xref: /dflybsd-src/sys/dev/netif/em/if_em.c (revision 744c01d0dc2aa1481a40e5b0988d15691602f5c9)
1 /*
2  *
3  * Copyright (c) 2004 Joerg Sonnenberger <joerg@bec.de>.  All rights reserved.
4  *
5  * Copyright (c) 2001-2006, Intel Corporation
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions are met:
10  *
11  *  1. Redistributions of source code must retain the above copyright notice,
12  *     this list of conditions and the following disclaimer.
13  *
14  *  2. Redistributions in binary form must reproduce the above copyright
15  *     notice, this list of conditions and the following disclaimer in the
16  *     documentation and/or other materials provided with the distribution.
17  *
18  *  3. Neither the name of the Intel Corporation nor the names of its
19  *     contributors may be used to endorse or promote products derived from
20  *     this software without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
23  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
26  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
27  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
28  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
29  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
30  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
31  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
32  * POSSIBILITY OF SUCH DAMAGE.
33  *
34  *
35  * Copyright (c) 2005 The DragonFly Project.  All rights reserved.
36  *
37  * This code is derived from software contributed to The DragonFly Project
38  * by Matthew Dillon <dillon@backplane.com>
39  *
40  * Redistribution and use in source and binary forms, with or without
41  * modification, are permitted provided that the following conditions
42  * are met:
43  *
44  * 1. Redistributions of source code must retain the above copyright
45  *    notice, this list of conditions and the following disclaimer.
46  * 2. Redistributions in binary form must reproduce the above copyright
47  *    notice, this list of conditions and the following disclaimer in
48  *    the documentation and/or other materials provided with the
49  *    distribution.
50  * 3. Neither the name of The DragonFly Project nor the names of its
51  *    contributors may be used to endorse or promote products derived
52  *    from this software without specific, prior written permission.
53  *
54  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
55  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
56  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
57  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
58  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
59  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
60  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
61  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
62  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
63  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
64  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
65  * SUCH DAMAGE.
66  *
67  * $DragonFly: src/sys/dev/netif/em/if_em.c,v 1.53 2006/12/23 10:39:16 sephe Exp $
68  * $FreeBSD$
69  */
70 /*
71  * SERIALIZATION API RULES:
72  *
73  * - If the driver uses the same serializer for the interrupt as for the
74  *   ifnet, most of the serialization will be done automatically for the
75  *   driver.
76  *
77  * - ifmedia entry points will be serialized by the ifmedia code using the
78  *   ifnet serializer.
79  *
80  * - if_* entry points except for if_input will be serialized by the IF
81  *   and protocol layers.
82  *
83  * - The device driver must be sure to serialize access from timeout code
84  *   installed by the device driver.
85  *
86  * - The device driver typically holds the serializer at the time it wishes
87  *   to call if_input.  If so, it should pass the serializer to if_input and
88  *   note that the serializer might be dropped temporarily by if_input
89  *   (e.g. in case it has to bridge the packet to another interface).
90  *
91  *   NOTE!  Since callers into the device driver hold the ifnet serializer,
92  *   the device driver may be holding a serializer at the time it calls
93  *   if_input even if it is not serializer-aware.
94  */
95 
96 #include "opt_polling.h"
97 #include "opt_inet.h"
98 
99 #include <sys/param.h>
100 #include <sys/bus.h>
101 #include <sys/endian.h>
102 #include <sys/kernel.h>
103 #include <sys/ktr.h>
104 #include <sys/malloc.h>
105 #include <sys/mbuf.h>
106 #include <sys/module.h>
107 #include <sys/rman.h>
108 #include <sys/serialize.h>
109 #include <sys/socket.h>
110 #include <sys/sockio.h>
111 #include <sys/sysctl.h>
112 
113 #include <net/bpf.h>
114 #include <net/ethernet.h>
115 #include <net/if.h>
116 #include <net/if_arp.h>
117 #include <net/if_dl.h>
118 #include <net/if_media.h>
119 #include <net/if_types.h>
120 #include <net/ifq_var.h>
121 #include <net/vlan/if_vlan_var.h>
122 
123 #ifdef INET
124 #include <netinet/in.h>
125 #include <netinet/in_systm.h>
126 #include <netinet/in_var.h>
127 #include <netinet/ip.h>
128 #include <netinet/tcp.h>
129 #include <netinet/udp.h>
130 #endif
131 
132 #include <dev/netif/em/if_em_hw.h>
133 #include <dev/netif/em/if_em.h>
134 
135 /*********************************************************************
136  *  Set this to one to display debug statistics
137  *********************************************************************/
138 int	em_display_debug_stats = 0;
139 
140 /*********************************************************************
141  *  Driver version
142  *********************************************************************/
143 
144 char em_driver_version[] = "6.2.9";
145 
146 
147 /*********************************************************************
148  *  PCI Device ID Table
149  *
150  *  Used by probe to select devices to load on
151  *  Last field stores an index into em_strings
152  *  Last entry must be all 0s
153  *
154  *  { Vendor ID, Device ID, SubVendor ID, SubDevice ID, String Index }
155  *********************************************************************/
156 
157 static em_vendor_info_t em_vendor_info_array[] =
158 {
159 	/* Intel(R) PRO/1000 Network Connection */
160 	{ 0x8086, E1000_DEV_ID_82540EM,		PCI_ANY_ID, PCI_ANY_ID, 0},
161 	{ 0x8086, E1000_DEV_ID_82540EM_LOM,	PCI_ANY_ID, PCI_ANY_ID, 0},
162 	{ 0x8086, E1000_DEV_ID_82540EP,		PCI_ANY_ID, PCI_ANY_ID, 0},
163 	{ 0x8086, E1000_DEV_ID_82540EP_LOM,	PCI_ANY_ID, PCI_ANY_ID, 0},
164 	{ 0x8086, E1000_DEV_ID_82540EP_LP,	PCI_ANY_ID, PCI_ANY_ID, 0},
165 
166 	{ 0x8086, E1000_DEV_ID_82541EI,		PCI_ANY_ID, PCI_ANY_ID, 0},
167 	{ 0x8086, E1000_DEV_ID_82541ER,		PCI_ANY_ID, PCI_ANY_ID, 0},
168 	{ 0x8086, E1000_DEV_ID_82541ER_LOM,	PCI_ANY_ID, PCI_ANY_ID, 0},
169 	{ 0x8086, E1000_DEV_ID_82541EI_MOBILE,	PCI_ANY_ID, PCI_ANY_ID, 0},
170 	{ 0x8086, E1000_DEV_ID_82541GI,		PCI_ANY_ID, PCI_ANY_ID, 0},
171 	{ 0x8086, E1000_DEV_ID_82541GI_LF,	PCI_ANY_ID, PCI_ANY_ID, 0},
172 	{ 0x8086, E1000_DEV_ID_82541GI_MOBILE,	PCI_ANY_ID, PCI_ANY_ID, 0},
173 
174 	{ 0x8086, E1000_DEV_ID_82542,		PCI_ANY_ID, PCI_ANY_ID, 0},
175 
176 	{ 0x8086, E1000_DEV_ID_82543GC_FIBER,	PCI_ANY_ID, PCI_ANY_ID, 0},
177 	{ 0x8086, E1000_DEV_ID_82543GC_COPPER,	PCI_ANY_ID, PCI_ANY_ID, 0},
178 
179 	{ 0x8086, E1000_DEV_ID_82544EI_COPPER,	PCI_ANY_ID, PCI_ANY_ID, 0},
180 	{ 0x8086, E1000_DEV_ID_82544EI_FIBER,	PCI_ANY_ID, PCI_ANY_ID, 0},
181 	{ 0x8086, E1000_DEV_ID_82544GC_COPPER,	PCI_ANY_ID, PCI_ANY_ID, 0},
182 	{ 0x8086, E1000_DEV_ID_82544GC_LOM,	PCI_ANY_ID, PCI_ANY_ID, 0},
183 
184 	{ 0x8086, E1000_DEV_ID_82545EM_COPPER,	PCI_ANY_ID, PCI_ANY_ID, 0},
185 	{ 0x8086, E1000_DEV_ID_82545EM_FIBER,	PCI_ANY_ID, PCI_ANY_ID, 0},
186 	{ 0x8086, E1000_DEV_ID_82545GM_COPPER,	PCI_ANY_ID, PCI_ANY_ID, 0},
187 	{ 0x8086, E1000_DEV_ID_82545GM_FIBER,	PCI_ANY_ID, PCI_ANY_ID, 0},
188 	{ 0x8086, E1000_DEV_ID_82545GM_SERDES,	PCI_ANY_ID, PCI_ANY_ID, 0},
189 
190 	{ 0x8086, E1000_DEV_ID_82546EB_COPPER,	PCI_ANY_ID, PCI_ANY_ID, 0},
191 	{ 0x8086, E1000_DEV_ID_82546EB_FIBER,	PCI_ANY_ID, PCI_ANY_ID, 0},
192 	{ 0x8086, E1000_DEV_ID_82546EB_QUAD_COPPER, PCI_ANY_ID, PCI_ANY_ID, 0},
193 	{ 0x8086, E1000_DEV_ID_82546GB_COPPER,	PCI_ANY_ID, PCI_ANY_ID, 0},
194 	{ 0x8086, E1000_DEV_ID_82546GB_FIBER,	PCI_ANY_ID, PCI_ANY_ID, 0},
195 	{ 0x8086, E1000_DEV_ID_82546GB_SERDES,	PCI_ANY_ID, PCI_ANY_ID, 0},
196 	{ 0x8086, E1000_DEV_ID_82546GB_PCIE,	PCI_ANY_ID, PCI_ANY_ID, 0},
197 	{ 0x8086, E1000_DEV_ID_82546GB_QUAD_COPPER, PCI_ANY_ID, PCI_ANY_ID, 0},
198 	{ 0x8086, E1000_DEV_ID_82546GB_QUAD_COPPER_KSP3,
199 						PCI_ANY_ID, PCI_ANY_ID, 0},
200 
201 	{ 0x8086, E1000_DEV_ID_82547EI,		PCI_ANY_ID, PCI_ANY_ID, 0},
202 	{ 0x8086, E1000_DEV_ID_82547EI_MOBILE,	PCI_ANY_ID, PCI_ANY_ID, 0},
203 	{ 0x8086, E1000_DEV_ID_82547GI,		PCI_ANY_ID, PCI_ANY_ID, 0},
204 
205 	{ 0x8086, E1000_DEV_ID_82571EB_COPPER,	PCI_ANY_ID, PCI_ANY_ID, 0},
206 	{ 0x8086, E1000_DEV_ID_82571EB_FIBER,	PCI_ANY_ID, PCI_ANY_ID, 0},
207 	{ 0x8086, E1000_DEV_ID_82571EB_SERDES,	PCI_ANY_ID, PCI_ANY_ID, 0},
208 	{ 0x8086, E1000_DEV_ID_82571EB_QUAD_COPPER,
209 						PCI_ANY_ID, PCI_ANY_ID, 0},
210 	{ 0x8086, E1000_DEV_ID_82571EB_QUAD_COPPER_LOWPROFILE,
211 						PCI_ANY_ID, PCI_ANY_ID, 0},
212 
213 	{ 0x8086, E1000_DEV_ID_82572EI_COPPER,	PCI_ANY_ID, PCI_ANY_ID, 0},
214 	{ 0x8086, E1000_DEV_ID_82572EI_FIBER,	PCI_ANY_ID, PCI_ANY_ID, 0},
215 	{ 0x8086, E1000_DEV_ID_82572EI_SERDES,	PCI_ANY_ID, PCI_ANY_ID, 0},
216 	{ 0x8086, E1000_DEV_ID_82572EI,		PCI_ANY_ID, PCI_ANY_ID, 0},
217 
218 	{ 0x8086, E1000_DEV_ID_82573E,		PCI_ANY_ID, PCI_ANY_ID, 0},
219 	{ 0x8086, E1000_DEV_ID_82573E_IAMT,	PCI_ANY_ID, PCI_ANY_ID, 0},
220 	{ 0x8086, E1000_DEV_ID_82573L,		PCI_ANY_ID, PCI_ANY_ID, 0},
221 
222 	{ 0x8086, E1000_DEV_ID_80003ES2LAN_COPPER_SPT,
223 						PCI_ANY_ID, PCI_ANY_ID, 0},
224 	{ 0x8086, E1000_DEV_ID_80003ES2LAN_SERDES_SPT,
225 						PCI_ANY_ID, PCI_ANY_ID, 0},
226 	{ 0x8086, E1000_DEV_ID_80003ES2LAN_COPPER_DPT,
227 						PCI_ANY_ID, PCI_ANY_ID, 0},
228 	{ 0x8086, E1000_DEV_ID_80003ES2LAN_SERDES_DPT,
229 						PCI_ANY_ID, PCI_ANY_ID, 0},
230 
231 	{ 0x8086, E1000_DEV_ID_ICH8_IGP_M_AMT,	PCI_ANY_ID, PCI_ANY_ID, 0},
232 	{ 0x8086, E1000_DEV_ID_ICH8_IGP_AMT,	PCI_ANY_ID, PCI_ANY_ID, 0},
233 	{ 0x8086, E1000_DEV_ID_ICH8_IGP_C,	PCI_ANY_ID, PCI_ANY_ID, 0},
234 	{ 0x8086, E1000_DEV_ID_ICH8_IFE,	PCI_ANY_ID, PCI_ANY_ID, 0},
235 	{ 0x8086, E1000_DEV_ID_ICH8_IFE_GT,	PCI_ANY_ID, PCI_ANY_ID, 0},
236 	{ 0x8086, E1000_DEV_ID_ICH8_IFE_G,	PCI_ANY_ID, PCI_ANY_ID, 0},
237 	{ 0x8086, E1000_DEV_ID_ICH8_IGP_M,	PCI_ANY_ID, PCI_ANY_ID, 0},
238 
239 	{ 0x8086, 0x101A, PCI_ANY_ID, PCI_ANY_ID, 0},
240 	{ 0x8086, 0x1014, PCI_ANY_ID, PCI_ANY_ID, 0},
241 	/* required last entry */
242 	{ 0, 0, 0, 0, 0}
243 };
244 
245 /*********************************************************************
246  *  Table of branding strings for all supported NICs.
247  *********************************************************************/
248 
249 static const char *em_strings[] = {
250 	"Intel(R) PRO/1000 Network Connection"
251 };
252 
253 /*********************************************************************
254  *  Function prototypes
255  *********************************************************************/
256 static int	em_probe(device_t);
257 static int	em_attach(device_t);
258 static int	em_detach(device_t);
259 static int	em_shutdown(device_t);
260 static void	em_intr(void *);
261 static int	em_suspend(device_t);
262 static int	em_resume(device_t);
263 static void	em_start(struct ifnet *);
264 static int	em_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
265 static void	em_watchdog(struct ifnet *);
266 static void	em_init(void *);
267 static void	em_stop(void *);
268 static void	em_media_status(struct ifnet *, struct ifmediareq *);
269 static int	em_media_change(struct ifnet *);
270 static void	em_identify_hardware(struct adapter *);
271 static int	em_allocate_pci_resources(device_t);
272 static void	em_free_pci_resources(device_t);
273 static void	em_local_timer(void *);
274 static int	em_hardware_init(struct adapter *);
275 static void	em_setup_interface(device_t, struct adapter *);
276 static int	em_setup_transmit_structures(struct adapter *);
277 static void	em_initialize_transmit_unit(struct adapter *);
278 static int	em_setup_receive_structures(struct adapter *);
279 static void	em_initialize_receive_unit(struct adapter *);
280 static void	em_enable_intr(struct adapter *);
281 static void	em_disable_intr(struct adapter *);
282 static void	em_free_transmit_structures(struct adapter *);
283 static void	em_free_receive_structures(struct adapter *);
284 static void	em_update_stats_counters(struct adapter *);
285 static void	em_txeof(struct adapter *);
286 static int	em_allocate_receive_structures(struct adapter *);
287 static void	em_rxeof(struct adapter *, int);
288 static void	em_receive_checksum(struct adapter *, struct em_rx_desc *,
289 				    struct mbuf *);
290 static void	em_transmit_checksum_setup(struct adapter *, struct mbuf *,
291 					   uint32_t *, uint32_t *);
292 static void	em_set_promisc(struct adapter *);
293 static void	em_disable_promisc(struct adapter *);
294 static void	em_set_multi(struct adapter *);
295 static void	em_print_hw_stats(struct adapter *);
296 static void	em_update_link_status(struct adapter *);
297 static int	em_get_buf(int i, struct adapter *, struct mbuf *, int how);
298 static void	em_enable_vlans(struct adapter *);
299 static void	em_disable_vlans(struct adapter *);
300 static int	em_encap(struct adapter *, struct mbuf *);
301 static void	em_smartspeed(struct adapter *);
302 static int	em_82547_fifo_workaround(struct adapter *, int);
303 static void	em_82547_update_fifo_head(struct adapter *, int);
304 static int	em_82547_tx_fifo_reset(struct adapter *);
305 static void	em_82547_move_tail(void *);
306 static void	em_82547_move_tail_serialized(struct adapter *);
307 static int	em_dma_malloc(struct adapter *, bus_size_t,
308 			      struct em_dma_alloc *);
309 static void	em_dma_free(struct adapter *, struct em_dma_alloc *);
310 static void	em_print_debug_info(struct adapter *);
311 static int	em_is_valid_ether_addr(uint8_t *);
312 static int	em_sysctl_stats(SYSCTL_HANDLER_ARGS);
313 static int	em_sysctl_debug_info(SYSCTL_HANDLER_ARGS);
314 static uint32_t	em_fill_descriptors(bus_addr_t address, uint32_t length,
315 				   PDESC_ARRAY desc_array);
316 static int	em_sysctl_int_delay(SYSCTL_HANDLER_ARGS);
317 static int	em_sysctl_int_throttle(SYSCTL_HANDLER_ARGS);
318 static void	em_add_int_delay_sysctl(struct adapter *, const char *,
319 					const char *,
320 					struct em_int_delay_info *, int, int);
321 
322 /*********************************************************************
323  *  FreeBSD Device Interface Entry Points
324  *********************************************************************/
325 
326 static device_method_t em_methods[] = {
327 	/* Device interface */
328 	DEVMETHOD(device_probe, em_probe),
329 	DEVMETHOD(device_attach, em_attach),
330 	DEVMETHOD(device_detach, em_detach),
331 	DEVMETHOD(device_shutdown, em_shutdown),
332 	DEVMETHOD(device_suspend, em_suspend),
333 	DEVMETHOD(device_resume, em_resume),
334 	{0, 0}
335 };
336 
337 static driver_t em_driver = {
338 	"em", em_methods, sizeof(struct adapter),
339 };
340 
341 static devclass_t em_devclass;
342 
343 DECLARE_DUMMY_MODULE(if_em);
344 DRIVER_MODULE(if_em, pci, em_driver, em_devclass, 0, 0);
345 
346 /*********************************************************************
347  *  Tunable default values.
348  *********************************************************************/
349 
350 #define E1000_TICKS_TO_USECS(ticks)	((1024 * (ticks) + 500) / 1000)
351 #define E1000_USECS_TO_TICKS(usecs)	((1000 * (usecs) + 512) / 1024)
352 
353 static int em_tx_int_delay_dflt = E1000_TICKS_TO_USECS(EM_TIDV);
354 static int em_rx_int_delay_dflt = E1000_TICKS_TO_USECS(EM_RDTR);
355 static int em_tx_abs_int_delay_dflt = E1000_TICKS_TO_USECS(EM_TADV);
356 static int em_rx_abs_int_delay_dflt = E1000_TICKS_TO_USECS(EM_RADV);
357 static int em_int_throttle_ceil = 10000;
358 static int em_rxd = EM_DEFAULT_RXD;
359 static int em_txd = EM_DEFAULT_TXD;
360 static int em_smart_pwr_down = FALSE;
361 
362 TUNABLE_INT("hw.em.tx_int_delay", &em_tx_int_delay_dflt);
363 TUNABLE_INT("hw.em.rx_int_delay", &em_rx_int_delay_dflt);
364 TUNABLE_INT("hw.em.tx_abs_int_delay", &em_tx_abs_int_delay_dflt);
365 TUNABLE_INT("hw.em.rx_abs_int_delay", &em_rx_abs_int_delay_dflt);
366 TUNABLE_INT("hw.em.int_throttle_ceil", &em_int_throttle_ceil);
367 TUNABLE_INT("hw.em.rxd", &em_rxd);
368 TUNABLE_INT("hw.em.txd", &em_txd);
369 TUNABLE_INT("hw.em.smart_pwr_down", &em_smart_pwr_down);
370 
371 /*
372  * Kernel trace for characterization of operations
373  */
374 #if !defined(KTR_IF_EM)
375 #define KTR_IF_EM	KTR_ALL
376 #endif
377 KTR_INFO_MASTER(if_em);
378 KTR_INFO(KTR_IF_EM, if_em, intr_beg, 0, "intr begin", 0);
379 KTR_INFO(KTR_IF_EM, if_em, intr_end, 1, "intr end", 0);
380 #ifdef DEVICE_POLLING
381 KTR_INFO(KTR_IF_EM, if_em, poll_beg, 2, "poll begin", 0);
382 KTR_INFO(KTR_IF_EM, if_em, poll_end, 3, "poll end", 0);
383 #endif
384 KTR_INFO(KTR_IF_EM, if_em, pkt_receive, 4, "rx packet", 0);
385 KTR_INFO(KTR_IF_EM, if_em, pkt_txqueue, 5, "tx packet", 0);
386 KTR_INFO(KTR_IF_EM, if_em, pkt_txclean, 6, "tx clean", 0);
387 #define logif(name)	KTR_LOG(if_em_ ## name)
388 
389 /*********************************************************************
390  *  Device identification routine
391  *
392  *  em_probe determines if the driver should be loaded on
393  *  adapter based on PCI vendor/device id of the adapter.
394  *
395  *  return 0 on success, positive on failure
396  *********************************************************************/
397 
398 static int
399 em_probe(device_t dev)
400 {
401 	em_vendor_info_t *ent;
402 
403 	uint16_t pci_vendor_id = 0;
404 	uint16_t pci_device_id = 0;
405 	uint16_t pci_subvendor_id = 0;
406 	uint16_t pci_subdevice_id = 0;
407 	char adapter_name[60];
408 
409 	INIT_DEBUGOUT("em_probe: begin");
410 
411 	pci_vendor_id = pci_get_vendor(dev);
412 	if (pci_vendor_id != EM_VENDOR_ID)
413 		return (ENXIO);
414 
415 	pci_device_id = pci_get_device(dev);
416 	pci_subvendor_id = pci_get_subvendor(dev);
417 	pci_subdevice_id = pci_get_subdevice(dev);
418 
419 	ent = em_vendor_info_array;
420 	while (ent->vendor_id != 0) {
421 		if ((pci_vendor_id == ent->vendor_id) &&
422 		    (pci_device_id == ent->device_id) &&
423 
424 		    ((pci_subvendor_id == ent->subvendor_id) ||
425 		     (ent->subvendor_id == PCI_ANY_ID)) &&
426 
427 		    ((pci_subdevice_id == ent->subdevice_id) ||
428 		     (ent->subdevice_id == PCI_ANY_ID))) {
429 			ksnprintf(adapter_name, sizeof(adapter_name),
430 				 "%s, Version - %s",  em_strings[ent->index],
431 				 em_driver_version);
432 			device_set_desc_copy(dev, adapter_name);
433 			return (0);
434 		}
435 		ent++;
436 	}
437 
438 	return (ENXIO);
439 }
440 
441 /*********************************************************************
442  *  Device initialization routine
443  *
444  *  The attach entry point is called when the driver is being loaded.
445  *  This routine identifies the type of hardware, allocates all resources
446  *  and initializes the hardware.
447  *
448  *  return 0 on success, positive on failure
449  *********************************************************************/
450 
451 static int
452 em_attach(device_t dev)
453 {
454 	struct adapter *adapter;
455 	int tsize, rsize;
456 	int error = 0;
457 
458 	INIT_DEBUGOUT("em_attach: begin");
459 
460 	adapter = device_get_softc(dev);
461 
462 	callout_init(&adapter->timer);
463 	callout_init(&adapter->tx_fifo_timer);
464 
465 	adapter->dev = dev;
466 	adapter->osdep.dev = dev;
467 
468 	/* SYSCTL stuff */
469 	sysctl_ctx_init(&adapter->sysctl_ctx);
470 	adapter->sysctl_tree = SYSCTL_ADD_NODE(&adapter->sysctl_ctx,
471 					       SYSCTL_STATIC_CHILDREN(_hw),
472 					       OID_AUTO,
473 					       device_get_nameunit(dev),
474 					       CTLFLAG_RD,
475 					       0, "");
476 
477 	if (adapter->sysctl_tree == NULL) {
478 		device_printf(dev, "Unable to create sysctl tree\n");
479 		return EIO;
480 	}
481 
482 	SYSCTL_ADD_PROC(&adapter->sysctl_ctx,
483 			SYSCTL_CHILDREN(adapter->sysctl_tree),
484 			OID_AUTO, "debug_info", CTLTYPE_INT|CTLFLAG_RW,
485 			(void *)adapter, 0,
486 			em_sysctl_debug_info, "I", "Debug Information");
487 
488 	SYSCTL_ADD_PROC(&adapter->sysctl_ctx,
489 			SYSCTL_CHILDREN(adapter->sysctl_tree),
490 			OID_AUTO, "stats", CTLTYPE_INT|CTLFLAG_RW,
491 			(void *)adapter, 0,
492 			em_sysctl_stats, "I", "Statistics");
493 
494 	/* Determine hardware revision */
495 	em_identify_hardware(adapter);
496 
497 	/* Set up some sysctls for the tunable interrupt delays */
498 	em_add_int_delay_sysctl(adapter, "rx_int_delay",
499 				"receive interrupt delay in usecs",
500 				&adapter->rx_int_delay,
501 				E1000_REG_OFFSET(&adapter->hw, RDTR),
502 				em_rx_int_delay_dflt);
503 	em_add_int_delay_sysctl(adapter, "tx_int_delay",
504 				"transmit interrupt delay in usecs",
505 				&adapter->tx_int_delay,
506 				E1000_REG_OFFSET(&adapter->hw, TIDV),
507 				em_tx_int_delay_dflt);
508 	if (adapter->hw.mac_type >= em_82540) {
509 		em_add_int_delay_sysctl(adapter, "rx_abs_int_delay",
510 					"receive interrupt delay limit in usecs",
511 					&adapter->rx_abs_int_delay,
512 					E1000_REG_OFFSET(&adapter->hw, RADV),
513 					em_rx_abs_int_delay_dflt);
514 		em_add_int_delay_sysctl(adapter, "tx_abs_int_delay",
515 					"transmit interrupt delay limit in usecs",
516 					&adapter->tx_abs_int_delay,
517 					E1000_REG_OFFSET(&adapter->hw, TADV),
518 					em_tx_abs_int_delay_dflt);
519 		SYSCTL_ADD_PROC(&adapter->sysctl_ctx,
520 			SYSCTL_CHILDREN(adapter->sysctl_tree),
521 			OID_AUTO, "int_throttle_ceil", CTLTYPE_INT|CTLFLAG_RW,
522 			adapter, 0, em_sysctl_int_throttle, "I", NULL);
523 	}
524 
525 	/*
526 	 * Validate number of transmit and receive descriptors. It
527 	 * must not exceed hardware maximum, and must be multiple
528 	 * of EM_DBA_ALIGN.
529 	 */
530 	if (((em_txd * sizeof(struct em_tx_desc)) % EM_DBA_ALIGN) != 0 ||
531 	    (adapter->hw.mac_type >= em_82544 && em_txd > EM_MAX_TXD) ||
532 	    (adapter->hw.mac_type < em_82544 && em_txd > EM_MAX_TXD_82543) ||
533 	    (em_txd < EM_MIN_TXD)) {
534 		device_printf(dev, "Using %d TX descriptors instead of %d!\n",
535 			      EM_DEFAULT_TXD, em_txd);
536 		adapter->num_tx_desc = EM_DEFAULT_TXD;
537 	} else {
538 		adapter->num_tx_desc = em_txd;
539 	}
540 
541 	if (((em_rxd * sizeof(struct em_rx_desc)) % EM_DBA_ALIGN) != 0 ||
542 	    (adapter->hw.mac_type >= em_82544 && em_rxd > EM_MAX_RXD) ||
543 	    (adapter->hw.mac_type < em_82544 && em_rxd > EM_MAX_RXD_82543) ||
544 	    (em_rxd < EM_MIN_RXD)) {
545 		device_printf(dev, "Using %d RX descriptors instead of %d!\n",
546 			      EM_DEFAULT_RXD, em_rxd);
547 		adapter->num_rx_desc = EM_DEFAULT_RXD;
548 	} else {
549 		adapter->num_rx_desc = em_rxd;
550 	}
551 
552 	adapter->hw.autoneg = DO_AUTO_NEG;
553 	adapter->hw.wait_autoneg_complete = WAIT_FOR_AUTO_NEG_DEFAULT;
554 	adapter->hw.autoneg_advertised = AUTONEG_ADV_DEFAULT;
555 	adapter->hw.tbi_compatibility_en = TRUE;
556 	adapter->rx_buffer_len = EM_RXBUFFER_2048;
557 
558 	adapter->hw.phy_init_script = 1;
559 	adapter->hw.phy_reset_disable = FALSE;
560 
561 #ifndef EM_MASTER_SLAVE
562 	adapter->hw.master_slave = em_ms_hw_default;
563 #else
564 	adapter->hw.master_slave = EM_MASTER_SLAVE;
565 #endif
566 
567 	/*
568 	 * Set the max frame size assuming standard ethernet
569 	 * sized frames.
570 	 */
571 	adapter->hw.max_frame_size = ETHERMTU + ETHER_HDR_LEN + ETHER_CRC_LEN;
572 
573 	adapter->hw.min_frame_size =
574 	    MINIMUM_ETHERNET_PACKET_SIZE + ETHER_CRC_LEN;
575 
576 	/*
577 	 * This controls when hardware reports transmit completion
578 	 * status.
579 	 */
580 	adapter->hw.report_tx_early = 1;
581 
582 	error = em_allocate_pci_resources(dev);
583 	if (error)
584 		goto fail;
585 
586 	/* Initialize eeprom parameters */
587 	em_init_eeprom_params(&adapter->hw);
588 
589 	tsize = roundup2(adapter->num_tx_desc * sizeof(struct em_tx_desc),
590 			 EM_DBA_ALIGN);
591 
592 	/* Allocate Transmit Descriptor ring */
593 	error = em_dma_malloc(adapter, tsize, &adapter->txdma);
594 	if (error) {
595 		device_printf(dev, "Unable to allocate TxDescriptor memory\n");
596 		goto fail;
597 	}
598 	adapter->tx_desc_base = (struct em_tx_desc *)adapter->txdma.dma_vaddr;
599 
600 	rsize = roundup2(adapter->num_rx_desc * sizeof(struct em_rx_desc),
601 			 EM_DBA_ALIGN);
602 
603 	/* Allocate Receive Descriptor ring */
604 	error = em_dma_malloc(adapter, rsize, &adapter->rxdma);
605 	if (error) {
606 		device_printf(dev, "Unable to allocate rx_desc memory\n");
607 		goto fail;
608 	}
609 	adapter->rx_desc_base = (struct em_rx_desc *)adapter->rxdma.dma_vaddr;
610 
611 	/* Initialize the hardware */
612 	if (em_hardware_init(adapter)) {
613 		device_printf(dev, "Unable to initialize the hardware\n");
614 		error = EIO;
615 		goto fail;
616 	}
617 
618 	/* Copy the permanent MAC address out of the EEPROM */
619 	if (em_read_mac_addr(&adapter->hw) < 0) {
620 		device_printf(dev,
621 			      "EEPROM read error while reading MAC address\n");
622 		error = EIO;
623 		goto fail;
624 	}
625 
626 	if (!em_is_valid_ether_addr(adapter->hw.mac_addr)) {
627 		device_printf(dev, "Invalid MAC address\n");
628 		error = EIO;
629 		goto fail;
630 	}
631 
632 	/* Setup OS specific network interface */
633 	em_setup_interface(dev, adapter);
634 
635 	/* Initialize statistics */
636 	em_clear_hw_cntrs(&adapter->hw);
637 	em_update_stats_counters(adapter);
638 	adapter->hw.get_link_status = 1;
639 	em_update_link_status(adapter);
640 
641 	/* Indicate SOL/IDER usage */
642 	if (em_check_phy_reset_block(&adapter->hw)) {
643 		device_printf(dev, "PHY reset is blocked due to "
644 			      "SOL/IDER session.\n");
645 	}
646 
647 	/* Identify 82544 on PCIX */
648 	em_get_bus_info(&adapter->hw);
649 	if (adapter->hw.bus_type == em_bus_type_pcix &&
650 	    adapter->hw.mac_type == em_82544)
651 		adapter->pcix_82544 = TRUE;
652 	else
653 		adapter->pcix_82544 = FALSE;
654 
655 	error = bus_setup_intr(dev, adapter->res_interrupt, INTR_NETSAFE,
656 			   em_intr, adapter,
657 			   &adapter->int_handler_tag,
658 			   adapter->interface_data.ac_if.if_serializer);
659 	if (error) {
660 		device_printf(dev, "Error registering interrupt handler!\n");
661 		ether_ifdetach(&adapter->interface_data.ac_if);
662 		goto fail;
663 	}
664 
665 	INIT_DEBUGOUT("em_attach: end");
666 	return(0);
667 
668 fail:
669 	em_detach(dev);
670 	return(error);
671 }
672 
673 /*********************************************************************
674  *  Device removal routine
675  *
676  *  The detach entry point is called when the driver is being removed.
677  *  This routine stops the adapter and deallocates all the resources
678  *  that were allocated for driver operation.
679  *
680  *  return 0 on success, positive on failure
681  *********************************************************************/
682 
683 static int
684 em_detach(device_t dev)
685 {
686 	struct adapter *adapter = device_get_softc(dev);
687 
688 	INIT_DEBUGOUT("em_detach: begin");
689 
690 	if (device_is_attached(dev)) {
691 		struct ifnet *ifp = &adapter->interface_data.ac_if;
692 
693 		lwkt_serialize_enter(ifp->if_serializer);
694 		adapter->in_detach = 1;
695 		em_stop(adapter);
696 		em_phy_hw_reset(&adapter->hw);
697 		bus_teardown_intr(dev, adapter->res_interrupt,
698 				  adapter->int_handler_tag);
699 		lwkt_serialize_exit(ifp->if_serializer);
700 
701 		ether_ifdetach(ifp);
702 	}
703 	bus_generic_detach(dev);
704 
705 	em_free_pci_resources(dev);
706 
707 	/* Free Transmit Descriptor ring */
708 	if (adapter->tx_desc_base != NULL) {
709 		em_dma_free(adapter, &adapter->txdma);
710 		adapter->tx_desc_base = NULL;
711 	}
712 
713 	/* Free Receive Descriptor ring */
714 	if (adapter->rx_desc_base != NULL) {
715 		em_dma_free(adapter, &adapter->rxdma);
716 		adapter->rx_desc_base = NULL;
717 	}
718 
719 	/* Free sysctl tree */
720 	if (adapter->sysctl_tree != NULL) {
721 		adapter->sysctl_tree = NULL;
722 		sysctl_ctx_free(&adapter->sysctl_ctx);
723 	}
724 
725 	return (0);
726 }
727 
728 /*********************************************************************
729  *
730  *  Shutdown entry point
731  *
732  **********************************************************************/
733 
734 static int
735 em_shutdown(device_t dev)
736 {
737 	struct adapter *adapter = device_get_softc(dev);
738 	struct ifnet *ifp = &adapter->interface_data.ac_if;
739 
740 	lwkt_serialize_enter(ifp->if_serializer);
741 	em_stop(adapter);
742 	lwkt_serialize_exit(ifp->if_serializer);
743 
744 	return (0);
745 }
746 
747 /*
748  * Suspend/resume device methods.
749  */
750 static int
751 em_suspend(device_t dev)
752 {
753 	struct adapter *adapter = device_get_softc(dev);
754 	struct ifnet *ifp = &adapter->interface_data.ac_if;
755 
756 	lwkt_serialize_enter(ifp->if_serializer);
757 	em_stop(adapter);
758 	lwkt_serialize_exit(ifp->if_serializer);
759 	return (0);
760 }
761 
762 static int
763 em_resume(device_t dev)
764 {
765 	struct adapter *adapter = device_get_softc(dev);
766 	struct ifnet *ifp = &adapter->interface_data.ac_if;
767 
768 	lwkt_serialize_enter(ifp->if_serializer);
769 	ifp->if_flags &= ~IFF_RUNNING;
770 	em_init(adapter);
771 	if ((ifp->if_flags & (IFF_UP | IFF_RUNNING)) == (IFF_UP | IFF_RUNNING))
772 		em_start(ifp);
773 	lwkt_serialize_exit(ifp->if_serializer);
774 
775 	return bus_generic_resume(dev);
776 }
777 
778 /*********************************************************************
779  *  Transmit entry point
780  *
781  *  em_start is called by the stack to initiate a transmit.
782  *  The driver will remain in this routine as long as there are
783  *  packets to transmit and transmit resources are available.
784  *  In case resources are not available stack is notified and
785  *  the packet is requeued.
786  **********************************************************************/
787 
788 static void
789 em_start(struct ifnet *ifp)
790 {
791 	struct mbuf *m_head;
792 	struct adapter *adapter = ifp->if_softc;
793 
794 	ASSERT_SERIALIZED(ifp->if_serializer);
795 
796 	if ((ifp->if_flags & (IFF_RUNNING | IFF_OACTIVE)) != IFF_RUNNING)
797 		return;
798 	if (!adapter->link_active)
799 		return;
800 	while (!ifq_is_empty(&ifp->if_snd)) {
801 		m_head = ifq_poll(&ifp->if_snd);
802 
803 		if (m_head == NULL)
804 			break;
805 
806 		logif(pkt_txqueue);
807 		if (em_encap(adapter, m_head)) {
808 			ifp->if_flags |= IFF_OACTIVE;
809 			break;
810 		}
811 		ifq_dequeue(&ifp->if_snd, m_head);
812 
813 		/* Send a copy of the frame to the BPF listener */
814 		BPF_MTAP(ifp, m_head);
815 
816 		/* Set timeout in case hardware has problems transmitting. */
817 		ifp->if_timer = EM_TX_TIMEOUT;
818 	}
819 }
820 
821 /*********************************************************************
822  *  Ioctl entry point
823  *
824  *  em_ioctl is called when the user wants to configure the
825  *  interface.
826  *
827  *  return 0 on success, positive on failure
828  **********************************************************************/
829 
830 static int
831 em_ioctl(struct ifnet *ifp, u_long command, caddr_t data, struct ucred *cr)
832 {
833 	int max_frame_size, mask, error = 0, reinit = 0;
834 	struct ifreq *ifr = (struct ifreq *) data;
835 	struct adapter *adapter = ifp->if_softc;
836 	uint16_t eeprom_data = 0;
837 
838 	ASSERT_SERIALIZED(ifp->if_serializer);
839 
840 	if (adapter->in_detach)
841 		return 0;
842 
843 	switch (command) {
844 	case SIOCSIFMTU:
845 		IOCTL_DEBUGOUT("ioctl rcv'd: SIOCSIFMTU (Set Interface MTU)");
846 		switch (adapter->hw.mac_type) {
847 		case em_82573:
848 			/*
849 			 * 82573 only supports jumbo frames
850 			 * if ASPM is disabled.
851 			 */
852 			em_read_eeprom(&adapter->hw, EEPROM_INIT_3GIO_3,
853 			    1, &eeprom_data);
854 			if (eeprom_data & EEPROM_WORD1A_ASPM_MASK) {
855 				max_frame_size = ETHER_MAX_LEN;
856 				break;
857 			}
858 			/* Allow Jumbo frames */
859 			/* FALLTHROUGH */
860 		case em_82571:
861 		case em_82572:
862 		case em_80003es2lan:	/* Limit Jumbo Frame size */
863 			max_frame_size = 9234;
864 			break;
865 		case em_ich8lan:
866 			/* ICH8 does not support jumbo frames */
867 			max_frame_size = ETHER_MAX_LEN;
868 			break;
869 		default:
870 			max_frame_size = MAX_JUMBO_FRAME_SIZE;
871 			break;
872 		}
873 		if (ifr->ifr_mtu >
874 			max_frame_size - ETHER_HDR_LEN - ETHER_CRC_LEN) {
875 			error = EINVAL;
876 		} else {
877 			ifp->if_mtu = ifr->ifr_mtu;
878 			adapter->hw.max_frame_size =
879 			ifp->if_mtu + ETHER_HDR_LEN + ETHER_CRC_LEN;
880 			ifp->if_flags &= ~IFF_RUNNING;
881 			em_init(adapter);
882 		}
883 		break;
884 	case SIOCSIFFLAGS:
885 		IOCTL_DEBUGOUT("ioctl rcv'd: SIOCSIFFLAGS "
886 			       "(Set Interface Flags)");
887 		if (ifp->if_flags & IFF_UP) {
888 			if (!(ifp->if_flags & IFF_RUNNING)) {
889 				em_init(adapter);
890 			} else if ((ifp->if_flags ^ adapter->if_flags) &
891 				   IFF_PROMISC) {
892 				em_disable_promisc(adapter);
893 				em_set_promisc(adapter);
894 			}
895 		} else {
896 			if (ifp->if_flags & IFF_RUNNING)
897 				em_stop(adapter);
898 		}
899 		adapter->if_flags = ifp->if_flags;
900 		break;
901 	case SIOCADDMULTI:
902 	case SIOCDELMULTI:
903 		IOCTL_DEBUGOUT("ioctl rcv'd: SIOC(ADD|DEL)MULTI");
904 		if (ifp->if_flags & IFF_RUNNING) {
905 			em_disable_intr(adapter);
906 			em_set_multi(adapter);
907 			if (adapter->hw.mac_type == em_82542_rev2_0)
908 				em_initialize_receive_unit(adapter);
909 #ifdef DEVICE_POLLING
910 			/* Do not enable interrupt if polling(4) is enabled */
911 			if ((ifp->if_flags & IFF_POLLING) == 0)
912 #endif
913 			em_enable_intr(adapter);
914 		}
915 		break;
916 	case SIOCSIFMEDIA:
917 		/* Check SOL/IDER usage */
918 		if (em_check_phy_reset_block(&adapter->hw)) {
919 			if_printf(ifp, "Media change is blocked due to "
920 				  "SOL/IDER session.\n");
921 			break;
922 		}
923 		/* FALLTHROUGH */
924 	case SIOCGIFMEDIA:
925 		IOCTL_DEBUGOUT("ioctl rcv'd: SIOCxIFMEDIA "
926 			       "(Get/Set Interface Media)");
927 		error = ifmedia_ioctl(ifp, ifr, &adapter->media, command);
928 		break;
929 	case SIOCSIFCAP:
930 		IOCTL_DEBUGOUT("ioctl rcv'd: SIOCSIFCAP (Set Capabilities)");
931 		mask = ifr->ifr_reqcap ^ ifp->if_capenable;
932 		if (mask & IFCAP_HWCSUM) {
933 			ifp->if_capenable ^= IFCAP_HWCSUM;
934 			reinit = 1;
935 		}
936 		if (mask & IFCAP_VLAN_HWTAGGING) {
937 			ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
938 			reinit = 1;
939 		}
940 		if (reinit && (ifp->if_flags & IFF_RUNNING)) {
941 			ifp->if_flags &= ~IFF_RUNNING;
942 			em_init(adapter);
943 		}
944 		break;
945 	default:
946 		error = ether_ioctl(ifp, command, data);
947 		break;
948 	}
949 
950 	return (error);
951 }
952 
953 /*********************************************************************
954  *  Watchdog entry point
955  *
956  *  This routine is called whenever hardware quits transmitting.
957  *
958  **********************************************************************/
959 
960 static void
961 em_watchdog(struct ifnet *ifp)
962 {
963 	struct adapter *adapter = ifp->if_softc;
964 
965 	/*
966 	 * If we are in this routine because of pause frames, then
967 	 * don't reset the hardware.
968 	 */
969 	if (E1000_READ_REG(&adapter->hw, STATUS) & E1000_STATUS_TXOFF) {
970 		ifp->if_timer = EM_TX_TIMEOUT;
971 		return;
972 	}
973 
974 	if (em_check_for_link(&adapter->hw) == 0)
975 		if_printf(ifp, "watchdog timeout -- resetting\n");
976 
977 	ifp->if_flags &= ~IFF_RUNNING;
978 	em_init(adapter);
979 
980 	adapter->watchdog_timeouts++;
981 }
982 
983 /*********************************************************************
984  *  Init entry point
985  *
986  *  This routine is used in two ways. It is used by the stack as
987  *  init entry point in network interface structure. It is also used
988  *  by the driver as a hw/sw initialization routine to get to a
989  *  consistent state.
990  *
991  *  return 0 on success, positive on failure
992  **********************************************************************/
993 
994 static void
995 em_init(void *arg)
996 {
997 	struct adapter *adapter = arg;
998 	uint32_t pba;
999 	struct ifnet *ifp = &adapter->interface_data.ac_if;
1000 
1001 	ASSERT_SERIALIZED(ifp->if_serializer);
1002 
1003 	INIT_DEBUGOUT("em_init: begin");
1004 
1005 	if (ifp->if_flags & IFF_RUNNING)
1006 		return;
1007 
1008 	em_stop(adapter);
1009 
1010 	/*
1011 	 * Packet Buffer Allocation (PBA)
1012 	 * Writing PBA sets the receive portion of the buffer
1013 	 * the remainder is used for the transmit buffer.
1014 	 *
1015 	 * Devices before the 82547 had a Packet Buffer of 64K.
1016 	 *   Default allocation: PBA=48K for Rx, leaving 16K for Tx.
1017 	 * After the 82547 the buffer was reduced to 40K.
1018 	 *   Default allocation: PBA=30K for Rx, leaving 10K for Tx.
1019 	 *   Note: default does not leave enough room for Jumbo Frame >10k.
1020 	 */
1021 	switch (adapter->hw.mac_type) {
1022 	case em_82547:
1023 	case em_82547_rev_2: /* 82547: Total Packet Buffer is 40K */
1024 		if (adapter->hw.max_frame_size > EM_RXBUFFER_8192)
1025 			pba = E1000_PBA_22K; /* 22K for Rx, 18K for Tx */
1026 		else
1027 			pba = E1000_PBA_30K; /* 30K for Rx, 10K for Tx */
1028 
1029 		adapter->tx_fifo_head = 0;
1030 		adapter->tx_head_addr = pba << EM_TX_HEAD_ADDR_SHIFT;
1031 		adapter->tx_fifo_size =
1032 			(E1000_PBA_40K - pba) << EM_PBA_BYTES_SHIFT;
1033 		break;
1034 	/* Total Packet Buffer on these is 48K */
1035 	case em_82571:
1036 	case em_82572:
1037 	case em_80003es2lan:
1038 		pba = E1000_PBA_32K; /* 32K for Rx, 16K for Tx */
1039 		break;
1040 	case em_82573: /* 82573: Total Packet Buffer is 32K */
1041 		pba = E1000_PBA_12K; /* 12K for Rx, 20K for Tx */
1042 		break;
1043 	case em_ich8lan:
1044 		pba = E1000_PBA_8K;
1045 		break;
1046 	default:
1047 		/* Devices before 82547 had a Packet Buffer of 64K.   */
1048 		if(adapter->hw.max_frame_size > EM_RXBUFFER_8192)
1049 			pba = E1000_PBA_40K; /* 40K for Rx, 24K for Tx */
1050 		else
1051 			pba = E1000_PBA_48K; /* 48K for Rx, 16K for Tx */
1052 	}
1053 
1054 	INIT_DEBUGOUT1("em_init: pba=%dK",pba);
1055 	E1000_WRITE_REG(&adapter->hw, PBA, pba);
1056 
1057 	/* Get the latest mac address, User can use a LAA */
1058 	bcopy(adapter->interface_data.ac_enaddr, adapter->hw.mac_addr,
1059 	      ETHER_ADDR_LEN);
1060 
1061 	/* Initialize the hardware */
1062 	if (em_hardware_init(adapter)) {
1063 		if_printf(ifp, "Unable to initialize the hardware\n");
1064 		return;
1065 	}
1066 	em_update_link_status(adapter);
1067 
1068 	if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING)
1069 		em_enable_vlans(adapter);
1070 
1071 	/* Set hardware offload abilities */
1072 	if (adapter->hw.mac_type >= em_82543) {
1073 		if (ifp->if_capenable & IFCAP_TXCSUM)
1074 			ifp->if_hwassist = EM_CHECKSUM_FEATURES;
1075 		else
1076 			ifp->if_hwassist = 0;
1077 	}
1078 
1079 	/* Prepare transmit descriptors and buffers */
1080 	if (em_setup_transmit_structures(adapter)) {
1081 		if_printf(ifp, "Could not setup transmit structures\n");
1082 		em_stop(adapter);
1083 		return;
1084 	}
1085 	em_initialize_transmit_unit(adapter);
1086 
1087 	/* Setup Multicast table */
1088 	em_set_multi(adapter);
1089 
1090 	/* Prepare receive descriptors and buffers */
1091 	if (em_setup_receive_structures(adapter)) {
1092 		if_printf(ifp, "Could not setup receive structures\n");
1093 		em_stop(adapter);
1094 		return;
1095 	}
1096 	em_initialize_receive_unit(adapter);
1097 
1098 	/* Don't lose promiscuous settings */
1099 	em_set_promisc(adapter);
1100 
1101 	ifp->if_flags |= IFF_RUNNING;
1102 	ifp->if_flags &= ~IFF_OACTIVE;
1103 
1104 	callout_reset(&adapter->timer, hz, em_local_timer, adapter);
1105 	em_clear_hw_cntrs(&adapter->hw);
1106 
1107 #ifdef DEVICE_POLLING
1108 	/* Do not enable interrupt if polling(4) is enabled */
1109 	if (ifp->if_flags & IFF_POLLING)
1110 		em_disable_intr(adapter);
1111 	else
1112 #endif
1113 	em_enable_intr(adapter);
1114 
1115 	/* Don't reset the phy next time init gets called */
1116 	adapter->hw.phy_reset_disable = TRUE;
1117 }
1118 
1119 #ifdef DEVICE_POLLING
1120 
1121 static void
1122 em_poll(struct ifnet *ifp, enum poll_cmd cmd, int count)
1123 {
1124 	struct adapter *adapter = ifp->if_softc;
1125 	uint32_t reg_icr;
1126 
1127 	logif(poll_beg);
1128 
1129 	ASSERT_SERIALIZED(ifp->if_serializer);
1130 
1131 	switch(cmd) {
1132 	case POLL_REGISTER:
1133 		em_disable_intr(adapter);
1134 		break;
1135 	case POLL_DEREGISTER:
1136 		em_enable_intr(adapter);
1137 		break;
1138 	case POLL_AND_CHECK_STATUS:
1139 		reg_icr = E1000_READ_REG(&adapter->hw, ICR);
1140 		if (reg_icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1141 			callout_stop(&adapter->timer);
1142 			adapter->hw.get_link_status = 1;
1143 			em_check_for_link(&adapter->hw);
1144 			em_update_link_status(adapter);
1145 			callout_reset(&adapter->timer, hz, em_local_timer,
1146 				      adapter);
1147 		}
1148 		/* fall through */
1149 	case POLL_ONLY:
1150 		if (ifp->if_flags & IFF_RUNNING) {
1151 			em_rxeof(adapter, count);
1152 			em_txeof(adapter);
1153 
1154 			if (!ifq_is_empty(&ifp->if_snd))
1155 				em_start(ifp);
1156 		}
1157 		break;
1158 	}
1159 	logif(poll_end);
1160 }
1161 
1162 #endif /* DEVICE_POLLING */
1163 
1164 /*********************************************************************
1165  *
1166  *  Interrupt Service routine
1167  *
1168  *********************************************************************/
1169 static void
1170 em_intr(void *arg)
1171 {
1172 	uint32_t reg_icr;
1173 	struct ifnet *ifp;
1174 	struct adapter *adapter = arg;
1175 
1176 	ifp = &adapter->interface_data.ac_if;
1177 
1178 	logif(intr_beg);
1179 	ASSERT_SERIALIZED(ifp->if_serializer);
1180 
1181 	reg_icr = E1000_READ_REG(&adapter->hw, ICR);
1182 	if ((adapter->hw.mac_type >= em_82571 &&
1183 	     (reg_icr & E1000_ICR_INT_ASSERTED) == 0) ||
1184 	    reg_icr == 0) {
1185 		logif(intr_end);
1186 		return;
1187 	}
1188 
1189 	/*
1190 	 * XXX: some laptops trigger several spurious interrupts on em(4)
1191 	 * when in the resume cycle. The ICR register reports all-ones
1192 	 * value in this case. Processing such interrupts would lead to
1193 	 * a freeze. I don't know why.
1194 	 */
1195 	if (reg_icr == 0xffffffff) {
1196 		logif(intr_end);
1197 		return;
1198 	}
1199 
1200 	/*
1201 	 * note: do not attempt to improve efficiency by looping.  This
1202 	 * only results in unnecessary piecemeal collection of received
1203 	 * packets and unnecessary piecemeal cleanups of the transmit ring.
1204 	 */
1205 	if (ifp->if_flags & IFF_RUNNING) {
1206 		em_rxeof(adapter, -1);
1207 		em_txeof(adapter);
1208 	}
1209 
1210 	/* Link status change */
1211 	if (reg_icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
1212 		callout_stop(&adapter->timer);
1213 		adapter->hw.get_link_status = 1;
1214 		em_check_for_link(&adapter->hw);
1215 		em_update_link_status(adapter);
1216 		callout_reset(&adapter->timer, hz, em_local_timer, adapter);
1217 	}
1218 
1219 	if (reg_icr & E1000_ICR_RXO)
1220 		adapter->rx_overruns++;
1221 
1222 	if ((ifp->if_flags & IFF_RUNNING) && !ifq_is_empty(&ifp->if_snd))
1223 		em_start(ifp);
1224 	logif(intr_end);
1225 }
1226 
1227 /*********************************************************************
1228  *
1229  *  Media Ioctl callback
1230  *
1231  *  This routine is called whenever the user queries the status of
1232  *  the interface using ifconfig.
1233  *
1234  **********************************************************************/
1235 static void
1236 em_media_status(struct ifnet *ifp, struct ifmediareq *ifmr)
1237 {
1238 	struct adapter *adapter = ifp->if_softc;
1239 	u_char fiber_type = IFM_1000_SX;
1240 
1241 	INIT_DEBUGOUT("em_media_status: begin");
1242 
1243 	ASSERT_SERIALIZED(ifp->if_serializer);
1244 
1245 	em_check_for_link(&adapter->hw);
1246 	em_update_link_status(adapter);
1247 
1248 	ifmr->ifm_status = IFM_AVALID;
1249 	ifmr->ifm_active = IFM_ETHER;
1250 
1251 	if (!adapter->link_active)
1252 		return;
1253 
1254 	ifmr->ifm_status |= IFM_ACTIVE;
1255 
1256 	if (adapter->hw.media_type == em_media_type_fiber ||
1257 	    adapter->hw.media_type == em_media_type_internal_serdes) {
1258 		if (adapter->hw.mac_type == em_82545)
1259 			fiber_type = IFM_1000_LX;
1260 		ifmr->ifm_active |= fiber_type | IFM_FDX;
1261 	} else {
1262 		switch (adapter->link_speed) {
1263 		case 10:
1264 			ifmr->ifm_active |= IFM_10_T;
1265 			break;
1266 		case 100:
1267 			ifmr->ifm_active |= IFM_100_TX;
1268 			break;
1269 		case 1000:
1270 			ifmr->ifm_active |= IFM_1000_T;
1271 			break;
1272 		}
1273 		if (adapter->link_duplex == FULL_DUPLEX)
1274 			ifmr->ifm_active |= IFM_FDX;
1275 		else
1276 			ifmr->ifm_active |= IFM_HDX;
1277 	}
1278 }
1279 
1280 /*********************************************************************
1281  *
1282  *  Media Ioctl callback
1283  *
1284  *  This routine is called when the user changes speed/duplex using
1285  *  media/mediopt option with ifconfig.
1286  *
1287  **********************************************************************/
1288 static int
1289 em_media_change(struct ifnet *ifp)
1290 {
1291 	struct adapter *adapter = ifp->if_softc;
1292 	struct ifmedia *ifm = &adapter->media;
1293 
1294 	INIT_DEBUGOUT("em_media_change: begin");
1295 
1296 	ASSERT_SERIALIZED(ifp->if_serializer);
1297 
1298 	if (IFM_TYPE(ifm->ifm_media) != IFM_ETHER)
1299 		return (EINVAL);
1300 
1301 	switch (IFM_SUBTYPE(ifm->ifm_media)) {
1302 	case IFM_AUTO:
1303 		adapter->hw.autoneg = DO_AUTO_NEG;
1304 		adapter->hw.autoneg_advertised = AUTONEG_ADV_DEFAULT;
1305 		break;
1306 	case IFM_1000_LX:
1307 	case IFM_1000_SX:
1308 	case IFM_1000_T:
1309 		adapter->hw.autoneg = DO_AUTO_NEG;
1310 		adapter->hw.autoneg_advertised = ADVERTISE_1000_FULL;
1311 		break;
1312 	case IFM_100_TX:
1313 		adapter->hw.autoneg = FALSE;
1314 		adapter->hw.autoneg_advertised = 0;
1315 		if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX)
1316 			adapter->hw.forced_speed_duplex = em_100_full;
1317 		else
1318 			adapter->hw.forced_speed_duplex = em_100_half;
1319 		break;
1320 	case IFM_10_T:
1321 		adapter->hw.autoneg = FALSE;
1322 		adapter->hw.autoneg_advertised = 0;
1323 		if ((ifm->ifm_media & IFM_GMASK) == IFM_FDX)
1324 			adapter->hw.forced_speed_duplex = em_10_full;
1325 		else
1326 			adapter->hw.forced_speed_duplex = em_10_half;
1327 		break;
1328 	default:
1329 		if_printf(ifp, "Unsupported media type\n");
1330 	}
1331 	/*
1332 	 * As the speed/duplex settings may have changed we need to
1333 	 * reset the PHY.
1334 	 */
1335 	adapter->hw.phy_reset_disable = FALSE;
1336 
1337 	ifp->if_flags &= ~IFF_RUNNING;
1338 	em_init(adapter);
1339 
1340 	return(0);
1341 }
1342 
1343 static void
1344 em_tx_cb(void *arg, bus_dma_segment_t *seg, int nsegs, bus_size_t mapsize,
1345 	 int error)
1346 {
1347 	struct em_q *q = arg;
1348 
1349 	if (error)
1350 		return;
1351 	KASSERT(nsegs <= EM_MAX_SCATTER,
1352 		("Too many DMA segments returned when mapping tx packet"));
1353 	q->nsegs = nsegs;
1354 	bcopy(seg, q->segs, nsegs * sizeof(seg[0]));
1355 }
1356 
1357 /*********************************************************************
1358  *
1359  *  This routine maps the mbufs to tx descriptors.
1360  *
1361  *  return 0 on success, positive on failure
1362  **********************************************************************/
1363 static int
1364 em_encap(struct adapter *adapter, struct mbuf *m_head)
1365 {
1366 	uint32_t txd_upper = 0, txd_lower = 0, txd_used = 0, txd_saved = 0;
1367 	int i, j, error, last = 0;
1368 
1369 	struct ifvlan *ifv = NULL;
1370 	struct em_q q;
1371 	struct em_buffer *tx_buffer = NULL, *tx_buffer_first;
1372 	bus_dmamap_t map;
1373 	struct em_tx_desc *current_tx_desc = NULL;
1374 	struct ifnet *ifp = &adapter->interface_data.ac_if;
1375 
1376 	/*
1377 	 * Force a cleanup if number of TX descriptors
1378 	 * available hits the threshold
1379 	 */
1380 	if (adapter->num_tx_desc_avail <= EM_TX_CLEANUP_THRESHOLD) {
1381 		em_txeof(adapter);
1382 		if (adapter->num_tx_desc_avail <= EM_TX_CLEANUP_THRESHOLD) {
1383 			adapter->no_tx_desc_avail1++;
1384 			return (ENOBUFS);
1385 		}
1386 	}
1387 
1388 	/*
1389 	 * Capture the first descriptor index, this descriptor will have
1390 	 * the index of the EOP which is the only one that now gets a
1391 	 * DONE bit writeback.
1392 	 */
1393 	tx_buffer_first = &adapter->tx_buffer_area[adapter->next_avail_tx_desc];
1394 
1395 	/*
1396 	 * Map the packet for DMA.
1397 	 */
1398 	map = tx_buffer_first->map;
1399 	error = bus_dmamap_load_mbuf(adapter->txtag, map, m_head, em_tx_cb,
1400 				     &q, BUS_DMA_NOWAIT);
1401 	if (error != 0) {
1402 		adapter->no_tx_dma_setup++;
1403 		return (error);
1404 	}
1405 	KASSERT(q.nsegs != 0, ("em_encap: empty packet"));
1406 
1407 	if (q.nsegs > (adapter->num_tx_desc_avail - 2)) {
1408 		adapter->no_tx_desc_avail2++;
1409 		error = ENOBUFS;
1410 		goto fail;
1411 	}
1412 
1413 	if (ifp->if_hwassist > 0) {
1414 		em_transmit_checksum_setup(adapter,  m_head,
1415 					   &txd_upper, &txd_lower);
1416 	}
1417 
1418 	/* Find out if we are in vlan mode */
1419 	if ((m_head->m_flags & (M_PROTO1|M_PKTHDR)) == (M_PROTO1|M_PKTHDR) &&
1420 	    m_head->m_pkthdr.rcvif != NULL &&
1421 	    m_head->m_pkthdr.rcvif->if_type == IFT_L2VLAN)
1422 		ifv = m_head->m_pkthdr.rcvif->if_softc;
1423 
1424 	i = adapter->next_avail_tx_desc;
1425 	if (adapter->pcix_82544)
1426 		txd_saved = i;
1427 
1428 	/* Set up our transmit descriptors */
1429 	for (j = 0; j < q.nsegs; j++) {
1430 		/* If adapter is 82544 and on PCIX bus */
1431 		if(adapter->pcix_82544) {
1432 			DESC_ARRAY desc_array;
1433 			uint32_t array_elements, counter;
1434 
1435 			/*
1436 			 * Check the Address and Length combination and
1437 			 * split the data accordingly
1438 			 */
1439 			array_elements = em_fill_descriptors(q.segs[j].ds_addr,
1440 						q.segs[j].ds_len, &desc_array);
1441 			for (counter = 0; counter < array_elements; counter++) {
1442 				if (txd_used == adapter->num_tx_desc_avail) {
1443 					adapter->next_avail_tx_desc = txd_saved;
1444 					adapter->no_tx_desc_avail2++;
1445 					error = ENOBUFS;
1446 					goto fail;
1447 				}
1448 				tx_buffer = &adapter->tx_buffer_area[i];
1449 				current_tx_desc = &adapter->tx_desc_base[i];
1450 				current_tx_desc->buffer_addr = htole64(
1451 					desc_array.descriptor[counter].address);
1452 				current_tx_desc->lower.data = htole32(
1453 					adapter->txd_cmd | txd_lower |
1454 					(uint16_t)desc_array.descriptor[counter].length);
1455 				current_tx_desc->upper.data = htole32(txd_upper);
1456 
1457 				last = i;
1458 				if (++i == adapter->num_tx_desc)
1459 					i = 0;
1460 
1461 				tx_buffer->m_head = NULL;
1462 				tx_buffer->next_eop = -1;
1463 				txd_used++;
1464 			}
1465 		} else {
1466 			tx_buffer = &adapter->tx_buffer_area[i];
1467 			current_tx_desc = &adapter->tx_desc_base[i];
1468 
1469 			current_tx_desc->buffer_addr = htole64(q.segs[j].ds_addr);
1470 			current_tx_desc->lower.data = htole32(
1471 				adapter->txd_cmd | txd_lower | q.segs[j].ds_len);
1472 			current_tx_desc->upper.data = htole32(txd_upper);
1473 
1474 			last = i;
1475 			if (++i == adapter->num_tx_desc)
1476 				i = 0;
1477 
1478 			tx_buffer->m_head = NULL;
1479 			tx_buffer->next_eop = -1;
1480 		}
1481 	}
1482 
1483 	adapter->next_avail_tx_desc = i;
1484 	if (adapter->pcix_82544)
1485 		adapter->num_tx_desc_avail -= txd_used;
1486 	else
1487 		adapter->num_tx_desc_avail -= q.nsegs;
1488 
1489 	if (ifv != NULL) {
1490 		/* Set the vlan id */
1491 		current_tx_desc->upper.fields.special = htole16(ifv->ifv_tag);
1492 
1493 		/* Tell hardware to add tag */
1494 		current_tx_desc->lower.data |= htole32(E1000_TXD_CMD_VLE);
1495 	}
1496 
1497 	tx_buffer->m_head = m_head;
1498 	tx_buffer_first->map = tx_buffer->map;
1499 	tx_buffer->map = map;
1500 	bus_dmamap_sync(adapter->txtag, map, BUS_DMASYNC_PREWRITE);
1501 
1502 	/*
1503 	 * Last Descriptor of Packet needs End Of Packet (EOP)
1504 	 * and Report Status (RS)
1505 	 */
1506 	current_tx_desc->lower.data |=
1507 		htole32(E1000_TXD_CMD_EOP | E1000_TXD_CMD_RS);
1508 
1509 	/*
1510 	 * Keep track in the first buffer which descriptor will be
1511 	 * written back.
1512 	 */
1513 	tx_buffer_first->next_eop = last;
1514 
1515 	bus_dmamap_sync(adapter->txdma.dma_tag, adapter->txdma.dma_map,
1516 			BUS_DMASYNC_PREWRITE);
1517 
1518 	/*
1519 	 * Advance the Transmit Descriptor Tail (Tdt), this tells the E1000
1520 	 * that this frame is available to transmit.
1521 	 */
1522 	if (adapter->hw.mac_type == em_82547 &&
1523 	    adapter->link_duplex == HALF_DUPLEX) {
1524 		em_82547_move_tail_serialized(adapter);
1525 	} else {
1526 		E1000_WRITE_REG(&adapter->hw, TDT, i);
1527 		if (adapter->hw.mac_type == em_82547) {
1528 			em_82547_update_fifo_head(adapter,
1529 						  m_head->m_pkthdr.len);
1530 		}
1531 	}
1532 
1533 	return (0);
1534 fail:
1535 	bus_dmamap_unload(adapter->txtag, map);
1536 	return error;
1537 }
1538 
1539 /*********************************************************************
1540  *
1541  * 82547 workaround to avoid controller hang in half-duplex environment.
1542  * The workaround is to avoid queuing a large packet that would span
1543  * the internal Tx FIFO ring boundary. We need to reset the FIFO pointers
1544  * in this case. We do that only when FIFO is quiescent.
1545  *
1546  **********************************************************************/
1547 static void
1548 em_82547_move_tail(void *arg)
1549 {
1550 	struct adapter *adapter = arg;
1551 	struct ifnet *ifp = &adapter->interface_data.ac_if;
1552 
1553 	lwkt_serialize_enter(ifp->if_serializer);
1554 	em_82547_move_tail_serialized(adapter);
1555 	lwkt_serialize_exit(ifp->if_serializer);
1556 }
1557 
1558 static void
1559 em_82547_move_tail_serialized(struct adapter *adapter)
1560 {
1561 	uint16_t hw_tdt;
1562 	uint16_t sw_tdt;
1563 	struct em_tx_desc *tx_desc;
1564 	uint16_t length = 0;
1565 	boolean_t eop = 0;
1566 
1567 	hw_tdt = E1000_READ_REG(&adapter->hw, TDT);
1568 	sw_tdt = adapter->next_avail_tx_desc;
1569 
1570 	while (hw_tdt != sw_tdt) {
1571 		tx_desc = &adapter->tx_desc_base[hw_tdt];
1572 		length += tx_desc->lower.flags.length;
1573 		eop = tx_desc->lower.data & E1000_TXD_CMD_EOP;
1574 		if (++hw_tdt == adapter->num_tx_desc)
1575 			hw_tdt = 0;
1576 
1577 		if (eop) {
1578 			if (em_82547_fifo_workaround(adapter, length)) {
1579 				adapter->tx_fifo_wrk_cnt++;
1580 				callout_reset(&adapter->tx_fifo_timer, 1,
1581 					em_82547_move_tail, adapter);
1582 				break;
1583 			}
1584 			E1000_WRITE_REG(&adapter->hw, TDT, hw_tdt);
1585 			em_82547_update_fifo_head(adapter, length);
1586 			length = 0;
1587 		}
1588 	}
1589 }
1590 
1591 static int
1592 em_82547_fifo_workaround(struct adapter *adapter, int len)
1593 {
1594 	int fifo_space, fifo_pkt_len;
1595 
1596 	fifo_pkt_len = roundup2(len + EM_FIFO_HDR, EM_FIFO_HDR);
1597 
1598 	if (adapter->link_duplex == HALF_DUPLEX) {
1599 		fifo_space = adapter->tx_fifo_size - adapter->tx_fifo_head;
1600 
1601 		if (fifo_pkt_len >= (EM_82547_PKT_THRESH + fifo_space)) {
1602 			if (em_82547_tx_fifo_reset(adapter))
1603 				return (0);
1604 			else
1605 				return (1);
1606 		}
1607 	}
1608 
1609 	return (0);
1610 }
1611 
1612 static void
1613 em_82547_update_fifo_head(struct adapter *adapter, int len)
1614 {
1615 	int fifo_pkt_len = roundup2(len + EM_FIFO_HDR, EM_FIFO_HDR);
1616 
1617 	/* tx_fifo_head is always 16 byte aligned */
1618 	adapter->tx_fifo_head += fifo_pkt_len;
1619 	if (adapter->tx_fifo_head >= adapter->tx_fifo_size)
1620 		adapter->tx_fifo_head -= adapter->tx_fifo_size;
1621 }
1622 
1623 static int
1624 em_82547_tx_fifo_reset(struct adapter *adapter)
1625 {
1626 	uint32_t tctl;
1627 
1628 	if (E1000_READ_REG(&adapter->hw, TDT) == E1000_READ_REG(&adapter->hw, TDH) &&
1629 	    E1000_READ_REG(&adapter->hw, TDFT) == E1000_READ_REG(&adapter->hw, TDFH) &&
1630 	    E1000_READ_REG(&adapter->hw, TDFTS) == E1000_READ_REG(&adapter->hw, TDFHS) &&
1631 	    E1000_READ_REG(&adapter->hw, TDFPC) == 0) {
1632 		/* Disable TX unit */
1633 		tctl = E1000_READ_REG(&adapter->hw, TCTL);
1634 		E1000_WRITE_REG(&adapter->hw, TCTL, tctl & ~E1000_TCTL_EN);
1635 
1636 		/* Reset FIFO pointers */
1637 		E1000_WRITE_REG(&adapter->hw, TDFT,  adapter->tx_head_addr);
1638 		E1000_WRITE_REG(&adapter->hw, TDFH,  adapter->tx_head_addr);
1639 		E1000_WRITE_REG(&adapter->hw, TDFTS, adapter->tx_head_addr);
1640 		E1000_WRITE_REG(&adapter->hw, TDFHS, adapter->tx_head_addr);
1641 
1642 		/* Re-enable TX unit */
1643 		E1000_WRITE_REG(&adapter->hw, TCTL, tctl);
1644 		E1000_WRITE_FLUSH(&adapter->hw);
1645 
1646 		adapter->tx_fifo_head = 0;
1647 		adapter->tx_fifo_reset_cnt++;
1648 
1649 		return (TRUE);
1650 	} else {
1651 		return (FALSE);
1652 	}
1653 }
1654 
1655 static void
1656 em_set_promisc(struct adapter *adapter)
1657 {
1658 	uint32_t reg_rctl;
1659 	struct ifnet *ifp = &adapter->interface_data.ac_if;
1660 
1661 	reg_rctl = E1000_READ_REG(&adapter->hw, RCTL);
1662 
1663 	adapter->em_insert_vlan_header = 0;
1664 	if (ifp->if_flags & IFF_PROMISC) {
1665 		reg_rctl |= (E1000_RCTL_UPE | E1000_RCTL_MPE);
1666 		E1000_WRITE_REG(&adapter->hw, RCTL, reg_rctl);
1667 
1668 		/*
1669 		 * Disable VLAN stripping in promiscous mode.
1670 		 * This enables bridging of vlan tagged frames to occur
1671 		 * and also allows vlan tags to be seen in tcpdump.
1672 		 */
1673 		if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING)
1674 			em_disable_vlans(adapter);
1675 		adapter->em_insert_vlan_header = 1;
1676 	} else if (ifp->if_flags & IFF_ALLMULTI) {
1677 		reg_rctl |= E1000_RCTL_MPE;
1678 		reg_rctl &= ~E1000_RCTL_UPE;
1679 		E1000_WRITE_REG(&adapter->hw, RCTL, reg_rctl);
1680 	}
1681 }
1682 
1683 static void
1684 em_disable_promisc(struct adapter *adapter)
1685 {
1686 	struct ifnet *ifp = &adapter->interface_data.ac_if;
1687 
1688 	uint32_t reg_rctl;
1689 
1690 	reg_rctl = E1000_READ_REG(&adapter->hw, RCTL);
1691 
1692 	reg_rctl &= (~E1000_RCTL_UPE);
1693 	reg_rctl &= (~E1000_RCTL_MPE);
1694 	E1000_WRITE_REG(&adapter->hw, RCTL, reg_rctl);
1695 
1696 	if (ifp->if_capenable & IFCAP_VLAN_HWTAGGING)
1697 		em_enable_vlans(adapter);
1698 	adapter->em_insert_vlan_header = 0;
1699 }
1700 
1701 /*********************************************************************
1702  *  Multicast Update
1703  *
1704  *  This routine is called whenever multicast address list is updated.
1705  *
1706  **********************************************************************/
1707 
1708 static void
1709 em_set_multi(struct adapter *adapter)
1710 {
1711 	uint32_t reg_rctl = 0;
1712 	uint8_t mta[MAX_NUM_MULTICAST_ADDRESSES * ETH_LENGTH_OF_ADDRESS];
1713 	struct ifmultiaddr *ifma;
1714 	int mcnt = 0;
1715 	struct ifnet *ifp = &adapter->interface_data.ac_if;
1716 
1717 	IOCTL_DEBUGOUT("em_set_multi: begin");
1718 
1719 	if (adapter->hw.mac_type == em_82542_rev2_0) {
1720 		reg_rctl = E1000_READ_REG(&adapter->hw, RCTL);
1721 		if (adapter->hw.pci_cmd_word & CMD_MEM_WRT_INVALIDATE)
1722 			em_pci_clear_mwi(&adapter->hw);
1723 		reg_rctl |= E1000_RCTL_RST;
1724 		E1000_WRITE_REG(&adapter->hw, RCTL, reg_rctl);
1725 		msec_delay(5);
1726 	}
1727 
1728 	LIST_FOREACH(ifma, &ifp->if_multiaddrs, ifma_link) {
1729 		if (ifma->ifma_addr->sa_family != AF_LINK)
1730 			continue;
1731 
1732 		if (mcnt == MAX_NUM_MULTICAST_ADDRESSES)
1733 			break;
1734 
1735 		bcopy(LLADDR((struct sockaddr_dl *)ifma->ifma_addr),
1736 		      &mta[mcnt*ETH_LENGTH_OF_ADDRESS], ETH_LENGTH_OF_ADDRESS);
1737 		mcnt++;
1738 	}
1739 
1740 	if (mcnt >= MAX_NUM_MULTICAST_ADDRESSES) {
1741 		reg_rctl = E1000_READ_REG(&adapter->hw, RCTL);
1742 		reg_rctl |= E1000_RCTL_MPE;
1743 		E1000_WRITE_REG(&adapter->hw, RCTL, reg_rctl);
1744 	} else {
1745 		em_mc_addr_list_update(&adapter->hw, mta, mcnt, 0, 1);
1746 	}
1747 
1748 	if (adapter->hw.mac_type == em_82542_rev2_0) {
1749 		reg_rctl = E1000_READ_REG(&adapter->hw, RCTL);
1750 		reg_rctl &= ~E1000_RCTL_RST;
1751 		E1000_WRITE_REG(&adapter->hw, RCTL, reg_rctl);
1752 		msec_delay(5);
1753 		if (adapter->hw.pci_cmd_word & CMD_MEM_WRT_INVALIDATE)
1754                         em_pci_set_mwi(&adapter->hw);
1755 	}
1756 }
1757 
1758 /*********************************************************************
1759  *  Timer routine
1760  *
1761  *  This routine checks for link status and updates statistics.
1762  *
1763  **********************************************************************/
1764 
1765 static void
1766 em_local_timer(void *arg)
1767 {
1768 	struct ifnet *ifp;
1769 	struct adapter *adapter = arg;
1770 	ifp = &adapter->interface_data.ac_if;
1771 
1772 	lwkt_serialize_enter(ifp->if_serializer);
1773 
1774 	em_check_for_link(&adapter->hw);
1775 	em_update_link_status(adapter);
1776 	em_update_stats_counters(adapter);
1777 	if (em_display_debug_stats && ifp->if_flags & IFF_RUNNING)
1778 		em_print_hw_stats(adapter);
1779 	em_smartspeed(adapter);
1780 
1781 	callout_reset(&adapter->timer, hz, em_local_timer, adapter);
1782 
1783 	lwkt_serialize_exit(ifp->if_serializer);
1784 }
1785 
1786 static void
1787 em_update_link_status(struct adapter *adapter)
1788 {
1789 	if (E1000_READ_REG(&adapter->hw, STATUS) & E1000_STATUS_LU) {
1790 		if (adapter->link_active == 0) {
1791 			em_get_speed_and_duplex(&adapter->hw,
1792 						&adapter->link_speed,
1793 						&adapter->link_duplex);
1794 			/* Check if we may set SPEED_MODE bit on PCI-E */
1795 			if (adapter->link_speed == SPEED_1000 &&
1796 			    (adapter->hw.mac_type == em_82571 ||
1797 			     adapter->hw.mac_type == em_82572)) {
1798 				int tarc0;
1799 
1800 				tarc0 = E1000_READ_REG(&adapter->hw, TARC0);
1801 				tarc0 |= SPEED_MODE_BIT;
1802 				E1000_WRITE_REG(&adapter->hw, TARC0, tarc0);
1803 			}
1804 			if (bootverbose) {
1805 				if_printf(&adapter->interface_data.ac_if,
1806 					  "Link is up %d Mbps %s\n",
1807 					  adapter->link_speed,
1808 					  adapter->link_duplex == FULL_DUPLEX ?
1809 						"Full Duplex" : "Half Duplex");
1810 			}
1811 			adapter->link_active = 1;
1812 			adapter->smartspeed = 0;
1813 #ifdef notyet
1814 			ifp->if_baudrate = adapter->link_speed * 1000000;
1815 			if_link_state_change(ifp, LINK_STATE_UP);
1816 #endif
1817 		}
1818 	} else {
1819 		if (adapter->link_active == 1) {
1820 #ifdef notyet
1821 			ifp->if_baudrate = 0;
1822 #endif
1823 			adapter->link_speed = 0;
1824 			adapter->link_duplex = 0;
1825 			if (bootverbose) {
1826 				if_printf(&adapter->interface_data.ac_if,
1827 					  "Link is Down\n");
1828 			}
1829 			adapter->link_active = 0;
1830 #ifdef notyet
1831 			if_link_state_change(ifp, LINK_STATE_DOWN);
1832 #endif
1833 		}
1834 	}
1835 }
1836 
1837 /*********************************************************************
1838  *
1839  *  This routine disables all traffic on the adapter by issuing a
1840  *  global reset on the MAC and deallocates TX/RX buffers.
1841  *
1842  **********************************************************************/
1843 
1844 static void
1845 em_stop(void *arg)
1846 {
1847 	struct ifnet   *ifp;
1848 	struct adapter * adapter = arg;
1849 	ifp = &adapter->interface_data.ac_if;
1850 
1851 	ASSERT_SERIALIZED(ifp->if_serializer);
1852 
1853 	INIT_DEBUGOUT("em_stop: begin");
1854 	em_disable_intr(adapter);
1855 	em_reset_hw(&adapter->hw);
1856 	callout_stop(&adapter->timer);
1857 	callout_stop(&adapter->tx_fifo_timer);
1858 	em_free_transmit_structures(adapter);
1859 	em_free_receive_structures(adapter);
1860 
1861 	/* Tell the stack that the interface is no longer active */
1862 	ifp->if_flags &= ~(IFF_RUNNING | IFF_OACTIVE);
1863 	ifp->if_timer = 0;
1864 }
1865 
1866 /*********************************************************************
1867  *
1868  *  Determine hardware revision.
1869  *
1870  **********************************************************************/
1871 static void
1872 em_identify_hardware(struct adapter *adapter)
1873 {
1874 	device_t dev = adapter->dev;
1875 
1876 	/* Make sure our PCI config space has the necessary stuff set */
1877 	adapter->hw.pci_cmd_word = pci_read_config(dev, PCIR_COMMAND, 2);
1878 	if (!((adapter->hw.pci_cmd_word & PCIM_CMD_BUSMASTEREN) &&
1879 	      (adapter->hw.pci_cmd_word & PCIM_CMD_MEMEN))) {
1880 		device_printf(dev, "Memory Access and/or Bus Master bits "
1881 			      "were not set!\n");
1882 		adapter->hw.pci_cmd_word |= PCIM_CMD_BUSMASTEREN |
1883 					    PCIM_CMD_MEMEN;
1884 		pci_write_config(dev, PCIR_COMMAND,
1885 				 adapter->hw.pci_cmd_word, 2);
1886 	}
1887 
1888 	/* Save off the information about this board */
1889 	adapter->hw.vendor_id = pci_get_vendor(dev);
1890 	adapter->hw.device_id = pci_get_device(dev);
1891 	adapter->hw.revision_id = pci_get_revid(dev);
1892 	adapter->hw.subsystem_vendor_id = pci_get_subvendor(dev);
1893 	adapter->hw.subsystem_id = pci_get_subdevice(dev);
1894 
1895 	/* Identify the MAC */
1896 	if (em_set_mac_type(&adapter->hw))
1897 		device_printf(dev, "Unknown MAC Type\n");
1898 
1899 	if (adapter->hw.mac_type == em_82541 ||
1900 	    adapter->hw.mac_type == em_82541_rev_2 ||
1901 	    adapter->hw.mac_type == em_82547 ||
1902 	    adapter->hw.mac_type == em_82547_rev_2)
1903 		adapter->hw.phy_init_script = TRUE;
1904 }
1905 
1906 static int
1907 em_allocate_pci_resources(device_t dev)
1908 {
1909 	struct adapter *adapter = device_get_softc(dev);
1910 	int rid;
1911 
1912 	rid = PCIR_BAR(0);
1913 	adapter->res_memory = bus_alloc_resource_any(dev, SYS_RES_MEMORY,
1914 						     &rid, RF_ACTIVE);
1915 	if (adapter->res_memory == NULL) {
1916 		device_printf(dev, "Unable to allocate bus resource: memory\n");
1917 		return ENXIO;
1918 	}
1919 	adapter->osdep.mem_bus_space_tag =
1920 		rman_get_bustag(adapter->res_memory);
1921 	adapter->osdep.mem_bus_space_handle =
1922 	    rman_get_bushandle(adapter->res_memory);
1923 	adapter->hw.hw_addr = (uint8_t *)&adapter->osdep.mem_bus_space_handle;
1924 
1925 	if (adapter->hw.mac_type > em_82543) {
1926 		/* Figure our where our IO BAR is ? */
1927 		for (rid = PCIR_BAR(0); rid < PCIR_CIS;) {
1928 			uint32_t val;
1929 
1930 			val = pci_read_config(dev, rid, 4);
1931 			if (EM_BAR_TYPE(val) == EM_BAR_TYPE_IO) {
1932 				adapter->io_rid = rid;
1933 				break;
1934 			}
1935 			rid += 4;
1936 			/* check for 64bit BAR */
1937 			if (EM_BAR_MEM_TYPE(val) == EM_BAR_MEM_TYPE_64BIT)
1938 				rid += 4;
1939 		}
1940 		if (rid >= PCIR_CIS) {
1941 			device_printf(dev, "Unable to locate IO BAR\n");
1942 			return (ENXIO);
1943  		}
1944 
1945 		adapter->res_ioport = bus_alloc_resource_any(dev,
1946 		    SYS_RES_IOPORT, &adapter->io_rid, RF_ACTIVE);
1947 		if (!(adapter->res_ioport)) {
1948 			device_printf(dev, "Unable to allocate bus resource: "
1949 				      "ioport\n");
1950 			return ENXIO;
1951 		}
1952 		adapter->hw.io_base = 0;
1953 		adapter->osdep.io_bus_space_tag =
1954 			rman_get_bustag(adapter->res_ioport);
1955 		adapter->osdep.io_bus_space_handle =
1956 			rman_get_bushandle(adapter->res_ioport);
1957 	}
1958 
1959 	/* For ICH8 we need to find the flash memory. */
1960 	if (adapter->hw.mac_type == em_ich8lan) {
1961 		rid = EM_FLASH;
1962 		adapter->flash_mem = bus_alloc_resource_any(dev,
1963 		    SYS_RES_MEMORY, &rid, RF_ACTIVE);
1964 		if (adapter->flash_mem == NULL) {
1965 			device_printf(dev, "Unable to allocate bus resource: "
1966 				      "flash memory\n");
1967 			return ENXIO;
1968 		}
1969 		adapter->osdep.flash_bus_space_tag =
1970 		    rman_get_bustag(adapter->flash_mem);
1971 		adapter->osdep.flash_bus_space_handle =
1972 		    rman_get_bushandle(adapter->flash_mem);
1973 	}
1974 
1975 	rid = 0x0;
1976 	adapter->res_interrupt = bus_alloc_resource_any(dev, SYS_RES_IRQ,
1977 	    &rid, RF_SHAREABLE | RF_ACTIVE);
1978 	if (adapter->res_interrupt == NULL) {
1979 		device_printf(dev, "Unable to allocate bus resource: "
1980 			      "interrupt\n");
1981 		return ENXIO;
1982 	}
1983 
1984 	adapter->hw.back = &adapter->osdep;
1985 
1986 	return 0;
1987 }
1988 
1989 static void
1990 em_free_pci_resources(device_t dev)
1991 {
1992 	struct adapter *adapter = device_get_softc(dev);
1993 
1994 	if (adapter->res_interrupt != NULL) {
1995 		bus_release_resource(dev, SYS_RES_IRQ, 0,
1996 				     adapter->res_interrupt);
1997 	}
1998 	if (adapter->res_memory != NULL) {
1999 		bus_release_resource(dev, SYS_RES_MEMORY, PCIR_BAR(0),
2000 				     adapter->res_memory);
2001 	}
2002 
2003 	if (adapter->res_ioport != NULL) {
2004 		bus_release_resource(dev, SYS_RES_IOPORT, adapter->io_rid,
2005 				     adapter->res_ioport);
2006 	}
2007 
2008 	if (adapter->flash_mem != NULL) {
2009 		bus_release_resource(dev, SYS_RES_MEMORY, EM_FLASH,
2010 				     adapter->flash_mem);
2011 	}
2012 }
2013 
2014 /*********************************************************************
2015  *
2016  *  Initialize the hardware to a configuration as specified by the
2017  *  adapter structure. The controller is reset, the EEPROM is
2018  *  verified, the MAC address is set, then the shared initialization
2019  *  routines are called.
2020  *
2021  **********************************************************************/
2022 static int
2023 em_hardware_init(struct adapter *adapter)
2024 {
2025 	uint16_t	rx_buffer_size;
2026 
2027 	INIT_DEBUGOUT("em_hardware_init: begin");
2028 	/* Issue a global reset */
2029 	em_reset_hw(&adapter->hw);
2030 
2031 	/* When hardware is reset, fifo_head is also reset */
2032 	adapter->tx_fifo_head = 0;
2033 
2034 	/* Make sure we have a good EEPROM before we read from it */
2035 	if (em_validate_eeprom_checksum(&adapter->hw) < 0) {
2036 		device_printf(adapter->dev,
2037 			      "The EEPROM Checksum Is Not Valid\n");
2038 		return (EIO);
2039 	}
2040 
2041 	if (em_read_part_num(&adapter->hw, &(adapter->part_num)) < 0) {
2042 		device_printf(adapter->dev,
2043 			      "EEPROM read error while reading part number\n");
2044 		return (EIO);
2045 	}
2046 
2047 	/* Set up smart power down as default off on newer adapters. */
2048 	if (!em_smart_pwr_down &&
2049 	    (adapter->hw.mac_type == em_82571 ||
2050 	     adapter->hw.mac_type == em_82572)) {
2051 		uint16_t phy_tmp = 0;
2052 
2053 		/* Speed up time to link by disabling smart power down. */
2054 		em_read_phy_reg(&adapter->hw, IGP02E1000_PHY_POWER_MGMT,
2055 				&phy_tmp);
2056 		phy_tmp &= ~IGP02E1000_PM_SPD;
2057 		em_write_phy_reg(&adapter->hw, IGP02E1000_PHY_POWER_MGMT,
2058 				 phy_tmp);
2059 	}
2060 
2061 	/*
2062 	 * These parameters control the automatic generation (Tx) and
2063 	 * response (Rx) to Ethernet PAUSE frames.
2064 	 * - High water mark should allow for at least two frames to be
2065 	 *   received after sending an XOFF.
2066 	 * - Low water mark works best when it is very near the high water mark.
2067 	 *   This allows the receiver to restart by sending XON when it has
2068 	 *   drained a bit.  Here we use an arbitary value of 1500 which will
2069 	 *   restart after one full frame is pulled from the buffer.  There
2070 	 *   could be several smaller frames in the buffer and if so they will
2071 	 *   not trigger the XON until their total number reduces the buffer
2072 	 *   by 1500.
2073 	 * - The pause time is fairly large at 1000 x 512ns = 512 usec.
2074 	 */
2075 	rx_buffer_size = ((E1000_READ_REG(&adapter->hw, PBA) & 0xffff) << 10);
2076 
2077 	adapter->hw.fc_high_water =
2078 	    rx_buffer_size - roundup2(adapter->hw.max_frame_size, 1024);
2079 	adapter->hw.fc_low_water = adapter->hw.fc_high_water - 1500;
2080 	if (adapter->hw.mac_type == em_80003es2lan)
2081 		adapter->hw.fc_pause_time = 0xFFFF;
2082 	else
2083 		adapter->hw.fc_pause_time = 0x1000;
2084 	adapter->hw.fc_send_xon = TRUE;
2085 	adapter->hw.fc = E1000_FC_FULL;
2086 
2087 	if (em_init_hw(&adapter->hw) < 0) {
2088 		device_printf(adapter->dev, "Hardware Initialization Failed");
2089 		return (EIO);
2090 	}
2091 
2092 	em_check_for_link(&adapter->hw);
2093 
2094 	return (0);
2095 }
2096 
2097 /*********************************************************************
2098  *
2099  *  Setup networking device structure and register an interface.
2100  *
2101  **********************************************************************/
2102 static void
2103 em_setup_interface(device_t dev, struct adapter *adapter)
2104 {
2105 	struct ifnet *ifp;
2106 	u_char fiber_type = IFM_1000_SX;	/* default type */
2107 	INIT_DEBUGOUT("em_setup_interface: begin");
2108 
2109 	ifp = &adapter->interface_data.ac_if;
2110 	if_initname(ifp, device_get_name(dev), device_get_unit(dev));
2111 	ifp->if_mtu = ETHERMTU;
2112 	ifp->if_baudrate = 1000000000;
2113 	ifp->if_init =  em_init;
2114 	ifp->if_softc = adapter;
2115 	ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
2116 	ifp->if_ioctl = em_ioctl;
2117 	ifp->if_start = em_start;
2118 #ifdef DEVICE_POLLING
2119 	ifp->if_poll = em_poll;
2120 #endif
2121 	ifp->if_watchdog = em_watchdog;
2122 	ifq_set_maxlen(&ifp->if_snd, adapter->num_tx_desc - 1);
2123 	ifq_set_ready(&ifp->if_snd);
2124 
2125 	if (adapter->hw.mac_type >= em_82543)
2126 		ifp->if_capabilities |= IFCAP_HWCSUM;
2127 
2128 	ifp->if_capenable = ifp->if_capabilities;
2129 
2130 	ether_ifattach(ifp, adapter->hw.mac_addr, NULL);
2131 
2132 	/*
2133 	 * Tell the upper layer(s) we support long frames.
2134 	 */
2135 	ifp->if_data.ifi_hdrlen = sizeof(struct ether_vlan_header);
2136 	ifp->if_capabilities |= IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU;
2137 #if 0
2138 	ifp->if_capenable |= IFCAP_VLAN_MTU;
2139 #endif
2140 
2141 	/*
2142 	 * Specify the media types supported by this adapter and register
2143 	 * callbacks to update media and link information
2144 	 */
2145 	ifmedia_init(&adapter->media, IFM_IMASK, em_media_change,
2146 		     em_media_status);
2147 	if (adapter->hw.media_type == em_media_type_fiber ||
2148 	    adapter->hw.media_type == em_media_type_internal_serdes) {
2149 		if (adapter->hw.mac_type == em_82545)
2150 			fiber_type = IFM_1000_LX;
2151 		ifmedia_add(&adapter->media, IFM_ETHER | fiber_type | IFM_FDX,
2152 			    0, NULL);
2153 		ifmedia_add(&adapter->media, IFM_ETHER | fiber_type, 0, NULL);
2154 	} else {
2155 		ifmedia_add(&adapter->media, IFM_ETHER | IFM_10_T, 0, NULL);
2156 		ifmedia_add(&adapter->media, IFM_ETHER | IFM_10_T | IFM_FDX,
2157 			    0, NULL);
2158 		ifmedia_add(&adapter->media, IFM_ETHER | IFM_100_TX,
2159 			    0, NULL);
2160 		ifmedia_add(&adapter->media, IFM_ETHER | IFM_100_TX | IFM_FDX,
2161 			    0, NULL);
2162 		ifmedia_add(&adapter->media, IFM_ETHER | IFM_1000_T | IFM_FDX,
2163 			    0, NULL);
2164 		ifmedia_add(&adapter->media, IFM_ETHER | IFM_1000_T, 0, NULL);
2165 	}
2166 	ifmedia_add(&adapter->media, IFM_ETHER | IFM_AUTO, 0, NULL);
2167 	ifmedia_set(&adapter->media, IFM_ETHER | IFM_AUTO);
2168 }
2169 
2170 /*********************************************************************
2171  *
2172  *  Workaround for SmartSpeed on 82541 and 82547 controllers
2173  *
2174  **********************************************************************/
2175 static void
2176 em_smartspeed(struct adapter *adapter)
2177 {
2178 	uint16_t phy_tmp;
2179 
2180 	if (adapter->link_active || (adapter->hw.phy_type != em_phy_igp) ||
2181 	    !adapter->hw.autoneg ||
2182 	    !(adapter->hw.autoneg_advertised & ADVERTISE_1000_FULL))
2183 		return;
2184 
2185 	if (adapter->smartspeed == 0) {
2186 		/*
2187 		 * If Master/Slave config fault is asserted twice,
2188 		 * we assume back-to-back.
2189 		 */
2190 		em_read_phy_reg(&adapter->hw, PHY_1000T_STATUS, &phy_tmp);
2191 		if (!(phy_tmp & SR_1000T_MS_CONFIG_FAULT))
2192 			return;
2193 		em_read_phy_reg(&adapter->hw, PHY_1000T_STATUS, &phy_tmp);
2194 		if (phy_tmp & SR_1000T_MS_CONFIG_FAULT) {
2195 			em_read_phy_reg(&adapter->hw, PHY_1000T_CTRL, &phy_tmp);
2196 			if (phy_tmp & CR_1000T_MS_ENABLE) {
2197 				phy_tmp &= ~CR_1000T_MS_ENABLE;
2198 				em_write_phy_reg(&adapter->hw,
2199 						 PHY_1000T_CTRL, phy_tmp);
2200 				adapter->smartspeed++;
2201 				if (adapter->hw.autoneg &&
2202 				    !em_phy_setup_autoneg(&adapter->hw) &&
2203 				    !em_read_phy_reg(&adapter->hw, PHY_CTRL,
2204 						     &phy_tmp)) {
2205 					phy_tmp |= (MII_CR_AUTO_NEG_EN |
2206 						    MII_CR_RESTART_AUTO_NEG);
2207 					em_write_phy_reg(&adapter->hw,
2208 							 PHY_CTRL, phy_tmp);
2209 				}
2210 			}
2211 		}
2212 		return;
2213 	} else if (adapter->smartspeed == EM_SMARTSPEED_DOWNSHIFT) {
2214 		/* If still no link, perhaps using 2/3 pair cable */
2215 		em_read_phy_reg(&adapter->hw, PHY_1000T_CTRL, &phy_tmp);
2216 		phy_tmp |= CR_1000T_MS_ENABLE;
2217 		em_write_phy_reg(&adapter->hw, PHY_1000T_CTRL, phy_tmp);
2218 		if (adapter->hw.autoneg &&
2219 		    !em_phy_setup_autoneg(&adapter->hw) &&
2220 		    !em_read_phy_reg(&adapter->hw, PHY_CTRL, &phy_tmp)) {
2221 			phy_tmp |= (MII_CR_AUTO_NEG_EN |
2222 				    MII_CR_RESTART_AUTO_NEG);
2223 			em_write_phy_reg(&adapter->hw, PHY_CTRL, phy_tmp);
2224 		}
2225 	}
2226 	/* Restart process after EM_SMARTSPEED_MAX iterations */
2227 	if (adapter->smartspeed++ == EM_SMARTSPEED_MAX)
2228 		adapter->smartspeed = 0;
2229 }
2230 
2231 /*
2232  * Manage DMA'able memory.
2233  */
2234 static void
2235 em_dmamap_cb(void *arg, bus_dma_segment_t *segs, int nseg, int error)
2236 {
2237 	if (error)
2238 		return;
2239 	*(bus_addr_t *)arg = segs->ds_addr;
2240 }
2241 
2242 static int
2243 em_dma_malloc(struct adapter *adapter, bus_size_t size,
2244 	      struct em_dma_alloc *dma)
2245 {
2246 	device_t dev = adapter->dev;
2247 	int error;
2248 
2249 	error = bus_dma_tag_create(NULL,		/* parent */
2250 				   EM_DBA_ALIGN, 0,	/* alignment, bounds */
2251 				   BUS_SPACE_MAXADDR,	/* lowaddr */
2252 				   BUS_SPACE_MAXADDR,	/* highaddr */
2253 				   NULL, NULL,		/* filter, filterarg */
2254 				   size,		/* maxsize */
2255 				   1,			/* nsegments */
2256 				   size,		/* maxsegsize */
2257 				   0,			/* flags */
2258 				   &dma->dma_tag);
2259 	if (error) {
2260 		device_printf(dev, "%s: bus_dma_tag_create failed; error %d\n",
2261 			      __func__, error);
2262 		return error;
2263 	}
2264 
2265 	error = bus_dmamem_alloc(dma->dma_tag, (void**)&dma->dma_vaddr,
2266 				 BUS_DMA_WAITOK, &dma->dma_map);
2267 	if (error) {
2268 		device_printf(dev, "%s: bus_dmammem_alloc failed; "
2269 			      "size %llu, error %d\n",
2270 			      __func__, (uintmax_t)size, error);
2271 		goto fail;
2272 	}
2273 
2274 	error = bus_dmamap_load(dma->dma_tag, dma->dma_map,
2275 				dma->dma_vaddr, size,
2276 				em_dmamap_cb, &dma->dma_paddr,
2277 				BUS_DMA_WAITOK);
2278 	if (error) {
2279 		device_printf(dev, "%s: bus_dmamap_load failed; error %u\n",
2280 			      __func__, error);
2281 		bus_dmamem_free(dma->dma_tag, dma->dma_vaddr, dma->dma_map);
2282 		goto fail;
2283 	}
2284 
2285 	return 0;
2286 fail:
2287 	bus_dma_tag_destroy(dma->dma_tag);
2288 	dma->dma_tag = NULL;
2289 	return error;
2290 }
2291 
2292 static void
2293 em_dma_free(struct adapter *adapter, struct em_dma_alloc *dma)
2294 {
2295 	if (dma->dma_tag != NULL) {
2296 		bus_dmamap_unload(dma->dma_tag, dma->dma_map);
2297 		bus_dmamem_free(dma->dma_tag, dma->dma_vaddr, dma->dma_map);
2298 		bus_dma_tag_destroy(dma->dma_tag);
2299 		dma->dma_tag = NULL;
2300 	}
2301 }
2302 
2303 /*********************************************************************
2304  *
2305  *  Allocate and initialize transmit structures.
2306  *
2307  **********************************************************************/
2308 static int
2309 em_setup_transmit_structures(struct adapter *adapter)
2310 {
2311 	struct em_buffer *tx_buffer;
2312 	bus_size_t size;
2313 	int error, i;
2314 
2315 	/*
2316 	 * Setup DMA descriptor areas.
2317 	 */
2318 	size = roundup2(adapter->hw.max_frame_size, MCLBYTES);
2319 	if (bus_dma_tag_create(NULL,			/* parent */
2320 			       1, 0,			/* alignment, bounds */
2321 			       BUS_SPACE_MAXADDR,	/* lowaddr */
2322 			       BUS_SPACE_MAXADDR,	/* highaddr */
2323 			       NULL, NULL,		/* filter, filterarg */
2324 			       size,			/* maxsize */
2325 			       EM_MAX_SCATTER,		/* nsegments */
2326 			       size,			/* maxsegsize */
2327 			       0,			/* flags */
2328 			       &adapter->txtag)) {
2329 		device_printf(adapter->dev, "Unable to allocate TX DMA tag\n");
2330 		return(ENOMEM);
2331 	}
2332 
2333 	adapter->tx_buffer_area =
2334 		kmalloc(sizeof(struct em_buffer) * adapter->num_tx_desc,
2335 			M_DEVBUF, M_WAITOK | M_ZERO);
2336 
2337 	bzero(adapter->tx_desc_base,
2338 	      sizeof(struct em_tx_desc) * adapter->num_tx_desc);
2339 	tx_buffer = adapter->tx_buffer_area;
2340 	for (i = 0; i < adapter->num_tx_desc; i++) {
2341 		error = bus_dmamap_create(adapter->txtag, 0, &tx_buffer->map);
2342 		if (error) {
2343 			device_printf(adapter->dev,
2344 				      "Unable to create TX DMA map\n");
2345 			goto fail;
2346 		}
2347 		tx_buffer++;
2348 	}
2349 
2350 	adapter->next_avail_tx_desc = 0;
2351 	adapter->next_tx_to_clean = 0;
2352 
2353 	/* Set number of descriptors available */
2354 	adapter->num_tx_desc_avail = adapter->num_tx_desc;
2355 
2356 	/* Set checksum context */
2357 	adapter->active_checksum_context = OFFLOAD_NONE;
2358 
2359 	bus_dmamap_sync(adapter->txdma.dma_tag, adapter->txdma.dma_map,
2360 			BUS_DMASYNC_PREWRITE);
2361 
2362 	return (0);
2363 fail:
2364 	em_free_transmit_structures(adapter);
2365 	return (error);
2366 }
2367 
2368 /*********************************************************************
2369  *
2370  *  Enable transmit unit.
2371  *
2372  **********************************************************************/
2373 static void
2374 em_initialize_transmit_unit(struct adapter *adapter)
2375 {
2376 	uint32_t reg_tctl;
2377 	uint32_t reg_tipg = 0;
2378 	uint64_t bus_addr;
2379 
2380 	INIT_DEBUGOUT("em_initialize_transmit_unit: begin");
2381 
2382 	/* Setup the Base and Length of the Tx Descriptor Ring */
2383 	bus_addr = adapter->txdma.dma_paddr;
2384 	E1000_WRITE_REG(&adapter->hw, TDLEN,
2385 			adapter->num_tx_desc * sizeof(struct em_tx_desc));
2386 	E1000_WRITE_REG(&adapter->hw, TDBAH, (uint32_t)(bus_addr >> 32));
2387 	E1000_WRITE_REG(&adapter->hw, TDBAL, (uint32_t)bus_addr);
2388 
2389 	/* Setup the HW Tx Head and Tail descriptor pointers */
2390 	E1000_WRITE_REG(&adapter->hw, TDT, 0);
2391 	E1000_WRITE_REG(&adapter->hw, TDH, 0);
2392 
2393 	HW_DEBUGOUT2("Base = %x, Length = %x\n",
2394 		     E1000_READ_REG(&adapter->hw, TDBAL),
2395 		     E1000_READ_REG(&adapter->hw, TDLEN));
2396 
2397 	/* Set the default values for the Tx Inter Packet Gap timer */
2398 	switch (adapter->hw.mac_type) {
2399 	case em_82542_rev2_0:
2400 	case em_82542_rev2_1:
2401 		reg_tipg = DEFAULT_82542_TIPG_IPGT;
2402 		reg_tipg |= DEFAULT_82542_TIPG_IPGR1 << E1000_TIPG_IPGR1_SHIFT;
2403 		reg_tipg |= DEFAULT_82542_TIPG_IPGR2 << E1000_TIPG_IPGR2_SHIFT;
2404 		break;
2405 	case em_80003es2lan:
2406 		reg_tipg = DEFAULT_82543_TIPG_IPGR1;
2407 		reg_tipg |=
2408 		    DEFAULT_80003ES2LAN_TIPG_IPGR2 << E1000_TIPG_IPGR2_SHIFT;
2409 		break;
2410 	default:
2411 		if (adapter->hw.media_type == em_media_type_fiber ||
2412 		    adapter->hw.media_type == em_media_type_internal_serdes)
2413 			reg_tipg = DEFAULT_82543_TIPG_IPGT_FIBER;
2414 		else
2415 			reg_tipg = DEFAULT_82543_TIPG_IPGT_COPPER;
2416 		reg_tipg |= DEFAULT_82543_TIPG_IPGR1 << E1000_TIPG_IPGR1_SHIFT;
2417 		reg_tipg |= DEFAULT_82543_TIPG_IPGR2 << E1000_TIPG_IPGR2_SHIFT;
2418 	}
2419 
2420 	E1000_WRITE_REG(&adapter->hw, TIPG, reg_tipg);
2421 	E1000_WRITE_REG(&adapter->hw, TIDV, adapter->tx_int_delay.value);
2422 	if (adapter->hw.mac_type >= em_82540) {
2423 		E1000_WRITE_REG(&adapter->hw, TADV,
2424 				adapter->tx_abs_int_delay.value);
2425 	}
2426 
2427 	/* Program the Transmit Control Register */
2428 	reg_tctl = E1000_TCTL_PSP | E1000_TCTL_EN |
2429 		   (E1000_COLLISION_THRESHOLD << E1000_CT_SHIFT);
2430 	if (adapter->hw.mac_type >= em_82571)
2431 		reg_tctl |= E1000_TCTL_MULR;
2432 	if (adapter->link_duplex == 1)
2433 		reg_tctl |= E1000_FDX_COLLISION_DISTANCE << E1000_COLD_SHIFT;
2434 	else
2435 		reg_tctl |= E1000_HDX_COLLISION_DISTANCE << E1000_COLD_SHIFT;
2436 
2437 	/* This write will effectively turn on the transmit unit. */
2438 	E1000_WRITE_REG(&adapter->hw, TCTL, reg_tctl);
2439 
2440 	/* Setup Transmit Descriptor Base Settings */
2441 	adapter->txd_cmd = E1000_TXD_CMD_IFCS;
2442 
2443 	if (adapter->tx_int_delay.value > 0)
2444 		adapter->txd_cmd |= E1000_TXD_CMD_IDE;
2445 }
2446 
2447 /*********************************************************************
2448  *
2449  *  Free all transmit related data structures.
2450  *
2451  **********************************************************************/
2452 static void
2453 em_free_transmit_structures(struct adapter *adapter)
2454 {
2455 	struct em_buffer *tx_buffer;
2456 	int i;
2457 
2458 	INIT_DEBUGOUT("free_transmit_structures: begin");
2459 
2460 	if (adapter->tx_buffer_area != NULL) {
2461 		tx_buffer = adapter->tx_buffer_area;
2462 		for (i = 0; i < adapter->num_tx_desc; i++, tx_buffer++) {
2463 			if (tx_buffer->m_head != NULL) {
2464 				bus_dmamap_unload(adapter->txtag,
2465 						  tx_buffer->map);
2466 				m_freem(tx_buffer->m_head);
2467 			}
2468 
2469 			if (tx_buffer->map != NULL) {
2470 				bus_dmamap_destroy(adapter->txtag, tx_buffer->map);
2471 				tx_buffer->map = NULL;
2472  			}
2473 			tx_buffer->m_head = NULL;
2474 		}
2475 	}
2476 	if (adapter->tx_buffer_area != NULL) {
2477 		kfree(adapter->tx_buffer_area, M_DEVBUF);
2478 		adapter->tx_buffer_area = NULL;
2479 	}
2480 	if (adapter->txtag != NULL) {
2481 		bus_dma_tag_destroy(adapter->txtag);
2482 		adapter->txtag = NULL;
2483 	}
2484 }
2485 
2486 /*********************************************************************
2487  *
2488  *  The offload context needs to be set when we transfer the first
2489  *  packet of a particular protocol (TCP/UDP). We change the
2490  *  context only if the protocol type changes.
2491  *
2492  **********************************************************************/
2493 static void
2494 em_transmit_checksum_setup(struct adapter *adapter,
2495 			   struct mbuf *mp,
2496 			   uint32_t *txd_upper,
2497 			   uint32_t *txd_lower)
2498 {
2499 	struct em_context_desc *TXD;
2500 	struct em_buffer *tx_buffer;
2501 	int curr_txd;
2502 
2503 	if (mp->m_pkthdr.csum_flags) {
2504 		if (mp->m_pkthdr.csum_flags & CSUM_TCP) {
2505 			*txd_upper = E1000_TXD_POPTS_TXSM << 8;
2506 			*txd_lower = E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
2507 			if (adapter->active_checksum_context == OFFLOAD_TCP_IP)
2508 				return;
2509 			else
2510 				adapter->active_checksum_context = OFFLOAD_TCP_IP;
2511 		} else if (mp->m_pkthdr.csum_flags & CSUM_UDP) {
2512 			*txd_upper = E1000_TXD_POPTS_TXSM << 8;
2513 			*txd_lower = E1000_TXD_CMD_DEXT | E1000_TXD_DTYP_D;
2514 			if (adapter->active_checksum_context == OFFLOAD_UDP_IP)
2515 				return;
2516 			else
2517 				adapter->active_checksum_context = OFFLOAD_UDP_IP;
2518 		} else {
2519 			*txd_upper = 0;
2520 			*txd_lower = 0;
2521 			return;
2522 		}
2523 	} else {
2524 		*txd_upper = 0;
2525 		*txd_lower = 0;
2526 		return;
2527 	}
2528 
2529 	/*
2530 	 * If we reach this point, the checksum offload context
2531 	 * needs to be reset.
2532 	 */
2533 	curr_txd = adapter->next_avail_tx_desc;
2534 	tx_buffer = &adapter->tx_buffer_area[curr_txd];
2535 	TXD = (struct em_context_desc *) &adapter->tx_desc_base[curr_txd];
2536 
2537 	TXD->lower_setup.ip_fields.ipcss = ETHER_HDR_LEN;
2538 	TXD->lower_setup.ip_fields.ipcso =
2539 	    ETHER_HDR_LEN + offsetof(struct ip, ip_sum);
2540 	TXD->lower_setup.ip_fields.ipcse =
2541 	    htole16(ETHER_HDR_LEN + sizeof(struct ip) - 1);
2542 
2543 	TXD->upper_setup.tcp_fields.tucss =
2544 	    ETHER_HDR_LEN + sizeof(struct ip);
2545 	TXD->upper_setup.tcp_fields.tucse = htole16(0);
2546 
2547 	if (adapter->active_checksum_context == OFFLOAD_TCP_IP) {
2548 		TXD->upper_setup.tcp_fields.tucso =
2549 			ETHER_HDR_LEN + sizeof(struct ip) +
2550 			offsetof(struct tcphdr, th_sum);
2551 	} else if (adapter->active_checksum_context == OFFLOAD_UDP_IP) {
2552 		TXD->upper_setup.tcp_fields.tucso =
2553 			ETHER_HDR_LEN + sizeof(struct ip) +
2554 			offsetof(struct udphdr, uh_sum);
2555 	}
2556 
2557 	TXD->tcp_seg_setup.data = htole32(0);
2558 	TXD->cmd_and_length = htole32(adapter->txd_cmd | E1000_TXD_CMD_DEXT);
2559 
2560 	tx_buffer->m_head = NULL;
2561 	tx_buffer->next_eop = -1;
2562 
2563 	if (++curr_txd == adapter->num_tx_desc)
2564 		curr_txd = 0;
2565 
2566 	adapter->num_tx_desc_avail--;
2567 	adapter->next_avail_tx_desc = curr_txd;
2568 }
2569 
2570 /**********************************************************************
2571  *
2572  *  Examine each tx_buffer in the used queue. If the hardware is done
2573  *  processing the packet then free associated resources. The
2574  *  tx_buffer is put back on the free queue.
2575  *
2576  **********************************************************************/
2577 
2578 static void
2579 em_txeof(struct adapter *adapter)
2580 {
2581 	int first, last, done, num_avail;
2582 	struct em_buffer *tx_buffer;
2583 	struct em_tx_desc *tx_desc, *eop_desc;
2584 	struct ifnet *ifp = &adapter->interface_data.ac_if;
2585 
2586 	if (adapter->num_tx_desc_avail == adapter->num_tx_desc)
2587 		return;
2588 
2589 	num_avail = adapter->num_tx_desc_avail;
2590 	first = adapter->next_tx_to_clean;
2591 	tx_desc = &adapter->tx_desc_base[first];
2592 	tx_buffer = &adapter->tx_buffer_area[first];
2593 	last = tx_buffer->next_eop;
2594 	KKASSERT(last >= 0 && last < adapter->num_tx_desc);
2595 	eop_desc = &adapter->tx_desc_base[last];
2596 
2597 	/*
2598 	 * Now caculate the terminating index for the cleanup loop below
2599 	 */
2600 	if (++last == adapter->num_tx_desc)
2601 		last = 0;
2602 	done = last;
2603 
2604 	bus_dmamap_sync(adapter->txdma.dma_tag, adapter->txdma.dma_map,
2605 			BUS_DMASYNC_POSTREAD);
2606 
2607 	while (eop_desc->upper.fields.status & E1000_TXD_STAT_DD) {
2608 		while (first != done) {
2609 			tx_desc->upper.data = 0;
2610 			tx_desc->lower.data = 0;
2611 			num_avail++;
2612 
2613 			logif(pkt_txclean);
2614 
2615 			if (tx_buffer->m_head) {
2616 				ifp->if_opackets++;
2617 				bus_dmamap_sync(adapter->txtag, tx_buffer->map,
2618 						BUS_DMASYNC_POSTWRITE);
2619 				bus_dmamap_unload(adapter->txtag,
2620 						  tx_buffer->map);
2621 
2622 				m_freem(tx_buffer->m_head);
2623 				tx_buffer->m_head = NULL;
2624 			}
2625 			tx_buffer->next_eop = -1;
2626 
2627 			if (++first == adapter->num_tx_desc)
2628 				first = 0;
2629 
2630 			tx_buffer = &adapter->tx_buffer_area[first];
2631 			tx_desc = &adapter->tx_desc_base[first];
2632 		}
2633 		/* See if we can continue to the next packet */
2634 		last = tx_buffer->next_eop;
2635 		if (last != -1) {
2636 			KKASSERT(last >= 0 && last < adapter->num_tx_desc);
2637 			eop_desc = &adapter->tx_desc_base[last];
2638 			if (++last == adapter->num_tx_desc)
2639 				last = 0;
2640 			done = last;
2641 		} else {
2642 			break;
2643 		}
2644 	}
2645 
2646 	bus_dmamap_sync(adapter->txdma.dma_tag, adapter->txdma.dma_map,
2647 			BUS_DMASYNC_PREWRITE);
2648 
2649 	adapter->next_tx_to_clean = first;
2650 
2651 	/*
2652 	 * If we have enough room, clear IFF_OACTIVE to tell the stack
2653 	 * that it is OK to send packets.
2654 	 * If there are no pending descriptors, clear the timeout. Otherwise,
2655 	 * if some descriptors have been freed, restart the timeout.
2656 	 */
2657 	if (num_avail > EM_TX_CLEANUP_THRESHOLD) {
2658 		ifp->if_flags &= ~IFF_OACTIVE;
2659 		if (num_avail == adapter->num_tx_desc)
2660 			ifp->if_timer = 0;
2661 		else if (num_avail == adapter->num_tx_desc_avail)
2662 			ifp->if_timer = EM_TX_TIMEOUT;
2663 	}
2664 	adapter->num_tx_desc_avail = num_avail;
2665 }
2666 
2667 /*********************************************************************
2668  *
2669  *  Get a buffer from system mbuf buffer pool.
2670  *
2671  **********************************************************************/
2672 static int
2673 em_get_buf(int i, struct adapter *adapter, struct mbuf *nmp, int how)
2674 {
2675 	struct mbuf *mp = nmp;
2676 	struct em_buffer *rx_buffer;
2677 	struct ifnet *ifp;
2678 	bus_addr_t paddr;
2679 	int error;
2680 
2681 	ifp = &adapter->interface_data.ac_if;
2682 
2683 	if (mp == NULL) {
2684 		mp = m_getcl(how, MT_DATA, M_PKTHDR);
2685 		if (mp == NULL) {
2686 			adapter->mbuf_cluster_failed++;
2687 			return (ENOBUFS);
2688 		}
2689 		mp->m_len = mp->m_pkthdr.len = MCLBYTES;
2690 	} else {
2691 		mp->m_len = mp->m_pkthdr.len = MCLBYTES;
2692 		mp->m_data = mp->m_ext.ext_buf;
2693 		mp->m_next = NULL;
2694 	}
2695 
2696 	if (ifp->if_mtu <= ETHERMTU)
2697 		m_adj(mp, ETHER_ALIGN);
2698 
2699 	rx_buffer = &adapter->rx_buffer_area[i];
2700 
2701 	/*
2702 	 * Using memory from the mbuf cluster pool, invoke the
2703 	 * bus_dma machinery to arrange the memory mapping.
2704 	 */
2705 	error = bus_dmamap_load(adapter->rxtag, rx_buffer->map,
2706 				mtod(mp, void *), mp->m_len,
2707 				em_dmamap_cb, &paddr, 0);
2708 	if (error) {
2709 		m_free(mp);
2710 		return (error);
2711 	}
2712 	rx_buffer->m_head = mp;
2713 	adapter->rx_desc_base[i].buffer_addr = htole64(paddr);
2714 	bus_dmamap_sync(adapter->rxtag, rx_buffer->map, BUS_DMASYNC_PREREAD);
2715 
2716 	return (0);
2717 }
2718 
2719 /*********************************************************************
2720  *
2721  *  Allocate memory for rx_buffer structures. Since we use one
2722  *  rx_buffer per received packet, the maximum number of rx_buffer's
2723  *  that we'll need is equal to the number of receive descriptors
2724  *  that we've allocated.
2725  *
2726  **********************************************************************/
2727 static int
2728 em_allocate_receive_structures(struct adapter *adapter)
2729 {
2730 	int i, error, size;
2731 	struct em_buffer *rx_buffer;
2732 
2733 	size = adapter->num_rx_desc * sizeof(struct em_buffer);
2734 	adapter->rx_buffer_area = kmalloc(size, M_DEVBUF, M_WAITOK | M_ZERO);
2735 
2736 	error = bus_dma_tag_create(NULL,		/* parent */
2737 				   1, 0,		/* alignment, bounds */
2738 				   BUS_SPACE_MAXADDR,	/* lowaddr */
2739 				   BUS_SPACE_MAXADDR,	/* highaddr */
2740 				   NULL, NULL,		/* filter, filterarg */
2741 				   MCLBYTES,		/* maxsize */
2742 				   1,			/* nsegments */
2743 				   MCLBYTES,		/* maxsegsize */
2744 				   0,			/* flags */
2745 				   &adapter->rxtag);
2746 	if (error) {
2747 		device_printf(adapter->dev, "%s: bus_dma_tag_create failed; "
2748 			      "error %u\n", __func__, error);
2749 		goto fail;
2750 	}
2751 
2752 	rx_buffer = adapter->rx_buffer_area;
2753 	for (i = 0; i < adapter->num_rx_desc; i++, rx_buffer++) {
2754 		error = bus_dmamap_create(adapter->rxtag, BUS_DMA_NOWAIT,
2755 					  &rx_buffer->map);
2756 		if (error) {
2757 			device_printf(adapter->dev,
2758 				      "%s: bus_dmamap_create failed; "
2759 				      "error %u\n", __func__, error);
2760 			goto fail;
2761 		}
2762 	}
2763 
2764 	for (i = 0; i < adapter->num_rx_desc; i++) {
2765 		error = em_get_buf(i, adapter, NULL, MB_WAIT);
2766 		if (error)
2767 			goto fail;
2768 	}
2769 
2770 	bus_dmamap_sync(adapter->rxdma.dma_tag, adapter->rxdma.dma_map,
2771 			BUS_DMASYNC_PREWRITE);
2772 
2773 	return (0);
2774 fail:
2775 	em_free_receive_structures(adapter);
2776 	return (error);
2777 }
2778 
2779 /*********************************************************************
2780  *
2781  *  Allocate and initialize receive structures.
2782  *
2783  **********************************************************************/
2784 static int
2785 em_setup_receive_structures(struct adapter *adapter)
2786 {
2787 	int error;
2788 
2789 	bzero(adapter->rx_desc_base,
2790 	      sizeof(struct em_rx_desc) * adapter->num_rx_desc);
2791 
2792 	error = em_allocate_receive_structures(adapter);
2793 	if (error)
2794 		return (error);
2795 
2796 	/* Setup our descriptor pointers */
2797 	adapter->next_rx_desc_to_check = 0;
2798 
2799 	return (0);
2800 }
2801 
2802 /*********************************************************************
2803  *
2804  *  Enable receive unit.
2805  *
2806  **********************************************************************/
2807 static void
2808 em_initialize_receive_unit(struct adapter *adapter)
2809 {
2810 	uint32_t reg_rctl;
2811 	uint32_t reg_rxcsum;
2812 	struct ifnet *ifp;
2813 	uint64_t bus_addr;
2814 
2815 	INIT_DEBUGOUT("em_initialize_receive_unit: begin");
2816 
2817 	ifp = &adapter->interface_data.ac_if;
2818 
2819 	/*
2820 	 * Make sure receives are disabled while setting
2821 	 * up the descriptor ring
2822 	 */
2823 	E1000_WRITE_REG(&adapter->hw, RCTL, 0);
2824 
2825 	/* Set the Receive Delay Timer Register */
2826 	E1000_WRITE_REG(&adapter->hw, RDTR,
2827 			adapter->rx_int_delay.value | E1000_RDT_FPDB);
2828 
2829 	if(adapter->hw.mac_type >= em_82540) {
2830 		E1000_WRITE_REG(&adapter->hw, RADV,
2831 				adapter->rx_abs_int_delay.value);
2832 
2833 		/* Set the interrupt throttling rate in 256ns increments */
2834 		if (em_int_throttle_ceil) {
2835 			E1000_WRITE_REG(&adapter->hw, ITR,
2836 				1000000000 / 256 / em_int_throttle_ceil);
2837 		} else {
2838 			E1000_WRITE_REG(&adapter->hw, ITR, 0);
2839 		}
2840 	}
2841 
2842 	/* Setup the Base and Length of the Rx Descriptor Ring */
2843 	bus_addr = adapter->rxdma.dma_paddr;
2844 	E1000_WRITE_REG(&adapter->hw, RDLEN, adapter->num_rx_desc *
2845 			sizeof(struct em_rx_desc));
2846 	E1000_WRITE_REG(&adapter->hw, RDBAH, (uint32_t)(bus_addr >> 32));
2847 	E1000_WRITE_REG(&adapter->hw, RDBAL, (uint32_t)bus_addr);
2848 
2849 	/* Setup the Receive Control Register */
2850 	reg_rctl = E1000_RCTL_EN | E1000_RCTL_BAM | E1000_RCTL_LBM_NO |
2851 		   E1000_RCTL_RDMTS_HALF |
2852 		   (adapter->hw.mc_filter_type << E1000_RCTL_MO_SHIFT);
2853 
2854 	if (adapter->hw.tbi_compatibility_on == TRUE)
2855 		reg_rctl |= E1000_RCTL_SBP;
2856 
2857 	switch (adapter->rx_buffer_len) {
2858 	default:
2859 	case EM_RXBUFFER_2048:
2860 		reg_rctl |= E1000_RCTL_SZ_2048;
2861 		break;
2862 	case EM_RXBUFFER_4096:
2863 		reg_rctl |= E1000_RCTL_SZ_4096 | E1000_RCTL_BSEX |
2864 			    E1000_RCTL_LPE;
2865 		break;
2866 	case EM_RXBUFFER_8192:
2867 		reg_rctl |= E1000_RCTL_SZ_8192 | E1000_RCTL_BSEX |
2868 			    E1000_RCTL_LPE;
2869 		break;
2870 	case EM_RXBUFFER_16384:
2871 		reg_rctl |= E1000_RCTL_SZ_16384 | E1000_RCTL_BSEX |
2872 			    E1000_RCTL_LPE;
2873 		break;
2874 	}
2875 
2876 	if (ifp->if_mtu > ETHERMTU)
2877 		reg_rctl |= E1000_RCTL_LPE;
2878 
2879 	/* Enable 82543 Receive Checksum Offload for TCP and UDP */
2880 	if ((adapter->hw.mac_type >= em_82543) &&
2881 	    (ifp->if_capenable & IFCAP_RXCSUM)) {
2882 		reg_rxcsum = E1000_READ_REG(&adapter->hw, RXCSUM);
2883 		reg_rxcsum |= (E1000_RXCSUM_IPOFL | E1000_RXCSUM_TUOFL);
2884 		E1000_WRITE_REG(&adapter->hw, RXCSUM, reg_rxcsum);
2885 	}
2886 
2887 	/* Enable Receives */
2888 	E1000_WRITE_REG(&adapter->hw, RCTL, reg_rctl);
2889 
2890 	/* Setup the HW Rx Head and Tail Descriptor Pointers */
2891 	E1000_WRITE_REG(&adapter->hw, RDH, 0);
2892 	E1000_WRITE_REG(&adapter->hw, RDT, adapter->num_rx_desc - 1);
2893 }
2894 
2895 /*********************************************************************
2896  *
2897  *  Free receive related data structures.
2898  *
2899  **********************************************************************/
2900 static void
2901 em_free_receive_structures(struct adapter *adapter)
2902 {
2903 	struct em_buffer *rx_buffer;
2904 	int i;
2905 
2906 	INIT_DEBUGOUT("free_receive_structures: begin");
2907 
2908 	if (adapter->rx_buffer_area != NULL) {
2909 		rx_buffer = adapter->rx_buffer_area;
2910 		for (i = 0; i < adapter->num_rx_desc; i++, rx_buffer++) {
2911 			if (rx_buffer->m_head != NULL) {
2912 				bus_dmamap_unload(adapter->rxtag,
2913 						  rx_buffer->map);
2914 				m_freem(rx_buffer->m_head);
2915 				rx_buffer->m_head = NULL;
2916 			}
2917 			if (rx_buffer->map != NULL) {
2918 				bus_dmamap_destroy(adapter->rxtag,
2919 						   rx_buffer->map);
2920 				rx_buffer->map = NULL;
2921 			}
2922 		}
2923 	}
2924 	if (adapter->rx_buffer_area != NULL) {
2925 		kfree(adapter->rx_buffer_area, M_DEVBUF);
2926 		adapter->rx_buffer_area = NULL;
2927 	}
2928 	if (adapter->rxtag != NULL) {
2929 		bus_dma_tag_destroy(adapter->rxtag);
2930 		adapter->rxtag = NULL;
2931 	}
2932 }
2933 
2934 /*********************************************************************
2935  *
2936  *  This routine executes in interrupt context. It replenishes
2937  *  the mbufs in the descriptor and sends data which has been
2938  *  dma'ed into host memory to upper layer.
2939  *
2940  *  We loop at most count times if count is > 0, or until done if
2941  *  count < 0.
2942  *
2943  *********************************************************************/
2944 static void
2945 em_rxeof(struct adapter *adapter, int count)
2946 {
2947 	struct ifnet *ifp;
2948 	struct mbuf *mp;
2949 	uint8_t accept_frame = 0;
2950 	uint8_t eop = 0;
2951 	uint16_t len, desc_len, prev_len_adj;
2952 	int i;
2953 
2954 	/* Pointer to the receive descriptor being examined. */
2955 	struct em_rx_desc *current_desc;
2956 
2957 	ifp = &adapter->interface_data.ac_if;
2958 	i = adapter->next_rx_desc_to_check;
2959 	current_desc = &adapter->rx_desc_base[i];
2960 
2961 	bus_dmamap_sync(adapter->rxdma.dma_tag, adapter->rxdma.dma_map,
2962 			BUS_DMASYNC_POSTREAD);
2963 
2964 	if (!(current_desc->status & E1000_RXD_STAT_DD))
2965 		return;
2966 
2967 	while ((current_desc->status & E1000_RXD_STAT_DD) && count != 0) {
2968 		logif(pkt_receive);
2969 		mp = adapter->rx_buffer_area[i].m_head;
2970 		bus_dmamap_sync(adapter->rxtag, adapter->rx_buffer_area[i].map,
2971 				BUS_DMASYNC_POSTREAD);
2972 		bus_dmamap_unload(adapter->rxtag,
2973 				  adapter->rx_buffer_area[i].map);
2974 
2975 		accept_frame = 1;
2976 		prev_len_adj = 0;
2977 		desc_len = le16toh(current_desc->length);
2978 		if (current_desc->status & E1000_RXD_STAT_EOP) {
2979 			count--;
2980 			eop = 1;
2981 			if (desc_len < ETHER_CRC_LEN) {
2982 				len = 0;
2983 				prev_len_adj = ETHER_CRC_LEN - desc_len;
2984 			} else {
2985 				len = desc_len - ETHER_CRC_LEN;
2986 			}
2987 		} else {
2988 			eop = 0;
2989 			len = desc_len;
2990 		}
2991 
2992 		if (current_desc->errors & E1000_RXD_ERR_FRAME_ERR_MASK) {
2993 			uint8_t last_byte;
2994 			uint32_t pkt_len = desc_len;
2995 
2996 			if (adapter->fmp != NULL)
2997 				pkt_len += adapter->fmp->m_pkthdr.len;
2998 
2999 			last_byte = *(mtod(mp, caddr_t) + desc_len - 1);
3000 
3001 			if (TBI_ACCEPT(&adapter->hw, current_desc->status,
3002 				       current_desc->errors,
3003 				       pkt_len, last_byte)) {
3004 				em_tbi_adjust_stats(&adapter->hw,
3005 						    &adapter->stats,
3006 						    pkt_len,
3007 						    adapter->hw.mac_addr);
3008 				if (len > 0)
3009 					len--;
3010 			} else {
3011 				accept_frame = 0;
3012 			}
3013 		}
3014 
3015 		if (accept_frame) {
3016 			if (em_get_buf(i, adapter, NULL, MB_DONTWAIT) == ENOBUFS) {
3017 				adapter->dropped_pkts++;
3018 				em_get_buf(i, adapter, mp, MB_DONTWAIT);
3019 				if (adapter->fmp != NULL)
3020 					m_freem(adapter->fmp);
3021 				adapter->fmp = NULL;
3022 				adapter->lmp = NULL;
3023 				goto skip;
3024 			}
3025 
3026 			/* Assign correct length to the current fragment */
3027 			mp->m_len = len;
3028 
3029 			if (adapter->fmp == NULL) {
3030 				mp->m_pkthdr.len = len;
3031 				adapter->fmp = mp;	 /* Store the first mbuf */
3032 				adapter->lmp = mp;
3033 			} else {
3034 				/* Chain mbuf's together */
3035 				/*
3036 				 * Adjust length of previous mbuf in chain if
3037 				 * we received less than 4 bytes in the last
3038 				 * descriptor.
3039 				 */
3040 				if (prev_len_adj > 0) {
3041 					adapter->lmp->m_len -= prev_len_adj;
3042 					adapter->fmp->m_pkthdr.len -= prev_len_adj;
3043 				}
3044 				adapter->lmp->m_next = mp;
3045 				adapter->lmp = adapter->lmp->m_next;
3046 				adapter->fmp->m_pkthdr.len += len;
3047 			}
3048 
3049 			if (eop) {
3050 				adapter->fmp->m_pkthdr.rcvif = ifp;
3051 				ifp->if_ipackets++;
3052 
3053 				em_receive_checksum(adapter, current_desc,
3054 						    adapter->fmp);
3055 				if (current_desc->status & E1000_RXD_STAT_VP) {
3056 					VLAN_INPUT_TAG(adapter->fmp,
3057 						       (current_desc->special &
3058 							E1000_RXD_SPC_VLAN_MASK));
3059 				} else {
3060 					ifp->if_input(ifp, adapter->fmp);
3061 				}
3062 				adapter->fmp = NULL;
3063 				adapter->lmp = NULL;
3064 			}
3065 		} else {
3066 			adapter->dropped_pkts++;
3067 			em_get_buf(i, adapter, mp, MB_DONTWAIT);
3068 			if (adapter->fmp != NULL)
3069 				m_freem(adapter->fmp);
3070 			adapter->fmp = NULL;
3071 			adapter->lmp = NULL;
3072 		}
3073 
3074 skip:
3075 		/* Zero out the receive descriptors status. */
3076 		current_desc->status = 0;
3077 
3078 		/* Advance our pointers to the next descriptor. */
3079 		if (++i == adapter->num_rx_desc) {
3080 			i = 0;
3081 			current_desc = adapter->rx_desc_base;
3082 		} else {
3083 			current_desc++;
3084 		}
3085 	}
3086 
3087 	bus_dmamap_sync(adapter->rxdma.dma_tag, adapter->rxdma.dma_map,
3088 			BUS_DMASYNC_PREWRITE);
3089 
3090 	adapter->next_rx_desc_to_check = i;
3091 
3092 	/* Advance the E1000's Receive Queue #0  "Tail Pointer". */
3093 	if (--i < 0)
3094 		i = adapter->num_rx_desc - 1;
3095 
3096 	E1000_WRITE_REG(&adapter->hw, RDT, i);
3097 }
3098 
3099 /*********************************************************************
3100  *
3101  *  Verify that the hardware indicated that the checksum is valid.
3102  *  Inform the stack about the status of checksum so that stack
3103  *  doesn't spend time verifying the checksum.
3104  *
3105  *********************************************************************/
3106 static void
3107 em_receive_checksum(struct adapter *adapter,
3108 		    struct em_rx_desc *rx_desc,
3109 		    struct mbuf *mp)
3110 {
3111 	/* 82543 or newer only */
3112 	if ((adapter->hw.mac_type < em_82543) ||
3113 	    /* Ignore Checksum bit is set */
3114 	    (rx_desc->status & E1000_RXD_STAT_IXSM)) {
3115 		mp->m_pkthdr.csum_flags = 0;
3116 		return;
3117 	}
3118 
3119 	if (rx_desc->status & E1000_RXD_STAT_IPCS) {
3120 		/* Did it pass? */
3121 		if (!(rx_desc->errors & E1000_RXD_ERR_IPE)) {
3122 			/* IP Checksum Good */
3123 			mp->m_pkthdr.csum_flags = CSUM_IP_CHECKED;
3124 			mp->m_pkthdr.csum_flags |= CSUM_IP_VALID;
3125 		} else {
3126 			mp->m_pkthdr.csum_flags = 0;
3127 		}
3128 	}
3129 
3130 	if (rx_desc->status & E1000_RXD_STAT_TCPCS) {
3131 		/* Did it pass? */
3132 		if (!(rx_desc->errors & E1000_RXD_ERR_TCPE)) {
3133 			mp->m_pkthdr.csum_flags |=
3134 			(CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
3135 			mp->m_pkthdr.csum_data = htons(0xffff);
3136 		}
3137 	}
3138 }
3139 
3140 
3141 static void
3142 em_enable_vlans(struct adapter *adapter)
3143 {
3144 	uint32_t ctrl;
3145 
3146 	E1000_WRITE_REG(&adapter->hw, VET, ETHERTYPE_VLAN);
3147 
3148 	ctrl = E1000_READ_REG(&adapter->hw, CTRL);
3149 	ctrl |= E1000_CTRL_VME;
3150 	E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
3151 }
3152 
3153 static void
3154 em_disable_vlans(struct adapter *adapter)
3155 {
3156 	uint32_t ctrl;
3157 
3158 	ctrl = E1000_READ_REG(&adapter->hw, CTRL);
3159 	ctrl &= ~E1000_CTRL_VME;
3160 	E1000_WRITE_REG(&adapter->hw, CTRL, ctrl);
3161 }
3162 
3163 /*
3164  * note: we must call bus_enable_intr() prior to enabling the hardware
3165  * interrupt and bus_disable_intr() after disabling the hardware interrupt
3166  * in order to avoid handler execution races from scheduled interrupt
3167  * threads.
3168  */
3169 static void
3170 em_enable_intr(struct adapter *adapter)
3171 {
3172 	struct ifnet *ifp = &adapter->interface_data.ac_if;
3173 
3174 	if ((ifp->if_flags & IFF_POLLING) == 0) {
3175 		lwkt_serialize_handler_enable(ifp->if_serializer);
3176 		E1000_WRITE_REG(&adapter->hw, IMS, (IMS_ENABLE_MASK));
3177 	}
3178 }
3179 
3180 static void
3181 em_disable_intr(struct adapter *adapter)
3182 {
3183 	/*
3184 	 * The first version of 82542 had an errata where when link was forced
3185 	 * it would stay up even up even if the cable was disconnected.
3186 	 * Sequence errors were used to detect the disconnect and then the
3187 	 * driver would unforce the link.  This code in the in the ISR.  For
3188 	 * this to work correctly the Sequence error interrupt had to be
3189 	 * enabled all the time.
3190 	 */
3191 	if (adapter->hw.mac_type == em_82542_rev2_0) {
3192 		E1000_WRITE_REG(&adapter->hw, IMC,
3193 				(0xffffffff & ~E1000_IMC_RXSEQ));
3194 	} else {
3195 		E1000_WRITE_REG(&adapter->hw, IMC, 0xffffffff);
3196 	}
3197 
3198 	lwkt_serialize_handler_disable(adapter->interface_data.ac_if.if_serializer);
3199 }
3200 
3201 static int
3202 em_is_valid_ether_addr(uint8_t *addr)
3203 {
3204 	static const char zero_addr[6] = { 0, 0, 0, 0, 0, 0 };
3205 
3206 	if ((addr[0] & 1) || !bcmp(addr, zero_addr, ETHER_ADDR_LEN))
3207 		return (FALSE);
3208 	else
3209 		return (TRUE);
3210 }
3211 
3212 void
3213 em_write_pci_cfg(struct em_hw *hw, uint32_t reg, uint16_t *value)
3214 {
3215 	pci_write_config(((struct em_osdep *)hw->back)->dev, reg, *value, 2);
3216 }
3217 
3218 void
3219 em_read_pci_cfg(struct em_hw *hw, uint32_t reg, uint16_t *value)
3220 {
3221 	*value = pci_read_config(((struct em_osdep *)hw->back)->dev, reg, 2);
3222 }
3223 
3224 void
3225 em_pci_set_mwi(struct em_hw *hw)
3226 {
3227 	pci_write_config(((struct em_osdep *)hw->back)->dev, PCIR_COMMAND,
3228 			 (hw->pci_cmd_word | CMD_MEM_WRT_INVALIDATE), 2);
3229 }
3230 
3231 void
3232 em_pci_clear_mwi(struct em_hw *hw)
3233 {
3234 	pci_write_config(((struct em_osdep *)hw->back)->dev, PCIR_COMMAND,
3235 			 (hw->pci_cmd_word & ~CMD_MEM_WRT_INVALIDATE), 2);
3236 }
3237 
3238 uint32_t
3239 em_io_read(struct em_hw *hw, unsigned long port)
3240 {
3241 	struct em_osdep *io = hw->back;
3242 
3243 	return bus_space_read_4(io->io_bus_space_tag,
3244 				io->io_bus_space_handle, port);
3245 }
3246 
3247 void
3248 em_io_write(struct em_hw *hw, unsigned long port, uint32_t value)
3249 {
3250 	struct em_osdep *io = hw->back;
3251 
3252 	bus_space_write_4(io->io_bus_space_tag,
3253 			  io->io_bus_space_handle, port, value);
3254 }
3255 
3256 /*
3257  * We may eventually really do this, but its unnecessary
3258  * for now so we just return unsupported.
3259  */
3260 int32_t
3261 em_read_pcie_cap_reg(struct em_hw *hw, uint32_t reg, uint16_t *value)
3262 {
3263 	return (0);
3264 }
3265 
3266 
3267 /*********************************************************************
3268  * 82544 Coexistence issue workaround.
3269  *    There are 2 issues.
3270  *	1. Transmit Hang issue.
3271  *    To detect this issue, following equation can be used...
3272  *          SIZE[3:0] + ADDR[2:0] = SUM[3:0].
3273  *          If SUM[3:0] is in between 1 to 4, we will have this issue.
3274  *
3275  *	2. DAC issue.
3276  *    To detect this issue, following equation can be used...
3277  *          SIZE[3:0] + ADDR[2:0] = SUM[3:0].
3278  *          If SUM[3:0] is in between 9 to c, we will have this issue.
3279  *
3280  *
3281  *    WORKAROUND:
3282  *          Make sure we do not have ending address as 1,2,3,4(Hang) or
3283  *          9,a,b,c (DAC)
3284  *
3285 *************************************************************************/
3286 static uint32_t
3287 em_fill_descriptors(bus_addr_t address, uint32_t length, PDESC_ARRAY desc_array)
3288 {
3289 	/* Since issue is sensitive to length and address.*/
3290 	/* Let us first check the address...*/
3291 	uint32_t safe_terminator;
3292 	if (length <= 4) {
3293 		desc_array->descriptor[0].address = address;
3294 		desc_array->descriptor[0].length = length;
3295 		desc_array->elements = 1;
3296 		return (desc_array->elements);
3297 	}
3298 	safe_terminator = (uint32_t)((((uint32_t)address & 0x7) + (length & 0xF)) & 0xF);
3299 	/* if it does not fall between 0x1 to 0x4 and 0x9 to 0xC then return */
3300 	if (safe_terminator == 0 ||
3301 	    (safe_terminator > 4 && safe_terminator < 9) ||
3302 	    (safe_terminator > 0xC && safe_terminator <= 0xF)) {
3303 		desc_array->descriptor[0].address = address;
3304 		desc_array->descriptor[0].length = length;
3305 		desc_array->elements = 1;
3306 		return (desc_array->elements);
3307 	}
3308 
3309 	desc_array->descriptor[0].address = address;
3310 	desc_array->descriptor[0].length = length - 4;
3311 	desc_array->descriptor[1].address = address + (length - 4);
3312 	desc_array->descriptor[1].length = 4;
3313 	desc_array->elements = 2;
3314 	return (desc_array->elements);
3315 }
3316 
3317 /**********************************************************************
3318  *
3319  *  Update the board statistics counters.
3320  *
3321  **********************************************************************/
3322 static void
3323 em_update_stats_counters(struct adapter *adapter)
3324 {
3325 	struct ifnet   *ifp;
3326 
3327 	if (adapter->hw.media_type == em_media_type_copper ||
3328 	    (E1000_READ_REG(&adapter->hw, STATUS) & E1000_STATUS_LU)) {
3329 		adapter->stats.symerrs += E1000_READ_REG(&adapter->hw, SYMERRS);
3330 		adapter->stats.sec += E1000_READ_REG(&adapter->hw, SEC);
3331 	}
3332 	adapter->stats.crcerrs += E1000_READ_REG(&adapter->hw, CRCERRS);
3333 	adapter->stats.mpc += E1000_READ_REG(&adapter->hw, MPC);
3334 	adapter->stats.scc += E1000_READ_REG(&adapter->hw, SCC);
3335 	adapter->stats.ecol += E1000_READ_REG(&adapter->hw, ECOL);
3336 
3337 	adapter->stats.mcc += E1000_READ_REG(&adapter->hw, MCC);
3338 	adapter->stats.latecol += E1000_READ_REG(&adapter->hw, LATECOL);
3339 	adapter->stats.colc += E1000_READ_REG(&adapter->hw, COLC);
3340 	adapter->stats.dc += E1000_READ_REG(&adapter->hw, DC);
3341 	adapter->stats.rlec += E1000_READ_REG(&adapter->hw, RLEC);
3342 	adapter->stats.xonrxc += E1000_READ_REG(&adapter->hw, XONRXC);
3343 	adapter->stats.xontxc += E1000_READ_REG(&adapter->hw, XONTXC);
3344 	adapter->stats.xoffrxc += E1000_READ_REG(&adapter->hw, XOFFRXC);
3345 	adapter->stats.xofftxc += E1000_READ_REG(&adapter->hw, XOFFTXC);
3346 	adapter->stats.fcruc += E1000_READ_REG(&adapter->hw, FCRUC);
3347 	adapter->stats.prc64 += E1000_READ_REG(&adapter->hw, PRC64);
3348 	adapter->stats.prc127 += E1000_READ_REG(&adapter->hw, PRC127);
3349 	adapter->stats.prc255 += E1000_READ_REG(&adapter->hw, PRC255);
3350 	adapter->stats.prc511 += E1000_READ_REG(&adapter->hw, PRC511);
3351 	adapter->stats.prc1023 += E1000_READ_REG(&adapter->hw, PRC1023);
3352 	adapter->stats.prc1522 += E1000_READ_REG(&adapter->hw, PRC1522);
3353 	adapter->stats.gprc += E1000_READ_REG(&adapter->hw, GPRC);
3354 	adapter->stats.bprc += E1000_READ_REG(&adapter->hw, BPRC);
3355 	adapter->stats.mprc += E1000_READ_REG(&adapter->hw, MPRC);
3356 	adapter->stats.gptc += E1000_READ_REG(&adapter->hw, GPTC);
3357 
3358 	/* For the 64-bit byte counters the low dword must be read first. */
3359 	/* Both registers clear on the read of the high dword */
3360 
3361 	adapter->stats.gorcl += E1000_READ_REG(&adapter->hw, GORCL);
3362 	adapter->stats.gorch += E1000_READ_REG(&adapter->hw, GORCH);
3363 	adapter->stats.gotcl += E1000_READ_REG(&adapter->hw, GOTCL);
3364 	adapter->stats.gotch += E1000_READ_REG(&adapter->hw, GOTCH);
3365 
3366 	adapter->stats.rnbc += E1000_READ_REG(&adapter->hw, RNBC);
3367 	adapter->stats.ruc += E1000_READ_REG(&adapter->hw, RUC);
3368 	adapter->stats.rfc += E1000_READ_REG(&adapter->hw, RFC);
3369 	adapter->stats.roc += E1000_READ_REG(&adapter->hw, ROC);
3370 	adapter->stats.rjc += E1000_READ_REG(&adapter->hw, RJC);
3371 
3372 	adapter->stats.torl += E1000_READ_REG(&adapter->hw, TORL);
3373 	adapter->stats.torh += E1000_READ_REG(&adapter->hw, TORH);
3374 	adapter->stats.totl += E1000_READ_REG(&adapter->hw, TOTL);
3375 	adapter->stats.toth += E1000_READ_REG(&adapter->hw, TOTH);
3376 
3377 	adapter->stats.tpr += E1000_READ_REG(&adapter->hw, TPR);
3378 	adapter->stats.tpt += E1000_READ_REG(&adapter->hw, TPT);
3379 	adapter->stats.ptc64 += E1000_READ_REG(&adapter->hw, PTC64);
3380 	adapter->stats.ptc127 += E1000_READ_REG(&adapter->hw, PTC127);
3381 	adapter->stats.ptc255 += E1000_READ_REG(&adapter->hw, PTC255);
3382 	adapter->stats.ptc511 += E1000_READ_REG(&adapter->hw, PTC511);
3383 	adapter->stats.ptc1023 += E1000_READ_REG(&adapter->hw, PTC1023);
3384 	adapter->stats.ptc1522 += E1000_READ_REG(&adapter->hw, PTC1522);
3385 	adapter->stats.mptc += E1000_READ_REG(&adapter->hw, MPTC);
3386 	adapter->stats.bptc += E1000_READ_REG(&adapter->hw, BPTC);
3387 
3388 	if (adapter->hw.mac_type >= em_82543) {
3389 		adapter->stats.algnerrc +=
3390 		    E1000_READ_REG(&adapter->hw, ALGNERRC);
3391 		adapter->stats.rxerrc +=
3392 		    E1000_READ_REG(&adapter->hw, RXERRC);
3393 		adapter->stats.tncrs +=
3394 		    E1000_READ_REG(&adapter->hw, TNCRS);
3395 		adapter->stats.cexterr +=
3396 		    E1000_READ_REG(&adapter->hw, CEXTERR);
3397 		adapter->stats.tsctc +=
3398 		    E1000_READ_REG(&adapter->hw, TSCTC);
3399 		adapter->stats.tsctfc +=
3400 		    E1000_READ_REG(&adapter->hw, TSCTFC);
3401 	}
3402 	ifp = &adapter->interface_data.ac_if;
3403 
3404 	/* Fill out the OS statistics structure */
3405 	ifp->if_collisions = adapter->stats.colc;
3406 
3407 	/* Rx Errors */
3408 	ifp->if_ierrors =
3409 		adapter->dropped_pkts +
3410 		adapter->stats.rxerrc +
3411 		adapter->stats.crcerrs +
3412 		adapter->stats.algnerrc +
3413 		adapter->stats.ruc + adapter->stats.roc +
3414 		adapter->stats.mpc + adapter->stats.cexterr +
3415 		adapter->rx_overruns;
3416 
3417 	/* Tx Errors */
3418 	ifp->if_oerrors = adapter->stats.ecol + adapter->stats.latecol +
3419 			  adapter->watchdog_timeouts;
3420 }
3421 
3422 
3423 /**********************************************************************
3424  *
3425  *  This routine is called only when em_display_debug_stats is enabled.
3426  *  This routine provides a way to take a look at important statistics
3427  *  maintained by the driver and hardware.
3428  *
3429  **********************************************************************/
3430 static void
3431 em_print_debug_info(struct adapter *adapter)
3432 {
3433 	device_t dev= adapter->dev;
3434 	uint8_t *hw_addr = adapter->hw.hw_addr;
3435 
3436 	device_printf(dev, "Adapter hardware address = %p \n", hw_addr);
3437 	device_printf(dev, "CTRL  = 0x%x RCTL = 0x%x\n",
3438 		      E1000_READ_REG(&adapter->hw, CTRL),
3439 		      E1000_READ_REG(&adapter->hw, RCTL));
3440 	device_printf(dev, "Packet buffer = Tx=%dk Rx=%dk\n",
3441 		      ((E1000_READ_REG(&adapter->hw, PBA) & 0xffff0000) >> 16),
3442 		      (E1000_READ_REG(&adapter->hw, PBA) & 0xffff));
3443 	device_printf(dev, "Flow control watermarks high = %d low = %d\n",
3444 		      adapter->hw.fc_high_water, adapter->hw.fc_low_water);
3445 	device_printf(dev, "tx_int_delay = %d, tx_abs_int_delay = %d\n",
3446 		      E1000_READ_REG(&adapter->hw, TIDV),
3447 		      E1000_READ_REG(&adapter->hw, TADV));
3448 	device_printf(dev, "rx_int_delay = %d, rx_abs_int_delay = %d\n",
3449 		      E1000_READ_REG(&adapter->hw, RDTR),
3450 		      E1000_READ_REG(&adapter->hw, RADV));
3451 	device_printf(dev, "fifo workaround = %lld, fifo_reset_count = %lld\n",
3452 		      (long long)adapter->tx_fifo_wrk_cnt,
3453 		      (long long)adapter->tx_fifo_reset_cnt);
3454 	device_printf(dev, "hw tdh = %d, hw tdt = %d\n",
3455 		      E1000_READ_REG(&adapter->hw, TDH),
3456 		      E1000_READ_REG(&adapter->hw, TDT));
3457 	device_printf(dev, "Num Tx descriptors avail = %d\n",
3458 		      adapter->num_tx_desc_avail);
3459 	device_printf(dev, "Tx Descriptors not avail1 = %ld\n",
3460 		      adapter->no_tx_desc_avail1);
3461 	device_printf(dev, "Tx Descriptors not avail2 = %ld\n",
3462 		      adapter->no_tx_desc_avail2);
3463 	device_printf(dev, "Std mbuf failed = %ld\n",
3464 		      adapter->mbuf_alloc_failed);
3465 	device_printf(dev, "Std mbuf cluster failed = %ld\n",
3466 		      adapter->mbuf_cluster_failed);
3467 	device_printf(dev, "Driver dropped packets = %ld\n",
3468 		      adapter->dropped_pkts);
3469 }
3470 
3471 static void
3472 em_print_hw_stats(struct adapter *adapter)
3473 {
3474 	device_t dev= adapter->dev;
3475 
3476 	device_printf(dev, "Excessive collisions = %lld\n",
3477 		      (long long)adapter->stats.ecol);
3478 	device_printf(dev, "Symbol errors = %lld\n",
3479 		      (long long)adapter->stats.symerrs);
3480 	device_printf(dev, "Sequence errors = %lld\n",
3481 		      (long long)adapter->stats.sec);
3482 	device_printf(dev, "Defer count = %lld\n",
3483 		      (long long)adapter->stats.dc);
3484 
3485 	device_printf(dev, "Missed Packets = %lld\n",
3486 		      (long long)adapter->stats.mpc);
3487 	device_printf(dev, "Receive No Buffers = %lld\n",
3488 		      (long long)adapter->stats.rnbc);
3489 	/* RLEC is inaccurate on some hardware, calculate our own. */
3490 	device_printf(dev, "Receive Length errors = %lld\n",
3491 		      (long long)adapter->stats.roc +
3492 		      (long long)adapter->stats.ruc);
3493 	device_printf(dev, "Receive errors = %lld\n",
3494 		      (long long)adapter->stats.rxerrc);
3495 	device_printf(dev, "Crc errors = %lld\n",
3496 		      (long long)adapter->stats.crcerrs);
3497 	device_printf(dev, "Alignment errors = %lld\n",
3498 		      (long long)adapter->stats.algnerrc);
3499 	device_printf(dev, "Carrier extension errors = %lld\n",
3500 		      (long long)adapter->stats.cexterr);
3501 	device_printf(dev, "RX overruns = %lu\n", adapter->rx_overruns);
3502 	device_printf(dev, "Watchdog timeouts = %lu\n",
3503 		      adapter->watchdog_timeouts);
3504 
3505 	device_printf(dev, "XON Rcvd = %lld\n",
3506 		      (long long)adapter->stats.xonrxc);
3507 	device_printf(dev, "XON Xmtd = %lld\n",
3508 		      (long long)adapter->stats.xontxc);
3509 	device_printf(dev, "XOFF Rcvd = %lld\n",
3510 		      (long long)adapter->stats.xoffrxc);
3511 	device_printf(dev, "XOFF Xmtd = %lld\n",
3512 		      (long long)adapter->stats.xofftxc);
3513 
3514 	device_printf(dev, "Good Packets Rcvd = %lld\n",
3515 		      (long long)adapter->stats.gprc);
3516 	device_printf(dev, "Good Packets Xmtd = %lld\n",
3517 		      (long long)adapter->stats.gptc);
3518 }
3519 
3520 static int
3521 em_sysctl_debug_info(SYSCTL_HANDLER_ARGS)
3522 {
3523 	int error;
3524 	int result;
3525 	struct adapter *adapter;
3526 
3527 	result = -1;
3528 	error = sysctl_handle_int(oidp, &result, 0, req);
3529 
3530 	if (error || !req->newptr)
3531 		return (error);
3532 
3533 	if (result == 1) {
3534 		adapter = (struct adapter *)arg1;
3535 		em_print_debug_info(adapter);
3536 	}
3537 
3538 	return (error);
3539 }
3540 
3541 static int
3542 em_sysctl_stats(SYSCTL_HANDLER_ARGS)
3543 {
3544 	int error;
3545 	int result;
3546 	struct adapter *adapter;
3547 
3548 	result = -1;
3549 	error = sysctl_handle_int(oidp, &result, 0, req);
3550 
3551 	if (error || !req->newptr)
3552 		return (error);
3553 
3554 	if (result == 1) {
3555 		adapter = (struct adapter *)arg1;
3556 		em_print_hw_stats(adapter);
3557 	}
3558 
3559 	return (error);
3560 }
3561 
3562 static int
3563 em_sysctl_int_delay(SYSCTL_HANDLER_ARGS)
3564 {
3565 	struct em_int_delay_info *info;
3566 	struct adapter *adapter;
3567 	uint32_t regval;
3568 	int error;
3569 	int usecs;
3570 	int ticks;
3571 
3572 	info = (struct em_int_delay_info *)arg1;
3573 	adapter = info->adapter;
3574 	usecs = info->value;
3575 	error = sysctl_handle_int(oidp, &usecs, 0, req);
3576 	if (error != 0 || req->newptr == NULL)
3577 		return (error);
3578 	if (usecs < 0 || usecs > E1000_TICKS_TO_USECS(65535))
3579 		return (EINVAL);
3580 	info->value = usecs;
3581 	ticks = E1000_USECS_TO_TICKS(usecs);
3582 
3583 	lwkt_serialize_enter(adapter->interface_data.ac_if.if_serializer);
3584 	regval = E1000_READ_OFFSET(&adapter->hw, info->offset);
3585 	regval = (regval & ~0xffff) | (ticks & 0xffff);
3586 	/* Handle a few special cases. */
3587 	switch (info->offset) {
3588 	case E1000_RDTR:
3589 	case E1000_82542_RDTR:
3590 		regval |= E1000_RDT_FPDB;
3591 		break;
3592 	case E1000_TIDV:
3593 	case E1000_82542_TIDV:
3594 		if (ticks == 0) {
3595 			adapter->txd_cmd &= ~E1000_TXD_CMD_IDE;
3596 			/* Don't write 0 into the TIDV register. */
3597 			regval++;
3598 		} else
3599 			adapter->txd_cmd |= E1000_TXD_CMD_IDE;
3600 		break;
3601 	}
3602 	E1000_WRITE_OFFSET(&adapter->hw, info->offset, regval);
3603 	lwkt_serialize_exit(adapter->interface_data.ac_if.if_serializer);
3604 	return (0);
3605 }
3606 
3607 static void
3608 em_add_int_delay_sysctl(struct adapter *adapter, const char *name,
3609 			const char *description, struct em_int_delay_info *info,
3610 			int offset, int value)
3611 {
3612 	info->adapter = adapter;
3613 	info->offset = offset;
3614 	info->value = value;
3615 	SYSCTL_ADD_PROC(&adapter->sysctl_ctx,
3616 			SYSCTL_CHILDREN(adapter->sysctl_tree),
3617 			OID_AUTO, name, CTLTYPE_INT|CTLFLAG_RW,
3618 			info, 0, em_sysctl_int_delay, "I", description);
3619 }
3620 
3621 static int
3622 em_sysctl_int_throttle(SYSCTL_HANDLER_ARGS)
3623 {
3624 	struct adapter *adapter = (void *)arg1;
3625 	int error;
3626 	int throttle;
3627 
3628 	throttle = em_int_throttle_ceil;
3629 	error = sysctl_handle_int(oidp, &throttle, 0, req);
3630 	if (error || req->newptr == NULL)
3631 		return error;
3632 	if (throttle < 0 || throttle > 1000000000 / 256)
3633 		return EINVAL;
3634 	if (throttle) {
3635 		/*
3636 		 * Set the interrupt throttling rate in 256ns increments,
3637 		 * recalculate sysctl value assignment to get exact frequency.
3638 		 */
3639 		throttle = 1000000000 / 256 / throttle;
3640 		lwkt_serialize_enter(adapter->interface_data.ac_if.if_serializer);
3641 		em_int_throttle_ceil = 1000000000 / 256 / throttle;
3642 		E1000_WRITE_REG(&adapter->hw, ITR, throttle);
3643 		lwkt_serialize_exit(adapter->interface_data.ac_if.if_serializer);
3644 	} else {
3645 		lwkt_serialize_enter(adapter->interface_data.ac_if.if_serializer);
3646 		em_int_throttle_ceil = 0;
3647 		E1000_WRITE_REG(&adapter->hw, ITR, 0);
3648 		lwkt_serialize_exit(adapter->interface_data.ac_if.if_serializer);
3649 	}
3650 	device_printf(adapter->dev, "Interrupt moderation set to %d/sec\n",
3651 			em_int_throttle_ceil);
3652 	return 0;
3653 }
3654