 		      "ASIC REV 0x%02x; CHIP REV 0x%02x\n",
2008 		      sc->bnx_chipid, sc->bnx_asicrev, sc->bnx_chiprev);
2009 
2010 	/*
2011 	 * Set various PHY quirk flags.
2012 	 */
2013 
2014 	capmask = MII_CAPMASK_DEFAULT;
2015 	if (product == PCI_PRODUCT_BROADCOM_BCM57791 ||
2016 	    product == PCI_PRODUCT_BROADCOM_BCM57795) {
2017 		/* 10/100 only */
2018 		capmask &= ~BMSR_EXTSTAT;
2019 	}
2020 
2021 	mii_priv |= BRGPHY_FLAG_WIRESPEED;
2022 	if (sc->bnx_chipid == BGE_CHIPID_BCM5762_A0)
2023 		mii_priv |= BRGPHY_FLAG_5762_A0;
2024 
2025 	/*
2026 	 * Chips with APE need BAR2 access for APE registers/memory.
2027 	 */
2028 	if (sc->bnx_flags & BNX_FLAG_APE) {
2029 		uint32_t pcistate;
2030 
2031 		rid = PCIR_BAR(2);
2032 		sc->bnx_res2 = bus_alloc_resource_any(dev, SYS_RES_MEMORY, &rid,
2033 		    RF_ACTIVE);
2034 		if (sc->bnx_res2 == NULL) {
2035 			device_printf(dev, "couldn't map BAR2 memory\n");
2036 			error = ENXIO;
2037 			goto fail;
2038 		}
2039 
2040 		/* Enable APE register/memory access by host driver. */
2041 		pcistate = pci_read_config(dev, BGE_PCI_PCISTATE, 4);
2042 		pcistate |= BGE_PCISTATE_ALLOW_APE_CTLSPC_WR |
2043 		    BGE_PCISTATE_ALLOW_APE_SHMEM_WR |
2044 		    BGE_PCISTATE_ALLOW_APE_PSPACE_WR;
2045 		pci_write_config(dev, BGE_PCI_PCISTATE, pcistate, 4);
2046 
2047 		bnx_ape_lock_init(sc);
2048 		bnx_ape_read_fw_ver(sc);
2049 	}
2050 
2051 	/* Initialize if_name earlier, so if_printf could be used */
2052 	ifp = &sc->arpcom.ac_if;
2053 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
2054 
2055 	/*
2056 	 * Try to reset the chip.
2057 	 */
2058 	bnx_sig_pre_reset(sc, BNX_RESET_SHUTDOWN);
2059 	bnx_reset(sc);
2060 	bnx_sig_post_reset(sc, BNX_RESET_SHUTDOWN);
2061 
2062 	if (bnx_chipinit(sc)) {
2063 		device_printf(dev, "chip initialization failed\n");
2064 		error = ENXIO;
2065 		goto fail;
2066 	}
2067 
2068 	/*
2069 	 * Get station address
2070 	 */
2071 	error = bnx_get_eaddr(sc, ether_addr);
2072 	if (error) {
2073 		device_printf(dev, "failed to read station address\n");
2074 		goto fail;
2075 	}
2076 
2077 	/* Setup RX/TX and interrupt count */
2078 	bnx_setup_ring_cnt(sc);
2079 
2080 	if ((sc->bnx_rx_retcnt == 1 && sc->bnx_tx_ringcnt == 1) ||
2081 	    (sc->bnx_rx_retcnt > 1 && sc->bnx_tx_ringcnt > 1)) {
2082 	    	/*
2083 		 * The RX ring and the corresponding TX ring processing
2084 		 * should be on the same CPU, since they share the same
2085 		 * status block.
2086 		 */
2087 		sc->bnx_flags |= BNX_FLAG_RXTX_BUNDLE;
2088 		if (bootverbose)
2089 			device_printf(dev, "RX/TX bundle\n");
2090 		if (sc->bnx_tx_ringcnt > 1) {
2091 			/*
2092 			 * Multiple TX rings do not share status block
2093 			 * with link status, so link status will have
2094 			 * to save its own status_tag.
2095 			 */
2096 			sc->bnx_flags |= BNX_FLAG_STATUS_HASTAG;
2097 			if (bootverbose)
2098 				device_printf(dev, "status needs tag\n");
2099 		}
2100 	} else {
2101 		KKASSERT(sc->bnx_rx_retcnt > 1 && sc->bnx_tx_ringcnt == 1);
2102 		if (bootverbose)
2103 			device_printf(dev, "RX/TX not bundled\n");
2104 	}
2105 
2106 	error = bnx_dma_alloc(dev);
2107 	if (error)
2108 		goto fail;
2109 
2110 #ifdef IFPOLL_ENABLE
2111 	if (sc->bnx_flags & BNX_FLAG_RXTX_BUNDLE) {
2112 		/*
2113 		 * NPOLLING RX/TX CPU offset
2114 		 */
2115 		if (sc->bnx_rx_retcnt == ncpus2) {
2116 			offset = 0;
2117 		} else {
2118 			offset_def =
2119 			(sc->bnx_rx_retcnt * device_get_unit(dev)) % ncpus2;
2120 			offset = device_getenv_int(dev, "npoll.offset",
2121 			    offset_def);
2122 			if (offset >= ncpus2 ||
2123 			    offset % sc->bnx_rx_retcnt != 0) {
2124 				device_printf(dev, "invalid npoll.offset %d, "
2125 				    "use %d\n", offset, offset_def);
2126 				offset = offset_def;
2127 			}
2128 		}
2129 		sc->bnx_npoll_rxoff = offset;
2130 		sc->bnx_npoll_txoff = offset;
2131 	} else {
2132 		/*
2133 		 * NPOLLING RX CPU offset
2134 		 */
2135 		if (sc->bnx_rx_retcnt == ncpus2) {
2136 			offset = 0;
2137 		} else {
2138 			offset_def =
2139 			(sc->bnx_rx_retcnt * device_get_unit(dev)) % ncpus2;
2140 			offset = device_getenv_int(dev, "npoll.rxoff",
2141 			    offset_def);
2142 			if (offset >= ncpus2 ||
2143 			    offset % sc->bnx_rx_retcnt != 0) {
2144 				device_printf(dev, "invalid npoll.rxoff %d, "
2145 				    "use %d\n", offset, offset_def);
2146 				offset = offset_def;
2147 			}
2148 		}
2149 		sc->bnx_npoll_rxoff = offset;
2150 
2151 		/*
2152 		 * NPOLLING TX CPU offset
2153 		 */
2154 		offset_def = device_get_unit(dev) % ncpus2;
2155 		offset = device_getenv_int(dev, "npoll.txoff", offset_def);
2156 		if (offset >= ncpus2) {
2157 			device_printf(dev, "invalid npoll.txoff %d, use %d\n",
2158 			    offset, offset_def);
2159 			offset = offset_def;
2160 		}
2161 		sc->bnx_npoll_txoff = offset;
2162 	}
2163 #endif	/* IFPOLL_ENABLE */
2164 
2165 	/*
2166 	 * Allocate interrupt
2167 	 */
2168 	error = bnx_alloc_intr(sc);
2169 	if (error)
2170 		goto fail;
2171 
2172 	/* Setup serializers */
2173 	bnx_setup_serialize(sc);
2174 
2175 	/* Set default tuneable values. */
2176 	sc->bnx_rx_coal_ticks = BNX_RX_COAL_TICKS_DEF;
2177 	sc->bnx_tx_coal_ticks = BNX_TX_COAL_TICKS_DEF;
2178 	sc->bnx_rx_coal_bds = BNX_RX_COAL_BDS_DEF;
2179 	sc->bnx_rx_coal_bds_poll = sc->bnx_rx_ret_ring[0].bnx_rx_cntmax;
2180 	sc->bnx_tx_coal_bds = BNX_TX_COAL_BDS_DEF;
2181 	sc->bnx_tx_coal_bds_poll = BNX_TX_COAL_BDS_POLL_DEF;
2182 	sc->bnx_rx_coal_bds_int = BNX_RX_COAL_BDS_INT_DEF;
2183 	sc->bnx_tx_coal_bds_int = BNX_TX_COAL_BDS_INT_DEF;
2184 
2185 	/* Set up ifnet structure */
2186 	ifp->if_softc = sc;
2187 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
2188 	ifp->if_ioctl = bnx_ioctl;
2189 	ifp->if_start = bnx_start;
2190 #ifdef IFPOLL_ENABLE
2191 	ifp->if_npoll = bnx_npoll;
2192 #endif
2193 	ifp->if_init = bnx_init;
2194 	ifp->if_serialize = bnx_serialize;
2195 	ifp->if_deserialize = bnx_deserialize;
2196 	ifp->if_tryserialize = bnx_tryserialize;
2197 #ifdef INVARIANTS
2198 	ifp->if_serialize_assert = bnx_serialize_assert;
2199 #endif
2200 	ifp->if_mtu = ETHERMTU;
2201 	ifp->if_capabilities = IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU;
2202 
2203 	ifp->if_capabilities |= IFCAP_HWCSUM;
2204 	ifp->if_hwassist = BNX_CSUM_FEATURES;
2205 	if (sc->bnx_flags & BNX_FLAG_TSO) {
2206 		ifp->if_capabilities |= IFCAP_TSO;
2207 		ifp->if_hwassist |= CSUM_TSO;
2208 	}
2209 	if (BNX_RSS_ENABLED(sc))
2210 		ifp->if_capabilities |= IFCAP_RSS;
2211 	ifp->if_capenable = ifp->if_capabilities;
2212 
2213 	ifq_set_maxlen(&ifp->if_snd, BGE_TX_RING_CNT - 1);
2214 	ifq_set_ready(&ifp->if_snd);
2215 	ifq_set_subq_cnt(&ifp->if_snd, sc->bnx_tx_ringcnt);
2216 
2217 	if (sc->bnx_tx_ringcnt > 1) {
2218 		ifp->if_mapsubq = ifq_mapsubq_mask;
2219 		ifq_set_subq_mask(&ifp->if_snd, sc->bnx_tx_ringcnt - 1);
2220 	}
2221 
2222 	/*
2223 	 * Figure out what sort of media we have by checking the
2224 	 * hardware config word in the first 32k of NIC internal memory,
2225 	 * or fall back to examining the EEPROM if necessary.
2226 	 * Note: on some BCM5700 cards, this value appears to be unset.
2227 	 * If that's the case, we have to rely on identifying the NIC
2228 	 * by its PCI subsystem ID, as we do below for the SysKonnect
2229 	 * SK-9D41.
2230 	 */
2231 	if (bnx_readmem_ind(sc, BGE_SRAM_DATA_SIG) == BGE_SRAM_DATA_SIG_MAGIC) {
2232 		hwcfg = bnx_readmem_ind(sc, BGE_SRAM_DATA_CFG);
2233 	} else {
2234 		if (bnx_read_eeprom(sc, (caddr_t)&hwcfg, BGE_EE_HWCFG_OFFSET,
2235 				    sizeof(hwcfg))) {
2236 			device_printf(dev, "failed to read EEPROM\n");
2237 			error = ENXIO;
2238 			goto fail;
2239 		}
2240 		hwcfg = ntohl(hwcfg);
2241 	}
2242 
2243 	/* The SysKonnect SK-9D41 is a 1000baseSX card. */
2244 	if (pci_get_subvendor(dev) == PCI_PRODUCT_SCHNEIDERKOCH_SK_9D41 ||
2245 	    (hwcfg & BGE_HWCFG_MEDIA) == BGE_MEDIA_FIBER)
2246 		sc->bnx_flags |= BNX_FLAG_TBI;
2247 
2248 	/* Setup MI MODE */
2249 	if (sc->bnx_flags & BNX_FLAG_CPMU)
2250 		sc->bnx_mi_mode = BGE_MIMODE_500KHZ_CONST;
2251 	else
2252 		sc->bnx_mi_mode = BGE_MIMODE_BASE;
2253 
2254 	/* Setup link status update stuffs */
2255 	if (sc->bnx_flags & BNX_FLAG_TBI) {
2256 		sc->bnx_link_upd = bnx_tbi_link_upd;
2257 		sc->bnx_link_chg = BGE_MACSTAT_LINK_CHANGED;
2258 	} else if (sc->bnx_mi_mode & BGE_MIMODE_AUTOPOLL) {
2259 		sc->bnx_link_upd = bnx_autopoll_link_upd;
2260 		sc->bnx_link_chg = BGE_MACSTAT_LINK_CHANGED;
2261 	} else {
2262 		sc->bnx_link_upd = bnx_copper_link_upd;
2263 		sc->bnx_link_chg = BGE_MACSTAT_LINK_CHANGED;
2264 	}
2265 
2266 	/* Set default PHY address */
2267 	sc->bnx_phyno = 1;
2268 
2269 	/*
2270 	 * PHY address mapping for various devices.
2271 	 *
2272 	 *          | F0 Cu | F0 Sr | F1 Cu | F1 Sr |
2273 	 * ---------+-------+-------+-------+-------+
2274 	 * BCM57XX  |   1   |   X   |   X   |   X   |
2275 	 * BCM5717  |   1   |   8   |   2   |   9   |
2276 	 * BCM5719  |   1   |   8   |   2   |   9   |
2277 	 * BCM5720  |   1   |   8   |   2   |   9   |
2278 	 *
2279 	 *          | F2 Cu | F2 Sr | F3 Cu | F3 Sr |
2280 	 * ---------+-------+-------+-------+-------+
2281 	 * BCM57XX  |   X   |   X   |   X   |   X   |
2282 	 * BCM5717  |   X   |   X   |   X   |   X   |
2283 	 * BCM5719  |   3   |   10  |   4   |   11  |
2284 	 * BCM5720  |   X   |   X   |   X   |   X   |
2285 	 *
2286 	 * Other addresses may respond but they are not
2287 	 * IEEE compliant PHYs and should be ignored.
2288 	 */
2289 	if (BNX_IS_5717_PLUS(sc)) {
2290 		if (sc->bnx_chipid == BGE_CHIPID_BCM5717_A0) {
2291 			if (CSR_READ_4(sc, BGE_SGDIG_STS) &
2292 			    BGE_SGDIGSTS_IS_SERDES)
2293 				sc->bnx_phyno = sc->bnx_func_addr + 8;
2294 			else
2295 				sc->bnx_phyno = sc->bnx_func_addr + 1;
2296 		} else {
2297 			if (CSR_READ_4(sc, BGE_CPMU_PHY_STRAP) &
2298 			    BGE_CPMU_PHY_STRAP_IS_SERDES)
2299 				sc->bnx_phyno = sc->bnx_func_addr + 8;
2300 			else
2301 				sc->bnx_phyno = sc->bnx_func_addr + 1;
2302 		}
2303 	}
2304 
2305 	if (sc->bnx_flags & BNX_FLAG_TBI) {
2306 		ifmedia_init(&sc->bnx_ifmedia, IFM_IMASK,
2307 		    bnx_ifmedia_upd, bnx_ifmedia_sts);
2308 		ifmedia_add(&sc->bnx_ifmedia, IFM_ETHER|IFM_1000_SX, 0, NULL);
2309 		ifmedia_add(&sc->bnx_ifmedia,
2310 		    IFM_ETHER|IFM_1000_SX|IFM_FDX, 0, NULL);
2311 		ifmedia_add(&sc->bnx_ifmedia, IFM_ETHER|IFM_AUTO, 0, NULL);
2312 		ifmedia_set(&sc->bnx_ifmedia, IFM_ETHER|IFM_AUTO);
2313 		sc->bnx_ifmedia.ifm_media = sc->bnx_ifmedia.ifm_cur->ifm_media;
2314 	} else {
2315 		struct mii_probe_args mii_args;
2316 
2317 		mii_probe_args_init(&mii_args, bnx_ifmedia_upd, bnx_ifmedia_sts);
2318 		mii_args.mii_probemask = 1 << sc->bnx_phyno;
2319 		mii_args.mii_capmask = capmask;
2320 		mii_args.mii_privtag = MII_PRIVTAG_BRGPHY;
2321 		mii_args.mii_priv = mii_priv;
2322 
2323 		error = mii_probe(dev, &sc->bnx_miibus, &mii_args);
2324 		if (error) {
2325 			device_printf(dev, "MII without any PHY!\n");
2326 			goto fail;
2327 		}
2328 	}
2329 
2330 	ctx = device_get_sysctl_ctx(sc->bnx_dev);
2331 	tree = SYSCTL_CHILDREN(device_get_sysctl_tree(sc->bnx_dev));
2332 
2333 	SYSCTL_ADD_INT(ctx, tree, OID_AUTO,
2334 	    "rx_rings", CTLFLAG_RD, &sc->bnx_rx_retcnt, 0, "# of RX rings");
2335 	SYSCTL_ADD_INT(ctx, tree, OID_AUTO,
2336 	    "tx_rings", CTLFLAG_RD, &sc->bnx_tx_ringcnt, 0, "# of TX rings");
2337 
2338 	SYSCTL_ADD_PROC(ctx, tree, OID_AUTO, "rx_coal_ticks",
2339 			CTLTYPE_INT | CTLFLAG_RW,
2340 			sc, 0, bnx_sysctl_rx_coal_ticks, "I",
2341 			"Receive coalescing ticks (usec).");
2342 	SYSCTL_ADD_PROC(ctx, tree, OID_AUTO, "tx_coal_ticks",
2343 			CTLTYPE_INT | CTLFLAG_RW,
2344 			sc, 0, bnx_sysctl_tx_coal_ticks, "I",
2345 			"Transmit coalescing ticks (usec).");
2346 	SYSCTL_ADD_PROC(ctx, tree, OID_AUTO, "rx_coal_bds",
2347 			CTLTYPE_INT | CTLFLAG_RW,
2348 			sc, 0, bnx_sysctl_rx_coal_bds, "I",
2349 			"Receive max coalesced BD count.");
2350 	SYSCTL_ADD_PROC(ctx, tree, OID_AUTO, "rx_coal_bds_poll",
2351 			CTLTYPE_INT | CTLFLAG_RW,
2352 			sc, 0, bnx_sysctl_rx_coal_bds_poll, "I",
2353 			"Receive max coalesced BD count in polling.");
2354 	SYSCTL_ADD_PROC(ctx, tree, OID_AUTO, "tx_coal_bds",
2355 			CTLTYPE_INT | CTLFLAG_RW,
2356 			sc, 0, bnx_sysctl_tx_coal_bds, "I",
2357 			"Transmit max coalesced BD count.");
2358 	SYSCTL_ADD_PROC(ctx, tree, OID_AUTO, "tx_coal_bds_poll",
2359 			CTLTYPE_INT | CTLFLAG_RW,
2360 			sc, 0, bnx_sysctl_tx_coal_bds_poll, "I",
2361 			"Transmit max coalesced BD count in polling.");
2362 	/*
2363 	 * A common design characteristic for many Broadcom
2364 	 * client controllers is that they only support a
2365 	 * single outstanding DMA read operation on the PCIe
2366 	 * bus. This means that it will take twice as long to
2367 	 * fetch a TX frame that is split into header and
2368 	 * payload buffers as it does to fetch a single,
2369 	 * contiguous TX frame (2 reads vs. 1 read). For these
2370 	 * controllers, coalescing buffers to reduce the number
2371 	 * of memory reads is effective way to get maximum
2372 	 * performance(about 940Mbps).  Without collapsing TX
2373 	 * buffers the maximum TCP bulk transfer performance
2374 	 * is about 850Mbps. However forcing coalescing mbufs
2375 	 * consumes a lot of CPU cycles, so leave it off by
2376 	 * default.
2377 	 */
2378 	SYSCTL_ADD_PROC(ctx, tree, OID_AUTO,
2379 	    "force_defrag", CTLTYPE_INT | CTLFLAG_RW,
2380 	    sc, 0, bnx_sysctl_force_defrag, "I",
2381 	    "Force defragment on TX path");
2382 
2383 	SYSCTL_ADD_PROC(ctx, tree, OID_AUTO,
2384 	    "tx_wreg", CTLTYPE_INT | CTLFLAG_RW,
2385 	    sc, 0, bnx_sysctl_tx_wreg, "I",
2386 	    "# of segments before writing to hardware register");
2387 
2388 	SYSCTL_ADD_PROC(ctx, tree, OID_AUTO,
2389 	    "std_refill", CTLTYPE_INT | CTLFLAG_RW,
2390 	    sc, 0, bnx_sysctl_std_refill, "I",
2391 	    "# of packets received before scheduling standard refilling");
2392 
2393 	SYSCTL_ADD_PROC(ctx, tree, OID_AUTO,
2394 	    "rx_coal_bds_int", CTLTYPE_INT | CTLFLAG_RW,
2395 	    sc, 0, bnx_sysctl_rx_coal_bds_int, "I",
2396 	    "Receive max coalesced BD count during interrupt.");
2397 	SYSCTL_ADD_PROC(ctx, tree, OID_AUTO,
2398 	    "tx_coal_bds_int", CTLTYPE_INT | CTLFLAG_RW,
2399 	    sc, 0, bnx_sysctl_tx_coal_bds_int, "I",
2400 	    "Transmit max coalesced BD count during interrupt.");
2401 
2402 #ifdef IFPOLL_ENABLE
2403 	if (sc->bnx_flags & BNX_FLAG_RXTX_BUNDLE) {
2404 		SYSCTL_ADD_PROC(ctx, tree, OID_AUTO,
2405 		    "npoll_offset", CTLTYPE_INT | CTLFLAG_RW,
2406 		    sc, 0, bnx_sysctl_npoll_offset, "I",
2407 		    "NPOLLING cpu offset");
2408 	} else {
2409 		SYSCTL_ADD_PROC(ctx, tree, OID_AUTO,
2410 		    "npoll_rxoff", CTLTYPE_INT | CTLFLAG_RW,
2411 		    sc, 0, bnx_sysctl_npoll_rxoff, "I",
2412 		    "NPOLLING RX cpu offset");
2413 		SYSCTL_ADD_PROC(ctx, tree, OID_AUTO,
2414 		    "npoll_txoff", CTLTYPE_INT | CTLFLAG_RW,
2415 		    sc, 0, bnx_sysctl_npoll_txoff, "I",
2416 		    "NPOLLING TX cpu offset");
2417 	}
2418 #endif
2419 
2420 #ifdef BNX_RSS_DEBUG
2421 	SYSCTL_ADD_INT(ctx, tree, OID_AUTO,
2422 	    "std_refill_mask", CTLFLAG_RD,
2423 	    &sc->bnx_rx_std_ring.bnx_rx_std_refill, 0, "");
2424 	SYSCTL_ADD_INT(ctx, tree, OID_AUTO,
2425 	    "std_used", CTLFLAG_RD,
2426 	    &sc->bnx_rx_std_ring.bnx_rx_std_used, 0, "");
2427 	SYSCTL_ADD_INT(ctx, tree, OID_AUTO,
2428 	    "rss_debug", CTLFLAG_RW, &sc->bnx_rss_debug, 0, "");
2429 	for (i = 0; i < sc->bnx_rx_retcnt; ++i) {
2430 		ksnprintf(desc, sizeof(desc), "rx_pkt%d", i);
2431 		SYSCTL_ADD_ULONG(ctx, tree, OID_AUTO,
2432 		    desc, CTLFLAG_RW, &sc->bnx_rx_ret_ring[i].bnx_rx_pkt, "");
2433 
2434 		ksnprintf(desc, sizeof(desc), "rx_force_sched%d", i);
2435 		SYSCTL_ADD_ULONG(ctx, tree, OID_AUTO,
2436 		    desc, CTLFLAG_RW,
2437 		    &sc->bnx_rx_ret_ring[i].bnx_rx_force_sched, "");
2438 	}
2439 #endif
2440 #ifdef BNX_TSS_DEBUG
2441 	for (i = 0; i < sc->bnx_tx_ringcnt; ++i) {
2442 		ksnprintf(desc, sizeof(desc), "tx_pkt%d", i);
2443 		SYSCTL_ADD_ULONG(ctx, tree, OID_AUTO,
2444 		    desc, CTLFLAG_RW, &sc->bnx_tx_ring[i].bnx_tx_pkt, "");
2445 	}
2446 #endif
2447 
2448 	SYSCTL_ADD_ULONG(ctx, tree, OID_AUTO,
2449 	    "norxbds", CTLFLAG_RW, &sc->bnx_norxbds, "");
2450 
2451 	SYSCTL_ADD_ULONG(ctx, tree, OID_AUTO,
2452 	    "errors", CTLFLAG_RW, &sc->bnx_errors, "");
2453 
2454 #ifdef BNX_TSO_DEBUG
2455 	for (i = 0; i < BNX_TSO_NSTATS; ++i) {
2456 		ksnprintf(desc, sizeof(desc), "tso%d", i + 1);
2457 		SYSCTL_ADD_ULONG(ctx, tree, OID_AUTO,
2458 		    desc, CTLFLAG_RW, &sc->bnx_tsosegs[i], "");
2459 	}
2460 #endif
2461 
2462 	/*
2463 	 * Call MI attach routine.
2464 	 */
2465 	ether_ifattach(ifp, ether_addr, NULL);
2466 
2467 	/* Setup TX rings and subqueues */
2468 	for (i = 0; i < sc->bnx_tx_ringcnt; ++i) {
2469 		struct ifaltq_subque *ifsq = ifq_get_subq(&ifp->if_snd, i);
2470 		struct bnx_tx_ring *txr = &sc->bnx_tx_ring[i];
2471 
2472 		ifsq_set_cpuid(ifsq, txr->bnx_tx_cpuid);
2473 		ifsq_set_hw_serialize(ifsq, &txr->bnx_tx_serialize);
2474 		ifsq_set_priv(ifsq, txr);
2475 		txr->bnx_ifsq = ifsq;
2476 
2477 		ifsq_watchdog_init(&txr->bnx_tx_watchdog, ifsq, bnx_watchdog);
2478 
2479 		if (bootverbose) {
2480 			device_printf(dev, "txr %d -> cpu%d\n", i,
2481 			    txr->bnx_tx_cpuid);
2482 		}
2483 	}
2484 
2485 	error = bnx_setup_intr(sc);
2486 	if (error) {
2487 		ether_ifdetach(ifp);
2488 		goto fail;
2489 	}
2490 	bnx_set_tick_cpuid(sc, FALSE);
2491 
2492 	/*
2493 	 * Create RX standard ring refilling thread
2494 	 */
2495 	std_cpuid_def = device_get_unit(dev) % ncpus;
2496 	std_cpuid = device_getenv_int(dev, "std.cpuid", std_cpuid_def);
2497 	if (std_cpuid < 0 || std_cpuid >= ncpus) {
2498 		device_printf(dev, "invalid std.cpuid %d, use %d\n",
2499 		    std_cpuid, std_cpuid_def);
2500 		std_cpuid = std_cpuid_def;
2501 	}
2502 
2503 	std = &sc->bnx_rx_std_ring;
2504 	lwkt_create(bnx_rx_std_refill_ithread, std, NULL,
2505 	    &std->bnx_rx_std_ithread, TDF_NOSTART | TDF_INTTHREAD, std_cpuid,
2506 	    "%s std", device_get_nameunit(dev));
2507 	lwkt_setpri(&std->bnx_rx_std_ithread, TDPRI_INT_MED);
2508 	std->bnx_rx_std_ithread.td_preemptable = lwkt_preempt;
2509 	sc->bnx_flags |= BNX_FLAG_STD_THREAD;
2510 
2511 	return(0);
2512 fail:
2513 	bnx_detach(dev);
2514 	return(error);
2515 }
2516 
2517 static int
2518 bnx_detach(device_t dev)
2519 {
2520 	struct bnx_softc *sc = device_get_softc(dev);
2521 
2522 	if (device_is_attached(dev)) {
2523 		struct ifnet *ifp = &sc->arpcom.ac_if;
2524 
2525 		ifnet_serialize_all(ifp);
2526 		bnx_stop(sc);
2527 		bnx_teardown_intr(sc, sc->bnx_intr_cnt);
2528 		ifnet_deserialize_all(ifp);
2529 
2530 		ether_ifdetach(ifp);
2531 	}
2532 
2533 	if (sc->bnx_flags & BNX_FLAG_STD_THREAD) {
2534 		struct bnx_rx_std_ring *std = &sc->bnx_rx_std_ring;
2535 
2536 		tsleep_interlock(std, 0);
2537 
2538 		if (std->bnx_rx_std_ithread.td_gd == mycpu) {
2539 			bnx_rx_std_refill_stop(std);
2540 		} else {
2541 			lwkt_send_ipiq(std->bnx_rx_std_ithread.td_gd,
2542 			    bnx_rx_std_refill_stop, std);
2543 		}
2544 
2545 		tsleep(std, PINTERLOCKED, "bnx_detach", 0);
2546 		if (bootverbose)
2547 			device_printf(dev, "RX std ithread exited\n");
2548 
2549 		lwkt_synchronize_ipiqs("bnx_detach_ipiq");
2550 	}
2551 
2552 	if (sc->bnx_flags & BNX_FLAG_TBI)
2553 		ifmedia_removeall(&sc->bnx_ifmedia);
2554 	if (sc->bnx_miibus)
2555 		device_delete_child(dev, sc->bnx_miibus);
2556 	bus_generic_detach(dev);
2557 
2558 	bnx_free_intr(sc);
2559 
2560 	if (sc->bnx_msix_mem_res != NULL) {
2561 		bus_release_resource(dev, SYS_RES_MEMORY, sc->bnx_msix_mem_rid,
2562 		    sc->bnx_msix_mem_res);
2563 	}
2564 	if (sc->bnx_res != NULL) {
2565 		bus_release_resource(dev, SYS_RES_MEMORY,
2566 		    BGE_PCI_BAR0, sc->bnx_res);
2567 	}
2568 	if (sc->bnx_res2 != NULL) {
2569 		bus_release_resource(dev, SYS_RES_MEMORY,
2570 		    PCIR_BAR(2), sc->bnx_res2);
2571 	}
2572 
2573 	bnx_dma_free(sc);
2574 
2575 	if (sc->bnx_serialize != NULL)
2576 		kfree(sc->bnx_serialize, M_DEVBUF);
2577 
2578 	return 0;
2579 }
2580 
2581 static void
2582 bnx_reset(struct bnx_softc *sc)
2583 {
2584 	device_t dev = sc->bnx_dev;
2585 	uint32_t cachesize, command, reset, mac_mode, mac_mode_mask;
2586 	void (*write_op)(struct bnx_softc *, uint32_t, uint32_t);
2587 	int i, val = 0;
2588 	uint16_t devctl;
2589 
2590 	mac_mode_mask = BGE_MACMODE_HALF_DUPLEX | BGE_MACMODE_PORTMODE;
2591 	if (sc->bnx_mfw_flags & BNX_MFW_ON_APE)
2592 		mac_mode_mask |= BGE_MACMODE_APE_RX_EN | BGE_MACMODE_APE_TX_EN;
2593 	mac_mode = CSR_READ_4(sc, BGE_MAC_MODE) & mac_mode_mask;
2594 
2595 	write_op = bnx_writemem_direct;
2596 
2597 	CSR_WRITE_4(sc, BGE_NVRAM_SWARB, BGE_NVRAMSWARB_SET1);
2598 	for (i = 0; i < 8000; i++) {
2599 		if (CSR_READ_4(sc, BGE_NVRAM_SWARB) & BGE_NVRAMSWARB_GNT1)
2600 			break;
2601 		DELAY(20);
2602 	}
2603 	if (i == 8000)
2604 		if_printf(&sc->arpcom.ac_if, "NVRAM lock timedout!\n");
2605 
2606 	/* Take APE lock when performing reset. */
2607 	bnx_ape_lock(sc, BGE_APE_LOCK_GRC);
2608 
2609 	/* Save some important PCI state. */
2610 	cachesize = pci_read_config(dev, BGE_PCI_CACHESZ, 4);
2611 	command = pci_read_config(dev, BGE_PCI_CMD, 4);
2612 
2613 	pci_write_config(dev, BGE_PCI_MISC_CTL,
2614 	    BGE_PCIMISCCTL_INDIRECT_ACCESS|BGE_PCIMISCCTL_MASK_PCI_INTR|
2615 	    BGE_HIF_SWAP_OPTIONS|BGE_PCIMISCCTL_PCISTATE_RW|
2616 	    BGE_PCIMISCCTL_TAGGED_STATUS, 4);
2617 
2618 	/* Disable fastboot on controllers that support it. */
2619 	if (bootverbose)
2620 		if_printf(&sc->arpcom.ac_if, "Disabling fastboot\n");
2621 	CSR_WRITE_4(sc, BGE_FASTBOOT_PC, 0x0);
2622 
2623 	/*
2624 	 * Write the magic number to SRAM at offset 0xB50.
2625 	 * When firmware finishes its initialization it will
2626 	 * write ~BGE_SRAM_FW_MB_MAGIC to the same location.
2627 	 */
2628 	bnx_writemem_ind(sc, BGE_SRAM_FW_MB, BGE_SRAM_FW_MB_MAGIC);
2629 
2630 	reset = BGE_MISCCFG_RESET_CORE_CLOCKS|(65<<1);
2631 
2632 	/* XXX: Broadcom Linux driver. */
2633 	/* Force PCI-E 1.0a mode */
2634 	if (!BNX_IS_57765_PLUS(sc) &&
2635 	    CSR_READ_4(sc, BGE_PCIE_PHY_TSTCTL) ==
2636 	    (BGE_PCIE_PHY_TSTCTL_PSCRAM |
2637 	     BGE_PCIE_PHY_TSTCTL_PCIE10)) {
2638 		CSR_WRITE_4(sc, BGE_PCIE_PHY_TSTCTL,
2639 		    BGE_PCIE_PHY_TSTCTL_PSCRAM);
2640 	}
2641 	if (sc->bnx_chipid != BGE_CHIPID_BCM5750_A0) {
2642 		/* Prevent PCIE link training during global reset */
2643 		CSR_WRITE_4(sc, BGE_MISC_CFG, (1<<29));
2644 		reset |= (1<<29);
2645 	}
2646 
2647 	/*
2648 	 * Set GPHY Power Down Override to leave GPHY
2649 	 * powered up in D0 uninitialized.
2650 	 */
2651 	if ((sc->bnx_flags & BNX_FLAG_CPMU) == 0)
2652 		reset |= BGE_MISCCFG_GPHY_PD_OVERRIDE;
2653 
2654 	/* Issue global reset */
2655 	write_op(sc, BGE_MISC_CFG, reset);
2656 
2657 	DELAY(100 * 1000);
2658 
2659 	/* XXX: Broadcom Linux driver. */
2660 	if (sc->bnx_chipid == BGE_CHIPID_BCM5750_A0) {
2661 		uint32_t v;
2662 
2663 		DELAY(500000); /* wait for link training to complete */
2664 		v = pci_read_config(dev, 0xc4, 4);
2665 		pci_write_config(dev, 0xc4, v | (1<<15), 4);
2666 	}
2667 
2668 	devctl = pci_read_config(dev, sc->bnx_pciecap + PCIER_DEVCTRL, 2);
2669 
2670 	/* Disable no snoop and disable relaxed ordering. */
2671 	devctl &= ~(PCIEM_DEVCTL_RELAX_ORDER | PCIEM_DEVCTL_NOSNOOP);
2672 
2673 	/* Old PCI-E chips only support 128 bytes Max PayLoad Size. */
2674 	if ((sc->bnx_flags & BNX_FLAG_CPMU) == 0) {
2675 		devctl &= ~PCIEM_DEVCTL_MAX_PAYLOAD_MASK;
2676 		devctl |= PCIEM_DEVCTL_MAX_PAYLOAD_128;
2677 	}
2678 
2679 	pci_write_config(dev, sc->bnx_pciecap + PCIER_DEVCTRL,
2680 	    devctl, 2);
2681 
2682 	/* Clear error status. */
2683 	pci_write_config(dev, sc->bnx_pciecap + PCIER_DEVSTS,
2684 	    PCIEM_DEVSTS_CORR_ERR |
2685 	    PCIEM_DEVSTS_NFATAL_ERR |
2686 	    PCIEM_DEVSTS_FATAL_ERR |
2687 	    PCIEM_DEVSTS_UNSUPP_REQ, 2);
2688 
2689 	/* Reset some of the PCI state that got zapped by reset */
2690 	pci_write_config(dev, BGE_PCI_MISC_CTL,
2691 	    BGE_PCIMISCCTL_INDIRECT_ACCESS|BGE_PCIMISCCTL_MASK_PCI_INTR|
2692 	    BGE_HIF_SWAP_OPTIONS|BGE_PCIMISCCTL_PCISTATE_RW|
2693 	    BGE_PCIMISCCTL_TAGGED_STATUS, 4);
2694 	val = BGE_PCISTATE_ROM_ENABLE | BGE_PCISTATE_ROM_RETRY_ENABLE;
2695 	if (sc->bnx_mfw_flags & BNX_MFW_ON_APE) {
2696 		val |= BGE_PCISTATE_ALLOW_APE_CTLSPC_WR |
2697 		    BGE_PCISTATE_ALLOW_APE_SHMEM_WR |
2698 		    BGE_PCISTATE_ALLOW_APE_PSPACE_WR;
2699 	}
2700 	pci_write_config(dev, BGE_PCI_PCISTATE, val, 4);
2701 	pci_write_config(dev, BGE_PCI_CACHESZ, cachesize, 4);
2702 	pci_write_config(dev, BGE_PCI_CMD, command, 4);
2703 
2704 	/* Enable memory arbiter */
2705 	CSR_WRITE_4(sc, BGE_MARB_MODE, BGE_MARBMODE_ENABLE);
2706 
2707 	/* Fix up byte swapping */
2708 	CSR_WRITE_4(sc, BGE_MODE_CTL, bnx_dma_swap_options(sc));
2709 
2710 	val = CSR_READ_4(sc, BGE_MAC_MODE);
2711 	val = (val & ~mac_mode_mask) | mac_mode;
2712 	CSR_WRITE_4(sc, BGE_MAC_MODE, val);
2713 	DELAY(40);
2714 
2715 	bnx_ape_unlock(sc, BGE_APE_LOCK_GRC);
2716 
2717 	/*
2718 	 * Poll until we see the 1's complement of the magic number.
2719 	 * This indicates that the firmware initialization is complete.
2720 	 */
2721 	for (i = 0; i < BNX_FIRMWARE_TIMEOUT; i++) {
2722 		val = bnx_readmem_ind(sc, BGE_SRAM_FW_MB);
2723 		if (val == ~BGE_SRAM_FW_MB_MAGIC)
2724 			break;
2725 		DELAY(10);
2726 	}
2727 	if (i == BNX_FIRMWARE_TIMEOUT) {
2728 		if_printf(&sc->arpcom.ac_if, "firmware handshake "
2729 			  "timed out, found 0x%08x\n", val);
2730 	}
2731 
2732 	/* BCM57765 A0 needs additional time before accessing. */
2733 	if (sc->bnx_chipid == BGE_CHIPID_BCM57765_A0)
2734 		DELAY(10 * 1000);
2735 
2736 	/*
2737 	 * The 5704 in TBI mode apparently needs some special
2738 	 * adjustment to insure the SERDES drive level is set
2739 	 * to 1.2V.
2740 	 */
2741 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5704 &&
2742 	    (sc->bnx_flags & BNX_FLAG_TBI)) {
2743 		uint32_t serdescfg;
2744 
2745 		serdescfg = CSR_READ_4(sc, BGE_SERDES_CFG);
2746 		serdescfg = (serdescfg & ~0xFFF) | 0x880;
2747 		CSR_WRITE_4(sc, BGE_SERDES_CFG, serdescfg);
2748 	}
2749 
2750 	CSR_WRITE_4(sc, BGE_MI_MODE,
2751 	    sc->bnx_mi_mode & ~BGE_MIMODE_AUTOPOLL);
2752 	DELAY(80);
2753 
2754 	/* XXX: Broadcom Linux driver. */
2755 	if (!BNX_IS_57765_PLUS(sc)) {
2756 		uint32_t v;
2757 
2758 		/* Enable Data FIFO protection. */
2759 		v = CSR_READ_4(sc, BGE_PCIE_TLDLPL_PORT);
2760 		CSR_WRITE_4(sc, BGE_PCIE_TLDLPL_PORT, v | (1 << 25));
2761 	}
2762 
2763 	DELAY(10000);
2764 
2765 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5720) {
2766 		BNX_CLRBIT(sc, BGE_CPMU_CLCK_ORIDE,
2767 		    CPMU_CLCK_ORIDE_MAC_ORIDE_EN);
2768 	}
2769 }
2770 
2771 /*
2772  * Frame reception handling. This is called if there's a frame
2773  * on the receive return list.
2774  *
2775  * Note: we have to be able to handle two possibilities here:
2776  * 1) the frame is from the jumbo recieve ring
2777  * 2) the frame is from the standard receive ring
2778  */
2779 
2780 static void
2781 bnx_rxeof(struct bnx_rx_ret_ring *ret, uint16_t rx_prod, int count)
2782 {
2783 	struct bnx_softc *sc = ret->bnx_sc;
2784 	struct bnx_rx_std_ring *std = ret->bnx_std;
2785 	struct ifnet *ifp = &sc->arpcom.ac_if;
2786 	int std_used = 0, cpuid = mycpuid;
2787 
2788 	while (ret->bnx_rx_saved_considx != rx_prod && count != 0) {
2789 		struct pktinfo pi0, *pi = NULL;
2790 		struct bge_rx_bd *cur_rx;
2791 		struct bnx_rx_buf *rb;
2792 		uint32_t rxidx;
2793 		struct mbuf *m = NULL;
2794 		uint16_t vlan_tag = 0;
2795 		int have_tag = 0;
2796 
2797 		--count;
2798 
2799 		cur_rx = &ret->bnx_rx_ret_ring[ret->bnx_rx_saved_considx];
2800 
2801 		rxidx = cur_rx->bge_idx;
2802 		KKASSERT(rxidx < BGE_STD_RX_RING_CNT);
2803 
2804 		BNX_INC(ret->bnx_rx_saved_considx, BNX_RETURN_RING_CNT);
2805 #ifdef BNX_RSS_DEBUG
2806 		ret->bnx_rx_pkt++;
2807 #endif
2808 
2809 		if (cur_rx->bge_flags & BGE_RXBDFLAG_VLAN_TAG) {
2810 			have_tag = 1;
2811 			vlan_tag = cur_rx->bge_vlan_tag;
2812 		}
2813 
2814 		if (ret->bnx_rx_cnt >= ret->bnx_rx_cntmax) {
2815 			atomic_add_int(&std->bnx_rx_std_used, std_used);
2816 			std_used = 0;
2817 
2818 			bnx_rx_std_refill_sched(ret, std);
2819 		}
2820 		ret->bnx_rx_cnt++;
2821 		++std_used;
2822 
2823 		rb = &std->bnx_rx_std_buf[rxidx];
2824 		m = rb->bnx_rx_mbuf;
2825 		if (cur_rx->bge_flags & BGE_RXBDFLAG_ERROR) {
2826 			IFNET_STAT_INC(ifp, ierrors, 1);
2827 			cpu_sfence();
2828 			rb->bnx_rx_refilled = 1;
2829 			continue;
2830 		}
2831 		if (bnx_newbuf_std(ret, rxidx, 0)) {
2832 			IFNET_STAT_INC(ifp, ierrors, 1);
2833 			continue;
2834 		}
2835 
2836 		IFNET_STAT_INC(ifp, ipackets, 1);
2837 		m->m_pkthdr.len = m->m_len = cur_rx->bge_len - ETHER_CRC_LEN;
2838 		m->m_pkthdr.rcvif = ifp;
2839 
2840 		if ((ifp->if_capenable & IFCAP_RXCSUM) &&
2841 		    (cur_rx->bge_flags & BGE_RXBDFLAG_IPV6) == 0) {
2842 			if (cur_rx->bge_flags & BGE_RXBDFLAG_IP_CSUM) {
2843 				m->m_pkthdr.csum_flags |= CSUM_IP_CHECKED;
2844 				if ((cur_rx->bge_error_flag &
2845 				    BGE_RXERRFLAG_IP_CSUM_NOK) == 0)
2846 					m->m_pkthdr.csum_flags |= CSUM_IP_VALID;
2847 			}
2848 			if (cur_rx->bge_flags & BGE_RXBDFLAG_TCP_UDP_CSUM) {
2849 				m->m_pkthdr.csum_data =
2850 				    cur_rx->bge_tcp_udp_csum;
2851 				m->m_pkthdr.csum_flags |= CSUM_DATA_VALID |
2852 				    CSUM_PSEUDO_HDR;
2853 			}
2854 		}
2855 		if (ifp->if_capenable & IFCAP_RSS) {
2856 			pi = bnx_rss_info(&pi0, cur_rx);
2857 			if (pi != NULL &&
2858 			    (cur_rx->bge_flags & BGE_RXBDFLAG_RSS_HASH)) {
2859 				m->m_flags |= M_HASH;
2860 				m->m_pkthdr.hash =
2861 				    toeplitz_hash(cur_rx->bge_hash);
2862 			}
2863 		}
2864 
2865 		/*
2866 		 * If we received a packet with a vlan tag, pass it
2867 		 * to vlan_input() instead of ether_input().
2868 		 */
2869 		if (have_tag) {
2870 			m->m_flags |= M_VLANTAG;
2871 			m->m_pkthdr.ether_vlantag = vlan_tag;
2872 		}
2873 		ifp->if_input(ifp, m, pi, cpuid);
2874 	}
2875 	bnx_writembx(sc, ret->bnx_rx_mbx, ret->bnx_rx_saved_considx);
2876 
2877 	if (std_used > 0) {
2878 		int cur_std_used;
2879 
2880 		cur_std_used = atomic_fetchadd_int(&std->bnx_rx_std_used,
2881 		    std_used);
2882 		if (cur_std_used + std_used >= (BGE_STD_RX_RING_CNT / 2)) {
2883 #ifdef BNX_RSS_DEBUG
2884 			ret->bnx_rx_force_sched++;
2885 #endif
2886 			bnx_rx_std_refill_sched(ret, std);
2887 		}
2888 	}
2889 }
2890 
2891 static void
2892 bnx_txeof(struct bnx_tx_ring *txr, uint16_t tx_cons)
2893 {
2894 	struct ifnet *ifp = &txr->bnx_sc->arpcom.ac_if;
2895 
2896 	/*
2897 	 * Go through our tx ring and free mbufs for those
2898 	 * frames that have been sent.
2899 	 */
2900 	while (txr->bnx_tx_saved_considx != tx_cons) {
2901 		struct bnx_tx_buf *buf;
2902 		uint32_t idx = 0;
2903 
2904 		idx = txr->bnx_tx_saved_considx;
2905 		buf = &txr->bnx_tx_buf[idx];
2906 		if (buf->bnx_tx_mbuf != NULL) {
2907 			IFNET_STAT_INC(ifp, opackets, 1);
2908 #ifdef BNX_TSS_DEBUG
2909 			txr->bnx_tx_pkt++;
2910 #endif
2911 			bus_dmamap_unload(txr->bnx_tx_mtag,
2912 			    buf->bnx_tx_dmamap);
2913 			m_freem(buf->bnx_tx_mbuf);
2914 			buf->bnx_tx_mbuf = NULL;
2915 		}
2916 		txr->bnx_tx_cnt--;
2917 		BNX_INC(txr->bnx_tx_saved_considx, BGE_TX_RING_CNT);
2918 	}
2919 
2920 	if ((BGE_TX_RING_CNT - txr->bnx_tx_cnt) >=
2921 	    (BNX_NSEG_RSVD + BNX_NSEG_SPARE))
2922 		ifsq_clr_oactive(txr->bnx_ifsq);
2923 
2924 	if (txr->bnx_tx_cnt == 0)
2925 		txr->bnx_tx_watchdog.wd_timer = 0;
2926 
2927 	if (!ifsq_is_empty(txr->bnx_ifsq))
2928 		ifsq_devstart(txr->bnx_ifsq);
2929 }
2930 
2931 static int
2932 bnx_handle_status(struct bnx_softc *sc)
2933 {
2934 	uint32_t status;
2935 	int handle = 0;
2936 
2937 	status = *sc->bnx_hw_status;
2938 
2939 	if (status & BGE_STATFLAG_ERROR) {
2940 		uint32_t val;
2941 		int reset = 0;
2942 
2943 		sc->bnx_errors++;
2944 
2945 		val = CSR_READ_4(sc, BGE_FLOW_ATTN);
2946 		if (val & ~BGE_FLOWATTN_MB_LOWAT) {
2947 			if_printf(&sc->arpcom.ac_if,
2948 			    "flow attn 0x%08x\n", val);
2949 			reset = 1;
2950 		}
2951 
2952 		val = CSR_READ_4(sc, BGE_MSI_STATUS);
2953 		if (val & ~BGE_MSISTAT_MSI_PCI_REQ) {
2954 			if_printf(&sc->arpcom.ac_if,
2955 			    "msi status 0x%08x\n", val);
2956 			reset = 1;
2957 		}
2958 
2959 		val = CSR_READ_4(sc, BGE_RDMA_STATUS);
2960 		if (val) {
2961 			if_printf(&sc->arpcom.ac_if,
2962 			    "rmda status 0x%08x\n", val);
2963 			reset = 1;
2964 		}
2965 
2966 		val = CSR_READ_4(sc, BGE_WDMA_STATUS);
2967 		if (val) {
2968 			if_printf(&sc->arpcom.ac_if,
2969 			    "wdma status 0x%08x\n", val);
2970 			reset = 1;
2971 		}
2972 
2973 		if (reset) {
2974 			bnx_serialize_skipmain(sc);
2975 			bnx_init(sc);
2976 			bnx_deserialize_skipmain(sc);
2977 		}
2978 		handle = 1;
2979 	}
2980 
2981 	if ((status & BGE_STATFLAG_LINKSTATE_CHANGED) || sc->bnx_link_evt) {
2982 		if (bootverbose) {
2983 			if_printf(&sc->arpcom.ac_if, "link change, "
2984 			    "link_evt %d\n", sc->bnx_link_evt);
2985 		}
2986 		bnx_link_poll(sc);
2987 		handle = 1;
2988 	}
2989 
2990 	return handle;
2991 }
2992 
2993 #ifdef IFPOLL_ENABLE
2994 
2995 static void
2996 bnx_npoll_rx(struct ifnet *ifp __unused, void *xret, int cycle)
2997 {
2998 	struct bnx_rx_ret_ring *ret = xret;
2999 	uint16_t rx_prod;
3000 
3001 	ASSERT_SERIALIZED(&ret->bnx_rx_ret_serialize);
3002 
3003 	ret->bnx_saved_status_tag = *ret->bnx_hw_status_tag;
3004 	cpu_lfence();
3005 
3006 	rx_prod = *ret->bnx_rx_considx;
3007 	if (ret->bnx_rx_saved_considx != rx_prod)
3008 		bnx_rxeof(ret, rx_prod, cycle);
3009 }
3010 
3011 static void
3012 bnx_npoll_tx_notag(struct ifnet *ifp __unused, void *xtxr, int cycle __unused)
3013 {
3014 	struct bnx_tx_ring *txr = xtxr;
3015 	uint16_t tx_cons;
3016 
3017 	ASSERT_SERIALIZED(&txr->bnx_tx_serialize);
3018 
3019 	tx_cons = *txr->bnx_tx_considx;
3020 	if (txr->bnx_tx_saved_considx != tx_cons)
3021 		bnx_txeof(txr, tx_cons);
3022 }
3023 
3024 static void
3025 bnx_npoll_tx(struct ifnet *ifp, void *xtxr, int cycle)
3026 {
3027 	struct bnx_tx_ring *txr = xtxr;
3028 
3029 	ASSERT_SERIALIZED(&txr->bnx_tx_serialize);
3030 
3031 	txr->bnx_saved_status_tag = *txr->bnx_hw_status_tag;
3032 	cpu_lfence();
3033 	bnx_npoll_tx_notag(ifp, txr, cycle);
3034 }
3035 
3036 static void
3037 bnx_npoll_status_notag(struct ifnet *ifp)
3038 {
3039 	struct bnx_softc *sc = ifp->if_softc;
3040 
3041 	ASSERT_SERIALIZED(&sc->bnx_main_serialize);
3042 
3043 	if (bnx_handle_status(sc)) {
3044 		/*
3045 		 * Status changes are handled; force the chip to
3046 		 * update the status block to reflect whether there
3047 		 * are more status changes or not, else staled status
3048 		 * changes are always seen.
3049 		 */
3050 		BNX_SETBIT(sc, BGE_HCC_MODE, BGE_HCCMODE_COAL_NOW);
3051 	}
3052 }
3053 
3054 static void
3055 bnx_npoll_status(struct ifnet *ifp)
3056 {
3057 	struct bnx_softc *sc = ifp->if_softc;
3058 
3059 	ASSERT_SERIALIZED(&sc->bnx_main_serialize);
3060 
3061 	sc->bnx_saved_status_tag = *sc->bnx_hw_status_tag;
3062 	cpu_lfence();
3063 	bnx_npoll_status_notag(ifp);
3064 }
3065 
3066 static void
3067 bnx_npoll(struct ifnet *ifp, struct ifpoll_info *info)
3068 {
3069 	struct bnx_softc *sc = ifp->if_softc;
3070 	int i;
3071 
3072 	ASSERT_IFNET_SERIALIZED_ALL(ifp);
3073 
3074 	if (info != NULL) {
3075 		if (sc->bnx_flags & BNX_FLAG_STATUS_HASTAG)
3076 			info->ifpi_status.status_func = bnx_npoll_status;
3077 		else
3078 			info->ifpi_status.status_func = bnx_npoll_status_notag;
3079 		info->ifpi_status.serializer = &sc->bnx_main_serialize;
3080 
3081 		for (i = 0; i < sc->bnx_tx_ringcnt; ++i) {
3082 			struct bnx_tx_ring *txr = &sc->bnx_tx_ring[i];
3083 			int idx = i + sc->bnx_npoll_txoff;
3084 
3085 			KKASSERT(idx < ncpus2);
3086 			if (sc->bnx_flags & BNX_FLAG_RXTX_BUNDLE) {
3087 				info->ifpi_tx[idx].poll_func =
3088 				    bnx_npoll_tx_notag;
3089 			} else {
3090 				info->ifpi_tx[idx].poll_func = bnx_npoll_tx;
3091 			}
3092 			info->ifpi_tx[idx].arg = txr;
3093 			info->ifpi_tx[idx].serializer = &txr->bnx_tx_serialize;
3094 			ifsq_set_cpuid(txr->bnx_ifsq, idx);
3095 		}
3096 
3097 		for (i = 0; i < sc->bnx_rx_retcnt; ++i) {
3098 			struct bnx_rx_ret_ring *ret = &sc->bnx_rx_ret_ring[i];
3099 			int idx = i + sc->bnx_npoll_rxoff;
3100 
3101 			KKASSERT(idx < ncpus2);
3102 			info->ifpi_rx[idx].poll_func = bnx_npoll_rx;
3103 			info->ifpi_rx[idx].arg = ret;
3104 			info->ifpi_rx[idx].serializer =
3105 			    &ret->bnx_rx_ret_serialize;
3106 		}
3107 
3108 		if (ifp->if_flags & IFF_RUNNING) {
3109 			bnx_disable_intr(sc);
3110 			bnx_set_tick_cpuid(sc, TRUE);
3111 
3112 			sc->bnx_coal_chg = BNX_TX_COAL_BDS_CHG |
3113 			    BNX_RX_COAL_BDS_CHG;
3114 			bnx_coal_change(sc);
3115 		}
3116 	} else {
3117 		for (i = 0; i < sc->bnx_tx_ringcnt; ++i) {
3118 			ifsq_set_cpuid(sc->bnx_tx_ring[i].bnx_ifsq,
3119 			    sc->bnx_tx_ring[i].bnx_tx_cpuid);
3120 		}
3121 		if (ifp->if_flags & IFF_RUNNING) {
3122 			sc->bnx_coal_chg = BNX_TX_COAL_BDS_CHG |
3123 			    BNX_RX_COAL_BDS_CHG;
3124 			bnx_coal_change(sc);
3125 
3126 			bnx_enable_intr(sc);
3127 			bnx_set_tick_cpuid(sc, FALSE);
3128 		}
3129 	}
3130 }
3131 
3132 #endif	/* IFPOLL_ENABLE */
3133 
3134 static void
3135 bnx_intr_legacy(void *xsc)
3136 {
3137 	struct bnx_softc *sc = xsc;
3138 	struct bnx_rx_ret_ring *ret = &sc->bnx_rx_ret_ring[0];
3139 
3140 	if (ret->bnx_saved_status_tag == *ret->bnx_hw_status_tag) {
3141 		uint32_t val;
3142 
3143 		val = pci_read_config(sc->bnx_dev, BGE_PCI_PCISTATE, 4);
3144 		if (val & BGE_PCISTAT_INTR_NOTACT)
3145 			return;
3146 	}
3147 
3148 	/*
3149 	 * NOTE:
3150 	 * Interrupt will have to be disabled if tagged status
3151 	 * is used, else interrupt will always be asserted on
3152 	 * certain chips (at least on BCM5750 AX/BX).
3153 	 */
3154 	bnx_writembx(sc, BGE_MBX_IRQ0_LO, 1);
3155 
3156 	bnx_intr(sc);
3157 }
3158 
3159 static void
3160 bnx_msi(void *xsc)
3161 {
3162 	bnx_intr(xsc);
3163 }
3164 
3165 static void
3166 bnx_intr(struct bnx_softc *sc)
3167 {
3168 	struct ifnet *ifp = &sc->arpcom.ac_if;
3169 	struct bnx_rx_ret_ring *ret = &sc->bnx_rx_ret_ring[0];
3170 
3171 	ASSERT_SERIALIZED(&sc->bnx_main_serialize);
3172 
3173 	ret->bnx_saved_status_tag = *ret->bnx_hw_status_tag;
3174 	/*
3175 	 * Use a load fence to ensure that status_tag is saved
3176 	 * before rx_prod, tx_cons and status.
3177 	 */
3178 	cpu_lfence();
3179 
3180 	bnx_handle_status(sc);
3181 
3182 	if (ifp->if_flags & IFF_RUNNING) {
3183 		struct bnx_tx_ring *txr = &sc->bnx_tx_ring[0];
3184 		uint16_t rx_prod, tx_cons;
3185 
3186 		lwkt_serialize_enter(&ret->bnx_rx_ret_serialize);
3187 		rx_prod = *ret->bnx_rx_considx;
3188 		if (ret->bnx_rx_saved_considx != rx_prod)
3189 			bnx_rxeof(ret, rx_prod, -1);
3190 		lwkt_serialize_exit(&ret->bnx_rx_ret_serialize);
3191 
3192 		lwkt_serialize_enter(&txr->bnx_tx_serialize);
3193 		tx_cons = *txr->bnx_tx_considx;
3194 		if (txr->bnx_tx_saved_considx != tx_cons)
3195 			bnx_txeof(txr, tx_cons);
3196 		lwkt_serialize_exit(&txr->bnx_tx_serialize);
3197 	}
3198 
3199 	bnx_writembx(sc, BGE_MBX_IRQ0_LO, ret->bnx_saved_status_tag << 24);
3200 }
3201 
3202 static void
3203 bnx_msix_tx_status(void *xtxr)
3204 {
3205 	struct bnx_tx_ring *txr = xtxr;
3206 	struct bnx_softc *sc = txr->bnx_sc;
3207 	struct ifnet *ifp = &sc->arpcom.ac_if;
3208 
3209 	ASSERT_SERIALIZED(&sc->bnx_main_serialize);
3210 
3211 	txr->bnx_saved_status_tag = *txr->bnx_hw_status_tag;
3212 	/*
3213 	 * Use a load fence to ensure that status_tag is saved
3214 	 * before tx_cons and status.
3215 	 */
3216 	cpu_lfence();
3217 
3218 	bnx_handle_status(sc);
3219 
3220 	if (ifp->if_flags & IFF_RUNNING) {
3221 		uint16_t tx_cons;
3222 
3223 		lwkt_serialize_enter(&txr->bnx_tx_serialize);
3224 		tx_cons = *txr->bnx_tx_considx;
3225 		if (txr->bnx_tx_saved_considx != tx_cons)
3226 			bnx_txeof(txr, tx_cons);
3227 		lwkt_serialize_exit(&txr->bnx_tx_serialize);
3228 	}
3229 
3230 	bnx_writembx(sc, BGE_MBX_IRQ0_LO, txr->bnx_saved_status_tag << 24);
3231 }
3232 
3233 static void
3234 bnx_msix_rx(void *xret)
3235 {
3236 	struct bnx_rx_ret_ring *ret = xret;
3237 	uint16_t rx_prod;
3238 
3239 	ASSERT_SERIALIZED(&ret->bnx_rx_ret_serialize);
3240 
3241 	ret->bnx_saved_status_tag = *ret->bnx_hw_status_tag;
3242 	/*
3243 	 * Use a load fence to ensure that status_tag is saved
3244 	 * before rx_prod.
3245 	 */
3246 	cpu_lfence();
3247 
3248 	rx_prod = *ret->bnx_rx_considx;
3249 	if (ret->bnx_rx_saved_considx != rx_prod)
3250 		bnx_rxeof(ret, rx_prod, -1);
3251 
3252 	bnx_writembx(ret->bnx_sc, ret->bnx_msix_mbx,
3253 	    ret->bnx_saved_status_tag << 24);
3254 }
3255 
3256 static void
3257 bnx_msix_rxtx(void *xret)
3258 {
3259 	struct bnx_rx_ret_ring *ret = xret;
3260 	struct bnx_tx_ring *txr = ret->bnx_txr;
3261 	uint16_t rx_prod, tx_cons;
3262 
3263 	ASSERT_SERIALIZED(&ret->bnx_rx_ret_serialize);
3264 
3265 	ret->bnx_saved_status_tag = *ret->bnx_hw_status_tag;
3266 	/*
3267 	 * Use a load fence to ensure that status_tag is saved
3268 	 * before rx_prod and tx_cons.
3269 	 */
3270 	cpu_lfence();
3271 
3272 	rx_prod = *ret->bnx_rx_considx;
3273 	if (ret->bnx_rx_saved_considx != rx_prod)
3274 		bnx_rxeof(ret, rx_prod, -1);
3275 
3276 	lwkt_serialize_enter(&txr->bnx_tx_serialize);
3277 	tx_cons = *txr->bnx_tx_considx;
3278 	if (txr->bnx_tx_saved_considx != tx_cons)
3279 		bnx_txeof(txr, tx_cons);
3280 	lwkt_serialize_exit(&txr->bnx_tx_serialize);
3281 
3282 	bnx_writembx(ret->bnx_sc, ret->bnx_msix_mbx,
3283 	    ret->bnx_saved_status_tag << 24);
3284 }
3285 
3286 static void
3287 bnx_msix_status(void *xsc)
3288 {
3289 	struct bnx_softc *sc = xsc;
3290 
3291 	ASSERT_SERIALIZED(&sc->bnx_main_serialize);
3292 
3293 	sc->bnx_saved_status_tag = *sc->bnx_hw_status_tag;
3294 	/*
3295 	 * Use a load fence to ensure that status_tag is saved
3296 	 * before status.
3297 	 */
3298 	cpu_lfence();
3299 
3300 	bnx_handle_status(sc);
3301 
3302 	bnx_writembx(sc, BGE_MBX_IRQ0_LO, sc->bnx_saved_status_tag << 24);
3303 }
3304 
3305 static void
3306 bnx_tick(void *xsc)
3307 {
3308 	struct bnx_softc *sc = xsc;
3309 
3310 	lwkt_serialize_enter(&sc->bnx_main_serialize);
3311 
3312 	bnx_stats_update_regs(sc);
3313 
3314 	if (sc->bnx_flags & BNX_FLAG_TBI) {
3315 		/*
3316 		 * Since in TBI mode auto-polling can't be used we should poll
3317 		 * link status manually. Here we register pending link event
3318 		 * and trigger interrupt.
3319 		 */
3320 		sc->bnx_link_evt++;
3321 		BNX_SETBIT(sc, BGE_HCC_MODE, BGE_HCCMODE_COAL_NOW);
3322 	} else if (!sc->bnx_link) {
3323 		mii_tick(device_get_softc(sc->bnx_miibus));
3324 	}
3325 
3326 	callout_reset_bycpu(&sc->bnx_tick_timer, hz, bnx_tick, sc,
3327 	    sc->bnx_tick_cpuid);
3328 
3329 	lwkt_serialize_exit(&sc->bnx_main_serialize);
3330 }
3331 
3332 static void
3333 bnx_stats_update_regs(struct bnx_softc *sc)
3334 {
3335 	struct ifnet *ifp = &sc->arpcom.ac_if;
3336 	struct bge_mac_stats_regs stats;
3337 	uint32_t *s, val;
3338 	int i;
3339 
3340 	s = (uint32_t *)&stats;
3341 	for (i = 0; i < sizeof(struct bge_mac_stats_regs); i += 4) {
3342 		*s = CSR_READ_4(sc, BGE_RX_STATS + i);
3343 		s++;
3344 	}
3345 
3346 	IFNET_STAT_SET(ifp, collisions,
3347 	   (stats.dot3StatsSingleCollisionFrames +
3348 	   stats.dot3StatsMultipleCollisionFrames +
3349 	   stats.dot3StatsExcessiveCollisions +
3350 	   stats.dot3StatsLateCollisions));
3351 
3352 	val = CSR_READ_4(sc, BGE_RXLP_LOCSTAT_OUT_OF_BDS);
3353 	sc->bnx_norxbds += val;
3354 }
3355 
3356 /*
3357  * Encapsulate an mbuf chain in the tx ring  by coupling the mbuf data
3358  * pointers to descriptors.
3359  */
3360 static int
3361 bnx_encap(struct bnx_tx_ring *txr, struct mbuf **m_head0, uint32_t *txidx,
3362     int *segs_used)
3363 {
3364 	struct bge_tx_bd *d = NULL;
3365 	uint16_t csum_flags = 0, vlan_tag = 0, mss = 0;
3366 	bus_dma_segment_t segs[BNX_NSEG_NEW];
3367 	bus_dmamap_t map;
3368 	int error, maxsegs, nsegs, idx, i;
3369 	struct mbuf *m_head = *m_head0, *m_new;
3370 
3371 	if (m_head->m_pkthdr.csum_flags & CSUM_TSO) {
3372 #ifdef BNX_TSO_DEBUG
3373 		int tso_nsegs;
3374 #endif
3375 
3376 		error = bnx_setup_tso(txr, m_head0, &mss, &csum_flags);
3377 		if (error)
3378 			return error;
3379 		m_head = *m_head0;
3380 
3381 #ifdef BNX_TSO_DEBUG
3382 		tso_nsegs = (m_head->m_pkthdr.len /
3383 		    m_head->m_pkthdr.tso_segsz) - 1;
3384 		if (tso_nsegs > (BNX_TSO_NSTATS - 1))
3385 			tso_nsegs = BNX_TSO_NSTATS - 1;
3386 		else if (tso_nsegs < 0)
3387 			tso_nsegs = 0;
3388 		txr->bnx_sc->bnx_tsosegs[tso_nsegs]++;
3389 #endif
3390 	} else if (m_head->m_pkthdr.csum_flags & BNX_CSUM_FEATURES) {
3391 		if (m_head->m_pkthdr.csum_flags & CSUM_IP)
3392 			csum_flags |= BGE_TXBDFLAG_IP_CSUM;
3393 		if (m_head->m_pkthdr.csum_flags & (CSUM_TCP | CSUM_UDP))
3394 			csum_flags |= BGE_TXBDFLAG_TCP_UDP_CSUM;
3395 		if (m_head->m_flags & M_LASTFRAG)
3396 			csum_flags |= BGE_TXBDFLAG_IP_FRAG_END;
3397 		else if (m_head->m_flags & M_FRAG)
3398 			csum_flags |= BGE_TXBDFLAG_IP_FRAG;
3399 	}
3400 	if (m_head->m_flags & M_VLANTAG) {
3401 		csum_flags |= BGE_TXBDFLAG_VLAN_TAG;
3402 		vlan_tag = m_head->m_pkthdr.ether_vlantag;
3403 	}
3404 
3405 	idx = *txidx;
3406 	map = txr->bnx_tx_buf[idx].bnx_tx_dmamap;
3407 
3408 	maxsegs = (BGE_TX_RING_CNT - txr->bnx_tx_cnt) - BNX_NSEG_RSVD;
3409 	KASSERT(maxsegs >= BNX_NSEG_SPARE,
3410 		("not enough segments %d", maxsegs));
3411 
3412 	if (maxsegs > BNX_NSEG_NEW)
3413 		maxsegs = BNX_NSEG_NEW;
3414 
3415 	/*
3416 	 * Pad outbound frame to BGE_MIN_FRAMELEN for an unusual reason.
3417 	 * The bge hardware will pad out Tx runts to BGE_MIN_FRAMELEN,
3418 	 * but when such padded frames employ the bge IP/TCP checksum
3419 	 * offload, the hardware checksum assist gives incorrect results
3420 	 * (possibly from incorporating its own padding into the UDP/TCP
3421 	 * checksum; who knows).  If we pad such runts with zeros, the
3422 	 * onboard checksum comes out correct.
3423 	 */
3424 	if ((csum_flags & BGE_TXBDFLAG_TCP_UDP_CSUM) &&
3425 	    m_head->m_pkthdr.len < BNX_MIN_FRAMELEN) {
3426 		error = m_devpad(m_head, BNX_MIN_FRAMELEN);
3427 		if (error)
3428 			goto back;
3429 	}
3430 
3431 	if ((txr->bnx_tx_flags & BNX_TX_FLAG_SHORTDMA) &&
3432 	    m_head->m_next != NULL) {
3433 		m_new = bnx_defrag_shortdma(m_head);
3434 		if (m_new == NULL) {
3435 			error = ENOBUFS;
3436 			goto back;
3437 		}
3438 		*m_head0 = m_head = m_new;
3439 	}
3440 	if ((m_head->m_pkthdr.csum_flags & CSUM_TSO) == 0 &&
3441 	    (txr->bnx_tx_flags & BNX_TX_FLAG_FORCE_DEFRAG) &&
3442 	    m_head->m_next != NULL) {
3443 		/*
3444 		 * Forcefully defragment mbuf chain to overcome hardware
3445 		 * limitation which only support a single outstanding
3446 		 * DMA read operation.  If it fails, keep moving on using
3447 		 * the original mbuf chain.
3448 		 */
3449 		m_new = m_defrag(m_head, MB_DONTWAIT);
3450 		if (m_new != NULL)
3451 			*m_head0 = m_head = m_new;
3452 	}
3453 
3454 	error = bus_dmamap_load_mbuf_defrag(txr->bnx_tx_mtag, map,
3455 	    m_head0, segs, maxsegs, &nsegs, BUS_DMA_NOWAIT);
3456 	if (error)
3457 		goto back;
3458 	*segs_used += nsegs;
3459 
3460 	m_head = *m_head0;
3461 	bus_dmamap_sync(txr->bnx_tx_mtag, map, BUS_DMASYNC_PREWRITE);
3462 
3463 	for (i = 0; ; i++) {
3464 		d = &txr->bnx_tx_ring[idx];
3465 
3466 		d->bge_addr.bge_addr_lo = BGE_ADDR_LO(segs[i].ds_addr);
3467 		d->bge_addr.bge_addr_hi = BGE_ADDR_HI(segs[i].ds_addr);
3468 		d->bge_len = segs[i].ds_len;
3469 		d->bge_flags = csum_flags;
3470 		d->bge_vlan_tag = vlan_tag;
3471 		d->bge_mss = mss;
3472 
3473 		if (i == nsegs - 1)
3474 			break;
3475 		BNX_INC(idx, BGE_TX_RING_CNT);
3476 	}
3477 	/* Mark the last segment as end of packet... */
3478 	d->bge_flags |= BGE_TXBDFLAG_END;
3479 
3480 	/*
3481 	 * Insure that the map for this transmission is placed at
3482 	 * the array index of the last descriptor in this chain.
3483 	 */
3484 	txr->bnx_tx_buf[*txidx].bnx_tx_dmamap = txr->bnx_tx_buf[idx].bnx_tx_dmamap;
3485 	txr->bnx_tx_buf[idx].bnx_tx_dmamap = map;
3486 	txr->bnx_tx_buf[idx].bnx_tx_mbuf = m_head;
3487 	txr->bnx_tx_cnt += nsegs;
3488 
3489 	BNX_INC(idx, BGE_TX_RING_CNT);
3490 	*txidx = idx;
3491 back:
3492 	if (error) {
3493 		m_freem(*m_head0);
3494 		*m_head0 = NULL;
3495 	}
3496 	return error;
3497 }
3498 
3499 /*
3500  * Main transmit routine. To avoid having to do mbuf copies, we put pointers
3501  * to the mbuf data regions directly in the transmit descriptors.
3502  */
3503 static void
3504 bnx_start(struct ifnet *ifp, struct ifaltq_subque *ifsq)
3505 {
3506 	struct bnx_tx_ring *txr = ifsq_get_priv(ifsq);
3507 	struct mbuf *m_head = NULL;
3508 	uint32_t prodidx;
3509 	int nsegs = 0;
3510 
3511 	KKASSERT(txr->bnx_ifsq == ifsq);
3512 	ASSERT_SERIALIZED(&txr->bnx_tx_serialize);
3513 
3514 	if ((ifp->if_flags & IFF_RUNNING) == 0 || ifsq_is_oactive(ifsq))
3515 		return;
3516 
3517 	prodidx = txr->bnx_tx_prodidx;
3518 
3519 	while (txr->bnx_tx_buf[prodidx].bnx_tx_mbuf == NULL) {
3520 		/*
3521 		 * Sanity check: avoid coming within BGE_NSEG_RSVD
3522 		 * descriptors of the end of the ring.  Also make
3523 		 * sure there are BGE_NSEG_SPARE descriptors for
3524 		 * jumbo buffers' or TSO segments' defragmentation.
3525 		 */
3526 		if ((BGE_TX_RING_CNT - txr->bnx_tx_cnt) <
3527 		    (BNX_NSEG_RSVD + BNX_NSEG_SPARE)) {
3528 			ifsq_set_oactive(ifsq);
3529 			break;
3530 		}
3531 
3532 		m_head = ifsq_dequeue(ifsq);
3533 		if (m_head == NULL)
3534 			break;
3535 
3536 		/*
3537 		 * Pack the data into the transmit ring. If we
3538 		 * don't have room, set the OACTIVE flag and wait
3539 		 * for the NIC to drain the ring.
3540 		 */
3541 		if (bnx_encap(txr, &m_head, &prodidx, &nsegs)) {
3542 			ifsq_set_oactive(ifsq);
3543 			IFNET_STAT_INC(ifp, oerrors, 1);
3544 			break;
3545 		}
3546 
3547 		if (nsegs >= txr->bnx_tx_wreg) {
3548 			/* Transmit */
3549 			bnx_writembx(txr->bnx_sc, txr->bnx_tx_mbx, prodidx);
3550 			nsegs = 0;
3551 		}
3552 
3553 		ETHER_BPF_MTAP(ifp, m_head);
3554 
3555 		/*
3556 		 * Set a timeout in case the chip goes out to lunch.
3557 		 */
3558 		txr->bnx_tx_watchdog.wd_timer = 5;
3559 	}
3560 
3561 	if (nsegs > 0) {
3562 		/* Transmit */
3563 		bnx_writembx(txr->bnx_sc, txr->bnx_tx_mbx, prodidx);
3564 	}
3565 	txr->bnx_tx_prodidx = prodidx;
3566 }
3567 
3568 static void
3569 bnx_init(void *xsc)
3570 {
3571 	struct bnx_softc *sc = xsc;
3572 	struct ifnet *ifp = &sc->arpcom.ac_if;
3573 	uint16_t *m;
3574 	uint32_t mode;
3575 	int i;
3576 	boolean_t polling;
3577 
3578 	ASSERT_IFNET_SERIALIZED_ALL(ifp);
3579 
3580 	/* Cancel pending I/O and flush buffers. */
3581 	bnx_stop(sc);
3582 
3583 	bnx_sig_pre_reset(sc, BNX_RESET_START);
3584 	bnx_reset(sc);
3585 	bnx_sig_post_reset(sc, BNX_RESET_START);
3586 
3587 	bnx_chipinit(sc);
3588 
3589 	/*
3590 	 * Init the various state machines, ring
3591 	 * control blocks and firmware.
3592 	 */
3593 	if (bnx_blockinit(sc)) {
3594 		if_printf(ifp, "initialization failure\n");
3595 		bnx_stop(sc);
3596 		return;
3597 	}
3598 
3599 	/* Specify MTU. */
3600 	CSR_WRITE_4(sc, BGE_RX_MTU, ifp->if_mtu +
3601 	    ETHER_HDR_LEN + ETHER_CRC_LEN + EVL_ENCAPLEN);
3602 
3603 	/* Load our MAC address. */
3604 	m = (uint16_t *)&sc->arpcom.ac_enaddr[0];
3605 	CSR_WRITE_4(sc, BGE_MAC_ADDR1_LO, htons(m[0]));
3606 	CSR_WRITE_4(sc, BGE_MAC_ADDR1_HI, (htons(m[1]) << 16) | htons(m[2]));
3607 
3608 	/* Enable or disable promiscuous mode as needed. */
3609 	bnx_setpromisc(sc);
3610 
3611 	/* Program multicast filter. */
3612 	bnx_setmulti(sc);
3613 
3614 	/* Init RX ring. */
3615 	if (bnx_init_rx_ring_std(&sc->bnx_rx_std_ring)) {
3616 		if_printf(ifp, "RX ring initialization failed\n");
3617 		bnx_stop(sc);
3618 		return;
3619 	}
3620 
3621 	/* Init jumbo RX ring. */
3622 	if (ifp->if_mtu > (ETHERMTU + ETHER_HDR_LEN + ETHER_CRC_LEN)) {
3623 		if (bnx_init_rx_ring_jumbo(sc)) {
3624 			if_printf(ifp, "Jumbo RX ring initialization failed\n");
3625 			bnx_stop(sc);
3626 			return;
3627 		}
3628 	}
3629 
3630 	/* Init our RX return ring index */
3631 	for (i = 0; i < sc->bnx_rx_retcnt; ++i) {
3632 		struct bnx_rx_ret_ring *ret = &sc->bnx_rx_ret_ring[i];
3633 
3634 		ret->bnx_rx_saved_considx = 0;
3635 		ret->bnx_rx_cnt = 0;
3636 	}
3637 
3638 	/* Init TX ring. */
3639 	for (i = 0; i < sc->bnx_tx_ringcnt; ++i)
3640 		bnx_init_tx_ring(&sc->bnx_tx_ring[i]);
3641 
3642 	/* Enable TX MAC state machine lockup fix. */
3643 	mode = CSR_READ_4(sc, BGE_TX_MODE);
3644 	mode |= BGE_TXMODE_MBUF_LOCKUP_FIX;
3645 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5720 ||
3646 	    sc->bnx_asicrev == BGE_ASICREV_BCM5762) {
3647 		mode &= ~(BGE_TXMODE_JMB_FRM_LEN | BGE_TXMODE_CNT_DN_MODE);
3648 		mode |= CSR_READ_4(sc, BGE_TX_MODE) &
3649 		    (BGE_TXMODE_JMB_FRM_LEN | BGE_TXMODE_CNT_DN_MODE);
3650 	}
3651 	/* Turn on transmitter */
3652 	CSR_WRITE_4(sc, BGE_TX_MODE, mode | BGE_TXMODE_ENABLE);
3653 	DELAY(100);
3654 
3655 	/* Initialize RSS */
3656 	mode = BGE_RXMODE_ENABLE | BGE_RXMODE_IPV6_ENABLE;
3657 	if (BNX_RSS_ENABLED(sc)) {
3658 		bnx_init_rss(sc);
3659 		mode |= BGE_RXMODE_RSS_ENABLE |
3660 		    BGE_RXMODE_RSS_HASH_MASK_BITS |
3661 		    BGE_RXMODE_RSS_IPV4_HASH |
3662 		    BGE_RXMODE_RSS_TCP_IPV4_HASH;
3663 	}
3664 	/* Turn on receiver */
3665 	BNX_SETBIT(sc, BGE_RX_MODE, mode);
3666 	DELAY(10);
3667 
3668 	/*
3669 	 * Set the number of good frames to receive after RX MBUF
3670 	 * Low Watermark has been reached.  After the RX MAC receives
3671 	 * this number of frames, it will drop subsequent incoming
3672 	 * frames until the MBUF High Watermark is reached.
3673 	 */
3674 	if (BNX_IS_57765_FAMILY(sc))
3675 		CSR_WRITE_4(sc, BGE_MAX_RX_FRAME_LOWAT, 1);
3676 	else
3677 		CSR_WRITE_4(sc, BGE_MAX_RX_FRAME_LOWAT, 2);
3678 
3679 	if (sc->bnx_intr_type == PCI_INTR_TYPE_MSI ||
3680 	    sc->bnx_intr_type == PCI_INTR_TYPE_MSIX) {
3681 		if (bootverbose) {
3682 			if_printf(ifp, "MSI_MODE: %#x\n",
3683 			    CSR_READ_4(sc, BGE_MSI_MODE));
3684 		}
3685 	}
3686 
3687 	/* Tell firmware we're alive. */
3688 	BNX_SETBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
3689 
3690 	/* Enable host interrupts if polling(4) is not enabled. */
3691 	PCI_SETBIT(sc->bnx_dev, BGE_PCI_MISC_CTL, BGE_PCIMISCCTL_CLEAR_INTA, 4);
3692 
3693 	polling = FALSE;
3694 #ifdef IFPOLL_ENABLE
3695 	if (ifp->if_flags & IFF_NPOLLING)
3696 		polling = TRUE;
3697 #endif
3698 	if (polling)
3699 		bnx_disable_intr(sc);
3700 	else
3701 		bnx_enable_intr(sc);
3702 	bnx_set_tick_cpuid(sc, polling);
3703 
3704 	ifp->if_flags |= IFF_RUNNING;
3705 	for (i = 0; i < sc->bnx_tx_ringcnt; ++i) {
3706 		struct bnx_tx_ring *txr = &sc->bnx_tx_ring[i];
3707 
3708 		ifsq_clr_oactive(txr->bnx_ifsq);
3709 		ifsq_watchdog_start(&txr->bnx_tx_watchdog);
3710 	}
3711 
3712 	bnx_ifmedia_upd(ifp);
3713 
3714 	callout_reset_bycpu(&sc->bnx_tick_timer, hz, bnx_tick, sc,
3715 	    sc->bnx_tick_cpuid);
3716 }
3717 
3718 /*
3719  * Set media options.
3720  */
3721 static int
3722 bnx_ifmedia_upd(struct ifnet *ifp)
3723 {
3724 	struct bnx_softc *sc = ifp->if_softc;
3725 
3726 	/* If this is a 1000baseX NIC, enable the TBI port. */
3727 	if (sc->bnx_flags & BNX_FLAG_TBI) {
3728 		struct ifmedia *ifm = &sc->bnx_ifmedia;
3729 
3730 		if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
3731 			return(EINVAL);
3732 
3733 		switch(IFM_SUBTYPE(ifm->ifm_media)) {
3734 		case IFM_AUTO:
3735 			break;
3736 
3737 		case IFM_1000_SX:
3738 			if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX) {
3739 				BNX_CLRBIT(sc, BGE_MAC_MODE,
3740 				    BGE_MACMODE_HALF_DUPLEX);
3741 			} else {
3742 				BNX_SETBIT(sc, BGE_MAC_MODE,
3743 				    BGE_MACMODE_HALF_DUPLEX);
3744 			}
3745 			DELAY(40);
3746 			break;
3747 		default:
3748 			return(EINVAL);
3749 		}
3750 	} else {
3751 		struct mii_data *mii = device_get_softc(sc->bnx_miibus);
3752 
3753 		sc->bnx_link_evt++;
3754 		sc->bnx_link = 0;
3755 		if (mii->mii_instance) {
3756 			struct mii_softc *miisc;
3757 
3758 			LIST_FOREACH(miisc, &mii->mii_phys, mii_list)
3759 				mii_phy_reset(miisc);
3760 		}
3761 		mii_mediachg(mii);
3762 
3763 		/*
3764 		 * Force an interrupt so that we will call bnx_link_upd
3765 		 * if needed and clear any pending link state attention.
3766 		 * Without this we are not getting any further interrupts
3767 		 * for link state changes and thus will not UP the link and
3768 		 * not be able to send in bnx_start.  The only way to get
3769 		 * things working was to receive a packet and get an RX
3770 		 * intr.
3771 		 *
3772 		 * bnx_tick should help for fiber cards and we might not
3773 		 * need to do this here if BNX_FLAG_TBI is set but as
3774 		 * we poll for fiber anyway it should not harm.
3775 		 */
3776 		BNX_SETBIT(sc, BGE_HCC_MODE, BGE_HCCMODE_COAL_NOW);
3777 	}
3778 	return(0);
3779 }
3780 
3781 /*
3782  * Report current media status.
3783  */
3784 static void
3785 bnx_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr)
3786 {
3787 	struct bnx_softc *sc = ifp->if_softc;
3788 
3789 	if ((ifp->if_flags & IFF_RUNNING) == 0)
3790 		return;
3791 
3792 	if (sc->bnx_flags & BNX_FLAG_TBI) {
3793 		ifmr->ifm_status = IFM_AVALID;
3794 		ifmr->ifm_active = IFM_ETHER;
3795 		if (CSR_READ_4(sc, BGE_MAC_STS) &
3796 		    BGE_MACSTAT_TBI_PCS_SYNCHED) {
3797 			ifmr->ifm_status |= IFM_ACTIVE;
3798 		} else {
3799 			ifmr->ifm_active |= IFM_NONE;
3800 			return;
3801 		}
3802 
3803 		ifmr->ifm_active |= IFM_1000_SX;
3804 		if (CSR_READ_4(sc, BGE_MAC_MODE) & BGE_MACMODE_HALF_DUPLEX)
3805 			ifmr->ifm_active |= IFM_HDX;
3806 		else
3807 			ifmr->ifm_active |= IFM_FDX;
3808 	} else {
3809 		struct mii_data *mii = device_get_softc(sc->bnx_miibus);
3810 
3811 		mii_pollstat(mii);
3812 		ifmr->ifm_active = mii->mii_media_active;
3813 		ifmr->ifm_status = mii->mii_media_status;
3814 	}
3815 }
3816 
3817 static int
3818 bnx_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
3819 {
3820 	struct bnx_softc *sc = ifp->if_softc;
3821 	struct ifreq *ifr = (struct ifreq *)data;
3822 	int mask, error = 0;
3823 
3824 	ASSERT_IFNET_SERIALIZED_ALL(ifp);
3825 
3826 	switch (command) {
3827 	case SIOCSIFMTU:
3828 		if ((!BNX_IS_JUMBO_CAPABLE(sc) && ifr->ifr_mtu > ETHERMTU) ||
3829 		    (BNX_IS_JUMBO_CAPABLE(sc) &&
3830 		     ifr->ifr_mtu > BNX_JUMBO_MTU)) {
3831 			error = EINVAL;
3832 		} else if (ifp->if_mtu != ifr->ifr_mtu) {
3833 			ifp->if_mtu = ifr->ifr_mtu;
3834 			if (ifp->if_flags & IFF_RUNNING)
3835 				bnx_init(sc);
3836 		}
3837 		break;
3838 	case SIOCSIFFLAGS:
3839 		if (ifp->if_flags & IFF_UP) {
3840 			if (ifp->if_flags & IFF_RUNNING) {
3841 				mask = ifp->if_flags ^ sc->bnx_if_flags;
3842 
3843 				/*
3844 				 * If only the state of the PROMISC flag
3845 				 * changed, then just use the 'set promisc
3846 				 * mode' command instead of reinitializing
3847 				 * the entire NIC. Doing a full re-init
3848 				 * means reloading the firmware and waiting
3849 				 * for it to start up, which may take a
3850 				 * second or two.  Similarly for ALLMULTI.
3851 				 */
3852 				if (mask & IFF_PROMISC)
3853 					bnx_setpromisc(sc);
3854 				if (mask & IFF_ALLMULTI)
3855 					bnx_setmulti(sc);
3856 			} else {
3857 				bnx_init(sc);
3858 			}
3859 		} else if (ifp->if_flags & IFF_RUNNING) {
3860 			bnx_stop(sc);
3861 		}
3862 		sc->bnx_if_flags = ifp->if_flags;
3863 		break;
3864 	case SIOCADDMULTI:
3865 	case SIOCDELMULTI:
3866 		if (ifp->if_flags & IFF_RUNNING)
3867 			bnx_setmulti(sc);
3868 		break;
3869 	case SIOCSIFMEDIA:
3870 	case SIOCGIFMEDIA:
3871 		if (sc->bnx_flags & BNX_FLAG_TBI) {
3872 			error = ifmedia_ioctl(ifp, ifr,
3873 			    &sc->bnx_ifmedia, command);
3874 		} else {
3875 			struct mii_data *mii;
3876 
3877 			mii = device_get_softc(sc->bnx_miibus);
3878 			error = ifmedia_ioctl(ifp, ifr,
3879 					      &mii->mii_media, command);
3880 		}
3881 		break;
3882         case SIOCSIFCAP:
3883 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
3884 		if (mask & IFCAP_HWCSUM) {
3885 			ifp->if_capenable ^= (mask & IFCAP_HWCSUM);
3886 			if (ifp->if_capenable & IFCAP_TXCSUM)
3887 				ifp->if_hwassist |= BNX_CSUM_FEATURES;
3888 			else
3889 				ifp->if_hwassist &= ~BNX_CSUM_FEATURES;
3890 		}
3891 		if (mask & IFCAP_TSO) {
3892 			ifp->if_capenable ^= (mask & IFCAP_TSO);
3893 			if (ifp->if_capenable & IFCAP_TSO)
3894 				ifp->if_hwassist |= CSUM_TSO;
3895 			else
3896 				ifp->if_hwassist &= ~CSUM_TSO;
3897 		}
3898 		if (mask & IFCAP_RSS)
3899 			ifp->if_capenable ^= IFCAP_RSS;
3900 		break;
3901 	default:
3902 		error = ether_ioctl(ifp, command, data);
3903 		break;
3904 	}
3905 	return error;
3906 }
3907 
3908 static void
3909 bnx_watchdog(struct ifaltq_subque *ifsq)
3910 {
3911 	struct ifnet *ifp = ifsq_get_ifp(ifsq);
3912 	struct bnx_softc *sc = ifp->if_softc;
3913 	int i;
3914 
3915 	ASSERT_IFNET_SERIALIZED_ALL(ifp);
3916 
3917 	if_printf(ifp, "watchdog timeout -- resetting\n");
3918 
3919 	bnx_init(sc);
3920 
3921 	IFNET_STAT_INC(ifp, oerrors, 1);
3922 
3923 	for (i = 0; i < sc->bnx_tx_ringcnt; ++i)
3924 		ifsq_devstart_sched(sc->bnx_tx_ring[i].bnx_ifsq);
3925 }
3926 
3927 /*
3928  * Stop the adapter and free any mbufs allocated to the
3929  * RX and TX lists.
3930  */
3931 static void
3932 bnx_stop(struct bnx_softc *sc)
3933 {
3934 	struct ifnet *ifp = &sc->arpcom.ac_if;
3935 	int i;
3936 
3937 	ASSERT_IFNET_SERIALIZED_ALL(ifp);
3938 
3939 	callout_stop(&sc->bnx_tick_timer);
3940 
3941 	/* Disable host interrupts. */
3942 	bnx_disable_intr(sc);
3943 
3944 	/*
3945 	 * Tell firmware we're shutting down.
3946 	 */
3947 	bnx_sig_pre_reset(sc, BNX_RESET_SHUTDOWN);
3948 
3949 	/*
3950 	 * Disable all of the receiver blocks
3951 	 */
3952 	bnx_stop_block(sc, BGE_RX_MODE, BGE_RXMODE_ENABLE);
3953 	bnx_stop_block(sc, BGE_RBDI_MODE, BGE_RBDIMODE_ENABLE);
3954 	bnx_stop_block(sc, BGE_RXLP_MODE, BGE_RXLPMODE_ENABLE);
3955 	bnx_stop_block(sc, BGE_RDBDI_MODE, BGE_RBDIMODE_ENABLE);
3956 	bnx_stop_block(sc, BGE_RDC_MODE, BGE_RDCMODE_ENABLE);
3957 	bnx_stop_block(sc, BGE_RBDC_MODE, BGE_RBDCMODE_ENABLE);
3958 
3959 	/*
3960 	 * Disable all of the transmit blocks
3961 	 */
3962 	bnx_stop_block(sc, BGE_SRS_MODE, BGE_SRSMODE_ENABLE);
3963 	bnx_stop_block(sc, BGE_SBDI_MODE, BGE_SBDIMODE_ENABLE);
3964 	bnx_stop_block(sc, BGE_SDI_MODE, BGE_SDIMODE_ENABLE);
3965 	bnx_stop_block(sc, BGE_RDMA_MODE, BGE_RDMAMODE_ENABLE);
3966 	bnx_stop_block(sc, BGE_SDC_MODE, BGE_SDCMODE_ENABLE);
3967 	bnx_stop_block(sc, BGE_SBDC_MODE, BGE_SBDCMODE_ENABLE);
3968 
3969 	/*
3970 	 * Shut down all of the memory managers and related
3971 	 * state machines.
3972 	 */
3973 	bnx_stop_block(sc, BGE_HCC_MODE, BGE_HCCMODE_ENABLE);
3974 	bnx_stop_block(sc, BGE_WDMA_MODE, BGE_WDMAMODE_ENABLE);
3975 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0xFFFFFFFF);
3976 	CSR_WRITE_4(sc, BGE_FTQ_RESET, 0);
3977 
3978 	bnx_reset(sc);
3979 	bnx_sig_post_reset(sc, BNX_RESET_SHUTDOWN);
3980 
3981 	/*
3982 	 * Tell firmware we're shutting down.
3983 	 */
3984 	BNX_CLRBIT(sc, BGE_MODE_CTL, BGE_MODECTL_STACKUP);
3985 
3986 	/* Free the RX lists. */
3987 	bnx_free_rx_ring_std(&sc->bnx_rx_std_ring);
3988 
3989 	/* Free jumbo RX list. */
3990 	if (BNX_IS_JUMBO_CAPABLE(sc))
3991 		bnx_free_rx_ring_jumbo(sc);
3992 
3993 	/* Free TX buffers. */
3994 	for (i = 0; i < sc->bnx_tx_ringcnt; ++i) {
3995 		struct bnx_tx_ring *txr = &sc->bnx_tx_ring[i];
3996 
3997 		txr->bnx_saved_status_tag = 0;
3998 		bnx_free_tx_ring(txr);
3999 	}
4000 
4001 	/* Clear saved status tag */
4002 	for (i = 0; i < sc->bnx_rx_retcnt; ++i)
4003 		sc->bnx_rx_ret_ring[i].bnx_saved_status_tag = 0;
4004 
4005 	sc->bnx_link = 0;
4006 	sc->bnx_coal_chg = 0;
4007 
4008 	ifp->if_flags &= ~IFF_RUNNING;
4009 	for (i = 0; i < sc->bnx_tx_ringcnt; ++i) {
4010 		struct bnx_tx_ring *txr = &sc->bnx_tx_ring[i];
4011 
4012 		ifsq_clr_oactive(txr->bnx_ifsq);
4013 		ifsq_watchdog_stop(&txr->bnx_tx_watchdog);
4014 	}
4015 }
4016 
4017 /*
4018  * Stop all chip I/O so that the kernel's probe routines don't
4019  * get confused by errant DMAs when rebooting.
4020  */
4021 static void
4022 bnx_shutdown(device_t dev)
4023 {
4024 	struct bnx_softc *sc = device_get_softc(dev);
4025 	struct ifnet *ifp = &sc->arpcom.ac_if;
4026 
4027 	ifnet_serialize_all(ifp);
4028 	bnx_stop(sc);
4029 	ifnet_deserialize_all(ifp);
4030 }
4031 
4032 static int
4033 bnx_suspend(device_t dev)
4034 {
4035 	struct bnx_softc *sc = device_get_softc(dev);
4036 	struct ifnet *ifp = &sc->arpcom.ac_if;
4037 
4038 	ifnet_serialize_all(ifp);
4039 	bnx_stop(sc);
4040 	ifnet_deserialize_all(ifp);
4041 
4042 	return 0;
4043 }
4044 
4045 static int
4046 bnx_resume(device_t dev)
4047 {
4048 	struct bnx_softc *sc = device_get_softc(dev);
4049 	struct ifnet *ifp = &sc->arpcom.ac_if;
4050 
4051 	ifnet_serialize_all(ifp);
4052 
4053 	if (ifp->if_flags & IFF_UP) {
4054 		int i;
4055 
4056 		bnx_init(sc);
4057 		for (i = 0; i < sc->bnx_tx_ringcnt; ++i)
4058 			ifsq_devstart_sched(sc->bnx_tx_ring[i].bnx_ifsq);
4059 	}
4060 
4061 	ifnet_deserialize_all(ifp);
4062 
4063 	return 0;
4064 }
4065 
4066 static void
4067 bnx_setpromisc(struct bnx_softc *sc)
4068 {
4069 	struct ifnet *ifp = &sc->arpcom.ac_if;
4070 
4071 	if (ifp->if_flags & IFF_PROMISC)
4072 		BNX_SETBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
4073 	else
4074 		BNX_CLRBIT(sc, BGE_RX_MODE, BGE_RXMODE_RX_PROMISC);
4075 }
4076 
4077 static void
4078 bnx_dma_free(struct bnx_softc *sc)
4079 {
4080 	struct bnx_rx_std_ring *std = &sc->bnx_rx_std_ring;
4081 	int i;
4082 
4083 	/* Destroy RX return rings */
4084 	if (sc->bnx_rx_ret_ring != NULL) {
4085 		for (i = 0; i < sc->bnx_rx_retcnt; ++i)
4086 			bnx_destroy_rx_ret_ring(&sc->bnx_rx_ret_ring[i]);
4087 		kfree(sc->bnx_rx_ret_ring, M_DEVBUF);
4088 	}
4089 
4090 	/* Destroy RX mbuf DMA stuffs. */
4091 	if (std->bnx_rx_mtag != NULL) {
4092 		for (i = 0; i < BGE_STD_RX_RING_CNT; i++) {
4093 			KKASSERT(std->bnx_rx_std_buf[i].bnx_rx_mbuf == NULL);
4094 			bus_dmamap_destroy(std->bnx_rx_mtag,
4095 			    std->bnx_rx_std_buf[i].bnx_rx_dmamap);
4096 		}
4097 		bus_dma_tag_destroy(std->bnx_rx_mtag);
4098 	}
4099 
4100 	/* Destroy standard RX ring */
4101 	bnx_dma_block_free(std->bnx_rx_std_ring_tag,
4102 	    std->bnx_rx_std_ring_map, std->bnx_rx_std_ring);
4103 
4104 	/* Destroy TX rings */
4105 	if (sc->bnx_tx_ring != NULL) {
4106 		for (i = 0; i < sc->bnx_tx_ringcnt; ++i)
4107 			bnx_destroy_tx_ring(&sc->bnx_tx_ring[i]);
4108 		kfree(sc->bnx_tx_ring, M_DEVBUF);
4109 	}
4110 
4111 	if (BNX_IS_JUMBO_CAPABLE(sc))
4112 		bnx_free_jumbo_mem(sc);
4113 
4114 	/* Destroy status blocks */
4115 	for (i = 0; i < sc->bnx_intr_cnt; ++i) {
4116 		struct bnx_intr_data *intr = &sc->bnx_intr_data[i];
4117 
4118 		bnx_dma_block_free(intr->bnx_status_tag,
4119 		    intr->bnx_status_map, intr->bnx_status_block);
4120 	}
4121 
4122 	/* Destroy the parent tag */
4123 	if (sc->bnx_cdata.bnx_parent_tag != NULL)
4124 		bus_dma_tag_destroy(sc->bnx_cdata.bnx_parent_tag);
4125 }
4126 
4127 static int
4128 bnx_dma_alloc(device_t dev)
4129 {
4130 	struct bnx_softc *sc = device_get_softc(dev);
4131 	struct bnx_rx_std_ring *std = &sc->bnx_rx_std_ring;
4132 	int i, error, mbx;
4133 
4134 	/*
4135 	 * Allocate the parent bus DMA tag appropriate for PCI.
4136 	 *
4137 	 * All of the NetExtreme/NetLink controllers have 4GB boundary
4138 	 * DMA bug.
4139 	 * Whenever an address crosses a multiple of the 4GB boundary
4140 	 * (including 4GB, 8Gb, 12Gb, etc.) and makes the transition
4141 	 * from 0xX_FFFF_FFFF to 0x(X+1)_0000_0000 an internal DMA
4142 	 * state machine will lockup and cause the device to hang.
4143 	 */
4144 	error = bus_dma_tag_create(NULL, 1, BGE_DMA_BOUNDARY_4G,
4145 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
4146 	    BUS_SPACE_MAXSIZE_32BIT, 0, BUS_SPACE_MAXSIZE_32BIT,
4147 	    0, &sc->bnx_cdata.bnx_parent_tag);
4148 	if (error) {
4149 		device_printf(dev, "could not create parent DMA tag\n");
4150 		return error;
4151 	}
4152 
4153 	/*
4154 	 * Create DMA stuffs for status blocks.
4155 	 */
4156 	for (i = 0; i < sc->bnx_intr_cnt; ++i) {
4157 		struct bnx_intr_data *intr = &sc->bnx_intr_data[i];
4158 
4159 		error = bnx_dma_block_alloc(sc,
4160 		    __VM_CACHELINE_ALIGN(BGE_STATUS_BLK_SZ),
4161 		    &intr->bnx_status_tag, &intr->bnx_status_map,
4162 		    (void *)&intr->bnx_status_block,
4163 		    &intr->bnx_status_block_paddr);
4164 		if (error) {
4165 			device_printf(dev,
4166 			    "could not create %dth status block\n", i);
4167 			return error;
4168 		}
4169 	}
4170 	sc->bnx_hw_status = &sc->bnx_intr_data[0].bnx_status_block->bge_status;
4171 	if (sc->bnx_flags & BNX_FLAG_STATUS_HASTAG) {
4172 		sc->bnx_hw_status_tag =
4173 		    &sc->bnx_intr_data[0].bnx_status_block->bge_status_tag;
4174 	}
4175 
4176 	/*
4177 	 * Create DMA tag and maps for RX mbufs.
4178 	 */
4179 	std->bnx_sc = sc;
4180 	lwkt_serialize_init(&std->bnx_rx_std_serialize);
4181 	error = bus_dma_tag_create(sc->bnx_cdata.bnx_parent_tag, 1, 0,
4182 	    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
4183 	    NULL, NULL, MCLBYTES, 1, MCLBYTES,
4184 	    BUS_DMA_ALLOCNOW | BUS_DMA_WAITOK, &std->bnx_rx_mtag);
4185 	if (error) {
4186 		device_printf(dev, "could not create RX mbuf DMA tag\n");
4187 		return error;
4188 	}
4189 
4190 	for (i = 0; i < BGE_STD_RX_RING_CNT; ++i) {
4191 		error = bus_dmamap_create(std->bnx_rx_mtag, BUS_DMA_WAITOK,
4192 		    &std->bnx_rx_std_buf[i].bnx_rx_dmamap);
4193 		if (error) {
4194 			int j;
4195 
4196 			for (j = 0; j < i; ++j) {
4197 				bus_dmamap_destroy(std->bnx_rx_mtag,
4198 				    std->bnx_rx_std_buf[j].bnx_rx_dmamap);
4199 			}
4200 			bus_dma_tag_destroy(std->bnx_rx_mtag);
4201 			std->bnx_rx_mtag = NULL;
4202 
4203 			device_printf(dev,
4204 			    "could not create %dth RX mbuf DMA map\n", i);
4205 			return error;
4206 		}
4207 	}
4208 
4209 	/*
4210 	 * Create DMA stuffs for standard RX ring.
4211 	 */
4212 	error = bnx_dma_block_alloc(sc, BGE_STD_RX_RING_SZ,
4213 	    &std->bnx_rx_std_ring_tag,
4214 	    &std->bnx_rx_std_ring_map,
4215 	    (void *)&std->bnx_rx_std_ring,
4216 	    &std->bnx_rx_std_ring_paddr);
4217 	if (error) {
4218 		device_printf(dev, "could not create std RX ring\n");
4219 		return error;
4220 	}
4221 
4222 	/*
4223 	 * Create RX return rings
4224 	 */
4225 	mbx = BGE_MBX_RX_CONS0_LO;
4226 	sc->bnx_rx_ret_ring = kmalloc_cachealign(
4227 	    sizeof(struct bnx_rx_ret_ring) * sc->bnx_rx_retcnt, M_DEVBUF,
4228 	    M_WAITOK | M_ZERO);
4229 	for (i = 0; i < sc->bnx_rx_retcnt; ++i) {
4230 		struct bnx_rx_ret_ring *ret = &sc->bnx_rx_ret_ring[i];
4231 		struct bnx_intr_data *intr;
4232 
4233 		ret->bnx_sc = sc;
4234 		ret->bnx_std = std;
4235 		ret->bnx_rx_mbx = mbx;
4236 		ret->bnx_rx_cntmax = (BGE_STD_RX_RING_CNT / 4) /
4237 		    sc->bnx_rx_retcnt;
4238 		ret->bnx_rx_mask = 1 << i;
4239 
4240 		if (!BNX_RSS_ENABLED(sc)) {
4241 			intr = &sc->bnx_intr_data[0];
4242 		} else {
4243 			KKASSERT(i + 1 < sc->bnx_intr_cnt);
4244 			intr = &sc->bnx_intr_data[i + 1];
4245 		}
4246 
4247 		if (i == 0) {
4248 			ret->bnx_rx_considx =
4249 			    &intr->bnx_status_block->bge_idx[0].bge_rx_prod_idx;
4250 		} else if (i == 1) {
4251 			ret->bnx_rx_considx =
4252 			    &intr->bnx_status_block->bge_rx_jumbo_cons_idx;
4253 		} else if (i == 2) {
4254 			ret->bnx_rx_considx =
4255 			    &intr->bnx_status_block->bge_rsvd1;
4256 		} else if (i == 3) {
4257 			ret->bnx_rx_considx =
4258 			    &intr->bnx_status_block->bge_rx_mini_cons_idx;
4259 		} else {
4260 			panic("unknown RX return ring %d\n", i);
4261 		}
4262 		ret->bnx_hw_status_tag =
4263 		    &intr->bnx_status_block->bge_status_tag;
4264 
4265 		error = bnx_create_rx_ret_ring(ret);
4266 		if (error) {
4267 			device_printf(dev,
4268 			    "could not create %dth RX ret ring\n", i);
4269 			return error;
4270 		}
4271 		mbx += 8;
4272 	}
4273 
4274 	/*
4275 	 * Create TX rings
4276 	 */
4277 	sc->bnx_tx_ring = kmalloc_cachealign(
4278 	    sizeof(struct bnx_tx_ring) * sc->bnx_tx_ringcnt, M_DEVBUF,
4279 	    M_WAITOK | M_ZERO);
4280 	for (i = 0; i < sc->bnx_tx_ringcnt; ++i) {
4281 		struct bnx_tx_ring *txr = &sc->bnx_tx_ring[i];
4282 		struct bnx_intr_data *intr;
4283 
4284 		txr->bnx_sc = sc;
4285 		txr->bnx_tx_mbx = bnx_tx_mailbox[i];
4286 
4287 		if (sc->bnx_tx_ringcnt == 1) {
4288 			intr = &sc->bnx_intr_data[0];
4289 		} else {
4290 			KKASSERT(i + 1 < sc->bnx_intr_cnt);
4291 			intr = &sc->bnx_intr_data[i + 1];
4292 		}
4293 
4294 		if ((sc->bnx_flags & BNX_FLAG_RXTX_BUNDLE) == 0) {
4295 			txr->bnx_hw_status_tag =
4296 			    &intr->bnx_status_block->bge_status_tag;
4297 		}
4298 		txr->bnx_tx_considx =
4299 		    &intr->bnx_status_block->bge_idx[0].bge_tx_cons_idx;
4300 
4301 		error = bnx_create_tx_ring(txr);
4302 		if (error) {
4303 			device_printf(dev,
4304 			    "could not create %dth TX ring\n", i);
4305 			return error;
4306 		}
4307 	}
4308 
4309 	/*
4310 	 * Create jumbo buffer pool.
4311 	 */
4312 	if (BNX_IS_JUMBO_CAPABLE(sc)) {
4313 		error = bnx_alloc_jumbo_mem(sc);
4314 		if (error) {
4315 			device_printf(dev,
4316 			    "could not create jumbo buffer pool\n");
4317 			return error;
4318 		}
4319 	}
4320 
4321 	return 0;
4322 }
4323 
4324 static int
4325 bnx_dma_block_alloc(struct bnx_softc *sc, bus_size_t size, bus_dma_tag_t *tag,
4326 		    bus_dmamap_t *map, void **addr, bus_addr_t *paddr)
4327 {
4328 	bus_dmamem_t dmem;
4329 	int error;
4330 
4331 	error = bus_dmamem_coherent(sc->bnx_cdata.bnx_parent_tag, PAGE_SIZE, 0,
4332 				    BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR,
4333 				    size, BUS_DMA_WAITOK | BUS_DMA_ZERO, &dmem);
4334 	if (error)
4335 		return error;
4336 
4337 	*tag = dmem.dmem_tag;
4338 	*map = dmem.dmem_map;
4339 	*addr = dmem.dmem_addr;
4340 	*paddr = dmem.dmem_busaddr;
4341 
4342 	return 0;
4343 }
4344 
4345 static void
4346 bnx_dma_block_free(bus_dma_tag_t tag, bus_dmamap_t map, void *addr)
4347 {
4348 	if (tag != NULL) {
4349 		bus_dmamap_unload(tag, map);
4350 		bus_dmamem_free(tag, addr, map);
4351 		bus_dma_tag_destroy(tag);
4352 	}
4353 }
4354 
4355 static void
4356 bnx_tbi_link_upd(struct bnx_softc *sc, uint32_t status)
4357 {
4358 	struct ifnet *ifp = &sc->arpcom.ac_if;
4359 
4360 #define PCS_ENCODE_ERR	(BGE_MACSTAT_PORT_DECODE_ERROR|BGE_MACSTAT_MI_COMPLETE)
4361 
4362 	/*
4363 	 * Sometimes PCS encoding errors are detected in
4364 	 * TBI mode (on fiber NICs), and for some reason
4365 	 * the chip will signal them as link changes.
4366 	 * If we get a link change event, but the 'PCS
4367 	 * encoding error' bit in the MAC status register
4368 	 * is set, don't bother doing a link check.
4369 	 * This avoids spurious "gigabit link up" messages
4370 	 * that sometimes appear on fiber NICs during
4371 	 * periods of heavy traffic.
4372 	 */
4373 	if (status & BGE_MACSTAT_TBI_PCS_SYNCHED) {
4374 		if (!sc->bnx_link) {
4375 			sc->bnx_link++;
4376 			if (sc->bnx_asicrev == BGE_ASICREV_BCM5704) {
4377 				BNX_CLRBIT(sc, BGE_MAC_MODE,
4378 				    BGE_MACMODE_TBI_SEND_CFGS);
4379 				DELAY(40);
4380 			}
4381 			CSR_WRITE_4(sc, BGE_MAC_STS, 0xFFFFFFFF);
4382 
4383 			if (bootverbose)
4384 				if_printf(ifp, "link UP\n");
4385 
4386 			ifp->if_link_state = LINK_STATE_UP;
4387 			if_link_state_change(ifp);
4388 		}
4389 	} else if ((status & PCS_ENCODE_ERR) != PCS_ENCODE_ERR) {
4390 		if (sc->bnx_link) {
4391 			sc->bnx_link = 0;
4392 
4393 			if (bootverbose)
4394 				if_printf(ifp, "link DOWN\n");
4395 
4396 			ifp->if_link_state = LINK_STATE_DOWN;
4397 			if_link_state_change(ifp);
4398 		}
4399 	}
4400 
4401 #undef PCS_ENCODE_ERR
4402 
4403 	/* Clear the attention. */
4404 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
4405 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
4406 	    BGE_MACSTAT_LINK_CHANGED);
4407 }
4408 
4409 static void
4410 bnx_copper_link_upd(struct bnx_softc *sc, uint32_t status __unused)
4411 {
4412 	struct ifnet *ifp = &sc->arpcom.ac_if;
4413 	struct mii_data *mii = device_get_softc(sc->bnx_miibus);
4414 
4415 	mii_pollstat(mii);
4416 	bnx_miibus_statchg(sc->bnx_dev);
4417 
4418 	if (bootverbose) {
4419 		if (sc->bnx_link)
4420 			if_printf(ifp, "link UP\n");
4421 		else
4422 			if_printf(ifp, "link DOWN\n");
4423 	}
4424 
4425 	/* Clear the attention. */
4426 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
4427 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
4428 	    BGE_MACSTAT_LINK_CHANGED);
4429 }
4430 
4431 static void
4432 bnx_autopoll_link_upd(struct bnx_softc *sc, uint32_t status __unused)
4433 {
4434 	struct ifnet *ifp = &sc->arpcom.ac_if;
4435 	struct mii_data *mii = device_get_softc(sc->bnx_miibus);
4436 
4437 	mii_pollstat(mii);
4438 
4439 	if (!sc->bnx_link &&
4440 	    (mii->mii_media_status & IFM_ACTIVE) &&
4441 	    IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE) {
4442 		sc->bnx_link++;
4443 		if (bootverbose)
4444 			if_printf(ifp, "link UP\n");
4445 	} else if (sc->bnx_link &&
4446 	    (!(mii->mii_media_status & IFM_ACTIVE) ||
4447 	    IFM_SUBTYPE(mii->mii_media_active) == IFM_NONE)) {
4448 		sc->bnx_link = 0;
4449 		if (bootverbose)
4450 			if_printf(ifp, "link DOWN\n");
4451 	}
4452 
4453 	/* Clear the attention. */
4454 	CSR_WRITE_4(sc, BGE_MAC_STS, BGE_MACSTAT_SYNC_CHANGED |
4455 	    BGE_MACSTAT_CFG_CHANGED | BGE_MACSTAT_MI_COMPLETE |
4456 	    BGE_MACSTAT_LINK_CHANGED);
4457 }
4458 
4459 static int
4460 bnx_sysctl_rx_coal_ticks(SYSCTL_HANDLER_ARGS)
4461 {
4462 	struct bnx_softc *sc = arg1;
4463 
4464 	return bnx_sysctl_coal_chg(oidp, arg1, arg2, req,
4465 	    &sc->bnx_rx_coal_ticks,
4466 	    BNX_RX_COAL_TICKS_MIN, BNX_RX_COAL_TICKS_MAX,
4467 	    BNX_RX_COAL_TICKS_CHG);
4468 }
4469 
4470 static int
4471 bnx_sysctl_tx_coal_ticks(SYSCTL_HANDLER_ARGS)
4472 {
4473 	struct bnx_softc *sc = arg1;
4474 
4475 	return bnx_sysctl_coal_chg(oidp, arg1, arg2, req,
4476 	    &sc->bnx_tx_coal_ticks,
4477 	    BNX_TX_COAL_TICKS_MIN, BNX_TX_COAL_TICKS_MAX,
4478 	    BNX_TX_COAL_TICKS_CHG);
4479 }
4480 
4481 static int
4482 bnx_sysctl_rx_coal_bds(SYSCTL_HANDLER_ARGS)
4483 {
4484 	struct bnx_softc *sc = arg1;
4485 
4486 	return bnx_sysctl_coal_chg(oidp, arg1, arg2, req,
4487 	    &sc->bnx_rx_coal_bds,
4488 	    BNX_RX_COAL_BDS_MIN, BNX_RX_COAL_BDS_MAX,
4489 	    BNX_RX_COAL_BDS_CHG);
4490 }
4491 
4492 static int
4493 bnx_sysctl_rx_coal_bds_poll(SYSCTL_HANDLER_ARGS)
4494 {
4495 	struct bnx_softc *sc = arg1;
4496 
4497 	return bnx_sysctl_coal_chg(oidp, arg1, arg2, req,
4498 	    &sc->bnx_rx_coal_bds_poll,
4499 	    BNX_RX_COAL_BDS_MIN, BNX_RX_COAL_BDS_MAX,
4500 	    BNX_RX_COAL_BDS_CHG);
4501 }
4502 
4503 static int
4504 bnx_sysctl_tx_coal_bds(SYSCTL_HANDLER_ARGS)
4505 {
4506 	struct bnx_softc *sc = arg1;
4507 
4508 	return bnx_sysctl_coal_chg(oidp, arg1, arg2, req,
4509 	    &sc->bnx_tx_coal_bds,
4510 	    BNX_TX_COAL_BDS_MIN, BNX_TX_COAL_BDS_MAX,
4511 	    BNX_TX_COAL_BDS_CHG);
4512 }
4513 
4514 static int
4515 bnx_sysctl_tx_coal_bds_poll(SYSCTL_HANDLER_ARGS)
4516 {
4517 	struct bnx_softc *sc = arg1;
4518 
4519 	return bnx_sysctl_coal_chg(oidp, arg1, arg2, req,
4520 	    &sc->bnx_tx_coal_bds_poll,
4521 	    BNX_TX_COAL_BDS_MIN, BNX_TX_COAL_BDS_MAX,
4522 	    BNX_TX_COAL_BDS_CHG);
4523 }
4524 
4525 static int
4526 bnx_sysctl_rx_coal_bds_int(SYSCTL_HANDLER_ARGS)
4527 {
4528 	struct bnx_softc *sc = arg1;
4529 
4530 	return bnx_sysctl_coal_chg(oidp, arg1, arg2, req,
4531 	    &sc->bnx_rx_coal_bds_int,
4532 	    BNX_RX_COAL_BDS_MIN, BNX_RX_COAL_BDS_MAX,
4533 	    BNX_RX_COAL_BDS_INT_CHG);
4534 }
4535 
4536 static int
4537 bnx_sysctl_tx_coal_bds_int(SYSCTL_HANDLER_ARGS)
4538 {
4539 	struct bnx_softc *sc = arg1;
4540 
4541 	return bnx_sysctl_coal_chg(oidp, arg1, arg2, req,
4542 	    &sc->bnx_tx_coal_bds_int,
4543 	    BNX_TX_COAL_BDS_MIN, BNX_TX_COAL_BDS_MAX,
4544 	    BNX_TX_COAL_BDS_INT_CHG);
4545 }
4546 
4547 static int
4548 bnx_sysctl_coal_chg(SYSCTL_HANDLER_ARGS, uint32_t *coal,
4549     int coal_min, int coal_max, uint32_t coal_chg_mask)
4550 {
4551 	struct bnx_softc *sc = arg1;
4552 	struct ifnet *ifp = &sc->arpcom.ac_if;
4553 	int error = 0, v;
4554 
4555 	ifnet_serialize_all(ifp);
4556 
4557 	v = *coal;
4558 	error = sysctl_handle_int(oidp, &v, 0, req);
4559 	if (!error && req->newptr != NULL) {
4560 		if (v < coal_min || v > coal_max) {
4561 			error = EINVAL;
4562 		} else {
4563 			*coal = v;
4564 			sc->bnx_coal_chg |= coal_chg_mask;
4565 
4566 			/* Commit changes */
4567 			bnx_coal_change(sc);
4568 		}
4569 	}
4570 
4571 	ifnet_deserialize_all(ifp);
4572 	return error;
4573 }
4574 
4575 static void
4576 bnx_coal_change(struct bnx_softc *sc)
4577 {
4578 	struct ifnet *ifp = &sc->arpcom.ac_if;
4579 	int i;
4580 
4581 	ASSERT_IFNET_SERIALIZED_ALL(ifp);
4582 
4583 	if (sc->bnx_coal_chg & BNX_RX_COAL_TICKS_CHG) {
4584 		if (sc->bnx_rx_retcnt == 1) {
4585 			CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS,
4586 			    sc->bnx_rx_coal_ticks);
4587 			i = 0;
4588 		} else {
4589 			CSR_WRITE_4(sc, BGE_HCC_RX_COAL_TICKS, 0);
4590 			for (i = 0; i < sc->bnx_rx_retcnt; ++i) {
4591 				CSR_WRITE_4(sc, BGE_VEC1_RX_COAL_TICKS +
4592 				    (i * BGE_VEC_COALSET_SIZE),
4593 				    sc->bnx_rx_coal_ticks);
4594 			}
4595 		}
4596 		for (; i < BNX_INTR_MAX - 1; ++i) {
4597 			CSR_WRITE_4(sc, BGE_VEC1_RX_COAL_TICKS +
4598 			    (i * BGE_VEC_COALSET_SIZE), 0);
4599 		}
4600 		if (bootverbose) {
4601 			if_printf(ifp, "rx_coal_ticks -> %u\n",
4602 			    sc->bnx_rx_coal_ticks);
4603 		}
4604 	}
4605 
4606 	if (sc->bnx_coal_chg & BNX_TX_COAL_TICKS_CHG) {
4607 		if (sc->bnx_tx_ringcnt == 1) {
4608 			CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS,
4609 			    sc->bnx_tx_coal_ticks);
4610 			i = 0;
4611 		} else {
4612 			CSR_WRITE_4(sc, BGE_HCC_TX_COAL_TICKS, 0);
4613 			for (i = 0; i < sc->bnx_tx_ringcnt; ++i) {
4614 				CSR_WRITE_4(sc, BGE_VEC1_TX_COAL_TICKS +
4615 				    (i * BGE_VEC_COALSET_SIZE),
4616 				    sc->bnx_tx_coal_ticks);
4617 			}
4618 		}
4619 		for (; i < BNX_INTR_MAX - 1; ++i) {
4620 			CSR_WRITE_4(sc, BGE_VEC1_TX_COAL_TICKS +
4621 			    (i * BGE_VEC_COALSET_SIZE), 0);
4622 		}
4623 		if (bootverbose) {
4624 			if_printf(ifp, "tx_coal_ticks -> %u\n",
4625 			    sc->bnx_tx_coal_ticks);
4626 		}
4627 	}
4628 
4629 	if (sc->bnx_coal_chg & BNX_RX_COAL_BDS_CHG) {
4630 		uint32_t rx_coal_bds;
4631 
4632 		if (ifp->if_flags & IFF_NPOLLING)
4633 			rx_coal_bds = sc->bnx_rx_coal_bds_poll;
4634 		else
4635 			rx_coal_bds = sc->bnx_rx_coal_bds;
4636 
4637 		if (sc->bnx_rx_retcnt == 1) {
4638 			CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS, rx_coal_bds);
4639 			i = 0;
4640 		} else {
4641 			CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS, 0);
4642 			for (i = 0; i < sc->bnx_rx_retcnt; ++i) {
4643 				CSR_WRITE_4(sc, BGE_VEC1_RX_MAX_COAL_BDS +
4644 				    (i * BGE_VEC_COALSET_SIZE), rx_coal_bds);
4645 			}
4646 		}
4647 		for (; i < BNX_INTR_MAX - 1; ++i) {
4648 			CSR_WRITE_4(sc, BGE_VEC1_RX_MAX_COAL_BDS +
4649 			    (i * BGE_VEC_COALSET_SIZE), 0);
4650 		}
4651 		if (bootverbose) {
4652 			if_printf(ifp, "%srx_coal_bds -> %u\n",
4653 			    (ifp->if_flags & IFF_NPOLLING) ? "polling " : "",
4654 			    rx_coal_bds);
4655 		}
4656 	}
4657 
4658 	if (sc->bnx_coal_chg & BNX_TX_COAL_BDS_CHG) {
4659 		uint32_t tx_coal_bds;
4660 
4661 		if (ifp->if_flags & IFF_NPOLLING)
4662 			tx_coal_bds = sc->bnx_tx_coal_bds_poll;
4663 		else
4664 			tx_coal_bds = sc->bnx_tx_coal_bds;
4665 
4666 		if (sc->bnx_tx_ringcnt == 1) {
4667 			CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS, tx_coal_bds);
4668 			i = 0;
4669 		} else {
4670 			CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS, 0);
4671 			for (i = 0; i < sc->bnx_tx_ringcnt; ++i) {
4672 				CSR_WRITE_4(sc, BGE_VEC1_TX_MAX_COAL_BDS +
4673 				    (i * BGE_VEC_COALSET_SIZE), tx_coal_bds);
4674 			}
4675 		}
4676 		for (; i < BNX_INTR_MAX - 1; ++i) {
4677 			CSR_WRITE_4(sc, BGE_VEC1_TX_MAX_COAL_BDS +
4678 			    (i * BGE_VEC_COALSET_SIZE), 0);
4679 		}
4680 		if (bootverbose) {
4681 			if_printf(ifp, "%stx_coal_bds -> %u\n",
4682 			    (ifp->if_flags & IFF_NPOLLING) ? "polling " : "",
4683 			    tx_coal_bds);
4684 		}
4685 	}
4686 
4687 	if (sc->bnx_coal_chg & BNX_RX_COAL_BDS_INT_CHG) {
4688 		if (sc->bnx_rx_retcnt == 1) {
4689 			CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT,
4690 			    sc->bnx_rx_coal_bds_int);
4691 			i = 0;
4692 		} else {
4693 			CSR_WRITE_4(sc, BGE_HCC_RX_MAX_COAL_BDS_INT, 0);
4694 			for (i = 0; i < sc->bnx_rx_retcnt; ++i) {
4695 				CSR_WRITE_4(sc, BGE_VEC1_RX_MAX_COAL_BDS_INT +
4696 				    (i * BGE_VEC_COALSET_SIZE),
4697 				    sc->bnx_rx_coal_bds_int);
4698 			}
4699 		}
4700 		for (; i < BNX_INTR_MAX - 1; ++i) {
4701 			CSR_WRITE_4(sc, BGE_VEC1_RX_MAX_COAL_BDS_INT +
4702 			    (i * BGE_VEC_COALSET_SIZE), 0);
4703 		}
4704 		if (bootverbose) {
4705 			if_printf(ifp, "rx_coal_bds_int -> %u\n",
4706 			    sc->bnx_rx_coal_bds_int);
4707 		}
4708 	}
4709 
4710 	if (sc->bnx_coal_chg & BNX_TX_COAL_BDS_INT_CHG) {
4711 		if (sc->bnx_tx_ringcnt == 1) {
4712 			CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT,
4713 			    sc->bnx_tx_coal_bds_int);
4714 			i = 0;
4715 		} else {
4716 			CSR_WRITE_4(sc, BGE_HCC_TX_MAX_COAL_BDS_INT, 0);
4717 			for (i = 0; i < sc->bnx_tx_ringcnt; ++i) {
4718 				CSR_WRITE_4(sc, BGE_VEC1_TX_MAX_COAL_BDS_INT +
4719 				    (i * BGE_VEC_COALSET_SIZE),
4720 				    sc->bnx_tx_coal_bds_int);
4721 			}
4722 		}
4723 		for (; i < BNX_INTR_MAX - 1; ++i) {
4724 			CSR_WRITE_4(sc, BGE_VEC1_TX_MAX_COAL_BDS_INT +
4725 			    (i * BGE_VEC_COALSET_SIZE), 0);
4726 		}
4727 		if (bootverbose) {
4728 			if_printf(ifp, "tx_coal_bds_int -> %u\n",
4729 			    sc->bnx_tx_coal_bds_int);
4730 		}
4731 	}
4732 
4733 	sc->bnx_coal_chg = 0;
4734 }
4735 
4736 static void
4737 bnx_check_intr_rxtx(void *xintr)
4738 {
4739 	struct bnx_intr_data *intr = xintr;
4740 	struct bnx_rx_ret_ring *ret;
4741 	struct bnx_tx_ring *txr;
4742 	struct ifnet *ifp;
4743 
4744 	lwkt_serialize_enter(intr->bnx_intr_serialize);
4745 
4746 	KKASSERT(mycpuid == intr->bnx_intr_cpuid);
4747 
4748 	ifp = &intr->bnx_sc->arpcom.ac_if;
4749 	if ((ifp->if_flags & (IFF_RUNNING | IFF_NPOLLING)) != IFF_RUNNING) {
4750 		lwkt_serialize_exit(intr->bnx_intr_serialize);
4751 		return;
4752 	}
4753 
4754 	txr = intr->bnx_txr;
4755 	ret = intr->bnx_ret;
4756 
4757 	if (*ret->bnx_rx_considx != ret->bnx_rx_saved_considx ||
4758 	    *txr->bnx_tx_considx != txr->bnx_tx_saved_considx) {
4759 		if (intr->bnx_rx_check_considx == ret->bnx_rx_saved_considx &&
4760 		    intr->bnx_tx_check_considx == txr->bnx_tx_saved_considx) {
4761 			if (!intr->bnx_intr_maylose) {
4762 				intr->bnx_intr_maylose = TRUE;
4763 				goto done;
4764 			}
4765 			if (bootverbose)
4766 				if_printf(ifp, "lost interrupt\n");
4767 			intr->bnx_intr_func(intr->bnx_intr_arg);
4768 		}
4769 	}
4770 	intr->bnx_intr_maylose = FALSE;
4771 	intr->bnx_rx_check_considx = ret->bnx_rx_saved_considx;
4772 	intr->bnx_tx_check_considx = txr->bnx_tx_saved_considx;
4773 
4774 done:
4775 	callout_reset(&intr->bnx_intr_timer, BNX_INTR_CKINTVL,
4776 	    intr->bnx_intr_check, intr);
4777 	lwkt_serialize_exit(intr->bnx_intr_serialize);
4778 }
4779 
4780 static void
4781 bnx_check_intr_tx(void *xintr)
4782 {
4783 	struct bnx_intr_data *intr = xintr;
4784 	struct bnx_tx_ring *txr;
4785 	struct ifnet *ifp;
4786 
4787 	lwkt_serialize_enter(intr->bnx_intr_serialize);
4788 
4789 	KKASSERT(mycpuid == intr->bnx_intr_cpuid);
4790 
4791 	ifp = &intr->bnx_sc->arpcom.ac_if;
4792 	if ((ifp->if_flags & (IFF_RUNNING | IFF_NPOLLING)) != IFF_RUNNING) {
4793 		lwkt_serialize_exit(intr->bnx_intr_serialize);
4794 		return;
4795 	}
4796 
4797 	txr = intr->bnx_txr;
4798 
4799 	if (*txr->bnx_tx_considx != txr->bnx_tx_saved_considx) {
4800 		if (intr->bnx_tx_check_considx == txr->bnx_tx_saved_considx) {
4801 			if (!intr->bnx_intr_maylose) {
4802 				intr->bnx_intr_maylose = TRUE;
4803 				goto done;
4804 			}
4805 			if (bootverbose)
4806 				if_printf(ifp, "lost interrupt\n");
4807 			intr->bnx_intr_func(intr->bnx_intr_arg);
4808 		}
4809 	}
4810 	intr->bnx_intr_maylose = FALSE;
4811 	intr->bnx_tx_check_considx = txr->bnx_tx_saved_considx;
4812 
4813 done:
4814 	callout_reset(&intr->bnx_intr_timer, BNX_INTR_CKINTVL,
4815 	    intr->bnx_intr_check, intr);
4816 	lwkt_serialize_exit(intr->bnx_intr_serialize);
4817 }
4818 
4819 static void
4820 bnx_check_intr_rx(void *xintr)
4821 {
4822 	struct bnx_intr_data *intr = xintr;
4823 	struct bnx_rx_ret_ring *ret;
4824 	struct ifnet *ifp;
4825 
4826 	lwkt_serialize_enter(intr->bnx_intr_serialize);
4827 
4828 	KKASSERT(mycpuid == intr->bnx_intr_cpuid);
4829 
4830 	ifp = &intr->bnx_sc->arpcom.ac_if;
4831 	if ((ifp->if_flags & (IFF_RUNNING | IFF_NPOLLING)) != IFF_RUNNING) {
4832 		lwkt_serialize_exit(intr->bnx_intr_serialize);
4833 		return;
4834 	}
4835 
4836 	ret = intr->bnx_ret;
4837 
4838 	if (*ret->bnx_rx_considx != ret->bnx_rx_saved_considx) {
4839 		if (intr->bnx_rx_check_considx == ret->bnx_rx_saved_considx) {
4840 			if (!intr->bnx_intr_maylose) {
4841 				intr->bnx_intr_maylose = TRUE;
4842 				goto done;
4843 			}
4844 			if (bootverbose)
4845 				if_printf(ifp, "lost interrupt\n");
4846 			intr->bnx_intr_func(intr->bnx_intr_arg);
4847 		}
4848 	}
4849 	intr->bnx_intr_maylose = FALSE;
4850 	intr->bnx_rx_check_considx = ret->bnx_rx_saved_considx;
4851 
4852 done:
4853 	callout_reset(&intr->bnx_intr_timer, BNX_INTR_CKINTVL,
4854 	    intr->bnx_intr_check, intr);
4855 	lwkt_serialize_exit(intr->bnx_intr_serialize);
4856 }
4857 
4858 static void
4859 bnx_enable_intr(struct bnx_softc *sc)
4860 {
4861 	struct ifnet *ifp = &sc->arpcom.ac_if;
4862 	int i;
4863 
4864 	for (i = 0; i < sc->bnx_intr_cnt; ++i) {
4865 		lwkt_serialize_handler_enable(
4866 		    sc->bnx_intr_data[i].bnx_intr_serialize);
4867 	}
4868 
4869 	/*
4870 	 * Enable interrupt.
4871 	 */
4872 	for (i = 0; i < sc->bnx_intr_cnt; ++i) {
4873 		struct bnx_intr_data *intr = &sc->bnx_intr_data[i];
4874 
4875 		bnx_writembx(sc, intr->bnx_intr_mbx,
4876 		    (*intr->bnx_saved_status_tag) << 24);
4877 		/* XXX Linux driver */
4878 		bnx_writembx(sc, intr->bnx_intr_mbx,
4879 		    (*intr->bnx_saved_status_tag) << 24);
4880 	}
4881 
4882 	/*
4883 	 * Unmask the interrupt when we stop polling.
4884 	 */
4885 	PCI_CLRBIT(sc->bnx_dev, BGE_PCI_MISC_CTL,
4886 	    BGE_PCIMISCCTL_MASK_PCI_INTR, 4);
4887 
4888 	/*
4889 	 * Trigger another interrupt, since above writing
4890 	 * to interrupt mailbox0 may acknowledge pending
4891 	 * interrupt.
4892 	 */
4893 	BNX_SETBIT(sc, BGE_MISC_LOCAL_CTL, BGE_MLC_INTR_SET);
4894 
4895 	if (sc->bnx_flags & BNX_FLAG_STATUSTAG_BUG) {
4896 		if (bootverbose)
4897 			if_printf(ifp, "status tag bug workaround\n");
4898 
4899 		for (i = 0; i < sc->bnx_intr_cnt; ++i) {
4900 			struct bnx_intr_data *intr = &sc->bnx_intr_data[i];
4901 
4902 			if (intr->bnx_intr_check == NULL)
4903 				continue;
4904 			intr->bnx_intr_maylose = FALSE;
4905 			intr->bnx_rx_check_considx = 0;
4906 			intr->bnx_tx_check_considx = 0;
4907 			callout_reset_bycpu(&intr->bnx_intr_timer,
4908 			    BNX_INTR_CKINTVL, intr->bnx_intr_check, intr,
4909 			    intr->bnx_intr_cpuid);
4910 		}
4911 	}
4912 }
4913 
4914 static void
4915 bnx_disable_intr(struct bnx_softc *sc)
4916 {
4917 	int i;
4918 
4919 	for (i = 0; i < sc->bnx_intr_cnt; ++i) {
4920 		struct bnx_intr_data *intr = &sc->bnx_intr_data[i];
4921 
4922 		callout_stop(&intr->bnx_intr_timer);
4923 		intr->bnx_intr_maylose = FALSE;
4924 		intr->bnx_rx_check_considx = 0;
4925 		intr->bnx_tx_check_considx = 0;
4926 	}
4927 
4928 	/*
4929 	 * Mask the interrupt when we start polling.
4930 	 */
4931 	PCI_SETBIT(sc->bnx_dev, BGE_PCI_MISC_CTL,
4932 	    BGE_PCIMISCCTL_MASK_PCI_INTR, 4);
4933 
4934 	/*
4935 	 * Acknowledge possible asserted interrupt.
4936 	 */
4937 	for (i = 0; i < BNX_INTR_MAX; ++i)
4938 		bnx_writembx(sc, sc->bnx_intr_data[i].bnx_intr_mbx, 1);
4939 
4940 	for (i = 0; i < sc->bnx_intr_cnt; ++i) {
4941 		lwkt_serialize_handler_disable(
4942 		    sc->bnx_intr_data[i].bnx_intr_serialize);
4943 	}
4944 }
4945 
4946 static int
4947 bnx_get_eaddr_mem(struct bnx_softc *sc, uint8_t ether_addr[])
4948 {
4949 	uint32_t mac_addr;
4950 	int ret = 1;
4951 
4952 	mac_addr = bnx_readmem_ind(sc, 0x0c14);
4953 	if ((mac_addr >> 16) == 0x484b) {
4954 		ether_addr[0] = (uint8_t)(mac_addr >> 8);
4955 		ether_addr[1] = (uint8_t)mac_addr;
4956 		mac_addr = bnx_readmem_ind(sc, 0x0c18);
4957 		ether_addr[2] = (uint8_t)(mac_addr >> 24);
4958 		ether_addr[3] = (uint8_t)(mac_addr >> 16);
4959 		ether_addr[4] = (uint8_t)(mac_addr >> 8);
4960 		ether_addr[5] = (uint8_t)mac_addr;
4961 		ret = 0;
4962 	}
4963 	return ret;
4964 }
4965 
4966 static int
4967 bnx_get_eaddr_nvram(struct bnx_softc *sc, uint8_t ether_addr[])
4968 {
4969 	int mac_offset = BGE_EE_MAC_OFFSET;
4970 
4971 	if (BNX_IS_5717_PLUS(sc)) {
4972 		int f;
4973 
4974 		f = pci_get_function(sc->bnx_dev);
4975 		if (f & 1)
4976 			mac_offset = BGE_EE_MAC_OFFSET_5717;
4977 		if (f > 1)
4978 			mac_offset += BGE_EE_MAC_OFFSET_5717_OFF;
4979 	}
4980 
4981 	return bnx_read_nvram(sc, ether_addr, mac_offset + 2, ETHER_ADDR_LEN);
4982 }
4983 
4984 static int
4985 bnx_get_eaddr_eeprom(struct bnx_softc *sc, uint8_t ether_addr[])
4986 {
4987 	if (sc->bnx_flags & BNX_FLAG_NO_EEPROM)
4988 		return 1;
4989 
4990 	return bnx_read_eeprom(sc, ether_addr, BGE_EE_MAC_OFFSET + 2,
4991 			       ETHER_ADDR_LEN);
4992 }
4993 
4994 static int
4995 bnx_get_eaddr(struct bnx_softc *sc, uint8_t eaddr[])
4996 {
4997 	static const bnx_eaddr_fcn_t bnx_eaddr_funcs[] = {
4998 		/* NOTE: Order is critical */
4999 		bnx_get_eaddr_mem,
5000 		bnx_get_eaddr_nvram,
5001 		bnx_get_eaddr_eeprom,
5002 		NULL
5003 	};
5004 	const bnx_eaddr_fcn_t *func;
5005 
5006 	for (func = bnx_eaddr_funcs; *func != NULL; ++func) {
5007 		if ((*func)(sc, eaddr) == 0)
5008 			break;
5009 	}
5010 	return (*func == NULL ? ENXIO : 0);
5011 }
5012 
5013 /*
5014  * NOTE: 'm' is not freed upon failure
5015  */
5016 struct mbuf *
5017 bnx_defrag_shortdma(struct mbuf *m)
5018 {
5019 	struct mbuf *n;
5020 	int found;
5021 
5022 	/*
5023 	 * If device receive two back-to-back send BDs with less than
5024 	 * or equal to 8 total bytes then the device may hang.  The two
5025 	 * back-to-back send BDs must in the same frame for this failure
5026 	 * to occur.  Scan mbuf chains and see whether two back-to-back
5027 	 * send BDs are there.  If this is the case, allocate new mbuf
5028 	 * and copy the frame to workaround the silicon bug.
5029 	 */
5030 	for (n = m, found = 0; n != NULL; n = n->m_next) {
5031 		if (n->m_len < 8) {
5032 			found++;
5033 			if (found > 1)
5034 				break;
5035 			continue;
5036 		}
5037 		found = 0;
5038 	}
5039 
5040 	if (found > 1)
5041 		n = m_defrag(m, MB_DONTWAIT);
5042 	else
5043 		n = m;
5044 	return n;
5045 }
5046 
5047 static void
5048 bnx_stop_block(struct bnx_softc *sc, bus_size_t reg, uint32_t bit)
5049 {
5050 	int i;
5051 
5052 	BNX_CLRBIT(sc, reg, bit);
5053 	for (i = 0; i < BNX_TIMEOUT; i++) {
5054 		if ((CSR_READ_4(sc, reg) & bit) == 0)
5055 			return;
5056 		DELAY(100);
5057 	}
5058 }
5059 
5060 static void
5061 bnx_link_poll(struct bnx_softc *sc)
5062 {
5063 	uint32_t status;
5064 
5065 	status = CSR_READ_4(sc, BGE_MAC_STS);
5066 	if ((status & sc->bnx_link_chg) || sc->bnx_link_evt) {
5067 		sc->bnx_link_evt = 0;
5068 		sc->bnx_link_upd(sc, status);
5069 	}
5070 }
5071 
5072 static void
5073 bnx_enable_msi(struct bnx_softc *sc, boolean_t is_msix)
5074 {
5075 	uint32_t msi_mode;
5076 
5077 	msi_mode = CSR_READ_4(sc, BGE_MSI_MODE);
5078 	msi_mode |= BGE_MSIMODE_ENABLE;
5079 	/*
5080 	 * NOTE:
5081 	 * 5718-PG105-R says that "one shot" mode does not work
5082 	 * if MSI is used, however, it obviously works.
5083 	 */
5084 	msi_mode &= ~BGE_MSIMODE_ONESHOT_DISABLE;
5085 	if (is_msix)
5086 		msi_mode |= BGE_MSIMODE_MSIX_MULTIMODE;
5087 	else
5088 		msi_mode &= ~BGE_MSIMODE_MSIX_MULTIMODE;
5089 	CSR_WRITE_4(sc, BGE_MSI_MODE, msi_mode);
5090 }
5091 
5092 static uint32_t
5093 bnx_dma_swap_options(struct bnx_softc *sc)
5094 {
5095 	uint32_t dma_options;
5096 
5097 	dma_options = BGE_MODECTL_WORDSWAP_NONFRAME |
5098 	    BGE_MODECTL_BYTESWAP_DATA | BGE_MODECTL_WORDSWAP_DATA;
5099 #if BYTE_ORDER == BIG_ENDIAN
5100 	dma_options |= BGE_MODECTL_BYTESWAP_NONFRAME;
5101 #endif
5102 	return dma_options;
5103 }
5104 
5105 static int
5106 bnx_setup_tso(struct bnx_tx_ring *txr, struct mbuf **mp,
5107     uint16_t *mss0, uint16_t *flags0)
5108 {
5109 	struct mbuf *m;
5110 	struct ip *ip;
5111 	struct tcphdr *th;
5112 	int thoff, iphlen, hoff, hlen;
5113 	uint16_t flags, mss;
5114 
5115 	m = *mp;
5116 	KASSERT(M_WRITABLE(m), ("TSO mbuf not writable"));
5117 
5118 	hoff = m->m_pkthdr.csum_lhlen;
5119 	iphlen = m->m_pkthdr.csum_iphlen;
5120 	thoff = m->m_pkthdr.csum_thlen;
5121 
5122 	KASSERT(hoff > 0, ("invalid ether header len"));
5123 	KASSERT(iphlen > 0, ("invalid ip header len"));
5124 	KASSERT(thoff > 0, ("invalid tcp header len"));
5125 
5126 	if (__predict_false(m->m_len < hoff + iphlen + thoff)) {
5127 		m = m_pullup(m, hoff + iphlen + thoff);
5128 		if (m == NULL) {
5129 			*mp = NULL;
5130 			return ENOBUFS;
5131 		}
5132 		*mp = m;
5133 	}
5134 	ip = mtodoff(m, struct ip *, hoff);
5135 	th = mtodoff(m, struct tcphdr *, hoff + iphlen);
5136 
5137 	mss = m->m_pkthdr.tso_segsz;
5138 	flags = BGE_TXBDFLAG_CPU_PRE_DMA | BGE_TXBDFLAG_CPU_POST_DMA;
5139 
5140 	ip->ip_len = htons(mss + iphlen + thoff);
5141 	th->th_sum = 0;
5142 
5143 	hlen = (iphlen + thoff) >> 2;
5144 	mss |= ((hlen & 0x3) << 14);
5145 	flags |= ((hlen & 0xf8) << 7) | ((hlen & 0x4) << 2);
5146 
5147 	*mss0 = mss;
5148 	*flags0 = flags;
5149 
5150 	return 0;
5151 }
5152 
5153 static int
5154 bnx_create_tx_ring(struct bnx_tx_ring *txr)
5155 {
5156 	bus_size_t txmaxsz, txmaxsegsz;
5157 	int i, error;
5158 
5159 	lwkt_serialize_init(&txr->bnx_tx_serialize);
5160 
5161 	/*
5162 	 * Create DMA tag and maps for TX mbufs.
5163 	 */
5164 	if (txr->bnx_sc->bnx_flags & BNX_FLAG_TSO)
5165 		txmaxsz = IP_MAXPACKET + sizeof(struct ether_vlan_header);
5166 	else
5167 		txmaxsz = BNX_JUMBO_FRAMELEN;
5168 	if (txr->bnx_sc->bnx_asicrev == BGE_ASICREV_BCM57766)
5169 		txmaxsegsz = MCLBYTES;
5170 	else
5171 		txmaxsegsz = PAGE_SIZE;
5172 	error = bus_dma_tag_create(txr->bnx_sc->bnx_cdata.bnx_parent_tag,
5173 	    1, 0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL,
5174 	    txmaxsz, BNX_NSEG_NEW, txmaxsegsz,
5175 	    BUS_DMA_ALLOCNOW | BUS_DMA_WAITOK | BUS_DMA_ONEBPAGE,
5176 	    &txr->bnx_tx_mtag);
5177 	if (error) {
5178 		device_printf(txr->bnx_sc->bnx_dev,
5179 		    "could not create TX mbuf DMA tag\n");
5180 		return error;
5181 	}
5182 
5183 	for (i = 0; i < BGE_TX_RING_CNT; i++) {
5184 		error = bus_dmamap_create(txr->bnx_tx_mtag,
5185 		    BUS_DMA_WAITOK | BUS_DMA_ONEBPAGE,
5186 		    &txr->bnx_tx_buf[i].bnx_tx_dmamap);
5187 		if (error) {
5188 			int j;
5189 
5190 			for (j = 0; j < i; ++j) {
5191 				bus_dmamap_destroy(txr->bnx_tx_mtag,
5192 				    txr->bnx_tx_buf[j].bnx_tx_dmamap);
5193 			}
5194 			bus_dma_tag_destroy(txr->bnx_tx_mtag);
5195 			txr->bnx_tx_mtag = NULL;
5196 
5197 			device_printf(txr->bnx_sc->bnx_dev,
5198 			    "could not create TX mbuf DMA map\n");
5199 			return error;
5200 		}
5201 	}
5202 
5203 	/*
5204 	 * Create DMA stuffs for TX ring.
5205 	 */
5206 	error = bnx_dma_block_alloc(txr->bnx_sc, BGE_TX_RING_SZ,
5207 	    &txr->bnx_tx_ring_tag,
5208 	    &txr->bnx_tx_ring_map,
5209 	    (void *)&txr->bnx_tx_ring,
5210 	    &txr->bnx_tx_ring_paddr);
5211 	if (error) {
5212 		device_printf(txr->bnx_sc->bnx_dev,
5213 		    "could not create TX ring\n");
5214 		return error;
5215 	}
5216 
5217 	txr->bnx_tx_flags |= BNX_TX_FLAG_SHORTDMA;
5218 	txr->bnx_tx_wreg = BNX_TX_WREG_NSEGS;
5219 
5220 	return 0;
5221 }
5222 
5223 static void
5224 bnx_destroy_tx_ring(struct bnx_tx_ring *txr)
5225 {
5226 	/* Destroy TX mbuf DMA stuffs. */
5227 	if (txr->bnx_tx_mtag != NULL) {
5228 		int i;
5229 
5230 		for (i = 0; i < BGE_TX_RING_CNT; i++) {
5231 			KKASSERT(txr->bnx_tx_buf[i].bnx_tx_mbuf == NULL);
5232 			bus_dmamap_destroy(txr->bnx_tx_mtag,
5233 			    txr->bnx_tx_buf[i].bnx_tx_dmamap);
5234 		}
5235 		bus_dma_tag_destroy(txr->bnx_tx_mtag);
5236 	}
5237 
5238 	/* Destroy TX ring */
5239 	bnx_dma_block_free(txr->bnx_tx_ring_tag,
5240 	    txr->bnx_tx_ring_map, txr->bnx_tx_ring);
5241 }
5242 
5243 static int
5244 bnx_sysctl_force_defrag(SYSCTL_HANDLER_ARGS)
5245 {
5246 	struct bnx_softc *sc = (void *)arg1;
5247 	struct ifnet *ifp = &sc->arpcom.ac_if;
5248 	struct bnx_tx_ring *txr = &sc->bnx_tx_ring[0];
5249 	int error, defrag, i;
5250 
5251 	if (txr->bnx_tx_flags & BNX_TX_FLAG_FORCE_DEFRAG)
5252 		defrag = 1;
5253 	else
5254 		defrag = 0;
5255 
5256 	error = sysctl_handle_int(oidp, &defrag, 0, req);
5257 	if (error || req->newptr == NULL)
5258 		return error;
5259 
5260 	ifnet_serialize_all(ifp);
5261 	for (i = 0; i < sc->bnx_tx_ringcnt; ++i) {
5262 		txr = &sc->bnx_tx_ring[i];
5263 		if (defrag)
5264 			txr->bnx_tx_flags |= BNX_TX_FLAG_FORCE_DEFRAG;
5265 		else
5266 			txr->bnx_tx_flags &= ~BNX_TX_FLAG_FORCE_DEFRAG;
5267 	}
5268 	ifnet_deserialize_all(ifp);
5269 
5270 	return 0;
5271 }
5272 
5273 static int
5274 bnx_sysctl_tx_wreg(SYSCTL_HANDLER_ARGS)
5275 {
5276 	struct bnx_softc *sc = (void *)arg1;
5277 	struct ifnet *ifp = &sc->arpcom.ac_if;
5278 	struct bnx_tx_ring *txr = &sc->bnx_tx_ring[0];
5279 	int error, tx_wreg, i;
5280 
5281 	tx_wreg = txr->bnx_tx_wreg;
5282 	error = sysctl_handle_int(oidp, &tx_wreg, 0, req);
5283 	if (error || req->newptr == NULL)
5284 		return error;
5285 
5286 	ifnet_serialize_all(ifp);
5287 	for (i = 0; i < sc->bnx_tx_ringcnt; ++i)
5288 		sc->bnx_tx_ring[i].bnx_tx_wreg = tx_wreg;
5289 	ifnet_deserialize_all(ifp);
5290 
5291 	return 0;
5292 }
5293 
5294 static int
5295 bnx_create_rx_ret_ring(struct bnx_rx_ret_ring *ret)
5296 {
5297 	int error;
5298 
5299 	lwkt_serialize_init(&ret->bnx_rx_ret_serialize);
5300 
5301 	/*
5302 	 * Create DMA stuffs for RX return ring.
5303 	 */
5304 	error = bnx_dma_block_alloc(ret->bnx_sc,
5305 	    BGE_RX_RTN_RING_SZ(BNX_RETURN_RING_CNT),
5306 	    &ret->bnx_rx_ret_ring_tag,
5307 	    &ret->bnx_rx_ret_ring_map,
5308 	    (void *)&ret->bnx_rx_ret_ring,
5309 	    &ret->bnx_rx_ret_ring_paddr);
5310 	if (error) {
5311 		device_printf(ret->bnx_sc->bnx_dev,
5312 		    "could not create RX ret ring\n");
5313 		return error;
5314 	}
5315 
5316 	/* Shadow standard ring's RX mbuf DMA tag */
5317 	ret->bnx_rx_mtag = ret->bnx_std->bnx_rx_mtag;
5318 
5319 	/*
5320 	 * Create tmp DMA map for RX mbufs.
5321 	 */
5322 	error = bus_dmamap_create(ret->bnx_rx_mtag, BUS_DMA_WAITOK,
5323 	    &ret->bnx_rx_tmpmap);
5324 	if (error) {
5325 		device_printf(ret->bnx_sc->bnx_dev,
5326 		    "could not create tmp RX mbuf DMA map\n");
5327 		ret->bnx_rx_mtag = NULL;
5328 		return error;
5329 	}
5330 	return 0;
5331 }
5332 
5333 static void
5334 bnx_destroy_rx_ret_ring(struct bnx_rx_ret_ring *ret)
5335 {
5336 	/* Destroy tmp RX mbuf DMA map */
5337 	if (ret->bnx_rx_mtag != NULL)
5338 		bus_dmamap_destroy(ret->bnx_rx_mtag, ret->bnx_rx_tmpmap);
5339 
5340 	/* Destroy RX return ring */
5341 	bnx_dma_block_free(ret->bnx_rx_ret_ring_tag,
5342 	    ret->bnx_rx_ret_ring_map, ret->bnx_rx_ret_ring);
5343 }
5344 
5345 static int
5346 bnx_alloc_intr(struct bnx_softc *sc)
5347 {
5348 	struct bnx_intr_data *intr;
5349 	u_int intr_flags;
5350 	int error;
5351 
5352 	if (sc->bnx_intr_cnt > 1) {
5353 		error = bnx_alloc_msix(sc);
5354 		if (error)
5355 			return error;
5356 		KKASSERT(sc->bnx_intr_type == PCI_INTR_TYPE_MSIX);
5357 		return 0;
5358 	}
5359 
5360 	KKASSERT(sc->bnx_intr_cnt == 1);
5361 
5362 	intr = &sc->bnx_intr_data[0];
5363 	intr->bnx_ret = &sc->bnx_rx_ret_ring[0];
5364 	intr->bnx_txr = &sc->bnx_tx_ring[0];
5365 	intr->bnx_intr_serialize = &sc->bnx_main_serialize;
5366 	intr->bnx_intr_check = bnx_check_intr_rxtx;
5367 	intr->bnx_saved_status_tag = &intr->bnx_ret->bnx_saved_status_tag;
5368 
5369 	sc->bnx_intr_type = pci_alloc_1intr(sc->bnx_dev, bnx_msi_enable,
5370 	    &intr->bnx_intr_rid, &intr_flags);
5371 
5372 	intr->bnx_intr_res = bus_alloc_resource_any(sc->bnx_dev, SYS_RES_IRQ,
5373 	    &intr->bnx_intr_rid, intr_flags);
5374 	if (intr->bnx_intr_res == NULL) {
5375 		device_printf(sc->bnx_dev, "could not alloc interrupt\n");
5376 		return ENXIO;
5377 	}
5378 
5379 	if (sc->bnx_intr_type == PCI_INTR_TYPE_MSI) {
5380 		bnx_enable_msi(sc, FALSE);
5381 		intr->bnx_intr_func = bnx_msi;
5382 		if (bootverbose)
5383 			device_printf(sc->bnx_dev, "oneshot MSI\n");
5384 	} else {
5385 		intr->bnx_intr_func = bnx_intr_legacy;
5386 	}
5387 	intr->bnx_intr_arg = sc;
5388 	intr->bnx_intr_cpuid = rman_get_cpuid(intr->bnx_intr_res);
5389 
5390 	intr->bnx_txr->bnx_tx_cpuid = intr->bnx_intr_cpuid;
5391 
5392 	return 0;
5393 }
5394 
5395 static int
5396 bnx_setup_intr(struct bnx_softc *sc)
5397 {
5398 	int error, i;
5399 
5400 	for (i = 0; i < sc->bnx_intr_cnt; ++i) {
5401 		struct bnx_intr_data *intr = &sc->bnx_intr_data[i];
5402 
5403 		error = bus_setup_intr_descr(sc->bnx_dev, intr->bnx_intr_res,
5404 		    INTR_MPSAFE, intr->bnx_intr_func, intr->bnx_intr_arg,
5405 		    &intr->bnx_intr_hand, intr->bnx_intr_serialize,
5406 		    intr->bnx_intr_desc);
5407 		if (error) {
5408 			device_printf(sc->bnx_dev,
5409 			    "could not set up %dth intr\n", i);
5410 			bnx_teardown_intr(sc, i);
5411 			return error;
5412 		}
5413 	}
5414 	return 0;
5415 }
5416 
5417 static void
5418 bnx_teardown_intr(struct bnx_softc *sc, int cnt)
5419 {
5420 	int i;
5421 
5422 	for (i = 0; i < cnt; ++i) {
5423 		struct bnx_intr_data *intr = &sc->bnx_intr_data[i];
5424 
5425 		bus_teardown_intr(sc->bnx_dev, intr->bnx_intr_res,
5426 		    intr->bnx_intr_hand);
5427 	}
5428 }
5429 
5430 static void
5431 bnx_free_intr(struct bnx_softc *sc)
5432 {
5433 	if (sc->bnx_intr_type != PCI_INTR_TYPE_MSIX) {
5434 		struct bnx_intr_data *intr;
5435 
5436 		KKASSERT(sc->bnx_intr_cnt <= 1);
5437 		intr = &sc->bnx_intr_data[0];
5438 
5439 		if (intr->bnx_intr_res != NULL) {
5440 			bus_release_resource(sc->bnx_dev, SYS_RES_IRQ,
5441 			    intr->bnx_intr_rid, intr->bnx_intr_res);
5442 		}
5443 		if (sc->bnx_intr_type == PCI_INTR_TYPE_MSI)
5444 			pci_release_msi(sc->bnx_dev);
5445 	} else {
5446 		bnx_free_msix(sc, TRUE);
5447 	}
5448 }
5449 
5450 static void
5451 bnx_setup_serialize(struct bnx_softc *sc)
5452 {
5453 	int i, j;
5454 
5455 	/*
5456 	 * Allocate serializer array
5457 	 */
5458 
5459 	/* Main + RX STD + TX + RX RET */
5460 	sc->bnx_serialize_cnt = 1 + 1 + sc->bnx_tx_ringcnt + sc->bnx_rx_retcnt;
5461 
5462 	sc->bnx_serialize =
5463 	    kmalloc(sc->bnx_serialize_cnt * sizeof(struct lwkt_serialize *),
5464 	        M_DEVBUF, M_WAITOK | M_ZERO);
5465 
5466 	/*
5467 	 * Setup serializers
5468 	 *
5469 	 * NOTE: Order is critical
5470 	 */
5471 
5472 	i = 0;
5473 
5474 	KKASSERT(i < sc->bnx_serialize_cnt);
5475 	sc->bnx_serialize[i++] = &sc->bnx_main_serialize;
5476 
5477 	KKASSERT(i < sc->bnx_serialize_cnt);
5478 	sc->bnx_serialize[i++] = &sc->bnx_rx_std_ring.bnx_rx_std_serialize;
5479 
5480 	for (j = 0; j < sc->bnx_rx_retcnt; ++j) {
5481 		KKASSERT(i < sc->bnx_serialize_cnt);
5482 		sc->bnx_serialize[i++] =
5483 		    &sc->bnx_rx_ret_ring[j].bnx_rx_ret_serialize;
5484 	}
5485 
5486 	for (j = 0; j < sc->bnx_tx_ringcnt; ++j) {
5487 		KKASSERT(i < sc->bnx_serialize_cnt);
5488 		sc->bnx_serialize[i++] =
5489 		    &sc->bnx_tx_ring[j].bnx_tx_serialize;
5490 	}
5491 
5492 	KKASSERT(i == sc->bnx_serialize_cnt);
5493 }
5494 
5495 static void
5496 bnx_serialize(struct ifnet *ifp, enum ifnet_serialize slz)
5497 {
5498 	struct bnx_softc *sc = ifp->if_softc;
5499 
5500 	ifnet_serialize_array_enter(sc->bnx_serialize,
5501 	    sc->bnx_serialize_cnt, slz);
5502 }
5503 
5504 static void
5505 bnx_deserialize(struct ifnet *ifp, enum ifnet_serialize slz)
5506 {
5507 	struct bnx_softc *sc = ifp->if_softc;
5508 
5509 	ifnet_serialize_array_exit(sc->bnx_serialize,
5510 	    sc->bnx_serialize_cnt, slz);
5511 }
5512 
5513 static int
5514 bnx_tryserialize(struct ifnet *ifp, enum ifnet_serialize slz)
5515 {
5516 	struct bnx_softc *sc = ifp->if_softc;
5517 
5518 	return ifnet_serialize_array_try(sc->bnx_serialize,
5519 	    sc->bnx_serialize_cnt, slz);
5520 }
5521 
5522 #ifdef INVARIANTS
5523 
5524 static void
5525 bnx_serialize_assert(struct ifnet *ifp, enum ifnet_serialize slz,
5526     boolean_t serialized)
5527 {
5528 	struct bnx_softc *sc = ifp->if_softc;
5529 
5530 	ifnet_serialize_array_assert(sc->bnx_serialize, sc->bnx_serialize_cnt,
5531 	    slz, serialized);
5532 }
5533 
5534 #endif	/* INVARIANTS */
5535 
5536 #ifdef IFPOLL_ENABLE
5537 
5538 static int
5539 bnx_sysctl_npoll_offset(SYSCTL_HANDLER_ARGS)
5540 {
5541 	struct bnx_softc *sc = (void *)arg1;
5542 	struct ifnet *ifp = &sc->arpcom.ac_if;
5543 	int error, off;
5544 
5545 	off = sc->bnx_npoll_rxoff;
5546 	error = sysctl_handle_int(oidp, &off, 0, req);
5547 	if (error || req->newptr == NULL)
5548 		return error;
5549 	if (off < 0)
5550 		return EINVAL;
5551 
5552 	ifnet_serialize_all(ifp);
5553 	if (off >= ncpus2 || off % sc->bnx_rx_retcnt != 0) {
5554 		error = EINVAL;
5555 	} else {
5556 		error = 0;
5557 		sc->bnx_npoll_txoff = off;
5558 		sc->bnx_npoll_rxoff = off;
5559 	}
5560 	ifnet_deserialize_all(ifp);
5561 
5562 	return error;
5563 }
5564 
5565 static int
5566 bnx_sysctl_npoll_rxoff(SYSCTL_HANDLER_ARGS)
5567 {
5568 	struct bnx_softc *sc = (void *)arg1;
5569 	struct ifnet *ifp = &sc->arpcom.ac_if;
5570 	int error, off;
5571 
5572 	off = sc->bnx_npoll_rxoff;
5573 	error = sysctl_handle_int(oidp, &off, 0, req);
5574 	if (error || req->newptr == NULL)
5575 		return error;
5576 	if (off < 0)
5577 		return EINVAL;
5578 
5579 	ifnet_serialize_all(ifp);
5580 	if (off >= ncpus2 || off % sc->bnx_rx_retcnt != 0) {
5581 		error = EINVAL;
5582 	} else {
5583 		error = 0;
5584 		sc->bnx_npoll_rxoff = off;
5585 	}
5586 	ifnet_deserialize_all(ifp);
5587 
5588 	return error;
5589 }
5590 
5591 static int
5592 bnx_sysctl_npoll_txoff(SYSCTL_HANDLER_ARGS)
5593 {
5594 	struct bnx_softc *sc = (void *)arg1;
5595 	struct ifnet *ifp = &sc->arpcom.ac_if;
5596 	int error, off;
5597 
5598 	off = sc->bnx_npoll_txoff;
5599 	error = sysctl_handle_int(oidp, &off, 0, req);
5600 	if (error || req->newptr == NULL)
5601 		return error;
5602 	if (off < 0)
5603 		return EINVAL;
5604 
5605 	ifnet_serialize_all(ifp);
5606 	if (off >= ncpus2) {
5607 		error = EINVAL;
5608 	} else {
5609 		error = 0;
5610 		sc->bnx_npoll_txoff = off;
5611 	}
5612 	ifnet_deserialize_all(ifp);
5613 
5614 	return error;
5615 }
5616 
5617 #endif	/* IFPOLL_ENABLE */
5618 
5619 static void
5620 bnx_set_tick_cpuid(struct bnx_softc *sc, boolean_t polling)
5621 {
5622 	if (polling)
5623 		sc->bnx_tick_cpuid = 0; /* XXX */
5624 	else
5625 		sc->bnx_tick_cpuid = sc->bnx_intr_data[0].bnx_intr_cpuid;
5626 }
5627 
5628 static void
5629 bnx_rx_std_refill_ithread(void *xstd)
5630 {
5631 	struct bnx_rx_std_ring *std = xstd;
5632 	struct globaldata *gd = mycpu;
5633 
5634 	crit_enter_gd(gd);
5635 
5636 	while (!std->bnx_rx_std_stop) {
5637 		if (std->bnx_rx_std_refill) {
5638 			lwkt_serialize_handler_call(
5639 			    &std->bnx_rx_std_serialize,
5640 			    bnx_rx_std_refill, std, NULL);
5641 		}
5642 
5643 		crit_exit_gd(gd);
5644 		crit_enter_gd(gd);
5645 
5646 		atomic_poll_release_int(&std->bnx_rx_std_running);
5647 		cpu_mfence();
5648 
5649 		if (!std->bnx_rx_std_refill && !std->bnx_rx_std_stop) {
5650 			lwkt_deschedule_self(gd->gd_curthread);
5651 			lwkt_switch();
5652 		}
5653 	}
5654 
5655 	crit_exit_gd(gd);
5656 
5657 	wakeup(std);
5658 
5659 	lwkt_exit();
5660 }
5661 
5662 static void
5663 bnx_rx_std_refill(void *xstd, void *frame __unused)
5664 {
5665 	struct bnx_rx_std_ring *std = xstd;
5666 	int cnt, refill_mask;
5667 
5668 again:
5669 	cnt = 0;
5670 
5671 	cpu_lfence();
5672 	refill_mask = std->bnx_rx_std_refill;
5673 	atomic_clear_int(&std->bnx_rx_std_refill, refill_mask);
5674 
5675 	while (refill_mask) {
5676 		uint16_t check_idx = std->bnx_rx_std;
5677 		int ret_idx;
5678 
5679 		ret_idx = bsfl(refill_mask);
5680 		for (;;) {
5681 			struct bnx_rx_buf *rb;
5682 			int refilled;
5683 
5684 			BNX_INC(check_idx, BGE_STD_RX_RING_CNT);
5685 			rb = &std->bnx_rx_std_buf[check_idx];
5686 			refilled = rb->bnx_rx_refilled;
5687 			cpu_lfence();
5688 			if (refilled) {
5689 				bnx_setup_rxdesc_std(std, check_idx);
5690 				std->bnx_rx_std = check_idx;
5691 				++cnt;
5692 				if (cnt >= 8) {
5693 					atomic_subtract_int(
5694 					    &std->bnx_rx_std_used, cnt);
5695 					bnx_writembx(std->bnx_sc,
5696 					    BGE_MBX_RX_STD_PROD_LO,
5697 					    std->bnx_rx_std);
5698 					cnt = 0;
5699 				}
5700 			} else {
5701 				break;
5702 			}
5703 		}
5704 		refill_mask &= ~(1 << ret_idx);
5705 	}
5706 
5707 	if (cnt) {
5708 		atomic_subtract_int(&std->bnx_rx_std_used, cnt);
5709 		bnx_writembx(std->bnx_sc, BGE_MBX_RX_STD_PROD_LO,
5710 		    std->bnx_rx_std);
5711 	}
5712 
5713 	if (std->bnx_rx_std_refill)
5714 		goto again;
5715 
5716 	atomic_poll_release_int(&std->bnx_rx_std_running);
5717 	cpu_mfence();
5718 
5719 	if (std->bnx_rx_std_refill)
5720 		goto again;
5721 }
5722 
5723 static int
5724 bnx_sysctl_std_refill(SYSCTL_HANDLER_ARGS)
5725 {
5726 	struct bnx_softc *sc = (void *)arg1;
5727 	struct ifnet *ifp = &sc->arpcom.ac_if;
5728 	struct bnx_rx_ret_ring *ret = &sc->bnx_rx_ret_ring[0];
5729 	int error, cntmax, i;
5730 
5731 	cntmax = ret->bnx_rx_cntmax;
5732 	error = sysctl_handle_int(oidp, &cntmax, 0, req);
5733 	if (error || req->newptr == NULL)
5734 		return error;
5735 
5736 	ifnet_serialize_all(ifp);
5737 
5738 	if ((cntmax * sc->bnx_rx_retcnt) >= BGE_STD_RX_RING_CNT / 2) {
5739 		error = EINVAL;
5740 		goto back;
5741 	}
5742 
5743 	for (i = 0; i < sc->bnx_tx_ringcnt; ++i)
5744 		sc->bnx_rx_ret_ring[i].bnx_rx_cntmax = cntmax;
5745 	error = 0;
5746 
5747 back:
5748 	ifnet_deserialize_all(ifp);
5749 
5750 	return error;
5751 }
5752 
5753 static void
5754 bnx_init_rss(struct bnx_softc *sc)
5755 {
5756 	uint8_t key[BGE_RSS_KEYREG_CNT * BGE_RSS_KEYREG_SIZE];
5757 	int i, j, r;
5758 
5759 	KKASSERT(BNX_RSS_ENABLED(sc));
5760 
5761 	/*
5762 	 * Configure RSS redirect table in following fashion:
5763 	 * (hash & ring_cnt_mask) == rdr_table[(hash & rdr_table_mask)]
5764 	 */
5765 	r = 0;
5766 	for (j = 0; j < BGE_RSS_INDIR_TBL_CNT; ++j) {
5767 		uint32_t tbl = 0;
5768 
5769 		for (i = 0; i < BGE_RSS_INDIR_TBLENT_CNT; ++i) {
5770 			uint32_t q;
5771 
5772 			q = r % sc->bnx_rx_retcnt;
5773 			tbl |= q << (BGE_RSS_INDIR_TBLENT_SHIFT *
5774 			    (BGE_RSS_INDIR_TBLENT_CNT - i - 1));
5775 			++r;
5776 		}
5777 
5778 		BNX_RSS_DPRINTF(sc, 1, "tbl%d %08x\n", j, tbl);
5779 		CSR_WRITE_4(sc, BGE_RSS_INDIR_TBL(j), tbl);
5780 	}
5781 
5782 	toeplitz_get_key(key, sizeof(key));
5783 	for (i = 0; i < BGE_RSS_KEYREG_CNT; ++i) {
5784 		uint32_t keyreg;
5785 
5786 		keyreg = BGE_RSS_KEYREG_VAL(key, i);
5787 
5788 		BNX_RSS_DPRINTF(sc, 1, "key%d %08x\n", i, keyreg);
5789 		CSR_WRITE_4(sc, BGE_RSS_KEYREG(i), keyreg);
5790 	}
5791 }
5792 
5793 static void
5794 bnx_setup_ring_cnt(struct bnx_softc *sc)
5795 {
5796 	int msix_enable, i, msix_cnt, msix_cnt2, ring_max;
5797 
5798 	sc->bnx_tx_ringcnt = 1;
5799 	sc->bnx_rx_retcnt = 1;
5800 	sc->bnx_intr_cnt = 1;
5801 
5802 	msix_enable = device_getenv_int(sc->bnx_dev, "msix.enable",
5803 	    bnx_msix_enable);
5804 	if (!msix_enable)
5805 		return;
5806 
5807 	if (ncpus2 == 1)
5808 		return;
5809 
5810 	msix_cnt = pci_msix_count(sc->bnx_dev);
5811 	if (msix_cnt <= 1)
5812 		return;
5813 
5814 	i = 0;
5815 	while ((1 << (i + 1)) <= msix_cnt)
5816 		++i;
5817 	msix_cnt2 = 1 << i;
5818 
5819 	/*
5820 	 * One MSI-X vector is dedicated to status or single TX queue,
5821 	 * so make sure that there are enough MSI-X vectors.
5822 	 */
5823 	if (msix_cnt == msix_cnt2) {
5824 		/*
5825 		 * XXX
5826 		 * This probably will not happen; 57785/5718 families
5827 		 * come with at least 5 MSI-X vectors.
5828 		 */
5829 		msix_cnt2 >>= 1;
5830 		if (msix_cnt2 <= 1) {
5831 			device_printf(sc->bnx_dev,
5832 			    "MSI-X count %d could not be used\n", msix_cnt);
5833 			return;
5834 		}
5835 		device_printf(sc->bnx_dev, "MSI-X count %d is power of 2\n",
5836 		    msix_cnt);
5837 	}
5838 
5839 	/*
5840 	 * Setup RX ring count
5841 	 */
5842 	ring_max = BNX_RX_RING_MAX;
5843 	if (ring_max > msix_cnt2)
5844 		ring_max = msix_cnt2;
5845 	sc->bnx_rx_retcnt = device_getenv_int(sc->bnx_dev, "rx_rings",
5846 	    bnx_rx_rings);
5847 	sc->bnx_rx_retcnt = if_ring_count2(sc->bnx_rx_retcnt, ring_max);
5848 
5849 	if (sc->bnx_rx_retcnt == 1)
5850 		return;
5851 
5852 	/*
5853 	 * We need one extra MSI-X vector for link status or
5854 	 * TX ring (if only one TX ring is enabled).
5855 	 */
5856 	sc->bnx_intr_cnt = sc->bnx_rx_retcnt + 1;
5857 
5858 	/*
5859 	 * Setup TX ring count
5860 	 *
5861 	 * Currently only BCM5719 and BCM5720 support multiple TX rings
5862 	 * and the TX ring count must be less than the RX ring count.
5863 	 */
5864 	if (sc->bnx_asicrev == BGE_ASICREV_BCM5719 ||
5865 	    sc->bnx_asicrev == BGE_ASICREV_BCM5720) {
5866 		ring_max = BNX_TX_RING_MAX;
5867 		if (ring_max > msix_cnt2)
5868 			ring_max = msix_cnt2;
5869 		if (ring_max > sc->bnx_rx_retcnt)
5870 			ring_max = sc->bnx_rx_retcnt;
5871 		sc->bnx_tx_ringcnt = device_getenv_int(sc->bnx_dev, "tx_rings",
5872 		    bnx_tx_rings);
5873 		sc->bnx_tx_ringcnt = if_ring_count2(sc->bnx_tx_ringcnt,
5874 		    ring_max);
5875 	}
5876 }
5877 
5878 static int
5879 bnx_alloc_msix(struct bnx_softc *sc)
5880 {
5881 	struct bnx_intr_data *intr;
5882 	boolean_t setup = FALSE;
5883 	int error, i, offset, offset_def;
5884 
5885 	KKASSERT(sc->bnx_intr_cnt > 1);
5886 	KKASSERT(sc->bnx_intr_cnt == sc->bnx_rx_retcnt + 1);
5887 
5888 	if (sc->bnx_flags & BNX_FLAG_RXTX_BUNDLE) {
5889 		/*
5890 		 * Link status
5891 		 */
5892 		intr = &sc->bnx_intr_data[0];
5893 
5894 		intr->bnx_intr_serialize = &sc->bnx_main_serialize;
5895 		intr->bnx_saved_status_tag = &sc->bnx_saved_status_tag;
5896 
5897 		intr->bnx_intr_func = bnx_msix_status;
5898 		intr->bnx_intr_arg = sc;
5899 		intr->bnx_intr_cpuid = 0; /* XXX */
5900 
5901 		ksnprintf(intr->bnx_intr_desc0, sizeof(intr->bnx_intr_desc0),
5902 		    "%s sts", device_get_nameunit(sc->bnx_dev));
5903 		intr->bnx_intr_desc = intr->bnx_intr_desc0;
5904 
5905 		/*
5906 		 * RX/TX rings
5907 		 */
5908 		if (sc->bnx_rx_retcnt == ncpus2) {
5909 			offset = 0;
5910 		} else {
5911 			offset_def = (sc->bnx_rx_retcnt *
5912 			    device_get_unit(sc->bnx_dev)) % ncpus2;
5913 
5914 			offset = device_getenv_int(sc->bnx_dev,
5915 			    "msix.offset", offset_def);
5916 			if (offset >= ncpus2 ||
5917 			    offset % sc->bnx_rx_retcnt != 0) {
5918 				device_printf(sc->bnx_dev,
5919 				    "invalid msix.offset %d, use %d\n",
5920 				    offset, offset_def);
5921 				offset = offset_def;
5922 			}
5923 		}
5924 
5925 		for (i = 1; i < sc->bnx_intr_cnt; ++i) {
5926 			int idx = i - 1;
5927 
5928 			intr = &sc->bnx_intr_data[i];
5929 
5930 			KKASSERT(idx < sc->bnx_rx_retcnt);
5931 			intr->bnx_ret = &sc->bnx_rx_ret_ring[idx];
5932 			if (idx < sc->bnx_tx_ringcnt) {
5933 				intr->bnx_txr = &sc->bnx_tx_ring[idx];
5934 				intr->bnx_ret->bnx_txr = intr->bnx_txr;
5935 			}
5936 
5937 			intr->bnx_intr_serialize =
5938 			    &intr->bnx_ret->bnx_rx_ret_serialize;
5939 			intr->bnx_saved_status_tag =
5940 			    &intr->bnx_ret->bnx_saved_status_tag;
5941 
5942 			intr->bnx_intr_arg = intr->bnx_ret;
5943 			KKASSERT(idx + offset < ncpus2);
5944 			intr->bnx_intr_cpuid = idx + offset;
5945 
5946 			if (intr->bnx_txr == NULL) {
5947 				intr->bnx_intr_check = bnx_check_intr_rx;
5948 				intr->bnx_intr_func = bnx_msix_rx;
5949 				ksnprintf(intr->bnx_intr_desc0,
5950 				    sizeof(intr->bnx_intr_desc0), "%s rx%d",
5951 				    device_get_nameunit(sc->bnx_dev), idx);
5952 			} else {
5953 				intr->bnx_intr_check = bnx_check_intr_rxtx;
5954 				intr->bnx_intr_func = bnx_msix_rxtx;
5955 				ksnprintf(intr->bnx_intr_desc0,
5956 				    sizeof(intr->bnx_intr_desc0), "%s rxtx%d",
5957 				    device_get_nameunit(sc->bnx_dev), idx);
5958 
5959 				intr->bnx_txr->bnx_tx_cpuid =
5960 				    intr->bnx_intr_cpuid;
5961 			}
5962 			intr->bnx_intr_desc = intr->bnx_intr_desc0;
5963 
5964 			intr->bnx_ret->bnx_msix_mbx = intr->bnx_intr_mbx;
5965 		}
5966 	} else {
5967 		/*
5968 		 * TX ring and link status
5969 		 */
5970 		offset_def = device_get_unit(sc->bnx_dev) % ncpus2;
5971 		offset = device_getenv_int(sc->bnx_dev, "msix.txoff",
5972 		    offset_def);
5973 		if (offset >= ncpus2) {
5974 			device_printf(sc->bnx_dev,
5975 			    "invalid msix.txoff %d, use %d\n",
5976 			    offset, offset_def);
5977 			offset = offset_def;
5978 		}
5979 
5980 		intr = &sc->bnx_intr_data[0];
5981 
5982 		intr->bnx_txr = &sc->bnx_tx_ring[0];
5983 		intr->bnx_intr_serialize = &sc->bnx_main_serialize;
5984 		intr->bnx_intr_check = bnx_check_intr_tx;
5985 		intr->bnx_saved_status_tag =
5986 		    &intr->bnx_txr->bnx_saved_status_tag;
5987 
5988 		intr->bnx_intr_func = bnx_msix_tx_status;
5989 		intr->bnx_intr_arg = intr->bnx_txr;
5990 		intr->bnx_intr_cpuid = offset;
5991 
5992 		ksnprintf(intr->bnx_intr_desc0, sizeof(intr->bnx_intr_desc0),
5993 		    "%s ststx", device_get_nameunit(sc->bnx_dev));
5994 		intr->bnx_intr_desc = intr->bnx_intr_desc0;
5995 
5996 		intr->bnx_txr->bnx_tx_cpuid = intr->bnx_intr_cpuid;
5997 
5998 		/*
5999 		 * RX rings
6000 		 */
6001 		if (sc->bnx_rx_retcnt == ncpus2) {
6002 			offset = 0;
6003 		} else {
6004 			offset_def = (sc->bnx_rx_retcnt *
6005 			    device_get_unit(sc->bnx_dev)) % ncpus2;
6006 
6007 			offset = device_getenv_int(sc->bnx_dev,
6008 			    "msix.rxoff", offset_def);
6009 			if (offset >= ncpus2 ||
6010 			    offset % sc->bnx_rx_retcnt != 0) {
6011 				device_printf(sc->bnx_dev,
6012 				    "invalid msix.rxoff %d, use %d\n",
6013 				    offset, offset_def);
6014 				offset = offset_def;
6015 			}
6016 		}
6017 
6018 		for (i = 1; i < sc->bnx_intr_cnt; ++i) {
6019 			int idx = i - 1;
6020 
6021 			intr = &sc->bnx_intr_data[i];
6022 
6023 			KKASSERT(idx < sc->bnx_rx_retcnt);
6024 			intr->bnx_ret = &sc->bnx_rx_ret_ring[idx];
6025 			intr->bnx_intr_serialize =
6026 			    &intr->bnx_ret->bnx_rx_ret_serialize;
6027 			intr->bnx_intr_check = bnx_check_intr_rx;
6028 			intr->bnx_saved_status_tag =
6029 			    &intr->bnx_ret->bnx_saved_status_tag;
6030 
6031 			intr->bnx_intr_func = bnx_msix_rx;
6032 			intr->bnx_intr_arg = intr->bnx_ret;
6033 			KKASSERT(idx + offset < ncpus2);
6034 			intr->bnx_intr_cpuid = idx + offset;
6035 
6036 			ksnprintf(intr->bnx_intr_desc0,
6037 			    sizeof(intr->bnx_intr_desc0), "%s rx%d",
6038 			    device_get_nameunit(sc->bnx_dev), idx);
6039 			intr->bnx_intr_desc = intr->bnx_intr_desc0;
6040 
6041 			intr->bnx_ret->bnx_msix_mbx = intr->bnx_intr_mbx;
6042 		}
6043 	}
6044 
6045 	if (BNX_IS_5717_PLUS(sc)) {
6046 		sc->bnx_msix_mem_rid = PCIR_BAR(4);
6047 	} else {
6048 		if (sc->bnx_res2 == NULL)
6049 			sc->bnx_msix_mem_rid = PCIR_BAR(2);
6050 	}
6051 	if (sc->bnx_msix_mem_rid != 0) {
6052 		sc->bnx_msix_mem_res = bus_alloc_resource_any(sc->bnx_dev,
6053 		    SYS_RES_MEMORY, &sc->bnx_msix_mem_rid, RF_ACTIVE);
6054 		if (sc->bnx_msix_mem_res == NULL) {
6055 			device_printf(sc->bnx_dev,
6056 			    "could not alloc MSI-X table\n");
6057 			return ENXIO;
6058 		}
6059 	}
6060 
6061 	bnx_enable_msi(sc, TRUE);
6062 
6063 	error = pci_setup_msix(sc->bnx_dev);
6064 	if (error) {
6065 		device_printf(sc->bnx_dev, "could not setup MSI-X\n");
6066 		goto back;
6067 	}
6068 	setup = TRUE;
6069 
6070 	for (i = 0; i < sc->bnx_intr_cnt; ++i) {
6071 		intr = &sc->bnx_intr_data[i];
6072 
6073 		error = pci_alloc_msix_vector(sc->bnx_dev, i,
6074 		    &intr->bnx_intr_rid, intr->bnx_intr_cpuid);
6075 		if (error) {
6076 			device_printf(sc->bnx_dev,
6077 			    "could not alloc MSI-X %d on cpu%d\n",
6078 			    i, intr->bnx_intr_cpuid);
6079 			goto back;
6080 		}
6081 
6082 		intr->bnx_intr_res = bus_alloc_resource_any(sc->bnx_dev,
6083 		    SYS_RES_IRQ, &intr->bnx_intr_rid, RF_ACTIVE);
6084 		if (intr->bnx_intr_res == NULL) {
6085 			device_printf(sc->bnx_dev,
6086 			    "could not alloc MSI-X %d resource\n", i);
6087 			error = ENXIO;
6088 			goto back;
6089 		}
6090 	}
6091 
6092 	pci_enable_msix(sc->bnx_dev);
6093 	sc->bnx_intr_type = PCI_INTR_TYPE_MSIX;
6094 back:
6095 	if (error)
6096 		bnx_free_msix(sc, setup);
6097 	return error;
6098 }
6099 
6100 static void
6101 bnx_free_msix(struct bnx_softc *sc, boolean_t setup)
6102 {
6103 	int i;
6104 
6105 	KKASSERT(sc->bnx_intr_cnt > 1);
6106 
6107 	for (i = 0; i < sc->bnx_intr_cnt; ++i) {
6108 		struct bnx_intr_data *intr = &sc->bnx_intr_data[i];
6109 
6110 		if (intr->bnx_intr_res != NULL) {
6111 			bus_release_resource(sc->bnx_dev, SYS_RES_IRQ,
6112 			    intr->bnx_intr_rid, intr->bnx_intr_res);
6113 		}
6114 		if (intr->bnx_intr_rid >= 0) {
6115 			pci_release_msix_vector(sc->bnx_dev,
6116 			    intr->bnx_intr_rid);
6117 		}
6118 	}
6119 	if (setup)
6120 		pci_teardown_msix(sc->bnx_dev);
6121 }
6122 
6123 static void
6124 bnx_rx_std_refill_sched_ipi(void *xret)
6125 {
6126 	struct bnx_rx_ret_ring *ret = xret;
6127 	struct bnx_rx_std_ring *std = ret->bnx_std;
6128 	struct globaldata *gd = mycpu;
6129 
6130 	crit_enter_gd(gd);
6131 
6132 	atomic_set_int(&std->bnx_rx_std_refill, ret->bnx_rx_mask);
6133 	cpu_sfence();
6134 
6135 	KKASSERT(std->bnx_rx_std_ithread.td_gd == gd);
6136 	lwkt_schedule(&std->bnx_rx_std_ithread);
6137 
6138 	crit_exit_gd(gd);
6139 }
6140 
6141 static void
6142 bnx_rx_std_refill_stop(void *xstd)
6143 {
6144 	struct bnx_rx_std_ring *std = xstd;
6145 	struct globaldata *gd = mycpu;
6146 
6147 	crit_enter_gd(gd);
6148 
6149 	std->bnx_rx_std_stop = 1;
6150 	cpu_sfence();
6151 
6152 	KKASSERT(std->bnx_rx_std_ithread.td_gd == gd);
6153 	lwkt_schedule(&std->bnx_rx_std_ithread);
6154 
6155 	crit_exit_gd(gd);
6156 }
6157 
6158 static void
6159 bnx_serialize_skipmain(struct bnx_softc *sc)
6160 {
6161 	lwkt_serialize_array_enter(sc->bnx_serialize,
6162 	    sc->bnx_serialize_cnt, 1);
6163 }
6164 
6165 static void
6166 bnx_deserialize_skipmain(struct bnx_softc *sc)
6167 {
6168 	lwkt_serialize_array_exit(sc->bnx_serialize,
6169 	    sc->bnx_serialize_cnt, 1);
6170 }
6171 
6172 static void
6173 bnx_rx_std_refill_sched(struct bnx_rx_ret_ring *ret,
6174     struct bnx_rx_std_ring *std)
6175 {
6176 	struct globaldata *gd = mycpu;
6177 
6178 	ret->bnx_rx_cnt = 0;
6179 	cpu_sfence();
6180 
6181 	crit_enter_gd(gd);
6182 
6183 	atomic_set_int(&std->bnx_rx_std_refill, ret->bnx_rx_mask);
6184 	cpu_sfence();
6185 	if (atomic_poll_acquire_int(&std->bnx_rx_std_running)) {
6186 		if (std->bnx_rx_std_ithread.td_gd == gd) {
6187 			lwkt_schedule(&std->bnx_rx_std_ithread);
6188 		} else {
6189 			lwkt_send_ipiq(
6190 			    std->bnx_rx_std_ithread.td_gd,
6191 			    bnx_rx_std_refill_sched_ipi, ret);
6192 		}
6193 	}
6194 
6195 	crit_exit_gd(gd);
6196 }
6197 
6198 static struct pktinfo *
6199 bnx_rss_info(struct pktinfo *pi, const struct bge_rx_bd *cur_rx)
6200 {
6201 	/* Don't pick up IPv6 packet */
6202 	if (cur_rx->bge_flags & BGE_RXBDFLAG_IPV6)
6203 		return NULL;
6204 
6205 	/* Don't pick up IP packet w/o IP checksum */
6206 	if ((cur_rx->bge_flags & BGE_RXBDFLAG_IP_CSUM) == 0 ||
6207 	    (cur_rx->bge_error_flag & BGE_RXERRFLAG_IP_CSUM_NOK))
6208 		return NULL;
6209 
6210 	/* Don't pick up IP packet w/o TCP/UDP checksum */
6211 	if ((cur_rx->bge_flags & BGE_RXBDFLAG_TCP_UDP_CSUM) == 0)
6212 		return NULL;
6213 
6214 	/* May be IP fragment */
6215 	if (cur_rx->bge_tcp_udp_csum != 0xffff)
6216 		return NULL;
6217 
6218 	if (cur_rx->bge_flags & BGE_RXBDFLAG_TCP_UDP_IS_TCP)
6219 		pi->pi_l3proto = IPPROTO_TCP;
6220 	else
6221 		pi->pi_l3proto = IPPROTO_UDP;
6222 	pi->pi_netisr = NETISR_IP;
6223 	pi->pi_flags = 0;
6224 
6225 	return pi;
6226 }
6227 
6228 static void
6229 bnx_sig_pre_reset(struct bnx_softc *sc, int type)
6230 {
6231 	if (type == BNX_RESET_START || type == BNX_RESET_SUSPEND)
6232 		bnx_ape_driver_state_change(sc, type);
6233 }
6234 
6235 static void
6236 bnx_sig_post_reset(struct bnx_softc *sc, int type)
6237 {
6238 	if (type == BNX_RESET_SHUTDOWN)
6239 		bnx_ape_driver_state_change(sc, type);
6240 }
6241 
6242 /*
6243  * Clear all stale locks and select the lock for this driver instance.
6244  */
6245 static void
6246 bnx_ape_lock_init(struct bnx_softc *sc)
6247 {
6248 	uint32_t bit, regbase;
6249 	int i;
6250 
6251 	regbase = BGE_APE_PER_LOCK_GRANT;
6252 
6253 	/* Clear any stale locks. */
6254 	for (i = BGE_APE_LOCK_PHY0; i <= BGE_APE_LOCK_GPIO; i++) {
6255 		switch (i) {
6256 		case BGE_APE_LOCK_PHY0:
6257 		case BGE_APE_LOCK_PHY1:
6258 		case BGE_APE_LOCK_PHY2:
6259 		case BGE_APE_LOCK_PHY3:
6260 			bit = BGE_APE_LOCK_GRANT_DRIVER0;
6261 			break;
6262 
6263 		default:
6264 			if (sc->bnx_func_addr == 0)
6265 				bit = BGE_APE_LOCK_GRANT_DRIVER0;
6266 			else
6267 				bit = 1 << sc->bnx_func_addr;
6268 			break;
6269 		}
6270 		APE_WRITE_4(sc, regbase + 4 * i, bit);
6271 	}
6272 
6273 	/* Select the PHY lock based on the device's function number. */
6274 	switch (sc->bnx_func_addr) {
6275 	case 0:
6276 		sc->bnx_phy_ape_lock = BGE_APE_LOCK_PHY0;
6277 		break;
6278 
6279 	case 1:
6280 		sc->bnx_phy_ape_lock = BGE_APE_LOCK_PHY1;
6281 		break;
6282 
6283 	case 2:
6284 		sc->bnx_phy_ape_lock = BGE_APE_LOCK_PHY2;
6285 		break;
6286 
6287 	case 3:
6288 		sc->bnx_phy_ape_lock = BGE_APE_LOCK_PHY3;
6289 		break;
6290 
6291 	default:
6292 		device_printf(sc->bnx_dev,
6293 		    "PHY lock not supported on this function\n");
6294 		break;
6295 	}
6296 }
6297 
6298 /*
6299  * Check for APE firmware, set flags, and print version info.
6300  */
6301 static void
6302 bnx_ape_read_fw_ver(struct bnx_softc *sc)
6303 {
6304 	const char *fwtype;
6305 	uint32_t apedata, features;
6306 
6307 	/* Check for a valid APE signature in shared memory. */
6308 	apedata = APE_READ_4(sc, BGE_APE_SEG_SIG);
6309 	if (apedata != BGE_APE_SEG_SIG_MAGIC) {
6310 		device_printf(sc->bnx_dev, "no APE signature\n");
6311 		sc->bnx_mfw_flags &= ~BNX_MFW_ON_APE;
6312 		return;
6313 	}
6314 
6315 	/* Check if APE firmware is running. */
6316 	apedata = APE_READ_4(sc, BGE_APE_FW_STATUS);
6317 	if ((apedata & BGE_APE_FW_STATUS_READY) == 0) {
6318 		device_printf(sc->bnx_dev, "APE signature found "
6319 		    "but FW status not ready! 0x%08x\n", apedata);
6320 		return;
6321 	}
6322 
6323 	sc->bnx_mfw_flags |= BNX_MFW_ON_APE;
6324 
6325 	/* Fetch the APE firwmare type and version. */
6326 	apedata = APE_READ_4(sc, BGE_APE_FW_VERSION);
6327 	features = APE_READ_4(sc, BGE_APE_FW_FEATURES);
6328 	if (features & BGE_APE_FW_FEATURE_NCSI) {
6329 		sc->bnx_mfw_flags |= BNX_MFW_TYPE_NCSI;
6330 		fwtype = "NCSI";
6331 	} else if (features & BGE_APE_FW_FEATURE_DASH) {
6332 		sc->bnx_mfw_flags |= BNX_MFW_TYPE_DASH;
6333 		fwtype = "DASH";
6334 	} else {
6335 		fwtype = "UNKN";
6336 	}
6337 
6338 	/* Print the APE firmware version. */
6339 	device_printf(sc->bnx_dev, "APE FW version: %s v%d.%d.%d.%d\n",
6340 	    fwtype,
6341 	    (apedata & BGE_APE_FW_VERSION_MAJMSK) >> BGE_APE_FW_VERSION_MAJSFT,
6342 	    (apedata & BGE_APE_FW_VERSION_MINMSK) >> BGE_APE_FW_VERSION_MINSFT,
6343 	    (apedata & BGE_APE_FW_VERSION_REVMSK) >> BGE_APE_FW_VERSION_REVSFT,
6344 	    (apedata & BGE_APE_FW_VERSION_BLDMSK));
6345 }
6346 
6347 static int
6348 bnx_ape_lock(struct bnx_softc *sc, int locknum)
6349 {
6350 	uint32_t bit, gnt, req, status;
6351 	int i, off;
6352 
6353 	if ((sc->bnx_mfw_flags & BNX_MFW_ON_APE) == 0)
6354 		return 0;
6355 
6356 	/* Lock request/grant registers have different bases. */
6357 	req = BGE_APE_PER_LOCK_REQ;
6358 	gnt = BGE_APE_PER_LOCK_GRANT;
6359 
6360 	off = 4 * locknum;
6361 
6362 	switch (locknum) {
6363 	case BGE_APE_LOCK_GPIO:
6364 		/* Lock required when using GPIO. */
6365 		if (sc->bnx_func_addr == 0)
6366 			bit = BGE_APE_LOCK_REQ_DRIVER0;
6367 		else
6368 			bit = 1 << sc->bnx_func_addr;
6369 		break;
6370 
6371 	case BGE_APE_LOCK_GRC:
6372 		/* Lock required to reset the device. */
6373 		if (sc->bnx_func_addr == 0)
6374 			bit = BGE_APE_LOCK_REQ_DRIVER0;
6375 		else
6376 			bit = 1 << sc->bnx_func_addr;
6377 		break;
6378 
6379 	case BGE_APE_LOCK_MEM:
6380 		/* Lock required when accessing certain APE memory. */
6381 		if (sc->bnx_func_addr == 0)
6382 			bit = BGE_APE_LOCK_REQ_DRIVER0;
6383 		else
6384 			bit = 1 << sc->bnx_func_addr;
6385 		break;
6386 
6387 	case BGE_APE_LOCK_PHY0:
6388 	case BGE_APE_LOCK_PHY1:
6389 	case BGE_APE_LOCK_PHY2:
6390 	case BGE_APE_LOCK_PHY3:
6391 		/* Lock required when accessing PHYs. */
6392 		bit = BGE_APE_LOCK_REQ_DRIVER0;
6393 		break;
6394 
6395 	default:
6396 		return EINVAL;
6397 	}
6398 
6399 	/* Request a lock. */
6400 	APE_WRITE_4(sc, req + off, bit);
6401 
6402 	/* Wait up to 1 second to acquire lock. */
6403 	for (i = 0; i < 20000; i++) {
6404 		status = APE_READ_4(sc, gnt + off);
6405 		if (status == bit)
6406 			break;
6407 		DELAY(50);
6408 	}
6409 
6410 	/* Handle any errors. */
6411 	if (status != bit) {
6412 		if_printf(&sc->arpcom.ac_if, "APE lock %d request failed! "
6413 		    "request = 0x%04x[0x%04x], status = 0x%04x[0x%04x]\n",
6414 		    locknum, req + off, bit & 0xFFFF, gnt + off,
6415 		    status & 0xFFFF);
6416 		/* Revoke the lock request. */
6417 		APE_WRITE_4(sc, gnt + off, bit);
6418 		return EBUSY;
6419 	}
6420 
6421 	return 0;
6422 }
6423 
6424 static void
6425 bnx_ape_unlock(struct bnx_softc *sc, int locknum)
6426 {
6427 	uint32_t bit, gnt;
6428 	int off;
6429 
6430 	if ((sc->bnx_mfw_flags & BNX_MFW_ON_APE) == 0)
6431 		return;
6432 
6433 	gnt = BGE_APE_PER_LOCK_GRANT;
6434 
6435 	off = 4 * locknum;
6436 
6437 	switch (locknum) {
6438 	case BGE_APE_LOCK_GPIO:
6439 		if (sc->bnx_func_addr == 0)
6440 			bit = BGE_APE_LOCK_GRANT_DRIVER0;
6441 		else
6442 			bit = 1 << sc->bnx_func_addr;
6443 		break;
6444 
6445 	case BGE_APE_LOCK_GRC:
6446 		if (sc->bnx_func_addr == 0)
6447 			bit = BGE_APE_LOCK_GRANT_DRIVER0;
6448 		else
6449 			bit = 1 << sc->bnx_func_addr;
6450 		break;
6451 
6452 	case BGE_APE_LOCK_MEM:
6453 		if (sc->bnx_func_addr == 0)
6454 			bit = BGE_APE_LOCK_GRANT_DRIVER0;
6455 		else
6456 			bit = 1 << sc->bnx_func_addr;
6457 		break;
6458 
6459 	case BGE_APE_LOCK_PHY0:
6460 	case BGE_APE_LOCK_PHY1:
6461 	case BGE_APE_LOCK_PHY2:
6462 	case BGE_APE_LOCK_PHY3:
6463 		bit = BGE_APE_LOCK_GRANT_DRIVER0;
6464 		break;
6465 
6466 	default:
6467 		return;
6468 	}
6469 
6470 	APE_WRITE_4(sc, gnt + off, bit);
6471 }
6472 
6473 /*
6474  * Send an event to the APE firmware.
6475  */
6476 static void
6477 bnx_ape_send_event(struct bnx_softc *sc, uint32_t event)
6478 {
6479 	uint32_t apedata;
6480 	int i;
6481 
6482 	/* NCSI does not support APE events. */
6483 	if ((sc->bnx_mfw_flags & BNX_MFW_ON_APE) == 0)
6484 		return;
6485 
6486 	/* Wait up to 1ms for APE to service previous event. */
6487 	for (i = 10; i > 0; i--) {
6488 		if (bnx_ape_lock(sc, BGE_APE_LOCK_MEM) != 0)
6489 			break;
6490 		apedata = APE_READ_4(sc, BGE_APE_EVENT_STATUS);
6491 		if ((apedata & BGE_APE_EVENT_STATUS_EVENT_PENDING) == 0) {
6492 			APE_WRITE_4(sc, BGE_APE_EVENT_STATUS, event |
6493 			    BGE_APE_EVENT_STATUS_EVENT_PENDING);
6494 			bnx_ape_unlock(sc, BGE_APE_LOCK_MEM);
6495 			APE_WRITE_4(sc, BGE_APE_EVENT, BGE_APE_EVENT_1);
6496 			break;
6497 		}
6498 		bnx_ape_unlock(sc, BGE_APE_LOCK_MEM);
6499 		DELAY(100);
6500 	}
6501 	if (i == 0) {
6502 		if_printf(&sc->arpcom.ac_if,
6503 		    "APE event 0x%08x send timed out\n", event);
6504 	}
6505 }
6506 
6507 static void
6508 bnx_ape_driver_state_change(struct bnx_softc *sc, int kind)
6509 {
6510 	uint32_t apedata, event;
6511 
6512 	if ((sc->bnx_mfw_flags & BNX_MFW_ON_APE) == 0)
6513 		return;
6514 
6515 	switch (kind) {
6516 	case BNX_RESET_START:
6517 		/* If this is the first load, clear the load counter. */
6518 		apedata = APE_READ_4(sc, BGE_APE_HOST_SEG_SIG);
6519 		if (apedata != BGE_APE_HOST_SEG_SIG_MAGIC) {
6520 			APE_WRITE_4(sc, BGE_APE_HOST_INIT_COUNT, 0);
6521 		} else {
6522 			apedata = APE_READ_4(sc, BGE_APE_HOST_INIT_COUNT);
6523 			APE_WRITE_4(sc, BGE_APE_HOST_INIT_COUNT, ++apedata);
6524 		}
6525 		APE_WRITE_4(sc, BGE_APE_HOST_SEG_SIG,
6526 		    BGE_APE_HOST_SEG_SIG_MAGIC);
6527 		APE_WRITE_4(sc, BGE_APE_HOST_SEG_LEN,
6528 		    BGE_APE_HOST_SEG_LEN_MAGIC);
6529 
6530 		/* Add some version info if bnx(4) supports it. */
6531 		APE_WRITE_4(sc, BGE_APE_HOST_DRIVER_ID,
6532 		    BGE_APE_HOST_DRIVER_ID_MAGIC(1, 0));
6533 		APE_WRITE_4(sc, BGE_APE_HOST_BEHAVIOR,
6534 		    BGE_APE_HOST_BEHAV_NO_PHYLOCK);
6535 		APE_WRITE_4(sc, BGE_APE_HOST_HEARTBEAT_INT_MS,
6536 		    BGE_APE_HOST_HEARTBEAT_INT_DISABLE);
6537 		APE_WRITE_4(sc, BGE_APE_HOST_DRVR_STATE,
6538 		    BGE_APE_HOST_DRVR_STATE_START);
6539 		event = BGE_APE_EVENT_STATUS_STATE_START;
6540 		break;
6541 
6542 	case BNX_RESET_SHUTDOWN:
6543 		APE_WRITE_4(sc, BGE_APE_HOST_DRVR_STATE,
6544 		    BGE_APE_HOST_DRVR_STATE_UNLOAD);
6545 		event = BGE_APE_EVENT_STATUS_STATE_UNLOAD;
6546 		break;
6547 
6548 	case BNX_RESET_SUSPEND:
6549 		event = BGE_APE_EVENT_STATUS_STATE_SUSPEND;
6550 		break;
6551 
6552 	default:
6553 		return;
6554 	}
6555 
6556 	bnx_ape_send_event(sc, event | BGE_APE_EVENT_STATUS_DRIVER_EVNT |
6557 	    BGE_APE_EVENT_STATUS_STATE_CHNGE);
6558 }
6559